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
);
62 EXPORT_SYMBOL_GPL(ktime_get
);
65 * ktime_get_real - get the real (wall-) time in ktime_t format
67 * returns the time in ktime_t format
69 ktime_t
ktime_get_real(void)
75 return timespec_to_ktime(now
);
78 EXPORT_SYMBOL_GPL(ktime_get_real
);
83 * Note: If we want to add new timer bases, we have to skip the two
84 * clock ids captured by the cpu-timers. We do this by holding empty
85 * entries rather than doing math adjustment of the clock ids.
86 * This ensures that we capture erroneous accesses to these clock ids
87 * rather than moving them into the range of valid clock id's.
89 DEFINE_PER_CPU(struct hrtimer_cpu_base
, hrtimer_bases
) =
95 .index
= CLOCK_REALTIME
,
96 .get_time
= &ktime_get_real
,
97 .resolution
= KTIME_LOW_RES
,
100 .index
= CLOCK_MONOTONIC
,
101 .get_time
= &ktime_get
,
102 .resolution
= KTIME_LOW_RES
,
108 * ktime_get_ts - get the monotonic clock in timespec format
109 * @ts: pointer to timespec variable
111 * The function calculates the monotonic clock from the realtime
112 * clock and the wall_to_monotonic offset and stores the result
113 * in normalized timespec format in the variable pointed to by @ts.
115 void ktime_get_ts(struct timespec
*ts
)
117 struct timespec tomono
;
121 seq
= read_seqbegin(&xtime_lock
);
123 tomono
= wall_to_monotonic
;
125 } while (read_seqretry(&xtime_lock
, seq
));
127 set_normalized_timespec(ts
, ts
->tv_sec
+ tomono
.tv_sec
,
128 ts
->tv_nsec
+ tomono
.tv_nsec
);
130 EXPORT_SYMBOL_GPL(ktime_get_ts
);
133 * Get the coarse grained time at the softirq based on xtime and
136 static void hrtimer_get_softirq_time(struct hrtimer_cpu_base
*base
)
138 ktime_t xtim
, tomono
;
139 struct timespec xts
, tom
;
143 seq
= read_seqbegin(&xtime_lock
);
145 getnstimeofday(&xts
);
149 tom
= wall_to_monotonic
;
150 } while (read_seqretry(&xtime_lock
, seq
));
152 xtim
= timespec_to_ktime(xts
);
153 tomono
= timespec_to_ktime(tom
);
154 base
->clock_base
[CLOCK_REALTIME
].softirq_time
= xtim
;
155 base
->clock_base
[CLOCK_MONOTONIC
].softirq_time
=
156 ktime_add(xtim
, tomono
);
160 * Helper function to check, whether the timer is running the callback
163 static inline int hrtimer_callback_running(struct hrtimer
*timer
)
165 return timer
->state
& HRTIMER_STATE_CALLBACK
;
169 * Functions and macros which are different for UP/SMP systems are kept in a
175 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
176 * means that all timers which are tied to this base via timer->base are
177 * locked, and the base itself is locked too.
179 * So __run_timers/migrate_timers can safely modify all timers which could
180 * be found on the lists/queues.
182 * When the timer's base is locked, and the timer removed from list, it is
183 * possible to set timer->base = NULL and drop the lock: the timer remains
187 struct hrtimer_clock_base
*lock_hrtimer_base(const struct hrtimer
*timer
,
188 unsigned long *flags
)
190 struct hrtimer_clock_base
*base
;
194 if (likely(base
!= NULL
)) {
195 spin_lock_irqsave(&base
->cpu_base
->lock
, *flags
);
196 if (likely(base
== timer
->base
))
198 /* The timer has migrated to another CPU: */
199 spin_unlock_irqrestore(&base
->cpu_base
->lock
, *flags
);
206 * Switch the timer base to the current CPU when possible.
208 static inline struct hrtimer_clock_base
*
209 switch_hrtimer_base(struct hrtimer
*timer
, struct hrtimer_clock_base
*base
)
211 struct hrtimer_clock_base
*new_base
;
212 struct hrtimer_cpu_base
*new_cpu_base
;
214 new_cpu_base
= &__get_cpu_var(hrtimer_bases
);
215 new_base
= &new_cpu_base
->clock_base
[base
->index
];
217 if (base
!= new_base
) {
219 * We are trying to schedule the timer on the local CPU.
220 * However we can't change timer's base while it is running,
221 * so we keep it on the same CPU. No hassle vs. reprogramming
222 * the event source in the high resolution case. The softirq
223 * code will take care of this when the timer function has
224 * completed. There is no conflict as we hold the lock until
225 * the timer is enqueued.
