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;
138 struct timespec xts, tom;
142 seq = read_seqbegin(&xtime_lock);
144 getnstimeofday(&xts);
148 tom = wall_to_monotonic;
149 } while (read_seqretry(&xtime_lock, seq));
151 xtim = timespec_to_ktime(xts);
152 tomono = timespec_to_ktime(tom);
153 base->clock_base[CLOCK_REALTIME].softirq_time = xtim;
154 base->clock_base[CLOCK_MONOTONIC].softirq_time =
155 ktime_add(xtim, tomono);
159 * Helper function to check, whether the timer is running the callback
162 static inline int hrtimer_callback_running(struct hrtimer *timer)
164 return timer->state & HRTIMER_STATE_CALLBACK;
168 * Functions and macros which are different for UP/SMP systems are kept in a
174 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
175 * means that all timers which are tied to this base via timer->base are
176 * locked, and the base itself is locked too.
178 * So __run_timers/migrate_timers can safely modify all timers which could
179 * be found on the lists/queues.
181 * When the timer's base is locked, and the timer removed from list, it is
182 * possible to set timer->base = NULL and drop the lock: the timer remains
186 struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
187 unsigned long *flags)
189 struct hrtimer_clock_base *base;
193 if (likely(base != NULL)) {
194 spin_lock_irqsave(&base->cpu_base->lock, *flags);
195 if (likely(base == timer->base))
197 /* The timer has migrated to another CPU: */
198 spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
205 * Switch the timer base to the current CPU when possible.
207 static inline struct hrtimer_clock_base *
208 switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base)
210 struct hrtimer_clock_base *new_base;
211 struct hrtimer_cpu_base *new_cpu_base;
213 new_cpu_base = &__get_cpu_var(hrtimer_bases);
214 new_base = &new_cpu_base->clock_base[base->index];
216 if (base != new_base) {
218 * We are trying to schedule the timer on the local CPU.
219 * However we can't change timer's base while it is running,
220 * so we keep it on the same CPU. No hassle vs. reprogramming
221 * the event source in the high resolution case. The softirq
222 * code will take care of this when the timer function has
223 * completed. There is no conflict as we hold the lock until
224 * the timer is enqueued.
226 if (unlikely(hrtimer_callback_running(timer)))
229 /* See the comment in lock_timer_base() */
231 spin_unlock(&base->cpu_base->lock);
232 spin_lock(&new_base->cpu_base->lock);
233 timer->base = new_base;
238 #else /* CONFIG_SMP */
240 static inline struct hrtimer_clock_base *
241 lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
243 struct hrtimer_clock_base *base = timer->base;
245 spin_lock_irqsave(&base->cpu_base->lock, *flags);
250 # define switch_hrtimer_base(t, b) (b)
252 #endif /* !CONFIG_SMP */
255 * Functions for the union type storage format of ktime_t which are
256 * too large for inlining:
258 #if BITS_PER_LONG < 64
259 # ifndef CONFIG_KTIME_SCALAR
261 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
263 * @nsec: the scalar nsec value to add
265 * Returns the sum of kt and nsec in ktime_t format
267 ktime_t ktime_add_ns(const ktime_t kt, u64 nsec)
271 if (likely(nsec < NSEC_PER_SEC)) {
274 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
276 tmp = ktime_set((long)nsec, rem);
279 return ktime_add(kt, tmp);
281 # endif /* !CONFIG_KTIME_SCALAR */
284 * Divide a ktime value by a nanosecond value
286 unsigned long ktime_divns(const ktime_t kt, s64 div)
291 dclc = dns = ktime_to_ns(kt);
293 /* Make sure the divisor is less than 2^32: */
299 do_div(dclc, (unsigned long) div);
301 return (unsigned long) dclc;
303 #endif /* BITS_PER_LONG >= 64 */
305 /* High resolution timer related functions */
306 #ifdef CONFIG_HIGH_RES_TIMERS
309 * High resolution timer enabled ?
