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/export.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/sched/sysctl.h>
48 #include <linux/sched/rt.h>
49 #include <linux/timer.h>
51 #include <asm/uaccess.h>
53 #include <trace/events/timer.h>
58 * There are more clockids then hrtimer bases. Thus, we index
59 * into the timer bases by the hrtimer_base_type enum. When trying
60 * to reach a base using a clockid, hrtimer_clockid_to_base()
61 * is used to convert from clockid to the proper hrtimer_base_type.
63 DEFINE_PER_CPU(struct hrtimer_cpu_base
, hrtimer_bases
) =
69 .index
= HRTIMER_BASE_MONOTONIC
,
70 .clockid
= CLOCK_MONOTONIC
,
71 .get_time
= &ktime_get
,
72 .resolution
= KTIME_LOW_RES
,
75 .index
= HRTIMER_BASE_REALTIME
,
76 .clockid
= CLOCK_REALTIME
,
77 .get_time
= &ktime_get_real
,
78 .resolution
= KTIME_LOW_RES
,
81 .index
= HRTIMER_BASE_BOOTTIME
,
82 .clockid
= CLOCK_BOOTTIME
,
83 .get_time
= &ktime_get_boottime
,
84 .resolution
= KTIME_LOW_RES
,
89 static const int hrtimer_clock_to_base_table
[MAX_CLOCKS
] = {
90 [CLOCK_REALTIME
] = HRTIMER_BASE_REALTIME
,
91 [CLOCK_MONOTONIC
] = HRTIMER_BASE_MONOTONIC
,
92 [CLOCK_BOOTTIME
] = HRTIMER_BASE_BOOTTIME
,
95 static inline int hrtimer_clockid_to_base(clockid_t clock_id
)
97 return hrtimer_clock_to_base_table
[clock_id
];
102 * Get the coarse grained time at the softirq based on xtime and
105 static void hrtimer_get_softirq_time(struct hrtimer_cpu_base
*base
)
107 ktime_t xtim
, mono
, boot
;
108 struct timespec xts
, tom
, slp
;
110 get_xtime_and_monotonic_and_sleep_offset(&xts
, &tom
, &slp
);
112 xtim
= timespec_to_ktime(xts
);
113 mono
= ktime_add(xtim
, timespec_to_ktime(tom
));
114 boot
= ktime_add(mono
, timespec_to_ktime(slp
));
115 base
->clock_base
[HRTIMER_BASE_REALTIME
].softirq_time
= xtim
;
116 base
->clock_base
[HRTIMER_BASE_MONOTONIC
].softirq_time
= mono
;
117 base
->clock_base
[HRTIMER_BASE_BOOTTIME
].softirq_time
= boot
;
121 * Functions and macros which are different for UP/SMP systems are kept in a
127 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
128 * means that all timers which are tied to this base via timer->base are
129 * locked, and the base itself is locked too.
131 * So __run_timers/migrate_timers can safely modify all timers which could
132 * be found on the lists/queues.
134 * When the timer's base is locked, and the timer removed from list, it is
135 * possible to set timer->base = NULL and drop the lock: the timer remains
139 struct hrtimer_clock_base
*lock_hrtimer_base(const struct hrtimer
*timer
,
140 unsigned long *flags
)
142 struct hrtimer_clock_base
*base
;
146 if (likely(base
!= NULL
)) {
147 raw_spin_lock_irqsave(&base
->cpu_base
->lock
, *flags
);
148 if (likely(base
== timer
->base
))
150 /* The timer has migrated to another CPU: */
151 raw_spin_unlock_irqrestore(&base
->cpu_base
->lock
, *flags
);
159 * Get the preferred target CPU for NOHZ
161 static int hrtimer_get_target(int this_cpu
, int pinned
)
164 if (!pinned
&& get_sysctl_timer_migration() && idle_cpu(this_cpu
))
165 return get_nohz_timer_target();
171 * With HIGHRES=y we do not migrate the timer when it is expiring
172 * before the next event on the target cpu because we cannot reprogram
173 * the target cpu hardware and we would cause it to fire late.
175 * Called with cpu_base->lock of target cpu held.
178 hrtimer_check_target(struct hrtimer
*timer
, struct hrtimer_clock_base
*new_base
)
180 #ifdef CONFIG_HIGH_RES_TIMERS
183 if (!new_base
->cpu_base
->hres_active
)
186 expires
= ktime_sub(hrtimer_get_expires(timer
), new_base
->offset
);
187 return expires
.tv64
<= new_base
->cpu_base
->expires_next
.tv64
;
194 * Switch the timer base to the current CPU when possible.
196 static inline struct hrtimer_clock_base
*
197 switch_hrtimer_base(struct hrtimer
*timer
, struct hrtimer_clock_base
*base
,
200 struct hrtimer_clock_base
*new_base
;
201 struct hrtimer_cpu_base
*new_cpu_base
;
202 int this_cpu
= smp_processor_id();
203 int cpu
= hrtimer_get_target(this_cpu
, pinned
);
204 int basenum
= base
->index
;
207 new_cpu_base
= &per_cpu(hrtimer_bases
, cpu
);
208 new_base
= &new_cpu_base
->clock_base
[basenum
];
210 if (base
!= new_base
) {
212 * We are trying to move timer to new_base.
213 * However we can't change timer's base while it is running,
214 * so we keep it on the same CPU. No hassle vs. reprogramming
215 * the event source in the high resolution case. The softirq
216 * code will take care of this when the timer function has
217 * completed. There is no conflict as we hold the lock until
218 * the timer is enqueued.
