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),
70 .seq = SEQCNT_ZERO(hrtimer_bases.seq),
74 .index = HRTIMER_BASE_MONOTONIC,
75 .clockid = CLOCK_MONOTONIC,
76 .get_time = &ktime_get,
79 .index = HRTIMER_BASE_REALTIME,
80 .clockid = CLOCK_REALTIME,
81 .get_time = &ktime_get_real,
84 .index = HRTIMER_BASE_BOOTTIME,
85 .clockid = CLOCK_BOOTTIME,
86 .get_time = &ktime_get_boottime,
89 .index = HRTIMER_BASE_TAI,
91 .get_time = &ktime_get_clocktai,
96 static const int hrtimer_clock_to_base_table[MAX_CLOCKS] = {
97 [CLOCK_REALTIME] = HRTIMER_BASE_REALTIME,
98 [CLOCK_MONOTONIC] = HRTIMER_BASE_MONOTONIC,
99 [CLOCK_BOOTTIME] = HRTIMER_BASE_BOOTTIME,
100 [CLOCK_TAI] = HRTIMER_BASE_TAI,
103 static inline int hrtimer_clockid_to_base(clockid_t clock_id)
105 return hrtimer_clock_to_base_table[clock_id];
109 * Functions and macros which are different for UP/SMP systems are kept in a
115 * We require the migration_base for lock_hrtimer_base()/switch_hrtimer_base()
116 * such that hrtimer_callback_running() can unconditionally dereference
117 * timer->base->cpu_base
119 static struct hrtimer_cpu_base migration_cpu_base = {
120 .seq = SEQCNT_ZERO(migration_cpu_base),
121 .clock_base = { { .cpu_base = &migration_cpu_base, }, },
124 #define migration_base migration_cpu_base.clock_base[0]
127 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
128 * means that all timers which are tied to this base via timer->base are
129 * locked, and the base itself is locked too.
131 * So __run_timers/migrate_timers can safely modify all timers which could
132 * be found on the lists/queues.
134 * When the timer's base is locked, and the timer removed from list, it is
135 * possible to set timer->base = &migration_base and drop the lock: the timer
139 struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
140 unsigned long *flags)
142 struct hrtimer_clock_base *base;
146 if (likely(base != &migration_base)) {
147 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
148 if (likely(base == timer->base))
150 /* The timer has migrated to another CPU: */
151 raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
158 * With HIGHRES=y we do not migrate the timer when it is expiring
159 * before the next event on the target cpu because we cannot reprogram
160 * the target cpu hardware and we would cause it to fire late.
162 * Called with cpu_base->lock of target cpu held.
165 hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base)
167 #ifdef CONFIG_HIGH_RES_TIMERS
170 if (!new_base->cpu_base->hres_active)
173 expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset);
174 return expires.tv64 <= new_base->cpu_base->expires_next.tv64;
181 * Switch the timer base to the current CPU when possible.
183 static inline struct hrtimer_clock_base *
184 switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
187 struct hrtimer_clock_base *new_base;
188 struct hrtimer_cpu_base *new_cpu_base;
189 int this_cpu = smp_processor_id();
190 int cpu = get_nohz_timer_target(pinned);
191 int basenum = base->index;
194 new_cpu_base = &per_cpu(hrtimer_bases, cpu);
195 new_base = &new_cpu_base->clock_base[basenum];
197 if (base != new_base) {
199 * We are trying to move timer to new_base.
200 * However we can't change timer's base while it is running,
201 * so we keep it on the same CPU. No hassle vs. reprogramming
202 * the event source in the high resolution case. The softirq
203 * code will take care of this when the timer function has
204 * completed. There is no conflict as we hold the lock until
205 * the timer is enqueued.
207 if (unlikely(hrtimer_callback_running(timer)))
210 /* See the comment in lock_hrtimer_base() */
211 timer->base = &migration_base;
212 raw_spin_unlock(&base->cpu_base->lock);
213 raw_spin_lock(&new_base->cpu_base->lock);
215 if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) {
217 raw_spin_unlock(&new_base->cpu_base->lock);
218 raw_spin_lock(&base->cpu_base->lock);
222 timer->base = new_base;
224 if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) {
232 #else /* CONFIG_SMP */
234 static inline struct hrtimer_clock_base *
235 lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
237 struct hrtimer_clock_base *base = timer->base;
239 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
244 # define switch_hrtimer_base(t, b, p) (b)
246 #endif /* !CONFIG_SMP */
249 * Functions for the union type storage format of ktime_t which are
250 * too large for inlining:
252 #if BITS_PER_LONG < 64
254 * Divide a ktime value by a nanosecond value
256 u64 __ktime_divns(const ktime_t kt, s64 div)
261 dclc = ktime_to_ns(kt);
262 /* Make sure the divisor is less than 2^32: */
268 do_div(dclc, (unsigned long) div);
272 EXPORT_SYMBOL_GPL(__ktime_divns);
273 #endif /* BITS_PER_LONG >= 64 */
276 * Add two ktime values and do a safety check for overflow:
278 ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
280 ktime_t res = ktime_add(lhs, rhs);
283 * We use KTIME_SEC_MAX here, the maximum timeout which we can
284 * return to user space in a timespec:
286 if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64)
287 res = ktime_set(KTIME_SEC_MAX, 0);
292 EXPORT_SYMBOL_GPL(ktime_add_safe);
