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/irq.h>
36 #include <linux/module.h>
37 #include <linux/percpu.h>
38 #include <linux/hrtimer.h>
39 #include <linux/notifier.h>
40 #include <linux/syscalls.h>
41 #include <linux/kallsyms.h>
42 #include <linux/interrupt.h>
43 #include <linux/tick.h>
44 #include <linux/seq_file.h>
45 #include <linux/err.h>
46 #include <linux/debugobjects.h>
48 #include <asm/uaccess.h>
51 * ktime_get - get the monotonic time in ktime_t format
53 * returns the time in ktime_t format
55 ktime_t ktime_get(void)
61 return timespec_to_ktime(now);
63 EXPORT_SYMBOL_GPL(ktime_get);
66 * ktime_get_real - get the real (wall-) time in ktime_t format
68 * returns the time in ktime_t format
70 ktime_t ktime_get_real(void)
76 return timespec_to_ktime(now);
79 EXPORT_SYMBOL_GPL(ktime_get_real);
84 * Note: If we want to add new timer bases, we have to skip the two
85 * clock ids captured by the cpu-timers. We do this by holding empty
86 * entries rather than doing math adjustment of the clock ids.
87 * This ensures that we capture erroneous accesses to these clock ids
88 * rather than moving them into the range of valid clock id's.
90 DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
96 .index = CLOCK_REALTIME,
97 .get_time = &ktime_get_real,
98 .resolution = KTIME_LOW_RES,
101 .index = CLOCK_MONOTONIC,
102 .get_time = &ktime_get,
103 .resolution = KTIME_LOW_RES,
109 * ktime_get_ts - get the monotonic clock in timespec format
110 * @ts: pointer to timespec variable
112 * The function calculates the monotonic clock from the realtime
113 * clock and the wall_to_monotonic offset and stores the result
114 * in normalized timespec format in the variable pointed to by @ts.
116 void ktime_get_ts(struct timespec *ts)
118 struct timespec tomono;
122 seq = read_seqbegin(&xtime_lock);
124 tomono = wall_to_monotonic;
126 } while (read_seqretry(&xtime_lock, seq));
128 set_normalized_timespec(ts, ts->tv_sec + tomono.tv_sec,
129 ts->tv_nsec + tomono.tv_nsec);
131 EXPORT_SYMBOL_GPL(ktime_get_ts);
134 * Get the coarse grained time at the softirq based on xtime and
137 static void hrtimer_get_softirq_time(struct hrtimer_cpu_base *base)
139 ktime_t xtim, tomono;
140 struct timespec xts, tom;
144 seq = read_seqbegin(&xtime_lock);
145 xts = current_kernel_time();
146 tom = wall_to_monotonic;
147 } while (read_seqretry(&xtime_lock, seq));
149 xtim = timespec_to_ktime(xts);
150 tomono = timespec_to_ktime(tom);
151 base->clock_base[CLOCK_REALTIME].softirq_time = xtim;
152 base->clock_base[CLOCK_MONOTONIC].softirq_time =
153 ktime_add(xtim, tomono);
157 * Functions and macros which are different for UP/SMP systems are kept in a
163 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
164 * means that all timers which are tied to this base via timer->base are
165 * locked, and the base itself is locked too.
167 * So __run_timers/migrate_timers can safely modify all timers which could
168 * be found on the lists/queues.
170 * When the timer's base is locked, and the timer removed from list, it is
171 * possible to set timer->base = NULL and drop the lock: the timer remains
175 struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
176 unsigned long *flags)
178 struct hrtimer_clock_base *base;
182 if (likely(base != NULL)) {
183 spin_lock_irqsave(&base->cpu_base->lock, *flags);
184 if (likely(base == timer->base))
186 /* The timer has migrated to another CPU: */
187 spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
194 * Switch the timer base to the current CPU when possible.
196 static inline struct hrtimer_clock_base *
197 switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base)
199 struct hrtimer_clock_base *new_base;
200 struct hrtimer_cpu_base *new_cpu_base;
202 new_cpu_base = &__get_cpu_var(hrtimer_bases);
203 new_base = &new_cpu_base->clock_base[base->index];
205 if (base != new_base) {
207 * We are trying to schedule the timer on the local CPU.
208 * However we can't change timer's base while it is running,
209 * so we keep it on the same CPU. No hassle vs. reprogramming
210 * the event source in the high resolution case. The softirq
211 * code will take care of this when the timer function has
212 * completed. There is no conflict as we hold the lock until
213 * the timer is enqueued.
