2 * linux/kernel/hrtimer.c
4 * Copyright(C) 2005-2006, Thomas Gleixner <tglx@linutronix.de>
5 * Copyright(C) 2005-2007, Red Hat, Inc., Ingo Molnar
6 * Copyright(C) 2006-2007 Timesys Corp., Thomas Gleixner
8 * High-resolution kernel timers
10 * In contrast to the low-resolution timeout API implemented in
11 * kernel/timer.c, hrtimers provide finer resolution and accuracy
12 * depending on system configuration and capabilities.
14 * These timers are currently used for:
18 * - precise in-kernel timing
20 * Started by: Thomas Gleixner and Ingo Molnar
23 * based on kernel/timer.c
25 * Help, testing, suggestions, bugfixes, improvements were
28 * George Anzinger, Andrew Morton, Steven Rostedt, Roman Zippel
31 * For licencing details see kernel-base/COPYING
34 #include <linux/cpu.h>
35 #include <linux/export.h>
36 #include <linux/percpu.h>
37 #include <linux/hrtimer.h>
38 #include <linux/notifier.h>
39 #include <linux/syscalls.h>
40 #include <linux/kallsyms.h>
41 #include <linux/interrupt.h>
42 #include <linux/tick.h>
43 #include <linux/seq_file.h>
44 #include <linux/err.h>
45 #include <linux/debugobjects.h>
46 #include <linux/sched.h>
47 #include <linux/sched/sysctl.h>
48 #include <linux/sched/rt.h>
49 #include <linux/sched/deadline.h>
50 #include <linux/timer.h>
51 #include <linux/freezer.h>
53 #include <asm/uaccess.h>
55 #include <trace/events/timer.h>
60 * There are more clockids then hrtimer bases. Thus, we index
61 * into the timer bases by the hrtimer_base_type enum. When trying
62 * to reach a base using a clockid, hrtimer_clockid_to_base()
63 * is used to convert from clockid to the proper hrtimer_base_type.
65 DEFINE_PER_CPU(struct hrtimer_cpu_base
, hrtimer_bases
) =
68 .lock
= __RAW_SPIN_LOCK_UNLOCKED(hrtimer_bases
.lock
),
72 .index
= HRTIMER_BASE_MONOTONIC
,
73 .clockid
= CLOCK_MONOTONIC
,
74 .get_time
= &ktime_get
,
75 .resolution
= KTIME_LOW_RES
,
78 .index
= HRTIMER_BASE_REALTIME
,
79 .clockid
= CLOCK_REALTIME
,
80 .get_time
= &ktime_get_real
,
81 .resolution
= KTIME_LOW_RES
,
84 .index
= HRTIMER_BASE_BOOTTIME
,
85 .clockid
= CLOCK_BOOTTIME
,
86 .get_time
= &ktime_get_boottime
,
87 .resolution
= KTIME_LOW_RES
,
90 .index
= HRTIMER_BASE_TAI
,
92 .get_time
= &ktime_get_clocktai
,
93 .resolution
= KTIME_LOW_RES
,
98 static const int hrtimer_clock_to_base_table
[MAX_CLOCKS
] = {
99 [CLOCK_REALTIME
] = HRTIMER_BASE_REALTIME
,
100 [CLOCK_MONOTONIC
] = HRTIMER_BASE_MONOTONIC
,
101 [CLOCK_BOOTTIME
] = HRTIMER_BASE_BOOTTIME
,
102 [CLOCK_TAI
] = HRTIMER_BASE_TAI
,
105 static inline int hrtimer_clockid_to_base(clockid_t clock_id
)
107 return hrtimer_clock_to_base_table
[clock_id
];
112 * Get the coarse grained time at the softirq based on xtime and
115 static void hrtimer_get_softirq_time(struct hrtimer_cpu_base
*base
)
117 ktime_t xtim
, mono
, boot
;
118 struct timespec xts
, tom
, slp
;
121 get_xtime_and_monotonic_and_sleep_offset(&xts
, &tom
, &slp
);
122 tai_offset
= timekeeping_get_tai_offset();
124 xtim
= timespec_to_ktime(xts
);
125 mono
= ktime_add(xtim
, timespec_to_ktime(tom
));
126 boot
= ktime_add(mono
, timespec_to_ktime(slp
));
127 base
->clock_base
[HRTIMER_BASE_REALTIME
].softirq_time
= xtim
;
128 base
->clock_base
[HRTIMER_BASE_MONOTONIC
].softirq_time
= mono
;
129 base
->clock_base
[HRTIMER_BASE_BOOTTIME
].softirq_time
= boot
;
130 base
->clock_base
[HRTIMER_BASE_TAI
].softirq_time
=
131 ktime_add(xtim
, ktime_set(tai_offset
, 0));
135 * Functions and macros which are different for UP/SMP systems are kept in a
141 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
142 * means that all timers which are tied to this base via timer->base are
143 * locked, and the base itself is locked too.
145 * So __run_timers/migrate_timers can safely modify all timers which could
146 * be found on the lists/queues.
148 * When the timer's base is locked, and the timer removed from list, it is
149 * possible to set timer->base = NULL and drop the lock: the timer remains
153 struct hrtimer_clock_base
*lock_hrtimer_base(const struct hrtimer
*timer
,
154 unsigned long *flags
)
156 struct hrtimer_clock_base
*base
;
160 if (likely(base
!= NULL
)) {
161 raw_spin_lock_irqsave(&base
->cpu_base
->lock
, *flags
);
162 if (likely(base
== timer
->base
))
164 /* The timer has migrated to another CPU: */
165 raw_spin_unlock_irqrestore(&base
->cpu_base
->lock
, *flags
);
172 * With HIGHRES=y we do not migrate the timer when it is expiring
173 * before the next event on the target cpu because we cannot reprogram
174 * the target cpu hardware and we would cause it to fire late.
176 * Called with cpu_base->lock of target cpu held.
179 hrtimer_check_target(struct hrtimer
*timer
, struct hrtimer_clock_base
*new_base
)
181 #ifdef CONFIG_HIGH_RES_TIMERS
184 if (!new_base
->cpu_base
->hres_active
)
187 expires
= ktime_sub(hrtimer_get_expires(timer
), new_base
->offset
);
188 return expires
.tv64
<= new_base
->cpu_base
->expires_next
.tv64
;
195 * Switch the timer base to the current CPU when possible.
197 static inline struct hrtimer_clock_base
*
198 switch_hrtimer_base(struct hrtimer
*timer
, struct hrtimer_clock_base
*base
,
201 struct hrtimer_clock_base
*new_base
;
202 struct hrtimer_cpu_base
*new_cpu_base
;
203 int this_cpu
= smp_processor_id();
204 int cpu
= get_nohz_timer_target(pinned
);
205 int basenum
= base
->index
;
208 new_cpu_base
= &per_cpu(hrtimer_bases
, cpu
);
209 new_base
= &new_cpu_base
->clock_base
[basenum
];
211 if (base
!= new_base
) {
213 * We are trying to move timer to new_base.
214 * However we can't change timer's base while it is running,
215 * so we keep it on the same CPU. No hassle vs. reprogramming
216 * the event source in the high resolution case. The softirq
217 * code will take care of this when the timer function has
218 * completed. There is no conflict as we hold the lock until
219 * the timer is enqueued.
