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/sched/deadline.h>
50 #include <linux/timer.h>
51 #include <linux/freezer.h>
53 #include <asm/uaccess.h>
55 #include <trace/events/timer.h>
57 #include "tick-internal.h"
62 * There are more clockids then hrtimer bases. Thus, we index
63 * into the timer bases by the hrtimer_base_type enum. When trying
64 * to reach a base using a clockid, hrtimer_clockid_to_base()
65 * is used to convert from clockid to the proper hrtimer_base_type.
67 DEFINE_PER_CPU(struct hrtimer_cpu_base
, hrtimer_bases
) =
70 .lock
= __RAW_SPIN_LOCK_UNLOCKED(hrtimer_bases
.lock
),
74 .index
= HRTIMER_BASE_MONOTONIC
,
75 .clockid
= CLOCK_MONOTONIC
,
76 .get_time
= &ktime_get
,
77 .resolution
= KTIME_LOW_RES
,
80 .index
= HRTIMER_BASE_REALTIME
,
81 .clockid
= CLOCK_REALTIME
,
82 .get_time
= &ktime_get_real
,
83 .resolution
= KTIME_LOW_RES
,
86 .index
= HRTIMER_BASE_BOOTTIME
,
87 .clockid
= CLOCK_BOOTTIME
,
88 .get_time
= &ktime_get_boottime
,
89 .resolution
= KTIME_LOW_RES
,
92 .index
= HRTIMER_BASE_TAI
,
94 .get_time
= &ktime_get_clocktai
,
95 .resolution
= KTIME_LOW_RES
,
100 static const int hrtimer_clock_to_base_table
[MAX_CLOCKS
] = {
101 [CLOCK_REALTIME
] = HRTIMER_BASE_REALTIME
,
102 [CLOCK_MONOTONIC
] = HRTIMER_BASE_MONOTONIC
,
103 [CLOCK_BOOTTIME
] = HRTIMER_BASE_BOOTTIME
,
104 [CLOCK_TAI
] = HRTIMER_BASE_TAI
,
107 static inline int hrtimer_clockid_to_base(clockid_t clock_id
)
109 return hrtimer_clock_to_base_table
[clock_id
];
114 * Get the coarse grained time at the softirq based on xtime and
117 static void hrtimer_get_softirq_time(struct hrtimer_cpu_base
*base
)
119 ktime_t xtim
, mono
, boot
, tai
;
120 ktime_t off_real
, off_boot
, off_tai
;
122 mono
= ktime_get_update_offsets_tick(&off_real
, &off_boot
, &off_tai
);
123 boot
= ktime_add(mono
, off_boot
);
124 xtim
= ktime_add(mono
, off_real
);
125 tai
= ktime_add(mono
, off_tai
);
127 base
->clock_base
[HRTIMER_BASE_REALTIME
].softirq_time
= xtim
;
128 base
->clock_base
[HRTIMER_BASE_MONOTONIC
].softirq_time
= mono
;
129 base
->clock_base
[HRTIMER_BASE_BOOTTIME
].softirq_time
= boot
;
130 base
->clock_base
[HRTIMER_BASE_TAI
].softirq_time
= tai
;
134 * Functions and macros which are different for UP/SMP systems are kept in a
140 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
141 * means that all timers which are tied to this base via timer->base are
142 * locked, and the base itself is locked too.
144 * So __run_timers/migrate_timers can safely modify all timers which could
145 * be found on the lists/queues.
147 * When the timer's base is locked, and the timer removed from list, it is
148 * possible to set timer->base = NULL and drop the lock: the timer remains
152 struct hrtimer_clock_base
*lock_hrtimer_base(const struct hrtimer
*timer
,
153 unsigned long *flags
)
155 struct hrtimer_clock_base
*base
;
159 if (likely(base
!= NULL
)) {
160 raw_spin_lock_irqsave(&base
->cpu_base
->lock
, *flags
);
161 if (likely(base
== timer
->base
))
163 /* The timer has migrated to another CPU: */
164 raw_spin_unlock_irqrestore(&base
->cpu_base
->lock
, *flags
);
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
= get_nohz_timer_target(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
;
237 if (cpu
!= this_cpu
&& hrtimer_check_target(timer
, new_base
)) {
245 #else /* CONFIG_SMP */
247 static inline struct hrtimer_clock_base
*
248 lock_hrtimer_base(const struct hrtimer
*timer
, unsigned long *flags
)
250 struct hrtimer_clock_base
*base
= timer
->base
;
252 raw_spin_lock_irqsave(&base
->cpu_base
->lock
, *flags
);
257 # define switch_hrtimer_base(t, b, p) (b)
259 #endif /* !CONFIG_SMP */
262 * Functions for the union type storage format of ktime_t which are
263 * too large for inlining:
265 #if BITS_PER_LONG < 64
267 * Divide a ktime value by a nanosecond value
269 u64
__ktime_divns(const ktime_t kt
, s64 div
)
274 dclc
= ktime_to_ns(kt
);
275 /* Make sure the divisor is less than 2^32: */
281 do_div(dclc
, (unsigned long) div
);
285 EXPORT_SYMBOL_GPL(__ktime_divns
);
286 #endif /* BITS_PER_LONG >= 64 */
289 * Add two ktime values and do a safety check for overflow:
291 ktime_t
ktime_add_safe(const ktime_t lhs
, const ktime_t rhs
)
293 ktime_t res
= ktime_add(lhs
, rhs
);
296 * We use KTIME_SEC_MAX here, the maximum timeout which we can
297 * return to user space in a timespec:
299 if (res
.tv64
< 0 || res
.tv64
< lhs
.tv64
|| res
.tv64
< rhs
.tv64
)
300 res
= ktime_set(KTIME_SEC_MAX
, 0);
305 EXPORT_SYMBOL_GPL(ktime_add_safe
);
307 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
309 static struct debug_obj_descr hrtimer_debug_descr
;
311 static void *hrtimer_debug_hint(void *addr
)
313 return ((struct hrtimer
*) addr
)->function
;
317 * fixup_init is called when:
318 * - an active object is initialized
320 static int hrtimer_fixup_init(void *addr
, enum debug_obj_state state
)
322 struct hrtimer
*timer
= addr
;
325 case ODEBUG_STATE_ACTIVE
:
326 hrtimer_cancel(timer
);
327 debug_object_init(timer
, &hrtimer_debug_descr
);
335 * fixup_activate is called when:
336 * - an active object is activated
337 * - an unknown object is activated (might be a statically initialized object)
339 static int hrtimer_fixup_activate(void *addr
