2 * linux/kernel/hrtimer.c
4 * Copyright(C) 2005-2006, Thomas Gleixner <tglx@linutronix.de>
5 * Copyright(C) 2005-2007, Red Hat, Inc., Ingo Molnar
6 * Copyright(C) 2006-2007 Timesys Corp., Thomas Gleixner
8 * High-resolution kernel timers
10 * In contrast to the low-resolution timeout API implemented in
11 * kernel/timer.c, hrtimers provide finer resolution and accuracy
12 * depending on system configuration and capabilities.
14 * These timers are currently used for:
18 * - precise in-kernel timing
20 * Started by: Thomas Gleixner and Ingo Molnar
23 * based on kernel/timer.c
25 * Help, testing, suggestions, bugfixes, improvements were
28 * George Anzinger, Andrew Morton, Steven Rostedt, Roman Zippel
31 * For licencing details see kernel-base/COPYING
34 #include <linux/cpu.h>
35 #include <linux/export.h>
36 #include <linux/percpu.h>
37 #include <linux/hrtimer.h>
38 #include <linux/notifier.h>
39 #include <linux/syscalls.h>
40 #include <linux/kallsyms.h>
41 #include <linux/interrupt.h>
42 #include <linux/tick.h>
43 #include <linux/seq_file.h>
44 #include <linux/err.h>
45 #include <linux/debugobjects.h>
46 #include <linux/sched.h>
47 #include <linux/sched/sysctl.h>
48 #include <linux/sched/rt.h>
49 #include <linux/sched/deadline.h>
50 #include <linux/timer.h>
51 #include <linux/freezer.h>
53 #include <asm/uaccess.h>
55 #include <trace/events/timer.h>
60 * There are more clockids then hrtimer bases. Thus, we index
61 * into the timer bases by the hrtimer_base_type enum. When trying
62 * to reach a base using a clockid, hrtimer_clockid_to_base()
63 * is used to convert from clockid to the proper hrtimer_base_type.
65 DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
68 .lock = __RAW_SPIN_LOCK_UNLOCKED(hrtimer_bases.lock),
72 .index = HRTIMER_BASE_MONOTONIC,
73 .clockid = CLOCK_MONOTONIC,
74 .get_time = &ktime_get,
75 .resolution = KTIME_LOW_RES,
78 .index = HRTIMER_BASE_REALTIME,
79 .clockid = CLOCK_REALTIME,
80 .get_time = &ktime_get_real,
81 .resolution = KTIME_LOW_RES,
84 .index = HRTIMER_BASE_BOOTTIME,
85 .clockid = CLOCK_BOOTTIME,
86 .get_time = &ktime_get_boottime,
87 .resolution = KTIME_LOW_RES,
90 .index = HRTIMER_BASE_TAI,
92 .get_time = &ktime_get_clocktai,
93 .resolution = KTIME_LOW_RES,
98 static const int hrtimer_clock_to_base_table[MAX_CLOCKS] = {
99 [CLOCK_REALTIME] = HRTIMER_BASE_REALTIME,
100 [CLOCK_MONOTONIC] = HRTIMER_BASE_MONOTONIC,
101 [CLOCK_BOOTTIME] = HRTIMER_BASE_BOOTTIME,
102 [CLOCK_TAI] = HRTIMER_BASE_TAI,
105 static inline int hrtimer_clockid_to_base(clockid_t clock_id)
107 return hrtimer_clock_to_base_table[clock_id];
112 * Get the coarse grained time at the softirq based on xtime and
115 static void hrtimer_get_softirq_time(struct hrtimer_cpu_base *base)
117 ktime_t xtim, mono, boot;
118 struct timespec xts, tom, slp;
121 get_xtime_and_monotonic_and_sleep_offset(&xts, &tom, &slp);
122 tai_offset = timekeeping_get_tai_offset();
124 xtim = timespec_to_ktime(xts);
125 mono = ktime_add(xtim, timespec_to_ktime(tom));
126 boot = ktime_add(mono, timespec_to_ktime(slp));
127 base->clock_base[HRTIMER_BASE_REALTIME].softirq_time = xtim;
128 base->clock_base[HRTIMER_BASE_MONOTONIC].softirq_time = mono;
129 base->clock_base[HRTIMER_BASE_BOOTTIME].softirq_time = boot;
130 base->clock_base[HRTIMER_BASE_TAI].softirq_time =
131 ktime_add(xtim, ktime_set(tai_offset, 0));
135 * Functions and macros which are different for UP/SMP systems are kept in a
141 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
142 * means that all timers which are tied to this base via timer->base are
143 * locked, and the base itself is locked too.
145 * So __run_timers/migrate_timers can safely modify all timers which could
146 * be found on the lists/queues.
148 * When the timer's base is locked, and the timer removed from list, it is
149 * possible to set timer->base = NULL and drop the lock: the timer remains
153 struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
154 unsigned long *flags)
156 struct hrtimer_clock_base *base;
160 if (likely(base != NULL)) {
161 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
162 if (likely(base == timer->base))
164 /* The timer has migrated to another CPU: */
165 raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
172 * With HIGHRES=y we do not migrate the timer when it is expiring
173 * before the next event on the target cpu because we cannot reprogram
174 * the target cpu hardware and we would cause it to fire late.
176 * Called with cpu_base->lock of target cpu held.
179 hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base)
181 #ifdef CONFIG_HIGH_RES_TIMERS
184 if (!new_base->cpu_base->hres_active)
187 expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset);
188 return expires.tv64 <= new_base->cpu_base->expires_next.tv64;
195 * Switch the timer base to the current CPU when possible.
197 static inline struct hrtimer_clock_base *
198 switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
201 struct hrtimer_clock_base *new_base;
202 struct hrtimer_cpu_base *new_cpu_base;
203 int this_cpu = smp_processor_id();
204 int cpu = get_nohz_timer_target(pinned);
205 int basenum = base->index;
208 new_cpu_base = &per_cpu(hrtimer_bases, cpu);
209 new_base = &new_cpu_base->clock_base[basenum];
211 if (base != new_base) {
213 * We are trying to move timer to new_base.
