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 <linux/uaccess.h>
55 #include <trace/events/timer.h>
57 #include "tick-internal.h"
62 * There are more clockids than 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
) =
69 .lock
= __RAW_SPIN_LOCK_UNLOCKED(hrtimer_bases
.lock
),
70 .seq
= SEQCNT_ZERO(hrtimer_bases
.seq
),
74 .index
= HRTIMER_BASE_MONOTONIC
,
75 .clockid
= CLOCK_MONOTONIC
,
76 .get_time
= &ktime_get
,
79 .index
= HRTIMER_BASE_REALTIME
,
80 .clockid
= CLOCK_REALTIME
,
81 .get_time
= &ktime_get_real
,
84 .index
= HRTIMER_BASE_BOOTTIME
,
85 .clockid
= CLOCK_BOOTTIME
,
86 .get_time
= &ktime_get_boottime
,
89 .index
= HRTIMER_BASE_TAI
,
91 .get_time
= &ktime_get_clocktai
,
96 static const int hrtimer_clock_to_base_table
[MAX_CLOCKS
] = {
97 /* Make sure we catch unsupported clockids */
98 [0 ... MAX_CLOCKS
- 1] = HRTIMER_MAX_CLOCK_BASES
,
100 [CLOCK_REALTIME
] = HRTIMER_BASE_REALTIME
,
101 [CLOCK_MONOTONIC
] = HRTIMER_BASE_MONOTONIC
,
102 [CLOCK_BOOTTIME
] = HRTIMER_BASE_BOOTTIME
,
103 [CLOCK_TAI
] = HRTIMER_BASE_TAI
,
107 * Functions and macros which are different for UP/SMP systems are kept in a
113 * We require the migration_base for lock_hrtimer_base()/switch_hrtimer_base()
114 * such that hrtimer_callback_running() can unconditionally dereference
115 * timer->base->cpu_base
117 static struct hrtimer_cpu_base migration_cpu_base
= {
118 .seq
= SEQCNT_ZERO(migration_cpu_base
),
119 .clock_base
= { { .cpu_base
= &migration_cpu_base
, }, },
122 #define migration_base migration_cpu_base.clock_base[0]
125 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
126 * means that all timers which are tied to this base via timer->base are
127 * locked, and the base itself is locked too.
129 * So __run_timers/migrate_timers can safely modify all timers which could
130 * be found on the lists/queues.
132 * When the timer's base is locked, and the timer removed from list, it is
133 * possible to set timer->base = &migration_base and drop the lock: the timer
137 struct hrtimer_clock_base
*lock_hrtimer_base(const struct hrtimer
*timer
,
138 unsigned long *flags
)
140 struct hrtimer_clock_base
*base
;
144 if (likely(base
!= &migration_base
)) {
145 raw_spin_lock_irqsave(&base
->cpu_base
->lock
, *flags
);
146 if (likely(base
== timer
->base
))
148 /* The timer has migrated to another CPU: */
149 raw_spin_unlock_irqrestore(&base
->cpu_base
->lock
, *flags
);
156 * With HIGHRES=y we do not migrate the timer when it is expiring
157 * before the next event on the target cpu because we cannot reprogram
158 * the target cpu hardware and we would cause it to fire late.
160 * Called with cpu_base->lock of target cpu held.
163 hrtimer_check_target(struct hrtimer
*timer
, struct hrtimer_clock_base
*new_base
)
165 #ifdef CONFIG_HIGH_RES_TIMERS
168 if (!new_base
->cpu_base
->hres_active
)
171 expires
= ktime_sub(hrtimer_get_expires(timer
), new_base
->offset
);
172 return expires
<= new_base
->cpu_base
->expires_next
;
178 #ifdef CONFIG_NO_HZ_COMMON
180 struct hrtimer_cpu_base
*get_target_base(struct hrtimer_cpu_base
*base
,
183 if (pinned
|| !base
->migration_enabled
)
185 return &per_cpu(hrtimer_bases
, get_nohz_timer_target());
189 struct hrtimer_cpu_base
*get_target_base(struct hrtimer_cpu_base
*base
,
197 * We switch the timer base to a power-optimized selected CPU target,
199 * - NO_HZ_COMMON is enabled
200 * - timer migration is enabled
201 * - the timer callback is not running
202 * - the timer is not the first expiring timer on the new target
204 * If one of the above requirements is not fulfilled we move the timer
205 * to the current CPU or leave it on the previously assigned CPU if
206 * the timer callback is currently running.
208 static inline struct hrtimer_clock_base
*
209 switch_hrtimer_base(struct hrtimer
*timer
, struct hrtimer_clock_base
*base
,
212 struct hrtimer_cpu_base
*new_cpu_base
, *this_cpu_base
;
213 struct hrtimer_clock_base
*new_base
;
214 int basenum
= base
->index
;
216 this_cpu_base
= this_cpu_ptr(&hrtimer_bases
);
217 new_cpu_base
= get_target_base(this_cpu_base
, pinned
);
219 new_base
= &new_cpu_base
->clock_base
[basenum
];
221 if (base
!= new_base
) {
223 * We are trying to move timer to new_base.
224 * However we can't change timer's base while it is running,
225 * so we keep it on the same CPU. No hassle vs. reprogramming
226 * the event source in the high resolution case. The softirq
227 * code will take care of this when the timer function has
228 * completed. There is no conflict as we hold the lock until
229 * the timer is enqueued.
