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
7 * Copyright (c) 2014, NVIDIA CORPORATION. All rights reserved.
9 * High-resolution kernel timers
11 * In contrast to the low-resolution timeout API implemented in
12 * kernel/timer.c, hrtimers provide finer resolution and accuracy
13 * depending on system configuration and capabilities.
15 * These timers are currently used for:
19 * - precise in-kernel timing
21 * Started by: Thomas Gleixner and Ingo Molnar
24 * based on kernel/timer.c
26 * Help, testing, suggestions, bugfixes, improvements were
29 * George Anzinger, Andrew Morton, Steven Rostedt, Roman Zippel
32 * For licencing details see kernel-base/COPYING
35 #include <linux/cpu.h>
36 #include <linux/export.h>
37 #include <linux/percpu.h>
38 #include <linux/hrtimer.h>
39 #include <linux/notifier.h>
40 #include <linux/syscalls.h>
41 #include <linux/kallsyms.h>
42 #include <linux/interrupt.h>
43 #include <linux/tick.h>
44 #include <linux/seq_file.h>
45 #include <linux/err.h>
46 #include <linux/debugobjects.h>
47 #include <linux/sched.h>
48 #include <linux/sched/sysctl.h>
49 #include <linux/sched/rt.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);
173 * Get the preferred target CPU for NOHZ
175 static int hrtimer_get_target(int this_cpu, int pinned)
177 #ifdef CONFIG_NO_HZ_COMMON
178 if (!pinned && get_sysctl_timer_migration() && idle_cpu(this_cpu))
179 return get_nohz_timer_target();
185 * With HIGHRES=y we do not migrate the timer when it is expiring
186 * before the next event on the target cpu because we cannot reprogram
187 * the target cpu hardware and we would cause it to fire late.
189 * Called with cpu_base->lock of target cpu held.
192 hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base)
194 #ifdef CONFIG_HIGH_RES_TIMERS
197 if (!new_base->cpu_base->hres_active)
200 expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset);
201 return expires.tv64 <= new_base->cpu_base->expires_next.tv64;
208 * Switch the timer base to the current CPU when possible.
210 static inline struct hrtimer_clock_base *
211 switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
214 struct hrtimer_clock_base *new_base;
215 struct hrtimer_cpu_base *new_cpu_base;
216 int this_cpu = smp_processor_id();
217 int cpu = hrtimer_get_target(this_cpu, pinned);
218 int basenum = base->index;
221 new_cpu_base = &per_cpu(hrtimer_bases, cpu);
222 new_base = &new_cpu_base->clock_base[basenum];
224 if (base != new_base) {
226 * We are trying to move timer to new_base.
227 * However we can't change timer's base while it is running,
228 * so we keep it on the same CPU. No hassle vs. reprogramming
229 * the event source in the high resolution case. The softirq
230 * code will take care of this when the timer function has
231 * completed. There is no conflict as we hold the lock until
232 * the timer is enqueued.
234 if (unlikely(hrtimer_callback_running(timer)))
237 /* See the comment in lock_timer_base() */
239 raw_spin_unlock(&base->cpu_base->lock);
240 raw_spin_lock(&new_base->cpu_base->lock);
242 if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) {
244 raw_spin_unlock(&new_base->cpu_base->lock);
245 raw_spin_lock(&base->cpu_base->lock);
249 timer->base = new_base;
251 if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) {
259 #else /* CONFIG_SMP */
261 static inline struct hrtimer_clock_base *
262 lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
264 struct hrtimer_clock_base *base = timer->base;
266 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
271 # define switch_hrtimer_base(t, b, p) (b)
273 #endif /* !CONFIG_SMP */
276 * Functions for the union type storage format of ktime_t which are
277 * too large for inlining:
279 #if BITS_PER_LONG < 64
280 # ifndef CONFIG_KTIME_SCALAR
282 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
284 * @nsec: the scalar nsec value to add
286 * Returns the sum of kt and nsec in ktime_t format
288 ktime_t ktime_add_ns(const ktime_t kt, u64 nsec)
292 if (likely(nsec < NSEC_PER_SEC)) {
295 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
297 /* Make sure nsec fits into long */
298 if (unlikely(nsec > KTIME_SEC_MAX))
299 return (ktime_t){ .tv64 = KTIME_MAX };
301 tmp = ktime_set((long)nsec, rem);
304 return ktime_add(kt, tmp);
307 EXPORT_SYMBOL_GPL(ktime_add_ns);
310 * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable
312 * @nsec: the scalar nsec value to subtract
314 * Returns the subtraction of @nsec from @kt in ktime_t format
316 ktime_t ktime_sub_ns(const ktime_t kt, u64 nsec)
320 if (likely(nsec < NSEC_PER_SEC)) {
323 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
325 tmp = ktime_set((long)nsec, rem);
328 return ktime_sub(kt, tmp);
331 EXPORT_SYMBOL_GPL(ktime_sub_ns);
332 # endif /* !CONFIG_KTIME_SCALAR */
335 * Divide a ktime value by a nanosecond value
337 u64 ktime_divns(const ktime_t kt, s64 div)
342 dclc = ktime_to_ns(kt);
343 /* Make sure the divisor is less than 2^32: */
349 do_div(dclc, (unsigned long) div);
353 #endif /* BITS_PER_LONG >= 64 */
356 * Add two ktime values and do a safety check for overflow:
358 ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
360 ktime_t res = ktime_add(lhs, rhs);
363 * We use KTIME_SEC_MAX here, the maximum timeout which we can
364 * return to user space in a timespec:
366 if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64)
367 res = ktime_set(KTIME_SEC_MAX, 0);
372 EXPORT_SYMBOL_GPL(ktime_add_safe);
