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);
64 * ktime_get_real - get the real (wall-) time in ktime_t format
66 * returns the time in ktime_t format
68 ktime_t ktime_get_real(void)
74 return timespec_to_ktime(now);
77 EXPORT_SYMBOL_GPL(ktime_get_real);
82 * Note: If we want to add new timer bases, we have to skip the two
83 * clock ids captured by the cpu-timers. We do this by holding empty
84 * entries rather than doing math adjustment of the clock ids.
85 * This ensures that we capture erroneous accesses to these clock ids
86 * rather than moving them into the range of valid clock id's.
88 DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
94 .index = CLOCK_REALTIME,
95 .get_time = &ktime_get_real,
96 .resolution = KTIME_LOW_RES,
99 .index = CLOCK_MONOTONIC,
100 .get_time = &ktime_get,
101 .resolution = KTIME_LOW_RES,
107 * ktime_get_ts - get the monotonic clock in timespec format
108 * @ts: pointer to timespec variable
110 * The function calculates the monotonic clock from the realtime
111 * clock and the wall_to_monotonic offset and stores the result
112 * in normalized timespec format in the variable pointed to by @ts.
114 void ktime_get_ts(struct timespec *ts)
116 struct timespec tomono;
120 seq = read_seqbegin(&xtime_lock);
122 tomono = wall_to_monotonic;
124 } while (read_seqretry(&xtime_lock, seq));
126 set_normalized_timespec(ts, ts->tv_sec + tomono.tv_sec,
127 ts->tv_nsec + tomono.tv_nsec);
129 EXPORT_SYMBOL_GPL(ktime_get_ts);
132 * Get the coarse grained time at the softirq based on xtime and
135 static void hrtimer_get_softirq_time(struct hrtimer_cpu_base *base)
137 ktime_t xtim, tomono;
142 seq = read_seqbegin(&xtime_lock);
144 getnstimeofday(&xts);
148 } while (read_seqretry(&xtime_lock, seq));
150 xtim = timespec_to_ktime(xts);
151 tomono = timespec_to_ktime(wall_to_monotonic);
152 base->clock_base[CLOCK_REALTIME].softirq_time = xtim;
153 base->clock_base[CLOCK_MONOTONIC].softirq_time =
154 ktime_add(xtim, tomono);
158 * Helper function to check, whether the timer is running the callback
161 static inline int hrtimer_callback_running(struct hrtimer *timer)
163 return timer->state & HRTIMER_STATE_CALLBACK;
167 * Functions and macros which are different for UP/SMP systems are kept in a
173 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
174 * means that all timers which are tied to this base via timer->base are
175 * locked, and the base itself is locked too.
177 * So __run_timers/migrate_timers can safely modify all timers which could
178 * be found on the lists/queues.
180 * When the timer's base is locked, and the timer removed from list, it is
181 * possible to set timer->base = NULL and drop the lock: the timer remains
185 struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
186 unsigned long *flags)
188 struct hrtimer_clock_base *base;
192 if (likely(base != NULL)) {
193 spin_lock_irqsave(&base->cpu_base->lock, *flags);
194 if (likely(base == timer->base))
196 /* The timer has migrated to another CPU: */
197 spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
204 * Switch the timer base to the current CPU when possible.
206 static inline struct hrtimer_clock_base *
207 switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base)
209 struct hrtimer_clock_base *new_base;
210 struct hrtimer_cpu_base *new_cpu_base;
212 new_cpu_base = &__get_cpu_var(hrtimer_bases);
213 new_base = &new_cpu_base->clock_base[base->index];
215 if (base != new_base) {
217 * We are trying to schedule the timer on the local CPU.
218 * However we can't change timer's base while it is running,
219 * so we keep it on the same CPU. No hassle vs. reprogramming
220 * the event source in the high resolution case. The softirq
221 * code will take care of this when the timer function has
222 * completed. There is no conflict as we hold the lock until
223 * the timer is enqueued.
