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/interrupt.h>
41 #include <linux/tick.h>
42 #include <linux/seq_file.h>
43 #include <linux/err.h>
44 #include <linux/debugobjects.h>
45 #include <linux/sched/signal.h>
46 #include <linux/sched/sysctl.h>
47 #include <linux/sched/rt.h>
48 #include <linux/sched/deadline.h>
49 #include <linux/sched/nohz.h>
50 #include <linux/sched/debug.h>
51 #include <linux/timer.h>
52 #include <linux/freezer.h>
53 #include <linux/compat.h>
55 #include <linux/uaccess.h>
57 #include <trace/events/timer.h>
59 #include "tick-internal.h"
62 * Masks for selecting the soft and hard context timers from
65 #define MASK_SHIFT (HRTIMER_BASE_MONOTONIC_SOFT)
66 #define HRTIMER_ACTIVE_HARD ((1U << MASK_SHIFT) - 1)
67 #define HRTIMER_ACTIVE_SOFT (HRTIMER_ACTIVE_HARD << MASK_SHIFT)
68 #define HRTIMER_ACTIVE_ALL (HRTIMER_ACTIVE_SOFT | HRTIMER_ACTIVE_HARD)
73 * There are more clockids than hrtimer bases. Thus, we index
74 * into the timer bases by the hrtimer_base_type enum. When trying
75 * to reach a base using a clockid, hrtimer_clockid_to_base()
76 * is used to convert from clockid to the proper hrtimer_base_type.
78 DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
80 .lock = __RAW_SPIN_LOCK_UNLOCKED(hrtimer_bases.lock),
84 .index = HRTIMER_BASE_MONOTONIC,
85 .clockid = CLOCK_MONOTONIC,
86 .get_time = &ktime_get,
89 .index = HRTIMER_BASE_REALTIME,
90 .clockid = CLOCK_REALTIME,
91 .get_time = &ktime_get_real,
94 .index = HRTIMER_BASE_TAI,
96 .get_time = &ktime_get_clocktai,
99 .index = HRTIMER_BASE_MONOTONIC_SOFT,
100 .clockid = CLOCK_MONOTONIC,
101 .get_time = &ktime_get,
104 .index = HRTIMER_BASE_REALTIME_SOFT,
105 .clockid = CLOCK_REALTIME,
106 .get_time = &ktime_get_real,
109 .index = HRTIMER_BASE_TAI_SOFT,
110 .clockid = CLOCK_TAI,
111 .get_time = &ktime_get_clocktai,
116 static const int hrtimer_clock_to_base_table[MAX_CLOCKS] = {
117 /* Make sure we catch unsupported clockids */
118 [0 ... MAX_CLOCKS - 1] = HRTIMER_MAX_CLOCK_BASES,
120 [CLOCK_REALTIME] = HRTIMER_BASE_REALTIME,
121 [CLOCK_MONOTONIC] = HRTIMER_BASE_MONOTONIC,
122 [CLOCK_BOOTTIME] = HRTIMER_BASE_MONOTONIC,
123 [CLOCK_TAI] = HRTIMER_BASE_TAI,
127 * Functions and macros which are different for UP/SMP systems are kept in a
133 * We require the migration_base for lock_hrtimer_base()/switch_hrtimer_base()
134 * such that hrtimer_callback_running() can unconditionally dereference
135 * timer->base->cpu_base
137 static struct hrtimer_cpu_base migration_cpu_base = {
138 .clock_base = { { .cpu_base = &migration_cpu_base, }, },
141 #define migration_base migration_cpu_base.clock_base[0]
144 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
145 * means that all timers which are tied to this base via timer->base are
146 * locked, and the base itself is locked too.
148 * So __run_timers/migrate_timers can safely modify all timers which could
149 * be found on the lists/queues.
151 * When the timer's base is locked, and the timer removed from list, it is
152 * possible to set timer->base = &migration_base and drop the lock: the timer
156 struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
157 unsigned long *flags)
159 struct hrtimer_clock_base *base;
163 if (likely(base != &migration_base)) {
164 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
165 if (likely(base == timer->base))
167 /* The timer has migrated to another CPU: */
168 raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
175 * We do not migrate the timer when it is expiring before the next
176 * event on the target cpu. When high resolution is enabled, we cannot
177 * reprogram the target cpu hardware and we would cause it to fire
178 * late. To keep it simple, we handle the high resolution enabled and
179 * disabled case similar.
181 * Called with cpu_base->lock of target cpu held.
184 hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base)
188 expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset);
189 return expires < new_base->cpu_base->expires_next;
193 struct hrtimer_cpu_base *get_target_base(struct hrtimer_cpu_base *base,
196 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
197 if (static_branch_likely(&timers_migration_enabled) && !pinned)
198 return &per_cpu(hrtimer_bases, get_nohz_timer_target());
204 * We switch the timer base to a power-optimized selected CPU target,
206 * - NO_HZ_COMMON is enabled
207 * - timer migration is enabled
208 * - the timer callback is not running
209 * - the timer is not the first expiring timer on the new target
211 * If one of the above requirements is not fulfilled we move the timer
212 * to the current CPU or leave it on the previously assigned CPU if
213 * the timer callback is currently running.
215 static inline struct hrtimer_clock_base *
216 switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
219 struct hrtimer_cpu_base *new_cpu_base, *this_cpu_base;
220 struct hrtimer_clock_base *new_base;
221 int basenum = base->index;
223 this_cpu_base = this_cpu_ptr(&hrtimer_bases);
224 new_cpu_base = get_target_base(this_cpu_base, pinned);
226 new_base = &new_cpu_base->clock_base[basenum];
228 if (base != new_base) {
230 * We are trying to move timer to new_base.
231 * However we can't change timer's base while it is running,
232 * so we keep it on the same CPU. No hassle vs. reprogramming
233 * the event source in the high resolution case. The softirq
234 * code will take care of this when the timer function has
235 * completed. There is no conflict as we hold the lock until
236 * the timer is enqueued.
