2 * linux/kernel/hrtimer.c
4 * Copyright(C) 2005-2006, Thomas Gleixner <tglx@linutronix.de>
5 * Copyright(C) 2005-2007, Red Hat, Inc., Ingo Molnar
6 * Copyright(C) 2006-2007 Timesys Corp., Thomas Gleixner
8 * High-resolution kernel timers
10 * In contrast to the low-resolution timeout API implemented in
11 * kernel/timer.c, hrtimers provide finer resolution and accuracy
12 * depending on system configuration and capabilities.
14 * These timers are currently used for:
18 * - precise in-kernel timing
20 * Started by: Thomas Gleixner and Ingo Molnar
23 * based on kernel/timer.c
25 * Help, testing, suggestions, bugfixes, improvements were
28 * George Anzinger, Andrew Morton, Steven Rostedt, Roman Zippel
31 * For licencing details see kernel-base/COPYING
34 #include <linux/cpu.h>
35 #include <linux/export.h>
36 #include <linux/percpu.h>
37 #include <linux/hrtimer.h>
38 #include <linux/notifier.h>
39 #include <linux/syscalls.h>
40 #include <linux/kallsyms.h>
41 #include <linux/interrupt.h>
42 #include <linux/tick.h>
43 #include <linux/seq_file.h>
44 #include <linux/err.h>
45 #include <linux/debugobjects.h>
46 #include <linux/sched.h>
47 #include <linux/sched/sysctl.h>
48 #include <linux/sched/rt.h>
49 #include <linux/sched/deadline.h>
50 #include <linux/timer.h>
51 #include <linux/freezer.h>
53 #include <asm/uaccess.h>
55 #include <trace/events/timer.h>
60 * There are more clockids then hrtimer bases. Thus, we index
61 * into the timer bases by the hrtimer_base_type enum. When trying
62 * to reach a base using a clockid, hrtimer_clockid_to_base()
63 * is used to convert from clockid to the proper hrtimer_base_type.
65 DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
68 .lock = __RAW_SPIN_LOCK_UNLOCKED(hrtimer_bases.lock),
72 .index = HRTIMER_BASE_MONOTONIC,
73 .clockid = CLOCK_MONOTONIC,
74 .get_time = &ktime_get,
75 .resolution = KTIME_LOW_RES,
78 .index = HRTIMER_BASE_REALTIME,
79 .clockid = CLOCK_REALTIME,
80 .get_time = &ktime_get_real,
81 .resolution = KTIME_LOW_RES,
84 .index = HRTIMER_BASE_BOOTTIME,
85 .clockid = CLOCK_BOOTTIME,
86 .get_time = &ktime_get_boottime,
87 .resolution = KTIME_LOW_RES,
90 .index = HRTIMER_BASE_TAI,
92 .get_time = &ktime_get_clocktai,
93 .resolution = KTIME_LOW_RES,
98 static const int hrtimer_clock_to_base_table[MAX_CLOCKS] = {
99 [CLOCK_REALTIME] = HRTIMER_BASE_REALTIME,
100 [CLOCK_MONOTONIC] = HRTIMER_BASE_MONOTONIC,
101 [CLOCK_BOOTTIME] = HRTIMER_BASE_BOOTTIME,
102 [CLOCK_TAI] = HRTIMER_BASE_TAI,
105 static inline int hrtimer_clockid_to_base(clockid_t clock_id)
107 return hrtimer_clock_to_base_table[clock_id];
112 * Get the coarse grained time at the softirq based on xtime and
115 static void hrtimer_get_softirq_time(struct hrtimer_cpu_base *base)
117 ktime_t xtim, mono, boot;
118 struct timespec xts, tom, slp;
121 get_xtime_and_monotonic_and_sleep_offset(&xts, &tom, &slp);
122 tai_offset = timekeeping_get_tai_offset();
124 xtim = timespec_to_ktime(xts);
125 mono = ktime_add(xtim, timespec_to_ktime(tom));
126 boot = ktime_add(mono, timespec_to_ktime(slp));
127 base->clock_base[HRTIMER_BASE_REALTIME].softirq_time = xtim;
128 base->clock_base[HRTIMER_BASE_MONOTONIC].softirq_time = mono;
129 base->clock_base[HRTIMER_BASE_BOOTTIME].softirq_time = boot;
130 base->clock_base[HRTIMER_BASE_TAI].softirq_time =
131 ktime_add(xtim, ktime_set(tai_offset, 0));
135 * Functions and macros which are different for UP/SMP systems are kept in a
141 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
142 * means that all timers which are tied to this base via timer->base are
143 * locked, and the base itself is locked too.
145 * So __run_timers/migrate_timers can safely modify all timers which could
146 * be found on the lists/queues.
148 * When the timer's base is locked, and the timer removed from list, it is
149 * possible to set timer->base = NULL and drop the lock: the timer remains
153 struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
154 unsigned long *flags)
156 struct hrtimer_clock_base *base;
160 if (likely(base != NULL)) {
161 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
162 if (likely(base == timer->base))
164 /* The timer has migrated to another CPU: */
165 raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
173 * Get the preferred target CPU for NOHZ
175 static int hrtimer_get_target(int this_cpu, int pinned)
177 #ifdef CONFIG_NO_HZ_COMMON
178 if (!pinned && get_sysctl_timer_migration() && idle_cpu(this_cpu))
179 return get_nohz_timer_target();
185 * With HIGHRES=y we do not migrate the timer when it is expiring
186 * before the next event on the target cpu because we cannot reprogram
187 * the target cpu hardware and we would cause it to fire late.
189 * Called with cpu_base->lock of target cpu held.
192 hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base)
194 #ifdef CONFIG_HIGH_RES_TIMERS
197 if (!new_base->cpu_base->hres_active)
200 expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset);
201 return expires.tv64 <= new_base->cpu_base->expires_next.tv64;
208 * Switch the timer base to the current CPU when possible.
