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
) =
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 [CLOCK_REALTIME
] = HRTIMER_BASE_REALTIME
,
98 [CLOCK_MONOTONIC
] = HRTIMER_BASE_MONOTONIC
,
99 [CLOCK_BOOTTIME
] = HRTIMER_BASE_BOOTTIME
,
100 [CLOCK_TAI
] = HRTIMER_BASE_TAI
,
103 static inline int hrtimer_clockid_to_base(clockid_t clock_id
)
105 return hrtimer_clock_to_base_table
[clock_id
];
109 * Functions and macros which are different for UP/SMP systems are kept in a
115 * We require the migration_base for lock_hrtimer_base()/switch_hrtimer_base()
116 * such that hrtimer_callback_running() can unconditionally dereference
117 * timer->base->cpu_base
119 static struct hrtimer_cpu_base migration_cpu_base
= {
120 .seq
= SEQCNT_ZERO(migration_cpu_base
),
121 .clock_base
= { { .cpu_base
= &migration_cpu_base
, }, },
124 #define migration_base migration_cpu_base.clock_base[0]
127 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
128 * means that all timers which are tied to this base via timer->base are
129 * locked, and the base itself is locked too.
131 * So __run_timers/migrate_timers can safely modify all timers which could
132 * be found on the lists/queues.
134 * When the timer's base is locked, and the timer removed from list, it is
135 * possible to set timer->base = &migration_base and drop the lock: the timer
139 struct hrtimer_clock_base
*lock_hrtimer_base(const struct hrtimer
*timer
,
140 unsigned long *flags
)
142 struct hrtimer_clock_base
*base
;
146 if (likely(base
!= &migration_base
)) {
147 raw_spin_lock_irqsave(&base
->cpu_base
->lock
, *flags
);
148 if (likely(base
== timer
->base
))
150 /* The timer has migrated to another CPU: */
151 raw_spin_unlock_irqrestore(&base
->cpu_base
->lock
, *flags
);
158 * With HIGHRES=y we do not migrate the timer when it is expiring
159 * before the next event on the target cpu because we cannot reprogram
160 * the target cpu hardware and we would cause it to fire late.
162 * Called with cpu_base->lock of target cpu held.
165 hrtimer_check_target(struct hrtimer
*timer
, struct hrtimer_clock_base
*new_base
)
167 #ifdef CONFIG_HIGH_RES_TIMERS
170 if (!new_base
->cpu_base
->hres_active
)
173 expires
= ktime_sub(hrtimer_get_expires(timer
), new_base
->offset
);
174 return expires
.tv64
<= new_base
->cpu_base
->expires_next
.tv64
;
180 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
182 struct hrtimer_cpu_base
*get_target_base(struct hrtimer_cpu_base
*base
,
185 if (pinned
|| !base
->migration_enabled
)
187 return &per_cpu(hrtimer_bases
, get_nohz_timer_target());
191 struct hrtimer_cpu_base
*get_target_base(struct hrtimer_cpu_base
*base
,
199 * We switch the timer base to a power-optimized selected CPU target,
201 * - NO_HZ_COMMON is enabled
202 * - timer migration is enabled
203 * - the timer callback is not running
204 * - the timer is not the first expiring timer on the new target
206 * If one of the above requirements is not fulfilled we move the timer
207 * to the current CPU or leave it on the previously assigned CPU if
208 * the timer callback is currently running.
210 static inline struct hrtimer_clock_base
*
211 switch_hrtimer_base(struct hrtimer
*timer
, struct hrtimer_clock_base
*base
,
214 struct hrtimer_cpu_base
*new_cpu_base
, *this_cpu_base
;
215 struct hrtimer_clock_base
*new_base
;
216 int basenum
= base
->index
;
218 this_cpu_base
= this_cpu_ptr(&hrtimer_bases
);
219 new_cpu_base
= get_target_base(this_cpu_base
, pinned
);
221 new_base
= &new_cpu_base
->clock_base
[basenum
];
223 if (base
!= new_base
) {
225 * We are trying to move timer to new_base.
226 * However we can't change timer's base while it is running,
227 * so we keep it on the same CPU. No hassle vs. reprogramming
228 * the event source in the high resolution case. The softirq
229 * code will take care of this when the timer function has
230 * completed. There is no conflict as we hold the lock until
231 * the timer is enqueued.
233 if (unlikely(hrtimer_callback_running(timer
)))
236 /* See the comment in lock_hrtimer_base() */
237 timer
->base
= &migration_base
;
238 raw_spin_unlock(&base
->cpu_base
->lock
);
239 raw_spin_lock(&new_base
->cpu_base
->lock
);
241 if (new_cpu_base
!= this_cpu_base
&&
242 hrtimer_check_target(timer
, new_base
)) {
243 raw_spin_unlock(&new_base
->cpu_base
->lock
);
244 raw_spin_lock(&base
->cpu_base
->lock
);
245 new_cpu_base
= this_cpu_base
;
249 timer
->base
= new_base
;
251 if (new_cpu_base
!= this_cpu_base
&&
252 hrtimer_check_target(timer
, new_base
)) {
253 new_cpu_base
= this_cpu_base
;
260 #else /* CONFIG_SMP */
262 static inline struct hrtimer_clock_base
*
263 lock_hrtimer_base(const struct hrtimer
*timer
, unsigned long *flags
)
265 struct hrtimer_clock_base
*base
= timer
->base
;
267 raw_spin_lock_irqsave(&base
->cpu_base
->lock
, *flags
);
272 # define switch_hrtimer_base(t, b, p) (b)
274 #endif /* !CONFIG_SMP */
277 * Functions for the union type storage format of ktime_t which are
278 * too large for inlining:
280 #if BITS_PER_LONG < 64
282 * Divide a ktime value by a nanosecond value
284 s64
__ktime_divns(const ktime_t kt
, s64 div
)
290 dclc
= ktime_to_ns(kt
);
291 tmp
= dclc
< 0 ? -dclc
: dclc
;
293 /* Make sure the divisor is less than 2^32: */
299 do_div(tmp
, (unsigned long) div
);
