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/module.h>
36 #include <linux/percpu.h>
37 #include <linux/hrtimer.h>
38 #include <linux/notifier.h>
39 #include <linux/syscalls.h>
40 #include <linux/kallsyms.h>
41 #include <linux/interrupt.h>
42 #include <linux/tick.h>
43 #include <linux/seq_file.h>
44 #include <linux/err.h>
45 #include <linux/debugobjects.h>
46 #include <linux/sched.h>
47 #include <linux/timer.h>
49 #include <asm/uaccess.h>
51 #include <trace/events/timer.h>
56 * There are more clockids then hrtimer bases. Thus, we index
57 * into the timer bases by the hrtimer_base_type enum. When trying
58 * to reach a base using a clockid, hrtimer_clockid_to_base()
59 * is used to convert from clockid to the proper hrtimer_base_type.
61 DEFINE_PER_CPU(struct hrtimer_cpu_base
, hrtimer_bases
) =
67 .index
= CLOCK_REALTIME
,
68 .get_time
= &ktime_get_real
,
69 .resolution
= KTIME_LOW_RES
,
72 .index
= CLOCK_MONOTONIC
,
73 .get_time
= &ktime_get
,
74 .resolution
= KTIME_LOW_RES
,
77 .index
= CLOCK_BOOTTIME
,
78 .get_time
= &ktime_get_boottime
,
79 .resolution
= KTIME_LOW_RES
,
84 static int hrtimer_clock_to_base_table
[MAX_CLOCKS
];
86 static inline int hrtimer_clockid_to_base(clockid_t clock_id
)
88 return hrtimer_clock_to_base_table
[clock_id
];
93 * Get the coarse grained time at the softirq based on xtime and
96 static void hrtimer_get_softirq_time(struct hrtimer_cpu_base
*base
)
98 ktime_t xtim
, mono
, boot
;
99 struct timespec xts
, tom
, slp
;
101 get_xtime_and_monotonic_and_sleep_offset(&xts
, &tom
, &slp
);
103 xtim
= timespec_to_ktime(xts
);
104 mono
= ktime_add(xtim
, timespec_to_ktime(tom
));
105 boot
= ktime_add(mono
, timespec_to_ktime(slp
));
106 base
->clock_base
[HRTIMER_BASE_REALTIME
].softirq_time
= xtim
;
107 base
->clock_base
[HRTIMER_BASE_MONOTONIC
].softirq_time
= mono
;
108 base
->clock_base
[HRTIMER_BASE_BOOTTIME
].softirq_time
= boot
;
112 * Functions and macros which are different for UP/SMP systems are kept in a
118 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
119 * means that all timers which are tied to this base via timer->base are
120 * locked, and the base itself is locked too.
122 * So __run_timers/migrate_timers can safely modify all timers which could
123 * be found on the lists/queues.
125 * When the timer's base is locked, and the timer removed from list, it is
126 * possible to set timer->base = NULL and drop the lock: the timer remains
130 struct hrtimer_clock_base
*lock_hrtimer_base(const struct hrtimer
*timer
,
131 unsigned long *flags
)
133 struct hrtimer_clock_base
*base
;
137 if (likely(base
!= NULL
)) {
138 raw_spin_lock_irqsave(&base
->cpu_base
->lock
, *flags
);
139 if (likely(base
== timer
->base
))
141 /* The timer has migrated to another CPU: */
142 raw_spin_unlock_irqrestore(&base
->cpu_base
->lock
, *flags
);
150 * Get the preferred target CPU for NOHZ
152 static int hrtimer_get_target(int this_cpu
, int pinned
)
155 if (!pinned
&& get_sysctl_timer_migration() && idle_cpu(this_cpu
))
156 return get_nohz_timer_target();
162 * With HIGHRES=y we do not migrate the timer when it is expiring
163 * before the next event on the target cpu because we cannot reprogram
164 * the target cpu hardware and we would cause it to fire late.
166 * Called with cpu_base->lock of target cpu held.
169 hrtimer_check_target(struct hrtimer
*timer
, struct hrtimer_clock_base
*new_base
)
171 #ifdef CONFIG_HIGH_RES_TIMERS
174 if (!new_base
->cpu_base
->hres_active
)
177 expires
= ktime_sub(hrtimer_get_expires(timer
), new_base
->offset
);
178 return expires
.tv64
<= new_base
->cpu_base
->expires_next
.tv64
;
185 * Switch the timer base to the current CPU when possible.
187 static inline struct hrtimer_clock_base
*
188 switch_hrtimer_base(struct hrtimer
*timer
, struct hrtimer_clock_base
*base
,
191 struct hrtimer_clock_base
*new_base
;
192 struct hrtimer_cpu_base
*new_cpu_base
;
193 int this_cpu
= smp_processor_id();
194 int cpu
= hrtimer_get_target(this_cpu
, pinned
);
195 int basenum
= hrtimer_clockid_to_base(base
->index
);
198 new_cpu_base
= &per_cpu(hrtimer_bases
, cpu
);
199 new_base
= &new_cpu_base
->clock_base
[basenum
];
201 if (base
!= new_base
) {
203 * We are trying to move timer to new_base.
204 * However we can't change timer's base while it is running,
205 * so we keep it on the same CPU. No hassle vs. reprogramming
206 * the event source in the high resolution case. The softirq
207 * code will take care of this when the timer function has
208 * completed. There is no conflict as we hold the lock until
209 * the timer is enqueued.