227 if (unlikely(hrtimer_callback_running(timer
)))
230 /* See the comment in lock_timer_base() */
232 spin_unlock(&base
->cpu_base
->lock
);
233 spin_lock(&new_base
->cpu_base
->lock
);
234 timer
->base
= new_base
;
239 #else /* CONFIG_SMP */
241 static inline struct hrtimer_clock_base
*
242 lock_hrtimer_base(const struct hrtimer
*timer
, unsigned long *flags
)
244 struct hrtimer_clock_base
*base
= timer
->base
;
246 spin_lock_irqsave(&base
->cpu_base
->lock
, *flags
);
251 # define switch_hrtimer_base(t, b) (b)
253 #endif /* !CONFIG_SMP */
256 * Functions for the union type storage format of ktime_t which are
257 * too large for inlining:
259 #if BITS_PER_LONG < 64
260 # ifndef CONFIG_KTIME_SCALAR
262 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
264 * @nsec: the scalar nsec value to add
266 * Returns the sum of kt and nsec in ktime_t format
268 ktime_t
ktime_add_ns(const ktime_t kt
, u64 nsec
)
272 if (likely(nsec
< NSEC_PER_SEC
)) {
275 unsigned long rem
= do_div(nsec
, NSEC_PER_SEC
);
277 tmp
= ktime_set((long)nsec
, rem
);
280 return ktime_add(kt
, tmp
);
283 EXPORT_SYMBOL_GPL(ktime_add_ns
);
284 # endif /* !CONFIG_KTIME_SCALAR */
287 * Divide a ktime value by a nanosecond value
289 unsigned long ktime_divns(const ktime_t kt
, s64 div
)
294 dclc
= dns
= ktime_to_ns(kt
);
296 /* Make sure the divisor is less than 2^32: */
302 do_div(dclc
, (unsigned long) div
);
304 return (unsigned long) dclc
;
306 #endif /* BITS_PER_LONG >= 64 */
308 /* High resolution timer related functions */
309 #ifdef CONFIG_HIGH_RES_TIMERS
312 * High resolution timer enabled ?
314 static int hrtimer_hres_enabled __read_mostly
= 1;
317 * Enable / Disable high resolution mode
319 static int __init
setup_hrtimer_hres(char *str
)
321 if (!strcmp(str
, "off"))
322 hrtimer_hres_enabled
= 0;
323 else if (!strcmp(str
, "on"))
324 hrtimer_hres_enabled
= 1;
330 __setup("highres=", setup_hrtimer_hres
);
333 * hrtimer_high_res_enabled - query, if the highres mode is enabled
335 static inline int hrtimer_is_hres_enabled(void)
337 return hrtimer_hres_enabled
;
341 * Is the high resolution mode active ?
343 static inline int hrtimer_hres_active(void)
345 return __get_cpu_var(hrtimer_bases
).hres_active
;
349 * Reprogram the event source with checking both queues for the
351 * Called with interrupts disabled and base->lock held
353 static void hrtimer_force_reprogram(struct hrtimer_cpu_base
*cpu_base
)
356 struct hrtimer_clock_base
*base
= cpu_base
->clock_base
;
359 cpu_base
->expires_next
.tv64
= KTIME_MAX
;
361 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++, base
++) {
362 struct hrtimer
*timer
;
366 timer
= rb_entry(base
->first
, struct hrtimer
, node
);
367 expires
= ktime_sub(timer
->expires
, base
->offset
);
368 if (expires
.tv64
< cpu_base
->expires_next
.tv64
)
369 cpu_base
->expires_next
= expires
;
372 if (cpu_base
->expires_next
.tv64
!= KTIME_MAX
)
373 tick_program_event(cpu_base
->expires_next
, 1);
377 * Shared reprogramming for clock_realtime and clock_monotonic
379 * When a timer is enqueued and expires earlier than the already enqueued
380 * timers, we have to check, whether it expires earlier than the timer for
381 * which the clock event device was armed.
383 * Called with interrupts disabled and base->cpu_base.lock held
385 static int hrtimer_reprogram(struct hrtimer
*timer
,
386 struct hrtimer_clock_base
*base
)
388 ktime_t
*expires_next
= &__get_cpu_var(hrtimer_bases
).expires_next
;
389 ktime_t expires
= ktime_sub(timer
->expires
, base
->offset
);
393 * When the callback is running, we do not reprogram the clock event
394 * device. The timer callback is either running on a different CPU or
395 * the callback is executed in the hrtimer_interupt context. The
396 * reprogramming is handled either by the softirq, which called the
397 * callback or at the end of the hrtimer_interrupt.
399 if (hrtimer_callback_running(timer
))
402 if (expires
.tv64
>= expires_next
->tv64
)
406 * Clockevents returns -ETIME, when the event was in the past.
408 res
= tick_program_event(expires
, 0);
409 if (!IS_ERR_VALUE(res
))
410 *expires_next
= expires
;
416 * Retrigger next event is called after clock was set
418 * Called with interrupts disabled via on_each_cpu()
420 static void retrigger_next_event(void *arg
)
422 struct hrtimer_cpu_base
*base
;
423 struct timespec realtime_offset
;
426 if (!hrtimer_hres_active())
430 seq
= read_seqbegin(&xtime_lock
);
431 set_normalized_timespec(&realtime_offset
,
432 -wall_to_monotonic
.tv_sec
,
433 -wall_to_monotonic
.tv_nsec
);
434 } while (read_seqretry(&xtime_lock
, seq
));
436 base
= &__get_cpu_var(hrtimer_bases
);
438 /* Adjust CLOCK_REALTIME offset */
439 spin_lock(&base
->lock
);
440 base
->clock_base
[CLOCK_REALTIME
].offset
=
441 timespec_to_ktime(realtime_offset
);
443 hrtimer_force_reprogram(base
);
444 spin_unlock(&base
->lock
);
448 * Clock realtime was set
450 * Change the offset of the realtime clock vs. the monotonic
453 * We might have to reprogram the high resolution timer interrupt. On
454 * SMP we call the architecture specific code to retrigger _all_ high
455 * resolution timer interrupts. On UP we just disable interrupts and
456 * call the high resolution interrupt code.