, enum debug_obj_state state
)
343 case ODEBUG_STATE_NOTAVAILABLE
:
347 case ODEBUG_STATE_ACTIVE
:
356 * fixup_free is called when:
357 * - an active object is freed
359 static int hrtimer_fixup_free(void *addr
, enum debug_obj_state state
)
361 struct hrtimer
*timer
= addr
;
364 case ODEBUG_STATE_ACTIVE
:
365 hrtimer_cancel(timer
);
366 debug_object_free(timer
, &hrtimer_debug_descr
);
373 static struct debug_obj_descr hrtimer_debug_descr
= {
375 .debug_hint
= hrtimer_debug_hint
,
376 .fixup_init
= hrtimer_fixup_init
,
377 .fixup_activate
= hrtimer_fixup_activate
,
378 .fixup_free
= hrtimer_fixup_free
,
381 static inline void debug_hrtimer_init(struct hrtimer
*timer
)
383 debug_object_init(timer
, &hrtimer_debug_descr
);
386 static inline void debug_hrtimer_activate(struct hrtimer
*timer
)
388 debug_object_activate(timer
, &hrtimer_debug_descr
);
391 static inline void debug_hrtimer_deactivate(struct hrtimer
*timer
)
393 debug_object_deactivate(timer
, &hrtimer_debug_descr
);
396 static inline void debug_hrtimer_free(struct hrtimer
*timer
)
398 debug_object_free(timer
, &hrtimer_debug_descr
);
401 static void __hrtimer_init(struct hrtimer
*timer
, clockid_t clock_id
,
402 enum hrtimer_mode mode
);
404 void hrtimer_init_on_stack(struct hrtimer
*timer
, clockid_t clock_id
,
405 enum hrtimer_mode mode
)
407 debug_object_init_on_stack(timer
, &hrtimer_debug_descr
);
408 __hrtimer_init(timer
, clock_id
, mode
);
410 EXPORT_SYMBOL_GPL(hrtimer_init_on_stack
);
412 void destroy_hrtimer_on_stack(struct hrtimer
*timer
)
414 debug_object_free(timer
, &hrtimer_debug_descr
);
418 static inline void debug_hrtimer_init(struct hrtimer
*timer
) { }
419 static inline void debug_hrtimer_activate(struct hrtimer
*timer
) { }
420 static inline void debug_hrtimer_deactivate(struct hrtimer
*timer
) { }
424 debug_init(struct hrtimer
*timer
, clockid_t clockid
,
425 enum hrtimer_mode mode
)
427 debug_hrtimer_init(timer
);
428 trace_hrtimer_init(timer
, clockid
, mode
);
431 static inline void debug_activate(struct hrtimer
*timer
)
433 debug_hrtimer_activate(timer
);
434 trace_hrtimer_start(timer
);
437 static inline void debug_deactivate(struct hrtimer
*timer
)
439 debug_hrtimer_deactivate(timer
);
440 trace_hrtimer_cancel(timer
);
443 #if defined(CONFIG_NO_HZ_COMMON) || defined(CONFIG_HIGH_RES_TIMERS)
444 static ktime_t
__hrtimer_get_next_event(struct hrtimer_cpu_base
*cpu_base
)
446 struct hrtimer_clock_base
*base
= cpu_base
->clock_base
;
447 ktime_t expires
, expires_next
= { .tv64
= KTIME_MAX
};
450 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++, base
++) {
451 struct timerqueue_node
*next
;
452 struct hrtimer
*timer
;
454 next
= timerqueue_getnext(&base
->active
);
458 timer
= container_of(next
, struct hrtimer
, node
);
459 expires
= ktime_sub(hrtimer_get_expires(timer
), base
->offset
);
460 if (expires
.tv64
< expires_next
.tv64
)
461 expires_next
= expires
;
464 * clock_was_set() might have changed base->offset of any of
465 * the clock bases so the result might be negative. Fix it up
466 * to prevent a false positive in clockevents_program_event().
468 if (expires_next
.tv64
< 0)
469 expires_next
.tv64
= 0;
474 /* High resolution timer related functions */
475 #ifdef CONFIG_HIGH_RES_TIMERS
478 * High resolution timer enabled ?
480 static int hrtimer_hres_enabled __read_mostly
= 1;
483 * Enable / Disable high resolution mode
485 static int __init
setup_hrtimer_hres(char *str
)
487 if (!strcmp(str
, "off"))
488 hrtimer_hres_enabled
= 0;
489 else if (!strcmp(str
, "on"))
490 hrtimer_hres_enabled
= 1;
496 __setup("highres=", setup_hrtimer_hres
);
499 * hrtimer_high_res_enabled - query, if the highres mode is enabled
501 static inline int hrtimer_is_hres_enabled(void)
503 return hrtimer_hres_enabled
;
507 * Is the high resolution mode active ?
509 static inline int hrtimer_hres_active(void)
511 return __this_cpu_read(hrtimer_bases
.hres_active
);
515 * Reprogram the event source with checking both queues for the
517 * Called with interrupts disabled and base->lock held
520 hrtimer_force_reprogram(struct hrtimer_cpu_base
*cpu_base
, int skip_equal
)
522 ktime_t expires_next
= __hrtimer_get_next_event(cpu_base
);
524 if (skip_equal
&& expires_next
.tv64
== cpu_base
->expires_next
.tv64
)
527 cpu_base
->expires_next
.tv64
= expires_next
.tv64
;
530 * If a hang was detected in the last timer interrupt then we
531 * leave the hang delay active in the hardware. We want the
532 * system to make progress. That also prevents the following
534 * T1 expires 50ms from now
535 * T2 expires 5s from now
537 * T1 is removed, so this code is called and would reprogram
538 * the hardware to 5s from now. Any hrtimer_start after that
539 * will not reprogram the hardware due to hang_detected being
540 * set. So we'd effectivly block all timers until the T2 event
543 if (cpu_base
->hang_detected
)
546 if (cpu_base
->expires_next
.tv64
!= KTIME_MAX
)
547 tick_program_event(cpu_base
->expires_next
, 1);
551 * Shared reprogramming for clock_realtime and clock_monotonic
553 * When a timer is enqueued and expires earlier than the already enqueued
554 * timers, we have to check, whether it expires earlier than the timer for
555 * which the clock event device was armed.