215 if (unlikely(hrtimer_callback_running(timer)))
218 /* See the comment in lock_timer_base() */
220 spin_unlock(&base->cpu_base->lock);
221 spin_lock(&new_base->cpu_base->lock);
222 timer->base = new_base;
227 #else /* CONFIG_SMP */
229 static inline struct hrtimer_clock_base *
230 lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
232 struct hrtimer_clock_base *base = timer->base;
234 spin_lock_irqsave(&base->cpu_base->lock, *flags);
239 # define switch_hrtimer_base(t, b) (b)
241 #endif /* !CONFIG_SMP */
244 * Functions for the union type storage format of ktime_t which are
245 * too large for inlining:
247 #if BITS_PER_LONG < 64
248 # ifndef CONFIG_KTIME_SCALAR
250 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
252 * @nsec: the scalar nsec value to add
254 * Returns the sum of kt and nsec in ktime_t format
256 ktime_t ktime_add_ns(const ktime_t kt, u64 nsec)
260 if (likely(nsec < NSEC_PER_SEC)) {
263 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
265 tmp = ktime_set((long)nsec, rem);
268 return ktime_add(kt, tmp);
271 EXPORT_SYMBOL_GPL(ktime_add_ns);
274 * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable
276 * @nsec: the scalar nsec value to subtract
278 * Returns the subtraction of @nsec from @kt in ktime_t format
280 ktime_t ktime_sub_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_sub(kt, tmp);
295 EXPORT_SYMBOL_GPL(ktime_sub_ns);
296 # endif /* !CONFIG_KTIME_SCALAR */
299 * Divide a ktime value by a nanosecond value
301 u64 ktime_divns(const ktime_t kt, s64 div)
306 dclc = ktime_to_ns(kt);
307 /* Make sure the divisor is less than 2^32: */
313 do_div(dclc, (unsigned long) div);
317 #endif /* BITS_PER_LONG >= 64 */
320 * Add two ktime values and do a safety check for overflow:
322 ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
324 ktime_t res = ktime_add(lhs, rhs);
327 * We use KTIME_SEC_MAX here, the maximum timeout which we can
328 * return to user space in a timespec:
330 if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64)
331 res = ktime_set(KTIME_SEC_MAX, 0);
336 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
338 static struct debug_obj_descr hrtimer_debug_descr;
341 * fixup_init is called when:
342 * - an active object is initialized
344 static int hrtimer_fixup_init(void *addr, enum debug_obj_state state)
346 struct hrtimer *timer = addr;
349 case ODEBUG_STATE_ACTIVE:
350 hrtimer_cancel(timer);
351 debug_object_init(timer, &hrtimer_debug_descr);
359 * fixup_activate is called when:
360 * - an active object is activated
361 * - an unknown object is activated (might be a statically initialized object)
363 static int hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
367 case ODEBUG_STATE_NOTAVAILABLE:
371 case ODEBUG_STATE_ACTIVE:
380 * fixup_free is called when:
381 * - an active object is freed
383 static int hrtimer_fixup_free(void *addr, enum debug_obj_state state)
385 struct hrtimer *timer = addr;
388 case ODEBUG_STATE_ACTIVE:
389 hrtimer_cancel(timer);
390 debug_object_free(timer, &hrtimer_debug_descr);
397 static struct debug_obj_descr hrtimer_debug_descr = {
399 .fixup_init = hrtimer_fixup_init,
400 .fixup_activate = hrtimer_fixup_activate,
401 .fixup_free = hrtimer_fixup_free,
404 static inline void debug_hrtimer_init(struct hrtimer *timer)
406 debug_object_init(timer, &hrtimer_debug_descr);
409 static inline void debug_hrtimer_activate(struct hrtimer *timer)
411 debug_object_activate(timer, &hrtimer_debug_descr);
414 static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
416 debug_object_deactivate(timer, &hrtimer_debug_descr);
419 static inline void debug_hrtimer_free(struct hrtimer *timer)
421 debug_object_free(timer, &hrtimer_debug_descr);
424 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
425 enum hrtimer_mode mode);
427 void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
428 enum hrtimer_mode mode)
430 debug_object_init_on_stack(timer, &hrtimer_debug_descr);
431 __hrtimer_init(timer, clock_id, mode);
434 void destroy_hrtimer_on_stack(struct hrtimer *timer)
436 debug_object_free(timer, &hrtimer_debug_descr);
440 static inline void debug_hrtimer_init(struct hrtimer *timer) { }
441 static inline void debug_hrtimer_activate(struct hrtimer *timer) { }
442 static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
446 * Check, whether the timer is on the callback pending list
448 static inline int hrtimer_cb_pending(const struct hrtimer *timer)
450 return timer->state & HRTIMER_STATE_PENDING;
454 * Remove a timer from the callback pending list
456 static inline void hrtimer_remove_cb_pending(struct hrtimer *timer)
458 list_del_init(&timer->cb_entry);
461 /* High resolution timer related functions */
462 #ifdef CONFIG_HIGH_RES_TIMERS
465 * High resolution timer enabled ?
467 static int hrtimer_hres_enabled __read_mostly = 1;
470 * Enable / Disable high resolution mode
472 static int __init setup_hrtimer_hres(char *str)
474 if (!strcmp(str, "off"))
475 hrtimer_hres_enabled = 0;
476 else if (!strcmp(str, "on"))
477 hrtimer_hres_enabled = 1;
483 __setup("highres=", setup_hrtimer_hres);
486 * hrtimer_high_res_enabled - query, if the highres mode is enabled
488 static inline int hrtimer_is_hres_enabled(void)
490 return hrtimer_hres_enabled;
494 * Is the high resolution mode active ?
496 static inline int hrtimer_hres_active(void)
498 return __get_cpu_var(hrtimer_bases).hres_active;
502 * Reprogram the event source with checking both queues for the
504 * Called with interrupts disabled and base->lock held
506 static void hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base)
509 struct hrtimer_clock_base *base = cpu_base->clock_base;
512 cpu_base->expires_next.tv64 = KTIME_MAX;
514 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
515 struct hrtimer *timer;
519 timer = rb_entry(base->first, struct hrtimer, node);
520 expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
521 if (expires.tv64 < cpu_base->expires_next.tv64)
522 cpu_base->expires_next = expires;
525 if (cpu_base->expires_next.tv64 != KTIME_MAX)
526 tick_program_event(cpu_base->expires_next, 1);
530 * Shared reprogramming for clock_realtime and clock_monotonic
532 * When a timer is enqueued and expires earlier than the already enqueued
533 * timers, we have to check, whether it expires earlier than the timer for
534 * which the clock event device was armed.
536 * Called with interrupts disabled and base->cpu_base.lock held
538 static int hrtimer_reprogram(struct hrtimer *timer,
539 struct hrtimer_clock_base *base)
541 ktime_t *expires_next = &__get_cpu_var(hrtimer_bases).expires_next;
542 ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
545 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
548 * When the callback is running, we do not reprogram the clock event
549 * device. The timer callback is either running on a different CPU or
550 * the callback is executed in the hrtimer_interrupt context. The
551 * reprogramming is handled either by the softirq, which called the
552 * callback or at the end of the hrtimer_interrupt.
