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);
282 # endif /* !CONFIG_KTIME_SCALAR */
285 * Divide a ktime value by a nanosecond value
287 unsigned long ktime_divns(const ktime_t kt, s64 div)
292 dclc = dns = ktime_to_ns(kt);
294 /* Make sure the divisor is less than 2^32: */
300 do_div(dclc, (unsigned long) div);
302 return (unsigned long) dclc;
304 #endif /* BITS_PER_LONG >= 64 */
306 /* High resolution timer related functions */
307 #ifdef CONFIG_HIGH_RES_TIMERS
310 * High resolution timer enabled ?
312 static int hrtimer_hres_enabled __read_mostly = 1;
315 * Enable / Disable high resolution mode
317 static int __init setup_hrtimer_hres(char *str)
319 if (!strcmp(str, "off"))
320 hrtimer_hres_enabled = 0;
321 else if (!strcmp(str, "on"))
322 hrtimer_hres_enabled = 1;
328 __setup("highres=", setup_hrtimer_hres);
331 * hrtimer_high_res_enabled - query, if the highres mode is enabled
333 static inline int hrtimer_is_hres_enabled(void)
335 return hrtimer_hres_enabled;
339 * Is the high resolution mode active ?
341 static inline int hrtimer_hres_active(void)
343 return __get_cpu_var(hrtimer_bases).hres_active;
347 * Reprogram the event source with checking both queues for the
349 * Called with interrupts disabled and base->lock held
351 static void hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base)
354 struct hrtimer_clock_base *base = cpu_base->clock_base;
357 cpu_base->expires_next.tv64 = KTIME_MAX;
359 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
360 struct hrtimer *timer;
364 timer = rb_entry(base->first, struct hrtimer, node);
365 expires = ktime_sub(timer->expires, base->offset);
366 if (expires.tv64 < cpu_base->expires_next.tv64)
367 cpu_base->expires_next = expires;
370 if (cpu_base->expires_next.tv64 != KTIME_MAX)
371 tick_program_event(cpu_base->expires_next, 1);
375 * Shared reprogramming for clock_realtime and clock_monotonic
377 * When a timer is enqueued and expires earlier than the already enqueued
378 * timers, we have to check, whether it expires earlier than the timer for
379 * which the clock event device was armed.
381 * Called with interrupts disabled and base->cpu_base.lock held
383 static int hrtimer_reprogram(struct hrtimer *timer,
384 struct hrtimer_clock_base *base)
386 ktime_t *expires_next = &__get_cpu_var(hrtimer_bases).expires_next;
387 ktime_t expires = ktime_sub(timer->expires, base->offset);
391 * When the callback is running, we do not reprogram the clock event
392 * device. The timer callback is either running on a different CPU or
393 * the callback is executed in the hrtimer_interupt context. The
394 * reprogramming is handled either by the softirq, which called the
395 * callback or at the end of the hrtimer_interrupt.
397 if (hrtimer_callback_running(timer))
400 if (expires.tv64 >= expires_next->tv64)
404 * Clockevents returns -ETIME, when the event was in the past.
406 res = tick_program_event(expires, 0);
407 if (!IS_ERR_VALUE(res))
408 *expires_next = expires;
414 * Retrigger next event is called after clock was set
416 * Called with interrupts disabled via on_each_cpu()
418 static void retrigger_next_event(void *arg)
420 struct hrtimer_cpu_base *base;
421 struct timespec realtime_offset;
424 if (!hrtimer_hres_active())
428 seq = read_seqbegin(&xtime_lock);
429 set_normalized_timespec(&realtime_offset,
430 -wall_to_monotonic.tv_sec,
431 -wall_to_monotonic.tv_nsec);
432 } while (read_seqretry(&xtime_lock, seq));
434 base = &__get_cpu_var(hrtimer_bases);
436 /* Adjust CLOCK_REALTIME offset */
437 spin_lock(&base->lock);
438 base->clock_base[CLOCK_REALTIME].offset =
439 timespec_to_ktime(realtime_offset);
441 hrtimer_force_reprogram(base);
442 spin_unlock(&base->lock);
446 * Clock realtime was set
448 * Change the offset of the realtime clock vs. the monotonic
451 * We might have to reprogram the high resolution timer interrupt. On
452 * SMP we call the architecture specific code to retrigger _all_ high
453 * resolution timer interrupts. On UP we just disable interrupts and
454 * call the high resolution interrupt code.
