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_BOOTTIME,
95 .clockid = CLOCK_BOOTTIME,
96 .get_time = &ktime_get_boottime,
99 .index = HRTIMER_BASE_TAI,
100 .clockid = CLOCK_TAI,
101 .get_time = &ktime_get_clocktai,
104 .index = HRTIMER_BASE_MONOTONIC_SOFT,
105 .clockid = CLOCK_MONOTONIC,
106 .get_time = &ktime_get,
109 .index = HRTIMER_BASE_REALTIME_SOFT,
110 .clockid = CLOCK_REALTIME,
111 .get_time = &ktime_get_real,
114 .index = HRTIMER_BASE_BOOTTIME_SOFT,
115 .clockid = CLOCK_BOOTTIME,
116 .get_time = &ktime_get_boottime,
119 .index = HRTIMER_BASE_TAI_SOFT,
120 .clockid = CLOCK_TAI,
121 .get_time = &ktime_get_clocktai,
126 static const int hrtimer_clock_to_base_table[MAX_CLOCKS] = {
127 /* Make sure we catch unsupported clockids */
128 [0 ... MAX_CLOCKS - 1] = HRTIMER_MAX_CLOCK_BASES,
130 [CLOCK_REALTIME] = HRTIMER_BASE_REALTIME,
131 [CLOCK_MONOTONIC] = HRTIMER_BASE_MONOTONIC,
132 [CLOCK_BOOTTIME] = HRTIMER_BASE_BOOTTIME,
133 [CLOCK_TAI] = HRTIMER_BASE_TAI,
137 * Functions and macros which are different for UP/SMP systems are kept in a
143 * We require the migration_base for lock_hrtimer_base()/switch_hrtimer_base()
144 * such that hrtimer_callback_running() can unconditionally dereference
145 * timer->base->cpu_base
147 static struct hrtimer_cpu_base migration_cpu_base = {
148 .clock_base = { { .cpu_base = &migration_cpu_base, }, },
151 #define migration_base migration_cpu_base.clock_base[0]
154 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
155 * means that all timers which are tied to this base via timer->base are
156 * locked, and the base itself is locked too.
158 * So __run_timers/migrate_timers can safely modify all timers which could
159 * be found on the lists/queues.
161 * When the timer's base is locked, and the timer removed from list, it is
162 * possible to set timer->base = &migration_base and drop the lock: the timer
166 struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
167 unsigned long *flags)
169 struct hrtimer_clock_base *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 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;
264 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 struct hrtimer_clock_base *
278 lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
280 struct hrtimer_clock_base *base = timer->base;
282 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
287 # define switch_hrtimer_base(t, b, p) (b)
289 #endif /* !CONFIG_SMP */
292 * Functions for the union type storage format of ktime_t which are
293 * too large for inlining:
295 #if BITS_PER_LONG < 64
297 * Divide a ktime value by a nanosecond value
299 s64 __ktime_divns(const ktime_t kt, s64 div)
305 dclc = ktime_to_ns(kt);
306 tmp = dclc < 0 ? -dclc : dclc;
308 /* Make sure the divisor is less than 2^32: */
314 do_div(tmp, (unsigned long) div);
315 return dclc < 0 ? -tmp : tmp;
317 EXPORT_SYMBOL_GPL(__ktime_divns);
318 #endif /* BITS_PER_LONG >= 64 */
321 * Add two ktime values and do a safety check for overflow:
323 ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
325 ktime_t res = ktime_add_unsafe(lhs, rhs);
328 * We use KTIME_SEC_MAX here, the maximum timeout which we can
329 * return to user space in a timespec:
331 if (res < 0 || res < lhs || res < rhs)
332 res = ktime_set(KTIME_SEC_MAX, 0);
337 EXPORT_SYMBOL_GPL(ktime_add_safe);
339 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
341 static struct debug_obj_descr hrtimer_debug_descr;
343 static void *hrtimer_debug_hint(void *addr)
345 return ((struct hrtimer *) addr)->function;
349 * fixup_init is called when:
350 * - an active object is initialized
352 static bool hrtimer_fixup_init(void *addr, enum debug_obj_state state)
354 struct hrtimer *timer = addr;
357 case ODEBUG_STATE_ACTIVE:
358 hrtimer_cancel(timer);
359 debug_object_init(timer, &hrtimer_debug_descr);
367 * fixup_activate is called when:
368 * - an active object is activated
369 * - an unknown non-static object is activated
371 static bool hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
374 case ODEBUG_STATE_ACTIVE:
383 * fixup_free is called when:
384 * - an active object is freed
386 static bool hrtimer_fixup_free(void *addr, enum debug_obj_state state)
388 struct hrtimer *timer = addr;
391 case ODEBUG_STATE_ACTIVE:
392 hrtimer_cancel(timer);
393 debug_object_free(timer, &hrtimer_debug_descr);
400 static struct debug_obj_descr hrtimer_debug_descr = {
402 .debug_hint = hrtimer_debug_hint,
403 .fixup_init = hrtimer_fixup_init,
404 .fixup_activate = hrtimer_fixup_activate,
405 .fixup_free = hrtimer_fixup_free,
408 static inline void debug_hrtimer_init(struct hrtimer *timer)
410 debug_object_init(timer, &hrtimer_debug_descr);
413 static inline void debug_hrtimer_activate(struct hrtimer *timer,
414 enum hrtimer_mode mode)
416 debug_object_activate(timer, &hrtimer_debug_descr);
419 static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
421 debug_object_deactivate(timer, &hrtimer_debug_descr);
424 static inline void debug_hrtimer_free(struct hrtimer *timer)
426 debug_object_free(timer, &hrtimer_debug_descr);
429 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
430 enum hrtimer_mode mode);
432 void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
433 enum hrtimer_mode mode)
435 debug_object_init_on_stack(timer, &hrtimer_debug_descr);
436 __hrtimer_init(timer, clock_id, mode);
438 EXPORT_SYMBOL_GPL(hrtimer_init_on_stack);
440 void destroy_hrtimer_on_stack(struct hrtimer *timer)
442 debug_object_free(timer, &hrtimer_debug_descr);
444 EXPORT_SYMBOL_GPL(destroy_hrtimer_on_stack);
448 static inline void debug_hrtimer_init(struct hrtimer *timer) { }
449 static inline void debug_hrtimer_activate(struct hrtimer *timer,
450 enum hrtimer_mode mode) { }
451 static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
455 debug_init(struct hrtimer *timer, clockid_t clockid,
456 enum hrtimer_mode mode)
458 debug_hrtimer_init(timer);
459 trace_hrtimer_init(timer, clockid, mode);
462 static inline void debug_activate(struct hrtimer *timer,
463 enum hrtimer_mode mode)
465 debug_hrtimer_activate(timer, mode);
466 trace_hrtimer_start(timer, mode);
469 static inline void debug_deactivate(struct hrtimer *timer)
471 debug_hrtimer_deactivate(timer);
472 trace_hrtimer_cancel(timer);
475 static struct hrtimer_clock_base *
476 __next_base(struct hrtimer_cpu_base *cpu_base, unsigned int *active)
483 idx = __ffs(*active);
484 *active &= ~(1U << idx);
486 return &cpu_base->clock_base[idx];
489 #define for_each_active_base(base, cpu_base, active) \
490 while ((base = __next_base((cpu_base), &(active))))
492 static ktime_t __hrtimer_next_event_base(struct hrtimer_cpu_base *cpu_base,
493 const struct hrtimer *exclude,
495 ktime_t expires_next)
497 struct hrtimer_clock_base *base;
500 for_each_active_base(base, cpu_base, active) {
501 struct timerqueue_node *next;
502 struct hrtimer *timer;
504 next = timerqueue_getnext(&base->active);
505 timer = container_of(next, struct hrtimer, node);
506 if (timer == exclude) {
507 /* Get to the next timer in the queue. */
508 next = timerqueue_iterate_next(next);
512 timer = container_of(next, struct hrtimer, node);
514 expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
515 if (expires < expires_next) {
516 expires_next = expires;
518 /* Skip cpu_base update if a timer is being excluded. */
523 cpu_base->softirq_next_timer = timer;
525 cpu_base->next_timer = timer;
529 * clock_was_set() might have changed base->offset of any of
530 * the clock bases so the result might be negative. Fix it up
531 * to prevent a false positive in clockevents_program_event().
