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 than 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;
180 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
182 struct hrtimer_cpu_base *get_target_base(struct hrtimer_cpu_base *base,
185 if (pinned || !base->migration_enabled)
187 return &per_cpu(hrtimer_bases, get_nohz_timer_target());
191 struct hrtimer_cpu_base *get_target_base(struct hrtimer_cpu_base *base,
199 * We switch the timer base to a power-optimized selected CPU target,
201 * - NO_HZ_COMMON is enabled
202 * - timer migration is enabled
203 * - the timer callback is not running
204 * - the timer is not the first expiring timer on the new target
206 * If one of the above requirements is not fulfilled we move the timer
207 * to the current CPU or leave it on the previously assigned CPU if
208 * the timer callback is currently running.
210 static inline struct hrtimer_clock_base *
211 switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
214 struct hrtimer_cpu_base *new_cpu_base, *this_cpu_base;
215 struct hrtimer_clock_base *new_base;
216 int basenum = base->index;
218 this_cpu_base = this_cpu_ptr(&hrtimer_bases);
219 new_cpu_base = get_target_base(this_cpu_base, pinned);
221 new_base = &new_cpu_base->clock_base[basenum];
223 if (base != new_base) {
225 * We are trying to move timer to new_base.
226 * However we can't change timer's base while it is running,
227 * so we keep it on the same CPU. No hassle vs. reprogramming
228 * the event source in the high resolution case. The softirq
229 * code will take care of this when the timer function has
230 * completed. There is no conflict as we hold the lock until
231 * the timer is enqueued.
233 if (unlikely(hrtimer_callback_running(timer)))
236 /* See the comment in lock_hrtimer_base() */
237 timer->base = &migration_base;
238 raw_spin_unlock(&base->cpu_base->lock);
239 raw_spin_lock(&new_base->cpu_base->lock);
241 if (new_cpu_base != this_cpu_base &&
242 hrtimer_check_target(timer, new_base)) {
243 raw_spin_unlock(&new_base->cpu_base->lock);
244 raw_spin_lock(&base->cpu_base->lock);
245 new_cpu_base = this_cpu_base;
249 timer->base = new_base;
251 if (new_cpu_base != this_cpu_base &&
252 hrtimer_check_target(timer, new_base)) {
253 new_cpu_base = this_cpu_base;
260 #else /* CONFIG_SMP */
262 static inline struct hrtimer_clock_base *
263 lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
265 struct hrtimer_clock_base *base = timer->base;
267 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
272 # define switch_hrtimer_base(t, b, p) (b)
274 #endif /* !CONFIG_SMP */
277 * Functions for the union type storage format of ktime_t which are
278 * too large for inlining:
280 #if BITS_PER_LONG < 64
282 * Divide a ktime value by a nanosecond value
284 s64 __ktime_divns(const ktime_t kt, s64 div)
290 dclc = ktime_to_ns(kt);
291 tmp = dclc < 0 ? -dclc : dclc;
293 /* Make sure the divisor is less than 2^32: */
299 do_div(tmp, (unsigned long) div);
300 return dclc < 0 ? -tmp : tmp;
302 EXPORT_SYMBOL_GPL(__ktime_divns);
303 #endif /* BITS_PER_LONG >= 64 */
306 * Add two ktime values and do a safety check for overflow:
308 ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
310 ktime_t res = ktime_add(lhs, rhs);
313 * We use KTIME_SEC_MAX here, the maximum timeout which we can
314 * return to user space in a timespec:
316 if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64)
317 res = ktime_set(KTIME_SEC_MAX, 0);
322 EXPORT_SYMBOL_GPL(ktime_add_safe);
324 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
326 static struct debug_obj_descr hrtimer_debug_descr;
328 static void *hrtimer_debug_hint(void *addr)
330 return ((struct hrtimer *) addr)->function;
334 * fixup_init is called when:
335 * - an active object is initialized
337 static bool hrtimer_fixup_init(void *addr, enum debug_obj_state state)
339 struct hrtimer *timer = addr;
342 case ODEBUG_STATE_ACTIVE:
343 hrtimer_cancel(timer);
344 debug_object_init(timer, &hrtimer_debug_descr);
352 * fixup_activate is called when:
353 * - an active object is activated
354 * - an unknown non-static object is activated
356 static bool hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
359 case ODEBUG_STATE_ACTIVE:
368 * fixup_free is called when:
369 * - an active object is freed
371 static bool hrtimer_fixup_free(void *addr, enum debug_obj_state state)
