2 * linux/kernel/hrtimer.c
4 * Copyright(C) 2005-2006, Thomas Gleixner <tglx@linutronix.de>
5 * Copyright(C) 2005-2007, Red Hat, Inc., Ingo Molnar
6 * Copyright(C) 2006-2007 Timesys Corp., Thomas Gleixner
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/irq.h>
36 #include <linux/module.h>
37 #include <linux/percpu.h>
38 #include <linux/hrtimer.h>
39 #include <linux/notifier.h>
40 #include <linux/syscalls.h>
41 #include <linux/kallsyms.h>
42 #include <linux/interrupt.h>
43 #include <linux/tick.h>
44 #include <linux/seq_file.h>
45 #include <linux/err.h>
47 #include <asm/uaccess.h>
50 * ktime_get - get the monotonic time in ktime_t format
52 * returns the time in ktime_t format
54 ktime_t
ktime_get(void)
60 return timespec_to_ktime(now
);
64 * ktime_get_real - get the real (wall-) time in ktime_t format
66 * returns the time in ktime_t format
68 ktime_t
ktime_get_real(void)
74 return timespec_to_ktime(now
);
77 EXPORT_SYMBOL_GPL(ktime_get_real
);
82 * Note: If we want to add new timer bases, we have to skip the two
83 * clock ids captured by the cpu-timers. We do this by holding empty
84 * entries rather than doing math adjustment of the clock ids.
85 * This ensures that we capture erroneous accesses to these clock ids
86 * rather than moving them into the range of valid clock id's.
88 DEFINE_PER_CPU(struct hrtimer_cpu_base
, hrtimer_bases
) =
94 .index
= CLOCK_REALTIME
,
95 .get_time
= &ktime_get_real
,
96 .resolution
= KTIME_LOW_RES
,
99 .index
= CLOCK_MONOTONIC
,
100 .get_time
= &ktime_get
,
101 .resolution
= KTIME_LOW_RES
,
107 * ktime_get_ts - get the monotonic clock in timespec format
108 * @ts: pointer to timespec variable
110 * The function calculates the monotonic clock from the realtime
111 * clock and the wall_to_monotonic offset and stores the result
112 * in normalized timespec format in the variable pointed to by @ts.
114 void ktime_get_ts(struct timespec
*ts
)
116 struct timespec tomono
;
120 seq
= read_seqbegin(&xtime_lock
);
122 tomono
= wall_to_monotonic
;
124 } while (read_seqretry(&xtime_lock
, seq
));
126 set_normalized_timespec(ts
, ts
->tv_sec
+ tomono
.tv_sec
,
127 ts
->tv_nsec
+ tomono
.tv_nsec
);
129 EXPORT_SYMBOL_GPL(ktime_get_ts
);
132 * Get the coarse grained time at the softirq based on xtime and
135 static void hrtimer_get_softirq_time(struct hrtimer_cpu_base
*base
)
137 ktime_t xtim
, tomono
;
142 seq
= read_seqbegin(&xtime_lock
);
144 getnstimeofday(&xts
);
148 } while (read_seqretry(&xtime_lock
, seq
));
150 xtim
= timespec_to_ktime(xts
);
151 tomono
= timespec_to_ktime(wall_to_monotonic
);
152 base
->clock_base
[CLOCK_REALTIME
].softirq_time
= xtim
;
153 base
->clock_base
[CLOCK_MONOTONIC
].softirq_time
=
154 ktime_add(xtim
, tomono
);
158 * Helper function to check, whether the timer is running the callback
161 static inline int hrtimer_callback_running(struct hrtimer
*timer
)
163 return timer
->state
& HRTIMER_STATE_CALLBACK
;
167 * Functions and macros which are different for UP/SMP systems are kept in a
173 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
174 * means that all timers which are tied to this base via timer->base are
175 * locked, and the base itself is locked too.
177 * So __run_timers/migrate_timers can safely modify all timers which could
178 * be found on the lists/queues.
180 * When the timer's base is locked, and the timer removed from list, it is
181 * possible to set timer->base = NULL and drop the lock: the timer remains
185 struct hrtimer_clock_base
*lock_hrtimer_base(const struct hrtimer
*timer
,
186 unsigned long *flags
)
188 struct hrtimer_clock_base
*base
;
192 if (likely(base
!= NULL
)) {
193 spin_lock_irqsave(&base
->cpu_base
->lock
, *flags
);
194 if (likely(base
== timer
->base
))
196 /* The timer has migrated to another CPU: */
197 spin_unlock_irqrestore(&base
->cpu_base
->lock
, *flags
);
204 * Switch the timer base to the current CPU when possible.
206 static inline struct hrtimer_clock_base
*
207 switch_hrtimer_base(struct hrtimer
*timer
, struct hrtimer_clock_base
*base
)
209 struct hrtimer_clock_base
*new_base
;
210 struct hrtimer_cpu_base
*new_cpu_base
;
212 new_cpu_base
= &__get_cpu_var(hrtimer_bases
);
213 new_base
= &new_cpu_base
->clock_base
[base
->index
];
215 if (base
!= new_base
) {
217 * We are trying to schedule the timer on the local CPU.
