2 * linux/kernel/hrtimer.c
4 * Copyright(C) 2005, Thomas Gleixner <tglx@linutronix.de>
5 * Copyright(C) 2005, Red Hat, Inc., Ingo Molnar
7 * High-resolution kernel timers
9 * In contrast to the low-resolution timeout API implemented in
10 * kernel/timer.c, hrtimers provide finer resolution and accuracy
11 * depending on system configuration and capabilities.
13 * These timers are currently used for:
17 * - precise in-kernel timing
19 * Started by: Thomas Gleixner and Ingo Molnar
22 * based on kernel/timer.c
24 * For licencing details see kernel-base/COPYING
27 #include <linux/cpu.h>
28 #include <linux/module.h>
29 #include <linux/percpu.h>
30 #include <linux/hrtimer.h>
31 #include <linux/notifier.h>
32 #include <linux/syscalls.h>
33 #include <linux/interrupt.h>
35 #include <asm/uaccess.h>
38 * ktime_get - get the monotonic time in ktime_t format
40 * returns the time in ktime_t format
42 static ktime_t ktime_get(void)
48 return timespec_to_ktime(now);
52 * ktime_get_real - get the real (wall-) time in ktime_t format
54 * returns the time in ktime_t format
56 static ktime_t ktime_get_real(void)
62 return timespec_to_ktime(now);
65 EXPORT_SYMBOL_GPL(ktime_get_real);
71 #define MAX_HRTIMER_BASES 2
73 static DEFINE_PER_CPU(struct hrtimer_base, hrtimer_bases[MAX_HRTIMER_BASES]) =
76 .index = CLOCK_REALTIME,
77 .get_time = &ktime_get_real,
78 .resolution = KTIME_REALTIME_RES,
81 .index = CLOCK_MONOTONIC,
82 .get_time = &ktime_get,
83 .resolution = KTIME_MONOTONIC_RES,
88 * ktime_get_ts - get the monotonic clock in timespec format
90 * @ts: pointer to timespec variable
92 * The function calculates the monotonic clock from the realtime
93 * clock and the wall_to_monotonic offset and stores the result
94 * in normalized timespec format in the variable pointed to by ts.
96 void ktime_get_ts(struct timespec *ts)
98 struct timespec tomono;
102 seq = read_seqbegin(&xtime_lock);
104 tomono = wall_to_monotonic;
106 } while (read_seqretry(&xtime_lock, seq));
108 set_normalized_timespec(ts, ts->tv_sec + tomono.tv_sec,
109 ts->tv_nsec + tomono.tv_nsec);
111 EXPORT_SYMBOL_GPL(ktime_get_ts);
114 * Functions and macros which are different for UP/SMP systems are kept in a
119 #define set_curr_timer(b, t) do { (b)->curr_timer = (t); } while (0)
122 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
123 * means that all timers which are tied to this base via timer->base are
124 * locked, and the base itself is locked too.
126 * So __run_timers/migrate_timers can safely modify all timers which could
127 * be found on the lists/queues.
129 * When the timer's base is locked, and the timer removed from list, it is
130 * possible to set timer->base = NULL and drop the lock: the timer remains
133 static struct hrtimer_base *lock_hrtimer_base(const struct hrtimer *timer,
134 unsigned long *flags)
136 struct hrtimer_base *base;
140 if (likely(base != NULL)) {
141 spin_lock_irqsave(&base->lock, *flags);
142 if (likely(base == timer->base))
144 /* The timer has migrated to another CPU: */
145 spin_unlock_irqrestore(&base->lock, *flags);
152 * Switch the timer base to the current CPU when possible.
154 static inline struct hrtimer_base *
155 switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_base *base)
157 struct hrtimer_base *new_base;
159 new_base = &__get_cpu_var(hrtimer_bases[base->index]);
161 if (base != new_base) {
163 * We are trying to schedule the timer on the local CPU.
164 * However we can't change timer's base while it is running,
165 * so we keep it on the same CPU. No hassle vs. reprogramming
166 * the event source in the high resolution case. The softirq
167 * code will take care of this when the timer function has
168 * completed. There is no conflict as we hold the lock until
169 * the timer is enqueued.