373 struct hrtimer *timer = addr;
376 case ODEBUG_STATE_ACTIVE:
377 hrtimer_cancel(timer);
378 debug_object_free(timer, &hrtimer_debug_descr);
385 static struct debug_obj_descr hrtimer_debug_descr = {
387 .debug_hint = hrtimer_debug_hint,
388 .fixup_init = hrtimer_fixup_init,
389 .fixup_activate = hrtimer_fixup_activate,
390 .fixup_free = hrtimer_fixup_free,
393 static inline void debug_hrtimer_init(struct hrtimer *timer)
395 debug_object_init(timer, &hrtimer_debug_descr);
398 static inline void debug_hrtimer_activate(struct hrtimer *timer)
400 debug_object_activate(timer, &hrtimer_debug_descr);
403 static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
405 debug_object_deactivate(timer, &hrtimer_debug_descr);
408 static inline void debug_hrtimer_free(struct hrtimer *timer)
410 debug_object_free(timer, &hrtimer_debug_descr);
413 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
414 enum hrtimer_mode mode);
416 void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
417 enum hrtimer_mode mode)
419 debug_object_init_on_stack(timer, &hrtimer_debug_descr);
420 __hrtimer_init(timer, clock_id, mode);
422 EXPORT_SYMBOL_GPL(hrtimer_init_on_stack);
424 void destroy_hrtimer_on_stack(struct hrtimer *timer)
426 debug_object_free(timer, &hrtimer_debug_descr);
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 ktime_t expires, expires_next = { .tv64 = KTIME_MAX };
468 unsigned int active = cpu_base->active_bases;
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.tv64 < expires_next.tv64) {
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.tv64 < 0)
492 expires_next.tv64 = 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.tv64 == cpu_base->expires_next.tv64)
566 cpu_base->expires_next.tv64 = expires_next.tv64;
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.
624 if (expires.tv64 < 0)
627 if (expires.tv64 >= cpu_base->expires_next.tv64)
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.tv64 = 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 reprogramm 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();
768 static inline void timer_stats_hrtimer_set_start_info(struct hrtimer *timer)
770 #ifdef CONFIG_TIMER_STATS
771 if (timer->start_site)
773 timer->start_site = __builtin_return_address(0);
774 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
775 timer->start_pid = current->pid;
779 static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer *timer)
781 #ifdef CONFIG_TIMER_STATS
782 timer->start_site = NULL;
786 static inline void timer_stats_account_hrtimer(struct hrtimer *timer)
788 #ifdef CONFIG_TIMER_STATS
789 if (likely(!timer_stats_active))
791 timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
792 timer->function, timer->start_comm, 0);
797 * Counterpart to lock_hrtimer_base above:
800 void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
802 raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
806 * hrtimer_forward - forward the timer expiry
807 * @timer: hrtimer to forward
808 * @now: forward past this time
809 * @interval: the interval to forward
811 * Forward the timer expiry so it will expire in the future.
812 * Returns the number of overruns.
814 * Can be safely called from the callback function of @timer. If
815 * called from other contexts @timer must neither be enqueued nor
816 * running the callback and the caller needs to take care of
819 * Note: This only updates the timer expiry value and does not requeue
822 u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
827 delta = ktime_sub(now, hrtimer_get_expires(timer));
832 if (WARN_ON(timer->state & HRTIMER_STATE_ENQUEUED))
835 if (interval.tv64 < hrtimer_resolution)
836 interval.tv64 = hrtimer_resolution;
838 if (unlikely(delta.tv64 >= interval.tv64)) {
839 s64 incr = ktime_to_ns(interval);
841 orun = ktime_divns(delta, incr);
842 hrtimer_add_expires_ns(timer, incr * orun);
843 if (hrtimer_get_expires_tv64(timer) > now.tv64)
846 * This (and the ktime_add() below) is the
847 * correction for exact:
851 hrtimer_add_expires(timer, interval);
855 EXPORT_SYMBOL_GPL(hrtimer_forward);
858 * enqueue_hrtimer - internal function to (re)start a timer
860 * The timer is inserted in expiry order. Insertion into the
861 * red black tree is O(log(n)). Must hold the base lock.
863 * Returns 1 when the new timer is the leftmost timer in the tree.
