2 * linux/kernel/hrtimer.c
4 * Copyright(C) 2005-2006, Thomas Gleixner <tglx@linutronix.de>
5 * Copyright(C) 2005-2006, Red Hat, Inc., Ingo Molnar
6 * Copyright(C) 2006 Timesys Corp., Thomas Gleixner <tglx@timesys.com>
8 * High-resolution kernel timers
10 * In contrast to the low-resolution timeout API implemented in
11 * kernel/timer.c, hrtimers provide finer resolution and accuracy
12 * depending on system configuration and capabilities.
14 * These timers are currently used for:
18 * - precise in-kernel timing
20 * Started by: Thomas Gleixner and Ingo Molnar
23 * based on kernel/timer.c
25 * Help, testing, suggestions, bugfixes, improvements were
28 * George Anzinger, Andrew Morton, Steven Rostedt, Roman Zippel
31 * For licencing details see kernel-base/COPYING
34 #include <linux/cpu.h>
35 #include <linux/module.h>
36 #include <linux/percpu.h>
37 #include <linux/hrtimer.h>
38 #include <linux/notifier.h>
39 #include <linux/syscalls.h>
40 #include <linux/interrupt.h>
42 #include <asm/uaccess.h>
45 * ktime_get - get the monotonic time in ktime_t format
47 * returns the time in ktime_t format
49 static ktime_t ktime_get(void)
55 return timespec_to_ktime(now);
59 * ktime_get_real - get the real (wall-) time in ktime_t format
61 * returns the time in ktime_t format
63 static ktime_t ktime_get_real(void)
69 return timespec_to_ktime(now);
72 EXPORT_SYMBOL_GPL(ktime_get_real);
77 * Note: If we want to add new timer bases, we have to skip the two
78 * clock ids captured by the cpu-timers. We do this by holding empty
79 * entries rather than doing math adjustment of the clock ids.
80 * This ensures that we capture erroneous accesses to these clock ids
81 * rather than moving them into the range of valid clock id's.
83 static DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
89 .index = CLOCK_REALTIME,
90 .get_time = &ktime_get_real,
91 .resolution = KTIME_REALTIME_RES,
94 .index = CLOCK_MONOTONIC,
95 .get_time = &ktime_get,
96 .resolution = KTIME_MONOTONIC_RES,
102 * ktime_get_ts - get the monotonic clock in timespec format
103 * @ts: pointer to timespec variable
105 * The function calculates the monotonic clock from the realtime
106 * clock and the wall_to_monotonic offset and stores the result
107 * in normalized timespec format in the variable pointed to by @ts.
109 void ktime_get_ts(struct timespec *ts)
111 struct timespec tomono;
115 seq = read_seqbegin(&xtime_lock);
117 tomono = wall_to_monotonic;
119 } while (read_seqretry(&xtime_lock, seq));
121 set_normalized_timespec(ts, ts->tv_sec + tomono.tv_sec,
122 ts->tv_nsec + tomono.tv_nsec);
124 EXPORT_SYMBOL_GPL(ktime_get_ts);
127 * Get the coarse grained time at the softirq based on xtime and
130 static void hrtimer_get_softirq_time(struct hrtimer_cpu_base *base)
132 ktime_t xtim, tomono;
137 seq = read_seqbegin(&xtime_lock);
139 getnstimeofday(&xts);
143 } while (read_seqretry(&xtime_lock, seq));
145 xtim = timespec_to_ktime(xts);
146 tomono = timespec_to_ktime(wall_to_monotonic);
147 base->clock_base[CLOCK_REALTIME].softirq_time = xtim;
148 base->clock_base[CLOCK_MONOTONIC].softirq_time =
149 ktime_add(xtim, tomono);
153 * Helper function to check, whether the timer is on one of the queues
155 static inline int hrtimer_is_queued(struct hrtimer *timer)
157 return timer->state & HRTIMER_STATE_ENQUEUED;
161 * Helper function to check, whether the timer is running the callback
164 static inline int hrtimer_callback_running(struct hrtimer *timer)
166 return timer->state & HRTIMER_STATE_CALLBACK;
170 * Functions and macros which are different for UP/SMP systems are kept in a
176 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
177 * means that all timers which are tied to this base via timer->base are
178 * locked, and the base itself is locked too.