458 void clock_was_set(void)
460 /* Retrigger the CPU local events everywhere */
461 on_each_cpu(retrigger_next_event
, NULL
, 0, 1);
465 * During resume we might have to reprogram the high resolution timer
466 * interrupt (on the local CPU):
468 void hres_timers_resume(void)
470 WARN_ON_ONCE(num_online_cpus() > 1);
472 /* Retrigger the CPU local events: */
473 retrigger_next_event(NULL
);
477 * Check, whether the timer is on the callback pending list
479 static inline int hrtimer_cb_pending(const struct hrtimer
*timer
)
481 return timer
->state
& HRTIMER_STATE_PENDING
;
485 * Remove a timer from the callback pending list
487 static inline void hrtimer_remove_cb_pending(struct hrtimer
*timer
)
489 list_del_init(&timer
->cb_entry
);
493 * Initialize the high resolution related parts of cpu_base
495 static inline void hrtimer_init_hres(struct hrtimer_cpu_base
*base
)
497 base
->expires_next
.tv64
= KTIME_MAX
;
498 base
->hres_active
= 0;
499 INIT_LIST_HEAD(&base
->cb_pending
);
503 * Initialize the high resolution related parts of a hrtimer
505 static inline void hrtimer_init_timer_hres(struct hrtimer
*timer
)
507 INIT_LIST_HEAD(&timer
->cb_entry
);
511 * When High resolution timers are active, try to reprogram. Note, that in case
512 * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry
513 * check happens. The timer gets enqueued into the rbtree. The reprogramming
514 * and expiry check is done in the hrtimer_interrupt or in the softirq.
516 static inline int hrtimer_enqueue_reprogram(struct hrtimer
*timer
,
517 struct hrtimer_clock_base
*base
)
519 if (base
->cpu_base
->hres_active
&& hrtimer_reprogram(timer
, base
)) {
521 /* Timer is expired, act upon the callback mode */
522 switch(timer
->cb_mode
) {
523 case HRTIMER_CB_IRQSAFE_NO_RESTART
:
525 * We can call the callback from here. No restart
526 * happens, so no danger of recursion
528 BUG_ON(timer
->function(timer
) != HRTIMER_NORESTART
);
530 case HRTIMER_CB_IRQSAFE_NO_SOFTIRQ
:
532 * This is solely for the sched tick emulation with
533 * dynamic tick support to ensure that we do not
534 * restart the tick right on the edge and end up with
535 * the tick timer in the softirq ! The calling site
536 * takes care of this.
539 case HRTIMER_CB_IRQSAFE
:
540 case HRTIMER_CB_SOFTIRQ
:
542 * Move everything else into the softirq pending list !
544 list_add_tail(&timer
->cb_entry
,
545 &base
->cpu_base
->cb_pending
);
546 timer
->state
= HRTIMER_STATE_PENDING
;
547 raise_softirq(HRTIMER_SOFTIRQ
);
557 * Switch to high resolution mode
559 static int hrtimer_switch_to_hres(void)
561 int cpu
= smp_processor_id();
562 struct hrtimer_cpu_base
*base
= &per_cpu(hrtimer_bases
, cpu
);
565 if (base
->hres_active
)
568 local_irq_save(flags
);
570 if (tick_init_highres()) {
571 local_irq_restore(flags
);
572 printk(KERN_WARNING
"Could not switch to high resolution "
573 "mode on CPU %d\n", cpu
);
576 base
->hres_active
= 1;
577 base
->clock_base
[CLOCK_REALTIME
].resolution
= KTIME_HIGH_RES
;
578 base
->clock_base
[CLOCK_MONOTONIC
].resolution
= KTIME_HIGH_RES
;
580 tick_setup_sched_timer();
582 /* "Retrigger" the interrupt to get things going */
583 retrigger_next_event(NULL
);
584 local_irq_restore(flags
);
585 printk(KERN_INFO
"Switched to high resolution mode on CPU %d\n",
592 static inline int hrtimer_hres_active(void) { return 0; }
593 static inline int hrtimer_is_hres_enabled(void) { return 0; }
594 static inline int hrtimer_switch_to_hres(void) { return 0; }
595 static inline void hrtimer_force_reprogram(struct hrtimer_cpu_base
*base
) { }
596 static inline int hrtimer_enqueue_reprogram(struct hrtimer
*timer
,
597 struct hrtimer_clock_base
*base
)
601 static inline int hrtimer_cb_pending(struct hrtimer
*timer
) { return 0; }
602 static inline void hrtimer_remove_cb_pending(struct hrtimer
*timer
) { }
603 static inline void hrtimer_init_hres(struct hrtimer_cpu_base
*base
) { }
604 static inline void hrtimer_init_timer_hres(struct hrtimer
*timer
) { }
606 #endif /* CONFIG_HIGH_RES_TIMERS */
608 #ifdef CONFIG_TIMER_STATS
609 void __timer_stats_hrtimer_set_start_info(struct hrtimer
*timer
, void *addr
)
611 if (timer
->start_site
)
614 timer
->start_site
= addr
;
615 memcpy(timer
->start_comm
, current
->comm
, TASK_COMM_LEN
);
616 timer
->start_pid
= current
->pid
;
621 * Counterpart to lock_timer_base above:
624 void unlock_hrtimer_base(const struct hrtimer
*timer
, unsigned long *flags
)
626 spin_unlock_irqrestore(&timer
->base
->cpu_base
->lock
, *flags
);
630 * hrtimer_forward - forward the timer expiry
631 * @timer: hrtimer to forward
632 * @now: forward past this time
633 * @interval: the interval to forward
635 * Forward the timer expiry so it will expire in the future.
