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
);
144 xts
= current_kernel_time();
145 tom
= wall_to_monotonic
;
146 } while (read_seqretry(&xtime_lock
, seq
));
148 xtim
= timespec_to_ktime(xts
);
149 tomono
= timespec_to_ktime(tom
);
150 base
->clock_base
[CLOCK_REALTIME
].softirq_time
= xtim
;
151 base
->clock_base
[CLOCK_MONOTONIC
].softirq_time
=
152 ktime_add(xtim
, tomono
);
156 * Helper function to check, whether the timer is running the callback
159 static inline int hrtimer_callback_running(struct hrtimer
*timer
)
161 return timer
->state
& HRTIMER_STATE_CALLBACK
;
165 * Functions and macros which are different for UP/SMP systems are kept in a
171 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
172 * means that all timers which are tied to this base via timer->base are
173 * locked, and the base itself is locked too.
175 * So __run_timers/migrate_timers can safely modify all timers which could
176 * be found on the lists/queues.
178 * When the timer's base is locked, and the timer removed from list, it is
179 * possible to set timer->base = NULL and drop the lock: the timer remains
183 struct hrtimer_clock_base
*lock_hrtimer_base(const struct hrtimer
*timer
,
184 unsigned long *flags
)
186 struct hrtimer_clock_base
*base
;
190 if (likely(base
!= NULL
)) {
191 spin_lock_irqsave(&base
->cpu_base
->lock
, *flags
);
192 if (likely(base
== timer
->base
))
194 /* The timer has migrated to another CPU: */
195 spin_unlock_irqrestore(&base
->cpu_base
->lock
, *flags
);
202 * Switch the timer base to the current CPU when possible.
204 static inline struct hrtimer_clock_base
*
205 switch_hrtimer_base(struct hrtimer
*timer
, struct hrtimer_clock_base
*base
)
207 struct hrtimer_clock_base
*new_base
;
208 struct hrtimer_cpu_base
*new_cpu_base
;
210 new_cpu_base
= &__get_cpu_var(hrtimer_bases
);
211 new_base
= &new_cpu_base
->clock_base
[base
->index
];
213 if (base
!= new_base
) {
215 * We are trying to schedule the timer on the local CPU.
216 * However we can't change timer's base while it is running,
217 * so we keep it on the same CPU. No hassle vs. reprogramming
218 * the event source in the high resolution case. The softirq
219 * code will take care of this when the timer function has
220 * completed. There is no conflict as we hold the lock until
221 * the timer is enqueued.
223 if (unlikely(hrtimer_callback_running(timer
)))
226 /* See the comment in lock_timer_base() */
228 spin_unlock(&base
->cpu_base
->lock
);
229 spin_lock(&new_base
->cpu_base
->lock
);
230 timer
->base
= new_base
;
235 #else /* CONFIG_SMP */
237 static inline struct hrtimer_clock_base
*
238 lock_hrtimer_base(const struct hrtimer
*timer
, unsigned long *flags
)
240 struct hrtimer_clock_base
*base
= timer
->base
;
242 spin_lock_irqsave(&base
->cpu_base
->lock
, *flags
);
247 # define switch_hrtimer_base(t, b) (b)
249 #endif /* !CONFIG_SMP */
252 * Functions for the union type storage format of ktime_t which are
253 * too large for inlining:
255 #if BITS_PER_LONG < 64
256 # ifndef CONFIG_KTIME_SCALAR
258 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
260 * @nsec: the scalar nsec value to add
262 * Returns the sum of kt and nsec in ktime_t format
264 ktime_t
ktime_add_ns(const ktime_t kt
, u64 nsec
)
268 if (likely(nsec
< NSEC_PER_SEC
)) {
271 unsigned long rem
= do_div(nsec
, NSEC_PER_SEC
);
273 tmp
= ktime_set((long)nsec
, rem
);
276 return ktime_add(kt
, tmp
);
279 EXPORT_SYMBOL_GPL(ktime_add_ns
);
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 */
305 * Add two ktime values and do a safety check for overflow:
308 ktime_t
ktime_add_safe(const ktime_t lhs
, const ktime_t rhs
)
310 ktime_t res
= ktime_add(lhs
, rhs
);
313 * We use KTIME_SEC_MAX here, the maximum timeout which we can
314 * return to user space in a timespec:
316 if (res
.tv64
< 0 || res
.tv64
< lhs
.tv64
|| res
.tv64
< rhs
.tv64
)
317 res
= ktime_set(KTIME_SEC_MAX
, 0);
322 /* High resolution timer related functions */
323 #ifdef CONFIG_HIGH_RES_TIMERS
326 * High resolution timer enabled ?
328 static int hrtimer_hres_enabled __read_mostly
= 1;
331 * Enable / Disable high resolution mode
333 static int __init
setup_hrtimer_hres(char *str
)
335 if (!strcmp(str
, "off"))
336 hrtimer_hres_enabled
= 0;
337 else if (!strcmp(str
, "on"))
338 hrtimer_hres_enabled
= 1;
344 __setup("highres=", setup_hrtimer_hres
);
347 * hrtimer_high_res_enabled - query, if the highres mode is enabled
349 static inline int hrtimer_is_hres_enabled(void)
351 return hrtimer_hres_enabled
;
355 * Is the high resolution mode active ?
