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>
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);
145 getnstimeofday(&xts);
149 tom = wall_to_monotonic;
150 } while (read_seqretry(&xtime_lock, seq));
152 xtim = timespec_to_ktime(xts);
153 tomono = timespec_to_ktime(tom);
154 base->clock_base[CLOCK_REALTIME].softirq_time = xtim;
155 base->clock_base[CLOCK_MONOTONIC].softirq_time =
156 ktime_add(xtim, tomono);
160 * Helper function to check, whether the timer is running the callback
163 static inline int hrtimer_callback_running(struct hrtimer *timer)
165 return timer->state & HRTIMER_STATE_CALLBACK;
169 * Functions and macros which are different for UP/SMP systems are kept in a
175 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
176 * means that all timers which are tied to this base via timer->base are
177 * locked, and the base itself is locked too.
179 * So __run_timers/migrate_timers can safely modify all timers which could
180 * be found on the lists/queues.
182 * When the timer's base is locked, and the timer removed from list, it is
183 * possible to set timer->base = NULL and drop the lock: the timer remains
187 struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
188 unsigned long *flags)
190 struct hrtimer_clock_base *base;
194 if (likely(base != NULL)) {
195 spin_lock_irqsave(&base->cpu_base->lock, *flags);
196 if (likely(base == timer->base))
198 /* The timer has migrated to another CPU: */
199 spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
206 * Switch the timer base to the current CPU when possible.
208 static inline struct hrtimer_clock_base *
209 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;
214 new_cpu_base = &__get_cpu_var(hrtimer_bases);
215 new_base = &new_cpu_base->clock_base[base->index];
217 if (base != new_base) {
219 * We are trying to schedule the timer on the local CPU.
220 * However we can't change timer's base while it is running,
221 * so we keep it on the same CPU. No hassle vs. reprogramming
222 * the event source in the high resolution case. The softirq
223 * code will take care of this when the timer function has
224 * completed. There is no conflict as we hold the lock until
225 * the timer is enqueued.
227 if (unlikely(hrtimer_callback_running(timer)))
230 /* See the comment in lock_timer_base() */
232 spin_unlock(&base->cpu_base->lock);
233 spin_lock(&new_base->cpu_base->lock);
234 timer->base = new_base;
239 #else /* CONFIG_SMP */
241 static inline struct hrtimer_clock_base *
242 lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
244 struct hrtimer_clock_base *base = timer->base;
246 spin_lock_irqsave(&base->cpu_base->lock, *flags);
251 # define switch_hrtimer_base(t, b) (b)
253 #endif /* !CONFIG_SMP */
256 * Functions for the union type storage format of ktime_t which are
257 * too large for inlining:
259 #if BITS_PER_LONG < 64
260 # ifndef CONFIG_KTIME_SCALAR
262 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
264 * @nsec: the scalar nsec value to add
266 * Returns the sum of kt and nsec in ktime_t format
268 ktime_t ktime_add_ns(const ktime_t kt, u64 nsec)
272 if (likely(nsec < NSEC_PER_SEC)) {
275 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
277 tmp = ktime_set((long)nsec, rem);
280 return ktime_add(kt, tmp);
283 EXPORT_SYMBOL_GPL(ktime_add_ns);
284 # endif /* !CONFIG_KTIME_SCALAR */
287 * Divide a ktime value by a nanosecond value
289 unsigned long ktime_divns(const ktime_t kt, s64 div)
294 dclc = dns = ktime_to_ns(kt);
296 /* Make sure the divisor is less than 2^32: */
302 do_div(dclc, (unsigned long) div);
304 return (unsigned long) dclc;
306 #endif /* BITS_PER_LONG >= 64 */
308 /* High resolution timer related functions */
309 #ifdef CONFIG_HIGH_RES_TIMERS
312 * High resolution timer enabled ?
314 static int hrtimer_hres_enabled __read_mostly = 1;
317 * Enable / Disable high resolution mode
319 static int __init setup_hrtimer_hres(char *str)
321 if (!strcmp(str, "off"))
322 hrtimer_hres_enabled = 0;
323 else if (!strcmp(str, "on"))
324 hrtimer_hres_enabled = 1;
330 __setup("highres=", setup_hrtimer_hres);
333 * hrtimer_high_res_enabled - query, if the highres mode is enabled
335 static inline int hrtimer_is_hres_enabled(void)
337 return hrtimer_hres_enabled;
341 * Is the high resolution mode active ?
