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/signal.h>
47 #include <linux/sched/sysctl.h>
48 #include <linux/sched/rt.h>
49 #include <linux/sched/deadline.h>
50 #include <linux/sched/nohz.h>
51 #include <linux/timer.h>
52 #include <linux/freezer.h>
54 #include <linux/uaccess.h>
56 #include <trace/events/timer.h>
58 #include "tick-internal.h"
63 * There are more clockids than hrtimer bases. Thus, we index
64 * into the timer bases by the hrtimer_base_type enum. When trying
65 * to reach a base using a clockid, hrtimer_clockid_to_base()
66 * is used to convert from clockid to the proper hrtimer_base_type.
68 DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
70 .lock = __RAW_SPIN_LOCK_UNLOCKED(hrtimer_bases.lock),
71 .seq = SEQCNT_ZERO(hrtimer_bases.seq),
75 .index = HRTIMER_BASE_MONOTONIC,
76 .clockid = CLOCK_MONOTONIC,
77 .get_time = &ktime_get,
80 .index = HRTIMER_BASE_REALTIME,
81 .clockid = CLOCK_REALTIME,
82 .get_time = &ktime_get_real,
85 .index = HRTIMER_BASE_BOOTTIME,
86 .clockid = CLOCK_BOOTTIME,
87 .get_time = &ktime_get_boottime,
90 .index = HRTIMER_BASE_TAI,
92 .get_time = &ktime_get_clocktai,
97 static const int hrtimer_clock_to_base_table[MAX_CLOCKS] = {
98 /* Make sure we catch unsupported clockids */
99 [0 ... MAX_CLOCKS - 1] = HRTIMER_MAX_CLOCK_BASES,
101 [CLOCK_REALTIME] = HRTIMER_BASE_REALTIME,
102 [CLOCK_MONOTONIC] = HRTIMER_BASE_MONOTONIC,
103 [CLOCK_BOOTTIME] = HRTIMER_BASE_BOOTTIME,
104 [CLOCK_TAI] = HRTIMER_BASE_TAI,
108 * Functions and macros which are different for UP/SMP systems are kept in a
114 * We require the migration_base for lock_hrtimer_base()/switch_hrtimer_base()
115 * such that hrtimer_callback_running() can unconditionally dereference
116 * timer->base->cpu_base
118 static struct hrtimer_cpu_base migration_cpu_base = {
119 .seq = SEQCNT_ZERO(migration_cpu_base),
120 .clock_base = { { .cpu_base = &migration_cpu_base, }, },
123 #define migration_base migration_cpu_base.clock_base[0]
126 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
127 * means that all timers which are tied to this base via timer->base are
128 * locked, and the base itself is locked too.
130 * So __run_timers/migrate_timers can safely modify all timers which could
131 * be found on the lists/queues.
133 * When the timer's base is locked, and the timer removed from list, it is
134 * possible to set timer->base = &migration_base and drop the lock: the timer
138 struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
139 unsigned long *flags)
141 struct hrtimer_clock_base *base;
145 if (likely(base != &migration_base)) {
146 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
147 if (likely(base == timer->base))
149 /* The timer has migrated to another CPU: */
150 raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
157 * With HIGHRES=y we do not migrate the timer when it is expiring
158 * before the next event on the target cpu because we cannot reprogram
159 * the target cpu hardware and we would cause it to fire late.
161 * Called with cpu_base->lock of target cpu held.
164 hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base)
166 #ifdef CONFIG_HIGH_RES_TIMERS
169 if (!new_base->cpu_base->hres_active)
172 expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset);
173 return expires <= new_base->cpu_base->expires_next;
179 #ifdef CONFIG_NO_HZ_COMMON
181 struct hrtimer_cpu_base *get_target_base(struct hrtimer_cpu_base *base,
184 if (pinned || !base->migration_enabled)
186 return &per_cpu(hrtimer_bases, get_nohz_timer_target());
190 struct hrtimer_cpu_base *get_target_base(struct hrtimer_cpu_base *base,
198 * We switch the timer base to a power-optimized selected CPU target,
200 * - NO_HZ_COMMON is enabled
201 * - timer migration is enabled
202 * - the timer callback is not running
203 * - the timer is not the first expiring timer on the new target
205 * If one of the above requirements is not fulfilled we move the timer
206 * to the current CPU or leave it on the previously assigned CPU if
207 * the timer callback is currently running.
209 static inline struct hrtimer_clock_base *
210 switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
213 struct hrtimer_cpu_base *new_cpu_base, *this_cpu_base;
214 struct hrtimer_clock_base *new_base;
215 int basenum = base->index;
217 this_cpu_base = this_cpu_ptr(&hrtimer_bases);
218 new_cpu_base = get_target_base(this_cpu_base, pinned);
220 new_base = &new_cpu_base->clock_base[basenum];
222 if (base != new_base) {
224 * We are trying to move timer to new_base.
225 * However we can't change timer's base while it is running,
226 * so we keep it on the same CPU. No hassle vs. reprogramming
227 * the event source in the high resolution case. The softirq
228 * code will take care of this when the timer function has
229 * completed. There is no conflict as we hold the lock until
230 * the timer is enqueued.
