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/timer.h>
49 #include <asm/uaccess.h>
51 #include <trace/events/timer.h>
56 * There are more clockids then hrtimer bases. Thus, we index
57 * into the timer bases by the hrtimer_base_type enum. When trying
58 * to reach a base using a clockid, hrtimer_clockid_to_base()
59 * is used to convert from clockid to the proper hrtimer_base_type.
61 DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
64 .lock = __RAW_SPIN_LOCK_UNLOCKED(hrtimer_bases.lock),
68 .index = HRTIMER_BASE_MONOTONIC,
69 .clockid = CLOCK_MONOTONIC,
70 .get_time = &ktime_get,
71 .resolution = KTIME_LOW_RES,
74 .index = HRTIMER_BASE_REALTIME,
75 .clockid = CLOCK_REALTIME,
76 .get_time = &ktime_get_real,
77 .resolution = KTIME_LOW_RES,
80 .index = HRTIMER_BASE_BOOTTIME,
81 .clockid = CLOCK_BOOTTIME,
82 .get_time = &ktime_get_boottime,
83 .resolution = KTIME_LOW_RES,
88 static const int hrtimer_clock_to_base_table[MAX_CLOCKS] = {
89 [CLOCK_REALTIME] = HRTIMER_BASE_REALTIME,
90 [CLOCK_MONOTONIC] = HRTIMER_BASE_MONOTONIC,
91 [CLOCK_BOOTTIME] = HRTIMER_BASE_BOOTTIME,
94 static inline int hrtimer_clockid_to_base(clockid_t clock_id)
96 return hrtimer_clock_to_base_table[clock_id];
101 * Get the coarse grained time at the softirq based on xtime and
104 static void hrtimer_get_softirq_time(struct hrtimer_cpu_base *base)
106 ktime_t xtim, mono, boot;
107 struct timespec xts, tom, slp;
109 get_xtime_and_monotonic_and_sleep_offset(&xts, &tom, &slp);
111 xtim = timespec_to_ktime(xts);
112 mono = ktime_add(xtim, timespec_to_ktime(tom));
113 boot = ktime_add(mono, timespec_to_ktime(slp));
114 base->clock_base[HRTIMER_BASE_REALTIME].softirq_time = xtim;
115 base->clock_base[HRTIMER_BASE_MONOTONIC].softirq_time = mono;
116 base->clock_base[HRTIMER_BASE_BOOTTIME].softirq_time = boot;
120 * Functions and macros which are different for UP/SMP systems are kept in a
126 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
127 * means that all timers which are tied to this base via timer->base are
128 * locked, and the base itself is locked too.
130 * So __run_timers/migrate_timers can safely modify all timers which could
131 * be found on the lists/queues.
133 * When the timer's base is locked, and the timer removed from list, it is
134 * possible to set timer->base = NULL and drop the lock: the timer remains
138 struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
139 unsigned long *flags)
141 struct hrtimer_clock_base *base;
145 if (likely(base != NULL)) {
146 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
147 if (likely(base == timer->base))
149 /* The timer has migrated to another CPU: */
150 raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
158 * Get the preferred target CPU for NOHZ
160 static int hrtimer_get_target(int this_cpu, int pinned)
163 if (!pinned && get_sysctl_timer_migration() && idle_cpu(this_cpu))
164 return get_nohz_timer_target();
170 * With HIGHRES=y we do not migrate the timer when it is expiring
171 * before the next event on the target cpu because we cannot reprogram
172 * the target cpu hardware and we would cause it to fire late.
174 * Called with cpu_base->lock of target cpu held.
177 hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base)
179 #ifdef CONFIG_HIGH_RES_TIMERS
182 if (!new_base->cpu_base->hres_active)
185 expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset);
186 return expires.tv64 <= new_base->cpu_base->expires_next.tv64;
193 * Switch the timer base to the current CPU when possible.
195 static inline struct hrtimer_clock_base *
196 switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
199 struct hrtimer_clock_base *new_base;
200 struct hrtimer_cpu_base *new_cpu_base;
201 int this_cpu = smp_processor_id();
202 int cpu = hrtimer_get_target(this_cpu, pinned);
203 int basenum = base->index;
206 new_cpu_base = &per_cpu(hrtimer_bases, cpu);
207 new_base = &new_cpu_base->clock_base[basenum];
209 if (base != new_base) {
211 * We are trying to move timer to new_base.
212 * However we can't change timer's base while it is running,
213 * so we keep it on the same CPU. No hassle vs. reprogramming
214 * the event source in the high resolution case. The softirq
215 * code will take care of this when the timer function has
216 * completed. There is no conflict as we hold the lock until
217 * the timer is enqueued.