214 * However we can't change timer's base while it is running,
215 * so we keep it on the same CPU. No hassle vs. reprogramming
216 * the event source in the high resolution case. The softirq
217 * code will take care of this when the timer function has
218 * completed. There is no conflict as we hold the lock until
219 * the timer is enqueued.
221 if (unlikely(hrtimer_callback_running(timer)))
224 /* See the comment in lock_timer_base() */
226 raw_spin_unlock(&base->cpu_base->lock);
227 raw_spin_lock(&new_base->cpu_base->lock);
229 if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) {
231 raw_spin_unlock(&new_base->cpu_base->lock);
232 raw_spin_lock(&base->cpu_base->lock);
236 timer->base = new_base;
241 #else /* CONFIG_SMP */
243 static inline struct hrtimer_clock_base *
244 lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
246 struct hrtimer_clock_base *base = timer->base;
248 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
253 # define switch_hrtimer_base(t, b, p) (b)
255 #endif /* !CONFIG_SMP */
258 * Functions for the union type storage format of ktime_t which are
259 * too large for inlining:
261 #if BITS_PER_LONG < 64
262 # ifndef CONFIG_KTIME_SCALAR
264 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
266 * @nsec: the scalar nsec value to add
268 * Returns the sum of kt and nsec in ktime_t format
270 ktime_t ktime_add_ns(const ktime_t kt, u64 nsec)
274 if (likely(nsec < NSEC_PER_SEC)) {
277 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
279 /* Make sure nsec fits into long */
280 if (unlikely(nsec > KTIME_SEC_MAX))
281 return (ktime_t){ .tv64 = KTIME_MAX };
283 tmp = ktime_set((long)nsec, rem);
286 return ktime_add(kt, tmp);
289 EXPORT_SYMBOL_GPL(ktime_add_ns);
292 * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable
294 * @nsec: the scalar nsec value to subtract
296 * Returns the subtraction of @nsec from @kt in ktime_t format
298 ktime_t ktime_sub_ns(const ktime_t kt, u64 nsec)
302 if (likely(nsec < NSEC_PER_SEC)) {
305 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
307 tmp = ktime_set((long)nsec, rem);
310 return ktime_sub(kt, tmp);
313 EXPORT_SYMBOL_GPL(ktime_sub_ns);
314 # endif /* !CONFIG_KTIME_SCALAR */
317 * Divide a ktime value by a nanosecond value
319 u64 ktime_divns(const ktime_t kt, s64 div)
324 dclc = ktime_to_ns(kt);
325 /* Make sure the divisor is less than 2^32: */
331 do_div(dclc, (unsigned long) div);
335 #endif /* BITS_PER_LONG >= 64 */
338 * Add two ktime values and do a safety check for overflow:
340 ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
342 ktime_t res = ktime_add(lhs, rhs);
345 * We use KTIME_SEC_MAX here, the maximum timeout which we can
346 * return to user space in a timespec:
348 if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64)
349 res = ktime_set(KTIME_SEC_MAX, 0);
354 EXPORT_SYMBOL_GPL(ktime_add_safe);
356 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
358 static struct debug_obj_descr hrtimer_debug_descr;
360 static void *hrtimer_debug_hint(void *addr)
362 return ((struct hrtimer *) addr)->function;
366 * fixup_init is called when:
367 * - an active object is initialized
369 static int hrtimer_fixup_init(void *addr, enum debug_obj_state state)
371 struct hrtimer *timer = addr;
374 case ODEBUG_STATE_ACTIVE:
375 hrtimer_cancel(timer);
376 debug_object_init(timer, &hrtimer_debug_descr);
384 * fixup_activate is called when:
385 * - an active object is activated
386 * - an unknown object is activated (might be a statically initialized object)
388 static int hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
392 case ODEBUG_STATE_NOTAVAILABLE:
396 case ODEBUG_STATE_ACTIVE:
405 * fixup_free is called when:
406 * - an active object is freed
408 static int hrtimer_fixup_free(void *addr, enum debug_obj_state state)
410 struct hrtimer *timer = addr;
413 case ODEBUG_STATE_ACTIVE:
414 hrtimer_cancel(timer);
415 debug_object_free(timer, &hrtimer_debug_descr);
422 static struct debug_obj_descr hrtimer_debug_descr = {
424 .debug_hint = hrtimer_debug_hint,
425 .fixup_init = hrtimer_fixup_init,
426 .fixup_activate = hrtimer_fixup_activate,
427 .fixup_free = hrtimer_fixup_free,
430 static inline void debug_hrtimer_init(struct hrtimer *timer)
432 debug_object_init(timer, &hrtimer_debug_descr);
435 static inline void debug_hrtimer_activate(struct hrtimer *timer)
437 debug_object_activate(timer, &hrtimer_debug_descr);
440 static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
442 debug_object_deactivate(timer, &hrtimer_debug_descr);
445 static inline void debug_hrtimer_free(struct hrtimer *timer)
447 debug_object_free(timer, &hrtimer_debug_descr);
450 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
451 enum hrtimer_mode mode);
453 void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
454 enum hrtimer_mode mode)
456 debug_object_init_on_stack(timer, &hrtimer_debug_descr);
457 __hrtimer_init(timer, clock_id, mode);
459 EXPORT_SYMBOL_GPL(hrtimer_init_on_stack);
461 void destroy_hrtimer_on_stack(struct hrtimer *timer)
463 debug_object_free(timer, &hrtimer_debug_descr);
467 static inline void debug_hrtimer_init(struct hrtimer *timer) { }
468 static inline void debug_hrtimer_activate(struct hrtimer *timer) { }
469 static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
473 debug_init(struct hrtimer *timer, clockid_t clockid,
474 enum hrtimer_mode mode)
476 debug_hrtimer_init(timer);
477 trace_hrtimer_init(timer, clockid, mode);
480 static inline void debug_activate(struct hrtimer *timer)
482 debug_hrtimer_activate(timer);
483 trace_hrtimer_start(timer);
486 static inline void debug_deactivate(struct hrtimer *timer)
488 debug_hrtimer_deactivate(timer);
489 trace_hrtimer_cancel(timer);
492 /* High resolution timer related functions */
493 #ifdef CONFIG_HIGH_RES_TIMERS
496 * High resolution timer enabled ?