374 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
376 static struct debug_obj_descr hrtimer_debug_descr;
378 static void *hrtimer_debug_hint(void *addr)
380 return ((struct hrtimer *) addr)->function;
384 * fixup_init is called when:
385 * - an active object is initialized
387 static int hrtimer_fixup_init(void *addr, enum debug_obj_state state)
389 struct hrtimer *timer = addr;
392 case ODEBUG_STATE_ACTIVE:
393 hrtimer_cancel(timer);
394 debug_object_init(timer, &hrtimer_debug_descr);
402 * fixup_activate is called when:
403 * - an active object is activated
404 * - an unknown object is activated (might be a statically initialized object)
406 static int hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
410 case ODEBUG_STATE_NOTAVAILABLE:
414 case ODEBUG_STATE_ACTIVE:
423 * fixup_free is called when:
424 * - an active object is freed
426 static int hrtimer_fixup_free(void *addr, enum debug_obj_state state)
428 struct hrtimer *timer = addr;
431 case ODEBUG_STATE_ACTIVE:
432 hrtimer_cancel(timer);
433 debug_object_free(timer, &hrtimer_debug_descr);
440 static struct debug_obj_descr hrtimer_debug_descr = {
442 .debug_hint = hrtimer_debug_hint,
443 .fixup_init = hrtimer_fixup_init,
444 .fixup_activate = hrtimer_fixup_activate,
445 .fixup_free = hrtimer_fixup_free,
448 static inline void debug_hrtimer_init(struct hrtimer *timer)
450 debug_object_init(timer, &hrtimer_debug_descr);
453 static inline void debug_hrtimer_activate(struct hrtimer *timer)
455 debug_object_activate(timer, &hrtimer_debug_descr);
458 static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
460 debug_object_deactivate(timer, &hrtimer_debug_descr);
463 static inline void debug_hrtimer_free(struct hrtimer *timer)
465 debug_object_free(timer, &hrtimer_debug_descr);
468 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
469 enum hrtimer_mode mode);
471 void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
472 enum hrtimer_mode mode)
474 debug_object_init_on_stack(timer, &hrtimer_debug_descr);
475 __hrtimer_init(timer, clock_id, mode);
477 EXPORT_SYMBOL_GPL(hrtimer_init_on_stack);
479 void destroy_hrtimer_on_stack(struct hrtimer *timer)
481 debug_object_free(timer, &hrtimer_debug_descr);
485 static inline void debug_hrtimer_init(struct hrtimer *timer) { }
486 static inline void debug_hrtimer_activate(struct hrtimer *timer) { }
487 static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
491 debug_init(struct hrtimer *timer, clockid_t clockid,
492 enum hrtimer_mode mode)
494 debug_hrtimer_init(timer);
495 trace_hrtimer_init(timer, clockid, mode);
498 static inline void debug_activate(struct hrtimer *timer)
500 debug_hrtimer_activate(timer);
501 trace_hrtimer_start(timer);
504 static inline void debug_deactivate(struct hrtimer *timer)
506 debug_hrtimer_deactivate(timer);
507 trace_hrtimer_cancel(timer);
510 /* High resolution timer related functions */
511 #ifdef CONFIG_HIGH_RES_TIMERS
514 * High resolution timer enabled ?
516 static int hrtimer_hres_enabled __read_mostly = 1;
519 * Enable / Disable high resolution mode
521 static int __init setup_hrtimer_hres(char *str)
523 if (!strcmp(str, "off"))
524 hrtimer_hres_enabled = 0;
525 else if (!strcmp(str, "on"))
526 hrtimer_hres_enabled = 1;
532 __setup("highres=", setup_hrtimer_hres);
535 * hrtimer_high_res_enabled - query, if the highres mode is enabled
537 static inline int hrtimer_is_hres_enabled(void)
539 return hrtimer_hres_enabled;
543 * Is the high resolution mode active ?
545 static inline int hrtimer_hres_active(void)
547 return __this_cpu_read(hrtimer_bases.hres_active);
551 * Reprogram the event source with checking both queues for the
553 * Called with interrupts disabled and base->lock held
556 hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
559 struct hrtimer_clock_base *base = cpu_base->clock_base;
560 ktime_t expires, expires_next;
562 expires_next.tv64 = KTIME_MAX;
564 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
565 struct hrtimer *timer;
566 struct timerqueue_node *next;
568 next = timerqueue_getnext(&base->active);
571 timer = container_of(next, struct hrtimer, node);
573 expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
575 * clock_was_set() has changed base->offset so the
576 * result might be negative. Fix it up to prevent a
577 * false positive in clockevents_program_event()
579 if (expires.tv64 < 0)
581 if (expires.tv64 < expires_next.tv64)
582 expires_next = expires;
585 if (skip_equal && expires_next.tv64 == cpu_base->expires_next.tv64)
588 cpu_base->expires_next.tv64 = expires_next.tv64;
591 * If a hang was detected in the last timer interrupt then we
592 * leave the hang delay active in the hardware. We want the
593 * system to make progress. That also prevents the following
595 * T1 expires 50ms from now
596 * T2 expires 5s from now
598 * T1 is removed, so this code is called and would reprogram
599 * the hardware to 5s from now. Any hrtimer_start after that
600 * will not reprogram the hardware due to hang_detected being
601 * set. So we'd effectivly block all timers until the T2 event
604 if (cpu_base->hang_detected)
607 if (cpu_base->expires_next.tv64 != KTIME_MAX)
608 tick_program_event(cpu_base->expires_next, 1);
612 * Shared reprogramming for clock_realtime and clock_monotonic
614 * When a timer is enqueued and expires earlier than the already enqueued
615 * timers, we have to check, whether it expires earlier than the timer for
616 * which the clock event device was armed.
