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/module.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>
47 #include <asm/uaccess.h>
50 * ktime_get - get the monotonic time in ktime_t format
52 * returns the time in ktime_t format
54 ktime_t ktime_get(void)
60 return timespec_to_ktime(now);
62 EXPORT_SYMBOL_GPL(ktime_get);
65 * ktime_get_real - get the real (wall-) time in ktime_t format
67 * returns the time in ktime_t format
69 ktime_t ktime_get_real(void)
75 return timespec_to_ktime(now);
78 EXPORT_SYMBOL_GPL(ktime_get_real);
83 * Note: If we want to add new timer bases, we have to skip the two
84 * clock ids captured by the cpu-timers. We do this by holding empty
85 * entries rather than doing math adjustment of the clock ids.
86 * This ensures that we capture erroneous accesses to these clock ids
87 * rather than moving them into the range of valid clock id's.
89 DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
95 .index = CLOCK_REALTIME,
96 .get_time = &ktime_get_real,
97 .resolution = KTIME_LOW_RES,
100 .index = CLOCK_MONOTONIC,
101 .get_time = &ktime_get,
102 .resolution = KTIME_LOW_RES,
108 * ktime_get_ts - get the monotonic clock in timespec format
109 * @ts: pointer to timespec variable
111 * The function calculates the monotonic clock from the realtime
112 * clock and the wall_to_monotonic offset and stores the result
113 * in normalized timespec format in the variable pointed to by @ts.
115 void ktime_get_ts(struct timespec *ts)
117 struct timespec tomono;
121 seq = read_seqbegin(&xtime_lock);
123 tomono = wall_to_monotonic;
125 } while (read_seqretry(&xtime_lock, seq));
127 set_normalized_timespec(ts, ts->tv_sec + tomono.tv_sec,
128 ts->tv_nsec + tomono.tv_nsec);
130 EXPORT_SYMBOL_GPL(ktime_get_ts);
133 * Get the coarse grained time at the softirq based on xtime and
136 static void hrtimer_get_softirq_time(struct hrtimer_cpu_base *base)
138 ktime_t xtim, tomono;
139 struct timespec xts, tom;
143 seq = read_seqbegin(&xtime_lock);
144 xts = current_kernel_time();
145 tom = wall_to_monotonic;
146 } while (read_seqretry(&xtime_lock, seq));
148 xtim = timespec_to_ktime(xts);
149 tomono = timespec_to_ktime(tom);
150 base->clock_base[CLOCK_REALTIME].softirq_time = xtim;
151 base->clock_base[CLOCK_MONOTONIC].softirq_time =
152 ktime_add(xtim, tomono);
156 * Functions and macros which are different for UP/SMP systems are kept in a
162 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
163 * means that all timers which are tied to this base via timer->base are
164 * locked, and the base itself is locked too.
166 * So __run_timers/migrate_timers can safely modify all timers which could
167 * be found on the lists/queues.
169 * When the timer's base is locked, and the timer removed from list, it is
170 * possible to set timer->base = NULL and drop the lock: the timer remains
174 struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
175 unsigned long *flags)
177 struct hrtimer_clock_base *base;
181 if (likely(base != NULL)) {
182 spin_lock_irqsave(&base->cpu_base->lock, *flags);
183 if (likely(base == timer->base))
185 /* The timer has migrated to another CPU: */
186 spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
193 * Switch the timer base to the current CPU when possible.
195 static inline struct hrtimer_clock_base *
196 switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base)
198 struct hrtimer_clock_base *new_base;
199 struct hrtimer_cpu_base *new_cpu_base;
201 new_cpu_base = &__get_cpu_var(hrtimer_bases);
202 new_base = &new_cpu_base->clock_base[base->index];
204 if (base != new_base) {
206 * We are trying to schedule the timer on the local CPU.
207 * However we can't change timer's base while it is running,
208 * so we keep it on the same CPU. No hassle vs. reprogramming
209 * the event source in the high resolution case. The softirq
210 * code will take care of this when the timer function has
211 * completed. There is no conflict as we hold the lock until
212 * the timer is enqueued.
