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>
57 #include "tick-internal.h"
62 * There are more clockids then hrtimer bases. Thus, we index
63 * into the timer bases by the hrtimer_base_type enum. When trying
64 * to reach a base using a clockid, hrtimer_clockid_to_base()
65 * is used to convert from clockid to the proper hrtimer_base_type.
67 DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
69 .lock = __RAW_SPIN_LOCK_UNLOCKED(hrtimer_bases.lock),
73 .index = HRTIMER_BASE_MONOTONIC,
74 .clockid = CLOCK_MONOTONIC,
75 .get_time = &ktime_get,
78 .index = HRTIMER_BASE_REALTIME,
79 .clockid = CLOCK_REALTIME,
80 .get_time = &ktime_get_real,
83 .index = HRTIMER_BASE_BOOTTIME,
84 .clockid = CLOCK_BOOTTIME,
85 .get_time = &ktime_get_boottime,
88 .index = HRTIMER_BASE_TAI,
90 .get_time = &ktime_get_clocktai,
95 static const int hrtimer_clock_to_base_table[MAX_CLOCKS] = {
96 [CLOCK_REALTIME] = HRTIMER_BASE_REALTIME,
97 [CLOCK_MONOTONIC] = HRTIMER_BASE_MONOTONIC,
98 [CLOCK_BOOTTIME] = HRTIMER_BASE_BOOTTIME,
99 [CLOCK_TAI] = HRTIMER_BASE_TAI,
102 static inline int hrtimer_clockid_to_base(clockid_t clock_id)
104 return hrtimer_clock_to_base_table[clock_id];
108 * Functions and macros which are different for UP/SMP systems are kept in a
114 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
115 * means that all timers which are tied to this base via timer->base are
116 * locked, and the base itself is locked too.
118 * So __run_timers/migrate_timers can safely modify all timers which could
119 * be found on the lists/queues.
121 * When the timer's base is locked, and the timer removed from list, it is
122 * possible to set timer->base = NULL and drop the lock: the timer remains
126 struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
127 unsigned long *flags)
129 struct hrtimer_clock_base *base;
133 if (likely(base != NULL)) {
134 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
135 if (likely(base == timer->base))
137 /* The timer has migrated to another CPU: */
138 raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
145 * With HIGHRES=y we do not migrate the timer when it is expiring
146 * before the next event on the target cpu because we cannot reprogram
147 * the target cpu hardware and we would cause it to fire late.
149 * Called with cpu_base->lock of target cpu held.
152 hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base)
154 #ifdef CONFIG_HIGH_RES_TIMERS
157 if (!new_base->cpu_base->hres_active)
160 expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset);
161 return expires.tv64 <= new_base->cpu_base->expires_next.tv64;
168 * Switch the timer base to the current CPU when possible.
170 static inline struct hrtimer_clock_base *
171 switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
174 struct hrtimer_clock_base *new_base;
175 struct hrtimer_cpu_base *new_cpu_base;
176 int this_cpu = smp_processor_id();
177 int cpu = get_nohz_timer_target(pinned);
178 int basenum = base->index;
181 new_cpu_base = &per_cpu(hrtimer_bases, cpu);
182 new_base = &new_cpu_base->clock_base[basenum];
184 if (base != new_base) {
186 * We are trying to move timer to new_base.
187 * However we can't change timer's base while it is running,
188 * so we keep it on the same CPU. No hassle vs. reprogramming
189 * the event source in the high resolution case. The softirq
190 * code will take care of this when the timer function has
191 * completed. There is no conflict as we hold the lock until
192 * the timer is enqueued.
194 if (unlikely(hrtimer_callback_running(timer)))
197 /* See the comment in lock_timer_base() */
199 raw_spin_unlock(&base->cpu_base->lock);
200 raw_spin_lock(&new_base->cpu_base->lock);
202 if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) {
204 raw_spin_unlock(&new_base->cpu_base->lock);
205 raw_spin_lock(&base->cpu_base->lock);
209 timer->base = new_base;
211 if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) {
219 #else /* CONFIG_SMP */
221 static inline struct hrtimer_clock_base *
222 lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
224 struct hrtimer_clock_base *base = timer->base;
226 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
231 # define switch_hrtimer_base(t, b, p) (b)
233 #endif /* !CONFIG_SMP */
236 * Functions for the union type storage format of ktime_t which are
237 * too large for inlining:
239 #if BITS_PER_LONG < 64
241 * Divide a ktime value by a nanosecond value
243 u64 __ktime_divns(const ktime_t kt, s64 div)
248 dclc = ktime_to_ns(kt);
249 /* Make sure the divisor is less than 2^32: */
255 do_div(dclc, (unsigned long) div);
259 EXPORT_SYMBOL_GPL(__ktime_divns);
260 #endif /* BITS_PER_LONG >= 64 */
263 * Add two ktime values and do a safety check for overflow:
265 ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
267 ktime_t res = ktime_add(lhs, rhs);
270 * We use KTIME_SEC_MAX here, the maximum timeout which we can
271 * return to user space in a timespec:
273 if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64)
274 res = ktime_set(KTIME_SEC_MAX, 0);
279 EXPORT_SYMBOL_GPL(ktime_add_safe);
281 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