232 if (unlikely(hrtimer_callback_running(timer)))
235 /* See the comment in lock_hrtimer_base() */
236 timer->base = &migration_base;
237 raw_spin_unlock(&base->cpu_base->lock);
238 raw_spin_lock(&new_base->cpu_base->lock);
240 if (new_cpu_base != this_cpu_base &&
241 hrtimer_check_target(timer, new_base)) {
242 raw_spin_unlock(&new_base->cpu_base->lock);
243 raw_spin_lock(&base->cpu_base->lock);
244 new_cpu_base = this_cpu_base;
248 timer->base = new_base;
250 if (new_cpu_base != this_cpu_base &&
251 hrtimer_check_target(timer, new_base)) {
252 new_cpu_base = this_cpu_base;
259 #else /* CONFIG_SMP */
261 static inline struct hrtimer_clock_base *
262 lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
264 struct hrtimer_clock_base *base = timer->base;
266 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
271 # define switch_hrtimer_base(t, b, p) (b)
273 #endif /* !CONFIG_SMP */
276 * Functions for the union type storage format of ktime_t which are
277 * too large for inlining:
279 #if BITS_PER_LONG < 64
281 * Divide a ktime value by a nanosecond value
283 s64 __ktime_divns(const ktime_t kt, s64 div)
289 dclc = ktime_to_ns(kt);
290 tmp = dclc < 0 ? -dclc : dclc;
292 /* Make sure the divisor is less than 2^32: */
298 do_div(tmp, (unsigned long) div);
299 return dclc < 0 ? -tmp : tmp;
301 EXPORT_SYMBOL_GPL(__ktime_divns);
302 #endif /* BITS_PER_LONG >= 64 */
305 * Add two ktime values and do a safety check for overflow:
307 ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
309 ktime_t res = ktime_add_unsafe(lhs, rhs);
312 * We use KTIME_SEC_MAX here, the maximum timeout which we can
313 * return to user space in a timespec:
315 if (res < 0 || res < lhs || res < rhs)
316 res = ktime_set(KTIME_SEC_MAX, 0);
321 EXPORT_SYMBOL_GPL(ktime_add_safe);
323 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
325 static struct debug_obj_descr hrtimer_debug_descr;
327 static void *hrtimer_debug_hint(void *addr)
329 return ((struct hrtimer *) addr)->function;
333 * fixup_init is called when:
334 * - an active object is initialized
336 static bool hrtimer_fixup_init(void *addr, enum debug_obj_state state)
338 struct hrtimer *timer = addr;
341 case ODEBUG_STATE_ACTIVE:
342 hrtimer_cancel(timer);
343 debug_object_init(timer, &hrtimer_debug_descr);
351 * fixup_activate is called when:
352 * - an active object is activated
353 * - an unknown non-static object is activated
355 static bool hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
358 case ODEBUG_STATE_ACTIVE:
367 * fixup_free is called when:
368 * - an active object is freed
370 static bool hrtimer_fixup_free(void *addr, enum debug_obj_state state)
372 struct hrtimer *timer = addr;
375 case ODEBUG_STATE_ACTIVE:
376 hrtimer_cancel(timer);
377 debug_object_free(timer, &hrtimer_debug_descr);
384 static struct debug_obj_descr hrtimer_debug_descr = {
386 .debug_hint = hrtimer_debug_hint,
387 .fixup_init = hrtimer_fixup_init,
388 .fixup_activate = hrtimer_fixup_activate,
389 .fixup_free = hrtimer_fixup_free,
392 static inline void debug_hrtimer_init(struct hrtimer *timer)
394 debug_object_init(timer, &hrtimer_debug_descr);
397 static inline void debug_hrtimer_activate(struct hrtimer *timer)
399 debug_object_activate(timer, &hrtimer_debug_descr);
402 static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
404 debug_object_deactivate(timer, &hrtimer_debug_descr);
407 static inline void debug_hrtimer_free(struct hrtimer *timer)
409 debug_object_free(timer, &hrtimer_debug_descr);
412 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
413 enum hrtimer_mode mode);
415 void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
416 enum hrtimer_mode mode)
418 debug_object_init_on_stack(timer, &hrtimer_debug_descr);
419 __hrtimer_init(timer, clock_id, mode);
421 EXPORT_SYMBOL_GPL(hrtimer_init_on_stack);
423 void destroy_hrtimer_on_stack(struct hrtimer *timer)
425 debug_object_free(timer, &hrtimer_debug_descr);
427 EXPORT_SYMBOL_GPL(destroy_hrtimer_on_stack);
430 static inline void debug_hrtimer_init(struct hrtimer *timer) { }
431 static inline void debug_hrtimer_activate(struct hrtimer *timer) { }
432 static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
436 debug_init(struct hrtimer *timer, clockid_t clockid,
437 enum hrtimer_mode mode)
439 debug_hrtimer_init(timer);
440 trace_hrtimer_init(timer, clockid, mode);
443 static inline void debug_activate(struct hrtimer *timer)
445 debug_hrtimer_activate(timer);
446 trace_hrtimer_start(timer);
449 static inline void debug_deactivate(struct hrtimer *timer)
451 debug_hrtimer_deactivate(timer);
452 trace_hrtimer_cancel(timer);
455 #if defined(CONFIG_NO_HZ_COMMON) || defined(CONFIG_HIGH_RES_TIMERS)
456 static inline void hrtimer_update_next_timer(struct hrtimer_cpu_base *cpu_base,
457 struct hrtimer *timer)
459 #ifdef CONFIG_HIGH_RES_TIMERS
460 cpu_base->next_timer = timer;
464 static ktime_t __hrtimer_get_next_event(struct hrtimer_cpu_base *cpu_base)
466 struct hrtimer_clock_base *base = cpu_base->clock_base;
467 unsigned int active = cpu_base->active_bases;
468 ktime_t expires, expires_next = KTIME_MAX;
470 hrtimer_update_next_timer(cpu_base, NULL);
471 for (; active; base++, active >>= 1) {
472 struct timerqueue_node *next;
473 struct hrtimer *timer;
475 if (!(active & 0x01))
478 next = timerqueue_getnext(&base->active);
479 timer = container_of(next, struct hrtimer, node);
480 expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
481 if (expires < expires_next) {
482 expires_next = expires;
483 hrtimer_update_next_timer(cpu_base, timer);
487 * clock_was_set() might have changed base->offset of any of
488 * the clock bases so the result might be negative. Fix it up
489 * to prevent a false positive in clockevents_program_event().