214 if (unlikely(hrtimer_callback_running(timer)))
217 /* See the comment in lock_timer_base() */
219 spin_unlock(&base->cpu_base->lock);
220 spin_lock(&new_base->cpu_base->lock);
221 timer->base = new_base;
226 #else /* CONFIG_SMP */
228 static inline struct hrtimer_clock_base *
229 lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
231 struct hrtimer_clock_base *base = timer->base;
233 spin_lock_irqsave(&base->cpu_base->lock, *flags);
238 # define switch_hrtimer_base(t, b) (b)
240 #endif /* !CONFIG_SMP */
243 * Functions for the union type storage format of ktime_t which are
244 * too large for inlining:
246 #if BITS_PER_LONG < 64
247 # ifndef CONFIG_KTIME_SCALAR
249 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
251 * @nsec: the scalar nsec value to add
253 * Returns the sum of kt and nsec in ktime_t format
255 ktime_t ktime_add_ns(const ktime_t kt, u64 nsec)
259 if (likely(nsec < NSEC_PER_SEC)) {
262 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
264 tmp = ktime_set((long)nsec, rem);
267 return ktime_add(kt, tmp);
270 EXPORT_SYMBOL_GPL(ktime_add_ns);
273 * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable
275 * @nsec: the scalar nsec value to subtract
277 * Returns the subtraction of @nsec from @kt in ktime_t format
279 ktime_t ktime_sub_ns(const ktime_t kt, u64 nsec)
283 if (likely(nsec < NSEC_PER_SEC)) {
286 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
288 tmp = ktime_set((long)nsec, rem);
291 return ktime_sub(kt, tmp);
294 EXPORT_SYMBOL_GPL(ktime_sub_ns);
295 # endif /* !CONFIG_KTIME_SCALAR */
298 * Divide a ktime value by a nanosecond value
300 u64 ktime_divns(const ktime_t kt, s64 div)
305 dclc = ktime_to_ns(kt);
306 /* Make sure the divisor is less than 2^32: */
312 do_div(dclc, (unsigned long) div);
316 #endif /* BITS_PER_LONG >= 64 */
319 * Add two ktime values and do a safety check for overflow:
321 ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
323 ktime_t res = ktime_add(lhs, rhs);
326 * We use KTIME_SEC_MAX here, the maximum timeout which we can
327 * return to user space in a timespec:
329 if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64)
330 res = ktime_set(KTIME_SEC_MAX, 0);
335 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
337 static struct debug_obj_descr hrtimer_debug_descr;
340 * fixup_init is called when:
341 * - an active object is initialized
343 static int hrtimer_fixup_init(void *addr, enum debug_obj_state state)
345 struct hrtimer *timer = addr;
348 case ODEBUG_STATE_ACTIVE:
349 hrtimer_cancel(timer);
350 debug_object_init(timer, &hrtimer_debug_descr);
358 * fixup_activate is called when:
359 * - an active object is activated
360 * - an unknown object is activated (might be a statically initialized object)
362 static int hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
366 case ODEBUG_STATE_NOTAVAILABLE:
370 case ODEBUG_STATE_ACTIVE:
379 * fixup_free is called when:
380 * - an active object is freed
382 static int hrtimer_fixup_free(void *addr, enum debug_obj_state state)
384 struct hrtimer *timer = addr;
387 case ODEBUG_STATE_ACTIVE:
388 hrtimer_cancel(timer);
389 debug_object_free(timer, &hrtimer_debug_descr);
396 static struct debug_obj_descr hrtimer_debug_descr = {
398 .fixup_init = hrtimer_fixup_init,
399 .fixup_activate = hrtimer_fixup_activate,
400 .fixup_free = hrtimer_fixup_free,
403 static inline void debug_hrtimer_init(struct hrtimer *timer)
405 debug_object_init(timer, &hrtimer_debug_descr);
408 static inline void debug_hrtimer_activate(struct hrtimer *timer)
410 debug_object_activate(timer, &hrtimer_debug_descr);
413 static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
415 debug_object_deactivate(timer, &hrtimer_debug_descr);
418 static inline void debug_hrtimer_free(struct hrtimer *timer)
420 debug_object_free(timer, &hrtimer_debug_descr);
423 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
424 enum hrtimer_mode mode);
426 void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
427 enum hrtimer_mode mode)
429 debug_object_init_on_stack(timer, &hrtimer_debug_descr);
430 __hrtimer_init(timer, clock_id, mode);
433 void destroy_hrtimer_on_stack(struct hrtimer *timer)
435 debug_object_free(timer, &hrtimer_debug_descr);
439 static inline void debug_hrtimer_init(struct hrtimer *timer) { }
440 static inline void debug_hrtimer_activate(struct hrtimer *timer) { }
441 static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
444 /* High resolution timer related functions */
445 #ifdef CONFIG_HIGH_RES_TIMERS
448 * High resolution timer enabled ?
450 static int hrtimer_hres_enabled __read_mostly = 1;
453 * Enable / Disable high resolution mode
455 static int __init setup_hrtimer_hres(char *str)
457 if (!strcmp(str, "off"))
458 hrtimer_hres_enabled = 0;
459 else if (!strcmp(str, "on"))
460 hrtimer_hres_enabled = 1;
466 __setup("highres=", setup_hrtimer_hres);
469 * hrtimer_high_res_enabled - query, if the highres mode is enabled
471 static inline int hrtimer_is_hres_enabled(void)
473 return hrtimer_hres_enabled;
477 * Is the high resolution mode active ?
479 static inline int hrtimer_hres_active(void)
481 return __get_cpu_var(hrtimer_bases).hres_active;
485 * Reprogram the event source with checking both queues for the
487 * Called with interrupts disabled and base->lock held
489 static void hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base)
492 struct hrtimer_clock_base *base = cpu_base->clock_base;
495 cpu_base->expires_next.tv64 = KTIME_MAX;
497 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
498 struct hrtimer *timer;
502 timer = rb_entry(base->first, struct hrtimer, node);
503 expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
505 * clock_was_set() has changed base->offset so the
506 * result might be negative. Fix it up to prevent a
507 * false positive in clockevents_program_event()
509 if (expires.tv64 < 0)
511 if (expires.tv64 < cpu_base->expires_next.tv64)
512 cpu_base->expires_next = expires;
515 if (cpu_base->expires_next.tv64 != KTIME_MAX)
516 tick_program_event(cpu_base->expires_next, 1);
520 * Shared reprogramming for clock_realtime and clock_monotonic
522 * When a timer is enqueued and expires earlier than the already enqueued
523 * timers, we have to check, whether it expires earlier than the timer for
524 * which the clock event device was armed.
