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 /* Make sure nsec fits into long */
290 if (unlikely(nsec > KTIME_SEC_MAX))
291 return (ktime_t){ .tv64 = KTIME_MAX };
293 tmp = ktime_set((long)nsec, rem);
296 return ktime_add(kt, tmp);
299 EXPORT_SYMBOL_GPL(ktime_add_ns);
302 * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable
304 * @nsec: the scalar nsec value to subtract
306 * Returns the subtraction of @nsec from @kt in ktime_t format
308 ktime_t ktime_sub_ns(const ktime_t kt, u64 nsec)
312 if (likely(nsec < NSEC_PER_SEC)) {
315 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
317 tmp = ktime_set((long)nsec, rem);
320 return ktime_sub(kt, tmp);
323 EXPORT_SYMBOL_GPL(ktime_sub_ns);
324 # endif /* !CONFIG_KTIME_SCALAR */
327 * Divide a ktime value by a nanosecond value
329 u64 ktime_divns(const ktime_t kt, s64 div)
334 dclc = ktime_to_ns(kt);
335 /* Make sure the divisor is less than 2^32: */
341 do_div(dclc, (unsigned long) div);
345 #endif /* BITS_PER_LONG >= 64 */
348 * Add two ktime values and do a safety check for overflow:
350 ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
352 ktime_t res = ktime_add(lhs, rhs);
355 * We use KTIME_SEC_MAX here, the maximum timeout which we can
356 * return to user space in a timespec:
358 if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64)
359 res = ktime_set(KTIME_SEC_MAX, 0);
364 EXPORT_SYMBOL_GPL(ktime_add_safe);
366 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
368 static struct debug_obj_descr hrtimer_debug_descr;
370 static void *hrtimer_debug_hint(void *addr)
372 return ((struct hrtimer *) addr)->function;
376 * fixup_init is called when:
377 * - an active object is initialized
379 static int hrtimer_fixup_init(void *addr, enum debug_obj_state state)
381 struct hrtimer *timer = addr;
384 case ODEBUG_STATE_ACTIVE:
385 hrtimer_cancel(timer);
386 debug_object_init(timer, &hrtimer_debug_descr);
394 * fixup_activate is called when:
395 * - an active object is activated
396 * - an unknown object is activated (might be a statically initialized object)
398 static int hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
402 case ODEBUG_STATE_NOTAVAILABLE:
406 case ODEBUG_STATE_ACTIVE:
415 * fixup_free is called when:
416 * - an active object is freed
418 static int hrtimer_fixup_free(void *addr, enum debug_obj_state state)
420 struct hrtimer *timer = addr;
423 case ODEBUG_STATE_ACTIVE:
424 hrtimer_cancel(timer);
425 debug_object_free(timer, &hrtimer_debug_descr);
432 static struct debug_obj_descr hrtimer_debug_descr = {
434 .debug_hint = hrtimer_debug_hint,
435 .fixup_init = hrtimer_fixup_init,
436 .fixup_activate = hrtimer_fixup_activate,
437 .fixup_free = hrtimer_fixup_free,
440 static inline void debug_hrtimer_init(struct hrtimer *timer)
442 debug_object_init(timer, &hrtimer_debug_descr);
445 static inline void debug_hrtimer_activate(struct hrtimer *timer)
447 debug_object_activate(timer, &hrtimer_debug_descr);
450 static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
452 debug_object_deactivate(timer, &hrtimer_debug_descr);
455 static inline void debug_hrtimer_free(struct hrtimer *timer)
457 debug_object_free(timer, &hrtimer_debug_descr);
460 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
461 enum hrtimer_mode mode);
463 void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
464 enum hrtimer_mode mode)
466 debug_object_init_on_stack(timer, &hrtimer_debug_descr);
467 __hrtimer_init(timer, clock_id, mode);
469 EXPORT_SYMBOL_GPL(hrtimer_init_on_stack);
471 void destroy_hrtimer_on_stack(struct hrtimer *timer)
473 debug_object_free(timer, &hrtimer_debug_descr);
477 static inline void debug_hrtimer_init(struct hrtimer *timer) { }
478 static inline void debug_hrtimer_activate(struct hrtimer *timer) { }
479 static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
483 debug_init(struct hrtimer *timer, clockid_t clockid,
484 enum hrtimer_mode mode)
486 debug_hrtimer_init(timer);
487 trace_hrtimer_init(timer, clockid, mode);
490 static inline void debug_activate(struct hrtimer *timer)
492 debug_hrtimer_activate(timer);
493 trace_hrtimer_start(timer);
496 static inline void debug_deactivate(struct hrtimer *timer)
498 debug_hrtimer_deactivate(timer);
499 trace_hrtimer_cancel(timer);
502 /* High resolution timer related functions */
503 #ifdef CONFIG_HIGH_RES_TIMERS
506 * High resolution timer enabled ?
