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/signal.h>
47 #include <linux/sched/sysctl.h>
48 #include <linux/sched/rt.h>
49 #include <linux/sched/deadline.h>
50 #include <linux/sched/nohz.h>
51 #include <linux/sched/debug.h>
52 #include <linux/timer.h>
53 #include <linux/freezer.h>
54 #include <linux/compat.h>
56 #include <linux/uaccess.h>
58 #include <trace/events/timer.h>
60 #include "tick-internal.h"
65 * There are more clockids than hrtimer bases. Thus, we index
66 * into the timer bases by the hrtimer_base_type enum. When trying
67 * to reach a base using a clockid, hrtimer_clockid_to_base()
68 * is used to convert from clockid to the proper hrtimer_base_type.
70 DEFINE_PER_CPU(struct hrtimer_cpu_base
, hrtimer_bases
) =
72 .lock
= __RAW_SPIN_LOCK_UNLOCKED(hrtimer_bases
.lock
),
73 .seq
= SEQCNT_ZERO(hrtimer_bases
.seq
),
77 .index
= HRTIMER_BASE_MONOTONIC
,
78 .clockid
= CLOCK_MONOTONIC
,
79 .get_time
= &ktime_get
,
82 .index
= HRTIMER_BASE_REALTIME
,
83 .clockid
= CLOCK_REALTIME
,
84 .get_time
= &ktime_get_real
,
87 .index
= HRTIMER_BASE_BOOTTIME
,
88 .clockid
= CLOCK_BOOTTIME
,
89 .get_time
= &ktime_get_boottime
,
92 .index
= HRTIMER_BASE_TAI
,
94 .get_time
= &ktime_get_clocktai
,
99 static const int hrtimer_clock_to_base_table
[MAX_CLOCKS
] = {
100 /* Make sure we catch unsupported clockids */
101 [0 ... MAX_CLOCKS
- 1] = HRTIMER_MAX_CLOCK_BASES
,
103 [CLOCK_REALTIME
] = HRTIMER_BASE_REALTIME
,
104 [CLOCK_MONOTONIC
] = HRTIMER_BASE_MONOTONIC
,
105 [CLOCK_BOOTTIME
] = HRTIMER_BASE_BOOTTIME
,
106 [CLOCK_TAI
] = HRTIMER_BASE_TAI
,
110 * Functions and macros which are different for UP/SMP systems are kept in a
116 * We require the migration_base for lock_hrtimer_base()/switch_hrtimer_base()
117 * such that hrtimer_callback_running() can unconditionally dereference
118 * timer->base->cpu_base
120 static struct hrtimer_cpu_base migration_cpu_base
= {
121 .seq
= SEQCNT_ZERO(migration_cpu_base
),
122 .clock_base
= { { .cpu_base
= &migration_cpu_base
, }, },
125 #define migration_base migration_cpu_base.clock_base[0]
128 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
129 * means that all timers which are tied to this base via timer->base are
130 * locked, and the base itself is locked too.
132 * So __run_timers/migrate_timers can safely modify all timers which could
133 * be found on the lists/queues.
135 * When the timer's base is locked, and the timer removed from list, it is
136 * possible to set timer->base = &migration_base and drop the lock: the timer
140 struct hrtimer_clock_base
*lock_hrtimer_base(const struct hrtimer
*timer
,
141 unsigned long *flags
)
143 struct hrtimer_clock_base
*base
;
147 if (likely(base
!= &migration_base
)) {
148 raw_spin_lock_irqsave(&base
->cpu_base
->lock
, *flags
);
149 if (likely(base
== timer
->base
))
151 /* The timer has migrated to another CPU: */
152 raw_spin_unlock_irqrestore(&base
->cpu_base
->lock
, *flags
);
159 * With HIGHRES=y we do not migrate the timer when it is expiring
160 * before the next event on the target cpu because we cannot reprogram
161 * the target cpu hardware and we would cause it to fire late.
163 * Called with cpu_base->lock of target cpu held.
166 hrtimer_check_target(struct hrtimer
*timer
, struct hrtimer_clock_base
*new_base
)
168 #ifdef CONFIG_HIGH_RES_TIMERS
171 if (!new_base
->cpu_base
->hres_active
)
174 expires
= ktime_sub(hrtimer_get_expires(timer
), new_base
->offset
);
175 return expires
<= new_base
->cpu_base
->expires_next
;
181 #ifdef CONFIG_NO_HZ_COMMON
183 struct hrtimer_cpu_base
*get_target_base(struct hrtimer_cpu_base
*base
,
186 if (pinned
|| !base
->migration_enabled
)
188 return &per_cpu(hrtimer_bases
, get_nohz_timer_target());
192 struct hrtimer_cpu_base
*get_target_base(struct hrtimer_cpu_base
*base
,
200 * We switch the timer base to a power-optimized selected CPU target,
202 * - NO_HZ_COMMON is enabled
203 * - timer migration is enabled
204 * - the timer callback is not running
205 * - the timer is not the first expiring timer on the new target
207 * If one of the above requirements is not fulfilled we move the timer
208 * to the current CPU or leave it on the previously assigned CPU if
209 * the timer callback is currently running.