526 * Called with interrupts disabled and base->cpu_base.lock held
528 static int hrtimer_reprogram(struct hrtimer *timer,
529 struct hrtimer_clock_base *base)
531 ktime_t *expires_next = &__get_cpu_var(hrtimer_bases).expires_next;
532 ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
535 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
538 * When the callback is running, we do not reprogram the clock event
539 * device. The timer callback is either running on a different CPU or
540 * the callback is executed in the hrtimer_interrupt context. The
541 * reprogramming is handled either by the softirq, which called the
542 * callback or at the end of the hrtimer_interrupt.
544 if (hrtimer_callback_running(timer))
548 * CLOCK_REALTIME timer might be requested with an absolute
549 * expiry time which is less than base->offset. Nothing wrong
550 * about that, just avoid to call into the tick code, which
551 * has now objections against negative expiry values.
553 if (expires.tv64 < 0)
556 if (expires.tv64 >= expires_next->tv64)
560 * Clockevents returns -ETIME, when the event was in the past.
562 res = tick_program_event(expires, 0);
563 if (!IS_ERR_VALUE(res))
564 *expires_next = expires;
570 * Retrigger next event is called after clock was set
572 * Called with interrupts disabled via on_each_cpu()
574 static void retrigger_next_event(void *arg)
576 struct hrtimer_cpu_base *base;
577 struct timespec realtime_offset;
580 if (!hrtimer_hres_active())
584 seq = read_seqbegin(&xtime_lock);
585 set_normalized_timespec(&realtime_offset,
586 -wall_to_monotonic.tv_sec,
587 -wall_to_monotonic.tv_nsec);
588 } while (read_seqretry(&xtime_lock, seq));
590 base = &__get_cpu_var(hrtimer_bases);
592 /* Adjust CLOCK_REALTIME offset */
593 spin_lock(&base->lock);
594 base->clock_base[CLOCK_REALTIME].offset =
595 timespec_to_ktime(realtime_offset);
597 hrtimer_force_reprogram(base);
598 spin_unlock(&base->lock);
602 * Clock realtime was set
604 * Change the offset of the realtime clock vs. the monotonic
607 * We might have to reprogram the high resolution timer interrupt. On
608 * SMP we call the architecture specific code to retrigger _all_ high
609 * resolution timer interrupts. On UP we just disable interrupts and
610 * call the high resolution interrupt code.
612 void clock_was_set(void)
614 /* Retrigger the CPU local events everywhere */
615 on_each_cpu(retrigger_next_event, NULL, 1);
619 * During resume we might have to reprogram the high resolution timer
620 * interrupt (on the local CPU):
622 void hres_timers_resume(void)
624 /* Retrigger the CPU local events: */
625 retrigger_next_event(NULL);
629 * Initialize the high resolution related parts of cpu_base
631 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
633 base->expires_next.tv64 = KTIME_MAX;
634 base->hres_active = 0;
638 * Initialize the high resolution related parts of a hrtimer
640 static inline void hrtimer_init_timer_hres(struct hrtimer *timer)
646 * When High resolution timers are active, try to reprogram. Note, that in case
647 * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry
648 * check happens. The timer gets enqueued into the rbtree. The reprogramming
649 * and expiry check is done in the hrtimer_interrupt or in the softirq.
651 static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
652 struct hrtimer_clock_base *base)
654 if (base->cpu_base->hres_active && hrtimer_reprogram(timer, base)) {
655 spin_unlock(&base->cpu_base->lock);
656 raise_softirq_irqoff(HRTIMER_SOFTIRQ);
657 spin_lock(&base->cpu_base->lock);
664 * Switch to high resolution mode
666 static int hrtimer_switch_to_hres(void)
668 int cpu = smp_processor_id();
669 struct hrtimer_cpu_base *base = &per_cpu(hrtimer_bases, cpu);
672 if (base->hres_active)
675 local_irq_save(flags);
677 if (tick_init_highres()) {
678 local_irq_restore(flags);
679 printk(KERN_WARNING "Could not switch to high resolution "
680 "mode on CPU %d\n", cpu);
683 base->hres_active = 1;
684 base->clock_base[CLOCK_REALTIME].resolution = KTIME_HIGH_RES;
685 base->clock_base[CLOCK_MONOTONIC].resolution = KTIME_HIGH_RES;
687 tick_setup_sched_timer();
689 /* "Retrigger" the interrupt to get things going */
690 retrigger_next_event(NULL);
691 local_irq_restore(flags);
692 printk(KERN_DEBUG "Switched to high resolution mode on CPU %d\n",
699 static inline int hrtimer_hres_active(void) { return 0; }
700 static inline int hrtimer_is_hres_enabled(void) { return 0; }
701 static inline int hrtimer_switch_to_hres(void) { return 0; }
702 static inline void hrtimer_force_reprogram(struct hrtimer_cpu_base *base) { }
703 static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
704 struct hrtimer_clock_base *base)
708 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { }
709 static inline void hrtimer_init_timer_hres(struct hrtimer *timer) { }
711 #endif /* CONFIG_HIGH_RES_TIMERS */
713 #ifdef CONFIG_TIMER_STATS
714 void __timer_stats_hrtimer_set_start_info(struct hrtimer *timer, void *addr)
716 if (timer->start_site)
719 timer->start_site = addr;
720 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
721 timer->start_pid = current->pid;
726 * Counterpart to lock_hrtimer_base above:
729 void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
731 spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
735 * hrtimer_forward - forward the timer expiry
736 * @timer: hrtimer to forward
737 * @now: forward past this time
738 * @interval: the interval to forward
740 * Forward the timer expiry so it will expire in the future.