238 if (unlikely(hrtimer_callback_running(timer)))
241 /* See the comment in lock_hrtimer_base() */
242 timer->base = &migration_base;
243 raw_spin_unlock(&base->cpu_base->lock);
244 raw_spin_lock(&new_base->cpu_base->lock);
246 if (new_cpu_base != this_cpu_base &&
247 hrtimer_check_target(timer, new_base)) {
248 raw_spin_unlock(&new_base->cpu_base->lock);
249 raw_spin_lock(&base->cpu_base->lock);
250 new_cpu_base = this_cpu_base;
254 timer->base = new_base;
256 if (new_cpu_base != this_cpu_base &&
257 hrtimer_check_target(timer, new_base)) {
258 new_cpu_base = this_cpu_base;
265 #else /* CONFIG_SMP */
267 static inline struct hrtimer_clock_base *
268 lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
270 struct hrtimer_clock_base *base = timer->base;
272 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
277 # define switch_hrtimer_base(t, b, p) (b)
279 #endif /* !CONFIG_SMP */
282 * Functions for the union type storage format of ktime_t which are
283 * too large for inlining:
285 #if BITS_PER_LONG < 64
287 * Divide a ktime value by a nanosecond value
289 s64 __ktime_divns(const ktime_t kt, s64 div)
295 dclc = ktime_to_ns(kt);
296 tmp = dclc < 0 ? -dclc : dclc;
298 /* Make sure the divisor is less than 2^32: */
304 do_div(tmp, (unsigned long) div);
305 return dclc < 0 ? -tmp : tmp;
307 EXPORT_SYMBOL_GPL(__ktime_divns);
308 #endif /* BITS_PER_LONG >= 64 */
311 * Add two ktime values and do a safety check for overflow:
313 ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
315 ktime_t res = ktime_add_unsafe(lhs, rhs);
318 * We use KTIME_SEC_MAX here, the maximum timeout which we can
319 * return to user space in a timespec:
321 if (res < 0 || res < lhs || res < rhs)
322 res = ktime_set(KTIME_SEC_MAX, 0);
327 EXPORT_SYMBOL_GPL(ktime_add_safe);
329 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
331 static struct debug_obj_descr hrtimer_debug_descr;
333 static void *hrtimer_debug_hint(void *addr)
335 return ((struct hrtimer *) addr)->function;
339 * fixup_init is called when:
340 * - an active object is initialized
342 static bool hrtimer_fixup_init(void *addr, enum debug_obj_state state)
344 struct hrtimer *timer = addr;
347 case ODEBUG_STATE_ACTIVE:
348 hrtimer_cancel(timer);
349 debug_object_init(timer, &hrtimer_debug_descr);
357 * fixup_activate is called when:
358 * - an active object is activated
359 * - an unknown non-static object is activated
361 static bool hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
364 case ODEBUG_STATE_ACTIVE:
373 * fixup_free is called when:
374 * - an active object is freed
376 static bool hrtimer_fixup_free(void *addr, enum debug_obj_state state)
378 struct hrtimer *timer = addr;
381 case ODEBUG_STATE_ACTIVE:
382 hrtimer_cancel(timer);
383 debug_object_free(timer, &hrtimer_debug_descr);
390 static struct debug_obj_descr hrtimer_debug_descr = {
392 .debug_hint = hrtimer_debug_hint,
393 .fixup_init = hrtimer_fixup_init,
394 .fixup_activate = hrtimer_fixup_activate,
395 .fixup_free = hrtimer_fixup_free,
398 static inline void debug_hrtimer_init(struct hrtimer *timer)
400 debug_object_init(timer, &hrtimer_debug_descr);
403 static inline void debug_hrtimer_activate(struct hrtimer *timer,
404 enum hrtimer_mode mode)
406 debug_object_activate(timer, &hrtimer_debug_descr);
409 static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
411 debug_object_deactivate(timer, &hrtimer_debug_descr);
414 static inline void debug_hrtimer_free(struct hrtimer *timer)
416 debug_object_free(timer, &hrtimer_debug_descr);
419 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
420 enum hrtimer_mode mode);
422 void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
423 enum hrtimer_mode mode)
425 debug_object_init_on_stack(timer, &hrtimer_debug_descr);
426 __hrtimer_init(timer, clock_id, mode);
428 EXPORT_SYMBOL_GPL(hrtimer_init_on_stack);
430 void destroy_hrtimer_on_stack(struct hrtimer *timer)
432 debug_object_free(timer, &hrtimer_debug_descr);
434 EXPORT_SYMBOL_GPL(destroy_hrtimer_on_stack);
438 static inline void debug_hrtimer_init(struct hrtimer *timer) { }
439 static inline void debug_hrtimer_activate(struct hrtimer *timer,
440 enum hrtimer_mode mode) { }
441 static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
445 debug_init(struct hrtimer *timer, clockid_t clockid,
446 enum hrtimer_mode mode)
448 debug_hrtimer_init(timer);
449 trace_hrtimer_init(timer, clockid, mode);
452 static inline void debug_activate(struct hrtimer *timer,
453 enum hrtimer_mode mode)
455 debug_hrtimer_activate(timer, mode);
456 trace_hrtimer_start(timer, mode);
459 static inline void debug_deactivate(struct hrtimer *timer)
461 debug_hrtimer_deactivate(timer);
462 trace_hrtimer_cancel(timer);
465 static struct hrtimer_clock_base *
466 __next_base(struct hrtimer_cpu_base *cpu_base, unsigned int *active)
473 idx = __ffs(*active);
474 *active &= ~(1U << idx);
476 return &cpu_base->clock_base[idx];
479 #define for_each_active_base(base, cpu_base, active) \
480 while ((base = __next_base((cpu_base), &(active))))
482 static ktime_t __hrtimer_next_event_base(struct hrtimer_cpu_base *cpu_base,
483 const struct hrtimer *exclude,
485 ktime_t expires_next)
487 struct hrtimer_clock_base *base;
490 for_each_active_base(base, cpu_base, active) {
491 struct timerqueue_node *next;
492 struct hrtimer *timer;
494 next = timerqueue_getnext(&base->active);
495 timer = container_of(next, struct hrtimer, node);
496 if (timer == exclude) {
497 /* Get to the next timer in the queue. */
498 next = timerqueue_iterate_next(next);
502 timer = container_of(next, struct hrtimer, node);
504 expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
505 if (expires < expires_next) {
506 expires_next = expires;
508 /* Skip cpu_base update if a timer is being excluded. */
513 cpu_base->softirq_next_timer = timer;
515 cpu_base->next_timer = timer;
519 * clock_was_set() might have changed base->offset of any of
520 * the clock bases so the result might be negative. Fix it up
521 * to prevent a false positive in clockevents_program_event().
523 if (expires_next < 0)
529 * Recomputes cpu_base::*next_timer and returns the earliest expires_next but
530 * does not set cpu_base::*expires_next, that is done by hrtimer_reprogram.
532 * When a softirq is pending, we can ignore the HRTIMER_ACTIVE_SOFT bases,
533 * those timers will get run whenever the softirq gets handled, at the end of
534 * hrtimer_run_softirq(), hrtimer_update_softirq_timer() will re-add these bases.
536 * Therefore softirq values are those from the HRTIMER_ACTIVE_SOFT clock bases.
537 * The !softirq values are the minima across HRTIMER_ACTIVE_ALL, unless an actual
538 * softirq is pending, in which case they're the minima of HRTIMER_ACTIVE_HARD.
540 * @active_mask must be one of:
541 * - HRTIMER_ACTIVE_ALL,
542 * - HRTIMER_ACTIVE_SOFT, or
543 * - HRTIMER_ACTIVE_HARD.