491 if (expires_next < 0)
497 static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base)
499 ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset;
500 ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset;
501 ktime_t *offs_tai = &base->clock_base[HRTIMER_BASE_TAI].offset;
503 return ktime_get_update_offsets_now(&base->clock_was_set_seq,
504 offs_real, offs_boot, offs_tai);
507 /* High resolution timer related functions */
508 #ifdef CONFIG_HIGH_RES_TIMERS
511 * High resolution timer enabled ?
513 static bool hrtimer_hres_enabled __read_mostly = true;
514 unsigned int hrtimer_resolution __read_mostly = LOW_RES_NSEC;
515 EXPORT_SYMBOL_GPL(hrtimer_resolution);
518 * Enable / Disable high resolution mode
520 static int __init setup_hrtimer_hres(char *str)
522 return (kstrtobool(str, &hrtimer_hres_enabled) == 0);
525 __setup("highres=", setup_hrtimer_hres);
528 * hrtimer_high_res_enabled - query, if the highres mode is enabled
530 static inline int hrtimer_is_hres_enabled(void)
532 return hrtimer_hres_enabled;
536 * Is the high resolution mode active ?
538 static inline int __hrtimer_hres_active(struct hrtimer_cpu_base *cpu_base)
540 return cpu_base->hres_active;
543 static inline int hrtimer_hres_active(void)
545 return __hrtimer_hres_active(this_cpu_ptr(&hrtimer_bases));
549 * Reprogram the event source with checking both queues for the
551 * Called with interrupts disabled and base->lock held
554 hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
556 ktime_t expires_next;
558 if (!cpu_base->hres_active)
561 expires_next = __hrtimer_get_next_event(cpu_base);
563 if (skip_equal && expires_next == cpu_base->expires_next)
566 cpu_base->expires_next = expires_next;
569 * If a hang was detected in the last timer interrupt then we
570 * leave the hang delay active in the hardware. We want the
571 * system to make progress. That also prevents the following
573 * T1 expires 50ms from now
574 * T2 expires 5s from now
576 * T1 is removed, so this code is called and would reprogram
577 * the hardware to 5s from now. Any hrtimer_start after that
578 * will not reprogram the hardware due to hang_detected being
579 * set. So we'd effectivly block all timers until the T2 event
582 if (cpu_base->hang_detected)
585 tick_program_event(cpu_base->expires_next, 1);
589 * When a timer is enqueued and expires earlier than the already enqueued
590 * timers, we have to check, whether it expires earlier than the timer for
591 * which the clock event device was armed.
593 * Called with interrupts disabled and base->cpu_base.lock held
595 static void hrtimer_reprogram(struct hrtimer *timer,
596 struct hrtimer_clock_base *base)
598 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
599 ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
601 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
604 * If the timer is not on the current cpu, we cannot reprogram
605 * the other cpus clock event device.
607 if (base->cpu_base != cpu_base)
611 * If the hrtimer interrupt is running, then it will
612 * reevaluate the clock bases and reprogram the clock event
613 * device. The callbacks are always executed in hard interrupt
614 * context so we don't need an extra check for a running
617 if (cpu_base->in_hrtirq)
621 * CLOCK_REALTIME timer might be requested with an absolute
622 * expiry time which is less than base->offset. Set it to 0.
627 if (expires >= cpu_base->expires_next)
630 /* Update the pointer to the next expiring timer */
631 cpu_base->next_timer = timer;
634 * If a hang was detected in the last timer interrupt then we
635 * do not schedule a timer which is earlier than the expiry
636 * which we enforced in the hang detection. We want the system
639 if (cpu_base->hang_detected)
643 * Program the timer hardware. We enforce the expiry for
644 * events which are already in the past.
646 cpu_base->expires_next = expires;
647 tick_program_event(expires, 1);
651 * Initialize the high resolution related parts of cpu_base
653 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
655 base->expires_next = KTIME_MAX;
656 base->hres_active = 0;
660 * Retrigger next event is called after clock was set
662 * Called with interrupts disabled via on_each_cpu()
664 static void retrigger_next_event(void *arg)
666 struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
668 if (!base->hres_active)
671 raw_spin_lock(&base->lock);
672 hrtimer_update_base(base);
673 hrtimer_force_reprogram(base, 0);
674 raw_spin_unlock(&base->lock);
678 * Switch to high resolution mode
680 static void hrtimer_switch_to_hres(void)
682 struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
684 if (tick_init_highres()) {
685 printk(KERN_WARNING "Could not switch to high resolution "
686 "mode on CPU %d\n", base->cpu);
689 base->hres_active = 1;
690 hrtimer_resolution = HIGH_RES_NSEC;
692 tick_setup_sched_timer();
693 /* "Retrigger" the interrupt to get things going */
694 retrigger_next_event(NULL);
697 static void clock_was_set_work(struct work_struct *work)
702 static DECLARE_WORK(hrtimer_work, clock_was_set_work);
705 * Called from timekeeping and resume code to reprogram the hrtimer
706 * interrupt device on all cpus.