180 * So __run_timers/migrate_timers can safely modify all timers which could
181 * be found on the lists/queues.
183 * When the timer's base is locked, and the timer removed from list, it is
184 * possible to set timer->base = NULL and drop the lock: the timer remains
188 struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
189 unsigned long *flags)
191 struct hrtimer_clock_base *base;
195 if (likely(base != NULL)) {
196 spin_lock_irqsave(&base->cpu_base->lock, *flags);
197 if (likely(base == timer->base))
199 /* The timer has migrated to another CPU: */
200 spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
207 * Switch the timer base to the current CPU when possible.
209 static inline struct hrtimer_clock_base *
210 switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base)
212 struct hrtimer_clock_base *new_base;
213 struct hrtimer_cpu_base *new_cpu_base;
215 new_cpu_base = &__get_cpu_var(hrtimer_bases);
216 new_base = &new_cpu_base->clock_base[base->index];
218 if (base != new_base) {
220 * We are trying to schedule the timer on the local CPU.
221 * However we can't change timer's base while it is running,
222 * so we keep it on the same CPU. No hassle vs. reprogramming
223 * the event source in the high resolution case. The softirq
224 * code will take care of this when the timer function has
225 * completed. There is no conflict as we hold the lock until
226 * the timer is enqueued.
228 if (unlikely(timer->state & HRTIMER_STATE_CALLBACK))
231 /* See the comment in lock_timer_base() */
233 spin_unlock(&base->cpu_base->lock);
234 spin_lock(&new_base->cpu_base->lock);
235 timer->base = new_base;
240 #else /* CONFIG_SMP */
242 static inline struct hrtimer_clock_base *
243 lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
245 struct hrtimer_clock_base *base = timer->base;
247 spin_lock_irqsave(&base->cpu_base->lock, *flags);
252 #define switch_hrtimer_base(t, b) (b)
254 #endif /* !CONFIG_SMP */
257 * Functions for the union type storage format of ktime_t which are
258 * too large for inlining:
260 #if BITS_PER_LONG < 64
261 # ifndef CONFIG_KTIME_SCALAR
263 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
265 * @nsec: the scalar nsec value to add
267 * Returns the sum of kt and nsec in ktime_t format
269 ktime_t ktime_add_ns(const ktime_t kt, u64 nsec)
273 if (likely(nsec < NSEC_PER_SEC)) {
276 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
278 tmp = ktime_set((long)nsec, rem);
281 return ktime_add(kt, tmp);
284 #else /* CONFIG_KTIME_SCALAR */
286 # endif /* !CONFIG_KTIME_SCALAR */
289 * Divide a ktime value by a nanosecond value
291 static unsigned long ktime_divns(const ktime_t kt, s64 div)
296 dclc = dns = ktime_to_ns(kt);
298 /* Make sure the divisor is less than 2^32: */
304 do_div(dclc, (unsigned long) div);
306 return (unsigned long) dclc;
309 #else /* BITS_PER_LONG < 64 */
310 # define ktime_divns(kt, div) (unsigned long)((kt).tv64 / (div))
311 #endif /* BITS_PER_LONG >= 64 */
314 * Timekeeping resumed notification
316 void hrtimer_notify_resume(void)
322 * Counterpart to lock_timer_base above:
325 void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
327 spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
331 * hrtimer_forward - forward the timer expiry
332 * @timer: hrtimer to forward
333 * @now: forward past this time
334 * @interval: the interval to forward
336 * Forward the timer expiry so it will expire in the future.
337 * Returns the number of overruns.
340 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
342 unsigned long orun = 1;
345 delta = ktime_sub(now, timer->expires);
350 if (interval.tv64 < timer->base->resolution.tv64)
351 interval.tv64 = timer->base->resolution.tv64;
353 if (unlikely(delta.tv64 >= interval.tv64)) {
354 s64 incr = ktime_to_ns(interval);
356 orun = ktime_divns(delta, incr);
357 timer->expires = ktime_add_ns(timer->expires, incr * orun);
358 if (timer->expires.tv64 > now.tv64)
361 * This (and the ktime_add() below) is the
362 * correction for exact:
366 timer->expires = ktime_add(timer->expires, interval);
372 * enqueue_hrtimer - internal function to (re)start a timer
374 * The timer is inserted in expiry order. Insertion into the
375 * red black tree is O(log(n)). Must hold the base lock.