231 if (unlikely(hrtimer_callback_running(timer
)))
234 /* See the comment in lock_hrtimer_base() */
235 timer
->base
= &migration_base
;
236 raw_spin_unlock(&base
->cpu_base
->lock
);
237 raw_spin_lock(&new_base
->cpu_base
->lock
);
239 if (new_cpu_base
!= this_cpu_base
&&
240 hrtimer_check_target(timer
, new_base
)) {
241 raw_spin_unlock(&new_base
->cpu_base
->lock
);
242 raw_spin_lock(&base
->cpu_base
->lock
);
243 new_cpu_base
= this_cpu_base
;
247 timer
->base
= new_base
;
249 if (new_cpu_base
!= this_cpu_base
&&
250 hrtimer_check_target(timer
, new_base
)) {
251 new_cpu_base
= this_cpu_base
;
258 #else /* CONFIG_SMP */
260 static inline struct hrtimer_clock_base
*
261 lock_hrtimer_base(const struct hrtimer
*timer
, unsigned long *flags
)
263 struct hrtimer_clock_base
*base
= timer
->base
;
265 raw_spin_lock_irqsave(&base
->cpu_base
->lock
, *flags
);
270 # define switch_hrtimer_base(t, b, p) (b)
272 #endif /* !CONFIG_SMP */
275 * Functions for the union type storage format of ktime_t which are
276 * too large for inlining:
278 #if BITS_PER_LONG < 64
280 * Divide a ktime value by a nanosecond value
282 s64
__ktime_divns(const ktime_t kt
, s64 div
)
288 dclc
= ktime_to_ns(kt
);
289 tmp
= dclc
< 0 ? -dclc
: dclc
;
291 /* Make sure the divisor is less than 2^32: */
297 do_div(tmp
, (unsigned long) div
);
298 return dclc
< 0 ? -tmp
: tmp
;
300 EXPORT_SYMBOL_GPL(__ktime_divns
);
301 #endif /* BITS_PER_LONG >= 64 */
304 * Add two ktime values and do a safety check for overflow:
306 ktime_t
ktime_add_safe(const ktime_t lhs
, const ktime_t rhs
)
308 ktime_t res
= ktime_add_unsafe(lhs
, rhs
);
311 * We use KTIME_SEC_MAX here, the maximum timeout which we can
312 * return to user space in a timespec:
314 if (res
< 0 || res
< lhs
|| res
< rhs
)
315 res
= ktime_set(KTIME_SEC_MAX
, 0);
320 EXPORT_SYMBOL_GPL(ktime_add_safe
);
322 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
324 static struct debug_obj_descr hrtimer_debug_descr
;
326 static void *hrtimer_debug_hint(void *addr
)
328 return ((struct hrtimer
*) addr
)->function
;
332 * fixup_init is called when:
333 * - an active object is initialized
335 static bool hrtimer_fixup_init(void *addr
, enum debug_obj_state state
)
337 struct hrtimer
*timer
= addr
;
340 case ODEBUG_STATE_ACTIVE
:
341 hrtimer_cancel(timer
);
342 debug_object_init(timer
, &hrtimer_debug_descr
);
350 * fixup_activate is called when:
351 * - an active object is activated
352 * - an unknown non-static object is activated
354 static bool hrtimer_fixup_activate(void *addr
, enum debug_obj_state state
)
357 case ODEBUG_STATE_ACTIVE
:
366 * fixup_free is called when:
367 * - an active object is freed
369 static bool hrtimer_fixup_free(void *addr
, enum debug_obj_state state
)
371 struct hrtimer
*timer
= addr
;
374 case ODEBUG_STATE_ACTIVE
:
375 hrtimer_cancel(timer
);
376 debug_object_free(timer
, &hrtimer_debug_descr
);
383 static struct debug_obj_descr hrtimer_debug_descr
= {
385 .debug_hint
= hrtimer_debug_hint
,
386 .fixup_init
= hrtimer_fixup_init
,
387 .fixup_activate
= hrtimer_fixup_activate
,
388 .fixup_free
= hrtimer_fixup_free
,
391 static inline void debug_hrtimer_init(struct hrtimer
*timer
)
393 debug_object_init(timer
, &hrtimer_debug_descr
);
396 static inline void debug_hrtimer_activate(struct hrtimer
*timer
)
398 debug_object_activate(timer
, &hrtimer_debug_descr
);
401 static inline void debug_hrtimer_deactivate(struct hrtimer
*timer
)
403 debug_object_deactivate(timer
, &hrtimer_debug_descr
);
406 static inline void debug_hrtimer_free(struct hrtimer
*timer
)
408 debug_object_free(timer
, &hrtimer_debug_descr
);
411 static void __hrtimer_init(struct hrtimer
*timer
, clockid_t clock_id
,
412 enum hrtimer_mode mode
);
414 void hrtimer_init_on_stack(struct hrtimer
*timer
, clockid_t clock_id
,
415 enum hrtimer_mode mode
)
417 debug_object_init_on_stack(timer
, &hrtimer_debug_descr
);
418 __hrtimer_init(timer
, clock_id
, mode
);
420 EXPORT_SYMBOL_GPL(hrtimer_init_on_stack
);
422 void destroy_hrtimer_on_stack(struct hrtimer
*timer
)
424 debug_object_free(timer
, &hrtimer_debug_descr
);
426 EXPORT_SYMBOL_GPL(destroy_hrtimer_on_stack
);
429 static inline void debug_hrtimer_init(struct hrtimer
*timer
) { }
430 static inline void debug_hrtimer_activate(struct hrtimer
*timer
) { }
431 static inline void debug_hrtimer_deactivate(struct hrtimer
*timer
) { }
435 debug_init(struct hrtimer
*timer
, clockid_t clockid
,
436 enum hrtimer_mode mode
)
438 debug_hrtimer_init(timer
);
439 trace_hrtimer_init(timer
, clockid
, mode
);
442 static inline void debug_activate(struct hrtimer
*timer
)
444 debug_hrtimer_activate(timer
);
445 trace_hrtimer_start(timer
);
448 static inline void debug_deactivate(struct hrtimer
*timer
)
450 debug_hrtimer_deactivate(timer
);
451 trace_hrtimer_cancel(timer
);
454 #if defined(CONFIG_NO_HZ_COMMON) || defined(CONFIG_HIGH_RES_TIMERS)
455 static inline void hrtimer_update_next_timer(struct hrtimer_cpu_base
*cpu_base
,
456 struct hrtimer
*timer
)
458 #ifdef CONFIG_HIGH_RES_TIMERS
459 cpu_base
->next_timer
= timer
;
463 static ktime_t
__hrtimer_get_next_event(struct hrtimer_cpu_base
*cpu_base
)
465 struct hrtimer_clock_base
*base
= cpu_base
->clock_base
;
466 unsigned int active
= cpu_base
->active_bases
;
467 ktime_t expires
, expires_next
= KTIME_MAX
;
469 hrtimer_update_next_timer(cpu_base
, NULL
);
470 for (; active
; base
++, active
>>= 1) {
471 struct timerqueue_node
*next
;
472 struct hrtimer
*timer
;
474 if (!(active
& 0x01))
477 next
= timerqueue_getnext(&base
->active
);
478 timer
= container_of(next
, struct hrtimer
, node
);
479 expires
= ktime_sub(hrtimer_get_expires(timer
), base
->offset
);
480 if (expires
< expires_next
) {
481 expires_next
= expires
;
482 hrtimer_update_next_timer(cpu_base
, timer
);
486 * clock_was_set() might have changed base->offset of any of
487 * the clock bases so the result might be negative. Fix it up
488 * to prevent a false positive in clockevents_program_event().
