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/timer.h>
50 #include <linux/freezer.h>
52 #include <asm/uaccess.h>
54 #include <trace/events/timer.h>
59 * There are more clockids then hrtimer bases. Thus, we index
60 * into the timer bases by the hrtimer_base_type enum. When trying
61 * to reach a base using a clockid, hrtimer_clockid_to_base()
62 * is used to convert from clockid to the proper hrtimer_base_type.
64 DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
67 .lock = __RAW_SPIN_LOCK_UNLOCKED(hrtimer_bases.lock),
71 .index = HRTIMER_BASE_MONOTONIC,
72 .clockid = CLOCK_MONOTONIC,
73 .get_time = &ktime_get,
74 .resolution = KTIME_LOW_RES,
77 .index = HRTIMER_BASE_REALTIME,
78 .clockid = CLOCK_REALTIME,
79 .get_time = &ktime_get_real,
80 .resolution = KTIME_LOW_RES,
83 .index = HRTIMER_BASE_BOOTTIME,
84 .clockid = CLOCK_BOOTTIME,
85 .get_time = &ktime_get_boottime,
86 .resolution = KTIME_LOW_RES,
89 .index = HRTIMER_BASE_TAI,
91 .get_time = &ktime_get_clocktai,
92 .resolution = KTIME_LOW_RES,
97 static const int hrtimer_clock_to_base_table[MAX_CLOCKS] = {
98 [CLOCK_REALTIME] = HRTIMER_BASE_REALTIME,
99 [CLOCK_MONOTONIC] = HRTIMER_BASE_MONOTONIC,
100 [CLOCK_BOOTTIME] = HRTIMER_BASE_BOOTTIME,
101 [CLOCK_TAI] = HRTIMER_BASE_TAI,
104 static inline int hrtimer_clockid_to_base(clockid_t clock_id)
106 return hrtimer_clock_to_base_table[clock_id];
111 * Get the coarse grained time at the softirq based on xtime and
114 static void hrtimer_get_softirq_time(struct hrtimer_cpu_base *base)
116 ktime_t xtim, mono, boot;
117 struct timespec xts, tom, slp;
120 get_xtime_and_monotonic_and_sleep_offset(&xts, &tom, &slp);
121 tai_offset = timekeeping_get_tai_offset();
123 xtim = timespec_to_ktime(xts);
124 mono = ktime_add(xtim, timespec_to_ktime(tom));
125 boot = ktime_add(mono, timespec_to_ktime(slp));
126 base->clock_base[HRTIMER_BASE_REALTIME].softirq_time = xtim;
127 base->clock_base[HRTIMER_BASE_MONOTONIC].softirq_time = mono;
128 base->clock_base[HRTIMER_BASE_BOOTTIME].softirq_time = boot;
129 base->clock_base[HRTIMER_BASE_TAI].softirq_time =
130 ktime_add(xtim, ktime_set(tai_offset, 0));
134 * Functions and macros which are different for UP/SMP systems are kept in a
140 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
141 * means that all timers which are tied to this base via timer->base are
142 * locked, and the base itself is locked too.
144 * So __run_timers/migrate_timers can safely modify all timers which could
145 * be found on the lists/queues.
147 * When the timer's base is locked, and the timer removed from list, it is
148 * possible to set timer->base = NULL and drop the lock: the timer remains
152 struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
153 unsigned long *flags)
155 struct hrtimer_clock_base *base;
159 if (likely(base != NULL)) {
160 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
161 if (likely(base == timer->base))
163 /* The timer has migrated to another CPU: */
164 raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
172 * Get the preferred target CPU for NOHZ
174 static int hrtimer_get_target(int this_cpu, int pinned)
176 #ifdef CONFIG_NO_HZ_COMMON
177 if (!pinned && get_sysctl_timer_migration() && idle_cpu(this_cpu))
178 return get_nohz_timer_target();
184 * With HIGHRES=y we do not migrate the timer when it is expiring
185 * before the next event on the target cpu because we cannot reprogram
186 * the target cpu hardware and we would cause it to fire late.
188 * Called with cpu_base->lock of target cpu held.
191 hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base)
193 #ifdef CONFIG_HIGH_RES_TIMERS
196 if (!new_base->cpu_base->hres_active)
199 expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset);
200 return expires.tv64 <= new_base->cpu_base->expires_next.tv64;
207 * Switch the timer base to the current CPU when possible.
209 static inline struct hrtimer_clock_base *
210 switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
213 struct hrtimer_clock_base *new_base;
214 struct hrtimer_cpu_base *new_cpu_base;
215 int this_cpu = smp_processor_id();
216 int cpu = hrtimer_get_target(this_cpu, pinned);
217 int basenum = base->index;
220 new_cpu_base = &per_cpu(hrtimer_bases, cpu);
221 new_base = &new_cpu_base->clock_base[basenum];
223 if (base != new_base) {
225 * We are trying to move timer to new_base.
226 * However we can't change timer's base while it is running,
227 * so we keep it on the same CPU. No hassle vs. reprogramming
228 * the event source in the high resolution case. The softirq
229 * code will take care of this when the timer function has
230 * completed. There is no conflict as we hold the lock until
231 * the timer is enqueued.