225 if (unlikely(hrtimer_callback_running(timer)))
228 /* See the comment in lock_timer_base() */
230 spin_unlock(&base->cpu_base->lock);
231 spin_lock(&new_base->cpu_base->lock);
232 timer->base = new_base;
237 #else /* CONFIG_SMP */
239 static inline struct hrtimer_clock_base *
240 lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
242 struct hrtimer_clock_base *base = timer->base;
244 spin_lock_irqsave(&base->cpu_base->lock, *flags);
249 # define switch_hrtimer_base(t, b) (b)
251 #endif /* !CONFIG_SMP */
254 * Functions for the union type storage format of ktime_t which are
255 * too large for inlining:
257 #if BITS_PER_LONG < 64
258 # ifndef CONFIG_KTIME_SCALAR
260 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
262 * @nsec: the scalar nsec value to add
264 * Returns the sum of kt and nsec in ktime_t format
266 ktime_t ktime_add_ns(const ktime_t kt, u64 nsec)
270 if (likely(nsec < NSEC_PER_SEC)) {
273 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
275 tmp = ktime_set((long)nsec, rem);
278 return ktime_add(kt, tmp);
280 # endif /* !CONFIG_KTIME_SCALAR */
283 * Divide a ktime value by a nanosecond value
285 unsigned long ktime_divns(const ktime_t kt, s64 div)
290 dclc = dns = ktime_to_ns(kt);
292 /* Make sure the divisor is less than 2^32: */
298 do_div(dclc, (unsigned long) div);
300 return (unsigned long) dclc;
302 #endif /* BITS_PER_LONG >= 64 */
304 /* High resolution timer related functions */
305 #ifdef CONFIG_HIGH_RES_TIMERS
308 * High resolution timer enabled ?
310 static int hrtimer_hres_enabled __read_mostly = 1;
313 * Enable / Disable high resolution mode
315 static int __init setup_hrtimer_hres(char *str)
317 if (!strcmp(str, "off"))
318 hrtimer_hres_enabled = 0;
319 else if (!strcmp(str, "on"))
320 hrtimer_hres_enabled = 1;
326 __setup("highres=", setup_hrtimer_hres);
329 * hrtimer_high_res_enabled - query, if the highres mode is enabled
331 static inline int hrtimer_is_hres_enabled(void)
333 return hrtimer_hres_enabled;
337 * Is the high resolution mode active ?
339 static inline int hrtimer_hres_active(void)
341 return __get_cpu_var(hrtimer_bases).hres_active;
345 * Reprogram the event source with checking both queues for the
347 * Called with interrupts disabled and base->lock held
349 static void hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base)
352 struct hrtimer_clock_base *base = cpu_base->clock_base;
355 cpu_base->expires_next.tv64 = KTIME_MAX;
357 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
358 struct hrtimer *timer;
362 timer = rb_entry(base->first, struct hrtimer, node);
363 expires = ktime_sub(timer->expires, base->offset);
364 if (expires.tv64 < cpu_base->expires_next.tv64)
365 cpu_base->expires_next = expires;
368 if (cpu_base->expires_next.tv64 != KTIME_MAX)
369 tick_program_event(cpu_base->expires_next, 1);
373 * Shared reprogramming for clock_realtime and clock_monotonic
375 * When a timer is enqueued and expires earlier than the already enqueued
376 * timers, we have to check, whether it expires earlier than the timer for
377 * which the clock event device was armed.
379 * Called with interrupts disabled and base->cpu_base.lock held
381 static int hrtimer_reprogram(struct hrtimer *timer,
382 struct hrtimer_clock_base *base)
384 ktime_t *expires_next = &__get_cpu_var(hrtimer_bases).expires_next;
385 ktime_t expires = ktime_sub(timer->expires, base->offset);
389 * When the callback is running, we do not reprogram the clock event
390 * device. The timer callback is either running on a different CPU or
391 * the callback is executed in the hrtimer_interupt context. The
392 * reprogramming is handled either by the softirq, which called the
393 * callback or at the end of the hrtimer_interrupt.
395 if (hrtimer_callback_running(timer))
398 if (expires.tv64 >= expires_next->tv64)
402 * Clockevents returns -ETIME, when the event was in the past.
