1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright(C) 2005-2006, Thomas Gleixner <tglx@linutronix.de>
4 * Copyright(C) 2005-2007, Red Hat, Inc., Ingo Molnar
5 * Copyright(C) 2006-2007 Timesys Corp., Thomas Gleixner
7 * High-resolution kernel timers
9 * In contrast to the low-resolution timeout API, aka timer wheel,
10 * hrtimers provide finer resolution and accuracy depending on system
11 * configuration and capabilities.
13 * Started by: Thomas Gleixner and Ingo Molnar
16 * Based on the original timer wheel code
18 * Help, testing, suggestions, bugfixes, improvements were
21 * George Anzinger, Andrew Morton, Steven Rostedt, Roman Zippel
25 #include <linux/cpu.h>
26 #include <linux/export.h>
27 #include <linux/percpu.h>
28 #include <linux/hrtimer.h>
29 #include <linux/notifier.h>
30 #include <linux/syscalls.h>
31 #include <linux/interrupt.h>
32 #include <linux/tick.h>
33 #include <linux/err.h>
34 #include <linux/debugobjects.h>
35 #include <linux/sched/signal.h>
36 #include <linux/sched/sysctl.h>
37 #include <linux/sched/rt.h>
38 #include <linux/sched/deadline.h>
39 #include <linux/sched/nohz.h>
40 #include <linux/sched/debug.h>
41 #include <linux/timer.h>
42 #include <linux/freezer.h>
43 #include <linux/compat.h>
45 #include <linux/uaccess.h>
47 #include <trace/events/timer.h>
49 #include "tick-internal.h"
52 * Masks for selecting the soft and hard context timers from
55 #define MASK_SHIFT (HRTIMER_BASE_MONOTONIC_SOFT)
56 #define HRTIMER_ACTIVE_HARD ((1U << MASK_SHIFT) - 1)
57 #define HRTIMER_ACTIVE_SOFT (HRTIMER_ACTIVE_HARD << MASK_SHIFT)
58 #define HRTIMER_ACTIVE_ALL (HRTIMER_ACTIVE_SOFT | HRTIMER_ACTIVE_HARD)
63 * There are more clockids than hrtimer bases. Thus, we index
64 * into the timer bases by the hrtimer_base_type enum. When trying
65 * to reach a base using a clockid, hrtimer_clockid_to_base()
66 * is used to convert from clockid to the proper hrtimer_base_type.
68 DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
70 .lock = __RAW_SPIN_LOCK_UNLOCKED(hrtimer_bases.lock),
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,
94 .index = HRTIMER_BASE_MONOTONIC_SOFT,
95 .clockid = CLOCK_MONOTONIC,
96 .get_time = &ktime_get,
99 .index = HRTIMER_BASE_REALTIME_SOFT,
100 .clockid = CLOCK_REALTIME,
101 .get_time = &ktime_get_real,
104 .index = HRTIMER_BASE_BOOTTIME_SOFT,
105 .clockid = CLOCK_BOOTTIME,
106 .get_time = &ktime_get_boottime,
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_BOOTTIME,
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 = {
139 .cpu_base = &migration_cpu_base,
140 .seq = SEQCNT_RAW_SPINLOCK_ZERO(migration_cpu_base.seq,
141 &migration_cpu_base.lock),
145 #define migration_base migration_cpu_base.clock_base[0]
147 static inline bool is_migration_base(struct hrtimer_clock_base *base)
149 return base == &migration_base;
153 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
154 * means that all timers which are tied to this base via timer->base are
155 * locked, and the base itself is locked too.
157 * So __run_timers/migrate_timers can safely modify all timers which could
158 * be found on the lists/queues.
160 * When the timer's base is locked, and the timer removed from list, it is
161 * possible to set timer->base = &migration_base and drop the lock: the timer
165 struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
166 unsigned long *flags)
167 __acquires(&timer->base->lock)
169 struct hrtimer_clock_base *base;
172 base = READ_ONCE(timer->base);
173 if (likely(base != &migration_base)) {
174 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
175 if (likely(base == timer->base))
177 /* The timer has migrated to another CPU: */
178 raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
185 * We do not migrate the timer when it is expiring before the next
186 * event on the target cpu. When high resolution is enabled, we cannot
187 * reprogram the target cpu hardware and we would cause it to fire
188 * late. To keep it simple, we handle the high resolution enabled and
189 * disabled case similar.
191 * Called with cpu_base->lock of target cpu held.
194 hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base)
198 expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset);
199 return expires < new_base->cpu_base->expires_next;
203 struct hrtimer_cpu_base *get_target_base(struct hrtimer_cpu_base *base,
206 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
207 if (static_branch_likely(&timers_migration_enabled) && !pinned)
208 return &per_cpu(hrtimer_bases, get_nohz_timer_target());
214 * We switch the timer base to a power-optimized selected CPU target,
216 * - NO_HZ_COMMON is enabled
217 * - timer migration is enabled
218 * - the timer callback is not running
219 * - the timer is not the first expiring timer on the new target
221 * If one of the above requirements is not fulfilled we move the timer
222 * to the current CPU or leave it on the previously assigned CPU if
223 * the timer callback is currently running.
225 static inline struct hrtimer_clock_base *
226 switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
229 struct hrtimer_cpu_base *new_cpu_base, *this_cpu_base;
230 struct hrtimer_clock_base *new_base;
231 int basenum = base->index;
233 this_cpu_base = this_cpu_ptr(&hrtimer_bases);
234 new_cpu_base = get_target_base(this_cpu_base, pinned);
236 new_base = &new_cpu_base->clock_base[basenum];
238 if (base != new_base) {
240 * We are trying to move timer to new_base.
241 * However we can't change timer's base while it is running,
242 * so we keep it on the same CPU. No hassle vs. reprogramming
243 * the event source in the high resolution case. The softirq
244 * code will take care of this when the timer function has
245 * completed. There is no conflict as we hold the lock until
246 * the timer is enqueued.
248 if (unlikely(hrtimer_callback_running(timer)))
251 /* See the comment in lock_hrtimer_base() */
252 WRITE_ONCE(timer->base, &migration_base);
253 raw_spin_unlock(&base->cpu_base->lock);
254 raw_spin_lock(&new_base->cpu_base->lock);
256 if (new_cpu_base != this_cpu_base &&
257 hrtimer_check_target(timer, new_base)) {
258 raw_spin_unlock(&new_base->cpu_base->lock);
259 raw_spin_lock(&base->cpu_base->lock);
260 new_cpu_base = this_cpu_base;
261 WRITE_ONCE(timer->base, base);
264 WRITE_ONCE(timer->base, new_base);
266 if (new_cpu_base != this_cpu_base &&
267 hrtimer_check_target(timer, new_base)) {
268 new_cpu_base = this_cpu_base;
275 #else /* CONFIG_SMP */
277 static inline bool is_migration_base(struct hrtimer_clock_base *base)
282 static inline struct hrtimer_clock_base *
283 lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
284 __acquires(&timer->base->cpu_base->lock)
286 struct hrtimer_clock_base *base = timer->base;
288 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
293 # define switch_hrtimer_base(t, b, p) (b)
295 #endif /* !CONFIG_SMP */
298 * Functions for the union type storage format of ktime_t which are
299 * too large for inlining:
301 #if BITS_PER_LONG < 64
303 * Divide a ktime value by a nanosecond value
305 s64 __ktime_divns(const ktime_t kt, s64 div)
311 dclc = ktime_to_ns(kt);
312 tmp = dclc < 0 ? -dclc : dclc;
314 /* Make sure the divisor is less than 2^32: */
320 do_div(tmp, (u32) div);
321 return dclc < 0 ? -tmp : tmp;
323 EXPORT_SYMBOL_GPL(__ktime_divns);
324 #endif /* BITS_PER_LONG >= 64 */
327 * Add two ktime values and do a safety check for overflow:
329 ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
331 ktime_t res = ktime_add_unsafe(lhs, rhs);
334 * We use KTIME_SEC_MAX here, the maximum timeout which we can
335 * return to user space in a timespec:
337 if (res < 0 || res < lhs || res < rhs)
338 res = ktime_set(KTIME_SEC_MAX, 0);