300 return dclc
< 0 ? -tmp
: tmp
;
302 EXPORT_SYMBOL_GPL(__ktime_divns
);
303 #endif /* BITS_PER_LONG >= 64 */
306 * Add two ktime values and do a safety check for overflow:
308 ktime_t
ktime_add_safe(const ktime_t lhs
, const ktime_t rhs
)
310 ktime_t res
= ktime_add(lhs
, rhs
);
313 * We use KTIME_SEC_MAX here, the maximum timeout which we can
314 * return to user space in a timespec:
316 if (res
.tv64
< 0 || res
.tv64
< lhs
.tv64
|| res
.tv64
< rhs
.tv64
)
317 res
= ktime_set(KTIME_SEC_MAX
, 0);
322 EXPORT_SYMBOL_GPL(ktime_add_safe
);
324 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
326 static struct debug_obj_descr hrtimer_debug_descr
;
328 static void *hrtimer_debug_hint(void *addr
)
330 return ((struct hrtimer
*) addr
)->function
;
334 * fixup_init is called when:
335 * - an active object is initialized
337 static int hrtimer_fixup_init(void *addr
, enum debug_obj_state state
)
339 struct hrtimer
*timer
= addr
;
342 case ODEBUG_STATE_ACTIVE
:
343 hrtimer_cancel(timer
);
344 debug_object_init(timer
, &hrtimer_debug_descr
);
352 * fixup_activate is called when:
353 * - an active object is activated
354 * - an unknown object is activated (might be a statically initialized object)
356 static int hrtimer_fixup_activate(void *addr
, enum debug_obj_state state
)
360 case ODEBUG_STATE_NOTAVAILABLE
:
364 case ODEBUG_STATE_ACTIVE
:
373 * fixup_free is called when:
374 * - an active object is freed
376 static int hrtimer_fixup_free(void *addr
, enum debug_obj_state state
)
378 struct hrtimer
*timer
= addr
;
381 case ODEBUG_STATE_ACTIVE
:
382 hrtimer_cancel(timer
);
383 debug_object_free(timer
, &hrtimer_debug_descr
);
390 static struct debug_obj_descr hrtimer_debug_descr
= {
392 .debug_hint
= hrtimer_debug_hint
,
393 .fixup_init
= hrtimer_fixup_init
,
394 .fixup_activate
= hrtimer_fixup_activate
,
395 .fixup_free
= hrtimer_fixup_free
,
398 static inline void debug_hrtimer_init(struct hrtimer
*timer
)
400 debug_object_init(timer
, &hrtimer_debug_descr
);
403 static inline void debug_hrtimer_activate(struct hrtimer
*timer
)
405 debug_object_activate(timer
, &hrtimer_debug_descr
);
408 static inline void debug_hrtimer_deactivate(struct hrtimer
*timer
)
410 debug_object_deactivate(timer
, &hrtimer_debug_descr
);
413 static inline void debug_hrtimer_free(struct hrtimer
*timer
)
415 debug_object_free(timer
, &hrtimer_debug_descr
);
418 static void __hrtimer_init(struct hrtimer
*timer
, clockid_t clock_id
,
419 enum hrtimer_mode mode
);
421 void hrtimer_init_on_stack(struct hrtimer
*timer
, clockid_t clock_id
,
422 enum hrtimer_mode mode
)
424 debug_object_init_on_stack(timer
, &hrtimer_debug_descr
);
425 __hrtimer_init(timer
, clock_id
, mode
);
427 EXPORT_SYMBOL_GPL(hrtimer_init_on_stack
);
429 void destroy_hrtimer_on_stack(struct hrtimer
*timer
)
431 debug_object_free(timer
, &hrtimer_debug_descr
);
435 static inline void debug_hrtimer_init(struct hrtimer
*timer
) { }
436 static inline void debug_hrtimer_activate(struct hrtimer
*timer
) { }
437 static inline void debug_hrtimer_deactivate(struct hrtimer
*timer
) { }
441 debug_init(struct hrtimer
*timer
, clockid_t clockid
,
442 enum hrtimer_mode mode
)
444 debug_hrtimer_init(timer
);
445 trace_hrtimer_init(timer
, clockid
, mode
);
448 static inline void debug_activate(struct hrtimer
*timer
)
450 debug_hrtimer_activate(timer
);
451 trace_hrtimer_start(timer
);
454 static inline void debug_deactivate(struct hrtimer
*timer
)
456 debug_hrtimer_deactivate(timer
);
457 trace_hrtimer_cancel(timer
);
460 #if defined(CONFIG_NO_HZ_COMMON) || defined(CONFIG_HIGH_RES_TIMERS)
461 static inline void hrtimer_update_next_timer(struct hrtimer_cpu_base
*cpu_base
,
462 struct hrtimer
*timer
)
464 #ifdef CONFIG_HIGH_RES_TIMERS
465 cpu_base
->next_timer
= timer
;
469 static ktime_t
__hrtimer_get_next_event(struct hrtimer_cpu_base
*cpu_base
)
471 struct hrtimer_clock_base
*base
= cpu_base
->clock_base
;
472 ktime_t expires
, expires_next
= { .tv64
= KTIME_MAX
};
473 unsigned int active
= cpu_base
->active_bases
;
475 hrtimer_update_next_timer(cpu_base
, NULL
);
476 for (; active
; base
++, active
>>= 1) {
477 struct timerqueue_node
*next
;
478 struct hrtimer
*timer
;
480 if (!(active
& 0x01))
483 next
= timerqueue_getnext(&base
->active
);
484 timer
= container_of(next
, struct hrtimer
, node
);
485 expires
= ktime_sub(hrtimer_get_expires(timer
), base
->offset
);
486 if (expires
.tv64
< expires_next
.tv64
) {
487 expires_next
= expires
;
488 hrtimer_update_next_timer(cpu_base
, timer
);
492 * clock_was_set() might have changed base->offset of any of
493 * the clock bases so the result might be negative. Fix it up
494 * to prevent a false positive in clockevents_program_event().
496 if (expires_next
.tv64
< 0)
497 expires_next
.tv64
= 0;
502 static inline ktime_t
hrtimer_update_base(struct hrtimer_cpu_base
*base
)
504 ktime_t
*offs_real
= &base
->clock_base
[HRTIMER_BASE_REALTIME
].offset
;
505 ktime_t
*offs_boot
= &base
->clock_base
[HRTIMER_BASE_BOOTTIME
].offset
;
506 ktime_t
*offs_tai
= &base
->clock_base
[HRTIMER_BASE_TAI
].offset
;
508 return ktime_get_update_offsets_now(&base
->clock_was_set_seq
,
509 offs_real
, offs_boot
, offs_tai
);
512 /* High resolution timer related functions */
513 #ifdef CONFIG_HIGH_RES_TIMERS
516 * High resolution timer enabled ?