708 void clock_was_set_delayed(void)
710 schedule_work(&hrtimer_work);
715 static inline int __hrtimer_hres_active(struct hrtimer_cpu_base *b) { return 0; }
716 static inline int hrtimer_hres_active(void) { return 0; }
717 static inline int hrtimer_is_hres_enabled(void) { return 0; }
718 static inline void hrtimer_switch_to_hres(void) { }
720 hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { }
721 static inline int hrtimer_reprogram(struct hrtimer *timer,
722 struct hrtimer_clock_base *base)
726 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { }
727 static inline void retrigger_next_event(void *arg) { }
729 #endif /* CONFIG_HIGH_RES_TIMERS */
732 * Clock realtime was set
734 * Change the offset of the realtime clock vs. the monotonic
737 * We might have to reprogram the high resolution timer interrupt. On
738 * SMP we call the architecture specific code to retrigger _all_ high
739 * resolution timer interrupts. On UP we just disable interrupts and
740 * call the high resolution interrupt code.
742 void clock_was_set(void)
744 #ifdef CONFIG_HIGH_RES_TIMERS
745 /* Retrigger the CPU local events everywhere */
746 on_each_cpu(retrigger_next_event, NULL, 1);
748 timerfd_clock_was_set();
752 * During resume we might have to reprogram the high resolution timer
753 * interrupt on all online CPUs. However, all other CPUs will be
754 * stopped with IRQs interrupts disabled so the clock_was_set() call
757 void hrtimers_resume(void)
759 WARN_ONCE(!irqs_disabled(),
760 KERN_INFO "hrtimers_resume() called with IRQs enabled!");
762 /* Retrigger on the local CPU */
763 retrigger_next_event(NULL);
764 /* And schedule a retrigger for all others */
765 clock_was_set_delayed();
769 * Counterpart to lock_hrtimer_base above:
772 void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
774 raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
778 * hrtimer_forward - forward the timer expiry
779 * @timer: hrtimer to forward
780 * @now: forward past this time
781 * @interval: the interval to forward
783 * Forward the timer expiry so it will expire in the future.
784 * Returns the number of overruns.
786 * Can be safely called from the callback function of @timer. If
787 * called from other contexts @timer must neither be enqueued nor
788 * running the callback and the caller needs to take care of
791 * Note: This only updates the timer expiry value and does not requeue
794 u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
799 delta = ktime_sub(now, hrtimer_get_expires(timer));
804 if (WARN_ON(timer->state & HRTIMER_STATE_ENQUEUED))
807 if (interval < hrtimer_resolution)
808 interval = hrtimer_resolution;
810 if (unlikely(delta >= interval)) {
811 s64 incr = ktime_to_ns(interval);
813 orun = ktime_divns(delta, incr);
814 hrtimer_add_expires_ns(timer, incr * orun);
815 if (hrtimer_get_expires_tv64(timer) > now)
818 * This (and the ktime_add() below) is the
819 * correction for exact:
823 hrtimer_add_expires(timer, interval);
827 EXPORT_SYMBOL_GPL(hrtimer_forward);
830 * enqueue_hrtimer - internal function to (re)start a timer
832 * The timer is inserted in expiry order. Insertion into the
833 * red black tree is O(log(n)). Must hold the base lock.
835 * Returns 1 when the new timer is the leftmost timer in the tree.
837 static int enqueue_hrtimer(struct hrtimer *timer,
838 struct hrtimer_clock_base *base)
840 debug_activate(timer);
842 base->cpu_base->active_bases |= 1 << base->index;
844 timer->state = HRTIMER_STATE_ENQUEUED;
846 return timerqueue_add(&base->active, &timer->node);
850 * __remove_hrtimer - internal function to remove a timer
852 * Caller must hold the base lock.
854 * High resolution timer mode reprograms the clock event device when the
855 * timer is the one which expires next. The caller can disable this by setting
856 * reprogram to zero. This is useful, when the context does a reprogramming
857 * anyway (e.g. timer interrupt)
859 static void __remove_hrtimer(struct hrtimer *timer,
860 struct hrtimer_clock_base *base,
861 u8 newstate, int reprogram)
863 struct hrtimer_cpu_base *cpu_base = base->cpu_base;
864 u8 state = timer->state;
866 timer->state = newstate;
867 if (!(state & HRTIMER_STATE_ENQUEUED))
870 if (!timerqueue_del(&base->active, &timer->node))
871 cpu_base->active_bases &= ~(1 << base->index);
873 #ifdef CONFIG_HIGH_RES_TIMERS
875 * Note: If reprogram is false we do not update
876 * cpu_base->next_timer. This happens when we remove the first
877 * timer on a remote cpu. No harm as we never dereference
878 * cpu_base->next_timer. So the worst thing what can happen is
879 * an superflous call to hrtimer_force_reprogram() on the
880 * remote cpu later on if the same timer gets enqueued again.
882 if (reprogram && timer == cpu_base->next_timer)
883 hrtimer_force_reprogram(cpu_base, 1);
888 * remove hrtimer, called with base lock held
891 remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base, bool restart)
893 if (hrtimer_is_queued(timer)) {
894 u8 state = timer->state;
898 * Remove the timer and force reprogramming when high
899 * resolution mode is active and the timer is on the current
900 * CPU. If we remove a timer on another CPU, reprogramming is
901 * skipped. The interrupt event on this CPU is fired and
902 * reprogramming happens in the interrupt handler. This is a
903 * rare case and less expensive than a smp call.
905 debug_deactivate(timer);
906 reprogram = base->cpu_base == this_cpu_ptr(&hrtimer_bases);
909 state = HRTIMER_STATE_INACTIVE;
911 __remove_hrtimer(timer, base, state, reprogram);
917 static inline ktime_t hrtimer_update_lowres(struct hrtimer *timer, ktime_t tim,
918 const enum hrtimer_mode mode)
920 #ifdef CONFIG_TIME_LOW_RES
922 * CONFIG_TIME_LOW_RES indicates that the system has no way to return
923 * granular time values. For relative timers we add hrtimer_resolution
924 * (i.e. one jiffie) to prevent short timeouts.