343 static inline int hrtimer_hres_active(void)
345 return __get_cpu_var(hrtimer_bases).hres_active;
349 * Reprogram the event source with checking both queues for the
351 * Called with interrupts disabled and base->lock held
353 static void hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base)
356 struct hrtimer_clock_base *base = cpu_base->clock_base;
359 cpu_base->expires_next.tv64 = KTIME_MAX;
361 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
362 struct hrtimer *timer;
366 timer = rb_entry(base->first, struct hrtimer, node);
367 expires = ktime_sub(timer->expires, base->offset);
368 if (expires.tv64 < cpu_base->expires_next.tv64)
369 cpu_base->expires_next = expires;
372 if (cpu_base->expires_next.tv64 != KTIME_MAX)
373 tick_program_event(cpu_base->expires_next, 1);
377 * Shared reprogramming for clock_realtime and clock_monotonic
379 * When a timer is enqueued and expires earlier than the already enqueued
380 * timers, we have to check, whether it expires earlier than the timer for
381 * which the clock event device was armed.
383 * Called with interrupts disabled and base->cpu_base.lock held
385 static int hrtimer_reprogram(struct hrtimer *timer,
386 struct hrtimer_clock_base *base)
388 ktime_t *expires_next = &__get_cpu_var(hrtimer_bases).expires_next;
389 ktime_t expires = ktime_sub(timer->expires, base->offset);
393 * When the callback is running, we do not reprogram the clock event
394 * device. The timer callback is either running on a different CPU or
395 * the callback is executed in the hrtimer_interupt context. The
396 * reprogramming is handled either by the softirq, which called the
397 * callback or at the end of the hrtimer_interrupt.
399 if (hrtimer_callback_running(timer))
402 if (expires.tv64 >= expires_next->tv64)
406 * Clockevents returns -ETIME, when the event was in the past.
408 res = tick_program_event(expires, 0);
409 if (!IS_ERR_VALUE(res))
410 *expires_next = expires;
416 * Retrigger next event is called after clock was set
418 * Called with interrupts disabled via on_each_cpu()
420 static void retrigger_next_event(void *arg)
422 struct hrtimer_cpu_base *base;
423 struct timespec realtime_offset;
426 if (!hrtimer_hres_active())
430 seq = read_seqbegin(&xtime_lock);
431 set_normalized_timespec(&realtime_offset,
432 -wall_to_monotonic.tv_sec,
433 -wall_to_monotonic.tv_nsec);
434 } while (read_seqretry(&xtime_lock, seq));
436 base = &__get_cpu_var(hrtimer_bases);
438 /* Adjust CLOCK_REALTIME offset */
439 spin_lock(&base->lock);
440 base->clock_base[CLOCK_REALTIME].offset =
441 timespec_to_ktime(realtime_offset);
443 hrtimer_force_reprogram(base);
444 spin_unlock(&base->lock);
448 * Clock realtime was set
450 * Change the offset of the realtime clock vs. the monotonic
453 * We might have to reprogram the high resolution timer interrupt. On
454 * SMP we call the architecture specific code to retrigger _all_ high
455 * resolution timer interrupts. On UP we just disable interrupts and
456 * call the high resolution interrupt code.
458 void clock_was_set(void)
460 /* Retrigger the CPU local events everywhere */
461 on_each_cpu(retrigger_next_event, NULL, 0, 1);
465 * During resume we might have to reprogram the high resolution timer
466 * interrupt (on the local CPU):
468 void hres_timers_resume(void)
470 WARN_ON_ONCE(num_online_cpus() > 1);
472 /* Retrigger the CPU local events: */
473 retrigger_next_event(NULL);
477 * Check, whether the timer is on the callback pending list
479 static inline int hrtimer_cb_pending(const struct hrtimer *timer)
481 return timer->state & HRTIMER_STATE_PENDING;
485 * Remove a timer from the callback pending list
487 static inline void hrtimer_remove_cb_pending(struct hrtimer *timer)
489 list_del_init(&timer->cb_entry);
493 * Initialize the high resolution related parts of cpu_base
495 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
497 base->expires_next.tv64 = KTIME_MAX;
498 base->hres_active = 0;
499 INIT_LIST_HEAD(&base->cb_pending);
503 * Initialize the high resolution related parts of a hrtimer
505 static inline void hrtimer_init_timer_hres(struct hrtimer *timer)
507 INIT_LIST_HEAD(&timer->cb_entry);
511 * When High resolution timers are active, try to reprogram. Note, that in case
512 * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry
513 * check happens. The timer gets enqueued into the rbtree. The reprogramming
514 * and expiry check is done in the hrtimer_interrupt or in the softirq.