518 static int hrtimer_hres_enabled __read_mostly
= 1;
519 unsigned int hrtimer_resolution __read_mostly
= LOW_RES_NSEC
;
520 EXPORT_SYMBOL_GPL(hrtimer_resolution
);
523 * Enable / Disable high resolution mode
525 static int __init
setup_hrtimer_hres(char *str
)
527 if (!strcmp(str
, "off"))
528 hrtimer_hres_enabled
= 0;
529 else if (!strcmp(str
, "on"))
530 hrtimer_hres_enabled
= 1;
536 __setup("highres=", setup_hrtimer_hres
);
539 * hrtimer_high_res_enabled - query, if the highres mode is enabled
541 static inline int hrtimer_is_hres_enabled(void)
543 return hrtimer_hres_enabled
;
547 * Is the high resolution mode active ?
549 static inline int __hrtimer_hres_active(struct hrtimer_cpu_base
*cpu_base
)
551 return cpu_base
->hres_active
;
554 static inline int hrtimer_hres_active(void)
556 return __hrtimer_hres_active(this_cpu_ptr(&hrtimer_bases
));
560 * Reprogram the event source with checking both queues for the
562 * Called with interrupts disabled and base->lock held
565 hrtimer_force_reprogram(struct hrtimer_cpu_base
*cpu_base
, int skip_equal
)
567 ktime_t expires_next
;
569 if (!cpu_base
->hres_active
)
572 expires_next
= __hrtimer_get_next_event(cpu_base
);
574 if (skip_equal
&& expires_next
.tv64
== cpu_base
->expires_next
.tv64
)
577 cpu_base
->expires_next
.tv64
= expires_next
.tv64
;
580 * If a hang was detected in the last timer interrupt then we
581 * leave the hang delay active in the hardware. We want the
582 * system to make progress. That also prevents the following
584 * T1 expires 50ms from now
585 * T2 expires 5s from now
587 * T1 is removed, so this code is called and would reprogram
588 * the hardware to 5s from now. Any hrtimer_start after that
589 * will not reprogram the hardware due to hang_detected being
590 * set. So we'd effectivly block all timers until the T2 event
593 if (cpu_base
->hang_detected
)
596 tick_program_event(cpu_base
->expires_next
, 1);
600 * When a timer is enqueued and expires earlier than the already enqueued
601 * timers, we have to check, whether it expires earlier than the timer for
602 * which the clock event device was armed.
604 * Called with interrupts disabled and base->cpu_base.lock held
606 static void hrtimer_reprogram(struct hrtimer
*timer
,
607 struct hrtimer_clock_base
*base
)
609 struct hrtimer_cpu_base
*cpu_base
= this_cpu_ptr(&hrtimer_bases
);
610 ktime_t expires
= ktime_sub(hrtimer_get_expires(timer
), base
->offset
);
612 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer
) < 0);
615 * If the timer is not on the current cpu, we cannot reprogram
616 * the other cpus clock event device.
618 if (base
->cpu_base
!= cpu_base
)
622 * If the hrtimer interrupt is running, then it will
623 * reevaluate the clock bases and reprogram the clock event
624 * device. The callbacks are always executed in hard interrupt
625 * context so we don't need an extra check for a running
628 if (cpu_base
->in_hrtirq
)
632 * CLOCK_REALTIME timer might be requested with an absolute
633 * expiry time which is less than base->offset. Set it to 0.
635 if (expires
.tv64
< 0)
638 if (expires
.tv64
>= cpu_base
->expires_next
.tv64
)
641 /* Update the pointer to the next expiring timer */
642 cpu_base
->next_timer
= timer
;
645 * If a hang was detected in the last timer interrupt then we
646 * do not schedule a timer which is earlier than the expiry
647 * which we enforced in the hang detection. We want the system
650 if (cpu_base
->hang_detected
)
654 * Program the timer hardware. We enforce the expiry for
655 * events which are already in the past.
657 cpu_base
->expires_next
= expires
;
658 tick_program_event(expires
, 1);
662 * Initialize the high resolution related parts of cpu_base
664 static inline void hrtimer_init_hres(struct hrtimer_cpu_base
*base
)
666 base
->expires_next
.tv64
= KTIME_MAX
;
667 base
->hres_active
= 0;
671 * Retrigger next event is called after clock was set
673 * Called with interrupts disabled via on_each_cpu()
675 static void retrigger_next_event(void *arg
)
677 struct hrtimer_cpu_base
*base
= this_cpu_ptr(&hrtimer_bases
);
679 if (!base
->hres_active
)
682 raw_spin_lock(&base
->lock
);
683 hrtimer_update_base(base
);
684 hrtimer_force_reprogram(base
, 0);
685 raw_spin_unlock(&base
->lock
);
689 * Switch to high resolution mode
691 static void hrtimer_switch_to_hres(void)
693 struct hrtimer_cpu_base
*base
= this_cpu_ptr(&hrtimer_bases
);
695 if (tick_init_highres()) {
696 printk(KERN_WARNING
"Could not switch to high resolution "
697 "mode on CPU %d\n", base
->cpu
);
699 base
->hres_active
= 1;
700 hrtimer_resolution
= HIGH_RES_NSEC
;
702 tick_setup_sched_timer();
703 /* "Retrigger" the interrupt to get things going */
704 retrigger_next_event(NULL
);
707 static void clock_was_set_work(struct work_struct
*work
)
712 static DECLARE_WORK(hrtimer_work
, clock_was_set_work
);
715 * Called from timekeeping and resume code to reprogramm the hrtimer
716 * interrupt device on all cpus.
718 void clock_was_set_delayed(void)
720 schedule_work(&hrtimer_work
);
725 static inline int __hrtimer_hres_active(struct hrtimer_cpu_base
*b
) { return 0; }
726 static inline int hrtimer_hres_active(void) { return 0; }
727 static inline int hrtimer_is_hres_enabled(void) { return 0; }
728 static inline void hrtimer_switch_to_hres(void) { }
730 hrtimer_force_reprogram(struct hrtimer_cpu_base
*base
, int skip_equal
) { }
731 static inline int hrtimer_reprogram(struct hrtimer
*timer
,
732 struct hrtimer_clock_base
*base
)
736 static inline void hrtimer_init_hres(struct hrtimer_cpu_base
*base
) { }
737 static inline void retrigger_next_event(void *arg
) { }
739 #endif /* CONFIG_HIGH_RES_TIMERS */
742 * Clock realtime was set
744 * Change the offset of the realtime clock vs. the monotonic
747 * We might have to reprogram the high resolution timer interrupt. On
748 * SMP we call the architecture specific code to retrigger _all_ high
749 * resolution timer interrupts. On UP we just disable interrupts and
750 * call the high resolution interrupt code.