741 * Returns the number of overruns.
743 u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
748 delta = ktime_sub(now, hrtimer_get_expires(timer));
753 if (interval.tv64 < timer->base->resolution.tv64)
754 interval.tv64 = timer->base->resolution.tv64;
756 if (unlikely(delta.tv64 >= interval.tv64)) {
757 s64 incr = ktime_to_ns(interval);
759 orun = ktime_divns(delta, incr);
760 hrtimer_add_expires_ns(timer, incr * orun);
761 if (hrtimer_get_expires_tv64(timer) > now.tv64)
764 * This (and the ktime_add() below) is the
765 * correction for exact:
769 hrtimer_add_expires(timer, interval);
773 EXPORT_SYMBOL_GPL(hrtimer_forward);
776 * enqueue_hrtimer - internal function to (re)start a timer
778 * The timer is inserted in expiry order. Insertion into the
779 * red black tree is O(log(n)). Must hold the base lock.
781 * Returns 1 when the new timer is the leftmost timer in the tree.
783 static int enqueue_hrtimer(struct hrtimer *timer,
784 struct hrtimer_clock_base *base)
786 struct rb_node **link = &base->active.rb_node;
787 struct rb_node *parent = NULL;
788 struct hrtimer *entry;
791 debug_hrtimer_activate(timer);
794 * Find the right place in the rbtree:
798 entry = rb_entry(parent, struct hrtimer, node);
800 * We dont care about collisions. Nodes with
801 * the same expiry time stay together.
803 if (hrtimer_get_expires_tv64(timer) <
804 hrtimer_get_expires_tv64(entry)) {
805 link = &(*link)->rb_left;
807 link = &(*link)->rb_right;
813 * Insert the timer to the rbtree and check whether it
814 * replaces the first pending timer
817 base->first = &timer->node;
819 rb_link_node(&timer->node, parent, link);
820 rb_insert_color(&timer->node, &base->active);
822 * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
823 * state of a possibly running callback.
825 timer->state |= HRTIMER_STATE_ENQUEUED;
831 * __remove_hrtimer - internal function to remove a timer
833 * Caller must hold the base lock.
835 * High resolution timer mode reprograms the clock event device when the
836 * timer is the one which expires next. The caller can disable this by setting
837 * reprogram to zero. This is useful, when the context does a reprogramming
838 * anyway (e.g. timer interrupt)
840 static void __remove_hrtimer(struct hrtimer *timer,
841 struct hrtimer_clock_base *base,
842 unsigned long newstate, int reprogram)
844 if (timer->state & HRTIMER_STATE_ENQUEUED) {
846 * Remove the timer from the rbtree and replace the
847 * first entry pointer if necessary.
849 if (base->first == &timer->node) {
850 base->first = rb_next(&timer->node);
851 /* Reprogram the clock event device. if enabled */
852 if (reprogram && hrtimer_hres_active())
853 hrtimer_force_reprogram(base->cpu_base);
855 rb_erase(&timer->node, &base->active);
857 timer->state = newstate;
861 * remove hrtimer, called with base lock held
864 remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base)
866 if (hrtimer_is_queued(timer)) {
870 * Remove the timer and force reprogramming when high
871 * resolution mode is active and the timer is on the current
872 * CPU. If we remove a timer on another CPU, reprogramming is
873 * skipped. The interrupt event on this CPU is fired and
874 * reprogramming happens in the interrupt handler. This is a
875 * rare case and less expensive than a smp call.
877 debug_hrtimer_deactivate(timer);
878 timer_stats_hrtimer_clear_start_info(timer);
879 reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases);
880 __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE,
888 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
889 * @timer: the timer to be added
891 * @delta_ns: "slack" range for the timer
892 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
896 * 1 when the timer was active
899 hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, unsigned long delta_ns,
900 const enum hrtimer_mode mode)
902 struct hrtimer_clock_base *base, *new_base;
906 base = lock_hrtimer_base(timer, &flags);
908 /* Remove an active timer from the queue: */
909 ret = remove_hrtimer(timer, base);
911 /* Switch the timer base, if necessary: */
912 new_base = switch_hrtimer_base(timer, base);
914 if (mode == HRTIMER_MODE_REL) {
915 tim = ktime_add_safe(tim, new_base->get_time());
917 * CONFIG_TIME_LOW_RES is a temporary way for architectures
918 * to signal that they simply return xtime in
919 * do_gettimeoffset(). In this case we want to round up by
920 * resolution when starting a relative timer, to avoid short
921 * timeouts. This will go away with the GTOD framework.