211 if (unlikely(hrtimer_callback_running(timer
)))
214 /* See the comment in lock_timer_base() */
216 raw_spin_unlock(&base
->cpu_base
->lock
);
217 raw_spin_lock(&new_base
->cpu_base
->lock
);
219 if (cpu
!= this_cpu
&& hrtimer_check_target(timer
, new_base
)) {
221 raw_spin_unlock(&new_base
->cpu_base
->lock
);
222 raw_spin_lock(&base
->cpu_base
->lock
);
226 timer
->base
= new_base
;
231 #else /* CONFIG_SMP */
233 static inline struct hrtimer_clock_base
*
234 lock_hrtimer_base(const struct hrtimer
*timer
, unsigned long *flags
)
236 struct hrtimer_clock_base
*base
= timer
->base
;
238 raw_spin_lock_irqsave(&base
->cpu_base
->lock
, *flags
);
243 # define switch_hrtimer_base(t, b, p) (b)
245 #endif /* !CONFIG_SMP */
248 * Functions for the union type storage format of ktime_t which are
249 * too large for inlining:
251 #if BITS_PER_LONG < 64
252 # ifndef CONFIG_KTIME_SCALAR
254 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
256 * @nsec: the scalar nsec value to add
258 * Returns the sum of kt and nsec in ktime_t format
260 ktime_t
ktime_add_ns(const ktime_t kt
, u64 nsec
)
264 if (likely(nsec
< NSEC_PER_SEC
)) {
267 unsigned long rem
= do_div(nsec
, NSEC_PER_SEC
);
269 tmp
= ktime_set((long)nsec
, rem
);
272 return ktime_add(kt
, tmp
);
275 EXPORT_SYMBOL_GPL(ktime_add_ns
);
278 * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable
280 * @nsec: the scalar nsec value to subtract
282 * Returns the subtraction of @nsec from @kt in ktime_t format
284 ktime_t
ktime_sub_ns(const ktime_t kt
, u64 nsec
)
288 if (likely(nsec
< NSEC_PER_SEC
)) {
291 unsigned long rem
= do_div(nsec
, NSEC_PER_SEC
);
293 tmp
= ktime_set((long)nsec
, rem
);
296 return ktime_sub(kt
, tmp
);
299 EXPORT_SYMBOL_GPL(ktime_sub_ns
);
300 # endif /* !CONFIG_KTIME_SCALAR */
303 * Divide a ktime value by a nanosecond value
305 u64
ktime_divns(const ktime_t kt
, s64 div
)
310 dclc
= ktime_to_ns(kt
);
311 /* Make sure the divisor is less than 2^32: */
317 do_div(dclc
, (unsigned long) div
);
321 #endif /* BITS_PER_LONG >= 64 */
324 * Add two ktime values and do a safety check for overflow:
326 ktime_t
ktime_add_safe(const ktime_t lhs
, const ktime_t rhs
)
328 ktime_t res
= ktime_add(lhs
, rhs
);
331 * We use KTIME_SEC_MAX here, the maximum timeout which we can
332 * return to user space in a timespec:
334 if (res
.tv64
< 0 || res
.tv64
< lhs
.tv64
|| res
.tv64
< rhs
.tv64
)
335 res
= ktime_set(KTIME_SEC_MAX
, 0);
340 EXPORT_SYMBOL_GPL(ktime_add_safe
);
342 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
344 static struct debug_obj_descr hrtimer_debug_descr
;
347 * fixup_init is called when:
348 * - an active object is initialized
350 static int hrtimer_fixup_init(void *addr
, enum debug_obj_state state
)
352 struct hrtimer
*timer
= addr
;
355 case ODEBUG_STATE_ACTIVE
:
356 hrtimer_cancel(timer
);
357 debug_object_init(timer
, &hrtimer_debug_descr
);
365 * fixup_activate is called when:
366 * - an active object is activated
367 * - an unknown object is activated (might be a statically initialized object)
369 static int hrtimer_fixup_activate(void *addr
, enum debug_obj_state state
)
373 case ODEBUG_STATE_NOTAVAILABLE
:
377 case ODEBUG_STATE_ACTIVE
:
386 * fixup_free is called when:
387 * - an active object is freed
389 static int hrtimer_fixup_free(void *addr
, enum debug_obj_state state
)
391 struct hrtimer
*timer
= addr
;
394 case ODEBUG_STATE_ACTIVE
:
395 hrtimer_cancel(timer
);
396 debug_object_free(timer
, &hrtimer_debug_descr
);
403 static struct debug_obj_descr hrtimer_debug_descr
= {
405 .fixup_init
= hrtimer_fixup_init
,
406 .fixup_activate
= hrtimer_fixup_activate
,
407 .fixup_free
= hrtimer_fixup_free
,
410 static inline void debug_hrtimer_init(struct hrtimer
*timer
)
412 debug_object_init(timer
, &hrtimer_debug_descr
);
415 static inline void debug_hrtimer_activate(struct hrtimer
*timer
)
417 debug_object_activate(timer
, &hrtimer_debug_descr
);
420 static inline void debug_hrtimer_deactivate(struct hrtimer
*timer
)
422 debug_object_deactivate(timer
, &hrtimer_debug_descr
);
425 static inline void debug_hrtimer_free(struct hrtimer
*timer
)
427 debug_object_free(timer
, &hrtimer_debug_descr
);
430 static void __hrtimer_init(struct hrtimer
*timer
, clockid_t clock_id
,
431 enum hrtimer_mode mode
);
433 void hrtimer_init_on_stack(struct hrtimer
*timer
, clockid_t clock_id
,
434 enum hrtimer_mode mode
)
436 debug_object_init_on_stack(timer
, &hrtimer_debug_descr
);
437 __hrtimer_init(timer
, clock_id
, mode
);
439 EXPORT_SYMBOL_GPL(hrtimer_init_on_stack
);
441 void destroy_hrtimer_on_stack(struct hrtimer
*timer
)
443 debug_object_free(timer
, &hrtimer_debug_descr
);
447 static inline void debug_hrtimer_init(struct hrtimer
*timer
) { }
448 static inline void debug_hrtimer_activate(struct hrtimer
*timer
) { }
449 static inline void debug_hrtimer_deactivate(struct hrtimer
*timer
) { }
453 debug_init(struct hrtimer
*timer
, clockid_t clockid
,
454 enum hrtimer_mode mode
)
456 debug_hrtimer_init(timer
);
457 trace_hrtimer_init(timer
, clockid
, mode
);
460 static inline void debug_activate(struct hrtimer
*timer
)
462 debug_hrtimer_activate(timer
);
463 trace_hrtimer_start(timer
);
466 static inline void debug_deactivate(struct hrtimer
*timer
)
468 debug_hrtimer_deactivate(timer
);
469 trace_hrtimer_cancel(timer
);
472 /* High resolution timer related functions */
473 #ifdef CONFIG_HIGH_RES_TIMERS
476 * High resolution timer enabled ?
478 static int hrtimer_hres_enabled __read_mostly
= 1;
481 * Enable / Disable high resolution mode
483 static int __init
setup_hrtimer_hres(char *str
)
485 if (!strcmp(str
, "off"))
486 hrtimer_hres_enabled
= 0;
487 else if (!strcmp(str
, "on"))
488 hrtimer_hres_enabled
= 1;
494 __setup("highres=", setup_hrtimer_hres
);
497 * hrtimer_high_res_enabled - query, if the highres mode is enabled
499 static inline int hrtimer_is_hres_enabled(void)
501 return hrtimer_hres_enabled
;
505 * Is the high resolution mode active ?
