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 * Returns the number of overruns.
278 hrtimer_forward(struct hrtimer *timer, ktime_t interval)
280 unsigned long orun = 1;
283 now = timer->base->get_time();
285 delta = ktime_sub(now, timer->expires);
290 if (interval.tv64 < timer->base->resolution.tv64)
291 interval.tv64 = timer->base->resolution.tv64;
293 if (unlikely(delta.tv64 >= interval.tv64)) {
294 nsec_t incr = ktime_to_ns(interval);
296 orun = ktime_divns(delta, incr);
297 timer->expires = ktime_add_ns(timer->expires, incr * orun);
298 if (timer->expires.tv64 > now.tv64)
301 * This (and the ktime_add() below) is the
302 * correction for exact:
306 timer->expires = ktime_add(timer->expires, interval);
312 * enqueue_hrtimer - internal function to (re)start a timer
314 * The timer is inserted in expiry order. Insertion into the
315 * red black tree is O(log(n)). Must hold the base lock.
317 static void enqueue_hrtimer(struct hrtimer *timer, struct hrtimer_base *base)
319 struct rb_node **link = &base->active.rb_node;
320 struct rb_node *parent = NULL;
321 struct hrtimer *entry;
324 * Find the right place in the rbtree:
328 entry = rb_entry(parent, struct hrtimer, node);
330 * We dont care about collisions. Nodes with
331 * the same expiry time stay together.
333 if (timer->expires.tv64 < entry->expires.tv64)
334 link = &(*link)->rb_left;
336 link = &(*link)->rb_right;
340 * Insert the timer to the rbtree and check whether it
341 * replaces the first pending timer
343 rb_link_node(&timer->node, parent, link);
344 rb_insert_color(&timer->node, &base->active);
346 timer->state = HRTIMER_PENDING;
348 if (!base->first || timer->expires.tv64 <
349 rb_entry(base->first, struct hrtimer, node)->expires.tv64)
350 base->first = &timer->node;
354 * __remove_hrtimer - internal function to remove a timer
356 * Caller must hold the base lock.
358 static void __remove_hrtimer(struct hrtimer *timer, struct hrtimer_base *base)
361 * Remove the timer from the rbtree and replace the
362 * first entry pointer if necessary.
364 if (base->first == &timer->node)
365 base->first = rb_next(&timer->node);
366 rb_erase(&timer->node, &base->active);
370 * remove hrtimer, called with base lock held
373 remove_hrtimer(struct hrtimer *timer, struct hrtimer_base *base)
375 if (hrtimer_active(timer)) {
376 __remove_hrtimer(timer, base);
377 timer->state = HRTIMER_INACTIVE;
384 * hrtimer_start - (re)start an relative timer on the current CPU
386 * @timer: the timer to be added
388 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
392 * 1 when the timer was active
395 hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode)
397 struct hrtimer_base *base, *new_base;
401 base = lock_hrtimer_base(timer, &flags);
403 /* Remove an active timer from the queue: */
404 ret = remove_hrtimer(timer, base);
406 /* Switch the timer base, if necessary: */
407 new_base = switch_hrtimer_base(timer, base);
409 if (mode == HRTIMER_REL)
410 tim = ktime_add(tim, new_base->get_time());
411 timer->expires = tim;
413 enqueue_hrtimer(timer, new_base);
415 unlock_hrtimer_base(timer, &flags);
421 * hrtimer_try_to_cancel - try to deactivate a timer
423 * @timer: hrtimer to stop
426 * 0 when the timer was not active
427 * 1 when the timer was active
428 * -1 when the timer is currently excuting the callback function and
431 int hrtimer_try_to_cancel(struct hrtimer *timer)
433 struct hrtimer_base *base;
437 base = lock_hrtimer_base(timer, &flags);
439 if (base->curr_timer != timer)
440 ret = remove_hrtimer(timer, base);
442 unlock_hrtimer_base(timer, &flags);
449 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
451 * @timer: the timer to be cancelled
454 * 0 when the timer was not active
455 * 1 when the timer was active
457 int hrtimer_cancel(struct hrtimer *timer)
460 int ret = hrtimer_try_to_cancel(timer);
468 * hrtimer_get_remaining - get remaining time for the timer
470 * @timer: the timer to read
472 ktime_t hrtimer_get_remaining(const struct hrtimer *timer)
474 struct hrtimer_base *base;
478 base = lock_hrtimer_base(timer, &flags);
479 rem = ktime_sub(timer->expires, timer->base->get_time());
480 unlock_hrtimer_base(timer, &flags);
486 * hrtimer_rebase - rebase an initialized hrtimer to a different base
488 * @timer: the timer to be rebased
489 * @clock_id: the clock to be used
491 void hrtimer_rebase(struct hrtimer *timer, const clockid_t clock_id)
493 struct hrtimer_base *bases;
495 bases = per_cpu(hrtimer_bases, raw_smp_processor_id());
496 timer->base = &bases[clock_id];
500 * hrtimer_init - initialize a timer to the given clock
502 * @timer: the timer to be initialized
503 * @clock_id: the clock to be used
505 void hrtimer_init(struct hrtimer *timer, const clockid_t clock_id)
507 memset(timer, 0, sizeof(struct hrtimer));
508 hrtimer_rebase(timer, clock_id);
512 * hrtimer_get_res - get the timer resolution for a clock
514 * @which_clock: which clock to query
515 * @tp: pointer to timespec variable to store the resolution
517 * Store the resolution of the clock selected by which_clock in the
518 * variable pointed to by tp.
