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);
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 all online CPUs. However, all other CPUs will be
804 * stopped with IRQs interrupts disabled so the clock_was_set() call
807 void hrtimers_resume(void)
809 WARN_ONCE(!irqs_disabled(),
810 KERN_INFO "hrtimers_resume() called with IRQs enabled!");
812 /* Retrigger on the local CPU */
813 retrigger_next_event(NULL);
814 /* And schedule a retrigger for all others */
815 clock_was_set_delayed();
818 static inline void timer_stats_hrtimer_set_start_info(struct hrtimer *timer)
820 #ifdef CONFIG_TIMER_STATS
821 if (timer->start_site)
823 timer->start_site = __builtin_return_address(0);
824 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
825 timer->start_pid = current->pid;
829 static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer *timer)
831 #ifdef CONFIG_TIMER_STATS
832 timer->start_site = NULL;
836 static inline void timer_stats_account_hrtimer(struct hrtimer *timer)
838 #ifdef CONFIG_TIMER_STATS
839 if (likely(!timer_stats_active))
841 timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
842 timer->function, timer->start_comm, 0);
847 * Counterpart to lock_hrtimer_base above:
850 void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
852 raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
856 * hrtimer_forward - forward the timer expiry
857 * @timer: hrtimer to forward
858 * @now: forward past this time
859 * @interval: the interval to forward
861 * Forward the timer expiry so it will expire in the future.
862 * Returns the number of overruns.
864 u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
869 delta = ktime_sub(now, hrtimer_get_expires(timer));
874 if (interval.tv64 < timer->base->resolution.tv64)
875 interval.tv64 = timer->base->resolution.tv64;
877 if (unlikely(delta.tv64 >= interval.tv64)) {
878 s64 incr = ktime_to_ns(interval);
880 orun = ktime_divns(delta, incr);
881 hrtimer_add_expires_ns(timer, incr * orun);
882 if (hrtimer_get_expires_tv64(timer) > now.tv64)
885 * This (and the ktime_add() below) is the
886 * correction for exact:
890 hrtimer_add_expires(timer, interval);
894 EXPORT_SYMBOL_GPL(hrtimer_forward);
897 * enqueue_hrtimer - internal function to (re)start a timer
899 * The timer is inserted in expiry order. Insertion into the
900 * red black tree is O(log(n)). Must hold the base lock.
902 * Returns 1 when the new timer is the leftmost timer in the tree.
904 static int enqueue_hrtimer(struct hrtimer *timer,
905 struct hrtimer_clock_base *base)
907 debug_activate(timer);
909 timerqueue_add(&base->active, &timer->node);
910 base->cpu_base->active_bases |= 1 << base->index;
913 * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
914 * state of a possibly running callback.
916 timer->state |= HRTIMER_STATE_ENQUEUED;
918 return (&timer->node == base->active.next);
922 * __remove_hrtimer - internal function to remove a timer
924 * Caller must hold the base lock.
926 * High resolution timer mode reprograms the clock event device when the
927 * timer is the one which expires next. The caller can disable this by setting
928 * reprogram to zero. This is useful, when the context does a reprogramming
929 * anyway (e.g. timer interrupt)
931 static void __remove_hrtimer(struct hrtimer *timer,
932 struct hrtimer_clock_base *base,
933 unsigned long newstate, int reprogram)
935 struct timerqueue_node *next_timer;
936 if (!(timer->state & HRTIMER_STATE_ENQUEUED))
939 next_timer = timerqueue_getnext(&base->active);
940 timerqueue_del(&base->active, &timer->node);
941 if (&timer->node == next_timer) {
942 #ifdef CONFIG_HIGH_RES_TIMERS
943 /* Reprogram the clock event device. if enabled */
944 if (reprogram && hrtimer_hres_active()) {
947 expires = ktime_sub(hrtimer_get_expires(timer),
949 if (base->cpu_base->expires_next.tv64 == expires.tv64)
950 hrtimer_force_reprogram(base->cpu_base, 1);
954 if (!timerqueue_getnext(&base->active))
955 base->cpu_base->active_bases &= ~(1 << base->index);
957 timer->state = newstate;
961 * remove hrtimer, called with base lock held
964 remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base)
966 if (hrtimer_is_queued(timer)) {
971 * Remove the timer and force reprogramming when high
972 * resolution mode is active and the timer is on the current
973 * CPU. If we remove a timer on another CPU, reprogramming is
974 * skipped. The interrupt event on this CPU is fired and
975 * reprogramming happens in the interrupt handler. This is a
976 * rare case and less expensive than a smp call.
