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 /* Make sure we catch unsupported clockids */
98 [0 ... MAX_CLOCKS - 1] = HRTIMER_MAX_CLOCK_BASES,
100 [CLOCK_REALTIME] = HRTIMER_BASE_REALTIME,
101 [CLOCK_MONOTONIC] = HRTIMER_BASE_MONOTONIC,
102 [CLOCK_BOOTTIME] = HRTIMER_BASE_BOOTTIME,
103 [CLOCK_TAI] = HRTIMER_BASE_TAI,
107 * Functions and macros which are different for UP/SMP systems are kept in a
113 * We require the migration_base for lock_hrtimer_base()/switch_hrtimer_base()
114 * such that hrtimer_callback_running() can unconditionally dereference
115 * timer->base->cpu_base
117 static struct hrtimer_cpu_base migration_cpu_base = {
118 .seq = SEQCNT_ZERO(migration_cpu_base),
119 .clock_base = { { .cpu_base = &migration_cpu_base, }, },
122 #define migration_base migration_cpu_base.clock_base[0]
125 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
126 * means that all timers which are tied to this base via timer->base are
127 * locked, and the base itself is locked too.
129 * So __run_timers/migrate_timers can safely modify all timers which could
130 * be found on the lists/queues.
132 * When the timer's base is locked, and the timer removed from list, it is
133 * possible to set timer->base = &migration_base and drop the lock: the timer
137 struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
138 unsigned long *flags)
140 struct hrtimer_clock_base *base;
144 if (likely(base != &migration_base)) {
145 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
146 if (likely(base == timer->base))
148 /* The timer has migrated to another CPU: */
149 raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
156 * With HIGHRES=y we do not migrate the timer when it is expiring
157 * before the next event on the target cpu because we cannot reprogram
158 * the target cpu hardware and we would cause it to fire late.
160 * Called with cpu_base->lock of target cpu held.
163 hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base)
165 #ifdef CONFIG_HIGH_RES_TIMERS
168 if (!new_base->cpu_base->hres_active)
171 expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset);
172 return expires.tv64 <= new_base->cpu_base->expires_next.tv64;
178 #ifdef CONFIG_NO_HZ_COMMON
180 struct hrtimer_cpu_base *get_target_base(struct hrtimer_cpu_base *base,
183 if (pinned || !base->migration_enabled)
185 return &per_cpu(hrtimer_bases, get_nohz_timer_target());
189 struct hrtimer_cpu_base *get_target_base(struct hrtimer_cpu_base *base,
197 * We switch the timer base to a power-optimized selected CPU target,
199 * - NO_HZ_COMMON is enabled
200 * - timer migration is enabled
201 * - the timer callback is not running
202 * - the timer is not the first expiring timer on the new target
204 * If one of the above requirements is not fulfilled we move the timer
205 * to the current CPU or leave it on the previously assigned CPU if
206 * the timer callback is currently running.
208 static inline struct hrtimer_clock_base *
209 switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
212 struct hrtimer_cpu_base *new_cpu_base, *this_cpu_base;
213 struct hrtimer_clock_base *new_base;
214 int basenum = base->index;
216 this_cpu_base = this_cpu_ptr(&hrtimer_bases);
217 new_cpu_base = get_target_base(this_cpu_base, pinned);
219 new_base = &new_cpu_base->clock_base[basenum];
221 if (base != new_base) {
223 * We are trying to move timer to new_base.
224 * However we can't change timer's base while it is running,
225 * so we keep it on the same CPU. No hassle vs. reprogramming
226 * the event source in the high resolution case. The softirq
227 * code will take care of this when the timer function has
228 * completed. There is no conflict as we hold the lock until
229 * the timer is enqueued.
231 if (unlikely(hrtimer_callback_running(timer)))
234 /* See the comment in lock_hrtimer_base() */
235 timer->base = &migration_base;
236 raw_spin_unlock(&base->cpu_base->lock);
237 raw_spin_lock(&new_base->cpu_base->lock);
239 if (new_cpu_base != this_cpu_base &&
240 hrtimer_check_target(timer, new_base)) {
241 raw_spin_unlock(&new_base->cpu_base->lock);
242 raw_spin_lock(&base->cpu_base->lock);
243 new_cpu_base = this_cpu_base;
247 timer->base = new_base;
249 if (new_cpu_base != this_cpu_base &&
250 hrtimer_check_target(timer, new_base)) {
251 new_cpu_base = this_cpu_base;
258 #else /* CONFIG_SMP */
260 static inline struct hrtimer_clock_base *
261 lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
263 struct hrtimer_clock_base *base = timer->base;
265 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
270 # define switch_hrtimer_base(t, b, p) (b)
272 #endif /* !CONFIG_SMP */
275 * Functions for the union type storage format of ktime_t which are
276 * too large for inlining:
278 #if BITS_PER_LONG < 64
280 * Divide a ktime value by a nanosecond value
282 s64 __ktime_divns(const ktime_t kt, s64 div)
288 dclc = ktime_to_ns(kt);
289 tmp = dclc < 0 ? -dclc : dclc;
291 /* Make sure the divisor is less than 2^32: */
297 do_div(tmp, (unsigned long) div);
298 return dclc < 0 ? -tmp : tmp;
300 EXPORT_SYMBOL_GPL(__ktime_divns);
301 #endif /* BITS_PER_LONG >= 64 */
304 * Add two ktime values and do a safety check for overflow:
306 ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
308 ktime_t res = ktime_add_unsafe(lhs, rhs);