546 __hrtimer_get_next_event(struct hrtimer_cpu_base *cpu_base, unsigned int active_mask)
549 struct hrtimer *next_timer = NULL;
550 ktime_t expires_next = KTIME_MAX;
552 if (!cpu_base->softirq_activated && (active_mask & HRTIMER_ACTIVE_SOFT)) {
553 active = cpu_base->active_bases & HRTIMER_ACTIVE_SOFT;
554 cpu_base->softirq_next_timer = NULL;
555 expires_next = __hrtimer_next_event_base(cpu_base, NULL,
558 next_timer = cpu_base->softirq_next_timer;
561 if (active_mask & HRTIMER_ACTIVE_HARD) {
562 active = cpu_base->active_bases & HRTIMER_ACTIVE_HARD;
563 cpu_base->next_timer = next_timer;
564 expires_next = __hrtimer_next_event_base(cpu_base, NULL, active,
571 static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base)
573 ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset;
574 ktime_t *offs_tai = &base->clock_base[HRTIMER_BASE_TAI].offset;
576 ktime_t now = ktime_get_update_offsets_now(&base->clock_was_set_seq,
577 offs_real, offs_tai);
579 base->clock_base[HRTIMER_BASE_REALTIME_SOFT].offset = *offs_real;
580 base->clock_base[HRTIMER_BASE_TAI_SOFT].offset = *offs_tai;
586 * Is the high resolution mode active ?
588 static inline int __hrtimer_hres_active(struct hrtimer_cpu_base *cpu_base)
590 return IS_ENABLED(CONFIG_HIGH_RES_TIMERS) ?
591 cpu_base->hres_active : 0;
594 static inline int hrtimer_hres_active(void)
596 return __hrtimer_hres_active(this_cpu_ptr(&hrtimer_bases));
600 * Reprogram the event source with checking both queues for the
602 * Called with interrupts disabled and base->lock held
605 hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
607 ktime_t expires_next;
610 * Find the current next expiration time.
612 expires_next = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_ALL);
614 if (cpu_base->next_timer && cpu_base->next_timer->is_soft) {
616 * When the softirq is activated, hrtimer has to be
617 * programmed with the first hard hrtimer because soft
618 * timer interrupt could occur too late.
620 if (cpu_base->softirq_activated)
621 expires_next = __hrtimer_get_next_event(cpu_base,
622 HRTIMER_ACTIVE_HARD);
624 cpu_base->softirq_expires_next = expires_next;
627 if (skip_equal && expires_next == cpu_base->expires_next)
630 cpu_base->expires_next = expires_next;
633 * If hres is not active, hardware does not have to be
636 * If a hang was detected in the last timer interrupt then we
637 * leave the hang delay active in the hardware. We want the
638 * system to make progress. That also prevents the following
640 * T1 expires 50ms from now
641 * T2 expires 5s from now
643 * T1 is removed, so this code is called and would reprogram
644 * the hardware to 5s from now. Any hrtimer_start after that
645 * will not reprogram the hardware due to hang_detected being
646 * set. So we'd effectivly block all timers until the T2 event
649 if (!__hrtimer_hres_active(cpu_base) || cpu_base->hang_detected)
652 tick_program_event(cpu_base->expires_next, 1);
655 /* High resolution timer related functions */
656 #ifdef CONFIG_HIGH_RES_TIMERS
659 * High resolution timer enabled ?
661 static bool hrtimer_hres_enabled __read_mostly = true;
662 unsigned int hrtimer_resolution __read_mostly = LOW_RES_NSEC;
663 EXPORT_SYMBOL_GPL(hrtimer_resolution);
666 * Enable / Disable high resolution mode
668 static int __init setup_hrtimer_hres(char *str)
670 return (kstrtobool(str, &hrtimer_hres_enabled) == 0);
673 __setup("highres=", setup_hrtimer_hres);
676 * hrtimer_high_res_enabled - query, if the highres mode is enabled
678 static inline int hrtimer_is_hres_enabled(void)
680 return hrtimer_hres_enabled;
684 * Retrigger next event is called after clock was set
686 * Called with interrupts disabled via on_each_cpu()
688 static void retrigger_next_event(void *arg)
690 struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
692 if (!__hrtimer_hres_active(base))
695 raw_spin_lock(&base->lock);
696 hrtimer_update_base(base);
697 hrtimer_force_reprogram(base, 0);
698 raw_spin_unlock(&base->lock);
702 * Switch to high resolution mode
704 static void hrtimer_switch_to_hres(void)
706 struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
708 if (tick_init_highres()) {
709 printk(KERN_WARNING "Could not switch to high resolution "
710 "mode on CPU %d\n", base->cpu);
713 base->hres_active = 1;
714 hrtimer_resolution = HIGH_RES_NSEC;
716 tick_setup_sched_timer();
717 /* "Retrigger" the interrupt to get things going */
718 retrigger_next_event(NULL);
721 static void clock_was_set_work(struct work_struct *work)
726 static DECLARE_WORK(hrtimer_work, clock_was_set_work);
729 * Called from timekeeping and resume code to reprogram the hrtimer
730 * interrupt device on all cpus.
732 void clock_was_set_delayed(void)
734 schedule_work(&hrtimer_work);
739 static inline int hrtimer_is_hres_enabled(void) { return 0; }
740 static inline void hrtimer_switch_to_hres(void) { }
741 static inline void retrigger_next_event(void *arg) { }
743 #endif /* CONFIG_HIGH_RES_TIMERS */
746 * When a timer is enqueued and expires earlier than the already enqueued
747 * timers, we have to check, whether it expires earlier than the timer for
748 * which the clock event device was armed.
750 * Called with interrupts disabled and base->cpu_base.lock held
752 static void hrtimer_reprogram(struct hrtimer *timer, bool reprogram)
754 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
755 struct hrtimer_clock_base *base = timer->base;
756 ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
758 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
761 * CLOCK_REALTIME timer might be requested with an absolute
762 * expiry time which is less than base->offset. Set it to 0.
767 if (timer->is_soft) {
769 * soft hrtimer could be started on a remote CPU. In this
770 * case softirq_expires_next needs to be updated on the
771 * remote CPU. The soft hrtimer will not expire before the
772 * first hard hrtimer on the remote CPU -
773 * hrtimer_check_target() prevents this case.
775 struct hrtimer_cpu_base *timer_cpu_base = base->cpu_base;
777 if (timer_cpu_base->softirq_activated)
780 if (!ktime_before(expires, timer_cpu_base->softirq_expires_next))
783 timer_cpu_base->softirq_next_timer = timer;
784 timer_cpu_base->softirq_expires_next = expires;
786 if (!ktime_before(expires, timer_cpu_base->expires_next) ||
792 * If the timer is not on the current cpu, we cannot reprogram
793 * the other cpus clock event device.