171 if (unlikely(base->curr_timer == timer))
174 /* See the comment in lock_timer_base() */
176 spin_unlock(&base->lock);
177 spin_lock(&new_base->lock);
178 timer->base = new_base;
183 #else /* CONFIG_SMP */
185 #define set_curr_timer(b, t) do { } while (0)
187 static inline struct hrtimer_base *
188 lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
190 struct hrtimer_base *base = timer->base;
192 spin_lock_irqsave(&base->lock, *flags);
197 #define switch_hrtimer_base(t, b) (b)
199 #endif /* !CONFIG_SMP */
202 * Functions for the union type storage format of ktime_t which are
203 * too large for inlining:
205 #if BITS_PER_LONG < 64
206 # ifndef CONFIG_KTIME_SCALAR
208 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
211 * @nsec: the scalar nsec value to add
213 * Returns the sum of kt and nsec in ktime_t format
215 ktime_t ktime_add_ns(const ktime_t kt, u64 nsec)
219 if (likely(nsec < NSEC_PER_SEC)) {
222 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
224 tmp = ktime_set((long)nsec, rem);
227 return ktime_add(kt, tmp);
230 #else /* CONFIG_KTIME_SCALAR */
232 # endif /* !CONFIG_KTIME_SCALAR */
235 * Divide a ktime value by a nanosecond value
237 static unsigned long ktime_divns(const ktime_t kt, nsec_t div)
242 dclc = dns = ktime_to_ns(kt);
244 /* Make sure the divisor is less than 2^32: */
250 do_div(dclc, (unsigned long) div);
252 return (unsigned long) dclc;
255 #else /* BITS_PER_LONG < 64 */
256 # define ktime_divns(kt, div) (unsigned long)((kt).tv64 / (div))
257 #endif /* BITS_PER_LONG >= 64 */
260 * Counterpart to lock_timer_base above:
263 void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
265 spin_unlock_irqrestore(&timer->base->lock, *flags);
269 * hrtimer_forward - forward the timer expiry
271 * @timer: hrtimer to forward
272 * @interval: the interval to forward
274 * Forward the timer expiry so it will expire in the future.
275 * The number of overruns is added to the overrun field.
278 hrtimer_forward(struct hrtimer *timer, const ktime_t interval)
280 unsigned long orun = 1;
283 now = timer->base->get_time();
285 delta = ktime_sub(now, timer->expires);
290 if (unlikely(delta.tv64 >= interval.tv64)) {
291 nsec_t incr = ktime_to_ns(interval);
293 orun = ktime_divns(delta, incr);
294 timer->expires = ktime_add_ns(timer->expires, incr * orun);
295 if (timer->expires.tv64 > now.tv64)
298 * This (and the ktime_add() below) is the
299 * correction for exact:
303 timer->expires = ktime_add(timer->expires, interval);
309 * enqueue_hrtimer - internal function to (re)start a timer
311 * The timer is inserted in expiry order. Insertion into the
312 * red black tree is O(log(n)). Must hold the base lock.
314 static void enqueue_hrtimer(struct hrtimer *timer, struct hrtimer_base *base)
316 struct rb_node **link = &base->active.rb_node;
317 struct list_head *prev = &base->pending;
318 struct rb_node *parent = NULL;
319 struct hrtimer *entry;
322 * Find the right place in the rbtree:
326 entry = rb_entry(parent, struct hrtimer, node);
328 * We dont care about collisions. Nodes with
329 * the same expiry time stay together.
331 if (timer->expires.tv64 < entry->expires.tv64)
332 link = &(*link)->rb_left;
334 link = &(*link)->rb_right;
340 * Insert the timer to the rbtree and to the sorted list:
342 rb_link_node(&timer->node, parent, link);
343 rb_insert_color(&timer->node, &base->active);
344 list_add(&timer->list, prev);
346 timer->state = HRTIMER_PENDING;
351 * __remove_hrtimer - internal function to remove a timer
353 * Caller must hold the base lock.
355 static void __remove_hrtimer(struct hrtimer *timer, struct hrtimer_base *base)
358 * Remove the timer from the sorted list and from the rbtree:
360 list_del(&timer->list);
361 rb_erase(&timer->node, &base->active);
365 * remove hrtimer, called with base lock held
368 remove_hrtimer(struct hrtimer *timer, struct hrtimer_base *base)
370 if (hrtimer_active(timer)) {
371 __remove_hrtimer(timer, base);
372 timer->state = HRTIMER_INACTIVE;
379 * hrtimer_start - (re)start an relative timer on the current CPU
381 * @timer: the timer to be added
383 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
387 * 1 when the timer was active
390 hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode)
392 struct hrtimer_base *base, *new_base;