498 static int hrtimer_hres_enabled __read_mostly = 1;
501 * Enable / Disable high resolution mode
503 static int __init setup_hrtimer_hres(char *str)
505 if (!strcmp(str, "off"))
506 hrtimer_hres_enabled = 0;
507 else if (!strcmp(str, "on"))
508 hrtimer_hres_enabled = 1;
514 __setup("highres=", setup_hrtimer_hres);
517 * hrtimer_high_res_enabled - query, if the highres mode is enabled
519 static inline int hrtimer_is_hres_enabled(void)
521 return hrtimer_hres_enabled;
525 * Is the high resolution mode active ?
527 static inline int hrtimer_hres_active(void)
529 return __this_cpu_read(hrtimer_bases.hres_active);
533 * Reprogram the event source with checking both queues for the
535 * Called with interrupts disabled and base->lock held
538 hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
541 struct hrtimer_clock_base *base = cpu_base->clock_base;
542 ktime_t expires, expires_next;
544 expires_next.tv64 = KTIME_MAX;
546 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
547 struct hrtimer *timer;
548 struct timerqueue_node *next;
550 next = timerqueue_getnext(&base->active);
553 timer = container_of(next, struct hrtimer, node);
555 expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
557 * clock_was_set() has changed base->offset so the
558 * result might be negative. Fix it up to prevent a
559 * false positive in clockevents_program_event()
561 if (expires.tv64 < 0)
563 if (expires.tv64 < expires_next.tv64)
564 expires_next = expires;
567 if (skip_equal && expires_next.tv64 == cpu_base->expires_next.tv64)
570 cpu_base->expires_next.tv64 = expires_next.tv64;
573 * If a hang was detected in the last timer interrupt then we
574 * leave the hang delay active in the hardware. We want the
575 * system to make progress. That also prevents the following
577 * T1 expires 50ms from now
578 * T2 expires 5s from now
580 * T1 is removed, so this code is called and would reprogram
581 * the hardware to 5s from now. Any hrtimer_start after that
582 * will not reprogram the hardware due to hang_detected being
583 * set. So we'd effectivly block all timers until the T2 event
586 if (cpu_base->hang_detected)
589 if (cpu_base->expires_next.tv64 != KTIME_MAX)
590 tick_program_event(cpu_base->expires_next, 1);
594 * Shared reprogramming for clock_realtime and clock_monotonic
596 * When a timer is enqueued and expires earlier than the already enqueued
597 * timers, we have to check, whether it expires earlier than the timer for
598 * which the clock event device was armed.
600 * Called with interrupts disabled and base->cpu_base.lock held
602 static int hrtimer_reprogram(struct hrtimer *timer,
603 struct hrtimer_clock_base *base)
605 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
606 ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
609 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
612 * When the callback is running, we do not reprogram the clock event
613 * device. The timer callback is either running on a different CPU or
614 * the callback is executed in the hrtimer_interrupt context. The
615 * reprogramming is handled either by the softirq, which called the
616 * callback or at the end of the hrtimer_interrupt.
618 if (hrtimer_callback_running(timer))
622 * CLOCK_REALTIME timer might be requested with an absolute
623 * expiry time which is less than base->offset. Nothing wrong
624 * about that, just avoid to call into the tick code, which
625 * has now objections against negative expiry values.
627 if (expires.tv64 < 0)
630 if (expires.tv64 >= cpu_base->expires_next.tv64)
634 * If a hang was detected in the last timer interrupt then we
635 * do not schedule a timer which is earlier than the expiry
636 * which we enforced in the hang detection. We want the system
639 if (cpu_base->hang_detected)
643 * Clockevents returns -ETIME, when the event was in the past.
645 res = tick_program_event(expires, 0);
646 if (!IS_ERR_VALUE(res))
647 cpu_base->expires_next = expires;
652 * Initialize the high resolution related parts of cpu_base
654 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
656 base->expires_next.tv64 = KTIME_MAX;
657 base->hres_active = 0;
661 * When High resolution timers are active, try to reprogram. Note, that in case
662 * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry
663 * check happens. The timer gets enqueued into the rbtree. The reprogramming
664 * and expiry check is done in the hrtimer_interrupt or in the softirq.
666 static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
667 struct hrtimer_clock_base *base)
669 return base->cpu_base->hres_active && hrtimer_reprogram(timer, base);
672 static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base)
674 ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset;
675 ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset;
676 ktime_t *offs_tai = &base->clock_base[HRTIMER_BASE_TAI].offset;
678 return ktime_get_update_offsets(offs_real, offs_boot, offs_tai);
682 * Retrigger next event is called after clock was set
684 * Called with interrupts disabled via on_each_cpu()
686 static void retrigger_next_event(void *arg)
688 struct hrtimer_cpu_base *base = &__get_cpu_var(hrtimer_bases);
690 if (!hrtimer_hres_active())
693 raw_spin_lock(&base->lock);
694 hrtimer_update_base(base);
695 hrtimer_force_reprogram(base, 0);
696 raw_spin_unlock(&base->lock);
700 * Switch to high resolution mode
702 static int hrtimer_switch_to_hres(void)
704 int i, cpu = smp_processor_id();
705 struct hrtimer_cpu_base *base = &per_cpu(hrtimer_bases, cpu);
708 if (base->hres_active)
711 local_irq_save(flags);
713 if (tick_init_highres()) {
714 local_irq_restore(flags);
715 printk(KERN_WARNING "Could not switch to high resolution "
716 "mode on CPU %d\n", cpu);
719 base->hres_active = 1;
720 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
721 base->clock_base[i].resolution = KTIME_HIGH_RES;
723 tick_setup_sched_timer();
724 /* "Retrigger" the interrupt to get things going */
725 retrigger_next_event(NULL);
726 local_irq_restore(flags);
730 static void clock_was_set_work(struct work_struct *work)
735 static DECLARE_WORK(hrtimer_work, clock_was_set_work);
738 * Called from timekeeping and resume code to reprogramm the hrtimer
739 * interrupt device on all cpus.