865 static int enqueue_hrtimer(struct hrtimer *timer,
866 struct hrtimer_clock_base *base)
868 debug_activate(timer);
870 base->cpu_base->active_bases |= 1 << base->index;
872 timer->state = HRTIMER_STATE_ENQUEUED;
874 return timerqueue_add(&base->active, &timer->node);
878 * __remove_hrtimer - internal function to remove a timer
880 * Caller must hold the base lock.
882 * High resolution timer mode reprograms the clock event device when the
883 * timer is the one which expires next. The caller can disable this by setting
884 * reprogram to zero. This is useful, when the context does a reprogramming
885 * anyway (e.g. timer interrupt)
887 static void __remove_hrtimer(struct hrtimer *timer,
888 struct hrtimer_clock_base *base,
889 u8 newstate, int reprogram)
891 struct hrtimer_cpu_base *cpu_base = base->cpu_base;
892 u8 state = timer->state;
894 timer->state = newstate;
895 if (!(state & HRTIMER_STATE_ENQUEUED))
898 if (!timerqueue_del(&base->active, &timer->node))
899 cpu_base->active_bases &= ~(1 << base->index);
901 #ifdef CONFIG_HIGH_RES_TIMERS
903 * Note: If reprogram is false we do not update
904 * cpu_base->next_timer. This happens when we remove the first
905 * timer on a remote cpu. No harm as we never dereference
906 * cpu_base->next_timer. So the worst thing what can happen is
907 * an superflous call to hrtimer_force_reprogram() on the
908 * remote cpu later on if the same timer gets enqueued again.
910 if (reprogram && timer == cpu_base->next_timer)
911 hrtimer_force_reprogram(cpu_base, 1);
916 * remove hrtimer, called with base lock held
919 remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base, bool restart)
921 if (hrtimer_is_queued(timer)) {
922 u8 state = timer->state;
926 * Remove the timer and force reprogramming when high
927 * resolution mode is active and the timer is on the current
928 * CPU. If we remove a timer on another CPU, reprogramming is
929 * skipped. The interrupt event on this CPU is fired and
930 * reprogramming happens in the interrupt handler. This is a
931 * rare case and less expensive than a smp call.
933 debug_deactivate(timer);
934 timer_stats_hrtimer_clear_start_info(timer);
935 reprogram = base->cpu_base == this_cpu_ptr(&hrtimer_bases);
938 state = HRTIMER_STATE_INACTIVE;
940 __remove_hrtimer(timer, base, state, reprogram);
946 static inline ktime_t hrtimer_update_lowres(struct hrtimer *timer, ktime_t tim,
947 const enum hrtimer_mode mode)
949 #ifdef CONFIG_TIME_LOW_RES
951 * CONFIG_TIME_LOW_RES indicates that the system has no way to return
952 * granular time values. For relative timers we add hrtimer_resolution
953 * (i.e. one jiffie) to prevent short timeouts.
955 timer->is_rel = mode & HRTIMER_MODE_REL;
957 tim = ktime_add_safe(tim, ktime_set(0, hrtimer_resolution));
963 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
964 * @timer: the timer to be added
966 * @delta_ns: "slack" range for the timer
967 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
968 * relative (HRTIMER_MODE_REL)
970 void hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
971 u64 delta_ns, const enum hrtimer_mode mode)
973 struct hrtimer_clock_base *base, *new_base;
977 base = lock_hrtimer_base(timer, &flags);
979 /* Remove an active timer from the queue: */
980 remove_hrtimer(timer, base, true);
982 if (mode & HRTIMER_MODE_REL)
983 tim = ktime_add_safe(tim, base->get_time());
985 tim = hrtimer_update_lowres(timer, tim, mode);
987 hrtimer_set_expires_range_ns(timer, tim, delta_ns);
989 /* Switch the timer base, if necessary: */
990 new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED);
992 timer_stats_hrtimer_set_start_info(timer);
994 leftmost = enqueue_hrtimer(timer, new_base);
998 if (!hrtimer_is_hres_active(timer)) {
1000 * Kick to reschedule the next tick to handle the new timer
1001 * on dynticks target.