218 * However we can't change timer's base while it is running,
219 * so we keep it on the same CPU. No hassle vs. reprogramming
220 * the event source in the high resolution case. The softirq
221 * code will take care of this when the timer function has
222 * completed. There is no conflict as we hold the lock until
223 * the timer is enqueued.
225 if (unlikely(hrtimer_callback_running(timer
)))
228 /* See the comment in lock_timer_base() */
230 spin_unlock(&base
->cpu_base
->lock
);
231 spin_lock(&new_base
->cpu_base
->lock
);
232 timer
->base
= new_base
;
237 #else /* CONFIG_SMP */
239 static inline struct hrtimer_clock_base
*
240 lock_hrtimer_base(const struct hrtimer
*timer
, unsigned long *flags
)
242 struct hrtimer_clock_base
*base
= timer
->base
;
244 spin_lock_irqsave(&base
->cpu_base
->lock
, *flags
);
249 # define switch_hrtimer_base(t, b) (b)
251 #endif /* !CONFIG_SMP */
254 * Functions for the union type storage format of ktime_t which are
255 * too large for inlining:
257 #if BITS_PER_LONG < 64
258 # ifndef CONFIG_KTIME_SCALAR
260 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
262 * @nsec: the scalar nsec value to add
264 * Returns the sum of kt and nsec in ktime_t format
266 ktime_t
ktime_add_ns(const ktime_t kt
, u64 nsec
)
270 if (likely(nsec
< NSEC_PER_SEC
)) {
273 unsigned long rem
= do_div(nsec
, NSEC_PER_SEC
);
275 tmp
= ktime_set((long)nsec
, rem
);
278 return ktime_add(kt
, tmp
);
280 # endif /* !CONFIG_KTIME_SCALAR */
283 * Divide a ktime value by a nanosecond value
285 unsigned long ktime_divns(const ktime_t kt
, s64 div
)
290 dclc
= dns
= ktime_to_ns(kt
);
292 /* Make sure the divisor is less than 2^32: */
298 do_div(dclc
, (unsigned long) div
);
300 return (unsigned long) dclc
;
302 #endif /* BITS_PER_LONG >= 64 */
304 /* High resolution timer related functions */
305 #ifdef CONFIG_HIGH_RES_TIMERS
308 * High resolution timer enabled ?
310 static int hrtimer_hres_enabled __read_mostly
= 1;
313 * Enable / Disable high resolution mode
315 static int __init
setup_hrtimer_hres(char *str
)
317 if (!strcmp(str
, "off"))
318 hrtimer_hres_enabled
= 0;
319 else if (!strcmp(str
, "on"))
320 hrtimer_hres_enabled
= 1;
326 __setup("highres=", setup_hrtimer_hres
);
329 * hrtimer_high_res_enabled - query, if the highres mode is enabled
331 static inline int hrtimer_is_hres_enabled(void)
333 return hrtimer_hres_enabled
;
337 * Is the high resolution mode active ?
339 static inline int hrtimer_hres_active(void)
341 return __get_cpu_var(hrtimer_bases
).hres_active
;
345 * Reprogram the event source with checking both queues for the
347 * Called with interrupts disabled and base->lock held
349 static void hrtimer_force_reprogram(struct hrtimer_cpu_base
*cpu_base
)
352 struct hrtimer_clock_base
*base
= cpu_base
->clock_base
;
355 cpu_base
->expires_next
.tv64
= KTIME_MAX
;
357 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++, base
++) {
358 struct hrtimer
*timer
;
362 timer
= rb_entry(base
->first
, struct hrtimer
, node
);
363 expires
= ktime_sub(timer
->expires
, base
->offset
);
364 if (expires
.tv64
< cpu_base
->expires_next
.tv64
)
365 cpu_base
->expires_next
= expires
;
368 if (cpu_base
->expires_next
.tv64
!= KTIME_MAX
)
369 tick_program_event(cpu_base
->expires_next
, 1);
373 * Shared reprogramming for clock_realtime and clock_monotonic
375 * When a timer is enqueued and expires earlier than the already enqueued
376 * timers, we have to check, whether it expires earlier than the timer for
377 * which the clock event device was armed.
379 * Called with interrupts disabled and base->cpu_base.lock held
381 static int hrtimer_reprogram(struct hrtimer
*timer
,
382 struct hrtimer_clock_base
*base
)
384 ktime_t
*expires_next
= &__get_cpu_var(hrtimer_bases
).expires_next
;
385 ktime_t expires
= ktime_sub(timer
->expires
, base
->offset
);
389 * When the callback is running, we do not reprogram the clock event
390 * device. The timer callback is either running on a different CPU or
391 * the callback is executed in the hrtimer_interupt context. The
392 * reprogramming is handled either by the softirq, which called the
393 * callback or at the end of the hrtimer_interrupt.
395 if (hrtimer_callback_running(timer
))
398 if (expires
.tv64
>= expires_next
->tv64
)
402 * Clockevents returns -ETIME, when the event was in the past.
