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
);
282 * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable
284 * @nsec: the scalar nsec value to subtract
286 * Returns the subtraction of @nsec from @kt in ktime_t format
288 ktime_t
ktime_sub_ns(const ktime_t kt
, u64 nsec
)
292 if (likely(nsec
< NSEC_PER_SEC
)) {
295 unsigned long rem
= do_div(nsec
, NSEC_PER_SEC
);
297 tmp
= ktime_set((long)nsec
, rem
);
300 return ktime_sub(kt
, tmp
);
303 EXPORT_SYMBOL_GPL(ktime_sub_ns
);
304 # endif /* !CONFIG_KTIME_SCALAR */
307 * Divide a ktime value by a nanosecond value
309 u64
ktime_divns(const ktime_t kt
, s64 div
)
314 dclc
= dns
= ktime_to_ns(kt
);
316 /* Make sure the divisor is less than 2^32: */
322 do_div(dclc
, (unsigned long) div
);
326 #endif /* BITS_PER_LONG >= 64 */
329 * Add two ktime values and do a safety check for overflow:
331 ktime_t
ktime_add_safe(const ktime_t lhs
, const ktime_t rhs
)
333 ktime_t res
= ktime_add(lhs
, rhs
);
336 * We use KTIME_SEC_MAX here, the maximum timeout which we can
337 * return to user space in a timespec:
339 if (res
.tv64
< 0 || res
.tv64
< lhs
.tv64
|| res
.tv64
< rhs
.tv64
)
340 res
= ktime_set(KTIME_SEC_MAX
, 0);
346 * Check, whether the timer is on the callback pending list
348 static inline int hrtimer_cb_pending(const struct hrtimer
*timer
)
350 return timer
->state
& HRTIMER_STATE_PENDING
;
354 * Remove a timer from the callback pending list
356 static inline void hrtimer_remove_cb_pending(struct hrtimer
*timer
)
358 list_del_init(&timer
->cb_entry
);
361 /* High resolution timer related functions */
362 #ifdef CONFIG_HIGH_RES_TIMERS
365 * High resolution timer enabled ?
367 static int hrtimer_hres_enabled __read_mostly
= 1;
370 * Enable / Disable high resolution mode
372 static int __init
setup_hrtimer_hres(char *str
)
374 if (!strcmp(str
, "off"))
375 hrtimer_hres_enabled
= 0;
376 else if (!strcmp(str
, "on"))
377 hrtimer_hres_enabled
= 1;
383 __setup("highres=", setup_hrtimer_hres
);
386 * hrtimer_high_res_enabled - query, if the highres mode is enabled
388 static inline int hrtimer_is_hres_enabled(void)
390 return hrtimer_hres_enabled
;
394 * Is the high resolution mode active ?
396 static inline int hrtimer_hres_active(void)
398 return __get_cpu_var(hrtimer_bases
).hres_active
;
402 * Reprogram the event source with checking both queues for the
404 * Called with interrupts disabled and base->lock held
406 static void hrtimer_force_reprogram(struct hrtimer_cpu_base
*cpu_base
)
409 struct hrtimer_clock_base
*base
= cpu_base
->clock_base
;
412 cpu_base
->expires_next
.tv64
= KTIME_MAX
;
414 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++, base
++) {
415 struct hrtimer
*timer
;
419 timer
= rb_entry(base
->first
, struct hrtimer
, node
);
420 expires
= ktime_sub(timer
->expires
, base
->offset
);
421 if (expires
.tv64
< cpu_base
->expires_next
.tv64
)
422 cpu_base
->expires_next
= expires
;
425 if (cpu_base
->expires_next
.tv64
!= KTIME_MAX
)
426 tick_program_event(cpu_base
->expires_next
, 1);
430 * Shared reprogramming for clock_realtime and clock_monotonic
432 * When a timer is enqueued and expires earlier than the already enqueued
433 * timers, we have to check, whether it expires earlier than the timer for
434 * which the clock event device was armed.
436 * Called with interrupts disabled and base->cpu_base.lock held
438 static int hrtimer_reprogram(struct hrtimer
*timer
,
439 struct hrtimer_clock_base
*base
)
441 ktime_t
*expires_next
= &__get_cpu_var(hrtimer_bases
).expires_next
;
442 ktime_t expires
= ktime_sub(timer
->expires
, base
->offset
);
445 WARN_ON_ONCE(timer
->expires
.tv64
< 0);
448 * When the callback is running, we do not reprogram the clock event
449 * device. The timer callback is either running on a different CPU or
450 * the callback is executed in the hrtimer_interrupt context. The
451 * reprogramming is handled either by the softirq, which called the
452 * callback or at the end of the hrtimer_interrupt.