211 static inline struct hrtimer_clock_base
*
212 switch_hrtimer_base(struct hrtimer
*timer
, struct hrtimer_clock_base
*base
,
215 struct hrtimer_cpu_base
*new_cpu_base
, *this_cpu_base
;
216 struct hrtimer_clock_base
*new_base
;
217 int basenum
= base
->index
;
219 this_cpu_base
= this_cpu_ptr(&hrtimer_bases
);
220 new_cpu_base
= get_target_base(this_cpu_base
, pinned
);
222 new_base
= &new_cpu_base
->clock_base
[basenum
];
224 if (base
!= new_base
) {
226 * We are trying to move timer to new_base.
227 * However we can't change timer's base while it is running,
228 * so we keep it on the same CPU. No hassle vs. reprogramming
229 * the event source in the high resolution case. The softirq
230 * code will take care of this when the timer function has
231 * completed. There is no conflict as we hold the lock until
232 * the timer is enqueued.
234 if (unlikely(hrtimer_callback_running(timer
)))
237 /* See the comment in lock_hrtimer_base() */
238 timer
->base
= &migration_base
;
239 raw_spin_unlock(&base
->cpu_base
->lock
);
240 raw_spin_lock(&new_base
->cpu_base
->lock
);
242 if (new_cpu_base
!= this_cpu_base
&&
243 hrtimer_check_target(timer
, new_base
)) {
244 raw_spin_unlock(&new_base
->cpu_base
->lock
);
245 raw_spin_lock(&base
->cpu_base
->lock
);
246 new_cpu_base
= this_cpu_base
;
250 timer
->base
= new_base
;
252 if (new_cpu_base
!= this_cpu_base
&&
253 hrtimer_check_target(timer
, new_base
)) {
254 new_cpu_base
= this_cpu_base
;
261 #else /* CONFIG_SMP */
263 static inline struct hrtimer_clock_base
*
264 lock_hrtimer_base(const struct hrtimer
*timer
, unsigned long *flags
)
266 struct hrtimer_clock_base
*base
= timer
->base
;
268 raw_spin_lock_irqsave(&base
->cpu_base
->lock
, *flags
);
273 # define switch_hrtimer_base(t, b, p) (b)
275 #endif /* !CONFIG_SMP */
278 * Functions for the union type storage format of ktime_t which are
279 * too large for inlining:
281 #if BITS_PER_LONG < 64
283 * Divide a ktime value by a nanosecond value
285 s64
__ktime_divns(const ktime_t kt
, s64 div
)
291 dclc
= ktime_to_ns(kt
);
292 tmp
= dclc
< 0 ? -dclc
: dclc
;
294 /* Make sure the divisor is less than 2^32: */
300 do_div(tmp
, (unsigned long) div
);
301 return dclc
< 0 ? -tmp
: tmp
;
303 EXPORT_SYMBOL_GPL(__ktime_divns
);
304 #endif /* BITS_PER_LONG >= 64 */
307 * Add two ktime values and do a safety check for overflow:
309 ktime_t
ktime_add_safe(const ktime_t lhs
, const ktime_t rhs
)
311 ktime_t res
= ktime_add_unsafe(lhs
, rhs
);
314 * We use KTIME_SEC_MAX here, the maximum timeout which we can
315 * return to user space in a timespec:
317 if (res
< 0 || res
< lhs
|| res
< rhs
)
318 res
= ktime_set(KTIME_SEC_MAX
, 0);
323 EXPORT_SYMBOL_GPL(ktime_add_safe
);
325 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
327 static struct debug_obj_descr hrtimer_debug_descr
;
329 static void *hrtimer_debug_hint(void *addr
)
331 return ((struct hrtimer
*) addr
)->function
;
335 * fixup_init is called when:
336 * - an active object is initialized
338 static bool hrtimer_fixup_init(void *addr
, enum debug_obj_state state
)
340 struct hrtimer
*timer
= addr
;
343 case ODEBUG_STATE_ACTIVE
:
344 hrtimer_cancel(timer
);
345 debug_object_init(timer
, &hrtimer_debug_descr
);
353 * fixup_activate is called when:
354 * - an active object is activated
355 * - an unknown non-static object is activated