456 void clock_was_set(void)
458 /* Retrigger the CPU local events everywhere */
459 on_each_cpu(retrigger_next_event, NULL, 0, 1);
463 * During resume we might have to reprogram the high resolution timer
464 * interrupt (on the local CPU):
466 void hres_timers_resume(void)
468 WARN_ON_ONCE(num_online_cpus() > 1);
470 /* Retrigger the CPU local events: */
471 retrigger_next_event(NULL);
475 * Check, whether the timer is on the callback pending list
477 static inline int hrtimer_cb_pending(const struct hrtimer *timer)
479 return timer->state & HRTIMER_STATE_PENDING;
483 * Remove a timer from the callback pending list
485 static inline void hrtimer_remove_cb_pending(struct hrtimer *timer)
487 list_del_init(&timer->cb_entry);
491 * Initialize the high resolution related parts of cpu_base
493 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
495 base->expires_next.tv64 = KTIME_MAX;
496 base->hres_active = 0;
497 INIT_LIST_HEAD(&base->cb_pending);
501 * Initialize the high resolution related parts of a hrtimer
503 static inline void hrtimer_init_timer_hres(struct hrtimer *timer)
505 INIT_LIST_HEAD(&timer->cb_entry);
509 * When High resolution timers are active, try to reprogram. Note, that in case
510 * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry
511 * check happens. The timer gets enqueued into the rbtree. The reprogramming
512 * and expiry check is done in the hrtimer_interrupt or in the softirq.
514 static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
515 struct hrtimer_clock_base *base)
517 if (base->cpu_base->hres_active && hrtimer_reprogram(timer, base)) {
519 /* Timer is expired, act upon the callback mode */
520 switch(timer->cb_mode) {
521 case HRTIMER_CB_IRQSAFE_NO_RESTART:
523 * We can call the callback from here. No restart
524 * happens, so no danger of recursion
526 BUG_ON(timer->function(timer) != HRTIMER_NORESTART);
528 case HRTIMER_CB_IRQSAFE_NO_SOFTIRQ:
530 * This is solely for the sched tick emulation with
531 * dynamic tick support to ensure that we do not
532 * restart the tick right on the edge and end up with
533 * the tick timer in the softirq ! The calling site
534 * takes care of this.
537 case HRTIMER_CB_IRQSAFE:
538 case HRTIMER_CB_SOFTIRQ:
540 * Move everything else into the softirq pending list !
542 list_add_tail(&timer->cb_entry,
543 &base->cpu_base->cb_pending);
544 timer->state = HRTIMER_STATE_PENDING;
545 raise_softirq(HRTIMER_SOFTIRQ);
555 * Switch to high resolution mode
557 static int hrtimer_switch_to_hres(void)
559 struct hrtimer_cpu_base *base = &__get_cpu_var(hrtimer_bases);
562 if (base->hres_active)
565 local_irq_save(flags);
567 if (tick_init_highres()) {
568 local_irq_restore(flags);
571 base->hres_active = 1;
572 base->clock_base[CLOCK_REALTIME].resolution = KTIME_HIGH_RES;
573 base->clock_base[CLOCK_MONOTONIC].resolution = KTIME_HIGH_RES;
575 tick_setup_sched_timer();
577 /* "Retrigger" the interrupt to get things going */
578 retrigger_next_event(NULL);
579 local_irq_restore(flags);
580 printk(KERN_INFO "Switched to high resolution mode on CPU %d\n",
587 static inline int hrtimer_hres_active(void) { return 0; }
588 static inline int hrtimer_is_hres_enabled(void) { return 0; }
589 static inline int hrtimer_switch_to_hres(void) { return 0; }
590 static inline void hrtimer_force_reprogram(struct hrtimer_cpu_base *base) { }
591 static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
592 struct hrtimer_clock_base *base)
596 static inline int hrtimer_cb_pending(struct hrtimer *timer) { return 0; }
597 static inline void hrtimer_remove_cb_pending(struct hrtimer *timer) { }
598 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { }
599 static inline void hrtimer_init_timer_hres(struct hrtimer *timer) { }
601 #endif /* CONFIG_HIGH_RES_TIMERS */
603 #ifdef CONFIG_TIMER_STATS
604 void __timer_stats_hrtimer_set_start_info(struct hrtimer *timer, void *addr)
606 if (timer->start_site)
609 timer->start_site = addr;
610 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
611 timer->start_pid = current->pid;
616 * Counterpart to lock_timer_base above:
619 void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
621 spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
625 * hrtimer_forward - forward the timer expiry
626 * @timer: hrtimer to forward
627 * @now: forward past this time
628 * @interval: the interval to forward
630 * Forward the timer expiry so it will expire in the future.