926 timer->is_rel = mode & HRTIMER_MODE_REL;
928 tim = ktime_add_safe(tim, hrtimer_resolution);
934 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
935 * @timer: the timer to be added
937 * @delta_ns: "slack" range for the timer
938 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
939 * relative (HRTIMER_MODE_REL)
941 void hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
942 u64 delta_ns, const enum hrtimer_mode mode)
944 struct hrtimer_clock_base *base, *new_base;
948 base = lock_hrtimer_base(timer, &flags);
950 /* Remove an active timer from the queue: */
951 remove_hrtimer(timer, base, true);
953 if (mode & HRTIMER_MODE_REL)
954 tim = ktime_add_safe(tim, base->get_time());
956 tim = hrtimer_update_lowres(timer, tim, mode);
958 hrtimer_set_expires_range_ns(timer, tim, delta_ns);
960 /* Switch the timer base, if necessary: */
961 new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED);
963 leftmost = enqueue_hrtimer(timer, new_base);
967 if (!hrtimer_is_hres_active(timer)) {
969 * Kick to reschedule the next tick to handle the new timer
970 * on dynticks target.
972 if (new_base->cpu_base->nohz_active)
973 wake_up_nohz_cpu(new_base->cpu_base->cpu);
975 hrtimer_reprogram(timer, new_base);
978 unlock_hrtimer_base(timer, &flags);
980 EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
983 * hrtimer_try_to_cancel - try to deactivate a timer
984 * @timer: hrtimer to stop
987 * 0 when the timer was not active
988 * 1 when the timer was active
989 * -1 when the timer is currently excuting the callback function and
992 int hrtimer_try_to_cancel(struct hrtimer *timer)
994 struct hrtimer_clock_base *base;
999 * Check lockless first. If the timer is not active (neither
1000 * enqueued nor running the callback, nothing to do here. The
1001 * base lock does not serialize against a concurrent enqueue,
1002 * so we can avoid taking it.
1004 if (!hrtimer_active(timer))
1007 base = lock_hrtimer_base(timer, &flags);
1009 if (!hrtimer_callback_running(timer))
1010 ret = remove_hrtimer(timer, base, false);
1012 unlock_hrtimer_base(timer, &flags);
1017 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1020 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1021 * @timer: the timer to be cancelled
1024 * 0 when the timer was not active
1025 * 1 when the timer was active
1027 int hrtimer_cancel(struct hrtimer *timer)
1030 int ret = hrtimer_try_to_cancel(timer);
1037 EXPORT_SYMBOL_GPL(hrtimer_cancel);
1040 * hrtimer_get_remaining - get remaining time for the timer
1041 * @timer: the timer to read
1042 * @adjust: adjust relative timers when CONFIG_TIME_LOW_RES=y
1044 ktime_t __hrtimer_get_remaining(const struct hrtimer *timer, bool adjust)
1046 unsigned long flags;
1049 lock_hrtimer_base(timer, &flags);
1050 if (IS_ENABLED(CONFIG_TIME_LOW_RES) && adjust)
1051 rem = hrtimer_expires_remaining_adjusted(timer);
1053 rem = hrtimer_expires_remaining(timer);
1054 unlock_hrtimer_base(timer, &flags);
1058 EXPORT_SYMBOL_GPL(__hrtimer_get_remaining);
1060 #ifdef CONFIG_NO_HZ_COMMON
1062 * hrtimer_get_next_event - get the time until next expiry event
1064 * Returns the next expiry time or KTIME_MAX if no timer is pending.
1066 u64 hrtimer_get_next_event(void)
1068 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1069 u64 expires = KTIME_MAX;
1070 unsigned long flags;
1072 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1074 if (!__hrtimer_hres_active(cpu_base))
1075 expires = __hrtimer_get_next_event(cpu_base);
1077 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1083 static inline int hrtimer_clockid_to_base(clockid_t clock_id)
1085 if (likely(clock_id < MAX_CLOCKS)) {
1086 int base = hrtimer_clock_to_base_table[clock_id];
1088 if (likely(base != HRTIMER_MAX_CLOCK_BASES))
1091 WARN(1, "Invalid clockid %d. Using MONOTONIC\n", clock_id);
1092 return HRTIMER_BASE_MONOTONIC;
1095 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1096 enum hrtimer_mode mode)
1098 struct hrtimer_cpu_base *cpu_base;
1101 memset(timer, 0, sizeof(struct hrtimer));
1103 cpu_base = raw_cpu_ptr(&hrtimer_bases);
1105 if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
1106 clock_id = CLOCK_MONOTONIC;
1108 base = hrtimer_clockid_to_base(clock_id);
1109 timer->base = &cpu_base->clock_base[base];
1110 timerqueue_init(&timer->node);
1114 * hrtimer_init - initialize a timer to the given clock
1115 * @timer: the timer to be initialized
1116 * @clock_id: the clock to be used
1117 * @mode: timer mode abs/rel
1119 void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1120 enum hrtimer_mode mode)
1122 debug_init(timer, clock_id, mode);
1123 __hrtimer_init(timer, clock_id, mode);
1125 EXPORT_SYMBOL_GPL(hrtimer_init);
1128 * A timer is active, when it is enqueued into the rbtree or the
1129 * callback function is running or it's in the state of being migrated
1132 * It is important for this function to not return a false negative.