636 * Returns the number of overruns.
639 hrtimer_forward(struct hrtimer
*timer
, ktime_t now
, ktime_t interval
)
641 unsigned long orun
= 1;
644 delta
= ktime_sub(now
, timer
->expires
);
649 if (interval
.tv64
< timer
->base
->resolution
.tv64
)
650 interval
.tv64
= timer
->base
->resolution
.tv64
;
652 if (unlikely(delta
.tv64
>= interval
.tv64
)) {
653 s64 incr
= ktime_to_ns(interval
);
655 orun
= ktime_divns(delta
, incr
);
656 timer
->expires
= ktime_add_ns(timer
->expires
, incr
* orun
);
657 if (timer
->expires
.tv64
> now
.tv64
)
660 * This (and the ktime_add() below) is the
661 * correction for exact:
665 timer
->expires
= ktime_add(timer
->expires
, interval
);
667 * Make sure, that the result did not wrap with a very large
670 if (timer
->expires
.tv64
< 0)
671 timer
->expires
= ktime_set(KTIME_SEC_MAX
, 0);
675 EXPORT_SYMBOL_GPL(hrtimer_forward
);
678 * enqueue_hrtimer - internal function to (re)start a timer
680 * The timer is inserted in expiry order. Insertion into the
681 * red black tree is O(log(n)). Must hold the base lock.
683 static void enqueue_hrtimer(struct hrtimer
*timer
,
684 struct hrtimer_clock_base
*base
, int reprogram
)
686 struct rb_node
**link
= &base
->active
.rb_node
;
687 struct rb_node
*parent
= NULL
;
688 struct hrtimer
*entry
;
692 * Find the right place in the rbtree:
696 entry
= rb_entry(parent
, struct hrtimer
, node
);
698 * We dont care about collisions. Nodes with
699 * the same expiry time stay together.
701 if (timer
->expires
.tv64
< entry
->expires
.tv64
) {
702 link
= &(*link
)->rb_left
;
704 link
= &(*link
)->rb_right
;
710 * Insert the timer to the rbtree and check whether it
711 * replaces the first pending timer
715 * Reprogram the clock event device. When the timer is already
716 * expired hrtimer_enqueue_reprogram has either called the
717 * callback or added it to the pending list and raised the
720 * This is a NOP for !HIGHRES
722 if (reprogram
&& hrtimer_enqueue_reprogram(timer
, base
))
725 base
->first
= &timer
->node
;
728 rb_link_node(&timer
->node
, parent
, link
);
729 rb_insert_color(&timer
->node
, &base
->active
);
731 * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
732 * state of a possibly running callback.
734 timer
->state
|= HRTIMER_STATE_ENQUEUED
;
738 * __remove_hrtimer - internal function to remove a timer
740 * Caller must hold the base lock.
742 * High resolution timer mode reprograms the clock event device when the
743 * timer is the one which expires next. The caller can disable this by setting
744 * reprogram to zero. This is useful, when the context does a reprogramming
745 * anyway (e.g. timer interrupt)
747 static void __remove_hrtimer(struct hrtimer
*timer
,
748 struct hrtimer_clock_base
*base
,
749 unsigned long newstate
, int reprogram
)
751 /* High res. callback list. NOP for !HIGHRES */
752 if (hrtimer_cb_pending(timer
))
753 hrtimer_remove_cb_pending(timer
);
756 * Remove the timer from the rbtree and replace the
757 * first entry pointer if necessary.
759 if (base
->first
== &timer
->node
) {
760 base
->first
= rb_next(&timer
->node
);
761 /* Reprogram the clock event device. if enabled */
762 if (reprogram
&& hrtimer_hres_active())
763 hrtimer_force_reprogram(base
->cpu_base
);
765 rb_erase(&timer
->node
, &base
->active
);
767 timer
->state
= newstate
;
771 * remove hrtimer, called with base lock held
774 remove_hrtimer(struct hrtimer
*timer
, struct hrtimer_clock_base
*base
)
776 if (hrtimer_is_queued(timer
)) {
780 * Remove the timer and force reprogramming when high
781 * resolution mode is active and the timer is on the current
782 * CPU. If we remove a timer on another CPU, reprogramming is
783 * skipped. The interrupt event on this CPU is fired and
784 * reprogramming happens in the interrupt handler. This is a
785 * rare case and less expensive than a smp call.