752 void clock_was_set(void)
754 #ifdef CONFIG_HIGH_RES_TIMERS
755 /* Retrigger the CPU local events everywhere */
756 on_each_cpu(retrigger_next_event
, NULL
, 1);
758 timerfd_clock_was_set();
762 * During resume we might have to reprogram the high resolution timer
763 * interrupt on all online CPUs. However, all other CPUs will be
764 * stopped with IRQs interrupts disabled so the clock_was_set() call
767 void hrtimers_resume(void)
769 WARN_ONCE(!irqs_disabled(),
770 KERN_INFO
"hrtimers_resume() called with IRQs enabled!");
772 /* Retrigger on the local CPU */
773 retrigger_next_event(NULL
);
774 /* And schedule a retrigger for all others */
775 clock_was_set_delayed();
778 static inline void timer_stats_hrtimer_set_start_info(struct hrtimer
*timer
)
780 #ifdef CONFIG_TIMER_STATS
781 if (timer
->start_site
)
783 timer
->start_site
= __builtin_return_address(0);
784 memcpy(timer
->start_comm
, current
->comm
, TASK_COMM_LEN
);
785 timer
->start_pid
= current
->pid
;
789 static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer
*timer
)
791 #ifdef CONFIG_TIMER_STATS
792 timer
->start_site
= NULL
;
796 static inline void timer_stats_account_hrtimer(struct hrtimer
*timer
)
798 #ifdef CONFIG_TIMER_STATS
799 if (likely(!timer_stats_active
))
801 timer_stats_update_stats(timer
, timer
->start_pid
, timer
->start_site
,
802 timer
->function
, timer
->start_comm
, 0);
807 * Counterpart to lock_hrtimer_base above:
810 void unlock_hrtimer_base(const struct hrtimer
*timer
, unsigned long *flags
)
812 raw_spin_unlock_irqrestore(&timer
->base
->cpu_base
->lock
, *flags
);
816 * hrtimer_forward - forward the timer expiry
817 * @timer: hrtimer to forward
818 * @now: forward past this time
819 * @interval: the interval to forward
821 * Forward the timer expiry so it will expire in the future.
822 * Returns the number of overruns.
824 * Can be safely called from the callback function of @timer. If
825 * called from other contexts @timer must neither be enqueued nor
826 * running the callback and the caller needs to take care of
829 * Note: This only updates the timer expiry value and does not requeue
832 u64
hrtimer_forward(struct hrtimer
*timer
, ktime_t now
, ktime_t interval
)
837 delta
= ktime_sub(now
, hrtimer_get_expires(timer
));
842 if (WARN_ON(timer
->state
& HRTIMER_STATE_ENQUEUED
))
845 if (interval
.tv64
< hrtimer_resolution
)
846 interval
.tv64
= hrtimer_resolution
;
848 if (unlikely(delta
.tv64
>= interval
.tv64
)) {
849 s64 incr
= ktime_to_ns(interval
);
851 orun
= ktime_divns(delta
, incr
);
852 hrtimer_add_expires_ns(timer
, incr
* orun
);
853 if (hrtimer_get_expires_tv64(timer
) > now
.tv64
)
856 * This (and the ktime_add() below) is the
857 * correction for exact:
861 hrtimer_add_expires(timer
, interval
);
865 EXPORT_SYMBOL_GPL(hrtimer_forward
);
868 * enqueue_hrtimer - internal function to (re)start a timer
870 * The timer is inserted in expiry order. Insertion into the
871 * red black tree is O(log(n)). Must hold the base lock.
873 * Returns 1 when the new timer is the leftmost timer in the tree.
875 static int enqueue_hrtimer(struct hrtimer
*timer
,
876 struct hrtimer_clock_base
*base
)
878 debug_activate(timer
);
880 base
->cpu_base
->active_bases
|= 1 << base
->index
;
882 timer
->state
= HRTIMER_STATE_ENQUEUED
;
884 return timerqueue_add(&base
->active
, &timer
->node
);
888 * __remove_hrtimer - internal function to remove a timer
890 * Caller must hold the base lock.
892 * High resolution timer mode reprograms the clock event device when the
893 * timer is the one which expires next. The caller can disable this by setting
894 * reprogram to zero. This is useful, when the context does a reprogramming
895 * anyway (e.g. timer interrupt)
897 static void __remove_hrtimer(struct hrtimer
*timer
,
898 struct hrtimer_clock_base
*base
,
899 unsigned long newstate
, int reprogram
)
901 struct hrtimer_cpu_base
*cpu_base
= base
->cpu_base
;
902 unsigned int state
= timer
->state
;
904 timer
->state
= newstate
;
905 if (!(state
& HRTIMER_STATE_ENQUEUED
))
908 if (!timerqueue_del(&base
->active
, &timer
->node
))
909 cpu_base
->active_bases
&= ~(1 << base
->index
);
911 #ifdef CONFIG_HIGH_RES_TIMERS
913 * Note: If reprogram is false we do not update
914 * cpu_base->next_timer. This happens when we remove the first
915 * timer on a remote cpu. No harm as we never dereference
916 * cpu_base->next_timer. So the worst thing what can happen is
917 * an superflous call to hrtimer_force_reprogram() on the
918 * remote cpu later on if the same timer gets enqueued again.
920 if (reprogram
&& timer
== cpu_base
->next_timer
)
921 hrtimer_force_reprogram(cpu_base
, 1);
926 * remove hrtimer, called with base lock held
929 remove_hrtimer(struct hrtimer
*timer
, struct hrtimer_clock_base
*base
, bool restart
)
931 if (hrtimer_is_queued(timer
)) {
932 unsigned long state
= timer
->state
;
936 * Remove the timer and force reprogramming when high
937 * resolution mode is active and the timer is on the current
938 * CPU. If we remove a timer on another CPU, reprogramming is
939 * skipped. The interrupt event on this CPU is fired and
940 * reprogramming happens in the interrupt handler. This is a
941 * rare case and less expensive than a smp call.
943 debug_deactivate(timer
);
944 timer_stats_hrtimer_clear_start_info(timer
);
945 reprogram
= base
->cpu_base
== this_cpu_ptr(&hrtimer_bases
);
948 state
= HRTIMER_STATE_INACTIVE
;
950 __remove_hrtimer(timer
, base
, state
, reprogram
);
957 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
958 * @timer: the timer to be added
960 * @delta_ns: "slack" range for the timer
961 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
962 * relative (HRTIMER_MODE_REL)
964 void hrtimer_start_range_ns(struct hrtimer
*timer
, ktime_t tim
,
965 unsigned long delta_ns
, const enum hrtimer_mode mode
)
967 struct hrtimer_clock_base
*base
, *new_base
;
971 base
= lock_hrtimer_base(timer
, &flags
);
973 /* Remove an active timer from the queue: */
974 remove_hrtimer(timer
, base
, true);
976 if (mode
& HRTIMER_MODE_REL
) {
977 tim
= ktime_add_safe(tim
, base
->get_time());
979 * CONFIG_TIME_LOW_RES is a temporary way for architectures
980 * to signal that they simply return xtime in
981 * do_gettimeoffset(). In this case we want to round up by
982 * resolution when starting a relative timer, to avoid short
983 * timeouts. This will go away with the GTOD framework.