294 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
296 static struct debug_obj_descr hrtimer_debug_descr;
298 static void *hrtimer_debug_hint(void *addr)
300 return ((struct hrtimer *) addr)->function;
304 * fixup_init is called when:
305 * - an active object is initialized
307 static int hrtimer_fixup_init(void *addr, enum debug_obj_state state)
309 struct hrtimer *timer = addr;
312 case ODEBUG_STATE_ACTIVE:
313 hrtimer_cancel(timer);
314 debug_object_init(timer, &hrtimer_debug_descr);
322 * fixup_activate is called when:
323 * - an active object is activated
324 * - an unknown object is activated (might be a statically initialized object)
326 static int hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
330 case ODEBUG_STATE_NOTAVAILABLE:
334 case ODEBUG_STATE_ACTIVE:
343 * fixup_free is called when:
344 * - an active object is freed
346 static int hrtimer_fixup_free(void *addr, enum debug_obj_state state)
348 struct hrtimer *timer = addr;
351 case ODEBUG_STATE_ACTIVE:
352 hrtimer_cancel(timer);
353 debug_object_free(timer, &hrtimer_debug_descr);
360 static struct debug_obj_descr hrtimer_debug_descr = {
362 .debug_hint = hrtimer_debug_hint,
363 .fixup_init = hrtimer_fixup_init,
364 .fixup_activate = hrtimer_fixup_activate,
365 .fixup_free = hrtimer_fixup_free,
368 static inline void debug_hrtimer_init(struct hrtimer *timer)
370 debug_object_init(timer, &hrtimer_debug_descr);
373 static inline void debug_hrtimer_activate(struct hrtimer *timer)
375 debug_object_activate(timer, &hrtimer_debug_descr);
378 static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
380 debug_object_deactivate(timer, &hrtimer_debug_descr);
383 static inline void debug_hrtimer_free(struct hrtimer *timer)
385 debug_object_free(timer, &hrtimer_debug_descr);
388 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
389 enum hrtimer_mode mode);
391 void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
392 enum hrtimer_mode mode)
394 debug_object_init_on_stack(timer, &hrtimer_debug_descr);
395 __hrtimer_init(timer, clock_id, mode);
397 EXPORT_SYMBOL_GPL(hrtimer_init_on_stack);
399 void destroy_hrtimer_on_stack(struct hrtimer *timer)
401 debug_object_free(timer, &hrtimer_debug_descr);
405 static inline void debug_hrtimer_init(struct hrtimer *timer) { }
406 static inline void debug_hrtimer_activate(struct hrtimer *timer) { }
407 static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
411 debug_init(struct hrtimer *timer, clockid_t clockid,
412 enum hrtimer_mode mode)
414 debug_hrtimer_init(timer);
415 trace_hrtimer_init(timer, clockid, mode);
418 static inline void debug_activate(struct hrtimer *timer)
420 debug_hrtimer_activate(timer);
421 trace_hrtimer_start(timer);
424 static inline void debug_deactivate(struct hrtimer *timer)
426 debug_hrtimer_deactivate(timer);
427 trace_hrtimer_cancel(timer);
430 #if defined(CONFIG_NO_HZ_COMMON) || defined(CONFIG_HIGH_RES_TIMERS)
431 static inline void hrtimer_update_next_timer(struct hrtimer_cpu_base *cpu_base,
432 struct hrtimer *timer)
434 #ifdef CONFIG_HIGH_RES_TIMERS
435 cpu_base->next_timer = timer;
439 static ktime_t __hrtimer_get_next_event(struct hrtimer_cpu_base *cpu_base)
441 struct hrtimer_clock_base *base = cpu_base->clock_base;
442 ktime_t expires, expires_next = { .tv64 = KTIME_MAX };
443 unsigned int active = cpu_base->active_bases;
445 hrtimer_update_next_timer(cpu_base, NULL);
446 for (; active; base++, active >>= 1) {
447 struct timerqueue_node *next;
448 struct hrtimer *timer;
450 if (!(active & 0x01))
453 next = timerqueue_getnext(&base->active);
454 timer = container_of(next, struct hrtimer, node);
455 expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
456 if (expires.tv64 < expires_next.tv64) {
457 expires_next = expires;
458 hrtimer_update_next_timer(cpu_base, timer);
462 * clock_was_set() might have changed base->offset of any of
463 * the clock bases so the result might be negative. Fix it up
464 * to prevent a false positive in clockevents_program_event().
466 if (expires_next.tv64 < 0)
467 expires_next.tv64 = 0;
472 static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base)
474 ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset;
475 ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset;
476 ktime_t *offs_tai = &base->clock_base[HRTIMER_BASE_TAI].offset;
478 return ktime_get_update_offsets_now(&base->clock_was_set_seq,
479 offs_real, offs_boot, offs_tai);
482 /* High resolution timer related functions */
483 #ifdef CONFIG_HIGH_RES_TIMERS
486 * High resolution timer enabled ?
488 static int hrtimer_hres_enabled __read_mostly = 1;
489 unsigned int hrtimer_resolution __read_mostly = LOW_RES_NSEC;
490 EXPORT_SYMBOL_GPL(hrtimer_resolution);
493 * Enable / Disable high resolution mode
495 static int __init setup_hrtimer_hres(char *str)
497 if (!strcmp(str, "off"))
498 hrtimer_hres_enabled = 0;
499 else if (!strcmp(str, "on"))
500 hrtimer_hres_enabled = 1;
506 __setup("highres=", setup_hrtimer_hres);
509 * hrtimer_high_res_enabled - query, if the highres mode is enabled
511 static inline int hrtimer_is_hres_enabled(void)
513 return hrtimer_hres_enabled;
517 * Is the high resolution mode active ?