311 static int hrtimer_hres_enabled __read_mostly = 1;
314 * Enable / Disable high resolution mode
316 static int __init setup_hrtimer_hres(char *str)
318 if (!strcmp(str, "off"))
319 hrtimer_hres_enabled = 0;
320 else if (!strcmp(str, "on"))
321 hrtimer_hres_enabled = 1;
327 __setup("highres=", setup_hrtimer_hres);
330 * hrtimer_high_res_enabled - query, if the highres mode is enabled
332 static inline int hrtimer_is_hres_enabled(void)
334 return hrtimer_hres_enabled;
338 * Is the high resolution mode active ?
340 static inline int hrtimer_hres_active(void)
342 return __get_cpu_var(hrtimer_bases).hres_active;
346 * Reprogram the event source with checking both queues for the
348 * Called with interrupts disabled and base->lock held
350 static void hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base)
353 struct hrtimer_clock_base *base = cpu_base->clock_base;
356 cpu_base->expires_next.tv64 = KTIME_MAX;
358 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
359 struct hrtimer *timer;
363 timer = rb_entry(base->first, struct hrtimer, node);
364 expires = ktime_sub(timer->expires, base->offset);
365 if (expires.tv64 < cpu_base->expires_next.tv64)
366 cpu_base->expires_next = expires;
369 if (cpu_base->expires_next.tv64 != KTIME_MAX)
370 tick_program_event(cpu_base->expires_next, 1);
374 * Shared reprogramming for clock_realtime and clock_monotonic
376 * When a timer is enqueued and expires earlier than the already enqueued
377 * timers, we have to check, whether it expires earlier than the timer for
378 * which the clock event device was armed.
380 * Called with interrupts disabled and base->cpu_base.lock held
382 static int hrtimer_reprogram(struct hrtimer *timer,
383 struct hrtimer_clock_base *base)
385 ktime_t *expires_next = &__get_cpu_var(hrtimer_bases).expires_next;
386 ktime_t expires = ktime_sub(timer->expires, base->offset);
390 * When the callback is running, we do not reprogram the clock event
391 * device. The timer callback is either running on a different CPU or
392 * the callback is executed in the hrtimer_interupt context. The
393 * reprogramming is handled either by the softirq, which called the
394 * callback or at the end of the hrtimer_interrupt.
396 if (hrtimer_callback_running(timer))
399 if (expires.tv64 >= expires_next->tv64)
403 * Clockevents returns -ETIME, when the event was in the past.
405 res = tick_program_event(expires, 0);
406 if (!IS_ERR_VALUE(res))
407 *expires_next = expires;
413 * Retrigger next event is called after clock was set
415 * Called with interrupts disabled via on_each_cpu()
417 static void retrigger_next_event(void *arg)
419 struct hrtimer_cpu_base *base;
420 struct timespec realtime_offset;
423 if (!hrtimer_hres_active())
427 seq = read_seqbegin(&xtime_lock);
428 set_normalized_timespec(&realtime_offset,
429 -wall_to_monotonic.tv_sec,
430 -wall_to_monotonic.tv_nsec);
431 } while (read_seqretry(&xtime_lock, seq));
433 base = &__get_cpu_var(hrtimer_bases);
435 /* Adjust CLOCK_REALTIME offset */
436 spin_lock(&base->lock);
437 base->clock_base[CLOCK_REALTIME].offset =
438 timespec_to_ktime(realtime_offset);
440 hrtimer_force_reprogram(base);
441 spin_unlock(&base->lock);
445 * Clock realtime was set
447 * Change the offset of the realtime clock vs. the monotonic
450 * We might have to reprogram the high resolution timer interrupt. On
451 * SMP we call the architecture specific code to retrigger _all_ high
452 * resolution timer interrupts. On UP we just disable interrupts and
453 * call the high resolution interrupt code.