557 * Note, that in case the state has HRTIMER_STATE_CALLBACK set, no reprogramming
558 * and no expiry check happens. The timer gets enqueued into the rbtree. The
559 * reprogramming and expiry check is done in the hrtimer_interrupt or in the
562 * Called with interrupts disabled and base->cpu_base.lock held
564 static int hrtimer_reprogram(struct hrtimer
*timer
,
565 struct hrtimer_clock_base
*base
)
567 struct hrtimer_cpu_base
*cpu_base
= this_cpu_ptr(&hrtimer_bases
);
568 ktime_t expires
= ktime_sub(hrtimer_get_expires(timer
), base
->offset
);
571 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer
) < 0);
574 * When the callback is running, we do not reprogram the clock event
575 * device. The timer callback is either running on a different CPU or
576 * the callback is executed in the hrtimer_interrupt context. The
577 * reprogramming is handled either by the softirq, which called the
578 * callback or at the end of the hrtimer_interrupt.
580 if (hrtimer_callback_running(timer
))
584 * CLOCK_REALTIME timer might be requested with an absolute
585 * expiry time which is less than base->offset. Nothing wrong
586 * about that, just avoid to call into the tick code, which
587 * has now objections against negative expiry values.
589 if (expires
.tv64
< 0)
592 if (expires
.tv64
>= cpu_base
->expires_next
.tv64
)
596 * When the target cpu of the timer is currently executing
597 * hrtimer_interrupt(), then we do not touch the clock event
598 * device. hrtimer_interrupt() will reevaluate all clock bases
599 * before reprogramming the device.
601 if (cpu_base
->in_hrtirq
)
605 * If a hang was detected in the last timer interrupt then we
606 * do not schedule a timer which is earlier than the expiry
607 * which we enforced in the hang detection. We want the system
610 if (cpu_base
->hang_detected
)
614 * Clockevents returns -ETIME, when the event was in the past.
616 res
= tick_program_event(expires
, 0);
617 if (!IS_ERR_VALUE(res
))
618 cpu_base
->expires_next
= expires
;
623 * Initialize the high resolution related parts of cpu_base
625 static inline void hrtimer_init_hres(struct hrtimer_cpu_base
*base
)
627 base
->expires_next
.tv64
= KTIME_MAX
;
628 base
->hres_active
= 0;
631 static inline ktime_t
hrtimer_update_base(struct hrtimer_cpu_base
*base
)
633 ktime_t
*offs_real
= &base
->clock_base
[HRTIMER_BASE_REALTIME
].offset
;
634 ktime_t
*offs_boot
= &base
->clock_base
[HRTIMER_BASE_BOOTTIME
].offset
;
635 ktime_t
*offs_tai
= &base
->clock_base
[HRTIMER_BASE_TAI
].offset
;
637 return ktime_get_update_offsets_now(offs_real
, offs_boot
, offs_tai
);
641 * Retrigger next event is called after clock was set
643 * Called with interrupts disabled via on_each_cpu()
645 static void retrigger_next_event(void *arg
)
647 struct hrtimer_cpu_base
*base
= this_cpu_ptr(&hrtimer_bases
);
649 if (!hrtimer_hres_active())
652 raw_spin_lock(&base
->lock
);
653 hrtimer_update_base(base
);
654 hrtimer_force_reprogram(base
, 0);
655 raw_spin_unlock(&base
->lock
);
659 * Switch to high resolution mode
661 static int hrtimer_switch_to_hres(void)
663 int i
, cpu
= smp_processor_id();
664 struct hrtimer_cpu_base
*base
= &per_cpu(hrtimer_bases
, cpu
);
667 if (base
->hres_active
)
670 local_irq_save(flags
);
672 if (tick_init_highres()) {
673 local_irq_restore(flags
);
674 printk(KERN_WARNING
"Could not switch to high resolution "
675 "mode on CPU %d\n", cpu
);
678 base
->hres_active
= 1;
679 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++)
680 base
->clock_base
[i
].resolution
= KTIME_HIGH_RES
;
682 tick_setup_sched_timer();
683 /* "Retrigger" the interrupt to get things going */
684 retrigger_next_event(NULL
);
685 local_irq_restore(flags
);
689 static void clock_was_set_work(struct work_struct
*work
)
694 static DECLARE_WORK(hrtimer_work
, clock_was_set_work
);
697 * Called from timekeeping and resume code to reprogramm the hrtimer
698 * interrupt device on all cpus.
700 void clock_was_set_delayed(void)
702 schedule_work(&hrtimer_work
);
707 static inline int hrtimer_hres_active(void) { return 0; }
708 static inline int hrtimer_is_hres_enabled(void) { return 0; }
709 static inline int hrtimer_switch_to_hres(void) { return 0; }
711 hrtimer_force_reprogram(struct hrtimer_cpu_base
*base
, int skip_equal
) { }
712 static inline int hrtimer_reprogram(struct hrtimer
*timer
,
713 struct hrtimer_clock_base
*base
)
717 static inline void hrtimer_init_hres(struct hrtimer_cpu_base
*base
) { }
718 static inline void retrigger_next_event(void *arg
) { }
720 #endif /* CONFIG_HIGH_RES_TIMERS */
723 * Clock realtime was set
725 * Change the offset of the realtime clock vs. the monotonic
728 * We might have to reprogram the high resolution timer interrupt. On
729 * SMP we call the architecture specific code to retrigger _all_ high
730 * resolution timer interrupts. On UP we just disable interrupts and
731 * call the high resolution interrupt code.