454 if (hrtimer_callback_running(timer
))
458 * CLOCK_REALTIME timer might be requested with an absolute
459 * expiry time which is less than base->offset. Nothing wrong
460 * about that, just avoid to call into the tick code, which
461 * has now objections against negative expiry values.
463 if (expires
.tv64
< 0)
466 if (expires
.tv64
>= expires_next
->tv64
)
470 * Clockevents returns -ETIME, when the event was in the past.
472 res
= tick_program_event(expires
, 0);
473 if (!IS_ERR_VALUE(res
))
474 *expires_next
= expires
;
480 * Retrigger next event is called after clock was set
482 * Called with interrupts disabled via on_each_cpu()
484 static void retrigger_next_event(void *arg
)
486 struct hrtimer_cpu_base
*base
;
487 struct timespec realtime_offset
;
490 if (!hrtimer_hres_active())
494 seq
= read_seqbegin(&xtime_lock
);
495 set_normalized_timespec(&realtime_offset
,
496 -wall_to_monotonic
.tv_sec
,
497 -wall_to_monotonic
.tv_nsec
);
498 } while (read_seqretry(&xtime_lock
, seq
));
500 base
= &__get_cpu_var(hrtimer_bases
);
502 /* Adjust CLOCK_REALTIME offset */
503 spin_lock(&base
->lock
);
504 base
->clock_base
[CLOCK_REALTIME
].offset
=
505 timespec_to_ktime(realtime_offset
);
507 hrtimer_force_reprogram(base
);
508 spin_unlock(&base
->lock
);
512 * Clock realtime was set
514 * Change the offset of the realtime clock vs. the monotonic
517 * We might have to reprogram the high resolution timer interrupt. On
518 * SMP we call the architecture specific code to retrigger _all_ high
519 * resolution timer interrupts. On UP we just disable interrupts and
520 * call the high resolution interrupt code.
522 void clock_was_set(void)
524 /* Retrigger the CPU local events everywhere */
525 on_each_cpu(retrigger_next_event
, NULL
, 0, 1);
529 * During resume we might have to reprogram the high resolution timer
530 * interrupt (on the local CPU):
532 void hres_timers_resume(void)
534 WARN_ON_ONCE(num_online_cpus() > 1);
536 /* Retrigger the CPU local events: */
537 retrigger_next_event(NULL
);
541 * Initialize the high resolution related parts of cpu_base
543 static inline void hrtimer_init_hres(struct hrtimer_cpu_base
*base
)
545 base
->expires_next
.tv64
= KTIME_MAX
;
546 base
->hres_active
= 0;
550 * Initialize the high resolution related parts of a hrtimer
552 static inline void hrtimer_init_timer_hres(struct hrtimer
*timer
)
557 * When High resolution timers are active, try to reprogram. Note, that in case
558 * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry
559 * check happens. The timer gets enqueued into the rbtree. The reprogramming
560 * and expiry check is done in the hrtimer_interrupt or in the softirq.
562 static inline int hrtimer_enqueue_reprogram(struct hrtimer
*timer
,
563 struct hrtimer_clock_base
*base
)
565 if (base
->cpu_base
->hres_active
&& hrtimer_reprogram(timer
, base
)) {
567 /* Timer is expired, act upon the callback mode */
568 switch(timer
->cb_mode
) {
569 case HRTIMER_CB_IRQSAFE_NO_RESTART
:
571 * We can call the callback from here. No restart
572 * happens, so no danger of recursion
574 BUG_ON(timer
->function(timer
) != HRTIMER_NORESTART
);
576 case HRTIMER_CB_IRQSAFE_NO_SOFTIRQ
:
578 * This is solely for the sched tick emulation with
579 * dynamic tick support to ensure that we do not
580 * restart the tick right on the edge and end up with
581 * the tick timer in the softirq ! The calling site
582 * takes care of this.
585 case HRTIMER_CB_IRQSAFE
:
586 case HRTIMER_CB_SOFTIRQ
:
588 * Move everything else into the softirq pending list !