507 static inline int hrtimer_hres_active(void)
509 return __this_cpu_read(hrtimer_bases
.hres_active
);
513 * Reprogram the event source with checking both queues for the
515 * Called with interrupts disabled and base->lock held
518 hrtimer_force_reprogram(struct hrtimer_cpu_base
*cpu_base
, int skip_equal
)
521 struct hrtimer_clock_base
*base
= cpu_base
->clock_base
;
522 ktime_t expires
, expires_next
;
524 expires_next
.tv64
= KTIME_MAX
;
526 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++, base
++) {
527 struct hrtimer
*timer
;
528 struct timerqueue_node
*next
;
530 next
= timerqueue_getnext(&base
->active
);
533 timer
= container_of(next
, struct hrtimer
, node
);
535 expires
= ktime_sub(hrtimer_get_expires(timer
), base
->offset
);
537 * clock_was_set() has changed base->offset so the
538 * result might be negative. Fix it up to prevent a
539 * false positive in clockevents_program_event()
541 if (expires
.tv64
< 0)
543 if (expires
.tv64
< expires_next
.tv64
)
544 expires_next
= expires
;
547 if (skip_equal
&& expires_next
.tv64
== cpu_base
->expires_next
.tv64
)
550 cpu_base
->expires_next
.tv64
= expires_next
.tv64
;
552 if (cpu_base
->expires_next
.tv64
!= KTIME_MAX
)
553 tick_program_event(cpu_base
->expires_next
, 1);
557 * Shared reprogramming for clock_realtime and clock_monotonic
559 * When a timer is enqueued and expires earlier than the already enqueued
560 * timers, we have to check, whether it expires earlier than the timer for
561 * which the clock event device was armed.
563 * Called with interrupts disabled and base->cpu_base.lock held
565 static int hrtimer_reprogram(struct hrtimer
*timer
,
566 struct hrtimer_clock_base
*base
)
568 struct hrtimer_cpu_base
*cpu_base
= &__get_cpu_var(hrtimer_bases
);
569 ktime_t expires
= ktime_sub(hrtimer_get_expires(timer
), base
->offset
);
572 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer
) < 0);
575 * When the callback is running, we do not reprogram the clock event
576 * device. The timer callback is either running on a different CPU or
577 * the callback is executed in the hrtimer_interrupt context. The
578 * reprogramming is handled either by the softirq, which called the
579 * callback or at the end of the hrtimer_interrupt.
581 if (hrtimer_callback_running(timer
))
585 * CLOCK_REALTIME timer might be requested with an absolute
586 * expiry time which is less than base->offset. Nothing wrong
587 * about that, just avoid to call into the tick code, which
588 * has now objections against negative expiry values.
590 if (expires
.tv64
< 0)
593 if (expires
.tv64
>= cpu_base
->expires_next
.tv64
)
597 * If a hang was detected in the last timer interrupt then we
598 * do not schedule a timer which is earlier than the expiry
599 * which we enforced in the hang detection. We want the system
602 if (cpu_base
->hang_detected
)
606 * Clockevents returns -ETIME, when the event was in the past.
608 res
= tick_program_event(expires
, 0);
609 if (!IS_ERR_VALUE(res
))
610 cpu_base
->expires_next
= expires
;
616 * Retrigger next event is called after clock was set
618 * Called with interrupts disabled via on_each_cpu()
620 static void retrigger_next_event(void *arg
)
622 struct hrtimer_cpu_base
*base
;
623 struct timespec realtime_offset
, wtm
, sleep
;
625 if (!hrtimer_hres_active())
628 get_xtime_and_monotonic_and_sleep_offset(&realtime_offset
, &wtm
,
630 set_normalized_timespec(&realtime_offset
, -wtm
.tv_sec
, -wtm
.tv_nsec
);
632 base
= &__get_cpu_var(hrtimer_bases
);
634 /* Adjust CLOCK_REALTIME offset */
635 raw_spin_lock(&base
->lock
);
636 base
->clock_base
[HRTIMER_BASE_REALTIME
].offset
=
637 timespec_to_ktime(realtime_offset
);
639 hrtimer_force_reprogram(base
, 0);
640 raw_spin_unlock(&base
->lock
);
644 * Clock realtime was set
646 * Change the offset of the realtime clock vs. the monotonic
649 * We might have to reprogram the high resolution timer interrupt. On
650 * SMP we call the architecture specific code to retrigger _all_ high
651 * resolution timer interrupts. On UP we just disable interrupts and
652 * call the high resolution interrupt code.
654 void clock_was_set(void)
656 /* Retrigger the CPU local events everywhere */
657 on_each_cpu(retrigger_next_event
, NULL
, 1);
661 * During resume we might have to reprogram the high resolution timer
662 * interrupt (on the local CPU):
664 void hres_timers_resume(void)
666 WARN_ONCE(!irqs_disabled(),
667 KERN_INFO
"hres_timers_resume() called with IRQs enabled!");
669 retrigger_next_event(NULL
);
673 * Initialize the high resolution related parts of cpu_base
675 static inline void hrtimer_init_hres(struct hrtimer_cpu_base
*base
)
677 base
->expires_next
.tv64
= KTIME_MAX
;
678 base
->hres_active
= 0;
682 * Initialize the high resolution related parts of a hrtimer
684 static inline void hrtimer_init_timer_hres(struct hrtimer
*timer
)
690 * When High resolution timers are active, try to reprogram. Note, that in case
691 * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry
692 * check happens. The timer gets enqueued into the rbtree. The reprogramming
693 * and expiry check is done in the hrtimer_interrupt or in the softirq.
695 static inline int hrtimer_enqueue_reprogram(struct hrtimer
*timer
,
696 struct hrtimer_clock_base
*base
,
699 if (base
->cpu_base
->hres_active
&& hrtimer_reprogram(timer
, base
)) {
701 raw_spin_unlock(&base
->cpu_base
->lock
);
702 raise_softirq_irqoff(HRTIMER_SOFTIRQ
);
703 raw_spin_lock(&base
->cpu_base
->lock
);
705 __raise_softirq_irqoff(HRTIMER_SOFTIRQ
);
714 * Switch to high resolution mode
716 static int hrtimer_switch_to_hres(void)
718 int cpu
= smp_processor_id();
719 struct hrtimer_cpu_base
*base
= &per_cpu(hrtimer_bases
, cpu
);
722 if (base
->hres_active
)
725 local_irq_save(flags
);
727 if (tick_init_highres()) {
728 local_irq_restore(flags
);
729 printk(KERN_WARNING
"Could not switch to high resolution "
730 "mode on CPU %d\n", cpu
);
733 base
->hres_active
= 1;
734 base
->clock_base
[HRTIMER_BASE_REALTIME
].resolution
= KTIME_HIGH_RES
;
735 base
->clock_base
[HRTIMER_BASE_MONOTONIC
].resolution
= KTIME_HIGH_RES
;
736 base
->clock_base
[HRTIMER_BASE_BOOTTIME
].resolution
= KTIME_HIGH_RES
;
738 tick_setup_sched_timer();
740 /* "Retrigger" the interrupt to get things going */
741 retrigger_next_event(NULL
);
742 local_irq_restore(flags
);
748 static inline int hrtimer_hres_active(void) { return 0; }
749 static inline int hrtimer_is_hres_enabled(void) { return 0; }
750 static inline int hrtimer_switch_to_hres(void) { return 0; }
752 hrtimer_force_reprogram(struct hrtimer_cpu_base
*base
, int skip_equal
) { }
753 static inline int hrtimer_enqueue_reprogram(struct hrtimer
*timer
,
754 struct hrtimer_clock_base
*base
,
759 static inline void hrtimer_init_hres(struct hrtimer_cpu_base
*base
) { }
760 static inline void hrtimer_init_timer_hres(struct hrtimer
*timer
) { }
762 #endif /* CONFIG_HIGH_RES_TIMERS */
764 static inline void timer_stats_hrtimer_set_start_info(struct hrtimer
*timer
)
766 #ifdef CONFIG_TIMER_STATS
767 if (timer
->start_site
)
769 timer
->start_site
= __builtin_return_address(0);
770 memcpy(timer
->start_comm
, current
->comm
, TASK_COMM_LEN
);
771 timer
->start_pid
= current
->pid
;
775 static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer
*timer
)
777 #ifdef CONFIG_TIMER_STATS
778 timer
->start_site
= NULL
;
782 static inline void timer_stats_account_hrtimer(struct hrtimer
*timer
)
784 #ifdef CONFIG_TIMER_STATS
785 if (likely(!timer_stats_active
))
787 timer_stats_update_stats(timer
, timer
->start_pid
, timer
->start_site
,
788 timer
->function
, timer
->start_comm
, 0);
793 * Counterpart to lock_hrtimer_base above:
796 void unlock_hrtimer_base(const struct hrtimer
*timer
, unsigned long *flags
)
798 raw_spin_unlock_irqrestore(&timer
->base
->cpu_base
->lock
, *flags
);
802 * hrtimer_forward - forward the timer expiry
803 * @timer: hrtimer to forward
804 * @now: forward past this time
805 * @interval: the interval to forward
807 * Forward the timer expiry so it will expire in the future.