631 * Returns the number of overruns.
634 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
636 unsigned long orun = 1;
639 delta = ktime_sub(now, timer->expires);
644 if (interval.tv64 < timer->base->resolution.tv64)
645 interval.tv64 = timer->base->resolution.tv64;
647 if (unlikely(delta.tv64 >= interval.tv64)) {
648 s64 incr = ktime_to_ns(interval);
650 orun = ktime_divns(delta, incr);
651 timer->expires = ktime_add_ns(timer->expires, incr * orun);
652 if (timer->expires.tv64 > now.tv64)
655 * This (and the ktime_add() below) is the
656 * correction for exact:
660 timer->expires = ktime_add(timer->expires, interval);
662 * Make sure, that the result did not wrap with a very large
665 if (timer->expires.tv64 < 0)
666 timer->expires = ktime_set(KTIME_SEC_MAX, 0);
672 * enqueue_hrtimer - internal function to (re)start a timer
674 * The timer is inserted in expiry order. Insertion into the
675 * red black tree is O(log(n)). Must hold the base lock.
677 static void enqueue_hrtimer(struct hrtimer *timer,
678 struct hrtimer_clock_base *base, int reprogram)
680 struct rb_node **link = &base->active.rb_node;
681 struct rb_node *parent = NULL;
682 struct hrtimer *entry;
685 * Find the right place in the rbtree:
689 entry = rb_entry(parent, struct hrtimer, node);
691 * We dont care about collisions. Nodes with
692 * the same expiry time stay together.
694 if (timer->expires.tv64 < entry->expires.tv64)
695 link = &(*link)->rb_left;
697 link = &(*link)->rb_right;
701 * Insert the timer to the rbtree and check whether it
702 * replaces the first pending timer
704 if (!base->first || timer->expires.tv64 <
705 rb_entry(base->first, struct hrtimer, node)->expires.tv64) {
707 * Reprogram the clock event device. When the timer is already
708 * expired hrtimer_enqueue_reprogram has either called the
709 * callback or added it to the pending list and raised the
712 * This is a NOP for !HIGHRES
714 if (reprogram && hrtimer_enqueue_reprogram(timer, base))
717 base->first = &timer->node;
720 rb_link_node(&timer->node, parent, link);
721 rb_insert_color(&timer->node, &base->active);
723 * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
724 * state of a possibly running callback.
726 timer->state |= HRTIMER_STATE_ENQUEUED;
730 * __remove_hrtimer - internal function to remove a timer
732 * Caller must hold the base lock.
734 * High resolution timer mode reprograms the clock event device when the
735 * timer is the one which expires next. The caller can disable this by setting
736 * reprogram to zero. This is useful, when the context does a reprogramming
737 * anyway (e.g. timer interrupt)
739 static void __remove_hrtimer(struct hrtimer *timer,
740 struct hrtimer_clock_base *base,
741 unsigned long newstate, int reprogram)
743 /* High res. callback list. NOP for !HIGHRES */
744 if (hrtimer_cb_pending(timer))
745 hrtimer_remove_cb_pending(timer);
748 * Remove the timer from the rbtree and replace the
749 * first entry pointer if necessary.
751 if (base->first == &timer->node) {
752 base->first = rb_next(&timer->node);
753 /* Reprogram the clock event device. if enabled */
754 if (reprogram && hrtimer_hres_active())
755 hrtimer_force_reprogram(base->cpu_base);
757 rb_erase(&timer->node, &base->active);
759 timer->state = newstate;
763 * remove hrtimer, called with base lock held
766 remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base)
768 if (hrtimer_is_queued(timer)) {
772 * Remove the timer and force reprogramming when high
773 * resolution mode is active and the timer is on the current
774 * CPU. If we remove a timer on another CPU, reprogramming is
775 * skipped. The interrupt event on this CPU is fired and
776 * reprogramming happens in the interrupt handler. This is a
777 * rare case and less expensive than a smp call.
