1 diff --git a/Documentation/hwlat_detector.txt b/Documentation/hwlat_detector.txt
3 index 000000000000..cb61516483d3
5 +++ b/Documentation/hwlat_detector.txt
10 +The module hwlat_detector is a special purpose kernel module that is used to
11 +detect large system latencies induced by the behavior of certain underlying
12 +hardware or firmware, independent of Linux itself. The code was developed
13 +originally to detect SMIs (System Management Interrupts) on x86 systems,
14 +however there is nothing x86 specific about this patchset. It was
15 +originally written for use by the "RT" patch since the Real Time
16 +kernel is highly latency sensitive.
18 +SMIs are usually not serviced by the Linux kernel, which typically does not
19 +even know that they are occuring. SMIs are instead are set up by BIOS code
20 +and are serviced by BIOS code, usually for "critical" events such as
21 +management of thermal sensors and fans. Sometimes though, SMIs are used for
22 +other tasks and those tasks can spend an inordinate amount of time in the
23 +handler (sometimes measured in milliseconds). Obviously this is a problem if
24 +you are trying to keep event service latencies down in the microsecond range.
26 +The hardware latency detector works by hogging all of the cpus for configurable
27 +amounts of time (by calling stop_machine()), polling the CPU Time Stamp Counter
28 +for some period, then looking for gaps in the TSC data. Any gap indicates a
29 +time when the polling was interrupted and since the machine is stopped and
30 +interrupts turned off the only thing that could do that would be an SMI.
32 +Note that the SMI detector should *NEVER* be used in a production environment.
33 +It is intended to be run manually to determine if the hardware platform has a
34 +problem with long system firmware service routines.
39 +Loading the module hwlat_detector passing the parameter "enabled=1" (or by
40 +setting the "enable" entry in "hwlat_detector" debugfs toggled on) is the only
41 +step required to start the hwlat_detector. It is possible to redefine the
42 +threshold in microseconds (us) above which latency spikes will be taken
43 +into account (parameter "threshold=").
47 + # modprobe hwlat_detector enabled=1 threshold=100
49 +After the module is loaded, it creates a directory named "hwlat_detector" under
50 +the debugfs mountpoint, "/debug/hwlat_detector" for this text. It is necessary
51 +to have debugfs mounted, which might be on /sys/debug on your system.
53 +The /debug/hwlat_detector interface contains the following files:
55 +count - number of latency spikes observed since last reset
56 +enable - a global enable/disable toggle (0/1), resets count
57 +max - maximum hardware latency actually observed (usecs)
58 +sample - a pipe from which to read current raw sample data
59 + in the format <timestamp> <latency observed usecs>
60 + (can be opened O_NONBLOCK for a single sample)
61 +threshold - minimum latency value to be considered (usecs)
62 +width - time period to sample with CPUs held (usecs)
63 + must be less than the total window size (enforced)
64 +window - total period of sampling, width being inside (usecs)
66 +By default we will set width to 500,000 and window to 1,000,000, meaning that
67 +we will sample every 1,000,000 usecs (1s) for 500,000 usecs (0.5s). If we
68 +observe any latencies that exceed the threshold (initially 100 usecs),
69 +then we write to a global sample ring buffer of 8K samples, which is
70 +consumed by reading from the "sample" (pipe) debugfs file interface.
71 diff --git a/Documentation/kernel-parameters.txt b/Documentation/kernel-parameters.txt
72 index ca64ca566099..19bef799c902 100644
73 --- a/Documentation/kernel-parameters.txt
74 +++ b/Documentation/kernel-parameters.txt
75 @@ -1640,6 +1640,15 @@ bytes respectively. Such letter suffixes can also be entirely omitted.
77 See Documentation/filesystems/nfs/nfsroot.txt.
79 + irqaffinity= [SMP] Set the default irq affinity mask
81 + <cpu number>,...,<cpu number>
83 + <cpu number>-<cpu number>
84 + (must be a positive range in ascending order)
86 + <cpu number>,...,<cpu number>-<cpu number>
89 When an interrupt is not handled search all handlers
90 for it. Intended to get systems with badly broken
91 diff --git a/Documentation/sysrq.txt b/Documentation/sysrq.txt
92 index 13f5619b2203..f64d075ba647 100644
93 --- a/Documentation/sysrq.txt
94 +++ b/Documentation/sysrq.txt
95 @@ -59,10 +59,17 @@ On PowerPC - Press 'ALT - Print Screen (or F13) - <command key>,
96 On other - If you know of the key combos for other architectures, please
97 let me know so I can add them to this section.
99 -On all - write a character to /proc/sysrq-trigger. e.g.:
101 +On all - write a character to /proc/sysrq-trigger, e.g.:
102 echo t > /proc/sysrq-trigger
104 +On all - Enable network SysRq by writing a cookie to icmp_echo_sysrq, e.g.
105 + echo 0x01020304 >/proc/sys/net/ipv4/icmp_echo_sysrq
106 + Send an ICMP echo request with this pattern plus the particular
107 + SysRq command key. Example:
108 + # ping -c1 -s57 -p0102030468
109 + will trigger the SysRq-H (help) command.
112 * What are the 'command' keys?
113 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
114 'b' - Will immediately reboot the system without syncing or unmounting
115 diff --git a/Documentation/trace/histograms.txt b/Documentation/trace/histograms.txt
117 index 000000000000..6f2aeabf7faa
119 +++ b/Documentation/trace/histograms.txt
121 + Using the Linux Kernel Latency Histograms
124 +This document gives a short explanation how to enable, configure and use
125 +latency histograms. Latency histograms are primarily relevant in the
126 +context of real-time enabled kernels (CONFIG_PREEMPT/CONFIG_PREEMPT_RT)
127 +and are used in the quality management of the Linux real-time
131 +* Purpose of latency histograms
133 +A latency histogram continuously accumulates the frequencies of latency
134 +data. There are two types of histograms
135 +- potential sources of latencies
136 +- effective latencies
139 +* Potential sources of latencies
141 +Potential sources of latencies are code segments where interrupts,
142 +preemption or both are disabled (aka critical sections). To create
143 +histograms of potential sources of latency, the kernel stores the time
144 +stamp at the start of a critical section, determines the time elapsed
145 +when the end of the section is reached, and increments the frequency
146 +counter of that latency value - irrespective of whether any concurrently
147 +running process is affected by latency or not.
148 +- Configuration items (in the Kernel hacking/Tracers submenu)
149 + CONFIG_INTERRUPT_OFF_LATENCY
150 + CONFIG_PREEMPT_OFF_LATENCY
153 +* Effective latencies
155 +Effective latencies are actually occuring during wakeup of a process. To
156 +determine effective latencies, the kernel stores the time stamp when a
157 +process is scheduled to be woken up, and determines the duration of the
158 +wakeup time shortly before control is passed over to this process. Note
159 +that the apparent latency in user space may be somewhat longer, since the
160 +process may be interrupted after control is passed over to it but before
161 +the execution in user space takes place. Simply measuring the interval
162 +between enqueuing and wakeup may also not appropriate in cases when a
163 +process is scheduled as a result of a timer expiration. The timer may have
164 +missed its deadline, e.g. due to disabled interrupts, but this latency
165 +would not be registered. Therefore, the offsets of missed timers are
166 +recorded in a separate histogram. If both wakeup latency and missed timer
167 +offsets are configured and enabled, a third histogram may be enabled that
168 +records the overall latency as a sum of the timer latency, if any, and the
169 +wakeup latency. This histogram is called "timerandwakeup".
170 +- Configuration items (in the Kernel hacking/Tracers submenu)
171 + CONFIG_WAKEUP_LATENCY
172 + CONFIG_MISSED_TIMER_OFSETS
177 +The interface to the administration of the latency histograms is located
178 +in the debugfs file system. To mount it, either enter
180 +mount -t sysfs nodev /sys
181 +mount -t debugfs nodev /sys/kernel/debug
183 +from shell command line level, or add
185 +nodev /sys sysfs defaults 0 0
186 +nodev /sys/kernel/debug debugfs defaults 0 0
188 +to the file /etc/fstab. All latency histogram related files are then
189 +available in the directory /sys/kernel/debug/tracing/latency_hist. A
190 +particular histogram type is enabled by writing non-zero to the related
191 +variable in the /sys/kernel/debug/tracing/latency_hist/enable directory.
192 +Select "preemptirqsoff" for the histograms of potential sources of
193 +latencies and "wakeup" for histograms of effective latencies etc. The
194 +histogram data - one per CPU - are available in the files
196 +/sys/kernel/debug/tracing/latency_hist/preemptoff/CPUx
197 +/sys/kernel/debug/tracing/latency_hist/irqsoff/CPUx
198 +/sys/kernel/debug/tracing/latency_hist/preemptirqsoff/CPUx
199 +/sys/kernel/debug/tracing/latency_hist/wakeup/CPUx
200 +/sys/kernel/debug/tracing/latency_hist/wakeup/sharedprio/CPUx
201 +/sys/kernel/debug/tracing/latency_hist/missed_timer_offsets/CPUx
202 +/sys/kernel/debug/tracing/latency_hist/timerandwakeup/CPUx
204 +The histograms are reset by writing non-zero to the file "reset" in a
205 +particular latency directory. To reset all latency data, use
209 +TRACINGDIR=/sys/kernel/debug/tracing
210 +HISTDIR=$TRACINGDIR/latency_hist
215 + for i in `find . | grep /reset$`
224 +Latency data are stored with a resolution of one microsecond. The
225 +maximum latency is 10,240 microseconds. The data are only valid, if the
226 +overflow register is empty. Every output line contains the latency in
227 +microseconds in the first row and the number of samples in the second
228 +row. To display only lines with a positive latency count, use, for
231 +grep -v " 0$" /sys/kernel/debug/tracing/latency_hist/preemptoff/CPU0
233 +#Minimum latency: 0 microseconds.
234 +#Average latency: 0 microseconds.
235 +#Maximum latency: 25 microseconds.
236 +#Total samples: 3104770694
237 +#There are 0 samples greater or equal than 10240 microseconds
266 +* Wakeup latency of a selected process
268 +To only collect wakeup latency data of a particular process, write the
269 +PID of the requested process to
271 +/sys/kernel/debug/tracing/latency_hist/wakeup/pid
273 +PIDs are not considered, if this variable is set to 0.
276 +* Details of the process with the highest wakeup latency so far
278 +Selected data of the process that suffered from the highest wakeup
279 +latency that occurred in a particular CPU are available in the file
281 +/sys/kernel/debug/tracing/latency_hist/wakeup/max_latency-CPUx.
283 +In addition, other relevant system data at the time when the
284 +latency occurred are given.
286 +The format of the data is (all in one line):
287 +<PID> <Priority> <Latency> (<Timeroffset>) <Command> \
288 +<- <PID> <Priority> <Command> <Timestamp>
290 +The value of <Timeroffset> is only relevant in the combined timer
291 +and wakeup latency recording. In the wakeup recording, it is
292 +always 0, in the missed_timer_offsets recording, it is the same
295 +When retrospectively searching for the origin of a latency and
296 +tracing was not enabled, it may be helpful to know the name and
297 +some basic data of the task that (finally) was switching to the
298 +late real-tlme task. In addition to the victim's data, also the
299 +data of the possible culprit are therefore displayed after the
302 +Finally, the timestamp of the time when the latency occurred
303 +in <seconds>.<microseconds> after the most recent system boot
306 +These data are also reset when the wakeup histogram is reset.
307 diff --git a/Makefile b/Makefile
308 index a5ecb29c6ed3..5521b0c3abf2 100644
311 @@ -785,6 +785,9 @@ KBUILD_CFLAGS += $(call cc-option,-Werror=strict-prototypes)
312 # Prohibit date/time macros, which would make the build non-deterministic
313 KBUILD_CFLAGS += $(call cc-option,-Werror=date-time)
315 +# enforce correct pointer usage
316 +KBUILD_CFLAGS += $(call cc-option,-Werror=incompatible-pointer-types)
318 # use the deterministic mode of AR if available
319 KBUILD_ARFLAGS := $(call ar-option,D)
321 diff --git a/arch/Kconfig b/arch/Kconfig
322 index 4e949e58b192..3b26d76933fb 100644
325 @@ -9,6 +9,7 @@ config OPROFILE
326 tristate "OProfile system profiling"
328 depends on HAVE_OPROFILE
329 + depends on !PREEMPT_RT_FULL
331 select RING_BUFFER_ALLOW_SWAP
333 @@ -52,6 +53,7 @@ config KPROBES
335 bool "Optimize very unlikely/likely branches"
336 depends on HAVE_ARCH_JUMP_LABEL
337 + depends on (!INTERRUPT_OFF_HIST && !PREEMPT_OFF_HIST && !WAKEUP_LATENCY_HIST && !MISSED_TIMER_OFFSETS_HIST)
339 This option enables a transparent branch optimization that
340 makes certain almost-always-true or almost-always-false branch
341 diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig
342 index 34e1569a11ee..79c4603e9453 100644
343 --- a/arch/arm/Kconfig
344 +++ b/arch/arm/Kconfig
345 @@ -33,7 +33,7 @@ config ARM
346 select HARDIRQS_SW_RESEND
347 select HAVE_ARCH_AUDITSYSCALL if (AEABI && !OABI_COMPAT)
348 select HAVE_ARCH_BITREVERSE if (CPU_32v7M || CPU_32v7) && !CPU_32v6
349 - select HAVE_ARCH_JUMP_LABEL if !XIP_KERNEL && !CPU_ENDIAN_BE32
350 + select HAVE_ARCH_JUMP_LABEL if !XIP_KERNEL && !CPU_ENDIAN_BE32 && !PREEMPT_RT_BASE
351 select HAVE_ARCH_KGDB if !CPU_ENDIAN_BE32
352 select HAVE_ARCH_SECCOMP_FILTER if (AEABI && !OABI_COMPAT)
353 select HAVE_ARCH_TRACEHOOK
354 @@ -68,6 +68,7 @@ config ARM
355 select HAVE_PERF_EVENTS
356 select HAVE_PERF_REGS
357 select HAVE_PERF_USER_STACK_DUMP
358 + select HAVE_PREEMPT_LAZY
359 select HAVE_RCU_TABLE_FREE if (SMP && ARM_LPAE)
360 select HAVE_REGS_AND_STACK_ACCESS_API
361 select HAVE_SYSCALL_TRACEPOINTS
362 diff --git a/arch/arm/include/asm/switch_to.h b/arch/arm/include/asm/switch_to.h
363 index 12ebfcc1d539..c962084605bc 100644
364 --- a/arch/arm/include/asm/switch_to.h
365 +++ b/arch/arm/include/asm/switch_to.h
368 #include <linux/thread_info.h>
370 +#if defined CONFIG_PREEMPT_RT_FULL && defined CONFIG_HIGHMEM
371 +void switch_kmaps(struct task_struct *prev_p, struct task_struct *next_p);
374 +switch_kmaps(struct task_struct *prev_p, struct task_struct *next_p) { }
378 * For v7 SMP cores running a preemptible kernel we may be pre-empted
379 * during a TLB maintenance operation, so execute an inner-shareable dsb
380 @@ -25,6 +32,7 @@ extern struct task_struct *__switch_to(struct task_struct *, struct thread_info
381 #define switch_to(prev,next,last) \
383 __complete_pending_tlbi(); \
384 + switch_kmaps(prev, next); \
385 last = __switch_to(prev,task_thread_info(prev), task_thread_info(next)); \
388 diff --git a/arch/arm/include/asm/thread_info.h b/arch/arm/include/asm/thread_info.h
389 index 776757d1604a..1f36a4eccc72 100644
390 --- a/arch/arm/include/asm/thread_info.h
391 +++ b/arch/arm/include/asm/thread_info.h
392 @@ -49,6 +49,7 @@ struct cpu_context_save {
394 unsigned long flags; /* low level flags */
395 int preempt_count; /* 0 => preemptable, <0 => bug */
396 + int preempt_lazy_count; /* 0 => preemptable, <0 => bug */
397 mm_segment_t addr_limit; /* address limit */
398 struct task_struct *task; /* main task structure */
400 @@ -142,7 +143,8 @@ extern int vfp_restore_user_hwstate(struct user_vfp __user *,
401 #define TIF_SYSCALL_TRACE 4 /* syscall trace active */
402 #define TIF_SYSCALL_AUDIT 5 /* syscall auditing active */
403 #define TIF_SYSCALL_TRACEPOINT 6 /* syscall tracepoint instrumentation */
404 -#define TIF_SECCOMP 7 /* seccomp syscall filtering active */
405 +#define TIF_SECCOMP 8 /* seccomp syscall filtering active */
406 +#define TIF_NEED_RESCHED_LAZY 7
408 #define TIF_NOHZ 12 /* in adaptive nohz mode */
409 #define TIF_USING_IWMMXT 17
410 @@ -152,6 +154,7 @@ extern int vfp_restore_user_hwstate(struct user_vfp __user *,
411 #define _TIF_SIGPENDING (1 << TIF_SIGPENDING)
412 #define _TIF_NEED_RESCHED (1 << TIF_NEED_RESCHED)
413 #define _TIF_NOTIFY_RESUME (1 << TIF_NOTIFY_RESUME)
414 +#define _TIF_NEED_RESCHED_LAZY (1 << TIF_NEED_RESCHED_LAZY)
415 #define _TIF_UPROBE (1 << TIF_UPROBE)
416 #define _TIF_SYSCALL_TRACE (1 << TIF_SYSCALL_TRACE)
417 #define _TIF_SYSCALL_AUDIT (1 << TIF_SYSCALL_AUDIT)
418 @@ -167,7 +170,8 @@ extern int vfp_restore_user_hwstate(struct user_vfp __user *,
419 * Change these and you break ASM code in entry-common.S
421 #define _TIF_WORK_MASK (_TIF_NEED_RESCHED | _TIF_SIGPENDING | \
422 - _TIF_NOTIFY_RESUME | _TIF_UPROBE)
423 + _TIF_NOTIFY_RESUME | _TIF_UPROBE | \
424 + _TIF_NEED_RESCHED_LAZY)
426 #endif /* __KERNEL__ */
427 #endif /* __ASM_ARM_THREAD_INFO_H */
428 diff --git a/arch/arm/kernel/asm-offsets.c b/arch/arm/kernel/asm-offsets.c
429 index 871b8267d211..4dbe70de7318 100644
430 --- a/arch/arm/kernel/asm-offsets.c
431 +++ b/arch/arm/kernel/asm-offsets.c
432 @@ -65,6 +65,7 @@ int main(void)
434 DEFINE(TI_FLAGS, offsetof(struct thread_info, flags));
435 DEFINE(TI_PREEMPT, offsetof(struct thread_info, preempt_count));
436 + DEFINE(TI_PREEMPT_LAZY, offsetof(struct thread_info, preempt_lazy_count));
437 DEFINE(TI_ADDR_LIMIT, offsetof(struct thread_info, addr_limit));
438 DEFINE(TI_TASK, offsetof(struct thread_info, task));
439 DEFINE(TI_CPU, offsetof(struct thread_info, cpu));
440 diff --git a/arch/arm/kernel/entry-armv.S b/arch/arm/kernel/entry-armv.S
441 index 3ce377f7251f..d044cea59f54 100644
442 --- a/arch/arm/kernel/entry-armv.S
443 +++ b/arch/arm/kernel/entry-armv.S
444 @@ -215,11 +215,18 @@ __irq_svc:
445 #ifdef CONFIG_PREEMPT
447 ldr r8, [tsk, #TI_PREEMPT] @ get preempt count
448 - ldr r0, [tsk, #TI_FLAGS] @ get flags
449 teq r8, #0 @ if preempt count != 0
450 + bne 1f @ return from exeption
451 + ldr r0, [tsk, #TI_FLAGS] @ get flags
452 + tst r0, #_TIF_NEED_RESCHED @ if NEED_RESCHED is set
453 + blne svc_preempt @ preempt!
455 + ldr r8, [tsk, #TI_PREEMPT_LAZY] @ get preempt lazy count
456 + teq r8, #0 @ if preempt lazy count != 0
457 movne r0, #0 @ force flags to 0
458 - tst r0, #_TIF_NEED_RESCHED
459 + tst r0, #_TIF_NEED_RESCHED_LAZY
464 svc_exit r5, irq = 1 @ return from exception
465 @@ -234,8 +241,14 @@ svc_preempt:
466 1: bl preempt_schedule_irq @ irq en/disable is done inside
467 ldr r0, [tsk, #TI_FLAGS] @ get new tasks TI_FLAGS
468 tst r0, #_TIF_NEED_RESCHED
470 + tst r0, #_TIF_NEED_RESCHED_LAZY
473 + ldr r0, [tsk, #TI_PREEMPT_LAZY] @ get preempt lazy count
474 + teq r0, #0 @ if preempt lazy count != 0
481 diff --git a/arch/arm/kernel/entry-common.S b/arch/arm/kernel/entry-common.S
482 index 30a7228eaceb..c3bd6cbfce4b 100644
483 --- a/arch/arm/kernel/entry-common.S
484 +++ b/arch/arm/kernel/entry-common.S
485 @@ -36,7 +36,9 @@ ret_fast_syscall:
487 disable_irq_notrace @ disable interrupts
488 ldr r1, [tsk, #TI_FLAGS] @ re-check for syscall tracing
489 - tst r1, #_TIF_SYSCALL_WORK | _TIF_WORK_MASK
490 + tst r1, #((_TIF_SYSCALL_WORK | _TIF_WORK_MASK) & ~_TIF_SECCOMP)
491 + bne fast_work_pending
492 + tst r1, #_TIF_SECCOMP
493 bne fast_work_pending
495 /* perform architecture specific actions before user return */
496 @@ -62,8 +64,11 @@ ret_fast_syscall:
497 str r0, [sp, #S_R0 + S_OFF]! @ save returned r0
498 disable_irq_notrace @ disable interrupts
499 ldr r1, [tsk, #TI_FLAGS] @ re-check for syscall tracing
500 - tst r1, #_TIF_SYSCALL_WORK | _TIF_WORK_MASK
501 + tst r1, #((_TIF_SYSCALL_WORK | _TIF_WORK_MASK) & ~_TIF_SECCOMP)
503 + tst r1, #_TIF_SECCOMP
507 ENDPROC(ret_fast_syscall)
509 diff --git a/arch/arm/kernel/patch.c b/arch/arm/kernel/patch.c
510 index 69bda1a5707e..1f665acaa6a9 100644
511 --- a/arch/arm/kernel/patch.c
512 +++ b/arch/arm/kernel/patch.c
513 @@ -15,7 +15,7 @@ struct patch {
517 -static DEFINE_SPINLOCK(patch_lock);
518 +static DEFINE_RAW_SPINLOCK(patch_lock);
520 static void __kprobes *patch_map(void *addr, int fixmap, unsigned long *flags)
521 __acquires(&patch_lock)
522 @@ -32,7 +32,7 @@ static void __kprobes *patch_map(void *addr, int fixmap, unsigned long *flags)
526 - spin_lock_irqsave(&patch_lock, *flags);
527 + raw_spin_lock_irqsave(&patch_lock, *flags);
529 __acquire(&patch_lock);
531 @@ -47,7 +47,7 @@ static void __kprobes patch_unmap(int fixmap, unsigned long *flags)
532 clear_fixmap(fixmap);
535 - spin_unlock_irqrestore(&patch_lock, *flags);
536 + raw_spin_unlock_irqrestore(&patch_lock, *flags);
538 __release(&patch_lock);
540 diff --git a/arch/arm/kernel/process.c b/arch/arm/kernel/process.c
541 index 4adfb46e3ee9..15f1d94b47c5 100644
542 --- a/arch/arm/kernel/process.c
543 +++ b/arch/arm/kernel/process.c
544 @@ -319,6 +319,30 @@ unsigned long arch_randomize_brk(struct mm_struct *mm)
549 + * CONFIG_SPLIT_PTLOCK_CPUS results in a page->ptl lock. If the lock is not
550 + * initialized by pgtable_page_ctor() then a coredump of the vector page will
553 +static int __init vectors_user_mapping_init_page(void)
556 + unsigned long addr = 0xffff0000;
561 + pgd = pgd_offset_k(addr);
562 + pud = pud_offset(pgd, addr);
563 + pmd = pmd_offset(pud, addr);
564 + page = pmd_page(*(pmd));
566 + pgtable_page_ctor(page);
570 +late_initcall(vectors_user_mapping_init_page);
572 #ifdef CONFIG_KUSER_HELPERS
574 * The vectors page is always readable from user space for the
575 diff --git a/arch/arm/kernel/signal.c b/arch/arm/kernel/signal.c
576 index 7b8f2141427b..96541e00b74a 100644
577 --- a/arch/arm/kernel/signal.c
578 +++ b/arch/arm/kernel/signal.c
579 @@ -572,7 +572,8 @@ do_work_pending(struct pt_regs *regs, unsigned int thread_flags, int syscall)
581 trace_hardirqs_off();
583 - if (likely(thread_flags & _TIF_NEED_RESCHED)) {
584 + if (likely(thread_flags & (_TIF_NEED_RESCHED |
585 + _TIF_NEED_RESCHED_LAZY))) {
588 if (unlikely(!user_mode(regs)))
589 diff --git a/arch/arm/kernel/smp.c b/arch/arm/kernel/smp.c
590 index b26361355dae..e5754e3b03c4 100644
591 --- a/arch/arm/kernel/smp.c
592 +++ b/arch/arm/kernel/smp.c
593 @@ -230,8 +230,6 @@ int __cpu_disable(void)
595 local_flush_tlb_all();
597 - clear_tasks_mm_cpumask(cpu);
602 @@ -247,6 +245,9 @@ void __cpu_die(unsigned int cpu)
603 pr_err("CPU%u: cpu didn't die\n", cpu);
607 + clear_tasks_mm_cpumask(cpu);
609 pr_notice("CPU%u: shutdown\n", cpu);
612 diff --git a/arch/arm/kernel/unwind.c b/arch/arm/kernel/unwind.c
613 index 0bee233fef9a..314cfb232a63 100644
614 --- a/arch/arm/kernel/unwind.c
615 +++ b/arch/arm/kernel/unwind.c
616 @@ -93,7 +93,7 @@ extern const struct unwind_idx __start_unwind_idx[];
617 static const struct unwind_idx *__origin_unwind_idx;
618 extern const struct unwind_idx __stop_unwind_idx[];
620 -static DEFINE_SPINLOCK(unwind_lock);
621 +static DEFINE_RAW_SPINLOCK(unwind_lock);
622 static LIST_HEAD(unwind_tables);
624 /* Convert a prel31 symbol to an absolute address */
625 @@ -201,7 +201,7 @@ static const struct unwind_idx *unwind_find_idx(unsigned long addr)
626 /* module unwind tables */
627 struct unwind_table *table;
629 - spin_lock_irqsave(&unwind_lock, flags);
630 + raw_spin_lock_irqsave(&unwind_lock, flags);
631 list_for_each_entry(table, &unwind_tables, list) {
632 if (addr >= table->begin_addr &&
633 addr < table->end_addr) {
634 @@ -213,7 +213,7 @@ static const struct unwind_idx *unwind_find_idx(unsigned long addr)
638 - spin_unlock_irqrestore(&unwind_lock, flags);
639 + raw_spin_unlock_irqrestore(&unwind_lock, flags);
642 pr_debug("%s: idx = %p\n", __func__, idx);
643 @@ -529,9 +529,9 @@ struct unwind_table *unwind_table_add(unsigned long start, unsigned long size,
644 tab->begin_addr = text_addr;
645 tab->end_addr = text_addr + text_size;
647 - spin_lock_irqsave(&unwind_lock, flags);
648 + raw_spin_lock_irqsave(&unwind_lock, flags);
649 list_add_tail(&tab->list, &unwind_tables);
650 - spin_unlock_irqrestore(&unwind_lock, flags);
651 + raw_spin_unlock_irqrestore(&unwind_lock, flags);
655 @@ -543,9 +543,9 @@ void unwind_table_del(struct unwind_table *tab)
659 - spin_lock_irqsave(&unwind_lock, flags);
660 + raw_spin_lock_irqsave(&unwind_lock, flags);
661 list_del(&tab->list);
662 - spin_unlock_irqrestore(&unwind_lock, flags);
663 + raw_spin_unlock_irqrestore(&unwind_lock, flags);
667 diff --git a/arch/arm/kvm/arm.c b/arch/arm/kvm/arm.c
668 index d7bef2144760..36a3e51492f7 100644
669 --- a/arch/arm/kvm/arm.c
670 +++ b/arch/arm/kvm/arm.c
671 @@ -496,18 +496,18 @@ static void kvm_arm_resume_guest(struct kvm *kvm)
672 struct kvm_vcpu *vcpu;
674 kvm_for_each_vcpu(i, vcpu, kvm) {
675 - wait_queue_head_t *wq = kvm_arch_vcpu_wq(vcpu);
676 + struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
678 vcpu->arch.pause = false;
679 - wake_up_interruptible(wq);
684 static void vcpu_sleep(struct kvm_vcpu *vcpu)
686 - wait_queue_head_t *wq = kvm_arch_vcpu_wq(vcpu);
687 + struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
689 - wait_event_interruptible(*wq, ((!vcpu->arch.power_off) &&
690 + swait_event_interruptible(*wq, ((!vcpu->arch.power_off) &&
691 (!vcpu->arch.pause)));
694 @@ -566,7 +566,7 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
695 * involves poking the GIC, which must be done in a
696 * non-preemptible context.
700 kvm_timer_flush_hwstate(vcpu);
701 kvm_vgic_flush_hwstate(vcpu);
703 @@ -585,7 +585,7 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
705 kvm_timer_sync_hwstate(vcpu);
706 kvm_vgic_sync_hwstate(vcpu);
712 @@ -639,7 +639,7 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
714 kvm_vgic_sync_hwstate(vcpu);
719 ret = handle_exit(vcpu, run, ret);
721 diff --git a/arch/arm/kvm/psci.c b/arch/arm/kvm/psci.c
722 index 443db0c43d7c..a08d7a93aebb 100644
723 --- a/arch/arm/kvm/psci.c
724 +++ b/arch/arm/kvm/psci.c
725 @@ -70,7 +70,7 @@ static unsigned long kvm_psci_vcpu_on(struct kvm_vcpu *source_vcpu)
727 struct kvm *kvm = source_vcpu->kvm;
728 struct kvm_vcpu *vcpu = NULL;
729 - wait_queue_head_t *wq;
730 + struct swait_queue_head *wq;
731 unsigned long cpu_id;
732 unsigned long context_id;
733 phys_addr_t target_pc;
734 @@ -119,7 +119,7 @@ static unsigned long kvm_psci_vcpu_on(struct kvm_vcpu *source_vcpu)
735 smp_mb(); /* Make sure the above is visible */
737 wq = kvm_arch_vcpu_wq(vcpu);
738 - wake_up_interruptible(wq);
741 return PSCI_RET_SUCCESS;
743 diff --git a/arch/arm/mach-at91/Kconfig b/arch/arm/mach-at91/Kconfig
744 index 28656c2b54a0..3f501305ca26 100644
745 --- a/arch/arm/mach-at91/Kconfig
746 +++ b/arch/arm/mach-at91/Kconfig
747 @@ -99,6 +99,7 @@ config HAVE_AT91_USB_CLK
748 config COMMON_CLK_AT91
755 diff --git a/arch/arm/mach-at91/at91rm9200.c b/arch/arm/mach-at91/at91rm9200.c
756 index c1a7c6cc00e1..63b4fa25b48a 100644
757 --- a/arch/arm/mach-at91/at91rm9200.c
758 +++ b/arch/arm/mach-at91/at91rm9200.c
760 #include <linux/of_platform.h>
762 #include <asm/mach/arch.h>
763 -#include <asm/system_misc.h>
767 @@ -33,7 +32,6 @@ static void __init at91rm9200_dt_device_init(void)
769 of_platform_populate(NULL, of_default_bus_match_table, NULL, soc_dev);
771 - arm_pm_idle = at91rm9200_idle;
772 at91rm9200_pm_init();
775 diff --git a/arch/arm/mach-at91/at91sam9.c b/arch/arm/mach-at91/at91sam9.c
776 index 7eb64f763034..cada2a6412b3 100644
777 --- a/arch/arm/mach-at91/at91sam9.c
778 +++ b/arch/arm/mach-at91/at91sam9.c
779 @@ -62,8 +62,6 @@ static void __init at91sam9_common_init(void)
780 soc_dev = soc_device_to_device(soc);
782 of_platform_populate(NULL, of_default_bus_match_table, NULL, soc_dev);
784 - arm_pm_idle = at91sam9_idle;
787 static void __init at91sam9_dt_device_init(void)
788 diff --git a/arch/arm/mach-at91/generic.h b/arch/arm/mach-at91/generic.h
789 index b0fa7dc7286d..28ca57a2060f 100644
790 --- a/arch/arm/mach-at91/generic.h
791 +++ b/arch/arm/mach-at91/generic.h
793 #ifndef _AT91_GENERIC_H
794 #define _AT91_GENERIC_H
796 -#include <linux/of.h>
797 -#include <linux/reboot.h>
800 -extern void __init at91_map_io(void);
801 -extern void __init at91_alt_map_io(void);
804 -extern void at91rm9200_idle(void);
805 -extern void at91sam9_idle(void);
808 extern void __init at91rm9200_pm_init(void);
809 extern void __init at91sam9260_pm_init(void);
810 extern void __init at91sam9g45_pm_init(void);
811 extern void __init at91sam9x5_pm_init(void);
812 +extern void __init sama5_pm_init(void);
814 static inline void __init at91rm9200_pm_init(void) { }
815 static inline void __init at91sam9260_pm_init(void) { }
816 static inline void __init at91sam9g45_pm_init(void) { }
817 static inline void __init at91sam9x5_pm_init(void) { }
818 +static inline void __init sama5_pm_init(void) { }
821 #endif /* _AT91_GENERIC_H */
822 diff --git a/arch/arm/mach-at91/pm.c b/arch/arm/mach-at91/pm.c
823 index d687f860a2da..ab53332a9637 100644
824 --- a/arch/arm/mach-at91/pm.c
825 +++ b/arch/arm/mach-at91/pm.c
827 #include <asm/mach/irq.h>
828 #include <asm/fncpy.h>
829 #include <asm/cacheflush.h>
830 +#include <asm/system_misc.h>
835 +static void __iomem *pmc;
838 * FIXME: this is needed to communicate between the pinctrl driver and
839 * the PM implementation in the machine. Possibly part of the PM
840 @@ -87,7 +90,7 @@ static int at91_pm_verify_clocks(void)
844 - scsr = at91_pmc_read(AT91_PMC_SCSR);
845 + scsr = readl(pmc + AT91_PMC_SCSR);
847 /* USB must not be using PLLB */
848 if ((scsr & at91_pm_data.uhp_udp_mask) != 0) {
849 @@ -101,8 +104,7 @@ static int at91_pm_verify_clocks(void)
851 if ((scsr & (AT91_PMC_PCK0 << i)) == 0)
854 - css = at91_pmc_read(AT91_PMC_PCKR(i)) & AT91_PMC_CSS;
855 + css = readl(pmc + AT91_PMC_PCKR(i)) & AT91_PMC_CSS;
856 if (css != AT91_PMC_CSS_SLOW) {
857 pr_err("AT91: PM - Suspend-to-RAM with PCK%d src %d\n", i, css);
859 @@ -145,8 +147,8 @@ static void at91_pm_suspend(suspend_state_t state)
863 - at91_suspend_sram_fn(at91_pmc_base, at91_ramc_base[0],
864 - at91_ramc_base[1], pm_data);
865 + at91_suspend_sram_fn(pmc, at91_ramc_base[0],
866 + at91_ramc_base[1], pm_data);
870 @@ -369,6 +371,21 @@ static __init void at91_dt_ramc(void)
871 at91_pm_set_standby(standby);
874 +void at91rm9200_idle(void)
877 + * Disable the processor clock. The processor will be automatically
878 + * re-enabled by an interrupt or by a reset.
880 + writel(AT91_PMC_PCK, pmc + AT91_PMC_SCDR);
883 +void at91sam9_idle(void)
885 + writel(AT91_PMC_PCK, pmc + AT91_PMC_SCDR);
889 static void __init at91_pm_sram_init(void)
891 struct gen_pool *sram_pool;
892 @@ -415,13 +432,36 @@ static void __init at91_pm_sram_init(void)
893 &at91_pm_suspend_in_sram, at91_pm_suspend_in_sram_sz);
896 -static void __init at91_pm_init(void)
897 +static const struct of_device_id atmel_pmc_ids[] __initconst = {
898 + { .compatible = "atmel,at91rm9200-pmc" },
899 + { .compatible = "atmel,at91sam9260-pmc" },
900 + { .compatible = "atmel,at91sam9g45-pmc" },
901 + { .compatible = "atmel,at91sam9n12-pmc" },
902 + { .compatible = "atmel,at91sam9x5-pmc" },
903 + { .compatible = "atmel,sama5d3-pmc" },
904 + { .compatible = "atmel,sama5d2-pmc" },
905 + { /* sentinel */ },
908 +static void __init at91_pm_init(void (*pm_idle)(void))
910 - at91_pm_sram_init();
911 + struct device_node *pmc_np;
913 if (at91_cpuidle_device.dev.platform_data)
914 platform_device_register(&at91_cpuidle_device);
916 + pmc_np = of_find_matching_node(NULL, atmel_pmc_ids);
917 + pmc = of_iomap(pmc_np, 0);
919 + pr_err("AT91: PM not supported, PMC not found\n");
924 + arm_pm_idle = pm_idle;
926 + at91_pm_sram_init();
928 if (at91_suspend_sram_fn)
929 suspend_set_ops(&at91_pm_ops);
931 @@ -440,7 +480,7 @@ void __init at91rm9200_pm_init(void)
932 at91_pm_data.uhp_udp_mask = AT91RM9200_PMC_UHP | AT91RM9200_PMC_UDP;
933 at91_pm_data.memctrl = AT91_MEMCTRL_MC;
936 + at91_pm_init(at91rm9200_idle);
939 void __init at91sam9260_pm_init(void)
940 @@ -448,7 +488,7 @@ void __init at91sam9260_pm_init(void)
942 at91_pm_data.memctrl = AT91_MEMCTRL_SDRAMC;
943 at91_pm_data.uhp_udp_mask = AT91SAM926x_PMC_UHP | AT91SAM926x_PMC_UDP;
944 - return at91_pm_init();
945 + at91_pm_init(at91sam9_idle);
948 void __init at91sam9g45_pm_init(void)
949 @@ -456,7 +496,7 @@ void __init at91sam9g45_pm_init(void)
951 at91_pm_data.uhp_udp_mask = AT91SAM926x_PMC_UHP;
952 at91_pm_data.memctrl = AT91_MEMCTRL_DDRSDR;
953 - return at91_pm_init();
954 + at91_pm_init(at91sam9_idle);
957 void __init at91sam9x5_pm_init(void)
958 @@ -464,5 +504,13 @@ void __init at91sam9x5_pm_init(void)
960 at91_pm_data.uhp_udp_mask = AT91SAM926x_PMC_UHP | AT91SAM926x_PMC_UDP;
961 at91_pm_data.memctrl = AT91_MEMCTRL_DDRSDR;
962 - return at91_pm_init();
963 + at91_pm_init(at91sam9_idle);
966 +void __init sama5_pm_init(void)
969 + at91_pm_data.uhp_udp_mask = AT91SAM926x_PMC_UHP | AT91SAM926x_PMC_UDP;
970 + at91_pm_data.memctrl = AT91_MEMCTRL_DDRSDR;
971 + at91_pm_init(NULL);
973 diff --git a/arch/arm/mach-at91/sama5.c b/arch/arm/mach-at91/sama5.c
974 index d9cf6799aec0..df8fdf1cf66d 100644
975 --- a/arch/arm/mach-at91/sama5.c
976 +++ b/arch/arm/mach-at91/sama5.c
977 @@ -51,7 +51,7 @@ static void __init sama5_dt_device_init(void)
978 soc_dev = soc_device_to_device(soc);
980 of_platform_populate(NULL, of_default_bus_match_table, NULL, soc_dev);
981 - at91sam9x5_pm_init();
985 static const char *const sama5_dt_board_compat[] __initconst = {
986 diff --git a/arch/arm/mach-exynos/platsmp.c b/arch/arm/mach-exynos/platsmp.c
987 index 98a2c0cbb833..310dce500d3e 100644
988 --- a/arch/arm/mach-exynos/platsmp.c
989 +++ b/arch/arm/mach-exynos/platsmp.c
990 @@ -230,7 +230,7 @@ static void __iomem *scu_base_addr(void)
991 return (void __iomem *)(S5P_VA_SCU);
994 -static DEFINE_SPINLOCK(boot_lock);
995 +static DEFINE_RAW_SPINLOCK(boot_lock);
997 static void exynos_secondary_init(unsigned int cpu)
999 @@ -243,8 +243,8 @@ static void exynos_secondary_init(unsigned int cpu)
1001 * Synchronise with the boot thread.
1003 - spin_lock(&boot_lock);
1004 - spin_unlock(&boot_lock);
1005 + raw_spin_lock(&boot_lock);
1006 + raw_spin_unlock(&boot_lock);
1009 int exynos_set_boot_addr(u32 core_id, unsigned long boot_addr)
1010 @@ -308,7 +308,7 @@ static int exynos_boot_secondary(unsigned int cpu, struct task_struct *idle)
1011 * Set synchronisation state between this boot processor
1012 * and the secondary one
1014 - spin_lock(&boot_lock);
1015 + raw_spin_lock(&boot_lock);
1018 * The secondary processor is waiting to be released from
1019 @@ -335,7 +335,7 @@ static int exynos_boot_secondary(unsigned int cpu, struct task_struct *idle)
1022 printk(KERN_ERR "cpu1 power enable failed");
1023 - spin_unlock(&boot_lock);
1024 + raw_spin_unlock(&boot_lock);
1028 @@ -381,7 +381,7 @@ static int exynos_boot_secondary(unsigned int cpu, struct task_struct *idle)
1029 * calibrations, then wait for it to finish
1032 - spin_unlock(&boot_lock);
1033 + raw_spin_unlock(&boot_lock);
1035 return pen_release != -1 ? ret : 0;
1037 diff --git a/arch/arm/mach-hisi/platmcpm.c b/arch/arm/mach-hisi/platmcpm.c
1038 index b5f8f5ffda79..9753a84df9c4 100644
1039 --- a/arch/arm/mach-hisi/platmcpm.c
1040 +++ b/arch/arm/mach-hisi/platmcpm.c
1043 static void __iomem *sysctrl, *fabric;
1044 static int hip04_cpu_table[HIP04_MAX_CLUSTERS][HIP04_MAX_CPUS_PER_CLUSTER];
1045 -static DEFINE_SPINLOCK(boot_lock);
1046 +static DEFINE_RAW_SPINLOCK(boot_lock);
1047 static u32 fabric_phys_addr;
1049 * [0]: bootwrapper physical address
1050 @@ -113,7 +113,7 @@ static int hip04_boot_secondary(unsigned int l_cpu, struct task_struct *idle)
1051 if (cluster >= HIP04_MAX_CLUSTERS || cpu >= HIP04_MAX_CPUS_PER_CLUSTER)
1054 - spin_lock_irq(&boot_lock);
1055 + raw_spin_lock_irq(&boot_lock);
1057 if (hip04_cpu_table[cluster][cpu])
1059 @@ -147,7 +147,7 @@ static int hip04_boot_secondary(unsigned int l_cpu, struct task_struct *idle)
1062 hip04_cpu_table[cluster][cpu]++;
1063 - spin_unlock_irq(&boot_lock);
1064 + raw_spin_unlock_irq(&boot_lock);
1068 @@ -162,11 +162,11 @@ static void hip04_cpu_die(unsigned int l_cpu)
1069 cpu = MPIDR_AFFINITY_LEVEL(mpidr, 0);
1070 cluster = MPIDR_AFFINITY_LEVEL(mpidr, 1);
1072 - spin_lock(&boot_lock);
1073 + raw_spin_lock(&boot_lock);
1074 hip04_cpu_table[cluster][cpu]--;
1075 if (hip04_cpu_table[cluster][cpu] == 1) {
1076 /* A power_up request went ahead of us. */
1077 - spin_unlock(&boot_lock);
1078 + raw_spin_unlock(&boot_lock);
1080 } else if (hip04_cpu_table[cluster][cpu] > 1) {
1081 pr_err("Cluster %d CPU%d boots multiple times\n", cluster, cpu);
1082 @@ -174,7 +174,7 @@ static void hip04_cpu_die(unsigned int l_cpu)
1085 last_man = hip04_cluster_is_down(cluster);
1086 - spin_unlock(&boot_lock);
1087 + raw_spin_unlock(&boot_lock);
1089 /* Since it's Cortex A15, disable L2 prefetching. */
1091 @@ -203,7 +203,7 @@ static int hip04_cpu_kill(unsigned int l_cpu)
1092 cpu >= HIP04_MAX_CPUS_PER_CLUSTER);
1094 count = TIMEOUT_MSEC / POLL_MSEC;
1095 - spin_lock_irq(&boot_lock);
1096 + raw_spin_lock_irq(&boot_lock);
1097 for (tries = 0; tries < count; tries++) {
1098 if (hip04_cpu_table[cluster][cpu])
1100 @@ -211,10 +211,10 @@ static int hip04_cpu_kill(unsigned int l_cpu)
1101 data = readl_relaxed(sysctrl + SC_CPU_RESET_STATUS(cluster));
1102 if (data & CORE_WFI_STATUS(cpu))
1104 - spin_unlock_irq(&boot_lock);
1105 + raw_spin_unlock_irq(&boot_lock);
1106 /* Wait for clean L2 when the whole cluster is down. */
1108 - spin_lock_irq(&boot_lock);
1109 + raw_spin_lock_irq(&boot_lock);
1113 @@ -231,10 +231,10 @@ static int hip04_cpu_kill(unsigned int l_cpu)
1115 if (hip04_cluster_is_down(cluster))
1116 hip04_set_snoop_filter(cluster, 0);
1117 - spin_unlock_irq(&boot_lock);
1118 + raw_spin_unlock_irq(&boot_lock);
1121 - spin_unlock_irq(&boot_lock);
1122 + raw_spin_unlock_irq(&boot_lock);
1126 diff --git a/arch/arm/mach-imx/Kconfig b/arch/arm/mach-imx/Kconfig
1127 index 8ceda2844c4f..08bcf8fb76f2 100644
1128 --- a/arch/arm/mach-imx/Kconfig
1129 +++ b/arch/arm/mach-imx/Kconfig
1130 @@ -524,7 +524,7 @@ config SOC_IMX6Q
1131 bool "i.MX6 Quad/DualLite support"
1132 select ARM_ERRATA_764369 if SMP
1133 select HAVE_ARM_SCU if SMP
1134 - select HAVE_ARM_TWD if SMP
1135 + select HAVE_ARM_TWD
1136 select PCI_DOMAINS if PCI
1137 select PINCTRL_IMX6Q
1139 diff --git a/arch/arm/mach-omap2/omap-smp.c b/arch/arm/mach-omap2/omap-smp.c
1140 index 79e1f876d1c9..7e625c17f78e 100644
1141 --- a/arch/arm/mach-omap2/omap-smp.c
1142 +++ b/arch/arm/mach-omap2/omap-smp.c
1144 /* SCU base address */
1145 static void __iomem *scu_base;
1147 -static DEFINE_SPINLOCK(boot_lock);
1148 +static DEFINE_RAW_SPINLOCK(boot_lock);
1150 void __iomem *omap4_get_scu_base(void)
1152 @@ -74,8 +74,8 @@ static void omap4_secondary_init(unsigned int cpu)
1154 * Synchronise with the boot thread.
1156 - spin_lock(&boot_lock);
1157 - spin_unlock(&boot_lock);
1158 + raw_spin_lock(&boot_lock);
1159 + raw_spin_unlock(&boot_lock);
1162 static int omap4_boot_secondary(unsigned int cpu, struct task_struct *idle)
1163 @@ -89,7 +89,7 @@ static int omap4_boot_secondary(unsigned int cpu, struct task_struct *idle)
1164 * Set synchronisation state between this boot processor
1165 * and the secondary one
1167 - spin_lock(&boot_lock);
1168 + raw_spin_lock(&boot_lock);
1171 * Update the AuxCoreBoot0 with boot state for secondary core.
1172 @@ -166,7 +166,7 @@ static int omap4_boot_secondary(unsigned int cpu, struct task_struct *idle)
1173 * Now the secondary core is starting up let it run its
1174 * calibrations, then wait for it to finish
1176 - spin_unlock(&boot_lock);
1177 + raw_spin_unlock(&boot_lock);
1181 diff --git a/arch/arm/mach-prima2/platsmp.c b/arch/arm/mach-prima2/platsmp.c
1182 index e46c91094dde..dcb3ed0c26da 100644
1183 --- a/arch/arm/mach-prima2/platsmp.c
1184 +++ b/arch/arm/mach-prima2/platsmp.c
1187 static void __iomem *clk_base;
1189 -static DEFINE_SPINLOCK(boot_lock);
1190 +static DEFINE_RAW_SPINLOCK(boot_lock);
1192 static void sirfsoc_secondary_init(unsigned int cpu)
1194 @@ -36,8 +36,8 @@ static void sirfsoc_secondary_init(unsigned int cpu)
1196 * Synchronise with the boot thread.
1198 - spin_lock(&boot_lock);
1199 - spin_unlock(&boot_lock);
1200 + raw_spin_lock(&boot_lock);
1201 + raw_spin_unlock(&boot_lock);
1204 static const struct of_device_id clk_ids[] = {
1205 @@ -75,7 +75,7 @@ static int sirfsoc_boot_secondary(unsigned int cpu, struct task_struct *idle)
1206 /* make sure write buffer is drained */
1209 - spin_lock(&boot_lock);
1210 + raw_spin_lock(&boot_lock);
1213 * The secondary processor is waiting to be released from
1214 @@ -107,7 +107,7 @@ static int sirfsoc_boot_secondary(unsigned int cpu, struct task_struct *idle)
1215 * now the secondary core is starting up let it run its
1216 * calibrations, then wait for it to finish
1218 - spin_unlock(&boot_lock);
1219 + raw_spin_unlock(&boot_lock);
1221 return pen_release != -1 ? -ENOSYS : 0;
1223 diff --git a/arch/arm/mach-qcom/platsmp.c b/arch/arm/mach-qcom/platsmp.c
1224 index 9b00123a315d..0a49fe1bc8cf 100644
1225 --- a/arch/arm/mach-qcom/platsmp.c
1226 +++ b/arch/arm/mach-qcom/platsmp.c
1229 extern void secondary_startup_arm(void);
1231 -static DEFINE_SPINLOCK(boot_lock);
1232 +static DEFINE_RAW_SPINLOCK(boot_lock);
1234 #ifdef CONFIG_HOTPLUG_CPU
1235 static void qcom_cpu_die(unsigned int cpu)
1236 @@ -60,8 +60,8 @@ static void qcom_secondary_init(unsigned int cpu)
1238 * Synchronise with the boot thread.
1240 - spin_lock(&boot_lock);
1241 - spin_unlock(&boot_lock);
1242 + raw_spin_lock(&boot_lock);
1243 + raw_spin_unlock(&boot_lock);
1246 static int scss_release_secondary(unsigned int cpu)
1247 @@ -284,7 +284,7 @@ static int qcom_boot_secondary(unsigned int cpu, int (*func)(unsigned int))
1248 * set synchronisation state between this boot processor
1249 * and the secondary one
1251 - spin_lock(&boot_lock);
1252 + raw_spin_lock(&boot_lock);
1255 * Send the secondary CPU a soft interrupt, thereby causing
1256 @@ -297,7 +297,7 @@ static int qcom_boot_secondary(unsigned int cpu, int (*func)(unsigned int))
1257 * now the secondary core is starting up let it run its
1258 * calibrations, then wait for it to finish
1260 - spin_unlock(&boot_lock);
1261 + raw_spin_unlock(&boot_lock);
1265 diff --git a/arch/arm/mach-spear/platsmp.c b/arch/arm/mach-spear/platsmp.c
1266 index fd4297713d67..b0553b2c2d53 100644
1267 --- a/arch/arm/mach-spear/platsmp.c
1268 +++ b/arch/arm/mach-spear/platsmp.c
1269 @@ -32,7 +32,7 @@ static void write_pen_release(int val)
1270 sync_cache_w(&pen_release);
1273 -static DEFINE_SPINLOCK(boot_lock);
1274 +static DEFINE_RAW_SPINLOCK(boot_lock);
1276 static void __iomem *scu_base = IOMEM(VA_SCU_BASE);
1278 @@ -47,8 +47,8 @@ static void spear13xx_secondary_init(unsigned int cpu)
1280 * Synchronise with the boot thread.
1282 - spin_lock(&boot_lock);
1283 - spin_unlock(&boot_lock);
1284 + raw_spin_lock(&boot_lock);
1285 + raw_spin_unlock(&boot_lock);
1288 static int spear13xx_boot_secondary(unsigned int cpu, struct task_struct *idle)
1289 @@ -59,7 +59,7 @@ static int spear13xx_boot_secondary(unsigned int cpu, struct task_struct *idle)
1290 * set synchronisation state between this boot processor
1291 * and the secondary one
1293 - spin_lock(&boot_lock);
1294 + raw_spin_lock(&boot_lock);
1297 * The secondary processor is waiting to be released from
1298 @@ -84,7 +84,7 @@ static int spear13xx_boot_secondary(unsigned int cpu, struct task_struct *idle)
1299 * now the secondary core is starting up let it run its
1300 * calibrations, then wait for it to finish
1302 - spin_unlock(&boot_lock);
1303 + raw_spin_unlock(&boot_lock);
1305 return pen_release != -1 ? -ENOSYS : 0;
1307 diff --git a/arch/arm/mach-sti/platsmp.c b/arch/arm/mach-sti/platsmp.c
1308 index c4ad6eae67fa..e830b20b212f 100644
1309 --- a/arch/arm/mach-sti/platsmp.c
1310 +++ b/arch/arm/mach-sti/platsmp.c
1311 @@ -35,7 +35,7 @@ static void write_pen_release(int val)
1312 sync_cache_w(&pen_release);
1315 -static DEFINE_SPINLOCK(boot_lock);
1316 +static DEFINE_RAW_SPINLOCK(boot_lock);
1318 static void sti_secondary_init(unsigned int cpu)
1320 @@ -48,8 +48,8 @@ static void sti_secondary_init(unsigned int cpu)
1322 * Synchronise with the boot thread.
1324 - spin_lock(&boot_lock);
1325 - spin_unlock(&boot_lock);
1326 + raw_spin_lock(&boot_lock);
1327 + raw_spin_unlock(&boot_lock);
1330 static int sti_boot_secondary(unsigned int cpu, struct task_struct *idle)
1331 @@ -60,7 +60,7 @@ static int sti_boot_secondary(unsigned int cpu, struct task_struct *idle)
1332 * set synchronisation state between this boot processor
1333 * and the secondary one
1335 - spin_lock(&boot_lock);
1336 + raw_spin_lock(&boot_lock);
1339 * The secondary processor is waiting to be released from
1340 @@ -91,7 +91,7 @@ static int sti_boot_secondary(unsigned int cpu, struct task_struct *idle)
1341 * now the secondary core is starting up let it run its
1342 * calibrations, then wait for it to finish
1344 - spin_unlock(&boot_lock);
1345 + raw_spin_unlock(&boot_lock);
1347 return pen_release != -1 ? -ENOSYS : 0;
1349 diff --git a/arch/arm/mm/fault.c b/arch/arm/mm/fault.c
1350 index c095455d496e..276ac73f8650 100644
1351 --- a/arch/arm/mm/fault.c
1352 +++ b/arch/arm/mm/fault.c
1353 @@ -430,6 +430,9 @@ do_translation_fault(unsigned long addr, unsigned int fsr,
1354 if (addr < TASK_SIZE)
1355 return do_page_fault(addr, fsr, regs);
1357 + if (interrupts_enabled(regs))
1358 + local_irq_enable();
1360 if (user_mode(regs))
1363 @@ -497,6 +500,9 @@ do_translation_fault(unsigned long addr, unsigned int fsr,
1365 do_sect_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
1367 + if (interrupts_enabled(regs))
1368 + local_irq_enable();
1370 do_bad_area(addr, fsr, regs);
1373 diff --git a/arch/arm/mm/highmem.c b/arch/arm/mm/highmem.c
1374 index d02f8187b1cc..542692dbd40a 100644
1375 --- a/arch/arm/mm/highmem.c
1376 +++ b/arch/arm/mm/highmem.c
1377 @@ -34,6 +34,11 @@ static inline pte_t get_fixmap_pte(unsigned long vaddr)
1381 +static unsigned int fixmap_idx(int type)
1383 + return FIX_KMAP_BEGIN + type + KM_TYPE_NR * smp_processor_id();
1386 void *kmap(struct page *page)
1389 @@ -54,12 +59,13 @@ EXPORT_SYMBOL(kunmap);
1391 void *kmap_atomic(struct page *page)
1393 + pte_t pte = mk_pte(page, kmap_prot);
1395 unsigned long vaddr;
1399 - preempt_disable();
1400 + preempt_disable_nort();
1401 pagefault_disable();
1402 if (!PageHighMem(page))
1403 return page_address(page);
1404 @@ -79,7 +85,7 @@ void *kmap_atomic(struct page *page)
1406 type = kmap_atomic_idx_push();
1408 - idx = FIX_KMAP_BEGIN + type + KM_TYPE_NR * smp_processor_id();
1409 + idx = fixmap_idx(type);
1410 vaddr = __fix_to_virt(idx);
1411 #ifdef CONFIG_DEBUG_HIGHMEM
1413 @@ -93,7 +99,10 @@ void *kmap_atomic(struct page *page)
1414 * in place, so the contained TLB flush ensures the TLB is updated
1415 * with the new mapping.
1417 - set_fixmap_pte(idx, mk_pte(page, kmap_prot));
1418 +#ifdef CONFIG_PREEMPT_RT_FULL
1419 + current->kmap_pte[type] = pte;
1421 + set_fixmap_pte(idx, pte);
1423 return (void *)vaddr;
1425 @@ -106,44 +115,75 @@ void __kunmap_atomic(void *kvaddr)
1427 if (kvaddr >= (void *)FIXADDR_START) {
1428 type = kmap_atomic_idx();
1429 - idx = FIX_KMAP_BEGIN + type + KM_TYPE_NR * smp_processor_id();
1430 + idx = fixmap_idx(type);
1432 if (cache_is_vivt())
1433 __cpuc_flush_dcache_area((void *)vaddr, PAGE_SIZE);
1434 +#ifdef CONFIG_PREEMPT_RT_FULL
1435 + current->kmap_pte[type] = __pte(0);
1437 #ifdef CONFIG_DEBUG_HIGHMEM
1438 BUG_ON(vaddr != __fix_to_virt(idx));
1439 - set_fixmap_pte(idx, __pte(0));
1441 (void) idx; /* to kill a warning */
1443 + set_fixmap_pte(idx, __pte(0));
1444 kmap_atomic_idx_pop();
1445 } else if (vaddr >= PKMAP_ADDR(0) && vaddr < PKMAP_ADDR(LAST_PKMAP)) {
1446 /* this address was obtained through kmap_high_get() */
1447 kunmap_high(pte_page(pkmap_page_table[PKMAP_NR(vaddr)]));
1451 + preempt_enable_nort();
1453 EXPORT_SYMBOL(__kunmap_atomic);
1455 void *kmap_atomic_pfn(unsigned long pfn)
1457 + pte_t pte = pfn_pte(pfn, kmap_prot);
1458 unsigned long vaddr;
1460 struct page *page = pfn_to_page(pfn);
1462 - preempt_disable();
1463 + preempt_disable_nort();
1464 pagefault_disable();
1465 if (!PageHighMem(page))
1466 return page_address(page);
1468 type = kmap_atomic_idx_push();
1469 - idx = FIX_KMAP_BEGIN + type + KM_TYPE_NR * smp_processor_id();
1470 + idx = fixmap_idx(type);
1471 vaddr = __fix_to_virt(idx);
1472 #ifdef CONFIG_DEBUG_HIGHMEM
1473 BUG_ON(!pte_none(get_fixmap_pte(vaddr)));
1475 - set_fixmap_pte(idx, pfn_pte(pfn, kmap_prot));
1476 +#ifdef CONFIG_PREEMPT_RT_FULL
1477 + current->kmap_pte[type] = pte;
1479 + set_fixmap_pte(idx, pte);
1481 return (void *)vaddr;
1483 +#if defined CONFIG_PREEMPT_RT_FULL
1484 +void switch_kmaps(struct task_struct *prev_p, struct task_struct *next_p)
1489 + * Clear @prev's kmap_atomic mappings
1491 + for (i = 0; i < prev_p->kmap_idx; i++) {
1492 + int idx = fixmap_idx(i);
1494 + set_fixmap_pte(idx, __pte(0));
1497 + * Restore @next_p's kmap_atomic mappings
1499 + for (i = 0; i < next_p->kmap_idx; i++) {
1500 + int idx = fixmap_idx(i);
1502 + if (!pte_none(next_p->kmap_pte[i]))
1503 + set_fixmap_pte(idx, next_p->kmap_pte[i]);
1507 diff --git a/arch/arm/plat-versatile/platsmp.c b/arch/arm/plat-versatile/platsmp.c
1508 index 53feb90c840c..b4a8d54fc3f3 100644
1509 --- a/arch/arm/plat-versatile/platsmp.c
1510 +++ b/arch/arm/plat-versatile/platsmp.c
1511 @@ -30,7 +30,7 @@ static void write_pen_release(int val)
1512 sync_cache_w(&pen_release);
1515 -static DEFINE_SPINLOCK(boot_lock);
1516 +static DEFINE_RAW_SPINLOCK(boot_lock);
1518 void versatile_secondary_init(unsigned int cpu)
1520 @@ -43,8 +43,8 @@ void versatile_secondary_init(unsigned int cpu)
1522 * Synchronise with the boot thread.
1524 - spin_lock(&boot_lock);
1525 - spin_unlock(&boot_lock);
1526 + raw_spin_lock(&boot_lock);
1527 + raw_spin_unlock(&boot_lock);
1530 int versatile_boot_secondary(unsigned int cpu, struct task_struct *idle)
1531 @@ -55,7 +55,7 @@ int versatile_boot_secondary(unsigned int cpu, struct task_struct *idle)
1532 * Set synchronisation state between this boot processor
1533 * and the secondary one
1535 - spin_lock(&boot_lock);
1536 + raw_spin_lock(&boot_lock);
1539 * This is really belt and braces; we hold unintended secondary
1540 @@ -85,7 +85,7 @@ int versatile_boot_secondary(unsigned int cpu, struct task_struct *idle)
1541 * now the secondary core is starting up let it run its
1542 * calibrations, then wait for it to finish
1544 - spin_unlock(&boot_lock);
1545 + raw_spin_unlock(&boot_lock);
1547 return pen_release != -1 ? -ENOSYS : 0;
1549 diff --git a/arch/arm64/Kconfig b/arch/arm64/Kconfig
1550 index 14cdc6dea493..9196cf82f7be 100644
1551 --- a/arch/arm64/Kconfig
1552 +++ b/arch/arm64/Kconfig
1553 @@ -76,6 +76,7 @@ config ARM64
1554 select HAVE_PERF_REGS
1555 select HAVE_PERF_USER_STACK_DUMP
1556 select HAVE_RCU_TABLE_FREE
1557 + select HAVE_PREEMPT_LAZY
1558 select HAVE_SYSCALL_TRACEPOINTS
1559 select IOMMU_DMA if IOMMU_SUPPORT
1561 @@ -582,7 +583,7 @@ config XEN_DOM0
1564 bool "Xen guest support on ARM64"
1565 - depends on ARM64 && OF
1566 + depends on ARM64 && OF && !PREEMPT_RT_FULL
1569 Say Y if you want to run Linux in a Virtual Machine on Xen on ARM64.
1570 diff --git a/arch/arm64/include/asm/thread_info.h b/arch/arm64/include/asm/thread_info.h
1571 index 90c7ff233735..5f4e89fbc290 100644
1572 --- a/arch/arm64/include/asm/thread_info.h
1573 +++ b/arch/arm64/include/asm/thread_info.h
1574 @@ -49,6 +49,7 @@ struct thread_info {
1575 mm_segment_t addr_limit; /* address limit */
1576 struct task_struct *task; /* main task structure */
1577 int preempt_count; /* 0 => preemptable, <0 => bug */
1578 + int preempt_lazy_count; /* 0 => preemptable, <0 => bug */
1582 @@ -103,6 +104,7 @@ static inline struct thread_info *current_thread_info(void)
1583 #define TIF_NEED_RESCHED 1
1584 #define TIF_NOTIFY_RESUME 2 /* callback before returning to user */
1585 #define TIF_FOREIGN_FPSTATE 3 /* CPU's FP state is not current's */
1586 +#define TIF_NEED_RESCHED_LAZY 4
1588 #define TIF_SYSCALL_TRACE 8
1589 #define TIF_SYSCALL_AUDIT 9
1590 @@ -118,6 +120,7 @@ static inline struct thread_info *current_thread_info(void)
1591 #define _TIF_NEED_RESCHED (1 << TIF_NEED_RESCHED)
1592 #define _TIF_NOTIFY_RESUME (1 << TIF_NOTIFY_RESUME)
1593 #define _TIF_FOREIGN_FPSTATE (1 << TIF_FOREIGN_FPSTATE)
1594 +#define _TIF_NEED_RESCHED_LAZY (1 << TIF_NEED_RESCHED_LAZY)
1595 #define _TIF_NOHZ (1 << TIF_NOHZ)
1596 #define _TIF_SYSCALL_TRACE (1 << TIF_SYSCALL_TRACE)
1597 #define _TIF_SYSCALL_AUDIT (1 << TIF_SYSCALL_AUDIT)
1598 @@ -126,7 +129,8 @@ static inline struct thread_info *current_thread_info(void)
1599 #define _TIF_32BIT (1 << TIF_32BIT)
1601 #define _TIF_WORK_MASK (_TIF_NEED_RESCHED | _TIF_SIGPENDING | \
1602 - _TIF_NOTIFY_RESUME | _TIF_FOREIGN_FPSTATE)
1603 + _TIF_NOTIFY_RESUME | _TIF_FOREIGN_FPSTATE | \
1604 + _TIF_NEED_RESCHED_LAZY)
1606 #define _TIF_SYSCALL_WORK (_TIF_SYSCALL_TRACE | _TIF_SYSCALL_AUDIT | \
1607 _TIF_SYSCALL_TRACEPOINT | _TIF_SECCOMP | \
1608 diff --git a/arch/arm64/kernel/asm-offsets.c b/arch/arm64/kernel/asm-offsets.c
1609 index 087cf9a65359..d74475928399 100644
1610 --- a/arch/arm64/kernel/asm-offsets.c
1611 +++ b/arch/arm64/kernel/asm-offsets.c
1612 @@ -35,6 +35,7 @@ int main(void)
1614 DEFINE(TI_FLAGS, offsetof(struct thread_info, flags));
1615 DEFINE(TI_PREEMPT, offsetof(struct thread_info, preempt_count));
1616 + DEFINE(TI_PREEMPT_LAZY, offsetof(struct thread_info, preempt_lazy_count));
1617 DEFINE(TI_ADDR_LIMIT, offsetof(struct thread_info, addr_limit));
1618 DEFINE(TI_TASK, offsetof(struct thread_info, task));
1619 DEFINE(TI_CPU, offsetof(struct thread_info, cpu));
1620 diff --git a/arch/arm64/kernel/entry.S b/arch/arm64/kernel/entry.S
1621 index bd14849beb73..cf92d4ee51db 100644
1622 --- a/arch/arm64/kernel/entry.S
1623 +++ b/arch/arm64/kernel/entry.S
1624 @@ -376,11 +376,16 @@ el1_irq:
1625 #ifdef CONFIG_PREEMPT
1627 ldr w24, [tsk, #TI_PREEMPT] // get preempt count
1628 - cbnz w24, 1f // preempt count != 0
1629 + cbnz w24, 2f // preempt count != 0
1630 ldr x0, [tsk, #TI_FLAGS] // get flags
1631 - tbz x0, #TIF_NEED_RESCHED, 1f // needs rescheduling?
1633 + tbnz x0, #TIF_NEED_RESCHED, 1f // needs rescheduling?
1635 + ldr w24, [tsk, #TI_PREEMPT_LAZY] // get preempt lazy count
1636 + cbnz w24, 2f // preempt lazy count != 0
1637 + tbz x0, #TIF_NEED_RESCHED_LAZY, 2f // needs rescheduling?
1642 #ifdef CONFIG_TRACE_IRQFLAGS
1643 bl trace_hardirqs_on
1644 @@ -394,6 +399,7 @@ el1_preempt:
1645 1: bl preempt_schedule_irq // irq en/disable is done inside
1646 ldr x0, [tsk, #TI_FLAGS] // get new tasks TI_FLAGS
1647 tbnz x0, #TIF_NEED_RESCHED, 1b // needs rescheduling?
1648 + tbnz x0, #TIF_NEED_RESCHED_LAZY, 1b // needs rescheduling?
1652 @@ -638,6 +644,7 @@ ret_fast_syscall_trace:
1655 tbnz x1, #TIF_NEED_RESCHED, work_resched
1656 + tbnz x1, #TIF_NEED_RESCHED_LAZY, work_resched
1657 /* TIF_SIGPENDING, TIF_NOTIFY_RESUME or TIF_FOREIGN_FPSTATE case */
1658 ldr x2, [sp, #S_PSTATE]
1659 mov x0, sp // 'regs'
1660 diff --git a/arch/mips/Kconfig b/arch/mips/Kconfig
1661 index 8b0424abc84c..5422d4c0bbdf 100644
1662 --- a/arch/mips/Kconfig
1663 +++ b/arch/mips/Kconfig
1664 @@ -2411,7 +2411,7 @@ config CPU_R4400_WORKAROUNDS
1667 bool "High Memory Support"
1668 - depends on 32BIT && CPU_SUPPORTS_HIGHMEM && SYS_SUPPORTS_HIGHMEM && !CPU_MIPS32_3_5_EVA
1669 + depends on 32BIT && CPU_SUPPORTS_HIGHMEM && SYS_SUPPORTS_HIGHMEM && !CPU_MIPS32_3_5_EVA && !PREEMPT_RT_FULL
1671 config CPU_SUPPORTS_HIGHMEM
1673 diff --git a/arch/mips/kvm/mips.c b/arch/mips/kvm/mips.c
1674 index a017b23ee4aa..8d4d9270140f 100644
1675 --- a/arch/mips/kvm/mips.c
1676 +++ b/arch/mips/kvm/mips.c
1677 @@ -454,8 +454,8 @@ int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1679 dvcpu->arch.wait = 0;
1681 - if (waitqueue_active(&dvcpu->wq))
1682 - wake_up_interruptible(&dvcpu->wq);
1683 + if (swait_active(&dvcpu->wq))
1684 + swake_up(&dvcpu->wq);
1688 @@ -1183,8 +1183,8 @@ static void kvm_mips_comparecount_func(unsigned long data)
1689 kvm_mips_callbacks->queue_timer_int(vcpu);
1691 vcpu->arch.wait = 0;
1692 - if (waitqueue_active(&vcpu->wq))
1693 - wake_up_interruptible(&vcpu->wq);
1694 + if (swait_active(&vcpu->wq))
1695 + swake_up(&vcpu->wq);
1698 /* low level hrtimer wake routine */
1699 diff --git a/arch/powerpc/Kconfig b/arch/powerpc/Kconfig
1700 index db49e0d796b1..1d2be228661c 100644
1701 --- a/arch/powerpc/Kconfig
1702 +++ b/arch/powerpc/Kconfig
1703 @@ -60,10 +60,11 @@ config LOCKDEP_SUPPORT
1705 config RWSEM_GENERIC_SPINLOCK
1707 + default y if PREEMPT_RT_FULL
1709 config RWSEM_XCHGADD_ALGORITHM
1712 + default y if !PREEMPT_RT_FULL
1714 config GENERIC_LOCKBREAK
1716 @@ -141,6 +142,7 @@ config PPC
1717 select ARCH_HAS_TICK_BROADCAST if GENERIC_CLOCKEVENTS_BROADCAST
1718 select GENERIC_STRNCPY_FROM_USER
1719 select GENERIC_STRNLEN_USER
1720 + select HAVE_PREEMPT_LAZY
1721 select HAVE_MOD_ARCH_SPECIFIC
1722 select MODULES_USE_ELF_RELA
1723 select CLONE_BACKWARDS
1724 @@ -319,7 +321,7 @@ menu "Kernel options"
1727 bool "High memory support"
1729 + depends on PPC32 && !PREEMPT_RT_FULL
1731 source kernel/Kconfig.hz
1732 source kernel/Kconfig.preempt
1733 diff --git a/arch/powerpc/include/asm/kvm_host.h b/arch/powerpc/include/asm/kvm_host.h
1734 index a92d95aee42d..20376580583f 100644
1735 --- a/arch/powerpc/include/asm/kvm_host.h
1736 +++ b/arch/powerpc/include/asm/kvm_host.h
1737 @@ -286,7 +286,7 @@ struct kvmppc_vcore {
1738 struct list_head runnable_threads;
1739 struct list_head preempt_list;
1741 - wait_queue_head_t wq;
1742 + struct swait_queue_head wq;
1743 spinlock_t stoltb_lock; /* protects stolen_tb and preempt_tb */
1746 @@ -627,7 +627,7 @@ struct kvm_vcpu_arch {
1750 - wait_queue_head_t *wqp;
1751 + struct swait_queue_head *wqp;
1752 struct kvmppc_vcore *vcore;
1755 diff --git a/arch/powerpc/include/asm/thread_info.h b/arch/powerpc/include/asm/thread_info.h
1756 index 7efee4a3240b..40e6fa1b85b2 100644
1757 --- a/arch/powerpc/include/asm/thread_info.h
1758 +++ b/arch/powerpc/include/asm/thread_info.h
1759 @@ -42,6 +42,8 @@ struct thread_info {
1760 int cpu; /* cpu we're on */
1761 int preempt_count; /* 0 => preemptable,
1763 + int preempt_lazy_count; /* 0 => preemptable,
1765 unsigned long local_flags; /* private flags for thread */
1767 /* low level flags - has atomic operations done on it */
1768 @@ -82,8 +84,7 @@ static inline struct thread_info *current_thread_info(void)
1769 #define TIF_SYSCALL_TRACE 0 /* syscall trace active */
1770 #define TIF_SIGPENDING 1 /* signal pending */
1771 #define TIF_NEED_RESCHED 2 /* rescheduling necessary */
1772 -#define TIF_POLLING_NRFLAG 3 /* true if poll_idle() is polling
1773 - TIF_NEED_RESCHED */
1774 +#define TIF_NEED_RESCHED_LAZY 3 /* lazy rescheduling necessary */
1775 #define TIF_32BIT 4 /* 32 bit binary */
1776 #define TIF_RESTORE_TM 5 /* need to restore TM FP/VEC/VSX */
1777 #define TIF_SYSCALL_AUDIT 7 /* syscall auditing active */
1778 @@ -101,6 +102,8 @@ static inline struct thread_info *current_thread_info(void)
1779 #if defined(CONFIG_PPC64)
1780 #define TIF_ELF2ABI 18 /* function descriptors must die! */
1782 +#define TIF_POLLING_NRFLAG 19 /* true if poll_idle() is polling
1783 + TIF_NEED_RESCHED */
1785 /* as above, but as bit values */
1786 #define _TIF_SYSCALL_TRACE (1<<TIF_SYSCALL_TRACE)
1787 @@ -119,14 +122,16 @@ static inline struct thread_info *current_thread_info(void)
1788 #define _TIF_SYSCALL_TRACEPOINT (1<<TIF_SYSCALL_TRACEPOINT)
1789 #define _TIF_EMULATE_STACK_STORE (1<<TIF_EMULATE_STACK_STORE)
1790 #define _TIF_NOHZ (1<<TIF_NOHZ)
1791 +#define _TIF_NEED_RESCHED_LAZY (1<<TIF_NEED_RESCHED_LAZY)
1792 #define _TIF_SYSCALL_DOTRACE (_TIF_SYSCALL_TRACE | _TIF_SYSCALL_AUDIT | \
1793 _TIF_SECCOMP | _TIF_SYSCALL_TRACEPOINT | \
1796 #define _TIF_USER_WORK_MASK (_TIF_SIGPENDING | _TIF_NEED_RESCHED | \
1797 _TIF_NOTIFY_RESUME | _TIF_UPROBE | \
1799 + _TIF_RESTORE_TM | _TIF_NEED_RESCHED_LAZY)
1800 #define _TIF_PERSYSCALL_MASK (_TIF_RESTOREALL|_TIF_NOERROR)
1801 +#define _TIF_NEED_RESCHED_MASK (_TIF_NEED_RESCHED | _TIF_NEED_RESCHED_LAZY)
1803 /* Bits in local_flags */
1804 /* Don't move TLF_NAPPING without adjusting the code in entry_32.S */
1805 diff --git a/arch/powerpc/kernel/asm-offsets.c b/arch/powerpc/kernel/asm-offsets.c
1806 index 40da69163d51..bd040815334b 100644
1807 --- a/arch/powerpc/kernel/asm-offsets.c
1808 +++ b/arch/powerpc/kernel/asm-offsets.c
1809 @@ -160,6 +160,7 @@ int main(void)
1810 DEFINE(TI_FLAGS, offsetof(struct thread_info, flags));
1811 DEFINE(TI_LOCAL_FLAGS, offsetof(struct thread_info, local_flags));
1812 DEFINE(TI_PREEMPT, offsetof(struct thread_info, preempt_count));
1813 + DEFINE(TI_PREEMPT_LAZY, offsetof(struct thread_info, preempt_lazy_count));
1814 DEFINE(TI_TASK, offsetof(struct thread_info, task));
1815 DEFINE(TI_CPU, offsetof(struct thread_info, cpu));
1817 diff --git a/arch/powerpc/kernel/entry_32.S b/arch/powerpc/kernel/entry_32.S
1818 index 2405631e91a2..c21b4b42eaa0 100644
1819 --- a/arch/powerpc/kernel/entry_32.S
1820 +++ b/arch/powerpc/kernel/entry_32.S
1821 @@ -818,7 +818,14 @@ resume_kernel:
1822 cmpwi 0,r0,0 /* if non-zero, just restore regs and return */
1824 andi. r8,r8,_TIF_NEED_RESCHED
1826 + lwz r0,TI_PREEMPT_LAZY(r9)
1827 + cmpwi 0,r0,0 /* if non-zero, just restore regs and return */
1829 + lwz r0,TI_FLAGS(r9)
1830 + andi. r0,r0,_TIF_NEED_RESCHED_LAZY
1834 andi. r0,r3,MSR_EE /* interrupts off? */
1835 beq restore /* don't schedule if so */
1836 @@ -829,11 +836,11 @@ resume_kernel:
1838 bl trace_hardirqs_off
1840 -1: bl preempt_schedule_irq
1841 +2: bl preempt_schedule_irq
1842 CURRENT_THREAD_INFO(r9, r1)
1844 - andi. r0,r3,_TIF_NEED_RESCHED
1846 + andi. r0,r3,_TIF_NEED_RESCHED_MASK
1848 #ifdef CONFIG_TRACE_IRQFLAGS
1849 /* And now, to properly rebalance the above, we tell lockdep they
1850 * are being turned back on, which will happen when we return
1851 @@ -1154,7 +1161,7 @@ global_dbcr0:
1852 #endif /* !(CONFIG_4xx || CONFIG_BOOKE) */
1854 do_work: /* r10 contains MSR_KERNEL here */
1855 - andi. r0,r9,_TIF_NEED_RESCHED
1856 + andi. r0,r9,_TIF_NEED_RESCHED_MASK
1859 do_resched: /* r10 contains MSR_KERNEL here */
1860 @@ -1175,7 +1182,7 @@ recheck:
1861 MTMSRD(r10) /* disable interrupts */
1862 CURRENT_THREAD_INFO(r9, r1)
1864 - andi. r0,r9,_TIF_NEED_RESCHED
1865 + andi. r0,r9,_TIF_NEED_RESCHED_MASK
1867 andi. r0,r9,_TIF_USER_WORK_MASK
1869 diff --git a/arch/powerpc/kernel/entry_64.S b/arch/powerpc/kernel/entry_64.S
1870 index f6fd0332c3a2..96235fe0a581 100644
1871 --- a/arch/powerpc/kernel/entry_64.S
1872 +++ b/arch/powerpc/kernel/entry_64.S
1873 @@ -683,7 +683,7 @@ _GLOBAL(ret_from_except_lite)
1877 -1: andi. r0,r4,_TIF_NEED_RESCHED
1878 +1: andi. r0,r4,_TIF_NEED_RESCHED_MASK
1880 bl restore_interrupts
1882 @@ -745,10 +745,18 @@ resume_kernel:
1884 #ifdef CONFIG_PREEMPT
1885 /* Check if we need to preempt */
1886 + lwz r8,TI_PREEMPT(r9)
1887 + cmpwi 0,r8,0 /* if non-zero, just restore regs and return */
1889 andi. r0,r4,_TIF_NEED_RESCHED
1892 + andi. r0,r4,_TIF_NEED_RESCHED_LAZY
1894 + lwz r8,TI_PREEMPT_LAZY(r9)
1896 /* Check that preempt_count() == 0 and interrupts are enabled */
1897 - lwz r8,TI_PREEMPT(r9)
1902 @@ -765,7 +773,7 @@ resume_kernel:
1903 /* Re-test flags and eventually loop */
1904 CURRENT_THREAD_INFO(r9, r1)
1906 - andi. r0,r4,_TIF_NEED_RESCHED
1907 + andi. r0,r4,_TIF_NEED_RESCHED_MASK
1911 diff --git a/arch/powerpc/kernel/irq.c b/arch/powerpc/kernel/irq.c
1912 index 290559df1e8b..070afa6da35d 100644
1913 --- a/arch/powerpc/kernel/irq.c
1914 +++ b/arch/powerpc/kernel/irq.c
1915 @@ -614,6 +614,7 @@ void irq_ctx_init(void)
1919 +#ifndef CONFIG_PREEMPT_RT_FULL
1920 void do_softirq_own_stack(void)
1922 struct thread_info *curtp, *irqtp;
1923 @@ -631,6 +632,7 @@ void do_softirq_own_stack(void)
1925 set_bits(irqtp->flags, &curtp->flags);
1929 irq_hw_number_t virq_to_hw(unsigned int virq)
1931 diff --git a/arch/powerpc/kernel/misc_32.S b/arch/powerpc/kernel/misc_32.S
1932 index df4efa304b2c..9cb0c2f6e7ac 100644
1933 --- a/arch/powerpc/kernel/misc_32.S
1934 +++ b/arch/powerpc/kernel/misc_32.S
1936 * We store the saved ksp_limit in the unused part
1937 * of the STACK_FRAME_OVERHEAD
1939 +#ifndef CONFIG_PREEMPT_RT_FULL
1940 _GLOBAL(call_do_softirq)
1943 @@ -56,6 +57,7 @@ _GLOBAL(call_do_softirq)
1944 stw r10,THREAD+KSP_LIMIT(r2)
1950 * void call_do_irq(struct pt_regs *regs, struct thread_info *irqtp);
1951 diff --git a/arch/powerpc/kernel/misc_64.S b/arch/powerpc/kernel/misc_64.S
1952 index db475d41b57a..96b7ef80e05d 100644
1953 --- a/arch/powerpc/kernel/misc_64.S
1954 +++ b/arch/powerpc/kernel/misc_64.S
1959 +#ifndef CONFIG_PREEMPT_RT_FULL
1960 _GLOBAL(call_do_softirq)
1963 @@ -40,6 +41,7 @@ _GLOBAL(call_do_softirq)
1969 _GLOBAL(call_do_irq)
1971 diff --git a/arch/powerpc/kvm/Kconfig b/arch/powerpc/kvm/Kconfig
1972 index c2024ac9d4e8..2303788da7e1 100644
1973 --- a/arch/powerpc/kvm/Kconfig
1974 +++ b/arch/powerpc/kvm/Kconfig
1975 @@ -172,6 +172,7 @@ config KVM_E500MC
1977 bool "KVM in-kernel MPIC emulation"
1978 depends on KVM && E500
1979 + depends on !PREEMPT_RT_FULL
1980 select HAVE_KVM_IRQCHIP
1981 select HAVE_KVM_IRQFD
1982 select HAVE_KVM_IRQ_ROUTING
1983 diff --git a/arch/powerpc/kvm/book3s_hv.c b/arch/powerpc/kvm/book3s_hv.c
1984 index 3c3a367b6e59..2ac7a8b108ac 100644
1985 --- a/arch/powerpc/kvm/book3s_hv.c
1986 +++ b/arch/powerpc/kvm/book3s_hv.c
1987 @@ -114,11 +114,11 @@ static bool kvmppc_ipi_thread(int cpu)
1988 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
1991 - wait_queue_head_t *wqp;
1992 + struct swait_queue_head *wqp;
1994 wqp = kvm_arch_vcpu_wq(vcpu);
1995 - if (waitqueue_active(wqp)) {
1996 - wake_up_interruptible(wqp);
1997 + if (swait_active(wqp)) {
1999 ++vcpu->stat.halt_wakeup;
2002 @@ -707,8 +707,8 @@ int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
2003 tvcpu->arch.prodded = 1;
2005 if (vcpu->arch.ceded) {
2006 - if (waitqueue_active(&vcpu->wq)) {
2007 - wake_up_interruptible(&vcpu->wq);
2008 + if (swait_active(&vcpu->wq)) {
2009 + swake_up(&vcpu->wq);
2010 vcpu->stat.halt_wakeup++;
2013 @@ -1453,7 +1453,7 @@ static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int core)
2014 INIT_LIST_HEAD(&vcore->runnable_threads);
2015 spin_lock_init(&vcore->lock);
2016 spin_lock_init(&vcore->stoltb_lock);
2017 - init_waitqueue_head(&vcore->wq);
2018 + init_swait_queue_head(&vcore->wq);
2019 vcore->preempt_tb = TB_NIL;
2020 vcore->lpcr = kvm->arch.lpcr;
2021 vcore->first_vcpuid = core * threads_per_subcore;
2022 @@ -2525,10 +2525,9 @@ static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
2024 struct kvm_vcpu *vcpu;
2026 + DECLARE_SWAITQUEUE(wait);
2028 - DEFINE_WAIT(wait);
2030 - prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
2031 + prepare_to_swait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
2034 * Check one last time for pending exceptions and ceded state after
2035 @@ -2542,7 +2541,7 @@ static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
2039 - finish_wait(&vc->wq, &wait);
2040 + finish_swait(&vc->wq, &wait);
2044 @@ -2550,7 +2549,7 @@ static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
2045 trace_kvmppc_vcore_blocked(vc, 0);
2046 spin_unlock(&vc->lock);
2048 - finish_wait(&vc->wq, &wait);
2049 + finish_swait(&vc->wq, &wait);
2050 spin_lock(&vc->lock);
2051 vc->vcore_state = VCORE_INACTIVE;
2052 trace_kvmppc_vcore_blocked(vc, 1);
2053 @@ -2606,7 +2605,7 @@ static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
2054 kvmppc_start_thread(vcpu, vc);
2055 trace_kvm_guest_enter(vcpu);
2056 } else if (vc->vcore_state == VCORE_SLEEPING) {
2058 + swake_up(&vc->wq);
2062 diff --git a/arch/powerpc/platforms/ps3/device-init.c b/arch/powerpc/platforms/ps3/device-init.c
2063 index 3f175e8aedb4..c4c02f91904c 100644
2064 --- a/arch/powerpc/platforms/ps3/device-init.c
2065 +++ b/arch/powerpc/platforms/ps3/device-init.c
2066 @@ -752,7 +752,7 @@ static int ps3_notification_read_write(struct ps3_notification_device *dev,
2068 pr_debug("%s:%u: notification %s issued\n", __func__, __LINE__, op);
2070 - res = wait_event_interruptible(dev->done.wait,
2071 + res = swait_event_interruptible(dev->done.wait,
2072 dev->done.done || kthread_should_stop());
2073 if (kthread_should_stop())
2075 diff --git a/arch/s390/include/asm/kvm_host.h b/arch/s390/include/asm/kvm_host.h
2076 index e9a983f40a24..bbdc539fb3c6 100644
2077 --- a/arch/s390/include/asm/kvm_host.h
2078 +++ b/arch/s390/include/asm/kvm_host.h
2079 @@ -427,7 +427,7 @@ struct kvm_s390_irq_payload {
2080 struct kvm_s390_local_interrupt {
2082 struct kvm_s390_float_interrupt *float_int;
2083 - wait_queue_head_t *wq;
2084 + struct swait_queue_head *wq;
2086 DECLARE_BITMAP(sigp_emerg_pending, KVM_MAX_VCPUS);
2087 struct kvm_s390_irq_payload irq;
2088 diff --git a/arch/s390/kvm/interrupt.c b/arch/s390/kvm/interrupt.c
2089 index 6a75352f453c..cc862c486002 100644
2090 --- a/arch/s390/kvm/interrupt.c
2091 +++ b/arch/s390/kvm/interrupt.c
2092 @@ -868,13 +868,13 @@ no_timer:
2094 void kvm_s390_vcpu_wakeup(struct kvm_vcpu *vcpu)
2096 - if (waitqueue_active(&vcpu->wq)) {
2097 + if (swait_active(&vcpu->wq)) {
2099 * The vcpu gave up the cpu voluntarily, mark it as a good
2102 vcpu->preempted = true;
2103 - wake_up_interruptible(&vcpu->wq);
2104 + swake_up(&vcpu->wq);
2105 vcpu->stat.halt_wakeup++;
2108 diff --git a/arch/sh/kernel/irq.c b/arch/sh/kernel/irq.c
2109 index 6c0378c0b8b5..abd58b4dff97 100644
2110 --- a/arch/sh/kernel/irq.c
2111 +++ b/arch/sh/kernel/irq.c
2112 @@ -147,6 +147,7 @@ void irq_ctx_exit(int cpu)
2113 hardirq_ctx[cpu] = NULL;
2116 +#ifndef CONFIG_PREEMPT_RT_FULL
2117 void do_softirq_own_stack(void)
2119 struct thread_info *curctx;
2120 @@ -174,6 +175,7 @@ void do_softirq_own_stack(void)
2121 "r5", "r6", "r7", "r8", "r9", "r15", "t", "pr"
2126 static inline void handle_one_irq(unsigned int irq)
2128 diff --git a/arch/sparc/Kconfig b/arch/sparc/Kconfig
2129 index 56442d2d7bbc..8c9598f534c9 100644
2130 --- a/arch/sparc/Kconfig
2131 +++ b/arch/sparc/Kconfig
2132 @@ -189,12 +189,10 @@ config NR_CPUS
2133 source kernel/Kconfig.hz
2135 config RWSEM_GENERIC_SPINLOCK
2137 - default y if SPARC32
2138 + def_bool PREEMPT_RT_FULL
2140 config RWSEM_XCHGADD_ALGORITHM
2142 - default y if SPARC64
2143 + def_bool !RWSEM_GENERIC_SPINLOCK && !PREEMPT_RT_FULL
2145 config GENERIC_HWEIGHT
2147 diff --git a/arch/sparc/kernel/irq_64.c b/arch/sparc/kernel/irq_64.c
2148 index e22416ce56ea..d359de71153a 100644
2149 --- a/arch/sparc/kernel/irq_64.c
2150 +++ b/arch/sparc/kernel/irq_64.c
2151 @@ -854,6 +854,7 @@ void __irq_entry handler_irq(int pil, struct pt_regs *regs)
2152 set_irq_regs(old_regs);
2155 +#ifndef CONFIG_PREEMPT_RT_FULL
2156 void do_softirq_own_stack(void)
2158 void *orig_sp, *sp = softirq_stack[smp_processor_id()];
2159 @@ -868,6 +869,7 @@ void do_softirq_own_stack(void)
2160 __asm__ __volatile__("mov %0, %%sp"
2165 #ifdef CONFIG_HOTPLUG_CPU
2166 void fixup_irqs(void)
2167 diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
2168 index 436639a31624..6ee1dd0deadc 100644
2169 --- a/arch/x86/Kconfig
2170 +++ b/arch/x86/Kconfig
2171 @@ -17,6 +17,7 @@ config X86_64
2175 + select HAVE_PREEMPT_LAZY
2176 select ACPI_LEGACY_TABLES_LOOKUP if ACPI
2177 select ACPI_SYSTEM_POWER_STATES_SUPPORT if ACPI
2179 @@ -212,8 +213,11 @@ config ARCH_MAY_HAVE_PC_FDC
2181 depends on ISA_DMA_API
2183 +config RWSEM_GENERIC_SPINLOCK
2184 + def_bool PREEMPT_RT_FULL
2186 config RWSEM_XCHGADD_ALGORITHM
2188 + def_bool !RWSEM_GENERIC_SPINLOCK && !PREEMPT_RT_FULL
2190 config GENERIC_CALIBRATE_DELAY
2192 @@ -848,7 +852,7 @@ config IOMMU_HELPER
2194 bool "Enable Maximum number of SMP Processors and NUMA Nodes"
2195 depends on X86_64 && SMP && DEBUG_KERNEL
2196 - select CPUMASK_OFFSTACK
2197 + select CPUMASK_OFFSTACK if !PREEMPT_RT_FULL
2199 Enable maximum number of CPUS and NUMA Nodes for this architecture.
2201 diff --git a/arch/x86/crypto/aesni-intel_glue.c b/arch/x86/crypto/aesni-intel_glue.c
2202 index 3633ad6145c5..c6d5458ee7f9 100644
2203 --- a/arch/x86/crypto/aesni-intel_glue.c
2204 +++ b/arch/x86/crypto/aesni-intel_glue.c
2205 @@ -383,14 +383,14 @@ static int ecb_encrypt(struct blkcipher_desc *desc,
2206 err = blkcipher_walk_virt(desc, &walk);
2207 desc->flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP;
2209 - kernel_fpu_begin();
2210 while ((nbytes = walk.nbytes)) {
2211 + kernel_fpu_begin();
2212 aesni_ecb_enc(ctx, walk.dst.virt.addr, walk.src.virt.addr,
2213 - nbytes & AES_BLOCK_MASK);
2214 + nbytes & AES_BLOCK_MASK);
2216 nbytes &= AES_BLOCK_SIZE - 1;
2217 err = blkcipher_walk_done(desc, &walk, nbytes);
2223 @@ -407,14 +407,14 @@ static int ecb_decrypt(struct blkcipher_desc *desc,
2224 err = blkcipher_walk_virt(desc, &walk);
2225 desc->flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP;
2227 - kernel_fpu_begin();
2228 while ((nbytes = walk.nbytes)) {
2229 + kernel_fpu_begin();
2230 aesni_ecb_dec(ctx, walk.dst.virt.addr, walk.src.virt.addr,
2231 nbytes & AES_BLOCK_MASK);
2233 nbytes &= AES_BLOCK_SIZE - 1;
2234 err = blkcipher_walk_done(desc, &walk, nbytes);
2240 @@ -431,14 +431,14 @@ static int cbc_encrypt(struct blkcipher_desc *desc,
2241 err = blkcipher_walk_virt(desc, &walk);
2242 desc->flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP;
2244 - kernel_fpu_begin();
2245 while ((nbytes = walk.nbytes)) {
2246 + kernel_fpu_begin();
2247 aesni_cbc_enc(ctx, walk.dst.virt.addr, walk.src.virt.addr,
2248 nbytes & AES_BLOCK_MASK, walk.iv);
2250 nbytes &= AES_BLOCK_SIZE - 1;
2251 err = blkcipher_walk_done(desc, &walk, nbytes);
2257 @@ -455,14 +455,14 @@ static int cbc_decrypt(struct blkcipher_desc *desc,
2258 err = blkcipher_walk_virt(desc, &walk);
2259 desc->flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP;
2261 - kernel_fpu_begin();
2262 while ((nbytes = walk.nbytes)) {
2263 + kernel_fpu_begin();
2264 aesni_cbc_dec(ctx, walk.dst.virt.addr, walk.src.virt.addr,
2265 nbytes & AES_BLOCK_MASK, walk.iv);
2267 nbytes &= AES_BLOCK_SIZE - 1;
2268 err = blkcipher_walk_done(desc, &walk, nbytes);
2274 @@ -514,18 +514,20 @@ static int ctr_crypt(struct blkcipher_desc *desc,
2275 err = blkcipher_walk_virt_block(desc, &walk, AES_BLOCK_SIZE);
2276 desc->flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP;
2278 - kernel_fpu_begin();
2279 while ((nbytes = walk.nbytes) >= AES_BLOCK_SIZE) {
2280 + kernel_fpu_begin();
2281 aesni_ctr_enc_tfm(ctx, walk.dst.virt.addr, walk.src.virt.addr,
2282 nbytes & AES_BLOCK_MASK, walk.iv);
2284 nbytes &= AES_BLOCK_SIZE - 1;
2285 err = blkcipher_walk_done(desc, &walk, nbytes);
2288 + kernel_fpu_begin();
2289 ctr_crypt_final(ctx, &walk);
2291 err = blkcipher_walk_done(desc, &walk, 0);
2297 diff --git a/arch/x86/crypto/cast5_avx_glue.c b/arch/x86/crypto/cast5_avx_glue.c
2298 index 8648158f3916..d7699130ee36 100644
2299 --- a/arch/x86/crypto/cast5_avx_glue.c
2300 +++ b/arch/x86/crypto/cast5_avx_glue.c
2301 @@ -59,7 +59,7 @@ static inline void cast5_fpu_end(bool fpu_enabled)
2302 static int ecb_crypt(struct blkcipher_desc *desc, struct blkcipher_walk *walk,
2305 - bool fpu_enabled = false;
2307 struct cast5_ctx *ctx = crypto_blkcipher_ctx(desc->tfm);
2308 const unsigned int bsize = CAST5_BLOCK_SIZE;
2309 unsigned int nbytes;
2310 @@ -75,7 +75,7 @@ static int ecb_crypt(struct blkcipher_desc *desc, struct blkcipher_walk *walk,
2311 u8 *wsrc = walk->src.virt.addr;
2312 u8 *wdst = walk->dst.virt.addr;
2314 - fpu_enabled = cast5_fpu_begin(fpu_enabled, nbytes);
2315 + fpu_enabled = cast5_fpu_begin(false, nbytes);
2317 /* Process multi-block batch */
2318 if (nbytes >= bsize * CAST5_PARALLEL_BLOCKS) {
2319 @@ -103,10 +103,9 @@ static int ecb_crypt(struct blkcipher_desc *desc, struct blkcipher_walk *walk,
2320 } while (nbytes >= bsize);
2323 + cast5_fpu_end(fpu_enabled);
2324 err = blkcipher_walk_done(desc, walk, nbytes);
2327 - cast5_fpu_end(fpu_enabled);
2331 @@ -227,7 +226,7 @@ done:
2332 static int cbc_decrypt(struct blkcipher_desc *desc, struct scatterlist *dst,
2333 struct scatterlist *src, unsigned int nbytes)
2335 - bool fpu_enabled = false;
2337 struct blkcipher_walk walk;
2340 @@ -236,12 +235,11 @@ static int cbc_decrypt(struct blkcipher_desc *desc, struct scatterlist *dst,
2341 desc->flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP;
2343 while ((nbytes = walk.nbytes)) {
2344 - fpu_enabled = cast5_fpu_begin(fpu_enabled, nbytes);
2345 + fpu_enabled = cast5_fpu_begin(false, nbytes);
2346 nbytes = __cbc_decrypt(desc, &walk);
2347 + cast5_fpu_end(fpu_enabled);
2348 err = blkcipher_walk_done(desc, &walk, nbytes);
2351 - cast5_fpu_end(fpu_enabled);
2355 @@ -311,7 +309,7 @@ done:
2356 static int ctr_crypt(struct blkcipher_desc *desc, struct scatterlist *dst,
2357 struct scatterlist *src, unsigned int nbytes)
2359 - bool fpu_enabled = false;
2361 struct blkcipher_walk walk;
2364 @@ -320,13 +318,12 @@ static int ctr_crypt(struct blkcipher_desc *desc, struct scatterlist *dst,
2365 desc->flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP;
2367 while ((nbytes = walk.nbytes) >= CAST5_BLOCK_SIZE) {
2368 - fpu_enabled = cast5_fpu_begin(fpu_enabled, nbytes);
2369 + fpu_enabled = cast5_fpu_begin(false, nbytes);
2370 nbytes = __ctr_crypt(desc, &walk);
2371 + cast5_fpu_end(fpu_enabled);
2372 err = blkcipher_walk_done(desc, &walk, nbytes);
2375 - cast5_fpu_end(fpu_enabled);
2378 ctr_crypt_final(desc, &walk);
2379 err = blkcipher_walk_done(desc, &walk, 0);
2380 diff --git a/arch/x86/crypto/glue_helper.c b/arch/x86/crypto/glue_helper.c
2381 index 6a85598931b5..3a506ce7ed93 100644
2382 --- a/arch/x86/crypto/glue_helper.c
2383 +++ b/arch/x86/crypto/glue_helper.c
2384 @@ -39,7 +39,7 @@ static int __glue_ecb_crypt_128bit(const struct common_glue_ctx *gctx,
2385 void *ctx = crypto_blkcipher_ctx(desc->tfm);
2386 const unsigned int bsize = 128 / 8;
2387 unsigned int nbytes, i, func_bytes;
2388 - bool fpu_enabled = false;
2392 err = blkcipher_walk_virt(desc, walk);
2393 @@ -49,7 +49,7 @@ static int __glue_ecb_crypt_128bit(const struct common_glue_ctx *gctx,
2394 u8 *wdst = walk->dst.virt.addr;
2396 fpu_enabled = glue_fpu_begin(bsize, gctx->fpu_blocks_limit,
2397 - desc, fpu_enabled, nbytes);
2398 + desc, false, nbytes);
2400 for (i = 0; i < gctx->num_funcs; i++) {
2401 func_bytes = bsize * gctx->funcs[i].num_blocks;
2402 @@ -71,10 +71,10 @@ static int __glue_ecb_crypt_128bit(const struct common_glue_ctx *gctx,
2406 + glue_fpu_end(fpu_enabled);
2407 err = blkcipher_walk_done(desc, walk, nbytes);
2410 - glue_fpu_end(fpu_enabled);
2414 @@ -194,7 +194,7 @@ int glue_cbc_decrypt_128bit(const struct common_glue_ctx *gctx,
2415 struct scatterlist *src, unsigned int nbytes)
2417 const unsigned int bsize = 128 / 8;
2418 - bool fpu_enabled = false;
2420 struct blkcipher_walk walk;
2423 @@ -203,12 +203,12 @@ int glue_cbc_decrypt_128bit(const struct common_glue_ctx *gctx,
2425 while ((nbytes = walk.nbytes)) {
2426 fpu_enabled = glue_fpu_begin(bsize, gctx->fpu_blocks_limit,
2427 - desc, fpu_enabled, nbytes);
2428 + desc, false, nbytes);
2429 nbytes = __glue_cbc_decrypt_128bit(gctx, desc, &walk);
2430 + glue_fpu_end(fpu_enabled);
2431 err = blkcipher_walk_done(desc, &walk, nbytes);
2434 - glue_fpu_end(fpu_enabled);
2437 EXPORT_SYMBOL_GPL(glue_cbc_decrypt_128bit);
2438 @@ -277,7 +277,7 @@ int glue_ctr_crypt_128bit(const struct common_glue_ctx *gctx,
2439 struct scatterlist *src, unsigned int nbytes)
2441 const unsigned int bsize = 128 / 8;
2442 - bool fpu_enabled = false;
2444 struct blkcipher_walk walk;
2447 @@ -286,13 +286,12 @@ int glue_ctr_crypt_128bit(const struct common_glue_ctx *gctx,
2449 while ((nbytes = walk.nbytes) >= bsize) {
2450 fpu_enabled = glue_fpu_begin(bsize, gctx->fpu_blocks_limit,
2451 - desc, fpu_enabled, nbytes);
2452 + desc, false, nbytes);
2453 nbytes = __glue_ctr_crypt_128bit(gctx, desc, &walk);
2454 + glue_fpu_end(fpu_enabled);
2455 err = blkcipher_walk_done(desc, &walk, nbytes);
2458 - glue_fpu_end(fpu_enabled);
2461 glue_ctr_crypt_final_128bit(
2462 gctx->funcs[gctx->num_funcs - 1].fn_u.ctr, desc, &walk);
2463 @@ -347,7 +346,7 @@ int glue_xts_crypt_128bit(const struct common_glue_ctx *gctx,
2464 void *tweak_ctx, void *crypt_ctx)
2466 const unsigned int bsize = 128 / 8;
2467 - bool fpu_enabled = false;
2469 struct blkcipher_walk walk;
2472 @@ -360,21 +359,21 @@ int glue_xts_crypt_128bit(const struct common_glue_ctx *gctx,
2474 /* set minimum length to bsize, for tweak_fn */
2475 fpu_enabled = glue_fpu_begin(bsize, gctx->fpu_blocks_limit,
2476 - desc, fpu_enabled,
2478 nbytes < bsize ? bsize : nbytes);
2480 /* calculate first value of T */
2481 tweak_fn(tweak_ctx, walk.iv, walk.iv);
2482 + glue_fpu_end(fpu_enabled);
2485 + fpu_enabled = glue_fpu_begin(bsize, gctx->fpu_blocks_limit,
2486 + desc, false, nbytes);
2487 nbytes = __glue_xts_crypt_128bit(gctx, crypt_ctx, desc, &walk);
2489 + glue_fpu_end(fpu_enabled);
2490 err = blkcipher_walk_done(desc, &walk, nbytes);
2491 nbytes = walk.nbytes;
2494 - glue_fpu_end(fpu_enabled);
2498 EXPORT_SYMBOL_GPL(glue_xts_crypt_128bit);
2499 diff --git a/arch/x86/entry/common.c b/arch/x86/entry/common.c
2500 index 1a4477cedc49..75a301b6a5b6 100644
2501 --- a/arch/x86/entry/common.c
2502 +++ b/arch/x86/entry/common.c
2503 @@ -220,7 +220,7 @@ long syscall_trace_enter(struct pt_regs *regs)
2505 #define EXIT_TO_USERMODE_LOOP_FLAGS \
2506 (_TIF_SIGPENDING | _TIF_NOTIFY_RESUME | _TIF_UPROBE | \
2507 - _TIF_NEED_RESCHED | _TIF_USER_RETURN_NOTIFY)
2508 + _TIF_NEED_RESCHED_MASK | _TIF_USER_RETURN_NOTIFY)
2510 static void exit_to_usermode_loop(struct pt_regs *regs, u32 cached_flags)
2512 @@ -236,9 +236,16 @@ static void exit_to_usermode_loop(struct pt_regs *regs, u32 cached_flags)
2513 /* We have work to do. */
2516 - if (cached_flags & _TIF_NEED_RESCHED)
2517 + if (cached_flags & _TIF_NEED_RESCHED_MASK)
2520 +#ifdef ARCH_RT_DELAYS_SIGNAL_SEND
2521 + if (unlikely(current->forced_info.si_signo)) {
2522 + struct task_struct *t = current;
2523 + force_sig_info(t->forced_info.si_signo, &t->forced_info, t);
2524 + t->forced_info.si_signo = 0;
2527 if (cached_flags & _TIF_UPROBE)
2528 uprobe_notify_resume(regs);
2530 diff --git a/arch/x86/entry/entry_32.S b/arch/x86/entry/entry_32.S
2531 index ae678ad128a9..3bcef8bdb911 100644
2532 --- a/arch/x86/entry/entry_32.S
2533 +++ b/arch/x86/entry/entry_32.S
2534 @@ -278,8 +278,24 @@ END(ret_from_exception)
2535 ENTRY(resume_kernel)
2536 DISABLE_INTERRUPTS(CLBR_ANY)
2538 + # preempt count == 0 + NEED_RS set?
2539 cmpl $0, PER_CPU_VAR(__preempt_count)
2540 +#ifndef CONFIG_PREEMPT_LAZY
2545 + # atleast preempt count == 0 ?
2546 + cmpl $_PREEMPT_ENABLED,PER_CPU_VAR(__preempt_count)
2549 + cmpl $0,TI_preempt_lazy_count(%ebp) # non-zero preempt_lazy_count ?
2552 + testl $_TIF_NEED_RESCHED_LAZY, TI_flags(%ebp)
2556 testl $X86_EFLAGS_IF, PT_EFLAGS(%esp) # interrupts off (exception path) ?
2558 call preempt_schedule_irq
2559 diff --git a/arch/x86/entry/entry_64.S b/arch/x86/entry/entry_64.S
2560 index a55697d19824..316081a2ca85 100644
2561 --- a/arch/x86/entry/entry_64.S
2562 +++ b/arch/x86/entry/entry_64.S
2563 @@ -579,7 +579,23 @@ retint_kernel:
2564 bt $9, EFLAGS(%rsp) /* were interrupts off? */
2566 0: cmpl $0, PER_CPU_VAR(__preempt_count)
2567 +#ifndef CONFIG_PREEMPT_LAZY
2570 + jz do_preempt_schedule_irq
2572 + # atleast preempt count == 0 ?
2573 + cmpl $_PREEMPT_ENABLED,PER_CPU_VAR(__preempt_count)
2576 + GET_THREAD_INFO(%rcx)
2577 + cmpl $0, TI_preempt_lazy_count(%rcx)
2580 + bt $TIF_NEED_RESCHED_LAZY,TI_flags(%rcx)
2582 +do_preempt_schedule_irq:
2584 call preempt_schedule_irq
2587 @@ -867,6 +883,7 @@ bad_gs:
2591 +#ifndef CONFIG_PREEMPT_RT_FULL
2592 /* Call softirq on interrupt stack. Interrupts are off. */
2593 ENTRY(do_softirq_own_stack)
2595 @@ -879,6 +896,7 @@ ENTRY(do_softirq_own_stack)
2596 decl PER_CPU_VAR(irq_count)
2598 END(do_softirq_own_stack)
2602 idtentry xen_hypervisor_callback xen_do_hypervisor_callback has_error_code=0
2603 diff --git a/arch/x86/include/asm/preempt.h b/arch/x86/include/asm/preempt.h
2604 index 01bcde84d3e4..6f432adc55cd 100644
2605 --- a/arch/x86/include/asm/preempt.h
2606 +++ b/arch/x86/include/asm/preempt.h
2607 @@ -79,17 +79,46 @@ static __always_inline void __preempt_count_sub(int val)
2608 * a decrement which hits zero means we have no preempt_count and should
2611 -static __always_inline bool __preempt_count_dec_and_test(void)
2612 +static __always_inline bool ____preempt_count_dec_and_test(void)
2614 GEN_UNARY_RMWcc("decl", __preempt_count, __percpu_arg(0), "e");
2617 +static __always_inline bool __preempt_count_dec_and_test(void)
2619 + if (____preempt_count_dec_and_test())
2621 +#ifdef CONFIG_PREEMPT_LAZY
2622 + if (current_thread_info()->preempt_lazy_count)
2624 + return test_thread_flag(TIF_NEED_RESCHED_LAZY);
2631 * Returns true when we need to resched and can (barring IRQ state).
2633 static __always_inline bool should_resched(int preempt_offset)
2635 +#ifdef CONFIG_PREEMPT_LAZY
2638 + tmp = raw_cpu_read_4(__preempt_count);
2639 + if (tmp == preempt_offset)
2642 + /* preempt count == 0 ? */
2643 + tmp &= ~PREEMPT_NEED_RESCHED;
2646 + if (current_thread_info()->preempt_lazy_count)
2648 + return test_thread_flag(TIF_NEED_RESCHED_LAZY);
2650 return unlikely(raw_cpu_read_4(__preempt_count) == preempt_offset);
2654 #ifdef CONFIG_PREEMPT
2655 diff --git a/arch/x86/include/asm/signal.h b/arch/x86/include/asm/signal.h
2656 index 2138c9ae19ee..3f5b4ee2e2c1 100644
2657 --- a/arch/x86/include/asm/signal.h
2658 +++ b/arch/x86/include/asm/signal.h
2659 @@ -23,6 +23,19 @@ typedef struct {
2660 unsigned long sig[_NSIG_WORDS];
2664 + * Because some traps use the IST stack, we must keep preemption
2665 + * disabled while calling do_trap(), but do_trap() may call
2666 + * force_sig_info() which will grab the signal spin_locks for the
2667 + * task, which in PREEMPT_RT_FULL are mutexes. By defining
2668 + * ARCH_RT_DELAYS_SIGNAL_SEND the force_sig_info() will set
2669 + * TIF_NOTIFY_RESUME and set up the signal to be sent on exit of the
2672 +#if defined(CONFIG_PREEMPT_RT_FULL)
2673 +#define ARCH_RT_DELAYS_SIGNAL_SEND
2676 #ifndef CONFIG_COMPAT
2677 typedef sigset_t compat_sigset_t;
2679 diff --git a/arch/x86/include/asm/stackprotector.h b/arch/x86/include/asm/stackprotector.h
2680 index 58505f01962f..02fa39652cd6 100644
2681 --- a/arch/x86/include/asm/stackprotector.h
2682 +++ b/arch/x86/include/asm/stackprotector.h
2685 static __always_inline void boot_init_stack_canary(void)
2688 + u64 uninitialized_var(canary);
2691 #ifdef CONFIG_X86_64
2692 @@ -70,8 +70,15 @@ static __always_inline void boot_init_stack_canary(void)
2693 * of randomness. The TSC only matters for very early init,
2694 * there it already has some randomness on most systems. Later
2695 * on during the bootup the random pool has true entropy too.
2697 + * For preempt-rt we need to weaken the randomness a bit, as
2698 + * we can't call into the random generator from atomic context
2699 + * due to locking constraints. We just leave canary
2700 + * uninitialized and use the TSC based randomness on top of it.
2702 +#ifndef CONFIG_PREEMPT_RT_FULL
2703 get_random_bytes(&canary, sizeof(canary));
2706 canary += tsc + (tsc << 32UL);
2708 diff --git a/arch/x86/include/asm/thread_info.h b/arch/x86/include/asm/thread_info.h
2709 index c7b551028740..ddb63bd90e3c 100644
2710 --- a/arch/x86/include/asm/thread_info.h
2711 +++ b/arch/x86/include/asm/thread_info.h
2712 @@ -58,6 +58,8 @@ struct thread_info {
2713 __u32 status; /* thread synchronous flags */
2714 __u32 cpu; /* current CPU */
2715 mm_segment_t addr_limit;
2716 + int preempt_lazy_count; /* 0 => lazy preemptable
2718 unsigned int sig_on_uaccess_error:1;
2719 unsigned int uaccess_err:1; /* uaccess failed */
2721 @@ -95,6 +97,7 @@ struct thread_info {
2722 #define TIF_SYSCALL_EMU 6 /* syscall emulation active */
2723 #define TIF_SYSCALL_AUDIT 7 /* syscall auditing active */
2724 #define TIF_SECCOMP 8 /* secure computing */
2725 +#define TIF_NEED_RESCHED_LAZY 9 /* lazy rescheduling necessary */
2726 #define TIF_USER_RETURN_NOTIFY 11 /* notify kernel of userspace return */
2727 #define TIF_UPROBE 12 /* breakpointed or singlestepping */
2728 #define TIF_NOTSC 16 /* TSC is not accessible in userland */
2729 @@ -119,6 +122,7 @@ struct thread_info {
2730 #define _TIF_SYSCALL_EMU (1 << TIF_SYSCALL_EMU)
2731 #define _TIF_SYSCALL_AUDIT (1 << TIF_SYSCALL_AUDIT)
2732 #define _TIF_SECCOMP (1 << TIF_SECCOMP)
2733 +#define _TIF_NEED_RESCHED_LAZY (1 << TIF_NEED_RESCHED_LAZY)
2734 #define _TIF_USER_RETURN_NOTIFY (1 << TIF_USER_RETURN_NOTIFY)
2735 #define _TIF_UPROBE (1 << TIF_UPROBE)
2736 #define _TIF_NOTSC (1 << TIF_NOTSC)
2737 @@ -152,6 +156,8 @@ struct thread_info {
2738 #define _TIF_WORK_CTXSW_PREV (_TIF_WORK_CTXSW|_TIF_USER_RETURN_NOTIFY)
2739 #define _TIF_WORK_CTXSW_NEXT (_TIF_WORK_CTXSW)
2741 +#define _TIF_NEED_RESCHED_MASK (_TIF_NEED_RESCHED | _TIF_NEED_RESCHED_LAZY)
2743 #define STACK_WARN (THREAD_SIZE/8)
2746 diff --git a/arch/x86/include/asm/uv/uv_bau.h b/arch/x86/include/asm/uv/uv_bau.h
2747 index fc808b83fccb..ebb40118abf5 100644
2748 --- a/arch/x86/include/asm/uv/uv_bau.h
2749 +++ b/arch/x86/include/asm/uv/uv_bau.h
2750 @@ -615,9 +615,9 @@ struct bau_control {
2751 cycles_t send_message;
2752 cycles_t period_end;
2753 cycles_t period_time;
2754 - spinlock_t uvhub_lock;
2755 - spinlock_t queue_lock;
2756 - spinlock_t disable_lock;
2757 + raw_spinlock_t uvhub_lock;
2758 + raw_spinlock_t queue_lock;
2759 + raw_spinlock_t disable_lock;
2762 int max_concurr_const;
2763 @@ -776,15 +776,15 @@ static inline int atom_asr(short i, struct atomic_short *v)
2764 * to be lowered below the current 'v'. atomic_add_unless can only stop
2767 -static inline int atomic_inc_unless_ge(spinlock_t *lock, atomic_t *v, int u)
2768 +static inline int atomic_inc_unless_ge(raw_spinlock_t *lock, atomic_t *v, int u)
2771 + raw_spin_lock(lock);
2772 if (atomic_read(v) >= u) {
2773 - spin_unlock(lock);
2774 + raw_spin_unlock(lock);
2778 - spin_unlock(lock);
2779 + raw_spin_unlock(lock);
2783 diff --git a/arch/x86/include/asm/uv/uv_hub.h b/arch/x86/include/asm/uv/uv_hub.h
2784 index ea7074784cc4..01ec643ce66e 100644
2785 --- a/arch/x86/include/asm/uv/uv_hub.h
2786 +++ b/arch/x86/include/asm/uv/uv_hub.h
2787 @@ -492,7 +492,7 @@ struct uv_blade_info {
2788 unsigned short nr_online_cpus;
2789 unsigned short pnode;
2791 - spinlock_t nmi_lock; /* obsolete, see uv_hub_nmi */
2792 + raw_spinlock_t nmi_lock; /* obsolete, see uv_hub_nmi */
2793 unsigned long nmi_count; /* obsolete, see uv_hub_nmi */
2795 extern struct uv_blade_info *uv_blade_info;
2796 diff --git a/arch/x86/kernel/acpi/boot.c b/arch/x86/kernel/acpi/boot.c
2797 index e75907601a41..a29fc4f84fc4 100644
2798 --- a/arch/x86/kernel/acpi/boot.c
2799 +++ b/arch/x86/kernel/acpi/boot.c
2800 @@ -87,7 +87,9 @@ static u64 acpi_lapic_addr __initdata = APIC_DEFAULT_PHYS_BASE;
2804 +#ifdef CONFIG_X86_IO_APIC
2805 static DEFINE_MUTEX(acpi_ioapic_lock);
2808 /* --------------------------------------------------------------------------
2809 Boot-time Configuration
2810 diff --git a/arch/x86/kernel/apic/io_apic.c b/arch/x86/kernel/apic/io_apic.c
2811 index 1e5d2f07416b..ac0127bb155e 100644
2812 --- a/arch/x86/kernel/apic/io_apic.c
2813 +++ b/arch/x86/kernel/apic/io_apic.c
2814 @@ -1711,7 +1711,8 @@ static bool io_apic_level_ack_pending(struct mp_chip_data *data)
2815 static inline bool ioapic_irqd_mask(struct irq_data *data)
2817 /* If we are moving the irq we need to mask it */
2818 - if (unlikely(irqd_is_setaffinity_pending(data))) {
2819 + if (unlikely(irqd_is_setaffinity_pending(data) &&
2820 + !irqd_irq_inprogress(data))) {
2821 mask_ioapic_irq(data);
2824 diff --git a/arch/x86/kernel/apic/x2apic_uv_x.c b/arch/x86/kernel/apic/x2apic_uv_x.c
2825 index 4a139465f1d4..ad2afff02b36 100644
2826 --- a/arch/x86/kernel/apic/x2apic_uv_x.c
2827 +++ b/arch/x86/kernel/apic/x2apic_uv_x.c
2828 @@ -947,7 +947,7 @@ void __init uv_system_init(void)
2829 uv_blade_info[blade].pnode = pnode;
2830 uv_blade_info[blade].nr_possible_cpus = 0;
2831 uv_blade_info[blade].nr_online_cpus = 0;
2832 - spin_lock_init(&uv_blade_info[blade].nmi_lock);
2833 + raw_spin_lock_init(&uv_blade_info[blade].nmi_lock);
2834 min_pnode = min(pnode, min_pnode);
2835 max_pnode = max(pnode, max_pnode);
2837 diff --git a/arch/x86/kernel/asm-offsets.c b/arch/x86/kernel/asm-offsets.c
2838 index 439df975bc7a..b7954ddd6a0a 100644
2839 --- a/arch/x86/kernel/asm-offsets.c
2840 +++ b/arch/x86/kernel/asm-offsets.c
2841 @@ -32,6 +32,7 @@ void common(void) {
2842 OFFSET(TI_flags, thread_info, flags);
2843 OFFSET(TI_status, thread_info, status);
2844 OFFSET(TI_addr_limit, thread_info, addr_limit);
2845 + OFFSET(TI_preempt_lazy_count, thread_info, preempt_lazy_count);
2848 OFFSET(crypto_tfm_ctx_offset, crypto_tfm, __crt_ctx);
2849 @@ -89,4 +90,5 @@ void common(void) {
2852 DEFINE(PTREGS_SIZE, sizeof(struct pt_regs));
2853 + DEFINE(_PREEMPT_ENABLED, PREEMPT_ENABLED);
2855 diff --git a/arch/x86/kernel/cpu/mcheck/mce.c b/arch/x86/kernel/cpu/mcheck/mce.c
2856 index 7e8a736d09db..430a4ec07811 100644
2857 --- a/arch/x86/kernel/cpu/mcheck/mce.c
2858 +++ b/arch/x86/kernel/cpu/mcheck/mce.c
2860 #include <linux/debugfs.h>
2861 #include <linux/irq_work.h>
2862 #include <linux/export.h>
2863 +#include <linux/jiffies.h>
2864 +#include <linux/swork.h>
2866 #include <asm/processor.h>
2867 #include <asm/traps.h>
2868 @@ -1236,7 +1238,7 @@ void mce_log_therm_throt_event(__u64 status)
2869 static unsigned long check_interval = INITIAL_CHECK_INTERVAL;
2871 static DEFINE_PER_CPU(unsigned long, mce_next_interval); /* in jiffies */
2872 -static DEFINE_PER_CPU(struct timer_list, mce_timer);
2873 +static DEFINE_PER_CPU(struct hrtimer, mce_timer);
2875 static unsigned long mce_adjust_timer_default(unsigned long interval)
2877 @@ -1245,32 +1247,18 @@ static unsigned long mce_adjust_timer_default(unsigned long interval)
2879 static unsigned long (*mce_adjust_timer)(unsigned long interval) = mce_adjust_timer_default;
2881 -static void __restart_timer(struct timer_list *t, unsigned long interval)
2882 +static enum hrtimer_restart __restart_timer(struct hrtimer *timer, unsigned long interval)
2884 - unsigned long when = jiffies + interval;
2885 - unsigned long flags;
2887 - local_irq_save(flags);
2889 - if (timer_pending(t)) {
2890 - if (time_before(when, t->expires))
2891 - mod_timer_pinned(t, when);
2893 - t->expires = round_jiffies(when);
2894 - add_timer_on(t, smp_processor_id());
2897 - local_irq_restore(flags);
2899 + return HRTIMER_NORESTART;
2900 + hrtimer_forward_now(timer, ns_to_ktime(jiffies_to_nsecs(interval)));
2901 + return HRTIMER_RESTART;
2904 -static void mce_timer_fn(unsigned long data)
2905 +static enum hrtimer_restart mce_timer_fn(struct hrtimer *timer)
2907 - struct timer_list *t = this_cpu_ptr(&mce_timer);
2908 - int cpu = smp_processor_id();
2911 - WARN_ON(cpu != data);
2913 iv = __this_cpu_read(mce_next_interval);
2915 if (mce_available(this_cpu_ptr(&cpu_info))) {
2916 @@ -1293,7 +1281,7 @@ static void mce_timer_fn(unsigned long data)
2919 __this_cpu_write(mce_next_interval, iv);
2920 - __restart_timer(t, iv);
2921 + return __restart_timer(timer, iv);
2925 @@ -1301,7 +1289,7 @@ done:
2927 void mce_timer_kick(unsigned long interval)
2929 - struct timer_list *t = this_cpu_ptr(&mce_timer);
2930 + struct hrtimer *t = this_cpu_ptr(&mce_timer);
2931 unsigned long iv = __this_cpu_read(mce_next_interval);
2933 __restart_timer(t, interval);
2934 @@ -1316,7 +1304,7 @@ static void mce_timer_delete_all(void)
2937 for_each_online_cpu(cpu)
2938 - del_timer_sync(&per_cpu(mce_timer, cpu));
2939 + hrtimer_cancel(&per_cpu(mce_timer, cpu));
2942 static void mce_do_trigger(struct work_struct *work)
2943 @@ -1326,6 +1314,56 @@ static void mce_do_trigger(struct work_struct *work)
2945 static DECLARE_WORK(mce_trigger_work, mce_do_trigger);
2947 +static void __mce_notify_work(struct swork_event *event)
2949 + /* Not more than two messages every minute */
2950 + static DEFINE_RATELIMIT_STATE(ratelimit, 60*HZ, 2);
2952 + /* wake processes polling /dev/mcelog */
2953 + wake_up_interruptible(&mce_chrdev_wait);
2956 + * There is no risk of missing notifications because
2957 + * work_pending is always cleared before the function is
2960 + if (mce_helper[0] && !work_pending(&mce_trigger_work))
2961 + schedule_work(&mce_trigger_work);
2963 + if (__ratelimit(&ratelimit))
2964 + pr_info(HW_ERR "Machine check events logged\n");
2967 +#ifdef CONFIG_PREEMPT_RT_FULL
2968 +static bool notify_work_ready __read_mostly;
2969 +static struct swork_event notify_work;
2971 +static int mce_notify_work_init(void)
2975 + err = swork_get();
2979 + INIT_SWORK(¬ify_work, __mce_notify_work);
2980 + notify_work_ready = true;
2984 +static void mce_notify_work(void)
2986 + if (notify_work_ready)
2987 + swork_queue(¬ify_work);
2990 +static void mce_notify_work(void)
2992 + __mce_notify_work(NULL);
2994 +static inline int mce_notify_work_init(void) { return 0; }
2998 * Notify the user(s) about new machine check events.
2999 * Can be called from interrupt context, but not from machine check/NMI
3000 @@ -1333,19 +1371,8 @@ static DECLARE_WORK(mce_trigger_work, mce_do_trigger);
3002 int mce_notify_irq(void)
3004 - /* Not more than two messages every minute */
3005 - static DEFINE_RATELIMIT_STATE(ratelimit, 60*HZ, 2);
3007 if (test_and_clear_bit(0, &mce_need_notify)) {
3008 - /* wake processes polling /dev/mcelog */
3009 - wake_up_interruptible(&mce_chrdev_wait);
3011 - if (mce_helper[0])
3012 - schedule_work(&mce_trigger_work);
3014 - if (__ratelimit(&ratelimit))
3015 - pr_info(HW_ERR "Machine check events logged\n");
3017 + mce_notify_work();
3021 @@ -1639,7 +1666,7 @@ static void __mcheck_cpu_clear_vendor(struct cpuinfo_x86 *c)
3025 -static void mce_start_timer(unsigned int cpu, struct timer_list *t)
3026 +static void mce_start_timer(unsigned int cpu, struct hrtimer *t)
3028 unsigned long iv = check_interval * HZ;
3030 @@ -1648,16 +1675,17 @@ static void mce_start_timer(unsigned int cpu, struct timer_list *t)
3032 per_cpu(mce_next_interval, cpu) = iv;
3034 - t->expires = round_jiffies(jiffies + iv);
3035 - add_timer_on(t, cpu);
3036 + hrtimer_start_range_ns(t, ns_to_ktime(jiffies_to_usecs(iv) * 1000ULL),
3037 + 0, HRTIMER_MODE_REL_PINNED);
3040 static void __mcheck_cpu_init_timer(void)
3042 - struct timer_list *t = this_cpu_ptr(&mce_timer);
3043 + struct hrtimer *t = this_cpu_ptr(&mce_timer);
3044 unsigned int cpu = smp_processor_id();
3046 - setup_timer(t, mce_timer_fn, cpu);
3047 + hrtimer_init(t, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
3048 + t->function = mce_timer_fn;
3049 mce_start_timer(cpu, t);
3052 @@ -2376,6 +2404,8 @@ static void mce_disable_cpu(void *h)
3053 if (!mce_available(raw_cpu_ptr(&cpu_info)))
3056 + hrtimer_cancel(this_cpu_ptr(&mce_timer));
3058 if (!(action & CPU_TASKS_FROZEN))
3061 @@ -2398,6 +2428,7 @@ static void mce_reenable_cpu(void *h)
3063 wrmsrl(MSR_IA32_MCx_CTL(i), b->ctl);
3065 + __mcheck_cpu_init_timer();
3068 /* Get notified when a cpu comes on/off. Be hotplug friendly. */
3069 @@ -2405,7 +2436,6 @@ static int
3070 mce_cpu_callback(struct notifier_block *nfb, unsigned long action, void *hcpu)
3072 unsigned int cpu = (unsigned long)hcpu;
3073 - struct timer_list *t = &per_cpu(mce_timer, cpu);
3075 switch (action & ~CPU_TASKS_FROZEN) {
3077 @@ -2425,11 +2455,9 @@ mce_cpu_callback(struct notifier_block *nfb, unsigned long action, void *hcpu)
3079 case CPU_DOWN_PREPARE:
3080 smp_call_function_single(cpu, mce_disable_cpu, &action, 1);
3081 - del_timer_sync(t);
3083 case CPU_DOWN_FAILED:
3084 smp_call_function_single(cpu, mce_reenable_cpu, &action, 1);
3085 - mce_start_timer(cpu, t);
3089 @@ -2468,6 +2496,10 @@ static __init int mcheck_init_device(void)
3093 + err = mce_notify_work_init();
3097 if (!zalloc_cpumask_var(&mce_device_initialized, GFP_KERNEL)) {
3100 diff --git a/arch/x86/kernel/cpu/perf_event_intel_rapl.c b/arch/x86/kernel/cpu/perf_event_intel_rapl.c
3101 index ed446bdcbf31..d2ac364e2118 100644
3102 --- a/arch/x86/kernel/cpu/perf_event_intel_rapl.c
3103 +++ b/arch/x86/kernel/cpu/perf_event_intel_rapl.c
3104 @@ -117,7 +117,7 @@ static struct perf_pmu_events_attr event_attr_##v = { \
3109 + raw_spinlock_t lock;
3110 int n_active; /* number of active events */
3111 struct list_head active_list;
3112 struct pmu *pmu; /* pointer to rapl_pmu_class */
3113 @@ -220,13 +220,13 @@ static enum hrtimer_restart rapl_hrtimer_handle(struct hrtimer *hrtimer)
3115 return HRTIMER_NORESTART;
3117 - spin_lock_irqsave(&pmu->lock, flags);
3118 + raw_spin_lock_irqsave(&pmu->lock, flags);
3120 list_for_each_entry(event, &pmu->active_list, active_entry) {
3121 rapl_event_update(event);
3124 - spin_unlock_irqrestore(&pmu->lock, flags);
3125 + raw_spin_unlock_irqrestore(&pmu->lock, flags);
3127 hrtimer_forward_now(hrtimer, pmu->timer_interval);
3129 @@ -263,9 +263,9 @@ static void rapl_pmu_event_start(struct perf_event *event, int mode)
3130 struct rapl_pmu *pmu = __this_cpu_read(rapl_pmu);
3131 unsigned long flags;
3133 - spin_lock_irqsave(&pmu->lock, flags);
3134 + raw_spin_lock_irqsave(&pmu->lock, flags);
3135 __rapl_pmu_event_start(pmu, event);
3136 - spin_unlock_irqrestore(&pmu->lock, flags);
3137 + raw_spin_unlock_irqrestore(&pmu->lock, flags);
3140 static void rapl_pmu_event_stop(struct perf_event *event, int mode)
3141 @@ -274,7 +274,7 @@ static void rapl_pmu_event_stop(struct perf_event *event, int mode)
3142 struct hw_perf_event *hwc = &event->hw;
3143 unsigned long flags;
3145 - spin_lock_irqsave(&pmu->lock, flags);
3146 + raw_spin_lock_irqsave(&pmu->lock, flags);
3148 /* mark event as deactivated and stopped */
3149 if (!(hwc->state & PERF_HES_STOPPED)) {
3150 @@ -299,7 +299,7 @@ static void rapl_pmu_event_stop(struct perf_event *event, int mode)
3151 hwc->state |= PERF_HES_UPTODATE;
3154 - spin_unlock_irqrestore(&pmu->lock, flags);
3155 + raw_spin_unlock_irqrestore(&pmu->lock, flags);
3158 static int rapl_pmu_event_add(struct perf_event *event, int mode)
3159 @@ -308,14 +308,14 @@ static int rapl_pmu_event_add(struct perf_event *event, int mode)
3160 struct hw_perf_event *hwc = &event->hw;
3161 unsigned long flags;
3163 - spin_lock_irqsave(&pmu->lock, flags);
3164 + raw_spin_lock_irqsave(&pmu->lock, flags);
3166 hwc->state = PERF_HES_UPTODATE | PERF_HES_STOPPED;
3168 if (mode & PERF_EF_START)
3169 __rapl_pmu_event_start(pmu, event);
3171 - spin_unlock_irqrestore(&pmu->lock, flags);
3172 + raw_spin_unlock_irqrestore(&pmu->lock, flags);
3176 @@ -603,7 +603,7 @@ static int rapl_cpu_prepare(int cpu)
3177 pmu = kzalloc_node(sizeof(*pmu), GFP_KERNEL, cpu_to_node(cpu));
3180 - spin_lock_init(&pmu->lock);
3181 + raw_spin_lock_init(&pmu->lock);
3183 INIT_LIST_HEAD(&pmu->active_list);
3185 diff --git a/arch/x86/kernel/dumpstack_32.c b/arch/x86/kernel/dumpstack_32.c
3186 index 464ffd69b92e..00db1aad1548 100644
3187 --- a/arch/x86/kernel/dumpstack_32.c
3188 +++ b/arch/x86/kernel/dumpstack_32.c
3189 @@ -42,7 +42,7 @@ void dump_trace(struct task_struct *task, struct pt_regs *regs,
3190 unsigned long *stack, unsigned long bp,
3191 const struct stacktrace_ops *ops, void *data)
3193 - const unsigned cpu = get_cpu();
3194 + const unsigned cpu = get_cpu_light();
3198 @@ -86,7 +86,7 @@ void dump_trace(struct task_struct *task, struct pt_regs *regs,
3200 touch_nmi_watchdog();
3205 EXPORT_SYMBOL(dump_trace);
3207 diff --git a/arch/x86/kernel/dumpstack_64.c b/arch/x86/kernel/dumpstack_64.c
3208 index 5f1c6266eb30..c331e3fef465 100644
3209 --- a/arch/x86/kernel/dumpstack_64.c
3210 +++ b/arch/x86/kernel/dumpstack_64.c
3211 @@ -152,7 +152,7 @@ void dump_trace(struct task_struct *task, struct pt_regs *regs,
3212 unsigned long *stack, unsigned long bp,
3213 const struct stacktrace_ops *ops, void *data)
3215 - const unsigned cpu = get_cpu();
3216 + const unsigned cpu = get_cpu_light();
3217 struct thread_info *tinfo;
3218 unsigned long *irq_stack = (unsigned long *)per_cpu(irq_stack_ptr, cpu);
3219 unsigned long dummy;
3220 @@ -241,7 +241,7 @@ void dump_trace(struct task_struct *task, struct pt_regs *regs,
3221 * This handles the process stack:
3223 bp = ops->walk_stack(tinfo, stack, bp, ops, data, NULL, &graph);
3227 EXPORT_SYMBOL(dump_trace);
3229 @@ -255,7 +255,7 @@ show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
3233 - preempt_disable();
3234 + migrate_disable();
3235 cpu = smp_processor_id();
3237 irq_stack_end = (unsigned long *)(per_cpu(irq_stack_ptr, cpu));
3238 @@ -291,7 +291,7 @@ show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
3239 pr_cont(" %016lx", *stack++);
3240 touch_nmi_watchdog();
3246 show_trace_log_lvl(task, regs, sp, bp, log_lvl);
3247 diff --git a/arch/x86/kernel/irq_32.c b/arch/x86/kernel/irq_32.c
3248 index 38da8f29a9c8..ce71f7098f15 100644
3249 --- a/arch/x86/kernel/irq_32.c
3250 +++ b/arch/x86/kernel/irq_32.c
3251 @@ -128,6 +128,7 @@ void irq_ctx_init(int cpu)
3252 cpu, per_cpu(hardirq_stack, cpu), per_cpu(softirq_stack, cpu));
3255 +#ifndef CONFIG_PREEMPT_RT_FULL
3256 void do_softirq_own_stack(void)
3258 struct thread_info *curstk;
3259 @@ -146,6 +147,7 @@ void do_softirq_own_stack(void)
3261 call_on_stack(__do_softirq, isp);
3265 bool handle_irq(struct irq_desc *desc, struct pt_regs *regs)
3267 diff --git a/arch/x86/kernel/kvm.c b/arch/x86/kernel/kvm.c
3268 index 47190bd399e7..807950860fb7 100644
3269 --- a/arch/x86/kernel/kvm.c
3270 +++ b/arch/x86/kernel/kvm.c
3272 #include <linux/kprobes.h>
3273 #include <linux/debugfs.h>
3274 #include <linux/nmi.h>
3275 +#include <linux/swait.h>
3276 #include <asm/timer.h>
3277 #include <asm/cpu.h>
3278 #include <asm/traps.h>
3279 @@ -91,14 +92,14 @@ static void kvm_io_delay(void)
3281 struct kvm_task_sleep_node {
3282 struct hlist_node link;
3283 - wait_queue_head_t wq;
3284 + struct swait_queue_head wq;
3290 static struct kvm_task_sleep_head {
3292 + raw_spinlock_t lock;
3293 struct hlist_head list;
3294 } async_pf_sleepers[KVM_TASK_SLEEP_HASHSIZE];
3296 @@ -122,17 +123,17 @@ void kvm_async_pf_task_wait(u32 token)
3297 u32 key = hash_32(token, KVM_TASK_SLEEP_HASHBITS);
3298 struct kvm_task_sleep_head *b = &async_pf_sleepers[key];
3299 struct kvm_task_sleep_node n, *e;
3300 - DEFINE_WAIT(wait);
3301 + DECLARE_SWAITQUEUE(wait);
3305 - spin_lock(&b->lock);
3306 + raw_spin_lock(&b->lock);
3307 e = _find_apf_task(b, token);
3309 /* dummy entry exist -> wake up was delivered ahead of PF */
3310 hlist_del(&e->link);
3312 - spin_unlock(&b->lock);
3313 + raw_spin_unlock(&b->lock);
3317 @@ -141,13 +142,13 @@ void kvm_async_pf_task_wait(u32 token)
3319 n.cpu = smp_processor_id();
3320 n.halted = is_idle_task(current) || preempt_count() > 1;
3321 - init_waitqueue_head(&n.wq);
3322 + init_swait_queue_head(&n.wq);
3323 hlist_add_head(&n.link, &b->list);
3324 - spin_unlock(&b->lock);
3325 + raw_spin_unlock(&b->lock);
3329 - prepare_to_wait(&n.wq, &wait, TASK_UNINTERRUPTIBLE);
3330 + prepare_to_swait(&n.wq, &wait, TASK_UNINTERRUPTIBLE);
3331 if (hlist_unhashed(&n.link))
3334 @@ -166,7 +167,7 @@ void kvm_async_pf_task_wait(u32 token)
3338 - finish_wait(&n.wq, &wait);
3339 + finish_swait(&n.wq, &wait);
3343 @@ -178,8 +179,8 @@ static void apf_task_wake_one(struct kvm_task_sleep_node *n)
3344 hlist_del_init(&n->link);
3346 smp_send_reschedule(n->cpu);
3347 - else if (waitqueue_active(&n->wq))
3349 + else if (swait_active(&n->wq))
3353 static void apf_task_wake_all(void)
3354 @@ -189,14 +190,14 @@ static void apf_task_wake_all(void)
3355 for (i = 0; i < KVM_TASK_SLEEP_HASHSIZE; i++) {
3356 struct hlist_node *p, *next;
3357 struct kvm_task_sleep_head *b = &async_pf_sleepers[i];
3358 - spin_lock(&b->lock);
3359 + raw_spin_lock(&b->lock);
3360 hlist_for_each_safe(p, next, &b->list) {
3361 struct kvm_task_sleep_node *n =
3362 hlist_entry(p, typeof(*n), link);
3363 if (n->cpu == smp_processor_id())
3364 apf_task_wake_one(n);
3366 - spin_unlock(&b->lock);
3367 + raw_spin_unlock(&b->lock);
3371 @@ -212,7 +213,7 @@ void kvm_async_pf_task_wake(u32 token)
3375 - spin_lock(&b->lock);
3376 + raw_spin_lock(&b->lock);
3377 n = _find_apf_task(b, token);
3380 @@ -225,17 +226,17 @@ again:
3381 * Allocation failed! Busy wait while other cpu
3384 - spin_unlock(&b->lock);
3385 + raw_spin_unlock(&b->lock);
3390 n->cpu = smp_processor_id();
3391 - init_waitqueue_head(&n->wq);
3392 + init_swait_queue_head(&n->wq);
3393 hlist_add_head(&n->link, &b->list);
3395 apf_task_wake_one(n);
3396 - spin_unlock(&b->lock);
3397 + raw_spin_unlock(&b->lock);
3400 EXPORT_SYMBOL_GPL(kvm_async_pf_task_wake);
3401 @@ -486,7 +487,7 @@ void __init kvm_guest_init(void)
3402 paravirt_ops_setup();
3403 register_reboot_notifier(&kvm_pv_reboot_nb);
3404 for (i = 0; i < KVM_TASK_SLEEP_HASHSIZE; i++)
3405 - spin_lock_init(&async_pf_sleepers[i].lock);
3406 + raw_spin_lock_init(&async_pf_sleepers[i].lock);
3407 if (kvm_para_has_feature(KVM_FEATURE_ASYNC_PF))
3408 x86_init.irqs.trap_init = kvm_apf_trap_init;
3410 diff --git a/arch/x86/kernel/nmi.c b/arch/x86/kernel/nmi.c
3411 index 697f90db0e37..424aec4a4c71 100644
3412 --- a/arch/x86/kernel/nmi.c
3413 +++ b/arch/x86/kernel/nmi.c
3414 @@ -231,7 +231,7 @@ pci_serr_error(unsigned char reason, struct pt_regs *regs)
3417 if (panic_on_unrecovered_nmi)
3418 - panic("NMI: Not continuing");
3419 + nmi_panic(regs, "NMI: Not continuing");
3421 pr_emerg("Dazed and confused, but trying to continue\n");
3423 @@ -255,8 +255,16 @@ io_check_error(unsigned char reason, struct pt_regs *regs)
3424 reason, smp_processor_id());
3427 - if (panic_on_io_nmi)
3428 - panic("NMI IOCK error: Not continuing");
3429 + if (panic_on_io_nmi) {
3430 + nmi_panic(regs, "NMI IOCK error: Not continuing");
3433 + * If we end up here, it means we have received an NMI while
3434 + * processing panic(). Simply return without delaying and
3435 + * re-enabling NMIs.
3440 /* Re-enable the IOCK line, wait for a few seconds */
3441 reason = (reason & NMI_REASON_CLEAR_MASK) | NMI_REASON_CLEAR_IOCHK;
3442 @@ -297,7 +305,7 @@ unknown_nmi_error(unsigned char reason, struct pt_regs *regs)
3444 pr_emerg("Do you have a strange power saving mode enabled?\n");
3445 if (unknown_nmi_panic || panic_on_unrecovered_nmi)
3446 - panic("NMI: Not continuing");
3447 + nmi_panic(regs, "NMI: Not continuing");
3449 pr_emerg("Dazed and confused, but trying to continue\n");
3451 diff --git a/arch/x86/kernel/process_32.c b/arch/x86/kernel/process_32.c
3452 index 9f950917528b..4dd4beae917a 100644
3453 --- a/arch/x86/kernel/process_32.c
3454 +++ b/arch/x86/kernel/process_32.c
3456 #include <linux/uaccess.h>
3457 #include <linux/io.h>
3458 #include <linux/kdebug.h>
3459 +#include <linux/highmem.h>
3461 #include <asm/pgtable.h>
3462 #include <asm/ldt.h>
3463 @@ -210,6 +211,35 @@ start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
3465 EXPORT_SYMBOL_GPL(start_thread);
3467 +#ifdef CONFIG_PREEMPT_RT_FULL
3468 +static void switch_kmaps(struct task_struct *prev_p, struct task_struct *next_p)
3473 + * Clear @prev's kmap_atomic mappings
3475 + for (i = 0; i < prev_p->kmap_idx; i++) {
3476 + int idx = i + KM_TYPE_NR * smp_processor_id();
3477 + pte_t *ptep = kmap_pte - idx;
3479 + kpte_clear_flush(ptep, __fix_to_virt(FIX_KMAP_BEGIN + idx));
3482 + * Restore @next_p's kmap_atomic mappings
3484 + for (i = 0; i < next_p->kmap_idx; i++) {
3485 + int idx = i + KM_TYPE_NR * smp_processor_id();
3487 + if (!pte_none(next_p->kmap_pte[i]))
3488 + set_pte(kmap_pte - idx, next_p->kmap_pte[i]);
3493 +switch_kmaps(struct task_struct *prev_p, struct task_struct *next_p) { }
3498 * switch_to(x,y) should switch tasks from x to y.
3499 @@ -286,6 +316,8 @@ __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
3500 task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT))
3501 __switch_to_xtra(prev_p, next_p, tss);
3503 + switch_kmaps(prev_p, next_p);
3506 * Leave lazy mode, flushing any hypercalls made here.
3507 * This must be done before restoring TLS segments so
3508 diff --git a/arch/x86/kernel/reboot.c b/arch/x86/kernel/reboot.c
3509 index f660d63f40fe..8384207adde2 100644
3510 --- a/arch/x86/kernel/reboot.c
3511 +++ b/arch/x86/kernel/reboot.c
3512 @@ -726,6 +726,7 @@ static int crashing_cpu;
3513 static nmi_shootdown_cb shootdown_callback;
3515 static atomic_t waiting_for_crash_ipi;
3516 +static int crash_ipi_issued;
3518 static int crash_nmi_callback(unsigned int val, struct pt_regs *regs)
3520 @@ -788,6 +789,9 @@ void nmi_shootdown_cpus(nmi_shootdown_cb callback)
3522 smp_send_nmi_allbutself();
3524 + /* Kick CPUs looping in NMI context. */
3525 + WRITE_ONCE(crash_ipi_issued, 1);
3527 msecs = 1000; /* Wait at most a second for the other cpus to stop */
3528 while ((atomic_read(&waiting_for_crash_ipi) > 0) && msecs) {
3530 @@ -796,6 +800,22 @@ void nmi_shootdown_cpus(nmi_shootdown_cb callback)
3532 /* Leave the nmi callback set */
3535 +/* Override the weak function in kernel/panic.c */
3536 +void nmi_panic_self_stop(struct pt_regs *regs)
3540 + * Wait for the crash dumping IPI to be issued, and then
3541 + * call its callback directly.
3543 + if (READ_ONCE(crash_ipi_issued))
3544 + crash_nmi_callback(0, regs); /* Don't return */
3550 #else /* !CONFIG_SMP */
3551 void nmi_shootdown_cpus(nmi_shootdown_cb callback)
3553 diff --git a/arch/x86/kvm/lapic.c b/arch/x86/kvm/lapic.c
3554 index 1c96f09367ae..ffb6d9859122 100644
3555 --- a/arch/x86/kvm/lapic.c
3556 +++ b/arch/x86/kvm/lapic.c
3557 @@ -1195,7 +1195,7 @@ static void apic_update_lvtt(struct kvm_lapic *apic)
3558 static void apic_timer_expired(struct kvm_lapic *apic)
3560 struct kvm_vcpu *vcpu = apic->vcpu;
3561 - wait_queue_head_t *q = &vcpu->wq;
3562 + struct swait_queue_head *q = &vcpu->wq;
3563 struct kvm_timer *ktimer = &apic->lapic_timer;
3565 if (atomic_read(&apic->lapic_timer.pending))
3566 @@ -1204,8 +1204,8 @@ static void apic_timer_expired(struct kvm_lapic *apic)
3567 atomic_inc(&apic->lapic_timer.pending);
3568 kvm_set_pending_timer(vcpu);
3570 - if (waitqueue_active(q))
3571 - wake_up_interruptible(q);
3572 + if (swait_active(q))
3575 if (apic_lvtt_tscdeadline(apic))
3576 ktimer->expired_tscdeadline = ktimer->tscdeadline;
3577 @@ -1801,6 +1801,7 @@ int kvm_create_lapic(struct kvm_vcpu *vcpu)
3578 hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
3580 apic->lapic_timer.timer.function = apic_timer_fn;
3581 + apic->lapic_timer.timer.irqsafe = 1;
3584 * APIC is created enabled. This will prevent kvm_lapic_set_base from
3585 diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
3586 index ae2b9cd358f2..50c9861d4219 100644
3587 --- a/arch/x86/kvm/x86.c
3588 +++ b/arch/x86/kvm/x86.c
3589 @@ -5827,6 +5827,13 @@ int kvm_arch_init(void *opaque)
3593 +#ifdef CONFIG_PREEMPT_RT_FULL
3594 + if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
3595 + printk(KERN_ERR "RT requires X86_FEATURE_CONSTANT_TSC\n");
3596 + return -EOPNOTSUPP;
3600 r = kvm_mmu_module_init();
3602 goto out_free_percpu;
3603 diff --git a/arch/x86/mm/highmem_32.c b/arch/x86/mm/highmem_32.c
3604 index a6d739258137..bd24ba1c4a86 100644
3605 --- a/arch/x86/mm/highmem_32.c
3606 +++ b/arch/x86/mm/highmem_32.c
3607 @@ -32,10 +32,11 @@ EXPORT_SYMBOL(kunmap);
3609 void *kmap_atomic_prot(struct page *page, pgprot_t prot)
3611 + pte_t pte = mk_pte(page, prot);
3612 unsigned long vaddr;
3615 - preempt_disable();
3616 + preempt_disable_nort();
3617 pagefault_disable();
3619 if (!PageHighMem(page))
3620 @@ -45,7 +46,10 @@ void *kmap_atomic_prot(struct page *page, pgprot_t prot)
3621 idx = type + KM_TYPE_NR*smp_processor_id();
3622 vaddr = __fix_to_virt(FIX_KMAP_BEGIN + idx);
3623 BUG_ON(!pte_none(*(kmap_pte-idx)));
3624 - set_pte(kmap_pte-idx, mk_pte(page, prot));
3625 +#ifdef CONFIG_PREEMPT_RT_FULL
3626 + current->kmap_pte[type] = pte;
3628 + set_pte(kmap_pte-idx, pte);
3629 arch_flush_lazy_mmu_mode();
3631 return (void *)vaddr;
3632 @@ -88,6 +92,9 @@ void __kunmap_atomic(void *kvaddr)
3633 * is a bad idea also, in case the page changes cacheability
3634 * attributes or becomes a protected page in a hypervisor.
3636 +#ifdef CONFIG_PREEMPT_RT_FULL
3637 + current->kmap_pte[type] = __pte(0);
3639 kpte_clear_flush(kmap_pte-idx, vaddr);
3640 kmap_atomic_idx_pop();
3641 arch_flush_lazy_mmu_mode();
3642 @@ -100,7 +107,7 @@ void __kunmap_atomic(void *kvaddr)
3647 + preempt_enable_nort();
3649 EXPORT_SYMBOL(__kunmap_atomic);
3651 diff --git a/arch/x86/mm/iomap_32.c b/arch/x86/mm/iomap_32.c
3652 index 9c0ff045fdd4..dd25dd1671b6 100644
3653 --- a/arch/x86/mm/iomap_32.c
3654 +++ b/arch/x86/mm/iomap_32.c
3655 @@ -56,6 +56,7 @@ EXPORT_SYMBOL_GPL(iomap_free);
3657 void *kmap_atomic_prot_pfn(unsigned long pfn, pgprot_t prot)
3659 + pte_t pte = pfn_pte(pfn, prot);
3660 unsigned long vaddr;
3663 @@ -65,7 +66,12 @@ void *kmap_atomic_prot_pfn(unsigned long pfn, pgprot_t prot)
3664 type = kmap_atomic_idx_push();
3665 idx = type + KM_TYPE_NR * smp_processor_id();
3666 vaddr = __fix_to_virt(FIX_KMAP_BEGIN + idx);
3667 - set_pte(kmap_pte - idx, pfn_pte(pfn, prot));
3668 + WARN_ON(!pte_none(*(kmap_pte - idx)));
3670 +#ifdef CONFIG_PREEMPT_RT_FULL
3671 + current->kmap_pte[type] = pte;
3673 + set_pte(kmap_pte - idx, pte);
3674 arch_flush_lazy_mmu_mode();
3676 return (void *)vaddr;
3677 @@ -113,6 +119,9 @@ iounmap_atomic(void __iomem *kvaddr)
3678 * is a bad idea also, in case the page changes cacheability
3679 * attributes or becomes a protected page in a hypervisor.