283 static struct debug_obj_descr hrtimer_debug_descr;
285 static void *hrtimer_debug_hint(void *addr)
287 return ((struct hrtimer *) addr)->function;
291 * fixup_init is called when:
292 * - an active object is initialized
294 static int hrtimer_fixup_init(void *addr, enum debug_obj_state state)
296 struct hrtimer *timer = addr;
299 case ODEBUG_STATE_ACTIVE:
300 hrtimer_cancel(timer);
301 debug_object_init(timer, &hrtimer_debug_descr);
309 * fixup_activate is called when:
310 * - an active object is activated
311 * - an unknown object is activated (might be a statically initialized object)
313 static int hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
317 case ODEBUG_STATE_NOTAVAILABLE:
321 case ODEBUG_STATE_ACTIVE:
330 * fixup_free is called when:
331 * - an active object is freed
333 static int hrtimer_fixup_free(void *addr, enum debug_obj_state state)
335 struct hrtimer *timer = addr;
338 case ODEBUG_STATE_ACTIVE:
339 hrtimer_cancel(timer);
340 debug_object_free(timer, &hrtimer_debug_descr);
347 static struct debug_obj_descr hrtimer_debug_descr = {
349 .debug_hint = hrtimer_debug_hint,
350 .fixup_init = hrtimer_fixup_init,
351 .fixup_activate = hrtimer_fixup_activate,
352 .fixup_free = hrtimer_fixup_free,
355 static inline void debug_hrtimer_init(struct hrtimer *timer)
357 debug_object_init(timer, &hrtimer_debug_descr);
360 static inline void debug_hrtimer_activate(struct hrtimer *timer)
362 debug_object_activate(timer, &hrtimer_debug_descr);
365 static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
367 debug_object_deactivate(timer, &hrtimer_debug_descr);
370 static inline void debug_hrtimer_free(struct hrtimer *timer)
372 debug_object_free(timer, &hrtimer_debug_descr);
375 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
376 enum hrtimer_mode mode);
378 void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
379 enum hrtimer_mode mode)
381 debug_object_init_on_stack(timer, &hrtimer_debug_descr);
382 __hrtimer_init(timer, clock_id, mode);
384 EXPORT_SYMBOL_GPL(hrtimer_init_on_stack);
386 void destroy_hrtimer_on_stack(struct hrtimer *timer)
388 debug_object_free(timer, &hrtimer_debug_descr);
392 static inline void debug_hrtimer_init(struct hrtimer *timer) { }
393 static inline void debug_hrtimer_activate(struct hrtimer *timer) { }
394 static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
398 debug_init(struct hrtimer *timer, clockid_t clockid,
399 enum hrtimer_mode mode)
401 debug_hrtimer_init(timer);
402 trace_hrtimer_init(timer, clockid, mode);
405 static inline void debug_activate(struct hrtimer *timer)
407 debug_hrtimer_activate(timer);
408 trace_hrtimer_start(timer);
411 static inline void debug_deactivate(struct hrtimer *timer)
413 debug_hrtimer_deactivate(timer);
414 trace_hrtimer_cancel(timer);
417 #if defined(CONFIG_NO_HZ_COMMON) || defined(CONFIG_HIGH_RES_TIMERS)
418 static inline void hrtimer_update_next_timer(struct hrtimer_cpu_base *cpu_base,
419 struct hrtimer *timer)
421 #ifdef CONFIG_HIGH_RES_TIMERS
422 cpu_base->next_timer = timer;
426 static ktime_t __hrtimer_get_next_event(struct hrtimer_cpu_base *cpu_base)
428 struct hrtimer_clock_base *base = cpu_base->clock_base;
429 ktime_t expires, expires_next = { .tv64 = KTIME_MAX };
430 unsigned int active = cpu_base->active_bases;
432 hrtimer_update_next_timer(cpu_base, NULL);
433 for (; active; base++, active >>= 1) {
434 struct timerqueue_node *next;
435 struct hrtimer *timer;
437 if (!(active & 0x01))
440 next = timerqueue_getnext(&base->active);
441 timer = container_of(next, struct hrtimer, node);
442 expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
443 if (expires.tv64 < expires_next.tv64) {
444 expires_next = expires;
445 hrtimer_update_next_timer(cpu_base, timer);
449 * clock_was_set() might have changed base->offset of any of
450 * the clock bases so the result might be negative. Fix it up
451 * to prevent a false positive in clockevents_program_event().
453 if (expires_next.tv64 < 0)
454 expires_next.tv64 = 0;
459 static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base)
461 ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset;
462 ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset;
463 ktime_t *offs_tai = &base->clock_base[HRTIMER_BASE_TAI].offset;
465 return ktime_get_update_offsets_now(&base->clock_was_set_seq,
466 offs_real, offs_boot, offs_tai);
469 /* High resolution timer related functions */
470 #ifdef CONFIG_HIGH_RES_TIMERS
473 * High resolution timer enabled ?
475 static int hrtimer_hres_enabled __read_mostly = 1;
476 unsigned int hrtimer_resolution __read_mostly = LOW_RES_NSEC;
477 EXPORT_SYMBOL_GPL(hrtimer_resolution);
480 * Enable / Disable high resolution mode
482 static int __init setup_hrtimer_hres(char *str)
484 if (!strcmp(str, "off"))
485 hrtimer_hres_enabled = 0;
486 else if (!strcmp(str, "on"))
487 hrtimer_hres_enabled = 1;
493 __setup("highres=", setup_hrtimer_hres);
496 * hrtimer_high_res_enabled - query, if the highres mode is enabled
498 static inline int hrtimer_is_hres_enabled(void)
500 return hrtimer_hres_enabled;
504 * Is the high resolution mode active ?