1003 if (new_base->cpu_base->nohz_active)
1004 wake_up_nohz_cpu(new_base->cpu_base->cpu);
1006 hrtimer_reprogram(timer, new_base);
1009 unlock_hrtimer_base(timer, &flags);
1011 EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
1014 * hrtimer_try_to_cancel - try to deactivate a timer
1015 * @timer: hrtimer to stop
1018 * 0 when the timer was not active
1019 * 1 when the timer was active
1020 * -1 when the timer is currently excuting the callback function and
1023 int hrtimer_try_to_cancel(struct hrtimer *timer)
1025 struct hrtimer_clock_base *base;
1026 unsigned long flags;
1030 * Check lockless first. If the timer is not active (neither
1031 * enqueued nor running the callback, nothing to do here. The
1032 * base lock does not serialize against a concurrent enqueue,
1033 * so we can avoid taking it.
1035 if (!hrtimer_active(timer))
1038 base = lock_hrtimer_base(timer, &flags);
1040 if (!hrtimer_callback_running(timer))
1041 ret = remove_hrtimer(timer, base, false);
1043 unlock_hrtimer_base(timer, &flags);
1048 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1051 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1052 * @timer: the timer to be cancelled
1055 * 0 when the timer was not active
1056 * 1 when the timer was active
1058 int hrtimer_cancel(struct hrtimer *timer)
1061 int ret = hrtimer_try_to_cancel(timer);
1068 EXPORT_SYMBOL_GPL(hrtimer_cancel);
1071 * hrtimer_get_remaining - get remaining time for the timer
1072 * @timer: the timer to read
1073 * @adjust: adjust relative timers when CONFIG_TIME_LOW_RES=y
1075 ktime_t __hrtimer_get_remaining(const struct hrtimer *timer, bool adjust)
1077 unsigned long flags;
1080 lock_hrtimer_base(timer, &flags);
1081 if (IS_ENABLED(CONFIG_TIME_LOW_RES) && adjust)
1082 rem = hrtimer_expires_remaining_adjusted(timer);
1084 rem = hrtimer_expires_remaining(timer);
1085 unlock_hrtimer_base(timer, &flags);
1089 EXPORT_SYMBOL_GPL(__hrtimer_get_remaining);
1091 #ifdef CONFIG_NO_HZ_COMMON
1093 * hrtimer_get_next_event - get the time until next expiry event
1095 * Returns the next expiry time or KTIME_MAX if no timer is pending.
1097 u64 hrtimer_get_next_event(void)
1099 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1100 u64 expires = KTIME_MAX;
1101 unsigned long flags;
1103 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1105 if (!__hrtimer_hres_active(cpu_base))
1106 expires = __hrtimer_get_next_event(cpu_base).tv64;
1108 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1114 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1115 enum hrtimer_mode mode)
1117 struct hrtimer_cpu_base *cpu_base;
1120 memset(timer, 0, sizeof(struct hrtimer));
1122 cpu_base = raw_cpu_ptr(&hrtimer_bases);
1124 if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
1125 clock_id = CLOCK_MONOTONIC;
1127 base = hrtimer_clockid_to_base(clock_id);
1128 timer->base = &cpu_base->clock_base[base];
1129 timerqueue_init(&timer->node);
1131 #ifdef CONFIG_TIMER_STATS
1132 timer->start_site = NULL;
1133 timer->start_pid = -1;
1134 memset(timer->start_comm, 0, TASK_COMM_LEN);
1139 * hrtimer_init - initialize a timer to the given clock
1140 * @timer: the timer to be initialized
1141 * @clock_id: the clock to be used
1142 * @mode: timer mode abs/rel
1144 void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1145 enum hrtimer_mode mode)
1147 debug_init(timer, clock_id, mode);
1148 __hrtimer_init(timer, clock_id, mode);
1150 EXPORT_SYMBOL_GPL(hrtimer_init);
1153 * A timer is active, when it is enqueued into the rbtree or the
1154 * callback function is running or it's in the state of being migrated
1157 * It is important for this function to not return a false negative.
1159 bool hrtimer_active(const struct hrtimer *timer)
1161 struct hrtimer_cpu_base *cpu_base;
1165 cpu_base = READ_ONCE(timer->base->cpu_base);
1166 seq = raw_read_seqcount_begin(&cpu_base->seq);
1168 if (timer->state != HRTIMER_STATE_INACTIVE ||
1169 cpu_base->running == timer)
1172 } while (read_seqcount_retry(&cpu_base->seq, seq) ||
1173 cpu_base != READ_ONCE(timer->base->cpu_base));
1177 EXPORT_SYMBOL_GPL(hrtimer_active);
1180 * The write_seqcount_barrier()s in __run_hrtimer() split the thing into 3
1181 * distinct sections:
1183 * - queued: the timer is queued
1184 * - callback: the timer is being ran
1185 * - post: the timer is inactive or (re)queued
1187 * On the read side we ensure we observe timer->state and cpu_base->running
1188 * from the same section, if anything changed while we looked at it, we retry.