404 res
= tick_program_event(expires
, 0);
405 if (!IS_ERR_VALUE(res
))
406 *expires_next
= expires
;
412 * Retrigger next event is called after clock was set
414 * Called with interrupts disabled via on_each_cpu()
416 static void retrigger_next_event(void *arg
)
418 struct hrtimer_cpu_base
*base
;
419 struct timespec realtime_offset
;
422 if (!hrtimer_hres_active())
426 seq
= read_seqbegin(&xtime_lock
);
427 set_normalized_timespec(&realtime_offset
,
428 -wall_to_monotonic
.tv_sec
,
429 -wall_to_monotonic
.tv_nsec
);
430 } while (read_seqretry(&xtime_lock
, seq
));
432 base
= &__get_cpu_var(hrtimer_bases
);
434 /* Adjust CLOCK_REALTIME offset */
435 spin_lock(&base
->lock
);
436 base
->clock_base
[CLOCK_REALTIME
].offset
=
437 timespec_to_ktime(realtime_offset
);
439 hrtimer_force_reprogram(base
);
440 spin_unlock(&base
->lock
);
444 * Clock realtime was set
446 * Change the offset of the realtime clock vs. the monotonic
449 * We might have to reprogram the high resolution timer interrupt. On
450 * SMP we call the architecture specific code to retrigger _all_ high
451 * resolution timer interrupts. On UP we just disable interrupts and
452 * call the high resolution interrupt code.
454 void clock_was_set(void)
456 /* Retrigger the CPU local events everywhere */
457 on_each_cpu(retrigger_next_event
, NULL
, 0, 1);
461 * Check, whether the timer is on the callback pending list
463 static inline int hrtimer_cb_pending(const struct hrtimer
*timer
)
465 return timer
->state
& HRTIMER_STATE_PENDING
;
469 * Remove a timer from the callback pending list
471 static inline void hrtimer_remove_cb_pending(struct hrtimer
*timer
)
473 list_del_init(&timer
->cb_entry
);
477 * Initialize the high resolution related parts of cpu_base
479 static inline void hrtimer_init_hres(struct hrtimer_cpu_base
*base
)
481 base
->expires_next
.tv64
= KTIME_MAX
;
482 base
->hres_active
= 0;
483 INIT_LIST_HEAD(&base
->cb_pending
);
487 * Initialize the high resolution related parts of a hrtimer
489 static inline void hrtimer_init_timer_hres(struct hrtimer
*timer
)
491 INIT_LIST_HEAD(&timer
->cb_entry
);
495 * When High resolution timers are active, try to reprogram. Note, that in case
496 * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry
497 * check happens. The timer gets enqueued into the rbtree. The reprogramming
498 * and expiry check is done in the hrtimer_interrupt or in the softirq.
500 static inline int hrtimer_enqueue_reprogram(struct hrtimer
*timer
,
501 struct hrtimer_clock_base
*base
)
503 if (base
->cpu_base
->hres_active
&& hrtimer_reprogram(timer
, base
)) {
505 /* Timer is expired, act upon the callback mode */
506 switch(timer
->cb_mode
) {
507 case HRTIMER_CB_IRQSAFE_NO_RESTART
:
509 * We can call the callback from here. No restart
510 * happens, so no danger of recursion
512 BUG_ON(timer
->function(timer
) != HRTIMER_NORESTART
);
514 case HRTIMER_CB_IRQSAFE_NO_SOFTIRQ
:
516 * This is solely for the sched tick emulation with
517 * dynamic tick support to ensure that we do not
518 * restart the tick right on the edge and end up with
519 * the tick timer in the softirq ! The calling site
520 * takes care of this.
523 case HRTIMER_CB_IRQSAFE
:
524 case HRTIMER_CB_SOFTIRQ
:
526 * Move everything else into the softirq pending list !
528 list_add_tail(&timer
->cb_entry
,
529 &base
->cpu_base
->cb_pending
);
530 timer
->state
= HRTIMER_STATE_PENDING
;
531 raise_softirq(HRTIMER_SOFTIRQ
);
541 * Switch to high resolution mode
543 static int hrtimer_switch_to_hres(void)
545 struct hrtimer_cpu_base
*base
= &__get_cpu_var(hrtimer_bases
);
548 if (base
->hres_active
)
551 local_irq_save(flags
);
553 if (tick_init_highres()) {
554 local_irq_restore(flags
);
557 base
->hres_active
= 1;
558 base
->clock_base
[CLOCK_REALTIME
].resolution
= KTIME_HIGH_RES
;
559 base
->clock_base
[CLOCK_MONOTONIC
].resolution
= KTIME_HIGH_RES
;
561 tick_setup_sched_timer();
563 /* "Retrigger" the interrupt to get things going */
564 retrigger_next_event(NULL
);
565 local_irq_restore(flags
);
566 printk(KERN_INFO
"Switched to high resolution mode on CPU %d\n",
573 static inline int hrtimer_hres_active(void) { return 0; }
574 static inline int hrtimer_is_hres_enabled(void) { return 0; }
575 static inline int hrtimer_switch_to_hres(void) { return 0; }
576 static inline void hrtimer_force_reprogram(struct hrtimer_cpu_base
*base
) { }
577 static inline int hrtimer_enqueue_reprogram(struct hrtimer
*timer
,
578 struct hrtimer_clock_base
*base
)
582 static inline int hrtimer_cb_pending(struct hrtimer
*timer
) { return 0; }
583 static inline void hrtimer_remove_cb_pending(struct hrtimer
*timer
) { }
584 static inline void hrtimer_init_hres(struct hrtimer_cpu_base
*base
) { }
585 static inline void hrtimer_init_timer_hres(struct hrtimer
*timer
) { }
587 #endif /* CONFIG_HIGH_RES_TIMERS */
589 #ifdef CONFIG_TIMER_STATS
590 void __timer_stats_hrtimer_set_start_info(struct hrtimer
*timer
, void *addr
)
592 if (timer
->start_site
)
595 timer
->start_site
= addr
;
596 memcpy(timer
->start_comm
, current
->comm
, TASK_COMM_LEN
);
597 timer
->start_pid
= current
->pid
;
602 * Counterpart to lock_timer_base above:
605 void unlock_hrtimer_base(const struct hrtimer
*timer
, unsigned long *flags
)
607 spin_unlock_irqrestore(&timer
->base
->cpu_base
->lock
, *flags
);
611 * hrtimer_forward - forward the timer expiry
612 * @timer: hrtimer to forward
613 * @now: forward past this time
614 * @interval: the interval to forward
616 * Forward the timer expiry so it will expire in the future.