377 static void enqueue_hrtimer(struct hrtimer *timer,
378 struct hrtimer_clock_base *base)
380 struct rb_node **link = &base->active.rb_node;
381 struct rb_node *parent = NULL;
382 struct hrtimer *entry;
385 * Find the right place in the rbtree:
389 entry = rb_entry(parent, struct hrtimer, node);
391 * We dont care about collisions. Nodes with
392 * the same expiry time stay together.
394 if (timer->expires.tv64 < entry->expires.tv64)
395 link = &(*link)->rb_left;
397 link = &(*link)->rb_right;
401 * Insert the timer to the rbtree and check whether it
402 * replaces the first pending timer
404 rb_link_node(&timer->node, parent, link);
405 rb_insert_color(&timer->node, &base->active);
407 * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
408 * state of a possibly running callback.
410 timer->state |= HRTIMER_STATE_ENQUEUED;
412 if (!base->first || timer->expires.tv64 <
413 rb_entry(base->first, struct hrtimer, node)->expires.tv64)
414 base->first = &timer->node;
418 * __remove_hrtimer - internal function to remove a timer
420 * Caller must hold the base lock.
422 static void __remove_hrtimer(struct hrtimer *timer,
423 struct hrtimer_clock_base *base,
424 unsigned long newstate)
427 * Remove the timer from the rbtree and replace the
428 * first entry pointer if necessary.
430 if (base->first == &timer->node)
431 base->first = rb_next(&timer->node);
432 rb_erase(&timer->node, &base->active);
433 timer->state = newstate;
437 * remove hrtimer, called with base lock held
440 remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base)
442 if (hrtimer_is_queued(timer)) {
443 __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE);
450 * hrtimer_start - (re)start an relative timer on the current CPU
451 * @timer: the timer to be added
453 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
457 * 1 when the timer was active
460 hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode)
462 struct hrtimer_clock_base *base, *new_base;
466 base = lock_hrtimer_base(timer, &flags);
468 /* Remove an active timer from the queue: */
469 ret = remove_hrtimer(timer, base);
471 /* Switch the timer base, if necessary: */
472 new_base = switch_hrtimer_base(timer, base);
474 if (mode == HRTIMER_MODE_REL) {
475 tim = ktime_add(tim, new_base->get_time());
477 * CONFIG_TIME_LOW_RES is a temporary way for architectures
478 * to signal that they simply return xtime in
479 * do_gettimeoffset(). In this case we want to round up by
480 * resolution when starting a relative timer, to avoid short
481 * timeouts. This will go away with the GTOD framework.
483 #ifdef CONFIG_TIME_LOW_RES
484 tim = ktime_add(tim, base->resolution);
487 timer->expires = tim;
489 enqueue_hrtimer(timer, new_base);
491 unlock_hrtimer_base(timer, &flags);
495 EXPORT_SYMBOL_GPL(hrtimer_start);
498 * hrtimer_try_to_cancel - try to deactivate a timer
499 * @timer: hrtimer to stop
502 * 0 when the timer was not active
503 * 1 when the timer was active
504 * -1 when the timer is currently excuting the callback function and
507 int hrtimer_try_to_cancel(struct hrtimer *timer)
509 struct hrtimer_clock_base *base;
513 base = lock_hrtimer_base(timer, &flags);
515 if (!hrtimer_callback_running(timer))
516 ret = remove_hrtimer(timer, base);
518 unlock_hrtimer_base(timer, &flags);
523 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
526 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
527 * @timer: the timer to be cancelled
530 * 0 when the timer was not active
531 * 1 when the timer was active
533 int hrtimer_cancel(struct hrtimer *timer)
536 int ret = hrtimer_try_to_cancel(timer);
543 EXPORT_SYMBOL_GPL(hrtimer_cancel);
546 * hrtimer_get_remaining - get remaining time for the timer
547 * @timer: the timer to read
549 ktime_t hrtimer_get_remaining(const struct hrtimer *timer)
551 struct hrtimer_clock_base *base;
555 base = lock_hrtimer_base(timer, &flags);