210 static inline struct hrtimer_clock_base *
211 switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
214 struct hrtimer_clock_base *new_base;
215 struct hrtimer_cpu_base *new_cpu_base;
216 int this_cpu = smp_processor_id();
217 int cpu = hrtimer_get_target(this_cpu, pinned);
218 int basenum = base->index;
221 new_cpu_base = &per_cpu(hrtimer_bases, cpu);
222 new_base = &new_cpu_base->clock_base[basenum];
224 if (base != new_base) {
226 * We are trying to move timer to new_base.
227 * However we can't change timer's base while it is running,
228 * so we keep it on the same CPU. No hassle vs. reprogramming
229 * the event source in the high resolution case. The softirq
230 * code will take care of this when the timer function has
231 * completed. There is no conflict as we hold the lock until
232 * the timer is enqueued.
234 if (unlikely(hrtimer_callback_running(timer)))
237 /* See the comment in lock_timer_base() */
239 raw_spin_unlock(&base->cpu_base->lock);
240 raw_spin_lock(&new_base->cpu_base->lock);
242 if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) {
244 raw_spin_unlock(&new_base->cpu_base->lock);
245 raw_spin_lock(&base->cpu_base->lock);
249 timer->base = new_base;
254 #else /* CONFIG_SMP */
256 static inline struct hrtimer_clock_base *
257 lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
259 struct hrtimer_clock_base *base = timer->base;
261 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
266 # define switch_hrtimer_base(t, b, p) (b)
268 #endif /* !CONFIG_SMP */
271 * Functions for the union type storage format of ktime_t which are
272 * too large for inlining:
274 #if BITS_PER_LONG < 64
275 # ifndef CONFIG_KTIME_SCALAR
277 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
279 * @nsec: the scalar nsec value to add
281 * Returns the sum of kt and nsec in ktime_t format
283 ktime_t ktime_add_ns(const ktime_t kt, u64 nsec)
287 if (likely(nsec < NSEC_PER_SEC)) {
290 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
292 /* Make sure nsec fits into long */
293 if (unlikely(nsec > KTIME_SEC_MAX))
294 return (ktime_t){ .tv64 = KTIME_MAX };
296 tmp = ktime_set((long)nsec, rem);
299 return ktime_add(kt, tmp);
302 EXPORT_SYMBOL_GPL(ktime_add_ns);
305 * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable
307 * @nsec: the scalar nsec value to subtract
309 * Returns the subtraction of @nsec from @kt in ktime_t format
311 ktime_t ktime_sub_ns(const ktime_t kt, u64 nsec)
315 if (likely(nsec < NSEC_PER_SEC)) {
318 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
320 tmp = ktime_set((long)nsec, rem);
323 return ktime_sub(kt, tmp);
326 EXPORT_SYMBOL_GPL(ktime_sub_ns);
327 # endif /* !CONFIG_KTIME_SCALAR */
330 * Divide a ktime value by a nanosecond value
332 u64 ktime_divns(const ktime_t kt, s64 div)
337 dclc = ktime_to_ns(kt);
338 /* Make sure the divisor is less than 2^32: */
344 do_div(dclc, (unsigned long) div);
348 #endif /* BITS_PER_LONG >= 64 */
351 * Add two ktime values and do a safety check for overflow:
353 ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
355 ktime_t res = ktime_add(lhs, rhs);
358 * We use KTIME_SEC_MAX here, the maximum timeout which we can
359 * return to user space in a timespec:
361 if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64)
362 res = ktime_set(KTIME_SEC_MAX, 0);
367 EXPORT_SYMBOL_GPL(ktime_add_safe);
369 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
371 static struct debug_obj_descr hrtimer_debug_descr;
373 static void *hrtimer_debug_hint(void *addr)
375 return ((struct hrtimer *) addr)->function;
379 * fixup_init is called when:
380 * - an active object is initialized
382 static int hrtimer_fixup_init(void *addr, enum debug_obj_state state)
384 struct hrtimer *timer = addr;
387 case ODEBUG_STATE_ACTIVE:
388 hrtimer_cancel(timer);
389 debug_object_init(timer, &hrtimer_debug_descr);
397 * fixup_activate is called when:
398 * - an active object is activated
399 * - an unknown object is activated (might be a statically initialized object)
401 static int hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
405 case ODEBUG_STATE_NOTAVAILABLE:
409 case ODEBUG_STATE_ACTIVE:
418 * fixup_free is called when:
419 * - an active object is freed
421 static int hrtimer_fixup_free(void *addr, enum debug_obj_state state)
423 struct hrtimer *timer = addr;
426 case ODEBUG_STATE_ACTIVE:
427 hrtimer_cancel(timer);
428 debug_object_free(timer, &hrtimer_debug_descr);
435 static struct debug_obj_descr hrtimer_debug_descr = {
437 .debug_hint = hrtimer_debug_hint,
438 .fixup_init = hrtimer_fixup_init,
439 .fixup_activate = hrtimer_fixup_activate,
440 .fixup_free = hrtimer_fixup_free,
443 static inline void debug_hrtimer_init(struct hrtimer *timer)
445 debug_object_init(timer, &hrtimer_debug_descr);
448 static inline void debug_hrtimer_activate(struct hrtimer *timer)
450 debug_object_activate(timer, &hrtimer_debug_descr);
453 static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
455 debug_object_deactivate(timer, &hrtimer_debug_descr);
458 static inline void debug_hrtimer_free(struct hrtimer *timer)
460 debug_object_free(timer, &hrtimer_debug_descr);
463 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
464 enum hrtimer_mode mode);
466 void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
467 enum hrtimer_mode mode)
469 debug_object_init_on_stack(timer, &hrtimer_debug_descr);
470 __hrtimer_init(timer, clock_id, mode);
472 EXPORT_SYMBOL_GPL(hrtimer_init_on_stack);
474 void destroy_hrtimer_on_stack(struct hrtimer *timer)
476 debug_object_free(timer, &hrtimer_debug_descr);
480 static inline void debug_hrtimer_init(struct hrtimer *timer) { }
481 static inline void debug_hrtimer_activate(struct hrtimer *timer) { }
482 static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
486 debug_init(struct hrtimer *timer, clockid_t clockid,
487 enum hrtimer_mode mode)
489 debug_hrtimer_init(timer);
490 trace_hrtimer_init(timer, clockid, mode);
493 static inline void debug_activate(struct hrtimer *timer)
495 debug_hrtimer_activate(timer);
496 trace_hrtimer_start(timer);
499 static inline void debug_deactivate(struct hrtimer *timer)
501 debug_hrtimer_deactivate(timer);
502 trace_hrtimer_cancel(timer);
505 /* High resolution timer related functions */
506 #ifdef CONFIG_HIGH_RES_TIMERS
509 * High resolution timer enabled ?