221 if (unlikely(hrtimer_callback_running(timer
)))
224 /* See the comment in lock_timer_base() */
226 raw_spin_unlock(&base
->cpu_base
->lock
);
227 raw_spin_lock(&new_base
->cpu_base
->lock
);
229 if (cpu
!= this_cpu
&& hrtimer_check_target(timer
, new_base
)) {
231 raw_spin_unlock(&new_base
->cpu_base
->lock
);
232 raw_spin_lock(&base
->cpu_base
->lock
);
236 timer
->base
= new_base
;
241 #else /* CONFIG_SMP */
243 static inline struct hrtimer_clock_base
*
244 lock_hrtimer_base(const struct hrtimer
*timer
, unsigned long *flags
)
246 struct hrtimer_clock_base
*base
= timer
->base
;
248 raw_spin_lock_irqsave(&base
->cpu_base
->lock
, *flags
);
253 # define switch_hrtimer_base(t, b, p) (b)
255 #endif /* !CONFIG_SMP */
258 * Functions for the union type storage format of ktime_t which are
259 * too large for inlining:
261 #if BITS_PER_LONG < 64
262 # ifndef CONFIG_KTIME_SCALAR
264 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
266 * @nsec: the scalar nsec value to add
268 * Returns the sum of kt and nsec in ktime_t format
270 ktime_t
ktime_add_ns(const ktime_t kt
, u64 nsec
)
274 if (likely(nsec
< NSEC_PER_SEC
)) {
277 unsigned long rem
= do_div(nsec
, NSEC_PER_SEC
);
279 /* Make sure nsec fits into long */
280 if (unlikely(nsec
> KTIME_SEC_MAX
))
281 return (ktime_t
){ .tv64
= KTIME_MAX
};
283 tmp
= ktime_set((long)nsec
, rem
);
286 return ktime_add(kt
, tmp
);
289 EXPORT_SYMBOL_GPL(ktime_add_ns
);
292 * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable
294 * @nsec: the scalar nsec value to subtract
296 * Returns the subtraction of @nsec from @kt in ktime_t format
298 ktime_t
ktime_sub_ns(const ktime_t kt
, u64 nsec
)
302 if (likely(nsec
< NSEC_PER_SEC
)) {
305 unsigned long rem
= do_div(nsec
, NSEC_PER_SEC
);
307 tmp
= ktime_set((long)nsec
, rem
);
310 return ktime_sub(kt
, tmp
);
313 EXPORT_SYMBOL_GPL(ktime_sub_ns
);
314 # endif /* !CONFIG_KTIME_SCALAR */
317 * Divide a ktime value by a nanosecond value
319 u64
ktime_divns(const ktime_t kt
, s64 div
)
324 dclc
= ktime_to_ns(kt
);
325 /* Make sure the divisor is less than 2^32: */
331 do_div(dclc
, (unsigned long) div
);
335 #endif /* BITS_PER_LONG >= 64 */
338 * Add two ktime values and do a safety check for overflow:
340 ktime_t
ktime_add_safe(const ktime_t lhs
, const ktime_t rhs
)
342 ktime_t res
= ktime_add(lhs
, rhs
);
345 * We use KTIME_SEC_MAX here, the maximum timeout which we can
346 * return to user space in a timespec:
348 if (res
.tv64
< 0 || res
.tv64
< lhs
.tv64
|| res
.tv64
< rhs
.tv64
)
349 res
= ktime_set(KTIME_SEC_MAX
, 0);
354 EXPORT_SYMBOL_GPL(ktime_add_safe
);
356 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
358 static struct debug_obj_descr hrtimer_debug_descr
;
360 static void *hrtimer_debug_hint(void *addr
)
362 return ((struct hrtimer
*) addr
)->function
;
366 * fixup_init is called when:
367 * - an active object is initialized
369 static int hrtimer_fixup_init(void *addr
, enum debug_obj_state state
)
371 struct hrtimer
*timer
= addr
;
374 case ODEBUG_STATE_ACTIVE
:
375 hrtimer_cancel(timer
);
376 debug_object_init(timer
, &hrtimer_debug_descr
);
384 * fixup_activate is called when:
385 * - an active object is activated
386 * - an unknown object is activated (might be a statically initialized object)
388 static int hrtimer_fixup_activate(void *addr
, enum debug_obj_state state
)
392 case ODEBUG_STATE_NOTAVAILABLE
:
396 case ODEBUG_STATE_ACTIVE
:
405 * fixup_free is called when:
406 * - an active object is freed
408 static int hrtimer_fixup_free(void *addr
, enum debug_obj_state state
)
410 struct hrtimer
*timer
= addr
;
413 case ODEBUG_STATE_ACTIVE
:
414 hrtimer_cancel(timer
);
415 debug_object_free(timer
, &hrtimer_debug_descr
);
422 static struct debug_obj_descr hrtimer_debug_descr
= {
424 .debug_hint
= hrtimer_debug_hint
,
425 .fixup_init
= hrtimer_fixup_init
,
426 .fixup_activate
= hrtimer_fixup_activate
,
427 .fixup_free
= hrtimer_fixup_free
,
430 static inline void debug_hrtimer_init(struct hrtimer
*timer
)
432 debug_object_init(timer
, &hrtimer_debug_descr
);
435 static inline void debug_hrtimer_activate(struct hrtimer
*timer
)
437 debug_object_activate(timer
, &hrtimer_debug_descr
);
440 static inline void debug_hrtimer_deactivate(struct hrtimer
*timer
)
442 debug_object_deactivate(timer
, &hrtimer_debug_descr
);
445 static inline void debug_hrtimer_free(struct hrtimer
*timer
)
447 debug_object_free(timer
, &hrtimer_debug_descr
);
450 static void __hrtimer_init(struct hrtimer
*timer
, clockid_t clock_id
,
451 enum hrtimer_mode mode
);
453 void hrtimer_init_on_stack(struct hrtimer
*timer
, clockid_t clock_id
,
454 enum hrtimer_mode mode
)
456 debug_object_init_on_stack(timer
, &hrtimer_debug_descr
);
457 __hrtimer_init(timer
, clock_id
, mode
);
459 EXPORT_SYMBOL_GPL(hrtimer_init_on_stack
);
461 void destroy_hrtimer_on_stack(struct hrtimer
*timer
)
463 debug_object_free(timer
, &hrtimer_debug_descr
);
467 static inline void debug_hrtimer_init(struct hrtimer
*timer
) { }
468 static inline void debug_hrtimer_activate(struct hrtimer
*timer
) { }
469 static inline void debug_hrtimer_deactivate(struct hrtimer
*timer
) { }
473 debug_init(struct hrtimer
*timer
, clockid_t clockid
,
474 enum hrtimer_mode mode
)
476 debug_hrtimer_init(timer
);
477 trace_hrtimer_init(timer
, clockid
, mode
);
480 static inline void debug_activate(struct hrtimer
*timer
)
482 debug_hrtimer_activate(timer
);
483 trace_hrtimer_start(timer
);
486 static inline void debug_deactivate(struct hrtimer
*timer
)
488 debug_hrtimer_deactivate(timer
);
489 trace_hrtimer_cancel(timer
);
492 /* High resolution timer related functions */
493 #ifdef CONFIG_HIGH_RES_TIMERS
496 * High resolution timer enabled ?
