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 * Help, testing, suggestions, bugfixes, improvements were
27 * George Anzinger, Andrew Morton, Steven Rostedt, Roman Zippel
30 * For licencing details see kernel-base/COPYING
33 #include <linux/cpu.h>
34 #include <linux/module.h>
35 #include <linux/percpu.h>
36 #include <linux/hrtimer.h>
37 #include <linux/notifier.h>
38 #include <linux/syscalls.h>
39 #include <linux/interrupt.h>
41 #include <asm/uaccess.h>
44 * ktime_get - get the monotonic time in ktime_t format
46 * returns the time in ktime_t format
48 static ktime_t
ktime_get(void)
54 return timespec_to_ktime(now
);
58 * ktime_get_real - get the real (wall-) time in ktime_t format
60 * returns the time in ktime_t format
62 static ktime_t
ktime_get_real(void)
68 return timespec_to_ktime(now
);
71 EXPORT_SYMBOL_GPL(ktime_get_real
);
76 * Note: If we want to add new timer bases, we have to skip the two
77 * clock ids captured by the cpu-timers. We do this by holding empty
78 * entries rather than doing math adjustment of the clock ids.
79 * This ensures that we capture erroneous accesses to these clock ids
80 * rather than moving them into the range of valid clock id's.
83 #define MAX_HRTIMER_BASES 2
85 static DEFINE_PER_CPU(struct hrtimer_base
, hrtimer_bases
[MAX_HRTIMER_BASES
]) =
88 .index
= CLOCK_REALTIME
,
89 .get_time
= &ktime_get_real
,
90 .resolution
= KTIME_REALTIME_RES
,
93 .index
= CLOCK_MONOTONIC
,
94 .get_time
= &ktime_get
,
95 .resolution
= KTIME_MONOTONIC_RES
,
100 * ktime_get_ts - get the monotonic clock in timespec format
101 * @ts: pointer to timespec variable
103 * The function calculates the monotonic clock from the realtime
104 * clock and the wall_to_monotonic offset and stores the result
105 * in normalized timespec format in the variable pointed to by @ts.
107 void ktime_get_ts(struct timespec
*ts
)
109 struct timespec tomono
;
113 seq
= read_seqbegin(&xtime_lock
);
115 tomono
= wall_to_monotonic
;
117 } while (read_seqretry(&xtime_lock
, seq
));
119 set_normalized_timespec(ts
, ts
->tv_sec
+ tomono
.tv_sec
,
120 ts
->tv_nsec
+ tomono
.tv_nsec
);
122 EXPORT_SYMBOL_GPL(ktime_get_ts
);
125 * Get the coarse grained time at the softirq based on xtime and
128 static void hrtimer_get_softirq_time(struct hrtimer_base
*base
)
130 ktime_t xtim
, tomono
;
135 seq
= read_seqbegin(&xtime_lock
);
137 getnstimeofday(&xts
);
141 } while (read_seqretry(&xtime_lock
, seq
));
143 xtim
= timespec_to_ktime(xts
);
144 tomono
= timespec_to_ktime(wall_to_monotonic
);
145 base
[CLOCK_REALTIME
].softirq_time
= xtim
;
146 base
[CLOCK_MONOTONIC
].softirq_time
= ktime_add(xtim
, tomono
);
150 * Functions and macros which are different for UP/SMP systems are kept in a
155 #define set_curr_timer(b, t) do { (b)->curr_timer = (t); } while (0)
158 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
159 * means that all timers which are tied to this base via timer->base are
160 * locked, and the base itself is locked too.
162 * So __run_timers/migrate_timers can safely modify all timers which could
163 * be found on the lists/queues.