508 static int hrtimer_hres_enabled __read_mostly = 1;
511 * Enable / Disable high resolution mode
513 static int __init setup_hrtimer_hres(char *str)
515 if (!strcmp(str, "off"))
516 hrtimer_hres_enabled = 0;
517 else if (!strcmp(str, "on"))
518 hrtimer_hres_enabled = 1;
524 __setup("highres=", setup_hrtimer_hres);
527 * hrtimer_high_res_enabled - query, if the highres mode is enabled
529 static inline int hrtimer_is_hres_enabled(void)
531 return hrtimer_hres_enabled;
535 * Is the high resolution mode active ?
537 static inline int hrtimer_hres_active(void)
539 return __this_cpu_read(hrtimer_bases.hres_active);
543 * Reprogram the event source with checking both queues for the
545 * Called with interrupts disabled and base->lock held
548 hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
551 struct hrtimer_clock_base *base = cpu_base->clock_base;
552 ktime_t expires, expires_next;
554 expires_next.tv64 = KTIME_MAX;
556 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
557 struct hrtimer *timer;
558 struct timerqueue_node *next;
560 next = timerqueue_getnext(&base->active);
563 timer = container_of(next, struct hrtimer, node);
565 expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
567 * clock_was_set() has changed base->offset so the
568 * result might be negative. Fix it up to prevent a
569 * false positive in clockevents_program_event()
571 if (expires.tv64 < 0)
573 if (expires.tv64 < expires_next.tv64)
574 expires_next = expires;
577 if (skip_equal && expires_next.tv64 == cpu_base->expires_next.tv64)
580 cpu_base->expires_next.tv64 = expires_next.tv64;
582 if (cpu_base->expires_next.tv64 != KTIME_MAX)
583 tick_program_event(cpu_base->expires_next, 1);
587 * Shared reprogramming for clock_realtime and clock_monotonic
589 * When a timer is enqueued and expires earlier than the already enqueued
590 * timers, we have to check, whether it expires earlier than the timer for
591 * which the clock event device was armed.
593 * Called with interrupts disabled and base->cpu_base.lock held
595 static int hrtimer_reprogram(struct hrtimer *timer,
596 struct hrtimer_clock_base *base)
598 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
599 ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
602 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
605 * When the callback is running, we do not reprogram the clock event
606 * device. The timer callback is either running on a different CPU or
607 * the callback is executed in the hrtimer_interrupt context. The
608 * reprogramming is handled either by the softirq, which called the
609 * callback or at the end of the hrtimer_interrupt.
611 if (hrtimer_callback_running(timer))
615 * CLOCK_REALTIME timer might be requested with an absolute
616 * expiry time which is less than base->offset. Nothing wrong
617 * about that, just avoid to call into the tick code, which
618 * has now objections against negative expiry values.
620 if (expires.tv64 < 0)
623 if (expires.tv64 >= cpu_base->expires_next.tv64)
627 * If a hang was detected in the last timer interrupt then we
628 * do not schedule a timer which is earlier than the expiry
629 * which we enforced in the hang detection. We want the system
632 if (cpu_base->hang_detected)
636 * Clockevents returns -ETIME, when the event was in the past.
638 res = tick_program_event(expires, 0);
639 if (!IS_ERR_VALUE(res))
640 cpu_base->expires_next = expires;
645 * Initialize the high resolution related parts of cpu_base
647 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
649 base->expires_next.tv64 = KTIME_MAX;
650 base->hres_active = 0;
654 * When High resolution timers are active, try to reprogram. Note, that in case
655 * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry
656 * check happens. The timer gets enqueued into the rbtree. The reprogramming
657 * and expiry check is done in the hrtimer_interrupt or in the softirq.
659 static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
660 struct hrtimer_clock_base *base)
662 return base->cpu_base->hres_active && hrtimer_reprogram(timer, base);
665 static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base)
667 ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset;
668 ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset;
669 ktime_t *offs_tai = &base->clock_base[HRTIMER_BASE_TAI].offset;
671 return ktime_get_update_offsets(offs_real, offs_boot, offs_tai);
675 * Retrigger next event is called after clock was set
677 * Called with interrupts disabled via on_each_cpu()
679 static void retrigger_next_event(void *arg)
681 struct hrtimer_cpu_base *base = &__get_cpu_var(hrtimer_bases);
683 if (!hrtimer_hres_active())
686 raw_spin_lock(&base->lock);
687 hrtimer_update_base(base);
688 hrtimer_force_reprogram(base, 0);
689 raw_spin_unlock(&base->lock);
693 * Switch to high resolution mode
695 static int hrtimer_switch_to_hres(void)
697 int i, cpu = smp_processor_id();
698 struct hrtimer_cpu_base *base = &per_cpu(hrtimer_bases, cpu);
701 if (base->hres_active)
704 local_irq_save(flags);
706 if (tick_init_highres()) {
707 local_irq_restore(flags);
708 printk(KERN_WARNING "Could not switch to high resolution "
709 "mode on CPU %d\n", cpu);
712 base->hres_active = 1;
713 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
714 base->clock_base[i].resolution = KTIME_HIGH_RES;
716 tick_setup_sched_timer();
717 /* "Retrigger" the interrupt to get things going */
718 retrigger_next_event(NULL);
719 local_irq_restore(flags);
724 * Called from timekeeping code to reprogramm the hrtimer interrupt
725 * device. If called from the timer interrupt context we defer it to
728 void clock_was_set_delayed(void)
730 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
732 cpu_base->clock_was_set = 1;
733 __raise_softirq_irqoff(HRTIMER_SOFTIRQ);
738 static inline int hrtimer_hres_active(void) { return 0; }
739 static inline int hrtimer_is_hres_enabled(void) { return 0; }
740 static inline int hrtimer_switch_to_hres(void) { return 0; }
742 hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { }
743 static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
744 struct hrtimer_clock_base *base)
748 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { }
749 static inline void retrigger_next_event(void *arg) { }
751 #endif /* CONFIG_HIGH_RES_TIMERS */
754 * Clock realtime was set
756 * Change the offset of the realtime clock vs. the monotonic
759 * We might have to reprogram the high resolution timer interrupt. On
760 * SMP we call the architecture specific code to retrigger _all_ high
761 * resolution timer interrupts. On UP we just disable interrupts and
762 * call the high resolution interrupt code.