311 * We use KTIME_SEC_MAX here, the maximum timeout which we can
312 * return to user space in a timespec:
314 if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64)
315 res = ktime_set(KTIME_SEC_MAX, 0);
320 EXPORT_SYMBOL_GPL(ktime_add_safe);
322 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
324 static struct debug_obj_descr hrtimer_debug_descr;
326 static void *hrtimer_debug_hint(void *addr)
328 return ((struct hrtimer *) addr)->function;
332 * fixup_init is called when:
333 * - an active object is initialized
335 static bool hrtimer_fixup_init(void *addr, enum debug_obj_state state)
337 struct hrtimer *timer = addr;
340 case ODEBUG_STATE_ACTIVE:
341 hrtimer_cancel(timer);
342 debug_object_init(timer, &hrtimer_debug_descr);
350 * fixup_activate is called when:
351 * - an active object is activated
352 * - an unknown non-static object is activated
354 static bool hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
357 case ODEBUG_STATE_ACTIVE:
366 * fixup_free is called when:
367 * - an active object is freed
369 static bool hrtimer_fixup_free(void *addr, enum debug_obj_state state)
371 struct hrtimer *timer = addr;
374 case ODEBUG_STATE_ACTIVE:
375 hrtimer_cancel(timer);
376 debug_object_free(timer, &hrtimer_debug_descr);
383 static struct debug_obj_descr hrtimer_debug_descr = {
385 .debug_hint = hrtimer_debug_hint,
386 .fixup_init = hrtimer_fixup_init,
387 .fixup_activate = hrtimer_fixup_activate,
388 .fixup_free = hrtimer_fixup_free,
391 static inline void debug_hrtimer_init(struct hrtimer *timer)
393 debug_object_init(timer, &hrtimer_debug_descr);
396 static inline void debug_hrtimer_activate(struct hrtimer *timer)
398 debug_object_activate(timer, &hrtimer_debug_descr);
401 static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
403 debug_object_deactivate(timer, &hrtimer_debug_descr);
406 static inline void debug_hrtimer_free(struct hrtimer *timer)
408 debug_object_free(timer, &hrtimer_debug_descr);
411 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
412 enum hrtimer_mode mode);
414 void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
415 enum hrtimer_mode mode)
417 debug_object_init_on_stack(timer, &hrtimer_debug_descr);
418 __hrtimer_init(timer, clock_id, mode);
420 EXPORT_SYMBOL_GPL(hrtimer_init_on_stack);
422 void destroy_hrtimer_on_stack(struct hrtimer *timer)
424 debug_object_free(timer, &hrtimer_debug_descr);
426 EXPORT_SYMBOL_GPL(destroy_hrtimer_on_stack);
429 static inline void debug_hrtimer_init(struct hrtimer *timer) { }
430 static inline void debug_hrtimer_activate(struct hrtimer *timer) { }
431 static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
435 debug_init(struct hrtimer *timer, clockid_t clockid,
436 enum hrtimer_mode mode)
438 debug_hrtimer_init(timer);
439 trace_hrtimer_init(timer, clockid, mode);
442 static inline void debug_activate(struct hrtimer *timer)
444 debug_hrtimer_activate(timer);
445 trace_hrtimer_start(timer);
448 static inline void debug_deactivate(struct hrtimer *timer)
450 debug_hrtimer_deactivate(timer);
451 trace_hrtimer_cancel(timer);
454 #if defined(CONFIG_NO_HZ_COMMON) || defined(CONFIG_HIGH_RES_TIMERS)
455 static inline void hrtimer_update_next_timer(struct hrtimer_cpu_base *cpu_base,
456 struct hrtimer *timer)
458 #ifdef CONFIG_HIGH_RES_TIMERS
459 cpu_base->next_timer = timer;
463 static ktime_t __hrtimer_get_next_event(struct hrtimer_cpu_base *cpu_base)
465 struct hrtimer_clock_base *base = cpu_base->clock_base;
466 ktime_t expires, expires_next = { .tv64 = KTIME_MAX };
467 unsigned int active = cpu_base->active_bases;
469 hrtimer_update_next_timer(cpu_base, NULL);
470 for (; active; base++, active >>= 1) {
471 struct timerqueue_node *next;
472 struct hrtimer *timer;
474 if (!(active & 0x01))
477 next = timerqueue_getnext(&base->active);
478 timer = container_of(next, struct hrtimer, node);
479 expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
480 if (expires.tv64 < expires_next.tv64) {
481 expires_next = expires;
482 hrtimer_update_next_timer(cpu_base, timer);
486 * clock_was_set() might have changed base->offset of any of
487 * the clock bases so the result might be negative. Fix it up
488 * to prevent a false positive in clockevents_program_event().
490 if (expires_next.tv64 < 0)
491 expires_next.tv64 = 0;
496 static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base)
498 ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset;
499 ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset;
500 ktime_t *offs_tai = &base->clock_base[HRTIMER_BASE_TAI].offset;
502 return ktime_get_update_offsets_now(&base->clock_was_set_seq,
503 offs_real, offs_boot, offs_tai);
506 /* High resolution timer related functions */
507 #ifdef CONFIG_HIGH_RES_TIMERS
510 * High resolution timer enabled ?
512 static bool hrtimer_hres_enabled __read_mostly = true;
513 unsigned int hrtimer_resolution __read_mostly = LOW_RES_NSEC;
514 EXPORT_SYMBOL_GPL(hrtimer_resolution);
517 * Enable / Disable high resolution mode
519 static int __init setup_hrtimer_hres(char *str)
521 return (kstrtobool(str, &hrtimer_hres_enabled) == 0);
524 __setup("highres=", setup_hrtimer_hres);
527 * hrtimer_high_res_enabled - query, if the highres mode is enabled
529 static inline int hrtimer_is_hres_enabled(void)
531 return hrtimer_hres_enabled;
535 * Is the high resolution mode active ?