618 * Called with interrupts disabled and base->cpu_base.lock held
620 static int hrtimer_reprogram(struct hrtimer *timer,
621 struct hrtimer_clock_base *base)
623 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
624 ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
627 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
630 * When the callback is running, we do not reprogram the clock event
631 * device. The timer callback is either running on a different CPU or
632 * the callback is executed in the hrtimer_interrupt context. The
633 * reprogramming is handled either by the softirq, which called the
634 * callback or at the end of the hrtimer_interrupt.
636 if (hrtimer_callback_running(timer))
640 * CLOCK_REALTIME timer might be requested with an absolute
641 * expiry time which is less than base->offset. Nothing wrong
642 * about that, just avoid to call into the tick code, which
643 * has now objections against negative expiry values.
645 if (expires.tv64 < 0)
648 if (expires.tv64 >= cpu_base->expires_next.tv64)
652 * If a hang was detected in the last timer interrupt then we
653 * do not schedule a timer which is earlier than the expiry
654 * which we enforced in the hang detection. We want the system
657 if (cpu_base->hang_detected)
661 * Clockevents returns -ETIME, when the event was in the past.
663 res = tick_program_event(expires, 0);
664 if (!IS_ERR_VALUE(res))
665 cpu_base->expires_next = expires;
670 * Initialize the high resolution related parts of cpu_base
672 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
674 base->expires_next.tv64 = KTIME_MAX;
675 base->hres_active = 0;
679 * When High resolution timers are active, try to reprogram. Note, that in case
680 * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry
681 * check happens. The timer gets enqueued into the rbtree. The reprogramming
682 * and expiry check is done in the hrtimer_interrupt or in the softirq.
684 static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
685 struct hrtimer_clock_base *base)
687 return base->cpu_base->hres_active && hrtimer_reprogram(timer, base);
690 static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base)
692 ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset;
693 ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset;
694 ktime_t *offs_tai = &base->clock_base[HRTIMER_BASE_TAI].offset;
696 return ktime_get_update_offsets(offs_real, offs_boot, offs_tai);
700 * Retrigger next event is called after clock was set
702 * Called with interrupts disabled via on_each_cpu()
704 static void retrigger_next_event(void *arg)
706 struct hrtimer_cpu_base *base = &__get_cpu_var(hrtimer_bases);
708 if (!hrtimer_hres_active())
711 raw_spin_lock(&base->lock);
712 hrtimer_update_base(base);
713 hrtimer_force_reprogram(base, 0);
714 raw_spin_unlock(&base->lock);
718 * Switch to high resolution mode
720 static int hrtimer_switch_to_hres(void)
722 int i, cpu = smp_processor_id();
723 struct hrtimer_cpu_base *base = &per_cpu(hrtimer_bases, cpu);
726 if (base->hres_active)
729 local_irq_save(flags);
731 if (tick_init_highres()) {
732 local_irq_restore(flags);
733 printk(KERN_WARNING "Could not switch to high resolution "
734 "mode on CPU %d\n", cpu);
737 base->hres_active = 1;
738 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
739 base->clock_base[i].resolution = KTIME_HIGH_RES;
741 tick_setup_sched_timer();
742 /* "Retrigger" the interrupt to get things going */
743 retrigger_next_event(NULL);
744 local_irq_restore(flags);
748 static void clock_was_set_work(struct work_struct *work)
753 static DECLARE_WORK(hrtimer_work, clock_was_set_work);
756 * Called from timekeeping and resume code to reprogramm the hrtimer
757 * interrupt device on all cpus.
759 void clock_was_set_delayed(void)
761 schedule_work(&hrtimer_work);
766 static inline int hrtimer_hres_active(void) { return 0; }
767 static inline int hrtimer_is_hres_enabled(void) { return 0; }
768 static inline int hrtimer_switch_to_hres(void) { return 0; }
770 hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { }
771 static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
772 struct hrtimer_clock_base *base)
776 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { }
777 static inline void retrigger_next_event(void *arg) { }
779 #endif /* CONFIG_HIGH_RES_TIMERS */
782 * Clock realtime was set
784 * Change the offset of the realtime clock vs. the monotonic
787 * We might have to reprogram the high resolution timer interrupt. On
788 * SMP we call the architecture specific code to retrigger _all_ high
789 * resolution timer interrupts. On UP we just disable interrupts and
790 * call the high resolution interrupt code.
792 void clock_was_set(void)
794 #ifdef CONFIG_HIGH_RES_TIMERS
795 /* Retrigger the CPU local events everywhere */
796 on_each_cpu(retrigger_next_event, NULL, 1);
798 timerfd_clock_was_set();
802 * During resume we might have to reprogram the high resolution timer
803 * interrupt (on the local CPU):
805 void hrtimers_resume(void)
807 WARN_ONCE(!irqs_disabled(),
808 KERN_INFO "hrtimers_resume() called with IRQs enabled!");
810 /* Retrigger on the local CPU */
811 retrigger_next_event(NULL);
812 /* And schedule a retrigger for all others */
813 clock_was_set_delayed();
816 static inline void timer_stats_hrtimer_set_start_info(struct hrtimer *timer)
818 #ifdef CONFIG_TIMER_STATS
819 if (timer->start_site)
821 timer->start_site = __builtin_return_address(0);
822 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
823 timer->start_pid = current->pid;
827 static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer *timer)
829 #ifdef CONFIG_TIMER_STATS
830 timer->start_site = NULL;
834 static inline void timer_stats_account_hrtimer(struct hrtimer *timer)
836 #ifdef CONFIG_TIMER_STATS
837 if (likely(!timer_stats_active))
839 timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
840 timer->function, timer->start_comm, 0);
845 * Counterpart to lock_hrtimer_base above:
848 void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
850 raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
854 * hrtimer_forward - forward the timer expiry
855 * @timer: hrtimer to forward
856 * @now: forward past this time
857 * @interval: the interval to forward
859 * Forward the timer expiry so it will expire in the future.