764 void clock_was_set(void)
766 #ifdef CONFIG_HIGH_RES_TIMERS
767 /* Retrigger the CPU local events everywhere */
768 on_each_cpu(retrigger_next_event, NULL, 1);
770 timerfd_clock_was_set();
774 * During resume we might have to reprogram the high resolution timer
775 * interrupt (on the local CPU):
777 void hrtimers_resume(void)
779 WARN_ONCE(!irqs_disabled(),
780 KERN_INFO "hrtimers_resume() called with IRQs enabled!");
782 retrigger_next_event(NULL);
783 timerfd_clock_was_set();
786 static inline void timer_stats_hrtimer_set_start_info(struct hrtimer *timer)
788 #ifdef CONFIG_TIMER_STATS
789 if (timer->start_site)
791 timer->start_site = __builtin_return_address(0);
792 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
793 timer->start_pid = current->pid;
797 static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer *timer)
799 #ifdef CONFIG_TIMER_STATS
800 timer->start_site = NULL;
804 static inline void timer_stats_account_hrtimer(struct hrtimer *timer)
806 #ifdef CONFIG_TIMER_STATS
807 if (likely(!timer_stats_active))
809 timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
810 timer->function, timer->start_comm, 0);
815 * Counterpart to lock_hrtimer_base above:
818 void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
820 raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
824 * hrtimer_forward - forward the timer expiry
825 * @timer: hrtimer to forward
826 * @now: forward past this time
827 * @interval: the interval to forward
829 * Forward the timer expiry so it will expire in the future.
830 * Returns the number of overruns.
832 u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
837 delta = ktime_sub(now, hrtimer_get_expires(timer));
842 if (interval.tv64 < timer->base->resolution.tv64)
843 interval.tv64 = timer->base->resolution.tv64;
845 if (unlikely(delta.tv64 >= interval.tv64)) {
846 s64 incr = ktime_to_ns(interval);
848 orun = ktime_divns(delta, incr);
849 hrtimer_add_expires_ns(timer, incr * orun);
850 if (hrtimer_get_expires_tv64(timer) > now.tv64)
853 * This (and the ktime_add() below) is the
854 * correction for exact:
858 hrtimer_add_expires(timer, interval);
862 EXPORT_SYMBOL_GPL(hrtimer_forward);
865 * enqueue_hrtimer - internal function to (re)start a timer
867 * The timer is inserted in expiry order. Insertion into the
868 * red black tree is O(log(n)). Must hold the base lock.
870 * Returns 1 when the new timer is the leftmost timer in the tree.
872 static int enqueue_hrtimer(struct hrtimer *timer,
873 struct hrtimer_clock_base *base)
875 debug_activate(timer);
877 timerqueue_add(&base->active, &timer->node);
878 base->cpu_base->active_bases |= 1 << base->index;
881 * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
882 * state of a possibly running callback.
884 timer->state |= HRTIMER_STATE_ENQUEUED;
886 return (&timer->node == base->active.next);
890 * __remove_hrtimer - internal function to remove a timer
892 * Caller must hold the base lock.
894 * High resolution timer mode reprograms the clock event device when the
895 * timer is the one which expires next. The caller can disable this by setting
896 * reprogram to zero. This is useful, when the context does a reprogramming
897 * anyway (e.g. timer interrupt)
899 static void __remove_hrtimer(struct hrtimer *timer,
900 struct hrtimer_clock_base *base,
901 unsigned long newstate, int reprogram)
903 struct timerqueue_node *next_timer;
904 if (!(timer->state & HRTIMER_STATE_ENQUEUED))
907 next_timer = timerqueue_getnext(&base->active);
908 timerqueue_del(&base->active, &timer->node);
909 if (&timer->node == next_timer) {
910 #ifdef CONFIG_HIGH_RES_TIMERS
911 /* Reprogram the clock event device. if enabled */
912 if (reprogram && hrtimer_hres_active()) {
915 expires = ktime_sub(hrtimer_get_expires(timer),
917 if (base->cpu_base->expires_next.tv64 == expires.tv64)
918 hrtimer_force_reprogram(base->cpu_base, 1);
922 if (!timerqueue_getnext(&base->active))
923 base->cpu_base->active_bases &= ~(1 << base->index);
925 timer->state = newstate;
929 * remove hrtimer, called with base lock held
932 remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base)
934 if (hrtimer_is_queued(timer)) {
939 * Remove the timer and force reprogramming when high
940 * resolution mode is active and the timer is on the current
941 * CPU. If we remove a timer on another CPU, reprogramming is
942 * skipped. The interrupt event on this CPU is fired and
943 * reprogramming happens in the interrupt handler. This is a
944 * rare case and less expensive than a smp call.