357 static inline int hrtimer_hres_active(void)
359 return __get_cpu_var(hrtimer_bases
).hres_active
;
363 * Reprogram the event source with checking both queues for the
365 * Called with interrupts disabled and base->lock held
367 static void hrtimer_force_reprogram(struct hrtimer_cpu_base
*cpu_base
)
370 struct hrtimer_clock_base
*base
= cpu_base
->clock_base
;
373 cpu_base
->expires_next
.tv64
= KTIME_MAX
;
375 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++, base
++) {
376 struct hrtimer
*timer
;
380 timer
= rb_entry(base
->first
, struct hrtimer
, node
);
381 expires
= ktime_sub(timer
->expires
, base
->offset
);
382 if (expires
.tv64
< cpu_base
->expires_next
.tv64
)
383 cpu_base
->expires_next
= expires
;
386 if (cpu_base
->expires_next
.tv64
!= KTIME_MAX
)
387 tick_program_event(cpu_base
->expires_next
, 1);
391 * Shared reprogramming for clock_realtime and clock_monotonic
393 * When a timer is enqueued and expires earlier than the already enqueued
394 * timers, we have to check, whether it expires earlier than the timer for
395 * which the clock event device was armed.
397 * Called with interrupts disabled and base->cpu_base.lock held
399 static int hrtimer_reprogram(struct hrtimer
*timer
,
400 struct hrtimer_clock_base
*base
)
402 ktime_t
*expires_next
= &__get_cpu_var(hrtimer_bases
).expires_next
;
403 ktime_t expires
= ktime_sub(timer
->expires
, base
->offset
);
407 * When the callback is running, we do not reprogram the clock event
408 * device. The timer callback is either running on a different CPU or
409 * the callback is executed in the hrtimer_interupt context. The
410 * reprogramming is handled either by the softirq, which called the
411 * callback or at the end of the hrtimer_interrupt.
413 if (hrtimer_callback_running(timer
))
416 if (expires
.tv64
>= expires_next
->tv64
)
420 * Clockevents returns -ETIME, when the event was in the past.
422 res
= tick_program_event(expires
, 0);
423 if (!IS_ERR_VALUE(res
))
424 *expires_next
= expires
;
430 * Retrigger next event is called after clock was set
432 * Called with interrupts disabled via on_each_cpu()
434 static void retrigger_next_event(void *arg
)
436 struct hrtimer_cpu_base
*base
;
437 struct timespec realtime_offset
;
440 if (!hrtimer_hres_active())
444 seq
= read_seqbegin(&xtime_lock
);
445 set_normalized_timespec(&realtime_offset
,
446 -wall_to_monotonic
.tv_sec
,
447 -wall_to_monotonic
.tv_nsec
);
448 } while (read_seqretry(&xtime_lock
, seq
));
450 base
= &__get_cpu_var(hrtimer_bases
);
452 /* Adjust CLOCK_REALTIME offset */
453 spin_lock(&base
->lock
);
454 base
->clock_base
[CLOCK_REALTIME
].offset
=
455 timespec_to_ktime(realtime_offset
);
457 hrtimer_force_reprogram(base
);
458 spin_unlock(&base
->lock
);
462 * Clock realtime was set
464 * Change the offset of the realtime clock vs. the monotonic
467 * We might have to reprogram the high resolution timer interrupt. On
468 * SMP we call the architecture specific code to retrigger _all_ high
469 * resolution timer interrupts. On UP we just disable interrupts and
470 * call the high resolution interrupt code.
472 void clock_was_set(void)
474 /* Retrigger the CPU local events everywhere */
475 on_each_cpu(retrigger_next_event
, NULL
, 0, 1);
479 * During resume we might have to reprogram the high resolution timer
480 * interrupt (on the local CPU):
482 void hres_timers_resume(void)
484 WARN_ON_ONCE(num_online_cpus() > 1);
486 /* Retrigger the CPU local events: */
487 retrigger_next_event(NULL
);
491 * Check, whether the timer is on the callback pending list
493 static inline int hrtimer_cb_pending(const struct hrtimer
*timer
)
495 return timer
->state
& HRTIMER_STATE_PENDING
;
499 * Remove a timer from the callback pending list
501 static inline void hrtimer_remove_cb_pending(struct hrtimer
*timer
)
503 list_del_init(&timer
->cb_entry
);
507 * Initialize the high resolution related parts of cpu_base
509 static inline void hrtimer_init_hres(struct hrtimer_cpu_base
*base
)
511 base
->expires_next
.tv64
= KTIME_MAX
;
512 base
->hres_active
= 0;
513 INIT_LIST_HEAD(&base
->cb_pending
);
517 * Initialize the high resolution related parts of a hrtimer
519 static inline void hrtimer_init_timer_hres(struct hrtimer
*timer
)
521 INIT_LIST_HEAD(&timer
->cb_entry
);
525 * When High resolution timers are active, try to reprogram. Note, that in case
526 * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry
527 * check happens. The timer gets enqueued into the rbtree. The reprogramming
528 * and expiry check is done in the hrtimer_interrupt or in the softirq.