787 timer_stats_hrtimer_clear_start_info(timer
);
788 reprogram
= base
->cpu_base
== &__get_cpu_var(hrtimer_bases
);
789 __remove_hrtimer(timer
, base
, HRTIMER_STATE_INACTIVE
,
797 * hrtimer_start - (re)start an relative timer on the current CPU
798 * @timer: the timer to be added
800 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
804 * 1 when the timer was active
807 hrtimer_start(struct hrtimer
*timer
, ktime_t tim
, const enum hrtimer_mode mode
)
809 struct hrtimer_clock_base
*base
, *new_base
;
813 base
= lock_hrtimer_base(timer
, &flags
);
815 /* Remove an active timer from the queue: */
816 ret
= remove_hrtimer(timer
, base
);
818 /* Switch the timer base, if necessary: */
819 new_base
= switch_hrtimer_base(timer
, base
);
821 if (mode
== HRTIMER_MODE_REL
) {
822 tim
= ktime_add(tim
, new_base
->get_time());
824 * CONFIG_TIME_LOW_RES is a temporary way for architectures
825 * to signal that they simply return xtime in
826 * do_gettimeoffset(). In this case we want to round up by
827 * resolution when starting a relative timer, to avoid short
828 * timeouts. This will go away with the GTOD framework.
830 #ifdef CONFIG_TIME_LOW_RES
831 tim
= ktime_add(tim
, base
->resolution
);
834 timer
->expires
= tim
;
836 timer_stats_hrtimer_set_start_info(timer
);
839 * Only allow reprogramming if the new base is on this CPU.
840 * (it might still be on another CPU if the timer was pending)
842 enqueue_hrtimer(timer
, new_base
,
843 new_base
->cpu_base
== &__get_cpu_var(hrtimer_bases
));
845 unlock_hrtimer_base(timer
, &flags
);
849 EXPORT_SYMBOL_GPL(hrtimer_start
);
852 * hrtimer_try_to_cancel - try to deactivate a timer
853 * @timer: hrtimer to stop
856 * 0 when the timer was not active
857 * 1 when the timer was active
858 * -1 when the timer is currently excuting the callback function and
861 int hrtimer_try_to_cancel(struct hrtimer
*timer
)
863 struct hrtimer_clock_base
*base
;
867 base
= lock_hrtimer_base(timer
, &flags
);
869 if (!hrtimer_callback_running(timer
))
870 ret
= remove_hrtimer(timer
, base
);
872 unlock_hrtimer_base(timer
, &flags
);
877 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel
);
880 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
881 * @timer: the timer to be cancelled
884 * 0 when the timer was not active
885 * 1 when the timer was active
887 int hrtimer_cancel(struct hrtimer
*timer
)
890 int ret
= hrtimer_try_to_cancel(timer
);
897 EXPORT_SYMBOL_GPL(hrtimer_cancel
);
900 * hrtimer_get_remaining - get remaining time for the timer
901 * @timer: the timer to read
903 ktime_t
hrtimer_get_remaining(const struct hrtimer
*timer
)
905 struct hrtimer_clock_base
*base
;
909 base
= lock_hrtimer_base(timer
, &flags
);
910 rem
= ktime_sub(timer
->expires
, base
->get_time());
911 unlock_hrtimer_base(timer
, &flags
);
915 EXPORT_SYMBOL_GPL(hrtimer_get_remaining
);
917 #if defined(CONFIG_NO_IDLE_HZ) || defined(CONFIG_NO_HZ)
919 * hrtimer_get_next_event - get the time until next expiry event
921 * Returns the delta to the next expiry event or KTIME_MAX if no timer
924 ktime_t
hrtimer_get_next_event(void)
926 struct hrtimer_cpu_base
*cpu_base
= &__get_cpu_var(hrtimer_bases
);
927 struct hrtimer_clock_base
*base
= cpu_base
->clock_base
;
928 ktime_t delta
, mindelta
= { .tv64
= KTIME_MAX
};
932 spin_lock_irqsave(&cpu_base
->lock
, flags
);
934 if (!hrtimer_hres_active()) {
935 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++, base
++) {
936 struct hrtimer
*timer
;
941 timer
= rb_entry(base
->first
, struct hrtimer
, node
);
942 delta
.tv64
= timer
->expires
.tv64
;
943 delta
= ktime_sub(delta
, base
->get_time());
944 if (delta
.tv64
< mindelta
.tv64
)
945 mindelta
.tv64
= delta
.tv64
;
949 spin_unlock_irqrestore(&cpu_base
->lock
, flags
);
951 if (mindelta
.tv64
< 0)
958 * hrtimer_init - initialize a timer to the given clock
959 * @timer: the timer to be initialized
960 * @clock_id: the clock to be used
961 * @mode: timer mode abs/rel
963 void hrtimer_init(struct hrtimer
*timer
, clockid_t clock_id
,
964 enum hrtimer_mode mode
)
966 struct hrtimer_cpu_base
*cpu_base
;
968 memset(timer
, 0, sizeof(struct hrtimer
));
970 cpu_base
= &__raw_get_cpu_var(hrtimer_bases
);
972 if (clock_id
== CLOCK_REALTIME
&& mode
!= HRTIMER_MODE_ABS
)
973 clock_id
= CLOCK_MONOTONIC
;
975 timer
->base
= &cpu_base
->clock_base
[clock_id
];
976 hrtimer_init_timer_hres(timer
);
978 #ifdef CONFIG_TIMER_STATS
979 timer
->start_site
= NULL
;
980 timer
->start_pid
= -1;
981 memset(timer
->start_comm
, 0, TASK_COMM_LEN
);
984 EXPORT_SYMBOL_GPL(hrtimer_init
);
987 * hrtimer_get_res - get the timer resolution for a clock
988 * @which_clock: which clock to query
989 * @tp: pointer to timespec variable to store the resolution
991 * Store the resolution of the clock selected by @which_clock in the
992 * variable pointed to by @tp.