220 if (unlikely(hrtimer_callback_running(timer
)))
223 /* See the comment in lock_timer_base() */
225 raw_spin_unlock(&base
->cpu_base
->lock
);
226 raw_spin_lock(&new_base
->cpu_base
->lock
);
228 if (cpu
!= this_cpu
&& hrtimer_check_target(timer
, new_base
)) {
230 raw_spin_unlock(&new_base
->cpu_base
->lock
);
231 raw_spin_lock(&base
->cpu_base
->lock
);
235 timer
->base
= new_base
;
240 #else /* CONFIG_SMP */
242 static inline struct hrtimer_clock_base
*
243 lock_hrtimer_base(const struct hrtimer
*timer
, unsigned long *flags
)
245 struct hrtimer_clock_base
*base
= timer
->base
;
247 raw_spin_lock_irqsave(&base
->cpu_base
->lock
, *flags
);
252 # define switch_hrtimer_base(t, b, p) (b)
254 #endif /* !CONFIG_SMP */
257 * Functions for the union type storage format of ktime_t which are
258 * too large for inlining:
260 #if BITS_PER_LONG < 64
261 # ifndef CONFIG_KTIME_SCALAR
263 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
265 * @nsec: the scalar nsec value to add
267 * Returns the sum of kt and nsec in ktime_t format
269 ktime_t
ktime_add_ns(const ktime_t kt
, u64 nsec
)
273 if (likely(nsec
< NSEC_PER_SEC
)) {
276 unsigned long rem
= do_div(nsec
, NSEC_PER_SEC
);
278 tmp
= ktime_set((long)nsec
, rem
);
281 return ktime_add(kt
, tmp
);
284 EXPORT_SYMBOL_GPL(ktime_add_ns
);
287 * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable
289 * @nsec: the scalar nsec value to subtract
291 * Returns the subtraction of @nsec from @kt in ktime_t format
293 ktime_t
ktime_sub_ns(const ktime_t kt
, u64 nsec
)
297 if (likely(nsec
< NSEC_PER_SEC
)) {
300 unsigned long rem
= do_div(nsec
, NSEC_PER_SEC
);
302 tmp
= ktime_set((long)nsec
, rem
);
305 return ktime_sub(kt
, tmp
);
308 EXPORT_SYMBOL_GPL(ktime_sub_ns
);
309 # endif /* !CONFIG_KTIME_SCALAR */
312 * Divide a ktime value by a nanosecond value
314 u64
ktime_divns(const ktime_t kt
, s64 div
)
319 dclc
= ktime_to_ns(kt
);
320 /* Make sure the divisor is less than 2^32: */
326 do_div(dclc
, (unsigned long) div
);
330 #endif /* BITS_PER_LONG >= 64 */
333 * Add two ktime values and do a safety check for overflow:
335 ktime_t
ktime_add_safe(const ktime_t lhs
, const ktime_t rhs
)
337 ktime_t res
= ktime_add(lhs
, rhs
);
340 * We use KTIME_SEC_MAX here, the maximum timeout which we can
341 * return to user space in a timespec:
343 if (res
.tv64
< 0 || res
.tv64
< lhs
.tv64
|| res
.tv64
< rhs
.tv64
)
344 res
= ktime_set(KTIME_SEC_MAX
, 0);
349 EXPORT_SYMBOL_GPL(ktime_add_safe
);
351 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
353 static struct debug_obj_descr hrtimer_debug_descr
;
355 static void *hrtimer_debug_hint(void *addr
)
357 return ((struct hrtimer
*) addr
)->function
;
361 * fixup_init is called when:
362 * - an active object is initialized
364 static int hrtimer_fixup_init(void *addr
, enum debug_obj_state state
)
366 struct hrtimer
*timer
= addr
;
369 case ODEBUG_STATE_ACTIVE
:
370 hrtimer_cancel(timer
);
371 debug_object_init(timer
, &hrtimer_debug_descr
);
379 * fixup_activate is called when:
380 * - an active object is activated
381 * - an unknown object is activated (might be a statically initialized object)
383 static int hrtimer_fixup_activate(void *addr
, enum debug_obj_state state
)
387 case ODEBUG_STATE_NOTAVAILABLE
:
391 case ODEBUG_STATE_ACTIVE
:
400 * fixup_free is called when:
401 * - an active object is freed
403 static int hrtimer_fixup_free(void *addr
, enum debug_obj_state state
)
405 struct hrtimer
*timer
= addr
;
408 case ODEBUG_STATE_ACTIVE
:
409 hrtimer_cancel(timer
);
410 debug_object_free(timer
, &hrtimer_debug_descr
);
417 static struct debug_obj_descr hrtimer_debug_descr
= {
419 .debug_hint
= hrtimer_debug_hint
,
420 .fixup_init
= hrtimer_fixup_init
,
421 .fixup_activate
= hrtimer_fixup_activate
,
422 .fixup_free
= hrtimer_fixup_free
,
425 static inline void debug_hrtimer_init(struct hrtimer
*timer
)
427 debug_object_init(timer
, &hrtimer_debug_descr
);
430 static inline void debug_hrtimer_activate(struct hrtimer
*timer
)
432 debug_object_activate(timer
, &hrtimer_debug_descr
);
435 static inline void debug_hrtimer_deactivate(struct hrtimer
*timer
)
437 debug_object_deactivate(timer
, &hrtimer_debug_descr
);
440 static inline void debug_hrtimer_free(struct hrtimer
*timer
)
442 debug_object_free(timer
, &hrtimer_debug_descr
);
445 static void __hrtimer_init(struct hrtimer
*timer
, clockid_t clock_id
,
446 enum hrtimer_mode mode
);
448 void hrtimer_init_on_stack(struct hrtimer
*timer
, clockid_t clock_id
,
449 enum hrtimer_mode mode
)
451 debug_object_init_on_stack(timer
, &hrtimer_debug_descr
);
452 __hrtimer_init(timer
, clock_id
, mode
);
454 EXPORT_SYMBOL_GPL(hrtimer_init_on_stack
);
456 void destroy_hrtimer_on_stack(struct hrtimer
*timer
)
458 debug_object_free(timer
, &hrtimer_debug_descr
);
462 static inline void debug_hrtimer_init(struct hrtimer
*timer
) { }
463 static inline void debug_hrtimer_activate(struct hrtimer
*timer
) { }
464 static inline void debug_hrtimer_deactivate(struct hrtimer
*timer
) { }
468 debug_init(struct hrtimer
*timer
, clockid_t clockid
,
469 enum hrtimer_mode mode
)
471 debug_hrtimer_init(timer
);
472 trace_hrtimer_init(timer
, clockid
, mode
);
475 static inline void debug_activate(struct hrtimer
*timer
)
477 debug_hrtimer_activate(timer
);
478 trace_hrtimer_start(timer
);
481 static inline void debug_deactivate(struct hrtimer
*timer
)
483 debug_hrtimer_deactivate(timer
);
484 trace_hrtimer_cancel(timer
);
487 /* High resolution timer related functions */
488 #ifdef CONFIG_HIGH_RES_TIMERS
491 * High resolution timer enabled ?
493 static int hrtimer_hres_enabled __read_mostly
= 1;
496 * Enable / Disable high resolution mode
498 static int __init
setup_hrtimer_hres(char *str
)
500 if (!strcmp(str
, "off"))
501 hrtimer_hres_enabled
= 0;
502 else if (!strcmp(str
, "on"))
503 hrtimer_hres_enabled
= 1;
509 __setup("highres=", setup_hrtimer_hres
);
512 * hrtimer_high_res_enabled - query, if the highres mode is enabled
514 static inline int hrtimer_is_hres_enabled(void)
516 return hrtimer_hres_enabled
;
520 * Is the high resolution mode active ?