985 #ifdef CONFIG_TIME_LOW_RES
986 tim
= ktime_add_safe(tim
, ktime_set(0, hrtimer_resolution
));
990 hrtimer_set_expires_range_ns(timer
, tim
, delta_ns
);
992 /* Switch the timer base, if necessary: */
993 new_base
= switch_hrtimer_base(timer
, base
, mode
& HRTIMER_MODE_PINNED
);
995 timer_stats_hrtimer_set_start_info(timer
);
997 leftmost
= enqueue_hrtimer(timer
, new_base
);
1001 if (!hrtimer_is_hres_active(timer
)) {
1003 * Kick to reschedule the next tick to handle the new timer
1004 * on dynticks target.
1006 if (new_base
->cpu_base
->nohz_active
)
1007 wake_up_nohz_cpu(new_base
->cpu_base
->cpu
);
1009 hrtimer_reprogram(timer
, new_base
);
1012 unlock_hrtimer_base(timer
, &flags
);
1014 EXPORT_SYMBOL_GPL(hrtimer_start_range_ns
);
1017 * hrtimer_try_to_cancel - try to deactivate a timer
1018 * @timer: hrtimer to stop
1021 * 0 when the timer was not active
1022 * 1 when the timer was active
1023 * -1 when the timer is currently excuting the callback function and
1026 int hrtimer_try_to_cancel(struct hrtimer
*timer
)
1028 struct hrtimer_clock_base
*base
;
1029 unsigned long flags
;
1033 * Check lockless first. If the timer is not active (neither
1034 * enqueued nor running the callback, nothing to do here. The
1035 * base lock does not serialize against a concurrent enqueue,
1036 * so we can avoid taking it.
1038 if (!hrtimer_active(timer
))
1041 base
= lock_hrtimer_base(timer
, &flags
);
1043 if (!hrtimer_callback_running(timer
))
1044 ret
= remove_hrtimer(timer
, base
, false);
1046 unlock_hrtimer_base(timer
, &flags
);
1051 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel
);
1054 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1055 * @timer: the timer to be cancelled
1058 * 0 when the timer was not active
1059 * 1 when the timer was active
1061 int hrtimer_cancel(struct hrtimer
*timer
)
1064 int ret
= hrtimer_try_to_cancel(timer
);
1071 EXPORT_SYMBOL_GPL(hrtimer_cancel
);
1074 * hrtimer_get_remaining - get remaining time for the timer
1075 * @timer: the timer to read
1077 ktime_t
hrtimer_get_remaining(const struct hrtimer
*timer
)
1079 unsigned long flags
;
1082 lock_hrtimer_base(timer
, &flags
);
1083 rem
= hrtimer_expires_remaining(timer
);
1084 unlock_hrtimer_base(timer
, &flags
);
1088 EXPORT_SYMBOL_GPL(hrtimer_get_remaining
);
1090 #ifdef CONFIG_NO_HZ_COMMON
1092 * hrtimer_get_next_event - get the time until next expiry event
1094 * Returns the next expiry time or KTIME_MAX if no timer is pending.
1096 u64
hrtimer_get_next_event(void)
1098 struct hrtimer_cpu_base
*cpu_base
= this_cpu_ptr(&hrtimer_bases
);
1099 u64 expires
= KTIME_MAX
;
1100 unsigned long flags
;
1102 raw_spin_lock_irqsave(&cpu_base
->lock
, flags
);
1104 if (!__hrtimer_hres_active(cpu_base
))
1105 expires
= __hrtimer_get_next_event(cpu_base
).tv64
;
1107 raw_spin_unlock_irqrestore(&cpu_base
->lock
, flags
);
1113 static void __hrtimer_init(struct hrtimer
*timer
, clockid_t clock_id
,
1114 enum hrtimer_mode mode
)
1116 struct hrtimer_cpu_base
*cpu_base
;
1119 memset(timer
, 0, sizeof(struct hrtimer
));
1121 cpu_base
= raw_cpu_ptr(&hrtimer_bases
);
1123 if (clock_id
== CLOCK_REALTIME
&& mode
!= HRTIMER_MODE_ABS
)
1124 clock_id
= CLOCK_MONOTONIC
;
1126 base
= hrtimer_clockid_to_base(clock_id
);
1127 timer
->base
= &cpu_base
->clock_base
[base
];
1128 timerqueue_init(&timer
->node
);
1130 #ifdef CONFIG_TIMER_STATS
1131 timer
->start_site
= NULL
;
1132 timer
->start_pid
= -1;
1133 memset(timer
->start_comm
, 0, TASK_COMM_LEN
);
1138 * hrtimer_init - initialize a timer to the given clock
1139 * @timer: the timer to be initialized
1140 * @clock_id: the clock to be used
1141 * @mode: timer mode abs/rel
1143 void hrtimer_init(struct hrtimer
*timer
, clockid_t clock_id
,
1144 enum hrtimer_mode mode
)
1146 debug_init(timer
, clock_id
, mode
);
1147 __hrtimer_init(timer
, clock_id
, mode
);
1149 EXPORT_SYMBOL_GPL(hrtimer_init
);
1152 * A timer is active, when it is enqueued into the rbtree or the
1153 * callback function is running or it's in the state of being migrated
1156 * It is important for this function to not return a false negative.
1158 bool hrtimer_active(const struct hrtimer
*timer
)
1160 struct hrtimer_cpu_base
*cpu_base
;
1164 cpu_base
= READ_ONCE(timer
->base
->cpu_base
);
1165 seq
= raw_read_seqcount_begin(&cpu_base
->seq
);
1167 if (timer
->state
!= HRTIMER_STATE_INACTIVE
||
1168 cpu_base
->running
== timer
)
1171 } while (read_seqcount_retry(&cpu_base
->seq
, seq
) ||
1172 cpu_base
!= READ_ONCE(timer
->base
->cpu_base
));
1176 EXPORT_SYMBOL_GPL(hrtimer_active
);
1179 * The write_seqcount_barrier()s in __run_hrtimer() split the thing into 3
1180 * distinct sections:
1182 * - queued: the timer is queued
1183 * - callback: the timer is being ran
1184 * - post: the timer is inactive or (re)queued
1186 * On the read side we ensure we observe timer->state and cpu_base->running
1187 * from the same section, if anything changed while we looked at it, we retry.