343 EXPORT_SYMBOL_GPL(ktime_add_safe);
345 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
347 static const struct debug_obj_descr hrtimer_debug_descr;
349 static void *hrtimer_debug_hint(void *addr)
351 return ((struct hrtimer *) addr)->function;
355 * fixup_init is called when:
356 * - an active object is initialized
358 static bool hrtimer_fixup_init(void *addr, enum debug_obj_state state)
360 struct hrtimer *timer = addr;
363 case ODEBUG_STATE_ACTIVE:
364 hrtimer_cancel(timer);
365 debug_object_init(timer, &hrtimer_debug_descr);
373 * fixup_activate is called when:
374 * - an active object is activated
375 * - an unknown non-static object is activated
377 static bool hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
380 case ODEBUG_STATE_ACTIVE:
389 * fixup_free is called when:
390 * - an active object is freed
392 static bool hrtimer_fixup_free(void *addr, enum debug_obj_state state)
394 struct hrtimer *timer = addr;
397 case ODEBUG_STATE_ACTIVE:
398 hrtimer_cancel(timer);
399 debug_object_free(timer, &hrtimer_debug_descr);
406 static const struct debug_obj_descr hrtimer_debug_descr = {
408 .debug_hint = hrtimer_debug_hint,
409 .fixup_init = hrtimer_fixup_init,
410 .fixup_activate = hrtimer_fixup_activate,
411 .fixup_free = hrtimer_fixup_free,
414 static inline void debug_hrtimer_init(struct hrtimer *timer)
416 debug_object_init(timer, &hrtimer_debug_descr);
419 static inline void debug_hrtimer_activate(struct hrtimer *timer,
420 enum hrtimer_mode mode)
422 debug_object_activate(timer, &hrtimer_debug_descr);
425 static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
427 debug_object_deactivate(timer, &hrtimer_debug_descr);
430 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
431 enum hrtimer_mode mode);
433 void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
434 enum hrtimer_mode mode)
436 debug_object_init_on_stack(timer, &hrtimer_debug_descr);
437 __hrtimer_init(timer, clock_id, mode);
439 EXPORT_SYMBOL_GPL(hrtimer_init_on_stack);
441 static void __hrtimer_init_sleeper(struct hrtimer_sleeper *sl,
442 clockid_t clock_id, enum hrtimer_mode mode);
444 void hrtimer_init_sleeper_on_stack(struct hrtimer_sleeper *sl,
445 clockid_t clock_id, enum hrtimer_mode mode)
447 debug_object_init_on_stack(&sl->timer, &hrtimer_debug_descr);
448 __hrtimer_init_sleeper(sl, clock_id, mode);
450 EXPORT_SYMBOL_GPL(hrtimer_init_sleeper_on_stack);
452 void destroy_hrtimer_on_stack(struct hrtimer *timer)
454 debug_object_free(timer, &hrtimer_debug_descr);
456 EXPORT_SYMBOL_GPL(destroy_hrtimer_on_stack);
460 static inline void debug_hrtimer_init(struct hrtimer *timer) { }
461 static inline void debug_hrtimer_activate(struct hrtimer *timer,
462 enum hrtimer_mode mode) { }
463 static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
467 debug_init(struct hrtimer *timer, clockid_t clockid,
468 enum hrtimer_mode mode)
470 debug_hrtimer_init(timer);
471 trace_hrtimer_init(timer, clockid, mode);
474 static inline void debug_activate(struct hrtimer *timer,
475 enum hrtimer_mode mode)
477 debug_hrtimer_activate(timer, mode);
478 trace_hrtimer_start(timer, mode);
481 static inline void debug_deactivate(struct hrtimer *timer)
483 debug_hrtimer_deactivate(timer);
484 trace_hrtimer_cancel(timer);
487 static struct hrtimer_clock_base *
488 __next_base(struct hrtimer_cpu_base *cpu_base, unsigned int *active)
495 idx = __ffs(*active);
496 *active &= ~(1U << idx);
498 return &cpu_base->clock_base[idx];
501 #define for_each_active_base(base, cpu_base, active) \
502 while ((base = __next_base((cpu_base), &(active))))
504 static ktime_t __hrtimer_next_event_base(struct hrtimer_cpu_base *cpu_base,
505 const struct hrtimer *exclude,
507 ktime_t expires_next)
509 struct hrtimer_clock_base *base;
512 for_each_active_base(base, cpu_base, active) {
513 struct timerqueue_node *next;
514 struct hrtimer *timer;
516 next = timerqueue_getnext(&base->active);
517 timer = container_of(next, struct hrtimer, node);
518 if (timer == exclude) {
519 /* Get to the next timer in the queue. */
520 next = timerqueue_iterate_next(next);
524 timer = container_of(next, struct hrtimer, node);
526 expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
527 if (expires < expires_next) {
528 expires_next = expires;
530 /* Skip cpu_base update if a timer is being excluded. */
535 cpu_base->softirq_next_timer = timer;
537 cpu_base->next_timer = timer;
541 * clock_was_set() might have changed base->offset of any of
542 * the clock bases so the result might be negative. Fix it up
543 * to prevent a false positive in clockevents_program_event().
545 if (expires_next < 0)
551 * Recomputes cpu_base::*next_timer and returns the earliest expires_next
552 * but does not set cpu_base::*expires_next, that is done by
553 * hrtimer[_force]_reprogram and hrtimer_interrupt only. When updating
554 * cpu_base::*expires_next right away, reprogramming logic would no longer
557 * When a softirq is pending, we can ignore the HRTIMER_ACTIVE_SOFT bases,
558 * those timers will get run whenever the softirq gets handled, at the end of
559 * hrtimer_run_softirq(), hrtimer_update_softirq_timer() will re-add these bases.
561 * Therefore softirq values are those from the HRTIMER_ACTIVE_SOFT clock bases.
562 * The !softirq values are the minima across HRTIMER_ACTIVE_ALL, unless an actual
563 * softirq is pending, in which case they're the minima of HRTIMER_ACTIVE_HARD.
565 * @active_mask must be one of:
566 * - HRTIMER_ACTIVE_ALL,
567 * - HRTIMER_ACTIVE_SOFT, or
568 * - HRTIMER_ACTIVE_HARD.
571 __hrtimer_get_next_event(struct hrtimer_cpu_base *cpu_base, unsigned int active_mask)
574 struct hrtimer *next_timer = NULL;
575 ktime_t expires_next = KTIME_MAX;
577 if (!cpu_base->softirq_activated && (active_mask & HRTIMER_ACTIVE_SOFT)) {
578 active = cpu_base->active_bases & HRTIMER_ACTIVE_SOFT;
579 cpu_base->softirq_next_timer = NULL;
580 expires_next = __hrtimer_next_event_base(cpu_base, NULL,
583 next_timer = cpu_base->softirq_next_timer;
586 if (active_mask & HRTIMER_ACTIVE_HARD) {
587 active = cpu_base->active_bases & HRTIMER_ACTIVE_HARD;
588 cpu_base->next_timer = next_timer;
589 expires_next = __hrtimer_next_event_base(cpu_base, NULL, active,
596 static ktime_t hrtimer_update_next_event(struct hrtimer_cpu_base *cpu_base)
598 ktime_t expires_next, soft = KTIME_MAX;
601 * If the soft interrupt has already been activated, ignore the
602 * soft bases. They will be handled in the already raised soft
605 if (!cpu_base->softirq_activated) {
606 soft = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_SOFT);
608 * Update the soft expiry time. clock_settime() might have
611 cpu_base->softirq_expires_next = soft;
614 expires_next = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_HARD);
616 * If a softirq timer is expiring first, update cpu_base->next_timer
617 * and program the hardware with the soft expiry time.
619 if (expires_next > soft) {
620 cpu_base->next_timer = cpu_base->softirq_next_timer;
627 static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base)
629 ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset;
630 ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset;
631 ktime_t *offs_tai = &base->clock_base[HRTIMER_BASE_TAI].offset;
633 ktime_t now = ktime_get_update_offsets_now(&base->clock_was_set_seq,
634 offs_real, offs_boot, offs_tai);
636 base->clock_base[HRTIMER_BASE_REALTIME_SOFT].offset = *offs_real;
637 base->clock_base[HRTIMER_BASE_BOOTTIME_SOFT].offset = *offs_boot;
638 base->clock_base[HRTIMER_BASE_TAI_SOFT].offset = *offs_tai;
644 * Is the high resolution mode active ?
646 static inline int __hrtimer_hres_active(struct hrtimer_cpu_base *cpu_base)
648 return IS_ENABLED(CONFIG_HIGH_RES_TIMERS) ?