923 #ifdef CONFIG_TIME_LOW_RES
924 tim = ktime_add_safe(tim, base->resolution);
928 hrtimer_set_expires_range_ns(timer, tim, delta_ns);
930 timer_stats_hrtimer_set_start_info(timer);
932 leftmost = enqueue_hrtimer(timer, new_base);
935 * Only allow reprogramming if the new base is on this CPU.
936 * (it might still be on another CPU if the timer was pending)
938 * XXX send_remote_softirq() ?
940 if (leftmost && new_base->cpu_base == &__get_cpu_var(hrtimer_bases))
941 hrtimer_enqueue_reprogram(timer, new_base);
943 unlock_hrtimer_base(timer, &flags);
947 EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
950 * hrtimer_start - (re)start an hrtimer on the current CPU
951 * @timer: the timer to be added
953 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
957 * 1 when the timer was active
960 hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode)
962 return hrtimer_start_range_ns(timer, tim, 0, mode);
964 EXPORT_SYMBOL_GPL(hrtimer_start);
968 * hrtimer_try_to_cancel - try to deactivate a timer
969 * @timer: hrtimer to stop
972 * 0 when the timer was not active
973 * 1 when the timer was active
974 * -1 when the timer is currently excuting the callback function and
977 int hrtimer_try_to_cancel(struct hrtimer *timer)
979 struct hrtimer_clock_base *base;
983 base = lock_hrtimer_base(timer, &flags);
985 if (!hrtimer_callback_running(timer))
986 ret = remove_hrtimer(timer, base);
988 unlock_hrtimer_base(timer, &flags);
993 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
996 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
997 * @timer: the timer to be cancelled
1000 * 0 when the timer was not active
1001 * 1 when the timer was active
1003 int hrtimer_cancel(struct hrtimer *timer)
1006 int ret = hrtimer_try_to_cancel(timer);
1013 EXPORT_SYMBOL_GPL(hrtimer_cancel);
1016 * hrtimer_get_remaining - get remaining time for the timer
1017 * @timer: the timer to read
1019 ktime_t hrtimer_get_remaining(const struct hrtimer *timer)
1021 struct hrtimer_clock_base *base;
1022 unsigned long flags;
1025 base = lock_hrtimer_base(timer, &flags);
1026 rem = hrtimer_expires_remaining(timer);
1027 unlock_hrtimer_base(timer, &flags);
1031 EXPORT_SYMBOL_GPL(hrtimer_get_remaining);
1035 * hrtimer_get_next_event - get the time until next expiry event
1037 * Returns the delta to the next expiry event or KTIME_MAX if no timer
1040 ktime_t hrtimer_get_next_event(void)
1042 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1043 struct hrtimer_clock_base *base = cpu_base->clock_base;
1044 ktime_t delta, mindelta = { .tv64 = KTIME_MAX };
1045 unsigned long flags;
1048 spin_lock_irqsave(&cpu_base->lock, flags);
1050 if (!hrtimer_hres_active()) {
1051 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
1052 struct hrtimer *timer;
1057 timer = rb_entry(base->first, struct hrtimer, node);
1058 delta.tv64 = hrtimer_get_expires_tv64(timer);
1059 delta = ktime_sub(delta, base->get_time());
1060 if (delta.tv64 < mindelta.tv64)
1061 mindelta.tv64 = delta.tv64;
1065 spin_unlock_irqrestore(&cpu_base->lock, flags);
1067 if (mindelta.tv64 < 0)
1073 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1074 enum hrtimer_mode mode)
1076 struct hrtimer_cpu_base *cpu_base;
1078 memset(timer, 0, sizeof(struct hrtimer));
1080 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1082 if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
1083 clock_id = CLOCK_MONOTONIC;
1085 timer->base = &cpu_base->clock_base[clock_id];
1086 INIT_LIST_HEAD(&timer->cb_entry);
1087 hrtimer_init_timer_hres(timer);
1089 #ifdef CONFIG_TIMER_STATS
1090 timer->start_site = NULL;
1091 timer->start_pid = -1;
1092 memset(timer->start_comm, 0, TASK_COMM_LEN);
1097 * hrtimer_init - initialize a timer to the given clock
1098 * @timer: the timer to be initialized
1099 * @clock_id: the clock to be used
1100 * @mode: timer mode abs/rel
1102 void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1103 enum hrtimer_mode mode)
1105 debug_hrtimer_init(timer);
1106 __hrtimer_init(timer, clock_id, mode);
1108 EXPORT_SYMBOL_GPL(hrtimer_init);
1111 * hrtimer_get_res - get the timer resolution for a clock
1112 * @which_clock: which clock to query
1113 * @tp: pointer to timespec variable to store the resolution
1115 * Store the resolution of the clock selected by @which_clock in the
1116 * variable pointed to by @tp.