530 static inline int hrtimer_enqueue_reprogram(struct hrtimer
*timer
,
531 struct hrtimer_clock_base
*base
)
533 if (base
->cpu_base
->hres_active
&& hrtimer_reprogram(timer
, base
)) {
535 /* Timer is expired, act upon the callback mode */
536 switch(timer
->cb_mode
) {
537 case HRTIMER_CB_IRQSAFE_NO_RESTART
:
539 * We can call the callback from here. No restart
540 * happens, so no danger of recursion
542 BUG_ON(timer
->function(timer
) != HRTIMER_NORESTART
);
544 case HRTIMER_CB_IRQSAFE_NO_SOFTIRQ
:
546 * This is solely for the sched tick emulation with
547 * dynamic tick support to ensure that we do not
548 * restart the tick right on the edge and end up with
549 * the tick timer in the softirq ! The calling site
550 * takes care of this.
553 case HRTIMER_CB_IRQSAFE
:
554 case HRTIMER_CB_SOFTIRQ
:
556 * Move everything else into the softirq pending list !
558 list_add_tail(&timer
->cb_entry
,
559 &base
->cpu_base
->cb_pending
);
560 timer
->state
= HRTIMER_STATE_PENDING
;
561 raise_softirq(HRTIMER_SOFTIRQ
);
571 * Switch to high resolution mode
573 static int hrtimer_switch_to_hres(void)
575 int cpu
= smp_processor_id();
576 struct hrtimer_cpu_base
*base
= &per_cpu(hrtimer_bases
, cpu
);
579 if (base
->hres_active
)
582 local_irq_save(flags
);
584 if (tick_init_highres()) {
585 local_irq_restore(flags
);
586 printk(KERN_WARNING
"Could not switch to high resolution "
587 "mode on CPU %d\n", cpu
);
590 base
->hres_active
= 1;
591 base
->clock_base
[CLOCK_REALTIME
].resolution
= KTIME_HIGH_RES
;
592 base
->clock_base
[CLOCK_MONOTONIC
].resolution
= KTIME_HIGH_RES
;
594 tick_setup_sched_timer();
596 /* "Retrigger" the interrupt to get things going */
597 retrigger_next_event(NULL
);
598 local_irq_restore(flags
);
599 printk(KERN_INFO
"Switched to high resolution mode on CPU %d\n",
606 static inline int hrtimer_hres_active(void) { return 0; }
607 static inline int hrtimer_is_hres_enabled(void) { return 0; }
608 static inline int hrtimer_switch_to_hres(void) { return 0; }
609 static inline void hrtimer_force_reprogram(struct hrtimer_cpu_base
*base
) { }
610 static inline int hrtimer_enqueue_reprogram(struct hrtimer
*timer
,
611 struct hrtimer_clock_base
*base
)
615 static inline int hrtimer_cb_pending(struct hrtimer
*timer
) { return 0; }
616 static inline void hrtimer_remove_cb_pending(struct hrtimer
*timer
) { }
617 static inline void hrtimer_init_hres(struct hrtimer_cpu_base
*base
) { }
618 static inline void hrtimer_init_timer_hres(struct hrtimer
*timer
) { }
620 #endif /* CONFIG_HIGH_RES_TIMERS */
622 #ifdef CONFIG_TIMER_STATS
623 void __timer_stats_hrtimer_set_start_info(struct hrtimer
*timer
, void *addr
)
625 if (timer
->start_site
)
628 timer
->start_site
= addr
;
629 memcpy(timer
->start_comm
, current
->comm
, TASK_COMM_LEN
);
630 timer
->start_pid
= current
->pid
;
635 * Counterpart to lock_timer_base above:
638 void unlock_hrtimer_base(const struct hrtimer
*timer
, unsigned long *flags
)
640 spin_unlock_irqrestore(&timer
->base
->cpu_base
->lock
, *flags
);
644 * hrtimer_forward - forward the timer expiry
645 * @timer: hrtimer to forward
646 * @now: forward past this time
647 * @interval: the interval to forward
649 * Forward the timer expiry so it will expire in the future.
650 * Returns the number of overruns.