533 if (expires_next < 0)
539 * Recomputes cpu_base::*next_timer and returns the earliest expires_next
540 * but does not set cpu_base::*expires_next, that is done by
541 * hrtimer[_force]_reprogram and hrtimer_interrupt only. When updating
542 * cpu_base::*expires_next right away, reprogramming logic would no longer
545 * When a softirq is pending, we can ignore the HRTIMER_ACTIVE_SOFT bases,
546 * those timers will get run whenever the softirq gets handled, at the end of
547 * hrtimer_run_softirq(), hrtimer_update_softirq_timer() will re-add these bases.
549 * Therefore softirq values are those from the HRTIMER_ACTIVE_SOFT clock bases.
550 * The !softirq values are the minima across HRTIMER_ACTIVE_ALL, unless an actual
551 * softirq is pending, in which case they're the minima of HRTIMER_ACTIVE_HARD.
553 * @active_mask must be one of:
554 * - HRTIMER_ACTIVE_ALL,
555 * - HRTIMER_ACTIVE_SOFT, or
556 * - HRTIMER_ACTIVE_HARD.
559 __hrtimer_get_next_event(struct hrtimer_cpu_base *cpu_base, unsigned int active_mask)
562 struct hrtimer *next_timer = NULL;
563 ktime_t expires_next = KTIME_MAX;
565 if (!cpu_base->softirq_activated && (active_mask & HRTIMER_ACTIVE_SOFT)) {
566 active = cpu_base->active_bases & HRTIMER_ACTIVE_SOFT;
567 cpu_base->softirq_next_timer = NULL;
568 expires_next = __hrtimer_next_event_base(cpu_base, NULL,
571 next_timer = cpu_base->softirq_next_timer;
574 if (active_mask & HRTIMER_ACTIVE_HARD) {
575 active = cpu_base->active_bases & HRTIMER_ACTIVE_HARD;
576 cpu_base->next_timer = next_timer;
577 expires_next = __hrtimer_next_event_base(cpu_base, NULL, active,
584 static ktime_t hrtimer_update_next_event(struct hrtimer_cpu_base *cpu_base)
586 ktime_t expires_next, soft = KTIME_MAX;
589 * If the soft interrupt has already been activated, ignore the
590 * soft bases. They will be handled in the already raised soft
593 if (!cpu_base->softirq_activated) {
594 soft = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_SOFT);
596 * Update the soft expiry time. clock_settime() might have
599 cpu_base->softirq_expires_next = soft;
602 expires_next = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_HARD);
604 * If a softirq timer is expiring first, update cpu_base->next_timer
605 * and program the hardware with the soft expiry time.
607 if (expires_next > soft) {
608 cpu_base->next_timer = cpu_base->softirq_next_timer;
615 static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base)
617 ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset;
618 ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset;
619 ktime_t *offs_tai = &base->clock_base[HRTIMER_BASE_TAI].offset;
621 ktime_t now = ktime_get_update_offsets_now(&base->clock_was_set_seq,
622 offs_real, offs_boot, offs_tai);
624 base->clock_base[HRTIMER_BASE_REALTIME_SOFT].offset = *offs_real;
625 base->clock_base[HRTIMER_BASE_BOOTTIME_SOFT].offset = *offs_boot;
626 base->clock_base[HRTIMER_BASE_TAI_SOFT].offset = *offs_tai;
632 * Is the high resolution mode active ?
634 static inline int __hrtimer_hres_active(struct hrtimer_cpu_base *cpu_base)
636 return IS_ENABLED(CONFIG_HIGH_RES_TIMERS) ?
637 cpu_base->hres_active : 0;
640 static inline int hrtimer_hres_active(void)
642 return __hrtimer_hres_active(this_cpu_ptr(&hrtimer_bases));
646 * Reprogram the event source with checking both queues for the
648 * Called with interrupts disabled and base->lock held
651 hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
653 ktime_t expires_next;
655 expires_next = hrtimer_update_next_event(cpu_base);
657 if (skip_equal && expires_next == cpu_base->expires_next)
660 cpu_base->expires_next = expires_next;
663 * If hres is not active, hardware does not have to be
666 * If a hang was detected in the last timer interrupt then we
667 * leave the hang delay active in the hardware. We want the
668 * system to make progress. That also prevents the following
670 * T1 expires 50ms from now
671 * T2 expires 5s from now
673 * T1 is removed, so this code is called and would reprogram
674 * the hardware to 5s from now. Any hrtimer_start after that
675 * will not reprogram the hardware due to hang_detected being
676 * set. So we'd effectivly block all timers until the T2 event
679 if (!__hrtimer_hres_active(cpu_base) || cpu_base->hang_detected)
682 tick_program_event(cpu_base->expires_next, 1);
685 /* High resolution timer related functions */
686 #ifdef CONFIG_HIGH_RES_TIMERS
689 * High resolution timer enabled ?