795 if (base->cpu_base != cpu_base)
799 * If the hrtimer interrupt is running, then it will
800 * reevaluate the clock bases and reprogram the clock event
801 * device. The callbacks are always executed in hard interrupt
802 * context so we don't need an extra check for a running
805 if (cpu_base->in_hrtirq)
808 if (expires >= cpu_base->expires_next)
811 /* Update the pointer to the next expiring timer */
812 cpu_base->next_timer = timer;
813 cpu_base->expires_next = expires;
816 * If hres is not active, hardware does not have to be
819 * If a hang was detected in the last timer interrupt then we
820 * do not schedule a timer which is earlier than the expiry
821 * which we enforced in the hang detection. We want the system
824 if (!__hrtimer_hres_active(cpu_base) || cpu_base->hang_detected)
828 * Program the timer hardware. We enforce the expiry for
829 * events which are already in the past.
831 tick_program_event(expires, 1);
835 * Clock realtime was set
837 * Change the offset of the realtime clock vs. the monotonic
840 * We might have to reprogram the high resolution timer interrupt. On
841 * SMP we call the architecture specific code to retrigger _all_ high
842 * resolution timer interrupts. On UP we just disable interrupts and
843 * call the high resolution interrupt code.
845 void clock_was_set(void)
847 #ifdef CONFIG_HIGH_RES_TIMERS
848 /* Retrigger the CPU local events everywhere */
849 on_each_cpu(retrigger_next_event, NULL, 1);
851 timerfd_clock_was_set();
855 * During resume we might have to reprogram the high resolution timer
856 * interrupt on all online CPUs. However, all other CPUs will be
857 * stopped with IRQs interrupts disabled so the clock_was_set() call
860 void hrtimers_resume(void)
862 lockdep_assert_irqs_disabled();
863 /* Retrigger on the local CPU */
864 retrigger_next_event(NULL);
865 /* And schedule a retrigger for all others */
866 clock_was_set_delayed();
870 * Counterpart to lock_hrtimer_base above:
873 void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
875 raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
879 * hrtimer_forward - forward the timer expiry
880 * @timer: hrtimer to forward
881 * @now: forward past this time
882 * @interval: the interval to forward
884 * Forward the timer expiry so it will expire in the future.
885 * Returns the number of overruns.
887 * Can be safely called from the callback function of @timer. If
888 * called from other contexts @timer must neither be enqueued nor
889 * running the callback and the caller needs to take care of
892 * Note: This only updates the timer expiry value and does not requeue
895 u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
900 delta = ktime_sub(now, hrtimer_get_expires(timer));
905 if (WARN_ON(timer->state & HRTIMER_STATE_ENQUEUED))
908 if (interval < hrtimer_resolution)
909 interval = hrtimer_resolution;
911 if (unlikely(delta >= interval)) {
912 s64 incr = ktime_to_ns(interval);
914 orun = ktime_divns(delta, incr);
915 hrtimer_add_expires_ns(timer, incr * orun);
916 if (hrtimer_get_expires_tv64(timer) > now)
919 * This (and the ktime_add() below) is the
920 * correction for exact:
924 hrtimer_add_expires(timer, interval);
928 EXPORT_SYMBOL_GPL(hrtimer_forward);
931 * enqueue_hrtimer - internal function to (re)start a timer
933 * The timer is inserted in expiry order. Insertion into the
934 * red black tree is O(log(n)). Must hold the base lock.
936 * Returns 1 when the new timer is the leftmost timer in the tree.
938 static int enqueue_hrtimer(struct hrtimer *timer,
939 struct hrtimer_clock_base *base,
940 enum hrtimer_mode mode)
942 debug_activate(timer, mode);
944 base->cpu_base->active_bases |= 1 << base->index;
946 timer->state = HRTIMER_STATE_ENQUEUED;
948 return timerqueue_add(&base->active, &timer->node);
952 * __remove_hrtimer - internal function to remove a timer
954 * Caller must hold the base lock.
956 * High resolution timer mode reprograms the clock event device when the
957 * timer is the one which expires next. The caller can disable this by setting
958 * reprogram to zero. This is useful, when the context does a reprogramming
959 * anyway (e.g. timer interrupt)
961 static void __remove_hrtimer(struct hrtimer *timer,
962 struct hrtimer_clock_base *base,
963 u8 newstate, int reprogram)
965 struct hrtimer_cpu_base *cpu_base = base->cpu_base;
966 u8 state = timer->state;
968 timer->state = newstate;
969 if (!(state & HRTIMER_STATE_ENQUEUED))
972 if (!timerqueue_del(&base->active, &timer->node))
973 cpu_base->active_bases &= ~(1 << base->index);
976 * Note: If reprogram is false we do not update
977 * cpu_base->next_timer. This happens when we remove the first
978 * timer on a remote cpu. No harm as we never dereference
979 * cpu_base->next_timer. So the worst thing what can happen is
980 * an superflous call to hrtimer_force_reprogram() on the
981 * remote cpu later on if the same timer gets enqueued again.
983 if (reprogram && timer == cpu_base->next_timer)
984 hrtimer_force_reprogram(cpu_base, 1);
988 * remove hrtimer, called with base lock held
991 remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base, bool restart)
993 if (hrtimer_is_queued(timer)) {
994 u8 state = timer->state;
998 * Remove the timer and force reprogramming when high
999 * resolution mode is active and the timer is on the current
1000 * CPU. If we remove a timer on another CPU, reprogramming is
1001 * skipped. The interrupt event on this CPU is fired and
1002 * reprogramming happens in the interrupt handler. This is a
1003 * rare case and less expensive than a smp call.
1005 debug_deactivate(timer);
1006 reprogram = base->cpu_base == this_cpu_ptr(&hrtimer_bases);
1009 state = HRTIMER_STATE_INACTIVE;
1011 __remove_hrtimer(timer, base, state, reprogram);
1017 static inline ktime_t hrtimer_update_lowres(struct hrtimer *timer, ktime_t tim,
1018 const enum hrtimer_mode mode)
1020 #ifdef CONFIG_TIME_LOW_RES
1022 * CONFIG_TIME_LOW_RES indicates that the system has no way to return
1023 * granular time values. For relative timers we add hrtimer_resolution
1024 * (i.e. one jiffie) to prevent short timeouts.
1026 timer->is_rel = mode & HRTIMER_MODE_REL;
1028 tim = ktime_add_safe(tim, hrtimer_resolution);
1034 hrtimer_update_softirq_timer(struct hrtimer_cpu_base *cpu_base, bool reprogram)
1039 * Find the next SOFT expiration.
1041 expires = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_SOFT);
1044 * reprogramming needs to be triggered, even if the next soft
1045 * hrtimer expires at the same time than the next hard
1046 * hrtimer. cpu_base->softirq_expires_next needs to be updated!
