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,
494 ktime_t expires_next)
496 struct hrtimer_clock_base *base;
499 for_each_active_base(base, cpu_base, active) {
500 struct timerqueue_node *next;
501 struct hrtimer *timer;
503 next = timerqueue_getnext(&base->active);
504 timer = container_of(next, struct hrtimer, node);
505 expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
506 if (expires < expires_next) {
507 expires_next = expires;
509 cpu_base->softirq_next_timer = timer;
511 cpu_base->next_timer = timer;
515 * clock_was_set() might have changed base->offset of any of
516 * the clock bases so the result might be negative. Fix it up
517 * to prevent a false positive in clockevents_program_event().
519 if (expires_next < 0)
525 * Recomputes cpu_base::*next_timer and returns the earliest expires_next but
526 * does not set cpu_base::*expires_next, that is done by hrtimer_reprogram.
528 * When a softirq is pending, we can ignore the HRTIMER_ACTIVE_SOFT bases,
529 * those timers will get run whenever the softirq gets handled, at the end of
530 * hrtimer_run_softirq(), hrtimer_update_softirq_timer() will re-add these bases.
532 * Therefore softirq values are those from the HRTIMER_ACTIVE_SOFT clock bases.
533 * The !softirq values are the minima across HRTIMER_ACTIVE_ALL, unless an actual
534 * softirq is pending, in which case they're the minima of HRTIMER_ACTIVE_HARD.
536 * @active_mask must be one of:
537 * - HRTIMER_ACTIVE_ALL,
538 * - HRTIMER_ACTIVE_SOFT, or
539 * - HRTIMER_ACTIVE_HARD.
542 __hrtimer_get_next_event(struct hrtimer_cpu_base *cpu_base, unsigned int active_mask)
545 struct hrtimer *next_timer = NULL;
546 ktime_t expires_next = KTIME_MAX;
548 if (!cpu_base->softirq_activated && (active_mask & HRTIMER_ACTIVE_SOFT)) {
549 active = cpu_base->active_bases & HRTIMER_ACTIVE_SOFT;
550 cpu_base->softirq_next_timer = NULL;
551 expires_next = __hrtimer_next_event_base(cpu_base, active, KTIME_MAX);
553 next_timer = cpu_base->softirq_next_timer;
556 if (active_mask & HRTIMER_ACTIVE_HARD) {
557 active = cpu_base->active_bases & HRTIMER_ACTIVE_HARD;
558 cpu_base->next_timer = next_timer;
559 expires_next = __hrtimer_next_event_base(cpu_base, active, expires_next);
565 static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base)
567 ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset;
568 ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset;
569 ktime_t *offs_tai = &base->clock_base[HRTIMER_BASE_TAI].offset;
571 ktime_t now = ktime_get_update_offsets_now(&base->clock_was_set_seq,
572 offs_real, offs_boot, offs_tai);
574 base->clock_base[HRTIMER_BASE_REALTIME_SOFT].offset = *offs_real;
575 base->clock_base[HRTIMER_BASE_BOOTTIME_SOFT].offset = *offs_boot;
576 base->clock_base[HRTIMER_BASE_TAI_SOFT].offset = *offs_tai;
582 * Is the high resolution mode active ?
584 static inline int __hrtimer_hres_active(struct hrtimer_cpu_base *cpu_base)
586 return IS_ENABLED(CONFIG_HIGH_RES_TIMERS) ?
587 cpu_base->hres_active : 0;
590 static inline int hrtimer_hres_active(void)
592 return __hrtimer_hres_active(this_cpu_ptr(&hrtimer_bases));
596 * Reprogram the event source with checking both queues for the
598 * Called with interrupts disabled and base->lock held
601 hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
603 ktime_t expires_next;
606 * Find the current next expiration time.
608 expires_next = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_ALL);
610 if (cpu_base->next_timer && cpu_base->next_timer->is_soft) {
612 * When the softirq is activated, hrtimer has to be
613 * programmed with the first hard hrtimer because soft
614 * timer interrupt could occur too late.
616 if (cpu_base->softirq_activated)
617 expires_next = __hrtimer_get_next_event(cpu_base,
618 HRTIMER_ACTIVE_HARD);
620 cpu_base->softirq_expires_next = expires_next;
623 if (skip_equal && expires_next == cpu_base->expires_next)
626 cpu_base->expires_next = expires_next;
629 * If hres is not active, hardware does not have to be
632 * If a hang was detected in the last timer interrupt then we
633 * leave the hang delay active in the hardware. We want the
634 * system to make progress. That also prevents the following
636 * T1 expires 50ms from now
637 * T2 expires 5s from now
639 * T1 is removed, so this code is called and would reprogram
640 * the hardware to 5s from now. Any hrtimer_start after that
641 * will not reprogram the hardware due to hang_detected being
642 * set. So we'd effectivly block all timers until the T2 event
645 if (!__hrtimer_hres_active(cpu_base) || cpu_base->hang_detected)
648 tick_program_event(cpu_base->expires_next, 1);
651 /* High resolution timer related functions */
652 #ifdef CONFIG_HIGH_RES_TIMERS
655 * High resolution timer enabled ?