649 cpu_base->hres_active : 0;
652 static inline int hrtimer_hres_active(void)
654 return __hrtimer_hres_active(this_cpu_ptr(&hrtimer_bases));
657 static void __hrtimer_reprogram(struct hrtimer_cpu_base *cpu_base,
658 struct hrtimer *next_timer,
659 ktime_t expires_next)
661 cpu_base->expires_next = expires_next;
664 * If hres is not active, hardware does not have to be
667 * If a hang was detected in the last timer interrupt then we
668 * leave the hang delay active in the hardware. We want the
669 * system to make progress. That also prevents the following
671 * T1 expires 50ms from now
672 * T2 expires 5s from now
674 * T1 is removed, so this code is called and would reprogram
675 * the hardware to 5s from now. Any hrtimer_start after that
676 * will not reprogram the hardware due to hang_detected being
677 * set. So we'd effectively block all timers until the T2 event
680 if (!__hrtimer_hres_active(cpu_base) || cpu_base->hang_detected)
683 tick_program_event(expires_next, 1);
687 * Reprogram the event source with checking both queues for the
689 * Called with interrupts disabled and base->lock held
692 hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
694 ktime_t expires_next;
696 expires_next = hrtimer_update_next_event(cpu_base);
698 if (skip_equal && expires_next == cpu_base->expires_next)
701 __hrtimer_reprogram(cpu_base, cpu_base->next_timer, expires_next);
704 /* High resolution timer related functions */
705 #ifdef CONFIG_HIGH_RES_TIMERS
708 * High resolution timer enabled ?
710 static bool hrtimer_hres_enabled __read_mostly = true;
711 unsigned int hrtimer_resolution __read_mostly = LOW_RES_NSEC;
712 EXPORT_SYMBOL_GPL(hrtimer_resolution);
715 * Enable / Disable high resolution mode
717 static int __init setup_hrtimer_hres(char *str)
719 return (kstrtobool(str, &hrtimer_hres_enabled) == 0);
722 __setup("highres=", setup_hrtimer_hres);
725 * hrtimer_high_res_enabled - query, if the highres mode is enabled
727 static inline int hrtimer_is_hres_enabled(void)
729 return hrtimer_hres_enabled;
732 static void retrigger_next_event(void *arg);
735 * Switch to high resolution mode
737 static void hrtimer_switch_to_hres(void)
739 struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
741 if (tick_init_highres()) {
742 pr_warn("Could not switch to high resolution mode on CPU %u\n",
746 base->hres_active = 1;
747 hrtimer_resolution = HIGH_RES_NSEC;
749 tick_setup_sched_timer();
750 /* "Retrigger" the interrupt to get things going */
751 retrigger_next_event(NULL);
756 static inline int hrtimer_is_hres_enabled(void) { return 0; }
757 static inline void hrtimer_switch_to_hres(void) { }
759 #endif /* CONFIG_HIGH_RES_TIMERS */
761 * Retrigger next event is called after clock was set with interrupts
762 * disabled through an SMP function call or directly from low level
765 * This is only invoked when:
766 * - CONFIG_HIGH_RES_TIMERS is enabled.
767 * - CONFIG_NOHZ_COMMON is enabled
769 * For the other cases this function is empty and because the call sites
770 * are optimized out it vanishes as well, i.e. no need for lots of
773 static void retrigger_next_event(void *arg)
775 struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
778 * When high resolution mode or nohz is active, then the offsets of
779 * CLOCK_REALTIME/TAI/BOOTTIME have to be updated. Otherwise the
780 * next tick will take care of that.
782 * If high resolution mode is active then the next expiring timer
783 * must be reevaluated and the clock event device reprogrammed if
786 * In the NOHZ case the update of the offset and the reevaluation
787 * of the next expiring timer is enough. The return from the SMP
788 * function call will take care of the reprogramming in case the
789 * CPU was in a NOHZ idle sleep.
791 if (!__hrtimer_hres_active(base) && !tick_nohz_active)
794 raw_spin_lock(&base->lock);
795 hrtimer_update_base(base);
796 if (__hrtimer_hres_active(base))
797 hrtimer_force_reprogram(base, 0);
799 hrtimer_update_next_event(base);
800 raw_spin_unlock(&base->lock);
804 * When a timer is enqueued and expires earlier than the already enqueued
805 * timers, we have to check, whether it expires earlier than the timer for
806 * which the clock event device was armed.
808 * Called with interrupts disabled and base->cpu_base.lock held
810 static void hrtimer_reprogram(struct hrtimer *timer, bool reprogram)
812 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
813 struct hrtimer_clock_base *base = timer->base;
814 ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
816 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
819 * CLOCK_REALTIME timer might be requested with an absolute
820 * expiry time which is less than base->offset. Set it to 0.
825 if (timer->is_soft) {
827 * soft hrtimer could be started on a remote CPU. In this
828 * case softirq_expires_next needs to be updated on the
829 * remote CPU. The soft hrtimer will not expire before the
830 * first hard hrtimer on the remote CPU -
831 * hrtimer_check_target() prevents this case.
833 struct hrtimer_cpu_base *timer_cpu_base = base->cpu_base;
835 if (timer_cpu_base->softirq_activated)
838 if (!ktime_before(expires, timer_cpu_base->softirq_expires_next))
841 timer_cpu_base->softirq_next_timer = timer;
842 timer_cpu_base->softirq_expires_next = expires;
844 if (!ktime_before(expires, timer_cpu_base->expires_next) ||
850 * If the timer is not on the current cpu, we cannot reprogram
851 * the other cpus clock event device.
853 if (base->cpu_base != cpu_base)
856 if (expires >= cpu_base->expires_next)
860 * If the hrtimer interrupt is running, then it will reevaluate the
861 * clock bases and reprogram the clock event device.
863 if (cpu_base->in_hrtirq)
866 cpu_base->next_timer = timer;
868 __hrtimer_reprogram(cpu_base, timer, expires);
871 static bool update_needs_ipi(struct hrtimer_cpu_base *cpu_base,
874 struct hrtimer_clock_base *base;
879 * Update the base offsets unconditionally so the following
880 * checks whether the SMP function call is required works.
882 * The update is safe even when the remote CPU is in the hrtimer
883 * interrupt or the hrtimer soft interrupt and expiring affected
884 * bases. Either it will see the update before handling a base or
885 * it will see it when it finishes the processing and reevaluates
886 * the next expiring timer.
888 seq = cpu_base->clock_was_set_seq;
889 hrtimer_update_base(cpu_base);
892 * If the sequence did not change over the update then the
893 * remote CPU already handled it.
895 if (seq == cpu_base->clock_was_set_seq)
899 * If the remote CPU is currently handling an hrtimer interrupt, it
900 * will reevaluate the first expiring timer of all clock bases
901 * before reprogramming. Nothing to do here.
903 if (cpu_base->in_hrtirq)
907 * Walk the affected clock bases and check whether the first expiring
908 * timer in a clock base is moving ahead of the first expiring timer of
909 * @cpu_base. If so, the IPI must be invoked because per CPU clock
910 * event devices cannot be remotely reprogrammed.
912 active &= cpu_base->active_bases;
914 for_each_active_base(base, cpu_base, active) {
915 struct timerqueue_node *next;
917 next = timerqueue_getnext(&base->active);
918 expires = ktime_sub(next->expires, base->offset);
919 if (expires < cpu_base->expires_next)
922 /* Extra check for softirq clock bases */
923 if (base->clockid < HRTIMER_BASE_MONOTONIC_SOFT)
925 if (cpu_base->softirq_activated)
927 if (expires < cpu_base->softirq_expires_next)
934 * Clock was set. This might affect CLOCK_REALTIME, CLOCK_TAI and
935 * CLOCK_BOOTTIME (for late sleep time injection).
937 * This requires to update the offsets for these clocks
938 * vs. CLOCK_MONOTONIC. When high resolution timers are enabled, then this
939 * also requires to eventually reprogram the per CPU clock event devices
940 * when the change moves an affected timer ahead of the first expiring
941 * timer on that CPU. Obviously remote per CPU clock event devices cannot
942 * be reprogrammed. The other reason why an IPI has to be sent is when the
943 * system is in !HIGH_RES and NOHZ mode. The NOHZ mode updates the offsets
944 * in the tick, which obviously might be stopped, so this has to bring out
945 * the remote CPU which might sleep in idle to get this sorted.
947 void clock_was_set(unsigned int bases)
949 struct hrtimer_cpu_base *cpu_base = raw_cpu_ptr(&hrtimer_bases);
953 if (!__hrtimer_hres_active(cpu_base) && !tick_nohz_active)
956 if (!zalloc_cpumask_var(&mask, GFP_KERNEL)) {
957 on_each_cpu(retrigger_next_event, NULL, 1);
961 /* Avoid interrupting CPUs if possible */
963 for_each_online_cpu(cpu) {
966 cpu_base = &per_cpu(hrtimer_bases, cpu);
967 raw_spin_lock_irqsave(&cpu_base->lock, flags);
969 if (update_needs_ipi(cpu_base, bases))
970 cpumask_set_cpu(cpu, mask);
972 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
976 smp_call_function_many(mask, retrigger_next_event, NULL, 1);
979 free_cpumask_var(mask);
982 timerfd_clock_was_set();
985 static void clock_was_set_work(struct work_struct *work)
987 clock_was_set(CLOCK_SET_WALL);
990 static DECLARE_WORK(hrtimer_work, clock_was_set_work);
993 * Called from timekeeping code to reprogram the hrtimer interrupt device
994 * on all cpus and to notify timerfd.