556 rem = ktime_sub(timer->expires, base->get_time());
557 unlock_hrtimer_base(timer, &flags);
561 EXPORT_SYMBOL_GPL(hrtimer_get_remaining);
563 #if defined(CONFIG_NO_IDLE_HZ) || defined(CONFIG_NO_HZ)
565 * hrtimer_get_next_event - get the time until next expiry event
567 * Returns the delta to the next expiry event or KTIME_MAX if no timer
570 ktime_t hrtimer_get_next_event(void)
572 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
573 struct hrtimer_clock_base *base = cpu_base->clock_base;
574 ktime_t delta, mindelta = { .tv64 = KTIME_MAX };
578 spin_lock_irqsave(&cpu_base->lock, flags);
580 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
581 struct hrtimer *timer;
586 timer = rb_entry(base->first, struct hrtimer, node);
587 delta.tv64 = timer->expires.tv64;
588 delta = ktime_sub(delta, base->get_time());
589 if (delta.tv64 < mindelta.tv64)
590 mindelta.tv64 = delta.tv64;
593 spin_unlock_irqrestore(&cpu_base->lock, flags);
595 if (mindelta.tv64 < 0)
602 * hrtimer_init - initialize a timer to the given clock
603 * @timer: the timer to be initialized
604 * @clock_id: the clock to be used
605 * @mode: timer mode abs/rel
607 void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
608 enum hrtimer_mode mode)
610 struct hrtimer_cpu_base *cpu_base;
612 memset(timer, 0, sizeof(struct hrtimer));
614 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
616 if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
617 clock_id = CLOCK_MONOTONIC;
619 timer->base = &cpu_base->clock_base[clock_id];
621 EXPORT_SYMBOL_GPL(hrtimer_init);
624 * hrtimer_get_res - get the timer resolution for a clock
625 * @which_clock: which clock to query
626 * @tp: pointer to timespec variable to store the resolution
628 * Store the resolution of the clock selected by @which_clock in the
629 * variable pointed to by @tp.
631 int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp)
633 struct hrtimer_cpu_base *cpu_base;
635 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
636 *tp = ktime_to_timespec(cpu_base->clock_base[which_clock].resolution);
640 EXPORT_SYMBOL_GPL(hrtimer_get_res);
643 * Expire the per base hrtimer-queue:
645 static inline void run_hrtimer_queue(struct hrtimer_cpu_base *cpu_base,
648 struct rb_node *node;
649 struct hrtimer_clock_base *base = &cpu_base->clock_base[index];
654 if (base->get_softirq_time)
655 base->softirq_time = base->get_softirq_time();
657 spin_lock_irq(&cpu_base->lock);
659 while ((node = base->first)) {
660 struct hrtimer *timer;
661 enum hrtimer_restart (*fn)(struct hrtimer *);
664 timer = rb_entry(node, struct hrtimer, node);
665 if (base->softirq_time.tv64 <= timer->expires.tv64)
668 fn = timer->function;
669 __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK);
670 spin_unlock_irq(&cpu_base->lock);
674 spin_lock_irq(&cpu_base->lock);
676 timer->state &= ~HRTIMER_STATE_CALLBACK;
677 if (restart != HRTIMER_NORESTART) {
678 BUG_ON(hrtimer_active(timer));
679 enqueue_hrtimer(timer, base);
682 spin_unlock_irq(&cpu_base->lock);
686 * Called from timer softirq every jiffy, expire hrtimers:
688 void hrtimer_run_queues(void)
690 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
693 hrtimer_get_softirq_time(cpu_base);
695 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
696 run_hrtimer_queue(cpu_base, i);
700 * Sleep related functions:
702 static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
704 struct hrtimer_sleeper *t =
705 container_of(timer, struct hrtimer_sleeper, timer);
706 struct task_struct *task = t->task;
710 wake_up_process(task);
712 return HRTIMER_NORESTART;
715 void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
717 sl->timer.function = hrtimer_wakeup;
721 static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
723 hrtimer_init_sleeper(t, current);
726 set_current_state(TASK_INTERRUPTIBLE);