498 static int hrtimer_hres_enabled __read_mostly
= 1;
501 * Enable / Disable high resolution mode
503 static int __init
setup_hrtimer_hres(char *str
)
505 if (!strcmp(str
, "off"))
506 hrtimer_hres_enabled
= 0;
507 else if (!strcmp(str
, "on"))
508 hrtimer_hres_enabled
= 1;
514 __setup("highres=", setup_hrtimer_hres
);
517 * hrtimer_high_res_enabled - query, if the highres mode is enabled
519 static inline int hrtimer_is_hres_enabled(void)
521 return hrtimer_hres_enabled
;
525 * Is the high resolution mode active ?
527 static inline int hrtimer_hres_active(void)
529 return __this_cpu_read(hrtimer_bases
.hres_active
);
533 * Reprogram the event source with checking both queues for the
535 * Called with interrupts disabled and base->lock held
538 hrtimer_force_reprogram(struct hrtimer_cpu_base
*cpu_base
, int skip_equal
)
541 struct hrtimer_clock_base
*base
= cpu_base
->clock_base
;
542 ktime_t expires
, expires_next
;
544 expires_next
.tv64
= KTIME_MAX
;
546 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++, base
++) {
547 struct hrtimer
*timer
;
548 struct timerqueue_node
*next
;
550 next
= timerqueue_getnext(&base
->active
);
553 timer
= container_of(next
, struct hrtimer
, node
);
555 expires
= ktime_sub(hrtimer_get_expires(timer
), base
->offset
);
557 * clock_was_set() has changed base->offset so the
558 * result might be negative. Fix it up to prevent a
559 * false positive in clockevents_program_event()
561 if (expires
.tv64
< 0)
563 if (expires
.tv64
< expires_next
.tv64
)
564 expires_next
= expires
;
567 if (skip_equal
&& expires_next
.tv64
== cpu_base
->expires_next
.tv64
)
570 cpu_base
->expires_next
.tv64
= expires_next
.tv64
;
572 if (cpu_base
->expires_next
.tv64
!= KTIME_MAX
)
573 tick_program_event(cpu_base
->expires_next
, 1);
577 * Shared reprogramming for clock_realtime and clock_monotonic
579 * When a timer is enqueued and expires earlier than the already enqueued
580 * timers, we have to check, whether it expires earlier than the timer for
581 * which the clock event device was armed.
583 * Called with interrupts disabled and base->cpu_base.lock held
585 static int hrtimer_reprogram(struct hrtimer
*timer
,
586 struct hrtimer_clock_base
*base
)
588 struct hrtimer_cpu_base
*cpu_base
= &__get_cpu_var(hrtimer_bases
);
589 ktime_t expires
= ktime_sub(hrtimer_get_expires(timer
), base
->offset
);
592 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer
) < 0);
595 * When the callback is running, we do not reprogram the clock event
596 * device. The timer callback is either running on a different CPU or
597 * the callback is executed in the hrtimer_interrupt context. The
598 * reprogramming is handled either by the softirq, which called the
599 * callback or at the end of the hrtimer_interrupt.
601 if (hrtimer_callback_running(timer
))
605 * CLOCK_REALTIME timer might be requested with an absolute
606 * expiry time which is less than base->offset. Nothing wrong
607 * about that, just avoid to call into the tick code, which
608 * has now objections against negative expiry values.
610 if (expires
.tv64
< 0)
613 if (expires
.tv64
>= cpu_base
->expires_next
.tv64
)
617 * If a hang was detected in the last timer interrupt then we
618 * do not schedule a timer which is earlier than the expiry
619 * which we enforced in the hang detection. We want the system
622 if (cpu_base
->hang_detected
)
626 * Clockevents returns -ETIME, when the event was in the past.
628 res
= tick_program_event(expires
, 0);
629 if (!IS_ERR_VALUE(res
))
630 cpu_base
->expires_next
= expires
;
635 * Initialize the high resolution related parts of cpu_base
637 static inline void hrtimer_init_hres(struct hrtimer_cpu_base
*base
)
639 base
->expires_next
.tv64
= KTIME_MAX
;
640 base
->hres_active
= 0;
644 * When High resolution timers are active, try to reprogram. Note, that in case
645 * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry
646 * check happens. The timer gets enqueued into the rbtree. The reprogramming
647 * and expiry check is done in the hrtimer_interrupt or in the softirq.
649 static inline int hrtimer_enqueue_reprogram(struct hrtimer
*timer
,
650 struct hrtimer_clock_base
*base
)
652 return base
->cpu_base
->hres_active
&& hrtimer_reprogram(timer
, base
);
655 static inline ktime_t
hrtimer_update_base(struct hrtimer_cpu_base
*base
)
657 ktime_t
*offs_real
= &base
->clock_base
[HRTIMER_BASE_REALTIME
].offset
;
658 ktime_t
*offs_boot
= &base
->clock_base
[HRTIMER_BASE_BOOTTIME
].offset
;
659 ktime_t
*offs_tai
= &base
->clock_base
[HRTIMER_BASE_TAI
].offset
;
661 return ktime_get_update_offsets(offs_real
, offs_boot
, offs_tai
);
665 * Retrigger next event is called after clock was set
667 * Called with interrupts disabled via on_each_cpu()
669 static void retrigger_next_event(void *arg
)
671 struct hrtimer_cpu_base
*base
= &__get_cpu_var(hrtimer_bases
);
673 if (!hrtimer_hres_active())
676 raw_spin_lock(&base
->lock
);
677 hrtimer_update_base(base
);
678 hrtimer_force_reprogram(base
, 0);
679 raw_spin_unlock(&base
->lock
);
683 * Switch to high resolution mode
685 static int hrtimer_switch_to_hres(void)
687 int i
, cpu
= smp_processor_id();
688 struct hrtimer_cpu_base
*base
= &per_cpu(hrtimer_bases
, cpu
);
691 if (base
->hres_active
)
694 local_irq_save(flags
);
696 if (tick_init_highres()) {
697 local_irq_restore(flags
);
698 printk(KERN_WARNING
"Could not switch to high resolution "
699 "mode on CPU %d\n", cpu
);
702 base
->hres_active
= 1;
703 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++)
704 base
->clock_base
[i
].resolution
= KTIME_HIGH_RES
;
706 tick_setup_sched_timer();
707 /* "Retrigger" the interrupt to get things going */
708 retrigger_next_event(NULL
);
709 local_irq_restore(flags
);
713 static void clock_was_set_work(struct work_struct
*work
)
718 static DECLARE_WORK(hrtimer_work
, clock_was_set_work
);
721 * Called from timekeeping and resume code to reprogramm the hrtimer
722 * interrupt device on all cpus.