691 static bool hrtimer_hres_enabled __read_mostly = true;
692 unsigned int hrtimer_resolution __read_mostly = LOW_RES_NSEC;
693 EXPORT_SYMBOL_GPL(hrtimer_resolution);
696 * Enable / Disable high resolution mode
698 static int __init setup_hrtimer_hres(char *str)
700 return (kstrtobool(str, &hrtimer_hres_enabled) == 0);
703 __setup("highres=", setup_hrtimer_hres);
706 * hrtimer_high_res_enabled - query, if the highres mode is enabled
708 static inline int hrtimer_is_hres_enabled(void)
710 return hrtimer_hres_enabled;
714 * Retrigger next event is called after clock was set
716 * Called with interrupts disabled via on_each_cpu()
718 static void retrigger_next_event(void *arg)
720 struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
722 if (!__hrtimer_hres_active(base))
725 raw_spin_lock(&base->lock);
726 hrtimer_update_base(base);
727 hrtimer_force_reprogram(base, 0);
728 raw_spin_unlock(&base->lock);
732 * Switch to high resolution mode
734 static void hrtimer_switch_to_hres(void)
736 struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
738 if (tick_init_highres()) {
739 pr_warn("Could not switch to high resolution mode on CPU %u\n",
743 base->hres_active = 1;
744 hrtimer_resolution = HIGH_RES_NSEC;
746 tick_setup_sched_timer();
747 /* "Retrigger" the interrupt to get things going */
748 retrigger_next_event(NULL);
751 static void clock_was_set_work(struct work_struct *work)
756 static DECLARE_WORK(hrtimer_work, clock_was_set_work);
759 * Called from timekeeping and resume code to reprogram the hrtimer
760 * interrupt device on all cpus.
762 void clock_was_set_delayed(void)
764 schedule_work(&hrtimer_work);
769 static inline int hrtimer_is_hres_enabled(void) { return 0; }
770 static inline void hrtimer_switch_to_hres(void) { }
771 static inline void retrigger_next_event(void *arg) { }
773 #endif /* CONFIG_HIGH_RES_TIMERS */
776 * When a timer is enqueued and expires earlier than the already enqueued
777 * timers, we have to check, whether it expires earlier than the timer for
778 * which the clock event device was armed.
780 * Called with interrupts disabled and base->cpu_base.lock held
782 static void hrtimer_reprogram(struct hrtimer *timer, bool reprogram)
784 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
785 struct hrtimer_clock_base *base = timer->base;
786 ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
788 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
791 * CLOCK_REALTIME timer might be requested with an absolute
792 * expiry time which is less than base->offset. Set it to 0.
797 if (timer->is_soft) {
799 * soft hrtimer could be started on a remote CPU. In this
800 * case softirq_expires_next needs to be updated on the
801 * remote CPU. The soft hrtimer will not expire before the
802 * first hard hrtimer on the remote CPU -
803 * hrtimer_check_target() prevents this case.
805 struct hrtimer_cpu_base *timer_cpu_base = base->cpu_base;
807 if (timer_cpu_base->softirq_activated)
810 if (!ktime_before(expires, timer_cpu_base->softirq_expires_next))
813 timer_cpu_base->softirq_next_timer = timer;
814 timer_cpu_base->softirq_expires_next = expires;
816 if (!ktime_before(expires, timer_cpu_base->expires_next) ||
822 * If the timer is not on the current cpu, we cannot reprogram
823 * the other cpus clock event device.
825 if (base->cpu_base != cpu_base)
829 * If the hrtimer interrupt is running, then it will
830 * reevaluate the clock bases and reprogram the clock event
831 * device. The callbacks are always executed in hard interrupt
832 * context so we don't need an extra check for a running
835 if (cpu_base->in_hrtirq)
838 if (expires >= cpu_base->expires_next)
841 /* Update the pointer to the next expiring timer */
842 cpu_base->next_timer = timer;
843 cpu_base->expires_next = expires;
846 * If hres is not active, hardware does not have to be
849 * If a hang was detected in the last timer interrupt then we
850 * do not schedule a timer which is earlier than the expiry
851 * which we enforced in the hang detection. We want the system
854 if (!__hrtimer_hres_active(cpu_base) || cpu_base->hang_detected)
858 * Program the timer hardware. We enforce the expiry for
859 * events which are already in the past.
861 tick_program_event(expires, 1);
865 * Clock realtime was set
867 * Change the offset of the realtime clock vs. the monotonic
870 * We might have to reprogram the high resolution timer interrupt. On
871 * SMP we call the architecture specific code to retrigger _all_ high
872 * resolution timer interrupts. On UP we just disable interrupts and
873 * call the high resolution interrupt code.
875 void clock_was_set(void)
877 #ifdef CONFIG_HIGH_RES_TIMERS
878 /* Retrigger the CPU local events everywhere */
879 on_each_cpu(retrigger_next_event, NULL, 1);
881 timerfd_clock_was_set();
885 * During resume we might have to reprogram the high resolution timer
886 * interrupt on all online CPUs. However, all other CPUs will be
887 * stopped with IRQs interrupts disabled so the clock_was_set() call
890 void hrtimers_resume(void)
892 lockdep_assert_irqs_disabled();
893 /* Retrigger on the local CPU */
894 retrigger_next_event(NULL);
895 /* And schedule a retrigger for all others */
896 clock_was_set_delayed();
900 * Counterpart to lock_hrtimer_base above:
903 void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
905 raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
909 * hrtimer_forward - forward the timer expiry
910 * @timer: hrtimer to forward
911 * @now: forward past this time
912 * @interval: the interval to forward
914 * Forward the timer expiry so it will expire in the future.
915 * Returns the number of overruns.
917 * Can be safely called from the callback function of @timer. If
918 * called from other contexts @timer must neither be enqueued nor
919 * running the callback and the caller needs to take care of
922 * Note: This only updates the timer expiry value and does not requeue
925 u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
930 delta = ktime_sub(now, hrtimer_get_expires(timer));
935 if (WARN_ON(timer->state & HRTIMER_STATE_ENQUEUED))
938 if (interval < hrtimer_resolution)
939 interval = hrtimer_resolution;
941 if (unlikely(delta >= interval)) {
942 s64 incr = ktime_to_ns(interval);
944 orun = ktime_divns(delta, incr);
945 hrtimer_add_expires_ns(timer, incr * orun);
946 if (hrtimer_get_expires_tv64(timer) > now)
949 * This (and the ktime_add() below) is the
950 * correction for exact:
954 hrtimer_add_expires(timer, interval);
958 EXPORT_SYMBOL_GPL(hrtimer_forward);
961 * enqueue_hrtimer - internal function to (re)start a timer
963 * The timer is inserted in expiry order. Insertion into the
964 * red black tree is O(log(n)). Must hold the base lock.