506 static inline int __hrtimer_hres_active(struct hrtimer_cpu_base *cpu_base)
508 return cpu_base->hres_active;
511 static inline int hrtimer_hres_active(void)
513 return __hrtimer_hres_active(this_cpu_ptr(&hrtimer_bases));
517 * Reprogram the event source with checking both queues for the
519 * Called with interrupts disabled and base->lock held
522 hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
524 ktime_t expires_next;
526 if (!cpu_base->hres_active)
529 expires_next = __hrtimer_get_next_event(cpu_base);
531 if (skip_equal && expires_next.tv64 == cpu_base->expires_next.tv64)
534 cpu_base->expires_next.tv64 = expires_next.tv64;
537 * If a hang was detected in the last timer interrupt then we
538 * leave the hang delay active in the hardware. We want the
539 * system to make progress. That also prevents the following
541 * T1 expires 50ms from now
542 * T2 expires 5s from now
544 * T1 is removed, so this code is called and would reprogram
545 * the hardware to 5s from now. Any hrtimer_start after that
546 * will not reprogram the hardware due to hang_detected being
547 * set. So we'd effectivly block all timers until the T2 event
550 if (cpu_base->hang_detected)
553 if (cpu_base->expires_next.tv64 != KTIME_MAX)
554 tick_program_event(cpu_base->expires_next, 1);
558 * When a timer is enqueued and expires earlier than the already enqueued
559 * timers, we have to check, whether it expires earlier than the timer for
560 * which the clock event device was armed.
562 * Called with interrupts disabled and base->cpu_base.lock held
564 static void hrtimer_reprogram(struct hrtimer *timer,
565 struct hrtimer_clock_base *base)
567 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
568 ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
570 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
573 * If the timer is not on the current cpu, we cannot reprogram
574 * the other cpus clock event device.
576 if (base->cpu_base != cpu_base)
580 * If the hrtimer interrupt is running, then it will
581 * reevaluate the clock bases and reprogram the clock event
582 * device. The callbacks are always executed in hard interrupt
583 * context so we don't need an extra check for a running
586 if (cpu_base->in_hrtirq)
590 * CLOCK_REALTIME timer might be requested with an absolute
591 * expiry time which is less than base->offset. Set it to 0.
593 if (expires.tv64 < 0)
596 if (expires.tv64 >= cpu_base->expires_next.tv64)
599 /* Update the pointer to the next expiring timer */
600 cpu_base->next_timer = timer;
603 * If a hang was detected in the last timer interrupt then we
604 * do not schedule a timer which is earlier than the expiry
605 * which we enforced in the hang detection. We want the system
608 if (cpu_base->hang_detected)
612 * Program the timer hardware. We enforce the expiry for
613 * events which are already in the past.
615 cpu_base->expires_next = expires;
616 tick_program_event(expires, 1);
620 * Initialize the high resolution related parts of cpu_base
622 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
624 base->expires_next.tv64 = KTIME_MAX;
625 base->hres_active = 0;
629 * Retrigger next event is called after clock was set
631 * Called with interrupts disabled via on_each_cpu()
633 static void retrigger_next_event(void *arg)
635 struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
637 if (!base->hres_active)
640 raw_spin_lock(&base->lock);
641 hrtimer_update_base(base);
642 hrtimer_force_reprogram(base, 0);
643 raw_spin_unlock(&base->lock);
647 * Switch to high resolution mode
649 static int hrtimer_switch_to_hres(void)
651 struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
653 if (tick_init_highres()) {
654 printk(KERN_WARNING "Could not switch to high resolution "
655 "mode on CPU %d\n", base->cpu);
658 base->hres_active = 1;
659 hrtimer_resolution = HIGH_RES_NSEC;
661 tick_setup_sched_timer();
662 /* "Retrigger" the interrupt to get things going */
663 retrigger_next_event(NULL);
667 static void clock_was_set_work(struct work_struct *work)
672 static DECLARE_WORK(hrtimer_work, clock_was_set_work);
675 * Called from timekeeping and resume code to reprogramm the hrtimer
676 * interrupt device on all cpus.
678 void clock_was_set_delayed(void)
680 schedule_work(&hrtimer_work);
685 static inline int __hrtimer_hres_active(struct hrtimer_cpu_base *b) { return 0; }
686 static inline int hrtimer_hres_active(void) { return 0; }
687 static inline int hrtimer_is_hres_enabled(void) { return 0; }
688 static inline int hrtimer_switch_to_hres(void) { return 0; }
690 hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { }
691 static inline int hrtimer_reprogram(struct hrtimer *timer,
692 struct hrtimer_clock_base *base)
696 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { }
697 static inline void retrigger_next_event(void *arg) { }
699 #endif /* CONFIG_HIGH_RES_TIMERS */
702 * Clock realtime was set
704 * Change the offset of the realtime clock vs. the monotonic
707 * We might have to reprogram the high resolution timer interrupt. On
708 * SMP we call the architecture specific code to retrigger _all_ high
709 * resolution timer interrupts. On UP we just disable interrupts and
710 * call the high resolution interrupt code.