724 void clock_was_set_delayed(void)
726 schedule_work(&hrtimer_work
);
731 static inline int hrtimer_hres_active(void) { return 0; }
732 static inline int hrtimer_is_hres_enabled(void) { return 0; }
733 static inline int hrtimer_switch_to_hres(void) { return 0; }
735 hrtimer_force_reprogram(struct hrtimer_cpu_base
*base
, int skip_equal
) { }
736 static inline int hrtimer_enqueue_reprogram(struct hrtimer
*timer
,
737 struct hrtimer_clock_base
*base
)
741 static inline void hrtimer_init_hres(struct hrtimer_cpu_base
*base
) { }
742 static inline void retrigger_next_event(void *arg
) { }
744 #endif /* CONFIG_HIGH_RES_TIMERS */
747 * Clock realtime was set
749 * Change the offset of the realtime clock vs. the monotonic
752 * We might have to reprogram the high resolution timer interrupt. On
753 * SMP we call the architecture specific code to retrigger _all_ high
754 * resolution timer interrupts. On UP we just disable interrupts and
755 * call the high resolution interrupt code.
757 void clock_was_set(void)
759 #ifdef CONFIG_HIGH_RES_TIMERS
760 /* Retrigger the CPU local events everywhere */
761 on_each_cpu(retrigger_next_event
, NULL
, 1);
763 timerfd_clock_was_set();
767 * During resume we might have to reprogram the high resolution timer
768 * interrupt on all online CPUs. However, all other CPUs will be
769 * stopped with IRQs interrupts disabled so the clock_was_set() call
772 void hrtimers_resume(void)
774 WARN_ONCE(!irqs_disabled(),
775 KERN_INFO
"hrtimers_resume() called with IRQs enabled!");
777 /* Retrigger on the local CPU */
778 retrigger_next_event(NULL
);
779 /* And schedule a retrigger for all others */
780 clock_was_set_delayed();
783 static inline void timer_stats_hrtimer_set_start_info(struct hrtimer
*timer
)
785 #ifdef CONFIG_TIMER_STATS
786 if (timer
->start_site
)
788 timer
->start_site
= __builtin_return_address(0);
789 memcpy(timer
->start_comm
, current
->comm
, TASK_COMM_LEN
);
790 timer
->start_pid
= current
->pid
;
794 static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer
*timer
)
796 #ifdef CONFIG_TIMER_STATS
797 timer
->start_site
= NULL
;
801 static inline void timer_stats_account_hrtimer(struct hrtimer
*timer
)
803 #ifdef CONFIG_TIMER_STATS
804 if (likely(!timer_stats_active
))
806 timer_stats_update_stats(timer
, timer
->start_pid
, timer
->start_site
,
807 timer
->function
, timer
->start_comm
, 0);
812 * Counterpart to lock_hrtimer_base above:
815 void unlock_hrtimer_base(const struct hrtimer
*timer
, unsigned long *flags
)
817 raw_spin_unlock_irqrestore(&timer
->base
->cpu_base
->lock
, *flags
);
821 * hrtimer_forward - forward the timer expiry
822 * @timer: hrtimer to forward
823 * @now: forward past this time
824 * @interval: the interval to forward
826 * Forward the timer expiry so it will expire in the future.
827 * Returns the number of overruns.
829 u64
hrtimer_forward(struct hrtimer
*timer
, ktime_t now
, ktime_t interval
)
834 delta
= ktime_sub(now
, hrtimer_get_expires(timer
));
839 if (interval
.tv64
< timer
->base
->resolution
.tv64
)
840 interval
.tv64
= timer
->base
->resolution
.tv64
;
842 if (unlikely(delta
.tv64
>= interval
.tv64
)) {
843 s64 incr
= ktime_to_ns(interval
);
845 orun
= ktime_divns(delta
, incr
);
846 hrtimer_add_expires_ns(timer
, incr
* orun
);
847 if (hrtimer_get_expires_tv64(timer
) > now
.tv64
)
850 * This (and the ktime_add() below) is the
851 * correction for exact:
855 hrtimer_add_expires(timer
, interval
);
859 EXPORT_SYMBOL_GPL(hrtimer_forward
);
862 * enqueue_hrtimer - internal function to (re)start a timer
864 * The timer is inserted in expiry order. Insertion into the
865 * red black tree is O(log(n)). Must hold the base lock.
867 * Returns 1 when the new timer is the leftmost timer in the tree.
869 static int enqueue_hrtimer(struct hrtimer
*timer
,
870 struct hrtimer_clock_base
*base
)
872 debug_activate(timer
);
874 timerqueue_add(&base
->active
, &timer
->node
);
875 base
->cpu_base
->active_bases
|= 1 << base
->index
;
878 * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
879 * state of a possibly running callback.
881 timer
->state
|= HRTIMER_STATE_ENQUEUED
;
883 return (&timer
->node
== base
->active
.next
);
887 * __remove_hrtimer - internal function to remove a timer
889 * Caller must hold the base lock.
891 * High resolution timer mode reprograms the clock event device when the
892 * timer is the one which expires next. The caller can disable this by setting
893 * reprogram to zero. This is useful, when the context does a reprogramming
894 * anyway (e.g. timer interrupt)
896 static void __remove_hrtimer(struct hrtimer
*timer
,
897 struct hrtimer_clock_base
*base
,
898 unsigned long newstate
, int reprogram
)
900 struct timerqueue_node
*next_timer
;
901 if (!(timer
->state
& HRTIMER_STATE_ENQUEUED
))
904 next_timer
= timerqueue_getnext(&base
->active
);
905 timerqueue_del(&base
->active
, &timer
->node
);
906 if (&timer
->node
== next_timer
) {
907 #ifdef CONFIG_HIGH_RES_TIMERS
908 /* Reprogram the clock event device. if enabled */
909 if (reprogram
&& hrtimer_hres_active()) {
912 expires
= ktime_sub(hrtimer_get_expires(timer
),
914 if (base
->cpu_base
->expires_next
.tv64
== expires
.tv64
)
915 hrtimer_force_reprogram(base
->cpu_base
, 1);
919 if (!timerqueue_getnext(&base
->active
))
920 base
->cpu_base
->active_bases
&= ~(1 << base
->index
);
922 timer
->state
= newstate
;
926 * remove hrtimer, called with base lock held
929 remove_hrtimer(struct hrtimer
*timer
, struct hrtimer_clock_base
*base
)
931 if (hrtimer_is_queued(timer
)) {
936 * Remove the timer and force reprogramming when high
937 * resolution mode is active and the timer is on the current
938 * CPU. If we remove a timer on another CPU, reprogramming is
939 * skipped. The interrupt event on this CPU is fired and
940 * reprogramming happens in the interrupt handler. This is a
941 * rare case and less expensive than a smp call.
943 debug_deactivate(timer
);
944 timer_stats_hrtimer_clear_start_info(timer
);
945 reprogram
= base
->cpu_base
== &__get_cpu_var(hrtimer_bases
);
947 * We must preserve the CALLBACK state flag here,
948 * otherwise we could move the timer base in
949 * switch_hrtimer_base.