860 * Returns the number of overruns.
862 u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
867 delta = ktime_sub(now, hrtimer_get_expires(timer));
872 if (interval.tv64 < timer->base->resolution.tv64)
873 interval.tv64 = timer->base->resolution.tv64;
875 if (unlikely(delta.tv64 >= interval.tv64)) {
876 s64 incr = ktime_to_ns(interval);
878 orun = ktime_divns(delta, incr);
879 hrtimer_add_expires_ns(timer, incr * orun);
880 if (hrtimer_get_expires_tv64(timer) > now.tv64)
883 * This (and the ktime_add() below) is the
884 * correction for exact:
888 hrtimer_add_expires(timer, interval);
892 EXPORT_SYMBOL_GPL(hrtimer_forward);
895 * enqueue_hrtimer - internal function to (re)start a timer
897 * The timer is inserted in expiry order. Insertion into the
898 * red black tree is O(log(n)). Must hold the base lock.
900 * Returns 1 when the new timer is the leftmost timer in the tree.
902 static int enqueue_hrtimer(struct hrtimer *timer,
903 struct hrtimer_clock_base *base)
905 debug_activate(timer);
907 timerqueue_add(&base->active, &timer->node);
908 base->cpu_base->active_bases |= 1 << base->index;
911 * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
912 * state of a possibly running callback.
914 timer->state |= HRTIMER_STATE_ENQUEUED;
916 return (&timer->node == base->active.next);
920 * __remove_hrtimer - internal function to remove a timer
922 * Caller must hold the base lock.
924 * High resolution timer mode reprograms the clock event device when the
925 * timer is the one which expires next. The caller can disable this by setting
926 * reprogram to zero. This is useful, when the context does a reprogramming
927 * anyway (e.g. timer interrupt)
929 static void __remove_hrtimer(struct hrtimer *timer,
930 struct hrtimer_clock_base *base,
931 unsigned long newstate, int reprogram)
933 struct timerqueue_node *next_timer;
934 if (!(timer->state & HRTIMER_STATE_ENQUEUED))
937 next_timer = timerqueue_getnext(&base->active);
938 timerqueue_del(&base->active, &timer->node);
939 if (&timer->node == next_timer) {
940 #ifdef CONFIG_HIGH_RES_TIMERS
941 /* Reprogram the clock event device. if enabled */
942 if (reprogram && hrtimer_hres_active()) {
945 expires = ktime_sub(hrtimer_get_expires(timer),
947 if (base->cpu_base->expires_next.tv64 == expires.tv64)
948 hrtimer_force_reprogram(base->cpu_base, 1);
952 if (!timerqueue_getnext(&base->active))
953 base->cpu_base->active_bases &= ~(1 << base->index);
955 timer->state = newstate;
959 * remove hrtimer, called with base lock held
962 remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base)
964 if (hrtimer_is_queued(timer)) {
969 * Remove the timer and force reprogramming when high
970 * resolution mode is active and the timer is on the current
971 * CPU. If we remove a timer on another CPU, reprogramming is
972 * skipped. The interrupt event on this CPU is fired and
973 * reprogramming happens in the interrupt handler. This is a
974 * rare case and less expensive than a smp call.
976 debug_deactivate(timer);
977 timer_stats_hrtimer_clear_start_info(timer);
978 reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases);
980 * We must preserve the CALLBACK state flag here,
981 * otherwise we could move the timer base in
982 * switch_hrtimer_base.
984 state = timer->state & HRTIMER_STATE_CALLBACK;
985 __remove_hrtimer(timer, base, state, reprogram);
991 int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
992 unsigned long delta_ns, const enum hrtimer_mode mode,
995 struct hrtimer_clock_base *base, *new_base;
999 base = lock_hrtimer_base(timer, &flags);
1001 /* Remove an active timer from the queue: */
1002 ret = remove_hrtimer(timer, base);
1004 if (mode & HRTIMER_MODE_REL) {
1005 tim = ktime_add_safe(tim, base->get_time());
1007 * CONFIG_TIME_LOW_RES is a temporary way for architectures
1008 * to signal that they simply return xtime in
1009 * do_gettimeoffset(). In this case we want to round up by
1010 * resolution when starting a relative timer, to avoid short
1011 * timeouts. This will go away with the GTOD framework.
1013 #ifdef CONFIG_TIME_LOW_RES
1014 tim = ktime_add_safe(tim, base->resolution);
1018 hrtimer_set_expires_range_ns(timer, tim, delta_ns);
1020 /* Switch the timer base, if necessary: */
1021 new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED);
1023 timer_stats_hrtimer_set_start_info(timer);
1025 leftmost = enqueue_hrtimer(timer, new_base);
1028 * Only allow reprogramming if the new base is on this CPU.
1029 * (it might still be on another CPU if the timer was pending)
1031 * XXX send_remote_softirq() ?
1033 if (leftmost && new_base->cpu_base == &__get_cpu_var(hrtimer_bases)
1034 && hrtimer_enqueue_reprogram(timer, new_base)) {
1037 * We need to drop cpu_base->lock to avoid a
1038 * lock ordering issue vs. rq->lock.