519 static inline int __hrtimer_hres_active(struct hrtimer_cpu_base *cpu_base)
521 return cpu_base->hres_active;
524 static inline int hrtimer_hres_active(void)
526 return __hrtimer_hres_active(this_cpu_ptr(&hrtimer_bases));
530 * Reprogram the event source with checking both queues for the
532 * Called with interrupts disabled and base->lock held
535 hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
537 ktime_t expires_next;
539 if (!cpu_base->hres_active)
542 expires_next = __hrtimer_get_next_event(cpu_base);
544 if (skip_equal && expires_next.tv64 == cpu_base->expires_next.tv64)
547 cpu_base->expires_next.tv64 = expires_next.tv64;
550 * If a hang was detected in the last timer interrupt then we
551 * leave the hang delay active in the hardware. We want the
552 * system to make progress. That also prevents the following
554 * T1 expires 50ms from now
555 * T2 expires 5s from now
557 * T1 is removed, so this code is called and would reprogram
558 * the hardware to 5s from now. Any hrtimer_start after that
559 * will not reprogram the hardware due to hang_detected being
560 * set. So we'd effectivly block all timers until the T2 event
563 if (cpu_base->hang_detected)
566 tick_program_event(cpu_base->expires_next, 1);
570 * When a timer is enqueued and expires earlier than the already enqueued
571 * timers, we have to check, whether it expires earlier than the timer for
572 * which the clock event device was armed.
574 * Called with interrupts disabled and base->cpu_base.lock held
576 static void hrtimer_reprogram(struct hrtimer *timer,
577 struct hrtimer_clock_base *base)
579 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
580 ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
582 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
585 * If the timer is not on the current cpu, we cannot reprogram
586 * the other cpus clock event device.
588 if (base->cpu_base != cpu_base)
592 * If the hrtimer interrupt is running, then it will
593 * reevaluate the clock bases and reprogram the clock event
594 * device. The callbacks are always executed in hard interrupt
595 * context so we don't need an extra check for a running
598 if (cpu_base->in_hrtirq)
602 * CLOCK_REALTIME timer might be requested with an absolute
603 * expiry time which is less than base->offset. Set it to 0.
605 if (expires.tv64 < 0)
608 if (expires.tv64 >= cpu_base->expires_next.tv64)
611 /* Update the pointer to the next expiring timer */
612 cpu_base->next_timer = timer;
615 * If a hang was detected in the last timer interrupt then we
616 * do not schedule a timer which is earlier than the expiry
617 * which we enforced in the hang detection. We want the system
620 if (cpu_base->hang_detected)
624 * Program the timer hardware. We enforce the expiry for
625 * events which are already in the past.
627 cpu_base->expires_next = expires;
628 tick_program_event(expires, 1);
632 * Initialize the high resolution related parts of cpu_base
634 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
636 base->expires_next.tv64 = KTIME_MAX;
637 base->hres_active = 0;
641 * Retrigger next event is called after clock was set
643 * Called with interrupts disabled via on_each_cpu()
645 static void retrigger_next_event(void *arg)
647 struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
649 if (!base->hres_active)
652 raw_spin_lock(&base->lock);
653 hrtimer_update_base(base);
654 hrtimer_force_reprogram(base, 0);
655 raw_spin_unlock(&base->lock);
659 * Switch to high resolution mode
661 static int hrtimer_switch_to_hres(void)
663 struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
665 if (tick_init_highres()) {
666 printk(KERN_WARNING "Could not switch to high resolution "
667 "mode on CPU %d\n", base->cpu);
670 base->hres_active = 1;
671 hrtimer_resolution = HIGH_RES_NSEC;
673 tick_setup_sched_timer();
674 /* "Retrigger" the interrupt to get things going */
675 retrigger_next_event(NULL);
679 static void clock_was_set_work(struct work_struct *work)
684 static DECLARE_WORK(hrtimer_work, clock_was_set_work);
687 * Called from timekeeping and resume code to reprogramm the hrtimer
688 * interrupt device on all cpus.
690 void clock_was_set_delayed(void)
692 schedule_work(&hrtimer_work);
697 static inline int __hrtimer_hres_active(struct hrtimer_cpu_base *b) { return 0; }
698 static inline int hrtimer_hres_active(void) { return 0; }
699 static inline int hrtimer_is_hres_enabled(void) { return 0; }
700 static inline int hrtimer_switch_to_hres(void) { return 0; }
702 hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { }
703 static inline int hrtimer_reprogram(struct hrtimer *timer,
704 struct hrtimer_clock_base *base)
708 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { }
709 static inline void retrigger_next_event(void *arg) { }
711 #endif /* CONFIG_HIGH_RES_TIMERS */
714 * Clock realtime was set
716 * Change the offset of the realtime clock vs. the monotonic
719 * We might have to reprogram the high resolution timer interrupt. On
720 * SMP we call the architecture specific code to retrigger _all_ high
721 * resolution timer interrupts. On UP we just disable interrupts and
722 * call the high resolution interrupt code.