978 debug_deactivate(timer);
979 timer_stats_hrtimer_clear_start_info(timer);
980 reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases);
982 * We must preserve the CALLBACK state flag here,
983 * otherwise we could move the timer base in
984 * switch_hrtimer_base.
986 state = timer->state & HRTIMER_STATE_CALLBACK;
987 __remove_hrtimer(timer, base, state, reprogram);
993 int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
994 unsigned long delta_ns, const enum hrtimer_mode mode,
997 struct hrtimer_clock_base *base, *new_base;
1001 base = lock_hrtimer_base(timer, &flags);
1003 /* Remove an active timer from the queue: */
1004 ret = remove_hrtimer(timer, base);
1006 if (mode & HRTIMER_MODE_REL) {
1007 tim = ktime_add_safe(tim, base->get_time());
1009 * CONFIG_TIME_LOW_RES is a temporary way for architectures
1010 * to signal that they simply return xtime in
1011 * do_gettimeoffset(). In this case we want to round up by
1012 * resolution when starting a relative timer, to avoid short
1013 * timeouts. This will go away with the GTOD framework.
1015 #ifdef CONFIG_TIME_LOW_RES
1016 tim = ktime_add_safe(tim, base->resolution);
1020 hrtimer_set_expires_range_ns(timer, tim, delta_ns);
1022 /* Switch the timer base, if necessary: */
1023 new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED);
1025 timer_stats_hrtimer_set_start_info(timer);
1027 leftmost = enqueue_hrtimer(timer, new_base);
1030 * Only allow reprogramming if the new base is on this CPU.
1031 * (it might still be on another CPU if the timer was pending)
1033 * XXX send_remote_softirq() ?
1035 if (leftmost && new_base->cpu_base == &__get_cpu_var(hrtimer_bases)
1036 && hrtimer_enqueue_reprogram(timer, new_base)) {
1039 * We need to drop cpu_base->lock to avoid a
1040 * lock ordering issue vs. rq->lock.
1042 raw_spin_unlock(&new_base->cpu_base->lock);
1043 raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1044 local_irq_restore(flags);
1047 __raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1051 unlock_hrtimer_base(timer, &flags);
1057 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
1058 * @timer: the timer to be added
1060 * @delta_ns: "slack" range for the timer
1061 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
1062 * relative (HRTIMER_MODE_REL)
1066 * 1 when the timer was active
1068 int hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1069 unsigned long delta_ns, const enum hrtimer_mode mode)
1071 return __hrtimer_start_range_ns(timer, tim, delta_ns, mode, 1);
1073 EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
1076 * hrtimer_start - (re)start an hrtimer on the current CPU
1077 * @timer: the timer to be added
1079 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
1080 * relative (HRTIMER_MODE_REL)
1084 * 1 when the timer was active
1087 hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode)
1089 return __hrtimer_start_range_ns(timer, tim, 0, mode, 1);
1091 EXPORT_SYMBOL_GPL(hrtimer_start);
1095 * hrtimer_try_to_cancel - try to deactivate a timer
1096 * @timer: hrtimer to stop
1099 * 0 when the timer was not active
1100 * 1 when the timer was active
1101 * -1 when the timer is currently excuting the callback function and
1104 int hrtimer_try_to_cancel(struct hrtimer *timer)