741 void clock_was_set_delayed(void)
743 schedule_work(&hrtimer_work);
748 static inline int hrtimer_hres_active(void) { return 0; }
749 static inline int hrtimer_is_hres_enabled(void) { return 0; }
750 static inline int hrtimer_switch_to_hres(void) { return 0; }
752 hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { }
753 static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
754 struct hrtimer_clock_base *base)
758 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { }
759 static inline void retrigger_next_event(void *arg) { }
761 #endif /* CONFIG_HIGH_RES_TIMERS */
764 * Clock realtime was set
766 * Change the offset of the realtime clock vs. the monotonic
769 * We might have to reprogram the high resolution timer interrupt. On
770 * SMP we call the architecture specific code to retrigger _all_ high
771 * resolution timer interrupts. On UP we just disable interrupts and
772 * call the high resolution interrupt code.
774 void clock_was_set(void)
776 #ifdef CONFIG_HIGH_RES_TIMERS
777 /* Retrigger the CPU local events everywhere */
778 on_each_cpu(retrigger_next_event, NULL, 1);
780 timerfd_clock_was_set();
784 * During resume we might have to reprogram the high resolution timer
785 * interrupt on all online CPUs. However, all other CPUs will be
786 * stopped with IRQs interrupts disabled so the clock_was_set() call
789 void hrtimers_resume(void)
791 WARN_ONCE(!irqs_disabled(),
792 KERN_INFO "hrtimers_resume() called with IRQs enabled!");
794 /* Retrigger on the local CPU */
795 retrigger_next_event(NULL);
796 /* And schedule a retrigger for all others */
797 clock_was_set_delayed();
800 static inline void timer_stats_hrtimer_set_start_info(struct hrtimer *timer)
802 #ifdef CONFIG_TIMER_STATS
803 if (timer->start_site)
805 timer->start_site = __builtin_return_address(0);
806 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
807 timer->start_pid = current->pid;
811 static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer *timer)
813 #ifdef CONFIG_TIMER_STATS
814 timer->start_site = NULL;
818 static inline void timer_stats_account_hrtimer(struct hrtimer *timer)
820 #ifdef CONFIG_TIMER_STATS
821 if (likely(!timer_stats_active))
823 timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
824 timer->function, timer->start_comm, 0);
829 * Counterpart to lock_hrtimer_base above:
832 void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
834 raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
838 * hrtimer_forward - forward the timer expiry
839 * @timer: hrtimer to forward
840 * @now: forward past this time
841 * @interval: the interval to forward
843 * Forward the timer expiry so it will expire in the future.
844 * Returns the number of overruns.
846 u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
851 delta = ktime_sub(now, hrtimer_get_expires(timer));
856 if (interval.tv64 < timer->base->resolution.tv64)
857 interval.tv64 = timer->base->resolution.tv64;
859 if (unlikely(delta.tv64 >= interval.tv64)) {
860 s64 incr = ktime_to_ns(interval);
862 orun = ktime_divns(delta, incr);
863 hrtimer_add_expires_ns(timer, incr * orun);
864 if (hrtimer_get_expires_tv64(timer) > now.tv64)
867 * This (and the ktime_add() below) is the
868 * correction for exact:
872 hrtimer_add_expires(timer, interval);
876 EXPORT_SYMBOL_GPL(hrtimer_forward);
879 * enqueue_hrtimer - internal function to (re)start a timer
881 * The timer is inserted in expiry order. Insertion into the
882 * red black tree is O(log(n)). Must hold the base lock.
884 * Returns 1 when the new timer is the leftmost timer in the tree.
886 static int enqueue_hrtimer(struct hrtimer *timer,
887 struct hrtimer_clock_base *base)
889 debug_activate(timer);
891 timerqueue_add(&base->active, &timer->node);
892 base->cpu_base->active_bases |= 1 << base->index;
895 * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
896 * state of a possibly running callback.
898 timer->state |= HRTIMER_STATE_ENQUEUED;
900 return (&timer->node == base->active.next);
904 * __remove_hrtimer - internal function to remove a timer
906 * Caller must hold the base lock.
908 * High resolution timer mode reprograms the clock event device when the
909 * timer is the one which expires next. The caller can disable this by setting
910 * reprogram to zero. This is useful, when the context does a reprogramming
911 * anyway (e.g. timer interrupt)
913 static void __remove_hrtimer(struct hrtimer *timer,
914 struct hrtimer_clock_base *base,
915 unsigned long newstate, int reprogram)
917 struct timerqueue_node *next_timer;
918 if (!(timer->state & HRTIMER_STATE_ENQUEUED))
921 next_timer = timerqueue_getnext(&base->active);
922 timerqueue_del(&base->active, &timer->node);
923 if (&timer->node == next_timer) {
924 #ifdef CONFIG_HIGH_RES_TIMERS
925 /* Reprogram the clock event device. if enabled */
926 if (reprogram && hrtimer_hres_active()) {
929 expires = ktime_sub(hrtimer_get_expires(timer),
931 if (base->cpu_base->expires_next.tv64 == expires.tv64)
932 hrtimer_force_reprogram(base->cpu_base, 1);
936 if (!timerqueue_getnext(&base->active))
937 base->cpu_base->active_bases &= ~(1 << base->index);
939 timer->state = newstate;
943 * remove hrtimer, called with base lock held
946 remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base)
948 if (hrtimer_is_queued(timer)) {
953 * Remove the timer and force reprogramming when high
954 * resolution mode is active and the timer is on the current
955 * CPU. If we remove a timer on another CPU, reprogramming is
956 * skipped. The interrupt event on this CPU is fired and
957 * reprogramming happens in the interrupt handler. This is a
958 * rare case and less expensive than a smp call.
960 debug_deactivate(timer);
961 timer_stats_hrtimer_clear_start_info(timer);
962 reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases);
964 * We must preserve the CALLBACK state flag here,
965 * otherwise we could move the timer base in
966 * switch_hrtimer_base.