712 void clock_was_set(void)
714 #ifdef CONFIG_HIGH_RES_TIMERS
715 /* Retrigger the CPU local events everywhere */
716 on_each_cpu(retrigger_next_event, NULL, 1);
718 timerfd_clock_was_set();
722 * During resume we might have to reprogram the high resolution timer
723 * interrupt on all online CPUs. However, all other CPUs will be
724 * stopped with IRQs interrupts disabled so the clock_was_set() call
727 void hrtimers_resume(void)
729 WARN_ONCE(!irqs_disabled(),
730 KERN_INFO "hrtimers_resume() called with IRQs enabled!");
732 /* Retrigger on the local CPU */
733 retrigger_next_event(NULL);
734 /* And schedule a retrigger for all others */
735 clock_was_set_delayed();
738 static inline void timer_stats_hrtimer_set_start_info(struct hrtimer *timer)
740 #ifdef CONFIG_TIMER_STATS
741 if (timer->start_site)
743 timer->start_site = __builtin_return_address(0);
744 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
745 timer->start_pid = current->pid;
749 static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer *timer)
751 #ifdef CONFIG_TIMER_STATS
752 timer->start_site = NULL;
756 static inline void timer_stats_account_hrtimer(struct hrtimer *timer)
758 #ifdef CONFIG_TIMER_STATS
759 if (likely(!timer_stats_active))
761 timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
762 timer->function, timer->start_comm, 0);
767 * Counterpart to lock_hrtimer_base above:
770 void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
772 raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
776 * hrtimer_forward - forward the timer expiry
777 * @timer: hrtimer to forward
778 * @now: forward past this time
779 * @interval: the interval to forward
781 * Forward the timer expiry so it will expire in the future.
782 * Returns the number of overruns.
784 * Can be safely called from the callback function of @timer. If
785 * called from other contexts @timer must neither be enqueued nor
786 * running the callback and the caller needs to take care of
789 * Note: This only updates the timer expiry value and does not requeue
792 u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
797 delta = ktime_sub(now, hrtimer_get_expires(timer));
802 if (interval.tv64 < hrtimer_resolution)
803 interval.tv64 = hrtimer_resolution;
805 if (unlikely(delta.tv64 >= interval.tv64)) {
806 s64 incr = ktime_to_ns(interval);
808 orun = ktime_divns(delta, incr);
809 hrtimer_add_expires_ns(timer, incr * orun);
810 if (hrtimer_get_expires_tv64(timer) > now.tv64)
813 * This (and the ktime_add() below) is the
814 * correction for exact:
818 hrtimer_add_expires(timer, interval);
822 EXPORT_SYMBOL_GPL(hrtimer_forward);
825 * enqueue_hrtimer - internal function to (re)start a timer
827 * The timer is inserted in expiry order. Insertion into the
828 * red black tree is O(log(n)). Must hold the base lock.
830 * Returns 1 when the new timer is the leftmost timer in the tree.
832 static int enqueue_hrtimer(struct hrtimer *timer,
833 struct hrtimer_clock_base *base)
835 debug_activate(timer);
837 base->cpu_base->active_bases |= 1 << base->index;
840 * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
841 * state of a possibly running callback.
843 timer->state |= HRTIMER_STATE_ENQUEUED;
845 return timerqueue_add(&base->active, &timer->node);
849 * __remove_hrtimer - internal function to remove a timer
851 * Caller must hold the base lock.
853 * High resolution timer mode reprograms the clock event device when the
854 * timer is the one which expires next. The caller can disable this by setting
855 * reprogram to zero. This is useful, when the context does a reprogramming
856 * anyway (e.g. timer interrupt)
858 static void __remove_hrtimer(struct hrtimer *timer,
859 struct hrtimer_clock_base *base,
860 unsigned long newstate, int reprogram)
862 struct hrtimer_cpu_base *cpu_base = base->cpu_base;
863 unsigned int state = timer->state;
865 timer->state = newstate;
866 if (!(state & HRTIMER_STATE_ENQUEUED))
869 if (!timerqueue_del(&base->active, &timer->node))
870 cpu_base->active_bases &= ~(1 << base->index);
872 #ifdef CONFIG_HIGH_RES_TIMERS
874 * Note: If reprogram is false we do not update
875 * cpu_base->next_timer. This happens when we remove the first
876 * timer on a remote cpu. No harm as we never dereference
877 * cpu_base->next_timer. So the worst thing what can happen is
878 * an superflous call to hrtimer_force_reprogram() on the
879 * remote cpu later on if the same timer gets enqueued again.
881 if (reprogram && timer == cpu_base->next_timer)
882 hrtimer_force_reprogram(cpu_base, 1);
887 * remove hrtimer, called with base lock held
890 remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base)
892 if (hrtimer_is_queued(timer)) {
897 * Remove the timer and force reprogramming when high
898 * resolution mode is active and the timer is on the current
899 * CPU. If we remove a timer on another CPU, reprogramming is
900 * skipped. The interrupt event on this CPU is fired and
901 * reprogramming happens in the interrupt handler. This is a
902 * rare case and less expensive than a smp call.