1106 struct hrtimer_clock_base *base;
1107 unsigned long flags;
1110 base = lock_hrtimer_base(timer, &flags);
1112 if (!hrtimer_callback_running(timer))
1113 ret = remove_hrtimer(timer, base);
1115 unlock_hrtimer_base(timer, &flags);
1120 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1123 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1124 * @timer: the timer to be cancelled
1127 * 0 when the timer was not active
1128 * 1 when the timer was active
1130 int hrtimer_cancel(struct hrtimer *timer)
1133 int ret = hrtimer_try_to_cancel(timer);
1140 EXPORT_SYMBOL_GPL(hrtimer_cancel);
1143 * hrtimer_get_remaining - get remaining time for the timer
1144 * @timer: the timer to read
1146 ktime_t hrtimer_get_remaining(const struct hrtimer *timer)
1148 unsigned long flags;
1151 lock_hrtimer_base(timer, &flags);
1152 rem = hrtimer_expires_remaining(timer);
1153 unlock_hrtimer_base(timer, &flags);
1157 EXPORT_SYMBOL_GPL(hrtimer_get_remaining);
1159 #ifdef CONFIG_NO_HZ_COMMON
1161 * hrtimer_get_next_event - get the time until next expiry event
1163 * Returns the delta to the next expiry event or KTIME_MAX if no timer
1166 ktime_t hrtimer_get_next_event(void)
1168 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1169 struct hrtimer_clock_base *base = cpu_base->clock_base;
1170 ktime_t delta, mindelta = { .tv64 = KTIME_MAX };
1171 unsigned long flags;
1174 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1176 if (!hrtimer_hres_active()) {
1177 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
1178 struct hrtimer *timer;
1179 struct timerqueue_node *next;
1181 next = timerqueue_getnext(&base->active);
1185 timer = container_of(next, struct hrtimer, node);
1186 delta.tv64 = hrtimer_get_expires_tv64(timer);
1187 delta = ktime_sub(delta, base->get_time());
1188 if (delta.tv64 < mindelta.tv64)
1189 mindelta.tv64 = delta.tv64;
1193 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1195 if (mindelta.tv64 < 0)
1201 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1202 enum hrtimer_mode mode)
1204 struct hrtimer_cpu_base *cpu_base;
1207 memset(timer, 0, sizeof(struct hrtimer));
1209 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1211 if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
1212 clock_id = CLOCK_MONOTONIC;
1214 base = hrtimer_clockid_to_base(clock_id);
1215 timer->base = &cpu_base->clock_base[base];
1216 timerqueue_init(&timer->node);
1218 #ifdef CONFIG_TIMER_STATS
1219 timer->start_site = NULL;
1220 timer->start_pid = -1;
1221 memset(timer->start_comm, 0, TASK_COMM_LEN);
1226 * hrtimer_init - initialize a timer to the given clock
1227 * @timer: the timer to be initialized
1228 * @clock_id: the clock to be used
1229 * @mode: timer mode abs/rel
1231 void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1232 enum hrtimer_mode mode)
1234 debug_init(timer, clock_id, mode);
1235 __hrtimer_init(timer, clock_id, mode);
1237 EXPORT_SYMBOL_GPL(hrtimer_init);
1240 * hrtimer_get_res - get the timer resolution for a clock
1241 * @which_clock: which clock to query
1242 * @tp: pointer to timespec variable to store the resolution
1244 * Store the resolution of the clock selected by @which_clock in the
1245 * variable pointed to by @tp.
