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/timer.h>
51 #include <asm/uaccess.h>
53 #include <trace/events/timer.h>
58 * There are more clockids then hrtimer bases. Thus, we index
59 * into the timer bases by the hrtimer_base_type enum. When trying
60 * to reach a base using a clockid, hrtimer_clockid_to_base()
61 * is used to convert from clockid to the proper hrtimer_base_type.
63 DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
69 .index = HRTIMER_BASE_MONOTONIC,
70 .clockid = CLOCK_MONOTONIC,
71 .get_time = &ktime_get,
72 .resolution = KTIME_LOW_RES,
75 .index = HRTIMER_BASE_REALTIME,
76 .clockid = CLOCK_REALTIME,
77 .get_time = &ktime_get_real,
78 .resolution = KTIME_LOW_RES,
81 .index = HRTIMER_BASE_BOOTTIME,
82 .clockid = CLOCK_BOOTTIME,
83 .get_time = &ktime_get_boottime,
84 .resolution = KTIME_LOW_RES,
87 .index = HRTIMER_BASE_TAI,
89 .get_time = &ktime_get_clocktai,
90 .resolution = KTIME_LOW_RES,
95 static const int hrtimer_clock_to_base_table[MAX_CLOCKS] = {
96 [CLOCK_REALTIME] = HRTIMER_BASE_REALTIME,
97 [CLOCK_MONOTONIC] = HRTIMER_BASE_MONOTONIC,
98 [CLOCK_BOOTTIME] = HRTIMER_BASE_BOOTTIME,
99 [CLOCK_TAI] = HRTIMER_BASE_TAI,
102 static inline int hrtimer_clockid_to_base(clockid_t clock_id)
104 return hrtimer_clock_to_base_table[clock_id];
109 * Get the coarse grained time at the softirq based on xtime and
112 static void hrtimer_get_softirq_time(struct hrtimer_cpu_base *base)
114 ktime_t xtim, mono, boot;
115 struct timespec xts, tom, slp;
118 get_xtime_and_monotonic_and_sleep_offset(&xts, &tom, &slp);
119 tai_offset = timekeeping_get_tai_offset();
121 xtim = timespec_to_ktime(xts);
122 mono = ktime_add(xtim, timespec_to_ktime(tom));
123 boot = ktime_add(mono, timespec_to_ktime(slp));
124 base->clock_base[HRTIMER_BASE_REALTIME].softirq_time = xtim;
125 base->clock_base[HRTIMER_BASE_MONOTONIC].softirq_time = mono;
126 base->clock_base[HRTIMER_BASE_BOOTTIME].softirq_time = boot;
127 base->clock_base[HRTIMER_BASE_TAI].softirq_time =
128 ktime_add(xtim, ktime_set(tai_offset, 0));
132 * Functions and macros which are different for UP/SMP systems are kept in a
138 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
139 * means that all timers which are tied to this base via timer->base are
140 * locked, and the base itself is locked too.
142 * So __run_timers/migrate_timers can safely modify all timers which could
143 * be found on the lists/queues.
145 * When the timer's base is locked, and the timer removed from list, it is
146 * possible to set timer->base = NULL and drop the lock: the timer remains
150 struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
151 unsigned long *flags)
153 struct hrtimer_clock_base *base;
157 if (likely(base != NULL)) {
158 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
159 if (likely(base == timer->base))
161 /* The timer has migrated to another CPU: */
162 raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
170 * Get the preferred target CPU for NOHZ
172 static int hrtimer_get_target(int this_cpu, int pinned)
175 if (!pinned && get_sysctl_timer_migration() && idle_cpu(this_cpu))
176 return get_nohz_timer_target();
182 * With HIGHRES=y we do not migrate the timer when it is expiring
183 * before the next event on the target cpu because we cannot reprogram
184 * the target cpu hardware and we would cause it to fire late.
186 * Called with cpu_base->lock of target cpu held.
189 hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base)
191 #ifdef CONFIG_HIGH_RES_TIMERS
194 if (!new_base->cpu_base->hres_active)
197 expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset);
198 return expires.tv64 <= new_base->cpu_base->expires_next.tv64;
205 * Switch the timer base to the current CPU when possible.
207 static inline struct hrtimer_clock_base *
208 switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
211 struct hrtimer_clock_base *new_base;
212 struct hrtimer_cpu_base *new_cpu_base;
213 int this_cpu = smp_processor_id();
214 int cpu = hrtimer_get_target(this_cpu, pinned);
215 int basenum = base->index;
218 new_cpu_base = &per_cpu(hrtimer_bases, cpu);
219 new_base = &new_cpu_base->clock_base[basenum];
221 if (base != new_base) {
223 * We are trying to move timer to new_base.
224 * However we can't change timer's base while it is running,
225 * so we keep it on the same CPU. No hassle vs. reprogramming
226 * the event source in the high resolution case. The softirq
227 * code will take care of this when the timer function has
228 * completed. There is no conflict as we hold the lock until
229 * the timer is enqueued.