590 list_add_tail(&timer
->cb_entry
,
591 &base
->cpu_base
->cb_pending
);
592 timer
->state
= HRTIMER_STATE_PENDING
;
593 raise_softirq(HRTIMER_SOFTIRQ
);
603 * Switch to high resolution mode
605 static int hrtimer_switch_to_hres(void)
607 int cpu
= smp_processor_id();
608 struct hrtimer_cpu_base
*base
= &per_cpu(hrtimer_bases
, cpu
);
611 if (base
->hres_active
)
614 local_irq_save(flags
);
616 if (tick_init_highres()) {
617 local_irq_restore(flags
);
618 printk(KERN_WARNING
"Could not switch to high resolution "
619 "mode on CPU %d\n", cpu
);
622 base
->hres_active
= 1;
623 base
->clock_base
[CLOCK_REALTIME
].resolution
= KTIME_HIGH_RES
;
624 base
->clock_base
[CLOCK_MONOTONIC
].resolution
= KTIME_HIGH_RES
;
626 tick_setup_sched_timer();
628 /* "Retrigger" the interrupt to get things going */
629 retrigger_next_event(NULL
);
630 local_irq_restore(flags
);
631 printk(KERN_DEBUG
"Switched to high resolution mode on CPU %d\n",
638 static inline int hrtimer_hres_active(void) { return 0; }
639 static inline int hrtimer_is_hres_enabled(void) { return 0; }
640 static inline int hrtimer_switch_to_hres(void) { return 0; }
641 static inline void hrtimer_force_reprogram(struct hrtimer_cpu_base
*base
) { }
642 static inline int hrtimer_enqueue_reprogram(struct hrtimer
*timer
,
643 struct hrtimer_clock_base
*base
)
647 static inline void hrtimer_init_hres(struct hrtimer_cpu_base
*base
) { }
648 static inline void hrtimer_init_timer_hres(struct hrtimer
*timer
) { }
649 static inline int hrtimer_reprogram(struct hrtimer
*timer
,
650 struct hrtimer_clock_base
*base
)
655 #endif /* CONFIG_HIGH_RES_TIMERS */
657 #ifdef CONFIG_TIMER_STATS
658 void __timer_stats_hrtimer_set_start_info(struct hrtimer
*timer
, void *addr
)
660 if (timer
->start_site
)
663 timer
->start_site
= addr
;
664 memcpy(timer
->start_comm
, current
->comm
, TASK_COMM_LEN
);
665 timer
->start_pid
= current
->pid
;
670 * Counterpart to lock_hrtimer_base above:
673 void unlock_hrtimer_base(const struct hrtimer
*timer
, unsigned long *flags
)
675 spin_unlock_irqrestore(&timer
->base
->cpu_base
->lock
, *flags
);
679 * hrtimer_forward - forward the timer expiry
680 * @timer: hrtimer to forward
681 * @now: forward past this time
682 * @interval: the interval to forward
684 * Forward the timer expiry so it will expire in the future.
685 * Returns the number of overruns.
687 u64
hrtimer_forward(struct hrtimer
*timer
, ktime_t now
, ktime_t interval
)
692 delta
= ktime_sub(now
, timer
->expires
);
697 if (interval
.tv64
< timer
->base
->resolution
.tv64
)
698 interval
.tv64
= timer
->base
->resolution
.tv64
;
700 if (unlikely(delta
.tv64
>= interval
.tv64
)) {
701 s64 incr
= ktime_to_ns(interval
);
703 orun
= ktime_divns(delta
, incr
);
704 timer
->expires
= ktime_add_ns(timer
->expires
, incr
* orun
);
705 if (timer
->expires
.tv64
> now
.tv64
)
708 * This (and the ktime_add() below) is the
709 * correction for exact:
713 timer
->expires
= ktime_add_safe(timer
->expires
, interval
);
717 EXPORT_SYMBOL_GPL(hrtimer_forward
);
720 * enqueue_hrtimer - internal function to (re)start a timer
722 * The timer is inserted in expiry order. Insertion into the
723 * red black tree is O(log(n)). Must hold the base lock.
725 static void enqueue_hrtimer(struct hrtimer
*timer
,
726 struct hrtimer_clock_base
*base
, int reprogram
)
728 struct rb_node
**link
= &base
->active
.rb_node
;
729 struct rb_node
*parent
= NULL
;
730 struct hrtimer
*entry
;
734 * Find the right place in the rbtree:
738 entry
= rb_entry(parent
, struct hrtimer
, node
);
740 * We dont care about collisions. Nodes with
741 * the same expiry time stay together.
743 if (timer
->expires
.tv64
< entry
->expires
.tv64
) {
744 link
= &(*link
)->rb_left
;
746 link
= &(*link
)->rb_right
;
752 * Insert the timer to the rbtree and check whether it
753 * replaces the first pending timer
757 * Reprogram the clock event device. When the timer is already
758 * expired hrtimer_enqueue_reprogram has either called the
759 * callback or added it to the pending list and raised the
762 * This is a NOP for !HIGHRES
764 if (reprogram
&& hrtimer_enqueue_reprogram(timer
, base
))
767 base
->first
= &timer
->node
;
770 rb_link_node(&timer
->node
, parent
, link
);
771 rb_insert_color(&timer
->node
, &base
->active
);
773 * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
774 * state of a possibly running callback.