455 void clock_was_set(void)
457 /* Retrigger the CPU local events everywhere */
458 on_each_cpu(retrigger_next_event, NULL, 0, 1);
462 * Check, whether the timer is on the callback pending list
464 static inline int hrtimer_cb_pending(const struct hrtimer *timer)
466 return timer->state & HRTIMER_STATE_PENDING;
470 * Remove a timer from the callback pending list
472 static inline void hrtimer_remove_cb_pending(struct hrtimer *timer)
474 list_del_init(&timer->cb_entry);
478 * Initialize the high resolution related parts of cpu_base
480 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
482 base->expires_next.tv64 = KTIME_MAX;
483 base->hres_active = 0;
484 INIT_LIST_HEAD(&base->cb_pending);
488 * Initialize the high resolution related parts of a hrtimer
490 static inline void hrtimer_init_timer_hres(struct hrtimer *timer)
492 INIT_LIST_HEAD(&timer->cb_entry);
496 * When High resolution timers are active, try to reprogram. Note, that in case
497 * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry
498 * check happens. The timer gets enqueued into the rbtree. The reprogramming
499 * and expiry check is done in the hrtimer_interrupt or in the softirq.
501 static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
502 struct hrtimer_clock_base *base)
504 if (base->cpu_base->hres_active && hrtimer_reprogram(timer, base)) {
506 /* Timer is expired, act upon the callback mode */
507 switch(timer->cb_mode) {
508 case HRTIMER_CB_IRQSAFE_NO_RESTART:
510 * We can call the callback from here. No restart
511 * happens, so no danger of recursion
513 BUG_ON(timer->function(timer) != HRTIMER_NORESTART);
515 case HRTIMER_CB_IRQSAFE_NO_SOFTIRQ:
517 * This is solely for the sched tick emulation with
518 * dynamic tick support to ensure that we do not
519 * restart the tick right on the edge and end up with
520 * the tick timer in the softirq ! The calling site
521 * takes care of this.
524 case HRTIMER_CB_IRQSAFE:
525 case HRTIMER_CB_SOFTIRQ:
527 * Move everything else into the softirq pending list !
529 list_add_tail(&timer->cb_entry,
530 &base->cpu_base->cb_pending);
531 timer->state = HRTIMER_STATE_PENDING;
532 raise_softirq(HRTIMER_SOFTIRQ);
542 * Switch to high resolution mode
544 static int hrtimer_switch_to_hres(void)
546 struct hrtimer_cpu_base *base = &__get_cpu_var(hrtimer_bases);
549 if (base->hres_active)
552 local_irq_save(flags);
554 if (tick_init_highres()) {
555 local_irq_restore(flags);
558 base->hres_active = 1;
559 base->clock_base[CLOCK_REALTIME].resolution = KTIME_HIGH_RES;
560 base->clock_base[CLOCK_MONOTONIC].resolution = KTIME_HIGH_RES;
562 tick_setup_sched_timer();
564 /* "Retrigger" the interrupt to get things going */
565 retrigger_next_event(NULL);
566 local_irq_restore(flags);
567 printk(KERN_INFO "Switched to high resolution mode on CPU %d\n",
574 static inline int hrtimer_hres_active(void) { return 0; }
575 static inline int hrtimer_is_hres_enabled(void) { return 0; }
576 static inline int hrtimer_switch_to_hres(void) { return 0; }
577 static inline void hrtimer_force_reprogram(struct hrtimer_cpu_base *base) { }
578 static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
579 struct hrtimer_clock_base *base)
583 static inline int hrtimer_cb_pending(struct hrtimer *timer) { return 0; }
584 static inline void hrtimer_remove_cb_pending(struct hrtimer *timer) { }
585 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { }
586 static inline void hrtimer_init_timer_hres(struct hrtimer *timer) { }
588 #endif /* CONFIG_HIGH_RES_TIMERS */
590 #ifdef CONFIG_TIMER_STATS
591 void __timer_stats_hrtimer_set_start_info(struct hrtimer *timer, void *addr)
593 if (timer->start_site)
596 timer->start_site = addr;
597 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
598 timer->start_pid = current->pid;
603 * Counterpart to lock_timer_base above:
606 void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
608 spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
612 * hrtimer_forward - forward the timer expiry
613 * @timer: hrtimer to forward
614 * @now: forward past this time
615 * @interval: the interval to forward
617 * Forward the timer expiry so it will expire in the future.
618 * Returns the number of overruns.