3681 +#ifdef CONFIG_PREEMPT_RT_FULL
3682 + current->kmap_pte[type] = __pte(0);
3684 kpte_clear_flush(kmap_pte-idx, vaddr);
3685 kmap_atomic_idx_pop();
3687 diff --git a/arch/x86/mm/pageattr.c b/arch/x86/mm/pageattr.c
3688 index b599a780a5a9..2e85c4117daf 100644
3689 --- a/arch/x86/mm/pageattr.c
3690 +++ b/arch/x86/mm/pageattr.c
3691 @@ -208,7 +208,15 @@ static void cpa_flush_array(unsigned long *start, int numpages, int cache,
3692 int in_flags, struct page **pages)
3694 unsigned int i, level;
3695 +#ifdef CONFIG_PREEMPT
3697 + * Avoid wbinvd() because it causes latencies on all CPUs,
3698 + * regardless of any CPU isolation that may be in effect.
3700 + unsigned long do_wbinvd = 0;
3702 unsigned long do_wbinvd = cache && numpages >= 1024; /* 4M threshold */
3705 BUG_ON(irqs_disabled());
3707 diff --git a/arch/x86/platform/uv/tlb_uv.c b/arch/x86/platform/uv/tlb_uv.c
3708 index 3b6ec42718e4..7871083de089 100644
3709 --- a/arch/x86/platform/uv/tlb_uv.c
3710 +++ b/arch/x86/platform/uv/tlb_uv.c
3711 @@ -714,9 +714,9 @@ static void destination_plugged(struct bau_desc *bau_desc,
3713 quiesce_local_uvhub(hmaster);
3715 - spin_lock(&hmaster->queue_lock);
3716 + raw_spin_lock(&hmaster->queue_lock);
3717 reset_with_ipi(&bau_desc->distribution, bcp);
3718 - spin_unlock(&hmaster->queue_lock);
3719 + raw_spin_unlock(&hmaster->queue_lock);
3721 end_uvhub_quiesce(hmaster);
3723 @@ -736,9 +736,9 @@ static void destination_timeout(struct bau_desc *bau_desc,
3725 quiesce_local_uvhub(hmaster);
3727 - spin_lock(&hmaster->queue_lock);
3728 + raw_spin_lock(&hmaster->queue_lock);
3729 reset_with_ipi(&bau_desc->distribution, bcp);
3730 - spin_unlock(&hmaster->queue_lock);
3731 + raw_spin_unlock(&hmaster->queue_lock);
3733 end_uvhub_quiesce(hmaster);
3735 @@ -759,7 +759,7 @@ static void disable_for_period(struct bau_control *bcp, struct ptc_stats *stat)
3738 hmaster = bcp->uvhub_master;
3739 - spin_lock(&hmaster->disable_lock);
3740 + raw_spin_lock(&hmaster->disable_lock);
3741 if (!bcp->baudisabled) {
3742 stat->s_bau_disabled++;
3744 @@ -772,7 +772,7 @@ static void disable_for_period(struct bau_control *bcp, struct ptc_stats *stat)
3748 - spin_unlock(&hmaster->disable_lock);
3749 + raw_spin_unlock(&hmaster->disable_lock);
3752 static void count_max_concurr(int stat, struct bau_control *bcp,
3753 @@ -835,7 +835,7 @@ static void record_send_stats(cycles_t time1, cycles_t time2,
3755 static void uv1_throttle(struct bau_control *hmaster, struct ptc_stats *stat)
3757 - spinlock_t *lock = &hmaster->uvhub_lock;
3758 + raw_spinlock_t *lock = &hmaster->uvhub_lock;
3761 v = &hmaster->active_descriptor_count;
3762 @@ -968,7 +968,7 @@ static int check_enable(struct bau_control *bcp, struct ptc_stats *stat)
3763 struct bau_control *hmaster;
3765 hmaster = bcp->uvhub_master;
3766 - spin_lock(&hmaster->disable_lock);
3767 + raw_spin_lock(&hmaster->disable_lock);
3768 if (bcp->baudisabled && (get_cycles() >= bcp->set_bau_on_time)) {
3769 stat->s_bau_reenabled++;
3770 for_each_present_cpu(tcpu) {
3771 @@ -980,10 +980,10 @@ static int check_enable(struct bau_control *bcp, struct ptc_stats *stat)
3772 tbcp->period_giveups = 0;
3775 - spin_unlock(&hmaster->disable_lock);
3776 + raw_spin_unlock(&hmaster->disable_lock);
3779 - spin_unlock(&hmaster->disable_lock);
3780 + raw_spin_unlock(&hmaster->disable_lock);
3784 @@ -1901,9 +1901,9 @@ static void __init init_per_cpu_tunables(void)
3785 bcp->cong_reps = congested_reps;
3786 bcp->disabled_period = sec_2_cycles(disabled_period);
3787 bcp->giveup_limit = giveup_limit;
3788 - spin_lock_init(&bcp->queue_lock);
3789 - spin_lock_init(&bcp->uvhub_lock);
3790 - spin_lock_init(&bcp->disable_lock);
3791 + raw_spin_lock_init(&bcp->queue_lock);
3792 + raw_spin_lock_init(&bcp->uvhub_lock);
3793 + raw_spin_lock_init(&bcp->disable_lock);
3797 diff --git a/arch/x86/platform/uv/uv_time.c b/arch/x86/platform/uv/uv_time.c
3798 index 2b158a9fa1d7..5e0b122620cb 100644
3799 --- a/arch/x86/platform/uv/uv_time.c
3800 +++ b/arch/x86/platform/uv/uv_time.c
3801 @@ -57,7 +57,7 @@ static DEFINE_PER_CPU(struct clock_event_device, cpu_ced);
3803 /* There is one of these allocated per node */
3804 struct uv_rtc_timer_head {
3806 + raw_spinlock_t lock;
3807 /* next cpu waiting for timer, local node relative: */
3809 /* number of cpus on this node: */
3810 @@ -177,7 +177,7 @@ static __init int uv_rtc_allocate_timers(void)
3811 uv_rtc_deallocate_timers();
3814 - spin_lock_init(&head->lock);
3815 + raw_spin_lock_init(&head->lock);
3816 head->ncpus = uv_blade_nr_possible_cpus(bid);
3817 head->next_cpu = -1;
3818 blade_info[bid] = head;
3819 @@ -231,7 +231,7 @@ static int uv_rtc_set_timer(int cpu, u64 expires)
3820 unsigned long flags;
3823 - spin_lock_irqsave(&head->lock, flags);
3824 + raw_spin_lock_irqsave(&head->lock, flags);
3826 next_cpu = head->next_cpu;
3828 @@ -243,12 +243,12 @@ static int uv_rtc_set_timer(int cpu, u64 expires)
3829 if (uv_setup_intr(cpu, expires)) {
3831 uv_rtc_find_next_timer(head, pnode);
3832 - spin_unlock_irqrestore(&head->lock, flags);
3833 + raw_spin_unlock_irqrestore(&head->lock, flags);
3838 - spin_unlock_irqrestore(&head->lock, flags);
3839 + raw_spin_unlock_irqrestore(&head->lock, flags);
3843 @@ -267,7 +267,7 @@ static int uv_rtc_unset_timer(int cpu, int force)
3844 unsigned long flags;
3847 - spin_lock_irqsave(&head->lock, flags);
3848 + raw_spin_lock_irqsave(&head->lock, flags);
3850 if ((head->next_cpu == bcpu && uv_read_rtc(NULL) >= *t) || force)
3852 @@ -279,7 +279,7 @@ static int uv_rtc_unset_timer(int cpu, int force)
3853 uv_rtc_find_next_timer(head, pnode);
3856 - spin_unlock_irqrestore(&head->lock, flags);
3857 + raw_spin_unlock_irqrestore(&head->lock, flags);
3861 @@ -299,13 +299,18 @@ static int uv_rtc_unset_timer(int cpu, int force)
3862 static cycle_t uv_read_rtc(struct clocksource *cs)
3864 unsigned long offset;
3867 + preempt_disable();
3868 if (uv_get_min_hub_revision_id() == 1)
3871 offset = (uv_blade_processor_id() * L1_CACHE_BYTES) % PAGE_SIZE;
3873 - return (cycle_t)uv_read_local_mmr(UVH_RTC | offset);
3874 + cycles = (cycle_t)uv_read_local_mmr(UVH_RTC | offset);
3881 diff --git a/block/blk-core.c b/block/blk-core.c
3882 index ef083e7a37c5..7a7926b1b4b5 100644
3883 --- a/block/blk-core.c
3884 +++ b/block/blk-core.c
3885 @@ -125,6 +125,9 @@ void blk_rq_init(struct request_queue *q, struct request *rq)
3887 INIT_LIST_HEAD(&rq->queuelist);
3888 INIT_LIST_HEAD(&rq->timeout_list);
3889 +#ifdef CONFIG_PREEMPT_RT_FULL
3890 + INIT_WORK(&rq->work, __blk_mq_complete_request_remote_work);
3894 rq->__sector = (sector_t) -1;
3895 @@ -233,7 +236,7 @@ EXPORT_SYMBOL(blk_start_queue_async);
3897 void blk_start_queue(struct request_queue *q)
3899 - WARN_ON(!irqs_disabled());
3900 + WARN_ON_NONRT(!irqs_disabled());
3902 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
3904 @@ -659,7 +662,7 @@ int blk_queue_enter(struct request_queue *q, gfp_t gfp)
3905 if (!gfpflags_allow_blocking(gfp))
3908 - ret = wait_event_interruptible(q->mq_freeze_wq,
3909 + ret = swait_event_interruptible(q->mq_freeze_wq,
3910 !atomic_read(&q->mq_freeze_depth) ||
3911 blk_queue_dying(q));
3912 if (blk_queue_dying(q))
3913 @@ -679,7 +682,7 @@ static void blk_queue_usage_counter_release(struct percpu_ref *ref)
3914 struct request_queue *q =
3915 container_of(ref, struct request_queue, q_usage_counter);
3917 - wake_up_all(&q->mq_freeze_wq);
3918 + swake_up_all(&q->mq_freeze_wq);
3921 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
3922 @@ -741,7 +744,7 @@ struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
3923 q->bypass_depth = 1;
3924 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
3926 - init_waitqueue_head(&q->mq_freeze_wq);
3927 + init_swait_queue_head(&q->mq_freeze_wq);
3930 * Init percpu_ref in atomic mode so that it's faster to shutdown.
3931 @@ -3222,7 +3225,7 @@ static void queue_unplugged(struct request_queue *q, unsigned int depth,
3932 blk_run_queue_async(q);
3935 - spin_unlock(q->queue_lock);
3936 + spin_unlock_irq(q->queue_lock);
3939 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
3940 @@ -3270,7 +3273,6 @@ EXPORT_SYMBOL(blk_check_plugged);
3941 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
3943 struct request_queue *q;
3944 - unsigned long flags;
3948 @@ -3290,11 +3292,6 @@ void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
3953 - * Save and disable interrupts here, to avoid doing it for every
3954 - * queue lock we have to take.
3956 - local_irq_save(flags);
3957 while (!list_empty(&list)) {
3958 rq = list_entry_rq(list.next);
3959 list_del_init(&rq->queuelist);
3960 @@ -3307,7 +3304,7 @@ void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
3961 queue_unplugged(q, depth, from_schedule);
3964 - spin_lock(q->queue_lock);
3965 + spin_lock_irq(q->queue_lock);
3969 @@ -3334,8 +3331,6 @@ void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
3972 queue_unplugged(q, depth, from_schedule);
3974 - local_irq_restore(flags);
3977 void blk_finish_plug(struct blk_plug *plug)
3978 diff --git a/block/blk-ioc.c b/block/blk-ioc.c
3979 index 381cb50a673c..dc8785233d94 100644
3980 --- a/block/blk-ioc.c
3981 +++ b/block/blk-ioc.c
3983 #include <linux/bio.h>
3984 #include <linux/blkdev.h>
3985 #include <linux/slab.h>
3986 +#include <linux/delay.h>
3990 @@ -109,7 +110,7 @@ static void ioc_release_fn(struct work_struct *work)
3991 spin_unlock(q->queue_lock);
3993 spin_unlock_irqrestore(&ioc->lock, flags);
3996 spin_lock_irqsave_nested(&ioc->lock, flags, 1);
3999 @@ -187,7 +188,7 @@ retry:
4000 spin_unlock(icq->q->queue_lock);
4002 spin_unlock_irqrestore(&ioc->lock, flags);
4008 diff --git a/block/blk-iopoll.c b/block/blk-iopoll.c
4009 index 0736729d6494..3e21e31d0d7e 100644
4010 --- a/block/blk-iopoll.c
4011 +++ b/block/blk-iopoll.c
4012 @@ -35,6 +35,7 @@ void blk_iopoll_sched(struct blk_iopoll *iop)
4013 list_add_tail(&iop->list, this_cpu_ptr(&blk_cpu_iopoll));
4014 __raise_softirq_irqoff(BLOCK_IOPOLL_SOFTIRQ);
4015 local_irq_restore(flags);
4016 + preempt_check_resched_rt();
4018 EXPORT_SYMBOL(blk_iopoll_sched);
4020 @@ -132,6 +133,7 @@ static void blk_iopoll_softirq(struct softirq_action *h)
4021 __raise_softirq_irqoff(BLOCK_IOPOLL_SOFTIRQ);
4024 + preempt_check_resched_rt();
4028 @@ -201,6 +203,7 @@ static int blk_iopoll_cpu_notify(struct notifier_block *self,
4029 this_cpu_ptr(&blk_cpu_iopoll));
4030 __raise_softirq_irqoff(BLOCK_IOPOLL_SOFTIRQ);
4032 + preempt_check_resched_rt();
4036 diff --git a/block/blk-mq-cpu.c b/block/blk-mq-cpu.c
4037 index bb3ed488f7b5..628c6c13c482 100644
4038 --- a/block/blk-mq-cpu.c
4039 +++ b/block/blk-mq-cpu.c
4043 static LIST_HEAD(blk_mq_cpu_notify_list);
4044 -static DEFINE_RAW_SPINLOCK(blk_mq_cpu_notify_lock);
4045 +static DEFINE_SPINLOCK(blk_mq_cpu_notify_lock);
4047 static int blk_mq_main_cpu_notify(struct notifier_block *self,
4048 unsigned long action, void *hcpu)
4049 @@ -25,7 +25,10 @@ static int blk_mq_main_cpu_notify(struct notifier_block *self,
4050 struct blk_mq_cpu_notifier *notify;
4051 int ret = NOTIFY_OK;
4053 - raw_spin_lock(&blk_mq_cpu_notify_lock);
4054 + if (action != CPU_POST_DEAD)
4057 + spin_lock(&blk_mq_cpu_notify_lock);
4059 list_for_each_entry(notify, &blk_mq_cpu_notify_list, list) {
4060 ret = notify->notify(notify->data, action, cpu);
4061 @@ -33,7 +36,7 @@ static int blk_mq_main_cpu_notify(struct notifier_block *self,
4065 - raw_spin_unlock(&blk_mq_cpu_notify_lock);
4066 + spin_unlock(&blk_mq_cpu_notify_lock);
4070 @@ -41,16 +44,16 @@ void blk_mq_register_cpu_notifier(struct blk_mq_cpu_notifier *notifier)
4072 BUG_ON(!notifier->notify);
4074 - raw_spin_lock(&blk_mq_cpu_notify_lock);
4075 + spin_lock(&blk_mq_cpu_notify_lock);
4076 list_add_tail(¬ifier->list, &blk_mq_cpu_notify_list);
4077 - raw_spin_unlock(&blk_mq_cpu_notify_lock);
4078 + spin_unlock(&blk_mq_cpu_notify_lock);
4081 void blk_mq_unregister_cpu_notifier(struct blk_mq_cpu_notifier *notifier)
4083 - raw_spin_lock(&blk_mq_cpu_notify_lock);
4084 + spin_lock(&blk_mq_cpu_notify_lock);
4085 list_del(¬ifier->list);
4086 - raw_spin_unlock(&blk_mq_cpu_notify_lock);
4087 + spin_unlock(&blk_mq_cpu_notify_lock);
4090 void blk_mq_init_cpu_notifier(struct blk_mq_cpu_notifier *notifier,
4091 diff --git a/block/blk-mq.c b/block/blk-mq.c
4092 index 0d1af3e44efb..e4fc80184dd8 100644
4093 --- a/block/blk-mq.c
4094 +++ b/block/blk-mq.c
4095 @@ -92,7 +92,7 @@ EXPORT_SYMBOL_GPL(blk_mq_freeze_queue_start);
4097 static void blk_mq_freeze_queue_wait(struct request_queue *q)
4099 - wait_event(q->mq_freeze_wq, percpu_ref_is_zero(&q->q_usage_counter));
4100 + swait_event(q->mq_freeze_wq, percpu_ref_is_zero(&q->q_usage_counter));
4104 @@ -130,7 +130,7 @@ void blk_mq_unfreeze_queue(struct request_queue *q)
4105 WARN_ON_ONCE(freeze_depth < 0);
4106 if (!freeze_depth) {
4107 percpu_ref_reinit(&q->q_usage_counter);
4108 - wake_up_all(&q->mq_freeze_wq);
4109 + swake_up_all(&q->mq_freeze_wq);
4112 EXPORT_SYMBOL_GPL(blk_mq_unfreeze_queue);
4113 @@ -149,7 +149,7 @@ void blk_mq_wake_waiters(struct request_queue *q)
4114 * dying, we need to ensure that processes currently waiting on
4115 * the queue are notified as well.
4117 - wake_up_all(&q->mq_freeze_wq);
4118 + swake_up_all(&q->mq_freeze_wq);
4121 bool blk_mq_can_queue(struct blk_mq_hw_ctx *hctx)
4122 @@ -196,6 +196,9 @@ static void blk_mq_rq_ctx_init(struct request_queue *q, struct blk_mq_ctx *ctx,
4126 +#ifdef CONFIG_PREEMPT_RT_FULL
4127 + INIT_WORK(&rq->work, __blk_mq_complete_request_remote_work);
4129 INIT_LIST_HEAD(&rq->timeout_list);
4132 @@ -325,6 +328,17 @@ void blk_mq_end_request(struct request *rq, int error)
4134 EXPORT_SYMBOL(blk_mq_end_request);
4136 +#ifdef CONFIG_PREEMPT_RT_FULL
4138 +void __blk_mq_complete_request_remote_work(struct work_struct *work)
4140 + struct request *rq = container_of(work, struct request, work);
4142 + rq->q->softirq_done_fn(rq);
4147 static void __blk_mq_complete_request_remote(void *data)
4149 struct request *rq = data;
4150 @@ -332,6 +346,8 @@ static void __blk_mq_complete_request_remote(void *data)
4151 rq->q->softirq_done_fn(rq);
4156 static void blk_mq_ipi_complete_request(struct request *rq)
4158 struct blk_mq_ctx *ctx = rq->mq_ctx;
4159 @@ -343,19 +359,23 @@ static void blk_mq_ipi_complete_request(struct request *rq)
4164 + cpu = get_cpu_light();
4165 if (!test_bit(QUEUE_FLAG_SAME_FORCE, &rq->q->queue_flags))
4166 shared = cpus_share_cache(cpu, ctx->cpu);
4168 if (cpu != ctx->cpu && !shared && cpu_online(ctx->cpu)) {
4169 +#ifdef CONFIG_PREEMPT_RT_FULL
4170 + schedule_work_on(ctx->cpu, &rq->work);
4172 rq->csd.func = __blk_mq_complete_request_remote;
4175 smp_call_function_single_async(ctx->cpu, &rq->csd);
4178 rq->q->softirq_done_fn(rq);
4184 static void __blk_mq_complete_request(struct request *rq)
4185 @@ -862,14 +882,14 @@ void blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx, bool async)
4189 - int cpu = get_cpu();
4190 + int cpu = get_cpu_light();
4191 if (cpumask_test_cpu(cpu, hctx->cpumask)) {
4192 __blk_mq_run_hw_queue(hctx);
4202 kblockd_schedule_delayed_work_on(blk_mq_hctx_next_cpu(hctx),
4203 @@ -1616,7 +1636,7 @@ static int blk_mq_hctx_notify(void *data, unsigned long action,
4205 struct blk_mq_hw_ctx *hctx = data;
4207 - if (action == CPU_DEAD || action == CPU_DEAD_FROZEN)
4208 + if (action == CPU_POST_DEAD)
4209 return blk_mq_hctx_cpu_offline(hctx, cpu);
4212 diff --git a/block/blk-mq.h b/block/blk-mq.h
4213 index 713820b47b31..3cb6feb4fe23 100644
4214 --- a/block/blk-mq.h
4215 +++ b/block/blk-mq.h
4216 @@ -74,7 +74,10 @@ struct blk_align_bitmap {
4217 static inline struct blk_mq_ctx *__blk_mq_get_ctx(struct request_queue *q,
4220 - return per_cpu_ptr(q->queue_ctx, cpu);
4221 + struct blk_mq_ctx *ctx;
4223 + ctx = per_cpu_ptr(q->queue_ctx, cpu);
4228 @@ -85,12 +88,12 @@ static inline struct blk_mq_ctx *__blk_mq_get_ctx(struct request_queue *q,
4230 static inline struct blk_mq_ctx *blk_mq_get_ctx(struct request_queue *q)
4232 - return __blk_mq_get_ctx(q, get_cpu());
4233 + return __blk_mq_get_ctx(q, get_cpu_light());
4236 static inline void blk_mq_put_ctx(struct blk_mq_ctx *ctx)
4242 struct blk_mq_alloc_data {
4243 diff --git a/block/blk-softirq.c b/block/blk-softirq.c
4244 index 53b1737e978d..81c3c0a62edf 100644
4245 --- a/block/blk-softirq.c
4246 +++ b/block/blk-softirq.c
4247 @@ -51,6 +51,7 @@ static void trigger_softirq(void *data)
4248 raise_softirq_irqoff(BLOCK_SOFTIRQ);
4250 local_irq_restore(flags);
4251 + preempt_check_resched_rt();
4255 @@ -93,6 +94,7 @@ static int blk_cpu_notify(struct notifier_block *self, unsigned long action,
4256 this_cpu_ptr(&blk_cpu_done));
4257 raise_softirq_irqoff(BLOCK_SOFTIRQ);
4259 + preempt_check_resched_rt();
4263 @@ -150,6 +152,7 @@ do_local:
4266 local_irq_restore(flags);
4267 + preempt_check_resched_rt();
4271 diff --git a/block/bounce.c b/block/bounce.c
4272 index 1cb5dd3a5da1..2f1ec8a67cbe 100644
4273 --- a/block/bounce.c
4274 +++ b/block/bounce.c
4275 @@ -55,11 +55,11 @@ static void bounce_copy_vec(struct bio_vec *to, unsigned char *vfrom)
4276 unsigned long flags;
4279 - local_irq_save(flags);
4280 + local_irq_save_nort(flags);
4281 vto = kmap_atomic(to->bv_page);
4282 memcpy(vto + to->bv_offset, vfrom, to->bv_len);
4284 - local_irq_restore(flags);
4285 + local_irq_restore_nort(flags);
4288 #else /* CONFIG_HIGHMEM */
4289 diff --git a/crypto/algapi.c b/crypto/algapi.c
4290 index 43f5bdb6b570..b54fd9187d12 100644
4291 --- a/crypto/algapi.c
4292 +++ b/crypto/algapi.c
4293 @@ -720,13 +720,13 @@ EXPORT_SYMBOL_GPL(crypto_spawn_tfm2);
4295 int crypto_register_notifier(struct notifier_block *nb)
4297 - return blocking_notifier_chain_register(&crypto_chain, nb);
4298 + return srcu_notifier_chain_register(&crypto_chain, nb);
4300 EXPORT_SYMBOL_GPL(crypto_register_notifier);
4302 int crypto_unregister_notifier(struct notifier_block *nb)
4304 - return blocking_notifier_chain_unregister(&crypto_chain, nb);
4305 + return srcu_notifier_chain_unregister(&crypto_chain, nb);
4307 EXPORT_SYMBOL_GPL(crypto_unregister_notifier);
4309 diff --git a/crypto/api.c b/crypto/api.c
4310 index bbc147cb5dec..bc1a848f02ec 100644
4313 @@ -31,7 +31,7 @@ EXPORT_SYMBOL_GPL(crypto_alg_list);
4314 DECLARE_RWSEM(crypto_alg_sem);
4315 EXPORT_SYMBOL_GPL(crypto_alg_sem);
4317 -BLOCKING_NOTIFIER_HEAD(crypto_chain);
4318 +SRCU_NOTIFIER_HEAD(crypto_chain);
4319 EXPORT_SYMBOL_GPL(crypto_chain);
4321 static struct crypto_alg *crypto_larval_wait(struct crypto_alg *alg);
4322 @@ -236,10 +236,10 @@ int crypto_probing_notify(unsigned long val, void *v)
4326 - ok = blocking_notifier_call_chain(&crypto_chain, val, v);
4327 + ok = srcu_notifier_call_chain(&crypto_chain, val, v);
4328 if (ok == NOTIFY_DONE) {
4329 request_module("cryptomgr");
4330 - ok = blocking_notifier_call_chain(&crypto_chain, val, v);
4331 + ok = srcu_notifier_call_chain(&crypto_chain, val, v);
4335 diff --git a/crypto/internal.h b/crypto/internal.h
4336 index 00e42a3ed814..2e85551e235f 100644
4337 --- a/crypto/internal.h
4338 +++ b/crypto/internal.h
4339 @@ -47,7 +47,7 @@ struct crypto_larval {
4341 extern struct list_head crypto_alg_list;
4342 extern struct rw_semaphore crypto_alg_sem;
4343 -extern struct blocking_notifier_head crypto_chain;
4344 +extern struct srcu_notifier_head crypto_chain;
4346 #ifdef CONFIG_PROC_FS
4347 void __init crypto_init_proc(void);
4348 @@ -143,7 +143,7 @@ static inline int crypto_is_moribund(struct crypto_alg *alg)
4350 static inline void crypto_notify(unsigned long val, void *v)
4352 - blocking_notifier_call_chain(&crypto_chain, val, v);
4353 + srcu_notifier_call_chain(&crypto_chain, val, v);
4356 #endif /* _CRYPTO_INTERNAL_H */
4357 diff --git a/drivers/acpi/acpica/acglobal.h b/drivers/acpi/acpica/acglobal.h
4358 index faa97604d878..941497f31cf0 100644
4359 --- a/drivers/acpi/acpica/acglobal.h
4360 +++ b/drivers/acpi/acpica/acglobal.h
4361 @@ -116,7 +116,7 @@ ACPI_GLOBAL(u8, acpi_gbl_global_lock_pending);
4364 ACPI_GLOBAL(acpi_spinlock, acpi_gbl_gpe_lock); /* For GPE data structs and registers */
4365 -ACPI_GLOBAL(acpi_spinlock, acpi_gbl_hardware_lock); /* For ACPI H/W except GPE registers */
4366 +ACPI_GLOBAL(acpi_raw_spinlock, acpi_gbl_hardware_lock); /* For ACPI H/W except GPE registers */
4367 ACPI_GLOBAL(acpi_spinlock, acpi_gbl_reference_count_lock);
4369 /* Mutex for _OSI support */
4370 diff --git a/drivers/acpi/acpica/hwregs.c b/drivers/acpi/acpica/hwregs.c
4371 index 3cf77afd142c..dc32e72132f1 100644
4372 --- a/drivers/acpi/acpica/hwregs.c
4373 +++ b/drivers/acpi/acpica/hwregs.c
4374 @@ -269,14 +269,14 @@ acpi_status acpi_hw_clear_acpi_status(void)
4375 ACPI_BITMASK_ALL_FIXED_STATUS,
4376 ACPI_FORMAT_UINT64(acpi_gbl_xpm1a_status.address)));
4378 - lock_flags = acpi_os_acquire_lock(acpi_gbl_hardware_lock);
4379 + raw_spin_lock_irqsave(acpi_gbl_hardware_lock, lock_flags);
4381 /* Clear the fixed events in PM1 A/B */
4383 status = acpi_hw_register_write(ACPI_REGISTER_PM1_STATUS,
4384 ACPI_BITMASK_ALL_FIXED_STATUS);
4386 - acpi_os_release_lock(acpi_gbl_hardware_lock, lock_flags);
4387 + raw_spin_unlock_irqrestore(acpi_gbl_hardware_lock, lock_flags);
4389 if (ACPI_FAILURE(status)) {
4391 diff --git a/drivers/acpi/acpica/hwxface.c b/drivers/acpi/acpica/hwxface.c
4392 index b2e50d8007fe..ff007084dc48 100644
4393 --- a/drivers/acpi/acpica/hwxface.c
4394 +++ b/drivers/acpi/acpica/hwxface.c
4395 @@ -374,7 +374,7 @@ acpi_status acpi_write_bit_register(u32 register_id, u32 value)
4396 return_ACPI_STATUS(AE_BAD_PARAMETER);
4399 - lock_flags = acpi_os_acquire_lock(acpi_gbl_hardware_lock);
4400 + raw_spin_lock_irqsave(acpi_gbl_hardware_lock, lock_flags);
4403 * At this point, we know that the parent register is one of the
4404 @@ -435,7 +435,7 @@ acpi_status acpi_write_bit_register(u32 register_id, u32 value)
4408 - acpi_os_release_lock(acpi_gbl_hardware_lock, lock_flags);
4409 + raw_spin_unlock_irqrestore(acpi_gbl_hardware_lock, lock_flags);
4410 return_ACPI_STATUS(status);
4413 diff --git a/drivers/acpi/acpica/utmutex.c b/drivers/acpi/acpica/utmutex.c
4414 index ce406e39b669..41a75eb3ae9d 100644
4415 --- a/drivers/acpi/acpica/utmutex.c
4416 +++ b/drivers/acpi/acpica/utmutex.c
4417 @@ -88,7 +88,7 @@ acpi_status acpi_ut_mutex_initialize(void)
4418 return_ACPI_STATUS (status);
4421 - status = acpi_os_create_lock (&acpi_gbl_hardware_lock);
4422 + status = acpi_os_create_raw_lock (&acpi_gbl_hardware_lock);
4423 if (ACPI_FAILURE (status)) {
4424 return_ACPI_STATUS (status);
4426 @@ -156,7 +156,7 @@ void acpi_ut_mutex_terminate(void)
4427 /* Delete the spinlocks */
4429 acpi_os_delete_lock(acpi_gbl_gpe_lock);
4430 - acpi_os_delete_lock(acpi_gbl_hardware_lock);
4431 + acpi_os_delete_raw_lock(acpi_gbl_hardware_lock);
4432 acpi_os_delete_lock(acpi_gbl_reference_count_lock);
4434 /* Delete the reader/writer lock */
4435 diff --git a/drivers/ata/libata-sff.c b/drivers/ata/libata-sff.c
4436 index 7dbba387d12a..65beb7abb4e7 100644
4437 --- a/drivers/ata/libata-sff.c
4438 +++ b/drivers/ata/libata-sff.c
4439 @@ -678,9 +678,9 @@ unsigned int ata_sff_data_xfer_noirq(struct ata_device *dev, unsigned char *buf,
4440 unsigned long flags;
4441 unsigned int consumed;
4443 - local_irq_save(flags);
4444 + local_irq_save_nort(flags);
4445 consumed = ata_sff_data_xfer32(dev, buf, buflen, rw);
4446 - local_irq_restore(flags);
4447 + local_irq_restore_nort(flags);
4451 @@ -719,7 +719,7 @@ static void ata_pio_sector(struct ata_queued_cmd *qc)
4452 unsigned long flags;
4454 /* FIXME: use a bounce buffer */
4455 - local_irq_save(flags);
4456 + local_irq_save_nort(flags);
4457 buf = kmap_atomic(page);
4459 /* do the actual data transfer */
4460 @@ -727,7 +727,7 @@ static void ata_pio_sector(struct ata_queued_cmd *qc)
4464 - local_irq_restore(flags);
4465 + local_irq_restore_nort(flags);
4467 buf = page_address(page);
4468 ap->ops->sff_data_xfer(qc->dev, buf + offset, qc->sect_size,
4469 @@ -864,7 +864,7 @@ next_sg:
4470 unsigned long flags;
4472 /* FIXME: use bounce buffer */
4473 - local_irq_save(flags);
4474 + local_irq_save_nort(flags);
4475 buf = kmap_atomic(page);
4477 /* do the actual data transfer */
4478 @@ -872,7 +872,7 @@ next_sg:
4482 - local_irq_restore(flags);
4483 + local_irq_restore_nort(flags);
4485 buf = page_address(page);
4486 consumed = ap->ops->sff_data_xfer(dev, buf + offset,
4487 diff --git a/drivers/block/zram/zram_drv.c b/drivers/block/zram/zram_drv.c
4488 index 62a93b685c54..5ad3c4e652a8 100644
4489 --- a/drivers/block/zram/zram_drv.c
4490 +++ b/drivers/block/zram/zram_drv.c
4491 @@ -520,6 +520,8 @@ static struct zram_meta *zram_meta_alloc(char *pool_name, u64 disksize)
4495 + zram_meta_init_table_locks(meta, disksize);
4500 @@ -568,12 +570,12 @@ static int zram_decompress_page(struct zram *zram, char *mem, u32 index)
4501 unsigned long handle;
4504 - bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
4505 + zram_lock_table(&meta->table[index]);
4506 handle = meta->table[index].handle;
4507 size = zram_get_obj_size(meta, index);
4509 if (!handle || zram_test_flag(meta, index, ZRAM_ZERO)) {
4510 - bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
4511 + zram_unlock_table(&meta->table[index]);
4512 memset(mem, 0, PAGE_SIZE);
4515 @@ -584,7 +586,7 @@ static int zram_decompress_page(struct zram *zram, char *mem, u32 index)
4517 ret = zcomp_decompress(zram->comp, cmem, size, mem);
4518 zs_unmap_object(meta->mem_pool, handle);
4519 - bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
4520 + zram_unlock_table(&meta->table[index]);
4522 /* Should NEVER happen. Return bio error if it does. */
4523 if (unlikely(ret)) {
4524 @@ -604,14 +606,14 @@ static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
4525 struct zram_meta *meta = zram->meta;
4526 page = bvec->bv_page;
4528 - bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
4529 + zram_lock_table(&meta->table[index]);
4530 if (unlikely(!meta->table[index].handle) ||
4531 zram_test_flag(meta, index, ZRAM_ZERO)) {
4532 - bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
4533 + zram_unlock_table(&meta->table[index]);
4534 handle_zero_page(bvec);
4537 - bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
4538 + zram_unlock_table(&meta->table[index]);
4540 if (is_partial_io(bvec))
4541 /* Use a temporary buffer to decompress the page */
4542 @@ -689,10 +691,10 @@ static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index,
4544 kunmap_atomic(user_mem);
4545 /* Free memory associated with this sector now. */
4546 - bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
4547 + zram_lock_table(&meta->table[index]);
4548 zram_free_page(zram, index);
4549 zram_set_flag(meta, index, ZRAM_ZERO);
4550 - bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
4551 + zram_unlock_table(&meta->table[index]);
4553 atomic64_inc(&zram->stats.zero_pages);
4555 @@ -752,12 +754,12 @@ static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index,
4556 * Free memory associated with this sector
4557 * before overwriting unused sectors.
4559 - bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
4560 + zram_lock_table(&meta->table[index]);
4561 zram_free_page(zram, index);
4563 meta->table[index].handle = handle;
4564 zram_set_obj_size(meta, index, clen);
4565 - bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
4566 + zram_unlock_table(&meta->table[index]);
4569 atomic64_add(clen, &zram->stats.compr_data_size);
4570 @@ -800,9 +802,9 @@ static void zram_bio_discard(struct zram *zram, u32 index,
4573 while (n >= PAGE_SIZE) {
4574 - bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
4575 + zram_lock_table(&meta->table[index]);
4576 zram_free_page(zram, index);
4577 - bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
4578 + zram_unlock_table(&meta->table[index]);
4579 atomic64_inc(&zram->stats.notify_free);
4582 @@ -928,9 +930,9 @@ static void zram_slot_free_notify(struct block_device *bdev,
4583 zram = bdev->bd_disk->private_data;
4586 - bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
4587 + zram_lock_table(&meta->table[index]);
4588 zram_free_page(zram, index);
4589 - bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
4590 + zram_unlock_table(&meta->table[index]);
4591 atomic64_inc(&zram->stats.notify_free);
4594 diff --git a/drivers/block/zram/zram_drv.h b/drivers/block/zram/zram_drv.h
4595 index 8e92339686d7..9e3e953d680e 100644
4596 --- a/drivers/block/zram/zram_drv.h
4597 +++ b/drivers/block/zram/zram_drv.h
4598 @@ -72,6 +72,9 @@ enum zram_pageflags {
4599 struct zram_table_entry {
4600 unsigned long handle;
4601 unsigned long value;
4602 +#ifdef CONFIG_PREEMPT_RT_BASE
4608 @@ -119,4 +122,42 @@ struct zram {
4610 bool claim; /* Protected by bdev->bd_mutex */
4613 +#ifndef CONFIG_PREEMPT_RT_BASE
4614 +static inline void zram_lock_table(struct zram_table_entry *table)
4616 + bit_spin_lock(ZRAM_ACCESS, &table->value);
4619 +static inline void zram_unlock_table(struct zram_table_entry *table)
4621 + bit_spin_unlock(ZRAM_ACCESS, &table->value);
4624 +static inline void zram_meta_init_table_locks(struct zram_meta *meta, u64 disksize) { }
4625 +#else /* CONFIG_PREEMPT_RT_BASE */
4626 +static inline void zram_lock_table(struct zram_table_entry *table)
4628 + spin_lock(&table->lock);
4629 + __set_bit(ZRAM_ACCESS, &table->value);
4632 +static inline void zram_unlock_table(struct zram_table_entry *table)
4634 + __clear_bit(ZRAM_ACCESS, &table->value);
4635 + spin_unlock(&table->lock);
4638 +static inline void zram_meta_init_table_locks(struct zram_meta *meta, u64 disksize)
4640 + size_t num_pages = disksize >> PAGE_SHIFT;
4643 + for (index = 0; index < num_pages; index++) {
4644 + spinlock_t *lock = &meta->table[index].lock;
4645 + spin_lock_init(lock);
4648 +#endif /* CONFIG_PREEMPT_RT_BASE */
4651 diff --git a/drivers/char/random.c b/drivers/char/random.c
4652 index 491a4dce13fe..cf69b6b42208 100644
4653 --- a/drivers/char/random.c
4654 +++ b/drivers/char/random.c
4655 @@ -799,8 +799,6 @@ static void add_timer_randomness(struct timer_rand_state *state, unsigned num)
4657 long delta, delta2, delta3;
4659 - preempt_disable();
4661 sample.jiffies = jiffies;
4662 sample.cycles = random_get_entropy();
4664 @@ -841,7 +839,6 @@ static void add_timer_randomness(struct timer_rand_state *state, unsigned num)
4666 credit_entropy_bits(r, min_t(int, fls(delta>>1), 11));
4671 void add_input_randomness(unsigned int type, unsigned int code,
4672 @@ -894,28 +891,27 @@ static __u32 get_reg(struct fast_pool *f, struct pt_regs *regs)
4673 return *(ptr + f->reg_idx++);
4676 -void add_interrupt_randomness(int irq, int irq_flags)
4677 +void add_interrupt_randomness(int irq, int irq_flags, __u64 ip)
4679 struct entropy_store *r;
4680 struct fast_pool *fast_pool = this_cpu_ptr(&irq_randomness);
4681 - struct pt_regs *regs = get_irq_regs();
4682 unsigned long now = jiffies;
4683 cycles_t cycles = random_get_entropy();
4684 __u32 c_high, j_high;
4690 - cycles = get_reg(fast_pool, regs);
4691 + cycles = get_reg(fast_pool, NULL);
4692 c_high = (sizeof(cycles) > 4) ? cycles >> 32 : 0;
4693 j_high = (sizeof(now) > 4) ? now >> 32 : 0;
4694 fast_pool->pool[0] ^= cycles ^ j_high ^ irq;
4695 fast_pool->pool[1] ^= now ^ c_high;
4696 - ip = regs ? instruction_pointer(regs) : _RET_IP_;
4699 fast_pool->pool[2] ^= ip;
4700 fast_pool->pool[3] ^= (sizeof(ip) > 4) ? ip >> 32 :
4701 - get_reg(fast_pool, regs);
4702 + get_reg(fast_pool, NULL);
4704 fast_mix(fast_pool);
4705 add_interrupt_bench(cycles);
4706 diff --git a/drivers/clk/at91/clk-generated.c b/drivers/clk/at91/clk-generated.c
4707 index abc80949e1dd..4ad3298eb372 100644
4708 --- a/drivers/clk/at91/clk-generated.c
4709 +++ b/drivers/clk/at91/clk-generated.c
4711 #include <linux/clkdev.h>
4712 #include <linux/clk/at91_pmc.h>
4713 #include <linux/of.h>
4714 -#include <linux/of_address.h>
4715 -#include <linux/io.h>
4716 +#include <linux/mfd/syscon.h>
4717 +#include <linux/regmap.h>
4723 struct clk_generated {
4725 - struct at91_pmc *pmc;
4726 + struct regmap *regmap;
4727 struct clk_range range;
4732 @@ -41,49 +42,52 @@ struct clk_generated {
4733 static int clk_generated_enable(struct clk_hw *hw)
4735 struct clk_generated *gck = to_clk_generated(hw);
4736 - struct at91_pmc *pmc = gck->pmc;
4738 + unsigned long flags;
4740 pr_debug("GCLK: %s, gckdiv = %d, parent id = %d\n",
4741 __func__, gck->gckdiv, gck->parent_id);
4744 - pmc_write(pmc, AT91_PMC_PCR, (gck->id & AT91_PMC_PCR_PID_MASK));
4745 - tmp = pmc_read(pmc, AT91_PMC_PCR) &
4746 - ~(AT91_PMC_PCR_GCKDIV_MASK | AT91_PMC_PCR_GCKCSS_MASK);
4747 - pmc_write(pmc, AT91_PMC_PCR, tmp | AT91_PMC_PCR_GCKCSS(gck->parent_id)
4748 - | AT91_PMC_PCR_CMD
4749 - | AT91_PMC_PCR_GCKDIV(gck->gckdiv)
4750 - | AT91_PMC_PCR_GCKEN);
4752 + spin_lock_irqsave(gck->lock, flags);
4753 + regmap_write(gck->regmap, AT91_PMC_PCR,
4754 + (gck->id & AT91_PMC_PCR_PID_MASK));
4755 + regmap_update_bits(gck->regmap, AT91_PMC_PCR,
4756 + AT91_PMC_PCR_GCKDIV_MASK | AT91_PMC_PCR_GCKCSS_MASK |
4757 + AT91_PMC_PCR_CMD | AT91_PMC_PCR_GCKEN,
4758 + AT91_PMC_PCR_GCKCSS(gck->parent_id) |
4759 + AT91_PMC_PCR_CMD |
4760 + AT91_PMC_PCR_GCKDIV(gck->gckdiv) |
4761 + AT91_PMC_PCR_GCKEN);
4762 + spin_unlock_irqrestore(gck->lock, flags);
4766 static void clk_generated_disable(struct clk_hw *hw)
4768 struct clk_generated *gck = to_clk_generated(hw);
4769 - struct at91_pmc *pmc = gck->pmc;
4773 - pmc_write(pmc, AT91_PMC_PCR, (gck->id & AT91_PMC_PCR_PID_MASK));
4774 - tmp = pmc_read(pmc, AT91_PMC_PCR) & ~AT91_PMC_PCR_GCKEN;
4775 - pmc_write(pmc, AT91_PMC_PCR, tmp | AT91_PMC_PCR_CMD);
4777 + unsigned long flags;
4779 + spin_lock_irqsave(gck->lock, flags);
4780 + regmap_write(gck->regmap, AT91_PMC_PCR,
4781 + (gck->id & AT91_PMC_PCR_PID_MASK));
4782 + regmap_update_bits(gck->regmap, AT91_PMC_PCR,
4783 + AT91_PMC_PCR_CMD | AT91_PMC_PCR_GCKEN,
4784 + AT91_PMC_PCR_CMD);
4785 + spin_unlock_irqrestore(gck->lock, flags);
4788 static int clk_generated_is_enabled(struct clk_hw *hw)
4790 struct clk_generated *gck = to_clk_generated(hw);
4791 - struct at91_pmc *pmc = gck->pmc;
4793 + unsigned long flags;
4794 + unsigned int status;
4797 - pmc_write(pmc, AT91_PMC_PCR, (gck->id & AT91_PMC_PCR_PID_MASK));
4798 - ret = !!(pmc_read(pmc, AT91_PMC_PCR) & AT91_PMC_PCR_GCKEN);
4800 + spin_lock_irqsave(gck->lock, flags);
4801 + regmap_write(gck->regmap, AT91_PMC_PCR,
4802 + (gck->id & AT91_PMC_PCR_PID_MASK));
4803 + regmap_read(gck->regmap, AT91_PMC_PCR, &status);
4804 + spin_unlock_irqrestore(gck->lock, flags);
4807 + return status & AT91_PMC_PCR_GCKEN ? 1 : 0;
4810 static unsigned long
4811 @@ -214,13 +218,14 @@ static const struct clk_ops generated_ops = {
4813 static void clk_generated_startup(struct clk_generated *gck)
4815 - struct at91_pmc *pmc = gck->pmc;
4817 + unsigned long flags;
4820 - pmc_write(pmc, AT91_PMC_PCR, (gck->id & AT91_PMC_PCR_PID_MASK));
4821 - tmp = pmc_read(pmc, AT91_PMC_PCR);
4823 + spin_lock_irqsave(gck->lock, flags);
4824 + regmap_write(gck->regmap, AT91_PMC_PCR,
4825 + (gck->id & AT91_PMC_PCR_PID_MASK));
4826 + regmap_read(gck->regmap, AT91_PMC_PCR, &tmp);
4827 + spin_unlock_irqrestore(gck->lock, flags);
4829 gck->parent_id = (tmp & AT91_PMC_PCR_GCKCSS_MASK)
4830 >> AT91_PMC_PCR_GCKCSS_OFFSET;
4831 @@ -229,8 +234,8 @@ static void clk_generated_startup(struct clk_generated *gck)
4834 static struct clk * __init
4835 -at91_clk_register_generated(struct at91_pmc *pmc, const char *name,
4836 - const char **parent_names, u8 num_parents,
4837 +at91_clk_register_generated(struct regmap *regmap, spinlock_t *lock, const char
4838 + *name, const char **parent_names, u8 num_parents,
4839 u8 id, const struct clk_range *range)
4841 struct clk_generated *gck;
4842 @@ -249,7 +254,8 @@ at91_clk_register_generated(struct at91_pmc *pmc, const char *name,
4845 gck->hw.init = &init;
4847 + gck->regmap = regmap;
4849 gck->range = *range;
4851 clk = clk_register(NULL, &gck->hw);
4852 @@ -261,8 +267,7 @@ at91_clk_register_generated(struct at91_pmc *pmc, const char *name,
4856 -void __init of_sama5d2_clk_generated_setup(struct device_node *np,
4857 - struct at91_pmc *pmc)
4858 +void __init of_sama5d2_clk_generated_setup(struct device_node *np)
4862 @@ -272,6 +277,7 @@ void __init of_sama5d2_clk_generated_setup(struct device_node *np,
4863 const char *parent_names[GENERATED_SOURCE_MAX];
4864 struct device_node *gcknp;
4865 struct clk_range range = CLK_RANGE(0, 0);
4866 + struct regmap *regmap;
4868 num_parents = of_clk_get_parent_count(np);
4869 if (num_parents <= 0 || num_parents > GENERATED_SOURCE_MAX)
4870 @@ -283,6 +289,10 @@ void __init of_sama5d2_clk_generated_setup(struct device_node *np,
4871 if (!num || num > PERIPHERAL_MAX)
4874 + regmap = syscon_node_to_regmap(of_get_parent(np));
4875 + if (IS_ERR(regmap))
4878 for_each_child_of_node(np, gcknp) {
4879 if (of_property_read_u32(gcknp, "reg", &id))
4881 @@ -296,11 +306,14 @@ void __init of_sama5d2_clk_generated_setup(struct device_node *np,
4882 of_at91_get_clk_range(gcknp, "atmel,clk-output-range",
4885 - clk = at91_clk_register_generated(pmc, name, parent_names,
4886 - num_parents, id, &range);
4887 + clk = at91_clk_register_generated(regmap, &pmc_pcr_lock, name,
4888 + parent_names, num_parents,
4893 of_clk_add_provider(gcknp, of_clk_src_simple_get, clk);
4896 +CLK_OF_DECLARE(of_sama5d2_clk_generated_setup, "atmel,sama5d2-clk-generated",
4897 + of_sama5d2_clk_generated_setup);
4898 diff --git a/drivers/clk/at91/clk-h32mx.c b/drivers/clk/at91/clk-h32mx.c
4899 index a165230e7eda..8e20c8a76db7 100644
4900 --- a/drivers/clk/at91/clk-h32mx.c
4901 +++ b/drivers/clk/at91/clk-h32mx.c
4903 #include <linux/clk-provider.h>
4904 #include <linux/clkdev.h>
4905 #include <linux/clk/at91_pmc.h>
4906 -#include <linux/delay.h>
4907 #include <linux/of.h>
4908 -#include <linux/of_address.h>
4909 -#include <linux/of_irq.h>
4910 -#include <linux/io.h>
4911 -#include <linux/interrupt.h>
4912 -#include <linux/irq.h>
4913 -#include <linux/sched.h>
4914 -#include <linux/wait.h>
4915 +#include <linux/regmap.h>
4916 +#include <linux/mfd/syscon.h>
4922 struct clk_sama5d4_h32mx {
4924 - struct at91_pmc *pmc;
4925 + struct regmap *regmap;
4928 #define to_clk_sama5d4_h32mx(hw) container_of(hw, struct clk_sama5d4_h32mx, hw)
4929 @@ -40,8 +34,10 @@ static unsigned long clk_sama5d4_h32mx_recalc_rate(struct clk_hw *hw,
4930 unsigned long parent_rate)
4932 struct clk_sama5d4_h32mx *h32mxclk = to_clk_sama5d4_h32mx(hw);
4933 + unsigned int mckr;
4935 - if (pmc_read(h32mxclk->pmc, AT91_PMC_MCKR) & AT91_PMC_H32MXDIV)
4936 + regmap_read(h32mxclk->regmap, AT91_PMC_MCKR, &mckr);
4937 + if (mckr & AT91_PMC_H32MXDIV)
4938 return parent_rate / 2;
4940 if (parent_rate > H32MX_MAX_FREQ)
4941 @@ -70,18 +66,16 @@ static int clk_sama5d4_h32mx_set_rate(struct clk_hw *hw, unsigned long rate,
4942 unsigned long parent_rate)
4944 struct clk_sama5d4_h32mx *h32mxclk = to_clk_sama5d4_h32mx(hw);
4945 - struct at91_pmc *pmc = h32mxclk->pmc;
4949 if (parent_rate != rate && (parent_rate / 2) != rate)
4953 - tmp = pmc_read(pmc, AT91_PMC_MCKR) & ~AT91_PMC_H32MXDIV;
4954 if ((parent_rate / 2) == rate)
4955 - tmp |= AT91_PMC_H32MXDIV;
4956 - pmc_write(pmc, AT91_PMC_MCKR, tmp);
4958 + mckr = AT91_PMC_H32MXDIV;
4960 + regmap_update_bits(h32mxclk->regmap, AT91_PMC_MCKR,
4961 + AT91_PMC_H32MXDIV, mckr);
4965 @@ -92,14 +86,18 @@ static const struct clk_ops h32mx_ops = {
4966 .set_rate = clk_sama5d4_h32mx_set_rate,
4969 -void __init of_sama5d4_clk_h32mx_setup(struct device_node *np,
4970 - struct at91_pmc *pmc)
4971 +static void __init of_sama5d4_clk_h32mx_setup(struct device_node *np)
4973 struct clk_sama5d4_h32mx *h32mxclk;
4974 struct clk_init_data init;
4975 const char *parent_name;
4976 + struct regmap *regmap;
4979 + regmap = syscon_node_to_regmap(of_get_parent(np));
4980 + if (IS_ERR(regmap))
4983 h32mxclk = kzalloc(sizeof(*h32mxclk), GFP_KERNEL);
4986 @@ -113,7 +111,7 @@ void __init of_sama5d4_clk_h32mx_setup(struct device_node *np,
4987 init.flags = CLK_SET_RATE_GATE;
4989 h32mxclk->hw.init = &init;
4990 - h32mxclk->pmc = pmc;
4991 + h32mxclk->regmap = regmap;
4993 clk = clk_register(NULL, &h32mxclk->hw);
4995 @@ -123,3 +121,5 @@ void __init of_sama5d4_clk_h32mx_setup(struct device_node *np,
4997 of_clk_add_provider(np, of_clk_src_simple_get, clk);
4999 +CLK_OF_DECLARE(of_sama5d4_clk_h32mx_setup, "atmel,sama5d4-clk-h32mx",
5000 + of_sama5d4_clk_h32mx_setup);
5001 diff --git a/drivers/clk/at91/clk-main.c b/drivers/clk/at91/clk-main.c
5002 index fd7247deabdc..4bfc94d6c26e 100644
5003 --- a/drivers/clk/at91/clk-main.c
5004 +++ b/drivers/clk/at91/clk-main.c
5006 #include <linux/clk/at91_pmc.h>
5007 #include <linux/delay.h>
5008 #include <linux/of.h>
5009 -#include <linux/of_address.h>
5010 -#include <linux/of_irq.h>
5011 -#include <linux/io.h>
5012 -#include <linux/interrupt.h>
5013 -#include <linux/irq.h>
5014 -#include <linux/sched.h>
5015 -#include <linux/wait.h>
5016 +#include <linux/mfd/syscon.h>
5017 +#include <linux/regmap.h>
5023 struct clk_main_osc {
5025 - struct at91_pmc *pmc;
5027 - wait_queue_head_t wait;
5028 + struct regmap *regmap;
5031 #define to_clk_main_osc(hw) container_of(hw, struct clk_main_osc, hw)
5033 struct clk_main_rc_osc {
5035 - struct at91_pmc *pmc;
5037 - wait_queue_head_t wait;
5038 + struct regmap *regmap;
5039 unsigned long frequency;
5040 unsigned long accuracy;
5042 @@ -54,51 +45,47 @@ struct clk_main_rc_osc {
5044 struct clk_rm9200_main {
5046 - struct at91_pmc *pmc;
5047 + struct regmap *regmap;
5050 #define to_clk_rm9200_main(hw) container_of(hw, struct clk_rm9200_main, hw)
5052 struct clk_sam9x5_main {
5054 - struct at91_pmc *pmc;
5056 - wait_queue_head_t wait;
5057 + struct regmap *regmap;
5061 #define to_clk_sam9x5_main(hw) container_of(hw, struct clk_sam9x5_main, hw)
5063 -static irqreturn_t clk_main_osc_irq_handler(int irq, void *dev_id)
5064 +static inline bool clk_main_osc_ready(struct regmap *regmap)
5066 - struct clk_main_osc *osc = dev_id;
5067 + unsigned int status;
5069 - wake_up(&osc->wait);
5070 - disable_irq_nosync(osc->irq);
5071 + regmap_read(regmap, AT91_PMC_SR, &status);
5073 - return IRQ_HANDLED;
5074 + return status & AT91_PMC_MOSCS;
5077 static int clk_main_osc_prepare(struct clk_hw *hw)
5079 struct clk_main_osc *osc = to_clk_main_osc(hw);
5080 - struct at91_pmc *pmc = osc->pmc;
5081 + struct regmap *regmap = osc->regmap;
5084 - tmp = pmc_read(pmc, AT91_CKGR_MOR) & ~MOR_KEY_MASK;
5085 + regmap_read(regmap, AT91_CKGR_MOR, &tmp);
5086 + tmp &= ~MOR_KEY_MASK;
5088 if (tmp & AT91_PMC_OSCBYPASS)
5091 if (!(tmp & AT91_PMC_MOSCEN)) {
5092 tmp |= AT91_PMC_MOSCEN | AT91_PMC_KEY;
5093 - pmc_write(pmc, AT91_CKGR_MOR, tmp);
5094 + regmap_write(regmap, AT91_CKGR_MOR, tmp);
5097 - while (!(pmc_read(pmc, AT91_PMC_SR) & AT91_PMC_MOSCS)) {
5098 - enable_irq(osc->irq);
5099 - wait_event(osc->wait,
5100 - pmc_read(pmc, AT91_PMC_SR) & AT91_PMC_MOSCS);
5102 + while (!clk_main_osc_ready(regmap))
5107 @@ -106,9 +93,10 @@ static int clk_main_osc_prepare(struct clk_hw *hw)
5108 static void clk_main_osc_unprepare(struct clk_hw *hw)
5110 struct clk_main_osc *osc = to_clk_main_osc(hw);
5111 - struct at91_pmc *pmc = osc->pmc;
5112 - u32 tmp = pmc_read(pmc, AT91_CKGR_MOR);
5113 + struct regmap *regmap = osc->regmap;
5116 + regmap_read(regmap, AT91_CKGR_MOR, &tmp);
5117 if (tmp & AT91_PMC_OSCBYPASS)
5120 @@ -116,20 +104,22 @@ static void clk_main_osc_unprepare(struct clk_hw *hw)
5123 tmp &= ~(AT91_PMC_KEY | AT91_PMC_MOSCEN);
5124 - pmc_write(pmc, AT91_CKGR_MOR, tmp | AT91_PMC_KEY);
5125 + regmap_write(regmap, AT91_CKGR_MOR, tmp | AT91_PMC_KEY);
5128 static int clk_main_osc_is_prepared(struct clk_hw *hw)
5130 struct clk_main_osc *osc = to_clk_main_osc(hw);
5131 - struct at91_pmc *pmc = osc->pmc;
5132 - u32 tmp = pmc_read(pmc, AT91_CKGR_MOR);
5133 + struct regmap *regmap = osc->regmap;
5136 + regmap_read(regmap, AT91_CKGR_MOR, &tmp);
5137 if (tmp & AT91_PMC_OSCBYPASS)
5140 - return !!((pmc_read(pmc, AT91_PMC_SR) & AT91_PMC_MOSCS) &&
5141 - (pmc_read(pmc, AT91_CKGR_MOR) & AT91_PMC_MOSCEN));
5142 + regmap_read(regmap, AT91_PMC_SR, &status);
5144 + return (status & AT91_PMC_MOSCS) && (tmp & AT91_PMC_MOSCEN);
5147 static const struct clk_ops main_osc_ops = {
5148 @@ -139,18 +129,16 @@ static const struct clk_ops main_osc_ops = {
5151 static struct clk * __init
5152 -at91_clk_register_main_osc(struct at91_pmc *pmc,
5154 +at91_clk_register_main_osc(struct regmap *regmap,
5156 const char *parent_name,
5160 struct clk_main_osc *osc;
5161 struct clk *clk = NULL;
5162 struct clk_init_data init;
5164 - if (!pmc || !irq || !name || !parent_name)
5165 + if (!name || !parent_name)
5166 return ERR_PTR(-EINVAL);
5168 osc = kzalloc(sizeof(*osc), GFP_KERNEL);
5169 @@ -164,85 +152,70 @@ at91_clk_register_main_osc(struct at91_pmc *pmc,
5170 init.flags = CLK_IGNORE_UNUSED;
5172 osc->hw.init = &init;
5176 - init_waitqueue_head(&osc->wait);
5177 - irq_set_status_flags(osc->irq, IRQ_NOAUTOEN);
5178 - ret = request_irq(osc->irq, clk_main_osc_irq_handler,
5179 - IRQF_TRIGGER_HIGH, name, osc);
5182 - return ERR_PTR(ret);
5184 + osc->regmap = regmap;
5187 - pmc_write(pmc, AT91_CKGR_MOR,
5188 - (pmc_read(pmc, AT91_CKGR_MOR) &
5189 - ~(MOR_KEY_MASK | AT91_PMC_MOSCEN)) |
5190 - AT91_PMC_OSCBYPASS | AT91_PMC_KEY);
5191 + regmap_update_bits(regmap,
5192 + AT91_CKGR_MOR, MOR_KEY_MASK |
5194 + AT91_PMC_OSCBYPASS | AT91_PMC_KEY);
5196 clk = clk_register(NULL, &osc->hw);
5197 - if (IS_ERR(clk)) {
5198 - free_irq(irq, osc);
5206 -void __init of_at91rm9200_clk_main_osc_setup(struct device_node *np,
5207 - struct at91_pmc *pmc)
5208 +static void __init of_at91rm9200_clk_main_osc_setup(struct device_node *np)
5212 const char *name = np->name;
5213 const char *parent_name;
5214 + struct regmap *regmap;
5217 of_property_read_string(np, "clock-output-names", &name);
5218 bypass = of_property_read_bool(np, "atmel,osc-bypass");
5219 parent_name = of_clk_get_parent_name(np, 0);
5221 - irq = irq_of_parse_and_map(np, 0);
5223 + regmap = syscon_node_to_regmap(of_get_parent(np));
5224 + if (IS_ERR(regmap))
5227 - clk = at91_clk_register_main_osc(pmc, irq, name, parent_name, bypass);
5228 + clk = at91_clk_register_main_osc(regmap, name, parent_name, bypass);
5232 of_clk_add_provider(np, of_clk_src_simple_get, clk);
5234 +CLK_OF_DECLARE(at91rm9200_clk_main_osc, "atmel,at91rm9200-clk-main-osc",
5235 + of_at91rm9200_clk_main_osc_setup);
5237 -static irqreturn_t clk_main_rc_osc_irq_handler(int irq, void *dev_id)
5238 +static bool clk_main_rc_osc_ready(struct regmap *regmap)
5240 - struct clk_main_rc_osc *osc = dev_id;
5241 + unsigned int status;
5243 - wake_up(&osc->wait);
5244 - disable_irq_nosync(osc->irq);
5245 + regmap_read(regmap, AT91_PMC_SR, &status);
5247 - return IRQ_HANDLED;
5248 + return status & AT91_PMC_MOSCRCS;
5251 static int clk_main_rc_osc_prepare(struct clk_hw *hw)
5253 struct clk_main_rc_osc *osc = to_clk_main_rc_osc(hw);
5254 - struct at91_pmc *pmc = osc->pmc;
5256 + struct regmap *regmap = osc->regmap;
5259 - tmp = pmc_read(pmc, AT91_CKGR_MOR) & ~MOR_KEY_MASK;
5260 + regmap_read(regmap, AT91_CKGR_MOR, &mor);
5262 - if (!(tmp & AT91_PMC_MOSCRCEN)) {
5263 - tmp |= AT91_PMC_MOSCRCEN | AT91_PMC_KEY;
5264 - pmc_write(pmc, AT91_CKGR_MOR, tmp);
5266 + if (!(mor & AT91_PMC_MOSCRCEN))
5267 + regmap_update_bits(regmap, AT91_CKGR_MOR,
5268 + MOR_KEY_MASK | AT91_PMC_MOSCRCEN,
5269 + AT91_PMC_MOSCRCEN | AT91_PMC_KEY);
5271 - while (!(pmc_read(pmc, AT91_PMC_SR) & AT91_PMC_MOSCRCS)) {
5272 - enable_irq(osc->irq);
5273 - wait_event(osc->wait,
5274 - pmc_read(pmc, AT91_PMC_SR) & AT91_PMC_MOSCRCS);
5276 + while (!clk_main_rc_osc_ready(regmap))
5281 @@ -250,23 +223,28 @@ static int clk_main_rc_osc_prepare(struct clk_hw *hw)
5282 static void clk_main_rc_osc_unprepare(struct clk_hw *hw)
5284 struct clk_main_rc_osc *osc = to_clk_main_rc_osc(hw);
5285 - struct at91_pmc *pmc = osc->pmc;
5286 - u32 tmp = pmc_read(pmc, AT91_CKGR_MOR);
5287 + struct regmap *regmap = osc->regmap;
5290 + regmap_read(regmap, AT91_CKGR_MOR, &mor);
5292 - if (!(tmp & AT91_PMC_MOSCRCEN))
5293 + if (!(mor & AT91_PMC_MOSCRCEN))
5296 - tmp &= ~(MOR_KEY_MASK | AT91_PMC_MOSCRCEN);
5297 - pmc_write(pmc, AT91_CKGR_MOR, tmp | AT91_PMC_KEY);
5298 + regmap_update_bits(regmap, AT91_CKGR_MOR,
5299 + MOR_KEY_MASK | AT91_PMC_MOSCRCEN, AT91_PMC_KEY);
5302 static int clk_main_rc_osc_is_prepared(struct clk_hw *hw)
5304 struct clk_main_rc_osc *osc = to_clk_main_rc_osc(hw);
5305 - struct at91_pmc *pmc = osc->pmc;
5306 + struct regmap *regmap = osc->regmap;
5307 + unsigned int mor, status;
5309 - return !!((pmc_read(pmc, AT91_PMC_SR) & AT91_PMC_MOSCRCS) &&
5310 - (pmc_read(pmc, AT91_CKGR_MOR) & AT91_PMC_MOSCRCEN));
5311 + regmap_read(regmap, AT91_CKGR_MOR, &mor);
5312 + regmap_read(regmap, AT91_PMC_SR, &status);
5314 + return (mor & AT91_PMC_MOSCRCEN) && (status & AT91_PMC_MOSCRCS);
5317 static unsigned long clk_main_rc_osc_recalc_rate(struct clk_hw *hw,
5318 @@ -294,17 +272,15 @@ static const struct clk_ops main_rc_osc_ops = {
5321 static struct clk * __init
5322 -at91_clk_register_main_rc_osc(struct at91_pmc *pmc,
5324 +at91_clk_register_main_rc_osc(struct regmap *regmap,
5326 u32 frequency, u32 accuracy)
5329 struct clk_main_rc_osc *osc;
5330 struct clk *clk = NULL;
5331 struct clk_init_data init;
5333 - if (!pmc || !irq || !name || !frequency)
5334 + if (!name || !frequency)
5335 return ERR_PTR(-EINVAL);
5337 osc = kzalloc(sizeof(*osc), GFP_KERNEL);
5338 @@ -318,63 +294,53 @@ at91_clk_register_main_rc_osc(struct at91_pmc *pmc,
5339 init.flags = CLK_IS_ROOT | CLK_IGNORE_UNUSED;
5341 osc->hw.init = &init;
5344 + osc->regmap = regmap;
5345 osc->frequency = frequency;
5346 osc->accuracy = accuracy;
5348 - init_waitqueue_head(&osc->wait);
5349 - irq_set_status_flags(osc->irq, IRQ_NOAUTOEN);
5350 - ret = request_irq(osc->irq, clk_main_rc_osc_irq_handler,
5351 - IRQF_TRIGGER_HIGH, name, osc);
5353 - return ERR_PTR(ret);
5355 clk = clk_register(NULL, &osc->hw);
5356 - if (IS_ERR(clk)) {
5357 - free_irq(irq, osc);
5365 -void __init of_at91sam9x5_clk_main_rc_osc_setup(struct device_node *np,
5366 - struct at91_pmc *pmc)
5367 +static void __init of_at91sam9x5_clk_main_rc_osc_setup(struct device_node *np)
5373 const char *name = np->name;
5374 + struct regmap *regmap;
5376 of_property_read_string(np, "clock-output-names", &name);
5377 of_property_read_u32(np, "clock-frequency", &frequency);
5378 of_property_read_u32(np, "clock-accuracy", &accuracy);
5380 - irq = irq_of_parse_and_map(np, 0);
5382 + regmap = syscon_node_to_regmap(of_get_parent(np));
5383 + if (IS_ERR(regmap))
5386 - clk = at91_clk_register_main_rc_osc(pmc, irq, name, frequency,
5388 + clk = at91_clk_register_main_rc_osc(regmap, name, frequency, accuracy);
5392 of_clk_add_provider(np, of_clk_src_simple_get, clk);
5394 +CLK_OF_DECLARE(at91sam9x5_clk_main_rc_osc, "atmel,at91sam9x5-clk-main-rc-osc",
5395 + of_at91sam9x5_clk_main_rc_osc_setup);
5398 -static int clk_main_probe_frequency(struct at91_pmc *pmc)
5399 +static int clk_main_probe_frequency(struct regmap *regmap)
5401 unsigned long prep_time, timeout;
5403 + unsigned int mcfr;
5405 timeout = jiffies + usecs_to_jiffies(MAINFRDY_TIMEOUT);
5407 prep_time = jiffies;
5408 - tmp = pmc_read(pmc, AT91_CKGR_MCFR);
5409 - if (tmp & AT91_PMC_MAINRDY)
5410 + regmap_read(regmap, AT91_CKGR_MCFR, &mcfr);
5411 + if (mcfr & AT91_PMC_MAINRDY)
5413 usleep_range(MAINF_LOOP_MIN_WAIT, MAINF_LOOP_MAX_WAIT);
5414 } while (time_before(prep_time, timeout));
5415 @@ -382,34 +348,37 @@ static int clk_main_probe_frequency(struct at91_pmc *pmc)
5419 -static unsigned long clk_main_recalc_rate(struct at91_pmc *pmc,
5420 +static unsigned long clk_main_recalc_rate(struct regmap *regmap,
5421 unsigned long parent_rate)
5424 + unsigned int mcfr;
5429 pr_warn("Main crystal frequency not set, using approximate value\n");
5430 - tmp = pmc_read(pmc, AT91_CKGR_MCFR);
5431 - if (!(tmp & AT91_PMC_MAINRDY))
5432 + regmap_read(regmap, AT91_CKGR_MCFR, &mcfr);
5433 + if (!(mcfr & AT91_PMC_MAINRDY))
5436 - return ((tmp & AT91_PMC_MAINF) * SLOW_CLOCK_FREQ) / MAINF_DIV;
5437 + return ((mcfr & AT91_PMC_MAINF) * SLOW_CLOCK_FREQ) / MAINF_DIV;
5440 static int clk_rm9200_main_prepare(struct clk_hw *hw)
5442 struct clk_rm9200_main *clkmain = to_clk_rm9200_main(hw);
5444 - return clk_main_probe_frequency(clkmain->pmc);
5445 + return clk_main_probe_frequency(clkmain->regmap);
5448 static int clk_rm9200_main_is_prepared(struct clk_hw *hw)
5450 struct clk_rm9200_main *clkmain = to_clk_rm9200_main(hw);
5451 + unsigned int status;
5453 + regmap_read(clkmain->regmap, AT91_CKGR_MCFR, &status);
5455 - return !!(pmc_read(clkmain->pmc, AT91_CKGR_MCFR) & AT91_PMC_MAINRDY);
5456 + return status & AT91_PMC_MAINRDY ? 1 : 0;
5459 static unsigned long clk_rm9200_main_recalc_rate(struct clk_hw *hw,
5460 @@ -417,7 +386,7 @@ static unsigned long clk_rm9200_main_recalc_rate(struct clk_hw *hw,
5462 struct clk_rm9200_main *clkmain = to_clk_rm9200_main(hw);
5464 - return clk_main_recalc_rate(clkmain->pmc, parent_rate);
5465 + return clk_main_recalc_rate(clkmain->regmap, parent_rate);
5468 static const struct clk_ops rm9200_main_ops = {
5469 @@ -427,7 +396,7 @@ static const struct clk_ops rm9200_main_ops = {
5472 static struct clk * __init
5473 -at91_clk_register_rm9200_main(struct at91_pmc *pmc,
5474 +at91_clk_register_rm9200_main(struct regmap *regmap,
5476 const char *parent_name)
5478 @@ -435,7 +404,7 @@ at91_clk_register_rm9200_main(struct at91_pmc *pmc,
5479 struct clk *clk = NULL;
5480 struct clk_init_data init;
5482 - if (!pmc || !name)
5484 return ERR_PTR(-EINVAL);
5487 @@ -452,7 +421,7 @@ at91_clk_register_rm9200_main(struct at91_pmc *pmc,
5490 clkmain->hw.init = &init;
5491 - clkmain->pmc = pmc;
5492 + clkmain->regmap = regmap;
5494 clk = clk_register(NULL, &clkmain->hw);
5496 @@ -461,52 +430,54 @@ at91_clk_register_rm9200_main(struct at91_pmc *pmc,
5500 -void __init of_at91rm9200_clk_main_setup(struct device_node *np,
5501 - struct at91_pmc *pmc)
5502 +static void __init of_at91rm9200_clk_main_setup(struct device_node *np)
5505 const char *parent_name;
5506 const char *name = np->name;
5507 + struct regmap *regmap;
5509 parent_name = of_clk_get_parent_name(np, 0);
5510 of_property_read_string(np, "clock-output-names", &name);
5512 - clk = at91_clk_register_rm9200_main(pmc, name, parent_name);
5513 + regmap = syscon_node_to_regmap(of_get_parent(np));
5514 + if (IS_ERR(regmap))
5517 + clk = at91_clk_register_rm9200_main(regmap, name, parent_name);
5521 of_clk_add_provider(np, of_clk_src_simple_get, clk);
5523 +CLK_OF_DECLARE(at91rm9200_clk_main, "atmel,at91rm9200-clk-main",
5524 + of_at91rm9200_clk_main_setup);
5526 -static irqreturn_t clk_sam9x5_main_irq_handler(int irq, void *dev_id)
5527 +static inline bool clk_sam9x5_main_ready(struct regmap *regmap)
5529 - struct clk_sam9x5_main *clkmain = dev_id;
5530 + unsigned int status;
5532 - wake_up(&clkmain->wait);
5533 - disable_irq_nosync(clkmain->irq);
5534 + regmap_read(regmap, AT91_PMC_SR, &status);
5536 - return IRQ_HANDLED;
5537 + return status & AT91_PMC_MOSCSELS ? 1 : 0;
5540 static int clk_sam9x5_main_prepare(struct clk_hw *hw)
5542 struct clk_sam9x5_main *clkmain = to_clk_sam9x5_main(hw);
5543 - struct at91_pmc *pmc = clkmain->pmc;
5544 + struct regmap *regmap = clkmain->regmap;
5546 - while (!(pmc_read(pmc, AT91_PMC_SR) & AT91_PMC_MOSCSELS)) {
5547 - enable_irq(clkmain->irq);
5548 - wait_event(clkmain->wait,
5549 - pmc_read(pmc, AT91_PMC_SR) & AT91_PMC_MOSCSELS);
5551 + while (!clk_sam9x5_main_ready(regmap))
5554 - return clk_main_probe_frequency(pmc);
5555 + return clk_main_probe_frequency(regmap);
5558 static int clk_sam9x5_main_is_prepared(struct clk_hw *hw)
5560 struct clk_sam9x5_main *clkmain = to_clk_sam9x5_main(hw);
5562 - return !!(pmc_read(clkmain->pmc, AT91_PMC_SR) & AT91_PMC_MOSCSELS);
5563 + return clk_sam9x5_main_ready(clkmain->regmap);
5566 static unsigned long clk_sam9x5_main_recalc_rate(struct clk_hw *hw,
5567 @@ -514,30 +485,28 @@ static unsigned long clk_sam9x5_main_recalc_rate(struct clk_hw *hw,
5569 struct clk_sam9x5_main *clkmain = to_clk_sam9x5_main(hw);
5571 - return clk_main_recalc_rate(clkmain->pmc, parent_rate);
5572 + return clk_main_recalc_rate(clkmain->regmap, parent_rate);
5575 static int clk_sam9x5_main_set_parent(struct clk_hw *hw, u8 index)
5577 struct clk_sam9x5_main *clkmain = to_clk_sam9x5_main(hw);
5578 - struct at91_pmc *pmc = clkmain->pmc;
5580 + struct regmap *regmap = clkmain->regmap;
5586 - tmp = pmc_read(pmc, AT91_CKGR_MOR) & ~MOR_KEY_MASK;
5587 + regmap_read(regmap, AT91_CKGR_MOR, &tmp);
5588 + tmp &= ~MOR_KEY_MASK;
5590 if (index && !(tmp & AT91_PMC_MOSCSEL))
5591 - pmc_write(pmc, AT91_CKGR_MOR, tmp | AT91_PMC_MOSCSEL);
5592 + regmap_write(regmap, AT91_CKGR_MOR, tmp | AT91_PMC_MOSCSEL);
5593 else if (!index && (tmp & AT91_PMC_MOSCSEL))
5594 - pmc_write(pmc, AT91_CKGR_MOR, tmp & ~AT91_PMC_MOSCSEL);
5595 + regmap_write(regmap, AT91_CKGR_MOR, tmp & ~AT91_PMC_MOSCSEL);
5597 - while (!(pmc_read(pmc, AT91_PMC_SR) & AT91_PMC_MOSCSELS)) {
5598 - enable_irq(clkmain->irq);
5599 - wait_event(clkmain->wait,
5600 - pmc_read(pmc, AT91_PMC_SR) & AT91_PMC_MOSCSELS);
5602 + while (!clk_sam9x5_main_ready(regmap))
5607 @@ -545,8 +514,11 @@ static int clk_sam9x5_main_set_parent(struct clk_hw *hw, u8 index)
5608 static u8 clk_sam9x5_main_get_parent(struct clk_hw *hw)
5610 struct clk_sam9x5_main *clkmain = to_clk_sam9x5_main(hw);
5611 + unsigned int status;
5613 + regmap_read(clkmain->regmap, AT91_CKGR_MOR, &status);
5615 - return !!(pmc_read(clkmain->pmc, AT91_CKGR_MOR) & AT91_PMC_MOSCEN);
5616 + return status & AT91_PMC_MOSCEN ? 1 : 0;
5619 static const struct clk_ops sam9x5_main_ops = {
5620 @@ -558,18 +530,17 @@ static const struct clk_ops sam9x5_main_ops = {
5623 static struct clk * __init
5624 -at91_clk_register_sam9x5_main(struct at91_pmc *pmc,
5626 +at91_clk_register_sam9x5_main(struct regmap *regmap,
5628 const char **parent_names,
5632 struct clk_sam9x5_main *clkmain;
5633 struct clk *clk = NULL;
5634 struct clk_init_data init;
5635 + unsigned int status;
5637 - if (!pmc || !irq || !name)
5639 return ERR_PTR(-EINVAL);
5641 if (!parent_names || !num_parents)
5642 @@ -586,51 +557,42 @@ at91_clk_register_sam9x5_main(struct at91_pmc *pmc,
5643 init.flags = CLK_SET_PARENT_GATE;
5645 clkmain->hw.init = &init;
5646 - clkmain->pmc = pmc;
5647 - clkmain->irq = irq;
5648 - clkmain->parent = !!(pmc_read(clkmain->pmc, AT91_CKGR_MOR) &
5650 - init_waitqueue_head(&clkmain->wait);
5651 - irq_set_status_flags(clkmain->irq, IRQ_NOAUTOEN);
5652 - ret = request_irq(clkmain->irq, clk_sam9x5_main_irq_handler,
5653 - IRQF_TRIGGER_HIGH, name, clkmain);
5655 - return ERR_PTR(ret);
5656 + clkmain->regmap = regmap;
5657 + regmap_read(clkmain->regmap, AT91_CKGR_MOR, &status);
5658 + clkmain->parent = status & AT91_PMC_MOSCEN ? 1 : 0;
5660 clk = clk_register(NULL, &clkmain->hw);
5661 - if (IS_ERR(clk)) {
5662 - free_irq(clkmain->irq, clkmain);
5670 -void __init of_at91sam9x5_clk_main_setup(struct device_node *np,
5671 - struct at91_pmc *pmc)
5672 +static void __init of_at91sam9x5_clk_main_setup(struct device_node *np)
5675 const char *parent_names[2];
5678 const char *name = np->name;
5679 + struct regmap *regmap;
5681 num_parents = of_clk_get_parent_count(np);
5682 if (num_parents <= 0 || num_parents > 2)
5685 of_clk_parent_fill(np, parent_names, num_parents);
5686 + regmap = syscon_node_to_regmap(of_get_parent(np));
5687 + if (IS_ERR(regmap))
5690 of_property_read_string(np, "clock-output-names", &name);
5692 - irq = irq_of_parse_and_map(np, 0);
5696 - clk = at91_clk_register_sam9x5_main(pmc, irq, name, parent_names,
5697 + clk = at91_clk_register_sam9x5_main(regmap, name, parent_names,
5702 of_clk_add_provider(np, of_clk_src_simple_get, clk);
5704 +CLK_OF_DECLARE(at91sam9x5_clk_main, "atmel,at91sam9x5-clk-main",
5705 + of_at91sam9x5_clk_main_setup);
5706 diff --git a/drivers/clk/at91/clk-master.c b/drivers/clk/at91/clk-master.c
5707 index 620ea323356b..7d4a1864ea7c 100644
5708 --- a/drivers/clk/at91/clk-master.c
5709 +++ b/drivers/clk/at91/clk-master.c
5711 #include <linux/clkdev.h>
5712 #include <linux/clk/at91_pmc.h>
5713 #include <linux/of.h>
5714 -#include <linux/of_address.h>
5715 -#include <linux/of_irq.h>
5716 -#include <linux/io.h>
5717 -#include <linux/wait.h>
5718 -#include <linux/sched.h>
5719 -#include <linux/interrupt.h>
5720 -#include <linux/irq.h>
5721 +#include <linux/mfd/syscon.h>
5722 +#include <linux/regmap.h>
5726 @@ -44,32 +39,26 @@ struct clk_master_layout {
5730 - struct at91_pmc *pmc;
5732 - wait_queue_head_t wait;
5733 + struct regmap *regmap;
5734 const struct clk_master_layout *layout;
5735 const struct clk_master_characteristics *characteristics;
5738 -static irqreturn_t clk_master_irq_handler(int irq, void *dev_id)
5739 +static inline bool clk_master_ready(struct regmap *regmap)
5741 - struct clk_master *master = (struct clk_master *)dev_id;
5742 + unsigned int status;
5744 - wake_up(&master->wait);
5745 - disable_irq_nosync(master->irq);
5746 + regmap_read(regmap, AT91_PMC_SR, &status);
5748 - return IRQ_HANDLED;
5749 + return status & AT91_PMC_MCKRDY ? 1 : 0;
5752 static int clk_master_prepare(struct clk_hw *hw)
5754 struct clk_master *master = to_clk_master(hw);
5755 - struct at91_pmc *pmc = master->pmc;
5757 - while (!(pmc_read(pmc, AT91_PMC_SR) & AT91_PMC_MCKRDY)) {
5758 - enable_irq(master->irq);
5759 - wait_event(master->wait,
5760 - pmc_read(pmc, AT91_PMC_SR) & AT91_PMC_MCKRDY);
5762 + while (!clk_master_ready(master->regmap))
5767 @@ -78,7 +67,7 @@ static int clk_master_is_prepared(struct clk_hw *hw)
5769 struct clk_master *master = to_clk_master(hw);
5771 - return !!(pmc_read(master->pmc, AT91_PMC_SR) & AT91_PMC_MCKRDY);
5772 + return clk_master_ready(master->regmap);
5775 static unsigned long clk_master_recalc_rate(struct clk_hw *hw,
5776 @@ -88,18 +77,16 @@ static unsigned long clk_master_recalc_rate(struct clk_hw *hw,
5778 unsigned long rate = parent_rate;
5779 struct clk_master *master = to_clk_master(hw);
5780 - struct at91_pmc *pmc = master->pmc;
5781 const struct clk_master_layout *layout = master->layout;
5782 const struct clk_master_characteristics *characteristics =
5783 master->characteristics;
5785 + unsigned int mckr;
5788 - tmp = pmc_read(pmc, AT91_PMC_MCKR) & layout->mask;
5790 + regmap_read(master->regmap, AT91_PMC_MCKR, &mckr);
5791 + mckr &= layout->mask;
5793 - pres = (tmp >> layout->pres_shift) & MASTER_PRES_MASK;
5794 - div = (tmp >> MASTER_DIV_SHIFT) & MASTER_DIV_MASK;
5795 + pres = (mckr >> layout->pres_shift) & MASTER_PRES_MASK;
5796 + div = (mckr >> MASTER_DIV_SHIFT) & MASTER_DIV_MASK;
5798 if (characteristics->have_div3_pres && pres == MASTER_PRES_MAX)
5800 @@ -119,9 +106,11 @@ static unsigned long clk_master_recalc_rate(struct clk_hw *hw,
5801 static u8 clk_master_get_parent(struct clk_hw *hw)
5803 struct clk_master *master = to_clk_master(hw);
5804 - struct at91_pmc *pmc = master->pmc;
5805 + unsigned int mckr;
5807 - return pmc_read(pmc, AT91_PMC_MCKR) & AT91_PMC_CSS;
5808 + regmap_read(master->regmap, AT91_PMC_MCKR, &mckr);
5810 + return mckr & AT91_PMC_CSS;
5813 static const struct clk_ops master_ops = {
5814 @@ -132,18 +121,17 @@ static const struct clk_ops master_ops = {
5817 static struct clk * __init
5818 -at91_clk_register_master(struct at91_pmc *pmc, unsigned int irq,
5819 +at91_clk_register_master(struct regmap *regmap,
5820 const char *name, int num_parents,
5821 const char **parent_names,
5822 const struct clk_master_layout *layout,
5823 const struct clk_master_characteristics *characteristics)
5826 struct clk_master *master;
5827 struct clk *clk = NULL;
5828 struct clk_init_data init;
5830 - if (!pmc || !irq || !name || !num_parents || !parent_names)
5831 + if (!name || !num_parents || !parent_names)
5832 return ERR_PTR(-EINVAL);
5834 master = kzalloc(sizeof(*master), GFP_KERNEL);
5835 @@ -159,20 +147,10 @@ at91_clk_register_master(struct at91_pmc *pmc, unsigned int irq,
5836 master->hw.init = &init;
5837 master->layout = layout;
5838 master->characteristics = characteristics;
5839 - master->pmc = pmc;
5840 - master->irq = irq;
5841 - init_waitqueue_head(&master->wait);
5842 - irq_set_status_flags(master->irq, IRQ_NOAUTOEN);
5843 - ret = request_irq(master->irq, clk_master_irq_handler,
5844 - IRQF_TRIGGER_HIGH, "clk-master", master);
5847 - return ERR_PTR(ret);
5849 + master->regmap = regmap;
5851 clk = clk_register(NULL, &master->hw);
5853 - free_irq(master->irq, master);
5857 @@ -217,15 +195,15 @@ out_free_characteristics:
5861 -of_at91_clk_master_setup(struct device_node *np, struct at91_pmc *pmc,
5862 +of_at91_clk_master_setup(struct device_node *np,
5863 const struct clk_master_layout *layout)
5868 const char *parent_names[MASTER_SOURCE_MAX];
5869 const char *name = np->name;
5870 struct clk_master_characteristics *characteristics;
5871 + struct regmap *regmap;
5873 num_parents = of_clk_get_parent_count(np);
5874 if (num_parents <= 0 || num_parents > MASTER_SOURCE_MAX)
5875 @@ -239,11 +217,11 @@ of_at91_clk_master_setup(struct device_node *np, struct at91_pmc *pmc,
5876 if (!characteristics)
5879 - irq = irq_of_parse_and_map(np, 0);
5881 - goto out_free_characteristics;
5882 + regmap = syscon_node_to_regmap(of_get_parent(np));
5883 + if (IS_ERR(regmap))
5886 - clk = at91_clk_register_master(pmc, irq, name, num_parents,
5887 + clk = at91_clk_register_master(regmap, name, num_parents,
5888 parent_names, layout,
5891 @@ -256,14 +234,16 @@ out_free_characteristics:
5892 kfree(characteristics);
5895 -void __init of_at91rm9200_clk_master_setup(struct device_node *np,
5896 - struct at91_pmc *pmc)
5897 +static void __init of_at91rm9200_clk_master_setup(struct device_node *np)
5899 - of_at91_clk_master_setup(np, pmc, &at91rm9200_master_layout);
5900 + of_at91_clk_master_setup(np, &at91rm9200_master_layout);
5902 +CLK_OF_DECLARE(at91rm9200_clk_master, "atmel,at91rm9200-clk-master",
5903 + of_at91rm9200_clk_master_setup);
5905 -void __init of_at91sam9x5_clk_master_setup(struct device_node *np,
5906 - struct at91_pmc *pmc)
5907 +static void __init of_at91sam9x5_clk_master_setup(struct device_node *np)
5909 - of_at91_clk_master_setup(np, pmc, &at91sam9x5_master_layout);
5910 + of_at91_clk_master_setup(np, &at91sam9x5_master_layout);
5912 +CLK_OF_DECLARE(at91sam9x5_clk_master, "atmel,at91sam9x5-clk-master",
5913 + of_at91sam9x5_clk_master_setup);
5914 diff --git a/drivers/clk/at91/clk-peripheral.c b/drivers/clk/at91/clk-peripheral.c
5915 index 58f3b568e9cb..d69cd2a121b1 100644
5916 --- a/drivers/clk/at91/clk-peripheral.c
5917 +++ b/drivers/clk/at91/clk-peripheral.c
5919 #include <linux/clkdev.h>
5920 #include <linux/clk/at91_pmc.h>
5921 #include <linux/of.h>
5922 -#include <linux/of_address.h>
5923 -#include <linux/io.h>
5924 +#include <linux/mfd/syscon.h>
5925 +#include <linux/regmap.h>
5929 +DEFINE_SPINLOCK(pmc_pcr_lock);
5931 #define PERIPHERAL_MAX 64
5933 #define PERIPHERAL_AT91RM9200 0
5936 struct clk_peripheral {
5938 - struct at91_pmc *pmc;
5939 + struct regmap *regmap;
5943 @@ -41,8 +43,9 @@ struct clk_peripheral {
5945 struct clk_sam9x5_peripheral {
5947 - struct at91_pmc *pmc;
5948 + struct regmap *regmap;
5949 struct clk_range range;
5954 @@ -54,7 +57,6 @@ struct clk_sam9x5_peripheral {
5955 static int clk_peripheral_enable(struct clk_hw *hw)
5957 struct clk_peripheral *periph = to_clk_peripheral(hw);
5958 - struct at91_pmc *pmc = periph->pmc;
5959 int offset = AT91_PMC_PCER;
5960 u32 id = periph->id;
5962 @@ -62,14 +64,14 @@ static int clk_peripheral_enable(struct clk_hw *hw)
5964 if (id > PERIPHERAL_ID_MAX)
5965 offset = AT91_PMC_PCER1;
5966 - pmc_write(pmc, offset, PERIPHERAL_MASK(id));
5967 + regmap_write(periph->regmap, offset, PERIPHERAL_MASK(id));
5972 static void clk_peripheral_disable(struct clk_hw *hw)
5974 struct clk_peripheral *periph = to_clk_peripheral(hw);
5975 - struct at91_pmc *pmc = periph->pmc;
5976 int offset = AT91_PMC_PCDR;
5977 u32 id = periph->id;
5979 @@ -77,21 +79,23 @@ static void clk_peripheral_disable(struct clk_hw *hw)
5981 if (id > PERIPHERAL_ID_MAX)
5982 offset = AT91_PMC_PCDR1;
5983 - pmc_write(pmc, offset, PERIPHERAL_MASK(id));
5984 + regmap_write(periph->regmap, offset, PERIPHERAL_MASK(id));
5987 static int clk_peripheral_is_enabled(struct clk_hw *hw)
5989 struct clk_peripheral *periph = to_clk_peripheral(hw);
5990 - struct at91_pmc *pmc = periph->pmc;
5991 int offset = AT91_PMC_PCSR;
5992 + unsigned int status;
5993 u32 id = periph->id;
5995 if (id < PERIPHERAL_ID_MIN)
5997 if (id > PERIPHERAL_ID_MAX)
5998 offset = AT91_PMC_PCSR1;
5999 - return !!(pmc_read(pmc, offset) & PERIPHERAL_MASK(id));
6000 + regmap_read(periph->regmap, offset, &status);
6002 + return status & PERIPHERAL_MASK(id) ? 1 : 0;
6005 static const struct clk_ops peripheral_ops = {
6006 @@ -101,14 +105,14 @@ static const struct clk_ops peripheral_ops = {
6009 static struct clk * __init
6010 -at91_clk_register_peripheral(struct at91_pmc *pmc, const char *name,
6011 +at91_clk_register_peripheral(struct regmap *regmap, const char *name,
6012 const char *parent_name, u32 id)
6014 struct clk_peripheral *periph;
6015 struct clk *clk = NULL;
6016 struct clk_init_data init;
6018 - if (!pmc || !name || !parent_name || id > PERIPHERAL_ID_MAX)
6019 + if (!name || !parent_name || id > PERIPHERAL_ID_MAX)
6020 return ERR_PTR(-EINVAL);
6022 periph = kzalloc(sizeof(*periph), GFP_KERNEL);
6023 @@ -123,7 +127,7 @@ at91_clk_register_peripheral(struct at91_pmc *pmc, const char *name,
6026 periph->hw.init = &init;
6027 - periph->pmc = pmc;
6028 + periph->regmap = regmap;
6030 clk = clk_register(NULL, &periph->hw);
6032 @@ -160,53 +164,58 @@ static void clk_sam9x5_peripheral_autodiv(struct clk_sam9x5_peripheral *periph)
6033 static int clk_sam9x5_peripheral_enable(struct clk_hw *hw)
6035 struct clk_sam9x5_peripheral *periph = to_clk_sam9x5_peripheral(hw);
6036 - struct at91_pmc *pmc = periph->pmc;
6038 + unsigned long flags;
6040 if (periph->id < PERIPHERAL_ID_MIN)
6044 - pmc_write(pmc, AT91_PMC_PCR, (periph->id & AT91_PMC_PCR_PID_MASK));
6045 - tmp = pmc_read(pmc, AT91_PMC_PCR) & ~AT91_PMC_PCR_DIV_MASK;
6046 - pmc_write(pmc, AT91_PMC_PCR, tmp | AT91_PMC_PCR_DIV(periph->div)
6047 - | AT91_PMC_PCR_CMD
6048 - | AT91_PMC_PCR_EN);
6050 + spin_lock_irqsave(periph->lock, flags);
6051 + regmap_write(periph->regmap, AT91_PMC_PCR,
6052 + (periph->id & AT91_PMC_PCR_PID_MASK));
6053 + regmap_update_bits(periph->regmap, AT91_PMC_PCR,
6054 + AT91_PMC_PCR_DIV_MASK | AT91_PMC_PCR_CMD |
6056 + AT91_PMC_PCR_DIV(periph->div) |
6057 + AT91_PMC_PCR_CMD |
6059 + spin_unlock_irqrestore(periph->lock, flags);
6064 static void clk_sam9x5_peripheral_disable(struct clk_hw *hw)
6066 struct clk_sam9x5_peripheral *periph = to_clk_sam9x5_peripheral(hw);
6067 - struct at91_pmc *pmc = periph->pmc;
6069 + unsigned long flags;
6071 if (periph->id < PERIPHERAL_ID_MIN)
6075 - pmc_write(pmc, AT91_PMC_PCR, (periph->id & AT91_PMC_PCR_PID_MASK));
6076 - tmp = pmc_read(pmc, AT91_PMC_PCR) & ~AT91_PMC_PCR_EN;
6077 - pmc_write(pmc, AT91_PMC_PCR, tmp | AT91_PMC_PCR_CMD);
6079 + spin_lock_irqsave(periph->lock, flags);
6080 + regmap_write(periph->regmap, AT91_PMC_PCR,
6081 + (periph->id & AT91_PMC_PCR_PID_MASK));
6082 + regmap_update_bits(periph->regmap, AT91_PMC_PCR,
6083 + AT91_PMC_PCR_EN | AT91_PMC_PCR_CMD,
6084 + AT91_PMC_PCR_CMD);
6085 + spin_unlock_irqrestore(periph->lock, flags);
6088 static int clk_sam9x5_peripheral_is_enabled(struct clk_hw *hw)
6090 struct clk_sam9x5_peripheral *periph = to_clk_sam9x5_peripheral(hw);
6091 - struct at91_pmc *pmc = periph->pmc;
6093 + unsigned long flags;
6094 + unsigned int status;
6096 if (periph->id < PERIPHERAL_ID_MIN)
6100 - pmc_write(pmc, AT91_PMC_PCR, (periph->id & AT91_PMC_PCR_PID_MASK));
6101 - ret = !!(pmc_read(pmc, AT91_PMC_PCR) & AT91_PMC_PCR_EN);
6103 + spin_lock_irqsave(periph->lock, flags);
6104 + regmap_write(periph->regmap, AT91_PMC_PCR,
6105 + (periph->id & AT91_PMC_PCR_PID_MASK));
6106 + regmap_read(periph->regmap, AT91_PMC_PCR, &status);
6107 + spin_unlock_irqrestore(periph->lock, flags);
6110 + return status & AT91_PMC_PCR_EN ? 1 : 0;
6113 static unsigned long
6114 @@ -214,19 +223,20 @@ clk_sam9x5_peripheral_recalc_rate(struct clk_hw *hw,
6115 unsigned long parent_rate)
6117 struct clk_sam9x5_peripheral *periph = to_clk_sam9x5_peripheral(hw);
6118 - struct at91_pmc *pmc = periph->pmc;
6120 + unsigned long flags;
6121 + unsigned int status;
6123 if (periph->id < PERIPHERAL_ID_MIN)
6127 - pmc_write(pmc, AT91_PMC_PCR, (periph->id & AT91_PMC_PCR_PID_MASK));
6128 - tmp = pmc_read(pmc, AT91_PMC_PCR);
6130 + spin_lock_irqsave(periph->lock, flags);
6131 + regmap_write(periph->regmap, AT91_PMC_PCR,
6132 + (periph->id & AT91_PMC_PCR_PID_MASK));
6133 + regmap_read(periph->regmap, AT91_PMC_PCR, &status);
6134 + spin_unlock_irqrestore(periph->lock, flags);
6136 - if (tmp & AT91_PMC_PCR_EN) {
6137 - periph->div = PERIPHERAL_RSHIFT(tmp);
6138 + if (status & AT91_PMC_PCR_EN) {
6139 + periph->div = PERIPHERAL_RSHIFT(status);
6140 periph->auto_div = false;
6142 clk_sam9x5_peripheral_autodiv(periph);
6143 @@ -318,15 +328,15 @@ static const struct clk_ops sam9x5_peripheral_ops = {
6146 static struct clk * __init
6147 -at91_clk_register_sam9x5_peripheral(struct at91_pmc *pmc, const char *name,
6148 - const char *parent_name, u32 id,
6149 - const struct clk_range *range)
6150 +at91_clk_register_sam9x5_peripheral(struct regmap *regmap, spinlock_t *lock,
6151 + const char *name, const char *parent_name,
6152 + u32 id, const struct clk_range *range)
6154 struct clk_sam9x5_peripheral *periph;
6155 struct clk *clk = NULL;
6156 struct clk_init_data init;
6158 - if (!pmc || !name || !parent_name)
6159 + if (!name || !parent_name)
6160 return ERR_PTR(-EINVAL);
6162 periph = kzalloc(sizeof(*periph), GFP_KERNEL);
6163 @@ -342,7 +352,8 @@ at91_clk_register_sam9x5_peripheral(struct at91_pmc *pmc, const char *name,
6165 periph->hw.init = &init;
6167 - periph->pmc = pmc;
6168 + periph->regmap = regmap;
6169 + periph->lock = lock;
6170 periph->auto_div = true;
6171 periph->range = *range;
6173 @@ -356,7 +367,7 @@ at91_clk_register_sam9x5_peripheral(struct at91_pmc *pmc, const char *name,
6177 -of_at91_clk_periph_setup(struct device_node *np, struct at91_pmc *pmc, u8 type)
6178 +of_at91_clk_periph_setup(struct device_node *np, u8 type)
6182 @@ -364,6 +375,7 @@ of_at91_clk_periph_setup(struct device_node *np, struct at91_pmc *pmc, u8 type)
6183 const char *parent_name;
6185 struct device_node *periphclknp;
6186 + struct regmap *regmap;
6188 parent_name = of_clk_get_parent_name(np, 0);
6190 @@ -373,6 +385,10 @@ of_at91_clk_periph_setup(struct device_node *np, struct at91_pmc *pmc, u8 type)
6191 if (!num || num > PERIPHERAL_MAX)
6194 + regmap = syscon_node_to_regmap(of_get_parent(np));
6195 + if (IS_ERR(regmap))
6198 for_each_child_of_node(np, periphclknp) {
6199 if (of_property_read_u32(periphclknp, "reg", &id))
6201 @@ -384,7 +400,7 @@ of_at91_clk_periph_setup(struct device_node *np, struct at91_pmc *pmc, u8 type)
6202 name = periphclknp->name;
6204 if (type == PERIPHERAL_AT91RM9200) {
6205 - clk = at91_clk_register_peripheral(pmc, name,
6206 + clk = at91_clk_register_peripheral(regmap, name,
6209 struct clk_range range = CLK_RANGE(0, 0);
6210 @@ -393,7 +409,9 @@ of_at91_clk_periph_setup(struct device_node *np, struct at91_pmc *pmc, u8 type)
6211 "atmel,clk-output-range",
6214 - clk = at91_clk_register_sam9x5_peripheral(pmc, name,
6215 + clk = at91_clk_register_sam9x5_peripheral(regmap,
6221 @@ -405,14 +423,16 @@ of_at91_clk_periph_setup(struct device_node *np, struct at91_pmc *pmc, u8 type)
6225 -void __init of_at91rm9200_clk_periph_setup(struct device_node *np,
6226 - struct at91_pmc *pmc)
6227 +static void __init of_at91rm9200_clk_periph_setup(struct device_node *np)
6229 - of_at91_clk_periph_setup(np, pmc, PERIPHERAL_AT91RM9200);
6230 + of_at91_clk_periph_setup(np, PERIPHERAL_AT91RM9200);
6232 +CLK_OF_DECLARE(at91rm9200_clk_periph, "atmel,at91rm9200-clk-peripheral",
6233 + of_at91rm9200_clk_periph_setup);
6235 -void __init of_at91sam9x5_clk_periph_setup(struct device_node *np,
6236 - struct at91_pmc *pmc)
6237 +static void __init of_at91sam9x5_clk_periph_setup(struct device_node *np)
6239 - of_at91_clk_periph_setup(np, pmc, PERIPHERAL_AT91SAM9X5);
6240 + of_at91_clk_periph_setup(np, PERIPHERAL_AT91SAM9X5);
6242 +CLK_OF_DECLARE(at91sam9x5_clk_periph, "atmel,at91sam9x5-clk-peripheral",
6243 + of_at91sam9x5_clk_periph_setup);
6244 diff --git a/drivers/clk/at91/clk-pll.c b/drivers/clk/at91/clk-pll.c
6245 index 18b60f4895a6..fb2e0b56d4b7 100644
6246 --- a/drivers/clk/at91/clk-pll.c
6247 +++ b/drivers/clk/at91/clk-pll.c
6249 #include <linux/clkdev.h>
6250 #include <linux/clk/at91_pmc.h>
6251 #include <linux/of.h>
6252 -#include <linux/of_address.h>
6253 -#include <linux/of_irq.h>
6254 -#include <linux/io.h>
6255 -#include <linux/kernel.h>
6256 -#include <linux/wait.h>
6257 -#include <linux/sched.h>
6258 -#include <linux/interrupt.h>
6259 -#include <linux/irq.h>
6260 +#include <linux/mfd/syscon.h>
6261 +#include <linux/regmap.h>
6265 @@ -58,9 +52,7 @@ struct clk_pll_layout {
6269 - struct at91_pmc *pmc;
6271 - wait_queue_head_t wait;
6272 + struct regmap *regmap;
6276 @@ -69,20 +61,19 @@ struct clk_pll {
6277 const struct clk_pll_characteristics *characteristics;
6280 -static irqreturn_t clk_pll_irq_handler(int irq, void *dev_id)
6281 +static inline bool clk_pll_ready(struct regmap *regmap, int id)
6283 - struct clk_pll *pll = (struct clk_pll *)dev_id;
6284 + unsigned int status;
6286 - wake_up(&pll->wait);
6287 - disable_irq_nosync(pll->irq);
6288 + regmap_read(regmap, AT91_PMC_SR, &status);
6290 - return IRQ_HANDLED;
6291 + return status & PLL_STATUS_MASK(id) ? 1 : 0;
6294 static int clk_pll_prepare(struct clk_hw *hw)
6296 struct clk_pll *pll = to_clk_pll(hw);
6297 - struct at91_pmc *pmc = pll->pmc;
6298 + struct regmap *regmap = pll->regmap;
6299 const struct clk_pll_layout *layout = pll->layout;
6300 const struct clk_pll_characteristics *characteristics =
6301 pll->characteristics;
6302 @@ -90,39 +81,34 @@ static int clk_pll_prepare(struct clk_hw *hw)
6303 u32 mask = PLL_STATUS_MASK(id);
6304 int offset = PLL_REG(id);
6307 + unsigned int pllr;
6308 + unsigned int status;
6312 - pllr = pmc_read(pmc, offset);
6313 + regmap_read(regmap, offset, &pllr);
6314 div = PLL_DIV(pllr);
6315 mul = PLL_MUL(pllr, layout);
6317 - if ((pmc_read(pmc, AT91_PMC_SR) & mask) &&
6318 + regmap_read(regmap, AT91_PMC_SR, &status);
6319 + if ((status & mask) &&
6320 (div == pll->div && mul == pll->mul))
6323 if (characteristics->out)
6324 out = characteristics->out[pll->range];
6325 - if (characteristics->icpll) {
6326 - icpr = pmc_read(pmc, AT91_PMC_PLLICPR) & ~PLL_ICPR_MASK(id);
6327 - icpr |= (characteristics->icpll[pll->range] <<
6328 - PLL_ICPR_SHIFT(id));
6329 - pmc_write(pmc, AT91_PMC_PLLICPR, icpr);
6332 - pllr &= ~layout->pllr_mask;
6333 - pllr |= layout->pllr_mask &
6334 - (pll->div | (PLL_MAX_COUNT << PLL_COUNT_SHIFT) |
6335 - (out << PLL_OUT_SHIFT) |
6336 - ((pll->mul & layout->mul_mask) << layout->mul_shift));
6337 - pmc_write(pmc, offset, pllr);
6339 - while (!(pmc_read(pmc, AT91_PMC_SR) & mask)) {
6340 - enable_irq(pll->irq);
6341 - wait_event(pll->wait,
6342 - pmc_read(pmc, AT91_PMC_SR) & mask);
6344 + if (characteristics->icpll)
6345 + regmap_update_bits(regmap, AT91_PMC_PLLICPR, PLL_ICPR_MASK(id),
6346 + characteristics->icpll[pll->range] << PLL_ICPR_SHIFT(id));
6348 + regmap_update_bits(regmap, offset, layout->pllr_mask,
6349 + pll->div | (PLL_MAX_COUNT << PLL_COUNT_SHIFT) |
6350 + (out << PLL_OUT_SHIFT) |
6351 + ((pll->mul & layout->mul_mask) << layout->mul_shift));
6353 + while (!clk_pll_ready(regmap, pll->id))
6358 @@ -130,32 +116,35 @@ static int clk_pll_prepare(struct clk_hw *hw)
6359 static int clk_pll_is_prepared(struct clk_hw *hw)
6361 struct clk_pll *pll = to_clk_pll(hw);
6362 - struct at91_pmc *pmc = pll->pmc;
6364 - return !!(pmc_read(pmc, AT91_PMC_SR) &
6365 - PLL_STATUS_MASK(pll->id));
6366 + return clk_pll_ready(pll->regmap, pll->id);
6369 static void clk_pll_unprepare(struct clk_hw *hw)
6371 struct clk_pll *pll = to_clk_pll(hw);
6372 - struct at91_pmc *pmc = pll->pmc;
6373 - const struct clk_pll_layout *layout = pll->layout;
6374 - int offset = PLL_REG(pll->id);
6375 - u32 tmp = pmc_read(pmc, offset) & ~(layout->pllr_mask);
6376 + unsigned int mask = pll->layout->pllr_mask;
6378 - pmc_write(pmc, offset, tmp);
6379 + regmap_update_bits(pll->regmap, PLL_REG(pll->id), mask, ~mask);
6382 static unsigned long clk_pll_recalc_rate(struct clk_hw *hw,
6383 unsigned long parent_rate)
6385 struct clk_pll *pll = to_clk_pll(hw);
6386 + unsigned int pllr;
6390 - if (!pll->div || !pll->mul)
6391 + regmap_read(pll->regmap, PLL_REG(pll->id), &pllr);
6393 + div = PLL_DIV(pllr);
6394 + mul = PLL_MUL(pllr, pll->layout);
6399 - return (parent_rate / pll->div) * (pll->mul + 1);
6400 + return (parent_rate / div) * (mul + 1);
6403 static long clk_pll_get_best_div_mul(struct clk_pll *pll, unsigned long rate,
6404 @@ -308,7 +297,7 @@ static const struct clk_ops pll_ops = {
6407 static struct clk * __init
6408 -at91_clk_register_pll(struct at91_pmc *pmc, unsigned int irq, const char *name,
6409 +at91_clk_register_pll(struct regmap *regmap, const char *name,
6410 const char *parent_name, u8 id,
6411 const struct clk_pll_layout *layout,
6412 const struct clk_pll_characteristics *characteristics)
6413 @@ -316,9 +305,8 @@ at91_clk_register_pll(struct at91_pmc *pmc, unsigned int irq, const char *name,
6414 struct clk_pll *pll;
6415 struct clk *clk = NULL;
6416 struct clk_init_data init;
6418 int offset = PLL_REG(id);
6420 + unsigned int pllr;
6422 if (id > PLL_MAX_ID)
6423 return ERR_PTR(-EINVAL);
6424 @@ -337,23 +325,13 @@ at91_clk_register_pll(struct at91_pmc *pmc, unsigned int irq, const char *name,
6425 pll->hw.init = &init;
6426 pll->layout = layout;
6427 pll->characteristics = characteristics;
6430 - tmp = pmc_read(pmc, offset) & layout->pllr_mask;
6431 - pll->div = PLL_DIV(tmp);
6432 - pll->mul = PLL_MUL(tmp, layout);
6433 - init_waitqueue_head(&pll->wait);
6434 - irq_set_status_flags(pll->irq, IRQ_NOAUTOEN);
6435 - ret = request_irq(pll->irq, clk_pll_irq_handler, IRQF_TRIGGER_HIGH,
6436 - id ? "clk-pllb" : "clk-plla", pll);
6439 - return ERR_PTR(ret);
6441 + pll->regmap = regmap;
6442 + regmap_read(regmap, offset, &pllr);
6443 + pll->div = PLL_DIV(pllr);
6444 + pll->mul = PLL_MUL(pllr, layout);
6446 clk = clk_register(NULL, &pll->hw);
6448 - free_irq(pll->irq, pll);
6452 @@ -483,12 +461,12 @@ out_free_characteristics:
6456 -of_at91_clk_pll_setup(struct device_node *np, struct at91_pmc *pmc,
6457 +of_at91_clk_pll_setup(struct device_node *np,
6458 const struct clk_pll_layout *layout)
6463 + struct regmap *regmap;
6464 const char *parent_name;
6465 const char *name = np->name;
6466 struct clk_pll_characteristics *characteristics;
6467 @@ -500,15 +478,15 @@ of_at91_clk_pll_setup(struct device_node *np, struct at91_pmc *pmc,
6469 of_property_read_string(np, "clock-output-names", &name);
6471 - characteristics = of_at91_clk_pll_get_characteristics(np);
6472 - if (!characteristics)
6473 + regmap = syscon_node_to_regmap(of_get_parent(np));
6474 + if (IS_ERR(regmap))
6477 - irq = irq_of_parse_and_map(np, 0);
6479 + characteristics = of_at91_clk_pll_get_characteristics(np);
6480 + if (!characteristics)
6483 - clk = at91_clk_register_pll(pmc, irq, name, parent_name, id, layout,
6484 + clk = at91_clk_register_pll(regmap, name, parent_name, id, layout,
6487 goto out_free_characteristics;
6488 @@ -520,26 +498,30 @@ out_free_characteristics:
6489 kfree(characteristics);
6492 -void __init of_at91rm9200_clk_pll_setup(struct device_node *np,
6493 - struct at91_pmc *pmc)
6494 +static void __init of_at91rm9200_clk_pll_setup(struct device_node *np)
6496 - of_at91_clk_pll_setup(np, pmc, &at91rm9200_pll_layout);
6497 + of_at91_clk_pll_setup(np, &at91rm9200_pll_layout);
6499 +CLK_OF_DECLARE(at91rm9200_clk_pll, "atmel,at91rm9200-clk-pll",
6500 + of_at91rm9200_clk_pll_setup);
6502 -void __init of_at91sam9g45_clk_pll_setup(struct device_node *np,
6503 - struct at91_pmc *pmc)
6504 +static void __init of_at91sam9g45_clk_pll_setup(struct device_node *np)
6506 - of_at91_clk_pll_setup(np, pmc, &at91sam9g45_pll_layout);
6507 + of_at91_clk_pll_setup(np, &at91sam9g45_pll_layout);
6509 +CLK_OF_DECLARE(at91sam9g45_clk_pll, "atmel,at91sam9g45-clk-pll",
6510 + of_at91sam9g45_clk_pll_setup);
6512 -void __init of_at91sam9g20_clk_pllb_setup(struct device_node *np,
6513 - struct at91_pmc *pmc)
6514 +static void __init of_at91sam9g20_clk_pllb_setup(struct device_node *np)
6516 - of_at91_clk_pll_setup(np, pmc, &at91sam9g20_pllb_layout);
6517 + of_at91_clk_pll_setup(np, &at91sam9g20_pllb_layout);
6519 +CLK_OF_DECLARE(at91sam9g20_clk_pllb, "atmel,at91sam9g20-clk-pllb",
6520 + of_at91sam9g20_clk_pllb_setup);
6522 -void __init of_sama5d3_clk_pll_setup(struct device_node *np,
6523 - struct at91_pmc *pmc)
6524 +static void __init of_sama5d3_clk_pll_setup(struct device_node *np)
6526 - of_at91_clk_pll_setup(np, pmc, &sama5d3_pll_layout);
6527 + of_at91_clk_pll_setup(np, &sama5d3_pll_layout);
6529 +CLK_OF_DECLARE(sama5d3_clk_pll, "atmel,sama5d3-clk-pll",
6530 + of_sama5d3_clk_pll_setup);
6531 diff --git a/drivers/clk/at91/clk-plldiv.c b/drivers/clk/at91/clk-plldiv.c
6532 index ea226562bb40..2bed26481027 100644
6533 --- a/drivers/clk/at91/clk-plldiv.c
6534 +++ b/drivers/clk/at91/clk-plldiv.c
6536 #include <linux/clkdev.h>
6537 #include <linux/clk/at91_pmc.h>
6538 #include <linux/of.h>
6539 -#include <linux/of_address.h>
6540 -#include <linux/io.h>
6541 +#include <linux/mfd/syscon.h>
6542 +#include <linux/regmap.h>
6550 - struct at91_pmc *pmc;
6551 + struct regmap *regmap;
6554 static unsigned long clk_plldiv_recalc_rate(struct clk_hw *hw,
6555 unsigned long parent_rate)
6557 struct clk_plldiv *plldiv = to_clk_plldiv(hw);
6558 - struct at91_pmc *pmc = plldiv->pmc;
6559 + unsigned int mckr;
6561 - if (pmc_read(pmc, AT91_PMC_MCKR) & AT91_PMC_PLLADIV2)
6562 + regmap_read(plldiv->regmap, AT91_PMC_MCKR, &mckr);
6564 + if (mckr & AT91_PMC_PLLADIV2)
6565 return parent_rate / 2;
6568 @@ -57,18 +59,12 @@ static int clk_plldiv_set_rate(struct clk_hw *hw, unsigned long rate,
6569 unsigned long parent_rate)
6571 struct clk_plldiv *plldiv = to_clk_plldiv(hw);
6572 - struct at91_pmc *pmc = plldiv->pmc;
6575 - if (parent_rate != rate && (parent_rate / 2) != rate)
6576 + if ((parent_rate != rate) && (parent_rate / 2 != rate))
6580 - tmp = pmc_read(pmc, AT91_PMC_MCKR) & ~AT91_PMC_PLLADIV2;
6581 - if ((parent_rate / 2) == rate)
6582 - tmp |= AT91_PMC_PLLADIV2;
6583 - pmc_write(pmc, AT91_PMC_MCKR, tmp);
6585 + regmap_update_bits(plldiv->regmap, AT91_PMC_MCKR, AT91_PMC_PLLADIV2,
6586 + parent_rate != rate ? AT91_PMC_PLLADIV2 : 0);
6590 @@ -80,7 +76,7 @@ static const struct clk_ops plldiv_ops = {
6593 static struct clk * __init
6594 -at91_clk_register_plldiv(struct at91_pmc *pmc, const char *name,
6595 +at91_clk_register_plldiv(struct regmap *regmap, const char *name,
6596 const char *parent_name)
6598 struct clk_plldiv *plldiv;
6599 @@ -98,7 +94,7 @@ at91_clk_register_plldiv(struct at91_pmc *pmc, const char *name,
6600 init.flags = CLK_SET_RATE_GATE;
6602 plldiv->hw.init = &init;
6603 - plldiv->pmc = pmc;
6604 + plldiv->regmap = regmap;
6606 clk = clk_register(NULL, &plldiv->hw);
6608 @@ -109,27 +105,27 @@ at91_clk_register_plldiv(struct at91_pmc *pmc, const char *name,
6612 -of_at91_clk_plldiv_setup(struct device_node *np, struct at91_pmc *pmc)
6613 +of_at91sam9x5_clk_plldiv_setup(struct device_node *np)
6616 const char *parent_name;
6617 const char *name = np->name;
6618 + struct regmap *regmap;
6620 parent_name = of_clk_get_parent_name(np, 0);
6622 of_property_read_string(np, "clock-output-names", &name);
6624 - clk = at91_clk_register_plldiv(pmc, name, parent_name);
6625 + regmap = syscon_node_to_regmap(of_get_parent(np));
6626 + if (IS_ERR(regmap))
6629 + clk = at91_clk_register_plldiv(regmap, name, parent_name);
6633 of_clk_add_provider(np, of_clk_src_simple_get, clk);
6637 -void __init of_at91sam9x5_clk_plldiv_setup(struct device_node *np,
6638 - struct at91_pmc *pmc)
6640 - of_at91_clk_plldiv_setup(np, pmc);
6642 +CLK_OF_DECLARE(at91sam9x5_clk_plldiv, "atmel,at91sam9x5-clk-plldiv",
6643 + of_at91sam9x5_clk_plldiv_setup);
6644 diff --git a/drivers/clk/at91/clk-programmable.c b/drivers/clk/at91/clk-programmable.c
6645 index 14b270b85fec..bc0be629671b 100644
6646 --- a/drivers/clk/at91/clk-programmable.c
6647 +++ b/drivers/clk/at91/clk-programmable.c
6649 #include <linux/clkdev.h>
6650 #include <linux/clk/at91_pmc.h>
6651 #include <linux/of.h>
6652 -#include <linux/of_address.h>
6653 -#include <linux/io.h>
6654 -#include <linux/wait.h>
6655 -#include <linux/sched.h>
6656 +#include <linux/mfd/syscon.h>
6657 +#include <linux/regmap.h>
6663 #define PROG_STATUS_MASK(id) (1 << ((id) + 8))
6664 #define PROG_PRES_MASK 0x7
6665 +#define PROG_PRES(layout, pckr) ((pckr >> layout->pres_shift) & PROG_PRES_MASK)
6666 #define PROG_MAX_RM9200_CSS 3
6668 struct clk_programmable_layout {
6669 @@ -34,7 +33,7 @@ struct clk_programmable_layout {
6671 struct clk_programmable {
6673 - struct at91_pmc *pmc;
6674 + struct regmap *regmap;
6676 const struct clk_programmable_layout *layout;
6678 @@ -44,14 +43,12 @@ struct clk_programmable {
6679 static unsigned long clk_programmable_recalc_rate(struct clk_hw *hw,
6680 unsigned long parent_rate)
6683 struct clk_programmable *prog = to_clk_programmable(hw);
6684 - struct at91_pmc *pmc = prog->pmc;
6685 - const struct clk_programmable_layout *layout = prog->layout;
6686 + unsigned int pckr;
6688 + regmap_read(prog->regmap, AT91_PMC_PCKR(prog->id), &pckr);
6690 - pres = (pmc_read(pmc, AT91_PMC_PCKR(prog->id)) >> layout->pres_shift) &
6692 - return parent_rate >> pres;
6693 + return parent_rate >> PROG_PRES(prog->layout, pckr);
6696 static int clk_programmable_determine_rate(struct clk_hw *hw,
6697 @@ -101,36 +98,36 @@ static int clk_programmable_set_parent(struct clk_hw *hw, u8 index)
6699 struct clk_programmable *prog = to_clk_programmable(hw);
6700 const struct clk_programmable_layout *layout = prog->layout;
6701 - struct at91_pmc *pmc = prog->pmc;
6702 - u32 tmp = pmc_read(pmc, AT91_PMC_PCKR(prog->id)) & ~layout->css_mask;
6703 + unsigned int mask = layout->css_mask;
6704 + unsigned int pckr = 0;
6706 if (layout->have_slck_mck)
6707 - tmp &= AT91_PMC_CSSMCK_MCK;
6708 + mask |= AT91_PMC_CSSMCK_MCK;
6710 if (index > layout->css_mask) {
6711 - if (index > PROG_MAX_RM9200_CSS && layout->have_slck_mck) {
6712 - tmp |= AT91_PMC_CSSMCK_MCK;
6715 + if (index > PROG_MAX_RM9200_CSS && !layout->have_slck_mck)
6719 + pckr |= AT91_PMC_CSSMCK_MCK;
6722 - pmc_write(pmc, AT91_PMC_PCKR(prog->id), tmp | index);
6723 + regmap_update_bits(prog->regmap, AT91_PMC_PCKR(prog->id), mask, pckr);
6728 static u8 clk_programmable_get_parent(struct clk_hw *hw)
6732 struct clk_programmable *prog = to_clk_programmable(hw);
6733 - struct at91_pmc *pmc = prog->pmc;
6734 const struct clk_programmable_layout *layout = prog->layout;
6735 + unsigned int pckr;
6738 + regmap_read(prog->regmap, AT91_PMC_PCKR(prog->id), &pckr);
6740 + ret = pckr & layout->css_mask;
6742 - tmp = pmc_read(pmc, AT91_PMC_PCKR(prog->id));
6743 - ret = tmp & layout->css_mask;
6744 - if (layout->have_slck_mck && (tmp & AT91_PMC_CSSMCK_MCK) && !ret)
6745 + if (layout->have_slck_mck && (pckr & AT91_PMC_CSSMCK_MCK) && !ret)
6746 ret = PROG_MAX_RM9200_CSS + 1;
6749 @@ -140,26 +137,27 @@ static int clk_programmable_set_rate(struct clk_hw *hw, unsigned long rate,
6750 unsigned long parent_rate)
6752 struct clk_programmable *prog = to_clk_programmable(hw);
6753 - struct at91_pmc *pmc = prog->pmc;
6754 const struct clk_programmable_layout *layout = prog->layout;
6755 unsigned long div = parent_rate / rate;
6756 + unsigned int pckr;
6758 - u32 tmp = pmc_read(pmc, AT91_PMC_PCKR(prog->id)) &
6759 - ~(PROG_PRES_MASK << layout->pres_shift);
6761 + regmap_read(prog->regmap, AT91_PMC_PCKR(prog->id), &pckr);
6766 shift = fls(div) - 1;
6768 - if (div != (1<<shift))
6769 + if (div != (1 << shift))
6772 if (shift >= PROG_PRES_MASK)
6775 - pmc_write(pmc, AT91_PMC_PCKR(prog->id),
6776 - tmp | (shift << layout->pres_shift));
6777 + regmap_update_bits(prog->regmap, AT91_PMC_PCKR(prog->id),
6778 + PROG_PRES_MASK << layout->pres_shift,
6779 + shift << layout->pres_shift);
6783 @@ -173,7 +171,7 @@ static const struct clk_ops programmable_ops = {
6786 static struct clk * __init
6787 -at91_clk_register_programmable(struct at91_pmc *pmc,
6788 +at91_clk_register_programmable(struct regmap *regmap,
6789 const char *name, const char **parent_names,
6790 u8 num_parents, u8 id,
6791 const struct clk_programmable_layout *layout)
6792 @@ -198,7 +196,7 @@ at91_clk_register_programmable(struct at91_pmc *pmc,
6794 prog->layout = layout;
6795 prog->hw.init = &init;
6797 + prog->regmap = regmap;
6799 clk = clk_register(NULL, &prog->hw);
6801 @@ -226,7 +224,7 @@ static const struct clk_programmable_layout at91sam9x5_programmable_layout = {
6805 -of_at91_clk_prog_setup(struct device_node *np, struct at91_pmc *pmc,
6806 +of_at91_clk_prog_setup(struct device_node *np,
6807 const struct clk_programmable_layout *layout)
6810 @@ -236,6 +234,7 @@ of_at91_clk_prog_setup(struct device_node *np, struct at91_pmc *pmc,
6811 const char *parent_names[PROG_SOURCE_MAX];
6813 struct device_node *progclknp;
6814 + struct regmap *regmap;
6816 num_parents = of_clk_get_parent_count(np);
6817 if (num_parents <= 0 || num_parents > PROG_SOURCE_MAX)
6818 @@ -247,6 +246,10 @@ of_at91_clk_prog_setup(struct device_node *np, struct at91_pmc *pmc,
6819 if (!num || num > (PROG_ID_MAX + 1))
6822 + regmap = syscon_node_to_regmap(of_get_parent(np));
6823 + if (IS_ERR(regmap))
6826 for_each_child_of_node(np, progclknp) {
6827 if (of_property_read_u32(progclknp, "reg", &id))
6829 @@ -254,7 +257,7 @@ of_at91_clk_prog_setup(struct device_node *np, struct at91_pmc *pmc,
6830 if (of_property_read_string(np, "clock-output-names", &name))
6831 name = progclknp->name;
6833 - clk = at91_clk_register_programmable(pmc, name,
6834 + clk = at91_clk_register_programmable(regmap, name,
6835 parent_names, num_parents,
6838 @@ -265,20 +268,23 @@ of_at91_clk_prog_setup(struct device_node *np, struct at91_pmc *pmc,
6842 -void __init of_at91rm9200_clk_prog_setup(struct device_node *np,
6843 - struct at91_pmc *pmc)
6844 +static void __init of_at91rm9200_clk_prog_setup(struct device_node *np)
6846 - of_at91_clk_prog_setup(np, pmc, &at91rm9200_programmable_layout);
6847 + of_at91_clk_prog_setup(np, &at91rm9200_programmable_layout);
6849 +CLK_OF_DECLARE(at91rm9200_clk_prog, "atmel,at91rm9200-clk-programmable",
6850 + of_at91rm9200_clk_prog_setup);
6852 -void __init of_at91sam9g45_clk_prog_setup(struct device_node *np,
6853 - struct at91_pmc *pmc)
6854 +static void __init of_at91sam9g45_clk_prog_setup(struct device_node *np)
6856 - of_at91_clk_prog_setup(np, pmc, &at91sam9g45_programmable_layout);
6857 + of_at91_clk_prog_setup(np, &at91sam9g45_programmable_layout);
6859 +CLK_OF_DECLARE(at91sam9g45_clk_prog, "atmel,at91sam9g45-clk-programmable",
6860 + of_at91sam9g45_clk_prog_setup);
6862 -void __init of_at91sam9x5_clk_prog_setup(struct device_node *np,
6863 - struct at91_pmc *pmc)
6864 +static void __init of_at91sam9x5_clk_prog_setup(struct device_node *np)
6866 - of_at91_clk_prog_setup(np, pmc, &at91sam9x5_programmable_layout);
6867 + of_at91_clk_prog_setup(np, &at91sam9x5_programmable_layout);
6869 +CLK_OF_DECLARE(at91sam9x5_clk_prog, "atmel,at91sam9x5-clk-programmable",
6870 + of_at91sam9x5_clk_prog_setup);
6871 diff --git a/drivers/clk/at91/clk-slow.c b/drivers/clk/at91/clk-slow.c
6872 index d0d5076a9b94..221c09684ba3 100644
6873 --- a/drivers/clk/at91/clk-slow.c
6874 +++ b/drivers/clk/at91/clk-slow.c
6876 #include <linux/clk.h>
6877 #include <linux/clk-provider.h>
6878 #include <linux/clkdev.h>
6879 -#include <linux/slab.h>
6880 #include <linux/clk/at91_pmc.h>
6881 #include <linux/delay.h>
6882 #include <linux/of.h>
6883 -#include <linux/of_address.h>
6884 -#include <linux/of_irq.h>
6885 -#include <linux/io.h>
6886 -#include <linux/interrupt.h>
6887 -#include <linux/irq.h>
6888 -#include <linux/sched.h>
6889 -#include <linux/wait.h>
6890 +#include <linux/mfd/syscon.h>
6891 +#include <linux/regmap.h>
6895 @@ -59,7 +53,7 @@ struct clk_slow_rc_osc {
6897 struct clk_sam9260_slow {
6899 - struct at91_pmc *pmc;
6900 + struct regmap *regmap;
6903 #define to_clk_sam9260_slow(hw) container_of(hw, struct clk_sam9260_slow, hw)
6904 @@ -393,8 +387,11 @@ void __init of_at91sam9x5_clk_slow_setup(struct device_node *np,
6905 static u8 clk_sam9260_slow_get_parent(struct clk_hw *hw)
6907 struct clk_sam9260_slow *slowck = to_clk_sam9260_slow(hw);
6908 + unsigned int status;
6910 - return !!(pmc_read(slowck->pmc, AT91_PMC_SR) & AT91_PMC_OSCSEL);
6911 + regmap_read(slowck->regmap, AT91_PMC_SR, &status);
6913 + return status & AT91_PMC_OSCSEL ? 1 : 0;
6916 static const struct clk_ops sam9260_slow_ops = {
6917 @@ -402,7 +399,7 @@ static const struct clk_ops sam9260_slow_ops = {
6920 static struct clk * __init
6921 -at91_clk_register_sam9260_slow(struct at91_pmc *pmc,
6922 +at91_clk_register_sam9260_slow(struct regmap *regmap,
6924 const char **parent_names,
6926 @@ -411,7 +408,7 @@ at91_clk_register_sam9260_slow(struct at91_pmc *pmc,
6927 struct clk *clk = NULL;
6928 struct clk_init_data init;
6930 - if (!pmc || !name)
6932 return ERR_PTR(-EINVAL);
6934 if (!parent_names || !num_parents)
6935 @@ -428,7 +425,7 @@ at91_clk_register_sam9260_slow(struct at91_pmc *pmc,
6938 slowck->hw.init = &init;
6939 - slowck->pmc = pmc;
6940 + slowck->regmap = regmap;
6942 clk = clk_register(NULL, &slowck->hw);
6944 @@ -439,29 +436,34 @@ at91_clk_register_sam9260_slow(struct at91_pmc *pmc,
6948 -void __init of_at91sam9260_clk_slow_setup(struct device_node *np,
6949 - struct at91_pmc *pmc)
6950 +static void __init of_at91sam9260_clk_slow_setup(struct device_node *np)
6953 const char *parent_names[2];
6955 const char *name = np->name;
6956 + struct regmap *regmap;
6958 num_parents = of_clk_get_parent_count(np);
6959 if (num_parents != 2)
6962 of_clk_parent_fill(np, parent_names, num_parents);
6963 + regmap = syscon_node_to_regmap(of_get_parent(np));
6964 + if (IS_ERR(regmap))
6967 of_property_read_string(np, "clock-output-names", &name);
6969 - clk = at91_clk_register_sam9260_slow(pmc, name, parent_names,
6970 + clk = at91_clk_register_sam9260_slow(regmap, name, parent_names,
6975 of_clk_add_provider(np, of_clk_src_simple_get, clk);
6977 +CLK_OF_DECLARE(at91sam9260_clk_slow, "atmel,at91sam9260-clk-slow",
6978 + of_at91sam9260_clk_slow_setup);
6981 * FIXME: All slow clk users are not properly claiming it (get + prepare +
6982 diff --git a/drivers/clk/at91/clk-smd.c b/drivers/clk/at91/clk-smd.c
6983 index a7f8501cfa05..e6948a52005a 100644
6984 --- a/drivers/clk/at91/clk-smd.c
6985 +++ b/drivers/clk/at91/clk-smd.c
6987 #include <linux/clkdev.h>
6988 #include <linux/clk/at91_pmc.h>
6989 #include <linux/of.h>
6990 -#include <linux/of_address.h>
6991 -#include <linux/io.h>
6992 +#include <linux/mfd/syscon.h>
6993 +#include <linux/regmap.h>
6999 struct at91sam9x5_clk_smd {
7001 - struct at91_pmc *pmc;
7002 + struct regmap *regmap;
7005 #define to_at91sam9x5_clk_smd(hw) \
7006 @@ -33,13 +33,13 @@ struct at91sam9x5_clk_smd {
7007 static unsigned long at91sam9x5_clk_smd_recalc_rate(struct clk_hw *hw,
7008 unsigned long parent_rate)
7012 struct at91sam9x5_clk_smd *smd = to_at91sam9x5_clk_smd(hw);
7013 - struct at91_pmc *pmc = smd->pmc;
7014 + unsigned int smdr;
7017 + regmap_read(smd->regmap, AT91_PMC_SMD, &smdr);
7018 + smddiv = (smdr & AT91_PMC_SMD_DIV) >> SMD_DIV_SHIFT;
7020 - tmp = pmc_read(pmc, AT91_PMC_SMD);
7021 - smddiv = (tmp & AT91_PMC_SMD_DIV) >> SMD_DIV_SHIFT;
7022 return parent_rate / (smddiv + 1);
7025 @@ -67,40 +67,38 @@ static long at91sam9x5_clk_smd_round_rate(struct clk_hw *hw, unsigned long rate,
7027 static int at91sam9x5_clk_smd_set_parent(struct clk_hw *hw, u8 index)
7030 struct at91sam9x5_clk_smd *smd = to_at91sam9x5_clk_smd(hw);
7031 - struct at91_pmc *pmc = smd->pmc;
7035 - tmp = pmc_read(pmc, AT91_PMC_SMD) & ~AT91_PMC_SMDS;
7037 - tmp |= AT91_PMC_SMDS;
7038 - pmc_write(pmc, AT91_PMC_SMD, tmp);
7040 + regmap_update_bits(smd->regmap, AT91_PMC_SMD, AT91_PMC_SMDS,
7041 + index ? AT91_PMC_SMDS : 0);
7046 static u8 at91sam9x5_clk_smd_get_parent(struct clk_hw *hw)
7048 struct at91sam9x5_clk_smd *smd = to_at91sam9x5_clk_smd(hw);
7049 - struct at91_pmc *pmc = smd->pmc;
7050 + unsigned int smdr;
7052 - return pmc_read(pmc, AT91_PMC_SMD) & AT91_PMC_SMDS;
7053 + regmap_read(smd->regmap, AT91_PMC_SMD, &smdr);
7055 + return smdr & AT91_PMC_SMDS;
7058 static int at91sam9x5_clk_smd_set_rate(struct clk_hw *hw, unsigned long rate,
7059 unsigned long parent_rate)
7062 struct at91sam9x5_clk_smd *smd = to_at91sam9x5_clk_smd(hw);
7063 - struct at91_pmc *pmc = smd->pmc;
7064 unsigned long div = parent_rate / rate;
7066 if (parent_rate % rate || div < 1 || div > (SMD_MAX_DIV + 1))
7068 - tmp = pmc_read(pmc, AT91_PMC_SMD) & ~AT91_PMC_SMD_DIV;
7069 - tmp |= (div - 1) << SMD_DIV_SHIFT;
7070 - pmc_write(pmc, AT91_PMC_SMD, tmp);
7072 + regmap_update_bits(smd->regmap, AT91_PMC_SMD, AT91_PMC_SMD_DIV,
7073 + (div - 1) << SMD_DIV_SHIFT);
7077 @@ -114,7 +112,7 @@ static const struct clk_ops at91sam9x5_smd_ops = {
7080 static struct clk * __init
7081 -at91sam9x5_clk_register_smd(struct at91_pmc *pmc, const char *name,
7082 +at91sam9x5_clk_register_smd(struct regmap *regmap, const char *name,
7083 const char **parent_names, u8 num_parents)
7085 struct at91sam9x5_clk_smd *smd;
7086 @@ -132,7 +130,7 @@ at91sam9x5_clk_register_smd(struct at91_pmc *pmc, const char *name,
7087 init.flags = CLK_SET_RATE_GATE | CLK_SET_PARENT_GATE;
7089 smd->hw.init = &init;
7091 + smd->regmap = regmap;
7093 clk = clk_register(NULL, &smd->hw);
7095 @@ -141,13 +139,13 @@ at91sam9x5_clk_register_smd(struct at91_pmc *pmc, const char *name,
7099 -void __init of_at91sam9x5_clk_smd_setup(struct device_node *np,
7100 - struct at91_pmc *pmc)
7101 +static void __init of_at91sam9x5_clk_smd_setup(struct device_node *np)
7105 const char *parent_names[SMD_SOURCE_MAX];
7106 const char *name = np->name;
7107 + struct regmap *regmap;
7109 num_parents = of_clk_get_parent_count(np);
7110 if (num_parents <= 0 || num_parents > SMD_SOURCE_MAX)
7111 @@ -157,10 +155,16 @@ void __init of_at91sam9x5_clk_smd_setup(struct device_node *np,
7113 of_property_read_string(np, "clock-output-names", &name);
7115 - clk = at91sam9x5_clk_register_smd(pmc, name, parent_names,
7116 + regmap = syscon_node_to_regmap(of_get_parent(np));
7117 + if (IS_ERR(regmap))
7120 + clk = at91sam9x5_clk_register_smd(regmap, name, parent_names,
7125 of_clk_add_provider(np, of_clk_src_simple_get, clk);
7127 +CLK_OF_DECLARE(at91sam9x5_clk_smd, "atmel,at91sam9x5-clk-smd",
7128 + of_at91sam9x5_clk_smd_setup);
7129 diff --git a/drivers/clk/at91/clk-system.c b/drivers/clk/at91/clk-system.c
7130 index 3f5314344286..8f35d8172909 100644
7131 --- a/drivers/clk/at91/clk-system.c
7132 +++ b/drivers/clk/at91/clk-system.c
7134 #include <linux/clkdev.h>
7135 #include <linux/clk/at91_pmc.h>
7136 #include <linux/of.h>
7137 -#include <linux/of_address.h>
7138 -#include <linux/io.h>
7139 -#include <linux/irq.h>
7140 -#include <linux/of_irq.h>
7141 -#include <linux/interrupt.h>
7142 -#include <linux/wait.h>
7143 -#include <linux/sched.h>
7144 +#include <linux/mfd/syscon.h>
7145 +#include <linux/regmap.h>
7150 #define to_clk_system(hw) container_of(hw, struct clk_system, hw)
7153 - struct at91_pmc *pmc;
7155 - wait_queue_head_t wait;
7156 + struct regmap *regmap;
7160 @@ -39,58 +32,54 @@ static inline int is_pck(int id)
7162 return (id >= 8) && (id <= 15);
7164 -static irqreturn_t clk_system_irq_handler(int irq, void *dev_id)
7166 +static inline bool clk_system_ready(struct regmap *regmap, int id)
7168 - struct clk_system *sys = (struct clk_system *)dev_id;
7169 + unsigned int status;
7171 - wake_up(&sys->wait);
7172 - disable_irq_nosync(sys->irq);
7173 + regmap_read(regmap, AT91_PMC_SR, &status);
7175 - return IRQ_HANDLED;
7176 + return status & (1 << id) ? 1 : 0;
7179 static int clk_system_prepare(struct clk_hw *hw)
7181 struct clk_system *sys = to_clk_system(hw);
7182 - struct at91_pmc *pmc = sys->pmc;
7183 - u32 mask = 1 << sys->id;
7185 - pmc_write(pmc, AT91_PMC_SCER, mask);
7186 + regmap_write(sys->regmap, AT91_PMC_SCER, 1 << sys->id);
7188 if (!is_pck(sys->id))
7191 - while (!(pmc_read(pmc, AT91_PMC_SR) & mask)) {
7193 - enable_irq(sys->irq);
7194 - wait_event(sys->wait,
7195 - pmc_read(pmc, AT91_PMC_SR) & mask);
7199 + while (!clk_system_ready(sys->regmap, sys->id))
7205 static void clk_system_unprepare(struct clk_hw *hw)
7207 struct clk_system *sys = to_clk_system(hw);
7208 - struct at91_pmc *pmc = sys->pmc;
7210 - pmc_write(pmc, AT91_PMC_SCDR, 1 << sys->id);
7211 + regmap_write(sys->regmap, AT91_PMC_SCDR, 1 << sys->id);
7214 static int clk_system_is_prepared(struct clk_hw *hw)
7216 struct clk_system *sys = to_clk_system(hw);
7217 - struct at91_pmc *pmc = sys->pmc;
7218 + unsigned int status;
7220 + regmap_read(sys->regmap, AT91_PMC_SCSR, &status);
7222 - if (!(pmc_read(pmc, AT91_PMC_SCSR) & (1 << sys->id)))
7223 + if (!(status & (1 << sys->id)))
7226 if (!is_pck(sys->id))
7229 - return !!(pmc_read(pmc, AT91_PMC_SR) & (1 << sys->id));
7230 + regmap_read(sys->regmap, AT91_PMC_SR, &status);
7232 + return status & (1 << sys->id) ? 1 : 0;
7235 static const struct clk_ops system_ops = {
7236 @@ -100,13 +89,12 @@ static const struct clk_ops system_ops = {
7239 static struct clk * __init
7240 -at91_clk_register_system(struct at91_pmc *pmc, const char *name,
7241 - const char *parent_name, u8 id, int irq)
7242 +at91_clk_register_system(struct regmap *regmap, const char *name,
7243 + const char *parent_name, u8 id)
7245 struct clk_system *sys;
7246 struct clk *clk = NULL;
7247 struct clk_init_data init;
7250 if (!parent_name || id > SYSTEM_MAX_ID)
7251 return ERR_PTR(-EINVAL);
7252 @@ -123,44 +111,33 @@ at91_clk_register_system(struct at91_pmc *pmc, const char *name,
7255 sys->hw.init = &init;
7259 - init_waitqueue_head(&sys->wait);
7260 - irq_set_status_flags(sys->irq, IRQ_NOAUTOEN);
7261 - ret = request_irq(sys->irq, clk_system_irq_handler,
7262 - IRQF_TRIGGER_HIGH, name, sys);
7265 - return ERR_PTR(ret);
7268 + sys->regmap = regmap;
7270 clk = clk_register(NULL, &sys->hw);
7271 - if (IS_ERR(clk)) {
7273 - free_irq(sys->irq, sys);
7282 -of_at91_clk_sys_setup(struct device_node *np, struct at91_pmc *pmc)
7283 +static void __init of_at91rm9200_clk_sys_setup(struct device_node *np)
7290 struct device_node *sysclknp;
7291 const char *parent_name;
7292 + struct regmap *regmap;
7294 num = of_get_child_count(np);
7295 if (num > (SYSTEM_MAX_ID + 1))
7298 + regmap = syscon_node_to_regmap(of_get_parent(np));
7299 + if (IS_ERR(regmap))
7302 for_each_child_of_node(np, sysclknp) {
7303 if (of_property_read_u32(sysclknp, "reg", &id))
7305 @@ -168,21 +145,14 @@ of_at91_clk_sys_setup(struct device_node *np, struct at91_pmc *pmc)
7306 if (of_property_read_string(np, "clock-output-names", &name))
7307 name = sysclknp->name;
7310 - irq = irq_of_parse_and_map(sysclknp, 0);
7312 parent_name = of_clk_get_parent_name(sysclknp, 0);
7314 - clk = at91_clk_register_system(pmc, name, parent_name, id, irq);
7315 + clk = at91_clk_register_system(regmap, name, parent_name, id);
7319 of_clk_add_provider(sysclknp, of_clk_src_simple_get, clk);
7323 -void __init of_at91rm9200_clk_sys_setup(struct device_node *np,
7324 - struct at91_pmc *pmc)
7326 - of_at91_clk_sys_setup(np, pmc);
7328 +CLK_OF_DECLARE(at91rm9200_clk_sys, "atmel,at91rm9200-clk-system",
7329 + of_at91rm9200_clk_sys_setup);
7330 diff --git a/drivers/clk/at91/clk-usb.c b/drivers/clk/at91/clk-usb.c
7331 index 8ab8502778a2..650ca45892c0 100644
7332 --- a/drivers/clk/at91/clk-usb.c
7333 +++ b/drivers/clk/at91/clk-usb.c
7335 #include <linux/clkdev.h>
7336 #include <linux/clk/at91_pmc.h>
7337 #include <linux/of.h>
7338 -#include <linux/of_address.h>
7339 -#include <linux/io.h>
7340 +#include <linux/mfd/syscon.h>
7341 +#include <linux/regmap.h>
7347 struct at91sam9x5_clk_usb {
7349 - struct at91_pmc *pmc;
7350 + struct regmap *regmap;
7353 #define to_at91sam9x5_clk_usb(hw) \
7354 @@ -35,7 +35,7 @@ struct at91sam9x5_clk_usb {
7356 struct at91rm9200_clk_usb {
7358 - struct at91_pmc *pmc;
7359 + struct regmap *regmap;
7363 @@ -45,13 +45,12 @@ struct at91rm9200_clk_usb {
7364 static unsigned long at91sam9x5_clk_usb_recalc_rate(struct clk_hw *hw,
7365 unsigned long parent_rate)
7369 struct at91sam9x5_clk_usb *usb = to_at91sam9x5_clk_usb(hw);
7370 - struct at91_pmc *pmc = usb->pmc;
7371 + unsigned int usbr;
7374 - tmp = pmc_read(pmc, AT91_PMC_USB);
7375 - usbdiv = (tmp & AT91_PMC_OHCIUSBDIV) >> SAM9X5_USB_DIV_SHIFT;
7376 + regmap_read(usb->regmap, AT91_PMC_USB, &usbr);
7377 + usbdiv = (usbr & AT91_PMC_OHCIUSBDIV) >> SAM9X5_USB_DIV_SHIFT;
7379 return DIV_ROUND_CLOSEST(parent_rate, (usbdiv + 1));
7381 @@ -109,33 +108,31 @@ static int at91sam9x5_clk_usb_determine_rate(struct clk_hw *hw,
7383 static int at91sam9x5_clk_usb_set_parent(struct clk_hw *hw, u8 index)
7386 struct at91sam9x5_clk_usb *usb = to_at91sam9x5_clk_usb(hw);
7387 - struct at91_pmc *pmc = usb->pmc;
7391 - tmp = pmc_read(pmc, AT91_PMC_USB) & ~AT91_PMC_USBS;
7393 - tmp |= AT91_PMC_USBS;
7394 - pmc_write(pmc, AT91_PMC_USB, tmp);
7396 + regmap_update_bits(usb->regmap, AT91_PMC_USB, AT91_PMC_USBS,
7397 + index ? AT91_PMC_USBS : 0);
7402 static u8 at91sam9x5_clk_usb_get_parent(struct clk_hw *hw)
7404 struct at91sam9x5_clk_usb *usb = to_at91sam9x5_clk_usb(hw);
7405 - struct at91_pmc *pmc = usb->pmc;
7406 + unsigned int usbr;
7408 - return pmc_read(pmc, AT91_PMC_USB) & AT91_PMC_USBS;
7409 + regmap_read(usb->regmap, AT91_PMC_USB, &usbr);
7411 + return usbr & AT91_PMC_USBS;
7414 static int at91sam9x5_clk_usb_set_rate(struct clk_hw *hw, unsigned long rate,
7415 unsigned long parent_rate)
7418 struct at91sam9x5_clk_usb *usb = to_at91sam9x5_clk_usb(hw);
7419 - struct at91_pmc *pmc = usb->pmc;
7423 @@ -145,9 +142,8 @@ static int at91sam9x5_clk_usb_set_rate(struct clk_hw *hw, unsigned long rate,
7424 if (div > SAM9X5_USB_MAX_DIV + 1 || !div)
7427 - tmp = pmc_read(pmc, AT91_PMC_USB) & ~AT91_PMC_OHCIUSBDIV;
7428 - tmp |= (div - 1) << SAM9X5_USB_DIV_SHIFT;
7429 - pmc_write(pmc, AT91_PMC_USB, tmp);
7430 + regmap_update_bits(usb->regmap, AT91_PMC_USB, AT91_PMC_OHCIUSBDIV,
7431 + (div - 1) << SAM9X5_USB_DIV_SHIFT);
7435 @@ -163,28 +159,28 @@ static const struct clk_ops at91sam9x5_usb_ops = {
7436 static int at91sam9n12_clk_usb_enable(struct clk_hw *hw)
7438 struct at91sam9x5_clk_usb *usb = to_at91sam9x5_clk_usb(hw);
7439 - struct at91_pmc *pmc = usb->pmc;
7441 - pmc_write(pmc, AT91_PMC_USB,
7442 - pmc_read(pmc, AT91_PMC_USB) | AT91_PMC_USBS);
7443 + regmap_update_bits(usb->regmap, AT91_PMC_USB, AT91_PMC_USBS,
7449 static void at91sam9n12_clk_usb_disable(struct clk_hw *hw)
7451 struct at91sam9x5_clk_usb *usb = to_at91sam9x5_clk_usb(hw);
7452 - struct at91_pmc *pmc = usb->pmc;
7454 - pmc_write(pmc, AT91_PMC_USB,
7455 - pmc_read(pmc, AT91_PMC_USB) & ~AT91_PMC_USBS);
7456 + regmap_update_bits(usb->regmap, AT91_PMC_USB, AT91_PMC_USBS, 0);
7459 static int at91sam9n12_clk_usb_is_enabled(struct clk_hw *hw)
7461 struct at91sam9x5_clk_usb *usb = to_at91sam9x5_clk_usb(hw);
7462 - struct at91_pmc *pmc = usb->pmc;
7463 + unsigned int usbr;
7465 - return !!(pmc_read(pmc, AT91_PMC_USB) & AT91_PMC_USBS);
7466 + regmap_read(usb->regmap, AT91_PMC_USB, &usbr);
7468 + return usbr & AT91_PMC_USBS;
7471 static const struct clk_ops at91sam9n12_usb_ops = {
7472 @@ -197,7 +193,7 @@ static const struct clk_ops at91sam9n12_usb_ops = {
7475 static struct clk * __init
7476 -at91sam9x5_clk_register_usb(struct at91_pmc *pmc, const char *name,
7477 +at91sam9x5_clk_register_usb(struct regmap *regmap, const char *name,
7478 const char **parent_names, u8 num_parents)
7480 struct at91sam9x5_clk_usb *usb;
7481 @@ -216,7 +212,7 @@ at91sam9x5_clk_register_usb(struct at91_pmc *pmc, const char *name,
7482 CLK_SET_RATE_PARENT;
7484 usb->hw.init = &init;
7486 + usb->regmap = regmap;
7488 clk = clk_register(NULL, &usb->hw);
7490 @@ -226,7 +222,7 @@ at91sam9x5_clk_register_usb(struct at91_pmc *pmc, const char *name,
7493 static struct clk * __init
7494 -at91sam9n12_clk_register_usb(struct at91_pmc *pmc, const char *name,
7495 +at91sam9n12_clk_register_usb(struct regmap *regmap, const char *name,
7496 const char *parent_name)
7498 struct at91sam9x5_clk_usb *usb;
7499 @@ -244,7 +240,7 @@ at91sam9n12_clk_register_usb(struct at91_pmc *pmc, const char *name,
7500 init.flags = CLK_SET_RATE_GATE | CLK_SET_RATE_PARENT;
7502 usb->hw.init = &init;
7504 + usb->regmap = regmap;
7506 clk = clk_register(NULL, &usb->hw);
7508 @@ -257,12 +253,12 @@ static unsigned long at91rm9200_clk_usb_recalc_rate(struct clk_hw *hw,
7509 unsigned long parent_rate)
7511 struct at91rm9200_clk_usb *usb = to_at91rm9200_clk_usb(hw);
7512 - struct at91_pmc *pmc = usb->pmc;
7514 + unsigned int pllbr;
7517 - tmp = pmc_read(pmc, AT91_CKGR_PLLBR);
7518 - usbdiv = (tmp & AT91_PMC_USBDIV) >> RM9200_USB_DIV_SHIFT;
7519 + regmap_read(usb->regmap, AT91_CKGR_PLLBR, &pllbr);
7521 + usbdiv = (pllbr & AT91_PMC_USBDIV) >> RM9200_USB_DIV_SHIFT;
7522 if (usb->divisors[usbdiv])
7523 return parent_rate / usb->divisors[usbdiv];
7525 @@ -310,10 +306,8 @@ static long at91rm9200_clk_usb_round_rate(struct clk_hw *hw, unsigned long rate,
7526 static int at91rm9200_clk_usb_set_rate(struct clk_hw *hw, unsigned long rate,
7527 unsigned long parent_rate)
7531 struct at91rm9200_clk_usb *usb = to_at91rm9200_clk_usb(hw);
7532 - struct at91_pmc *pmc = usb->pmc;
7536 @@ -323,10 +317,10 @@ static int at91rm9200_clk_usb_set_rate(struct clk_hw *hw, unsigned long rate,
7538 for (i = 0; i < RM9200_USB_DIV_TAB_SIZE; i++) {
7539 if (usb->divisors[i] == div) {
7540 - tmp = pmc_read(pmc, AT91_CKGR_PLLBR) &
7542 - tmp |= i << RM9200_USB_DIV_SHIFT;
7543 - pmc_write(pmc, AT91_CKGR_PLLBR, tmp);
7544 + regmap_update_bits(usb->regmap, AT91_CKGR_PLLBR,
7546 + i << RM9200_USB_DIV_SHIFT);
7551 @@ -341,7 +335,7 @@ static const struct clk_ops at91rm9200_usb_ops = {
7554 static struct clk * __init
7555 -at91rm9200_clk_register_usb(struct at91_pmc *pmc, const char *name,
7556 +at91rm9200_clk_register_usb(struct regmap *regmap, const char *name,
7557 const char *parent_name, const u32 *divisors)
7559 struct at91rm9200_clk_usb *usb;
7560 @@ -359,7 +353,7 @@ at91rm9200_clk_register_usb(struct at91_pmc *pmc, const char *name,
7561 init.flags = CLK_SET_RATE_PARENT;
7563 usb->hw.init = &init;
7565 + usb->regmap = regmap;
7566 memcpy(usb->divisors, divisors, sizeof(usb->divisors));
7568 clk = clk_register(NULL, &usb->hw);
7569 @@ -369,13 +363,13 @@ at91rm9200_clk_register_usb(struct at91_pmc *pmc, const char *name,
7573 -void __init of_at91sam9x5_clk_usb_setup(struct device_node *np,
7574 - struct at91_pmc *pmc)
7575 +static void __init of_at91sam9x5_clk_usb_setup(struct device_node *np)
7579 const char *parent_names[USB_SOURCE_MAX];
7580 const char *name = np->name;
7581 + struct regmap *regmap;
7583 num_parents = of_clk_get_parent_count(np);
7584 if (num_parents <= 0 || num_parents > USB_SOURCE_MAX)
7585 @@ -385,19 +379,26 @@ void __init of_at91sam9x5_clk_usb_setup(struct device_node *np,
7587 of_property_read_string(np, "clock-output-names", &name);
7589 - clk = at91sam9x5_clk_register_usb(pmc, name, parent_names, num_parents);
7590 + regmap = syscon_node_to_regmap(of_get_parent(np));
7591 + if (IS_ERR(regmap))
7594 + clk = at91sam9x5_clk_register_usb(regmap, name, parent_names,
7599 of_clk_add_provider(np, of_clk_src_simple_get, clk);
7601 +CLK_OF_DECLARE(at91sam9x5_clk_usb, "atmel,at91sam9x5-clk-usb",
7602 + of_at91sam9x5_clk_usb_setup);
7604 -void __init of_at91sam9n12_clk_usb_setup(struct device_node *np,
7605 - struct at91_pmc *pmc)
7606 +static void __init of_at91sam9n12_clk_usb_setup(struct device_node *np)
7609 const char *parent_name;
7610 const char *name = np->name;
7611 + struct regmap *regmap;
7613 parent_name = of_clk_get_parent_name(np, 0);
7615 @@ -405,20 +406,26 @@ void __init of_at91sam9n12_clk_usb_setup(struct device_node *np,
7617 of_property_read_string(np, "clock-output-names", &name);
7619 - clk = at91sam9n12_clk_register_usb(pmc, name, parent_name);
7620 + regmap = syscon_node_to_regmap(of_get_parent(np));
7621 + if (IS_ERR(regmap))
7624 + clk = at91sam9n12_clk_register_usb(regmap, name, parent_name);
7628 of_clk_add_provider(np, of_clk_src_simple_get, clk);
7630 +CLK_OF_DECLARE(at91sam9n12_clk_usb, "atmel,at91sam9n12-clk-usb",
7631 + of_at91sam9n12_clk_usb_setup);
7633 -void __init of_at91rm9200_clk_usb_setup(struct device_node *np,
7634 - struct at91_pmc *pmc)
7635 +static void __init of_at91rm9200_clk_usb_setup(struct device_node *np)
7638 const char *parent_name;
7639 const char *name = np->name;
7640 u32 divisors[4] = {0, 0, 0, 0};
7641 + struct regmap *regmap;
7643 parent_name = of_clk_get_parent_name(np, 0);
7645 @@ -430,9 +437,15 @@ void __init of_at91rm9200_clk_usb_setup(struct device_node *np,
7647 of_property_read_string(np, "clock-output-names", &name);
7649 - clk = at91rm9200_clk_register_usb(pmc, name, parent_name, divisors);
7650 + regmap = syscon_node_to_regmap(of_get_parent(np));
7651 + if (IS_ERR(regmap))
7654 + clk = at91rm9200_clk_register_usb(regmap, name, parent_name, divisors);
7658 of_clk_add_provider(np, of_clk_src_simple_get, clk);
7660 +CLK_OF_DECLARE(at91rm9200_clk_usb, "atmel,at91rm9200-clk-usb",
7661 + of_at91rm9200_clk_usb_setup);
7662 diff --git a/drivers/clk/at91/clk-utmi.c b/drivers/clk/at91/clk-utmi.c
7663 index ca561e90a60f..61fcf399e58c 100644
7664 --- a/drivers/clk/at91/clk-utmi.c
7665 +++ b/drivers/clk/at91/clk-utmi.c
7667 #include <linux/clk-provider.h>
7668 #include <linux/clkdev.h>
7669 #include <linux/clk/at91_pmc.h>
7670 -#include <linux/interrupt.h>
7671 -#include <linux/irq.h>
7672 #include <linux/of.h>
7673 -#include <linux/of_address.h>
7674 -#include <linux/of_irq.h>
7675 -#include <linux/io.h>
7676 -#include <linux/sched.h>
7677 -#include <linux/wait.h>
7678 +#include <linux/mfd/syscon.h>
7679 +#include <linux/regmap.h>
7687 - struct at91_pmc *pmc;
7689 - wait_queue_head_t wait;
7690 + struct regmap *regmap;
7693 #define to_clk_utmi(hw) container_of(hw, struct clk_utmi, hw)
7695 -static irqreturn_t clk_utmi_irq_handler(int irq, void *dev_id)
7696 +static inline bool clk_utmi_ready(struct regmap *regmap)
7698 - struct clk_utmi *utmi = (struct clk_utmi *)dev_id;
7699 + unsigned int status;
7701 - wake_up(&utmi->wait);
7702 - disable_irq_nosync(utmi->irq);
7703 + regmap_read(regmap, AT91_PMC_SR, &status);
7705 - return IRQ_HANDLED;
7706 + return status & AT91_PMC_LOCKU;
7709 static int clk_utmi_prepare(struct clk_hw *hw)
7711 struct clk_utmi *utmi = to_clk_utmi(hw);
7712 - struct at91_pmc *pmc = utmi->pmc;
7713 - u32 tmp = pmc_read(pmc, AT91_CKGR_UCKR) | AT91_PMC_UPLLEN |
7714 - AT91_PMC_UPLLCOUNT | AT91_PMC_BIASEN;
7715 + unsigned int uckr = AT91_PMC_UPLLEN | AT91_PMC_UPLLCOUNT |
7718 - pmc_write(pmc, AT91_CKGR_UCKR, tmp);
7719 + regmap_update_bits(utmi->regmap, AT91_CKGR_UCKR, uckr, uckr);
7721 - while (!(pmc_read(pmc, AT91_PMC_SR) & AT91_PMC_LOCKU)) {
7722 - enable_irq(utmi->irq);
7723 - wait_event(utmi->wait,
7724 - pmc_read(pmc, AT91_PMC_SR) & AT91_PMC_LOCKU);
7726 + while (!clk_utmi_ready(utmi->regmap))
7731 @@ -64,18 +52,15 @@ static int clk_utmi_prepare(struct clk_hw *hw)
7732 static int clk_utmi_is_prepared(struct clk_hw *hw)
7734 struct clk_utmi *utmi = to_clk_utmi(hw);
7735 - struct at91_pmc *pmc = utmi->pmc;
7737 - return !!(pmc_read(pmc, AT91_PMC_SR) & AT91_PMC_LOCKU);
7738 + return clk_utmi_ready(utmi->regmap);
7741 static void clk_utmi_unprepare(struct clk_hw *hw)
7743 struct clk_utmi *utmi = to_clk_utmi(hw);
7744 - struct at91_pmc *pmc = utmi->pmc;
7745 - u32 tmp = pmc_read(pmc, AT91_CKGR_UCKR) & ~AT91_PMC_UPLLEN;
7747 - pmc_write(pmc, AT91_CKGR_UCKR, tmp);
7748 + regmap_update_bits(utmi->regmap, AT91_CKGR_UCKR, AT91_PMC_UPLLEN, 0);
7751 static unsigned long clk_utmi_recalc_rate(struct clk_hw *hw,
7752 @@ -93,10 +78,9 @@ static const struct clk_ops utmi_ops = {
7755 static struct clk * __init
7756 -at91_clk_register_utmi(struct at91_pmc *pmc, unsigned int irq,
7757 +at91_clk_register_utmi(struct regmap *regmap,
7758 const char *name, const char *parent_name)
7761 struct clk_utmi *utmi;
7762 struct clk *clk = NULL;
7763 struct clk_init_data init;
7764 @@ -112,52 +96,36 @@ at91_clk_register_utmi(struct at91_pmc *pmc, unsigned int irq,
7765 init.flags = CLK_SET_RATE_GATE;
7767 utmi->hw.init = &init;
7770 - init_waitqueue_head(&utmi->wait);
7771 - irq_set_status_flags(utmi->irq, IRQ_NOAUTOEN);
7772 - ret = request_irq(utmi->irq, clk_utmi_irq_handler,
7773 - IRQF_TRIGGER_HIGH, "clk-utmi", utmi);
7776 - return ERR_PTR(ret);
7778 + utmi->regmap = regmap;
7780 clk = clk_register(NULL, &utmi->hw);
7781 - if (IS_ERR(clk)) {
7782 - free_irq(utmi->irq, utmi);
7791 -of_at91_clk_utmi_setup(struct device_node *np, struct at91_pmc *pmc)
7792 +static void __init of_at91sam9x5_clk_utmi_setup(struct device_node *np)
7796 const char *parent_name;
7797 const char *name = np->name;
7798 + struct regmap *regmap;
7800 parent_name = of_clk_get_parent_name(np, 0);
7802 of_property_read_string(np, "clock-output-names", &name);
7804 - irq = irq_of_parse_and_map(np, 0);
7806 + regmap = syscon_node_to_regmap(of_get_parent(np));
7807 + if (IS_ERR(regmap))
7810 - clk = at91_clk_register_utmi(pmc, irq, name, parent_name);
7811 + clk = at91_clk_register_utmi(regmap, name, parent_name);
7815 of_clk_add_provider(np, of_clk_src_simple_get, clk);
7819 -void __init of_at91sam9x5_clk_utmi_setup(struct device_node *np,
7820 - struct at91_pmc *pmc)
7822 - of_at91_clk_utmi_setup(np, pmc);
7824 +CLK_OF_DECLARE(at91sam9x5_clk_utmi, "atmel,at91sam9x5-clk-utmi",
7825 + of_at91sam9x5_clk_utmi_setup);
7826 diff --git a/drivers/clk/at91/pmc.c b/drivers/clk/at91/pmc.c
7827 index 8476b570779b..526df5ba042d 100644
7828 --- a/drivers/clk/at91/pmc.c
7829 +++ b/drivers/clk/at91/pmc.c
7831 #include <linux/clkdev.h>
7832 #include <linux/clk/at91_pmc.h>
7833 #include <linux/of.h>
7834 -#include <linux/of_address.h>
7835 -#include <linux/io.h>
7836 -#include <linux/interrupt.h>
7837 -#include <linux/irq.h>
7838 -#include <linux/irqchip/chained_irq.h>
7839 -#include <linux/irqdomain.h>
7840 -#include <linux/of_irq.h>
7841 +#include <linux/mfd/syscon.h>
7842 +#include <linux/regmap.h>
7844 #include <asm/proc-fns.h>
7848 -void __iomem *at91_pmc_base;
7849 -EXPORT_SYMBOL_GPL(at91_pmc_base);
7851 -void at91rm9200_idle(void)
7854 - * Disable the processor clock. The processor will be automatically
7855 - * re-enabled by an interrupt or by a reset.