733 void clock_was_set(void)
735 #ifdef CONFIG_HIGH_RES_TIMERS
736 /* Retrigger the CPU local events everywhere */
737 on_each_cpu(retrigger_next_event
, NULL
, 1);
739 timerfd_clock_was_set();
743 * During resume we might have to reprogram the high resolution timer
744 * interrupt on all online CPUs. However, all other CPUs will be
745 * stopped with IRQs interrupts disabled so the clock_was_set() call
748 void hrtimers_resume(void)
750 WARN_ONCE(!irqs_disabled(),
751 KERN_INFO
"hrtimers_resume() called with IRQs enabled!");
753 /* Retrigger on the local CPU */
754 retrigger_next_event(NULL
);
755 /* And schedule a retrigger for all others */
756 clock_was_set_delayed();
759 static inline void timer_stats_hrtimer_set_start_info(struct hrtimer
*timer
)
761 #ifdef CONFIG_TIMER_STATS
762 if (timer
->start_site
)
764 timer
->start_site
= __builtin_return_address(0);
765 memcpy(timer
->start_comm
, current
->comm
, TASK_COMM_LEN
);
766 timer
->start_pid
= current
->pid
;
770 static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer
*timer
)
772 #ifdef CONFIG_TIMER_STATS
773 timer
->start_site
= NULL
;
777 static inline void timer_stats_account_hrtimer(struct hrtimer
*timer
)
779 #ifdef CONFIG_TIMER_STATS
780 if (likely(!timer_stats_active
))
782 timer_stats_update_stats(timer
, timer
->start_pid
, timer
->start_site
,
783 timer
->function
, timer
->start_comm
, 0);
788 * Counterpart to lock_hrtimer_base above:
791 void unlock_hrtimer_base(const struct hrtimer
*timer
, unsigned long *flags
)
793 raw_spin_unlock_irqrestore(&timer
->base
->cpu_base
->lock
, *flags
);
797 * hrtimer_forward - forward the timer expiry
798 * @timer: hrtimer to forward
799 * @now: forward past this time
800 * @interval: the interval to forward
802 * Forward the timer expiry so it will expire in the future.
803 * Returns the number of overruns.
805 u64
hrtimer_forward(struct hrtimer
*timer
, ktime_t now
, ktime_t interval
)
810 delta
= ktime_sub(now
, hrtimer_get_expires(timer
));
815 if (interval
.tv64
< timer
->base
->resolution
.tv64
)
816 interval
.tv64
= timer
->base
->resolution
.tv64
;
818 if (unlikely(delta
.tv64
>= interval
.tv64
)) {
819 s64 incr
= ktime_to_ns(interval
);
821 orun
= ktime_divns(delta
, incr
);
822 hrtimer_add_expires_ns(timer
, incr
* orun
);
823 if (hrtimer_get_expires_tv64(timer
) > now
.tv64
)
826 * This (and the ktime_add() below) is the
827 * correction for exact:
831 hrtimer_add_expires(timer
, interval
);
835 EXPORT_SYMBOL_GPL(hrtimer_forward
);
838 * enqueue_hrtimer - internal function to (re)start a timer
840 * The timer is inserted in expiry order. Insertion into the
841 * red black tree is O(log(n)). Must hold the base lock.
843 * Returns 1 when the new timer is the leftmost timer in the tree.
845 static int enqueue_hrtimer(struct hrtimer
*timer
,
846 struct hrtimer_clock_base
*base
)
848 debug_activate(timer
);
850 timerqueue_add(&base
->active
, &timer
->node
);
851 base
->cpu_base
->active_bases
|= 1 << base
->index
;
854 * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
855 * state of a possibly running callback.
857 timer
->state
|= HRTIMER_STATE_ENQUEUED
;
859 return (&timer
->node
== base
->active
.next
);
863 * __remove_hrtimer - internal function to remove a timer
865 * Caller must hold the base lock.
867 * High resolution timer mode reprograms the clock event device when the
868 * timer is the one which expires next. The caller can disable this by setting
869 * reprogram to zero. This is useful, when the context does a reprogramming
870 * anyway (e.g. timer interrupt)
872 static void __remove_hrtimer(struct hrtimer
*timer
,
873 struct hrtimer_clock_base
*base
,
874 unsigned long newstate
, int reprogram
)
876 struct timerqueue_node
*next_timer
;
877 if (!(timer
->state
& HRTIMER_STATE_ENQUEUED
))
880 next_timer
= timerqueue_getnext(&base
->active
);
881 timerqueue_del(&base
->active
, &timer
->node
);
882 if (&timer
->node
== next_timer
) {
883 #ifdef CONFIG_HIGH_RES_TIMERS
884 /* Reprogram the clock event device. if enabled */
885 if (reprogram
&& hrtimer_hres_active()) {
888 expires
= ktime_sub(hrtimer_get_expires(timer
),
890 if (base
->cpu_base
->expires_next
.tv64
== expires
.tv64
)
891 hrtimer_force_reprogram(base
->cpu_base
, 1);
895 if (!timerqueue_getnext(&base
->active
))
896 base
->cpu_base
->active_bases
&= ~(1 << base
->index
);
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
== this_cpu_ptr(&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 if (mode
& HRTIMER_MODE_REL
) {
948 tim
= ktime_add_safe(tim
, base
->get_time());
950 * CONFIG_TIME_LOW_RES is a temporary way for architectures
951 * to signal that they simply return xtime in
952 * do_gettimeoffset(). In this case we want to round up by
953 * resolution when starting a relative timer, to avoid short
954 * timeouts. This will go away with the GTOD framework.
956 #ifdef CONFIG_TIME_LOW_RES
957 tim
= ktime_add_safe(tim
, base
->resolution
);
961 hrtimer_set_expires_range_ns(timer
, tim
, delta_ns
);
963 /* Switch the timer base, if necessary: */
964 new_base
= switch_hrtimer_base(timer
, base
, mode
& HRTIMER_MODE_PINNED
);
966 timer_stats_hrtimer_set_start_info(timer
);
968 leftmost
= enqueue_hrtimer(timer
, new_base
);
971 unlock_hrtimer_base(timer
, &flags
);
975 if (!hrtimer_is_hres_active(timer
)) {
977 * Kick to reschedule the next tick to handle the new timer
978 * on dynticks target.
980 wake_up_nohz_cpu(new_base
->cpu_base
->cpu
);
981 } else if (new_base
->cpu_base
== this_cpu_ptr(&hrtimer_bases
) &&
982 hrtimer_reprogram(timer
, new_base
)) {
984 * Only allow reprogramming if the new base is on this CPU.
985 * (it might still be on another CPU if the timer was pending)
987 * XXX send_remote_softirq() ?