1188 * This includes timer->base changing because sequence numbers alone are
1189 * insufficient for that.
1191 * The sequence numbers are required because otherwise we could still observe
1192 * a false negative if the read side got smeared over multiple consequtive
1193 * __run_hrtimer() invocations.
1196 static void __run_hrtimer(struct hrtimer_cpu_base
*cpu_base
,
1197 struct hrtimer_clock_base
*base
,
1198 struct hrtimer
*timer
, ktime_t
*now
)
1200 enum hrtimer_restart (*fn
)(struct hrtimer
*);
1203 lockdep_assert_held(&cpu_base
->lock
);
1205 debug_deactivate(timer
);
1206 cpu_base
->running
= timer
;
1209 * Separate the ->running assignment from the ->state assignment.
1211 * As with a regular write barrier, this ensures the read side in
1212 * hrtimer_active() cannot observe cpu_base->running == NULL &&
1213 * timer->state == INACTIVE.
1215 raw_write_seqcount_barrier(&cpu_base
->seq
);
1217 __remove_hrtimer(timer
, base
, HRTIMER_STATE_INACTIVE
, 0);
1218 timer_stats_account_hrtimer(timer
);
1219 fn
= timer
->function
;
1222 * Because we run timers from hardirq context, there is no chance
1223 * they get migrated to another cpu, therefore its safe to unlock
1226 raw_spin_unlock(&cpu_base
->lock
);
1227 trace_hrtimer_expire_entry(timer
, now
);
1228 restart
= fn(timer
);
1229 trace_hrtimer_expire_exit(timer
);
1230 raw_spin_lock(&cpu_base
->lock
);
1233 * Note: We clear the running state after enqueue_hrtimer and
1234 * we do not reprogramm the event hardware. Happens either in
1235 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1237 * Note: Because we dropped the cpu_base->lock above,
1238 * hrtimer_start_range_ns() can have popped in and enqueued the timer
1241 if (restart
!= HRTIMER_NORESTART
&&
1242 !(timer
->state
& HRTIMER_STATE_ENQUEUED
))
1243 enqueue_hrtimer(timer
, base
);
1246 * Separate the ->running assignment from the ->state assignment.
1248 * As with a regular write barrier, this ensures the read side in
1249 * hrtimer_active() cannot observe cpu_base->running == NULL &&
1250 * timer->state == INACTIVE.
1252 raw_write_seqcount_barrier(&cpu_base
->seq
);
1254 WARN_ON_ONCE(cpu_base
->running
!= timer
);
1255 cpu_base
->running
= NULL
;
1258 static void __hrtimer_run_queues(struct hrtimer_cpu_base
*cpu_base
, ktime_t now
)
1260 struct hrtimer_clock_base
*base
= cpu_base
->clock_base
;
1261 unsigned int active
= cpu_base
->active_bases
;
1263 for (; active
; base
++, active
>>= 1) {
1264 struct timerqueue_node
*node
;
1267 if (!(active
& 0x01))
1270 basenow
= ktime_add(now
, base
->offset
);
1272 while ((node
= timerqueue_getnext(&base
->active
))) {
1273 struct hrtimer
*timer
;
1275 timer
= container_of(node
, struct hrtimer
, node
);
1278 * The immediate goal for using the softexpires is
1279 * minimizing wakeups, not running timers at the
1280 * earliest interrupt after their soft expiration.
1281 * This allows us to avoid using a Priority Search
1282 * Tree, which can answer a stabbing querry for
1283 * overlapping intervals and instead use the simple
1284 * BST we already have.
1285 * We don't add extra wakeups by delaying timers that
1286 * are right-of a not yet expired timer, because that
1287 * timer will have to trigger a wakeup anyway.
1289 if (basenow
.tv64
< hrtimer_get_softexpires_tv64(timer
))
1292 __run_hrtimer(cpu_base
, base
, timer
, &basenow
);
1297 #ifdef CONFIG_HIGH_RES_TIMERS
1300 * High resolution timer interrupt
1301 * Called with interrupts disabled
1303 void hrtimer_interrupt(struct clock_event_device
*dev
)
1305 struct hrtimer_cpu_base
*cpu_base
= this_cpu_ptr(&hrtimer_bases
);
1306 ktime_t expires_next
, now
, entry_time
, delta
;
1309 BUG_ON(!cpu_base
->hres_active
);
1310 cpu_base
->nr_events
++;
1311 dev
->next_event
.tv64
= KTIME_MAX
;
1313 raw_spin_lock(&cpu_base
->lock
);
1314 entry_time
= now
= hrtimer_update_base(cpu_base
);
1316 cpu_base
->in_hrtirq
= 1;
1318 * We set expires_next to KTIME_MAX here with cpu_base->lock
1319 * held to prevent that a timer is enqueued in our queue via
1320 * the migration code. This does not affect enqueueing of
1321 * timers which run their callback and need to be requeued on
1324 cpu_base
->expires_next
.tv64
= KTIME_MAX
;
1326 __hrtimer_run_queues(cpu_base
, now
);
1328 /* Reevaluate the clock bases for the next expiry */
1329 expires_next
= __hrtimer_get_next_event(cpu_base
);
1331 * Store the new expiry value so the migration code can verify
1334 cpu_base
->expires_next
= expires_next
;
1335 cpu_base
->in_hrtirq
= 0;
1336 raw_spin_unlock(&cpu_base
->lock
);
1338 /* Reprogramming necessary ? */
1339 if (!tick_program_event(expires_next
, 0)) {
1340 cpu_base
->hang_detected
= 0;
1345 * The next timer was already expired due to:
1347 * - long lasting callbacks
1348 * - being scheduled away when running in a VM
1350 * We need to prevent that we loop forever in the hrtimer
1351 * interrupt routine. We give it 3 attempts to avoid
1352 * overreacting on some spurious event.
1354 * Acquire base lock for updating the offsets and retrieving
1357 raw_spin_lock(&cpu_base
->lock
);
1358 now
= hrtimer_update_base(cpu_base
);
1359 cpu_base
->nr_retries
++;
1363 * Give the system a chance to do something else than looping
1364 * here. We stored the entry time, so we know exactly how long
1365 * we spent here. We schedule the next event this amount of
1368 cpu_base
->nr_hangs
++;
1369 cpu_base
->hang_detected
= 1;
1370 raw_spin_unlock(&cpu_base
->lock
);
1371 delta
= ktime_sub(now
, entry_time
);
1372 if ((unsigned int)delta
.tv64
> cpu_base
->max_hang_time
)
1373 cpu_base
->max_hang_time
= (unsigned int) delta
.tv64
;
1375 * Limit it to a sensible value as we enforce a longer
1376 * delay. Give the CPU at least 100ms to catch up.