724 void clock_was_set(void)
726 #ifdef CONFIG_HIGH_RES_TIMERS
727 /* Retrigger the CPU local events everywhere */
728 on_each_cpu(retrigger_next_event, NULL, 1);
730 timerfd_clock_was_set();
734 * During resume we might have to reprogram the high resolution timer
735 * interrupt on all online CPUs. However, all other CPUs will be
736 * stopped with IRQs interrupts disabled so the clock_was_set() call
739 void hrtimers_resume(void)
741 WARN_ONCE(!irqs_disabled(),
742 KERN_INFO "hrtimers_resume() called with IRQs enabled!");
744 /* Retrigger on the local CPU */
745 retrigger_next_event(NULL);
746 /* And schedule a retrigger for all others */
747 clock_was_set_delayed();
750 static inline void timer_stats_hrtimer_set_start_info(struct hrtimer *timer)
752 #ifdef CONFIG_TIMER_STATS
753 if (timer->start_site)
755 timer->start_site = __builtin_return_address(0);
756 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
757 timer->start_pid = current->pid;
761 static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer *timer)
763 #ifdef CONFIG_TIMER_STATS
764 timer->start_site = NULL;
768 static inline void timer_stats_account_hrtimer(struct hrtimer *timer)
770 #ifdef CONFIG_TIMER_STATS
771 if (likely(!timer_stats_active))
773 timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
774 timer->function, timer->start_comm, 0);
779 * Counterpart to lock_hrtimer_base above:
782 void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
784 raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
788 * hrtimer_forward - forward the timer expiry
789 * @timer: hrtimer to forward
790 * @now: forward past this time
791 * @interval: the interval to forward
793 * Forward the timer expiry so it will expire in the future.
794 * Returns the number of overruns.
796 * Can be safely called from the callback function of @timer. If
797 * called from other contexts @timer must neither be enqueued nor
798 * running the callback and the caller needs to take care of
801 * Note: This only updates the timer expiry value and does not requeue
804 u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
809 delta = ktime_sub(now, hrtimer_get_expires(timer));
814 if (WARN_ON(timer->state & HRTIMER_STATE_ENQUEUED))
817 if (interval.tv64 < hrtimer_resolution)
818 interval.tv64 = hrtimer_resolution;
820 if (unlikely(delta.tv64 >= interval.tv64)) {
821 s64 incr = ktime_to_ns(interval);
823 orun = ktime_divns(delta, incr);
824 hrtimer_add_expires_ns(timer, incr * orun);
825 if (hrtimer_get_expires_tv64(timer) > now.tv64)
828 * This (and the ktime_add() below) is the
829 * correction for exact:
833 hrtimer_add_expires(timer, interval);
837 EXPORT_SYMBOL_GPL(hrtimer_forward);
840 * enqueue_hrtimer - internal function to (re)start a timer
842 * The timer is inserted in expiry order. Insertion into the
843 * red black tree is O(log(n)). Must hold the base lock.
845 * Returns 1 when the new timer is the leftmost timer in the tree.
847 static int enqueue_hrtimer(struct hrtimer *timer,
848 struct hrtimer_clock_base *base)
850 debug_activate(timer);
852 base->cpu_base->active_bases |= 1 << base->index;
854 timer->state = HRTIMER_STATE_ENQUEUED;
856 return timerqueue_add(&base->active, &timer->node);
860 * __remove_hrtimer - internal function to remove a timer
862 * Caller must hold the base lock.
864 * High resolution timer mode reprograms the clock event device when the
865 * timer is the one which expires next. The caller can disable this by setting
866 * reprogram to zero. This is useful, when the context does a reprogramming
867 * anyway (e.g. timer interrupt)
869 static void __remove_hrtimer(struct hrtimer *timer,
870 struct hrtimer_clock_base *base,
871 unsigned long newstate, int reprogram)
873 struct hrtimer_cpu_base *cpu_base = base->cpu_base;
874 unsigned int state = timer->state;
876 timer->state = newstate;
877 if (!(state & HRTIMER_STATE_ENQUEUED))
880 if (!timerqueue_del(&base->active, &timer->node))
881 cpu_base->active_bases &= ~(1 << base->index);
883 #ifdef CONFIG_HIGH_RES_TIMERS
885 * Note: If reprogram is false we do not update
886 * cpu_base->next_timer. This happens when we remove the first
887 * timer on a remote cpu. No harm as we never dereference
888 * cpu_base->next_timer. So the worst thing what can happen is
889 * an superflous call to hrtimer_force_reprogram() on the
890 * remote cpu later on if the same timer gets enqueued again.
892 if (reprogram && timer == cpu_base->next_timer)
893 hrtimer_force_reprogram(cpu_base, 1);
898 * remove hrtimer, called with base lock held
901 remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base, bool restart)
903 if (hrtimer_is_queued(timer)) {
904 unsigned long state = timer->state;
908 * Remove the timer and force reprogramming when high
909 * resolution mode is active and the timer is on the current
910 * CPU. If we remove a timer on another CPU, reprogramming is
911 * skipped. The interrupt event on this CPU is fired and
912 * reprogramming happens in the interrupt handler. This is a
913 * rare case and less expensive than a smp call.
915 debug_deactivate(timer);
916 timer_stats_hrtimer_clear_start_info(timer);
917 reprogram = base->cpu_base == this_cpu_ptr(&hrtimer_bases);
920 state = HRTIMER_STATE_INACTIVE;
922 __remove_hrtimer(timer, base, state, reprogram);
929 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
930 * @timer: the timer to be added
932 * @delta_ns: "slack" range for the timer
933 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
934 * relative (HRTIMER_MODE_REL)
936 void hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
937 unsigned long delta_ns, const enum hrtimer_mode mode)
939 struct hrtimer_clock_base *base, *new_base;
943 base = lock_hrtimer_base(timer, &flags);
945 /* Remove an active timer from the queue: */
946 remove_hrtimer(timer, base, true);
948 if (mode & HRTIMER_MODE_REL) {
949 tim = ktime_add_safe(tim, base->get_time());
951 * CONFIG_TIME_LOW_RES is a temporary way for architectures
952 * to signal that they simply return xtime in
953 * do_gettimeoffset(). In this case we want to round up by
954 * resolution when starting a relative timer, to avoid short
955 * timeouts. This will go away with the GTOD framework.