231 if (unlikely(hrtimer_callback_running(timer)))
234 /* See the comment in lock_timer_base() */
236 raw_spin_unlock(&base->cpu_base->lock);
237 raw_spin_lock(&new_base->cpu_base->lock);
239 if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) {
241 raw_spin_unlock(&new_base->cpu_base->lock);
242 raw_spin_lock(&base->cpu_base->lock);
246 timer->base = new_base;
251 #else /* CONFIG_SMP */
253 static inline struct hrtimer_clock_base *
254 lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
256 struct hrtimer_clock_base *base = timer->base;
258 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
263 # define switch_hrtimer_base(t, b, p) (b)
265 #endif /* !CONFIG_SMP */
268 * Functions for the union type storage format of ktime_t which are
269 * too large for inlining:
271 #if BITS_PER_LONG < 64
272 # ifndef CONFIG_KTIME_SCALAR
274 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
276 * @nsec: the scalar nsec value to add
278 * Returns the sum of kt and nsec in ktime_t format
280 ktime_t ktime_add_ns(const ktime_t kt, u64 nsec)
284 if (likely(nsec < NSEC_PER_SEC)) {
287 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
289 tmp = ktime_set((long)nsec, rem);
292 return ktime_add(kt, tmp);
295 EXPORT_SYMBOL_GPL(ktime_add_ns);
298 * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable
300 * @nsec: the scalar nsec value to subtract
302 * Returns the subtraction of @nsec from @kt in ktime_t format
304 ktime_t ktime_sub_ns(const ktime_t kt, u64 nsec)
308 if (likely(nsec < NSEC_PER_SEC)) {
311 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
313 tmp = ktime_set((long)nsec, rem);
316 return ktime_sub(kt, tmp);
319 EXPORT_SYMBOL_GPL(ktime_sub_ns);
320 # endif /* !CONFIG_KTIME_SCALAR */
323 * Divide a ktime value by a nanosecond value
325 u64 ktime_divns(const ktime_t kt, s64 div)
330 dclc = ktime_to_ns(kt);
331 /* Make sure the divisor is less than 2^32: */
337 do_div(dclc, (unsigned long) div);
341 #endif /* BITS_PER_LONG >= 64 */
344 * Add two ktime values and do a safety check for overflow:
346 ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
348 ktime_t res = ktime_add(lhs, rhs);
351 * We use KTIME_SEC_MAX here, the maximum timeout which we can
352 * return to user space in a timespec:
354 if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64)
355 res = ktime_set(KTIME_SEC_MAX, 0);
360 EXPORT_SYMBOL_GPL(ktime_add_safe);
362 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
364 static struct debug_obj_descr hrtimer_debug_descr;
366 static void *hrtimer_debug_hint(void *addr)
368 return ((struct hrtimer *) addr)->function;
372 * fixup_init is called when:
373 * - an active object is initialized
375 static int hrtimer_fixup_init(void *addr, enum debug_obj_state state)
377 struct hrtimer *timer = addr;
380 case ODEBUG_STATE_ACTIVE:
381 hrtimer_cancel(timer);
382 debug_object_init(timer, &hrtimer_debug_descr);
390 * fixup_activate is called when:
391 * - an active object is activated
392 * - an unknown object is activated (might be a statically initialized object)
394 static int hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
398 case ODEBUG_STATE_NOTAVAILABLE:
402 case ODEBUG_STATE_ACTIVE:
411 * fixup_free is called when:
412 * - an active object is freed
414 static int hrtimer_fixup_free(void *addr, enum debug_obj_state state)
416 struct hrtimer *timer = addr;
419 case ODEBUG_STATE_ACTIVE:
420 hrtimer_cancel(timer);
421 debug_object_free(timer, &hrtimer_debug_descr);
428 static struct debug_obj_descr hrtimer_debug_descr = {
430 .debug_hint = hrtimer_debug_hint,
431 .fixup_init = hrtimer_fixup_init,
432 .fixup_activate = hrtimer_fixup_activate,
433 .fixup_free = hrtimer_fixup_free,
436 static inline void debug_hrtimer_init(struct hrtimer *timer)
438 debug_object_init(timer, &hrtimer_debug_descr);
441 static inline void debug_hrtimer_activate(struct hrtimer *timer)
443 debug_object_activate(timer, &hrtimer_debug_descr);
446 static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
448 debug_object_deactivate(timer, &hrtimer_debug_descr);
451 static inline void debug_hrtimer_free(struct hrtimer *timer)
453 debug_object_free(timer, &hrtimer_debug_descr);
456 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
457 enum hrtimer_mode mode);
459 void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
460 enum hrtimer_mode mode)
462 debug_object_init_on_stack(timer, &hrtimer_debug_descr);
463 __hrtimer_init(timer, clock_id, mode);
465 EXPORT_SYMBOL_GPL(hrtimer_init_on_stack);
467 void destroy_hrtimer_on_stack(struct hrtimer *timer)
469 debug_object_free(timer, &hrtimer_debug_descr);
473 static inline void debug_hrtimer_init(struct hrtimer *timer) { }
474 static inline void debug_hrtimer_activate(struct hrtimer *timer) { }
475 static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
479 debug_init(struct hrtimer *timer, clockid_t clockid,
480 enum hrtimer_mode mode)
482 debug_hrtimer_init(timer);
483 trace_hrtimer_init(timer, clockid, mode);
486 static inline void debug_activate(struct hrtimer *timer)
488 debug_hrtimer_activate(timer);
489 trace_hrtimer_start(timer);
492 static inline void debug_deactivate(struct hrtimer *timer)
494 debug_hrtimer_deactivate(timer);
495 trace_hrtimer_cancel(timer);
498 /* High resolution timer related functions */
499 #ifdef CONFIG_HIGH_RES_TIMERS
502 * High resolution timer enabled ?
504 static int hrtimer_hres_enabled __read_mostly = 1;
507 * Enable / Disable high resolution mode
509 static int __init setup_hrtimer_hres(char *str)
511 if (!strcmp(str, "off"))
512 hrtimer_hres_enabled = 0;
513 else if (!strcmp(str, "on"))
514 hrtimer_hres_enabled = 1;
520 __setup("highres=", setup_hrtimer_hres);
523 * hrtimer_high_res_enabled - query, if the highres mode is enabled
525 static inline int hrtimer_is_hres_enabled(void)
527 return hrtimer_hres_enabled;
531 * Is the high resolution mode active ?