653 hrtimer_forward(struct hrtimer
*timer
, ktime_t now
, ktime_t interval
)
655 unsigned long orun
= 1;
658 delta
= ktime_sub(now
, timer
->expires
);
663 if (interval
.tv64
< timer
->base
->resolution
.tv64
)
664 interval
.tv64
= timer
->base
->resolution
.tv64
;
666 if (unlikely(delta
.tv64
>= interval
.tv64
)) {
667 s64 incr
= ktime_to_ns(interval
);
669 orun
= ktime_divns(delta
, incr
);
670 timer
->expires
= ktime_add_ns(timer
->expires
, incr
* orun
);
671 if (timer
->expires
.tv64
> now
.tv64
)
674 * This (and the ktime_add() below) is the
675 * correction for exact:
679 timer
->expires
= ktime_add_safe(timer
->expires
, interval
);
683 EXPORT_SYMBOL_GPL(hrtimer_forward
);
686 * enqueue_hrtimer - internal function to (re)start a timer
688 * The timer is inserted in expiry order. Insertion into the
689 * red black tree is O(log(n)). Must hold the base lock.
691 static void enqueue_hrtimer(struct hrtimer
*timer
,
692 struct hrtimer_clock_base
*base
, int reprogram
)
694 struct rb_node
**link
= &base
->active
.rb_node
;
695 struct rb_node
*parent
= NULL
;
696 struct hrtimer
*entry
;
700 * Find the right place in the rbtree:
704 entry
= rb_entry(parent
, struct hrtimer
, node
);
706 * We dont care about collisions. Nodes with
707 * the same expiry time stay together.
709 if (timer
->expires
.tv64
< entry
->expires
.tv64
) {
710 link
= &(*link
)->rb_left
;
712 link
= &(*link
)->rb_right
;
718 * Insert the timer to the rbtree and check whether it
719 * replaces the first pending timer
723 * Reprogram the clock event device. When the timer is already
724 * expired hrtimer_enqueue_reprogram has either called the
725 * callback or added it to the pending list and raised the
728 * This is a NOP for !HIGHRES
730 if (reprogram
&& hrtimer_enqueue_reprogram(timer
, base
))
733 base
->first
= &timer
->node
;
736 rb_link_node(&timer
->node
, parent
, link
);
737 rb_insert_color(&timer
->node
, &base
->active
);
739 * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
740 * state of a possibly running callback.
742 timer
->state
|= HRTIMER_STATE_ENQUEUED
;
746 * __remove_hrtimer - internal function to remove a timer
748 * Caller must hold the base lock.
750 * High resolution timer mode reprograms the clock event device when the
751 * timer is the one which expires next. The caller can disable this by setting
752 * reprogram to zero. This is useful, when the context does a reprogramming
753 * anyway (e.g. timer interrupt)
755 static void __remove_hrtimer(struct hrtimer
*timer
,
756 struct hrtimer_clock_base
*base
,
757 unsigned long newstate
, int reprogram
)
759 /* High res. callback list. NOP for !HIGHRES */
760 if (hrtimer_cb_pending(timer
))
761 hrtimer_remove_cb_pending(timer
);
764 * Remove the timer from the rbtree and replace the
765 * first entry pointer if necessary.
767 if (base
->first
== &timer
->node
) {
768 base
->first
= rb_next(&timer
->node
);
769 /* Reprogram the clock event device. if enabled */
770 if (reprogram
&& hrtimer_hres_active())
771 hrtimer_force_reprogram(base
->cpu_base
);
773 rb_erase(&timer
->node
, &base
->active
);
775 timer
->state
= newstate
;
779 * remove hrtimer, called with base lock held
782 remove_hrtimer(struct hrtimer
*timer
, struct hrtimer_clock_base
*base
)
784 if (hrtimer_is_queued(timer
)) {
788 * Remove the timer and force reprogramming when high
789 * resolution mode is active and the timer is on the current
790 * CPU. If we remove a timer on another CPU, reprogramming is
791 * skipped. The interrupt event on this CPU is fired and
792 * reprogramming happens in the interrupt handler. This is a
793 * rare case and less expensive than a smp call.
795 timer_stats_hrtimer_clear_start_info(timer
);
796 reprogram
= base
->cpu_base
== &__get_cpu_var(hrtimer_bases
);
797 __remove_hrtimer(timer
, base
, HRTIMER_STATE_INACTIVE
,
805 * hrtimer_start - (re)start an relative timer on the current CPU
806 * @timer: the timer to be added
808 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
812 * 1 when the timer was active
815 hrtimer_start(struct hrtimer
*timer
, ktime_t tim
, const enum hrtimer_mode mode
)
817 struct hrtimer_clock_base
*base
, *new_base
;
821 base
= lock_hrtimer_base(timer
, &flags
);
823 /* Remove an active timer from the queue: */
824 ret
= remove_hrtimer(timer
, base
);
826 /* Switch the timer base, if necessary: */
827 new_base
= switch_hrtimer_base(timer
, base
);
829 if (mode
== HRTIMER_MODE_REL
) {
830 tim
= ktime_add_safe(tim
, new_base
->get_time());
832 * CONFIG_TIME_LOW_RES is a temporary way for architectures
833 * to signal that they simply return xtime in
834 * do_gettimeoffset(). In this case we want to round up by
835 * resolution when starting a relative timer, to avoid short
836 * timeouts. This will go away with the GTOD framework.