511 static int hrtimer_hres_enabled __read_mostly = 1;
514 * Enable / Disable high resolution mode
516 static int __init setup_hrtimer_hres(char *str)
518 if (!strcmp(str, "off"))
519 hrtimer_hres_enabled = 0;
520 else if (!strcmp(str, "on"))
521 hrtimer_hres_enabled = 1;
527 __setup("highres=", setup_hrtimer_hres);
530 * hrtimer_high_res_enabled - query, if the highres mode is enabled
532 static inline int hrtimer_is_hres_enabled(void)
534 return hrtimer_hres_enabled;
538 * Is the high resolution mode active ?
540 static inline int hrtimer_hres_active(void)
542 return __this_cpu_read(hrtimer_bases.hres_active);
546 * Reprogram the event source with checking both queues for the
548 * Called with interrupts disabled and base->lock held
551 hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
554 struct hrtimer_clock_base *base = cpu_base->clock_base;
555 ktime_t expires, expires_next;
557 expires_next.tv64 = KTIME_MAX;
559 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
560 struct hrtimer *timer;
561 struct timerqueue_node *next;
563 next = timerqueue_getnext(&base->active);
566 timer = container_of(next, struct hrtimer, node);
568 expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
570 * clock_was_set() has changed base->offset so the
571 * result might be negative. Fix it up to prevent a
572 * false positive in clockevents_program_event()
574 if (expires.tv64 < 0)
576 if (expires.tv64 < expires_next.tv64)
577 expires_next = expires;
580 if (skip_equal && expires_next.tv64 == cpu_base->expires_next.tv64)
583 cpu_base->expires_next.tv64 = expires_next.tv64;
586 * If a hang was detected in the last timer interrupt then we
587 * leave the hang delay active in the hardware. We want the
588 * system to make progress. That also prevents the following
590 * T1 expires 50ms from now
591 * T2 expires 5s from now
593 * T1 is removed, so this code is called and would reprogram
594 * the hardware to 5s from now. Any hrtimer_start after that
595 * will not reprogram the hardware due to hang_detected being
596 * set. So we'd effectivly block all timers until the T2 event
599 if (cpu_base->hang_detected)
602 if (cpu_base->expires_next.tv64 != KTIME_MAX)
603 tick_program_event(cpu_base->expires_next, 1);
607 * Shared reprogramming for clock_realtime and clock_monotonic
609 * When a timer is enqueued and expires earlier than the already enqueued
610 * timers, we have to check, whether it expires earlier than the timer for
611 * which the clock event device was armed.
613 * Called with interrupts disabled and base->cpu_base.lock held
615 static int hrtimer_reprogram(struct hrtimer *timer,
616 struct hrtimer_clock_base *base)
618 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
619 ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
622 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
625 * When the callback is running, we do not reprogram the clock event
626 * device. The timer callback is either running on a different CPU or
627 * the callback is executed in the hrtimer_interrupt context. The
628 * reprogramming is handled either by the softirq, which called the
629 * callback or at the end of the hrtimer_interrupt.
631 if (hrtimer_callback_running(timer))
635 * CLOCK_REALTIME timer might be requested with an absolute
636 * expiry time which is less than base->offset. Nothing wrong
637 * about that, just avoid to call into the tick code, which
638 * has now objections against negative expiry values.
640 if (expires.tv64 < 0)
643 if (expires.tv64 >= cpu_base->expires_next.tv64)
647 * If a hang was detected in the last timer interrupt then we
648 * do not schedule a timer which is earlier than the expiry
649 * which we enforced in the hang detection. We want the system
652 if (cpu_base->hang_detected)
656 * Clockevents returns -ETIME, when the event was in the past.
658 res = tick_program_event(expires, 0);
659 if (!IS_ERR_VALUE(res))
660 cpu_base->expires_next = expires;
665 * Initialize the high resolution related parts of cpu_base
667 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
669 base->expires_next.tv64 = KTIME_MAX;
670 base->hres_active = 0;
674 * When High resolution timers are active, try to reprogram. Note, that in case
675 * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry
676 * check happens. The timer gets enqueued into the rbtree. The reprogramming
677 * and expiry check is done in the hrtimer_interrupt or in the softirq.
679 static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
680 struct hrtimer_clock_base *base)
682 return base->cpu_base->hres_active && hrtimer_reprogram(timer, base);
685 static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base)
687 ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset;
688 ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset;
689 ktime_t *offs_tai = &base->clock_base[HRTIMER_BASE_TAI].offset;
691 return ktime_get_update_offsets(offs_real, offs_boot, offs_tai);
695 * Retrigger next event is called after clock was set
697 * Called with interrupts disabled via on_each_cpu()
699 static void retrigger_next_event(void *arg)
701 struct hrtimer_cpu_base *base = &__get_cpu_var(hrtimer_bases);
703 if (!hrtimer_hres_active())
706 raw_spin_lock(&base->lock);
707 hrtimer_update_base(base);
708 hrtimer_force_reprogram(base, 0);
709 raw_spin_unlock(&base->lock);
713 * Switch to high resolution mode
715 static int hrtimer_switch_to_hres(void)
717 int i, cpu = smp_processor_id();
718 struct hrtimer_cpu_base *base = &per_cpu(hrtimer_bases, cpu);
721 if (base->hres_active)
724 local_irq_save(flags);
726 if (tick_init_highres()) {
727 local_irq_restore(flags);
728 printk(KERN_WARNING "Could not switch to high resolution "
729 "mode on CPU %d\n", cpu);
732 base->hres_active = 1;
733 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
734 base->clock_base[i].resolution = KTIME_HIGH_RES;
736 tick_setup_sched_timer();
737 /* "Retrigger" the interrupt to get things going */
738 retrigger_next_event(NULL);
739 local_irq_restore(flags);
743 static void clock_was_set_work(struct work_struct *work)
748 static DECLARE_WORK(hrtimer_work, clock_was_set_work);
751 * Called from timekeeping and resume code to reprogramm the hrtimer
752 * interrupt device on all cpus.