951 state
= timer
->state
& HRTIMER_STATE_CALLBACK
;
952 __remove_hrtimer(timer
, base
, state
, reprogram
);
958 int __hrtimer_start_range_ns(struct hrtimer
*timer
, ktime_t tim
,
959 unsigned long delta_ns
, const enum hrtimer_mode mode
,
962 struct hrtimer_clock_base
*base
, *new_base
;
966 base
= lock_hrtimer_base(timer
, &flags
);
968 /* Remove an active timer from the queue: */
969 ret
= remove_hrtimer(timer
, base
);
971 /* Switch the timer base, if necessary: */
972 new_base
= switch_hrtimer_base(timer
, base
, mode
& HRTIMER_MODE_PINNED
);
974 if (mode
& HRTIMER_MODE_REL
) {
975 tim
= ktime_add_safe(tim
, new_base
->get_time());
977 * CONFIG_TIME_LOW_RES is a temporary way for architectures
978 * to signal that they simply return xtime in
979 * do_gettimeoffset(). In this case we want to round up by
980 * resolution when starting a relative timer, to avoid short
981 * timeouts. This will go away with the GTOD framework.
983 #ifdef CONFIG_TIME_LOW_RES
984 tim
= ktime_add_safe(tim
, base
->resolution
);
988 hrtimer_set_expires_range_ns(timer
, tim
, delta_ns
);
990 timer_stats_hrtimer_set_start_info(timer
);
992 leftmost
= enqueue_hrtimer(timer
, new_base
);
995 * Only allow reprogramming if the new base is on this CPU.
996 * (it might still be on another CPU if the timer was pending)
998 * XXX send_remote_softirq() ?
1000 if (leftmost
&& new_base
->cpu_base
== &__get_cpu_var(hrtimer_bases
)
1001 && hrtimer_enqueue_reprogram(timer
, new_base
)) {
1004 * We need to drop cpu_base->lock to avoid a
1005 * lock ordering issue vs. rq->lock.
1007 raw_spin_unlock(&new_base
->cpu_base
->lock
);
1008 raise_softirq_irqoff(HRTIMER_SOFTIRQ
);
1009 local_irq_restore(flags
);
1012 __raise_softirq_irqoff(HRTIMER_SOFTIRQ
);
1016 unlock_hrtimer_base(timer
, &flags
);
1022 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
1023 * @timer: the timer to be added
1025 * @delta_ns: "slack" range for the timer
1026 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
1027 * relative (HRTIMER_MODE_REL)
1031 * 1 when the timer was active
1033 int hrtimer_start_range_ns(struct hrtimer
*timer
, ktime_t tim
,
1034 unsigned long delta_ns
, const enum hrtimer_mode mode
)
1036 return __hrtimer_start_range_ns(timer
, tim
, delta_ns
, mode
, 1);
1038 EXPORT_SYMBOL_GPL(hrtimer_start_range_ns
);
1041 * hrtimer_start - (re)start an hrtimer on the current CPU
1042 * @timer: the timer to be added
1044 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
1045 * relative (HRTIMER_MODE_REL)
1049 * 1 when the timer was active
1052 hrtimer_start(struct hrtimer
*timer
, ktime_t tim
, const enum hrtimer_mode mode
)
1054 return __hrtimer_start_range_ns(timer
, tim
, 0, mode
, 1);
1056 EXPORT_SYMBOL_GPL(hrtimer_start
);
1060 * hrtimer_try_to_cancel - try to deactivate a timer
1061 * @timer: hrtimer to stop
1064 * 0 when the timer was not active
1065 * 1 when the timer was active
1066 * -1 when the timer is currently excuting the callback function and
1069 int hrtimer_try_to_cancel(struct hrtimer
*timer
)
1071 struct hrtimer_clock_base
*base
;
1072 unsigned long flags
;
1075 base
= lock_hrtimer_base(timer
, &flags
);
1077 if (!hrtimer_callback_running(timer
))
1078 ret
= remove_hrtimer(timer
, base
);
1080 unlock_hrtimer_base(timer
, &flags
);
1085 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel
);
1088 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1089 * @timer: the timer to be cancelled
1092 * 0 when the timer was not active
1093 * 1 when the timer was active
1095 int hrtimer_cancel(struct hrtimer
*timer
)
1098 int ret
= hrtimer_try_to_cancel(timer
);
1105 EXPORT_SYMBOL_GPL(hrtimer_cancel
);
1108 * hrtimer_get_remaining - get remaining time for the timer
1109 * @timer: the timer to read
1111 ktime_t
hrtimer_get_remaining(const struct hrtimer
*timer
)
1113 unsigned long flags
;
1116 lock_hrtimer_base(timer
, &flags
);
1117 rem
= hrtimer_expires_remaining(timer
);
1118 unlock_hrtimer_base(timer
, &flags
);
1122 EXPORT_SYMBOL_GPL(hrtimer_get_remaining
);
1124 #ifdef CONFIG_NO_HZ_COMMON
1126 * hrtimer_get_next_event - get the time until next expiry event
1128 * Returns the delta to the next expiry event or KTIME_MAX if no timer
1131 ktime_t
hrtimer_get_next_event(void)
1133 struct hrtimer_cpu_base
*cpu_base
= &__get_cpu_var(hrtimer_bases
);
1134 struct hrtimer_clock_base
*base
= cpu_base
->clock_base
;
1135 ktime_t delta
, mindelta
= { .tv64
= KTIME_MAX
};
1136 unsigned long flags
;
1139 raw_spin_lock_irqsave(&cpu_base
->lock
, flags
);
1141 if (!hrtimer_hres_active()) {
1142 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++, base
++) {
1143 struct hrtimer
*timer
;
1144 struct timerqueue_node
*next
;
1146 next
= timerqueue_getnext(&base
->active
);
1150 timer
= container_of(next
, struct hrtimer
, node
);
1151 delta
.tv64
= hrtimer_get_expires_tv64(timer
);
1152 delta
= ktime_sub(delta
, base
->get_time());
1153 if (delta
.tv64
< mindelta
.tv64
)
1154 mindelta
.tv64
= delta
.tv64
;
1158 raw_spin_unlock_irqrestore(&cpu_base
->lock
, flags
);
1160 if (mindelta
.tv64
< 0)
1166 static void __hrtimer_init(struct hrtimer
*timer
, clockid_t clock_id
,
1167 enum hrtimer_mode mode
)
1169 struct hrtimer_cpu_base
*cpu_base
;
1172 memset(timer
, 0, sizeof(struct hrtimer
));
1174 cpu_base
= &__raw_get_cpu_var(hrtimer_bases
);
1176 if (clock_id
== CLOCK_REALTIME
&& mode
!= HRTIMER_MODE_ABS
)
1177 clock_id
= CLOCK_MONOTONIC
;
1179 base
= hrtimer_clockid_to_base(clock_id
);
1180 timer
->base
= &cpu_base
->clock_base
[base
];
1181 timerqueue_init(&timer
->node
);
1183 #ifdef CONFIG_TIMER_STATS
1184 timer
->start_site
= NULL
;
1185 timer
->start_pid
= -1;
1186 memset(timer
->start_comm
, 0, TASK_COMM_LEN
);
1191 * hrtimer_init - initialize a timer to the given clock
1192 * @timer: the timer to be initialized
1193 * @clock_id: the clock to be used
1194 * @mode: timer mode abs/rel
1196 void hrtimer_init(struct hrtimer
*timer
, clockid_t clock_id
,
1197 enum hrtimer_mode mode
)
1199 debug_init(timer
, clock_id
, mode
);
1200 __hrtimer_init(timer
, clock_id
, mode
);
1202 EXPORT_SYMBOL_GPL(hrtimer_init
);
1205 * hrtimer_get_res - get the timer resolution for a clock
1206 * @which_clock: which clock to query
1207 * @tp: pointer to timespec variable to store the resolution
1209 * Store the resolution of the clock selected by @which_clock in the
1210 * variable pointed to by @tp.