516 static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
517 struct hrtimer_clock_base *base)
519 if (base->cpu_base->hres_active && hrtimer_reprogram(timer, base)) {
521 /* Timer is expired, act upon the callback mode */
522 switch(timer->cb_mode) {
523 case HRTIMER_CB_IRQSAFE_NO_RESTART:
525 * We can call the callback from here. No restart
526 * happens, so no danger of recursion
528 BUG_ON(timer->function(timer) != HRTIMER_NORESTART);
530 case HRTIMER_CB_IRQSAFE_NO_SOFTIRQ:
532 * This is solely for the sched tick emulation with
533 * dynamic tick support to ensure that we do not
534 * restart the tick right on the edge and end up with
535 * the tick timer in the softirq ! The calling site
536 * takes care of this.
539 case HRTIMER_CB_IRQSAFE:
540 case HRTIMER_CB_SOFTIRQ:
542 * Move everything else into the softirq pending list !
544 list_add_tail(&timer->cb_entry,
545 &base->cpu_base->cb_pending);
546 timer->state = HRTIMER_STATE_PENDING;
547 raise_softirq(HRTIMER_SOFTIRQ);
557 * Switch to high resolution mode
559 static int hrtimer_switch_to_hres(void)
561 struct hrtimer_cpu_base *base = &__get_cpu_var(hrtimer_bases);
564 if (base->hres_active)
567 local_irq_save(flags);
569 if (tick_init_highres()) {
570 local_irq_restore(flags);
573 base->hres_active = 1;
574 base->clock_base[CLOCK_REALTIME].resolution = KTIME_HIGH_RES;
575 base->clock_base[CLOCK_MONOTONIC].resolution = KTIME_HIGH_RES;
577 tick_setup_sched_timer();
579 /* "Retrigger" the interrupt to get things going */
580 retrigger_next_event(NULL);
581 local_irq_restore(flags);
582 printk(KERN_INFO "Switched to high resolution mode on CPU %d\n",
589 static inline int hrtimer_hres_active(void) { return 0; }
590 static inline int hrtimer_is_hres_enabled(void) { return 0; }
591 static inline int hrtimer_switch_to_hres(void) { return 0; }
592 static inline void hrtimer_force_reprogram(struct hrtimer_cpu_base *base) { }
593 static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
594 struct hrtimer_clock_base *base)
598 static inline int hrtimer_cb_pending(struct hrtimer *timer) { return 0; }
599 static inline void hrtimer_remove_cb_pending(struct hrtimer *timer) { }
600 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { }
601 static inline void hrtimer_init_timer_hres(struct hrtimer *timer) { }
603 #endif /* CONFIG_HIGH_RES_TIMERS */
605 #ifdef CONFIG_TIMER_STATS
606 void __timer_stats_hrtimer_set_start_info(struct hrtimer *timer, void *addr)
608 if (timer->start_site)
611 timer->start_site = addr;
612 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
613 timer->start_pid = current->pid;
618 * Counterpart to lock_timer_base above:
621 void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
623 spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
627 * hrtimer_forward - forward the timer expiry
628 * @timer: hrtimer to forward
629 * @now: forward past this time
630 * @interval: the interval to forward
632 * Forward the timer expiry so it will expire in the future.
633 * Returns the number of overruns.