554 if (hrtimer_callback_running(timer))
558 * CLOCK_REALTIME timer might be requested with an absolute
559 * expiry time which is less than base->offset. Nothing wrong
560 * about that, just avoid to call into the tick code, which
561 * has now objections against negative expiry values.
563 if (expires.tv64 < 0)
566 if (expires.tv64 >= expires_next->tv64)
570 * Clockevents returns -ETIME, when the event was in the past.
572 res = tick_program_event(expires, 0);
573 if (!IS_ERR_VALUE(res))
574 *expires_next = expires;
580 * Retrigger next event is called after clock was set
582 * Called with interrupts disabled via on_each_cpu()
584 static void retrigger_next_event(void *arg)
586 struct hrtimer_cpu_base *base;
587 struct timespec realtime_offset;
590 if (!hrtimer_hres_active())
594 seq = read_seqbegin(&xtime_lock);
595 set_normalized_timespec(&realtime_offset,
596 -wall_to_monotonic.tv_sec,
597 -wall_to_monotonic.tv_nsec);
598 } while (read_seqretry(&xtime_lock, seq));
600 base = &__get_cpu_var(hrtimer_bases);
602 /* Adjust CLOCK_REALTIME offset */
603 spin_lock(&base->lock);
604 base->clock_base[CLOCK_REALTIME].offset =
605 timespec_to_ktime(realtime_offset);
607 hrtimer_force_reprogram(base);
608 spin_unlock(&base->lock);
612 * Clock realtime was set
614 * Change the offset of the realtime clock vs. the monotonic
617 * We might have to reprogram the high resolution timer interrupt. On
618 * SMP we call the architecture specific code to retrigger _all_ high
619 * resolution timer interrupts. On UP we just disable interrupts and
620 * call the high resolution interrupt code.
622 void clock_was_set(void)
624 /* Retrigger the CPU local events everywhere */
625 on_each_cpu(retrigger_next_event, NULL, 1);
629 * During resume we might have to reprogram the high resolution timer
630 * interrupt (on the local CPU):
632 void hres_timers_resume(void)
634 /* Retrigger the CPU local events: */
635 retrigger_next_event(NULL);
639 * Initialize the high resolution related parts of cpu_base
641 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
643 base->expires_next.tv64 = KTIME_MAX;
644 base->hres_active = 0;
648 * Initialize the high resolution related parts of a hrtimer
650 static inline void hrtimer_init_timer_hres(struct hrtimer *timer)
655 * When High resolution timers are active, try to reprogram. Note, that in case
656 * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry
657 * check happens. The timer gets enqueued into the rbtree. The reprogramming
658 * and expiry check is done in the hrtimer_interrupt or in the softirq.
660 static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
661 struct hrtimer_clock_base *base)
663 if (base->cpu_base->hres_active && hrtimer_reprogram(timer, base)) {
665 /* Timer is expired, act upon the callback mode */
666 switch(timer->cb_mode) {
667 case HRTIMER_CB_IRQSAFE_PERCPU:
668 case HRTIMER_CB_IRQSAFE_UNLOCKED:
670 * This is solely for the sched tick emulation with
671 * dynamic tick support to ensure that we do not
672 * restart the tick right on the edge and end up with
673 * the tick timer in the softirq ! The calling site
674 * takes care of this. Also used for hrtimer sleeper !
676 debug_hrtimer_deactivate(timer);
678 case HRTIMER_CB_SOFTIRQ:
680 * Move everything else into the softirq pending list !
682 list_add_tail(&timer->cb_entry,
683 &base->cpu_base->cb_pending);
684 timer->state = HRTIMER_STATE_PENDING;
694 * Switch to high resolution mode
696 static int hrtimer_switch_to_hres(void)
698 int cpu = smp_processor_id();
699 struct hrtimer_cpu_base *base = &per_cpu(hrtimer_bases, cpu);
702 if (base->hres_active)
705 local_irq_save(flags);
707 if (tick_init_highres()) {
708 local_irq_restore(flags);
709 printk(KERN_WARNING "Could not switch to high resolution "
710 "mode on CPU %d\n", cpu);
713 base->hres_active = 1;
714 base->clock_base[CLOCK_REALTIME].resolution = KTIME_HIGH_RES;
715 base->clock_base[CLOCK_MONOTONIC].resolution = KTIME_HIGH_RES;
717 tick_setup_sched_timer();
719 /* "Retrigger" the interrupt to get things going */
720 retrigger_next_event(NULL);
721 local_irq_restore(flags);
722 printk(KERN_DEBUG "Switched to high resolution mode on CPU %d\n",
727 static inline void hrtimer_raise_softirq(void)
729 raise_softirq(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; }
737 static inline void hrtimer_force_reprogram(struct hrtimer_cpu_base *base) { }
738 static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
739 struct hrtimer_clock_base *base)
743 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { }
744 static inline void hrtimer_init_timer_hres(struct hrtimer *timer) { }
745 static inline int hrtimer_reprogram(struct hrtimer *timer,
746 struct hrtimer_clock_base *base)
750 static inline void hrtimer_raise_softirq(void) { }
752 #endif /* CONFIG_HIGH_RES_TIMERS */
754 #ifdef CONFIG_TIMER_STATS
755 void __timer_stats_hrtimer_set_start_info(struct hrtimer *timer, void *addr)
757 if (timer->start_site)
760 timer->start_site = addr;
761 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
762 timer->start_pid = current->pid;
767 * Counterpart to lock_hrtimer_base above:
770 void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
772 spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
776 * hrtimer_forward - forward the timer expiry
777 * @timer: hrtimer to forward
778 * @now: forward past this time
779 * @interval: the interval to forward
781 * Forward the timer expiry so it will expire in the future.
782 * Returns the number of overruns.