657 static bool hrtimer_hres_enabled __read_mostly = true;
658 unsigned int hrtimer_resolution __read_mostly = LOW_RES_NSEC;
659 EXPORT_SYMBOL_GPL(hrtimer_resolution);
662 * Enable / Disable high resolution mode
664 static int __init setup_hrtimer_hres(char *str)
666 return (kstrtobool(str, &hrtimer_hres_enabled) == 0);
669 __setup("highres=", setup_hrtimer_hres);
672 * hrtimer_high_res_enabled - query, if the highres mode is enabled
674 static inline int hrtimer_is_hres_enabled(void)
676 return hrtimer_hres_enabled;
680 * Retrigger next event is called after clock was set
682 * Called with interrupts disabled via on_each_cpu()
684 static void retrigger_next_event(void *arg)
686 struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
688 if (!__hrtimer_hres_active(base))
691 raw_spin_lock(&base->lock);
692 hrtimer_update_base(base);
693 hrtimer_force_reprogram(base, 0);
694 raw_spin_unlock(&base->lock);
698 * Switch to high resolution mode
700 static void hrtimer_switch_to_hres(void)
702 struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
704 if (tick_init_highres()) {
705 printk(KERN_WARNING "Could not switch to high resolution "
706 "mode on CPU %d\n", base->cpu);
709 base->hres_active = 1;
710 hrtimer_resolution = HIGH_RES_NSEC;
712 tick_setup_sched_timer();
713 /* "Retrigger" the interrupt to get things going */
714 retrigger_next_event(NULL);
717 static void clock_was_set_work(struct work_struct *work)
722 static DECLARE_WORK(hrtimer_work, clock_was_set_work);
725 * Called from timekeeping and resume code to reprogram the hrtimer
726 * interrupt device on all cpus.
728 void clock_was_set_delayed(void)
730 schedule_work(&hrtimer_work);
735 static inline int hrtimer_is_hres_enabled(void) { return 0; }
736 static inline void hrtimer_switch_to_hres(void) { }
737 static inline void retrigger_next_event(void *arg) { }
739 #endif /* CONFIG_HIGH_RES_TIMERS */
742 * When a timer is enqueued and expires earlier than the already enqueued
743 * timers, we have to check, whether it expires earlier than the timer for
744 * which the clock event device was armed.
746 * Called with interrupts disabled and base->cpu_base.lock held
748 static void hrtimer_reprogram(struct hrtimer *timer, bool reprogram)
750 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
751 struct hrtimer_clock_base *base = timer->base;
752 ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
754 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
757 * CLOCK_REALTIME timer might be requested with an absolute
758 * expiry time which is less than base->offset. Set it to 0.
763 if (timer->is_soft) {
765 * soft hrtimer could be started on a remote CPU. In this
766 * case softirq_expires_next needs to be updated on the
767 * remote CPU. The soft hrtimer will not expire before the
768 * first hard hrtimer on the remote CPU -
769 * hrtimer_check_target() prevents this case.
771 struct hrtimer_cpu_base *timer_cpu_base = base->cpu_base;
773 if (timer_cpu_base->softirq_activated)
776 if (!ktime_before(expires, timer_cpu_base->softirq_expires_next))
779 timer_cpu_base->softirq_next_timer = timer;
780 timer_cpu_base->softirq_expires_next = expires;
782 if (!ktime_before(expires, timer_cpu_base->expires_next) ||
788 * If the timer is not on the current cpu, we cannot reprogram
789 * the other cpus clock event device.
791 if (base->cpu_base != cpu_base)
795 * If the hrtimer interrupt is running, then it will
796 * reevaluate the clock bases and reprogram the clock event
797 * device. The callbacks are always executed in hard interrupt
798 * context so we don't need an extra check for a running
801 if (cpu_base->in_hrtirq)
804 if (expires >= cpu_base->expires_next)
807 /* Update the pointer to the next expiring timer */
808 cpu_base->next_timer = timer;
809 cpu_base->expires_next = expires;
812 * If hres is not active, hardware does not have to be
815 * If a hang was detected in the last timer interrupt then we
816 * do not schedule a timer which is earlier than the expiry
817 * which we enforced in the hang detection. We want the system
820 if (!__hrtimer_hres_active(cpu_base) || cpu_base->hang_detected)
824 * Program the timer hardware. We enforce the expiry for
825 * events which are already in the past.
827 tick_program_event(expires, 1);
831 * Clock realtime was set
833 * Change the offset of the realtime clock vs. the monotonic
836 * We might have to reprogram the high resolution timer interrupt. On
837 * SMP we call the architecture specific code to retrigger _all_ high
838 * resolution timer interrupts. On UP we just disable interrupts and
839 * call the high resolution interrupt code.
841 void clock_was_set(void)
843 #ifdef CONFIG_HIGH_RES_TIMERS
844 /* Retrigger the CPU local events everywhere */
845 on_each_cpu(retrigger_next_event, NULL, 1);
847 timerfd_clock_was_set();
851 * During resume we might have to reprogram the high resolution timer
852 * interrupt on all online CPUs. However, all other CPUs will be
853 * stopped with IRQs interrupts disabled so the clock_was_set() call
856 void hrtimers_resume(void)
858 lockdep_assert_irqs_disabled();
859 /* Retrigger on the local CPU */
860 retrigger_next_event(NULL);
861 /* And schedule a retrigger for all others */
862 clock_was_set_delayed();
866 * Counterpart to lock_hrtimer_base above:
869 void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
871 raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
875 * hrtimer_forward - forward the timer expiry
876 * @timer: hrtimer to forward
877 * @now: forward past this time
878 * @interval: the interval to forward
880 * Forward the timer expiry so it will expire in the future.