165 * When the timer's base is locked, and the timer removed from list, it is
166 * possible to set timer->base = NULL and drop the lock: the timer remains
169 static struct hrtimer_base
*lock_hrtimer_base(const struct hrtimer
*timer
,
170 unsigned long *flags
)
172 struct hrtimer_base
*base
;
176 if (likely(base
!= NULL
)) {
177 spin_lock_irqsave(&base
->lock
, *flags
);
178 if (likely(base
== timer
->base
))
180 /* The timer has migrated to another CPU: */
181 spin_unlock_irqrestore(&base
->lock
, *flags
);
188 * Switch the timer base to the current CPU when possible.
190 static inline struct hrtimer_base
*
191 switch_hrtimer_base(struct hrtimer
*timer
, struct hrtimer_base
*base
)
193 struct hrtimer_base
*new_base
;
195 new_base
= &__get_cpu_var(hrtimer_bases
)[base
->index
];
197 if (base
!= new_base
) {
199 * We are trying to schedule the timer on the local CPU.
200 * However we can't change timer's base while it is running,
201 * so we keep it on the same CPU. No hassle vs. reprogramming
202 * the event source in the high resolution case. The softirq
203 * code will take care of this when the timer function has
204 * completed. There is no conflict as we hold the lock until
205 * the timer is enqueued.
207 if (unlikely(base
->curr_timer
== timer
))
210 /* See the comment in lock_timer_base() */
212 spin_unlock(&base
->lock
);
213 spin_lock(&new_base
->lock
);
214 timer
->base
= new_base
;
219 #else /* CONFIG_SMP */
221 #define set_curr_timer(b, t) do { } while (0)
223 static inline struct hrtimer_base
*
224 lock_hrtimer_base(const struct hrtimer
*timer
, unsigned long *flags
)
226 struct hrtimer_base
*base
= timer
->base
;
228 spin_lock_irqsave(&base
->lock
, *flags
);
233 #define switch_hrtimer_base(t, b) (b)
235 #endif /* !CONFIG_SMP */
238 * Functions for the union type storage format of ktime_t which are
239 * too large for inlining:
241 #if BITS_PER_LONG < 64
242 # ifndef CONFIG_KTIME_SCALAR
244 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
246 * @nsec: the scalar nsec value to add
248 * Returns the sum of kt and nsec in ktime_t format
250 ktime_t
ktime_add_ns(const ktime_t kt
, u64 nsec
)
254 if (likely(nsec
< NSEC_PER_SEC
)) {
257 unsigned long rem
= do_div(nsec
, NSEC_PER_SEC
);
259 tmp
= ktime_set((long)nsec
, rem
);
262 return ktime_add(kt
, tmp
);
265 #else /* CONFIG_KTIME_SCALAR */
267 # endif /* !CONFIG_KTIME_SCALAR */
270 * Divide a ktime value by a nanosecond value
272 static unsigned long ktime_divns(const ktime_t kt
, s64 div
)
277 dclc
= dns
= ktime_to_ns(kt
);
279 /* Make sure the divisor is less than 2^32: */
285 do_div(dclc
, (unsigned long) div
);
287 return (unsigned long) dclc
;
290 #else /* BITS_PER_LONG < 64 */
291 # define ktime_divns(kt, div) (unsigned long)((kt).tv64 / (div))
292 #endif /* BITS_PER_LONG >= 64 */
295 * Counterpart to lock_timer_base above:
298 void unlock_hrtimer_base(const struct hrtimer
*timer
, unsigned long *flags
)
300 spin_unlock_irqrestore(&timer
->base
->lock
, *flags
);
304 * hrtimer_forward - forward the timer expiry
305 * @timer: hrtimer to forward
306 * @now: forward past this time
307 * @interval: the interval to forward
309 * Forward the timer expiry so it will expire in the future.
310 * Returns the number of overruns.