784 u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
789 delta = ktime_sub(now, hrtimer_get_expires(timer));
794 if (interval.tv64 < timer->base->resolution.tv64)
795 interval.tv64 = timer->base->resolution.tv64;
797 if (unlikely(delta.tv64 >= interval.tv64)) {
798 s64 incr = ktime_to_ns(interval);
800 orun = ktime_divns(delta, incr);
801 hrtimer_add_expires_ns(timer, incr * orun);
802 if (hrtimer_get_expires_tv64(timer) > now.tv64)
805 * This (and the ktime_add() below) is the
806 * correction for exact:
810 hrtimer_add_expires(timer, interval);
814 EXPORT_SYMBOL_GPL(hrtimer_forward);
817 * enqueue_hrtimer - internal function to (re)start a timer
819 * The timer is inserted in expiry order. Insertion into the
820 * red black tree is O(log(n)). Must hold the base lock.
822 static void enqueue_hrtimer(struct hrtimer *timer,
823 struct hrtimer_clock_base *base, int reprogram)
825 struct rb_node **link = &base->active.rb_node;
826 struct rb_node *parent = NULL;
827 struct hrtimer *entry;
830 debug_hrtimer_activate(timer);
833 * Find the right place in the rbtree:
837 entry = rb_entry(parent, struct hrtimer, node);
839 * We dont care about collisions. Nodes with
840 * the same expiry time stay together.
842 if (hrtimer_get_expires_tv64(timer) <
843 hrtimer_get_expires_tv64(entry)) {
844 link = &(*link)->rb_left;
846 link = &(*link)->rb_right;
852 * Insert the timer to the rbtree and check whether it
853 * replaces the first pending timer
857 * Reprogram the clock event device. When the timer is already
858 * expired hrtimer_enqueue_reprogram has either called the
859 * callback or added it to the pending list and raised the
862 * This is a NOP for !HIGHRES
864 if (reprogram && hrtimer_enqueue_reprogram(timer, base))
867 base->first = &timer->node;
870 rb_link_node(&timer->node, parent, link);
871 rb_insert_color(&timer->node, &base->active);
873 * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
874 * state of a possibly running callback.
876 timer->state |= HRTIMER_STATE_ENQUEUED;
880 * __remove_hrtimer - internal function to remove a timer
882 * Caller must hold the base lock.
884 * High resolution timer mode reprograms the clock event device when the
885 * timer is the one which expires next. The caller can disable this by setting
886 * reprogram to zero. This is useful, when the context does a reprogramming
887 * anyway (e.g. timer interrupt)
889 static void __remove_hrtimer(struct hrtimer *timer,
890 struct hrtimer_clock_base *base,
891 unsigned long newstate, int reprogram)
893 /* High res. callback list. NOP for !HIGHRES */
894 if (hrtimer_cb_pending(timer))
895 hrtimer_remove_cb_pending(timer);
898 * Remove the timer from the rbtree and replace the
899 * first entry pointer if necessary.
901 if (base->first == &timer->node) {
902 base->first = rb_next(&timer->node);
903 /* Reprogram the clock event device. if enabled */
904 if (reprogram && hrtimer_hres_active())
905 hrtimer_force_reprogram(base->cpu_base);
907 rb_erase(&timer->node, &base->active);
909 timer->state = newstate;
913 * remove hrtimer, called with base lock held
916 remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base)
918 if (hrtimer_is_queued(timer)) {
922 * Remove the timer and force reprogramming when high
923 * resolution mode is active and the timer is on the current
924 * CPU. If we remove a timer on another CPU, reprogramming is
925 * skipped. The interrupt event on this CPU is fired and
926 * reprogramming happens in the interrupt handler. This is a
927 * rare case and less expensive than a smp call.
929 debug_hrtimer_deactivate(timer);
930 timer_stats_hrtimer_clear_start_info(timer);
931 reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases);
932 __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE,
940 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
941 * @timer: the timer to be added
943 * @delta_ns: "slack" range for the timer
944 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
948 * 1 when the timer was active
951 hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, unsigned long delta_ns,
952 const enum hrtimer_mode mode)
954 struct hrtimer_clock_base *base, *new_base;
958 base = lock_hrtimer_base(timer, &flags);
960 /* Remove an active timer from the queue: */
961 ret = remove_hrtimer(timer, base);
963 /* Switch the timer base, if necessary: */
964 new_base = switch_hrtimer_base(timer, base);
966 if (mode == HRTIMER_MODE_REL) {
967 tim = ktime_add_safe(tim, new_base->get_time());
969 * CONFIG_TIME_LOW_RES is a temporary way for architectures
970 * to signal that they simply return xtime in
971 * do_gettimeoffset(). In this case we want to round up by
972 * resolution when starting a relative timer, to avoid short
973 * timeouts. This will go away with the GTOD framework.
975 #ifdef CONFIG_TIME_LOW_RES
976 tim = ktime_add_safe(tim, base->resolution);
980 hrtimer_set_expires_range_ns(timer, tim, delta_ns);
982 timer_stats_hrtimer_set_start_info(timer);
985 * Only allow reprogramming if the new base is on this CPU.
986 * (it might still be on another CPU if the timer was pending)
988 enqueue_hrtimer(timer, new_base,
989 new_base->cpu_base == &__get_cpu_var(hrtimer_bases));
992 * The timer may be expired and moved to the cb_pending
993 * list. We can not raise the softirq with base lock held due
994 * to a possible deadlock with runqueue lock.
996 raise = timer->state == HRTIMER_STATE_PENDING;
999 * We use preempt_disable to prevent this task from migrating after
1000 * setting up the softirq and raising it. Otherwise, if me migrate
1001 * we will raise the softirq on the wrong CPU.