537 static inline int __hrtimer_hres_active(struct hrtimer_cpu_base *cpu_base)
539 return cpu_base->hres_active;
542 static inline int hrtimer_hres_active(void)
544 return __hrtimer_hres_active(this_cpu_ptr(&hrtimer_bases));
548 * Reprogram the event source with checking both queues for the
550 * Called with interrupts disabled and base->lock held
553 hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
555 ktime_t expires_next;
557 if (!cpu_base->hres_active)
560 expires_next = __hrtimer_get_next_event(cpu_base);
562 if (skip_equal && expires_next.tv64 == cpu_base->expires_next.tv64)
565 cpu_base->expires_next.tv64 = expires_next.tv64;
568 * If a hang was detected in the last timer interrupt then we
569 * leave the hang delay active in the hardware. We want the
570 * system to make progress. That also prevents the following
572 * T1 expires 50ms from now
573 * T2 expires 5s from now
575 * T1 is removed, so this code is called and would reprogram
576 * the hardware to 5s from now. Any hrtimer_start after that
577 * will not reprogram the hardware due to hang_detected being
578 * set. So we'd effectivly block all timers until the T2 event
581 if (cpu_base->hang_detected)
584 tick_program_event(cpu_base->expires_next, 1);
588 * When a timer is enqueued and expires earlier than the already enqueued
589 * timers, we have to check, whether it expires earlier than the timer for
590 * which the clock event device was armed.
592 * Called with interrupts disabled and base->cpu_base.lock held
594 static void hrtimer_reprogram(struct hrtimer *timer,
595 struct hrtimer_clock_base *base)
597 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
598 ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
600 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
603 * If the timer is not on the current cpu, we cannot reprogram
604 * the other cpus clock event device.
606 if (base->cpu_base != cpu_base)
610 * If the hrtimer interrupt is running, then it will
611 * reevaluate the clock bases and reprogram the clock event
612 * device. The callbacks are always executed in hard interrupt
613 * context so we don't need an extra check for a running
616 if (cpu_base->in_hrtirq)
620 * CLOCK_REALTIME timer might be requested with an absolute
621 * expiry time which is less than base->offset. Set it to 0.
623 if (expires.tv64 < 0)
626 if (expires.tv64 >= cpu_base->expires_next.tv64)
629 /* Update the pointer to the next expiring timer */
630 cpu_base->next_timer = timer;
633 * If a hang was detected in the last timer interrupt then we
634 * do not schedule a timer which is earlier than the expiry
635 * which we enforced in the hang detection. We want the system
638 if (cpu_base->hang_detected)
642 * Program the timer hardware. We enforce the expiry for
643 * events which are already in the past.
645 cpu_base->expires_next = expires;
646 tick_program_event(expires, 1);
650 * Initialize the high resolution related parts of cpu_base
652 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
654 base->expires_next.tv64 = KTIME_MAX;
655 base->hang_detected = 0;
656 base->hres_active = 0;
657 base->next_timer = NULL;
661 * Retrigger next event is called after clock was set
663 * Called with interrupts disabled via on_each_cpu()
665 static void retrigger_next_event(void *arg)
667 struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
669 if (!base->hres_active)
672 raw_spin_lock(&base->lock);
673 hrtimer_update_base(base);
674 hrtimer_force_reprogram(base, 0);
675 raw_spin_unlock(&base->lock);
679 * Switch to high resolution mode
681 static void hrtimer_switch_to_hres(void)
683 struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
685 if (tick_init_highres()) {
686 printk(KERN_WARNING "Could not switch to high resolution "
687 "mode on CPU %d\n", base->cpu);
690 base->hres_active = 1;
691 hrtimer_resolution = HIGH_RES_NSEC;
693 tick_setup_sched_timer();
694 /* "Retrigger" the interrupt to get things going */
695 retrigger_next_event(NULL);
698 static void clock_was_set_work(struct work_struct *work)
703 static DECLARE_WORK(hrtimer_work, clock_was_set_work);
706 * Called from timekeeping and resume code to reprogram the hrtimer
707 * interrupt device on all cpus.
709 void clock_was_set_delayed(void)
711 schedule_work(&hrtimer_work);
716 static inline int __hrtimer_hres_active(struct hrtimer_cpu_base *b) { return 0; }
717 static inline int hrtimer_hres_active(void) { return 0; }
718 static inline int hrtimer_is_hres_enabled(void) { return 0; }
719 static inline void hrtimer_switch_to_hres(void) { }
721 hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { }
722 static inline int hrtimer_reprogram(struct hrtimer *timer,
723 struct hrtimer_clock_base *base)
727 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { }
728 static inline void retrigger_next_event(void *arg) { }
730 #endif /* CONFIG_HIGH_RES_TIMERS */
733 * Clock realtime was set
735 * Change the offset of the realtime clock vs. the monotonic
738 * We might have to reprogram the high resolution timer interrupt. On
739 * SMP we call the architecture specific code to retrigger _all_ high
740 * resolution timer interrupts. On UP we just disable interrupts and
741 * call the high resolution interrupt code.