754 void clock_was_set_delayed(void)
756 schedule_work(&hrtimer_work);
761 static inline int hrtimer_hres_active(void) { return 0; }
762 static inline int hrtimer_is_hres_enabled(void) { return 0; }
763 static inline int hrtimer_switch_to_hres(void) { return 0; }
765 hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { }
766 static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
767 struct hrtimer_clock_base *base)
771 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { }
772 static inline void retrigger_next_event(void *arg) { }
774 #endif /* CONFIG_HIGH_RES_TIMERS */
777 * Clock realtime was set
779 * Change the offset of the realtime clock vs. the monotonic
782 * We might have to reprogram the high resolution timer interrupt. On
783 * SMP we call the architecture specific code to retrigger _all_ high
784 * resolution timer interrupts. On UP we just disable interrupts and
785 * call the high resolution interrupt code.
787 void clock_was_set(void)
789 #ifdef CONFIG_HIGH_RES_TIMERS
790 /* Retrigger the CPU local events everywhere */
791 on_each_cpu(retrigger_next_event, NULL, 1);
793 timerfd_clock_was_set();
797 * During resume we might have to reprogram the high resolution timer
798 * interrupt on all online CPUs. However, all other CPUs will be
799 * stopped with IRQs interrupts disabled so the clock_was_set() call
802 void hrtimers_resume(void)
804 WARN_ONCE(!irqs_disabled(),
805 KERN_INFO "hrtimers_resume() called with IRQs enabled!");
807 /* Retrigger on the local CPU */
808 retrigger_next_event(NULL);
809 /* And schedule a retrigger for all others */
810 clock_was_set_delayed();
813 static inline void timer_stats_hrtimer_set_start_info(struct hrtimer *timer)
815 #ifdef CONFIG_TIMER_STATS
816 if (timer->start_site)
818 timer->start_site = __builtin_return_address(0);
819 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
820 timer->start_pid = current->pid;
824 static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer *timer)
826 #ifdef CONFIG_TIMER_STATS
827 timer->start_site = NULL;
831 static inline void timer_stats_account_hrtimer(struct hrtimer *timer)
833 #ifdef CONFIG_TIMER_STATS
834 if (likely(!timer_stats_active))
836 timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
837 timer->function, timer->start_comm, 0);
842 * Counterpart to lock_hrtimer_base above:
845 void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
847 raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
851 * hrtimer_forward - forward the timer expiry
852 * @timer: hrtimer to forward
853 * @now: forward past this time
854 * @interval: the interval to forward
856 * Forward the timer expiry so it will expire in the future.
857 * Returns the number of overruns.
859 u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
864 delta = ktime_sub(now, hrtimer_get_expires(timer));
869 if (interval.tv64 < timer->base->resolution.tv64)
870 interval.tv64 = timer->base->resolution.tv64;
872 if (unlikely(delta.tv64 >= interval.tv64)) {
873 s64 incr = ktime_to_ns(interval);
875 orun = ktime_divns(delta, incr);
876 hrtimer_add_expires_ns(timer, incr * orun);
877 if (hrtimer_get_expires_tv64(timer) > now.tv64)
880 * This (and the ktime_add() below) is the
881 * correction for exact:
885 hrtimer_add_expires(timer, interval);
889 EXPORT_SYMBOL_GPL(hrtimer_forward);
892 * enqueue_hrtimer - internal function to (re)start a timer
894 * The timer is inserted in expiry order. Insertion into the
895 * red black tree is O(log(n)). Must hold the base lock.
897 * Returns 1 when the new timer is the leftmost timer in the tree.
899 static int enqueue_hrtimer(struct hrtimer *timer,
900 struct hrtimer_clock_base *base)
902 debug_activate(timer);
904 timerqueue_add(&base->active, &timer->node);
905 base->cpu_base->active_bases |= 1 << base->index;
908 * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
909 * state of a possibly running callback.
911 timer->state |= HRTIMER_STATE_ENQUEUED;
913 return (&timer->node == base->active.next);
917 * __remove_hrtimer - internal function to remove a timer
919 * Caller must hold the base lock.
921 * High resolution timer mode reprograms the clock event device when the
922 * timer is the one which expires next. The caller can disable this by setting
923 * reprogram to zero. This is useful, when the context does a reprogramming
924 * anyway (e.g. timer interrupt)
926 static void __remove_hrtimer(struct hrtimer *timer,
927 struct hrtimer_clock_base *base,
928 unsigned long newstate, int reprogram)
930 struct timerqueue_node *next_timer;
931 if (!(timer->state & HRTIMER_STATE_ENQUEUED))
934 next_timer = timerqueue_getnext(&base->active);
935 timerqueue_del(&base->active, &timer->node);
936 if (&timer->node == next_timer) {
937 #ifdef CONFIG_HIGH_RES_TIMERS
938 /* Reprogram the clock event device. if enabled */
939 if (reprogram && hrtimer_hres_active()) {
942 expires = ktime_sub(hrtimer_get_expires(timer),
944 if (base->cpu_base->expires_next.tv64 == expires.tv64)
945 hrtimer_force_reprogram(base->cpu_base, 1);
949 if (!timerqueue_getnext(&base->active))
950 base->cpu_base->active_bases &= ~(1 << base->index);
952 timer->state = newstate;
956 * remove hrtimer, called with base lock held
959 remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base)
961 if (hrtimer_is_queued(timer)) {
966 * Remove the timer and force reprogramming when high
967 * resolution mode is active and the timer is on the current
968 * CPU. If we remove a timer on another CPU, reprogramming is
969 * skipped. The interrupt event on this CPU is fired and
970 * reprogramming happens in the interrupt handler. This is a
971 * rare case and less expensive than a smp call.