1134 bool hrtimer_active(const struct hrtimer *timer)
1136 struct hrtimer_cpu_base *cpu_base;
1140 cpu_base = READ_ONCE(timer->base->cpu_base);
1141 seq = raw_read_seqcount_begin(&cpu_base->seq);
1143 if (timer->state != HRTIMER_STATE_INACTIVE ||
1144 cpu_base->running == timer)
1147 } while (read_seqcount_retry(&cpu_base->seq, seq) ||
1148 cpu_base != READ_ONCE(timer->base->cpu_base));
1152 EXPORT_SYMBOL_GPL(hrtimer_active);
1155 * The write_seqcount_barrier()s in __run_hrtimer() split the thing into 3
1156 * distinct sections:
1158 * - queued: the timer is queued
1159 * - callback: the timer is being ran
1160 * - post: the timer is inactive or (re)queued
1162 * On the read side we ensure we observe timer->state and cpu_base->running
1163 * from the same section, if anything changed while we looked at it, we retry.
1164 * This includes timer->base changing because sequence numbers alone are
1165 * insufficient for that.
1167 * The sequence numbers are required because otherwise we could still observe
1168 * a false negative if the read side got smeared over multiple consequtive
1169 * __run_hrtimer() invocations.
1172 static void __run_hrtimer(struct hrtimer_cpu_base *cpu_base,
1173 struct hrtimer_clock_base *base,
1174 struct hrtimer *timer, ktime_t *now)
1176 enum hrtimer_restart (*fn)(struct hrtimer *);
1179 lockdep_assert_held(&cpu_base->lock);
1181 debug_deactivate(timer);
1182 cpu_base->running = timer;
1185 * Separate the ->running assignment from the ->state assignment.
1187 * As with a regular write barrier, this ensures the read side in
1188 * hrtimer_active() cannot observe cpu_base->running == NULL &&
1189 * timer->state == INACTIVE.
1191 raw_write_seqcount_barrier(&cpu_base->seq);
1193 __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE, 0);
1194 fn = timer->function;
1197 * Clear the 'is relative' flag for the TIME_LOW_RES case. If the
1198 * timer is restarted with a period then it becomes an absolute
1199 * timer. If its not restarted it does not matter.
1201 if (IS_ENABLED(CONFIG_TIME_LOW_RES))
1202 timer->is_rel = false;
1205 * Because we run timers from hardirq context, there is no chance
1206 * they get migrated to another cpu, therefore its safe to unlock
1209 raw_spin_unlock(&cpu_base->lock);
1210 trace_hrtimer_expire_entry(timer, now);
1211 restart = fn(timer);
1212 trace_hrtimer_expire_exit(timer);
1213 raw_spin_lock(&cpu_base->lock);
1216 * Note: We clear the running state after enqueue_hrtimer and
1217 * we do not reprogram the event hardware. Happens either in
1218 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1220 * Note: Because we dropped the cpu_base->lock above,
1221 * hrtimer_start_range_ns() can have popped in and enqueued the timer
1224 if (restart != HRTIMER_NORESTART &&
1225 !(timer->state & HRTIMER_STATE_ENQUEUED))
1226 enqueue_hrtimer(timer, base);
1229 * Separate the ->running assignment from the ->state assignment.
1231 * As with a regular write barrier, this ensures the read side in
1232 * hrtimer_active() cannot observe cpu_base->running == NULL &&
1233 * timer->state == INACTIVE.
1235 raw_write_seqcount_barrier(&cpu_base->seq);
1237 WARN_ON_ONCE(cpu_base->running != timer);
1238 cpu_base->running = NULL;
1241 static void __hrtimer_run_queues(struct hrtimer_cpu_base *cpu_base, ktime_t now)
1243 struct hrtimer_clock_base *base = cpu_base->clock_base;
1244 unsigned int active = cpu_base->active_bases;
1246 for (; active; base++, active >>= 1) {
1247 struct timerqueue_node *node;
1250 if (!(active & 0x01))
1253 basenow = ktime_add(now, base->offset);
1255 while ((node = timerqueue_getnext(&base->active))) {
1256 struct hrtimer *timer;
1258 timer = container_of(node, struct hrtimer, node);
1261 * The immediate goal for using the softexpires is
1262 * minimizing wakeups, not running timers at the
1263 * earliest interrupt after their soft expiration.
1264 * This allows us to avoid using a Priority Search
1265 * Tree, which can answer a stabbing querry for
1266 * overlapping intervals and instead use the simple
1267 * BST we already have.
1268 * We don't add extra wakeups by delaying timers that
1269 * are right-of a not yet expired timer, because that
1270 * timer will have to trigger a wakeup anyway.
1272 if (basenow < hrtimer_get_softexpires_tv64(timer))
1275 __run_hrtimer(cpu_base, base, timer, &basenow);
1280 #ifdef CONFIG_HIGH_RES_TIMERS
1283 * High resolution timer interrupt
1284 * Called with interrupts disabled
1286 void hrtimer_interrupt(struct clock_event_device *dev)
1288 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1289 ktime_t expires_next, now, entry_time, delta;
1292 BUG_ON(!cpu_base->hres_active);
1293 cpu_base->nr_events++;
1294 dev->next_event = KTIME_MAX;
1296 raw_spin_lock(&cpu_base->lock);
1297 entry_time = now = hrtimer_update_base(cpu_base);
1299 cpu_base->in_hrtirq = 1;
1301 * We set expires_next to KTIME_MAX here with cpu_base->lock
1302 * held to prevent that a timer is enqueued in our queue via
1303 * the migration code. This does not affect enqueueing of
1304 * timers which run their callback and need to be requeued on
1307 cpu_base->expires_next = KTIME_MAX;
1309 __hrtimer_run_queues(cpu_base, now);
1311 /* Reevaluate the clock bases for the next expiry */
1312 expires_next = __hrtimer_get_next_event(cpu_base);
1314 * Store the new expiry value so the migration code can verify
1317 cpu_base->expires_next = expires_next;
1318 cpu_base->in_hrtirq = 0;
1319 raw_spin_unlock(&cpu_base->lock);
1321 /* Reprogramming necessary ? */
1322 if (!tick_program_event(expires_next, 0)) {
1323 cpu_base->hang_detected = 0;
1328 * The next timer was already expired due to:
1330 * - long lasting callbacks
1331 * - being scheduled away when running in a VM
1333 * We need to prevent that we loop forever in the hrtimer
1334 * interrupt routine. We give it 3 attempts to avoid
1335 * overreacting on some spurious event.