957 #ifdef CONFIG_TIME_LOW_RES
958 tim = ktime_add_safe(tim, ktime_set(0, hrtimer_resolution));
962 hrtimer_set_expires_range_ns(timer, tim, delta_ns);
964 /* Switch the timer base, if necessary: */
965 new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED);
967 timer_stats_hrtimer_set_start_info(timer);
969 leftmost = enqueue_hrtimer(timer, new_base);
973 if (!hrtimer_is_hres_active(timer)) {
975 * Kick to reschedule the next tick to handle the new timer
976 * on dynticks target.
978 wake_up_nohz_cpu(new_base->cpu_base->cpu);
980 hrtimer_reprogram(timer, new_base);
983 unlock_hrtimer_base(timer, &flags);
985 EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
988 * hrtimer_try_to_cancel - try to deactivate a timer
989 * @timer: hrtimer to stop
992 * 0 when the timer was not active
993 * 1 when the timer was active
994 * -1 when the timer is currently excuting the callback function and
997 int hrtimer_try_to_cancel(struct hrtimer *timer)
999 struct hrtimer_clock_base *base;
1000 unsigned long flags;
1004 * Check lockless first. If the timer is not active (neither
1005 * enqueued nor running the callback, nothing to do here. The
1006 * base lock does not serialize against a concurrent enqueue,
1007 * so we can avoid taking it.
1009 if (!hrtimer_active(timer))
1012 base = lock_hrtimer_base(timer, &flags);
1014 if (!hrtimer_callback_running(timer))
1015 ret = remove_hrtimer(timer, base, false);
1017 unlock_hrtimer_base(timer, &flags);
1022 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1025 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1026 * @timer: the timer to be cancelled
1029 * 0 when the timer was not active
1030 * 1 when the timer was active
1032 int hrtimer_cancel(struct hrtimer *timer)
1035 int ret = hrtimer_try_to_cancel(timer);
1042 EXPORT_SYMBOL_GPL(hrtimer_cancel);
1045 * hrtimer_get_remaining - get remaining time for the timer
1046 * @timer: the timer to read
1048 ktime_t hrtimer_get_remaining(const struct hrtimer *timer)
1050 unsigned long flags;
1053 lock_hrtimer_base(timer, &flags);
1054 rem = hrtimer_expires_remaining(timer);
1055 unlock_hrtimer_base(timer, &flags);
1059 EXPORT_SYMBOL_GPL(hrtimer_get_remaining);
1061 #ifdef CONFIG_NO_HZ_COMMON
1063 * hrtimer_get_next_event - get the time until next expiry event
1065 * Returns the next expiry time or KTIME_MAX if no timer is pending.
1067 u64 hrtimer_get_next_event(void)
1069 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1070 u64 expires = KTIME_MAX;
1071 unsigned long flags;
1073 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1075 if (!__hrtimer_hres_active(cpu_base))
1076 expires = __hrtimer_get_next_event(cpu_base).tv64;
1078 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1084 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1085 enum hrtimer_mode mode)
1087 struct hrtimer_cpu_base *cpu_base;
1090 memset(timer, 0, sizeof(struct hrtimer));
1092 cpu_base = raw_cpu_ptr(&hrtimer_bases);
1094 if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
1095 clock_id = CLOCK_MONOTONIC;
1097 base = hrtimer_clockid_to_base(clock_id);
1098 timer->base = &cpu_base->clock_base[base];
1099 timerqueue_init(&timer->node);
1101 #ifdef CONFIG_TIMER_STATS
1102 timer->start_site = NULL;
1103 timer->start_pid = -1;
1104 memset(timer->start_comm, 0, TASK_COMM_LEN);
1109 * hrtimer_init - initialize a timer to the given clock
1110 * @timer: the timer to be initialized
1111 * @clock_id: the clock to be used
1112 * @mode: timer mode abs/rel
1114 void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1115 enum hrtimer_mode mode)
1117 debug_init(timer, clock_id, mode);
1118 __hrtimer_init(timer, clock_id, mode);
1120 EXPORT_SYMBOL_GPL(hrtimer_init);
1123 * A timer is active, when it is enqueued into the rbtree or the
1124 * callback function is running or it's in the state of being migrated
1127 * It is important for this function to not return a false negative.
1129 bool hrtimer_active(const struct hrtimer *timer)
1131 struct hrtimer_cpu_base *cpu_base;
1135 cpu_base = READ_ONCE(timer->base->cpu_base);
1136 seq = raw_read_seqcount_begin(&cpu_base->seq);
1138 if (timer->state != HRTIMER_STATE_INACTIVE ||
1139 cpu_base->running == timer)
1142 } while (read_seqcount_retry(&cpu_base->seq, seq) ||
1143 cpu_base != READ_ONCE(timer->base->cpu_base));
1147 EXPORT_SYMBOL_GPL(hrtimer_active);
1150 * The write_seqcount_barrier()s in __run_hrtimer() split the thing into 3
1151 * distinct sections:
1153 * - queued: the timer is queued
1154 * - callback: the timer is being ran
1155 * - post: the timer is inactive or (re)queued
1157 * On the read side we ensure we observe timer->state and cpu_base->running
1158 * from the same section, if anything changed while we looked at it, we retry.