973 debug_deactivate(timer);
974 timer_stats_hrtimer_clear_start_info(timer);
975 reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases);
977 * We must preserve the CALLBACK state flag here,
978 * otherwise we could move the timer base in
979 * switch_hrtimer_base.
981 state = timer->state & HRTIMER_STATE_CALLBACK;
982 __remove_hrtimer(timer, base, state, reprogram);
988 int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
989 unsigned long delta_ns, const enum hrtimer_mode mode,
992 struct hrtimer_clock_base *base, *new_base;
996 base = lock_hrtimer_base(timer, &flags);
998 /* Remove an active timer from the queue: */
999 ret = remove_hrtimer(timer, base);
1001 /* Switch the timer base, if necessary: */
1002 new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED);
1004 if (mode & HRTIMER_MODE_REL) {
1005 tim = ktime_add_safe(tim, new_base->get_time());
1007 * CONFIG_TIME_LOW_RES is a temporary way for architectures
1008 * to signal that they simply return xtime in
1009 * do_gettimeoffset(). In this case we want to round up by
1010 * resolution when starting a relative timer, to avoid short
1011 * timeouts. This will go away with the GTOD framework.
1013 #ifdef CONFIG_TIME_LOW_RES
1014 tim = ktime_add_safe(tim, base->resolution);
1018 hrtimer_set_expires_range_ns(timer, tim, delta_ns);
1020 timer_stats_hrtimer_set_start_info(timer);
1022 leftmost = enqueue_hrtimer(timer, new_base);
1025 * Only allow reprogramming if the new base is on this CPU.
1026 * (it might still be on another CPU if the timer was pending)
1028 * XXX send_remote_softirq() ?
1030 if (leftmost && new_base->cpu_base == &__get_cpu_var(hrtimer_bases)
1031 && hrtimer_enqueue_reprogram(timer, new_base)) {
1034 * We need to drop cpu_base->lock to avoid a
1035 * lock ordering issue vs. rq->lock.
1037 raw_spin_unlock(&new_base->cpu_base->lock);
1038 raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1039 local_irq_restore(flags);
1042 __raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1046 unlock_hrtimer_base(timer, &flags);
1052 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
1053 * @timer: the timer to be added
1055 * @delta_ns: "slack" range for the timer
1056 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
1057 * relative (HRTIMER_MODE_REL)
1061 * 1 when the timer was active
1063 int hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1064 unsigned long delta_ns, const enum hrtimer_mode mode)
1066 return __hrtimer_start_range_ns(timer, tim, delta_ns, mode, 1);
1068 EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
1071 * hrtimer_start - (re)start an hrtimer on the current CPU
1072 * @timer: the timer to be added
1074 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
1075 * relative (HRTIMER_MODE_REL)
1079 * 1 when the timer was active
1082 hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode)
1084 return __hrtimer_start_range_ns(timer, tim, 0, mode, 1);
1086 EXPORT_SYMBOL_GPL(hrtimer_start);
1090 * hrtimer_try_to_cancel - try to deactivate a timer
1091 * @timer: hrtimer to stop
1094 * 0 when the timer was not active
1095 * 1 when the timer was active
1096 * -1 when the timer is currently excuting the callback function and
1099 int hrtimer_try_to_cancel(struct hrtimer *timer)
1101 struct hrtimer_clock_base *base;
1102 unsigned long flags;
1105 base = lock_hrtimer_base(timer, &flags);
1107 if (!hrtimer_callback_running(timer))
1108 ret = remove_hrtimer(timer, base);
1110 unlock_hrtimer_base(timer, &flags);
1115 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1118 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1119 * @timer: the timer to be cancelled
1122 * 0 when the timer was not active
1123 * 1 when the timer was active
1125 int hrtimer_cancel(struct hrtimer *timer)
1128 int ret = hrtimer_try_to_cancel(timer);
1135 EXPORT_SYMBOL_GPL(hrtimer_cancel);
1138 * hrtimer_get_remaining - get remaining time for the timer
1139 * @timer: the timer to read
1141 ktime_t hrtimer_get_remaining(const struct hrtimer *timer)
1143 unsigned long flags;
1146 lock_hrtimer_base(timer, &flags);
1147 rem = hrtimer_expires_remaining(timer);
1148 unlock_hrtimer_base(timer, &flags);
1152 EXPORT_SYMBOL_GPL(hrtimer_get_remaining);
1154 #ifdef CONFIG_NO_HZ_COMMON
1156 * hrtimer_get_next_event - get the time until next expiry event
1158 * Returns the delta to the next expiry event or KTIME_MAX if no timer
1161 ktime_t hrtimer_get_next_event(void)
1163 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1164 struct hrtimer_clock_base *base = cpu_base->clock_base;
1165 ktime_t delta, mindelta = { .tv64 = KTIME_MAX };
1166 unsigned long flags;
1169 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1171 if (!hrtimer_hres_active()) {
1172 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
1173 struct hrtimer *timer;
1174 struct timerqueue_node *next;
1176 next = timerqueue_getnext(&base->active);
1180 timer = container_of(next, struct hrtimer, node);
1181 delta.tv64 = hrtimer_get_expires_tv64(timer);
1182 delta = ktime_sub(delta, base->get_time());
1183 if (delta.tv64 < mindelta.tv64)
1184 mindelta.tv64 = delta.tv64;
1188 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1190 if (mindelta.tv64 < 0)
1196 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1197 enum hrtimer_mode mode)
1199 struct hrtimer_cpu_base *cpu_base;
1202 memset(timer, 0, sizeof(struct hrtimer));
1204 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1206 if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
1207 clock_id = CLOCK_MONOTONIC;
1209 base = hrtimer_clockid_to_base(clock_id);
1210 timer->base = &cpu_base->clock_base[base];
1211 timerqueue_init(&timer->node);
1213 #ifdef CONFIG_TIMER_STATS
1214 timer->start_site = NULL;
1215 timer->start_pid = -1;
1216 memset(timer->start_comm, 0, TASK_COMM_LEN);
1221 * hrtimer_init - initialize a timer to the given clock
1222 * @timer: the timer to be initialized
1223 * @clock_id: the clock to be used
1224 * @mode: timer mode abs/rel
1226 void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1227 enum hrtimer_mode mode)
1229 debug_init(timer, clock_id, mode);
1230 __hrtimer_init(timer, clock_id, mode);
1232 EXPORT_SYMBOL_GPL(hrtimer_init);
1235 * hrtimer_get_res - get the timer resolution for a clock
1236 * @which_clock: which clock to query
1237 * @tp: pointer to timespec variable to store the resolution
1239 * Store the resolution of the clock selected by @which_clock in the
1240 * variable pointed to by @tp.