233 if (unlikely(hrtimer_callback_running(timer)))
236 /* See the comment in lock_timer_base() */
238 raw_spin_unlock(&base->cpu_base->lock);
239 raw_spin_lock(&new_base->cpu_base->lock);
241 if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) {
243 raw_spin_unlock(&new_base->cpu_base->lock);
244 raw_spin_lock(&base->cpu_base->lock);
248 timer->base = new_base;
250 if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) {
258 #else /* CONFIG_SMP */
260 static inline struct hrtimer_clock_base *
261 lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
263 struct hrtimer_clock_base *base = timer->base;
265 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
270 # define switch_hrtimer_base(t, b, p) (b)
272 #endif /* !CONFIG_SMP */
275 * Functions for the union type storage format of ktime_t which are
276 * too large for inlining:
278 #if BITS_PER_LONG < 64
279 # ifndef CONFIG_KTIME_SCALAR
281 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
283 * @nsec: the scalar nsec value to add
285 * Returns the sum of kt and nsec in ktime_t format
287 ktime_t ktime_add_ns(const ktime_t kt, u64 nsec)
291 if (likely(nsec < NSEC_PER_SEC)) {
294 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
296 /* Make sure nsec fits into long */
297 if (unlikely(nsec > KTIME_SEC_MAX))
298 return (ktime_t){ .tv64 = KTIME_MAX };
300 tmp = ktime_set((long)nsec, rem);
303 return ktime_add(kt, tmp);
306 EXPORT_SYMBOL_GPL(ktime_add_ns);
309 * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable
311 * @nsec: the scalar nsec value to subtract
313 * Returns the subtraction of @nsec from @kt in ktime_t format
315 ktime_t ktime_sub_ns(const ktime_t kt, u64 nsec)
319 if (likely(nsec < NSEC_PER_SEC)) {
322 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
324 tmp = ktime_set((long)nsec, rem);
327 return ktime_sub(kt, tmp);
330 EXPORT_SYMBOL_GPL(ktime_sub_ns);
331 # endif /* !CONFIG_KTIME_SCALAR */
334 * Divide a ktime value by a nanosecond value
336 u64 ktime_divns(const ktime_t kt, s64 div)
341 dclc = ktime_to_ns(kt);
342 /* Make sure the divisor is less than 2^32: */
348 do_div(dclc, (unsigned long) div);
352 #endif /* BITS_PER_LONG >= 64 */
355 * Add two ktime values and do a safety check for overflow:
357 ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
359 ktime_t res = ktime_add(lhs, rhs);
362 * We use KTIME_SEC_MAX here, the maximum timeout which we can
363 * return to user space in a timespec:
365 if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64)
366 res = ktime_set(KTIME_SEC_MAX, 0);
371 EXPORT_SYMBOL_GPL(ktime_add_safe);
373 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
375 static struct debug_obj_descr hrtimer_debug_descr;
377 static void *hrtimer_debug_hint(void *addr)
379 return ((struct hrtimer *) addr)->function;
383 * fixup_init is called when:
384 * - an active object is initialized
386 static int hrtimer_fixup_init(void *addr, enum debug_obj_state state)
388 struct hrtimer *timer = addr;
391 case ODEBUG_STATE_ACTIVE:
392 hrtimer_cancel(timer);
393 debug_object_init(timer, &hrtimer_debug_descr);
401 * fixup_activate is called when:
402 * - an active object is activated
403 * - an unknown object is activated (might be a statically initialized object)
405 static int hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
409 case ODEBUG_STATE_NOTAVAILABLE:
413 case ODEBUG_STATE_ACTIVE:
422 * fixup_free is called when:
423 * - an active object is freed
425 static int hrtimer_fixup_free(void *addr, enum debug_obj_state state)
427 struct hrtimer *timer = addr;
430 case ODEBUG_STATE_ACTIVE:
431 hrtimer_cancel(timer);
432 debug_object_free(timer, &hrtimer_debug_descr);
439 static struct debug_obj_descr hrtimer_debug_descr = {
441 .debug_hint = hrtimer_debug_hint,
442 .fixup_init = hrtimer_fixup_init,
443 .fixup_activate = hrtimer_fixup_activate,
444 .fixup_free = hrtimer_fixup_free,
447 static inline void debug_hrtimer_init(struct hrtimer *timer)
449 debug_object_init(timer, &hrtimer_debug_descr);
452 static inline void debug_hrtimer_activate(struct hrtimer *timer)
454 debug_object_activate(timer, &hrtimer_debug_descr);
457 static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
459 debug_object_deactivate(timer, &hrtimer_debug_descr);
462 static inline void debug_hrtimer_free(struct hrtimer *timer)
464 debug_object_free(timer, &hrtimer_debug_descr);
467 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
468 enum hrtimer_mode mode);
470 void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
471 enum hrtimer_mode mode)
473 debug_object_init_on_stack(timer, &hrtimer_debug_descr);
474 __hrtimer_init(timer, clock_id, mode);
476 EXPORT_SYMBOL_GPL(hrtimer_init_on_stack);
478 void destroy_hrtimer_on_stack(struct hrtimer *timer)
480 debug_object_free(timer, &hrtimer_debug_descr);
484 static inline void debug_hrtimer_init(struct hrtimer *timer) { }
485 static inline void debug_hrtimer_activate(struct hrtimer *timer) { }
486 static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
490 debug_init(struct hrtimer *timer, clockid_t clockid,
491 enum hrtimer_mode mode)
493 debug_hrtimer_init(timer);
494 trace_hrtimer_init(timer, clockid, mode);
497 static inline void debug_activate(struct hrtimer *timer)
499 debug_hrtimer_activate(timer);
500 trace_hrtimer_start(timer);
503 static inline void debug_deactivate(struct hrtimer *timer)
505 debug_hrtimer_deactivate(timer);
506 trace_hrtimer_cancel(timer);
509 /* High resolution timer related functions */
510 #ifdef CONFIG_HIGH_RES_TIMERS
513 * High resolution timer enabled ?