808 * Returns the number of overruns.
810 u64
hrtimer_forward(struct hrtimer
*timer
, ktime_t now
, ktime_t interval
)
815 delta
= ktime_sub(now
, hrtimer_get_expires(timer
));
820 if (interval
.tv64
< timer
->base
->resolution
.tv64
)
821 interval
.tv64
= timer
->base
->resolution
.tv64
;
823 if (unlikely(delta
.tv64
>= interval
.tv64
)) {
824 s64 incr
= ktime_to_ns(interval
);
826 orun
= ktime_divns(delta
, incr
);
827 hrtimer_add_expires_ns(timer
, incr
* orun
);
828 if (hrtimer_get_expires_tv64(timer
) > now
.tv64
)
831 * This (and the ktime_add() below) is the
832 * correction for exact:
836 hrtimer_add_expires(timer
, interval
);
840 EXPORT_SYMBOL_GPL(hrtimer_forward
);
843 * enqueue_hrtimer - internal function to (re)start a timer
845 * The timer is inserted in expiry order. Insertion into the
846 * red black tree is O(log(n)). Must hold the base lock.
848 * Returns 1 when the new timer is the leftmost timer in the tree.
850 static int enqueue_hrtimer(struct hrtimer
*timer
,
851 struct hrtimer_clock_base
*base
)
853 debug_activate(timer
);
855 timerqueue_add(&base
->active
, &timer
->node
);
858 * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
859 * state of a possibly running callback.
861 timer
->state
|= HRTIMER_STATE_ENQUEUED
;
863 return (&timer
->node
== base
->active
.next
);
867 * __remove_hrtimer - internal function to remove a timer
869 * Caller must hold the base lock.
871 * High resolution timer mode reprograms the clock event device when the
872 * timer is the one which expires next. The caller can disable this by setting
873 * reprogram to zero. This is useful, when the context does a reprogramming
874 * anyway (e.g. timer interrupt)
876 static void __remove_hrtimer(struct hrtimer
*timer
,
877 struct hrtimer_clock_base
*base
,
878 unsigned long newstate
, int reprogram
)
880 if (!(timer
->state
& HRTIMER_STATE_ENQUEUED
))
883 if (&timer
->node
== timerqueue_getnext(&base
->active
)) {
884 #ifdef CONFIG_HIGH_RES_TIMERS
885 /* Reprogram the clock event device. if enabled */
886 if (reprogram
&& hrtimer_hres_active()) {
889 expires
= ktime_sub(hrtimer_get_expires(timer
),
891 if (base
->cpu_base
->expires_next
.tv64
== expires
.tv64
)
892 hrtimer_force_reprogram(base
->cpu_base
, 1);
896 timerqueue_del(&base
->active
, &timer
->node
);
898 timer
->state
= newstate
;
902 * remove hrtimer, called with base lock held
905 remove_hrtimer(struct hrtimer
*timer
, struct hrtimer_clock_base
*base
)
907 if (hrtimer_is_queued(timer
)) {
912 * Remove the timer and force reprogramming when high
913 * resolution mode is active and the timer is on the current
914 * CPU. If we remove a timer on another CPU, reprogramming is
915 * skipped. The interrupt event on this CPU is fired and
916 * reprogramming happens in the interrupt handler. This is a
917 * rare case and less expensive than a smp call.
919 debug_deactivate(timer
);
920 timer_stats_hrtimer_clear_start_info(timer
);
921 reprogram
= base
->cpu_base
== &__get_cpu_var(hrtimer_bases
);
923 * We must preserve the CALLBACK state flag here,
924 * otherwise we could move the timer base in
925 * switch_hrtimer_base.
927 state
= timer
->state
& HRTIMER_STATE_CALLBACK
;
928 __remove_hrtimer(timer
, base
, state
, reprogram
);
934 int __hrtimer_start_range_ns(struct hrtimer
*timer
, ktime_t tim
,
935 unsigned long delta_ns
, const enum hrtimer_mode mode
,
938 struct hrtimer_clock_base
*base
, *new_base
;
942 base
= lock_hrtimer_base(timer
, &flags
);
944 /* Remove an active timer from the queue: */
945 ret
= remove_hrtimer(timer
, base
);
947 /* Switch the timer base, if necessary: */
948 new_base
= switch_hrtimer_base(timer
, base
, mode
& HRTIMER_MODE_PINNED
);
950 if (mode
& HRTIMER_MODE_REL
) {
951 tim
= ktime_add_safe(tim
, new_base
->get_time());
953 * CONFIG_TIME_LOW_RES is a temporary way for architectures
954 * to signal that they simply return xtime in
955 * do_gettimeoffset(). In this case we want to round up by
956 * resolution when starting a relative timer, to avoid short
957 * timeouts. This will go away with the GTOD framework.
959 #ifdef CONFIG_TIME_LOW_RES
960 tim
= ktime_add_safe(tim
, base
->resolution
);
964 hrtimer_set_expires_range_ns(timer
, tim
, delta_ns
);
966 timer_stats_hrtimer_set_start_info(timer
);
968 leftmost
= enqueue_hrtimer(timer
, new_base
);
971 * Only allow reprogramming if the new base is on this CPU.