1212 int hrtimer_get_res(const clockid_t which_clock
, struct timespec
*tp
)
1214 struct hrtimer_cpu_base
*cpu_base
;
1215 int base
= hrtimer_clockid_to_base(which_clock
);
1217 cpu_base
= &__raw_get_cpu_var(hrtimer_bases
);
1218 *tp
= ktime_to_timespec(cpu_base
->clock_base
[base
].resolution
);
1222 EXPORT_SYMBOL_GPL(hrtimer_get_res
);
1224 static void __run_hrtimer(struct hrtimer
*timer
, ktime_t
*now
)
1226 struct hrtimer_clock_base
*base
= timer
->base
;
1227 struct hrtimer_cpu_base
*cpu_base
= base
->cpu_base
;
1228 enum hrtimer_restart (*fn
)(struct hrtimer
*);
1231 WARN_ON(!irqs_disabled());
1233 debug_deactivate(timer
);
1234 __remove_hrtimer(timer
, base
, HRTIMER_STATE_CALLBACK
, 0);
1235 timer_stats_account_hrtimer(timer
);
1236 fn
= timer
->function
;
1239 * Because we run timers from hardirq context, there is no chance
1240 * they get migrated to another cpu, therefore its safe to unlock
1243 raw_spin_unlock(&cpu_base
->lock
);
1244 trace_hrtimer_expire_entry(timer
, now
);
1245 restart
= fn(timer
);
1246 trace_hrtimer_expire_exit(timer
);
1247 raw_spin_lock(&cpu_base
->lock
);
1250 * Note: We clear the CALLBACK bit after enqueue_hrtimer and
1251 * we do not reprogramm the event hardware. Happens either in
1252 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1254 if (restart
!= HRTIMER_NORESTART
) {
1255 BUG_ON(timer
->state
!= HRTIMER_STATE_CALLBACK
);
1256 enqueue_hrtimer(timer
, base
);
1259 WARN_ON_ONCE(!(timer
->state
& HRTIMER_STATE_CALLBACK
));
1261 timer
->state
&= ~HRTIMER_STATE_CALLBACK
;
1264 #ifdef CONFIG_HIGH_RES_TIMERS
1267 * High resolution timer interrupt
1268 * Called with interrupts disabled
1270 void hrtimer_interrupt(struct clock_event_device
*dev
)
1272 struct hrtimer_cpu_base
*cpu_base
= &__get_cpu_var(hrtimer_bases
);
1273 ktime_t expires_next
, now
, entry_time
, delta
;
1276 BUG_ON(!cpu_base
->hres_active
);
1277 cpu_base
->nr_events
++;
1278 dev
->next_event
.tv64
= KTIME_MAX
;
1280 raw_spin_lock(&cpu_base
->lock
);
1281 entry_time
= now
= hrtimer_update_base(cpu_base
);
1283 expires_next
.tv64
= KTIME_MAX
;
1285 * We set expires_next to KTIME_MAX here with cpu_base->lock
1286 * held to prevent that a timer is enqueued in our queue via
1287 * the migration code. This does not affect enqueueing of
1288 * timers which run their callback and need to be requeued on
1291 cpu_base
->expires_next
.tv64
= KTIME_MAX
;
1293 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++) {
1294 struct hrtimer_clock_base
*base
;
1295 struct timerqueue_node
*node
;
1298 if (!(cpu_base
->active_bases
& (1 << i
)))
1301 base
= cpu_base
->clock_base
+ i
;
1302 basenow
= ktime_add(now
, base
->offset
);
1304 while ((node
= timerqueue_getnext(&base
->active
))) {
1305 struct hrtimer
*timer
;
1307 timer
= container_of(node
, struct hrtimer
, node
);
1310 * The immediate goal for using the softexpires is
1311 * minimizing wakeups, not running timers at the
1312 * earliest interrupt after their soft expiration.
1313 * This allows us to avoid using a Priority Search
1314 * Tree, which can answer a stabbing querry for
1315 * overlapping intervals and instead use the simple
1316 * BST we already have.
1317 * We don't add extra wakeups by delaying timers that
1318 * are right-of a not yet expired timer, because that
1319 * timer will have to trigger a wakeup anyway.
1322 if (basenow
.tv64
< hrtimer_get_softexpires_tv64(timer
)) {
1325 expires
= ktime_sub(hrtimer_get_expires(timer
),
1327 if (expires
.tv64
< 0)
1328 expires
.tv64
= KTIME_MAX
;
1329 if (expires
.tv64
< expires_next
.tv64
)
1330 expires_next
= expires
;
1334 __run_hrtimer(timer
, &basenow
);
1339 * Store the new expiry value so the migration code can verify
1342 cpu_base
->expires_next
= expires_next
;
1343 raw_spin_unlock(&cpu_base
->lock
);
1345 /* Reprogramming necessary ? */
1346 if (expires_next
.tv64
== KTIME_MAX
||
1347 !tick_program_event(expires_next
, 0)) {
1348 cpu_base
->hang_detected
= 0;
1353 * The next timer was already expired due to:
1355 * - long lasting callbacks
1356 * - being scheduled away when running in a VM
1358 * We need to prevent that we loop forever in the hrtimer
1359 * interrupt routine. We give it 3 attempts to avoid
1360 * overreacting on some spurious event.
1362 * Acquire base lock for updating the offsets and retrieving
1365 raw_spin_lock(&cpu_base
->lock
);
1366 now
= hrtimer_update_base(cpu_base
);
1367 cpu_base
->nr_retries
++;
1371 * Give the system a chance to do something else than looping
1372 * here. We stored the entry time, so we know exactly how long
1373 * we spent here. We schedule the next event this amount of
1376 cpu_base
->nr_hangs
++;
1377 cpu_base
->hang_detected
= 1;
1378 raw_spin_unlock(&cpu_base
->lock
);
1379 delta
= ktime_sub(now
, entry_time
);
1380 if (delta
.tv64
> cpu_base
->max_hang_time
.tv64
)
1381 cpu_base
->max_hang_time
= delta
;
1383 * Limit it to a sensible value as we enforce a longer
1384 * delay. Give the CPU at least 100ms to catch up.
1386 if (delta
.tv64
> 100 * NSEC_PER_MSEC
)
1387 expires_next
= ktime_add_ns(now
, 100 * NSEC_PER_MSEC
);
1389 expires_next
= ktime_add(now
, delta
);
1390 tick_program_event(expires_next
, 1);
1391 printk_once(KERN_WARNING
"hrtimer: interrupt took %llu ns\n",
1392 ktime_to_ns(delta
));
1396 * local version of hrtimer_peek_ahead_timers() called with interrupts
1399 static void __hrtimer_peek_ahead_timers(void)
1401 struct tick_device
*td
;
1403 if (!hrtimer_hres_active())
1406 td
= &__get_cpu_var(tick_cpu_device
);
1407 if (td
&& td
->evtdev
)
1408 hrtimer_interrupt(td
->evtdev
);
1412 * hrtimer_peek_ahead_timers -- run soft-expired timers now
1414 * hrtimer_peek_ahead_timers will peek at the timer queue of
1415 * the current cpu and check if there are any timers for which
1416 * the soft expires time has passed. If any such timers exist,
1417 * they are run immediately and then removed from the timer queue.