515 static int hrtimer_hres_enabled __read_mostly = 1;
518 * Enable / Disable high resolution mode
520 static int __init setup_hrtimer_hres(char *str)
522 if (!strcmp(str, "off"))
523 hrtimer_hres_enabled = 0;
524 else if (!strcmp(str, "on"))
525 hrtimer_hres_enabled = 1;
531 __setup("highres=", setup_hrtimer_hres);
534 * hrtimer_high_res_enabled - query, if the highres mode is enabled
536 static inline int hrtimer_is_hres_enabled(void)
538 return hrtimer_hres_enabled;
542 * Is the high resolution mode active ?
544 static inline int hrtimer_hres_active(void)
546 return __this_cpu_read(hrtimer_bases.hres_active);
550 * Reprogram the event source with checking both queues for the
552 * Called with interrupts disabled and base->lock held
555 hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
558 struct hrtimer_clock_base *base = cpu_base->clock_base;
559 ktime_t expires, expires_next;
561 expires_next.tv64 = KTIME_MAX;
563 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
564 struct hrtimer *timer;
565 struct timerqueue_node *next;
567 next = timerqueue_getnext(&base->active);
570 timer = container_of(next, struct hrtimer, node);
572 expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
574 * clock_was_set() has changed base->offset so the
575 * result might be negative. Fix it up to prevent a
576 * false positive in clockevents_program_event()
578 if (expires.tv64 < 0)
580 if (expires.tv64 < expires_next.tv64)
581 expires_next = expires;
584 if (skip_equal && expires_next.tv64 == cpu_base->expires_next.tv64)
587 cpu_base->expires_next.tv64 = expires_next.tv64;
590 * If a hang was detected in the last timer interrupt then we
591 * leave the hang delay active in the hardware. We want the
592 * system to make progress. That also prevents the following
594 * T1 expires 50ms from now
595 * T2 expires 5s from now
597 * T1 is removed, so this code is called and would reprogram
598 * the hardware to 5s from now. Any hrtimer_start after that
599 * will not reprogram the hardware due to hang_detected being
600 * set. So we'd effectivly block all timers until the T2 event
603 if (cpu_base->hang_detected)
606 if (cpu_base->expires_next.tv64 != KTIME_MAX)
607 tick_program_event(cpu_base->expires_next, 1);
611 * Shared reprogramming for clock_realtime and clock_monotonic
613 * When a timer is enqueued and expires earlier than the already enqueued
614 * timers, we have to check, whether it expires earlier than the timer for
615 * which the clock event device was armed.
617 * Called with interrupts disabled and base->cpu_base.lock held
619 static int hrtimer_reprogram(struct hrtimer *timer,
620 struct hrtimer_clock_base *base)
622 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
623 ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
626 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
629 * When the callback is running, we do not reprogram the clock event
630 * device. The timer callback is either running on a different CPU or
631 * the callback is executed in the hrtimer_interrupt context. The
632 * reprogramming is handled either by the softirq, which called the
633 * callback or at the end of the hrtimer_interrupt.
635 if (hrtimer_callback_running(timer))
639 * CLOCK_REALTIME timer might be requested with an absolute
640 * expiry time which is less than base->offset. Nothing wrong
641 * about that, just avoid to call into the tick code, which
642 * has now objections against negative expiry values.
644 if (expires.tv64 < 0)
647 if (expires.tv64 >= cpu_base->expires_next.tv64)
651 * If a hang was detected in the last timer interrupt then we
652 * do not schedule a timer which is earlier than the expiry
653 * which we enforced in the hang detection. We want the system
656 if (cpu_base->hang_detected)
660 * Clockevents returns -ETIME, when the event was in the past.
662 res = tick_program_event(expires, 0);
663 if (!IS_ERR_VALUE(res))
664 cpu_base->expires_next = expires;
669 * Initialize the high resolution related parts of cpu_base
671 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
673 base->expires_next.tv64 = KTIME_MAX;
674 base->hres_active = 0;
678 * When High resolution timers are active, try to reprogram. Note, that in case
679 * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry
680 * check happens. The timer gets enqueued into the rbtree. The reprogramming
681 * and expiry check is done in the hrtimer_interrupt or in the softirq.
683 static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
684 struct hrtimer_clock_base *base)
686 return base->cpu_base->hres_active && hrtimer_reprogram(timer, base);
689 static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base)
691 ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset;
692 ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset;
693 ktime_t *offs_tai = &base->clock_base[HRTIMER_BASE_TAI].offset;
695 return ktime_get_update_offsets(offs_real, offs_boot, offs_tai);
699 * Retrigger next event is called after clock was set
701 * Called with interrupts disabled via on_each_cpu()
703 static void retrigger_next_event(void *arg)
705 struct hrtimer_cpu_base *base = &__get_cpu_var(hrtimer_bases);
707 if (!hrtimer_hres_active())
710 raw_spin_lock(&base->lock);
711 hrtimer_update_base(base);
712 hrtimer_force_reprogram(base, 0);
713 raw_spin_unlock(&base->lock);
717 * Switch to high resolution mode
719 static int hrtimer_switch_to_hres(void)
721 int i, cpu = smp_processor_id();
722 struct hrtimer_cpu_base *base = &per_cpu(hrtimer_bases, cpu);
725 if (base->hres_active)
728 local_irq_save(flags);
730 if (tick_init_highres()) {
731 local_irq_restore(flags);
732 printk(KERN_WARNING "Could not switch to high resolution "
733 "mode on CPU %d\n", cpu);
736 base->hres_active = 1;
737 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
738 base->clock_base[i].resolution = KTIME_HIGH_RES;
740 tick_setup_sched_timer();
741 /* "Retrigger" the interrupt to get things going */
742 retrigger_next_event(NULL);
743 local_irq_restore(flags);
747 static void clock_was_set_work(struct work_struct *work)
752 static DECLARE_WORK(hrtimer_work, clock_was_set_work);
755 * Called from timekeeping and resume code to reprogramm the hrtimer
756 * interrupt device on all cpus.