881 * Returns the number of overruns.
883 * Can be safely called from the callback function of @timer. If
884 * called from other contexts @timer must neither be enqueued nor
885 * running the callback and the caller needs to take care of
888 * Note: This only updates the timer expiry value and does not requeue
891 u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
896 delta = ktime_sub(now, hrtimer_get_expires(timer));
901 if (WARN_ON(timer->state & HRTIMER_STATE_ENQUEUED))
904 if (interval < hrtimer_resolution)
905 interval = hrtimer_resolution;
907 if (unlikely(delta >= interval)) {
908 s64 incr = ktime_to_ns(interval);
910 orun = ktime_divns(delta, incr);
911 hrtimer_add_expires_ns(timer, incr * orun);
912 if (hrtimer_get_expires_tv64(timer) > now)
915 * This (and the ktime_add() below) is the
916 * correction for exact:
920 hrtimer_add_expires(timer, interval);
924 EXPORT_SYMBOL_GPL(hrtimer_forward);
927 * enqueue_hrtimer - internal function to (re)start a timer
929 * The timer is inserted in expiry order. Insertion into the
930 * red black tree is O(log(n)). Must hold the base lock.
932 * Returns 1 when the new timer is the leftmost timer in the tree.
934 static int enqueue_hrtimer(struct hrtimer *timer,
935 struct hrtimer_clock_base *base,
936 enum hrtimer_mode mode)
938 debug_activate(timer, mode);
940 base->cpu_base->active_bases |= 1 << base->index;
942 timer->state = HRTIMER_STATE_ENQUEUED;
944 return timerqueue_add(&base->active, &timer->node);
948 * __remove_hrtimer - internal function to remove a timer
950 * Caller must hold the base lock.
952 * High resolution timer mode reprograms the clock event device when the
953 * timer is the one which expires next. The caller can disable this by setting
954 * reprogram to zero. This is useful, when the context does a reprogramming
955 * anyway (e.g. timer interrupt)
957 static void __remove_hrtimer(struct hrtimer *timer,
958 struct hrtimer_clock_base *base,
959 u8 newstate, int reprogram)
961 struct hrtimer_cpu_base *cpu_base = base->cpu_base;
962 u8 state = timer->state;
964 timer->state = newstate;
965 if (!(state & HRTIMER_STATE_ENQUEUED))
968 if (!timerqueue_del(&base->active, &timer->node))
969 cpu_base->active_bases &= ~(1 << base->index);
972 * Note: If reprogram is false we do not update
973 * cpu_base->next_timer. This happens when we remove the first
974 * timer on a remote cpu. No harm as we never dereference
975 * cpu_base->next_timer. So the worst thing what can happen is
976 * an superflous call to hrtimer_force_reprogram() on the
977 * remote cpu later on if the same timer gets enqueued again.
979 if (reprogram && timer == cpu_base->next_timer)
980 hrtimer_force_reprogram(cpu_base, 1);
984 * remove hrtimer, called with base lock held
987 remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base, bool restart)
989 if (hrtimer_is_queued(timer)) {
990 u8 state = timer->state;
994 * Remove the timer and force reprogramming when high
995 * resolution mode is active and the timer is on the current
996 * CPU. If we remove a timer on another CPU, reprogramming is
997 * skipped. The interrupt event on this CPU is fired and
998 * reprogramming happens in the interrupt handler. This is a
999 * rare case and less expensive than a smp call.
1001 debug_deactivate(timer);
1002 reprogram = base->cpu_base == this_cpu_ptr(&hrtimer_bases);
1005 state = HRTIMER_STATE_INACTIVE;
1007 __remove_hrtimer(timer, base, state, reprogram);
1013 static inline ktime_t hrtimer_update_lowres(struct hrtimer *timer, ktime_t tim,
1014 const enum hrtimer_mode mode)
1016 #ifdef CONFIG_TIME_LOW_RES
1018 * CONFIG_TIME_LOW_RES indicates that the system has no way to return
1019 * granular time values. For relative timers we add hrtimer_resolution
1020 * (i.e. one jiffie) to prevent short timeouts.
1022 timer->is_rel = mode & HRTIMER_MODE_REL;
1024 tim = ktime_add_safe(tim, hrtimer_resolution);
1030 hrtimer_update_softirq_timer(struct hrtimer_cpu_base *cpu_base, bool reprogram)
1035 * Find the next SOFT expiration.
1037 expires = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_SOFT);
1040 * reprogramming needs to be triggered, even if the next soft
1041 * hrtimer expires at the same time than the next hard
1042 * hrtimer. cpu_base->softirq_expires_next needs to be updated!