522 static inline int hrtimer_hres_active(void)
524 return __this_cpu_read(hrtimer_bases
.hres_active
);
528 * Reprogram the event source with checking both queues for the
530 * Called with interrupts disabled and base->lock held
533 hrtimer_force_reprogram(struct hrtimer_cpu_base
*cpu_base
, int skip_equal
)
536 struct hrtimer_clock_base
*base
= cpu_base
->clock_base
;
537 ktime_t expires
, expires_next
;
539 expires_next
.tv64
= KTIME_MAX
;
541 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++, base
++) {
542 struct hrtimer
*timer
;
543 struct timerqueue_node
*next
;
545 next
= timerqueue_getnext(&base
->active
);
548 timer
= container_of(next
, struct hrtimer
, node
);
550 expires
= ktime_sub(hrtimer_get_expires(timer
), base
->offset
);
552 * clock_was_set() has changed base->offset so the
553 * result might be negative. Fix it up to prevent a
554 * false positive in clockevents_program_event()
556 if (expires
.tv64
< 0)
558 if (expires
.tv64
< expires_next
.tv64
)
559 expires_next
= expires
;
562 if (skip_equal
&& expires_next
.tv64
== cpu_base
->expires_next
.tv64
)
565 cpu_base
->expires_next
.tv64
= expires_next
.tv64
;
567 if (cpu_base
->expires_next
.tv64
!= KTIME_MAX
)
568 tick_program_event(cpu_base
->expires_next
, 1);
572 * Shared reprogramming for clock_realtime and clock_monotonic
574 * When a timer is enqueued and expires earlier than the already enqueued
575 * timers, we have to check, whether it expires earlier than the timer for
576 * which the clock event device was armed.
578 * Called with interrupts disabled and base->cpu_base.lock held
580 static int hrtimer_reprogram(struct hrtimer
*timer
,
581 struct hrtimer_clock_base
*base
)
583 struct hrtimer_cpu_base
*cpu_base
= &__get_cpu_var(hrtimer_bases
);
584 ktime_t expires
= ktime_sub(hrtimer_get_expires(timer
), base
->offset
);
587 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer
) < 0);
590 * When the callback is running, we do not reprogram the clock event
591 * device. The timer callback is either running on a different CPU or
592 * the callback is executed in the hrtimer_interrupt context. The
593 * reprogramming is handled either by the softirq, which called the
594 * callback or at the end of the hrtimer_interrupt.
596 if (hrtimer_callback_running(timer
))
600 * CLOCK_REALTIME timer might be requested with an absolute
601 * expiry time which is less than base->offset. Nothing wrong
602 * about that, just avoid to call into the tick code, which
603 * has now objections against negative expiry values.
605 if (expires
.tv64
< 0)
608 if (expires
.tv64
>= cpu_base
->expires_next
.tv64
)
612 * If a hang was detected in the last timer interrupt then we
613 * do not schedule a timer which is earlier than the expiry
614 * which we enforced in the hang detection. We want the system
617 if (cpu_base
->hang_detected
)
621 * Clockevents returns -ETIME, when the event was in the past.
623 res
= tick_program_event(expires
, 0);
624 if (!IS_ERR_VALUE(res
))
625 cpu_base
->expires_next
= expires
;
630 * Initialize the high resolution related parts of cpu_base
632 static inline void hrtimer_init_hres(struct hrtimer_cpu_base
*base
)
634 base
->expires_next
.tv64
= KTIME_MAX
;
635 base
->hres_active
= 0;
639 * When High resolution timers are active, try to reprogram. Note, that in case
640 * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry
641 * check happens. The timer gets enqueued into the rbtree. The reprogramming
642 * and expiry check is done in the hrtimer_interrupt or in the softirq.
644 static inline int hrtimer_enqueue_reprogram(struct hrtimer
*timer
,
645 struct hrtimer_clock_base
*base
)
647 return base
->cpu_base
->hres_active
&& hrtimer_reprogram(timer
, base
);
650 static inline ktime_t
hrtimer_update_base(struct hrtimer_cpu_base
*base
)
652 ktime_t
*offs_real
= &base
->clock_base
[HRTIMER_BASE_REALTIME
].offset
;
653 ktime_t
*offs_boot
= &base
->clock_base
[HRTIMER_BASE_BOOTTIME
].offset
;
655 return ktime_get_update_offsets(offs_real
, offs_boot
);
659 * Retrigger next event is called after clock was set
661 * Called with interrupts disabled via on_each_cpu()
663 static void retrigger_next_event(void *arg
)
665 struct hrtimer_cpu_base
*base
= &__get_cpu_var(hrtimer_bases
);
667 if (!hrtimer_hres_active())
670 raw_spin_lock(&base
->lock
);
671 hrtimer_update_base(base
);
672 hrtimer_force_reprogram(base
, 0);
673 raw_spin_unlock(&base
->lock
);
677 * Switch to high resolution mode
679 static int hrtimer_switch_to_hres(void)
681 int i
, cpu
= smp_processor_id();
682 struct hrtimer_cpu_base
*base
= &per_cpu(hrtimer_bases
, cpu
);
685 if (base
->hres_active
)
688 local_irq_save(flags
);
690 if (tick_init_highres()) {
691 local_irq_restore(flags
);
692 printk(KERN_WARNING
"Could not switch to high resolution "
693 "mode on CPU %d\n", cpu
);
696 base
->hres_active
= 1;
697 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++)
698 base
->clock_base
[i
].resolution
= KTIME_HIGH_RES
;
700 tick_setup_sched_timer();
701 /* "Retrigger" the interrupt to get things going */
702 retrigger_next_event(NULL
);
703 local_irq_restore(flags
);
708 * Called from timekeeping code to reprogramm the hrtimer interrupt
709 * device. If called from the timer interrupt context we defer it to
712 void clock_was_set_delayed(void)
714 struct hrtimer_cpu_base
*cpu_base
= &__get_cpu_var(hrtimer_bases
);
716 cpu_base
->clock_was_set
= 1;
717 __raise_softirq_irqoff(HRTIMER_SOFTIRQ
);
722 static inline int hrtimer_hres_active(void) { return 0; }
723 static inline int hrtimer_is_hres_enabled(void) { return 0; }
724 static inline int hrtimer_switch_to_hres(void) { return 0; }
726 hrtimer_force_reprogram(struct hrtimer_cpu_base
*base
, int skip_equal
) { }
727 static inline int hrtimer_enqueue_reprogram(struct hrtimer
*timer
,
728 struct hrtimer_clock_base
*base
)
732 static inline void hrtimer_init_hres(struct hrtimer_cpu_base
*base
) { }
733 static inline void retrigger_next_event(void *arg
) { }
735 #endif /* CONFIG_HIGH_RES_TIMERS */
738 * Clock realtime was set
740 * Change the offset of the realtime clock vs. the monotonic
743 * We might have to reprogram the high resolution timer interrupt. On
744 * SMP we call the architecture specific code to retrigger _all_ high
745 * resolution timer interrupts. On UP we just disable interrupts and
746 * call the high resolution interrupt code.