1378 if (delta
.tv64
> 100 * NSEC_PER_MSEC
)
1379 expires_next
= ktime_add_ns(now
, 100 * NSEC_PER_MSEC
);
1381 expires_next
= ktime_add(now
, delta
);
1382 tick_program_event(expires_next
, 1);
1383 printk_once(KERN_WARNING
"hrtimer: interrupt took %llu ns\n",
1384 ktime_to_ns(delta
));
1388 * local version of hrtimer_peek_ahead_timers() called with interrupts
1391 static inline void __hrtimer_peek_ahead_timers(void)
1393 struct tick_device
*td
;
1395 if (!hrtimer_hres_active())
1398 td
= this_cpu_ptr(&tick_cpu_device
);
1399 if (td
&& td
->evtdev
)
1400 hrtimer_interrupt(td
->evtdev
);
1403 #else /* CONFIG_HIGH_RES_TIMERS */
1405 static inline void __hrtimer_peek_ahead_timers(void) { }
1407 #endif /* !CONFIG_HIGH_RES_TIMERS */
1410 * Called from run_local_timers in hardirq context every jiffy
1412 void hrtimer_run_queues(void)
1414 struct hrtimer_cpu_base
*cpu_base
= this_cpu_ptr(&hrtimer_bases
);
1417 if (__hrtimer_hres_active(cpu_base
))
1421 * This _is_ ugly: We have to check periodically, whether we
1422 * can switch to highres and / or nohz mode. The clocksource
1423 * switch happens with xtime_lock held. Notification from
1424 * there only sets the check bit in the tick_oneshot code,
1425 * otherwise we might deadlock vs. xtime_lock.
1427 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) {
1428 hrtimer_switch_to_hres();
1432 raw_spin_lock(&cpu_base
->lock
);
1433 now
= hrtimer_update_base(cpu_base
);
1434 __hrtimer_run_queues(cpu_base
, now
);
1435 raw_spin_unlock(&cpu_base
->lock
);
1439 * Sleep related functions:
1441 static enum hrtimer_restart
hrtimer_wakeup(struct hrtimer
*timer
)
1443 struct hrtimer_sleeper
*t
=
1444 container_of(timer
, struct hrtimer_sleeper
, timer
);
1445 struct task_struct
*task
= t
->task
;
1449 wake_up_process(task
);
1451 return HRTIMER_NORESTART
;
1454 void hrtimer_init_sleeper(struct hrtimer_sleeper
*sl
, struct task_struct
*task
)
1456 sl
->timer
.function
= hrtimer_wakeup
;
1459 EXPORT_SYMBOL_GPL(hrtimer_init_sleeper
);
1461 static int __sched
do_nanosleep(struct hrtimer_sleeper
*t
, enum hrtimer_mode mode
)
1463 hrtimer_init_sleeper(t
, current
);
1466 set_current_state(TASK_INTERRUPTIBLE
);
1467 hrtimer_start_expires(&t
->timer
, mode
);
1469 if (likely(t
->task
))
1470 freezable_schedule();
1472 hrtimer_cancel(&t
->timer
);
1473 mode
= HRTIMER_MODE_ABS
;
1475 } while (t
->task
&& !signal_pending(current
));
1477 __set_current_state(TASK_RUNNING
);
1479 return t
->task
== NULL
;
1482 static int update_rmtp(struct hrtimer
*timer
, struct timespec __user
*rmtp
)
1484 struct timespec rmt
;
1487 rem
= hrtimer_expires_remaining(timer
);
1490 rmt
= ktime_to_timespec(rem
);
1492 if (copy_to_user(rmtp
, &rmt
, sizeof(*rmtp
)))
1498 long __sched
hrtimer_nanosleep_restart(struct restart_block
*restart
)
1500 struct hrtimer_sleeper t
;
1501 struct timespec __user
*rmtp
;
1504 hrtimer_init_on_stack(&t
.timer
, restart
->nanosleep
.clockid
,
1506 hrtimer_set_expires_tv64(&t
.timer
, restart
->nanosleep
.expires
);
1508 if (do_nanosleep(&t
, HRTIMER_MODE_ABS
))
1511 rmtp
= restart
->nanosleep
.rmtp
;
1513 ret
= update_rmtp(&t
.timer
, rmtp
);
1518 /* The other values in restart are already filled in */
1519 ret
= -ERESTART_RESTARTBLOCK
;
1521 destroy_hrtimer_on_stack(&t
.timer
);
1525 long hrtimer_nanosleep(struct timespec
*rqtp
, struct timespec __user
*rmtp
,
1526 const enum hrtimer_mode mode
, const clockid_t clockid
)
1528 struct restart_block
*restart
;
1529 struct hrtimer_sleeper t
;
1531 unsigned long slack
;
1533 slack
= current
->timer_slack_ns
;
1534 if (dl_task(current
) || rt_task(current
))
1537 hrtimer_init_on_stack(&t
.timer
, clockid
, mode
);
1538 hrtimer_set_expires_range_ns(&t
.timer
, timespec_to_ktime(*rqtp
), slack
);
1539 if (do_nanosleep(&t
, mode
))
1542 /* Absolute timers do not update the rmtp value and restart: */
1543 if (mode
== HRTIMER_MODE_ABS
) {
1544 ret
= -ERESTARTNOHAND
;
1549 ret
= update_rmtp(&t
.timer
, rmtp
);
1554 restart
= ¤t
->restart_block
;
1555 restart
->fn
= hrtimer_nanosleep_restart
;
1556 restart
->nanosleep
.clockid
= t
.timer
.base
->clockid
;
1557 restart
->nanosleep
.rmtp
= rmtp
;
1558 restart
->nanosleep
.expires
= hrtimer_get_expires_tv64(&t
.timer
);
1560 ret
= -ERESTART_RESTARTBLOCK
;
1562 destroy_hrtimer_on_stack(&t
.timer
);
1566 SYSCALL_DEFINE2(nanosleep
, struct timespec __user
*, rqtp
,
1567 struct timespec __user
*, rmtp
)
1571 if (copy_from_user(&tu
, rqtp
, sizeof(tu
)))
1574 if (!timespec_valid(&tu
))
1577 return hrtimer_nanosleep(&tu
, rmtp
, HRTIMER_MODE_REL
, CLOCK_MONOTONIC
);
1581 * Functions related to boot-time initialization:
1583 static void init_hrtimers_cpu(int cpu
)
1585 struct hrtimer_cpu_base
*cpu_base
= &per_cpu(hrtimer_bases
, cpu
);
1588 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++) {
1589 cpu_base
->clock_base
[i
].cpu_base
= cpu_base
;
1590 timerqueue_init_head(&cpu_base
->clock_base
[i
].active
);
1593 cpu_base
->cpu
= cpu
;
1594 hrtimer_init_hres(cpu_base
);
1597 #ifdef CONFIG_HOTPLUG_CPU
1599 static void migrate_hrtimer_list(struct hrtimer_clock_base
*old_base
,
1600 struct hrtimer_clock_base
*new_base
)
1602 struct hrtimer
*timer
;
1603 struct timerqueue_node
*node
;
1605 while ((node
= timerqueue_getnext(&old_base
->active
))) {
1606 timer
= container_of(node