636 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
638 unsigned long orun = 1;
641 delta = ktime_sub(now, timer->expires);
646 if (interval.tv64 < timer->base->resolution.tv64)
647 interval.tv64 = timer->base->resolution.tv64;
649 if (unlikely(delta.tv64 >= interval.tv64)) {
650 s64 incr = ktime_to_ns(interval);
652 orun = ktime_divns(delta, incr);
653 timer->expires = ktime_add_ns(timer->expires, incr * orun);
654 if (timer->expires.tv64 > now.tv64)
657 * This (and the ktime_add() below) is the
658 * correction for exact:
662 timer->expires = ktime_add(timer->expires, interval);
664 * Make sure, that the result did not wrap with a very large
667 if (timer->expires.tv64 < 0)
668 timer->expires = ktime_set(KTIME_SEC_MAX, 0);
672 EXPORT_SYMBOL_GPL(hrtimer_forward);
675 * enqueue_hrtimer - internal function to (re)start a timer
677 * The timer is inserted in expiry order. Insertion into the
678 * red black tree is O(log(n)). Must hold the base lock.
680 static void enqueue_hrtimer(struct hrtimer *timer,
681 struct hrtimer_clock_base *base, int reprogram)
683 struct rb_node **link = &base->active.rb_node;
684 struct rb_node *parent = NULL;
685 struct hrtimer *entry;
688 * Find the right place in the rbtree:
692 entry = rb_entry(parent, struct hrtimer, node);
694 * We dont care about collisions. Nodes with
695 * the same expiry time stay together.
697 if (timer->expires.tv64 < entry->expires.tv64)
698 link = &(*link)->rb_left;
700 link = &(*link)->rb_right;
704 * Insert the timer to the rbtree and check whether it
705 * replaces the first pending timer
707 if (!base->first || timer->expires.tv64 <
708 rb_entry(base->first, struct hrtimer, node)->expires.tv64) {
710 * Reprogram the clock event device. When the timer is already
711 * expired hrtimer_enqueue_reprogram has either called the
712 * callback or added it to the pending list and raised the
715 * This is a NOP for !HIGHRES
717 if (reprogram && hrtimer_enqueue_reprogram(timer, base))
720 base->first = &timer->node;
723 rb_link_node(&timer->node, parent, link);
724 rb_insert_color(&timer->node, &base->active);
726 * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
727 * state of a possibly running callback.
729 timer->state |= HRTIMER_STATE_ENQUEUED;
733 * __remove_hrtimer - internal function to remove a timer
735 * Caller must hold the base lock.
737 * High resolution timer mode reprograms the clock event device when the
738 * timer is the one which expires next. The caller can disable this by setting
739 * reprogram to zero. This is useful, when the context does a reprogramming
740 * anyway (e.g. timer interrupt)
742 static void __remove_hrtimer(struct hrtimer *timer,
743 struct hrtimer_clock_base *base,
744 unsigned long newstate, int reprogram)
746 /* High res. callback list. NOP for !HIGHRES */
747 if (hrtimer_cb_pending(timer))
748 hrtimer_remove_cb_pending(timer);
751 * Remove the timer from the rbtree and replace the
752 * first entry pointer if necessary.
754 if (base->first == &timer->node) {
755 base->first = rb_next(&timer->node);
756 /* Reprogram the clock event device. if enabled */
757 if (reprogram && hrtimer_hres_active())
758 hrtimer_force_reprogram(base->cpu_base);
760 rb_erase(&timer->node, &base->active);
762 timer->state = newstate;
766 * remove hrtimer, called with base lock held
769 remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base)
771 if (hrtimer_is_queued(timer)) {
775 * Remove the timer and force reprogramming when high
776 * resolution mode is active and the timer is on the current
777 * CPU. If we remove a timer on another CPU, reprogramming is
778 * skipped. The interrupt event on this CPU is fired and
779 * reprogramming happens in the interrupt handler. This is a
780 * rare case and less expensive than a smp call.
782 timer_stats_hrtimer_clear_start_info(timer);
783 reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases);
784 __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE,
792 * hrtimer_start - (re)start an relative timer on the current CPU
793 * @timer: the timer to be added
795 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
799 * 1 when the timer was active
802 hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode)
804 struct hrtimer_clock_base *base, *new_base;
808 base = lock_hrtimer_base(timer, &flags);
810 /* Remove an active timer from the queue: */
811 ret = remove_hrtimer(timer, base);
813 /* Switch the timer base, if necessary: */
814 new_base = switch_hrtimer_base(timer, base);
816 if (mode == HRTIMER_MODE_REL) {
817 tim = ktime_add(tim, new_base->get_time());
819 * CONFIG_TIME_LOW_RES is a temporary way for architectures
820 * to signal that they simply return xtime in
821 * do_gettimeoffset(). In this case we want to round up by
822 * resolution when starting a relative timer, to avoid short
823 * timeouts. This will go away with the GTOD framework.