776 timer
->state
|= HRTIMER_STATE_ENQUEUED
;
780 * __remove_hrtimer - internal function to remove a timer
782 * Caller must hold the base lock.
784 * High resolution timer mode reprograms the clock event device when the
785 * timer is the one which expires next. The caller can disable this by setting
786 * reprogram to zero. This is useful, when the context does a reprogramming
787 * anyway (e.g. timer interrupt)
789 static void __remove_hrtimer(struct hrtimer
*timer
,
790 struct hrtimer_clock_base
*base
,
791 unsigned long newstate
, int reprogram
)
793 /* High res. callback list. NOP for !HIGHRES */
794 if (hrtimer_cb_pending(timer
))
795 hrtimer_remove_cb_pending(timer
);
798 * Remove the timer from the rbtree and replace the
799 * first entry pointer if necessary.
801 if (base
->first
== &timer
->node
) {
802 base
->first
= rb_next(&timer
->node
);
803 /* Reprogram the clock event device. if enabled */
804 if (reprogram
&& hrtimer_hres_active())
805 hrtimer_force_reprogram(base
->cpu_base
);
807 rb_erase(&timer
->node
, &base
->active
);
809 timer
->state
= newstate
;
813 * remove hrtimer, called with base lock held
816 remove_hrtimer(struct hrtimer
*timer
, struct hrtimer_clock_base
*base
)
818 if (hrtimer_is_queued(timer
)) {
822 * Remove the timer and force reprogramming when high
823 * resolution mode is active and the timer is on the current
824 * CPU. If we remove a timer on another CPU, reprogramming is
825 * skipped. The interrupt event on this CPU is fired and
826 * reprogramming happens in the interrupt handler. This is a
827 * rare case and less expensive than a smp call.
829 timer_stats_hrtimer_clear_start_info(timer
);
830 reprogram
= base
->cpu_base
== &__get_cpu_var(hrtimer_bases
);
831 __remove_hrtimer(timer
, base
, HRTIMER_STATE_INACTIVE
,
839 * hrtimer_start - (re)start an relative timer on the current CPU
840 * @timer: the timer to be added
842 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
846 * 1 when the timer was active
849 hrtimer_start(struct hrtimer
*timer
, ktime_t tim
, const enum hrtimer_mode mode
)
851 struct hrtimer_clock_base
*base
, *new_base
;
855 base
= lock_hrtimer_base(timer
, &flags
);
857 /* Remove an active timer from the queue: */
858 ret
= remove_hrtimer(timer
, base
);
860 /* Switch the timer base, if necessary: */
861 new_base
= switch_hrtimer_base(timer
, base
);
863 if (mode
== HRTIMER_MODE_REL
) {
864 tim
= ktime_add_safe(tim
, new_base
->get_time());
866 * CONFIG_TIME_LOW_RES is a temporary way for architectures
867 * to signal that they simply return xtime in
868 * do_gettimeoffset(). In this case we want to round up by
869 * resolution when starting a relative timer, to avoid short
870 * timeouts. This will go away with the GTOD framework.
872 #ifdef CONFIG_TIME_LOW_RES
873 tim
= ktime_add_safe(tim
, base
->resolution
);
876 timer
->expires
= tim
;
878 timer_stats_hrtimer_set_start_info(timer
);
881 * Only allow reprogramming if the new base is on this CPU.
882 * (it might still be on another CPU if the timer was pending)
884 enqueue_hrtimer(timer
, new_base
,
885 new_base
->cpu_base
== &__get_cpu_var(hrtimer_bases
));
887 unlock_hrtimer_base(timer
, &flags
);
891 EXPORT_SYMBOL_GPL(hrtimer_start
);
894 * hrtimer_try_to_cancel - try to deactivate a timer
895 * @timer: hrtimer to stop
898 * 0 when the timer was not active
899 * 1 when the timer was active
900 * -1 when the timer is currently excuting the callback function and
903 int hrtimer_try_to_cancel(struct hrtimer
*timer
)
905 struct hrtimer_clock_base
*base
;
909 base
= lock_hrtimer_base(timer
, &flags
);
911 if (!hrtimer_callback_running(timer
))
912 ret
= remove_hrtimer(timer
, base
);
914 unlock_hrtimer_base(timer
, &flags
);
919 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel
);
922 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
923 * @timer: the timer to be cancelled
926 * 0 when the timer was not active
927 * 1 when the timer was active
929 int hrtimer_cancel(struct hrtimer
*timer
)
932 int ret
= hrtimer_try_to_cancel(timer
);
939 EXPORT_SYMBOL_GPL(hrtimer_cancel
);
942 * hrtimer_get_remaining - get remaining time for the timer
943 * @timer: the timer to read
945 ktime_t