1242 int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp)
1244 struct hrtimer_cpu_base *cpu_base;
1245 int base = hrtimer_clockid_to_base(which_clock);
1247 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1248 *tp = ktime_to_timespec(cpu_base->clock_base[base].resolution);
1252 EXPORT_SYMBOL_GPL(hrtimer_get_res);
1254 static void __run_hrtimer(struct hrtimer *timer, ktime_t *now)
1256 struct hrtimer_clock_base *base = timer->base;
1257 struct hrtimer_cpu_base *cpu_base = base->cpu_base;
1258 enum hrtimer_restart (*fn)(struct hrtimer *);
1261 WARN_ON(!irqs_disabled());
1263 debug_deactivate(timer);
1264 __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0);
1265 timer_stats_account_hrtimer(timer);
1266 fn = timer->function;
1269 * Because we run timers from hardirq context, there is no chance
1270 * they get migrated to another cpu, therefore its safe to unlock
1273 raw_spin_unlock(&cpu_base->lock);
1274 trace_hrtimer_expire_entry(timer, now);
1275 restart = fn(timer);
1276 trace_hrtimer_expire_exit(timer);
1277 raw_spin_lock(&cpu_base->lock);
1280 * Note: We clear the CALLBACK bit after enqueue_hrtimer and
1281 * we do not reprogramm the event hardware. Happens either in
1282 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1284 if (restart != HRTIMER_NORESTART) {
1285 BUG_ON(timer->state != HRTIMER_STATE_CALLBACK);
1286 enqueue_hrtimer(timer, base);
1289 WARN_ON_ONCE(!(timer->state & HRTIMER_STATE_CALLBACK));
1291 timer->state &= ~HRTIMER_STATE_CALLBACK;
1294 #ifdef CONFIG_HIGH_RES_TIMERS
1297 * High resolution timer interrupt
1298 * Called with interrupts disabled
1300 void hrtimer_interrupt(struct clock_event_device *dev)
1302 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1303 ktime_t expires_next, now, entry_time, delta;
1306 BUG_ON(!cpu_base->hres_active);
1307 cpu_base->nr_events++;
1308 dev->next_event.tv64 = KTIME_MAX;
1310 raw_spin_lock(&cpu_base->lock);
1311 entry_time = now = hrtimer_update_base(cpu_base);
1313 expires_next.tv64 = KTIME_MAX;
1315 * We set expires_next to KTIME_MAX here with cpu_base->lock
1316 * held to prevent that a timer is enqueued in our queue via
1317 * the migration code. This does not affect enqueueing of
1318 * timers which run their callback and need to be requeued on
1321 cpu_base->expires_next.tv64 = KTIME_MAX;
1323 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1324 struct hrtimer_clock_base *base;
1325 struct timerqueue_node *node;
1328 if (!(cpu_base->active_bases & (1 << i)))
1331 base = cpu_base->clock_base + i;
1332 basenow = ktime_add(now, base->offset);
1334 while ((node = timerqueue_getnext(&base->active))) {
1335 struct hrtimer *timer;
1337 timer = container_of(node, struct hrtimer, node);
1340 * The immediate goal for using the softexpires is
1341 * minimizing wakeups, not running timers at the
1342 * earliest interrupt after their soft expiration.
1343 * This allows us to avoid using a Priority Search
1344 * Tree, which can answer a stabbing querry for
1345 * overlapping intervals and instead use the simple
1346 * BST we already have.
1347 * We don't add extra wakeups by delaying timers that
1348 * are right-of a not yet expired timer, because that
1349 * timer will have to trigger a wakeup anyway.
1352 if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer)) {
1355 expires = ktime_sub(hrtimer_get_expires(timer),
1357 if (expires.tv64 < 0)
1358 expires.tv64 = KTIME_MAX;
1359 if (expires.tv64 < expires_next.tv64)
1360 expires_next = expires;
1364 __run_hrtimer(timer, &basenow);
1369 * Store the new expiry value so the migration code can verify
1372 cpu_base->expires_next = expires_next;
1373 raw_spin_unlock(&cpu_base->lock);
1375 /* Reprogramming necessary ? */
1376 if (expires_next.tv64 == KTIME_MAX ||
1377 !tick_program_event(expires_next, 0)) {
1378 cpu_base->hang_detected = 0;
1383 * The next timer was already expired due to:
1385 * - long lasting callbacks
1386 * - being scheduled away when running in a VM
1388 * We need to prevent that we loop forever in the hrtimer
1389 * interrupt routine. We give it 3 attempts to avoid
1390 * overreacting on some spurious event.
1392 * Acquire base lock for updating the offsets and retrieving
1395 raw_spin_lock(&cpu_base->lock);
1396 now = hrtimer_update_base(cpu_base);
1397 cpu_base->nr_retries++;
1401 * Give the system a chance to do something else than looping
1402 * here. We stored the entry time, so we know exactly how long
1403 * we spent here. We schedule the next event this amount of
1406 cpu_base->nr_hangs++;
1407 cpu_base->hang_detected = 1;
1408 raw_spin_unlock(&cpu_base->lock);
1409 delta = ktime_sub(now, entry_time);
1410 if (delta.tv64 > cpu_base->max_hang_time.tv64)
1411 cpu_base->max_hang_time = delta;
1413 * Limit it to a sensible value as we enforce a longer
1414 * delay. Give the CPU at least 100ms to catch up.