7857 - at91_pmc_write(AT91_PMC_SCDR, AT91_PMC_PCK);
7860 -void at91sam9_idle(void)
7862 - at91_pmc_write(AT91_PMC_SCDR, AT91_PMC_PCK);
7866 int of_at91_get_clk_range(struct device_node *np, const char *propname,
7867 struct clk_range *range)
7869 @@ -64,402 +41,3 @@ int of_at91_get_clk_range(struct device_node *np, const char *propname,
7872 EXPORT_SYMBOL_GPL(of_at91_get_clk_range);
7874 -static void pmc_irq_mask(struct irq_data *d)
7876 - struct at91_pmc *pmc = irq_data_get_irq_chip_data(d);
7878 - pmc_write(pmc, AT91_PMC_IDR, 1 << d->hwirq);
7881 -static void pmc_irq_unmask(struct irq_data *d)
7883 - struct at91_pmc *pmc = irq_data_get_irq_chip_data(d);
7885 - pmc_write(pmc, AT91_PMC_IER, 1 << d->hwirq);
7888 -static int pmc_irq_set_type(struct irq_data *d, unsigned type)
7890 - if (type != IRQ_TYPE_LEVEL_HIGH) {
7891 - pr_warn("PMC: type not supported (support only IRQ_TYPE_LEVEL_HIGH type)\n");
7898 -static void pmc_irq_suspend(struct irq_data *d)
7900 - struct at91_pmc *pmc = irq_data_get_irq_chip_data(d);
7902 - pmc->imr = pmc_read(pmc, AT91_PMC_IMR);
7903 - pmc_write(pmc, AT91_PMC_IDR, pmc->imr);
7906 -static void pmc_irq_resume(struct irq_data *d)
7908 - struct at91_pmc *pmc = irq_data_get_irq_chip_data(d);
7910 - pmc_write(pmc, AT91_PMC_IER, pmc->imr);
7913 -static struct irq_chip pmc_irq = {
7915 - .irq_disable = pmc_irq_mask,
7916 - .irq_mask = pmc_irq_mask,
7917 - .irq_unmask = pmc_irq_unmask,
7918 - .irq_set_type = pmc_irq_set_type,
7919 - .irq_suspend = pmc_irq_suspend,
7920 - .irq_resume = pmc_irq_resume,
7923 -static struct lock_class_key pmc_lock_class;
7925 -static int pmc_irq_map(struct irq_domain *h, unsigned int virq,
7926 - irq_hw_number_t hw)
7928 - struct at91_pmc *pmc = h->host_data;
7930 - irq_set_lockdep_class(virq, &pmc_lock_class);
7932 - irq_set_chip_and_handler(virq, &pmc_irq,
7933 - handle_level_irq);
7934 - irq_set_chip_data(virq, pmc);
7939 -static int pmc_irq_domain_xlate(struct irq_domain *d,
7940 - struct device_node *ctrlr,
7941 - const u32 *intspec, unsigned int intsize,
7942 - irq_hw_number_t *out_hwirq,
7943 - unsigned int *out_type)
7945 - struct at91_pmc *pmc = d->host_data;
7946 - const struct at91_pmc_caps *caps = pmc->caps;
7948 - if (WARN_ON(intsize < 1))
7951 - *out_hwirq = intspec[0];
7953 - if (!(caps->available_irqs & (1 << *out_hwirq)))
7956 - *out_type = IRQ_TYPE_LEVEL_HIGH;
7961 -static const struct irq_domain_ops pmc_irq_ops = {
7962 - .map = pmc_irq_map,
7963 - .xlate = pmc_irq_domain_xlate,
7966 -static irqreturn_t pmc_irq_handler(int irq, void *data)
7968 - struct at91_pmc *pmc = (struct at91_pmc *)data;
7972 - sr = pmc_read(pmc, AT91_PMC_SR) & pmc_read(pmc, AT91_PMC_IMR);
7976 - for_each_set_bit(n, &sr, BITS_PER_LONG)
7977 - generic_handle_irq(irq_find_mapping(pmc->irqdomain, n));
7979 - return IRQ_HANDLED;
7982 -static const struct at91_pmc_caps at91rm9200_caps = {
7983 - .available_irqs = AT91_PMC_MOSCS | AT91_PMC_LOCKA | AT91_PMC_LOCKB |
7984 - AT91_PMC_MCKRDY | AT91_PMC_PCK0RDY |
7985 - AT91_PMC_PCK1RDY | AT91_PMC_PCK2RDY |
7989 -static const struct at91_pmc_caps at91sam9260_caps = {
7990 - .available_irqs = AT91_PMC_MOSCS | AT91_PMC_LOCKA | AT91_PMC_LOCKB |
7991 - AT91_PMC_MCKRDY | AT91_PMC_PCK0RDY |
7995 -static const struct at91_pmc_caps at91sam9g45_caps = {
7996 - .available_irqs = AT91_PMC_MOSCS | AT91_PMC_LOCKA | AT91_PMC_MCKRDY |
7997 - AT91_PMC_LOCKU | AT91_PMC_PCK0RDY |
8001 -static const struct at91_pmc_caps at91sam9n12_caps = {
8002 - .available_irqs = AT91_PMC_MOSCS | AT91_PMC_LOCKA | AT91_PMC_LOCKB |
8003 - AT91_PMC_MCKRDY | AT91_PMC_PCK0RDY |
8004 - AT91_PMC_PCK1RDY | AT91_PMC_MOSCSELS |
8005 - AT91_PMC_MOSCRCS | AT91_PMC_CFDEV,
8008 -static const struct at91_pmc_caps at91sam9x5_caps = {
8009 - .available_irqs = AT91_PMC_MOSCS | AT91_PMC_LOCKA | AT91_PMC_MCKRDY |
8010 - AT91_PMC_LOCKU | AT91_PMC_PCK0RDY |
8011 - AT91_PMC_PCK1RDY | AT91_PMC_MOSCSELS |
8012 - AT91_PMC_MOSCRCS | AT91_PMC_CFDEV,
8015 -static const struct at91_pmc_caps sama5d2_caps = {
8016 - .available_irqs = AT91_PMC_MOSCS | AT91_PMC_LOCKA | AT91_PMC_MCKRDY |
8017 - AT91_PMC_LOCKU | AT91_PMC_PCK0RDY |
8018 - AT91_PMC_PCK1RDY | AT91_PMC_PCK2RDY |
8019 - AT91_PMC_MOSCSELS | AT91_PMC_MOSCRCS |
8020 - AT91_PMC_CFDEV | AT91_PMC_GCKRDY,
8023 -static const struct at91_pmc_caps sama5d3_caps = {
8024 - .available_irqs = AT91_PMC_MOSCS | AT91_PMC_LOCKA | AT91_PMC_MCKRDY |
8025 - AT91_PMC_LOCKU | AT91_PMC_PCK0RDY |
8026 - AT91_PMC_PCK1RDY | AT91_PMC_PCK2RDY |
8027 - AT91_PMC_MOSCSELS | AT91_PMC_MOSCRCS |
8031 -static struct at91_pmc *__init at91_pmc_init(struct device_node *np,
8032 - void __iomem *regbase, int virq,
8033 - const struct at91_pmc_caps *caps)
8035 - struct at91_pmc *pmc;
8037 - if (!regbase || !virq || !caps)
8040 - at91_pmc_base = regbase;
8042 - pmc = kzalloc(sizeof(*pmc), GFP_KERNEL);
8046 - spin_lock_init(&pmc->lock);
8047 - pmc->regbase = regbase;
8051 - pmc->irqdomain = irq_domain_add_linear(np, 32, &pmc_irq_ops, pmc);
8053 - if (!pmc->irqdomain)
8054 - goto out_free_pmc;
8056 - pmc_write(pmc, AT91_PMC_IDR, 0xffffffff);
8057 - if (request_irq(pmc->virq, pmc_irq_handler,
8058 - IRQF_SHARED | IRQF_COND_SUSPEND, "pmc", pmc))
8059 - goto out_remove_irqdomain;
8063 -out_remove_irqdomain:
8064 - irq_domain_remove(pmc->irqdomain);
8071 -static const struct of_device_id pmc_clk_ids[] __initconst = {
8072 - /* Slow oscillator */
8074 - .compatible = "atmel,at91sam9260-clk-slow",
8075 - .data = of_at91sam9260_clk_slow_setup,
8079 - .compatible = "atmel,at91rm9200-clk-main-osc",
8080 - .data = of_at91rm9200_clk_main_osc_setup,
8083 - .compatible = "atmel,at91sam9x5-clk-main-rc-osc",
8084 - .data = of_at91sam9x5_clk_main_rc_osc_setup,
8087 - .compatible = "atmel,at91rm9200-clk-main",
8088 - .data = of_at91rm9200_clk_main_setup,
8091 - .compatible = "atmel,at91sam9x5-clk-main",
8092 - .data = of_at91sam9x5_clk_main_setup,
8096 - .compatible = "atmel,at91rm9200-clk-pll",
8097 - .data = of_at91rm9200_clk_pll_setup,
8100 - .compatible = "atmel,at91sam9g45-clk-pll",
8101 - .data = of_at91sam9g45_clk_pll_setup,
8104 - .compatible = "atmel,at91sam9g20-clk-pllb",
8105 - .data = of_at91sam9g20_clk_pllb_setup,
8108 - .compatible = "atmel,sama5d3-clk-pll",
8109 - .data = of_sama5d3_clk_pll_setup,
8112 - .compatible = "atmel,at91sam9x5-clk-plldiv",
8113 - .data = of_at91sam9x5_clk_plldiv_setup,
8115 - /* Master clock */
8117 - .compatible = "atmel,at91rm9200-clk-master",
8118 - .data = of_at91rm9200_clk_master_setup,
8121 - .compatible = "atmel,at91sam9x5-clk-master",
8122 - .data = of_at91sam9x5_clk_master_setup,
8124 - /* System clocks */
8126 - .compatible = "atmel,at91rm9200-clk-system",
8127 - .data = of_at91rm9200_clk_sys_setup,
8129 - /* Peripheral clocks */
8131 - .compatible = "atmel,at91rm9200-clk-peripheral",
8132 - .data = of_at91rm9200_clk_periph_setup,
8135 - .compatible = "atmel,at91sam9x5-clk-peripheral",
8136 - .data = of_at91sam9x5_clk_periph_setup,
8138 - /* Programmable clocks */
8140 - .compatible = "atmel,at91rm9200-clk-programmable",
8141 - .data = of_at91rm9200_clk_prog_setup,
8144 - .compatible = "atmel,at91sam9g45-clk-programmable",
8145 - .data = of_at91sam9g45_clk_prog_setup,
8148 - .compatible = "atmel,at91sam9x5-clk-programmable",
8149 - .data = of_at91sam9x5_clk_prog_setup,
8152 -#if defined(CONFIG_HAVE_AT91_UTMI)
8154 - .compatible = "atmel,at91sam9x5-clk-utmi",
8155 - .data = of_at91sam9x5_clk_utmi_setup,
8159 -#if defined(CONFIG_HAVE_AT91_USB_CLK)
8161 - .compatible = "atmel,at91rm9200-clk-usb",
8162 - .data = of_at91rm9200_clk_usb_setup,
8165 - .compatible = "atmel,at91sam9x5-clk-usb",
8166 - .data = of_at91sam9x5_clk_usb_setup,
8169 - .compatible = "atmel,at91sam9n12-clk-usb",
8170 - .data = of_at91sam9n12_clk_usb_setup,
8174 -#if defined(CONFIG_HAVE_AT91_SMD)
8176 - .compatible = "atmel,at91sam9x5-clk-smd",
8177 - .data = of_at91sam9x5_clk_smd_setup,
8180 -#if defined(CONFIG_HAVE_AT91_H32MX)
8182 - .compatible = "atmel,sama5d4-clk-h32mx",
8183 - .data = of_sama5d4_clk_h32mx_setup,
8186 -#if defined(CONFIG_HAVE_AT91_GENERATED_CLK)
8188 - .compatible = "atmel,sama5d2-clk-generated",
8189 - .data = of_sama5d2_clk_generated_setup,
8195 -static void __init of_at91_pmc_setup(struct device_node *np,
8196 - const struct at91_pmc_caps *caps)
8198 - struct at91_pmc *pmc;
8199 - struct device_node *childnp;
8200 - void (*clk_setup)(struct device_node *, struct at91_pmc *);
8201 - const struct of_device_id *clk_id;
8202 - void __iomem *regbase = of_iomap(np, 0);
8208 - virq = irq_of_parse_and_map(np, 0);
8212 - pmc = at91_pmc_init(np, regbase, virq, caps);
8215 - for_each_child_of_node(np, childnp) {
8216 - clk_id = of_match_node(pmc_clk_ids, childnp);
8219 - clk_setup = clk_id->data;
8220 - clk_setup(childnp, pmc);
8224 -static void __init of_at91rm9200_pmc_setup(struct device_node *np)
8226 - of_at91_pmc_setup(np, &at91rm9200_caps);
8228 -CLK_OF_DECLARE(at91rm9200_clk_pmc, "atmel,at91rm9200-pmc",
8229 - of_at91rm9200_pmc_setup);
8231 -static void __init of_at91sam9260_pmc_setup(struct device_node *np)
8233 - of_at91_pmc_setup(np, &at91sam9260_caps);
8235 -CLK_OF_DECLARE(at91sam9260_clk_pmc, "atmel,at91sam9260-pmc",
8236 - of_at91sam9260_pmc_setup);
8238 -static void __init of_at91sam9g45_pmc_setup(struct device_node *np)
8240 - of_at91_pmc_setup(np, &at91sam9g45_caps);
8242 -CLK_OF_DECLARE(at91sam9g45_clk_pmc, "atmel,at91sam9g45-pmc",
8243 - of_at91sam9g45_pmc_setup);
8245 -static void __init of_at91sam9n12_pmc_setup(struct device_node *np)
8247 - of_at91_pmc_setup(np, &at91sam9n12_caps);
8249 -CLK_OF_DECLARE(at91sam9n12_clk_pmc, "atmel,at91sam9n12-pmc",
8250 - of_at91sam9n12_pmc_setup);
8252 -static void __init of_at91sam9x5_pmc_setup(struct device_node *np)
8254 - of_at91_pmc_setup(np, &at91sam9x5_caps);
8256 -CLK_OF_DECLARE(at91sam9x5_clk_pmc, "atmel,at91sam9x5-pmc",
8257 - of_at91sam9x5_pmc_setup);
8259 -static void __init of_sama5d2_pmc_setup(struct device_node *np)
8261 - of_at91_pmc_setup(np, &sama5d2_caps);
8263 -CLK_OF_DECLARE(sama5d2_clk_pmc, "atmel,sama5d2-pmc",
8264 - of_sama5d2_pmc_setup);
8266 -static void __init of_sama5d3_pmc_setup(struct device_node *np)
8268 - of_at91_pmc_setup(np, &sama5d3_caps);
8270 -CLK_OF_DECLARE(sama5d3_clk_pmc, "atmel,sama5d3-pmc",
8271 - of_sama5d3_pmc_setup);
8272 diff --git a/drivers/clk/at91/pmc.h b/drivers/clk/at91/pmc.h
8273 index f65739272779..5771fff0ee3f 100644
8274 --- a/drivers/clk/at91/pmc.h
8275 +++ b/drivers/clk/at91/pmc.h
8278 #include <linux/io.h>
8279 #include <linux/irqdomain.h>
8280 +#include <linux/regmap.h>
8281 #include <linux/spinlock.h>
8283 +extern spinlock_t pmc_pcr_lock;
8288 @@ -23,102 +26,7 @@ struct clk_range {
8290 #define CLK_RANGE(MIN, MAX) {.min = MIN, .max = MAX,}
8292 -struct at91_pmc_caps {
8293 - u32 available_irqs;
8297 - void __iomem *regbase;
8300 - const struct at91_pmc_caps *caps;
8301 - struct irq_domain *irqdomain;
8305 -static inline void pmc_lock(struct at91_pmc *pmc)
8307 - spin_lock(&pmc->lock);
8310 -static inline void pmc_unlock(struct at91_pmc *pmc)
8312 - spin_unlock(&pmc->lock);
8315 -static inline u32 pmc_read(struct at91_pmc *pmc, int offset)
8317 - return readl(pmc->regbase + offset);
8320 -static inline void pmc_write(struct at91_pmc *pmc, int offset, u32 value)
8322 - writel(value, pmc->regbase + offset);
8325 int of_at91_get_clk_range(struct device_node *np, const char *propname,
8326 struct clk_range *range);
8328 -void of_at91sam9260_clk_slow_setup(struct device_node *np,
8329 - struct at91_pmc *pmc);
8331 -void of_at91rm9200_clk_main_osc_setup(struct device_node *np,
8332 - struct at91_pmc *pmc);
8333 -void of_at91sam9x5_clk_main_rc_osc_setup(struct device_node *np,
8334 - struct at91_pmc *pmc);
8335 -void of_at91rm9200_clk_main_setup(struct device_node *np,
8336 - struct at91_pmc *pmc);
8337 -void of_at91sam9x5_clk_main_setup(struct device_node *np,
8338 - struct at91_pmc *pmc);
8340 -void of_at91rm9200_clk_pll_setup(struct device_node *np,
8341 - struct at91_pmc *pmc);
8342 -void of_at91sam9g45_clk_pll_setup(struct device_node *np,
8343 - struct at91_pmc *pmc);
8344 -void of_at91sam9g20_clk_pllb_setup(struct device_node *np,
8345 - struct at91_pmc *pmc);
8346 -void of_sama5d3_clk_pll_setup(struct device_node *np,
8347 - struct at91_pmc *pmc);
8348 -void of_at91sam9x5_clk_plldiv_setup(struct device_node *np,
8349 - struct at91_pmc *pmc);
8351 -void of_at91rm9200_clk_master_setup(struct device_node *np,
8352 - struct at91_pmc *pmc);
8353 -void of_at91sam9x5_clk_master_setup(struct device_node *np,
8354 - struct at91_pmc *pmc);
8356 -void of_at91rm9200_clk_sys_setup(struct device_node *np,
8357 - struct at91_pmc *pmc);
8359 -void of_at91rm9200_clk_periph_setup(struct device_node *np,
8360 - struct at91_pmc *pmc);
8361 -void of_at91sam9x5_clk_periph_setup(struct device_node *np,
8362 - struct at91_pmc *pmc);
8364 -void of_at91rm9200_clk_prog_setup(struct device_node *np,
8365 - struct at91_pmc *pmc);
8366 -void of_at91sam9g45_clk_prog_setup(struct device_node *np,
8367 - struct at91_pmc *pmc);
8368 -void of_at91sam9x5_clk_prog_setup(struct device_node *np,
8369 - struct at91_pmc *pmc);
8371 -void of_at91sam9x5_clk_utmi_setup(struct device_node *np,
8372 - struct at91_pmc *pmc);
8374 -void of_at91rm9200_clk_usb_setup(struct device_node *np,
8375 - struct at91_pmc *pmc);
8376 -void of_at91sam9x5_clk_usb_setup(struct device_node *np,
8377 - struct at91_pmc *pmc);
8378 -void of_at91sam9n12_clk_usb_setup(struct device_node *np,
8379 - struct at91_pmc *pmc);
8381 -void of_at91sam9x5_clk_smd_setup(struct device_node *np,
8382 - struct at91_pmc *pmc);
8384 -void of_sama5d4_clk_h32mx_setup(struct device_node *np,
8385 - struct at91_pmc *pmc);
8387 -void of_sama5d2_clk_generated_setup(struct device_node *np,
8388 - struct at91_pmc *pmc);
8390 #endif /* __PMC_H_ */
8391 diff --git a/drivers/clocksource/tcb_clksrc.c b/drivers/clocksource/tcb_clksrc.c
8392 index 4da2af9694a2..5b6f57f500b8 100644
8393 --- a/drivers/clocksource/tcb_clksrc.c
8394 +++ b/drivers/clocksource/tcb_clksrc.c
8396 * this 32 bit free-running counter. the second channel is not used.
8398 * - The third channel may be used to provide a 16-bit clockevent
8399 - * source, used in either periodic or oneshot mode. This runs
8400 - * at 32 KiHZ, and can handle delays of up to two seconds.
8401 + * source, used in either periodic or oneshot mode.
8403 * A boot clocksource and clockevent source are also currently needed,
8404 * unless the relevant platforms (ARM/AT91, AVR32/AT32) are changed so
8405 @@ -74,6 +73,8 @@ static struct clocksource clksrc = {
8406 struct tc_clkevt_device {
8407 struct clock_event_device clkevt;
8414 @@ -82,15 +83,26 @@ static struct tc_clkevt_device *to_tc_clkevt(struct clock_event_device *clkevt)
8415 return container_of(clkevt, struct tc_clkevt_device, clkevt);
8418 -/* For now, we always use the 32K clock ... this optimizes for NO_HZ,
8419 - * because using one of the divided clocks would usually mean the
8420 - * tick rate can never be less than several dozen Hz (vs 0.5 Hz).
8422 - * A divided clock could be good for high resolution timers, since
8423 - * 30.5 usec resolution can seem "low".
8425 static u32 timer_clock;
8427 +static void tc_clk_disable(struct clock_event_device *d)
8429 + struct tc_clkevt_device *tcd = to_tc_clkevt(d);
8431 + clk_disable(tcd->clk);
8432 + tcd->clk_enabled = false;
8435 +static void tc_clk_enable(struct clock_event_device *d)
8437 + struct tc_clkevt_device *tcd = to_tc_clkevt(d);
8439 + if (tcd->clk_enabled)
8441 + clk_enable(tcd->clk);
8442 + tcd->clk_enabled = true;
8445 static int tc_shutdown(struct clock_event_device *d)
8447 struct tc_clkevt_device *tcd = to_tc_clkevt(d);
8448 @@ -98,8 +110,14 @@ static int tc_shutdown(struct clock_event_device *d)
8450 __raw_writel(0xff, regs + ATMEL_TC_REG(2, IDR));
8451 __raw_writel(ATMEL_TC_CLKDIS, regs + ATMEL_TC_REG(2, CCR));
8455 +static int tc_shutdown_clk_off(struct clock_event_device *d)
8458 if (!clockevent_state_detached(d))
8459 - clk_disable(tcd->clk);
8460 + tc_clk_disable(d);
8464 @@ -112,9 +130,9 @@ static int tc_set_oneshot(struct clock_event_device *d)
8465 if (clockevent_state_oneshot(d) || clockevent_state_periodic(d))
8468 - clk_enable(tcd->clk);
8471 - /* slow clock, count up to RC, then irq and stop */
8472 + /* count up to RC, then irq and stop */
8473 __raw_writel(timer_clock | ATMEL_TC_CPCSTOP | ATMEL_TC_WAVE |
8474 ATMEL_TC_WAVESEL_UP_AUTO, regs + ATMEL_TC_REG(2, CMR));
8475 __raw_writel(ATMEL_TC_CPCS, regs + ATMEL_TC_REG(2, IER));
8476 @@ -134,12 +152,12 @@ static int tc_set_periodic(struct clock_event_device *d)
8477 /* By not making the gentime core emulate periodic mode on top
8478 * of oneshot, we get lower overhead and improved accuracy.
8480 - clk_enable(tcd->clk);
8483 - /* slow clock, count up to RC, then irq and restart */
8484 + /* count up to RC, then irq and restart */
8485 __raw_writel(timer_clock | ATMEL_TC_WAVE | ATMEL_TC_WAVESEL_UP_AUTO,
8486 regs + ATMEL_TC_REG(2, CMR));
8487 - __raw_writel((32768 + HZ / 2) / HZ, tcaddr + ATMEL_TC_REG(2, RC));
8488 + __raw_writel((tcd->freq + HZ / 2) / HZ, tcaddr + ATMEL_TC_REG(2, RC));
8490 /* Enable clock and interrupts on RC compare */
8491 __raw_writel(ATMEL_TC_CPCS, regs + ATMEL_TC_REG(2, IER));
8492 @@ -166,9 +184,13 @@ static struct tc_clkevt_device clkevt = {
8493 .features = CLOCK_EVT_FEAT_PERIODIC |
8494 CLOCK_EVT_FEAT_ONESHOT,
8495 /* Should be lower than at91rm9200's system timer */
8496 +#ifdef CONFIG_ATMEL_TCB_CLKSRC_USE_SLOW_CLOCK
8501 .set_next_event = tc_next_event,
8502 - .set_state_shutdown = tc_shutdown,
8503 + .set_state_shutdown = tc_shutdown_clk_off,
8504 .set_state_periodic = tc_set_periodic,
8505 .set_state_oneshot = tc_set_oneshot,
8507 @@ -188,8 +210,9 @@ static irqreturn_t ch2_irq(int irq, void *handle)
8511 -static int __init setup_clkevents(struct atmel_tc *tc, int clk32k_divisor_idx)
8512 +static int __init setup_clkevents(struct atmel_tc *tc, int divisor_idx)
8514 + unsigned divisor = atmel_tc_divisors[divisor_idx];
8516 struct clk *t2_clk = tc->clk[2];
8517 int irq = tc->irq[2];
8518 @@ -210,7 +233,11 @@ static int __init setup_clkevents(struct atmel_tc *tc, int clk32k_divisor_idx)
8519 clkevt.regs = tc->regs;
8520 clkevt.clk = t2_clk;
8522 - timer_clock = clk32k_divisor_idx;
8523 + timer_clock = divisor_idx;
8525 + clkevt.freq = 32768;
8527 + clkevt.freq = clk_get_rate(t2_clk) / divisor;
8529 clkevt.clkevt.cpumask = cpumask_of(0);
8531 @@ -221,7 +248,7 @@ static int __init setup_clkevents(struct atmel_tc *tc, int clk32k_divisor_idx)
8535 - clockevents_config_and_register(&clkevt.clkevt, 32768, 1, 0xffff);
8536 + clockevents_config_and_register(&clkevt.clkevt, clkevt.freq, 1, 0xffff);
8540 @@ -358,7 +385,11 @@ static int __init tcb_clksrc_init(void)
8541 goto err_disable_t1;
8543 /* channel 2: periodic and oneshot timer support */
8544 +#ifdef CONFIG_ATMEL_TCB_CLKSRC_USE_SLOW_CLOCK
8545 ret = setup_clkevents(tc, clk32k_divisor_idx);
8547 + ret = setup_clkevents(tc, best_divisor_idx);
8550 goto err_unregister_clksrc;
8552 diff --git a/drivers/clocksource/timer-atmel-pit.c b/drivers/clocksource/timer-atmel-pit.c
8553 index d911c5dca8f1..7a40f7e88468 100644
8554 --- a/drivers/clocksource/timer-atmel-pit.c
8555 +++ b/drivers/clocksource/timer-atmel-pit.c
8556 @@ -46,6 +46,7 @@ struct pit_data {
8560 + bool irq_requested;
8564 @@ -96,15 +97,29 @@ static int pit_clkevt_shutdown(struct clock_event_device *dev)
8566 /* disable irq, leaving the clocksource active */
8567 pit_write(data->base, AT91_PIT_MR, (data->cycle - 1) | AT91_PIT_PITEN);
8568 + if (data->irq_requested) {
8569 + free_irq(data->irq, data);
8570 + data->irq_requested = false;
8575 +static irqreturn_t at91sam926x_pit_interrupt(int irq, void *dev_id);
8577 * Clockevent device: interrupts every 1/HZ (== pit_cycles * MCK/16)
8579 static int pit_clkevt_set_periodic(struct clock_event_device *dev)
8581 struct pit_data *data = clkevt_to_pit_data(dev);
8584 + ret = request_irq(data->irq, at91sam926x_pit_interrupt,
8585 + IRQF_SHARED | IRQF_TIMER | IRQF_IRQPOLL,
8586 + "at91_tick", data);
8588 + panic(pr_fmt("Unable to setup IRQ\n"));
8590 + data->irq_requested = true;
8592 /* update clocksource counter */
8593 data->cnt += data->cycle * PIT_PICNT(pit_read(data->base, AT91_PIT_PIVR));
8594 @@ -181,7 +196,6 @@ static void __init at91sam926x_pit_common_init(struct pit_data *data)
8596 unsigned long pit_rate;
8601 * Use our actual MCK to figure out how many MCK/16 ticks per
8602 @@ -206,13 +220,6 @@ static void __init at91sam926x_pit_common_init(struct pit_data *data)
8603 data->clksrc.flags = CLOCK_SOURCE_IS_CONTINUOUS;
8604 clocksource_register_hz(&data->clksrc, pit_rate);
8606 - /* Set up irq handler */
8607 - ret = request_irq(data->irq, at91sam926x_pit_interrupt,
8608 - IRQF_SHARED | IRQF_TIMER | IRQF_IRQPOLL,
8609 - "at91_tick", data);
8611 - panic(pr_fmt("Unable to setup IRQ\n"));
8613 /* Set up and register clockevents */
8614 data->clkevt.name = "pit";
8615 data->clkevt.features = CLOCK_EVT_FEAT_PERIODIC;
8616 diff --git a/drivers/clocksource/timer-atmel-st.c b/drivers/clocksource/timer-atmel-st.c
8617 index 29d21d68df5a..103d0fd70cc4 100644
8618 --- a/drivers/clocksource/timer-atmel-st.c
8619 +++ b/drivers/clocksource/timer-atmel-st.c
8620 @@ -115,18 +115,29 @@ static void clkdev32k_disable_and_flush_irq(void)
8621 last_crtr = read_CRTR();
8624 +static int atmel_st_irq;
8626 static int clkevt32k_shutdown(struct clock_event_device *evt)
8628 clkdev32k_disable_and_flush_irq();
8630 regmap_write(regmap_st, AT91_ST_IER, irqmask);
8631 + free_irq(atmel_st_irq, regmap_st);
8635 static int clkevt32k_set_oneshot(struct clock_event_device *dev)
8639 clkdev32k_disable_and_flush_irq();
8641 + ret = request_irq(atmel_st_irq, at91rm9200_timer_interrupt,
8642 + IRQF_SHARED | IRQF_TIMER | IRQF_IRQPOLL,
8643 + "at91_tick", regmap_st);
8645 + panic(pr_fmt("Unable to setup IRQ\n"));
8648 * ALM for oneshot irqs, set by next_event()
8649 * before 32 seconds have passed.
8650 @@ -139,8 +150,16 @@ static int clkevt32k_set_oneshot(struct clock_event_device *dev)
8652 static int clkevt32k_set_periodic(struct clock_event_device *dev)
8656 clkdev32k_disable_and_flush_irq();
8658 + ret = request_irq(atmel_st_irq, at91rm9200_timer_interrupt,
8659 + IRQF_SHARED | IRQF_TIMER | IRQF_IRQPOLL,
8660 + "at91_tick", regmap_st);
8662 + panic(pr_fmt("Unable to setup IRQ\n"));
8664 /* PIT for periodic irqs; fixed rate of 1/HZ */
8665 irqmask = AT91_ST_PITS;
8666 regmap_write(regmap_st, AT91_ST_PIMR, timer_latch);
8667 @@ -198,7 +217,7 @@ static void __init atmel_st_timer_init(struct device_node *node)
8670 unsigned int sclk_rate, val;
8674 regmap_st = syscon_node_to_regmap(node);
8675 if (IS_ERR(regmap_st))
8676 @@ -210,17 +229,10 @@ static void __init atmel_st_timer_init(struct device_node *node)
8677 regmap_read(regmap_st, AT91_ST_SR, &val);
8679 /* Get the interrupts property */
8680 - irq = irq_of_parse_and_map(node, 0);
8682 + atmel_st_irq = irq_of_parse_and_map(node, 0);
8683 + if (!atmel_st_irq)
8684 panic(pr_fmt("Unable to get IRQ from DT\n"));
8686 - /* Make IRQs happen for the system timer */
8687 - ret = request_irq(irq, at91rm9200_timer_interrupt,
8688 - IRQF_SHARED | IRQF_TIMER | IRQF_IRQPOLL,
8689 - "at91_tick", regmap_st);
8691 - panic(pr_fmt("Unable to setup IRQ\n"));
8693 sclk = of_clk_get(node, 0);
8695 panic(pr_fmt("Unable to get slow clock\n"));
8696 diff --git a/drivers/cpufreq/Kconfig.x86 b/drivers/cpufreq/Kconfig.x86
8697 index c59bdcb83217..8f23161d80be 100644
8698 --- a/drivers/cpufreq/Kconfig.x86
8699 +++ b/drivers/cpufreq/Kconfig.x86
8700 @@ -123,7 +123,7 @@ config X86_POWERNOW_K7_ACPI
8702 config X86_POWERNOW_K8
8703 tristate "AMD Opteron/Athlon64 PowerNow!"
8704 - depends on ACPI && ACPI_PROCESSOR && X86_ACPI_CPUFREQ
8705 + depends on ACPI && ACPI_PROCESSOR && X86_ACPI_CPUFREQ && !PREEMPT_RT_BASE
8707 This adds the CPUFreq driver for K8/early Opteron/Athlon64 processors.
8708 Support for K10 and newer processors is now in acpi-cpufreq.
8709 diff --git a/drivers/cpuidle/coupled.c b/drivers/cpuidle/coupled.c
8710 index 344058f8501a..d5657d50ac40 100644
8711 --- a/drivers/cpuidle/coupled.c
8712 +++ b/drivers/cpuidle/coupled.c
8713 @@ -119,7 +119,6 @@ struct cpuidle_coupled {
8715 #define CPUIDLE_COUPLED_NOT_IDLE (-1)
8717 -static DEFINE_MUTEX(cpuidle_coupled_lock);
8718 static DEFINE_PER_CPU(struct call_single_data, cpuidle_coupled_poke_cb);
8721 diff --git a/drivers/gpu/drm/i915/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
8722 index 6ed7d63a0688..9da7482ad256 100644
8723 --- a/drivers/gpu/drm/i915/i915_gem_execbuffer.c
8724 +++ b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
8725 @@ -1264,7 +1264,9 @@ i915_gem_ringbuffer_submission(struct i915_execbuffer_params *params,
8729 +#ifndef CONFIG_PREEMPT_RT_BASE
8730 trace_i915_gem_ring_dispatch(params->request, params->dispatch_flags);
8733 i915_gem_execbuffer_move_to_active(vmas, params->request);
8734 i915_gem_execbuffer_retire_commands(params);
8735 diff --git a/drivers/gpu/drm/i915/i915_gem_shrinker.c b/drivers/gpu/drm/i915/i915_gem_shrinker.c
8736 index c0a96f1ee18e..deb1e207fa3c 100644
8737 --- a/drivers/gpu/drm/i915/i915_gem_shrinker.c
8738 +++ b/drivers/gpu/drm/i915/i915_gem_shrinker.c
8739 @@ -39,7 +39,7 @@ static bool mutex_is_locked_by(struct mutex *mutex, struct task_struct *task)
8740 if (!mutex_is_locked(mutex))
8743 -#if defined(CONFIG_DEBUG_MUTEXES) || defined(CONFIG_MUTEX_SPIN_ON_OWNER)
8744 +#if (defined(CONFIG_DEBUG_MUTEXES) || defined(CONFIG_MUTEX_SPIN_ON_OWNER)) && !defined(CONFIG_PREEMPT_RT_BASE)
8745 return mutex->owner == task;
8747 /* Since UP may be pre-empted, we cannot assume that we own the lock */
8748 diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c
8749 index b7b0a38acd67..148dcb1349d5 100644
8750 --- a/drivers/gpu/drm/i915/i915_irq.c
8751 +++ b/drivers/gpu/drm/i915/i915_irq.c
8752 @@ -812,6 +812,7 @@ static int i915_get_crtc_scanoutpos(struct drm_device *dev, unsigned int pipe,
8753 spin_lock_irqsave(&dev_priv->uncore.lock, irqflags);
8755 /* preempt_disable_rt() should go right here in PREEMPT_RT patchset. */
8756 + preempt_disable_rt();
8758 /* Get optional system timestamp before query. */
8760 @@ -863,6 +864,7 @@ static int i915_get_crtc_scanoutpos(struct drm_device *dev, unsigned int pipe,
8761 *etime = ktime_get();
8763 /* preempt_enable_rt() should go right here in PREEMPT_RT patchset. */
8764 + preempt_enable_rt();
8766 spin_unlock_irqrestore(&dev_priv->uncore.lock, irqflags);
8768 diff --git a/drivers/gpu/drm/i915/intel_display.c b/drivers/gpu/drm/i915/intel_display.c
8769 index 4f5d07bb3511..8ecd5c016dba 100644
8770 --- a/drivers/gpu/drm/i915/intel_display.c
8771 +++ b/drivers/gpu/drm/i915/intel_display.c
8772 @@ -11400,7 +11400,7 @@ void intel_check_page_flip(struct drm_device *dev, int pipe)
8773 struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
8774 struct intel_unpin_work *work;
8776 - WARN_ON(!in_interrupt());
8777 + WARN_ON_NONRT(!in_interrupt());
8781 diff --git a/drivers/gpu/drm/i915/intel_sprite.c b/drivers/gpu/drm/i915/intel_sprite.c
8782 index 2cc6aa072f4c..b79d33f14868 100644
8783 --- a/drivers/gpu/drm/i915/intel_sprite.c
8784 +++ b/drivers/gpu/drm/i915/intel_sprite.c
8786 #include "intel_drv.h"
8787 #include <drm/i915_drm.h>
8788 #include "i915_drv.h"
8789 +#include <linux/locallock.h>
8792 format_is_yuv(uint32_t format)
8793 @@ -64,6 +65,8 @@ static int usecs_to_scanlines(const struct drm_display_mode *adjusted_mode,
8794 1000 * adjusted_mode->crtc_htotal);
8797 +static DEFINE_LOCAL_IRQ_LOCK(pipe_update_lock);
8800 * intel_pipe_update_start() - start update of a set of display registers
8801 * @crtc: the crtc of which the registers are going to be updated
8802 @@ -96,7 +99,7 @@ void intel_pipe_update_start(struct intel_crtc *crtc)
8803 min = vblank_start - usecs_to_scanlines(adjusted_mode, 100);
8804 max = vblank_start - 1;
8806 - local_irq_disable();
8807 + local_lock_irq(pipe_update_lock);
8809 if (min <= 0 || max <= 0)
8811 @@ -126,11 +129,11 @@ void intel_pipe_update_start(struct intel_crtc *crtc)
8815 - local_irq_enable();
8816 + local_unlock_irq(pipe_update_lock);
8818 timeout = schedule_timeout(timeout);
8820 - local_irq_disable();
8821 + local_lock_irq(pipe_update_lock);
8824 finish_wait(wq, &wait);
8825 @@ -164,7 +167,7 @@ void intel_pipe_update_end(struct intel_crtc *crtc)
8827 trace_i915_pipe_update_end(crtc, end_vbl_count, scanline_end);
8829 - local_irq_enable();
8830 + local_unlock_irq(pipe_update_lock);
8832 if (crtc->debug.start_vbl_count &&
8833 crtc->debug.start_vbl_count != end_vbl_count) {
8834 diff --git a/drivers/gpu/drm/radeon/radeon_display.c b/drivers/gpu/drm/radeon/radeon_display.c
8835 index 3645b223aa37..642854b2ed2c 100644
8836 --- a/drivers/gpu/drm/radeon/radeon_display.c
8837 +++ b/drivers/gpu/drm/radeon/radeon_display.c
8838 @@ -1862,6 +1862,7 @@ int radeon_get_crtc_scanoutpos(struct drm_device *dev, unsigned int pipe,
8839 struct radeon_device *rdev = dev->dev_private;
8841 /* preempt_disable_rt() should go right here in PREEMPT_RT patchset. */
8842 + preempt_disable_rt();
8844 /* Get optional system timestamp before query. */
8846 @@ -1954,6 +1955,7 @@ int radeon_get_crtc_scanoutpos(struct drm_device *dev, unsigned int pipe,
8847 *etime = ktime_get();
8849 /* preempt_enable_rt() should go right here in PREEMPT_RT patchset. */
8850 + preempt_enable_rt();
8852 /* Decode into vertical and horizontal scanout position. */
8853 *vpos = position & 0x1fff;
8854 diff --git a/drivers/hv/vmbus_drv.c b/drivers/hv/vmbus_drv.c
8855 index 802dcb409030..d6d9427860d8 100644
8856 --- a/drivers/hv/vmbus_drv.c
8857 +++ b/drivers/hv/vmbus_drv.c
8858 @@ -820,7 +820,7 @@ static void vmbus_isr(void)
8859 tasklet_schedule(&msg_dpc);
8862 - add_interrupt_randomness(HYPERVISOR_CALLBACK_VECTOR, 0);
8863 + add_interrupt_randomness(HYPERVISOR_CALLBACK_VECTOR, 0, 0);
8867 diff --git a/drivers/i2c/busses/i2c-omap.c b/drivers/i2c/busses/i2c-omap.c
8868 index 08d26ba61ed3..46b89dd42b10 100644
8869 --- a/drivers/i2c/busses/i2c-omap.c
8870 +++ b/drivers/i2c/busses/i2c-omap.c
8871 @@ -995,15 +995,12 @@ omap_i2c_isr(int irq, void *dev_id)
8875 - spin_lock(&omap->lock);
8876 - mask = omap_i2c_read_reg(omap, OMAP_I2C_IE_REG);
8877 stat = omap_i2c_read_reg(omap, OMAP_I2C_STAT_REG);
8878 + mask = omap_i2c_read_reg(omap, OMAP_I2C_IE_REG);
8881 ret = IRQ_WAKE_THREAD;
8883 - spin_unlock(&omap->lock);
8888 diff --git a/drivers/ide/alim15x3.c b/drivers/ide/alim15x3.c
8889 index 36f76e28a0bf..394f142f90c7 100644
8890 --- a/drivers/ide/alim15x3.c
8891 +++ b/drivers/ide/alim15x3.c
8892 @@ -234,7 +234,7 @@ static int init_chipset_ali15x3(struct pci_dev *dev)
8894 isa_dev = pci_get_device(PCI_VENDOR_ID_AL, PCI_DEVICE_ID_AL_M1533, NULL);
8896 - local_irq_save(flags);
8897 + local_irq_save_nort(flags);
8899 if (m5229_revision < 0xC2) {
8901 @@ -325,7 +325,7 @@ out:
8904 pci_dev_put(isa_dev);
8905 - local_irq_restore(flags);
8906 + local_irq_restore_nort(flags);
8910 diff --git a/drivers/ide/hpt366.c b/drivers/ide/hpt366.c
8911 index 696b6c1ec940..0d0a96629b73 100644
8912 --- a/drivers/ide/hpt366.c
8913 +++ b/drivers/ide/hpt366.c
8914 @@ -1241,7 +1241,7 @@ static int init_dma_hpt366(ide_hwif_t *hwif,
8916 dma_old = inb(base + 2);
8918 - local_irq_save(flags);
8919 + local_irq_save_nort(flags);
8922 pci_read_config_byte(dev, hwif->channel ? 0x4b : 0x43, &masterdma);
8923 @@ -1252,7 +1252,7 @@ static int init_dma_hpt366(ide_hwif_t *hwif,
8924 if (dma_new != dma_old)
8925 outb(dma_new, base + 2);
8927 - local_irq_restore(flags);
8928 + local_irq_restore_nort(flags);
8930 printk(KERN_INFO " %s: BM-DMA at 0x%04lx-0x%04lx\n",
8931 hwif->name, base, base + 7);
8932 diff --git a/drivers/ide/ide-io-std.c b/drivers/ide/ide-io-std.c
8933 index 19763977568c..4169433faab5 100644
8934 --- a/drivers/ide/ide-io-std.c
8935 +++ b/drivers/ide/ide-io-std.c
8936 @@ -175,7 +175,7 @@ void ide_input_data(ide_drive_t *drive, struct ide_cmd *cmd, void *buf,
8937 unsigned long uninitialized_var(flags);
8939 if ((io_32bit & 2) && !mmio) {
8940 - local_irq_save(flags);
8941 + local_irq_save_nort(flags);
8942 ata_vlb_sync(io_ports->nsect_addr);
8945 @@ -186,7 +186,7 @@ void ide_input_data(ide_drive_t *drive, struct ide_cmd *cmd, void *buf,
8946 insl(data_addr, buf, words);
8948 if ((io_32bit & 2) && !mmio)
8949 - local_irq_restore(flags);
8950 + local_irq_restore_nort(flags);
8952 if (((len + 1) & 3) < 2)
8954 @@ -219,7 +219,7 @@ void ide_output_data(ide_drive_t *drive, struct ide_cmd *cmd, void *buf,
8955 unsigned long uninitialized_var(flags);
8957 if ((io_32bit & 2) && !mmio) {
8958 - local_irq_save(flags);
8959 + local_irq_save_nort(flags);
8960 ata_vlb_sync(io_ports->nsect_addr);
8963 @@ -230,7 +230,7 @@ void ide_output_data(ide_drive_t *drive, struct ide_cmd *cmd, void *buf,
8964 outsl(data_addr, buf, words);
8966 if ((io_32bit & 2) && !mmio)
8967 - local_irq_restore(flags);
8968 + local_irq_restore_nort(flags);
8970 if (((len + 1) & 3) < 2)
8972 diff --git a/drivers/ide/ide-io.c b/drivers/ide/ide-io.c
8973 index 669ea1e45795..e12e43e62245 100644
8974 --- a/drivers/ide/ide-io.c
8975 +++ b/drivers/ide/ide-io.c
8976 @@ -659,7 +659,7 @@ void ide_timer_expiry (unsigned long data)
8977 /* disable_irq_nosync ?? */
8978 disable_irq(hwif->irq);
8979 /* local CPU only, as if we were handling an interrupt */
8980 - local_irq_disable();
8981 + local_irq_disable_nort();
8982 if (hwif->polling) {
8983 startstop = handler(drive);
8984 } else if (drive_is_ready(drive)) {
8985 diff --git a/drivers/ide/ide-iops.c b/drivers/ide/ide-iops.c
8986 index 376f2dc410c5..f014dd1b73dc 100644
8987 --- a/drivers/ide/ide-iops.c
8988 +++ b/drivers/ide/ide-iops.c
8989 @@ -129,12 +129,12 @@ int __ide_wait_stat(ide_drive_t *drive, u8 good, u8 bad,
8990 if ((stat & ATA_BUSY) == 0)
8993 - local_irq_restore(flags);
8994 + local_irq_restore_nort(flags);
8999 - local_irq_restore(flags);
9000 + local_irq_restore_nort(flags);
9003 * Allow status to settle, then read it again.
9004 diff --git a/drivers/ide/ide-probe.c b/drivers/ide/ide-probe.c
9005 index 0b63facd1d87..4ceba37afc0c 100644
9006 --- a/drivers/ide/ide-probe.c
9007 +++ b/drivers/ide/ide-probe.c
9008 @@ -196,10 +196,10 @@ static void do_identify(ide_drive_t *drive, u8 cmd, u16 *id)
9011 /* local CPU only; some systems need this */
9012 - local_irq_save(flags);
9013 + local_irq_save_nort(flags);
9014 /* read 512 bytes of id info */
9015 hwif->tp_ops->input_data(drive, NULL, id, SECTOR_SIZE);
9016 - local_irq_restore(flags);
9017 + local_irq_restore_nort(flags);
9019 drive->dev_flags |= IDE_DFLAG_ID_READ;
9021 diff --git a/drivers/ide/ide-taskfile.c b/drivers/ide/ide-taskfile.c
9022 index a716693417a3..be0568c722d6 100644
9023 --- a/drivers/ide/ide-taskfile.c
9024 +++ b/drivers/ide/ide-taskfile.c
9025 @@ -250,7 +250,7 @@ void ide_pio_bytes(ide_drive_t *drive, struct ide_cmd *cmd,
9027 page_is_high = PageHighMem(page);
9029 - local_irq_save(flags);
9030 + local_irq_save_nort(flags);
9032 buf = kmap_atomic(page) + offset;
9034 @@ -271,7 +271,7 @@ void ide_pio_bytes(ide_drive_t *drive, struct ide_cmd *cmd,
9038 - local_irq_restore(flags);
9039 + local_irq_restore_nort(flags);
9043 @@ -414,7 +414,7 @@ static ide_startstop_t pre_task_out_intr(ide_drive_t *drive,
9046 if ((drive->dev_flags & IDE_DFLAG_UNMASK) == 0)
9047 - local_irq_disable();
9048 + local_irq_disable_nort();
9050 ide_set_handler(drive, &task_pio_intr, WAIT_WORSTCASE);
9052 diff --git a/drivers/infiniband/ulp/ipoib/ipoib_multicast.c b/drivers/infiniband/ulp/ipoib/ipoib_multicast.c
9053 index 5580ab0b5781..a123d0439c4c 100644
9054 --- a/drivers/infiniband/ulp/ipoib/ipoib_multicast.c
9055 +++ b/drivers/infiniband/ulp/ipoib/ipoib_multicast.c
9056 @@ -862,7 +862,7 @@ void ipoib_mcast_restart_task(struct work_struct *work)
9058 ipoib_dbg_mcast(priv, "restarting multicast task\n");
9060 - local_irq_save(flags);
9061 + local_irq_save_nort(flags);
9062 netif_addr_lock(dev);
9063 spin_lock(&priv->lock);
9065 @@ -944,7 +944,7 @@ void ipoib_mcast_restart_task(struct work_struct *work)
9067 spin_unlock(&priv->lock);
9068 netif_addr_unlock(dev);
9069 - local_irq_restore(flags);
9070 + local_irq_restore_nort(flags);
9073 * make sure the in-flight joins have finished before we attempt
9074 diff --git a/drivers/input/gameport/gameport.c b/drivers/input/gameport/gameport.c
9075 index 4a2a9e370be7..e970d9afd179 100644
9076 --- a/drivers/input/gameport/gameport.c
9077 +++ b/drivers/input/gameport/gameport.c
9078 @@ -91,13 +91,13 @@ static int gameport_measure_speed(struct gameport *gameport)
9081 for (i = 0; i < 50; i++) {
9082 - local_irq_save(flags);
9083 + local_irq_save_nort(flags);
9084 t1 = ktime_get_ns();
9085 for (t = 0; t < 50; t++)
9086 gameport_read(gameport);
9087 t2 = ktime_get_ns();
9088 t3 = ktime_get_ns();
9089 - local_irq_restore(flags);
9090 + local_irq_restore_nort(flags);
9092 t = (t2 - t1) - (t3 - t2);
9094 @@ -124,12 +124,12 @@ static int old_gameport_measure_speed(struct gameport *gameport)
9097 for(i = 0; i < 50; i++) {
9098 - local_irq_save(flags);
9099 + local_irq_save_nort(flags);
9101 for (t = 0; t < 50; t++) gameport_read(gameport);
9104 - local_irq_restore(flags);
9105 + local_irq_restore_nort(flags);
9107 if ((t = DELTA(t2,t1) - DELTA(t3,t2)) < tx) tx = t;
9109 @@ -148,11 +148,11 @@ static int old_gameport_measure_speed(struct gameport *gameport)
9112 for(i = 0; i < 50; i++) {
9113 - local_irq_save(flags);
9114 + local_irq_save_nort(flags);
9116 for (t = 0; t < 50; t++) gameport_read(gameport);
9118 - local_irq_restore(flags);
9119 + local_irq_restore_nort(flags);
9121 if (t2 - t1 < tx) tx = t2 - t1;
9123 diff --git a/drivers/iommu/amd_iommu.c b/drivers/iommu/amd_iommu.c
9124 index a0ef57483ebb..b08732bae11a 100644
9125 --- a/drivers/iommu/amd_iommu.c
9126 +++ b/drivers/iommu/amd_iommu.c
9127 @@ -2022,10 +2022,10 @@ static int __attach_device(struct iommu_dev_data *dev_data,
9131 - * Must be called with IRQs disabled. Warn here to detect early
9133 + * Must be called with IRQs disabled on a non RT kernel. Warn here to
9134 + * detect early when its not.
9136 - WARN_ON(!irqs_disabled());
9137 + WARN_ON_NONRT(!irqs_disabled());
9140 spin_lock(&domain->lock);
9141 @@ -2188,10 +2188,10 @@ static void __detach_device(struct iommu_dev_data *dev_data)
9142 struct protection_domain *domain;
9145 - * Must be called with IRQs disabled. Warn here to detect early
9147 + * Must be called with IRQs disabled on a non RT kernel. Warn here to
9148 + * detect early when its not.
9150 - WARN_ON(!irqs_disabled());
9151 + WARN_ON_NONRT(!irqs_disabled());
9153 if (WARN_ON(!dev_data->domain))
9155 diff --git a/drivers/leds/trigger/Kconfig b/drivers/leds/trigger/Kconfig
9156 index 5bda6a9b56bb..d6286584c807 100644
9157 --- a/drivers/leds/trigger/Kconfig
9158 +++ b/drivers/leds/trigger/Kconfig
9159 @@ -61,7 +61,7 @@ config LEDS_TRIGGER_BACKLIGHT
9161 config LEDS_TRIGGER_CPU
9162 bool "LED CPU Trigger"
9163 - depends on LEDS_TRIGGERS
9164 + depends on LEDS_TRIGGERS && !PREEMPT_RT_BASE
9166 This allows LEDs to be controlled by active CPUs. This shows
9167 the active CPUs across an array of LEDs so you can see which
9168 diff --git a/drivers/md/bcache/Kconfig b/drivers/md/bcache/Kconfig
9169 index 4d200883c505..98b64ed5cb81 100644
9170 --- a/drivers/md/bcache/Kconfig
9171 +++ b/drivers/md/bcache/Kconfig
9175 tristate "Block device as cache"
9176 + depends on !PREEMPT_RT_FULL
9178 Allows a block device to be used as cache for other devices; uses
9179 a btree for indexing and the layout is optimized for SSDs.
9180 diff --git a/drivers/md/dm.c b/drivers/md/dm.c
9181 index 320eb3c4bb6b..aba2d71149e3 100644
9182 --- a/drivers/md/dm.c
9183 +++ b/drivers/md/dm.c
9184 @@ -2185,7 +2185,7 @@ static void dm_request_fn(struct request_queue *q)
9185 /* Establish tio->ti before queuing work (map_tio_request) */
9187 queue_kthread_work(&md->kworker, &tio->work);
9188 - BUG_ON(!irqs_disabled());
9189 + BUG_ON_NONRT(!irqs_disabled());
9193 diff --git a/drivers/md/raid5.c b/drivers/md/raid5.c
9194 index 4384b46cee1a..6415f094a4eb 100644
9195 --- a/drivers/md/raid5.c
9196 +++ b/drivers/md/raid5.c
9197 @@ -1920,8 +1920,9 @@ static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request)
9198 struct raid5_percpu *percpu;
9202 + cpu = get_cpu_light();
9203 percpu = per_cpu_ptr(conf->percpu, cpu);
9204 + spin_lock(&percpu->lock);
9205 if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) {
9206 ops_run_biofill(sh);
9208 @@ -1977,7 +1978,8 @@ static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request)
9209 if (test_and_clear_bit(R5_Overlap, &dev->flags))
9210 wake_up(&sh->raid_conf->wait_for_overlap);
9213 + spin_unlock(&percpu->lock);
9217 static struct stripe_head *alloc_stripe(struct kmem_cache *sc, gfp_t gfp)
9218 @@ -6416,6 +6418,7 @@ static int raid5_alloc_percpu(struct r5conf *conf)
9222 + spin_lock_init(&per_cpu_ptr(conf->percpu, cpu)->lock);
9226 diff --git a/drivers/md/raid5.h b/drivers/md/raid5.h
9227 index 517d4b68a1be..efe91887ecd7 100644
9228 --- a/drivers/md/raid5.h
9229 +++ b/drivers/md/raid5.h
9230 @@ -504,6 +504,7 @@ struct r5conf {
9231 int recovery_disabled;
9232 /* per cpu variables */
9233 struct raid5_percpu {
9234 + spinlock_t lock; /* Protection for -RT */
9235 struct page *spare_page; /* Used when checking P/Q in raid6 */
9236 struct flex_array *scribble; /* space for constructing buffer
9237 * lists and performing address
9238 diff --git a/drivers/media/platform/vsp1/vsp1_video.c b/drivers/media/platform/vsp1/vsp1_video.c
9239 index 5ce88e1f5d71..b4f8cd74ecb8 100644
9240 --- a/drivers/media/platform/vsp1/vsp1_video.c
9241 +++ b/drivers/media/platform/vsp1/vsp1_video.c
9242 @@ -520,7 +520,7 @@ static bool vsp1_pipeline_stopped(struct vsp1_pipeline *pipe)
9245 spin_lock_irqsave(&pipe->irqlock, flags);
9246 - stopped = pipe->state == VSP1_PIPELINE_STOPPED,
9247 + stopped = pipe->state == VSP1_PIPELINE_STOPPED;
9248 spin_unlock_irqrestore(&pipe->irqlock, flags);
9251 diff --git a/drivers/misc/Kconfig b/drivers/misc/Kconfig
9252 index 4bf7d50b1bc7..6f7e99ad6e29 100644
9253 --- a/drivers/misc/Kconfig
9254 +++ b/drivers/misc/Kconfig
9255 @@ -54,6 +54,7 @@ config AD525X_DPOT_SPI
9257 bool "Atmel AT32/AT91 Timer/Counter Library"
9258 depends on (AVR32 || ARCH_AT91)
9259 + default y if PREEMPT_RT_FULL
9261 Select this if you want a library to allocate the Timer/Counter
9262 blocks found on many Atmel processors. This facilitates using
9263 @@ -69,8 +70,7 @@ config ATMEL_TCB_CLKSRC
9264 are combined to make a single 32-bit timer.
9266 When GENERIC_CLOCKEVENTS is defined, the third timer channel
9267 - may be used as a clock event device supporting oneshot mode
9268 - (delays of up to two seconds) based on the 32 KiHz clock.
9269 + may be used as a clock event device supporting oneshot mode.
9271 config ATMEL_TCB_CLKSRC_BLOCK
9273 @@ -84,6 +84,15 @@ config ATMEL_TCB_CLKSRC_BLOCK
9274 TC can be used for other purposes, such as PWM generation and
9277 +config ATMEL_TCB_CLKSRC_USE_SLOW_CLOCK
9278 + bool "TC Block use 32 KiHz clock"
9279 + depends on ATMEL_TCB_CLKSRC
9280 + default y if !PREEMPT_RT_FULL
9282 + Select this to use 32 KiHz base clock rate as TC block clock
9283 + source for clock events.
9287 tristate "Dummy IRQ handler"
9289 @@ -113,6 +122,35 @@ config IBM_ASM
9290 for information on the specific driver level and support statement
9291 for your IBM server.
9293 +config HWLAT_DETECTOR
9294 + tristate "Testing module to detect hardware-induced latencies"
9295 + depends on DEBUG_FS
9296 + depends on RING_BUFFER
9299 + A simple hardware latency detector. Use this module to detect
9300 + large latencies introduced by the behavior of the underlying
9301 + system firmware external to Linux. We do this using periodic
9302 + use of stop_machine to grab all available CPUs and measure
9303 + for unexplainable gaps in the CPU timestamp counter(s). By
9304 + default, the module is not enabled until the "enable" file
9305 + within the "hwlat_detector" debugfs directory is toggled.
9307 + This module is often used to detect SMI (System Management
9308 + Interrupts) on x86 systems, though is not x86 specific. To
9309 + this end, we default to using a sample window of 1 second,
9310 + during which we will sample for 0.5 seconds. If an SMI or
9311 + similar event occurs during that time, it is recorded
9312 + into an 8K samples global ring buffer until retreived.
9314 + WARNING: This software should never be enabled (it can be built
9315 + but should not be turned on after it is loaded) in a production
9316 + environment where high latencies are a concern since the
9317 + sampling mechanism actually introduces latencies for
9318 + regular tasks while the CPU(s) are being held.
9323 tristate "Sensable PHANToM (PCI)"
9325 diff --git a/drivers/misc/Makefile b/drivers/misc/Makefile
9326 index 537d7f3b78da..ec4aecba0656 100644
9327 --- a/drivers/misc/Makefile
9328 +++ b/drivers/misc/Makefile
9329 @@ -39,6 +39,7 @@ obj-$(CONFIG_C2PORT) += c2port/
9330 obj-$(CONFIG_HMC6352) += hmc6352.o
9333 +obj-$(CONFIG_HWLAT_DETECTOR) += hwlat_detector.o
9334 obj-$(CONFIG_SPEAR13XX_PCIE_GADGET) += spear13xx_pcie_gadget.o
9335 obj-$(CONFIG_VMWARE_BALLOON) += vmw_balloon.o
9336 obj-$(CONFIG_ARM_CHARLCD) += arm-charlcd.o
9337 diff --git a/drivers/misc/hwlat_detector.c b/drivers/misc/hwlat_detector.c
9338 new file mode 100644
9339 index 000000000000..52f5ad5fd9c0
9341 +++ b/drivers/misc/hwlat_detector.c
9344 + * hwlat_detector.c - A simple Hardware Latency detector.
9346 + * Use this module to detect large system latencies induced by the behavior of
9347 + * certain underlying system hardware or firmware, independent of Linux itself.
9348 + * The code was developed originally to detect the presence of SMIs on Intel
9349 + * and AMD systems, although there is no dependency upon x86 herein.
9351 + * The classical example usage of this module is in detecting the presence of
9352 + * SMIs or System Management Interrupts on Intel and AMD systems. An SMI is a
9353 + * somewhat special form of hardware interrupt spawned from earlier CPU debug
9354 + * modes in which the (BIOS/EFI/etc.) firmware arranges for the South Bridge
9355 + * LPC (or other device) to generate a special interrupt under certain
9356 + * circumstances, for example, upon expiration of a special SMI timer device,
9357 + * due to certain external thermal readings, on certain I/O address accesses,
9358 + * and other situations. An SMI hits a special CPU pin, triggers a special
9359 + * SMI mode (complete with special memory map), and the OS is unaware.
9361 + * Although certain hardware-inducing latencies are necessary (for example,
9362 + * a modern system often requires an SMI handler for correct thermal control
9363 + * and remote management) they can wreak havoc upon any OS-level performance
9364 + * guarantees toward low-latency, especially when the OS is not even made
9365 + * aware of the presence of these interrupts. For this reason, we need a
9366 + * somewhat brute force mechanism to detect these interrupts. In this case,
9367 + * we do it by hogging all of the CPU(s) for configurable timer intervals,
9368 + * sampling the built-in CPU timer, looking for discontiguous readings.
9370 + * WARNING: This implementation necessarily introduces latencies. Therefore,
9371 + * you should NEVER use this module in a production environment
9372 + * requiring any kind of low-latency performance guarantee(s).
9374 + * Copyright (C) 2008-2009 Jon Masters, Red Hat, Inc. <jcm@redhat.com>
9376 + * Includes useful feedback from Clark Williams <clark@redhat.com>
9378 + * This file is licensed under the terms of the GNU General Public
9379 + * License version 2. This program is licensed "as is" without any
9380 + * warranty of any kind, whether express or implied.
9383 +#include <linux/module.h>
9384 +#include <linux/init.h>
9385 +#include <linux/ring_buffer.h>
9386 +#include <linux/time.h>
9387 +#include <linux/hrtimer.h>
9388 +#include <linux/kthread.h>
9389 +#include <linux/debugfs.h>
9390 +#include <linux/seq_file.h>
9391 +#include <linux/uaccess.h>
9392 +#include <linux/version.h>
9393 +#include <linux/delay.h>
9394 +#include <linux/slab.h>
9395 +#include <linux/trace_clock.h>
9397 +#define BUF_SIZE_DEFAULT 262144UL /* 8K*(sizeof(entry)) */
9398 +#define BUF_FLAGS (RB_FL_OVERWRITE) /* no block on full */
9399 +#define U64STR_SIZE 22 /* 20 digits max */
9401 +#define VERSION "1.0.0"
9402 +#define BANNER "hwlat_detector: "
9403 +#define DRVNAME "hwlat_detector"
9404 +#define DEFAULT_SAMPLE_WINDOW 1000000 /* 1s */
9405 +#define DEFAULT_SAMPLE_WIDTH 500000 /* 0.5s */
9406 +#define DEFAULT_LAT_THRESHOLD 10 /* 10us */
9408 +/* Module metadata */
9410 +MODULE_LICENSE("GPL");
9411 +MODULE_AUTHOR("Jon Masters <jcm@redhat.com>");
9412 +MODULE_DESCRIPTION("A simple hardware latency detector");
9413 +MODULE_VERSION(VERSION);
9415 +/* Module parameters */
9418 +static int enabled;
9419 +static int threshold;
9421 +module_param(debug, int, 0); /* enable debug */
9422 +module_param(enabled, int, 0); /* enable detector */
9423 +module_param(threshold, int, 0); /* latency threshold */
9425 +/* Buffering and sampling */
9427 +static struct ring_buffer *ring_buffer; /* sample buffer */
9428 +static DEFINE_MUTEX(ring_buffer_mutex); /* lock changes */
9429 +static unsigned long buf_size = BUF_SIZE_DEFAULT;
9430 +static struct task_struct *kthread; /* sampling thread */
9432 +/* DebugFS filesystem entries */
9434 +static struct dentry *debug_dir; /* debugfs directory */
9435 +static struct dentry *debug_max; /* maximum TSC delta */
9436 +static struct dentry *debug_count; /* total detect count */
9437 +static struct dentry *debug_sample_width; /* sample width us */
9438 +static struct dentry *debug_sample_window; /* sample window us */
9439 +static struct dentry *debug_sample; /* raw samples us */
9440 +static struct dentry *debug_threshold; /* threshold us */
9441 +static struct dentry *debug_enable; /* enable/disable */
9443 +/* Individual samples and global state */
9445 +struct sample; /* latency sample */
9446 +struct data; /* Global state */
9448 +/* Sampling functions */
9449 +static int __buffer_add_sample(struct sample *sample);
9450 +static struct sample *buffer_get_sample(struct sample *sample);
9452 +/* Threading and state */
9453 +static int kthread_fn(void *unused);
9454 +static int start_kthread(void);
9455 +static int stop_kthread(void);
9456 +static void __reset_stats(void);
9457 +static int init_stats(void);
9459 +/* Debugfs interface */
9460 +static ssize_t simple_data_read(struct file *filp, char __user *ubuf,
9461 + size_t cnt, loff_t *ppos, const u64 *entry);
9462 +static ssize_t simple_data_write(struct file *filp, const char __user *ubuf,
9463 + size_t cnt, loff_t *ppos, u64 *entry);
9464 +static int debug_sample_fopen(struct inode *inode, struct file *filp);
9465 +static ssize_t debug_sample_fread(struct file *filp, char __user *ubuf,
9466 + size_t cnt, loff_t *ppos);
9467 +static int debug_sample_release(struct inode *inode, struct file *filp);
9468 +static int debug_enable_fopen(struct inode *inode, struct file *filp);
9469 +static ssize_t debug_enable_fread(struct file *filp, char __user *ubuf,
9470 + size_t cnt, loff_t *ppos);
9471 +static ssize_t debug_enable_fwrite(struct file *file,
9472 + const char __user *user_buffer,
9473 + size_t user_size, loff_t *offset);
9475 +/* Initialization functions */
9476 +static int init_debugfs(void);
9477 +static void free_debugfs(void);
9478 +static int detector_init(void);
9479 +static void detector_exit(void);
9481 +/* Individual latency samples are stored here when detected and packed into
9482 + * the ring_buffer circular buffer, where they are overwritten when
9483 + * more than buf_size/sizeof(sample) samples are received. */
9485 + u64 seqnum; /* unique sequence */
9486 + u64 duration; /* ktime delta */
9487 + u64 outer_duration; /* ktime delta (outer loop) */
9488 + struct timespec timestamp; /* wall time */
9489 + unsigned long lost;
9492 +/* keep the global state somewhere. */
9493 +static struct data {
9495 + struct mutex lock; /* protect changes */
9497 + u64 count; /* total since reset */
9498 + u64 max_sample; /* max hardware latency */
9499 + u64 threshold; /* sample threshold level */
9501 + u64 sample_window; /* total sampling window (on+off) */
9502 + u64 sample_width; /* active sampling portion of window */
9504 + atomic_t sample_open; /* whether the sample file is open */
9506 + wait_queue_head_t wq; /* waitqeue for new sample values */
9511 + * __buffer_add_sample - add a new latency sample recording to the ring buffer
9512 + * @sample: The new latency sample value
9514 + * This receives a new latency sample and records it in a global ring buffer.
9515 + * No additional locking is used in this case.
9517 +static int __buffer_add_sample(struct sample *sample)
9519 + return ring_buffer_write(ring_buffer,
9520 + sizeof(struct sample), sample);
9524 + * buffer_get_sample - remove a hardware latency sample from the ring buffer
9525 + * @sample: Pre-allocated storage for the sample
9527 + * This retrieves a hardware latency sample from the global circular buffer
9529 +static struct sample *buffer_get_sample(struct sample *sample)
9531 + struct ring_buffer_event *e = NULL;
9532 + struct sample *s = NULL;
9533 + unsigned int cpu = 0;
9538 + mutex_lock(&ring_buffer_mutex);
9539 + for_each_online_cpu(cpu) {
9540 + e = ring_buffer_consume(ring_buffer, cpu, NULL, &sample->lost);
9546 + s = ring_buffer_event_data(e);
9547 + memcpy(sample, s, sizeof(struct sample));
9550 + mutex_unlock(&ring_buffer_mutex);
9555 +#ifndef CONFIG_TRACING
9556 +#define time_type ktime_t
9557 +#define time_get() ktime_get()
9558 +#define time_to_us(x) ktime_to_us(x)
9559 +#define time_sub(a, b) ktime_sub(a, b)
9560 +#define init_time(a, b) (a).tv64 = b
9561 +#define time_u64(a) ((a).tv64)
9563 +#define time_type u64
9564 +#define time_get() trace_clock_local()
9565 +#define time_to_us(x) div_u64(x, 1000)
9566 +#define time_sub(a, b) ((a) - (b))
9567 +#define init_time(a, b) (a = b)
9568 +#define time_u64(a) a
9571 + * get_sample - sample the CPU TSC and look for likely hardware latencies
9573 + * Used to repeatedly capture the CPU TSC (or similar), looking for potential
9574 + * hardware-induced latency. Called with interrupts disabled and with
9577 +static int get_sample(void)
9579 + time_type start, t1, t2, last_t2;
9580 + s64 diff, total = 0;
9582 + u64 outer_sample = 0;
9585 + init_time(last_t2, 0);
9586 + start = time_get(); /* start timestamp */
9590 + t1 = time_get(); /* we'll look for a discontinuity */
9593 + if (time_u64(last_t2)) {
9594 + /* Check the delta from outer loop (t2 to next t1) */
9595 + diff = time_to_us(time_sub(t1, last_t2));
9596 + /* This shouldn't happen */
9598 + pr_err(BANNER "time running backwards\n");
9601 + if (diff > outer_sample)
9602 + outer_sample = diff;
9606 + total = time_to_us(time_sub(t2, start)); /* sample width */
9608 + /* This checks the inner loop (t1 to t2) */
9609 + diff = time_to_us(time_sub(t2, t1)); /* current diff */
9611 + /* This shouldn't happen */
9613 + pr_err(BANNER "time running backwards\n");
9617 + if (diff > sample)
9618 + sample = diff; /* only want highest value */
9620 + } while (total <= data.sample_width);
9624 + /* If we exceed the threshold value, we have found a hardware latency */
9625 + if (sample > data.threshold || outer_sample > data.threshold) {
9631 + s.seqnum = data.count;
9632 + s.duration = sample;
9633 + s.outer_duration = outer_sample;
9634 + s.timestamp = CURRENT_TIME;
9635 + __buffer_add_sample(&s);
9637 + /* Keep a running maximum ever recorded hardware latency */
9638 + if (sample > data.max_sample)
9639 + data.max_sample = sample;
9647 + * kthread_fn - The CPU time sampling/hardware latency detection kernel thread
9648 + * @unused: A required part of the kthread API.
9650 + * Used to periodically sample the CPU TSC via a call to get_sample. We
9651 + * disable interrupts, which does (intentionally) introduce latency since we
9652 + * need to ensure nothing else might be running (and thus pre-empting).
9653 + * Obviously this should never be used in production environments.
9655 + * Currently this runs on which ever CPU it was scheduled on, but most
9656 + * real-worald hardware latency situations occur across several CPUs,
9657 + * but we might later generalize this if we find there are any actualy
9658 + * systems with alternate SMI delivery or other hardware latencies.
9660 +static int kthread_fn(void *unused)
9665 + while (!kthread_should_stop()) {
9667 + mutex_lock(&data.lock);
9669 + local_irq_disable();
9670 + ret = get_sample();
9671 + local_irq_enable();
9674 + wake_up(&data.wq); /* wake up reader(s) */
9676 + interval = data.sample_window - data.sample_width;
9677 + do_div(interval, USEC_PER_MSEC); /* modifies interval value */
9679 + mutex_unlock(&data.lock);
9681 + if (msleep_interruptible(interval))
9689 + * start_kthread - Kick off the hardware latency sampling/detector kthread
9691 + * This starts a kernel thread that will sit and sample the CPU timestamp
9692 + * counter (TSC or similar) and look for potential hardware latencies.
9694 +static int start_kthread(void)
9696 + kthread = kthread_run(kthread_fn, NULL,
9698 + if (IS_ERR(kthread)) {
9699 + pr_err(BANNER "could not start sampling thread\n");
9708 + * stop_kthread - Inform the hardware latency samping/detector kthread to stop
9710 + * This kicks the running hardware latency sampling/detector kernel thread and
9711 + * tells it to stop sampling now. Use this on unload and at system shutdown.
9713 +static int stop_kthread(void)
9717 + ret = kthread_stop(kthread);
9723 + * __reset_stats - Reset statistics for the hardware latency detector
9725 + * We use data to store various statistics and global state. We call this
9726 + * function in order to reset those when "enable" is toggled on or off, and
9727 + * also at initialization. Should be called with data.lock held.
9729 +static void __reset_stats(void)
9732 + data.max_sample = 0;
9733 + ring_buffer_reset(ring_buffer); /* flush out old sample entries */
9737 + * init_stats - Setup global state statistics for the hardware latency detector
9739 + * We use data to store various statistics and global state. We also use
9740 + * a global ring buffer (ring_buffer) to keep raw samples of detected hardware
9741 + * induced system latencies. This function initializes these structures and
9742 + * allocates the global ring buffer also.
9744 +static int init_stats(void)
9746 + int ret = -ENOMEM;
9748 + mutex_init(&data.lock);
9749 + init_waitqueue_head(&data.wq);
9750 + atomic_set(&data.sample_open, 0);
9752 + ring_buffer = ring_buffer_alloc(buf_size, BUF_FLAGS);
9754 + if (WARN(!ring_buffer, KERN_ERR BANNER
9755 + "failed to allocate ring buffer!\n"))
9759 + data.threshold = threshold ?: DEFAULT_LAT_THRESHOLD; /* threshold us */
9760 + data.sample_window = DEFAULT_SAMPLE_WINDOW; /* window us */
9761 + data.sample_width = DEFAULT_SAMPLE_WIDTH; /* width us */
9771 + * simple_data_read - Wrapper read function for global state debugfs entries
9772 + * @filp: The active open file structure for the debugfs "file"
9773 + * @ubuf: The userspace provided buffer to read value into
9774 + * @cnt: The maximum number of bytes to read
9775 + * @ppos: The current "file" position
9776 + * @entry: The entry to read from
9778 + * This function provides a generic read implementation for the global state
9779 + * "data" structure debugfs filesystem entries. It would be nice to use
9780 + * simple_attr_read directly, but we need to make sure that the data.lock
9781 + * is held during the actual read.
9783 +static ssize_t simple_data_read(struct file *filp, char __user *ubuf,
9784 + size_t cnt, loff_t *ppos, const u64 *entry)
9786 + char buf[U64STR_SIZE];
9790 + memset(buf, 0, sizeof(buf));
9795 + mutex_lock(&data.lock);
9797 + mutex_unlock(&data.lock);
9799 + len = snprintf(buf, sizeof(buf), "%llu\n", (unsigned long long)val);
9801 + return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
9806 + * simple_data_write - Wrapper write function for global state debugfs entries
9807 + * @filp: The active open file structure for the debugfs "file"
9808 + * @ubuf: The userspace provided buffer to write value from
9809 + * @cnt: The maximum number of bytes to write
9810 + * @ppos: The current "file" position
9811 + * @entry: The entry to write to
9813 + * This function provides a generic write implementation for the global state
9814 + * "data" structure debugfs filesystem entries. It would be nice to use
9815 + * simple_attr_write directly, but we need to make sure that the data.lock
9816 + * is held during the actual write.
9818 +static ssize_t simple_data_write(struct file *filp, const char __user *ubuf,
9819 + size_t cnt, loff_t *ppos, u64 *entry)
9821 + char buf[U64STR_SIZE];
9822 + int csize = min(cnt, sizeof(buf));
9826 + memset(buf, '\0', sizeof(buf));
9827 + if (copy_from_user(buf, ubuf, csize))
9830 + buf[U64STR_SIZE-1] = '\0'; /* just in case */
9831 + err = kstrtoull(buf, 10, &val);
9835 + mutex_lock(&data.lock);
9837 + mutex_unlock(&data.lock);
9843 + * debug_count_fopen - Open function for "count" debugfs entry
9844 + * @inode: The in-kernel inode representation of the debugfs "file"
9845 + * @filp: The active open file structure for the debugfs "file"
9847 + * This function provides an open implementation for the "count" debugfs
9848 + * interface to the hardware latency detector.
9850 +static int debug_count_fopen(struct inode *inode, struct file *filp)
9856 + * debug_count_fread - Read function for "count" debugfs entry
9857 + * @filp: The active open file structure for the debugfs "file"
9858 + * @ubuf: The userspace provided buffer to read value into
9859 + * @cnt: The maximum number of bytes to read
9860 + * @ppos: The current "file" position
9862 + * This function provides a read implementation for the "count" debugfs
9863 + * interface to the hardware latency detector. Can be used to read the
9864 + * number of latency readings exceeding the configured threshold since
9865 + * the detector was last reset (e.g. by writing a zero into "count").
9867 +static ssize_t debug_count_fread(struct file *filp, char __user *ubuf,
9868 + size_t cnt, loff_t *ppos)
9870 + return simple_data_read(filp, ubuf, cnt, ppos, &data.count);
9874 + * debug_count_fwrite - Write function for "count" debugfs entry
9875 + * @filp: The active open file structure for the debugfs "file"
9876 + * @ubuf: The user buffer that contains the value to write
9877 + * @cnt: The maximum number of bytes to write to "file"
9878 + * @ppos: The current position in the debugfs "file"
9880 + * This function provides a write implementation for the "count" debugfs
9881 + * interface to the hardware latency detector. Can be used to write a
9882 + * desired value, especially to zero the total count.
9884 +static ssize_t debug_count_fwrite(struct file *filp,
9885 + const char __user *ubuf,
9889 + return simple_data_write(filp, ubuf, cnt, ppos, &data.count);
9893 + * debug_enable_fopen - Dummy open function for "enable" debugfs interface
9894 + * @inode: The in-kernel inode representation of the debugfs "file"
9895 + * @filp: The active open file structure for the debugfs "file"
9897 + * This function provides an open implementation for the "enable" debugfs
9898 + * interface to the hardware latency detector.
9900 +static int debug_enable_fopen(struct inode *inode, struct file *filp)
9906 + * debug_enable_fread - Read function for "enable" debugfs interface
9907 + * @filp: The active open file structure for the debugfs "file"
9908 + * @ubuf: The userspace provided buffer to read value into
9909 + * @cnt: The maximum number of bytes to read
9910 + * @ppos: The current "file" position
9912 + * This function provides a read implementation for the "enable" debugfs
9913 + * interface to the hardware latency detector. Can be used to determine
9914 + * whether the detector is currently enabled ("0\n" or "1\n" returned).
9916 +static ssize_t debug_enable_fread(struct file *filp, char __user *ubuf,
9917 + size_t cnt, loff_t *ppos)
9921 + if ((cnt < sizeof(buf)) || (*ppos))
9924 + buf[0] = enabled ? '1' : '0';
9927 + if (copy_to_user(ubuf, buf, strlen(buf)))
9929 + return *ppos = strlen(buf);
9933 + * debug_enable_fwrite - Write function for "enable" debugfs interface
9934 + * @filp: The active open file structure for the debugfs "file"
9935 + * @ubuf: The user buffer that contains the value to write
9936 + * @cnt: The maximum number of bytes to write to "file"
9937 + * @ppos: The current position in the debugfs "file"
9939 + * This function provides a write implementation for the "enable" debugfs
9940 + * interface to the hardware latency detector. Can be used to enable or
9941 + * disable the detector, which will have the side-effect of possibly
9942 + * also resetting the global stats and kicking off the measuring
9943 + * kthread (on an enable) or the converse (upon a disable).
9945 +static ssize_t debug_enable_fwrite(struct file *filp,
9946 + const char __user *ubuf,
9951 + int csize = min(cnt, sizeof(buf));
9955 + memset(buf, '\0', sizeof(buf));
9956 + if (copy_from_user(buf, ubuf, csize))
9959 + buf[sizeof(buf)-1] = '\0'; /* just in case */
9960 + err = kstrtoul(buf, 10, &val);
9969 + if (start_kthread())
9975 + err = stop_kthread();
9977 + pr_err(BANNER "cannot stop kthread\n");
9980 + wake_up(&data.wq); /* reader(s) should return */
9987 + * debug_max_fopen - Open function for "max" debugfs entry
9988 + * @inode: The in-kernel inode representation of the debugfs "file"
9989 + * @filp: The active open file structure for the debugfs "file"
9991 + * This function provides an open implementation for the "max" debugfs
9992 + * interface to the hardware latency detector.
9994 +static int debug_max_fopen(struct inode *inode, struct file *filp)
10000 + * debug_max_fread - Read function for "max" debugfs entry
10001 + * @filp: The active open file structure for the debugfs "file"
10002 + * @ubuf: The userspace provided buffer to read value into
10003 + * @cnt: The maximum number of bytes to read
10004 + * @ppos: The current "file" position
10006 + * This function provides a read implementation for the "max" debugfs
10007 + * interface to the hardware latency detector. Can be used to determine
10008 + * the maximum latency value observed since it was last reset.
10010 +static ssize_t debug_max_fread(struct file *filp, char __user *ubuf,
10011 + size_t cnt, loff_t *ppos)
10013 + return simple_data_read(filp, ubuf, cnt, ppos, &data.max_sample);
10017 + * debug_max_fwrite - Write function for "max" debugfs entry
10018 + * @filp: The active open file structure for the debugfs "file"
10019 + * @ubuf: The user buffer that contains the value to write
10020 + * @cnt: The maximum number of bytes to write to "file"
10021 + * @ppos: The current position in the debugfs "file"
10023 + * This function provides a write implementation for the "max" debugfs
10024 + * interface to the hardware latency detector. Can be used to reset the
10025 + * maximum or set it to some other desired value - if, then, subsequent
10026 + * measurements exceed this value, the maximum will be updated.
10028 +static ssize_t debug_max_fwrite(struct file *filp,
10029 + const char __user *ubuf,
10033 + return simple_data_write(filp, ubuf, cnt, ppos, &data.max_sample);
10038 + * debug_sample_fopen - An open function for "sample" debugfs interface
10039 + * @inode: The in-kernel inode representation of this debugfs "file"
10040 + * @filp: The active open file structure for the debugfs "file"
10042 + * This function handles opening the "sample" file within the hardware
10043 + * latency detector debugfs directory interface. This file is used to read
10044 + * raw samples from the global ring_buffer and allows the user to see a
10045 + * running latency history. Can be opened blocking or non-blocking,
10046 + * affecting whether it behaves as a buffer read pipe, or does not.
10047 + * Implements simple locking to prevent multiple simultaneous use.
10049 +static int debug_sample_fopen(struct inode *inode, struct file *filp)
10051 + if (!atomic_add_unless(&data.sample_open, 1, 1))
10058 + * debug_sample_fread - A read function for "sample" debugfs interface
10059 + * @filp: The active open file structure for the debugfs "file"
10060 + * @ubuf: The user buffer that will contain the samples read
10061 + * @cnt: The maximum bytes to read from the debugfs "file"
10062 + * @ppos: The current position in the debugfs "file"
10064 + * This function handles reading from the "sample" file within the hardware
10065 + * latency detector debugfs directory interface. This file is used to read
10066 + * raw samples from the global ring_buffer and allows the user to see a
10067 + * running latency history. By default this will block pending a new
10068 + * value written into the sample buffer, unless there are already a
10069 + * number of value(s) waiting in the buffer, or the sample file was
10070 + * previously opened in a non-blocking mode of operation.
10072 +static ssize_t debug_sample_fread(struct file *filp, char __user *ubuf,
10073 + size_t cnt, loff_t *ppos)
10077 + struct sample *sample = NULL;
10082 + sample = kzalloc(sizeof(struct sample), GFP_KERNEL);
10086 + while (!buffer_get_sample(sample)) {
10088 + DEFINE_WAIT(wait);
10090 + if (filp->f_flags & O_NONBLOCK) {
10095 + prepare_to_wait(&data.wq, &wait, TASK_INTERRUPTIBLE);
10097 + finish_wait(&data.wq, &wait);
10099 + if (signal_pending(current)) {
10104 + if (!enabled) { /* enable was toggled */
10110 + len = snprintf(buf, sizeof(buf), "%010lu.%010lu\t%llu\t%llu\n",
10111 + sample->timestamp.tv_sec,
10112 + sample->timestamp.tv_nsec,
10113 + sample->duration,
10114 + sample->outer_duration);
10117 + /* handling partial reads is more trouble than it's worth */
10121 + if (copy_to_user(ubuf, buf, len))
10130 + * debug_sample_release - Release function for "sample" debugfs interface
10131 + * @inode: The in-kernel inode represenation of the debugfs "file"
10132 + * @filp: The active open file structure for the debugfs "file"
10134 + * This function completes the close of the debugfs interface "sample" file.
10135 + * Frees the sample_open "lock" so that other users may open the interface.
10137 +static int debug_sample_release(struct inode *inode, struct file *filp)
10139 + atomic_dec(&data.sample_open);
10145 + * debug_threshold_fopen - Open function for "threshold" debugfs entry
10146 + * @inode: The in-kernel inode representation of the debugfs "file"
10147 + * @filp: The active open file structure for the debugfs "file"
10149 + * This function provides an open implementation for the "threshold" debugfs
10150 + * interface to the hardware latency detector.
10152 +static int debug_threshold_fopen(struct inode *inode, struct file *filp)
10158 + * debug_threshold_fread - Read function for "threshold" debugfs entry
10159 + * @filp: The active open file structure for the debugfs "file"
10160 + * @ubuf: The userspace provided buffer to read value into
10161 + * @cnt: The maximum number of bytes to read
10162 + * @ppos: The current "file" position
10164 + * This function provides a read implementation for the "threshold" debugfs
10165 + * interface to the hardware latency detector. It can be used to determine
10166 + * the current threshold level at which a latency will be recorded in the
10167 + * global ring buffer, typically on the order of 10us.
10169 +static ssize_t debug_threshold_fread(struct file *filp, char __user *ubuf,
10170 + size_t cnt, loff_t *ppos)
10172 + return simple_data_read(filp, ubuf, cnt, ppos, &data.threshold);
10176 + * debug_threshold_fwrite - Write function for "threshold" debugfs entry
10177 + * @filp: The active open file structure for the debugfs "file"
10178 + * @ubuf: The user buffer that contains the value to write
10179 + * @cnt: The maximum number of bytes to write to "file"
10180 + * @ppos: The current position in the debugfs "file"
10182 + * This function provides a write implementation for the "threshold" debugfs
10183 + * interface to the hardware latency detector. It can be used to configure
10184 + * the threshold level at which any subsequently detected latencies will
10185 + * be recorded into the global ring buffer.
10187 +static ssize_t debug_threshold_fwrite(struct file *filp,
10188 + const char __user *ubuf,
10194 + ret = simple_data_write(filp, ubuf, cnt, ppos, &data.threshold);
10197 + wake_up_process(kthread);
10203 + * debug_width_fopen - Open function for "width" debugfs entry
10204 + * @inode: The in-kernel inode representation of the debugfs "file"
10205 + * @filp: The active open file structure for the debugfs "file"
10207 + * This function provides an open implementation for the "width" debugfs
10208 + * interface to the hardware latency detector.
10210 +static int debug_width_fopen(struct inode *inode, struct file *filp)
10216 + * debug_width_fread - Read function for "width" debugfs entry
10217 + * @filp: The active open file structure for the debugfs "file"
10218 + * @ubuf: The userspace provided buffer to read value into
10219 + * @cnt: The maximum number of bytes to read
10220 + * @ppos: The current "file" position
10222 + * This function provides a read implementation for the "width" debugfs
10223 + * interface to the hardware latency detector. It can be used to determine
10224 + * for how many us of the total window us we will actively sample for any
10225 + * hardware-induced latecy periods. Obviously, it is not possible to
10226 + * sample constantly and have the system respond to a sample reader, or,
10227 + * worse, without having the system appear to have gone out to lunch.
10229 +static ssize_t debug_width_fread(struct file *filp, char __user *ubuf,
10230 + size_t cnt, loff_t *ppos)
10232 + return simple_data_read(filp, ubuf, cnt, ppos, &data.sample_width);
10236 + * debug_width_fwrite - Write function for "width" debugfs entry
10237 + * @filp: The active open file structure for the debugfs "file"
10238 + * @ubuf: The user buffer that contains the value to write
10239 + * @cnt: The maximum number of bytes to write to "file"
10240 + * @ppos: The current position in the debugfs "file"
10242 + * This function provides a write implementation for the "width" debugfs
10243 + * interface to the hardware latency detector. It can be used to configure
10244 + * for how many us of the total window us we will actively sample for any
10245 + * hardware-induced latency periods. Obviously, it is not possible to
10246 + * sample constantly and have the system respond to a sample reader, or,
10247 + * worse, without having the system appear to have gone out to lunch. It
10248 + * is enforced that width is less that the total window size.
10250 +static ssize_t debug_width_fwrite(struct file *filp,
10251 + const char __user *ubuf,
10255 + char buf[U64STR_SIZE];
10256 + int csize = min(cnt, sizeof(buf));
10260 + memset(buf, '\0', sizeof(buf));
10261 + if (copy_from_user(buf, ubuf, csize))
10264 + buf[U64STR_SIZE-1] = '\0'; /* just in case */
10265 + err = kstrtoull(buf, 10, &val);
10269 + mutex_lock(&data.lock);
10270 + if (val < data.sample_window)
10271 + data.sample_width = val;
10273 + mutex_unlock(&data.lock);
10276 + mutex_unlock(&data.lock);
10279 + wake_up_process(kthread);
10285 + * debug_window_fopen - Open function for "window" debugfs entry
10286 + * @inode: The in-kernel inode representation of the debugfs "file"
10287 + * @filp: The active open file structure for the debugfs "file"
10289 + * This function provides an open implementation for the "window" debugfs
10290 + * interface to the hardware latency detector. The window is the total time
10291 + * in us that will be considered one sample period. Conceptually, windows
10292 + * occur back-to-back and contain a sample width period during which
10293 + * actual sampling occurs.
10295 +static int debug_window_fopen(struct inode *inode, struct file *filp)
10301 + * debug_window_fread - Read function for "window" debugfs entry
10302 + * @filp: The active open file structure for the debugfs "file"
10303 + * @ubuf: The userspace provided buffer to read value into
10304 + * @cnt: The maximum number of bytes to read
10305 + * @ppos: The current "file" position
10307 + * This function provides a read implementation for the "window" debugfs
10308 + * interface to the hardware latency detector. The window is the total time
10309 + * in us that will be considered one sample period. Conceptually, windows
10310 + * occur back-to-back and contain a sample width period during which
10311 + * actual sampling occurs. Can be used to read the total window size.
10313 +static ssize_t debug_window_fread(struct file *filp, char __user *ubuf,
10314 + size_t cnt, loff_t *ppos)
10316 + return simple_data_read(filp, ubuf, cnt, ppos, &data.sample_window);
10320 + * debug_window_fwrite - Write function for "window" debugfs entry
10321 + * @filp: The active open file structure for the debugfs "file"
10322 + * @ubuf: The user buffer that contains the value to write
10323 + * @cnt: The maximum number of bytes to write to "file"
10324 + * @ppos: The current position in the debugfs "file"
10326 + * This function provides a write implementation for the "window" debufds
10327 + * interface to the hardware latency detetector. The window is the total time
10328 + * in us that will be considered one sample period. Conceptually, windows
10329 + * occur back-to-back and contain a sample width period during which
10330 + * actual sampling occurs. Can be used to write a new total window size. It
10331 + * is enfoced that any value written must be greater than the sample width
10332 + * size, or an error results.
10334 +static ssize_t debug_window_fwrite(struct file *filp,
10335 + const char __user *ubuf,
10339 + char buf[U64STR_SIZE];
10340 + int csize = min(cnt, sizeof(buf));
10344 + memset(buf, '\0', sizeof(buf));
10345 + if (copy_from_user(buf, ubuf, csize))
10348 + buf[U64STR_SIZE-1] = '\0'; /* just in case */
10349 + err = kstrtoull(buf, 10, &val);
10353 + mutex_lock(&data.lock);
10354 + if (data.sample_width < val)
10355 + data.sample_window = val;
10357 + mutex_unlock(&data.lock);
10360 + mutex_unlock(&data.lock);
10366 + * Function pointers for the "count" debugfs file operations
10368 +static const struct file_operations count_fops = {
10369 + .open = debug_count_fopen,
10370 + .read = debug_count_fread,
10371 + .write = debug_count_fwrite,
10372 + .owner = THIS_MODULE,
10376 + * Function pointers for the "enable" debugfs file operations
10378 +static const struct file_operations enable_fops = {
10379 + .open = debug_enable_fopen,
10380 + .read = debug_enable_fread,
10381 + .write = debug_enable_fwrite,
10382 + .owner = THIS_MODULE,
10386 + * Function pointers for the "max" debugfs file operations
10388 +static const struct file_operations max_fops = {
10389 + .open = debug_max_fopen,
10390 + .read = debug_max_fread,
10391 + .write = debug_max_fwrite,
10392 + .owner = THIS_MODULE,
10396 + * Function pointers for the "sample" debugfs file operations
10398 +static const struct file_operations sample_fops = {
10399 + .open = debug_sample_fopen,
10400 + .read = debug_sample_fread,
10401 + .release = debug_sample_release,
10402 + .owner = THIS_MODULE,
10406 + * Function pointers for the "threshold" debugfs file operations
10408 +static const struct file_operations threshold_fops = {
10409 + .open = debug_threshold_fopen,
10410 + .read = debug_threshold_fread,
10411 + .write = debug_threshold_fwrite,
10412 + .owner = THIS_MODULE,
10416 + * Function pointers for the "width" debugfs file operations
10418 +static const struct file_operations width_fops = {
10419 + .open = debug_width_fopen,
10420 + .read = debug_width_fread,
10421 + .write = debug_width_fwrite,
10422 + .owner = THIS_MODULE,
10426 + * Function pointers for the "window" debugfs file operations
10428 +static const struct file_operations window_fops = {
10429 + .open = debug_window_fopen,
10430 + .read = debug_window_fread,
10431 + .write = debug_window_fwrite,
10432 + .owner = THIS_MODULE,
10436 + * init_debugfs - A function to initialize the debugfs interface files
10438 + * This function creates entries in debugfs for "hwlat_detector", including
10439 + * files to read values from the detector, current samples, and the
10440 + * maximum sample that has been captured since the hardware latency
10441 + * dectector was started.
10443 +static int init_debugfs(void)
10445 + int ret = -ENOMEM;
10447 + debug_dir = debugfs_create_dir(DRVNAME, NULL);
10449 + goto err_debug_dir;
10451 + debug_sample = debugfs_create_file("sample", 0444,
10454 + if (!debug_sample)
10457 + debug_count = debugfs_create_file("count", 0444,
10460 + if (!debug_count)
10463 + debug_max = debugfs_create_file("max", 0444,
10469 + debug_sample_window = debugfs_create_file("window", 0644,
10472 + if (!debug_sample_window)
10475 + debug_sample_width = debugfs_create_file("width", 0644,
10478 + if (!debug_sample_width)
10481 + debug_threshold = debugfs_create_file("threshold", 0644,
10483 + &threshold_fops);
10484 + if (!debug_threshold)
10485 + goto err_threshold;
10487 + debug_enable = debugfs_create_file("enable", 0644,
10488 + debug_dir, &enabled,
10490 + if (!debug_enable)
10499 + debugfs_remove(debug_threshold);
10501 + debugfs_remove(debug_sample_width);
10503 + debugfs_remove(debug_sample_window);
10505 + debugfs_remove(debug_max);
10507 + debugfs_remove(debug_count);
10509 + debugfs_remove(debug_sample);
10511 + debugfs_remove(debug_dir);
10518 + * free_debugfs - A function to cleanup the debugfs file interface
10520 +static void free_debugfs(void)
10522 + /* could also use a debugfs_remove_recursive */
10523 + debugfs_remove(debug_enable);
10524 + debugfs_remove(debug_threshold);
10525 + debugfs_remove(debug_sample_width);
10526 + debugfs_remove(debug_sample_window);
10527 + debugfs_remove(debug_max);
10528 + debugfs_remove(debug_count);
10529 + debugfs_remove(debug_sample);
10530 + debugfs_remove(debug_dir);
10534 + * detector_init - Standard module initialization code
10536 +static int detector_init(void)
10538 + int ret = -ENOMEM;
10540 + pr_info(BANNER "version %s\n", VERSION);
10542 + ret = init_stats();
10546 + ret = init_debugfs();
10551 + ret = start_kthread();
10556 + ring_buffer_free(ring_buffer);
10563 + * detector_exit - Standard module cleanup code
10565 +static void detector_exit(void)
10571 + err = stop_kthread();
10573 + pr_err(BANNER "cannot stop kthread\n");
10577 + ring_buffer_free(ring_buffer); /* free up the ring buffer */
10581 +module_init(detector_init);
10582 +module_exit(detector_exit);
10583 diff --git a/drivers/mmc/host/mmci.c b/drivers/mmc/host/mmci.c
10584 index acece3299756..58ea04a03fa9 100644
10585 --- a/drivers/mmc/host/mmci.c
10586 +++ b/drivers/mmc/host/mmci.c
10587 @@ -1155,15 +1155,12 @@ static irqreturn_t mmci_pio_irq(int irq, void *dev_id)
10588 struct sg_mapping_iter *sg_miter = &host->sg_miter;
10589 struct variant_data *variant = host->variant;
10590 void __iomem *base = host->base;
10591 - unsigned long flags;
10594 status = readl(base + MMCISTATUS);
10596 dev_dbg(mmc_dev(host->mmc), "irq1 (pio) %08x\n", status);
10598 - local_irq_save(flags);
10601 unsigned int remain, len;
10603 @@ -1203,8 +1200,6 @@ static irqreturn_t mmci_pio_irq(int irq, void *dev_id)
10605 sg_miter_stop(sg_miter);
10607 - local_irq_restore(flags);
10610 * If we have less than the fifo 'half-full' threshold to transfer,
10611 * trigger a PIO interrupt as soon as any data is available.
10612 diff --git a/drivers/net/ethernet/3com/3c59x.c b/drivers/net/ethernet/3com/3c59x.c
10613 index 2839af00f20c..4348b9c850d3 100644
10614 --- a/drivers/net/ethernet/3com/3c59x.c
10615 +++ b/drivers/net/ethernet/3com/3c59x.c
10616 @@ -842,9 +842,9 @@ static void poll_vortex(struct net_device *dev)
10618 struct vortex_private *vp = netdev_priv(dev);
10619 unsigned long flags;
10620 - local_irq_save(flags);
10621 + local_irq_save_nort(flags);
10622 (vp->full_bus_master_rx ? boomerang_interrupt:vortex_interrupt)(dev->irq,dev);
10623 - local_irq_restore(flags);
10624 + local_irq_restore_nort(flags);
10628 @@ -1916,12 +1916,12 @@ static void vortex_tx_timeout(struct net_device *dev)
10629 * Block interrupts because vortex_interrupt does a bare spin_lock()
10631 unsigned long flags;
10632 - local_irq_save(flags);
10633 + local_irq_save_nort(flags);
10634 if (vp->full_bus_master_tx)
10635 boomerang_interrupt(dev->irq, dev);
10637 vortex_interrupt(dev->irq, dev);
10638 - local_irq_restore(flags);
10639 + local_irq_restore_nort(flags);
10643 diff --git a/drivers/net/ethernet/atheros/atl1c/atl1c_main.c b/drivers/net/ethernet/atheros/atl1c/atl1c_main.c
10644 index 8b5988e210d5..cf9928ccdd7e 100644
10645 --- a/drivers/net/ethernet/atheros/atl1c/atl1c_main.c
10646 +++ b/drivers/net/ethernet/atheros/atl1c/atl1c_main.c
10647 @@ -2221,11 +2221,7 @@ static netdev_tx_t atl1c_xmit_frame(struct sk_buff *skb,
10650 tpd_req = atl1c_cal_tpd_req(skb);
10651 - if (!spin_trylock_irqsave(&adapter->tx_lock, flags)) {
10652 - if (netif_msg_pktdata(adapter))
10653 - dev_info(&adapter->pdev->dev, "tx locked\n");
10654 - return NETDEV_TX_LOCKED;
10656 + spin_lock_irqsave(&adapter->tx_lock, flags);
10658 if (atl1c_tpd_avail(adapter, type) < tpd_req) {
10659 /* no enough descriptor, just stop queue */
10660 diff --git a/drivers/net/ethernet/atheros/atl1e/atl1e_main.c b/drivers/net/ethernet/atheros/atl1e/atl1e_main.c
10661 index 59a03a193e83..734f7a7ad2c3 100644
10662 --- a/drivers/net/ethernet/atheros/atl1e/atl1e_main.c
10663 +++ b/drivers/net/ethernet/atheros/atl1e/atl1e_main.c
10664 @@ -1880,8 +1880,7 @@ static netdev_tx_t atl1e_xmit_frame(struct sk_buff *skb,
10665 return NETDEV_TX_OK;
10667 tpd_req = atl1e_cal_tdp_req(skb);
10668 - if (!spin_trylock_irqsave(&adapter->tx_lock, flags))
10669 - return NETDEV_TX_LOCKED;
10670 + spin_lock_irqsave(&adapter->tx_lock, flags);
10672 if (atl1e_tpd_avail(adapter) < tpd_req) {
10673 /* no enough descriptor, just stop queue */
10674 diff --git a/drivers/net/ethernet/chelsio/cxgb/sge.c b/drivers/net/ethernet/chelsio/cxgb/sge.c
10675 index 526ea74e82d9..86f467a2c485 100644
10676 --- a/drivers/net/ethernet/chelsio/cxgb/sge.c
10677 +++ b/drivers/net/ethernet/chelsio/cxgb/sge.c
10678 @@ -1664,8 +1664,7 @@ static int t1_sge_tx(struct sk_buff *skb, struct adapter *adapter,
10679 struct cmdQ *q = &sge->cmdQ[qid];
10680 unsigned int credits, pidx, genbit, count, use_sched_skb = 0;
10682 - if (!spin_trylock(&q->lock))
10683 - return NETDEV_TX_LOCKED;
10684 + spin_lock(&q->lock);
10686 reclaim_completed_tx(sge, q);
10688 diff --git a/drivers/net/ethernet/neterion/s2io.c b/drivers/net/ethernet/neterion/s2io.c
10689 index 9ba975853ec6..813cfa698160 100644
10690 --- a/drivers/net/ethernet/neterion/s2io.c
10691 +++ b/drivers/net/ethernet/neterion/s2io.c
10692 @@ -4084,12 +4084,7 @@ static netdev_tx_t s2io_xmit(struct sk_buff *skb, struct net_device *dev)
10693 [skb->priority & (MAX_TX_FIFOS - 1)];
10694 fifo = &mac_control->fifos[queue];
10696 - if (do_spin_lock)
10697 - spin_lock_irqsave(&fifo->tx_lock, flags);
10699 - if (unlikely(!spin_trylock_irqsave(&fifo->tx_lock, flags)))
10700 - return NETDEV_TX_LOCKED;
10702 + spin_lock_irqsave(&fifo->tx_lock, flags);
10704 if (sp->config.multiq) {
10705 if (__netif_subqueue_stopped(dev, fifo->fifo_no)) {
10706 diff --git a/drivers/net/ethernet/oki-semi/pch_gbe/pch_gbe_main.c b/drivers/net/ethernet/oki-semi/pch_gbe/pch_gbe_main.c
10707 index 3b98b263bad0..ca4add749410 100644
10708 --- a/drivers/net/ethernet/oki-semi/pch_gbe/pch_gbe_main.c
10709 +++ b/drivers/net/ethernet/oki-semi/pch_gbe/pch_gbe_main.c
10710 @@ -2137,10 +2137,8 @@ static int pch_gbe_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
10711 struct pch_gbe_tx_ring *tx_ring = adapter->tx_ring;
10712 unsigned long flags;
10714 - if (!spin_trylock_irqsave(&tx_ring->tx_lock, flags)) {
10715 - /* Collision - tell upper layer to requeue */
10716 - return NETDEV_TX_LOCKED;
10718 + spin_lock_irqsave(&tx_ring->tx_lock, flags);
10720 if (unlikely(!PCH_GBE_DESC_UNUSED(tx_ring))) {
10721 netif_stop_queue(netdev);
10722 spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
10723 diff --git a/drivers/net/ethernet/realtek/8139too.c b/drivers/net/ethernet/realtek/8139too.c
10724 index ef668d300800..d987d571fdd6 100644
10725 --- a/drivers/net/ethernet/realtek/8139too.c
10726 +++ b/drivers/net/ethernet/realtek/8139too.c
10727 @@ -2229,7 +2229,7 @@ static void rtl8139_poll_controller(struct net_device *dev)
10728 struct rtl8139_private *tp = netdev_priv(dev);
10729 const int irq = tp->pci_dev->irq;
10731 - disable_irq(irq);
10732 + disable_irq_nosync(irq);
10733 rtl8139_interrupt(irq, dev);
10736 diff --git a/drivers/net/ethernet/tehuti/tehuti.c b/drivers/net/ethernet/tehuti/tehuti.c
10737 index 14c9d1baa85c..e1a5305418a8 100644
10738 --- a/drivers/net/ethernet/tehuti/tehuti.c
10739 +++ b/drivers/net/ethernet/tehuti/tehuti.c
10740 @@ -1629,13 +1629,8 @@ static netdev_tx_t bdx_tx_transmit(struct sk_buff *skb,
10741 unsigned long flags;
10744 - local_irq_save(flags);
10745 - if (!spin_trylock(&priv->tx_lock)) {
10746 - local_irq_restore(flags);
10747 - DBG("%s[%s]: TX locked, returning NETDEV_TX_LOCKED\n",
10748 - BDX_DRV_NAME, ndev->name);
10749 - return NETDEV_TX_LOCKED;
10752 + spin_lock_irqsave(&priv->tx_lock, flags);
10754 /* build tx descriptor */
10755 BDX_ASSERT(f->m.wptr >= f->m.memsz); /* started with valid wptr */
10756 diff --git a/drivers/net/rionet.c b/drivers/net/rionet.c
10757 index e7034c55e796..2e4ee0f912bf 100644
10758 --- a/drivers/net/rionet.c
10759 +++ b/drivers/net/rionet.c
10760 @@ -174,11 +174,7 @@ static int rionet_start_xmit(struct sk_buff *skb, struct net_device *ndev)
10761 unsigned long flags;
10764 - local_irq_save(flags);
10765 - if (!spin_trylock(&rnet->tx_lock)) {
10766 - local_irq_restore(flags);
10767 - return NETDEV_TX_LOCKED;
10769 + spin_lock_irqsave(&rnet->tx_lock, flags);
10771 if (is_multicast_ether_addr(eth->h_dest))
10772 add_num = nets[rnet->mport->id].nact;
10773 diff --git a/drivers/net/wireless/orinoco/orinoco_usb.c b/drivers/net/wireless/orinoco/orinoco_usb.c
10774 index f2cd513d54b2..6c0f4c9638a2 100644
10775 --- a/drivers/net/wireless/orinoco/orinoco_usb.c
10776 +++ b/drivers/net/wireless/orinoco/orinoco_usb.c
10777 @@ -697,7 +697,7 @@ static void ezusb_req_ctx_wait(struct ezusb_priv *upriv,
10778 while (!ctx->done.done && msecs--)
10781 - wait_event_interruptible(ctx->done.wait,
10782 + swait_event_interruptible(ctx->done.wait,
10786 diff --git a/drivers/pci/access.c b/drivers/pci/access.c
10787 index 59ac36fe7c42..7a45a20af78a 100644
10788 --- a/drivers/pci/access.c
10789 +++ b/drivers/pci/access.c
10790 @@ -561,7 +561,7 @@ void pci_cfg_access_unlock(struct pci_dev *dev)
10791 WARN_ON(!dev->block_cfg_access);
10793 dev->block_cfg_access = 0;
10794 - wake_up_all(&pci_cfg_wait);
10795 + wake_up_all_locked(&pci_cfg_wait);
10796 raw_spin_unlock_irqrestore(&pci_lock, flags);
10798 EXPORT_SYMBOL_GPL(pci_cfg_access_unlock);
10799 diff --git a/drivers/pinctrl/qcom/pinctrl-msm.c b/drivers/pinctrl/qcom/pinctrl-msm.c
10800 index 9736f9be5447..5fe9b173dcb3 100644
10801 --- a/drivers/pinctrl/qcom/pinctrl-msm.c
10802 +++ b/drivers/pinctrl/qcom/pinctrl-msm.c
10803 @@ -60,7 +60,7 @@ struct msm_pinctrl {
10804 struct notifier_block restart_nb;
10808 + raw_spinlock_t lock;
10810 DECLARE_BITMAP(dual_edge_irqs, MAX_NR_GPIO);
10811 DECLARE_BITMAP(enabled_irqs, MAX_NR_GPIO);
10812 @@ -156,14 +156,14 @@ static int msm_pinmux_set_mux(struct pinctrl_dev *pctldev,
10813 if (WARN_ON(i == g->nfuncs))
10816 - spin_lock_irqsave(&pctrl->lock, flags);
10817 + raw_spin_lock_irqsave(&pctrl->lock, flags);
10819 val = readl(pctrl->regs + g->ctl_reg);
10820 val &= ~(0x7 << g->mux_bit);
10821 val |= i << g->mux_bit;
10822 writel(val, pctrl->regs + g->ctl_reg);
10824 - spin_unlock_irqrestore(&pctrl->lock, flags);
10825 + raw_spin_unlock_irqrestore(&pctrl->lock, flags);
10829 @@ -326,14 +326,14 @@ static int msm_config_group_set(struct pinctrl_dev *pctldev,
10831 case PIN_CONFIG_OUTPUT:
10832 /* set output value */
10833 - spin_lock_irqsave(&pctrl->lock, flags);
10834 + raw_spin_lock_irqsave(&pctrl->lock, flags);
10835 val = readl(pctrl->regs + g->io_reg);
10837 val |= BIT(g->out_bit);
10839 val &= ~BIT(g->out_bit);
10840 writel(val, pctrl->regs + g->io_reg);
10841 - spin_unlock_irqrestore(&pctrl->lock, flags);
10842 + raw_spin_unlock_irqrestore(&pctrl->lock, flags);
10844 /* enable output */
10846 @@ -354,12 +354,12 @@ static int msm_config_group_set(struct pinctrl_dev *pctldev,
10850 - spin_lock_irqsave(&pctrl->lock, flags);
10851 + raw_spin_lock_irqsave(&pctrl->lock, flags);
10852 val = readl(pctrl->regs + g->ctl_reg);
10853 val &= ~(mask << bit);
10855 writel(val, pctrl->regs + g->ctl_reg);
10856 - spin_unlock_irqrestore(&pctrl->lock, flags);
10857 + raw_spin_unlock_irqrestore(&pctrl->lock, flags);
10861 @@ -387,13 +387,13 @@ static int msm_gpio_direction_input(struct gpio_chip *chip, unsigned offset)
10863 g = &pctrl->soc->groups[offset];
10865 - spin_lock_irqsave(&pctrl->lock, flags);
10866 + raw_spin_lock_irqsave(&pctrl->lock, flags);
10868 val = readl(pctrl->regs + g->ctl_reg);
10869 val &= ~BIT(g->oe_bit);
10870 writel(val, pctrl->regs + g->ctl_reg);
10872 - spin_unlock_irqrestore(&pctrl->lock, flags);
10873 + raw_spin_unlock_irqrestore(&pctrl->lock, flags);
10877 @@ -407,7 +407,7 @@ static int msm_gpio_direction_output(struct gpio_chip *chip, unsigned offset, in
10879 g = &pctrl->soc->groups[offset];
10881 - spin_lock_irqsave(&pctrl->lock, flags);
10882 + raw_spin_lock_irqsave(&pctrl->lock, flags);
10884 val = readl(pctrl->regs + g->io_reg);
10886 @@ -420,7 +420,7 @@ static int msm_gpio_direction_output(struct gpio_chip *chip, unsigned offset, in
10887 val |= BIT(g->oe_bit);
10888 writel(val, pctrl->regs + g->ctl_reg);
10890 - spin_unlock_irqrestore(&pctrl->lock, flags);
10891 + raw_spin_unlock_irqrestore(&pctrl->lock, flags);
10895 @@ -446,7 +446,7 @@ static void msm_gpio_set(struct gpio_chip *chip, unsigned offset, int value)
10897 g = &pctrl->soc->groups[offset];
10899 - spin_lock_irqsave(&pctrl->lock, flags);
10900 + raw_spin_lock_irqsave(&pctrl->lock, flags);
10902 val = readl(pctrl->regs + g->io_reg);
10904 @@ -455,7 +455,7 @@ static void msm_gpio_set(struct gpio_chip *chip, unsigned offset, int value)
10905 val &= ~BIT(g->out_bit);
10906 writel(val, pctrl->regs + g->io_reg);
10908 - spin_unlock_irqrestore(&pctrl->lock, flags);
10909 + raw_spin_unlock_irqrestore(&pctrl->lock, flags);
10912 #ifdef CONFIG_DEBUG_FS
10913 @@ -574,7 +574,7 @@ static void msm_gpio_irq_mask(struct irq_data *d)
10915 g = &pctrl->soc->groups[d->hwirq];
10917 - spin_lock_irqsave(&pctrl->lock, flags);
10918 + raw_spin_lock_irqsave(&pctrl->lock, flags);
10920 val = readl(pctrl->regs + g->intr_cfg_reg);
10921 val &= ~BIT(g->intr_enable_bit);
10922 @@ -582,7 +582,7 @@ static void msm_gpio_irq_mask(struct irq_data *d)
10924 clear_bit(d->hwirq, pctrl->enabled_irqs);
10926 - spin_unlock_irqrestore(&pctrl->lock, flags);
10927 + raw_spin_unlock_irqrestore(&pctrl->lock, flags);
10930 static void msm_gpio_irq_unmask(struct irq_data *d)
10931 @@ -595,7 +595,7 @@ static void msm_gpio_irq_unmask(struct irq_data *d)
10933 g = &pctrl->soc->groups[d->hwirq];
10935 - spin_lock_irqsave(&pctrl->lock, flags);
10936 + raw_spin_lock_irqsave(&pctrl->lock, flags);
10938 val = readl(pctrl->regs + g->intr_cfg_reg);
10939 val |= BIT(g->intr_enable_bit);
10940 @@ -603,7 +603,7 @@ static void msm_gpio_irq_unmask(struct irq_data *d)
10942 set_bit(d->hwirq, pctrl->enabled_irqs);
10944 - spin_unlock_irqrestore(&pctrl->lock, flags);
10945 + raw_spin_unlock_irqrestore(&pctrl->lock, flags);
10948 static void msm_gpio_irq_ack(struct irq_data *d)
10949 @@ -616,7 +616,7 @@ static void msm_gpio_irq_ack(struct irq_data *d)
10951 g = &pctrl->soc->groups[d->hwirq];
10953 - spin_lock_irqsave(&pctrl->lock, flags);
10954 + raw_spin_lock_irqsave(&pctrl->lock, flags);
10956 val = readl(pctrl->regs + g->intr_status_reg);
10957 if (g->intr_ack_high)
10958 @@ -628,7 +628,7 @@ static void msm_gpio_irq_ack(struct irq_data *d)
10959 if (test_bit(d->hwirq, pctrl->dual_edge_irqs))
10960 msm_gpio_update_dual_edge_pos(pctrl, g, d);
10962 - spin_unlock_irqrestore(&pctrl->lock, flags);
10963 + raw_spin_unlock_irqrestore(&pctrl->lock, flags);
10966 static int msm_gpio_irq_set_type(struct irq_data *d, unsigned int type)
10967 @@ -641,7 +641,7 @@ static int msm_gpio_irq_set_type(struct irq_data *d, unsigned int type)
10969 g = &pctrl->soc->groups[d->hwirq];
10971 - spin_lock_irqsave(&pctrl->lock, flags);
10972 + raw_spin_lock_irqsave(&pctrl->lock, flags);
10975 * For hw without possibility of detecting both edges
10976 @@ -715,7 +715,7 @@ static int msm_gpio_irq_set_type(struct irq_data *d, unsigned int type)
10977 if (test_bit(d->hwirq, pctrl->dual_edge_irqs))
10978 msm_gpio_update_dual_edge_pos(pctrl, g, d);
10980 - spin_unlock_irqrestore(&pctrl->lock, flags);
10981 + raw_spin_unlock_irqrestore(&pctrl->lock, flags);
10983 if (type & (IRQ_TYPE_LEVEL_LOW | IRQ_TYPE_LEVEL_HIGH))
10984 irq_set_handler_locked(d, handle_level_irq);
10985 @@ -731,11 +731,11 @@ static int msm_gpio_irq_set_wake(struct irq_data *d, unsigned int on)
10986 struct msm_pinctrl *pctrl = to_msm_pinctrl(gc);
10987 unsigned long flags;
10989 - spin_lock_irqsave(&pctrl->lock, flags);
10990 + raw_spin_lock_irqsave(&pctrl->lock, flags);
10992 irq_set_irq_wake(pctrl->irq, on);
10994 - spin_unlock_irqrestore(&pctrl->lock, flags);
10995 + raw_spin_unlock_irqrestore(&pctrl->lock, flags);
10999 @@ -881,7 +881,7 @@ int msm_pinctrl_probe(struct platform_device *pdev,
11000 pctrl->soc = soc_data;
11001 pctrl->chip = msm_gpio_template;
11003 - spin_lock_init(&pctrl->lock);
11004 + raw_spin_lock_init(&pctrl->lock);
11006 res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
11007 pctrl->regs = devm_ioremap_resource(&pdev->dev, res);
11008 diff --git a/drivers/scsi/fcoe/fcoe.c b/drivers/scsi/fcoe/fcoe.c
11009 index f4424063b860..cbbbebd86c6e 100644
11010 --- a/drivers/scsi/fcoe/fcoe.c
11011 +++ b/drivers/scsi/fcoe/fcoe.c
11012 @@ -1286,7 +1286,7 @@ static void fcoe_percpu_thread_destroy(unsigned int cpu)
11013 struct sk_buff *skb;
11015 struct fcoe_percpu_s *p0;
11016 - unsigned targ_cpu = get_cpu();
11017 + unsigned targ_cpu = get_cpu_light();
11018 #endif /* CONFIG_SMP */
11020 FCOE_DBG("Destroying receive thread for CPU %d\n", cpu);
11021 @@ -1342,7 +1342,7 @@ static void fcoe_percpu_thread_destroy(unsigned int cpu)
11023 spin_unlock_bh(&p->fcoe_rx_list.lock);
11029 * This a non-SMP scenario where the singular Rx thread is
11030 @@ -1566,11 +1566,11 @@ err2:
11031 static int fcoe_alloc_paged_crc_eof(struct sk_buff *skb, int tlen)
11033 struct fcoe_percpu_s *fps;
11035 + int rc, cpu = get_cpu_light();
11037 - fps = &get_cpu_var(fcoe_percpu);
11038 + fps = &per_cpu(fcoe_percpu, cpu);
11039 rc = fcoe_get_paged_crc_eof(skb, tlen, fps);
11040 - put_cpu_var(fcoe_percpu);
11045 @@ -1766,11 +1766,11 @@ static inline int fcoe_filter_frames(struct fc_lport *lport,
11049 - stats = per_cpu_ptr(lport->stats, get_cpu());
11050 + stats = per_cpu_ptr(lport->stats, get_cpu_light());
11051 stats->InvalidCRCCount++;
11052 if (stats->InvalidCRCCount < 5)
11053 printk(KERN_WARNING "fcoe: dropping frame with CRC error\n");
11059 @@ -1814,7 +1814,7 @@ static void fcoe_recv_frame(struct sk_buff *skb)
11061 hp = (struct fcoe_hdr *) skb_network_header(skb);
11063 - stats = per_cpu_ptr(lport->stats, get_cpu());
11064 + stats = per_cpu_ptr(lport->stats, get_cpu_light());
11065 if (unlikely(FC_FCOE_DECAPS_VER(hp) != FC_FCOE_VER)) {
11066 if (stats->ErrorFrames < 5)
11067 printk(KERN_WARNING "fcoe: FCoE version "
11068 @@ -1846,13 +1846,13 @@ static void fcoe_recv_frame(struct sk_buff *skb)
11071 if (!fcoe_filter_frames(lport, fp)) {
11074 fc_exch_recv(lport, fp);
11078 stats->ErrorFrames++;
11084 diff --git a/drivers/scsi/fcoe/fcoe_ctlr.c b/drivers/scsi/fcoe/fcoe_ctlr.c
11085 index 34a1b1f333b4..d91131210695 100644
11086 --- a/drivers/scsi/fcoe/fcoe_ctlr.c
11087 +++ b/drivers/scsi/fcoe/fcoe_ctlr.c
11088 @@ -831,7 +831,7 @@ static unsigned long fcoe_ctlr_age_fcfs(struct fcoe_ctlr *fip)
11090 INIT_LIST_HEAD(&del_list);
11092 - stats = per_cpu_ptr(fip->lp->stats, get_cpu());
11093 + stats = per_cpu_ptr(fip->lp->stats, get_cpu_light());
11095 list_for_each_entry_safe(fcf, next, &fip->fcfs, list) {
11096 deadline = fcf->time + fcf->fka_period + fcf->fka_period / 2;
11097 @@ -867,7 +867,7 @@ static unsigned long fcoe_ctlr_age_fcfs(struct fcoe_ctlr *fip)
11098 sel_time = fcf->time;
11104 list_for_each_entry_safe(fcf, next, &del_list, list) {
11105 /* Removes fcf from current list */
11106 diff --git a/drivers/scsi/libfc/fc_exch.c b/drivers/scsi/libfc/fc_exch.c
11107 index 30f9ef0c0d4f..6c686bc01a82 100644
11108 --- a/drivers/scsi/libfc/fc_exch.c
11109 +++ b/drivers/scsi/libfc/fc_exch.c
11110 @@ -814,10 +814,10 @@ static struct fc_exch *fc_exch_em_alloc(struct fc_lport *lport,
11112 memset(ep, 0, sizeof(*ep));
11115 + cpu = get_cpu_light();
11116 pool = per_cpu_ptr(mp->pool, cpu);
11117 spin_lock_bh(&pool->lock);
11121 /* peek cache of free slot */
11122 if (pool->left != FC_XID_UNKNOWN) {
11123 diff --git a/drivers/scsi/libsas/sas_ata.c b/drivers/scsi/libsas/sas_ata.c
11124 index 6f5e2720ffad..ee8a8ed49b89 100644
11125 --- a/drivers/scsi/libsas/sas_ata.c
11126 +++ b/drivers/scsi/libsas/sas_ata.c
11127 @@ -190,7 +190,7 @@ static unsigned int sas_ata_qc_issue(struct ata_queued_cmd *qc)
11128 /* TODO: audit callers to ensure they are ready for qc_issue to
11129 * unconditionally re-enable interrupts
11131 - local_irq_save(flags);
11132 + local_irq_save_nort(flags);
11133 spin_unlock(ap->lock);
11135 /* If the device fell off, no sense in issuing commands */
11136 @@ -255,7 +255,7 @@ static unsigned int sas_ata_qc_issue(struct ata_queued_cmd *qc)
11139 spin_lock(ap->lock);
11140 - local_irq_restore(flags);
11141 + local_irq_restore_nort(flags);
11145 diff --git a/drivers/scsi/qla2xxx/qla_inline.h b/drivers/scsi/qla2xxx/qla_inline.h
11146 index fee9eb7c8a60..b42d4adc42dc 100644
11147 --- a/drivers/scsi/qla2xxx/qla_inline.h
11148 +++ b/drivers/scsi/qla2xxx/qla_inline.h
11149 @@ -59,12 +59,12 @@ qla2x00_poll(struct rsp_que *rsp)
11151 unsigned long flags;
11152 struct qla_hw_data *ha = rsp->hw;
11153 - local_irq_save(flags);
11154 + local_irq_save_nort(flags);
11155 if (IS_P3P_TYPE(ha))
11156 qla82xx_poll(0, rsp);
11158 ha->isp_ops->intr_handler(0, rsp);
11159 - local_irq_restore(flags);
11160 + local_irq_restore_nort(flags);
11163 static inline uint8_t *
11164 diff --git a/drivers/thermal/x86_pkg_temp_thermal.c b/drivers/thermal/x86_pkg_temp_thermal.c
11165 index 7fc919f7da4d..e03fa17b8670 100644
11166 --- a/drivers/thermal/x86_pkg_temp_thermal.c
11167 +++ b/drivers/thermal/x86_pkg_temp_thermal.c
11169 #include <linux/pm.h>
11170 #include <linux/thermal.h>
11171 #include <linux/debugfs.h>
11172 +#include <linux/swork.h>
11173 #include <asm/cpu_device_id.h>
11174 #include <asm/mce.h>
11176 @@ -352,7 +353,7 @@ static void pkg_temp_thermal_threshold_work_fn(struct work_struct *work)
11180 -static int pkg_temp_thermal_platform_thermal_notify(__u64 msr_val)
11181 +static void platform_thermal_notify_work(struct swork_event *event)
11183 unsigned long flags;
11184 int cpu = smp_processor_id();
11185 @@ -369,7 +370,7 @@ static int pkg_temp_thermal_platform_thermal_notify(__u64 msr_val)
11186 pkg_work_scheduled[phy_id]) {
11187 disable_pkg_thres_interrupt();
11188 spin_unlock_irqrestore(&pkg_work_lock, flags);
11192 pkg_work_scheduled[phy_id] = 1;
11193 spin_unlock_irqrestore(&pkg_work_lock, flags);
11194 @@ -378,9 +379,48 @@ static int pkg_temp_thermal_platform_thermal_notify(__u64 msr_val)
11195 schedule_delayed_work_on(cpu,
11196 &per_cpu(pkg_temp_thermal_threshold_work, cpu),
11197 msecs_to_jiffies(notify_delay_ms));
11200 +#ifdef CONFIG_PREEMPT_RT_FULL
11201 +static struct swork_event notify_work;
11203 +static int thermal_notify_work_init(void)
11207 + err = swork_get();
11211 + INIT_SWORK(¬ify_work, platform_thermal_notify_work);
11215 +static void thermal_notify_work_cleanup(void)
11220 +static int pkg_temp_thermal_platform_thermal_notify(__u64 msr_val)
11222 + swork_queue(¬ify_work);
11226 +#else /* !CONFIG_PREEMPT_RT_FULL */
11228 +static int thermal_notify_work_init(void) { return 0; }
11230 +static void thermal_notify_work_cleanup(void) { }
11232 +static int pkg_temp_thermal_platform_thermal_notify(__u64 msr_val)
11234 + platform_thermal_notify_work(NULL);
11238 +#endif /* CONFIG_PREEMPT_RT_FULL */
11240 static int find_siblings_cpu(int cpu)
11243 @@ -584,6 +624,9 @@ static int __init pkg_temp_thermal_init(void)
11244 if (!x86_match_cpu(pkg_temp_thermal_ids))
11247 + if (!thermal_notify_work_init())
11250 spin_lock_init(&pkg_work_lock);
11251 platform_thermal_package_notify =
11252 pkg_temp_thermal_platform_thermal_notify;
11253 @@ -608,7 +651,7 @@ err_ret:
11254 kfree(pkg_work_scheduled);
11255 platform_thermal_package_notify = NULL;
11256 platform_thermal_package_rate_control = NULL;
11258 + thermal_notify_work_cleanup();
11262 @@ -633,6 +676,7 @@ static void __exit pkg_temp_thermal_exit(void)
11263 mutex_unlock(&phy_dev_list_mutex);
11264 platform_thermal_package_notify = NULL;
11265 platform_thermal_package_rate_control = NULL;
11266 + thermal_notify_work_cleanup();
11267 for_each_online_cpu(i)
11268 cancel_delayed_work_sync(
11269 &per_cpu(pkg_temp_thermal_threshold_work, i));
11270 diff --git a/drivers/tty/serial/8250/8250_core.c b/drivers/tty/serial/8250/8250_core.c
11271 index 39126460c1f5..af7701ca4d48 100644
11272 --- a/drivers/tty/serial/8250/8250_core.c
11273 +++ b/drivers/tty/serial/8250/8250_core.c
11274 @@ -58,7 +58,16 @@ static struct uart_driver serial8250_reg;
11276 static unsigned int skip_txen_test; /* force skip of txen test at init time */
11278 -#define PASS_LIMIT 512
11280 + * On -rt we can have a more delays, and legitimately
11281 + * so - so don't drop work spuriously and spam the
11284 +#ifdef CONFIG_PREEMPT_RT_FULL
11285 +# define PASS_LIMIT 1000000
11287 +# define PASS_LIMIT 512
11290 #include <asm/serial.h>
11292 diff --git a/drivers/tty/serial/8250/8250_port.c b/drivers/tty/serial/8250/8250_port.c
11293 index 56ccbcefdd85..a0b9e854672c 100644
11294 --- a/drivers/tty/serial/8250/8250_port.c
11295 +++ b/drivers/tty/serial/8250/8250_port.c
11297 #include <linux/nmi.h>
11298 #include <linux/mutex.h>
11299 #include <linux/slab.h>
11300 +#include <linux/kdb.h>
11301 #include <linux/uaccess.h>
11302 #include <linux/pm_runtime.h>
11304 @@ -2843,9 +2844,9 @@ void serial8250_console_write(struct uart_8250_port *up, const char *s,
11306 serial8250_rpm_get(up);
11309 + if (port->sysrq || oops_in_progress)
11311 - else if (oops_in_progress)
11312 + else if (in_kdb_printk())
11313 locked = spin_trylock_irqsave(&port->lock, flags);
11315 spin_lock_irqsave(&port->lock, flags);
11316 diff --git a/drivers/tty/serial/amba-pl011.c b/drivers/tty/serial/amba-pl011.c
11317 index 899a77187bde..3ff6363b3751 100644
11318 --- a/drivers/tty/serial/amba-pl011.c
11319 +++ b/drivers/tty/serial/amba-pl011.c
11320 @@ -2067,13 +2067,19 @@ pl011_console_write(struct console *co, const char *s, unsigned int count)
11322 clk_enable(uap->clk);
11324 - local_irq_save(flags);
11326 + * local_irq_save(flags);
11328 + * This local_irq_save() is nonsense. If we come in via sysrq
11329 + * handling then interrupts are already disabled. Aside of
11330 + * that the port.sysrq check is racy on SMP regardless.
11332 if (uap->port.sysrq)
11334 else if (oops_in_progress)
11335 - locked = spin_trylock(&uap->port.lock);
11336 + locked = spin_trylock_irqsave(&uap->port.lock, flags);
11338 - spin_lock(&uap->port.lock);
11339 + spin_lock_irqsave(&uap->port.lock, flags);
11342 * First save the CR then disable the interrupts
11343 @@ -2098,8 +2104,7 @@ pl011_console_write(struct console *co, const char *s, unsigned int count)
11344 writew(old_cr, uap->port.membase + UART011_CR);
11347 - spin_unlock(&uap->port.lock);
11348 - local_irq_restore(flags);
11349 + spin_unlock_irqrestore(&uap->port.lock, flags);
11351 clk_disable(uap->clk);
11353 diff --git a/drivers/tty/serial/omap-serial.c b/drivers/tty/serial/omap-serial.c
11354 index de1c143b475f..3fef536a1844 100644
11355 --- a/drivers/tty/serial/omap-serial.c
11356 +++ b/drivers/tty/serial/omap-serial.c
11357 @@ -1257,13 +1257,10 @@ serial_omap_console_write(struct console *co, const char *s,
11359 pm_runtime_get_sync(up->dev);
11361 - local_irq_save(flags);
11362 - if (up->port.sysrq)
11364 - else if (oops_in_progress)
11365 - locked = spin_trylock(&up->port.lock);
11366 + if (up->port.sysrq || oops_in_progress)
11367 + locked = spin_trylock_irqsave(&up->port.lock, flags);
11369 - spin_lock(&up->port.lock);
11370 + spin_lock_irqsave(&up->port.lock, flags);
11373 * First save the IER then disable the interrupts
11374 @@ -1292,8 +1289,7 @@ serial_omap_console_write(struct console *co, const char *s,
11375 pm_runtime_mark_last_busy(up->dev);
11376 pm_runtime_put_autosuspend(up->dev);
11378 - spin_unlock(&up->port.lock);
11379 - local_irq_restore(flags);
11380 + spin_unlock_irqrestore(&up->port.lock, flags);
11384 diff --git a/drivers/usb/core/hcd.c b/drivers/usb/core/hcd.c
11385 index c3f4f2ab7b33..ae1ffb5f7610 100644
11386 --- a/drivers/usb/core/hcd.c
11387 +++ b/drivers/usb/core/hcd.c
11388 @@ -1738,9 +1738,9 @@ static void __usb_hcd_giveback_urb(struct urb *urb)
11389 * and no one may trigger the above deadlock situation when
11390 * running complete() in tasklet.
11392 - local_irq_save(flags);
11393 + local_irq_save_nort(flags);
11394 urb->complete(urb);
11395 - local_irq_restore(flags);
11396 + local_irq_restore_nort(flags);
11398 usb_anchor_resume_wakeups(anchor);
11399 atomic_dec(&urb->use_count);
11400 diff --git a/drivers/usb/gadget/function/f_fs.c b/drivers/usb/gadget/function/f_fs.c
11401 index 9ad5145d3103..bf883695cdfb 100644
11402 --- a/drivers/usb/gadget/function/f_fs.c
11403 +++ b/drivers/usb/gadget/function/f_fs.c
11404 @@ -1404,7 +1404,7 @@ static void ffs_data_put(struct ffs_data *ffs)
11405 pr_info("%s(): freeing\n", __func__);
11406 ffs_data_clear(ffs);
11407 BUG_ON(waitqueue_active(&ffs->ev.waitq) ||
11408 - waitqueue_active(&ffs->ep0req_completion.wait));
11409 + swait_active(&ffs->ep0req_completion.wait));
11410 kfree(ffs->dev_name);
11413 diff --git a/drivers/usb/gadget/legacy/inode.c b/drivers/usb/gadget/legacy/inode.c
11414 index de014436fb22..e52700c7555e 100644
11415 --- a/drivers/usb/gadget/legacy/inode.c
11416 +++ b/drivers/usb/gadget/legacy/inode.c
11417 @@ -345,7 +345,7 @@ ep_io (struct ep_data *epdata, void *buf, unsigned len)
11418 spin_unlock_irq (&epdata->dev->lock);
11420 if (likely (value == 0)) {
11421 - value = wait_event_interruptible (done.wait, done.done);
11422 + value = swait_event_interruptible (done.wait, done.done);
11424 spin_lock_irq (&epdata->dev->lock);
11425 if (likely (epdata->ep != NULL)) {
11426 @@ -354,7 +354,7 @@ ep_io (struct ep_data *epdata, void *buf, unsigned len)
11427 usb_ep_dequeue (epdata->ep, epdata->req);
11428 spin_unlock_irq (&epdata->dev->lock);
11430 - wait_event (done.wait, done.done);
11431 + swait_event (done.wait, done.done);
11432 if (epdata->status == -ECONNRESET)
11433 epdata->status = -EINTR;
11435 diff --git a/drivers/usb/gadget/udc/atmel_usba_udc.c b/drivers/usb/gadget/udc/atmel_usba_udc.c
11436 index f92f5aff0dd5..f9bba26e3655 100644
11437 --- a/drivers/usb/gadget/udc/atmel_usba_udc.c
11438 +++ b/drivers/usb/gadget/udc/atmel_usba_udc.c
11440 #include <linux/device.h>
11441 #include <linux/dma-mapping.h>
11442 #include <linux/list.h>
11443 +#include <linux/mfd/syscon.h>
11444 #include <linux/platform_device.h>
11445 +#include <linux/regmap.h>
11446 #include <linux/usb/ch9.h>
11447 #include <linux/usb/gadget.h>
11448 #include <linux/usb/atmel_usba_udc.h>
11449 @@ -1888,20 +1890,15 @@ static int atmel_usba_stop(struct usb_gadget *gadget)
11451 static void at91sam9rl_toggle_bias(struct usba_udc *udc, int is_on)
11453 - unsigned int uckr = at91_pmc_read(AT91_CKGR_UCKR);
11456 - at91_pmc_write(AT91_CKGR_UCKR, uckr | AT91_PMC_BIASEN);
11458 - at91_pmc_write(AT91_CKGR_UCKR, uckr & ~(AT91_PMC_BIASEN));
11459 + regmap_update_bits(udc->pmc, AT91_CKGR_UCKR, AT91_PMC_BIASEN,
11460 + is_on ? AT91_PMC_BIASEN : 0);
11463 static void at91sam9g45_pulse_bias(struct usba_udc *udc)
11465 - unsigned int uckr = at91_pmc_read(AT91_CKGR_UCKR);
11467 - at91_pmc_write(AT91_CKGR_UCKR, uckr & ~(AT91_PMC_BIASEN));
11468 - at91_pmc_write(AT91_CKGR_UCKR, uckr | AT91_PMC_BIASEN);
11469 + regmap_update_bits(udc->pmc, AT91_CKGR_UCKR, AT91_PMC_BIASEN, 0);
11470 + regmap_update_bits(udc->pmc, AT91_CKGR_UCKR, AT91_PMC_BIASEN,
11471 + AT91_PMC_BIASEN);
11474 static const struct usba_udc_errata at91sam9rl_errata = {
11475 @@ -1938,6 +1935,9 @@ static struct usba_ep * atmel_udc_of_init(struct platform_device *pdev,
11476 return ERR_PTR(-EINVAL);
11478 udc->errata = match->data;
11479 + udc->pmc = syscon_regmap_lookup_by_compatible("atmel,at91sam9g45-pmc");
11480 + if (udc->errata && IS_ERR(udc->pmc))
11481 + return ERR_CAST(udc->pmc);
11485 diff --git a/drivers/usb/gadget/udc/atmel_usba_udc.h b/drivers/usb/gadget/udc/atmel_usba_udc.h
11486 index ea448a344767..3e1c9d589dfa 100644
11487 --- a/drivers/usb/gadget/udc/atmel_usba_udc.h
11488 +++ b/drivers/usb/gadget/udc/atmel_usba_udc.h
11489 @@ -354,6 +354,8 @@ struct usba_udc {
11490 struct dentry *debugfs_root;
11491 struct dentry *debugfs_regs;
11494 + struct regmap *pmc;
11497 static inline struct usba_ep *to_usba_ep(struct usb_ep *ep)
11498 diff --git a/fs/aio.c b/fs/aio.c
11499 index fe4f49212b99..c3194afdc3df 100644
11503 #include <linux/ramfs.h>
11504 #include <linux/percpu-refcount.h>
11505 #include <linux/mount.h>
11506 +#include <linux/swork.h>
11508 #include <asm/kmap_types.h>
11509 #include <asm/uaccess.h>
11510 @@ -115,7 +116,7 @@ struct kioctx {
11511 struct page **ring_pages;
11514 - struct work_struct free_work;
11515 + struct swork_event free_work;
11518 * signals when all in-flight requests are done
11519 @@ -258,6 +259,7 @@ static int __init aio_setup(void)
11520 .mount = aio_mount,
11521 .kill_sb = kill_anon_super,
11523 + BUG_ON(swork_get());
11524 aio_mnt = kern_mount(&aio_fs);
11525 if (IS_ERR(aio_mnt))
11526 panic("Failed to create aio fs mount.");
11527 @@ -573,9 +575,9 @@ static int kiocb_cancel(struct aio_kiocb *kiocb)
11528 return cancel(&kiocb->common);
11531 -static void free_ioctx(struct work_struct *work)
11532 +static void free_ioctx(struct swork_event *sev)
11534 - struct kioctx *ctx = container_of(work, struct kioctx, free_work);
11535 + struct kioctx *ctx = container_of(sev, struct kioctx, free_work);
11537 pr_debug("freeing %p\n", ctx);
11539 @@ -594,8 +596,8 @@ static void free_ioctx_reqs(struct percpu_ref *ref)
11540 if (ctx->rq_wait && atomic_dec_and_test(&ctx->rq_wait->count))
11541 complete(&ctx->rq_wait->comp);
11543 - INIT_WORK(&ctx->free_work, free_ioctx);
11544 - schedule_work(&ctx->free_work);
11545 + INIT_SWORK(&ctx->free_work, free_ioctx);
11546 + swork_queue(&ctx->free_work);
11550 @@ -603,9 +605,9 @@ static void free_ioctx_reqs(struct percpu_ref *ref)
11551 * and ctx->users has dropped to 0, so we know no more kiocbs can be submitted -
11552 * now it's safe to cancel any that need to be.