996 void clock_was_set_delayed(void)
998 schedule_work(&hrtimer_work);
1002 * Called during resume either directly from via timekeeping_resume()
1003 * or in the case of s2idle from tick_unfreeze() to ensure that the
1004 * hrtimers are up to date.
1006 void hrtimers_resume_local(void)
1008 lockdep_assert_irqs_disabled();
1009 /* Retrigger on the local CPU */
1010 retrigger_next_event(NULL);
1014 * Counterpart to lock_hrtimer_base above:
1017 void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
1018 __releases(&timer->base->cpu_base->lock)
1020 raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
1024 * hrtimer_forward() - forward the timer expiry
1025 * @timer: hrtimer to forward
1026 * @now: forward past this time
1027 * @interval: the interval to forward
1029 * Forward the timer expiry so it will expire in the future.
1032 * This only updates the timer expiry value and does not requeue the timer.
1034 * There is also a variant of the function hrtimer_forward_now().
1036 * Context: Can be safely called from the callback function of @timer. If called
1037 * from other contexts @timer must neither be enqueued nor running the
1038 * callback and the caller needs to take care of serialization.
1040 * Return: The number of overruns are returned.
1042 u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
1047 delta = ktime_sub(now, hrtimer_get_expires(timer));
1052 if (WARN_ON(timer->state & HRTIMER_STATE_ENQUEUED))
1055 if (interval < hrtimer_resolution)
1056 interval = hrtimer_resolution;
1058 if (unlikely(delta >= interval)) {
1059 s64 incr = ktime_to_ns(interval);
1061 orun = ktime_divns(delta, incr);
1062 hrtimer_add_expires_ns(timer, incr * orun);
1063 if (hrtimer_get_expires_tv64(timer) > now)
1066 * This (and the ktime_add() below) is the
1067 * correction for exact:
1071 hrtimer_add_expires(timer, interval);
1075 EXPORT_SYMBOL_GPL(hrtimer_forward);
1078 * enqueue_hrtimer - internal function to (re)start a timer
1080 * The timer is inserted in expiry order. Insertion into the
1081 * red black tree is O(log(n)). Must hold the base lock.
1083 * Returns 1 when the new timer is the leftmost timer in the tree.
1085 static int enqueue_hrtimer(struct hrtimer *timer,
1086 struct hrtimer_clock_base *base,
1087 enum hrtimer_mode mode)
1089 debug_activate(timer, mode);
1090 WARN_ON_ONCE(!base->cpu_base->online);
1092 base->cpu_base->active_bases |= 1 << base->index;
1094 /* Pairs with the lockless read in hrtimer_is_queued() */
1095 WRITE_ONCE(timer->state, HRTIMER_STATE_ENQUEUED);
1097 return timerqueue_add(&base->active, &timer->node);
1101 * __remove_hrtimer - internal function to remove a timer
1103 * Caller must hold the base lock.
1105 * High resolution timer mode reprograms the clock event device when the
1106 * timer is the one which expires next. The caller can disable this by setting
1107 * reprogram to zero. This is useful, when the context does a reprogramming
1108 * anyway (e.g. timer interrupt)
1110 static void __remove_hrtimer(struct hrtimer *timer,
1111 struct hrtimer_clock_base *base,
1112 u8 newstate, int reprogram)
1114 struct hrtimer_cpu_base *cpu_base = base->cpu_base;
1115 u8 state = timer->state;
1117 /* Pairs with the lockless read in hrtimer_is_queued() */
1118 WRITE_ONCE(timer->state, newstate);
1119 if (!(state & HRTIMER_STATE_ENQUEUED))
1122 if (!timerqueue_del(&base->active, &timer->node))
1123 cpu_base->active_bases &= ~(1 << base->index);
1126 * Note: If reprogram is false we do not update
1127 * cpu_base->next_timer. This happens when we remove the first
1128 * timer on a remote cpu. No harm as we never dereference
1129 * cpu_base->next_timer. So the worst thing what can happen is
1130 * an superfluous call to hrtimer_force_reprogram() on the
1131 * remote cpu later on if the same timer gets enqueued again.
1133 if (reprogram && timer == cpu_base->next_timer)
1134 hrtimer_force_reprogram(cpu_base, 1);
1138 * remove hrtimer, called with base lock held
1141 remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base,
1142 bool restart, bool keep_local)
1144 u8 state = timer->state;
1146 if (state & HRTIMER_STATE_ENQUEUED) {
1150 * Remove the timer and force reprogramming when high
1151 * resolution mode is active and the timer is on the current
1152 * CPU. If we remove a timer on another CPU, reprogramming is
1153 * skipped. The interrupt event on this CPU is fired and
1154 * reprogramming happens in the interrupt handler. This is a
1155 * rare case and less expensive than a smp call.
1157 debug_deactivate(timer);
1158 reprogram = base->cpu_base == this_cpu_ptr(&hrtimer_bases);
1161 * If the timer is not restarted then reprogramming is
1162 * required if the timer is local. If it is local and about
1163 * to be restarted, avoid programming it twice (on removal
1164 * and a moment later when it's requeued).
1167 state = HRTIMER_STATE_INACTIVE;
1169 reprogram &= !keep_local;
1171 __remove_hrtimer(timer, base, state, reprogram);
1177 static inline ktime_t hrtimer_update_lowres(struct hrtimer *timer, ktime_t tim,
1178 const enum hrtimer_mode mode)
1180 #ifdef CONFIG_TIME_LOW_RES
1182 * CONFIG_TIME_LOW_RES indicates that the system has no way to return
1183 * granular time values. For relative timers we add hrtimer_resolution
1184 * (i.e. one jiffie) to prevent short timeouts.
1186 timer->is_rel = mode & HRTIMER_MODE_REL;
1188 tim = ktime_add_safe(tim, hrtimer_resolution);
1194 hrtimer_update_softirq_timer(struct hrtimer_cpu_base *cpu_base, bool reprogram)
1199 * Find the next SOFT expiration.
1201 expires = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_SOFT);
1204 * reprogramming needs to be triggered, even if the next soft
1205 * hrtimer expires at the same time than the next hard
1206 * hrtimer. cpu_base->softirq_expires_next needs to be updated!
1208 if (expires == KTIME_MAX)
1212 * cpu_base->*next_timer is recomputed by __hrtimer_get_next_event()
1213 * cpu_base->*expires_next is only set by hrtimer_reprogram()
1215 hrtimer_reprogram(cpu_base->softirq_next_timer, reprogram);
1218 static int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1219 u64 delta_ns, const enum hrtimer_mode mode,
1220 struct hrtimer_clock_base *base)
1222 struct hrtimer_clock_base *new_base;
1223 bool force_local, first;
1226 * If the timer is on the local cpu base and is the first expiring
1227 * timer then this might end up reprogramming the hardware twice
1228 * (on removal and on enqueue). To avoid that by prevent the
1229 * reprogram on removal, keep the timer local to the current CPU
1230 * and enforce reprogramming after it is queued no matter whether
1231 * it is the new first expiring timer again or not.
1233 force_local = base->cpu_base == this_cpu_ptr(&hrtimer_bases);
1234 force_local &= base->cpu_base->next_timer == timer;
1237 * Remove an active timer from the queue. In case it is not queued
1238 * on the current CPU, make sure that remove_hrtimer() updates the
1239 * remote data correctly.
1241 * If it's on the current CPU and the first expiring timer, then
1242 * skip reprogramming, keep the timer local and enforce
1243 * reprogramming later if it was the first expiring timer. This
1244 * avoids programming the underlying clock event twice (once at
1245 * removal and once after enqueue).
1247 remove_hrtimer(timer, base, true, force_local);
1249 if (mode & HRTIMER_MODE_REL)
1250 tim = ktime_add_safe(tim, base->get_time());
1252 tim = hrtimer_update_lowres(timer, tim, mode);
1254 hrtimer_set_expires_range_ns(timer, tim, delta_ns);
1256 /* Switch the timer base, if necessary: */
1258 new_base = switch_hrtimer_base(timer, base,
1259 mode & HRTIMER_MODE_PINNED);
1264 first = enqueue_hrtimer(timer, new_base, mode);
1269 * Timer was forced to stay on the current CPU to avoid
1270 * reprogramming on removal and enqueue. Force reprogram the
1271 * hardware by evaluating the new first expiring timer.