533 static inline int hrtimer_hres_active(void)
535 return __this_cpu_read(hrtimer_bases.hres_active);
539 * Reprogram the event source with checking both queues for the
541 * Called with interrupts disabled and base->lock held
544 hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
547 struct hrtimer_clock_base *base = cpu_base->clock_base;
548 ktime_t expires, expires_next;
550 expires_next.tv64 = KTIME_MAX;
552 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
553 struct hrtimer *timer;
554 struct timerqueue_node *next;
556 next = timerqueue_getnext(&base->active);
559 timer = container_of(next, struct hrtimer, node);
561 expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
563 * clock_was_set() has changed base->offset so the
564 * result might be negative. Fix it up to prevent a
565 * false positive in clockevents_program_event()
567 if (expires.tv64 < 0)
569 if (expires.tv64 < expires_next.tv64)
570 expires_next = expires;
573 if (skip_equal && expires_next.tv64 == cpu_base->expires_next.tv64)
576 cpu_base->expires_next.tv64 = expires_next.tv64;
578 if (cpu_base->expires_next.tv64 != KTIME_MAX)
579 tick_program_event(cpu_base->expires_next, 1);
583 * Shared reprogramming for clock_realtime and clock_monotonic
585 * When a timer is enqueued and expires earlier than the already enqueued
586 * timers, we have to check, whether it expires earlier than the timer for
587 * which the clock event device was armed.
589 * Called with interrupts disabled and base->cpu_base.lock held
591 static int hrtimer_reprogram(struct hrtimer *timer,
592 struct hrtimer_clock_base *base)
594 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
595 ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
598 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
601 * When the callback is running, we do not reprogram the clock event
602 * device. The timer callback is either running on a different CPU or
603 * the callback is executed in the hrtimer_interrupt context. The
604 * reprogramming is handled either by the softirq, which called the
605 * callback or at the end of the hrtimer_interrupt.
607 if (hrtimer_callback_running(timer))
611 * CLOCK_REALTIME timer might be requested with an absolute
612 * expiry time which is less than base->offset. Nothing wrong
613 * about that, just avoid to call into the tick code, which
614 * has now objections against negative expiry values.
616 if (expires.tv64 < 0)
619 if (expires.tv64 >= cpu_base->expires_next.tv64)
623 * If a hang was detected in the last timer interrupt then we
624 * do not schedule a timer which is earlier than the expiry
625 * which we enforced in the hang detection. We want the system
628 if (cpu_base->hang_detected)
632 * Clockevents returns -ETIME, when the event was in the past.
634 res = tick_program_event(expires, 0);
635 if (!IS_ERR_VALUE(res))
636 cpu_base->expires_next = expires;
641 * Initialize the high resolution related parts of cpu_base
643 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
645 base->expires_next.tv64 = KTIME_MAX;
646 base->hres_active = 0;
650 * When High resolution timers are active, try to reprogram. Note, that in case
651 * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry
652 * check happens. The timer gets enqueued into the rbtree. The reprogramming
653 * and expiry check is done in the hrtimer_interrupt or in the softirq.
655 static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
656 struct hrtimer_clock_base *base)
658 return base->cpu_base->hres_active && hrtimer_reprogram(timer, base);
661 static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base)
663 ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset;
664 ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset;
665 ktime_t *offs_tai = &base->clock_base[HRTIMER_BASE_TAI].offset;
667 return ktime_get_update_offsets(offs_real, offs_boot, offs_tai);
671 * Retrigger next event is called after clock was set
673 * Called with interrupts disabled via on_each_cpu()
675 static void retrigger_next_event(void *arg)
677 struct hrtimer_cpu_base *base = &__get_cpu_var(hrtimer_bases);
679 if (!hrtimer_hres_active())
682 raw_spin_lock(&base->lock);
683 hrtimer_update_base(base);
684 hrtimer_force_reprogram(base, 0);
685 raw_spin_unlock(&base->lock);
689 * Switch to high resolution mode
691 static int hrtimer_switch_to_hres(void)
693 int i, cpu = smp_processor_id();
694 struct hrtimer_cpu_base *base = &per_cpu(hrtimer_bases, cpu);
697 if (base->hres_active)
700 local_irq_save(flags);
702 if (tick_init_highres()) {
703 local_irq_restore(flags);
704 printk(KERN_WARNING "Could not switch to high resolution "
705 "mode on CPU %d\n", cpu);
708 base->hres_active = 1;
709 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
710 base->clock_base[i].resolution = KTIME_HIGH_RES;
712 tick_setup_sched_timer();
713 /* "Retrigger" the interrupt to get things going */
714 retrigger_next_event(NULL);
715 local_irq_restore(flags);
720 * Called from timekeeping code to reprogramm the hrtimer interrupt
721 * device. If called from the timer interrupt context we defer it to
724 void clock_was_set_delayed(void)
726 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
728 cpu_base->clock_was_set = 1;
729 __raise_softirq_irqoff(HRTIMER_SOFTIRQ);
734 static inline int hrtimer_hres_active(void) { return 0; }
735 static inline int hrtimer_is_hres_enabled(void) { return 0; }
736 static inline int hrtimer_switch_to_hres(void) { return 0; }
738 hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { }
739 static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
740 struct hrtimer_clock_base *base)
744 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { }
745 static inline void retrigger_next_event(void *arg) { }
747 #endif /* CONFIG_HIGH_RES_TIMERS */
750 * Clock realtime was set
752 * Change the offset of the realtime clock vs. the monotonic
755 * We might have to reprogram the high resolution timer interrupt. On
756 * SMP we call the architecture specific code to retrigger _all_ high
757 * resolution timer interrupts. On UP we just disable interrupts and
758 * call the high resolution interrupt code.