748 void clock_was_set(void)
750 #ifdef CONFIG_HIGH_RES_TIMERS
751 /* Retrigger the CPU local events everywhere */
752 on_each_cpu(retrigger_next_event
, NULL
, 1);
754 timerfd_clock_was_set();
758 * During resume we might have to reprogram the high resolution timer
759 * interrupt (on the local CPU):
761 void hrtimers_resume(void)
763 WARN_ONCE(!irqs_disabled(),
764 KERN_INFO
"hrtimers_resume() called with IRQs enabled!");
766 retrigger_next_event(NULL
);
767 timerfd_clock_was_set();
770 static inline void timer_stats_hrtimer_set_start_info(struct hrtimer
*timer
)
772 #ifdef CONFIG_TIMER_STATS
773 if (timer
->start_site
)
775 timer
->start_site
= __builtin_return_address(0);
776 memcpy(timer
->start_comm
, current
->comm
, TASK_COMM_LEN
);
777 timer
->start_pid
= current
->pid
;
781 static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer
*timer
)
783 #ifdef CONFIG_TIMER_STATS
784 timer
->start_site
= NULL
;
788 static inline void timer_stats_account_hrtimer(struct hrtimer
*timer
)
790 #ifdef CONFIG_TIMER_STATS
791 if (likely(!timer_stats_active
))
793 timer_stats_update_stats(timer
, timer
->start_pid
, timer
->start_site
,
794 timer
->function
, timer
->start_comm
, 0);
799 * Counterpart to lock_hrtimer_base above:
802 void unlock_hrtimer_base(const struct hrtimer
*timer
, unsigned long *flags
)
804 raw_spin_unlock_irqrestore(&timer
->base
->cpu_base
->lock
, *flags
);
808 * hrtimer_forward - forward the timer expiry
809 * @timer: hrtimer to forward
810 * @now: forward past this time
811 * @interval: the interval to forward
813 * Forward the timer expiry so it will expire in the future.
814 * Returns the number of overruns.
816 u64
hrtimer_forward(struct hrtimer
*timer
, ktime_t now
, ktime_t interval
)
821 delta
= ktime_sub(now
, hrtimer_get_expires(timer
));
826 if (interval
.tv64
< timer
->base
->resolution
.tv64
)
827 interval
.tv64
= timer
->base
->resolution
.tv64
;
829 if (unlikely(delta
.tv64
>= interval
.tv64
)) {
830 s64 incr
= ktime_to_ns(interval
);
832 orun
= ktime_divns(delta
, incr
);
833 hrtimer_add_expires_ns(timer
, incr
* orun
);
834 if (hrtimer_get_expires_tv64(timer
) > now
.tv64
)
837 * This (and the ktime_add() below) is the
838 * correction for exact:
842 hrtimer_add_expires(timer
, interval
);
846 EXPORT_SYMBOL_GPL(hrtimer_forward
);
849 * enqueue_hrtimer - internal function to (re)start a timer
851 * The timer is inserted in expiry order. Insertion into the
852 * red black tree is O(log(n)). Must hold the base lock.
854 * Returns 1 when the new timer is the leftmost timer in the tree.
856 static int enqueue_hrtimer(struct hrtimer
*timer
,
857 struct hrtimer_clock_base
*base
)
859 debug_activate(timer
);
861 timerqueue_add(&base
->active
, &timer
->node
);
862 base
->cpu_base
->active_bases
|= 1 << base
->index
;
865 * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
866 * state of a possibly running callback.
868 timer
->state
|= HRTIMER_STATE_ENQUEUED
;
870 return (&timer
->node
== base
->active
.next
);
874 * __remove_hrtimer - internal function to remove a timer
876 * Caller must hold the base lock.
878 * High resolution timer mode reprograms the clock event device when the
879 * timer is the one which expires next. The caller can disable this by setting
880 * reprogram to zero. This is useful, when the context does a reprogramming
881 * anyway (e.g. timer interrupt)
883 static void __remove_hrtimer(struct hrtimer
*timer
,
884 struct hrtimer_clock_base
*base
,
885 unsigned long newstate
, int reprogram
)
887 struct timerqueue_node
*next_timer
;
888 if (!(timer
->state
& HRTIMER_STATE_ENQUEUED
))
891 next_timer
= timerqueue_getnext(&base
->active
);
892 timerqueue_del(&base
->active
, &timer
->node
);
893 if (&timer
->node
== next_timer
) {
894 #ifdef CONFIG_HIGH_RES_TIMERS
895 /* Reprogram the clock event device. if enabled */
896 if (reprogram
&& hrtimer_hres_active()) {
899 expires
= ktime_sub(hrtimer_get_expires(timer
),
901 if (base
->cpu_base
->expires_next
.tv64
== expires
.tv64
)
902 hrtimer_force_reprogram(base
->cpu_base
, 1);
906 if (!timerqueue_getnext(&base
->active
))
907 base
->cpu_base
->active_bases
&= ~(1 << base
->index
);
909 timer
->state
= newstate
;
913 * remove hrtimer, called with base lock held
916 remove_hrtimer(struct hrtimer
*timer
, struct hrtimer_clock_base
*base
)
918 if (hrtimer_is_queued(timer
)) {
923 * Remove the timer and force reprogramming when high
924 * resolution mode is active and the timer is on the current
925 * CPU. If we remove a timer on another CPU, reprogramming is
926 * skipped. The interrupt event on this CPU is fired and
927 * reprogramming happens in the interrupt handler. This is a
928 * rare case and less expensive than a smp call.
930 debug_deactivate(timer
);
931 timer_stats_hrtimer_clear_start_info(timer
);
932 reprogram
= base
->cpu_base
== &__get_cpu_var(hrtimer_bases
);
934 * We must preserve the CALLBACK state flag here,
935 * otherwise we could move the timer base in
936 * switch_hrtimer_base.
938 state
= timer
->state
& HRTIMER_STATE_CALLBACK
;
939 __remove_hrtimer(timer
, base
, state
, reprogram
);
945 int __hrtimer_start_range_ns(struct hrtimer
*timer
, ktime_t tim
,
946 unsigned long delta_ns
, const enum hrtimer_mode mode
,
949 struct hrtimer_clock_base
*base
, *new_base
;
953 base
= lock_hrtimer_base(timer
, &flags
);
955 /* Remove an active timer from the queue: */
956 ret
= remove_hrtimer(timer
, base
);
958 /* Switch the timer base, if necessary: */
959 new_base
= switch_hrtimer_base(timer
, base
, mode
& HRTIMER_MODE_PINNED
);
961 if (mode
& HRTIMER_MODE_REL
) {
962 tim
= ktime_add_safe(tim
, new_base
->get_time());
964 * CONFIG_TIME_LOW_RES is a temporary way for architectures
965 * to signal that they simply return xtime in
966 * do_gettimeoffset(). In this case we want to round up by
967 * resolution when starting a relative timer, to avoid short
968 * timeouts. This will go away with the GTOD framework.
970 #ifdef CONFIG_TIME_LOW_RES
971 tim
= ktime_add_safe(tim
, base
->resolution
);
975 hrtimer_set_expires_range_ns(timer
, tim
, delta_ns
);
977 timer_stats_hrtimer_set_start_info(timer
);
979 leftmost
= enqueue_hrtimer(timer
, new_base
);
982 * Only allow reprogramming if the new base is on this CPU.
983 * (it might still be on another CPU if the timer was pending)
985 * XXX send_remote_softirq() ?