1044 if (expires == KTIME_MAX)
1048 * cpu_base->*next_timer is recomputed by __hrtimer_get_next_event()
1049 * cpu_base->*expires_next is only set by hrtimer_reprogram()
1051 hrtimer_reprogram(cpu_base->softirq_next_timer, reprogram);
1054 static int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1055 u64 delta_ns, const enum hrtimer_mode mode,
1056 struct hrtimer_clock_base *base)
1058 struct hrtimer_clock_base *new_base;
1060 /* Remove an active timer from the queue: */
1061 remove_hrtimer(timer, base, true);
1063 if (mode & HRTIMER_MODE_REL)
1064 tim = ktime_add_safe(tim, base->get_time());
1066 tim = hrtimer_update_lowres(timer, tim, mode);
1068 hrtimer_set_expires_range_ns(timer, tim, delta_ns);
1070 /* Switch the timer base, if necessary: */
1071 new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED);
1073 return enqueue_hrtimer(timer, new_base, mode);
1077 * hrtimer_start_range_ns - (re)start an hrtimer
1078 * @timer: the timer to be added
1080 * @delta_ns: "slack" range for the timer
1081 * @mode: timer mode: absolute (HRTIMER_MODE_ABS) or
1082 * relative (HRTIMER_MODE_REL), and pinned (HRTIMER_MODE_PINNED);
1083 * softirq based mode is considered for debug purpose only!
1085 void hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1086 u64 delta_ns, const enum hrtimer_mode mode)
1088 struct hrtimer_clock_base *base;
1089 unsigned long flags;
1092 * Check whether the HRTIMER_MODE_SOFT bit and hrtimer.is_soft
1095 WARN_ON_ONCE(!(mode & HRTIMER_MODE_SOFT) ^ !timer->is_soft);
1097 base = lock_hrtimer_base(timer, &flags);
1099 if (__hrtimer_start_range_ns(timer, tim, delta_ns, mode, base))
1100 hrtimer_reprogram(timer, true);
1102 unlock_hrtimer_base(timer, &flags);
1104 EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
1107 * hrtimer_try_to_cancel - try to deactivate a timer
1108 * @timer: hrtimer to stop
1111 * 0 when the timer was not active
1112 * 1 when the timer was active
1113 * -1 when the timer is currently executing the callback function and
1116 int hrtimer_try_to_cancel(struct hrtimer *timer)
1118 struct hrtimer_clock_base *base;
1119 unsigned long flags;
1123 * Check lockless first. If the timer is not active (neither
1124 * enqueued nor running the callback, nothing to do here. The
1125 * base lock does not serialize against a concurrent enqueue,
1126 * so we can avoid taking it.
1128 if (!hrtimer_active(timer))
1131 base = lock_hrtimer_base(timer, &flags);
1133 if (!hrtimer_callback_running(timer))
1134 ret = remove_hrtimer(timer, base, false);
1136 unlock_hrtimer_base(timer, &flags);
1141 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1144 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1145 * @timer: the timer to be cancelled
1148 * 0 when the timer was not active
1149 * 1 when the timer was active
1151 int hrtimer_cancel(struct hrtimer *timer)
1154 int ret = hrtimer_try_to_cancel(timer);
1161 EXPORT_SYMBOL_GPL(hrtimer_cancel);
1164 * hrtimer_get_remaining - get remaining time for the timer
1165 * @timer: the timer to read
1166 * @adjust: adjust relative timers when CONFIG_TIME_LOW_RES=y
1168 ktime_t __hrtimer_get_remaining(const struct hrtimer *timer, bool adjust)
1170 unsigned long flags;
1173 lock_hrtimer_base(timer, &flags);
1174 if (IS_ENABLED(CONFIG_TIME_LOW_RES) && adjust)
1175 rem = hrtimer_expires_remaining_adjusted(timer);
1177 rem = hrtimer_expires_remaining(timer);
1178 unlock_hrtimer_base(timer, &flags);
1182 EXPORT_SYMBOL_GPL(__hrtimer_get_remaining);
1184 #ifdef CONFIG_NO_HZ_COMMON
1186 * hrtimer_get_next_event - get the time until next expiry event
1188 * Returns the next expiry time or KTIME_MAX if no timer is pending.
1190 u64 hrtimer_get_next_event(void)
1192 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1193 u64 expires = KTIME_MAX;
1194 unsigned long flags;
1196 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1198 if (!__hrtimer_hres_active(cpu_base))
1199 expires = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_ALL);
1201 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1207 static inline int hrtimer_clockid_to_base(clockid_t clock_id)
1209 if (likely(clock_id < MAX_CLOCKS)) {
1210 int base = hrtimer_clock_to_base_table[clock_id];
1212 if (likely(base != HRTIMER_MAX_CLOCK_BASES))
1215 WARN(1, "Invalid clockid %d. Using MONOTONIC\n", clock_id);
1216 return HRTIMER_BASE_MONOTONIC;
1219 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1220 enum hrtimer_mode mode)
1222 bool softtimer = !!(mode & HRTIMER_MODE_SOFT);
1223 int base = softtimer ? HRTIMER_MAX_CLOCK_BASES / 2 : 0;
1224 struct hrtimer_cpu_base *cpu_base;
1226 memset(timer, 0, sizeof(struct hrtimer));
1228 cpu_base = raw_cpu_ptr(&hrtimer_bases);
1231 * POSIX magic: Relative CLOCK_REALTIME timers are not affected by
1232 * clock modifications, so they needs to become CLOCK_MONOTONIC to
1233 * ensure POSIX compliance.