396 base = lock_hrtimer_base(timer, &flags);
398 /* Remove an active timer from the queue: */
399 ret = remove_hrtimer(timer, base);
401 /* Switch the timer base, if necessary: */
402 new_base = switch_hrtimer_base(timer, base);
404 if (mode == HRTIMER_REL)
405 tim = ktime_add(tim, new_base->get_time());
406 timer->expires = tim;
408 enqueue_hrtimer(timer, new_base);
410 unlock_hrtimer_base(timer, &flags);
416 * hrtimer_try_to_cancel - try to deactivate a timer
418 * @timer: hrtimer to stop
421 * 0 when the timer was not active
422 * 1 when the timer was active
423 * -1 when the timer is currently excuting the callback function and
426 int hrtimer_try_to_cancel(struct hrtimer *timer)
428 struct hrtimer_base *base;
432 base = lock_hrtimer_base(timer, &flags);
434 if (base->curr_timer != timer)
435 ret = remove_hrtimer(timer, base);
437 unlock_hrtimer_base(timer, &flags);
444 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
446 * @timer: the timer to be cancelled
449 * 0 when the timer was not active
450 * 1 when the timer was active
452 int hrtimer_cancel(struct hrtimer *timer)
455 int ret = hrtimer_try_to_cancel(timer);
463 * hrtimer_get_remaining - get remaining time for the timer
465 * @timer: the timer to read
467 ktime_t hrtimer_get_remaining(const struct hrtimer *timer)
469 struct hrtimer_base *base;
473 base = lock_hrtimer_base(timer, &flags);
474 rem = ktime_sub(timer->expires, timer->base->get_time());
475 unlock_hrtimer_base(timer, &flags);
481 * hrtimer_rebase - rebase an initialized hrtimer to a different base
483 * @timer: the timer to be rebased
484 * @clock_id: the clock to be used
486 void hrtimer_rebase(struct hrtimer *timer, const clockid_t clock_id)
488 struct hrtimer_base *bases;
490 bases = per_cpu(hrtimer_bases, raw_smp_processor_id());
491 timer->base = &bases[clock_id];
495 * hrtimer_init - initialize a timer to the given clock
497 * @timer: the timer to be initialized
498 * @clock_id: the clock to be used
500 void hrtimer_init(struct hrtimer *timer, const clockid_t clock_id)
502 memset(timer, 0, sizeof(struct hrtimer));
503 hrtimer_rebase(timer, clock_id);
507 * hrtimer_get_res - get the timer resolution for a clock
509 * @which_clock: which clock to query
510 * @tp: pointer to timespec variable to store the resolution
512 * Store the resolution of the clock selected by which_clock in the
513 * variable pointed to by tp.
515 int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp)
517 struct hrtimer_base *bases;
520 bases = per_cpu(hrtimer_bases, raw_smp_processor_id());
521 tp->tv_nsec = bases[which_clock].resolution;
527 * Expire the per base hrtimer-queue:
529 static inline void run_hrtimer_queue(struct hrtimer_base *base)
531 ktime_t now = base->get_time();
533 spin_lock_irq(&base->lock);
535 while (!list_empty(&base->pending)) {
536 struct hrtimer *timer;
541 timer = list_entry(base->pending.next, struct hrtimer, list);
542 if (now.tv64 <= timer->expires.tv64)
545 fn = timer->function;
547 set_curr_timer(base, timer);
548 __remove_hrtimer(timer, base);
549 spin_unlock_irq(&base->lock);
552 * fn == NULL is special case for the simplest timer
553 * variant - wake up process and do not restart:
556 wake_up_process(data);
557 restart = HRTIMER_NORESTART;
561 spin_lock_irq(&base->lock);
563 if (restart == HRTIMER_RESTART)
564 enqueue_hrtimer(timer, base);
566 timer->state = HRTIMER_EXPIRED;
568 set_curr_timer(base, NULL);
569 spin_unlock_irq(&base->lock);
573 * Called from timer softirq every jiffy, expire hrtimers:
575 void hrtimer_run_queues(void)
577 struct hrtimer_base *base = __get_cpu_var(hrtimer_bases);
580 for (i = 0; i < MAX_HRTIMER_BASES; i++)
581 run_hrtimer_queue(&base[i]);
585 * Sleep related functions:
589 * schedule_hrtimer - sleep until timeout
591 * @timer: hrtimer variable initialized with the correct clock base
592 * @mode: timeout value is abs/rel
594 * Make the current task sleep until @timeout is
597 * You can set the task state as follows -
599 * %TASK_UNINTERRUPTIBLE - at least @timeout is guaranteed to
600 * pass before the routine returns. The routine will return 0
602 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
603 * delivered to the current task. In this case the remaining time
606 * The current task state is guaranteed to be TASK_RUNNING when this
609 static ktime_t __sched