1048 if (expires == KTIME_MAX)
1052 * cpu_base->*next_timer is recomputed by __hrtimer_get_next_event()
1053 * cpu_base->*expires_next is only set by hrtimer_reprogram()
1055 hrtimer_reprogram(cpu_base->softirq_next_timer, reprogram);
1058 static int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1059 u64 delta_ns, const enum hrtimer_mode mode,
1060 struct hrtimer_clock_base *base)
1062 struct hrtimer_clock_base *new_base;
1064 /* Remove an active timer from the queue: */
1065 remove_hrtimer(timer, base, true);
1067 if (mode & HRTIMER_MODE_REL)
1068 tim = ktime_add_safe(tim, base->get_time());
1070 tim = hrtimer_update_lowres(timer, tim, mode);
1072 hrtimer_set_expires_range_ns(timer, tim, delta_ns);
1074 /* Switch the timer base, if necessary: */
1075 new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED);
1077 return enqueue_hrtimer(timer, new_base, mode);
1081 * hrtimer_start_range_ns - (re)start an hrtimer
1082 * @timer: the timer to be added
1084 * @delta_ns: "slack" range for the timer
1085 * @mode: timer mode: absolute (HRTIMER_MODE_ABS) or
1086 * relative (HRTIMER_MODE_REL), and pinned (HRTIMER_MODE_PINNED);
1087 * softirq based mode is considered for debug purpose only!
1089 void hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1090 u64 delta_ns, const enum hrtimer_mode mode)
1092 struct hrtimer_clock_base *base;
1093 unsigned long flags;
1096 * Check whether the HRTIMER_MODE_SOFT bit and hrtimer.is_soft
1099 WARN_ON_ONCE(!(mode & HRTIMER_MODE_SOFT) ^ !timer->is_soft);
1101 base = lock_hrtimer_base(timer, &flags);
1103 if (__hrtimer_start_range_ns(timer, tim, delta_ns, mode, base))
1104 hrtimer_reprogram(timer, true);
1106 unlock_hrtimer_base(timer, &flags);
1108 EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
1111 * hrtimer_try_to_cancel - try to deactivate a timer
1112 * @timer: hrtimer to stop
1115 * 0 when the timer was not active
1116 * 1 when the timer was active
1117 * -1 when the timer is currently executing the callback function and
1120 int hrtimer_try_to_cancel(struct hrtimer *timer)
1122 struct hrtimer_clock_base *base;
1123 unsigned long flags;
1127 * Check lockless first. If the timer is not active (neither
1128 * enqueued nor running the callback, nothing to do here. The
1129 * base lock does not serialize against a concurrent enqueue,
1130 * so we can avoid taking it.
1132 if (!hrtimer_active(timer))
1135 base = lock_hrtimer_base(timer, &flags);
1137 if (!hrtimer_callback_running(timer))
1138 ret = remove_hrtimer(timer, base, false);
1140 unlock_hrtimer_base(timer, &flags);
1145 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1148 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1149 * @timer: the timer to be cancelled
1152 * 0 when the timer was not active
1153 * 1 when the timer was active
1155 int hrtimer_cancel(struct hrtimer *timer)
1158 int ret = hrtimer_try_to_cancel(timer);
1165 EXPORT_SYMBOL_GPL(hrtimer_cancel);
1168 * hrtimer_get_remaining - get remaining time for the timer
1169 * @timer: the timer to read
1170 * @adjust: adjust relative timers when CONFIG_TIME_LOW_RES=y
1172 ktime_t __hrtimer_get_remaining(const struct hrtimer *timer, bool adjust)
1174 unsigned long flags;
1177 lock_hrtimer_base(timer, &flags);
1178 if (IS_ENABLED(CONFIG_TIME_LOW_RES) && adjust)
1179 rem = hrtimer_expires_remaining_adjusted(timer);
1181 rem = hrtimer_expires_remaining(timer);
1182 unlock_hrtimer_base(timer, &flags);
1186 EXPORT_SYMBOL_GPL(__hrtimer_get_remaining);
1188 #ifdef CONFIG_NO_HZ_COMMON
1190 * hrtimer_get_next_event - get the time until next expiry event
1192 * Returns the next expiry time or KTIME_MAX if no timer is pending.
1194 u64 hrtimer_get_next_event(void)
1196 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1197 u64 expires = KTIME_MAX;
1198 unsigned long flags;
1200 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1202 if (!__hrtimer_hres_active(cpu_base))
1203 expires = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_ALL);
1205 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1211 * hrtimer_next_event_without - time until next expiry event w/o one timer
1212 * @exclude: timer to exclude
1214 * Returns the next expiry time over all timers except for the @exclude one or
1215 * KTIME_MAX if none of them is pending.
1217 u64 hrtimer_next_event_without(const struct hrtimer *exclude)
1219 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1220 u64 expires = KTIME_MAX;
1221 unsigned long flags;
1223 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1225 if (__hrtimer_hres_active(cpu_base)) {
1226 unsigned int active;
1228 if (!cpu_base->softirq_activated) {
1229 active = cpu_base->active_bases & HRTIMER_ACTIVE_SOFT;
1230 expires = __hrtimer_next_event_base(cpu_base, exclude,
1233 active = cpu_base->active_bases & HRTIMER_ACTIVE_HARD;
1234 expires = __hrtimer_next_event_base(cpu_base, exclude, active,
1238 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1244 static inline int hrtimer_clockid_to_base(clockid_t clock_id)
1246 if (likely(clock_id < MAX_CLOCKS)) {
1247 int base = hrtimer_clock_to_base_table[clock_id];
1249 if (likely(base != HRTIMER_MAX_CLOCK_BASES))
1252 WARN(1, "Invalid clockid %d. Using MONOTONIC\n", clock_id);
1253 return HRTIMER_BASE_MONOTONIC;
1256 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1257 enum hrtimer_mode mode)
1259 bool softtimer = !!(mode & HRTIMER_MODE_SOFT);
1260 int base = softtimer ? HRTIMER_MAX_CLOCK_BASES / 2 : 0;
1261 struct hrtimer_cpu_base *cpu_base;
1263 memset(timer, 0, sizeof(struct hrtimer));
1265 cpu_base = raw_cpu_ptr(&hrtimer_bases);
1268 * POSIX magic: Relative CLOCK_REALTIME timers are not affected by
1269 * clock modifications, so they needs to become CLOCK_MONOTONIC to
1270 * ensure POSIX compliance.