838 #ifdef CONFIG_TIME_LOW_RES
839 tim
= ktime_add_safe(tim
, base
->resolution
);
842 timer
->expires
= tim
;
844 timer_stats_hrtimer_set_start_info(timer
);
847 * Only allow reprogramming if the new base is on this CPU.
848 * (it might still be on another CPU if the timer was pending)
850 enqueue_hrtimer(timer
, new_base
,
851 new_base
->cpu_base
== &__get_cpu_var(hrtimer_bases
));
853 unlock_hrtimer_base(timer
, &flags
);
857 EXPORT_SYMBOL_GPL(hrtimer_start
);
860 * hrtimer_try_to_cancel - try to deactivate a timer
861 * @timer: hrtimer to stop
864 * 0 when the timer was not active
865 * 1 when the timer was active
866 * -1 when the timer is currently excuting the callback function and
869 int hrtimer_try_to_cancel(struct hrtimer
*timer
)
871 struct hrtimer_clock_base
*base
;
875 base
= lock_hrtimer_base(timer
, &flags
);
877 if (!hrtimer_callback_running(timer
))
878 ret
= remove_hrtimer(timer
, base
);
880 unlock_hrtimer_base(timer
, &flags
);
885 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel
);
888 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
889 * @timer: the timer to be cancelled
892 * 0 when the timer was not active
893 * 1 when the timer was active
895 int hrtimer_cancel(struct hrtimer
*timer
)
898 int ret
= hrtimer_try_to_cancel(timer
);
905 EXPORT_SYMBOL_GPL(hrtimer_cancel
);
908 * hrtimer_get_remaining - get remaining time for the timer
909 * @timer: the timer to read
911 ktime_t
hrtimer_get_remaining(const struct hrtimer
*timer
)
913 struct hrtimer_clock_base
*base
;
917 base
= lock_hrtimer_base(timer
, &flags
);
918 rem
= ktime_sub(timer
->expires
, base
->get_time());
919 unlock_hrtimer_base(timer
, &flags
);
923 EXPORT_SYMBOL_GPL(hrtimer_get_remaining
);
925 #if defined(CONFIG_NO_IDLE_HZ) || defined(CONFIG_NO_HZ)
927 * hrtimer_get_next_event - get the time until next expiry event
929 * Returns the delta to the next expiry event or KTIME_MAX if no timer
932 ktime_t
hrtimer_get_next_event(void)
934 struct hrtimer_cpu_base
*cpu_base
= &__get_cpu_var(hrtimer_bases
);
935 struct hrtimer_clock_base
*base
= cpu_base
->clock_base
;
936 ktime_t delta
, mindelta
= { .tv64
= KTIME_MAX
};
940 spin_lock_irqsave(&cpu_base
->lock
, flags
);
942 if (!hrtimer_hres_active()) {
943 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++, base
++) {
944 struct hrtimer
*timer
;
949 timer
= rb_entry(base
->first
, struct hrtimer
, node
);
950 delta
.tv64
= timer
->expires
.tv64
;
951 delta
= ktime_sub(delta
, base
->get_time());
952 if (delta
.tv64
< mindelta
.tv64
)
953 mindelta
.tv64
= delta
.tv64
;
957 spin_unlock_irqrestore(&cpu_base
->lock
, flags
);
959 if (mindelta
.tv64
< 0)
966 * hrtimer_init - initialize a timer to the given clock
967 * @timer: the timer to be initialized
968 * @clock_id: the clock to be used
969 * @mode: timer mode abs/rel
971 void hrtimer_init(struct hrtimer
*timer
, clockid_t clock_id
,
972 enum hrtimer_mode mode
)
974 struct hrtimer_cpu_base
*cpu_base
;
976 memset(timer
, 0, sizeof(struct hrtimer
));
978 cpu_base
= &__raw_get_cpu_var(hrtimer_bases
);
980 if (clock_id
== CLOCK_REALTIME
&& mode
!= HRTIMER_MODE_ABS
)
981 clock_id
= CLOCK_MONOTONIC
;
983 timer
->base
= &cpu_base
->clock_base
[clock_id
];
984 hrtimer_init_timer_hres(timer
);
986 #ifdef CONFIG_TIMER_STATS
987 timer
->start_site
= NULL
;
988 timer
->start_pid
= -1;
989 memset(timer
->start_comm
, 0, TASK_COMM_LEN
);
992 EXPORT_SYMBOL_GPL(hrtimer_init
);
995 * hrtimer_get_res - get the timer resolution for a clock
996 * @which_clock: which clock to query
997 * @tp: pointer to timespec variable to store the resolution
999 * Store the resolution of the clock selected by @which_clock in the
1000 * variable pointed to by @tp.