994 int hrtimer_get_res(const clockid_t which_clock
, struct timespec
*tp
)
996 struct hrtimer_cpu_base
*cpu_base
;
998 cpu_base
= &__raw_get_cpu_var(hrtimer_bases
);
999 *tp
= ktime_to_timespec(cpu_base
->clock_base
[which_clock
].resolution
);
1003 EXPORT_SYMBOL_GPL(hrtimer_get_res
);
1005 #ifdef CONFIG_HIGH_RES_TIMERS
1008 * High resolution timer interrupt
1009 * Called with interrupts disabled
1011 void hrtimer_interrupt(struct clock_event_device
*dev
)
1013 struct hrtimer_cpu_base
*cpu_base
= &__get_cpu_var(hrtimer_bases
);
1014 struct hrtimer_clock_base
*base
;
1015 ktime_t expires_next
, now
;
1018 BUG_ON(!cpu_base
->hres_active
);
1019 cpu_base
->nr_events
++;
1020 dev
->next_event
.tv64
= KTIME_MAX
;
1025 expires_next
.tv64
= KTIME_MAX
;
1027 base
= cpu_base
->clock_base
;
1029 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++) {
1031 struct rb_node
*node
;
1033 spin_lock(&cpu_base
->lock
);
1035 basenow
= ktime_add(now
, base
->offset
);
1037 while ((node
= base
->first
)) {
1038 struct hrtimer
*timer
;
1040 timer
= rb_entry(node
, struct hrtimer
, node
);
1042 if (basenow
.tv64
< timer
->expires
.tv64
) {
1045 expires
= ktime_sub(timer
->expires
,
1047 if (expires
.tv64
< expires_next
.tv64
)
1048 expires_next
= expires
;
1052 /* Move softirq callbacks to the pending list */
1053 if (timer
->cb_mode
== HRTIMER_CB_SOFTIRQ
) {
1054 __remove_hrtimer(timer
, base
,
1055 HRTIMER_STATE_PENDING
, 0);
1056 list_add_tail(&timer
->cb_entry
,
1057 &base
->cpu_base
->cb_pending
);
1062 __remove_hrtimer(timer
, base
,
1063 HRTIMER_STATE_CALLBACK
, 0);
1064 timer_stats_account_hrtimer(timer
);
1067 * Note: We clear the CALLBACK bit after
1068 * enqueue_hrtimer to avoid reprogramming of
1069 * the event hardware. This happens at the end
1070 * of this function anyway.
1072 if (timer
->function(timer
) != HRTIMER_NORESTART
) {
1073 BUG_ON(timer
->state
!= HRTIMER_STATE_CALLBACK
);
1074 enqueue_hrtimer(timer
, base
, 0);
1076 timer
->state
&= ~HRTIMER_STATE_CALLBACK
;
1078 spin_unlock(&cpu_base
->lock
);
1082 cpu_base
->expires_next
= expires_next
;
1084 /* Reprogramming necessary ? */
1085 if (expires_next
.tv64
!= KTIME_MAX
) {
1086 if (tick_program_event(expires_next
, 0))
1090 /* Raise softirq ? */
1092 raise_softirq(HRTIMER_SOFTIRQ
);
1095 static void run_hrtimer_softirq(struct softirq_action
*h
)
1097 struct hrtimer_cpu_base
*cpu_base
= &__get_cpu_var(hrtimer_bases
);
1099 spin_lock_irq(&cpu_base
->lock
);
1101 while (!list_empty(&cpu_base
->cb_pending
)) {
1102 enum hrtimer_restart (*fn
)(struct hrtimer
*);
1103 struct hrtimer
*timer
;
1106 timer
= list_entry(cpu_base
->cb_pending
.next
,
1107 struct hrtimer
, cb_entry
);
1109 timer_stats_account_hrtimer(timer
);
1111 fn
= timer
->function
;
1112 __remove_hrtimer(timer
, timer
->base
, HRTIMER_STATE_CALLBACK
, 0);
1113 spin_unlock_irq(&cpu_base
->lock
);
1115 restart
= fn(timer
);
1117 spin_lock_irq(&cpu_base
->lock
);
1119 timer
->state
&= ~HRTIMER_STATE_CALLBACK
;
1120 if (restart
== HRTIMER_RESTART
) {
1121 BUG_ON(hrtimer_active(timer
));
1123 * Enqueue the timer, allow reprogramming of the event
1126 enqueue_hrtimer(timer
, timer
->base
, 1);
1127 } else if (hrtimer_active(timer
)) {
1129 * If the timer was rearmed on another CPU, reprogram
1132 if (timer
->base
->first
== &timer
->node
)
1133 hrtimer_reprogram(timer
, timer
->base
);
1136 spin_unlock_irq(&cpu_base
->lock
);
1139 #endif /* CONFIG_HIGH_RES_TIMERS */
1142 * Expire the per base hrtimer-queue:
1144 static inline void run_hrtimer_queue(struct hrtimer_cpu_base
*cpu_base
,
1147 struct rb_node
*node
;
1148 struct hrtimer_clock_base
*base
= &cpu_base
->clock_base
[index