966 * Returns 1 when the new timer is the leftmost timer in the tree.
968 static int enqueue_hrtimer(struct hrtimer *timer,
969 struct hrtimer_clock_base *base,
970 enum hrtimer_mode mode)
972 debug_activate(timer, mode);
974 base->cpu_base->active_bases |= 1 << base->index;
976 /* Pairs with the lockless read in hrtimer_is_queued() */
977 WRITE_ONCE(timer->state, HRTIMER_STATE_ENQUEUED);
979 return timerqueue_add(&base->active, &timer->node);
983 * __remove_hrtimer - internal function to remove a timer
985 * Caller must hold the base lock.
987 * High resolution timer mode reprograms the clock event device when the
988 * timer is the one which expires next. The caller can disable this by setting
989 * reprogram to zero. This is useful, when the context does a reprogramming
990 * anyway (e.g. timer interrupt)
992 static void __remove_hrtimer(struct hrtimer *timer,
993 struct hrtimer_clock_base *base,
994 u8 newstate, int reprogram)
996 struct hrtimer_cpu_base *cpu_base = base->cpu_base;
997 u8 state = timer->state;
999 /* Pairs with the lockless read in hrtimer_is_queued() */
1000 WRITE_ONCE(timer->state, newstate);
1001 if (!(state & HRTIMER_STATE_ENQUEUED))
1004 if (!timerqueue_del(&base->active, &timer->node))
1005 cpu_base->active_bases &= ~(1 << base->index);
1008 * Note: If reprogram is false we do not update
1009 * cpu_base->next_timer. This happens when we remove the first
1010 * timer on a remote cpu. No harm as we never dereference
1011 * cpu_base->next_timer. So the worst thing what can happen is
1012 * an superflous call to hrtimer_force_reprogram() on the
1013 * remote cpu later on if the same timer gets enqueued again.
1015 if (reprogram && timer == cpu_base->next_timer)
1016 hrtimer_force_reprogram(cpu_base, 1);
1020 * remove hrtimer, called with base lock held
1023 remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base,
1024 bool restart, bool keep_local)
1026 u8 state = timer->state;
1028 if (state & HRTIMER_STATE_ENQUEUED) {
1032 * Remove the timer and force reprogramming when high
1033 * resolution mode is active and the timer is on the current
1034 * CPU. If we remove a timer on another CPU, reprogramming is
1035 * skipped. The interrupt event on this CPU is fired and
1036 * reprogramming happens in the interrupt handler. This is a
1037 * rare case and less expensive than a smp call.
1039 debug_deactivate(timer);
1040 reprogram = base->cpu_base == this_cpu_ptr(&hrtimer_bases);
1043 * If the timer is not restarted then reprogramming is
1044 * required if the timer is local. If it is local and about
1045 * to be restarted, avoid programming it twice (on removal
1046 * and a moment later when it's requeued).
1049 state = HRTIMER_STATE_INACTIVE;
1051 reprogram &= !keep_local;
1053 __remove_hrtimer(timer, base, state, reprogram);
1059 static inline ktime_t hrtimer_update_lowres(struct hrtimer *timer, ktime_t tim,
1060 const enum hrtimer_mode mode)
1062 #ifdef CONFIG_TIME_LOW_RES
1064 * CONFIG_TIME_LOW_RES indicates that the system has no way to return
1065 * granular time values. For relative timers we add hrtimer_resolution
1066 * (i.e. one jiffie) to prevent short timeouts.
1068 timer->is_rel = mode & HRTIMER_MODE_REL;
1070 tim = ktime_add_safe(tim, hrtimer_resolution);
1076 hrtimer_update_softirq_timer(struct hrtimer_cpu_base *cpu_base, bool reprogram)
1081 * Find the next SOFT expiration.
1083 expires = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_SOFT);
1086 * reprogramming needs to be triggered, even if the next soft
1087 * hrtimer expires at the same time than the next hard
1088 * hrtimer. cpu_base->softirq_expires_next needs to be updated!
1090 if (expires == KTIME_MAX)
1094 * cpu_base->*next_timer is recomputed by __hrtimer_get_next_event()
1095 * cpu_base->*expires_next is only set by hrtimer_reprogram()
1097 hrtimer_reprogram(cpu_base->softirq_next_timer, reprogram);
1100 static int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1101 u64 delta_ns, const enum hrtimer_mode mode,
1102 struct hrtimer_clock_base *base)
1104 struct hrtimer_clock_base *new_base;
1105 bool force_local, first;
1108 * If the timer is on the local cpu base and is the first expiring
1109 * timer then this might end up reprogramming the hardware twice
1110 * (on removal and on enqueue). To avoid that by prevent the
1111 * reprogram on removal, keep the timer local to the current CPU
1112 * and enforce reprogramming after it is queued no matter whether
1113 * it is the new first expiring timer again or not.
1115 force_local = base->cpu_base == this_cpu_ptr(&hrtimer_bases);
1116 force_local &= base->cpu_base->next_timer == timer;
1119 * Remove an active timer from the queue. In case it is not queued
1120 * on the current CPU, make sure that remove_hrtimer() updates the
1121 * remote data correctly.
1123 * If it's on the current CPU and the first expiring timer, then
1124 * skip reprogramming, keep the timer local and enforce
1125 * reprogramming later if it was the first expiring timer. This
1126 * avoids programming the underlying clock event twice (once at
1127 * removal and once after enqueue).
1129 remove_hrtimer(timer, base, true, force_local);
1131 if (mode & HRTIMER_MODE_REL)
1132 tim = ktime_add_safe(tim, base->get_time());
1134 tim = hrtimer_update_lowres(timer, tim, mode);
1136 hrtimer_set_expires_range_ns(timer, tim, delta_ns);
1138 /* Switch the timer base, if necessary: */
1140 new_base = switch_hrtimer_base(timer, base,
1141 mode & HRTIMER_MODE_PINNED);
1146 first = enqueue_hrtimer(timer, new_base, mode);
1151 * Timer was forced to stay on the current CPU to avoid
1152 * reprogramming on removal and enqueue. Force reprogram the
1153 * hardware by evaluating the new first expiring timer.
1155 hrtimer_force_reprogram(new_base->cpu_base, 1);
1160 * hrtimer_start_range_ns - (re)start an hrtimer
1161 * @timer: the timer to be added
1163 * @delta_ns: "slack" range for the timer
1164 * @mode: timer mode: absolute (HRTIMER_MODE_ABS) or
1165 * relative (HRTIMER_MODE_REL), and pinned (HRTIMER_MODE_PINNED);
1166 * softirq based mode is considered for debug purpose only!