1002 int hrtimer_get_res(const clockid_t which_clock
, struct timespec
*tp
)
1004 struct hrtimer_cpu_base
*cpu_base
;
1006 cpu_base
= &__raw_get_cpu_var(hrtimer_bases
);
1007 *tp
= ktime_to_timespec(cpu_base
->clock_base
[which_clock
].resolution
);
1011 EXPORT_SYMBOL_GPL(hrtimer_get_res
);
1013 #ifdef CONFIG_HIGH_RES_TIMERS
1016 * High resolution timer interrupt
1017 * Called with interrupts disabled
1019 void hrtimer_interrupt(struct clock_event_device
*dev
)
1021 struct hrtimer_cpu_base
*cpu_base
= &__get_cpu_var(hrtimer_bases
);
1022 struct hrtimer_clock_base
*base
;
1023 ktime_t expires_next
, now
;
1026 BUG_ON(!cpu_base
->hres_active
);
1027 cpu_base
->nr_events
++;
1028 dev
->next_event
.tv64
= KTIME_MAX
;
1033 expires_next
.tv64
= KTIME_MAX
;
1035 base
= cpu_base
->clock_base
;
1037 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++) {
1039 struct rb_node
*node
;
1041 spin_lock(&cpu_base
->lock
);
1043 basenow
= ktime_add(now
, base
->offset
);
1045 while ((node
= base
->first
)) {
1046 struct hrtimer
*timer
;
1048 timer
= rb_entry(node
, struct hrtimer
, node
);
1050 if (basenow
.tv64
< timer
->expires
.tv64
) {
1053 expires
= ktime_sub(timer
->expires
,
1055 if (expires
.tv64
< expires_next
.tv64
)
1056 expires_next
= expires
;
1060 /* Move softirq callbacks to the pending list */
1061 if (timer
->cb_mode
== HRTIMER_CB_SOFTIRQ
) {
1062 __remove_hrtimer(timer
, base
,
1063 HRTIMER_STATE_PENDING
, 0);
1064 list_add_tail(&timer
->cb_entry
,
1065 &base
->cpu_base
->cb_pending
);
1070 __remove_hrtimer(timer
, base
,
1071 HRTIMER_STATE_CALLBACK
, 0);
1072 timer_stats_account_hrtimer(timer
);
1075 * Note: We clear the CALLBACK bit after
1076 * enqueue_hrtimer to avoid reprogramming of
1077 * the event hardware. This happens at the end
1078 * of this function anyway.
1080 if (timer
->function(timer
) != HRTIMER_NORESTART
) {
1081 BUG_ON(timer
->state
!= HRTIMER_STATE_CALLBACK
);
1082 enqueue_hrtimer(timer
, base
, 0);
1084 timer
->state
&= ~HRTIMER_STATE_CALLBACK
;
1086 spin_unlock(&cpu_base
->lock
);
1090 cpu_base
->expires_next
= expires_next
;
1092 /* Reprogramming necessary ? */
1093 if (expires_next
.tv64
!= KTIME_MAX
) {
1094 if (tick_program_event(expires_next
, 0))
1098 /* Raise softirq ? */
1100 raise_softirq(HRTIMER_SOFTIRQ
);
1103 static void run_hrtimer_softirq(struct softirq_action
*h
)
1105 struct hrtimer_cpu_base
*cpu_base
= &__get_cpu_var(hrtimer_bases
);
1107 spin_lock_irq(&cpu_base
->lock
);
1109 while (!list_empty(&cpu_base
->cb_pending
)) {
1110 enum hrtimer_restart (*fn
)(struct hrtimer
*);
1111 struct hrtimer
*timer
;
1114 timer
= list_entry(cpu_base
->cb_pending
.next
,
1115 struct hrtimer
, cb_entry
);
1117 timer_stats_account_hrtimer(timer
);
1119 fn
= timer
->function
;
1120 __remove_hrtimer(timer
, timer
->base
, HRTIMER_STATE_CALLBACK
, 0);
1121 spin_unlock_irq(&cpu_base
->lock
);
1123 restart
= fn(timer
);
1125 spin_lock_irq(&cpu_base
->lock
);
1127 timer
->state
&= ~HRTIMER_STATE_CALLBACK
;
1128 if (restart
== HRTIMER_RESTART
) {
1129 BUG_ON(hrtimer_active(timer
));
1131 * Enqueue the timer, allow reprogramming of the event
1134 enqueue_hrtimer(timer
, timer
->base
, 1);
1135 } else if (hrtimer_active(timer
)) {
1137 * If the timer was rearmed on another CPU, reprogram
1140 if (timer
->base
->first
== &timer
->node
)
1141 hrtimer_reprogram(timer
, timer
->base
);
1144 spin_unlock_irq(&cpu_base
->lock
);
1147 #endif /* CONFIG_HIGH_RES_TIMERS */
1150 * Expire the per base hrtimer-queue:
1152 static inline void run_hrtimer_queue(struct hrtimer_cpu_base
*cpu_base
,
1155 struct rb_node
*node
;
1156 struct hrtimer_clock_base
*base
= &cpu_base
->clock_base
[index