404 res = tick_program_event(expires, 0);
405 if (!IS_ERR_VALUE(res))
406 *expires_next = expires;
412 * Retrigger next event is called after clock was set
414 * Called with interrupts disabled via on_each_cpu()
416 static void retrigger_next_event(void *arg)
418 struct hrtimer_cpu_base *base;
419 struct timespec realtime_offset;
422 if (!hrtimer_hres_active())
426 seq = read_seqbegin(&xtime_lock);
427 set_normalized_timespec(&realtime_offset,
428 -wall_to_monotonic.tv_sec,
429 -wall_to_monotonic.tv_nsec);
430 } while (read_seqretry(&xtime_lock, seq));
432 base = &__get_cpu_var(hrtimer_bases);
434 /* Adjust CLOCK_REALTIME offset */
435 spin_lock(&base->lock);
436 base->clock_base[CLOCK_REALTIME].offset =
437 timespec_to_ktime(realtime_offset);
439 hrtimer_force_reprogram(base);
440 spin_unlock(&base->lock);
444 * Clock realtime was set
446 * Change the offset of the realtime clock vs. the monotonic
449 * We might have to reprogram the high resolution timer interrupt. On
450 * SMP we call the architecture specific code to retrigger _all_ high
451 * resolution timer interrupts. On UP we just disable interrupts and
452 * call the high resolution interrupt code.
454 void clock_was_set(void)
456 /* Retrigger the CPU local events everywhere */
457 on_each_cpu(retrigger_next_event, NULL, 0, 1);
461 * Check, whether the timer is on the callback pending list
463 static inline int hrtimer_cb_pending(const struct hrtimer *timer)
465 return timer->state & HRTIMER_STATE_PENDING;
469 * Remove a timer from the callback pending list
471 static inline void hrtimer_remove_cb_pending(struct hrtimer *timer)
473 list_del_init(&timer->cb_entry);
477 * Initialize the high resolution related parts of cpu_base
479 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
481 base->expires_next.tv64 = KTIME_MAX;
482 base->hres_active = 0;
483 INIT_LIST_HEAD(&base->cb_pending);
487 * Initialize the high resolution related parts of a hrtimer
489 static inline void hrtimer_init_timer_hres(struct hrtimer *timer)
491 INIT_LIST_HEAD(&timer->cb_entry);
495 * When High resolution timers are active, try to reprogram. Note, that in case
496 * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry
497 * check happens. The timer gets enqueued into the rbtree. The reprogramming
498 * and expiry check is done in the hrtimer_interrupt or in the softirq.
500 static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
501 struct hrtimer_clock_base *base)
503 if (base->cpu_base->hres_active && hrtimer_reprogram(timer, base)) {
505 /* Timer is expired, act upon the callback mode */
506 switch(timer->cb_mode) {
507 case HRTIMER_CB_IRQSAFE_NO_RESTART:
509 * We can call the callback from here. No restart
510 * happens, so no danger of recursion
512 BUG_ON(timer->function(timer) != HRTIMER_NORESTART);
514 case HRTIMER_CB_IRQSAFE_NO_SOFTIRQ:
516 * This is solely for the sched tick emulation with
517 * dynamic tick support to ensure that we do not
518 * restart the tick right on the edge and end up with
519 * the tick timer in the softirq ! The calling site
520 * takes care of this.
523 case HRTIMER_CB_IRQSAFE:
524 case HRTIMER_CB_SOFTIRQ:
526 * Move everything else into the softirq pending list !
528 list_add_tail(&timer->cb_entry,
529 &base->cpu_base->cb_pending);
530 timer->state = HRTIMER_STATE_PENDING;
531 raise_softirq(HRTIMER_SOFTIRQ);
541 * Switch to high resolution mode
543 static void hrtimer_switch_to_hres(void)
545 struct hrtimer_cpu_base *base = &__get_cpu_var(hrtimer_bases);
548 if (base->hres_active)
551 local_irq_save(flags);
553 if (tick_init_highres()) {
554 local_irq_restore(flags);
557 base->hres_active = 1;
558 base->clock_base[CLOCK_REALTIME].resolution = KTIME_HIGH_RES;
559 base->clock_base[CLOCK_MONOTONIC].resolution = KTIME_HIGH_RES;
561 tick_setup_sched_timer();
563 /* "Retrigger" the interrupt to get things going */
564 retrigger_next_event(NULL);
565 local_irq_restore(flags);
566 printk(KERN_INFO "Switched to high resolution mode on CPU %d\n",
572 static inline int hrtimer_hres_active(void) { return 0; }
573 static inline int hrtimer_is_hres_enabled(void) { return 0; }
574 static inline void hrtimer_switch_to_hres(void) { }
575 static inline void hrtimer_force_reprogram(struct hrtimer_cpu_base *base) { }
576 static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
577 struct hrtimer_clock_base *base)
581 static inline int hrtimer_cb_pending(struct hrtimer *timer) { return 0; }
582 static inline void hrtimer_remove_cb_pending(struct hrtimer *timer) { }
583 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { }
584 static inline void hrtimer_init_timer_hres(struct hrtimer *timer) { }
586 #endif /* CONFIG_HIGH_RES_TIMERS */
589 * Counterpart to lock_timer_base above:
592 void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
594 spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
598 * hrtimer_forward - forward the timer expiry
599 * @timer: hrtimer to forward
600 * @now: forward past this time
601 * @interval: the interval to forward
603 * Forward the timer expiry so it will expire in the future.