760 void clock_was_set(void)
762 #ifdef CONFIG_HIGH_RES_TIMERS
763 /* Retrigger the CPU local events everywhere */
764 on_each_cpu(retrigger_next_event, NULL, 1);
766 timerfd_clock_was_set();
770 * During resume we might have to reprogram the high resolution timer
771 * interrupt (on the local CPU):
773 void hrtimers_resume(void)
775 WARN_ONCE(!irqs_disabled(),
776 KERN_INFO "hrtimers_resume() called with IRQs enabled!");
778 retrigger_next_event(NULL);
779 timerfd_clock_was_set();
782 static inline void timer_stats_hrtimer_set_start_info(struct hrtimer *timer)
784 #ifdef CONFIG_TIMER_STATS
785 if (timer->start_site)
787 timer->start_site = __builtin_return_address(0);
788 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
789 timer->start_pid = current->pid;
793 static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer *timer)
795 #ifdef CONFIG_TIMER_STATS
796 timer->start_site = NULL;
800 static inline void timer_stats_account_hrtimer(struct hrtimer *timer)
802 #ifdef CONFIG_TIMER_STATS
803 if (likely(!timer_stats_active))
805 timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
806 timer->function, timer->start_comm, 0);
811 * Counterpart to lock_hrtimer_base above:
814 void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
816 raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
820 * hrtimer_forward - forward the timer expiry
821 * @timer: hrtimer to forward
822 * @now: forward past this time
823 * @interval: the interval to forward
825 * Forward the timer expiry so it will expire in the future.
826 * Returns the number of overruns.
828 u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
833 delta = ktime_sub(now, hrtimer_get_expires(timer));
838 if (interval.tv64 < timer->base->resolution.tv64)
839 interval.tv64 = timer->base->resolution.tv64;
841 if (unlikely(delta.tv64 >= interval.tv64)) {
842 s64 incr = ktime_to_ns(interval);
844 orun = ktime_divns(delta, incr);
845 hrtimer_add_expires_ns(timer, incr * orun);
846 if (hrtimer_get_expires_tv64(timer) > now.tv64)
849 * This (and the ktime_add() below) is the
850 * correction for exact:
854 hrtimer_add_expires(timer, interval);
858 EXPORT_SYMBOL_GPL(hrtimer_forward);
861 * enqueue_hrtimer - internal function to (re)start a timer
863 * The timer is inserted in expiry order. Insertion into the
864 * red black tree is O(log(n)). Must hold the base lock.
866 * Returns 1 when the new timer is the leftmost timer in the tree.
868 static int enqueue_hrtimer(struct hrtimer *timer,
869 struct hrtimer_clock_base *base)
871 debug_activate(timer);
873 timerqueue_add(&base->active, &timer->node);
874 base->cpu_base->active_bases |= 1 << base->index;
877 * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
878 * state of a possibly running callback.
880 timer->state |= HRTIMER_STATE_ENQUEUED;
882 return (&timer->node == base->active.next);
886 * __remove_hrtimer - internal function to remove a timer
888 * Caller must hold the base lock.
890 * High resolution timer mode reprograms the clock event device when the
891 * timer is the one which expires next. The caller can disable this by setting
892 * reprogram to zero. This is useful, when the context does a reprogramming
893 * anyway (e.g. timer interrupt)
895 static void __remove_hrtimer(struct hrtimer *timer,
896 struct hrtimer_clock_base *base,
897 unsigned long newstate, int reprogram)
899 struct timerqueue_node *next_timer;
900 if (!(timer->state & HRTIMER_STATE_ENQUEUED))
903 next_timer = timerqueue_getnext(&base->active);
904 timerqueue_del(&base->active, &timer->node);
905 if (&timer->node == next_timer) {
906 #ifdef CONFIG_HIGH_RES_TIMERS
907 /* Reprogram the clock event device. if enabled */
908 if (reprogram && hrtimer_hres_active()) {
911 expires = ktime_sub(hrtimer_get_expires(timer),
913 if (base->cpu_base->expires_next.tv64 == expires.tv64)
914 hrtimer_force_reprogram(base->cpu_base, 1);
918 if (!timerqueue_getnext(&base->active))
919 base->cpu_base->active_bases &= ~(1 << base->index);
921 timer->state = newstate;
925 * remove hrtimer, called with base lock held
928 remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base)
930 if (hrtimer_is_queued(timer)) {
935 * Remove the timer and force reprogramming when high
936 * resolution mode is active and the timer is on the current
937 * CPU. If we remove a timer on another CPU, reprogramming is
938 * skipped. The interrupt event on this CPU is fired and
939 * reprogramming happens in the interrupt handler. This is a
940 * rare case and less expensive than a smp call.
942 debug_deactivate(timer);
943 timer_stats_hrtimer_clear_start_info(timer);
944 reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases);
946 * We must preserve the CALLBACK state flag here,
947 * otherwise we could move the timer base in
948 * switch_hrtimer_base.
950 state = timer->state & HRTIMER_STATE_CALLBACK;
951 __remove_hrtimer(timer, base, state, reprogram);
957 int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
958 unsigned long delta_ns, const enum hrtimer_mode mode,
961 struct hrtimer_clock_base *base, *new_base;
965 base = lock_hrtimer_base(timer, &flags);
967 /* Remove an active timer from the queue: */
968 ret = remove_hrtimer(timer, base);
970 /* Switch the timer base, if necessary: */
971 new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED);
973 if (mode & HRTIMER_MODE_REL) {
974 tim = ktime_add_safe(tim, new_base->get_time());
976 * CONFIG_TIME_LOW_RES is a temporary way for architectures
977 * to signal that they simply return xtime in
978 * do_gettimeoffset(). In this case we want to round up by
979 * resolution when starting a relative timer, to avoid short
980 * timeouts. This will go away with the GTOD framework.