1247 int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp)
1249 struct hrtimer_cpu_base *cpu_base;
1250 int base = hrtimer_clockid_to_base(which_clock);
1252 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1253 *tp = ktime_to_timespec(cpu_base->clock_base[base].resolution);
1257 EXPORT_SYMBOL_GPL(hrtimer_get_res);
1259 static void __run_hrtimer(struct hrtimer *timer, ktime_t *now)
1261 struct hrtimer_clock_base *base = timer->base;
1262 struct hrtimer_cpu_base *cpu_base = base->cpu_base;
1263 enum hrtimer_restart (*fn)(struct hrtimer *);
1266 WARN_ON(!irqs_disabled());
1268 debug_deactivate(timer);
1269 __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0);
1270 timer_stats_account_hrtimer(timer);
1271 fn = timer->function;
1274 * Because we run timers from hardirq context, there is no chance
1275 * they get migrated to another cpu, therefore its safe to unlock
1278 raw_spin_unlock(&cpu_base->lock);
1279 trace_hrtimer_expire_entry(timer, now);
1280 restart = fn(timer);
1281 trace_hrtimer_expire_exit(timer);
1282 raw_spin_lock(&cpu_base->lock);
1285 * Note: We clear the CALLBACK bit after enqueue_hrtimer and
1286 * we do not reprogramm the event hardware. Happens either in
1287 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1289 if (restart != HRTIMER_NORESTART) {
1290 BUG_ON(timer->state != HRTIMER_STATE_CALLBACK);
1291 enqueue_hrtimer(timer, base);
1294 WARN_ON_ONCE(!(timer->state & HRTIMER_STATE_CALLBACK));
1296 timer->state &= ~HRTIMER_STATE_CALLBACK;
1299 #ifdef CONFIG_HIGH_RES_TIMERS
1302 * High resolution timer interrupt
1303 * Called with interrupts disabled
1305 void hrtimer_interrupt(struct clock_event_device *dev)
1307 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1308 ktime_t expires_next, now, entry_time, delta;
1311 BUG_ON(!cpu_base->hres_active);
1312 cpu_base->nr_events++;
1313 dev->next_event.tv64 = KTIME_MAX;
1315 raw_spin_lock(&cpu_base->lock);
1316 entry_time = now = hrtimer_update_base(cpu_base);
1318 expires_next.tv64 = KTIME_MAX;
1320 * We set expires_next to KTIME_MAX here with cpu_base->lock
1321 * held to prevent that a timer is enqueued in our queue via
1322 * the migration code. This does not affect enqueueing of
1323 * timers which run their callback and need to be requeued on
1326 cpu_base->expires_next.tv64 = KTIME_MAX;
1328 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1329 struct hrtimer_clock_base *base;
1330 struct timerqueue_node *node;
1333 if (!(cpu_base->active_bases & (1 << i)))
1336 base = cpu_base->clock_base + i;
1337 basenow = ktime_add(now, base->offset);
1339 while ((node = timerqueue_getnext(&base->active))) {
1340 struct hrtimer *timer;
1342 timer = container_of(node, struct hrtimer, node);
1345 * The immediate goal for using the softexpires is
1346 * minimizing wakeups, not running timers at the
1347 * earliest interrupt after their soft expiration.
1348 * This allows us to avoid using a Priority Search
1349 * Tree, which can answer a stabbing querry for
1350 * overlapping intervals and instead use the simple
1351 * BST we already have.
1352 * We don't add extra wakeups by delaying timers that
1353 * are right-of a not yet expired timer, because that
1354 * timer will have to trigger a wakeup anyway.
1357 if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer)) {
1360 expires = ktime_sub(hrtimer_get_expires(timer),
1362 if (expires.tv64 < 0)
1363 expires.tv64 = KTIME_MAX;
1364 if (expires.tv64 < expires_next.tv64)
1365 expires_next = expires;
1369 __run_hrtimer(timer, &basenow);
1374 * Store the new expiry value so the migration code can verify
1377 cpu_base->expires_next = expires_next;
1378 raw_spin_unlock(&cpu_base->lock);
1380 /* Reprogramming necessary ? */
1381 if (expires_next.tv64 == KTIME_MAX ||
1382 !tick_program_event(expires_next, 0)) {
1383 cpu_base->hang_detected = 0;
1388 * The next timer was already expired due to:
1390 * - long lasting callbacks
1391 * - being scheduled away when running in a VM
1393 * We need to prevent that we loop forever in the hrtimer
1394 * interrupt routine. We give it 3 attempts to avoid
1395 * overreacting on some spurious event.
1397 * Acquire base lock for updating the offsets and retrieving
1400 raw_spin_lock(&cpu_base->lock);
1401 now = hrtimer_update_base(cpu_base);
1402 cpu_base->nr_retries++;
1406 * Give the system a chance to do something else than looping
1407 * here. We stored the entry time, so we know exactly how long
1408 * we spent here. We schedule the next event this amount of
1411 cpu_base->nr_hangs++;
1412 cpu_base->hang_detected = 1;
1413 raw_spin_unlock(&cpu_base->lock);
1414 delta = ktime_sub(now, entry_time);
1415 if (delta.tv64 > cpu_base->max_hang_time.tv64)
1416 cpu_base->max_hang_time = delta;
1418 * Limit it to a sensible value as we enforce a longer
1419 * delay. Give the CPU at least 100ms to catch up.