743 void clock_was_set(void)
745 #ifdef CONFIG_HIGH_RES_TIMERS
746 /* Retrigger the CPU local events everywhere */
747 on_each_cpu(retrigger_next_event, NULL, 1);
749 timerfd_clock_was_set();
753 * During resume we might have to reprogram the high resolution timer
754 * interrupt on all online CPUs. However, all other CPUs will be
755 * stopped with IRQs interrupts disabled so the clock_was_set() call
758 void hrtimers_resume(void)
760 WARN_ONCE(!irqs_disabled(),
761 KERN_INFO "hrtimers_resume() called with IRQs enabled!");
763 /* Retrigger on the local CPU */
764 retrigger_next_event(NULL);
765 /* And schedule a retrigger for all others */
766 clock_was_set_delayed();
769 static inline void timer_stats_hrtimer_set_start_info(struct hrtimer *timer)
771 #ifdef CONFIG_TIMER_STATS
772 if (timer->start_site)
774 timer->start_site = __builtin_return_address(0);
775 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
776 timer->start_pid = current->pid;
780 static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer *timer)
782 #ifdef CONFIG_TIMER_STATS
783 timer->start_site = NULL;
787 static inline void timer_stats_account_hrtimer(struct hrtimer *timer)
789 #ifdef CONFIG_TIMER_STATS
790 if (likely(!timer_stats_active))
792 timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
793 timer->function, timer->start_comm, 0);
798 * Counterpart to lock_hrtimer_base above:
801 void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
803 raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
807 * hrtimer_forward - forward the timer expiry
808 * @timer: hrtimer to forward
809 * @now: forward past this time
810 * @interval: the interval to forward
812 * Forward the timer expiry so it will expire in the future.
813 * Returns the number of overruns.
815 * Can be safely called from the callback function of @timer. If
816 * called from other contexts @timer must neither be enqueued nor
817 * running the callback and the caller needs to take care of
820 * Note: This only updates the timer expiry value and does not requeue
823 u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
828 delta = ktime_sub(now, hrtimer_get_expires(timer));
833 if (WARN_ON(timer->state & HRTIMER_STATE_ENQUEUED))
836 if (interval.tv64 < hrtimer_resolution)
837 interval.tv64 = hrtimer_resolution;
839 if (unlikely(delta.tv64 >= interval.tv64)) {
840 s64 incr = ktime_to_ns(interval);
842 orun = ktime_divns(delta, incr);
843 hrtimer_add_expires_ns(timer, incr * orun);
844 if (hrtimer_get_expires_tv64(timer) > now.tv64)
847 * This (and the ktime_add() below) is the
848 * correction for exact:
852 hrtimer_add_expires(timer, interval);
856 EXPORT_SYMBOL_GPL(hrtimer_forward);
859 * enqueue_hrtimer - internal function to (re)start a timer
861 * The timer is inserted in expiry order. Insertion into the
862 * red black tree is O(log(n)). Must hold the base lock.
864 * Returns 1 when the new timer is the leftmost timer in the tree.
866 static int enqueue_hrtimer(struct hrtimer *timer,
867 struct hrtimer_clock_base *base)
869 debug_activate(timer);
871 base->cpu_base->active_bases |= 1 << base->index;
873 /* Pairs with the lockless read in hrtimer_is_queued() */
874 WRITE_ONCE(timer->state, HRTIMER_STATE_ENQUEUED);
876 return timerqueue_add(&base->active, &timer->node);
880 * __remove_hrtimer - internal function to remove a timer
882 * Caller must hold the base lock.
884 * High resolution timer mode reprograms the clock event device when the
885 * timer is the one which expires next. The caller can disable this by setting
886 * reprogram to zero. This is useful, when the context does a reprogramming
887 * anyway (e.g. timer interrupt)
889 static void __remove_hrtimer(struct hrtimer *timer,
890 struct hrtimer_clock_base *base,
891 u8 newstate, int reprogram)
893 struct hrtimer_cpu_base *cpu_base = base->cpu_base;
894 u8 state = timer->state;
896 /* Pairs with the lockless read in hrtimer_is_queued() */
897 WRITE_ONCE(timer->state, newstate);
898 if (!(state & HRTIMER_STATE_ENQUEUED))
901 if (!timerqueue_del(&base->active, &timer->node))
902 cpu_base->active_bases &= ~(1 << base->index);
904 #ifdef CONFIG_HIGH_RES_TIMERS
906 * Note: If reprogram is false we do not update
907 * cpu_base->next_timer. This happens when we remove the first
908 * timer on a remote cpu. No harm as we never dereference
909 * cpu_base->next_timer. So the worst thing what can happen is
910 * an superflous call to hrtimer_force_reprogram() on the
911 * remote cpu later on if the same timer gets enqueued again.
913 if (reprogram && timer == cpu_base->next_timer)
914 hrtimer_force_reprogram(cpu_base, 1);
919 * remove hrtimer, called with base lock held
922 remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base, bool restart)
924 u8 state = timer->state;
926 if (state & HRTIMER_STATE_ENQUEUED) {
930 * Remove the timer and force reprogramming when high
931 * resolution mode is active and the timer is on the current
932 * CPU. If we remove a timer on another CPU, reprogramming is
933 * skipped. The interrupt event on this CPU is fired and
934 * reprogramming happens in the interrupt handler. This is a
935 * rare case and less expensive than a smp call.
937 debug_deactivate(timer);
938 timer_stats_hrtimer_clear_start_info(timer);
939 reprogram = base->cpu_base == this_cpu_ptr(&hrtimer_bases);
942 state = HRTIMER_STATE_INACTIVE;
944 __remove_hrtimer(timer, base, state, reprogram);
950 static inline ktime_t hrtimer_update_lowres(struct hrtimer *timer, ktime_t tim,
951 const enum hrtimer_mode mode)
953 #ifdef CONFIG_TIME_LOW_RES
955 * CONFIG_TIME_LOW_RES indicates that the system has no way to return
956 * granular time values. For relative timers we add hrtimer_resolution
957 * (i.e. one jiffie) to prevent short timeouts.