313 hrtimer_forward(struct hrtimer
*timer
, ktime_t now
, ktime_t interval
)
315 unsigned long orun
= 1;
318 delta
= ktime_sub(now
, timer
->expires
);
323 if (interval
.tv64
< timer
->base
->resolution
.tv64
)
324 interval
.tv64
= timer
->base
->resolution
.tv64
;
326 if (unlikely(delta
.tv64
>= interval
.tv64
)) {
327 s64 incr
= ktime_to_ns(interval
);
329 orun
= ktime_divns(delta
, incr
);
330 timer
->expires
= ktime_add_ns(timer
->expires
, incr
* orun
);
331 if (timer
->expires
.tv64
> now
.tv64
)
334 * This (and the ktime_add() below) is the
335 * correction for exact:
339 timer
->expires
= ktime_add(timer
->expires
, interval
);
345 * enqueue_hrtimer - internal function to (re)start a timer
347 * The timer is inserted in expiry order. Insertion into the
348 * red black tree is O(log(n)). Must hold the base lock.
350 static void enqueue_hrtimer(struct hrtimer
*timer
, struct hrtimer_base
*base
)
352 struct rb_node
**link
= &base
->active
.rb_node
;
353 struct rb_node
*parent
= NULL
;
354 struct hrtimer
*entry
;
357 * Find the right place in the rbtree:
361 entry
= rb_entry(parent
, struct hrtimer
, node
);
363 * We dont care about collisions. Nodes with
364 * the same expiry time stay together.
366 if (timer
->expires
.tv64
< entry
->expires
.tv64
)
367 link
= &(*link
)->rb_left
;
369 link
= &(*link
)->rb_right
;
373 * Insert the timer to the rbtree and check whether it
374 * replaces the first pending timer
376 rb_link_node(&timer
->node
, parent
, link
);
377 rb_insert_color(&timer
->node
, &base
->active
);
379 if (!base
->first
|| timer
->expires
.tv64
<
380 rb_entry(base
->first
, struct hrtimer
, node
)->expires
.tv64
)
381 base
->first
= &timer
->node
;
385 * __remove_hrtimer - internal function to remove a timer
387 * Caller must hold the base lock.
389 static void __remove_hrtimer(struct hrtimer
*timer
, struct hrtimer_base
*base
)
392 * Remove the timer from the rbtree and replace the
393 * first entry pointer if necessary.
395 if (base
->first
== &timer
->node
)
396 base
->first
= rb_next(&timer
->node
);
397 rb_erase(&timer
->node
, &base
->active
);
398 rb_set_parent(&timer
->node
, &timer
->node
);
402 * remove hrtimer, called with base lock held
405 remove_hrtimer(struct hrtimer
*timer
, struct hrtimer_base
*base
)
407 if (hrtimer_active(timer
)) {
408 __remove_hrtimer(timer
, base
);
415 * hrtimer_start - (re)start an relative timer on the current CPU
416 * @timer: the timer to be added
418 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
422 * 1 when the timer was active
425 hrtimer_start(struct hrtimer
*timer
, ktime_t tim
, const enum hrtimer_mode mode
)
427 struct hrtimer_base
*base
, *new_base
;
431 base
= lock_hrtimer_base(timer
, &flags
);
433 /* Remove an active timer from the queue: */
434 ret
= remove_hrtimer(timer
, base
);
436 /* Switch the timer base, if necessary: */
437 new_base
= switch_hrtimer_base(timer
, base
);
439 if (mode
== HRTIMER_REL
) {
440 tim
= ktime_add(tim
, new_base
->get_time());
442 * CONFIG_TIME_LOW_RES is a temporary way for architectures
443 * to signal that they simply return xtime in
444 * do_gettimeoffset(). In this case we want to round up by
445 * resolution when starting a relative timer, to avoid short
446 * timeouts. This will go away with the GTOD framework.