987 if (leftmost
&& new_base
->cpu_base
== &__get_cpu_var(hrtimer_bases
)
988 && hrtimer_enqueue_reprogram(timer
, new_base
)) {
991 * We need to drop cpu_base->lock to avoid a
992 * lock ordering issue vs. rq->lock.
994 raw_spin_unlock(&new_base
->cpu_base
->lock
);
995 raise_softirq_irqoff(HRTIMER_SOFTIRQ
);
996 local_irq_restore(flags
);
999 __raise_softirq_irqoff(HRTIMER_SOFTIRQ
);
1003 unlock_hrtimer_base(timer
, &flags
);
1009 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
1010 * @timer: the timer to be added
1012 * @delta_ns: "slack" range for the timer
1013 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
1014 * relative (HRTIMER_MODE_REL)
1018 * 1 when the timer was active
1020 int hrtimer_start_range_ns(struct hrtimer
*timer
, ktime_t tim
,
1021 unsigned long delta_ns
, const enum hrtimer_mode mode
)
1023 return __hrtimer_start_range_ns(timer
, tim
, delta_ns
, mode
, 1);
1025 EXPORT_SYMBOL_GPL(hrtimer_start_range_ns
);
1028 * hrtimer_start - (re)start an hrtimer on the current CPU
1029 * @timer: the timer to be added
1031 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
1032 * relative (HRTIMER_MODE_REL)
1036 * 1 when the timer was active
1039 hrtimer_start(struct hrtimer
*timer
, ktime_t tim
, const enum hrtimer_mode mode
)
1041 return __hrtimer_start_range_ns(timer
, tim
, 0, mode
, 1);
1043 EXPORT_SYMBOL_GPL(hrtimer_start
);
1047 * hrtimer_try_to_cancel - try to deactivate a timer
1048 * @timer: hrtimer to stop
1051 * 0 when the timer was not active
1052 * 1 when the timer was active
1053 * -1 when the timer is currently excuting the callback function and
1056 int hrtimer_try_to_cancel(struct hrtimer
*timer
)
1058 struct hrtimer_clock_base
*base
;
1059 unsigned long flags
;
1062 base
= lock_hrtimer_base(timer
, &flags
);
1064 if (!hrtimer_callback_running(timer
))
1065 ret
= remove_hrtimer(timer
, base
);
1067 unlock_hrtimer_base(timer
, &flags
);
1072 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel
);
1075 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1076 * @timer: the timer to be cancelled
1079 * 0 when the timer was not active
1080 * 1 when the timer was active
1082 int hrtimer_cancel(struct hrtimer
*timer
)
1085 int ret
= hrtimer_try_to_cancel(timer
);
1092 EXPORT_SYMBOL_GPL(hrtimer_cancel
);
1095 * hrtimer_get_remaining - get remaining time for the timer
1096 * @timer: the timer to read
1098 ktime_t
hrtimer_get_remaining(const struct hrtimer
*timer
)
1100 unsigned long flags
;
1103 lock_hrtimer_base(timer
, &flags
);
1104 rem
= hrtimer_expires_remaining(timer
);
1105 unlock_hrtimer_base(timer
, &flags
);
1109 EXPORT_SYMBOL_GPL(hrtimer_get_remaining
);
1113 * hrtimer_get_next_event - get the time until next expiry event
1115 * Returns the delta to the next expiry event or KTIME_MAX if no timer
1118 ktime_t
hrtimer_get_next_event(void)
1120 struct hrtimer_cpu_base
*cpu_base
= &__get_cpu_var(hrtimer_bases
);
1121 struct hrtimer_clock_base
*base
= cpu_base
->clock_base
;
1122 ktime_t delta
, mindelta
= { .tv64
= KTIME_MAX
};
1123 unsigned long flags
;
1126 raw_spin_lock_irqsave(&cpu_base
->lock
, flags
);
1128 if (!hrtimer_hres_active()) {
1129 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++, base
++) {
1130 struct hrtimer
*timer
;
1131 struct timerqueue_node
*next
;
1133 next
= timerqueue_getnext(&base
->active
);
1137 timer
= container_of(next
, struct hrtimer
, node
);
1138 delta
.tv64
= hrtimer_get_expires_tv64(timer
);
1139 delta
= ktime_sub(delta
, base
->get_time());
1140 if (delta
.tv64
< mindelta
.tv64
)
1141 mindelta
.tv64
= delta
.tv64
;
1145 raw_spin_unlock_irqrestore(&cpu_base
->lock
, flags
);
1147 if (mindelta
.tv64
< 0)
1153 static void __hrtimer_init(struct hrtimer
*timer
, clockid_t clock_id
,
1154 enum hrtimer_mode mode
)
1156 struct hrtimer_cpu_base
*cpu_base
;
1159 memset(timer
, 0, sizeof(struct hrtimer
));
1161 cpu_base
= &__raw_get_cpu_var(hrtimer_bases
);
1163 if (clock_id
== CLOCK_REALTIME
&& mode
!= HRTIMER_MODE_ABS
)
1164 clock_id
= CLOCK_MONOTONIC
;
1166 base
= hrtimer_clockid_to_base(clock_id
);
1167 timer
->base
= &cpu_base
->clock_base
[base
];
1168 timerqueue_init(&timer
->node
);
1170 #ifdef CONFIG_TIMER_STATS
1171 timer
->start_site
= NULL
;
1172 timer
->start_pid
= -1;
1173 memset(timer
->start_comm
, 0, TASK_COMM_LEN
);
1178 * hrtimer_init - initialize a timer to the given clock
1179 * @timer: the timer to be initialized
1180 * @clock_id: the clock to be used
1181 * @mode: timer mode abs/rel
1183 void hrtimer_init(struct hrtimer
*timer
, clockid_t clock_id
,
1184 enum hrtimer_mode mode
)
1186 debug_init(timer
, clock_id
, mode
);
1187 __hrtimer_init(timer
, clock_id
, mode
);
1189 EXPORT_SYMBOL_GPL(hrtimer_init
);
1192 * hrtimer_get_res - get the timer resolution for a clock
1193 * @which_clock: which clock to query
1194 * @tp: pointer to timespec variable to store the resolution
1196 * Store the resolution of the clock selected by @which_clock in the
1197 * variable pointed to by @tp.