1416 if (delta.tv64 > 100 * NSEC_PER_MSEC)
1417 expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1419 expires_next = ktime_add(now, delta);
1420 tick_program_event(expires_next, 1);
1421 printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n",
1422 ktime_to_ns(delta));
1426 * local version of hrtimer_peek_ahead_timers() called with interrupts
1429 static void __hrtimer_peek_ahead_timers(void)
1431 struct tick_device *td;
1433 if (!hrtimer_hres_active())
1436 td = &__get_cpu_var(tick_cpu_device);
1437 if (td && td->evtdev)
1438 hrtimer_interrupt(td->evtdev);
1442 * hrtimer_peek_ahead_timers -- run soft-expired timers now
1444 * hrtimer_peek_ahead_timers will peek at the timer queue of
1445 * the current cpu and check if there are any timers for which
1446 * the soft expires time has passed. If any such timers exist,
1447 * they are run immediately and then removed from the timer queue.
1450 void hrtimer_peek_ahead_timers(void)
1452 unsigned long flags;
1454 local_irq_save(flags);
1455 __hrtimer_peek_ahead_timers();
1456 local_irq_restore(flags);
1459 static void run_hrtimer_softirq(struct softirq_action *h)
1461 hrtimer_peek_ahead_timers();
1464 #else /* CONFIG_HIGH_RES_TIMERS */
1466 static inline void __hrtimer_peek_ahead_timers(void) { }
1468 #endif /* !CONFIG_HIGH_RES_TIMERS */
1471 * Called from timer softirq every jiffy, expire hrtimers:
1473 * For HRT its the fall back code to run the softirq in the timer
1474 * softirq context in case the hrtimer initialization failed or has
1475 * not been done yet.
1477 void hrtimer_run_pending(void)
1479 if (hrtimer_hres_active())
1483 * This _is_ ugly: We have to check in the softirq context,
1484 * whether we can switch to highres and / or nohz mode. The
1485 * clocksource switch happens in the timer interrupt with
1486 * xtime_lock held. Notification from there only sets the
1487 * check bit in the tick_oneshot code, otherwise we might
1488 * deadlock vs. xtime_lock.
1490 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1491 hrtimer_switch_to_hres();
1495 * Called from hardirq context every jiffy
1497 void hrtimer_run_queues(void)
1499 struct timerqueue_node *node;
1500 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1501 struct hrtimer_clock_base *base;
1502 int index, gettime = 1;
1504 if (hrtimer_hres_active())
1507 for (index = 0; index < HRTIMER_MAX_CLOCK_BASES; index++) {
1508 base = &cpu_base->clock_base[index];
1509 if (!timerqueue_getnext(&base->active))
1513 hrtimer_get_softirq_time(cpu_base);
1517 raw_spin_lock(&cpu_base->lock);
1519 while ((node = timerqueue_getnext(&base->active))) {
1520 struct hrtimer *timer;
1522 timer = container_of(node, struct hrtimer, node);
1523 if (base->softirq_time.tv64 <=
1524 hrtimer_get_expires_tv64(timer))
1527 __run_hrtimer(timer, &base->softirq_time);
1529 raw_spin_unlock(&cpu_base->lock);
1534 * Sleep related functions:
1536 static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1538 struct hrtimer_sleeper *t =
1539 container_of(timer, struct hrtimer_sleeper, timer);
1540 struct task_struct *task = t->task;
1544 wake_up_process(task);
1546 return HRTIMER_NORESTART;
1549 void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1551 sl->timer.function = hrtimer_wakeup;
1554 EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
1556 static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1558 hrtimer_init_sleeper(t, current);
1561 set_current_state(TASK_INTERRUPTIBLE);
1562 hrtimer_start_expires(&t->timer, mode);
1563 if (!hrtimer_active(&t->timer))
1566 if (likely(t->task))
1567 freezable_schedule();
1569 hrtimer_cancel(&t->timer);
1570 mode = HRTIMER_MODE_ABS;
1572 } while (t->task && !signal_pending(current));
1574 __set_current_state(TASK_RUNNING);
1576 return t->task == NULL;
1579 static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp)
1581 struct timespec rmt;
1584 rem = hrtimer_expires_remaining(timer);
1587 rmt = ktime_to_timespec(rem);
1589 if (copy_to_user(rmtp, &rmt, sizeof(*rmtp)))
1595 long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1597 struct hrtimer_sleeper t;
1598 struct timespec __user *rmtp;
1601 hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid,
1603 hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
1605 if (do_nanosleep(&t, HRTIMER_MODE_ABS))
1608 rmtp = restart->nanosleep.rmtp;
1610 ret = update_rmtp(&t.timer, rmtp);
1615 /* The other values in restart are already filled in */
1616 ret = -ERESTART_RESTARTBLOCK;
1618 destroy_hrtimer_on_stack(&t.timer);
1622 long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
1623 const enum hrtimer_mode mode, const clockid_t clockid)
1625 struct restart_block *restart;
1626 struct hrtimer_sleeper t;
1628 unsigned long slack;
1630 slack = current->timer_slack_ns;
1631 if (dl_task(current) || rt_task(current))
1634 hrtimer_init_on_stack(&t.timer, clockid, mode);
1635 hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack);
1636 if (do_nanosleep(&t, mode))
1639 /* Absolute timers do not update the rmtp value and restart: */
1640 if (mode == HRTIMER_MODE_ABS) {
1641 ret = -ERESTARTNOHAND;
1646 ret = update_rmtp(&t.timer, rmtp);
1651 restart = ¤t_thread_info()->restart_block;
1652 restart->fn = hrtimer_nanosleep_restart;
1653 restart->nanosleep.clockid = t.timer.base->clockid;
1654 restart->nanosleep.rmtp = rmtp;
1655 restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
1657 ret = -ERESTART_RESTARTBLOCK;
1659 destroy_hrtimer_on_stack(&t.timer);
1663 SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp,
1664 struct timespec __user *, rmtp)
1668 if (copy_from_user(&tu, rqtp, sizeof(tu)))
1671 if (!timespec_valid(&tu))
1674 return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1678 * Functions related to boot-time initialization:
1680 static void init_hrtimers_cpu(int cpu)
1682 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1685 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1686 cpu_base->clock_base[i].cpu_base = cpu_base;
1687 timerqueue_init_head(&cpu_base->clock_base[i].active);
1690 hrtimer_init_hres(cpu_base);
1693 #ifdef CONFIG_HOTPLUG_CPU
1695 static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1696 struct hrtimer_clock_base *new_base)
1698 struct hrtimer *timer;
1699 struct timerqueue_node *node;
1701 while ((node = timerqueue_getnext(&old_base->active))) {
1702 timer = container_of(node, struct hrtimer, node);
1703 BUG_ON(hrtimer_callback_running(timer));
1704 debug_deactivate(timer);
1707 * Mark it as STATE_MIGRATE not INACTIVE otherwise the
1708 * timer could be seen as !active and just vanish away
1709 * under us on another CPU
1711 __remove_hrtimer(timer, old_base, HRTIMER_STATE_MIGRATE, 0);
1712 timer->base = new_base;
1714 * Enqueue the timers on the new cpu. This does not
1715 * reprogram the event device in case the timer
1716 * expires before the earliest on this CPU, but we run
1717 * hrtimer_interrupt after we migrated everything to
1718 * sort out already expired timers and reprogram the
1721 enqueue_hrtimer(timer, new_base);
1723 /* Clear the migration state bit */
1724 timer->state &= ~HRTIMER_STATE_MIGRATE;
1728 static void migrate_hrtimers(int scpu)
1730 struct hrtimer_cpu_base *old_base, *new_base;
1733 BUG_ON(cpu_online(scpu));
1734 tick_cancel_sched_timer(scpu);
1736 local_irq_disable();
1737 old_base = &per_cpu(hrtimer_bases, scpu);
1738 new_base = &__get_cpu_var(hrtimer_bases);
1740 * The caller is globally serialized and nobody else
1741 * takes two locks at once, deadlock is not possible.