779 timer_stats_hrtimer_clear_start_info(timer);
780 reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases);
781 __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE,
789 * hrtimer_start - (re)start an relative timer on the current CPU
790 * @timer: the timer to be added
792 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
796 * 1 when the timer was active
799 hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode)
801 struct hrtimer_clock_base *base, *new_base;
805 base = lock_hrtimer_base(timer, &flags);
807 /* Remove an active timer from the queue: */
808 ret = remove_hrtimer(timer, base);
810 /* Switch the timer base, if necessary: */
811 new_base = switch_hrtimer_base(timer, base);
813 if (mode == HRTIMER_MODE_REL) {
814 tim = ktime_add(tim, new_base->get_time());
816 * CONFIG_TIME_LOW_RES is a temporary way for architectures
817 * to signal that they simply return xtime in
818 * do_gettimeoffset(). In this case we want to round up by
819 * resolution when starting a relative timer, to avoid short
820 * timeouts. This will go away with the GTOD framework.
822 #ifdef CONFIG_TIME_LOW_RES
823 tim = ktime_add(tim, base->resolution);
826 timer->expires = tim;
828 timer_stats_hrtimer_set_start_info(timer);
831 * Only allow reprogramming if the new base is on this CPU.
832 * (it might still be on another CPU if the timer was pending)
834 enqueue_hrtimer(timer, new_base,
835 new_base->cpu_base == &__get_cpu_var(hrtimer_bases));
837 unlock_hrtimer_base(timer, &flags);
841 EXPORT_SYMBOL_GPL(hrtimer_start);
844 * hrtimer_try_to_cancel - try to deactivate a timer
845 * @timer: hrtimer to stop
848 * 0 when the timer was not active
849 * 1 when the timer was active
850 * -1 when the timer is currently excuting the callback function and
853 int hrtimer_try_to_cancel(struct hrtimer *timer)
855 struct hrtimer_clock_base *base;
859 base = lock_hrtimer_base(timer, &flags);
861 if (!hrtimer_callback_running(timer))
862 ret = remove_hrtimer(timer, base);
864 unlock_hrtimer_base(timer, &flags);
869 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
872 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
873 * @timer: the timer to be cancelled
876 * 0 when the timer was not active
877 * 1 when the timer was active
879 int hrtimer_cancel(struct hrtimer *timer)
882 int ret = hrtimer_try_to_cancel(timer);
889 EXPORT_SYMBOL_GPL(hrtimer_cancel);
892 * hrtimer_get_remaining - get remaining time for the timer
893 * @timer: the timer to read
895 ktime_t hrtimer_get_remaining(const struct hrtimer *timer)
897 struct hrtimer_clock_base *base;
901 base = lock_hrtimer_base(timer, &flags);
902 rem = ktime_sub(timer->expires, base->get_time());
903 unlock_hrtimer_base(timer, &flags);
907 EXPORT_SYMBOL_GPL(hrtimer_get_remaining);
909 #if defined(CONFIG_NO_IDLE_HZ) || defined(CONFIG_NO_HZ)
911 * hrtimer_get_next_event - get the time until next expiry event
913 * Returns the delta to the next expiry event or KTIME_MAX if no timer
916 ktime_t hrtimer_get_next_event(void)
918 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
919 struct hrtimer_clock_base *base = cpu_base->clock_base;
920 ktime_t delta, mindelta = { .tv64 = KTIME_MAX };
924 spin_lock_irqsave(&cpu_base->lock, flags);
926 if (!hrtimer_hres_active()) {
927 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
928 struct hrtimer *timer;
933 timer = rb_entry(base->first, struct hrtimer, node);
934 delta.tv64 = timer->expires.tv64;
935 delta = ktime_sub(delta, base->get_time());
936 if (delta.tv64 < mindelta.tv64)
937 mindelta.tv64 = delta.tv64;
941 spin_unlock_irqrestore(&cpu_base->lock, flags);
943 if (mindelta.tv64 < 0)
950 * hrtimer_init - initialize a timer to the given clock
951 * @timer: the timer to be initialized
952 * @clock_id: the clock to be used
953 * @mode: timer mode abs/rel
955 void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
956 enum hrtimer_mode mode)
958 struct hrtimer_cpu_base *cpu_base;
960 memset(timer, 0, sizeof(struct hrtimer));
962 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
964 if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
965 clock_id = CLOCK_MONOTONIC;
967 timer->base = &cpu_base->clock_base[clock_id];
968 hrtimer_init_timer_hres(timer);
970 #ifdef CONFIG_TIMER_STATS
971 timer->start_site = NULL;
972 timer->start_pid = -1;
973 memset(timer->start_comm, 0, TASK_COMM_LEN);
976 EXPORT_SYMBOL_GPL(hrtimer_init);
979 * hrtimer_get_res - get the timer resolution for a clock
980 * @which_clock: which clock to query
981 * @tp: pointer to timespec variable to store the resolution
983 * Store the resolution of the clock selected by @which_clock in the
984 * variable pointed to by @tp.