1272 if (clock_id == CLOCK_REALTIME && mode & HRTIMER_MODE_REL)
1273 clock_id = CLOCK_MONOTONIC;
1275 base += hrtimer_clockid_to_base(clock_id);
1276 timer->is_soft = softtimer;
1277 timer->base = &cpu_base->clock_base[base];
1278 timerqueue_init(&timer->node);
1282 * hrtimer_init - initialize a timer to the given clock
1283 * @timer: the timer to be initialized
1284 * @clock_id: the clock to be used
1285 * @mode: The modes which are relevant for intitialization:
1286 * HRTIMER_MODE_ABS, HRTIMER_MODE_REL, HRTIMER_MODE_ABS_SOFT,
1287 * HRTIMER_MODE_REL_SOFT
1289 * The PINNED variants of the above can be handed in,
1290 * but the PINNED bit is ignored as pinning happens
1291 * when the hrtimer is started
1293 void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1294 enum hrtimer_mode mode)
1296 debug_init(timer, clock_id, mode);
1297 __hrtimer_init(timer, clock_id, mode);
1299 EXPORT_SYMBOL_GPL(hrtimer_init);
1302 * A timer is active, when it is enqueued into the rbtree or the
1303 * callback function is running or it's in the state of being migrated
1306 * It is important for this function to not return a false negative.
1308 bool hrtimer_active(const struct hrtimer *timer)
1310 struct hrtimer_clock_base *base;
1314 base = READ_ONCE(timer->base);
1315 seq = raw_read_seqcount_begin(&base->seq);
1317 if (timer->state != HRTIMER_STATE_INACTIVE ||
1318 base->running == timer)
1321 } while (read_seqcount_retry(&base->seq, seq) ||
1322 base != READ_ONCE(timer->base));
1326 EXPORT_SYMBOL_GPL(hrtimer_active);
1329 * The write_seqcount_barrier()s in __run_hrtimer() split the thing into 3
1330 * distinct sections:
1332 * - queued: the timer is queued
1333 * - callback: the timer is being ran
1334 * - post: the timer is inactive or (re)queued
1336 * On the read side we ensure we observe timer->state and cpu_base->running
1337 * from the same section, if anything changed while we looked at it, we retry.
1338 * This includes timer->base changing because sequence numbers alone are
1339 * insufficient for that.
1341 * The sequence numbers are required because otherwise we could still observe
1342 * a false negative if the read side got smeared over multiple consequtive
1343 * __run_hrtimer() invocations.
1346 static void __run_hrtimer(struct hrtimer_cpu_base *cpu_base,
1347 struct hrtimer_clock_base *base,
1348 struct hrtimer *timer, ktime_t *now,
1349 unsigned long flags)
1351 enum hrtimer_restart (*fn)(struct hrtimer *);
1354 lockdep_assert_held(&cpu_base->lock);
1356 debug_deactivate(timer);
1357 base->running = timer;
1360 * Separate the ->running assignment from the ->state assignment.
1362 * As with a regular write barrier, this ensures the read side in
1363 * hrtimer_active() cannot observe base->running == NULL &&
1364 * timer->state == INACTIVE.
1366 raw_write_seqcount_barrier(&base->seq);
1368 __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE, 0);
1369 fn = timer->function;
1372 * Clear the 'is relative' flag for the TIME_LOW_RES case. If the
1373 * timer is restarted with a period then it becomes an absolute
1374 * timer. If its not restarted it does not matter.
1376 if (IS_ENABLED(CONFIG_TIME_LOW_RES))
1377 timer->is_rel = false;
1380 * The timer is marked as running in the CPU base, so it is
1381 * protected against migration to a different CPU even if the lock
1384 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1385 trace_hrtimer_expire_entry(timer, now);
1386 restart = fn(timer);
1387 trace_hrtimer_expire_exit(timer);
1388 raw_spin_lock_irq(&cpu_base->lock);
1391 * Note: We clear the running state after enqueue_hrtimer and
1392 * we do not reprogram the event hardware. Happens either in
1393 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1395 * Note: Because we dropped the cpu_base->lock above,
1396 * hrtimer_start_range_ns() can have popped in and enqueued the timer
1399 if (restart != HRTIMER_NORESTART &&
1400 !(timer->state & HRTIMER_STATE_ENQUEUED))
1401 enqueue_hrtimer(timer, base, HRTIMER_MODE_ABS);
1404 * Separate the ->running assignment from the ->state assignment.
1406 * As with a regular write barrier, this ensures the read side in
1407 * hrtimer_active() cannot observe base->running.timer == NULL &&
1408 * timer->state == INACTIVE.
1410 raw_write_seqcount_barrier(&base->seq);
1412 WARN_ON_ONCE(base->running != timer);
1413 base->running = NULL;
1416 static void __hrtimer_run_queues(struct hrtimer_cpu_base *cpu_base, ktime_t now,
1417 unsigned long flags, unsigned int active_mask)
1419 struct hrtimer_clock_base *base;
1420 unsigned int active = cpu_base->active_bases & active_mask;
1422 for_each_active_base(base, cpu_base, active) {
1423 struct timerqueue_node *node;
1426 basenow = ktime_add(now, base->offset);
1428 while ((node = timerqueue_getnext(&base->active))) {
1429 struct hrtimer *timer;
1431 timer = container_of(node, struct hrtimer, node);
1434 * The immediate goal for using the softexpires is
1435 * minimizing wakeups, not running timers at the
1436 * earliest interrupt after their soft expiration.
1437 * This allows us to avoid using a Priority Search
1438 * Tree, which can answer a stabbing querry for
1439 * overlapping intervals and instead use the simple
1440 * BST we already have.
1441 * We don't add extra wakeups by delaying timers that
1442 * are right-of a not yet expired timer, because that
1443 * timer will have to trigger a wakeup anyway.
1445 if (basenow < hrtimer_get_softexpires_tv64(timer))
1448 __run_hrtimer(cpu_base, base, timer, &basenow, flags);
1453 static __latent_entropy void hrtimer_run_softirq(struct softirq_action *h)
1455 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1456 unsigned long flags;
1459 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1461 now = hrtimer_update_base(cpu_base);
1462 __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_SOFT);
1464 cpu_base->softirq_activated = 0;
1465 hrtimer_update_softirq_timer(cpu_base, true);
1467 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1470 #ifdef CONFIG_HIGH_RES_TIMERS
1473 * High resolution timer interrupt
1474 * Called with interrupts disabled
1476 void hrtimer_interrupt(struct clock_event_device *dev)
1478 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1479 ktime_t expires_next, now, entry_time, delta;
1480 unsigned long flags;
1483 BUG_ON(!cpu_base->hres_active);
1484 cpu_base->nr_events++;
1485 dev->next_event = KTIME_MAX;
1487 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1488 entry_time = now = hrtimer_update_base(cpu_base);
1490 cpu_base->in_hrtirq = 1;
1492 * We set expires_next to KTIME_MAX here with cpu_base->lock
1493 * held to prevent that a timer is enqueued in our queue via
1494 * the migration code. This does not affect enqueueing of
1495 * timers which run their callback and need to be requeued on
1498 cpu_base->expires_next = KTIME_MAX;
1500 if (!ktime_before(now, cpu_base->softirq_expires_next)) {
1501 cpu_base->softirq_expires_next = KTIME_MAX;
1502 cpu_base->softirq_activated = 1;
1503 raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1506 __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_HARD);
1508 /* Reevaluate the clock bases for the next expiry */
1509 expires_next = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_ALL);
1511 * Store the new expiry value so the migration code can verify
1514 cpu_base->expires_next = expires_next;
1515 cpu_base->in_hrtirq = 0;
1516 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1518 /* Reprogramming necessary ? */
1519 if (!tick_program_event(expires_next, 0)) {
1520 cpu_base->hang_detected = 0;
1525 * The next timer was already expired due to:
1527 * - long lasting callbacks
1528 * - being scheduled away when running in a VM
1530 * We need to prevent that we loop forever in the hrtimer
1531 * interrupt routine. We give it 3 attempts to avoid
1532 * overreacting on some spurious event.