1159 * This includes timer->base changing because sequence numbers alone are
1160 * insufficient for that.
1162 * The sequence numbers are required because otherwise we could still observe
1163 * a false negative if the read side got smeared over multiple consequtive
1164 * __run_hrtimer() invocations.
1167 static void __run_hrtimer(struct hrtimer_cpu_base *cpu_base,
1168 struct hrtimer_clock_base *base,
1169 struct hrtimer *timer, ktime_t *now)
1171 enum hrtimer_restart (*fn)(struct hrtimer *);
1174 lockdep_assert_held(&cpu_base->lock);
1176 debug_deactivate(timer);
1177 cpu_base->running = timer;
1180 * Separate the ->running assignment from the ->state assignment.
1182 * As with a regular write barrier, this ensures the read side in
1183 * hrtimer_active() cannot observe cpu_base->running == NULL &&
1184 * timer->state == INACTIVE.
1186 raw_write_seqcount_barrier(&cpu_base->seq);
1188 __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE, 0);
1189 timer_stats_account_hrtimer(timer);
1190 fn = timer->function;
1193 * Because we run timers from hardirq context, there is no chance
1194 * they get migrated to another cpu, therefore its safe to unlock
1197 raw_spin_unlock(&cpu_base->lock);
1198 trace_hrtimer_expire_entry(timer, now);
1199 restart = fn(timer);
1200 trace_hrtimer_expire_exit(timer);
1201 raw_spin_lock(&cpu_base->lock);
1204 * Note: We clear the running state after enqueue_hrtimer and
1205 * we do not reprogramm the event hardware. Happens either in
1206 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1208 * Note: Because we dropped the cpu_base->lock above,
1209 * hrtimer_start_range_ns() can have popped in and enqueued the timer
1212 if (restart != HRTIMER_NORESTART &&
1213 !(timer->state & HRTIMER_STATE_ENQUEUED))
1214 enqueue_hrtimer(timer, base);
1217 * Separate the ->running assignment from the ->state assignment.
1219 * As with a regular write barrier, this ensures the read side in
1220 * hrtimer_active() cannot observe cpu_base->running == NULL &&
1221 * timer->state == INACTIVE.
1223 raw_write_seqcount_barrier(&cpu_base->seq);
1225 WARN_ON_ONCE(cpu_base->running != timer);
1226 cpu_base->running = NULL;
1229 static void __hrtimer_run_queues(struct hrtimer_cpu_base *cpu_base, ktime_t now)
1231 struct hrtimer_clock_base *base = cpu_base->clock_base;
1232 unsigned int active = cpu_base->active_bases;
1234 for (; active; base++, active >>= 1) {
1235 struct timerqueue_node *node;
1238 if (!(active & 0x01))
1241 basenow = ktime_add(now, base->offset);
1243 while ((node = timerqueue_getnext(&base->active))) {
1244 struct hrtimer *timer;
1246 timer = container_of(node, struct hrtimer, node);
1249 * The immediate goal for using the softexpires is
1250 * minimizing wakeups, not running timers at the
1251 * earliest interrupt after their soft expiration.
1252 * This allows us to avoid using a Priority Search
1253 * Tree, which can answer a stabbing querry for
1254 * overlapping intervals and instead use the simple
1255 * BST we already have.
1256 * We don't add extra wakeups by delaying timers that
1257 * are right-of a not yet expired timer, because that
1258 * timer will have to trigger a wakeup anyway.
1260 if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer))
1263 __run_hrtimer(cpu_base, base, timer, &basenow);
1268 #ifdef CONFIG_HIGH_RES_TIMERS
1271 * High resolution timer interrupt
1272 * Called with interrupts disabled
1274 void hrtimer_interrupt(struct clock_event_device *dev)
1276 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1277 ktime_t expires_next, now, entry_time, delta;
1280 BUG_ON(!cpu_base->hres_active);
1281 cpu_base->nr_events++;
1282 dev->next_event.tv64 = KTIME_MAX;
1284 raw_spin_lock(&cpu_base->lock);
1285 entry_time = now = hrtimer_update_base(cpu_base);
1287 cpu_base->in_hrtirq = 1;
1289 * We set expires_next to KTIME_MAX here with cpu_base->lock
1290 * held to prevent that a timer is enqueued in our queue via
1291 * the migration code. This does not affect enqueueing of
1292 * timers which run their callback and need to be requeued on
1295 cpu_base->expires_next.tv64 = KTIME_MAX;
1297 __hrtimer_run_queues(cpu_base, now);
1299 /* Reevaluate the clock bases for the next expiry */
1300 expires_next = __hrtimer_get_next_event(cpu_base);
1302 * Store the new expiry value so the migration code can verify
1305 cpu_base->expires_next = expires_next;
1306 cpu_base->in_hrtirq = 0;
1307 raw_spin_unlock(&cpu_base->lock);
1309 /* Reprogramming necessary ? */
1310 if (!tick_program_event(expires_next, 0)) {
1311 cpu_base->hang_detected = 0;
1316 * The next timer was already expired due to:
1318 * - long lasting callbacks
1319 * - being scheduled away when running in a VM
1321 * We need to prevent that we loop forever in the hrtimer
1322 * interrupt routine. We give it 3 attempts to avoid
1323 * overreacting on some spurious event.