1168 void hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1169 u64 delta_ns, const enum hrtimer_mode mode)
1171 struct hrtimer_clock_base *base;
1172 unsigned long flags;
1175 * Check whether the HRTIMER_MODE_SOFT bit and hrtimer.is_soft
1178 WARN_ON_ONCE(!(mode & HRTIMER_MODE_SOFT) ^ !timer->is_soft);
1180 base = lock_hrtimer_base(timer, &flags);
1182 if (__hrtimer_start_range_ns(timer, tim, delta_ns, mode, base))
1183 hrtimer_reprogram(timer, true);
1185 unlock_hrtimer_base(timer, &flags);
1187 EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
1190 * hrtimer_try_to_cancel - try to deactivate a timer
1191 * @timer: hrtimer to stop
1194 * 0 when the timer was not active
1195 * 1 when the timer was active
1196 * -1 when the timer is currently executing the callback function and
1199 int hrtimer_try_to_cancel(struct hrtimer *timer)
1201 struct hrtimer_clock_base *base;
1202 unsigned long flags;
1206 * Check lockless first. If the timer is not active (neither
1207 * enqueued nor running the callback, nothing to do here. The
1208 * base lock does not serialize against a concurrent enqueue,
1209 * so we can avoid taking it.
1211 if (!hrtimer_active(timer))
1214 base = lock_hrtimer_base(timer, &flags);
1216 if (!hrtimer_callback_running(timer))
1217 ret = remove_hrtimer(timer, base, false, false);
1219 unlock_hrtimer_base(timer, &flags);
1224 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1227 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1228 * @timer: the timer to be cancelled
1231 * 0 when the timer was not active
1232 * 1 when the timer was active
1234 int hrtimer_cancel(struct hrtimer *timer)
1237 int ret = hrtimer_try_to_cancel(timer);
1244 EXPORT_SYMBOL_GPL(hrtimer_cancel);
1247 * hrtimer_get_remaining - get remaining time for the timer
1248 * @timer: the timer to read
1249 * @adjust: adjust relative timers when CONFIG_TIME_LOW_RES=y
1251 ktime_t __hrtimer_get_remaining(const struct hrtimer *timer, bool adjust)
1253 unsigned long flags;
1256 lock_hrtimer_base(timer, &flags);
1257 if (IS_ENABLED(CONFIG_TIME_LOW_RES) && adjust)
1258 rem = hrtimer_expires_remaining_adjusted(timer);
1260 rem = hrtimer_expires_remaining(timer);
1261 unlock_hrtimer_base(timer, &flags);
1265 EXPORT_SYMBOL_GPL(__hrtimer_get_remaining);
1267 #ifdef CONFIG_NO_HZ_COMMON
1269 * hrtimer_get_next_event - get the time until next expiry event
1271 * Returns the next expiry time or KTIME_MAX if no timer is pending.
1273 u64 hrtimer_get_next_event(void)
1275 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1276 u64 expires = KTIME_MAX;
1277 unsigned long flags;
1279 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1281 if (!__hrtimer_hres_active(cpu_base))
1282 expires = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_ALL);
1284 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1290 * hrtimer_next_event_without - time until next expiry event w/o one timer
1291 * @exclude: timer to exclude
1293 * Returns the next expiry time over all timers except for the @exclude one or
1294 * KTIME_MAX if none of them is pending.
1296 u64 hrtimer_next_event_without(const struct hrtimer *exclude)
1298 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1299 u64 expires = KTIME_MAX;
1300 unsigned long flags;
1302 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1304 if (__hrtimer_hres_active(cpu_base)) {
1305 unsigned int active;
1307 if (!cpu_base->softirq_activated) {
1308 active = cpu_base->active_bases & HRTIMER_ACTIVE_SOFT;
1309 expires = __hrtimer_next_event_base(cpu_base, exclude,
1312 active = cpu_base->active_bases & HRTIMER_ACTIVE_HARD;
1313 expires = __hrtimer_next_event_base(cpu_base, exclude, active,
1317 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1323 static inline int hrtimer_clockid_to_base(clockid_t clock_id)
1325 if (likely(clock_id < MAX_CLOCKS)) {
1326 int base = hrtimer_clock_to_base_table[clock_id];
1328 if (likely(base != HRTIMER_MAX_CLOCK_BASES))
1331 WARN(1, "Invalid clockid %d. Using MONOTONIC\n", clock_id);
1332 return HRTIMER_BASE_MONOTONIC;
1335 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1336 enum hrtimer_mode mode)
1338 bool softtimer = !!(mode & HRTIMER_MODE_SOFT);
1339 int base = softtimer ? HRTIMER_MAX_CLOCK_BASES / 2 : 0;
1340 struct hrtimer_cpu_base *cpu_base;
1342 memset(timer, 0, sizeof(struct hrtimer));
1344 cpu_base = raw_cpu_ptr(&hrtimer_bases);
1347 * POSIX magic: Relative CLOCK_REALTIME timers are not affected by
1348 * clock modifications, so they needs to become CLOCK_MONOTONIC to
1349 * ensure POSIX compliance.
1351 if (clock_id == CLOCK_REALTIME && mode & HRTIMER_MODE_REL)
1352 clock_id = CLOCK_MONOTONIC;
1354 base += hrtimer_clockid_to_base(clock_id);
1355 timer->is_soft = softtimer;
1356 timer->base = &cpu_base->clock_base[base];
1357 timerqueue_init(&timer->node);
1361 * hrtimer_init - initialize a timer to the given clock
1362 * @timer: the timer to be initialized
1363 * @clock_id: the clock to be used
1364 * @mode: The modes which are relevant for intitialization:
1365 * HRTIMER_MODE_ABS, HRTIMER_MODE_REL, HRTIMER_MODE_ABS_SOFT,
1366 * HRTIMER_MODE_REL_SOFT
1368 * The PINNED variants of the above can be handed in,
1369 * but the PINNED bit is ignored as pinning happens
1370 * when the hrtimer is started
1372 void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1373 enum hrtimer_mode mode)
1375 debug_init(timer, clock_id, mode);
1376 __hrtimer_init(timer, clock_id, mode);
1378 EXPORT_SYMBOL_GPL(hrtimer_init);
1381 * A timer is active, when it is enqueued into the rbtree or the
1382 * callback function is running or it's in the state of being migrated
1385 * It is important for this function to not return a false negative.