1118 int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp)
1120 struct hrtimer_cpu_base *cpu_base;
1122 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1123 *tp = ktime_to_timespec(cpu_base->clock_base[which_clock].resolution);
1127 EXPORT_SYMBOL_GPL(hrtimer_get_res);
1129 static void __run_hrtimer(struct hrtimer *timer)
1131 struct hrtimer_clock_base *base = timer->base;
1132 struct hrtimer_cpu_base *cpu_base = base->cpu_base;
1133 enum hrtimer_restart (*fn)(struct hrtimer *);
1136 WARN_ON(!irqs_disabled());
1138 debug_hrtimer_deactivate(timer);
1139 __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0);
1140 timer_stats_account_hrtimer(timer);
1141 fn = timer->function;
1144 * Because we run timers from hardirq context, there is no chance
1145 * they get migrated to another cpu, therefore its safe to unlock
1148 spin_unlock(&cpu_base->lock);
1149 restart = fn(timer);
1150 spin_lock(&cpu_base->lock);
1153 * Note: We clear the CALLBACK bit after enqueue_hrtimer and
1154 * we do not reprogramm the event hardware. Happens either in
1155 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1157 if (restart != HRTIMER_NORESTART) {
1158 BUG_ON(timer->state != HRTIMER_STATE_CALLBACK);
1159 enqueue_hrtimer(timer, base);
1161 timer->state &= ~HRTIMER_STATE_CALLBACK;
1164 #ifdef CONFIG_HIGH_RES_TIMERS
1166 static int force_clock_reprogram;
1169 * After 5 iteration's attempts, we consider that hrtimer_interrupt()
1170 * is hanging, which could happen with something that slows the interrupt
1171 * such as the tracing. Then we force the clock reprogramming for each future
1172 * hrtimer interrupts to avoid infinite loops and use the min_delta_ns
1173 * threshold that we will overwrite.
1174 * The next tick event will be scheduled to 3 times we currently spend on
1175 * hrtimer_interrupt(). This gives a good compromise, the cpus will spend
1176 * 1/4 of their time to process the hrtimer interrupts. This is enough to
1177 * let it running without serious starvation.
1181 hrtimer_interrupt_hanging(struct clock_event_device *dev,
1184 force_clock_reprogram = 1;
1185 dev->min_delta_ns = (unsigned long)try_time.tv64 * 3;
1186 printk(KERN_WARNING "hrtimer: interrupt too slow, "
1187 "forcing clock min delta to %lu ns\n", dev->min_delta_ns);
1190 * High resolution timer interrupt
1191 * Called with interrupts disabled
1193 void hrtimer_interrupt(struct clock_event_device *dev)
1195 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1196 struct hrtimer_clock_base *base;
1197 ktime_t expires_next, now;
1201 BUG_ON(!cpu_base->hres_active);
1202 cpu_base->nr_events++;
1203 dev->next_event.tv64 = KTIME_MAX;
1206 /* 5 retries is enough to notice a hang */
1207 if (!(++nr_retries % 5))
1208 hrtimer_interrupt_hanging(dev, ktime_sub(ktime_get(), now));
1212 expires_next.tv64 = KTIME_MAX;
1214 base = cpu_base->clock_base;
1216 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1218 struct rb_node *node;
1220 spin_lock(&cpu_base->lock);
1222 basenow = ktime_add(now, base->offset);
1224 while ((node = base->first)) {
1225 struct hrtimer *timer;
1227 timer = rb_entry(node, struct hrtimer, node);
1230 * The immediate goal for using the softexpires is
1231 * minimizing wakeups, not running timers at the
1232 * earliest interrupt after their soft expiration.
1233 * This allows us to avoid using a Priority Search
1234 * Tree, which can answer a stabbing querry for
1235 * overlapping intervals and instead use the simple
1236 * BST we already have.
1237 * We don't add extra wakeups by delaying timers that
1238 * are right-of a not yet expired timer, because that
1239 * timer will have to trigger a wakeup anyway.
1242 if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer)) {
1245 expires = ktime_sub(hrtimer_get_expires(timer),
1247 if (expires.tv64 < expires_next.tv64)
1248 expires_next = expires;
1252 __run_hrtimer(timer);
1254 spin_unlock(&cpu_base->lock);
1258 cpu_base->expires_next = expires_next;
1260 /* Reprogramming necessary ? */
1261 if (expires_next.tv64 != KTIME_MAX) {
1262 if (tick_program_event(expires_next, force_clock_reprogram))
1268 * local version of hrtimer_peek_ahead_timers() called with interrupts
1271 static void __hrtimer_peek_ahead_timers(void)
1273 struct tick_device *td;
1275 if (!hrtimer_hres_active())
1278 td = &__get_cpu_var(tick_cpu_device);
1279 if (td && td->evtdev)
1280 hrtimer_interrupt(td->evtdev);
1284 * hrtimer_peek_ahead_timers -- run soft-expired timers now
1286 * hrtimer_peek_ahead_timers will peek at the timer queue of
1287 * the current cpu and check if there are any timers for which
1288 * the soft expires time has passed. If any such timers exist,
1289 * they are run immediately and then removed from the timer queue.
1292 void hrtimer_peek_ahead_timers(void)
1294 unsigned long flags;
1296 local_irq_save(flags);
1297 __hrtimer_peek_ahead_timers();
1298 local_irq_restore(flags);
1301 static void run_hrtimer_softirq(struct softirq_action *h)
1303 hrtimer_peek_ahead_timers();
1306 #else /* CONFIG_HIGH_RES_TIMERS */
1308 static inline void __hrtimer_peek_ahead_timers(void) { }
1310 #endif /* !CONFIG_HIGH_RES_TIMERS */
1313 * Called from timer softirq every jiffy, expire hrtimers:
1315 * For HRT its the fall back code to run the softirq in the timer
1316 * softirq context in case the hrtimer initialization failed or has
1317 * not been done yet.