1325 * Acquire base lock for updating the offsets and retrieving
1328 raw_spin_lock(&cpu_base->lock);
1329 now = hrtimer_update_base(cpu_base);
1330 cpu_base->nr_retries++;
1334 * Give the system a chance to do something else than looping
1335 * here. We stored the entry time, so we know exactly how long
1336 * we spent here. We schedule the next event this amount of
1339 cpu_base->nr_hangs++;
1340 cpu_base->hang_detected = 1;
1341 raw_spin_unlock(&cpu_base->lock);
1342 delta = ktime_sub(now, entry_time);
1343 if ((unsigned int)delta.tv64 > cpu_base->max_hang_time)
1344 cpu_base->max_hang_time = (unsigned int) delta.tv64;
1346 * Limit it to a sensible value as we enforce a longer
1347 * delay. Give the CPU at least 100ms to catch up.
1349 if (delta.tv64 > 100 * NSEC_PER_MSEC)
1350 expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1352 expires_next = ktime_add(now, delta);
1353 tick_program_event(expires_next, 1);
1354 printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n",
1355 ktime_to_ns(delta));
1359 * local version of hrtimer_peek_ahead_timers() called with interrupts
1362 static inline void __hrtimer_peek_ahead_timers(void)
1364 struct tick_device *td;
1366 if (!hrtimer_hres_active())
1369 td = this_cpu_ptr(&tick_cpu_device);
1370 if (td && td->evtdev)
1371 hrtimer_interrupt(td->evtdev);
1374 #else /* CONFIG_HIGH_RES_TIMERS */
1376 static inline void __hrtimer_peek_ahead_timers(void) { }
1378 #endif /* !CONFIG_HIGH_RES_TIMERS */
1381 * Called from run_local_timers in hardirq context every jiffy
1383 void hrtimer_run_queues(void)
1385 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1388 if (__hrtimer_hres_active(cpu_base))
1392 * This _is_ ugly: We have to check periodically, whether we
1393 * can switch to highres and / or nohz mode. The clocksource
1394 * switch happens with xtime_lock held. Notification from
1395 * there only sets the check bit in the tick_oneshot code,
1396 * otherwise we might deadlock vs. xtime_lock.
1398 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) {
1399 hrtimer_switch_to_hres();
1403 raw_spin_lock(&cpu_base->lock);
1404 now = hrtimer_update_base(cpu_base);
1405 __hrtimer_run_queues(cpu_base, now);
1406 raw_spin_unlock(&cpu_base->lock);
1410 * Sleep related functions:
1412 static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1414 struct hrtimer_sleeper *t =
1415 container_of(timer, struct hrtimer_sleeper, timer);
1416 struct task_struct *task = t->task;
1420 wake_up_process(task);
1422 return HRTIMER_NORESTART;
1425 void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1427 sl->timer.function = hrtimer_wakeup;
1430 EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
1432 static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1434 hrtimer_init_sleeper(t, current);
1437 set_current_state(TASK_INTERRUPTIBLE);
1438 hrtimer_start_expires(&t->timer, mode);
1440 if (likely(t->task))
1441 freezable_schedule();
1443 hrtimer_cancel(&t->timer);
1444 mode = HRTIMER_MODE_ABS;
1446 } while (t->task && !signal_pending(current));
1448 __set_current_state(TASK_RUNNING);
1450 return t->task == NULL;
1453 static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp)
1455 struct timespec rmt;
1458 rem = hrtimer_expires_remaining(timer);
1461 rmt = ktime_to_timespec(rem);
1463 if (copy_to_user(rmtp, &rmt, sizeof(*rmtp)))
1469 long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1471 struct hrtimer_sleeper t;
1472 struct timespec __user *rmtp;
1475 hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid,
1477 hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
1479 if (do_nanosleep(&t, HRTIMER_MODE_ABS))
1482 rmtp = restart->nanosleep.rmtp;
1484 ret = update_rmtp(&t.timer, rmtp);
1489 /* The other values in restart are already filled in */
1490 ret = -ERESTART_RESTARTBLOCK;
1492 destroy_hrtimer_on_stack(&t.timer);
1496 long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
1497 const enum hrtimer_mode mode, const clockid_t clockid)
1499 struct restart_block *restart;
1500 struct hrtimer_sleeper t;
1502 unsigned long slack;
1504 slack = current->timer_slack_ns;
1505 if (dl_task(current) || rt_task(current))
1508 hrtimer_init_on_stack(&t.timer, clockid, mode);
1509 hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack);
1510 if (do_nanosleep(&t, mode))
1513 /* Absolute timers do not update the rmtp value and restart: */
1514 if (mode == HRTIMER_MODE_ABS) {
1515 ret = -ERESTARTNOHAND;
1520 ret = update_rmtp(&t.timer, rmtp);
1525 restart = ¤t->restart_block;
1526 restart->fn = hrtimer_nanosleep_restart;
1527 restart->nanosleep.clockid = t.timer.base->clockid;
1528 restart->nanosleep.rmtp = rmtp;
1529 restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
1531 ret = -ERESTART_RESTARTBLOCK;
1533 destroy_hrtimer_on_stack(&t.timer);
1537 SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp,
1538 struct timespec __user *, rmtp)
1542 if (copy_from_user(&tu, rqtp, sizeof(tu)))
1545 if (!timespec_valid(&tu))
1548 return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1552 * Functions related to boot-time initialization:
1554 static void init_hrtimers_cpu(int cpu)
1556 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1559 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1560 cpu_base->clock_base[i].cpu_base = cpu_base;
1561 timerqueue_init_head(&cpu_base->clock_base[i].active);
1564 cpu_base->cpu = cpu;
1565 hrtimer_init_hres(cpu_base);
1568 #ifdef CONFIG_HOTPLUG_CPU
1570 static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1571 struct hrtimer_clock_base *new_base)
1573 struct hrtimer *timer;
1574 struct timerqueue_node *node;
1576 while ((node = timerqueue_getnext(&old_base->active))) {
1577 timer = container_of(node, struct hrtimer, node);
1578 BUG_ON(hrtimer_callback_running(timer));
1579 debug_deactivate(timer);
1582 * Mark it as ENQUEUED not INACTIVE otherwise the
1583 * timer could be seen as !active and just vanish away
1584 * under us on another CPU
1586 __remove_hrtimer(timer, old_base, HRTIMER_STATE_ENQUEUED, 0);
1587 timer->base = new_base;
1589 * Enqueue the timers on the new cpu. This does not
1590 * reprogram the event device in case the timer
1591 * expires before the earliest on this CPU, but we run
1592 * hrtimer_interrupt after we migrated everything to
1593 * sort out already expired timers and reprogram the
1596 enqueue_hrtimer(timer, new_base);
1600 static void migrate_hrtimers(int scpu)
1602 struct hrtimer_cpu_base *old_base, *new_base;
1605 BUG_ON(cpu_online(scpu));
1606 tick_cancel_sched_timer(scpu);
1608 local_irq_disable();
1609 old_base = &per_cpu(hrtimer_bases, scpu);
1610 new_base = this_cpu_ptr(&hrtimer_bases);
1612 * The caller is globally serialized and nobody else
1613 * takes two locks at once, deadlock is not possible.