];
1161 if (base
->get_softirq_time
)
1162 base
->softirq_time
= base
->get_softirq_time();
1164 spin_lock_irq(&cpu_base
->lock
);
1166 while ((node
= base
->first
)) {
1167 struct hrtimer
*timer
;
1168 enum hrtimer_restart (*fn
)(struct hrtimer
*);
1171 timer
= rb_entry(node
, struct hrtimer
, node
);
1172 if (base
->softirq_time
.tv64
<= timer
->expires
.tv64
)
1175 #ifdef CONFIG_HIGH_RES_TIMERS
1176 WARN_ON_ONCE(timer
->cb_mode
== HRTIMER_CB_IRQSAFE_NO_SOFTIRQ
);
1178 timer_stats_account_hrtimer(timer
);
1180 fn
= timer
->function
;
1181 __remove_hrtimer(timer
, base
, HRTIMER_STATE_CALLBACK
, 0);
1182 spin_unlock_irq(&cpu_base
->lock
);
1184 restart
= fn(timer
);
1186 spin_lock_irq(&cpu_base
->lock
);
1188 timer
->state
&= ~HRTIMER_STATE_CALLBACK
;
1189 if (restart
!= HRTIMER_NORESTART
) {
1190 BUG_ON(hrtimer_active(timer
));
1191 enqueue_hrtimer(timer
, base
, 0);
1194 spin_unlock_irq(&cpu_base
->lock
);
1198 * Called from timer softirq every jiffy, expire hrtimers:
1200 * For HRT its the fall back code to run the softirq in the timer
1201 * softirq context in case the hrtimer initialization failed or has
1202 * not been done yet.
1204 void hrtimer_run_queues(void)
1206 struct hrtimer_cpu_base
*cpu_base
= &__get_cpu_var(hrtimer_bases
);
1209 if (hrtimer_hres_active())
1213 * This _is_ ugly: We have to check in the softirq context,
1214 * whether we can switch to highres and / or nohz mode. The
1215 * clocksource switch happens in the timer interrupt with
1216 * xtime_lock held. Notification from there only sets the
1217 * check bit in the tick_oneshot code, otherwise we might
1218 * deadlock vs. xtime_lock.
1220 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1221 if (hrtimer_switch_to_hres())
1224 hrtimer_get_softirq_time(cpu_base
);
1226 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++)
1227 run_hrtimer_queue(cpu_base
, i
);
1231 * Sleep related functions:
1233 static enum hrtimer_restart
hrtimer_wakeup(struct hrtimer
*timer
)
1235 struct hrtimer_sleeper
*t
=
1236 container_of(timer
, struct hrtimer_sleeper
, timer
);
1237 struct task_struct
*task
= t
->task
;
1241 wake_up_process(task
);
1243 return HRTIMER_NORESTART
;
1246 void hrtimer_init_sleeper(struct hrtimer_sleeper
*sl
, struct task_struct
*task
)
1248 sl
->timer
.function
= hrtimer_wakeup
;
1250 #ifdef CONFIG_HIGH_RES_TIMERS
1251 sl
->timer
.cb_mode
= HRTIMER_CB_IRQSAFE_NO_RESTART
;
1255 static int __sched
do_nanosleep(struct hrtimer_sleeper
*t
, enum hrtimer_mode mode
)
1257 hrtimer_init_sleeper(t
, current
);
1260 set_current_state(TASK_INTERRUPTIBLE
);
1261 hrtimer_start(&t
->timer
, t
->timer
.expires
, mode
);
1263 if (likely(t
->task
))
1266 hrtimer_cancel(&t
->timer
);
1267 mode
= HRTIMER_MODE_ABS
;
1269 } while (t
->task
&& !signal_pending(current
));
1271 return t
->task
== NULL
;
1274 long __sched
hrtimer_nanosleep_restart(struct restart_block
*restart
)
1276 struct hrtimer_sleeper t
;
1277 struct timespec __user
*rmtp
;
1281 restart
->fn
= do_no_restart_syscall
;
1283 hrtimer_init(&t
.timer
, restart
->arg0
, HRTIMER_MODE_ABS
);
1284 t
.timer
.expires
.tv64
= ((u64
)restart
->arg3
<< 32) | (u64
) restart
->arg2
;
1286 if (do_nanosleep(&t
, HRTIMER_MODE_ABS
))
1289 rmtp
= (struct timespec __user
*) restart
->arg1
;
1291 time
= ktime_sub(t
.timer
.expires
, t
.timer
.base
->get_time());
1294 tu
= ktime_to_timespec(time
);
1295 if (copy_to_user(rmtp
, &tu
, sizeof(tu
)))
1299 restart
->fn
= hrtimer_nanosleep_restart
;
1301 /* The other values in restart are already filled in */
1302 return -ERESTART_RESTARTBLOCK
;
1305 long hrtimer_nanosleep(struct timespec
*rqtp
, struct timespec __user
*rmtp
,