hrtimer_get_remaining(const struct hrtimer
*timer
)
947 struct hrtimer_clock_base
*base
;
951 base
= lock_hrtimer_base(timer
, &flags
);
952 rem
= ktime_sub(timer
->expires
, base
->get_time());
953 unlock_hrtimer_base(timer
, &flags
);
957 EXPORT_SYMBOL_GPL(hrtimer_get_remaining
);
959 #if defined(CONFIG_NO_IDLE_HZ) || defined(CONFIG_NO_HZ)
961 * hrtimer_get_next_event - get the time until next expiry event
963 * Returns the delta to the next expiry event or KTIME_MAX if no timer
966 ktime_t
hrtimer_get_next_event(void)
968 struct hrtimer_cpu_base
*cpu_base
= &__get_cpu_var(hrtimer_bases
);
969 struct hrtimer_clock_base
*base
= cpu_base
->clock_base
;
970 ktime_t delta
, mindelta
= { .tv64
= KTIME_MAX
};
974 spin_lock_irqsave(&cpu_base
->lock
, flags
);
976 if (!hrtimer_hres_active()) {
977 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++, base
++) {
978 struct hrtimer
*timer
;
983 timer
= rb_entry(base
->first
, struct hrtimer
, node
);
984 delta
.tv64
= timer
->expires
.tv64
;
985 delta
= ktime_sub(delta
, base
->get_time());
986 if (delta
.tv64
< mindelta
.tv64
)
987 mindelta
.tv64
= delta
.tv64
;
991 spin_unlock_irqrestore(&cpu_base
->lock
, flags
);
993 if (mindelta
.tv64
< 0)
1000 * hrtimer_init - initialize a timer to the given clock
1001 * @timer: the timer to be initialized
1002 * @clock_id: the clock to be used
1003 * @mode: timer mode abs/rel
1005 void hrtimer_init(struct hrtimer
*timer
, clockid_t clock_id
,
1006 enum hrtimer_mode mode
)
1008 struct hrtimer_cpu_base
*cpu_base
;
1010 memset(timer
, 0, sizeof(struct hrtimer
));
1012 cpu_base
= &__raw_get_cpu_var(hrtimer_bases
);
1014 if (clock_id
== CLOCK_REALTIME
&& mode
!= HRTIMER_MODE_ABS
)
1015 clock_id
= CLOCK_MONOTONIC
;
1017 timer
->base
= &cpu_base
->clock_base
[clock_id
];
1018 INIT_LIST_HEAD(&timer
->cb_entry
);
1019 hrtimer_init_timer_hres(timer
);
1021 #ifdef CONFIG_TIMER_STATS
1022 timer
->start_site
= NULL
;
1023 timer
->start_pid
= -1;
1024 memset(timer
->start_comm
, 0, TASK_COMM_LEN
);
1027 EXPORT_SYMBOL_GPL(hrtimer_init
);
1030 * hrtimer_get_res - get the timer resolution for a clock
1031 * @which_clock: which clock to query
1032 * @tp: pointer to timespec variable to store the resolution
1034 * Store the resolution of the clock selected by @which_clock in the
1035 * variable pointed to by @tp.
1037 int hrtimer_get_res(const clockid_t which_clock
, struct timespec
*tp
)
1039 struct hrtimer_cpu_base
*cpu_base
;
1041 cpu_base
= &__raw_get_cpu_var(hrtimer_bases
);
1042 *tp
= ktime_to_timespec(cpu_base
->clock_base
[which_clock
].resolution
);
1046 EXPORT_SYMBOL_GPL(hrtimer_get_res
);
1048 static void run_hrtimer_pending(struct hrtimer_cpu_base
*cpu_base
)
1050 spin_lock_irq(&cpu_base
->lock
);
1052 while (!list_empty(&cpu_base
->cb_pending
)) {
1053 enum hrtimer_restart (*fn
)(struct hrtimer
*);
1054 struct hrtimer
*timer
;
1057 timer
= list_entry(cpu_base
->cb_pending
.next
,
1058 struct hrtimer
, cb_entry
);
1060 timer_stats_account_hrtimer(timer
);
1062 fn
= timer
->function
;
1063 __remove_hrtimer(timer
, timer
->base
, HRTIMER_STATE_CALLBACK
, 0);
1064 spin_unlock_irq(&cpu_base
->lock
);
1066 restart
= fn(timer
);
1068 spin_lock_irq(&cpu_base
->lock
);
1070 timer
->state
&= ~HRTIMER_STATE_CALLBACK
;
1071 if (restart
== HRTIMER_RESTART
) {
1072 BUG_ON(hrtimer_active(timer
));
1074 * Enqueue the timer, allow reprogramming of the event
1077 enqueue_hrtimer(timer
, timer
->base
, 1);
1078 } else if (hrtimer_active(timer
)) {
1080 * If the timer was rearmed on another CPU, reprogram
1083 if (timer
->base
->first
== &timer
->node
)
1084 hrtimer_reprogram(timer
, timer
->base
);
1087 spin_unlock_irq(&cpu_base
->lock
);
1090 static void __run_hrtimer(struct hrtimer
*timer
)
1092 struct hrtimer_clock_base
*base
= timer
->base
;
1093 struct hrtimer_cpu_base
*cpu_base
= base
->cpu_base
;
1094 enum hrtimer_restart (*fn
)(struct hrtimer
*);
1097 __remove_hrtimer(timer
, base
, HRTIMER_STATE_CALLBACK
, 0);
1098 timer_stats_account_hrtimer(timer
);
1100 fn
= timer
->function
;
1101 if (timer
->cb_mode
== HRTIMER_CB_IRQSAFE_NO_SOFTIRQ
) {
1103 * Used for scheduler timers, avoid lock inversion with
1104 * rq->lock and tasklist_lock.