986 int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp)
988 struct hrtimer_cpu_base *cpu_base;
990 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
991 *tp = ktime_to_timespec(cpu_base->clock_base[which_clock].resolution);
995 EXPORT_SYMBOL_GPL(hrtimer_get_res);
997 #ifdef CONFIG_HIGH_RES_TIMERS
1000 * High resolution timer interrupt
1001 * Called with interrupts disabled
1003 void hrtimer_interrupt(struct clock_event_device *dev)
1005 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1006 struct hrtimer_clock_base *base;
1007 ktime_t expires_next, now;
1010 BUG_ON(!cpu_base->hres_active);
1011 cpu_base->nr_events++;
1012 dev->next_event.tv64 = KTIME_MAX;
1017 expires_next.tv64 = KTIME_MAX;
1019 base = cpu_base->clock_base;
1021 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1023 struct rb_node *node;
1025 spin_lock(&cpu_base->lock);
1027 basenow = ktime_add(now, base->offset);
1029 while ((node = base->first)) {
1030 struct hrtimer *timer;
1032 timer = rb_entry(node, struct hrtimer, node);
1034 if (basenow.tv64 < timer->expires.tv64) {
1037 expires = ktime_sub(timer->expires,
1039 if (expires.tv64 < expires_next.tv64)
1040 expires_next = expires;
1044 /* Move softirq callbacks to the pending list */
1045 if (timer->cb_mode == HRTIMER_CB_SOFTIRQ) {
1046 __remove_hrtimer(timer, base,
1047 HRTIMER_STATE_PENDING, 0);
1048 list_add_tail(&timer->cb_entry,
1049 &base->cpu_base->cb_pending);
1054 __remove_hrtimer(timer, base,
1055 HRTIMER_STATE_CALLBACK, 0);
1056 timer_stats_account_hrtimer(timer);
1059 * Note: We clear the CALLBACK bit after
1060 * enqueue_hrtimer to avoid reprogramming of
1061 * the event hardware. This happens at the end
1062 * of this function anyway.
1064 if (timer->function(timer) != HRTIMER_NORESTART) {
1065 BUG_ON(timer->state != HRTIMER_STATE_CALLBACK);
1066 enqueue_hrtimer(timer, base, 0);
1068 timer->state &= ~HRTIMER_STATE_CALLBACK;
1070 spin_unlock(&cpu_base->lock);
1074 cpu_base->expires_next = expires_next;
1076 /* Reprogramming necessary ? */
1077 if (expires_next.tv64 != KTIME_MAX) {
1078 if (tick_program_event(expires_next, 0))
1082 /* Raise softirq ? */
1084 raise_softirq(HRTIMER_SOFTIRQ);
1087 static void run_hrtimer_softirq(struct softirq_action *h)
1089 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1091 spin_lock_irq(&cpu_base->lock);
1093 while (!list_empty(&cpu_base->cb_pending)) {
1094 enum hrtimer_restart (*fn)(struct hrtimer *);
1095 struct hrtimer *timer;
1098 timer = list_entry(cpu_base->cb_pending.next,
1099 struct hrtimer, cb_entry);
1101 timer_stats_account_hrtimer(timer);
1103 fn = timer->function;
1104 __remove_hrtimer(timer, timer->base, HRTIMER_STATE_CALLBACK, 0);
1105 spin_unlock_irq(&cpu_base->lock);
1107 restart = fn(timer);
1109 spin_lock_irq(&cpu_base->lock);
1111 timer->state &= ~HRTIMER_STATE_CALLBACK;
1112 if (restart == HRTIMER_RESTART) {
1113 BUG_ON(hrtimer_active(timer));
1115 * Enqueue the timer, allow reprogramming of the event
1118 enqueue_hrtimer(timer, timer->base, 1);
1119 } else if (hrtimer_active(timer)) {
1121 * If the timer was rearmed on another CPU, reprogram
1124 if (timer->base->first == &timer->node)
1125 hrtimer_reprogram(timer, timer->base);
1128 spin_unlock_irq(&cpu_base->lock);
1131 #endif /* CONFIG_HIGH_RES_TIMERS */
1134 * Expire the per base hrtimer-queue:
1136 static inline void run_hrtimer_queue(struct hrtimer_cpu_base *cpu_base,
1139 struct rb_node *node;