11554 -static void free_ioctx_users(struct percpu_ref *ref)
11555 +static void free_ioctx_users_work(struct swork_event *sev)
11557 - struct kioctx *ctx = container_of(ref, struct kioctx, users);
11558 + struct kioctx *ctx = container_of(sev, struct kioctx, free_work);
11559 struct aio_kiocb *req;
11561 spin_lock_irq(&ctx->ctx_lock);
11562 @@ -624,6 +626,14 @@ static void free_ioctx_users(struct percpu_ref *ref)
11563 percpu_ref_put(&ctx->reqs);
11566 +static void free_ioctx_users(struct percpu_ref *ref)
11568 + struct kioctx *ctx = container_of(ref, struct kioctx, users);
11570 + INIT_SWORK(&ctx->free_work, free_ioctx_users_work);
11571 + swork_queue(&ctx->free_work);
11574 static int ioctx_add_table(struct kioctx *ctx, struct mm_struct *mm)
11576 unsigned i, new_nr;
11577 diff --git a/fs/autofs4/autofs_i.h b/fs/autofs4/autofs_i.h
11578 index 502d3892d8a4..05af8d3e6e88 100644
11579 --- a/fs/autofs4/autofs_i.h
11580 +++ b/fs/autofs4/autofs_i.h
11582 #include <linux/sched.h>
11583 #include <linux/mount.h>
11584 #include <linux/namei.h>
11585 +#include <linux/delay.h>
11586 #include <asm/current.h>
11587 #include <asm/uaccess.h>
11589 diff --git a/fs/autofs4/expire.c b/fs/autofs4/expire.c
11590 index 7a5a598a2d94..d08bcdc30566 100644
11591 --- a/fs/autofs4/expire.c
11592 +++ b/fs/autofs4/expire.c
11593 @@ -150,7 +150,7 @@ again:
11594 parent = p->d_parent;
11595 if (!spin_trylock(&parent->d_lock)) {
11596 spin_unlock(&p->d_lock);
11601 spin_unlock(&p->d_lock);
11602 diff --git a/fs/buffer.c b/fs/buffer.c
11603 index 4f4cd959da7c..72b27e17b907 100644
11606 @@ -305,8 +305,7 @@ static void end_buffer_async_read(struct buffer_head *bh, int uptodate)
11607 * decide that the page is now completely done.
11609 first = page_buffers(page);
11610 - local_irq_save(flags);
11611 - bit_spin_lock(BH_Uptodate_Lock, &first->b_state);
11612 + flags = bh_uptodate_lock_irqsave(first);
11613 clear_buffer_async_read(bh);
11616 @@ -319,8 +318,7 @@ static void end_buffer_async_read(struct buffer_head *bh, int uptodate)
11618 tmp = tmp->b_this_page;
11619 } while (tmp != bh);
11620 - bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
11621 - local_irq_restore(flags);
11622 + bh_uptodate_unlock_irqrestore(first, flags);
11625 * If none of the buffers had errors and they are all
11626 @@ -332,9 +330,7 @@ static void end_buffer_async_read(struct buffer_head *bh, int uptodate)
11630 - bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
11631 - local_irq_restore(flags);
11633 + bh_uptodate_unlock_irqrestore(first, flags);
11637 @@ -362,8 +358,7 @@ void end_buffer_async_write(struct buffer_head *bh, int uptodate)
11640 first = page_buffers(page);
11641 - local_irq_save(flags);
11642 - bit_spin_lock(BH_Uptodate_Lock, &first->b_state);
11643 + flags = bh_uptodate_lock_irqsave(first);
11645 clear_buffer_async_write(bh);
11647 @@ -375,15 +370,12 @@ void end_buffer_async_write(struct buffer_head *bh, int uptodate)
11649 tmp = tmp->b_this_page;
11651 - bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
11652 - local_irq_restore(flags);
11653 + bh_uptodate_unlock_irqrestore(first, flags);
11654 end_page_writeback(page);
11658 - bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
11659 - local_irq_restore(flags);
11661 + bh_uptodate_unlock_irqrestore(first, flags);
11663 EXPORT_SYMBOL(end_buffer_async_write);
11665 @@ -3325,6 +3317,7 @@ struct buffer_head *alloc_buffer_head(gfp_t gfp_flags)
11666 struct buffer_head *ret = kmem_cache_zalloc(bh_cachep, gfp_flags);
11668 INIT_LIST_HEAD(&ret->b_assoc_buffers);
11669 + buffer_head_init_locks(ret);
11671 __this_cpu_inc(bh_accounting.nr);
11673 diff --git a/fs/dcache.c b/fs/dcache.c
11674 index 849c1c1e787b..995995130191 100644
11678 #include <linux/mm.h>
11679 #include <linux/fs.h>
11680 #include <linux/fsnotify.h>
11681 +#include <linux/delay.h>
11682 #include <linux/slab.h>
11683 #include <linux/init.h>
11684 #include <linux/hash.h>
11685 @@ -747,6 +748,8 @@ static inline bool fast_dput(struct dentry *dentry)
11687 void dput(struct dentry *dentry)
11689 + struct dentry *parent;
11691 if (unlikely(!dentry))
11694 @@ -783,9 +786,18 @@ repeat:
11698 - dentry = dentry_kill(dentry);
11701 + parent = dentry_kill(dentry);
11705 + if (parent == dentry) {
11706 + /* the task with the highest priority won't schedule */
11707 + r = cond_resched();
11716 @@ -2397,7 +2409,7 @@ again:
11717 if (dentry->d_lockref.count == 1) {
11718 if (!spin_trylock(&inode->i_lock)) {
11719 spin_unlock(&dentry->d_lock);
11724 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
11725 diff --git a/fs/eventpoll.c b/fs/eventpoll.c
11726 index 1e009cad8d5c..d0c12504d3b4 100644
11727 --- a/fs/eventpoll.c
11728 +++ b/fs/eventpoll.c
11729 @@ -505,12 +505,12 @@ static int ep_poll_wakeup_proc(void *priv, void *cookie, int call_nests)
11731 static void ep_poll_safewake(wait_queue_head_t *wq)
11733 - int this_cpu = get_cpu();
11734 + int this_cpu = get_cpu_light();
11736 ep_call_nested(&poll_safewake_ncalls, EP_MAX_NESTS,
11737 ep_poll_wakeup_proc, NULL, wq, (void *) (long) this_cpu);
11743 static void ep_remove_wait_queue(struct eppoll_entry *pwq)
11744 diff --git a/fs/exec.c b/fs/exec.c
11745 index 3a6de10d3891..74962ac0a738 100644
11748 @@ -866,12 +866,14 @@ static int exec_mmap(struct mm_struct *mm)
11752 + preempt_disable_rt();
11753 active_mm = tsk->active_mm;
11755 tsk->active_mm = mm;
11756 activate_mm(active_mm, mm);
11757 tsk->mm->vmacache_seqnum = 0;
11758 vmacache_flush(tsk);
11759 + preempt_enable_rt();
11762 up_read(&old_mm->mmap_sem);
11763 diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
11764 index 2871576fbca4..d1137790ea58 100644
11765 --- a/fs/f2fs/f2fs.h
11766 +++ b/fs/f2fs/f2fs.h
11769 #ifdef CONFIG_F2FS_CHECK_FS
11770 #define f2fs_bug_on(sbi, condition) BUG_ON(condition)
11771 -#define f2fs_down_write(x, y) down_write_nest_lock(x, y)
11773 #define f2fs_bug_on(sbi, condition) \
11776 set_sbi_flag(sbi, SBI_NEED_FSCK); \
11779 -#define f2fs_down_write(x, y) down_write(x)
11783 @@ -959,7 +957,7 @@ static inline void f2fs_unlock_op(struct f2fs_sb_info *sbi)
11785 static inline void f2fs_lock_all(struct f2fs_sb_info *sbi)
11787 - f2fs_down_write(&sbi->cp_rwsem, &sbi->cp_mutex);
11788 + down_write(&sbi->cp_rwsem);
11791 static inline void f2fs_unlock_all(struct f2fs_sb_info *sbi)
11792 diff --git a/fs/jbd2/checkpoint.c b/fs/jbd2/checkpoint.c
11793 index 684996c8a3a4..6e18a06aaabe 100644
11794 --- a/fs/jbd2/checkpoint.c
11795 +++ b/fs/jbd2/checkpoint.c
11796 @@ -116,6 +116,8 @@ void __jbd2_log_wait_for_space(journal_t *journal)
11797 nblocks = jbd2_space_needed(journal);
11798 while (jbd2_log_space_left(journal) < nblocks) {
11799 write_unlock(&journal->j_state_lock);
11800 + if (current->plug)
11802 mutex_lock(&journal->j_checkpoint_mutex);
11805 diff --git a/fs/namespace.c b/fs/namespace.c
11806 index f26d18d69712..6e0e024715e8 100644
11807 --- a/fs/namespace.c
11808 +++ b/fs/namespace.c
11810 #include <linux/mnt_namespace.h>
11811 #include <linux/user_namespace.h>
11812 #include <linux/namei.h>
11813 +#include <linux/delay.h>
11814 #include <linux/security.h>
11815 #include <linux/idr.h>
11816 #include <linux/init.h> /* init_rootfs */
11817 @@ -356,8 +357,11 @@ int __mnt_want_write(struct vfsmount *m)
11818 * incremented count after it has set MNT_WRITE_HOLD.
11821 - while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
11823 + while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) {
11824 + preempt_enable();
11826 + preempt_disable();
11829 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
11830 * be set to match its requirements. So we must not load that until
11831 diff --git a/fs/ntfs/aops.c b/fs/ntfs/aops.c
11832 index 7521e11db728..f0de4b6b8bf3 100644
11833 --- a/fs/ntfs/aops.c
11834 +++ b/fs/ntfs/aops.c
11835 @@ -107,8 +107,7 @@ static void ntfs_end_buffer_async_read(struct buffer_head *bh, int uptodate)
11836 "0x%llx.", (unsigned long long)bh->b_blocknr);
11838 first = page_buffers(page);
11839 - local_irq_save(flags);
11840 - bit_spin_lock(BH_Uptodate_Lock, &first->b_state);
11841 + flags = bh_uptodate_lock_irqsave(first);
11842 clear_buffer_async_read(bh);
11845 @@ -123,8 +122,7 @@ static void ntfs_end_buffer_async_read(struct buffer_head *bh, int uptodate)
11847 tmp = tmp->b_this_page;
11848 } while (tmp != bh);
11849 - bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
11850 - local_irq_restore(flags);
11851 + bh_uptodate_unlock_irqrestore(first, flags);
11853 * If none of the buffers had errors then we can set the page uptodate,
11854 * but we first have to perform the post read mst fixups, if the
11855 @@ -145,13 +143,13 @@ static void ntfs_end_buffer_async_read(struct buffer_head *bh, int uptodate)
11856 recs = PAGE_CACHE_SIZE / rec_size;
11857 /* Should have been verified before we got here... */
11859 - local_irq_save(flags);
11860 + local_irq_save_nort(flags);
11861 kaddr = kmap_atomic(page);
11862 for (i = 0; i < recs; i++)
11863 post_read_mst_fixup((NTFS_RECORD*)(kaddr +
11864 i * rec_size), rec_size);
11865 kunmap_atomic(kaddr);
11866 - local_irq_restore(flags);
11867 + local_irq_restore_nort(flags);
11868 flush_dcache_page(page);
11869 if (likely(page_uptodate && !PageError(page)))
11870 SetPageUptodate(page);
11871 @@ -159,9 +157,7 @@ static void ntfs_end_buffer_async_read(struct buffer_head *bh, int uptodate)
11875 - bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
11876 - local_irq_restore(flags);
11878 + bh_uptodate_unlock_irqrestore(first, flags);
11882 diff --git a/fs/timerfd.c b/fs/timerfd.c
11883 index 1327a02ec778..4260febcb029 100644
11886 @@ -461,7 +461,10 @@ static int do_timerfd_settime(int ufd, int flags,
11889 spin_unlock_irq(&ctx->wqh.lock);
11891 + if (isalarm(ctx))
11892 + hrtimer_wait_for_timer(&ctx->t.alarm.timer);
11894 + hrtimer_wait_for_timer(&ctx->t.tmr);
11898 diff --git a/include/acpi/platform/aclinux.h b/include/acpi/platform/aclinux.h
11899 index 323e5daece54..cc5fbd534fd4 100644
11900 --- a/include/acpi/platform/aclinux.h
11901 +++ b/include/acpi/platform/aclinux.h
11902 @@ -127,6 +127,7 @@
11904 #define acpi_cache_t struct kmem_cache
11905 #define acpi_spinlock spinlock_t *
11906 +#define acpi_raw_spinlock raw_spinlock_t *
11907 #define acpi_cpu_flags unsigned long
11909 /* Use native linux version of acpi_os_allocate_zeroed */
11910 @@ -145,6 +146,20 @@
11911 #define ACPI_USE_ALTERNATE_PROTOTYPE_acpi_os_get_thread_id
11912 #define ACPI_USE_ALTERNATE_PROTOTYPE_acpi_os_create_lock
11914 +#define acpi_os_create_raw_lock(__handle) \
11916 + raw_spinlock_t *lock = ACPI_ALLOCATE(sizeof(*lock)); \
11919 + *(__handle) = lock; \
11920 + raw_spin_lock_init(*(__handle)); \
11922 + lock ? AE_OK : AE_NO_MEMORY; \
11925 +#define acpi_os_delete_raw_lock(__handle) kfree(__handle)
11929 * OSL interfaces used by debugger/disassembler
11931 diff --git a/include/asm-generic/bug.h b/include/asm-generic/bug.h
11932 index 630dd2372238..850e4d993a88 100644
11933 --- a/include/asm-generic/bug.h
11934 +++ b/include/asm-generic/bug.h
11935 @@ -206,6 +206,20 @@ extern void warn_slowpath_null(const char *file, const int line);
11936 # define WARN_ON_SMP(x) ({0;})
11939 +#ifdef CONFIG_PREEMPT_RT_BASE
11940 +# define BUG_ON_RT(c) BUG_ON(c)
11941 +# define BUG_ON_NONRT(c) do { } while (0)
11942 +# define WARN_ON_RT(condition) WARN_ON(condition)
11943 +# define WARN_ON_NONRT(condition) do { } while (0)
11944 +# define WARN_ON_ONCE_NONRT(condition) do { } while (0)
11946 +# define BUG_ON_RT(c) do { } while (0)
11947 +# define BUG_ON_NONRT(c) BUG_ON(c)
11948 +# define WARN_ON_RT(condition) do { } while (0)
11949 +# define WARN_ON_NONRT(condition) WARN_ON(condition)
11950 +# define WARN_ON_ONCE_NONRT(condition) WARN_ON_ONCE(condition)
11953 #endif /* __ASSEMBLY__ */
11956 diff --git a/include/asm-generic/preempt.h b/include/asm-generic/preempt.h
11957 index 5d8ffa3e6f8c..c1cde3577551 100644
11958 --- a/include/asm-generic/preempt.h
11959 +++ b/include/asm-generic/preempt.h
11962 static __always_inline int preempt_count(void)
11964 - return current_thread_info()->preempt_count;
11965 + return READ_ONCE(current_thread_info()->preempt_count);
11968 -static __always_inline int *preempt_count_ptr(void)
11969 +static __always_inline volatile int *preempt_count_ptr(void)
11971 return ¤t_thread_info()->preempt_count;
11973 diff --git a/include/linux/blk-mq.h b/include/linux/blk-mq.h
11974 index daf17d70aeca..463df8954255 100644
11975 --- a/include/linux/blk-mq.h
11976 +++ b/include/linux/blk-mq.h
11977 @@ -212,6 +212,7 @@ static inline u16 blk_mq_unique_tag_to_tag(u32 unique_tag)
11979 struct blk_mq_hw_ctx *blk_mq_map_queue(struct request_queue *, const int ctx_index);
11980 struct blk_mq_hw_ctx *blk_mq_alloc_single_hw_queue(struct blk_mq_tag_set *, unsigned int, int);
11981 +void __blk_mq_complete_request_remote_work(struct work_struct *work);
11983 int blk_mq_request_started(struct request *rq);
11984 void blk_mq_start_request(struct request *rq);
11985 diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
11986 index fe14382f9664..a82143ad6702 100644
11987 --- a/include/linux/blkdev.h
11988 +++ b/include/linux/blkdev.h
11989 @@ -89,6 +89,7 @@ struct request {
11990 struct list_head queuelist;
11992 struct call_single_data csd;
11993 + struct work_struct work;
11994 unsigned long fifo_time;
11997 @@ -455,7 +456,7 @@ struct request_queue {
11998 struct throtl_data *td;
12000 struct rcu_head rcu_head;
12001 - wait_queue_head_t mq_freeze_wq;
12002 + struct swait_queue_head mq_freeze_wq;
12003 struct percpu_ref q_usage_counter;
12004 struct list_head all_q_node;
12006 diff --git a/include/linux/bottom_half.h b/include/linux/bottom_half.h
12007 index 8fdcb783197d..d07dbeec7bc1 100644
12008 --- a/include/linux/bottom_half.h
12009 +++ b/include/linux/bottom_half.h
12012 #include <linux/preempt.h>
12014 +#ifdef CONFIG_PREEMPT_RT_FULL
12016 +extern void __local_bh_disable(void);
12017 +extern void _local_bh_enable(void);
12018 +extern void __local_bh_enable(void);
12020 +static inline void local_bh_disable(void)
12022 + __local_bh_disable();
12025 +static inline void __local_bh_disable_ip(unsigned long ip, unsigned int cnt)
12027 + __local_bh_disable();
12030 +static inline void local_bh_enable(void)
12032 + __local_bh_enable();
12035 +static inline void __local_bh_enable_ip(unsigned long ip, unsigned int cnt)
12037 + __local_bh_enable();
12040 +static inline void local_bh_enable_ip(unsigned long ip)
12042 + __local_bh_enable();
12047 #ifdef CONFIG_TRACE_IRQFLAGS
12048 extern void __local_bh_disable_ip(unsigned long ip, unsigned int cnt);
12050 @@ -30,5 +63,6 @@ static inline void local_bh_enable(void)
12052 __local_bh_enable_ip(_THIS_IP_, SOFTIRQ_DISABLE_OFFSET);
12056 #endif /* _LINUX_BH_H */
12057 diff --git a/include/linux/buffer_head.h b/include/linux/buffer_head.h
12058 index 89d9aa9e79bf..4a201008b02d 100644
12059 --- a/include/linux/buffer_head.h
12060 +++ b/include/linux/buffer_head.h
12061 @@ -75,8 +75,50 @@ struct buffer_head {
12062 struct address_space *b_assoc_map; /* mapping this buffer is
12064 atomic_t b_count; /* users using this buffer_head */
12065 +#ifdef CONFIG_PREEMPT_RT_BASE
12066 + spinlock_t b_uptodate_lock;
12067 +#if IS_ENABLED(CONFIG_JBD2)
12068 + spinlock_t b_state_lock;
12069 + spinlock_t b_journal_head_lock;
12074 +static inline unsigned long bh_uptodate_lock_irqsave(struct buffer_head *bh)
12076 + unsigned long flags;
12078 +#ifndef CONFIG_PREEMPT_RT_BASE
12079 + local_irq_save(flags);
12080 + bit_spin_lock(BH_Uptodate_Lock, &bh->b_state);
12082 + spin_lock_irqsave(&bh->b_uptodate_lock, flags);
12087 +static inline void
12088 +bh_uptodate_unlock_irqrestore(struct buffer_head *bh, unsigned long flags)
12090 +#ifndef CONFIG_PREEMPT_RT_BASE
12091 + bit_spin_unlock(BH_Uptodate_Lock, &bh->b_state);
12092 + local_irq_restore(flags);
12094 + spin_unlock_irqrestore(&bh->b_uptodate_lock, flags);
12098 +static inline void buffer_head_init_locks(struct buffer_head *bh)
12100 +#ifdef CONFIG_PREEMPT_RT_BASE
12101 + spin_lock_init(&bh->b_uptodate_lock);
12102 +#if IS_ENABLED(CONFIG_JBD2)
12103 + spin_lock_init(&bh->b_state_lock);
12104 + spin_lock_init(&bh->b_journal_head_lock);
12110 * macro tricks to expand the set_buffer_foo(), clear_buffer_foo()
12111 * and buffer_foo() functions.
12112 diff --git a/include/linux/cgroup-defs.h b/include/linux/cgroup-defs.h
12113 index 8da263299754..0cc474291e08 100644
12114 --- a/include/linux/cgroup-defs.h
12115 +++ b/include/linux/cgroup-defs.h
12117 #include <linux/percpu-refcount.h>
12118 #include <linux/percpu-rwsem.h>
12119 #include <linux/workqueue.h>
12120 +#include <linux/swork.h>
12122 #ifdef CONFIG_CGROUPS
12124 @@ -142,6 +143,7 @@ struct cgroup_subsys_state {
12125 /* percpu_ref killing and RCU release */
12126 struct rcu_head rcu_head;
12127 struct work_struct destroy_work;
12128 + struct swork_event destroy_swork;
12132 diff --git a/include/linux/clk/at91_pmc.h b/include/linux/clk/at91_pmc.h
12133 index 1e6932222e11..17f413bbbedf 100644
12134 --- a/include/linux/clk/at91_pmc.h
12135 +++ b/include/linux/clk/at91_pmc.h
12140 -#ifndef __ASSEMBLY__
12141 -extern void __iomem *at91_pmc_base;
12143 -#define at91_pmc_read(field) \
12144 - readl_relaxed(at91_pmc_base + field)
12146 -#define at91_pmc_write(field, value) \
12147 - writel_relaxed(value, at91_pmc_base + field)
12149 -.extern at91_pmc_base
12152 #define AT91_PMC_SCER 0x00 /* System Clock Enable Register */
12153 #define AT91_PMC_SCDR 0x04 /* System Clock Disable Register */
12155 diff --git a/include/linux/completion.h b/include/linux/completion.h
12156 index 5d5aaae3af43..3bca1590e29f 100644
12157 --- a/include/linux/completion.h
12158 +++ b/include/linux/completion.h
12160 * Atomic wait-for-completion handler data structures.
12161 * See kernel/sched/completion.c for details.
12164 -#include <linux/wait.h>
12165 +#include <linux/swait.h>
12168 * struct completion - structure used to maintain state for a "completion"
12169 @@ -24,11 +23,11 @@
12171 struct completion {
12173 - wait_queue_head_t wait;
12174 + struct swait_queue_head wait;
12177 #define COMPLETION_INITIALIZER(work) \
12178 - { 0, __WAIT_QUEUE_HEAD_INITIALIZER((work).wait) }
12179 + { 0, __SWAIT_QUEUE_HEAD_INITIALIZER((work).wait) }
12181 #define COMPLETION_INITIALIZER_ONSTACK(work) \
12182 ({ init_completion(&work); work; })
12183 @@ -73,7 +72,7 @@ struct completion {
12184 static inline void init_completion(struct completion *x)
12187 - init_waitqueue_head(&x->wait);
12188 + init_swait_queue_head(&x->wait);
12192 diff --git a/include/linux/cpu.h b/include/linux/cpu.h
12193 index 3ea9aae2387d..5de5f28a5ef3 100644
12194 --- a/include/linux/cpu.h
12195 +++ b/include/linux/cpu.h
12196 @@ -224,6 +224,8 @@ extern void get_online_cpus(void);
12197 extern void put_online_cpus(void);
12198 extern void cpu_hotplug_disable(void);
12199 extern void cpu_hotplug_enable(void);
12200 +extern void pin_current_cpu(void);
12201 +extern void unpin_current_cpu(void);
12202 #define hotcpu_notifier(fn, pri) cpu_notifier(fn, pri)
12203 #define __hotcpu_notifier(fn, pri) __cpu_notifier(fn, pri)
12204 #define register_hotcpu_notifier(nb) register_cpu_notifier(nb)
12205 @@ -241,6 +243,8 @@ static inline void cpu_hotplug_done(void) {}
12206 #define put_online_cpus() do { } while (0)
12207 #define cpu_hotplug_disable() do { } while (0)
12208 #define cpu_hotplug_enable() do { } while (0)
12209 +static inline void pin_current_cpu(void) { }
12210 +static inline void unpin_current_cpu(void) { }
12211 #define hotcpu_notifier(fn, pri) do { (void)(fn); } while (0)
12212 #define __hotcpu_notifier(fn, pri) do { (void)(fn); } while (0)
12213 /* These aren't inline functions due to a GCC bug. */
12214 diff --git a/include/linux/delay.h b/include/linux/delay.h
12215 index a6ecb34cf547..37caab306336 100644
12216 --- a/include/linux/delay.h
12217 +++ b/include/linux/delay.h
12218 @@ -52,4 +52,10 @@ static inline void ssleep(unsigned int seconds)
12219 msleep(seconds * 1000);
12222 +#ifdef CONFIG_PREEMPT_RT_FULL
12223 +extern void cpu_chill(void);
12225 +# define cpu_chill() cpu_relax()
12228 #endif /* defined(_LINUX_DELAY_H) */
12229 diff --git a/include/linux/ftrace.h b/include/linux/ftrace.h
12230 index 60048c50404e..f2cd67624f18 100644
12231 --- a/include/linux/ftrace.h
12232 +++ b/include/linux/ftrace.h
12233 @@ -694,6 +694,18 @@ static inline void __ftrace_enabled_restore(int enabled)
12234 #define CALLER_ADDR5 ((unsigned long)ftrace_return_address(5))
12235 #define CALLER_ADDR6 ((unsigned long)ftrace_return_address(6))
12237 +static inline unsigned long get_lock_parent_ip(void)
12239 + unsigned long addr = CALLER_ADDR0;
12241 + if (!in_lock_functions(addr))
12243 + addr = CALLER_ADDR1;
12244 + if (!in_lock_functions(addr))
12246 + return CALLER_ADDR2;
12249 #ifdef CONFIG_IRQSOFF_TRACER
12250 extern void time_hardirqs_on(unsigned long a0, unsigned long a1);
12251 extern void time_hardirqs_off(unsigned long a0, unsigned long a1);
12252 diff --git a/include/linux/highmem.h b/include/linux/highmem.h
12253 index bb3f3297062a..a117a33ef72c 100644
12254 --- a/include/linux/highmem.h
12255 +++ b/include/linux/highmem.h
12257 #include <linux/mm.h>
12258 #include <linux/uaccess.h>
12259 #include <linux/hardirq.h>
12260 +#include <linux/sched.h>
12262 #include <asm/cacheflush.h>
12264 @@ -65,7 +66,7 @@ static inline void kunmap(struct page *page)
12266 static inline void *kmap_atomic(struct page *page)
12268 - preempt_disable();
12269 + preempt_disable_nort();
12270 pagefault_disable();
12271 return page_address(page);
12273 @@ -74,7 +75,7 @@ static inline void *kmap_atomic(struct page *page)
12274 static inline void __kunmap_atomic(void *addr)
12276 pagefault_enable();
12277 - preempt_enable();
12278 + preempt_enable_nort();
12281 #define kmap_atomic_pfn(pfn) kmap_atomic(pfn_to_page(pfn))
12282 @@ -86,32 +87,51 @@ static inline void __kunmap_atomic(void *addr)
12284 #if defined(CONFIG_HIGHMEM) || defined(CONFIG_X86_32)
12286 +#ifndef CONFIG_PREEMPT_RT_FULL
12287 DECLARE_PER_CPU(int, __kmap_atomic_idx);
12290 static inline int kmap_atomic_idx_push(void)
12292 +#ifndef CONFIG_PREEMPT_RT_FULL
12293 int idx = __this_cpu_inc_return(__kmap_atomic_idx) - 1;
12295 -#ifdef CONFIG_DEBUG_HIGHMEM
12296 +# ifdef CONFIG_DEBUG_HIGHMEM
12297 WARN_ON_ONCE(in_irq() && !irqs_disabled());
12298 BUG_ON(idx >= KM_TYPE_NR);
12303 + current->kmap_idx++;
12304 + BUG_ON(current->kmap_idx > KM_TYPE_NR);
12305 + return current->kmap_idx - 1;
12309 static inline int kmap_atomic_idx(void)
12311 +#ifndef CONFIG_PREEMPT_RT_FULL
12312 return __this_cpu_read(__kmap_atomic_idx) - 1;
12314 + return current->kmap_idx - 1;
12318 static inline void kmap_atomic_idx_pop(void)
12320 -#ifdef CONFIG_DEBUG_HIGHMEM
12321 +#ifndef CONFIG_PREEMPT_RT_FULL
12322 +# ifdef CONFIG_DEBUG_HIGHMEM
12323 int idx = __this_cpu_dec_return(__kmap_atomic_idx);
12328 __this_cpu_dec(__kmap_atomic_idx);
12331 + current->kmap_idx--;
12332 +# ifdef CONFIG_DEBUG_HIGHMEM
12333 + BUG_ON(current->kmap_idx < 0);
12338 diff --git a/include/linux/hrtimer.h b/include/linux/hrtimer.h
12339 index 2ead22dd74a0..8fbcdfa5dc77 100644
12340 --- a/include/linux/hrtimer.h
12341 +++ b/include/linux/hrtimer.h
12342 @@ -87,6 +87,9 @@ enum hrtimer_restart {
12343 * @function: timer expiry callback function
12344 * @base: pointer to the timer base (per cpu and per clock)
12345 * @state: state information (See bit values above)
12346 + * @cb_entry: list entry to defer timers from hardirq context
12347 + * @irqsafe: timer can run in hardirq context
12348 + * @praecox: timer expiry time if expired at the time of programming
12349 * @is_rel: Set if the timer was armed relative
12350 * @start_pid: timer statistics field to store the pid of the task which
12351 * started the timer
12352 @@ -103,6 +106,11 @@ struct hrtimer {
12353 enum hrtimer_restart (*function)(struct hrtimer *);
12354 struct hrtimer_clock_base *base;
12356 + struct list_head cb_entry;
12358 +#ifdef CONFIG_MISSED_TIMER_OFFSETS_HIST
12362 #ifdef CONFIG_TIMER_STATS
12364 @@ -123,11 +131,7 @@ struct hrtimer_sleeper {
12365 struct task_struct *task;
12368 -#ifdef CONFIG_64BIT
12369 # define HRTIMER_CLOCK_BASE_ALIGN 64
12371 -# define HRTIMER_CLOCK_BASE_ALIGN 32
12375 * struct hrtimer_clock_base - the timer base for a specific clock
12376 @@ -136,6 +140,7 @@ struct hrtimer_sleeper {
12377 * timer to a base on another cpu.
12378 * @clockid: clock id for per_cpu support
12379 * @active: red black tree root node for the active timers
12380 + * @expired: list head for deferred timers.
12381 * @get_time: function to retrieve the current time of the clock
12382 * @offset: offset of this clock to the monotonic base
12384 @@ -144,6 +149,7 @@ struct hrtimer_clock_base {
12387 struct timerqueue_head active;
12388 + struct list_head expired;
12389 ktime_t (*get_time)(void);
12391 } __attribute__((__aligned__(HRTIMER_CLOCK_BASE_ALIGN)));
12392 @@ -187,6 +193,7 @@ struct hrtimer_cpu_base {
12393 raw_spinlock_t lock;
12395 struct hrtimer *running;
12396 + struct hrtimer *running_soft;
12398 unsigned int active_bases;
12399 unsigned int clock_was_set_seq;
12400 @@ -203,6 +210,9 @@ struct hrtimer_cpu_base {
12401 unsigned int nr_hangs;
12402 unsigned int max_hang_time;
12404 +#ifdef CONFIG_PREEMPT_RT_BASE
12405 + wait_queue_head_t wait;
12407 struct hrtimer_clock_base clock_base[HRTIMER_MAX_CLOCK_BASES];
12408 } ____cacheline_aligned;
12410 @@ -412,6 +422,13 @@ static inline void hrtimer_restart(struct hrtimer *timer)
12411 hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
12414 +/* Softirq preemption could deadlock timer removal */
12415 +#ifdef CONFIG_PREEMPT_RT_BASE
12416 + extern void hrtimer_wait_for_timer(const struct hrtimer *timer);
12418 +# define hrtimer_wait_for_timer(timer) do { cpu_relax(); } while (0)
12421 /* Query timers: */
12422 extern ktime_t __hrtimer_get_remaining(const struct hrtimer *timer, bool adjust);
12424 @@ -436,7 +453,7 @@ static inline int hrtimer_is_queued(struct hrtimer *timer)
12425 * Helper function to check, whether the timer is running the callback
12428 -static inline int hrtimer_callback_running(struct hrtimer *timer)
12429 +static inline int hrtimer_callback_running(const struct hrtimer *timer)
12431 return timer->base->cpu_base->running == timer;
12433 diff --git a/include/linux/idr.h b/include/linux/idr.h
12434 index 013fd9bc4cb6..f62be0aec911 100644
12435 --- a/include/linux/idr.h
12436 +++ b/include/linux/idr.h
12437 @@ -95,10 +95,14 @@ bool idr_is_empty(struct idr *idp);
12438 * Each idr_preload() should be matched with an invocation of this
12439 * function. See idr_preload() for details.
12441 +#ifdef CONFIG_PREEMPT_RT_FULL
12442 +void idr_preload_end(void);
12444 static inline void idr_preload_end(void)
12451 * idr_find - return pointer for given id
12452 diff --git a/include/linux/init_task.h b/include/linux/init_task.h
12453 index 1c1ff7e4faa4..60fadde71a44 100644
12454 --- a/include/linux/init_task.h
12455 +++ b/include/linux/init_task.h
12456 @@ -148,9 +148,15 @@ extern struct task_group root_task_group;
12457 # define INIT_PERF_EVENTS(tsk)
12460 +#ifdef CONFIG_PREEMPT_RT_BASE
12461 +# define INIT_TIMER_LIST .posix_timer_list = NULL,
12463 +# define INIT_TIMER_LIST
12466 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
12467 # define INIT_VTIME(tsk) \
12468 - .vtime_seqlock = __SEQLOCK_UNLOCKED(tsk.vtime_seqlock), \
12469 + .vtime_seqcount = SEQCNT_ZERO(tsk.vtime_seqcount), \
12471 .vtime_snap_whence = VTIME_SYS,
12473 @@ -239,6 +245,7 @@ extern struct task_group root_task_group;
12474 .cpu_timers = INIT_CPU_TIMERS(tsk.cpu_timers), \
12475 .pi_lock = __RAW_SPIN_LOCK_UNLOCKED(tsk.pi_lock), \
12476 .timer_slack_ns = 50000, /* 50 usec default slack */ \
12477 + INIT_TIMER_LIST \
12479 [PIDTYPE_PID] = INIT_PID_LINK(PIDTYPE_PID), \
12480 [PIDTYPE_PGID] = INIT_PID_LINK(PIDTYPE_PGID), \
12481 diff --git a/include/linux/interrupt.h b/include/linux/interrupt.h
12482 index ad16809c8596..655cee096aed 100644
12483 --- a/include/linux/interrupt.h
12484 +++ b/include/linux/interrupt.h
12486 * interrupt handler after suspending interrupts. For system
12487 * wakeup devices users need to implement wakeup detection in
12488 * their interrupt handlers.
12489 + * IRQF_NO_SOFTIRQ_CALL - Do not process softirqs in the irq thread context (RT)
12491 #define IRQF_SHARED 0x00000080
12492 #define IRQF_PROBE_SHARED 0x00000100
12494 #define IRQF_NO_THREAD 0x00010000
12495 #define IRQF_EARLY_RESUME 0x00020000
12496 #define IRQF_COND_SUSPEND 0x00040000
12497 +#define IRQF_NO_SOFTIRQ_CALL 0x00080000
12499 #define IRQF_TIMER (__IRQF_TIMER | IRQF_NO_SUSPEND | IRQF_NO_THREAD)
12501 @@ -186,7 +188,7 @@ extern void devm_free_irq(struct device *dev, unsigned int irq, void *dev_id);
12502 #ifdef CONFIG_LOCKDEP
12503 # define local_irq_enable_in_hardirq() do { } while (0)
12505 -# define local_irq_enable_in_hardirq() local_irq_enable()
12506 +# define local_irq_enable_in_hardirq() local_irq_enable_nort()
12509 extern void disable_irq_nosync(unsigned int irq);
12510 @@ -206,6 +208,7 @@ extern void resume_device_irqs(void);
12511 * @irq: Interrupt to which notification applies
12512 * @kref: Reference count, for internal use
12513 * @work: Work item, for internal use
12514 + * @list: List item for deferred callbacks
12515 * @notify: Function to be called on change. This will be
12516 * called in process context.
12517 * @release: Function to be called on release. This will be
12518 @@ -217,6 +220,7 @@ struct irq_affinity_notify {
12521 struct work_struct work;
12522 + struct list_head list;
12523 void (*notify)(struct irq_affinity_notify *, const cpumask_t *mask);
12524 void (*release)(struct kref *ref);
12526 @@ -379,9 +383,13 @@ extern int irq_set_irqchip_state(unsigned int irq, enum irqchip_irq_state which,
12529 #ifdef CONFIG_IRQ_FORCED_THREADING
12530 +# ifndef CONFIG_PREEMPT_RT_BASE
12531 extern bool force_irqthreads;
12533 +# define force_irqthreads (true)
12536 -#define force_irqthreads (0)
12537 +#define force_irqthreads (false)
12540 #ifndef __ARCH_SET_SOFTIRQ_PENDING
12541 @@ -438,9 +446,10 @@ struct softirq_action
12542 void (*action)(struct softirq_action *);
12545 +#ifndef CONFIG_PREEMPT_RT_FULL
12546 asmlinkage void do_softirq(void);
12547 asmlinkage void __do_softirq(void);
12549 +static inline void thread_do_softirq(void) { do_softirq(); }
12550 #ifdef __ARCH_HAS_DO_SOFTIRQ
12551 void do_softirq_own_stack(void);
12553 @@ -449,13 +458,25 @@ static inline void do_softirq_own_stack(void)
12558 +extern void thread_do_softirq(void);
12561 extern void open_softirq(int nr, void (*action)(struct softirq_action *));
12562 extern void softirq_init(void);
12563 extern void __raise_softirq_irqoff(unsigned int nr);
12564 +#ifdef CONFIG_PREEMPT_RT_FULL
12565 +extern void __raise_softirq_irqoff_ksoft(unsigned int nr);
12567 +static inline void __raise_softirq_irqoff_ksoft(unsigned int nr)
12569 + __raise_softirq_irqoff(nr);
12573 extern void raise_softirq_irqoff(unsigned int nr);
12574 extern void raise_softirq(unsigned int nr);
12575 +extern void softirq_check_pending_idle(void);
12577 DECLARE_PER_CPU(struct task_struct *, ksoftirqd);
12579 @@ -477,8 +498,9 @@ static inline struct task_struct *this_cpu_ksoftirqd(void)
12580 to be executed on some cpu at least once after this.
12581 * If the tasklet is already scheduled, but its execution is still not
12582 started, it will be executed only once.
12583 - * If this tasklet is already running on another CPU (or schedule is called
12584 - from tasklet itself), it is rescheduled for later.
12585 + * If this tasklet is already running on another CPU, it is rescheduled
12587 + * Schedule must not be called from the tasklet itself (a lockup occurs)
12588 * Tasklet is strictly serialized wrt itself, but not
12589 wrt another tasklets. If client needs some intertask synchronization,
12590 he makes it with spinlocks.
12591 @@ -503,27 +525,36 @@ struct tasklet_struct name = { NULL, 0, ATOMIC_INIT(1), func, data }
12594 TASKLET_STATE_SCHED, /* Tasklet is scheduled for execution */
12595 - TASKLET_STATE_RUN /* Tasklet is running (SMP only) */
12596 + TASKLET_STATE_RUN, /* Tasklet is running (SMP only) */
12597 + TASKLET_STATE_PENDING /* Tasklet is pending */
12601 +#define TASKLET_STATEF_SCHED (1 << TASKLET_STATE_SCHED)
12602 +#define TASKLET_STATEF_RUN (1 << TASKLET_STATE_RUN)
12603 +#define TASKLET_STATEF_PENDING (1 << TASKLET_STATE_PENDING)
12605 +#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT_RT_FULL)
12606 static inline int tasklet_trylock(struct tasklet_struct *t)
12608 return !test_and_set_bit(TASKLET_STATE_RUN, &(t)->state);
12611 +static inline int tasklet_tryunlock(struct tasklet_struct *t)
12613 + return cmpxchg(&t->state, TASKLET_STATEF_RUN, 0) == TASKLET_STATEF_RUN;
12616 static inline void tasklet_unlock(struct tasklet_struct *t)
12618 smp_mb__before_atomic();
12619 clear_bit(TASKLET_STATE_RUN, &(t)->state);
12622 -static inline void tasklet_unlock_wait(struct tasklet_struct *t)
12624 - while (test_bit(TASKLET_STATE_RUN, &(t)->state)) { barrier(); }
12626 +extern void tasklet_unlock_wait(struct tasklet_struct *t);
12629 #define tasklet_trylock(t) 1
12630 +#define tasklet_tryunlock(t) 1
12631 #define tasklet_unlock_wait(t) do { } while (0)
12632 #define tasklet_unlock(t) do { } while (0)
12634 @@ -572,12 +603,7 @@ static inline void tasklet_disable(struct tasklet_struct *t)
12638 -static inline void tasklet_enable(struct tasklet_struct *t)
12640 - smp_mb__before_atomic();
12641 - atomic_dec(&t->count);
12644 +extern void tasklet_enable(struct tasklet_struct *t);
12645 extern void tasklet_kill(struct tasklet_struct *t);
12646 extern void tasklet_kill_immediate(struct tasklet_struct *t, unsigned int cpu);
12647 extern void tasklet_init(struct tasklet_struct *t,
12648 @@ -608,6 +634,12 @@ void tasklet_hrtimer_cancel(struct tasklet_hrtimer *ttimer)
12649 tasklet_kill(&ttimer->tasklet);
12652 +#ifdef CONFIG_PREEMPT_RT_FULL
12653 +extern void softirq_early_init(void);
12655 +static inline void softirq_early_init(void) { }
12659 * Autoprobing for irqs:
12661 diff --git a/include/linux/irq.h b/include/linux/irq.h
12662 index f7cade00c525..dac9e11ba037 100644
12663 --- a/include/linux/irq.h
12664 +++ b/include/linux/irq.h
12665 @@ -72,6 +72,7 @@ enum irqchip_irq_state;
12666 * IRQ_IS_POLLED - Always polled by another interrupt. Exclude
12667 * it from the spurious interrupt detection
12668 * mechanism and from core side polling.
12669 + * IRQ_NO_SOFTIRQ_CALL - No softirq processing in the irq thread context (RT)
12670 * IRQ_DISABLE_UNLAZY - Disable lazy irq disable
12673 @@ -99,13 +100,14 @@ enum {
12674 IRQ_PER_CPU_DEVID = (1 << 17),
12675 IRQ_IS_POLLED = (1 << 18),
12676 IRQ_DISABLE_UNLAZY = (1 << 19),
12677 + IRQ_NO_SOFTIRQ_CALL = (1 << 20),
12680 #define IRQF_MODIFY_MASK \
12681 (IRQ_TYPE_SENSE_MASK | IRQ_NOPROBE | IRQ_NOREQUEST | \
12682 IRQ_NOAUTOEN | IRQ_MOVE_PCNTXT | IRQ_LEVEL | IRQ_NO_BALANCING | \
12683 IRQ_PER_CPU | IRQ_NESTED_THREAD | IRQ_NOTHREAD | IRQ_PER_CPU_DEVID | \
12684 - IRQ_IS_POLLED | IRQ_DISABLE_UNLAZY)
12685 + IRQ_IS_POLLED | IRQ_DISABLE_UNLAZY | IRQ_NO_SOFTIRQ_CALL)
12687 #define IRQ_NO_BALANCING_MASK (IRQ_PER_CPU | IRQ_NO_BALANCING)
12689 diff --git a/include/linux/irq_work.h b/include/linux/irq_work.h
12690 index 47b9ebd4a74f..2543aab05daa 100644
12691 --- a/include/linux/irq_work.h
12692 +++ b/include/linux/irq_work.h
12694 #define IRQ_WORK_BUSY 2UL
12695 #define IRQ_WORK_FLAGS 3UL
12696 #define IRQ_WORK_LAZY 4UL /* Doesn't want IPI, wait for tick */
12697 +#define IRQ_WORK_HARD_IRQ 8UL /* Run hard IRQ context, even on RT */
12700 unsigned long flags;
12701 @@ -51,4 +52,10 @@ static inline bool irq_work_needs_cpu(void) { return false; }
12702 static inline void irq_work_run(void) { }
12705 +#if defined(CONFIG_IRQ_WORK) && defined(CONFIG_PREEMPT_RT_FULL)
12706 +void irq_work_tick_soft(void);
12708 +static inline void irq_work_tick_soft(void) { }
12711 #endif /* _LINUX_IRQ_WORK_H */
12712 diff --git a/include/linux/irqdesc.h b/include/linux/irqdesc.h
12713 index a587a33363c7..ad57402a242d 100644
12714 --- a/include/linux/irqdesc.h
12715 +++ b/include/linux/irqdesc.h
12716 @@ -61,6 +61,7 @@ struct irq_desc {
12717 unsigned int irqs_unhandled;
12718 atomic_t threads_handled;
12719 int threads_handled_last;
12721 raw_spinlock_t lock;
12722 struct cpumask *percpu_enabled;
12724 diff --git a/include/linux/irqflags.h b/include/linux/irqflags.h
12725 index 5dd1272d1ab2..9b77034f7c5e 100644
12726 --- a/include/linux/irqflags.h
12727 +++ b/include/linux/irqflags.h
12729 # define trace_softirqs_enabled(p) ((p)->softirqs_enabled)
12730 # define trace_hardirq_enter() do { current->hardirq_context++; } while (0)
12731 # define trace_hardirq_exit() do { current->hardirq_context--; } while (0)
12732 -# define lockdep_softirq_enter() do { current->softirq_context++; } while (0)
12733 -# define lockdep_softirq_exit() do { current->softirq_context--; } while (0)
12734 # define INIT_TRACE_IRQFLAGS .softirqs_enabled = 1,
12736 # define trace_hardirqs_on() do { } while (0)
12738 # define trace_softirqs_enabled(p) 0
12739 # define trace_hardirq_enter() do { } while (0)
12740 # define trace_hardirq_exit() do { } while (0)
12741 +# define INIT_TRACE_IRQFLAGS
12744 +#if defined(CONFIG_TRACE_IRQFLAGS) && !defined(CONFIG_PREEMPT_RT_FULL)
12745 +# define lockdep_softirq_enter() do { current->softirq_context++; } while (0)
12746 +# define lockdep_softirq_exit() do { current->softirq_context--; } while (0)
12748 # define lockdep_softirq_enter() do { } while (0)
12749 # define lockdep_softirq_exit() do { } while (0)
12750 -# define INIT_TRACE_IRQFLAGS
12753 #if defined(CONFIG_IRQSOFF_TRACER) || \
12754 @@ -148,4 +152,23 @@
12756 #define irqs_disabled_flags(flags) raw_irqs_disabled_flags(flags)
12759 + * local_irq* variants depending on RT/!RT
12761 +#ifdef CONFIG_PREEMPT_RT_FULL
12762 +# define local_irq_disable_nort() do { } while (0)
12763 +# define local_irq_enable_nort() do { } while (0)
12764 +# define local_irq_save_nort(flags) local_save_flags(flags)
12765 +# define local_irq_restore_nort(flags) (void)(flags)
12766 +# define local_irq_disable_rt() local_irq_disable()
12767 +# define local_irq_enable_rt() local_irq_enable()
12769 +# define local_irq_disable_nort() local_irq_disable()
12770 +# define local_irq_enable_nort() local_irq_enable()
12771 +# define local_irq_save_nort(flags) local_irq_save(flags)
12772 +# define local_irq_restore_nort(flags) local_irq_restore(flags)
12773 +# define local_irq_disable_rt() do { } while (0)
12774 +# define local_irq_enable_rt() do { } while (0)
12778 diff --git a/include/linux/jbd2.h b/include/linux/jbd2.h
12779 index 65407f6c9120..eb5aabe4e18c 100644
12780 --- a/include/linux/jbd2.h
12781 +++ b/include/linux/jbd2.h
12782 @@ -352,32 +352,56 @@ static inline struct journal_head *bh2jh(struct buffer_head *bh)
12784 static inline void jbd_lock_bh_state(struct buffer_head *bh)
12786 +#ifndef CONFIG_PREEMPT_RT_BASE
12787 bit_spin_lock(BH_State, &bh->b_state);
12789 + spin_lock(&bh->b_state_lock);
12793 static inline int jbd_trylock_bh_state(struct buffer_head *bh)
12795 +#ifndef CONFIG_PREEMPT_RT_BASE
12796 return bit_spin_trylock(BH_State, &bh->b_state);
12798 + return spin_trylock(&bh->b_state_lock);
12802 static inline int jbd_is_locked_bh_state(struct buffer_head *bh)
12804 +#ifndef CONFIG_PREEMPT_RT_BASE
12805 return bit_spin_is_locked(BH_State, &bh->b_state);
12807 + return spin_is_locked(&bh->b_state_lock);
12811 static inline void jbd_unlock_bh_state(struct buffer_head *bh)
12813 +#ifndef CONFIG_PREEMPT_RT_BASE
12814 bit_spin_unlock(BH_State, &bh->b_state);
12816 + spin_unlock(&bh->b_state_lock);
12820 static inline void jbd_lock_bh_journal_head(struct buffer_head *bh)
12822 +#ifndef CONFIG_PREEMPT_RT_BASE
12823 bit_spin_lock(BH_JournalHead, &bh->b_state);
12825 + spin_lock(&bh->b_journal_head_lock);
12829 static inline void jbd_unlock_bh_journal_head(struct buffer_head *bh)
12831 +#ifndef CONFIG_PREEMPT_RT_BASE
12832 bit_spin_unlock(BH_JournalHead, &bh->b_state);
12834 + spin_unlock(&bh->b_journal_head_lock);
12838 #define J_ASSERT(assert) BUG_ON(!(assert))
12839 diff --git a/include/linux/kdb.h b/include/linux/kdb.h
12840 index a19bcf9e762e..897495386446 100644
12841 --- a/include/linux/kdb.h
12842 +++ b/include/linux/kdb.h
12843 @@ -167,6 +167,7 @@ extern __printf(2, 0) int vkdb_printf(enum kdb_msgsrc src, const char *fmt,
12844 extern __printf(1, 2) int kdb_printf(const char *, ...);
12845 typedef __printf(1, 2) int (*kdb_printf_t)(const char *, ...);
12847 +#define in_kdb_printk() (kdb_trap_printk)
12848 extern void kdb_init(int level);
12850 /* Access to kdb specific polling devices */
12851 @@ -201,6 +202,7 @@ extern int kdb_register_flags(char *, kdb_func_t, char *, char *,
12852 extern int kdb_unregister(char *);
12853 #else /* ! CONFIG_KGDB_KDB */
12854 static inline __printf(1, 2) int kdb_printf(const char *fmt, ...) { return 0; }
12855 +#define in_kdb_printk() (0)
12856 static inline void kdb_init(int level) {}
12857 static inline int kdb_register(char *cmd, kdb_func_t func, char *usage,
12858 char *help, short minlen) { return 0; }
12859 diff --git a/include/linux/kernel.h b/include/linux/kernel.h
12860 index 50220cab738c..d68f639f7330 100644
12861 --- a/include/linux/kernel.h
12862 +++ b/include/linux/kernel.h
12863 @@ -188,6 +188,9 @@ extern int _cond_resched(void);
12865 # define might_sleep() \
12866 do { __might_sleep(__FILE__, __LINE__, 0); might_resched(); } while (0)
12868 +# define might_sleep_no_state_check() \
12869 + do { ___might_sleep(__FILE__, __LINE__, 0); might_resched(); } while (0)
12870 # define sched_annotate_sleep() (current->task_state_change = 0)
12872 static inline void ___might_sleep(const char *file, int line,
12873 @@ -195,6 +198,7 @@ extern int _cond_resched(void);
12874 static inline void __might_sleep(const char *file, int line,
12875 int preempt_offset) { }
12876 # define might_sleep() do { might_resched(); } while (0)
12877 +# define might_sleep_no_state_check() do { might_resched(); } while (0)
12878 # define sched_annotate_sleep() do { } while (0)
12881 @@ -255,6 +259,7 @@ extern long (*panic_blink)(int state);
12883 void panic(const char *fmt, ...)
12885 +void nmi_panic(struct pt_regs *regs, const char *msg);
12886 extern void oops_enter(void);
12887 extern void oops_exit(void);
12888 void print_oops_end_marker(void);
12889 @@ -448,6 +453,14 @@ extern int sysctl_panic_on_stackoverflow;
12890 extern bool crash_kexec_post_notifiers;
12893 + * panic_cpu is used for synchronizing panic() and crash_kexec() execution. It
12894 + * holds a CPU number which is executing panic() currently. A value of
12895 + * PANIC_CPU_INVALID means no CPU has entered panic() or crash_kexec().
12897 +extern atomic_t panic_cpu;
12898 +#define PANIC_CPU_INVALID -1
12901 * Only to be used by arch init code. If the user over-wrote the default
12902 * CONFIG_PANIC_TIMEOUT, honor it.
12904 @@ -475,6 +488,7 @@ extern enum system_states {
12911 #define TAINT_PROPRIETARY_MODULE 0
12912 diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h
12913 index d7ce4e3280db..0b6d392b38e7 100644
12914 --- a/include/linux/kvm_host.h
12915 +++ b/include/linux/kvm_host.h
12917 #include <linux/irqflags.h>
12918 #include <linux/context_tracking.h>
12919 #include <linux/irqbypass.h>
12920 +#include <linux/swait.h>
12921 #include <asm/signal.h>
12923 #include <linux/kvm.h>
12924 @@ -243,7 +244,7 @@ struct kvm_vcpu {
12926 int guest_fpu_loaded, guest_xcr0_loaded;
12927 unsigned char fpu_counter;
12928 - wait_queue_head_t wq;
12929 + struct swait_queue_head wq;
12933 @@ -794,7 +795,7 @@ static inline bool kvm_arch_has_assigned_device(struct kvm *kvm)
12937 -static inline wait_queue_head_t *kvm_arch_vcpu_wq(struct kvm_vcpu *vcpu)
12938 +static inline struct swait_queue_head *kvm_arch_vcpu_wq(struct kvm_vcpu *vcpu)
12940 #ifdef __KVM_HAVE_ARCH_WQP
12941 return vcpu->arch.wqp;
12942 diff --git a/include/linux/lglock.h b/include/linux/lglock.h
12943 index c92ebd100d9b..6f035f635d0e 100644
12944 --- a/include/linux/lglock.h
12945 +++ b/include/linux/lglock.h
12946 @@ -34,13 +34,30 @@
12950 +#ifdef CONFIG_PREEMPT_RT_FULL
12951 + struct rt_mutex __percpu *lock;
12953 arch_spinlock_t __percpu *lock;
12955 #ifdef CONFIG_DEBUG_LOCK_ALLOC
12956 struct lock_class_key lock_key;
12957 struct lockdep_map lock_dep_map;
12961 +#ifdef CONFIG_PREEMPT_RT_FULL
12962 +# define DEFINE_LGLOCK(name) \
12963 + static DEFINE_PER_CPU(struct rt_mutex, name ## _lock) \
12964 + = __RT_MUTEX_INITIALIZER( name ## _lock); \
12965 + struct lglock name = { .lock = &name ## _lock }
12967 +# define DEFINE_STATIC_LGLOCK(name) \
12968 + static DEFINE_PER_CPU(struct rt_mutex, name ## _lock) \
12969 + = __RT_MUTEX_INITIALIZER( name ## _lock); \
12970 + static struct lglock name = { .lock = &name ## _lock }
12974 #define DEFINE_LGLOCK(name) \
12975 static DEFINE_PER_CPU(arch_spinlock_t, name ## _lock) \
12976 = __ARCH_SPIN_LOCK_UNLOCKED; \
12977 @@ -50,6 +67,7 @@ struct lglock {
12978 static DEFINE_PER_CPU(arch_spinlock_t, name ## _lock) \
12979 = __ARCH_SPIN_LOCK_UNLOCKED; \
12980 static struct lglock name = { .lock = &name ## _lock }
12983 void lg_lock_init(struct lglock *lg, char *name);
12985 @@ -64,6 +82,12 @@ void lg_double_unlock(struct lglock *lg, int cpu1, int cpu2);
12986 void lg_global_lock(struct lglock *lg);
12987 void lg_global_unlock(struct lglock *lg);
12989 +#ifndef CONFIG_PREEMPT_RT_FULL
12990 +#define lg_global_trylock_relax(name) lg_global_lock(name)
12992 +void lg_global_trylock_relax(struct lglock *lg);
12996 /* When !CONFIG_SMP, map lglock to spinlock */
12997 #define lglock spinlock
12998 diff --git a/include/linux/list_bl.h b/include/linux/list_bl.h
12999 index 8132214e8efd..89ffaa7bd342 100644
13000 --- a/include/linux/list_bl.h
13001 +++ b/include/linux/list_bl.h
13003 #define _LINUX_LIST_BL_H
13005 #include <linux/list.h>
13006 +#include <linux/spinlock.h>
13007 #include <linux/bit_spinlock.h>
13010 @@ -32,13 +33,24 @@
13012 struct hlist_bl_head {
13013 struct hlist_bl_node *first;
13014 +#ifdef CONFIG_PREEMPT_RT_BASE
13015 + raw_spinlock_t lock;
13019 struct hlist_bl_node {
13020 struct hlist_bl_node *next, **pprev;
13022 -#define INIT_HLIST_BL_HEAD(ptr) \
13023 - ((ptr)->first = NULL)
13025 +#ifdef CONFIG_PREEMPT_RT_BASE
13026 +#define INIT_HLIST_BL_HEAD(h) \
13028 + (h)->first = NULL; \
13029 + raw_spin_lock_init(&(h)->lock); \
13032 +#define INIT_HLIST_BL_HEAD(h) (h)->first = NULL
13035 static inline void INIT_HLIST_BL_NODE(struct hlist_bl_node *h)
13037 @@ -118,12 +130,26 @@ static inline void hlist_bl_del_init(struct hlist_bl_node *n)
13039 static inline void hlist_bl_lock(struct hlist_bl_head *b)
13041 +#ifndef CONFIG_PREEMPT_RT_BASE
13042 bit_spin_lock(0, (unsigned long *)b);
13044 + raw_spin_lock(&b->lock);
13045 +#if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK)
13046 + __set_bit(0, (unsigned long *)b);
13051 static inline void hlist_bl_unlock(struct hlist_bl_head *b)
13053 +#ifndef CONFIG_PREEMPT_RT_BASE
13054 __bit_spin_unlock(0, (unsigned long *)b);
13056 +#if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK)
13057 + __clear_bit(0, (unsigned long *)b);
13059 + raw_spin_unlock(&b->lock);
13063 static inline bool hlist_bl_is_locked(struct hlist_bl_head *b)
13064 diff --git a/include/linux/locallock.h b/include/linux/locallock.h
13065 new file mode 100644
13066 index 000000000000..e572a3971631
13068 +++ b/include/linux/locallock.h
13070 +#ifndef _LINUX_LOCALLOCK_H
13071 +#define _LINUX_LOCALLOCK_H
13073 +#include <linux/percpu.h>
13074 +#include <linux/spinlock.h>
13076 +#ifdef CONFIG_PREEMPT_RT_BASE
13078 +#ifdef CONFIG_DEBUG_SPINLOCK
13079 +# define LL_WARN(cond) WARN_ON(cond)
13081 +# define LL_WARN(cond) do { } while (0)
13085 + * per cpu lock based substitute for local_irq_*()
13087 +struct local_irq_lock {
13089 + struct task_struct *owner;
13091 + unsigned long flags;
13094 +#define DEFINE_LOCAL_IRQ_LOCK(lvar) \
13095 + DEFINE_PER_CPU(struct local_irq_lock, lvar) = { \
13096 + .lock = __SPIN_LOCK_UNLOCKED((lvar).lock) }
13098 +#define DECLARE_LOCAL_IRQ_LOCK(lvar) \
13099 + DECLARE_PER_CPU(struct local_irq_lock, lvar)
13101 +#define local_irq_lock_init(lvar) \
13104 + for_each_possible_cpu(__cpu) \
13105 + spin_lock_init(&per_cpu(lvar, __cpu).lock); \
13109 + * spin_lock|trylock|unlock_local flavour that does not migrate disable
13110 + * used for __local_lock|trylock|unlock where get_local_var/put_local_var
13111 + * already takes care of the migrate_disable/enable
13112 + * for CONFIG_PREEMPT_BASE map to the normal spin_* calls.
13114 +#ifdef CONFIG_PREEMPT_RT_FULL
13115 +# define spin_lock_local(lock) rt_spin_lock__no_mg(lock)
13116 +# define spin_trylock_local(lock) rt_spin_trylock__no_mg(lock)
13117 +# define spin_unlock_local(lock) rt_spin_unlock__no_mg(lock)
13119 +# define spin_lock_local(lock) spin_lock(lock)
13120 +# define spin_trylock_local(lock) spin_trylock(lock)
13121 +# define spin_unlock_local(lock) spin_unlock(lock)
13124 +static inline void __local_lock(struct local_irq_lock *lv)
13126 + if (lv->owner != current) {
13127 + spin_lock_local(&lv->lock);
13128 + LL_WARN(lv->owner);
13129 + LL_WARN(lv->nestcnt);
13130 + lv->owner = current;
13135 +#define local_lock(lvar) \
13136 + do { __local_lock(&get_local_var(lvar)); } while (0)
13138 +#define local_lock_on(lvar, cpu) \
13139 + do { __local_lock(&per_cpu(lvar, cpu)); } while (0)
13141 +static inline int __local_trylock(struct local_irq_lock *lv)
13143 + if (lv->owner != current && spin_trylock_local(&lv->lock)) {
13144 + LL_WARN(lv->owner);
13145 + LL_WARN(lv->nestcnt);
13146 + lv->owner = current;
13153 +#define local_trylock(lvar) \
13156 + __locked = __local_trylock(&get_local_var(lvar)); \
13158 + put_local_var(lvar); \
13162 +static inline void __local_unlock(struct local_irq_lock *lv)
13164 + LL_WARN(lv->nestcnt == 0);
13165 + LL_WARN(lv->owner != current);
13166 + if (--lv->nestcnt)
13169 + lv->owner = NULL;
13170 + spin_unlock_local(&lv->lock);
13173 +#define local_unlock(lvar) \
13175 + __local_unlock(this_cpu_ptr(&lvar)); \
13176 + put_local_var(lvar); \
13179 +#define local_unlock_on(lvar, cpu) \
13180 + do { __local_unlock(&per_cpu(lvar, cpu)); } while (0)
13182 +static inline void __local_lock_irq(struct local_irq_lock *lv)
13184 + spin_lock_irqsave(&lv->lock, lv->flags);
13185 + LL_WARN(lv->owner);
13186 + LL_WARN(lv->nestcnt);
13187 + lv->owner = current;
13191 +#define local_lock_irq(lvar) \
13192 + do { __local_lock_irq(&get_local_var(lvar)); } while (0)
13194 +#define local_lock_irq_on(lvar, cpu) \
13195 + do { __local_lock_irq(&per_cpu(lvar, cpu)); } while (0)
13197 +static inline void __local_unlock_irq(struct local_irq_lock *lv)
13199 + LL_WARN(!lv->nestcnt);
13200 + LL_WARN(lv->owner != current);
13201 + lv->owner = NULL;
13203 + spin_unlock_irq(&lv->lock);
13206 +#define local_unlock_irq(lvar) \
13208 + __local_unlock_irq(this_cpu_ptr(&lvar)); \
13209 + put_local_var(lvar); \
13212 +#define local_unlock_irq_on(lvar, cpu) \
13214 + __local_unlock_irq(&per_cpu(lvar, cpu)); \
13217 +static inline int __local_lock_irqsave(struct local_irq_lock *lv)
13219 + if (lv->owner != current) {
13220 + __local_lock_irq(lv);
13228 +#define local_lock_irqsave(lvar, _flags) \
13230 + if (__local_lock_irqsave(&get_local_var(lvar))) \
13231 + put_local_var(lvar); \
13232 + _flags = __this_cpu_read(lvar.flags); \
13235 +#define local_lock_irqsave_on(lvar, _flags, cpu) \
13237 + __local_lock_irqsave(&per_cpu(lvar, cpu)); \
13238 + _flags = per_cpu(lvar, cpu).flags; \
13241 +static inline int __local_unlock_irqrestore(struct local_irq_lock *lv,
13242 + unsigned long flags)
13244 + LL_WARN(!lv->nestcnt);
13245 + LL_WARN(lv->owner != current);
13246 + if (--lv->nestcnt)
13249 + lv->owner = NULL;
13250 + spin_unlock_irqrestore(&lv->lock, lv->flags);
13254 +#define local_unlock_irqrestore(lvar, flags) \
13256 + if (__local_unlock_irqrestore(this_cpu_ptr(&lvar), flags)) \
13257 + put_local_var(lvar); \
13260 +#define local_unlock_irqrestore_on(lvar, flags, cpu) \
13262 + __local_unlock_irqrestore(&per_cpu(lvar, cpu), flags); \
13265 +#define local_spin_trylock_irq(lvar, lock) \
13268 + local_lock_irq(lvar); \
13269 + __locked = spin_trylock(lock); \
13271 + local_unlock_irq(lvar); \
13275 +#define local_spin_lock_irq(lvar, lock) \
13277 + local_lock_irq(lvar); \
13278 + spin_lock(lock); \
13281 +#define local_spin_unlock_irq(lvar, lock) \
13283 + spin_unlock(lock); \
13284 + local_unlock_irq(lvar); \
13287 +#define local_spin_lock_irqsave(lvar, lock, flags) \
13289 + local_lock_irqsave(lvar, flags); \
13290 + spin_lock(lock); \
13293 +#define local_spin_unlock_irqrestore(lvar, lock, flags) \
13295 + spin_unlock(lock); \
13296 + local_unlock_irqrestore(lvar, flags); \
13299 +#define get_locked_var(lvar, var) \
13301 + local_lock(lvar); \
13302 + this_cpu_ptr(&var); \
13305 +#define put_locked_var(lvar, var) local_unlock(lvar);
13307 +#define local_lock_cpu(lvar) \
13309 + local_lock(lvar); \
13310 + smp_processor_id(); \
13313 +#define local_unlock_cpu(lvar) local_unlock(lvar)
13315 +#else /* PREEMPT_RT_BASE */
13317 +#define DEFINE_LOCAL_IRQ_LOCK(lvar) __typeof__(const int) lvar
13318 +#define DECLARE_LOCAL_IRQ_LOCK(lvar) extern __typeof__(const int) lvar
13320 +static inline void local_irq_lock_init(int lvar) { }
13322 +#define local_lock(lvar) preempt_disable()
13323 +#define local_unlock(lvar) preempt_enable()
13324 +#define local_lock_irq(lvar) local_irq_disable()
13325 +#define local_unlock_irq(lvar) local_irq_enable()
13326 +#define local_lock_irqsave(lvar, flags) local_irq_save(flags)
13327 +#define local_unlock_irqrestore(lvar, flags) local_irq_restore(flags)
13329 +#define local_spin_trylock_irq(lvar, lock) spin_trylock_irq(lock)
13330 +#define local_spin_lock_irq(lvar, lock) spin_lock_irq(lock)
13331 +#define local_spin_unlock_irq(lvar, lock) spin_unlock_irq(lock)
13332 +#define local_spin_lock_irqsave(lvar, lock, flags) \
13333 + spin_lock_irqsave(lock, flags)
13334 +#define local_spin_unlock_irqrestore(lvar, lock, flags) \
13335 + spin_unlock_irqrestore(lock, flags)
13337 +#define get_locked_var(lvar, var) get_cpu_var(var)
13338 +#define put_locked_var(lvar, var) put_cpu_var(var)
13340 +#define local_lock_cpu(lvar) get_cpu()
13341 +#define local_unlock_cpu(lvar) put_cpu()
13346 diff --git a/include/linux/mm_types.h b/include/linux/mm_types.h
13347 index 2ccccbfcd532..4b016eb6265b 100644
13348 --- a/include/linux/mm_types.h
13349 +++ b/include/linux/mm_types.h
13351 #include <linux/completion.h>
13352 #include <linux/cpumask.h>
13353 #include <linux/uprobes.h>
13354 +#include <linux/rcupdate.h>
13355 #include <linux/page-flags-layout.h>
13356 #include <asm/page.h>
13357 #include <asm/mmu.h>
13358 @@ -505,6 +506,9 @@ struct mm_struct {
13359 bool tlb_flush_pending;
13361 struct uprobes_state uprobes_state;
13362 +#ifdef CONFIG_PREEMPT_RT_BASE
13363 + struct rcu_head delayed_drop;
13365 #ifdef CONFIG_X86_INTEL_MPX
13366 /* address of the bounds directory */
13367 void __user *bd_addr;
13368 diff --git a/include/linux/module.h b/include/linux/module.h
13369 index b229a9961d02..5fea847cf95c 100644
13370 --- a/include/linux/module.h
13371 +++ b/include/linux/module.h
13372 @@ -500,6 +500,7 @@ static inline int module_is_live(struct module *mod)
13373 struct module *__module_text_address(unsigned long addr);
13374 struct module *__module_address(unsigned long addr);
13375 bool is_module_address(unsigned long addr);
13376 +bool __is_module_percpu_address(unsigned long addr, unsigned long *can_addr);
13377 bool is_module_percpu_address(unsigned long addr);
13378 bool is_module_text_address(unsigned long addr);
13380 @@ -665,6 +666,11 @@ static inline bool is_module_percpu_address(unsigned long addr)
13384 +static inline bool __is_module_percpu_address(unsigned long addr, unsigned long *can_addr)
13389 static inline bool is_module_text_address(unsigned long addr)
13392 diff --git a/include/linux/mutex.h b/include/linux/mutex.h
13393 index 2cb7531e7d7a..b3fdfc820216 100644
13394 --- a/include/linux/mutex.h
13395 +++ b/include/linux/mutex.h
13397 #include <asm/processor.h>
13398 #include <linux/osq_lock.h>
13400 +#ifdef CONFIG_DEBUG_LOCK_ALLOC
13401 +# define __DEP_MAP_MUTEX_INITIALIZER(lockname) \
13402 + , .dep_map = { .name = #lockname }
13404 +# define __DEP_MAP_MUTEX_INITIALIZER(lockname)
13407 +#ifdef CONFIG_PREEMPT_RT_FULL
13408 +# include <linux/mutex_rt.h>
13412 * Simple, straightforward mutexes with strict semantics:
13414 @@ -99,13 +110,6 @@ do { \
13415 static inline void mutex_destroy(struct mutex *lock) {}
13418 -#ifdef CONFIG_DEBUG_LOCK_ALLOC
13419 -# define __DEP_MAP_MUTEX_INITIALIZER(lockname) \
13420 - , .dep_map = { .name = #lockname }
13422 -# define __DEP_MAP_MUTEX_INITIALIZER(lockname)
13425 #define __MUTEX_INITIALIZER(lockname) \
13426 { .count = ATOMIC_INIT(1) \
13427 , .wait_lock = __SPIN_LOCK_UNLOCKED(lockname.wait_lock) \
13428 @@ -173,6 +177,8 @@ extern int __must_check mutex_lock_killable(struct mutex *lock);
13429 extern int mutex_trylock(struct mutex *lock);
13430 extern void mutex_unlock(struct mutex *lock);
13432 +#endif /* !PREEMPT_RT_FULL */
13434 extern int atomic_dec_and_mutex_lock(atomic_t *cnt, struct mutex *lock);
13436 #endif /* __LINUX_MUTEX_H */
13437 diff --git a/include/linux/mutex_rt.h b/include/linux/mutex_rt.h
13438 new file mode 100644
13439 index 000000000000..e0284edec655
13441 +++ b/include/linux/mutex_rt.h
13443 +#ifndef __LINUX_MUTEX_RT_H
13444 +#define __LINUX_MUTEX_RT_H
13446 +#ifndef __LINUX_MUTEX_H
13447 +#error "Please include mutex.h"
13450 +#include <linux/rtmutex.h>
13452 +/* FIXME: Just for __lockfunc */
13453 +#include <linux/spinlock.h>
13456 + struct rt_mutex lock;
13457 +#ifdef CONFIG_DEBUG_LOCK_ALLOC
13458 + struct lockdep_map dep_map;
13462 +#define __MUTEX_INITIALIZER(mutexname) \
13464 + .lock = __RT_MUTEX_INITIALIZER(mutexname.lock) \
13465 + __DEP_MAP_MUTEX_INITIALIZER(mutexname) \
13468 +#define DEFINE_MUTEX(mutexname) \
13469 + struct mutex mutexname = __MUTEX_INITIALIZER(mutexname)
13471 +extern void __mutex_do_init(struct mutex *lock, const char *name, struct lock_class_key *key);
13472 +extern void __lockfunc _mutex_lock(struct mutex *lock);
13473 +extern int __lockfunc _mutex_lock_interruptible(struct mutex *lock);
13474 +extern int __lockfunc _mutex_lock_killable(struct mutex *lock);
13475 +extern void __lockfunc _mutex_lock_nested(struct mutex *lock, int subclass);
13476 +extern void __lockfunc _mutex_lock_nest_lock(struct mutex *lock, struct lockdep_map *nest_lock);
13477 +extern int __lockfunc _mutex_lock_interruptible_nested(struct mutex *lock, int subclass);
13478 +extern int __lockfunc _mutex_lock_killable_nested(struct mutex *lock, int subclass);
13479 +extern int __lockfunc _mutex_trylock(struct mutex *lock);
13480 +extern void __lockfunc _mutex_unlock(struct mutex *lock);
13482 +#define mutex_is_locked(l) rt_mutex_is_locked(&(l)->lock)
13483 +#define mutex_lock(l) _mutex_lock(l)
13484 +#define mutex_lock_interruptible(l) _mutex_lock_interruptible(l)
13485 +#define mutex_lock_killable(l) _mutex_lock_killable(l)
13486 +#define mutex_trylock(l) _mutex_trylock(l)
13487 +#define mutex_unlock(l) _mutex_unlock(l)
13489 +#ifdef CONFIG_DEBUG_MUTEXES
13490 +#define mutex_destroy(l) rt_mutex_destroy(&(l)->lock)
13492 +static inline void mutex_destroy(struct mutex *lock) {}
13495 +#ifdef CONFIG_DEBUG_LOCK_ALLOC
13496 +# define mutex_lock_nested(l, s) _mutex_lock_nested(l, s)
13497 +# define mutex_lock_interruptible_nested(l, s) \
13498 + _mutex_lock_interruptible_nested(l, s)
13499 +# define mutex_lock_killable_nested(l, s) \
13500 + _mutex_lock_killable_nested(l, s)
13502 +# define mutex_lock_nest_lock(lock, nest_lock) \
13504 + typecheck(struct lockdep_map *, &(nest_lock)->dep_map); \
13505 + _mutex_lock_nest_lock(lock, &(nest_lock)->dep_map); \
13509 +# define mutex_lock_nested(l, s) _mutex_lock(l)
13510 +# define mutex_lock_interruptible_nested(l, s) \
13511 + _mutex_lock_interruptible(l)
13512 +# define mutex_lock_killable_nested(l, s) \
13513 + _mutex_lock_killable(l)
13514 +# define mutex_lock_nest_lock(lock, nest_lock) mutex_lock(lock)
13517 +# define mutex_init(mutex) \
13519 + static struct lock_class_key __key; \
13521 + rt_mutex_init(&(mutex)->lock); \
13522 + __mutex_do_init((mutex), #mutex, &__key); \
13525 +# define __mutex_init(mutex, name, key) \
13527 + rt_mutex_init(&(mutex)->lock); \
13528 + __mutex_do_init((mutex), name, key); \
13532 diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
13533 index 4035bbe40971..81ebe70d107d 100644
13534 --- a/include/linux/netdevice.h
13535 +++ b/include/linux/netdevice.h
13536 @@ -390,7 +390,19 @@ typedef enum rx_handler_result rx_handler_result_t;
13537 typedef rx_handler_result_t rx_handler_func_t(struct sk_buff **pskb);
13539 void __napi_schedule(struct napi_struct *n);
13542 + * When PREEMPT_RT_FULL is defined, all device interrupt handlers
13543 + * run as threads, and they can also be preempted (without PREEMPT_RT
13544 + * interrupt threads can not be preempted). Which means that calling
13545 + * __napi_schedule_irqoff() from an interrupt handler can be preempted
13546 + * and can corrupt the napi->poll_list.
13548 +#ifdef CONFIG_PREEMPT_RT_FULL
13549 +#define __napi_schedule_irqoff(n) __napi_schedule(n)
13551 void __napi_schedule_irqoff(struct napi_struct *n);
13554 static inline bool napi_disable_pending(struct napi_struct *n)
13556 @@ -2288,11 +2300,20 @@ void netdev_freemem(struct net_device *dev);
13557 void synchronize_net(void);
13558 int init_dummy_netdev(struct net_device *dev);
13560 +#ifdef CONFIG_PREEMPT_RT_FULL
13561 +static inline int dev_recursion_level(void)
13563 + return current->xmit_recursion;
13568 DECLARE_PER_CPU(int, xmit_recursion);
13569 static inline int dev_recursion_level(void)
13571 return this_cpu_read(xmit_recursion);
13575 struct net_device *dev_get_by_index(struct net *net, int ifindex);
13576 struct net_device *__dev_get_by_index(struct net *net, int ifindex);
13577 @@ -2610,6 +2631,7 @@ struct softnet_data {
13578 unsigned int dropped;
13579 struct sk_buff_head input_pkt_queue;
13580 struct napi_struct backlog;
13581 + struct sk_buff_head tofree_queue;
13585 diff --git a/include/linux/netfilter/x_tables.h b/include/linux/netfilter/x_tables.h
13586 index 04078e8a4803..a61c9609e32f 100644
13587 --- a/include/linux/netfilter/x_tables.h
13588 +++ b/include/linux/netfilter/x_tables.h
13591 #include <linux/netdevice.h>
13592 #include <linux/static_key.h>
13593 +#include <linux/locallock.h>
13594 #include <uapi/linux/netfilter/x_tables.h>
13597 @@ -289,6 +290,8 @@ void xt_free_table_info(struct xt_table_info *info);
13599 DECLARE_PER_CPU(seqcount_t, xt_recseq);
13601 +DECLARE_LOCAL_IRQ_LOCK(xt_write_lock);
13603 /* xt_tee_enabled - true if x_tables needs to handle reentrancy
13605 * Enabled if current ip(6)tables ruleset has at least one -j TEE rule.
13606 @@ -309,6 +312,9 @@ static inline unsigned int xt_write_recseq_begin(void)
13608 unsigned int addend;
13610 + /* RT protection */
13611 + local_lock(xt_write_lock);
13614 * Low order bit of sequence is set if we already
13615 * called xt_write_recseq_begin().
13616 @@ -339,6 +345,7 @@ static inline void xt_write_recseq_end(unsigned int addend)
13617 /* this is kind of a write_seqcount_end(), but addend is 0 or 1 */
13619 __this_cpu_add(xt_recseq.sequence, addend);
13620 + local_unlock(xt_write_lock);
13624 diff --git a/include/linux/notifier.h b/include/linux/notifier.h
13625 index d14a4c362465..2e4414a0c1c4 100644
13626 --- a/include/linux/notifier.h
13627 +++ b/include/linux/notifier.h
13630 * Alan Cox <Alan.Cox@linux.org>
13634 #ifndef _LINUX_NOTIFIER_H
13635 #define _LINUX_NOTIFIER_H
13636 #include <linux/errno.h>
13638 * in srcu_notifier_call_chain(): no cache bounces and no memory barriers.
13639 * As compensation, srcu_notifier_chain_unregister() is rather expensive.
13640 * SRCU notifier chains should be used when the chain will be called very
13641 - * often but notifier_blocks will seldom be removed. Also, SRCU notifier
13642 - * chains are slightly more difficult to use because they require special
13643 - * runtime initialization.
13644 + * often but notifier_blocks will seldom be removed.
13647 typedef int (*notifier_fn_t)(struct notifier_block *nb,
13648 @@ -88,7 +86,7 @@ struct srcu_notifier_head {
13649 (name)->head = NULL; \
13652 -/* srcu_notifier_heads must be initialized and cleaned up dynamically */
13653 +/* srcu_notifier_heads must be cleaned up dynamically */
13654 extern void srcu_init_notifier_head(struct srcu_notifier_head *nh);
13655 #define srcu_cleanup_notifier_head(name) \
13656 cleanup_srcu_struct(&(name)->srcu);
13657 @@ -101,7 +99,13 @@ extern void srcu_init_notifier_head(struct srcu_notifier_head *nh);
13659 #define RAW_NOTIFIER_INIT(name) { \
13661 -/* srcu_notifier_heads cannot be initialized statically */
13663 +#define SRCU_NOTIFIER_INIT(name, pcpu) \
13665 + .mutex = __MUTEX_INITIALIZER(name.mutex), \
13667 + .srcu = __SRCU_STRUCT_INIT(name.srcu, pcpu), \
13670 #define ATOMIC_NOTIFIER_HEAD(name) \
13671 struct atomic_notifier_head name = \
13672 @@ -113,6 +117,18 @@ extern void srcu_init_notifier_head(struct srcu_notifier_head *nh);
13673 struct raw_notifier_head name = \
13674 RAW_NOTIFIER_INIT(name)
13676 +#define _SRCU_NOTIFIER_HEAD(name, mod) \
13677 + static DEFINE_PER_CPU(struct srcu_struct_array, \
13678 + name##_head_srcu_array); \
13679 + mod struct srcu_notifier_head name = \
13680 + SRCU_NOTIFIER_INIT(name, name##_head_srcu_array)
13682 +#define SRCU_NOTIFIER_HEAD(name) \
13683 + _SRCU_NOTIFIER_HEAD(name, )
13685 +#define SRCU_NOTIFIER_HEAD_STATIC(name) \
13686 + _SRCU_NOTIFIER_HEAD(name, static)
13690 extern int atomic_notifier_chain_register(struct atomic_notifier_head *nh,
13691 @@ -182,12 +198,12 @@ static inline int notifier_to_errno(int ret)
13694 * Declared notifiers so far. I can imagine quite a few more chains
13695 - * over time (eg laptop power reset chains, reboot chain (to clean
13696 + * over time (eg laptop power reset chains, reboot chain (to clean
13697 * device units up), device [un]mount chain, module load/unload chain,
13698 - * low memory chain, screenblank chain (for plug in modular screenblankers)
13699 + * low memory chain, screenblank chain (for plug in modular screenblankers)
13700 * VC switch chains (for loadable kernel svgalib VC switch helpers) etc...
13704 /* CPU notfiers are defined in include/linux/cpu.h. */
13706 /* netdevice notifiers are defined in include/linux/netdevice.h */
13707 diff --git a/include/linux/percpu.h b/include/linux/percpu.h
13708 index caebf2a758dc..4ecc057b6e27 100644
13709 --- a/include/linux/percpu.h
13710 +++ b/include/linux/percpu.h
13712 PERCPU_MODULE_RESERVE)
13715 +#ifdef CONFIG_PREEMPT_RT_FULL
13717 +#define get_local_var(var) (*({ \
13718 + migrate_disable(); \
13719 + this_cpu_ptr(&var); }))
13721 +#define put_local_var(var) do { \
13723 + migrate_enable(); \
13726 +# define get_local_ptr(var) ({ \
13727 + migrate_disable(); \
13728 + this_cpu_ptr(var); })
13730 +# define put_local_ptr(var) do { \
13732 + migrate_enable(); \
13737 +#define get_local_var(var) get_cpu_var(var)
13738 +#define put_local_var(var) put_cpu_var(var)
13739 +#define get_local_ptr(var) get_cpu_ptr(var)
13740 +#define put_local_ptr(var) put_cpu_ptr(var)
13744 /* minimum unit size, also is the maximum supported allocation size */
13745 #define PCPU_MIN_UNIT_SIZE PFN_ALIGN(32 << 10)
13747 @@ -116,6 +145,7 @@ extern int __init pcpu_page_first_chunk(size_t reserved_size,
13750 extern void __percpu *__alloc_reserved_percpu(size_t size, size_t align);
13751 +extern bool __is_kernel_percpu_address(unsigned long addr, unsigned long *can_addr);
13752 extern bool is_kernel_percpu_address(unsigned long addr);
13754 #if !defined(CONFIG_SMP) || !defined(CONFIG_HAVE_SETUP_PER_CPU_AREA)
13755 diff --git a/include/linux/pid.h b/include/linux/pid.h
13756 index 23705a53abba..2cc64b779f03 100644
13757 --- a/include/linux/pid.h
13758 +++ b/include/linux/pid.h
13760 #define _LINUX_PID_H
13762 #include <linux/rcupdate.h>
13763 +#include <linux/atomic.h>
13767 diff --git a/include/linux/preempt.h b/include/linux/preempt.h
13768 index 75e4e30677f1..1cfb1cb72354 100644
13769 --- a/include/linux/preempt.h
13770 +++ b/include/linux/preempt.h
13772 #define HARDIRQ_OFFSET (1UL << HARDIRQ_SHIFT)
13773 #define NMI_OFFSET (1UL << NMI_SHIFT)
13775 -#define SOFTIRQ_DISABLE_OFFSET (2 * SOFTIRQ_OFFSET)
13776 +#ifndef CONFIG_PREEMPT_RT_FULL
13777 +# define SOFTIRQ_DISABLE_OFFSET (2 * SOFTIRQ_OFFSET)
13779 +# define SOFTIRQ_DISABLE_OFFSET (0)
13782 /* We use the MSB mostly because its available */
13783 #define PREEMPT_NEED_RESCHED 0x80000000
13785 #include <asm/preempt.h>
13787 #define hardirq_count() (preempt_count() & HARDIRQ_MASK)
13788 -#define softirq_count() (preempt_count() & SOFTIRQ_MASK)
13789 #define irq_count() (preempt_count() & (HARDIRQ_MASK | SOFTIRQ_MASK \
13791 +#ifndef CONFIG_PREEMPT_RT_FULL
13792 +# define softirq_count() (preempt_count() & SOFTIRQ_MASK)
13793 +# define in_serving_softirq() (softirq_count() & SOFTIRQ_OFFSET)
13795 +# define softirq_count() (0UL)
13796 +extern int in_serving_softirq(void);
13800 * Are we doing bottom half or hardware interrupt processing?
13802 #define in_irq() (hardirq_count())
13803 #define in_softirq() (softirq_count())
13804 #define in_interrupt() (irq_count())
13805 -#define in_serving_softirq() (softirq_count() & SOFTIRQ_OFFSET)
13808 * Are we in NMI context?
13809 @@ -91,7 +100,11 @@
13811 * The preempt_count offset after spin_lock()
13813 +#if !defined(CONFIG_PREEMPT_RT_FULL)
13814 #define PREEMPT_LOCK_OFFSET PREEMPT_DISABLE_OFFSET
13816 +#define PREEMPT_LOCK_OFFSET 0
13820 * The preempt_count offset needed for things like:
13821 @@ -140,6 +153,20 @@ extern void preempt_count_sub(int val);
13822 #define preempt_count_inc() preempt_count_add(1)
13823 #define preempt_count_dec() preempt_count_sub(1)
13825 +#ifdef CONFIG_PREEMPT_LAZY
13826 +#define add_preempt_lazy_count(val) do { preempt_lazy_count() += (val); } while (0)
13827 +#define sub_preempt_lazy_count(val) do { preempt_lazy_count() -= (val); } while (0)
13828 +#define inc_preempt_lazy_count() add_preempt_lazy_count(1)
13829 +#define dec_preempt_lazy_count() sub_preempt_lazy_count(1)
13830 +#define preempt_lazy_count() (current_thread_info()->preempt_lazy_count)
13832 +#define add_preempt_lazy_count(val) do { } while (0)
13833 +#define sub_preempt_lazy_count(val) do { } while (0)
13834 +#define inc_preempt_lazy_count() do { } while (0)
13835 +#define dec_preempt_lazy_count() do { } while (0)
13836 +#define preempt_lazy_count() (0)
13839 #ifdef CONFIG_PREEMPT_COUNT
13841 #define preempt_disable() \
13842 @@ -148,13 +175,25 @@ do { \
13846 +#define preempt_lazy_disable() \
13848 + inc_preempt_lazy_count(); \
13852 #define sched_preempt_enable_no_resched() \
13855 preempt_count_dec(); \
13858 -#define preempt_enable_no_resched() sched_preempt_enable_no_resched()
13859 +#ifdef CONFIG_PREEMPT_RT_BASE
13860 +# define preempt_enable_no_resched() sched_preempt_enable_no_resched()
13861 +# define preempt_check_resched_rt() preempt_check_resched()
13863 +# define preempt_enable_no_resched() preempt_enable()
13864 +# define preempt_check_resched_rt() barrier();
13867 #define preemptible() (preempt_count() == 0 && !irqs_disabled())
13869 @@ -179,6 +218,13 @@ do { \
13870 __preempt_schedule(); \
13873 +#define preempt_lazy_enable() \
13875 + dec_preempt_lazy_count(); \
13877 + preempt_check_resched(); \
13880 #else /* !CONFIG_PREEMPT */
13881 #define preempt_enable() \
13883 @@ -224,6 +270,7 @@ do { \
13884 #define preempt_disable_notrace() barrier()
13885 #define preempt_enable_no_resched_notrace() barrier()
13886 #define preempt_enable_notrace() barrier()
13887 +#define preempt_check_resched_rt() barrier()
13888 #define preemptible() 0
13890 #endif /* CONFIG_PREEMPT_COUNT */
13891 @@ -244,10 +291,31 @@ do { \
13893 #define preempt_fold_need_resched() \
13895 - if (tif_need_resched()) \
13896 + if (tif_need_resched_now()) \
13897 set_preempt_need_resched(); \
13900 +#ifdef CONFIG_PREEMPT_RT_FULL
13901 +# define preempt_disable_rt() preempt_disable()
13902 +# define preempt_enable_rt() preempt_enable()
13903 +# define preempt_disable_nort() barrier()
13904 +# define preempt_enable_nort() barrier()
13905 +# ifdef CONFIG_SMP
13906 + extern void migrate_disable(void);
13907 + extern void migrate_enable(void);
13908 +# else /* CONFIG_SMP */
13909 +# define migrate_disable() barrier()
13910 +# define migrate_enable() barrier()
13911 +# endif /* CONFIG_SMP */
13913 +# define preempt_disable_rt() barrier()
13914 +# define preempt_enable_rt() barrier()
13915 +# define preempt_disable_nort() preempt_disable()
13916 +# define preempt_enable_nort() preempt_enable()
13917 +# define migrate_disable() preempt_disable()
13918 +# define migrate_enable() preempt_enable()
13921 #ifdef CONFIG_PREEMPT_NOTIFIERS
13923 struct preempt_notifier;
13924 diff --git a/include/linux/printk.h b/include/linux/printk.h
13925 index 9729565c25ff..9cdca696b718 100644
13926 --- a/include/linux/printk.h
13927 +++ b/include/linux/printk.h
13928 @@ -117,9 +117,11 @@ int no_printk(const char *fmt, ...)
13929 #ifdef CONFIG_EARLY_PRINTK
13930 extern asmlinkage __printf(1, 2)
13931 void early_printk(const char *fmt, ...);
13932 +extern void printk_kill(void);
13934 static inline __printf(1, 2) __cold
13935 void early_printk(const char *s, ...) { }
13936 +static inline void printk_kill(void) { }
13939 typedef __printf(1, 0) int (*printk_func_t)(const char *fmt, va_list args);
13940 diff --git a/include/linux/radix-tree.h b/include/linux/radix-tree.h
13941 index 5d5174b59802..327dddaf4c8f 100644
13942 --- a/include/linux/radix-tree.h
13943 +++ b/include/linux/radix-tree.h
13944 @@ -279,6 +279,8 @@ unsigned int radix_tree_gang_lookup_slot(struct radix_tree_root *root,
13945 unsigned long first_index, unsigned int max_items);
13946 int radix_tree_preload(gfp_t gfp_mask);
13947 int radix_tree_maybe_preload(gfp_t gfp_mask);
13948 +void radix_tree_preload_end(void);
13950 void radix_tree_init(void);
13951 void *radix_tree_tag_set(struct radix_tree_root *root,
13952 unsigned long index, unsigned int tag);
13953 @@ -301,11 +303,6 @@ unsigned long radix_tree_range_tag_if_tagged(struct radix_tree_root *root,
13954 int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag);
13955 unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item);
13957 -static inline void radix_tree_preload_end(void)
13959 - preempt_enable();
13963 * struct radix_tree_iter - radix tree iterator state
13965 diff --git a/include/linux/random.h b/include/linux/random.h
13966 index a75840c1aa71..1a804361670c 100644
13967 --- a/include/linux/random.h
13968 +++ b/include/linux/random.h
13969 @@ -20,7 +20,7 @@ struct random_ready_callback {
13970 extern void add_device_randomness(const void *, unsigned int);
13971 extern void add_input_randomness(unsigned int type, unsigned int code,
13972 unsigned int value);
13973 -extern void add_interrupt_randomness(int irq, int irq_flags);
13974 +extern void add_interrupt_randomness(int irq, int irq_flags, __u64 ip);
13976 extern void get_random_bytes(void *buf, int nbytes);
13977 extern int add_random_ready_callback(struct random_ready_callback *rdy);
13978 diff --git a/include/linux/rbtree.h b/include/linux/rbtree.h
13979 index a5aa7ae671f4..24ddffd25492 100644
13980 --- a/include/linux/rbtree.h
13981 +++ b/include/linux/rbtree.h
13984 #include <linux/kernel.h>
13985 #include <linux/stddef.h>
13986 -#include <linux/rcupdate.h>
13989 unsigned long __rb_parent_color;
13990 @@ -86,14 +85,8 @@ static inline void rb_link_node(struct rb_node *node, struct rb_node *parent,
13994 -static inline void rb_link_node_rcu(struct rb_node *node, struct rb_node *parent,
13995 - struct rb_node **rb_link)
13997 - node->__rb_parent_color = (unsigned long)parent;
13998 - node->rb_left = node->rb_right = NULL;
14000 - rcu_assign_pointer(*rb_link, node);
14002 +void rb_link_node_rcu(struct rb_node *node, struct rb_node *parent,
14003 + struct rb_node **rb_link);
14005 #define rb_entry_safe(ptr, type, member) \
14006 ({ typeof(ptr) ____ptr = (ptr); \
14007 diff --git a/include/linux/rcupdate.h b/include/linux/rcupdate.h
14008 index a0189ba67fde..c2f5f955163d 100644
14009 --- a/include/linux/rcupdate.h
14010 +++ b/include/linux/rcupdate.h
14011 @@ -169,6 +169,9 @@ void call_rcu(struct rcu_head *head,
14013 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */
14015 +#ifdef CONFIG_PREEMPT_RT_FULL
14016 +#define call_rcu_bh call_rcu
14019 * call_rcu_bh() - Queue an RCU for invocation after a quicker grace period.
14020 * @head: structure to be used for queueing the RCU updates.
14021 @@ -192,6 +195,7 @@ void call_rcu(struct rcu_head *head,
14023 void call_rcu_bh(struct rcu_head *head,
14024 rcu_callback_t func);
14028 * call_rcu_sched() - Queue an RCU for invocation after sched grace period.
14029 @@ -292,6 +296,11 @@ void synchronize_rcu(void);
14030 * types of kernel builds, the rcu_read_lock() nesting depth is unknowable.
14032 #define rcu_preempt_depth() (current->rcu_read_lock_nesting)
14033 +#ifndef CONFIG_PREEMPT_RT_FULL
14034 +#define sched_rcu_preempt_depth() rcu_preempt_depth()
14036 +static inline int sched_rcu_preempt_depth(void) { return 0; }
14039 #else /* #ifdef CONFIG_PREEMPT_RCU */
14041 @@ -317,6 +326,8 @@ static inline int rcu_preempt_depth(void)
14045 +#define sched_rcu_preempt_depth() rcu_preempt_depth()
14047 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */
14049 /* Internal to kernel */
14050 @@ -489,7 +500,14 @@ extern struct lockdep_map rcu_callback_map;
14051 int debug_lockdep_rcu_enabled(void);
14053 int rcu_read_lock_held(void);
14054 +#ifdef CONFIG_PREEMPT_RT_FULL
14055 +static inline int rcu_read_lock_bh_held(void)
14057 + return rcu_read_lock_held();
14060 int rcu_read_lock_bh_held(void);
14064 * rcu_read_lock_sched_held() - might we be in RCU-sched read-side critical section?
14065 @@ -937,10 +955,14 @@ static inline void rcu_read_unlock(void)
14066 static inline void rcu_read_lock_bh(void)
14068 local_bh_disable();
14069 +#ifdef CONFIG_PREEMPT_RT_FULL
14073 rcu_lock_acquire(&rcu_bh_lock_map);
14074 RCU_LOCKDEP_WARN(!rcu_is_watching(),
14075 "rcu_read_lock_bh() used illegally while idle");
14080 @@ -950,10 +972,14 @@ static inline void rcu_read_lock_bh(void)
14082 static inline void rcu_read_unlock_bh(void)
14084 +#ifdef CONFIG_PREEMPT_RT_FULL
14085 + rcu_read_unlock();
14087 RCU_LOCKDEP_WARN(!rcu_is_watching(),
14088 "rcu_read_unlock_bh() used illegally while idle");
14089 rcu_lock_release(&rcu_bh_lock_map);
14095 diff --git a/include/linux/rcutree.h b/include/linux/rcutree.h
14096 index 60d15a080d7c..436c9e62bfc6 100644
14097 --- a/include/linux/rcutree.h
14098 +++ b/include/linux/rcutree.h
14099 @@ -44,7 +44,11 @@ static inline void rcu_virt_note_context_switch(int cpu)
14100 rcu_note_context_switch();
14103 +#ifdef CONFIG_PREEMPT_RT_FULL
14104 +# define synchronize_rcu_bh synchronize_rcu
14106 void synchronize_rcu_bh(void);
14108 void synchronize_sched_expedited(void);
14109 void synchronize_rcu_expedited(void);
14111 @@ -72,7 +76,11 @@ static inline void synchronize_rcu_bh_expedited(void)
14114 void rcu_barrier(void);
14115 +#ifdef CONFIG_PREEMPT_RT_FULL
14116 +# define rcu_barrier_bh rcu_barrier
14118 void rcu_barrier_bh(void);
14120 void rcu_barrier_sched(void);
14121 unsigned long get_state_synchronize_rcu(void);
14122 void cond_synchronize_rcu(unsigned long oldstate);
14123 @@ -85,12 +93,10 @@ unsigned long rcu_batches_started(void);
14124 unsigned long rcu_batches_started_bh(void);
14125 unsigned long rcu_batches_started_sched(void);
14126 unsigned long rcu_batches_completed(void);
14127 -unsigned long rcu_batches_completed_bh(void);
14128 unsigned long rcu_batches_completed_sched(void);
14129 void show_rcu_gp_kthreads(void);
14131 void rcu_force_quiescent_state(void);
14132 -void rcu_bh_force_quiescent_state(void);
14133 void rcu_sched_force_quiescent_state(void);
14135 void rcu_idle_enter(void);
14136 @@ -105,6 +111,14 @@ extern int rcu_scheduler_active __read_mostly;
14138 bool rcu_is_watching(void);
14140 +#ifndef CONFIG_PREEMPT_RT_FULL
14141 +void rcu_bh_force_quiescent_state(void);
14142 +unsigned long rcu_batches_completed_bh(void);
14144 +# define rcu_bh_force_quiescent_state rcu_force_quiescent_state
14145 +# define rcu_batches_completed_bh rcu_batches_completed
14148 void rcu_all_qs(void);
14150 #endif /* __LINUX_RCUTREE_H */
14151 diff --git a/include/linux/rtmutex.h b/include/linux/rtmutex.h
14152 index 1abba5ce2a2f..30211c627511 100644
14153 --- a/include/linux/rtmutex.h
14154 +++ b/include/linux/rtmutex.h
14155 @@ -13,11 +13,15 @@
14156 #define __LINUX_RT_MUTEX_H
14158 #include <linux/linkage.h>
14159 +#include <linux/spinlock_types_raw.h>
14160 #include <linux/rbtree.h>
14161 -#include <linux/spinlock_types.h>
14163 extern int max_lock_depth; /* for sysctl */
14165 +#ifdef CONFIG_DEBUG_MUTEXES
14166 +#include <linux/debug_locks.h>
14170 * The rt_mutex structure
14172 @@ -31,8 +35,8 @@ struct rt_mutex {
14173 struct rb_root waiters;
14174 struct rb_node *waiters_leftmost;
14175 struct task_struct *owner;
14176 -#ifdef CONFIG_DEBUG_RT_MUTEXES
14178 +#ifdef CONFIG_DEBUG_RT_MUTEXES
14179 const char *name, *file;
14182 @@ -55,22 +59,33 @@ struct hrtimer_sleeper;
14183 # define rt_mutex_debug_check_no_locks_held(task) do { } while (0)
14186 +# define rt_mutex_init(mutex) \
14188 + raw_spin_lock_init(&(mutex)->wait_lock); \
14189 + __rt_mutex_init(mutex, #mutex); \
14192 #ifdef CONFIG_DEBUG_RT_MUTEXES
14193 # define __DEBUG_RT_MUTEX_INITIALIZER(mutexname) \
14194 , .name = #mutexname, .file = __FILE__, .line = __LINE__
14195 -# define rt_mutex_init(mutex) __rt_mutex_init(mutex, __func__)
14196 extern void rt_mutex_debug_task_free(struct task_struct *tsk);
14198 # define __DEBUG_RT_MUTEX_INITIALIZER(mutexname)
14199 -# define rt_mutex_init(mutex) __rt_mutex_init(mutex, NULL)
14200 # define rt_mutex_debug_task_free(t) do { } while (0)
14203 -#define __RT_MUTEX_INITIALIZER(mutexname) \
14204 - { .wait_lock = __RAW_SPIN_LOCK_UNLOCKED(mutexname.wait_lock) \
14205 +#define __RT_MUTEX_INITIALIZER_PLAIN(mutexname) \
14206 + .wait_lock = __RAW_SPIN_LOCK_UNLOCKED(mutexname.wait_lock) \
14207 , .waiters = RB_ROOT \
14209 - __DEBUG_RT_MUTEX_INITIALIZER(mutexname)}
14210 + __DEBUG_RT_MUTEX_INITIALIZER(mutexname)
14212 +#define __RT_MUTEX_INITIALIZER(mutexname) \
14213 + { __RT_MUTEX_INITIALIZER_PLAIN(mutexname) }
14215 +#define __RT_MUTEX_INITIALIZER_SAVE_STATE(mutexname) \
14216 + { __RT_MUTEX_INITIALIZER_PLAIN(mutexname) \
14217 + , .save_state = 1 }
14219 #define DEFINE_RT_MUTEX(mutexname) \
14220 struct rt_mutex mutexname = __RT_MUTEX_INITIALIZER(mutexname)
14221 @@ -91,6 +106,7 @@ extern void rt_mutex_destroy(struct rt_mutex *lock);
14223 extern void rt_mutex_lock(struct rt_mutex *lock);
14224 extern int rt_mutex_lock_interruptible(struct rt_mutex *lock);
14225 +extern int rt_mutex_lock_killable(struct rt_mutex *lock);
14226 extern int rt_mutex_timed_lock(struct rt_mutex *lock,
14227 struct hrtimer_sleeper *timeout);
14229 diff --git a/include/linux/rwlock_rt.h b/include/linux/rwlock_rt.h
14230 new file mode 100644
14231 index 000000000000..49ed2d45d3be
14233 +++ b/include/linux/rwlock_rt.h
14235 +#ifndef __LINUX_RWLOCK_RT_H
14236 +#define __LINUX_RWLOCK_RT_H
14238 +#ifndef __LINUX_SPINLOCK_H
14239 +#error Do not include directly. Use spinlock.h
14242 +#define rwlock_init(rwl) \
14244 + static struct lock_class_key __key; \
14246 + rt_mutex_init(&(rwl)->lock); \
14247 + __rt_rwlock_init(rwl, #rwl, &__key); \
14250 +extern void __lockfunc rt_write_lock(rwlock_t *rwlock);
14251 +extern void __lockfunc rt_read_lock(rwlock_t *rwlock);
14252 +extern int __lockfunc rt_write_trylock(rwlock_t *rwlock);
14253 +extern int __lockfunc rt_write_trylock_irqsave(rwlock_t *trylock, unsigned long *flags);
14254 +extern int __lockfunc rt_read_trylock(rwlock_t *rwlock);
14255 +extern void __lockfunc rt_write_unlock(rwlock_t *rwlock);
14256 +extern void __lockfunc rt_read_unlock(rwlock_t *rwlock);
14257 +extern unsigned long __lockfunc rt_write_lock_irqsave(rwlock_t *rwlock);
14258 +extern unsigned long __lockfunc rt_read_lock_irqsave(rwlock_t *rwlock);
14259 +extern void __rt_rwlock_init(rwlock_t *rwlock, char *name, struct lock_class_key *key);
14261 +#define read_trylock(lock) __cond_lock(lock, rt_read_trylock(lock))
14262 +#define write_trylock(lock) __cond_lock(lock, rt_write_trylock(lock))
14264 +#define write_trylock_irqsave(lock, flags) \
14265 + __cond_lock(lock, rt_write_trylock_irqsave(lock, &flags))
14267 +#define read_lock_irqsave(lock, flags) \
14269 + typecheck(unsigned long, flags); \
14270 + flags = rt_read_lock_irqsave(lock); \
14273 +#define write_lock_irqsave(lock, flags) \
14275 + typecheck(unsigned long, flags); \
14276 + flags = rt_write_lock_irqsave(lock); \
14279 +#define read_lock(lock) rt_read_lock(lock)
14281 +#define read_lock_bh(lock) \
14283 + local_bh_disable(); \
14284 + rt_read_lock(lock); \
14287 +#define read_lock_irq(lock) read_lock(lock)
14289 +#define write_lock(lock) rt_write_lock(lock)
14291 +#define write_lock_bh(lock) \
14293 + local_bh_disable(); \
14294 + rt_write_lock(lock); \
14297 +#define write_lock_irq(lock) write_lock(lock)
14299 +#define read_unlock(lock) rt_read_unlock(lock)
14301 +#define read_unlock_bh(lock) \
14303 + rt_read_unlock(lock); \
14304 + local_bh_enable(); \
14307 +#define read_unlock_irq(lock) read_unlock(lock)
14309 +#define write_unlock(lock) rt_write_unlock(lock)
14311 +#define write_unlock_bh(lock) \
14313 + rt_write_unlock(lock); \
14314 + local_bh_enable(); \
14317 +#define write_unlock_irq(lock) write_unlock(lock)
14319 +#define read_unlock_irqrestore(lock, flags) \
14321 + typecheck(unsigned long, flags); \
14323 + rt_read_unlock(lock); \
14326 +#define write_unlock_irqrestore(lock, flags) \
14328 + typecheck(unsigned long, flags); \
14330 + rt_write_unlock(lock); \
14334 diff --git a/include/linux/rwlock_types.h b/include/linux/rwlock_types.h
14335 index cc0072e93e36..d0da966ad7a0 100644
14336 --- a/include/linux/rwlock_types.h
14337 +++ b/include/linux/rwlock_types.h
14339 #ifndef __LINUX_RWLOCK_TYPES_H
14340 #define __LINUX_RWLOCK_TYPES_H
14342 +#if !defined(__LINUX_SPINLOCK_TYPES_H)
14343 +# error "Do not include directly, include spinlock_types.h"
14347 * include/linux/rwlock_types.h - generic rwlock type definitions
14349 @@ -43,6 +47,7 @@ typedef struct {
14350 RW_DEP_MAP_INIT(lockname) }
14353 -#define DEFINE_RWLOCK(x) rwlock_t x = __RW_LOCK_UNLOCKED(x)
14354 +#define DEFINE_RWLOCK(name) \
14355 + rwlock_t name __cacheline_aligned_in_smp = __RW_LOCK_UNLOCKED(name)
14357 #endif /* __LINUX_RWLOCK_TYPES_H */
14358 diff --git a/include/linux/rwlock_types_rt.h b/include/linux/rwlock_types_rt.h
14359 new file mode 100644
14360 index 000000000000..b13832119591
14362 +++ b/include/linux/rwlock_types_rt.h
14364 +#ifndef __LINUX_RWLOCK_TYPES_RT_H
14365 +#define __LINUX_RWLOCK_TYPES_RT_H
14367 +#ifndef __LINUX_SPINLOCK_TYPES_H
14368 +#error "Do not include directly. Include spinlock_types.h instead"
14372 + * rwlocks - rtmutex which allows single reader recursion
14375 + struct rt_mutex lock;
14377 + unsigned int break_lock;
14378 +#ifdef CONFIG_DEBUG_LOCK_ALLOC
14379 + struct lockdep_map dep_map;
14383 +#ifdef CONFIG_DEBUG_LOCK_ALLOC
14384 +# define RW_DEP_MAP_INIT(lockname) .dep_map = { .name = #lockname }
14386 +# define RW_DEP_MAP_INIT(lockname)
14389 +#define __RW_LOCK_UNLOCKED(name) \
14390 + { .lock = __RT_MUTEX_INITIALIZER_SAVE_STATE(name.lock), \
14391 + RW_DEP_MAP_INIT(name) }
14393 +#define DEFINE_RWLOCK(name) \
14394 + rwlock_t name __cacheline_aligned_in_smp = __RW_LOCK_UNLOCKED(name)
14397 diff --git a/include/linux/rwsem.h b/include/linux/rwsem.h
14398 index 8f498cdde280..2b2148431f14 100644
14399 --- a/include/linux/rwsem.h
14400 +++ b/include/linux/rwsem.h
14402 #include <linux/osq_lock.h>
14405 +#ifdef CONFIG_PREEMPT_RT_FULL
14406 +#include <linux/rwsem_rt.h>
14407 +#else /* PREEMPT_RT_FULL */
14409 struct rw_semaphore;
14411 #ifdef CONFIG_RWSEM_GENERIC_SPINLOCK
14412 @@ -177,4 +181,6 @@ extern void up_read_non_owner(struct rw_semaphore *sem);
14413 # define up_read_non_owner(sem) up_read(sem)
14416 +#endif /* !PREEMPT_RT_FULL */
14418 #endif /* _LINUX_RWSEM_H */
14419 diff --git a/include/linux/rwsem_rt.h b/include/linux/rwsem_rt.h
14420 new file mode 100644
14421 index 000000000000..f97860b2e2a4
14423 +++ b/include/linux/rwsem_rt.h
14425 +#ifndef _LINUX_RWSEM_RT_H
14426 +#define _LINUX_RWSEM_RT_H
14428 +#ifndef _LINUX_RWSEM_H
14429 +#error "Include rwsem.h"
14433 + * RW-semaphores are a spinlock plus a reader-depth count.
14435 + * Note that the semantics are different from the usual
14436 + * Linux rw-sems, in PREEMPT_RT mode we do not allow
14437 + * multiple readers to hold the lock at once, we only allow
14438 + * a read-lock owner to read-lock recursively. This is
14439 + * better for latency, makes the implementation inherently
14440 + * fair and makes it simpler as well.
14443 +#include <linux/rtmutex.h>
14445 +struct rw_semaphore {
14446 + struct rt_mutex lock;
14448 +#ifdef CONFIG_DEBUG_LOCK_ALLOC
14449 + struct lockdep_map dep_map;
14453 +#define __RWSEM_INITIALIZER(name) \
14454 + { .lock = __RT_MUTEX_INITIALIZER(name.lock), \
14455 + RW_DEP_MAP_INIT(name) }
14457 +#define DECLARE_RWSEM(lockname) \
14458 + struct rw_semaphore lockname = __RWSEM_INITIALIZER(lockname)
14460 +extern void __rt_rwsem_init(struct rw_semaphore *rwsem, const char *name,
14461 + struct lock_class_key *key);
14463 +#define __rt_init_rwsem(sem, name, key) \
14465 + rt_mutex_init(&(sem)->lock); \
14466 + __rt_rwsem_init((sem), (name), (key));\
14469 +#define __init_rwsem(sem, name, key) __rt_init_rwsem(sem, name, key)
14471 +# define rt_init_rwsem(sem) \
14473 + static struct lock_class_key __key; \
14475 + __rt_init_rwsem((sem), #sem, &__key); \
14478 +extern void rt_down_write(struct rw_semaphore *rwsem);
14479 +extern void rt_down_read_nested(struct rw_semaphore *rwsem, int subclass);
14480 +extern void rt_down_write_nested(struct rw_semaphore *rwsem, int subclass);
14481 +extern void rt_down_write_nested_lock(struct rw_semaphore *rwsem,
14482 + struct lockdep_map *nest);
14483 +extern void rt__down_read(struct rw_semaphore *rwsem);
14484 +extern void rt_down_read(struct rw_semaphore *rwsem);
14485 +extern int rt_down_write_trylock(struct rw_semaphore *rwsem);
14486 +extern int rt__down_read_trylock(struct rw_semaphore *rwsem);
14487 +extern int rt_down_read_trylock(struct rw_semaphore *rwsem);
14488 +extern void __rt_up_read(struct rw_semaphore *rwsem);
14489 +extern void rt_up_read(struct rw_semaphore *rwsem);
14490 +extern void rt_up_write(struct rw_semaphore *rwsem);
14491 +extern void rt_downgrade_write(struct rw_semaphore *rwsem);
14493 +#define init_rwsem(sem) rt_init_rwsem(sem)
14494 +#define rwsem_is_locked(s) rt_mutex_is_locked(&(s)->lock)
14496 +static inline int rwsem_is_contended(struct rw_semaphore *sem)
14498 + /* rt_mutex_has_waiters() */
14499 + return !RB_EMPTY_ROOT(&sem->lock.waiters);
14502 +static inline void __down_read(struct rw_semaphore *sem)
14504 + rt__down_read(sem);
14507 +static inline void down_read(struct rw_semaphore *sem)
14509 + rt_down_read(sem);
14512 +static inline int __down_read_trylock(struct rw_semaphore *sem)
14514 + return rt__down_read_trylock(sem);
14517 +static inline int down_read_trylock(struct rw_semaphore *sem)
14519 + return rt_down_read_trylock(sem);
14522 +static inline void down_write(struct rw_semaphore *sem)
14524 + rt_down_write(sem);
14527 +static inline int down_write_trylock(struct rw_semaphore *sem)
14529 + return rt_down_write_trylock(sem);
14532 +static inline void __up_read(struct rw_semaphore *sem)
14534 + __rt_up_read(sem);
14537 +static inline void up_read(struct rw_semaphore *sem)
14542 +static inline void up_write(struct rw_semaphore *sem)
14544 + rt_up_write(sem);
14547 +static inline void downgrade_write(struct rw_semaphore *sem)
14549 + rt_downgrade_write(sem);
14552 +static inline void down_read_nested(struct rw_semaphore *sem, int subclass)
14554 + return rt_down_read_nested(sem, subclass);
14557 +static inline void down_write_nested(struct rw_semaphore *sem, int subclass)
14559 + rt_down_write_nested(sem, subclass);
14561 +#ifdef CONFIG_DEBUG_LOCK_ALLOC
14562 +static inline void down_write_nest_lock(struct rw_semaphore *sem,
14563 + struct rw_semaphore *nest_lock)
14565 + rt_down_write_nested_lock(sem, &nest_lock->dep_map);
14570 +static inline void down_write_nest_lock(struct rw_semaphore *sem,
14571 + struct rw_semaphore *nest_lock)
14573 + rt_down_write_nested_lock(sem, NULL);
14577 diff --git a/include/linux/sched.h b/include/linux/sched.h
14578 index 352213b360d7..1c7aa9446684 100644
14579 --- a/include/linux/sched.h
14580 +++ b/include/linux/sched.h
14581 @@ -26,6 +26,7 @@ struct sched_param {
14582 #include <linux/nodemask.h>
14583 #include <linux/mm_types.h>
14584 #include <linux/preempt.h>
14585 +#include <asm/kmap_types.h>
14587 #include <asm/page.h>
14588 #include <asm/ptrace.h>
14589 @@ -182,8 +183,6 @@ extern void update_cpu_load_nohz(void);
14590 static inline void update_cpu_load_nohz(void) { }
14593 -extern unsigned long get_parent_ip(unsigned long addr);
14595 extern void dump_cpu_task(int cpu);
14598 @@ -242,10 +241,7 @@ extern char ___assert_task_state[1 - 2*!!(
14599 TASK_UNINTERRUPTIBLE | __TASK_STOPPED | \
14600 __TASK_TRACED | EXIT_ZOMBIE | EXIT_DEAD)
14602 -#define task_is_traced(task) ((task->state & __TASK_TRACED) != 0)
14603 #define task_is_stopped(task) ((task->state & __TASK_STOPPED) != 0)
14604 -#define task_is_stopped_or_traced(task) \
14605 - ((task->state & (__TASK_STOPPED | __TASK_TRACED)) != 0)
14606 #define task_contributes_to_load(task) \
14607 ((task->state & TASK_UNINTERRUPTIBLE) != 0 && \
14608 (task->flags & PF_FROZEN) == 0 && \
14609 @@ -311,6 +307,11 @@ extern char ___assert_task_state[1 - 2*!!(
14613 +#define __set_current_state_no_track(state_value) \
14614 + do { current->state = (state_value); } while (0)
14615 +#define set_current_state_no_track(state_value) \
14616 + set_mb(current->state, (state_value))
14618 /* Task command name length */
14619 #define TASK_COMM_LEN 16
14621 @@ -970,8 +971,18 @@ struct wake_q_head {
14622 struct wake_q_head name = { WAKE_Q_TAIL, &name.first }
14624 extern void wake_q_add(struct wake_q_head *head,
14625 - struct task_struct *task);
14626 -extern void wake_up_q(struct wake_q_head *head);
14627 + struct task_struct *task);
14628 +extern void __wake_up_q(struct wake_q_head *head, bool sleeper);
14630 +static inline void wake_up_q(struct wake_q_head *head)
14632 + __wake_up_q(head, false);
14635 +static inline void wake_up_q_sleeper(struct wake_q_head *head)
14637 + __wake_up_q(head, true);
14641 * sched-domains (multiprocessor balancing) declarations:
14642 @@ -1379,6 +1390,7 @@ struct tlbflush_unmap_batch {
14644 struct task_struct {
14645 volatile long state; /* -1 unrunnable, 0 runnable, >0 stopped */
14646 + volatile long saved_state; /* saved state for "spinlock sleepers" */
14649 unsigned int flags; /* per process flags, defined below */
14650 @@ -1415,6 +1427,12 @@ struct task_struct {
14653 unsigned int policy;
14654 +#ifdef CONFIG_PREEMPT_RT_FULL
14655 + int migrate_disable;
14656 +# ifdef CONFIG_SCHED_DEBUG
14657 + int migrate_disable_atomic;
14660 int nr_cpus_allowed;
14661 cpumask_t cpus_allowed;
14663 @@ -1526,11 +1544,14 @@ struct task_struct {
14665 struct prev_cputime prev_cputime;
14666 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
14667 - seqlock_t vtime_seqlock;
14668 + seqcount_t vtime_seqcount;
14669 unsigned long long vtime_snap;
14671 - VTIME_SLEEPING = 0,
14672 + /* Task is sleeping or running in a CPU with VTIME inactive */
14673 + VTIME_INACTIVE = 0,
14674 + /* Task runs in userspace in a CPU with VTIME active */
14676 + /* Task runs in kernelspace in a CPU with VTIME active */
14678 } vtime_snap_whence;
14680 @@ -1542,6 +1563,9 @@ struct task_struct {
14682 struct task_cputime cputime_expires;
14683 struct list_head cpu_timers[3];
14684 +#ifdef CONFIG_PREEMPT_RT_BASE
14685 + struct task_struct *posix_timer_list;
14688 /* process credentials */
14689 const struct cred __rcu *ptracer_cred; /* Tracer's credentials at attach */
14690 @@ -1573,10 +1597,15 @@ struct task_struct {
14691 /* signal handlers */
14692 struct signal_struct *signal;
14693 struct sighand_struct *sighand;
14694 + struct sigqueue *sigqueue_cache;
14696 sigset_t blocked, real_blocked;
14697 sigset_t saved_sigmask; /* restored if set_restore_sigmask() was used */
14698 struct sigpending pending;
14699 +#ifdef CONFIG_PREEMPT_RT_FULL
14700 + /* TODO: move me into ->restart_block ? */
14701 + struct siginfo forced_info;
14704 unsigned long sas_ss_sp;
14705 size_t sas_ss_size;
14706 @@ -1800,6 +1829,12 @@ struct task_struct {
14707 unsigned long trace;
14708 /* bitmask and counter of trace recursion */
14709 unsigned long trace_recursion;
14710 +#ifdef CONFIG_WAKEUP_LATENCY_HIST
14711 + u64 preempt_timestamp_hist;
14712 +#ifdef CONFIG_MISSED_TIMER_OFFSETS_HIST
14713 + long timer_offset;
14716 #endif /* CONFIG_TRACING */
14717 #ifdef CONFIG_MEMCG
14718 struct mem_cgroup *memcg_in_oom;
14719 @@ -1816,9 +1851,23 @@ struct task_struct {
14720 unsigned int sequential_io;
14721 unsigned int sequential_io_avg;
14723 +#ifdef CONFIG_PREEMPT_RT_BASE
14724 + struct rcu_head put_rcu;
14725 + int softirq_nestcnt;
14726 + unsigned int softirqs_raised;
14728 +#ifdef CONFIG_PREEMPT_RT_FULL
14729 +# if defined CONFIG_HIGHMEM || defined CONFIG_X86_32
14731 + pte_t kmap_pte[KM_TYPE_NR];
14734 #ifdef CONFIG_DEBUG_ATOMIC_SLEEP
14735 unsigned long task_state_change;
14737 +#ifdef CONFIG_PREEMPT_RT_FULL
14738 + int xmit_recursion;
14740 int pagefault_disabled;
14741 /* CPU-specific state of this task */
14742 struct thread_struct thread;
14743 @@ -1836,9 +1885,6 @@ extern int arch_task_struct_size __read_mostly;
14744 # define arch_task_struct_size (sizeof(struct task_struct))
14747 -/* Future-safe accessor for struct task_struct's cpus_allowed. */
14748 -#define tsk_cpus_allowed(tsk) (&(tsk)->cpus_allowed)
14750 #define TNF_MIGRATED 0x01
14751 #define TNF_NO_GROUP 0x02
14752 #define TNF_SHARED 0x04
14753 @@ -2028,6 +2074,15 @@ extern struct pid *cad_pid;
14754 extern void free_task(struct task_struct *tsk);
14755 #define get_task_struct(tsk) do { atomic_inc(&(tsk)->usage); } while(0)
14757 +#ifdef CONFIG_PREEMPT_RT_BASE
14758 +extern void __put_task_struct_cb(struct rcu_head *rhp);
14760 +static inline void put_task_struct(struct task_struct *t)
14762 + if (atomic_dec_and_test(&t->usage))
14763 + call_rcu(&t->put_rcu, __put_task_struct_cb);
14766 extern void __put_task_struct(struct task_struct *t);
14768 static inline void put_task_struct(struct task_struct *t)
14769 @@ -2035,6 +2090,7 @@ static inline void put_task_struct(struct task_struct *t)
14770 if (atomic_dec_and_test(&t->usage))
14771 __put_task_struct(t);
14775 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
14776 extern void task_cputime(struct task_struct *t,
14777 @@ -2073,6 +2129,7 @@ extern void thread_group_cputime_adjusted(struct task_struct *p, cputime_t *ut,
14779 * Per process flags
14781 +#define PF_IN_SOFTIRQ 0x00000001 /* Task is serving softirq */
14782 #define PF_EXITING 0x00000004 /* getting shut down */
14783 #define PF_EXITPIDONE 0x00000008 /* pi exit done on shut down */
14784 #define PF_VCPU 0x00000010 /* I'm a virtual CPU */
14785 @@ -2237,6 +2294,10 @@ extern void do_set_cpus_allowed(struct task_struct *p,
14787 extern int set_cpus_allowed_ptr(struct task_struct *p,
14788 const struct cpumask *new_mask);
14789 +int migrate_me(void);
14790 +void tell_sched_cpu_down_begin(int cpu);
14791 +void tell_sched_cpu_down_done(int cpu);
14794 static inline void do_set_cpus_allowed(struct task_struct *p,
14795 const struct cpumask *new_mask)
14796 @@ -2249,6 +2310,9 @@ static inline int set_cpus_allowed_ptr(struct task_struct *p,
14800 +static inline int migrate_me(void) { return 0; }
14801 +static inline void tell_sched_cpu_down_begin(int cpu) { }
14802 +static inline void tell_sched_cpu_down_done(int cpu) { }
14805 #ifdef CONFIG_NO_HZ_COMMON
14806 @@ -2458,6 +2522,7 @@ extern void xtime_update(unsigned long ticks);
14808 extern int wake_up_state(struct task_struct *tsk, unsigned int state);
14809 extern int wake_up_process(struct task_struct *tsk);
14810 +extern int wake_up_lock_sleeper(struct task_struct * tsk);
14811 extern void wake_up_new_task(struct task_struct *tsk);
14813 extern void kick_process(struct task_struct *tsk);
14814 @@ -2581,12 +2646,24 @@ extern struct mm_struct * mm_alloc(void);
14816 /* mmdrop drops the mm and the page tables */
14817 extern void __mmdrop(struct mm_struct *);
14819 static inline void mmdrop(struct mm_struct * mm)
14821 if (unlikely(atomic_dec_and_test(&mm->mm_count)))
14825 +#ifdef CONFIG_PREEMPT_RT_BASE
14826 +extern void __mmdrop_delayed(struct rcu_head *rhp);
14827 +static inline void mmdrop_delayed(struct mm_struct *mm)
14829 + if (atomic_dec_and_test(&mm->mm_count))
14830 + call_rcu(&mm->delayed_drop, __mmdrop_delayed);
14833 +# define mmdrop_delayed(mm) mmdrop(mm)
14836 /* mmput gets rid of the mappings and all user-space */
14837 extern void mmput(struct mm_struct *);
14838 /* Grab a reference to a task's mm, if it is not already going away */
14839 @@ -2896,6 +2973,43 @@ static inline int test_tsk_need_resched(struct task_struct *tsk)
14840 return unlikely(test_tsk_thread_flag(tsk,TIF_NEED_RESCHED));
14843 +#ifdef CONFIG_PREEMPT_LAZY
14844 +static inline void set_tsk_need_resched_lazy(struct task_struct *tsk)
14846 + set_tsk_thread_flag(tsk,TIF_NEED_RESCHED_LAZY);
14849 +static inline void clear_tsk_need_resched_lazy(struct task_struct *tsk)
14851 + clear_tsk_thread_flag(tsk,TIF_NEED_RESCHED_LAZY);
14854 +static inline int test_tsk_need_resched_lazy(struct task_struct *tsk)
14856 + return unlikely(test_tsk_thread_flag(tsk,TIF_NEED_RESCHED_LAZY));
14859 +static inline int need_resched_lazy(void)
14861 + return test_thread_flag(TIF_NEED_RESCHED_LAZY);
14864 +static inline int need_resched_now(void)
14866 + return test_thread_flag(TIF_NEED_RESCHED);
14870 +static inline void clear_tsk_need_resched_lazy(struct task_struct *tsk) { }
14871 +static inline int need_resched_lazy(void) { return 0; }
14873 +static inline int need_resched_now(void)
14875 + return test_thread_flag(TIF_NEED_RESCHED);
14880 static inline int restart_syscall(void)
14882 set_tsk_thread_flag(current, TIF_SIGPENDING);
14883 @@ -2927,6 +3041,51 @@ static inline int signal_pending_state(long state, struct task_struct *p)
14884 return (state & TASK_INTERRUPTIBLE) || __fatal_signal_pending(p);
14887 +static inline bool __task_is_stopped_or_traced(struct task_struct *task)
14889 + if (task->state & (__TASK_STOPPED | __TASK_TRACED))
14891 +#ifdef CONFIG_PREEMPT_RT_FULL
14892 + if (task->saved_state & (__TASK_STOPPED | __TASK_TRACED))
14898 +static inline bool task_is_stopped_or_traced(struct task_struct *task)
14900 + bool traced_stopped;
14902 +#ifdef CONFIG_PREEMPT_RT_FULL
14903 + unsigned long flags;
14905 + raw_spin_lock_irqsave(&task->pi_lock, flags);
14906 + traced_stopped = __task_is_stopped_or_traced(task);
14907 + raw_spin_unlock_irqrestore(&task->pi_lock, flags);
14909 + traced_stopped = __task_is_stopped_or_traced(task);
14911 + return traced_stopped;
14914 +static inline bool task_is_traced(struct task_struct *task)
14916 + bool traced = false;
14918 + if (task->state & __TASK_TRACED)
14920 +#ifdef CONFIG_PREEMPT_RT_FULL
14921 + /* in case the task is sleeping on tasklist_lock */
14922 + raw_spin_lock_irq(&task->pi_lock);
14923 + if (task->state & __TASK_TRACED)
14925 + else if (task->saved_state & __TASK_TRACED)
14927 + raw_spin_unlock_irq(&task->pi_lock);
14933 * cond_resched() and cond_resched_lock(): latency reduction via
14934 * explicit rescheduling in places that are safe. The return
14935 @@ -2948,12 +3107,16 @@ extern int __cond_resched_lock(spinlock_t *lock);
14936 __cond_resched_lock(lock); \
14939 +#ifndef CONFIG_PREEMPT_RT_FULL
14940 extern int __cond_resched_softirq(void);
14942 #define cond_resched_softirq() ({ \
14943 ___might_sleep(__FILE__, __LINE__, SOFTIRQ_DISABLE_OFFSET); \
14944 __cond_resched_softirq(); \
14947 +# define cond_resched_softirq() cond_resched()
14950 static inline void cond_resched_rcu(void)
14952 @@ -3115,6 +3278,31 @@ static inline void set_task_cpu(struct task_struct *p, unsigned int cpu)
14954 #endif /* CONFIG_SMP */
14956 +static inline int __migrate_disabled(struct task_struct *p)
14958 +#ifdef CONFIG_PREEMPT_RT_FULL
14959 + return p->migrate_disable;
14965 +/* Future-safe accessor for struct task_struct's cpus_allowed. */
14966 +static inline const struct cpumask *tsk_cpus_allowed(struct task_struct *p)
14968 + if (__migrate_disabled(p))
14969 + return cpumask_of(task_cpu(p));
14971 + return &p->cpus_allowed;
14974 +static inline int tsk_nr_cpus_allowed(struct task_struct *p)
14976 + if (__migrate_disabled(p))
14978 + return p->nr_cpus_allowed;
14981 extern long sched_setaffinity(pid_t pid, const struct cpumask *new_mask);
14982 extern long sched_getaffinity(pid_t pid, struct cpumask *mask);
14984 diff --git a/include/linux/seqlock.h b/include/linux/seqlock.h
14985 index e0582106ef4f..b14f4d2368aa 100644
14986 --- a/include/linux/seqlock.h
14987 +++ b/include/linux/seqlock.h
14988 @@ -220,20 +220,30 @@ static inline int read_seqcount_retry(const seqcount_t *s, unsigned start)
14989 return __read_seqcount_retry(s, start);
14994 -static inline void raw_write_seqcount_begin(seqcount_t *s)
14995 +static inline void __raw_write_seqcount_begin(seqcount_t *s)
15001 -static inline void raw_write_seqcount_end(seqcount_t *s)
15002 +static inline void raw_write_seqcount_begin(seqcount_t *s)
15004 + preempt_disable_rt();
15005 + __raw_write_seqcount_begin(s);
15008 +static inline void __raw_write_seqcount_end(seqcount_t *s)
15014 +static inline void raw_write_seqcount_end(seqcount_t *s)
15016 + __raw_write_seqcount_end(s);
15017 + preempt_enable_rt();
15021 * raw_write_seqcount_barrier - do a seq write barrier
15022 * @s: pointer to seqcount_t
15023 @@ -425,10 +435,32 @@ typedef struct {
15025 * Read side functions for starting and finalizing a read side section.
15027 +#ifndef CONFIG_PREEMPT_RT_FULL
15028 static inline unsigned read_seqbegin(const seqlock_t *sl)
15030 return read_seqcount_begin(&sl->seqcount);
15034 + * Starvation safe read side for RT
15036 +static inline unsigned read_seqbegin(seqlock_t *sl)
15041 + ret = ACCESS_ONCE(sl->seqcount.sequence);
15042 + if (unlikely(ret & 1)) {
15044 + * Take the lock and let the writer proceed (i.e. evtl
15045 + * boost it), otherwise we could loop here forever.
15047 + spin_unlock_wait(&sl->lock);
15054 static inline unsigned read_seqretry(const seqlock_t *sl, unsigned start)
15056 @@ -443,36 +475,36 @@ static inline unsigned read_seqretry(const seqlock_t *sl, unsigned start)
15057 static inline void write_seqlock(seqlock_t *sl)
15059 spin_lock(&sl->lock);
15060 - write_seqcount_begin(&sl->seqcount);
15061 + __raw_write_seqcount_begin(&sl->seqcount);
15064 static inline void write_sequnlock(seqlock_t *sl)
15066 - write_seqcount_end(&sl->seqcount);
15067 + __raw_write_seqcount_end(&sl->seqcount);
15068 spin_unlock(&sl->lock);
15071 static inline void write_seqlock_bh(seqlock_t *sl)
15073 spin_lock_bh(&sl->lock);
15074 - write_seqcount_begin(&sl->seqcount);
15075 + __raw_write_seqcount_begin(&sl->seqcount);
15078 static inline void write_sequnlock_bh(seqlock_t *sl)
15080 - write_seqcount_end(&sl->seqcount);
15081 + __raw_write_seqcount_end(&sl->seqcount);
15082 spin_unlock_bh(&sl->lock);
15085 static inline void write_seqlock_irq(seqlock_t *sl)
15087 spin_lock_irq(&sl->lock);
15088 - write_seqcount_begin(&sl->seqcount);
15089 + __raw_write_seqcount_begin(&sl->seqcount);
15092 static inline void write_sequnlock_irq(seqlock_t *sl)
15094 - write_seqcount_end(&sl->seqcount);
15095 + __raw_write_seqcount_end(&sl->seqcount);
15096 spin_unlock_irq(&sl->lock);
15099 @@ -481,7 +513,7 @@ static inline unsigned long __write_seqlock_irqsave(seqlock_t *sl)
15100 unsigned long flags;
15102 spin_lock_irqsave(&sl->lock, flags);
15103 - write_seqcount_begin(&sl->seqcount);
15104 + __raw_write_seqcount_begin(&sl->seqcount);
15108 @@ -491,7 +523,7 @@ static inline unsigned long __write_seqlock_irqsave(seqlock_t *sl)
15110 write_sequnlock_irqrestore(seqlock_t *sl, unsigned long flags)
15112 - write_seqcount_end(&sl->seqcount);
15113 + __raw_write_seqcount_end(&sl->seqcount);
15114 spin_unlock_irqrestore(&sl->lock, flags);
15117 diff --git a/include/linux/signal.h b/include/linux/signal.h
15118 index d80259afb9e5..ddd1e6866a54 100644
15119 --- a/include/linux/signal.h
15120 +++ b/include/linux/signal.h
15121 @@ -233,6 +233,7 @@ static inline void init_sigpending(struct sigpending *sig)
15124 extern void flush_sigqueue(struct sigpending *queue);
15125 +extern void flush_task_sigqueue(struct task_struct *tsk);
15127 /* Test if 'sig' is valid signal. Use this instead of testing _NSIG directly */
15128 static inline int valid_signal(unsigned long sig)
15129 diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
15130 index d443d9ab0236..2d1c7f9b7fd0 100644
15131 --- a/include/linux/skbuff.h
15132 +++ b/include/linux/skbuff.h
15133 @@ -203,6 +203,7 @@ struct sk_buff_head {
15137 + raw_spinlock_t raw_lock;
15141 @@ -1465,6 +1466,12 @@ static inline void skb_queue_head_init(struct sk_buff_head *list)
15142 __skb_queue_head_init(list);
15145 +static inline void skb_queue_head_init_raw(struct sk_buff_head *list)
15147 + raw_spin_lock_init(&list->raw_lock);
15148 + __skb_queue_head_init(list);
15151 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
15152 struct lock_class_key *class)
15154 diff --git a/include/linux/smp.h b/include/linux/smp.h
15155 index c4414074bd88..e6ab36aeaaab 100644
15156 --- a/include/linux/smp.h
15157 +++ b/include/linux/smp.h
15158 @@ -185,6 +185,9 @@ static inline void smp_init(void) { }
15159 #define get_cpu() ({ preempt_disable(); smp_processor_id(); })
15160 #define put_cpu() preempt_enable()
15162 +#define get_cpu_light() ({ migrate_disable(); smp_processor_id(); })
15163 +#define put_cpu_light() migrate_enable()
15166 * Callback to arch code if there's nosmp or maxcpus=0 on the
15167 * boot command line:
15168 diff --git a/include/linux/spinlock.h b/include/linux/spinlock.h
15169 index 47dd0cebd204..02928fa5499d 100644
15170 --- a/include/linux/spinlock.h
15171 +++ b/include/linux/spinlock.h
15172 @@ -271,7 +271,11 @@ static inline void do_raw_spin_unlock(raw_spinlock_t *lock) __releases(lock)
15173 #define raw_spin_can_lock(lock) (!raw_spin_is_locked(lock))
15175 /* Include rwlock functions */
15176 -#include <linux/rwlock.h>
15177 +#ifdef CONFIG_PREEMPT_RT_FULL
15178 +# include <linux/rwlock_rt.h>
15180 +# include <linux/rwlock.h>
15184 * Pull the _spin_*()/_read_*()/_write_*() functions/declarations:
15185 @@ -282,6 +286,10 @@ static inline void do_raw_spin_unlock(raw_spinlock_t *lock) __releases(lock)
15186 # include <linux/spinlock_api_up.h>
15189 +#ifdef CONFIG_PREEMPT_RT_FULL
15190 +# include <linux/spinlock_rt.h>
15191 +#else /* PREEMPT_RT_FULL */
15194 * Map the spin_lock functions to the raw variants for PREEMPT_RT=n
15196 @@ -347,6 +355,12 @@ static __always_inline void spin_unlock(spinlock_t *lock)
15197 raw_spin_unlock(&lock->rlock);
15200 +static __always_inline int spin_unlock_no_deboost(spinlock_t *lock)
15202 + raw_spin_unlock(&lock->rlock);
15206 static __always_inline void spin_unlock_bh(spinlock_t *lock)
15208 raw_spin_unlock_bh(&lock->rlock);
15209 @@ -416,4 +430,6 @@ extern int _atomic_dec_and_lock(atomic_t *atomic, spinlock_t *lock);
15210 #define atomic_dec_and_lock(atomic, lock) \
15211 __cond_lock(lock, _atomic_dec_and_lock(atomic, lock))
15213 +#endif /* !PREEMPT_RT_FULL */
15215 #endif /* __LINUX_SPINLOCK_H */
15216 diff --git a/include/linux/spinlock_api_smp.h b/include/linux/spinlock_api_smp.h
15217 index 5344268e6e62..043263f30e81 100644
15218 --- a/include/linux/spinlock_api_smp.h
15219 +++ b/include/linux/spinlock_api_smp.h
15220 @@ -189,6 +189,8 @@ static inline int __raw_spin_trylock_bh(raw_spinlock_t *lock)
15224 -#include <linux/rwlock_api_smp.h>
15225 +#ifndef CONFIG_PREEMPT_RT_FULL
15226 +# include <linux/rwlock_api_smp.h>
15229 #endif /* __LINUX_SPINLOCK_API_SMP_H */
15230 diff --git a/include/linux/spinlock_rt.h b/include/linux/spinlock_rt.h
15231 new file mode 100644
15232 index 000000000000..7eb87584e843
15234 +++ b/include/linux/spinlock_rt.h
15236 +#ifndef __LINUX_SPINLOCK_RT_H
15237 +#define __LINUX_SPINLOCK_RT_H
15239 +#ifndef __LINUX_SPINLOCK_H
15240 +#error Do not include directly. Use spinlock.h
15243 +#include <linux/bug.h>
15246 +__rt_spin_lock_init(spinlock_t *lock, char *name, struct lock_class_key *key);
15248 +#define spin_lock_init(slock) \
15250 + static struct lock_class_key __key; \
15252 + rt_mutex_init(&(slock)->lock); \
15253 + __rt_spin_lock_init(slock, #slock, &__key); \
15256 +void __lockfunc rt_spin_lock__no_mg(spinlock_t *lock);
15257 +void __lockfunc rt_spin_unlock__no_mg(spinlock_t *lock);
15258 +int __lockfunc rt_spin_trylock__no_mg(spinlock_t *lock);
15260 +extern void __lockfunc rt_spin_lock(spinlock_t *lock);
15261 +extern unsigned long __lockfunc rt_spin_lock_trace_flags(spinlock_t *lock);
15262 +extern void __lockfunc rt_spin_lock_nested(spinlock_t *lock, int subclass);
15263 +extern void __lockfunc rt_spin_unlock(spinlock_t *lock);
15264 +extern int __lockfunc rt_spin_unlock_no_deboost(spinlock_t *lock);
15265 +extern void __lockfunc rt_spin_unlock_wait(spinlock_t *lock);
15266 +extern int __lockfunc rt_spin_trylock_irqsave(spinlock_t *lock, unsigned long *flags);
15267 +extern int __lockfunc rt_spin_trylock_bh(spinlock_t *lock);
15268 +extern int __lockfunc rt_spin_trylock(spinlock_t *lock);
15269 +extern int atomic_dec_and_spin_lock(atomic_t *atomic, spinlock_t *lock);
15272 + * lockdep-less calls, for derived types like rwlock:
15273 + * (for trylock they can use rt_mutex_trylock() directly.
15275 +extern void __lockfunc __rt_spin_lock__no_mg(struct rt_mutex *lock);
15276 +extern void __lockfunc __rt_spin_lock(struct rt_mutex *lock);
15277 +extern void __lockfunc __rt_spin_unlock(struct rt_mutex *lock);
15278 +extern int __lockfunc __rt_spin_trylock(struct rt_mutex *lock);
15280 +#define spin_lock(lock) rt_spin_lock(lock)
15282 +#define spin_lock_bh(lock) \
15284 + local_bh_disable(); \
15285 + rt_spin_lock(lock); \
15288 +#define spin_lock_irq(lock) spin_lock(lock)
15290 +#define spin_do_trylock(lock) __cond_lock(lock, rt_spin_trylock(lock))
15292 +#define spin_trylock(lock) \
15295 + __locked = spin_do_trylock(lock); \
15299 +#ifdef CONFIG_LOCKDEP
15300 +# define spin_lock_nested(lock, subclass) \
15302 + rt_spin_lock_nested(lock, subclass); \
15305 +#define spin_lock_bh_nested(lock, subclass) \
15307 + local_bh_disable(); \
15308 + rt_spin_lock_nested(lock, subclass); \
15311 +# define spin_lock_irqsave_nested(lock, flags, subclass) \
15313 + typecheck(unsigned long, flags); \
15315 + rt_spin_lock_nested(lock, subclass); \
15318 +# define spin_lock_nested(lock, subclass) spin_lock(lock)
15319 +# define spin_lock_bh_nested(lock, subclass) spin_lock_bh(lock)
15321 +# define spin_lock_irqsave_nested(lock, flags, subclass) \
15323 + typecheck(unsigned long, flags); \
15325 + spin_lock(lock); \
15329 +#define spin_lock_irqsave(lock, flags) \
15331 + typecheck(unsigned long, flags); \
15333 + spin_lock(lock); \
15336 +static inline unsigned long spin_lock_trace_flags(spinlock_t *lock)
15338 + unsigned long flags = 0;
15339 +#ifdef CONFIG_TRACE_IRQFLAGS
15340 + flags = rt_spin_lock_trace_flags(lock);
15342 + spin_lock(lock); /* lock_local */
15347 +/* FIXME: we need rt_spin_lock_nest_lock */
15348 +#define spin_lock_nest_lock(lock, nest_lock) spin_lock_nested(lock, 0)
15350 +#define spin_unlock(lock) rt_spin_unlock(lock)
15351 +#define spin_unlock_no_deboost(lock) rt_spin_unlock_no_deboost(lock)
15353 +#define spin_unlock_bh(lock) \
15355 + rt_spin_unlock(lock); \
15356 + local_bh_enable(); \
15359 +#define spin_unlock_irq(lock) spin_unlock(lock)
15361 +#define spin_unlock_irqrestore(lock, flags) \
15363 + typecheck(unsigned long, flags); \
15365 + spin_unlock(lock); \
15368 +#define spin_trylock_bh(lock) __cond_lock(lock, rt_spin_trylock_bh(lock))
15369 +#define spin_trylock_irq(lock) spin_trylock(lock)
15371 +#define spin_trylock_irqsave(lock, flags) \
15372 + rt_spin_trylock_irqsave(lock, &(flags))
15374 +#define spin_unlock_wait(lock) rt_spin_unlock_wait(lock)
15376 +#ifdef CONFIG_GENERIC_LOCKBREAK
15377 +# define spin_is_contended(lock) ((lock)->break_lock)
15379 +# define spin_is_contended(lock) (((void)(lock), 0))
15382 +static inline int spin_can_lock(spinlock_t *lock)
15384 + return !rt_mutex_is_locked(&lock->lock);
15387 +static inline int spin_is_locked(spinlock_t *lock)
15389 + return rt_mutex_is_locked(&lock->lock);
15392 +static inline void assert_spin_locked(spinlock_t *lock)
15394 + BUG_ON(!spin_is_locked(lock));
15397 +#define atomic_dec_and_lock(atomic, lock) \
15398 + atomic_dec_and_spin_lock(atomic, lock)
15401 diff --git a/include/linux/spinlock_types.h b/include/linux/spinlock_types.h
15402 index 73548eb13a5d..10bac715ea96 100644
15403 --- a/include/linux/spinlock_types.h
15404 +++ b/include/linux/spinlock_types.h
15406 * Released under the General Public License (GPL).