959 timer->is_rel = mode & HRTIMER_MODE_REL;
961 tim = ktime_add_safe(tim, ktime_set(0, hrtimer_resolution));
967 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
968 * @timer: the timer to be added
970 * @delta_ns: "slack" range for the timer
971 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
972 * relative (HRTIMER_MODE_REL)
974 void hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
975 u64 delta_ns, const enum hrtimer_mode mode)
977 struct hrtimer_clock_base *base, *new_base;
981 base = lock_hrtimer_base(timer, &flags);
983 /* Remove an active timer from the queue: */
984 remove_hrtimer(timer, base, true);
986 if (mode & HRTIMER_MODE_REL)
987 tim = ktime_add_safe(tim, base->get_time());
989 tim = hrtimer_update_lowres(timer, tim, mode);
991 hrtimer_set_expires_range_ns(timer, tim, delta_ns);
993 /* Switch the timer base, if necessary: */
994 new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED);
996 timer_stats_hrtimer_set_start_info(timer);
998 leftmost = enqueue_hrtimer(timer, new_base);
1002 if (!hrtimer_is_hres_active(timer)) {
1004 * Kick to reschedule the next tick to handle the new timer
1005 * on dynticks target.
1007 if (new_base->cpu_base->nohz_active)
1008 wake_up_nohz_cpu(new_base->cpu_base->cpu);
1010 hrtimer_reprogram(timer, new_base);
1013 unlock_hrtimer_base(timer, &flags);
1015 EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
1018 * hrtimer_try_to_cancel - try to deactivate a timer
1019 * @timer: hrtimer to stop
1022 * 0 when the timer was not active
1023 * 1 when the timer was active
1024 * -1 when the timer is currently excuting the callback function and
1027 int hrtimer_try_to_cancel(struct hrtimer *timer)
1029 struct hrtimer_clock_base *base;
1030 unsigned long flags;
1034 * Check lockless first. If the timer is not active (neither
1035 * enqueued nor running the callback, nothing to do here. The
1036 * base lock does not serialize against a concurrent enqueue,
1037 * so we can avoid taking it.
1039 if (!hrtimer_active(timer))
1042 base = lock_hrtimer_base(timer, &flags);
1044 if (!hrtimer_callback_running(timer))
1045 ret = remove_hrtimer(timer, base, false);
1047 unlock_hrtimer_base(timer, &flags);
1052 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1055 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1056 * @timer: the timer to be cancelled
1059 * 0 when the timer was not active
1060 * 1 when the timer was active
1062 int hrtimer_cancel(struct hrtimer *timer)
1065 int ret = hrtimer_try_to_cancel(timer);
1072 EXPORT_SYMBOL_GPL(hrtimer_cancel);
1075 * hrtimer_get_remaining - get remaining time for the timer
1076 * @timer: the timer to read
1077 * @adjust: adjust relative timers when CONFIG_TIME_LOW_RES=y
1079 ktime_t __hrtimer_get_remaining(const struct hrtimer *timer, bool adjust)
1081 unsigned long flags;
1084 lock_hrtimer_base(timer, &flags);
1085 if (IS_ENABLED(CONFIG_TIME_LOW_RES) && adjust)
1086 rem = hrtimer_expires_remaining_adjusted(timer);
1088 rem = hrtimer_expires_remaining(timer);
1089 unlock_hrtimer_base(timer, &flags);
1093 EXPORT_SYMBOL_GPL(__hrtimer_get_remaining);
1095 #ifdef CONFIG_NO_HZ_COMMON
1097 * hrtimer_get_next_event - get the time until next expiry event
1099 * Returns the next expiry time or KTIME_MAX if no timer is pending.
1101 u64 hrtimer_get_next_event(void)
1103 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1104 u64 expires = KTIME_MAX;
1105 unsigned long flags;
1107 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1109 if (!__hrtimer_hres_active(cpu_base))
1110 expires = __hrtimer_get_next_event(cpu_base).tv64;
1112 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1118 static inline int hrtimer_clockid_to_base(clockid_t clock_id)
1120 if (likely(clock_id < MAX_CLOCKS)) {
1121 int base = hrtimer_clock_to_base_table[clock_id];
1123 if (likely(base != HRTIMER_MAX_CLOCK_BASES))
1126 WARN(1, "Invalid clockid %d. Using MONOTONIC\n", clock_id);
1127 return HRTIMER_BASE_MONOTONIC;
1130 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1131 enum hrtimer_mode mode)
1133 struct hrtimer_cpu_base *cpu_base;
1136 memset(timer, 0, sizeof(struct hrtimer));
1138 cpu_base = raw_cpu_ptr(&hrtimer_bases);
1141 * POSIX magic: Relative CLOCK_REALTIME timers are not affected by
1142 * clock modifications, so they needs to become CLOCK_MONOTONIC to
1143 * ensure POSIX compliance.
1145 if (clock_id == CLOCK_REALTIME && mode & HRTIMER_MODE_REL)
1146 clock_id = CLOCK_MONOTONIC;
1148 base = hrtimer_clockid_to_base(clock_id);
1149 timer->base = &cpu_base->clock_base[base];
1150 timerqueue_init(&timer->node);
1152 #ifdef CONFIG_TIMER_STATS
1153 timer->start_site = NULL;
1154 timer->start_pid = -1;
1155 memset(timer->start_comm, 0, TASK_COMM_LEN);
1160 * hrtimer_init - initialize a timer to the given clock
1161 * @timer: the timer to be initialized
1162 * @clock_id: the clock to be used
1163 * @mode: timer mode abs/rel
1165 void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1166 enum hrtimer_mode mode)
1168 debug_init(timer, clock_id, mode);
1169 __hrtimer_init(timer, clock_id, mode);
1171 EXPORT_SYMBOL_GPL(hrtimer_init);
1174 * A timer is active, when it is enqueued into the rbtree or the
1175 * callback function is running or it's in the state of being migrated
1178 * It is important for this function to not return a false negative.