1273 hrtimer_force_reprogram(new_base->cpu_base, 1);
1278 * hrtimer_start_range_ns - (re)start an hrtimer
1279 * @timer: the timer to be added
1281 * @delta_ns: "slack" range for the timer
1282 * @mode: timer mode: absolute (HRTIMER_MODE_ABS) or
1283 * relative (HRTIMER_MODE_REL), and pinned (HRTIMER_MODE_PINNED);
1284 * softirq based mode is considered for debug purpose only!
1286 void hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1287 u64 delta_ns, const enum hrtimer_mode mode)
1289 struct hrtimer_clock_base *base;
1290 unsigned long flags;
1293 * Check whether the HRTIMER_MODE_SOFT bit and hrtimer.is_soft
1294 * match on CONFIG_PREEMPT_RT = n. With PREEMPT_RT check the hard
1295 * expiry mode because unmarked timers are moved to softirq expiry.
1297 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
1298 WARN_ON_ONCE(!(mode & HRTIMER_MODE_SOFT) ^ !timer->is_soft);
1300 WARN_ON_ONCE(!(mode & HRTIMER_MODE_HARD) ^ !timer->is_hard);
1302 base = lock_hrtimer_base(timer, &flags);
1304 if (__hrtimer_start_range_ns(timer, tim, delta_ns, mode, base))
1305 hrtimer_reprogram(timer, true);
1307 unlock_hrtimer_base(timer, &flags);
1309 EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
1312 * hrtimer_try_to_cancel - try to deactivate a timer
1313 * @timer: hrtimer to stop
1317 * * 0 when the timer was not active
1318 * * 1 when the timer was active
1319 * * -1 when the timer is currently executing the callback function and
1322 int hrtimer_try_to_cancel(struct hrtimer *timer)
1324 struct hrtimer_clock_base *base;
1325 unsigned long flags;
1329 * Check lockless first. If the timer is not active (neither
1330 * enqueued nor running the callback, nothing to do here. The
1331 * base lock does not serialize against a concurrent enqueue,
1332 * so we can avoid taking it.
1334 if (!hrtimer_active(timer))
1337 base = lock_hrtimer_base(timer, &flags);
1339 if (!hrtimer_callback_running(timer))
1340 ret = remove_hrtimer(timer, base, false, false);
1342 unlock_hrtimer_base(timer, &flags);
1347 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1349 #ifdef CONFIG_PREEMPT_RT
1350 static void hrtimer_cpu_base_init_expiry_lock(struct hrtimer_cpu_base *base)
1352 spin_lock_init(&base->softirq_expiry_lock);
1355 static void hrtimer_cpu_base_lock_expiry(struct hrtimer_cpu_base *base)
1357 spin_lock(&base->softirq_expiry_lock);
1360 static void hrtimer_cpu_base_unlock_expiry(struct hrtimer_cpu_base *base)
1362 spin_unlock(&base->softirq_expiry_lock);
1366 * The counterpart to hrtimer_cancel_wait_running().
1368 * If there is a waiter for cpu_base->expiry_lock, then it was waiting for
1369 * the timer callback to finish. Drop expiry_lock and reacquire it. That
1370 * allows the waiter to acquire the lock and make progress.
1372 static void hrtimer_sync_wait_running(struct hrtimer_cpu_base *cpu_base,
1373 unsigned long flags)
1375 if (atomic_read(&cpu_base->timer_waiters)) {
1376 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1377 spin_unlock(&cpu_base->softirq_expiry_lock);
1378 spin_lock(&cpu_base->softirq_expiry_lock);
1379 raw_spin_lock_irq(&cpu_base->lock);
1384 * This function is called on PREEMPT_RT kernels when the fast path
1385 * deletion of a timer failed because the timer callback function was
1388 * This prevents priority inversion: if the soft irq thread is preempted
1389 * in the middle of a timer callback, then calling del_timer_sync() can
1390 * lead to two issues:
1392 * - If the caller is on a remote CPU then it has to spin wait for the timer
1393 * handler to complete. This can result in unbound priority inversion.
1395 * - If the caller originates from the task which preempted the timer
1396 * handler on the same CPU, then spin waiting for the timer handler to
1397 * complete is never going to end.
1399 void hrtimer_cancel_wait_running(const struct hrtimer *timer)
1401 /* Lockless read. Prevent the compiler from reloading it below */
1402 struct hrtimer_clock_base *base = READ_ONCE(timer->base);
1405 * Just relax if the timer expires in hard interrupt context or if
1406 * it is currently on the migration base.
1408 if (!timer->is_soft || is_migration_base(base)) {
1414 * Mark the base as contended and grab the expiry lock, which is
1415 * held by the softirq across the timer callback. Drop the lock
1416 * immediately so the softirq can expire the next timer. In theory
1417 * the timer could already be running again, but that's more than
1418 * unlikely and just causes another wait loop.
1420 atomic_inc(&base->cpu_base->timer_waiters);
1421 spin_lock_bh(&base->cpu_base->softirq_expiry_lock);
1422 atomic_dec(&base->cpu_base->timer_waiters);
1423 spin_unlock_bh(&base->cpu_base->softirq_expiry_lock);
1427 hrtimer_cpu_base_init_expiry_lock(struct hrtimer_cpu_base *base) { }
1429 hrtimer_cpu_base_lock_expiry(struct hrtimer_cpu_base *base) { }
1431 hrtimer_cpu_base_unlock_expiry(struct hrtimer_cpu_base *base) { }
1432 static inline void hrtimer_sync_wait_running(struct hrtimer_cpu_base *base,
1433 unsigned long flags) { }
1437 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1438 * @timer: the timer to be cancelled
1441 * 0 when the timer was not active
1442 * 1 when the timer was active
1444 int hrtimer_cancel(struct hrtimer *timer)
1449 ret = hrtimer_try_to_cancel(timer);
1452 hrtimer_cancel_wait_running(timer);
1456 EXPORT_SYMBOL_GPL(hrtimer_cancel);
1459 * __hrtimer_get_remaining - get remaining time for the timer
1460 * @timer: the timer to read
1461 * @adjust: adjust relative timers when CONFIG_TIME_LOW_RES=y
1463 ktime_t __hrtimer_get_remaining(const struct hrtimer *timer, bool adjust)
1465 unsigned long flags;
1468 lock_hrtimer_base(timer, &flags);
1469 if (IS_ENABLED(CONFIG_TIME_LOW_RES) && adjust)
1470 rem = hrtimer_expires_remaining_adjusted(timer);
1472 rem = hrtimer_expires_remaining(timer);
1473 unlock_hrtimer_base(timer, &flags);
1477 EXPORT_SYMBOL_GPL(__hrtimer_get_remaining);
1479 #ifdef CONFIG_NO_HZ_COMMON
1481 * hrtimer_get_next_event - get the time until next expiry event
1483 * Returns the next expiry time or KTIME_MAX if no timer is pending.
1485 u64 hrtimer_get_next_event(void)
1487 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1488 u64 expires = KTIME_MAX;
1489 unsigned long flags;
1491 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1493 if (!__hrtimer_hres_active(cpu_base))
1494 expires = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_ALL);
1496 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1502 * hrtimer_next_event_without - time until next expiry event w/o one timer
1503 * @exclude: timer to exclude
1505 * Returns the next expiry time over all timers except for the @exclude one or
1506 * KTIME_MAX if none of them is pending.
1508 u64 hrtimer_next_event_without(const struct hrtimer *exclude)
1510 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1511 u64 expires = KTIME_MAX;
1512 unsigned long flags;
1514 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1516 if (__hrtimer_hres_active(cpu_base)) {
1517 unsigned int active;
1519 if (!cpu_base->softirq_activated) {
1520 active = cpu_base->active_bases & HRTIMER_ACTIVE_SOFT;
1521 expires = __hrtimer_next_event_base(cpu_base, exclude,
1524 active = cpu_base->active_bases & HRTIMER_ACTIVE_HARD;
1525 expires = __hrtimer_next_event_base(cpu_base, exclude, active,
1529 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1535 static inline int hrtimer_clockid_to_base(clockid_t clock_id)
1537 if (likely(clock_id < MAX_CLOCKS)) {
1538 int base = hrtimer_clock_to_base_table[clock_id];
1540 if (likely(base != HRTIMER_MAX_CLOCK_BASES))
1543 WARN(1, "Invalid clockid %d. Using MONOTONIC\n", clock_id);
1544 return HRTIMER_BASE_MONOTONIC;
1547 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1548 enum hrtimer_mode mode)
1550 bool softtimer = !!(mode & HRTIMER_MODE_SOFT);
1551 struct hrtimer_cpu_base *cpu_base;
1555 * On PREEMPT_RT enabled kernels hrtimers which are not explicitly
1556 * marked for hard interrupt expiry mode are moved into soft
1557 * interrupt context for latency reasons and because the callbacks
1558 * can invoke functions which might sleep on RT, e.g. spin_lock().