968 state = timer->state & HRTIMER_STATE_CALLBACK;
969 __remove_hrtimer(timer, base, state, reprogram);
975 int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
976 unsigned long delta_ns, const enum hrtimer_mode mode,
979 struct hrtimer_clock_base *base, *new_base;
983 base = lock_hrtimer_base(timer, &flags);
985 /* Remove an active timer from the queue: */
986 ret = remove_hrtimer(timer, base);
988 /* Switch the timer base, if necessary: */
989 new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED);
991 if (mode & HRTIMER_MODE_REL) {
992 tim = ktime_add_safe(tim, new_base->get_time());
994 * CONFIG_TIME_LOW_RES is a temporary way for architectures
995 * to signal that they simply return xtime in
996 * do_gettimeoffset(). In this case we want to round up by
997 * resolution when starting a relative timer, to avoid short
998 * timeouts. This will go away with the GTOD framework.
1000 #ifdef CONFIG_TIME_LOW_RES
1001 tim = ktime_add_safe(tim, base->resolution);
1005 hrtimer_set_expires_range_ns(timer, tim, delta_ns);
1007 timer_stats_hrtimer_set_start_info(timer);
1009 leftmost = enqueue_hrtimer(timer, new_base);
1012 * Only allow reprogramming if the new base is on this CPU.
1013 * (it might still be on another CPU if the timer was pending)
1015 * XXX send_remote_softirq() ?
1017 if (leftmost && new_base->cpu_base == &__get_cpu_var(hrtimer_bases)
1018 && hrtimer_enqueue_reprogram(timer, new_base)) {
1021 * We need to drop cpu_base->lock to avoid a
1022 * lock ordering issue vs. rq->lock.
1024 raw_spin_unlock(&new_base->cpu_base->lock);
1025 raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1026 local_irq_restore(flags);
1029 __raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1033 unlock_hrtimer_base(timer, &flags);
1039 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
1040 * @timer: the timer to be added
1042 * @delta_ns: "slack" range for the timer
1043 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
1044 * relative (HRTIMER_MODE_REL)
1048 * 1 when the timer was active
1050 int hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1051 unsigned long delta_ns, const enum hrtimer_mode mode)
1053 return __hrtimer_start_range_ns(timer, tim, delta_ns, mode, 1);
1055 EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
1058 * hrtimer_start - (re)start an hrtimer on the current CPU
1059 * @timer: the timer to be added
1061 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
1062 * relative (HRTIMER_MODE_REL)
1066 * 1 when the timer was active
1069 hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode)
1071 return __hrtimer_start_range_ns(timer, tim, 0, mode, 1);
1073 EXPORT_SYMBOL_GPL(hrtimer_start);
1077 * hrtimer_try_to_cancel - try to deactivate a timer
1078 * @timer: hrtimer to stop
1081 * 0 when the timer was not active
1082 * 1 when the timer was active
1083 * -1 when the timer is currently excuting the callback function and
1086 int hrtimer_try_to_cancel(struct hrtimer *timer)
1088 struct hrtimer_clock_base *base;
1089 unsigned long flags;
1092 base = lock_hrtimer_base(timer, &flags);
1094 if (!hrtimer_callback_running(timer))
1095 ret = remove_hrtimer(timer, base);
1097 unlock_hrtimer_base(timer, &flags);
1102 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1105 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1106 * @timer: the timer to be cancelled
1109 * 0 when the timer was not active
1110 * 1 when the timer was active
1112 int hrtimer_cancel(struct hrtimer *timer)
1115 int ret = hrtimer_try_to_cancel(timer);
1122 EXPORT_SYMBOL_GPL(hrtimer_cancel);
1125 * hrtimer_get_remaining - get remaining time for the timer
1126 * @timer: the timer to read
1128 ktime_t hrtimer_get_remaining(const struct hrtimer *timer)
1130 unsigned long flags;
1133 lock_hrtimer_base(timer, &flags);
1134 rem = hrtimer_expires_remaining(timer);
1135 unlock_hrtimer_base(timer, &flags);
1139 EXPORT_SYMBOL_GPL(hrtimer_get_remaining);
1141 #ifdef CONFIG_NO_HZ_COMMON
1143 * hrtimer_get_next_event - get the time until next expiry event
1145 * Returns the delta to the next expiry event or KTIME_MAX if no timer
1148 ktime_t hrtimer_get_next_event(void)
1150 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1151 struct hrtimer_clock_base *base = cpu_base->clock_base;
1152 ktime_t delta, mindelta = { .tv64 = KTIME_MAX };
1153 unsigned long flags;
1156 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1158 if (!hrtimer_hres_active()) {
1159 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
1160 struct hrtimer *timer;
1161 struct timerqueue_node *next;
1163 next = timerqueue_getnext(&base->active);
1167 timer = container_of(next, struct hrtimer, node);
1168 delta.tv64 = hrtimer_get_expires_tv64(timer);
1169 delta = ktime_sub(delta, base->get_time());
1170 if (delta.tv64 < mindelta.tv64)
1171 mindelta.tv64 = delta.tv64;
1175 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1177 if (mindelta.tv64 < 0)
1183 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1184 enum hrtimer_mode mode)
1186 struct hrtimer_cpu_base *cpu_base;
1189 memset(timer, 0, sizeof(struct hrtimer));
1191 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1193 if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
1194 clock_id = CLOCK_MONOTONIC;
1196 base = hrtimer_clockid_to_base(clock_id);
1197 timer->base = &cpu_base->clock_base[base];
1198 timerqueue_init(&timer->node);
1200 #ifdef CONFIG_TIMER_STATS
1201 timer->start_site = NULL;
1202 timer->start_pid = -1;
1203 memset(timer->start_comm, 0, TASK_COMM_LEN);
1208 * hrtimer_init - initialize a timer to the given clock
1209 * @timer: the timer to be initialized
1210 * @clock_id: the clock to be used
1211 * @mode: timer mode abs/rel
1213 void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1214 enum hrtimer_mode mode)
1216 debug_init(timer, clock_id, mode);
1217 __hrtimer_init(timer, clock_id, mode);
1219 EXPORT_SYMBOL_GPL(hrtimer_init);
1222 * hrtimer_get_res - get the timer resolution for a clock
1223 * @which_clock: which clock to query
1224 * @tp: pointer to timespec variable to store the resolution
1226 * Store the resolution of the clock selected by @which_clock in the
1227 * variable pointed to by @tp.