972 * (it might still be on another CPU if the timer was pending)
974 * XXX send_remote_softirq() ?
976 if (leftmost
&& new_base
->cpu_base
== &__get_cpu_var(hrtimer_bases
))
977 hrtimer_enqueue_reprogram(timer
, new_base
, wakeup
);
979 unlock_hrtimer_base(timer
, &flags
);
985 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
986 * @timer: the timer to be added
988 * @delta_ns: "slack" range for the timer
989 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
993 * 1 when the timer was active
995 int hrtimer_start_range_ns(struct hrtimer
*timer
, ktime_t tim
,
996 unsigned long delta_ns
, const enum hrtimer_mode mode
)
998 return __hrtimer_start_range_ns(timer
, tim
, delta_ns
, mode
, 1);
1000 EXPORT_SYMBOL_GPL(hrtimer_start_range_ns
);
1003 * hrtimer_start - (re)start an hrtimer on the current CPU
1004 * @timer: the timer to be added
1006 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
1010 * 1 when the timer was active
1013 hrtimer_start(struct hrtimer
*timer
, ktime_t tim
, const enum hrtimer_mode mode
)
1015 return __hrtimer_start_range_ns(timer
, tim
, 0, mode
, 1);
1017 EXPORT_SYMBOL_GPL(hrtimer_start
);
1021 * hrtimer_try_to_cancel - try to deactivate a timer
1022 * @timer: hrtimer to stop
1025 * 0 when the timer was not active
1026 * 1 when the timer was active
1027 * -1 when the timer is currently excuting the callback function and
1030 int hrtimer_try_to_cancel(struct hrtimer
*timer
)
1032 struct hrtimer_clock_base
*base
;
1033 unsigned long flags
;
1036 base
= lock_hrtimer_base(timer
, &flags
);
1038 if (!hrtimer_callback_running(timer
))
1039 ret
= remove_hrtimer(timer
, base
);
1041 unlock_hrtimer_base(timer
, &flags
);
1046 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel
);
1049 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1050 * @timer: the timer to be cancelled
1053 * 0 when the timer was not active
1054 * 1 when the timer was active
1056 int hrtimer_cancel(struct hrtimer
*timer
)
1059 int ret
= hrtimer_try_to_cancel(timer
);
1066 EXPORT_SYMBOL_GPL(hrtimer_cancel
);
1069 * hrtimer_get_remaining - get remaining time for the timer
1070 * @timer: the timer to read
1072 ktime_t
hrtimer_get_remaining(const struct hrtimer
*timer
)
1074 unsigned long flags
;
1077 lock_hrtimer_base(timer
, &flags
);
1078 rem
= hrtimer_expires_remaining(timer
);
1079 unlock_hrtimer_base(timer
, &flags
);
1083 EXPORT_SYMBOL_GPL(hrtimer_get_remaining
);
1087 * hrtimer_get_next_event - get the time until next expiry event
1089 * Returns the delta to the next expiry event or KTIME_MAX if no timer
1092 ktime_t
hrtimer_get_next_event(void)
1094 struct hrtimer_cpu_base
*cpu_base
= &__get_cpu_var(hrtimer_bases
);
1095 struct hrtimer_clock_base
*base
= cpu_base
->clock_base
;
1096 ktime_t delta
, mindelta
= { .tv64
= KTIME_MAX
};
1097 unsigned long flags
;
1100 raw_spin_lock_irqsave(&cpu_base
->lock
, flags
);
1102 if (!hrtimer_hres_active()) {
1103 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++, base
++) {
1104 struct hrtimer
*timer
;
1105 struct timerqueue_node
*next
;
1107 next
= timerqueue_getnext(&base
->active
);
1111 timer
= container_of(next
, struct hrtimer
, node
);
1112 delta
.tv64
= hrtimer_get_expires_tv64(timer
);
1113 delta
= ktime_sub(delta
, base
->get_time());
1114 if (delta
.tv64
< mindelta
.tv64
)
1115 mindelta
.tv64
= delta
.tv64
;
1119 raw_spin_unlock_irqrestore(&cpu_base
->lock
, flags
);
1121 if (mindelta
.tv64
< 0)
1127 static void __hrtimer_init(struct hrtimer
*timer
, clockid_t clock_id
,
1128 enum hrtimer_mode mode
)
1130 struct hrtimer_cpu_base
*cpu_base
;
1133 memset(timer
, 0, sizeof(struct hrtimer
));
1135 cpu_base
= &__raw_get_cpu_var(hrtimer_bases
);
1137 if (clock_id
== CLOCK_REALTIME
&& mode
!= HRTIMER_MODE_ABS
)
1138 clock_id
= CLOCK_MONOTONIC
;
1140 base
= hrtimer_clockid_to_base(clock_id
);
1141 timer
->base
= &cpu_base
->clock_base
[base
];
1142 hrtimer_init_timer_hres(timer
);
1143 timerqueue_init(&timer
->node
);
1145 #ifdef CONFIG_TIMER_STATS
1146 timer
->start_site
= NULL
;
1147 timer
->start_pid
= -1;
1148 memset(timer
->start_comm
, 0, TASK_COMM_LEN
);
1153 * hrtimer_init - initialize a timer to the given clock
1154 * @timer: the timer to be initialized
1155 * @clock_id: the clock to be used
1156 * @mode: timer mode abs/rel
1158 void hrtimer_init(struct hrtimer
*timer
, clockid_t clock_id
,
1159 enum hrtimer_mode mode
)
1161 debug_init(timer
, clock_id
, mode
);
1162 __hrtimer_init(timer
, clock_id
, mode
);
1164 EXPORT_SYMBOL_GPL(hrtimer_init
);
1167 * hrtimer_get_res - get the timer resolution for a clock
1168 * @which_clock: which clock to query
1169 * @tp: pointer to timespec variable to store the resolution
1171 * Store the resolution of the clock selected by @which_clock in the
1172 * variable pointed to by @tp.