758 void clock_was_set_delayed(void)
760 schedule_work(&hrtimer_work);
765 static inline int hrtimer_hres_active(void) { return 0; }
766 static inline int hrtimer_is_hres_enabled(void) { return 0; }
767 static inline int hrtimer_switch_to_hres(void) { return 0; }
769 hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { }
770 static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
771 struct hrtimer_clock_base *base)
775 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { }
776 static inline void retrigger_next_event(void *arg) { }
778 #endif /* CONFIG_HIGH_RES_TIMERS */
781 * Clock realtime was set
783 * Change the offset of the realtime clock vs. the monotonic
786 * We might have to reprogram the high resolution timer interrupt. On
787 * SMP we call the architecture specific code to retrigger _all_ high
788 * resolution timer interrupts. On UP we just disable interrupts and
789 * call the high resolution interrupt code.
791 void clock_was_set(void)
793 #ifdef CONFIG_HIGH_RES_TIMERS
794 /* Retrigger the CPU local events everywhere */
795 on_each_cpu(retrigger_next_event, NULL, 1);
797 timerfd_clock_was_set();
801 * During resume we might have to reprogram the high resolution timer
802 * interrupt (on the local CPU):
804 void hrtimers_resume(void)
806 WARN_ONCE(!irqs_disabled(),
807 KERN_INFO "hrtimers_resume() called with IRQs enabled!");
809 /* Retrigger on the local CPU */
810 retrigger_next_event(NULL);
811 /* And schedule a retrigger for all others */
812 clock_was_set_delayed();
815 static inline void timer_stats_hrtimer_set_start_info(struct hrtimer *timer)
817 #ifdef CONFIG_TIMER_STATS
818 if (timer->start_site)
820 timer->start_site = __builtin_return_address(0);
821 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
822 timer->start_pid = current->pid;
826 static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer *timer)
828 #ifdef CONFIG_TIMER_STATS
829 timer->start_site = NULL;
833 static inline void timer_stats_account_hrtimer(struct hrtimer *timer)
835 #ifdef CONFIG_TIMER_STATS
836 if (likely(!timer_stats_active))
838 timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
839 timer->function, timer->start_comm, 0);
844 * Counterpart to lock_hrtimer_base above:
847 void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
849 raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
853 * hrtimer_forward - forward the timer expiry
854 * @timer: hrtimer to forward
855 * @now: forward past this time
856 * @interval: the interval to forward
858 * Forward the timer expiry so it will expire in the future.
859 * Returns the number of overruns.
861 u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
866 delta = ktime_sub(now, hrtimer_get_expires(timer));
871 if (interval.tv64 < timer->base->resolution.tv64)
872 interval.tv64 = timer->base->resolution.tv64;
874 if (unlikely(delta.tv64 >= interval.tv64)) {
875 s64 incr = ktime_to_ns(interval);
877 orun = ktime_divns(delta, incr);
878 hrtimer_add_expires_ns(timer, incr * orun);
879 if (hrtimer_get_expires_tv64(timer) > now.tv64)
882 * This (and the ktime_add() below) is the
883 * correction for exact:
887 hrtimer_add_expires(timer, interval);
891 EXPORT_SYMBOL_GPL(hrtimer_forward);
894 * enqueue_hrtimer - internal function to (re)start a timer
896 * The timer is inserted in expiry order. Insertion into the
897 * red black tree is O(log(n)). Must hold the base lock.
899 * Returns 1 when the new timer is the leftmost timer in the tree.
901 static int enqueue_hrtimer(struct hrtimer *timer,
902 struct hrtimer_clock_base *base)
904 debug_activate(timer);
906 timerqueue_add(&base->active, &timer->node);
907 base->cpu_base->active_bases |= 1 << base->index;
910 * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
911 * state of a possibly running callback.
913 timer->state |= HRTIMER_STATE_ENQUEUED;
915 return (&timer->node == base->active.next);
919 * __remove_hrtimer - internal function to remove a timer
921 * Caller must hold the base lock.
923 * High resolution timer mode reprograms the clock event device when the
924 * timer is the one which expires next. The caller can disable this by setting
925 * reprogram to zero. This is useful, when the context does a reprogramming
926 * anyway (e.g. timer interrupt)
928 static void __remove_hrtimer(struct hrtimer *timer,
929 struct hrtimer_clock_base *base,
930 unsigned long newstate, int reprogram)
932 struct timerqueue_node *next_timer;
933 if (!(timer->state & HRTIMER_STATE_ENQUEUED))
936 next_timer = timerqueue_getnext(&base->active);
937 timerqueue_del(&base->active, &timer->node);
938 if (&timer->node == next_timer) {
939 #ifdef CONFIG_HIGH_RES_TIMERS
940 /* Reprogram the clock event device. if enabled */
941 if (reprogram && hrtimer_hres_active()) {
944 expires = ktime_sub(hrtimer_get_expires(timer),
946 if (base->cpu_base->expires_next.tv64 == expires.tv64)
947 hrtimer_force_reprogram(base->cpu_base, 1);
951 if (!timerqueue_getnext(&base->active))
952 base->cpu_base->active_bases &= ~(1 << base->index);
954 timer->state = newstate;
958 * remove hrtimer, called with base lock held
961 remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base)
963 if (hrtimer_is_queued(timer)) {
968 * Remove the timer and force reprogramming when high
969 * resolution mode is active and the timer is on the current
970 * CPU. If we remove a timer on another CPU, reprogramming is
971 * skipped. The interrupt event on this CPU is fired and
972 * reprogramming happens in the interrupt handler. This is a
973 * rare case and less expensive than a smp call.