1337 * Acquire base lock for updating the offsets and retrieving
1340 raw_spin_lock(&cpu_base->lock);
1341 now = hrtimer_update_base(cpu_base);
1342 cpu_base->nr_retries++;
1346 * Give the system a chance to do something else than looping
1347 * here. We stored the entry time, so we know exactly how long
1348 * we spent here. We schedule the next event this amount of
1351 cpu_base->nr_hangs++;
1352 cpu_base->hang_detected = 1;
1353 raw_spin_unlock(&cpu_base->lock);
1354 delta = ktime_sub(now, entry_time);
1355 if ((unsigned int)delta > cpu_base->max_hang_time)
1356 cpu_base->max_hang_time = (unsigned int) delta;
1358 * Limit it to a sensible value as we enforce a longer
1359 * delay. Give the CPU at least 100ms to catch up.
1361 if (delta > 100 * NSEC_PER_MSEC)
1362 expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1364 expires_next = ktime_add(now, delta);
1365 tick_program_event(expires_next, 1);
1366 printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n",
1367 ktime_to_ns(delta));
1371 * local version of hrtimer_peek_ahead_timers() called with interrupts
1374 static inline void __hrtimer_peek_ahead_timers(void)
1376 struct tick_device *td;
1378 if (!hrtimer_hres_active())
1381 td = this_cpu_ptr(&tick_cpu_device);
1382 if (td && td->evtdev)
1383 hrtimer_interrupt(td->evtdev);
1386 #else /* CONFIG_HIGH_RES_TIMERS */
1388 static inline void __hrtimer_peek_ahead_timers(void) { }
1390 #endif /* !CONFIG_HIGH_RES_TIMERS */
1393 * Called from run_local_timers in hardirq context every jiffy
1395 void hrtimer_run_queues(void)
1397 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1400 if (__hrtimer_hres_active(cpu_base))
1404 * This _is_ ugly: We have to check periodically, whether we
1405 * can switch to highres and / or nohz mode. The clocksource
1406 * switch happens with xtime_lock held. Notification from
1407 * there only sets the check bit in the tick_oneshot code,
1408 * otherwise we might deadlock vs. xtime_lock.
1410 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) {
1411 hrtimer_switch_to_hres();
1415 raw_spin_lock(&cpu_base->lock);
1416 now = hrtimer_update_base(cpu_base);
1417 __hrtimer_run_queues(cpu_base, now);
1418 raw_spin_unlock(&cpu_base->lock);
1422 * Sleep related functions:
1424 static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1426 struct hrtimer_sleeper *t =
1427 container_of(timer, struct hrtimer_sleeper, timer);
1428 struct task_struct *task = t->task;
1432 wake_up_process(task);
1434 return HRTIMER_NORESTART;
1437 void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1439 sl->timer.function = hrtimer_wakeup;
1442 EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
1444 static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1446 hrtimer_init_sleeper(t, current);
1449 set_current_state(TASK_INTERRUPTIBLE);
1450 hrtimer_start_expires(&t->timer, mode);
1452 if (likely(t->task))
1453 freezable_schedule();
1455 hrtimer_cancel(&t->timer);
1456 mode = HRTIMER_MODE_ABS;
1458 } while (t->task && !signal_pending(current));
1460 __set_current_state(TASK_RUNNING);
1462 return t->task == NULL;
1465 static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp)
1467 struct timespec rmt;
1470 rem = hrtimer_expires_remaining(timer);
1473 rmt = ktime_to_timespec(rem);
1475 if (copy_to_user(rmtp, &rmt, sizeof(*rmtp)))
1481 long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1483 struct hrtimer_sleeper t;
1484 struct timespec __user *rmtp;
1487 hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid,
1489 hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
1491 if (do_nanosleep(&t, HRTIMER_MODE_ABS))
1494 rmtp = restart->nanosleep.rmtp;
1496 ret = update_rmtp(&t.timer, rmtp);
1501 /* The other values in restart are already filled in */
1502 ret = -ERESTART_RESTARTBLOCK;
1504 destroy_hrtimer_on_stack(&t.timer);
1508 long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
1509 const enum hrtimer_mode mode, const clockid_t clockid)
1511 struct restart_block *restart;
1512 struct hrtimer_sleeper t;
1516 slack = current->timer_slack_ns;
1517 if (dl_task(current) || rt_task(current))
1520 hrtimer_init_on_stack(&t.timer, clockid, mode);
1521 hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack);
1522 if (do_nanosleep(&t, mode))
1525 /* Absolute timers do not update the rmtp value and restart: */
1526 if (mode == HRTIMER_MODE_ABS) {
1527 ret = -ERESTARTNOHAND;
1532 ret = update_rmtp(&t.timer, rmtp);
1537 restart = ¤t->restart_block;
1538 restart->fn = hrtimer_nanosleep_restart;
1539 restart->nanosleep.clockid = t.timer.base->clockid;
1540 restart->nanosleep.rmtp = rmtp;
1541 restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
1543 ret = -ERESTART_RESTARTBLOCK;
1545 destroy_hrtimer_on_stack(&t.timer);
1549 SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp,
1550 struct timespec __user *, rmtp)
1554 if (copy_from_user(&tu, rqtp, sizeof(tu)))
1557 if (!timespec_valid(&tu))
1560 return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1564 * Functions related to boot-time initialization:
1566 int hrtimers_prepare_cpu(unsigned int cpu)
1568 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1571 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1572 cpu_base->clock_base[i].cpu_base = cpu_base;
1573 timerqueue_init_head(&cpu_base->clock_base[i].active);
1576 cpu_base->cpu = cpu;
1577 hrtimer_init_hres(cpu_base);
1581 #ifdef CONFIG_HOTPLUG_CPU
1583 static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1584 struct hrtimer_clock_base *new_base)