1743 raw_spin_lock(&new_base->lock);
1744 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1746 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1747 migrate_hrtimer_list(&old_base->clock_base[i],
1748 &new_base->clock_base[i]);
1751 raw_spin_unlock(&old_base->lock);
1752 raw_spin_unlock(&new_base->lock);
1754 /* Check, if we got expired work to do */
1755 __hrtimer_peek_ahead_timers();
1759 #endif /* CONFIG_HOTPLUG_CPU */
1761 static int hrtimer_cpu_notify(struct notifier_block *self,
1762 unsigned long action, void *hcpu)
1764 int scpu = (long)hcpu;
1768 case CPU_UP_PREPARE:
1769 case CPU_UP_PREPARE_FROZEN:
1770 init_hrtimers_cpu(scpu);
1773 #ifdef CONFIG_HOTPLUG_CPU
1775 case CPU_DYING_FROZEN:
1776 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DYING, &scpu);
1779 case CPU_DEAD_FROZEN:
1781 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &scpu);
1782 migrate_hrtimers(scpu);
1794 static struct notifier_block hrtimers_nb = {
1795 .notifier_call = hrtimer_cpu_notify,
1798 void __init hrtimers_init(void)
1800 hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
1801 (void *)(long)smp_processor_id());
1802 register_cpu_notifier(&hrtimers_nb);
1803 #ifdef CONFIG_HIGH_RES_TIMERS
1804 open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq);
1809 * schedule_hrtimeout_range_clock - sleep until timeout
1810 * @expires: timeout value (ktime_t)
1811 * @delta: slack in expires timeout (ktime_t)
1812 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1813 * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME
1816 schedule_hrtimeout_range_clock(ktime_t *expires, unsigned long delta,
1817 const enum hrtimer_mode mode, int clock)
1819 struct hrtimer_sleeper t;
1822 * Optimize when a zero timeout value is given. It does not
1823 * matter whether this is an absolute or a relative time.
1825 if (expires && !expires->tv64) {
1826 __set_current_state(TASK_RUNNING);
1831 * A NULL parameter means "infinite"
1835 __set_current_state(TASK_RUNNING);
1839 hrtimer_init_on_stack(&t.timer, clock, mode);
1840 hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
1842 hrtimer_init_sleeper(&t, current);
1844 hrtimer_start_expires(&t.timer, mode);
1845 if (!hrtimer_active(&t.timer))
1851 hrtimer_cancel(&t.timer);
1852 destroy_hrtimer_on_stack(&t.timer);
1854 __set_current_state(TASK_RUNNING);
1856 return !t.task ? 0 : -EINTR;
1860 * schedule_hrtimeout_range - sleep until timeout
1861 * @expires: timeout value (ktime_t)
1862 * @delta: slack in expires timeout (ktime_t)
1863 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1865 * Make the current task sleep until the given expiry time has
1866 * elapsed. The routine will return immediately unless
1867 * the current task state has been set (see set_current_state()).
1869 * The @delta argument gives the kernel the freedom to schedule the
1870 * actual wakeup to a time that is both power and performance friendly.
1871 * The kernel give the normal best effort behavior for "@expires+@delta",
1872 * but may decide to fire the timer earlier, but no earlier than @expires.
1874 * You can set the task state as follows -
1876 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1877 * pass before the routine returns.
1879 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1880 * delivered to the current task.
1882 * The current task state is guaranteed to be TASK_RUNNING when this
1885 * Returns 0 when the timer has expired otherwise -EINTR
1887 int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta,
1888 const enum hrtimer_mode mode)
1890 return schedule_hrtimeout_range_clock(expires, delta, mode,
1893 EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
1896 * schedule_hrtimeout - sleep until timeout
1897 * @expires: timeout value (ktime_t)
1898 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1900 * Make the current task sleep until the given expiry time has
1901 * elapsed. The routine will return immediately unless
1902 * the current task state has been set (see set_current_state()).
1904 * You can set the task state as follows -
1906 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1907 * pass before the routine returns.
1909 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1910 * delivered to the current task.
1912 * The current task state is guaranteed to be TASK_RUNNING when this
1915 * Returns 0 when the timer has expired otherwise -EINTR
1917 int __sched schedule_hrtimeout(ktime_t *expires,
1918 const enum hrtimer_mode mode)
1920 return schedule_hrtimeout_range(expires, 0, mode);
1922 EXPORT_SYMBOL_GPL(schedule_hrtimeout);