219 if (unlikely(hrtimer_callback_running(timer)))
222 /* See the comment in lock_timer_base() */
224 raw_spin_unlock(&base->cpu_base->lock);
225 raw_spin_lock(&new_base->cpu_base->lock);
227 if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) {
229 raw_spin_unlock(&new_base->cpu_base->lock);
230 raw_spin_lock(&base->cpu_base->lock);
234 timer->base = new_base;
236 if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) {
244 #else /* CONFIG_SMP */
246 static inline struct hrtimer_clock_base *
247 lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
249 struct hrtimer_clock_base *base = timer->base;
251 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
256 # define switch_hrtimer_base(t, b, p) (b)
258 #endif /* !CONFIG_SMP */
261 * Functions for the union type storage format of ktime_t which are
262 * too large for inlining:
264 #if BITS_PER_LONG < 64
265 # ifndef CONFIG_KTIME_SCALAR
267 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
269 * @nsec: the scalar nsec value to add
271 * Returns the sum of kt and nsec in ktime_t format
273 ktime_t ktime_add_ns(const ktime_t kt, u64 nsec)
277 if (likely(nsec < NSEC_PER_SEC)) {
280 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
282 tmp = ktime_set((long)nsec, rem);
285 return ktime_add(kt, tmp);
288 EXPORT_SYMBOL_GPL(ktime_add_ns);
291 * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable
293 * @nsec: the scalar nsec value to subtract
295 * Returns the subtraction of @nsec from @kt in ktime_t format
297 ktime_t ktime_sub_ns(const ktime_t kt, u64 nsec)
301 if (likely(nsec < NSEC_PER_SEC)) {
304 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
306 /* Make sure nsec fits into long */
307 if (unlikely(nsec > KTIME_SEC_MAX))
308 return (ktime_t){ .tv64 = KTIME_MAX };
310 tmp = ktime_set((long)nsec, rem);
313 return ktime_sub(kt, tmp);
316 EXPORT_SYMBOL_GPL(ktime_sub_ns);
317 # endif /* !CONFIG_KTIME_SCALAR */
320 * Divide a ktime value by a nanosecond value
322 u64 ktime_divns(const ktime_t kt, s64 div)
327 dclc = ktime_to_ns(kt);
328 /* Make sure the divisor is less than 2^32: */
334 do_div(dclc, (unsigned long) div);
338 #endif /* BITS_PER_LONG >= 64 */
341 * Add two ktime values and do a safety check for overflow:
343 ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
345 ktime_t res = ktime_add(lhs, rhs);
348 * We use KTIME_SEC_MAX here, the maximum timeout which we can
349 * return to user space in a timespec:
351 if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64)
352 res = ktime_set(KTIME_SEC_MAX, 0);
357 EXPORT_SYMBOL_GPL(ktime_add_safe);
359 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
361 static struct debug_obj_descr hrtimer_debug_descr;
363 static void *hrtimer_debug_hint(void *addr)
365 return ((struct hrtimer *) addr)->function;
369 * fixup_init is called when:
370 * - an active object is initialized
372 static int hrtimer_fixup_init(void *addr, enum debug_obj_state state)
374 struct hrtimer *timer = addr;
377 case ODEBUG_STATE_ACTIVE:
378 hrtimer_cancel(timer);
379 debug_object_init(timer, &hrtimer_debug_descr);
387 * fixup_activate is called when:
388 * - an active object is activated
389 * - an unknown object is activated (might be a statically initialized object)
391 static int hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
395 case ODEBUG_STATE_NOTAVAILABLE:
399 case ODEBUG_STATE_ACTIVE:
408 * fixup_free is called when:
409 * - an active object is freed
411 static int hrtimer_fixup_free(void *addr, enum debug_obj_state state)
413 struct hrtimer *timer = addr;
416 case ODEBUG_STATE_ACTIVE:
417 hrtimer_cancel(timer);
418 debug_object_free(timer, &hrtimer_debug_descr);
425 static struct debug_obj_descr hrtimer_debug_descr = {
427 .debug_hint = hrtimer_debug_hint,
428 .fixup_init = hrtimer_fixup_init,
429 .fixup_activate = hrtimer_fixup_activate,
430 .fixup_free = hrtimer_fixup_free,
433 static inline void debug_hrtimer_init(struct hrtimer *timer)
435 debug_object_init(timer, &hrtimer_debug_descr);
438 static inline void debug_hrtimer_activate(struct hrtimer *timer)
440 debug_object_activate(timer, &hrtimer_debug_descr);
443 static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
445 debug_object_deactivate(timer, &hrtimer_debug_descr);
448 static inline void debug_hrtimer_free(struct hrtimer *timer)
450 debug_object_free(timer, &hrtimer_debug_descr);
453 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
454 enum hrtimer_mode mode);
456 void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
457 enum hrtimer_mode mode)
459 debug_object_init_on_stack(timer, &hrtimer_debug_descr);
460 __hrtimer_init(timer, clock_id, mode);
462 EXPORT_SYMBOL_GPL(hrtimer_init_on_stack);
464 void destroy_hrtimer_on_stack(struct hrtimer *timer)
466 debug_object_free(timer, &hrtimer_debug_descr);
470 static inline void debug_hrtimer_init(struct hrtimer *timer) { }
471 static inline void debug_hrtimer_activate(struct hrtimer *timer) { }
472 static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
476 debug_init(struct hrtimer *timer, clockid_t clockid,
477 enum hrtimer_mode mode)
479 debug_hrtimer_init(timer);
480 trace_hrtimer_init(timer, clockid, mode);
483 static inline void debug_activate(struct hrtimer *timer)
485 debug_hrtimer_activate(timer);
486 trace_hrtimer_start(timer);
489 static inline void debug_deactivate(struct hrtimer *timer)
491 debug_hrtimer_deactivate(timer);
492 trace_hrtimer_cancel(timer);
495 /* High resolution timer related functions */
496 #ifdef CONFIG_HIGH_RES_TIMERS
499 * High resolution timer enabled ?
501 static int hrtimer_hres_enabled __read_mostly = 1;
504 * Enable / Disable high resolution mode
506 static int __init setup_hrtimer_hres(char *str)
508 if (!strcmp(str, "off"))
509 hrtimer_hres_enabled = 0;
510 else if (!strcmp(str, "on"))
511 hrtimer_hres_enabled = 1;
517 __setup("highres=", setup_hrtimer_hres);
520 * hrtimer_high_res_enabled - query, if the highres mode is enabled
522 static inline int hrtimer_is_hres_enabled(void)
524 return hrtimer_hres_enabled;
528 * Is the high resolution mode active ?