1387 bool hrtimer_active(const struct hrtimer *timer)
1389 struct hrtimer_clock_base *base;
1393 base = READ_ONCE(timer->base);
1394 seq = raw_read_seqcount_begin(&base->seq);
1396 if (timer->state != HRTIMER_STATE_INACTIVE ||
1397 base->running == timer)
1400 } while (read_seqcount_retry(&base->seq, seq) ||
1401 base != READ_ONCE(timer->base));
1405 EXPORT_SYMBOL_GPL(hrtimer_active);
1408 * The write_seqcount_barrier()s in __run_hrtimer() split the thing into 3
1409 * distinct sections:
1411 * - queued: the timer is queued
1412 * - callback: the timer is being ran
1413 * - post: the timer is inactive or (re)queued
1415 * On the read side we ensure we observe timer->state and cpu_base->running
1416 * from the same section, if anything changed while we looked at it, we retry.
1417 * This includes timer->base changing because sequence numbers alone are
1418 * insufficient for that.
1420 * The sequence numbers are required because otherwise we could still observe
1421 * a false negative if the read side got smeared over multiple consequtive
1422 * __run_hrtimer() invocations.
1425 static void __run_hrtimer(struct hrtimer_cpu_base *cpu_base,
1426 struct hrtimer_clock_base *base,
1427 struct hrtimer *timer, ktime_t *now,
1428 unsigned long flags)
1430 enum hrtimer_restart (*fn)(struct hrtimer *);
1433 lockdep_assert_held(&cpu_base->lock);
1435 debug_deactivate(timer);
1436 base->running = timer;
1439 * Separate the ->running assignment from the ->state assignment.
1441 * As with a regular write barrier, this ensures the read side in
1442 * hrtimer_active() cannot observe base->running == NULL &&
1443 * timer->state == INACTIVE.
1445 raw_write_seqcount_barrier(&base->seq);
1447 __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE, 0);
1448 fn = timer->function;
1451 * Clear the 'is relative' flag for the TIME_LOW_RES case. If the
1452 * timer is restarted with a period then it becomes an absolute
1453 * timer. If its not restarted it does not matter.
1455 if (IS_ENABLED(CONFIG_TIME_LOW_RES))
1456 timer->is_rel = false;
1459 * The timer is marked as running in the CPU base, so it is
1460 * protected against migration to a different CPU even if the lock
1463 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1464 trace_hrtimer_expire_entry(timer, now);
1465 restart = fn(timer);
1466 trace_hrtimer_expire_exit(timer);
1467 raw_spin_lock_irq(&cpu_base->lock);
1470 * Note: We clear the running state after enqueue_hrtimer and
1471 * we do not reprogram the event hardware. Happens either in
1472 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1474 * Note: Because we dropped the cpu_base->lock above,
1475 * hrtimer_start_range_ns() can have popped in and enqueued the timer
1478 if (restart != HRTIMER_NORESTART &&
1479 !(timer->state & HRTIMER_STATE_ENQUEUED))
1480 enqueue_hrtimer(timer, base, HRTIMER_MODE_ABS);
1483 * Separate the ->running assignment from the ->state assignment.
1485 * As with a regular write barrier, this ensures the read side in
1486 * hrtimer_active() cannot observe base->running.timer == NULL &&
1487 * timer->state == INACTIVE.
1489 raw_write_seqcount_barrier(&base->seq);
1491 WARN_ON_ONCE(base->running != timer);
1492 base->running = NULL;
1495 static void __hrtimer_run_queues(struct hrtimer_cpu_base *cpu_base, ktime_t now,
1496 unsigned long flags, unsigned int active_mask)
1498 struct hrtimer_clock_base *base;
1499 unsigned int active = cpu_base->active_bases & active_mask;
1501 for_each_active_base(base, cpu_base, active) {
1502 struct timerqueue_node *node;
1505 basenow = ktime_add(now, base->offset);
1507 while ((node = timerqueue_getnext(&base->active))) {
1508 struct hrtimer *timer;
1510 timer = container_of(node, struct hrtimer, node);
1513 * The immediate goal for using the softexpires is
1514 * minimizing wakeups, not running timers at the
1515 * earliest interrupt after their soft expiration.
1516 * This allows us to avoid using a Priority Search
1517 * Tree, which can answer a stabbing querry for
1518 * overlapping intervals and instead use the simple
1519 * BST we already have.
1520 * We don't add extra wakeups by delaying timers that
1521 * are right-of a not yet expired timer, because that
1522 * timer will have to trigger a wakeup anyway.
1524 if (basenow < hrtimer_get_softexpires_tv64(timer))
1527 __run_hrtimer(cpu_base, base, timer, &basenow, flags);
1532 static __latent_entropy void hrtimer_run_softirq(struct softirq_action *h)
1534 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1535 unsigned long flags;
1538 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1540 now = hrtimer_update_base(cpu_base);
1541 __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_SOFT);
1543 cpu_base->softirq_activated = 0;
1544 hrtimer_update_softirq_timer(cpu_base, true);
1546 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1549 #ifdef CONFIG_HIGH_RES_TIMERS
1552 * High resolution timer interrupt
1553 * Called with interrupts disabled
1555 void hrtimer_interrupt(struct clock_event_device *dev)
1557 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1558 ktime_t expires_next, now, entry_time, delta;
1559 unsigned long flags;
1562 BUG_ON(!cpu_base->hres_active);
1563 cpu_base->nr_events++;
1564 dev->next_event = KTIME_MAX;
1566 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1567 entry_time = now = hrtimer_update_base(cpu_base);
1569 cpu_base->in_hrtirq = 1;
1571 * We set expires_next to KTIME_MAX here with cpu_base->lock
1572 * held to prevent that a timer is enqueued in our queue via
1573 * the migration code. This does not affect enqueueing of
1574 * timers which run their callback and need to be requeued on
1577 cpu_base->expires_next = KTIME_MAX;
1579 if (!ktime_before(now, cpu_base->softirq_expires_next)) {
1580 cpu_base->softirq_expires_next = KTIME_MAX;
1581 cpu_base->softirq_activated = 1;
1582 raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1585 __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_HARD);
1587 /* Reevaluate the clock bases for the [soft] next expiry */
1588 expires_next = hrtimer_update_next_event(cpu_base);
1590 * Store the new expiry value so the migration code can verify
1593 cpu_base->expires_next = expires_next;
1594 cpu_base->in_hrtirq = 0;
1595 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1597 /* Reprogramming necessary ? */
1598 if (!tick_program_event(expires_next, 0)) {
1599 cpu_base->hang_detected = 0;
1604 * The next timer was already expired due to:
1606 * - long lasting callbacks
1607 * - being scheduled away when running in a VM
1609 * We need to prevent that we loop forever in the hrtimer
1610 * interrupt routine. We give it 3 attempts to avoid
1611 * overreacting on some spurious event.