15409 -#if defined(CONFIG_SMP)
15410 -# include <asm/spinlock_types.h>
15412 -# include <linux/spinlock_types_up.h>
15415 -#include <linux/lockdep.h>
15417 -typedef struct raw_spinlock {
15418 - arch_spinlock_t raw_lock;
15419 -#ifdef CONFIG_GENERIC_LOCKBREAK
15420 - unsigned int break_lock;
15422 -#ifdef CONFIG_DEBUG_SPINLOCK
15423 - unsigned int magic, owner_cpu;
15426 -#ifdef CONFIG_DEBUG_LOCK_ALLOC
15427 - struct lockdep_map dep_map;
15431 -#define SPINLOCK_MAGIC 0xdead4ead
15433 -#define SPINLOCK_OWNER_INIT ((void *)-1L)
15435 -#ifdef CONFIG_DEBUG_LOCK_ALLOC
15436 -# define SPIN_DEP_MAP_INIT(lockname) .dep_map = { .name = #lockname }
15438 -# define SPIN_DEP_MAP_INIT(lockname)
15440 +#include <linux/spinlock_types_raw.h>
15442 -#ifdef CONFIG_DEBUG_SPINLOCK
15443 -# define SPIN_DEBUG_INIT(lockname) \
15444 - .magic = SPINLOCK_MAGIC, \
15445 - .owner_cpu = -1, \
15446 - .owner = SPINLOCK_OWNER_INIT,
15447 +#ifndef CONFIG_PREEMPT_RT_FULL
15448 +# include <linux/spinlock_types_nort.h>
15449 +# include <linux/rwlock_types.h>
15451 -# define SPIN_DEBUG_INIT(lockname)
15452 +# include <linux/rtmutex.h>
15453 +# include <linux/spinlock_types_rt.h>
15454 +# include <linux/rwlock_types_rt.h>
15457 -#define __RAW_SPIN_LOCK_INITIALIZER(lockname) \
15459 - .raw_lock = __ARCH_SPIN_LOCK_UNLOCKED, \
15460 - SPIN_DEBUG_INIT(lockname) \
15461 - SPIN_DEP_MAP_INIT(lockname) }
15463 -#define __RAW_SPIN_LOCK_UNLOCKED(lockname) \
15464 - (raw_spinlock_t) __RAW_SPIN_LOCK_INITIALIZER(lockname)
15466 -#define DEFINE_RAW_SPINLOCK(x) raw_spinlock_t x = __RAW_SPIN_LOCK_UNLOCKED(x)
15468 -typedef struct spinlock {
15470 - struct raw_spinlock rlock;
15472 -#ifdef CONFIG_DEBUG_LOCK_ALLOC
15473 -# define LOCK_PADSIZE (offsetof(struct raw_spinlock, dep_map))
15475 - u8 __padding[LOCK_PADSIZE];
15476 - struct lockdep_map dep_map;
15482 -#define __SPIN_LOCK_INITIALIZER(lockname) \
15483 - { { .rlock = __RAW_SPIN_LOCK_INITIALIZER(lockname) } }
15485 -#define __SPIN_LOCK_UNLOCKED(lockname) \
15486 - (spinlock_t ) __SPIN_LOCK_INITIALIZER(lockname)
15488 -#define DEFINE_SPINLOCK(x) spinlock_t x = __SPIN_LOCK_UNLOCKED(x)
15490 -#include <linux/rwlock_types.h>
15492 #endif /* __LINUX_SPINLOCK_TYPES_H */
15493 diff --git a/include/linux/spinlock_types_nort.h b/include/linux/spinlock_types_nort.h
15494 new file mode 100644
15495 index 000000000000..f1dac1fb1d6a
15497 +++ b/include/linux/spinlock_types_nort.h
15499 +#ifndef __LINUX_SPINLOCK_TYPES_NORT_H
15500 +#define __LINUX_SPINLOCK_TYPES_NORT_H
15502 +#ifndef __LINUX_SPINLOCK_TYPES_H
15503 +#error "Do not include directly. Include spinlock_types.h instead"
15507 + * The non RT version maps spinlocks to raw_spinlocks
15509 +typedef struct spinlock {
15511 + struct raw_spinlock rlock;
15513 +#ifdef CONFIG_DEBUG_LOCK_ALLOC
15514 +# define LOCK_PADSIZE (offsetof(struct raw_spinlock, dep_map))
15516 + u8 __padding[LOCK_PADSIZE];
15517 + struct lockdep_map dep_map;
15523 +#define __SPIN_LOCK_INITIALIZER(lockname) \
15524 + { { .rlock = __RAW_SPIN_LOCK_INITIALIZER(lockname) } }
15526 +#define __SPIN_LOCK_UNLOCKED(lockname) \
15527 + (spinlock_t ) __SPIN_LOCK_INITIALIZER(lockname)
15529 +#define DEFINE_SPINLOCK(x) spinlock_t x = __SPIN_LOCK_UNLOCKED(x)
15532 diff --git a/include/linux/spinlock_types_raw.h b/include/linux/spinlock_types_raw.h
15533 new file mode 100644
15534 index 000000000000..edffc4d53fc9
15536 +++ b/include/linux/spinlock_types_raw.h
15538 +#ifndef __LINUX_SPINLOCK_TYPES_RAW_H
15539 +#define __LINUX_SPINLOCK_TYPES_RAW_H
15541 +#if defined(CONFIG_SMP)
15542 +# include <asm/spinlock_types.h>
15544 +# include <linux/spinlock_types_up.h>
15547 +#include <linux/lockdep.h>
15549 +typedef struct raw_spinlock {
15550 + arch_spinlock_t raw_lock;
15551 +#ifdef CONFIG_GENERIC_LOCKBREAK
15552 + unsigned int break_lock;
15554 +#ifdef CONFIG_DEBUG_SPINLOCK
15555 + unsigned int magic, owner_cpu;
15558 +#ifdef CONFIG_DEBUG_LOCK_ALLOC
15559 + struct lockdep_map dep_map;
15563 +#define SPINLOCK_MAGIC 0xdead4ead
15565 +#define SPINLOCK_OWNER_INIT ((void *)-1L)
15567 +#ifdef CONFIG_DEBUG_LOCK_ALLOC
15568 +# define SPIN_DEP_MAP_INIT(lockname) .dep_map = { .name = #lockname }
15570 +# define SPIN_DEP_MAP_INIT(lockname)
15573 +#ifdef CONFIG_DEBUG_SPINLOCK
15574 +# define SPIN_DEBUG_INIT(lockname) \
15575 + .magic = SPINLOCK_MAGIC, \
15576 + .owner_cpu = -1, \
15577 + .owner = SPINLOCK_OWNER_INIT,
15579 +# define SPIN_DEBUG_INIT(lockname)
15582 +#define __RAW_SPIN_LOCK_INITIALIZER(lockname) \
15584 + .raw_lock = __ARCH_SPIN_LOCK_UNLOCKED, \
15585 + SPIN_DEBUG_INIT(lockname) \
15586 + SPIN_DEP_MAP_INIT(lockname) }
15588 +#define __RAW_SPIN_LOCK_UNLOCKED(lockname) \
15589 + (raw_spinlock_t) __RAW_SPIN_LOCK_INITIALIZER(lockname)
15591 +#define DEFINE_RAW_SPINLOCK(x) raw_spinlock_t x = __RAW_SPIN_LOCK_UNLOCKED(x)
15594 diff --git a/include/linux/spinlock_types_rt.h b/include/linux/spinlock_types_rt.h
15595 new file mode 100644
15596 index 000000000000..9fd431967abc
15598 +++ b/include/linux/spinlock_types_rt.h
15600 +#ifndef __LINUX_SPINLOCK_TYPES_RT_H
15601 +#define __LINUX_SPINLOCK_TYPES_RT_H
15603 +#ifndef __LINUX_SPINLOCK_TYPES_H
15604 +#error "Do not include directly. Include spinlock_types.h instead"
15607 +#include <linux/cache.h>
15610 + * PREEMPT_RT: spinlocks - an RT mutex plus lock-break field:
15612 +typedef struct spinlock {
15613 + struct rt_mutex lock;
15614 + unsigned int break_lock;
15615 +#ifdef CONFIG_DEBUG_LOCK_ALLOC
15616 + struct lockdep_map dep_map;
15620 +#ifdef CONFIG_DEBUG_RT_MUTEXES
15621 +# define __RT_SPIN_INITIALIZER(name) \
15623 + .wait_lock = __RAW_SPIN_LOCK_UNLOCKED(name.wait_lock), \
15624 + .save_state = 1, \
15625 + .file = __FILE__, \
15626 + .line = __LINE__ , \
15629 +# define __RT_SPIN_INITIALIZER(name) \
15631 + .wait_lock = __RAW_SPIN_LOCK_UNLOCKED(name.wait_lock), \
15632 + .save_state = 1, \
15637 +.wait_list = PLIST_HEAD_INIT_RAW((name).lock.wait_list, (name).lock.wait_lock)
15640 +#define __SPIN_LOCK_UNLOCKED(name) \
15641 + { .lock = __RT_SPIN_INITIALIZER(name.lock), \
15642 + SPIN_DEP_MAP_INIT(name) }
15644 +#define __DEFINE_SPINLOCK(name) \
15645 + spinlock_t name = __SPIN_LOCK_UNLOCKED(name)
15647 +#define DEFINE_SPINLOCK(name) \
15648 + spinlock_t name __cacheline_aligned_in_smp = __SPIN_LOCK_UNLOCKED(name)
15651 diff --git a/include/linux/srcu.h b/include/linux/srcu.h
15652 index f5f80c5643ac..ec1a8f01563c 100644
15653 --- a/include/linux/srcu.h
15654 +++ b/include/linux/srcu.h
15655 @@ -84,10 +84,10 @@ int init_srcu_struct(struct srcu_struct *sp);
15657 void process_srcu(struct work_struct *work);
15659 -#define __SRCU_STRUCT_INIT(name) \
15660 +#define __SRCU_STRUCT_INIT(name, pcpu_name) \
15662 .completed = -300, \
15663 - .per_cpu_ref = &name##_srcu_array, \
15664 + .per_cpu_ref = &pcpu_name, \
15665 .queue_lock = __SPIN_LOCK_UNLOCKED(name.queue_lock), \
15666 .running = false, \
15667 .batch_queue = RCU_BATCH_INIT(name.batch_queue), \
15668 @@ -104,7 +104,7 @@ void process_srcu(struct work_struct *work);
15670 #define __DEFINE_SRCU(name, is_static) \
15671 static DEFINE_PER_CPU(struct srcu_struct_array, name##_srcu_array);\
15672 - is_static struct srcu_struct name = __SRCU_STRUCT_INIT(name)
15673 + is_static struct srcu_struct name = __SRCU_STRUCT_INIT(name, name##_srcu_array)
15674 #define DEFINE_SRCU(name) __DEFINE_SRCU(name, /* not static */)
15675 #define DEFINE_STATIC_SRCU(name) __DEFINE_SRCU(name, static)
15677 diff --git a/include/linux/suspend.h b/include/linux/suspend.h
15678 index 8b6ec7ef0854..9b77d4cc929f 100644
15679 --- a/include/linux/suspend.h
15680 +++ b/include/linux/suspend.h
15681 @@ -194,6 +194,12 @@ struct platform_freeze_ops {
15685 +#if defined(CONFIG_SUSPEND) || defined(CONFIG_HIBERNATION)
15686 +extern bool pm_in_action;
15688 +# define pm_in_action false
15691 #ifdef CONFIG_SUSPEND
15693 * suspend_set_ops - set platform dependent suspend operations
15694 diff --git a/include/linux/swait.h b/include/linux/swait.h
15695 new file mode 100644
15696 index 000000000000..83f004a72320
15698 +++ b/include/linux/swait.h
15700 +#ifndef _LINUX_SWAIT_H
15701 +#define _LINUX_SWAIT_H
15703 +#include <linux/list.h>
15704 +#include <linux/stddef.h>
15705 +#include <linux/spinlock.h>
15706 +#include <asm/current.h>
15709 + * Simple wait queues
15711 + * While these are very similar to the other/complex wait queues (wait.h) the
15712 + * most important difference is that the simple waitqueue allows for
15713 + * deterministic behaviour -- IOW it has strictly bounded IRQ and lock hold
15716 + * In order to make this so, we had to drop a fair number of features of the
15717 + * other waitqueue code; notably:
15719 + * - mixing INTERRUPTIBLE and UNINTERRUPTIBLE sleeps on the same waitqueue;
15720 + * all wakeups are TASK_NORMAL in order to avoid O(n) lookups for the right
15723 + * - the exclusive mode; because this requires preserving the list order
15724 + * and this is hard.
15726 + * - custom wake functions; because you cannot give any guarantees about
15729 + * As a side effect of this; the data structures are slimmer.
15731 + * One would recommend using this wait queue where possible.
15734 +struct task_struct;
15736 +struct swait_queue_head {
15737 + raw_spinlock_t lock;
15738 + struct list_head task_list;
15741 +struct swait_queue {
15742 + struct task_struct *task;
15743 + struct list_head task_list;
15746 +#define __SWAITQUEUE_INITIALIZER(name) { \
15747 + .task = current, \
15748 + .task_list = LIST_HEAD_INIT((name).task_list), \
15751 +#define DECLARE_SWAITQUEUE(name) \
15752 + struct swait_queue name = __SWAITQUEUE_INITIALIZER(name)
15754 +#define __SWAIT_QUEUE_HEAD_INITIALIZER(name) { \
15755 + .lock = __RAW_SPIN_LOCK_UNLOCKED(name.lock), \
15756 + .task_list = LIST_HEAD_INIT((name).task_list), \
15759 +#define DECLARE_SWAIT_QUEUE_HEAD(name) \
15760 + struct swait_queue_head name = __SWAIT_QUEUE_HEAD_INITIALIZER(name)
15762 +extern void __init_swait_queue_head(struct swait_queue_head *q, const char *name,
15763 + struct lock_class_key *key);
15765 +#define init_swait_queue_head(q) \
15767 + static struct lock_class_key __key; \
15768 + __init_swait_queue_head((q), #q, &__key); \
15771 +#ifdef CONFIG_LOCKDEP
15772 +# define __SWAIT_QUEUE_HEAD_INIT_ONSTACK(name) \
15773 + ({ init_swait_queue_head(&name); name; })
15774 +# define DECLARE_SWAIT_QUEUE_HEAD_ONSTACK(name) \
15775 + struct swait_queue_head name = __SWAIT_QUEUE_HEAD_INIT_ONSTACK(name)
15777 +# define DECLARE_SWAIT_QUEUE_HEAD_ONSTACK(name) \
15778 + DECLARE_SWAIT_QUEUE_HEAD(name)
15781 +static inline int swait_active(struct swait_queue_head *q)
15783 + return !list_empty(&q->task_list);
15786 +extern void swake_up(struct swait_queue_head *q);
15787 +extern void swake_up_all(struct swait_queue_head *q);
15788 +extern void swake_up_locked(struct swait_queue_head *q);
15789 +extern void swake_up_all_locked(struct swait_queue_head *q);
15791 +extern void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait);
15792 +extern void prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait, int state);
15793 +extern long prepare_to_swait_event(struct swait_queue_head *q, struct swait_queue *wait, int state);
15795 +extern void __finish_swait(struct swait_queue_head *q, struct swait_queue *wait);
15796 +extern void finish_swait(struct swait_queue_head *q, struct swait_queue *wait);
15798 +/* as per ___wait_event() but for swait, therefore "exclusive == 0" */
15799 +#define ___swait_event(wq, condition, state, ret, cmd) \
15801 + struct swait_queue __wait; \
15802 + long __ret = ret; \
15804 + INIT_LIST_HEAD(&__wait.task_list); \
15806 + long __int = prepare_to_swait_event(&wq, &__wait, state);\
15811 + if (___wait_is_interruptible(state) && __int) { \
15818 + finish_swait(&wq, &__wait); \
15822 +#define __swait_event(wq, condition) \
15823 + (void)___swait_event(wq, condition, TASK_UNINTERRUPTIBLE, 0, \
15826 +#define swait_event(wq, condition) \
15830 + __swait_event(wq, condition); \
15833 +#define __swait_event_timeout(wq, condition, timeout) \
15834 + ___swait_event(wq, ___wait_cond_timeout(condition), \
15835 + TASK_UNINTERRUPTIBLE, timeout, \
15836 + __ret = schedule_timeout(__ret))
15838 +#define swait_event_timeout(wq, condition, timeout) \
15840 + long __ret = timeout; \
15841 + if (!___wait_cond_timeout(condition)) \
15842 + __ret = __swait_event_timeout(wq, condition, timeout); \
15846 +#define __swait_event_interruptible(wq, condition) \
15847 + ___swait_event(wq, condition, TASK_INTERRUPTIBLE, 0, \
15850 +#define swait_event_interruptible(wq, condition) \
15853 + if (!(condition)) \
15854 + __ret = __swait_event_interruptible(wq, condition); \
15858 +#define __swait_event_interruptible_timeout(wq, condition, timeout) \
15859 + ___swait_event(wq, ___wait_cond_timeout(condition), \
15860 + TASK_INTERRUPTIBLE, timeout, \
15861 + __ret = schedule_timeout(__ret))
15863 +#define swait_event_interruptible_timeout(wq, condition, timeout) \
15865 + long __ret = timeout; \
15866 + if (!___wait_cond_timeout(condition)) \
15867 + __ret = __swait_event_interruptible_timeout(wq, \
15868 + condition, timeout); \
15872 +#endif /* _LINUX_SWAIT_H */
15873 diff --git a/include/linux/swap.h b/include/linux/swap.h
15874 index d8ca2eaa3a8b..19e038054914 100644
15875 --- a/include/linux/swap.h
15876 +++ b/include/linux/swap.h
15878 #include <linux/fs.h>
15879 #include <linux/atomic.h>
15880 #include <linux/page-flags.h>
15881 +#include <linux/locallock.h>
15882 #include <asm/page.h>
15884 struct notifier_block;
15885 @@ -252,7 +253,8 @@ struct swap_info_struct {
15886 void *workingset_eviction(struct address_space *mapping, struct page *page);
15887 bool workingset_refault(void *shadow);
15888 void workingset_activation(struct page *page);
15889 -extern struct list_lru workingset_shadow_nodes;
15890 +extern struct list_lru __workingset_shadow_nodes;
15891 +DECLARE_LOCAL_IRQ_LOCK(workingset_shadow_lock);
15893 static inline unsigned int workingset_node_pages(struct radix_tree_node *node)
15895 @@ -298,6 +300,7 @@ extern unsigned long nr_free_pagecache_pages(void);
15898 /* linux/mm/swap.c */
15899 +DECLARE_LOCAL_IRQ_LOCK(swapvec_lock);
15900 extern void lru_cache_add(struct page *);
15901 extern void lru_cache_add_anon(struct page *page);
15902 extern void lru_cache_add_file(struct page *page);
15903 diff --git a/include/linux/swork.h b/include/linux/swork.h
15904 new file mode 100644
15905 index 000000000000..f175fa9a6016
15907 +++ b/include/linux/swork.h
15909 +#ifndef _LINUX_SWORK_H
15910 +#define _LINUX_SWORK_H
15912 +#include <linux/list.h>
15914 +struct swork_event {
15915 + struct list_head item;
15916 + unsigned long flags;
15917 + void (*func)(struct swork_event *);
15920 +static inline void INIT_SWORK(struct swork_event *event,
15921 + void (*func)(struct swork_event *))
15923 + event->flags = 0;
15924 + event->func = func;
15927 +bool swork_queue(struct swork_event *sev);
15929 +int swork_get(void);
15930 +void swork_put(void);
15932 +#endif /* _LINUX_SWORK_H */
15933 diff --git a/include/linux/thread_info.h b/include/linux/thread_info.h
15934 index ff307b548ed3..be9f9dc6a4e1 100644
15935 --- a/include/linux/thread_info.h
15936 +++ b/include/linux/thread_info.h
15937 @@ -102,7 +102,17 @@ static inline int test_ti_thread_flag(struct thread_info *ti, int flag)
15938 #define test_thread_flag(flag) \
15939 test_ti_thread_flag(current_thread_info(), flag)
15941 -#define tif_need_resched() test_thread_flag(TIF_NEED_RESCHED)
15942 +#ifdef CONFIG_PREEMPT_LAZY
15943 +#define tif_need_resched() (test_thread_flag(TIF_NEED_RESCHED) || \
15944 + test_thread_flag(TIF_NEED_RESCHED_LAZY))
15945 +#define tif_need_resched_now() (test_thread_flag(TIF_NEED_RESCHED))
15946 +#define tif_need_resched_lazy() test_thread_flag(TIF_NEED_RESCHED_LAZY))
15949 +#define tif_need_resched() test_thread_flag(TIF_NEED_RESCHED)
15950 +#define tif_need_resched_now() test_thread_flag(TIF_NEED_RESCHED)
15951 +#define tif_need_resched_lazy() 0
15954 #if defined TIF_RESTORE_SIGMASK && !defined HAVE_SET_RESTORE_SIGMASK
15956 diff --git a/include/linux/timer.h b/include/linux/timer.h
15957 index 61aa61dc410c..299d2b78591f 100644
15958 --- a/include/linux/timer.h
15959 +++ b/include/linux/timer.h
15960 @@ -225,7 +225,7 @@ extern void add_timer(struct timer_list *timer);
15962 extern int try_to_del_timer_sync(struct timer_list *timer);
15965 +#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT_RT_FULL)
15966 extern int del_timer_sync(struct timer_list *timer);
15968 # define del_timer_sync(t) del_timer(t)
15969 diff --git a/include/linux/trace_events.h b/include/linux/trace_events.h
15970 index 925730bc9fc1..a591f414da6f 100644
15971 --- a/include/linux/trace_events.h
15972 +++ b/include/linux/trace_events.h
15973 @@ -66,6 +66,9 @@ struct trace_entry {
15974 unsigned char flags;
15975 unsigned char preempt_count;
15977 + unsigned short migrate_disable;
15978 + unsigned short padding;
15979 + unsigned char preempt_lazy_count;
15982 #define TRACE_EVENT_TYPE_MAX \
15983 diff --git a/include/linux/uaccess.h b/include/linux/uaccess.h
15984 index 558129af828a..cf5c472bbc79 100644
15985 --- a/include/linux/uaccess.h
15986 +++ b/include/linux/uaccess.h
15987 @@ -24,6 +24,7 @@ static __always_inline void pagefault_disabled_dec(void)
15989 static inline void pagefault_disable(void)
15991 + migrate_disable();
15992 pagefault_disabled_inc();
15994 * make sure to have issued the store before a pagefault
15995 @@ -40,6 +41,7 @@ static inline void pagefault_enable(void)
15998 pagefault_disabled_dec();
15999 + migrate_enable();
16003 diff --git a/include/linux/uprobes.h b/include/linux/uprobes.h
16004 index 4a29c75b146e..0a294e950df8 100644
16005 --- a/include/linux/uprobes.h
16006 +++ b/include/linux/uprobes.h
16008 #include <linux/errno.h>
16009 #include <linux/rbtree.h>
16010 #include <linux/types.h>
16011 +#include <linux/wait.h>
16013 struct vm_area_struct;
16015 diff --git a/include/linux/vmstat.h b/include/linux/vmstat.h
16016 index 3e5d9075960f..7eaa847cd5a5 100644
16017 --- a/include/linux/vmstat.h
16018 +++ b/include/linux/vmstat.h
16019 @@ -33,7 +33,9 @@ DECLARE_PER_CPU(struct vm_event_state, vm_event_states);
16021 static inline void __count_vm_event(enum vm_event_item item)
16023 + preempt_disable_rt();
16024 raw_cpu_inc(vm_event_states.event[item]);
16025 + preempt_enable_rt();
16028 static inline void count_vm_event(enum vm_event_item item)
16029 @@ -43,7 +45,9 @@ static inline void count_vm_event(enum vm_event_item item)
16031 static inline void __count_vm_events(enum vm_event_item item, long delta)
16033 + preempt_disable_rt();
16034 raw_cpu_add(vm_event_states.event[item], delta);
16035 + preempt_enable_rt();
16038 static inline void count_vm_events(enum vm_event_item item, long delta)
16039 diff --git a/include/linux/wait.h b/include/linux/wait.h
16040 index 513b36f04dfd..981c8a840f96 100644
16041 --- a/include/linux/wait.h
16042 +++ b/include/linux/wait.h
16044 #include <linux/spinlock.h>
16045 #include <asm/current.h>
16046 #include <uapi/linux/wait.h>
16047 +#include <linux/atomic.h>
16049 typedef struct __wait_queue wait_queue_t;
16050 typedef int (*wait_queue_func_t)(wait_queue_t *wait, unsigned mode, int flags, void *key);
16051 diff --git a/include/net/dst.h b/include/net/dst.h
16052 index c7329dcd90cc..35c3dba16728 100644
16053 --- a/include/net/dst.h
16054 +++ b/include/net/dst.h
16055 @@ -437,7 +437,7 @@ static inline void dst_confirm(struct dst_entry *dst)
16056 static inline int dst_neigh_output(struct dst_entry *dst, struct neighbour *n,
16057 struct sk_buff *skb)
16059 - const struct hh_cache *hh;
16060 + struct hh_cache *hh;
16062 if (dst->pending_confirm) {
16063 unsigned long now = jiffies;
16064 diff --git a/include/net/neighbour.h b/include/net/neighbour.h
16065 index 8b683841e574..bf656008f6e7 100644
16066 --- a/include/net/neighbour.h
16067 +++ b/include/net/neighbour.h
16068 @@ -446,7 +446,7 @@ static inline int neigh_hh_bridge(struct hh_cache *hh, struct sk_buff *skb)
16072 -static inline int neigh_hh_output(const struct hh_cache *hh, struct sk_buff *skb)
16073 +static inline int neigh_hh_output(struct hh_cache *hh, struct sk_buff *skb)
16077 @@ -501,7 +501,7 @@ struct neighbour_cb {
16079 #define NEIGH_CB(skb) ((struct neighbour_cb *)(skb)->cb)
16081 -static inline void neigh_ha_snapshot(char *dst, const struct neighbour *n,
16082 +static inline void neigh_ha_snapshot(char *dst, struct neighbour *n,
16083 const struct net_device *dev)
16086 diff --git a/include/net/netns/ipv4.h b/include/net/netns/ipv4.h
16087 index c68926b4899c..dd0751e76065 100644
16088 --- a/include/net/netns/ipv4.h
16089 +++ b/include/net/netns/ipv4.h
16090 @@ -70,6 +70,7 @@ struct netns_ipv4 {
16092 int sysctl_icmp_echo_ignore_all;
16093 int sysctl_icmp_echo_ignore_broadcasts;
16094 + int sysctl_icmp_echo_sysrq;
16095 int sysctl_icmp_ignore_bogus_error_responses;
16096 int sysctl_icmp_ratelimit;
16097 int sysctl_icmp_ratemask;
16098 diff --git a/include/trace/events/hist.h b/include/trace/events/hist.h
16099 new file mode 100644
16100 index 000000000000..f7710de1b1f3
16102 +++ b/include/trace/events/hist.h
16104 +#undef TRACE_SYSTEM
16105 +#define TRACE_SYSTEM hist
16107 +#if !defined(_TRACE_HIST_H) || defined(TRACE_HEADER_MULTI_READ)
16108 +#define _TRACE_HIST_H
16110 +#include "latency_hist.h"
16111 +#include <linux/tracepoint.h>
16113 +#if !defined(CONFIG_PREEMPT_OFF_HIST) && !defined(CONFIG_INTERRUPT_OFF_HIST)
16114 +#define trace_preemptirqsoff_hist(a, b)
16115 +#define trace_preemptirqsoff_hist_rcuidle(a, b)
16117 +TRACE_EVENT(preemptirqsoff_hist,
16119 + TP_PROTO(int reason, int starthist),
16121 + TP_ARGS(reason, starthist),
16123 + TP_STRUCT__entry(
16124 + __field(int, reason)
16125 + __field(int, starthist)
16129 + __entry->reason = reason;
16130 + __entry->starthist = starthist;
16133 + TP_printk("reason=%s starthist=%s", getaction(__entry->reason),
16134 + __entry->starthist ? "start" : "stop")
16138 +#ifndef CONFIG_MISSED_TIMER_OFFSETS_HIST
16139 +#define trace_hrtimer_interrupt(a, b, c, d)
16141 +TRACE_EVENT(hrtimer_interrupt,
16143 + TP_PROTO(int cpu, long long offset, struct task_struct *curr,
16144 + struct task_struct *task),
16146 + TP_ARGS(cpu, offset, curr, task),
16148 + TP_STRUCT__entry(
16149 + __field(int, cpu)
16150 + __field(long long, offset)
16151 + __array(char, ccomm, TASK_COMM_LEN)
16152 + __field(int, cprio)
16153 + __array(char, tcomm, TASK_COMM_LEN)
16154 + __field(int, tprio)
16158 + __entry->cpu = cpu;
16159 + __entry->offset = offset;
16160 + memcpy(__entry->ccomm, curr->comm, TASK_COMM_LEN);
16161 + __entry->cprio = curr->prio;
16162 + memcpy(__entry->tcomm, task != NULL ? task->comm : "<none>",
16163 + task != NULL ? TASK_COMM_LEN : 7);
16164 + __entry->tprio = task != NULL ? task->prio : -1;
16167 + TP_printk("cpu=%d offset=%lld curr=%s[%d] thread=%s[%d]",
16168 + __entry->cpu, __entry->offset, __entry->ccomm,
16169 + __entry->cprio, __entry->tcomm, __entry->tprio)
16173 +#endif /* _TRACE_HIST_H */
16175 +/* This part must be outside protection */
16176 +#include <trace/define_trace.h>
16177 diff --git a/include/trace/events/latency_hist.h b/include/trace/events/latency_hist.h
16178 new file mode 100644
16179 index 000000000000..d3f2fbd560b1
16181 +++ b/include/trace/events/latency_hist.h
16183 +#ifndef _LATENCY_HIST_H
16184 +#define _LATENCY_HIST_H
16186 +enum hist_action {
16195 +static char *actions[] = {
16204 +static inline char *getaction(int action)
16206 + if (action >= 0 && action <= sizeof(actions)/sizeof(actions[0]))
16207 + return actions[action];
16208 + return "unknown";
16211 +#endif /* _LATENCY_HIST_H */
16212 diff --git a/include/trace/events/writeback.h b/include/trace/events/writeback.h
16213 index fff846b512e6..73614ce1d204 100644
16214 --- a/include/trace/events/writeback.h
16215 +++ b/include/trace/events/writeback.h
16216 @@ -134,58 +134,28 @@ DEFINE_EVENT(writeback_dirty_inode_template, writeback_dirty_inode,
16217 #ifdef CREATE_TRACE_POINTS
16218 #ifdef CONFIG_CGROUP_WRITEBACK
16220 -static inline size_t __trace_wb_cgroup_size(struct bdi_writeback *wb)
16221 +static inline unsigned int __trace_wb_assign_cgroup(struct bdi_writeback *wb)
16223 - return kernfs_path_len(wb->memcg_css->cgroup->kn) + 1;
16224 + return wb->memcg_css->cgroup->kn->ino;
16227 -static inline void __trace_wb_assign_cgroup(char *buf, struct bdi_writeback *wb)
16229 - struct cgroup *cgrp = wb->memcg_css->cgroup;
16232 - path = cgroup_path(cgrp, buf, kernfs_path_len(cgrp->kn) + 1);
16233 - WARN_ON_ONCE(path != buf);
16236 -static inline size_t __trace_wbc_cgroup_size(struct writeback_control *wbc)
16239 - return __trace_wb_cgroup_size(wbc->wb);
16244 -static inline void __trace_wbc_assign_cgroup(char *buf,
16245 - struct writeback_control *wbc)
16246 +static inline unsigned int __trace_wbc_assign_cgroup(struct writeback_control *wbc)
16249 - __trace_wb_assign_cgroup(buf, wbc->wb);
16250 + return __trace_wb_assign_cgroup(wbc->wb);
16252 - strcpy(buf, "/");
16256 #else /* CONFIG_CGROUP_WRITEBACK */
16258 -static inline size_t __trace_wb_cgroup_size(struct bdi_writeback *wb)
16263 -static inline void __trace_wb_assign_cgroup(char *buf, struct bdi_writeback *wb)
16265 - strcpy(buf, "/");
16268 -static inline size_t __trace_wbc_cgroup_size(struct writeback_control *wbc)
16269 +static inline unsigned int __trace_wb_assign_cgroup(struct bdi_writeback *wb)
16275 -static inline void __trace_wbc_assign_cgroup(char *buf,
16276 - struct writeback_control *wbc)
16277 +static inline unsigned int __trace_wbc_assign_cgroup(struct writeback_control *wbc)
16279 - strcpy(buf, "/");
16283 #endif /* CONFIG_CGROUP_WRITEBACK */
16284 @@ -201,7 +171,7 @@ DECLARE_EVENT_CLASS(writeback_write_inode_template,
16285 __array(char, name, 32)
16286 __field(unsigned long, ino)
16287 __field(int, sync_mode)
16288 - __dynamic_array(char, cgroup, __trace_wbc_cgroup_size(wbc))
16289 + __field(unsigned int, cgroup_ino)
16293 @@ -209,14 +179,14 @@ DECLARE_EVENT_CLASS(writeback_write_inode_template,
16294 dev_name(inode_to_bdi(inode)->dev), 32);
16295 __entry->ino = inode->i_ino;
16296 __entry->sync_mode = wbc->sync_mode;
16297 - __trace_wbc_assign_cgroup(__get_str(cgroup), wbc);
16298 + __entry->cgroup_ino = __trace_wbc_assign_cgroup(wbc);
16301 - TP_printk("bdi %s: ino=%lu sync_mode=%d cgroup=%s",
16302 + TP_printk("bdi %s: ino=%lu sync_mode=%d cgroup_ino=%u",
16305 __entry->sync_mode,
16306 - __get_str(cgroup)
16307 + __entry->cgroup_ino
16311 @@ -246,7 +216,7 @@ DECLARE_EVENT_CLASS(writeback_work_class,
16312 __field(int, range_cyclic)
16313 __field(int, for_background)
16314 __field(int, reason)
16315 - __dynamic_array(char, cgroup, __trace_wb_cgroup_size(wb))
16316 + __field(unsigned int, cgroup_ino)
16319 strncpy(__entry->name,
16320 @@ -258,10 +228,10 @@ DECLARE_EVENT_CLASS(writeback_work_class,
16321 __entry->range_cyclic = work->range_cyclic;
16322 __entry->for_background = work->for_background;
16323 __entry->reason = work->reason;
16324 - __trace_wb_assign_cgroup(__get_str(cgroup), wb);
16325 + __entry->cgroup_ino = __trace_wb_assign_cgroup(wb);
16327 TP_printk("bdi %s: sb_dev %d:%d nr_pages=%ld sync_mode=%d "
16328 - "kupdate=%d range_cyclic=%d background=%d reason=%s cgroup=%s",
16329 + "kupdate=%d range_cyclic=%d background=%d reason=%s cgroup_ino=%u",
16331 MAJOR(__entry->sb_dev), MINOR(__entry->sb_dev),
16333 @@ -270,7 +240,7 @@ DECLARE_EVENT_CLASS(writeback_work_class,
16334 __entry->range_cyclic,
16335 __entry->for_background,
16336 __print_symbolic(__entry->reason, WB_WORK_REASON),
16337 - __get_str(cgroup)
16338 + __entry->cgroup_ino
16341 #define DEFINE_WRITEBACK_WORK_EVENT(name) \
16342 @@ -300,15 +270,15 @@ DECLARE_EVENT_CLASS(writeback_class,
16345 __array(char, name, 32)
16346 - __dynamic_array(char, cgroup, __trace_wb_cgroup_size(wb))
16347 + __field(unsigned int, cgroup_ino)
16350 strncpy(__entry->name, dev_name(wb->bdi->dev), 32);
16351 - __trace_wb_assign_cgroup(__get_str(cgroup), wb);
16352 + __entry->cgroup_ino = __trace_wb_assign_cgroup(wb);
16354 - TP_printk("bdi %s: cgroup=%s",
16355 + TP_printk("bdi %s: cgroup_ino=%u",
16357 - __get_str(cgroup)
16358 + __entry->cgroup_ino
16361 #define DEFINE_WRITEBACK_EVENT(name) \
16362 @@ -347,7 +317,7 @@ DECLARE_EVENT_CLASS(wbc_class,
16363 __field(int, range_cyclic)
16364 __field(long, range_start)
16365 __field(long, range_end)
16366 - __dynamic_array(char, cgroup, __trace_wbc_cgroup_size(wbc))
16367 + __field(unsigned int, cgroup_ino)
16371 @@ -361,12 +331,12 @@ DECLARE_EVENT_CLASS(wbc_class,
16372 __entry->range_cyclic = wbc->range_cyclic;
16373 __entry->range_start = (long)wbc->range_start;
16374 __entry->range_end = (long)wbc->range_end;
16375 - __trace_wbc_assign_cgroup(__get_str(cgroup), wbc);
16376 + __entry->cgroup_ino = __trace_wbc_assign_cgroup(wbc);
16379 TP_printk("bdi %s: towrt=%ld skip=%ld mode=%d kupd=%d "
16380 "bgrd=%d reclm=%d cyclic=%d "
16381 - "start=0x%lx end=0x%lx cgroup=%s",
16382 + "start=0x%lx end=0x%lx cgroup_ino=%u",
16384 __entry->nr_to_write,
16385 __entry->pages_skipped,
16386 @@ -377,7 +347,7 @@ DECLARE_EVENT_CLASS(wbc_class,
16387 __entry->range_cyclic,
16388 __entry->range_start,
16389 __entry->range_end,
16390 - __get_str(cgroup)
16391 + __entry->cgroup_ino
16395 @@ -398,7 +368,7 @@ TRACE_EVENT(writeback_queue_io,
16397 __field(int, moved)
16398 __field(int, reason)
16399 - __dynamic_array(char, cgroup, __trace_wb_cgroup_size(wb))
16400 + __field(unsigned int, cgroup_ino)
16403 unsigned long *older_than_this = work->older_than_this;
16404 @@ -408,15 +378,15 @@ TRACE_EVENT(writeback_queue_io,
16405 (jiffies - *older_than_this) * 1000 / HZ : -1;
16406 __entry->moved = moved;
16407 __entry->reason = work->reason;
16408 - __trace_wb_assign_cgroup(__get_str(cgroup), wb);
16409 + __entry->cgroup_ino = __trace_wb_assign_cgroup(wb);
16411 - TP_printk("bdi %s: older=%lu age=%ld enqueue=%d reason=%s cgroup=%s",
16412 + TP_printk("bdi %s: older=%lu age=%ld enqueue=%d reason=%s cgroup_ino=%u",
16414 __entry->older, /* older_than_this in jiffies */
16415 __entry->age, /* older_than_this in relative milliseconds */
16417 __print_symbolic(__entry->reason, WB_WORK_REASON),
16418 - __get_str(cgroup)
16419 + __entry->cgroup_ino
16423 @@ -484,7 +454,7 @@ TRACE_EVENT(bdi_dirty_ratelimit,
16424 __field(unsigned long, dirty_ratelimit)
16425 __field(unsigned long, task_ratelimit)
16426 __field(unsigned long, balanced_dirty_ratelimit)
16427 - __dynamic_array(char, cgroup, __trace_wb_cgroup_size(wb))
16428 + __field(unsigned int, cgroup_ino)
16432 @@ -496,13 +466,13 @@ TRACE_EVENT(bdi_dirty_ratelimit,
16433 __entry->task_ratelimit = KBps(task_ratelimit);
16434 __entry->balanced_dirty_ratelimit =
16435 KBps(wb->balanced_dirty_ratelimit);
16436 - __trace_wb_assign_cgroup(__get_str(cgroup), wb);
16437 + __entry->cgroup_ino = __trace_wb_assign_cgroup(wb);
16440 TP_printk("bdi %s: "
16441 "write_bw=%lu awrite_bw=%lu dirty_rate=%lu "
16442 "dirty_ratelimit=%lu task_ratelimit=%lu "
16443 - "balanced_dirty_ratelimit=%lu cgroup=%s",
16444 + "balanced_dirty_ratelimit=%lu cgroup_ino=%u",
16446 __entry->write_bw, /* write bandwidth */
16447 __entry->avg_write_bw, /* avg write bandwidth */
16448 @@ -510,7 +480,7 @@ TRACE_EVENT(bdi_dirty_ratelimit,
16449 __entry->dirty_ratelimit, /* base ratelimit */
16450 __entry->task_ratelimit, /* ratelimit with position control */
16451 __entry->balanced_dirty_ratelimit, /* the balanced ratelimit */
16452 - __get_str(cgroup)
16453 + __entry->cgroup_ino
16457 @@ -548,7 +518,7 @@ TRACE_EVENT(balance_dirty_pages,
16458 __field( long, pause)
16459 __field(unsigned long, period)
16460 __field( long, think)
16461 - __dynamic_array(char, cgroup, __trace_wb_cgroup_size(wb))
16462 + __field(unsigned int, cgroup_ino)
16466 @@ -571,7 +541,7 @@ TRACE_EVENT(balance_dirty_pages,
16467 __entry->period = period * 1000 / HZ;
16468 __entry->pause = pause * 1000 / HZ;
16469 __entry->paused = (jiffies - start_time) * 1000 / HZ;
16470 - __trace_wb_assign_cgroup(__get_str(cgroup), wb);
16471 + __entry->cgroup_ino = __trace_wb_assign_cgroup(wb);
16475 @@ -580,7 +550,7 @@ TRACE_EVENT(balance_dirty_pages,
16476 "bdi_setpoint=%lu bdi_dirty=%lu "
16477 "dirty_ratelimit=%lu task_ratelimit=%lu "
16478 "dirtied=%u dirtied_pause=%u "
16479 - "paused=%lu pause=%ld period=%lu think=%ld cgroup=%s",
16480 + "paused=%lu pause=%ld period=%lu think=%ld cgroup_ino=%u",
16484 @@ -595,7 +565,7 @@ TRACE_EVENT(balance_dirty_pages,
16485 __entry->pause, /* ms */
16486 __entry->period, /* ms */
16487 __entry->think, /* ms */
16488 - __get_str(cgroup)
16489 + __entry->cgroup_ino
16493 @@ -609,8 +579,7 @@ TRACE_EVENT(writeback_sb_inodes_requeue,
16494 __field(unsigned long, ino)
16495 __field(unsigned long, state)
16496 __field(unsigned long, dirtied_when)
16497 - __dynamic_array(char, cgroup,
16498 - __trace_wb_cgroup_size(inode_to_wb(inode)))
16499 + __field(unsigned int, cgroup_ino)
16503 @@ -619,16 +588,16 @@ TRACE_EVENT(writeback_sb_inodes_requeue,
16504 __entry->ino = inode->i_ino;
16505 __entry->state = inode->i_state;
16506 __entry->dirtied_when = inode->dirtied_when;
16507 - __trace_wb_assign_cgroup(__get_str(cgroup), inode_to_wb(inode));
16508 + __entry->cgroup_ino = __trace_wb_assign_cgroup(inode_to_wb(inode));
16511 - TP_printk("bdi %s: ino=%lu state=%s dirtied_when=%lu age=%lu cgroup=%s",
16512 + TP_printk("bdi %s: ino=%lu state=%s dirtied_when=%lu age=%lu cgroup_ino=%u",
16515 show_inode_state(__entry->state),
16516 __entry->dirtied_when,
16517 (jiffies - __entry->dirtied_when) / HZ,
16518 - __get_str(cgroup)
16519 + __entry->cgroup_ino
16523 @@ -684,7 +653,7 @@ DECLARE_EVENT_CLASS(writeback_single_inode_template,
16524 __field(unsigned long, writeback_index)
16525 __field(long, nr_to_write)
16526 __field(unsigned long, wrote)
16527 - __dynamic_array(char, cgroup, __trace_wbc_cgroup_size(wbc))
16528 + __field(unsigned int, cgroup_ino)
16532 @@ -696,11 +665,11 @@ DECLARE_EVENT_CLASS(writeback_single_inode_template,
16533 __entry->writeback_index = inode->i_mapping->writeback_index;
16534 __entry->nr_to_write = nr_to_write;
16535 __entry->wrote = nr_to_write - wbc->nr_to_write;
16536 - __trace_wbc_assign_cgroup(__get_str(cgroup), wbc);
16537 + __entry->cgroup_ino = __trace_wbc_assign_cgroup(wbc);
16540 TP_printk("bdi %s: ino=%lu state=%s dirtied_when=%lu age=%lu "
16541 - "index=%lu to_write=%ld wrote=%lu cgroup=%s",
16542 + "index=%lu to_write=%ld wrote=%lu cgroup_ino=%u",
16545 show_inode_state(__entry->state),
16546 @@ -709,7 +678,7 @@ DECLARE_EVENT_CLASS(writeback_single_inode_template,
16547 __entry->writeback_index,
16548 __entry->nr_to_write,
16550 - __get_str(cgroup)
16551 + __entry->cgroup_ino
16555 diff --git a/init/Kconfig b/init/Kconfig
16556 index 235c7a2c0d20..a7c81c0911da 100644
16559 @@ -498,7 +498,7 @@ config TINY_RCU
16562 bool "Make expert-level adjustments to RCU configuration"
16564 + default y if PREEMPT_RT_FULL
16566 This option needs to be enabled if you wish to make
16567 expert-level adjustments to RCU configuration. By default,
16568 @@ -614,7 +614,7 @@ config RCU_FANOUT_LEAF
16570 config RCU_FAST_NO_HZ
16571 bool "Accelerate last non-dyntick-idle CPU's grace periods"
16572 - depends on NO_HZ_COMMON && SMP && RCU_EXPERT
16573 + depends on NO_HZ_COMMON && SMP && RCU_EXPERT && !PREEMPT_RT_FULL
16576 This option permits CPUs to enter dynticks-idle state even if
16577 @@ -641,7 +641,7 @@ config TREE_RCU_TRACE
16579 bool "Enable RCU priority boosting"
16580 depends on RT_MUTEXES && PREEMPT_RCU && RCU_EXPERT
16582 + default y if PREEMPT_RT_FULL
16584 This option boosts the priority of preempted RCU readers that
16585 block the current preemptible RCU grace period for too long.
16586 @@ -1106,6 +1106,7 @@ config CFS_BANDWIDTH
16587 config RT_GROUP_SCHED
16588 bool "Group scheduling for SCHED_RR/FIFO"
16589 depends on CGROUP_SCHED
16590 + depends on !PREEMPT_RT_FULL
16593 This feature lets you explicitly allocate real CPU bandwidth
16594 @@ -1719,6 +1720,7 @@ choice
16598 + depends on !PREEMPT_RT_FULL
16600 The regular slab allocator that is established and known to work
16601 well in all environments. It organizes cache hot objects in
16602 @@ -1737,6 +1739,7 @@ config SLUB
16605 bool "SLOB (Simple Allocator)"
16606 + depends on !PREEMPT_RT_FULL
16608 SLOB replaces the stock allocator with a drastically simpler
16609 allocator. SLOB is generally more space efficient but
16610 @@ -1746,7 +1749,7 @@ endchoice
16612 config SLUB_CPU_PARTIAL
16614 - depends on SLUB && SMP
16615 + depends on SLUB && SMP && !PREEMPT_RT_FULL
16616 bool "SLUB per cpu partial cache"
16618 Per cpu partial caches accellerate objects allocation and freeing
16619 diff --git a/init/Makefile b/init/Makefile
16620 index 7bc47ee31c36..88cf473554e0 100644
16621 --- a/init/Makefile
16622 +++ b/init/Makefile
16623 @@ -33,4 +33,4 @@ silent_chk_compile.h = :
16624 include/generated/compile.h: FORCE
16625 @$($(quiet)chk_compile.h)
16626 $(Q)$(CONFIG_SHELL) $(srctree)/scripts/mkcompile_h $@ \
16627 - "$(UTS_MACHINE)" "$(CONFIG_SMP)" "$(CONFIG_PREEMPT)" "$(CC) $(KBUILD_CFLAGS)"
16628 + "$(UTS_MACHINE)" "$(CONFIG_SMP)" "$(CONFIG_PREEMPT)" "$(CONFIG_PREEMPT_RT_FULL)" "$(CC) $(KBUILD_CFLAGS)"
16629 diff --git a/init/main.c b/init/main.c
16630 index 9e64d7097f1a..4a76e629c137 100644
16633 @@ -530,6 +530,7 @@ asmlinkage __visible void __init start_kernel(void)
16634 setup_command_line(command_line);
16635 setup_nr_cpu_ids();
16636 setup_per_cpu_areas();
16637 + softirq_early_init();
16638 smp_prepare_boot_cpu(); /* arch-specific boot-cpu hooks */
16640 build_all_zonelists(NULL, NULL);
16641 diff --git a/ipc/msg.c b/ipc/msg.c
16642 index c6521c205cb4..996d89023552 100644
16645 @@ -183,20 +183,14 @@ static void ss_wakeup(struct list_head *h, int kill)
16649 -static void expunge_all(struct msg_queue *msq, int res)
16650 +static void expunge_all(struct msg_queue *msq, int res,
16651 + struct wake_q_head *wake_q)
16653 struct msg_receiver *msr, *t;
16655 list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
16656 - msr->r_msg = NULL; /* initialize expunge ordering */
16657 - wake_up_process(msr->r_tsk);
16659 - * Ensure that the wakeup is visible before setting r_msg as
16660 - * the receiving end depends on it: either spinning on a nil,
16661 - * or dealing with -EAGAIN cases. See lockless receive part 1
16662 - * and 2 in do_msgrcv().
16664 - smp_wmb(); /* barrier (B) */
16666 + wake_q_add(wake_q, msr->r_tsk);
16667 msr->r_msg = ERR_PTR(res);
16670 @@ -213,11 +207,13 @@ static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
16672 struct msg_msg *msg, *t;
16673 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
16676 - expunge_all(msq, -EIDRM);
16677 + expunge_all(msq, -EIDRM, &wake_q);
16678 ss_wakeup(&msq->q_senders, 1);
16680 ipc_unlock_object(&msq->q_perm);
16681 + wake_up_q(&wake_q);
16684 list_for_each_entry_safe(msg, t, &msq->q_messages, m_list) {
16685 @@ -342,6 +338,7 @@ static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
16686 struct kern_ipc_perm *ipcp;
16687 struct msqid64_ds uninitialized_var(msqid64);
16688 struct msg_queue *msq;
16692 if (cmd == IPC_SET) {
16693 @@ -389,7 +386,7 @@ static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
16694 /* sleeping receivers might be excluded by
16695 * stricter permissions.
16697 - expunge_all(msq, -EAGAIN);
16698 + expunge_all(msq, -EAGAIN, &wake_q);
16699 /* sleeping senders might be able to send
16700 * due to a larger queue size.
16702 @@ -402,6 +399,7 @@ static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
16705 ipc_unlock_object(&msq->q_perm);
16706 + wake_up_q(&wake_q);
16710 @@ -566,7 +564,8 @@ static int testmsg(struct msg_msg *msg, long type, int mode)
16714 -static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg)
16715 +static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg,
16716 + struct wake_q_head *wake_q)
16718 struct msg_receiver *msr, *t;
16720 @@ -577,27 +576,13 @@ static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg)
16722 list_del(&msr->r_list);
16723 if (msr->r_maxsize < msg->m_ts) {
16724 - /* initialize pipelined send ordering */
16725 - msr->r_msg = NULL;
16726 - wake_up_process(msr->r_tsk);
16727 - /* barrier (B) see barrier comment below */
16729 + wake_q_add(wake_q, msr->r_tsk);
16730 msr->r_msg = ERR_PTR(-E2BIG);
16732 - msr->r_msg = NULL;
16733 msq->q_lrpid = task_pid_vnr(msr->r_tsk);
16734 msq->q_rtime = get_seconds();
16735 - wake_up_process(msr->r_tsk);
16737 - * Ensure that the wakeup is visible before
16738 - * setting r_msg, as the receiving can otherwise
16739 - * exit - once r_msg is set, the receiver can
16740 - * continue. See lockless receive part 1 and 2
16741 - * in do_msgrcv(). Barrier (B).
16744 + wake_q_add(wake_q, msr->r_tsk);
16750 @@ -613,6 +598,7 @@ long do_msgsnd(int msqid, long mtype, void __user *mtext,
16751 struct msg_msg *msg;
16753 struct ipc_namespace *ns;
16756 ns = current->nsproxy->ipc_ns;
16758 @@ -698,7 +684,7 @@ long do_msgsnd(int msqid, long mtype, void __user *mtext,
16759 msq->q_lspid = task_tgid_vnr(current);
16760 msq->q_stime = get_seconds();
16762 - if (!pipelined_send(msq, msg)) {
16763 + if (!pipelined_send(msq, msg, &wake_q)) {
16764 /* no one is waiting for this message, enqueue it */
16765 list_add_tail(&msg->m_list, &msq->q_messages);
16766 msq->q_cbytes += msgsz;
16767 @@ -712,6 +698,7 @@ long do_msgsnd(int msqid, long mtype, void __user *mtext,
16770 ipc_unlock_object(&msq->q_perm);
16771 + wake_up_q(&wake_q);
16775 @@ -932,57 +919,25 @@ long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp, int msgfl
16778 /* Lockless receive, part 2:
16779 - * Wait until pipelined_send or expunge_all are outside of
16780 - * wake_up_process(). There is a race with exit(), see
16781 - * ipc/mqueue.c for the details. The correct serialization
16782 - * ensures that a receiver cannot continue without the wakeup
16783 - * being visibible _before_ setting r_msg:
16784 + * The work in pipelined_send() and expunge_all():
16785 + * - Set pointer to message
16786 + * - Queue the receiver task for later wakeup
16787 + * - Wake up the process after the lock is dropped.
16790 - * <loop receiver>
16791 - * smp_rmb(); (A) <-- pair -. <waker thread>
16792 - * <load ->r_msg> | msr->r_msg = NULL;
16793 - * | wake_up_process();
16794 - * <continue> `------> smp_wmb(); (B)
16795 - * msr->r_msg = msg;
16797 - * Where (A) orders the message value read and where (B) orders
16798 - * the write to the r_msg -- done in both pipelined_send and
16800 + * Should the process wake up before this wakeup (due to a
16801 + * signal) it will either see the message and continue …
16805 - * Pairs with writer barrier in pipelined_send
16806 - * or expunge_all.
16808 - smp_rmb(); /* barrier (A) */
16809 - msg = (struct msg_msg *)msr_d.r_msg;
16814 - * The cpu_relax() call is a compiler barrier
16815 - * which forces everything in this loop to be
16821 - /* Lockless receive, part 3:
16822 - * If there is a message or an error then accept it without
16825 + msg = (struct msg_msg *)msr_d.r_msg;
16826 if (msg != ERR_PTR(-EAGAIN))
16829 - /* Lockless receive, part 3:
16830 - * Acquire the queue spinlock.
16833 + * … or see -EAGAIN, acquire the lock to check the message
16836 ipc_lock_object(&msq->q_perm);
16838 - /* Lockless receive, part 4:
16839 - * Repeat test after acquiring the spinlock.
16841 msg = (struct msg_msg *)msr_d.r_msg;
16842 if (msg != ERR_PTR(-EAGAIN))
16844 diff --git a/ipc/sem.c b/ipc/sem.c
16845 index 9862c3d1c26d..ef34d7376697 100644
16848 @@ -708,6 +708,13 @@ undo:
16849 static void wake_up_sem_queue_prepare(struct list_head *pt,
16850 struct sem_queue *q, int error)
16852 +#ifdef CONFIG_PREEMPT_RT_BASE
16853 + struct task_struct *p = q->sleeper;
16854 + get_task_struct(p);
16855 + q->status = error;
16856 + wake_up_process(p);
16857 + put_task_struct(p);
16859 if (list_empty(pt)) {
16861 * Hold preempt off so that we don't get preempted and have the
16862 @@ -719,6 +726,7 @@ static void wake_up_sem_queue_prepare(struct list_head *pt,
16865 list_add_tail(&q->list, pt);
16870 @@ -732,6 +740,7 @@ static void wake_up_sem_queue_prepare(struct list_head *pt,
16872 static void wake_up_sem_queue_do(struct list_head *pt)
16874 +#ifndef CONFIG_PREEMPT_RT_BASE
16875 struct sem_queue *q, *t;
16878 @@ -744,6 +753,7 @@ static void wake_up_sem_queue_do(struct list_head *pt)
16885 static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
16886 diff --git a/kernel/Kconfig.locks b/kernel/Kconfig.locks
16887 index ebdb0043203a..b9e6aa7e5aa6 100644
16888 --- a/kernel/Kconfig.locks
16889 +++ b/kernel/Kconfig.locks
16890 @@ -225,11 +225,11 @@ config ARCH_SUPPORTS_ATOMIC_RMW
16892 config MUTEX_SPIN_ON_OWNER
16894 - depends on SMP && !DEBUG_MUTEXES && ARCH_SUPPORTS_ATOMIC_RMW
16895 + depends on SMP && !DEBUG_MUTEXES && ARCH_SUPPORTS_ATOMIC_RMW && !PREEMPT_RT_FULL
16897 config RWSEM_SPIN_ON_OWNER
16899 - depends on SMP && RWSEM_XCHGADD_ALGORITHM && ARCH_SUPPORTS_ATOMIC_RMW
16900 + depends on SMP && RWSEM_XCHGADD_ALGORITHM && ARCH_SUPPORTS_ATOMIC_RMW && !PREEMPT_RT_FULL
16902 config LOCK_SPIN_ON_OWNER
16904 diff --git a/kernel/Kconfig.preempt b/kernel/Kconfig.preempt
16905 index 3f9c97419f02..11dbe26a8279 100644
16906 --- a/kernel/Kconfig.preempt
16907 +++ b/kernel/Kconfig.preempt
16911 + select PREEMPT_COUNT
16913 +config PREEMPT_RT_BASE
16917 +config HAVE_PREEMPT_LAZY
16920 +config PREEMPT_LAZY
16921 + def_bool y if HAVE_PREEMPT_LAZY && PREEMPT_RT_FULL
16924 prompt "Preemption Model"
16925 @@ -33,9 +46,9 @@ config PREEMPT_VOLUNTARY
16927 Select this if you are building a kernel for a desktop system.
16930 +config PREEMPT__LL
16931 bool "Preemptible Kernel (Low-Latency Desktop)"
16932 - select PREEMPT_COUNT
16934 select UNINLINE_SPIN_UNLOCK if !ARCH_INLINE_SPIN_UNLOCK
16936 This option reduces the latency of the kernel by making
16937 @@ -52,6 +65,22 @@ config PREEMPT
16938 embedded system with latency requirements in the milliseconds
16941 +config PREEMPT_RTB
16942 + bool "Preemptible Kernel (Basic RT)"
16943 + select PREEMPT_RT_BASE
16945 + This option is basically the same as (Low-Latency Desktop) but
16946 + enables changes which are preliminary for the full preemptible
16949 +config PREEMPT_RT_FULL
16950 + bool "Fully Preemptible Kernel (RT)"
16951 + depends on IRQ_FORCED_THREADING
16952 + select PREEMPT_RT_BASE
16953 + select PREEMPT_RCU
16955 + All and everything
16959 config PREEMPT_COUNT
16960 diff --git a/kernel/cgroup.c b/kernel/cgroup.c
16961 index 4cb94b678e9f..8c41ee8a6fee 100644
16962 --- a/kernel/cgroup.c
16963 +++ b/kernel/cgroup.c
16964 @@ -4741,10 +4741,10 @@ static void css_free_rcu_fn(struct rcu_head *rcu_head)
16965 queue_work(cgroup_destroy_wq, &css->destroy_work);
16968 -static void css_release_work_fn(struct work_struct *work)
16969 +static void css_release_work_fn(struct swork_event *sev)
16971 struct cgroup_subsys_state *css =
16972 - container_of(work, struct cgroup_subsys_state, destroy_work);
16973 + container_of(sev, struct cgroup_subsys_state, destroy_swork);
16974 struct cgroup_subsys *ss = css->ss;
16975 struct cgroup *cgrp = css->cgroup;
16977 @@ -4783,8 +4783,8 @@ static void css_release(struct percpu_ref *ref)
16978 struct cgroup_subsys_state *css =
16979 container_of(ref, struct cgroup_subsys_state, refcnt);
16981 - INIT_WORK(&css->destroy_work, css_release_work_fn);
16982 - queue_work(cgroup_destroy_wq, &css->destroy_work);
16983 + INIT_SWORK(&css->destroy_swork, css_release_work_fn);
16984 + swork_queue(&css->destroy_swork);
16987 static void init_and_link_css(struct cgroup_subsys_state *css,
16988 @@ -5401,6 +5401,7 @@ static int __init cgroup_wq_init(void)
16990 cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1);
16991 BUG_ON(!cgroup_destroy_wq);
16992 + BUG_ON(swork_get());
16995 * Used to destroy pidlists and separate to serve as flush domain.
16996 diff --git a/kernel/cpu.c b/kernel/cpu.c
16997 index 40d20bf5de28..0be18c1684d8 100644
17000 @@ -75,8 +75,8 @@ static struct {
17003 .active_writer = NULL,
17004 - .wq = __WAIT_QUEUE_HEAD_INITIALIZER(cpu_hotplug.wq),
17005 .lock = __MUTEX_INITIALIZER(cpu_hotplug.lock),
17006 + .wq = __WAIT_QUEUE_HEAD_INITIALIZER(cpu_hotplug.wq),
17007 #ifdef CONFIG_DEBUG_LOCK_ALLOC
17008 .dep_map = {.name = "cpu_hotplug.lock" },
17010 @@ -89,6 +89,289 @@ static struct {
17011 #define cpuhp_lock_acquire() lock_map_acquire(&cpu_hotplug.dep_map)
17012 #define cpuhp_lock_release() lock_map_release(&cpu_hotplug.dep_map)
17015 + * hotplug_pcp - per cpu hotplug descriptor
17016 + * @unplug: set when pin_current_cpu() needs to sync tasks
17017 + * @sync_tsk: the task that waits for tasks to finish pinned sections
17018 + * @refcount: counter of tasks in pinned sections
17019 + * @grab_lock: set when the tasks entering pinned sections should wait
17020 + * @synced: notifier for @sync_tsk to tell cpu_down it's finished
17021 + * @mutex: the mutex to make tasks wait (used when @grab_lock is true)
17022 + * @mutex_init: zero if the mutex hasn't been initialized yet.
17024 + * Although @unplug and @sync_tsk may point to the same task, the @unplug
17025 + * is used as a flag and still exists after @sync_tsk has exited and
17026 + * @sync_tsk set to NULL.
17028 +struct hotplug_pcp {
17029 + struct task_struct *unplug;
17030 + struct task_struct *sync_tsk;
17033 + struct completion synced;
17034 + struct completion unplug_wait;
17035 +#ifdef CONFIG_PREEMPT_RT_FULL
17037 + * Note, on PREEMPT_RT, the hotplug lock must save the state of
17038 + * the task, otherwise the mutex will cause the task to fail
17039 + * to sleep when required. (Because it's called from migrate_disable())
17041 + * The spinlock_t on PREEMPT_RT is a mutex that saves the task's
17046 + struct mutex mutex;
17051 +#ifdef CONFIG_PREEMPT_RT_FULL
17052 +# define hotplug_lock(hp) rt_spin_lock__no_mg(&(hp)->lock)
17053 +# define hotplug_unlock(hp) rt_spin_unlock__no_mg(&(hp)->lock)
17055 +# define hotplug_lock(hp) mutex_lock(&(hp)->mutex)
17056 +# define hotplug_unlock(hp) mutex_unlock(&(hp)->mutex)
17059 +static DEFINE_PER_CPU(struct hotplug_pcp, hotplug_pcp);
17062 + * pin_current_cpu - Prevent the current cpu from being unplugged
17064 + * Lightweight version of get_online_cpus() to prevent cpu from being
17065 + * unplugged when code runs in a migration disabled region.
17067 + * Must be called with preemption disabled (preempt_count = 1)!
17069 +void pin_current_cpu(void)
17071 + struct hotplug_pcp *hp;
17075 + hp = this_cpu_ptr(&hotplug_pcp);
17077 + if (!hp->unplug || hp->refcount || force || preempt_count() > 1 ||
17078 + hp->unplug == current) {
17082 + if (hp->grab_lock) {
17083 + preempt_enable();
17084 + hotplug_lock(hp);
17085 + hotplug_unlock(hp);
17087 + preempt_enable();
17089 + * Try to push this task off of this CPU.
17091 + if (!migrate_me()) {
17092 + preempt_disable();
17093 + hp = this_cpu_ptr(&hotplug_pcp);
17094 + if (!hp->grab_lock) {
17096 + * Just let it continue it's already pinned
17097 + * or about to sleep.
17102 + preempt_enable();
17105 + preempt_disable();
17110 + * unpin_current_cpu - Allow unplug of current cpu
17112 + * Must be called with preemption or interrupts disabled!
17114 +void unpin_current_cpu(void)
17116 + struct hotplug_pcp *hp = this_cpu_ptr(&hotplug_pcp);
17118 + WARN_ON(hp->refcount <= 0);
17120 + /* This is safe. sync_unplug_thread is pinned to this cpu */
17121 + if (!--hp->refcount && hp->unplug && hp->unplug != current)
17122 + wake_up_process(hp->unplug);
17125 +static void wait_for_pinned_cpus(struct hotplug_pcp *hp)
17127 + set_current_state(TASK_UNINTERRUPTIBLE);
17128 + while (hp->refcount) {
17129 + schedule_preempt_disabled();
17130 + set_current_state(TASK_UNINTERRUPTIBLE);
17134 +static int sync_unplug_thread(void *data)
17136 + struct hotplug_pcp *hp = data;
17138 + wait_for_completion(&hp->unplug_wait);
17139 + preempt_disable();
17140 + hp->unplug = current;
17141 + wait_for_pinned_cpus(hp);
17144 + * This thread will synchronize the cpu_down() with threads
17145 + * that have pinned the CPU. When the pinned CPU count reaches
17146 + * zero, we inform the cpu_down code to continue to the next step.
17148 + set_current_state(TASK_UNINTERRUPTIBLE);
17149 + preempt_enable();
17150 + complete(&hp->synced);
17153 + * If all succeeds, the next step will need tasks to wait till
17154 + * the CPU is offline before continuing. To do this, the grab_lock
17155 + * is set and tasks going into pin_current_cpu() will block on the
17156 + * mutex. But we still need to wait for those that are already in
17157 + * pinned CPU sections. If the cpu_down() failed, the kthread_should_stop()
17158 + * will kick this thread out.
17160 + while (!hp->grab_lock && !kthread_should_stop()) {
17162 + set_current_state(TASK_UNINTERRUPTIBLE);
17165 + /* Make sure grab_lock is seen before we see a stale completion */
17169 + * Now just before cpu_down() enters stop machine, we need to make
17170 + * sure all tasks that are in pinned CPU sections are out, and new
17171 + * tasks will now grab the lock, keeping them from entering pinned
17174 + if (!kthread_should_stop()) {
17175 + preempt_disable();
17176 + wait_for_pinned_cpus(hp);
17177 + preempt_enable();
17178 + complete(&hp->synced);
17181 + set_current_state(TASK_UNINTERRUPTIBLE);
17182 + while (!kthread_should_stop()) {
17184 + set_current_state(TASK_UNINTERRUPTIBLE);
17186 + set_current_state(TASK_RUNNING);
17189 + * Force this thread off this CPU as it's going down and
17190 + * we don't want any more work on this CPU.
17192 + current->flags &= ~PF_NO_SETAFFINITY;
17193 + set_cpus_allowed_ptr(current, cpu_present_mask);
17198 +static void __cpu_unplug_sync(struct hotplug_pcp *hp)
17200 + wake_up_process(hp->sync_tsk);
17201 + wait_for_completion(&hp->synced);
17204 +static void __cpu_unplug_wait(unsigned int cpu)
17206 + struct hotplug_pcp *hp = &per_cpu(hotplug_pcp, cpu);
17208 + complete(&hp->unplug_wait);
17209 + wait_for_completion(&hp->synced);
17213 + * Start the sync_unplug_thread on the target cpu and wait for it to
17216 +static int cpu_unplug_begin(unsigned int cpu)
17218 + struct hotplug_pcp *hp = &per_cpu(hotplug_pcp, cpu);
17221 + /* Protected by cpu_hotplug.lock */
17222 + if (!hp->mutex_init) {
17223 +#ifdef CONFIG_PREEMPT_RT_FULL
17224 + spin_lock_init(&hp->lock);
17226 + mutex_init(&hp->mutex);
17228 + hp->mutex_init = 1;
17231 + /* Inform the scheduler to migrate tasks off this CPU */
17232 + tell_sched_cpu_down_begin(cpu);
17234 + init_completion(&hp->synced);
17235 + init_completion(&hp->unplug_wait);
17237 + hp->sync_tsk = kthread_create(sync_unplug_thread, hp, "sync_unplug/%d", cpu);
17238 + if (IS_ERR(hp->sync_tsk)) {
17239 + err = PTR_ERR(hp->sync_tsk);
17240 + hp->sync_tsk = NULL;
17243 + kthread_bind(hp->sync_tsk, cpu);
17246 + * Wait for tasks to get out of the pinned sections,
17247 + * it's still OK if new tasks enter. Some CPU notifiers will
17248 + * wait for tasks that are going to enter these sections and
17249 + * we must not have them block.
17251 + wake_up_process(hp->sync_tsk);
17255 +static void cpu_unplug_sync(unsigned int cpu)
17257 + struct hotplug_pcp *hp = &per_cpu(hotplug_pcp, cpu);
17259 + init_completion(&hp->synced);
17260 + /* The completion needs to be initialzied before setting grab_lock */
17263 + /* Grab the mutex before setting grab_lock */
17264 + hotplug_lock(hp);
17265 + hp->grab_lock = 1;
17268 + * The CPU notifiers have been completed.
17269 + * Wait for tasks to get out of pinned CPU sections and have new
17270 + * tasks block until the CPU is completely down.
17272 + __cpu_unplug_sync(hp);
17274 + /* All done with the sync thread */
17275 + kthread_stop(hp->sync_tsk);
17276 + hp->sync_tsk = NULL;
17279 +static void cpu_unplug_done(unsigned int cpu)
17281 + struct hotplug_pcp *hp = &per_cpu(hotplug_pcp, cpu);
17283 + hp->unplug = NULL;
17284 + /* Let all tasks know cpu unplug is finished before cleaning up */
17287 + if (hp->sync_tsk)
17288 + kthread_stop(hp->sync_tsk);
17290 + if (hp->grab_lock) {
17291 + hotplug_unlock(hp);
17292 + /* protected by cpu_hotplug.lock */
17293 + hp->grab_lock = 0;
17295 + tell_sched_cpu_down_done(cpu);
17298 void get_online_cpus(void)
17300 @@ -338,13 +621,15 @@ static int take_cpu_down(void *_param)
17301 /* Requires cpu_add_remove_lock to be held */
17302 static int _cpu_down(unsigned int cpu, int tasks_frozen)
17304 - int err, nr_calls = 0;
17305 + int mycpu, err, nr_calls = 0;
17306 void *hcpu = (void *)(long)cpu;
17307 unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0;
17308 struct take_cpu_down_param tcd_param = {
17312 + cpumask_var_t cpumask;
17313 + cpumask_var_t cpumask_org;
17315 if (num_online_cpus() == 1)
17317 @@ -352,7 +637,34 @@ static int _cpu_down(unsigned int cpu, int tasks_frozen)
17318 if (!cpu_online(cpu))
17321 + /* Move the downtaker off the unplug cpu */
17322 + if (!alloc_cpumask_var(&cpumask, GFP_KERNEL))
17324 + if (!alloc_cpumask_var(&cpumask_org, GFP_KERNEL)) {
17325 + free_cpumask_var(cpumask);
17329 + cpumask_copy(cpumask_org, tsk_cpus_allowed(current));
17330 + cpumask_andnot(cpumask, cpu_online_mask, cpumask_of(cpu));
17331 + set_cpus_allowed_ptr(current, cpumask);
17332 + free_cpumask_var(cpumask);
17333 + migrate_disable();
17334 + mycpu = smp_processor_id();
17335 + if (mycpu == cpu) {
17336 + printk(KERN_ERR "Yuck! Still on unplug CPU\n!");
17337 + migrate_enable();
17339 + goto restore_cpus;
17341 + migrate_enable();
17343 cpu_hotplug_begin();
17344 + err = cpu_unplug_begin(cpu);
17346 + printk("cpu_unplug_begin(%d) failed\n", cpu);
17350 err = __cpu_notify(CPU_DOWN_PREPARE | mod, hcpu, -1, &nr_calls);
17352 @@ -378,8 +690,12 @@ static int _cpu_down(unsigned int cpu, int tasks_frozen)
17356 + __cpu_unplug_wait(cpu);
17357 smpboot_park_threads(cpu);
17359 + /* Notifiers are done. Don't let any more tasks pin this CPU. */
17360 + cpu_unplug_sync(cpu);
17363 * Prevent irq alloc/free while the dying cpu reorganizes the
17364 * interrupt affinities.
17365 @@ -424,9 +740,14 @@ static int _cpu_down(unsigned int cpu, int tasks_frozen)
17366 check_for_tasks(cpu);
17369 + cpu_unplug_done(cpu);
17371 cpu_hotplug_done();
17373 cpu_notify_nofail(CPU_POST_DEAD | mod, hcpu);
17375 + set_cpus_allowed_ptr(current, cpumask_org);
17376 + free_cpumask_var(cpumask_org);
17380 diff --git a/kernel/cpuset.c b/kernel/cpuset.c
17381 index b271353d5202..dd7b87b7f618 100644
17382 --- a/kernel/cpuset.c
17383 +++ b/kernel/cpuset.c
17384 @@ -283,7 +283,7 @@ static struct cpuset top_cpuset = {
17387 static DEFINE_MUTEX(cpuset_mutex);
17388 -static DEFINE_SPINLOCK(callback_lock);
17389 +static DEFINE_RAW_SPINLOCK(callback_lock);
17391 static struct workqueue_struct *cpuset_migrate_mm_wq;
17393 @@ -906,9 +906,9 @@ static void update_cpumasks_hier(struct cpuset *cs, struct cpumask *new_cpus)
17397 - spin_lock_irq(&callback_lock);
17398 + raw_spin_lock_irq(&callback_lock);
17399 cpumask_copy(cp->effective_cpus, new_cpus);
17400 - spin_unlock_irq(&callback_lock);
17401 + raw_spin_unlock_irq(&callback_lock);
17403 WARN_ON(!cgroup_subsys_on_dfl(cpuset_cgrp_subsys) &&
17404 !cpumask_equal(cp->cpus_allowed, cp->effective_cpus));
17405 @@ -973,9 +973,9 @@ static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs,
17409 - spin_lock_irq(&callback_lock);
17410 + raw_spin_lock_irq(&callback_lock);
17411 cpumask_copy(cs->cpus_allowed, trialcs->cpus_allowed);
17412 - spin_unlock_irq(&callback_lock);
17413 + raw_spin_unlock_irq(&callback_lock);
17415 /* use trialcs->cpus_allowed as a temp variable */
17416 update_cpumasks_hier(cs, trialcs->cpus_allowed);
17417 @@ -1184,9 +1184,9 @@ static void update_nodemasks_hier(struct cpuset *cs, nodemask_t *new_mems)
17421 - spin_lock_irq(&callback_lock);
17422 + raw_spin_lock_irq(&callback_lock);
17423 cp->effective_mems = *new_mems;
17424 - spin_unlock_irq(&callback_lock);
17425 + raw_spin_unlock_irq(&callback_lock);
17427 WARN_ON(!cgroup_subsys_on_dfl(cpuset_cgrp_subsys) &&
17428 !nodes_equal(cp->mems_allowed, cp->effective_mems));
17429 @@ -1254,9 +1254,9 @@ static int update_nodemask(struct cpuset *cs, struct cpuset *trialcs,
17433 - spin_lock_irq(&callback_lock);
17434 + raw_spin_lock_irq(&callback_lock);
17435 cs->mems_allowed = trialcs->mems_allowed;
17436 - spin_unlock_irq(&callback_lock);
17437 + raw_spin_unlock_irq(&callback_lock);
17439 /* use trialcs->mems_allowed as a temp variable */
17440 update_nodemasks_hier(cs, &trialcs->mems_allowed);
17441 @@ -1347,9 +1347,9 @@ static int update_flag(cpuset_flagbits_t bit, struct cpuset *cs,
17442 spread_flag_changed = ((is_spread_slab(cs) != is_spread_slab(trialcs))
17443 || (is_spread_page(cs) != is_spread_page(trialcs)));
17445 - spin_lock_irq(&callback_lock);
17446 + raw_spin_lock_irq(&callback_lock);
17447 cs->flags = trialcs->flags;
17448 - spin_unlock_irq(&callback_lock);
17449 + raw_spin_unlock_irq(&callback_lock);
17451 if (!cpumask_empty(trialcs->cpus_allowed) && balance_flag_changed)
17452 rebuild_sched_domains_locked();
17453 @@ -1761,7 +1761,7 @@ static int cpuset_common_seq_show(struct seq_file *sf, void *v)
17454 cpuset_filetype_t type = seq_cft(sf)->private;
17457 - spin_lock_irq(&callback_lock);
17458 + raw_spin_lock_irq(&callback_lock);
17462 @@ -1780,7 +1780,7 @@ static int cpuset_common_seq_show(struct seq_file *sf, void *v)
17466 - spin_unlock_irq(&callback_lock);
17467 + raw_spin_unlock_irq(&callback_lock);
17471 @@ -1994,12 +1994,12 @@ static int cpuset_css_online(struct cgroup_subsys_state *css)
17475 - spin_lock_irq(&callback_lock);
17476 + raw_spin_lock_irq(&callback_lock);
17477 if (cgroup_subsys_on_dfl(cpuset_cgrp_subsys)) {
17478 cpumask_copy(cs->effective_cpus, parent->effective_cpus);
17479 cs->effective_mems = parent->effective_mems;
17481 - spin_unlock_irq(&callback_lock);
17482 + raw_spin_unlock_irq(&callback_lock);
17484 if (!test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags))
17486 @@ -2026,12 +2026,12 @@ static int cpuset_css_online(struct cgroup_subsys_state *css)
17490 - spin_lock_irq(&callback_lock);
17491 + raw_spin_lock_irq(&callback_lock);
17492 cs->mems_allowed = parent->mems_allowed;
17493 cs->effective_mems = parent->mems_allowed;
17494 cpumask_copy(cs->cpus_allowed, parent->cpus_allowed);
17495 cpumask_copy(cs->effective_cpus, parent->cpus_allowed);
17496 - spin_unlock_irq(&callback_lock);
17497 + raw_spin_unlock_irq(&callback_lock);
17499 mutex_unlock(&cpuset_mutex);
17501 @@ -2070,7 +2070,7 @@ static void cpuset_css_free(struct cgroup_subsys_state *css)
17502 static void cpuset_bind(struct cgroup_subsys_state *root_css)
17504 mutex_lock(&cpuset_mutex);
17505 - spin_lock_irq(&callback_lock);
17506 + raw_spin_lock_irq(&callback_lock);
17508 if (cgroup_subsys_on_dfl(cpuset_cgrp_subsys)) {
17509 cpumask_copy(top_cpuset.cpus_allowed, cpu_possible_mask);
17510 @@ -2081,7 +2081,7 @@ static void cpuset_bind(struct cgroup_subsys_state *root_css)
17511 top_cpuset.mems_allowed = top_cpuset.effective_mems;
17514 - spin_unlock_irq(&callback_lock);
17515 + raw_spin_unlock_irq(&callback_lock);
17516 mutex_unlock(&cpuset_mutex);
17519 @@ -2182,12 +2182,12 @@ hotplug_update_tasks_legacy(struct cpuset *cs,
17523 - spin_lock_irq(&callback_lock);
17524 + raw_spin_lock_irq(&callback_lock);
17525 cpumask_copy(cs->cpus_allowed, new_cpus);
17526 cpumask_copy(cs->effective_cpus, new_cpus);
17527 cs->mems_allowed = *new_mems;
17528 cs->effective_mems = *new_mems;
17529 - spin_unlock_irq(&callback_lock);
17530 + raw_spin_unlock_irq(&callback_lock);
17533 * Don't call update_tasks_cpumask() if the cpuset becomes empty,
17534 @@ -2224,10 +2224,10 @@ hotplug_update_tasks(struct cpuset *cs,
17535 if (nodes_empty(*new_mems))
17536 *new_mems = parent_cs(cs)->effective_mems;
17538 - spin_lock_irq(&callback_lock);
17539 + raw_spin_lock_irq(&callback_lock);
17540 cpumask_copy(cs->effective_cpus, new_cpus);
17541 cs->effective_mems = *new_mems;
17542 - spin_unlock_irq(&callback_lock);
17543 + raw_spin_unlock_irq(&callback_lock);
17546 update_tasks_cpumask(cs);
17547 @@ -2313,21 +2313,21 @@ static void cpuset_hotplug_workfn(struct work_struct *work)
17549 /* synchronize cpus_allowed to cpu_active_mask */
17550 if (cpus_updated) {
17551 - spin_lock_irq(&callback_lock);
17552 + raw_spin_lock_irq(&callback_lock);
17554 cpumask_copy(top_cpuset.cpus_allowed, &new_cpus);
17555 cpumask_copy(top_cpuset.effective_cpus, &new_cpus);
17556 - spin_unlock_irq(&callback_lock);
17557 + raw_spin_unlock_irq(&callback_lock);
17558 /* we don't mess with cpumasks of tasks in top_cpuset */
17561 /* synchronize mems_allowed to N_MEMORY */
17562 if (mems_updated) {
17563 - spin_lock_irq(&callback_lock);
17564 + raw_spin_lock_irq(&callback_lock);
17566 top_cpuset.mems_allowed = new_mems;
17567 top_cpuset.effective_mems = new_mems;
17568 - spin_unlock_irq(&callback_lock);
17569 + raw_spin_unlock_irq(&callback_lock);
17570 update_tasks_nodemask(&top_cpuset);
17573 @@ -2425,11 +2425,11 @@ void cpuset_cpus_allowed(struct task_struct *tsk, struct cpumask *pmask)
17575 unsigned long flags;
17577 - spin_lock_irqsave(&callback_lock, flags);
17578 + raw_spin_lock_irqsave(&callback_lock, flags);
17580 guarantee_online_cpus(task_cs(tsk), pmask);
17582 - spin_unlock_irqrestore(&callback_lock, flags);
17583 + raw_spin_unlock_irqrestore(&callback_lock, flags);
17586 void cpuset_cpus_allowed_fallback(struct task_struct *tsk)
17587 @@ -2477,11 +2477,11 @@ nodemask_t cpuset_mems_allowed(struct task_struct *tsk)
17589 unsigned long flags;
17591 - spin_lock_irqsave(&callback_lock, flags);
17592 + raw_spin_lock_irqsave(&callback_lock, flags);
17594 guarantee_online_mems(task_cs(tsk), &mask);
17596 - spin_unlock_irqrestore(&callback_lock, flags);
17597 + raw_spin_unlock_irqrestore(&callback_lock, flags);
17601 @@ -2573,14 +2573,14 @@ int __cpuset_node_allowed(int node, gfp_t gfp_mask)
17604 /* Not hardwall and node outside mems_allowed: scan up cpusets */
17605 - spin_lock_irqsave(&callback_lock, flags);
17606 + raw_spin_lock_irqsave(&callback_lock, flags);
17609 cs = nearest_hardwall_ancestor(task_cs(current));
17610 allowed = node_isset(node, cs->mems_allowed);
17613 - spin_unlock_irqrestore(&callback_lock, flags);
17614 + raw_spin_unlock_irqrestore(&callback_lock, flags);
17618 diff --git a/kernel/debug/kdb/kdb_io.c b/kernel/debug/kdb/kdb_io.c
17619 index fc1ef736253c..83c666537a7a 100644
17620 --- a/kernel/debug/kdb/kdb_io.c
17621 +++ b/kernel/debug/kdb/kdb_io.c
17622 @@ -554,7 +554,6 @@ int vkdb_printf(enum kdb_msgsrc src, const char *fmt, va_list ap)
17625 int logging, saved_loglevel = 0;
17626 - int saved_trap_printk;
17627 int got_printf_lock = 0;
17630 @@ -565,8 +564,6 @@ int vkdb_printf(enum kdb_msgsrc src, const char *fmt, va_list ap)
17631 unsigned long uninitialized_var(flags);
17634 - saved_trap_printk = kdb_trap_printk;
17635 - kdb_trap_printk = 0;
17637 /* Serialize kdb_printf if multiple cpus try to write at once.
17638 * But if any cpu goes recursive in kdb, just print the output,
17639 @@ -855,7 +852,6 @@ kdb_print_out:
17641 __release(kdb_printf_lock);
17643 - kdb_trap_printk = saved_trap_printk;
17647 @@ -865,9 +861,11 @@ int kdb_printf(const char *fmt, ...)
17651 + kdb_trap_printk++;
17653 r = vkdb_printf(KDB_MSGSRC_INTERNAL, fmt, ap);
17655 + kdb_trap_printk--;
17659 diff --git a/kernel/events/core.c b/kernel/events/core.c
17660 index 784ab8fe8714..03166fb2364e 100644
17661 --- a/kernel/events/core.c
17662 +++ b/kernel/events/core.c
17663 @@ -802,6 +802,7 @@ static void __perf_mux_hrtimer_init(struct perf_cpu_context *cpuctx, int cpu)
17664 raw_spin_lock_init(&cpuctx->hrtimer_lock);
17665 hrtimer_init(timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_PINNED);
17666 timer->function = perf_mux_hrtimer_handler;
17667 + timer->irqsafe = 1;
17670 static int perf_mux_hrtimer_restart(struct perf_cpu_context *cpuctx)
17671 @@ -7240,6 +7241,7 @@ static void perf_swevent_init_hrtimer(struct perf_event *event)
17673 hrtimer_init(&hwc->hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
17674 hwc->hrtimer.function = perf_swevent_hrtimer;
17675 + hwc->hrtimer.irqsafe = 1;
17678 * Since hrtimers have a fixed rate, we can do a static freq->period
17679 diff --git a/kernel/exit.c b/kernel/exit.c
17680 index ffba5df4abd5..e199407f8831 100644
17681 --- a/kernel/exit.c
17682 +++ b/kernel/exit.c
17683 @@ -144,7 +144,7 @@ static void __exit_signal(struct task_struct *tsk)
17684 * Do this under ->siglock, we can race with another thread
17685 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
17687 - flush_sigqueue(&tsk->pending);
17688 + flush_task_sigqueue(tsk);
17689 tsk->sighand = NULL;
17690 spin_unlock(&sighand->siglock);
17692 diff --git a/kernel/fork.c b/kernel/fork.c
17693 index 0ee630f3ad4b..0fe2c188e94d 100644
17694 --- a/kernel/fork.c
17695 +++ b/kernel/fork.c
17696 @@ -108,7 +108,7 @@ int max_threads; /* tunable limit on nr_threads */
17698 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
17700 -__cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
17701 +DEFINE_RWLOCK(tasklist_lock); /* outer */
17703 #ifdef CONFIG_PROVE_RCU
17704 int lockdep_tasklist_lock_is_held(void)
17705 @@ -244,7 +244,9 @@ static inline void put_signal_struct(struct signal_struct *sig)
17706 if (atomic_dec_and_test(&sig->sigcnt))
17707 free_signal_struct(sig);
17710 +#ifdef CONFIG_PREEMPT_RT_BASE
17713 void __put_task_struct(struct task_struct *tsk)
17715 WARN_ON(!tsk->exit_state);
17716 @@ -261,7 +263,18 @@ void __put_task_struct(struct task_struct *tsk)
17717 if (!profile_handoff_task(tsk))
17720 +#ifndef CONFIG_PREEMPT_RT_BASE
17721 EXPORT_SYMBOL_GPL(__put_task_struct);
17723 +void __put_task_struct_cb(struct rcu_head *rhp)
17725 + struct task_struct *tsk = container_of(rhp, struct task_struct, put_rcu);
17727 + __put_task_struct(tsk);
17730 +EXPORT_SYMBOL_GPL(__put_task_struct_cb);
17733 void __init __weak arch_task_cache_init(void) { }
17735 @@ -693,6 +706,19 @@ void __mmdrop(struct mm_struct *mm)
17737 EXPORT_SYMBOL_GPL(__mmdrop);
17739 +#ifdef CONFIG_PREEMPT_RT_BASE
17741 + * RCU callback for delayed mm drop. Not strictly rcu, but we don't
17742 + * want another facility to make this work.
17744 +void __mmdrop_delayed(struct rcu_head *rhp)
17746 + struct mm_struct *mm = container_of(rhp, struct mm_struct, delayed_drop);
17753 * Decrement the use count and release all resources for an mm.
17755 @@ -1243,6 +1269,9 @@ static void rt_mutex_init_task(struct task_struct *p)
17757 static void posix_cpu_timers_init(struct task_struct *tsk)
17759 +#ifdef CONFIG_PREEMPT_RT_BASE
17760 + tsk->posix_timer_list = NULL;
17762 tsk->cputime_expires.prof_exp = 0;
17763 tsk->cputime_expires.virt_exp = 0;
17764 tsk->cputime_expires.sched_exp = 0;
17765 @@ -1369,15 +1398,16 @@ static struct task_struct *copy_process(unsigned long clone_flags,
17766 spin_lock_init(&p->alloc_lock);
17768 init_sigpending(&p->pending);
17769 + p->sigqueue_cache = NULL;
17771 p->utime = p->stime = p->gtime = 0;
17772 p->utimescaled = p->stimescaled = 0;
17773 prev_cputime_init(&p->prev_cputime);
17775 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
17776 - seqlock_init(&p->vtime_seqlock);
17777 + seqcount_init(&p->vtime_seqcount);
17779 - p->vtime_snap_whence = VTIME_SLEEPING;
17780 + p->vtime_snap_whence = VTIME_INACTIVE;
17783 #if defined(SPLIT_RSS_COUNTING)
17784 diff --git a/kernel/futex.c b/kernel/futex.c
17785 index 3057dabf726f..9604b73ada47 100644
17786 --- a/kernel/futex.c
17787 +++ b/kernel/futex.c
17788 @@ -815,7 +815,9 @@ void exit_pi_state_list(struct task_struct *curr)
17789 * task still owns the PI-state:
17791 if (head->next != next) {
17792 + raw_spin_unlock_irq(&curr->pi_lock);
17793 spin_unlock(&hb->lock);
17794 + raw_spin_lock_irq(&curr->pi_lock);
17798 @@ -1210,6 +1212,7 @@ static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this,
17799 struct futex_pi_state *pi_state = this->pi_state;
17800 u32 uninitialized_var(curval), newval;
17802 + WAKE_Q(wake_sleeper_q);
17806 @@ -1223,7 +1226,7 @@ static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this,
17807 if (pi_state->owner != current)
17810 - raw_spin_lock(&pi_state->pi_mutex.wait_lock);
17811 + raw_spin_lock_irq(&pi_state->pi_mutex.wait_lock);
17812 new_owner = rt_mutex_next_owner(&pi_state->pi_mutex);
17815 @@ -1259,24 +1262,25 @@ static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this,
17819 - raw_spin_unlock(&pi_state->pi_mutex.wait_lock);
17820 + raw_spin_unlock_irq(&pi_state->pi_mutex.wait_lock);
17824 - raw_spin_lock_irq(&pi_state->owner->pi_lock);
17825 + raw_spin_lock(&pi_state->owner->pi_lock);
17826 WARN_ON(list_empty(&pi_state->list));
17827 list_del_init(&pi_state->list);
17828 - raw_spin_unlock_irq(&pi_state->owner->pi_lock);
17829 + raw_spin_unlock(&pi_state->owner->pi_lock);
17831 - raw_spin_lock_irq(&new_owner->pi_lock);
17832 + raw_spin_lock(&new_owner->pi_lock);
17833 WARN_ON(!list_empty(&pi_state->list));
17834 list_add(&pi_state->list, &new_owner->pi_state_list);
17835 pi_state->owner = new_owner;
17836 - raw_spin_unlock_irq(&new_owner->pi_lock);
17837 + raw_spin_unlock(&new_owner->pi_lock);
17839 - raw_spin_unlock(&pi_state->pi_mutex.wait_lock);
17840 + raw_spin_unlock_irq(&pi_state->pi_mutex.wait_lock);
17842 - deboost = rt_mutex_futex_unlock(&pi_state->pi_mutex, &wake_q);
17843 + deboost = rt_mutex_futex_unlock(&pi_state->pi_mutex, &wake_q,
17844 + &wake_sleeper_q);
17847 * First unlock HB so the waiter does not spin on it once he got woken
17848 @@ -1284,8 +1288,9 @@ static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this,
17849 * deboost first (and lose our higher priority), then the task might get
17850 * scheduled away before the wake up can take place.
17852 - spin_unlock(&hb->lock);
17853 + deboost |= spin_unlock_no_deboost(&hb->lock);
17854 wake_up_q(&wake_q);
17855 + wake_up_q_sleeper(&wake_sleeper_q);
17857 rt_mutex_adjust_prio(current);
17859 @@ -1822,6 +1827,16 @@ retry_private:
17860 requeue_pi_wake_futex(this, &key2, hb2);
17863 + } else if (ret == -EAGAIN) {
17865 + * Waiter was woken by timeout or
17866 + * signal and has set pi_blocked_on to
17867 + * PI_WAKEUP_INPROGRESS before we
17868 + * tried to enqueue it on the rtmutex.
17870 + this->pi_state = NULL;
17871 + free_pi_state(pi_state);
17875 this->pi_state = NULL;
17876 @@ -2139,11 +2154,11 @@ static int fixup_owner(u32 __user *uaddr, struct futex_q *q, int locked)
17877 * we returned due to timeout or signal without taking the
17878 * rt_mutex. Too late.
17880 - raw_spin_lock(&q->pi_state->pi_mutex.wait_lock);
17881 + raw_spin_lock_irq(&q->pi_state->pi_mutex.wait_lock);
17882 owner = rt_mutex_owner(&q->pi_state->pi_mutex);
17884 owner = rt_mutex_next_owner(&q->pi_state->pi_mutex);
17885 - raw_spin_unlock(&q->pi_state->pi_mutex.wait_lock);
17886 + raw_spin_unlock_irq(&q->pi_state->pi_mutex.wait_lock);
17887 ret = fixup_pi_state_owner(uaddr, q, owner);
17890 @@ -2690,7 +2705,7 @@ static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags,
17892 struct hrtimer_sleeper timeout, *to = NULL;
17893 struct rt_mutex_waiter rt_waiter;
17894 - struct futex_hash_bucket *hb;
17895 + struct futex_hash_bucket *hb, *hb2;
17896 union futex_key key2 = FUTEX_KEY_INIT;
17897 struct futex_q q = futex_q_init;
17899 @@ -2715,10 +2730,7 @@ static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags,
17900 * The waiter is allocated on our stack, manipulated by the requeue
17901 * code while we sleep on uaddr.
17903 - debug_rt_mutex_init_waiter(&rt_waiter);
17904 - RB_CLEAR_NODE(&rt_waiter.pi_tree_entry);
17905 - RB_CLEAR_NODE(&rt_waiter.tree_entry);
17906 - rt_waiter.task = NULL;
17907 + rt_mutex_init_waiter(&rt_waiter, false);
17909 ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, VERIFY_WRITE);
17910 if (unlikely(ret != 0))
17911 @@ -2749,20 +2761,55 @@ static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags,
17912 /* Queue the futex_q, drop the hb lock, wait for wakeup. */
17913 futex_wait_queue_me(hb, &q, to);
17915 - spin_lock(&hb->lock);
17916 - ret = handle_early_requeue_pi_wakeup(hb, &q, &key2, to);
17917 - spin_unlock(&hb->lock);
17919 - goto out_put_keys;
17921 + * On RT we must avoid races with requeue and trying to block
17922 + * on two mutexes (hb->lock and uaddr2's rtmutex) by
17923 + * serializing access to pi_blocked_on with pi_lock.
17925 + raw_spin_lock_irq(¤t->pi_lock);
17926 + if (current->pi_blocked_on) {
17928 + * We have been requeued or are in the process of
17929 + * being requeued.
17931 + raw_spin_unlock_irq(¤t->pi_lock);
17934 + * Setting pi_blocked_on to PI_WAKEUP_INPROGRESS
17935 + * prevents a concurrent requeue from moving us to the
17936 + * uaddr2 rtmutex. After that we can safely acquire
17937 + * (and possibly block on) hb->lock.
17939 + current->pi_blocked_on = PI_WAKEUP_INPROGRESS;
17940 + raw_spin_unlock_irq(¤t->pi_lock);
17942 + spin_lock(&hb->lock);
17945 + * Clean up pi_blocked_on. We might leak it otherwise
17946 + * when we succeeded with the hb->lock in the fast
17949 + raw_spin_lock_irq(¤t->pi_lock);
17950 + current->pi_blocked_on = NULL;
17951 + raw_spin_unlock_irq(¤t->pi_lock);
17953 + ret = handle_early_requeue_pi_wakeup(hb, &q, &key2, to);
17954 + spin_unlock(&hb->lock);
17956 + goto out_put_keys;
17960 - * In order for us to be here, we know our q.key == key2, and since
17961 - * we took the hb->lock above, we also know that futex_requeue() has
17962 - * completed and we no longer have to concern ourselves with a wakeup
17963 - * race with the atomic proxy lock acquisition by the requeue code. The
17964 - * futex_requeue dropped our key1 reference and incremented our key2
17965 - * reference count.
17966 + * In order to be here, we have either been requeued, are in
17967 + * the process of being requeued, or requeue successfully
17968 + * acquired uaddr2 on our behalf. If pi_blocked_on was
17969 + * non-null above, we may be racing with a requeue. Do not
17970 + * rely on q->lock_ptr to be hb2->lock until after blocking on
17971 + * hb->lock or hb2->lock. The futex_requeue dropped our key1
17972 + * reference and incremented our key2 reference count.
17974 + hb2 = hash_futex(&key2);
17976 /* Check if the requeue code acquired the second futex for us. */
17977 if (!q.rt_waiter) {
17978 @@ -2771,7 +2818,8 @@ static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags,
17979 * did a lock-steal - fix up the PI-state in that case.
17981 if (q.pi_state && (q.pi_state->owner != current)) {
17982 - spin_lock(q.lock_ptr);
17983 + spin_lock(&hb2->lock);
17984 + BUG_ON(&hb2->lock != q.lock_ptr);
17985 ret = fixup_pi_state_owner(uaddr2, &q, current);
17986 if (ret && rt_mutex_owner(&q.pi_state->pi_mutex) == current)
17987 rt_mutex_unlock(&q.pi_state->pi_mutex);
17988 @@ -2780,7 +2828,7 @@ static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags,
17989 * the requeue_pi() code acquired for us.
17991 free_pi_state(q.pi_state);
17992 - spin_unlock(q.lock_ptr);
17993 + spin_unlock(&hb2->lock);
17996 struct rt_mutex *pi_mutex;
17997 @@ -2795,7 +2843,8 @@ static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags,
17998 ret = rt_mutex_finish_proxy_lock(pi_mutex, to, &rt_waiter);
17999 debug_rt_mutex_free_waiter(&rt_waiter);
18001 - spin_lock(q.lock_ptr);
18002 + spin_lock(&hb2->lock);
18003 + BUG_ON(&hb2->lock != q.lock_ptr);
18005 * Fixup the pi_state owner and possibly acquire the lock if we
18007 diff --git a/kernel/irq/handle.c b/kernel/irq/handle.c
18008 index 57bff7857e87..6c65c9252991 100644
18009 --- a/kernel/irq/handle.c
18010 +++ b/kernel/irq/handle.c
18011 @@ -134,6 +134,8 @@ void __irq_wake_thread(struct irq_desc *desc, struct irqaction *action)
18013 irqreturn_t handle_irq_event_percpu(struct irq_desc *desc)
18015 + struct pt_regs *regs = get_irq_regs();
18016 + u64 ip = regs ? instruction_pointer(regs) : 0;
18017 irqreturn_t retval = IRQ_NONE;
18018 unsigned int flags = 0, irq = desc->irq_data.irq;
18019 struct irqaction *action = desc->action;
18020 @@ -176,7 +178,11 @@ irqreturn_t handle_irq_event_percpu(struct irq_desc *desc)
18021 action = action->next;
18024 - add_interrupt_randomness(irq, flags);
18025 +#ifdef CONFIG_PREEMPT_RT_FULL
18026 + desc->random_ip = ip;
18028 + add_interrupt_randomness(irq, flags, ip);
18032 note_interrupt(desc, retval);
18033 diff --git a/kernel/irq/irqdesc.c b/kernel/irq/irqdesc.c
18034 index 239e2ae2c947..0b73349a42d5 100644
18035 --- a/kernel/irq/irqdesc.c
18036 +++ b/kernel/irq/irqdesc.c
18037 @@ -24,10 +24,27 @@
18038 static struct lock_class_key irq_desc_lock_class;
18040 #if defined(CONFIG_SMP)
18041 +static int __init irq_affinity_setup(char *str)
18043 + zalloc_cpumask_var(&irq_default_affinity, GFP_NOWAIT);
18044 + cpulist_parse(str, irq_default_affinity);
18046 + * Set at least the boot cpu. We don't want to end up with
18047 + * bugreports caused by random comandline masks
18049 + cpumask_set_cpu(smp_processor_id(), irq_default_affinity);
18052 +__setup("irqaffinity=", irq_affinity_setup);
18054 static void __init init_irq_default_affinity(void)
18056 - alloc_cpumask_var(&irq_default_affinity, GFP_NOWAIT);
18057 - cpumask_setall(irq_default_affinity);
18058 +#ifdef CONFIG_CPUMASK_OFFSTACK
18059 + if (!irq_default_affinity)
18060 + zalloc_cpumask_var(&irq_default_affinity, GFP_NOWAIT);
18062 + if (cpumask_empty(irq_default_affinity))
18063 + cpumask_setall(irq_default_affinity);
18066 static void __init init_irq_default_affinity(void)
18067 diff --git a/kernel/irq/manage.c b/kernel/irq/manage.c
18068 index 6ead200370da..8e89554aa345 100644
18069 --- a/kernel/irq/manage.c
18070 +++ b/kernel/irq/manage.c
18072 #include "internals.h"
18074 #ifdef CONFIG_IRQ_FORCED_THREADING
18075 +# ifndef CONFIG_PREEMPT_RT_BASE
18076 __read_mostly bool force_irqthreads;
18078 static int __init setup_forced_irqthreads(char *arg)
18079 @@ -30,6 +31,7 @@ static int __init setup_forced_irqthreads(char *arg)
18082 early_param("threadirqs", setup_forced_irqthreads);
18086 static void __synchronize_hardirq(struct irq_desc *desc)
18087 @@ -181,6 +183,62 @@ static inline void
18088 irq_get_pending(struct cpumask *mask, struct irq_desc *desc) { }
18091 +#ifdef CONFIG_PREEMPT_RT_FULL
18092 +static void _irq_affinity_notify(struct irq_affinity_notify *notify);
18093 +static struct task_struct *set_affinity_helper;
18094 +static LIST_HEAD(affinity_list);
18095 +static DEFINE_RAW_SPINLOCK(affinity_list_lock);
18097 +static int set_affinity_thread(void *unused)
18100 + struct irq_affinity_notify *notify;
18103 + set_current_state(TASK_INTERRUPTIBLE);
18105 + raw_spin_lock_irq(&affinity_list_lock);
18106 + empty = list_empty(&affinity_list);
18107 + raw_spin_unlock_irq(&affinity_list_lock);
18111 + if (kthread_should_stop())
18113 + set_current_state(TASK_RUNNING);
18117 + raw_spin_lock_irq(&affinity_list_lock);
18118 + if (!list_empty(&affinity_list)) {
18119 + notify = list_first_entry(&affinity_list,
18120 + struct irq_affinity_notify, list);
18121 + list_del_init(¬ify->list);
18123 + raw_spin_unlock_irq(&affinity_list_lock);
18127 + _irq_affinity_notify(notify);
18133 +static void init_helper_thread(void)
18135 + if (set_affinity_helper)
18137 + set_affinity_helper = kthread_run(set_affinity_thread, NULL,
18139 + WARN_ON(IS_ERR(set_affinity_helper));
18143 +static inline void init_helper_thread(void) { }
18147 int irq_do_set_affinity(struct irq_data *data, const struct cpumask *mask,
18150 @@ -220,7 +278,17 @@ int irq_set_affinity_locked(struct irq_data *data, const struct cpumask *mask,
18152 if (desc->affinity_notify) {
18153 kref_get(&desc->affinity_notify->kref);
18155 +#ifdef CONFIG_PREEMPT_RT_FULL
18156 + raw_spin_lock(&affinity_list_lock);
18157 + if (list_empty(&desc->affinity_notify->list))
18158 + list_add_tail(&affinity_list,
18159 + &desc->affinity_notify->list);
18160 + raw_spin_unlock(&affinity_list_lock);
18161 + wake_up_process(set_affinity_helper);
18163 schedule_work(&desc->affinity_notify->work);
18166 irqd_set(data, IRQD_AFFINITY_SET);
18168 @@ -258,10 +326,8 @@ int irq_set_affinity_hint(unsigned int irq, const struct cpumask *m)
18170 EXPORT_SYMBOL_GPL(irq_set_affinity_hint);
18172 -static void irq_affinity_notify(struct work_struct *work)
18173 +static void _irq_affinity_notify(struct irq_affinity_notify *notify)
18175 - struct irq_affinity_notify *notify =
18176 - container_of(work, struct irq_affinity_notify, work);
18177 struct irq_desc *desc = irq_to_desc(notify->irq);
18178 cpumask_var_t cpumask;
18179 unsigned long flags;
18180 @@ -283,6 +349,13 @@ out:
18181 kref_put(¬ify->kref, notify->release);
18184 +static void irq_affinity_notify(struct work_struct *work)
18186 + struct irq_affinity_notify *notify =
18187 + container_of(work, struct irq_affinity_notify, work);
18188 + _irq_affinity_notify(notify);
18192 * irq_set_affinity_notifier - control notification of IRQ affinity changes
18193 * @irq: Interrupt for which to enable/disable notification
18194 @@ -312,6 +385,8 @@ irq_set_affinity_notifier(unsigned int irq, struct irq_affinity_notify *notify)
18196 kref_init(¬ify->kref);
18197 INIT_WORK(¬ify->work, irq_affinity_notify);
18198 + INIT_LIST_HEAD(¬ify->list);
18199 + init_helper_thread();
18202 raw_spin_lock_irqsave(&desc->lock, flags);
18203 @@ -865,7 +940,15 @@ irq_forced_thread_fn(struct irq_desc *desc, struct irqaction *action)
18204 local_bh_disable();
18205 ret = action->thread_fn(action->irq, action->dev_id);
18206 irq_finalize_oneshot(desc, action);
18207 - local_bh_enable();
18209 + * Interrupts which have real time requirements can be set up
18210 + * to avoid softirq processing in the thread handler. This is
18211 + * safe as these interrupts do not raise soft interrupts.
18213 + if (irq_settings_no_softirq_call(desc))
18214 + _local_bh_enable();
18216 + local_bh_enable();
18220 @@ -962,6 +1045,12 @@ static int irq_thread(void *data)
18221 if (action_ret == IRQ_WAKE_THREAD)
18222 irq_wake_secondary(desc, action);
18224 +#ifdef CONFIG_PREEMPT_RT_FULL
18225 + migrate_disable();
18226 + add_interrupt_randomness(action->irq, 0,
18227 + desc->random_ip ^ (unsigned long) action);
18228 + migrate_enable();
18230 wake_threads_waitq(desc);
18233 @@ -1315,6 +1404,9 @@ __setup_irq(unsigned int irq, struct irq_desc *desc, struct irqaction *new)
18234 irqd_set(&desc->irq_data, IRQD_NO_BALANCING);
18237 + if (new->flags & IRQF_NO_SOFTIRQ_CALL)
18238 + irq_settings_set_no_softirq_call(desc);
18240 /* Set default affinity mask once everything is setup */
18241 setup_affinity(desc, mask);
18243 @@ -1968,7 +2060,7 @@ EXPORT_SYMBOL_GPL(irq_get_irqchip_state);
18244 * This call sets the internal irqchip state of an interrupt,
18245 * depending on the value of @which.
18247 - * This function should be called with preemption disabled if the
18248 + * This function should be called with migration disabled if the
18249 * interrupt controller has per-cpu registers.
18251 int irq_set_irqchip_state(unsigned int irq, enum irqchip_irq_state which,
18252 diff --git a/kernel/irq/settings.h b/kernel/irq/settings.h
18253 index 320579d89091..2df2d4445b1e 100644
18254 --- a/kernel/irq/settings.h
18255 +++ b/kernel/irq/settings.h
18256 @@ -16,6 +16,7 @@ enum {
18257 _IRQ_PER_CPU_DEVID = IRQ_PER_CPU_DEVID,
18258 _IRQ_IS_POLLED = IRQ_IS_POLLED,
18259 _IRQ_DISABLE_UNLAZY = IRQ_DISABLE_UNLAZY,
18260 + _IRQ_NO_SOFTIRQ_CALL = IRQ_NO_SOFTIRQ_CALL,
18261 _IRQF_MODIFY_MASK = IRQF_MODIFY_MASK,
18264 @@ -30,6 +31,7 @@ enum {
18265 #define IRQ_PER_CPU_DEVID GOT_YOU_MORON
18266 #define IRQ_IS_POLLED GOT_YOU_MORON
18267 #define IRQ_DISABLE_UNLAZY GOT_YOU_MORON
18268 +#define IRQ_NO_SOFTIRQ_CALL GOT_YOU_MORON
18269 #undef IRQF_MODIFY_MASK
18270 #define IRQF_MODIFY_MASK GOT_YOU_MORON
18272 @@ -40,6 +42,16 @@ irq_settings_clr_and_set(struct irq_desc *desc, u32 clr, u32 set)
18273 desc->status_use_accessors |= (set & _IRQF_MODIFY_MASK);
18276 +static inline bool irq_settings_no_softirq_call(struct irq_desc *desc)
18278 + return desc->status_use_accessors & _IRQ_NO_SOFTIRQ_CALL;
18281 +static inline void irq_settings_set_no_softirq_call(struct irq_desc *desc)
18283 + desc->status_use_accessors |= _IRQ_NO_SOFTIRQ_CALL;
18286 static inline bool irq_settings_is_per_cpu(struct irq_desc *desc)
18288 return desc->status_use_accessors & _IRQ_PER_CPU;
18289 diff --git a/kernel/irq/spurious.c b/kernel/irq/spurious.c
18290 index 32144175458d..ed26f2554972 100644
18291 --- a/kernel/irq/spurious.c
18292 +++ b/kernel/irq/spurious.c
18293 @@ -444,6 +444,10 @@ MODULE_PARM_DESC(noirqdebug, "Disable irq lockup detection when true");
18295 static int __init irqfixup_setup(char *str)
18297 +#ifdef CONFIG_PREEMPT_RT_BASE
18298 + pr_warn("irqfixup boot option not supported w/ CONFIG_PREEMPT_RT_BASE\n");
18302 printk(KERN_WARNING "Misrouted IRQ fixup support enabled.\n");
18303 printk(KERN_WARNING "This may impact system performance.\n");
18304 @@ -456,6 +460,10 @@ module_param(irqfixup, int, 0644);
18306 static int __init irqpoll_setup(char *str)
18308 +#ifdef CONFIG_PREEMPT_RT_BASE
18309 + pr_warn("irqpoll boot option not supported w/ CONFIG_PREEMPT_RT_BASE\n");
18313 printk(KERN_WARNING "Misrouted IRQ fixup and polling support "
18315 diff --git a/kernel/irq_work.c b/kernel/irq_work.c
18316 index bcf107ce0854..2899ba0d23d1 100644
18317 --- a/kernel/irq_work.c
18318 +++ b/kernel/irq_work.c
18320 #include <linux/cpu.h>
18321 #include <linux/notifier.h>
18322 #include <linux/smp.h>
18323 +#include <linux/interrupt.h>
18324 #include <asm/processor.h>
18327 @@ -65,6 +66,8 @@ void __weak arch_irq_work_raise(void)
18329 bool irq_work_queue_on(struct irq_work *work, int cpu)
18331 + struct llist_head *list;
18333 /* All work should have been flushed before going offline */
18334 WARN_ON_ONCE(cpu_is_offline(cpu));
18336 @@ -75,7 +78,12 @@ bool irq_work_queue_on(struct irq_work *work, int cpu)
18337 if (!irq_work_claim(work))
18340 - if (llist_add(&work->llnode, &per_cpu(raised_list, cpu)))
18341 + if (IS_ENABLED(CONFIG_PREEMPT_RT_FULL) && !(work->flags & IRQ_WORK_HARD_IRQ))
18342 + list = &per_cpu(lazy_list, cpu);
18344 + list = &per_cpu(raised_list, cpu);
18346 + if (llist_add(&work->llnode, list))
18347 arch_send_call_function_single_ipi(cpu);
18350 @@ -86,6 +94,9 @@ EXPORT_SYMBOL_GPL(irq_work_queue_on);
18351 /* Enqueue the irq work @work on the current CPU */
18352 bool irq_work_queue(struct irq_work *work)
18354 + struct llist_head *list;
18355 + bool lazy_work, realtime = IS_ENABLED(CONFIG_PREEMPT_RT_FULL);
18357 /* Only queue if not already pending */
18358 if (!irq_work_claim(work))
18360 @@ -93,13 +104,15 @@ bool irq_work_queue(struct irq_work *work)
18361 /* Queue the entry and raise the IPI if needed. */
18364 - /* If the work is "lazy", handle it from next tick if any */
18365 - if (work->flags & IRQ_WORK_LAZY) {
18366 - if (llist_add(&work->llnode, this_cpu_ptr(&lazy_list)) &&
18367 - tick_nohz_tick_stopped())
18368 - arch_irq_work_raise();
18370 - if (llist_add(&work->llnode, this_cpu_ptr(&raised_list)))
18371 + lazy_work = work->flags & IRQ_WORK_LAZY;
18373 + if (lazy_work || (realtime && !(work->flags & IRQ_WORK_HARD_IRQ)))
18374 + list = this_cpu_ptr(&lazy_list);
18376 + list = this_cpu_ptr(&raised_list);
18378 + if (llist_add(&work->llnode, list)) {
18379 + if (!lazy_work || tick_nohz_tick_stopped())
18380 arch_irq_work_raise();
18383 @@ -116,9 +129,8 @@ bool irq_work_needs_cpu(void)
18384 raised = this_cpu_ptr(&raised_list);
18385 lazy = this_cpu_ptr(&lazy_list);
18387 - if (llist_empty(raised) || arch_irq_work_has_interrupt())
18388 - if (llist_empty(lazy))
18390 + if (llist_empty(raised) && llist_empty(lazy))
18393 /* All work should have been flushed before going offline */
18394 WARN_ON_ONCE(cpu_is_offline(smp_processor_id()));
18395 @@ -132,7 +144,7 @@ static void irq_work_run_list(struct llist_head *list)
18396 struct irq_work *work;
18397 struct llist_node *llnode;
18399 - BUG_ON(!irqs_disabled());
18400 + BUG_ON_NONRT(!irqs_disabled());
18402 if (llist_empty(list))
18404 @@ -169,7 +181,16 @@ static void irq_work_run_list(struct llist_head *list)
18405 void irq_work_run(void)
18407 irq_work_run_list(this_cpu_ptr(&raised_list));
18408 - irq_work_run_list(this_cpu_ptr(&lazy_list));
18409 + if (IS_ENABLED(CONFIG_PREEMPT_RT_FULL)) {
18411 + * NOTE: we raise softirq via IPI for safety,
18412 + * and execute in irq_work_tick() to move the
18413 + * overhead from hard to soft irq context.
18415 + if (!llist_empty(this_cpu_ptr(&lazy_list)))
18416 + raise_softirq(TIMER_SOFTIRQ);
18418 + irq_work_run_list(this_cpu_ptr(&lazy_list));
18420 EXPORT_SYMBOL_GPL(irq_work_run);
18422 @@ -179,8 +200,17 @@ void irq_work_tick(void)
18424 if (!llist_empty(raised) && !arch_irq_work_has_interrupt())
18425 irq_work_run_list(raised);
18427 + if (!IS_ENABLED(CONFIG_PREEMPT_RT_FULL))
18428 + irq_work_run_list(this_cpu_ptr(&lazy_list));
18431 +#if defined(CONFIG_IRQ_WORK) && defined(CONFIG_PREEMPT_RT_FULL)
18432 +void irq_work_tick_soft(void)
18434 irq_work_run_list(this_cpu_ptr(&lazy_list));
18439 * Synchronize against the irq_work @entry, ensures the entry is not
18440 diff --git a/kernel/ksysfs.c b/kernel/ksysfs.c
18441 index e83b26464061..c0e08d1cf33e 100644
18442 --- a/kernel/ksysfs.c
18443 +++ b/kernel/ksysfs.c
18444 @@ -136,6 +136,15 @@ KERNEL_ATTR_RO(vmcoreinfo);
18446 #endif /* CONFIG_KEXEC_CORE */
18448 +#if defined(CONFIG_PREEMPT_RT_FULL)
18449 +static ssize_t realtime_show(struct kobject *kobj,
18450 + struct kobj_attribute *attr, char *buf)
18452 + return sprintf(buf, "%d\n", 1);
18454 +KERNEL_ATTR_RO(realtime);
18457 /* whether file capabilities are enabled */
18458 static ssize_t fscaps_show(struct kobject *kobj,
18459 struct kobj_attribute *attr, char *buf)
18460 @@ -203,6 +212,9 @@ static struct attribute * kernel_attrs[] = {
18461 &vmcoreinfo_attr.attr,
18463 &rcu_expedited_attr.attr,
18464 +#ifdef CONFIG_PREEMPT_RT_FULL
18465 + &realtime_attr.attr,
18470 diff --git a/kernel/locking/Makefile b/kernel/locking/Makefile
18471 index 8e96f6cc2a4a..447b03082d88 100644
18472 --- a/kernel/locking/Makefile
18473 +++ b/kernel/locking/Makefile
18476 -obj-y += mutex.o semaphore.o rwsem.o percpu-rwsem.o
18477 +obj-y += semaphore.o percpu-rwsem.o
18479 ifdef CONFIG_FUNCTION_TRACER
18480 CFLAGS_REMOVE_lockdep.o = $(CC_FLAGS_FTRACE)
18481 @@ -8,7 +8,11 @@ CFLAGS_REMOVE_mutex-debug.o = $(CC_FLAGS_FTRACE)
18482 CFLAGS_REMOVE_rtmutex-debug.o = $(CC_FLAGS_FTRACE)
18485 +ifneq ($(CONFIG_PREEMPT_RT_FULL),y)
18487 obj-$(CONFIG_DEBUG_MUTEXES) += mutex-debug.o
18490 obj-$(CONFIG_LOCKDEP) += lockdep.o
18491 ifeq ($(CONFIG_PROC_FS),y)
18492 obj-$(CONFIG_LOCKDEP) += lockdep_proc.o
18493 @@ -22,7 +26,10 @@ obj-$(CONFIG_RT_MUTEXES) += rtmutex.o
18494 obj-$(CONFIG_DEBUG_RT_MUTEXES) += rtmutex-debug.o
18495 obj-$(CONFIG_DEBUG_SPINLOCK) += spinlock.o
18496 obj-$(CONFIG_DEBUG_SPINLOCK) += spinlock_debug.o
18497 +ifneq ($(CONFIG_PREEMPT_RT_FULL),y)
18498 obj-$(CONFIG_RWSEM_GENERIC_SPINLOCK) += rwsem-spinlock.o
18499 obj-$(CONFIG_RWSEM_XCHGADD_ALGORITHM) += rwsem-xadd.o
18501 +obj-$(CONFIG_PREEMPT_RT_FULL) += rt.o
18502 obj-$(CONFIG_QUEUED_RWLOCKS) += qrwlock.o
18503 obj-$(CONFIG_LOCK_TORTURE_TEST) += locktorture.o
18504 diff --git a/kernel/locking/lglock.c b/kernel/locking/lglock.c
18505 index 951cfcd10b4a..57e0ea72c28a 100644
18506 --- a/kernel/locking/lglock.c
18507 +++ b/kernel/locking/lglock.c
18509 #include <linux/cpu.h>
18510 #include <linux/string.h>
18512 +#ifndef CONFIG_PREEMPT_RT_FULL
18513 +# define lg_lock_ptr arch_spinlock_t
18514 +# define lg_do_lock(l) arch_spin_lock(l)
18515 +# define lg_do_unlock(l) arch_spin_unlock(l)
18517 +# define lg_lock_ptr struct rt_mutex
18518 +# define lg_do_lock(l) __rt_spin_lock__no_mg(l)
18519 +# define lg_do_unlock(l) __rt_spin_unlock(l)
18522 * Note there is no uninit, so lglocks cannot be defined in
18523 * modules (but it's fine to use them from there)
18524 @@ -12,51 +21,60 @@
18526 void lg_lock_init(struct lglock *lg, char *name)
18528 +#ifdef CONFIG_PREEMPT_RT_FULL
18531 + for_each_possible_cpu(i) {
18532 + struct rt_mutex *lock = per_cpu_ptr(lg->lock, i);
18534 + rt_mutex_init(lock);
18537 LOCKDEP_INIT_MAP(&lg->lock_dep_map, name, &lg->lock_key, 0);
18539 EXPORT_SYMBOL(lg_lock_init);
18541 void lg_local_lock(struct lglock *lg)
18543 - arch_spinlock_t *lock;
18544 + lg_lock_ptr *lock;
18546 - preempt_disable();
18547 + migrate_disable();
18548 lock_acquire_shared(&lg->lock_dep_map, 0, 0, NULL, _RET_IP_);
18549 lock = this_cpu_ptr(lg->lock);
18550 - arch_spin_lock(lock);
18551 + lg_do_lock(lock);
18553 EXPORT_SYMBOL(lg_local_lock);
18555 void lg_local_unlock(struct lglock *lg)
18557 - arch_spinlock_t *lock;
18558 + lg_lock_ptr *lock;
18560 lock_release(&lg->lock_dep_map, 1, _RET_IP_);
18561 lock = this_cpu_ptr(lg->lock);
18562 - arch_spin_unlock(lock);
18563 - preempt_enable();
18564 + lg_do_unlock(lock);
18565 + migrate_enable();
18567 EXPORT_SYMBOL(lg_local_unlock);
18569 void lg_local_lock_cpu(struct lglock *lg, int cpu)
18571 - arch_spinlock_t *lock;
18572 + lg_lock_ptr *lock;
18574 - preempt_disable();
18575 + preempt_disable_nort();
18576 lock_acquire_shared(&lg->lock_dep_map, 0, 0, NULL, _RET_IP_);
18577 lock = per_cpu_ptr(lg->lock, cpu);
18578 - arch_spin_lock(lock);
18579 + lg_do_lock(lock);
18581 EXPORT_SYMBOL(lg_local_lock_cpu);
18583 void lg_local_unlock_cpu(struct lglock *lg, int cpu)
18585 - arch_spinlock_t *lock;
18586 + lg_lock_ptr *lock;
18588 lock_release(&lg->lock_dep_map, 1, _RET_IP_);
18589 lock = per_cpu_ptr(lg->lock, cpu);
18590 - arch_spin_unlock(lock);
18591 - preempt_enable();
18592 + lg_do_unlock(lock);
18593 + preempt_enable_nort();
18595 EXPORT_SYMBOL(lg_local_unlock_cpu);
18597 @@ -68,30 +86,30 @@ void lg_double_lock(struct lglock *lg, int cpu1, int cpu2)
18601 - preempt_disable();
18602 + preempt_disable_nort();
18603 lock_acquire_shared(&lg->lock_dep_map, 0, 0, NULL, _RET_IP_);
18604 - arch_spin_lock(per_cpu_ptr(lg->lock, cpu1));
18605 - arch_spin_lock(per_cpu_ptr(lg->lock, cpu2));
18606 + lg_do_lock(per_cpu_ptr(lg->lock, cpu1));
18607 + lg_do_lock(per_cpu_ptr(lg->lock, cpu2));
18610 void lg_double_unlock(struct lglock *lg, int cpu1, int cpu2)
18612 lock_release(&lg->lock_dep_map, 1, _RET_IP_);
18613 - arch_spin_unlock(per_cpu_ptr(lg->lock, cpu1));
18614 - arch_spin_unlock(per_cpu_ptr(lg->lock, cpu2));
18615 - preempt_enable();
18616 + lg_do_unlock(per_cpu_ptr(lg->lock, cpu1));
18617 + lg_do_unlock(per_cpu_ptr(lg->lock, cpu2));
18618 + preempt_enable_nort();
18621 void lg_global_lock(struct lglock *lg)
18625 - preempt_disable();
18626 + preempt_disable_nort();
18627 lock_acquire_exclusive(&lg->lock_dep_map, 0, 0, NULL, _RET_IP_);
18628 for_each_possible_cpu(i) {
18629 - arch_spinlock_t *lock;
18630 + lg_lock_ptr *lock;
18631 lock = per_cpu_ptr(lg->lock, i);
18632 - arch_spin_lock(lock);
18633 + lg_do_lock(lock);
18636 EXPORT_SYMBOL(lg_global_lock);
18637 @@ -102,10 +120,35 @@ void lg_global_unlock(struct lglock *lg)
18639 lock_release(&lg->lock_dep_map, 1, _RET_IP_);
18640 for_each_possible_cpu(i) {
18641 - arch_spinlock_t *lock;
18642 + lg_lock_ptr *lock;
18643 lock = per_cpu_ptr(lg->lock, i);
18644 - arch_spin_unlock(lock);
18645 + lg_do_unlock(lock);
18647 - preempt_enable();
18648 + preempt_enable_nort();
18650 EXPORT_SYMBOL(lg_global_unlock);
18652 +#ifdef CONFIG_PREEMPT_RT_FULL
18654 + * HACK: If you use this, you get to keep the pieces.
18655 + * Used in queue_stop_cpus_work() when stop machinery
18656 + * is called from inactive CPU, so we can't schedule.
18658 +# define lg_do_trylock_relax(l) \
18660 + while (!__rt_spin_trylock(l)) \
18664 +void lg_global_trylock_relax(struct lglock *lg)
18668 + lock_acquire_exclusive(&lg->lock_dep_map, 0, 0, NULL, _RET_IP_);
18669 + for_each_possible_cpu(i) {
18670 + lg_lock_ptr *lock;
18671 + lock = per_cpu_ptr(lg->lock, i);
18672 + lg_do_trylock_relax(lock);
18676 diff --git a/kernel/locking/lockdep.c b/kernel/locking/lockdep.c
18677 index 60ace56618f6..fd54dbf686cc 100644
18678 --- a/kernel/locking/lockdep.c
18679 +++ b/kernel/locking/lockdep.c
18680 @@ -668,6 +668,7 @@ look_up_lock_class(struct lockdep_map *lock, unsigned int subclass)
18681 struct lockdep_subclass_key *key;
18682 struct list_head *hash_head;
18683 struct lock_class *class;
18684 + bool is_static = false;
18686 #ifdef CONFIG_DEBUG_LOCKDEP
18688 @@ -695,10 +696,23 @@ look_up_lock_class(struct lockdep_map *lock, unsigned int subclass)
18691 * Static locks do not have their class-keys yet - for them the key
18692 - * is the lock object itself:
18693 + * is the lock object itself. If the lock is in the per cpu area,
18694 + * the canonical address of the lock (per cpu offset removed) is
18697 - if (unlikely(!lock->key))
18698 - lock->key = (void *)lock;
18699 + if (unlikely(!lock->key)) {
18700 + unsigned long can_addr, addr = (unsigned long)lock;
18702 + if (__is_kernel_percpu_address(addr, &can_addr))
18703 + lock->key = (void *)can_addr;
18704 + else if (__is_module_percpu_address(addr, &can_addr))
18705 + lock->key = (void *)can_addr;
18706 + else if (static_obj(lock))
18707 + lock->key = (void *)lock;
18709 + return ERR_PTR(-EINVAL);
18710 + is_static = true;
18714 * NOTE: the class-key must be unique. For dynamic locks, a static
18715 @@ -730,7 +744,7 @@ look_up_lock_class(struct lockdep_map *lock, unsigned int subclass)
18720 + return is_static || static_obj(lock->key) ? NULL : ERR_PTR(-EINVAL);
18724 @@ -748,19 +762,18 @@ register_lock_class(struct lockdep_map *lock, unsigned int subclass, int force)
18725 DEBUG_LOCKS_WARN_ON(!irqs_disabled());
18727 class = look_up_lock_class(lock, subclass);
18728 - if (likely(class))
18729 + if (likely(!IS_ERR_OR_NULL(class)))
18730 goto out_set_class_cache;
18733 * Debug-check: all keys must be persistent!
18735 - if (!static_obj(lock->key)) {
18737 + if (IS_ERR(class)) {
18739 printk("INFO: trying to register non-static key.\n");
18740 printk("the code is fine but needs lockdep annotation.\n");
18741 printk("turning off the locking correctness validator.\n");
18747 @@ -3278,7 +3291,7 @@ static int match_held_lock(struct held_lock *hlock, struct lockdep_map *lock)
18748 * Clearly if the lock hasn't been acquired _ever_, we're not
18749 * holding it either, so report failure.
18752 + if (IS_ERR_OR_NULL(class))
18756 @@ -3525,6 +3538,7 @@ static void check_flags(unsigned long flags)
18760 +#ifndef CONFIG_PREEMPT_RT_FULL
18762 * We dont accurately track softirq state in e.g.
18763 * hardirq contexts (such as on 4KSTACKS), so only
18764 @@ -3539,6 +3553,7 @@ static void check_flags(unsigned long flags)
18765 DEBUG_LOCKS_WARN_ON(!current->softirqs_enabled);
18771 print_irqtrace_events(current);
18772 @@ -3977,7 +3992,7 @@ void lockdep_reset_lock(struct lockdep_map *lock)
18773 * If the class exists we look it up and zap it:
18775 class = look_up_lock_class(lock, j);
18777 + if (!IS_ERR_OR_NULL(class))
18781 diff --git a/kernel/locking/locktorture.c b/kernel/locking/locktorture.c
18782 index 8ef1919d63b2..291fc19e28e0 100644
18783 --- a/kernel/locking/locktorture.c
18784 +++ b/kernel/locking/locktorture.c
18786 #include <linux/kthread.h>
18787 #include <linux/sched/rt.h>
18788 #include <linux/spinlock.h>
18789 -#include <linux/rwlock.h>
18790 #include <linux/mutex.h>
18791 #include <linux/rwsem.h>
18792 #include <linux/smp.h>
18793 diff --git a/kernel/locking/rt.c b/kernel/locking/rt.c
18794 new file mode 100644
18795 index 000000000000..d4ab61c1848b
18797 +++ b/kernel/locking/rt.c
18802 + * Real-Time Preemption Support
18804 + * started by Ingo Molnar:
18806 + * Copyright (C) 2004-2006 Red Hat, Inc., Ingo Molnar <mingo@redhat.com>
18807 + * Copyright (C) 2006, Timesys Corp., Thomas Gleixner <tglx@timesys.com>
18809 + * historic credit for proving that Linux spinlocks can be implemented via
18810 + * RT-aware mutexes goes to many people: The Pmutex project (Dirk Grambow
18811 + * and others) who prototyped it on 2.4 and did lots of comparative
18812 + * research and analysis; TimeSys, for proving that you can implement a
18813 + * fully preemptible kernel via the use of IRQ threading and mutexes;
18814 + * Bill Huey for persuasively arguing on lkml that the mutex model is the
18815 + * right one; and to MontaVista, who ported pmutexes to 2.6.
18817 + * This code is a from-scratch implementation and is not based on pmutexes,
18818 + * but the idea of converting spinlocks to mutexes is used here too.
18820 + * lock debugging, locking tree, deadlock detection:
18822 + * Copyright (C) 2004, LynuxWorks, Inc., Igor Manyilov, Bill Huey
18823 + * Released under the General Public License (GPL).
18825 + * Includes portions of the generic R/W semaphore implementation from:
18827 + * Copyright (c) 2001 David Howells (dhowells@redhat.com).
18828 + * - Derived partially from idea by Andrea Arcangeli <andrea@suse.de>
18829 + * - Derived also from comments by Linus
18831 + * Pending ownership of locks and ownership stealing:
18833 + * Copyright (C) 2005, Kihon Technologies Inc., Steven Rostedt
18835 + * (also by Steven Rostedt)
18836 + * - Converted single pi_lock to individual task locks.
18838 + * By Esben Nielsen:
18839 + * Doing priority inheritance with help of the scheduler.
18841 + * Copyright (C) 2006, Timesys Corp., Thomas Gleixner <tglx@timesys.com>
18842 + * - major rework based on Esben Nielsens initial patch
18843 + * - replaced thread_info references by task_struct refs
18844 + * - removed task->pending_owner dependency
18845 + * - BKL drop/reacquire for semaphore style locks to avoid deadlocks
18846 + * in the scheduler return path as discussed with Steven Rostedt
18848 + * Copyright (C) 2006, Kihon Technologies Inc.
18849 + * Steven Rostedt <rostedt@goodmis.org>
18850 + * - debugged and patched Thomas Gleixner's rework.
18851 + * - added back the cmpxchg to the rework.
18852 + * - turned atomic require back on for SMP.
18855 +#include <linux/spinlock.h>
18856 +#include <linux/rtmutex.h>
18857 +#include <linux/sched.h>
18858 +#include <linux/delay.h>
18859 +#include <linux/module.h>
18860 +#include <linux/kallsyms.h>
18861 +#include <linux/syscalls.h>
18862 +#include <linux/interrupt.h>
18863 +#include <linux/plist.h>
18864 +#include <linux/fs.h>
18865 +#include <linux/futex.h>
18866 +#include <linux/hrtimer.h>
18868 +#include "rtmutex_common.h"
18871 + * struct mutex functions
18873 +void __mutex_do_init(struct mutex *mutex, const char *name,
18874 + struct lock_class_key *key)
18876 +#ifdef CONFIG_DEBUG_LOCK_ALLOC
18878 + * Make sure we are not reinitializing a held lock:
18880 + debug_check_no_locks_freed((void *)mutex, sizeof(*mutex));
18881 + lockdep_init_map(&mutex->dep_map, name, key, 0);
18883 + mutex->lock.save_state = 0;
18885 +EXPORT_SYMBOL(__mutex_do_init);
18887 +void __lockfunc _mutex_lock(struct mutex *lock)
18889 + mutex_acquire(&lock->dep_map, 0, 0, _RET_IP_);
18890 + rt_mutex_lock(&lock->lock);
18892 +EXPORT_SYMBOL(_mutex_lock);
18894 +int __lockfunc _mutex_lock_interruptible(struct mutex *lock)
18898 + mutex_acquire(&lock->dep_map, 0, 0, _RET_IP_);
18899 + ret = rt_mutex_lock_interruptible(&lock->lock);
18901 + mutex_release(&lock->dep_map, 1, _RET_IP_);
18904 +EXPORT_SYMBOL(_mutex_lock_interruptible);
18906 +int __lockfunc _mutex_lock_killable(struct mutex *lock)
18910 + mutex_acquire(&lock->dep_map, 0, 0, _RET_IP_);
18911 + ret = rt_mutex_lock_killable(&lock->lock);
18913 + mutex_release(&lock->dep_map, 1, _RET_IP_);
18916 +EXPORT_SYMBOL(_mutex_lock_killable);
18918 +#ifdef CONFIG_DEBUG_LOCK_ALLOC
18919 +void __lockfunc _mutex_lock_nested(struct mutex *lock, int subclass)
18921 + mutex_acquire_nest(&lock->dep_map, subclass, 0, NULL, _RET_IP_);
18922 + rt_mutex_lock(&lock->lock);
18924 +EXPORT_SYMBOL(_mutex_lock_nested);
18926 +void __lockfunc _mutex_lock_nest_lock(struct mutex *lock, struct lockdep_map *nest)
18928 + mutex_acquire_nest(&lock->dep_map, 0, 0, nest, _RET_IP_);
18929 + rt_mutex_lock(&lock->lock);
18931 +EXPORT_SYMBOL(_mutex_lock_nest_lock);
18933 +int __lockfunc _mutex_lock_interruptible_nested(struct mutex *lock, int subclass)
18937 + mutex_acquire_nest(&lock->dep_map, subclass, 0, NULL, _RET_IP_);
18938 + ret = rt_mutex_lock_interruptible(&lock->lock);
18940 + mutex_release(&lock->dep_map, 1, _RET_IP_);
18943 +EXPORT_SYMBOL(_mutex_lock_interruptible_nested);
18945 +int __lockfunc _mutex_lock_killable_nested(struct mutex *lock, int subclass)
18949 + mutex_acquire(&lock->dep_map, subclass, 0, _RET_IP_);
18950 + ret = rt_mutex_lock_killable(&lock->lock);
18952 + mutex_release(&lock->dep_map, 1, _RET_IP_);
18955 +EXPORT_SYMBOL(_mutex_lock_killable_nested);
18958 +int __lockfunc _mutex_trylock(struct mutex *lock)
18960 + int ret = rt_mutex_trylock(&lock->lock);
18963 + mutex_acquire(&lock->dep_map, 0, 1, _RET_IP_);
18967 +EXPORT_SYMBOL(_mutex_trylock);
18969 +void __lockfunc _mutex_unlock(struct mutex *lock)
18971 + mutex_release(&lock->dep_map, 1, _RET_IP_);
18972 + rt_mutex_unlock(&lock->lock);
18974 +EXPORT_SYMBOL(_mutex_unlock);
18977 + * rwlock_t functions
18979 +int __lockfunc rt_write_trylock(rwlock_t *rwlock)
18983 + migrate_disable();
18984 + ret = rt_mutex_trylock(&rwlock->lock);
18986 + rwlock_acquire(&rwlock->dep_map, 0, 1, _RET_IP_);
18988 + migrate_enable();
18992 +EXPORT_SYMBOL(rt_write_trylock);
18994 +int __lockfunc rt_write_trylock_irqsave(rwlock_t *rwlock, unsigned long *flags)
18999 + ret = rt_write_trylock(rwlock);
19002 +EXPORT_SYMBOL(rt_write_trylock_irqsave);
19004 +int __lockfunc rt_read_trylock(rwlock_t *rwlock)
19006 + struct rt_mutex *lock = &rwlock->lock;
19010 + * recursive read locks succeed when current owns the lock,
19011 + * but not when read_depth == 0 which means that the lock is
19014 + if (rt_mutex_owner(lock) != current) {
19015 + migrate_disable();
19016 + ret = rt_mutex_trylock(lock);
19018 + rwlock_acquire(&rwlock->dep_map, 0, 1, _RET_IP_);
19020 + migrate_enable();
19022 + } else if (!rwlock->read_depth) {
19027 + rwlock->read_depth++;
19031 +EXPORT_SYMBOL(rt_read_trylock);
19033 +void __lockfunc rt_write_lock(rwlock_t *rwlock)
19035 + rwlock_acquire(&rwlock->dep_map, 0, 0, _RET_IP_);
19036 + __rt_spin_lock(&rwlock->lock);
19038 +EXPORT_SYMBOL(rt_write_lock);
19040 +void __lockfunc rt_read_lock(rwlock_t *rwlock)
19042 + struct rt_mutex *lock = &rwlock->lock;
19046 + * recursive read locks succeed when current owns the lock
19048 + if (rt_mutex_owner(lock) != current) {
19049 + rwlock_acquire(&rwlock->dep_map, 0, 0, _RET_IP_);
19050 + __rt_spin_lock(lock);
19052 + rwlock->read_depth++;
19055 +EXPORT_SYMBOL(rt_read_lock);
19057 +void __lockfunc rt_write_unlock(rwlock_t *rwlock)
19059 + /* NOTE: we always pass in '1' for nested, for simplicity */
19060 + rwlock_release(&rwlock->dep_map, 1, _RET_IP_);
19061 + __rt_spin_unlock(&rwlock->lock);
19062 + migrate_enable();
19064 +EXPORT_SYMBOL(rt_write_unlock);
19066 +void __lockfunc rt_read_unlock(rwlock_t *rwlock)
19068 + /* Release the lock only when read_depth is down to 0 */
19069 + if (--rwlock->read_depth == 0) {
19070 + rwlock_release(&rwlock->dep_map, 1, _RET_IP_);
19071 + __rt_spin_unlock(&rwlock->lock);
19072 + migrate_enable();
19075 +EXPORT_SYMBOL(rt_read_unlock);
19077 +unsigned long __lockfunc rt_write_lock_irqsave(rwlock_t *rwlock)
19079 + rt_write_lock(rwlock);
19083 +EXPORT_SYMBOL(rt_write_lock_irqsave);
19085 +unsigned long __lockfunc rt_read_lock_irqsave(rwlock_t *rwlock)
19087 + rt_read_lock(rwlock);
19091 +EXPORT_SYMBOL(rt_read_lock_irqsave);
19093 +void __rt_rwlock_init(rwlock_t *rwlock, char *name, struct lock_class_key *key)
19095 +#ifdef CONFIG_DEBUG_LOCK_ALLOC
19097 + * Make sure we are not reinitializing a held lock:
19099 + debug_check_no_locks_freed((void *)rwlock, sizeof(*rwlock));
19100 + lockdep_init_map(&rwlock->dep_map, name, key, 0);
19102 + rwlock->lock.save_state = 1;
19103 + rwlock->read_depth = 0;
19105 +EXPORT_SYMBOL(__rt_rwlock_init);
19111 +void rt_up_write(struct rw_semaphore *rwsem)
19113 + rwsem_release(&rwsem->dep_map, 1, _RET_IP_);
19114 + rt_mutex_unlock(&rwsem->lock);
19116 +EXPORT_SYMBOL(rt_up_write);
19118 +void __rt_up_read(struct rw_semaphore *rwsem)
19120 + if (--rwsem->read_depth == 0)
19121 + rt_mutex_unlock(&rwsem->lock);
19124 +void rt_up_read(struct rw_semaphore *rwsem)
19126 + rwsem_release(&rwsem->dep_map, 1, _RET_IP_);
19127 + __rt_up_read(rwsem);
19129 +EXPORT_SYMBOL(rt_up_read);
19132 + * downgrade a write lock into a read lock
19133 + * - just wake up any readers at the front of the queue
19135 +void rt_downgrade_write(struct rw_semaphore *rwsem)
19137 + BUG_ON(rt_mutex_owner(&rwsem->lock) != current);
19138 + rwsem->read_depth = 1;
19140 +EXPORT_SYMBOL(rt_downgrade_write);
19142 +int rt_down_write_trylock(struct rw_semaphore *rwsem)
19144 + int ret = rt_mutex_trylock(&rwsem->lock);
19147 + rwsem_acquire(&rwsem->dep_map, 0, 1, _RET_IP_);
19150 +EXPORT_SYMBOL(rt_down_write_trylock);
19152 +void rt_down_write(struct rw_semaphore *rwsem)
19154 + rwsem_acquire(&rwsem->dep_map, 0, 0, _RET_IP_);
19155 + rt_mutex_lock(&rwsem->lock);
19157 +EXPORT_SYMBOL(rt_down_write);
19159 +void rt_down_write_nested(struct rw_semaphore *rwsem, int subclass)
19161 + rwsem_acquire(&rwsem->dep_map, subclass, 0, _RET_IP_);
19162 + rt_mutex_lock(&rwsem->lock);
19164 +EXPORT_SYMBOL(rt_down_write_nested);
19166 +void rt_down_write_nested_lock(struct rw_semaphore *rwsem,
19167 + struct lockdep_map *nest)
19169 + rwsem_acquire_nest(&rwsem->dep_map, 0, 0, nest, _RET_IP_);
19170 + rt_mutex_lock(&rwsem->lock);
19172 +EXPORT_SYMBOL(rt_down_write_nested_lock);
19174 +int rt__down_read_trylock(struct rw_semaphore *rwsem)
19176 + struct rt_mutex *lock = &rwsem->lock;
19180 + * recursive read locks succeed when current owns the rwsem,
19181 + * but not when read_depth == 0 which means that the rwsem is
19184 + if (rt_mutex_owner(lock) != current)
19185 + ret = rt_mutex_trylock(&rwsem->lock);
19186 + else if (!rwsem->read_depth)
19190 + rwsem->read_depth++;
19195 +int rt_down_read_trylock(struct rw_semaphore *rwsem)
19199 + ret = rt__down_read_trylock(rwsem);
19201 + rwsem_acquire(&rwsem->dep_map, 0, 1, _RET_IP_);
19205 +EXPORT_SYMBOL(rt_down_read_trylock);
19207 +void rt__down_read(struct rw_semaphore *rwsem)
19209 + struct rt_mutex *lock = &rwsem->lock;
19211 + if (rt_mutex_owner(lock) != current)
19212 + rt_mutex_lock(&rwsem->lock);
19213 + rwsem->read_depth++;
19215 +EXPORT_SYMBOL(rt__down_read);
19217 +static void __rt_down_read(struct rw_semaphore *rwsem, int subclass)
19219 + rwsem_acquire_read(&rwsem->dep_map, subclass, 0, _RET_IP_);
19220 + rt__down_read(rwsem);
19223 +void rt_down_read(struct rw_semaphore *rwsem)
19225 + __rt_down_read(rwsem, 0);
19227 +EXPORT_SYMBOL(rt_down_read);
19229 +void rt_down_read_nested(struct rw_semaphore *rwsem, int subclass)
19231 + __rt_down_read(rwsem, subclass);
19233 +EXPORT_SYMBOL(rt_down_read_nested);
19235 +void __rt_rwsem_init(struct rw_semaphore *rwsem, const char *name,
19236 + struct lock_class_key *key)
19238 +#ifdef CONFIG_DEBUG_LOCK_ALLOC
19240 + * Make sure we are not reinitializing a held lock:
19242 + debug_check_no_locks_freed((void *)rwsem, sizeof(*rwsem));
19243 + lockdep_init_map(&rwsem->dep_map, name, key, 0);
19245 + rwsem->read_depth = 0;
19246 + rwsem->lock.save_state = 0;
19248 +EXPORT_SYMBOL(__rt_rwsem_init);
19251 + * atomic_dec_and_mutex_lock - return holding mutex if we dec to 0
19252 + * @cnt: the atomic which we are to dec
19253 + * @lock: the mutex to return holding if we dec to 0
19255 + * return true and hold lock if we dec to 0, return false otherwise
19257 +int atomic_dec_and_mutex_lock(atomic_t *cnt, struct mutex *lock)
19259 + /* dec if we can't possibly hit 0 */
19260 + if (atomic_add_unless(cnt, -1, 1))
19262 + /* we might hit 0, so take the lock */
19263 + mutex_lock(lock);
19264 + if (!atomic_dec_and_test(cnt)) {
19265 + /* when we actually did the dec, we didn't hit 0 */
19266 + mutex_unlock(lock);
19269 + /* we hit 0, and we hold the lock */
19272 +EXPORT_SYMBOL(atomic_dec_and_mutex_lock);
19273 diff --git a/kernel/locking/rtmutex.c b/kernel/locking/rtmutex.c
19274 index b066724d7a5b..0e9a6260441d 100644
19275 --- a/kernel/locking/rtmutex.c
19276 +++ b/kernel/locking/rtmutex.c
19278 * Copyright (C) 2005-2006 Timesys Corp., Thomas Gleixner <tglx@timesys.com>
19279 * Copyright (C) 2005 Kihon Technologies Inc., Steven Rostedt
19280 * Copyright (C) 2006 Esben Nielsen
19281 + * Adaptive Spinlocks:
19282 + * Copyright (C) 2008 Novell, Inc., Gregory Haskins, Sven Dietrich,
19283 + * and Peter Morreale,
19284 + * Adaptive Spinlocks simplification:
19285 + * Copyright (C) 2008 Red Hat, Inc., Steven Rostedt <srostedt@redhat.com>
19287 * See Documentation/locking/rt-mutex-design.txt for details.
19290 #include <linux/sched/rt.h>
19291 #include <linux/sched/deadline.h>
19292 #include <linux/timer.h>
19293 +#include <linux/ww_mutex.h>
19295 #include "rtmutex_common.h"
19297 @@ -133,6 +139,12 @@ static void fixup_rt_mutex_waiters(struct rt_mutex *lock)
19298 WRITE_ONCE(*p, owner & ~RT_MUTEX_HAS_WAITERS);
19301 +static int rt_mutex_real_waiter(struct rt_mutex_waiter *waiter)
19303 + return waiter && waiter != PI_WAKEUP_INPROGRESS &&
19304 + waiter != PI_REQUEUE_INPROGRESS;
19308 * We can speed up the acquire/release, if there's no debugging state to be
19310 @@ -163,13 +175,14 @@ static inline void mark_rt_mutex_waiters(struct rt_mutex *lock)
19311 * 2) Drop lock->wait_lock
19312 * 3) Try to unlock the lock with cmpxchg
19314 -static inline bool unlock_rt_mutex_safe(struct rt_mutex *lock)
19315 +static inline bool unlock_rt_mutex_safe(struct rt_mutex *lock,
19316 + unsigned long flags)
19317 __releases(lock->wait_lock)
19319 struct task_struct *owner = rt_mutex_owner(lock);
19321 clear_rt_mutex_waiters(lock);
19322 - raw_spin_unlock(&lock->wait_lock);
19323 + raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
19325 * If a new waiter comes in between the unlock and the cmpxchg
19326 * we have two situations:
19327 @@ -211,11 +224,12 @@ static inline void mark_rt_mutex_waiters(struct rt_mutex *lock)
19329 * Simple slow path only version: lock->owner is protected by lock->wait_lock.
19331 -static inline bool unlock_rt_mutex_safe(struct rt_mutex *lock)
19332 +static inline bool unlock_rt_mutex_safe(struct rt_mutex *lock,
19333 + unsigned long flags)
19334 __releases(lock->wait_lock)
19336 lock->owner = NULL;
19337 - raw_spin_unlock(&lock->wait_lock);
19338 + raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
19342 @@ -412,6 +426,14 @@ static bool rt_mutex_cond_detect_deadlock(struct rt_mutex_waiter *waiter,
19343 return debug_rt_mutex_detect_deadlock(waiter, chwalk);
19346 +static void rt_mutex_wake_waiter(struct rt_mutex_waiter *waiter)
19348 + if (waiter->savestate)
19349 + wake_up_lock_sleeper(waiter->task);
19351 + wake_up_process(waiter->task);
19355 * Max number of times we'll walk the boosting chain:
19357 @@ -419,7 +441,8 @@ int max_lock_depth = 1024;
19359 static inline struct rt_mutex *task_blocked_on_lock(struct task_struct *p)
19361 - return p->pi_blocked_on ? p->pi_blocked_on->lock : NULL;
19362 + return rt_mutex_real_waiter(p->pi_blocked_on) ?
19363 + p->pi_blocked_on->lock : NULL;
19367 @@ -497,7 +520,6 @@ static int rt_mutex_adjust_prio_chain(struct task_struct *task,
19368 int ret = 0, depth = 0;
19369 struct rt_mutex *lock;
19370 bool detect_deadlock;
19371 - unsigned long flags;
19372 bool requeue = true;
19374 detect_deadlock = rt_mutex_cond_detect_deadlock(orig_waiter, chwalk);
19375 @@ -540,7 +562,7 @@ static int rt_mutex_adjust_prio_chain(struct task_struct *task,
19377 * [1] Task cannot go away as we did a get_task() before !
19379 - raw_spin_lock_irqsave(&task->pi_lock, flags);
19380 + raw_spin_lock_irq(&task->pi_lock);
19383 * [2] Get the waiter on which @task is blocked on.
19384 @@ -556,7 +578,7 @@ static int rt_mutex_adjust_prio_chain(struct task_struct *task,
19385 * reached or the state of the chain has changed while we
19386 * dropped the locks.
19389 + if (!rt_mutex_real_waiter(waiter))
19390 goto out_unlock_pi;
19393 @@ -624,7 +646,7 @@ static int rt_mutex_adjust_prio_chain(struct task_struct *task,
19396 if (!raw_spin_trylock(&lock->wait_lock)) {
19397 - raw_spin_unlock_irqrestore(&task->pi_lock, flags);
19398 + raw_spin_unlock_irq(&task->pi_lock);
19402 @@ -655,7 +677,7 @@ static int rt_mutex_adjust_prio_chain(struct task_struct *task,
19404 * No requeue[7] here. Just release @task [8]
19406 - raw_spin_unlock_irqrestore(&task->pi_lock, flags);
19407 + raw_spin_unlock(&task->pi_lock);
19408 put_task_struct(task);
19411 @@ -663,14 +685,14 @@ static int rt_mutex_adjust_prio_chain(struct task_struct *task,
19412 * If there is no owner of the lock, end of chain.
19414 if (!rt_mutex_owner(lock)) {
19415 - raw_spin_unlock(&lock->wait_lock);
19416 + raw_spin_unlock_irq(&lock->wait_lock);
19420 /* [10] Grab the next task, i.e. owner of @lock */
19421 task = rt_mutex_owner(lock);
19422 get_task_struct(task);
19423 - raw_spin_lock_irqsave(&task->pi_lock, flags);
19424 + raw_spin_lock(&task->pi_lock);
19427 * No requeue [11] here. We just do deadlock detection.
19428 @@ -685,8 +707,8 @@ static int rt_mutex_adjust_prio_chain(struct task_struct *task,
19429 top_waiter = rt_mutex_top_waiter(lock);
19431 /* [13] Drop locks */
19432 - raw_spin_unlock_irqrestore(&task->pi_lock, flags);
19433 - raw_spin_unlock(&lock->wait_lock);
19434 + raw_spin_unlock(&task->pi_lock);
19435 + raw_spin_unlock_irq(&lock->wait_lock);
19437 /* If owner is not blocked, end of chain. */
19439 @@ -707,7 +729,7 @@ static int rt_mutex_adjust_prio_chain(struct task_struct *task,
19440 rt_mutex_enqueue(lock, waiter);
19442 /* [8] Release the task */
19443 - raw_spin_unlock_irqrestore(&task->pi_lock, flags);
19444 + raw_spin_unlock(&task->pi_lock);
19445 put_task_struct(task);
19448 @@ -718,21 +740,24 @@ static int rt_mutex_adjust_prio_chain(struct task_struct *task,
19449 * follow here. This is the end of the chain we are walking.
19451 if (!rt_mutex_owner(lock)) {
19452 + struct rt_mutex_waiter *lock_top_waiter;
19455 * If the requeue [7] above changed the top waiter,
19456 * then we need to wake the new top waiter up to try
19459 - if (prerequeue_top_waiter != rt_mutex_top_waiter(lock))
19460 - wake_up_process(rt_mutex_top_waiter(lock)->task);
19461 - raw_spin_unlock(&lock->wait_lock);
19462 + lock_top_waiter = rt_mutex_top_waiter(lock);
19463 + if (prerequeue_top_waiter != lock_top_waiter)
19464 + rt_mutex_wake_waiter(lock_top_waiter);
19465 + raw_spin_unlock_irq(&lock->wait_lock);
19469 /* [10] Grab the next task, i.e. the owner of @lock */
19470 task = rt_mutex_owner(lock);
19471 get_task_struct(task);
19472 - raw_spin_lock_irqsave(&task->pi_lock, flags);
19473 + raw_spin_lock(&task->pi_lock);
19475 /* [11] requeue the pi waiters if necessary */
19476 if (waiter == rt_mutex_top_waiter(lock)) {
19477 @@ -786,8 +811,8 @@ static int rt_mutex_adjust_prio_chain(struct task_struct *task,
19478 top_waiter = rt_mutex_top_waiter(lock);
19480 /* [13] Drop the locks */
19481 - raw_spin_unlock_irqrestore(&task->pi_lock, flags);
19482 - raw_spin_unlock(&lock->wait_lock);
19483 + raw_spin_unlock(&task->pi_lock);
19484 + raw_spin_unlock_irq(&lock->wait_lock);
19487 * Make the actual exit decisions [12], based on the stored
19488 @@ -810,28 +835,46 @@ static int rt_mutex_adjust_prio_chain(struct task_struct *task,
19492 - raw_spin_unlock_irqrestore(&task->pi_lock, flags);
19493 + raw_spin_unlock_irq(&task->pi_lock);
19495 put_task_struct(task);
19501 +#define STEAL_NORMAL 0
19502 +#define STEAL_LATERAL 1
19505 + * Note that RT tasks are excluded from lateral-steals to prevent the
19506 + * introduction of an unbounded latency
19508 +static inline int lock_is_stealable(struct task_struct *task,
19509 + struct task_struct *pendowner, int mode)
19511 + if (mode == STEAL_NORMAL || rt_task(task)) {
19512 + if (task->prio >= pendowner->prio)
19514 + } else if (task->prio > pendowner->prio)
19520 * Try to take an rt-mutex
19522 - * Must be called with lock->wait_lock held.
19523 + * Must be called with lock->wait_lock held and interrupts disabled
19525 * @lock: The lock to be acquired.
19526 * @task: The task which wants to acquire the lock
19527 * @waiter: The waiter that is queued to the lock's wait tree if the
19528 * callsite called task_blocked_on_lock(), otherwise NULL
19530 -static int try_to_take_rt_mutex(struct rt_mutex *lock, struct task_struct *task,
19531 - struct rt_mutex_waiter *waiter)
19532 +static int __try_to_take_rt_mutex(struct rt_mutex *lock,
19533 + struct task_struct *task,
19534 + struct rt_mutex_waiter *waiter, int mode)
19536 - unsigned long flags;
19539 * Before testing whether we can acquire @lock, we set the
19540 * RT_MUTEX_HAS_WAITERS bit in @lock->owner. This forces all
19541 @@ -867,8 +910,10 @@ static int try_to_take_rt_mutex(struct rt_mutex *lock, struct task_struct *task,
19542 * If waiter is not the highest priority waiter of
19545 - if (waiter != rt_mutex_top_waiter(lock))
19546 + if (waiter != rt_mutex_top_waiter(lock)) {
19547 + /* XXX lock_is_stealable() ? */
19552 * We can acquire the lock. Remove the waiter from the
19553 @@ -886,14 +931,10 @@ static int try_to_take_rt_mutex(struct rt_mutex *lock, struct task_struct *task,
19554 * not need to be dequeued.