1180 bool hrtimer_active(const struct hrtimer *timer)
1182 struct hrtimer_cpu_base *cpu_base;
1186 cpu_base = READ_ONCE(timer->base->cpu_base);
1187 seq = raw_read_seqcount_begin(&cpu_base->seq);
1189 if (timer->state != HRTIMER_STATE_INACTIVE ||
1190 cpu_base->running == timer)
1193 } while (read_seqcount_retry(&cpu_base->seq, seq) ||
1194 cpu_base != READ_ONCE(timer->base->cpu_base));
1198 EXPORT_SYMBOL_GPL(hrtimer_active);
1201 * The write_seqcount_barrier()s in __run_hrtimer() split the thing into 3
1202 * distinct sections:
1204 * - queued: the timer is queued
1205 * - callback: the timer is being ran
1206 * - post: the timer is inactive or (re)queued
1208 * On the read side we ensure we observe timer->state and cpu_base->running
1209 * from the same section, if anything changed while we looked at it, we retry.
1210 * This includes timer->base changing because sequence numbers alone are
1211 * insufficient for that.
1213 * The sequence numbers are required because otherwise we could still observe
1214 * a false negative if the read side got smeared over multiple consequtive
1215 * __run_hrtimer() invocations.
1218 static void __run_hrtimer(struct hrtimer_cpu_base *cpu_base,
1219 struct hrtimer_clock_base *base,
1220 struct hrtimer *timer, ktime_t *now)
1222 enum hrtimer_restart (*fn)(struct hrtimer *);
1225 lockdep_assert_held(&cpu_base->lock);
1227 debug_deactivate(timer);
1228 cpu_base->running = timer;
1231 * Separate the ->running assignment from the ->state assignment.
1233 * As with a regular write barrier, this ensures the read side in
1234 * hrtimer_active() cannot observe cpu_base->running == NULL &&
1235 * timer->state == INACTIVE.
1237 raw_write_seqcount_barrier(&cpu_base->seq);
1239 __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE, 0);
1240 timer_stats_account_hrtimer(timer);
1241 fn = timer->function;
1244 * Clear the 'is relative' flag for the TIME_LOW_RES case. If the
1245 * timer is restarted with a period then it becomes an absolute
1246 * timer. If its not restarted it does not matter.
1248 if (IS_ENABLED(CONFIG_TIME_LOW_RES))
1249 timer->is_rel = false;
1252 * Because we run timers from hardirq context, there is no chance
1253 * they get migrated to another cpu, therefore its safe to unlock
1256 raw_spin_unlock(&cpu_base->lock);
1257 trace_hrtimer_expire_entry(timer, now);
1258 restart = fn(timer);
1259 trace_hrtimer_expire_exit(timer);
1260 raw_spin_lock(&cpu_base->lock);
1263 * Note: We clear the running state after enqueue_hrtimer and
1264 * we do not reprogram the event hardware. Happens either in
1265 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1267 * Note: Because we dropped the cpu_base->lock above,
1268 * hrtimer_start_range_ns() can have popped in and enqueued the timer
1271 if (restart != HRTIMER_NORESTART &&
1272 !(timer->state & HRTIMER_STATE_ENQUEUED))
1273 enqueue_hrtimer(timer, base);
1276 * Separate the ->running assignment from the ->state assignment.
1278 * As with a regular write barrier, this ensures the read side in
1279 * hrtimer_active() cannot observe cpu_base->running == NULL &&
1280 * timer->state == INACTIVE.
1282 raw_write_seqcount_barrier(&cpu_base->seq);
1284 WARN_ON_ONCE(cpu_base->running != timer);
1285 cpu_base->running = NULL;
1288 static void __hrtimer_run_queues(struct hrtimer_cpu_base *cpu_base, ktime_t now)
1290 struct hrtimer_clock_base *base = cpu_base->clock_base;
1291 unsigned int active = cpu_base->active_bases;
1293 for (; active; base++, active >>= 1) {
1294 struct timerqueue_node *node;
1297 if (!(active & 0x01))
1300 basenow = ktime_add(now, base->offset);
1302 while ((node = timerqueue_getnext(&base->active))) {
1303 struct hrtimer *timer;
1305 timer = container_of(node, struct hrtimer, node);
1308 * The immediate goal for using the softexpires is
1309 * minimizing wakeups, not running timers at the
1310 * earliest interrupt after their soft expiration.
1311 * This allows us to avoid using a Priority Search
1312 * Tree, which can answer a stabbing querry for
1313 * overlapping intervals and instead use the simple
1314 * BST we already have.
1315 * We don't add extra wakeups by delaying timers that
1316 * are right-of a not yet expired timer, because that
1317 * timer will have to trigger a wakeup anyway.