946 debug_deactivate(timer);
947 timer_stats_hrtimer_clear_start_info(timer);
948 reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases);
950 * We must preserve the CALLBACK state flag here,
951 * otherwise we could move the timer base in
952 * switch_hrtimer_base.
954 state = timer->state & HRTIMER_STATE_CALLBACK;
955 __remove_hrtimer(timer, base, state, reprogram);
961 int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
962 unsigned long delta_ns, const enum hrtimer_mode mode,
965 struct hrtimer_clock_base *base, *new_base;
969 base = lock_hrtimer_base(timer, &flags);
971 /* Remove an active timer from the queue: */
972 ret = remove_hrtimer(timer, base);
974 /* Switch the timer base, if necessary: */
975 new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED);
977 if (mode & HRTIMER_MODE_REL) {
978 tim = ktime_add_safe(tim, new_base->get_time());
980 * CONFIG_TIME_LOW_RES is a temporary way for architectures
981 * to signal that they simply return xtime in
982 * do_gettimeoffset(). In this case we want to round up by
983 * resolution when starting a relative timer, to avoid short
984 * timeouts. This will go away with the GTOD framework.
986 #ifdef CONFIG_TIME_LOW_RES
987 tim = ktime_add_safe(tim, base->resolution);
991 hrtimer_set_expires_range_ns(timer, tim, delta_ns);
993 timer_stats_hrtimer_set_start_info(timer);
995 leftmost = enqueue_hrtimer(timer, new_base);
998 * Only allow reprogramming if the new base is on this CPU.
999 * (it might still be on another CPU if the timer was pending)
1001 * XXX send_remote_softirq() ?
1003 if (leftmost && new_base->cpu_base == &__get_cpu_var(hrtimer_bases)
1004 && hrtimer_enqueue_reprogram(timer, new_base)) {
1007 * We need to drop cpu_base->lock to avoid a
1008 * lock ordering issue vs. rq->lock.
1010 raw_spin_unlock(&new_base->cpu_base->lock);
1011 raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1012 local_irq_restore(flags);
1015 __raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1019 unlock_hrtimer_base(timer, &flags);
1025 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
1026 * @timer: the timer to be added
1028 * @delta_ns: "slack" range for the timer
1029 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
1030 * relative (HRTIMER_MODE_REL)
1034 * 1 when the timer was active
1036 int hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1037 unsigned long delta_ns, const enum hrtimer_mode mode)
1039 return __hrtimer_start_range_ns(timer, tim, delta_ns, mode, 1);
1041 EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
1044 * hrtimer_start - (re)start an hrtimer on the current CPU
1045 * @timer: the timer to be added
1047 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
1048 * relative (HRTIMER_MODE_REL)
1052 * 1 when the timer was active
1055 hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode)
1057 return __hrtimer_start_range_ns(timer, tim, 0, mode, 1);
1059 EXPORT_SYMBOL_GPL(hrtimer_start);
1063 * hrtimer_try_to_cancel - try to deactivate a timer
1064 * @timer: hrtimer to stop
1067 * 0 when the timer was not active
1068 * 1 when the timer was active
1069 * -1 when the timer is currently excuting the callback function and
1072 int hrtimer_try_to_cancel(struct hrtimer *timer)
1074 struct hrtimer_clock_base *base;
1075 unsigned long flags;
1078 base = lock_hrtimer_base(timer, &flags);
1080 if (!hrtimer_callback_running(timer))
1081 ret = remove_hrtimer(timer, base);
1083 unlock_hrtimer_base(timer, &flags);
1088 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1091 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1092 * @timer: the timer to be cancelled
1095 * 0 when the timer was not active
1096 * 1 when the timer was active
1098 int hrtimer_cancel(struct hrtimer *timer)
1101 int ret = hrtimer_try_to_cancel(timer);
1108 EXPORT_SYMBOL_GPL(hrtimer_cancel);
1111 * hrtimer_get_remaining - get remaining time for the timer
1112 * @timer: the timer to read
1114 ktime_t hrtimer_get_remaining(const struct hrtimer *timer)
1116 unsigned long flags;
1119 lock_hrtimer_base(timer, &flags);
1120 rem = hrtimer_expires_remaining(timer);
1121 unlock_hrtimer_base(timer, &flags);
1125 EXPORT_SYMBOL_GPL(hrtimer_get_remaining);
1129 * hrtimer_get_next_event - get the time until next expiry event
1131 * Returns the delta to the next expiry event or KTIME_MAX if no timer
1134 ktime_t hrtimer_get_next_event(void)
1136 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1137 struct hrtimer_clock_base *base = cpu_base->clock_base;
1138 ktime_t delta, mindelta = { .tv64 = KTIME_MAX };
1139 unsigned long flags;
1142 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1144 if (!hrtimer_hres_active()) {