];
1153 if (base
->get_softirq_time
)
1154 base
->softirq_time
= base
->get_softirq_time();
1156 spin_lock_irq(&cpu_base
->lock
);
1158 while ((node
= base
->first
)) {
1159 struct hrtimer
*timer
;
1160 enum hrtimer_restart (*fn
)(struct hrtimer
*);
1163 timer
= rb_entry(node
, struct hrtimer
, node
);
1164 if (base
->softirq_time
.tv64
<= timer
->expires
.tv64
)
1167 #ifdef CONFIG_HIGH_RES_TIMERS
1168 WARN_ON_ONCE(timer
->cb_mode
== HRTIMER_CB_IRQSAFE_NO_SOFTIRQ
);
1170 timer_stats_account_hrtimer(timer
);
1172 fn
= timer
->function
;
1173 __remove_hrtimer(timer
, base
, HRTIMER_STATE_CALLBACK
, 0);
1174 spin_unlock_irq(&cpu_base
->lock
);
1176 restart
= fn(timer
);
1178 spin_lock_irq(&cpu_base
->lock
);
1180 timer
->state
&= ~HRTIMER_STATE_CALLBACK
;
1181 if (restart
!= HRTIMER_NORESTART
) {
1182 BUG_ON(hrtimer_active(timer
));
1183 enqueue_hrtimer(timer
, base
, 0);
1186 spin_unlock_irq(&cpu_base
->lock
);
1190 * Called from timer softirq every jiffy, expire hrtimers:
1192 * For HRT its the fall back code to run the softirq in the timer
1193 * softirq context in case the hrtimer initialization failed or has
1194 * not been done yet.
1196 void hrtimer_run_queues(void)
1198 struct hrtimer_cpu_base
*cpu_base
= &__get_cpu_var(hrtimer_bases
);
1201 if (hrtimer_hres_active())
1205 * This _is_ ugly: We have to check in the softirq context,
1206 * whether we can switch to highres and / or nohz mode. The
1207 * clocksource switch happens in the timer interrupt with
1208 * xtime_lock held. Notification from there only sets the
1209 * check bit in the tick_oneshot code, otherwise we might
1210 * deadlock vs. xtime_lock.
1212 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1213 if (hrtimer_switch_to_hres())
1216 hrtimer_get_softirq_time(cpu_base
);
1218 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++)
1219 run_hrtimer_queue(cpu_base
, i
);
1223 * Sleep related functions:
1225 static enum hrtimer_restart
hrtimer_wakeup(struct hrtimer
*timer
)
1227 struct hrtimer_sleeper
*t
=
1228 container_of(timer
, struct hrtimer_sleeper
, timer
);
1229 struct task_struct
*task
= t
->task
;
1233 wake_up_process(task
);
1235 return HRTIMER_NORESTART
;
1238 void hrtimer_init_sleeper(struct hrtimer_sleeper
*sl
, struct task_struct
*task
)
1240 sl
->timer
.function
= hrtimer_wakeup
;
1242 #ifdef CONFIG_HIGH_RES_TIMERS
1243 sl
->timer
.cb_mode
= HRTIMER_CB_IRQSAFE_NO_RESTART
;
1247 static int __sched
do_nanosleep(struct hrtimer_sleeper
*t
, enum hrtimer_mode mode
)
1249 hrtimer_init_sleeper(t
, current
);
1252 set_current_state(TASK_INTERRUPTIBLE
);
1253 hrtimer_start(&t
->timer
, t
->timer
.expires
, mode
);
1255 if (likely(t
->task
))
1258 hrtimer_cancel(&t
->timer
);
1259 mode
= HRTIMER_MODE_ABS
;
1261 } while (t
->task
&& !signal_pending(current
));
1263 return t
->task
== NULL
;
1266 long __sched
hrtimer_nanosleep_restart(struct restart_block
*restart
)
1268 struct hrtimer_sleeper t
;
1269 struct timespec __user
*rmtp
;
1273 restart
->fn
= do_no_restart_syscall
;
1275 hrtimer_init(&t
.timer
, restart
->arg0
, HRTIMER_MODE_ABS
);
1276 t
.timer
.expires
.tv64
= ((u64
)restart
->arg3
<< 32) | (u64
) restart
->arg2
;
1278 if (do_nanosleep(&t
, HRTIMER_MODE_ABS
))
1281 rmtp
= (struct timespec __user
*) restart
->arg1
;
1283 time
= ktime_sub(t
.timer
.expires
, t
.timer
.base
->get_time());
1286 tu
= ktime_to_timespec(time
);
1287 if (copy_to_user(rmtp
, &tu
, sizeof(tu
)))
1291 restart
->fn
= hrtimer_nanosleep_restart
;
1293 /* The other values in restart are already filled in */
1294 return -ERESTART_RESTARTBLOCK
;
1297 long hrtimer_nanosleep(struct timespec
*rqtp
, struct timespec __user
*rmtp
,