1560 if (IS_ENABLED(CONFIG_PREEMPT_RT) && !(mode & HRTIMER_MODE_HARD))
1563 memset(timer, 0, sizeof(struct hrtimer));
1565 cpu_base = raw_cpu_ptr(&hrtimer_bases);
1568 * POSIX magic: Relative CLOCK_REALTIME timers are not affected by
1569 * clock modifications, so they needs to become CLOCK_MONOTONIC to
1570 * ensure POSIX compliance.
1572 if (clock_id == CLOCK_REALTIME && mode & HRTIMER_MODE_REL)
1573 clock_id = CLOCK_MONOTONIC;
1575 base = softtimer ? HRTIMER_MAX_CLOCK_BASES / 2 : 0;
1576 base += hrtimer_clockid_to_base(clock_id);
1577 timer->is_soft = softtimer;
1578 timer->is_hard = !!(mode & HRTIMER_MODE_HARD);
1579 timer->base = &cpu_base->clock_base[base];
1580 timerqueue_init(&timer->node);
1584 * hrtimer_init - initialize a timer to the given clock
1585 * @timer: the timer to be initialized
1586 * @clock_id: the clock to be used
1587 * @mode: The modes which are relevant for initialization:
1588 * HRTIMER_MODE_ABS, HRTIMER_MODE_REL, HRTIMER_MODE_ABS_SOFT,
1589 * HRTIMER_MODE_REL_SOFT
1591 * The PINNED variants of the above can be handed in,
1592 * but the PINNED bit is ignored as pinning happens
1593 * when the hrtimer is started
1595 void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1596 enum hrtimer_mode mode)
1598 debug_init(timer, clock_id, mode);
1599 __hrtimer_init(timer, clock_id, mode);
1601 EXPORT_SYMBOL_GPL(hrtimer_init);
1604 * A timer is active, when it is enqueued into the rbtree or the
1605 * callback function is running or it's in the state of being migrated
1608 * It is important for this function to not return a false negative.
1610 bool hrtimer_active(const struct hrtimer *timer)
1612 struct hrtimer_clock_base *base;
1616 base = READ_ONCE(timer->base);
1617 seq = raw_read_seqcount_begin(&base->seq);
1619 if (timer->state != HRTIMER_STATE_INACTIVE ||
1620 base->running == timer)
1623 } while (read_seqcount_retry(&base->seq, seq) ||
1624 base != READ_ONCE(timer->base));
1628 EXPORT_SYMBOL_GPL(hrtimer_active);
1631 * The write_seqcount_barrier()s in __run_hrtimer() split the thing into 3
1632 * distinct sections:
1634 * - queued: the timer is queued
1635 * - callback: the timer is being ran
1636 * - post: the timer is inactive or (re)queued
1638 * On the read side we ensure we observe timer->state and cpu_base->running
1639 * from the same section, if anything changed while we looked at it, we retry.
1640 * This includes timer->base changing because sequence numbers alone are
1641 * insufficient for that.
1643 * The sequence numbers are required because otherwise we could still observe
1644 * a false negative if the read side got smeared over multiple consecutive
1645 * __run_hrtimer() invocations.
1648 static void __run_hrtimer(struct hrtimer_cpu_base *cpu_base,
1649 struct hrtimer_clock_base *base,
1650 struct hrtimer *timer, ktime_t *now,
1651 unsigned long flags) __must_hold(&cpu_base->lock)
1653 enum hrtimer_restart (*fn)(struct hrtimer *);
1654 bool expires_in_hardirq;
1657 lockdep_assert_held(&cpu_base->lock);
1659 debug_deactivate(timer);
1660 base->running = timer;
1663 * Separate the ->running assignment from the ->state assignment.
1665 * As with a regular write barrier, this ensures the read side in
1666 * hrtimer_active() cannot observe base->running == NULL &&
1667 * timer->state == INACTIVE.
1669 raw_write_seqcount_barrier(&base->seq);
1671 __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE, 0);
1672 fn = timer->function;
1675 * Clear the 'is relative' flag for the TIME_LOW_RES case. If the
1676 * timer is restarted with a period then it becomes an absolute
1677 * timer. If its not restarted it does not matter.
1679 if (IS_ENABLED(CONFIG_TIME_LOW_RES))
1680 timer->is_rel = false;
1683 * The timer is marked as running in the CPU base, so it is
1684 * protected against migration to a different CPU even if the lock
1687 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1688 trace_hrtimer_expire_entry(timer, now);
1689 expires_in_hardirq = lockdep_hrtimer_enter(timer);
1691 restart = fn(timer);
1693 lockdep_hrtimer_exit(expires_in_hardirq);
1694 trace_hrtimer_expire_exit(timer);
1695 raw_spin_lock_irq(&cpu_base->lock);
1698 * Note: We clear the running state after enqueue_hrtimer and
1699 * we do not reprogram the event hardware. Happens either in
1700 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1702 * Note: Because we dropped the cpu_base->lock above,
1703 * hrtimer_start_range_ns() can have popped in and enqueued the timer
1706 if (restart != HRTIMER_NORESTART &&
1707 !(timer->state & HRTIMER_STATE_ENQUEUED))
1708 enqueue_hrtimer(timer, base, HRTIMER_MODE_ABS);
1711 * Separate the ->running assignment from the ->state assignment.
1713 * As with a regular write barrier, this ensures the read side in
1714 * hrtimer_active() cannot observe base->running.timer == NULL &&
1715 * timer->state == INACTIVE.
1717 raw_write_seqcount_barrier(&base->seq);
1719 WARN_ON_ONCE(base->running != timer);
1720 base->running = NULL;
1723 static void __hrtimer_run_queues(struct hrtimer_cpu_base *cpu_base, ktime_t now,
1724 unsigned long flags, unsigned int active_mask)
1726 struct hrtimer_clock_base *base;
1727 unsigned int active = cpu_base->active_bases & active_mask;
1729 for_each_active_base(base, cpu_base, active) {
1730 struct timerqueue_node *node;
1733 basenow = ktime_add(now, base->offset);
1735 while ((node = timerqueue_getnext(&base->active))) {
1736 struct hrtimer *timer;
1738 timer = container_of(node, struct hrtimer, node);
1741 * The immediate goal for using the softexpires is
1742 * minimizing wakeups, not running timers at the
1743 * earliest interrupt after their soft expiration.
1744 * This allows us to avoid using a Priority Search
1745 * Tree, which can answer a stabbing query for
1746 * overlapping intervals and instead use the simple
1747 * BST we already have.
1748 * We don't add extra wakeups by delaying timers that
1749 * are right-of a not yet expired timer, because that
1750 * timer will have to trigger a wakeup anyway.
1752 if (basenow < hrtimer_get_softexpires_tv64(timer))
1755 __run_hrtimer(cpu_base, base, timer, &basenow, flags);
1756 if (active_mask == HRTIMER_ACTIVE_SOFT)
1757 hrtimer_sync_wait_running(cpu_base, flags);
1762 static __latent_entropy void hrtimer_run_softirq(struct softirq_action *h)
1764 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1765 unsigned long flags;
1768 hrtimer_cpu_base_lock_expiry(cpu_base);
1769 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1771 now = hrtimer_update_base(cpu_base);
1772 __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_SOFT);
1774 cpu_base->softirq_activated = 0;
1775 hrtimer_update_softirq_timer(cpu_base, true);
1777 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1778 hrtimer_cpu_base_unlock_expiry(cpu_base);
1781 #ifdef CONFIG_HIGH_RES_TIMERS
1784 * High resolution timer interrupt
1785 * Called with interrupts disabled
1787 void hrtimer_interrupt(struct clock_event_device *dev)
1789 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1790 ktime_t expires_next, now, entry_time, delta;
1791 unsigned long flags;
1794 BUG_ON(!cpu_base->hres_active);
1795 cpu_base->nr_events++;
1796 dev->next_event = KTIME_MAX;
1798 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1799 entry_time = now = hrtimer_update_base(cpu_base);
1801 cpu_base->in_hrtirq = 1;
1803 * We set expires_next to KTIME_MAX here with cpu_base->lock
1804 * held to prevent that a timer is enqueued in our queue via
1805 * the migration code. This does not affect enqueueing of
1806 * timers which run their callback and need to be requeued on
1809 cpu_base->expires_next = KTIME_MAX;
1811 if (!ktime_before(now, cpu_base->softirq_expires_next)) {
1812 cpu_base->softirq_expires_next = KTIME_MAX;
1813 cpu_base->softirq_activated = 1;
1814 raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1817 __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_HARD);
1819 /* Reevaluate the clock bases for the [soft] next expiry */
1820 expires_next = hrtimer_update_next_event(cpu_base);
1822 * Store the new expiry value so the migration code can verify
1825 cpu_base->expires_next = expires_next;
1826 cpu_base->in_hrtirq = 0;
1827 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1829 /* Reprogramming necessary ? */
1830 if (!tick_program_event(expires_next, 0)) {
1831 cpu_base->hang_detected = 0;
1836 * The next timer was already expired due to:
1838 * - long lasting callbacks
1839 * - being scheduled away when running in a VM
1841 * We need to prevent that we loop forever in the hrtimer
1842 * interrupt routine. We give it 3 attempts to avoid
1843 * overreacting on some spurious event.