975 debug_deactivate(timer);
976 timer_stats_hrtimer_clear_start_info(timer);
977 reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases);
979 * We must preserve the CALLBACK state flag here,
980 * otherwise we could move the timer base in
981 * switch_hrtimer_base.
983 state = timer->state & HRTIMER_STATE_CALLBACK;
984 __remove_hrtimer(timer, base, state, reprogram);
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 * CONFIG_TIME_LOW_RES is a temporary way for architectures
1007 * to signal that they simply return xtime in
1008 * do_gettimeoffset(). In this case we want to round up by
1009 * resolution when starting a relative timer, to avoid short
1010 * timeouts. This will go away with the GTOD framework.
1012 #ifdef CONFIG_TIME_LOW_RES
1013 tim = ktime_add_safe(tim, base->resolution);
1017 hrtimer_set_expires_range_ns(timer, tim, delta_ns);
1019 /* Switch the timer base, if necessary: */
1020 new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED);
1022 timer_stats_hrtimer_set_start_info(timer);
1024 leftmost = enqueue_hrtimer(timer, new_base);
1027 * Only allow reprogramming if the new base is on this CPU.
1028 * (it might still be on another CPU if the timer was pending)
1030 * XXX send_remote_softirq() ?
1032 if (leftmost && new_base->cpu_base == &__get_cpu_var(hrtimer_bases)
1033 && hrtimer_enqueue_reprogram(timer, new_base)) {
1036 * We need to drop cpu_base->lock to avoid a
1037 * lock ordering issue vs. rq->lock.
1039 raw_spin_unlock(&new_base->cpu_base->lock);
1040 raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1041 local_irq_restore(flags);
1044 __raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1048 unlock_hrtimer_base(timer, &flags);
1054 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
1055 * @timer: the timer to be added
1057 * @delta_ns: "slack" range for the timer
1058 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
1059 * relative (HRTIMER_MODE_REL)
1063 * 1 when the timer was active
1065 int hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1066 unsigned long delta_ns, const enum hrtimer_mode mode)
1068 return __hrtimer_start_range_ns(timer, tim, delta_ns, mode, 1);
1070 EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
1073 * hrtimer_start - (re)start an hrtimer on the current CPU
1074 * @timer: the timer to be added
1076 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
1077 * relative (HRTIMER_MODE_REL)
1081 * 1 when the timer was active
1084 hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode)
1086 return __hrtimer_start_range_ns(timer, tim, 0, mode, 1);
1088 EXPORT_SYMBOL_GPL(hrtimer_start);
1092 * hrtimer_try_to_cancel - try to deactivate a timer
1093 * @timer: hrtimer to stop
1096 * 0 when the timer was not active
1097 * 1 when the timer was active
1098 * -1 when the timer is currently excuting the callback function and
1101 int hrtimer_try_to_cancel(struct hrtimer *timer)
1103 struct hrtimer_clock_base *base;
1104 unsigned long flags;
1107 base = lock_hrtimer_base(timer, &flags);
1109 if (!hrtimer_callback_running(timer))
1110 ret = remove_hrtimer(timer, base);
1112 unlock_hrtimer_base(timer, &flags);
1117 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1120 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1121 * @timer: the timer to be cancelled
1124 * 0 when the timer was not active
1125 * 1 when the timer was active
1127 int hrtimer_cancel(struct hrtimer *timer)
1130 int ret = hrtimer_try_to_cancel(timer);
1137 EXPORT_SYMBOL_GPL(hrtimer_cancel);
1140 * hrtimer_get_remaining - get remaining time for the timer
1141 * @timer: the timer to read
1143 ktime_t hrtimer_get_remaining(const struct hrtimer *timer)
1145 unsigned long flags;
1148 lock_hrtimer_base(timer, &flags);
1149 rem = hrtimer_expires_remaining(timer);
1150 unlock_hrtimer_base(timer, &flags);
1154 EXPORT_SYMBOL_GPL(hrtimer_get_remaining);
1156 #ifdef CONFIG_NO_HZ_COMMON
1158 * hrtimer_get_next_event - get the time until next expiry event
1160 * Returns the delta to the next expiry event or KTIME_MAX if no timer
1163 ktime_t hrtimer_get_next_event(void)
1165 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1166 struct hrtimer_clock_base *base = cpu_base->clock_base;
1167 ktime_t delta, mindelta = { .tv64 = KTIME_MAX };
1168 unsigned long flags;
1171 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1173 if (!hrtimer_hres_active()) {
1174 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
1175 struct hrtimer *timer;
1176 struct timerqueue_node *next;
1178 next = timerqueue_getnext(&base->active);
1182 timer = container_of(next, struct hrtimer, node);
1183 delta.tv64 = hrtimer_get_expires_tv64(timer);
1184 delta = ktime_sub(delta, base->get_time());
1185 if (delta.tv64 < mindelta.tv64)
1186 mindelta.tv64 = delta.tv64;
1190 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1192 if (mindelta.tv64 < 0)
1198 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1199 enum hrtimer_mode mode)
1201 struct hrtimer_cpu_base *cpu_base;
1204 memset(timer, 0, sizeof(struct hrtimer));
1206 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1208 if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