825 #ifdef CONFIG_TIME_LOW_RES
826 tim = ktime_add(tim, base->resolution);
829 timer->expires = tim;
831 timer_stats_hrtimer_set_start_info(timer);
834 * Only allow reprogramming if the new base is on this CPU.
835 * (it might still be on another CPU if the timer was pending)
837 enqueue_hrtimer(timer, new_base,
838 new_base->cpu_base == &__get_cpu_var(hrtimer_bases));
840 unlock_hrtimer_base(timer, &flags);
844 EXPORT_SYMBOL_GPL(hrtimer_start);
847 * hrtimer_try_to_cancel - try to deactivate a timer
848 * @timer: hrtimer to stop
851 * 0 when the timer was not active
852 * 1 when the timer was active
853 * -1 when the timer is currently excuting the callback function and
856 int hrtimer_try_to_cancel(struct hrtimer *timer)
858 struct hrtimer_clock_base *base;
862 base = lock_hrtimer_base(timer, &flags);
864 if (!hrtimer_callback_running(timer))
865 ret = remove_hrtimer(timer, base);
867 unlock_hrtimer_base(timer, &flags);
872 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
875 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
876 * @timer: the timer to be cancelled
879 * 0 when the timer was not active
880 * 1 when the timer was active
882 int hrtimer_cancel(struct hrtimer *timer)
885 int ret = hrtimer_try_to_cancel(timer);
892 EXPORT_SYMBOL_GPL(hrtimer_cancel);
895 * hrtimer_get_remaining - get remaining time for the timer
896 * @timer: the timer to read
898 ktime_t hrtimer_get_remaining(const struct hrtimer *timer)
900 struct hrtimer_clock_base *base;
904 base = lock_hrtimer_base(timer, &flags);
905 rem = ktime_sub(timer->expires, base->get_time());
906 unlock_hrtimer_base(timer, &flags);
910 EXPORT_SYMBOL_GPL(hrtimer_get_remaining);
912 #if defined(CONFIG_NO_IDLE_HZ) || defined(CONFIG_NO_HZ)
914 * hrtimer_get_next_event - get the time until next expiry event
916 * Returns the delta to the next expiry event or KTIME_MAX if no timer
919 ktime_t hrtimer_get_next_event(void)
921 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
922 struct hrtimer_clock_base *base = cpu_base->clock_base;
923 ktime_t delta, mindelta = { .tv64 = KTIME_MAX };
927 spin_lock_irqsave(&cpu_base->lock, flags);
929 if (!hrtimer_hres_active()) {
930 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
931 struct hrtimer *timer;
936 timer = rb_entry(base->first, struct hrtimer, node);
937 delta.tv64 = timer->expires.tv64;
938 delta = ktime_sub(delta, base->get_time());
939 if (delta.tv64 < mindelta.tv64)
940 mindelta.tv64 = delta.tv64;
944 spin_unlock_irqrestore(&cpu_base->lock, flags);
946 if (mindelta.tv64 < 0)
953 * hrtimer_init - initialize a timer to the given clock
954 * @timer: the timer to be initialized
955 * @clock_id: the clock to be used
956 * @mode: timer mode abs/rel
958 void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
959 enum hrtimer_mode mode)
961 struct hrtimer_cpu_base *cpu_base;
963 memset(timer, 0, sizeof(struct hrtimer));
965 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
967 if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
968 clock_id = CLOCK_MONOTONIC;
970 timer->base = &cpu_base->clock_base[clock_id];
971 hrtimer_init_timer_hres(timer);
973 #ifdef CONFIG_TIMER_STATS
974 timer->start_site = NULL;
975 timer->start_pid = -1;
976 memset(timer->start_comm, 0, TASK_COMM_LEN);
979 EXPORT_SYMBOL_GPL(hrtimer_init);
982 * hrtimer_get_res - get the timer resolution for a clock
983 * @which_clock: which clock to query
984 * @tp: pointer to timespec variable to store the resolution
986 * Store the resolution of the clock selected by @which_clock in the
987 * variable pointed to by @tp.