1420 void hrtimer_peek_ahead_timers(void)
1422 unsigned long flags
;
1424 local_irq_save(flags
);
1425 __hrtimer_peek_ahead_timers();
1426 local_irq_restore(flags
);
1429 static void run_hrtimer_softirq(struct softirq_action
*h
)
1431 hrtimer_peek_ahead_timers();
1434 #else /* CONFIG_HIGH_RES_TIMERS */
1436 static inline void __hrtimer_peek_ahead_timers(void) { }
1438 #endif /* !CONFIG_HIGH_RES_TIMERS */
1441 * Called from timer softirq every jiffy, expire hrtimers:
1443 * For HRT its the fall back code to run the softirq in the timer
1444 * softirq context in case the hrtimer initialization failed or has
1445 * not been done yet.
1447 void hrtimer_run_pending(void)
1449 if (hrtimer_hres_active())
1453 * This _is_ ugly: We have to check in the softirq context,
1454 * whether we can switch to highres and / or nohz mode. The
1455 * clocksource switch happens in the timer interrupt with
1456 * xtime_lock held. Notification from there only sets the
1457 * check bit in the tick_oneshot code, otherwise we might
1458 * deadlock vs. xtime_lock.
1460 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1461 hrtimer_switch_to_hres();
1465 * Called from hardirq context every jiffy
1467 void hrtimer_run_queues(void)
1469 struct timerqueue_node
*node
;
1470 struct hrtimer_cpu_base
*cpu_base
= &__get_cpu_var(hrtimer_bases
);
1471 struct hrtimer_clock_base
*base
;
1472 int index
, gettime
= 1;
1474 if (hrtimer_hres_active())
1477 for (index
= 0; index
< HRTIMER_MAX_CLOCK_BASES
; index
++) {
1478 base
= &cpu_base
->clock_base
[index
];
1479 if (!timerqueue_getnext(&base
->active
))
1483 hrtimer_get_softirq_time(cpu_base
);
1487 raw_spin_lock(&cpu_base
->lock
);
1489 while ((node
= timerqueue_getnext(&base
->active
))) {
1490 struct hrtimer
*timer
;
1492 timer
= container_of(node
, struct hrtimer
, node
);
1493 if (base
->softirq_time
.tv64
<=
1494 hrtimer_get_expires_tv64(timer
))
1497 __run_hrtimer(timer
, &base
->softirq_time
);
1499 raw_spin_unlock(&cpu_base
->lock
);
1504 * Sleep related functions:
1506 static enum hrtimer_restart
hrtimer_wakeup(struct hrtimer
*timer
)
1508 struct hrtimer_sleeper
*t
=
1509 container_of(timer
, struct hrtimer_sleeper
, timer
);
1510 struct task_struct
*task
= t
->task
;
1514 wake_up_process(task
);
1516 return HRTIMER_NORESTART
;
1519 void hrtimer_init_sleeper(struct hrtimer_sleeper
*sl
, struct task_struct
*task
)
1521 sl
->timer
.function
= hrtimer_wakeup
;
1524 EXPORT_SYMBOL_GPL(hrtimer_init_sleeper
);
1526 static int __sched
do_nanosleep(struct hrtimer_sleeper
*t
, enum hrtimer_mode mode
)
1528 hrtimer_init_sleeper(t
, current
);
1531 set_current_state(TASK_INTERRUPTIBLE
);
1532 hrtimer_start_expires(&t
->timer
, mode
);
1533 if (!hrtimer_active(&t
->timer
))
1536 if (likely(t
->task
))
1537 freezable_schedule();
1539 hrtimer_cancel(&t
->timer
);
1540 mode
= HRTIMER_MODE_ABS
;
1542 } while (t
->task
&& !signal_pending(current
));
1544 __set_current_state(TASK_RUNNING
);
1546 return t
->task
== NULL
;
1549 static int update_rmtp(struct hrtimer
*timer
, struct timespec __user
*rmtp
)
1551 struct timespec rmt
;
1554 rem
= hrtimer_expires_remaining(timer
);
1557 rmt
= ktime_to_timespec(rem
);
1559 if (copy_to_user(rmtp
, &rmt
, sizeof(*rmtp
)))
1565 long __sched
hrtimer_nanosleep_restart(struct restart_block
*restart
)
1567 struct hrtimer_sleeper t
;
1568 struct timespec __user
*rmtp
;
1571 hrtimer_init_on_stack(&t
.timer
, restart
->nanosleep
.clockid
,
1573 hrtimer_set_expires_tv64(&t
.timer
, restart
->nanosleep
.expires
);
1575 if (do_nanosleep(&t
, HRTIMER_MODE_ABS
))
1578 rmtp
= restart
->nanosleep
.rmtp
;
1580 ret
= update_rmtp(&t
.timer
, rmtp
);
1585 /* The other values in restart are already filled in */
1586 ret
= -ERESTART_RESTARTBLOCK
;
1588 destroy_hrtimer_on_stack(&t
.timer
);
1592 long hrtimer_nanosleep(struct timespec
*rqtp
, struct timespec __user
*rmtp
,
1593 const enum hrtimer_mode mode
, const clockid_t clockid
)
1595 struct restart_block
*restart
;
1596 struct hrtimer_sleeper t
;
1598 unsigned long slack
;
1600 slack
= current
->timer_slack_ns
;
1601 if (dl_task(current
) || rt_task(current
))
1604 hrtimer_init_on_stack(&t
.timer
, clockid
, mode
);
1605 hrtimer_set_expires_range_ns(&t
.timer
, timespec_to_ktime(*rqtp
), slack
);
1606 if (do_nanosleep(&t
, mode
))
1609 /* Absolute timers do not update the rmtp value and restart: */
1610 if (mode
== HRTIMER_MODE_ABS
) {
1611 ret
= -ERESTARTNOHAND
;
1616 ret
= update_rmtp(&t
.timer
, rmtp
);
1621 restart
= ¤t_thread_info()->restart_block
;
1622 restart
->fn
= hrtimer_nanosleep_restart
;
1623 restart
->nanosleep
.clockid
= t
.timer
.base
->clockid
;
1624 restart
->nanosleep
.rmtp
= rmtp
;
1625 restart
->nanosleep
.expires
= hrtimer_get_expires_tv64(&t
.timer
);
1627 ret
= -ERESTART_RESTARTBLOCK
;
1629 destroy_hrtimer_on_stack(&t
.timer
);
1633 SYSCALL_DEFINE2(nanosleep
, struct timespec __user
*, rqtp
,
1634 struct timespec __user
*, rmtp
)
1638 if (copy_from_user(&tu
, rqtp
, sizeof(tu
)))
1641 if (!timespec_valid(&tu
))
1644 return hrtimer_nanosleep(&tu
, rmtp
, HRTIMER_MODE_REL
, CLOCK_MONOTONIC
);
1648 * Functions related to boot-time initialization:
1650 static void init_hrtimers_cpu(int cpu
)
1652 struct hrtimer_cpu_base
*cpu_base
= &per_cpu(hrtimer_bases
, cpu
);
1655 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++) {
1656 cpu_base
->clock_base
[i
].cpu_base
= cpu_base
;
1657 timerqueue_init_head(&cpu_base
->clock_base
[i
].active
);
1660 hrtimer_init_hres(cpu_base
);
1663 #ifdef CONFIG_HOTPLUG_CPU
1665 static void migrate_hrtimer_list(struct hrtimer_clock_base
*old_base
,
1666 struct hrtimer_clock_base
*new_base
)
1668 struct hrtimer
*timer
;
1669 struct timerqueue_node
*node
;
1671 while ((node
= timerqueue_getnext(&old_base
->active
))) {
1672 timer
= container_of(node
, struct hrtimer
, node
);
1673 BUG_ON(hrtimer_callback_running(timer
));
1674 debug_deactivate(timer
);
1677 * Mark it as STATE_MIGRATE not INACTIVE otherwise the
1678 * timer could be seen as !active and just vanish away
1679 * under us on another CPU
1681 __remove_hrtimer(timer
, old_base
, HRTIMER_STATE_MIGRATE
, 0);
1682 timer
->base
= new_base
;
1684 * Enqueue the timers on the new cpu. This does not
1685 * reprogram the event device in case the timer
1686 * expires before the earliest on this CPU, but we run
1687 * hrtimer_interrupt after we migrated everything to
1688 * sort out already expired timers and reprogram the
1691 enqueue_hrtimer(timer
, new_base
);
1693 /* Clear the migration state bit */
1694 timer
->state
&= ~HRTIMER_STATE_MIGRATE
;
1698 static void migrate_hrtimers(int scpu
)
1700 struct hrtimer_cpu_base
*old_base
, *new_base
;
1703 BUG_ON(cpu_online(scpu
));
1704 tick_cancel_sched_timer(scpu
);
1706 local_irq_disable();
1707 old_base
= &per_cpu(hrtimer_bases
, scpu
);
1708 new_base
= &__get_cpu_var(hrtimer_bases
);
1710 * The caller is globally serialized and nobody else
1711 * takes two locks at once, deadlock is not possible.