1229 int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp)
1231 struct hrtimer_cpu_base *cpu_base;
1232 int base = hrtimer_clockid_to_base(which_clock);
1234 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1235 *tp = ktime_to_timespec(cpu_base->clock_base[base].resolution);
1239 EXPORT_SYMBOL_GPL(hrtimer_get_res);
1241 static void __run_hrtimer(struct hrtimer *timer, ktime_t *now)
1243 struct hrtimer_clock_base *base = timer->base;
1244 struct hrtimer_cpu_base *cpu_base = base->cpu_base;
1245 enum hrtimer_restart (*fn)(struct hrtimer *);
1248 WARN_ON(!irqs_disabled());
1250 debug_deactivate(timer);
1251 __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0);
1252 timer_stats_account_hrtimer(timer);
1253 fn = timer->function;
1256 * Because we run timers from hardirq context, there is no chance
1257 * they get migrated to another cpu, therefore its safe to unlock
1260 raw_spin_unlock(&cpu_base->lock);
1261 trace_hrtimer_expire_entry(timer, now);
1262 restart = fn(timer);
1263 trace_hrtimer_expire_exit(timer);
1264 raw_spin_lock(&cpu_base->lock);
1267 * Note: We clear the CALLBACK bit after enqueue_hrtimer and
1268 * we do not reprogramm the event hardware. Happens either in
1269 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1271 if (restart != HRTIMER_NORESTART) {
1272 BUG_ON(timer->state != HRTIMER_STATE_CALLBACK);
1273 enqueue_hrtimer(timer, base);
1276 WARN_ON_ONCE(!(timer->state & HRTIMER_STATE_CALLBACK));
1278 timer->state &= ~HRTIMER_STATE_CALLBACK;
1281 #ifdef CONFIG_HIGH_RES_TIMERS
1284 * High resolution timer interrupt
1285 * Called with interrupts disabled
1287 void hrtimer_interrupt(struct clock_event_device *dev)
1289 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1290 ktime_t expires_next, now, entry_time, delta;
1293 BUG_ON(!cpu_base->hres_active);
1294 cpu_base->nr_events++;
1295 dev->next_event.tv64 = KTIME_MAX;
1297 raw_spin_lock(&cpu_base->lock);
1298 entry_time = now = hrtimer_update_base(cpu_base);
1300 expires_next.tv64 = KTIME_MAX;
1302 * We set expires_next to KTIME_MAX here with cpu_base->lock
1303 * held to prevent that a timer is enqueued in our queue via
1304 * the migration code. This does not affect enqueueing of
1305 * timers which run their callback and need to be requeued on
1308 cpu_base->expires_next.tv64 = KTIME_MAX;
1310 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1311 struct hrtimer_clock_base *base;
1312 struct timerqueue_node *node;
1315 if (!(cpu_base->active_bases & (1 << i)))
1318 base = cpu_base->clock_base + i;
1319 basenow = ktime_add(now, base->offset);
1321 while ((node = timerqueue_getnext(&base->active))) {
1322 struct hrtimer *timer;
1324 timer = container_of(node, struct hrtimer, node);
1327 * The immediate goal for using the softexpires is
1328 * minimizing wakeups, not running timers at the
1329 * earliest interrupt after their soft expiration.
1330 * This allows us to avoid using a Priority Search
1331 * Tree, which can answer a stabbing querry for
1332 * overlapping intervals and instead use the simple
1333 * BST we already have.
1334 * We don't add extra wakeups by delaying timers that
1335 * are right-of a not yet expired timer, because that
1336 * timer will have to trigger a wakeup anyway.
1339 if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer)) {
1342 expires = ktime_sub(hrtimer_get_expires(timer),
1344 if (expires.tv64 < 0)
1345 expires.tv64 = KTIME_MAX;
1346 if (expires.tv64 < expires_next.tv64)
1347 expires_next = expires;
1351 __run_hrtimer(timer, &basenow);
1356 * Store the new expiry value so the migration code can verify
1359 cpu_base->expires_next = expires_next;
1360 raw_spin_unlock(&cpu_base->lock);
1362 /* Reprogramming necessary ? */
1363 if (expires_next.tv64 == KTIME_MAX ||
1364 !tick_program_event(expires_next, 0)) {
1365 cpu_base->hang_detected = 0;
1370 * The next timer was already expired due to:
1372 * - long lasting callbacks
1373 * - being scheduled away when running in a VM
1375 * We need to prevent that we loop forever in the hrtimer
1376 * interrupt routine. We give it 3 attempts to avoid
1377 * overreacting on some spurious event.
1379 * Acquire base lock for updating the offsets and retrieving
1382 raw_spin_lock(&cpu_base->lock);
1383 now = hrtimer_update_base(cpu_base);
1384 cpu_base->nr_retries++;
1388 * Give the system a chance to do something else than looping
1389 * here. We stored the entry time, so we know exactly how long
1390 * we spent here. We schedule the next event this amount of
1393 cpu_base->nr_hangs++;
1394 cpu_base->hang_detected = 1;
1395 raw_spin_unlock(&cpu_base->lock);
1396 delta = ktime_sub(now, entry_time);
1397 if (delta.tv64 > cpu_base->max_hang_time.tv64)
1398 cpu_base->max_hang_time = delta;
1400 * Limit it to a sensible value as we enforce a longer
1401 * delay. Give the CPU at least 100ms to catch up.