490 if (expires_next
< 0)
496 static inline ktime_t
hrtimer_update_base(struct hrtimer_cpu_base
*base
)
498 ktime_t
*offs_real
= &base
->clock_base
[HRTIMER_BASE_REALTIME
].offset
;
499 ktime_t
*offs_boot
= &base
->clock_base
[HRTIMER_BASE_BOOTTIME
].offset
;
500 ktime_t
*offs_tai
= &base
->clock_base
[HRTIMER_BASE_TAI
].offset
;
502 return ktime_get_update_offsets_now(&base
->clock_was_set_seq
,
503 offs_real
, offs_boot
, offs_tai
);
506 /* High resolution timer related functions */
507 #ifdef CONFIG_HIGH_RES_TIMERS
510 * High resolution timer enabled ?
512 static bool hrtimer_hres_enabled __read_mostly
= true;
513 unsigned int hrtimer_resolution __read_mostly
= LOW_RES_NSEC
;
514 EXPORT_SYMBOL_GPL(hrtimer_resolution
);
517 * Enable / Disable high resolution mode
519 static int __init
setup_hrtimer_hres(char *str
)
521 return (kstrtobool(str
, &hrtimer_hres_enabled
) == 0);
524 __setup("highres=", setup_hrtimer_hres
);
527 * hrtimer_high_res_enabled - query, if the highres mode is enabled
529 static inline int hrtimer_is_hres_enabled(void)
531 return hrtimer_hres_enabled
;
535 * Is the high resolution mode active ?
537 static inline int __hrtimer_hres_active(struct hrtimer_cpu_base
*cpu_base
)
539 return cpu_base
->hres_active
;
542 static inline int hrtimer_hres_active(void)
544 return __hrtimer_hres_active(this_cpu_ptr(&hrtimer_bases
));
548 * Reprogram the event source with checking both queues for the
550 * Called with interrupts disabled and base->lock held
553 hrtimer_force_reprogram(struct hrtimer_cpu_base
*cpu_base
, int skip_equal
)
555 ktime_t expires_next
;
557 if (!cpu_base
->hres_active
)
560 expires_next
= __hrtimer_get_next_event(cpu_base
);
562 if (skip_equal
&& expires_next
== cpu_base
->expires_next
)
565 cpu_base
->expires_next
= expires_next
;
568 * If a hang was detected in the last timer interrupt then we
569 * leave the hang delay active in the hardware. We want the
570 * system to make progress. That also prevents the following
572 * T1 expires 50ms from now
573 * T2 expires 5s from now
575 * T1 is removed, so this code is called and would reprogram
576 * the hardware to 5s from now. Any hrtimer_start after that
577 * will not reprogram the hardware due to hang_detected being
578 * set. So we'd effectivly block all timers until the T2 event
581 if (cpu_base
->hang_detected
)
584 tick_program_event(cpu_base
->expires_next
, 1);
588 * When a timer is enqueued and expires earlier than the already enqueued
589 * timers, we have to check, whether it expires earlier than the timer for
590 * which the clock event device was armed.
592 * Called with interrupts disabled and base->cpu_base.lock held
594 static void hrtimer_reprogram(struct hrtimer
*timer
,
595 struct hrtimer_clock_base
*base
)
597 struct hrtimer_cpu_base
*cpu_base
= this_cpu_ptr(&hrtimer_bases
);
598 ktime_t expires
= ktime_sub(hrtimer_get_expires(timer
), base
->offset
);
600 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer
) < 0);
603 * If the timer is not on the current cpu, we cannot reprogram
604 * the other cpus clock event device.
606 if (base
->cpu_base
!= cpu_base
)
610 * If the hrtimer interrupt is running, then it will
611 * reevaluate the clock bases and reprogram the clock event
612 * device. The callbacks are always executed in hard interrupt
613 * context so we don't need an extra check for a running
616 if (cpu_base
->in_hrtirq
)
620 * CLOCK_REALTIME timer might be requested with an absolute
621 * expiry time which is less than base->offset. Set it to 0.
626 if (expires
>= cpu_base
->expires_next
)
629 /* Update the pointer to the next expiring timer */
630 cpu_base
->next_timer
= timer
;
633 * If a hang was detected in the last timer interrupt then we
634 * do not schedule a timer which is earlier than the expiry
635 * which we enforced in the hang detection. We want the system
638 if (cpu_base
->hang_detected
)
642 * Program the timer hardware. We enforce the expiry for
643 * events which are already in the past.
645 cpu_base
->expires_next
= expires
;
646 tick_program_event(expires
, 1);
650 * Initialize the high resolution related parts of cpu_base
652 static inline void hrtimer_init_hres(struct hrtimer_cpu_base
*base
)
654 base
->expires_next
= KTIME_MAX
;
655 base
->hres_active
= 0;
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
= this_cpu_ptr(&hrtimer_bases
);
667 if (!base
->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 void hrtimer_switch_to_hres(void)
681 struct hrtimer_cpu_base
*base
= this_cpu_ptr(&hrtimer_bases
);
683 if (tick_init_highres()) {
684 printk(KERN_WARNING
"Could not switch to high resolution "
685 "mode on CPU %d\n", base
->cpu
);
688 base
->hres_active
= 1;
689 hrtimer_resolution
= HIGH_RES_NSEC
;
691 tick_setup_sched_timer();
692 /* "Retrigger" the interrupt to get things going */
693 retrigger_next_event(NULL
);
696 static void clock_was_set_work(struct work_struct
*work
)
701 static DECLARE_WORK(hrtimer_work
, clock_was_set_work
);
704 * Called from timekeeping and resume code to reprogram the hrtimer
705 * interrupt device on all cpus.
707 void clock_was_set_delayed(void)
709 schedule_work(&hrtimer_work
);
714 static inline int __hrtimer_hres_active(struct hrtimer_cpu_base
*b
) { return 0; }
715 static inline int hrtimer_hres_active(void) { return 0; }
716 static inline int hrtimer_is_hres_enabled(void) { return 0; }
717 static inline void hrtimer_switch_to_hres(void) { }
719 hrtimer_force_reprogram(struct hrtimer_cpu_base
*base
, int skip_equal
) { }
720 static inline int hrtimer_reprogram(struct hrtimer
*timer
,
721 struct hrtimer_clock_base
*base
)
725 static inline void hrtimer_init_hres(struct hrtimer_cpu_base
*base
) { }
726 static inline void retrigger_next_event(void *arg
) { }
728 #endif /* CONFIG_HIGH_RES_TIMERS */
731 * Clock realtime was set
733 * Change the offset of the realtime clock vs. the monotonic
736 * We might have to reprogram the high resolution timer interrupt. On
737 * SMP we call the architecture specific code to retrigger _all_ high
738 * resolution timer interrupts. On UP we just disable interrupts and
739 * call the high resolution interrupt code.