1145 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
1146 struct hrtimer *timer;
1147 struct timerqueue_node *next;
1149 next = timerqueue_getnext(&base->active);
1153 timer = container_of(next, struct hrtimer, node);
1154 delta.tv64 = hrtimer_get_expires_tv64(timer);
1155 delta = ktime_sub(delta, base->get_time());
1156 if (delta.tv64 < mindelta.tv64)
1157 mindelta.tv64 = delta.tv64;
1161 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1163 if (mindelta.tv64 < 0)
1169 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1170 enum hrtimer_mode mode)
1172 struct hrtimer_cpu_base *cpu_base;
1175 memset(timer, 0, sizeof(struct hrtimer));
1177 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1179 if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
1180 clock_id = CLOCK_MONOTONIC;
1182 base = hrtimer_clockid_to_base(clock_id);
1183 timer->base = &cpu_base->clock_base[base];
1184 timerqueue_init(&timer->node);
1186 #ifdef CONFIG_TIMER_STATS
1187 timer->start_site = NULL;
1188 timer->start_pid = -1;
1189 memset(timer->start_comm, 0, TASK_COMM_LEN);
1194 * hrtimer_init - initialize a timer to the given clock
1195 * @timer: the timer to be initialized
1196 * @clock_id: the clock to be used
1197 * @mode: timer mode abs/rel
1199 void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1200 enum hrtimer_mode mode)
1202 debug_init(timer, clock_id, mode);
1203 __hrtimer_init(timer, clock_id, mode);
1205 EXPORT_SYMBOL_GPL(hrtimer_init);
1208 * hrtimer_get_res - get the timer resolution for a clock
1209 * @which_clock: which clock to query
1210 * @tp: pointer to timespec variable to store the resolution
1212 * Store the resolution of the clock selected by @which_clock in the
1213 * variable pointed to by @tp.
1215 int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp)
1217 struct hrtimer_cpu_base *cpu_base;
1218 int base = hrtimer_clockid_to_base(which_clock);
1220 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1221 *tp = ktime_to_timespec(cpu_base->clock_base[base].resolution);
1225 EXPORT_SYMBOL_GPL(hrtimer_get_res);
1227 static void __run_hrtimer(struct hrtimer *timer, ktime_t *now)
1229 struct hrtimer_clock_base *base = timer->base;
1230 struct hrtimer_cpu_base *cpu_base = base->cpu_base;
1231 enum hrtimer_restart (*fn)(struct hrtimer *);
1234 WARN_ON(!irqs_disabled());
1236 debug_deactivate(timer);
1237 __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0);
1238 timer_stats_account_hrtimer(timer);
1239 fn = timer->function;
1242 * Because we run timers from hardirq context, there is no chance
1243 * they get migrated to another cpu, therefore its safe to unlock
1246 raw_spin_unlock(&cpu_base->lock);
1247 trace_hrtimer_expire_entry(timer, now);
1248 restart = fn(timer);
1249 trace_hrtimer_expire_exit(timer);
1250 raw_spin_lock(&cpu_base->lock);
1253 * Note: We clear the CALLBACK bit after enqueue_hrtimer and
1254 * we do not reprogramm the event hardware. Happens either in
1255 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1257 if (restart != HRTIMER_NORESTART) {
1258 BUG_ON(timer->state != HRTIMER_STATE_CALLBACK);
1259 enqueue_hrtimer(timer, base);
1262 WARN_ON_ONCE(!(timer->state & HRTIMER_STATE_CALLBACK));
1264 timer->state &= ~HRTIMER_STATE_CALLBACK;
1267 #ifdef CONFIG_HIGH_RES_TIMERS
1270 * High resolution timer interrupt
1271 * Called with interrupts disabled
1273 void hrtimer_interrupt(struct clock_event_device *dev)
1275 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1276 ktime_t expires_next, now, entry_time, delta;
1279 BUG_ON(!cpu_base->hres_active);
1280 cpu_base->nr_events++;
1281 dev->next_event.tv64 = KTIME_MAX;
1283 raw_spin_lock(&cpu_base->lock);
1284 entry_time = now = hrtimer_update_base(cpu_base);
1286 expires_next.tv64 = KTIME_MAX;
1288 * We set expires_next to KTIME_MAX here with cpu_base->lock
1289 * held to prevent that a timer is enqueued in our queue via
1290 * the migration code. This does not affect enqueueing of
1291 * timers which run their callback and need to be requeued on
1294 cpu_base->expires_next.tv64 = KTIME_MAX;
1296 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1297 struct hrtimer_clock_base *base;
1298 struct timerqueue_node *node;
1301 if (!(cpu_base->active_bases & (1 << i)))
1304 base = cpu_base->clock_base + i;
1305 basenow = ktime_add(now, base->offset);
1307 while ((node = timerqueue_getnext(&base->active))) {
1308 struct hrtimer *timer;
1310 timer = container_of(node, struct hrtimer, node);
1313 * The immediate goal for using the softexpires is
1314 * minimizing wakeups, not running timers at the
1315 * earliest interrupt after their soft expiration.