1298 const enum hrtimer_mode mode
, const clockid_t clockid
)
1300 struct restart_block
*restart
;
1301 struct hrtimer_sleeper t
;
1305 hrtimer_init(&t
.timer
, clockid
, mode
);
1306 t
.timer
.expires
= timespec_to_ktime(*rqtp
);
1307 if (do_nanosleep(&t
, mode
))
1310 /* Absolute timers do not update the rmtp value and restart: */
1311 if (mode
== HRTIMER_MODE_ABS
)
1312 return -ERESTARTNOHAND
;
1315 rem
= ktime_sub(t
.timer
.expires
, t
.timer
.base
->get_time());
1318 tu
= ktime_to_timespec(rem
);
1319 if (copy_to_user(rmtp
, &tu
, sizeof(tu
)))
1323 restart
= ¤t_thread_info()->restart_block
;
1324 restart
->fn
= hrtimer_nanosleep_restart
;
1325 restart
->arg0
= (unsigned long) t
.timer
.base
->index
;
1326 restart
->arg1
= (unsigned long) rmtp
;
1327 restart
->arg2
= t
.timer
.expires
.tv64
& 0xFFFFFFFF;
1328 restart
->arg3
= t
.timer
.expires
.tv64
>> 32;
1330 return -ERESTART_RESTARTBLOCK
;
1334 sys_nanosleep(struct timespec __user
*rqtp
, struct timespec __user
*rmtp
)
1338 if (copy_from_user(&tu
, rqtp
, sizeof(tu
)))
1341 if (!timespec_valid(&tu
))
1344 return hrtimer_nanosleep(&tu
, rmtp
, HRTIMER_MODE_REL
, CLOCK_MONOTONIC
);
1348 * Functions related to boot-time initialization:
1350 static void __devinit
init_hrtimers_cpu(int cpu
)
1352 struct hrtimer_cpu_base
*cpu_base
= &per_cpu(hrtimer_bases
, cpu
);
1355 spin_lock_init(&cpu_base
->lock
);
1356 lockdep_set_class(&cpu_base
->lock
, &cpu_base
->lock_key
);
1358 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++)
1359 cpu_base
->clock_base
[i
].cpu_base
= cpu_base
;
1361 hrtimer_init_hres(cpu_base
);
1364 #ifdef CONFIG_HOTPLUG_CPU
1366 static void migrate_hrtimer_list(struct hrtimer_clock_base
*old_base
,
1367 struct hrtimer_clock_base
*new_base
)
1369 struct hrtimer
*timer
;
1370 struct rb_node
*node
;
1372 while ((node
= rb_first(&old_base
->active
))) {
1373 timer
= rb_entry(node
, struct hrtimer
, node
);
1374 BUG_ON(hrtimer_callback_running(timer
));
1375 __remove_hrtimer(timer
, old_base
, HRTIMER_STATE_INACTIVE
, 0);
1376 timer
->base
= new_base
;
1378 * Enqueue the timer. Allow reprogramming of the event device
1380 enqueue_hrtimer(timer
, new_base
, 1);
1384 static void migrate_hrtimers(int cpu
)
1386 struct hrtimer_cpu_base
*old_base
, *new_base
;
1389 BUG_ON(cpu_online(cpu
));
1390 old_base
= &per_cpu(hrtimer_bases
, cpu
);
1391 new_base
= &get_cpu_var(hrtimer_bases
);
1393 tick_cancel_sched_timer(cpu
);
1395 local_irq_disable();
1396 double_spin_lock(&new_base
->lock
, &old_base
->lock
,
1397 smp_processor_id() < cpu
);
1399 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++) {
1400 migrate_hrtimer_list(&old_base
->clock_base
[i
],
1401 &new_base
->clock_base
[i
]);
1404 double_spin_unlock(&new_base
->lock
, &old_base
->lock
,
1405 smp_processor_id() < cpu
);
1407 put_cpu_var(hrtimer_bases
);
1409 #endif /* CONFIG_HOTPLUG_CPU */
1411 static int __cpuinit
hrtimer_cpu_notify(struct notifier_block
*self
,
1412 unsigned long action
, void *hcpu
)
1414 unsigned int cpu
= (long)hcpu
;
1418 case CPU_UP_PREPARE
:
1419 case CPU_UP_PREPARE_FROZEN
:
1420 init_hrtimers_cpu(cpu
);
1423 #ifdef CONFIG_HOTPLUG_CPU
1425 case CPU_DEAD_FROZEN
:
1426 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD
, &cpu
);
1427 migrate_hrtimers(cpu
);
1438 static struct notifier_block __cpuinitdata hrtimers_nb
= {
1439 .notifier_call
= hrtimer_cpu_notify
,
1442 void __init
hrtimers_init(void)
1444 hrtimer_cpu_notify(&hrtimers_nb
, (unsigned long)CPU_UP_PREPARE
,
1445 (void *)(long)smp_processor_id());
1446 register_cpu_notifier(&hrtimers_nb
);
1447 #ifdef CONFIG_HIGH_RES_TIMERS
1448 open_softirq(HRTIMER_SOFTIRQ
, run_hrtimer_softirq
, NULL
);