1005 unlock_hrtimer_base(timer, &flags);
1008 hrtimer_raise_softirq();
1013 EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
1016 * hrtimer_start - (re)start an hrtimer on the current CPU
1017 * @timer: the timer to be added
1019 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
1023 * 1 when the timer was active
1026 hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode)
1028 return hrtimer_start_range_ns(timer, tim, 0, mode);
1030 EXPORT_SYMBOL_GPL(hrtimer_start);
1034 * hrtimer_try_to_cancel - try to deactivate a timer
1035 * @timer: hrtimer to stop
1038 * 0 when the timer was not active
1039 * 1 when the timer was active
1040 * -1 when the timer is currently excuting the callback function and
1043 int hrtimer_try_to_cancel(struct hrtimer *timer)
1045 struct hrtimer_clock_base *base;
1046 unsigned long flags;
1049 base = lock_hrtimer_base(timer, &flags);
1051 if (!hrtimer_callback_running(timer))
1052 ret = remove_hrtimer(timer, base);
1054 unlock_hrtimer_base(timer, &flags);
1059 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1062 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1063 * @timer: the timer to be cancelled
1066 * 0 when the timer was not active
1067 * 1 when the timer was active
1069 int hrtimer_cancel(struct hrtimer *timer)
1072 int ret = hrtimer_try_to_cancel(timer);
1079 EXPORT_SYMBOL_GPL(hrtimer_cancel);
1082 * hrtimer_get_remaining - get remaining time for the timer
1083 * @timer: the timer to read
1085 ktime_t hrtimer_get_remaining(const struct hrtimer *timer)
1087 struct hrtimer_clock_base *base;
1088 unsigned long flags;
1091 base = lock_hrtimer_base(timer, &flags);
1092 rem = hrtimer_expires_remaining(timer);
1093 unlock_hrtimer_base(timer, &flags);
1097 EXPORT_SYMBOL_GPL(hrtimer_get_remaining);
1101 * hrtimer_get_next_event - get the time until next expiry event
1103 * Returns the delta to the next expiry event or KTIME_MAX if no timer
1106 ktime_t hrtimer_get_next_event(void)
1108 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1109 struct hrtimer_clock_base *base = cpu_base->clock_base;
1110 ktime_t delta, mindelta = { .tv64 = KTIME_MAX };
1111 unsigned long flags;
1114 spin_lock_irqsave(&cpu_base->lock, flags);
1116 if (!hrtimer_hres_active()) {
1117 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
1118 struct hrtimer *timer;
1123 timer = rb_entry(base->first, struct hrtimer, node);
1124 delta.tv64 = hrtimer_get_expires_tv64(timer);
1125 delta = ktime_sub(delta, base->get_time());
1126 if (delta.tv64 < mindelta.tv64)
1127 mindelta.tv64 = delta.tv64;
1131 spin_unlock_irqrestore(&cpu_base->lock, flags);
1133 if (mindelta.tv64 < 0)
1139 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1140 enum hrtimer_mode mode)
1142 struct hrtimer_cpu_base *cpu_base;
1144 memset(timer, 0, sizeof(struct hrtimer));
1146 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1148 if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
1149 clock_id = CLOCK_MONOTONIC;
1151 timer->base = &cpu_base->clock_base[clock_id];
1152 INIT_LIST_HEAD(&timer->cb_entry);
1153 hrtimer_init_timer_hres(timer);
1155 #ifdef CONFIG_TIMER_STATS
1156 timer->start_site = NULL;
1157 timer->start_pid = -1;
1158 memset(timer->start_comm, 0, TASK_COMM_LEN);
1163 * hrtimer_init - initialize a timer to the given clock
1164 * @timer: the timer to be initialized
1165 * @clock_id: the clock to be used
1166 * @mode: timer mode abs/rel
1168 void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1169 enum hrtimer_mode mode)
1171 debug_hrtimer_init(timer);
1172 __hrtimer_init(timer, clock_id, mode);
1174 EXPORT_SYMBOL_GPL(hrtimer_init);
1177 * hrtimer_get_res - get the timer resolution for a clock
1178 * @which_clock: which clock to query
1179 * @tp: pointer to timespec variable to store the resolution
1181 * Store the resolution of the clock selected by @which_clock in the
1182 * variable pointed to by @tp.
1184 int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp)
1186 struct hrtimer_cpu_base *cpu_base;
1188 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1189 *tp = ktime_to_timespec(cpu_base->clock_base[which_clock].resolution);
1193 EXPORT_SYMBOL_GPL(hrtimer_get_res);
1195 static void run_hrtimer_pending(struct hrtimer_cpu_base *cpu_base)
1197 spin_lock_irq(&cpu_base->lock);
1199 while (!list_empty(&cpu_base->cb_pending)) {
1200 enum hrtimer_restart (*fn)(struct hrtimer *);
1201 struct hrtimer *timer;
1203 int emulate_hardirq_ctx = 0;
1205 timer = list_entry(cpu_base->cb_pending.next,
1206 struct hrtimer, cb_entry);
1208 debug_hrtimer_deactivate(timer);
1209 timer_stats_account_hrtimer(timer);
1211 fn = timer->function;
1213 * A timer might have been added to the cb_pending list
1214 * when it was migrated during a cpu-offline operation.
1215 * Emulate hardirq context for such timers.
1217 if (timer->cb_mode == HRTIMER_CB_IRQSAFE_PERCPU ||
1218 timer->cb_mode == HRTIMER_CB_IRQSAFE_UNLOCKED)
1219 emulate_hardirq_ctx = 1;
1221 __remove_hrtimer(timer, timer->base, HRTIMER_STATE_CALLBACK, 0);
1222 spin_unlock_irq(&cpu_base->lock);
1224 if (unlikely(emulate_hardirq_ctx)) {
1225 local_irq_disable();
1226 restart = fn(timer);
1229 restart = fn(timer);
1231 spin_lock_irq(&cpu_base->lock);
1233 timer->state &= ~HRTIMER_STATE_CALLBACK;
1234 if (restart == HRTIMER_RESTART) {
1235 BUG_ON(hrtimer_active(timer));
1237 * Enqueue the timer, allow reprogramming of the event
1240 enqueue_hrtimer(timer, timer->base, 1);
1241 } else if (hrtimer_active(timer)) {
1243 * If the timer was rearmed on another CPU, reprogram
1246 struct hrtimer_clock_base *base = timer->base;
1248 if (base->first == &timer->node &&
1249 hrtimer_reprogram(timer, base)) {
1251 * Timer is expired. Thus move it from tree to
1252 * pending list again.