1403 if (delta.tv64 > 100 * NSEC_PER_MSEC)
1404 expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1406 expires_next = ktime_add(now, delta);
1407 tick_program_event(expires_next, 1);
1408 printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n",
1409 ktime_to_ns(delta));
1413 * local version of hrtimer_peek_ahead_timers() called with interrupts
1416 static void __hrtimer_peek_ahead_timers(void)
1418 struct tick_device *td;
1420 if (!hrtimer_hres_active())
1423 td = &__get_cpu_var(tick_cpu_device);
1424 if (td && td->evtdev)
1425 hrtimer_interrupt(td->evtdev);
1429 * hrtimer_peek_ahead_timers -- run soft-expired timers now
1431 * hrtimer_peek_ahead_timers will peek at the timer queue of
1432 * the current cpu and check if there are any timers for which
1433 * the soft expires time has passed. If any such timers exist,
1434 * they are run immediately and then removed from the timer queue.
1437 void hrtimer_peek_ahead_timers(void)
1439 unsigned long flags;
1441 local_irq_save(flags);
1442 __hrtimer_peek_ahead_timers();
1443 local_irq_restore(flags);
1446 static void run_hrtimer_softirq(struct softirq_action *h)
1448 hrtimer_peek_ahead_timers();
1451 #else /* CONFIG_HIGH_RES_TIMERS */
1453 static inline void __hrtimer_peek_ahead_timers(void) { }
1455 #endif /* !CONFIG_HIGH_RES_TIMERS */
1458 * Called from timer softirq every jiffy, expire hrtimers:
1460 * For HRT its the fall back code to run the softirq in the timer
1461 * softirq context in case the hrtimer initialization failed or has
1462 * not been done yet.
1464 void hrtimer_run_pending(void)
1466 if (hrtimer_hres_active())
1470 * This _is_ ugly: We have to check in the softirq context,
1471 * whether we can switch to highres and / or nohz mode. The
1472 * clocksource switch happens in the timer interrupt with
1473 * xtime_lock held. Notification from there only sets the
1474 * check bit in the tick_oneshot code, otherwise we might
1475 * deadlock vs. xtime_lock.
1477 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1478 hrtimer_switch_to_hres();
1482 * Called from hardirq context every jiffy
1484 void hrtimer_run_queues(void)
1486 struct timerqueue_node *node;
1487 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1488 struct hrtimer_clock_base *base;
1489 int index, gettime = 1;
1491 if (hrtimer_hres_active())
1494 for (index = 0; index < HRTIMER_MAX_CLOCK_BASES; index++) {
1495 base = &cpu_base->clock_base[index];
1496 if (!timerqueue_getnext(&base->active))
1500 hrtimer_get_softirq_time(cpu_base);
1504 raw_spin_lock(&cpu_base->lock);
1506 while ((node = timerqueue_getnext(&base->active))) {
1507 struct hrtimer *timer;
1509 timer = container_of(node, struct hrtimer, node);
1510 if (base->softirq_time.tv64 <=
1511 hrtimer_get_expires_tv64(timer))
1514 __run_hrtimer(timer, &base->softirq_time);
1516 raw_spin_unlock(&cpu_base->lock);
1521 * Sleep related functions:
1523 static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1525 struct hrtimer_sleeper *t =
1526 container_of(timer, struct hrtimer_sleeper, timer);
1527 struct task_struct *task = t->task;
1531 wake_up_process(task);
1533 return HRTIMER_NORESTART;
1536 void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1538 sl->timer.function = hrtimer_wakeup;
1541 EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
1543 static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1545 hrtimer_init_sleeper(t, current);
1548 set_current_state(TASK_INTERRUPTIBLE);
1549 hrtimer_start_expires(&t->timer, mode);
1550 if (!hrtimer_active(&t->timer))
1553 if (likely(t->task))
1554 freezable_schedule();
1556 hrtimer_cancel(&t->timer);
1557 mode = HRTIMER_MODE_ABS;
1559 } while (t->task && !signal_pending(current));
1561 __set_current_state(TASK_RUNNING);
1563 return t->task == NULL;
1566 static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp)
1568 struct timespec rmt;
1571 rem = hrtimer_expires_remaining(timer);
1574 rmt = ktime_to_timespec(rem);
1576 if (copy_to_user(rmtp, &rmt, sizeof(*rmtp)))
1582 long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1584 struct hrtimer_sleeper t;
1585 struct timespec __user *rmtp;
1588 hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid,
1590 hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
1592 if (do_nanosleep(&t, HRTIMER_MODE_ABS))
1595 rmtp = restart->nanosleep.rmtp;
1597 ret = update_rmtp(&t.timer, rmtp);
1602 /* The other values in restart are already filled in */
1603 ret = -ERESTART_RESTARTBLOCK;
1605 destroy_hrtimer_on_stack(&t.timer);
1609 long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
1610 const enum hrtimer_mode mode, const clockid_t clockid)
1612 struct restart_block *restart;
1613 struct hrtimer_sleeper t;
1615 unsigned long slack;
1617 slack = current->timer_slack_ns;
1618 if (dl_task(current) || rt_task(current))
1621 hrtimer_init_on_stack(&t.timer, clockid, mode);
1622 hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack);
1623 if (do_nanosleep(&t, mode))
1626 /* Absolute timers do not update the rmtp value and restart: */
1627 if (mode == HRTIMER_MODE_ABS) {
1628 ret = -ERESTARTNOHAND;
1633 ret = update_rmtp(&t.timer, rmtp);
1638 restart = ¤t_thread_info()->restart_block;
1639 restart->fn = hrtimer_nanosleep_restart;
1640 restart->nanosleep.clockid = t.timer.base->clockid;
1641 restart->nanosleep.rmtp = rmtp;
1642 restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
1644 ret = -ERESTART_RESTARTBLOCK;
1646 destroy_hrtimer_on_stack(&t.timer);
1650 SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp,
1651 struct timespec __user *, rmtp)
1655 if (copy_from_user(&tu, rqtp, sizeof(tu)))
1658 if (!timespec_valid(&tu))
1661 return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1665 * Functions related to boot-time initialization:
1667 static void init_hrtimers_cpu(int cpu)
1669 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1672 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1673 cpu_base->clock_base[i].cpu_base = cpu_base;
1674 timerqueue_init_head(&cpu_base->clock_base[i].active);
1677 hrtimer_init_hres(cpu_base);
1680 #ifdef CONFIG_HOTPLUG_CPU
1682 static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1683 struct hrtimer_clock_base *new_base)
1685 struct hrtimer *timer;
1686 struct timerqueue_node *node;
1688 while ((node = timerqueue_getnext(&old_base->active))) {
1689 timer = container_of(node, struct hrtimer, node);
1690 BUG_ON(hrtimer_callback_running(timer));
1691 debug_deactivate(timer);
1694 * Mark it as STATE_MIGRATE not INACTIVE otherwise the
1695 * timer could be seen as !active and just vanish away
1696 * under us on another CPU
1698 __remove_hrtimer(timer, old_base, HRTIMER_STATE_MIGRATE, 0);
1699 timer->base = new_base;
1701 * Enqueue the timers on the new cpu. This does not
1702 * reprogram the event device in case the timer
1703 * expires before the earliest on this CPU, but we run
1704 * hrtimer_interrupt after we migrated everything to
1705 * sort out already expired timers and reprogram the
1708 enqueue_hrtimer(timer, new_base);
1710 /* Clear the migration state bit */
1711 timer->state &= ~HRTIMER_STATE_MIGRATE;
1715 static void migrate_hrtimers(int scpu)
1717 struct hrtimer_cpu_base *old_base, *new_base;
1720 BUG_ON(cpu_online(scpu));
1721 tick_cancel_sched_timer(scpu);
1723 local_irq_disable();
1724 old_base = &per_cpu(hrtimer_bases, scpu);
1725 new_base = &__get_cpu_var(hrtimer_bases);
1727 * The caller is globally serialized and nobody else
1728 * takes two locks at once, deadlock is not possible.