1140 struct hrtimer_clock_base *base = &cpu_base->clock_base[index];
1145 if (base->get_softirq_time)
1146 base->softirq_time = base->get_softirq_time();
1148 spin_lock_irq(&cpu_base->lock);
1150 while ((node = base->first)) {
1151 struct hrtimer *timer;
1152 enum hrtimer_restart (*fn)(struct hrtimer *);
1155 timer = rb_entry(node, struct hrtimer, node);
1156 if (base->softirq_time.tv64 <= timer->expires.tv64)
1159 #ifdef CONFIG_HIGH_RES_TIMERS
1160 WARN_ON_ONCE(timer->cb_mode == HRTIMER_CB_IRQSAFE_NO_SOFTIRQ);
1162 timer_stats_account_hrtimer(timer);
1164 fn = timer->function;
1165 __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0);
1166 spin_unlock_irq(&cpu_base->lock);
1168 restart = fn(timer);
1170 spin_lock_irq(&cpu_base->lock);
1172 timer->state &= ~HRTIMER_STATE_CALLBACK;
1173 if (restart != HRTIMER_NORESTART) {
1174 BUG_ON(hrtimer_active(timer));
1175 enqueue_hrtimer(timer, base, 0);
1178 spin_unlock_irq(&cpu_base->lock);
1182 * Called from timer softirq every jiffy, expire hrtimers:
1184 * For HRT its the fall back code to run the softirq in the timer
1185 * softirq context in case the hrtimer initialization failed or has
1186 * not been done yet.
1188 void hrtimer_run_queues(void)
1190 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1193 if (hrtimer_hres_active())
1197 * This _is_ ugly: We have to check in the softirq context,
1198 * whether we can switch to highres and / or nohz mode. The
1199 * clocksource switch happens in the timer interrupt with
1200 * xtime_lock held. Notification from there only sets the
1201 * check bit in the tick_oneshot code, otherwise we might
1202 * deadlock vs. xtime_lock.
1204 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1205 if (hrtimer_switch_to_hres())
1208 hrtimer_get_softirq_time(cpu_base);
1210 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
1211 run_hrtimer_queue(cpu_base, i);
1215 * Sleep related functions:
1217 static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1219 struct hrtimer_sleeper *t =
1220 container_of(timer, struct hrtimer_sleeper, timer);
1221 struct task_struct *task = t->task;
1225 wake_up_process(task);
1227 return HRTIMER_NORESTART;
1230 void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1232 sl->timer.function = hrtimer_wakeup;
1234 #ifdef CONFIG_HIGH_RES_TIMERS
1235 sl->timer.cb_mode = HRTIMER_CB_IRQSAFE_NO_RESTART;
1239 static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1241 hrtimer_init_sleeper(t, current);
1244 set_current_state(TASK_INTERRUPTIBLE);
1245 hrtimer_start(&t->timer, t->timer.expires, mode);
1247 if (likely(t->task))
1250 hrtimer_cancel(&t->timer);
1251 mode = HRTIMER_MODE_ABS;
1253 } while (t->task && !signal_pending(current));
1255 return t->task == NULL;
1258 long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1260 struct hrtimer_sleeper t;
1261 struct timespec __user *rmtp;
1265 restart->fn = do_no_restart_syscall;
1267 hrtimer_init(&t.timer, restart->arg0, HRTIMER_MODE_ABS);
1268 t.timer.expires.tv64 = ((u64)restart->arg3 << 32) | (u64) restart->arg2;
1270 if (do_nanosleep(&t, HRTIMER_MODE_ABS))
1273 rmtp = (struct timespec __user *) restart->arg1;
1275 time = ktime_sub(t.timer.expires, t.timer.base->get_time());
1278 tu = ktime_to_timespec(time);
1279 if (copy_to_user(rmtp, &tu, sizeof(tu)))
1283 restart->fn = hrtimer_nanosleep_restart;
1285 /* The other values in restart are already filled in */
1286 return -ERESTART_RESTARTBLOCK;
1289 long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