1189 * This includes timer->base changing because sequence numbers alone are
1190 * insufficient for that.
1192 * The sequence numbers are required because otherwise we could still observe
1193 * a false negative if the read side got smeared over multiple consequtive
1194 * __run_hrtimer() invocations.
1197 static void __run_hrtimer(struct hrtimer_cpu_base *cpu_base,
1198 struct hrtimer_clock_base *base,
1199 struct hrtimer *timer, ktime_t *now)
1201 enum hrtimer_restart (*fn)(struct hrtimer *);
1204 lockdep_assert_held(&cpu_base->lock);
1206 debug_deactivate(timer);
1207 cpu_base->running = timer;
1210 * Separate the ->running assignment from the ->state assignment.
1212 * As with a regular write barrier, this ensures the read side in
1213 * hrtimer_active() cannot observe cpu_base->running == NULL &&
1214 * timer->state == INACTIVE.
1216 raw_write_seqcount_barrier(&cpu_base->seq);
1218 __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE, 0);
1219 timer_stats_account_hrtimer(timer);
1220 fn = timer->function;
1223 * Clear the 'is relative' flag for the TIME_LOW_RES case. If the
1224 * timer is restarted with a period then it becomes an absolute
1225 * timer. If its not restarted it does not matter.
1227 if (IS_ENABLED(CONFIG_TIME_LOW_RES))
1228 timer->is_rel = false;
1231 * Because we run timers from hardirq context, there is no chance
1232 * they get migrated to another cpu, therefore its safe to unlock
1235 raw_spin_unlock(&cpu_base->lock);
1236 trace_hrtimer_expire_entry(timer, now);
1237 restart = fn(timer);
1238 trace_hrtimer_expire_exit(timer);
1239 raw_spin_lock(&cpu_base->lock);
1242 * Note: We clear the running state after enqueue_hrtimer and
1243 * we do not reprogramm the event hardware. Happens either in
1244 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1246 * Note: Because we dropped the cpu_base->lock above,
1247 * hrtimer_start_range_ns() can have popped in and enqueued the timer
1250 if (restart != HRTIMER_NORESTART &&
1251 !(timer->state & HRTIMER_STATE_ENQUEUED))
1252 enqueue_hrtimer(timer, base);
1255 * Separate the ->running assignment from the ->state assignment.
1257 * As with a regular write barrier, this ensures the read side in
1258 * hrtimer_active() cannot observe cpu_base->running == NULL &&
1259 * timer->state == INACTIVE.
1261 raw_write_seqcount_barrier(&cpu_base->seq);
1263 WARN_ON_ONCE(cpu_base->running != timer);
1264 cpu_base->running = NULL;
1267 static void __hrtimer_run_queues(struct hrtimer_cpu_base *cpu_base, ktime_t now)
1269 struct hrtimer_clock_base *base = cpu_base->clock_base;
1270 unsigned int active = cpu_base->active_bases;
1272 for (; active; base++, active >>= 1) {
1273 struct timerqueue_node *node;
1276 if (!(active & 0x01))
1279 basenow = ktime_add(now, base->offset);
1281 while ((node = timerqueue_getnext(&base->active))) {
1282 struct hrtimer *timer;
1284 timer = container_of(node, struct hrtimer, node);
1287 * The immediate goal for using the softexpires is
1288 * minimizing wakeups, not running timers at the
1289 * earliest interrupt after their soft expiration.
1290 * This allows us to avoid using a Priority Search
1291 * Tree, which can answer a stabbing querry for
1292 * overlapping intervals and instead use the simple
1293 * BST we already have.
1294 * We don't add extra wakeups by delaying timers that
1295 * are right-of a not yet expired timer, because that
1296 * timer will have to trigger a wakeup anyway.