621 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
623 unsigned long orun = 1;
626 delta = ktime_sub(now, timer->expires);
631 if (interval.tv64 < timer->base->resolution.tv64)
632 interval.tv64 = timer->base->resolution.tv64;
634 if (unlikely(delta.tv64 >= interval.tv64)) {
635 s64 incr = ktime_to_ns(interval);
637 orun = ktime_divns(delta, incr);
638 timer->expires = ktime_add_ns(timer->expires, incr * orun);
639 if (timer->expires.tv64 > now.tv64)
642 * This (and the ktime_add() below) is the
643 * correction for exact:
647 timer->expires = ktime_add(timer->expires, interval);
649 * Make sure, that the result did not wrap with a very large
652 if (timer->expires.tv64 < 0)
653 timer->expires = ktime_set(KTIME_SEC_MAX, 0);
659 * enqueue_hrtimer - internal function to (re)start a timer
661 * The timer is inserted in expiry order. Insertion into the
662 * red black tree is O(log(n)). Must hold the base lock.
664 static void enqueue_hrtimer(struct hrtimer *timer,
665 struct hrtimer_clock_base *base, int reprogram)
667 struct rb_node **link = &base->active.rb_node;
668 struct rb_node *parent = NULL;
669 struct hrtimer *entry;
672 * Find the right place in the rbtree:
676 entry = rb_entry(parent, struct hrtimer, node);
678 * We dont care about collisions. Nodes with
679 * the same expiry time stay together.
681 if (timer->expires.tv64 < entry->expires.tv64)
682 link = &(*link)->rb_left;
684 link = &(*link)->rb_right;
688 * Insert the timer to the rbtree and check whether it
689 * replaces the first pending timer
691 if (!base->first || timer->expires.tv64 <
692 rb_entry(base->first, struct hrtimer, node)->expires.tv64) {
694 * Reprogram the clock event device. When the timer is already
695 * expired hrtimer_enqueue_reprogram has either called the
696 * callback or added it to the pending list and raised the
699 * This is a NOP for !HIGHRES
701 if (reprogram && hrtimer_enqueue_reprogram(timer, base))
704 base->first = &timer->node;
707 rb_link_node(&timer->node, parent, link);
708 rb_insert_color(&timer->node, &base->active);
710 * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
711 * state of a possibly running callback.
713 timer->state |= HRTIMER_STATE_ENQUEUED;
717 * __remove_hrtimer - internal function to remove a timer
719 * Caller must hold the base lock.
721 * High resolution timer mode reprograms the clock event device when the
722 * timer is the one which expires next. The caller can disable this by setting
723 * reprogram to zero. This is useful, when the context does a reprogramming
724 * anyway (e.g. timer interrupt)
726 static void __remove_hrtimer(struct hrtimer *timer,
727 struct hrtimer_clock_base *base,
728 unsigned long newstate, int reprogram)
730 /* High res. callback list. NOP for !HIGHRES */
731 if (hrtimer_cb_pending(timer))
732 hrtimer_remove_cb_pending(timer);
735 * Remove the timer from the rbtree and replace the
736 * first entry pointer if necessary.
738 if (base->first == &timer->node) {
739 base->first = rb_next(&timer->node);
740 /* Reprogram the clock event device. if enabled */
741 if (reprogram && hrtimer_hres_active())
742 hrtimer_force_reprogram(base->cpu_base);
744 rb_erase(&timer->node, &base->active);
746 timer->state = newstate;
750 * remove hrtimer, called with base lock held
753 remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base)
755 if (hrtimer_is_queued(timer)) {
759 * Remove the timer and force reprogramming when high
760 * resolution mode is active and the timer is on the current
761 * CPU. If we remove a timer on another CPU, reprogramming is
762 * skipped. The interrupt event on this CPU is fired and
763 * reprogramming happens in the interrupt handler. This is a
764 * rare case and less expensive than a smp call.
766 timer_stats_hrtimer_clear_start_info(timer);
767 reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases);
768 __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE,
776 * hrtimer_start - (re)start an relative timer on the current CPU
777 * @timer: the timer to be added
779 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
783 * 1 when the timer was active
786 hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode)
788 struct hrtimer_clock_base *base, *new_base;
792 base = lock_hrtimer_base(timer, &flags);
794 /* Remove an active timer from the queue: */
795 ret = remove_hrtimer(timer, base);
797 /* Switch the timer base, if necessary: */
798 new_base = switch_hrtimer_base(timer, base);
800 if (mode == HRTIMER_MODE_REL) {
801 tim = ktime_add(tim, new_base->get_time());
803 * CONFIG_TIME_LOW_RES is a temporary way for architectures
804 * to signal that they simply return xtime in
805 * do_gettimeoffset(). In this case we want to round up by
806 * resolution when starting a relative timer, to avoid short
807 * timeouts. This will go away with the GTOD framework.