1713 raw_spin_lock(&new_base
->lock
);
1714 raw_spin_lock_nested(&old_base
->lock
, SINGLE_DEPTH_NESTING
);
1716 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++) {
1717 migrate_hrtimer_list(&old_base
->clock_base
[i
],
1718 &new_base
->clock_base
[i
]);
1721 raw_spin_unlock(&old_base
->lock
);
1722 raw_spin_unlock(&new_base
->lock
);
1724 /* Check, if we got expired work to do */
1725 __hrtimer_peek_ahead_timers();
1729 #endif /* CONFIG_HOTPLUG_CPU */
1731 static int hrtimer_cpu_notify(struct notifier_block
*self
,
1732 unsigned long action
, void *hcpu
)
1734 int scpu
= (long)hcpu
;
1738 case CPU_UP_PREPARE
:
1739 case CPU_UP_PREPARE_FROZEN
:
1740 init_hrtimers_cpu(scpu
);
1743 #ifdef CONFIG_HOTPLUG_CPU
1745 case CPU_DYING_FROZEN
:
1746 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DYING
, &scpu
);
1749 case CPU_DEAD_FROZEN
:
1751 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD
, &scpu
);
1752 migrate_hrtimers(scpu
);
1764 static struct notifier_block hrtimers_nb
= {
1765 .notifier_call
= hrtimer_cpu_notify
,
1768 void __init
hrtimers_init(void)
1770 hrtimer_cpu_notify(&hrtimers_nb
, (unsigned long)CPU_UP_PREPARE
,
1771 (void *)(long)smp_processor_id());
1772 register_cpu_notifier(&hrtimers_nb
);
1773 #ifdef CONFIG_HIGH_RES_TIMERS
1774 open_softirq(HRTIMER_SOFTIRQ
, run_hrtimer_softirq
);
1779 * schedule_hrtimeout_range_clock - sleep until timeout
1780 * @expires: timeout value (ktime_t)
1781 * @delta: slack in expires timeout (ktime_t)
1782 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1783 * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME
1786 schedule_hrtimeout_range_clock(ktime_t
*expires
, unsigned long delta
,
1787 const enum hrtimer_mode mode
, int clock
)
1789 struct hrtimer_sleeper t
;
1792 * Optimize when a zero timeout value is given. It does not
1793 * matter whether this is an absolute or a relative time.
1795 if (expires
&& !expires
->tv64
) {
1796 __set_current_state(TASK_RUNNING
);
1801 * A NULL parameter means "infinite"
1805 __set_current_state(TASK_RUNNING
);
1809 hrtimer_init_on_stack(&t
.timer
, clock
, mode
);
1810 hrtimer_set_expires_range_ns(&t
.timer
, *expires
, delta
);
1812 hrtimer_init_sleeper(&t
, current
);
1814 hrtimer_start_expires(&t
.timer
, mode
);
1815 if (!hrtimer_active(&t
.timer
))
1821 hrtimer_cancel(&t
.timer
);
1822 destroy_hrtimer_on_stack(&t
.timer
);
1824 __set_current_state(TASK_RUNNING
);
1826 return !t
.task
? 0 : -EINTR
;
1830 * schedule_hrtimeout_range - sleep until timeout
1831 * @expires: timeout value (ktime_t)
1832 * @delta: slack in expires timeout (ktime_t)
1833 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1835 * Make the current task sleep until the given expiry time has
1836 * elapsed. The routine will return immediately unless
1837 * the current task state has been set (see set_current_state()).
1839 * The @delta argument gives the kernel the freedom to schedule the
1840 * actual wakeup to a time that is both power and performance friendly.
1841 * The kernel give the normal best effort behavior for "@expires+@delta",
1842 * but may decide to fire the timer earlier, but no earlier than @expires.
1844 * You can set the task state as follows -
1846 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1847 * pass before the routine returns.
1849 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1850 * delivered to the current task.
1852 * The current task state is guaranteed to be TASK_RUNNING when this
1855 * Returns 0 when the timer has expired otherwise -EINTR
1857 int __sched
schedule_hrtimeout_range(ktime_t
*expires
, unsigned long delta
,
1858 const enum hrtimer_mode mode
)
1860 return schedule_hrtimeout_range_clock(expires
, delta
, mode
,
1863 EXPORT_SYMBOL_GPL(schedule_hrtimeout_range
);
1866 * schedule_hrtimeout - sleep until timeout
1867 * @expires: timeout value (ktime_t)
1868 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1870 * Make the current task sleep until the given expiry time has
1871 * elapsed. The routine will return immediately unless
1872 * the current task state has been set (see set_current_state()).
1874 * You can set the task state as follows -
1876 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1877 * pass before the routine returns.
1879 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1880 * delivered to the current task.
1882 * The current task state is guaranteed to be TASK_RUNNING when this
1885 * Returns 0 when the timer has expired otherwise -EINTR
1887 int __sched
schedule_hrtimeout(ktime_t
*expires
,
1888 const enum hrtimer_mode mode
)
1890 return schedule_hrtimeout_range(expires
, 0, mode
);
1892 EXPORT_SYMBOL_GPL(schedule_hrtimeout
);