, struct hrtimer
, node
);
1607 BUG_ON(hrtimer_callback_running(timer
));
1608 debug_deactivate(timer
);
1611 * Mark it as ENQUEUED not INACTIVE otherwise the
1612 * timer could be seen as !active and just vanish away
1613 * under us on another CPU
1615 __remove_hrtimer(timer
, old_base
, HRTIMER_STATE_ENQUEUED
, 0);
1616 timer
->base
= new_base
;
1618 * Enqueue the timers on the new cpu. This does not
1619 * reprogram the event device in case the timer
1620 * expires before the earliest on this CPU, but we run
1621 * hrtimer_interrupt after we migrated everything to
1622 * sort out already expired timers and reprogram the
1625 enqueue_hrtimer(timer
, new_base
);
1629 static void migrate_hrtimers(int scpu
)
1631 struct hrtimer_cpu_base
*old_base
, *new_base
;
1634 BUG_ON(cpu_online(scpu
));
1635 tick_cancel_sched_timer(scpu
);
1637 local_irq_disable();
1638 old_base
= &per_cpu(hrtimer_bases
, scpu
);
1639 new_base
= this_cpu_ptr(&hrtimer_bases
);
1641 * The caller is globally serialized and nobody else
1642 * takes two locks at once, deadlock is not possible.
1644 raw_spin_lock(&new_base
->lock
);
1645 raw_spin_lock_nested(&old_base
->lock
, SINGLE_DEPTH_NESTING
);
1647 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++) {
1648 migrate_hrtimer_list(&old_base
->clock_base
[i
],
1649 &new_base
->clock_base
[i
]);
1652 raw_spin_unlock(&old_base
->lock
);
1653 raw_spin_unlock(&new_base
->lock
);
1655 /* Check, if we got expired work to do */
1656 __hrtimer_peek_ahead_timers();
1660 #endif /* CONFIG_HOTPLUG_CPU */
1662 static int hrtimer_cpu_notify(struct notifier_block
*self
,
1663 unsigned long action
, void *hcpu
)
1665 int scpu
= (long)hcpu
;
1669 case CPU_UP_PREPARE
:
1670 case CPU_UP_PREPARE_FROZEN
:
1671 init_hrtimers_cpu(scpu
);
1674 #ifdef CONFIG_HOTPLUG_CPU
1676 case CPU_DEAD_FROZEN
:
1677 migrate_hrtimers(scpu
);
1688 static struct notifier_block hrtimers_nb
= {
1689 .notifier_call
= hrtimer_cpu_notify
,
1692 void __init
hrtimers_init(void)
1694 hrtimer_cpu_notify(&hrtimers_nb
, (unsigned long)CPU_UP_PREPARE
,
1695 (void *)(long)smp_processor_id());
1696 register_cpu_notifier(&hrtimers_nb
);
1700 * schedule_hrtimeout_range_clock - 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
1704 * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME
1707 schedule_hrtimeout_range_clock(ktime_t
*expires
, unsigned long delta
,
1708 const enum hrtimer_mode mode
, int clock
)
1710 struct hrtimer_sleeper t
;
1713 * Optimize when a zero timeout value is given. It does not
1714 * matter whether this is an absolute or a relative time.
1716 if (expires
&& !expires
->tv64
) {
1717 __set_current_state(TASK_RUNNING
);
1722 * A NULL parameter means "infinite"
1729 hrtimer_init_on_stack(&t
.timer
, clock
, mode
);
1730 hrtimer_set_expires_range_ns(&t
.timer
, *expires
, delta
);
1732 hrtimer_init_sleeper(&t
, current
);
1734 hrtimer_start_expires(&t
.timer
, mode
);
1739 hrtimer_cancel(&t
.timer
);
1740 destroy_hrtimer_on_stack(&t
.timer
);
1742 __set_current_state(TASK_RUNNING
);
1744 return !t
.task
? 0 : -EINTR
;
1748 * schedule_hrtimeout_range - sleep until timeout
1749 * @expires: timeout value (ktime_t)
1750 * @delta: slack in expires timeout (ktime_t)
1751 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1753 * Make the current task sleep until the given expiry time has
1754 * elapsed. The routine will return immediately unless
1755 * the current task state has been set (see set_current_state()).
1757 * The @delta argument gives the kernel the freedom to schedule the
1758 * actual wakeup to a time that is both power and performance friendly.
1759 * The kernel give the normal best effort behavior for "@expires+@delta",
1760 * but may decide to fire the timer earlier, but no earlier than @expires.
1762 * You can set the task state as follows -
1764 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1765 * pass before the routine returns.
1767 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1768 * delivered to the current task.
1770 * The current task state is guaranteed to be TASK_RUNNING when this
1773 * Returns 0 when the timer has expired otherwise -EINTR
1775 int __sched
schedule_hrtimeout_range(ktime_t
*expires
, unsigned long delta
,
1776 const enum hrtimer_mode mode
)
1778 return schedule_hrtimeout_range_clock(expires
, delta
, mode
,
1781 EXPORT_SYMBOL_GPL(schedule_hrtimeout_range
);
1784 * schedule_hrtimeout - sleep until timeout
1785 * @expires: timeout value (ktime_t)
1786 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1788 * Make the current task sleep until the given expiry time has
1789 * elapsed. The routine will return immediately unless
1790 * the current task state has been set (see set_current_state()).
1792 * You can set the task state as follows -
1794 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1795 * pass before the routine returns.
1797 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1798 * delivered to the current task.
1800 * The current task state is guaranteed to be TASK_RUNNING when this
1803 * Returns 0 when the timer has expired otherwise -EINTR
1805 int __sched
schedule_hrtimeout(ktime_t
*expires
,
1806 const enum hrtimer_mode mode
)
1808 return schedule_hrtimeout_range(expires
, 0, mode
);
1810 EXPORT_SYMBOL_GPL(schedule_hrtimeout
);