809 #ifdef CONFIG_TIME_LOW_RES
810 tim = ktime_add(tim, base->resolution);
813 timer->expires = tim;
815 timer_stats_hrtimer_set_start_info(timer);
817 enqueue_hrtimer(timer, new_base, base == new_base);
819 unlock_hrtimer_base(timer, &flags);
823 EXPORT_SYMBOL_GPL(hrtimer_start);
826 * hrtimer_try_to_cancel - try to deactivate a timer
827 * @timer: hrtimer to stop
830 * 0 when the timer was not active
831 * 1 when the timer was active
832 * -1 when the timer is currently excuting the callback function and
835 int hrtimer_try_to_cancel(struct hrtimer *timer)
837 struct hrtimer_clock_base *base;
841 base = lock_hrtimer_base(timer, &flags);
843 if (!hrtimer_callback_running(timer))
844 ret = remove_hrtimer(timer, base);
846 unlock_hrtimer_base(timer, &flags);
851 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
854 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
855 * @timer: the timer to be cancelled
858 * 0 when the timer was not active
859 * 1 when the timer was active
861 int hrtimer_cancel(struct hrtimer *timer)
864 int ret = hrtimer_try_to_cancel(timer);
871 EXPORT_SYMBOL_GPL(hrtimer_cancel);
874 * hrtimer_get_remaining - get remaining time for the timer
875 * @timer: the timer to read
877 ktime_t hrtimer_get_remaining(const struct hrtimer *timer)
879 struct hrtimer_clock_base *base;
883 base = lock_hrtimer_base(timer, &flags);
884 rem = ktime_sub(timer->expires, base->get_time());
885 unlock_hrtimer_base(timer, &flags);
889 EXPORT_SYMBOL_GPL(hrtimer_get_remaining);
891 #if defined(CONFIG_NO_IDLE_HZ) || defined(CONFIG_NO_HZ)
893 * hrtimer_get_next_event - get the time until next expiry event
895 * Returns the delta to the next expiry event or KTIME_MAX if no timer
898 ktime_t hrtimer_get_next_event(void)
900 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
901 struct hrtimer_clock_base *base = cpu_base->clock_base;
902 ktime_t delta, mindelta = { .tv64 = KTIME_MAX };
906 spin_lock_irqsave(&cpu_base->lock, flags);
908 if (!hrtimer_hres_active()) {
909 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
910 struct hrtimer *timer;
915 timer = rb_entry(base->first, struct hrtimer, node);
916 delta.tv64 = timer->expires.tv64;
917 delta = ktime_sub(delta, base->get_time());
918 if (delta.tv64 < mindelta.tv64)
919 mindelta.tv64 = delta.tv64;
923 spin_unlock_irqrestore(&cpu_base->lock, flags);
925 if (mindelta.tv64 < 0)
932 * hrtimer_init - initialize a timer to the given clock
933 * @timer: the timer to be initialized
934 * @clock_id: the clock to be used
935 * @mode: timer mode abs/rel
937 void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
938 enum hrtimer_mode mode)
940 struct hrtimer_cpu_base *cpu_base;
942 memset(timer, 0, sizeof(struct hrtimer));
944 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
946 if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
947 clock_id = CLOCK_MONOTONIC;
949 timer->base = &cpu_base->clock_base[clock_id];
950 hrtimer_init_timer_hres(timer);
952 #ifdef CONFIG_TIMER_STATS
953 timer->start_site = NULL;
954 timer->start_pid = -1;
955 memset(timer->start_comm, 0, TASK_COMM_LEN);
958 EXPORT_SYMBOL_GPL(hrtimer_init);
961 * hrtimer_get_res - get the timer resolution for a clock
962 * @which_clock: which clock to query
963 * @tp: pointer to timespec variable to store the resolution
965 * Store the resolution of the clock selected by @which_clock in the
966 * variable pointed to by @tp.