1199 int hrtimer_get_res(const clockid_t which_clock
, struct timespec
*tp
)
1201 struct hrtimer_cpu_base
*cpu_base
;
1202 int base
= hrtimer_clockid_to_base(which_clock
);
1204 cpu_base
= &__raw_get_cpu_var(hrtimer_bases
);
1205 *tp
= ktime_to_timespec(cpu_base
->clock_base
[base
].resolution
);
1209 EXPORT_SYMBOL_GPL(hrtimer_get_res
);
1211 static void __run_hrtimer(struct hrtimer
*timer
, ktime_t
*now
)
1213 struct hrtimer_clock_base
*base
= timer
->base
;
1214 struct hrtimer_cpu_base
*cpu_base
= base
->cpu_base
;
1215 enum hrtimer_restart (*fn
)(struct hrtimer
*);
1218 WARN_ON(!irqs_disabled());
1220 debug_deactivate(timer
);
1221 __remove_hrtimer(timer
, base
, HRTIMER_STATE_CALLBACK
, 0);
1222 timer_stats_account_hrtimer(timer
);
1223 fn
= timer
->function
;
1226 * Because we run timers from hardirq context, there is no chance
1227 * they get migrated to another cpu, therefore its safe to unlock
1230 raw_spin_unlock(&cpu_base
->lock
);
1231 trace_hrtimer_expire_entry(timer
, now
);
1232 restart
= fn(timer
);
1233 trace_hrtimer_expire_exit(timer
);
1234 raw_spin_lock(&cpu_base
->lock
);
1237 * Note: We clear the CALLBACK bit after enqueue_hrtimer and
1238 * we do not reprogramm the event hardware. Happens either in
1239 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1241 if (restart
!= HRTIMER_NORESTART
) {
1242 BUG_ON(timer
->state
!= HRTIMER_STATE_CALLBACK
);
1243 enqueue_hrtimer(timer
, base
);
1246 WARN_ON_ONCE(!(timer
->state
& HRTIMER_STATE_CALLBACK
));
1248 timer
->state
&= ~HRTIMER_STATE_CALLBACK
;
1251 #ifdef CONFIG_HIGH_RES_TIMERS
1254 * High resolution timer interrupt
1255 * Called with interrupts disabled
1257 void hrtimer_interrupt(struct clock_event_device
*dev
)
1259 struct hrtimer_cpu_base
*cpu_base
= &__get_cpu_var(hrtimer_bases
);
1260 ktime_t expires_next
, now
, entry_time
, delta
;
1263 BUG_ON(!cpu_base
->hres_active
);
1264 cpu_base
->nr_events
++;
1265 dev
->next_event
.tv64
= KTIME_MAX
;
1267 raw_spin_lock(&cpu_base
->lock
);
1268 entry_time
= now
= hrtimer_update_base(cpu_base
);
1270 expires_next
.tv64
= KTIME_MAX
;
1272 * We set expires_next to KTIME_MAX here with cpu_base->lock
1273 * held to prevent that a timer is enqueued in our queue via
1274 * the migration code. This does not affect enqueueing of
1275 * timers which run their callback and need to be requeued on
1278 cpu_base
->expires_next
.tv64
= KTIME_MAX
;
1280 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++) {
1281 struct hrtimer_clock_base
*base
;
1282 struct timerqueue_node
*node
;
1285 if (!(cpu_base
->active_bases
& (1 << i
)))
1288 base
= cpu_base
->clock_base
+ i
;
1289 basenow
= ktime_add(now
, base
->offset
);
1291 while ((node
= timerqueue_getnext(&base
->active
))) {
1292 struct hrtimer
*timer
;
1294 timer
= container_of(node
, struct hrtimer
, node
);
1297 * The immediate goal for using the softexpires is
1298 * minimizing wakeups, not running timers at the
1299 * earliest interrupt after their soft expiration.
1300 * This allows us to avoid using a Priority Search
1301 * Tree, which can answer a stabbing querry for
1302 * overlapping intervals and instead use the simple
1303 * BST we already have.
1304 * We don't add extra wakeups by delaying timers that
1305 * are right-of a not yet expired timer, because that
1306 * timer will have to trigger a wakeup anyway.
1309 if (basenow
.tv64
< hrtimer_get_softexpires_tv64(timer
)) {
1312 expires
= ktime_sub(hrtimer_get_expires(timer
),
1314 if (expires
.tv64
< expires_next
.tv64
)
1315 expires_next
= expires
;
1319 __run_hrtimer(timer
, &basenow
);
1324 * Store the new expiry value so the migration code can verify
1327 cpu_base
->expires_next
= expires_next
;
1328 raw_spin_unlock(&cpu_base
->lock
);
1330 /* Reprogramming necessary ? */
1331 if (expires_next
.tv64
== KTIME_MAX
||
1332 !tick_program_event(expires_next
, 0)) {
1333 cpu_base
->hang_detected
= 0;
1338 * The next timer was already expired due to:
1340 * - long lasting callbacks
1341 * - being scheduled away when running in a VM
1343 * We need to prevent that we loop forever in the hrtimer
1344 * interrupt routine. We give it 3 attempts to avoid
1345 * overreacting on some spurious event.
1347 * Acquire base lock for updating the offsets and retrieving
1350 raw_spin_lock(&cpu_base
->lock
);
1351 now
= hrtimer_update_base(cpu_base
);
1352 cpu_base
->nr_retries
++;
1356 * Give the system a chance to do something else than looping
1357 * here. We stored the entry time, so we know exactly how long
1358 * we spent here. We schedule the next event this amount of
1361 cpu_base
->nr_hangs
++;
1362 cpu_base
->hang_detected
= 1;
1363 raw_spin_unlock(&cpu_base
->lock
);
1364 delta
= ktime_sub(now
, entry_time
);
1365 if (delta
.tv64
> cpu_base
->max_hang_time
.tv64
)
1366 cpu_base
->max_hang_time
= delta
;
1368 * Limit it to a sensible value as we enforce a longer
1369 * delay. Give the CPU at least 100ms to catch up.
1371 if (delta
.tv64
> 100 * NSEC_PER_MSEC
)
1372 expires_next
= ktime_add_ns(now
, 100 * NSEC_PER_MSEC
);
1374 expires_next
= ktime_add(now
, delta
);
1375 tick_program_event(expires_next
, 1);
1376 printk_once(KERN_WARNING
"hrtimer: interrupt took %llu ns\n",
1377 ktime_to_ns(delta
));
1381 * local version of hrtimer_peek_ahead_timers() called with interrupts
1384 static void __hrtimer_peek_ahead_timers(void)
1386 struct tick_device
*td
;
1388 if (!hrtimer_hres_active())
1391 td
= &__get_cpu_var(tick_cpu_device
);
1392 if (td
&& td
->evtdev
)
1393 hrtimer_interrupt(td
->evtdev
);
1397 * hrtimer_peek_ahead_timers -- run soft-expired timers now
1399 * hrtimer_peek_ahead_timers will peek at the timer queue of
1400 * the current cpu and check if there are any timers for which
1401 * the soft expires time has passed. If any such timers exist,
1402 * they are run immediately and then removed from the timer queue.
1405 void hrtimer_peek_ahead_timers(void)
1407 unsigned long flags
;
1409 local_irq_save(flags
);
1410 __hrtimer_peek_ahead_timers();
1411 local_irq_restore(flags
);
1414 static void run_hrtimer_softirq(struct softirq_action
*h
)
1416 struct hrtimer_cpu_base
*cpu_base
= &__get_cpu_var(hrtimer_bases
);
1418 if (cpu_base
->clock_was_set
) {
1419 cpu_base
->clock_was_set
= 0;
1423 hrtimer_peek_ahead_timers();
1426 #else /* CONFIG_HIGH_RES_TIMERS */
1428 static inline void __hrtimer_peek_ahead_timers(void) { }
1430 #endif /* !CONFIG_HIGH_RES_TIMERS */
1433 * Called from timer softirq every jiffy, expire hrtimers:
1435 * For HRT its the fall back code to run the softirq in the timer
1436 * softirq context in case the hrtimer initialization failed or has
1437 * not been done yet.