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 * Because we run timers from hardirq context, there is no chance
1272 * they get migrated to another cpu, therefore its safe to unlock
1275 raw_spin_unlock(&cpu_base->lock);
1276 trace_hrtimer_expire_entry(timer, now);
1277 restart = fn(timer);
1278 trace_hrtimer_expire_exit(timer);
1279 raw_spin_lock(&cpu_base->lock);
1282 * Note: We clear the CALLBACK bit after enqueue_hrtimer and
1283 * we do not reprogramm the event hardware. Happens either in
1284 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1286 if (restart != HRTIMER_NORESTART) {
1287 BUG_ON(timer->state != HRTIMER_STATE_CALLBACK);
1288 enqueue_hrtimer(timer, base);
1291 WARN_ON_ONCE(!(timer->state & HRTIMER_STATE_CALLBACK));
1293 timer->state &= ~HRTIMER_STATE_CALLBACK;
1296 #ifdef CONFIG_HIGH_RES_TIMERS
1299 * High resolution timer interrupt
1300 * Called with interrupts disabled
1302 void hrtimer_interrupt(struct clock_event_device *dev)
1304 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1305 ktime_t expires_next, now, entry_time, delta;
1308 BUG_ON(!cpu_base->hres_active);
1309 cpu_base->nr_events++;
1310 dev->next_event.tv64 = KTIME_MAX;
1312 raw_spin_lock(&cpu_base->lock);
1313 entry_time = now = hrtimer_update_base(cpu_base);
1315 expires_next.tv64 = KTIME_MAX;
1317 * We set expires_next to KTIME_MAX here with cpu_base->lock
1318 * held to prevent that a timer is enqueued in our queue via
1319 * the migration code. This does not affect enqueueing of
1320 * timers which run their callback and need to be requeued on
1323 cpu_base->expires_next.tv64 = KTIME_MAX;
1325 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1326 struct hrtimer_clock_base *base;
1327 struct timerqueue_node *node;
1330 if (!(cpu_base->active_bases & (1 << i)))
1333 base = cpu_base->clock_base + i;
1334 basenow = ktime_add(now, base->offset);
1336 while ((node = timerqueue_getnext(&base->active))) {
1337 struct hrtimer *timer;
1339 timer = container_of(node, struct hrtimer, node);
1342 * The immediate goal for using the softexpires is
1343 * minimizing wakeups, not running timers at the
1344 * earliest interrupt after their soft expiration.
1345 * This allows us to avoid using a Priority Search
1346 * Tree, which can answer a stabbing querry for
1347 * overlapping intervals and instead use the simple
1348 * BST we already have.
1349 * We don't add extra wakeups by delaying timers that
1350 * are right-of a not yet expired timer, because that
1351 * timer will have to trigger a wakeup anyway.
1354 if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer)) {
1357 expires = ktime_sub(hrtimer_get_expires(timer),
1359 if (expires.tv64 < 0)
1360 expires.tv64 = KTIME_MAX;
1361 if (expires.tv64 < expires_next.tv64)
1362 expires_next = expires;
1366 __run_hrtimer(timer, &basenow);
1371 * Store the new expiry value so the migration code can verify
1374 cpu_base->expires_next = expires_next;
1375 raw_spin_unlock(&cpu_base->lock);
1377 /* Reprogramming necessary ? */
1378 if (expires_next.tv64 == KTIME_MAX ||
1379 !tick_program_event(expires_next, 0)) {
1380 cpu_base->hang_detected = 0;
1385 * The next timer was already expired due to:
1387 * - long lasting callbacks
1388 * - being scheduled away when running in a VM
1390 * We need to prevent that we loop forever in the hrtimer
1391 * interrupt routine. We give it 3 attempts to avoid
1392 * overreacting on some spurious event.
1394 * Acquire base lock for updating the offsets and retrieving
1397 raw_spin_lock(&cpu_base->lock);
1398 now = hrtimer_update_base(cpu_base);
1399 cpu_base->nr_retries++;
1403 * Give the system a chance to do something else than looping
1404 * here. We stored the entry time, so we know exactly how long
1405 * we spent here. We schedule the next event this amount of
1408 cpu_base->nr_hangs++;
1409 cpu_base->hang_detected = 1;
1410 raw_spin_unlock(&cpu_base->lock);
1411 delta = ktime_sub(now, entry_time);
1412 if (delta.tv64 > cpu_base->max_hang_time.tv64)
1413 cpu_base->max_hang_time = delta;
1415 * Limit it to a sensible value as we enforce a longer
1416 * delay. Give the CPU at least 100ms to catch up.
1418 if (delta.tv64 > 100 * NSEC_PER_MSEC)
1419 expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1421 expires_next = ktime_add(now, delta);
1422 tick_program_event(expires_next, 1);
1423 printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n",
1424 ktime_to_ns(delta));
1428 * local version of hrtimer_peek_ahead_timers() called with interrupts
1431 static void __hrtimer_peek_ahead_timers(void)
1433 struct tick_device *td;
1435 if (!hrtimer_hres_active())
1438 td = &__get_cpu_var(tick_cpu_device);
1439 if (td && td->evtdev)
1440 hrtimer_interrupt(td->evtdev);
1444 * hrtimer_peek_ahead_timers -- run soft-expired timers now
1446 * hrtimer_peek_ahead_timers will peek at the timer queue of
1447 * the current cpu and check if there are any timers for which
1448 * the soft expires time has passed. If any such timers exist,
1449 * they are run immediately and then removed from the timer queue.