604 * Returns the number of overruns.
607 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
609 unsigned long orun = 1;
612 delta = ktime_sub(now, timer->expires);
617 if (interval.tv64 < timer->base->resolution.tv64)
618 interval.tv64 = timer->base->resolution.tv64;
620 if (unlikely(delta.tv64 >= interval.tv64)) {
621 s64 incr = ktime_to_ns(interval);
623 orun = ktime_divns(delta, incr);
624 timer->expires = ktime_add_ns(timer->expires, incr * orun);
625 if (timer->expires.tv64 > now.tv64)
628 * This (and the ktime_add() below) is the
629 * correction for exact:
633 timer->expires = ktime_add(timer->expires, interval);
639 * enqueue_hrtimer - internal function to (re)start a timer
641 * The timer is inserted in expiry order. Insertion into the
642 * red black tree is O(log(n)). Must hold the base lock.
644 static void enqueue_hrtimer(struct hrtimer *timer,
645 struct hrtimer_clock_base *base, int reprogram)
647 struct rb_node **link = &base->active.rb_node;
648 struct rb_node *parent = NULL;
649 struct hrtimer *entry;
652 * Find the right place in the rbtree:
656 entry = rb_entry(parent, struct hrtimer, node);
658 * We dont care about collisions. Nodes with
659 * the same expiry time stay together.
661 if (timer->expires.tv64 < entry->expires.tv64)
662 link = &(*link)->rb_left;
664 link = &(*link)->rb_right;
668 * Insert the timer to the rbtree and check whether it
669 * replaces the first pending timer
671 if (!base->first || timer->expires.tv64 <
672 rb_entry(base->first, struct hrtimer, node)->expires.tv64) {
674 * Reprogram the clock event device. When the timer is already
675 * expired hrtimer_enqueue_reprogram has either called the
676 * callback or added it to the pending list and raised the
679 * This is a NOP for !HIGHRES
681 if (reprogram && hrtimer_enqueue_reprogram(timer, base))
684 base->first = &timer->node;
687 rb_link_node(&timer->node, parent, link);
688 rb_insert_color(&timer->node, &base->active);
690 * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
691 * state of a possibly running callback.
693 timer->state |= HRTIMER_STATE_ENQUEUED;
697 * __remove_hrtimer - internal function to remove a timer
699 * Caller must hold the base lock.
701 * High resolution timer mode reprograms the clock event device when the
702 * timer is the one which expires next. The caller can disable this by setting
703 * reprogram to zero. This is useful, when the context does a reprogramming
704 * anyway (e.g. timer interrupt)
706 static void __remove_hrtimer(struct hrtimer *timer,
707 struct hrtimer_clock_base *base,
708 unsigned long newstate, int reprogram)
710 /* High res. callback list. NOP for !HIGHRES */
711 if (hrtimer_cb_pending(timer))
712 hrtimer_remove_cb_pending(timer);
715 * Remove the timer from the rbtree and replace the
716 * first entry pointer if necessary.