1298 if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer))
1301 __run_hrtimer(cpu_base, base, timer, &basenow);
1306 #ifdef CONFIG_HIGH_RES_TIMERS
1309 * High resolution timer interrupt
1310 * Called with interrupts disabled
1312 void hrtimer_interrupt(struct clock_event_device *dev)
1314 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1315 ktime_t expires_next, now, entry_time, delta;
1318 BUG_ON(!cpu_base->hres_active);
1319 cpu_base->nr_events++;
1320 dev->next_event.tv64 = KTIME_MAX;
1322 raw_spin_lock(&cpu_base->lock);
1323 entry_time = now = hrtimer_update_base(cpu_base);
1325 cpu_base->in_hrtirq = 1;
1327 * We set expires_next to KTIME_MAX here with cpu_base->lock
1328 * held to prevent that a timer is enqueued in our queue via
1329 * the migration code. This does not affect enqueueing of
1330 * timers which run their callback and need to be requeued on
1333 cpu_base->expires_next.tv64 = KTIME_MAX;
1335 __hrtimer_run_queues(cpu_base, now);
1337 /* Reevaluate the clock bases for the next expiry */
1338 expires_next = __hrtimer_get_next_event(cpu_base);
1340 * Store the new expiry value so the migration code can verify
1343 cpu_base->expires_next = expires_next;
1344 cpu_base->in_hrtirq = 0;
1345 raw_spin_unlock(&cpu_base->lock);
1347 /* Reprogramming necessary ? */
1348 if (!tick_program_event(expires_next, 0)) {
1349 cpu_base->hang_detected = 0;
1354 * The next timer was already expired due to:
1356 * - long lasting callbacks
1357 * - being scheduled away when running in a VM
1359 * We need to prevent that we loop forever in the hrtimer
1360 * interrupt routine. We give it 3 attempts to avoid
1361 * overreacting on some spurious event.
1363 * Acquire base lock for updating the offsets and retrieving
1366 raw_spin_lock(&cpu_base->lock);
1367 now = hrtimer_update_base(cpu_base);
1368 cpu_base->nr_retries++;
1372 * Give the system a chance to do something else than looping
1373 * here. We stored the entry time, so we know exactly how long
1374 * we spent here. We schedule the next event this amount of
1377 cpu_base->nr_hangs++;
1378 cpu_base->hang_detected = 1;
1379 raw_spin_unlock(&cpu_base->lock);
1380 delta = ktime_sub(now, entry_time);
1381 if ((unsigned int)delta.tv64 > cpu_base->max_hang_time)
1382 cpu_base->max_hang_time = (unsigned int) delta.tv64;
1384 * Limit it to a sensible value as we enforce a longer
1385 * delay. Give the CPU at least 100ms to catch up.
1387 if (delta.tv64 > 100 * NSEC_PER_MSEC)
1388 expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1390 expires_next = ktime_add(now, delta);
1391 tick_program_event(expires_next, 1);
1392 printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n",
1393 ktime_to_ns(delta));
1397 * local version of hrtimer_peek_ahead_timers() called with interrupts
1400 static inline void __hrtimer_peek_ahead_timers(void)
1402 struct tick_device *td;
1404 if (!hrtimer_hres_active())
1407 td = this_cpu_ptr(&tick_cpu_device);
1408 if (td && td->evtdev)
1409 hrtimer_interrupt(td->evtdev);
1412 #else /* CONFIG_HIGH_RES_TIMERS */
1414 static inline void __hrtimer_peek_ahead_timers(void) { }
1416 #endif /* !CONFIG_HIGH_RES_TIMERS */
1419 * Called from run_local_timers in hardirq context every jiffy
1421 void hrtimer_run_queues(void)
1423 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1426 if (__hrtimer_hres_active(cpu_base))
1430 * This _is_ ugly: We have to check periodically, whether we
1431 * can switch to highres and / or nohz mode. The clocksource
1432 * switch happens with xtime_lock held. Notification from
1433 * there only sets the check bit in the tick_oneshot code,
1434 * otherwise we might deadlock vs. xtime_lock.