1421 if (delta.tv64 > 100 * NSEC_PER_MSEC)
1422 expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1424 expires_next = ktime_add(now, delta);
1425 tick_program_event(expires_next, 1);
1426 printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n",
1427 ktime_to_ns(delta));
1431 * local version of hrtimer_peek_ahead_timers() called with interrupts
1434 static void __hrtimer_peek_ahead_timers(void)
1436 struct tick_device *td;
1438 if (!hrtimer_hres_active())
1441 td = &__get_cpu_var(tick_cpu_device);
1442 if (td && td->evtdev)
1443 hrtimer_interrupt(td->evtdev);
1447 * hrtimer_peek_ahead_timers -- run soft-expired timers now
1449 * hrtimer_peek_ahead_timers will peek at the timer queue of
1450 * the current cpu and check if there are any timers for which
1451 * the soft expires time has passed. If any such timers exist,
1452 * they are run immediately and then removed from the timer queue.
1455 void hrtimer_peek_ahead_timers(void)
1457 unsigned long flags;
1459 local_irq_save(flags);
1460 __hrtimer_peek_ahead_timers();
1461 local_irq_restore(flags);
1464 static void run_hrtimer_softirq(struct softirq_action *h)
1466 hrtimer_peek_ahead_timers();
1469 #else /* CONFIG_HIGH_RES_TIMERS */
1471 static inline void __hrtimer_peek_ahead_timers(void) { }
1473 #endif /* !CONFIG_HIGH_RES_TIMERS */
1476 * Called from timer softirq every jiffy, expire hrtimers:
1478 * For HRT its the fall back code to run the softirq in the timer
1479 * softirq context in case the hrtimer initialization failed or has
1480 * not been done yet.
1482 void hrtimer_run_pending(void)
1484 if (hrtimer_hres_active())
1488 * This _is_ ugly: We have to check in the softirq context,
1489 * whether we can switch to highres and / or nohz mode. The
1490 * clocksource switch happens in the timer interrupt with
1491 * xtime_lock held. Notification from there only sets the
1492 * check bit in the tick_oneshot code, otherwise we might
1493 * deadlock vs. xtime_lock.
1495 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1496 hrtimer_switch_to_hres();
1500 * Called from hardirq context every jiffy
1502 void hrtimer_run_queues(void)
1504 struct timerqueue_node *node;
1505 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1506 struct hrtimer_clock_base *base;
1507 int index, gettime = 1;
1509 if (hrtimer_hres_active())
1512 for (index = 0; index < HRTIMER_MAX_CLOCK_BASES; index++) {
1513 base = &cpu_base->clock_base[index];
1514 if (!timerqueue_getnext(&base->active))
1518 hrtimer_get_softirq_time(cpu_base);
1522 raw_spin_lock(&cpu_base->lock);
1524 while ((node = timerqueue_getnext(&base->active))) {
1525 struct hrtimer *timer;
1527 timer = container_of(node, struct hrtimer, node);
1528 if (base->softirq_time.tv64 <=
1529 hrtimer_get_expires_tv64(timer))
1532 __run_hrtimer(timer, &base->softirq_time);
1534 raw_spin_unlock(&cpu_base->lock);
1539 * Sleep related functions:
1541 static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1543 struct hrtimer_sleeper *t =
1544 container_of(timer, struct hrtimer_sleeper, timer);
1545 struct task_struct *task = t->task;
1549 wake_up_process(task);
1551 return HRTIMER_NORESTART;
1554 void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1556 sl->timer.function = hrtimer_wakeup;
1559 EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
1561 static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1563 hrtimer_init_sleeper(t, current);
1566 set_current_state(TASK_INTERRUPTIBLE);
1567 hrtimer_start_expires(&t->timer, mode);
1568 if (!hrtimer_active(&t->timer))
1571 if (likely(t->task))
1572 freezable_schedule();
1574 hrtimer_cancel(&t->timer);
1575 mode = HRTIMER_MODE_ABS;
1577 } while (t->task && !signal_pending(current));
1579 __set_current_state(TASK_RUNNING);