1106 * These timers are required to deal with enqueue expiry
1107 * themselves and are not allowed to migrate.
1109 spin_unlock(&cpu_base
->lock
);
1110 restart
= fn(timer
);
1111 spin_lock(&cpu_base
->lock
);
1113 restart
= fn(timer
);
1116 * Note: We clear the CALLBACK bit after enqueue_hrtimer to avoid
1117 * reprogramming of the event hardware. This happens at the end of this
1120 if (restart
!= HRTIMER_NORESTART
) {
1121 BUG_ON(timer
->state
!= HRTIMER_STATE_CALLBACK
);
1122 enqueue_hrtimer(timer
, base
, 0);
1124 timer
->state
&= ~HRTIMER_STATE_CALLBACK
;
1127 #ifdef CONFIG_HIGH_RES_TIMERS
1130 * High resolution timer interrupt
1131 * Called with interrupts disabled
1133 void hrtimer_interrupt(struct clock_event_device
*dev
)
1135 struct hrtimer_cpu_base
*cpu_base
= &__get_cpu_var(hrtimer_bases
);
1136 struct hrtimer_clock_base
*base
;
1137 ktime_t expires_next
, now
;
1140 BUG_ON(!cpu_base
->hres_active
);
1141 cpu_base
->nr_events
++;
1142 dev
->next_event
.tv64
= KTIME_MAX
;
1147 expires_next
.tv64
= KTIME_MAX
;
1149 base
= cpu_base
->clock_base
;
1151 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++) {
1153 struct rb_node
*node
;
1155 spin_lock(&cpu_base
->lock
);
1157 basenow
= ktime_add(now
, base
->offset
);
1159 while ((node
= base
->first
)) {
1160 struct hrtimer
*timer
;
1162 timer
= rb_entry(node
, struct hrtimer
, node
);
1164 if (basenow
.tv64
< timer
->expires
.tv64
) {
1167 expires
= ktime_sub(timer
->expires
,
1169 if (expires
.tv64
< expires_next
.tv64
)
1170 expires_next
= expires
;
1174 /* Move softirq callbacks to the pending list */
1175 if (timer
->cb_mode
== HRTIMER_CB_SOFTIRQ
) {
1176 __remove_hrtimer(timer
, base
,
1177 HRTIMER_STATE_PENDING
, 0);
1178 list_add_tail(&timer
->cb_entry
,
1179 &base
->cpu_base
->cb_pending
);
1184 __run_hrtimer(timer
);
1186 spin_unlock(&cpu_base
->lock
);
1190 cpu_base
->expires_next
= expires_next
;
1192 /* Reprogramming necessary ? */
1193 if (expires_next
.tv64
!= KTIME_MAX
) {
1194 if (tick_program_event(expires_next
, 0))
1198 /* Raise softirq ? */
1200 raise_softirq(HRTIMER_SOFTIRQ
);
1203 static void run_hrtimer_softirq(struct softirq_action
*h
)
1205 run_hrtimer_pending(&__get_cpu_var(hrtimer_bases
));
1208 #endif /* CONFIG_HIGH_RES_TIMERS */
1211 * Called from timer softirq every jiffy, expire hrtimers:
1213 * For HRT its the fall back code to run the softirq in the timer
1214 * softirq context in case the hrtimer initialization failed or has
1215 * not been done yet.
1217 void hrtimer_run_pending(void)
1219 struct hrtimer_cpu_base
*cpu_base
= &__get_cpu_var(hrtimer_bases
);
1221 if (hrtimer_hres_active())
1225 * This _is_ ugly: We have to check in the softirq context,
1226 * whether we can switch to highres and / or nohz mode. The
1227 * clocksource switch happens in the timer interrupt with
1228 * xtime_lock held. Notification from there only sets the
1229 * check bit in the tick_oneshot code, otherwise we might
1230 * deadlock vs. xtime_lock.