530 static inline int hrtimer_hres_active(void)
532 return __this_cpu_read(hrtimer_bases.hres_active);
536 * Reprogram the event source with checking both queues for the
538 * Called with interrupts disabled and base->lock held
541 hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
544 struct hrtimer_clock_base *base = cpu_base->clock_base;
545 ktime_t expires, expires_next;
547 expires_next.tv64 = KTIME_MAX;
549 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
550 struct hrtimer *timer;
551 struct timerqueue_node *next;
553 next = timerqueue_getnext(&base->active);
556 timer = container_of(next, struct hrtimer, node);
558 expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
560 * clock_was_set() has changed base->offset so the
561 * result might be negative. Fix it up to prevent a
562 * false positive in clockevents_program_event()
564 if (expires.tv64 < 0)
566 if (expires.tv64 < expires_next.tv64)
567 expires_next = expires;
570 if (skip_equal && expires_next.tv64 == cpu_base->expires_next.tv64)
573 cpu_base->expires_next.tv64 = expires_next.tv64;
576 * If a hang was detected in the last timer interrupt then we
577 * leave the hang delay active in the hardware. We want the
578 * system to make progress. That also prevents the following
580 * T1 expires 50ms from now
581 * T2 expires 5s from now
583 * T1 is removed, so this code is called and would reprogram
584 * the hardware to 5s from now. Any hrtimer_start after that
585 * will not reprogram the hardware due to hang_detected being
586 * set. So we'd effectivly block all timers until the T2 event
589 if (cpu_base->hang_detected)
592 if (cpu_base->expires_next.tv64 != KTIME_MAX)
593 tick_program_event(cpu_base->expires_next, 1);
597 * Shared reprogramming for clock_realtime and clock_monotonic
599 * When a timer is enqueued and expires earlier than the already enqueued
600 * timers, we have to check, whether it expires earlier than the timer for
601 * which the clock event device was armed.
603 * Called with interrupts disabled and base->cpu_base.lock held
605 static int hrtimer_reprogram(struct hrtimer *timer,
606 struct hrtimer_clock_base *base)
608 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
609 ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
612 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
615 * When the callback is running, we do not reprogram the clock event
616 * device. The timer callback is either running on a different CPU or
617 * the callback is executed in the hrtimer_interrupt context. The
618 * reprogramming is handled either by the softirq, which called the
619 * callback or at the end of the hrtimer_interrupt.
621 if (hrtimer_callback_running(timer))
625 * CLOCK_REALTIME timer might be requested with an absolute
626 * expiry time which is less than base->offset. Nothing wrong
627 * about that, just avoid to call into the tick code, which
628 * has now objections against negative expiry values.
630 if (expires.tv64 < 0)
633 if (expires.tv64 >= cpu_base->expires_next.tv64)
637 * If a hang was detected in the last timer interrupt then we
638 * do not schedule a timer which is earlier than the expiry
639 * which we enforced in the hang detection. We want the system
642 if (cpu_base->hang_detected)
646 * Clockevents returns -ETIME, when the event was in the past.
648 res = tick_program_event(expires, 0);
649 if (!IS_ERR_VALUE(res))
650 cpu_base->expires_next = expires;
655 * Initialize the high resolution related parts of cpu_base
657 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
659 base->expires_next.tv64 = KTIME_MAX;
660 base->hang_detected = 0;
661 base->hres_active = 0;
665 * When High resolution timers are active, try to reprogram. Note, that in case
666 * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry
667 * check happens. The timer gets enqueued into the rbtree. The reprogramming
668 * and expiry check is done in the hrtimer_interrupt or in the softirq.
670 static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
671 struct hrtimer_clock_base *base)
673 return base->cpu_base->hres_active && hrtimer_reprogram(timer, base);
676 static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base)
678 ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset;
679 ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset;
681 return ktime_get_update_offsets(offs_real, offs_boot);
685 * Retrigger next event is called after clock was set
687 * Called with interrupts disabled via on_each_cpu()
689 static void retrigger_next_event(void *arg)
691 struct hrtimer_cpu_base *base = &__get_cpu_var(hrtimer_bases);
693 if (!hrtimer_hres_active())
696 raw_spin_lock(&base->lock);
697 hrtimer_update_base(base);
698 hrtimer_force_reprogram(base, 0);
699 raw_spin_unlock(&base->lock);
703 * Switch to high resolution mode
705 static int hrtimer_switch_to_hres(void)
707 int i, cpu = smp_processor_id();
708 struct hrtimer_cpu_base *base = &per_cpu(hrtimer_bases, cpu);
711 if (base->hres_active)
714 local_irq_save(flags);
716 if (tick_init_highres()) {
717 local_irq_restore(flags);
718 printk(KERN_WARNING "Could not switch to high resolution "
719 "mode on CPU %d\n", cpu);
722 base->hres_active = 1;
723 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
724 base->clock_base[i].resolution = KTIME_HIGH_RES;
726 tick_setup_sched_timer();
727 /* "Retrigger" the interrupt to get things going */
728 retrigger_next_event(NULL);
729 local_irq_restore(flags);
734 * Called from timekeeping code to reprogramm the hrtimer interrupt
735 * device. If called from the timer interrupt context we defer it to
738 void clock_was_set_delayed(void)
740 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
742 cpu_base->clock_was_set = 1;
743 __raise_softirq_irqoff(HRTIMER_SOFTIRQ);
748 static inline int hrtimer_hres_active(void) { return 0; }
749 static inline int hrtimer_is_hres_enabled(void) { return 0; }
750 static inline int hrtimer_switch_to_hres(void) { return 0; }
752 hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { }
753 static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
754 struct hrtimer_clock_base *base)
758 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { }
759 static inline void retrigger_next_event(void *arg) { }
761 #endif /* CONFIG_HIGH_RES_TIMERS */
764 * Clock realtime was set
766 * Change the offset of the realtime clock vs. the monotonic
769 * We might have to reprogram the high resolution timer interrupt. On
770 * SMP we call the architecture specific code to retrigger _all_ high
771 * resolution timer interrupts. On UP we just disable interrupts and
772 * call the high resolution interrupt code.