968 int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp)
970 struct hrtimer_cpu_base *cpu_base;
972 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
973 *tp = ktime_to_timespec(cpu_base->clock_base[which_clock].resolution);
977 EXPORT_SYMBOL_GPL(hrtimer_get_res);
979 #ifdef CONFIG_HIGH_RES_TIMERS
982 * High resolution timer interrupt
983 * Called with interrupts disabled
985 void hrtimer_interrupt(struct clock_event_device *dev)
987 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
988 struct hrtimer_clock_base *base;
989 ktime_t expires_next, now;
992 BUG_ON(!cpu_base->hres_active);
993 cpu_base->nr_events++;
994 dev->next_event.tv64 = KTIME_MAX;
999 expires_next.tv64 = KTIME_MAX;
1001 base = cpu_base->clock_base;
1003 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1005 struct rb_node *node;
1007 spin_lock(&cpu_base->lock);
1009 basenow = ktime_add(now, base->offset);
1011 while ((node = base->first)) {
1012 struct hrtimer *timer;
1014 timer = rb_entry(node, struct hrtimer, node);
1016 if (basenow.tv64 < timer->expires.tv64) {
1019 expires = ktime_sub(timer->expires,
1021 if (expires.tv64 < expires_next.tv64)
1022 expires_next = expires;
1026 /* Move softirq callbacks to the pending list */
1027 if (timer->cb_mode == HRTIMER_CB_SOFTIRQ) {
1028 __remove_hrtimer(timer, base,
1029 HRTIMER_STATE_PENDING, 0);
1030 list_add_tail(&timer->cb_entry,
1031 &base->cpu_base->cb_pending);
1036 __remove_hrtimer(timer, base,
1037 HRTIMER_STATE_CALLBACK, 0);
1038 timer_stats_account_hrtimer(timer);
1041 * Note: We clear the CALLBACK bit after
1042 * enqueue_hrtimer to avoid reprogramming of
1043 * the event hardware. This happens at the end
1044 * of this function anyway.
1046 if (timer->function(timer) != HRTIMER_NORESTART) {
1047 BUG_ON(timer->state != HRTIMER_STATE_CALLBACK);
1048 enqueue_hrtimer(timer, base, 0);
1050 timer->state &= ~HRTIMER_STATE_CALLBACK;
1052 spin_unlock(&cpu_base->lock);
1056 cpu_base->expires_next = expires_next;
1058 /* Reprogramming necessary ? */
1059 if (expires_next.tv64 != KTIME_MAX) {
1060 if (tick_program_event(expires_next, 0))
1064 /* Raise softirq ? */
1066 raise_softirq(HRTIMER_SOFTIRQ);
1069 static void run_hrtimer_softirq(struct softirq_action *h)
1071 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1073 spin_lock_irq(&cpu_base->lock);
1075 while (!list_empty(&cpu_base->cb_pending)) {
1076 enum hrtimer_restart (*fn)(struct hrtimer *);
1077 struct hrtimer *timer;
1080 timer = list_entry(cpu_base->cb_pending.next,
1081 struct hrtimer, cb_entry);
1083 timer_stats_account_hrtimer(timer);
1085 fn = timer->function;
1086 __remove_hrtimer(timer, timer->base, HRTIMER_STATE_CALLBACK, 0);
1087 spin_unlock_irq(&cpu_base->lock);
1089 restart = fn(timer);
1091 spin_lock_irq(&cpu_base->lock);
1093 timer->state &= ~HRTIMER_STATE_CALLBACK;
1094 if (restart == HRTIMER_RESTART) {
1095 BUG_ON(hrtimer_active(timer));
1097 * Enqueue the timer, allow reprogramming of the event
1100 enqueue_hrtimer(timer, timer->base, 1);
1101 } else if (hrtimer_active(timer)) {
1103 * If the timer was rearmed on another CPU, reprogram
1106 if (timer->base->first == &timer->node)
1107 hrtimer_reprogram(timer, timer->base);
1110 spin_unlock_irq(&cpu_base->lock);
1113 #endif /* CONFIG_HIGH_RES_TIMERS */
1116 * Expire the per base hrtimer-queue:
1118 static inline void run_hrtimer_queue(struct hrtimer_cpu_base *cpu_base,
1121 struct rb_node *node;