989 int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp)
991 struct hrtimer_cpu_base *cpu_base;
993 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
994 *tp = ktime_to_timespec(cpu_base->clock_base[which_clock].resolution);
998 EXPORT_SYMBOL_GPL(hrtimer_get_res);
1000 #ifdef CONFIG_HIGH_RES_TIMERS
1003 * High resolution timer interrupt
1004 * Called with interrupts disabled
1006 void hrtimer_interrupt(struct clock_event_device *dev)
1008 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1009 struct hrtimer_clock_base *base;
1010 ktime_t expires_next, now;
1013 BUG_ON(!cpu_base->hres_active);
1014 cpu_base->nr_events++;
1015 dev->next_event.tv64 = KTIME_MAX;
1020 expires_next.tv64 = KTIME_MAX;
1022 base = cpu_base->clock_base;
1024 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1026 struct rb_node *node;
1028 spin_lock(&cpu_base->lock);
1030 basenow = ktime_add(now, base->offset);
1032 while ((node = base->first)) {
1033 struct hrtimer *timer;
1035 timer = rb_entry(node, struct hrtimer, node);
1037 if (basenow.tv64 < timer->expires.tv64) {
1040 expires = ktime_sub(timer->expires,
1042 if (expires.tv64 < expires_next.tv64)
1043 expires_next = expires;
1047 /* Move softirq callbacks to the pending list */
1048 if (timer->cb_mode == HRTIMER_CB_SOFTIRQ) {
1049 __remove_hrtimer(timer, base,
1050 HRTIMER_STATE_PENDING, 0);
1051 list_add_tail(&timer->cb_entry,
1052 &base->cpu_base->cb_pending);
1057 __remove_hrtimer(timer, base,
1058 HRTIMER_STATE_CALLBACK, 0);
1059 timer_stats_account_hrtimer(timer);
1062 * Note: We clear the CALLBACK bit after
1063 * enqueue_hrtimer to avoid reprogramming of
1064 * the event hardware. This happens at the end
1065 * of this function anyway.
1067 if (timer->function(timer) != HRTIMER_NORESTART) {
1068 BUG_ON(timer->state != HRTIMER_STATE_CALLBACK);
1069 enqueue_hrtimer(timer, base, 0);
1071 timer->state &= ~HRTIMER_STATE_CALLBACK;
1073 spin_unlock(&cpu_base->lock);
1077 cpu_base->expires_next = expires_next;
1079 /* Reprogramming necessary ? */
1080 if (expires_next.tv64 != KTIME_MAX) {
1081 if (tick_program_event(expires_next, 0))
1085 /* Raise softirq ? */
1087 raise_softirq(HRTIMER_SOFTIRQ);
1090 static void run_hrtimer_softirq(struct softirq_action *h)
1092 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1094 spin_lock_irq(&cpu_base->lock);
1096 while (!list_empty(&cpu_base->cb_pending)) {
1097 enum hrtimer_restart (*fn)(struct hrtimer *);
1098 struct hrtimer *timer;
1101 timer = list_entry(cpu_base->cb_pending.next,
1102 struct hrtimer, cb_entry);
1104 timer_stats_account_hrtimer(timer);
1106 fn = timer->function;
1107 __remove_hrtimer(timer, timer->base, HRTIMER_STATE_CALLBACK, 0);
1108 spin_unlock_irq(&cpu_base->lock);
1110 restart = fn(timer);
1112 spin_lock_irq(&cpu_base->lock);
1114 timer->state &= ~HRTIMER_STATE_CALLBACK;
1115 if (restart == HRTIMER_RESTART) {
1116 BUG_ON(hrtimer_active(timer));
1118 * Enqueue the timer, allow reprogramming of the event
1121 enqueue_hrtimer(timer, timer->base, 1);
1122 } else if (hrtimer_active(timer)) {
1124 * If the timer was rearmed on another CPU, reprogram
1127 if (timer->base->first == &timer->node)
1128 hrtimer_reprogram(timer, timer->base);
1131 spin_unlock_irq(&cpu_base->lock);
1134 #endif /* CONFIG_HIGH_RES_TIMERS */
1137 * Expire the per base hrtimer-queue:
1139 static inline void run_hrtimer_queue(struct hrtimer_cpu_base *cpu_base,
1142 struct rb_node *node;