741 void clock_was_set(void)
743 #ifdef CONFIG_HIGH_RES_TIMERS
744 /* Retrigger the CPU local events everywhere */
745 on_each_cpu(retrigger_next_event
, NULL
, 1);
747 timerfd_clock_was_set();
751 * During resume we might have to reprogram the high resolution timer
752 * interrupt on all online CPUs. However, all other CPUs will be
753 * stopped with IRQs interrupts disabled so the clock_was_set() call
756 void hrtimers_resume(void)
758 WARN_ONCE(!irqs_disabled(),
759 KERN_INFO
"hrtimers_resume() called with IRQs enabled!");
761 /* Retrigger on the local CPU */
762 retrigger_next_event(NULL
);
763 /* And schedule a retrigger for all others */
764 clock_was_set_delayed();
768 * Counterpart to lock_hrtimer_base above:
771 void unlock_hrtimer_base(const struct hrtimer
*timer
, unsigned long *flags
)
773 raw_spin_unlock_irqrestore(&timer
->base
->cpu_base
->lock
, *flags
);
777 * hrtimer_forward - forward the timer expiry
778 * @timer: hrtimer to forward
779 * @now: forward past this time
780 * @interval: the interval to forward
782 * Forward the timer expiry so it will expire in the future.
783 * Returns the number of overruns.
785 * Can be safely called from the callback function of @timer. If
786 * called from other contexts @timer must neither be enqueued nor
787 * running the callback and the caller needs to take care of
790 * Note: This only updates the timer expiry value and does not requeue
793 u64
hrtimer_forward(struct hrtimer
*timer
, ktime_t now
, ktime_t interval
)
798 delta
= ktime_sub(now
, hrtimer_get_expires(timer
));
803 if (WARN_ON(timer
->state
& HRTIMER_STATE_ENQUEUED
))
806 if (interval
< hrtimer_resolution
)
807 interval
= hrtimer_resolution
;
809 if (unlikely(delta
>= interval
)) {
810 s64 incr
= ktime_to_ns(interval
);
812 orun
= ktime_divns(delta
, incr
);
813 hrtimer_add_expires_ns(timer
, incr
* orun
);
814 if (hrtimer_get_expires_tv64(timer
) > now
)
817 * This (and the ktime_add() below) is the
818 * correction for exact:
822 hrtimer_add_expires(timer
, interval
);
826 EXPORT_SYMBOL_GPL(hrtimer_forward
);
829 * enqueue_hrtimer - internal function to (re)start a timer
831 * The timer is inserted in expiry order. Insertion into the
832 * red black tree is O(log(n)). Must hold the base lock.
834 * Returns 1 when the new timer is the leftmost timer in the tree.
836 static int enqueue_hrtimer(struct hrtimer
*timer
,
837 struct hrtimer_clock_base
*base
)
839 debug_activate(timer
);
841 base
->cpu_base
->active_bases
|= 1 << base
->index
;
843 timer
->state
= HRTIMER_STATE_ENQUEUED
;
845 return timerqueue_add(&base
->active
, &timer
->node
);
849 * __remove_hrtimer - internal function to remove a timer
851 * Caller must hold the base lock.
853 * High resolution timer mode reprograms the clock event device when the
854 * timer is the one which expires next. The caller can disable this by setting
855 * reprogram to zero. This is useful, when the context does a reprogramming
856 * anyway (e.g. timer interrupt)
858 static void __remove_hrtimer(struct hrtimer
*timer
,
859 struct hrtimer_clock_base
*base
,
860 u8 newstate
, int reprogram
)
862 struct hrtimer_cpu_base
*cpu_base
= base
->cpu_base
;
863 u8 state
= timer
->state
;
865 timer
->state
= newstate
;
866 if (!(state
& HRTIMER_STATE_ENQUEUED
))
869 if (!timerqueue_del(&base
->active
, &timer
->node
))
870 cpu_base
->active_bases
&= ~(1 << base
->index
);
872 #ifdef CONFIG_HIGH_RES_TIMERS
874 * Note: If reprogram is false we do not update
875 * cpu_base->next_timer. This happens when we remove the first
876 * timer on a remote cpu. No harm as we never dereference
877 * cpu_base->next_timer. So the worst thing what can happen is
878 * an superflous call to hrtimer_force_reprogram() on the
879 * remote cpu later on if the same timer gets enqueued again.
881 if (reprogram
&& timer
== cpu_base
->next_timer
)
882 hrtimer_force_reprogram(cpu_base
, 1);
887 * remove hrtimer, called with base lock held
890 remove_hrtimer(struct hrtimer
*timer
, struct hrtimer_clock_base
*base
, bool restart
)
892 if (hrtimer_is_queued(timer
)) {
893 u8 state
= timer
->state
;
897 * Remove the timer and force reprogramming when high
898 * resolution mode is active and the timer is on the current
899 * CPU. If we remove a timer on another CPU, reprogramming is
900 * skipped. The interrupt event on this CPU is fired and
901 * reprogramming happens in the interrupt handler. This is a
902 * rare case and less expensive than a smp call.
904 debug_deactivate(timer
);
905 reprogram
= base
->cpu_base
== this_cpu_ptr(&hrtimer_bases
);
908 state
= HRTIMER_STATE_INACTIVE
;
910 __remove_hrtimer(timer
, base
, state
, reprogram
);
916 static inline ktime_t
hrtimer_update_lowres(struct hrtimer
*timer
, ktime_t tim
,
917 const enum hrtimer_mode mode
)
919 #ifdef CONFIG_TIME_LOW_RES
921 * CONFIG_TIME_LOW_RES indicates that the system has no way to return
922 * granular time values. For relative timers we add hrtimer_resolution
923 * (i.e. one jiffie) to prevent short timeouts.
925 timer
->is_rel
= mode
& HRTIMER_MODE_REL
;
927 tim
= ktime_add_safe(tim
, hrtimer_resolution
);
933 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
934 * @timer: the timer to be added
936 * @delta_ns: "slack" range for the timer
937 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
938 * relative (HRTIMER_MODE_REL)
940 void hrtimer_start_range_ns(struct hrtimer
*timer
, ktime_t tim
,
941 u64 delta_ns
, const enum hrtimer_mode mode
)
943 struct hrtimer_clock_base
*base
, *new_base
;
947 base
= lock_hrtimer_base(timer
, &flags
);
949 /* Remove an active timer from the queue: */
950 remove_hrtimer(timer
, base
, true);
952 if (mode
& HRTIMER_MODE_REL
)
953 tim
= ktime_add_safe(tim
, base
->get_time());
955 tim
= hrtimer_update_lowres(timer
, tim
, mode
);
957 hrtimer_set_expires_range_ns(timer
, tim
, delta_ns
);
959 /* Switch the timer base, if necessary: */
960 new_base
= switch_hrtimer_base(timer
, base
, mode
& HRTIMER_MODE_PINNED
);
962 leftmost
= enqueue_hrtimer(timer
, new_base
);
966 if (!hrtimer_is_hres_active(timer
)) {
968 * Kick to reschedule the next tick to handle the new timer
969 * on dynticks target.