1040 raw_spin_unlock(&new_base->cpu_base->lock);
1041 raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1042 local_irq_restore(flags);
1045 __raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1049 unlock_hrtimer_base(timer, &flags);
1055 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
1056 * @timer: the timer to be added
1058 * @delta_ns: "slack" range for the timer
1059 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
1060 * relative (HRTIMER_MODE_REL)
1064 * 1 when the timer was active
1066 int hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1067 unsigned long delta_ns, const enum hrtimer_mode mode)
1069 return __hrtimer_start_range_ns(timer, tim, delta_ns, mode, 1);
1071 EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
1074 * hrtimer_start - (re)start an hrtimer on the current CPU
1075 * @timer: the timer to be added
1077 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
1078 * relative (HRTIMER_MODE_REL)
1082 * 1 when the timer was active
1085 hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode)
1087 return __hrtimer_start_range_ns(timer, tim, 0, mode, 1);
1089 EXPORT_SYMBOL_GPL(hrtimer_start);
1093 * hrtimer_try_to_cancel - try to deactivate a timer
1094 * @timer: hrtimer to stop
1097 * 0 when the timer was not active
1098 * 1 when the timer was active
1099 * -1 when the timer is currently excuting the callback function and
1102 int hrtimer_try_to_cancel(struct hrtimer *timer)
1104 struct hrtimer_clock_base *base;
1105 unsigned long flags;
1108 base = lock_hrtimer_base(timer, &flags);
1110 if (!hrtimer_callback_running(timer))
1111 ret = remove_hrtimer(timer, base);
1113 unlock_hrtimer_base(timer, &flags);
1118 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1120 int hrtimer_try_to_cancel_relaxed(struct hrtimer *timer)
1122 struct hrtimer_clock_base *base;
1123 unsigned long flags;
1126 base = lock_hrtimer_base(timer, &flags);
1128 if (!hrtimer_callback_running_relaxed(timer))
1129 ret = remove_hrtimer(timer, base);
1131 unlock_hrtimer_base(timer, &flags);
1135 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel_relaxed);
1138 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1139 * @timer: the timer to be cancelled
1142 * 0 when the timer was not active
1143 * 1 when the timer was active
1145 int hrtimer_cancel(struct hrtimer *timer)
1148 int ret = hrtimer_try_to_cancel_relaxed(timer);
1154 * The corresponding ldax call for cpu_read_relax() is
1155 * present in hrtimer_callback_running_relaxed() as
1156 * hrtimer_try_to_cancel_relaxed() ultimately invokes
1157 * hrtimer_callback_running_relaxed().
1162 EXPORT_SYMBOL_GPL(hrtimer_cancel);
1165 * hrtimer_get_remaining - get remaining time for the timer
1166 * @timer: the timer to read
1168 ktime_t hrtimer_get_remaining(const struct hrtimer *timer)
1170 unsigned long flags;
1173 lock_hrtimer_base(timer, &flags);
1174 rem = hrtimer_expires_remaining(timer);
1175 unlock_hrtimer_base(timer, &flags);
1179 EXPORT_SYMBOL_GPL(hrtimer_get_remaining);
1181 #ifdef CONFIG_NO_HZ_COMMON
1183 * hrtimer_get_next_event - get the time until next expiry event
1185 * Returns the delta to the next expiry event or KTIME_MAX if no timer
1188 ktime_t hrtimer_get_next_event(void)
1190 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1191 struct hrtimer_clock_base *base = cpu_base->clock_base;
1192 ktime_t delta, mindelta = { .tv64 = KTIME_MAX };
1193 unsigned long flags;
1196 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1198 if (!hrtimer_hres_active()) {
1199 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
1200 struct hrtimer *timer;
1201 struct timerqueue_node *next;
1203 next = timerqueue_getnext(&base->active);
1207 timer = container_of(next, struct hrtimer, node);
1208 delta.tv64 = hrtimer_get_expires_tv64(timer);
1209 delta = ktime_sub(delta, base->get_time());
1210 if (delta.tv64 < mindelta.tv64)
1211 mindelta.tv64 = delta.tv64;
1215 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1217 if (mindelta.tv64 < 0)
1223 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1224 enum hrtimer_mode mode)
1226 struct hrtimer_cpu_base *cpu_base;
1229 memset(timer, 0, sizeof(struct hrtimer));
1231 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1233 if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
1234 clock_id = CLOCK_MONOTONIC;
1236 base = hrtimer_clockid_to_base(clock_id);
1237 timer->base = &cpu_base->clock_base[base];
1238 timerqueue_init(&timer->node);
1240 #ifdef CONFIG_TIMER_STATS
1241 timer->start_site = NULL;
1242 timer->start_pid = -1;
1243 memset(timer->start_comm, 0, TASK_COMM_LEN);
1248 * hrtimer_init - initialize a timer to the given clock
1249 * @timer: the timer to be initialized
1250 * @clock_id: the clock to be used
1251 * @mode: timer mode abs/rel
1253 void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1254 enum hrtimer_mode mode)
1256 debug_init(timer, clock_id, mode);
1257 __hrtimer_init(timer, clock_id, mode);
1259 EXPORT_SYMBOL_GPL(hrtimer_init);
1262 * hrtimer_get_res - get the timer resolution for a clock
1263 * @which_clock: which clock to query
1264 * @tp: pointer to timespec variable to store the resolution
1266 * Store the resolution of the clock selected by @which_clock in the
1267 * variable pointed to by @tp.