718 if (base->first == &timer->node) {
719 base->first = rb_next(&timer->node);
720 /* Reprogram the clock event device. if enabled */
721 if (reprogram && hrtimer_hres_active())
722 hrtimer_force_reprogram(base->cpu_base);
724 rb_erase(&timer->node, &base->active);
726 timer->state = newstate;
730 * remove hrtimer, called with base lock held
733 remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base)
735 if (hrtimer_is_queued(timer)) {
739 * Remove the timer and force reprogramming when high
740 * resolution mode is active and the timer is on the current
741 * CPU. If we remove a timer on another CPU, reprogramming is
742 * skipped. The interrupt event on this CPU is fired and
743 * reprogramming happens in the interrupt handler. This is a
744 * rare case and less expensive than a smp call.
746 reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases);
747 __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE,
755 * hrtimer_start - (re)start an relative timer on the current CPU
756 * @timer: the timer to be added
758 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
762 * 1 when the timer was active
765 hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode)
767 struct hrtimer_clock_base *base, *new_base;
771 base = lock_hrtimer_base(timer, &flags);
773 /* Remove an active timer from the queue: */
774 ret = remove_hrtimer(timer, base);
776 /* Switch the timer base, if necessary: */
777 new_base = switch_hrtimer_base(timer, base);
779 if (mode == HRTIMER_MODE_REL) {
780 tim = ktime_add(tim, new_base->get_time());
782 * CONFIG_TIME_LOW_RES is a temporary way for architectures
783 * to signal that they simply return xtime in
784 * do_gettimeoffset(). In this case we want to round up by
785 * resolution when starting a relative timer, to avoid short
786 * timeouts. This will go away with the GTOD framework.
788 #ifdef CONFIG_TIME_LOW_RES
789 tim = ktime_add(tim, base->resolution);
792 timer->expires = tim;
794 enqueue_hrtimer(timer, new_base, base == new_base);
796 unlock_hrtimer_base(timer, &flags);
800 EXPORT_SYMBOL_GPL(hrtimer_start);
803 * hrtimer_try_to_cancel - try to deactivate a timer
804 * @timer: hrtimer to stop
807 * 0 when the timer was not active
808 * 1 when the timer was active
809 * -1 when the timer is currently excuting the callback function and
812 int hrtimer_try_to_cancel(struct hrtimer *timer)
814 struct hrtimer_clock_base *base;
818 base = lock_hrtimer_base(timer, &flags);
820 if (!hrtimer_callback_running(timer))
821 ret = remove_hrtimer(timer, base);
823 unlock_hrtimer_base(timer, &flags);
828 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
831 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
832 * @timer: the timer to be cancelled
835 * 0 when the timer was not active
836 * 1 when the timer was active
838 int hrtimer_cancel(struct hrtimer *timer)
841 int ret = hrtimer_try_to_cancel(timer);
848 EXPORT_SYMBOL_GPL(hrtimer_cancel);
851 * hrtimer_get_remaining - get remaining time for the timer
852 * @timer: the timer to read
854 ktime_t hrtimer_get_remaining(const struct hrtimer *timer)
856 struct hrtimer_clock_base *base;
860 base = lock_hrtimer_base(timer, &flags);
861 rem = ktime_sub(timer->expires, base->get_time());
862 unlock_hrtimer_base(timer, &flags);
866 EXPORT_SYMBOL_GPL(hrtimer_get_remaining);
868 #if defined(CONFIG_NO_IDLE_HZ) || defined(CONFIG_NO_HZ)
870 * hrtimer_get_next_event - get the time until next expiry event
872 * Returns the delta to the next expiry event or KTIME_MAX if no timer
875 ktime_t hrtimer_get_next_event(void)
877 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
878 struct hrtimer_clock_base *base = cpu_base->clock_base;
879 ktime_t delta, mindelta = { .tv64 = KTIME_MAX };
883 spin_lock_irqsave(&cpu_base->lock, flags);
885 if (!hrtimer_hres_active()) {
886 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
887 struct hrtimer *timer;
892 timer = rb_entry(base->first, struct hrtimer, node);
893 delta.tv64 = timer->expires.tv64;
894 delta = ktime_sub(delta, base->get_time());
895 if (delta.tv64 < mindelta.tv64)
896 mindelta.tv64 = delta.tv64;
900 spin_unlock_irqrestore(&cpu_base->lock, flags);
902 if (mindelta.tv64 < 0)
909 * hrtimer_init - initialize a timer to the given clock
910 * @timer: the timer to be initialized
911 * @clock_id: the clock to be used
912 * @mode: timer mode abs/rel
914 void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
915 enum hrtimer_mode mode)
917 struct hrtimer_cpu_base *cpu_base;
919 memset(timer, 0, sizeof(struct hrtimer));
921 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
923 if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
924 clock_id = CLOCK_MONOTONIC;
926 timer->base = &cpu_base->clock_base[clock_id];
927 hrtimer_init_timer_hres(timer);
929 EXPORT_SYMBOL_GPL(hrtimer_init);
932 * hrtimer_get_res - get the timer resolution for a clock
933 * @which_clock: which clock to query
934 * @tp: pointer to timespec variable to store the resolution
936 * Store the resolution of the clock selected by @which_clock in the
937 * variable pointed to by @tp.