904 debug_deactivate(timer);
905 timer_stats_hrtimer_clear_start_info(timer);
906 reprogram = base->cpu_base == this_cpu_ptr(&hrtimer_bases);
908 * We must preserve the CALLBACK state flag here,
909 * otherwise we could move the timer base in
910 * switch_hrtimer_base.
912 state = timer->state & HRTIMER_STATE_CALLBACK;
913 __remove_hrtimer(timer, base, state, reprogram);
920 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
921 * @timer: the timer to be added
923 * @delta_ns: "slack" range for the timer
924 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
925 * relative (HRTIMER_MODE_REL)
927 void hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
928 unsigned long delta_ns, const enum hrtimer_mode mode)
930 struct hrtimer_clock_base *base, *new_base;
934 base = lock_hrtimer_base(timer, &flags);
936 /* Remove an active timer from the queue: */
937 remove_hrtimer(timer, base);
939 if (mode & HRTIMER_MODE_REL) {
940 tim = ktime_add_safe(tim, base->get_time());
942 * CONFIG_TIME_LOW_RES is a temporary way for architectures
943 * to signal that they simply return xtime in
944 * do_gettimeoffset(). In this case we want to round up by
945 * resolution when starting a relative timer, to avoid short
946 * timeouts. This will go away with the GTOD framework.
948 #ifdef CONFIG_TIME_LOW_RES
949 tim = ktime_add_safe(tim, ktime_set(0, hrtimer_resolution));
953 hrtimer_set_expires_range_ns(timer, tim, delta_ns);
955 /* Switch the timer base, if necessary: */
956 new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED);
958 timer_stats_hrtimer_set_start_info(timer);
960 leftmost = enqueue_hrtimer(timer, new_base);
964 if (!hrtimer_is_hres_active(timer)) {
966 * Kick to reschedule the next tick to handle the new timer
967 * on dynticks target.
969 wake_up_nohz_cpu(new_base->cpu_base->cpu);
971 hrtimer_reprogram(timer, new_base);
974 unlock_hrtimer_base(timer, &flags);
976 EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
979 * hrtimer_try_to_cancel - try to deactivate a timer
980 * @timer: hrtimer to stop
983 * 0 when the timer was not active
984 * 1 when the timer was active
985 * -1 when the timer is currently excuting the callback function and
988 int hrtimer_try_to_cancel(struct hrtimer *timer)
990 struct hrtimer_clock_base *base;
995 * Check lockless first. If the timer is not active (neither
996 * enqueued nor running the callback, nothing to do here. The
997 * base lock does not serialize against a concurrent enqueue,
998 * so we can avoid taking it.
1000 if (!hrtimer_active(timer))
1003 base = lock_hrtimer_base(timer, &flags);
1005 if (!hrtimer_callback_running(timer))
1006 ret = remove_hrtimer(timer, base);
1008 unlock_hrtimer_base(timer, &flags);
1013 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1016 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1017 * @timer: the timer to be cancelled
1020 * 0 when the timer was not active
1021 * 1 when the timer was active
1023 int hrtimer_cancel(struct hrtimer *timer)
1026 int ret = hrtimer_try_to_cancel(timer);
1033 EXPORT_SYMBOL_GPL(hrtimer_cancel);
1036 * hrtimer_get_remaining - get remaining time for the timer
1037 * @timer: the timer to read
1039 ktime_t hrtimer_get_remaining(const struct hrtimer *timer)
1041 unsigned long flags;
1044 lock_hrtimer_base(timer, &flags);
1045 rem = hrtimer_expires_remaining(timer);
1046 unlock_hrtimer_base(timer, &flags);
1050 EXPORT_SYMBOL_GPL(hrtimer_get_remaining);
1052 #ifdef CONFIG_NO_HZ_COMMON
1054 * hrtimer_get_next_event - get the time until next expiry event
1056 * Returns the next expiry time or KTIME_MAX if no timer is pending.
