2 * SMP initialisation and IPI support
3 * Based on arch/arm/kernel/smp.c
5 * Copyright (C) 2012 ARM Ltd.
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
16 * You should have received a copy of the GNU General Public License
17 * along with this program. If not, see <http://www.gnu.org/licenses/>.
20 #include <linux/acpi.h>
21 #include <linux/delay.h>
22 #include <linux/init.h>
23 #include <linux/spinlock.h>
24 #include <linux/sched.h>
25 #include <linux/interrupt.h>
26 #include <linux/cache.h>
27 #include <linux/profile.h>
28 #include <linux/errno.h>
30 #include <linux/err.h>
31 #include <linux/cpu.h>
32 #include <linux/smp.h>
33 #include <linux/seq_file.h>
34 #include <linux/irq.h>
35 #include <linux/percpu.h>
36 #include <linux/clockchips.h>
37 #include <linux/completion.h>
39 #include <linux/irq_work.h>
41 #include <asm/alternative.h>
42 #include <asm/atomic.h>
43 #include <asm/cacheflush.h>
45 #include <asm/cputype.h>
46 #include <asm/cpu_ops.h>
47 #include <asm/mmu_context.h>
48 #include <asm/pgtable.h>
49 #include <asm/pgalloc.h>
50 #include <asm/processor.h>
51 #include <asm/smp_plat.h>
52 #include <asm/sections.h>
53 #include <asm/tlbflush.h>
54 #include <asm/ptrace.h>
56 #define CREATE_TRACE_POINTS
57 #include <trace/events/ipi.h>
60 * as from 2.5, kernels no longer have an init_tasks structure
61 * so we need some other way of telling a new secondary core
62 * where to place its SVC stack
64 struct secondary_data secondary_data
;
75 * Boot a secondary CPU, and assign it the specified idle task.
76 * This also gives us the initial stack to use for this CPU.
78 static int boot_secondary(unsigned int cpu
, struct task_struct
*idle
)
80 if (cpu_ops
[cpu
]->cpu_boot
)
81 return cpu_ops
[cpu
]->cpu_boot(cpu
);
86 static DECLARE_COMPLETION(cpu_running
);
88 int __cpu_up(unsigned int cpu
, struct task_struct
*idle
)
93 * We need to tell the secondary core where to find its stack and the
96 secondary_data
.stack
= task_stack_page(idle
) + THREAD_START_SP
;
97 __flush_dcache_area(&secondary_data
, sizeof(secondary_data
));
100 * Now bring the CPU into our world.
102 ret
= boot_secondary(cpu
, idle
);
105 * CPU was successfully started, wait for it to come online or
108 wait_for_completion_timeout(&cpu_running
,
109 msecs_to_jiffies(1000));
111 if (!cpu_online(cpu
)) {
112 pr_crit("CPU%u: failed to come online\n", cpu
);
116 pr_err("CPU%u: failed to boot: %d\n", cpu
, ret
);
119 secondary_data
.stack
= NULL
;
124 static void smp_store_cpu_info(unsigned int cpuid
)
126 store_cpu_topology(cpuid
);
130 * This is the secondary CPU boot entry. We're using this CPUs
131 * idle thread stack, but a set of temporary page tables.
133 asmlinkage
void secondary_start_kernel(void)
135 struct mm_struct
*mm
= &init_mm
;
136 unsigned int cpu
= smp_processor_id();
139 * All kernel threads share the same mm context; grab a
140 * reference and switch to it.
142 atomic_inc(&mm
->mm_count
);
143 current
->active_mm
= mm
;
144 cpumask_set_cpu(cpu
, mm_cpumask(mm
));
146 set_my_cpu_offset(per_cpu_offset(smp_processor_id()));
147 printk("CPU%u: Booted secondary processor\n", cpu
);
150 * TTBR0 is only used for the identity mapping at this stage. Make it
151 * point to zero page to avoid speculatively fetching new entries.