448 #ifdef CONFIG_TIME_LOW_RES
449 tim
= ktime_add(tim
, base
->resolution
);
452 timer
->expires
= tim
;
454 enqueue_hrtimer(timer
, new_base
);
456 unlock_hrtimer_base(timer
, &flags
);
460 EXPORT_SYMBOL_GPL(hrtimer_start
);
463 * hrtimer_try_to_cancel - try to deactivate a timer
464 * @timer: hrtimer to stop
467 * 0 when the timer was not active
468 * 1 when the timer was active
469 * -1 when the timer is currently excuting the callback function and
472 int hrtimer_try_to_cancel(struct hrtimer
*timer
)
474 struct hrtimer_base
*base
;
478 base
= lock_hrtimer_base(timer
, &flags
);
480 if (base
->curr_timer
!= timer
)
481 ret
= remove_hrtimer(timer
, base
);
483 unlock_hrtimer_base(timer
, &flags
);
488 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel
);
491 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
492 * @timer: the timer to be cancelled
495 * 0 when the timer was not active
496 * 1 when the timer was active
498 int hrtimer_cancel(struct hrtimer
*timer
)
501 int ret
= hrtimer_try_to_cancel(timer
);
508 EXPORT_SYMBOL_GPL(hrtimer_cancel
);
511 * hrtimer_get_remaining - get remaining time for the timer
512 * @timer: the timer to read
514 ktime_t
hrtimer_get_remaining(const struct hrtimer
*timer
)
516 struct hrtimer_base
*base
;
520 base
= lock_hrtimer_base(timer
, &flags
);
521 rem
= ktime_sub(timer
->expires
, timer
->base
->get_time());
522 unlock_hrtimer_base(timer
, &flags
);
526 EXPORT_SYMBOL_GPL(hrtimer_get_remaining
);
528 #ifdef CONFIG_NO_IDLE_HZ
530 * hrtimer_get_next_event - get the time until next expiry event
532 * Returns the delta to the next expiry event or KTIME_MAX if no timer
535 ktime_t
hrtimer_get_next_event(void)
537 struct hrtimer_base
*base
= __get_cpu_var(hrtimer_bases
);
538 ktime_t delta
, mindelta
= { .tv64
= KTIME_MAX
};
542 for (i
= 0; i
< MAX_HRTIMER_BASES
; i
++, base
++) {
543 struct hrtimer
*timer
;
545 spin_lock_irqsave(&base
->lock
, flags
);
547 spin_unlock_irqrestore(&base
->lock
, flags
);
550 timer
= rb_entry(base
->first
, struct hrtimer
, node
);
551 delta
.tv64
= timer
->expires
.tv64
;
552 spin_unlock_irqrestore(&base
->lock
, flags
);
553 delta
= ktime_sub(delta
, base
->get_time());
554 if (delta
.tv64
< mindelta
.tv64
)
555 mindelta
.tv64
= delta
.tv64
;
557 if (mindelta
.tv64
< 0)
564 * hrtimer_init - initialize a timer to the given clock
565 * @timer: the timer to be initialized
566 * @clock_id: the clock to be used
567 * @mode: timer mode abs/rel
569 void hrtimer_init(struct hrtimer
*timer
, clockid_t clock_id
,
570 enum hrtimer_mode mode
)
572 struct hrtimer_base
*bases
;
574 memset(timer
, 0, sizeof(struct hrtimer
));
576 bases
= __raw_get_cpu_var(hrtimer_bases
);
578 if (clock_id
== CLOCK_REALTIME
&& mode
!= HRTIMER_ABS
)
579 clock_id
= CLOCK_MONOTONIC
;
581 timer
->base
= &bases
[clock_id
];
582 rb_set_parent(&timer
->node
, &timer
->node
);
584 EXPORT_SYMBOL_GPL(hrtimer_init
);
587 * hrtimer_get_res - get the timer resolution for a clock
588 * @which_clock: which clock to query
589 * @tp: pointer to timespec variable to store the resolution
591 * Store the resolution of the clock selected by @which_clock in the
592 * variable pointed to by @tp.