1058 u64 hrtimer_get_next_event(void)
1060 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1061 u64 expires = KTIME_MAX;
1062 unsigned long flags;
1064 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1066 if (!__hrtimer_hres_active(cpu_base))
1067 expires = __hrtimer_get_next_event(cpu_base).tv64;
1069 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1075 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1076 enum hrtimer_mode mode)
1078 struct hrtimer_cpu_base *cpu_base;
1081 memset(timer, 0, sizeof(struct hrtimer));
1083 cpu_base = raw_cpu_ptr(&hrtimer_bases);
1085 if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
1086 clock_id = CLOCK_MONOTONIC;
1088 base = hrtimer_clockid_to_base(clock_id);
1089 timer->base = &cpu_base->clock_base[base];
1090 timerqueue_init(&timer->node);
1092 #ifdef CONFIG_TIMER_STATS
1093 timer->start_site = NULL;
1094 timer->start_pid = -1;
1095 memset(timer->start_comm, 0, TASK_COMM_LEN);
1100 * hrtimer_init - initialize a timer to the given clock
1101 * @timer: the timer to be initialized
1102 * @clock_id: the clock to be used
1103 * @mode: timer mode abs/rel
1105 void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1106 enum hrtimer_mode mode)
1108 debug_init(timer, clock_id, mode);
1109 __hrtimer_init(timer, clock_id, mode);
1111 EXPORT_SYMBOL_GPL(hrtimer_init);
1113 static void __run_hrtimer(struct hrtimer_cpu_base *cpu_base,
1114 struct hrtimer_clock_base *base,
1115 struct hrtimer *timer, ktime_t *now)
1117 enum hrtimer_restart (*fn)(struct hrtimer *);
1120 WARN_ON(!irqs_disabled());
1122 debug_deactivate(timer);
1123 __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0);
1124 timer_stats_account_hrtimer(timer);
1125 fn = timer->function;
1128 * Because we run timers from hardirq context, there is no chance
1129 * they get migrated to another cpu, therefore its safe to unlock
1132 raw_spin_unlock(&cpu_base->lock);
1133 trace_hrtimer_expire_entry(timer, now);
1134 restart = fn(timer);
1135 trace_hrtimer_expire_exit(timer);
1136 raw_spin_lock(&cpu_base->lock);
1139 * Note: We clear the CALLBACK bit after enqueue_hrtimer and
1140 * we do not reprogramm the event hardware. Happens either in
1141 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1143 if (restart != HRTIMER_NORESTART) {
1144 BUG_ON(timer->state != HRTIMER_STATE_CALLBACK);
1145 enqueue_hrtimer(timer, base);
1148 WARN_ON_ONCE(!(timer->state & HRTIMER_STATE_CALLBACK));
1150 timer->state &= ~HRTIMER_STATE_CALLBACK;
1153 static void __hrtimer_run_queues(struct hrtimer_cpu_base *cpu_base, ktime_t now)
1155 struct hrtimer_clock_base *base = cpu_base->clock_base;
1156 unsigned int active = cpu_base->active_bases;
1158 for (; active; base++, active >>= 1) {
1159 struct timerqueue_node *node;
1162 if (!(active & 0x01))
1165 basenow = ktime_add(now, base->offset);
1167 while ((node = timerqueue_getnext(&base->active))) {
1168 struct hrtimer *timer;
1170 timer = container_of(node, struct hrtimer, node);
1173 * The immediate goal for using the softexpires is
1174 * minimizing wakeups, not running timers at the
1175 * earliest interrupt after their soft expiration.
1176 * This allows us to avoid using a Priority Search
1177 * Tree, which can answer a stabbing querry for
1178 * overlapping intervals and instead use the simple
1179 * BST we already have.
1180 * We don't add extra wakeups by delaying timers that
1181 * are right-of a not yet expired timer, because that
1182 * timer will have to trigger a wakeup anyway.
1184 if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer))
1187 __run_hrtimer(cpu_base, base, timer, &basenow);
1192 #ifdef CONFIG_HIGH_RES_TIMERS
1195 * High resolution timer interrupt
1196 * Called with interrupts disabled
1198 void hrtimer_interrupt(struct clock_event_device *dev)
1200 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1201 ktime_t expires_next, now, entry_time, delta;
1204 BUG_ON(!cpu_base->hres_active);
1205 cpu_base->nr_events++;
1206 dev->next_event.tv64 = KTIME_MAX;
1208 raw_spin_lock(&cpu_base->lock);
1209 entry_time = now = hrtimer_update_base(cpu_base);
1211 cpu_base->in_hrtirq = 1;
1213 * We set expires_next to KTIME_MAX here with cpu_base->lock
1214 * held to prevent that a timer is enqueued in our queue via
1215 * the migration code. This does not affect enqueueing of
1216 * timers which run their callback and need to be requeued on
1219 cpu_base->expires_next.tv64 = KTIME_MAX;
1221 __hrtimer_run_queues(cpu_base, now);
1223 /* Reevaluate the clock bases for the next expiry */
1224 expires_next = __hrtimer_get_next_event(cpu_base);
1226 * Store the new expiry value so the migration code can verify
1229 cpu_base->expires_next = expires_next;
1230 cpu_base->in_hrtirq = 0;
1231 raw_spin_unlock(&cpu_base->lock);
1233 /* Reprogramming necessary ? */
1234 if (expires_next.tv64 == KTIME_MAX ||
1235 !tick_program_event(expires_next, 0)) {
1236 cpu_base->hang_detected = 0;
1241 * The next timer was already expired due to:
1243 * - long lasting callbacks
1244 * - being scheduled away when running in a VM
1246 * We need to prevent that we loop forever in the hrtimer
1247 * interrupt routine. We give it 3 attempts to avoid
1248 * overreacting on some spurious event.
1250 * Acquire base lock for updating the offsets and retrieving
1253 raw_spin_lock(&cpu_base->lock);
1254 now = hrtimer_update_base(cpu_base);
1255 cpu_base->nr_retries++;
1259 * Give the system a chance to do something else than looping
1260 * here. We stored the entry time, so we know exactly how long
1261 * we spent here. We schedule the next event this amount of
1264 cpu_base->nr_hangs++;
1265 cpu_base->hang_detected = 1;
1266 raw_spin_unlock(&cpu_base->lock);
1267 delta = ktime_sub(now, entry_time);
1268 if ((unsigned int)delta.tv64 > cpu_base->max_hang_time)
1269 cpu_base->max_hang_time = (unsigned int) delta.tv64;
1271 * Limit it to a sensible value as we enforce a longer
1272 * delay. Give the CPU at least 100ms to catch up.