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
);
1007 EXPORT_SYMBOL_GPL(__hrtimer_start_range_ns
);
1010 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
1011 * @timer: the timer to be added
1013 * @delta_ns: "slack" range for the timer
1014 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
1015 * relative (HRTIMER_MODE_REL)
1019 * 1 when the timer was active
1021 int hrtimer_start_range_ns(struct hrtimer
*timer
, ktime_t tim
,
1022 unsigned long delta_ns
, const enum hrtimer_mode mode
)
1024 return __hrtimer_start_range_ns(timer
, tim
, delta_ns
, mode
, 1);
1026 EXPORT_SYMBOL_GPL(hrtimer_start_range_ns
);
1029 * hrtimer_start - (re)start an hrtimer on the current CPU
1030 * @timer: the timer to be added
1032 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
1033 * relative (HRTIMER_MODE_REL)
1037 * 1 when the timer was active
1040 hrtimer_start(struct hrtimer
*timer
, ktime_t tim
, const enum hrtimer_mode mode
)
1042 return __hrtimer_start_range_ns(timer
, tim
, 0, mode
, 1);
1044 EXPORT_SYMBOL_GPL(hrtimer_start
);
1048 * hrtimer_try_to_cancel - try to deactivate a timer
1049 * @timer: hrtimer to stop
1052 * 0 when the timer was not active
1053 * 1 when the timer was active
1054 * -1 when the timer is currently excuting the callback function and
1057 int hrtimer_try_to_cancel(struct hrtimer
*timer
)
1059 struct hrtimer_clock_base
*base
;
1060 unsigned long flags
;
1063 base
= lock_hrtimer_base(timer
, &flags
);
1065 if (!hrtimer_callback_running(timer
))
1066 ret
= remove_hrtimer(timer
, base
);
1068 unlock_hrtimer_base(timer
, &flags
);
1073 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel
);
1076 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1077 * @timer: the timer to be cancelled
1080 * 0 when the timer was not active
1081 * 1 when the timer was active
1083 int hrtimer_cancel(struct hrtimer
*timer
)
1086 int ret
= hrtimer_try_to_cancel(timer
);
1093 EXPORT_SYMBOL_GPL(hrtimer_cancel
);
1096 * hrtimer_get_remaining - get remaining time for the timer
1097 * @timer: the timer to read
1099 ktime_t
hrtimer_get_remaining(const struct hrtimer
*timer
)
1101 unsigned long flags
;
1104 lock_hrtimer_base(timer
, &flags
);
1105 rem
= hrtimer_expires_remaining(timer
);
1106 unlock_hrtimer_base(timer
, &flags
);
1110 EXPORT_SYMBOL_GPL(hrtimer_get_remaining
);
1112 #ifdef CONFIG_NO_HZ_COMMON
1114 * hrtimer_get_next_event - get the time until next expiry event
1116 * Returns the delta to the next expiry event or KTIME_MAX if no timer
1119 ktime_t
hrtimer_get_next_event(void)
1121 struct hrtimer_cpu_base
*cpu_base
= this_cpu_ptr(&hrtimer_bases
);
1122 ktime_t mindelta
= { .tv64
= KTIME_MAX
};
1123 unsigned long flags
;
1125 raw_spin_lock_irqsave(&cpu_base
->lock
, flags
);
1127 if (!hrtimer_hres_active())
1128 mindelta
= ktime_sub(__hrtimer_get_next_event(cpu_base
),
1131 raw_spin_unlock_irqrestore(&cpu_base
->lock
, flags
);
1133 if (mindelta
.tv64
< 0)
1139 static void __hrtimer_init(struct hrtimer
*timer
, clockid_t clock_id
,
1140 enum hrtimer_mode mode
)
1142 struct hrtimer_cpu_base
*cpu_base
;
1145 memset(timer
, 0, sizeof(struct hrtimer
));
1147 cpu_base
= raw_cpu_ptr(&hrtimer_bases
);
1149 if (clock_id
== CLOCK_REALTIME
&& mode
!= HRTIMER_MODE_ABS
)
1150 clock_id
= CLOCK_MONOTONIC
;
1152 base
= hrtimer_clockid_to_base(clock_id
);
1153 timer
->base
= &cpu_base
->clock_base
[base
];
1154 timerqueue_init(&timer
->node
);
1156 #ifdef CONFIG_TIMER_STATS
1157 timer
->start_site
= NULL
;
1158 timer
->start_pid
= -1;
1159 memset(timer
->start_comm
, 0, TASK_COMM_LEN
);
1164 * hrtimer_init - initialize a timer to the given clock
1165 * @timer: the timer to be initialized
1166 * @clock_id: the clock to be used
1167 * @mode: timer mode abs/rel
1169 void hrtimer_init(struct hrtimer
*timer
, clockid_t clock_id
,
1170 enum hrtimer_mode mode
)
1172 debug_init(timer
, clock_id
, mode
);
1173 __hrtimer_init(timer
, clock_id
, mode
);
1175 EXPORT_SYMBOL_GPL(hrtimer_init
);
1178 * hrtimer_get_res - get the timer resolution for a clock
1179 * @which_clock: which clock to query
1180 * @tp: pointer to timespec variable to store the resolution
1182 * Store the resolution of the clock selected by @which_clock in the
1183 * variable pointed to by @tp.
1185 int hrtimer_get_res(const clockid_t which_clock
, struct timespec
*tp
)
1187 struct hrtimer_cpu_base
*cpu_base
;
1188 int base
= hrtimer_clockid_to_base(which_clock
);
1190 cpu_base
= raw_cpu_ptr(&hrtimer_bases
);
1191 *tp
= ktime_to_timespec(cpu_base
->clock_base
[base
].resolution
);
1195 EXPORT_SYMBOL_GPL(hrtimer_get_res
);
1197 static void __run_hrtimer(struct hrtimer
*timer
, ktime_t
*now
)
1199 struct hrtimer_clock_base
*base
= timer
->base
;
1200 struct hrtimer_cpu_base
*cpu_base
= base
->cpu_base
;
1201 enum hrtimer_restart (*fn
)(struct hrtimer
*);
1204 WARN_ON(!irqs_disabled());
1206 debug_deactivate(timer
);
1207 __remove_hrtimer(timer
, base
, HRTIMER_STATE_CALLBACK
, 0);
1208 timer_stats_account_hrtimer(timer
);
1209 fn
= timer
->function
;
1212 * Because we run timers from hardirq context, there is no chance
1213 * they get migrated to another cpu, therefore its safe to unlock
1216 raw_spin_unlock(&cpu_base
->lock
);
1217 trace_hrtimer_expire_entry(timer
, now
);
1218 restart
= fn(timer
);
1219 trace_hrtimer_expire_exit(timer
);
1220 raw_spin_lock(&cpu_base
->lock
);
1223 * Note: We clear the CALLBACK bit after enqueue_hrtimer and
1224 * we do not reprogramm the event hardware. Happens either in
1225 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1227 if (restart
!= HRTIMER_NORESTART
) {
1228 BUG_ON(timer
->state
!= HRTIMER_STATE_CALLBACK
);
1229 enqueue_hrtimer(timer
, base
);
1232 WARN_ON_ONCE(!(timer
->state
& HRTIMER_STATE_CALLBACK
));
1234 timer
->state
&= ~HRTIMER_STATE_CALLBACK
;
1237 #ifdef CONFIG_HIGH_RES_TIMERS
1240 * High resolution timer interrupt
1241 * Called with interrupts disabled
1243 void hrtimer_interrupt(struct clock_event_device
*dev
)
1245 struct hrtimer_cpu_base
*cpu_base
= this_cpu_ptr(&hrtimer_bases
);
1246 ktime_t expires_next
, now
, entry_time
, delta
;
1249 BUG_ON(!cpu_base
->hres_active
);
1250 cpu_base
->nr_events
++;
1251 dev
->next_event
.tv64
= KTIME_MAX
;
1253 raw_spin_lock(&cpu_base
->lock
);
1254 entry_time
= now
= hrtimer_update_base(cpu_base
);
1256 cpu_base
->in_hrtirq
= 1;
1258 * We set expires_next to KTIME_MAX here with cpu_base->lock
1259 * held to prevent that a timer is enqueued in our queue via
1260 * the migration code. This does not affect enqueueing of
1261 * timers which run their callback and need to be requeued on
1264 cpu_base
->expires_next
.tv64
= KTIME_MAX
;
1266 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++) {
1267 struct hrtimer_clock_base
*base
;
1268 struct timerqueue_node
*node
;
1271 if (!(cpu_base
->active_bases
& (1 << i
)))
1274 base
= cpu_base
->clock_base
+ i
;
1275 basenow
= ktime_add(now
, base
->offset
);
1277 while ((node
= timerqueue_getnext(&base
->active
))) {
1278 struct hrtimer
*timer
;
1280 timer
= container_of(node
, struct hrtimer
, node
);
1283 * The immediate goal for using the softexpires is
1284 * minimizing wakeups, not running timers at the
1285 * earliest interrupt after their soft expiration.