1122 struct hrtimer_clock_base *base = &cpu_base->clock_base[index];
1127 if (base->get_softirq_time)
1128 base->softirq_time = base->get_softirq_time();
1130 spin_lock_irq(&cpu_base->lock);
1132 while ((node = base->first)) {
1133 struct hrtimer *timer;
1134 enum hrtimer_restart (*fn)(struct hrtimer *);
1137 timer = rb_entry(node, struct hrtimer, node);
1138 if (base->softirq_time.tv64 <= timer->expires.tv64)
1141 #ifdef CONFIG_HIGH_RES_TIMERS
1142 WARN_ON_ONCE(timer->cb_mode == HRTIMER_CB_IRQSAFE_NO_SOFTIRQ);
1144 timer_stats_account_hrtimer(timer);
1146 fn = timer->function;
1147 __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0);
1148 spin_unlock_irq(&cpu_base->lock);
1150 restart = fn(timer);
1152 spin_lock_irq(&cpu_base->lock);
1154 timer->state &= ~HRTIMER_STATE_CALLBACK;
1155 if (restart != HRTIMER_NORESTART) {
1156 BUG_ON(hrtimer_active(timer));
1157 enqueue_hrtimer(timer, base, 0);
1160 spin_unlock_irq(&cpu_base->lock);
1164 * Called from timer softirq every jiffy, expire hrtimers:
1166 * For HRT its the fall back code to run the softirq in the timer
1167 * softirq context in case the hrtimer initialization failed or has
1168 * not been done yet.
1170 void hrtimer_run_queues(void)
1172 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1175 if (hrtimer_hres_active())
1179 * This _is_ ugly: We have to check in the softirq context,
1180 * whether we can switch to highres and / or nohz mode. The
1181 * clocksource switch happens in the timer interrupt with
1182 * xtime_lock held. Notification from there only sets the
1183 * check bit in the tick_oneshot code, otherwise we might
1184 * deadlock vs. xtime_lock.
1186 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1187 if (hrtimer_switch_to_hres())
1190 hrtimer_get_softirq_time(cpu_base);
1192 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
1193 run_hrtimer_queue(cpu_base, i);
1197 * Sleep related functions:
1199 static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1201 struct hrtimer_sleeper *t =
1202 container_of(timer, struct hrtimer_sleeper, timer);
1203 struct task_struct *task = t->task;
1207 wake_up_process(task);
1209 return HRTIMER_NORESTART;
1212 void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1214 sl->timer.function = hrtimer_wakeup;
1216 #ifdef CONFIG_HIGH_RES_TIMERS
1217 sl->timer.cb_mode = HRTIMER_CB_IRQSAFE_NO_RESTART;
1221 static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1223 hrtimer_init_sleeper(t, current);
1226 set_current_state(TASK_INTERRUPTIBLE);
1227 hrtimer_start(&t->timer, t->timer.expires, mode);
1229 if (likely(t->task))
1232 hrtimer_cancel(&t->timer);
1233 mode = HRTIMER_MODE_ABS;
1235 } while (t->task && !signal_pending(current));
1237 return t->task == NULL;
1240 long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1242 struct hrtimer_sleeper t;
1243 struct timespec __user *rmtp;
1247 restart->fn = do_no_restart_syscall;
1249 hrtimer_init(&t.timer, restart->arg0, HRTIMER_MODE_ABS);
1250 t.timer.expires.tv64 = ((u64)restart->arg3 << 32) | (u64) restart->arg2;
1252 if (do_nanosleep(&t, HRTIMER_MODE_ABS))
1255 rmtp = (struct timespec __user *) restart->arg1;
1257 time = ktime_sub(t.timer.expires, t.timer.base->get_time());
1260 tu = ktime_to_timespec(time);
1261 if (copy_to_user(rmtp, &tu, sizeof(tu)))
1265 restart->fn = hrtimer_nanosleep_restart;
1267 /* The other values in restart are already filled in */
1268 return -ERESTART_RESTARTBLOCK;
1271 long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