971 if (new_base
->cpu_base
->nohz_active
)
972 wake_up_nohz_cpu(new_base
->cpu_base
->cpu
);
974 hrtimer_reprogram(timer
, new_base
);
977 unlock_hrtimer_base(timer
, &flags
);
979 EXPORT_SYMBOL_GPL(hrtimer_start_range_ns
);
982 * hrtimer_try_to_cancel - try to deactivate a timer
983 * @timer: hrtimer to stop
986 * 0 when the timer was not active
987 * 1 when the timer was active
988 * -1 when the timer is currently excuting the callback function and
991 int hrtimer_try_to_cancel(struct hrtimer
*timer
)
993 struct hrtimer_clock_base
*base
;
998 * Check lockless first. If the timer is not active (neither
999 * enqueued nor running the callback, nothing to do here. The
1000 * base lock does not serialize against a concurrent enqueue,
1001 * so we can avoid taking it.
1003 if (!hrtimer_active(timer
))
1006 base
= lock_hrtimer_base(timer
, &flags
);
1008 if (!hrtimer_callback_running(timer
))
1009 ret
= remove_hrtimer(timer
, base
, false);
1011 unlock_hrtimer_base(timer
, &flags
);
1016 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel
);
1019 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1020 * @timer: the timer to be cancelled
1023 * 0 when the timer was not active
1024 * 1 when the timer was active
1026 int hrtimer_cancel(struct hrtimer
*timer
)
1029 int ret
= hrtimer_try_to_cancel(timer
);
1036 EXPORT_SYMBOL_GPL(hrtimer_cancel
);
1039 * hrtimer_get_remaining - get remaining time for the timer
1040 * @timer: the timer to read
1041 * @adjust: adjust relative timers when CONFIG_TIME_LOW_RES=y
1043 ktime_t
__hrtimer_get_remaining(const struct hrtimer
*timer
, bool adjust
)
1045 unsigned long flags
;
1048 lock_hrtimer_base(timer
, &flags
);
1049 if (IS_ENABLED(CONFIG_TIME_LOW_RES
) && adjust
)
1050 rem
= hrtimer_expires_remaining_adjusted(timer
);
1052 rem
= hrtimer_expires_remaining(timer
);
1053 unlock_hrtimer_base(timer
, &flags
);
1057 EXPORT_SYMBOL_GPL(__hrtimer_get_remaining
);
1059 #ifdef CONFIG_NO_HZ_COMMON
1061 * hrtimer_get_next_event - get the time until next expiry event
1063 * Returns the next expiry time or KTIME_MAX if no timer is pending.
1065 u64
hrtimer_get_next_event(void)
1067 struct hrtimer_cpu_base
*cpu_base
= this_cpu_ptr(&hrtimer_bases
);
1068 u64 expires
= KTIME_MAX
;
1069 unsigned long flags
;
1071 raw_spin_lock_irqsave(&cpu_base
->lock
, flags
);
1073 if (!__hrtimer_hres_active(cpu_base
))
1074 expires
= __hrtimer_get_next_event(cpu_base
);
1076 raw_spin_unlock_irqrestore(&cpu_base
->lock
, flags
);
1082 static inline int hrtimer_clockid_to_base(clockid_t clock_id
)
1084 if (likely(clock_id
< MAX_CLOCKS
)) {
1085 int base
= hrtimer_clock_to_base_table
[clock_id
];
1087 if (likely(base
!= HRTIMER_MAX_CLOCK_BASES
))
1090 WARN(1, "Invalid clockid %d. Using MONOTONIC\n", clock_id
);
1091 return HRTIMER_BASE_MONOTONIC
;
1094 static void __hrtimer_init(struct hrtimer
*timer
, clockid_t clock_id
,
1095 enum hrtimer_mode mode
)
1097 struct hrtimer_cpu_base
*cpu_base
;
1100 memset(timer
, 0, sizeof(struct hrtimer
));
1102 cpu_base
= raw_cpu_ptr(&hrtimer_bases
);
1104 if (clock_id
== CLOCK_REALTIME
&& mode
!= HRTIMER_MODE_ABS
)
1105 clock_id
= CLOCK_MONOTONIC
;
1107 base
= hrtimer_clockid_to_base(clock_id
);
1108 timer
->base
= &cpu_base
->clock_base
[base
];
1109 timerqueue_init(&timer
->node
);
1113 * hrtimer_init - initialize a timer to the given clock
1114 * @timer: the timer to be initialized
1115 * @clock_id: the clock to be used
1116 * @mode: timer mode abs/rel
1118 void hrtimer_init(struct hrtimer
*timer
, clockid_t clock_id
,
1119 enum hrtimer_mode mode
)
1121 debug_init(timer
, clock_id
, mode
);
1122 __hrtimer_init(timer
, clock_id
, mode
);
1124 EXPORT_SYMBOL_GPL(hrtimer_init
);
1127 * A timer is active, when it is enqueued into the rbtree or the
1128 * callback function is running or it's in the state of being migrated
1131 * It is important for this function to not return a false negative.
1133 bool hrtimer_active(const struct hrtimer
*timer
)
1135 struct hrtimer_cpu_base
*cpu_base
;
1139 cpu_base
= READ_ONCE(timer
->base
->cpu_base
);
1140 seq
= raw_read_seqcount_begin(&cpu_base
->seq
);
1142 if (timer
->state
!= HRTIMER_STATE_INACTIVE
||
1143 cpu_base
->running
== timer
)
1146 } while (read_seqcount_retry(&cpu_base
->seq
, seq
) ||
1147 cpu_base
!= READ_ONCE(timer
->base
->cpu_base
));
1151 EXPORT_SYMBOL_GPL(hrtimer_active
);
1154 * The write_seqcount_barrier()s in __run_hrtimer() split the thing into 3
1155 * distinct sections:
1157 * - queued: the timer is queued
1158 * - callback: the timer is being ran
1159 * - post: the timer is inactive or (re)queued
1161 * On the read side we ensure we observe timer->state and cpu_base->running
1162 * from the same section, if anything changed while we looked at it, we retry.