1436 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) {
1437 hrtimer_switch_to_hres();
1441 raw_spin_lock(&cpu_base->lock);
1442 now = hrtimer_update_base(cpu_base);
1443 __hrtimer_run_queues(cpu_base, now);
1444 raw_spin_unlock(&cpu_base->lock);
1448 * Sleep related functions:
1450 static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1452 struct hrtimer_sleeper *t =
1453 container_of(timer, struct hrtimer_sleeper, timer);
1454 struct task_struct *task = t->task;
1458 wake_up_process(task);
1460 return HRTIMER_NORESTART;
1463 void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1465 sl->timer.function = hrtimer_wakeup;
1468 EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
1470 static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1472 hrtimer_init_sleeper(t, current);
1475 set_current_state(TASK_INTERRUPTIBLE);
1476 hrtimer_start_expires(&t->timer, mode);
1478 if (likely(t->task))
1479 freezable_schedule();
1481 hrtimer_cancel(&t->timer);
1482 mode = HRTIMER_MODE_ABS;
1484 } while (t->task && !signal_pending(current));
1486 __set_current_state(TASK_RUNNING);
1488 return t->task == NULL;
1491 static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp)
1493 struct timespec rmt;
1496 rem = hrtimer_expires_remaining(timer);
1499 rmt = ktime_to_timespec(rem);
1501 if (copy_to_user(rmtp, &rmt, sizeof(*rmtp)))
1507 long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1509 struct hrtimer_sleeper t;
1510 struct timespec __user *rmtp;
1513 hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid,
1515 hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
1517 if (do_nanosleep(&t, HRTIMER_MODE_ABS))
1520 rmtp = restart->nanosleep.rmtp;
1522 ret = update_rmtp(&t.timer, rmtp);
1527 /* The other values in restart are already filled in */
1528 ret = -ERESTART_RESTARTBLOCK;
1530 destroy_hrtimer_on_stack(&t.timer);
1534 long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
1535 const enum hrtimer_mode mode, const clockid_t clockid)
1537 struct restart_block *restart;
1538 struct hrtimer_sleeper t;
1542 slack = current->timer_slack_ns;
1543 if (dl_task(current) || rt_task(current))
1546 hrtimer_init_on_stack(&t.timer, clockid, mode);
1547 hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack);
1548 if (do_nanosleep(&t, mode))
1551 /* Absolute timers do not update the rmtp value and restart: */
1552 if (mode == HRTIMER_MODE_ABS) {
1553 ret = -ERESTARTNOHAND;
1558 ret = update_rmtp(&t.timer, rmtp);
1563 restart = ¤t->restart_block;
1564 restart->fn = hrtimer_nanosleep_restart;
1565 restart->nanosleep.clockid = t.timer.base->clockid;
1566 restart->nanosleep.rmtp = rmtp;
1567 restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
1569 ret = -ERESTART_RESTARTBLOCK;
1571 destroy_hrtimer_on_stack(&t.timer);
1575 SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp,
1576 struct timespec __user *, rmtp)
1580 if (copy_from_user(&tu, rqtp, sizeof(tu)))
1583 if (!timespec_valid(&tu))
1586 return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1590 * Functions related to boot-time initialization:
1592 static void init_hrtimers_cpu(int cpu)
1594 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1597 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1598 cpu_base->clock_base[i].cpu_base = cpu_base;
1599 timerqueue_init_head(&cpu_base->clock_base[i].active);
1602 cpu_base->cpu = cpu;
1603 hrtimer_init_hres(cpu_base);
1606 #ifdef CONFIG_HOTPLUG_CPU
1608 static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1609 struct hrtimer_clock_base *new_base)
1611 struct hrtimer *timer;
1612 struct timerqueue_node *node;
1614 while ((node = timerqueue_getnext(&old_base->active))) {
1615 timer = container_of(node, struct hrtimer, node);
1616 BUG_ON(hrtimer_callback_running(timer));
1617 debug_deactivate(timer);
1620 * Mark it as ENQUEUED not INACTIVE otherwise the
1621 * timer could be seen as !active and just vanish away
1622 * under us on another CPU
1624 __remove_hrtimer(timer, old_base, HRTIMER_STATE_ENQUEUED, 0);
1625 timer->base = new_base;
1627 * Enqueue the timers on the new cpu. This does not
1628 * reprogram the event device in case the timer
1629 * expires before the earliest on this CPU, but we run
1630 * hrtimer_interrupt after we migrated everything to
1631 * sort out already expired timers and reprogram the
1634 enqueue_hrtimer(timer, new_base);
1638 static void migrate_hrtimers(int scpu)
1640 struct hrtimer_cpu_base *old_base, *new_base;
1643 BUG_ON(cpu_online(scpu));
1644 tick_cancel_sched_timer(scpu);
1646 local_irq_disable();
1647 old_base = &per_cpu(hrtimer_bases, scpu);
1648 new_base = this_cpu_ptr(&hrtimer_bases);
1650 * The caller is globally serialized and nobody else
1651 * takes two locks at once, deadlock is not possible.