1452 void hrtimer_peek_ahead_timers(void)
1454 unsigned long flags;
1456 local_irq_save(flags);
1457 __hrtimer_peek_ahead_timers();
1458 local_irq_restore(flags);
1461 static void run_hrtimer_softirq(struct softirq_action *h)
1463 hrtimer_peek_ahead_timers();
1466 #else /* CONFIG_HIGH_RES_TIMERS */
1468 static inline void __hrtimer_peek_ahead_timers(void) { }
1470 #endif /* !CONFIG_HIGH_RES_TIMERS */
1473 * Called from timer softirq every jiffy, expire hrtimers:
1475 * For HRT its the fall back code to run the softirq in the timer
1476 * softirq context in case the hrtimer initialization failed or has
1477 * not been done yet.
1479 void hrtimer_run_pending(void)
1481 if (hrtimer_hres_active())
1485 * This _is_ ugly: We have to check in the softirq context,
1486 * whether we can switch to highres and / or nohz mode. The
1487 * clocksource switch happens in the timer interrupt with
1488 * xtime_lock held. Notification from there only sets the
1489 * check bit in the tick_oneshot code, otherwise we might
1490 * deadlock vs. xtime_lock.
1492 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1493 hrtimer_switch_to_hres();
1497 * Called from hardirq context every jiffy
1499 void hrtimer_run_queues(void)
1501 struct timerqueue_node *node;
1502 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1503 struct hrtimer_clock_base *base;
1504 int index, gettime = 1;
1506 if (hrtimer_hres_active())
1509 for (index = 0; index < HRTIMER_MAX_CLOCK_BASES; index++) {
1510 base = &cpu_base->clock_base[index];
1511 if (!timerqueue_getnext(&base->active))
1515 hrtimer_get_softirq_time(cpu_base);
1519 raw_spin_lock(&cpu_base->lock);
1521 while ((node = timerqueue_getnext(&base->active))) {
1522 struct hrtimer *timer;
1524 timer = container_of(node, struct hrtimer, node);
1525 if (base->softirq_time.tv64 <=
1526 hrtimer_get_expires_tv64(timer))
1529 __run_hrtimer(timer, &base->softirq_time);
1531 raw_spin_unlock(&cpu_base->lock);
1536 * Sleep related functions:
1538 static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1540 struct hrtimer_sleeper *t =
1541 container_of(timer, struct hrtimer_sleeper, timer);
1542 struct task_struct *task = t->task;
1546 wake_up_process(task);
1548 return HRTIMER_NORESTART;
1551 void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1553 sl->timer.function = hrtimer_wakeup;
1556 EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
1558 static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1560 hrtimer_init_sleeper(t, current);
1563 set_current_state(TASK_INTERRUPTIBLE);
1564 hrtimer_start_expires(&t->timer, mode);
1565 if (!hrtimer_active(&t->timer))
1568 if (likely(t->task))
1569 freezable_schedule();
1571 hrtimer_cancel(&t->timer);
1572 mode = HRTIMER_MODE_ABS;
1574 } while (t->task && !signal_pending(current));
1576 __set_current_state(TASK_RUNNING);
1578 return t->task == NULL;
1581 static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp)
1583 struct timespec rmt;
1586 rem = hrtimer_expires_remaining(timer);
1589 rmt = ktime_to_timespec(rem);
1591 if (copy_to_user(rmtp, &rmt, sizeof(*rmtp)))
1597 long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1599 struct hrtimer_sleeper t;
1600 struct timespec __user *rmtp;
1603 hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid,
1605 hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
1607 if (do_nanosleep(&t, HRTIMER_MODE_ABS))
1610 rmtp = restart->nanosleep.rmtp;
1612 ret = update_rmtp(&t.timer, rmtp);
1617 /* The other values in restart are already filled in */
1618 ret = -ERESTART_RESTARTBLOCK;
1620 destroy_hrtimer_on_stack(&t.timer);
1624 long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
1625 const enum hrtimer_mode mode, const clockid_t clockid)
1627 struct restart_block *restart;
1628 struct hrtimer_sleeper t;
1630 unsigned long slack;
1632 slack = current->timer_slack_ns;
1633 if (rt_task(current))
1636 hrtimer_init_on_stack(&t.timer, clockid, mode);
1637 hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack);
1638 if (do_nanosleep(&t, mode))
1641 /* Absolute timers do not update the rmtp value and restart: */
1642 if (mode == HRTIMER_MODE_ABS) {
1643 ret = -ERESTARTNOHAND;
1648 ret = update_rmtp(&t.timer, rmtp);
1653 restart = ¤t_thread_info()->restart_block;
1654 restart->fn = hrtimer_nanosleep_restart;
1655 restart->nanosleep.clockid = t.timer.base->clockid;
1656 restart->nanosleep.rmtp = rmtp;
1657 restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
1659 ret = -ERESTART_RESTARTBLOCK;
1661 destroy_hrtimer_on_stack(&t.timer);
1665 SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp,
1666 struct timespec __user *, rmtp)
1670 if (copy_from_user(&tu, rqtp, sizeof(tu)))
1673 if (!timespec_valid(&tu))
1676 return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1680 * Functions related to boot-time initialization:
1682 static void __cpuinit init_hrtimers_cpu(int cpu)
1684 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1687 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1688 cpu_base->clock_base[i].cpu_base = cpu_base;
1689 timerqueue_init_head(&cpu_base->clock_base[i].active);
1692 hrtimer_init_hres(cpu_base);
1695 #ifdef CONFIG_HOTPLUG_CPU
1697 static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1698 struct hrtimer_clock_base *new_base)
1700 struct hrtimer *timer;
1701 struct timerqueue_node *node;
1703 while ((node = timerqueue_getnext(&old_base->active))) {
1704 timer = container_of(node, struct hrtimer, node);
1705 BUG_ON(hrtimer_callback_running(timer));
1706 debug_deactivate(timer);
1709 * Mark it as STATE_MIGRATE not INACTIVE otherwise the
1710 * timer could be seen as !active and just vanish away
1711 * under us on another CPU
1713 __remove_hrtimer(timer, old_base, HRTIMER_STATE_MIGRATE, 0);
1714 timer->base = new_base;
1716 * Enqueue the timers on the new cpu. This does not
1717 * reprogram the event device in case the timer
1718 * expires before the earliest on this CPU, but we run
1719 * hrtimer_interrupt after we migrated everything to
1720 * sort out already expired timers and reprogram the
1723 enqueue_hrtimer(timer, new_base);
1725 /* Clear the migration state bit */
1726 timer->state &= ~HRTIMER_STATE_MIGRATE;
1730 static void migrate_hrtimers(int scpu)
1732 struct hrtimer_cpu_base *old_base, *new_base;
1735 BUG_ON(cpu_online(scpu));
1736 tick_cancel_sched_timer(scpu);
1738 local_irq_disable();
1739 old_base = &per_cpu(hrtimer_bases, scpu);
1740 new_base = &__get_cpu_var(hrtimer_bases);
1742 * The caller is globally serialized and nobody else
1743 * takes two locks at once, deadlock is not possible.