617 * Returns the number of overruns.
620 hrtimer_forward(struct hrtimer
*timer
, ktime_t now
, ktime_t interval
)
622 unsigned long orun
= 1;
625 delta
= ktime_sub(now
, timer
->expires
);
630 if (interval
.tv64
< timer
->base
->resolution
.tv64
)
631 interval
.tv64
= timer
->base
->resolution
.tv64
;
633 if (unlikely(delta
.tv64
>= interval
.tv64
)) {
634 s64 incr
= ktime_to_ns(interval
);
636 orun
= ktime_divns(delta
, incr
);
637 timer
->expires
= ktime_add_ns(timer
->expires
, incr
* orun
);
638 if (timer
->expires
.tv64
> now
.tv64
)
641 * This (and the ktime_add() below) is the
642 * correction for exact:
646 timer
->expires
= ktime_add(timer
->expires
, interval
);
652 * enqueue_hrtimer - internal function to (re)start a timer
654 * The timer is inserted in expiry order. Insertion into the
655 * red black tree is O(log(n)). Must hold the base lock.
657 static void enqueue_hrtimer(struct hrtimer
*timer
,
658 struct hrtimer_clock_base
*base
, int reprogram
)
660 struct rb_node
**link
= &base
->active
.rb_node
;
661 struct rb_node
*parent
= NULL
;
662 struct hrtimer
*entry
;
665 * Find the right place in the rbtree:
669 entry
= rb_entry(parent
, struct hrtimer
, node
);
671 * We dont care about collisions. Nodes with
672 * the same expiry time stay together.
674 if (timer
->expires
.tv64
< entry
->expires
.tv64
)
675 link
= &(*link
)->rb_left
;
677 link
= &(*link
)->rb_right
;
681 * Insert the timer to the rbtree and check whether it
682 * replaces the first pending timer
684 if (!base
->first
|| timer
->expires
.tv64
<
685 rb_entry(base
->first
, struct hrtimer
, node
)->expires
.tv64
) {
687 * Reprogram the clock event device. When the timer is already
688 * expired hrtimer_enqueue_reprogram has either called the
689 * callback or added it to the pending list and raised the
692 * This is a NOP for !HIGHRES
694 if (reprogram
&& hrtimer_enqueue_reprogram(timer
, base
))
697 base
->first
= &timer
->node
;
700 rb_link_node(&timer
->node
, parent
, link
);
701 rb_insert_color(&timer
->node
, &base
->active
);
703 * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
704 * state of a possibly running callback.
706 timer
->state
|= HRTIMER_STATE_ENQUEUED
;
710 * __remove_hrtimer - internal function to remove a timer
712 * Caller must hold the base lock.
714 * High resolution timer mode reprograms the clock event device when the
715 * timer is the one which expires next. The caller can disable this by setting
716 * reprogram to zero. This is useful, when the context does a reprogramming
717 * anyway (e.g. timer interrupt)
719 static void __remove_hrtimer(struct hrtimer
*timer
,
720 struct hrtimer_clock_base
*base
,
721 unsigned long newstate
, int reprogram
)
723 /* High res. callback list. NOP for !HIGHRES */
724 if (hrtimer_cb_pending(timer
))
725 hrtimer_remove_cb_pending(timer
);
728 * Remove the timer from the rbtree and replace the
729 * first entry pointer if necessary.
731 if (base
->first
== &timer
->node
) {
732 base
->first
= rb_next(&timer
->node
);
733 /* Reprogram the clock event device. if enabled */
734 if (reprogram
&& hrtimer_hres_active())
735 hrtimer_force_reprogram(base
->cpu_base
);
737 rb_erase(&timer
->node
, &base
->active
);
739 timer
->state
= newstate
;
743 * remove hrtimer, called with base lock held
746 remove_hrtimer(struct hrtimer
*timer
, struct hrtimer_clock_base
*base
)
748 if (hrtimer_is_queued(timer
)) {
752 * Remove the timer and force reprogramming when high
753 * resolution mode is active and the timer is on the current
754 * CPU. If we remove a timer on another CPU, reprogramming is
755 * skipped. The interrupt event on this CPU is fired and
756 * reprogramming happens in the interrupt handler. This is a
757 * rare case and less expensive than a smp call.