594 int hrtimer_get_res(const clockid_t which_clock
, struct timespec
*tp
)
596 struct hrtimer_base
*bases
;
598 bases
= __raw_get_cpu_var(hrtimer_bases
);
599 *tp
= ktime_to_timespec(bases
[which_clock
].resolution
);
603 EXPORT_SYMBOL_GPL(hrtimer_get_res
);
606 * Expire the per base hrtimer-queue:
608 static inline void run_hrtimer_queue(struct hrtimer_base
*base
)
610 struct rb_node
*node
;
615 if (base
->get_softirq_time
)
616 base
->softirq_time
= base
->get_softirq_time();
618 spin_lock_irq(&base
->lock
);
620 while ((node
= base
->first
)) {
621 struct hrtimer
*timer
;
622 int (*fn
)(struct hrtimer
*);
625 timer
= rb_entry(node
, struct hrtimer
, node
);
626 if (base
->softirq_time
.tv64
<= timer
->expires
.tv64
)
629 fn
= timer
->function
;
630 set_curr_timer(base
, timer
);
631 __remove_hrtimer(timer
, base
);
632 spin_unlock_irq(&base
->lock
);
636 spin_lock_irq(&base
->lock
);
638 if (restart
!= HRTIMER_NORESTART
) {
639 BUG_ON(hrtimer_active(timer
));
640 enqueue_hrtimer(timer
, base
);
643 set_curr_timer(base
, NULL
);
644 spin_unlock_irq(&base
->lock
);
648 * Called from timer softirq every jiffy, expire hrtimers:
650 void hrtimer_run_queues(void)
652 struct hrtimer_base
*base
= __get_cpu_var(hrtimer_bases
);
655 hrtimer_get_softirq_time(base
);
657 for (i
= 0; i
< MAX_HRTIMER_BASES
; i
++)
658 run_hrtimer_queue(&base
[i
]);
662 * Sleep related functions:
664 static int hrtimer_wakeup(struct hrtimer
*timer
)
666 struct hrtimer_sleeper
*t
=
667 container_of(timer
, struct hrtimer_sleeper
, timer
);
668 struct task_struct
*task
= t
->task
;
672 wake_up_process(task
);
674 return HRTIMER_NORESTART
;
677 void hrtimer_init_sleeper(struct hrtimer_sleeper
*sl
, struct task_struct
*task
)
679 sl
->timer
.function
= hrtimer_wakeup
;
683 static int __sched
do_nanosleep(struct hrtimer_sleeper
*t
, enum hrtimer_mode mode
)
685 hrtimer_init_sleeper(t
, current
);
688 set_current_state(TASK_INTERRUPTIBLE
);
689 hrtimer_start(&t
->timer
, t
->timer
.expires
, mode
);
693 hrtimer_cancel(&t
->timer
);
696 } while (t
->task
&& !signal_pending(current
));
698 return t
->task
== NULL
;
701 long __sched
hrtimer_nanosleep_restart(struct restart_block
*restart
)
703 struct hrtimer_sleeper t
;
704 struct timespec __user
*rmtp
;
708 restart
->fn
= do_no_restart_syscall
;
710 hrtimer_init(&t
.timer
, restart
->arg0
, HRTIMER_ABS
);
711 t
.timer
.expires
.tv64
= ((u64
)restart
->arg3
<< 32) | (u64
) restart
->arg2
;
713 if (do_nanosleep(&t
, HRTIMER_ABS
))
716 rmtp
= (struct timespec __user
*) restart
->arg1
;
718 time
= ktime_sub(t
.timer
.expires
, t
.timer
.base
->get_time());
721 tu
= ktime_to_timespec(time
);
722 if (copy_to_user(rmtp
, &tu
, sizeof(tu
)))
726 restart
->fn
= hrtimer_nanosleep_restart
;
728 /* The other values in restart are already filled in */
729 return -ERESTART_RESTARTBLOCK
;
732 long hrtimer_nanosleep(struct timespec
*rqtp
, struct timespec __user
*rmtp
,