153 cpu_set_reserved_ttbr0();
155 cpu_set_default_tcr_t0sz();
158 trace_hardirqs_off();
160 if (cpu_ops
[cpu
]->cpu_postboot
)
161 cpu_ops
[cpu
]->cpu_postboot();
164 * Log the CPU info before it is marked online and might get read.
169 * Enable GIC and timers.
171 notify_cpu_starting(cpu
);
173 smp_store_cpu_info(cpu
);
176 * OK, now it's safe to let the boot CPU continue. Wait for
177 * the CPU migration code to notice that the CPU is online
178 * before we continue.
180 set_cpu_online(cpu
, true);
181 complete(&cpu_running
);
185 local_async_enable();
188 * OK, it's off to the idle thread for us
190 cpu_startup_entry(CPUHP_ONLINE
);
193 #ifdef CONFIG_HOTPLUG_CPU
194 static int op_cpu_disable(unsigned int cpu
)
197 * If we don't have a cpu_die method, abort before we reach the point
198 * of no return. CPU0 may not have an cpu_ops, so test for it.
200 if (!cpu_ops
[cpu
] || !cpu_ops
[cpu
]->cpu_die
)
204 * We may need to abort a hot unplug for some other mechanism-specific
207 if (cpu_ops
[cpu
]->cpu_disable
)
208 return cpu_ops
[cpu
]->cpu_disable(cpu
);
214 * __cpu_disable runs on the processor to be shutdown.
216 int __cpu_disable(void)
218 unsigned int cpu
= smp_processor_id();
221 ret
= op_cpu_disable(cpu
);
226 * Take this CPU offline. Once we clear this, we can't return,
227 * and we must not schedule until we're ready to give up the cpu.
229 set_cpu_online(cpu
, false);
232 * OK - migrate IRQs away from this CPU
237 * Remove this CPU from the vm mask set of all processes.
239 clear_tasks_mm_cpumask(cpu
);
244 static int op_cpu_kill(unsigned int cpu
)
247 * If we have no means of synchronising with the dying CPU, then assume
248 * that it is really dead. We can only wait for an arbitrary length of
249 * time and hope that it's dead, so let's skip the wait and just hope.
251 if (!cpu_ops
[cpu
]->cpu_kill
)
254 return cpu_ops
[cpu
]->cpu_kill(cpu
);
258 * called on the thread which is asking for a CPU to be shutdown -
259 * waits until shutdown has completed, or it is timed out.
261 void __cpu_die(unsigned int cpu
)
263 if (!cpu_wait_death(cpu
, 5)) {
264 pr_crit("CPU%u: cpu didn't die\n", cpu
);
267 pr_notice("CPU%u: shutdown\n", cpu
);
270 * Now that the dying CPU is beyond the point of no return w.r.t.
271 * in-kernel synchronisation, try to get the firwmare to help us to
272 * verify that it has really left the kernel before we consider
273 * clobbering anything it might still be using.
275 if (!op_cpu_kill(cpu
))
276 pr_warn("CPU%d may not have shut down cleanly\n", cpu
);
280 * Called from the idle thread for the CPU which has been shutdown.
282 * Note that we disable IRQs here, but do not re-enable them
283 * before returning to the caller. This is also the behaviour
284 * of the other hotplug-cpu capable cores, so presumably coming
285 * out of idle fixes this.
289 unsigned int cpu
= smp_processor_id();
295 /* Tell __cpu_die() that this CPU is now safe to dispose of */
296 (void)cpu_report_death();
299 * Actually shutdown the CPU. This must never fail. The specific hotplug
300 * mechanism must perform all required cache maintenance to ensure that
301 * no dirty lines are lost in the process of shutting down the CPU.