1232 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1233 hrtimer_switch_to_hres();
1235 run_hrtimer_pending(cpu_base
);
1239 * Called from hardirq context every jiffy
1241 void hrtimer_run_queues(void)
1243 struct rb_node
*node
;
1244 struct hrtimer_cpu_base
*cpu_base
= &__get_cpu_var(hrtimer_bases
);
1245 struct hrtimer_clock_base
*base
;
1246 int index
, gettime
= 1;
1248 if (hrtimer_hres_active())
1251 for (index
= 0; index
< HRTIMER_MAX_CLOCK_BASES
; index
++) {
1252 base
= &cpu_base
->clock_base
[index
];
1257 if (base
->get_softirq_time
)
1258 base
->softirq_time
= base
->get_softirq_time();
1260 hrtimer_get_softirq_time(cpu_base
);
1264 spin_lock(&cpu_base
->lock
);
1266 while ((node
= base
->first
)) {
1267 struct hrtimer
*timer
;
1269 timer
= rb_entry(node
, struct hrtimer
, node
);
1270 if (base
->softirq_time
.tv64
<= timer
->expires
.tv64
)
1273 if (timer
->cb_mode
== HRTIMER_CB_SOFTIRQ
) {
1274 __remove_hrtimer(timer
, base
,
1275 HRTIMER_STATE_PENDING
, 0);
1276 list_add_tail(&timer
->cb_entry
,
1277 &base
->cpu_base
->cb_pending
);
1281 __run_hrtimer(timer
);
1283 spin_unlock(&cpu_base
->lock
);
1288 * Sleep related functions:
1290 static enum hrtimer_restart
hrtimer_wakeup(struct hrtimer
*timer
)
1292 struct hrtimer_sleeper
*t
=
1293 container_of(timer
, struct hrtimer_sleeper
, timer
);
1294 struct task_struct
*task
= t
->task
;
1298 wake_up_process(task
);
1300 return HRTIMER_NORESTART
;
1303 void hrtimer_init_sleeper(struct hrtimer_sleeper
*sl
, struct task_struct
*task
)
1305 sl
->timer
.function
= hrtimer_wakeup
;
1307 #ifdef CONFIG_HIGH_RES_TIMERS
1308 sl
->timer
.cb_mode
= HRTIMER_CB_IRQSAFE_NO_SOFTIRQ
;
1312 static int __sched
do_nanosleep(struct hrtimer_sleeper
*t
, enum hrtimer_mode mode
)
1314 hrtimer_init_sleeper(t
, current
);
1317 set_current_state(TASK_INTERRUPTIBLE
);
1318 hrtimer_start(&t
->timer
, t
->timer
.expires
, mode
);
1319 if (!hrtimer_active(&t
->timer
))
1322 if (likely(t
->task
))
1325 hrtimer_cancel(&t
->timer
);
1326 mode
= HRTIMER_MODE_ABS
;
1328 } while (t
->task
&& !signal_pending(current
));
1330 __set_current_state(TASK_RUNNING
);
1332 return t
->task
== NULL
;
1335 static int update_rmtp(struct hrtimer
*timer
, struct timespec __user
*rmtp
)
1337 struct timespec rmt
;
1340 rem
= ktime_sub(timer
->expires
, timer
->base
->get_time());
1343 rmt
= ktime_to_timespec(rem
);
1345 if (copy_to_user(rmtp
, &rmt
, sizeof(*rmtp
)))
1351 long __sched
hrtimer_nanosleep_restart(struct restart_block
*restart
)
1353 struct hrtimer_sleeper t
;
1354 struct timespec __user
*rmtp
;
1356 hrtimer_init(&t
.timer
, restart
->nanosleep
.index
, HRTIMER_MODE_ABS
);
1357 t
.timer
.expires
.tv64
= restart
->nanosleep
.expires
;
1359 if (do_nanosleep(&t
, HRTIMER_MODE_ABS
))
1362 rmtp
= restart
->nanosleep
.rmtp
;
1364 int ret
= update_rmtp(&t
.timer
, rmtp
);
1369 /* The other values in restart are already filled in */
1370 return -ERESTART_RESTARTBLOCK
;
1373 long hrtimer_nanosleep(struct timespec
*rqtp
, struct timespec __user
*rmtp
,
1374 const enum hrtimer_mode mode
, const clockid_t clockid
)
1376 struct restart_block
*restart
;
1377 struct hrtimer_sleeper t
;
1379 hrtimer_init(&t
.timer
, clockid
, mode
);
1380 t
.timer
.expires
= timespec_to_ktime(*rqtp