1163 * This includes timer->base changing because sequence numbers alone are
1164 * insufficient for that.
1166 * The sequence numbers are required because otherwise we could still observe
1167 * a false negative if the read side got smeared over multiple consequtive
1168 * __run_hrtimer() invocations.
1171 static void __run_hrtimer(struct hrtimer_cpu_base
*cpu_base
,
1172 struct hrtimer_clock_base
*base
,
1173 struct hrtimer
*timer
, ktime_t
*now
)
1175 enum hrtimer_restart (*fn
)(struct hrtimer
*);
1178 lockdep_assert_held(&cpu_base
->lock
);
1180 debug_deactivate(timer
);
1181 cpu_base
->running
= timer
;
1184 * Separate the ->running assignment from the ->state assignment.
1186 * As with a regular write barrier, this ensures the read side in
1187 * hrtimer_active() cannot observe cpu_base->running == NULL &&
1188 * timer->state == INACTIVE.
1190 raw_write_seqcount_barrier(&cpu_base
->seq
);
1192 __remove_hrtimer(timer
, base
, HRTIMER_STATE_INACTIVE
, 0);
1193 fn
= timer
->function
;
1196 * Clear the 'is relative' flag for the TIME_LOW_RES case. If the
1197 * timer is restarted with a period then it becomes an absolute
1198 * timer. If its not restarted it does not matter.
1200 if (IS_ENABLED(CONFIG_TIME_LOW_RES
))
1201 timer
->is_rel
= false;
1204 * Because we run timers from hardirq context, there is no chance
1205 * they get migrated to another cpu, therefore its safe to unlock
1208 raw_spin_unlock(&cpu_base
->lock
);
1209 trace_hrtimer_expire_entry(timer
, now
);
1210 restart
= fn(timer
);
1211 trace_hrtimer_expire_exit(timer
);
1212 raw_spin_lock(&cpu_base
->lock
);
1215 * Note: We clear the running state after enqueue_hrtimer and
1216 * we do not reprogram the event hardware. Happens either in
1217 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1219 * Note: Because we dropped the cpu_base->lock above,
1220 * hrtimer_start_range_ns() can have popped in and enqueued the timer
1223 if (restart
!= HRTIMER_NORESTART
&&
1224 !(timer
->state
& HRTIMER_STATE_ENQUEUED
))
1225 enqueue_hrtimer(timer
, base
);
1228 * Separate the ->running assignment from the ->state assignment.
1230 * As with a regular write barrier, this ensures the read side in
1231 * hrtimer_active() cannot observe cpu_base->running == NULL &&
1232 * timer->state == INACTIVE.
1234 raw_write_seqcount_barrier(&cpu_base
->seq
);
1236 WARN_ON_ONCE(cpu_base
->running
!= timer
);
1237 cpu_base
->running
= NULL
;
1240 static void __hrtimer_run_queues(struct hrtimer_cpu_base
*cpu_base
, ktime_t now
)
1242 struct hrtimer_clock_base
*base
= cpu_base
->clock_base
;
1243 unsigned int active
= cpu_base
->active_bases
;
1245 for (; active
; base
++, active
>>= 1) {
1246 struct timerqueue_node
*node
;
1249 if (!(active
& 0x01))
1252 basenow
= ktime_add(now
, base
->offset
);
1254 while ((node
= timerqueue_getnext(&base
->active
))) {
1255 struct hrtimer
*timer
;
1257 timer
= container_of(node
, struct hrtimer
, node
);
1260 * The immediate goal for using the softexpires is
1261 * minimizing wakeups, not running timers at the
1262 * earliest interrupt after their soft expiration.
1263 * This allows us to avoid using a Priority Search
1264 * Tree, which can answer a stabbing querry for
1265 * overlapping intervals and instead use the simple
1266 * BST we already have.
1267 * We don't add extra wakeups by delaying timers that
1268 * are right-of a not yet expired timer, because that
1269 * timer will have to trigger a wakeup anyway.
1271 if (basenow
< hrtimer_get_softexpires_tv64(timer
))
1274 __run_hrtimer(cpu_base
, base
, timer
, &basenow
);
1279 #ifdef CONFIG_HIGH_RES_TIMERS
1282 * High resolution timer interrupt
1283 * Called with interrupts disabled
1285 void hrtimer_interrupt(struct clock_event_device
*dev
)
1287 struct hrtimer_cpu_base
*cpu_base
= this_cpu_ptr(&hrtimer_bases
);
1288 ktime_t expires_next
, now
, entry_time
, delta
;
1291 BUG_ON(!cpu_base
->hres_active
);
1292 cpu_base
->nr_events
++;
1293 dev
->next_event
= KTIME_MAX
;
1295 raw_spin_lock(&cpu_base
->lock
);
1296 entry_time
= now
= hrtimer_update_base(cpu_base
);
1298 cpu_base
->in_hrtirq
= 1;
1300 * We set expires_next to KTIME_MAX here with cpu_base->lock
1301 * held to prevent that a timer is enqueued in our queue via
1302 * the migration code. This does not affect enqueueing of
1303 * timers which run their callback and need to be requeued on
1306 cpu_base
->expires_next
= KTIME_MAX
;
1308 __hrtimer_run_queues(cpu_base
, now
);
1310 /* Reevaluate the clock bases for the next expiry */
1311 expires_next
= __hrtimer_get_next_event(cpu_base
);
1313 * Store the new expiry value so the migration code can verify
1316 cpu_base
->expires_next
= expires_next
;
1317 cpu_base
->in_hrtirq
= 0;
1318 raw_spin_unlock(&cpu_base
->lock
);
1320 /* Reprogramming necessary ? */
1321 if (!tick_program_event(expires_next
, 0)) {
1322 cpu_base
->hang_detected
= 0;
1327 * The next timer was already expired due to:
1329 * - long lasting callbacks
1330 * - being scheduled away when running in a VM
1332 * We need to prevent that we loop forever in the hrtimer
1333 * interrupt routine. We give it 3 attempts to avoid
1334 * overreacting on some spurious event.
1336 * Acquire base lock for updating the offsets and retrieving
1339 raw_spin_lock(&cpu_base
->lock
);
1340 now
= hrtimer_update_base(cpu_base
);
1341 cpu_base
->nr_retries
++;
1345 * Give the system a chance to do something else than looping
1346 * here. We stored the entry time, so we know exactly how long
1347 * we spent here. We schedule the next event this amount of
1350 cpu_base
->nr_hangs
++;
1351 cpu_base
->hang_detected
= 1;
1352 raw_spin_unlock(&cpu_base
->lock
);
1353 delta
= ktime_sub(now
, entry_time
);
1354 if ((unsigned int)delta
> cpu_base
->max_hang_time
)
1355 cpu_base
->max_hang_time
= (unsigned int) delta
;
1357 * Limit it to a sensible value as we enforce a longer
1358 * delay. Give the CPU at least 100ms to catch up.