1319 void hrtimer_run_pending(void)
1321 if (hrtimer_hres_active())
1325 * This _is_ ugly: We have to check in the softirq context,
1326 * whether we can switch to highres and / or nohz mode. The
1327 * clocksource switch happens in the timer interrupt with
1328 * xtime_lock held. Notification from there only sets the
1329 * check bit in the tick_oneshot code, otherwise we might
1330 * deadlock vs. xtime_lock.
1332 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1333 hrtimer_switch_to_hres();
1337 * Called from hardirq context every jiffy
1339 void hrtimer_run_queues(void)
1341 struct rb_node *node;
1342 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1343 struct hrtimer_clock_base *base;
1344 int index, gettime = 1;
1346 if (hrtimer_hres_active())
1349 for (index = 0; index < HRTIMER_MAX_CLOCK_BASES; index++) {
1350 base = &cpu_base->clock_base[index];
1356 hrtimer_get_softirq_time(cpu_base);
1360 spin_lock(&cpu_base->lock);
1362 while ((node = base->first)) {
1363 struct hrtimer *timer;
1365 timer = rb_entry(node, struct hrtimer, node);
1366 if (base->softirq_time.tv64 <=
1367 hrtimer_get_expires_tv64(timer))
1370 __run_hrtimer(timer);
1372 spin_unlock(&cpu_base->lock);
1377 * Sleep related functions:
1379 static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1381 struct hrtimer_sleeper *t =
1382 container_of(timer, struct hrtimer_sleeper, timer);
1383 struct task_struct *task = t->task;
1387 wake_up_process(task);
1389 return HRTIMER_NORESTART;
1392 void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1394 sl->timer.function = hrtimer_wakeup;
1398 static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1400 hrtimer_init_sleeper(t, current);
1403 set_current_state(TASK_INTERRUPTIBLE);
1404 hrtimer_start_expires(&t->timer, mode);
1405 if (!hrtimer_active(&t->timer))
1408 if (likely(t->task))
1411 hrtimer_cancel(&t->timer);
1412 mode = HRTIMER_MODE_ABS;
1414 } while (t->task && !signal_pending(current));
1416 __set_current_state(TASK_RUNNING);
1418 return t->task == NULL;
1421 static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp)
1423 struct timespec rmt;
1426 rem = hrtimer_expires_remaining(timer);
1429 rmt = ktime_to_timespec(rem);
1431 if (copy_to_user(rmtp, &rmt, sizeof(*rmtp)))
1437 long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1439 struct hrtimer_sleeper t;
1440 struct timespec __user *rmtp;
1443 hrtimer_init_on_stack(&t.timer, restart->nanosleep.index,
1445 hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
1447 if (do_nanosleep(&t, HRTIMER_MODE_ABS))
1450 rmtp = restart->nanosleep.rmtp;
1452 ret = update_rmtp(&t.timer, rmtp);
1457 /* The other values in restart are already filled in */
1458 ret = -ERESTART_RESTARTBLOCK;
1460 destroy_hrtimer_on_stack(&t.timer);
1464 long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
1465 const enum hrtimer_mode mode, const clockid_t clockid)
1467 struct restart_block *restart;
1468 struct hrtimer_sleeper t;
1470 unsigned long slack;
1472 slack = current->timer_slack_ns;
1473 if (rt_task(current))
1476 hrtimer_init_on_stack(&t.timer, clockid, mode);
1477 hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack);
1478 if (do_nanosleep(&t, mode))
1481 /* Absolute timers do not update the rmtp value and restart: */
1482 if (mode == HRTIMER_MODE_ABS) {
1483 ret = -ERESTARTNOHAND;
1488 ret = update_rmtp(&t.timer, rmtp);
1493 restart = ¤t_thread_info()->restart_block;
1494 restart->fn = hrtimer_nanosleep_restart;
1495 restart->nanosleep.index = t.timer.base->index;
1496 restart->nanosleep.rmtp = rmtp;
1497 restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
1499 ret = -ERESTART_RESTARTBLOCK;
1501 destroy_hrtimer_on_stack(&t.timer);
1505 SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp,
1506 struct timespec __user *, rmtp)
1510 if (copy_from_user(&tu, rqtp, sizeof(tu)))
1513 if (!timespec_valid(&tu))
1516 return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1520 * Functions related to boot-time initialization:
1522 static void __cpuinit init_hrtimers_cpu(int cpu)
1524 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1527 spin_lock_init(&cpu_base->lock);
1529 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
1530 cpu_base->clock_base[i].cpu_base = cpu_base;
1532 hrtimer_init_hres(cpu_base);
1535 #ifdef CONFIG_HOTPLUG_CPU
1537 static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1538 struct hrtimer_clock_base *new_base)
1540 struct hrtimer *timer;
1541 struct rb_node *node;
1543 while ((node = rb_first(&old_base->active))) {
1544 timer = rb_entry(node, struct hrtimer, node);
1545 BUG_ON(hrtimer_callback_running(timer));
1546 debug_hrtimer_deactivate(timer);
1549 * Mark it as STATE_MIGRATE not INACTIVE otherwise the
1550 * timer could be seen as !active and just vanish away
1551 * under us on another CPU
1553 __remove_hrtimer(timer, old_base, HRTIMER_STATE_MIGRATE, 0);
1554 timer->base = new_base;
1556 * Enqueue the timers on the new cpu. This does not
1557 * reprogram the event device in case the timer
1558 * expires before the earliest on this CPU, but we run
1559 * hrtimer_interrupt after we migrated everything to
1560 * sort out already expired timers and reprogram the
1563 enqueue_hrtimer(timer, new_base);
1565 /* Clear the migration state bit */
1566 timer->state &= ~HRTIMER_STATE_MIGRATE;
1570 static void migrate_hrtimers(int scpu)
1572 struct hrtimer_cpu_base *old_base, *new_base;
1575 BUG_ON(cpu_online(scpu));
1576 tick_cancel_sched_timer(scpu);
1578 local_irq_disable();
1579 old_base = &per_cpu(hrtimer_bases, scpu);
1580 new_base = &__get_cpu_var(hrtimer_bases);
1582 * The caller is globally serialized and nobody else
1583 * takes two locks at once, deadlock is not possible.