727 hrtimer_start(&t->timer, t->timer.expires, mode);
731 hrtimer_cancel(&t->timer);
732 mode = HRTIMER_MODE_ABS;
734 } while (t->task && !signal_pending(current));
736 return t->task == NULL;
739 long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
741 struct hrtimer_sleeper t;
742 struct timespec __user *rmtp;
746 restart->fn = do_no_restart_syscall;
748 hrtimer_init(&t.timer, restart->arg0, HRTIMER_MODE_ABS);
749 t.timer.expires.tv64 = ((u64)restart->arg3 << 32) | (u64) restart->arg2;
751 if (do_nanosleep(&t, HRTIMER_MODE_ABS))
754 rmtp = (struct timespec __user *) restart->arg1;
756 time = ktime_sub(t.timer.expires, t.timer.base->get_time());
759 tu = ktime_to_timespec(time);
760 if (copy_to_user(rmtp, &tu, sizeof(tu)))
764 restart->fn = hrtimer_nanosleep_restart;
766 /* The other values in restart are already filled in */
767 return -ERESTART_RESTARTBLOCK;
770 long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
771 const enum hrtimer_mode mode, const clockid_t clockid)
773 struct restart_block *restart;
774 struct hrtimer_sleeper t;
778 hrtimer_init(&t.timer, clockid, mode);
779 t.timer.expires = timespec_to_ktime(*rqtp);
780 if (do_nanosleep(&t, mode))
783 /* Absolute timers do not update the rmtp value and restart: */
784 if (mode == HRTIMER_MODE_ABS)
785 return -ERESTARTNOHAND;
788 rem = ktime_sub(t.timer.expires, t.timer.base->get_time());
791 tu = ktime_to_timespec(rem);
792 if (copy_to_user(rmtp, &tu, sizeof(tu)))
796 restart = ¤t_thread_info()->restart_block;
797 restart->fn = hrtimer_nanosleep_restart;
798 restart->arg0 = (unsigned long) t.timer.base->index;
799 restart->arg1 = (unsigned long) rmtp;
800 restart->arg2 = t.timer.expires.tv64 & 0xFFFFFFFF;
801 restart->arg3 = t.timer.expires.tv64 >> 32;
803 return -ERESTART_RESTARTBLOCK;
807 sys_nanosleep(struct timespec __user *rqtp, struct timespec __user *rmtp)
811 if (copy_from_user(&tu, rqtp, sizeof(tu)))
814 if (!timespec_valid(&tu))
817 return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
821 * Functions related to boot-time initialization:
823 static void __devinit init_hrtimers_cpu(int cpu)
825 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
828 spin_lock_init(&cpu_base->lock);
829 lockdep_set_class(&cpu_base->lock, &cpu_base->lock_key);
831 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
832 cpu_base->clock_base[i].cpu_base = cpu_base;
836 #ifdef CONFIG_HOTPLUG_CPU
838 static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
839 struct hrtimer_clock_base *new_base)
841 struct hrtimer *timer;
842 struct rb_node *node;
844 while ((node = rb_first(&old_base->active))) {
845 timer = rb_entry(node, struct hrtimer, node);
846 BUG_ON(timer->state & HRTIMER_STATE_CALLBACK);
847 __remove_hrtimer(timer, old_base, HRTIMER_STATE_INACTIVE);
848 timer->base = new_base;
849 enqueue_hrtimer(timer, new_base);
853 static void migrate_hrtimers(int cpu)
855 struct hrtimer_cpu_base *old_base, *new_base;
858 BUG_ON(cpu_online(cpu));
859 old_base = &per_cpu(hrtimer_bases, cpu);
860 new_base = &get_cpu_var(hrtimer_bases);
864 spin_lock(&new_base->lock);
865 spin_lock(&old_base->lock);
867 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
868 migrate_hrtimer_list(&old_base->clock_base[i],
869 &new_base->clock_base[i]);
871 spin_unlock(&old_base->lock);
872 spin_unlock(&new_base->lock);
875 put_cpu_var(hrtimer_bases);
877 #endif /* CONFIG_HOTPLUG_CPU */
879 static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self,
880 unsigned long action, void *hcpu)
882 long cpu = (long)hcpu;
887 init_hrtimers_cpu(cpu);
890 #ifdef CONFIG_HOTPLUG_CPU
892 migrate_hrtimers(cpu);
903 static struct notifier_block __cpuinitdata hrtimers_nb = {
904 .notifier_call = hrtimer_cpu_notify,
907 void __init hrtimers_init(void)
909 hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
910 (void *)(long)smp_processor_id());
911 register_cpu_notifier(&hrtimers_nb);