982 #ifdef CONFIG_TIME_LOW_RES
983 tim = ktime_add_safe(tim, base->resolution);
987 hrtimer_set_expires_range_ns(timer, tim, delta_ns);
989 timer_stats_hrtimer_set_start_info(timer);
991 leftmost = enqueue_hrtimer(timer, new_base);
994 * Only allow reprogramming if the new base is on this CPU.
995 * (it might still be on another CPU if the timer was pending)
997 * XXX send_remote_softirq() ?
999 if (leftmost && new_base->cpu_base == &__get_cpu_var(hrtimer_bases)
1000 && hrtimer_enqueue_reprogram(timer, new_base)) {
1003 * We need to drop cpu_base->lock to avoid a
1004 * lock ordering issue vs. rq->lock.
1006 raw_spin_unlock(&new_base->cpu_base->lock);
1007 raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1008 local_irq_restore(flags);
1011 __raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1015 unlock_hrtimer_base(timer, &flags);
1021 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
1022 * @timer: the timer to be added
1024 * @delta_ns: "slack" range for the timer
1025 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
1029 * 1 when the timer was active
1031 int hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1032 unsigned long delta_ns, const enum hrtimer_mode mode)
1034 return __hrtimer_start_range_ns(timer, tim, delta_ns, mode, 1);
1036 EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
1039 * hrtimer_start - (re)start an hrtimer on the current CPU
1040 * @timer: the timer to be added
1042 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
1046 * 1 when the timer was active
1049 hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode)
1051 return __hrtimer_start_range_ns(timer, tim, 0, mode, 1);
1053 EXPORT_SYMBOL_GPL(hrtimer_start);
1057 * hrtimer_try_to_cancel - try to deactivate a timer
1058 * @timer: hrtimer to stop
1061 * 0 when the timer was not active
1062 * 1 when the timer was active
1063 * -1 when the timer is currently excuting the callback function and
1066 int hrtimer_try_to_cancel(struct hrtimer *timer)
1068 struct hrtimer_clock_base *base;
1069 unsigned long flags;
1072 base = lock_hrtimer_base(timer, &flags);
1074 if (!hrtimer_callback_running(timer))
1075 ret = remove_hrtimer(timer, base);
1077 unlock_hrtimer_base(timer, &flags);
1082 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1085 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1086 * @timer: the timer to be cancelled
1089 * 0 when the timer was not active
1090 * 1 when the timer was active
1092 int hrtimer_cancel(struct hrtimer *timer)
1095 int ret = hrtimer_try_to_cancel(timer);
1102 EXPORT_SYMBOL_GPL(hrtimer_cancel);
1105 * hrtimer_get_remaining - get remaining time for the timer
1106 * @timer: the timer to read
1108 ktime_t hrtimer_get_remaining(const struct hrtimer *timer)
1110 unsigned long flags;
1113 lock_hrtimer_base(timer, &flags);
1114 rem = hrtimer_expires_remaining(timer);
1115 unlock_hrtimer_base(timer, &flags);
1119 EXPORT_SYMBOL_GPL(hrtimer_get_remaining);
1123 * hrtimer_get_next_event - get the time until next expiry event
1125 * Returns the delta to the next expiry event or KTIME_MAX if no timer
1128 ktime_t hrtimer_get_next_event(void)
1130 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1131 struct hrtimer_clock_base *base = cpu_base->clock_base;
1132 ktime_t delta, mindelta = { .tv64 = KTIME_MAX };
1133 unsigned long flags;
1136 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1138 if (!hrtimer_hres_active()) {
1139 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
1140 struct hrtimer *timer;
1141 struct timerqueue_node *next;
1143 next = timerqueue_getnext(&base->active);
1147 timer = container_of(next, struct hrtimer, node);
1148 delta.tv64 = hrtimer_get_expires_tv64(timer);
1149 delta = ktime_sub(delta, base->get_time());
1150 if (delta.tv64 < mindelta.tv64)
1151 mindelta.tv64 = delta.tv64;
1155 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1157 if (mindelta.tv64 < 0)
1163 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1164 enum hrtimer_mode mode)
1166 struct hrtimer_cpu_base *cpu_base;
1169 memset(timer, 0, sizeof(struct hrtimer));
1171 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1173 if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
1174 clock_id = CLOCK_MONOTONIC;
1176 base = hrtimer_clockid_to_base(clock_id);
1177 timer->base = &cpu_base->clock_base[base];
1178 timerqueue_init(&timer->node);
1180 #ifdef CONFIG_TIMER_STATS
1181 timer->start_site = NULL;
1182 timer->start_pid = -1;
1183 memset(timer->start_comm, 0, TASK_COMM_LEN);
1188 * hrtimer_init - initialize a timer to the given clock
1189 * @timer: the timer to be initialized
1190 * @clock_id: the clock to be used
1191 * @mode: timer mode abs/rel
1193 void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1194 enum hrtimer_mode mode)
1196 debug_init(timer, clock_id, mode);
1197 __hrtimer_init(timer, clock_id, mode);
1199 EXPORT_SYMBOL_GPL(hrtimer_init);
1202 * hrtimer_get_res - get the timer resolution for a clock
1203 * @which_clock: which clock to query
1204 * @tp: pointer to timespec variable to store the resolution
1206 * Store the resolution of the clock selected by @which_clock in the
1207 * variable pointed to by @tp.