939 int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp)
941 struct hrtimer_cpu_base *cpu_base;
943 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
944 *tp = ktime_to_timespec(cpu_base->clock_base[which_clock].resolution);
948 EXPORT_SYMBOL_GPL(hrtimer_get_res);
950 #ifdef CONFIG_HIGH_RES_TIMERS
953 * High resolution timer interrupt
954 * Called with interrupts disabled
956 void hrtimer_interrupt(struct clock_event_device *dev)
958 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
959 struct hrtimer_clock_base *base;
960 ktime_t expires_next, now;
963 BUG_ON(!cpu_base->hres_active);
964 cpu_base->nr_events++;
965 dev->next_event.tv64 = KTIME_MAX;
970 expires_next.tv64 = KTIME_MAX;
972 base = cpu_base->clock_base;
974 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
976 struct rb_node *node;
978 spin_lock(&cpu_base->lock);
980 basenow = ktime_add(now, base->offset);
982 while ((node = base->first)) {
983 struct hrtimer *timer;
985 timer = rb_entry(node, struct hrtimer, node);
987 if (basenow.tv64 < timer->expires.tv64) {
990 expires = ktime_sub(timer->expires,
992 if (expires.tv64 < expires_next.tv64)
993 expires_next = expires;
997 /* Move softirq callbacks to the pending list */
998 if (timer->cb_mode == HRTIMER_CB_SOFTIRQ) {
999 __remove_hrtimer(timer, base,
1000 HRTIMER_STATE_PENDING, 0);
1001 list_add_tail(&timer->cb_entry,
1002 &base->cpu_base->cb_pending);
1007 __remove_hrtimer(timer, base,
1008 HRTIMER_STATE_CALLBACK, 0);
1011 * Note: We clear the CALLBACK bit after
1012 * enqueue_hrtimer to avoid reprogramming of
1013 * the event hardware. This happens at the end
1014 * of this function anyway.
1016 if (timer->function(timer) != HRTIMER_NORESTART) {
1017 BUG_ON(timer->state != HRTIMER_STATE_CALLBACK);
1018 enqueue_hrtimer(timer, base, 0);
1020 timer->state &= ~HRTIMER_STATE_CALLBACK;
1022 spin_unlock(&cpu_base->lock);
1026 cpu_base->expires_next = expires_next;
1028 /* Reprogramming necessary ? */
1029 if (expires_next.tv64 != KTIME_MAX) {
1030 if (tick_program_event(expires_next, 0))
1034 /* Raise softirq ? */
1036 raise_softirq(HRTIMER_SOFTIRQ);
1039 static void run_hrtimer_softirq(struct softirq_action *h)
1041 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1043 spin_lock_irq(&cpu_base->lock);
1045 while (!list_empty(&cpu_base->cb_pending)) {
1046 enum hrtimer_restart (*fn)(struct hrtimer *);
1047 struct hrtimer *timer;
1050 timer = list_entry(cpu_base->cb_pending.next,
1051 struct hrtimer, cb_entry);
1053 fn = timer->function;
1054 __remove_hrtimer(timer, timer->base, HRTIMER_STATE_CALLBACK, 0);
1055 spin_unlock_irq(&cpu_base->lock);
1057 restart = fn(timer);
1059 spin_lock_irq(&cpu_base->lock);
1061 timer->state &= ~HRTIMER_STATE_CALLBACK;
1062 if (restart == HRTIMER_RESTART) {
1063 BUG_ON(hrtimer_active(timer));
1065 * Enqueue the timer, allow reprogramming of the event
1068 enqueue_hrtimer(timer, timer->base, 1);
1069 } else if (hrtimer_active(timer)) {
1071 * If the timer was rearmed on another CPU, reprogram
1074 if (timer->base->first == &timer->node)
1075 hrtimer_reprogram(timer, timer->base);
1078 spin_unlock_irq(&cpu_base->lock);
1081 #endif /* CONFIG_HIGH_RES_TIMERS */
1084 * Expire the per base hrtimer-queue:
1086 static inline void run_hrtimer_queue(struct hrtimer_cpu_base *cpu_base,
1089 struct rb_node *node;