1254 __remove_hrtimer(timer, base,
1255 HRTIMER_STATE_PENDING, 0);
1256 list_add_tail(&timer->cb_entry,
1257 &base->cpu_base->cb_pending);
1261 spin_unlock_irq(&cpu_base->lock);
1264 static void __run_hrtimer(struct hrtimer *timer)
1266 struct hrtimer_clock_base *base = timer->base;
1267 struct hrtimer_cpu_base *cpu_base = base->cpu_base;
1268 enum hrtimer_restart (*fn)(struct hrtimer *);
1271 debug_hrtimer_deactivate(timer);
1272 __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0);
1273 timer_stats_account_hrtimer(timer);
1275 fn = timer->function;
1276 if (timer->cb_mode == HRTIMER_CB_IRQSAFE_PERCPU ||
1277 timer->cb_mode == HRTIMER_CB_IRQSAFE_UNLOCKED) {
1279 * Used for scheduler timers, avoid lock inversion with
1280 * rq->lock and tasklist_lock.
1282 * These timers are required to deal with enqueue expiry
1283 * themselves and are not allowed to migrate.
1285 spin_unlock(&cpu_base->lock);
1286 restart = fn(timer);
1287 spin_lock(&cpu_base->lock);
1289 restart = fn(timer);
1292 * Note: We clear the CALLBACK bit after enqueue_hrtimer to avoid
1293 * reprogramming of the event hardware. This happens at the end of this
1296 if (restart != HRTIMER_NORESTART) {
1297 BUG_ON(timer->state != HRTIMER_STATE_CALLBACK);
1298 enqueue_hrtimer(timer, base, 0);
1300 timer->state &= ~HRTIMER_STATE_CALLBACK;
1303 #ifdef CONFIG_HIGH_RES_TIMERS
1306 * High resolution timer interrupt
1307 * Called with interrupts disabled
1309 void hrtimer_interrupt(struct clock_event_device *dev)
1311 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1312 struct hrtimer_clock_base *base;
1313 ktime_t expires_next, now;
1316 BUG_ON(!cpu_base->hres_active);
1317 cpu_base->nr_events++;
1318 dev->next_event.tv64 = KTIME_MAX;
1323 expires_next.tv64 = KTIME_MAX;
1325 base = cpu_base->clock_base;
1327 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1329 struct rb_node *node;
1331 spin_lock(&cpu_base->lock);
1333 basenow = ktime_add(now, base->offset);
1335 while ((node = base->first)) {
1336 struct hrtimer *timer;
1338 timer = rb_entry(node, struct hrtimer, node);
1341 * The immediate goal for using the softexpires is
1342 * minimizing wakeups, not running timers at the
1343 * earliest interrupt after their soft expiration.
1344 * This allows us to avoid using a Priority Search
1345 * Tree, which can answer a stabbing querry for
1346 * overlapping intervals and instead use the simple
1347 * BST we already have.
1348 * We don't add extra wakeups by delaying timers that
1349 * are right-of a not yet expired timer, because that
1350 * timer will have to trigger a wakeup anyway.
1353 if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer)) {
1356 expires = ktime_sub(hrtimer_get_expires(timer),
1358 if (expires.tv64 < expires_next.tv64)
1359 expires_next = expires;
1363 /* Move softirq callbacks to the pending list */
1364 if (timer->cb_mode == HRTIMER_CB_SOFTIRQ) {
1365 __remove_hrtimer(timer, base,
1366 HRTIMER_STATE_PENDING, 0);
1367 list_add_tail(&timer->cb_entry,
1368 &base->cpu_base->cb_pending);
1373 __run_hrtimer(timer);
1375 spin_unlock(&cpu_base->lock);
1379 cpu_base->expires_next = expires_next;
1381 /* Reprogramming necessary ? */
1382 if (expires_next.tv64 != KTIME_MAX) {
1383 if (tick_program_event(expires_next, 0))
1387 /* Raise softirq ? */
1389 raise_softirq(HRTIMER_SOFTIRQ);
1393 * hrtimer_peek_ahead_timers -- run soft-expired timers now
1395 * hrtimer_peek_ahead_timers will peek at the timer queue of
1396 * the current cpu and check if there are any timers for which
1397 * the soft expires time has passed. If any such timers exist,
1398 * they are run immediately and then removed from the timer queue.
1401 void hrtimer_peek_ahead_timers(void)
1403 struct tick_device *td;
1404 unsigned long flags;
1406 if (!hrtimer_hres_active())
1409 local_irq_save(flags);
1410 td = &__get_cpu_var(tick_cpu_device);
1411 if (td && td->evtdev)
1412 hrtimer_interrupt(td->evtdev);
1413 local_irq_restore(flags);
1416 static void run_hrtimer_softirq(struct softirq_action *h)
1418 run_hrtimer_pending(&__get_cpu_var(hrtimer_bases));
1421 #endif /* CONFIG_HIGH_RES_TIMERS */
1424 * Called from timer softirq every jiffy, expire hrtimers:
1426 * For HRT its the fall back code to run the softirq in the timer
1427 * softirq context in case the hrtimer initialization failed or has
1428 * not been done yet.
1430 void hrtimer_run_pending(void)
1432 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1434 if (hrtimer_hres_active())
1438 * This _is_ ugly: We have to check in the softirq context,
1439 * whether we can switch to highres and / or nohz mode. The
1440 * clocksource switch happens in the timer interrupt with
1441 * xtime_lock held. Notification from there only sets the
1442 * check bit in the tick_oneshot code, otherwise we might
1443 * deadlock vs. xtime_lock.