1235 if (clock_id == CLOCK_REALTIME && mode & HRTIMER_MODE_REL)
1236 clock_id = CLOCK_MONOTONIC;
1238 base += hrtimer_clockid_to_base(clock_id);
1239 timer->is_soft = softtimer;
1240 timer->base = &cpu_base->clock_base[base];
1241 timerqueue_init(&timer->node);
1245 * hrtimer_init - initialize a timer to the given clock
1246 * @timer: the timer to be initialized
1247 * @clock_id: the clock to be used
1248 * @mode: The modes which are relevant for intitialization:
1249 * HRTIMER_MODE_ABS, HRTIMER_MODE_REL, HRTIMER_MODE_ABS_SOFT,
1250 * HRTIMER_MODE_REL_SOFT
1252 * The PINNED variants of the above can be handed in,
1253 * but the PINNED bit is ignored as pinning happens
1254 * when the hrtimer is started
1256 void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1257 enum hrtimer_mode mode)
1259 debug_init(timer, clock_id, mode);
1260 __hrtimer_init(timer, clock_id, mode);
1262 EXPORT_SYMBOL_GPL(hrtimer_init);
1265 * A timer is active, when it is enqueued into the rbtree or the
1266 * callback function is running or it's in the state of being migrated
1269 * It is important for this function to not return a false negative.
1271 bool hrtimer_active(const struct hrtimer *timer)
1273 struct hrtimer_clock_base *base;
1277 base = READ_ONCE(timer->base);
1278 seq = raw_read_seqcount_begin(&base->seq);
1280 if (timer->state != HRTIMER_STATE_INACTIVE ||
1281 base->running == timer)
1284 } while (read_seqcount_retry(&base->seq, seq) ||
1285 base != READ_ONCE(timer->base));
1289 EXPORT_SYMBOL_GPL(hrtimer_active);
1292 * The write_seqcount_barrier()s in __run_hrtimer() split the thing into 3
1293 * distinct sections:
1295 * - queued: the timer is queued
1296 * - callback: the timer is being ran
1297 * - post: the timer is inactive or (re)queued
1299 * On the read side we ensure we observe timer->state and cpu_base->running
1300 * from the same section, if anything changed while we looked at it, we retry.
1301 * This includes timer->base changing because sequence numbers alone are
1302 * insufficient for that.
1304 * The sequence numbers are required because otherwise we could still observe
1305 * a false negative if the read side got smeared over multiple consequtive
1306 * __run_hrtimer() invocations.
1309 static void __run_hrtimer(struct hrtimer_cpu_base *cpu_base,
1310 struct hrtimer_clock_base *base,
1311 struct hrtimer *timer, ktime_t *now,
1312 unsigned long flags)
1314 enum hrtimer_restart (*fn)(struct hrtimer *);
1317 lockdep_assert_held(&cpu_base->lock);
1319 debug_deactivate(timer);
1320 base->running = timer;
1323 * Separate the ->running assignment from the ->state assignment.
1325 * As with a regular write barrier, this ensures the read side in
1326 * hrtimer_active() cannot observe base->running == NULL &&
1327 * timer->state == INACTIVE.
1329 raw_write_seqcount_barrier(&base->seq);
1331 __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE, 0);
1332 fn = timer->function;
1335 * Clear the 'is relative' flag for the TIME_LOW_RES case. If the
1336 * timer is restarted with a period then it becomes an absolute
1337 * timer. If its not restarted it does not matter.
1339 if (IS_ENABLED(CONFIG_TIME_LOW_RES))
1340 timer->is_rel = false;
1343 * The timer is marked as running in the CPU base, so it is
1344 * protected against migration to a different CPU even if the lock
1347 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1348 trace_hrtimer_expire_entry(timer, now);
1349 restart = fn(timer);
1350 trace_hrtimer_expire_exit(timer);
1351 raw_spin_lock_irq(&cpu_base->lock);
1354 * Note: We clear the running state after enqueue_hrtimer and
1355 * we do not reprogram the event hardware. Happens either in
1356 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1358 * Note: Because we dropped the cpu_base->lock above,
1359 * hrtimer_start_range_ns() can have popped in and enqueued the timer
1362 if (restart != HRTIMER_NORESTART &&
1363 !(timer->state & HRTIMER_STATE_ENQUEUED))
1364 enqueue_hrtimer(timer, base, HRTIMER_MODE_ABS);
1367 * Separate the ->running assignment from the ->state assignment.
1369 * As with a regular write barrier, this ensures the read side in
1370 * hrtimer_active() cannot observe base->running.timer == NULL &&
1371 * timer->state == INACTIVE.
1373 raw_write_seqcount_barrier(&base->seq);
1375 WARN_ON_ONCE(base->running != timer);
1376 base->running = NULL;
1379 static void __hrtimer_run_queues(struct hrtimer_cpu_base *cpu_base, ktime_t now,
1380 unsigned long flags, unsigned int active_mask)
1382 struct hrtimer_clock_base *base;
1383 unsigned int active = cpu_base->active_bases & active_mask;
1385 for_each_active_base(base, cpu_base, active) {
1386 struct timerqueue_node *node;
1389 basenow = ktime_add(now, base->offset);
1391 while ((node = timerqueue_getnext(&base->active))) {
1392 struct hrtimer *timer;
1394 timer = container_of(node, struct hrtimer, node);
1397 * The immediate goal for using the softexpires is
1398 * minimizing wakeups, not running timers at the
1399 * earliest interrupt after their soft expiration.