774 void clock_was_set(void)
776 #ifdef CONFIG_HIGH_RES_TIMERS
777 /* Retrigger the CPU local events everywhere */
778 on_each_cpu(retrigger_next_event, NULL, 1);
780 timerfd_clock_was_set();
784 * During resume we might have to reprogram the high resolution timer
785 * interrupt (on the local CPU):
787 void hrtimers_resume(void)
789 WARN_ONCE(!irqs_disabled(),
790 KERN_INFO "hrtimers_resume() called with IRQs enabled!");
792 retrigger_next_event(NULL);
793 timerfd_clock_was_set();
796 static inline void timer_stats_hrtimer_set_start_info(struct hrtimer *timer)
798 #ifdef CONFIG_TIMER_STATS
799 if (timer->start_site)
801 timer->start_site = __builtin_return_address(0);
802 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
803 timer->start_pid = current->pid;
807 static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer *timer)
809 #ifdef CONFIG_TIMER_STATS
810 timer->start_site = NULL;
814 static inline void timer_stats_account_hrtimer(struct hrtimer *timer)
816 #ifdef CONFIG_TIMER_STATS
817 if (likely(!timer_stats_active))
819 timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
820 timer->function, timer->start_comm, 0);
825 * Counterpart to lock_hrtimer_base above:
828 void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
830 raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
834 * hrtimer_forward - forward the timer expiry
835 * @timer: hrtimer to forward
836 * @now: forward past this time
837 * @interval: the interval to forward
839 * Forward the timer expiry so it will expire in the future.
840 * Returns the number of overruns.
842 u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
847 delta = ktime_sub(now, hrtimer_get_expires(timer));
852 if (WARN_ON(timer->state & HRTIMER_STATE_ENQUEUED))
855 if (interval.tv64 < timer->base->resolution.tv64)
856 interval.tv64 = timer->base->resolution.tv64;
858 if (unlikely(delta.tv64 >= interval.tv64)) {
859 s64 incr = ktime_to_ns(interval);
861 orun = ktime_divns(delta, incr);
862 hrtimer_add_expires_ns(timer, incr * orun);
863 if (hrtimer_get_expires_tv64(timer) > now.tv64)
866 * This (and the ktime_add() below) is the
867 * correction for exact:
871 hrtimer_add_expires(timer, interval);
875 EXPORT_SYMBOL_GPL(hrtimer_forward);
878 * enqueue_hrtimer - internal function to (re)start a timer
880 * The timer is inserted in expiry order. Insertion into the
881 * red black tree is O(log(n)). Must hold the base lock.
883 * Returns 1 when the new timer is the leftmost timer in the tree.
885 static int enqueue_hrtimer(struct hrtimer *timer,
886 struct hrtimer_clock_base *base)
888 debug_activate(timer);
890 timerqueue_add(&base->active, &timer->node);
891 base->cpu_base->active_bases |= 1 << base->index;
894 * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
895 * state of a possibly running callback.
897 timer->state |= HRTIMER_STATE_ENQUEUED;
899 return (&timer->node == base->active.next);
903 * __remove_hrtimer - internal function to remove a timer
905 * Caller must hold the base lock.
907 * High resolution timer mode reprograms the clock event device when the
908 * timer is the one which expires next. The caller can disable this by setting
909 * reprogram to zero. This is useful, when the context does a reprogramming
910 * anyway (e.g. timer interrupt)
912 static void __remove_hrtimer(struct hrtimer *timer,
913 struct hrtimer_clock_base *base,
914 u8 newstate, int reprogram)
916 struct timerqueue_node *next_timer;
917 if (!(timer->state & HRTIMER_STATE_ENQUEUED))
920 next_timer = timerqueue_getnext(&base->active);
921 timerqueue_del(&base->active, &timer->node);
922 if (&timer->node == next_timer) {
923 #ifdef CONFIG_HIGH_RES_TIMERS
924 /* Reprogram the clock event device. if enabled */
925 if (reprogram && hrtimer_hres_active()) {
928 expires = ktime_sub(hrtimer_get_expires(timer),
930 if (base->cpu_base->expires_next.tv64 == expires.tv64)
931 hrtimer_force_reprogram(base->cpu_base, 1);
935 if (!timerqueue_getnext(&base->active))
936 base->cpu_base->active_bases &= ~(1 << base->index);
938 timer->state = newstate;
942 * remove hrtimer, called with base lock held
945 remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base)
947 if (hrtimer_is_queued(timer)) {
952 * Remove the timer and force reprogramming when high
953 * resolution mode is active and the timer is on the current
954 * CPU. If we remove a timer on another CPU, reprogramming is
955 * skipped. The interrupt event on this CPU is fired and
956 * reprogramming happens in the interrupt handler. This is a
957 * rare case and less expensive than a smp call.
959 debug_deactivate(timer);
960 timer_stats_hrtimer_clear_start_info(timer);
961 reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases);
963 * We must preserve the CALLBACK state flag here,
964 * otherwise we could move the timer base in
965 * switch_hrtimer_base.
967 state = timer->state & HRTIMER_STATE_CALLBACK;
968 __remove_hrtimer(timer, base, state, reprogram);
974 static inline ktime_t hrtimer_update_lowres(struct hrtimer *timer, ktime_t tim,
975 const enum hrtimer_mode mode)
977 #ifdef CONFIG_TIME_LOW_RES
979 * CONFIG_TIME_LOW_RES indicates that the system has no way to return
980 * granular time values. For relative timers we add KTIME_LOW_RES
981 * (i.e. one jiffie) to prevent short timeouts.
983 timer->is_rel = mode & HRTIMER_MODE_REL;
985 tim = ktime_add_safe(tim, KTIME_LOW_RES);
990 int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
991 unsigned long delta_ns, const enum hrtimer_mode mode,
994 struct hrtimer_clock_base *base, *new_base;
998 base = lock_hrtimer_base(timer, &flags);
1000 /* Remove an active timer from the queue: */
1001 ret = remove_hrtimer(timer, base);
1003 if (mode & HRTIMER_MODE_REL)
1004 tim = ktime_add_safe(tim, base->get_time());
1006 tim = hrtimer_update_lowres(timer, tim, mode);
1008 hrtimer_set_expires_range_ns(timer, tim, delta_ns);
1010 /* Switch the timer base, if necessary: */
1011 new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED);
1013 timer_stats_hrtimer_set_start_info(timer);
1015 leftmost = enqueue_hrtimer(timer, new_base);
1018 * Only allow reprogramming if the new base is on this CPU.