1581 return t->task == NULL;
1584 static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp)
1586 struct timespec rmt;
1589 rem = hrtimer_expires_remaining(timer);
1592 rmt = ktime_to_timespec(rem);
1594 if (copy_to_user(rmtp, &rmt, sizeof(*rmtp)))
1600 long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1602 struct hrtimer_sleeper t;
1603 struct timespec __user *rmtp;
1606 hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid,
1608 hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
1610 if (do_nanosleep(&t, HRTIMER_MODE_ABS))
1613 rmtp = restart->nanosleep.rmtp;
1615 ret = update_rmtp(&t.timer, rmtp);
1620 /* The other values in restart are already filled in */
1621 ret = -ERESTART_RESTARTBLOCK;
1623 destroy_hrtimer_on_stack(&t.timer);
1627 long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
1628 const enum hrtimer_mode mode, const clockid_t clockid)
1630 struct restart_block *restart;
1631 struct hrtimer_sleeper t;
1633 unsigned long slack;
1635 slack = current->timer_slack_ns;
1636 if (dl_task(current) || rt_task(current))
1639 hrtimer_init_on_stack(&t.timer, clockid, mode);
1640 hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack);
1641 if (do_nanosleep(&t, mode))
1644 /* Absolute timers do not update the rmtp value and restart: */
1645 if (mode == HRTIMER_MODE_ABS) {
1646 ret = -ERESTARTNOHAND;
1651 ret = update_rmtp(&t.timer, rmtp);
1656 restart = ¤t_thread_info()->restart_block;
1657 restart->fn = hrtimer_nanosleep_restart;
1658 restart->nanosleep.clockid = t.timer.base->clockid;
1659 restart->nanosleep.rmtp = rmtp;
1660 restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
1662 ret = -ERESTART_RESTARTBLOCK;
1664 destroy_hrtimer_on_stack(&t.timer);
1668 SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp,
1669 struct timespec __user *, rmtp)
1673 if (copy_from_user(&tu, rqtp, sizeof(tu)))
1676 if (!timespec_valid(&tu))
1679 return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1683 * Functions related to boot-time initialization:
1685 static void init_hrtimers_cpu(int cpu)
1687 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1690 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1691 cpu_base->clock_base[i].cpu_base = cpu_base;
1692 timerqueue_init_head(&cpu_base->clock_base[i].active);
1695 hrtimer_init_hres(cpu_base);
1698 #ifdef CONFIG_HOTPLUG_CPU
1700 static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1701 struct hrtimer_clock_base *new_base)
1703 struct hrtimer *timer;
1704 struct timerqueue_node *node;
1706 while ((node = timerqueue_getnext(&old_base->active))) {
1707 timer = container_of(node, struct hrtimer, node);
1708 BUG_ON(hrtimer_callback_running(timer));
1709 debug_deactivate(timer);
1712 * Mark it as STATE_MIGRATE not INACTIVE otherwise the
1713 * timer could be seen as !active and just vanish away
1714 * under us on another CPU
1716 __remove_hrtimer(timer, old_base, HRTIMER_STATE_MIGRATE, 0);
1717 timer->base = new_base;
1719 * Enqueue the timers on the new cpu. This does not
1720 * reprogram the event device in case the timer
1721 * expires before the earliest on this CPU, but we run
1722 * hrtimer_interrupt after we migrated everything to
1723 * sort out already expired timers and reprogram the
1726 enqueue_hrtimer(timer, new_base);
1728 /* Clear the migration state bit */
1729 timer->state &= ~HRTIMER_STATE_MIGRATE;
1733 static void migrate_hrtimers(int scpu)
1735 struct hrtimer_cpu_base *old_base, *new_base;
1738 BUG_ON(cpu_online(scpu));
1739 tick_cancel_sched_timer(scpu);
1741 local_irq_disable();
1742 old_base = &per_cpu(hrtimer_bases, scpu);
1743 new_base = &__get_cpu_var(hrtimer_bases);
1745 * The caller is globally serialized and nobody else
1746 * takes two locks at once, deadlock is not possible.