1845 * Acquire base lock for updating the offsets and retrieving
1848 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1849 now = hrtimer_update_base(cpu_base);
1850 cpu_base->nr_retries++;
1854 * Give the system a chance to do something else than looping
1855 * here. We stored the entry time, so we know exactly how long
1856 * we spent here. We schedule the next event this amount of
1859 cpu_base->nr_hangs++;
1860 cpu_base->hang_detected = 1;
1861 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1863 delta = ktime_sub(now, entry_time);
1864 if ((unsigned int)delta > cpu_base->max_hang_time)
1865 cpu_base->max_hang_time = (unsigned int) delta;
1867 * Limit it to a sensible value as we enforce a longer
1868 * delay. Give the CPU at least 100ms to catch up.
1870 if (delta > 100 * NSEC_PER_MSEC)
1871 expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1873 expires_next = ktime_add(now, delta);
1874 tick_program_event(expires_next, 1);
1875 pr_warn_once("hrtimer: interrupt took %llu ns\n", ktime_to_ns(delta));
1878 /* called with interrupts disabled */
1879 static inline void __hrtimer_peek_ahead_timers(void)
1881 struct tick_device *td;
1883 if (!hrtimer_hres_active())
1886 td = this_cpu_ptr(&tick_cpu_device);
1887 if (td && td->evtdev)
1888 hrtimer_interrupt(td->evtdev);
1891 #else /* CONFIG_HIGH_RES_TIMERS */
1893 static inline void __hrtimer_peek_ahead_timers(void) { }
1895 #endif /* !CONFIG_HIGH_RES_TIMERS */
1898 * Called from run_local_timers in hardirq context every jiffy
1900 void hrtimer_run_queues(void)
1902 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1903 unsigned long flags;
1906 if (__hrtimer_hres_active(cpu_base))
1910 * This _is_ ugly: We have to check periodically, whether we
1911 * can switch to highres and / or nohz mode. The clocksource
1912 * switch happens with xtime_lock held. Notification from
1913 * there only sets the check bit in the tick_oneshot code,
1914 * otherwise we might deadlock vs. xtime_lock.
1916 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) {
1917 hrtimer_switch_to_hres();
1921 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1922 now = hrtimer_update_base(cpu_base);
1924 if (!ktime_before(now, cpu_base->softirq_expires_next)) {
1925 cpu_base->softirq_expires_next = KTIME_MAX;
1926 cpu_base->softirq_activated = 1;
1927 raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1930 __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_HARD);
1931 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1935 * Sleep related functions:
1937 static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1939 struct hrtimer_sleeper *t =
1940 container_of(timer, struct hrtimer_sleeper, timer);
1941 struct task_struct *task = t->task;
1945 wake_up_process(task);
1947 return HRTIMER_NORESTART;
1951 * hrtimer_sleeper_start_expires - Start a hrtimer sleeper timer
1952 * @sl: sleeper to be started
1953 * @mode: timer mode abs/rel
1955 * Wrapper around hrtimer_start_expires() for hrtimer_sleeper based timers
1956 * to allow PREEMPT_RT to tweak the delivery mode (soft/hardirq context)
1958 void hrtimer_sleeper_start_expires(struct hrtimer_sleeper *sl,
1959 enum hrtimer_mode mode)
1962 * Make the enqueue delivery mode check work on RT. If the sleeper
1963 * was initialized for hard interrupt delivery, force the mode bit.
1964 * This is a special case for hrtimer_sleepers because
1965 * hrtimer_init_sleeper() determines the delivery mode on RT so the
1966 * fiddling with this decision is avoided at the call sites.
1968 if (IS_ENABLED(CONFIG_PREEMPT_RT) && sl->timer.is_hard)
1969 mode |= HRTIMER_MODE_HARD;
1971 hrtimer_start_expires(&sl->timer, mode);
1973 EXPORT_SYMBOL_GPL(hrtimer_sleeper_start_expires);
1975 static void __hrtimer_init_sleeper(struct hrtimer_sleeper *sl,
1976 clockid_t clock_id, enum hrtimer_mode mode)
1979 * On PREEMPT_RT enabled kernels hrtimers which are not explicitly
1980 * marked for hard interrupt expiry mode are moved into soft
1981 * interrupt context either for latency reasons or because the
1982 * hrtimer callback takes regular spinlocks or invokes other
1983 * functions which are not suitable for hard interrupt context on
1986 * The hrtimer_sleeper callback is RT compatible in hard interrupt
1987 * context, but there is a latency concern: Untrusted userspace can
1988 * spawn many threads which arm timers for the same expiry time on
1989 * the same CPU. That causes a latency spike due to the wakeup of
1990 * a gazillion threads.
1992 * OTOH, privileged real-time user space applications rely on the
1993 * low latency of hard interrupt wakeups. If the current task is in
1994 * a real-time scheduling class, mark the mode for hard interrupt
1997 if (IS_ENABLED(CONFIG_PREEMPT_RT)) {
1998 if (task_is_realtime(current) && !(mode & HRTIMER_MODE_SOFT))
1999 mode |= HRTIMER_MODE_HARD;
2002 __hrtimer_init(&sl->timer, clock_id, mode);
2003 sl->timer.function = hrtimer_wakeup;
2008 * hrtimer_init_sleeper - initialize sleeper to the given clock
2009 * @sl: sleeper to be initialized
2010 * @clock_id: the clock to be used
2011 * @mode: timer mode abs/rel
2013 void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, clockid_t clock_id,
2014 enum hrtimer_mode mode)
2016 debug_init(&sl->timer, clock_id, mode);
2017 __hrtimer_init_sleeper(sl, clock_id, mode);
2020 EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
2022 int nanosleep_copyout(struct restart_block *restart, struct timespec64 *ts)
2024 switch(restart->nanosleep.type) {
2025 #ifdef CONFIG_COMPAT_32BIT_TIME
2027 if (put_old_timespec32(ts, restart->nanosleep.compat_rmtp))
2032 if (put_timespec64(ts, restart->nanosleep.rmtp))
2038 return -ERESTART_RESTARTBLOCK;
2041 static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
2043 struct restart_block *restart;
2046 set_current_state(TASK_INTERRUPTIBLE|TASK_FREEZABLE);
2047 hrtimer_sleeper_start_expires(t, mode);
2049 if (likely(t->task))
2052 hrtimer_cancel(&t->timer);
2053 mode = HRTIMER_MODE_ABS;
2055 } while (t->task && !signal_pending(current));
2057 __set_current_state(TASK_RUNNING);
2062 restart = ¤t->restart_block;
2063 if (restart->nanosleep.type != TT_NONE) {
2064 ktime_t rem = hrtimer_expires_remaining(&t->timer);
2065 struct timespec64 rmt;
2069 rmt = ktime_to_timespec64(rem);
2071 return nanosleep_copyout(restart, &rmt);
2073 return -ERESTART_RESTARTBLOCK;
2076 static long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
2078 struct hrtimer_sleeper t;
2081 hrtimer_init_sleeper_on_stack(&t, restart->nanosleep.clockid,
2083 hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
2084 ret = do_nanosleep(&t, HRTIMER_MODE_ABS);
2085 destroy_hrtimer_on_stack(&t.timer);
2089 long hrtimer_nanosleep(ktime_t rqtp, const enum hrtimer_mode mode,
2090 const clockid_t clockid)
2092 struct restart_block *restart;
2093 struct hrtimer_sleeper t;
2097 slack = current->timer_slack_ns;
2098 if (rt_task(current))
2101 hrtimer_init_sleeper_on_stack(&t, clockid, mode);
2102 hrtimer_set_expires_range_ns(&t.timer, rqtp, slack);
2103 ret = do_nanosleep(&t, mode);
2104 if (ret != -ERESTART_RESTARTBLOCK)
2107 /* Absolute timers do not update the rmtp value and restart: */
2108 if (mode == HRTIMER_MODE_ABS) {
2109 ret = -ERESTARTNOHAND;
2113 restart = ¤t->restart_block;
2114 restart->nanosleep.clockid = t.timer.base->clockid;
2115 restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
2116 set_restart_fn(restart, hrtimer_nanosleep_restart);
2118 destroy_hrtimer_on_stack(&t.timer);
2124 SYSCALL_DEFINE2(nanosleep, struct __kernel_timespec __user *, rqtp,
2125 struct __kernel_timespec __user *, rmtp)