1400 * This allows us to avoid using a Priority Search
1401 * Tree, which can answer a stabbing querry for
1402 * overlapping intervals and instead use the simple
1403 * BST we already have.
1404 * We don't add extra wakeups by delaying timers that
1405 * are right-of a not yet expired timer, because that
1406 * timer will have to trigger a wakeup anyway.
1408 if (basenow < hrtimer_get_softexpires_tv64(timer))
1411 __run_hrtimer(cpu_base, base, timer, &basenow, flags);
1416 static __latent_entropy void hrtimer_run_softirq(struct softirq_action *h)
1418 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1419 unsigned long flags;
1422 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1424 now = hrtimer_update_base(cpu_base);
1425 __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_SOFT);
1427 cpu_base->softirq_activated = 0;
1428 hrtimer_update_softirq_timer(cpu_base, true);
1430 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1433 #ifdef CONFIG_HIGH_RES_TIMERS
1436 * High resolution timer interrupt
1437 * Called with interrupts disabled
1439 void hrtimer_interrupt(struct clock_event_device *dev)
1441 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1442 ktime_t expires_next, now, entry_time, delta;
1443 unsigned long flags;
1446 BUG_ON(!cpu_base->hres_active);
1447 cpu_base->nr_events++;
1448 dev->next_event = KTIME_MAX;
1450 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1451 entry_time = now = hrtimer_update_base(cpu_base);
1453 cpu_base->in_hrtirq = 1;
1455 * We set expires_next to KTIME_MAX here with cpu_base->lock
1456 * held to prevent that a timer is enqueued in our queue via
1457 * the migration code. This does not affect enqueueing of
1458 * timers which run their callback and need to be requeued on
1461 cpu_base->expires_next = KTIME_MAX;
1463 if (!ktime_before(now, cpu_base->softirq_expires_next)) {
1464 cpu_base->softirq_expires_next = KTIME_MAX;
1465 cpu_base->softirq_activated = 1;
1466 raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1469 __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_HARD);
1471 /* Reevaluate the clock bases for the next expiry */
1472 expires_next = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_ALL);
1474 * Store the new expiry value so the migration code can verify
1477 cpu_base->expires_next = expires_next;
1478 cpu_base->in_hrtirq = 0;
1479 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1481 /* Reprogramming necessary ? */
1482 if (!tick_program_event(expires_next, 0)) {
1483 cpu_base->hang_detected = 0;
1488 * The next timer was already expired due to:
1490 * - long lasting callbacks
1491 * - being scheduled away when running in a VM
1493 * We need to prevent that we loop forever in the hrtimer
1494 * interrupt routine. We give it 3 attempts to avoid
1495 * overreacting on some spurious event.
1497 * Acquire base lock for updating the offsets and retrieving
1500 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1501 now = hrtimer_update_base(cpu_base);
1502 cpu_base->nr_retries++;
1506 * Give the system a chance to do something else than looping
1507 * here. We stored the entry time, so we know exactly how long
1508 * we spent here. We schedule the next event this amount of
1511 cpu_base->nr_hangs++;
1512 cpu_base->hang_detected = 1;
1513 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1515 delta = ktime_sub(now, entry_time);
1516 if ((unsigned int)delta > cpu_base->max_hang_time)
1517 cpu_base->max_hang_time = (unsigned int) delta;
1519 * Limit it to a sensible value as we enforce a longer
1520 * delay. Give the CPU at least 100ms to catch up.
1522 if (delta > 100 * NSEC_PER_MSEC)
1523 expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1525 expires_next = ktime_add(now, delta);
1526 tick_program_event(expires_next, 1);
1527 printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n",
1528 ktime_to_ns(delta));
1531 /* called with interrupts disabled */
1532 static inline void __hrtimer_peek_ahead_timers(void)
1534 struct tick_device *td;
1536 if (!hrtimer_hres_active())
1539 td = this_cpu_ptr(&tick_cpu_device);
1540 if (td && td->evtdev)
1541 hrtimer_interrupt(td->evtdev);
1544 #else /* CONFIG_HIGH_RES_TIMERS */
1546 static inline void __hrtimer_peek_ahead_timers(void) { }
1548 #endif /* !CONFIG_HIGH_RES_TIMERS */
1551 * Called from run_local_timers in hardirq context every jiffy
1553 void hrtimer_run_queues(void)
1555 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1556 unsigned long flags;
1559 if (__hrtimer_hres_active(cpu_base))
1563 * This _is_ ugly: We have to check periodically, whether we
1564 * can switch to highres and / or nohz mode. The clocksource
1565 * switch happens with xtime_lock held. Notification from
1566 * there only sets the check bit in the tick_oneshot code,
1567 * otherwise we might deadlock vs. xtime_lock.