1445 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1446 hrtimer_switch_to_hres();
1448 run_hrtimer_pending(cpu_base);
1452 * Called from hardirq context every jiffy
1454 void hrtimer_run_queues(void)
1456 struct rb_node *node;
1457 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1458 struct hrtimer_clock_base *base;
1459 int index, gettime = 1;
1461 if (hrtimer_hres_active())
1464 for (index = 0; index < HRTIMER_MAX_CLOCK_BASES; index++) {
1465 base = &cpu_base->clock_base[index];
1471 hrtimer_get_softirq_time(cpu_base);
1475 spin_lock(&cpu_base->lock);
1477 while ((node = base->first)) {
1478 struct hrtimer *timer;
1480 timer = rb_entry(node, struct hrtimer, node);
1481 if (base->softirq_time.tv64 <=
1482 hrtimer_get_expires_tv64(timer))
1485 if (timer->cb_mode == HRTIMER_CB_SOFTIRQ) {
1486 __remove_hrtimer(timer, base,
1487 HRTIMER_STATE_PENDING, 0);
1488 list_add_tail(&timer->cb_entry,
1489 &base->cpu_base->cb_pending);
1493 __run_hrtimer(timer);
1495 spin_unlock(&cpu_base->lock);
1500 * Sleep related functions:
1502 static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1504 struct hrtimer_sleeper *t =
1505 container_of(timer, struct hrtimer_sleeper, timer);
1506 struct task_struct *task = t->task;
1510 wake_up_process(task);
1512 return HRTIMER_NORESTART;
1515 void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1517 sl->timer.function = hrtimer_wakeup;
1519 #ifdef CONFIG_HIGH_RES_TIMERS
1520 sl->timer.cb_mode = HRTIMER_CB_IRQSAFE_UNLOCKED;
1524 static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1526 hrtimer_init_sleeper(t, current);
1529 set_current_state(TASK_INTERRUPTIBLE);
1530 hrtimer_start_expires(&t->timer, mode);
1531 if (!hrtimer_active(&t->timer))
1534 if (likely(t->task))
1537 hrtimer_cancel(&t->timer);
1538 mode = HRTIMER_MODE_ABS;
1540 } while (t->task && !signal_pending(current));
1542 __set_current_state(TASK_RUNNING);
1544 return t->task == NULL;
1547 static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp)
1549 struct timespec rmt;
1552 rem = hrtimer_expires_remaining(timer);
1555 rmt = ktime_to_timespec(rem);
1557 if (copy_to_user(rmtp, &rmt, sizeof(*rmtp)))
1563 long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1565 struct hrtimer_sleeper t;
1566 struct timespec __user *rmtp;
1569 hrtimer_init_on_stack(&t.timer, restart->nanosleep.index,
1571 hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
1573 if (do_nanosleep(&t, HRTIMER_MODE_ABS))
1576 rmtp = restart->nanosleep.rmtp;
1578 ret = update_rmtp(&t.timer, rmtp);
1583 /* The other values in restart are already filled in */
1584 ret = -ERESTART_RESTARTBLOCK;
1586 destroy_hrtimer_on_stack(&t.timer);
1590 long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
1591 const enum hrtimer_mode mode, const clockid_t clockid)
1593 struct restart_block *restart;
1594 struct hrtimer_sleeper t;
1596 unsigned long slack;
1598 slack = current->timer_slack_ns;
1599 if (rt_task(current))
1602 hrtimer_init_on_stack(&t.timer, clockid, mode);
1603 hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack);
1604 if (do_nanosleep(&t, mode))
1607 /* Absolute timers do not update the rmtp value and restart: */
1608 if (mode == HRTIMER_MODE_ABS) {
1609 ret = -ERESTARTNOHAND;
1614 ret = update_rmtp(&t.timer, rmtp);
1619 restart = ¤t_thread_info()->restart_block;
1620 restart->fn = hrtimer_nanosleep_restart;
1621 restart->nanosleep.index = t.timer.base->index;
1622 restart->nanosleep.rmtp = rmtp;
1623 restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
1625 ret = -ERESTART_RESTARTBLOCK;
1627 destroy_hrtimer_on_stack(&t.timer);
1632 sys_nanosleep(struct timespec __user *rqtp, struct timespec __user *rmtp)
1636 if (copy_from_user(&tu, rqtp, sizeof(tu)))
1639 if (!timespec_valid(&tu))
1642 return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1646 * Functions related to boot-time initialization:
1648 static void __cpuinit init_hrtimers_cpu(int cpu)
1650 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1653 spin_lock_init(&cpu_base->lock);
1655 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
1656 cpu_base->clock_base[i].cpu_base = cpu_base;
1658 INIT_LIST_HEAD(&cpu_base->cb_pending);
1659 hrtimer_init_hres(cpu_base);
1662 #ifdef CONFIG_HOTPLUG_CPU
1664 static int migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1665 struct hrtimer_clock_base *new_base, int dcpu)
1667 struct hrtimer *timer;
1668 struct rb_node *node;
1671 while ((node = rb_first(&old_base->active))) {
1672 timer = rb_entry(node, struct hrtimer, node);
1673 BUG_ON(hrtimer_callback_running(timer));
1674 debug_hrtimer_deactivate(timer);
1677 * Should not happen. Per CPU timers should be
1678 * canceled _before_ the migration code is called
1680 if (timer->cb_mode == HRTIMER_CB_IRQSAFE_PERCPU) {
1681 __remove_hrtimer(timer, old_base,
1682 HRTIMER_STATE_INACTIVE, 0);
1683 WARN(1, "hrtimer (%p %p)active but cpu %d dead\n",
1684 timer, timer->function, dcpu);
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 timer. Allow reprogramming of the event device
1698 enqueue_hrtimer(timer, new_base, 1);
1700 #ifdef CONFIG_HIGH_RES_TIMERS
1702 * Happens with high res enabled when the timer was
1703 * already expired and the callback mode is
1704 * HRTIMER_CB_IRQSAFE_UNLOCKED (hrtimer_sleeper). The
1705 * enqueue code does not move them to the soft irq
1706 * pending list for performance/latency reasons, but
1707 * in the migration state, we need to do that
1708 * otherwise we end up with a stale timer.