520 int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp)
522 struct hrtimer_base *bases;
524 bases = per_cpu(hrtimer_bases, raw_smp_processor_id());
525 *tp = ktime_to_timespec(bases[which_clock].resolution);
531 * Expire the per base hrtimer-queue:
533 static inline void run_hrtimer_queue(struct hrtimer_base *base)
535 ktime_t now = base->get_time();
536 struct rb_node *node;
538 spin_lock_irq(&base->lock);
540 while ((node = base->first)) {
541 struct hrtimer *timer;
546 timer = rb_entry(node, struct hrtimer, node);
547 if (now.tv64 <= timer->expires.tv64)
550 fn = timer->function;
552 set_curr_timer(base, timer);
553 __remove_hrtimer(timer, base);
554 spin_unlock_irq(&base->lock);
557 * fn == NULL is special case for the simplest timer
558 * variant - wake up process and do not restart:
561 wake_up_process(data);
562 restart = HRTIMER_NORESTART;
566 spin_lock_irq(&base->lock);
568 if (restart == HRTIMER_RESTART)
569 enqueue_hrtimer(timer, base);
571 timer->state = HRTIMER_EXPIRED;
573 set_curr_timer(base, NULL);
574 spin_unlock_irq(&base->lock);
578 * Called from timer softirq every jiffy, expire hrtimers:
580 void hrtimer_run_queues(void)
582 struct hrtimer_base *base = __get_cpu_var(hrtimer_bases);
585 for (i = 0; i < MAX_HRTIMER_BASES; i++)
586 run_hrtimer_queue(&base[i]);
590 * Sleep related functions:
594 * schedule_hrtimer - sleep until timeout
596 * @timer: hrtimer variable initialized with the correct clock base
597 * @mode: timeout value is abs/rel
599 * Make the current task sleep until @timeout is
602 * You can set the task state as follows -
604 * %TASK_UNINTERRUPTIBLE - at least @timeout is guaranteed to
605 * pass before the routine returns. The routine will return 0
607 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
608 * delivered to the current task. In this case the remaining time
611 * The current task state is guaranteed to be TASK_RUNNING when this
614 static ktime_t __sched
615 schedule_hrtimer(struct hrtimer *timer, const enum hrtimer_mode mode)
617 /* fn stays NULL, meaning single-shot wakeup: */
618 timer->data = current;
620 hrtimer_start(timer, timer->expires, mode);
623 hrtimer_cancel(timer);
625 /* Return the remaining time: */
626 if (timer->state != HRTIMER_EXPIRED)
627 return ktime_sub(timer->expires, timer->base->get_time());
629 return (ktime_t) {.tv64 = 0 };
632 static inline ktime_t __sched
633 schedule_hrtimer_interruptible(struct hrtimer *timer,
634 const enum hrtimer_mode mode)
636 set_current_state(TASK_INTERRUPTIBLE);
638 return schedule_hrtimer(timer, mode);
642 nanosleep_restart(struct restart_block *restart, clockid_t clockid)
644 struct timespec __user *rmtp;
646 void *rfn_save = restart->fn;
647 struct hrtimer timer;
650 restart->fn = do_no_restart_syscall;
652 hrtimer_init(&timer, clockid);
654 timer.expires.tv64 = ((u64)restart->arg1 << 32) | (u64) restart->arg0;
656 rem = schedule_hrtimer_interruptible(&timer, HRTIMER_ABS);
661 rmtp = (struct timespec __user *) restart->arg2;
662 tu = ktime_to_timespec(rem);
663 if (rmtp && copy_to_user(rmtp, &tu, sizeof(tu)))
666 restart->fn = rfn_save;