1272 const enum hrtimer_mode mode, const clockid_t clockid)
1274 struct restart_block *restart;
1275 struct hrtimer_sleeper t;
1279 hrtimer_init(&t.timer, clockid, mode);
1280 t.timer.expires = timespec_to_ktime(*rqtp);
1281 if (do_nanosleep(&t, mode))
1284 /* Absolute timers do not update the rmtp value and restart: */
1285 if (mode == HRTIMER_MODE_ABS)
1286 return -ERESTARTNOHAND;
1289 rem = ktime_sub(t.timer.expires, t.timer.base->get_time());
1292 tu = ktime_to_timespec(rem);
1293 if (copy_to_user(rmtp, &tu, sizeof(tu)))
1297 restart = ¤t_thread_info()->restart_block;
1298 restart->fn = hrtimer_nanosleep_restart;
1299 restart->arg0 = (unsigned long) t.timer.base->index;
1300 restart->arg1 = (unsigned long) rmtp;
1301 restart->arg2 = t.timer.expires.tv64 & 0xFFFFFFFF;
1302 restart->arg3 = t.timer.expires.tv64 >> 32;
1304 return -ERESTART_RESTARTBLOCK;
1308 sys_nanosleep(struct timespec __user *rqtp, struct timespec __user *rmtp)
1312 if (copy_from_user(&tu, rqtp, sizeof(tu)))
1315 if (!timespec_valid(&tu))
1318 return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1322 * Functions related to boot-time initialization:
1324 static void __devinit init_hrtimers_cpu(int cpu)
1326 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1329 spin_lock_init(&cpu_base->lock);
1330 lockdep_set_class(&cpu_base->lock, &cpu_base->lock_key);
1332 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
1333 cpu_base->clock_base[i].cpu_base = cpu_base;
1335 hrtimer_init_hres(cpu_base);
1338 #ifdef CONFIG_HOTPLUG_CPU
1340 static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1341 struct hrtimer_clock_base *new_base)
1343 struct hrtimer *timer;
1344 struct rb_node *node;
1346 while ((node = rb_first(&old_base->active))) {
1347 timer = rb_entry(node, struct hrtimer, node);
1348 BUG_ON(hrtimer_callback_running(timer));
1349 __remove_hrtimer(timer, old_base, HRTIMER_STATE_INACTIVE, 0);
1350 timer->base = new_base;
1352 * Enqueue the timer. Allow reprogramming of the event device
1354 enqueue_hrtimer(timer, new_base, 1);
1358 static void migrate_hrtimers(int cpu)
1360 struct hrtimer_cpu_base *old_base, *new_base;
1363 BUG_ON(cpu_online(cpu));
1364 old_base = &per_cpu(hrtimer_bases, cpu);
1365 new_base = &get_cpu_var(hrtimer_bases);
1367 tick_cancel_sched_timer(cpu);
1369 local_irq_disable();
1370 double_spin_lock(&new_base->lock, &old_base->lock,
1371 smp_processor_id() < cpu);
1373 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1374 migrate_hrtimer_list(&old_base->clock_base[i],
1375 &new_base->clock_base[i]);
1378 double_spin_unlock(&new_base->lock, &old_base->lock,
1379 smp_processor_id() < cpu);
1381 put_cpu_var(hrtimer_bases);
1383 #endif /* CONFIG_HOTPLUG_CPU */
1385 static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self,
1386 unsigned long action, void *hcpu)
1388 long cpu = (long)hcpu;
1392 case CPU_UP_PREPARE:
1393 init_hrtimers_cpu(cpu);
1396 #ifdef CONFIG_HOTPLUG_CPU
1398 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &cpu);
1399 migrate_hrtimers(cpu);
1410 static struct notifier_block __cpuinitdata hrtimers_nb = {
1411 .notifier_call = hrtimer_cpu_notify,
1414 void __init hrtimers_init(void)
1416 hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
1417 (void *)(long)smp_processor_id());
1418 register_cpu_notifier(&hrtimers_nb);
1419 #ifdef CONFIG_HIGH_RES_TIMERS
1420 open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq, NULL);