19556 if (rt_mutex_has_waiters(lock)) {
19558 - * If @task->prio is greater than or equal to
19559 - * the top waiter priority (kernel view),
19562 - if (task->prio >= rt_mutex_top_waiter(lock)->prio)
19564 + struct task_struct *pown = rt_mutex_top_waiter(lock)->task;
19566 + if (task != pown && !lock_is_stealable(task, pown, mode))
19569 * The current top waiter stays enqueued. We
19570 * don't have to change anything in the lock
19571 @@ -916,7 +957,7 @@ static int try_to_take_rt_mutex(struct rt_mutex *lock, struct task_struct *task,
19572 * case, but conditionals are more expensive than a redundant
19575 - raw_spin_lock_irqsave(&task->pi_lock, flags);
19576 + raw_spin_lock(&task->pi_lock);
19577 task->pi_blocked_on = NULL;
19579 * Finish the lock acquisition. @task is the new owner. If
19580 @@ -925,7 +966,7 @@ static int try_to_take_rt_mutex(struct rt_mutex *lock, struct task_struct *task,
19582 if (rt_mutex_has_waiters(lock))
19583 rt_mutex_enqueue_pi(task, rt_mutex_top_waiter(lock));
19584 - raw_spin_unlock_irqrestore(&task->pi_lock, flags);
19585 + raw_spin_unlock(&task->pi_lock);
19588 /* We got the lock. */
19589 @@ -942,12 +983,444 @@ takeit:
19593 +#ifdef CONFIG_PREEMPT_RT_FULL
19595 + * preemptible spin_lock functions:
19597 +static inline void rt_spin_lock_fastlock(struct rt_mutex *lock,
19598 + void (*slowfn)(struct rt_mutex *lock,
19602 + might_sleep_no_state_check();
19605 + migrate_disable();
19607 + if (likely(rt_mutex_cmpxchg_acquire(lock, NULL, current)))
19608 + rt_mutex_deadlock_account_lock(lock, current);
19610 + slowfn(lock, do_mig_dis);
19613 +static inline int rt_spin_lock_fastunlock(struct rt_mutex *lock,
19614 + int (*slowfn)(struct rt_mutex *lock))
19616 + if (likely(rt_mutex_cmpxchg_release(lock, current, NULL))) {
19617 + rt_mutex_deadlock_account_unlock(current);
19620 + return slowfn(lock);
19624 + * Note that owner is a speculative pointer and dereferencing relies
19625 + * on rcu_read_lock() and the check against the lock owner.
19627 +static int adaptive_wait(struct rt_mutex *lock,
19628 + struct task_struct *owner)
19634 + if (owner != rt_mutex_owner(lock))
19637 + * Ensure that owner->on_cpu is dereferenced _after_
19638 + * checking the above to be valid.
19641 + if (!owner->on_cpu) {
19647 + rcu_read_unlock();
19651 +static int adaptive_wait(struct rt_mutex *lock,
19652 + struct task_struct *orig_owner)
19658 +static int task_blocks_on_rt_mutex(struct rt_mutex *lock,
19659 + struct rt_mutex_waiter *waiter,
19660 + struct task_struct *task,
19661 + enum rtmutex_chainwalk chwalk);
19663 + * Slow path lock function spin_lock style: this variant is very
19664 + * careful not to miss any non-lock wakeups.
19666 + * We store the current state under p->pi_lock in p->saved_state and
19667 + * the try_to_wake_up() code handles this accordingly.
19669 +static void noinline __sched rt_spin_lock_slowlock(struct rt_mutex *lock,
19672 + struct task_struct *lock_owner, *self = current;
19673 + struct rt_mutex_waiter waiter, *top_waiter;
19674 + unsigned long flags;
19677 + rt_mutex_init_waiter(&waiter, true);
19679 + raw_spin_lock_irqsave(&lock->wait_lock, flags);
19681 + if (__try_to_take_rt_mutex(lock, self, NULL, STEAL_LATERAL)) {
19682 + raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
19686 + BUG_ON(rt_mutex_owner(lock) == self);
19689 + * We save whatever state the task is in and we'll restore it
19690 + * after acquiring the lock taking real wakeups into account
19691 + * as well. We are serialized via pi_lock against wakeups. See
19692 + * try_to_wake_up().
19694 + raw_spin_lock(&self->pi_lock);
19695 + self->saved_state = self->state;
19696 + __set_current_state_no_track(TASK_UNINTERRUPTIBLE);
19697 + raw_spin_unlock(&self->pi_lock);
19699 + ret = task_blocks_on_rt_mutex(lock, &waiter, self, RT_MUTEX_MIN_CHAINWALK);
19703 + /* Try to acquire the lock again. */
19704 + if (__try_to_take_rt_mutex(lock, self, &waiter, STEAL_LATERAL))
19707 + top_waiter = rt_mutex_top_waiter(lock);
19708 + lock_owner = rt_mutex_owner(lock);
19710 + raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
19712 + debug_rt_mutex_print_deadlock(&waiter);
19714 + if (top_waiter != &waiter || adaptive_wait(lock, lock_owner)) {
19716 + migrate_enable();
19719 + migrate_disable();
19722 + raw_spin_lock_irqsave(&lock->wait_lock, flags);
19724 + raw_spin_lock(&self->pi_lock);
19725 + __set_current_state_no_track(TASK_UNINTERRUPTIBLE);
19726 + raw_spin_unlock(&self->pi_lock);
19730 + * Restore the task state to current->saved_state. We set it
19731 + * to the original state above and the try_to_wake_up() code
19732 + * has possibly updated it when a real (non-rtmutex) wakeup
19733 + * happened while we were blocked. Clear saved_state so
19734 + * try_to_wakeup() does not get confused.
19736 + raw_spin_lock(&self->pi_lock);
19737 + __set_current_state_no_track(self->saved_state);
19738 + self->saved_state = TASK_RUNNING;
19739 + raw_spin_unlock(&self->pi_lock);
19742 + * try_to_take_rt_mutex() sets the waiter bit
19743 + * unconditionally. We might have to fix that up:
19745 + fixup_rt_mutex_waiters(lock);
19747 + BUG_ON(rt_mutex_has_waiters(lock) && &waiter == rt_mutex_top_waiter(lock));
19748 + BUG_ON(!RB_EMPTY_NODE(&waiter.tree_entry));
19750 + raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
19752 + debug_rt_mutex_free_waiter(&waiter);
19755 +static void mark_wakeup_next_waiter(struct wake_q_head *wake_q,
19756 + struct wake_q_head *wake_sleeper_q,
19757 + struct rt_mutex *lock);
19759 + * Slow path to release a rt_mutex spin_lock style
19761 +static int noinline __sched rt_spin_lock_slowunlock(struct rt_mutex *lock)
19763 + unsigned long flags;
19765 + WAKE_Q(wake_sleeper_q);
19767 + raw_spin_lock_irqsave(&lock->wait_lock, flags);
19769 + debug_rt_mutex_unlock(lock);
19771 + rt_mutex_deadlock_account_unlock(current);
19773 + if (!rt_mutex_has_waiters(lock)) {
19774 + lock->owner = NULL;
19775 + raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
19779 + mark_wakeup_next_waiter(&wake_q, &wake_sleeper_q, lock);
19781 + raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
19782 + wake_up_q(&wake_q);
19783 + wake_up_q_sleeper(&wake_sleeper_q);
19785 + /* Undo pi boosting.when necessary */
19786 + rt_mutex_adjust_prio(current);
19790 +static int noinline __sched rt_spin_lock_slowunlock_no_deboost(struct rt_mutex *lock)
19792 + unsigned long flags;
19794 + WAKE_Q(wake_sleeper_q);
19796 + raw_spin_lock_irqsave(&lock->wait_lock, flags);
19798 + debug_rt_mutex_unlock(lock);
19800 + rt_mutex_deadlock_account_unlock(current);
19802 + if (!rt_mutex_has_waiters(lock)) {
19803 + lock->owner = NULL;
19804 + raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
19808 + mark_wakeup_next_waiter(&wake_q, &wake_sleeper_q, lock);
19810 + raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
19811 + wake_up_q(&wake_q);
19812 + wake_up_q_sleeper(&wake_sleeper_q);
19816 +void __lockfunc rt_spin_lock__no_mg(spinlock_t *lock)
19818 + rt_spin_lock_fastlock(&lock->lock, rt_spin_lock_slowlock, false);
19819 + spin_acquire(&lock->dep_map, 0, 0, _RET_IP_);
19821 +EXPORT_SYMBOL(rt_spin_lock__no_mg);
19823 +void __lockfunc rt_spin_lock(spinlock_t *lock)
19825 + rt_spin_lock_fastlock(&lock->lock, rt_spin_lock_slowlock, true);
19826 + spin_acquire(&lock->dep_map, 0, 0, _RET_IP_);
19828 +EXPORT_SYMBOL(rt_spin_lock);
19830 +void __lockfunc __rt_spin_lock(struct rt_mutex *lock)
19832 + rt_spin_lock_fastlock(lock, rt_spin_lock_slowlock, true);
19834 +EXPORT_SYMBOL(__rt_spin_lock);
19836 +void __lockfunc __rt_spin_lock__no_mg(struct rt_mutex *lock)
19838 + rt_spin_lock_fastlock(lock, rt_spin_lock_slowlock, false);
19840 +EXPORT_SYMBOL(__rt_spin_lock__no_mg);
19842 +#ifdef CONFIG_DEBUG_LOCK_ALLOC
19843 +void __lockfunc rt_spin_lock_nested(spinlock_t *lock, int subclass)
19845 + spin_acquire(&lock->dep_map, subclass, 0, _RET_IP_);
19846 + rt_spin_lock_fastlock(&lock->lock, rt_spin_lock_slowlock, true);
19848 +EXPORT_SYMBOL(rt_spin_lock_nested);
19851 +void __lockfunc rt_spin_unlock__no_mg(spinlock_t *lock)
19853 + /* NOTE: we always pass in '1' for nested, for simplicity */
19854 + spin_release(&lock->dep_map, 1, _RET_IP_);
19855 + rt_spin_lock_fastunlock(&lock->lock, rt_spin_lock_slowunlock);
19857 +EXPORT_SYMBOL(rt_spin_unlock__no_mg);
19859 +void __lockfunc rt_spin_unlock(spinlock_t *lock)
19861 + /* NOTE: we always pass in '1' for nested, for simplicity */
19862 + spin_release(&lock->dep_map, 1, _RET_IP_);
19863 + rt_spin_lock_fastunlock(&lock->lock, rt_spin_lock_slowunlock);
19864 + migrate_enable();
19866 +EXPORT_SYMBOL(rt_spin_unlock);
19868 +int __lockfunc rt_spin_unlock_no_deboost(spinlock_t *lock)
19872 + /* NOTE: we always pass in '1' for nested, for simplicity */
19873 + spin_release(&lock->dep_map, 1, _RET_IP_);
19874 + ret = rt_spin_lock_fastunlock(&lock->lock, rt_spin_lock_slowunlock_no_deboost);
19875 + migrate_enable();
19879 +void __lockfunc __rt_spin_unlock(struct rt_mutex *lock)
19881 + rt_spin_lock_fastunlock(lock, rt_spin_lock_slowunlock);
19883 +EXPORT_SYMBOL(__rt_spin_unlock);
19886 + * Wait for the lock to get unlocked: instead of polling for an unlock
19887 + * (like raw spinlocks do), we lock and unlock, to force the kernel to
19888 + * schedule if there's contention:
19890 +void __lockfunc rt_spin_unlock_wait(spinlock_t *lock)
19893 + spin_unlock(lock);
19895 +EXPORT_SYMBOL(rt_spin_unlock_wait);
19897 +int __lockfunc __rt_spin_trylock(struct rt_mutex *lock)
19899 + return rt_mutex_trylock(lock);
19902 +int __lockfunc rt_spin_trylock__no_mg(spinlock_t *lock)
19906 + ret = rt_mutex_trylock(&lock->lock);
19908 + spin_acquire(&lock->dep_map, 0, 1, _RET_IP_);
19911 +EXPORT_SYMBOL(rt_spin_trylock__no_mg);
19913 +int __lockfunc rt_spin_trylock(spinlock_t *lock)
19917 + migrate_disable();
19918 + ret = rt_mutex_trylock(&lock->lock);
19920 + spin_acquire(&lock->dep_map, 0, 1, _RET_IP_);
19922 + migrate_enable();
19925 +EXPORT_SYMBOL(rt_spin_trylock);
19927 +int __lockfunc rt_spin_trylock_bh(spinlock_t *lock)
19931 + local_bh_disable();
19932 + ret = rt_mutex_trylock(&lock->lock);
19934 + migrate_disable();
19935 + spin_acquire(&lock->dep_map, 0, 1, _RET_IP_);
19937 + local_bh_enable();
19940 +EXPORT_SYMBOL(rt_spin_trylock_bh);
19942 +int __lockfunc rt_spin_trylock_irqsave(spinlock_t *lock, unsigned long *flags)
19947 + ret = rt_mutex_trylock(&lock->lock);
19949 + migrate_disable();
19950 + spin_acquire(&lock->dep_map, 0, 1, _RET_IP_);
19954 +EXPORT_SYMBOL(rt_spin_trylock_irqsave);
19956 +int atomic_dec_and_spin_lock(atomic_t *atomic, spinlock_t *lock)
19958 + /* Subtract 1 from counter unless that drops it to 0 (ie. it was 1) */
19959 + if (atomic_add_unless(atomic, -1, 1))
19961 + rt_spin_lock(lock);
19962 + if (atomic_dec_and_test(atomic))
19964 + rt_spin_unlock(lock);
19967 +EXPORT_SYMBOL(atomic_dec_and_spin_lock);
19970 +__rt_spin_lock_init(spinlock_t *lock, char *name, struct lock_class_key *key)
19972 +#ifdef CONFIG_DEBUG_LOCK_ALLOC
19974 + * Make sure we are not reinitializing a held lock:
19976 + debug_check_no_locks_freed((void *)lock, sizeof(*lock));
19977 + lockdep_init_map(&lock->dep_map, name, key, 0);
19980 +EXPORT_SYMBOL(__rt_spin_lock_init);
19982 +#endif /* PREEMPT_RT_FULL */
19984 +#ifdef CONFIG_PREEMPT_RT_FULL
19985 + static inline int __sched
19986 +__mutex_lock_check_stamp(struct rt_mutex *lock, struct ww_acquire_ctx *ctx)
19988 + struct ww_mutex *ww = container_of(lock, struct ww_mutex, base.lock);
19989 + struct ww_acquire_ctx *hold_ctx = ACCESS_ONCE(ww->ctx);
19994 + if (unlikely(ctx == hold_ctx))
19995 + return -EALREADY;
19997 + if (ctx->stamp - hold_ctx->stamp <= LONG_MAX &&
19998 + (ctx->stamp != hold_ctx->stamp || ctx > hold_ctx)) {
19999 +#ifdef CONFIG_DEBUG_MUTEXES
20000 + DEBUG_LOCKS_WARN_ON(ctx->contending_lock);
20001 + ctx->contending_lock = ww;
20009 + static inline int __sched
20010 +__mutex_lock_check_stamp(struct rt_mutex *lock, struct ww_acquire_ctx *ctx)
20019 +try_to_take_rt_mutex(struct rt_mutex *lock, struct task_struct *task,
20020 + struct rt_mutex_waiter *waiter)
20022 + return __try_to_take_rt_mutex(lock, task, waiter, STEAL_NORMAL);
20026 * Task blocks on lock.
20028 * Prepare waiter and propagate pi chain
20030 - * This must be called with lock->wait_lock held.
20031 + * This must be called with lock->wait_lock held and interrupts disabled
20033 static int task_blocks_on_rt_mutex(struct rt_mutex *lock,
20034 struct rt_mutex_waiter *waiter,
20035 @@ -958,7 +1431,6 @@ static int task_blocks_on_rt_mutex(struct rt_mutex *lock,
20036 struct rt_mutex_waiter *top_waiter = waiter;
20037 struct rt_mutex *next_lock;
20038 int chain_walk = 0, res;
20039 - unsigned long flags;
20042 * Early deadlock detection. We really don't want the task to
20043 @@ -972,7 +1444,24 @@ static int task_blocks_on_rt_mutex(struct rt_mutex *lock,
20047 - raw_spin_lock_irqsave(&task->pi_lock, flags);
20048 + raw_spin_lock(&task->pi_lock);
20051 + * In the case of futex requeue PI, this will be a proxy
20052 + * lock. The task will wake unaware that it is enqueueed on
20053 + * this lock. Avoid blocking on two locks and corrupting
20054 + * pi_blocked_on via the PI_WAKEUP_INPROGRESS
20055 + * flag. futex_wait_requeue_pi() sets this when it wakes up
20056 + * before requeue (due to a signal or timeout). Do not enqueue
20057 + * the task if PI_WAKEUP_INPROGRESS is set.
20059 + if (task != current && task->pi_blocked_on == PI_WAKEUP_INPROGRESS) {
20060 + raw_spin_unlock(&task->pi_lock);
20064 + BUG_ON(rt_mutex_real_waiter(task->pi_blocked_on));
20066 __rt_mutex_adjust_prio(task);
20067 waiter->task = task;
20068 waiter->lock = lock;
20069 @@ -985,18 +1474,18 @@ static int task_blocks_on_rt_mutex(struct rt_mutex *lock,
20071 task->pi_blocked_on = waiter;
20073 - raw_spin_unlock_irqrestore(&task->pi_lock, flags);
20074 + raw_spin_unlock(&task->pi_lock);
20079 - raw_spin_lock_irqsave(&owner->pi_lock, flags);
20080 + raw_spin_lock(&owner->pi_lock);
20081 if (waiter == rt_mutex_top_waiter(lock)) {
20082 rt_mutex_dequeue_pi(owner, top_waiter);
20083 rt_mutex_enqueue_pi(owner, waiter);
20085 __rt_mutex_adjust_prio(owner);
20086 - if (owner->pi_blocked_on)
20087 + if (rt_mutex_real_waiter(owner->pi_blocked_on))
20089 } else if (rt_mutex_cond_detect_deadlock(waiter, chwalk)) {
20091 @@ -1005,7 +1494,7 @@ static int task_blocks_on_rt_mutex(struct rt_mutex *lock,
20092 /* Store the lock on which owner is blocked or NULL */
20093 next_lock = task_blocked_on_lock(owner);
20095 - raw_spin_unlock_irqrestore(&owner->pi_lock, flags);
20096 + raw_spin_unlock(&owner->pi_lock);
20098 * Even if full deadlock detection is on, if the owner is not
20099 * blocked itself, we can avoid finding this out in the chain
20100 @@ -1021,12 +1510,12 @@ static int task_blocks_on_rt_mutex(struct rt_mutex *lock,
20102 get_task_struct(owner);
20104 - raw_spin_unlock(&lock->wait_lock);
20105 + raw_spin_unlock_irq(&lock->wait_lock);
20107 res = rt_mutex_adjust_prio_chain(owner, chwalk, lock,
20108 next_lock, waiter, task);
20110 - raw_spin_lock(&lock->wait_lock);
20111 + raw_spin_lock_irq(&lock->wait_lock);
20115 @@ -1035,15 +1524,15 @@ static int task_blocks_on_rt_mutex(struct rt_mutex *lock,
20116 * Remove the top waiter from the current tasks pi waiter tree and
20119 - * Called with lock->wait_lock held.
20120 + * Called with lock->wait_lock held and interrupts disabled.
20122 static void mark_wakeup_next_waiter(struct wake_q_head *wake_q,
20123 + struct wake_q_head *wake_sleeper_q,
20124 struct rt_mutex *lock)
20126 struct rt_mutex_waiter *waiter;
20127 - unsigned long flags;
20129 - raw_spin_lock_irqsave(¤t->pi_lock, flags);
20130 + raw_spin_lock(¤t->pi_lock);
20132 waiter = rt_mutex_top_waiter(lock);
20134 @@ -1065,15 +1554,18 @@ static void mark_wakeup_next_waiter(struct wake_q_head *wake_q,
20136 lock->owner = (void *) RT_MUTEX_HAS_WAITERS;
20138 - raw_spin_unlock_irqrestore(¤t->pi_lock, flags);
20139 + raw_spin_unlock(¤t->pi_lock);
20141 - wake_q_add(wake_q, waiter->task);
20142 + if (waiter->savestate)
20143 + wake_q_add(wake_sleeper_q, waiter->task);
20145 + wake_q_add(wake_q, waiter->task);
20149 * Remove a waiter from a lock and give up
20151 - * Must be called with lock->wait_lock held and
20152 + * Must be called with lock->wait_lock held and interrupts disabled. I must
20153 * have just failed to try_to_take_rt_mutex().
20155 static void remove_waiter(struct rt_mutex *lock,
20156 @@ -1081,13 +1573,12 @@ static void remove_waiter(struct rt_mutex *lock,
20158 bool is_top_waiter = (waiter == rt_mutex_top_waiter(lock));
20159 struct task_struct *owner = rt_mutex_owner(lock);
20160 - struct rt_mutex *next_lock;
20161 - unsigned long flags;
20162 + struct rt_mutex *next_lock = NULL;
20164 - raw_spin_lock_irqsave(¤t->pi_lock, flags);
20165 + raw_spin_lock(¤t->pi_lock);
20166 rt_mutex_dequeue(lock, waiter);
20167 current->pi_blocked_on = NULL;
20168 - raw_spin_unlock_irqrestore(¤t->pi_lock, flags);
20169 + raw_spin_unlock(¤t->pi_lock);
20172 * Only update priority if the waiter was the highest priority
20173 @@ -1096,7 +1587,7 @@ static void remove_waiter(struct rt_mutex *lock,
20174 if (!owner || !is_top_waiter)
20177 - raw_spin_lock_irqsave(&owner->pi_lock, flags);
20178 + raw_spin_lock(&owner->pi_lock);
20180 rt_mutex_dequeue_pi(owner, waiter);
20182 @@ -1106,9 +1597,10 @@ static void remove_waiter(struct rt_mutex *lock,
20183 __rt_mutex_adjust_prio(owner);
20185 /* Store the lock on which owner is blocked or NULL */
20186 - next_lock = task_blocked_on_lock(owner);
20187 + if (rt_mutex_real_waiter(owner->pi_blocked_on))
20188 + next_lock = task_blocked_on_lock(owner);
20190 - raw_spin_unlock_irqrestore(&owner->pi_lock, flags);
20191 + raw_spin_unlock(&owner->pi_lock);
20194 * Don't walk the chain, if the owner task is not blocked
20195 @@ -1120,12 +1612,12 @@ static void remove_waiter(struct rt_mutex *lock,
20196 /* gets dropped in rt_mutex_adjust_prio_chain()! */
20197 get_task_struct(owner);
20199 - raw_spin_unlock(&lock->wait_lock);
20200 + raw_spin_unlock_irq(&lock->wait_lock);
20202 rt_mutex_adjust_prio_chain(owner, RT_MUTEX_MIN_CHAINWALK, lock,
20203 next_lock, NULL, current);
20205 - raw_spin_lock(&lock->wait_lock);
20206 + raw_spin_lock_irq(&lock->wait_lock);
20210 @@ -1142,17 +1634,17 @@ void rt_mutex_adjust_pi(struct task_struct *task)
20211 raw_spin_lock_irqsave(&task->pi_lock, flags);
20213 waiter = task->pi_blocked_on;
20214 - if (!waiter || (waiter->prio == task->prio &&
20215 + if (!rt_mutex_real_waiter(waiter) || (waiter->prio == task->prio &&
20216 !dl_prio(task->prio))) {
20217 raw_spin_unlock_irqrestore(&task->pi_lock, flags);
20220 next_lock = waiter->lock;
20221 - raw_spin_unlock_irqrestore(&task->pi_lock, flags);
20223 /* gets dropped in rt_mutex_adjust_prio_chain()! */
20224 get_task_struct(task);
20226 + raw_spin_unlock_irqrestore(&task->pi_lock, flags);
20227 rt_mutex_adjust_prio_chain(task, RT_MUTEX_MIN_CHAINWALK, NULL,
20228 next_lock, NULL, task);
20230 @@ -1161,16 +1653,17 @@ void rt_mutex_adjust_pi(struct task_struct *task)
20231 * __rt_mutex_slowlock() - Perform the wait-wake-try-to-take loop
20232 * @lock: the rt_mutex to take
20233 * @state: the state the task should block in (TASK_INTERRUPTIBLE
20234 - * or TASK_UNINTERRUPTIBLE)
20235 + * or TASK_UNINTERRUPTIBLE)
20236 * @timeout: the pre-initialized and started timer, or NULL for none
20237 * @waiter: the pre-initialized rt_mutex_waiter
20239 - * lock->wait_lock must be held by the caller.
20240 + * Must be called with lock->wait_lock held and interrupts disabled
20243 __rt_mutex_slowlock(struct rt_mutex *lock, int state,
20244 struct hrtimer_sleeper *timeout,
20245 - struct rt_mutex_waiter *waiter)
20246 + struct rt_mutex_waiter *waiter,
20247 + struct ww_acquire_ctx *ww_ctx)
20251 @@ -1193,13 +1686,19 @@ __rt_mutex_slowlock(struct rt_mutex *lock, int state,
20255 - raw_spin_unlock(&lock->wait_lock);
20256 + if (ww_ctx && ww_ctx->acquired > 0) {
20257 + ret = __mutex_lock_check_stamp(lock, ww_ctx);
20262 + raw_spin_unlock_irq(&lock->wait_lock);
20264 debug_rt_mutex_print_deadlock(waiter);
20268 - raw_spin_lock(&lock->wait_lock);
20269 + raw_spin_lock_irq(&lock->wait_lock);
20270 set_current_state(state);
20273 @@ -1227,26 +1726,112 @@ static void rt_mutex_handle_deadlock(int res, int detect_deadlock,
20277 +static __always_inline void ww_mutex_lock_acquired(struct ww_mutex *ww,
20278 + struct ww_acquire_ctx *ww_ctx)
20280 +#ifdef CONFIG_DEBUG_MUTEXES
20282 + * If this WARN_ON triggers, you used ww_mutex_lock to acquire,
20283 + * but released with a normal mutex_unlock in this call.
20285 + * This should never happen, always use ww_mutex_unlock.
20287 + DEBUG_LOCKS_WARN_ON(ww->ctx);
20290 + * Not quite done after calling ww_acquire_done() ?
20292 + DEBUG_LOCKS_WARN_ON(ww_ctx->done_acquire);
20294 + if (ww_ctx->contending_lock) {
20296 + * After -EDEADLK you tried to
20297 + * acquire a different ww_mutex? Bad!
20299 + DEBUG_LOCKS_WARN_ON(ww_ctx->contending_lock != ww);
20302 + * You called ww_mutex_lock after receiving -EDEADLK,
20303 + * but 'forgot' to unlock everything else first?
20305 + DEBUG_LOCKS_WARN_ON(ww_ctx->acquired > 0);
20306 + ww_ctx->contending_lock = NULL;
20310 + * Naughty, using a different class will lead to undefined behavior!
20312 + DEBUG_LOCKS_WARN_ON(ww_ctx->ww_class != ww->ww_class);
20314 + ww_ctx->acquired++;
20317 +#ifdef CONFIG_PREEMPT_RT_FULL
20318 +static void ww_mutex_account_lock(struct rt_mutex *lock,
20319 + struct ww_acquire_ctx *ww_ctx)
20321 + struct ww_mutex *ww = container_of(lock, struct ww_mutex, base.lock);
20322 + struct rt_mutex_waiter *waiter, *n;
20325 + * This branch gets optimized out for the common case,
20326 + * and is only important for ww_mutex_lock.
20328 + ww_mutex_lock_acquired(ww, ww_ctx);
20329 + ww->ctx = ww_ctx;
20332 + * Give any possible sleeping processes the chance to wake up,
20333 + * so they can recheck if they have to back off.
20335 + rbtree_postorder_for_each_entry_safe(waiter, n, &lock->waiters,
20337 + /* XXX debug rt mutex waiter wakeup */
20339 + BUG_ON(waiter->lock != lock);
20340 + rt_mutex_wake_waiter(waiter);
20346 +static void ww_mutex_account_lock(struct rt_mutex *lock,
20347 + struct ww_acquire_ctx *ww_ctx)
20354 * Slow path lock function:
20357 rt_mutex_slowlock(struct rt_mutex *lock, int state,
20358 struct hrtimer_sleeper *timeout,
20359 - enum rtmutex_chainwalk chwalk)
20360 + enum rtmutex_chainwalk chwalk,
20361 + struct ww_acquire_ctx *ww_ctx)
20363 struct rt_mutex_waiter waiter;
20364 + unsigned long flags;
20367 - debug_rt_mutex_init_waiter(&waiter);
20368 - RB_CLEAR_NODE(&waiter.pi_tree_entry);
20369 - RB_CLEAR_NODE(&waiter.tree_entry);
20370 + rt_mutex_init_waiter(&waiter, false);
20372 - raw_spin_lock(&lock->wait_lock);
20374 + * Technically we could use raw_spin_[un]lock_irq() here, but this can
20375 + * be called in early boot if the cmpxchg() fast path is disabled
20376 + * (debug, no architecture support). In this case we will acquire the
20377 + * rtmutex with lock->wait_lock held. But we cannot unconditionally
20378 + * enable interrupts in that early boot case. So we need to use the
20379 + * irqsave/restore variants.
20381 + raw_spin_lock_irqsave(&lock->wait_lock, flags);
20383 /* Try to acquire the lock again: */
20384 if (try_to_take_rt_mutex(lock, current, NULL)) {
20385 - raw_spin_unlock(&lock->wait_lock);
20387 + ww_mutex_account_lock(lock, ww_ctx);
20388 + raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
20392 @@ -1260,13 +1845,23 @@ rt_mutex_slowlock(struct rt_mutex *lock, int state,
20395 /* sleep on the mutex */
20396 - ret = __rt_mutex_slowlock(lock, state, timeout, &waiter);
20397 + ret = __rt_mutex_slowlock(lock, state, timeout, &waiter,
20399 + else if (ww_ctx) {
20400 + /* ww_mutex received EDEADLK, let it become EALREADY */
20401 + ret = __mutex_lock_check_stamp(lock, ww_ctx);
20405 if (unlikely(ret)) {
20406 __set_current_state(TASK_RUNNING);
20407 if (rt_mutex_has_waiters(lock))
20408 remove_waiter(lock, &waiter);
20409 - rt_mutex_handle_deadlock(ret, chwalk, &waiter);
20410 + /* ww_mutex want to report EDEADLK/EALREADY, let them */
20412 + rt_mutex_handle_deadlock(ret, chwalk, &waiter);
20413 + } else if (ww_ctx) {
20414 + ww_mutex_account_lock(lock, ww_ctx);
20418 @@ -1275,7 +1870,7 @@ rt_mutex_slowlock(struct rt_mutex *lock, int state,
20420 fixup_rt_mutex_waiters(lock);
20422 - raw_spin_unlock(&lock->wait_lock);
20423 + raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
20425 /* Remove pending timer: */
20426 if (unlikely(timeout))
20427 @@ -1291,6 +1886,7 @@ rt_mutex_slowlock(struct rt_mutex *lock, int state,
20429 static inline int rt_mutex_slowtrylock(struct rt_mutex *lock)
20431 + unsigned long flags;
20435 @@ -1302,10 +1898,10 @@ static inline int rt_mutex_slowtrylock(struct rt_mutex *lock)
20439 - * The mutex has currently no owner. Lock the wait lock and
20440 - * try to acquire the lock.
20441 + * The mutex has currently no owner. Lock the wait lock and try to
20442 + * acquire the lock. We use irqsave here to support early boot calls.
20444 - raw_spin_lock(&lock->wait_lock);
20445 + raw_spin_lock_irqsave(&lock->wait_lock, flags);
20447 ret = try_to_take_rt_mutex(lock, current, NULL);
20449 @@ -1315,7 +1911,7 @@ static inline int rt_mutex_slowtrylock(struct rt_mutex *lock)
20451 fixup_rt_mutex_waiters(lock);
20453 - raw_spin_unlock(&lock->wait_lock);
20454 + raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
20458 @@ -1325,9 +1921,13 @@ static inline int rt_mutex_slowtrylock(struct rt_mutex *lock)
20459 * Return whether the current task needs to undo a potential priority boosting.
20461 static bool __sched rt_mutex_slowunlock(struct rt_mutex *lock,
20462 - struct wake_q_head *wake_q)
20463 + struct wake_q_head *wake_q,
20464 + struct wake_q_head *wake_sleeper_q)
20466 - raw_spin_lock(&lock->wait_lock);
20467 + unsigned long flags;
20469 + /* irqsave required to support early boot calls */
20470 + raw_spin_lock_irqsave(&lock->wait_lock, flags);
20472 debug_rt_mutex_unlock(lock);
20474 @@ -1366,10 +1966,10 @@ static bool __sched rt_mutex_slowunlock(struct rt_mutex *lock,
20476 while (!rt_mutex_has_waiters(lock)) {
20477 /* Drops lock->wait_lock ! */
20478 - if (unlock_rt_mutex_safe(lock) == true)
20479 + if (unlock_rt_mutex_safe(lock, flags) == true)
20481 /* Relock the rtmutex and try again */
20482 - raw_spin_lock(&lock->wait_lock);
20483 + raw_spin_lock_irqsave(&lock->wait_lock, flags);
20487 @@ -1378,9 +1978,9 @@ static bool __sched rt_mutex_slowunlock(struct rt_mutex *lock,
20489 * Queue the next waiter for wakeup once we release the wait_lock.
20491 - mark_wakeup_next_waiter(wake_q, lock);
20492 + mark_wakeup_next_waiter(wake_q, wake_sleeper_q, lock);
20494 - raw_spin_unlock(&lock->wait_lock);
20495 + raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
20497 /* check PI boosting */
20499 @@ -1394,31 +1994,36 @@ static bool __sched rt_mutex_slowunlock(struct rt_mutex *lock,
20502 rt_mutex_fastlock(struct rt_mutex *lock, int state,
20503 + struct ww_acquire_ctx *ww_ctx,
20504 int (*slowfn)(struct rt_mutex *lock, int state,
20505 struct hrtimer_sleeper *timeout,
20506 - enum rtmutex_chainwalk chwalk))
20507 + enum rtmutex_chainwalk chwalk,
20508 + struct ww_acquire_ctx *ww_ctx))
20510 if (likely(rt_mutex_cmpxchg_acquire(lock, NULL, current))) {
20511 rt_mutex_deadlock_account_lock(lock, current);
20514 - return slowfn(lock, state, NULL, RT_MUTEX_MIN_CHAINWALK);
20515 + return slowfn(lock, state, NULL, RT_MUTEX_MIN_CHAINWALK,
20520 rt_mutex_timed_fastlock(struct rt_mutex *lock, int state,
20521 struct hrtimer_sleeper *timeout,
20522 enum rtmutex_chainwalk chwalk,
20523 + struct ww_acquire_ctx *ww_ctx,
20524 int (*slowfn)(struct rt_mutex *lock, int state,
20525 struct hrtimer_sleeper *timeout,
20526 - enum rtmutex_chainwalk chwalk))
20527 + enum rtmutex_chainwalk chwalk,
20528 + struct ww_acquire_ctx *ww_ctx))
20530 if (chwalk == RT_MUTEX_MIN_CHAINWALK &&
20531 likely(rt_mutex_cmpxchg_acquire(lock, NULL, current))) {
20532 rt_mutex_deadlock_account_lock(lock, current);
20535 - return slowfn(lock, state, timeout, chwalk);
20536 + return slowfn(lock, state, timeout, chwalk, ww_ctx);
20540 @@ -1435,17 +2040,20 @@ rt_mutex_fasttrylock(struct rt_mutex *lock,
20542 rt_mutex_fastunlock(struct rt_mutex *lock,
20543 bool (*slowfn)(struct rt_mutex *lock,
20544 - struct wake_q_head *wqh))
20545 + struct wake_q_head *wqh,
20546 + struct wake_q_head *wq_sleeper))
20549 + WAKE_Q(wake_sleeper_q);
20551 if (likely(rt_mutex_cmpxchg_release(lock, current, NULL))) {
20552 rt_mutex_deadlock_account_unlock(current);
20555 - bool deboost = slowfn(lock, &wake_q);
20556 + bool deboost = slowfn(lock, &wake_q, &wake_sleeper_q);
20558 wake_up_q(&wake_q);
20559 + wake_up_q_sleeper(&wake_sleeper_q);
20561 /* Undo pi boosting if necessary: */
20563 @@ -1462,7 +2070,7 @@ void __sched rt_mutex_lock(struct rt_mutex *lock)
20567 - rt_mutex_fastlock(lock, TASK_UNINTERRUPTIBLE, rt_mutex_slowlock);
20568 + rt_mutex_fastlock(lock, TASK_UNINTERRUPTIBLE, NULL, rt_mutex_slowlock);
20570 EXPORT_SYMBOL_GPL(rt_mutex_lock);
20572 @@ -1479,7 +2087,7 @@ int __sched rt_mutex_lock_interruptible(struct rt_mutex *lock)
20576 - return rt_mutex_fastlock(lock, TASK_INTERRUPTIBLE, rt_mutex_slowlock);
20577 + return rt_mutex_fastlock(lock, TASK_INTERRUPTIBLE, NULL, rt_mutex_slowlock);
20579 EXPORT_SYMBOL_GPL(rt_mutex_lock_interruptible);
20581 @@ -1492,11 +2100,30 @@ int rt_mutex_timed_futex_lock(struct rt_mutex *lock,
20584 return rt_mutex_timed_fastlock(lock, TASK_INTERRUPTIBLE, timeout,
20585 - RT_MUTEX_FULL_CHAINWALK,
20586 + RT_MUTEX_FULL_CHAINWALK, NULL,
20587 rt_mutex_slowlock);
20591 + * rt_mutex_lock_killable - lock a rt_mutex killable
20593 + * @lock: the rt_mutex to be locked
20594 + * @detect_deadlock: deadlock detection on/off
20598 + * -EINTR when interrupted by a signal
20599 + * -EDEADLK when the lock would deadlock (when deadlock detection is on)
20601 +int __sched rt_mutex_lock_killable(struct rt_mutex *lock)
20605 + return rt_mutex_fastlock(lock, TASK_KILLABLE, NULL, rt_mutex_slowlock);
20607 +EXPORT_SYMBOL_GPL(rt_mutex_lock_killable);
20610 * rt_mutex_timed_lock - lock a rt_mutex interruptible
20611 * the timeout structure is provided
20613 @@ -1516,6 +2143,7 @@ rt_mutex_timed_lock(struct rt_mutex *lock, struct hrtimer_sleeper *timeout)
20615 return rt_mutex_timed_fastlock(lock, TASK_INTERRUPTIBLE, timeout,
20616 RT_MUTEX_MIN_CHAINWALK,
20618 rt_mutex_slowlock);
20620 EXPORT_SYMBOL_GPL(rt_mutex_timed_lock);
20621 @@ -1533,7 +2161,11 @@ EXPORT_SYMBOL_GPL(rt_mutex_timed_lock);
20623 int __sched rt_mutex_trylock(struct rt_mutex *lock)
20625 +#ifdef CONFIG_PREEMPT_RT_FULL
20626 + if (WARN_ON_ONCE(in_irq() || in_nmi()))
20628 if (WARN_ON(in_irq() || in_nmi() || in_serving_softirq()))
20632 return rt_mutex_fasttrylock(lock, rt_mutex_slowtrylock);
20633 @@ -1559,13 +2191,14 @@ EXPORT_SYMBOL_GPL(rt_mutex_unlock);
20636 bool __sched rt_mutex_futex_unlock(struct rt_mutex *lock,
20637 - struct wake_q_head *wqh)
20638 + struct wake_q_head *wqh,
20639 + struct wake_q_head *wq_sleeper)
20641 if (likely(rt_mutex_cmpxchg_release(lock, current, NULL))) {
20642 rt_mutex_deadlock_account_unlock(current);
20645 - return rt_mutex_slowunlock(lock, wqh);
20646 + return rt_mutex_slowunlock(lock, wqh, wq_sleeper);
20650 @@ -1598,13 +2231,12 @@ EXPORT_SYMBOL_GPL(rt_mutex_destroy);
20651 void __rt_mutex_init(struct rt_mutex *lock, const char *name)
20653 lock->owner = NULL;
20654 - raw_spin_lock_init(&lock->wait_lock);
20655 lock->waiters = RB_ROOT;
20656 lock->waiters_leftmost = NULL;
20658 debug_rt_mutex_init(lock, name);
20660 -EXPORT_SYMBOL_GPL(__rt_mutex_init);
20661 +EXPORT_SYMBOL(__rt_mutex_init);
20664 * rt_mutex_init_proxy_locked - initialize and lock a rt_mutex on behalf of a
20665 @@ -1619,7 +2251,7 @@ EXPORT_SYMBOL_GPL(__rt_mutex_init);
20666 void rt_mutex_init_proxy_locked(struct rt_mutex *lock,
20667 struct task_struct *proxy_owner)
20669 - __rt_mutex_init(lock, NULL);
20670 + rt_mutex_init(lock);
20671 debug_rt_mutex_proxy_lock(lock, proxy_owner);
20672 rt_mutex_set_owner(lock, proxy_owner);
20673 rt_mutex_deadlock_account_lock(lock, proxy_owner);
20674 @@ -1660,13 +2292,42 @@ int rt_mutex_start_proxy_lock(struct rt_mutex *lock,
20678 - raw_spin_lock(&lock->wait_lock);
20679 + raw_spin_lock_irq(&lock->wait_lock);
20681 if (try_to_take_rt_mutex(lock, task, NULL)) {
20682 - raw_spin_unlock(&lock->wait_lock);
20683 + raw_spin_unlock_irq(&lock->wait_lock);
20687 +#ifdef CONFIG_PREEMPT_RT_FULL
20689 + * In PREEMPT_RT there's an added race.
20690 + * If the task, that we are about to requeue, times out,
20691 + * it can set the PI_WAKEUP_INPROGRESS. This tells the requeue
20692 + * to skip this task. But right after the task sets
20693 + * its pi_blocked_on to PI_WAKEUP_INPROGRESS it can then
20694 + * block on the spin_lock(&hb->lock), which in RT is an rtmutex.
20695 + * This will replace the PI_WAKEUP_INPROGRESS with the actual
20696 + * lock that it blocks on. We *must not* place this task
20697 + * on this proxy lock in that case.
20699 + * To prevent this race, we first take the task's pi_lock
20700 + * and check if it has updated its pi_blocked_on. If it has,
20701 + * we assume that it woke up and we return -EAGAIN.
20702 + * Otherwise, we set the task's pi_blocked_on to
20703 + * PI_REQUEUE_INPROGRESS, so that if the task is waking up
20704 + * it will know that we are in the process of requeuing it.
20706 + raw_spin_lock(&task->pi_lock);
20707 + if (task->pi_blocked_on) {
20708 + raw_spin_unlock(&task->pi_lock);
20709 + raw_spin_unlock_irq(&lock->wait_lock);
20712 + task->pi_blocked_on = PI_REQUEUE_INPROGRESS;
20713 + raw_spin_unlock(&task->pi_lock);
20716 /* We enforce deadlock detection for futexes */
20717 ret = task_blocks_on_rt_mutex(lock, waiter, task,
20718 RT_MUTEX_FULL_CHAINWALK);
20719 @@ -1681,10 +2342,10 @@ int rt_mutex_start_proxy_lock(struct rt_mutex *lock,
20723 - if (unlikely(ret))
20724 + if (ret && rt_mutex_has_waiters(lock))
20725 remove_waiter(lock, waiter);
20727 - raw_spin_unlock(&lock->wait_lock);
20728 + raw_spin_unlock_irq(&lock->wait_lock);
20730 debug_rt_mutex_print_deadlock(waiter);
20732 @@ -1732,12 +2393,12 @@ int rt_mutex_finish_proxy_lock(struct rt_mutex *lock,
20736 - raw_spin_lock(&lock->wait_lock);
20737 + raw_spin_lock_irq(&lock->wait_lock);
20739 set_current_state(TASK_INTERRUPTIBLE);
20741 /* sleep on the mutex */
20742 - ret = __rt_mutex_slowlock(lock, TASK_INTERRUPTIBLE, to, waiter);
20743 + ret = __rt_mutex_slowlock(lock, TASK_INTERRUPTIBLE, to, waiter, NULL);
20746 remove_waiter(lock, waiter);
20747 @@ -1748,7 +2409,93 @@ int rt_mutex_finish_proxy_lock(struct rt_mutex *lock,
20749 fixup_rt_mutex_waiters(lock);
20751 - raw_spin_unlock(&lock->wait_lock);
20752 + raw_spin_unlock_irq(&lock->wait_lock);
20758 +ww_mutex_deadlock_injection(struct ww_mutex *lock, struct ww_acquire_ctx *ctx)
20760 +#ifdef CONFIG_DEBUG_WW_MUTEX_SLOWPATH
20763 + if (ctx->deadlock_inject_countdown-- == 0) {
20764 + tmp = ctx->deadlock_inject_interval;
20765 + if (tmp > UINT_MAX/4)
20768 + tmp = tmp*2 + tmp + tmp/2;
20770 + ctx->deadlock_inject_interval = tmp;
20771 + ctx->deadlock_inject_countdown = tmp;
20772 + ctx->contending_lock = lock;
20774 + ww_mutex_unlock(lock);
20783 +#ifdef CONFIG_PREEMPT_RT_FULL
20785 +__ww_mutex_lock_interruptible(struct ww_mutex *lock, struct ww_acquire_ctx *ww_ctx)
20791 + mutex_acquire_nest(&lock->base.dep_map, 0, 0, &ww_ctx->dep_map, _RET_IP_);
20792 + ret = rt_mutex_slowlock(&lock->base.lock, TASK_INTERRUPTIBLE, NULL, 0, ww_ctx);
20794 + mutex_release(&lock->base.dep_map, 1, _RET_IP_);
20795 + else if (!ret && ww_ctx->acquired > 1)
20796 + return ww_mutex_deadlock_injection(lock, ww_ctx);
20800 +EXPORT_SYMBOL_GPL(__ww_mutex_lock_interruptible);
20803 +__ww_mutex_lock(struct ww_mutex *lock, struct ww_acquire_ctx *ww_ctx)
20809 + mutex_acquire_nest(&lock->base.dep_map, 0, 0, &ww_ctx->dep_map, _RET_IP_);
20810 + ret = rt_mutex_slowlock(&lock->base.lock, TASK_UNINTERRUPTIBLE, NULL, 0, ww_ctx);
20812 + mutex_release(&lock->base.dep_map, 1, _RET_IP_);
20813 + else if (!ret && ww_ctx->acquired > 1)
20814 + return ww_mutex_deadlock_injection(lock, ww_ctx);
20818 +EXPORT_SYMBOL_GPL(__ww_mutex_lock);
20820 +void __sched ww_mutex_unlock(struct ww_mutex *lock)
20822 + int nest = !!lock->ctx;
20825 + * The unlocking fastpath is the 0->1 transition from 'locked'
20826 + * into 'unlocked' state:
20829 +#ifdef CONFIG_DEBUG_MUTEXES
20830 + DEBUG_LOCKS_WARN_ON(!lock->ctx->acquired);
20832 + if (lock->ctx->acquired > 0)
20833 + lock->ctx->acquired--;
20834 + lock->ctx = NULL;
20837 + mutex_release(&lock->base.dep_map, nest, _RET_IP_);
20838 + rt_mutex_unlock(&lock->base.lock);
20840 +EXPORT_SYMBOL(ww_mutex_unlock);
20842 diff --git a/kernel/locking/rtmutex_common.h b/kernel/locking/rtmutex_common.h
20843 index e317e1cbb3eb..f457c7574920 100644
20844 --- a/kernel/locking/rtmutex_common.h
20845 +++ b/kernel/locking/rtmutex_common.h
20846 @@ -27,6 +27,7 @@ struct rt_mutex_waiter {
20847 struct rb_node pi_tree_entry;
20848 struct task_struct *task;
20849 struct rt_mutex *lock;
20851 #ifdef CONFIG_DEBUG_RT_MUTEXES
20853 struct pid *deadlock_task_pid;
20854 @@ -98,6 +99,9 @@ enum rtmutex_chainwalk {
20856 * PI-futex support (proxy locking functions, etc.):
20858 +#define PI_WAKEUP_INPROGRESS ((struct rt_mutex_waiter *) 1)
20859 +#define PI_REQUEUE_INPROGRESS ((struct rt_mutex_waiter *) 2)
20861 extern struct task_struct *rt_mutex_next_owner(struct rt_mutex *lock);
20862 extern void rt_mutex_init_proxy_locked(struct rt_mutex *lock,
20863 struct task_struct *proxy_owner);
20864 @@ -111,7 +115,8 @@ extern int rt_mutex_finish_proxy_lock(struct rt_mutex *lock,
20865 struct rt_mutex_waiter *waiter);
20866 extern int rt_mutex_timed_futex_lock(struct rt_mutex *l, struct hrtimer_sleeper *to);
20867 extern bool rt_mutex_futex_unlock(struct rt_mutex *lock,
20868 - struct wake_q_head *wqh);
20869 + struct wake_q_head *wqh,
20870 + struct wake_q_head *wq_sleeper);
20871 extern void rt_mutex_adjust_prio(struct task_struct *task);
20873 #ifdef CONFIG_DEBUG_RT_MUTEXES
20874 @@ -120,4 +125,14 @@ extern void rt_mutex_adjust_prio(struct task_struct *task);
20875 # include "rtmutex.h"
20878 +static inline void
20879 +rt_mutex_init_waiter(struct rt_mutex_waiter *waiter, bool savestate)
20881 + debug_rt_mutex_init_waiter(waiter);
20882 + waiter->task = NULL;
20883 + waiter->savestate = savestate;
20884 + RB_CLEAR_NODE(&waiter->pi_tree_entry);
20885 + RB_CLEAR_NODE(&waiter->tree_entry);
20889 diff --git a/kernel/locking/spinlock.c b/kernel/locking/spinlock.c
20890 index db3ccb1dd614..909779647bd1 100644
20891 --- a/kernel/locking/spinlock.c
20892 +++ b/kernel/locking/spinlock.c
20893 @@ -124,8 +124,11 @@ void __lockfunc __raw_##op##_lock_bh(locktype##_t *lock) \
20894 * __[spin|read|write]_lock_bh()
20896 BUILD_LOCK_OPS(spin, raw_spinlock);
20898 +#ifndef CONFIG_PREEMPT_RT_FULL
20899 BUILD_LOCK_OPS(read, rwlock);
20900 BUILD_LOCK_OPS(write, rwlock);
20905 @@ -209,6 +212,8 @@ void __lockfunc _raw_spin_unlock_bh(raw_spinlock_t *lock)
20906 EXPORT_SYMBOL(_raw_spin_unlock_bh);
20909 +#ifndef CONFIG_PREEMPT_RT_FULL
20911 #ifndef CONFIG_INLINE_READ_TRYLOCK
20912 int __lockfunc _raw_read_trylock(rwlock_t *lock)
20914 @@ -353,6 +358,8 @@ void __lockfunc _raw_write_unlock_bh(rwlock_t *lock)
20915 EXPORT_SYMBOL(_raw_write_unlock_bh);
20918 +#endif /* !PREEMPT_RT_FULL */
20920 #ifdef CONFIG_DEBUG_LOCK_ALLOC
20922 void __lockfunc _raw_spin_lock_nested(raw_spinlock_t *lock, int subclass)
20923 diff --git a/kernel/locking/spinlock_debug.c b/kernel/locking/spinlock_debug.c
20924 index 0374a596cffa..94970338d518 100644
20925 --- a/kernel/locking/spinlock_debug.c
20926 +++ b/kernel/locking/spinlock_debug.c
20927 @@ -31,6 +31,7 @@ void __raw_spin_lock_init(raw_spinlock_t *lock, const char *name,
20929 EXPORT_SYMBOL(__raw_spin_lock_init);
20931 +#ifndef CONFIG_PREEMPT_RT_FULL
20932 void __rwlock_init(rwlock_t *lock, const char *name,
20933 struct lock_class_key *key)
20935 @@ -48,6 +49,7 @@ void __rwlock_init(rwlock_t *lock, const char *name,
20938 EXPORT_SYMBOL(__rwlock_init);
20941 static void spin_dump(raw_spinlock_t *lock, const char *msg)
20943 @@ -159,6 +161,7 @@ void do_raw_spin_unlock(raw_spinlock_t *lock)
20944 arch_spin_unlock(&lock->raw_lock);
20947 +#ifndef CONFIG_PREEMPT_RT_FULL
20948 static void rwlock_bug(rwlock_t *lock, const char *msg)
20950 if (!debug_locks_off())
20951 @@ -300,3 +303,5 @@ void do_raw_write_unlock(rwlock_t *lock)
20952 debug_write_unlock(lock);
20953 arch_write_unlock(&lock->raw_lock);
20957 diff --git a/kernel/module.c b/kernel/module.c
20958 index b14a4f31221f..be3f497a089d 100644
20959 --- a/kernel/module.c
20960 +++ b/kernel/module.c
20961 @@ -682,16 +682,7 @@ static void percpu_modcopy(struct module *mod,
20962 memcpy(per_cpu_ptr(mod->percpu, cpu), from, size);
20966 - * is_module_percpu_address - test whether address is from module static percpu
20967 - * @addr: address to test
20969 - * Test whether @addr belongs to module static percpu area.
20972 - * %true if @addr is from module static percpu area
20974 -bool is_module_percpu_address(unsigned long addr)
20975 +bool __is_module_percpu_address(unsigned long addr, unsigned long *can_addr)
20977 struct module *mod;
20979 @@ -705,9 +696,11 @@ bool is_module_percpu_address(unsigned long addr)
20981 for_each_possible_cpu(cpu) {
20982 void *start = per_cpu_ptr(mod->percpu, cpu);
20983 + void *va = (void *)addr;
20985 - if ((void *)addr >= start &&
20986 - (void *)addr < start + mod->percpu_size) {
20987 + if (va >= start && va < start + mod->percpu_size) {
20989 + *can_addr = (unsigned long) (va - start);
20993 @@ -718,6 +711,20 @@ bool is_module_percpu_address(unsigned long addr)
20998 + * is_module_percpu_address - test whether address is from module static percpu
20999 + * @addr: address to test
21001 + * Test whether @addr belongs to module static percpu area.
21004 + * %true if @addr is from module static percpu area
21006 +bool is_module_percpu_address(unsigned long addr)
21008 + return __is_module_percpu_address(addr, NULL);
21011 #else /* ... !CONFIG_SMP */
21013 static inline void __percpu *mod_percpu(struct module *mod)
21014 @@ -749,6 +756,11 @@ bool is_module_percpu_address(unsigned long addr)
21018 +bool __is_module_percpu_address(unsigned long addr, unsigned long *can_addr)
21023 #endif /* CONFIG_SMP */
21025 #define MODINFO_ATTR(field) \
21026 diff --git a/kernel/panic.c b/kernel/panic.c
21027 index 41e2b54f36b5..3535f802953a 100644
21028 --- a/kernel/panic.c
21029 +++ b/kernel/panic.c
21030 @@ -61,6 +61,37 @@ void __weak panic_smp_self_stop(void)
21035 + * Stop ourselves in NMI context if another CPU has already panicked. Arch code
21036 + * may override this to prepare for crash dumping, e.g. save regs info.
21038 +void __weak nmi_panic_self_stop(struct pt_regs *regs)
21040 + panic_smp_self_stop();
21043 +atomic_t panic_cpu = ATOMIC_INIT(PANIC_CPU_INVALID);
21046 + * A variant of panic() called from NMI context. We return if we've already
21047 + * panicked on this CPU. If another CPU already panicked, loop in
21048 + * nmi_panic_self_stop() which can provide architecture dependent code such
21049 + * as saving register state for crash dump.
21051 +void nmi_panic(struct pt_regs *regs, const char *msg)
21053 + int old_cpu, cpu;
21055 + cpu = raw_smp_processor_id();
21056 + old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, cpu);
21058 + if (old_cpu == PANIC_CPU_INVALID)
21059 + panic("%s", msg);
21060 + else if (old_cpu != cpu)
21061 + nmi_panic_self_stop(regs);
21063 +EXPORT_SYMBOL(nmi_panic);
21066 * panic - halt the system
21067 * @fmt: The text string to print
21068 @@ -71,17 +102,17 @@ void __weak panic_smp_self_stop(void)
21070 void panic(const char *fmt, ...)
21072 - static DEFINE_SPINLOCK(panic_lock);
21073 static char buf[1024];
21075 long i, i_next = 0;
21077 + int old_cpu, this_cpu;
21080 * Disable local interrupts. This will prevent panic_smp_self_stop
21081 * from deadlocking the first cpu that invokes the panic, since
21082 * there is nothing to prevent an interrupt handler (that runs
21083 - * after the panic_lock is acquired) from invoking panic again.
21084 + * after setting panic_cpu) from invoking panic() again.
21086 local_irq_disable();
21088 @@ -94,8 +125,16 @@ void panic(const char *fmt, ...)
21089 * multiple parallel invocations of panic, all other CPUs either
21090 * stop themself or will wait until they are stopped by the 1st CPU
21091 * with smp_send_stop().
21093 + * `old_cpu == PANIC_CPU_INVALID' means this is the 1st CPU which
21094 + * comes here, so go ahead.
21095 + * `old_cpu == this_cpu' means we came from nmi_panic() which sets
21096 + * panic_cpu to this CPU. In this case, this is also the 1st CPU.
21098 - if (!spin_trylock(&panic_lock))
21099 + this_cpu = raw_smp_processor_id();
21100 + old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu);
21102 + if (old_cpu != PANIC_CPU_INVALID && old_cpu != this_cpu)
21103 panic_smp_self_stop();
21106 @@ -400,9 +439,11 @@ static u64 oops_id;
21108 static int init_oops_id(void)
21110 +#ifndef CONFIG_PREEMPT_RT_FULL
21112 get_random_bytes(&oops_id, sizeof(oops_id));
21118 diff --git a/kernel/power/hibernate.c b/kernel/power/hibernate.c
21119 index 3124cebaec31..c1b981521dd0 100644
21120 --- a/kernel/power/hibernate.c
21121 +++ b/kernel/power/hibernate.c
21122 @@ -285,6 +285,8 @@ static int create_image(int platform_mode)
21124 local_irq_disable();
21126 + system_state = SYSTEM_SUSPEND;
21128 error = syscore_suspend();
21130 printk(KERN_ERR "PM: Some system devices failed to power down, "
21131 @@ -314,6 +316,7 @@ static int create_image(int platform_mode)
21135 + system_state = SYSTEM_RUNNING;
21136 local_irq_enable();
21139 @@ -438,6 +441,7 @@ static int resume_target_kernel(bool platform_mode)
21142 local_irq_disable();
21143 + system_state = SYSTEM_SUSPEND;
21145 error = syscore_suspend();
21147 @@ -471,6 +475,7 @@ static int resume_target_kernel(bool platform_mode)
21151 + system_state = SYSTEM_RUNNING;
21152 local_irq_enable();
21155 @@ -556,6 +561,7 @@ int hibernation_platform_enter(void)
21158 local_irq_disable();
21159 + system_state = SYSTEM_SUSPEND;
21161 if (pm_wakeup_pending()) {
21163 @@ -568,6 +574,7 @@ int hibernation_platform_enter(void)
21167 + system_state = SYSTEM_RUNNING;
21168 local_irq_enable();
21171 @@ -642,6 +649,10 @@ static void power_down(void)
21175 +#ifndef CONFIG_SUSPEND
21176 +bool pm_in_action;
21180 * hibernate - Carry out system hibernation, including saving the image.
21182 @@ -654,6 +665,8 @@ int hibernate(void)
21186 + pm_in_action = true;
21188 lock_system_sleep();
21189 /* The snapshot device should not be opened while we're running */
21190 if (!atomic_add_unless(&snapshot_device_available, -1, 0)) {
21191 @@ -719,6 +732,7 @@ int hibernate(void)
21192 atomic_inc(&snapshot_device_available);
21194 unlock_system_sleep();
21195 + pm_in_action = false;
21199 diff --git a/kernel/power/suspend.c b/kernel/power/suspend.c
21200 index f9fe133c13e2..393bc342c586 100644
21201 --- a/kernel/power/suspend.c
21202 +++ b/kernel/power/suspend.c
21203 @@ -359,6 +359,8 @@ static int suspend_enter(suspend_state_t state, bool *wakeup)
21204 arch_suspend_disable_irqs();
21205 BUG_ON(!irqs_disabled());
21207 + system_state = SYSTEM_SUSPEND;
21209 error = syscore_suspend();
21211 *wakeup = pm_wakeup_pending();
21212 @@ -375,6 +377,8 @@ static int suspend_enter(suspend_state_t state, bool *wakeup)
21216 + system_state = SYSTEM_RUNNING;
21218 arch_suspend_enable_irqs();
21219 BUG_ON(irqs_disabled());
21221 @@ -518,6 +522,8 @@ static int enter_state(suspend_state_t state)
21225 +bool pm_in_action;
21228 * pm_suspend - Externally visible function for suspending the system.
21229 * @state: System sleep state to enter.
21230 @@ -532,6 +538,8 @@ int pm_suspend(suspend_state_t state)
21231 if (state <= PM_SUSPEND_ON || state >= PM_SUSPEND_MAX)
21234 + pm_in_action = true;
21236 error = enter_state(state);
21238 suspend_stats.fail++;
21239 @@ -539,6 +547,7 @@ int pm_suspend(suspend_state_t state)
21241 suspend_stats.success++;
21243 + pm_in_action = false;
21246 EXPORT_SYMBOL(pm_suspend);
21247 diff --git a/kernel/printk/printk.c b/kernel/printk/printk.c
21248 index 0b5613554769..99deb2617308 100644
21249 --- a/kernel/printk/printk.c
21250 +++ b/kernel/printk/printk.c
21251 @@ -241,6 +241,65 @@ struct printk_log {
21253 static DEFINE_RAW_SPINLOCK(logbuf_lock);
21255 +#ifdef CONFIG_EARLY_PRINTK
21256 +struct console *early_console;
21258 +static void early_vprintk(const char *fmt, va_list ap)
21260 + if (early_console) {
21262 + int n = vscnprintf(buf, sizeof(buf), fmt, ap);
21264 + early_console->write(early_console, buf, n);
21268 +asmlinkage void early_printk(const char *fmt, ...)
21272 + va_start(ap, fmt);
21273 + early_vprintk(fmt, ap);
21278 + * This is independent of any log levels - a global
21279 + * kill switch that turns off all of printk.
21281 + * Used by the NMI watchdog if early-printk is enabled.
21283 +static bool __read_mostly printk_killswitch;
21285 +static int __init force_early_printk_setup(char *str)
21287 + printk_killswitch = true;
21290 +early_param("force_early_printk", force_early_printk_setup);
21292 +void printk_kill(void)
21294 + printk_killswitch = true;
21297 +#ifdef CONFIG_PRINTK
21298 +static int forced_early_printk(const char *fmt, va_list ap)
21300 + if (!printk_killswitch)
21302 + early_vprintk(fmt, ap);
21308 +static inline int forced_early_printk(const char *fmt, va_list ap)
21314 #ifdef CONFIG_PRINTK
21315 DECLARE_WAIT_QUEUE_HEAD(log_wait);
21316 /* the next printk record to read by syslog(READ) or /proc/kmsg */
21317 @@ -1203,6 +1262,7 @@ static int syslog_print_all(char __user *buf, int size, bool clear)
21321 + int attempts = 0;
21323 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
21325 @@ -1214,7 +1274,14 @@ static int syslog_print_all(char __user *buf, int size, bool clear)
21328 enum log_flags prev;
21333 + if (attempts > 10) {
21338 if (clear_seq < log_first_seq) {
21339 /* messages are gone, move to first available one */
21340 clear_seq = log_first_seq;
21341 @@ -1235,6 +1302,14 @@ static int syslog_print_all(char __user *buf, int size, bool clear)
21343 idx = log_next(idx);
21346 + if (num_msg > 5) {
21348 + raw_spin_unlock_irq(&logbuf_lock);
21349 + raw_spin_lock_irq(&logbuf_lock);
21350 + if (clear_seq < log_first_seq)
21355 /* move first record forward until length fits into the buffer */
21356 @@ -1248,6 +1323,14 @@ static int syslog_print_all(char __user *buf, int size, bool clear)
21358 idx = log_next(idx);
21361 + if (num_msg > 5) {
21363 + raw_spin_unlock_irq(&logbuf_lock);
21364 + raw_spin_lock_irq(&logbuf_lock);
21365 + if (clear_seq < log_first_seq)
21370 /* last message fitting into this dump */
21371 @@ -1288,6 +1371,7 @@ static int syslog_print_all(char __user *buf, int size, bool clear)
21372 clear_seq = log_next_seq;
21373 clear_idx = log_next_idx;
21376 raw_spin_unlock_irq(&logbuf_lock);
21379 @@ -1443,6 +1527,12 @@ static void call_console_drivers(int level,
21380 if (!console_drivers)
21383 + if (IS_ENABLED(CONFIG_PREEMPT_RT_BASE)) {
21384 + if (in_irq() || in_nmi())
21388 + migrate_disable();
21389 for_each_console(con) {
21390 if (exclusive_console && con != exclusive_console)
21392 @@ -1458,6 +1548,7 @@ static void call_console_drivers(int level,
21394 con->write(con, text, len);
21396 + migrate_enable();
21400 @@ -1518,6 +1609,15 @@ static inline int can_use_console(unsigned int cpu)
21401 static int console_trylock_for_printk(void)
21403 unsigned int cpu = smp_processor_id();
21404 +#ifdef CONFIG_PREEMPT_RT_FULL
21405 + int lock = !early_boot_irqs_disabled && (preempt_count() == 0) &&
21406 + !irqs_disabled();
21414 if (!console_trylock())
21416 @@ -1672,6 +1772,13 @@ asmlinkage int vprintk_emit(int facility, int level,
21417 /* cpu currently holding logbuf_lock in this function */
21418 static unsigned int logbuf_cpu = UINT_MAX;
21421 + * Fall back to early_printk if a debugging subsystem has
21422 + * killed printk output
21424 + if (unlikely(forced_early_printk(fmt, args)))
21427 if (level == LOGLEVEL_SCHED) {
21428 level = LOGLEVEL_DEFAULT;
21430 @@ -1813,8 +1920,7 @@ asmlinkage int vprintk_emit(int facility, int level,
21431 * console_sem which would prevent anyone from printing to
21434 - preempt_disable();
21436 + migrate_disable();
21438 * Try to acquire and then immediately release the console
21439 * semaphore. The release will print out buffers and wake up
21440 @@ -1822,7 +1928,7 @@ asmlinkage int vprintk_emit(int facility, int level,
21442 if (console_trylock_for_printk())
21444 - preempt_enable();
21445 + migrate_enable();
21449 @@ -1961,26 +2067,6 @@ DEFINE_PER_CPU(printk_func_t, printk_func);
21451 #endif /* CONFIG_PRINTK */
21453 -#ifdef CONFIG_EARLY_PRINTK
21454 -struct console *early_console;
21456 -asmlinkage __visible void early_printk(const char *fmt, ...)
21462 - if (!early_console)
21465 - va_start(ap, fmt);
21466 - n = vscnprintf(buf, sizeof(buf), fmt, ap);
21469 - early_console->write(early_console, buf, n);
21473 static int __add_preferred_console(char *name, int idx, char *options,
21476 @@ -2202,11 +2288,16 @@ static void console_cont_flush(char *text, size_t size)
21479 len = cont_print_text(text, size);
21480 +#ifdef CONFIG_PREEMPT_RT_FULL
21481 + raw_spin_unlock_irqrestore(&logbuf_lock, flags);
21482 + call_console_drivers(cont.level, NULL, 0, text, len);
21484 raw_spin_unlock(&logbuf_lock);
21485 stop_critical_timings();
21486 call_console_drivers(cont.level, NULL, 0, text, len);
21487 start_critical_timings();
21488 local_irq_restore(flags);
21492 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
21493 @@ -2316,13 +2407,17 @@ skip:
21494 console_idx = log_next(console_idx);
21496 console_prev = msg->flags;
21497 +#ifdef CONFIG_PREEMPT_RT_FULL
21498 + raw_spin_unlock_irqrestore(&logbuf_lock, flags);
21499 + call_console_drivers(level, ext_text, ext_len, text, len);
21501 raw_spin_unlock(&logbuf_lock);
21503 stop_critical_timings(); /* don't trace print latency */
21504 call_console_drivers(level, ext_text, ext_len, text, len);
21505 start_critical_timings();
21506 local_irq_restore(flags);
21509 if (do_cond_resched)
21512 @@ -2374,6 +2469,11 @@ void console_unblank(void)
21516 + if (IS_ENABLED(CONFIG_PREEMPT_RT_BASE)) {
21517 + if (in_irq() || in_nmi())
21522 * console_unblank can no longer be called in interrupt context unless
21523 * oops_in_progress is set to 1..
21524 diff --git a/kernel/ptrace.c b/kernel/ptrace.c
21525 index c7e8ed99c953..2a19cc5001b4 100644
21526 --- a/kernel/ptrace.c
21527 +++ b/kernel/ptrace.c
21528 @@ -136,7 +136,14 @@ static bool ptrace_freeze_traced(struct task_struct *task)
21530 spin_lock_irq(&task->sighand->siglock);
21531 if (task_is_traced(task) && !__fatal_signal_pending(task)) {
21532 - task->state = __TASK_TRACED;
21533 + unsigned long flags;
21535 + raw_spin_lock_irqsave(&task->pi_lock, flags);
21536 + if (task->state & __TASK_TRACED)
21537 + task->state = __TASK_TRACED;
21539 + task->saved_state = __TASK_TRACED;
21540 + raw_spin_unlock_irqrestore(&task->pi_lock, flags);
21543 spin_unlock_irq(&task->sighand->siglock);
21544 diff --git a/kernel/rcu/rcutorture.c b/kernel/rcu/rcutorture.c
21545 index d89328e260df..5bb3364a6284 100644
21546 --- a/kernel/rcu/rcutorture.c
21547 +++ b/kernel/rcu/rcutorture.c
21548 @@ -390,6 +390,7 @@ static struct rcu_torture_ops rcu_ops = {
21552 +#ifndef CONFIG_PREEMPT_RT_FULL
21554 * Definitions for rcu_bh torture testing.
21556 @@ -429,6 +430,12 @@ static struct rcu_torture_ops rcu_bh_ops = {
21561 +static struct rcu_torture_ops rcu_bh_ops = {
21562 + .ttype = INVALID_RCU_FLAVOR,
21567 * Don't even think about trying any of these in real life!!!
21568 * The names includes "busted", and they really means it!
21569 diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
21570 index f07343b54fe5..d862a203fce0 100644
21571 --- a/kernel/rcu/tree.c
21572 +++ b/kernel/rcu/tree.c
21574 #include <linux/random.h>
21575 #include <linux/trace_events.h>
21576 #include <linux/suspend.h>
21577 +#include <linux/delay.h>
21578 +#include <linux/gfp.h>
21579 +#include <linux/oom.h>
21580 +#include <linux/smpboot.h>
21581 +#include "../time/tick-internal.h"
21585 @@ -266,6 +271,19 @@ void rcu_sched_qs(void)
21589 +#ifdef CONFIG_PREEMPT_RT_FULL
21590 +static void rcu_preempt_qs(void);
21592 +void rcu_bh_qs(void)
21594 + unsigned long flags;
21596 + /* Callers to this function, rcu_preempt_qs(), must disable irqs. */
21597 + local_irq_save(flags);
21598 + rcu_preempt_qs();
21599 + local_irq_restore(flags);
21602 void rcu_bh_qs(void)
21604 if (__this_cpu_read(rcu_bh_data.cpu_no_qs.s)) {
21605 @@ -275,6 +293,7 @@ void rcu_bh_qs(void)
21606 __this_cpu_write(rcu_bh_data.cpu_no_qs.b.norm, false);
21611 static DEFINE_PER_CPU(int, rcu_sched_qs_mask);
21613 @@ -435,11 +454,13 @@ EXPORT_SYMBOL_GPL(rcu_batches_started_sched);
21615 * Return the number of RCU BH batches started thus far for debug & stats.
21617 +#ifndef CONFIG_PREEMPT_RT_FULL
21618 unsigned long rcu_batches_started_bh(void)
21620 return rcu_bh_state.gpnum;
21622 EXPORT_SYMBOL_GPL(rcu_batches_started_bh);
21626 * Return the number of RCU batches completed thus far for debug & stats.
21627 @@ -459,6 +480,7 @@ unsigned long rcu_batches_completed_sched(void)
21629 EXPORT_SYMBOL_GPL(rcu_batches_completed_sched);
21631 +#ifndef CONFIG_PREEMPT_RT_FULL
21633 * Return the number of RCU BH batches completed thus far for debug & stats.
21635 @@ -486,6 +508,13 @@ void rcu_bh_force_quiescent_state(void)
21637 EXPORT_SYMBOL_GPL(rcu_bh_force_quiescent_state);
21640 +void rcu_force_quiescent_state(void)
21643 +EXPORT_SYMBOL_GPL(rcu_force_quiescent_state);
21647 * Force a quiescent state for RCU-sched.
21649 @@ -536,9 +565,11 @@ void rcutorture_get_gp_data(enum rcutorture_type test_type, int *flags,
21653 +#ifndef CONFIG_PREEMPT_RT_FULL
21654 case RCU_BH_FLAVOR:
21655 rsp = &rcu_bh_state;
21658 case RCU_SCHED_FLAVOR:
21659 rsp = &rcu_sched_state;
21661 @@ -1590,7 +1621,6 @@ static int rcu_future_gp_cleanup(struct rcu_state *rsp, struct rcu_node *rnp)
21663 struct rcu_data *rdp = this_cpu_ptr(rsp->rda);
21665 - rcu_nocb_gp_cleanup(rsp, rnp);
21666 rnp->need_future_gp[c & 0x1] = 0;
21667 needmore = rnp->need_future_gp[(c + 1) & 0x1];
21668 trace_rcu_future_gp(rnp, rdp, c,
21669 @@ -1611,7 +1641,7 @@ static void rcu_gp_kthread_wake(struct rcu_state *rsp)
21670 !READ_ONCE(rsp->gp_flags) ||
21673 - wake_up(&rsp->gp_wq);
21674 + swake_up(&rsp->gp_wq);
21678 @@ -1991,6 +2021,7 @@ static void rcu_gp_cleanup(struct rcu_state *rsp)
21680 struct rcu_data *rdp;
21681 struct rcu_node *rnp = rcu_get_root(rsp);
21682 + struct swait_queue_head *sq;
21684 WRITE_ONCE(rsp->gp_activity, jiffies);
21685 raw_spin_lock_irq(&rnp->lock);
21686 @@ -2029,7 +2060,9 @@ static void rcu_gp_cleanup(struct rcu_state *rsp)
21687 needgp = __note_gp_changes(rsp, rnp, rdp) || needgp;
21688 /* smp_mb() provided by prior unlock-lock pair. */
21689 nocb += rcu_future_gp_cleanup(rsp, rnp);
21690 + sq = rcu_nocb_gp_get(rnp);
21691 raw_spin_unlock_irq(&rnp->lock);
21692 + rcu_nocb_gp_cleanup(sq);
21693 cond_resched_rcu_qs();
21694 WRITE_ONCE(rsp->gp_activity, jiffies);
21695 rcu_gp_slow(rsp, gp_cleanup_delay);
21696 @@ -2076,7 +2109,7 @@ static int __noreturn rcu_gp_kthread(void *arg)
21697 READ_ONCE(rsp->gpnum),
21699 rsp->gp_state = RCU_GP_WAIT_GPS;
21700 - wait_event_interruptible(rsp->gp_wq,
21701 + swait_event_interruptible(rsp->gp_wq,
21702 READ_ONCE(rsp->gp_flags) &
21704 rsp->gp_state = RCU_GP_DONE_GPS;
21705 @@ -2106,7 +2139,7 @@ static int __noreturn rcu_gp_kthread(void *arg)
21706 READ_ONCE(rsp->gpnum),
21708 rsp->gp_state = RCU_GP_WAIT_FQS;
21709 - ret = wait_event_interruptible_timeout(rsp->gp_wq,
21710 + ret = swait_event_interruptible_timeout(rsp->gp_wq,
21711 rcu_gp_fqs_check_wake(rsp, &gf), j);
21712 rsp->gp_state = RCU_GP_DOING_FQS;
21713 /* Locking provides needed memory barriers. */
21714 @@ -2230,7 +2263,7 @@ static void rcu_report_qs_rsp(struct rcu_state *rsp, unsigned long flags)
21715 WARN_ON_ONCE(!rcu_gp_in_progress(rsp));
21716 WRITE_ONCE(rsp->gp_flags, READ_ONCE(rsp->gp_flags) | RCU_GP_FLAG_FQS);
21717 raw_spin_unlock_irqrestore(&rcu_get_root(rsp)->lock, flags);
21718 - rcu_gp_kthread_wake(rsp);
21719 + swake_up(&rsp->gp_wq); /* Memory barrier implied by swake_up() path. */
21723 @@ -2891,7 +2924,7 @@ static void force_quiescent_state(struct rcu_state *rsp)
21725 WRITE_ONCE(rsp->gp_flags, READ_ONCE(rsp->gp_flags) | RCU_GP_FLAG_FQS);
21726 raw_spin_unlock_irqrestore(&rnp_old->lock, flags);
21727 - rcu_gp_kthread_wake(rsp);
21728 + swake_up(&rsp->gp_wq); /* Memory barrier implied by swake_up() path. */
21732 @@ -2934,18 +2967,17 @@ __rcu_process_callbacks(struct rcu_state *rsp)
21734 * Do RCU core processing for the current CPU.
21736 -static void rcu_process_callbacks(struct softirq_action *unused)
21737 +static void rcu_process_callbacks(void)
21739 struct rcu_state *rsp;
21741 if (cpu_is_offline(smp_processor_id()))
21743 - trace_rcu_utilization(TPS("Start RCU core"));
21744 for_each_rcu_flavor(rsp)
21745 __rcu_process_callbacks(rsp);
21746 - trace_rcu_utilization(TPS("End RCU core"));
21749 +static DEFINE_PER_CPU(struct task_struct *, rcu_cpu_kthread_task);
21751 * Schedule RCU callback invocation. If the specified type of RCU
21752 * does not support RCU priority boosting, just do a direct call,
21753 @@ -2957,18 +2989,105 @@ static void invoke_rcu_callbacks(struct rcu_state *rsp, struct rcu_data *rdp)
21755 if (unlikely(!READ_ONCE(rcu_scheduler_fully_active)))
21757 - if (likely(!rsp->boost)) {
21758 - rcu_do_batch(rsp, rdp);
21761 - invoke_rcu_callbacks_kthread();
21762 + rcu_do_batch(rsp, rdp);
21765 +static void rcu_wake_cond(struct task_struct *t, int status)
21768 + * If the thread is yielding, only wake it when this
21769 + * is invoked from idle
21771 + if (t && (status != RCU_KTHREAD_YIELDING || is_idle_task(current)))
21772 + wake_up_process(t);
21776 + * Wake up this CPU's rcuc kthread to do RCU core processing.
21778 static void invoke_rcu_core(void)
21780 - if (cpu_online(smp_processor_id()))
21781 - raise_softirq(RCU_SOFTIRQ);
21782 + unsigned long flags;
21783 + struct task_struct *t;
21785 + if (!cpu_online(smp_processor_id()))
21787 + local_irq_save(flags);
21788 + __this_cpu_write(rcu_cpu_has_work, 1);
21789 + t = __this_cpu_read(rcu_cpu_kthread_task);
21790 + if (t != NULL && current != t)
21791 + rcu_wake_cond(t, __this_cpu_read(rcu_cpu_kthread_status));
21792 + local_irq_restore(flags);
21795 +static void rcu_cpu_kthread_park(unsigned int cpu)
21797 + per_cpu(rcu_cpu_kthread_status, cpu) = RCU_KTHREAD_OFFCPU;
21800 +static int rcu_cpu_kthread_should_run(unsigned int cpu)
21802 + return __this_cpu_read(rcu_cpu_has_work);
21806 + * Per-CPU kernel thread that invokes RCU callbacks. This replaces the
21807 + * RCU softirq used in flavors and configurations of RCU that do not
21808 + * support RCU priority boosting.
21810 +static void rcu_cpu_kthread(unsigned int cpu)
21812 + unsigned int *statusp = this_cpu_ptr(&rcu_cpu_kthread_status);
21813 + char work, *workp = this_cpu_ptr(&rcu_cpu_has_work);
21816 + for (spincnt = 0; spincnt < 10; spincnt++) {
21817 + trace_rcu_utilization(TPS("Start CPU kthread@rcu_wait"));
21818 + local_bh_disable();
21819 + *statusp = RCU_KTHREAD_RUNNING;
21820 + this_cpu_inc(rcu_cpu_kthread_loops);
21821 + local_irq_disable();
21824 + local_irq_enable();
21826 + rcu_process_callbacks();
21827 + local_bh_enable();
21828 + if (*workp == 0) {
21829 + trace_rcu_utilization(TPS("End CPU kthread@rcu_wait"));
21830 + *statusp = RCU_KTHREAD_WAITING;
21834 + *statusp = RCU_KTHREAD_YIELDING;
21835 + trace_rcu_utilization(TPS("Start CPU kthread@rcu_yield"));
21836 + schedule_timeout_interruptible(2);
21837 + trace_rcu_utilization(TPS("End CPU kthread@rcu_yield"));
21838 + *statusp = RCU_KTHREAD_WAITING;
21841 +static struct smp_hotplug_thread rcu_cpu_thread_spec = {
21842 + .store = &rcu_cpu_kthread_task,
21843 + .thread_should_run = rcu_cpu_kthread_should_run,
21844 + .thread_fn = rcu_cpu_kthread,
21845 + .thread_comm = "rcuc/%u",
21846 + .setup = rcu_cpu_kthread_setup,
21847 + .park = rcu_cpu_kthread_park,
21851 + * Spawn per-CPU RCU core processing kthreads.
21853 +static int __init rcu_spawn_core_kthreads(void)
21857 + for_each_possible_cpu(cpu)
21858 + per_cpu(rcu_cpu_has_work, cpu) = 0;
21859 + BUG_ON(smpboot_register_percpu_thread(&rcu_cpu_thread_spec));
21862 +early_initcall(rcu_spawn_core_kthreads);
21865 * Handle any core-RCU processing required by a call_rcu() invocation.
21866 @@ -3114,6 +3233,7 @@ void call_rcu_sched(struct rcu_head *head, rcu_callback_t func)
21868 EXPORT_SYMBOL_GPL(call_rcu_sched);
21870 +#ifndef CONFIG_PREEMPT_RT_FULL
21872 * Queue an RCU callback for invocation after a quicker grace period.
21874 @@ -3122,6 +3242,7 @@ void call_rcu_bh(struct rcu_head *head, rcu_callback_t func)
21875 __call_rcu(head, func, &rcu_bh_state, -1, 0);
21877 EXPORT_SYMBOL_GPL(call_rcu_bh);
21881 * Queue an RCU callback for lazy invocation after a grace period.
21882 @@ -3213,6 +3334,7 @@ void synchronize_sched(void)
21884 EXPORT_SYMBOL_GPL(synchronize_sched);
21886 +#ifndef CONFIG_PREEMPT_RT_FULL
21888 * synchronize_rcu_bh - wait until an rcu_bh grace period has elapsed.
21890 @@ -3239,6 +3361,7 @@ void synchronize_rcu_bh(void)
21891 wait_rcu_gp(call_rcu_bh);
21893 EXPORT_SYMBOL_GPL(synchronize_rcu_bh);
21897 * get_state_synchronize_rcu - Snapshot current RCU state
21898 @@ -3524,7 +3647,7 @@ static void __rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp,
21899 raw_spin_unlock_irqrestore(&rnp->lock, flags);
21901 smp_mb(); /* EGP done before wake_up(). */
21902 - wake_up(&rsp->expedited_wq);
21903 + swake_up(&rsp->expedited_wq);
21907 @@ -3781,7 +3904,7 @@ static void synchronize_sched_expedited_wait(struct rcu_state *rsp)
21908 jiffies_start = jiffies;
21911 - ret = wait_event_interruptible_timeout(
21912 + ret = swait_event_timeout(
21914 sync_rcu_preempt_exp_done(rnp_root),
21916 @@ -3789,7 +3912,7 @@ static void synchronize_sched_expedited_wait(struct rcu_state *rsp)
21919 /* Hit a signal, disable CPU stall warnings. */
21920 - wait_event(rsp->expedited_wq,
21921 + swait_event(rsp->expedited_wq,
21922 sync_rcu_preempt_exp_done(rnp_root));
21925 @@ -4101,6 +4224,7 @@ static void _rcu_barrier(struct rcu_state *rsp)
21926 mutex_unlock(&rsp->barrier_mutex);
21929 +#ifndef CONFIG_PREEMPT_RT_FULL
21931 * rcu_barrier_bh - Wait until all in-flight call_rcu_bh() callbacks complete.
21933 @@ -4109,6 +4233,7 @@ void rcu_barrier_bh(void)
21934 _rcu_barrier(&rcu_bh_state);
21936 EXPORT_SYMBOL_GPL(rcu_barrier_bh);
21940 * rcu_barrier_sched - Wait for in-flight call_rcu_sched() callbacks.
21941 @@ -4455,8 +4580,8 @@ static void __init rcu_init_one(struct rcu_state *rsp,
21945 - init_waitqueue_head(&rsp->gp_wq);
21946 - init_waitqueue_head(&rsp->expedited_wq);
21947 + init_swait_queue_head(&rsp->gp_wq);
21948 + init_swait_queue_head(&rsp->expedited_wq);
21949 rnp = rsp->level[rcu_num_lvls - 1];
21950 for_each_possible_cpu(i) {
21951 while (i > rnp->grphi)
21952 @@ -4576,12 +4701,13 @@ void __init rcu_init(void)
21954 rcu_bootup_announce();
21955 rcu_init_geometry();
21956 +#ifndef CONFIG_PREEMPT_RT_FULL
21957 rcu_init_one(&rcu_bh_state, &rcu_bh_data);
21959 rcu_init_one(&rcu_sched_state, &rcu_sched_data);
21961 rcu_dump_rcu_node_tree(&rcu_sched_state);
21962 __rcu_init_preempt();
21963 - open_softirq(RCU_SOFTIRQ, rcu_process_callbacks);
21966 * We don't need protection against CPU-hotplug here because
21967 diff --git a/kernel/rcu/tree.h b/kernel/rcu/tree.h
21968 index 9fb4e238d4dc..c75834d8de24 100644
21969 --- a/kernel/rcu/tree.h
21970 +++ b/kernel/rcu/tree.h
21972 #include <linux/threads.h>
21973 #include <linux/cpumask.h>
21974 #include <linux/seqlock.h>
21975 +#include <linux/swait.h>
21976 #include <linux/stop_machine.h>
21979 @@ -241,7 +242,7 @@ struct rcu_node {
21980 /* Refused to boost: not sure why, though. */
21981 /* This can happen due to race conditions. */
21982 #ifdef CONFIG_RCU_NOCB_CPU
21983 - wait_queue_head_t nocb_gp_wq[2];
21984 + struct swait_queue_head nocb_gp_wq[2];
21985 /* Place for rcu_nocb_kthread() to wait GP. */
21986 #endif /* #ifdef CONFIG_RCU_NOCB_CPU */
21987 int need_future_gp[2];
21988 @@ -393,7 +394,7 @@ struct rcu_data {
21989 atomic_long_t nocb_q_count_lazy; /* invocation (all stages). */
21990 struct rcu_head *nocb_follower_head; /* CBs ready to invoke. */
21991 struct rcu_head **nocb_follower_tail;
21992 - wait_queue_head_t nocb_wq; /* For nocb kthreads to sleep on. */
21993 + struct swait_queue_head nocb_wq; /* For nocb kthreads to sleep on. */
21994 struct task_struct *nocb_kthread;
21995 int nocb_defer_wakeup; /* Defer wakeup of nocb_kthread. */
21997 @@ -472,7 +473,7 @@ struct rcu_state {
21998 unsigned long gpnum; /* Current gp number. */
21999 unsigned long completed; /* # of last completed gp. */
22000 struct task_struct *gp_kthread; /* Task for grace periods. */
22001 - wait_queue_head_t gp_wq; /* Where GP task waits. */
22002 + struct swait_queue_head gp_wq; /* Where GP task waits. */
22003 short gp_flags; /* Commands for GP task. */
22004 short gp_state; /* GP kthread sleep state. */
22006 @@ -504,7 +505,7 @@ struct rcu_state {
22007 atomic_long_t expedited_workdone3; /* # done by others #3. */
22008 atomic_long_t expedited_normal; /* # fallbacks to normal. */
22009 atomic_t expedited_need_qs; /* # CPUs left to check in. */
22010 - wait_queue_head_t expedited_wq; /* Wait for check-ins. */
22011 + struct swait_queue_head expedited_wq; /* Wait for check-ins. */
22012 int ncpus_snap; /* # CPUs seen last time. */
22014 unsigned long jiffies_force_qs; /* Time at which to invoke */
22015 @@ -556,18 +557,18 @@ extern struct list_head rcu_struct_flavors;
22017 extern struct rcu_state rcu_sched_state;
22019 +#ifndef CONFIG_PREEMPT_RT_FULL
22020 extern struct rcu_state rcu_bh_state;
22023 #ifdef CONFIG_PREEMPT_RCU
22024 extern struct rcu_state rcu_preempt_state;
22025 #endif /* #ifdef CONFIG_PREEMPT_RCU */
22027 -#ifdef CONFIG_RCU_BOOST
22028 DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_status);
22029 DECLARE_PER_CPU(int, rcu_cpu_kthread_cpu);
22030 DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_loops);
22031 DECLARE_PER_CPU(char, rcu_cpu_has_work);
22032 -#endif /* #ifdef CONFIG_RCU_BOOST */
22034 #ifndef RCU_TREE_NONCORE
22036 @@ -587,10 +588,9 @@ void call_rcu(struct rcu_head *head, rcu_callback_t func);
22037 static void __init __rcu_init_preempt(void);
22038 static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags);
22039 static void rcu_preempt_boost_start_gp(struct rcu_node *rnp);
22040 -static void invoke_rcu_callbacks_kthread(void);
22041 static bool rcu_is_callbacks_kthread(void);
22042 +static void rcu_cpu_kthread_setup(unsigned int cpu);
22043 #ifdef CONFIG_RCU_BOOST
22044 -static void rcu_preempt_do_callbacks(void);
22045 static int rcu_spawn_one_boost_kthread(struct rcu_state *rsp,
22046 struct rcu_node *rnp);
22047 #endif /* #ifdef CONFIG_RCU_BOOST */
22048 @@ -607,7 +607,8 @@ static void zero_cpu_stall_ticks(struct rcu_data *rdp);
22049 static void increment_cpu_stall_ticks(void);
22050 static bool rcu_nocb_cpu_needs_barrier(struct rcu_state *rsp, int cpu);
22051 static void rcu_nocb_gp_set(struct rcu_node *rnp, int nrq);
22052 -static void rcu_nocb_gp_cleanup(struct rcu_state *rsp, struct rcu_node *rnp);
22053 +static struct swait_queue_head *rcu_nocb_gp_get(struct rcu_node *rnp);
22054 +static void rcu_nocb_gp_cleanup(struct swait_queue_head *sq);
22055 static void rcu_init_one_nocb(struct rcu_node *rnp);
22056 static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp,
22057 bool lazy, unsigned long flags);
22058 diff --git a/kernel/rcu/tree_plugin.h b/kernel/rcu/tree_plugin.h
22059 index 32cbe72bf545..45e3e3e02a5c 100644
22060 --- a/kernel/rcu/tree_plugin.h
22061 +++ b/kernel/rcu/tree_plugin.h
22062 @@ -24,25 +24,10 @@
22063 * Paul E. McKenney <paulmck@linux.vnet.ibm.com>
22066 -#include <linux/delay.h>
22067 -#include <linux/gfp.h>
22068 -#include <linux/oom.h>
22069 -#include <linux/smpboot.h>
22070 -#include "../time/tick-internal.h"
22072 #ifdef CONFIG_RCU_BOOST
22074 #include "../locking/rtmutex_common.h"
22077 - * Control variables for per-CPU and per-rcu_node kthreads. These
22078 - * handle all flavors of RCU.
22080 -static DEFINE_PER_CPU(struct task_struct *, rcu_cpu_kthread_task);
22081 -DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_status);
22082 -DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_loops);
22083 -DEFINE_PER_CPU(char, rcu_cpu_has_work);
22085 #else /* #ifdef CONFIG_RCU_BOOST */
22088 @@ -55,6 +40,14 @@ DEFINE_PER_CPU(char, rcu_cpu_has_work);
22090 #endif /* #else #ifdef CONFIG_RCU_BOOST */
22093 + * Control variables for per-CPU and per-rcu_node kthreads. These
22094 + * handle all flavors of RCU.
22096 +DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_status);
22097 +DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_loops);
22098 +DEFINE_PER_CPU(char, rcu_cpu_has_work);
22100 #ifdef CONFIG_RCU_NOCB_CPU
22101 static cpumask_var_t rcu_nocb_mask; /* CPUs to have callbacks offloaded. */
22102 static bool have_rcu_nocb_mask; /* Was rcu_nocb_mask allocated? */
22103 @@ -432,7 +425,7 @@ void rcu_read_unlock_special(struct task_struct *t)
22106 /* Hardware IRQ handlers cannot block, complain if they get here. */
22107 - if (in_irq() || in_serving_softirq()) {
22108 + if (preempt_count() & (HARDIRQ_MASK | SOFTIRQ_OFFSET)) {
22109 lockdep_rcu_suspicious(__FILE__, __LINE__,
22110 "rcu_read_unlock() from irq or softirq with blocking in critical section!!!\n");
22111 pr_alert("->rcu_read_unlock_special: %#x (b: %d, enq: %d nq: %d)\n",
22112 @@ -645,15 +638,6 @@ static void rcu_preempt_check_callbacks(void)
22113 t->rcu_read_unlock_special.b.need_qs = true;
22116 -#ifdef CONFIG_RCU_BOOST
22118 -static void rcu_preempt_do_callbacks(void)
22120 - rcu_do_batch(rcu_state_p, this_cpu_ptr(rcu_data_p));
22123 -#endif /* #ifdef CONFIG_RCU_BOOST */
22126 * Queue a preemptible-RCU callback for invocation after a grace period.
22128 @@ -930,6 +914,19 @@ void exit_rcu(void)
22130 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */
22133 + * If boosting, set rcuc kthreads to realtime priority.
22135 +static void rcu_cpu_kthread_setup(unsigned int cpu)
22137 +#ifdef CONFIG_RCU_BOOST
22138 + struct sched_param sp;
22140 + sp.sched_priority = kthread_prio;
22141 + sched_setscheduler_nocheck(current, SCHED_FIFO, &sp);
22142 +#endif /* #ifdef CONFIG_RCU_BOOST */
22145 #ifdef CONFIG_RCU_BOOST
22147 #include "../locking/rtmutex_common.h"
22148 @@ -961,16 +958,6 @@ static void rcu_initiate_boost_trace(struct rcu_node *rnp)
22150 #endif /* #else #ifdef CONFIG_RCU_TRACE */
22152 -static void rcu_wake_cond(struct task_struct *t, int status)
22155 - * If the thread is yielding, only wake it when this
22156 - * is invoked from idle
22158 - if (status != RCU_KTHREAD_YIELDING || is_idle_task(current))
22159 - wake_up_process(t);
22163 * Carry out RCU priority boosting on the task indicated by ->exp_tasks
22164 * or ->boost_tasks, advancing the pointer to the next task in the
22165 @@ -1115,23 +1102,6 @@ static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags)
22169 - * Wake up the per-CPU kthread to invoke RCU callbacks.
22171 -static void invoke_rcu_callbacks_kthread(void)
22173 - unsigned long flags;
22175 - local_irq_save(flags);
22176 - __this_cpu_write(rcu_cpu_has_work, 1);
22177 - if (__this_cpu_read(rcu_cpu_kthread_task) != NULL &&
22178 - current != __this_cpu_read(rcu_cpu_kthread_task)) {
22179 - rcu_wake_cond(__this_cpu_read(rcu_cpu_kthread_task),
22180 - __this_cpu_read(rcu_cpu_kthread_status));
22182 - local_irq_restore(flags);
22186 * Is the current CPU running the RCU-callbacks kthread?
22187 * Caller must have preemption disabled.
22189 @@ -1186,67 +1156,6 @@ static int rcu_spawn_one_boost_kthread(struct rcu_state *rsp,
22193 -static void rcu_kthread_do_work(void)
22195 - rcu_do_batch(&rcu_sched_state, this_cpu_ptr(&rcu_sched_data));
22196 - rcu_do_batch(&rcu_bh_state, this_cpu_ptr(&rcu_bh_data));
22197 - rcu_preempt_do_callbacks();
22200 -static void rcu_cpu_kthread_setup(unsigned int cpu)
22202 - struct sched_param sp;
22204 - sp.sched_priority = kthread_prio;
22205 - sched_setscheduler_nocheck(current, SCHED_FIFO, &sp);
22208 -static void rcu_cpu_kthread_park(unsigned int cpu)
22210 - per_cpu(rcu_cpu_kthread_status, cpu) = RCU_KTHREAD_OFFCPU;
22213 -static int rcu_cpu_kthread_should_run(unsigned int cpu)
22215 - return __this_cpu_read(rcu_cpu_has_work);
22219 - * Per-CPU kernel thread that invokes RCU callbacks. This replaces the
22220 - * RCU softirq used in flavors and configurations of RCU that do not
22221 - * support RCU priority boosting.
22223 -static void rcu_cpu_kthread(unsigned int cpu)
22225 - unsigned int *statusp = this_cpu_ptr(&rcu_cpu_kthread_status);
22226 - char work, *workp = this_cpu_ptr(&rcu_cpu_has_work);
22229 - for (spincnt = 0; spincnt < 10; spincnt++) {
22230 - trace_rcu_utilization(TPS("Start CPU kthread@rcu_wait"));
22231 - local_bh_disable();
22232 - *statusp = RCU_KTHREAD_RUNNING;
22233 - this_cpu_inc(rcu_cpu_kthread_loops);
22234 - local_irq_disable();
22237 - local_irq_enable();
22239 - rcu_kthread_do_work();
22240 - local_bh_enable();
22241 - if (*workp == 0) {
22242 - trace_rcu_utilization(TPS("End CPU kthread@rcu_wait"));
22243 - *statusp = RCU_KTHREAD_WAITING;
22247 - *statusp = RCU_KTHREAD_YIELDING;
22248 - trace_rcu_utilization(TPS("Start CPU kthread@rcu_yield"));
22249 - schedule_timeout_interruptible(2);
22250 - trace_rcu_utilization(TPS("End CPU kthread@rcu_yield"));
22251 - *statusp = RCU_KTHREAD_WAITING;
22255 * Set the per-rcu_node kthread's affinity to cover all CPUs that are
22256 * served by the rcu_node in question. The CPU hotplug lock is still
22257 @@ -1276,26 +1185,12 @@ static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu)
22258 free_cpumask_var(cm);
22261 -static struct smp_hotplug_thread rcu_cpu_thread_spec = {
22262 - .store = &rcu_cpu_kthread_task,
22263 - .thread_should_run = rcu_cpu_kthread_should_run,
22264 - .thread_fn = rcu_cpu_kthread,
22265 - .thread_comm = "rcuc/%u",
22266 - .setup = rcu_cpu_kthread_setup,
22267 - .park = rcu_cpu_kthread_park,
22271 * Spawn boost kthreads -- called as soon as the scheduler is running.
22273 static void __init rcu_spawn_boost_kthreads(void)
22275 struct rcu_node *rnp;
22278 - for_each_possible_cpu(cpu)
22279 - per_cpu(rcu_cpu_has_work, cpu) = 0;
22280 - BUG_ON(smpboot_register_percpu_thread(&rcu_cpu_thread_spec));
22281 rcu_for_each_leaf_node(rcu_state_p, rnp)
22282 (void)rcu_spawn_one_boost_kthread(rcu_state_p, rnp);
22284 @@ -1318,11 +1213,6 @@ static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags)
22285 raw_spin_unlock_irqrestore(&rnp->lock, flags);
22288 -static void invoke_rcu_callbacks_kthread(void)
22293 static bool rcu_is_callbacks_kthread(void)
22296 @@ -1346,7 +1236,7 @@ static void rcu_prepare_kthreads(int cpu)
22298 #endif /* #else #ifdef CONFIG_RCU_BOOST */
22300 -#if !defined(CONFIG_RCU_FAST_NO_HZ)
22301 +#if !defined(CONFIG_RCU_FAST_NO_HZ) || defined(CONFIG_PREEMPT_RT_FULL)
22304 * Check to see if any future RCU-related work will need to be done
22305 @@ -1363,7 +1253,9 @@ int rcu_needs_cpu(u64 basemono, u64 *nextevt)
22306 return IS_ENABLED(CONFIG_RCU_NOCB_CPU_ALL)
22307 ? 0 : rcu_cpu_has_callbacks(NULL);
22309 +#endif /* !defined(CONFIG_RCU_FAST_NO_HZ) || defined(CONFIG_PREEMPT_RT_FULL) */
22311 +#if !defined(CONFIG_RCU_FAST_NO_HZ)
22313 * Because we do not have RCU_FAST_NO_HZ, don't bother cleaning up
22315 @@ -1459,6 +1351,8 @@ static bool __maybe_unused rcu_try_advance_all_cbs(void)
22319 +#ifndef CONFIG_PREEMPT_RT_FULL
22322 * Allow the CPU to enter dyntick-idle mode unless it has callbacks ready
22323 * to invoke. If the CPU has callbacks, try to advance them. Tell the
22324 @@ -1504,6 +1398,7 @@ int rcu_needs_cpu(u64 basemono, u64 *nextevt)
22325 *nextevt = basemono + dj * TICK_NSEC;
22328 +#endif /* #ifndef CONFIG_PREEMPT_RT_FULL */
22331 * Prepare a CPU for idle from an RCU perspective. The first major task
22332 @@ -1822,9 +1717,9 @@ early_param("rcu_nocb_poll", parse_rcu_nocb_poll);
22333 * Wake up any no-CBs CPUs' kthreads that were waiting on the just-ended
22336 -static void rcu_nocb_gp_cleanup(struct rcu_state *rsp, struct rcu_node *rnp)
22337 +static void rcu_nocb_gp_cleanup(struct swait_queue_head *sq)
22339 - wake_up_all(&rnp->nocb_gp_wq[rnp->completed & 0x1]);
22340 + swake_up_all(sq);
22344 @@ -1840,10 +1735,15 @@ static void rcu_nocb_gp_set(struct rcu_node *rnp, int nrq)
22345 rnp->need_future_gp[(rnp->completed + 1) & 0x1] += nrq;
22348 +static struct swait_queue_head *rcu_nocb_gp_get(struct rcu_node *rnp)
22350 + return &rnp->nocb_gp_wq[rnp->completed & 0x1];
22353 static void rcu_init_one_nocb(struct rcu_node *rnp)
22355 - init_waitqueue_head(&rnp->nocb_gp_wq[0]);
22356 - init_waitqueue_head(&rnp->nocb_gp_wq[1]);
22357 + init_swait_queue_head(&rnp->nocb_gp_wq[0]);
22358 + init_swait_queue_head(&rnp->nocb_gp_wq[1]);
22361 #ifndef CONFIG_RCU_NOCB_CPU_ALL
22362 @@ -1868,7 +1768,7 @@ static void wake_nocb_leader(struct rcu_data *rdp, bool force)
22363 if (READ_ONCE(rdp_leader->nocb_leader_sleep) || force) {
22364 /* Prior smp_mb__after_atomic() orders against prior enqueue. */
22365 WRITE_ONCE(rdp_leader->nocb_leader_sleep, false);
22366 - wake_up(&rdp_leader->nocb_wq);
22367 + swake_up(&rdp_leader->nocb_wq);
22371 @@ -2081,7 +1981,7 @@ static void rcu_nocb_wait_gp(struct rcu_data *rdp)
22373 trace_rcu_future_gp(rnp, rdp, c, TPS("StartWait"));
22375 - wait_event_interruptible(
22376 + swait_event_interruptible(
22377 rnp->nocb_gp_wq[c & 0x1],
22378 (d = ULONG_CMP_GE(READ_ONCE(rnp->completed), c)));
22380 @@ -2109,7 +2009,7 @@ wait_again:
22381 /* Wait for callbacks to appear. */
22382 if (!rcu_nocb_poll) {
22383 trace_rcu_nocb_wake(my_rdp->rsp->name, my_rdp->cpu, "Sleep");
22384 - wait_event_interruptible(my_rdp->nocb_wq,
22385 + swait_event_interruptible(my_rdp->nocb_wq,
22386 !READ_ONCE(my_rdp->nocb_leader_sleep));
22387 /* Memory barrier handled by smp_mb() calls below and repoll. */
22388 } else if (firsttime) {
22389 @@ -2184,7 +2084,7 @@ wait_again:
22390 * List was empty, wake up the follower.
22391 * Memory barriers supplied by atomic_long_add().
22393 - wake_up(&rdp->nocb_wq);
22394 + swake_up(&rdp->nocb_wq);
22398 @@ -2205,7 +2105,7 @@ static void nocb_follower_wait(struct rcu_data *rdp)
22399 if (!rcu_nocb_poll) {
22400 trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu,
22402 - wait_event_interruptible(rdp->nocb_wq,
22403 + swait_event_interruptible(rdp->nocb_wq,
22404 READ_ONCE(rdp->nocb_follower_head));
22405 } else if (firsttime) {
22406 /* Don't drown trace log with "Poll"! */
22407 @@ -2365,7 +2265,7 @@ void __init rcu_init_nohz(void)
22408 static void __init rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp)
22410 rdp->nocb_tail = &rdp->nocb_head;
22411 - init_waitqueue_head(&rdp->nocb_wq);
22412 + init_swait_queue_head(&rdp->nocb_wq);
22413 rdp->nocb_follower_tail = &rdp->nocb_follower_head;
22416 @@ -2515,7 +2415,7 @@ static bool rcu_nocb_cpu_needs_barrier(struct rcu_state *rsp, int cpu)
22420 -static void rcu_nocb_gp_cleanup(struct rcu_state *rsp, struct rcu_node *rnp)
22421 +static void rcu_nocb_gp_cleanup(struct swait_queue_head *sq)
22425 @@ -2523,6 +2423,11 @@ static void rcu_nocb_gp_set(struct rcu_node *rnp, int nrq)
22429 +static struct swait_queue_head *rcu_nocb_gp_get(struct rcu_node *rnp)
22434 static void rcu_init_one_nocb(struct rcu_node *rnp)
22437 diff --git a/kernel/rcu/update.c b/kernel/rcu/update.c
22438 index 5f748c5a40f0..9a3904603ff6 100644
22439 --- a/kernel/rcu/update.c
22440 +++ b/kernel/rcu/update.c
22441 @@ -276,6 +276,7 @@ int rcu_read_lock_held(void)
22443 EXPORT_SYMBOL_GPL(rcu_read_lock_held);
22445 +#ifndef CONFIG_PREEMPT_RT_FULL
22447 * rcu_read_lock_bh_held() - might we be in RCU-bh read-side critical section?
22449 @@ -302,6 +303,7 @@ int rcu_read_lock_bh_held(void)
22450 return in_softirq() || irqs_disabled();
22452 EXPORT_SYMBOL_GPL(rcu_read_lock_bh_held);
22455 #endif /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
22457 diff --git a/kernel/relay.c b/kernel/relay.c
22458 index 0b4570cfacae..60684be39f22 100644
22459 --- a/kernel/relay.c
22460 +++ b/kernel/relay.c
22461 @@ -336,6 +336,10 @@ static void wakeup_readers(unsigned long data)
22463 struct rchan_buf *buf = (struct rchan_buf *)data;
22464 wake_up_interruptible(&buf->read_wait);
22466 + * Stupid polling for now:
22468 + mod_timer(&buf->timer, jiffies + 1);
22472 @@ -353,6 +357,7 @@ static void __relay_reset(struct rchan_buf *buf, unsigned int init)
22473 init_waitqueue_head(&buf->read_wait);
22474 kref_init(&buf->kref);
22475 setup_timer(&buf->timer, wakeup_readers, (unsigned long)buf);
22476 + mod_timer(&buf->timer, jiffies + 1);
22478 del_timer_sync(&buf->timer);
22480 @@ -736,15 +741,6 @@ size_t relay_switch_subbuf(struct rchan_buf *buf, size_t length)
22482 buf->early_bytes += buf->chan->subbuf_size -
22483 buf->padding[old_subbuf];
22485 - if (waitqueue_active(&buf->read_wait))
22487 - * Calling wake_up_interruptible() from here
22488 - * will deadlock if we happen to be logging
22489 - * from the scheduler (trying to re-grab
22490 - * rq->lock), so defer it.
22492 - mod_timer(&buf->timer, jiffies + 1);
22496 diff --git a/kernel/sched/Makefile b/kernel/sched/Makefile
22497 index 67687973ce80..01b9994b367a 100644
22498 --- a/kernel/sched/Makefile
22499 +++ b/kernel/sched/Makefile
22500 @@ -13,7 +13,7 @@ endif
22502 obj-y += core.o loadavg.o clock.o cputime.o
22503 obj-y += idle_task.o fair.o rt.o deadline.o stop_task.o
22504 -obj-y += wait.o completion.o idle.o
22505 +obj-y += wait.o swait.o swork.o completion.o idle.o
22506 obj-$(CONFIG_SMP) += cpupri.o cpudeadline.o
22507 obj-$(CONFIG_SCHED_AUTOGROUP) += auto_group.o
22508 obj-$(CONFIG_SCHEDSTATS) += stats.o
22509 diff --git a/kernel/sched/completion.c b/kernel/sched/completion.c
22510 index 8d0f35debf35..b62cf6400fe0 100644
22511 --- a/kernel/sched/completion.c
22512 +++ b/kernel/sched/completion.c
22513 @@ -30,10 +30,10 @@ void complete(struct completion *x)
22515 unsigned long flags;
22517 - spin_lock_irqsave(&x->wait.lock, flags);
22518 + raw_spin_lock_irqsave(&x->wait.lock, flags);
22520 - __wake_up_locked(&x->wait, TASK_NORMAL, 1);
22521 - spin_unlock_irqrestore(&x->wait.lock, flags);
22522 + swake_up_locked(&x->wait);
22523 + raw_spin_unlock_irqrestore(&x->wait.lock, flags);
22525 EXPORT_SYMBOL(complete);
22527 @@ -50,10 +50,10 @@ void complete_all(struct completion *x)
22529 unsigned long flags;
22531 - spin_lock_irqsave(&x->wait.lock, flags);
22532 + raw_spin_lock_irqsave(&x->wait.lock, flags);
22533 x->done += UINT_MAX/2;
22534 - __wake_up_locked(&x->wait, TASK_NORMAL, 0);
22535 - spin_unlock_irqrestore(&x->wait.lock, flags);
22536 + swake_up_all_locked(&x->wait);
22537 + raw_spin_unlock_irqrestore(&x->wait.lock, flags);
22539 EXPORT_SYMBOL(complete_all);
22541 @@ -62,20 +62,20 @@ do_wait_for_common(struct completion *x,
22542 long (*action)(long), long timeout, int state)
22545 - DECLARE_WAITQUEUE(wait, current);
22546 + DECLARE_SWAITQUEUE(wait);
22548 - __add_wait_queue_tail_exclusive(&x->wait, &wait);
22549 + __prepare_to_swait(&x->wait, &wait);
22551 if (signal_pending_state(state, current)) {
22552 timeout = -ERESTARTSYS;
22555 __set_current_state(state);
22556 - spin_unlock_irq(&x->wait.lock);
22557 + raw_spin_unlock_irq(&x->wait.lock);
22558 timeout = action(timeout);
22559 - spin_lock_irq(&x->wait.lock);
22560 + raw_spin_lock_irq(&x->wait.lock);
22561 } while (!x->done && timeout);
22562 - __remove_wait_queue(&x->wait, &wait);
22563 + __finish_swait(&x->wait, &wait);
22567 @@ -89,9 +89,9 @@ __wait_for_common(struct completion *x,
22571 - spin_lock_irq(&x->wait.lock);
22572 + raw_spin_lock_irq(&x->wait.lock);
22573 timeout = do_wait_for_common(x, action, timeout, state);
22574 - spin_unlock_irq(&x->wait.lock);
22575 + raw_spin_unlock_irq(&x->wait.lock);
22579 @@ -277,12 +277,12 @@ bool try_wait_for_completion(struct completion *x)
22580 if (!READ_ONCE(x->done))
22583 - spin_lock_irqsave(&x->wait.lock, flags);
22584 + raw_spin_lock_irqsave(&x->wait.lock, flags);
22589 - spin_unlock_irqrestore(&x->wait.lock, flags);
22590 + raw_spin_unlock_irqrestore(&x->wait.lock, flags);
22593 EXPORT_SYMBOL(try_wait_for_completion);
22594 @@ -311,7 +311,7 @@ bool completion_done(struct completion *x)
22595 * after it's acquired the lock.
22598 - spin_unlock_wait(&x->wait.lock);
22599 + raw_spin_unlock_wait(&x->wait.lock);
22602 EXPORT_SYMBOL(completion_done);
22603 diff --git a/kernel/sched/core.c b/kernel/sched/core.c
22604 index 20253dbc8610..e9b8d518202e 100644
22605 --- a/kernel/sched/core.c
22606 +++ b/kernel/sched/core.c
22607 @@ -260,7 +260,11 @@ late_initcall(sched_init_debug);
22608 * Number of tasks to iterate in a single balance run.
22609 * Limited because this is done with IRQs disabled.
22611 +#ifndef CONFIG_PREEMPT_RT_FULL
22612 const_debug unsigned int sysctl_sched_nr_migrate = 32;
22614 +const_debug unsigned int sysctl_sched_nr_migrate = 8;
22618 * period over which we average the RT time consumption, measured
22619 @@ -438,6 +442,7 @@ static void init_rq_hrtick(struct rq *rq)
22621 hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
22622 rq->hrtick_timer.function = hrtick;
22623 + rq->hrtick_timer.irqsafe = 1;
22625 #else /* CONFIG_SCHED_HRTICK */
22626 static inline void hrtick_clear(struct rq *rq)
22627 @@ -542,7 +547,7 @@ void wake_q_add(struct wake_q_head *head, struct task_struct *task)
22628 head->lastp = &node->next;
22631 -void wake_up_q(struct wake_q_head *head)
22632 +void __wake_up_q(struct wake_q_head *head, bool sleeper)
22634 struct wake_q_node *node = head->first;
22636 @@ -559,7 +564,10 @@ void wake_up_q(struct wake_q_head *head)
22637 * wake_up_process() implies a wmb() to pair with the queueing
22638 * in wake_q_add() so as not to miss wakeups.
22640 - wake_up_process(task);
22642 + wake_up_lock_sleeper(task);
22644 + wake_up_process(task);
22645 put_task_struct(task);
22648 @@ -595,6 +603,38 @@ void resched_curr(struct rq *rq)
22649 trace_sched_wake_idle_without_ipi(cpu);
22652 +#ifdef CONFIG_PREEMPT_LAZY
22653 +void resched_curr_lazy(struct rq *rq)
22655 + struct task_struct *curr = rq->curr;
22658 + if (!sched_feat(PREEMPT_LAZY)) {
22659 + resched_curr(rq);
22663 + lockdep_assert_held(&rq->lock);
22665 + if (test_tsk_need_resched(curr))
22668 + if (test_tsk_need_resched_lazy(curr))
22671 + set_tsk_need_resched_lazy(curr);
22673 + cpu = cpu_of(rq);
22674 + if (cpu == smp_processor_id())
22677 + /* NEED_RESCHED_LAZY must be visible before we test polling */
22679 + if (!tsk_is_polling(curr))
22680 + smp_send_reschedule(cpu);
22684 void resched_cpu(int cpu)
22686 struct rq *rq = cpu_rq(cpu);
22687 @@ -618,11 +658,14 @@ void resched_cpu(int cpu)
22689 int get_nohz_timer_target(void)
22691 - int i, cpu = smp_processor_id();
22693 struct sched_domain *sd;
22695 + preempt_disable_rt();
22696 + cpu = smp_processor_id();
22698 if (!idle_cpu(cpu) && is_housekeeping_cpu(cpu))
22700 + goto preempt_en_rt;
22703 for_each_domain(cpu, sd) {
22704 @@ -641,6 +684,8 @@ int get_nohz_timer_target(void)
22705 cpu = housekeeping_any_cpu();
22709 + preempt_enable_rt();
22713 @@ -1174,6 +1219,11 @@ void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask)
22715 lockdep_assert_held(&p->pi_lock);
22717 + if (__migrate_disabled(p)) {
22718 + cpumask_copy(&p->cpus_allowed, new_mask);
22722 queued = task_on_rq_queued(p);
22723 running = task_current(rq, p);
22725 @@ -1196,6 +1246,84 @@ void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask)
22726 enqueue_task(rq, p, ENQUEUE_RESTORE);
22729 +static DEFINE_PER_CPU(struct cpumask, sched_cpumasks);
22730 +static DEFINE_MUTEX(sched_down_mutex);
22731 +static cpumask_t sched_down_cpumask;
22733 +void tell_sched_cpu_down_begin(int cpu)
22735 + mutex_lock(&sched_down_mutex);
22736 + cpumask_set_cpu(cpu, &sched_down_cpumask);
22737 + mutex_unlock(&sched_down_mutex);
22740 +void tell_sched_cpu_down_done(int cpu)
22742 + mutex_lock(&sched_down_mutex);
22743 + cpumask_clear_cpu(cpu, &sched_down_cpumask);
22744 + mutex_unlock(&sched_down_mutex);
22748 + * migrate_me - try to move the current task off this cpu
22750 + * Used by the pin_current_cpu() code to try to get tasks
22751 + * to move off the current CPU as it is going down.
22752 + * It will only move the task if the task isn't pinned to
22753 + * the CPU (with migrate_disable, affinity or NO_SETAFFINITY)
22754 + * and the task has to be in a RUNNING state. Otherwise the
22755 + * movement of the task will wake it up (change its state
22756 + * to running) when the task did not expect it.
22758 + * Returns 1 if it succeeded in moving the current task
22761 +int migrate_me(void)
22763 + struct task_struct *p = current;
22764 + struct migration_arg arg;
22765 + struct cpumask *cpumask;
22766 + struct cpumask *mask;
22767 + unsigned long flags;
22768 + unsigned int dest_cpu;
22772 + * We can not migrate tasks bounded to a CPU or tasks not
22773 + * running. The movement of the task will wake it up.
22775 + if (p->flags & PF_NO_SETAFFINITY || p->state)
22778 + mutex_lock(&sched_down_mutex);
22779 + rq = task_rq_lock(p, &flags);
22781 + cpumask = this_cpu_ptr(&sched_cpumasks);
22782 + mask = &p->cpus_allowed;
22784 + cpumask_andnot(cpumask, mask, &sched_down_cpumask);
22786 + if (!cpumask_weight(cpumask)) {
22787 + /* It's only on this CPU? */
22788 + task_rq_unlock(rq, p, &flags);
22789 + mutex_unlock(&sched_down_mutex);
22793 + dest_cpu = cpumask_any_and(cpu_active_mask, cpumask);
22796 + arg.dest_cpu = dest_cpu;
22798 + task_rq_unlock(rq, p, &flags);
22800 + stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg);
22801 + tlb_migrate_finish(p->mm);
22802 + mutex_unlock(&sched_down_mutex);
22808 * Change a given task's CPU affinity. Migrate the thread to a
22809 * proper CPU and schedule it away if the CPU it's executing on
22810 @@ -1235,7 +1363,7 @@ static int __set_cpus_allowed_ptr(struct task_struct *p,
22811 do_set_cpus_allowed(p, new_mask);
22813 /* Can the task run on the task's current CPU? If so, we're done */
22814 - if (cpumask_test_cpu(task_cpu(p), new_mask))
22815 + if (cpumask_test_cpu(task_cpu(p), new_mask) || __migrate_disabled(p))
22818 dest_cpu = cpumask_any_and(cpu_active_mask, new_mask);
22819 @@ -1411,6 +1539,18 @@ out:
22823 +static bool check_task_state(struct task_struct *p, long match_state)
22825 + bool match = false;
22827 + raw_spin_lock_irq(&p->pi_lock);
22828 + if (p->state == match_state || p->saved_state == match_state)
22830 + raw_spin_unlock_irq(&p->pi_lock);
22836 * wait_task_inactive - wait for a thread to unschedule.
22838 @@ -1455,7 +1595,7 @@ unsigned long wait_task_inactive(struct task_struct *p, long match_state)
22839 * is actually now running somewhere else!
22841 while (task_running(rq, p)) {
22842 - if (match_state && unlikely(p->state != match_state))
22843 + if (match_state && !check_task_state(p, match_state))
22847 @@ -1470,7 +1610,8 @@ unsigned long wait_task_inactive(struct task_struct *p, long match_state)
22848 running = task_running(rq, p);
22849 queued = task_on_rq_queued(p);
22851 - if (!match_state || p->state == match_state)
22852 + if (!match_state || p->state == match_state ||
22853 + p->saved_state == match_state)
22854 ncsw = p->nvcsw | LONG_MIN; /* sets MSB */
22855 task_rq_unlock(rq, p, &flags);
22857 @@ -1627,7 +1768,7 @@ int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags)
22859 lockdep_assert_held(&p->pi_lock);
22861 - if (p->nr_cpus_allowed > 1)
22862 + if (tsk_nr_cpus_allowed(p) > 1)
22863 cpu = p->sched_class->select_task_rq(p, cpu, sd_flags, wake_flags);
22866 @@ -1707,10 +1848,6 @@ static inline void ttwu_activate(struct rq *rq, struct task_struct *p, int en_fl
22868 activate_task(rq, p, en_flags);
22869 p->on_rq = TASK_ON_RQ_QUEUED;
22871 - /* if a worker is waking up, notify workqueue */
22872 - if (p->flags & PF_WQ_WORKER)
22873 - wq_worker_waking_up(p, cpu_of(rq));
22877 @@ -1937,8 +2074,27 @@ try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags)
22879 smp_mb__before_spinlock();
22880 raw_spin_lock_irqsave(&p->pi_lock, flags);
22881 - if (!(p->state & state))
22882 + if (!(p->state & state)) {
22884 + * The task might be running due to a spinlock sleeper
22885 + * wakeup. Check the saved state and set it to running
22886 + * if the wakeup condition is true.
22888 + if (!(wake_flags & WF_LOCK_SLEEPER)) {
22889 + if (p->saved_state & state) {
22890 + p->saved_state = TASK_RUNNING;
22898 + * If this is a regular wakeup, then we can unconditionally
22899 + * clear the saved state of a "lock sleeper".
22901 + if (!(wake_flags & WF_LOCK_SLEEPER))
22902 + p->saved_state = TASK_RUNNING;
22904 trace_sched_waking(p);
22906 @@ -2030,52 +2186,6 @@ out:
22910 - * try_to_wake_up_local - try to wake up a local task with rq lock held
22911 - * @p: the thread to be awakened
22913 - * Put @p on the run-queue if it's not already there. The caller must
22914 - * ensure that this_rq() is locked, @p is bound to this_rq() and not
22915 - * the current task.
22917 -static void try_to_wake_up_local(struct task_struct *p)
22919 - struct rq *rq = task_rq(p);
22921 - if (WARN_ON_ONCE(rq != this_rq()) ||
22922 - WARN_ON_ONCE(p == current))
22925 - lockdep_assert_held(&rq->lock);
22927 - if (!raw_spin_trylock(&p->pi_lock)) {
22929 - * This is OK, because current is on_cpu, which avoids it being
22930 - * picked for load-balance and preemption/IRQs are still
22931 - * disabled avoiding further scheduler activity on it and we've
22932 - * not yet picked a replacement task.
22934 - lockdep_unpin_lock(&rq->lock);
22935 - raw_spin_unlock(&rq->lock);
22936 - raw_spin_lock(&p->pi_lock);
22937 - raw_spin_lock(&rq->lock);
22938 - lockdep_pin_lock(&rq->lock);
22941 - if (!(p->state & TASK_NORMAL))
22944 - trace_sched_waking(p);
22946 - if (!task_on_rq_queued(p))
22947 - ttwu_activate(rq, p, ENQUEUE_WAKEUP);
22949 - ttwu_do_wakeup(rq, p, 0);
22950 - ttwu_stat(p, smp_processor_id(), 0);
22952 - raw_spin_unlock(&p->pi_lock);
22956 * wake_up_process - Wake up a specific process
22957 * @p: The process to be woken up.
22959 @@ -2093,6 +2203,18 @@ int wake_up_process(struct task_struct *p)
22961 EXPORT_SYMBOL(wake_up_process);
22964 + * wake_up_lock_sleeper - Wake up a specific process blocked on a "sleeping lock"
22965 + * @p: The process to be woken up.
22967 + * Same as wake_up_process() above, but wake_flags=WF_LOCK_SLEEPER to indicate
22968 + * the nature of the wakeup.
22970 +int wake_up_lock_sleeper(struct task_struct *p)
22972 + return try_to_wake_up(p, TASK_ALL, WF_LOCK_SLEEPER);
22975 int wake_up_state(struct task_struct *p, unsigned int state)
22977 return try_to_wake_up(p, state, 0);
22978 @@ -2279,6 +2401,9 @@ int sched_fork(unsigned long clone_flags, struct task_struct *p)
22981 init_task_preempt_count(p);
22982 +#ifdef CONFIG_HAVE_PREEMPT_LAZY
22983 + task_thread_info(p)->preempt_lazy_count = 0;
22986 plist_node_init(&p->pushable_tasks, MAX_PRIO);
22987 RB_CLEAR_NODE(&p->pushable_dl_tasks);
22988 @@ -2603,8 +2728,12 @@ static struct rq *finish_task_switch(struct task_struct *prev)
22989 finish_arch_post_lock_switch();
22991 fire_sched_in_preempt_notifiers(current);
22993 + * We use mmdrop_delayed() here so we don't have to do the
22994 + * full __mmdrop() when we are the last user.
22998 + mmdrop_delayed(mm);
22999 if (unlikely(prev_state == TASK_DEAD)) {
23000 if (prev->sched_class->task_dead)
23001 prev->sched_class->task_dead(prev);
23002 @@ -2935,16 +3064,6 @@ u64 scheduler_tick_max_deferment(void)
23006 -notrace unsigned long get_parent_ip(unsigned long addr)
23008 - if (in_lock_functions(addr)) {
23009 - addr = CALLER_ADDR2;
23010 - if (in_lock_functions(addr))
23011 - addr = CALLER_ADDR3;
23016 #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \
23017 defined(CONFIG_PREEMPT_TRACER))
23019 @@ -2966,7 +3085,7 @@ void preempt_count_add(int val)
23020 PREEMPT_MASK - 10);
23022 if (preempt_count() == val) {
23023 - unsigned long ip = get_parent_ip(CALLER_ADDR1);
23024 + unsigned long ip = get_lock_parent_ip();
23025 #ifdef CONFIG_DEBUG_PREEMPT
23026 current->preempt_disable_ip = ip;
23028 @@ -2993,7 +3112,7 @@ void preempt_count_sub(int val)
23031 if (preempt_count() == val)
23032 - trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1));
23033 + trace_preempt_on(CALLER_ADDR0, get_lock_parent_ip());
23034 __preempt_count_sub(val);
23036 EXPORT_SYMBOL(preempt_count_sub);
23037 @@ -3048,6 +3167,77 @@ static inline void schedule_debug(struct task_struct *prev)
23038 schedstat_inc(this_rq(), sched_count);
23041 +#if defined(CONFIG_PREEMPT_RT_FULL) && defined(CONFIG_SMP)
23043 +void migrate_disable(void)
23045 + struct task_struct *p = current;
23047 + if (in_atomic() || irqs_disabled()) {
23048 +#ifdef CONFIG_SCHED_DEBUG
23049 + p->migrate_disable_atomic++;
23054 +#ifdef CONFIG_SCHED_DEBUG
23055 + if (unlikely(p->migrate_disable_atomic)) {
23061 + if (p->migrate_disable) {
23062 + p->migrate_disable++;
23066 + preempt_disable();
23067 + preempt_lazy_disable();
23068 + pin_current_cpu();
23069 + p->migrate_disable = 1;
23070 + preempt_enable();
23072 +EXPORT_SYMBOL(migrate_disable);
23074 +void migrate_enable(void)
23076 + struct task_struct *p = current;
23078 + if (in_atomic() || irqs_disabled()) {
23079 +#ifdef CONFIG_SCHED_DEBUG
23080 + p->migrate_disable_atomic--;
23085 +#ifdef CONFIG_SCHED_DEBUG
23086 + if (unlikely(p->migrate_disable_atomic)) {
23091 + WARN_ON_ONCE(p->migrate_disable <= 0);
23093 + if (p->migrate_disable > 1) {
23094 + p->migrate_disable--;
23098 + preempt_disable();
23100 + * Clearing migrate_disable causes tsk_cpus_allowed to
23101 + * show the tasks original cpu affinity.
23103 + p->migrate_disable = 0;
23105 + unpin_current_cpu();
23106 + preempt_enable();
23107 + preempt_lazy_enable();
23109 +EXPORT_SYMBOL(migrate_enable);
23113 * Pick up the highest-prio task:
23115 @@ -3172,19 +3362,6 @@ static void __sched notrace __schedule(bool preempt)
23117 deactivate_task(rq, prev, DEQUEUE_SLEEP);
23121 - * If a worker went to sleep, notify and ask workqueue
23122 - * whether it wants to wake up a task to maintain
23125 - if (prev->flags & PF_WQ_WORKER) {
23126 - struct task_struct *to_wakeup;
23128 - to_wakeup = wq_worker_sleeping(prev, cpu);
23130 - try_to_wake_up_local(to_wakeup);
23133 switch_count = &prev->nvcsw;
23135 @@ -3194,6 +3371,7 @@ static void __sched notrace __schedule(bool preempt)
23137 next = pick_next_task(rq, prev);
23138 clear_tsk_need_resched(prev);
23139 + clear_tsk_need_resched_lazy(prev);
23140 clear_preempt_need_resched();
23141 rq->clock_skip_update = 0;
23143 @@ -3215,9 +3393,20 @@ static void __sched notrace __schedule(bool preempt)
23145 static inline void sched_submit_work(struct task_struct *tsk)
23147 - if (!tsk->state || tsk_is_pi_blocked(tsk))
23151 + * If a worker went to sleep, notify and ask workqueue whether
23152 + * it wants to wake up a task to maintain concurrency.
23154 + if (tsk->flags & PF_WQ_WORKER)
23155 + wq_worker_sleeping(tsk);
23158 + if (tsk_is_pi_blocked(tsk))
23162 * If we are going to sleep and we have plugged IO queued,
23163 * make sure to submit it to avoid deadlocks.
23165 @@ -3225,6 +3414,12 @@ static inline void sched_submit_work(struct task_struct *tsk)
23166 blk_schedule_flush_plug(tsk);
23169 +static void sched_update_worker(struct task_struct *tsk)
23171 + if (tsk->flags & PF_WQ_WORKER)
23172 + wq_worker_running(tsk);
23175 asmlinkage __visible void __sched schedule(void)
23177 struct task_struct *tsk = current;
23178 @@ -3235,6 +3430,7 @@ asmlinkage __visible void __sched schedule(void)
23180 sched_preempt_enable_no_resched();
23181 } while (need_resched());
23182 + sched_update_worker(tsk);
23184 EXPORT_SYMBOL(schedule);
23186 @@ -3283,6 +3479,30 @@ static void __sched notrace preempt_schedule_common(void)
23187 } while (need_resched());
23190 +#ifdef CONFIG_PREEMPT_LAZY
23192 + * If TIF_NEED_RESCHED is then we allow to be scheduled away since this is
23193 + * set by a RT task. Oterwise we try to avoid beeing scheduled out as long as
23194 + * preempt_lazy_count counter >0.
23196 +static __always_inline int preemptible_lazy(void)
23198 + if (test_thread_flag(TIF_NEED_RESCHED))
23200 + if (current_thread_info()->preempt_lazy_count)
23207 +static inline int preemptible_lazy(void)
23214 #ifdef CONFIG_PREEMPT
23216 * this is the entry point to schedule() from in-kernel preemption
23217 @@ -3297,6 +3517,8 @@ asmlinkage __visible void __sched notrace preempt_schedule(void)
23219 if (likely(!preemptible()))
23221 + if (!preemptible_lazy())
23224 preempt_schedule_common();
23226 @@ -3323,6 +3545,8 @@ asmlinkage __visible void __sched notrace preempt_schedule_notrace(void)
23228 if (likely(!preemptible()))
23230 + if (!preemptible_lazy())
23234 preempt_disable_notrace();
23235 @@ -3332,7 +3556,16 @@ asmlinkage __visible void __sched notrace preempt_schedule_notrace(void)
23236 * an infinite recursion.
23238 prev_ctx = exception_enter();
23240 + * The add/subtract must not be traced by the function
23241 + * tracer. But we still want to account for the
23242 + * preempt off latency tracer. Since the _notrace versions
23243 + * of add/subtract skip the accounting for latency tracer
23244 + * we must force it manually.
23246 + start_critical_timings();
23248 + stop_critical_timings();
23249 exception_exit(prev_ctx);
23251 preempt_enable_no_resched_notrace();
23252 @@ -4676,6 +4909,7 @@ int __cond_resched_lock(spinlock_t *lock)
23254 EXPORT_SYMBOL(__cond_resched_lock);
23256 +#ifndef CONFIG_PREEMPT_RT_FULL
23257 int __sched __cond_resched_softirq(void)
23259 BUG_ON(!in_softirq());
23260 @@ -4689,6 +4923,7 @@ int __sched __cond_resched_softirq(void)
23263 EXPORT_SYMBOL(__cond_resched_softirq);
23267 * yield - yield the current processor to other threads.
23268 @@ -5055,7 +5290,9 @@ void init_idle(struct task_struct *idle, int cpu)
23270 /* Set the preempt count _outside_ the spinlocks! */
23271 init_idle_preempt_count(idle, cpu);
23273 +#ifdef CONFIG_HAVE_PREEMPT_LAZY
23274 + task_thread_info(idle)->preempt_lazy_count = 0;
23277 * The idle tasks have their own, simple scheduling class:
23279 @@ -5196,6 +5433,8 @@ void sched_setnuma(struct task_struct *p, int nid)
23280 #endif /* CONFIG_NUMA_BALANCING */
23282 #ifdef CONFIG_HOTPLUG_CPU
23283 +static DEFINE_PER_CPU(struct mm_struct *, idle_last_mm);
23286 * Ensures that the idle task is using init_mm right before its cpu goes
23288 @@ -5210,7 +5449,11 @@ void idle_task_exit(void)
23289 switch_mm(mm, &init_mm, current);
23290 finish_arch_post_lock_switch();
23294 + * Defer the cleanup to an alive cpu. On RT we can neither
23295 + * call mmdrop() nor mmdrop_delayed() from here.
23297 + per_cpu(idle_last_mm, smp_processor_id()) = mm;
23301 @@ -5583,6 +5826,10 @@ migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu)
23304 calc_load_migrate(rq);
23305 + if (per_cpu(idle_last_mm, cpu)) {
23306 + mmdrop(per_cpu(idle_last_mm, cpu));
23307 + per_cpu(idle_last_mm, cpu) = NULL;
23312 @@ -7566,7 +7813,7 @@ void __init sched_init(void)
23313 #ifdef CONFIG_DEBUG_ATOMIC_SLEEP
23314 static inline int preempt_count_equals(int preempt_offset)
23316 - int nested = preempt_count() + rcu_preempt_depth();
23317 + int nested = preempt_count() + sched_rcu_preempt_depth();
23319 return (nested == preempt_offset);
23321 diff --git a/kernel/sched/cpudeadline.c b/kernel/sched/cpudeadline.c
23322 index 5a75b08cfd85..5be58820465c 100644
23323 --- a/kernel/sched/cpudeadline.c
23324 +++ b/kernel/sched/cpudeadline.c
23325 @@ -103,10 +103,10 @@ int cpudl_find(struct cpudl *cp, struct task_struct *p,
23326 const struct sched_dl_entity *dl_se = &p->dl;
23329 - cpumask_and(later_mask, cp->free_cpus, &p->cpus_allowed)) {
23330 + cpumask_and(later_mask, cp->free_cpus, tsk_cpus_allowed(p))) {
23331 best_cpu = cpumask_any(later_mask);
23333 - } else if (cpumask_test_cpu(cpudl_maximum(cp), &p->cpus_allowed) &&
23334 + } else if (cpumask_test_cpu(cpudl_maximum(cp), tsk_cpus_allowed(p)) &&
23335 dl_time_before(dl_se->deadline, cp->elements[0].dl)) {
23336 best_cpu = cpudl_maximum(cp);
23338 diff --git a/kernel/sched/cpupri.c b/kernel/sched/cpupri.c
23339 index 981fcd7dc394..11e9705bf937 100644
23340 --- a/kernel/sched/cpupri.c
23341 +++ b/kernel/sched/cpupri.c
23342 @@ -103,11 +103,11 @@ int cpupri_find(struct cpupri *cp, struct task_struct *p,
23346 - if (cpumask_any_and(&p->cpus_allowed, vec->mask) >= nr_cpu_ids)
23347 + if (cpumask_any_and(tsk_cpus_allowed(p), vec->mask) >= nr_cpu_ids)
23351 - cpumask_and(lowest_mask, &p->cpus_allowed, vec->mask);
23352 + cpumask_and(lowest_mask, tsk_cpus_allowed(p), vec->mask);
23355 * We have to ensure that we have at least one bit
23356 diff --git a/kernel/sched/cputime.c b/kernel/sched/cputime.c
23357 index a1aecbedf5b1..558b98af241d 100644
23358 --- a/kernel/sched/cputime.c
23359 +++ b/kernel/sched/cputime.c
23360 @@ -685,7 +685,7 @@ static cputime_t get_vtime_delta(struct task_struct *tsk)
23362 unsigned long long delta = vtime_delta(tsk);
23364 - WARN_ON_ONCE(tsk->vtime_snap_whence == VTIME_SLEEPING);
23365 + WARN_ON_ONCE(tsk->vtime_snap_whence == VTIME_INACTIVE);
23366 tsk->vtime_snap += delta;
23368 /* CHECKME: always safe to convert nsecs to cputime? */
23369 @@ -701,37 +701,37 @@ static void __vtime_account_system(struct task_struct *tsk)
23371 void vtime_account_system(struct task_struct *tsk)
23373 - write_seqlock(&tsk->vtime_seqlock);
23374 + write_seqcount_begin(&tsk->vtime_seqcount);
23375 __vtime_account_system(tsk);
23376 - write_sequnlock(&tsk->vtime_seqlock);
23377 + write_seqcount_end(&tsk->vtime_seqcount);
23380 void vtime_gen_account_irq_exit(struct task_struct *tsk)
23382 - write_seqlock(&tsk->vtime_seqlock);
23383 + write_seqcount_begin(&tsk->vtime_seqcount);
23384 __vtime_account_system(tsk);
23385 if (context_tracking_in_user())
23386 tsk->vtime_snap_whence = VTIME_USER;
23387 - write_sequnlock(&tsk->vtime_seqlock);
23388 + write_seqcount_end(&tsk->vtime_seqcount);
23391 void vtime_account_user(struct task_struct *tsk)
23393 cputime_t delta_cpu;
23395 - write_seqlock(&tsk->vtime_seqlock);
23396 + write_seqcount_begin(&tsk->vtime_seqcount);
23397 delta_cpu = get_vtime_delta(tsk);
23398 tsk->vtime_snap_whence = VTIME_SYS;
23399 account_user_time(tsk, delta_cpu, cputime_to_scaled(delta_cpu));
23400 - write_sequnlock(&tsk->vtime_seqlock);
23401 + write_seqcount_end(&tsk->vtime_seqcount);
23404 void vtime_user_enter(struct task_struct *tsk)
23406 - write_seqlock(&tsk->vtime_seqlock);
23407 + write_seqcount_begin(&tsk->vtime_seqcount);
23408 __vtime_account_system(tsk);
23409 tsk->vtime_snap_whence = VTIME_USER;
23410 - write_sequnlock(&tsk->vtime_seqlock);
23411 + write_seqcount_end(&tsk->vtime_seqcount);
23414 void vtime_guest_enter(struct task_struct *tsk)
23415 @@ -743,19 +743,19 @@ void vtime_guest_enter(struct task_struct *tsk)
23416 * synchronization against the reader (task_gtime())
23417 * that can thus safely catch up with a tickless delta.
23419 - write_seqlock(&tsk->vtime_seqlock);
23420 + write_seqcount_begin(&tsk->vtime_seqcount);
23421 __vtime_account_system(tsk);
23422 current->flags |= PF_VCPU;
23423 - write_sequnlock(&tsk->vtime_seqlock);
23424 + write_seqcount_end(&tsk->vtime_seqcount);
23426 EXPORT_SYMBOL_GPL(vtime_guest_enter);
23428 void vtime_guest_exit(struct task_struct *tsk)
23430 - write_seqlock(&tsk->vtime_seqlock);
23431 + write_seqcount_begin(&tsk->vtime_seqcount);
23432 __vtime_account_system(tsk);
23433 current->flags &= ~PF_VCPU;
23434 - write_sequnlock(&tsk->vtime_seqlock);
23435 + write_seqcount_end(&tsk->vtime_seqcount);
23437 EXPORT_SYMBOL_GPL(vtime_guest_exit);
23439 @@ -768,24 +768,26 @@ void vtime_account_idle(struct task_struct *tsk)
23441 void arch_vtime_task_switch(struct task_struct *prev)
23443 - write_seqlock(&prev->vtime_seqlock);
23444 - prev->vtime_snap_whence = VTIME_SLEEPING;
23445 - write_sequnlock(&prev->vtime_seqlock);
23446 + write_seqcount_begin(&prev->vtime_seqcount);
23447 + prev->vtime_snap_whence = VTIME_INACTIVE;
23448 + write_seqcount_end(&prev->vtime_seqcount);
23450 - write_seqlock(¤t->vtime_seqlock);
23451 + write_seqcount_begin(¤t->vtime_seqcount);
23452 current->vtime_snap_whence = VTIME_SYS;
23453 current->vtime_snap = sched_clock_cpu(smp_processor_id());
23454 - write_sequnlock(¤t->vtime_seqlock);
23455 + write_seqcount_end(¤t->vtime_seqcount);
23458 void vtime_init_idle(struct task_struct *t, int cpu)
23460 unsigned long flags;
23462 - write_seqlock_irqsave(&t->vtime_seqlock, flags);
23463 + local_irq_save(flags);
23464 + write_seqcount_begin(&t->vtime_seqcount);
23465 t->vtime_snap_whence = VTIME_SYS;
23466 t->vtime_snap = sched_clock_cpu(cpu);
23467 - write_sequnlock_irqrestore(&t->vtime_seqlock, flags);
23468 + write_seqcount_end(&t->vtime_seqcount);
23469 + local_irq_restore(flags);
23472 cputime_t task_gtime(struct task_struct *t)
23473 @@ -797,13 +799,13 @@ cputime_t task_gtime(struct task_struct *t)
23477 - seq = read_seqbegin(&t->vtime_seqlock);
23478 + seq = read_seqcount_begin(&t->vtime_seqcount);
23481 if (t->flags & PF_VCPU)
23482 gtime += vtime_delta(t);
23484 - } while (read_seqretry(&t->vtime_seqlock, seq));
23485 + } while (read_seqcount_retry(&t->vtime_seqcount, seq));
23489 @@ -826,7 +828,7 @@ fetch_task_cputime(struct task_struct *t,
23493 - seq = read_seqbegin(&t->vtime_seqlock);
23494 + seq = read_seqcount_begin(&t->vtime_seqcount);
23498 @@ -834,7 +836,7 @@ fetch_task_cputime(struct task_struct *t,
23501 /* Task is sleeping, nothing to add */
23502 - if (t->vtime_snap_whence == VTIME_SLEEPING ||
23503 + if (t->vtime_snap_whence == VTIME_INACTIVE ||
23507 @@ -850,7 +852,7 @@ fetch_task_cputime(struct task_struct *t,
23508 if (t->vtime_snap_whence == VTIME_SYS)
23511 - } while (read_seqretry(&t->vtime_seqlock, seq));
23512 + } while (read_seqcount_retry(&t->vtime_seqcount, seq));
23516 diff --git a/kernel/sched/deadline.c b/kernel/sched/deadline.c
23517 index e984f059e5fc..20e0c9b9ace5 100644
23518 --- a/kernel/sched/deadline.c
23519 +++ b/kernel/sched/deadline.c
23520 @@ -134,7 +134,7 @@ static void inc_dl_migration(struct sched_dl_entity *dl_se, struct dl_rq *dl_rq)
23522 struct task_struct *p = dl_task_of(dl_se);
23524 - if (p->nr_cpus_allowed > 1)
23525 + if (tsk_nr_cpus_allowed(p) > 1)
23526 dl_rq->dl_nr_migratory++;
23528 update_dl_migration(dl_rq);
23529 @@ -144,7 +144,7 @@ static void dec_dl_migration(struct sched_dl_entity *dl_se, struct dl_rq *dl_rq)
23531 struct task_struct *p = dl_task_of(dl_se);
23533 - if (p->nr_cpus_allowed > 1)
23534 + if (tsk_nr_cpus_allowed(p) > 1)
23535 dl_rq->dl_nr_migratory--;
23537 update_dl_migration(dl_rq);
23538 @@ -697,6 +697,7 @@ void init_dl_task_timer(struct sched_dl_entity *dl_se)
23540 hrtimer_init(timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
23541 timer->function = dl_task_timer;
23542 + timer->irqsafe = 1;
23546 @@ -989,7 +990,7 @@ static void enqueue_task_dl(struct rq *rq, struct task_struct *p, int flags)
23548 enqueue_dl_entity(&p->dl, pi_se, flags);
23550 - if (!task_current(rq, p) && p->nr_cpus_allowed > 1)
23551 + if (!task_current(rq, p) && tsk_nr_cpus_allowed(p) > 1)
23552 enqueue_pushable_dl_task(rq, p);
23555 @@ -1067,9 +1068,9 @@ select_task_rq_dl(struct task_struct *p, int cpu, int sd_flag, int flags)
23556 * try to make it stay here, it might be important.
23558 if (unlikely(dl_task(curr)) &&
23559 - (curr->nr_cpus_allowed < 2 ||
23560 + (tsk_nr_cpus_allowed(curr) < 2 ||
23561 !dl_entity_preempt(&p->dl, &curr->dl)) &&
23562 - (p->nr_cpus_allowed > 1)) {
23563 + (tsk_nr_cpus_allowed(p) > 1)) {
23564 int target = find_later_rq(p);
23566 if (target != -1 &&
23567 @@ -1090,7 +1091,7 @@ static void check_preempt_equal_dl(struct rq *rq, struct task_struct *p)
23568 * Current can't be migrated, useless to reschedule,
23569 * let's hope p can move out.
23571 - if (rq->curr->nr_cpus_allowed == 1 ||
23572 + if (tsk_nr_cpus_allowed(rq->curr) == 1 ||
23573 cpudl_find(&rq->rd->cpudl, rq->curr, NULL) == -1)
23576 @@ -1098,7 +1099,7 @@ static void check_preempt_equal_dl(struct rq *rq, struct task_struct *p)
23577 * p is migratable, so let's not schedule it and
23578 * see if it is pushed or pulled somewhere else.
23580 - if (p->nr_cpus_allowed != 1 &&
23581 + if (tsk_nr_cpus_allowed(p) != 1 &&
23582 cpudl_find(&rq->rd->cpudl, p, NULL) != -1)
23585 @@ -1212,7 +1213,7 @@ static void put_prev_task_dl(struct rq *rq, struct task_struct *p)
23587 update_curr_dl(rq);
23589 - if (on_dl_rq(&p->dl) && p->nr_cpus_allowed > 1)
23590 + if (on_dl_rq(&p->dl) && tsk_nr_cpus_allowed(p) > 1)
23591 enqueue_pushable_dl_task(rq, p);
23594 @@ -1335,7 +1336,7 @@ static int find_later_rq(struct task_struct *task)
23595 if (unlikely(!later_mask))
23598 - if (task->nr_cpus_allowed == 1)
23599 + if (tsk_nr_cpus_allowed(task) == 1)
23603 @@ -1441,7 +1442,7 @@ static struct rq *find_lock_later_rq(struct task_struct *task, struct rq *rq)
23604 if (double_lock_balance(rq, later_rq)) {
23605 if (unlikely(task_rq(task) != rq ||
23606 !cpumask_test_cpu(later_rq->cpu,
23607 - &task->cpus_allowed) ||
23608 + tsk_cpus_allowed(task)) ||
23609 task_running(rq, task) ||
23610 !task_on_rq_queued(task))) {
23611 double_unlock_balance(rq, later_rq);
23612 @@ -1480,7 +1481,7 @@ static struct task_struct *pick_next_pushable_dl_task(struct rq *rq)
23614 BUG_ON(rq->cpu != task_cpu(p));
23615 BUG_ON(task_current(rq, p));
23616 - BUG_ON(p->nr_cpus_allowed <= 1);
23617 + BUG_ON(tsk_nr_cpus_allowed(p) <= 1);
23619 BUG_ON(!task_on_rq_queued(p));
23620 BUG_ON(!dl_task(p));
23621 @@ -1519,7 +1520,7 @@ retry:
23623 if (dl_task(rq->curr) &&
23624 dl_time_before(next_task->dl.deadline, rq->curr->dl.deadline) &&
23625 - rq->curr->nr_cpus_allowed > 1) {
23626 + tsk_nr_cpus_allowed(rq->curr) > 1) {
23630 @@ -1666,9 +1667,9 @@ static void task_woken_dl(struct rq *rq, struct task_struct *p)
23632 if (!task_running(rq, p) &&
23633 !test_tsk_need_resched(rq->curr) &&
23634 - p->nr_cpus_allowed > 1 &&
23635 + tsk_nr_cpus_allowed(p) > 1 &&
23636 dl_task(rq->curr) &&
23637 - (rq->curr->nr_cpus_allowed < 2 ||
23638 + (tsk_nr_cpus_allowed(rq->curr) < 2 ||
23639 !dl_entity_preempt(&p->dl, &rq->curr->dl))) {
23642 @@ -1769,7 +1770,7 @@ static void switched_to_dl(struct rq *rq, struct task_struct *p)
23644 if (task_on_rq_queued(p) && rq->curr != p) {
23646 - if (p->nr_cpus_allowed > 1 && rq->dl.overloaded)
23647 + if (tsk_nr_cpus_allowed(p) > 1 && rq->dl.overloaded)
23648 queue_push_tasks(rq);
23650 if (dl_task(rq->curr))
23651 diff --git a/kernel/sched/debug.c b/kernel/sched/debug.c
23652 index 641511771ae6..a2d69b883623 100644
23653 --- a/kernel/sched/debug.c
23654 +++ b/kernel/sched/debug.c
23655 @@ -251,6 +251,9 @@ void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
23660 + P(rt_nr_migratory);
23665 @@ -635,6 +638,10 @@ void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
23669 +#ifdef CONFIG_PREEMPT_RT_FULL
23670 + P(migrate_disable);
23672 + P(nr_cpus_allowed);
23676 diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
23677 index 812069b66f47..ddf1424bcc78 100644
23678 --- a/kernel/sched/fair.c
23679 +++ b/kernel/sched/fair.c
23680 @@ -3166,7 +3166,7 @@ check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr)
23681 ideal_runtime = sched_slice(cfs_rq, curr);
23682 delta_exec = curr->sum_exec_runtime - curr->prev_sum_exec_runtime;
23683 if (delta_exec > ideal_runtime) {
23684 - resched_curr(rq_of(cfs_rq));
23685 + resched_curr_lazy(rq_of(cfs_rq));
23687 * The current task ran long enough, ensure it doesn't get
23688 * re-elected due to buddy favours.
23689 @@ -3190,7 +3190,7 @@ check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr)
23692 if (delta > ideal_runtime)
23693 - resched_curr(rq_of(cfs_rq));
23694 + resched_curr_lazy(rq_of(cfs_rq));
23698 @@ -3330,7 +3330,7 @@ entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr, int queued)
23699 * validating it and just reschedule.
23702 - resched_curr(rq_of(cfs_rq));
23703 + resched_curr_lazy(rq_of(cfs_rq));
23707 @@ -3512,7 +3512,7 @@ static void __account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec)
23708 * hierarchy can be throttled
23710 if (!assign_cfs_rq_runtime(cfs_rq) && likely(cfs_rq->curr))
23711 - resched_curr(rq_of(cfs_rq));
23712 + resched_curr_lazy(rq_of(cfs_rq));
23715 static __always_inline
23716 @@ -4144,7 +4144,7 @@ static void hrtick_start_fair(struct rq *rq, struct task_struct *p)
23720 - resched_curr(rq);
23721 + resched_curr_lazy(rq);
23724 hrtick_start(rq, delta);
23725 @@ -5232,7 +5232,7 @@ static void check_preempt_wakeup(struct rq *rq, struct task_struct *p, int wake_
23729 - resched_curr(rq);
23730 + resched_curr_lazy(rq);
23732 * Only set the backward buddy when the current task is still
23733 * on the rq. This can happen when a wakeup gets interleaved
23734 @@ -7983,7 +7983,7 @@ static void task_fork_fair(struct task_struct *p)
23735 * 'current' within the tree based on its new key value.