1439 void hrtimer_run_pending(void)
1441 if (hrtimer_hres_active())
1445 * This _is_ ugly: We have to check in the softirq context,
1446 * whether we can switch to highres and / or nohz mode. The
1447 * clocksource switch happens in the timer interrupt with
1448 * xtime_lock held. Notification from there only sets the
1449 * check bit in the tick_oneshot code, otherwise we might
1450 * deadlock vs. xtime_lock.
1452 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1453 hrtimer_switch_to_hres();
1457 * Called from hardirq context every jiffy
1459 void hrtimer_run_queues(void)
1461 struct timerqueue_node
*node
;
1462 struct hrtimer_cpu_base
*cpu_base
= &__get_cpu_var(hrtimer_bases
);
1463 struct hrtimer_clock_base
*base
;
1464 int index
, gettime
= 1;
1466 if (hrtimer_hres_active())
1469 for (index
= 0; index
< HRTIMER_MAX_CLOCK_BASES
; index
++) {
1470 base
= &cpu_base
->clock_base
[index
];
1471 if (!timerqueue_getnext(&base
->active
))
1475 hrtimer_get_softirq_time(cpu_base
);
1479 raw_spin_lock(&cpu_base
->lock
);
1481 while ((node
= timerqueue_getnext(&base
->active
))) {
1482 struct hrtimer
*timer
;
1484 timer
= container_of(node
, struct hrtimer
, node
);
1485 if (base
->softirq_time
.tv64
<=
1486 hrtimer_get_expires_tv64(timer
))
1489 __run_hrtimer(timer
, &base
->softirq_time
);
1491 raw_spin_unlock(&cpu_base
->lock
);
1496 * Sleep related functions:
1498 static enum hrtimer_restart
hrtimer_wakeup(struct hrtimer
*timer
)
1500 struct hrtimer_sleeper
*t
=
1501 container_of(timer
, struct hrtimer_sleeper
, timer
);
1502 struct task_struct
*task
= t
->task
;
1506 wake_up_process(task
);
1508 return HRTIMER_NORESTART
;
1511 void hrtimer_init_sleeper(struct hrtimer_sleeper
*sl
, struct task_struct
*task
)
1513 sl
->timer
.function
= hrtimer_wakeup
;
1516 EXPORT_SYMBOL_GPL(hrtimer_init_sleeper
);
1518 static int __sched
do_nanosleep(struct hrtimer_sleeper
*t
, enum hrtimer_mode mode
)
1520 hrtimer_init_sleeper(t
, current
);
1523 set_current_state(TASK_INTERRUPTIBLE
);
1524 hrtimer_start_expires(&t
->timer
, mode
);
1525 if (!hrtimer_active(&t
->timer
))
1528 if (likely(t
->task
))
1531 hrtimer_cancel(&t
->timer
);
1532 mode
= HRTIMER_MODE_ABS
;
1534 } while (t
->task
&& !signal_pending(current
));
1536 __set_current_state(TASK_RUNNING
);
1538 return t
->task
== NULL
;
1541 static int update_rmtp(struct hrtimer
*timer
, struct timespec __user
*rmtp
)
1543 struct timespec rmt
;
1546 rem
= hrtimer_expires_remaining(timer
);
1549 rmt
= ktime_to_timespec(rem
);
1551 if (copy_to_user(rmtp
, &rmt
, sizeof(*rmtp
)))
1557 long __sched
hrtimer_nanosleep_restart(struct restart_block
*restart
)
1559 struct hrtimer_sleeper t
;
1560 struct timespec __user
*rmtp
;
1563 hrtimer_init_on_stack(&t
.timer
, restart
->nanosleep
.clockid
,
1565 hrtimer_set_expires_tv64(&t
.timer
, restart
->nanosleep
.expires
);
1567 if (do_nanosleep(&t
, HRTIMER_MODE_ABS
))
1570 rmtp
= restart
->nanosleep
.rmtp
;
1572 ret
= update_rmtp(&t
.timer
, rmtp
);
1577 /* The other values in restart are already filled in */
1578 ret
= -ERESTART_RESTARTBLOCK
;
1580 destroy_hrtimer_on_stack(&t
.timer
);
1584 long hrtimer_nanosleep(struct timespec
*rqtp
, struct timespec __user
*rmtp
,
1585 const enum hrtimer_mode mode
, const clockid_t clockid
)
1587 struct restart_block
*restart
;
1588 struct hrtimer_sleeper t
;
1590 unsigned long slack
;
1592 slack
= current
->timer_slack_ns
;
1593 if (rt_task(current
))
1596 hrtimer_init_on_stack(&t
.timer
, clockid
, mode
);
1597 hrtimer_set_expires_range_ns(&t
.timer
, timespec_to_ktime(*rqtp
), slack
);
1598 if (do_nanosleep(&t
, mode
))
1601 /* Absolute timers do not update the rmtp value and restart: */
1602 if (mode
== HRTIMER_MODE_ABS
) {
1603 ret
= -ERESTARTNOHAND
;
1608 ret
= update_rmtp(&t
.timer
, rmtp
);
1613 restart
= ¤t_thread_info()->restart_block
;
1614 restart
->fn
= hrtimer_nanosleep_restart
;
1615 restart
->nanosleep
.clockid
= t
.timer
.base
->clockid
;
1616 restart
->nanosleep
.rmtp
= rmtp
;
1617 restart
->nanosleep
.expires
= hrtimer_get_expires_tv64(&t
.timer
);
1619 ret
= -ERESTART_RESTARTBLOCK
;
1621 destroy_hrtimer_on_stack(&t
.timer
);
1625 SYSCALL_DEFINE2(nanosleep
, struct timespec __user
*, rqtp
,
1626 struct timespec __user
*, rmtp
)
1630 if (copy_from_user(&tu
, rqtp
, sizeof(tu
)))
1633 if (!timespec_valid(&tu
))
1636 return hrtimer_nanosleep(&tu
, rmtp
, HRTIMER_MODE_REL
, CLOCK_MONOTONIC
);
1640 * Functions related to boot-time initialization:
1642 static void __cpuinit
init_hrtimers_cpu(int cpu
)
1644 struct hrtimer_cpu_base
*cpu_base
= &per_cpu(hrtimer_bases
, cpu
);
1647 raw_spin_lock_init(&cpu_base
->lock
);
1649 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++) {
1650 cpu_base
->clock_base
[i
].cpu_base
= cpu_base
;
1651 timerqueue_init_head(&cpu_base
->clock_base
[i
].active
);
1654 hrtimer_init_hres(cpu_base
);
1657 #ifdef CONFIG_HOTPLUG_CPU
1659 static void migrate_hrtimer_list(struct hrtimer_clock_base
*old_base
,
1660 struct hrtimer_clock_base
*new_base
)
1662 struct hrtimer
*timer
;
1663 struct timerqueue_node
*node
;
1665 while ((node
= timerqueue_getnext(&old_base
->active
))) {
1666 timer
= container_of(node
, struct hrtimer
, node
);
1667 BUG_ON(hrtimer_callback_running(timer
));
1668 debug_deactivate(timer
);
1671 * Mark it as STATE_MIGRATE not INACTIVE otherwise the
1672 * timer could be seen as !active and just vanish away
1673 * under us on another CPU
1675 __remove_hrtimer(timer
, old_base
, HRTIMER_STATE_MIGRATE
, 0);
1676 timer
->base
= new_base
;
1678 * Enqueue the timers on the new cpu. This does not
1679 * reprogram the event device in case the timer
1680 * expires before the earliest on this CPU, but we run
1681 * hrtimer_interrupt after we migrated everything to
1682 * sort out already expired timers and reprogram the
1685 enqueue_hrtimer(timer
, new_base
);
1687 /* Clear the migration state bit */
1688 timer
->state
&= ~HRTIMER_STATE_MIGRATE
;
1692 static void migrate_hrtimers(int scpu
)
1694 struct hrtimer_cpu_base
*old_base
, *new_base
;
1697 BUG_ON(cpu_online(scpu
));
1698 tick_cancel_sched_timer(scpu
);
1700 local_irq_disable();
1701 old_base
= &per_cpu(hrtimer_bases
, scpu
);
1702 new_base
= &__get_cpu_var(hrtimer_bases
);
1704 * The caller is globally serialized and nobody else
1705 * takes two locks at once, deadlock is not possible.