303 cpu_ops
[cpu
]->cpu_die(cpu
);
309 void __init
smp_cpus_done(unsigned int max_cpus
)
311 pr_info("SMP: Total of %d processors activated.\n", num_online_cpus());
312 do_post_cpus_up_work();
315 void __init
smp_prepare_boot_cpu(void)
317 set_my_cpu_offset(per_cpu_offset(smp_processor_id()));
320 static u64 __init
of_get_cpu_mpidr(struct device_node
*dn
)
326 * A cpu node with missing "reg" property is
327 * considered invalid to build a cpu_logical_map
330 cell
= of_get_property(dn
, "reg", NULL
);
332 pr_err("%s: missing reg property\n", dn
->full_name
);
336 hwid
= of_read_number(cell
, of_n_addr_cells(dn
));
338 * Non affinity bits must be set to 0 in the DT
340 if (hwid
& ~MPIDR_HWID_BITMASK
) {
341 pr_err("%s: invalid reg property\n", dn
->full_name
);
348 * Duplicate MPIDRs are a recipe for disaster. Scan all initialized
349 * entries and check for duplicates. If any is found just ignore the
350 * cpu. cpu_logical_map was initialized to INVALID_HWID to avoid
351 * matching valid MPIDR values.
353 static bool __init
is_mpidr_duplicate(unsigned int cpu
, u64 hwid
)
357 for (i
= 1; (i
< cpu
) && (i
< NR_CPUS
); i
++)
358 if (cpu_logical_map(i
) == hwid
)
364 * Initialize cpu operations for a logical cpu and
365 * set it in the possible mask on success
367 static int __init
smp_cpu_setup(int cpu
)
369 if (cpu_read_ops(cpu
))
372 if (cpu_ops
[cpu
]->cpu_init(cpu
))
375 set_cpu_possible(cpu
, true);
380 static bool bootcpu_valid __initdata
;
381 static unsigned int cpu_count
= 1;
385 * acpi_map_gic_cpu_interface - parse processor MADT entry
387 * Carry out sanity checks on MADT processor entry and initialize
388 * cpu_logical_map on success
391 acpi_map_gic_cpu_interface(struct acpi_madt_generic_interrupt
*processor
)
393 u64 hwid
= processor
->arm_mpidr
;
395 if (hwid
& ~MPIDR_HWID_BITMASK
|| hwid
== INVALID_HWID
) {
396 pr_err("skipping CPU entry with invalid MPIDR 0x%llx\n", hwid
);
400 if (!(processor
->flags
& ACPI_MADT_ENABLED
)) {
401 pr_err("skipping disabled CPU entry with 0x%llx MPIDR\n", hwid
);
405 if (is_mpidr_duplicate(cpu_count
, hwid
)) {
406 pr_err("duplicate CPU MPIDR 0x%llx in MADT\n", hwid
);
410 /* Check if GICC structure of boot CPU is available in the MADT */
411 if (cpu_logical_map(0) == hwid
) {
413 pr_err("duplicate boot CPU MPIDR: 0x%llx in MADT\n",
417 bootcpu_valid
= true;
421 if (cpu_count
>= NR_CPUS
)
424 /* map the logical cpu id to cpu MPIDR */
425 cpu_logical_map(cpu_count
) = hwid
;
431 acpi_parse_gic_cpu_interface(struct acpi_subtable_header
*header
,
432 const unsigned long end
)
434 struct acpi_madt_generic_interrupt
*processor
;
436 processor
= (struct acpi_madt_generic_interrupt
*)header
;
437 if (BAD_MADT_ENTRY(processor
, end
))
440 acpi_table_print_madt_entry(header
);
442 acpi_map_gic_cpu_interface(processor
);
447 #define acpi_table_parse_madt(...) do { } while (0)
451 * Enumerate the possible CPU set from the device tree and build the
452 * cpu logical map array containing MPIDR values related to logical
453 * cpus. Assumes that cpu_logical_map(0) has already been initialized.