1090 struct hrtimer_clock_base *base = &cpu_base->clock_base[index];
1095 if (base->get_softirq_time)
1096 base->softirq_time = base->get_softirq_time();
1098 spin_lock_irq(&cpu_base->lock);
1100 while ((node = base->first)) {
1101 struct hrtimer *timer;
1102 enum hrtimer_restart (*fn)(struct hrtimer *);
1105 timer = rb_entry(node, struct hrtimer, node);
1106 if (base->softirq_time.tv64 <= timer->expires.tv64)
1109 fn = timer->function;
1110 __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0);
1111 spin_unlock_irq(&cpu_base->lock);
1113 restart = fn(timer);
1115 spin_lock_irq(&cpu_base->lock);
1117 timer->state &= ~HRTIMER_STATE_CALLBACK;
1118 if (restart != HRTIMER_NORESTART) {
1119 BUG_ON(hrtimer_active(timer));
1120 enqueue_hrtimer(timer, base, 0);
1123 spin_unlock_irq(&cpu_base->lock);
1127 * Called from timer softirq every jiffy, expire hrtimers:
1129 * For HRT its the fall back code to run the softirq in the timer
1130 * softirq context in case the hrtimer initialization failed or has
1131 * not been done yet.
1133 void hrtimer_run_queues(void)
1135 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1138 if (hrtimer_hres_active())
1142 * This _is_ ugly: We have to check in the softirq context,
1143 * whether we can switch to highres and / or nohz mode. The
1144 * clocksource switch happens in the timer interrupt with
1145 * xtime_lock held. Notification from there only sets the
1146 * check bit in the tick_oneshot code, otherwise we might
1147 * deadlock vs. xtime_lock.
1149 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1150 hrtimer_switch_to_hres();
1152 hrtimer_get_softirq_time(cpu_base);
1154 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
1155 run_hrtimer_queue(cpu_base, i);
1159 * Sleep related functions:
1161 static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1163 struct hrtimer_sleeper *t =
1164 container_of(timer, struct hrtimer_sleeper, timer);
1165 struct task_struct *task = t->task;
1169 wake_up_process(task);
1171 return HRTIMER_NORESTART;
1174 void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1176 sl->timer.function = hrtimer_wakeup;
1178 #ifdef CONFIG_HIGH_RES_TIMERS
1179 sl->timer.cb_mode = HRTIMER_CB_IRQSAFE_NO_RESTART;
1183 static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1185 hrtimer_init_sleeper(t, current);
1188 set_current_state(TASK_INTERRUPTIBLE);
1189 hrtimer_start(&t->timer, t->timer.expires, mode);
1191 if (likely(t->task))
1194 hrtimer_cancel(&t->timer);
1195 mode = HRTIMER_MODE_ABS;
1197 } while (t->task && !signal_pending(current));
1199 return t->task == NULL;
1202 long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1204 struct hrtimer_sleeper t;
1205 struct timespec __user *rmtp;
1209 restart->fn = do_no_restart_syscall;
1211 hrtimer_init(&t.timer, restart->arg0, HRTIMER_MODE_ABS);
1212 t.timer.expires.tv64 = ((u64)restart->arg3 << 32) | (u64) restart->arg2;
1214 if (do_nanosleep(&t, HRTIMER_MODE_ABS))
1217 rmtp = (struct timespec __user *) restart->arg1;
1219 time = ktime_sub(t.timer.expires, t.timer.base->get_time());
1222 tu = ktime_to_timespec(time);
1223 if (copy_to_user(rmtp, &tu, sizeof(tu)))
1227 restart->fn = hrtimer_nanosleep_restart;
1229 /* The other values in restart are already filled in */
1230 return -ERESTART_RESTARTBLOCK;
1233 long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