1710 if (timer->state == HRTIMER_STATE_MIGRATE) {
1711 timer->state = HRTIMER_STATE_PENDING;
1712 list_add_tail(&timer->cb_entry,
1713 &new_base->cpu_base->cb_pending);
1717 /* Clear the migration state bit */
1718 timer->state &= ~HRTIMER_STATE_MIGRATE;
1723 #ifdef CONFIG_HIGH_RES_TIMERS
1724 static int migrate_hrtimer_pending(struct hrtimer_cpu_base *old_base,
1725 struct hrtimer_cpu_base *new_base)
1727 struct hrtimer *timer;
1730 while (!list_empty(&old_base->cb_pending)) {
1731 timer = list_entry(old_base->cb_pending.next,
1732 struct hrtimer, cb_entry);
1734 __remove_hrtimer(timer, timer->base, HRTIMER_STATE_PENDING, 0);
1735 timer->base = &new_base->clock_base[timer->base->index];
1736 list_add_tail(&timer->cb_entry, &new_base->cb_pending);
1742 static int migrate_hrtimer_pending(struct hrtimer_cpu_base *old_base,
1743 struct hrtimer_cpu_base *new_base)
1749 static void migrate_hrtimers(int cpu)
1751 struct hrtimer_cpu_base *old_base, *new_base;
1754 BUG_ON(cpu_online(cpu));
1755 old_base = &per_cpu(hrtimer_bases, cpu);
1756 new_base = &get_cpu_var(hrtimer_bases);
1758 tick_cancel_sched_timer(cpu);
1760 * The caller is globally serialized and nobody else
1761 * takes two locks at once, deadlock is not possible.
1763 spin_lock_irq(&new_base->lock);
1764 spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1766 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1767 if (migrate_hrtimer_list(&old_base->clock_base[i],
1768 &new_base->clock_base[i], cpu))
1772 if (migrate_hrtimer_pending(old_base, new_base))
1775 spin_unlock(&old_base->lock);
1776 spin_unlock_irq(&new_base->lock);
1777 put_cpu_var(hrtimer_bases);
1780 hrtimer_raise_softirq();
1782 #endif /* CONFIG_HOTPLUG_CPU */
1784 static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self,
1785 unsigned long action, void *hcpu)
1787 unsigned int cpu = (long)hcpu;
1791 case CPU_UP_PREPARE:
1792 case CPU_UP_PREPARE_FROZEN:
1793 init_hrtimers_cpu(cpu);
1796 #ifdef CONFIG_HOTPLUG_CPU
1798 case CPU_DEAD_FROZEN:
1799 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &cpu);
1800 migrate_hrtimers(cpu);
1811 static struct notifier_block __cpuinitdata hrtimers_nb = {
1812 .notifier_call = hrtimer_cpu_notify,
1815 void __init hrtimers_init(void)
1817 hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
1818 (void *)(long)smp_processor_id());
1819 register_cpu_notifier(&hrtimers_nb);
1820 #ifdef CONFIG_HIGH_RES_TIMERS
1821 open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq);
1826 * schedule_hrtimeout_range - sleep until timeout
1827 * @expires: timeout value (ktime_t)
1828 * @delta: slack in expires timeout (ktime_t)
1829 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1831 * Make the current task sleep until the given expiry time has
1832 * elapsed. The routine will return immediately unless
1833 * the current task state has been set (see set_current_state()).
1835 * The @delta argument gives the kernel the freedom to schedule the
1836 * actual wakeup to a time that is both power and performance friendly.
1837 * The kernel give the normal best effort behavior for "@expires+@delta",
1838 * but may decide to fire the timer earlier, but no earlier than @expires.
1840 * You can set the task state as follows -
1842 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1843 * pass before the routine returns.
1845 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1846 * delivered to the current task.
1848 * The current task state is guaranteed to be TASK_RUNNING when this
1851 * Returns 0 when the timer has expired otherwise -EINTR
1853 int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta,
1854 const enum hrtimer_mode mode)
1856 struct hrtimer_sleeper t;
1859 * Optimize when a zero timeout value is given. It does not
1860 * matter whether this is an absolute or a relative time.
1862 if (expires && !expires->tv64) {
1863 __set_current_state(TASK_RUNNING);
1868 * A NULL parameter means "inifinte"
1872 __set_current_state(TASK_RUNNING);
1876 hrtimer_init_on_stack(&t.timer, CLOCK_MONOTONIC, mode);
1877 hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
1879 hrtimer_init_sleeper(&t, current);
1881 hrtimer_start_expires(&t.timer, mode);
1882 if (!hrtimer_active(&t.timer))
1888 hrtimer_cancel(&t.timer);
1889 destroy_hrtimer_on_stack(&t.timer);
1891 __set_current_state(TASK_RUNNING);
1893 return !t.task ? 0 : -EINTR;
1895 EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
1898 * schedule_hrtimeout - sleep until timeout
1899 * @expires: timeout value (ktime_t)
1900 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1902 * Make the current task sleep until the given expiry time has
1903 * elapsed. The routine will return immediately unless
1904 * the current task state has been set (see set_current_state()).
1906 * You can set the task state as follows -
1908 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1909 * pass before the routine returns.
1911 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1912 * delivered to the current task.
1914 * The current task state is guaranteed to be TASK_RUNNING when this
1917 * Returns 0 when the timer has expired otherwise -EINTR
1919 int __sched schedule_hrtimeout(ktime_t *expires,
1920 const enum hrtimer_mode mode)
1922 return schedule_hrtimeout_range(expires, 0, mode);
1924 EXPORT_SYMBOL_GPL(schedule_hrtimeout);