1745 raw_spin_lock(&new_base->lock);
1746 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1748 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1749 migrate_hrtimer_list(&old_base->clock_base[i],
1750 &new_base->clock_base[i]);
1753 raw_spin_unlock(&old_base->lock);
1754 raw_spin_unlock(&new_base->lock);
1756 /* Check, if we got expired work to do */
1757 __hrtimer_peek_ahead_timers();
1761 #endif /* CONFIG_HOTPLUG_CPU */
1763 static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self,
1764 unsigned long action, void *hcpu)
1766 int scpu = (long)hcpu;
1770 case CPU_UP_PREPARE:
1771 case CPU_UP_PREPARE_FROZEN:
1772 init_hrtimers_cpu(scpu);
1775 #ifdef CONFIG_HOTPLUG_CPU
1777 case CPU_DYING_FROZEN:
1778 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DYING, &scpu);
1781 case CPU_DEAD_FROZEN:
1783 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &scpu);
1784 migrate_hrtimers(scpu);
1796 static struct notifier_block __cpuinitdata hrtimers_nb = {
1797 .notifier_call = hrtimer_cpu_notify,
1800 void __init hrtimers_init(void)
1802 hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
1803 (void *)(long)smp_processor_id());
1804 register_cpu_notifier(&hrtimers_nb);
1805 #ifdef CONFIG_HIGH_RES_TIMERS
1806 open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq);
1811 * schedule_hrtimeout_range_clock - sleep until timeout
1812 * @expires: timeout value (ktime_t)
1813 * @delta: slack in expires timeout (ktime_t)
1814 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1815 * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME
1818 schedule_hrtimeout_range_clock(ktime_t *expires, unsigned long delta,
1819 const enum hrtimer_mode mode, int clock)
1821 struct hrtimer_sleeper t;
1824 * Optimize when a zero timeout value is given. It does not
1825 * matter whether this is an absolute or a relative time.
1827 if (expires && !expires->tv64) {
1828 __set_current_state(TASK_RUNNING);
1833 * A NULL parameter means "infinite"
1837 __set_current_state(TASK_RUNNING);
1841 hrtimer_init_on_stack(&t.timer, clock, mode);
1842 hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
1844 hrtimer_init_sleeper(&t, current);
1846 hrtimer_start_expires(&t.timer, mode);
1847 if (!hrtimer_active(&t.timer))
1853 hrtimer_cancel(&t.timer);
1854 destroy_hrtimer_on_stack(&t.timer);
1856 __set_current_state(TASK_RUNNING);
1858 return !t.task ? 0 : -EINTR;
1862 * schedule_hrtimeout_range - sleep until timeout
1863 * @expires: timeout value (ktime_t)
1864 * @delta: slack in expires timeout (ktime_t)
1865 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1867 * Make the current task sleep until the given expiry time has
1868 * elapsed. The routine will return immediately unless
1869 * the current task state has been set (see set_current_state()).
1871 * The @delta argument gives the kernel the freedom to schedule the
1872 * actual wakeup to a time that is both power and performance friendly.
1873 * The kernel give the normal best effort behavior for "@expires+@delta",
1874 * but may decide to fire the timer earlier, but no earlier than @expires.
1876 * You can set the task state as follows -
1878 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1879 * pass before the routine returns.
1881 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1882 * delivered to the current task.
1884 * The current task state is guaranteed to be TASK_RUNNING when this
1887 * Returns 0 when the timer has expired otherwise -EINTR
1889 int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta,
1890 const enum hrtimer_mode mode)
1892 return schedule_hrtimeout_range_clock(expires, delta, mode,
1895 EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
1898 * schedule_hrtimeout - sleep until timeout
1899 * @expires: timeout value (ktime_t)
1900 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1902 * Make the current task sleep until the given expiry time has
1903 * elapsed. The routine will return immediately unless
1904 * the current task state has been set (see set_current_state()).
1906 * You can set the task state as follows -
1908 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1909 * pass before the routine returns.
1911 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1912 * delivered to the current task.
1914 * The current task state is guaranteed to be TASK_RUNNING when this
1917 * Returns 0 when the timer has expired otherwise -EINTR
1919 int __sched schedule_hrtimeout(ktime_t *expires,
1920 const enum hrtimer_mode mode)
1922 return schedule_hrtimeout_range(expires, 0, mode);
1924 EXPORT_SYMBOL_GPL(schedule_hrtimeout);