1534 * Acquire base lock for updating the offsets and retrieving
1537 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1538 now = hrtimer_update_base(cpu_base);
1539 cpu_base->nr_retries++;
1543 * Give the system a chance to do something else than looping
1544 * here. We stored the entry time, so we know exactly how long
1545 * we spent here. We schedule the next event this amount of
1548 cpu_base->nr_hangs++;
1549 cpu_base->hang_detected = 1;
1550 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1552 delta = ktime_sub(now, entry_time);
1553 if ((unsigned int)delta > cpu_base->max_hang_time)
1554 cpu_base->max_hang_time = (unsigned int) delta;
1556 * Limit it to a sensible value as we enforce a longer
1557 * delay. Give the CPU at least 100ms to catch up.
1559 if (delta > 100 * NSEC_PER_MSEC)
1560 expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1562 expires_next = ktime_add(now, delta);
1563 tick_program_event(expires_next, 1);
1564 printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n",
1565 ktime_to_ns(delta));
1568 /* called with interrupts disabled */
1569 static inline void __hrtimer_peek_ahead_timers(void)
1571 struct tick_device *td;
1573 if (!hrtimer_hres_active())
1576 td = this_cpu_ptr(&tick_cpu_device);
1577 if (td && td->evtdev)
1578 hrtimer_interrupt(td->evtdev);
1581 #else /* CONFIG_HIGH_RES_TIMERS */
1583 static inline void __hrtimer_peek_ahead_timers(void) { }
1585 #endif /* !CONFIG_HIGH_RES_TIMERS */
1588 * Called from run_local_timers in hardirq context every jiffy
1590 void hrtimer_run_queues(void)
1592 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1593 unsigned long flags;
1596 if (__hrtimer_hres_active(cpu_base))
1600 * This _is_ ugly: We have to check periodically, whether we
1601 * can switch to highres and / or nohz mode. The clocksource
1602 * switch happens with xtime_lock held. Notification from
1603 * there only sets the check bit in the tick_oneshot code,
1604 * otherwise we might deadlock vs. xtime_lock.
1606 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) {
1607 hrtimer_switch_to_hres();
1611 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1612 now = hrtimer_update_base(cpu_base);
1614 if (!ktime_before(now, cpu_base->softirq_expires_next)) {
1615 cpu_base->softirq_expires_next = KTIME_MAX;
1616 cpu_base->softirq_activated = 1;
1617 raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1620 __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_HARD);
1621 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1625 * Sleep related functions:
1627 static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1629 struct hrtimer_sleeper *t =
1630 container_of(timer, struct hrtimer_sleeper, timer);
1631 struct task_struct *task = t->task;
1635 wake_up_process(task);
1637 return HRTIMER_NORESTART;
1640 void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1642 sl->timer.function = hrtimer_wakeup;
1645 EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
1647 int nanosleep_copyout(struct restart_block *restart, struct timespec64 *ts)
1649 switch(restart->nanosleep.type) {
1650 #ifdef CONFIG_COMPAT
1652 if (compat_put_timespec64(ts, restart->nanosleep.compat_rmtp))
1657 if (put_timespec64(ts, restart->nanosleep.rmtp))
1663 return -ERESTART_RESTARTBLOCK;
1666 static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1668 struct restart_block *restart;
1670 hrtimer_init_sleeper(t, current);
1673 set_current_state(TASK_INTERRUPTIBLE);
1674 hrtimer_start_expires(&t->timer, mode);
1676 if (likely(t->task))
1677 freezable_schedule();
1679 hrtimer_cancel(&t->timer);
1680 mode = HRTIMER_MODE_ABS;
1682 } while (t->task && !signal_pending(current));
1684 __set_current_state(TASK_RUNNING);
1689 restart = ¤t->restart_block;
1690 if (restart->nanosleep.type != TT_NONE) {
1691 ktime_t rem = hrtimer_expires_remaining(&t->timer);
1692 struct timespec64 rmt;
1696 rmt = ktime_to_timespec64(rem);
1698 return nanosleep_copyout(restart, &rmt);
1700 return -ERESTART_RESTARTBLOCK;
1703 static long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1705 struct hrtimer_sleeper t;
1708 hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid,
1710 hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
1712 ret = do_nanosleep(&t, HRTIMER_MODE_ABS);
1713 destroy_hrtimer_on_stack(&t.timer);
1717 long hrtimer_nanosleep(const struct timespec64 *rqtp,
1718 const enum hrtimer_mode mode, const clockid_t clockid)
1720 struct restart_block *restart;
1721 struct hrtimer_sleeper t;
1725 slack = current->timer_slack_ns;
1726 if (dl_task(current) || rt_task(current))
1729 hrtimer_init_on_stack(&t.timer, clockid, mode);
1730 hrtimer_set_expires_range_ns(&t.timer, timespec64_to_ktime(*rqtp), slack);
1731 ret = do_nanosleep(&t, mode);
1732 if (ret != -ERESTART_RESTARTBLOCK)
1735 /* Absolute timers do not update the rmtp value and restart: */
1736 if (mode == HRTIMER_MODE_ABS) {
1737 ret = -ERESTARTNOHAND;
1741 restart = ¤t->restart_block;
1742 restart->fn = hrtimer_nanosleep_restart;
1743 restart->nanosleep.clockid = t.timer.base->clockid;
1744 restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
1746 destroy_hrtimer_on_stack(&t.timer);
1750 SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp,
1751 struct timespec __user *, rmtp)
1753 struct timespec64 tu;
1755 if (get_timespec64(&tu, rqtp))
1758 if (!timespec64_valid(&tu))
1761 current->restart_block.nanosleep.type = rmtp ? TT_NATIVE : TT_NONE;
1762 current->restart_block.nanosleep.rmtp = rmtp;
1763 return hrtimer_nanosleep(&tu, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1766 #ifdef CONFIG_COMPAT
1768 COMPAT_SYSCALL_DEFINE2(nanosleep, struct compat_timespec __user *, rqtp,
1769 struct compat_timespec __user *, rmtp)
1771 struct timespec64 tu;
1773 if (compat_get_timespec64(&tu, rqtp))
1776 if (!timespec64_valid(&tu))
1779 current->restart_block.nanosleep.type = rmtp ? TT_COMPAT : TT_NONE;
1780 current->restart_block.nanosleep.compat_rmtp = rmtp;
1781 return hrtimer_nanosleep(&tu, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1786 * Functions related to boot-time initialization:
1788 int hrtimers_prepare_cpu(unsigned int cpu)