1569 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) {
1570 hrtimer_switch_to_hres();
1574 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1575 now = hrtimer_update_base(cpu_base);
1577 if (!ktime_before(now, cpu_base->softirq_expires_next)) {
1578 cpu_base->softirq_expires_next = KTIME_MAX;
1579 cpu_base->softirq_activated = 1;
1580 raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1583 __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_HARD);
1584 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1588 * Sleep related functions:
1590 static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1592 struct hrtimer_sleeper *t =
1593 container_of(timer, struct hrtimer_sleeper, timer);
1594 struct task_struct *task = t->task;
1598 wake_up_process(task);
1600 return HRTIMER_NORESTART;
1603 void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1605 sl->timer.function = hrtimer_wakeup;
1608 EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
1610 int nanosleep_copyout(struct restart_block *restart, struct timespec64 *ts)
1612 switch(restart->nanosleep.type) {
1613 #ifdef CONFIG_COMPAT
1615 if (compat_put_timespec64(ts, restart->nanosleep.compat_rmtp))
1620 if (put_timespec64(ts, restart->nanosleep.rmtp))
1626 return -ERESTART_RESTARTBLOCK;
1629 static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1631 struct restart_block *restart;
1633 hrtimer_init_sleeper(t, current);
1636 set_current_state(TASK_INTERRUPTIBLE);
1637 hrtimer_start_expires(&t->timer, mode);
1639 if (likely(t->task))
1640 freezable_schedule();
1642 hrtimer_cancel(&t->timer);
1643 mode = HRTIMER_MODE_ABS;
1645 } while (t->task && !signal_pending(current));
1647 __set_current_state(TASK_RUNNING);
1652 restart = ¤t->restart_block;
1653 if (restart->nanosleep.type != TT_NONE) {
1654 ktime_t rem = hrtimer_expires_remaining(&t->timer);
1655 struct timespec64 rmt;
1659 rmt = ktime_to_timespec64(rem);
1661 return nanosleep_copyout(restart, &rmt);
1663 return -ERESTART_RESTARTBLOCK;
1666 static long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1668 struct hrtimer_sleeper t;
1671 hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid,
1673 hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
1675 ret = do_nanosleep(&t, HRTIMER_MODE_ABS);
1676 destroy_hrtimer_on_stack(&t.timer);
1680 long hrtimer_nanosleep(const struct timespec64 *rqtp,
1681 const enum hrtimer_mode mode, const clockid_t clockid)
1683 struct restart_block *restart;
1684 struct hrtimer_sleeper t;
1688 slack = current->timer_slack_ns;
1689 if (dl_task(current) || rt_task(current))
1692 hrtimer_init_on_stack(&t.timer, clockid, mode);
1693 hrtimer_set_expires_range_ns(&t.timer, timespec64_to_ktime(*rqtp), slack);
1694 ret = do_nanosleep(&t, mode);
1695 if (ret != -ERESTART_RESTARTBLOCK)
1698 /* Absolute timers do not update the rmtp value and restart: */
1699 if (mode == HRTIMER_MODE_ABS) {
1700 ret = -ERESTARTNOHAND;
1704 restart = ¤t->restart_block;
1705 restart->fn = hrtimer_nanosleep_restart;
1706 restart->nanosleep.clockid = t.timer.base->clockid;
1707 restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
1709 destroy_hrtimer_on_stack(&t.timer);
1713 SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp,
1714 struct timespec __user *, rmtp)
1716 struct timespec64 tu;
1718 if (get_timespec64(&tu, rqtp))
1721 if (!timespec64_valid(&tu))
1724 current->restart_block.nanosleep.type = rmtp ? TT_NATIVE : TT_NONE;
1725 current->restart_block.nanosleep.rmtp = rmtp;
1726 return hrtimer_nanosleep(&tu, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1729 #ifdef CONFIG_COMPAT
1731 COMPAT_SYSCALL_DEFINE2(nanosleep, struct compat_timespec __user *, rqtp,
1732 struct compat_timespec __user *, rmtp)
1734 struct timespec64 tu;
1736 if (compat_get_timespec64(&tu, rqtp))
1739 if (!timespec64_valid(&tu))
1742 current->restart_block.nanosleep.type = rmtp ? TT_COMPAT : TT_NONE;
1743 current->restart_block.nanosleep.compat_rmtp = rmtp;
1744 return hrtimer_nanosleep(&tu, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1749 * Functions related to boot-time initialization:
1751 int hrtimers_prepare_cpu(unsigned int cpu)
1753 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1756 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1757 cpu_base->clock_base[i].cpu_base = cpu_base;
1758 timerqueue_init_head(&cpu_base->clock_base[i].active);
1761 cpu_base->cpu = cpu;
1762 cpu_base->active_bases = 0;
1763 cpu_base->hres_active = 0;
1764 cpu_base->hang_detected = 0;
1765 cpu_base->next_timer = NULL;
1766 cpu_base->softirq_next_timer = NULL;
1767 cpu_base->expires_next = KTIME_MAX;
1768 cpu_base->softirq_expires_next = KTIME_MAX;
1772 #ifdef CONFIG_HOTPLUG_CPU
1774 static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1775 struct hrtimer_clock_base *new_base)
1777 struct hrtimer *timer;
1778 struct timerqueue_node *node;
1780 while ((node = timerqueue_getnext(&old_base->active))) {
1781 timer = container_of(node, struct hrtimer, node);
1782 BUG_ON(hrtimer_callback_running(timer));
1783 debug_deactivate(timer);
1786 * Mark it as ENQUEUED not INACTIVE otherwise the
1787 * timer could be seen as !active and just vanish away
1788 * under us on another CPU
1790 __remove_hrtimer(timer, old_base, HRTIMER_STATE_ENQUEUED, 0);
1791 timer->base = new_base;
1793 * Enqueue the timers on the new cpu. This does not
1794 * reprogram the event device in case the timer
1795 * expires before the earliest on this CPU, but we run
1796 * hrtimer_interrupt after we migrated everything to
1797 * sort out already expired timers and reprogram the
1800 enqueue_hrtimer(timer, new_base, HRTIMER_MODE_ABS);
1804 int hrtimers_dead_cpu(unsigned int scpu)
1806 struct hrtimer_cpu_base *old_base, *new_base;
1809 BUG_ON(cpu_online(scpu));
1810 tick_cancel_sched_timer(scpu);
1813 * this BH disable ensures that raise_softirq_irqoff() does
1814 * not wakeup ksoftirqd (and acquire the pi-lock) while
1815 * holding the cpu_base lock
1818 local_irq_disable();
1819 old_base = &per_cpu(hrtimer_bases, scpu);
1820 new_base = this_cpu_ptr(&hrtimer_bases);
1822 * The caller is globally serialized and nobody else
1823 * takes two locks at once, deadlock is not possible.