1615 raw_spin_lock(&new_base->lock);
1616 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1618 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1619 migrate_hrtimer_list(&old_base->clock_base[i],
1620 &new_base->clock_base[i]);
1623 raw_spin_unlock(&old_base->lock);
1624 raw_spin_unlock(&new_base->lock);
1626 /* Check, if we got expired work to do */
1627 __hrtimer_peek_ahead_timers();
1631 #endif /* CONFIG_HOTPLUG_CPU */
1633 static int hrtimer_cpu_notify(struct notifier_block *self,
1634 unsigned long action, void *hcpu)
1636 int scpu = (long)hcpu;
1640 case CPU_UP_PREPARE:
1641 case CPU_UP_PREPARE_FROZEN:
1642 init_hrtimers_cpu(scpu);
1645 #ifdef CONFIG_HOTPLUG_CPU
1647 case CPU_DEAD_FROZEN:
1648 migrate_hrtimers(scpu);
1659 static struct notifier_block hrtimers_nb = {
1660 .notifier_call = hrtimer_cpu_notify,
1663 void __init hrtimers_init(void)
1665 hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
1666 (void *)(long)smp_processor_id());
1667 register_cpu_notifier(&hrtimers_nb);
1671 * schedule_hrtimeout_range_clock - sleep until timeout
1672 * @expires: timeout value (ktime_t)
1673 * @delta: slack in expires timeout (ktime_t)
1674 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1675 * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME
1678 schedule_hrtimeout_range_clock(ktime_t *expires, unsigned long delta,
1679 const enum hrtimer_mode mode, int clock)
1681 struct hrtimer_sleeper t;
1684 * Optimize when a zero timeout value is given. It does not
1685 * matter whether this is an absolute or a relative time.
1687 if (expires && !expires->tv64) {
1688 __set_current_state(TASK_RUNNING);
1693 * A NULL parameter means "infinite"
1700 hrtimer_init_on_stack(&t.timer, clock, mode);
1701 hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
1703 hrtimer_init_sleeper(&t, current);
1705 hrtimer_start_expires(&t.timer, mode);
1710 hrtimer_cancel(&t.timer);
1711 destroy_hrtimer_on_stack(&t.timer);
1713 __set_current_state(TASK_RUNNING);
1715 return !t.task ? 0 : -EINTR;
1719 * schedule_hrtimeout_range - sleep until timeout
1720 * @expires: timeout value (ktime_t)
1721 * @delta: slack in expires timeout (ktime_t)
1722 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1724 * Make the current task sleep until the given expiry time has
1725 * elapsed. The routine will return immediately unless
1726 * the current task state has been set (see set_current_state()).
1728 * The @delta argument gives the kernel the freedom to schedule the
1729 * actual wakeup to a time that is both power and performance friendly.
1730 * The kernel give the normal best effort behavior for "@expires+@delta",
1731 * but may decide to fire the timer earlier, but no earlier than @expires.
1733 * You can set the task state as follows -
1735 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1736 * pass before the routine returns.
1738 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1739 * delivered to the current task.
1741 * The current task state is guaranteed to be TASK_RUNNING when this
1744 * Returns 0 when the timer has expired otherwise -EINTR
1746 int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta,
1747 const enum hrtimer_mode mode)
1749 return schedule_hrtimeout_range_clock(expires, delta, mode,
1752 EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
1755 * schedule_hrtimeout - sleep until timeout
1756 * @expires: timeout value (ktime_t)
1757 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1759 * Make the current task sleep until the given expiry time has
1760 * elapsed. The routine will return immediately unless
1761 * the current task state has been set (see set_current_state()).
1763 * You can set the task state as follows -
1765 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1766 * pass before the routine returns.
1768 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1769 * delivered to the current task.
1771 * The current task state is guaranteed to be TASK_RUNNING when this
1774 * Returns 0 when the timer has expired otherwise -EINTR
1776 int __sched schedule_hrtimeout(ktime_t *expires,
1777 const enum hrtimer_mode mode)
1779 return schedule_hrtimeout_range(expires, 0, mode);
1781 EXPORT_SYMBOL_GPL(schedule_hrtimeout);