455 void __init
of_parse_and_init_cpus(void)
457 struct device_node
*dn
= NULL
;
459 while ((dn
= of_find_node_by_type(dn
, "cpu"))) {
460 u64 hwid
= of_get_cpu_mpidr(dn
);
462 if (hwid
== INVALID_HWID
)
465 if (is_mpidr_duplicate(cpu_count
, hwid
)) {
466 pr_err("%s: duplicate cpu reg properties in the DT\n",
472 * The numbering scheme requires that the boot CPU
473 * must be assigned logical id 0. Record it so that
474 * the logical map built from DT is validated and can
477 if (hwid
== cpu_logical_map(0)) {
479 pr_err("%s: duplicate boot cpu reg property in DT\n",
484 bootcpu_valid
= true;
487 * cpu_logical_map has already been
488 * initialized and the boot cpu doesn't need
489 * the enable-method so continue without
495 if (cpu_count
>= NR_CPUS
)
498 pr_debug("cpu logical map 0x%llx\n", hwid
);
499 cpu_logical_map(cpu_count
) = hwid
;
506 * Enumerate the possible CPU set from the device tree or ACPI and build the
507 * cpu logical map array containing MPIDR values related to logical
508 * cpus. Assumes that cpu_logical_map(0) has already been initialized.
510 void __init
smp_init_cpus(void)
515 of_parse_and_init_cpus();
518 * do a walk of MADT to determine how many CPUs
519 * we have including disabled CPUs, and get information
520 * we need for SMP init
522 acpi_table_parse_madt(ACPI_MADT_TYPE_GENERIC_INTERRUPT
,
523 acpi_parse_gic_cpu_interface
, 0);
525 if (cpu_count
> NR_CPUS
)
526 pr_warn("no. of cores (%d) greater than configured maximum of %d - clipping\n",
529 if (!bootcpu_valid
) {
530 pr_err("missing boot CPU MPIDR, not enabling secondaries\n");
535 * We need to set the cpu_logical_map entries before enabling
536 * the cpus so that cpu processor description entries (DT cpu nodes
537 * and ACPI MADT entries) can be retrieved by matching the cpu hwid
538 * with entries in cpu_logical_map while initializing the cpus.
539 * If the cpu set-up fails, invalidate the cpu_logical_map entry.
541 for (i
= 1; i
< NR_CPUS
; i
++) {
542 if (cpu_logical_map(i
) != INVALID_HWID
) {
543 if (smp_cpu_setup(i
))
544 cpu_logical_map(i
) = INVALID_HWID
;
549 void __init
smp_prepare_cpus(unsigned int max_cpus
)
552 unsigned int cpu
, ncores
= num_possible_cpus();
556 smp_store_cpu_info(smp_processor_id());
559 * are we trying to boot more cores than exist?
561 if (max_cpus
> ncores
)
564 /* Don't bother if we're effectively UP */
569 * Initialise the present map (which describes the set of CPUs
570 * actually populated at the present time) and release the
571 * secondaries from the bootloader.
573 * Make sure we online at most (max_cpus - 1) additional CPUs.