1360 if (delta
> 100 * NSEC_PER_MSEC
)
1361 expires_next
= ktime_add_ns(now
, 100 * NSEC_PER_MSEC
);
1363 expires_next
= ktime_add(now
, delta
);
1364 tick_program_event(expires_next
, 1);
1365 printk_once(KERN_WARNING
"hrtimer: interrupt took %llu ns\n",
1366 ktime_to_ns(delta
));
1370 * local version of hrtimer_peek_ahead_timers() called with interrupts
1373 static inline void __hrtimer_peek_ahead_timers(void)
1375 struct tick_device
*td
;
1377 if (!hrtimer_hres_active())
1380 td
= this_cpu_ptr(&tick_cpu_device
);
1381 if (td
&& td
->evtdev
)
1382 hrtimer_interrupt(td
->evtdev
);
1385 #else /* CONFIG_HIGH_RES_TIMERS */
1387 static inline void __hrtimer_peek_ahead_timers(void) { }
1389 #endif /* !CONFIG_HIGH_RES_TIMERS */
1392 * Called from run_local_timers in hardirq context every jiffy
1394 void hrtimer_run_queues(void)
1396 struct hrtimer_cpu_base
*cpu_base
= this_cpu_ptr(&hrtimer_bases
);
1399 if (__hrtimer_hres_active(cpu_base
))
1403 * This _is_ ugly: We have to check periodically, whether we
1404 * can switch to highres and / or nohz mode. The clocksource
1405 * switch happens with xtime_lock held. Notification from
1406 * there only sets the check bit in the tick_oneshot code,
1407 * otherwise we might deadlock vs. xtime_lock.
1409 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) {
1410 hrtimer_switch_to_hres();
1414 raw_spin_lock(&cpu_base
->lock
);
1415 now
= hrtimer_update_base(cpu_base
);
1416 __hrtimer_run_queues(cpu_base
, now
);
1417 raw_spin_unlock(&cpu_base
->lock
);
1421 * Sleep related functions:
1423 static enum hrtimer_restart
hrtimer_wakeup(struct hrtimer
*timer
)
1425 struct hrtimer_sleeper
*t
=
1426 container_of(timer
, struct hrtimer_sleeper
, timer
);
1427 struct task_struct
*task
= t
->task
;
1431 wake_up_process(task
);
1433 return HRTIMER_NORESTART
;
1436 void hrtimer_init_sleeper(struct hrtimer_sleeper
*sl
, struct task_struct
*task
)
1438 sl
->timer
.function
= hrtimer_wakeup
;
1441 EXPORT_SYMBOL_GPL(hrtimer_init_sleeper
);
1443 static int __sched
do_nanosleep(struct hrtimer_sleeper
*t
, enum hrtimer_mode mode
)
1445 hrtimer_init_sleeper(t
, current
);
1448 set_current_state(TASK_INTERRUPTIBLE
);
1449 hrtimer_start_expires(&t
->timer
, mode
);
1451 if (likely(t
->task
))
1452 freezable_schedule();
1454 hrtimer_cancel(&t
->timer
);
1455 mode
= HRTIMER_MODE_ABS
;
1457 } while (t
->task
&& !signal_pending(current
));
1459 __set_current_state(TASK_RUNNING
);
1461 return t
->task
== NULL
;
1464 static int update_rmtp(struct hrtimer
*timer
, struct timespec __user
*rmtp
)
1466 struct timespec rmt
;
1469 rem
= hrtimer_expires_remaining(timer
);
1472 rmt
= ktime_to_timespec(rem
);
1474 if (copy_to_user(rmtp
, &rmt
, sizeof(*rmtp
)))
1480 long __sched
hrtimer_nanosleep_restart(struct restart_block
*restart
)
1482 struct hrtimer_sleeper t
;
1483 struct timespec __user
*rmtp
;
1486 hrtimer_init_on_stack(&t
.timer
, restart
->nanosleep
.clockid
,
1488 hrtimer_set_expires_tv64(&t
.timer
, restart
->nanosleep
.expires
);
1490 if (do_nanosleep(&t
, HRTIMER_MODE_ABS
))
1493 rmtp
= restart
->nanosleep
.rmtp
;
1495 ret
= update_rmtp(&t
.timer
, rmtp
);
1500 /* The other values in restart are already filled in */
1501 ret
= -ERESTART_RESTARTBLOCK
;
1503 destroy_hrtimer_on_stack(&t
.timer
);
1507 long hrtimer_nanosleep(struct timespec
*rqtp
, struct timespec __user
*rmtp
,
1508 const enum hrtimer_mode mode
, const clockid_t clockid
)
1510 struct restart_block
*restart
;
1511 struct hrtimer_sleeper t
;
1515 slack
= current
->timer_slack_ns
;
1516 if (dl_task(current
) || rt_task(current
))
1519 hrtimer_init_on_stack(&t
.timer
, clockid
, mode
);
1520 hrtimer_set_expires_range_ns(&t
.timer
, timespec_to_ktime(*rqtp
), slack
);
1521 if (do_nanosleep(&t
, mode
))
1524 /* Absolute timers do not update the rmtp value and restart: */
1525 if (mode
== HRTIMER_MODE_ABS
) {
1526 ret
= -ERESTARTNOHAND
;
1531 ret
= update_rmtp(&t
.timer
, rmtp
);
1536 restart
= ¤t
->restart_block
;
1537 restart
->fn
= hrtimer_nanosleep_restart
;
1538 restart
->nanosleep
.clockid
= t
.timer
.base
->clockid
;
1539 restart
->nanosleep
.rmtp
= rmtp
;
1540 restart
->nanosleep
.expires
= hrtimer_get_expires_tv64(&t
.timer
);
1542 ret
= -ERESTART_RESTARTBLOCK
;
1544 destroy_hrtimer_on_stack(&t
.timer
);
1548 SYSCALL_DEFINE2(nanosleep
, struct timespec __user
*, rqtp
,
1549 struct timespec __user
*, rmtp
)
1553 if (copy_from_user(&tu
, rqtp
, sizeof(tu
)))
1556 if (!timespec_valid(&tu
))
1559 return hrtimer_nanosleep(&tu
, rmtp
, HRTIMER_MODE_REL
, CLOCK_MONOTONIC
);
1563 * Functions related to boot-time initialization:
1565 int hrtimers_prepare_cpu(unsigned int cpu
)
1567 struct hrtimer_cpu_base
*cpu_base
= &per_cpu(hrtimer_bases
, cpu
);
1570 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++) {
1571 cpu_base
->clock_base
[i
].cpu_base
= cpu_base
;
1572 timerqueue_init_head(&cpu_base
->clock_base
[i
].active
);
1575 cpu_base
->cpu
= cpu
;
1576 hrtimer_init_hres(cpu_base
);
1580 #ifdef CONFIG_HOTPLUG_CPU
1582 static void migrate_hrtimer_list(struct hrtimer_clock_base
*old_base
,
1583 struct hrtimer_clock_base
*new_base
)
1585 struct hrtimer
*timer
;
1586 struct timerqueue_node
*node
;
1588 while ((node
= timerqueue_getnext(&old_base
->active
))) {
1589 timer
= container_of(node
, struct hrtimer
, node
);
1590 BUG_ON(hrtimer_callback_running(timer