23737 swap(curr->vruntime, se->vruntime);
23738 - resched_curr(rq);
23739 + resched_curr_lazy(rq);
23742 se->vruntime -= cfs_rq->min_vruntime;
23743 @@ -8008,7 +8008,7 @@ prio_changed_fair(struct rq *rq, struct task_struct *p, int oldprio)
23745 if (rq->curr == p) {
23746 if (p->prio > oldprio)
23747 - resched_curr(rq);
23748 + resched_curr_lazy(rq);
23750 check_preempt_curr(rq, p, 0);
23752 diff --git a/kernel/sched/features.h b/kernel/sched/features.h
23753 index 69631fa46c2f..6d28fcd08872 100644
23754 --- a/kernel/sched/features.h
23755 +++ b/kernel/sched/features.h
23756 @@ -45,11 +45,19 @@ SCHED_FEAT(LB_BIAS, true)
23758 SCHED_FEAT(NONTASK_CAPACITY, true)
23760 +#ifdef CONFIG_PREEMPT_RT_FULL
23761 +SCHED_FEAT(TTWU_QUEUE, false)
23762 +# ifdef CONFIG_PREEMPT_LAZY
23763 +SCHED_FEAT(PREEMPT_LAZY, true)
23768 * Queue remote wakeups on the target CPU and process them
23769 * using the scheduler IPI. Reduces rq->lock contention/bounces.
23771 SCHED_FEAT(TTWU_QUEUE, true)
23774 #ifdef HAVE_RT_PUSH_IPI
23776 diff --git a/kernel/sched/rt.c b/kernel/sched/rt.c
23777 index 78ae5c1d9412..4ac6937f4a65 100644
23778 --- a/kernel/sched/rt.c
23779 +++ b/kernel/sched/rt.c
23780 @@ -47,6 +47,7 @@ void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime)
23782 hrtimer_init(&rt_b->rt_period_timer,
23783 CLOCK_MONOTONIC, HRTIMER_MODE_REL);
23784 + rt_b->rt_period_timer.irqsafe = 1;
23785 rt_b->rt_period_timer.function = sched_rt_period_timer;
23788 @@ -93,6 +94,7 @@ void init_rt_rq(struct rt_rq *rt_rq)
23789 rt_rq->push_cpu = nr_cpu_ids;
23790 raw_spin_lock_init(&rt_rq->push_lock);
23791 init_irq_work(&rt_rq->push_work, push_irq_work_func);
23792 + rt_rq->push_work.flags |= IRQ_WORK_HARD_IRQ;
23794 #endif /* CONFIG_SMP */
23795 /* We start is dequeued state, because no RT tasks are queued */
23796 @@ -326,7 +328,7 @@ static void inc_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
23797 rt_rq = &rq_of_rt_rq(rt_rq)->rt;
23799 rt_rq->rt_nr_total++;
23800 - if (p->nr_cpus_allowed > 1)
23801 + if (tsk_nr_cpus_allowed(p) > 1)
23802 rt_rq->rt_nr_migratory++;
23804 update_rt_migration(rt_rq);
23805 @@ -343,7 +345,7 @@ static void dec_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
23806 rt_rq = &rq_of_rt_rq(rt_rq)->rt;
23808 rt_rq->rt_nr_total--;
23809 - if (p->nr_cpus_allowed > 1)
23810 + if (tsk_nr_cpus_allowed(p) > 1)
23811 rt_rq->rt_nr_migratory--;
23813 update_rt_migration(rt_rq);
23814 @@ -1262,7 +1264,7 @@ enqueue_task_rt(struct rq *rq, struct task_struct *p, int flags)
23816 enqueue_rt_entity(rt_se, flags & ENQUEUE_HEAD);
23818 - if (!task_current(rq, p) && p->nr_cpus_allowed > 1)
23819 + if (!task_current(rq, p) && tsk_nr_cpus_allowed(p) > 1)
23820 enqueue_pushable_task(rq, p);
23823 @@ -1351,7 +1353,7 @@ select_task_rq_rt(struct task_struct *p, int cpu, int sd_flag, int flags)
23824 * will have to sort it out.
23826 if (curr && unlikely(rt_task(curr)) &&
23827 - (curr->nr_cpus_allowed < 2 ||
23828 + (tsk_nr_cpus_allowed(curr) < 2 ||
23829 curr->prio <= p->prio)) {
23830 int target = find_lowest_rq(p);
23832 @@ -1375,7 +1377,7 @@ static void check_preempt_equal_prio(struct rq *rq, struct task_struct *p)
23833 * Current can't be migrated, useless to reschedule,
23834 * let's hope p can move out.
23836 - if (rq->curr->nr_cpus_allowed == 1 ||
23837 + if (tsk_nr_cpus_allowed(rq->curr) == 1 ||
23838 !cpupri_find(&rq->rd->cpupri, rq->curr, NULL))
23841 @@ -1383,7 +1385,7 @@ static void check_preempt_equal_prio(struct rq *rq, struct task_struct *p)
23842 * p is migratable, so let's not schedule it and
23843 * see if it is pushed or pulled somewhere else.
23845 - if (p->nr_cpus_allowed != 1
23846 + if (tsk_nr_cpus_allowed(p) != 1
23847 && cpupri_find(&rq->rd->cpupri, p, NULL))
23850 @@ -1517,7 +1519,7 @@ static void put_prev_task_rt(struct rq *rq, struct task_struct *p)
23851 * The previous task needs to be made eligible for pushing
23852 * if it is still active
23854 - if (on_rt_rq(&p->rt) && p->nr_cpus_allowed > 1)
23855 + if (on_rt_rq(&p->rt) && tsk_nr_cpus_allowed(p) > 1)
23856 enqueue_pushable_task(rq, p);
23859 @@ -1567,7 +1569,7 @@ static int find_lowest_rq(struct task_struct *task)
23860 if (unlikely(!lowest_mask))
23863 - if (task->nr_cpus_allowed == 1)
23864 + if (tsk_nr_cpus_allowed(task) == 1)
23865 return -1; /* No other targets possible */
23867 if (!cpupri_find(&task_rq(task)->rd->cpupri, task, lowest_mask))
23868 @@ -1699,7 +1701,7 @@ static struct task_struct *pick_next_pushable_task(struct rq *rq)
23870 BUG_ON(rq->cpu != task_cpu(p));
23871 BUG_ON(task_current(rq, p));
23872 - BUG_ON(p->nr_cpus_allowed <= 1);
23873 + BUG_ON(tsk_nr_cpus_allowed(p) <= 1);
23875 BUG_ON(!task_on_rq_queued(p));
23876 BUG_ON(!rt_task(p));
23877 @@ -2059,9 +2061,9 @@ static void task_woken_rt(struct rq *rq, struct task_struct *p)
23879 if (!task_running(rq, p) &&
23880 !test_tsk_need_resched(rq->curr) &&
23881 - p->nr_cpus_allowed > 1 &&
23882 + tsk_nr_cpus_allowed(p) > 1 &&
23883 (dl_task(rq->curr) || rt_task(rq->curr)) &&
23884 - (rq->curr->nr_cpus_allowed < 2 ||
23885 + (tsk_nr_cpus_allowed(rq->curr) < 2 ||
23886 rq->curr->prio <= p->prio))
23889 @@ -2134,7 +2136,7 @@ static void switched_to_rt(struct rq *rq, struct task_struct *p)
23891 if (task_on_rq_queued(p) && rq->curr != p) {
23893 - if (p->nr_cpus_allowed > 1 && rq->rt.overloaded)
23894 + if (tsk_nr_cpus_allowed(p) > 1 && rq->rt.overloaded)
23895 queue_push_tasks(rq);
23896 #endif /* CONFIG_SMP */
23897 if (p->prio < rq->curr->prio)
23898 diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
23899 index 4e5db65d1aab..ae26df3bd63f 100644
23900 --- a/kernel/sched/sched.h
23901 +++ b/kernel/sched/sched.h
23902 @@ -1100,6 +1100,7 @@ static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
23903 #define WF_SYNC 0x01 /* waker goes to sleep after wakeup */
23904 #define WF_FORK 0x02 /* child wakeup after fork */
23905 #define WF_MIGRATED 0x4 /* internal use, task got migrated */
23906 +#define WF_LOCK_SLEEPER 0x08 /* wakeup spinlock "sleeper" */
23909 * To aid in avoiding the subversion of "niceness" due to uneven distribution
23910 @@ -1299,6 +1300,15 @@ extern void init_sched_fair_class(void);
23911 extern void resched_curr(struct rq *rq);
23912 extern void resched_cpu(int cpu);
23914 +#ifdef CONFIG_PREEMPT_LAZY
23915 +extern void resched_curr_lazy(struct rq *rq);
23917 +static inline void resched_curr_lazy(struct rq *rq)
23919 + resched_curr(rq);
23923 extern struct rt_bandwidth def_rt_bandwidth;
23924 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
23926 diff --git a/kernel/sched/swait.c b/kernel/sched/swait.c
23927 new file mode 100644
23928 index 000000000000..205fe36868f9
23930 +++ b/kernel/sched/swait.c
23932 +#include <linux/sched.h>
23933 +#include <linux/swait.h>
23934 +#include <linux/suspend.h>
23936 +void __init_swait_queue_head(struct swait_queue_head *q, const char *name,
23937 + struct lock_class_key *key)
23939 + raw_spin_lock_init(&q->lock);
23940 + lockdep_set_class_and_name(&q->lock, key, name);
23941 + INIT_LIST_HEAD(&q->task_list);
23943 +EXPORT_SYMBOL(__init_swait_queue_head);
23946 + * The thing about the wake_up_state() return value; I think we can ignore it.
23948 + * If for some reason it would return 0, that means the previously waiting
23949 + * task is already running, so it will observe condition true (or has already).
23951 +void swake_up_locked(struct swait_queue_head *q)
23953 + struct swait_queue *curr;
23955 + if (list_empty(&q->task_list))
23958 + curr = list_first_entry(&q->task_list, typeof(*curr), task_list);
23959 + wake_up_process(curr->task);
23960 + list_del_init(&curr->task_list);
23962 +EXPORT_SYMBOL(swake_up_locked);
23964 +void swake_up_all_locked(struct swait_queue_head *q)
23966 + struct swait_queue *curr;
23969 + while (!list_empty(&q->task_list)) {
23971 + curr = list_first_entry(&q->task_list, typeof(*curr),
23973 + wake_up_process(curr->task);
23974 + list_del_init(&curr->task_list);
23977 + if (pm_in_action)
23979 + WARN(wakes > 2, "complate_all() with %d waiters\n", wakes);
23981 +EXPORT_SYMBOL(swake_up_all_locked);
23983 +void swake_up(struct swait_queue_head *q)
23985 + unsigned long flags;
23987 + if (!swait_active(q))
23990 + raw_spin_lock_irqsave(&q->lock, flags);
23991 + swake_up_locked(q);
23992 + raw_spin_unlock_irqrestore(&q->lock, flags);
23994 +EXPORT_SYMBOL(swake_up);
23997 + * Does not allow usage from IRQ disabled, since we must be able to
23998 + * release IRQs to guarantee bounded hold time.
24000 +void swake_up_all(struct swait_queue_head *q)
24002 + struct swait_queue *curr;
24005 + if (!swait_active(q))
24008 + raw_spin_lock_irq(&q->lock);
24009 + list_splice_init(&q->task_list, &tmp);
24010 + while (!list_empty(&tmp)) {
24011 + curr = list_first_entry(&tmp, typeof(*curr), task_list);
24013 + wake_up_state(curr->task, TASK_NORMAL);
24014 + list_del_init(&curr->task_list);
24016 + if (list_empty(&tmp))
24019 + raw_spin_unlock_irq(&q->lock);
24020 + raw_spin_lock_irq(&q->lock);
24022 + raw_spin_unlock_irq(&q->lock);
24024 +EXPORT_SYMBOL(swake_up_all);
24026 +void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait)
24028 + wait->task = current;
24029 + if (list_empty(&wait->task_list))
24030 + list_add(&wait->task_list, &q->task_list);
24033 +void prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait, int state)
24035 + unsigned long flags;
24037 + raw_spin_lock_irqsave(&q->lock, flags);
24038 + __prepare_to_swait(q, wait);
24039 + set_current_state(state);
24040 + raw_spin_unlock_irqrestore(&q->lock, flags);
24042 +EXPORT_SYMBOL(prepare_to_swait);
24044 +long prepare_to_swait_event(struct swait_queue_head *q, struct swait_queue *wait, int state)
24046 + if (signal_pending_state(state, current))
24047 + return -ERESTARTSYS;
24049 + prepare_to_swait(q, wait, state);
24053 +EXPORT_SYMBOL(prepare_to_swait_event);
24055 +void __finish_swait(struct swait_queue_head *q, struct swait_queue *wait)
24057 + __set_current_state(TASK_RUNNING);
24058 + if (!list_empty(&wait->task_list))
24059 + list_del_init(&wait->task_list);
24062 +void finish_swait(struct swait_queue_head *q, struct swait_queue *wait)
24064 + unsigned long flags;
24066 + __set_current_state(TASK_RUNNING);
24068 + if (!list_empty_careful(&wait->task_list)) {
24069 + raw_spin_lock_irqsave(&q->lock, flags);
24070 + list_del_init(&wait->task_list);
24071 + raw_spin_unlock_irqrestore(&q->lock, flags);
24074 +EXPORT_SYMBOL(finish_swait);
24075 diff --git a/kernel/sched/swork.c b/kernel/sched/swork.c
24076 new file mode 100644
24077 index 000000000000..1950f40ca725
24079 +++ b/kernel/sched/swork.c
24082 + * Copyright (C) 2014 BMW Car IT GmbH, Daniel Wagner daniel.wagner@bmw-carit.de
24084 + * Provides a framework for enqueuing callbacks from irq context
24085 + * PREEMPT_RT_FULL safe. The callbacks are executed in kthread context.
24088 +#include <linux/swait.h>
24089 +#include <linux/swork.h>
24090 +#include <linux/kthread.h>
24091 +#include <linux/slab.h>
24092 +#include <linux/spinlock.h>
24093 +#include <linux/export.h>
24095 +#define SWORK_EVENT_PENDING (1 << 0)
24097 +static DEFINE_MUTEX(worker_mutex);
24098 +static struct sworker *glob_worker;
24101 + struct list_head events;
24102 + struct swait_queue_head wq;
24104 + raw_spinlock_t lock;
24106 + struct task_struct *task;
24110 +static bool swork_readable(struct sworker *worker)
24114 + if (kthread_should_stop())
24117 + raw_spin_lock_irq(&worker->lock);
24118 + r = !list_empty(&worker->events);
24119 + raw_spin_unlock_irq(&worker->lock);
24124 +static int swork_kthread(void *arg)
24126 + struct sworker *worker = arg;
24129 + swait_event_interruptible(worker->wq,
24130 + swork_readable(worker));
24131 + if (kthread_should_stop())
24134 + raw_spin_lock_irq(&worker->lock);
24135 + while (!list_empty(&worker->events)) {
24136 + struct swork_event *sev;
24138 + sev = list_first_entry(&worker->events,
24139 + struct swork_event, item);
24140 + list_del(&sev->item);
24141 + raw_spin_unlock_irq(&worker->lock);
24143 + WARN_ON_ONCE(!test_and_clear_bit(SWORK_EVENT_PENDING,
24146 + raw_spin_lock_irq(&worker->lock);
24148 + raw_spin_unlock_irq(&worker->lock);
24153 +static struct sworker *swork_create(void)
24155 + struct sworker *worker;
24157 + worker = kzalloc(sizeof(*worker), GFP_KERNEL);
24159 + return ERR_PTR(-ENOMEM);
24161 + INIT_LIST_HEAD(&worker->events);
24162 + raw_spin_lock_init(&worker->lock);
24163 + init_swait_queue_head(&worker->wq);
24165 + worker->task = kthread_run(swork_kthread, worker, "kswork");
24166 + if (IS_ERR(worker->task)) {
24168 + return ERR_PTR(-ENOMEM);
24174 +static void swork_destroy(struct sworker *worker)
24176 + kthread_stop(worker->task);
24178 + WARN_ON(!list_empty(&worker->events));
24183 + * swork_queue - queue swork
24185 + * Returns %false if @work was already on a queue, %true otherwise.
24187 + * The work is queued and processed on a random CPU
24189 +bool swork_queue(struct swork_event *sev)
24191 + unsigned long flags;
24193 + if (test_and_set_bit(SWORK_EVENT_PENDING, &sev->flags))
24196 + raw_spin_lock_irqsave(&glob_worker->lock, flags);
24197 + list_add_tail(&sev->item, &glob_worker->events);
24198 + raw_spin_unlock_irqrestore(&glob_worker->lock, flags);
24200 + swake_up(&glob_worker->wq);
24203 +EXPORT_SYMBOL_GPL(swork_queue);
24206 + * swork_get - get an instance of the sworker
24208 + * Returns an negative error code if the initialization if the worker did not
24209 + * work, %0 otherwise.
24212 +int swork_get(void)
24214 + struct sworker *worker;
24216 + mutex_lock(&worker_mutex);
24217 + if (!glob_worker) {
24218 + worker = swork_create();
24219 + if (IS_ERR(worker)) {
24220 + mutex_unlock(&worker_mutex);
24224 + glob_worker = worker;
24227 + glob_worker->refs++;
24228 + mutex_unlock(&worker_mutex);
24232 +EXPORT_SYMBOL_GPL(swork_get);
24235 + * swork_put - puts an instance of the sworker
24237 + * Will destroy the sworker thread. This function must not be called until all
24238 + * queued events have been completed.
24240 +void swork_put(void)
24242 + mutex_lock(&worker_mutex);
24244 + glob_worker->refs--;
24245 + if (glob_worker->refs > 0)
24248 + swork_destroy(glob_worker);
24249 + glob_worker = NULL;
24251 + mutex_unlock(&worker_mutex);
24253 +EXPORT_SYMBOL_GPL(swork_put);
24254 diff --git a/kernel/signal.c b/kernel/signal.c
24255 index f3f1f7a972fd..bc2c990f3f63 100644
24256 --- a/kernel/signal.c
24257 +++ b/kernel/signal.c
24259 #include <linux/export.h>
24260 #include <linux/init.h>
24261 #include <linux/sched.h>
24262 +#include <linux/sched/rt.h>
24263 #include <linux/fs.h>
24264 #include <linux/tty.h>
24265 #include <linux/binfmts.h>
24266 @@ -352,13 +353,30 @@ static bool task_participate_group_stop(struct task_struct *task)
24270 +static inline struct sigqueue *get_task_cache(struct task_struct *t)
24272 + struct sigqueue *q = t->sigqueue_cache;
24274 + if (cmpxchg(&t->sigqueue_cache, q, NULL) != q)
24279 +static inline int put_task_cache(struct task_struct *t, struct sigqueue *q)
24281 + if (cmpxchg(&t->sigqueue_cache, NULL, q) == NULL)
24287 * allocate a new signal queue record
24288 * - this may be called without locks if and only if t == current, otherwise an
24289 * appropriate lock must be held to stop the target task from exiting
24291 static struct sigqueue *
24292 -__sigqueue_alloc(int sig, struct task_struct *t, gfp_t flags, int override_rlimit)
24293 +__sigqueue_do_alloc(int sig, struct task_struct *t, gfp_t flags,
24294 + int override_rlimit, int fromslab)
24296 struct sigqueue *q = NULL;
24297 struct user_struct *user;
24298 @@ -375,7 +393,10 @@ __sigqueue_alloc(int sig, struct task_struct *t, gfp_t flags, int override_rlimi
24299 if (override_rlimit ||
24300 atomic_read(&user->sigpending) <=
24301 task_rlimit(t, RLIMIT_SIGPENDING)) {
24302 - q = kmem_cache_alloc(sigqueue_cachep, flags);
24304 + q = get_task_cache(t);
24306 + q = kmem_cache_alloc(sigqueue_cachep, flags);
24308 print_dropped_signal(sig);
24310 @@ -392,6 +413,13 @@ __sigqueue_alloc(int sig, struct task_struct *t, gfp_t flags, int override_rlimi
24314 +static struct sigqueue *
24315 +__sigqueue_alloc(int sig, struct task_struct *t, gfp_t flags,
24316 + int override_rlimit)
24318 + return __sigqueue_do_alloc(sig, t, flags, override_rlimit, 0);
24321 static void __sigqueue_free(struct sigqueue *q)
24323 if (q->flags & SIGQUEUE_PREALLOC)
24324 @@ -401,6 +429,21 @@ static void __sigqueue_free(struct sigqueue *q)
24325 kmem_cache_free(sigqueue_cachep, q);
24328 +static void sigqueue_free_current(struct sigqueue *q)
24330 + struct user_struct *up;
24332 + if (q->flags & SIGQUEUE_PREALLOC)
24336 + if (rt_prio(current->normal_prio) && !put_task_cache(current, q)) {
24337 + atomic_dec(&up->sigpending);
24340 + __sigqueue_free(q);
24343 void flush_sigqueue(struct sigpending *queue)
24345 struct sigqueue *q;
24346 @@ -414,6 +457,21 @@ void flush_sigqueue(struct sigpending *queue)
24350 + * Called from __exit_signal. Flush tsk->pending and
24351 + * tsk->sigqueue_cache
24353 +void flush_task_sigqueue(struct task_struct *tsk)
24355 + struct sigqueue *q;
24357 + flush_sigqueue(&tsk->pending);
24359 + q = get_task_cache(tsk);
24361 + kmem_cache_free(sigqueue_cachep, q);
24365 * Flush all pending signals for this kthread.
24367 void flush_signals(struct task_struct *t)
24368 @@ -525,7 +583,7 @@ static void collect_signal(int sig, struct sigpending *list, siginfo_t *info)
24370 list_del_init(&first->list);
24371 copy_siginfo(info, &first->info);
24372 - __sigqueue_free(first);
24373 + sigqueue_free_current(first);
24376 * Ok, it wasn't in the queue. This must be
24377 @@ -560,6 +618,8 @@ int dequeue_signal(struct task_struct *tsk, sigset_t *mask, siginfo_t *info)
24381 + WARN_ON_ONCE(tsk != current);
24383 /* We only dequeue private signals from ourselves, we don't let
24384 * signalfd steal them
24386 @@ -1156,8 +1216,8 @@ int do_send_sig_info(int sig, struct siginfo *info, struct task_struct *p,
24387 * We don't want to have recursive SIGSEGV's etc, for example,
24388 * that is why we also clear SIGNAL_UNKILLABLE.
24391 -force_sig_info(int sig, struct siginfo *info, struct task_struct *t)
24393 +do_force_sig_info(int sig, struct siginfo *info, struct task_struct *t)
24395 unsigned long int flags;
24396 int ret, blocked, ignored;
24397 @@ -1182,6 +1242,39 @@ force_sig_info(int sig, struct siginfo *info, struct task_struct *t)
24401 +int force_sig_info(int sig, struct siginfo *info, struct task_struct *t)
24404 + * On some archs, PREEMPT_RT has to delay sending a signal from a trap
24405 + * since it can not enable preemption, and the signal code's spin_locks
24406 + * turn into mutexes. Instead, it must set TIF_NOTIFY_RESUME which will
24407 + * send the signal on exit of the trap.
24409 +#ifdef ARCH_RT_DELAYS_SIGNAL_SEND
24410 + if (in_atomic()) {
24411 + if (WARN_ON_ONCE(t != current))
24413 + if (WARN_ON_ONCE(t->forced_info.si_signo))
24416 + if (is_si_special(info)) {
24417 + WARN_ON_ONCE(info != SEND_SIG_PRIV);
24418 + t->forced_info.si_signo = sig;
24419 + t->forced_info.si_errno = 0;
24420 + t->forced_info.si_code = SI_KERNEL;
24421 + t->forced_info.si_pid = 0;
24422 + t->forced_info.si_uid = 0;
24424 + t->forced_info = *info;
24427 + set_tsk_thread_flag(t, TIF_NOTIFY_RESUME);
24431 + return do_force_sig_info(sig, info, t);
24435 * Nuke all other threads in the group.
24437 @@ -1216,12 +1309,12 @@ struct sighand_struct *__lock_task_sighand(struct task_struct *tsk,
24438 * Disable interrupts early to avoid deadlocks.
24439 * See rcu_read_unlock() comment header for details.
24441 - local_irq_save(*flags);
24442 + local_irq_save_nort(*flags);
24444 sighand = rcu_dereference(tsk->sighand);
24445 if (unlikely(sighand == NULL)) {
24447 - local_irq_restore(*flags);
24448 + local_irq_restore_nort(*flags);
24452 @@ -1242,7 +1335,7 @@ struct sighand_struct *__lock_task_sighand(struct task_struct *tsk,
24454 spin_unlock(&sighand->siglock);
24456 - local_irq_restore(*flags);
24457 + local_irq_restore_nort(*flags);
24461 @@ -1485,7 +1578,8 @@ EXPORT_SYMBOL(kill_pid);
24463 struct sigqueue *sigqueue_alloc(void)
24465 - struct sigqueue *q = __sigqueue_alloc(-1, current, GFP_KERNEL, 0);
24466 + /* Preallocated sigqueue objects always from the slabcache ! */
24467 + struct sigqueue *q = __sigqueue_do_alloc(-1, current, GFP_KERNEL, 0, 1);
24470 q->flags |= SIGQUEUE_PREALLOC;
24471 @@ -1846,15 +1940,7 @@ static void ptrace_stop(int exit_code, int why, int clear_code, siginfo_t *info)
24472 if (gstop_done && ptrace_reparented(current))
24473 do_notify_parent_cldstop(current, false, why);
24476 - * Don't want to allow preemption here, because
24477 - * sys_ptrace() needs this task to be inactive.
24479 - * XXX: implement read_unlock_no_resched().
24481 - preempt_disable();
24482 read_unlock(&tasklist_lock);
24483 - preempt_enable_no_resched();
24484 freezable_schedule();
24487 diff --git a/kernel/softirq.c b/kernel/softirq.c
24488 index 479e4436f787..cb9c1d5dee10 100644
24489 --- a/kernel/softirq.c
24490 +++ b/kernel/softirq.c
24491 @@ -21,10 +21,12 @@
24492 #include <linux/freezer.h>
24493 #include <linux/kthread.h>
24494 #include <linux/rcupdate.h>
24495 +#include <linux/delay.h>
24496 #include <linux/ftrace.h>
24497 #include <linux/smp.h>
24498 #include <linux/smpboot.h>
24499 #include <linux/tick.h>
24500 +#include <linux/locallock.h>
24501 #include <linux/irq.h>
24503 #define CREATE_TRACE_POINTS
24504 @@ -56,12 +58,108 @@ EXPORT_SYMBOL(irq_stat);
24505 static struct softirq_action softirq_vec[NR_SOFTIRQS] __cacheline_aligned_in_smp;
24507 DEFINE_PER_CPU(struct task_struct *, ksoftirqd);
24508 +#ifdef CONFIG_PREEMPT_RT_FULL
24509 +#define TIMER_SOFTIRQS ((1 << TIMER_SOFTIRQ) | (1 << HRTIMER_SOFTIRQ))
24510 +DEFINE_PER_CPU(struct task_struct *, ktimer_softirqd);
24513 const char * const softirq_to_name[NR_SOFTIRQS] = {
24514 "HI", "TIMER", "NET_TX", "NET_RX", "BLOCK", "BLOCK_IOPOLL",
24515 "TASKLET", "SCHED", "HRTIMER", "RCU"
24518 +#ifdef CONFIG_NO_HZ_COMMON
24519 +# ifdef CONFIG_PREEMPT_RT_FULL
24521 +struct softirq_runner {
24522 + struct task_struct *runner[NR_SOFTIRQS];
24525 +static DEFINE_PER_CPU(struct softirq_runner, softirq_runners);
24527 +static inline void softirq_set_runner(unsigned int sirq)
24529 + struct softirq_runner *sr = this_cpu_ptr(&softirq_runners);
24531 + sr->runner[sirq] = current;
24534 +static inline void softirq_clr_runner(unsigned int sirq)
24536 + struct softirq_runner *sr = this_cpu_ptr(&softirq_runners);
24538 + sr->runner[sirq] = NULL;
24542 + * On preempt-rt a softirq running context might be blocked on a
24543 + * lock. There might be no other runnable task on this CPU because the
24544 + * lock owner runs on some other CPU. So we have to go into idle with
24545 + * the pending bit set. Therefor we need to check this otherwise we
24546 + * warn about false positives which confuses users and defeats the
24547 + * whole purpose of this test.
24549 + * This code is called with interrupts disabled.
24551 +void softirq_check_pending_idle(void)
24553 + static int rate_limit;
24554 + struct softirq_runner *sr = this_cpu_ptr(&softirq_runners);
24558 + if (rate_limit >= 10)
24561 + warnpending = local_softirq_pending() & SOFTIRQ_STOP_IDLE_MASK;
24562 + for (i = 0; i < NR_SOFTIRQS; i++) {
24563 + struct task_struct *tsk = sr->runner[i];
24566 + * The wakeup code in rtmutex.c wakes up the task
24567 + * _before_ it sets pi_blocked_on to NULL under
24568 + * tsk->pi_lock. So we need to check for both: state
24569 + * and pi_blocked_on.
24572 + raw_spin_lock(&tsk->pi_lock);
24573 + if (tsk->pi_blocked_on || tsk->state == TASK_RUNNING) {
24574 + /* Clear all bits pending in that task */
24575 + warnpending &= ~(tsk->softirqs_raised);
24576 + warnpending &= ~(1 << i);
24578 + raw_spin_unlock(&tsk->pi_lock);
24582 + if (warnpending) {
24583 + printk(KERN_ERR "NOHZ: local_softirq_pending %02x\n",
24590 + * On !PREEMPT_RT we just printk rate limited:
24592 +void softirq_check_pending_idle(void)
24594 + static int rate_limit;
24596 + if (rate_limit < 10 &&
24597 + (local_softirq_pending() & SOFTIRQ_STOP_IDLE_MASK)) {
24598 + printk(KERN_ERR "NOHZ: local_softirq_pending %02x\n",
24599 + local_softirq_pending());
24605 +#else /* !CONFIG_NO_HZ_COMMON */
24606 +static inline void softirq_set_runner(unsigned int sirq) { }
24607 +static inline void softirq_clr_runner(unsigned int sirq) { }
24611 * we cannot loop indefinitely here to avoid userspace starvation,
24612 * but we also don't want to introduce a worst case 1/HZ latency
24613 @@ -77,6 +175,79 @@ static void wakeup_softirqd(void)
24614 wake_up_process(tsk);
24617 +#ifdef CONFIG_PREEMPT_RT_FULL
24618 +static void wakeup_timer_softirqd(void)
24620 + /* Interrupts are disabled: no need to stop preemption */
24621 + struct task_struct *tsk = __this_cpu_read(ktimer_softirqd);
24623 + if (tsk && tsk->state != TASK_RUNNING)
24624 + wake_up_process(tsk);
24628 +static void handle_softirq(unsigned int vec_nr)
24630 + struct softirq_action *h = softirq_vec + vec_nr;
24633 + prev_count = preempt_count();
24635 + kstat_incr_softirqs_this_cpu(vec_nr);
24637 + trace_softirq_entry(vec_nr);
24639 + trace_softirq_exit(vec_nr);
24640 + if (unlikely(prev_count != preempt_count())) {
24641 + pr_err("huh, entered softirq %u %s %p with preempt_count %08x, exited with %08x?\n",
24642 + vec_nr, softirq_to_name[vec_nr], h->action,
24643 + prev_count, preempt_count());
24644 + preempt_count_set(prev_count);
24648 +#ifndef CONFIG_PREEMPT_RT_FULL
24649 +static inline int ksoftirqd_softirq_pending(void)
24651 + return local_softirq_pending();
24654 +static void handle_pending_softirqs(u32 pending)
24656 + struct softirq_action *h = softirq_vec;
24659 + local_irq_enable();
24663 + while ((softirq_bit = ffs(pending))) {
24664 + unsigned int vec_nr;
24666 + h += softirq_bit - 1;
24667 + vec_nr = h - softirq_vec;
24668 + handle_softirq(vec_nr);
24671 + pending >>= softirq_bit;
24675 + local_irq_disable();
24678 +static void run_ksoftirqd(unsigned int cpu)
24680 + local_irq_disable();
24681 + if (ksoftirqd_softirq_pending()) {
24683 + local_irq_enable();
24684 + cond_resched_rcu_qs();
24687 + local_irq_enable();
24691 * preempt_count and SOFTIRQ_OFFSET usage:
24692 * - preempt_count is changed by SOFTIRQ_OFFSET on entering or leaving
24693 @@ -116,9 +287,9 @@ void __local_bh_disable_ip(unsigned long ip, unsigned int cnt)
24695 if (preempt_count() == cnt) {
24696 #ifdef CONFIG_DEBUG_PREEMPT
24697 - current->preempt_disable_ip = get_parent_ip(CALLER_ADDR1);
24698 + current->preempt_disable_ip = get_lock_parent_ip();
24700 - trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1));
24701 + trace_preempt_off(CALLER_ADDR0, get_lock_parent_ip());
24704 EXPORT_SYMBOL(__local_bh_disable_ip);
24705 @@ -232,10 +403,8 @@ asmlinkage __visible void __do_softirq(void)
24706 unsigned long end = jiffies + MAX_SOFTIRQ_TIME;
24707 unsigned long old_flags = current->flags;
24708 int max_restart = MAX_SOFTIRQ_RESTART;
24709 - struct softirq_action *h;
24715 * Mask out PF_MEMALLOC s current task context is borrowed for the
24716 @@ -254,36 +423,7 @@ restart:
24717 /* Reset the pending bitmask before enabling irqs */
24718 set_softirq_pending(0);
24720 - local_irq_enable();
24724 - while ((softirq_bit = ffs(pending))) {
24725 - unsigned int vec_nr;
24728 - h += softirq_bit - 1;
24730 - vec_nr = h - softirq_vec;
24731 - prev_count = preempt_count();
24733 - kstat_incr_softirqs_this_cpu(vec_nr);
24735 - trace_softirq_entry(vec_nr);
24737 - trace_softirq_exit(vec_nr);
24738 - if (unlikely(prev_count != preempt_count())) {
24739 - pr_err("huh, entered softirq %u %s %p with preempt_count %08x, exited with %08x?\n",
24740 - vec_nr, softirq_to_name[vec_nr], h->action,
24741 - prev_count, preempt_count());
24742 - preempt_count_set(prev_count);
24745 - pending >>= softirq_bit;
24749 - local_irq_disable();
24750 + handle_pending_softirqs(pending);
24752 pending = local_softirq_pending();
24754 @@ -320,6 +460,310 @@ asmlinkage __visible void do_softirq(void)
24758 + * This function must run with irqs disabled!
24760 +void raise_softirq_irqoff(unsigned int nr)
24762 + __raise_softirq_irqoff(nr);
24765 + * If we're in an interrupt or softirq, we're done
24766 + * (this also catches softirq-disabled code). We will
24767 + * actually run the softirq once we return from
24768 + * the irq or softirq.
24770 + * Otherwise we wake up ksoftirqd to make sure we
24771 + * schedule the softirq soon.
24773 + if (!in_interrupt())
24774 + wakeup_softirqd();
24777 +void __raise_softirq_irqoff(unsigned int nr)
24779 + trace_softirq_raise(nr);
24780 + or_softirq_pending(1UL << nr);
24783 +static inline void local_bh_disable_nort(void) { local_bh_disable(); }
24784 +static inline void _local_bh_enable_nort(void) { _local_bh_enable(); }
24785 +static void ksoftirqd_set_sched_params(unsigned int cpu) { }
24787 +#else /* !PREEMPT_RT_FULL */
24790 + * On RT we serialize softirq execution with a cpu local lock per softirq
24792 +static DEFINE_PER_CPU(struct local_irq_lock [NR_SOFTIRQS], local_softirq_locks);
24794 +void __init softirq_early_init(void)
24798 + for (i = 0; i < NR_SOFTIRQS; i++)
24799 + local_irq_lock_init(local_softirq_locks[i]);
24802 +static void lock_softirq(int which)
24804 + local_lock(local_softirq_locks[which]);
24807 +static void unlock_softirq(int which)
24809 + local_unlock(local_softirq_locks[which]);
24812 +static void do_single_softirq(int which)
24814 + unsigned long old_flags = current->flags;
24816 + current->flags &= ~PF_MEMALLOC;
24817 + vtime_account_irq_enter(current);
24818 + current->flags |= PF_IN_SOFTIRQ;
24819 + lockdep_softirq_enter();
24820 + local_irq_enable();
24821 + handle_softirq(which);
24822 + local_irq_disable();
24823 + lockdep_softirq_exit();
24824 + current->flags &= ~PF_IN_SOFTIRQ;
24825 + vtime_account_irq_enter(current);
24826 + tsk_restore_flags(current, old_flags, PF_MEMALLOC);
24830 + * Called with interrupts disabled. Process softirqs which were raised
24831 + * in current context (or on behalf of ksoftirqd).
24833 +static void do_current_softirqs(void)
24835 + while (current->softirqs_raised) {
24836 + int i = __ffs(current->softirqs_raised);
24837 + unsigned int pending, mask = (1U << i);
24839 + current->softirqs_raised &= ~mask;
24840 + local_irq_enable();
24843 + * If the lock is contended, we boost the owner to
24844 + * process the softirq or leave the critical section
24848 + local_irq_disable();
24849 + softirq_set_runner(i);
24851 + * Check with the local_softirq_pending() bits,
24852 + * whether we need to process this still or if someone
24853 + * else took care of it.
24855 + pending = local_softirq_pending();
24856 + if (pending & mask) {
24857 + set_softirq_pending(pending & ~mask);
24858 + do_single_softirq(i);
24860 + softirq_clr_runner(i);
24861 + WARN_ON(current->softirq_nestcnt != 1);
24862 + local_irq_enable();
24863 + unlock_softirq(i);
24864 + local_irq_disable();
24868 +void __local_bh_disable(void)
24870 + if (++current->softirq_nestcnt == 1)
24871 + migrate_disable();
24873 +EXPORT_SYMBOL(__local_bh_disable);
24875 +void __local_bh_enable(void)
24877 + if (WARN_ON(current->softirq_nestcnt == 0))
24880 + local_irq_disable();
24881 + if (current->softirq_nestcnt == 1 && current->softirqs_raised)
24882 + do_current_softirqs();
24883 + local_irq_enable();
24885 + if (--current->softirq_nestcnt == 0)
24886 + migrate_enable();
24888 +EXPORT_SYMBOL(__local_bh_enable);
24890 +void _local_bh_enable(void)
24892 + if (WARN_ON(current->softirq_nestcnt == 0))
24894 + if (--current->softirq_nestcnt == 0)
24895 + migrate_enable();
24897 +EXPORT_SYMBOL(_local_bh_enable);
24899 +int in_serving_softirq(void)
24901 + return current->flags & PF_IN_SOFTIRQ;
24903 +EXPORT_SYMBOL(in_serving_softirq);
24905 +/* Called with preemption disabled */
24906 +static void run_ksoftirqd(unsigned int cpu)
24908 + local_irq_disable();
24909 + current->softirq_nestcnt++;
24911 + do_current_softirqs();
24912 + current->softirq_nestcnt--;
24913 + local_irq_enable();
24914 + cond_resched_rcu_qs();
24918 + * Called from netif_rx_ni(). Preemption enabled, but migration
24919 + * disabled. So the cpu can't go away under us.
24921 +void thread_do_softirq(void)
24923 + if (!in_serving_softirq() && current->softirqs_raised) {
24924 + current->softirq_nestcnt++;
24925 + do_current_softirqs();
24926 + current->softirq_nestcnt--;
24930 +static void do_raise_softirq_irqoff(unsigned int nr)
24932 + unsigned int mask;
24934 + mask = 1UL << nr;
24936 + trace_softirq_raise(nr);
24937 + or_softirq_pending(mask);
24940 + * If we are not in a hard interrupt and inside a bh disabled
24941 + * region, we simply raise the flag on current. local_bh_enable()
24942 + * will make sure that the softirq is executed. Otherwise we
24943 + * delegate it to ksoftirqd.
24945 + if (!in_irq() && current->softirq_nestcnt)
24946 + current->softirqs_raised |= mask;
24947 + else if (!__this_cpu_read(ksoftirqd) || !__this_cpu_read(ktimer_softirqd))
24950 + if (mask & TIMER_SOFTIRQS)
24951 + __this_cpu_read(ktimer_softirqd)->softirqs_raised |= mask;
24953 + __this_cpu_read(ksoftirqd)->softirqs_raised |= mask;
24956 +static void wakeup_proper_softirq(unsigned int nr)
24958 + if ((1UL << nr) & TIMER_SOFTIRQS)
24959 + wakeup_timer_softirqd();
24961 + wakeup_softirqd();
24965 +void __raise_softirq_irqoff(unsigned int nr)
24967 + do_raise_softirq_irqoff(nr);
24968 + if (!in_irq() && !current->softirq_nestcnt)
24969 + wakeup_proper_softirq(nr);
24973 + * Same as __raise_softirq_irqoff() but will process them in ksoftirqd
24975 +void __raise_softirq_irqoff_ksoft(unsigned int nr)
24977 + unsigned int mask;
24979 + if (WARN_ON_ONCE(!__this_cpu_read(ksoftirqd) ||
24980 + !__this_cpu_read(ktimer_softirqd)))
24982 + mask = 1UL << nr;
24984 + trace_softirq_raise(nr);
24985 + or_softirq_pending(mask);
24986 + if (mask & TIMER_SOFTIRQS)
24987 + __this_cpu_read(ktimer_softirqd)->softirqs_raised |= mask;
24989 + __this_cpu_read(ksoftirqd)->softirqs_raised |= mask;
24990 + wakeup_proper_softirq(nr);
24994 + * This function must run with irqs disabled!
24996 +void raise_softirq_irqoff(unsigned int nr)
24998 + do_raise_softirq_irqoff(nr);
25001 + * If we're in an hard interrupt we let irq return code deal
25002 + * with the wakeup of ksoftirqd.
25007 + * If we are in thread context but outside of a bh disabled
25008 + * region, we need to wake ksoftirqd as well.
25010 + * CHECKME: Some of the places which do that could be wrapped
25011 + * into local_bh_disable/enable pairs. Though it's unclear
25012 + * whether this is worth the effort. To find those places just
25013 + * raise a WARN() if the condition is met.
25015 + if (!current->softirq_nestcnt)
25016 + wakeup_proper_softirq(nr);
25019 +static inline int ksoftirqd_softirq_pending(void)
25021 + return current->softirqs_raised;
25024 +static inline void local_bh_disable_nort(void) { }
25025 +static inline void _local_bh_enable_nort(void) { }
25027 +static inline void ksoftirqd_set_sched_params(unsigned int cpu)
25029 + /* Take over all but timer pending softirqs when starting */
25030 + local_irq_disable();
25031 + current->softirqs_raised = local_softirq_pending() & ~TIMER_SOFTIRQS;
25032 + local_irq_enable();
25035 +static inline void ktimer_softirqd_set_sched_params(unsigned int cpu)
25037 + struct sched_param param = { .sched_priority = 1 };
25039 + sched_setscheduler(current, SCHED_FIFO, ¶m);
25041 + /* Take over timer pending softirqs when starting */
25042 + local_irq_disable();
25043 + current->softirqs_raised = local_softirq_pending() & TIMER_SOFTIRQS;
25044 + local_irq_enable();
25047 +static inline void ktimer_softirqd_clr_sched_params(unsigned int cpu,
25050 + struct sched_param param = { .sched_priority = 0 };
25052 + sched_setscheduler(current, SCHED_NORMAL, ¶m);
25055 +static int ktimer_softirqd_should_run(unsigned int cpu)
25057 + return current->softirqs_raised;
25060 +#endif /* PREEMPT_RT_FULL */
25062 * Enter an interrupt context.
25064 void irq_enter(void)
25065 @@ -330,9 +774,9 @@ void irq_enter(void)
25066 * Prevent raise_softirq from needlessly waking up ksoftirqd
25067 * here, as softirq will be serviced on return from interrupt.
25069 - local_bh_disable();
25070 + local_bh_disable_nort();
25072 - _local_bh_enable();
25073 + _local_bh_enable_nort();
25077 @@ -340,6 +784,7 @@ void irq_enter(void)
25079 static inline void invoke_softirq(void)
25081 +#ifndef CONFIG_PREEMPT_RT_FULL
25082 if (!force_irqthreads) {
25083 #ifdef CONFIG_HAVE_IRQ_EXIT_ON_IRQ_STACK
25085 @@ -359,6 +804,18 @@ static inline void invoke_softirq(void)
25089 +#else /* PREEMPT_RT_FULL */
25090 + unsigned long flags;
25092 + local_irq_save(flags);
25093 + if (__this_cpu_read(ksoftirqd) &&
25094 + __this_cpu_read(ksoftirqd)->softirqs_raised)
25095 + wakeup_softirqd();
25096 + if (__this_cpu_read(ktimer_softirqd) &&
25097 + __this_cpu_read(ktimer_softirqd)->softirqs_raised)
25098 + wakeup_timer_softirqd();
25099 + local_irq_restore(flags);
25103 static inline void tick_irq_exit(void)
25104 @@ -395,26 +852,6 @@ void irq_exit(void)
25105 trace_hardirq_exit(); /* must be last! */
25109 - * This function must run with irqs disabled!
25111 -inline void raise_softirq_irqoff(unsigned int nr)
25113 - __raise_softirq_irqoff(nr);
25116 - * If we're in an interrupt or softirq, we're done
25117 - * (this also catches softirq-disabled code). We will
25118 - * actually run the softirq once we return from
25119 - * the irq or softirq.
25121 - * Otherwise we wake up ksoftirqd to make sure we
25122 - * schedule the softirq soon.
25124 - if (!in_interrupt())
25125 - wakeup_softirqd();
25128 void raise_softirq(unsigned int nr)
25130 unsigned long flags;
25131 @@ -424,12 +861,6 @@ void raise_softirq(unsigned int nr)
25132 local_irq_restore(flags);
25135 -void __raise_softirq_irqoff(unsigned int nr)
25137 - trace_softirq_raise(nr);
25138 - or_softirq_pending(1UL << nr);
25141 void open_softirq(int nr, void (*action)(struct softirq_action *))
25143 softirq_vec[nr].action = action;
25144 @@ -446,15 +877,45 @@ struct tasklet_head {
25145 static DEFINE_PER_CPU(struct tasklet_head, tasklet_vec);
25146 static DEFINE_PER_CPU(struct tasklet_head, tasklet_hi_vec);
25148 +static void inline
25149 +__tasklet_common_schedule(struct tasklet_struct *t, struct tasklet_head *head, unsigned int nr)
25151 + if (tasklet_trylock(t)) {
25153 + /* We may have been preempted before tasklet_trylock
25154 + * and __tasklet_action may have already run.
25155 + * So double check the sched bit while the takslet
25156 + * is locked before adding it to the list.
25158 + if (test_bit(TASKLET_STATE_SCHED, &t->state)) {
25161 + head->tail = &(t->next);
25162 + raise_softirq_irqoff(nr);
25163 + tasklet_unlock(t);
25165 + /* This is subtle. If we hit the corner case above
25166 + * It is possible that we get preempted right here,
25167 + * and another task has successfully called
25168 + * tasklet_schedule(), then this function, and
25169 + * failed on the trylock. Thus we must be sure
25170 + * before releasing the tasklet lock, that the
25171 + * SCHED_BIT is clear. Otherwise the tasklet
25172 + * may get its SCHED_BIT set, but not added to the
25175 + if (!tasklet_tryunlock(t))
25181 void __tasklet_schedule(struct tasklet_struct *t)
25183 unsigned long flags;
25185 local_irq_save(flags);
25187 - *__this_cpu_read(tasklet_vec.tail) = t;
25188 - __this_cpu_write(tasklet_vec.tail, &(t->next));
25189 - raise_softirq_irqoff(TASKLET_SOFTIRQ);
25190 + __tasklet_common_schedule(t, this_cpu_ptr(&tasklet_vec), TASKLET_SOFTIRQ);
25191 local_irq_restore(flags);
25193 EXPORT_SYMBOL(__tasklet_schedule);
25194 @@ -464,10 +925,7 @@ void __tasklet_hi_schedule(struct tasklet_struct *t)
25195 unsigned long flags;
25197 local_irq_save(flags);
25199 - *__this_cpu_read(tasklet_hi_vec.tail) = t;
25200 - __this_cpu_write(tasklet_hi_vec.tail, &(t->next));
25201 - raise_softirq_irqoff(HI_SOFTIRQ);
25202 + __tasklet_common_schedule(t, this_cpu_ptr(&tasklet_hi_vec), HI_SOFTIRQ);
25203 local_irq_restore(flags);
25205 EXPORT_SYMBOL(__tasklet_hi_schedule);
25206 @@ -476,82 +934,122 @@ void __tasklet_hi_schedule_first(struct tasklet_struct *t)
25208 BUG_ON(!irqs_disabled());
25210 - t->next = __this_cpu_read(tasklet_hi_vec.head);
25211 - __this_cpu_write(tasklet_hi_vec.head, t);
25212 - __raise_softirq_irqoff(HI_SOFTIRQ);
25213 + __tasklet_hi_schedule(t);
25215 EXPORT_SYMBOL(__tasklet_hi_schedule_first);
25217 -static void tasklet_action(struct softirq_action *a)
25218 +void tasklet_enable(struct tasklet_struct *t)
25220 - struct tasklet_struct *list;
25221 + if (!atomic_dec_and_test(&t->count))
25223 + if (test_and_clear_bit(TASKLET_STATE_PENDING, &t->state))
25224 + tasklet_schedule(t);
25226 +EXPORT_SYMBOL(tasklet_enable);
25228 - local_irq_disable();
25229 - list = __this_cpu_read(tasklet_vec.head);
25230 - __this_cpu_write(tasklet_vec.head, NULL);
25231 - __this_cpu_write(tasklet_vec.tail, this_cpu_ptr(&tasklet_vec.head));
25232 - local_irq_enable();
25233 +static void __tasklet_action(struct softirq_action *a,
25234 + struct tasklet_struct *list)
25236 + int loops = 1000000;
25239 struct tasklet_struct *t = list;
25243 - if (tasklet_trylock(t)) {
25244 - if (!atomic_read(&t->count)) {
25245 - if (!test_and_clear_bit(TASKLET_STATE_SCHED,
25248 - t->func(t->data);
25249 - tasklet_unlock(t);
25252 - tasklet_unlock(t);
25254 + * Should always succeed - after a tasklist got on the
25255 + * list (after getting the SCHED bit set from 0 to 1),
25256 + * nothing but the tasklet softirq it got queued to can
25259 + if (!tasklet_trylock(t)) {
25264 - local_irq_disable();
25266 - *__this_cpu_read(tasklet_vec.tail) = t;
25267 - __this_cpu_write(tasklet_vec.tail, &(t->next));
25268 - __raise_softirq_irqoff(TASKLET_SOFTIRQ);
25269 - local_irq_enable();
25272 + * If we cannot handle the tasklet because it's disabled,
25273 + * mark it as pending. tasklet_enable() will later
25274 + * re-schedule the tasklet.
25276 + if (unlikely(atomic_read(&t->count))) {
25278 + /* implicit unlock: */
25280 + t->state = TASKLET_STATEF_PENDING;
25285 + * After this point on the tasklet might be rescheduled
25286 + * on another CPU, but it can only be added to another
25287 + * CPU's tasklet list if we unlock the tasklet (which we
25290 + if (!test_and_clear_bit(TASKLET_STATE_SCHED, &t->state))
25294 + t->func(t->data);
25297 + * Try to unlock the tasklet. We must use cmpxchg, because
25298 + * another CPU might have scheduled or disabled the tasklet.
25299 + * We only allow the STATE_RUN -> 0 transition here.
25301 + while (!tasklet_tryunlock(t)) {
25303 + * If it got disabled meanwhile, bail out:
25305 + if (atomic_read(&t->count))
25306 + goto out_disabled;
25308 + * If it got scheduled meanwhile, re-execute
25309 + * the tasklet function:
25311 + if (test_and_clear_bit(TASKLET_STATE_SCHED, &t->state))
25314 + printk("hm, tasklet state: %08lx\n", t->state);
25316 + tasklet_unlock(t);
25323 +static void tasklet_action(struct softirq_action *a)
25325 + struct tasklet_struct *list;
25327 + local_irq_disable();
25329 + list = __this_cpu_read(tasklet_vec.head);
25330 + __this_cpu_write(tasklet_vec.head, NULL);
25331 + __this_cpu_write(tasklet_vec.tail, this_cpu_ptr(&tasklet_vec.head));
25333 + local_irq_enable();
25335 + __tasklet_action(a, list);
25338 static void tasklet_hi_action(struct softirq_action *a)
25340 struct tasklet_struct *list;
25342 local_irq_disable();
25344 list = __this_cpu_read(tasklet_hi_vec.head);
25345 __this_cpu_write(tasklet_hi_vec.head, NULL);
25346 __this_cpu_write(tasklet_hi_vec.tail, this_cpu_ptr(&tasklet_hi_vec.head));
25347 - local_irq_enable();
25350 - struct tasklet_struct *t = list;
25352 - list = list->next;
25354 - if (tasklet_trylock(t)) {
25355 - if (!atomic_read(&t->count)) {
25356 - if (!test_and_clear_bit(TASKLET_STATE_SCHED,
25359 - t->func(t->data);
25360 - tasklet_unlock(t);
25363 - tasklet_unlock(t);
25365 + local_irq_enable();
25367 - local_irq_disable();
25369 - *__this_cpu_read(tasklet_hi_vec.tail) = t;
25370 - __this_cpu_write(tasklet_hi_vec.tail, &(t->next));
25371 - __raise_softirq_irqoff(HI_SOFTIRQ);
25372 - local_irq_enable();
25374 + __tasklet_action(a, list);
25377 void tasklet_init(struct tasklet_struct *t,
25378 @@ -572,7 +1070,7 @@ void tasklet_kill(struct tasklet_struct *t)
25380 while (test_and_set_bit(TASKLET_STATE_SCHED, &t->state)) {
25384 } while (test_bit(TASKLET_STATE_SCHED, &t->state));
25386 tasklet_unlock_wait(t);
25387 @@ -646,25 +1144,26 @@ void __init softirq_init(void)
25388 open_softirq(HI_SOFTIRQ, tasklet_hi_action);
25391 -static int ksoftirqd_should_run(unsigned int cpu)
25393 - return local_softirq_pending();
25396 -static void run_ksoftirqd(unsigned int cpu)
25397 +#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT_RT_FULL)
25398 +void tasklet_unlock_wait(struct tasklet_struct *t)
25400 - local_irq_disable();
25401 - if (local_softirq_pending()) {
25402 + while (test_bit(TASKLET_STATE_RUN, &(t)->state)) {
25404 - * We can safely run softirq on inline stack, as we are not deep
25405 - * in the task stack here.
25406 + * Hack for now to avoid this busy-loop:
25409 - local_irq_enable();
25410 - cond_resched_rcu_qs();
25412 +#ifdef CONFIG_PREEMPT_RT_FULL
25418 - local_irq_enable();
25420 +EXPORT_SYMBOL(tasklet_unlock_wait);
25423 +static int ksoftirqd_should_run(unsigned int cpu)
25425 + return ksoftirqd_softirq_pending();
25428 #ifdef CONFIG_HOTPLUG_CPU
25429 @@ -746,16 +1245,31 @@ static struct notifier_block cpu_nfb = {
25431 static struct smp_hotplug_thread softirq_threads = {
25432 .store = &ksoftirqd,
25433 + .setup = ksoftirqd_set_sched_params,
25434 .thread_should_run = ksoftirqd_should_run,
25435 .thread_fn = run_ksoftirqd,
25436 .thread_comm = "ksoftirqd/%u",
25439 +#ifdef CONFIG_PREEMPT_RT_FULL
25440 +static struct smp_hotplug_thread softirq_timer_threads = {
25441 + .store = &ktimer_softirqd,
25442 + .setup = ktimer_softirqd_set_sched_params,
25443 + .cleanup = ktimer_softirqd_clr_sched_params,
25444 + .thread_should_run = ktimer_softirqd_should_run,
25445 + .thread_fn = run_ksoftirqd,
25446 + .thread_comm = "ktimersoftd/%u",
25450 static __init int spawn_ksoftirqd(void)
25452 register_cpu_notifier(&cpu_nfb);
25454 BUG_ON(smpboot_register_percpu_thread(&softirq_threads));
25455 +#ifdef CONFIG_PREEMPT_RT_FULL
25456 + BUG_ON(smpboot_register_percpu_thread(&softirq_timer_threads));
25461 diff --git a/kernel/stop_machine.c b/kernel/stop_machine.c
25462 index a3bbaee77c58..f84d3b45cda7 100644
25463 --- a/kernel/stop_machine.c
25464 +++ b/kernel/stop_machine.c
25465 @@ -37,7 +37,7 @@ struct cpu_stop_done {
25466 struct cpu_stopper {
25467 struct task_struct *thread;
25470 + raw_spinlock_t lock;
25471 bool enabled; /* is this stopper enabled? */
25472 struct list_head works; /* list of pending works */
25474 @@ -86,12 +86,12 @@ static void cpu_stop_queue_work(unsigned int cpu, struct cpu_stop_work *work)
25475 struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu);
25476 unsigned long flags;
25478 - spin_lock_irqsave(&stopper->lock, flags);
25479 + raw_spin_lock_irqsave(&stopper->lock, flags);
25480 if (stopper->enabled)
25481 __cpu_stop_queue_work(stopper, work);
25483 cpu_stop_signal_done(work->done, false);
25484 - spin_unlock_irqrestore(&stopper->lock, flags);
25485 + raw_spin_unlock_irqrestore(&stopper->lock, flags);
25489 @@ -224,8 +224,8 @@ static int cpu_stop_queue_two_works(int cpu1, struct cpu_stop_work *work1,
25492 lg_double_lock(&stop_cpus_lock, cpu1, cpu2);
25493 - spin_lock_irq(&stopper1->lock);
25494 - spin_lock_nested(&stopper2->lock, SINGLE_DEPTH_NESTING);
25495 + raw_spin_lock_irq(&stopper1->lock);
25496 + raw_spin_lock_nested(&stopper2->lock, SINGLE_DEPTH_NESTING);
25499 if (!stopper1->enabled || !stopper2->enabled)
25500 @@ -235,8 +235,8 @@ static int cpu_stop_queue_two_works(int cpu1, struct cpu_stop_work *work1,
25501 __cpu_stop_queue_work(stopper1, work1);
25502 __cpu_stop_queue_work(stopper2, work2);
25504 - spin_unlock(&stopper2->lock);
25505 - spin_unlock_irq(&stopper1->lock);
25506 + raw_spin_unlock(&stopper2->lock);
25507 + raw_spin_unlock_irq(&stopper1->lock);
25508 lg_double_unlock(&stop_cpus_lock, cpu1, cpu2);
25511 @@ -258,7 +258,7 @@ int stop_two_cpus(unsigned int cpu1, unsigned int cpu2, cpu_stop_fn_t fn, void *
25512 struct cpu_stop_work work1, work2;
25513 struct multi_stop_data msdata;
25515 - preempt_disable();
25516 + preempt_disable_nort();
25517 msdata = (struct multi_stop_data){
25520 @@ -278,11 +278,11 @@ int stop_two_cpus(unsigned int cpu1, unsigned int cpu2, cpu_stop_fn_t fn, void *
25523 if (cpu_stop_queue_two_works(cpu1, &work1, cpu2, &work2)) {
25524 - preempt_enable();
25525 + preempt_enable_nort();
25529 - preempt_enable();
25530 + preempt_enable_nort();
25532 wait_for_completion(&done.completion);
25534 @@ -315,17 +315,20 @@ static DEFINE_MUTEX(stop_cpus_mutex);
25536 static void queue_stop_cpus_work(const struct cpumask *cpumask,
25537 cpu_stop_fn_t fn, void *arg,
25538 - struct cpu_stop_done *done)
25539 + struct cpu_stop_done *done, bool inactive)
25541 struct cpu_stop_work *work;
25545 - * Disable preemption while queueing to avoid getting
25546 - * preempted by a stopper which might wait for other stoppers
25547 - * to enter @fn which can lead to deadlock.
25548 + * Make sure that all work is queued on all cpus before
25549 + * any of the cpus can execute it.
25551 - lg_global_lock(&stop_cpus_lock);
25553 + lg_global_lock(&stop_cpus_lock);
25555 + lg_global_trylock_relax(&stop_cpus_lock);
25557 for_each_cpu(cpu, cpumask) {
25558 work = &per_cpu(cpu_stopper.stop_work, cpu);
25560 @@ -342,7 +345,7 @@ static int __stop_cpus(const struct cpumask *cpumask,
25561 struct cpu_stop_done done;
25563 cpu_stop_init_done(&done, cpumask_weight(cpumask));
25564 - queue_stop_cpus_work(cpumask, fn, arg, &done);
25565 + queue_stop_cpus_work(cpumask, fn, arg, &done, false);
25566 wait_for_completion(&done.completion);
25567 return done.executed ? done.ret : -ENOENT;
25569 @@ -422,9 +425,9 @@ static int cpu_stop_should_run(unsigned int cpu)
25570 unsigned long flags;
25573 - spin_lock_irqsave(&stopper->lock, flags);
25574 + raw_spin_lock_irqsave(&stopper->lock, flags);
25575 run = !list_empty(&stopper->works);
25576 - spin_unlock_irqrestore(&stopper->lock, flags);
25577 + raw_spin_unlock_irqrestore(&stopper->lock, flags);
25581 @@ -436,13 +439,13 @@ static void cpu_stopper_thread(unsigned int cpu)
25585 - spin_lock_irq(&stopper->lock);
25586 + raw_spin_lock_irq(&stopper->lock);
25587 if (!list_empty(&stopper->works)) {
25588 work = list_first_entry(&stopper->works,
25589 struct cpu_stop_work, list);
25590 list_del_init(&work->list);
25592 - spin_unlock_irq(&stopper->lock);
25593 + raw_spin_unlock_irq(&stopper->lock);
25596 cpu_stop_fn_t fn = work->fn;
25597 @@ -450,6 +453,16 @@ repeat:
25598 struct cpu_stop_done *done = work->done;
25599 char ksym_buf[KSYM_NAME_LEN] __maybe_unused;
25602 + * Wait until the stopper finished scheduling on all
25605 + lg_global_lock(&stop_cpus_lock);
25607 + * Let other cpu threads continue as well
25609 + lg_global_unlock(&stop_cpus_lock);
25611 /* cpu stop callbacks are not allowed to sleep */
25614 @@ -520,10 +533,12 @@ static int __init cpu_stop_init(void)
25615 for_each_possible_cpu(cpu) {
25616 struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu);
25618 - spin_lock_init(&stopper->lock);
25619 + raw_spin_lock_init(&stopper->lock);
25620 INIT_LIST_HEAD(&stopper->works);
25623 + lg_lock_init(&stop_cpus_lock, "stop_cpus_lock");
25625 BUG_ON(smpboot_register_percpu_thread(&cpu_stop_threads));
25626 stop_machine_unpark(raw_smp_processor_id());
25627 stop_machine_initialized = true;
25628 @@ -620,7 +635,7 @@ int stop_machine_from_inactive_cpu(cpu_stop_fn_t fn, void *data,
25629 set_state(&msdata, MULTI_STOP_PREPARE);
25630 cpu_stop_init_done(&done, num_active_cpus());
25631 queue_stop_cpus_work(cpu_active_mask, multi_cpu_stop, &msdata,
25634 ret = multi_cpu_stop(&msdata);
25636 /* Busy wait for completion. */
25637 diff --git a/kernel/time/hrtimer.c b/kernel/time/hrtimer.c
25638 index 17f7bcff1e02..ba3d60144838 100644
25639 --- a/kernel/time/hrtimer.c
25640 +++ b/kernel/time/hrtimer.c
25641 @@ -48,11 +48,13 @@
25642 #include <linux/sched/rt.h>
25643 #include <linux/sched/deadline.h>
25644 #include <linux/timer.h>
25645 +#include <linux/kthread.h>
25646 #include <linux/freezer.h>
25648 #include <asm/uaccess.h>
25650 #include <trace/events/timer.h>
25651 +#include <trace/events/hist.h>
25653 #include "tick-internal.h"
25655 @@ -717,6 +719,44 @@ static void clock_was_set_work(struct work_struct *work)
25657 static DECLARE_WORK(hrtimer_work, clock_was_set_work);
25659 +#ifdef CONFIG_PREEMPT_RT_FULL
25661 + * RT can not call schedule_work from real interrupt context.
25662 + * Need to make a thread to do the real work.
25664 +static struct task_struct *clock_set_delay_thread;
25665 +static bool do_clock_set_delay;
25667 +static int run_clock_set_delay(void *ignore)
25669 + while (!kthread_should_stop()) {
25670 + set_current_state(TASK_INTERRUPTIBLE);
25671 + if (do_clock_set_delay) {
25672 + do_clock_set_delay = false;
25673 + schedule_work(&hrtimer_work);
25677 + __set_current_state(TASK_RUNNING);
25681 +void clock_was_set_delayed(void)
25683 + do_clock_set_delay = true;
25684 + /* Make visible before waking up process */
25686 + wake_up_process(clock_set_delay_thread);
25689 +static __init int create_clock_set_delay_thread(void)
25691 + clock_set_delay_thread = kthread_run(run_clock_set_delay, NULL, "kclksetdelayd");
25692 + BUG_ON(!clock_set_delay_thread);
25695 +early_initcall(create_clock_set_delay_thread);
25696 +#else /* PREEMPT_RT_FULL */
25698 * Called from timekeeping and resume code to reprogramm the hrtimer
25699 * interrupt device on all cpus.
25700 @@ -725,6 +765,7 @@ void clock_was_set_delayed(void)
25702 schedule_work(&hrtimer_work);
25708 @@ -734,11 +775,8 @@ static inline int hrtimer_is_hres_enabled(void) { return 0; }
25709 static inline void hrtimer_switch_to_hres(void) { }
25711 hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { }
25712 -static inline int hrtimer_reprogram(struct hrtimer *timer,
25713 - struct hrtimer_clock_base *base)
25717 +static inline void hrtimer_reprogram(struct hrtimer *timer,
25718 + struct hrtimer_clock_base *base) { }
25719 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { }
25720 static inline void retrigger_next_event(void *arg) { }
25722 @@ -870,6 +908,32 @@ u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
25724 EXPORT_SYMBOL_GPL(hrtimer_forward);
25726 +#ifdef CONFIG_PREEMPT_RT_BASE
25727 +# define wake_up_timer_waiters(b) wake_up(&(b)->wait)
25730 + * hrtimer_wait_for_timer - Wait for a running timer
25732 + * @timer: timer to wait for
25734 + * The function waits in case the timers callback function is
25735 + * currently executed on the waitqueue of the timer base. The
25736 + * waitqueue is woken up after the timer callback function has
25737 + * finished execution.
25739 +void hrtimer_wait_for_timer(const struct hrtimer *timer)
25741 + struct hrtimer_clock_base *base = timer->base;
25743 + if (base && base->cpu_base && !timer->irqsafe)
25744 + wait_event(base->cpu_base->wait,
25745 + !(hrtimer_callback_running(timer)));
25749 +# define wake_up_timer_waiters(b) do { } while (0)
25753 * enqueue_hrtimer - internal function to (re)start a timer
25755 @@ -911,6 +975,11 @@ static void __remove_hrtimer(struct hrtimer *timer,
25756 if (!(state & HRTIMER_STATE_ENQUEUED))
25759 + if (unlikely(!list_empty(&timer->cb_entry))) {
25760 + list_del_init(&timer->cb_entry);
25764 if (!timerqueue_del(&base->active, &timer->node))
25765 cpu_base->active_bases &= ~(1 << base->index);
25767 @@ -1006,7 +1075,16 @@ void hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
25768 new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED);
25770 timer_stats_hrtimer_set_start_info(timer);
25771 +#ifdef CONFIG_MISSED_TIMER_OFFSETS_HIST
25773 + ktime_t now = new_base->get_time();
25775 + if (ktime_to_ns(tim) < ktime_to_ns(now))
25776 + timer->praecox = now;
25778 + timer->praecox = ktime_set(0, 0);
25781 leftmost = enqueue_hrtimer(timer, new_base);
25784 @@ -1078,7 +1156,7 @@ int hrtimer_cancel(struct hrtimer *timer)
25789 + hrtimer_wait_for_timer(timer);
25792 EXPORT_SYMBOL_GPL(hrtimer_cancel);
25793 @@ -1142,6 +1220,7 @@ static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
25795 base = hrtimer_clockid_to_base(clock_id);
25796 timer->base = &cpu_base->clock_base[base];
25797 + INIT_LIST_HEAD(&timer->cb_entry);
25798 timerqueue_init(&timer->node);
25800 #ifdef CONFIG_TIMER_STATS
25801 @@ -1182,6 +1261,7 @@ bool hrtimer_active(const struct hrtimer *timer)
25802 seq = raw_read_seqcount_begin(&cpu_base->seq);
25804 if (timer->state != HRTIMER_STATE_INACTIVE ||
25805 + cpu_base->running_soft == timer ||
25806 cpu_base->running == timer)
25809 @@ -1280,10 +1360,112 @@ static void __run_hrtimer(struct hrtimer_cpu_base *cpu_base,
25810 cpu_base->running = NULL;
25813 +#ifdef CONFIG_PREEMPT_RT_BASE
25814 +static void hrtimer_rt_reprogram(int restart, struct hrtimer *timer,
25815 + struct hrtimer_clock_base *base)
25819 + if (restart != HRTIMER_NORESTART &&
25820 + !(timer->state & HRTIMER_STATE_ENQUEUED)) {
25822 + leftmost = enqueue_hrtimer(timer, base);
25825 +#ifdef CONFIG_HIGH_RES_TIMERS
25826 + if (!hrtimer_is_hres_active(timer)) {
25828 + * Kick to reschedule the next tick to handle the new timer
25829 + * on dynticks target.
25831 + if (base->cpu_base->nohz_active)
25832 + wake_up_nohz_cpu(base->cpu_base->cpu);
25835 + hrtimer_reprogram(timer, base);
25842 + * The changes in mainline which removed the callback modes from
25843 + * hrtimer are not yet working with -rt. The non wakeup_process()
25844 + * based callbacks which involve sleeping locks need to be treated
25847 +static void hrtimer_rt_run_pending(void)
25849 + enum hrtimer_restart (*fn)(struct hrtimer *);
25850 + struct hrtimer_cpu_base *cpu_base;
25851 + struct hrtimer_clock_base *base;
25852 + struct hrtimer *timer;
25853 + int index, restart;
25855 + local_irq_disable();
25856 + cpu_base = &per_cpu(hrtimer_bases, smp_processor_id());
25858 + raw_spin_lock(&cpu_base->lock);
25860 + for (index = 0; index < HRTIMER_MAX_CLOCK_BASES; index++) {
25861 + base = &cpu_base->clock_base[index];
25863 + while (!list_empty(&base->expired)) {
25864 + timer = list_first_entry(&base->expired,
25865 + struct hrtimer, cb_entry);
25868 + * Same as the above __run_hrtimer function
25869 + * just we run with interrupts enabled.
25871 + debug_deactivate(timer);
25872 + cpu_base->running_soft = timer;
25873 + raw_write_seqcount_barrier(&cpu_base->seq);
25875 + __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE, 0);
25876 + timer_stats_account_hrtimer(timer);
25877 + fn = timer->function;
25879 + raw_spin_unlock_irq(&cpu_base->lock);
25880 + restart = fn(timer);
25881 + raw_spin_lock_irq(&cpu_base->lock);
25883 + hrtimer_rt_reprogram(restart, timer, base);
25884 + raw_write_seqcount_barrier(&cpu_base->seq);
25886 + WARN_ON_ONCE(cpu_base->running_soft != timer);
25887 + cpu_base->running_soft = NULL;
25891 + raw_spin_unlock_irq(&cpu_base->lock);
25893 + wake_up_timer_waiters(cpu_base);
25896 +static int hrtimer_rt_defer(struct hrtimer *timer)
25898 + if (timer->irqsafe)
25901 + __remove_hrtimer(timer, timer->base, timer->state, 0);
25902 + list_add_tail(&timer->cb_entry, &timer->base->expired);
25908 +static inline int hrtimer_rt_defer(struct hrtimer *timer) { return 0; }
25912 +static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer);
25914 static void __hrtimer_run_queues(struct hrtimer_cpu_base *cpu_base, ktime_t now)
25916 struct hrtimer_clock_base *base = cpu_base->clock_base;
25917 unsigned int active = cpu_base->active_bases;
25920 for (; active; base++, active >>= 1) {
25921 struct timerqueue_node *node;
25922 @@ -1299,6 +1481,15 @@ static void __hrtimer_run_queues(struct hrtimer_cpu_base *cpu_base, ktime_t now)
25924 timer = container_of(node, struct hrtimer, node);
25926 + trace_hrtimer_interrupt(raw_smp_processor_id(),
25927 + ktime_to_ns(ktime_sub(ktime_to_ns(timer->praecox) ?
25928 + timer->praecox : hrtimer_get_expires(timer),
25931 + timer->function == hrtimer_wakeup ?
25932 + container_of(timer, struct hrtimer_sleeper,
25933 + timer)->task : NULL);
25936 * The immediate goal for using the softexpires is
25937 * minimizing wakeups, not running timers at the
25938 @@ -1314,9 +1505,14 @@ static void __hrtimer_run_queues(struct hrtimer_cpu_base *cpu_base, ktime_t now)
25939 if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer))
25942 - __run_hrtimer(cpu_base, base, timer, &basenow);
25943 + if (!hrtimer_rt_defer(timer))
25944 + __run_hrtimer(cpu_base, base, timer, &basenow);
25950 + raise_softirq_irqoff(HRTIMER_SOFTIRQ);
25953 #ifdef CONFIG_HIGH_RES_TIMERS
25954 @@ -1479,16 +1675,18 @@ static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
25955 void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
25957 sl->timer.function = hrtimer_wakeup;
25958 + sl->timer.irqsafe = 1;
25961 EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
25963 -static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
25964 +static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode,
25965 + unsigned long state)
25967 hrtimer_init_sleeper(t, current);
25970 - set_current_state(TASK_INTERRUPTIBLE);
25971 + set_current_state(state);
25972 hrtimer_start_expires(&t->timer, mode);
25974 if (likely(t->task))
25975 @@ -1530,7 +1728,8 @@ long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
25977 hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
25979 - if (do_nanosleep(&t, HRTIMER_MODE_ABS))
25980 + /* cpu_chill() does not care about restart state. */
25981 + if (do_nanosleep(&t, HRTIMER_MODE_ABS, TASK_INTERRUPTIBLE))
25984 rmtp = restart->nanosleep.rmtp;
25985 @@ -1547,8 +1746,10 @@ out:
25989 -long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
25990 - const enum hrtimer_mode mode, const clockid_t clockid)
25992 +__hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
25993 + const enum hrtimer_mode mode, const clockid_t clockid,
25994 + unsigned long state)
25996 struct restart_block *restart;
25997 struct hrtimer_sleeper t;
25998 @@ -1561,7 +1762,7 @@ long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
26000 hrtimer_init_on_stack(&t.timer, clockid, mode);
26001 hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack);
26002 - if (do_nanosleep(&t, mode))
26003 + if (do_nanosleep(&t, mode, state))
26006 /* Absolute timers do not update the rmtp value and restart: */
26007 @@ -1588,6 +1789,12 @@ out:
26011 +long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
26012 + const enum hrtimer_mode mode, const clockid_t clockid)
26014 + return __hrtimer_nanosleep(rqtp, rmtp, mode, clockid, TASK_INTERRUPTIBLE);
26017 SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp,
26018 struct timespec __user *, rmtp)
26020 @@ -1602,6 +1809,26 @@ SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp,
26021 return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
26024 +#ifdef CONFIG_PREEMPT_RT_FULL
26026 + * Sleep for 1 ms in hope whoever holds what we want will let it go.
26028 +void cpu_chill(void)
26030 + struct timespec tu = {
26031 + .tv_nsec = NSEC_PER_MSEC,
26033 + unsigned int freeze_flag = current->flags & PF_NOFREEZE;
26035 + current->flags |= PF_NOFREEZE;
26036 + __hrtimer_nanosleep(&tu, NULL, HRTIMER_MODE_REL, CLOCK_MONOTONIC,
26037 + TASK_UNINTERRUPTIBLE);
26038 + if (!freeze_flag)
26039 + current->flags &= ~PF_NOFREEZE;
26041 +EXPORT_SYMBOL(cpu_chill);
26045 * Functions related to boot-time initialization:
26047 @@ -1613,10 +1840,14 @@ static void init_hrtimers_cpu(int cpu)
26048 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
26049 cpu_base->clock_base[i].cpu_base = cpu_base;
26050 timerqueue_init_head(&cpu_base->clock_base[i].active);
26051 + INIT_LIST_HEAD(&cpu_base->clock_base[i].expired);
26054 cpu_base->cpu = cpu;
26055 hrtimer_init_hres(cpu_base);
26056 +#ifdef CONFIG_PREEMPT_RT_BASE
26057 + init_waitqueue_head(&cpu_base->wait);
26061 #ifdef CONFIG_HOTPLUG_CPU
26062 @@ -1714,11 +1945,21 @@ static struct notifier_block hrtimers_nb = {
26063 .notifier_call = hrtimer_cpu_notify,
26066 +#ifdef CONFIG_PREEMPT_RT_BASE
26067 +static void run_hrtimer_softirq(struct softirq_action *h)
26069 + hrtimer_rt_run_pending();
26073 void __init hrtimers_init(void)
26075 hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
26076 (void *)(long)smp_processor_id());
26077 register_cpu_notifier(&hrtimers_nb);
26078 +#ifdef CONFIG_PREEMPT_RT_BASE
26079 + open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq);
26084 diff --git a/kernel/time/itimer.c b/kernel/time/itimer.c
26085 index 1d5c7204ddc9..184de6751180 100644
26086 --- a/kernel/time/itimer.c
26087 +++ b/kernel/time/itimer.c
26088 @@ -213,6 +213,7 @@ again:
26089 /* We are sharing ->siglock with it_real_fn() */
26090 if (hrtimer_try_to_cancel(timer) < 0) {
26091 spin_unlock_irq(&tsk->sighand->siglock);
26092 + hrtimer_wait_for_timer(&tsk->signal->real_timer);
26095 expires = timeval_to_ktime(value->it_value);
26096 diff --git a/kernel/time/jiffies.c b/kernel/time/jiffies.c
26097 index 347fecf86a3f..2ede47408a3e 100644
26098 --- a/kernel/time/jiffies.c
26099 +++ b/kernel/time/jiffies.c
26100 @@ -74,7 +74,8 @@ static struct clocksource clocksource_jiffies = {
26104 -__cacheline_aligned_in_smp DEFINE_SEQLOCK(jiffies_lock);
26105 +__cacheline_aligned_in_smp DEFINE_RAW_SPINLOCK(jiffies_lock);
26106 +__cacheline_aligned_in_smp seqcount_t jiffies_seq;
26108 #if (BITS_PER_LONG < 64)
26109 u64 get_jiffies_64(void)
26110 @@ -83,9 +84,9 @@ u64 get_jiffies_64(void)
26114 - seq = read_seqbegin(&jiffies_lock);
26115 + seq = read_seqcount_begin(&jiffies_seq);
26117 - } while (read_seqretry(&jiffies_lock, seq));
26118 + } while (read_seqcount_retry(&jiffies_seq, seq));
26121 EXPORT_SYMBOL(get_jiffies_64);
26122 diff --git a/kernel/time/ntp.c b/kernel/time/ntp.c
26123 index ab861771e37f..0f6868fd2de6 100644
26124 --- a/kernel/time/ntp.c
26125 +++ b/kernel/time/ntp.c
26127 #include <linux/workqueue.h>
26128 #include <linux/hrtimer.h>
26129 #include <linux/jiffies.h>
26130 +#include <linux/kthread.h>
26131 #include <linux/math64.h>
26132 #include <linux/timex.h>
26133 #include <linux/time.h>
26134 @@ -562,10 +563,52 @@ static void sync_cmos_clock(struct work_struct *work)
26135 &sync_cmos_work, timespec64_to_jiffies(&next));
26138 +#ifdef CONFIG_PREEMPT_RT_FULL
26140 + * RT can not call schedule_delayed_work from real interrupt context.
26141 + * Need to make a thread to do the real work.
26143 +static struct task_struct *cmos_delay_thread;
26144 +static bool do_cmos_delay;
26146 +static int run_cmos_delay(void *ignore)
26148 + while (!kthread_should_stop()) {
26149 + set_current_state(TASK_INTERRUPTIBLE);
26150 + if (do_cmos_delay) {
26151 + do_cmos_delay = false;
26152 + queue_delayed_work(system_power_efficient_wq,
26153 + &sync_cmos_work, 0);
26157 + __set_current_state(TASK_RUNNING);
26161 +void ntp_notify_cmos_timer(void)
26163 + do_cmos_delay = true;
26164 + /* Make visible before waking up process */
26166 + wake_up_process(cmos_delay_thread);
26169 +static __init int create_cmos_delay_thread(void)
26171 + cmos_delay_thread = kthread_run(run_cmos_delay, NULL, "kcmosdelayd");
26172 + BUG_ON(!cmos_delay_thread);
26175 +early_initcall(create_cmos_delay_thread);
26179 void ntp_notify_cmos_timer(void)
26181 queue_delayed_work(system_power_efficient_wq, &sync_cmos_work, 0);
26183 +#endif /* CONFIG_PREEMPT_RT_FULL */
26186 void ntp_notify_cmos_timer(void) { }
26187 diff --git a/kernel/time/posix-cpu-timers.c b/kernel/time/posix-cpu-timers.c
26188 index 80016b329d94..b7342b6e6a5a 100644
26189 --- a/kernel/time/posix-cpu-timers.c
26190 +++ b/kernel/time/posix-cpu-timers.c
26194 #include <linux/sched.h>
26195 +#include <linux/sched/rt.h>
26196 #include <linux/posix-timers.h>
26197 #include <linux/errno.h>
26198 #include <linux/math64.h>
26199 @@ -650,7 +651,7 @@ static int posix_cpu_timer_set(struct k_itimer *timer, int timer_flags,
26201 * Disarm any old timer after extracting its expiry time.
26203 - WARN_ON_ONCE(!irqs_disabled());
26204 + WARN_ON_ONCE_NONRT(!irqs_disabled());
26207 old_incr = timer->it.cpu.incr;
26208 @@ -1092,7 +1093,7 @@ void posix_cpu_timer_schedule(struct k_itimer *timer)
26210 * Now re-arm for the new expiry time.
26212 - WARN_ON_ONCE(!irqs_disabled());
26213 + WARN_ON_ONCE_NONRT(!irqs_disabled());
26215 unlock_task_sighand(p, &flags);
26217 @@ -1183,13 +1184,13 @@ static inline int fastpath_timer_check(struct task_struct *tsk)
26218 * already updated our counts. We need to check if any timers fire now.
26219 * Interrupts are disabled.
26221 -void run_posix_cpu_timers(struct task_struct *tsk)
26222 +static void __run_posix_cpu_timers(struct task_struct *tsk)
26225 struct k_itimer *timer, *next;
26226 unsigned long flags;
26228 - WARN_ON_ONCE(!irqs_disabled());
26229 + WARN_ON_ONCE_NONRT(!irqs_disabled());
26232 * The fast path checks that there are no expired thread or thread
26233 @@ -1243,6 +1244,190 @@ void run_posix_cpu_timers(struct task_struct *tsk)
26237 +#ifdef CONFIG_PREEMPT_RT_BASE
26238 +#include <linux/kthread.h>
26239 +#include <linux/cpu.h>
26240 +DEFINE_PER_CPU(struct task_struct *, posix_timer_task);
26241 +DEFINE_PER_CPU(struct task_struct *, posix_timer_tasklist);
26243 +static int posix_cpu_timers_thread(void *data)
26245 + int cpu = (long)data;
26247 + BUG_ON(per_cpu(posix_timer_task,cpu) != current);
26249 + while (!kthread_should_stop()) {
26250 + struct task_struct *tsk = NULL;
26251 + struct task_struct *next = NULL;
26253 + if (cpu_is_offline(cpu))
26254 + goto wait_to_die;
26256 + /* grab task list */
26257 + raw_local_irq_disable();
26258 + tsk = per_cpu(posix_timer_tasklist, cpu);
26259 + per_cpu(posix_timer_tasklist, cpu) = NULL;
26260 + raw_local_irq_enable();
26262 + /* its possible the list is empty, just return */
26264 + set_current_state(TASK_INTERRUPTIBLE);
26266 + __set_current_state(TASK_RUNNING);
26270 + /* Process task list */
26273 + next = tsk->posix_timer_list;
26275 + /* run the task timers, clear its ptr and
26278 + __run_posix_cpu_timers(tsk);
26279 + tsk->posix_timer_list = NULL;
26280 + put_task_struct(tsk);
26282 + /* check if this is the last on the list */
26291 + /* Wait for kthread_stop */
26292 + set_current_state(TASK_INTERRUPTIBLE);
26293 + while (!kthread_should_stop()) {
26295 + set_current_state(TASK_INTERRUPTIBLE);
26297 + __set_current_state(TASK_RUNNING);
26301 +static inline int __fastpath_timer_check(struct task_struct *tsk)
26303 + /* tsk == current, ensure it is safe to use ->signal/sighand */
26304 + if (unlikely(tsk->exit_state))
26307 + if (!task_cputime_zero(&tsk->cputime_expires))
26310 + if (!task_cputime_zero(&tsk->signal->cputime_expires))
26316 +void run_posix_cpu_timers(struct task_struct *tsk)
26318 + unsigned long cpu = smp_processor_id();
26319 + struct task_struct *tasklist;
26321 + BUG_ON(!irqs_disabled());
26322 + if(!per_cpu(posix_timer_task, cpu))
26324 + /* get per-cpu references */
26325 + tasklist = per_cpu(posix_timer_tasklist, cpu);
26327 + /* check to see if we're already queued */
26328 + if (!tsk->posix_timer_list && __fastpath_timer_check(tsk)) {
26329 + get_task_struct(tsk);
26331 + tsk->posix_timer_list = tasklist;
26334 + * The list is terminated by a self-pointing
26337 + tsk->posix_timer_list = tsk;
26339 + per_cpu(posix_timer_tasklist, cpu) = tsk;
26341 + wake_up_process(per_cpu(posix_timer_task, cpu));
26346 + * posix_cpu_thread_call - callback that gets triggered when a CPU is added.
26347 + * Here we can start up the necessary migration thread for the new CPU.
26349 +static int posix_cpu_thread_call(struct notifier_block *nfb,
26350 + unsigned long action, void *hcpu)
26352 + int cpu = (long)hcpu;
26353 + struct task_struct *p;
26354 + struct sched_param param;
26356 + switch (action) {
26357 + case CPU_UP_PREPARE:
26358 + p = kthread_create(posix_cpu_timers_thread, hcpu,
26359 + "posixcputmr/%d",cpu);
26361 + return NOTIFY_BAD;
26362 + p->flags |= PF_NOFREEZE;
26363 + kthread_bind(p, cpu);
26364 + /* Must be high prio to avoid getting starved */
26365 + param.sched_priority = MAX_RT_PRIO-1;
26366 + sched_setscheduler(p, SCHED_FIFO, ¶m);
26367 + per_cpu(posix_timer_task,cpu) = p;
26370 + /* Strictly unneccessary, as first user will wake it. */
26371 + wake_up_process(per_cpu(posix_timer_task,cpu));
26373 +#ifdef CONFIG_HOTPLUG_CPU
26374 + case CPU_UP_CANCELED:
26375 + /* Unbind it from offline cpu so it can run. Fall thru. */
26376 + kthread_bind(per_cpu(posix_timer_task, cpu),
26377 + cpumask_any(cpu_online_mask));
26378 + kthread_stop(per_cpu(posix_timer_task,cpu));
26379 + per_cpu(posix_timer_task,cpu) = NULL;
26382 + kthread_stop(per_cpu(posix_timer_task,cpu));
26383 + per_cpu(posix_timer_task,cpu) = NULL;
26387 + return NOTIFY_OK;
26390 +/* Register at highest priority so that task migration (migrate_all_tasks)
26391 + * happens before everything else.
26393 +static struct notifier_block posix_cpu_thread_notifier = {
26394 + .notifier_call = posix_cpu_thread_call,
26398 +static int __init posix_cpu_thread_init(void)
26400 + void *hcpu = (void *)(long)smp_processor_id();
26401 + /* Start one for boot CPU. */
26402 + unsigned long cpu;
26404 + /* init the per-cpu posix_timer_tasklets */
26405 + for_each_possible_cpu(cpu)
26406 + per_cpu(posix_timer_tasklist, cpu) = NULL;
26408 + posix_cpu_thread_call(&posix_cpu_thread_notifier, CPU_UP_PREPARE, hcpu);
26409 + posix_cpu_thread_call(&posix_cpu_thread_notifier, CPU_ONLINE, hcpu);
26410 + register_cpu_notifier(&posix_cpu_thread_notifier);
26413 +early_initcall(posix_cpu_thread_init);
26414 +#else /* CONFIG_PREEMPT_RT_BASE */
26415 +void run_posix_cpu_timers(struct task_struct *tsk)
26417 + __run_posix_cpu_timers(tsk);
26419 +#endif /* CONFIG_PREEMPT_RT_BASE */
26422 * Set one of the process-wide special case CPU timers or RLIMIT_CPU.
26423 * The tsk->sighand->siglock must be held by the caller.
26424 diff --git a/kernel/time/posix-timers.c b/kernel/time/posix-timers.c
26425 index f2826c35e918..464a98155a0e 100644
26426 --- a/kernel/time/posix-timers.c
26427 +++ b/kernel/time/posix-timers.c
26428 @@ -506,6 +506,7 @@ static enum hrtimer_restart posix_timer_fn(struct hrtimer *timer)
26429 static struct pid *good_sigevent(sigevent_t * event)
26431 struct task_struct *rtn = current->group_leader;
26432 + int sig = event->sigev_signo;
26434 if ((event->sigev_notify & SIGEV_THREAD_ID ) &&
26435 (!(rtn = find_task_by_vpid(event->sigev_notify_thread_id)) ||
26436 @@ -514,7 +515,8 @@ static struct pid *good_sigevent(sigevent_t * event)
26439 if (((event->sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE) &&
26440 - ((event->sigev_signo <= 0) || (event->sigev_signo > SIGRTMAX)))
26441 + (sig <= 0 || sig > SIGRTMAX || sig_kernel_only(sig) ||
26442 + sig_kernel_coredump(sig)))
26445 return task_pid(rtn);
26446 @@ -826,6 +828,20 @@ SYSCALL_DEFINE1(timer_getoverrun, timer_t, timer_id)
26451 + * Protected by RCU!
26453 +static void timer_wait_for_callback(struct k_clock *kc, struct k_itimer *timr)
26455 +#ifdef CONFIG_PREEMPT_RT_FULL
26456 + if (kc->timer_set == common_timer_set)
26457 + hrtimer_wait_for_timer(&timr->it.real.timer);
26459 + /* FIXME: Whacky hack for posix-cpu-timers */
26460 + schedule_timeout(1);
26464 /* Set a POSIX.1b interval timer. */
26465 /* timr->it_lock is taken. */
26467 @@ -903,6 +919,7 @@ retry:
26472 kc = clockid_to_kclock(timr->it_clock);
26473 if (WARN_ON_ONCE(!kc || !kc->timer_set))
26475 @@ -911,9 +928,12 @@ retry:
26477 unlock_timer(timr, flag);
26478 if (error == TIMER_RETRY) {
26479 + timer_wait_for_callback(kc, timr);
26480 rtn = NULL; // We already got the old time...
26481 + rcu_read_unlock();
26484 + rcu_read_unlock();
26486 if (old_setting && !error &&
26487 copy_to_user(old_setting, &old_spec, sizeof (old_spec)))
26488 @@ -951,10 +971,15 @@ retry_delete:
26493 if (timer_delete_hook(timer) == TIMER_RETRY) {
26494 unlock_timer(timer, flags);
26495 + timer_wait_for_callback(clockid_to_kclock(timer->it_clock),
26497 + rcu_read_unlock();
26500 + rcu_read_unlock();
26502 spin_lock(¤t->sighand->siglock);
26503 list_del(&timer->list);
26504 @@ -980,8 +1005,18 @@ static void itimer_delete(struct k_itimer *timer)
26506 spin_lock_irqsave(&timer->it_lock, flags);
26508 + /* On RT we can race with a deletion */
26509 + if (!timer->it_signal) {
26510 + unlock_timer(timer, flags);
26514 if (timer_delete_hook(timer) == TIMER_RETRY) {
26516 unlock_timer(timer, flags);
26517 + timer_wait_for_callback(clockid_to_kclock(timer->it_clock),
26519 + rcu_read_unlock();
26522 list_del(&timer->list);
26523 diff --git a/kernel/time/tick-broadcast-hrtimer.c b/kernel/time/tick-broadcast-hrtimer.c
26524 index 53d7184da0be..1b4ac3361c3f 100644
26525 --- a/kernel/time/tick-broadcast-hrtimer.c
26526 +++ b/kernel/time/tick-broadcast-hrtimer.c
26527 @@ -106,5 +106,6 @@ void tick_setup_hrtimer_broadcast(void)
26529 hrtimer_init(&bctimer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
26530 bctimer.function = bc_handler;
26531 + bctimer.irqsafe = true;
26532 clockevents_register_device(&ce_broadcast_hrtimer);
26534 diff --git a/kernel/time/tick-common.c b/kernel/time/tick-common.c
26535 index 4fcd99e12aa0..5a47f2e98faf 100644
26536 --- a/kernel/time/tick-common.c
26537 +++ b/kernel/time/tick-common.c
26538 @@ -79,13 +79,15 @@ int tick_is_oneshot_available(void)
26539 static void tick_periodic(int cpu)
26541 if (tick_do_timer_cpu == cpu) {
26542 - write_seqlock(&jiffies_lock);
26543 + raw_spin_lock(&jiffies_lock);
26544 + write_seqcount_begin(&jiffies_seq);
26546 /* Keep track of the next tick event */
26547 tick_next_period = ktime_add(tick_next_period, tick_period);
26550 - write_sequnlock(&jiffies_lock);
26551 + write_seqcount_end(&jiffies_seq);
26552 + raw_spin_unlock(&jiffies_lock);
26553 update_wall_time();
26556 @@ -157,9 +159,9 @@ void tick_setup_periodic(struct clock_event_device *dev, int broadcast)
26560 - seq = read_seqbegin(&jiffies_lock);
26561 + seq = read_seqcount_begin(&jiffies_seq);
26562 next = tick_next_period;
26563 - } while (read_seqretry(&jiffies_lock, seq));
26564 + } while (read_seqcount_retry(&jiffies_seq, seq));
26566 clockevents_switch_state(dev, CLOCK_EVT_STATE_ONESHOT);
26568 diff --git a/kernel/time/tick-sched.c b/kernel/time/tick-sched.c
26569 index 22c57e191a23..d536824cbd36 100644
26570 --- a/kernel/time/tick-sched.c
26571 +++ b/kernel/time/tick-sched.c
26572 @@ -62,7 +62,8 @@ static void tick_do_update_jiffies64(ktime_t now)
26575 /* Reevalute with jiffies_lock held */
26576 - write_seqlock(&jiffies_lock);
26577 + raw_spin_lock(&jiffies_lock);
26578 + write_seqcount_begin(&jiffies_seq);
26580 delta = ktime_sub(now, last_jiffies_update);
26581 if (delta.tv64 >= tick_period.tv64) {
26582 @@ -85,10 +86,12 @@ static void tick_do_update_jiffies64(ktime_t now)
26583 /* Keep the tick_next_period variable up to date */
26584 tick_next_period = ktime_add(last_jiffies_update, tick_period);
26586 - write_sequnlock(&jiffies_lock);
26587 + write_seqcount_end(&jiffies_seq);
26588 + raw_spin_unlock(&jiffies_lock);
26591 - write_sequnlock(&jiffies_lock);
26592 + write_seqcount_end(&jiffies_seq);
26593 + raw_spin_unlock(&jiffies_lock);
26594 update_wall_time();
26597 @@ -99,12 +102,14 @@ static ktime_t tick_init_jiffy_update(void)
26601 - write_seqlock(&jiffies_lock);
26602 + raw_spin_lock(&jiffies_lock);
26603 + write_seqcount_begin(&jiffies_seq);
26604 /* Did we start the jiffies update yet ? */
26605 if (last_jiffies_update.tv64 == 0)
26606 last_jiffies_update = tick_next_period;
26607 period = last_jiffies_update;
26608 - write_sequnlock(&jiffies_lock);
26609 + write_seqcount_end(&jiffies_seq);
26610 + raw_spin_unlock(&jiffies_lock);
26614 @@ -176,6 +181,11 @@ static bool can_stop_full_tick(void)
26618 + if (!arch_irq_work_has_interrupt()) {
26619 + trace_tick_stop(0, "missing irq work interrupt\n");
26623 /* sched_clock_tick() needs us? */
26624 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
26626 @@ -204,6 +214,7 @@ static void nohz_full_kick_work_func(struct irq_work *work)
26628 static DEFINE_PER_CPU(struct irq_work, nohz_full_kick_work) = {
26629 .func = nohz_full_kick_work_func,
26630 + .flags = IRQ_WORK_HARD_IRQ,
26634 @@ -578,10 +589,10 @@ static ktime_t tick_nohz_stop_sched_tick(struct tick_sched *ts,
26636 /* Read jiffies and the time when jiffies were updated last */
26638 - seq = read_seqbegin(&jiffies_lock);
26639 + seq = read_seqcount_begin(&jiffies_seq);
26640 basemono = last_jiffies_update.tv64;
26641 basejiff = jiffies;
26642 - } while (read_seqretry(&jiffies_lock, seq));
26643 + } while (read_seqcount_retry(&jiffies_seq, seq));
26644 ts->last_jiffies = basejiff;
26646 if (rcu_needs_cpu(basemono, &next_rcu) ||
26647 @@ -753,14 +764,7 @@ static bool can_stop_idle_tick(int cpu, struct tick_sched *ts)
26650 if (unlikely(local_softirq_pending() && cpu_online(cpu))) {
26651 - static int ratelimit;
26653 - if (ratelimit < 10 &&
26654 - (local_softirq_pending() & SOFTIRQ_STOP_IDLE_MASK)) {
26655 - pr_warn("NOHZ: local_softirq_pending %02x\n",
26656 - (unsigned int) local_softirq_pending());
26659 + softirq_check_pending_idle();
26663 @@ -1100,6 +1104,7 @@ void tick_setup_sched_timer(void)
26664 * Emulate tick processing via per-CPU hrtimers:
26666 hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
26667 + ts->sched_timer.irqsafe = 1;
26668 ts->sched_timer.function = tick_sched_timer;
26670 /* Get the next period (per cpu) */
26671 diff --git a/kernel/time/timekeeping.c b/kernel/time/timekeeping.c
26672 index 738012d68117..e060b34d5603 100644
26673 --- a/kernel/time/timekeeping.c
26674 +++ b/kernel/time/timekeeping.c
26675 @@ -2070,8 +2070,10 @@ EXPORT_SYMBOL(hardpps);
26677 void xtime_update(unsigned long ticks)
26679 - write_seqlock(&jiffies_lock);
26680 + raw_spin_lock(&jiffies_lock);
26681 + write_seqcount_begin(&jiffies_seq);
26683 - write_sequnlock(&jiffies_lock);
26684 + write_seqcount_end(&jiffies_seq);
26685 + raw_spin_unlock(&jiffies_lock);
26686 update_wall_time();
26688 diff --git a/kernel/time/timekeeping.h b/kernel/time/timekeeping.h
26689 index 704f595ce83f..763a3e5121ff 100644
26690 --- a/kernel/time/timekeeping.h
26691 +++ b/kernel/time/timekeeping.h
26692 @@ -19,7 +19,8 @@ extern void timekeeping_resume(void);
26693 extern void do_timer(unsigned long ticks);
26694 extern void update_wall_time(void);
26696 -extern seqlock_t jiffies_lock;
26697 +extern raw_spinlock_t jiffies_lock;
26698 +extern seqcount_t jiffies_seq;
26700 #define CS_NAME_LEN 32
26702 diff --git a/kernel/time/timer.c b/kernel/time/timer.c
26703 index bbc5d1114583..603699ff9411 100644
26704 --- a/kernel/time/timer.c
26705 +++ b/kernel/time/timer.c
26706 @@ -80,6 +80,9 @@ struct tvec_root {
26709 struct timer_list *running_timer;
26710 +#ifdef CONFIG_PREEMPT_RT_FULL
26711 + wait_queue_head_t wait_for_running_timer;
26713 unsigned long timer_jiffies;
26714 unsigned long next_timer;
26715 unsigned long active_timers;
26716 @@ -777,6 +780,39 @@ static struct tvec_base *lock_timer_base(struct timer_list *timer,
26720 +#ifdef CONFIG_PREEMPT_RT_FULL
26721 +static inline struct tvec_base *switch_timer_base(struct timer_list *timer,
26722 + struct tvec_base *old,
26723 + struct tvec_base *new)
26726 + * We cannot do the below because we might be preempted and
26727 + * then the preempter would see NULL and loop forever.
26729 + if (spin_trylock(&new->lock)) {
26730 + WRITE_ONCE(timer->flags,
26731 + (timer->flags & ~TIMER_BASEMASK) | new->cpu);
26732 + spin_unlock(&old->lock);
26739 +static inline struct tvec_base *switch_timer_base(struct timer_list *timer,
26740 + struct tvec_base *old,
26741 + struct tvec_base *new)
26743 + /* See the comment in lock_timer_base() */
26744 + timer->flags |= TIMER_MIGRATING;
26746 + spin_unlock(&old->lock);
26747 + spin_lock(&new->lock);
26748 + WRITE_ONCE(timer->flags,
26749 + (timer->flags & ~TIMER_BASEMASK) | new->cpu);
26755 __mod_timer(struct timer_list *timer, unsigned long expires,
26756 @@ -807,16 +843,8 @@ __mod_timer(struct timer_list *timer, unsigned long expires,
26757 * handler yet has not finished. This also guarantees that
26758 * the timer is serialized wrt itself.
26760 - if (likely(base->running_timer != timer)) {
26761 - /* See the comment in lock_timer_base() */
26762 - timer->flags |= TIMER_MIGRATING;
26764 - spin_unlock(&base->lock);
26766 - spin_lock(&base->lock);
26767 - WRITE_ONCE(timer->flags,
26768 - (timer->flags & ~TIMER_BASEMASK) | base->cpu);
26770 + if (likely(base->running_timer != timer))
26771 + base = switch_timer_base(timer, base, new_base);
26774 timer->expires = expires;
26775 @@ -1006,6 +1034,33 @@ void add_timer_on(struct timer_list *timer, int cpu)
26777 EXPORT_SYMBOL_GPL(add_timer_on);
26779 +#ifdef CONFIG_PREEMPT_RT_FULL
26781 + * Wait for a running timer
26783 +static void wait_for_running_timer(struct timer_list *timer)
26785 + struct tvec_base *base;
26786 + u32 tf = timer->flags;
26788 + if (tf & TIMER_MIGRATING)
26791 + base = per_cpu_ptr(&tvec_bases, tf & TIMER_CPUMASK);
26792 + wait_event(base->wait_for_running_timer,
26793 + base->running_timer != timer);
26796 +# define wakeup_timer_waiters(b) wake_up_all(&(b)->wait_for_running_timer)
26798 +static inline void wait_for_running_timer(struct timer_list *timer)
26803 +# define wakeup_timer_waiters(b) do { } while (0)
26807 * del_timer - deactive a timer.
26808 * @timer: the timer to be deactivated
26809 @@ -1063,7 +1118,7 @@ int try_to_del_timer_sync(struct timer_list *timer)
26811 EXPORT_SYMBOL(try_to_del_timer_sync);
26814 +#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT_RT_FULL)
26816 * del_timer_sync - deactivate a timer and wait for the handler to finish.
26817 * @timer: the timer to be deactivated
26818 @@ -1123,7 +1178,7 @@ int del_timer_sync(struct timer_list *timer)
26819 int ret = try_to_del_timer_sync(timer);
26823 + wait_for_running_timer(timer);
26826 EXPORT_SYMBOL(del_timer_sync);
26827 @@ -1248,16 +1303,18 @@ static inline void __run_timers(struct tvec_base *base)
26829 spin_unlock(&base->lock);
26830 call_timer_fn(timer, fn, data);
26831 + base->running_timer = NULL;
26832 spin_lock(&base->lock);
26834 spin_unlock_irq(&base->lock);
26835 call_timer_fn(timer, fn, data);
26836 + base->running_timer = NULL;
26837 spin_lock_irq(&base->lock);
26841 - base->running_timer = NULL;
26842 spin_unlock_irq(&base->lock);
26843 + wakeup_timer_waiters(base);
26846 #ifdef CONFIG_NO_HZ_COMMON
26847 @@ -1390,6 +1447,14 @@ u64 get_next_timer_interrupt(unsigned long basej, u64 basem)
26848 if (cpu_is_offline(smp_processor_id()))
26851 +#ifdef CONFIG_PREEMPT_RT_FULL
26853 + * On PREEMPT_RT we cannot sleep here. As a result we can't take
26854 + * the base lock to check when the next timer is pending and so
26855 + * we assume the next jiffy.
26857 + return basem + TICK_NSEC;
26859 spin_lock(&base->lock);
26860 if (base->active_timers) {
26861 if (time_before_eq(base->next_timer, base->timer_jiffies))
26862 @@ -1416,13 +1481,13 @@ void update_process_times(int user_tick)
26864 /* Note: this timer irq context must be accounted for as well. */
26865 account_process_tick(p, user_tick);
26866 + scheduler_tick();
26867 run_local_timers();
26868 rcu_check_callbacks(user_tick);
26869 -#ifdef CONFIG_IRQ_WORK
26870 +#if defined(CONFIG_IRQ_WORK)
26874 - scheduler_tick();
26875 run_posix_cpu_timers(p);
26878 @@ -1433,6 +1498,8 @@ static void run_timer_softirq(struct softirq_action *h)
26880 struct tvec_base *base = this_cpu_ptr(&tvec_bases);
26882 + irq_work_tick_soft();
26884 if (time_after_eq(jiffies, base->timer_jiffies))
26885 __run_timers(base);
26887 @@ -1589,7 +1656,7 @@ static void migrate_timers(int cpu)
26889 BUG_ON(cpu_online(cpu));
26890 old_base = per_cpu_ptr(&tvec_bases, cpu);
26891 - new_base = get_cpu_ptr(&tvec_bases);
26892 + new_base = get_local_ptr(&tvec_bases);
26894 * The caller is globally serialized and nobody else
26895 * takes two locks at once, deadlock is not possible.
26896 @@ -1613,7 +1680,7 @@ static void migrate_timers(int cpu)
26898 spin_unlock(&old_base->lock);
26899 spin_unlock_irq(&new_base->lock);
26900 - put_cpu_ptr(&tvec_bases);
26901 + put_local_ptr(&tvec_bases);
26904 static int timer_cpu_notify(struct notifier_block *self,
26905 @@ -1645,6 +1712,9 @@ static void __init init_timer_cpu(int cpu)
26908 spin_lock_init(&base->lock);
26909 +#ifdef CONFIG_PREEMPT_RT_FULL
26910 + init_waitqueue_head(&base->wait_for_running_timer);
26913 base->timer_jiffies = jiffies;
26914 base->next_timer = base->timer_jiffies;
26915 diff --git a/kernel/trace/Kconfig b/kernel/trace/Kconfig
26916 index e45db6b0d878..364ccd0eb57b 100644
26917 --- a/kernel/trace/Kconfig
26918 +++ b/kernel/trace/Kconfig
26919 @@ -187,6 +187,24 @@ config IRQSOFF_TRACER
26920 enabled. This option and the preempt-off timing option can be
26921 used together or separately.)
26923 +config INTERRUPT_OFF_HIST
26924 + bool "Interrupts-off Latency Histogram"
26925 + depends on IRQSOFF_TRACER
26927 + This option generates continuously updated histograms (one per cpu)
26928 + of the duration of time periods with interrupts disabled. The
26929 + histograms are disabled by default. To enable them, write a non-zero
26932 + /sys/kernel/debug/tracing/latency_hist/enable/preemptirqsoff
26934 + If PREEMPT_OFF_HIST is also selected, additional histograms (one
26935 + per cpu) are generated that accumulate the duration of time periods
26936 + when both interrupts and preemption are disabled. The histogram data
26937 + will be located in the debug file system at
26939 + /sys/kernel/debug/tracing/latency_hist/irqsoff
26941 config PREEMPT_TRACER
26942 bool "Preemption-off Latency Tracer"
26944 @@ -211,6 +229,24 @@ config PREEMPT_TRACER
26945 enabled. This option and the irqs-off timing option can be
26946 used together or separately.)
26948 +config PREEMPT_OFF_HIST
26949 + bool "Preemption-off Latency Histogram"
26950 + depends on PREEMPT_TRACER
26952 + This option generates continuously updated histograms (one per cpu)
26953 + of the duration of time periods with preemption disabled. The
26954 + histograms are disabled by default. To enable them, write a non-zero
26957 + /sys/kernel/debug/tracing/latency_hist/enable/preemptirqsoff
26959 + If INTERRUPT_OFF_HIST is also selected, additional histograms (one
26960 + per cpu) are generated that accumulate the duration of time periods
26961 + when both interrupts and preemption are disabled. The histogram data
26962 + will be located in the debug file system at
26964 + /sys/kernel/debug/tracing/latency_hist/preemptoff
26966 config SCHED_TRACER
26967 bool "Scheduling Latency Tracer"
26968 select GENERIC_TRACER
26969 @@ -221,6 +257,74 @@ config SCHED_TRACER
26970 This tracer tracks the latency of the highest priority task
26971 to be scheduled in, starting from the point it has woken up.
26973 +config WAKEUP_LATENCY_HIST
26974 + bool "Scheduling Latency Histogram"
26975 + depends on SCHED_TRACER
26977 + This option generates continuously updated histograms (one per cpu)
26978 + of the scheduling latency of the highest priority task.
26979 + The histograms are disabled by default. To enable them, write a
26980 + non-zero number to
26982 + /sys/kernel/debug/tracing/latency_hist/enable/wakeup
26984 + Two different algorithms are used, one to determine the latency of
26985 + processes that exclusively use the highest priority of the system and
26986 + another one to determine the latency of processes that share the
26987 + highest system priority with other processes. The former is used to
26988 + improve hardware and system software, the latter to optimize the
26989 + priority design of a given system. The histogram data will be
26990 + located in the debug file system at
26992 + /sys/kernel/debug/tracing/latency_hist/wakeup
26996 + /sys/kernel/debug/tracing/latency_hist/wakeup/sharedprio
26998 + If both Scheduling Latency Histogram and Missed Timer Offsets
26999 + Histogram are selected, additional histogram data will be collected
27000 + that contain, in addition to the wakeup latency, the timer latency, in
27001 + case the wakeup was triggered by an expired timer. These histograms
27002 + are available in the
27004 + /sys/kernel/debug/tracing/latency_hist/timerandwakeup
27006 + directory. They reflect the apparent interrupt and scheduling latency
27007 + and are best suitable to determine the worst-case latency of a given
27008 + system. To enable these histograms, write a non-zero number to
27010 + /sys/kernel/debug/tracing/latency_hist/enable/timerandwakeup
27012 +config MISSED_TIMER_OFFSETS_HIST
27013 + depends on HIGH_RES_TIMERS
27014 + select GENERIC_TRACER
27015 + bool "Missed Timer Offsets Histogram"
27017 + Generate a histogram of missed timer offsets in microseconds. The
27018 + histograms are disabled by default. To enable them, write a non-zero
27021 + /sys/kernel/debug/tracing/latency_hist/enable/missed_timer_offsets
27023 + The histogram data will be located in the debug file system at
27025 + /sys/kernel/debug/tracing/latency_hist/missed_timer_offsets
27027 + If both Scheduling Latency Histogram and Missed Timer Offsets
27028 + Histogram are selected, additional histogram data will be collected
27029 + that contain, in addition to the wakeup latency, the timer latency, in
27030 + case the wakeup was triggered by an expired timer. These histograms
27031 + are available in the
27033 + /sys/kernel/debug/tracing/latency_hist/timerandwakeup
27035 + directory. They reflect the apparent interrupt and scheduling latency
27036 + and are best suitable to determine the worst-case latency of a given
27037 + system. To enable these histograms, write a non-zero number to
27039 + /sys/kernel/debug/tracing/latency_hist/enable/timerandwakeup
27041 config ENABLE_DEFAULT_TRACERS
27042 bool "Trace process context switches and events"
27043 depends on !GENERIC_TRACER
27044 diff --git a/kernel/trace/Makefile b/kernel/trace/Makefile
27045 index 05ea5167e6bb..bc08c67301ae 100644
27046 --- a/kernel/trace/Makefile
27047 +++ b/kernel/trace/Makefile
27048 @@ -40,6 +40,10 @@ obj-$(CONFIG_FUNCTION_TRACER) += trace_functions.o
27049 obj-$(CONFIG_IRQSOFF_TRACER) += trace_irqsoff.o
27050 obj-$(CONFIG_PREEMPT_TRACER) += trace_irqsoff.o
27051 obj-$(CONFIG_SCHED_TRACER) += trace_sched_wakeup.o
27052 +obj-$(CONFIG_INTERRUPT_OFF_HIST) += latency_hist.o
27053 +obj-$(CONFIG_PREEMPT_OFF_HIST) += latency_hist.o
27054 +obj-$(CONFIG_WAKEUP_LATENCY_HIST) += latency_hist.o
27055 +obj-$(CONFIG_MISSED_TIMER_OFFSETS_HIST) += latency_hist.o
27056 obj-$(CONFIG_NOP_TRACER) += trace_nop.o
27057 obj-$(CONFIG_STACK_TRACER) += trace_stack.o
27058 obj-$(CONFIG_MMIOTRACE) += trace_mmiotrace.o
27059 diff --git a/kernel/trace/latency_hist.c b/kernel/trace/latency_hist.c
27060 new file mode 100644
27061 index 000000000000..7f6ee70dea41
27063 +++ b/kernel/trace/latency_hist.c
27066 + * kernel/trace/latency_hist.c
27068 + * Add support for histograms of preemption-off latency and
27069 + * interrupt-off latency and wakeup latency, it depends on
27070 + * Real-Time Preemption Support.
27072 + * Copyright (C) 2005 MontaVista Software, Inc.
27073 + * Yi Yang <yyang@ch.mvista.com>
27075 + * Converted to work with the new latency tracer.
27076 + * Copyright (C) 2008 Red Hat, Inc.
27077 + * Steven Rostedt <srostedt@redhat.com>
27080 +#include <linux/module.h>
27081 +#include <linux/debugfs.h>
27082 +#include <linux/seq_file.h>
27083 +#include <linux/percpu.h>
27084 +#include <linux/kallsyms.h>
27085 +#include <linux/uaccess.h>
27086 +#include <linux/sched.h>
27087 +#include <linux/sched/rt.h>
27088 +#include <linux/slab.h>
27089 +#include <linux/atomic.h>
27090 +#include <asm/div64.h>
27092 +#include "trace.h"
27093 +#include <trace/events/sched.h>
27095 +#define NSECS_PER_USECS 1000L
27097 +#define CREATE_TRACE_POINTS
27098 +#include <trace/events/hist.h>
27101 + IRQSOFF_LATENCY = 0,
27102 + PREEMPTOFF_LATENCY,
27103 + PREEMPTIRQSOFF_LATENCY,
27105 + WAKEUP_LATENCY_SHAREDPRIO,
27106 + MISSED_TIMER_OFFSETS,
27107 + TIMERANDWAKEUP_LATENCY,
27108 + MAX_LATENCY_TYPE,
27111 +#define MAX_ENTRY_NUM 10240
27113 +struct hist_data {
27114 + atomic_t hist_mode; /* 0 log, 1 don't log */
27115 + long offset; /* set it to MAX_ENTRY_NUM/2 for a bipolar scale */
27118 + unsigned long long below_hist_bound_samples;
27119 + unsigned long long above_hist_bound_samples;
27120 + long long accumulate_lat;
27121 + unsigned long long total_samples;
27122 + unsigned long long hist_array[MAX_ENTRY_NUM];
27125 +struct enable_data {
27126 + int latency_type;
27130 +static char *latency_hist_dir_root = "latency_hist";
27132 +#ifdef CONFIG_INTERRUPT_OFF_HIST
27133 +static DEFINE_PER_CPU(struct hist_data, irqsoff_hist);
27134 +static char *irqsoff_hist_dir = "irqsoff";
27135 +static DEFINE_PER_CPU(cycles_t, hist_irqsoff_start);
27136 +static DEFINE_PER_CPU(int, hist_irqsoff_counting);
27139 +#ifdef CONFIG_PREEMPT_OFF_HIST
27140 +static DEFINE_PER_CPU(struct hist_data, preemptoff_hist);
27141 +static char *preemptoff_hist_dir = "preemptoff";
27142 +static DEFINE_PER_CPU(cycles_t, hist_preemptoff_start);
27143 +static DEFINE_PER_CPU(int, hist_preemptoff_counting);
27146 +#if defined(CONFIG_PREEMPT_OFF_HIST) && defined(CONFIG_INTERRUPT_OFF_HIST)
27147 +static DEFINE_PER_CPU(struct hist_data, preemptirqsoff_hist);
27148 +static char *preemptirqsoff_hist_dir = "preemptirqsoff";
27149 +static DEFINE_PER_CPU(cycles_t, hist_preemptirqsoff_start);
27150 +static DEFINE_PER_CPU(int, hist_preemptirqsoff_counting);
27153 +#if defined(CONFIG_PREEMPT_OFF_HIST) || defined(CONFIG_INTERRUPT_OFF_HIST)
27154 +static notrace void probe_preemptirqsoff_hist(void *v, int reason, int start);
27155 +static struct enable_data preemptirqsoff_enabled_data = {
27156 + .latency_type = PREEMPTIRQSOFF_LATENCY,
27161 +#if defined(CONFIG_WAKEUP_LATENCY_HIST) || \
27162 + defined(CONFIG_MISSED_TIMER_OFFSETS_HIST)
27163 +struct maxlatproc_data {
27164 + char comm[FIELD_SIZEOF(struct task_struct, comm)];
27165 + char current_comm[FIELD_SIZEOF(struct task_struct, comm)];
27169 + int current_prio;
27171 + long timeroffset;
27172 + cycle_t timestamp;
27176 +#ifdef CONFIG_WAKEUP_LATENCY_HIST
27177 +static DEFINE_PER_CPU(struct hist_data, wakeup_latency_hist);
27178 +static DEFINE_PER_CPU(struct hist_data, wakeup_latency_hist_sharedprio);
27179 +static char *wakeup_latency_hist_dir = "wakeup";
27180 +static char *wakeup_latency_hist_dir_sharedprio = "sharedprio";
27181 +static notrace void probe_wakeup_latency_hist_start(void *v,
27182 + struct task_struct *p);
27183 +static notrace void probe_wakeup_latency_hist_stop(void *v,
27184 + bool preempt, struct task_struct *prev, struct task_struct *next);
27185 +static notrace void probe_sched_migrate_task(void *,
27186 + struct task_struct *task, int cpu);
27187 +static struct enable_data wakeup_latency_enabled_data = {
27188 + .latency_type = WAKEUP_LATENCY,
27191 +static DEFINE_PER_CPU(struct maxlatproc_data, wakeup_maxlatproc);
27192 +static DEFINE_PER_CPU(struct maxlatproc_data, wakeup_maxlatproc_sharedprio);
27193 +static DEFINE_PER_CPU(struct task_struct *, wakeup_task);
27194 +static DEFINE_PER_CPU(int, wakeup_sharedprio);
27195 +static unsigned long wakeup_pid;
27198 +#ifdef CONFIG_MISSED_TIMER_OFFSETS_HIST
27199 +static DEFINE_PER_CPU(struct hist_data, missed_timer_offsets);
27200 +static char *missed_timer_offsets_dir = "missed_timer_offsets";
27201 +static notrace void probe_hrtimer_interrupt(void *v, int cpu,
27202 + long long offset, struct task_struct *curr, struct task_struct *task);
27203 +static struct enable_data missed_timer_offsets_enabled_data = {
27204 + .latency_type = MISSED_TIMER_OFFSETS,
27207 +static DEFINE_PER_CPU(struct maxlatproc_data, missed_timer_offsets_maxlatproc);
27208 +static unsigned long missed_timer_offsets_pid;
27211 +#if defined(CONFIG_WAKEUP_LATENCY_HIST) && \
27212 + defined(CONFIG_MISSED_TIMER_OFFSETS_HIST)
27213 +static DEFINE_PER_CPU(struct hist_data, timerandwakeup_latency_hist);
27214 +static char *timerandwakeup_latency_hist_dir = "timerandwakeup";
27215 +static struct enable_data timerandwakeup_enabled_data = {
27216 + .latency_type = TIMERANDWAKEUP_LATENCY,
27219 +static DEFINE_PER_CPU(struct maxlatproc_data, timerandwakeup_maxlatproc);
27222 +void notrace latency_hist(int latency_type, int cpu, long latency,
27223 + long timeroffset, cycle_t stop,
27224 + struct task_struct *p)
27226 + struct hist_data *my_hist;
27227 +#if defined(CONFIG_WAKEUP_LATENCY_HIST) || \
27228 + defined(CONFIG_MISSED_TIMER_OFFSETS_HIST)
27229 + struct maxlatproc_data *mp = NULL;
27232 + if (!cpu_possible(cpu) || latency_type < 0 ||
27233 + latency_type >= MAX_LATENCY_TYPE)
27236 + switch (latency_type) {
27237 +#ifdef CONFIG_INTERRUPT_OFF_HIST
27238 + case IRQSOFF_LATENCY:
27239 + my_hist = &per_cpu(irqsoff_hist, cpu);
27242 +#ifdef CONFIG_PREEMPT_OFF_HIST
27243 + case PREEMPTOFF_LATENCY:
27244 + my_hist = &per_cpu(preemptoff_hist, cpu);
27247 +#if defined(CONFIG_PREEMPT_OFF_HIST) && defined(CONFIG_INTERRUPT_OFF_HIST)
27248 + case PREEMPTIRQSOFF_LATENCY:
27249 + my_hist = &per_cpu(preemptirqsoff_hist, cpu);
27252 +#ifdef CONFIG_WAKEUP_LATENCY_HIST
27253 + case WAKEUP_LATENCY:
27254 + my_hist = &per_cpu(wakeup_latency_hist, cpu);
27255 + mp = &per_cpu(wakeup_maxlatproc, cpu);
27257 + case WAKEUP_LATENCY_SHAREDPRIO:
27258 + my_hist = &per_cpu(wakeup_latency_hist_sharedprio, cpu);
27259 + mp = &per_cpu(wakeup_maxlatproc_sharedprio, cpu);
27262 +#ifdef CONFIG_MISSED_TIMER_OFFSETS_HIST
27263 + case MISSED_TIMER_OFFSETS:
27264 + my_hist = &per_cpu(missed_timer_offsets, cpu);
27265 + mp = &per_cpu(missed_timer_offsets_maxlatproc, cpu);
27268 +#if defined(CONFIG_WAKEUP_LATENCY_HIST) && \
27269 + defined(CONFIG_MISSED_TIMER_OFFSETS_HIST)
27270 + case TIMERANDWAKEUP_LATENCY:
27271 + my_hist = &per_cpu(timerandwakeup_latency_hist, cpu);
27272 + mp = &per_cpu(timerandwakeup_maxlatproc, cpu);
27280 + latency += my_hist->offset;
27282 + if (atomic_read(&my_hist->hist_mode) == 0)
27285 + if (latency < 0 || latency >= MAX_ENTRY_NUM) {
27287 + my_hist->below_hist_bound_samples++;
27289 + my_hist->above_hist_bound_samples++;
27291 + my_hist->hist_array[latency]++;
27293 + if (unlikely(latency > my_hist->max_lat ||
27294 + my_hist->min_lat == LONG_MAX)) {
27295 +#if defined(CONFIG_WAKEUP_LATENCY_HIST) || \
27296 + defined(CONFIG_MISSED_TIMER_OFFSETS_HIST)
27297 + if (latency_type == WAKEUP_LATENCY ||
27298 + latency_type == WAKEUP_LATENCY_SHAREDPRIO ||
27299 + latency_type == MISSED_TIMER_OFFSETS ||
27300 + latency_type == TIMERANDWAKEUP_LATENCY) {
27301 + strncpy(mp->comm, p->comm, sizeof(mp->comm));
27302 + strncpy(mp->current_comm, current->comm,
27303 + sizeof(mp->current_comm));
27304 + mp->pid = task_pid_nr(p);
27305 + mp->current_pid = task_pid_nr(current);
27306 + mp->prio = p->prio;
27307 + mp->current_prio = current->prio;
27308 + mp->latency = latency;
27309 + mp->timeroffset = timeroffset;
27310 + mp->timestamp = stop;
27313 + my_hist->max_lat = latency;
27315 + if (unlikely(latency < my_hist->min_lat))
27316 + my_hist->min_lat = latency;
27317 + my_hist->total_samples++;
27318 + my_hist->accumulate_lat += latency;
27321 +static void *l_start(struct seq_file *m, loff_t *pos)
27323 + loff_t *index_ptr = NULL;
27324 + loff_t index = *pos;
27325 + struct hist_data *my_hist = m->private;
27327 + if (index == 0) {
27328 + char minstr[32], avgstr[32], maxstr[32];
27330 + atomic_dec(&my_hist->hist_mode);
27332 + if (likely(my_hist->total_samples)) {
27333 + long avg = (long) div64_s64(my_hist->accumulate_lat,
27334 + my_hist->total_samples);
27335 + snprintf(minstr, sizeof(minstr), "%ld",
27336 + my_hist->min_lat - my_hist->offset);
27337 + snprintf(avgstr, sizeof(avgstr), "%ld",
27338 + avg - my_hist->offset);
27339 + snprintf(maxstr, sizeof(maxstr), "%ld",
27340 + my_hist->max_lat - my_hist->offset);
27342 + strcpy(minstr, "<undef>");
27343 + strcpy(avgstr, minstr);
27344 + strcpy(maxstr, minstr);
27347 + seq_printf(m, "#Minimum latency: %s microseconds\n"
27348 + "#Average latency: %s microseconds\n"
27349 + "#Maximum latency: %s microseconds\n"
27350 + "#Total samples: %llu\n"
27351 + "#There are %llu samples lower than %ld"
27352 + " microseconds.\n"
27353 + "#There are %llu samples greater or equal"
27354 + " than %ld microseconds.\n"
27355 + "#usecs\t%16s\n",
27356 + minstr, avgstr, maxstr,
27357 + my_hist->total_samples,
27358 + my_hist->below_hist_bound_samples,
27359 + -my_hist->offset,
27360 + my_hist->above_hist_bound_samples,
27361 + MAX_ENTRY_NUM - my_hist->offset,
27364 + if (index < MAX_ENTRY_NUM) {
27365 + index_ptr = kmalloc(sizeof(loff_t), GFP_KERNEL);
27367 + *index_ptr = index;
27370 + return index_ptr;
27373 +static void *l_next(struct seq_file *m, void *p, loff_t *pos)
27375 + loff_t *index_ptr = p;
27376 + struct hist_data *my_hist = m->private;
27378 + if (++*pos >= MAX_ENTRY_NUM) {
27379 + atomic_inc(&my_hist->hist_mode);
27382 + *index_ptr = *pos;
27383 + return index_ptr;
27386 +static void l_stop(struct seq_file *m, void *p)
27391 +static int l_show(struct seq_file *m, void *p)
27393 + int index = *(loff_t *) p;
27394 + struct hist_data *my_hist = m->private;
27396 + seq_printf(m, "%6ld\t%16llu\n", index - my_hist->offset,
27397 + my_hist->hist_array[index]);
27401 +static const struct seq_operations latency_hist_seq_op = {
27402 + .start = l_start,
27408 +static int latency_hist_open(struct inode *inode, struct file *file)
27412 + ret = seq_open(file, &latency_hist_seq_op);
27414 + struct seq_file *seq = file->private_data;
27415 + seq->private = inode->i_private;
27420 +static const struct file_operations latency_hist_fops = {
27421 + .open = latency_hist_open,
27422 + .read = seq_read,
27423 + .llseek = seq_lseek,
27424 + .release = seq_release,
27427 +#if defined(CONFIG_WAKEUP_LATENCY_HIST) || \
27428 + defined(CONFIG_MISSED_TIMER_OFFSETS_HIST)
27429 +static void clear_maxlatprocdata(struct maxlatproc_data *mp)
27431 + mp->comm[0] = mp->current_comm[0] = '\0';
27432 + mp->prio = mp->current_prio = mp->pid = mp->current_pid =
27433 + mp->latency = mp->timeroffset = -1;
27434 + mp->timestamp = 0;
27438 +static void hist_reset(struct hist_data *hist)
27440 + atomic_dec(&hist->hist_mode);
27442 + memset(hist->hist_array, 0, sizeof(hist->hist_array));
27443 + hist->below_hist_bound_samples = 0ULL;
27444 + hist->above_hist_bound_samples = 0ULL;
27445 + hist->min_lat = LONG_MAX;
27446 + hist->max_lat = LONG_MIN;
27447 + hist->total_samples = 0ULL;
27448 + hist->accumulate_lat = 0LL;
27450 + atomic_inc(&hist->hist_mode);
27454 +latency_hist_reset(struct file *file, const char __user *a,
27455 + size_t size, loff_t *off)
27458 + struct hist_data *hist = NULL;
27459 +#if defined(CONFIG_WAKEUP_LATENCY_HIST) || \
27460 + defined(CONFIG_MISSED_TIMER_OFFSETS_HIST)
27461 + struct maxlatproc_data *mp = NULL;
27463 + off_t latency_type = (off_t) file->private_data;
27465 + for_each_online_cpu(cpu) {
27467 + switch (latency_type) {
27468 +#ifdef CONFIG_PREEMPT_OFF_HIST
27469 + case PREEMPTOFF_LATENCY:
27470 + hist = &per_cpu(preemptoff_hist, cpu);
27473 +#ifdef CONFIG_INTERRUPT_OFF_HIST
27474 + case IRQSOFF_LATENCY:
27475 + hist = &per_cpu(irqsoff_hist, cpu);
27478 +#if defined(CONFIG_INTERRUPT_OFF_HIST) && defined(CONFIG_PREEMPT_OFF_HIST)
27479 + case PREEMPTIRQSOFF_LATENCY:
27480 + hist = &per_cpu(preemptirqsoff_hist, cpu);
27483 +#ifdef CONFIG_WAKEUP_LATENCY_HIST
27484 + case WAKEUP_LATENCY:
27485 + hist = &per_cpu(wakeup_latency_hist, cpu);
27486 + mp = &per_cpu(wakeup_maxlatproc, cpu);
27488 + case WAKEUP_LATENCY_SHAREDPRIO:
27489 + hist = &per_cpu(wakeup_latency_hist_sharedprio, cpu);
27490 + mp = &per_cpu(wakeup_maxlatproc_sharedprio, cpu);
27493 +#ifdef CONFIG_MISSED_TIMER_OFFSETS_HIST
27494 + case MISSED_TIMER_OFFSETS:
27495 + hist = &per_cpu(missed_timer_offsets, cpu);
27496 + mp = &per_cpu(missed_timer_offsets_maxlatproc, cpu);
27499 +#if defined(CONFIG_WAKEUP_LATENCY_HIST) && \
27500 + defined(CONFIG_MISSED_TIMER_OFFSETS_HIST)
27501 + case TIMERANDWAKEUP_LATENCY:
27502 + hist = &per_cpu(timerandwakeup_latency_hist, cpu);
27503 + mp = &per_cpu(timerandwakeup_maxlatproc, cpu);
27508 + hist_reset(hist);
27509 +#if defined(CONFIG_WAKEUP_LATENCY_HIST) || \
27510 + defined(CONFIG_MISSED_TIMER_OFFSETS_HIST)
27511 + if (latency_type == WAKEUP_LATENCY ||
27512 + latency_type == WAKEUP_LATENCY_SHAREDPRIO ||
27513 + latency_type == MISSED_TIMER_OFFSETS ||
27514 + latency_type == TIMERANDWAKEUP_LATENCY)
27515 + clear_maxlatprocdata(mp);
27522 +#if defined(CONFIG_WAKEUP_LATENCY_HIST) || \
27523 + defined(CONFIG_MISSED_TIMER_OFFSETS_HIST)
27525 +show_pid(struct file *file, char __user *ubuf, size_t cnt, loff_t *ppos)
27529 + unsigned long *this_pid = file->private_data;
27531 + r = snprintf(buf, sizeof(buf), "%lu\n", *this_pid);
27532 + return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
27535 +static ssize_t do_pid(struct file *file, const char __user *ubuf,
27536 + size_t cnt, loff_t *ppos)
27539 + unsigned long pid;
27540 + unsigned long *this_pid = file->private_data;
27542 + if (cnt >= sizeof(buf))
27545 + if (copy_from_user(&buf, ubuf, cnt))
27550 + if (kstrtoul(buf, 10, &pid))
27559 +#if defined(CONFIG_WAKEUP_LATENCY_HIST) || \
27560 + defined(CONFIG_MISSED_TIMER_OFFSETS_HIST)
27562 +show_maxlatproc(struct file *file, char __user *ubuf, size_t cnt, loff_t *ppos)
27565 + struct maxlatproc_data *mp = file->private_data;
27566 + int strmaxlen = (TASK_COMM_LEN * 2) + (8 * 8);
27567 + unsigned long long t;
27568 + unsigned long usecs, secs;
27571 + if (mp->pid == -1 || mp->current_pid == -1) {
27572 + buf = "(none)\n";
27573 + return simple_read_from_buffer(ubuf, cnt, ppos, buf,
27577 + buf = kmalloc(strmaxlen, GFP_KERNEL);
27581 + t = ns2usecs(mp->timestamp);
27582 + usecs = do_div(t, USEC_PER_SEC);
27583 + secs = (unsigned long) t;
27584 + r = snprintf(buf, strmaxlen,
27585 + "%d %d %ld (%ld) %s <- %d %d %s %lu.%06lu\n", mp->pid,
27586 + MAX_RT_PRIO-1 - mp->prio, mp->latency, mp->timeroffset, mp->comm,
27587 + mp->current_pid, MAX_RT_PRIO-1 - mp->current_prio, mp->current_comm,
27589 + r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
27596 +show_enable(struct file *file, char __user *ubuf, size_t cnt, loff_t *ppos)
27599 + struct enable_data *ed = file->private_data;
27602 + r = snprintf(buf, sizeof(buf), "%d\n", ed->enabled);
27603 + return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
27607 +do_enable(struct file *file, const char __user *ubuf, size_t cnt, loff_t *ppos)
27611 + struct enable_data *ed = file->private_data;
27613 + if (cnt >= sizeof(buf))
27616 + if (copy_from_user(&buf, ubuf, cnt))
27621 + if (kstrtoul(buf, 10, &enable))
27624 + if ((enable && ed->enabled) || (!enable && !ed->enabled))
27630 + switch (ed->latency_type) {
27631 +#if defined(CONFIG_INTERRUPT_OFF_HIST) || defined(CONFIG_PREEMPT_OFF_HIST)
27632 + case PREEMPTIRQSOFF_LATENCY:
27633 + ret = register_trace_preemptirqsoff_hist(
27634 + probe_preemptirqsoff_hist, NULL);
27636 + pr_info("wakeup trace: Couldn't assign "
27637 + "probe_preemptirqsoff_hist "
27638 + "to trace_preemptirqsoff_hist\n");
27643 +#ifdef CONFIG_WAKEUP_LATENCY_HIST
27644 + case WAKEUP_LATENCY:
27645 + ret = register_trace_sched_wakeup(
27646 + probe_wakeup_latency_hist_start, NULL);
27648 + pr_info("wakeup trace: Couldn't assign "
27649 + "probe_wakeup_latency_hist_start "
27650 + "to trace_sched_wakeup\n");
27653 + ret = register_trace_sched_wakeup_new(
27654 + probe_wakeup_latency_hist_start, NULL);
27656 + pr_info("wakeup trace: Couldn't assign "
27657 + "probe_wakeup_latency_hist_start "
27658 + "to trace_sched_wakeup_new\n");
27659 + unregister_trace_sched_wakeup(
27660 + probe_wakeup_latency_hist_start, NULL);
27663 + ret = register_trace_sched_switch(
27664 + probe_wakeup_latency_hist_stop, NULL);
27666 + pr_info("wakeup trace: Couldn't assign "
27667 + "probe_wakeup_latency_hist_stop "
27668 + "to trace_sched_switch\n");
27669 + unregister_trace_sched_wakeup(
27670 + probe_wakeup_latency_hist_start, NULL);
27671 + unregister_trace_sched_wakeup_new(
27672 + probe_wakeup_latency_hist_start, NULL);
27675 + ret = register_trace_sched_migrate_task(
27676 + probe_sched_migrate_task, NULL);
27678 + pr_info("wakeup trace: Couldn't assign "
27679 + "probe_sched_migrate_task "
27680 + "to trace_sched_migrate_task\n");
27681 + unregister_trace_sched_wakeup(
27682 + probe_wakeup_latency_hist_start, NULL);
27683 + unregister_trace_sched_wakeup_new(
27684 + probe_wakeup_latency_hist_start, NULL);
27685 + unregister_trace_sched_switch(
27686 + probe_wakeup_latency_hist_stop, NULL);
27691 +#ifdef CONFIG_MISSED_TIMER_OFFSETS_HIST
27692 + case MISSED_TIMER_OFFSETS:
27693 + ret = register_trace_hrtimer_interrupt(
27694 + probe_hrtimer_interrupt, NULL);
27696 + pr_info("wakeup trace: Couldn't assign "
27697 + "probe_hrtimer_interrupt "
27698 + "to trace_hrtimer_interrupt\n");
27703 +#if defined(CONFIG_WAKEUP_LATENCY_HIST) && \
27704 + defined(CONFIG_MISSED_TIMER_OFFSETS_HIST)
27705 + case TIMERANDWAKEUP_LATENCY:
27706 + if (!wakeup_latency_enabled_data.enabled ||
27707 + !missed_timer_offsets_enabled_data.enabled)
27715 + switch (ed->latency_type) {
27716 +#if defined(CONFIG_INTERRUPT_OFF_HIST) || defined(CONFIG_PREEMPT_OFF_HIST)
27717 + case PREEMPTIRQSOFF_LATENCY:
27721 + unregister_trace_preemptirqsoff_hist(
27722 + probe_preemptirqsoff_hist, NULL);
27723 + for_each_online_cpu(cpu) {
27724 +#ifdef CONFIG_INTERRUPT_OFF_HIST
27725 + per_cpu(hist_irqsoff_counting,
27728 +#ifdef CONFIG_PREEMPT_OFF_HIST
27729 + per_cpu(hist_preemptoff_counting,
27732 +#if defined(CONFIG_INTERRUPT_OFF_HIST) && defined(CONFIG_PREEMPT_OFF_HIST)
27733 + per_cpu(hist_preemptirqsoff_counting,
27740 +#ifdef CONFIG_WAKEUP_LATENCY_HIST
27741 + case WAKEUP_LATENCY:
27745 + unregister_trace_sched_wakeup(
27746 + probe_wakeup_latency_hist_start, NULL);
27747 + unregister_trace_sched_wakeup_new(
27748 + probe_wakeup_latency_hist_start, NULL);
27749 + unregister_trace_sched_switch(
27750 + probe_wakeup_latency_hist_stop, NULL);
27751 + unregister_trace_sched_migrate_task(
27752 + probe_sched_migrate_task, NULL);
27754 + for_each_online_cpu(cpu) {
27755 + per_cpu(wakeup_task, cpu) = NULL;
27756 + per_cpu(wakeup_sharedprio, cpu) = 0;
27759 +#ifdef CONFIG_MISSED_TIMER_OFFSETS_HIST
27760 + timerandwakeup_enabled_data.enabled = 0;
27764 +#ifdef CONFIG_MISSED_TIMER_OFFSETS_HIST
27765 + case MISSED_TIMER_OFFSETS:
27766 + unregister_trace_hrtimer_interrupt(
27767 + probe_hrtimer_interrupt, NULL);
27768 +#ifdef CONFIG_WAKEUP_LATENCY_HIST
27769 + timerandwakeup_enabled_data.enabled = 0;
27777 + ed->enabled = enable;
27781 +static const struct file_operations latency_hist_reset_fops = {
27782 + .open = tracing_open_generic,
27783 + .write = latency_hist_reset,
27786 +static const struct file_operations enable_fops = {
27787 + .open = tracing_open_generic,
27788 + .read = show_enable,
27789 + .write = do_enable,
27792 +#if defined(CONFIG_WAKEUP_LATENCY_HIST) || \
27793 + defined(CONFIG_MISSED_TIMER_OFFSETS_HIST)
27794 +static const struct file_operations pid_fops = {
27795 + .open = tracing_open_generic,
27796 + .read = show_pid,
27800 +static const struct file_operations maxlatproc_fops = {
27801 + .open = tracing_open_generic,
27802 + .read = show_maxlatproc,
27806 +#if defined(CONFIG_INTERRUPT_OFF_HIST) || defined(CONFIG_PREEMPT_OFF_HIST)
27807 +static notrace void probe_preemptirqsoff_hist(void *v, int reason,
27810 + int cpu = raw_smp_processor_id();
27811 + int time_set = 0;
27814 + cycle_t uninitialized_var(start);
27816 + if (!preempt_count() && !irqs_disabled())
27819 +#ifdef CONFIG_INTERRUPT_OFF_HIST
27820 + if ((reason == IRQS_OFF || reason == TRACE_START) &&
27821 + !per_cpu(hist_irqsoff_counting, cpu)) {
27822 + per_cpu(hist_irqsoff_counting, cpu) = 1;
27823 + start = ftrace_now(cpu);
27825 + per_cpu(hist_irqsoff_start, cpu) = start;
27829 +#ifdef CONFIG_PREEMPT_OFF_HIST
27830 + if ((reason == PREEMPT_OFF || reason == TRACE_START) &&
27831 + !per_cpu(hist_preemptoff_counting, cpu)) {
27832 + per_cpu(hist_preemptoff_counting, cpu) = 1;
27833 + if (!(time_set++))
27834 + start = ftrace_now(cpu);
27835 + per_cpu(hist_preemptoff_start, cpu) = start;
27839 +#if defined(CONFIG_INTERRUPT_OFF_HIST) && defined(CONFIG_PREEMPT_OFF_HIST)
27840 + if (per_cpu(hist_irqsoff_counting, cpu) &&
27841 + per_cpu(hist_preemptoff_counting, cpu) &&
27842 + !per_cpu(hist_preemptirqsoff_counting, cpu)) {
27843 + per_cpu(hist_preemptirqsoff_counting, cpu) = 1;
27845 + start = ftrace_now(cpu);
27846 + per_cpu(hist_preemptirqsoff_start, cpu) = start;
27850 + cycle_t uninitialized_var(stop);
27852 +#ifdef CONFIG_INTERRUPT_OFF_HIST
27853 + if ((reason == IRQS_ON || reason == TRACE_STOP) &&
27854 + per_cpu(hist_irqsoff_counting, cpu)) {
27855 + cycle_t start = per_cpu(hist_irqsoff_start, cpu);
27857 + stop = ftrace_now(cpu);
27860 + long latency = ((long) (stop - start)) /
27863 + latency_hist(IRQSOFF_LATENCY, cpu, latency, 0,
27866 + per_cpu(hist_irqsoff_counting, cpu) = 0;
27870 +#ifdef CONFIG_PREEMPT_OFF_HIST
27871 + if ((reason == PREEMPT_ON || reason == TRACE_STOP) &&
27872 + per_cpu(hist_preemptoff_counting, cpu)) {
27873 + cycle_t start = per_cpu(hist_preemptoff_start, cpu);
27875 + if (!(time_set++))
27876 + stop = ftrace_now(cpu);
27878 + long latency = ((long) (stop - start)) /
27881 + latency_hist(PREEMPTOFF_LATENCY, cpu, latency,
27884 + per_cpu(hist_preemptoff_counting, cpu) = 0;
27888 +#if defined(CONFIG_INTERRUPT_OFF_HIST) && defined(CONFIG_PREEMPT_OFF_HIST)
27889 + if ((!per_cpu(hist_irqsoff_counting, cpu) ||
27890 + !per_cpu(hist_preemptoff_counting, cpu)) &&
27891 + per_cpu(hist_preemptirqsoff_counting, cpu)) {
27892 + cycle_t start = per_cpu(hist_preemptirqsoff_start, cpu);
27895 + stop = ftrace_now(cpu);
27897 + long latency = ((long) (stop - start)) /
27900 + latency_hist(PREEMPTIRQSOFF_LATENCY, cpu,
27901 + latency, 0, stop, NULL);
27903 + per_cpu(hist_preemptirqsoff_counting, cpu) = 0;
27910 +#ifdef CONFIG_WAKEUP_LATENCY_HIST
27911 +static DEFINE_RAW_SPINLOCK(wakeup_lock);
27912 +static notrace void probe_sched_migrate_task(void *v, struct task_struct *task,
27915 + int old_cpu = task_cpu(task);
27917 + if (cpu != old_cpu) {
27918 + unsigned long flags;
27919 + struct task_struct *cpu_wakeup_task;
27921 + raw_spin_lock_irqsave(&wakeup_lock, flags);
27923 + cpu_wakeup_task = per_cpu(wakeup_task, old_cpu);
27924 + if (task == cpu_wakeup_task) {
27925 + put_task_struct(cpu_wakeup_task);
27926 + per_cpu(wakeup_task, old_cpu) = NULL;
27927 + cpu_wakeup_task = per_cpu(wakeup_task, cpu) = task;
27928 + get_task_struct(cpu_wakeup_task);
27931 + raw_spin_unlock_irqrestore(&wakeup_lock, flags);
27935 +static notrace void probe_wakeup_latency_hist_start(void *v,
27936 + struct task_struct *p)
27938 + unsigned long flags;
27939 + struct task_struct *curr = current;
27940 + int cpu = task_cpu(p);
27941 + struct task_struct *cpu_wakeup_task;
27943 + raw_spin_lock_irqsave(&wakeup_lock, flags);
27945 + cpu_wakeup_task = per_cpu(wakeup_task, cpu);
27947 + if (wakeup_pid) {
27948 + if ((cpu_wakeup_task && p->prio == cpu_wakeup_task->prio) ||
27949 + p->prio == curr->prio)
27950 + per_cpu(wakeup_sharedprio, cpu) = 1;
27951 + if (likely(wakeup_pid != task_pid_nr(p)))
27954 + if (likely(!rt_task(p)) ||
27955 + (cpu_wakeup_task && p->prio > cpu_wakeup_task->prio) ||
27956 + p->prio > curr->prio)
27958 + if ((cpu_wakeup_task && p->prio == cpu_wakeup_task->prio) ||
27959 + p->prio == curr->prio)
27960 + per_cpu(wakeup_sharedprio, cpu) = 1;
27963 + if (cpu_wakeup_task)
27964 + put_task_struct(cpu_wakeup_task);
27965 + cpu_wakeup_task = per_cpu(wakeup_task, cpu) = p;
27966 + get_task_struct(cpu_wakeup_task);
27967 + cpu_wakeup_task->preempt_timestamp_hist =
27968 + ftrace_now(raw_smp_processor_id());
27970 + raw_spin_unlock_irqrestore(&wakeup_lock, flags);
27973 +static notrace void probe_wakeup_latency_hist_stop(void *v,
27974 + bool preempt, struct task_struct *prev, struct task_struct *next)
27976 + unsigned long flags;
27977 + int cpu = task_cpu(next);
27980 + struct task_struct *cpu_wakeup_task;
27982 + raw_spin_lock_irqsave(&wakeup_lock, flags);
27984 + cpu_wakeup_task = per_cpu(wakeup_task, cpu);
27986 + if (cpu_wakeup_task == NULL)
27989 + /* Already running? */
27990 + if (unlikely(current == cpu_wakeup_task))
27993 + if (next != cpu_wakeup_task) {
27994 + if (next->prio < cpu_wakeup_task->prio)
27997 + if (next->prio == cpu_wakeup_task->prio)
27998 + per_cpu(wakeup_sharedprio, cpu) = 1;
28003 + if (current->prio == cpu_wakeup_task->prio)
28004 + per_cpu(wakeup_sharedprio, cpu) = 1;
28007 + * The task we are waiting for is about to be switched to.
28008 + * Calculate latency and store it in histogram.