1143 struct hrtimer_clock_base *base = &cpu_base->clock_base[index];
1148 if (base->get_softirq_time)
1149 base->softirq_time = base->get_softirq_time();
1151 spin_lock_irq(&cpu_base->lock);
1153 while ((node = base->first)) {
1154 struct hrtimer *timer;
1155 enum hrtimer_restart (*fn)(struct hrtimer *);
1158 timer = rb_entry(node, struct hrtimer, node);
1159 if (base->softirq_time.tv64 <= timer->expires.tv64)
1162 #ifdef CONFIG_HIGH_RES_TIMERS
1163 WARN_ON_ONCE(timer->cb_mode == HRTIMER_CB_IRQSAFE_NO_SOFTIRQ);
1165 timer_stats_account_hrtimer(timer);
1167 fn = timer->function;
1168 __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0);
1169 spin_unlock_irq(&cpu_base->lock);
1171 restart = fn(timer);
1173 spin_lock_irq(&cpu_base->lock);
1175 timer->state &= ~HRTIMER_STATE_CALLBACK;
1176 if (restart != HRTIMER_NORESTART) {
1177 BUG_ON(hrtimer_active(timer));
1178 enqueue_hrtimer(timer, base, 0);
1181 spin_unlock_irq(&cpu_base->lock);
1185 * Called from timer softirq every jiffy, expire hrtimers:
1187 * For HRT its the fall back code to run the softirq in the timer
1188 * softirq context in case the hrtimer initialization failed or has
1189 * not been done yet.
1191 void hrtimer_run_queues(void)
1193 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1196 if (hrtimer_hres_active())
1200 * This _is_ ugly: We have to check in the softirq context,
1201 * whether we can switch to highres and / or nohz mode. The
1202 * clocksource switch happens in the timer interrupt with
1203 * xtime_lock held. Notification from there only sets the
1204 * check bit in the tick_oneshot code, otherwise we might
1205 * deadlock vs. xtime_lock.
1207 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1208 if (hrtimer_switch_to_hres())
1211 hrtimer_get_softirq_time(cpu_base);
1213 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
1214 run_hrtimer_queue(cpu_base, i);
1218 * Sleep related functions:
1220 static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1222 struct hrtimer_sleeper *t =
1223 container_of(timer, struct hrtimer_sleeper, timer);
1224 struct task_struct *task = t->task;
1228 wake_up_process(task);
1230 return HRTIMER_NORESTART;
1233 void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1235 sl->timer.function = hrtimer_wakeup;
1237 #ifdef CONFIG_HIGH_RES_TIMERS
1238 sl->timer.cb_mode = HRTIMER_CB_IRQSAFE_NO_RESTART;
1242 static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1244 hrtimer_init_sleeper(t, current);
1247 set_current_state(TASK_INTERRUPTIBLE);
1248 hrtimer_start(&t->timer, t->timer.expires, mode);
1250 if (likely(t->task))
1253 hrtimer_cancel(&t->timer);
1254 mode = HRTIMER_MODE_ABS;
1256 } while (t->task && !signal_pending(current));
1258 return t->task == NULL;
1261 long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1263 struct hrtimer_sleeper t;
1264 struct timespec __user *rmtp;
1268 restart->fn = do_no_restart_syscall;
1270 hrtimer_init(&t.timer, restart->arg0, HRTIMER_MODE_ABS);
1271 t.timer.expires.tv64 = ((u64)restart->arg3 << 32) | (u64) restart->arg2;
1273 if (do_nanosleep(&t, HRTIMER_MODE_ABS))
1276 rmtp = (struct timespec __user *) restart->arg1;
1278 time = ktime_sub(t.timer.expires, t.timer.base->get_time());
1281 tu = ktime_to_timespec(time);
1282 if (copy_to_user(rmtp, &tu, sizeof(tu)))
1286 restart->fn = hrtimer_nanosleep_restart;
1288 /* The other values in restart are already filled in */
1289 return -ERESTART_RESTARTBLOCK;
1292 long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