2127 struct timespec64 tu;
2129 if (get_timespec64(&tu, rqtp))
2132 if (!timespec64_valid(&tu))
2135 current->restart_block.fn = do_no_restart_syscall;
2136 current->restart_block.nanosleep.type = rmtp ? TT_NATIVE : TT_NONE;
2137 current->restart_block.nanosleep.rmtp = rmtp;
2138 return hrtimer_nanosleep(timespec64_to_ktime(tu), HRTIMER_MODE_REL,
2144 #ifdef CONFIG_COMPAT_32BIT_TIME
2146 SYSCALL_DEFINE2(nanosleep_time32, struct old_timespec32 __user *, rqtp,
2147 struct old_timespec32 __user *, rmtp)
2149 struct timespec64 tu;
2151 if (get_old_timespec32(&tu, rqtp))
2154 if (!timespec64_valid(&tu))
2157 current->restart_block.fn = do_no_restart_syscall;
2158 current->restart_block.nanosleep.type = rmtp ? TT_COMPAT : TT_NONE;
2159 current->restart_block.nanosleep.compat_rmtp = rmtp;
2160 return hrtimer_nanosleep(timespec64_to_ktime(tu), HRTIMER_MODE_REL,
2166 * Functions related to boot-time initialization:
2168 int hrtimers_prepare_cpu(unsigned int cpu)
2170 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
2173 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
2174 struct hrtimer_clock_base *clock_b = &cpu_base->clock_base[i];
2176 clock_b->cpu_base = cpu_base;
2177 seqcount_raw_spinlock_init(&clock_b->seq, &cpu_base->lock);
2178 timerqueue_init_head(&clock_b->active);
2181 cpu_base->cpu = cpu;
2182 cpu_base->active_bases = 0;
2183 cpu_base->hres_active = 0;
2184 cpu_base->hang_detected = 0;
2185 cpu_base->next_timer = NULL;
2186 cpu_base->softirq_next_timer = NULL;
2187 cpu_base->expires_next = KTIME_MAX;
2188 cpu_base->softirq_expires_next = KTIME_MAX;
2189 cpu_base->online = 1;
2190 hrtimer_cpu_base_init_expiry_lock(cpu_base);
2194 #ifdef CONFIG_HOTPLUG_CPU
2196 static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
2197 struct hrtimer_clock_base *new_base)
2199 struct hrtimer *timer;
2200 struct timerqueue_node *node;
2202 while ((node = timerqueue_getnext(&old_base->active))) {
2203 timer = container_of(node, struct hrtimer, node);
2204 BUG_ON(hrtimer_callback_running(timer));
2205 debug_deactivate(timer);
2208 * Mark it as ENQUEUED not INACTIVE otherwise the
2209 * timer could be seen as !active and just vanish away
2210 * under us on another CPU
2212 __remove_hrtimer(timer, old_base, HRTIMER_STATE_ENQUEUED, 0);
2213 timer->base = new_base;
2215 * Enqueue the timers on the new cpu. This does not
2216 * reprogram the event device in case the timer
2217 * expires before the earliest on this CPU, but we run
2218 * hrtimer_interrupt after we migrated everything to
2219 * sort out already expired timers and reprogram the
2222 enqueue_hrtimer(timer, new_base, HRTIMER_MODE_ABS);
2226 int hrtimers_cpu_dying(unsigned int dying_cpu)
2228 struct hrtimer_cpu_base *old_base, *new_base;
2229 int i, ncpu = cpumask_first(cpu_active_mask);
2231 tick_cancel_sched_timer(dying_cpu);
2233 old_base = this_cpu_ptr(&hrtimer_bases);
2234 new_base = &per_cpu(hrtimer_bases, ncpu);
2237 * The caller is globally serialized and nobody else
2238 * takes two locks at once, deadlock is not possible.
2240 raw_spin_lock(&old_base->lock);
2241 raw_spin_lock_nested(&new_base->lock, SINGLE_DEPTH_NESTING);
2243 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
2244 migrate_hrtimer_list(&old_base->clock_base[i],
2245 &new_base->clock_base[i]);
2249 * The migration might have changed the first expiring softirq
2250 * timer on this CPU. Update it.
2252 __hrtimer_get_next_event(new_base, HRTIMER_ACTIVE_SOFT);
2253 /* Tell the other CPU to retrigger the next event */
2254 smp_call_function_single(ncpu, retrigger_next_event, NULL, 0);
2256 raw_spin_unlock(&new_base->lock);
2257 old_base->online = 0;
2258 raw_spin_unlock(&old_base->lock);
2263 #endif /* CONFIG_HOTPLUG_CPU */
2265 void __init hrtimers_init(void)
2267 hrtimers_prepare_cpu(smp_processor_id());
2268 open_softirq(HRTIMER_SOFTIRQ, hrtimer_run_softirq);
2272 * schedule_hrtimeout_range_clock - sleep until timeout
2273 * @expires: timeout value (ktime_t)
2274 * @delta: slack in expires timeout (ktime_t) for SCHED_OTHER tasks
2276 * @clock_id: timer clock to be used
2279 schedule_hrtimeout_range_clock(ktime_t *expires, u64 delta,
2280 const enum hrtimer_mode mode, clockid_t clock_id)
2282 struct hrtimer_sleeper t;
2285 * Optimize when a zero timeout value is given. It does not
2286 * matter whether this is an absolute or a relative time.
2288 if (expires && *expires == 0) {
2289 __set_current_state(TASK_RUNNING);
2294 * A NULL parameter means "infinite"
2302 * Override any slack passed by the user if under
2305 if (rt_task(current))
2308 hrtimer_init_sleeper_on_stack(&t, clock_id, mode);
2309 hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
2310 hrtimer_sleeper_start_expires(&t, mode);
2315 hrtimer_cancel(&t.timer);
2316 destroy_hrtimer_on_stack(&t.timer);
2318 __set_current_state(TASK_RUNNING);
2320 return !t.task ? 0 : -EINTR;
2322 EXPORT_SYMBOL_GPL(schedule_hrtimeout_range_clock);
2325 * schedule_hrtimeout_range - sleep until timeout
2326 * @expires: timeout value (ktime_t)
2327 * @delta: slack in expires timeout (ktime_t) for SCHED_OTHER tasks
2330 * Make the current task sleep until the given expiry time has
2331 * elapsed. The routine will return immediately unless
2332 * the current task state has been set (see set_current_state()).
2334 * The @delta argument gives the kernel the freedom to schedule the
2335 * actual wakeup to a time that is both power and performance friendly
2336 * for regular (non RT/DL) tasks.
2337 * The kernel give the normal best effort behavior for "@expires+@delta",
2338 * but may decide to fire the timer earlier, but no earlier than @expires.
2340 * You can set the task state as follows -
2342 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
2343 * pass before the routine returns unless the current task is explicitly
2344 * woken up, (e.g. by wake_up_process()).
2346 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
2347 * delivered to the current task or the current task is explicitly woken
2350 * The current task state is guaranteed to be TASK_RUNNING when this
2353 * Returns 0 when the timer has expired. If the task was woken before the
2354 * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or
2355 * by an explicit wakeup, it returns -EINTR.
2357 int __sched schedule_hrtimeout_range(ktime_t *expires, u64 delta,
2358 const enum hrtimer_mode mode)
2360 return schedule_hrtimeout_range_clock(expires, delta, mode,
2363 EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
2366 * schedule_hrtimeout - sleep until timeout
2367 * @expires: timeout value (ktime_t)
2370 * Make the current task sleep until the given expiry time has
2371 * elapsed. The routine will return immediately unless
2372 * the current task state has been set (see set_current_state()).
2374 * You can set the task state as follows -
2376 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
2377 * pass before the routine returns unless the current task is explicitly
2378 * woken up, (e.g. by wake_up_process()).
2380 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
2381 * delivered to the current task or the current task is explicitly woken
2384 * The current task state is guaranteed to be TASK_RUNNING when this
2387 * Returns 0 when the timer has expired. If the task was woken before the
2388 * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or
2389 * by an explicit wakeup, it returns -EINTR.
2391 int __sched schedule_hrtimeout(ktime_t *expires,
2392 const enum hrtimer_mode mode)
2394 return schedule_hrtimeout_range(expires, 0, mode);
2396 EXPORT_SYMBOL_GPL(schedule_hrtimeout);