1174 int hrtimer_get_res(const clockid_t which_clock
, struct timespec
*tp
)
1176 struct hrtimer_cpu_base
*cpu_base
;
1177 int base
= hrtimer_clockid_to_base(which_clock
);
1179 cpu_base
= &__raw_get_cpu_var(hrtimer_bases
);
1180 *tp
= ktime_to_timespec(cpu_base
->clock_base
[base
].resolution
);
1184 EXPORT_SYMBOL_GPL(hrtimer_get_res
);
1186 static void __run_hrtimer(struct hrtimer
*timer
, ktime_t
*now
)
1188 struct hrtimer_clock_base
*base
= timer
->base
;
1189 struct hrtimer_cpu_base
*cpu_base
= base
->cpu_base
;
1190 enum hrtimer_restart (*fn
)(struct hrtimer
*);
1193 WARN_ON(!irqs_disabled());
1195 debug_deactivate(timer
);
1196 __remove_hrtimer(timer
, base
, HRTIMER_STATE_CALLBACK
, 0);
1197 timer_stats_account_hrtimer(timer
);
1198 fn
= timer
->function
;
1201 * Because we run timers from hardirq context, there is no chance
1202 * they get migrated to another cpu, therefore its safe to unlock
1205 raw_spin_unlock(&cpu_base
->lock
);
1206 trace_hrtimer_expire_entry(timer
, now
);
1207 restart
= fn(timer
);
1208 trace_hrtimer_expire_exit(timer
);
1209 raw_spin_lock(&cpu_base
->lock
);
1212 * Note: We clear the CALLBACK bit after enqueue_hrtimer and
1213 * we do not reprogramm the event hardware. Happens either in
1214 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1216 if (restart
!= HRTIMER_NORESTART
) {
1217 BUG_ON(timer
->state
!= HRTIMER_STATE_CALLBACK
);
1218 enqueue_hrtimer(timer
, base
);
1221 WARN_ON_ONCE(!(timer
->state
& HRTIMER_STATE_CALLBACK
));
1223 timer
->state
&= ~HRTIMER_STATE_CALLBACK
;
1226 #ifdef CONFIG_HIGH_RES_TIMERS
1229 * High resolution timer interrupt
1230 * Called with interrupts disabled
1232 void hrtimer_interrupt(struct clock_event_device
*dev
)
1234 struct hrtimer_cpu_base
*cpu_base
= &__get_cpu_var(hrtimer_bases
);
1235 struct hrtimer_clock_base
*base
;
1236 ktime_t expires_next
, now
, entry_time
, delta
;
1239 BUG_ON(!cpu_base
->hres_active
);
1240 cpu_base
->nr_events
++;
1241 dev
->next_event
.tv64
= KTIME_MAX
;
1243 entry_time
= now
= ktime_get();
1245 expires_next
.tv64
= KTIME_MAX
;
1247 raw_spin_lock(&cpu_base
->lock
);
1249 * We set expires_next to KTIME_MAX here with cpu_base->lock
1250 * held to prevent that a timer is enqueued in our queue via
1251 * the migration code. This does not affect enqueueing of
1252 * timers which run their callback and need to be requeued on
1255 cpu_base
->expires_next
.tv64
= KTIME_MAX
;
1257 base
= cpu_base
->clock_base
;
1259 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++) {
1261 struct timerqueue_node
*node
;
1263 basenow
= ktime_add(now
, base
->offset
);
1265 while ((node
= timerqueue_getnext(&base
->active
))) {
1266 struct hrtimer
*timer
;
1268 timer
= container_of(node
, struct hrtimer
, node
);
1271 * The immediate goal for using the softexpires is
1272 * minimizing wakeups, not running timers at the
1273 * earliest interrupt after their soft expiration.
1274 * This allows us to avoid using a Priority Search
1275 * Tree, which can answer a stabbing querry for
1276 * overlapping intervals and instead use the simple
1277 * BST we already have.
1278 * We don't add extra wakeups by delaying timers that
1279 * are right-of a not yet expired timer, because that
1280 * timer will have to trigger a wakeup anyway.
1283 if (basenow
.tv64
< hrtimer_get_softexpires_tv64(timer
)) {
1286 expires
= ktime_sub(hrtimer_get_expires(timer
),
1288 if (expires
.tv64
< expires_next
.tv64
)
1289 expires_next
= expires
;
1293 __run_hrtimer(timer
, &basenow
);
1299 * Store the new expiry value so the migration code can verify
1302 cpu_base
->expires_next
= expires_next
;
1303 raw_spin_unlock(&cpu_base
->lock
);
1305 /* Reprogramming necessary ? */
1306 if (expires_next
.tv64
== KTIME_MAX
||
1307 !tick_program_event(expires_next
, 0)) {
1308 cpu_base
->hang_detected
= 0;
1313 * The next timer was already expired due to:
1315 * - long lasting callbacks
1316 * - being scheduled away when running in a VM
1318 * We need to prevent that we loop forever in the hrtimer
1319 * interrupt routine. We give it 3 attempts to avoid
1320 * overreacting on some spurious event.
1323 cpu_base
->nr_retries
++;
1327 * Give the system a chance to do something else than looping
1328 * here. We stored the entry time, so we know exactly how long
1329 * we spent here. We schedule the next event this amount of
1332 cpu_base
->nr_hangs
++;
1333 cpu_base
->hang_detected
= 1;
1334 delta
= ktime_sub(now
, entry_time
);
1335 if (delta
.tv64
> cpu_base
->max_hang_time
.tv64
)
1336 cpu_base
->max_hang_time
= delta
;
1338 * Limit it to a sensible value as we enforce a longer
1339 * delay. Give the CPU at least 100ms to catch up.
1341 if (delta
.tv64
> 100 * NSEC_PER_MSEC
)
1342 expires_next
= ktime_add_ns(now
, 100 * NSEC_PER_MSEC
);
1344 expires_next
= ktime_add(now
, delta
);
1345 tick_program_event(expires_next
, 1);
1346 printk_once(KERN_WARNING
"hrtimer: interrupt took %llu ns\n",
1347 ktime_to_ns(delta
));
1351 * local version of hrtimer_peek_ahead_timers() called with interrupts
1354 static void __hrtimer_peek_ahead_timers(void)
1356 struct tick_device
*td
;
1358 if (!hrtimer_hres_active())
1361 td
= &__get_cpu_var(tick_cpu_device
);
1362 if (td
&& td
->evtdev
)
1363 hrtimer_interrupt(td
->evtdev
);
1367 * hrtimer_peek_ahead_timers -- run soft-expired timers now
1369 * hrtimer_peek_ahead_timers will peek at the timer queue of
1370 * the current cpu and check if there are any timers for which
1371 * the soft expires time has passed. If any such timers exist,
1372 * they are run immediately and then removed from the timer queue.
1375 void hrtimer_peek_ahead_timers(void)
1377 unsigned long flags
;
1379 local_irq_save(flags
);
1380 __hrtimer_peek_ahead_timers();
1381 local_irq_restore(flags
);
1384 static void run_hrtimer_softirq(struct softirq_action
*h
)
1386 hrtimer_peek_ahead_timers();
1389 #else /* CONFIG_HIGH_RES_TIMERS */
1391 static inline void __hrtimer_peek_ahead_timers(void) { }
1393 #endif /* !CONFIG_HIGH_RES_TIMERS */
1396 * Called from timer softirq every jiffy, expire hrtimers:
1398 * For HRT its the fall back code to run the softirq in the timer
1399 * softirq context in case the hrtimer initialization failed or has
1400 * not been done yet.
1402 void hrtimer_run_pending(void)
1404 if (hrtimer_hres_active())
1408 * This _is_ ugly: We have to check in the softirq context,
1409 * whether we can switch to highres and / or nohz mode. The
1410 * clocksource switch happens in the timer interrupt with
1411 * xtime_lock held. Notification from there only sets the
1412 * check bit in the tick_oneshot code, otherwise we might
1413 * deadlock vs. xtime_lock.
