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);
113 * Functions and macros which are different for UP/SMP systems are kept in a
118 #define set_curr_timer(b, t) do { (b)->curr_timer = (t); } while (0)
121 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
122 * means that all timers which are tied to this base via timer->base are
123 * locked, and the base itself is locked too.
125 * So __run_timers/migrate_timers can safely modify all timers which could
126 * be found on the lists/queues.
128 * When the timer's base is locked, and the timer removed from list, it is
129 * possible to set timer->base = NULL and drop the lock: the timer remains
132 static struct hrtimer_base *lock_hrtimer_base(const struct hrtimer *timer,
133 unsigned long *flags)
135 struct hrtimer_base *base;
139 if (likely(base != NULL)) {
140 spin_lock_irqsave(&base->lock, *flags);
141 if (likely(base == timer->base))
143 /* The timer has migrated to another CPU: */
144 spin_unlock_irqrestore(&base->lock, *flags);
151 * Switch the timer base to the current CPU when possible.
153 static inline struct hrtimer_base *
154 switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_base *base)
156 struct hrtimer_base *new_base;
158 new_base = &__get_cpu_var(hrtimer_bases[base->index]);
160 if (base != new_base) {
162 * We are trying to schedule the timer on the local CPU.
163 * However we can't change timer's base while it is running,
164 * so we keep it on the same CPU. No hassle vs. reprogramming
165 * the event source in the high resolution case. The softirq
166 * code will take care of this when the timer function has
167 * completed. There is no conflict as we hold the lock until
168 * the timer is enqueued.
170 if (unlikely(base->curr_timer == timer))
173 /* See the comment in lock_timer_base() */
175 spin_unlock(&base->lock);
176 spin_lock(&new_base->lock);
177 timer->base = new_base;
182 #else /* CONFIG_SMP */
184 #define set_curr_timer(b, t) do { } while (0)
186 static inline struct hrtimer_base *
187 lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
189 struct hrtimer_base *base = timer->base;
191 spin_lock_irqsave(&base->lock, *flags);
196 #define switch_hrtimer_base(t, b) (b)
198 #endif /* !CONFIG_SMP */
201 * Functions for the union type storage format of ktime_t which are
202 * too large for inlining:
204 #if BITS_PER_LONG < 64
205 # ifndef CONFIG_KTIME_SCALAR
207 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
210 * @nsec: the scalar nsec value to add
212 * Returns the sum of kt and nsec in ktime_t format
214 ktime_t ktime_add_ns(const ktime_t kt, u64 nsec)
218 if (likely(nsec < NSEC_PER_SEC)) {
221 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
223 tmp = ktime_set((long)nsec, rem);
226 return ktime_add(kt, tmp);
229 #else /* CONFIG_KTIME_SCALAR */
231 # endif /* !CONFIG_KTIME_SCALAR */
234 * Divide a ktime value by a nanosecond value
236 static unsigned long ktime_divns(const ktime_t kt, nsec_t div)
241 dclc = dns = ktime_to_ns(kt);
243 /* Make sure the divisor is less than 2^32: */
249 do_div(dclc, (unsigned long) div);
251 return (unsigned long) dclc;
254 #else /* BITS_PER_LONG < 64 */
255 # define ktime_divns(kt, div) (unsigned long)((kt).tv64 / (div))
256 #endif /* BITS_PER_LONG >= 64 */
259 * Counterpart to lock_timer_base above:
262 void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
264 spin_unlock_irqrestore(&timer->base->lock, *flags);
268 * hrtimer_forward - forward the timer expiry
270 * @timer: hrtimer to forward
271 * @interval: the interval to forward
273 * Forward the timer expiry so it will expire in the future.
274 * The number of overruns is added to the overrun field.