1613 * Acquire base lock for updating the offsets and retrieving
1616 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1617 now = hrtimer_update_base(cpu_base);
1618 cpu_base->nr_retries++;
1622 * Give the system a chance to do something else than looping
1623 * here. We stored the entry time, so we know exactly how long
1624 * we spent here. We schedule the next event this amount of
1627 cpu_base->nr_hangs++;
1628 cpu_base->hang_detected = 1;
1629 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1631 delta = ktime_sub(now, entry_time);
1632 if ((unsigned int)delta > cpu_base->max_hang_time)
1633 cpu_base->max_hang_time = (unsigned int) delta;
1635 * Limit it to a sensible value as we enforce a longer
1636 * delay. Give the CPU at least 100ms to catch up.
1638 if (delta > 100 * NSEC_PER_MSEC)
1639 expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1641 expires_next = ktime_add(now, delta);
1642 tick_program_event(expires_next, 1);
1643 pr_warn_once("hrtimer: interrupt took %llu ns\n", ktime_to_ns(delta));
1646 /* called with interrupts disabled */
1647 static inline void __hrtimer_peek_ahead_timers(void)
1649 struct tick_device *td;
1651 if (!hrtimer_hres_active())
1654 td = this_cpu_ptr(&tick_cpu_device);
1655 if (td && td->evtdev)
1656 hrtimer_interrupt(td->evtdev);
1659 #else /* CONFIG_HIGH_RES_TIMERS */
1661 static inline void __hrtimer_peek_ahead_timers(void) { }
1663 #endif /* !CONFIG_HIGH_RES_TIMERS */
1666 * Called from run_local_timers in hardirq context every jiffy
1668 void hrtimer_run_queues(void)
1670 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1671 unsigned long flags;
1674 if (__hrtimer_hres_active(cpu_base))
1678 * This _is_ ugly: We have to check periodically, whether we
1679 * can switch to highres and / or nohz mode. The clocksource
1680 * switch happens with xtime_lock held. Notification from
1681 * there only sets the check bit in the tick_oneshot code,
1682 * otherwise we might deadlock vs. xtime_lock.
1684 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) {
1685 hrtimer_switch_to_hres();
1689 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1690 now = hrtimer_update_base(cpu_base);
1692 if (!ktime_before(now, cpu_base->softirq_expires_next)) {
1693 cpu_base->softirq_expires_next = KTIME_MAX;
1694 cpu_base->softirq_activated = 1;
1695 raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1698 __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_HARD);
1699 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1703 * Sleep related functions:
1705 static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1707 struct hrtimer_sleeper *t =
1708 container_of(timer, struct hrtimer_sleeper, timer);
1709 struct task_struct *task = t->task;
1713 wake_up_process(task);
1715 return HRTIMER_NORESTART;
1718 void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1720 sl->timer.function = hrtimer_wakeup;
1723 EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
1725 int nanosleep_copyout(struct restart_block *restart, struct timespec64 *ts)
1727 switch(restart->nanosleep.type) {
1728 #ifdef CONFIG_COMPAT_32BIT_TIME
1730 if (compat_put_timespec64(ts, restart->nanosleep.compat_rmtp))
1735 if (put_timespec64(ts, restart->nanosleep.rmtp))
1741 return -ERESTART_RESTARTBLOCK;
1744 static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1746 struct restart_block *restart;
1748 hrtimer_init_sleeper(t, current);
1751 set_current_state(TASK_INTERRUPTIBLE);
1752 hrtimer_start_expires(&t->timer, mode);
1754 if (likely(t->task))
1755 freezable_schedule();
1757 hrtimer_cancel(&t->timer);
1758 mode = HRTIMER_MODE_ABS;
1760 } while (t->task && !signal_pending(current));
1762 __set_current_state(TASK_RUNNING);
1767 restart = ¤t->restart_block;
1768 if (restart->nanosleep.type != TT_NONE) {
1769 ktime_t rem = hrtimer_expires_remaining(&t->timer);
1770 struct timespec64 rmt;
1774 rmt = ktime_to_timespec64(rem);
1776 return nanosleep_copyout(restart, &rmt);
1778 return -ERESTART_RESTARTBLOCK;
1781 static long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1783 struct hrtimer_sleeper t;
1786 hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid,
1788 hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
1790 ret = do_nanosleep(&t, HRTIMER_MODE_ABS);
1791 destroy_hrtimer_on_stack(&t.timer);
1795 long hrtimer_nanosleep(const struct timespec64 *rqtp,
1796 const enum hrtimer_mode mode, const clockid_t clockid)
1798 struct restart_block *restart;
1799 struct hrtimer_sleeper t;
1803 slack = current->timer_slack_ns;
1804 if (dl_task(current) || rt_task(current))
1807 hrtimer_init_on_stack(&t.timer, clockid, mode);
1808 hrtimer_set_expires_range_ns(&t.timer, timespec64_to_ktime(*rqtp), slack);
1809 ret = do_nanosleep(&t, mode);
1810 if (ret != -ERESTART_RESTARTBLOCK)
1813 /* Absolute timers do not update the rmtp value and restart: */
1814 if (mode == HRTIMER_MODE_ABS) {
1815 ret = -ERESTARTNOHAND;
1819 restart = ¤t->restart_block;
1820 restart->nanosleep.clockid = t.timer.base->clockid;
1821 restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
1822 set_restart_fn(restart, hrtimer_nanosleep_restart);
1824 destroy_hrtimer_on_stack(&t.timer);
1828 #if !defined(CONFIG_64BIT_TIME) || defined(CONFIG_64BIT)
1830 SYSCALL_DEFINE2(nanosleep, struct __kernel_timespec __user *, rqtp,
1831 struct __kernel_timespec __user *, rmtp)
1833 struct timespec64 tu;
1835 if (get_timespec64(&tu, rqtp))
1838 if (!timespec64_valid(&tu))
1841 current->restart_block.nanosleep.type = rmtp ? TT_NATIVE : TT_NONE;
1842 current->restart_block.nanosleep.rmtp = rmtp;
1843 return hrtimer_nanosleep(&tu, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1848 #ifdef CONFIG_COMPAT_32BIT_TIME
1850 COMPAT_SYSCALL_DEFINE2(nanosleep, struct compat_timespec __user *, rqtp,
1851 struct compat_timespec __user *, rmtp)
1853 struct timespec64 tu;
1855 if (compat_get_timespec64(&tu, rqtp))
1858 if (!timespec64_valid(&tu))
1861 current->restart_block.nanosleep.type = rmtp ? TT_COMPAT : TT_NONE;
1862 current->restart_block.nanosleep.compat_rmtp = rmtp;
1863 return hrtimer_nanosleep(&tu, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1868 * Functions related to boot-time initialization:
1870 int hrtimers_prepare_cpu(unsigned int cpu)