1316 * This allows us to avoid using a Priority Search
1317 * Tree, which can answer a stabbing querry for
1318 * overlapping intervals and instead use the simple
1319 * BST we already have.
1320 * We don't add extra wakeups by delaying timers that
1321 * are right-of a not yet expired timer, because that
1322 * timer will have to trigger a wakeup anyway.
1325 if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer)) {
1328 expires = ktime_sub(hrtimer_get_expires(timer),
1330 if (expires.tv64 < 0)
1331 expires.tv64 = KTIME_MAX;
1332 if (expires.tv64 < expires_next.tv64)
1333 expires_next = expires;
1337 __run_hrtimer(timer, &basenow);
1342 * Store the new expiry value so the migration code can verify
1345 cpu_base->expires_next = expires_next;
1346 raw_spin_unlock(&cpu_base->lock);
1348 /* Reprogramming necessary ? */
1349 if (expires_next.tv64 == KTIME_MAX ||
1350 !tick_program_event(expires_next, 0)) {
1351 cpu_base->hang_detected = 0;
1356 * The next timer was already expired due to:
1358 * - long lasting callbacks
1359 * - being scheduled away when running in a VM
1361 * We need to prevent that we loop forever in the hrtimer
1362 * interrupt routine. We give it 3 attempts to avoid
1363 * overreacting on some spurious event.
1365 * Acquire base lock for updating the offsets and retrieving
1368 raw_spin_lock(&cpu_base->lock);
1369 now = hrtimer_update_base(cpu_base);
1370 cpu_base->nr_retries++;
1374 * Give the system a chance to do something else than looping
1375 * here. We stored the entry time, so we know exactly how long
1376 * we spent here. We schedule the next event this amount of
1379 cpu_base->nr_hangs++;
1380 cpu_base->hang_detected = 1;
1381 raw_spin_unlock(&cpu_base->lock);
1382 delta = ktime_sub(now, entry_time);
1383 if (delta.tv64 > cpu_base->max_hang_time.tv64)
1384 cpu_base->max_hang_time = delta;
1386 * Limit it to a sensible value as we enforce a longer
1387 * delay. Give the CPU at least 100ms to catch up.
1389 if (delta.tv64 > 100 * NSEC_PER_MSEC)
1390 expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1392 expires_next = ktime_add(now, delta);
1393 tick_program_event(expires_next, 1);
1394 printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n",
1395 ktime_to_ns(delta));
1399 * local version of hrtimer_peek_ahead_timers() called with interrupts
1402 static void __hrtimer_peek_ahead_timers(void)
1404 struct tick_device *td;
1406 if (!hrtimer_hres_active())
1409 td = &__get_cpu_var(tick_cpu_device);
1410 if (td && td->evtdev)
1411 hrtimer_interrupt(td->evtdev);
1415 * hrtimer_peek_ahead_timers -- run soft-expired timers now
1417 * hrtimer_peek_ahead_timers will peek at the timer queue of
1418 * the current cpu and check if there are any timers for which
1419 * the soft expires time has passed. If any such timers exist,
1420 * they are run immediately and then removed from the timer queue.
1423 void hrtimer_peek_ahead_timers(void)
1425 unsigned long flags;
1427 local_irq_save(flags);
1428 __hrtimer_peek_ahead_timers();
1429 local_irq_restore(flags);
1432 static void run_hrtimer_softirq(struct softirq_action *h)
1434 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1436 if (cpu_base->clock_was_set) {
1437 cpu_base->clock_was_set = 0;
1441 hrtimer_peek_ahead_timers();
1444 #else /* CONFIG_HIGH_RES_TIMERS */
1446 static inline void __hrtimer_peek_ahead_timers(void) { }
1448 #endif /* !CONFIG_HIGH_RES_TIMERS */
1451 * Called from timer softirq every jiffy, expire hrtimers:
1453 * For HRT its the fall back code to run the softirq in the timer
1454 * softirq context in case the hrtimer initialization failed or has
1455 * not been done yet.
1457 void hrtimer_run_pending(void)
1459 if (hrtimer_hres_active())
1463 * This _is_ ugly: We have to check in the softirq context,
1464 * whether we can switch to highres and / or nohz mode. The
1465 * clocksource switch happens in the timer interrupt with
1466 * xtime_lock held. Notification from there only sets the
1467 * check bit in the tick_oneshot code, otherwise we might
1468 * deadlock vs. xtime_lock.