277 hrtimer_forward(struct hrtimer *timer, const ktime_t interval)
279 unsigned long orun = 1;
282 now = timer->base->get_time();
284 delta = ktime_sub(now, timer->expires);
289 if (unlikely(delta.tv64 >= interval.tv64)) {
290 nsec_t incr = ktime_to_ns(interval);
292 orun = ktime_divns(delta, incr);
293 timer->expires = ktime_add_ns(timer->expires, incr * orun);
294 if (timer->expires.tv64 > now.tv64)
297 * This (and the ktime_add() below) is the
298 * correction for exact:
302 timer->expires = ktime_add(timer->expires, interval);
308 * enqueue_hrtimer - internal function to (re)start a timer
310 * The timer is inserted in expiry order. Insertion into the
311 * red black tree is O(log(n)). Must hold the base lock.
313 static void enqueue_hrtimer(struct hrtimer *timer, struct hrtimer_base *base)
315 struct rb_node **link = &base->active.rb_node;
316 struct list_head *prev = &base->pending;
317 struct rb_node *parent = NULL;
318 struct hrtimer *entry;
321 * Find the right place in the rbtree:
325 entry = rb_entry(parent, struct hrtimer, node);
327 * We dont care about collisions. Nodes with
328 * the same expiry time stay together.
330 if (timer->expires.tv64 < entry->expires.tv64)
331 link = &(*link)->rb_left;
333 link = &(*link)->rb_right;
339 * Insert the timer to the rbtree and to the sorted list:
341 rb_link_node(&timer->node, parent, link);
342 rb_insert_color(&timer->node, &base->active);
343 list_add(&timer->list, prev);
345 timer->state = HRTIMER_PENDING;
350 * __remove_hrtimer - internal function to remove a timer
352 * Caller must hold the base lock.
354 static void __remove_hrtimer(struct hrtimer *timer, struct hrtimer_base *base)
357 * Remove the timer from the sorted list and from the rbtree:
359 list_del(&timer->list);
360 rb_erase(&timer->node, &base->active);
364 * remove hrtimer, called with base lock held
367 remove_hrtimer(struct hrtimer *timer, struct hrtimer_base *base)
369 if (hrtimer_active(timer)) {
370 __remove_hrtimer(timer, base);
371 timer->state = HRTIMER_INACTIVE;
378 * hrtimer_start - (re)start an relative timer on the current CPU
380 * @timer: the timer to be added
382 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
386 * 1 when the timer was active
389 hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode)
391 struct hrtimer_base *base, *new_base;
395 base = lock_hrtimer_base(timer, &flags);
397 /* Remove an active timer from the queue: */
398 ret = remove_hrtimer(timer, base);
400 /* Switch the timer base, if necessary: */
401 new_base = switch_hrtimer_base(timer, base);
403 if (mode == HRTIMER_REL)
404 tim = ktime_add(tim, new_base->get_time());
405 timer->expires = tim;
407 enqueue_hrtimer(timer, new_base);
409 unlock_hrtimer_base(timer, &flags);
415 * hrtimer_try_to_cancel - try to deactivate a timer
417 * @timer: hrtimer to stop
420 * 0 when the timer was not active
421 * 1 when the timer was active
422 * -1 when the timer is currently excuting the callback function and
425 int hrtimer_try_to_cancel(struct hrtimer *timer)
427 struct hrtimer_base *base;
431 base = lock_hrtimer_base(timer, &flags);
433 if (base->curr_timer != timer)
434 ret = remove_hrtimer(timer, base);
436 unlock_hrtimer_base(timer, &flags);
443 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
445 * @timer: the timer to be cancelled
448 * 0 when the timer was not active
449 * 1 when the timer was active
451 int hrtimer_cancel(struct hrtimer *timer)
454 int ret = hrtimer_try_to_cancel(timer);
462 * hrtimer_get_remaining - get remaining time for the timer
464 * @timer: the timer to read
466 ktime_t hrtimer_get_remaining(const struct hrtimer *timer)
468 struct hrtimer_base *base;
472 base = lock_hrtimer_base(timer, &flags);
473 rem = ktime_sub(timer->expires, timer->base->get_time());
474 unlock_hrtimer_base(timer, &flags);
480 * hrtimer_rebase - rebase an initialized hrtimer to a different base
482 * @timer: the timer to be rebased
483 * @clock_id: the clock to be used
485 void hrtimer_rebase(struct hrtimer *timer, const clockid_t clock_id)
487 struct hrtimer_base *bases;
489 bases = per_cpu(hrtimer_bases, raw_smp_processor_id());
490 timer->base = &bases[clock_id];
494 * hrtimer_init - initialize a timer to the given clock
496 * @timer: the timer to be initialized
497 * @clock_id: the clock to be used
499 void hrtimer_init(struct hrtimer *timer, const clockid_t clock_id)
501 memset(timer, 0, sizeof(struct hrtimer));
502 hrtimer_rebase(timer, clock_id);
506 * hrtimer_get_res - get the timer resolution for a clock
508 * @which_clock: which clock to query
509 * @tp: pointer to timespec variable to store the resolution
511 * Store the resolution of the clock selected by which_clock in the
512 * variable pointed to by tp.