1872 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1875 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1876 cpu_base->clock_base[i].cpu_base = cpu_base;
1877 timerqueue_init_head(&cpu_base->clock_base[i].active);
1880 cpu_base->cpu = cpu;
1881 cpu_base->active_bases = 0;
1882 cpu_base->hres_active = 0;
1883 cpu_base->hang_detected = 0;
1884 cpu_base->next_timer = NULL;
1885 cpu_base->softirq_next_timer = NULL;
1886 cpu_base->expires_next = KTIME_MAX;
1887 cpu_base->softirq_expires_next = KTIME_MAX;
1891 #ifdef CONFIG_HOTPLUG_CPU
1893 static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1894 struct hrtimer_clock_base *new_base)
1896 struct hrtimer *timer;
1897 struct timerqueue_node *node;
1899 while ((node = timerqueue_getnext(&old_base->active))) {
1900 timer = container_of(node, struct hrtimer, node);
1901 BUG_ON(hrtimer_callback_running(timer));
1902 debug_deactivate(timer);
1905 * Mark it as ENQUEUED not INACTIVE otherwise the
1906 * timer could be seen as !active and just vanish away
1907 * under us on another CPU
1909 __remove_hrtimer(timer, old_base, HRTIMER_STATE_ENQUEUED, 0);
1910 timer->base = new_base;
1912 * Enqueue the timers on the new cpu. This does not
1913 * reprogram the event device in case the timer
1914 * expires before the earliest on this CPU, but we run
1915 * hrtimer_interrupt after we migrated everything to
1916 * sort out already expired timers and reprogram the
1919 enqueue_hrtimer(timer, new_base, HRTIMER_MODE_ABS);
1923 int hrtimers_dead_cpu(unsigned int scpu)
1925 struct hrtimer_cpu_base *old_base, *new_base;
1928 BUG_ON(cpu_online(scpu));
1929 tick_cancel_sched_timer(scpu);
1932 * this BH disable ensures that raise_softirq_irqoff() does
1933 * not wakeup ksoftirqd (and acquire the pi-lock) while
1934 * holding the cpu_base lock
1937 local_irq_disable();
1938 old_base = &per_cpu(hrtimer_bases, scpu);
1939 new_base = this_cpu_ptr(&hrtimer_bases);
1941 * The caller is globally serialized and nobody else
1942 * takes two locks at once, deadlock is not possible.
1944 raw_spin_lock(&new_base->lock);
1945 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1947 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1948 migrate_hrtimer_list(&old_base->clock_base[i],
1949 &new_base->clock_base[i]);
1953 * The migration might have changed the first expiring softirq
1954 * timer on this CPU. Update it.
1956 hrtimer_update_softirq_timer(new_base, false);
1958 raw_spin_unlock(&old_base->lock);
1959 raw_spin_unlock(&new_base->lock);
1961 /* Check, if we got expired work to do */
1962 __hrtimer_peek_ahead_timers();
1968 #endif /* CONFIG_HOTPLUG_CPU */
1970 void __init hrtimers_init(void)
1972 hrtimers_prepare_cpu(smp_processor_id());
1973 open_softirq(HRTIMER_SOFTIRQ, hrtimer_run_softirq);
1977 * schedule_hrtimeout_range_clock - sleep until timeout
1978 * @expires: timeout value (ktime_t)
1979 * @delta: slack in expires timeout (ktime_t)
1981 * @clock_id: timer clock to be used
1984 schedule_hrtimeout_range_clock(ktime_t *expires, u64 delta,
1985 const enum hrtimer_mode mode, clockid_t clock_id)
1987 struct hrtimer_sleeper t;
1990 * Optimize when a zero timeout value is given. It does not
1991 * matter whether this is an absolute or a relative time.
1993 if (expires && *expires == 0) {
1994 __set_current_state(TASK_RUNNING);
1999 * A NULL parameter means "infinite"
2006 hrtimer_init_on_stack(&t.timer, clock_id, mode);
2007 hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
2009 hrtimer_init_sleeper(&t, current);
2011 hrtimer_start_expires(&t.timer, mode);
2016 hrtimer_cancel(&t.timer);
2017 destroy_hrtimer_on_stack(&t.timer);
2019 __set_current_state(TASK_RUNNING);
2021 return !t.task ? 0 : -EINTR;
2025 * schedule_hrtimeout_range - sleep until timeout
2026 * @expires: timeout value (ktime_t)
2027 * @delta: slack in expires timeout (ktime_t)
2030 * Make the current task sleep until the given expiry time has
2031 * elapsed. The routine will return immediately unless
2032 * the current task state has been set (see set_current_state()).
2034 * The @delta argument gives the kernel the freedom to schedule the
2035 * actual wakeup to a time that is both power and performance friendly.
2036 * The kernel give the normal best effort behavior for "@expires+@delta",
2037 * but may decide to fire the timer earlier, but no earlier than @expires.
2039 * You can set the task state as follows -
2041 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
2042 * pass before the routine returns unless the current task is explicitly
2043 * woken up, (e.g. by wake_up_process()).
2045 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
2046 * delivered to the current task or the current task is explicitly woken
2049 * The current task state is guaranteed to be TASK_RUNNING when this
2052 * Returns 0 when the timer has expired. If the task was woken before the
2053 * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or
2054 * by an explicit wakeup, it returns -EINTR.
2056 int __sched schedule_hrtimeout_range(ktime_t *expires, u64 delta,
2057 const enum hrtimer_mode mode)
2059 return schedule_hrtimeout_range_clock(expires, delta, mode,
2062 EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
2065 * schedule_hrtimeout - sleep until timeout
2066 * @expires: timeout value (ktime_t)
2069 * Make the current task sleep until the given expiry time has
2070 * elapsed. The routine will return immediately unless
2071 * the current task state has been set (see set_current_state()).
2073 * You can set the task state as follows -
2075 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
2076 * pass before the routine returns unless the current task is explicitly
2077 * woken up, (e.g. by wake_up_process()).
2079 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
2080 * delivered to the current task or the current task is explicitly woken
2083 * The current task state is guaranteed to be TASK_RUNNING when this
2086 * Returns 0 when the timer has expired. If the task was woken before the
2087 * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or
2088 * by an explicit wakeup, it returns -EINTR.
2090 int __sched schedule_hrtimeout(ktime_t *expires,
2091 const enum hrtimer_mode mode)
2093 return schedule_hrtimeout_range(expires, 0, mode);
2095 EXPORT_SYMBOL_GPL(schedule_hrtimeout);