1274 if (delta.tv64 > 100 * NSEC_PER_MSEC)
1275 expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1277 expires_next = ktime_add(now, delta);
1278 tick_program_event(expires_next, 1);
1279 printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n",
1280 ktime_to_ns(delta));
1284 * local version of hrtimer_peek_ahead_timers() called with interrupts
1287 static inline void __hrtimer_peek_ahead_timers(void)
1289 struct tick_device *td;
1291 if (!hrtimer_hres_active())
1294 td = this_cpu_ptr(&tick_cpu_device);
1295 if (td && td->evtdev)
1296 hrtimer_interrupt(td->evtdev);
1299 #else /* CONFIG_HIGH_RES_TIMERS */
1301 static inline void __hrtimer_peek_ahead_timers(void) { }
1303 #endif /* !CONFIG_HIGH_RES_TIMERS */
1306 * Called from run_local_timers in hardirq context every jiffy
1308 void hrtimer_run_queues(void)
1310 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1313 if (__hrtimer_hres_active(cpu_base))
1317 * This _is_ ugly: We have to check periodically, whether we
1318 * can switch to highres and / or nohz mode. The clocksource
1319 * switch happens with xtime_lock held. Notification from
1320 * there only sets the check bit in the tick_oneshot code,
1321 * otherwise we might deadlock vs. xtime_lock.
1323 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) {
1324 hrtimer_switch_to_hres();
1328 raw_spin_lock(&cpu_base->lock);
1329 now = hrtimer_update_base(cpu_base);
1330 __hrtimer_run_queues(cpu_base, now);
1331 raw_spin_unlock(&cpu_base->lock);
1335 * Sleep related functions:
1337 static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1339 struct hrtimer_sleeper *t =
1340 container_of(timer, struct hrtimer_sleeper, timer);
1341 struct task_struct *task = t->task;
1345 wake_up_process(task);
1347 return HRTIMER_NORESTART;
1350 void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1352 sl->timer.function = hrtimer_wakeup;
1355 EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
1357 static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1359 hrtimer_init_sleeper(t, current);
1362 set_current_state(TASK_INTERRUPTIBLE);
1363 hrtimer_start_expires(&t->timer, mode);
1365 if (likely(t->task))
1366 freezable_schedule();
1368 hrtimer_cancel(&t->timer);
1369 mode = HRTIMER_MODE_ABS;
1371 } while (t->task && !signal_pending(current));
1373 __set_current_state(TASK_RUNNING);
1375 return t->task == NULL;
1378 static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp)
1380 struct timespec rmt;
1383 rem = hrtimer_expires_remaining(timer);
1386 rmt = ktime_to_timespec(rem);
1388 if (copy_to_user(rmtp, &rmt, sizeof(*rmtp)))
1394 long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1396 struct hrtimer_sleeper t;
1397 struct timespec __user *rmtp;
1400 hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid,
1402 hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
1404 if (do_nanosleep(&t, HRTIMER_MODE_ABS))
1407 rmtp = restart->nanosleep.rmtp;
1409 ret = update_rmtp(&t.timer, rmtp);
1414 /* The other values in restart are already filled in */
1415 ret = -ERESTART_RESTARTBLOCK;
1417 destroy_hrtimer_on_stack(&t.timer);
1421 long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
1422 const enum hrtimer_mode mode, const clockid_t clockid)
1424 struct restart_block *restart;
1425 struct hrtimer_sleeper t;
1427 unsigned long slack;
1429 slack = current->timer_slack_ns;
1430 if (dl_task(current) || rt_task(current))
1433 hrtimer_init_on_stack(&t.timer, clockid, mode);
1434 hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack);
1435 if (do_nanosleep(&t, mode))
1438 /* Absolute timers do not update the rmtp value and restart: */
1439 if (mode == HRTIMER_MODE_ABS) {
1440 ret = -ERESTARTNOHAND;
1445 ret = update_rmtp(&t.timer, rmtp);
1450 restart = ¤t->restart_block;
1451 restart->fn = hrtimer_nanosleep_restart;
1452 restart->nanosleep.clockid = t.timer.base->clockid;
1453 restart->nanosleep.rmtp = rmtp;
1454 restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
1456 ret = -ERESTART_RESTARTBLOCK;
1458 destroy_hrtimer_on_stack(&t.timer);
1462 SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp,
1463 struct timespec __user *, rmtp)
1467 if (copy_from_user(&tu, rqtp, sizeof(tu)))
1470 if (!timespec_valid(&tu))
1473 return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1477 * Functions related to boot-time initialization:
1479 static void init_hrtimers_cpu(int cpu)
1481 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1484 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1485 cpu_base->clock_base[i].cpu_base = cpu_base;
1486 timerqueue_init_head(&cpu_base->clock_base[i].active);
1489 cpu_base->cpu = cpu;
1490 hrtimer_init_hres(cpu_base);
1493 #ifdef CONFIG_HOTPLUG_CPU
1495 static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1496 struct hrtimer_clock_base *new_base)
1498 struct hrtimer *timer;
1499 struct timerqueue_node *node;
1501 while ((node = timerqueue_getnext(&old_base->active))) {
1502 timer = container_of(node, struct hrtimer, node);
1503 BUG_ON(hrtimer_callback_running(timer));
1504 debug_deactivate(timer);
1507 * Mark it as STATE_MIGRATE not INACTIVE otherwise the
1508 * timer could be seen as !active and just vanish away
1509 * under us on another CPU
1511 __remove_hrtimer(timer, old_base, HRTIMER_STATE_MIGRATE, 0);
1512 timer->base = new_base;
1514 * Enqueue the timers on the new cpu. This does not
1515 * reprogram the event device in case the timer
1516 * expires before the earliest on this CPU, but we run
1517 * hrtimer_interrupt after we migrated everything to
1518 * sort out already expired timers and reprogram the
1521 enqueue_hrtimer(timer, new_base);
1523 /* Clear the migration state bit */
1524 timer->state &= ~HRTIMER_STATE_MIGRATE;
1528 static void migrate_hrtimers(int scpu)
1530 struct hrtimer_cpu_base *old_base, *new_base;
1533 BUG_ON(cpu_online(scpu));
1534 tick_cancel_sched_timer(scpu);
1536 local_irq_disable();
1537 old_base = &per_cpu(hrtimer_bases, scpu);
1538 new_base = this_cpu_ptr(&hrtimer_bases);
1540 * The caller is globally serialized and nobody else
1541 * takes two locks at once, deadlock is not possible.