1790 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1793 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1794 cpu_base->clock_base[i].cpu_base = cpu_base;
1795 timerqueue_init_head(&cpu_base->clock_base[i].active);
1798 cpu_base->cpu = cpu;
1799 cpu_base->active_bases = 0;
1800 cpu_base->hres_active = 0;
1801 cpu_base->hang_detected = 0;
1802 cpu_base->next_timer = NULL;
1803 cpu_base->softirq_next_timer = NULL;
1804 cpu_base->expires_next = KTIME_MAX;
1805 cpu_base->softirq_expires_next = KTIME_MAX;
1809 #ifdef CONFIG_HOTPLUG_CPU
1811 static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1812 struct hrtimer_clock_base *new_base)
1814 struct hrtimer *timer;
1815 struct timerqueue_node *node;
1817 while ((node = timerqueue_getnext(&old_base->active))) {
1818 timer = container_of(node, struct hrtimer, node);
1819 BUG_ON(hrtimer_callback_running(timer));
1820 debug_deactivate(timer);
1823 * Mark it as ENQUEUED not INACTIVE otherwise the
1824 * timer could be seen as !active and just vanish away
1825 * under us on another CPU
1827 __remove_hrtimer(timer, old_base, HRTIMER_STATE_ENQUEUED, 0);
1828 timer->base = new_base;
1830 * Enqueue the timers on the new cpu. This does not
1831 * reprogram the event device in case the timer
1832 * expires before the earliest on this CPU, but we run
1833 * hrtimer_interrupt after we migrated everything to
1834 * sort out already expired timers and reprogram the
1837 enqueue_hrtimer(timer, new_base, HRTIMER_MODE_ABS);
1841 int hrtimers_dead_cpu(unsigned int scpu)
1843 struct hrtimer_cpu_base *old_base, *new_base;
1846 BUG_ON(cpu_online(scpu));
1847 tick_cancel_sched_timer(scpu);
1850 * this BH disable ensures that raise_softirq_irqoff() does
1851 * not wakeup ksoftirqd (and acquire the pi-lock) while
1852 * holding the cpu_base lock
1855 local_irq_disable();
1856 old_base = &per_cpu(hrtimer_bases, scpu);
1857 new_base = this_cpu_ptr(&hrtimer_bases);
1859 * The caller is globally serialized and nobody else
1860 * takes two locks at once, deadlock is not possible.
1862 raw_spin_lock(&new_base->lock);
1863 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1865 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1866 migrate_hrtimer_list(&old_base->clock_base[i],
1867 &new_base->clock_base[i]);
1871 * The migration might have changed the first expiring softirq
1872 * timer on this CPU. Update it.
1874 hrtimer_update_softirq_timer(new_base, false);
1876 raw_spin_unlock(&old_base->lock);
1877 raw_spin_unlock(&new_base->lock);
1879 /* Check, if we got expired work to do */
1880 __hrtimer_peek_ahead_timers();
1886 #endif /* CONFIG_HOTPLUG_CPU */
1888 void __init hrtimers_init(void)
1890 hrtimers_prepare_cpu(smp_processor_id());
1891 open_softirq(HRTIMER_SOFTIRQ, hrtimer_run_softirq);
1895 * schedule_hrtimeout_range_clock - sleep until timeout
1896 * @expires: timeout value (ktime_t)
1897 * @delta: slack in expires timeout (ktime_t)
1899 * @clock_id: timer clock to be used
1902 schedule_hrtimeout_range_clock(ktime_t *expires, u64 delta,
1903 const enum hrtimer_mode mode, clockid_t clock_id)
1905 struct hrtimer_sleeper t;
1908 * Optimize when a zero timeout value is given. It does not
1909 * matter whether this is an absolute or a relative time.
1911 if (expires && *expires == 0) {
1912 __set_current_state(TASK_RUNNING);
1917 * A NULL parameter means "infinite"
1924 hrtimer_init_on_stack(&t.timer, clock_id, mode);
1925 hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
1927 hrtimer_init_sleeper(&t, current);
1929 hrtimer_start_expires(&t.timer, mode);
1934 hrtimer_cancel(&t.timer);
1935 destroy_hrtimer_on_stack(&t.timer);
1937 __set_current_state(TASK_RUNNING);
1939 return !t.task ? 0 : -EINTR;
1943 * schedule_hrtimeout_range - sleep until timeout
1944 * @expires: timeout value (ktime_t)
1945 * @delta: slack in expires timeout (ktime_t)
1948 * Make the current task sleep until the given expiry time has
1949 * elapsed. The routine will return immediately unless
1950 * the current task state has been set (see set_current_state()).
1952 * The @delta argument gives the kernel the freedom to schedule the
1953 * actual wakeup to a time that is both power and performance friendly.
1954 * The kernel give the normal best effort behavior for "@expires+@delta",
1955 * but may decide to fire the timer earlier, but no earlier than @expires.
1957 * You can set the task state as follows -
1959 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1960 * pass before the routine returns unless the current task is explicitly
1961 * woken up, (e.g. by wake_up_process()).
1963 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1964 * delivered to the current task or the current task is explicitly woken
1967 * The current task state is guaranteed to be TASK_RUNNING when this
1970 * Returns 0 when the timer has expired. If the task was woken before the
1971 * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or
1972 * by an explicit wakeup, it returns -EINTR.
1974 int __sched schedule_hrtimeout_range(ktime_t *expires, u64 delta,
1975 const enum hrtimer_mode mode)
1977 return schedule_hrtimeout_range_clock(expires, delta, mode,
1980 EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
1983 * schedule_hrtimeout - sleep until timeout
1984 * @expires: timeout value (ktime_t)
1987 * Make the current task sleep until the given expiry time has
1988 * elapsed. The routine will return immediately unless
1989 * the current task state has been set (see set_current_state()).
1991 * You can set the task state as follows -
1993 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1994 * pass before the routine returns unless the current task is explicitly
1995 * woken up, (e.g. by wake_up_process()).
1997 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1998 * delivered to the current task or the current task is explicitly woken
2001 * The current task state is guaranteed to be TASK_RUNNING when this
2004 * Returns 0 when the timer has expired. If the task was woken before the
2005 * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or
2006 * by an explicit wakeup, it returns -EINTR.
2008 int __sched schedule_hrtimeout(ktime_t *expires,
2009 const enum hrtimer_mode mode)
2011 return schedule_hrtimeout_range(expires, 0, mode);
2013 EXPORT_SYMBOL_GPL(schedule_hrtimeout);