1470 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1471 hrtimer_switch_to_hres();
1475 * Called from hardirq context every jiffy
1477 void hrtimer_run_queues(void)
1479 struct timerqueue_node *node;
1480 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1481 struct hrtimer_clock_base *base;
1482 int index, gettime = 1;
1484 if (hrtimer_hres_active())
1487 for (index = 0; index < HRTIMER_MAX_CLOCK_BASES; index++) {
1488 base = &cpu_base->clock_base[index];
1489 if (!timerqueue_getnext(&base->active))
1493 hrtimer_get_softirq_time(cpu_base);
1497 raw_spin_lock(&cpu_base->lock);
1499 while ((node = timerqueue_getnext(&base->active))) {
1500 struct hrtimer *timer;
1502 timer = container_of(node, struct hrtimer, node);
1503 if (base->softirq_time.tv64 <=
1504 hrtimer_get_expires_tv64(timer))
1507 __run_hrtimer(timer, &base->softirq_time);
1509 raw_spin_unlock(&cpu_base->lock);
1514 * Sleep related functions:
1516 static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1518 struct hrtimer_sleeper *t =
1519 container_of(timer, struct hrtimer_sleeper, timer);
1520 struct task_struct *task = t->task;
1524 wake_up_process(task);
1526 return HRTIMER_NORESTART;
1529 void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1531 sl->timer.function = hrtimer_wakeup;
1534 EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
1536 static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1538 hrtimer_init_sleeper(t, current);
1541 set_current_state(TASK_INTERRUPTIBLE);
1542 hrtimer_start_expires(&t->timer, mode);
1543 if (!hrtimer_active(&t->timer))
1546 if (likely(t->task))
1549 hrtimer_cancel(&t->timer);
1550 mode = HRTIMER_MODE_ABS;
1552 } while (t->task && !signal_pending(current));
1554 __set_current_state(TASK_RUNNING);
1556 return t->task == NULL;
1559 static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp)
1561 struct timespec rmt;
1564 rem = hrtimer_expires_remaining(timer);
1567 rmt = ktime_to_timespec(rem);
1569 if (copy_to_user(rmtp, &rmt, sizeof(*rmtp)))
1575 long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1577 struct hrtimer_sleeper t;
1578 struct timespec __user *rmtp;
1581 hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid,
1583 hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
1585 if (do_nanosleep(&t, HRTIMER_MODE_ABS))
1588 rmtp = restart->nanosleep.rmtp;
1590 ret = update_rmtp(&t.timer, rmtp);
1595 /* The other values in restart are already filled in */
1596 ret = -ERESTART_RESTARTBLOCK;
1598 destroy_hrtimer_on_stack(&t.timer);
1602 long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
1603 const enum hrtimer_mode mode, const clockid_t clockid)
1605 struct restart_block *restart;
1606 struct hrtimer_sleeper t;
1608 unsigned long slack;
1610 slack = current->timer_slack_ns;
1611 if (rt_task(current))
1614 hrtimer_init_on_stack(&t.timer, clockid, mode);
1615 hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack);
1616 if (do_nanosleep(&t, mode))
1619 /* Absolute timers do not update the rmtp value and restart: */
1620 if (mode == HRTIMER_MODE_ABS) {
1621 ret = -ERESTARTNOHAND;
1626 ret = update_rmtp(&t.timer, rmtp);
1631 restart = ¤t_thread_info()->restart_block;
1632 restart->fn = hrtimer_nanosleep_restart;
1633 restart->nanosleep.clockid = t.timer.base->clockid;
1634 restart->nanosleep.rmtp = rmtp;
1635 restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
1637 ret = -ERESTART_RESTARTBLOCK;
1639 destroy_hrtimer_on_stack(&t.timer);
1643 SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp,
1644 struct timespec __user *, rmtp)
1648 if (copy_from_user(&tu, rqtp, sizeof(tu)))
1651 if (!timespec_valid(&tu))
1654 return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1658 * Functions related to boot-time initialization:
1660 static void __cpuinit init_hrtimers_cpu(int cpu)
1662 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1665 raw_spin_lock_init(&cpu_base->lock);
1667 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1668 cpu_base->clock_base[i].cpu_base = cpu_base;
1669 timerqueue_init_head(&cpu_base->clock_base[i].active);
1672 hrtimer_init_hres(cpu_base);
1675 #ifdef CONFIG_HOTPLUG_CPU
1677 static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1678 struct hrtimer_clock_base *new_base)
1680 struct hrtimer *timer;
1681 struct timerqueue_node *node;
1683 while ((node = timerqueue_getnext(&old_base->active))) {
1684 timer = container_of(node, struct hrtimer, node);
1685 BUG_ON(hrtimer_callback_running(timer));
1686 debug_deactivate(timer);
1689 * Mark it as STATE_MIGRATE not INACTIVE otherwise the
1690 * timer could be seen as !active and just vanish away
1691 * under us on another CPU
1693 __remove_hrtimer(timer, old_base, HRTIMER_STATE_MIGRATE, 0);
1694 timer->base = new_base;
1696 * Enqueue the timers on the new cpu. This does not
1697 * reprogram the event device in case the timer
1698 * expires before the earliest on this CPU, but we run
1699 * hrtimer_interrupt after we migrated everything to
1700 * sort out already expired timers and reprogram the
1703 enqueue_hrtimer(timer, new_base);
1705 /* Clear the migration state bit */
1706 timer->state &= ~HRTIMER_STATE_MIGRATE;
1710 static void migrate_hrtimers(int scpu)
1712 struct hrtimer_cpu_base *old_base, *new_base;
1715 BUG_ON(cpu_online(scpu));
1716 tick_cancel_sched_timer(scpu);
1718 local_irq_disable();
1719 old_base = &per_cpu(hrtimer_bases, scpu);
1720 new_base = &__get_cpu_var(hrtimer_bases);
1722 * The caller is globally serialized and nobody else
1723 * takes two locks at once, deadlock is not possible.