1543 raw_spin_lock(&new_base->lock);
1544 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1546 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1547 migrate_hrtimer_list(&old_base->clock_base[i],
1548 &new_base->clock_base[i]);
1551 raw_spin_unlock(&old_base->lock);
1552 raw_spin_unlock(&new_base->lock);
1554 /* Check, if we got expired work to do */
1555 __hrtimer_peek_ahead_timers();
1559 #endif /* CONFIG_HOTPLUG_CPU */
1561 static int hrtimer_cpu_notify(struct notifier_block *self,
1562 unsigned long action, void *hcpu)
1564 int scpu = (long)hcpu;
1568 case CPU_UP_PREPARE:
1569 case CPU_UP_PREPARE_FROZEN:
1570 init_hrtimers_cpu(scpu);
1573 #ifdef CONFIG_HOTPLUG_CPU
1575 case CPU_DEAD_FROZEN:
1576 migrate_hrtimers(scpu);
1587 static struct notifier_block hrtimers_nb = {
1588 .notifier_call = hrtimer_cpu_notify,
1591 void __init hrtimers_init(void)
1593 hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
1594 (void *)(long)smp_processor_id());
1595 register_cpu_notifier(&hrtimers_nb);
1599 * schedule_hrtimeout_range_clock - sleep until timeout
1600 * @expires: timeout value (ktime_t)
1601 * @delta: slack in expires timeout (ktime_t)
1602 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1603 * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME
1606 schedule_hrtimeout_range_clock(ktime_t *expires, unsigned long delta,
1607 const enum hrtimer_mode mode, int clock)
1609 struct hrtimer_sleeper t;
1612 * Optimize when a zero timeout value is given. It does not
1613 * matter whether this is an absolute or a relative time.
1615 if (expires && !expires->tv64) {
1616 __set_current_state(TASK_RUNNING);
1621 * A NULL parameter means "infinite"
1628 hrtimer_init_on_stack(&t.timer, clock, mode);
1629 hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
1631 hrtimer_init_sleeper(&t, current);
1633 hrtimer_start_expires(&t.timer, mode);
1638 hrtimer_cancel(&t.timer);
1639 destroy_hrtimer_on_stack(&t.timer);
1641 __set_current_state(TASK_RUNNING);
1643 return !t.task ? 0 : -EINTR;
1647 * schedule_hrtimeout_range - sleep until timeout
1648 * @expires: timeout value (ktime_t)
1649 * @delta: slack in expires timeout (ktime_t)
1650 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1652 * Make the current task sleep until the given expiry time has
1653 * elapsed. The routine will return immediately unless
1654 * the current task state has been set (see set_current_state()).
1656 * The @delta argument gives the kernel the freedom to schedule the
1657 * actual wakeup to a time that is both power and performance friendly.
1658 * The kernel give the normal best effort behavior for "@expires+@delta",
1659 * but may decide to fire the timer earlier, but no earlier than @expires.
1661 * You can set the task state as follows -
1663 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1664 * pass before the routine returns.
1666 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1667 * delivered to the current task.
1669 * The current task state is guaranteed to be TASK_RUNNING when this
1672 * Returns 0 when the timer has expired otherwise -EINTR
1674 int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta,
1675 const enum hrtimer_mode mode)
1677 return schedule_hrtimeout_range_clock(expires, delta, mode,
1680 EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
1683 * schedule_hrtimeout - sleep until timeout
1684 * @expires: timeout value (ktime_t)
1685 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1687 * Make the current task sleep until the given expiry time has
1688 * elapsed. The routine will return immediately unless
1689 * the current task state has been set (see set_current_state()).
1691 * You can set the task state as follows -
1693 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1694 * pass before the routine returns.
1696 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1697 * delivered to the current task.
1699 * The current task state is guaranteed to be TASK_RUNNING when this
1702 * Returns 0 when the timer has expired otherwise -EINTR
1704 int __sched schedule_hrtimeout(ktime_t *expires,
1705 const enum hrtimer_mode mode)
1707 return schedule_hrtimeout_range(expires, 0, mode);
1709 EXPORT_SYMBOL_GPL(schedule_hrtimeout);