1209 int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp)
1211 struct hrtimer_cpu_base *cpu_base;
1212 int base = hrtimer_clockid_to_base(which_clock);
1214 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1215 *tp = ktime_to_timespec(cpu_base->clock_base[base].resolution);
1219 EXPORT_SYMBOL_GPL(hrtimer_get_res);
1221 static void __run_hrtimer(struct hrtimer *timer, ktime_t *now)
1223 struct hrtimer_clock_base *base = timer->base;
1224 struct hrtimer_cpu_base *cpu_base = base->cpu_base;
1225 enum hrtimer_restart (*fn)(struct hrtimer *);
1228 WARN_ON(!irqs_disabled());
1230 debug_deactivate(timer);
1231 __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0);
1232 timer_stats_account_hrtimer(timer);
1233 fn = timer->function;
1236 * Because we run timers from hardirq context, there is no chance
1237 * they get migrated to another cpu, therefore its safe to unlock
1240 raw_spin_unlock(&cpu_base->lock);
1241 trace_hrtimer_expire_entry(timer, now);
1242 restart = fn(timer);
1243 trace_hrtimer_expire_exit(timer);
1244 raw_spin_lock(&cpu_base->lock);
1247 * Note: We clear the CALLBACK bit after enqueue_hrtimer and
1248 * we do not reprogramm the event hardware. Happens either in
1249 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1251 if (restart != HRTIMER_NORESTART) {
1252 BUG_ON(timer->state != HRTIMER_STATE_CALLBACK);
1253 enqueue_hrtimer(timer, base);
1256 WARN_ON_ONCE(!(timer->state & HRTIMER_STATE_CALLBACK));
1258 timer->state &= ~HRTIMER_STATE_CALLBACK;
1261 #ifdef CONFIG_HIGH_RES_TIMERS
1264 * High resolution timer interrupt
1265 * Called with interrupts disabled
1267 void hrtimer_interrupt(struct clock_event_device *dev)
1269 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1270 ktime_t expires_next, now, entry_time, delta;
1273 BUG_ON(!cpu_base->hres_active);
1274 cpu_base->nr_events++;
1275 dev->next_event.tv64 = KTIME_MAX;
1277 raw_spin_lock(&cpu_base->lock);
1278 entry_time = now = hrtimer_update_base(cpu_base);
1280 expires_next.tv64 = KTIME_MAX;
1282 * We set expires_next to KTIME_MAX here with cpu_base->lock
1283 * held to prevent that a timer is enqueued in our queue via
1284 * the migration code. This does not affect enqueueing of
1285 * timers which run their callback and need to be requeued on
1288 cpu_base->expires_next.tv64 = KTIME_MAX;
1290 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1291 struct hrtimer_clock_base *base;
1292 struct timerqueue_node *node;
1295 if (!(cpu_base->active_bases & (1 << i)))
1298 base = cpu_base->clock_base + i;
1299 basenow = ktime_add(now, base->offset);
1301 while ((node = timerqueue_getnext(&base->active))) {
1302 struct hrtimer *timer;
1304 timer = container_of(node, struct hrtimer, node);
1307 * The immediate goal for using the softexpires is
1308 * minimizing wakeups, not running timers at the
1309 * earliest interrupt after their soft expiration.
1310 * This allows us to avoid using a Priority Search
1311 * Tree, which can answer a stabbing querry for
1312 * overlapping intervals and instead use the simple
1313 * BST we already have.
1314 * We don't add extra wakeups by delaying timers that
1315 * are right-of a not yet expired timer, because that
1316 * timer will have to trigger a wakeup anyway.
1319 if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer)) {
1322 expires = ktime_sub(hrtimer_get_expires(timer),
1324 if (expires.tv64 < expires_next.tv64)
1325 expires_next = expires;
1329 __run_hrtimer(timer, &basenow);
1334 * Store the new expiry value so the migration code can verify
1337 cpu_base->expires_next = expires_next;
1338 raw_spin_unlock(&cpu_base->lock);
1340 /* Reprogramming necessary ? */
1341 if (expires_next.tv64 == KTIME_MAX ||
1342 !tick_program_event(expires_next, 0)) {
1343 cpu_base->hang_detected = 0;
1348 * The next timer was already expired due to:
1350 * - long lasting callbacks
1351 * - being scheduled away when running in a VM
1353 * We need to prevent that we loop forever in the hrtimer
1354 * interrupt routine. We give it 3 attempts to avoid
1355 * overreacting on some spurious event.
1357 * Acquire base lock for updating the offsets and retrieving
1360 raw_spin_lock(&cpu_base->lock);
1361 now = hrtimer_update_base(cpu_base);
1362 cpu_base->nr_retries++;
1366 * Give the system a chance to do something else than looping
1367 * here. We stored the entry time, so we know exactly how long
1368 * we spent here. We schedule the next event this amount of
1371 cpu_base->nr_hangs++;
1372 cpu_base->hang_detected = 1;
1373 raw_spin_unlock(&cpu_base->lock);
1374 delta = ktime_sub(now, entry_time);
1375 if (delta.tv64 > cpu_base->max_hang_time.tv64)
1376 cpu_base->max_hang_time = delta;
1378 * Limit it to a sensible value as we enforce a longer
1379 * delay. Give the CPU at least 100ms to catch up.
1381 if (delta.tv64 > 100 * NSEC_PER_MSEC)
1382 expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1384 expires_next = ktime_add(now, delta);
1385 tick_program_event(expires_next, 1);
1386 printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n",
1387 ktime_to_ns(delta));
1391 * local version of hrtimer_peek_ahead_timers() called with interrupts
1394 static void __hrtimer_peek_ahead_timers(void)
1396 struct tick_device *td;
1398 if (!hrtimer_hres_active())
1401 td = &__get_cpu_var(tick_cpu_device);
1402 if (td && td->evtdev)
1403 hrtimer_interrupt(td->evtdev);
1407 * hrtimer_peek_ahead_timers -- run soft-expired timers now
1409 * hrtimer_peek_ahead_timers will peek at the timer queue of
1410 * the current cpu and check if there are any timers for which
1411 * the soft expires time has passed. If any such timers exist,
1412 * they are run immediately and then removed from the timer queue.