1306 const enum hrtimer_mode mode
, const clockid_t clockid
)
1308 struct restart_block
*restart
;
1309 struct hrtimer_sleeper t
;
1313 hrtimer_init(&t
.timer
, clockid
, mode
);
1314 t
.timer
.expires
= timespec_to_ktime(*rqtp
);
1315 if (do_nanosleep(&t
, mode
))
1318 /* Absolute timers do not update the rmtp value and restart: */
1319 if (mode
== HRTIMER_MODE_ABS
)
1320 return -ERESTARTNOHAND
;
1323 rem
= ktime_sub(t
.timer
.expires
, t
.timer
.base
->get_time());
1326 tu
= ktime_to_timespec(rem
);
1327 if (copy_to_user(rmtp
, &tu
, sizeof(tu
)))
1331 restart
= ¤t_thread_info()->restart_block
;
1332 restart
->fn
= hrtimer_nanosleep_restart
;
1333 restart
->arg0
= (unsigned long) t
.timer
.base
->index
;
1334 restart
->arg1
= (unsigned long) rmtp
;
1335 restart
->arg2
= t
.timer
.expires
.tv64
& 0xFFFFFFFF;
1336 restart
->arg3
= t
.timer
.expires
.tv64
>> 32;
1338 return -ERESTART_RESTARTBLOCK
;
1342 sys_nanosleep(struct timespec __user
*rqtp
, struct timespec __user
*rmtp
)
1346 if (copy_from_user(&tu
, rqtp
, sizeof(tu
)))
1349 if (!timespec_valid(&tu
))
1352 return hrtimer_nanosleep(&tu
, rmtp
, HRTIMER_MODE_REL
, CLOCK_MONOTONIC
);
1356 * Functions related to boot-time initialization:
1358 static void __devinit
init_hrtimers_cpu(int cpu
)
1360 struct hrtimer_cpu_base
*cpu_base
= &per_cpu(hrtimer_bases
, cpu
);
1363 spin_lock_init(&cpu_base
->lock
);
1364 lockdep_set_class(&cpu_base
->lock
, &cpu_base
->lock_key
);
1366 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++)
1367 cpu_base
->clock_base
[i
].cpu_base
= cpu_base
;
1369 hrtimer_init_hres(cpu_base
);
1372 #ifdef CONFIG_HOTPLUG_CPU
1374 static void migrate_hrtimer_list(struct hrtimer_clock_base
*old_base
,
1375 struct hrtimer_clock_base
*new_base
)
1377 struct hrtimer
*timer
;
1378 struct rb_node
*node
;
1380 while ((node
= rb_first(&old_base
->active
))) {
1381 timer
= rb_entry(node
, struct hrtimer
, node
);
1382 BUG_ON(hrtimer_callback_running(timer
));
1383 __remove_hrtimer(timer
, old_base
, HRTIMER_STATE_INACTIVE
, 0);
1384 timer
->base
= new_base
;
1386 * Enqueue the timer. Allow reprogramming of the event device
1388 enqueue_hrtimer(timer
, new_base
, 1);
1392 static void migrate_hrtimers(int cpu
)
1394 struct hrtimer_cpu_base
*old_base
, *new_base
;
1397 BUG_ON(cpu_online(cpu
));
1398 old_base
= &per_cpu(hrtimer_bases
, cpu
);
1399 new_base
= &get_cpu_var(hrtimer_bases
);
1401 tick_cancel_sched_timer(cpu
);
1403 local_irq_disable();
1404 double_spin_lock(&new_base
->lock
, &old_base
->lock
,
1405 smp_processor_id() < cpu
);
1407 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++) {
1408 migrate_hrtimer_list(&old_base
->clock_base
[i
],
1409 &new_base
->clock_base
[i
]);
1412 double_spin_unlock(&new_base
->lock
, &old_base
->lock
,
1413 smp_processor_id() < cpu
);
1415 put_cpu_var(hrtimer_bases
);
1417 #endif /* CONFIG_HOTPLUG_CPU */
1419 static int __cpuinit
hrtimer_cpu_notify(struct notifier_block
*self
,
1420 unsigned long action
, void *hcpu
)
1422 unsigned int cpu
= (long)hcpu
;
1426 case CPU_UP_PREPARE
:
1427 case CPU_UP_PREPARE_FROZEN
:
1428 init_hrtimers_cpu(cpu
);
1431 #ifdef CONFIG_HOTPLUG_CPU
1433 case CPU_DEAD_FROZEN
:
1434 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD
, &cpu
);
1435 migrate_hrtimers(cpu
);
1446 static struct notifier_block __cpuinitdata hrtimers_nb
= {
1447 .notifier_call
= hrtimer_cpu_notify
,
1450 void __init
hrtimers_init(void)
1452 hrtimer_cpu_notify(&hrtimers_nb
, (unsigned long)CPU_UP_PREPARE
,
1453 (void *)(long)smp_processor_id());
1454 register_cpu_notifier(&hrtimers_nb
);
1455 #ifdef CONFIG_HIGH_RES_TIMERS
1456 open_softirq(HRTIMER_SOFTIRQ
, run_hrtimer_softirq
, NULL
);