1707 raw_spin_lock(&new_base
->lock
);
1708 raw_spin_lock_nested(&old_base
->lock
, SINGLE_DEPTH_NESTING
);
1710 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++) {
1711 migrate_hrtimer_list(&old_base
->clock_base
[i
],
1712 &new_base
->clock_base
[i
]);
1715 raw_spin_unlock(&old_base
->lock
);
1716 raw_spin_unlock(&new_base
->lock
);
1718 /* Check, if we got expired work to do */
1719 __hrtimer_peek_ahead_timers();
1723 #endif /* CONFIG_HOTPLUG_CPU */
1725 static int __cpuinit
hrtimer_cpu_notify(struct notifier_block
*self
,
1726 unsigned long action
, void *hcpu
)
1728 int scpu
= (long)hcpu
;
1732 case CPU_UP_PREPARE
:
1733 case CPU_UP_PREPARE_FROZEN
:
1734 init_hrtimers_cpu(scpu
);
1737 #ifdef CONFIG_HOTPLUG_CPU
1739 case CPU_DYING_FROZEN
:
1740 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DYING
, &scpu
);
1743 case CPU_DEAD_FROZEN
:
1745 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD
, &scpu
);
1746 migrate_hrtimers(scpu
);
1758 static struct notifier_block __cpuinitdata hrtimers_nb
= {
1759 .notifier_call
= hrtimer_cpu_notify
,
1762 void __init
hrtimers_init(void)
1764 hrtimer_cpu_notify(&hrtimers_nb
, (unsigned long)CPU_UP_PREPARE
,
1765 (void *)(long)smp_processor_id());
1766 register_cpu_notifier(&hrtimers_nb
);
1767 #ifdef CONFIG_HIGH_RES_TIMERS
1768 open_softirq(HRTIMER_SOFTIRQ
, run_hrtimer_softirq
);
1773 * schedule_hrtimeout_range_clock - sleep until timeout
1774 * @expires: timeout value (ktime_t)
1775 * @delta: slack in expires timeout (ktime_t)
1776 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1777 * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME
1780 schedule_hrtimeout_range_clock(ktime_t
*expires
, unsigned long delta
,
1781 const enum hrtimer_mode mode
, int clock
)
1783 struct hrtimer_sleeper t
;
1786 * Optimize when a zero timeout value is given. It does not
1787 * matter whether this is an absolute or a relative time.
1789 if (expires
&& !expires
->tv64
) {
1790 __set_current_state(TASK_RUNNING
);
1795 * A NULL parameter means "infinite"
1799 __set_current_state(TASK_RUNNING
);
1803 hrtimer_init_on_stack(&t
.timer
, clock
, mode
);
1804 hrtimer_set_expires_range_ns(&t
.timer
, *expires
, delta
);
1806 hrtimer_init_sleeper(&t
, current
);
1808 hrtimer_start_expires(&t
.timer
, mode
);
1809 if (!hrtimer_active(&t
.timer
))
1815 hrtimer_cancel(&t
.timer
);
1816 destroy_hrtimer_on_stack(&t
.timer
);
1818 __set_current_state(TASK_RUNNING
);
1820 return !t
.task
? 0 : -EINTR
;
1824 * schedule_hrtimeout_range - sleep until timeout
1825 * @expires: timeout value (ktime_t)
1826 * @delta: slack in expires timeout (ktime_t)
1827 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1829 * Make the current task sleep until the given expiry time has
1830 * elapsed. The routine will return immediately unless
1831 * the current task state has been set (see set_current_state()).
1833 * The @delta argument gives the kernel the freedom to schedule the
1834 * actual wakeup to a time that is both power and performance friendly.
1835 * The kernel give the normal best effort behavior for "@expires+@delta",
1836 * but may decide to fire the timer earlier, but no earlier than @expires.
1838 * You can set the task state as follows -
1840 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1841 * pass before the routine returns.
1843 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1844 * delivered to the current task.
1846 * The current task state is guaranteed to be TASK_RUNNING when this
1849 * Returns 0 when the timer has expired otherwise -EINTR
1851 int __sched
schedule_hrtimeout_range(ktime_t
*expires
, unsigned long delta
,
1852 const enum hrtimer_mode mode
)
1854 return schedule_hrtimeout_range_clock(expires
, delta
, mode
,
1857 EXPORT_SYMBOL_GPL(schedule_hrtimeout_range
);
1860 * schedule_hrtimeout - sleep until timeout
1861 * @expires: timeout value (ktime_t)
1862 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1864 * Make the current task sleep until the given expiry time has
1865 * elapsed. The routine will return immediately unless
1866 * the current task state has been set (see set_current_state()).
1868 * You can set the task state as follows -
1870 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1871 * pass before the routine returns.
1873 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1874 * delivered to the current task.
1876 * The current task state is guaranteed to be TASK_RUNNING when this
1879 * Returns 0 when the timer has expired otherwise -EINTR
1881 int __sched
schedule_hrtimeout(ktime_t
*expires
,
1882 const enum hrtimer_mode mode
)
1884 return schedule_hrtimeout_range(expires
, 0, mode
);
1886 EXPORT_SYMBOL_GPL(schedule_hrtimeout
);