1319 if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer))
1322 __run_hrtimer(cpu_base, base, timer, &basenow);
1327 #ifdef CONFIG_HIGH_RES_TIMERS
1330 * High resolution timer interrupt
1331 * Called with interrupts disabled
1333 void hrtimer_interrupt(struct clock_event_device *dev)
1335 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1336 ktime_t expires_next, now, entry_time, delta;
1339 BUG_ON(!cpu_base->hres_active);
1340 cpu_base->nr_events++;
1341 dev->next_event.tv64 = KTIME_MAX;
1343 raw_spin_lock(&cpu_base->lock);
1344 entry_time = now = hrtimer_update_base(cpu_base);
1346 cpu_base->in_hrtirq = 1;
1348 * We set expires_next to KTIME_MAX here with cpu_base->lock
1349 * held to prevent that a timer is enqueued in our queue via
1350 * the migration code. This does not affect enqueueing of
1351 * timers which run their callback and need to be requeued on
1354 cpu_base->expires_next.tv64 = KTIME_MAX;
1356 __hrtimer_run_queues(cpu_base, now);
1358 /* Reevaluate the clock bases for the next expiry */
1359 expires_next = __hrtimer_get_next_event(cpu_base);
1361 * Store the new expiry value so the migration code can verify
1364 cpu_base->expires_next = expires_next;
1365 cpu_base->in_hrtirq = 0;
1366 raw_spin_unlock(&cpu_base->lock);
1368 /* Reprogramming necessary ? */
1369 if (!tick_program_event(expires_next, 0)) {
1370 cpu_base->hang_detected = 0;
1375 * The next timer was already expired due to:
1377 * - long lasting callbacks
1378 * - being scheduled away when running in a VM
1380 * We need to prevent that we loop forever in the hrtimer
1381 * interrupt routine. We give it 3 attempts to avoid
1382 * overreacting on some spurious event.
1384 * Acquire base lock for updating the offsets and retrieving
1387 raw_spin_lock(&cpu_base->lock);
1388 now = hrtimer_update_base(cpu_base);
1389 cpu_base->nr_retries++;
1393 * Give the system a chance to do something else than looping
1394 * here. We stored the entry time, so we know exactly how long
1395 * we spent here. We schedule the next event this amount of
1398 cpu_base->nr_hangs++;
1399 cpu_base->hang_detected = 1;
1400 raw_spin_unlock(&cpu_base->lock);
1401 delta = ktime_sub(now, entry_time);
1402 if ((unsigned int)delta.tv64 > cpu_base->max_hang_time)
1403 cpu_base->max_hang_time = (unsigned int) delta.tv64;
1405 * Limit it to a sensible value as we enforce a longer
1406 * delay. Give the CPU at least 100ms to catch up.
1408 if (delta.tv64 > 100 * NSEC_PER_MSEC)
1409 expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1411 expires_next = ktime_add(now, delta);
1412 tick_program_event(expires_next, 1);
1413 printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n",
1414 ktime_to_ns(delta));
1418 * local version of hrtimer_peek_ahead_timers() called with interrupts
1421 static inline void __hrtimer_peek_ahead_timers(void)
1423 struct tick_device *td;
1425 if (!hrtimer_hres_active())
1428 td = this_cpu_ptr(&tick_cpu_device);
1429 if (td && td->evtdev)
1430 hrtimer_interrupt(td->evtdev);
1433 #else /* CONFIG_HIGH_RES_TIMERS */
1435 static inline void __hrtimer_peek_ahead_timers(void) { }
1437 #endif /* !CONFIG_HIGH_RES_TIMERS */
1440 * Called from run_local_timers in hardirq context every jiffy
1442 void hrtimer_run_queues(void)
1444 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1447 if (__hrtimer_hres_active(cpu_base))
1451 * This _is_ ugly: We have to check periodically, whether we
1452 * can switch to highres and / or nohz mode. The clocksource
1453 * switch happens with xtime_lock held. Notification from
1454 * there only sets the check bit in the tick_oneshot code,
1455 * otherwise we might deadlock vs. xtime_lock.
1457 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) {
1458 hrtimer_switch_to_hres();
1462 raw_spin_lock(&cpu_base->lock);
1463 now = hrtimer_update_base(cpu_base);
1464 __hrtimer_run_queues(cpu_base, now);
1465 raw_spin_unlock(&cpu_base->lock);
1469 * Sleep related functions:
1471 static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1473 struct hrtimer_sleeper *t =
1474 container_of(timer, struct hrtimer_sleeper, timer);
1475 struct task_struct *task = t->task;
1479 wake_up_process(task);
1481 return HRTIMER_NORESTART;
1484 void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1486 sl->timer.function = hrtimer_wakeup;
1489 EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
1491 static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1493 hrtimer_init_sleeper(t, current);
1496 set_current_state(TASK_INTERRUPTIBLE);
1497 hrtimer_start_expires(&t->timer, mode);
1499 if (likely(t->task))
1500 freezable_schedule();
1502 hrtimer_cancel(&t->timer);
1503 mode = HRTIMER_MODE_ABS;
1505 } while (t->task && !signal_pending(current));
1507 __set_current_state(TASK_RUNNING);
1509 return t->task == NULL;
1512 static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp)
1514 struct timespec rmt;
1517 rem = hrtimer_expires_remaining(timer);
1520 rmt = ktime_to_timespec(rem);
1522 if (copy_to_user(rmtp, &rmt, sizeof(*rmtp)))
1528 long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1530 struct hrtimer_sleeper t;
1531 struct timespec __user *rmtp;
1534 hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid,
1536 hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
1538 if (do_nanosleep(&t, HRTIMER_MODE_ABS))
1541 rmtp = restart->nanosleep.rmtp;
1543 ret = update_rmtp(&t.timer, rmtp);
1548 /* The other values in restart are already filled in */
1549 ret = -ERESTART_RESTARTBLOCK;
1551 destroy_hrtimer_on_stack(&t.timer);
1555 long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
1556 const enum hrtimer_mode mode, const clockid_t clockid)
1558 struct restart_block *restart;
1559 struct hrtimer_sleeper t;
1563 slack = current->timer_slack_ns;
1564 if (dl_task(current) || rt_task(current))
1567 hrtimer_init_on_stack(&t.timer, clockid, mode);