514 int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp)
516 struct hrtimer_base *bases;
519 bases = per_cpu(hrtimer_bases, raw_smp_processor_id());
520 tp->tv_nsec = bases[which_clock].resolution;
526 * Expire the per base hrtimer-queue:
528 static inline void run_hrtimer_queue(struct hrtimer_base *base)
530 ktime_t now = base->get_time();
532 spin_lock_irq(&base->lock);
534 while (!list_empty(&base->pending)) {
535 struct hrtimer *timer;
540 timer = list_entry(base->pending.next, struct hrtimer, list);
541 if (now.tv64 <= timer->expires.tv64)
544 fn = timer->function;
546 set_curr_timer(base, timer);
547 __remove_hrtimer(timer, base);
548 spin_unlock_irq(&base->lock);
551 * fn == NULL is special case for the simplest timer
552 * variant - wake up process and do not restart:
555 wake_up_process(data);
556 restart = HRTIMER_NORESTART;
560 spin_lock_irq(&base->lock);
562 if (restart == HRTIMER_RESTART)
563 enqueue_hrtimer(timer, base);
565 timer->state = HRTIMER_EXPIRED;
567 set_curr_timer(base, NULL);
568 spin_unlock_irq(&base->lock);
572 * Called from timer softirq every jiffy, expire hrtimers:
574 void hrtimer_run_queues(void)
576 struct hrtimer_base *base = __get_cpu_var(hrtimer_bases);
579 for (i = 0; i < MAX_HRTIMER_BASES; i++)
580 run_hrtimer_queue(&base[i]);
584 * Sleep related functions:
588 * schedule_hrtimer - sleep until timeout
590 * @timer: hrtimer variable initialized with the correct clock base
591 * @mode: timeout value is abs/rel
593 * Make the current task sleep until @timeout is
596 * You can set the task state as follows -
598 * %TASK_UNINTERRUPTIBLE - at least @timeout is guaranteed to
599 * pass before the routine returns. The routine will return 0
601 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
602 * delivered to the current task. In this case the remaining time
605 * The current task state is guaranteed to be TASK_RUNNING when this
608 static ktime_t __sched
609 schedule_hrtimer(struct hrtimer *timer, const enum hrtimer_mode mode)
611 /* fn stays NULL, meaning single-shot wakeup: */
612 timer->data = current;
614 hrtimer_start(timer, timer->expires, mode);
617 hrtimer_cancel(timer);
619 /* Return the remaining time: */
620 if (timer->state != HRTIMER_EXPIRED)
621 return ktime_sub(timer->expires, timer->base->get_time());
623 return (ktime_t) {.tv64 = 0 };
626 static inline ktime_t __sched
627 schedule_hrtimer_interruptible(struct hrtimer *timer,
628 const enum hrtimer_mode mode)
630 set_current_state(TASK_INTERRUPTIBLE);
632 return schedule_hrtimer(timer, mode);
636 nanosleep_restart(struct restart_block *restart, clockid_t clockid)
638 struct timespec __user *rmtp, tu;
639 void *rfn_save = restart->fn;
640 struct hrtimer timer;
643 restart->fn = do_no_restart_syscall;
645 hrtimer_init(&timer, clockid);
647 timer.expires.tv64 = ((u64)restart->arg1 << 32) | (u64) restart->arg0;
649 rem = schedule_hrtimer_interruptible(&timer, HRTIMER_ABS);
654 rmtp = (struct timespec __user *) restart->arg2;
655 tu = ktime_to_timespec(rem);