));
1591 debug_deactivate(timer
);
1594 * Mark it as ENQUEUED not INACTIVE otherwise the
1595 * timer could be seen as !active and just vanish away
1596 * under us on another CPU
1598 __remove_hrtimer(timer
, old_base
, HRTIMER_STATE_ENQUEUED
, 0);
1599 timer
->base
= new_base
;
1601 * Enqueue the timers on the new cpu. This does not
1602 * reprogram the event device in case the timer
1603 * expires before the earliest on this CPU, but we run
1604 * hrtimer_interrupt after we migrated everything to
1605 * sort out already expired timers and reprogram the
1608 enqueue_hrtimer(timer
, new_base
);
1612 int hrtimers_dead_cpu(unsigned int scpu
)
1614 struct hrtimer_cpu_base
*old_base
, *new_base
;
1617 BUG_ON(cpu_online(scpu
));
1618 tick_cancel_sched_timer(scpu
);
1620 local_irq_disable();
1621 old_base
= &per_cpu(hrtimer_bases
, scpu
);
1622 new_base
= this_cpu_ptr(&hrtimer_bases
);
1624 * The caller is globally serialized and nobody else
1625 * takes two locks at once, deadlock is not possible.
1627 raw_spin_lock(&new_base
->lock
);
1628 raw_spin_lock_nested(&old_base
->lock
, SINGLE_DEPTH_NESTING
);
1630 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++) {
1631 migrate_hrtimer_list(&old_base
->clock_base
[i
],
1632 &new_base
->clock_base
[i
]);
1635 raw_spin_unlock(&old_base
->lock
);
1636 raw_spin_unlock(&new_base
->lock
);
1638 /* Check, if we got expired work to do */
1639 __hrtimer_peek_ahead_timers();
1644 #endif /* CONFIG_HOTPLUG_CPU */
1646 void __init
hrtimers_init(void)
1648 hrtimers_prepare_cpu(smp_processor_id());
1652 * schedule_hrtimeout_range_clock - sleep until timeout
1653 * @expires: timeout value (ktime_t)
1654 * @delta: slack in expires timeout (ktime_t)
1655 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1656 * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME
1659 schedule_hrtimeout_range_clock(ktime_t
*expires
, u64 delta
,
1660 const enum hrtimer_mode mode
, int clock
)
1662 struct hrtimer_sleeper t
;
1665 * Optimize when a zero timeout value is given. It does not
1666 * matter whether this is an absolute or a relative time.
1668 if (expires
&& *expires
== 0) {
1669 __set_current_state(TASK_RUNNING
);
1674 * A NULL parameter means "infinite"
1681 hrtimer_init_on_stack(&t
.timer
, clock
, mode
);
1682 hrtimer_set_expires_range_ns(&t
.timer
, *expires
, delta
);
1684 hrtimer_init_sleeper(&t
, current
);
1686 hrtimer_start_expires(&t
.timer
, mode
);
1691 hrtimer_cancel(&t
.timer
);
1692 destroy_hrtimer_on_stack(&t
.timer
);
1694 __set_current_state(TASK_RUNNING
);
1696 return !t
.task
? 0 : -EINTR
;
1700 * schedule_hrtimeout_range - sleep until timeout
1701 * @expires: timeout value (ktime_t)
1702 * @delta: slack in expires timeout (ktime_t)
1703 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1705 * Make the current task sleep until the given expiry time has
1706 * elapsed. The routine will return immediately unless
1707 * the current task state has been set (see set_current_state()).
1709 * The @delta argument gives the kernel the freedom to schedule the
1710 * actual wakeup to a time that is both power and performance friendly.
1711 * The kernel give the normal best effort behavior for "@expires+@delta",
1712 * but may decide to fire the timer earlier, but no earlier than @expires.
1714 * You can set the task state as follows -
1716 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1717 * pass before the routine returns unless the current task is explicitly
1718 * woken up, (e.g. by wake_up_process()).
1720 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1721 * delivered to the current task or the current task is explicitly woken
1724 * The current task state is guaranteed to be TASK_RUNNING when this
1727 * Returns 0 when the timer has expired. If the task was woken before the
1728 * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or
1729 * by an explicit wakeup, it returns -EINTR.
1731 int __sched
schedule_hrtimeout_range(ktime_t
*expires
, u64 delta
,
1732 const enum hrtimer_mode mode
)
1734 return schedule_hrtimeout_range_clock(expires
, delta
, mode
,
1737 EXPORT_SYMBOL_GPL(schedule_hrtimeout_range
);
1740 * schedule_hrtimeout - sleep until timeout
1741 * @expires: timeout value (ktime_t)
1742 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1744 * Make the current task sleep until the given expiry time has
1745 * elapsed. The routine will return immediately unless
1746 * the current task state has been set (see set_current_state()).
1748 * You can set the task state as follows -
1750 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1751 * pass before the routine returns unless the current task is explicitly
1752 * woken up, (e.g. by wake_up_process()).
1754 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1755 * delivered to the current task or the current task is explicitly woken
1758 * The current task state is guaranteed to be TASK_RUNNING when this
1761 * Returns 0 when the timer has expired. If the task was woken before the
1762 * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or
1763 * by an explicit wakeup, it returns -EINTR.
1765 int __sched
schedule_hrtimeout(ktime_t
*expires
,
1766 const enum hrtimer_mode mode
)
1768 return schedule_hrtimeout_range(expires
, 0, mode
);
1770 EXPORT_SYMBOL_GPL(schedule_hrtimeout
);