759 timer_stats_hrtimer_clear_start_info(timer
);
760 reprogram
= base
->cpu_base
== &__get_cpu_var(hrtimer_bases
);
761 __remove_hrtimer(timer
, base
, HRTIMER_STATE_INACTIVE
,
769 * hrtimer_start - (re)start an relative timer on the current CPU
770 * @timer: the timer to be added
772 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
776 * 1 when the timer was active
779 hrtimer_start(struct hrtimer
*timer
, ktime_t tim
, const enum hrtimer_mode mode
)
781 struct hrtimer_clock_base
*base
, *new_base
;
785 base
= lock_hrtimer_base(timer
, &flags
);
787 /* Remove an active timer from the queue: */
788 ret
= remove_hrtimer(timer
, base
);
790 /* Switch the timer base, if necessary: */
791 new_base
= switch_hrtimer_base(timer
, base
);
793 if (mode
== HRTIMER_MODE_REL
) {
794 tim
= ktime_add(tim
, new_base
->get_time());
796 * CONFIG_TIME_LOW_RES is a temporary way for architectures
797 * to signal that they simply return xtime in
798 * do_gettimeoffset(). In this case we want to round up by
799 * resolution when starting a relative timer, to avoid short
800 * timeouts. This will go away with the GTOD framework.
802 #ifdef CONFIG_TIME_LOW_RES
803 tim
= ktime_add(tim
, base
->resolution
);
806 timer
->expires
= tim
;
808 timer_stats_hrtimer_set_start_info(timer
);
810 enqueue_hrtimer(timer
, new_base
, base
== new_base
);
812 unlock_hrtimer_base(timer
, &flags
);
816 EXPORT_SYMBOL_GPL(hrtimer_start
);
819 * hrtimer_try_to_cancel - try to deactivate a timer
820 * @timer: hrtimer to stop
823 * 0 when the timer was not active
824 * 1 when the timer was active
825 * -1 when the timer is currently excuting the callback function and
828 int hrtimer_try_to_cancel(struct hrtimer
*timer
)
830 struct hrtimer_clock_base
*base
;
834 base
= lock_hrtimer_base(timer
, &flags
);
836 if (!hrtimer_callback_running(timer
))
837 ret
= remove_hrtimer(timer
, base
);
839 unlock_hrtimer_base(timer
, &flags
);
844 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel
);
847 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
848 * @timer: the timer to be cancelled
851 * 0 when the timer was not active
852 * 1 when the timer was active
854 int hrtimer_cancel(struct hrtimer
*timer
)
857 int ret
= hrtimer_try_to_cancel(timer
);
864 EXPORT_SYMBOL_GPL(hrtimer_cancel
);
867 * hrtimer_get_remaining - get remaining time for the timer
868 * @timer: the timer to read
870 ktime_t
hrtimer_get_remaining(const struct hrtimer
*timer
)
872 struct hrtimer_clock_base
*base
;
876 base
= lock_hrtimer_base(timer
, &flags
);
877 rem
= ktime_sub(timer
->expires
, base
->get_time());
878 unlock_hrtimer_base(timer
, &flags
);
882 EXPORT_SYMBOL_GPL(hrtimer_get_remaining
);
884 #if defined(CONFIG_NO_IDLE_HZ) || defined(CONFIG_NO_HZ)
886 * hrtimer_get_next_event - get the time until next expiry event
888 * Returns the delta to the next expiry event or KTIME_MAX if no timer
891 ktime_t
hrtimer_get_next_event(void)
893 struct hrtimer_cpu_base
*cpu_base
= &__get_cpu_var(hrtimer_bases
);
894 struct hrtimer_clock_base
*base
= cpu_base
->clock_base
;
895 ktime_t delta
, mindelta
= { .tv64
= KTIME_MAX
};
899 spin_lock_irqsave(&cpu_base
->lock
, flags
);
901 if (!hrtimer_hres_active()) {
902 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++, base
++) {
903 struct hrtimer
*timer
;
908 timer
= rb_entry(base
->first
, struct hrtimer
, node
);
909 delta
.tv64
= timer
->expires
.tv64
;
910 delta
= ktime_sub(delta
, base
->get_time());
911 if (delta
.tv64
< mindelta
.tv64
)
912 mindelta
.tv64
= delta
.tv64
;
916 spin_unlock_irqrestore(&cpu_base
->lock
, flags
);
918 if (mindelta
.tv64
< 0)
925 * hrtimer_init - initialize a timer to the given clock
926 * @timer: the timer to be initialized
927 * @clock_id: the clock to be used
928 * @mode: timer mode abs/rel
930 void hrtimer_init(struct hrtimer
*timer
, clockid_t clock_id
,
931 enum hrtimer_mode mode
)
933 struct hrtimer_cpu_base
*cpu_base
;
935 memset(timer
, 0, sizeof(struct hrtimer
));
937 cpu_base
= &__raw_get_cpu_var(hrtimer_bases
);
939 if (clock_id
== CLOCK_REALTIME
&& mode
!= HRTIMER_MODE_ABS
)
940 clock_id
= CLOCK_MONOTONIC
;
942 timer
->base
= &cpu_base
->clock_base
[clock_id
];
943 hrtimer_init_timer_hres(timer
);
945 #ifdef CONFIG_TIMER_STATS
946 timer
->start_site
= NULL
;
947 timer
->start_pid
= -1;
948 memset(timer
->start_comm
, 0, TASK_COMM_LEN
);
951 EXPORT_SYMBOL_GPL(hrtimer_init
);
954 * hrtimer_get_res - get the timer resolution for a clock
955 * @which_clock: which clock to query
956 * @tp: pointer to timespec variable to store the resolution
958 * Store the resolution of the clock selected by @which_clock in the
959 * variable pointed to by @tp.