656 if (rmtp && copy_to_user(rmtp, &tu, sizeof(tu)))
659 restart->fn = rfn_save;
661 /* The other values in restart are already filled in */
662 return -ERESTART_RESTARTBLOCK;
665 static long __sched nanosleep_restart_mono(struct restart_block *restart)
667 return nanosleep_restart(restart, CLOCK_MONOTONIC);
670 static long __sched nanosleep_restart_real(struct restart_block *restart)
672 return nanosleep_restart(restart, CLOCK_REALTIME);
675 long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
676 const enum hrtimer_mode mode, const clockid_t clockid)
678 struct restart_block *restart;
679 struct hrtimer timer;
683 hrtimer_init(&timer, clockid);
685 timer.expires = timespec_to_ktime(*rqtp);
687 rem = schedule_hrtimer_interruptible(&timer, mode);
691 /* Absolute timers do not update the rmtp value: */
692 if (mode == HRTIMER_ABS)
693 return -ERESTARTNOHAND;
695 tu = ktime_to_timespec(rem);
697 if (rmtp && copy_to_user(rmtp, &tu, sizeof(tu)))
700 restart = ¤t_thread_info()->restart_block;
701 restart->fn = (clockid == CLOCK_MONOTONIC) ?
702 nanosleep_restart_mono : nanosleep_restart_real;
703 restart->arg0 = timer.expires.tv64 & 0xFFFFFFFF;
704 restart->arg1 = timer.expires.tv64 >> 32;
705 restart->arg2 = (unsigned long) rmtp;
707 return -ERESTART_RESTARTBLOCK;
711 * Functions related to boot-time initialization:
713 static void __devinit init_hrtimers_cpu(int cpu)
715 struct hrtimer_base *base = per_cpu(hrtimer_bases, cpu);
718 for (i = 0; i < MAX_HRTIMER_BASES; i++) {
719 spin_lock_init(&base->lock);
720 INIT_LIST_HEAD(&base->pending);
725 #ifdef CONFIG_HOTPLUG_CPU
727 static void migrate_hrtimer_list(struct hrtimer_base *old_base,
728 struct hrtimer_base *new_base)
730 struct hrtimer *timer;
731 struct rb_node *node;
733 while ((node = rb_first(&old_base->active))) {
734 timer = rb_entry(node, struct hrtimer, node);
735 __remove_hrtimer(timer, old_base);
736 timer->base = new_base;
737 enqueue_hrtimer(timer, new_base);
741 static void migrate_hrtimers(int cpu)
743 struct hrtimer_base *old_base, *new_base;
746 BUG_ON(cpu_online(cpu));
747 old_base = per_cpu(hrtimer_bases, cpu);
748 new_base = get_cpu_var(hrtimer_bases);
752 for (i = 0; i < MAX_HRTIMER_BASES; i++) {
754 spin_lock(&new_base->lock);
755 spin_lock(&old_base->lock);
757 BUG_ON(old_base->curr_timer);
759 migrate_hrtimer_list(old_base, new_base);
761 spin_unlock(&old_base->lock);
762 spin_unlock(&new_base->lock);
768 put_cpu_var(hrtimer_bases);
770 #endif /* CONFIG_HOTPLUG_CPU */
772 static int __devinit hrtimer_cpu_notify(struct notifier_block *self,
773 unsigned long action, void *hcpu)
775 long cpu = (long)hcpu;
780 init_hrtimers_cpu(cpu);
783 #ifdef CONFIG_HOTPLUG_CPU
785 migrate_hrtimers(cpu);
796 static struct notifier_block __devinitdata hrtimers_nb = {
797 .notifier_call = hrtimer_cpu_notify,
800 void __init hrtimers_init(void)
802 hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
803 (void *)(long)smp_processor_id());
804 register_cpu_notifier(&hrtimers_nb);