1269 int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp)
1271 struct hrtimer_cpu_base *cpu_base;
1272 int base = hrtimer_clockid_to_base(which_clock);
1274 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1275 *tp = ktime_to_timespec(cpu_base->clock_base[base].resolution);
1279 EXPORT_SYMBOL_GPL(hrtimer_get_res);
1281 static void __run_hrtimer(struct hrtimer *timer, ktime_t *now)
1283 struct hrtimer_clock_base *base = timer->base;
1284 struct hrtimer_cpu_base *cpu_base = base->cpu_base;
1285 enum hrtimer_restart (*fn)(struct hrtimer *);
1288 WARN_ON(!irqs_disabled());
1290 debug_deactivate(timer);
1291 __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0);
1292 timer_stats_account_hrtimer(timer);
1293 fn = timer->function;
1296 * Because we run timers from hardirq context, there is no chance
1297 * they get migrated to another cpu, therefore its safe to unlock
1300 raw_spin_unlock(&cpu_base->lock);
1301 trace_hrtimer_expire_entry(timer, now);
1302 restart = fn(timer);
1303 trace_hrtimer_expire_exit(timer);
1304 raw_spin_lock(&cpu_base->lock);
1307 * Note: We clear the CALLBACK bit after enqueue_hrtimer and
1308 * we do not reprogramm the event hardware. Happens either in
1309 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1311 if (restart != HRTIMER_NORESTART) {
1312 BUG_ON(timer->state != HRTIMER_STATE_CALLBACK);
1313 enqueue_hrtimer(timer, base);
1316 WARN_ON_ONCE(!(timer->state & HRTIMER_STATE_CALLBACK));
1318 timer->state &= ~HRTIMER_STATE_CALLBACK;
1321 #ifdef CONFIG_HIGH_RES_TIMERS
1324 * High resolution timer interrupt
1325 * Called with interrupts disabled
1327 void hrtimer_interrupt(struct clock_event_device *dev)
1329 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1330 ktime_t expires_next, now, entry_time, delta;
1333 BUG_ON(!cpu_base->hres_active);
1334 cpu_base->nr_events++;
1335 dev->next_event.tv64 = KTIME_MAX;
1337 raw_spin_lock(&cpu_base->lock);
1338 entry_time = now = hrtimer_update_base(cpu_base);
1340 expires_next.tv64 = KTIME_MAX;
1342 * We set expires_next to KTIME_MAX here with cpu_base->lock
1343 * held to prevent that a timer is enqueued in our queue via
1344 * the migration code. This does not affect enqueueing of
1345 * timers which run their callback and need to be requeued on
1348 cpu_base->expires_next.tv64 = KTIME_MAX;
1350 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1351 struct hrtimer_clock_base *base;
1352 struct timerqueue_node *node;
1355 if (!(cpu_base->active_bases & (1 << i)))
1358 base = cpu_base->clock_base + i;
1359 basenow = ktime_add(now, base->offset);
1361 while ((node = timerqueue_getnext(&base->active))) {
1362 struct hrtimer *timer;
1364 timer = container_of(node, struct hrtimer, node);
1367 * The immediate goal for using the softexpires is
1368 * minimizing wakeups, not running timers at the
1369 * earliest interrupt after their soft expiration.
1370 * This allows us to avoid using a Priority Search
1371 * Tree, which can answer a stabbing querry for
1372 * overlapping intervals and instead use the simple
1373 * BST we already have.
1374 * We don't add extra wakeups by delaying timers that
1375 * are right-of a not yet expired timer, because that
1376 * timer will have to trigger a wakeup anyway.
1379 if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer)) {
1382 expires = ktime_sub(hrtimer_get_expires(timer),
1384 if (expires.tv64 < 0)
1385 expires.tv64 = KTIME_MAX;
1386 if (expires.tv64 < expires_next.tv64)
1387 expires_next = expires;
1391 __run_hrtimer(timer, &basenow);
1396 * Store the new expiry value so the migration code can verify
1399 cpu_base->expires_next = expires_next;
1400 raw_spin_unlock(&cpu_base->lock);
1402 /* Reprogramming necessary ? */
1403 if (expires_next.tv64 == KTIME_MAX ||
1404 !tick_program_event(expires_next, 0)) {
1405 cpu_base->hang_detected = 0;
1410 * The next timer was already expired due to:
1412 * - long lasting callbacks
1413 * - being scheduled away when running in a VM
1415 * We need to prevent that we loop forever in the hrtimer
1416 * interrupt routine. We give it 3 attempts to avoid
1417 * overreacting on some spurious event.
1419 * Acquire base lock for updating the offsets and retrieving
1422 raw_spin_lock(&cpu_base->lock);
1423 now = hrtimer_update_base(cpu_base);
1424 cpu_base->nr_retries++;
1428 * Give the system a chance to do something else than looping
1429 * here. We stored the entry time, so we know exactly how long
1430 * we spent here. We schedule the next event this amount of
1433 cpu_base->nr_hangs++;
1434 cpu_base->hang_detected = 1;
1435 raw_spin_unlock(&cpu_base->lock);
1436 delta = ktime_sub(now, entry_time);
1437 if (delta.tv64 > cpu_base->max_hang_time.tv64)
1438 cpu_base->max_hang_time = delta;
1440 * Limit it to a sensible value as we enforce a longer
1441 * delay. Give the CPU at least 100ms to catch up.
1443 if (delta.tv64 > 100 * NSEC_PER_MSEC)
1444 expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1446 expires_next = ktime_add(now, delta);
1447 tick_program_event(expires_next, 1);
1448 printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n",
1449 ktime_to_ns(delta));
1453 * local version of hrtimer_peek_ahead_timers() called with interrupts
1456 static void __hrtimer_peek_ahead_timers(void)
1458 struct tick_device *td;
1460 if (!hrtimer_hres_active())
1463 td = &__get_cpu_var(tick_cpu_device);
1464 if (td && td->evtdev)
1465 hrtimer_interrupt(td->evtdev);
1469 * hrtimer_peek_ahead_timers -- run soft-expired timers now
1471 * hrtimer_peek_ahead_timers will peek at the timer queue of
1472 * the current cpu and check if there are any timers for which
1473 * the soft expires time has passed. If any such timers exist,
1474 * they are run immediately and then removed from the timer queue.