1019 * (it might still be on another CPU if the timer was pending)
1021 * XXX send_remote_softirq() ?
1023 if (leftmost && new_base->cpu_base == &__get_cpu_var(hrtimer_bases)
1024 && hrtimer_enqueue_reprogram(timer, new_base)) {
1027 * We need to drop cpu_base->lock to avoid a
1028 * lock ordering issue vs. rq->lock.
1030 raw_spin_unlock(&new_base->cpu_base->lock);
1031 raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1032 local_irq_restore(flags);
1035 __raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1039 unlock_hrtimer_base(timer, &flags);
1045 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
1046 * @timer: the timer to be added
1048 * @delta_ns: "slack" range for the timer
1049 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
1053 * 1 when the timer was active
1055 int hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1056 unsigned long delta_ns, const enum hrtimer_mode mode)
1058 return __hrtimer_start_range_ns(timer, tim, delta_ns, mode, 1);
1060 EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
1063 * hrtimer_start - (re)start an hrtimer on the current CPU
1064 * @timer: the timer to be added
1066 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
1070 * 1 when the timer was active
1073 hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode)
1075 return __hrtimer_start_range_ns(timer, tim, 0, mode, 1);
1077 EXPORT_SYMBOL_GPL(hrtimer_start);
1081 * hrtimer_try_to_cancel - try to deactivate a timer
1082 * @timer: hrtimer to stop
1085 * 0 when the timer was not active
1086 * 1 when the timer was active
1087 * -1 when the timer is currently excuting the callback function and
1090 int hrtimer_try_to_cancel(struct hrtimer *timer)
1092 struct hrtimer_clock_base *base;
1093 unsigned long flags;
1096 base = lock_hrtimer_base(timer, &flags);
1098 if (!hrtimer_callback_running(timer))
1099 ret = remove_hrtimer(timer, base);
1101 unlock_hrtimer_base(timer, &flags);
1106 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1109 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1110 * @timer: the timer to be cancelled
1113 * 0 when the timer was not active
1114 * 1 when the timer was active
1116 int hrtimer_cancel(struct hrtimer *timer)
1119 int ret = hrtimer_try_to_cancel(timer);
1126 EXPORT_SYMBOL_GPL(hrtimer_cancel);
1129 * hrtimer_get_remaining - get remaining time for the timer
1130 * @timer: the timer to read
1131 * @adjust: adjust relative timers when CONFIG_TIME_LOW_RES=y
1133 ktime_t __hrtimer_get_remaining(const struct hrtimer *timer, bool adjust)
1135 unsigned long flags;
1138 lock_hrtimer_base(timer, &flags);
1139 #ifdef CONFIG_TIME_LOW_RES
1141 rem = hrtimer_expires_remaining_adjusted(timer);
1144 rem = hrtimer_expires_remaining(timer);
1145 unlock_hrtimer_base(timer, &flags);
1149 EXPORT_SYMBOL_GPL(__hrtimer_get_remaining);
1153 * hrtimer_get_next_event - get the time until next expiry event
1155 * Returns the delta to the next expiry event or KTIME_MAX if no timer
1158 ktime_t hrtimer_get_next_event(void)
1160 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1161 struct hrtimer_clock_base *base = cpu_base->clock_base;
1162 ktime_t delta, mindelta = { .tv64 = KTIME_MAX };
1163 unsigned long flags;
1166 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1168 if (!hrtimer_hres_active()) {
1169 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
1170 struct hrtimer *timer;
1171 struct timerqueue_node *next;
1173 next = timerqueue_getnext(&base->active);
1177 timer = container_of(next, struct hrtimer, node);
1178 delta.tv64 = hrtimer_get_expires_tv64(timer);
1179 delta = ktime_sub(delta, base->get_time());
1180 if (delta.tv64 < mindelta.tv64)
1181 mindelta.tv64 = delta.tv64;
1185 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1187 if (mindelta.tv64 < 0)
1193 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1194 enum hrtimer_mode mode)
1196 struct hrtimer_cpu_base *cpu_base;
1199 memset(timer, 0, sizeof(struct hrtimer));
1201 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1204 * POSIX magic: Relative CLOCK_REALTIME timers are not affected by
1205 * clock modifications, so they needs to become CLOCK_MONOTONIC to
1206 * ensure POSIX compliance.
1208 if (clock_id == CLOCK_REALTIME && mode & HRTIMER_MODE_REL)
1209 clock_id = CLOCK_MONOTONIC;
1211 base = hrtimer_clockid_to_base(clock_id);
1212 timer->base = &cpu_base->clock_base[base];
1213 timerqueue_init(&timer->node);
1215 #ifdef CONFIG_TIMER_STATS
1216 timer->start_site = NULL;
1217 timer->start_pid = -1;
1218 memset(timer->start_comm, 0, TASK_COMM_LEN);
1223 * hrtimer_init - initialize a timer to the given clock
1224 * @timer: the timer to be initialized
1225 * @clock_id: the clock to be used
1226 * @mode: timer mode abs/rel
1228 void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1229 enum hrtimer_mode mode)
1231 debug_init(timer, clock_id, mode);
1232 __hrtimer_init(timer, clock_id, mode);
1234 EXPORT_SYMBOL_GPL(hrtimer_init);
1237 * hrtimer_get_res - get the timer resolution for a clock
1238 * @which_clock: which clock to query
1239 * @tp: pointer to timespec variable to store the resolution
1241 * Store the resolution of the clock selected by @which_clock in the
1242 * variable pointed to by @tp.