1748 raw_spin_lock(&new_base->lock);
1749 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1751 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1752 migrate_hrtimer_list(&old_base->clock_base[i],
1753 &new_base->clock_base[i]);
1756 raw_spin_unlock(&old_base->lock);
1757 raw_spin_unlock(&new_base->lock);
1759 /* Check, if we got expired work to do */
1760 __hrtimer_peek_ahead_timers();
1764 #endif /* CONFIG_HOTPLUG_CPU */
1766 static int hrtimer_cpu_notify(struct notifier_block *self,
1767 unsigned long action, void *hcpu)
1769 int scpu = (long)hcpu;
1773 case CPU_UP_PREPARE:
1774 case CPU_UP_PREPARE_FROZEN:
1775 init_hrtimers_cpu(scpu);
1778 #ifdef CONFIG_HOTPLUG_CPU
1780 case CPU_DYING_FROZEN:
1781 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DYING, &scpu);
1784 case CPU_DEAD_FROZEN:
1786 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &scpu);
1787 migrate_hrtimers(scpu);
1799 static struct notifier_block hrtimers_nb = {
1800 .notifier_call = hrtimer_cpu_notify,
1803 void __init hrtimers_init(void)
1805 hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
1806 (void *)(long)smp_processor_id());
1807 register_cpu_notifier(&hrtimers_nb);
1808 #ifdef CONFIG_HIGH_RES_TIMERS
1809 open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq);
1814 * schedule_hrtimeout_range_clock - sleep until timeout
1815 * @expires: timeout value (ktime_t)
1816 * @delta: slack in expires timeout (ktime_t)
1817 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1818 * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME
1821 schedule_hrtimeout_range_clock(ktime_t *expires, unsigned long delta,
1822 const enum hrtimer_mode mode, int clock)
1824 struct hrtimer_sleeper t;
1827 * Optimize when a zero timeout value is given. It does not
1828 * matter whether this is an absolute or a relative time.
1830 if (expires && !expires->tv64) {
1831 __set_current_state(TASK_RUNNING);
1836 * A NULL parameter means "infinite"
1840 __set_current_state(TASK_RUNNING);
1844 hrtimer_init_on_stack(&t.timer, clock, mode);
1845 hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
1847 hrtimer_init_sleeper(&t, current);
1849 hrtimer_start_expires(&t.timer, mode);
1850 if (!hrtimer_active(&t.timer))
1856 hrtimer_cancel(&t.timer);
1857 destroy_hrtimer_on_stack(&t.timer);
1859 __set_current_state(TASK_RUNNING);
1861 return !t.task ? 0 : -EINTR;
1865 * schedule_hrtimeout_range - sleep until timeout
1866 * @expires: timeout value (ktime_t)
1867 * @delta: slack in expires timeout (ktime_t)
1868 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1870 * Make the current task sleep until the given expiry time has
1871 * elapsed. The routine will return immediately unless
1872 * the current task state has been set (see set_current_state()).
1874 * The @delta argument gives the kernel the freedom to schedule the
1875 * actual wakeup to a time that is both power and performance friendly.
1876 * The kernel give the normal best effort behavior for "@expires+@delta",
1877 * but may decide to fire the timer earlier, but no earlier than @expires.
1879 * You can set the task state as follows -
1881 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1882 * pass before the routine returns.
1884 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1885 * delivered to the current task.
1887 * The current task state is guaranteed to be TASK_RUNNING when this
1890 * Returns 0 when the timer has expired otherwise -EINTR
1892 int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta,
1893 const enum hrtimer_mode mode)
1895 return schedule_hrtimeout_range_clock(expires, delta, mode,
1898 EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
1901 * schedule_hrtimeout - sleep until timeout
1902 * @expires: timeout value (ktime_t)
1903 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1905 * Make the current task sleep until the given expiry time has
1906 * elapsed. The routine will return immediately unless
1907 * the current task state has been set (see set_current_state()).
1909 * You can set the task state as follows -
1911 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1912 * pass before the routine returns.
1914 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1915 * delivered to the current task.
1917 * The current task state is guaranteed to be TASK_RUNNING when this
1920 * Returns 0 when the timer has expired otherwise -EINTR
1922 int __sched schedule_hrtimeout(ktime_t *expires,
1923 const enum hrtimer_mode mode)
1925 return schedule_hrtimeout_range(expires, 0, mode);
1927 EXPORT_SYMBOL_GPL(schedule_hrtimeout);