1477 void hrtimer_peek_ahead_timers(void)
1479 unsigned long flags;
1481 local_irq_save(flags);
1482 __hrtimer_peek_ahead_timers();
1483 local_irq_restore(flags);
1486 static void run_hrtimer_softirq(struct softirq_action *h)
1488 hrtimer_peek_ahead_timers();
1491 #else /* CONFIG_HIGH_RES_TIMERS */
1493 static inline void __hrtimer_peek_ahead_timers(void) { }
1495 #endif /* !CONFIG_HIGH_RES_TIMERS */
1498 * Called from timer softirq every jiffy, expire hrtimers:
1500 * For HRT its the fall back code to run the softirq in the timer
1501 * softirq context in case the hrtimer initialization failed or has
1502 * not been done yet.
1504 void hrtimer_run_pending(void)
1506 if (hrtimer_hres_active())
1510 * This _is_ ugly: We have to check in the softirq context,
1511 * whether we can switch to highres and / or nohz mode. The
1512 * clocksource switch happens in the timer interrupt with
1513 * xtime_lock held. Notification from there only sets the
1514 * check bit in the tick_oneshot code, otherwise we might
1515 * deadlock vs. xtime_lock.
1517 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1518 hrtimer_switch_to_hres();
1522 * Called from hardirq context every jiffy
1524 void hrtimer_run_queues(void)
1526 struct timerqueue_node *node;
1527 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1528 struct hrtimer_clock_base *base;
1529 int index, gettime = 1;
1531 if (hrtimer_hres_active())
1534 for (index = 0; index < HRTIMER_MAX_CLOCK_BASES; index++) {
1535 base = &cpu_base->clock_base[index];
1536 if (!timerqueue_getnext(&base->active))
1540 hrtimer_get_softirq_time(cpu_base);
1544 raw_spin_lock(&cpu_base->lock);
1546 while ((node = timerqueue_getnext(&base->active))) {
1547 struct hrtimer *timer;
1549 timer = container_of(node, struct hrtimer, node);
1550 if (base->softirq_time.tv64 <=
1551 hrtimer_get_expires_tv64(timer))
1554 __run_hrtimer(timer, &base->softirq_time);
1556 raw_spin_unlock(&cpu_base->lock);
1561 * Sleep related functions:
1563 static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1565 struct hrtimer_sleeper *t =
1566 container_of(timer, struct hrtimer_sleeper, timer);
1567 struct task_struct *task = t->task;
1571 wake_up_process(task);
1573 return HRTIMER_NORESTART;
1576 void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1578 sl->timer.function = hrtimer_wakeup;
1581 EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
1583 static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1585 hrtimer_init_sleeper(t, current);
1588 set_current_state(TASK_INTERRUPTIBLE);
1589 hrtimer_start_expires(&t->timer, mode);
1590 if (!hrtimer_active(&t->timer))
1593 if (likely(t->task))
1594 freezable_schedule();
1596 hrtimer_cancel(&t->timer);
1597 mode = HRTIMER_MODE_ABS;
1599 } while (t->task && !signal_pending(current));
1601 __set_current_state(TASK_RUNNING);
1603 return t->task == NULL;
1606 static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp)
1608 struct timespec rmt;
1611 rem = hrtimer_expires_remaining(timer);
1614 rmt = ktime_to_timespec(rem);
1616 if (copy_to_user(rmtp, &rmt, sizeof(*rmtp)))
1622 long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1624 struct hrtimer_sleeper t;
1625 struct timespec __user *rmtp;
1628 hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid,
1630 hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
1632 if (do_nanosleep(&t, HRTIMER_MODE_ABS))
1635 rmtp = restart->nanosleep.rmtp;
1637 ret = update_rmtp(&t.timer, rmtp);
1642 /* The other values in restart are already filled in */
1643 ret = -ERESTART_RESTARTBLOCK;
1645 destroy_hrtimer_on_stack(&t.timer);
1649 long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
1650 const enum hrtimer_mode mode, const clockid_t clockid)
1652 struct restart_block *restart;
1653 struct hrtimer_sleeper t;
1655 unsigned long slack;
1657 slack = current->timer_slack_ns;
1658 if (rt_task(current))
1661 hrtimer_init_on_stack(&t.timer, clockid, mode);
1662 hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack);
1663 if (do_nanosleep(&t, mode))
1666 /* Absolute timers do not update the rmtp value and restart: */
1667 if (mode == HRTIMER_MODE_ABS) {
1668 ret = -ERESTARTNOHAND;
1673 ret = update_rmtp(&t.timer, rmtp);
1678 restart = ¤t_thread_info()->restart_block;
1679 restart->fn = hrtimer_nanosleep_restart;
1680 restart->nanosleep.clockid = t.timer.base->clockid;
1681 restart->nanosleep.rmtp = rmtp;
1682 restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
1684 ret = -ERESTART_RESTARTBLOCK;
1686 destroy_hrtimer_on_stack(&t.timer);
1690 SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp,
1691 struct timespec __user *, rmtp)
1695 if (copy_from_user(&tu, rqtp, sizeof(tu)))
1698 if (!timespec_valid(&tu))
1701 return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1705 * Functions related to boot-time initialization:
1707 static void __cpuinit init_hrtimers_cpu(int cpu)
1709 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1712 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1713 cpu_base->clock_base[i].cpu_base = cpu_base;
1714 timerqueue_init_head(&cpu_base->clock_base[i].active);
1717 hrtimer_init_hres(cpu_base);
1720 #ifdef CONFIG_HOTPLUG_CPU
1722 static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1723 struct hrtimer_clock_base *new_base)
1725 struct hrtimer *timer;
1726 struct timerqueue_node *node;
1728 while ((node = timerqueue_getnext(&old_base->active))) {
1729 timer = container_of(node, struct hrtimer, node);
1730 BUG_ON(hrtimer_callback_running(timer));
1731 debug_deactivate(timer);
1734 * Mark it as STATE_MIGRATE not INACTIVE otherwise the
1735 * timer could be seen as !active and just vanish away
1736 * under us on another CPU
1738 __remove_hrtimer(timer, old_base, HRTIMER_STATE_MIGRATE, 0);
1739 timer->base = new_base;
1741 * Enqueue the timers on the new cpu. This does not
1742 * reprogram the event device in case the timer
1743 * expires before the earliest on this CPU, but we run
1744 * hrtimer_interrupt after we migrated everything to
1745 * sort out already expired timers and reprogram the
1748 enqueue_hrtimer(timer, new_base);
1750 /* Clear the migration state bit */
1751 timer->state &= ~HRTIMER_STATE_MIGRATE;
1755 static void migrate_hrtimers(int scpu)
1757 struct hrtimer_cpu_base *old_base, *new_base;
1760 BUG_ON(cpu_online(scpu));
1761 tick_cancel_sched_timer(scpu);
1763 local_irq_disable();
1764 old_base = &per_cpu(hrtimer_bases, scpu);
1765 new_base = &__get_cpu_var(hrtimer_bases);
1767 * The caller is globally serialized and nobody else
1768 * takes two locks at once, deadlock is not possible.