668 /* The other values in restart are already filled in */
669 return -ERESTART_RESTARTBLOCK;
672 static long __sched nanosleep_restart_mono(struct restart_block *restart)
674 return nanosleep_restart(restart, CLOCK_MONOTONIC);
677 static long __sched nanosleep_restart_real(struct restart_block *restart)
679 return nanosleep_restart(restart, CLOCK_REALTIME);
682 long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
683 const enum hrtimer_mode mode, const clockid_t clockid)
685 struct restart_block *restart;
686 struct hrtimer timer;
690 hrtimer_init(&timer, clockid);
692 timer.expires = timespec_to_ktime(*rqtp);
694 rem = schedule_hrtimer_interruptible(&timer, mode);
698 /* Absolute timers do not update the rmtp value: */
699 if (mode == HRTIMER_ABS)
700 return -ERESTARTNOHAND;
702 tu = ktime_to_timespec(rem);
704 if (rmtp && copy_to_user(rmtp, &tu, sizeof(tu)))
707 restart = ¤t_thread_info()->restart_block;
708 restart->fn = (clockid == CLOCK_MONOTONIC) ?
709 nanosleep_restart_mono : nanosleep_restart_real;
710 restart->arg0 = timer.expires.tv64 & 0xFFFFFFFF;
711 restart->arg1 = timer.expires.tv64 >> 32;
712 restart->arg2 = (unsigned long) rmtp;
714 return -ERESTART_RESTARTBLOCK;
718 sys_nanosleep(struct timespec __user *rqtp, struct timespec __user *rmtp)
722 if (copy_from_user(&tu, rqtp, sizeof(tu)))
725 if (!timespec_valid(&tu))
728 return hrtimer_nanosleep(&tu, rmtp, HRTIMER_REL, CLOCK_MONOTONIC);
732 * Functions related to boot-time initialization:
734 static void __devinit init_hrtimers_cpu(int cpu)
736 struct hrtimer_base *base = per_cpu(hrtimer_bases, cpu);
739 for (i = 0; i < MAX_HRTIMER_BASES; i++) {
740 spin_lock_init(&base->lock);
745 #ifdef CONFIG_HOTPLUG_CPU
747 static void migrate_hrtimer_list(struct hrtimer_base *old_base,
748 struct hrtimer_base *new_base)
750 struct hrtimer *timer;
751 struct rb_node *node;
753 while ((node = rb_first(&old_base->active))) {
754 timer = rb_entry(node, struct hrtimer, node);
755 __remove_hrtimer(timer, old_base);
756 timer->base = new_base;
757 enqueue_hrtimer(timer, new_base);
761 static void migrate_hrtimers(int cpu)
763 struct hrtimer_base *old_base, *new_base;
766 BUG_ON(cpu_online(cpu));
767 old_base = per_cpu(hrtimer_bases, cpu);
768 new_base = get_cpu_var(hrtimer_bases);
772 for (i = 0; i < MAX_HRTIMER_BASES; i++) {
774 spin_lock(&new_base->lock);
775 spin_lock(&old_base->lock);
777 BUG_ON(old_base->curr_timer);
779 migrate_hrtimer_list(old_base, new_base);
781 spin_unlock(&old_base->lock);
782 spin_unlock(&new_base->lock);
788 put_cpu_var(hrtimer_bases);
790 #endif /* CONFIG_HOTPLUG_CPU */
792 static int __devinit hrtimer_cpu_notify(struct notifier_block *self,
793 unsigned long action, void *hcpu)
795 long cpu = (long)hcpu;
800 init_hrtimers_cpu(cpu);
803 #ifdef CONFIG_HOTPLUG_CPU
805 migrate_hrtimers(cpu);
816 static struct notifier_block __devinitdata hrtimers_nb = {
817 .notifier_call = hrtimer_cpu_notify,
820 void __init hrtimers_init(void)
822 hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
823 (void *)(long)smp_processor_id());
824 register_cpu_notifier(&hrtimers_nb);