1585 spin_lock(&new_base->lock);
1586 spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1588 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1589 migrate_hrtimer_list(&old_base->clock_base[i],
1590 &new_base->clock_base[i]);
1593 spin_unlock(&old_base->lock);
1594 spin_unlock(&new_base->lock);
1596 /* Check, if we got expired work to do */
1597 __hrtimer_peek_ahead_timers();
1601 #endif /* CONFIG_HOTPLUG_CPU */
1603 static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self,
1604 unsigned long action, void *hcpu)
1606 int scpu = (long)hcpu;
1610 case CPU_UP_PREPARE:
1611 case CPU_UP_PREPARE_FROZEN:
1612 init_hrtimers_cpu(scpu);
1615 #ifdef CONFIG_HOTPLUG_CPU
1617 case CPU_DYING_FROZEN:
1618 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DYING, &scpu);
1621 case CPU_DEAD_FROZEN:
1623 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &scpu);
1624 migrate_hrtimers(scpu);
1636 static struct notifier_block __cpuinitdata hrtimers_nb = {
1637 .notifier_call = hrtimer_cpu_notify,
1640 void __init hrtimers_init(void)
1642 hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
1643 (void *)(long)smp_processor_id());
1644 register_cpu_notifier(&hrtimers_nb);
1645 #ifdef CONFIG_HIGH_RES_TIMERS
1646 open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq);
1651 * schedule_hrtimeout_range - sleep until timeout
1652 * @expires: timeout value (ktime_t)
1653 * @delta: slack in expires timeout (ktime_t)
1654 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1656 * Make the current task sleep until the given expiry time has
1657 * elapsed. The routine will return immediately unless
1658 * the current task state has been set (see set_current_state()).
1660 * The @delta argument gives the kernel the freedom to schedule the
1661 * actual wakeup to a time that is both power and performance friendly.
1662 * The kernel give the normal best effort behavior for "@expires+@delta",
1663 * but may decide to fire the timer earlier, but no earlier than @expires.
1665 * You can set the task state as follows -
1667 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1668 * pass before the routine returns.
1670 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1671 * delivered to the current task.
1673 * The current task state is guaranteed to be TASK_RUNNING when this
1676 * Returns 0 when the timer has expired otherwise -EINTR
1678 int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta,
1679 const enum hrtimer_mode mode)
1681 struct hrtimer_sleeper t;
1684 * Optimize when a zero timeout value is given. It does not
1685 * matter whether this is an absolute or a relative time.
1687 if (expires && !expires->tv64) {
1688 __set_current_state(TASK_RUNNING);
1693 * A NULL parameter means "inifinte"
1697 __set_current_state(TASK_RUNNING);
1701 hrtimer_init_on_stack(&t.timer, CLOCK_MONOTONIC, mode);
1702 hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
1704 hrtimer_init_sleeper(&t, current);
1706 hrtimer_start_expires(&t.timer, mode);
1707 if (!hrtimer_active(&t.timer))
1713 hrtimer_cancel(&t.timer);
1714 destroy_hrtimer_on_stack(&t.timer);
1716 __set_current_state(TASK_RUNNING);
1718 return !t.task ? 0 : -EINTR;
1720 EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
1723 * schedule_hrtimeout - sleep until timeout
1724 * @expires: timeout value (ktime_t)
1725 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1727 * Make the current task sleep until the given expiry time has
1728 * elapsed. The routine will return immediately unless
1729 * the current task state has been set (see set_current_state()).
1731 * You can set the task state as follows -
1733 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1734 * pass before the routine returns.
1736 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1737 * delivered to the current task.
1739 * The current task state is guaranteed to be TASK_RUNNING when this
1742 * Returns 0 when the timer has expired otherwise -EINTR
1744 int __sched schedule_hrtimeout(ktime_t *expires,
1745 const enum hrtimer_mode mode)
1747 return schedule_hrtimeout_range(expires, 0, mode);
1749 EXPORT_SYMBOL_GPL(schedule_hrtimeout);