961 int hrtimer_get_res(const clockid_t which_clock
, struct timespec
*tp
)
963 struct hrtimer_cpu_base
*cpu_base
;
965 cpu_base
= &__raw_get_cpu_var(hrtimer_bases
);
966 *tp
= ktime_to_timespec(cpu_base
->clock_base
[which_clock
].resolution
);
970 EXPORT_SYMBOL_GPL(hrtimer_get_res
);
972 #ifdef CONFIG_HIGH_RES_TIMERS
975 * High resolution timer interrupt
976 * Called with interrupts disabled
978 void hrtimer_interrupt(struct clock_event_device
*dev
)
980 struct hrtimer_cpu_base
*cpu_base
= &__get_cpu_var(hrtimer_bases
);
981 struct hrtimer_clock_base
*base
;
982 ktime_t expires_next
, now
;
985 BUG_ON(!cpu_base
->hres_active
);
986 cpu_base
->nr_events
++;
987 dev
->next_event
.tv64
= KTIME_MAX
;
992 expires_next
.tv64
= KTIME_MAX
;
994 base
= cpu_base
->clock_base
;
996 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++) {
998 struct rb_node
*node
;
1000 spin_lock(&cpu_base
->lock
);
1002 basenow
= ktime_add(now
, base
->offset
);
1004 while ((node
= base
->first
)) {
1005 struct hrtimer
*timer
;
1007 timer
= rb_entry(node
, struct hrtimer
, node
);
1009 if (basenow
.tv64
< timer
->expires
.tv64
) {
1012 expires
= ktime_sub(timer
->expires
,
1014 if (expires
.tv64
< expires_next
.tv64
)
1015 expires_next
= expires
;
1019 /* Move softirq callbacks to the pending list */
1020 if (timer
->cb_mode
== HRTIMER_CB_SOFTIRQ
) {
1021 __remove_hrtimer(timer
, base
,
1022 HRTIMER_STATE_PENDING
, 0);
1023 list_add_tail(&timer
->cb_entry
,
1024 &base
->cpu_base
->cb_pending
);
1029 __remove_hrtimer(timer
, base
,
1030 HRTIMER_STATE_CALLBACK
, 0);
1031 timer_stats_account_hrtimer(timer
);
1034 * Note: We clear the CALLBACK bit after
1035 * enqueue_hrtimer to avoid reprogramming of
1036 * the event hardware. This happens at the end
1037 * of this function anyway.
1039 if (timer
->function(timer
) != HRTIMER_NORESTART
) {
1040 BUG_ON(timer
->state
!= HRTIMER_STATE_CALLBACK
);
1041 enqueue_hrtimer(timer
, base
, 0);
1043 timer
->state
&= ~HRTIMER_STATE_CALLBACK
;
1045 spin_unlock(&cpu_base
->lock
);
1049 cpu_base
->expires_next
= expires_next
;
1051 /* Reprogramming necessary ? */
1052 if (expires_next
.tv64
!= KTIME_MAX
) {
1053 if (tick_program_event(expires_next
, 0))
1057 /* Raise softirq ? */
1059 raise_softirq(HRTIMER_SOFTIRQ
);
1062 static void run_hrtimer_softirq(struct softirq_action
*h
)
1064 struct hrtimer_cpu_base
*cpu_base
= &__get_cpu_var(hrtimer_bases
);
1066 spin_lock_irq(&cpu_base
->lock
);
1068 while (!list_empty(&cpu_base
->cb_pending
)) {
1069 enum hrtimer_restart (*fn
)(struct hrtimer
*);
1070 struct hrtimer
*timer
;
1073 timer
= list_entry(cpu_base
->cb_pending
.next
,
1074 struct hrtimer
, cb_entry
);
1076 timer_stats_account_hrtimer(timer
);
1078 fn
= timer
->function
;
1079 __remove_hrtimer(timer
, timer
->base
, HRTIMER_STATE_CALLBACK
, 0);
1080 spin_unlock_irq(&cpu_base
->lock
);
1082 restart
= fn(timer
);
1084 spin_lock_irq(&cpu_base
->lock
);
1086 timer
->state
&= ~HRTIMER_STATE_CALLBACK
;
1087 if (restart
== HRTIMER_RESTART
) {
1088 BUG_ON(hrtimer_active(timer
));
1090 * Enqueue the timer, allow reprogramming of the event
1093 enqueue_hrtimer(timer
, timer
->base
, 1);
1094 } else if (hrtimer_active(timer
)) {
1096 * If the timer was rearmed on another CPU, reprogram
1099 if (timer
->base
->first
== &timer
->node
)
1100 hrtimer_reprogram(timer
, timer
->base
);
1103 spin_unlock_irq(&cpu_base
->lock
);
1106 #endif /* CONFIG_HIGH_RES_TIMERS */
1109 * Expire the per base hrtimer-queue:
1111 static inline void run_hrtimer_queue(struct hrtimer_cpu_base
*cpu_base
,
1114 struct rb_node
*node
;
1115 struct hrtimer_clock_base
*base
= &cpu_base
->clock_base