1244 int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp)
1246 struct hrtimer_cpu_base *cpu_base;
1247 int base = hrtimer_clockid_to_base(which_clock);
1249 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1250 *tp = ktime_to_timespec(cpu_base->clock_base[base].resolution);
1254 EXPORT_SYMBOL_GPL(hrtimer_get_res);
1256 static void __run_hrtimer(struct hrtimer *timer, ktime_t *now)
1258 struct hrtimer_clock_base *base = timer->base;
1259 struct hrtimer_cpu_base *cpu_base = base->cpu_base;
1260 enum hrtimer_restart (*fn)(struct hrtimer *);
1263 WARN_ON(!irqs_disabled());
1265 debug_deactivate(timer);
1266 __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0);
1267 timer_stats_account_hrtimer(timer);
1268 fn = timer->function;
1271 * Clear the 'is relative' flag for the TIME_LOW_RES case. If the
1272 * timer is restarted with a period then it becomes an absolute
1273 * timer. If its not restarted it does not matter.
1275 #ifdef CONFIG_TIME_LOW_RES
1276 timer->is_rel = false;
1280 * Because we run timers from hardirq context, there is no chance
1281 * they get migrated to another cpu, therefore its safe to unlock
1284 raw_spin_unlock(&cpu_base->lock);
1285 trace_hrtimer_expire_entry(timer, now);
1286 restart = fn(timer);
1287 trace_hrtimer_expire_exit(timer);
1288 raw_spin_lock(&cpu_base->lock);
1291 * Note: We clear the CALLBACK bit after enqueue_hrtimer and
1292 * we do not reprogramm the event hardware. Happens either in
1293 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1295 * Note: Because we dropped the cpu_base->lock above,
1296 * hrtimer_start_range_ns() can have popped in and enqueued the timer
1299 if (restart != HRTIMER_NORESTART &&
1300 !(timer->state & HRTIMER_STATE_ENQUEUED))
1301 enqueue_hrtimer(timer, base);
1303 WARN_ON_ONCE(!(timer->state & HRTIMER_STATE_CALLBACK));
1305 timer->state &= ~HRTIMER_STATE_CALLBACK;
1308 #ifdef CONFIG_HIGH_RES_TIMERS
1311 * High resolution timer interrupt
1312 * Called with interrupts disabled
1314 void hrtimer_interrupt(struct clock_event_device *dev)
1316 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1317 ktime_t expires_next, now, entry_time, delta;
1320 BUG_ON(!cpu_base->hres_active);
1321 cpu_base->nr_events++;
1322 dev->next_event.tv64 = KTIME_MAX;
1324 raw_spin_lock(&cpu_base->lock);
1325 entry_time = now = hrtimer_update_base(cpu_base);
1327 expires_next.tv64 = KTIME_MAX;
1329 * We set expires_next to KTIME_MAX here with cpu_base->lock
1330 * held to prevent that a timer is enqueued in our queue via
1331 * the migration code. This does not affect enqueueing of
1332 * timers which run their callback and need to be requeued on
1335 cpu_base->expires_next.tv64 = KTIME_MAX;
1337 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1338 struct hrtimer_clock_base *base;
1339 struct timerqueue_node *node;
1342 if (!(cpu_base->active_bases & (1 << i)))
1345 base = cpu_base->clock_base + i;
1346 basenow = ktime_add(now, base->offset);
1348 while ((node = timerqueue_getnext(&base->active))) {
1349 struct hrtimer *timer;
1351 timer = container_of(node, struct hrtimer, node);
1354 * The immediate goal for using the softexpires is
1355 * minimizing wakeups, not running timers at the
1356 * earliest interrupt after their soft expiration.
1357 * This allows us to avoid using a Priority Search
1358 * Tree, which can answer a stabbing querry for
1359 * overlapping intervals and instead use the simple
1360 * BST we already have.
1361 * We don't add extra wakeups by delaying timers that
1362 * are right-of a not yet expired timer, because that
1363 * timer will have to trigger a wakeup anyway.
1366 if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer)) {
1369 expires = ktime_sub(hrtimer_get_expires(timer),
1371 if (expires.tv64 < 0)
1372 expires.tv64 = KTIME_MAX;
1373 if (expires.tv64 < expires_next.tv64)
1374 expires_next = expires;
1378 __run_hrtimer(timer, &basenow);
1383 * Store the new expiry value so the migration code can verify
1386 cpu_base->expires_next = expires_next;
1387 raw_spin_unlock(&cpu_base->lock);
1389 /* Reprogramming necessary ? */
1390 if (expires_next.tv64 == KTIME_MAX ||
1391 !tick_program_event(expires_next, 0)) {
1392 cpu_base->hang_detected = 0;
1397 * The next timer was already expired due to:
1399 * - long lasting callbacks
1400 * - being scheduled away when running in a VM
1402 * We need to prevent that we loop forever in the hrtimer
1403 * interrupt routine. We give it 3 attempts to avoid
1404 * overreacting on some spurious event.
1406 * Acquire base lock for updating the offsets and retrieving
1409 raw_spin_lock(&cpu_base->lock);
1410 now = hrtimer_update_base(cpu_base);
1411 cpu_base->nr_retries++;
1415 * Give the system a chance to do something else than looping
1416 * here. We stored the entry time, so we know exactly how long
1417 * we spent here. We schedule the next event this amount of
1420 cpu_base->nr_hangs++;
1421 cpu_base->hang_detected = 1;
1422 raw_spin_unlock(&cpu_base->lock);
1423 delta = ktime_sub(now, entry_time);
1424 if (delta.tv64 > cpu_base->max_hang_time.tv64)
1425 cpu_base->max_hang_time = delta;
1427 * Limit it to a sensible value as we enforce a longer
1428 * delay. Give the CPU at least 100ms to catch up.
1430 if (delta.tv64 > 100 * NSEC_PER_MSEC)
1431 expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1433 expires_next = ktime_add(now, delta);
1434 tick_program_event(expires_next, 1);
1435 printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n",
1436 ktime_to_ns(delta));
1440 * local version of hrtimer_peek_ahead_timers() called with interrupts
1443 static void __hrtimer_peek_ahead_timers(void)
1445 struct tick_device *td;
1447 if (!hrtimer_hres_active())
1450 td = &__get_cpu_var(tick_cpu_device);
1451 if (td && td->evtdev)
1452 hrtimer_interrupt(td->evtdev);
1456 * hrtimer_peek_ahead_timers -- run soft-expired timers now
1458 * hrtimer_peek_ahead_timers will peek at the timer queue of
1459 * the current cpu and check if there are any timers for which
1460 * the soft expires time has passed. If any such timers exist,
1461 * they are run immediately and then removed from the timer queue.