1293 const enum hrtimer_mode mode, const clockid_t clockid)
1295 struct restart_block *restart;
1296 struct hrtimer_sleeper t;
1300 hrtimer_init(&t.timer, clockid, mode);
1301 t.timer.expires = timespec_to_ktime(*rqtp);
1302 if (do_nanosleep(&t, mode))
1305 /* Absolute timers do not update the rmtp value and restart: */
1306 if (mode == HRTIMER_MODE_ABS)
1307 return -ERESTARTNOHAND;
1310 rem = ktime_sub(t.timer.expires, t.timer.base->get_time());
1313 tu = ktime_to_timespec(rem);
1314 if (copy_to_user(rmtp, &tu, sizeof(tu)))
1318 restart = ¤t_thread_info()->restart_block;
1319 restart->fn = hrtimer_nanosleep_restart;
1320 restart->arg0 = (unsigned long) t.timer.base->index;
1321 restart->arg1 = (unsigned long) rmtp;
1322 restart->arg2 = t.timer.expires.tv64 & 0xFFFFFFFF;
1323 restart->arg3 = t.timer.expires.tv64 >> 32;
1325 return -ERESTART_RESTARTBLOCK;
1329 sys_nanosleep(struct timespec __user *rqtp, struct timespec __user *rmtp)
1333 if (copy_from_user(&tu, rqtp, sizeof(tu)))
1336 if (!timespec_valid(&tu))
1339 return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1343 * Functions related to boot-time initialization:
1345 static void __devinit init_hrtimers_cpu(int cpu)
1347 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1350 spin_lock_init(&cpu_base->lock);
1351 lockdep_set_class(&cpu_base->lock, &cpu_base->lock_key);
1353 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
1354 cpu_base->clock_base[i].cpu_base = cpu_base;
1356 hrtimer_init_hres(cpu_base);
1359 #ifdef CONFIG_HOTPLUG_CPU
1361 static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1362 struct hrtimer_clock_base *new_base)
1364 struct hrtimer *timer;
1365 struct rb_node *node;
1367 while ((node = rb_first(&old_base->active))) {
1368 timer = rb_entry(node, struct hrtimer, node);
1369 BUG_ON(hrtimer_callback_running(timer));
1370 __remove_hrtimer(timer, old_base, HRTIMER_STATE_INACTIVE, 0);
1371 timer->base = new_base;
1373 * Enqueue the timer. Allow reprogramming of the event device
1375 enqueue_hrtimer(timer, new_base, 1);
1379 static void migrate_hrtimers(int cpu)
1381 struct hrtimer_cpu_base *old_base, *new_base;
1384 BUG_ON(cpu_online(cpu));
1385 old_base = &per_cpu(hrtimer_bases, cpu);
1386 new_base = &get_cpu_var(hrtimer_bases);
1388 tick_cancel_sched_timer(cpu);
1390 local_irq_disable();
1391 double_spin_lock(&new_base->lock, &old_base->lock,
1392 smp_processor_id() < cpu);
1394 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1395 migrate_hrtimer_list(&old_base->clock_base[i],
1396 &new_base->clock_base[i]);
1399 double_spin_unlock(&new_base->lock, &old_base->lock,
1400 smp_processor_id() < cpu);
1402 put_cpu_var(hrtimer_bases);
1404 #endif /* CONFIG_HOTPLUG_CPU */
1406 static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self,
1407 unsigned long action, void *hcpu)
1409 long cpu = (long)hcpu;
1413 case CPU_UP_PREPARE:
1414 init_hrtimers_cpu(cpu);
1417 #ifdef CONFIG_HOTPLUG_CPU
1419 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &cpu);
1420 migrate_hrtimers(cpu);
1431 static struct notifier_block __cpuinitdata hrtimers_nb = {
1432 .notifier_call = hrtimer_cpu_notify,
1435 void __init hrtimers_init(void)
1437 hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
1438 (void *)(long)smp_processor_id());
1439 register_cpu_notifier(&hrtimers_nb);
1440 #ifdef CONFIG_HIGH_RES_TIMERS
1441 open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq, NULL);