[index
];
1120 if (base
->get_softirq_time
)
1121 base
->softirq_time
= base
->get_softirq_time();
1123 spin_lock_irq(&cpu_base
->lock
);
1125 while ((node
= base
->first
)) {
1126 struct hrtimer
*timer
;
1127 enum hrtimer_restart (*fn
)(struct hrtimer
*);
1130 timer
= rb_entry(node
, struct hrtimer
, node
);
1131 if (base
->softirq_time
.tv64
<= timer
->expires
.tv64
)
1134 #ifdef CONFIG_HIGH_RES_TIMERS
1135 WARN_ON_ONCE(timer
->cb_mode
== HRTIMER_CB_IRQSAFE_NO_SOFTIRQ
);
1137 timer_stats_account_hrtimer(timer
);
1139 fn
= timer
->function
;
1140 __remove_hrtimer(timer
, base
, HRTIMER_STATE_CALLBACK
, 0);
1141 spin_unlock_irq(&cpu_base
->lock
);
1143 restart
= fn(timer
);
1145 spin_lock_irq(&cpu_base
->lock
);
1147 timer
->state
&= ~HRTIMER_STATE_CALLBACK
;
1148 if (restart
!= HRTIMER_NORESTART
) {
1149 BUG_ON(hrtimer_active(timer
));
1150 enqueue_hrtimer(timer
, base
, 0);
1153 spin_unlock_irq(&cpu_base
->lock
);
1157 * Called from timer softirq every jiffy, expire hrtimers:
1159 * For HRT its the fall back code to run the softirq in the timer
1160 * softirq context in case the hrtimer initialization failed or has
1161 * not been done yet.
1163 void hrtimer_run_queues(void)
1165 struct hrtimer_cpu_base
*cpu_base
= &__get_cpu_var(hrtimer_bases
);
1168 if (hrtimer_hres_active())
1172 * This _is_ ugly: We have to check in the softirq context,
1173 * whether we can switch to highres and / or nohz mode. The
1174 * clocksource switch happens in the timer interrupt with
1175 * xtime_lock held. Notification from there only sets the
1176 * check bit in the tick_oneshot code, otherwise we might
1177 * deadlock vs. xtime_lock.
1179 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1180 if (hrtimer_switch_to_hres())
1183 hrtimer_get_softirq_time(cpu_base
);
1185 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++)
1186 run_hrtimer_queue(cpu_base
, i
);
1190 * Sleep related functions:
1192 static enum hrtimer_restart
hrtimer_wakeup(struct hrtimer
*timer
)
1194 struct hrtimer_sleeper
*t
=
1195 container_of(timer
, struct hrtimer_sleeper
, timer
);
1196 struct task_struct
*task
= t
->task
;
1200 wake_up_process(task
);
1202 return HRTIMER_NORESTART
;
1205 void hrtimer_init_sleeper(struct hrtimer_sleeper
*sl
, struct task_struct
*task
)
1207 sl
->timer
.function
= hrtimer_wakeup
;
1209 #ifdef CONFIG_HIGH_RES_TIMERS
1210 sl
->timer
.cb_mode
= HRTIMER_CB_IRQSAFE_NO_RESTART
;
1214 static int __sched
do_nanosleep(struct hrtimer_sleeper
*t
, enum hrtimer_mode mode
)
1216 hrtimer_init_sleeper(t
, current
);
1219 set_current_state(TASK_INTERRUPTIBLE
);
1220 hrtimer_start(&t
->timer
, t
->timer
.expires
, mode
);
1222 if (likely(t
->task
))
1225 hrtimer_cancel(&t
->timer
);
1226 mode
= HRTIMER_MODE_ABS
;
1228 } while (t
->task
&& !signal_pending(current
));
1230 return t
->task
== NULL
;
1233 long __sched
hrtimer_nanosleep_restart(struct restart_block
*restart
)
1235 struct hrtimer_sleeper t
;
1236 struct timespec __user
*rmtp
;
1240 restart
->fn
= do_no_restart_syscall
;
1242 hrtimer_init(&t
.timer
, restart
->arg0
, HRTIMER_MODE_ABS
);
1243 t
.timer
.expires
.tv64
= ((u64
)restart
->arg3
<< 32) | (u64
) restart
->arg2
;
1245 if (do_nanosleep(&t
, HRTIMER_MODE_ABS
))
1248 rmtp
= (struct timespec __user
*) restart
->arg1
;
1250 time
= ktime_sub(t
.timer
.expires
, t
.timer
.base
->get_time());
1253 tu
= ktime_to_timespec(time
);
1254 if (copy_to_user(rmtp
, &tu
, sizeof(tu
)))
1258 restart
->fn
= hrtimer_nanosleep_restart
;
1260 /* The other values in restart are already filled in */
1261 return -ERESTART_RESTARTBLOCK
;
1264 long hrtimer_nanosleep(struct timespec
*rqtp