1415 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1416 hrtimer_switch_to_hres();
1420 * Called from hardirq context every jiffy
1422 void hrtimer_run_queues(void)
1424 struct timerqueue_node
*node
;
1425 struct hrtimer_cpu_base
*cpu_base
= &__get_cpu_var(hrtimer_bases
);
1426 struct hrtimer_clock_base
*base
;
1427 int index
, gettime
= 1;
1429 if (hrtimer_hres_active())
1432 for (index
= 0; index
< HRTIMER_MAX_CLOCK_BASES
; index
++) {
1433 base
= &cpu_base
->clock_base
[index
];
1434 if (!timerqueue_getnext(&base
->active
))
1438 hrtimer_get_softirq_time(cpu_base
);
1442 raw_spin_lock(&cpu_base
->lock
);
1444 while ((node
= timerqueue_getnext(&base
->active
))) {
1445 struct hrtimer
*timer
;
1447 timer
= container_of(node
, struct hrtimer
, node
);
1448 if (base
->softirq_time
.tv64
<=
1449 hrtimer_get_expires_tv64(timer
))
1452 __run_hrtimer(timer
, &base
->softirq_time
);
1454 raw_spin_unlock(&cpu_base
->lock
);
1459 * Sleep related functions:
1461 static enum hrtimer_restart
hrtimer_wakeup(struct hrtimer
*timer
)
1463 struct hrtimer_sleeper
*t
=
1464 container_of(timer
, struct hrtimer_sleeper
, timer
);
1465 struct task_struct
*task
= t
->task
;
1469 wake_up_process(task
);
1471 return HRTIMER_NORESTART
;
1474 void hrtimer_init_sleeper(struct hrtimer_sleeper
*sl
, struct task_struct
*task
)
1476 sl
->timer
.function
= hrtimer_wakeup
;
1479 EXPORT_SYMBOL_GPL(hrtimer_init_sleeper
);
1481 static int __sched
do_nanosleep(struct hrtimer_sleeper
*t
, enum hrtimer_mode mode
)
1483 hrtimer_init_sleeper(t
, current
);
1486 set_current_state(TASK_INTERRUPTIBLE
);
1487 hrtimer_start_expires(&t
->timer
, mode
);
1488 if (!hrtimer_active(&t
->timer
))
1491 if (likely(t
->task
))
1494 hrtimer_cancel(&t
->timer
);
1495 mode
= HRTIMER_MODE_ABS
;
1497 } while (t
->task
&& !signal_pending(current
));
1499 __set_current_state(TASK_RUNNING
);
1501 return t
->task
== NULL
;
1504 static int update_rmtp(struct hrtimer
*timer
, struct timespec __user
*rmtp
)
1506 struct timespec rmt
;
1509 rem
= hrtimer_expires_remaining(timer
);
1512 rmt
= ktime_to_timespec(rem
);
1514 if (copy_to_user(rmtp
, &rmt
, sizeof(*rmtp
)))
1520 long __sched
hrtimer_nanosleep_restart(struct restart_block
*restart
)
1522 struct hrtimer_sleeper t
;
1523 struct timespec __user
*rmtp
;
1526 hrtimer_init_on_stack(&t
.timer
, restart
->nanosleep
.index
,
1528 hrtimer_set_expires_tv64(&t
.timer
, restart
->nanosleep
.expires
);
1530 if (do_nanosleep(&t
, HRTIMER_MODE_ABS
))
1533 rmtp
= restart
->nanosleep
.rmtp
;
1535 ret
= update_rmtp(&t
.timer
, rmtp
);
1540 /* The other values in restart are already filled in */
1541 ret
= -ERESTART_RESTARTBLOCK
;
1543 destroy_hrtimer_on_stack(&t
.timer
);
1547 long hrtimer_nanosleep(struct timespec
*rqtp
, struct timespec __user
*rmtp
,
1548 const enum hrtimer_mode mode
, const clockid_t clockid
)
1550 struct restart_block
*restart
;
1551 struct hrtimer_sleeper t
;
1553 unsigned long slack
;
1555 slack
= current
->timer_slack_ns
;
1556 if (rt_task(current
))
1559 hrtimer_init_on_stack(&t
.timer
, clockid
, mode
);
1560 hrtimer_set_expires_range_ns(&t
.timer
, timespec_to_ktime(*rqtp
), slack
);
1561 if (do_nanosleep(&t
, mode
))
1564 /* Absolute timers do not update the rmtp value and restart: */
1565 if (mode
== HRTIMER_MODE_ABS
) {
1566 ret
= -ERESTARTNOHAND
;
1571 ret
= update_rmtp(&t
.timer
, rmtp
);
1576 restart
= ¤t_thread_info()->restart_block
;
1577 restart
->fn
= hrtimer_nanosleep_restart
;
1578 restart
->nanosleep
.index
= t
.timer
.base
->index
;
1579 restart
->nanosleep
.rmtp
= rmtp
;
1580 restart
->nanosleep
.expires
= hrtimer_get_expires_tv64(&t
.timer
);
1582 ret
= -ERESTART_RESTARTBLOCK
;
1584 destroy_hrtimer_on_stack(&t
.timer
);
1588 SYSCALL_DEFINE2(nanosleep
, struct timespec __user
*, rqtp
,
1589 struct timespec __user
*, rmtp
)
1593 if (copy_from_user(&tu
, rqtp
, sizeof(tu
)))
1596 if (!timespec_valid(&tu
))
1599 return hrtimer_nanosleep(&tu
, rmtp
, HRTIMER_MODE_REL
, CLOCK_MONOTONIC
);
1603 * Functions related to boot-time initialization:
1605 static void __cpuinit
init_hrtimers_cpu(int cpu
)
1607 struct hrtimer_cpu_base
*cpu_base
= &per_cpu(hrtimer_bases
, cpu
);
1610 raw_spin_lock_init(&cpu_base
->lock
);
1612 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++) {
1613 cpu_base
->clock_base
[i
].cpu_base
= cpu_base
;
1614 timerqueue_init_head(&cpu_base
->clock_base
[i
].active
);
1617 hrtimer_init_hres(cpu_base
);
1620 #ifdef CONFIG_HOTPLUG_CPU
1622 static void migrate_hrtimer_list(struct hrtimer_clock_base
*old_base
,
1623 struct hrtimer_clock_base
*new_base
)
1625 struct hrtimer
*timer
;
1626 struct timerqueue_node
*node
;
1628 while ((node
= timerqueue_getnext(&old_base
->active
))) {
1629 timer
= container_of(node
, struct hrtimer
, node
);
1630 BUG_ON(hrtimer_callback_running(timer
));
1631 debug_deactivate(timer
);
1634 * Mark it as STATE_MIGRATE not INACTIVE otherwise the
1635 * timer could be seen as !active and just vanish away
1636 * under us on another CPU
1638 __remove_hrtimer(timer
, old_base
, HRTIMER_STATE_MIGRATE
, 0);
1639 timer
->base
= new_base
;
1641 * Enqueue the timers on the new cpu. This does not
1642 * reprogram the event device in case the timer
1643 * expires before the earliest on this CPU, but we run
1644 * hrtimer_interrupt after we migrated everything to
1645 * sort out already expired timers and reprogram the
1648 enqueue_hrtimer(timer
, new_base
);
1650 /* Clear the migration state bit */
1651 timer
->state
&= ~HRTIMER_STATE_MIGRATE
;
1655 static void migrate_hrtimers(int scpu
)
1657 struct hrtimer_cpu_base
*old_base
, *new_base
;
1660 BUG_ON(cpu_online(scpu
));
1661 tick_cancel_sched_timer(scpu
);
1663 local_irq_disable();
1664 old_base
= &per_cpu(hrtimer_bases
, scpu
);
1665 new_base
= &__get_cpu_var(hrtimer_bases
);
1667 * The caller is globally serialized and nobody else
1668 * takes two locks at once, deadlock is not possible.