1653 raw_spin_lock(&new_base->lock);
1654 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1656 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1657 migrate_hrtimer_list(&old_base->clock_base[i],
1658 &new_base->clock_base[i]);
1661 raw_spin_unlock(&old_base->lock);
1662 raw_spin_unlock(&new_base->lock);
1664 /* Check, if we got expired work to do */
1665 __hrtimer_peek_ahead_timers();
1669 #endif /* CONFIG_HOTPLUG_CPU */
1671 static int hrtimer_cpu_notify(struct notifier_block *self,
1672 unsigned long action, void *hcpu)
1674 int scpu = (long)hcpu;
1678 case CPU_UP_PREPARE:
1679 case CPU_UP_PREPARE_FROZEN:
1680 init_hrtimers_cpu(scpu);
1683 #ifdef CONFIG_HOTPLUG_CPU
1685 case CPU_DEAD_FROZEN:
1686 migrate_hrtimers(scpu);
1697 static struct notifier_block hrtimers_nb = {
1698 .notifier_call = hrtimer_cpu_notify,
1701 void __init hrtimers_init(void)
1703 hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
1704 (void *)(long)smp_processor_id());
1705 register_cpu_notifier(&hrtimers_nb);
1709 * schedule_hrtimeout_range_clock - sleep until timeout
1710 * @expires: timeout value (ktime_t)
1711 * @delta: slack in expires timeout (ktime_t)
1712 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1713 * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME
1716 schedule_hrtimeout_range_clock(ktime_t *expires, u64 delta,
1717 const enum hrtimer_mode mode, int clock)
1719 struct hrtimer_sleeper t;
1722 * Optimize when a zero timeout value is given. It does not
1723 * matter whether this is an absolute or a relative time.
1725 if (expires && !expires->tv64) {
1726 __set_current_state(TASK_RUNNING);
1731 * A NULL parameter means "infinite"
1738 hrtimer_init_on_stack(&t.timer, clock, mode);
1739 hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
1741 hrtimer_init_sleeper(&t, current);
1743 hrtimer_start_expires(&t.timer, mode);
1748 hrtimer_cancel(&t.timer);
1749 destroy_hrtimer_on_stack(&t.timer);
1751 __set_current_state(TASK_RUNNING);
1753 return !t.task ? 0 : -EINTR;
1757 * schedule_hrtimeout_range - sleep until timeout
1758 * @expires: timeout value (ktime_t)
1759 * @delta: slack in expires timeout (ktime_t)
1760 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1762 * Make the current task sleep until the given expiry time has
1763 * elapsed. The routine will return immediately unless
1764 * the current task state has been set (see set_current_state()).
1766 * The @delta argument gives the kernel the freedom to schedule the
1767 * actual wakeup to a time that is both power and performance friendly.
1768 * The kernel give the normal best effort behavior for "@expires+@delta",
1769 * but may decide to fire the timer earlier, but no earlier than @expires.
1771 * You can set the task state as follows -
1773 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1774 * pass before the routine returns.
1776 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1777 * delivered to the current task.
1779 * The current task state is guaranteed to be TASK_RUNNING when this
1782 * Returns 0 when the timer has expired otherwise -EINTR
1784 int __sched schedule_hrtimeout_range(ktime_t *expires, u64 delta,
1785 const enum hrtimer_mode mode)
1787 return schedule_hrtimeout_range_clock(expires, delta, mode,
1790 EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
1793 * schedule_hrtimeout - sleep until timeout
1794 * @expires: timeout value (ktime_t)
1795 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1797 * Make the current task sleep until the given expiry time has
1798 * elapsed. The routine will return immediately unless
1799 * the current task state has been set (see set_current_state()).
1801 * You can set the task state as follows -
1803 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1804 * pass before the routine returns.
1806 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1807 * delivered to the current task.
1809 * The current task state is guaranteed to be TASK_RUNNING when this
1812 * Returns 0 when the timer has expired otherwise -EINTR
1814 int __sched schedule_hrtimeout(ktime_t *expires,
1815 const enum hrtimer_mode mode)
1817 return schedule_hrtimeout_range(expires, 0, mode);
1819 EXPORT_SYMBOL_GPL(schedule_hrtimeout);