1286 * This allows us to avoid using a Priority Search
1287 * Tree, which can answer a stabbing querry for
1288 * overlapping intervals and instead use the simple
1289 * BST we already have.
1290 * We don't add extra wakeups by delaying timers that
1291 * are right-of a not yet expired timer, because that
1292 * timer will have to trigger a wakeup anyway.
1294 if (basenow
.tv64
< hrtimer_get_softexpires_tv64(timer
))
1297 __run_hrtimer(timer
, &basenow
);
1300 /* Reevaluate the clock bases for the next expiry */
1301 expires_next
= __hrtimer_get_next_event(cpu_base
);
1303 * Store the new expiry value so the migration code can verify
1306 cpu_base
->expires_next
= expires_next
;
1307 cpu_base
->in_hrtirq
= 0;
1308 raw_spin_unlock(&cpu_base
->lock
);
1310 /* Reprogramming necessary ? */
1311 if (expires_next
.tv64
== KTIME_MAX
||
1312 !tick_program_event(expires_next
, 0)) {
1313 cpu_base
->hang_detected
= 0;
1318 * The next timer was already expired due to:
1320 * - long lasting callbacks
1321 * - being scheduled away when running in a VM
1323 * We need to prevent that we loop forever in the hrtimer
1324 * interrupt routine. We give it 3 attempts to avoid
1325 * overreacting on some spurious event.
1327 * Acquire base lock for updating the offsets and retrieving
1330 raw_spin_lock(&cpu_base
->lock
);
1331 now
= hrtimer_update_base(cpu_base
);
1332 cpu_base
->nr_retries
++;
1336 * Give the system a chance to do something else than looping
1337 * here. We stored the entry time, so we know exactly how long
1338 * we spent here. We schedule the next event this amount of
1341 cpu_base
->nr_hangs
++;
1342 cpu_base
->hang_detected
= 1;
1343 raw_spin_unlock(&cpu_base
->lock
);
1344 delta
= ktime_sub(now
, entry_time
);
1345 if (delta
.tv64
> cpu_base
->max_hang_time
.tv64
)
1346 cpu_base
->max_hang_time
= delta
;
1348 * Limit it to a sensible value as we enforce a longer
1349 * delay. Give the CPU at least 100ms to catch up.
1351 if (delta
.tv64
> 100 * NSEC_PER_MSEC
)
1352 expires_next
= ktime_add_ns(now
, 100 * NSEC_PER_MSEC
);
1354 expires_next
= ktime_add(now
, delta
);
1355 tick_program_event(expires_next
, 1);
1356 printk_once(KERN_WARNING
"hrtimer: interrupt took %llu ns\n",
1357 ktime_to_ns(delta
));
1361 * local version of hrtimer_peek_ahead_timers() called with interrupts
1364 static void __hrtimer_peek_ahead_timers(void)
1366 struct tick_device
*td
;
1368 if (!hrtimer_hres_active())
1371 td
= this_cpu_ptr(&tick_cpu_device
);
1372 if (td
&& td
->evtdev
)
1373 hrtimer_interrupt(td
->evtdev
);
1377 * hrtimer_peek_ahead_timers -- run soft-expired timers now
1379 * hrtimer_peek_ahead_timers will peek at the timer queue of
1380 * the current cpu and check if there are any timers for which
1381 * the soft expires time has passed. If any such timers exist,
1382 * they are run immediately and then removed from the timer queue.
1385 void hrtimer_peek_ahead_timers(void)
1387 unsigned long flags
;
1389 local_irq_save(flags
);
1390 __hrtimer_peek_ahead_timers();
1391 local_irq_restore(flags
);
1394 static void run_hrtimer_softirq(struct softirq_action
*h
)
1396 hrtimer_peek_ahead_timers();
1399 #else /* CONFIG_HIGH_RES_TIMERS */
1401 static inline void __hrtimer_peek_ahead_timers(void) { }
1403 #endif /* !CONFIG_HIGH_RES_TIMERS */
1406 * Called from timer softirq every jiffy, expire hrtimers:
1408 * For HRT its the fall back code to run the softirq in the timer
1409 * softirq context in case the hrtimer initialization failed or has
1410 * not been done yet.
1412 void hrtimer_run_pending(void)
1414 if (hrtimer_hres_active())
1418 * This _is_ ugly: We have to check in the softirq context,
1419 * whether we can switch to highres and / or nohz mode. The
1420 * clocksource switch happens in the timer interrupt with
1421 * xtime_lock held. Notification from there only sets the
1422 * check bit in the tick_oneshot code, otherwise we might
1423 * deadlock vs. xtime_lock.