1464 void hrtimer_peek_ahead_timers(void)
1466 unsigned long flags;
1468 local_irq_save(flags);
1469 __hrtimer_peek_ahead_timers();
1470 local_irq_restore(flags);
1473 static void run_hrtimer_softirq(struct softirq_action *h)
1475 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1477 if (cpu_base->clock_was_set) {
1478 cpu_base->clock_was_set = 0;
1482 hrtimer_peek_ahead_timers();
1485 #else /* CONFIG_HIGH_RES_TIMERS */
1487 static inline void __hrtimer_peek_ahead_timers(void) { }
1489 #endif /* !CONFIG_HIGH_RES_TIMERS */
1492 * Called from timer softirq every jiffy, expire hrtimers:
1494 * For HRT its the fall back code to run the softirq in the timer
1495 * softirq context in case the hrtimer initialization failed or has
1496 * not been done yet.
1498 void hrtimer_run_pending(void)
1500 if (hrtimer_hres_active())
1504 * This _is_ ugly: We have to check in the softirq context,
1505 * whether we can switch to highres and / or nohz mode. The
1506 * clocksource switch happens in the timer interrupt with
1507 * xtime_lock held. Notification from there only sets the
1508 * check bit in the tick_oneshot code, otherwise we might
1509 * deadlock vs. xtime_lock.
1511 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1512 hrtimer_switch_to_hres();
1516 * Called from hardirq context every jiffy
1518 void hrtimer_run_queues(void)
1520 struct timerqueue_node *node;
1521 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1522 struct hrtimer_clock_base *base;
1523 int index, gettime = 1;
1525 if (hrtimer_hres_active())
1528 for (index = 0; index < HRTIMER_MAX_CLOCK_BASES; index++) {
1529 base = &cpu_base->clock_base[index];
1530 if (!timerqueue_getnext(&base->active))
1534 hrtimer_get_softirq_time(cpu_base);
1538 raw_spin_lock(&cpu_base->lock);
1540 while ((node = timerqueue_getnext(&base->active))) {
1541 struct hrtimer *timer;
1543 timer = container_of(node, struct hrtimer, node);
1544 if (base->softirq_time.tv64 <=
1545 hrtimer_get_expires_tv64(timer))
1548 __run_hrtimer(timer, &base->softirq_time);
1550 raw_spin_unlock(&cpu_base->lock);
1555 * Sleep related functions:
1557 static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1559 struct hrtimer_sleeper *t =
1560 container_of(timer, struct hrtimer_sleeper, timer);
1561 struct task_struct *task = t->task;
1565 wake_up_process(task);
1567 return HRTIMER_NORESTART;
1570 void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1572 sl->timer.function = hrtimer_wakeup;
1575 EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
1577 static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1579 hrtimer_init_sleeper(t, current);
1582 set_current_state(TASK_INTERRUPTIBLE);
1583 hrtimer_start_expires(&t->timer, mode);
1584 if (!hrtimer_active(&t->timer))
1587 if (likely(t->task))
1590 hrtimer_cancel(&t->timer);
1591 mode = HRTIMER_MODE_ABS;
1593 } while (t->task && !signal_pending(current));
1595 __set_current_state(TASK_RUNNING);
1597 return t->task == NULL;
1600 static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp)
1602 struct timespec rmt;
1605 rem = hrtimer_expires_remaining(timer);
1608 rmt = ktime_to_timespec(rem);
1610 if (copy_to_user(rmtp, &rmt, sizeof(*rmtp)))
1616 long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1618 struct hrtimer_sleeper t;
1619 struct timespec __user *rmtp;
1622 hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid,
1624 hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
1626 if (do_nanosleep(&t, HRTIMER_MODE_ABS))
1629 rmtp = restart->nanosleep.rmtp;
1631 ret = update_rmtp(&t.timer, rmtp);
1636 /* The other values in restart are already filled in */
1637 ret = -ERESTART_RESTARTBLOCK;
1639 destroy_hrtimer_on_stack(&t.timer);
1643 long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
1644 const enum hrtimer_mode mode, const clockid_t clockid)
1646 struct restart_block *restart;
1647 struct hrtimer_sleeper t;
1649 unsigned long slack;
1651 slack = current->timer_slack_ns;
1652 if (rt_task(current))
1655 hrtimer_init_on_stack(&t.timer, clockid, mode);
1656 hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack);
1657 if (do_nanosleep(&t, mode))
1660 /* Absolute timers do not update the rmtp value and restart: */
1661 if (mode == HRTIMER_MODE_ABS) {
1662 ret = -ERESTARTNOHAND;
1667 ret = update_rmtp(&t.timer, rmtp);
1672 restart = ¤t_thread_info()->restart_block;
1673 restart->fn = hrtimer_nanosleep_restart;
1674 restart->nanosleep.clockid = t.timer.base->clockid;
1675 restart->nanosleep.rmtp = rmtp;
1676 restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
1678 ret = -ERESTART_RESTARTBLOCK;
1680 destroy_hrtimer_on_stack(&t.timer);
1684 SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp,
1685 struct timespec __user *, rmtp)
1689 if (copy_from_user(&tu, rqtp, sizeof(tu)))
1692 if (!timespec_valid(&tu))
1695 return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1699 * Functions related to boot-time initialization:
1701 static void __cpuinit init_hrtimers_cpu(int cpu)
1703 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1706 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1707 cpu_base->clock_base[i].cpu_base = cpu_base;
1708 timerqueue_init_head(&cpu_base->clock_base[i].active);
1711 cpu_base->active_bases = 0;
1712 hrtimer_init_hres(cpu_base);
1715 #ifdef CONFIG_HOTPLUG_CPU
1717 static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1718 struct hrtimer_clock_base *new_base)
1720 struct hrtimer *timer;
1721 struct timerqueue_node *node;
1723 while ((node = timerqueue_getnext(&old_base->active))) {
1724 timer = container_of(node, struct hrtimer, node);
1725 BUG_ON(hrtimer_callback_running(timer));
1726 debug_deactivate(timer);
1729 * Mark it as STATE_MIGRATE not INACTIVE otherwise the
1730 * timer could be seen as !active and just vanish away
1731 * under us on another CPU
1733 __remove_hrtimer(timer, old_base, HRTIMER_STATE_MIGRATE, 0);
1734 timer->base = new_base;
1736 * Enqueue the timers on the new cpu. This does not
1737 * reprogram the event device in case the timer
1738 * expires before the earliest on this CPU, but we run
1739 * hrtimer_interrupt after we migrated everything to
1740 * sort out already expired timers and reprogram the
1743 enqueue_hrtimer(timer, new_base);
1745 /* Clear the migration state bit */
1746 timer->state &= ~HRTIMER_STATE_MIGRATE;
1750 static void migrate_hrtimers(int scpu)
1752 struct hrtimer_cpu_base *old_base, *new_base;
1755 BUG_ON(cpu_online(scpu));
1756 tick_cancel_sched_timer(scpu);
1758 local_irq_disable();
1759 old_base = &per_cpu(hrtimer_bases, scpu);
1760 new_base = &__get_cpu_var(hrtimer_bases);
1762 * The caller is globally serialized and nobody else
1763 * takes two locks at once, deadlock is not possible.