1415 void hrtimer_peek_ahead_timers(void)
1417 unsigned long flags;
1419 local_irq_save(flags);
1420 __hrtimer_peek_ahead_timers();
1421 local_irq_restore(flags);
1424 static void run_hrtimer_softirq(struct softirq_action *h)
1426 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1428 if (cpu_base->clock_was_set) {
1429 cpu_base->clock_was_set = 0;
1433 hrtimer_peek_ahead_timers();
1436 #else /* CONFIG_HIGH_RES_TIMERS */
1438 static inline void __hrtimer_peek_ahead_timers(void) { }
1440 #endif /* !CONFIG_HIGH_RES_TIMERS */
1443 * Called from timer softirq every jiffy, expire hrtimers:
1445 * For HRT its the fall back code to run the softirq in the timer
1446 * softirq context in case the hrtimer initialization failed or has
1447 * not been done yet.
1449 void hrtimer_run_pending(void)
1451 if (hrtimer_hres_active())
1455 * This _is_ ugly: We have to check in the softirq context,
1456 * whether we can switch to highres and / or nohz mode. The
1457 * clocksource switch happens in the timer interrupt with
1458 * xtime_lock held. Notification from there only sets the
1459 * check bit in the tick_oneshot code, otherwise we might
1460 * deadlock vs. xtime_lock.
1462 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1463 hrtimer_switch_to_hres();
1467 * Called from hardirq context every jiffy
1469 void hrtimer_run_queues(void)
1471 struct timerqueue_node *node;
1472 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1473 struct hrtimer_clock_base *base;
1474 int index, gettime = 1;
1476 if (hrtimer_hres_active())
1479 for (index = 0; index < HRTIMER_MAX_CLOCK_BASES; index++) {
1480 base = &cpu_base->clock_base[index];
1481 if (!timerqueue_getnext(&base->active))
1485 hrtimer_get_softirq_time(cpu_base);
1489 raw_spin_lock(&cpu_base->lock);
1491 while ((node = timerqueue_getnext(&base->active))) {
1492 struct hrtimer *timer;
1494 timer = container_of(node, struct hrtimer, node);
1495 if (base->softirq_time.tv64 <=
1496 hrtimer_get_expires_tv64(timer))
1499 __run_hrtimer(timer, &base->softirq_time);
1501 raw_spin_unlock(&cpu_base->lock);
1506 * Sleep related functions:
1508 static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1510 struct hrtimer_sleeper *t =
1511 container_of(timer, struct hrtimer_sleeper, timer);
1512 struct task_struct *task = t->task;
1516 wake_up_process(task);
1518 return HRTIMER_NORESTART;
1521 void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1523 sl->timer.function = hrtimer_wakeup;
1526 EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
1528 static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1530 hrtimer_init_sleeper(t, current);
1533 set_current_state(TASK_INTERRUPTIBLE);
1534 hrtimer_start_expires(&t->timer, mode);
1535 if (!hrtimer_active(&t->timer))
1538 if (likely(t->task))
1541 hrtimer_cancel(&t->timer);
1542 mode = HRTIMER_MODE_ABS;
1544 } while (t->task && !signal_pending(current));
1546 __set_current_state(TASK_RUNNING);
1548 return t->task == NULL;
1551 static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp)
1553 struct timespec rmt;
1556 rem = hrtimer_expires_remaining(timer);
1559 rmt = ktime_to_timespec(rem);
1561 if (copy_to_user(rmtp, &rmt, sizeof(*rmtp)))
1567 long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1569 struct hrtimer_sleeper t;
1570 struct timespec __user *rmtp;
1573 hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid,
1575 hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
1577 if (do_nanosleep(&t, HRTIMER_MODE_ABS))
1580 rmtp = restart->nanosleep.rmtp;
1582 ret = update_rmtp(&t.timer, rmtp);
1587 /* The other values in restart are already filled in */
1588 ret = -ERESTART_RESTARTBLOCK;
1590 destroy_hrtimer_on_stack(&t.timer);
1594 long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
1595 const enum hrtimer_mode mode, const clockid_t clockid)
1597 struct restart_block *restart;
1598 struct hrtimer_sleeper t;
1600 unsigned long slack;
1602 slack = current->timer_slack_ns;
1603 if (rt_task(current))
1606 hrtimer_init_on_stack(&t.timer, clockid, mode);
1607 hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack);
1608 if (do_nanosleep(&t, mode))
1611 /* Absolute timers do not update the rmtp value and restart: */
1612 if (mode == HRTIMER_MODE_ABS) {
1613 ret = -ERESTARTNOHAND;
1618 ret = update_rmtp(&t.timer, rmtp);
1623 restart = ¤t_thread_info()->restart_block;
1624 restart->fn = hrtimer_nanosleep_restart;
1625 restart->nanosleep.clockid = t.timer.base->clockid;
1626 restart->nanosleep.rmtp = rmtp;
1627 restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
1629 ret = -ERESTART_RESTARTBLOCK;
1631 destroy_hrtimer_on_stack(&t.timer);
1635 SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp,
1636 struct timespec __user *, rmtp)
1640 if (copy_from_user(&tu, rqtp, sizeof(tu)))
1643 if (!timespec_valid(&tu))
1646 return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1650 * Functions related to boot-time initialization:
1652 static void __cpuinit init_hrtimers_cpu(int cpu)
1654 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1657 raw_spin_lock_init(&cpu_base->lock);
1659 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1660 cpu_base->clock_base[i].cpu_base = cpu_base;
1661 timerqueue_init_head(&cpu_base->clock_base[i].active);
1664 hrtimer_init_hres(cpu_base);
1667 #ifdef CONFIG_HOTPLUG_CPU
1669 static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1670 struct hrtimer_clock_base *new_base)
1672 struct hrtimer *timer;
1673 struct timerqueue_node *node;
1675 while ((node = timerqueue_getnext(&old_base->active))) {
1676 timer = container_of(node, struct hrtimer, node);
1677 BUG_ON(hrtimer_callback_running(timer));
1678 debug_deactivate(timer);
1681 * Mark it as STATE_MIGRATE not INACTIVE otherwise the
1682 * timer could be seen as !active and just vanish away
1683 * under us on another CPU
1685 __remove_hrtimer(timer, old_base, HRTIMER_STATE_MIGRATE, 0);
1686 timer->base = new_base;
1688 * Enqueue the timers on the new cpu. This does not
1689 * reprogram the event device in case the timer
1690 * expires before the earliest on this CPU, but we run
1691 * hrtimer_interrupt after we migrated everything to
1692 * sort out already expired timers and reprogram the
1695 enqueue_hrtimer(timer, new_base);
1697 /* Clear the migration state bit */
1698 timer->state &= ~HRTIMER_STATE_MIGRATE;
1702 static void migrate_hrtimers(int scpu)
1704 struct hrtimer_cpu_base *old_base, *new_base;
1707 BUG_ON(cpu_online(scpu));
1708 tick_cancel_sched_timer(scpu);
1710 local_irq_disable();
1711 old_base = &per_cpu(hrtimer_bases, scpu);
1712 new_base = &__get_cpu_var(hrtimer_bases);
1714 * The caller is globally serialized and nobody else
1715 * takes two locks at once, deadlock is not possible.