1290 const enum hrtimer_mode mode, const clockid_t clockid)
1292 struct restart_block *restart;
1293 struct hrtimer_sleeper t;
1297 hrtimer_init(&t.timer, clockid, mode);
1298 t.timer.expires = timespec_to_ktime(*rqtp);
1299 if (do_nanosleep(&t, mode))
1302 /* Absolute timers do not update the rmtp value and restart: */
1303 if (mode == HRTIMER_MODE_ABS)
1304 return -ERESTARTNOHAND;
1307 rem = ktime_sub(t.timer.expires, t.timer.base->get_time());
1310 tu = ktime_to_timespec(rem);
1311 if (copy_to_user(rmtp, &tu, sizeof(tu)))
1315 restart = ¤t_thread_info()->restart_block;
1316 restart->fn = hrtimer_nanosleep_restart;
1317 restart->arg0 = (unsigned long) t.timer.base->index;
1318 restart->arg1 = (unsigned long) rmtp;
1319 restart->arg2 = t.timer.expires.tv64 & 0xFFFFFFFF;
1320 restart->arg3 = t.timer.expires.tv64 >> 32;
1322 return -ERESTART_RESTARTBLOCK;
1326 sys_nanosleep(struct timespec __user *rqtp, struct timespec __user *rmtp)
1330 if (copy_from_user(&tu, rqtp, sizeof(tu)))
1333 if (!timespec_valid(&tu))
1336 return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1340 * Functions related to boot-time initialization:
1342 static void __devinit init_hrtimers_cpu(int cpu)
1344 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1347 spin_lock_init(&cpu_base->lock);
1348 lockdep_set_class(&cpu_base->lock, &cpu_base->lock_key);
1350 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
1351 cpu_base->clock_base[i].cpu_base = cpu_base;
1353 hrtimer_init_hres(cpu_base);
1356 #ifdef CONFIG_HOTPLUG_CPU
1358 static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1359 struct hrtimer_clock_base *new_base)
1361 struct hrtimer *timer;
1362 struct rb_node *node;
1364 while ((node = rb_first(&old_base->active))) {
1365 timer = rb_entry(node, struct hrtimer, node);
1366 BUG_ON(hrtimer_callback_running(timer));
1367 __remove_hrtimer(timer, old_base, HRTIMER_STATE_INACTIVE, 0);
1368 timer->base = new_base;
1370 * Enqueue the timer. Allow reprogramming of the event device
1372 enqueue_hrtimer(timer, new_base, 1);
1376 static void migrate_hrtimers(int cpu)
1378 struct hrtimer_cpu_base *old_base, *new_base;
1381 BUG_ON(cpu_online(cpu));
1382 old_base = &per_cpu(hrtimer_bases, cpu);
1383 new_base = &get_cpu_var(hrtimer_bases);
1385 tick_cancel_sched_timer(cpu);
1387 local_irq_disable();
1388 double_spin_lock(&new_base->lock, &old_base->lock,
1389 smp_processor_id() < cpu);
1391 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1392 migrate_hrtimer_list(&old_base->clock_base[i],
1393 &new_base->clock_base[i]);
1396 double_spin_unlock(&new_base->lock, &old_base->lock,
1397 smp_processor_id() < cpu);
1399 put_cpu_var(hrtimer_bases);
1401 #endif /* CONFIG_HOTPLUG_CPU */
1403 static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self,
1404 unsigned long action, void *hcpu)
1406 long cpu = (long)hcpu;
1410 case CPU_UP_PREPARE:
1411 init_hrtimers_cpu(cpu);
1414 #ifdef CONFIG_HOTPLUG_CPU
1416 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &cpu);
1417 migrate_hrtimers(cpu);
1428 static struct notifier_block __cpuinitdata hrtimers_nb = {
1429 .notifier_call = hrtimer_cpu_notify,
1432 void __init hrtimers_init(void)
1434 hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
1435 (void *)(long)smp_processor_id());
1436 register_cpu_notifier(&hrtimers_nb);
1437 #ifdef CONFIG_HIGH_RES_TIMERS
1438 open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq, NULL);