1425 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1426 hrtimer_switch_to_hres();
1430 * Called from hardirq context every jiffy
1432 void hrtimer_run_queues(void)
1434 struct timerqueue_node
*node
;
1435 struct hrtimer_cpu_base
*cpu_base
= this_cpu_ptr(&hrtimer_bases
);
1436 struct hrtimer_clock_base
*base
;
1437 int index
, gettime
= 1;
1439 if (hrtimer_hres_active())
1442 for (index
= 0; index
< HRTIMER_MAX_CLOCK_BASES
; index
++) {
1443 base
= &cpu_base
->clock_base
[index
];
1444 if (!timerqueue_getnext(&base
->active
))
1448 hrtimer_get_softirq_time(cpu_base
);
1452 raw_spin_lock(&cpu_base
->lock
);
1454 while ((node
= timerqueue_getnext(&base
->active
))) {
1455 struct hrtimer
*timer
;
1457 timer
= container_of(node
, struct hrtimer
, node
);
1458 if (base
->softirq_time
.tv64
<=
1459 hrtimer_get_expires_tv64(timer
))
1462 __run_hrtimer(timer
, &base
->softirq_time
);
1464 raw_spin_unlock(&cpu_base
->lock
);
1469 * Sleep related functions:
1471 static enum hrtimer_restart
hrtimer_wakeup(struct hrtimer
*timer
)
1473 struct hrtimer_sleeper
*t
=
1474 container_of(timer
, struct hrtimer_sleeper
, timer
);
1475 struct task_struct
*task
= t
->task
;
1479 wake_up_process(task
);
1481 return HRTIMER_NORESTART
;
1484 void hrtimer_init_sleeper(struct hrtimer_sleeper
*sl
, struct task_struct
*task
)
1486 sl
->timer
.function
= hrtimer_wakeup
;
1489 EXPORT_SYMBOL_GPL(hrtimer_init_sleeper
);
1491 static int __sched
do_nanosleep(struct hrtimer_sleeper
*t
, enum hrtimer_mode mode
)
1493 hrtimer_init_sleeper(t
, current
);
1496 set_current_state(TASK_INTERRUPTIBLE
);
1497 hrtimer_start_expires(&t
->timer
, mode
);
1498 if (!hrtimer_active(&t
->timer
))
1501 if (likely(t
->task
))
1502 freezable_schedule();
1504 hrtimer_cancel(&t
->timer
);
1505 mode
= HRTIMER_MODE_ABS
;
1507 } while (t
->task
&& !signal_pending(current
));
1509 __set_current_state(TASK_RUNNING
);
1511 return t
->task
== NULL
;
1514 static int update_rmtp(struct hrtimer
*timer
, struct timespec __user
*rmtp
)
1516 struct timespec rmt
;
1519 rem
= hrtimer_expires_remaining(timer
);
1522 rmt
= ktime_to_timespec(rem
);
1524 if (copy_to_user(rmtp
, &rmt
, sizeof(*rmtp
)))
1530 long __sched
hrtimer_nanosleep_restart(struct restart_block
*restart
)
1532 struct hrtimer_sleeper t
;
1533 struct timespec __user
*rmtp
;
1536 hrtimer_init_on_stack(&t
.timer
, restart
->nanosleep
.clockid
,
1538 hrtimer_set_expires_tv64(&t
.timer
, restart
->nanosleep
.expires
);
1540 if (do_nanosleep(&t
, HRTIMER_MODE_ABS
))
1543 rmtp
= restart
->nanosleep
.rmtp
;
1545 ret
= update_rmtp(&t
.timer
, rmtp
);
1550 /* The other values in restart are already filled in */
1551 ret
= -ERESTART_RESTARTBLOCK
;
1553 destroy_hrtimer_on_stack(&t
.timer
);
1557 long hrtimer_nanosleep(struct timespec
*rqtp
, struct timespec __user
*rmtp
,
1558 const enum hrtimer_mode mode
, const clockid_t clockid
)
1560 struct restart_block
*restart
;
1561 struct hrtimer_sleeper t
;
1563 unsigned long slack
;
1565 slack
= current
->timer_slack_ns
;
1566 if (dl_task(current
) || rt_task(current
))
1569 hrtimer_init_on_stack(&t
.timer
, clockid
, mode
);
1570 hrtimer_set_expires_range_ns(&t
.timer
, timespec_to_ktime(*rqtp
), slack
);
1571 if (do_nanosleep(&t
, mode
))
1574 /* Absolute timers do not update the rmtp value and restart: */
1575 if (mode
== HRTIMER_MODE_ABS
) {
1576 ret
= -ERESTARTNOHAND
;
1581 ret
= update_rmtp(&t
.timer
, rmtp
);
1586 restart
= ¤t
->restart_block
;
1587 restart
->fn
= hrtimer_nanosleep_restart
;
1588 restart
->nanosleep
.clockid
= t
.timer
.base
->clockid
;
1589 restart
->nanosleep
.rmtp
= rmtp
;
1590 restart
->nanosleep
.expires
= hrtimer_get_expires_tv64(&t
.timer
);
1592 ret
= -ERESTART_RESTARTBLOCK
;
1594 destroy_hrtimer_on_stack(&t
.timer
);
1598 SYSCALL_DEFINE2(nanosleep
, struct timespec __user
*, rqtp
,
1599 struct timespec __user
*, rmtp
)
1603 if (copy_from_user(&tu
, rqtp
, sizeof(tu
)))
1606 if (!timespec_valid(&tu
))
1609 return hrtimer_nanosleep(&tu
, rmtp
, HRTIMER_MODE_REL
, CLOCK_MONOTONIC
);
1613 * Functions related to boot-time initialization:
1615 static void init_hrtimers_cpu(int cpu
)
1617 struct hrtimer_cpu_base
*cpu_base
= &per_cpu(hrtimer_bases
, cpu
);
1620 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++) {
1621 cpu_base
->clock_base
[i
].cpu_base
= cpu_base
;
1622 timerqueue_init_head(&cpu_base
->clock_base
[i
].active
);
1625 cpu_base
->cpu
= cpu
;
1626 hrtimer_init_hres(cpu_base
);
1629 #ifdef CONFIG_HOTPLUG_CPU
1631 static void migrate_hrtimer_list(struct hrtimer_clock_base
*old_base
,
1632 struct hrtimer_clock_base
*new_base
)
1634 struct hrtimer
*timer
;
1635 struct timerqueue_node
*node
;
1637 while ((node
= timerqueue_getnext(&old_base
->active
))) {
1638 timer
= container_of(node
, struct hrtimer
, node
);
1639 BUG_ON(hrtimer_callback_running(timer
));
1640 debug_deactivate(timer
);
1643 * Mark it as STATE_MIGRATE not INACTIVE otherwise the
1644 * timer could be seen as !active and just vanish away
1645 * under us on another CPU
1647 __remove_hrtimer(timer
, old_base
, HRTIMER_STATE_MIGRATE
, 0);
1648 timer
->base
= new_base
;
1650 * Enqueue the timers on the new cpu. This does not
1651 * reprogram the event device in case the timer
1652 * expires before the earliest on this CPU, but we run
1653 * hrtimer_interrupt after we migrated everything to
1654 * sort out already expired timers and reprogram the
1657 enqueue_hrtimer(timer
, new_base
);
1659 /* Clear the migration state bit */
1660 timer
->state
&= ~HRTIMER_STATE_MIGRATE
;
1664 static void migrate_hrtimers(int scpu
)
1666 struct hrtimer_cpu_base
*old_base
, *new_base
;
1669 BUG_ON(cpu_online(scpu
));
1670 tick_cancel_sched_timer(scpu
);
1672 local_irq_disable();
1673 old_base
= &per_cpu(hrtimer_bases
, scpu
);
1674 new_base
= this_cpu_ptr(&hrtimer_bases
);
1676 * The caller is globally serialized and nobody else
1677 * takes two locks at once, deadlock is not possible.