1765 raw_spin_lock(&new_base->lock);
1766 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1768 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1769 migrate_hrtimer_list(&old_base->clock_base[i],
1770 &new_base->clock_base[i]);
1773 raw_spin_unlock(&old_base->lock);
1774 raw_spin_unlock(&new_base->lock);
1776 /* Check, if we got expired work to do */
1777 __hrtimer_peek_ahead_timers();
1781 #endif /* CONFIG_HOTPLUG_CPU */
1783 static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self,
1784 unsigned long action, void *hcpu)
1786 int scpu = (long)hcpu;
1790 case CPU_UP_PREPARE:
1791 case CPU_UP_PREPARE_FROZEN:
1792 init_hrtimers_cpu(scpu);
1795 #ifdef CONFIG_HOTPLUG_CPU
1797 case CPU_DYING_FROZEN:
1798 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DYING, &scpu);
1801 case CPU_DEAD_FROZEN:
1803 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &scpu);
1804 migrate_hrtimers(scpu);
1816 static struct notifier_block __cpuinitdata hrtimers_nb = {
1817 .notifier_call = hrtimer_cpu_notify,
1820 void __init hrtimers_init(void)
1822 hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
1823 (void *)(long)smp_processor_id());
1824 register_cpu_notifier(&hrtimers_nb);
1825 #ifdef CONFIG_HIGH_RES_TIMERS
1826 open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq);
1831 * schedule_hrtimeout_range_clock - sleep until timeout
1832 * @expires: timeout value (ktime_t)
1833 * @delta: slack in expires timeout (ktime_t)
1834 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1835 * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME
1838 schedule_hrtimeout_range_clock(ktime_t *expires, unsigned long delta,
1839 const enum hrtimer_mode mode, int clock)
1841 struct hrtimer_sleeper t;
1844 * Optimize when a zero timeout value is given. It does not
1845 * matter whether this is an absolute or a relative time.
1847 if (expires && !expires->tv64) {
1848 __set_current_state(TASK_RUNNING);
1853 * A NULL parameter means "infinite"
1857 __set_current_state(TASK_RUNNING);
1861 hrtimer_init_on_stack(&t.timer, clock, mode);
1862 hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
1864 hrtimer_init_sleeper(&t, current);
1866 hrtimer_start_expires(&t.timer, mode);
1867 if (!hrtimer_active(&t.timer))
1873 hrtimer_cancel(&t.timer);
1874 destroy_hrtimer_on_stack(&t.timer);
1876 __set_current_state(TASK_RUNNING);
1878 return !t.task ? 0 : -EINTR;
1882 * schedule_hrtimeout_range - sleep until timeout
1883 * @expires: timeout value (ktime_t)
1884 * @delta: slack in expires timeout (ktime_t)
1885 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1887 * Make the current task sleep until the given expiry time has
1888 * elapsed. The routine will return immediately unless
1889 * the current task state has been set (see set_current_state()).
1891 * The @delta argument gives the kernel the freedom to schedule the
1892 * actual wakeup to a time that is both power and performance friendly.
1893 * The kernel give the normal best effort behavior for "@expires+@delta",
1894 * but may decide to fire the timer earlier, but no earlier than @expires.
1896 * You can set the task state as follows -
1898 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1899 * pass before the routine returns.
1901 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1902 * delivered to the current task.
1904 * The current task state is guaranteed to be TASK_RUNNING when this
1907 * Returns 0 when the timer has expired otherwise -EINTR
1909 int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta,
1910 const enum hrtimer_mode mode)
1912 return schedule_hrtimeout_range_clock(expires, delta, mode,
1915 EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
1918 * schedule_hrtimeout - sleep until timeout
1919 * @expires: timeout value (ktime_t)
1920 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1922 * Make the current task sleep until the given expiry time has
1923 * elapsed. The routine will return immediately unless
1924 * the current task state has been set (see set_current_state()).
1926 * You can set the task state as follows -
1928 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1929 * pass before the routine returns.
1931 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1932 * delivered to the current task.
1934 * The current task state is guaranteed to be TASK_RUNNING when this
1937 * Returns 0 when the timer has expired otherwise -EINTR
1939 int __sched schedule_hrtimeout(ktime_t *expires,
1940 const enum hrtimer_mode mode)
1942 return schedule_hrtimeout_range(expires, 0, mode);
1944 EXPORT_SYMBOL_GPL(schedule_hrtimeout);