1770 raw_spin_lock(&new_base->lock);
1771 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1773 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1774 migrate_hrtimer_list(&old_base->clock_base[i],
1775 &new_base->clock_base[i]);
1778 raw_spin_unlock(&old_base->lock);
1779 raw_spin_unlock(&new_base->lock);
1781 /* Check, if we got expired work to do */
1782 __hrtimer_peek_ahead_timers();
1786 #endif /* CONFIG_HOTPLUG_CPU */
1788 static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self,
1789 unsigned long action, void *hcpu)
1791 int scpu = (long)hcpu;
1795 case CPU_UP_PREPARE:
1796 case CPU_UP_PREPARE_FROZEN:
1797 init_hrtimers_cpu(scpu);
1800 #ifdef CONFIG_HOTPLUG_CPU
1802 case CPU_DYING_FROZEN:
1803 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DYING, &scpu);
1806 case CPU_DEAD_FROZEN:
1808 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &scpu);
1809 migrate_hrtimers(scpu);
1821 static struct notifier_block __cpuinitdata hrtimers_nb = {
1822 .notifier_call = hrtimer_cpu_notify,
1825 void __init hrtimers_init(void)
1827 hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
1828 (void *)(long)smp_processor_id());
1829 register_cpu_notifier(&hrtimers_nb);
1830 #ifdef CONFIG_HIGH_RES_TIMERS
1831 open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq);
1836 * schedule_hrtimeout_range_clock - sleep until timeout
1837 * @expires: timeout value (ktime_t)
1838 * @delta: slack in expires timeout (ktime_t)
1839 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1840 * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME
1843 schedule_hrtimeout_range_clock(ktime_t *expires, unsigned long delta,
1844 const enum hrtimer_mode mode, int clock)
1846 struct hrtimer_sleeper t;
1849 * Optimize when a zero timeout value is given. It does not
1850 * matter whether this is an absolute or a relative time.
1852 if (expires && !expires->tv64) {
1853 __set_current_state(TASK_RUNNING);
1858 * A NULL parameter means "infinite"
1862 __set_current_state(TASK_RUNNING);
1866 hrtimer_init_on_stack(&t.timer, clock, mode);
1867 hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
1869 hrtimer_init_sleeper(&t, current);
1871 hrtimer_start_expires(&t.timer, mode);
1872 if (!hrtimer_active(&t.timer))
1878 hrtimer_cancel(&t.timer);
1879 destroy_hrtimer_on_stack(&t.timer);
1881 __set_current_state(TASK_RUNNING);
1883 return !t.task ? 0 : -EINTR;
1887 * schedule_hrtimeout_range - sleep until timeout
1888 * @expires: timeout value (ktime_t)
1889 * @delta: slack in expires timeout (ktime_t)
1890 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1892 * Make the current task sleep until the given expiry time has
1893 * elapsed. The routine will return immediately unless
1894 * the current task state has been set (see set_current_state()).
1896 * The @delta argument gives the kernel the freedom to schedule the
1897 * actual wakeup to a time that is both power and performance friendly.
1898 * The kernel give the normal best effort behavior for "@expires+@delta",
1899 * but may decide to fire the timer earlier, but no earlier than @expires.
1901 * You can set the task state as follows -
1903 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1904 * pass before the routine returns.
1906 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1907 * delivered to the current task.
1909 * The current task state is guaranteed to be TASK_RUNNING when this
1912 * Returns 0 when the timer has expired otherwise -EINTR
1914 int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta,
1915 const enum hrtimer_mode mode)
1917 return schedule_hrtimeout_range_clock(expires, delta, mode,
1920 EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
1923 * schedule_hrtimeout - sleep until timeout
1924 * @expires: timeout value (ktime_t)
1925 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1927 * Make the current task sleep until the given expiry time has
1928 * elapsed. The routine will return immediately unless
1929 * the current task state has been set (see set_current_state()).
1931 * You can set the task state as follows -
1933 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1934 * pass before the routine returns.
1936 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1937 * delivered to the current task.
1939 * The current task state is guaranteed to be TASK_RUNNING when this
1942 * Returns 0 when the timer has expired otherwise -EINTR
1944 int __sched schedule_hrtimeout(ktime_t *expires,
1945 const enum hrtimer_mode mode)
1947 return schedule_hrtimeout_range(expires, 0, mode);
1949 EXPORT_SYMBOL_GPL(schedule_hrtimeout);