610 schedule_hrtimer(struct hrtimer *timer, const enum hrtimer_mode mode)
612 /* fn stays NULL, meaning single-shot wakeup: */
613 timer->data = current;
615 hrtimer_start(timer, timer->expires, mode);
618 hrtimer_cancel(timer);
620 /* Return the remaining time: */
621 if (timer->state != HRTIMER_EXPIRED)
622 return ktime_sub(timer->expires, timer->base->get_time());
624 return (ktime_t) {.tv64 = 0 };
627 static inline ktime_t __sched
628 schedule_hrtimer_interruptible(struct hrtimer *timer,
629 const enum hrtimer_mode mode)
631 set_current_state(TASK_INTERRUPTIBLE);
633 return schedule_hrtimer(timer, mode);
637 nanosleep_restart(struct restart_block *restart, clockid_t clockid)
639 struct timespec __user *rmtp, tu;
640 void *rfn_save = restart->fn;
641 struct hrtimer timer;
644 restart->fn = do_no_restart_syscall;
646 hrtimer_init(&timer, clockid);
648 timer.expires.tv64 = ((u64)restart->arg1 << 32) | (u64) restart->arg0;
650 rem = schedule_hrtimer_interruptible(&timer, HRTIMER_ABS);
655 rmtp = (struct timespec __user *) restart->arg2;
656 tu = ktime_to_timespec(rem);
657 if (rmtp && copy_to_user(rmtp, &tu, sizeof(tu)))
660 restart->fn = rfn_save;
662 /* The other values in restart are already filled in */
663 return -ERESTART_RESTARTBLOCK;
666 static long __sched nanosleep_restart_mono(struct restart_block *restart)
668 return nanosleep_restart(restart, CLOCK_MONOTONIC);
671 static long __sched nanosleep_restart_real(struct restart_block *restart)
673 return nanosleep_restart(restart, CLOCK_REALTIME);
676 long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
677 const enum hrtimer_mode mode, const clockid_t clockid)
679 struct restart_block *restart;
680 struct hrtimer timer;
684 hrtimer_init(&timer, clockid);
686 timer.expires = timespec_to_ktime(*rqtp);
688 rem = schedule_hrtimer_interruptible(&timer, mode);
692 /* Absolute timers do not update the rmtp value: */
693 if (mode == HRTIMER_ABS)
694 return -ERESTARTNOHAND;
696 tu = ktime_to_timespec(rem);
698 if (rmtp && copy_to_user(rmtp, &tu, sizeof(tu)))
701 restart = ¤t_thread_info()->restart_block;
702 restart->fn = (clockid == CLOCK_MONOTONIC) ?
703 nanosleep_restart_mono : nanosleep_restart_real;
704 restart->arg0 = timer.expires.tv64 & 0xFFFFFFFF;
705 restart->arg1 = timer.expires.tv64 >> 32;
706 restart->arg2 = (unsigned long) rmtp;
708 return -ERESTART_RESTARTBLOCK;
712 sys_nanosleep(struct timespec __user *rqtp, struct timespec __user *rmtp)
716 if (copy_from_user(&tu, rqtp, sizeof(tu)))
719 if (!timespec_valid(&tu))
722 return hrtimer_nanosleep(&tu, rmtp, HRTIMER_REL, CLOCK_MONOTONIC);
726 * Functions related to boot-time initialization:
728 static void __devinit init_hrtimers_cpu(int cpu)
730 struct hrtimer_base *base = per_cpu(hrtimer_bases, cpu);
733 for (i = 0; i < MAX_HRTIMER_BASES; i++) {
734 spin_lock_init(&base->lock);
735 INIT_LIST_HEAD(&base->pending);
740 #ifdef CONFIG_HOTPLUG_CPU
742 static void migrate_hrtimer_list(struct hrtimer_base *old_base,
743 struct hrtimer_base *new_base)
745 struct hrtimer *timer;
746 struct rb_node *node;
748 while ((node = rb_first(&old_base->active))) {
749 timer = rb_entry(node, struct hrtimer, node);
750 __remove_hrtimer(timer, old_base);
751 timer->base = new_base;
752 enqueue_hrtimer(timer, new_base);
756 static void migrate_hrtimers(int cpu)
758 struct hrtimer_base *old_base, *new_base;
761 BUG_ON(cpu_online(cpu));
762 old_base = per_cpu(hrtimer_bases, cpu);
763 new_base = get_cpu_var(hrtimer_bases);
767 for (i = 0; i < MAX_HRTIMER_BASES; i++) {
769 spin_lock(&new_base->lock);
770 spin_lock(&old_base->lock);
772 BUG_ON(old_base->curr_timer);
774 migrate_hrtimer_list(old_base, new_base);
776 spin_unlock(&old_base->lock);
777 spin_unlock(&new_base->lock);
783 put_cpu_var(hrtimer_bases);
785 #endif /* CONFIG_HOTPLUG_CPU */
787 static int __devinit hrtimer_cpu_notify(struct notifier_block *self,
788 unsigned long action, void *hcpu)
790 long cpu = (long)hcpu;
795 init_hrtimers_cpu(cpu);
798 #ifdef CONFIG_HOTPLUG_CPU
800 migrate_hrtimers(cpu);
811 static struct notifier_block __devinitdata hrtimers_nb = {
812 .notifier_call = hrtimer_cpu_notify,
815 void __init hrtimers_init(void)
817 hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
818 (void *)(long)smp_processor_id());
819 register_cpu_notifier(&hrtimers_nb);