576 for_each_possible_cpu(cpu
) {
580 if (cpu
== smp_processor_id())
586 err
= cpu_ops
[cpu
]->cpu_prepare(cpu
);
590 set_cpu_present(cpu
, true);
595 void (*__smp_cross_call
)(const struct cpumask
*, unsigned int);
597 void __init
set_smp_cross_call(void (*fn
)(const struct cpumask
*, unsigned int))
599 __smp_cross_call
= fn
;
602 static const char *ipi_types
[NR_IPI
] __tracepoint_string
= {
603 #define S(x,s) [x] = s
604 S(IPI_RESCHEDULE
, "Rescheduling interrupts"),
605 S(IPI_CALL_FUNC
, "Function call interrupts"),
606 S(IPI_CPU_STOP
, "CPU stop interrupts"),
607 S(IPI_TIMER
, "Timer broadcast interrupts"),
608 S(IPI_IRQ_WORK
, "IRQ work interrupts"),
611 static void smp_cross_call(const struct cpumask
*target
, unsigned int ipinr
)
613 trace_ipi_raise(target
, ipi_types
[ipinr
]);
614 __smp_cross_call(target
, ipinr
);
617 void show_ipi_list(struct seq_file
*p
, int prec
)
621 for (i
= 0; i
< NR_IPI
; i
++) {
622 seq_printf(p
, "%*s%u:%s", prec
- 1, "IPI", i
,
623 prec
>= 4 ? " " : "");
624 for_each_online_cpu(cpu
)
625 seq_printf(p
, "%10u ",
626 __get_irq_stat(cpu
, ipi_irqs
[i
]));
627 seq_printf(p
, " %s\n", ipi_types
[i
]);
631 u64
smp_irq_stat_cpu(unsigned int cpu
)
636 for (i
= 0; i
< NR_IPI
; i
++)
637 sum
+= __get_irq_stat(cpu
, ipi_irqs
[i
]);
642 void arch_send_call_function_ipi_mask(const struct cpumask
*mask
)
644 smp_cross_call(mask
, IPI_CALL_FUNC
);
647 void arch_send_call_function_single_ipi(int cpu
)
649 smp_cross_call(cpumask_of(cpu
), IPI_CALL_FUNC
);
652 #ifdef CONFIG_IRQ_WORK
653 void arch_irq_work_raise(void)
655 if (__smp_cross_call
)
656 smp_cross_call(cpumask_of(smp_processor_id()), IPI_IRQ_WORK
);
660 static DEFINE_RAW_SPINLOCK(stop_lock
);
663 * ipi_cpu_stop - handle IPI from smp_send_stop()
665 static void ipi_cpu_stop(unsigned int cpu
)
667 if (system_state
== SYSTEM_BOOTING
||
668 system_state
== SYSTEM_RUNNING
) {
669 raw_spin_lock(&stop_lock
);
670 pr_crit("CPU%u: stopping\n", cpu
);
672 raw_spin_unlock(&stop_lock
);
675 set_cpu_online(cpu
, false);
684 * Main handler for inter-processor interrupts
686 void handle_IPI(int ipinr
, struct pt_regs
*regs
)
688 unsigned int cpu
= smp_processor_id();
689 struct pt_regs
*old_regs
= set_irq_regs(regs
);
691 if ((unsigned)ipinr
< NR_IPI
) {
692 trace_ipi_entry(ipi_types
[ipinr
]);
693 __inc_irq_stat(cpu
, ipi_irqs
[ipinr
]);
703 generic_smp_call_function_interrupt();
713 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
716 tick_receive_broadcast();
721 #ifdef CONFIG_IRQ_WORK
730 pr_crit("CPU%u: Unknown IPI message 0x%x\n", cpu
, ipinr
);
734 if ((unsigned)ipinr
< NR_IPI
)
735 trace_ipi_exit(ipi_types
[ipinr
]);
736 set_irq_regs(old_regs
);
739 void smp_send_reschedule(int cpu
)
741 smp_cross_call(cpumask_of(cpu
), IPI_RESCHEDULE
);
744 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
745 void tick_broadcast(const struct cpumask
*mask
)
747 smp_cross_call(mask
, IPI_TIMER
);
751 void smp_send_stop(void)
753 unsigned long timeout
;
755 if (num_online_cpus() > 1) {
758 cpumask_copy(&mask
, cpu_online_mask
);
759 cpumask_clear_cpu(smp_processor_id(), &mask
);
761 smp_cross_call(&mask
, IPI_CPU_STOP
);
764 /* Wait up to one second for other CPUs to stop */
765 timeout
= USEC_PER_SEC
;
766 while (num_online_cpus() > 1 && timeout
--)
769 if (num_online_cpus() > 1)
770 pr_warning("SMP: failed to stop secondary CPUs\n");
776 int setup_profiling_timer(unsigned int multiplier
)