1586 struct hrtimer *timer;
1587 struct timerqueue_node *node;
1589 while ((node = timerqueue_getnext(&old_base->active))) {
1590 timer = container_of(node, struct hrtimer, node);
1591 BUG_ON(hrtimer_callback_running(timer));
1592 debug_deactivate(timer);
1595 * Mark it as ENQUEUED not INACTIVE otherwise the
1596 * timer could be seen as !active and just vanish away
1597 * under us on another CPU
1599 __remove_hrtimer(timer, old_base, HRTIMER_STATE_ENQUEUED, 0);
1600 timer->base = new_base;
1602 * Enqueue the timers on the new cpu. This does not
1603 * reprogram the event device in case the timer
1604 * expires before the earliest on this CPU, but we run
1605 * hrtimer_interrupt after we migrated everything to
1606 * sort out already expired timers and reprogram the
1609 enqueue_hrtimer(timer, new_base);
1613 int hrtimers_dead_cpu(unsigned int scpu)
1615 struct hrtimer_cpu_base *old_base, *new_base;
1618 BUG_ON(cpu_online(scpu));
1619 tick_cancel_sched_timer(scpu);
1621 local_irq_disable();
1622 old_base = &per_cpu(hrtimer_bases, scpu);
1623 new_base = this_cpu_ptr(&hrtimer_bases);
1625 * The caller is globally serialized and nobody else
1626 * takes two locks at once, deadlock is not possible.
1628 raw_spin_lock(&new_base->lock);
1629 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1631 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1632 migrate_hrtimer_list(&old_base->clock_base[i],
1633 &new_base->clock_base[i]);
1636 raw_spin_unlock(&old_base->lock);
1637 raw_spin_unlock(&new_base->lock);
1639 /* Check, if we got expired work to do */
1640 __hrtimer_peek_ahead_timers();
1645 #endif /* CONFIG_HOTPLUG_CPU */
1647 void __init hrtimers_init(void)
1649 hrtimers_prepare_cpu(smp_processor_id());
1653 * schedule_hrtimeout_range_clock - sleep until timeout
1654 * @expires: timeout value (ktime_t)
1655 * @delta: slack in expires timeout (ktime_t)
1656 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1657 * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME
1660 schedule_hrtimeout_range_clock(ktime_t *expires, u64 delta,
1661 const enum hrtimer_mode mode, int clock)
1663 struct hrtimer_sleeper t;
1666 * Optimize when a zero timeout value is given. It does not
1667 * matter whether this is an absolute or a relative time.
1669 if (expires && *expires == 0) {
1670 __set_current_state(TASK_RUNNING);
1675 * A NULL parameter means "infinite"
1682 hrtimer_init_on_stack(&t.timer, clock, mode);
1683 hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
1685 hrtimer_init_sleeper(&t, current);
1687 hrtimer_start_expires(&t.timer, mode);
1692 hrtimer_cancel(&t.timer);
1693 destroy_hrtimer_on_stack(&t.timer);
1695 __set_current_state(TASK_RUNNING);
1697 return !t.task ? 0 : -EINTR;
1701 * schedule_hrtimeout_range - sleep until timeout
1702 * @expires: timeout value (ktime_t)
1703 * @delta: slack in expires timeout (ktime_t)
1704 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1706 * Make the current task sleep until the given expiry time has
1707 * elapsed. The routine will return immediately unless
1708 * the current task state has been set (see set_current_state()).
1710 * The @delta argument gives the kernel the freedom to schedule the
1711 * actual wakeup to a time that is both power and performance friendly.
1712 * The kernel give the normal best effort behavior for "@expires+@delta",
1713 * but may decide to fire the timer earlier, but no earlier than @expires.
1715 * You can set the task state as follows -
1717 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1718 * pass before the routine returns unless the current task is explicitly
1719 * woken up, (e.g. by wake_up_process()).
1721 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1722 * delivered to the current task or the current task is explicitly woken
1725 * The current task state is guaranteed to be TASK_RUNNING when this
1728 * Returns 0 when the timer has expired. If the task was woken before the
1729 * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or
1730 * by an explicit wakeup, it returns -EINTR.
1732 int __sched schedule_hrtimeout_range(ktime_t *expires, u64 delta,
1733 const enum hrtimer_mode mode)
1735 return schedule_hrtimeout_range_clock(expires, delta, mode,
1738 EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
1741 * schedule_hrtimeout - sleep until timeout
1742 * @expires: timeout value (ktime_t)
1743 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1745 * Make the current task sleep until the given expiry time has
1746 * elapsed. The routine will return immediately unless
1747 * the current task state has been set (see set_current_state()).
1749 * You can set the task state as follows -
1751 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1752 * pass before the routine returns unless the current task is explicitly
1753 * woken up, (e.g. by wake_up_process()).
1755 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1756 * delivered to the current task or the current task is explicitly woken
1759 * The current task state is guaranteed to be TASK_RUNNING when this
1762 * Returns 0 when the timer has expired. If the task was woken before the
1763 * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or
1764 * by an explicit wakeup, it returns -EINTR.
1766 int __sched schedule_hrtimeout(ktime_t *expires,
1767 const enum hrtimer_mode mode)
1769 return schedule_hrtimeout_range(expires, 0, mode);
1771 EXPORT_SYMBOL_GPL(schedule_hrtimeout);