28010 + stop = ftrace_now(raw_smp_processor_id());
28012 + latency = ((long) (stop - next->preempt_timestamp_hist)) /
28015 + if (per_cpu(wakeup_sharedprio, cpu)) {
28016 + latency_hist(WAKEUP_LATENCY_SHAREDPRIO, cpu, latency, 0, stop,
28018 + per_cpu(wakeup_sharedprio, cpu) = 0;
28020 + latency_hist(WAKEUP_LATENCY, cpu, latency, 0, stop, next);
28021 +#ifdef CONFIG_MISSED_TIMER_OFFSETS_HIST
28022 + if (timerandwakeup_enabled_data.enabled) {
28023 + latency_hist(TIMERANDWAKEUP_LATENCY, cpu,
28024 + next->timer_offset + latency, next->timer_offset,
28031 +#ifdef CONFIG_MISSED_TIMER_OFFSETS_HIST
28032 + next->timer_offset = 0;
28034 + put_task_struct(cpu_wakeup_task);
28035 + per_cpu(wakeup_task, cpu) = NULL;
28037 + raw_spin_unlock_irqrestore(&wakeup_lock, flags);
28041 +#ifdef CONFIG_MISSED_TIMER_OFFSETS_HIST
28042 +static notrace void probe_hrtimer_interrupt(void *v, int cpu,
28043 + long long latency_ns, struct task_struct *curr,
28044 + struct task_struct *task)
28046 + if (latency_ns <= 0 && task != NULL && rt_task(task) &&
28047 + (task->prio < curr->prio ||
28048 + (task->prio == curr->prio &&
28049 + !cpumask_test_cpu(cpu, &task->cpus_allowed)))) {
28053 + if (missed_timer_offsets_pid) {
28054 + if (likely(missed_timer_offsets_pid !=
28055 + task_pid_nr(task)))
28059 + now = ftrace_now(cpu);
28060 + latency = (long) div_s64(-latency_ns, NSECS_PER_USECS);
28061 + latency_hist(MISSED_TIMER_OFFSETS, cpu, latency, latency, now,
28063 +#ifdef CONFIG_WAKEUP_LATENCY_HIST
28064 + task->timer_offset = latency;
28070 +static __init int latency_hist_init(void)
28072 + struct dentry *latency_hist_root = NULL;
28073 + struct dentry *dentry;
28074 +#ifdef CONFIG_WAKEUP_LATENCY_HIST
28075 + struct dentry *dentry_sharedprio;
28077 + struct dentry *entry;
28078 + struct dentry *enable_root;
28080 + struct hist_data *my_hist;
28082 + char *cpufmt = "CPU%d";
28083 +#if defined(CONFIG_WAKEUP_LATENCY_HIST) || \
28084 + defined(CONFIG_MISSED_TIMER_OFFSETS_HIST)
28085 + char *cpufmt_maxlatproc = "max_latency-CPU%d";
28086 + struct maxlatproc_data *mp = NULL;
28089 + dentry = tracing_init_dentry();
28090 + latency_hist_root = debugfs_create_dir(latency_hist_dir_root, dentry);
28091 + enable_root = debugfs_create_dir("enable", latency_hist_root);
28093 +#ifdef CONFIG_INTERRUPT_OFF_HIST
28094 + dentry = debugfs_create_dir(irqsoff_hist_dir, latency_hist_root);
28095 + for_each_possible_cpu(i) {
28096 + sprintf(name, cpufmt, i);
28097 + entry = debugfs_create_file(name, 0444, dentry,
28098 + &per_cpu(irqsoff_hist, i), &latency_hist_fops);
28099 + my_hist = &per_cpu(irqsoff_hist, i);
28100 + atomic_set(&my_hist->hist_mode, 1);
28101 + my_hist->min_lat = LONG_MAX;
28103 + entry = debugfs_create_file("reset", 0644, dentry,
28104 + (void *)IRQSOFF_LATENCY, &latency_hist_reset_fops);
28107 +#ifdef CONFIG_PREEMPT_OFF_HIST
28108 + dentry = debugfs_create_dir(preemptoff_hist_dir,
28109 + latency_hist_root);
28110 + for_each_possible_cpu(i) {
28111 + sprintf(name, cpufmt, i);
28112 + entry = debugfs_create_file(name, 0444, dentry,
28113 + &per_cpu(preemptoff_hist, i), &latency_hist_fops);
28114 + my_hist = &per_cpu(preemptoff_hist, i);
28115 + atomic_set(&my_hist->hist_mode, 1);
28116 + my_hist->min_lat = LONG_MAX;
28118 + entry = debugfs_create_file("reset", 0644, dentry,
28119 + (void *)PREEMPTOFF_LATENCY, &latency_hist_reset_fops);
28122 +#if defined(CONFIG_INTERRUPT_OFF_HIST) && defined(CONFIG_PREEMPT_OFF_HIST)
28123 + dentry = debugfs_create_dir(preemptirqsoff_hist_dir,
28124 + latency_hist_root);
28125 + for_each_possible_cpu(i) {
28126 + sprintf(name, cpufmt, i);
28127 + entry = debugfs_create_file(name, 0444, dentry,
28128 + &per_cpu(preemptirqsoff_hist, i), &latency_hist_fops);
28129 + my_hist = &per_cpu(preemptirqsoff_hist, i);
28130 + atomic_set(&my_hist->hist_mode, 1);
28131 + my_hist->min_lat = LONG_MAX;
28133 + entry = debugfs_create_file("reset", 0644, dentry,
28134 + (void *)PREEMPTIRQSOFF_LATENCY, &latency_hist_reset_fops);
28137 +#if defined(CONFIG_INTERRUPT_OFF_HIST) || defined(CONFIG_PREEMPT_OFF_HIST)
28138 + entry = debugfs_create_file("preemptirqsoff", 0644,
28139 + enable_root, (void *)&preemptirqsoff_enabled_data,
28143 +#ifdef CONFIG_WAKEUP_LATENCY_HIST
28144 + dentry = debugfs_create_dir(wakeup_latency_hist_dir,
28145 + latency_hist_root);
28146 + dentry_sharedprio = debugfs_create_dir(
28147 + wakeup_latency_hist_dir_sharedprio, dentry);
28148 + for_each_possible_cpu(i) {
28149 + sprintf(name, cpufmt, i);
28151 + entry = debugfs_create_file(name, 0444, dentry,
28152 + &per_cpu(wakeup_latency_hist, i),
28153 + &latency_hist_fops);
28154 + my_hist = &per_cpu(wakeup_latency_hist, i);
28155 + atomic_set(&my_hist->hist_mode, 1);
28156 + my_hist->min_lat = LONG_MAX;
28158 + entry = debugfs_create_file(name, 0444, dentry_sharedprio,
28159 + &per_cpu(wakeup_latency_hist_sharedprio, i),
28160 + &latency_hist_fops);
28161 + my_hist = &per_cpu(wakeup_latency_hist_sharedprio, i);
28162 + atomic_set(&my_hist->hist_mode, 1);
28163 + my_hist->min_lat = LONG_MAX;
28165 + sprintf(name, cpufmt_maxlatproc, i);
28167 + mp = &per_cpu(wakeup_maxlatproc, i);
28168 + entry = debugfs_create_file(name, 0444, dentry, mp,
28169 + &maxlatproc_fops);
28170 + clear_maxlatprocdata(mp);
28172 + mp = &per_cpu(wakeup_maxlatproc_sharedprio, i);
28173 + entry = debugfs_create_file(name, 0444, dentry_sharedprio, mp,
28174 + &maxlatproc_fops);
28175 + clear_maxlatprocdata(mp);
28177 + entry = debugfs_create_file("pid", 0644, dentry,
28178 + (void *)&wakeup_pid, &pid_fops);
28179 + entry = debugfs_create_file("reset", 0644, dentry,
28180 + (void *)WAKEUP_LATENCY, &latency_hist_reset_fops);
28181 + entry = debugfs_create_file("reset", 0644, dentry_sharedprio,
28182 + (void *)WAKEUP_LATENCY_SHAREDPRIO, &latency_hist_reset_fops);
28183 + entry = debugfs_create_file("wakeup", 0644,
28184 + enable_root, (void *)&wakeup_latency_enabled_data,
28188 +#ifdef CONFIG_MISSED_TIMER_OFFSETS_HIST
28189 + dentry = debugfs_create_dir(missed_timer_offsets_dir,
28190 + latency_hist_root);
28191 + for_each_possible_cpu(i) {
28192 + sprintf(name, cpufmt, i);
28193 + entry = debugfs_create_file(name, 0444, dentry,
28194 + &per_cpu(missed_timer_offsets, i), &latency_hist_fops);
28195 + my_hist = &per_cpu(missed_timer_offsets, i);
28196 + atomic_set(&my_hist->hist_mode, 1);
28197 + my_hist->min_lat = LONG_MAX;
28199 + sprintf(name, cpufmt_maxlatproc, i);
28200 + mp = &per_cpu(missed_timer_offsets_maxlatproc, i);
28201 + entry = debugfs_create_file(name, 0444, dentry, mp,
28202 + &maxlatproc_fops);
28203 + clear_maxlatprocdata(mp);
28205 + entry = debugfs_create_file("pid", 0644, dentry,
28206 + (void *)&missed_timer_offsets_pid, &pid_fops);
28207 + entry = debugfs_create_file("reset", 0644, dentry,
28208 + (void *)MISSED_TIMER_OFFSETS, &latency_hist_reset_fops);
28209 + entry = debugfs_create_file("missed_timer_offsets", 0644,
28210 + enable_root, (void *)&missed_timer_offsets_enabled_data,
28214 +#if defined(CONFIG_WAKEUP_LATENCY_HIST) && \
28215 + defined(CONFIG_MISSED_TIMER_OFFSETS_HIST)
28216 + dentry = debugfs_create_dir(timerandwakeup_latency_hist_dir,
28217 + latency_hist_root);
28218 + for_each_possible_cpu(i) {
28219 + sprintf(name, cpufmt, i);
28220 + entry = debugfs_create_file(name, 0444, dentry,
28221 + &per_cpu(timerandwakeup_latency_hist, i),
28222 + &latency_hist_fops);
28223 + my_hist = &per_cpu(timerandwakeup_latency_hist, i);
28224 + atomic_set(&my_hist->hist_mode, 1);
28225 + my_hist->min_lat = LONG_MAX;
28227 + sprintf(name, cpufmt_maxlatproc, i);
28228 + mp = &per_cpu(timerandwakeup_maxlatproc, i);
28229 + entry = debugfs_create_file(name, 0444, dentry, mp,
28230 + &maxlatproc_fops);
28231 + clear_maxlatprocdata(mp);
28233 + entry = debugfs_create_file("reset", 0644, dentry,
28234 + (void *)TIMERANDWAKEUP_LATENCY, &latency_hist_reset_fops);
28235 + entry = debugfs_create_file("timerandwakeup", 0644,
28236 + enable_root, (void *)&timerandwakeup_enabled_data,
28242 +device_initcall(latency_hist_init);
28243 diff --git a/kernel/trace/trace.c b/kernel/trace/trace.c
28244 index 4c21c0b7dc91..d0e52efed27a 100644
28245 --- a/kernel/trace/trace.c
28246 +++ b/kernel/trace/trace.c
28247 @@ -1652,6 +1652,7 @@ tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
28248 struct task_struct *tsk = current;
28250 entry->preempt_count = pc & 0xff;
28251 + entry->preempt_lazy_count = preempt_lazy_count();
28252 entry->pid = (tsk) ? tsk->pid : 0;
28254 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
28255 @@ -1661,8 +1662,11 @@ tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
28257 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
28258 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
28259 - (tif_need_resched() ? TRACE_FLAG_NEED_RESCHED : 0) |
28260 + (tif_need_resched_now() ? TRACE_FLAG_NEED_RESCHED : 0) |
28261 + (need_resched_lazy() ? TRACE_FLAG_NEED_RESCHED_LAZY : 0) |
28262 (test_preempt_need_resched() ? TRACE_FLAG_PREEMPT_RESCHED : 0);
28264 + entry->migrate_disable = (tsk) ? __migrate_disabled(tsk) & 0xFF : 0;
28266 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
28268 @@ -2555,14 +2559,17 @@ get_total_entries(struct trace_buffer *buf,
28270 static void print_lat_help_header(struct seq_file *m)
28272 - seq_puts(m, "# _------=> CPU# \n"
28273 - "# / _-----=> irqs-off \n"
28274 - "# | / _----=> need-resched \n"
28275 - "# || / _---=> hardirq/softirq \n"
28276 - "# ||| / _--=> preempt-depth \n"
28277 - "# |||| / delay \n"
28278 - "# cmd pid ||||| time | caller \n"
28279 - "# \\ / ||||| \\ | / \n");
28280 + seq_puts(m, "# _--------=> CPU# \n"
28281 + "# / _-------=> irqs-off \n"
28282 + "# | / _------=> need-resched \n"
28283 + "# || / _-----=> need-resched_lazy \n"
28284 + "# ||| / _----=> hardirq/softirq \n"
28285 + "# |||| / _---=> preempt-depth \n"
28286 + "# ||||| / _--=> preempt-lazy-depth\n"
28287 + "# |||||| / _-=> migrate-disable \n"
28288 + "# ||||||| / delay \n"
28289 + "# cmd pid |||||||| time | caller \n"
28290 + "# \\ / |||||||| \\ | / \n");
28293 static void print_event_info(struct trace_buffer *buf, struct seq_file *m)
28294 @@ -2588,11 +2595,14 @@ static void print_func_help_header_irq(struct trace_buffer *buf, struct seq_file
28295 print_event_info(buf, m);
28296 seq_puts(m, "# _-----=> irqs-off\n"
28297 "# / _----=> need-resched\n"
28298 - "# | / _---=> hardirq/softirq\n"
28299 - "# || / _--=> preempt-depth\n"
28300 - "# ||| / delay\n"
28301 - "# TASK-PID CPU# |||| TIMESTAMP FUNCTION\n"
28302 - "# | | | |||| | |\n");
28303 + "# |/ _-----=> need-resched_lazy\n"
28304 + "# || / _---=> hardirq/softirq\n"
28305 + "# ||| / _--=> preempt-depth\n"
28306 + "# |||| / _-=> preempt-lazy-depth\n"
28307 + "# ||||| / _-=> migrate-disable \n"
28308 + "# |||||| / delay\n"
28309 + "# TASK-PID CPU# ||||||| TIMESTAMP FUNCTION\n"
28310 + "# | | | ||||||| | |\n");
28314 diff --git a/kernel/trace/trace.h b/kernel/trace/trace.h
28315 index 919d9d07686f..3bf86ece683c 100644
28316 --- a/kernel/trace/trace.h
28317 +++ b/kernel/trace/trace.h
28318 @@ -117,6 +117,7 @@ struct kretprobe_trace_entry_head {
28319 * NEED_RESCHED - reschedule is requested
28320 * HARDIRQ - inside an interrupt handler
28321 * SOFTIRQ - inside a softirq handler
28322 + * NEED_RESCHED_LAZY - lazy reschedule is requested
28324 enum trace_flag_type {
28325 TRACE_FLAG_IRQS_OFF = 0x01,
28326 @@ -125,6 +126,7 @@ enum trace_flag_type {
28327 TRACE_FLAG_HARDIRQ = 0x08,
28328 TRACE_FLAG_SOFTIRQ = 0x10,
28329 TRACE_FLAG_PREEMPT_RESCHED = 0x20,
28330 + TRACE_FLAG_NEED_RESCHED_LAZY = 0x40,
28333 #define TRACE_BUF_SIZE 1024
28334 diff --git a/kernel/trace/trace_events.c b/kernel/trace/trace_events.c
28335 index 996f0fd34312..5bd79b347398 100644
28336 --- a/kernel/trace/trace_events.c
28337 +++ b/kernel/trace/trace_events.c
28338 @@ -188,6 +188,8 @@ static int trace_define_common_fields(void)
28339 __common_field(unsigned char, flags);
28340 __common_field(unsigned char, preempt_count);
28341 __common_field(int, pid);
28342 + __common_field(unsigned short, migrate_disable);
28343 + __common_field(unsigned short, padding);
28347 @@ -244,6 +246,14 @@ void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
28349 local_save_flags(fbuffer->flags);
28350 fbuffer->pc = preempt_count();
28352 + * If CONFIG_PREEMPT is enabled, then the tracepoint itself disables
28353 + * preemption (adding one to the preempt_count). Since we are
28354 + * interested in the preempt_count at the time the tracepoint was
28355 + * hit, we need to subtract one to offset the increment.
28357 + if (IS_ENABLED(CONFIG_PREEMPT))
28359 fbuffer->trace_file = trace_file;
28362 diff --git a/kernel/trace/trace_irqsoff.c b/kernel/trace/trace_irqsoff.c
28363 index be3222b7d72e..553e71254ad6 100644
28364 --- a/kernel/trace/trace_irqsoff.c
28365 +++ b/kernel/trace/trace_irqsoff.c
28367 #include <linux/uaccess.h>
28368 #include <linux/module.h>
28369 #include <linux/ftrace.h>
28370 +#include <trace/events/hist.h>
28374 @@ -424,11 +425,13 @@ void start_critical_timings(void)
28376 if (preempt_trace() || irq_trace())
28377 start_critical_timing(CALLER_ADDR0, CALLER_ADDR1);
28378 + trace_preemptirqsoff_hist_rcuidle(TRACE_START, 1);
28380 EXPORT_SYMBOL_GPL(start_critical_timings);
28382 void stop_critical_timings(void)
28384 + trace_preemptirqsoff_hist_rcuidle(TRACE_STOP, 0);
28385 if (preempt_trace() || irq_trace())
28386 stop_critical_timing(CALLER_ADDR0, CALLER_ADDR1);
28388 @@ -438,6 +441,7 @@ EXPORT_SYMBOL_GPL(stop_critical_timings);
28389 #ifdef CONFIG_PROVE_LOCKING
28390 void time_hardirqs_on(unsigned long a0, unsigned long a1)
28392 + trace_preemptirqsoff_hist_rcuidle(IRQS_ON, 0);
28393 if (!preempt_trace() && irq_trace())
28394 stop_critical_timing(a0, a1);
28396 @@ -446,6 +450,7 @@ void time_hardirqs_off(unsigned long a0, unsigned long a1)
28398 if (!preempt_trace() && irq_trace())
28399 start_critical_timing(a0, a1);
28400 + trace_preemptirqsoff_hist_rcuidle(IRQS_OFF, 1);
28403 #else /* !CONFIG_PROVE_LOCKING */
28404 @@ -471,6 +476,7 @@ inline void print_irqtrace_events(struct task_struct *curr)
28406 void trace_hardirqs_on(void)
28408 + trace_preemptirqsoff_hist(IRQS_ON, 0);
28409 if (!preempt_trace() && irq_trace())
28410 stop_critical_timing(CALLER_ADDR0, CALLER_ADDR1);
28412 @@ -480,11 +486,13 @@ void trace_hardirqs_off(void)
28414 if (!preempt_trace() && irq_trace())
28415 start_critical_timing(CALLER_ADDR0, CALLER_ADDR1);
28416 + trace_preemptirqsoff_hist(IRQS_OFF, 1);
28418 EXPORT_SYMBOL(trace_hardirqs_off);
28420 __visible void trace_hardirqs_on_caller(unsigned long caller_addr)
28422 + trace_preemptirqsoff_hist(IRQS_ON, 0);
28423 if (!preempt_trace() && irq_trace())
28424 stop_critical_timing(CALLER_ADDR0, caller_addr);
28426 @@ -494,6 +502,7 @@ __visible void trace_hardirqs_off_caller(unsigned long caller_addr)
28428 if (!preempt_trace() && irq_trace())
28429 start_critical_timing(CALLER_ADDR0, caller_addr);
28430 + trace_preemptirqsoff_hist(IRQS_OFF, 1);
28432 EXPORT_SYMBOL(trace_hardirqs_off_caller);
28434 @@ -503,12 +512,14 @@ EXPORT_SYMBOL(trace_hardirqs_off_caller);
28435 #ifdef CONFIG_PREEMPT_TRACER
28436 void trace_preempt_on(unsigned long a0, unsigned long a1)
28438 + trace_preemptirqsoff_hist(PREEMPT_ON, 0);
28439 if (preempt_trace() && !irq_trace())
28440 stop_critical_timing(a0, a1);
28443 void trace_preempt_off(unsigned long a0, unsigned long a1)
28445 + trace_preemptirqsoff_hist(PREEMPT_ON, 1);
28446 if (preempt_trace() && !irq_trace())
28447 start_critical_timing(a0, a1);
28449 diff --git a/kernel/trace/trace_output.c b/kernel/trace/trace_output.c
28450 index 282982195e09..9f19d839a756 100644
28451 --- a/kernel/trace/trace_output.c
28452 +++ b/kernel/trace/trace_output.c
28453 @@ -386,6 +386,7 @@ int trace_print_lat_fmt(struct trace_seq *s, struct trace_entry *entry)
28457 + char need_resched_lazy;
28461 @@ -413,6 +414,8 @@ int trace_print_lat_fmt(struct trace_seq *s, struct trace_entry *entry)
28462 need_resched = '.';
28465 + need_resched_lazy =
28466 + (entry->flags & TRACE_FLAG_NEED_RESCHED_LAZY) ? 'L' : '.';
28469 (hardirq && softirq) ? 'H' :
28470 @@ -420,14 +423,25 @@ int trace_print_lat_fmt(struct trace_seq *s, struct trace_entry *entry)
28474 - trace_seq_printf(s, "%c%c%c",
28475 - irqs_off, need_resched, hardsoft_irq);
28476 + trace_seq_printf(s, "%c%c%c%c",
28477 + irqs_off, need_resched, need_resched_lazy,
28480 if (entry->preempt_count)
28481 trace_seq_printf(s, "%x", entry->preempt_count);
28483 trace_seq_putc(s, '.');
28485 + if (entry->preempt_lazy_count)
28486 + trace_seq_printf(s, "%x", entry->preempt_lazy_count);
28488 + trace_seq_putc(s, '.');
28490 + if (entry->migrate_disable)
28491 + trace_seq_printf(s, "%x", entry->migrate_disable);
28493 + trace_seq_putc(s, '.');
28495 return !trace_seq_has_overflowed(s);
28498 diff --git a/kernel/user.c b/kernel/user.c
28499 index b069ccbfb0b0..1a2e88e98b5e 100644
28500 --- a/kernel/user.c
28501 +++ b/kernel/user.c
28502 @@ -161,11 +161,11 @@ void free_uid(struct user_struct *up)
28506 - local_irq_save(flags);
28507 + local_irq_save_nort(flags);
28508 if (atomic_dec_and_lock(&up->__count, &uidhash_lock))
28509 free_user(up, flags);
28511 - local_irq_restore(flags);
28512 + local_irq_restore_nort(flags);
28515 struct user_struct *alloc_uid(kuid_t uid)
28516 diff --git a/kernel/watchdog.c b/kernel/watchdog.c
28517 index c1e0b5f429b6..fa2e079cc314 100644
28518 --- a/kernel/watchdog.c
28519 +++ b/kernel/watchdog.c
28520 @@ -299,6 +299,8 @@ static int is_softlockup(unsigned long touch_ts)
28522 #ifdef CONFIG_HARDLOCKUP_DETECTOR
28524 +static DEFINE_RAW_SPINLOCK(watchdog_output_lock);
28526 static struct perf_event_attr wd_hw_attr = {
28527 .type = PERF_TYPE_HARDWARE,
28528 .config = PERF_COUNT_HW_CPU_CYCLES,
28529 @@ -332,6 +334,13 @@ static void watchdog_overflow_callback(struct perf_event *event,
28530 /* only print hardlockups once */
28531 if (__this_cpu_read(hard_watchdog_warn) == true)
28534 + * If early-printk is enabled then make sure we do not
28535 + * lock up in printk() and kill console logging:
28539 + raw_spin_lock(&watchdog_output_lock);
28541 pr_emerg("Watchdog detected hard LOCKUP on cpu %d", this_cpu);
28543 @@ -349,8 +358,9 @@ static void watchdog_overflow_callback(struct perf_event *event,
28544 !test_and_set_bit(0, &hardlockup_allcpu_dumped))
28545 trigger_allbutself_cpu_backtrace();
28547 + raw_spin_unlock(&watchdog_output_lock);
28548 if (hardlockup_panic)
28549 - panic("Hard LOCKUP");
28550 + nmi_panic(regs, "Hard LOCKUP");
28552 __this_cpu_write(hard_watchdog_warn, true);
28554 @@ -496,6 +506,7 @@ static void watchdog_enable(unsigned int cpu)
28555 /* kick off the timer for the hardlockup detector */
28556 hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
28557 hrtimer->function = watchdog_timer_fn;
28558 + hrtimer->irqsafe = 1;
28560 /* Enable the perf event */
28561 watchdog_nmi_enable(cpu);
28562 diff --git a/kernel/workqueue.c b/kernel/workqueue.c
28563 index 2c2f971f3e75..d5b0f4fc0eb0 100644
28564 --- a/kernel/workqueue.c
28565 +++ b/kernel/workqueue.c
28567 #include <linux/nodemask.h>
28568 #include <linux/moduleparam.h>
28569 #include <linux/uaccess.h>
28570 +#include <linux/locallock.h>
28571 +#include <linux/delay.h>
28573 #include "workqueue_internal.h"
28575 @@ -121,11 +123,16 @@ enum {
28576 * cpu or grabbing pool->lock is enough for read access. If
28577 * POOL_DISASSOCIATED is set, it's identical to L.
28579 + * On RT we need the extra protection via rt_lock_idle_list() for
28580 + * the list manipulations against read access from
28581 + * wq_worker_sleeping(). All other places are nicely serialized via
28584 * A: pool->attach_mutex protected.
28586 * PL: wq_pool_mutex protected.
28588 - * PR: wq_pool_mutex protected for writes. Sched-RCU protected for reads.
28589 + * PR: wq_pool_mutex protected for writes. RCU protected for reads.
28591 * PW: wq_pool_mutex and wq->mutex protected for writes. Either for reads.
28593 @@ -134,7 +141,7 @@ enum {
28595 * WQ: wq->mutex protected.
28597 - * WR: wq->mutex protected for writes. Sched-RCU protected for reads.
28598 + * WR: wq->mutex protected for writes. RCU protected for reads.
28600 * MD: wq_mayday_lock protected.
28602 @@ -183,7 +190,7 @@ struct worker_pool {
28603 atomic_t nr_running ____cacheline_aligned_in_smp;
28606 - * Destruction of pool is sched-RCU protected to allow dereferences
28607 + * Destruction of pool is RCU protected to allow dereferences
28608 * from get_work_pool().
28610 struct rcu_head rcu;
28611 @@ -212,7 +219,7 @@ struct pool_workqueue {
28613 * Release of unbound pwq is punted to system_wq. See put_pwq()
28614 * and pwq_unbound_release_workfn() for details. pool_workqueue
28615 - * itself is also sched-RCU protected so that the first pwq can be
28616 + * itself is also RCU protected so that the first pwq can be
28617 * determined without grabbing wq->mutex.
28619 struct work_struct unbound_release_work;
28620 @@ -331,6 +338,8 @@ EXPORT_SYMBOL_GPL(system_power_efficient_wq);
28621 struct workqueue_struct *system_freezable_power_efficient_wq __read_mostly;
28622 EXPORT_SYMBOL_GPL(system_freezable_power_efficient_wq);
28624 +static DEFINE_LOCAL_IRQ_LOCK(pendingb_lock);
28626 static int worker_thread(void *__worker);
28627 static void workqueue_sysfs_unregister(struct workqueue_struct *wq);
28629 @@ -338,20 +347,20 @@ static void workqueue_sysfs_unregister(struct workqueue_struct *wq);
28630 #include <trace/events/workqueue.h>
28632 #define assert_rcu_or_pool_mutex() \
28633 - RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held() && \
28634 + RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \
28635 !lockdep_is_held(&wq_pool_mutex), \
28636 - "sched RCU or wq_pool_mutex should be held")
28637 + "RCU or wq_pool_mutex should be held")
28639 #define assert_rcu_or_wq_mutex(wq) \
28640 - RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held() && \
28641 + RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \
28642 !lockdep_is_held(&wq->mutex), \
28643 - "sched RCU or wq->mutex should be held")
28644 + "RCU or wq->mutex should be held")
28646 #define assert_rcu_or_wq_mutex_or_pool_mutex(wq) \
28647 - RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held() && \
28648 + RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \
28649 !lockdep_is_held(&wq->mutex) && \
28650 !lockdep_is_held(&wq_pool_mutex), \
28651 - "sched RCU, wq->mutex or wq_pool_mutex should be held")
28652 + "RCU, wq->mutex or wq_pool_mutex should be held")
28654 #define for_each_cpu_worker_pool(pool, cpu) \
28655 for ((pool) = &per_cpu(cpu_worker_pools, cpu)[0]; \
28656 @@ -363,7 +372,7 @@ static void workqueue_sysfs_unregister(struct workqueue_struct *wq);
28657 * @pool: iteration cursor
28658 * @pi: integer used for iteration
28660 - * This must be called either with wq_pool_mutex held or sched RCU read
28661 + * This must be called either with wq_pool_mutex held or RCU read
28662 * locked. If the pool needs to be used beyond the locking in effect, the
28663 * caller is responsible for guaranteeing that the pool stays online.
28665 @@ -395,7 +404,7 @@ static void workqueue_sysfs_unregister(struct workqueue_struct *wq);
28666 * @pwq: iteration cursor
28667 * @wq: the target workqueue
28669 - * This must be called either with wq->mutex held or sched RCU read locked.
28670 + * This must be called either with wq->mutex held or RCU read locked.
28671 * If the pwq needs to be used beyond the locking in effect, the caller is
28672 * responsible for guaranteeing that the pwq stays online.
28674 @@ -407,6 +416,31 @@ static void workqueue_sysfs_unregister(struct workqueue_struct *wq);
28675 if (({ assert_rcu_or_wq_mutex(wq); false; })) { } \
28678 +#ifdef CONFIG_PREEMPT_RT_BASE
28679 +static inline void rt_lock_idle_list(struct worker_pool *pool)
28681 + preempt_disable();
28683 +static inline void rt_unlock_idle_list(struct worker_pool *pool)
28685 + preempt_enable();
28687 +static inline void sched_lock_idle_list(struct worker_pool *pool) { }
28688 +static inline void sched_unlock_idle_list(struct worker_pool *pool) { }
28690 +static inline void rt_lock_idle_list(struct worker_pool *pool) { }
28691 +static inline void rt_unlock_idle_list(struct worker_pool *pool) { }
28692 +static inline void sched_lock_idle_list(struct worker_pool *pool)
28694 + spin_lock_irq(&pool->lock);
28696 +static inline void sched_unlock_idle_list(struct worker_pool *pool)
28698 + spin_unlock_irq(&pool->lock);
28703 #ifdef CONFIG_DEBUG_OBJECTS_WORK
28705 static struct debug_obj_descr work_debug_descr;
28706 @@ -557,7 +591,7 @@ static int worker_pool_assign_id(struct worker_pool *pool)
28707 * @wq: the target workqueue
28708 * @node: the node ID
28710 - * This must be called with any of wq_pool_mutex, wq->mutex or sched RCU
28711 + * This must be called with any of wq_pool_mutex, wq->mutex or RCU
28713 * If the pwq needs to be used beyond the locking in effect, the caller is
28714 * responsible for guaranteeing that the pwq stays online.
28715 @@ -701,8 +735,8 @@ static struct pool_workqueue *get_work_pwq(struct work_struct *work)
28716 * @work: the work item of interest
28718 * Pools are created and destroyed under wq_pool_mutex, and allows read
28719 - * access under sched-RCU read lock. As such, this function should be
28720 - * called under wq_pool_mutex or with preemption disabled.
28721 + * access under RCU read lock. As such, this function should be
28722 + * called under wq_pool_mutex or inside of a rcu_read_lock() region.
28724 * All fields of the returned pool are accessible as long as the above
28725 * mentioned locking is in effect. If the returned pool needs to be used
28726 @@ -839,51 +873,44 @@ static struct worker *first_idle_worker(struct worker_pool *pool)
28728 static void wake_up_worker(struct worker_pool *pool)
28730 - struct worker *worker = first_idle_worker(pool);
28731 + struct worker *worker;
28733 + rt_lock_idle_list(pool);
28735 + worker = first_idle_worker(pool);
28737 if (likely(worker))
28738 wake_up_process(worker->task);
28740 + rt_unlock_idle_list(pool);
28744 - * wq_worker_waking_up - a worker is waking up
28745 - * @task: task waking up
28746 - * @cpu: CPU @task is waking up to
28747 + * wq_worker_running - a worker is running again
28748 + * @task: task returning from sleep
28750 - * This function is called during try_to_wake_up() when a worker is
28754 - * spin_lock_irq(rq->lock)
28755 + * This function is called when a worker returns from schedule()
28757 -void wq_worker_waking_up(struct task_struct *task, int cpu)
28758 +void wq_worker_running(struct task_struct *task)
28760 struct worker *worker = kthread_data(task);
28762 - if (!(worker->flags & WORKER_NOT_RUNNING)) {
28763 - WARN_ON_ONCE(worker->pool->cpu != cpu);
28764 + if (!worker->sleeping)
28766 + if (!(worker->flags & WORKER_NOT_RUNNING))
28767 atomic_inc(&worker->pool->nr_running);
28769 + worker->sleeping = 0;
28773 * wq_worker_sleeping - a worker is going to sleep
28774 * @task: task going to sleep
28775 - * @cpu: CPU in question, must be the current CPU number
28777 - * This function is called during schedule() when a busy worker is
28778 - * going to sleep. Worker on the same cpu can be woken up by
28779 - * returning pointer to its task.
28782 - * spin_lock_irq(rq->lock)
28785 - * Worker task on @cpu to wake up, %NULL if none.
28786 + * This function is called from schedule() when a busy worker is
28787 + * going to sleep.
28789 -struct task_struct *wq_worker_sleeping(struct task_struct *task, int cpu)
28790 +void wq_worker_sleeping(struct task_struct *task)
28792 - struct worker *worker = kthread_data(task), *to_wakeup = NULL;
28793 + struct worker *worker = kthread_data(task);
28794 struct worker_pool *pool;
28797 @@ -892,29 +919,26 @@ struct task_struct *wq_worker_sleeping(struct task_struct *task, int cpu)
28798 * checking NOT_RUNNING.
28800 if (worker->flags & WORKER_NOT_RUNNING)
28804 pool = worker->pool;
28806 - /* this can only happen on the local cpu */
28807 - if (WARN_ON_ONCE(cpu != raw_smp_processor_id() || pool->cpu != cpu))
28809 + if (WARN_ON_ONCE(worker->sleeping))
28812 + worker->sleeping = 1;
28815 * The counterpart of the following dec_and_test, implied mb,
28816 * worklist not empty test sequence is in insert_work().
28817 * Please read comment there.
28819 - * NOT_RUNNING is clear. This means that we're bound to and
28820 - * running on the local cpu w/ rq lock held and preemption
28821 - * disabled, which in turn means that none else could be
28822 - * manipulating idle_list, so dereferencing idle_list without pool
28825 if (atomic_dec_and_test(&pool->nr_running) &&
28826 - !list_empty(&pool->worklist))
28827 - to_wakeup = first_idle_worker(pool);
28828 - return to_wakeup ? to_wakeup->task : NULL;
28829 + !list_empty(&pool->worklist)) {
28830 + sched_lock_idle_list(pool);
28831 + wake_up_worker(pool);
28832 + sched_unlock_idle_list(pool);
28837 @@ -1108,12 +1132,14 @@ static void put_pwq_unlocked(struct pool_workqueue *pwq)
28841 - * As both pwqs and pools are sched-RCU protected, the
28842 + * As both pwqs and pools are RCU protected, the
28843 * following lock operations are safe.
28845 - spin_lock_irq(&pwq->pool->lock);
28847 + local_spin_lock_irq(pendingb_lock, &pwq->pool->lock);
28849 - spin_unlock_irq(&pwq->pool->lock);
28850 + local_spin_unlock_irq(pendingb_lock, &pwq->pool->lock);
28851 + rcu_read_unlock();
28855 @@ -1215,7 +1241,7 @@ static int try_to_grab_pending(struct work_struct *work, bool is_dwork,
28856 struct worker_pool *pool;
28857 struct pool_workqueue *pwq;
28859 - local_irq_save(*flags);
28860 + local_lock_irqsave(pendingb_lock, *flags);
28862 /* try to steal the timer if it exists */
28864 @@ -1234,6 +1260,7 @@ static int try_to_grab_pending(struct work_struct *work, bool is_dwork,
28865 if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work)))
28870 * The queueing is in progress, or it is already queued. Try to
28871 * steal it from ->worklist without clearing WORK_STRUCT_PENDING.
28872 @@ -1272,14 +1299,16 @@ static int try_to_grab_pending(struct work_struct *work, bool is_dwork,
28873 set_work_pool_and_keep_pending(work, pool->id);
28875 spin_unlock(&pool->lock);
28876 + rcu_read_unlock();
28879 spin_unlock(&pool->lock);
28881 - local_irq_restore(*flags);
28882 + rcu_read_unlock();
28883 + local_unlock_irqrestore(pendingb_lock, *flags);
28884 if (work_is_canceling(work))
28891 @@ -1348,7 +1377,7 @@ static void __queue_work(int cpu, struct workqueue_struct *wq,
28892 * queued or lose PENDING. Grabbing PENDING and queueing should
28893 * happen with IRQ disabled.
28895 - WARN_ON_ONCE(!irqs_disabled());
28896 + WARN_ON_ONCE_NONRT(!irqs_disabled());
28898 debug_work_activate(work);
28900 @@ -1356,6 +1385,8 @@ static void __queue_work(int cpu, struct workqueue_struct *wq,
28901 if (unlikely(wq->flags & __WQ_DRAINING) &&
28902 WARN_ON_ONCE(!is_chained_work(wq)))
28907 if (req_cpu == WORK_CPU_UNBOUND)
28908 cpu = raw_smp_processor_id();
28909 @@ -1412,10 +1443,8 @@ retry:
28910 /* pwq determined, queue */
28911 trace_workqueue_queue_work(req_cpu, pwq, work);
28913 - if (WARN_ON(!list_empty(&work->entry))) {
28914 - spin_unlock(&pwq->pool->lock);
28917 + if (WARN_ON(!list_empty(&work->entry)))
28920 pwq->nr_in_flight[pwq->work_color]++;
28921 work_flags = work_color_to_flags(pwq->work_color);
28922 @@ -1431,7 +1460,9 @@ retry:
28924 insert_work(pwq, work, worklist, work_flags);
28927 spin_unlock(&pwq->pool->lock);
28928 + rcu_read_unlock();
28932 @@ -1451,14 +1482,14 @@ bool queue_work_on(int cpu, struct workqueue_struct *wq,
28934 unsigned long flags;
28936 - local_irq_save(flags);
28937 + local_lock_irqsave(pendingb_lock,flags);
28939 if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) {
28940 __queue_work(cpu, wq, work);
28944 - local_irq_restore(flags);
28945 + local_unlock_irqrestore(pendingb_lock, flags);
28948 EXPORT_SYMBOL(queue_work_on);
28949 @@ -1525,14 +1556,14 @@ bool queue_delayed_work_on(int cpu, struct workqueue_struct *wq,
28950 unsigned long flags;
28952 /* read the comment in __queue_work() */
28953 - local_irq_save(flags);
28954 + local_lock_irqsave(pendingb_lock, flags);
28956 if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) {
28957 __queue_delayed_work(cpu, wq, dwork, delay);
28961 - local_irq_restore(flags);
28962 + local_unlock_irqrestore(pendingb_lock, flags);
28965 EXPORT_SYMBOL(queue_delayed_work_on);
28966 @@ -1567,7 +1598,7 @@ bool mod_delayed_work_on(int cpu, struct workqueue_struct *wq,
28968 if (likely(ret >= 0)) {
28969 __queue_delayed_work(cpu, wq, dwork, delay);
28970 - local_irq_restore(flags);
28971 + local_unlock_irqrestore(pendingb_lock, flags);
28974 /* -ENOENT from try_to_grab_pending() becomes %true */
28975 @@ -1600,7 +1631,9 @@ static void worker_enter_idle(struct worker *worker)
28976 worker->last_active = jiffies;
28978 /* idle_list is LIFO */
28979 + rt_lock_idle_list(pool);
28980 list_add(&worker->entry, &pool->idle_list);
28981 + rt_unlock_idle_list(pool);
28983 if (too_many_workers(pool) && !timer_pending(&pool->idle_timer))
28984 mod_timer(&pool->idle_timer, jiffies + IDLE_WORKER_TIMEOUT);
28985 @@ -1633,7 +1666,9 @@ static void worker_leave_idle(struct worker *worker)
28987 worker_clr_flags(worker, WORKER_IDLE);
28989 + rt_lock_idle_list(pool);
28990 list_del_init(&worker->entry);
28991 + rt_unlock_idle_list(pool);
28994 static struct worker *alloc_worker(int node)
28995 @@ -1799,7 +1834,9 @@ static void destroy_worker(struct worker *worker)
28996 pool->nr_workers--;
28999 + rt_lock_idle_list(pool);
29000 list_del_init(&worker->entry);
29001 + rt_unlock_idle_list(pool);
29002 worker->flags |= WORKER_DIE;
29003 wake_up_process(worker->task);
29005 @@ -2716,14 +2753,14 @@ static bool start_flush_work(struct work_struct *work, struct wq_barrier *barr)
29009 - local_irq_disable();
29011 pool = get_work_pool(work);
29013 - local_irq_enable();
29014 + rcu_read_unlock();
29018 - spin_lock(&pool->lock);
29019 + spin_lock_irq(&pool->lock);
29020 /* see the comment in try_to_grab_pending() with the same code */
29021 pwq = get_work_pwq(work);
29023 @@ -2750,10 +2787,11 @@ static bool start_flush_work(struct work_struct *work, struct wq_barrier *barr)
29025 lock_map_acquire_read(&pwq->wq->lockdep_map);
29026 lock_map_release(&pwq->wq->lockdep_map);
29028 + rcu_read_unlock();
29031 spin_unlock_irq(&pool->lock);
29032 + rcu_read_unlock();
29036 @@ -2840,7 +2878,7 @@ static bool __cancel_work_timer(struct work_struct *work, bool is_dwork)
29038 /* tell other tasks trying to grab @work to back off */
29039 mark_work_canceling(work);
29040 - local_irq_restore(flags);
29041 + local_unlock_irqrestore(pendingb_lock, flags);
29044 clear_work_data(work);
29045 @@ -2895,10 +2933,10 @@ EXPORT_SYMBOL_GPL(cancel_work_sync);
29047 bool flush_delayed_work(struct delayed_work *dwork)
29049 - local_irq_disable();
29050 + local_lock_irq(pendingb_lock);
29051 if (del_timer_sync(&dwork->timer))
29052 __queue_work(dwork->cpu, dwork->wq, &dwork->work);
29053 - local_irq_enable();
29054 + local_unlock_irq(pendingb_lock);
29055 return flush_work(&dwork->work);
29057 EXPORT_SYMBOL(flush_delayed_work);
29058 @@ -2933,7 +2971,7 @@ bool cancel_delayed_work(struct delayed_work *dwork)
29060 set_work_pool_and_clear_pending(&dwork->work,
29061 get_work_pool_id(&dwork->work));
29062 - local_irq_restore(flags);
29063 + local_unlock_irqrestore(pendingb_lock, flags);
29066 EXPORT_SYMBOL(cancel_delayed_work);
29067 @@ -3161,7 +3199,7 @@ static void rcu_free_pool(struct rcu_head *rcu)
29068 * put_unbound_pool - put a worker_pool
29069 * @pool: worker_pool to put
29071 - * Put @pool. If its refcnt reaches zero, it gets destroyed in sched-RCU
29072 + * Put @pool. If its refcnt reaches zero, it gets destroyed in RCU
29073 * safe manner. get_unbound_pool() calls this function on its failure path
29074 * and this function should be able to release pools which went through,
29075 * successfully or not, init_worker_pool().
29076 @@ -3215,8 +3253,8 @@ static void put_unbound_pool(struct worker_pool *pool)
29077 del_timer_sync(&pool->idle_timer);
29078 del_timer_sync(&pool->mayday_timer);
29080 - /* sched-RCU protected to allow dereferences from get_work_pool() */
29081 - call_rcu_sched(&pool->rcu, rcu_free_pool);
29082 + /* RCU protected to allow dereferences from get_work_pool() */
29083 + call_rcu(&pool->rcu, rcu_free_pool);
29087 @@ -3323,14 +3361,14 @@ static void pwq_unbound_release_workfn(struct work_struct *work)
29088 put_unbound_pool(pool);
29089 mutex_unlock(&wq_pool_mutex);
29091 - call_rcu_sched(&pwq->rcu, rcu_free_pwq);
29092 + call_rcu(&pwq->rcu, rcu_free_pwq);
29095 * If we're the last pwq going away, @wq is already dead and no one
29096 * is gonna access it anymore. Schedule RCU free.
29099 - call_rcu_sched(&wq->rcu, rcu_free_wq);
29100 + call_rcu(&wq->rcu, rcu_free_wq);
29104 @@ -3983,7 +4021,7 @@ void destroy_workqueue(struct workqueue_struct *wq)
29105 * The base ref is never dropped on per-cpu pwqs. Directly
29106 * schedule RCU free.
29108 - call_rcu_sched(&wq->rcu, rcu_free_wq);
29109 + call_rcu(&wq->rcu, rcu_free_wq);
29112 * We're the sole accessor of @wq at this point. Directly
29113 @@ -4076,7 +4114,8 @@ bool workqueue_congested(int cpu, struct workqueue_struct *wq)
29114 struct pool_workqueue *pwq;
29117 - rcu_read_lock_sched();
29119 + preempt_disable();
29121 if (cpu == WORK_CPU_UNBOUND)
29122 cpu = smp_processor_id();
29123 @@ -4087,7 +4126,8 @@ bool workqueue_congested(int cpu, struct workqueue_struct *wq)
29124 pwq = unbound_pwq_by_node(wq, cpu_to_node(cpu));
29126 ret = !list_empty(&pwq->delayed_works);
29127 - rcu_read_unlock_sched();
29128 + preempt_enable();
29129 + rcu_read_unlock();
29133 @@ -4113,15 +4153,15 @@ unsigned int work_busy(struct work_struct *work)
29134 if (work_pending(work))
29135 ret |= WORK_BUSY_PENDING;
29137 - local_irq_save(flags);
29139 pool = get_work_pool(work);
29141 - spin_lock(&pool->lock);
29142 + spin_lock_irqsave(&pool->lock, flags);
29143 if (find_worker_executing_work(pool, work))
29144 ret |= WORK_BUSY_RUNNING;
29145 - spin_unlock(&pool->lock);
29146 + spin_unlock_irqrestore(&pool->lock, flags);
29148 - local_irq_restore(flags);
29149 + rcu_read_unlock();
29153 @@ -4310,7 +4350,7 @@ void show_workqueue_state(void)
29154 unsigned long flags;
29157 - rcu_read_lock_sched();
29160 pr_info("Showing busy workqueues and worker pools:\n");
29162 @@ -4361,7 +4401,7 @@ void show_workqueue_state(void)
29163 spin_unlock_irqrestore(&pool->lock, flags);
29166 - rcu_read_unlock_sched();
29167 + rcu_read_unlock();
29171 @@ -4722,16 +4762,16 @@ bool freeze_workqueues_busy(void)
29172 * nr_active is monotonically decreasing. It's safe
29173 * to peek without lock.
29175 - rcu_read_lock_sched();
29177 for_each_pwq(pwq, wq) {
29178 WARN_ON_ONCE(pwq->nr_active < 0);
29179 if (pwq->nr_active) {
29181 - rcu_read_unlock_sched();
29182 + rcu_read_unlock();
29186 - rcu_read_unlock_sched();
29187 + rcu_read_unlock();
29190 mutex_unlock(&wq_pool_mutex);
29191 @@ -4921,7 +4961,8 @@ static ssize_t wq_pool_ids_show(struct device *dev,
29192 const char *delim = "";
29193 int node, written = 0;
29195 - rcu_read_lock_sched();
29196 + get_online_cpus();
29198 for_each_node(node) {
29199 written += scnprintf(buf + written, PAGE_SIZE - written,
29200 "%s%d:%d", delim, node,
29201 @@ -4929,7 +4970,8 @@ static ssize_t wq_pool_ids_show(struct device *dev,
29204 written += scnprintf(buf + written, PAGE_SIZE - written, "\n");
29205 - rcu_read_unlock_sched();
29206 + rcu_read_unlock();
29207 + put_online_cpus();
29211 diff --git a/kernel/workqueue_internal.h b/kernel/workqueue_internal.h
29212 index 45215870ac6c..f000c4d6917e 100644
29213 --- a/kernel/workqueue_internal.h
29214 +++ b/kernel/workqueue_internal.h
29215 @@ -43,6 +43,7 @@ struct worker {
29216 unsigned long last_active; /* L: last active timestamp */
29217 unsigned int flags; /* X: flags */
29218 int id; /* I: worker id */
29219 + int sleeping; /* None */
29222 * Opaque string set with work_set_desc(). Printed out with task
29223 @@ -68,7 +69,7 @@ static inline struct worker *current_wq_worker(void)
29224 * Scheduler hooks for concurrency managed workqueue. Only to be used from
29225 * sched/core.c and workqueue.c.
29227 -void wq_worker_waking_up(struct task_struct *task, int cpu);
29228 -struct task_struct *wq_worker_sleeping(struct task_struct *task, int cpu);
29229 +void wq_worker_running(struct task_struct *task);
29230 +void wq_worker_sleeping(struct task_struct *task);
29232 #endif /* _KERNEL_WORKQUEUE_INTERNAL_H */
29233 diff --git a/lib/Kconfig b/lib/Kconfig
29234 index 1a48744253d7..f75de578cca8 100644
29237 @@ -397,6 +397,7 @@ config CHECK_SIGNATURE
29239 config CPUMASK_OFFSTACK
29240 bool "Force CPU masks off stack" if DEBUG_PER_CPU_MAPS
29241 + depends on !PREEMPT_RT_FULL
29243 Use dynamic allocation for cpumask_var_t, instead of putting
29244 them on the stack. This is a bit more expensive, but avoids
29245 diff --git a/lib/debugobjects.c b/lib/debugobjects.c
29246 index 547f7f923dbc..8fcdbc2fc6d0 100644
29247 --- a/lib/debugobjects.c
29248 +++ b/lib/debugobjects.c
29249 @@ -309,7 +309,10 @@ __debug_object_init(void *addr, struct debug_obj_descr *descr, int onstack)
29250 struct debug_obj *obj;
29251 unsigned long flags;
29254 +#ifdef CONFIG_PREEMPT_RT_FULL
29255 + if (preempt_count() == 0 && !irqs_disabled())
29259 db = get_bucket((unsigned long) addr);
29261 diff --git a/lib/idr.c b/lib/idr.c
29262 index 6098336df267..9decbe914595 100644
29266 #include <linux/idr.h>
29267 #include <linux/spinlock.h>
29268 #include <linux/percpu.h>
29269 +#include <linux/locallock.h>
29271 #define MAX_IDR_SHIFT (sizeof(int) * 8 - 1)
29272 #define MAX_IDR_BIT (1U << MAX_IDR_SHIFT)
29273 @@ -45,6 +46,37 @@ static DEFINE_PER_CPU(struct idr_layer *, idr_preload_head);
29274 static DEFINE_PER_CPU(int, idr_preload_cnt);
29275 static DEFINE_SPINLOCK(simple_ida_lock);
29277 +#ifdef CONFIG_PREEMPT_RT_FULL
29278 +static DEFINE_LOCAL_IRQ_LOCK(idr_lock);
29280 +static inline void idr_preload_lock(void)
29282 + local_lock(idr_lock);
29285 +static inline void idr_preload_unlock(void)
29287 + local_unlock(idr_lock);
29290 +void idr_preload_end(void)
29292 + idr_preload_unlock();
29294 +EXPORT_SYMBOL(idr_preload_end);
29296 +static inline void idr_preload_lock(void)
29298 + preempt_disable();
29301 +static inline void idr_preload_unlock(void)
29303 + preempt_enable();
29308 /* the maximum ID which can be allocated given idr->layers */
29309 static int idr_max(int layers)
29311 @@ -115,14 +147,14 @@ static struct idr_layer *idr_layer_alloc(gfp_t gfp_mask, struct idr *layer_idr)
29312 * context. See idr_preload() for details.
29314 if (!in_interrupt()) {
29315 - preempt_disable();
29316 + idr_preload_lock();
29317 new = __this_cpu_read(idr_preload_head);
29319 __this_cpu_write(idr_preload_head, new->ary[0]);
29320 __this_cpu_dec(idr_preload_cnt);
29321 new->ary[0] = NULL;
29323 - preempt_enable();
29324 + idr_preload_unlock();
29328 @@ -366,7 +398,6 @@ static void idr_fill_slot(struct idr *idr, void *ptr, int id,
29329 idr_mark_full(pa, id);
29334 * idr_preload - preload for idr_alloc()
29335 * @gfp_mask: allocation mask to use for preloading
29336 @@ -401,7 +432,7 @@ void idr_preload(gfp_t gfp_mask)
29337 WARN_ON_ONCE(in_interrupt());
29338 might_sleep_if(gfpflags_allow_blocking(gfp_mask));
29340 - preempt_disable();
29341 + idr_preload_lock();
29344 * idr_alloc() is likely to succeed w/o full idr_layer buffer and
29345 @@ -413,9 +444,9 @@ void idr_preload(gfp_t gfp_mask)
29346 while (__this_cpu_read(idr_preload_cnt) < MAX_IDR_FREE) {
29347 struct idr_layer *new;
29349 - preempt_enable();
29350 + idr_preload_unlock();
29351 new = kmem_cache_zalloc(idr_layer_cache, gfp_mask);
29352 - preempt_disable();
29353 + idr_preload_lock();
29357 diff --git a/lib/locking-selftest.c b/lib/locking-selftest.c
29358 index 872a15a2a637..b93a6103fa4d 100644
29359 --- a/lib/locking-selftest.c
29360 +++ b/lib/locking-selftest.c
29361 @@ -590,6 +590,8 @@ GENERATE_TESTCASE(init_held_rsem)
29362 #include "locking-selftest-spin-hardirq.h"
29363 GENERATE_PERMUTATIONS_2_EVENTS(irqsafe1_hard_spin)
29365 +#ifndef CONFIG_PREEMPT_RT_FULL
29367 #include "locking-selftest-rlock-hardirq.h"
29368 GENERATE_PERMUTATIONS_2_EVENTS(irqsafe1_hard_rlock)
29370 @@ -605,9 +607,12 @@ GENERATE_PERMUTATIONS_2_EVENTS(irqsafe1_soft_rlock)
29371 #include "locking-selftest-wlock-softirq.h"
29372 GENERATE_PERMUTATIONS_2_EVENTS(irqsafe1_soft_wlock)
29379 +#ifndef CONFIG_PREEMPT_RT_FULL
29381 * Enabling hardirqs with a softirq-safe lock held:
29383 @@ -640,6 +645,8 @@ GENERATE_PERMUTATIONS_2_EVENTS(irqsafe2A_rlock)
29390 * Enabling irqs with an irq-safe lock held:
29392 @@ -663,6 +670,8 @@ GENERATE_PERMUTATIONS_2_EVENTS(irqsafe2A_rlock)
29393 #include "locking-selftest-spin-hardirq.h"
29394 GENERATE_PERMUTATIONS_2_EVENTS(irqsafe2B_hard_spin)
29396 +#ifndef CONFIG_PREEMPT_RT_FULL
29398 #include "locking-selftest-rlock-hardirq.h"
29399 GENERATE_PERMUTATIONS_2_EVENTS(irqsafe2B_hard_rlock)
29401 @@ -678,6 +687,8 @@ GENERATE_PERMUTATIONS_2_EVENTS(irqsafe2B_soft_rlock)
29402 #include "locking-selftest-wlock-softirq.h"
29403 GENERATE_PERMUTATIONS_2_EVENTS(irqsafe2B_soft_wlock)
29410 @@ -709,6 +720,8 @@ GENERATE_PERMUTATIONS_2_EVENTS(irqsafe2B_soft_wlock)
29411 #include "locking-selftest-spin-hardirq.h"
29412 GENERATE_PERMUTATIONS_3_EVENTS(irqsafe3_hard_spin)
29414 +#ifndef CONFIG_PREEMPT_RT_FULL
29416 #include "locking-selftest-rlock-hardirq.h"
29417 GENERATE_PERMUTATIONS_3_EVENTS(irqsafe3_hard_rlock)
29419 @@ -724,6 +737,8 @@ GENERATE_PERMUTATIONS_3_EVENTS(irqsafe3_soft_rlock)
29420 #include "locking-selftest-wlock-softirq.h"
29421 GENERATE_PERMUTATIONS_3_EVENTS(irqsafe3_soft_wlock)
29428 @@ -757,6 +772,8 @@ GENERATE_PERMUTATIONS_3_EVENTS(irqsafe3_soft_wlock)
29429 #include "locking-selftest-spin-hardirq.h"
29430 GENERATE_PERMUTATIONS_3_EVENTS(irqsafe4_hard_spin)
29432 +#ifndef CONFIG_PREEMPT_RT_FULL
29434 #include "locking-selftest-rlock-hardirq.h"
29435 GENERATE_PERMUTATIONS_3_EVENTS(irqsafe4_hard_rlock)
29437 @@ -772,10 +789,14 @@ GENERATE_PERMUTATIONS_3_EVENTS(irqsafe4_soft_rlock)
29438 #include "locking-selftest-wlock-softirq.h"
29439 GENERATE_PERMUTATIONS_3_EVENTS(irqsafe4_soft_wlock)
29447 +#ifndef CONFIG_PREEMPT_RT_FULL
29450 * read-lock / write-lock irq inversion.
29452 @@ -838,6 +859,10 @@ GENERATE_PERMUTATIONS_3_EVENTS(irq_inversion_soft_wlock)
29458 +#ifndef CONFIG_PREEMPT_RT_FULL
29461 * read-lock / write-lock recursion that is actually safe.
29463 @@ -876,6 +901,8 @@ GENERATE_PERMUTATIONS_3_EVENTS(irq_read_recursion_soft)
29470 * read-lock / write-lock recursion that is unsafe.
29472 @@ -1858,6 +1885,7 @@ void locking_selftest(void)
29474 printk(" --------------------------------------------------------------------------\n");
29476 +#ifndef CONFIG_PREEMPT_RT_FULL
29478 * irq-context testcases:
29480 @@ -1870,6 +1898,28 @@ void locking_selftest(void)
29482 DO_TESTCASE_6x2("irq read-recursion", irq_read_recursion);
29483 // DO_TESTCASE_6x2B("irq read-recursion #2", irq_read_recursion2);
29485 + /* On -rt, we only do hardirq context test for raw spinlock */
29486 + DO_TESTCASE_1B("hard-irqs-on + irq-safe-A", irqsafe1_hard_spin, 12);
29487 + DO_TESTCASE_1B("hard-irqs-on + irq-safe-A", irqsafe1_hard_spin, 21);
29489 + DO_TESTCASE_1B("hard-safe-A + irqs-on", irqsafe2B_hard_spin, 12);
29490 + DO_TESTCASE_1B("hard-safe-A + irqs-on", irqsafe2B_hard_spin, 21);
29492 + DO_TESTCASE_1B("hard-safe-A + unsafe-B #1", irqsafe3_hard_spin, 123);
29493 + DO_TESTCASE_1B("hard-safe-A + unsafe-B #1", irqsafe3_hard_spin, 132);
29494 + DO_TESTCASE_1B("hard-safe-A + unsafe-B #1", irqsafe3_hard_spin, 213);
29495 + DO_TESTCASE_1B("hard-safe-A + unsafe-B #1", irqsafe3_hard_spin, 231);
29496 + DO_TESTCASE_1B("hard-safe-A + unsafe-B #1", irqsafe3_hard_spin, 312);
29497 + DO_TESTCASE_1B("hard-safe-A + unsafe-B #1", irqsafe3_hard_spin, 321);
29499 + DO_TESTCASE_1B("hard-safe-A + unsafe-B #2", irqsafe4_hard_spin, 123);
29500 + DO_TESTCASE_1B("hard-safe-A + unsafe-B #2", irqsafe4_hard_spin, 132);
29501 + DO_TESTCASE_1B("hard-safe-A + unsafe-B #2", irqsafe4_hard_spin, 213);
29502 + DO_TESTCASE_1B("hard-safe-A + unsafe-B #2", irqsafe4_hard_spin, 231);
29503 + DO_TESTCASE_1B("hard-safe-A + unsafe-B #2", irqsafe4_hard_spin, 312);
29504 + DO_TESTCASE_1B("hard-safe-A + unsafe-B #2", irqsafe4_hard_spin, 321);
29509 diff --git a/lib/percpu_ida.c b/lib/percpu_ida.c
29510 index 6d40944960de..822a2c027e72 100644
29511 --- a/lib/percpu_ida.c
29512 +++ b/lib/percpu_ida.c
29514 #include <linux/string.h>
29515 #include <linux/spinlock.h>
29516 #include <linux/percpu_ida.h>
29517 +#include <linux/locallock.h>
29519 +static DEFINE_LOCAL_IRQ_LOCK(irq_off_lock);
29521 struct percpu_ida_cpu {
29523 @@ -148,13 +151,13 @@ int percpu_ida_alloc(struct percpu_ida *pool, int state)
29524 unsigned long flags;
29527 - local_irq_save(flags);
29528 + local_lock_irqsave(irq_off_lock, flags);
29529 tags = this_cpu_ptr(pool->tag_cpu);
29532 tag = alloc_local_tag(tags);
29533 if (likely(tag >= 0)) {
29534 - local_irq_restore(flags);
29535 + local_unlock_irqrestore(irq_off_lock, flags);
29539 @@ -173,6 +176,7 @@ int percpu_ida_alloc(struct percpu_ida *pool, int state)
29541 if (!tags->nr_free)
29542 alloc_global_tags(pool, tags);
29544 if (!tags->nr_free)
29545 steal_tags(pool, tags);
29547 @@ -184,7 +188,7 @@ int percpu_ida_alloc(struct percpu_ida *pool, int state)
29550 spin_unlock(&pool->lock);
29551 - local_irq_restore(flags);
29552 + local_unlock_irqrestore(irq_off_lock, flags);
29554 if (tag >= 0 || state == TASK_RUNNING)
29556 @@ -196,7 +200,7 @@ int percpu_ida_alloc(struct percpu_ida *pool, int state)
29560 - local_irq_save(flags);
29561 + local_lock_irqsave(irq_off_lock, flags);
29562 tags = this_cpu_ptr(pool->tag_cpu);
29564 if (state != TASK_RUNNING)
29565 @@ -221,7 +225,7 @@ void percpu_ida_free(struct percpu_ida *pool, unsigned tag)
29567 BUG_ON(tag >= pool->nr_tags);
29569 - local_irq_save(flags);
29570 + local_lock_irqsave(irq_off_lock, flags);
29571 tags = this_cpu_ptr(pool->tag_cpu);
29573 spin_lock(&tags->lock);
29574 @@ -253,7 +257,7 @@ void percpu_ida_free(struct percpu_ida *pool, unsigned tag)
29575 spin_unlock(&pool->lock);
29578 - local_irq_restore(flags);
29579 + local_unlock_irqrestore(irq_off_lock, flags);
29581 EXPORT_SYMBOL_GPL(percpu_ida_free);
29583 @@ -345,7 +349,7 @@ int percpu_ida_for_each_free(struct percpu_ida *pool, percpu_ida_cb fn,
29584 struct percpu_ida_cpu *remote;
29585 unsigned cpu, i, err = 0;
29587 - local_irq_save(flags);
29588 + local_lock_irqsave(irq_off_lock, flags);
29589 for_each_possible_cpu(cpu) {
29590 remote = per_cpu_ptr(pool->tag_cpu, cpu);
29591 spin_lock(&remote->lock);
29592 @@ -367,7 +371,7 @@ int percpu_ida_for_each_free(struct percpu_ida *pool, percpu_ida_cb fn,
29594 spin_unlock(&pool->lock);
29596 - local_irq_restore(flags);
29597 + local_unlock_irqrestore(irq_off_lock, flags);
29600 EXPORT_SYMBOL_GPL(percpu_ida_for_each_free);
29601 diff --git a/lib/radix-tree.c b/lib/radix-tree.c
29602 index 6b79e9026e24..44bf36a396a9 100644
29603 --- a/lib/radix-tree.c
29604 +++ b/lib/radix-tree.c
29606 #include <linux/bitops.h>
29607 #include <linux/rcupdate.h>
29608 #include <linux/preempt.h> /* in_interrupt() */
29610 +#include <linux/locallock.h>
29613 * The height_to_maxindex array needs to be one deeper than the maximum
29614 @@ -69,6 +69,7 @@ struct radix_tree_preload {
29615 struct radix_tree_node *nodes;
29617 static DEFINE_PER_CPU(struct radix_tree_preload, radix_tree_preloads) = { 0, };
29618 +static DEFINE_LOCAL_IRQ_LOCK(radix_tree_preloads_lock);
29620 static inline void *ptr_to_indirect(void *ptr)
29622 @@ -196,13 +197,14 @@ radix_tree_node_alloc(struct radix_tree_root *root)
29623 * succeed in getting a node here (and never reach
29624 * kmem_cache_alloc)
29626 - rtp = this_cpu_ptr(&radix_tree_preloads);
29627 + rtp = &get_locked_var(radix_tree_preloads_lock, radix_tree_preloads);
29630 rtp->nodes = ret->private_data;
29631 ret->private_data = NULL;
29634 + put_locked_var(radix_tree_preloads_lock, radix_tree_preloads);
29636 * Update the allocation stack trace as this is more useful
29638 @@ -257,14 +259,14 @@ static int __radix_tree_preload(gfp_t gfp_mask)
29639 struct radix_tree_node *node;
29642 - preempt_disable();
29643 + local_lock(radix_tree_preloads_lock);
29644 rtp = this_cpu_ptr(&radix_tree_preloads);
29645 while (rtp->nr < RADIX_TREE_PRELOAD_SIZE) {
29646 - preempt_enable();
29647 + local_unlock(radix_tree_preloads_lock);
29648 node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask);
29651 - preempt_disable();
29652 + local_lock(radix_tree_preloads_lock);
29653 rtp = this_cpu_ptr(&radix_tree_preloads);
29654 if (rtp->nr < RADIX_TREE_PRELOAD_SIZE) {
29655 node->private_data = rtp->nodes;
29656 @@ -306,11 +308,17 @@ int radix_tree_maybe_preload(gfp_t gfp_mask)
29657 if (gfpflags_allow_blocking(gfp_mask))
29658 return __radix_tree_preload(gfp_mask);
29659 /* Preloading doesn't help anything with this gfp mask, skip it */
29660 - preempt_disable();
29661 + local_lock(radix_tree_preloads_lock);
29664 EXPORT_SYMBOL(radix_tree_maybe_preload);
29666 +void radix_tree_preload_end(void)
29668 + local_unlock(radix_tree_preloads_lock);
29670 +EXPORT_SYMBOL(radix_tree_preload_end);
29673 * Return the maximum key which can be store into a
29674 * radix tree with height HEIGHT.
29675 diff --git a/lib/rbtree.c b/lib/rbtree.c
29676 index 1356454e36de..d15d6c4327f1 100644
29681 #include <linux/rbtree_augmented.h>
29682 #include <linux/export.h>
29683 +#include <linux/rcupdate.h>
29686 * red-black trees properties: http://en.wikipedia.org/wiki/Rbtree
29687 @@ -590,3 +591,13 @@ struct rb_node *rb_first_postorder(const struct rb_root *root)
29688 return rb_left_deepest_node(root->rb_node);
29690 EXPORT_SYMBOL(rb_first_postorder);
29692 +void rb_link_node_rcu(struct rb_node *node, struct rb_node *parent,
29693 + struct rb_node **rb_link)
29695 + node->__rb_parent_color = (unsigned long)parent;
29696 + node->rb_left = node->rb_right = NULL;
29698 + rcu_assign_pointer(*rb_link, node);
29700 +EXPORT_SYMBOL(rb_link_node_rcu);
29701 diff --git a/lib/scatterlist.c b/lib/scatterlist.c
29702 index bafa9933fa76..ebe3b7edd086 100644
29703 --- a/lib/scatterlist.c
29704 +++ b/lib/scatterlist.c
29705 @@ -620,7 +620,7 @@ void sg_miter_stop(struct sg_mapping_iter *miter)
29706 flush_kernel_dcache_page(miter->page);
29708 if (miter->__flags & SG_MITER_ATOMIC) {
29709 - WARN_ON_ONCE(preemptible());
29710 + WARN_ON_ONCE(!pagefault_disabled());
29711 kunmap_atomic(miter->addr);
29713 kunmap(miter->page);
29714 @@ -664,7 +664,7 @@ size_t sg_copy_buffer(struct scatterlist *sgl, unsigned int nents, void *buf,
29715 if (!sg_miter_skip(&miter, skip))
29718 - local_irq_save(flags);
29719 + local_irq_save_nort(flags);
29721 while (sg_miter_next(&miter) && offset < buflen) {
29723 @@ -681,7 +681,7 @@ size_t sg_copy_buffer(struct scatterlist *sgl, unsigned int nents, void *buf,
29725 sg_miter_stop(&miter);
29727 - local_irq_restore(flags);
29728 + local_irq_restore_nort(flags);
29731 EXPORT_SYMBOL(sg_copy_buffer);
29732 diff --git a/lib/smp_processor_id.c b/lib/smp_processor_id.c
29733 index 1afec32de6f2..11fa431046a8 100644
29734 --- a/lib/smp_processor_id.c
29735 +++ b/lib/smp_processor_id.c
29736 @@ -39,8 +39,9 @@ notrace static unsigned int check_preemption_disabled(const char *what1,
29737 if (!printk_ratelimit())
29740 - printk(KERN_ERR "BUG: using %s%s() in preemptible [%08x] code: %s/%d\n",
29741 - what1, what2, preempt_count() - 1, current->comm, current->pid);
29742 + printk(KERN_ERR "BUG: using %s%s() in preemptible [%08x %08x] code: %s/%d\n",
29743 + what1, what2, preempt_count() - 1, __migrate_disabled(current),
29744 + current->comm, current->pid);
29746 print_symbol("caller is %s\n", (long)__builtin_return_address(0));
29748 diff --git a/mm/Kconfig b/mm/Kconfig
29749 index 97a4e06b15c0..9614351e68b8 100644
29752 @@ -392,7 +392,7 @@ config NOMMU_INITIAL_TRIM_EXCESS
29754 config TRANSPARENT_HUGEPAGE
29755 bool "Transparent Hugepage Support"
29756 - depends on HAVE_ARCH_TRANSPARENT_HUGEPAGE
29757 + depends on HAVE_ARCH_TRANSPARENT_HUGEPAGE && !PREEMPT_RT_FULL
29760 Transparent Hugepages allows the kernel to use huge pages and
29761 diff --git a/mm/backing-dev.c b/mm/backing-dev.c
29762 index a988d4ef39da..f2c2ee1d5191 100644
29763 --- a/mm/backing-dev.c
29764 +++ b/mm/backing-dev.c
29765 @@ -457,9 +457,9 @@ void wb_congested_put(struct bdi_writeback_congested *congested)
29767 unsigned long flags;
29769 - local_irq_save(flags);
29770 + local_irq_save_nort(flags);
29771 if (!atomic_dec_and_lock(&congested->refcnt, &cgwb_lock)) {
29772 - local_irq_restore(flags);
29773 + local_irq_restore_nort(flags);
29777 diff --git a/mm/compaction.c b/mm/compaction.c
29778 index dba02dec7195..51963f58a29b 100644
29779 --- a/mm/compaction.c
29780 +++ b/mm/compaction.c
29781 @@ -1430,10 +1430,12 @@ check_drain:
29782 cc->migrate_pfn & ~((1UL << cc->order) - 1);
29784 if (cc->last_migrated_pfn < current_block_start) {
29786 + cpu = get_cpu_light();
29787 + local_lock_irq(swapvec_lock);
29788 lru_add_drain_cpu(cpu);
29789 + local_unlock_irq(swapvec_lock);
29790 drain_local_pages(zone);
29793 /* No more flushing until we migrate again */
29794 cc->last_migrated_pfn = 0;
29796 diff --git a/mm/filemap.c b/mm/filemap.c
29797 index 69f75c77c098..b203169ca0b4 100644
29800 @@ -144,9 +144,12 @@ static int page_cache_tree_insert(struct address_space *mapping,
29801 * node->private_list is protected by
29802 * mapping->tree_lock.
29804 - if (!list_empty(&node->private_list))
29805 - list_lru_del(&workingset_shadow_nodes,
29806 + if (!list_empty(&node->private_list)) {
29807 + local_lock(workingset_shadow_lock);
29808 + list_lru_del(&__workingset_shadow_nodes,
29809 &node->private_list);
29810 + local_unlock(workingset_shadow_lock);
29815 @@ -218,7 +221,9 @@ static void page_cache_tree_delete(struct address_space *mapping,
29816 if (!workingset_node_pages(node) &&
29817 list_empty(&node->private_list)) {
29818 node->private_data = mapping;
29819 - list_lru_add(&workingset_shadow_nodes, &node->private_list);
29820 + local_lock(workingset_shadow_lock);
29821 + list_lru_add(&__workingset_shadow_nodes, &node->private_list);
29822 + local_unlock(workingset_shadow_lock);
29826 diff --git a/mm/highmem.c b/mm/highmem.c
29827 index 123bcd3ed4f2..16e8cf26d38a 100644
29830 @@ -29,10 +29,11 @@
29831 #include <linux/kgdb.h>
29832 #include <asm/tlbflush.h>
29835 +#ifndef CONFIG_PREEMPT_RT_FULL
29836 #if defined(CONFIG_HIGHMEM) || defined(CONFIG_X86_32)
29837 DEFINE_PER_CPU(int, __kmap_atomic_idx);
29842 * Virtual_count is not a pure "count".
29843 @@ -107,8 +108,9 @@ static inline wait_queue_head_t *get_pkmap_wait_queue_head(unsigned int color)
29844 unsigned long totalhigh_pages __read_mostly;
29845 EXPORT_SYMBOL(totalhigh_pages);
29848 +#ifndef CONFIG_PREEMPT_RT_FULL
29849 EXPORT_PER_CPU_SYMBOL(__kmap_atomic_idx);
29852 unsigned int nr_free_highpages (void)
29854 diff --git a/mm/memcontrol.c b/mm/memcontrol.c
29855 index e25b93a4267d..1c619267d9da 100644
29856 --- a/mm/memcontrol.c
29857 +++ b/mm/memcontrol.c
29859 #include <net/sock.h>
29860 #include <net/ip.h>
29861 #include <net/tcp_memcontrol.h>
29862 +#include <linux/locallock.h>
29866 #include <asm/uaccess.h>
29867 @@ -87,6 +89,7 @@ int do_swap_account __read_mostly;
29868 #define do_swap_account 0
29871 +static DEFINE_LOCAL_IRQ_LOCK(event_lock);
29872 static const char * const mem_cgroup_stat_names[] = {
29875 @@ -1922,14 +1925,17 @@ static void drain_local_stock(struct work_struct *dummy)
29877 static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
29879 - struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
29880 + struct memcg_stock_pcp *stock;
29881 + int cpu = get_cpu_light();
29883 + stock = &per_cpu(memcg_stock, cpu);
29885 if (stock->cached != memcg) { /* reset if necessary */
29886 drain_stock(stock);
29887 stock->cached = memcg;
29889 stock->nr_pages += nr_pages;
29890 - put_cpu_var(memcg_stock);
29895 @@ -1945,7 +1951,7 @@ static void drain_all_stock(struct mem_cgroup *root_memcg)
29897 /* Notify other cpus that system-wide "drain" is running */
29899 - curcpu = get_cpu();
29900 + curcpu = get_cpu_light();
29901 for_each_online_cpu(cpu) {
29902 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
29903 struct mem_cgroup *memcg;
29904 @@ -1962,7 +1968,7 @@ static void drain_all_stock(struct mem_cgroup *root_memcg)
29905 schedule_work_on(cpu, &stock->work);
29911 mutex_unlock(&percpu_charge_mutex);
29913 @@ -4691,12 +4697,12 @@ static int mem_cgroup_move_account(struct page *page,
29917 - local_irq_disable();
29918 + local_lock_irq(event_lock);
29919 mem_cgroup_charge_statistics(to, page, nr_pages);
29920 memcg_check_events(to, page);
29921 mem_cgroup_charge_statistics(from, page, -nr_pages);
29922 memcg_check_events(from, page);
29923 - local_irq_enable();
29924 + local_unlock_irq(event_lock);
29928 @@ -5486,10 +5492,10 @@ void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
29929 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
29932 - local_irq_disable();
29933 + local_lock_irq(event_lock);
29934 mem_cgroup_charge_statistics(memcg, page, nr_pages);
29935 memcg_check_events(memcg, page);
29936 - local_irq_enable();
29937 + local_unlock_irq(event_lock);
29939 if (do_swap_account && PageSwapCache(page)) {
29940 swp_entry_t entry = { .val = page_private(page) };
29941 @@ -5545,14 +5551,14 @@ static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout,
29942 memcg_oom_recover(memcg);
29945 - local_irq_save(flags);
29946 + local_lock_irqsave(event_lock, flags);
29947 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon);
29948 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file);
29949 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge);
29950 __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout);
29951 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
29952 memcg_check_events(memcg, dummy_page);
29953 - local_irq_restore(flags);
29954 + local_unlock_irqrestore(event_lock, flags);
29956 if (!mem_cgroup_is_root(memcg))
29957 css_put_many(&memcg->css, nr_pages);
29958 @@ -5762,6 +5768,7 @@ void mem_cgroup_swapout(struct page *page, swp_entry_t entry)
29960 struct mem_cgroup *memcg, *swap_memcg;
29961 unsigned short oldid;
29962 + unsigned long flags;
29964 VM_BUG_ON_PAGE(PageLRU(page), page);
29965 VM_BUG_ON_PAGE(page_count(page), page);
29966 @@ -5802,12 +5809,16 @@ void mem_cgroup_swapout(struct page *page, swp_entry_t entry)
29967 * important here to have the interrupts disabled because it is the
29968 * only synchronisation we have for udpating the per-CPU variables.
29970 + local_lock_irqsave(event_lock, flags);
29971 +#ifndef CONFIG_PREEMPT_RT_BASE
29972 VM_BUG_ON(!irqs_disabled());
29974 mem_cgroup_charge_statistics(memcg, page, -1);
29975 memcg_check_events(memcg, page);
29977 if (!mem_cgroup_is_root(memcg))
29978 css_put(&memcg->css);
29979 + local_unlock_irqrestore(event_lock, flags);
29983 diff --git a/mm/mmu_context.c b/mm/mmu_context.c
29984 index f802c2d216a7..b1b6f238e42d 100644
29985 --- a/mm/mmu_context.c
29986 +++ b/mm/mmu_context.c
29987 @@ -23,6 +23,7 @@ void use_mm(struct mm_struct *mm)
29988 struct task_struct *tsk = current;
29991 + preempt_disable_rt();
29992 active_mm = tsk->active_mm;
29993 if (active_mm != mm) {
29994 atomic_inc(&mm->mm_count);
29995 @@ -30,6 +31,7 @@ void use_mm(struct mm_struct *mm)
29998 switch_mm(active_mm, mm, tsk);
29999 + preempt_enable_rt();
30001 #ifdef finish_arch_post_lock_switch
30002 finish_arch_post_lock_switch();
30003 diff --git a/mm/page_alloc.c b/mm/page_alloc.c
30004 index 6f9005dcca2e..e8ec72251705 100644
30005 --- a/mm/page_alloc.c
30006 +++ b/mm/page_alloc.c
30008 #include <linux/page_ext.h>
30009 #include <linux/hugetlb.h>
30010 #include <linux/sched/rt.h>
30011 +#include <linux/locallock.h>
30012 #include <linux/page_owner.h>
30013 #include <linux/kthread.h>
30015 @@ -264,6 +265,18 @@ EXPORT_SYMBOL(nr_node_ids);
30016 EXPORT_SYMBOL(nr_online_nodes);
30019 +static DEFINE_LOCAL_IRQ_LOCK(pa_lock);
30021 +#ifdef CONFIG_PREEMPT_RT_BASE
30022 +# define cpu_lock_irqsave(cpu, flags) \
30023 + local_lock_irqsave_on(pa_lock, flags, cpu)
30024 +# define cpu_unlock_irqrestore(cpu, flags) \
30025 + local_unlock_irqrestore_on(pa_lock, flags, cpu)
30027 +# define cpu_lock_irqsave(cpu, flags) local_irq_save(flags)
30028 +# define cpu_unlock_irqrestore(cpu, flags) local_irq_restore(flags)
30031 int page_group_by_mobility_disabled __read_mostly;
30033 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
30034 @@ -786,7 +799,7 @@ static inline int free_pages_check(struct page *page)
30038 - * Frees a number of pages from the PCP lists
30039 + * Frees a number of pages which have been collected from the pcp lists.
30040 * Assumes all pages on list are in same zone, and of same order.
30041 * count is the number of pages to free.
30043 @@ -797,18 +810,53 @@ static inline int free_pages_check(struct page *page)
30044 * pinned" detection logic.
30046 static void free_pcppages_bulk(struct zone *zone, int count,
30047 - struct per_cpu_pages *pcp)
30048 + struct list_head *list)
30050 - int migratetype = 0;
30051 - int batch_free = 0;
30052 int to_free = count;
30053 unsigned long nr_scanned;
30054 + unsigned long flags;
30056 + spin_lock_irqsave(&zone->lock, flags);
30058 - spin_lock(&zone->lock);
30059 nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED);
30061 __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
30063 + while (!list_empty(list)) {
30064 + struct page *page = list_first_entry(list, struct page, lru);
30065 + int mt; /* migratetype of the to-be-freed page */
30067 + /* must delete as __free_one_page list manipulates */
30068 + list_del(&page->lru);
30070 + mt = get_pcppage_migratetype(page);
30071 + /* MIGRATE_ISOLATE page should not go to pcplists */
30072 + VM_BUG_ON_PAGE(is_migrate_isolate(mt), page);
30073 + /* Pageblock could have been isolated meanwhile */
30074 + if (unlikely(has_isolate_pageblock(zone)))
30075 + mt = get_pageblock_migratetype(page);
30077 + __free_one_page(page, page_to_pfn(page), zone, 0, mt);
30078 + trace_mm_page_pcpu_drain(page, 0, mt);
30081 + WARN_ON(to_free != 0);
30082 + spin_unlock_irqrestore(&zone->lock, flags);
30086 + * Moves a number of pages from the PCP lists to free list which
30087 + * is freed outside of the locked region.
30089 + * Assumes all pages on list are in same zone, and of same order.
30090 + * count is the number of pages to free.
30092 +static void isolate_pcp_pages(int to_free, struct per_cpu_pages *src,
30093 + struct list_head *dst)
30095 + int migratetype = 0;
30096 + int batch_free = 0;
30100 struct list_head *list;
30101 @@ -824,7 +872,7 @@ static void free_pcppages_bulk(struct zone *zone, int count,
30103 if (++migratetype == MIGRATE_PCPTYPES)
30105 - list = &pcp->lists[migratetype];
30106 + list = &src->lists[migratetype];
30107 } while (list_empty(list));
30109 /* This is the only non-empty list. Free them all. */
30110 @@ -832,24 +880,12 @@ static void free_pcppages_bulk(struct zone *zone, int count,
30111 batch_free = to_free;
30114 - int mt; /* migratetype of the to-be-freed page */
30116 - page = list_entry(list->prev, struct page, lru);
30117 - /* must delete as __free_one_page list manipulates */
30118 + page = list_last_entry(list, struct page, lru);
30119 list_del(&page->lru);
30121 - mt = get_pcppage_migratetype(page);
30122 - /* MIGRATE_ISOLATE page should not go to pcplists */
30123 - VM_BUG_ON_PAGE(is_migrate_isolate(mt), page);
30124 - /* Pageblock could have been isolated meanwhile */
30125 - if (unlikely(has_isolate_pageblock(zone)))
30126 - mt = get_pageblock_migratetype(page);
30128 - __free_one_page(page, page_to_pfn(page), zone, 0, mt);
30129 - trace_mm_page_pcpu_drain(page, 0, mt);
30130 + list_add(&page->lru, dst);
30131 } while (--to_free && --batch_free && !list_empty(list));
30133 - spin_unlock(&zone->lock);
30136 static void free_one_page(struct zone *zone,
30137 @@ -858,7 +894,9 @@ static void free_one_page(struct zone *zone,
30140 unsigned long nr_scanned;
30141 - spin_lock(&zone->lock);
30142 + unsigned long flags;
30144 + spin_lock_irqsave(&zone->lock, flags);
30145 nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED);
30147 __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
30148 @@ -868,7 +906,7 @@ static void free_one_page(struct zone *zone,
30149 migratetype = get_pfnblock_migratetype(page, pfn);
30151 __free_one_page(page, pfn, zone, order, migratetype);
30152 - spin_unlock(&zone->lock);
30153 + spin_unlock_irqrestore(&zone->lock, flags);
30156 static int free_tail_pages_check(struct page *head_page, struct page *page)
30157 @@ -1019,10 +1057,10 @@ static void __free_pages_ok(struct page *page, unsigned int order)
30160 migratetype = get_pfnblock_migratetype(page, pfn);
30161 - local_irq_save(flags);
30162 + local_lock_irqsave(pa_lock, flags);
30163 __count_vm_events(PGFREE, 1 << order);
30164 free_one_page(page_zone(page), page, pfn, order, migratetype);
30165 - local_irq_restore(flags);
30166 + local_unlock_irqrestore(pa_lock, flags);
30169 static void __init __free_pages_boot_core(struct page *page,
30170 @@ -1879,16 +1917,18 @@ static int rmqueue_bulk(struct zone *zone, unsigned int order,
30171 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
30173 unsigned long flags;
30175 int to_drain, batch;
30177 - local_irq_save(flags);
30178 + local_lock_irqsave(pa_lock, flags);
30179 batch = READ_ONCE(pcp->batch);
30180 to_drain = min(pcp->count, batch);
30181 if (to_drain > 0) {
30182 - free_pcppages_bulk(zone, to_drain, pcp);
30183 + isolate_pcp_pages(to_drain, pcp, &dst);
30184 pcp->count -= to_drain;
30186 - local_irq_restore(flags);
30187 + local_unlock_irqrestore(pa_lock, flags);
30188 + free_pcppages_bulk(zone, to_drain, &dst);
30192 @@ -1904,16 +1944,21 @@ static void drain_pages_zone(unsigned int cpu, struct zone *zone)
30193 unsigned long flags;
30194 struct per_cpu_pageset *pset;
30195 struct per_cpu_pages *pcp;
30199 - local_irq_save(flags);
30200 + cpu_lock_irqsave(cpu, flags);
30201 pset = per_cpu_ptr(zone->pageset, cpu);
30204 - if (pcp->count) {
30205 - free_pcppages_bulk(zone, pcp->count, pcp);
30206 + count = pcp->count;
30208 + isolate_pcp_pages(count, pcp, &dst);
30211 - local_irq_restore(flags);
30212 + cpu_unlock_irqrestore(cpu, flags);
30214 + free_pcppages_bulk(zone, count, &dst);
30218 @@ -1999,8 +2044,17 @@ void drain_all_pages(struct zone *zone)
30220 cpumask_clear_cpu(cpu, &cpus_with_pcps);
30222 +#ifndef CONFIG_PREEMPT_RT_BASE
30223 on_each_cpu_mask(&cpus_with_pcps, (smp_call_func_t) drain_local_pages,
30226 + for_each_cpu(cpu, &cpus_with_pcps) {
30228 + drain_pages_zone(cpu, zone);
30230 + drain_pages(cpu);
30235 #ifdef CONFIG_HIBERNATION
30236 @@ -2056,7 +2110,7 @@ void free_hot_cold_page(struct page *page, bool cold)
30238 migratetype = get_pfnblock_migratetype(page, pfn);
30239 set_pcppage_migratetype(page, migratetype);
30240 - local_irq_save(flags);
30241 + local_lock_irqsave(pa_lock, flags);
30242 __count_vm_event(PGFREE);
30245 @@ -2082,12 +2136,17 @@ void free_hot_cold_page(struct page *page, bool cold)
30247 if (pcp->count >= pcp->high) {
30248 unsigned long batch = READ_ONCE(pcp->batch);
30249 - free_pcppages_bulk(zone, batch, pcp);
30252 + isolate_pcp_pages(batch, pcp, &dst);
30253 pcp->count -= batch;
30254 + local_unlock_irqrestore(pa_lock, flags);
30255 + free_pcppages_bulk(zone, batch, &dst);
30260 - local_irq_restore(flags);
30261 + local_unlock_irqrestore(pa_lock, flags);
30265 @@ -2222,7 +2281,7 @@ struct page *buffered_rmqueue(struct zone *preferred_zone,
30266 struct per_cpu_pages *pcp;
30267 struct list_head *list;
30269 - local_irq_save(flags);
30270 + local_lock_irqsave(pa_lock, flags);
30271 pcp = &this_cpu_ptr(zone->pageset)->pcp;
30272 list = &pcp->lists[migratetype];
30273 if (list_empty(list)) {
30274 @@ -2254,7 +2313,7 @@ struct page *buffered_rmqueue(struct zone *preferred_zone,
30276 WARN_ON_ONCE(order > 1);
30278 - spin_lock_irqsave(&zone->lock, flags);
30279 + local_spin_lock_irqsave(pa_lock, &zone->lock, flags);
30282 if (alloc_flags & ALLOC_HARDER) {
30283 @@ -2264,11 +2323,13 @@ struct page *buffered_rmqueue(struct zone *preferred_zone,
30286 page = __rmqueue(zone, order, migratetype, gfp_flags);
30287 - spin_unlock(&zone->lock);
30290 + spin_unlock(&zone->lock);
30293 __mod_zone_freepage_state(zone, -(1 << order),
30294 get_pcppage_migratetype(page));
30295 + spin_unlock(&zone->lock);
30298 __mod_zone_page_state(zone, NR_ALLOC_BATCH, -(1 << order));
30299 @@ -2278,13 +2339,13 @@ struct page *buffered_rmqueue(struct zone *preferred_zone,
30301 __count_zone_vm_events(PGALLOC, zone, 1 << order);
30302 zone_statistics(preferred_zone, zone, gfp_flags);
30303 - local_irq_restore(flags);
30304 + local_unlock_irqrestore(pa_lock, flags);
30306 VM_BUG_ON_PAGE(bad_range(zone, page), page);
30310 - local_irq_restore(flags);
30311 + local_unlock_irqrestore(pa_lock, flags);
30315 @@ -5953,6 +6014,7 @@ static int page_alloc_cpu_notify(struct notifier_block *self,
30316 void __init page_alloc_init(void)
30318 hotcpu_notifier(page_alloc_cpu_notify, 0);
30319 + local_irq_lock_init(pa_lock);
30323 @@ -6847,7 +6909,7 @@ void zone_pcp_reset(struct zone *zone)
30324 struct per_cpu_pageset *pset;
30326 /* avoid races with drain_pages() */
30327 - local_irq_save(flags);
30328 + local_lock_irqsave(pa_lock, flags);
30329 if (zone->pageset != &boot_pageset) {
30330 for_each_online_cpu(cpu) {
30331 pset = per_cpu_ptr(zone->pageset, cpu);
30332 @@ -6856,7 +6918,7 @@ void zone_pcp_reset(struct zone *zone)
30333 free_percpu(zone->pageset);
30334 zone->pageset = &boot_pageset;
30336 - local_irq_restore(flags);
30337 + local_unlock_irqrestore(pa_lock, flags);
30340 #ifdef CONFIG_MEMORY_HOTREMOVE
30341 diff --git a/mm/percpu.c b/mm/percpu.c
30342 index ef6353f0adbd..33ccbac7cdb8 100644
30345 @@ -1285,18 +1285,7 @@ void free_percpu(void __percpu *ptr)
30347 EXPORT_SYMBOL_GPL(free_percpu);
30350 - * is_kernel_percpu_address - test whether address is from static percpu area
30351 - * @addr: address to test
30353 - * Test whether @addr belongs to in-kernel static percpu area. Module
30354 - * static percpu areas are not considered. For those, use
30355 - * is_module_percpu_address().
30358 - * %true if @addr is from in-kernel static percpu area, %false otherwise.
30360 -bool is_kernel_percpu_address(unsigned long addr)
30361 +bool __is_kernel_percpu_address(unsigned long addr, unsigned long *can_addr)
30364 const size_t static_size = __per_cpu_end - __per_cpu_start;
30365 @@ -1305,16 +1294,36 @@ bool is_kernel_percpu_address(unsigned long addr)
30367 for_each_possible_cpu(cpu) {
30368 void *start = per_cpu_ptr(base, cpu);
30369 + void *va = (void *)addr;
30371 - if ((void *)addr >= start && (void *)addr < start + static_size)
30372 + if (va >= start && va < start + static_size) {
30374 + *can_addr = (unsigned long) (va - start);
30380 /* on UP, can't distinguish from other static vars, always false */
30385 + * is_kernel_percpu_address - test whether address is from static percpu area
30386 + * @addr: address to test
30388 + * Test whether @addr belongs to in-kernel static percpu area. Module
30389 + * static percpu areas are not considered. For those, use
30390 + * is_module_percpu_address().
30393 + * %true if @addr is from in-kernel static percpu area, %false otherwise.
30395 +bool is_kernel_percpu_address(unsigned long addr)
30397 + return __is_kernel_percpu_address(addr, NULL);
30401 * per_cpu_ptr_to_phys - convert translated percpu address to physical address
30402 * @addr: the address to be converted to physical address
30404 diff --git a/mm/slab.h b/mm/slab.h
30405 index 7b6087197997..afdc57941179 100644
30408 @@ -324,7 +324,11 @@ static inline struct kmem_cache *cache_from_obj(struct kmem_cache *s, void *x)
30409 * The slab lists for all objects.
30411 struct kmem_cache_node {
30412 +#ifdef CONFIG_SLUB
30413 + raw_spinlock_t list_lock;
30415 spinlock_t list_lock;
30419 struct list_head slabs_partial; /* partial list first, better asm code */
30420 diff --git a/mm/slub.c b/mm/slub.c
30421 index 65d5f92d51d2..feb4a445a546 100644
30424 @@ -1075,7 +1075,7 @@ static noinline struct kmem_cache_node *free_debug_processing(
30425 void *object = head;
30428 - spin_lock_irqsave(&n->list_lock, *flags);
30429 + raw_spin_lock_irqsave(&n->list_lock, *flags);
30432 if (!check_slab(s, page))
30433 @@ -1136,7 +1136,7 @@ out:
30437 - spin_unlock_irqrestore(&n->list_lock, *flags);
30438 + raw_spin_unlock_irqrestore(&n->list_lock, *flags);
30439 slab_fix(s, "Object at 0x%p not freed", object);
30442 @@ -1263,6 +1263,12 @@ static inline void dec_slabs_node(struct kmem_cache *s, int node,
30444 #endif /* CONFIG_SLUB_DEBUG */
30446 +struct slub_free_list {
30447 + raw_spinlock_t lock;
30448 + struct list_head list;
30450 +static DEFINE_PER_CPU(struct slub_free_list, slub_free_list);
30453 * Hooks for other subsystems that check memory allocations. In a typical
30454 * production configuration these hooks all should produce no code at all.
30455 @@ -1399,10 +1405,17 @@ static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
30459 + bool enableirqs = false;
30461 flags &= gfp_allowed_mask;
30463 if (gfpflags_allow_blocking(flags))
30464 + enableirqs = true;
30465 +#ifdef CONFIG_PREEMPT_RT_FULL
30466 + if (system_state == SYSTEM_RUNNING)
30467 + enableirqs = true;
30470 local_irq_enable();
30472 flags |= s->allocflags;
30473 @@ -1473,7 +1486,7 @@ static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
30477 - if (gfpflags_allow_blocking(flags))
30479 local_irq_disable();
30482 @@ -1529,6 +1542,16 @@ static void __free_slab(struct kmem_cache *s, struct page *page)
30483 __free_kmem_pages(page, order);
30486 +static void free_delayed(struct list_head *h)
30488 + while(!list_empty(h)) {
30489 + struct page *page = list_first_entry(h, struct page, lru);
30491 + list_del(&page->lru);
30492 + __free_slab(page->slab_cache, page);
30496 #define need_reserve_slab_rcu \
30497 (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head))
30499 @@ -1560,6 +1583,12 @@ static void free_slab(struct kmem_cache *s, struct page *page)
30502 call_rcu(head, rcu_free_slab);
30503 + } else if (irqs_disabled()) {
30504 + struct slub_free_list *f = this_cpu_ptr(&slub_free_list);
30506 + raw_spin_lock(&f->lock);
30507 + list_add(&page->lru, &f->list);
30508 + raw_spin_unlock(&f->lock);
30510 __free_slab(s, page);
30512 @@ -1673,7 +1702,7 @@ static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n,
30513 if (!n || !n->nr_partial)
30516 - spin_lock(&n->list_lock);
30517 + raw_spin_lock(&n->list_lock);
30518 list_for_each_entry_safe(page, page2, &n->partial, lru) {
30521 @@ -1698,7 +1727,7 @@ static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n,
30525 - spin_unlock(&n->list_lock);
30526 + raw_spin_unlock(&n->list_lock);
30530 @@ -1944,7 +1973,7 @@ redo:
30531 * that acquire_slab() will see a slab page that
30534 - spin_lock(&n->list_lock);
30535 + raw_spin_lock(&n->list_lock);
30539 @@ -1955,7 +1984,7 @@ redo:
30540 * slabs from diagnostic functions will not see
30541 * any frozen slabs.
30543 - spin_lock(&n->list_lock);
30544 + raw_spin_lock(&n->list_lock);
30548 @@ -1990,7 +2019,7 @@ redo:
30552 - spin_unlock(&n->list_lock);
30553 + raw_spin_unlock(&n->list_lock);
30556 stat(s, DEACTIVATE_EMPTY);
30557 @@ -2022,10 +2051,10 @@ static void unfreeze_partials(struct kmem_cache *s,
30558 n2 = get_node(s, page_to_nid(page));
30561 - spin_unlock(&n->list_lock);
30562 + raw_spin_unlock(&n->list_lock);
30565 - spin_lock(&n->list_lock);
30566 + raw_spin_lock(&n->list_lock);
30570 @@ -2054,7 +2083,7 @@ static void unfreeze_partials(struct kmem_cache *s,
30574 - spin_unlock(&n->list_lock);
30575 + raw_spin_unlock(&n->list_lock);
30577 while (discard_page) {
30578 page = discard_page;
30579 @@ -2093,14 +2122,21 @@ static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain)
30580 pobjects = oldpage->pobjects;
30581 pages = oldpage->pages;
30582 if (drain && pobjects > s->cpu_partial) {
30583 + struct slub_free_list *f;
30584 unsigned long flags;
30585 + LIST_HEAD(tofree);
30587 * partial array is full. Move the existing
30588 * set to the per node partial list.
30590 local_irq_save(flags);
30591 unfreeze_partials(s, this_cpu_ptr(s->cpu_slab));
30592 + f = this_cpu_ptr(&slub_free_list);
30593 + raw_spin_lock(&f->lock);
30594 + list_splice_init(&f->list, &tofree);
30595 + raw_spin_unlock(&f->lock);
30596 local_irq_restore(flags);
30597 + free_delayed(&tofree);
30601 @@ -2172,7 +2208,22 @@ static bool has_cpu_slab(int cpu, void *info)
30603 static void flush_all(struct kmem_cache *s)
30605 + LIST_HEAD(tofree);
30608 on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC);
30609 + for_each_online_cpu(cpu) {
30610 + struct slub_free_list *f;
30612 + if (!has_cpu_slab(cpu, s))
30615 + f = &per_cpu(slub_free_list, cpu);
30616 + raw_spin_lock_irq(&f->lock);
30617 + list_splice_init(&f->list, &tofree);
30618 + raw_spin_unlock_irq(&f->lock);
30619 + free_delayed(&tofree);
30624 @@ -2208,10 +2259,10 @@ static unsigned long count_partial(struct kmem_cache_node *n,
30625 unsigned long x = 0;
30628 - spin_lock_irqsave(&n->list_lock, flags);
30629 + raw_spin_lock_irqsave(&n->list_lock, flags);
30630 list_for_each_entry(page, &n->partial, lru)
30631 x += get_count(page);
30632 - spin_unlock_irqrestore(&n->list_lock, flags);
30633 + raw_spin_unlock_irqrestore(&n->list_lock, flags);
30636 #endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */
30637 @@ -2349,8 +2400,10 @@ static inline void *get_freelist(struct kmem_cache *s, struct page *page)
30638 * already disabled (which is the case for bulk allocation).
30640 static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
30641 - unsigned long addr, struct kmem_cache_cpu *c)
30642 + unsigned long addr, struct kmem_cache_cpu *c,
30643 + struct list_head *to_free)
30645 + struct slub_free_list *f;
30649 @@ -2410,6 +2463,13 @@ load_freelist:
30650 VM_BUG_ON(!c->page->frozen);
30651 c->freelist = get_freepointer(s, freelist);
30652 c->tid = next_tid(c->tid);
30655 + f = this_cpu_ptr(&slub_free_list);
30656 + raw_spin_lock(&f->lock);
30657 + list_splice_init(&f->list, to_free);
30658 + raw_spin_unlock(&f->lock);
30663 @@ -2441,7 +2501,7 @@ new_slab:
30664 deactivate_slab(s, page, get_freepointer(s, freelist));
30666 c->freelist = NULL;
30672 @@ -2453,6 +2513,7 @@ static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
30675 unsigned long flags;
30676 + LIST_HEAD(tofree);
30678 local_irq_save(flags);
30679 #ifdef CONFIG_PREEMPT
30680 @@ -2464,8 +2525,9 @@ static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
30681 c = this_cpu_ptr(s->cpu_slab);
30684 - p = ___slab_alloc(s, gfpflags, node, addr, c);
30685 + p = ___slab_alloc(s, gfpflags, node, addr, c, &tofree);
30686 local_irq_restore(flags);
30687 + free_delayed(&tofree);
30691 @@ -2652,7 +2714,7 @@ static void __slab_free(struct kmem_cache *s, struct page *page,
30695 - spin_unlock_irqrestore(&n->list_lock, flags);
30696 + raw_spin_unlock_irqrestore(&n->list_lock, flags);
30699 prior = page->freelist;
30700 @@ -2684,7 +2746,7 @@ static void __slab_free(struct kmem_cache *s, struct page *page,
30701 * Otherwise the list_lock will synchronize with
30702 * other processors updating the list of slabs.
30704 - spin_lock_irqsave(&n->list_lock, flags);
30705 + raw_spin_lock_irqsave(&n->list_lock, flags);
30709 @@ -2726,7 +2788,7 @@ static void __slab_free(struct kmem_cache *s, struct page *page,
30710 add_partial(n, page, DEACTIVATE_TO_TAIL);
30711 stat(s, FREE_ADD_PARTIAL);
30713 - spin_unlock_irqrestore(&n->list_lock, flags);
30714 + raw_spin_unlock_irqrestore(&n->list_lock, flags);
30718 @@ -2741,7 +2803,7 @@ slab_empty:
30719 remove_full(s, n, page);
30722 - spin_unlock_irqrestore(&n->list_lock, flags);
30723 + raw_spin_unlock_irqrestore(&n->list_lock, flags);
30724 stat(s, FREE_SLAB);
30725 discard_slab(s, page);
30727 @@ -2913,6 +2975,7 @@ int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size,
30730 struct kmem_cache_cpu *c;
30731 + LIST_HEAD(to_free);
30734 /* memcg and kmem_cache debug support */
30735 @@ -2936,7 +2999,7 @@ int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size,
30736 * of re-populating per CPU c->freelist
30738 p[i] = ___slab_alloc(s, flags, NUMA_NO_NODE,
30740 + _RET_IP_, c, &to_free);
30741 if (unlikely(!p[i]))
30744 @@ -2948,6 +3011,7 @@ int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size,
30746 c->tid = next_tid(c->tid);
30747 local_irq_enable();
30748 + free_delayed(&to_free);
30750 /* Clear memory outside IRQ disabled fastpath loop */
30751 if (unlikely(flags & __GFP_ZERO)) {
30752 @@ -3095,7 +3159,7 @@ static void
30753 init_kmem_cache_node(struct kmem_cache_node *n)
30756 - spin_lock_init(&n->list_lock);
30757 + raw_spin_lock_init(&n->list_lock);
30758 INIT_LIST_HEAD(&n->partial);
30759 #ifdef CONFIG_SLUB_DEBUG
30760 atomic_long_set(&n->nr_slabs, 0);
30761 @@ -3677,7 +3741,7 @@ int __kmem_cache_shrink(struct kmem_cache *s, bool deactivate)
30762 for (i = 0; i < SHRINK_PROMOTE_MAX; i++)
30763 INIT_LIST_HEAD(promote + i);
30765 - spin_lock_irqsave(&n->list_lock, flags);
30766 + raw_spin_lock_irqsave(&n->list_lock, flags);
30769 * Build lists of slabs to discard or promote.
30770 @@ -3708,7 +3772,7 @@ int __kmem_cache_shrink(struct kmem_cache *s, bool deactivate)
30771 for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--)
30772 list_splice(promote + i, &n->partial);
30774 - spin_unlock_irqrestore(&n->list_lock, flags);
30775 + raw_spin_unlock_irqrestore(&n->list_lock, flags);
30777 /* Release empty slabs */
30778 list_for_each_entry_safe(page, t, &discard, lru)
30779 @@ -3884,6 +3948,12 @@ void __init kmem_cache_init(void)
30781 static __initdata struct kmem_cache boot_kmem_cache,
30782 boot_kmem_cache_node;
30785 + for_each_possible_cpu(cpu) {
30786 + raw_spin_lock_init(&per_cpu(slub_free_list, cpu).lock);
30787 + INIT_LIST_HEAD(&per_cpu(slub_free_list, cpu).list);
30790 if (debug_guardpage_minorder())
30791 slub_max_order = 0;
30792 @@ -4127,7 +4197,7 @@ static int validate_slab_node(struct kmem_cache *s,
30794 unsigned long flags;
30796 - spin_lock_irqsave(&n->list_lock, flags);
30797 + raw_spin_lock_irqsave(&n->list_lock, flags);
30799 list_for_each_entry(page, &n->partial, lru) {
30800 validate_slab_slab(s, page, map);
30801 @@ -4149,7 +4219,7 @@ static int validate_slab_node(struct kmem_cache *s,
30802 s->name, count, atomic_long_read(&n->nr_slabs));
30805 - spin_unlock_irqrestore(&n->list_lock, flags);
30806 + raw_spin_unlock_irqrestore(&n->list_lock, flags);
30810 @@ -4337,12 +4407,12 @@ static int list_locations(struct kmem_cache *s, char *buf,
30811 if (!atomic_long_read(&n->nr_slabs))
30814 - spin_lock_irqsave(&n->list_lock, flags);
30815 + raw_spin_lock_irqsave(&n->list_lock, flags);
30816 list_for_each_entry(page, &n->partial, lru)
30817 process_slab(&t, s, page, alloc, map);
30818 list_for_each_entry(page, &n->full, lru)
30819 process_slab(&t, s, page, alloc, map);
30820 - spin_unlock_irqrestore(&n->list_lock, flags);
30821 + raw_spin_unlock_irqrestore(&n->list_lock, flags);
30824 for (i = 0; i < t.count; i++) {
30825 diff --git a/mm/swap.c b/mm/swap.c
30826 index 39395fb549c0..ad16649221d7 100644
30830 #include <linux/memcontrol.h>
30831 #include <linux/gfp.h>
30832 #include <linux/uio.h>
30833 +#include <linux/locallock.h>
30834 #include <linux/hugetlb.h>
30835 #include <linux/page_idle.h>
30837 @@ -46,6 +47,9 @@ static DEFINE_PER_CPU(struct pagevec, lru_add_pvec);
30838 static DEFINE_PER_CPU(struct pagevec, lru_rotate_pvecs);
30839 static DEFINE_PER_CPU(struct pagevec, lru_deactivate_file_pvecs);
30841 +static DEFINE_LOCAL_IRQ_LOCK(rotate_lock);
30842 +DEFINE_LOCAL_IRQ_LOCK(swapvec_lock);
30845 * This path almost never happens for VM activity - pages are normally
30846 * freed via pagevecs. But it gets used by networking.
30847 @@ -481,11 +485,11 @@ void rotate_reclaimable_page(struct page *page)
30848 unsigned long flags;
30850 page_cache_get(page);
30851 - local_irq_save(flags);
30852 + local_lock_irqsave(rotate_lock, flags);
30853 pvec = this_cpu_ptr(&lru_rotate_pvecs);
30854 if (!pagevec_add(pvec, page))
30855 pagevec_move_tail(pvec);
30856 - local_irq_restore(flags);
30857 + local_unlock_irqrestore(rotate_lock, flags);
30861 @@ -536,12 +540,13 @@ static bool need_activate_page_drain(int cpu)
30862 void activate_page(struct page *page)
30864 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
30865 - struct pagevec *pvec = &get_cpu_var(activate_page_pvecs);
30866 + struct pagevec *pvec = &get_locked_var(swapvec_lock,
30867 + activate_page_pvecs);
30869 page_cache_get(page);
30870 if (!pagevec_add(pvec, page))
30871 pagevec_lru_move_fn(pvec, __activate_page, NULL);
30872 - put_cpu_var(activate_page_pvecs);
30873 + put_locked_var(swapvec_lock, activate_page_pvecs);
30877 @@ -567,7 +572,7 @@ void activate_page(struct page *page)
30879 static void __lru_cache_activate_page(struct page *page)
30881 - struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
30882 + struct pagevec *pvec = &get_locked_var(swapvec_lock, lru_add_pvec);
30886 @@ -589,7 +594,7 @@ static void __lru_cache_activate_page(struct page *page)
30890 - put_cpu_var(lru_add_pvec);
30891 + put_locked_var(swapvec_lock, lru_add_pvec);
30895 @@ -630,13 +635,13 @@ EXPORT_SYMBOL(mark_page_accessed);
30897 static void __lru_cache_add(struct page *page)
30899 - struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
30900 + struct pagevec *pvec = &get_locked_var(swapvec_lock, lru_add_pvec);
30902 page_cache_get(page);
30903 if (!pagevec_space(pvec))
30904 __pagevec_lru_add(pvec);
30905 pagevec_add(pvec, page);
30906 - put_cpu_var(lru_add_pvec);
30907 + put_locked_var(swapvec_lock, lru_add_pvec);
30911 @@ -816,9 +821,15 @@ void lru_add_drain_cpu(int cpu)
30912 unsigned long flags;
30914 /* No harm done if a racing interrupt already did this */
30915 - local_irq_save(flags);
30916 +#ifdef CONFIG_PREEMPT_RT_BASE
30917 + local_lock_irqsave_on(rotate_lock, flags, cpu);
30918 + pagevec_move_tail(pvec);
30919 + local_unlock_irqrestore_on(rotate_lock, flags, cpu);
30921 + local_lock_irqsave(rotate_lock, flags);
30922 pagevec_move_tail(pvec);
30923 - local_irq_restore(flags);
30924 + local_unlock_irqrestore(rotate_lock, flags);
30928 pvec = &per_cpu(lru_deactivate_file_pvecs, cpu);
30929 @@ -846,26 +857,47 @@ void deactivate_file_page(struct page *page)
30932 if (likely(get_page_unless_zero(page))) {
30933 - struct pagevec *pvec = &get_cpu_var(lru_deactivate_file_pvecs);
30934 + struct pagevec *pvec = &get_locked_var(swapvec_lock,
30935 + lru_deactivate_file_pvecs);
30937 if (!pagevec_add(pvec, page))
30938 pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL);
30939 - put_cpu_var(lru_deactivate_file_pvecs);
30940 + put_locked_var(swapvec_lock, lru_deactivate_file_pvecs);
30944 void lru_add_drain(void)
30946 - lru_add_drain_cpu(get_cpu());
30948 + lru_add_drain_cpu(local_lock_cpu(swapvec_lock));
30949 + local_unlock_cpu(swapvec_lock);
30953 +#ifdef CONFIG_PREEMPT_RT_BASE
30954 +static inline void remote_lru_add_drain(int cpu, struct cpumask *has_work)
30956 + local_lock_on(swapvec_lock, cpu);
30957 + lru_add_drain_cpu(cpu);
30958 + local_unlock_on(swapvec_lock, cpu);
30963 static void lru_add_drain_per_cpu(struct work_struct *dummy)
30968 static DEFINE_PER_CPU(struct work_struct, lru_add_drain_work);
30969 +static inline void remote_lru_add_drain(int cpu, struct cpumask *has_work)
30971 + struct work_struct *work = &per_cpu(lru_add_drain_work, cpu);
30973 + INIT_WORK(work, lru_add_drain_per_cpu);
30974 + schedule_work_on(cpu, work);
30975 + cpumask_set_cpu(cpu, has_work);
30979 void lru_add_drain_all(void)
30981 @@ -878,20 +910,17 @@ void lru_add_drain_all(void)
30982 cpumask_clear(&has_work);
30984 for_each_online_cpu(cpu) {
30985 - struct work_struct *work = &per_cpu(lru_add_drain_work, cpu);
30987 if (pagevec_count(&per_cpu(lru_add_pvec, cpu)) ||
30988 pagevec_count(&per_cpu(lru_rotate_pvecs, cpu)) ||
30989 pagevec_count(&per_cpu(lru_deactivate_file_pvecs, cpu)) ||
30990 - need_activate_page_drain(cpu)) {
30991 - INIT_WORK(work, lru_add_drain_per_cpu);
30992 - schedule_work_on(cpu, work);
30993 - cpumask_set_cpu(cpu, &has_work);
30995 + need_activate_page_drain(cpu))
30996 + remote_lru_add_drain(cpu, &has_work);
30999 +#ifndef CONFIG_PREEMPT_RT_BASE
31000 for_each_cpu(cpu, &has_work)
31001 flush_work(&per_cpu(lru_add_drain_work, cpu));
31005 mutex_unlock(&lock);
31006 diff --git a/mm/truncate.c b/mm/truncate.c
31007 index 76e35ad97102..5f196420020c 100644
31008 --- a/mm/truncate.c
31009 +++ b/mm/truncate.c
31010 @@ -56,8 +56,11 @@ static void clear_exceptional_entry(struct address_space *mapping,
31011 * protected by mapping->tree_lock.
31013 if (!workingset_node_shadows(node) &&
31014 - !list_empty(&node->private_list))
31015 - list_lru_del(&workingset_shadow_nodes, &node->private_list);
31016 + !list_empty(&node->private_list)) {
31017 + local_lock(workingset_shadow_lock);
31018 + list_lru_del(&__workingset_shadow_nodes, &node->private_list);
31019 + local_unlock(workingset_shadow_lock);
31021 __radix_tree_delete_node(&mapping->page_tree, node);
31023 spin_unlock_irq(&mapping->tree_lock);
31024 diff --git a/mm/vmalloc.c b/mm/vmalloc.c
31025 index 8e3c9c5a3042..68740314ad54 100644
31028 @@ -821,7 +821,7 @@ static void *new_vmap_block(unsigned int order, gfp_t gfp_mask)
31029 struct vmap_block *vb;
31030 struct vmap_area *va;
31031 unsigned long vb_idx;
31033 + int node, err, cpu;
31036 node = numa_node_id();
31037 @@ -864,11 +864,12 @@ static void *new_vmap_block(unsigned int order, gfp_t gfp_mask)
31039 radix_tree_preload_end();
31041 - vbq = &get_cpu_var(vmap_block_queue);
31042 + cpu = get_cpu_light();
31043 + vbq = this_cpu_ptr(&vmap_block_queue);
31044 spin_lock(&vbq->lock);
31045 list_add_tail_rcu(&vb->free_list, &vbq->free);
31046 spin_unlock(&vbq->lock);
31047 - put_cpu_var(vmap_block_queue);
31052 @@ -937,6 +938,7 @@ static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
31053 struct vmap_block *vb;
31054 void *vaddr = NULL;
31055 unsigned int order;
31058 BUG_ON(offset_in_page(size));
31059 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
31060 @@ -951,7 +953,8 @@ static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
31061 order = get_order(size);
31064 - vbq = &get_cpu_var(vmap_block_queue);
31065 + cpu = get_cpu_light();
31066 + vbq = this_cpu_ptr(&vmap_block_queue);
31067 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
31068 unsigned long pages_off;
31070 @@ -974,7 +977,7 @@ static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
31074 - put_cpu_var(vmap_block_queue);
31078 /* Allocate new block if nothing was found */
31079 diff --git a/mm/vmstat.c b/mm/vmstat.c
31080 index c54fd2924f25..64416fd7c209 100644
31083 @@ -226,6 +226,7 @@ void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
31087 + preempt_disable_rt();
31088 x = delta + __this_cpu_read(*p);
31090 t = __this_cpu_read(pcp->stat_threshold);
31091 @@ -235,6 +236,7 @@ void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
31094 __this_cpu_write(*p, x);
31095 + preempt_enable_rt();
31097 EXPORT_SYMBOL(__mod_zone_page_state);
31099 @@ -267,6 +269,7 @@ void __inc_zone_state(struct zone *zone, enum zone_stat_item item)
31100 s8 __percpu *p = pcp->vm_stat_diff + item;
31103 + preempt_disable_rt();
31104 v = __this_cpu_inc_return(*p);
31105 t = __this_cpu_read(pcp->stat_threshold);
31106 if (unlikely(v > t)) {
31107 @@ -275,6 +278,7 @@ void __inc_zone_state(struct zone *zone, enum zone_stat_item item)
31108 zone_page_state_add(v + overstep, zone, item);
31109 __this_cpu_write(*p, -overstep);
31111 + preempt_enable_rt();
31114 void __inc_zone_page_state(struct page *page, enum zone_stat_item item)
31115 @@ -289,6 +293,7 @@ void __dec_zone_state(struct zone *zone, enum zone_stat_item item)
31116 s8 __percpu *p = pcp->vm_stat_diff + item;
31119 + preempt_disable_rt();
31120 v = __this_cpu_dec_return(*p);
31121 t = __this_cpu_read(pcp->stat_threshold);
31122 if (unlikely(v < - t)) {
31123 @@ -297,6 +302,7 @@ void __dec_zone_state(struct zone *zone, enum zone_stat_item item)
31124 zone_page_state_add(v - overstep, zone, item);
31125 __this_cpu_write(*p, overstep);
31127 + preempt_enable_rt();
31130 void __dec_zone_page_state(struct page *page, enum zone_stat_item item)
31131 diff --git a/mm/workingset.c b/mm/workingset.c
31132 index df66f426fdcf..6db7b243fa0d 100644
31133 --- a/mm/workingset.c
31134 +++ b/mm/workingset.c
31135 @@ -264,7 +264,8 @@ void workingset_activation(struct page *page)
31136 * point where they would still be useful.
31139 -struct list_lru workingset_shadow_nodes;
31140 +struct list_lru __workingset_shadow_nodes;
31141 +DEFINE_LOCAL_IRQ_LOCK(workingset_shadow_lock);
31143 static unsigned long count_shadow_nodes(struct shrinker *shrinker,
31144 struct shrink_control *sc)
31145 @@ -274,9 +275,9 @@ static unsigned long count_shadow_nodes(struct shrinker *shrinker,
31146 unsigned long pages;
31148 /* list_lru lock nests inside IRQ-safe mapping->tree_lock */
31149 - local_irq_disable();
31150 - shadow_nodes = list_lru_shrink_count(&workingset_shadow_nodes, sc);
31151 - local_irq_enable();
31152 + local_lock_irq(workingset_shadow_lock);
31153 + shadow_nodes = list_lru_shrink_count(&__workingset_shadow_nodes, sc);
31154 + local_unlock_irq(workingset_shadow_lock);
31156 pages = node_present_pages(sc->nid);
31158 @@ -361,9 +362,9 @@ static enum lru_status shadow_lru_isolate(struct list_head *item,
31159 spin_unlock(&mapping->tree_lock);
31160 ret = LRU_REMOVED_RETRY;
31162 - local_irq_enable();
31163 + local_unlock_irq(workingset_shadow_lock);
31165 - local_irq_disable();
31166 + local_lock_irq(workingset_shadow_lock);
31167 spin_lock(lru_lock);
31170 @@ -374,10 +375,10 @@ static unsigned long scan_shadow_nodes(struct shrinker *shrinker,
31173 /* list_lru lock nests inside IRQ-safe mapping->tree_lock */
31174 - local_irq_disable();
31175 - ret = list_lru_shrink_walk(&workingset_shadow_nodes, sc,
31176 + local_lock_irq(workingset_shadow_lock);
31177 + ret = list_lru_shrink_walk(&__workingset_shadow_nodes, sc,
31178 shadow_lru_isolate, NULL);
31179 - local_irq_enable();
31180 + local_unlock_irq(workingset_shadow_lock);
31184 @@ -398,7 +399,7 @@ static int __init workingset_init(void)
31188 - ret = list_lru_init_key(&workingset_shadow_nodes, &shadow_nodes_key);
31189 + ret = list_lru_init_key(&__workingset_shadow_nodes, &shadow_nodes_key);
31192 ret = register_shrinker(&workingset_shadow_shrinker);
31193 @@ -406,7 +407,7 @@ static int __init workingset_init(void)
31197 - list_lru_destroy(&workingset_shadow_nodes);
31198 + list_lru_destroy(&__workingset_shadow_nodes);
31202 diff --git a/mm/zsmalloc.c b/mm/zsmalloc.c
31203 index c1ea19478119..529552c3716d 100644
31204 --- a/mm/zsmalloc.c
31205 +++ b/mm/zsmalloc.c
31207 #include <linux/debugfs.h>
31208 #include <linux/zsmalloc.h>
31209 #include <linux/zpool.h>
31210 +#include <linux/locallock.h>
31213 * This must be power of 2 and greater than of equal to sizeof(link_free).
31214 @@ -403,6 +404,7 @@ static unsigned int get_maxobj_per_zspage(int size, int pages_per_zspage)
31216 /* per-cpu VM mapping areas for zspage accesses that cross page boundaries */
31217 static DEFINE_PER_CPU(struct mapping_area, zs_map_area);
31218 +static DEFINE_LOCAL_IRQ_LOCK(zs_map_area_lock);
31220 static int is_first_page(struct page *page)
31222 @@ -1289,7 +1291,7 @@ void *zs_map_object(struct zs_pool *pool, unsigned long handle,
31223 class = pool->size_class[class_idx];
31224 off = obj_idx_to_offset(page, obj_idx, class->size);
31226 - area = &get_cpu_var(zs_map_area);
31227 + area = &get_locked_var(zs_map_area_lock, zs_map_area);
31229 if (off + class->size <= PAGE_SIZE) {
31230 /* this object is contained entirely within a page */
31231 @@ -1342,7 +1344,7 @@ void zs_unmap_object(struct zs_pool *pool, unsigned long handle)
31233 __zs_unmap_object(area, pages, off, class->size);
31235 - put_cpu_var(zs_map_area);
31236 + put_locked_var(zs_map_area_lock, zs_map_area);
31239 EXPORT_SYMBOL_GPL(zs_unmap_object);
31240 diff --git a/net/core/dev.c b/net/core/dev.c
31241 index 48399d8ce614..2ccc02633e97 100644
31242 --- a/net/core/dev.c
31243 +++ b/net/core/dev.c
31244 @@ -186,6 +186,7 @@ static unsigned int napi_gen_id;
31245 static DEFINE_HASHTABLE(napi_hash, 8);
31247 static seqcount_t devnet_rename_seq;
31248 +static DEFINE_MUTEX(devnet_rename_mutex);
31250 static inline void dev_base_seq_inc(struct net *net)
31252 @@ -207,14 +208,14 @@ static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
31253 static inline void rps_lock(struct softnet_data *sd)
31256 - spin_lock(&sd->input_pkt_queue.lock);
31257 + raw_spin_lock(&sd->input_pkt_queue.raw_lock);
31261 static inline void rps_unlock(struct softnet_data *sd)
31264 - spin_unlock(&sd->input_pkt_queue.lock);
31265 + raw_spin_unlock(&sd->input_pkt_queue.raw_lock);
31269 @@ -884,7 +885,8 @@ retry:
31270 strcpy(name, dev->name);
31272 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
31274 + mutex_lock(&devnet_rename_mutex);
31275 + mutex_unlock(&devnet_rename_mutex);
31279 @@ -1153,20 +1155,17 @@ int dev_change_name(struct net_device *dev, const char *newname)
31280 if (dev->flags & IFF_UP)
31283 - write_seqcount_begin(&devnet_rename_seq);
31284 + mutex_lock(&devnet_rename_mutex);
31285 + __raw_write_seqcount_begin(&devnet_rename_seq);
31287 - if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
31288 - write_seqcount_end(&devnet_rename_seq);
31291 + if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
31294 memcpy(oldname, dev->name, IFNAMSIZ);
31296 err = dev_get_valid_name(net, dev, newname);
31298 - write_seqcount_end(&devnet_rename_seq);
31304 if (oldname[0] && !strchr(oldname, '%'))
31305 netdev_info(dev, "renamed from %s\n", oldname);
31306 @@ -1179,11 +1178,12 @@ rollback:
31308 memcpy(dev->name, oldname, IFNAMSIZ);
31309 dev->name_assign_type = old_assign_type;
31310 - write_seqcount_end(&devnet_rename_seq);
31316 - write_seqcount_end(&devnet_rename_seq);
31317 + __raw_write_seqcount_end(&devnet_rename_seq);
31318 + mutex_unlock(&devnet_rename_mutex);
31320 netdev_adjacent_rename_links(dev, oldname);
31322 @@ -1204,7 +1204,8 @@ rollback:
31323 /* err >= 0 after dev_alloc_name() or stores the first errno */
31326 - write_seqcount_begin(&devnet_rename_seq);
31327 + mutex_lock(&devnet_rename_mutex);
31328 + __raw_write_seqcount_begin(&devnet_rename_seq);
31329 memcpy(dev->name, oldname, IFNAMSIZ);
31330 memcpy(oldname, newname, IFNAMSIZ);
31331 dev->name_assign_type = old_assign_type;
31332 @@ -1217,6 +1218,11 @@ rollback:
31338 + __raw_write_seqcount_end(&devnet_rename_seq);
31339 + mutex_unlock(&devnet_rename_mutex);
31344 @@ -2268,6 +2274,7 @@ static inline void __netif_reschedule(struct Qdisc *q)
31345 sd->output_queue_tailp = &q->next_sched;
31346 raise_softirq_irqoff(NET_TX_SOFTIRQ);
31347 local_irq_restore(flags);
31348 + preempt_check_resched_rt();
31351 void __netif_schedule(struct Qdisc *q)
31352 @@ -2349,6 +2356,7 @@ void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
31353 __this_cpu_write(softnet_data.completion_queue, skb);
31354 raise_softirq_irqoff(NET_TX_SOFTIRQ);
31355 local_irq_restore(flags);
31356 + preempt_check_resched_rt();
31358 EXPORT_SYMBOL(__dev_kfree_skb_irq);
31360 @@ -2906,7 +2914,11 @@ static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
31361 * This permits __QDISC___STATE_RUNNING owner to get the lock more
31362 * often and dequeue packets faster.
31364 +#ifdef CONFIG_PREEMPT_RT_FULL
31365 + contended = true;
31367 contended = qdisc_is_running(q);
31369 if (unlikely(contended))
31370 spin_lock(&q->busylock);
31372 @@ -2966,9 +2978,44 @@ static void skb_update_prio(struct sk_buff *skb)
31373 #define skb_update_prio(skb)
31376 +#ifdef CONFIG_PREEMPT_RT_FULL
31378 +static inline int xmit_rec_read(void)
31380 + return current->xmit_recursion;
31383 +static inline void xmit_rec_inc(void)
31385 + current->xmit_recursion++;
31388 +static inline void xmit_rec_dec(void)
31390 + current->xmit_recursion--;
31395 DEFINE_PER_CPU(int, xmit_recursion);
31396 EXPORT_SYMBOL(xmit_recursion);
31398 +static inline int xmit_rec_read(void)
31400 + return __this_cpu_read(xmit_recursion);
31403 +static inline void xmit_rec_inc(void)
31405 + __this_cpu_inc(xmit_recursion);
31408 +static inline void xmit_rec_dec(void)
31410 + __this_cpu_dec(xmit_recursion);
31414 #define RECURSION_LIMIT 10
31417 @@ -3161,7 +3208,7 @@ static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
31419 if (txq->xmit_lock_owner != cpu) {
31421 - if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
31422 + if (xmit_rec_read() > RECURSION_LIMIT)
31423 goto recursion_alert;
31425 skb = validate_xmit_skb(skb, dev);
31426 @@ -3171,9 +3218,9 @@ static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
31427 HARD_TX_LOCK(dev, txq, cpu);
31429 if (!netif_xmit_stopped(txq)) {
31430 - __this_cpu_inc(xmit_recursion);
31432 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
31433 - __this_cpu_dec(xmit_recursion);
31435 if (dev_xmit_complete(rc)) {
31436 HARD_TX_UNLOCK(dev, txq);
31438 @@ -3547,6 +3594,7 @@ drop:
31441 local_irq_restore(flags);
31442 + preempt_check_resched_rt();
31444 atomic_long_inc(&skb->dev->rx_dropped);
31446 @@ -3565,7 +3613,7 @@ static int netif_rx_internal(struct sk_buff *skb)
31447 struct rps_dev_flow voidflow, *rflow = &voidflow;
31450 - preempt_disable();
31451 + migrate_disable();
31454 cpu = get_rps_cpu(skb->dev, skb, &rflow);
31455 @@ -3575,13 +3623,13 @@ static int netif_rx_internal(struct sk_buff *skb)
31456 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
31459 - preempt_enable();
31460 + migrate_enable();
31464 unsigned int qtail;
31465 - ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
31467 + ret = enqueue_to_backlog(skb, get_cpu_light(), &qtail);
31472 @@ -3615,16 +3663,44 @@ int netif_rx_ni(struct sk_buff *skb)
31474 trace_netif_rx_ni_entry(skb);
31476 - preempt_disable();
31477 + local_bh_disable();
31478 err = netif_rx_internal(skb);
31479 - if (local_softirq_pending())
31481 - preempt_enable();
31482 + local_bh_enable();
31486 EXPORT_SYMBOL(netif_rx_ni);
31488 +#ifdef CONFIG_PREEMPT_RT_FULL
31490 + * RT runs ksoftirqd as a real time thread and the root_lock is a
31491 + * "sleeping spinlock". If the trylock fails then we can go into an
31492 + * infinite loop when ksoftirqd preempted the task which actually
31493 + * holds the lock, because we requeue q and raise NET_TX softirq
31494 + * causing ksoftirqd to loop forever.
31496 + * It's safe to use spin_lock on RT here as softirqs run in thread
31497 + * context and cannot deadlock against the thread which is holding
31500 + * On !RT the trylock might fail, but there we bail out from the
31501 + * softirq loop after 10 attempts which we can't do on RT. And the
31502 + * task holding root_lock cannot be preempted, so the only downside of
31503 + * that trylock is that we need 10 loops to decide that we should have
31504 + * given up in the first one :)
31506 +static inline int take_root_lock(spinlock_t *lock)
31512 +static inline int take_root_lock(spinlock_t *lock)
31514 + return spin_trylock(lock);
31518 static void net_tx_action(struct softirq_action *h)
31520 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
31521 @@ -3666,7 +3742,7 @@ static void net_tx_action(struct softirq_action *h)
31522 head = head->next_sched;
31524 root_lock = qdisc_lock(q);
31525 - if (spin_trylock(root_lock)) {
31526 + if (take_root_lock(root_lock)) {
31527 smp_mb__before_atomic();
31528 clear_bit(__QDISC_STATE_SCHED,
31530 @@ -4088,7 +4164,7 @@ static void flush_backlog(void *arg)
31531 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
31532 if (skb->dev == dev) {
31533 __skb_unlink(skb, &sd->input_pkt_queue);
31535 + __skb_queue_tail(&sd->tofree_queue, skb);
31536 input_queue_head_incr(sd);
31539 @@ -4097,10 +4173,13 @@ static void flush_backlog(void *arg)
31540 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
31541 if (skb->dev == dev) {
31542 __skb_unlink(skb, &sd->process_queue);
31544 + __skb_queue_tail(&sd->tofree_queue, skb);
31545 input_queue_head_incr(sd);
31549 + if (!skb_queue_empty(&sd->tofree_queue))
31550 + raise_softirq_irqoff(NET_RX_SOFTIRQ);
31553 static int napi_gro_complete(struct sk_buff *skb)
31554 @@ -4557,6 +4636,7 @@ static void net_rps_action_and_irq_enable(struct softnet_data *sd)
31555 sd->rps_ipi_list = NULL;
31557 local_irq_enable();
31558 + preempt_check_resched_rt();
31560 /* Send pending IPI's to kick RPS processing on remote cpus. */
31562 @@ -4570,6 +4650,7 @@ static void net_rps_action_and_irq_enable(struct softnet_data *sd)
31565 local_irq_enable();
31566 + preempt_check_resched_rt();
31569 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
31570 @@ -4651,9 +4732,11 @@ void __napi_schedule(struct napi_struct *n)
31571 local_irq_save(flags);
31572 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
31573 local_irq_restore(flags);
31574 + preempt_check_resched_rt();
31576 EXPORT_SYMBOL(__napi_schedule);
31578 +#ifndef CONFIG_PREEMPT_RT_FULL
31580 * __napi_schedule_irqoff - schedule for receive
31581 * @n: entry to schedule
31582 @@ -4665,6 +4748,7 @@ void __napi_schedule_irqoff(struct napi_struct *n)
31583 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
31585 EXPORT_SYMBOL(__napi_schedule_irqoff);
31588 void __napi_complete(struct napi_struct *n)
31590 @@ -4891,13 +4975,21 @@ static void net_rx_action(struct softirq_action *h)
31591 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
31592 unsigned long time_limit = jiffies + 2;
31593 int budget = netdev_budget;
31594 + struct sk_buff_head tofree_q;
31595 + struct sk_buff *skb;
31599 + __skb_queue_head_init(&tofree_q);
31601 local_irq_disable();
31602 + skb_queue_splice_init(&sd->tofree_queue, &tofree_q);
31603 list_splice_init(&sd->poll_list, &list);
31604 local_irq_enable();
31606 + while ((skb = __skb_dequeue(&tofree_q)))
31610 struct napi_struct *n;
31612 @@ -4927,7 +5019,7 @@ static void net_rx_action(struct softirq_action *h)
31613 list_splice_tail(&repoll, &list);
31614 list_splice(&list, &sd->poll_list);
31615 if (!list_empty(&sd->poll_list))
31616 - __raise_softirq_irqoff(NET_RX_SOFTIRQ);
31617 + __raise_softirq_irqoff_ksoft(NET_RX_SOFTIRQ);
31619 net_rps_action_and_irq_enable(sd);
31621 @@ -7266,7 +7358,7 @@ EXPORT_SYMBOL(free_netdev);
31622 void synchronize_net(void)
31625 - if (rtnl_is_locked())
31626 + if (rtnl_is_locked() && !IS_ENABLED(CONFIG_PREEMPT_RT_FULL))
31627 synchronize_rcu_expedited();
31630 @@ -7507,16 +7599,20 @@ static int dev_cpu_callback(struct notifier_block *nfb,
31632 raise_softirq_irqoff(NET_TX_SOFTIRQ);
31633 local_irq_enable();
31634 + preempt_check_resched_rt();
31636 /* Process offline CPU's input_pkt_queue */
31637 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
31639 input_queue_head_incr(oldsd);
31641 - while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
31642 + while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
31644 input_queue_head_incr(oldsd);
31646 + while ((skb = __skb_dequeue(&oldsd->tofree_queue))) {
31652 @@ -7818,8 +7914,9 @@ static int __init net_dev_init(void)
31653 for_each_possible_cpu(i) {
31654 struct softnet_data *sd = &per_cpu(softnet_data, i);
31656 - skb_queue_head_init(&sd->input_pkt_queue);
31657 - skb_queue_head_init(&sd->process_queue);
31658 + skb_queue_head_init_raw(&sd->input_pkt_queue);
31659 + skb_queue_head_init_raw(&sd->process_queue);
31660 + skb_queue_head_init_raw(&sd->tofree_queue);
31661 INIT_LIST_HEAD(&sd->poll_list);
31662 sd->output_queue_tailp = &sd->output_queue;
31664 diff --git a/net/core/skbuff.c b/net/core/skbuff.c
31665 index 73dfd7729bc9..62f62810cddd 100644
31666 --- a/net/core/skbuff.c
31667 +++ b/net/core/skbuff.c
31669 #include <linux/errqueue.h>
31670 #include <linux/prefetch.h>
31671 #include <linux/if_vlan.h>
31672 +#include <linux/locallock.h>
31674 #include <net/protocol.h>
31675 #include <net/dst.h>
31676 @@ -351,6 +352,8 @@ EXPORT_SYMBOL(build_skb);
31678 static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache);
31679 static DEFINE_PER_CPU(struct page_frag_cache, napi_alloc_cache);
31680 +static DEFINE_LOCAL_IRQ_LOCK(netdev_alloc_lock);
31681 +static DEFINE_LOCAL_IRQ_LOCK(napi_alloc_cache_lock);
31683 static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
31685 @@ -358,10 +361,10 @@ static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
31686 unsigned long flags;
31689 - local_irq_save(flags);
31690 + local_lock_irqsave(netdev_alloc_lock, flags);
31691 nc = this_cpu_ptr(&netdev_alloc_cache);
31692 data = __alloc_page_frag(nc, fragsz, gfp_mask);
31693 - local_irq_restore(flags);
31694 + local_unlock_irqrestore(netdev_alloc_lock, flags);
31698 @@ -380,9 +383,13 @@ EXPORT_SYMBOL(netdev_alloc_frag);
31700 static void *__napi_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
31702 - struct page_frag_cache *nc = this_cpu_ptr(&napi_alloc_cache);
31703 + struct page_frag_cache *nc;
31706 - return __alloc_page_frag(nc, fragsz, gfp_mask);
31707 + nc = &get_locked_var(napi_alloc_cache_lock, napi_alloc_cache);
31708 + data = __alloc_page_frag(nc, fragsz, gfp_mask);
31709 + put_locked_var(napi_alloc_cache_lock, napi_alloc_cache);
31713 void *napi_alloc_frag(unsigned int fragsz)
31714 @@ -429,13 +436,13 @@ struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len,
31715 if (sk_memalloc_socks())
31716 gfp_mask |= __GFP_MEMALLOC;
31718 - local_irq_save(flags);
31719 + local_lock_irqsave(netdev_alloc_lock, flags);
31721 nc = this_cpu_ptr(&netdev_alloc_cache);
31722 data = __alloc_page_frag(nc, len, gfp_mask);
31723 pfmemalloc = nc->pfmemalloc;
31725 - local_irq_restore(flags);
31726 + local_unlock_irqrestore(netdev_alloc_lock, flags);
31728 if (unlikely(!data))
31730 @@ -476,9 +483,10 @@ EXPORT_SYMBOL(__netdev_alloc_skb);
31731 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len,
31734 - struct page_frag_cache *nc = this_cpu_ptr(&napi_alloc_cache);
31735 + struct page_frag_cache *nc;
31736 struct sk_buff *skb;
31740 len += NET_SKB_PAD + NET_IP_ALIGN;
31742 @@ -496,7 +504,11 @@ struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len,
31743 if (sk_memalloc_socks())
31744 gfp_mask |= __GFP_MEMALLOC;
31746 + nc = &get_locked_var(napi_alloc_cache_lock, napi_alloc_cache);
31747 data = __alloc_page_frag(nc, len, gfp_mask);
31748 + pfmemalloc = nc->pfmemalloc;
31749 + put_locked_var(napi_alloc_cache_lock, napi_alloc_cache);
31751 if (unlikely(!data))
31754 @@ -507,7 +519,7 @@ struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len,
31757 /* use OR instead of assignment to avoid clearing of bits in mask */
31758 - if (nc->pfmemalloc)
31760 skb->pfmemalloc = 1;
31761 skb->head_frag = 1;
31763 diff --git a/net/core/sock.c b/net/core/sock.c
31764 index 9c708a5fb751..823377a7e63a 100644
31765 --- a/net/core/sock.c
31766 +++ b/net/core/sock.c
31767 @@ -2447,12 +2447,11 @@ void lock_sock_nested(struct sock *sk, int subclass)
31768 if (sk->sk_lock.owned)
31770 sk->sk_lock.owned = 1;
31771 - spin_unlock(&sk->sk_lock.slock);
31772 + spin_unlock_bh(&sk->sk_lock.slock);
31774 * The sk_lock has mutex_lock() semantics here:
31776 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
31777 - local_bh_enable();
31779 EXPORT_SYMBOL(lock_sock_nested);
31781 diff --git a/net/ipv4/icmp.c b/net/ipv4/icmp.c
31782 index 36e26977c908..ff2593269089 100644
31783 --- a/net/ipv4/icmp.c
31784 +++ b/net/ipv4/icmp.c
31786 #include <linux/jiffies.h>
31787 #include <linux/kernel.h>
31788 #include <linux/fcntl.h>
31789 +#include <linux/sysrq.h>
31790 #include <linux/socket.h>
31791 #include <linux/in.h>
31792 #include <linux/inet.h>
31794 #include <linux/string.h>
31795 #include <linux/netfilter_ipv4.h>
31796 #include <linux/slab.h>
31797 +#include <linux/locallock.h>
31798 #include <net/snmp.h>
31799 #include <net/ip.h>
31800 #include <net/route.h>
31801 @@ -204,6 +206,8 @@ static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
31803 * On SMP we have one ICMP socket per-cpu.
31805 +static DEFINE_LOCAL_IRQ_LOCK(icmp_sk_lock);
31807 static struct sock *icmp_sk(struct net *net)
31809 return *this_cpu_ptr(net->ipv4.icmp_sk);
31810 @@ -215,12 +219,14 @@ static inline struct sock *icmp_xmit_lock(struct net *net)
31812 local_bh_disable();
31814 + local_lock(icmp_sk_lock);
31817 if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
31818 /* This can happen if the output path signals a
31819 * dst_link_failure() for an outgoing ICMP packet.
31821 + local_unlock(icmp_sk_lock);
31825 @@ -230,6 +236,7 @@ static inline struct sock *icmp_xmit_lock(struct net *net)
31826 static inline void icmp_xmit_unlock(struct sock *sk)
31828 spin_unlock_bh(&sk->sk_lock.slock);
31829 + local_unlock(icmp_sk_lock);
31832 int sysctl_icmp_msgs_per_sec __read_mostly = 1000;
31833 @@ -358,6 +365,7 @@ static void icmp_push_reply(struct icmp_bxm *icmp_param,
31835 struct sk_buff *skb;
31837 + local_lock(icmp_sk_lock);
31838 sk = icmp_sk(dev_net((*rt)->dst.dev));
31839 if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
31840 icmp_param->data_len+icmp_param->head_len,
31841 @@ -380,6 +388,7 @@ static void icmp_push_reply(struct icmp_bxm *icmp_param,
31842 skb->ip_summed = CHECKSUM_NONE;
31843 ip_push_pending_frames(sk, fl4);
31845 + local_unlock(icmp_sk_lock);
31849 @@ -891,6 +900,30 @@ static bool icmp_redirect(struct sk_buff *skb)
31853 + * 32bit and 64bit have different timestamp length, so we check for
31854 + * the cookie at offset 20 and verify it is repeated at offset 50
31856 +#define CO_POS0 20
31857 +#define CO_POS1 50
31858 +#define CO_SIZE sizeof(int)
31859 +#define ICMP_SYSRQ_SIZE 57
31862 + * We got a ICMP_SYSRQ_SIZE sized ping request. Check for the cookie
31863 + * pattern and if it matches send the next byte as a trigger to sysrq.
31865 +static void icmp_check_sysrq(struct net *net, struct sk_buff *skb)
31867 + int cookie = htonl(net->ipv4.sysctl_icmp_echo_sysrq);
31868 + char *p = skb->data;
31870 + if (!memcmp(&cookie, p + CO_POS0, CO_SIZE) &&
31871 + !memcmp(&cookie, p + CO_POS1, CO_SIZE) &&
31872 + p[CO_POS0 + CO_SIZE] == p[CO_POS1 + CO_SIZE])
31873 + handle_sysrq(p[CO_POS0 + CO_SIZE]);
31877 * Handle ICMP_ECHO ("ping") requests.
31879 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
31880 @@ -917,6 +950,11 @@ static bool icmp_echo(struct sk_buff *skb)
31881 icmp_param.data_len = skb->len;
31882 icmp_param.head_len = sizeof(struct icmphdr);
31883 icmp_reply(&icmp_param, skb);
31885 + if (skb->len == ICMP_SYSRQ_SIZE &&
31886 + net->ipv4.sysctl_icmp_echo_sysrq) {
31887 + icmp_check_sysrq(net, skb);
31890 /* should there be an ICMP stat for ignored echos? */
31892 diff --git a/net/ipv4/sysctl_net_ipv4.c b/net/ipv4/sysctl_net_ipv4.c
31893 index 70fb352e317f..1bcd436709a4 100644
31894 --- a/net/ipv4/sysctl_net_ipv4.c
31895 +++ b/net/ipv4/sysctl_net_ipv4.c
31896 @@ -818,6 +818,13 @@ static struct ctl_table ipv4_net_table[] = {
31897 .proc_handler = proc_dointvec
31900 + .procname = "icmp_echo_sysrq",
31901 + .data = &init_net.ipv4.sysctl_icmp_echo_sysrq,
31902 + .maxlen = sizeof(int),
31904 + .proc_handler = proc_dointvec
31907 .procname = "icmp_ignore_bogus_error_responses",
31908 .data = &init_net.ipv4.sysctl_icmp_ignore_bogus_error_responses,
31909 .maxlen = sizeof(int),
31910 diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c
31911 index 198fc2314c82..6d84e7c8c8a9 100644
31912 --- a/net/ipv4/tcp_ipv4.c
31913 +++ b/net/ipv4/tcp_ipv4.c
31915 #include <linux/init.h>
31916 #include <linux/times.h>
31917 #include <linux/slab.h>
31918 +#include <linux/locallock.h>
31920 #include <net/net_namespace.h>
31921 #include <net/icmp.h>
31922 @@ -570,6 +571,7 @@ void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb)
31924 EXPORT_SYMBOL(tcp_v4_send_check);
31926 +static DEFINE_LOCAL_IRQ_LOCK(tcp_sk_lock);
31928 * This routine will send an RST to the other tcp.
31930 @@ -691,10 +693,13 @@ static void tcp_v4_send_reset(const struct sock *sk, struct sk_buff *skb)
31931 arg.bound_dev_if = sk->sk_bound_dev_if;
31933 arg.tos = ip_hdr(skb)->tos;
31935 + local_lock(tcp_sk_lock);
31936 ip_send_unicast_reply(*this_cpu_ptr(net->ipv4.tcp_sk),
31937 skb, &TCP_SKB_CB(skb)->header.h4.opt,
31938 ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
31939 &arg, arg.iov[0].iov_len);
31940 + local_unlock(tcp_sk_lock);
31942 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
31943 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
31944 @@ -776,10 +781,12 @@ static void tcp_v4_send_ack(struct net *net,
31946 arg.bound_dev_if = oif;
31948 + local_lock(tcp_sk_lock);
31949 ip_send_unicast_reply(*this_cpu_ptr(net->ipv4.tcp_sk),
31950 skb, &TCP_SKB_CB(skb)->header.h4.opt,
31951 ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
31952 &arg, arg.iov[0].iov_len);
31953 + local_unlock(tcp_sk_lock);
31955 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
31957 diff --git a/net/mac80211/rx.c b/net/mac80211/rx.c
31958 index 9f0915f72702..28d234c03cb5 100644
31959 --- a/net/mac80211/rx.c
31960 +++ b/net/mac80211/rx.c
31961 @@ -3601,7 +3601,7 @@ void ieee80211_rx_napi(struct ieee80211_hw *hw, struct sk_buff *skb,
31962 struct ieee80211_supported_band *sband;
31963 struct ieee80211_rx_status *status = IEEE80211_SKB_RXCB(skb);
31965 - WARN_ON_ONCE(softirq_count() == 0);
31966 + WARN_ON_ONCE_NONRT(softirq_count() == 0);
31968 if (WARN_ON(status->band >= IEEE80211_NUM_BANDS))
31970 diff --git a/net/netfilter/core.c b/net/netfilter/core.c
31971 index f39276d1c2d7..10880c89d62f 100644
31972 --- a/net/netfilter/core.c
31973 +++ b/net/netfilter/core.c
31974 @@ -22,11 +22,17 @@
31975 #include <linux/proc_fs.h>
31976 #include <linux/mutex.h>
31977 #include <linux/slab.h>
31978 +#include <linux/locallock.h>
31979 #include <net/net_namespace.h>
31980 #include <net/sock.h>
31982 #include "nf_internals.h"
31984 +#ifdef CONFIG_PREEMPT_RT_BASE
31985 +DEFINE_LOCAL_IRQ_LOCK(xt_write_lock);
31986 +EXPORT_PER_CPU_SYMBOL(xt_write_lock);
31989 static DEFINE_MUTEX(afinfo_mutex);
31991 const struct nf_afinfo __rcu *nf_afinfo[NFPROTO_NUMPROTO] __read_mostly;
31992 diff --git a/net/packet/af_packet.c b/net/packet/af_packet.c
31993 index f8d6a0ca9c03..f6be45510cf6 100644
31994 --- a/net/packet/af_packet.c
31995 +++ b/net/packet/af_packet.c
31997 #include <linux/if_packet.h>
31998 #include <linux/wireless.h>
31999 #include <linux/kernel.h>
32000 +#include <linux/delay.h>
32001 #include <linux/kmod.h>
32002 #include <linux/slab.h>
32003 #include <linux/vmalloc.h>
32004 @@ -694,7 +695,7 @@ static void prb_retire_rx_blk_timer_expired(unsigned long data)
32005 if (BLOCK_NUM_PKTS(pbd)) {
32006 while (atomic_read(&pkc->blk_fill_in_prog)) {
32007 /* Waiting for skb_copy_bits to finish... */
32013 @@ -956,7 +957,7 @@ static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
32014 if (!(status & TP_STATUS_BLK_TMO)) {
32015 while (atomic_read(&pkc->blk_fill_in_prog)) {
32016 /* Waiting for skb_copy_bits to finish... */
32021 prb_close_block(pkc, pbd, po, status);
32022 diff --git a/net/rds/ib_rdma.c b/net/rds/ib_rdma.c
32023 index a2340748ec86..19123a97b354 100644
32024 --- a/net/rds/ib_rdma.c
32025 +++ b/net/rds/ib_rdma.c
32027 #include <linux/slab.h>
32028 #include <linux/rculist.h>
32029 #include <linux/llist.h>
32030 +#include <linux/delay.h>
32034 @@ -313,7 +314,7 @@ static inline void wait_clean_list_grace(void)
32035 for_each_online_cpu(cpu) {
32036 flag = &per_cpu(clean_list_grace, cpu);
32037 while (test_bit(CLEAN_LIST_BUSY_BIT, flag))
32043 diff --git a/net/sched/sch_generic.c b/net/sched/sch_generic.c
32044 index aa4725038f94..00b81cab28f3 100644
32045 --- a/net/sched/sch_generic.c
32046 +++ b/net/sched/sch_generic.c
32047 @@ -893,7 +893,7 @@ void dev_deactivate_many(struct list_head *head)
32048 /* Wait for outstanding qdisc_run calls. */
32049 list_for_each_entry(dev, head, close_list)
32050 while (some_qdisc_is_busy(dev))
32055 void dev_deactivate(struct net_device *dev)
32056 diff --git a/net/sunrpc/svc_xprt.c b/net/sunrpc/svc_xprt.c
32057 index a6cbb2104667..5b69bb580617 100644
32058 --- a/net/sunrpc/svc_xprt.c
32059 +++ b/net/sunrpc/svc_xprt.c
32060 @@ -340,7 +340,7 @@ void svc_xprt_do_enqueue(struct svc_xprt *xprt)
32065 + cpu = get_cpu_light();
32066 pool = svc_pool_for_cpu(xprt->xpt_server, cpu);
32068 atomic_long_inc(&pool->sp_stats.packets);
32069 @@ -376,7 +376,7 @@ redo_search:
32071 atomic_long_inc(&pool->sp_stats.threads_woken);
32072 wake_up_process(rqstp->rq_task);
32078 @@ -397,7 +397,7 @@ redo_search:
32085 trace_svc_xprt_do_enqueue(xprt, rqstp);
32087 diff --git a/scripts/mkcompile_h b/scripts/mkcompile_h
32088 index 6fdc97ef6023..523e0420d7f0 100755
32089 --- a/scripts/mkcompile_h
32090 +++ b/scripts/mkcompile_h
32091 @@ -4,7 +4,8 @@ TARGET=$1
32099 vecho() { [ "${quiet}" = "silent_" ] || echo "$@" ; }
32101 @@ -57,6 +58,7 @@ UTS_VERSION="#$VERSION"
32103 if [ -n "$SMP" ] ; then CONFIG_FLAGS="SMP"; fi
32104 if [ -n "$PREEMPT" ] ; then CONFIG_FLAGS="$CONFIG_FLAGS PREEMPT"; fi
32105 +if [ -n "$RT" ] ; then CONFIG_FLAGS="$CONFIG_FLAGS RT"; fi
32106 UTS_VERSION="$UTS_VERSION $CONFIG_FLAGS $TIMESTAMP"
32108 # Truncate to maximum length
32109 diff --git a/sound/core/pcm_native.c b/sound/core/pcm_native.c
32110 index 4ba64fd49759..34e50186885d 100644
32111 --- a/sound/core/pcm_native.c
32112 +++ b/sound/core/pcm_native.c
32113 @@ -135,7 +135,7 @@ EXPORT_SYMBOL_GPL(snd_pcm_stream_unlock);
32114 void snd_pcm_stream_lock_irq(struct snd_pcm_substream *substream)
32116 if (!substream->pcm->nonatomic)
32117 - local_irq_disable();
32118 + local_irq_disable_nort();
32119 snd_pcm_stream_lock(substream);
32121 EXPORT_SYMBOL_GPL(snd_pcm_stream_lock_irq);
32122 @@ -150,7 +150,7 @@ void snd_pcm_stream_unlock_irq(struct snd_pcm_substream *substream)
32124 snd_pcm_stream_unlock(substream);
32125 if (!substream->pcm->nonatomic)
32126 - local_irq_enable();
32127 + local_irq_enable_nort();
32129 EXPORT_SYMBOL_GPL(snd_pcm_stream_unlock_irq);
32131 @@ -158,7 +158,7 @@ unsigned long _snd_pcm_stream_lock_irqsave(struct snd_pcm_substream *substream)
32133 unsigned long flags = 0;
32134 if (!substream->pcm->nonatomic)
32135 - local_irq_save(flags);
32136 + local_irq_save_nort(flags);
32137 snd_pcm_stream_lock(substream);
32140 @@ -176,7 +176,7 @@ void snd_pcm_stream_unlock_irqrestore(struct snd_pcm_substream *substream,
32142 snd_pcm_stream_unlock(substream);
32143 if (!substream->pcm->nonatomic)
32144 - local_irq_restore(flags);
32145 + local_irq_restore_nort(flags);
32147 EXPORT_SYMBOL_GPL(snd_pcm_stream_unlock_irqrestore);
32149 diff --git a/virt/kvm/async_pf.c b/virt/kvm/async_pf.c
32150 index 4f70d12e392d..9378d0919ed8 100644
32151 --- a/virt/kvm/async_pf.c
32152 +++ b/virt/kvm/async_pf.c
32153 @@ -98,8 +98,8 @@ static void async_pf_execute(struct work_struct *work)
32154 * This memory barrier pairs with prepare_to_wait's set_current_state()
32157 - if (waitqueue_active(&vcpu->wq))
32158 - wake_up_interruptible(&vcpu->wq);
32159 + if (swait_active(&vcpu->wq))
32160 + swake_up(&vcpu->wq);
32163 kvm_put_kvm(vcpu->kvm);
32164 diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
32165 index cb092bd9965b..133e7fa2ec8b 100644
32166 --- a/virt/kvm/kvm_main.c
32167 +++ b/virt/kvm/kvm_main.c
32168 @@ -228,8 +228,7 @@ int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
32170 vcpu->vcpu_id = id;
32172 - vcpu->halt_poll_ns = 0;
32173 - init_waitqueue_head(&vcpu->wq);
32174 + init_swait_queue_head(&vcpu->wq);
32175 kvm_async_pf_vcpu_init(vcpu);
32177 vcpu->pre_pcpu = -1;
32178 @@ -2008,7 +2007,7 @@ static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
32179 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
32181 ktime_t start, cur;
32182 - DEFINE_WAIT(wait);
32183 + DECLARE_SWAITQUEUE(wait);
32184 bool waited = false;
32187 @@ -2033,7 +2032,7 @@ void kvm_vcpu_block(struct kvm_vcpu *vcpu)
32188 kvm_arch_vcpu_blocking(vcpu);
32191 - prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
32192 + prepare_to_swait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
32194 if (kvm_vcpu_check_block(vcpu) < 0)
32196 @@ -2042,7 +2041,7 @@ void kvm_vcpu_block(struct kvm_vcpu *vcpu)
32200 - finish_wait(&vcpu->wq, &wait);
32201 + finish_swait(&vcpu->wq, &wait);
32204 kvm_arch_vcpu_unblocking(vcpu);
32205 @@ -2074,11 +2073,11 @@ void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
32208 int cpu = vcpu->cpu;
32209 - wait_queue_head_t *wqp;
32210 + struct swait_queue_head *wqp;
32212 wqp = kvm_arch_vcpu_wq(vcpu);
32213 - if (waitqueue_active(wqp)) {
32214 - wake_up_interruptible(wqp);
32215 + if (swait_active(wqp)) {
32217 ++vcpu->stat.halt_wakeup;
32220 @@ -2179,7 +2178,7 @@ void kvm_vcpu_on_spin(struct kvm_vcpu *me)
32224 - if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
32225 + if (swait_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
32227 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))