1679 raw_spin_lock(&new_base
->lock
);
1680 raw_spin_lock_nested(&old_base
->lock
, SINGLE_DEPTH_NESTING
);
1682 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++) {
1683 migrate_hrtimer_list(&old_base
->clock_base
[i
],
1684 &new_base
->clock_base
[i
]);
1687 raw_spin_unlock(&old_base
->lock
);
1688 raw_spin_unlock(&new_base
->lock
);
1690 /* Check, if we got expired work to do */
1691 __hrtimer_peek_ahead_timers();
1695 #endif /* CONFIG_HOTPLUG_CPU */
1697 static int hrtimer_cpu_notify(struct notifier_block
*self
,
1698 unsigned long action
, void *hcpu
)
1700 int scpu
= (long)hcpu
;
1704 case CPU_UP_PREPARE
:
1705 case CPU_UP_PREPARE_FROZEN
:
1706 init_hrtimers_cpu(scpu
);
1709 #ifdef CONFIG_HOTPLUG_CPU
1711 case CPU_DEAD_FROZEN
:
1712 migrate_hrtimers(scpu
);
1723 static struct notifier_block hrtimers_nb
= {
1724 .notifier_call
= hrtimer_cpu_notify
,
1727 void __init
hrtimers_init(void)
1729 hrtimer_cpu_notify(&hrtimers_nb
, (unsigned long)CPU_UP_PREPARE
,
1730 (void *)(long)smp_processor_id());
1731 register_cpu_notifier(&hrtimers_nb
);
1732 #ifdef CONFIG_HIGH_RES_TIMERS
1733 open_softirq(HRTIMER_SOFTIRQ
, run_hrtimer_softirq
);
1738 * schedule_hrtimeout_range_clock - sleep until timeout
1739 * @expires: timeout value (ktime_t)
1740 * @delta: slack in expires timeout (ktime_t)
1741 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1742 * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME
1745 schedule_hrtimeout_range_clock(ktime_t
*expires
, unsigned long delta
,
1746 const enum hrtimer_mode mode
, int clock
)
1748 struct hrtimer_sleeper t
;
1751 * Optimize when a zero timeout value is given. It does not
1752 * matter whether this is an absolute or a relative time.
1754 if (expires
&& !expires
->tv64
) {
1755 __set_current_state(TASK_RUNNING
);
1760 * A NULL parameter means "infinite"
1767 hrtimer_init_on_stack(&t
.timer
, clock
, mode
);
1768 hrtimer_set_expires_range_ns(&t
.timer
, *expires
, delta
);
1770 hrtimer_init_sleeper(&t
, current
);
1772 hrtimer_start_expires(&t
.timer
, mode
);
1773 if (!hrtimer_active(&t
.timer
))
1779 hrtimer_cancel(&t
.timer
);
1780 destroy_hrtimer_on_stack(&t
.timer
);
1782 __set_current_state(TASK_RUNNING
);
1784 return !t
.task
? 0 : -EINTR
;
1788 * schedule_hrtimeout_range - sleep until timeout
1789 * @expires: timeout value (ktime_t)
1790 * @delta: slack in expires timeout (ktime_t)
1791 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1793 * Make the current task sleep until the given expiry time has
1794 * elapsed. The routine will return immediately unless
1795 * the current task state has been set (see set_current_state()).
1797 * The @delta argument gives the kernel the freedom to schedule the
1798 * actual wakeup to a time that is both power and performance friendly.
1799 * The kernel give the normal best effort behavior for "@expires+@delta",
1800 * but may decide to fire the timer earlier, but no earlier than @expires.
1802 * You can set the task state as follows -
1804 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1805 * pass before the routine returns.
1807 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1808 * delivered to the current task.
1810 * The current task state is guaranteed to be TASK_RUNNING when this
1813 * Returns 0 when the timer has expired otherwise -EINTR
1815 int __sched
schedule_hrtimeout_range(ktime_t
*expires
, unsigned long delta
,
1816 const enum hrtimer_mode mode
)
1818 return schedule_hrtimeout_range_clock(expires
, delta
, mode
,
1821 EXPORT_SYMBOL_GPL(schedule_hrtimeout_range
);
1824 * schedule_hrtimeout - sleep until timeout
1825 * @expires: timeout value (ktime_t)
1826 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1828 * Make the current task sleep until the given expiry time has
1829 * elapsed. The routine will return immediately unless
1830 * the current task state has been set (see set_current_state()).
1832 * You can set the task state as follows -
1834 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1835 * pass before the routine returns.
1837 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1838 * delivered to the current task.
1840 * The current task state is guaranteed to be TASK_RUNNING when this
1843 * Returns 0 when the timer has expired otherwise -EINTR
1845 int __sched
schedule_hrtimeout(ktime_t
*expires
,
1846 const enum hrtimer_mode mode
)
1848 return schedule_hrtimeout_range(expires
, 0, mode
);
1850 EXPORT_SYMBOL_GPL(schedule_hrtimeout
);