1670 raw_spin_lock(&new_base
->lock
);
1671 raw_spin_lock_nested(&old_base
->lock
, SINGLE_DEPTH_NESTING
);
1673 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++) {
1674 migrate_hrtimer_list(&old_base
->clock_base
[i
],
1675 &new_base
->clock_base
[i
]);
1678 raw_spin_unlock(&old_base
->lock
);
1679 raw_spin_unlock(&new_base
->lock
);
1681 /* Check, if we got expired work to do */
1682 __hrtimer_peek_ahead_timers();
1686 #endif /* CONFIG_HOTPLUG_CPU */
1688 static int __cpuinit
hrtimer_cpu_notify(struct notifier_block
*self
,
1689 unsigned long action
, void *hcpu
)
1691 int scpu
= (long)hcpu
;
1695 case CPU_UP_PREPARE
:
1696 case CPU_UP_PREPARE_FROZEN
:
1697 init_hrtimers_cpu(scpu
);
1700 #ifdef CONFIG_HOTPLUG_CPU
1702 case CPU_DYING_FROZEN
:
1703 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DYING
, &scpu
);
1706 case CPU_DEAD_FROZEN
:
1708 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD
, &scpu
);
1709 migrate_hrtimers(scpu
);
1721 static struct notifier_block __cpuinitdata hrtimers_nb
= {
1722 .notifier_call
= hrtimer_cpu_notify
,
1725 void __init
hrtimers_init(void)
1727 hrtimer_clock_to_base_table
[CLOCK_REALTIME
] = HRTIMER_BASE_REALTIME
;
1728 hrtimer_clock_to_base_table
[CLOCK_MONOTONIC
] = HRTIMER_BASE_MONOTONIC
;
1729 hrtimer_clock_to_base_table
[CLOCK_BOOTTIME
] = HRTIMER_BASE_BOOTTIME
;
1731 hrtimer_cpu_notify(&hrtimers_nb
, (unsigned long)CPU_UP_PREPARE
,
1732 (void *)(long)smp_processor_id());
1733 register_cpu_notifier(&hrtimers_nb
);
1734 #ifdef CONFIG_HIGH_RES_TIMERS
1735 open_softirq(HRTIMER_SOFTIRQ
, run_hrtimer_softirq
);
1740 * schedule_hrtimeout_range_clock - sleep until timeout
1741 * @expires: timeout value (ktime_t)
1742 * @delta: slack in expires timeout (ktime_t)
1743 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1744 * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME
1747 schedule_hrtimeout_range_clock(ktime_t
*expires
, unsigned long delta
,
1748 const enum hrtimer_mode mode
, int clock
)
1750 struct hrtimer_sleeper t
;
1753 * Optimize when a zero timeout value is given. It does not
1754 * matter whether this is an absolute or a relative time.
1756 if (expires
&& !expires
->tv64
) {
1757 __set_current_state(TASK_RUNNING
);
1762 * A NULL parameter means "infinite"
1766 __set_current_state(TASK_RUNNING
);
1770 hrtimer_init_on_stack(&t
.timer
, clock
, mode
);
1771 hrtimer_set_expires_range_ns(&t
.timer
, *expires
, delta
);
1773 hrtimer_init_sleeper(&t
, current
);
1775 hrtimer_start_expires(&t
.timer
, mode
);
1776 if (!hrtimer_active(&t
.timer
))
1782 hrtimer_cancel(&t
.timer
);
1783 destroy_hrtimer_on_stack(&t
.timer
);
1785 __set_current_state(TASK_RUNNING
);
1787 return !t
.task
? 0 : -EINTR
;
1791 * schedule_hrtimeout_range - sleep until timeout
1792 * @expires: timeout value (ktime_t)
1793 * @delta: slack in expires timeout (ktime_t)
1794 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1796 * Make the current task sleep until the given expiry time has
1797 * elapsed. The routine will return immediately unless
1798 * the current task state has been set (see set_current_state()).
1800 * The @delta argument gives the kernel the freedom to schedule the
1801 * actual wakeup to a time that is both power and performance friendly.
1802 * The kernel give the normal best effort behavior for "@expires+@delta",
1803 * but may decide to fire the timer earlier, but no earlier than @expires.
1805 * You can set the task state as follows -
1807 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1808 * pass before the routine returns.
1810 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1811 * delivered to the current task.
1813 * The current task state is guaranteed to be TASK_RUNNING when this
1816 * Returns 0 when the timer has expired otherwise -EINTR
1818 int __sched
schedule_hrtimeout_range(ktime_t
*expires
, unsigned long delta
,
1819 const enum hrtimer_mode mode
)
1821 return schedule_hrtimeout_range_clock(expires
, delta
, mode
,
1824 EXPORT_SYMBOL_GPL(schedule_hrtimeout_range
);
1827 * schedule_hrtimeout - sleep until timeout
1828 * @expires: timeout value (ktime_t)
1829 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1831 * Make the current task sleep until the given expiry time has
1832 * elapsed. The routine will return immediately unless
1833 * the current task state has been set (see set_current_state()).
1835 * You can set the task state as follows -
1837 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1838 * pass before the routine returns.
1840 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1841 * delivered to the current task.
1843 * The current task state is guaranteed to be TASK_RUNNING when this
1846 * Returns 0 when the timer has expired otherwise -EINTR
1848 int __sched
schedule_hrtimeout(ktime_t
*expires
,
1849 const enum hrtimer_mode mode
)
1851 return schedule_hrtimeout_range(expires
, 0, mode
);
1853 EXPORT_SYMBOL_GPL(schedule_hrtimeout
);