1568 hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack);
1569 if (do_nanosleep(&t, mode))
1572 /* Absolute timers do not update the rmtp value and restart: */
1573 if (mode == HRTIMER_MODE_ABS) {
1574 ret = -ERESTARTNOHAND;
1579 ret = update_rmtp(&t.timer, rmtp);
1584 restart = ¤t->restart_block;
1585 restart->nanosleep.clockid = t.timer.base->clockid;
1586 restart->nanosleep.rmtp = rmtp;
1587 restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
1588 set_restart_fn(restart, hrtimer_nanosleep_restart);
1590 ret = -ERESTART_RESTARTBLOCK;
1592 destroy_hrtimer_on_stack(&t.timer);
1596 SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp,
1597 struct timespec __user *, rmtp)
1601 if (copy_from_user(&tu, rqtp, sizeof(tu)))
1604 if (!timespec_valid(&tu))
1607 return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1611 * Functions related to boot-time initialization:
1613 int hrtimers_prepare_cpu(unsigned int cpu)
1615 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1618 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1619 cpu_base->clock_base[i].cpu_base = cpu_base;
1620 timerqueue_init_head(&cpu_base->clock_base[i].active);
1623 cpu_base->active_bases = 0;
1624 cpu_base->cpu = cpu;
1625 hrtimer_init_hres(cpu_base);
1629 #ifdef CONFIG_HOTPLUG_CPU
1631 static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1632 struct hrtimer_clock_base *new_base)
1634 struct hrtimer *timer;
1635 struct timerqueue_node *node;
1637 while ((node = timerqueue_getnext(&old_base->active))) {
1638 timer = container_of(node, struct hrtimer, node);
1639 BUG_ON(hrtimer_callback_running(timer));
1640 debug_deactivate(timer);
1643 * Mark it as ENQUEUED not INACTIVE otherwise the
1644 * timer could be seen as !active and just vanish away
1645 * under us on another CPU
1647 __remove_hrtimer(timer, old_base, HRTIMER_STATE_ENQUEUED, 0);
1648 timer->base = new_base;
1650 * Enqueue the timers on the new cpu. This does not
1651 * reprogram the event device in case the timer
1652 * expires before the earliest on this CPU, but we run
1653 * hrtimer_interrupt after we migrated everything to
1654 * sort out already expired timers and reprogram the
1657 enqueue_hrtimer(timer, new_base);
1661 int hrtimers_dead_cpu(unsigned int scpu)
1663 struct hrtimer_cpu_base *old_base, *new_base;
1666 BUG_ON(cpu_online(scpu));
1667 tick_cancel_sched_timer(scpu);
1669 local_irq_disable();
1670 old_base = &per_cpu(hrtimer_bases, scpu);
1671 new_base = this_cpu_ptr(&hrtimer_bases);
1673 * The caller is globally serialized and nobody else
1674 * takes two locks at once, deadlock is not possible.
1676 raw_spin_lock(&new_base->lock);
1677 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1679 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1680 migrate_hrtimer_list(&old_base->clock_base[i],
1681 &new_base->clock_base[i]);
1684 raw_spin_unlock(&old_base->lock);
1685 raw_spin_unlock(&new_base->lock);
1687 /* Check, if we got expired work to do */
1688 __hrtimer_peek_ahead_timers();
1693 #endif /* CONFIG_HOTPLUG_CPU */
1695 void __init hrtimers_init(void)
1697 hrtimers_prepare_cpu(smp_processor_id());
1701 * schedule_hrtimeout_range_clock - 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
1705 * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME
1708 schedule_hrtimeout_range_clock(ktime_t *expires, u64 delta,
1709 const enum hrtimer_mode mode, int clock)
1711 struct hrtimer_sleeper t;
1714 * Optimize when a zero timeout value is given. It does not
1715 * matter whether this is an absolute or a relative time.
1717 if (expires && !expires->tv64) {
1718 __set_current_state(TASK_RUNNING);
1723 * A NULL parameter means "infinite"
1730 hrtimer_init_on_stack(&t.timer, clock, mode);
1731 hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
1733 hrtimer_init_sleeper(&t, current);
1735 hrtimer_start_expires(&t.timer, mode);
1740 hrtimer_cancel(&t.timer);
1741 destroy_hrtimer_on_stack(&t.timer);
1743 __set_current_state(TASK_RUNNING);
1745 return !t.task ? 0 : -EINTR;
1749 * schedule_hrtimeout_range - sleep until timeout
1750 * @expires: timeout value (ktime_t)
1751 * @delta: slack in expires timeout (ktime_t)
1752 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1754 * Make the current task sleep until the given expiry time has
1755 * elapsed. The routine will return immediately unless
1756 * the current task state has been set (see set_current_state()).
1758 * The @delta argument gives the kernel the freedom to schedule the
1759 * actual wakeup to a time that is both power and performance friendly.
1760 * The kernel give the normal best effort behavior for "@expires+@delta",
1761 * but may decide to fire the timer earlier, but no earlier than @expires.
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_range(ktime_t *expires, u64 delta,
1777 const enum hrtimer_mode mode)
1779 return schedule_hrtimeout_range_clock(expires, delta, mode,
1782 EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
1785 * schedule_hrtimeout - sleep until timeout
1786 * @expires: timeout value (ktime_t)
1787 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1789 * Make the current task sleep until the given expiry time has
1790 * elapsed. The routine will return immediately unless
1791 * the current task state has been set (see set_current_state()).
1793 * You can set the task state as follows -
1795 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1796 * pass before the routine returns.
1798 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1799 * delivered to the current task.
1801 * The current task state is guaranteed to be TASK_RUNNING when this
1804 * Returns 0 when the timer has expired otherwise -EINTR
1806 int __sched schedule_hrtimeout(ktime_t *expires,
1807 const enum hrtimer_mode mode)
1809 return schedule_hrtimeout_range(expires, 0, mode);
1811 EXPORT_SYMBOL_GPL(schedule_hrtimeout);