1730 raw_spin_lock(&new_base->lock);
1731 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1733 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1734 migrate_hrtimer_list(&old_base->clock_base[i],
1735 &new_base->clock_base[i]);
1738 raw_spin_unlock(&old_base->lock);
1739 raw_spin_unlock(&new_base->lock);
1741 /* Check, if we got expired work to do */
1742 __hrtimer_peek_ahead_timers();
1746 #endif /* CONFIG_HOTPLUG_CPU */
1748 static int hrtimer_cpu_notify(struct notifier_block *self,
1749 unsigned long action, void *hcpu)
1751 int scpu = (long)hcpu;
1755 case CPU_UP_PREPARE:
1756 case CPU_UP_PREPARE_FROZEN:
1757 init_hrtimers_cpu(scpu);
1760 #ifdef CONFIG_HOTPLUG_CPU
1762 case CPU_DYING_FROZEN:
1763 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DYING, &scpu);
1766 case CPU_DEAD_FROZEN:
1768 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &scpu);
1769 migrate_hrtimers(scpu);
1781 static struct notifier_block hrtimers_nb = {
1782 .notifier_call = hrtimer_cpu_notify,
1785 void __init hrtimers_init(void)
1787 hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
1788 (void *)(long)smp_processor_id());
1789 register_cpu_notifier(&hrtimers_nb);
1790 #ifdef CONFIG_HIGH_RES_TIMERS
1791 open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq);
1796 * schedule_hrtimeout_range_clock - sleep until timeout
1797 * @expires: timeout value (ktime_t)
1798 * @delta: slack in expires timeout (ktime_t)
1799 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1800 * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME
1803 schedule_hrtimeout_range_clock(ktime_t *expires, unsigned long delta,
1804 const enum hrtimer_mode mode, int clock)
1806 struct hrtimer_sleeper t;
1809 * Optimize when a zero timeout value is given. It does not
1810 * matter whether this is an absolute or a relative time.
1812 if (expires && !expires->tv64) {
1813 __set_current_state(TASK_RUNNING);
1818 * A NULL parameter means "infinite"
1822 __set_current_state(TASK_RUNNING);
1826 hrtimer_init_on_stack(&t.timer, clock, mode);
1827 hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
1829 hrtimer_init_sleeper(&t, current);
1831 hrtimer_start_expires(&t.timer, mode);
1832 if (!hrtimer_active(&t.timer))
1838 hrtimer_cancel(&t.timer);
1839 destroy_hrtimer_on_stack(&t.timer);
1841 __set_current_state(TASK_RUNNING);
1843 return !t.task ? 0 : -EINTR;
1847 * schedule_hrtimeout_range - sleep until timeout
1848 * @expires: timeout value (ktime_t)
1849 * @delta: slack in expires timeout (ktime_t)
1850 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1852 * Make the current task sleep until the given expiry time has
1853 * elapsed. The routine will return immediately unless
1854 * the current task state has been set (see set_current_state()).
1856 * The @delta argument gives the kernel the freedom to schedule the
1857 * actual wakeup to a time that is both power and performance friendly.
1858 * The kernel give the normal best effort behavior for "@expires+@delta",
1859 * but may decide to fire the timer earlier, but no earlier than @expires.
1861 * You can set the task state as follows -
1863 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1864 * pass before the routine returns.
1866 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1867 * delivered to the current task.
1869 * The current task state is guaranteed to be TASK_RUNNING when this
1872 * Returns 0 when the timer has expired otherwise -EINTR
1874 int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta,
1875 const enum hrtimer_mode mode)
1877 return schedule_hrtimeout_range_clock(expires, delta, mode,
1880 EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
1883 * schedule_hrtimeout - sleep until timeout
1884 * @expires: timeout value (ktime_t)
1885 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1887 * Make the current task sleep until the given expiry time has
1888 * elapsed. The routine will return immediately unless
1889 * the current task state has been set (see set_current_state()).
1891 * You can set the task state as follows -
1893 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1894 * pass before the routine returns.
1896 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1897 * delivered to the current task.
1899 * The current task state is guaranteed to be TASK_RUNNING when this
1902 * Returns 0 when the timer has expired otherwise -EINTR
1904 int __sched schedule_hrtimeout(ktime_t *expires,
1905 const enum hrtimer_mode mode)
1907 return schedule_hrtimeout_range(expires, 0, mode);
1909 EXPORT_SYMBOL_GPL(schedule_hrtimeout);