1825 raw_spin_lock(&new_base->lock);
1826 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1828 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1829 migrate_hrtimer_list(&old_base->clock_base[i],
1830 &new_base->clock_base[i]);
1834 * The migration might have changed the first expiring softirq
1835 * timer on this CPU. Update it.
1837 hrtimer_update_softirq_timer(new_base, false);
1839 raw_spin_unlock(&old_base->lock);
1840 raw_spin_unlock(&new_base->lock);
1842 /* Check, if we got expired work to do */
1843 __hrtimer_peek_ahead_timers();
1849 #endif /* CONFIG_HOTPLUG_CPU */
1851 void __init hrtimers_init(void)
1853 hrtimers_prepare_cpu(smp_processor_id());
1854 open_softirq(HRTIMER_SOFTIRQ, hrtimer_run_softirq);
1858 * schedule_hrtimeout_range_clock - sleep until timeout
1859 * @expires: timeout value (ktime_t)
1860 * @delta: slack in expires timeout (ktime_t)
1862 * @clock_id: timer clock to be used
1865 schedule_hrtimeout_range_clock(ktime_t *expires, u64 delta,
1866 const enum hrtimer_mode mode, clockid_t clock_id)
1868 struct hrtimer_sleeper t;
1871 * Optimize when a zero timeout value is given. It does not
1872 * matter whether this is an absolute or a relative time.
1874 if (expires && *expires == 0) {
1875 __set_current_state(TASK_RUNNING);
1880 * A NULL parameter means "infinite"
1887 hrtimer_init_on_stack(&t.timer, clock_id, mode);
1888 hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
1890 hrtimer_init_sleeper(&t, current);
1892 hrtimer_start_expires(&t.timer, mode);
1897 hrtimer_cancel(&t.timer);
1898 destroy_hrtimer_on_stack(&t.timer);
1900 __set_current_state(TASK_RUNNING);
1902 return !t.task ? 0 : -EINTR;
1906 * schedule_hrtimeout_range - sleep until timeout
1907 * @expires: timeout value (ktime_t)
1908 * @delta: slack in expires timeout (ktime_t)
1911 * Make the current task sleep until the given expiry time has
1912 * elapsed. The routine will return immediately unless
1913 * the current task state has been set (see set_current_state()).
1915 * The @delta argument gives the kernel the freedom to schedule the
1916 * actual wakeup to a time that is both power and performance friendly.
1917 * The kernel give the normal best effort behavior for "@expires+@delta",
1918 * but may decide to fire the timer earlier, but no earlier than @expires.
1920 * You can set the task state as follows -
1922 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1923 * pass before the routine returns unless the current task is explicitly
1924 * woken up, (e.g. by wake_up_process()).
1926 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1927 * delivered to the current task or the current task is explicitly woken
1930 * The current task state is guaranteed to be TASK_RUNNING when this
1933 * Returns 0 when the timer has expired. If the task was woken before the
1934 * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or
1935 * by an explicit wakeup, it returns -EINTR.
1937 int __sched schedule_hrtimeout_range(ktime_t *expires, u64 delta,
1938 const enum hrtimer_mode mode)
1940 return schedule_hrtimeout_range_clock(expires, delta, mode,
1943 EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
1946 * schedule_hrtimeout - sleep until timeout
1947 * @expires: timeout value (ktime_t)
1950 * Make the current task sleep until the given expiry time has
1951 * elapsed. The routine will return immediately unless
1952 * the current task state has been set (see set_current_state()).
1954 * You can set the task state as follows -
1956 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1957 * pass before the routine returns unless the current task is explicitly
1958 * woken up, (e.g. by wake_up_process()).
1960 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1961 * delivered to the current task or the current task is explicitly woken
1964 * The current task state is guaranteed to be TASK_RUNNING when this
1967 * Returns 0 when the timer has expired. If the task was woken before the
1968 * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or
1969 * by an explicit wakeup, it returns -EINTR.
1971 int __sched schedule_hrtimeout(ktime_t *expires,
1972 const enum hrtimer_mode mode)
1974 return schedule_hrtimeout_range(expires, 0, mode);
1976 EXPORT_SYMBOL_GPL(schedule_hrtimeout);