, struct timespec __user
*rmtp
,
1265 const enum hrtimer_mode mode
, const clockid_t clockid
)
1267 struct restart_block
*restart
;
1268 struct hrtimer_sleeper t
;
1272 hrtimer_init(&t
.timer
, clockid
, mode
);
1273 t
.timer
.expires
= timespec_to_ktime(*rqtp
);
1274 if (do_nanosleep(&t
, mode
))
1277 /* Absolute timers do not update the rmtp value and restart: */
1278 if (mode
== HRTIMER_MODE_ABS
)
1279 return -ERESTARTNOHAND
;
1282 rem
= ktime_sub(t
.timer
.expires
, t
.timer
.base
->get_time());
1285 tu
= ktime_to_timespec(rem
);
1286 if (copy_to_user(rmtp
, &tu
, sizeof(tu
)))
1290 restart
= ¤t_thread_info()->restart_block
;
1291 restart
->fn
= hrtimer_nanosleep_restart
;
1292 restart
->arg0
= (unsigned long) t
.timer
.base
->index
;
1293 restart
->arg1
= (unsigned long) rmtp
;
1294 restart
->arg2
= t
.timer
.expires
.tv64
& 0xFFFFFFFF;
1295 restart
->arg3
= t
.timer
.expires
.tv64
>> 32;
1297 return -ERESTART_RESTARTBLOCK
;
1301 sys_nanosleep(struct timespec __user
*rqtp
, struct timespec __user
*rmtp
)
1305 if (copy_from_user(&tu
, rqtp
, sizeof(tu
)))
1308 if (!timespec_valid(&tu
))
1311 return hrtimer_nanosleep(&tu
, rmtp
, HRTIMER_MODE_REL
, CLOCK_MONOTONIC
);
1315 * Functions related to boot-time initialization:
1317 static void __devinit
init_hrtimers_cpu(int cpu
)
1319 struct hrtimer_cpu_base
*cpu_base
= &per_cpu(hrtimer_bases
, cpu
);
1322 spin_lock_init(&cpu_base
->lock
);
1323 lockdep_set_class(&cpu_base
->lock
, &cpu_base
->lock_key
);
1325 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++)
1326 cpu_base
->clock_base
[i
].cpu_base
= cpu_base
;
1328 hrtimer_init_hres(cpu_base
);
1331 #ifdef CONFIG_HOTPLUG_CPU
1333 static void migrate_hrtimer_list(struct hrtimer_clock_base
*old_base
,
1334 struct hrtimer_clock_base
*new_base
)
1336 struct hrtimer
*timer
;
1337 struct rb_node
*node
;
1339 while ((node
= rb_first(&old_base
->active
))) {
1340 timer
= rb_entry(node
, struct hrtimer
, node
);
1341 BUG_ON(hrtimer_callback_running(timer
));
1342 __remove_hrtimer(timer
, old_base
, HRTIMER_STATE_INACTIVE
, 0);
1343 timer
->base
= new_base
;
1345 * Enqueue the timer. Allow reprogramming of the event device
1347 enqueue_hrtimer(timer
, new_base
, 1);
1351 static void migrate_hrtimers(int cpu
)
1353 struct hrtimer_cpu_base
*old_base
, *new_base
;
1356 BUG_ON(cpu_online(cpu
));
1357 old_base
= &per_cpu(hrtimer_bases
, cpu
);
1358 new_base
= &get_cpu_var(hrtimer_bases
);
1360 tick_cancel_sched_timer(cpu
);
1362 local_irq_disable();
1363 double_spin_lock(&new_base
->lock
, &old_base
->lock
,
1364 smp_processor_id() < cpu
);
1366 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++) {
1367 migrate_hrtimer_list(&old_base
->clock_base
[i
],
1368 &new_base
->clock_base
[i
]);
1371 double_spin_unlock(&new_base
->lock
, &old_base
->lock
,
1372 smp_processor_id() < cpu
);
1374 put_cpu_var(hrtimer_bases
);
1376 #endif /* CONFIG_HOTPLUG_CPU */
1378 static int __cpuinit
hrtimer_cpu_notify(struct notifier_block
*self
,
1379 unsigned long action
, void *hcpu
)
1381 long cpu
= (long)hcpu
;
1385 case CPU_UP_PREPARE
:
1386 init_hrtimers_cpu(cpu
);
1389 #ifdef CONFIG_HOTPLUG_CPU
1391 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD
, &cpu
);
1392 migrate_hrtimers(cpu
);
1403 static struct notifier_block __cpuinitdata hrtimers_nb
= {
1404 .notifier_call
= hrtimer_cpu_notify
,
1407 void __init
hrtimers_init(void)
1409 hrtimer_cpu_notify(&hrtimers_nb
, (unsigned long)CPU_UP_PREPARE
,
1410 (void *)(long)smp_processor_id());
1411 register_cpu_notifier(&hrtimers_nb
);
1412 #ifdef CONFIG_HIGH_RES_TIMERS
1413 open_softirq(HRTIMER_SOFTIRQ
, run_hrtimer_softirq
, NULL
);