1234 const enum hrtimer_mode mode, const clockid_t clockid)
1236 struct restart_block *restart;
1237 struct hrtimer_sleeper t;
1241 hrtimer_init(&t.timer, clockid, mode);
1242 t.timer.expires = timespec_to_ktime(*rqtp);
1243 if (do_nanosleep(&t, mode))
1246 /* Absolute timers do not update the rmtp value and restart: */
1247 if (mode == HRTIMER_MODE_ABS)
1248 return -ERESTARTNOHAND;
1251 rem = ktime_sub(t.timer.expires, t.timer.base->get_time());
1254 tu = ktime_to_timespec(rem);
1255 if (copy_to_user(rmtp, &tu, sizeof(tu)))
1259 restart = ¤t_thread_info()->restart_block;
1260 restart->fn = hrtimer_nanosleep_restart;
1261 restart->arg0 = (unsigned long) t.timer.base->index;
1262 restart->arg1 = (unsigned long) rmtp;
1263 restart->arg2 = t.timer.expires.tv64 & 0xFFFFFFFF;
1264 restart->arg3 = t.timer.expires.tv64 >> 32;
1266 return -ERESTART_RESTARTBLOCK;
1270 sys_nanosleep(struct timespec __user *rqtp, struct timespec __user *rmtp)
1274 if (copy_from_user(&tu, rqtp, sizeof(tu)))
1277 if (!timespec_valid(&tu))
1280 return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1284 * Functions related to boot-time initialization:
1286 static void __devinit init_hrtimers_cpu(int cpu)
1288 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1291 spin_lock_init(&cpu_base->lock);
1292 lockdep_set_class(&cpu_base->lock, &cpu_base->lock_key);
1294 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
1295 cpu_base->clock_base[i].cpu_base = cpu_base;
1297 hrtimer_init_hres(cpu_base);
1300 #ifdef CONFIG_HOTPLUG_CPU
1302 static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1303 struct hrtimer_clock_base *new_base)
1305 struct hrtimer *timer;
1306 struct rb_node *node;
1308 while ((node = rb_first(&old_base->active))) {
1309 timer = rb_entry(node, struct hrtimer, node);
1310 BUG_ON(hrtimer_callback_running(timer));
1311 __remove_hrtimer(timer, old_base, HRTIMER_STATE_INACTIVE, 0);
1312 timer->base = new_base;
1314 * Enqueue the timer. Allow reprogramming of the event device
1316 enqueue_hrtimer(timer, new_base, 1);
1320 static void migrate_hrtimers(int cpu)
1322 struct hrtimer_cpu_base *old_base, *new_base;
1325 BUG_ON(cpu_online(cpu));
1326 old_base = &per_cpu(hrtimer_bases, cpu);
1327 new_base = &get_cpu_var(hrtimer_bases);
1329 tick_cancel_sched_timer(cpu);
1331 local_irq_disable();
1333 spin_lock(&new_base->lock);
1334 spin_lock(&old_base->lock);
1336 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1337 migrate_hrtimer_list(&old_base->clock_base[i],
1338 &new_base->clock_base[i]);
1340 spin_unlock(&old_base->lock);
1341 spin_unlock(&new_base->lock);
1344 put_cpu_var(hrtimer_bases);
1346 #endif /* CONFIG_HOTPLUG_CPU */
1348 static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self,
1349 unsigned long action, void *hcpu)
1351 long cpu = (long)hcpu;
1355 case CPU_UP_PREPARE:
1356 init_hrtimers_cpu(cpu);
1359 #ifdef CONFIG_HOTPLUG_CPU
1361 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &cpu);
1362 migrate_hrtimers(cpu);
1373 static struct notifier_block __cpuinitdata hrtimers_nb = {
1374 .notifier_call = hrtimer_cpu_notify,
1377 void __init hrtimers_init(void)
1379 hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
1380 (void *)(long)smp_processor_id());
1381 register_cpu_notifier(&hrtimers_nb);
1382 #ifdef CONFIG_HIGH_RES_TIMERS
1383 open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq, NULL);