);
1381 if (do_nanosleep(&t
, mode
))
1384 /* Absolute timers do not update the rmtp value and restart: */
1385 if (mode
== HRTIMER_MODE_ABS
)
1386 return -ERESTARTNOHAND
;
1389 int ret
= update_rmtp(&t
.timer
, rmtp
);
1394 restart
= ¤t_thread_info()->restart_block
;
1395 restart
->fn
= hrtimer_nanosleep_restart
;
1396 restart
->nanosleep
.index
= t
.timer
.base
->index
;
1397 restart
->nanosleep
.rmtp
= rmtp
;
1398 restart
->nanosleep
.expires
= t
.timer
.expires
.tv64
;
1400 return -ERESTART_RESTARTBLOCK
;
1404 sys_nanosleep(struct timespec __user
*rqtp
, struct timespec __user
*rmtp
)
1408 if (copy_from_user(&tu
, rqtp
, sizeof(tu
)))
1411 if (!timespec_valid(&tu
))
1414 return hrtimer_nanosleep(&tu
, rmtp
, HRTIMER_MODE_REL
, CLOCK_MONOTONIC
);
1418 * Functions related to boot-time initialization:
1420 static void __cpuinit
init_hrtimers_cpu(int cpu
)
1422 struct hrtimer_cpu_base
*cpu_base
= &per_cpu(hrtimer_bases
, cpu
);
1425 spin_lock_init(&cpu_base
->lock
);
1427 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++)
1428 cpu_base
->clock_base
[i
].cpu_base
= cpu_base
;
1430 INIT_LIST_HEAD(&cpu_base
->cb_pending
);
1431 hrtimer_init_hres(cpu_base
);
1434 #ifdef CONFIG_HOTPLUG_CPU
1436 static void migrate_hrtimer_list(struct hrtimer_clock_base
*old_base
,
1437 struct hrtimer_clock_base
*new_base
)
1439 struct hrtimer
*timer
;
1440 struct rb_node
*node
;
1442 while ((node
= rb_first(&old_base
->active
))) {
1443 timer
= rb_entry(node
, struct hrtimer
, node
);
1444 BUG_ON(hrtimer_callback_running(timer
));
1445 __remove_hrtimer(timer
, old_base
, HRTIMER_STATE_INACTIVE
, 0);
1446 timer
->base
= new_base
;
1448 * Enqueue the timer. Allow reprogramming of the event device
1450 enqueue_hrtimer(timer
, new_base
, 1);
1454 static void migrate_hrtimers(int cpu
)
1456 struct hrtimer_cpu_base
*old_base
, *new_base
;
1459 BUG_ON(cpu_online(cpu
));
1460 old_base
= &per_cpu(hrtimer_bases
, cpu
);
1461 new_base
= &get_cpu_var(hrtimer_bases
);
1463 tick_cancel_sched_timer(cpu
);
1465 local_irq_disable();
1466 spin_lock(&new_base
->lock
);
1467 spin_lock_nested(&old_base
->lock
, SINGLE_DEPTH_NESTING
);
1469 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++) {
1470 migrate_hrtimer_list(&old_base
->clock_base
[i
],
1471 &new_base
->clock_base
[i
]);
1474 spin_unlock(&old_base
->lock
);
1475 spin_unlock(&new_base
->lock
);
1477 put_cpu_var(hrtimer_bases
);
1479 #endif /* CONFIG_HOTPLUG_CPU */
1481 static int __cpuinit
hrtimer_cpu_notify(struct notifier_block
*self
,
1482 unsigned long action
, void *hcpu
)
1484 unsigned int cpu
= (long)hcpu
;
1488 case CPU_UP_PREPARE
:
1489 case CPU_UP_PREPARE_FROZEN
:
1490 init_hrtimers_cpu(cpu
);
1493 #ifdef CONFIG_HOTPLUG_CPU
1495 case CPU_DEAD_FROZEN
:
1496 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD
, &cpu
);
1497 migrate_hrtimers(cpu
);
1508 static struct notifier_block __cpuinitdata hrtimers_nb
= {
1509 .notifier_call
= hrtimer_cpu_notify
,
1512 void __init
hrtimers_init(void)
1514 hrtimer_cpu_notify(&hrtimers_nb
, (unsigned long)CPU_UP_PREPARE
,
1515 (void *)(long)smp_processor_id());
1516 register_cpu_notifier(&hrtimers_nb
);
1517 #ifdef CONFIG_HIGH_RES_TIMERS
1518 open_softirq(HRTIMER_SOFTIRQ
, run_hrtimer_softirq
, NULL
);