1725 raw_spin_lock(&new_base->lock);
1726 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1728 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1729 migrate_hrtimer_list(&old_base->clock_base[i],
1730 &new_base->clock_base[i]);
1733 raw_spin_unlock(&old_base->lock);
1734 raw_spin_unlock(&new_base->lock);
1736 /* Check, if we got expired work to do */
1737 __hrtimer_peek_ahead_timers();
1741 #endif /* CONFIG_HOTPLUG_CPU */
1743 static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self,
1744 unsigned long action, void *hcpu)
1746 int scpu = (long)hcpu;
1750 case CPU_UP_PREPARE:
1751 case CPU_UP_PREPARE_FROZEN:
1752 init_hrtimers_cpu(scpu);
1755 #ifdef CONFIG_HOTPLUG_CPU
1757 case CPU_DYING_FROZEN:
1758 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DYING, &scpu);
1761 case CPU_DEAD_FROZEN:
1763 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &scpu);
1764 migrate_hrtimers(scpu);
1776 static struct notifier_block __cpuinitdata hrtimers_nb = {
1777 .notifier_call = hrtimer_cpu_notify,
1780 void __init hrtimers_init(void)
1782 hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
1783 (void *)(long)smp_processor_id());
1784 register_cpu_notifier(&hrtimers_nb);
1785 #ifdef CONFIG_HIGH_RES_TIMERS
1786 open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq);
1791 * schedule_hrtimeout_range_clock - sleep until timeout
1792 * @expires: timeout value (ktime_t)
1793 * @delta: slack in expires timeout (ktime_t)
1794 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1795 * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME
1798 schedule_hrtimeout_range_clock(ktime_t *expires, unsigned long delta,
1799 const enum hrtimer_mode mode, int clock)
1801 struct hrtimer_sleeper t;
1804 * Optimize when a zero timeout value is given. It does not
1805 * matter whether this is an absolute or a relative time.
1807 if (expires && !expires->tv64) {
1808 __set_current_state(TASK_RUNNING);
1813 * A NULL parameter means "infinite"
1817 __set_current_state(TASK_RUNNING);
1821 hrtimer_init_on_stack(&t.timer, clock, mode);
1822 hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
1824 hrtimer_init_sleeper(&t, current);
1826 hrtimer_start_expires(&t.timer, mode);
1827 if (!hrtimer_active(&t.timer))
1833 hrtimer_cancel(&t.timer);
1834 destroy_hrtimer_on_stack(&t.timer);
1836 __set_current_state(TASK_RUNNING);
1838 return !t.task ? 0 : -EINTR;
1842 * schedule_hrtimeout_range - sleep until timeout
1843 * @expires: timeout value (ktime_t)
1844 * @delta: slack in expires timeout (ktime_t)
1845 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1847 * Make the current task sleep until the given expiry time has
1848 * elapsed. The routine will return immediately unless
1849 * the current task state has been set (see set_current_state()).
1851 * The @delta argument gives the kernel the freedom to schedule the
1852 * actual wakeup to a time that is both power and performance friendly.
1853 * The kernel give the normal best effort behavior for "@expires+@delta",
1854 * but may decide to fire the timer earlier, but no earlier than @expires.
1856 * You can set the task state as follows -
1858 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1859 * pass before the routine returns.
1861 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1862 * delivered to the current task.
1864 * The current task state is guaranteed to be TASK_RUNNING when this
1867 * Returns 0 when the timer has expired otherwise -EINTR
1869 int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta,
1870 const enum hrtimer_mode mode)
1872 return schedule_hrtimeout_range_clock(expires, delta, mode,
1875 EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
1878 * schedule_hrtimeout - sleep until timeout
1879 * @expires: timeout value (ktime_t)
1880 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1882 * Make the current task sleep until the given expiry time has
1883 * elapsed. The routine will return immediately unless
1884 * the current task state has been set (see set_current_state()).
1886 * You can set the task state as follows -
1888 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1889 * pass before the routine returns.
1891 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1892 * delivered to the current task.
1894 * The current task state is guaranteed to be TASK_RUNNING when this
1897 * Returns 0 when the timer has expired otherwise -EINTR
1899 int __sched schedule_hrtimeout(ktime_t *expires,
1900 const enum hrtimer_mode mode)
1902 return schedule_hrtimeout_range(expires, 0, mode);
1904 EXPORT_SYMBOL_GPL(schedule_hrtimeout);