1717 raw_spin_lock(&new_base->lock);
1718 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1720 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1721 migrate_hrtimer_list(&old_base->clock_base[i],
1722 &new_base->clock_base[i]);
1725 raw_spin_unlock(&old_base->lock);
1726 raw_spin_unlock(&new_base->lock);
1728 /* Check, if we got expired work to do */
1729 __hrtimer_peek_ahead_timers();
1733 #endif /* CONFIG_HOTPLUG_CPU */
1735 static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self,
1736 unsigned long action, void *hcpu)
1738 int scpu = (long)hcpu;
1742 case CPU_UP_PREPARE:
1743 case CPU_UP_PREPARE_FROZEN:
1744 init_hrtimers_cpu(scpu);
1747 #ifdef CONFIG_HOTPLUG_CPU
1749 case CPU_DYING_FROZEN:
1750 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DYING, &scpu);
1753 case CPU_DEAD_FROZEN:
1755 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &scpu);
1756 migrate_hrtimers(scpu);
1768 static struct notifier_block __cpuinitdata hrtimers_nb = {
1769 .notifier_call = hrtimer_cpu_notify,
1772 void __init hrtimers_init(void)
1774 hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
1775 (void *)(long)smp_processor_id());
1776 register_cpu_notifier(&hrtimers_nb);
1777 #ifdef CONFIG_HIGH_RES_TIMERS
1778 open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq);
1783 * schedule_hrtimeout_range_clock - sleep until timeout
1784 * @expires: timeout value (ktime_t)
1785 * @delta: slack in expires timeout (ktime_t)
1786 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1787 * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME
1790 schedule_hrtimeout_range_clock(ktime_t *expires, unsigned long delta,
1791 const enum hrtimer_mode mode, int clock)
1793 struct hrtimer_sleeper t;
1796 * Optimize when a zero timeout value is given. It does not
1797 * matter whether this is an absolute or a relative time.
1799 if (expires && !expires->tv64) {
1800 __set_current_state(TASK_RUNNING);
1805 * A NULL parameter means "infinite"
1809 __set_current_state(TASK_RUNNING);
1813 hrtimer_init_on_stack(&t.timer, clock, mode);
1814 hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
1816 hrtimer_init_sleeper(&t, current);
1818 hrtimer_start_expires(&t.timer, mode);
1819 if (!hrtimer_active(&t.timer))
1825 hrtimer_cancel(&t.timer);
1826 destroy_hrtimer_on_stack(&t.timer);
1828 __set_current_state(TASK_RUNNING);
1830 return !t.task ? 0 : -EINTR;
1834 * schedule_hrtimeout_range - sleep until timeout
1835 * @expires: timeout value (ktime_t)
1836 * @delta: slack in expires timeout (ktime_t)
1837 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1839 * Make the current task sleep until the given expiry time has
1840 * elapsed. The routine will return immediately unless
1841 * the current task state has been set (see set_current_state()).
1843 * The @delta argument gives the kernel the freedom to schedule the
1844 * actual wakeup to a time that is both power and performance friendly.
1845 * The kernel give the normal best effort behavior for "@expires+@delta",
1846 * but may decide to fire the timer earlier, but no earlier than @expires.
1848 * You can set the task state as follows -
1850 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1851 * pass before the routine returns.
1853 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1854 * delivered to the current task.
1856 * The current task state is guaranteed to be TASK_RUNNING when this
1859 * Returns 0 when the timer has expired otherwise -EINTR
1861 int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta,
1862 const enum hrtimer_mode mode)
1864 return schedule_hrtimeout_range_clock(expires, delta, mode,
1867 EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
1870 * schedule_hrtimeout - sleep until timeout
1871 * @expires: timeout value (ktime_t)
1872 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1874 * Make the current task sleep until the given expiry time has
1875 * elapsed. The routine will return immediately unless
1876 * the current task state has been set (see set_current_state()).
1878 * You can set the task state as follows -
1880 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1881 * pass before the routine returns.
1883 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1884 * delivered to the current task.
1886 * The current task state is guaranteed to be TASK_RUNNING when this
1889 * Returns 0 when the timer has expired otherwise -EINTR
1891 int __sched schedule_hrtimeout(ktime_t *expires,
1892 const enum hrtimer_mode mode)
1894 return schedule_hrtimeout_range(expires, 0, mode);
1896 EXPORT_SYMBOL_GPL(schedule_hrtimeout);