733 const enum hrtimer_mode mode
, const clockid_t clockid
)
735 struct restart_block
*restart
;
736 struct hrtimer_sleeper t
;
740 hrtimer_init(&t
.timer
, clockid
, mode
);
741 t
.timer
.expires
= timespec_to_ktime(*rqtp
);
742 if (do_nanosleep(&t
, mode
))
745 /* Absolute timers do not update the rmtp value and restart: */
746 if (mode
== HRTIMER_ABS
)
747 return -ERESTARTNOHAND
;
750 rem
= ktime_sub(t
.timer
.expires
, t
.timer
.base
->get_time());
753 tu
= ktime_to_timespec(rem
);
754 if (copy_to_user(rmtp
, &tu
, sizeof(tu
)))
758 restart
= ¤t_thread_info()->restart_block
;
759 restart
->fn
= hrtimer_nanosleep_restart
;
760 restart
->arg0
= (unsigned long) t
.timer
.base
->index
;
761 restart
->arg1
= (unsigned long) rmtp
;
762 restart
->arg2
= t
.timer
.expires
.tv64
& 0xFFFFFFFF;
763 restart
->arg3
= t
.timer
.expires
.tv64
>> 32;
765 return -ERESTART_RESTARTBLOCK
;
769 sys_nanosleep(struct timespec __user
*rqtp
, struct timespec __user
*rmtp
)
773 if (copy_from_user(&tu
, rqtp
, sizeof(tu
)))
776 if (!timespec_valid(&tu
))
779 return hrtimer_nanosleep(&tu
, rmtp
, HRTIMER_REL
, CLOCK_MONOTONIC
);
783 * Functions related to boot-time initialization:
785 static void __devinit
init_hrtimers_cpu(int cpu
)
787 struct hrtimer_base
*base
= per_cpu(hrtimer_bases
, cpu
);
790 for (i
= 0; i
< MAX_HRTIMER_BASES
; i
++, base
++) {
791 spin_lock_init(&base
->lock
);
792 lockdep_set_class(&base
->lock
, &base
->lock_key
);
796 #ifdef CONFIG_HOTPLUG_CPU
798 static void migrate_hrtimer_list(struct hrtimer_base
*old_base
,
799 struct hrtimer_base
*new_base
)
801 struct hrtimer
*timer
;
802 struct rb_node
*node
;
804 while ((node
= rb_first(&old_base
->active
))) {
805 timer
= rb_entry(node
, struct hrtimer
, node
);
806 __remove_hrtimer(timer
, old_base
);
807 timer
->base
= new_base
;
808 enqueue_hrtimer(timer
, new_base
);
812 static void migrate_hrtimers(int cpu
)
814 struct hrtimer_base
*old_base
, *new_base
;
817 BUG_ON(cpu_online(cpu
));
818 old_base
= per_cpu(hrtimer_bases
, cpu
);
819 new_base
= get_cpu_var(hrtimer_bases
);
823 for (i
= 0; i
< MAX_HRTIMER_BASES
; i
++) {
825 spin_lock(&new_base
->lock
);
826 spin_lock(&old_base
->lock
);
828 BUG_ON(old_base
->curr_timer
);
830 migrate_hrtimer_list(old_base
, new_base
);
832 spin_unlock(&old_base
->lock
);
833 spin_unlock(&new_base
->lock
);
839 put_cpu_var(hrtimer_bases
);
841 #endif /* CONFIG_HOTPLUG_CPU */
843 static int __cpuinit
hrtimer_cpu_notify(struct notifier_block
*self
,
844 unsigned long action
, void *hcpu
)
846 long cpu
= (long)hcpu
;
851 init_hrtimers_cpu(cpu
);
854 #ifdef CONFIG_HOTPLUG_CPU
856 migrate_hrtimers(cpu
);
867 static struct notifier_block __cpuinitdata hrtimers_nb
= {
868 .notifier_call
= hrtimer_cpu_notify
,
871 void __init
hrtimers_init(void)
873 hrtimer_cpu_notify(&hrtimers_nb
, (unsigned long)CPU_UP_PREPARE
,
874 (void *)(long)smp_processor_id());
875 register_cpu_notifier(&hrtimers_nb
);