357 static bool hrtimer_fixup_activate(void *addr
, enum debug_obj_state state
)
360 case ODEBUG_STATE_ACTIVE
:
369 * fixup_free is called when:
370 * - an active object is freed
372 static bool hrtimer_fixup_free(void *addr
, enum debug_obj_state state
)
374 struct hrtimer
*timer
= addr
;
377 case ODEBUG_STATE_ACTIVE
:
378 hrtimer_cancel(timer
);
379 debug_object_free(timer
, &hrtimer_debug_descr
);
386 static struct debug_obj_descr hrtimer_debug_descr
= {
388 .debug_hint
= hrtimer_debug_hint
,
389 .fixup_init
= hrtimer_fixup_init
,
390 .fixup_activate
= hrtimer_fixup_activate
,
391 .fixup_free
= hrtimer_fixup_free
,
394 static inline void debug_hrtimer_init(struct hrtimer
*timer
)
396 debug_object_init(timer
, &hrtimer_debug_descr
);
399 static inline void debug_hrtimer_activate(struct hrtimer
*timer
)
401 debug_object_activate(timer
, &hrtimer_debug_descr
);
404 static inline void debug_hrtimer_deactivate(struct hrtimer
*timer
)
406 debug_object_deactivate(timer
, &hrtimer_debug_descr
);
409 static inline void debug_hrtimer_free(struct hrtimer
*timer
)
411 debug_object_free(timer
, &hrtimer_debug_descr
);
414 static void __hrtimer_init(struct hrtimer
*timer
, clockid_t clock_id
,
415 enum hrtimer_mode mode
);
417 void hrtimer_init_on_stack(struct hrtimer
*timer
, clockid_t clock_id
,
418 enum hrtimer_mode mode
)
420 debug_object_init_on_stack(timer
, &hrtimer_debug_descr
);
421 __hrtimer_init(timer
, clock_id
, mode
);
423 EXPORT_SYMBOL_GPL(hrtimer_init_on_stack
);
425 void destroy_hrtimer_on_stack(struct hrtimer
*timer
)
427 debug_object_free(timer
, &hrtimer_debug_descr
);
429 EXPORT_SYMBOL_GPL(destroy_hrtimer_on_stack
);
432 static inline void debug_hrtimer_init(struct hrtimer
*timer
) { }
433 static inline void debug_hrtimer_activate(struct hrtimer
*timer
) { }
434 static inline void debug_hrtimer_deactivate(struct hrtimer
*timer
) { }
438 debug_init(struct hrtimer
*timer
, clockid_t clockid
,
439 enum hrtimer_mode mode
)
441 debug_hrtimer_init(timer
);
442 trace_hrtimer_init(timer
, clockid
, mode
);
445 static inline void debug_activate(struct hrtimer
*timer
)
447 debug_hrtimer_activate(timer
);
448 trace_hrtimer_start(timer
);
451 static inline void debug_deactivate(struct hrtimer
*timer
)
453 debug_hrtimer_deactivate(timer
);
454 trace_hrtimer_cancel(timer
);
457 #if defined(CONFIG_NO_HZ_COMMON) || defined(CONFIG_HIGH_RES_TIMERS)
458 static inline void hrtimer_update_next_timer(struct hrtimer_cpu_base
*cpu_base
,
459 struct hrtimer
*timer
)
461 #ifdef CONFIG_HIGH_RES_TIMERS
462 cpu_base
->next_timer
= timer
;
466 static ktime_t
__hrtimer_get_next_event(struct hrtimer_cpu_base
*cpu_base
)
468 struct hrtimer_clock_base
*base
= cpu_base
->clock_base
;
469 unsigned int active
= cpu_base
->active_bases
;
470 ktime_t expires
, expires_next
= KTIME_MAX
;
472 hrtimer_update_next_timer(cpu_base
, NULL
);
473 for (; active
; base
++, active
>>= 1) {
474 struct timerqueue_node
*next
;
475 struct hrtimer
*timer
;
477 if (!(active
& 0x01))
480 next
= timerqueue_getnext(&base
->active
);
481 timer
= container_of(next
, struct hrtimer
, node
);
482 expires
= ktime_sub(hrtimer_get_expires(timer
), base
->offset
);
483 if (expires
< expires_next
) {
484 expires_next
= expires
;
485 hrtimer_update_next_timer(cpu_base
, timer
);
489 * clock_was_set() might have changed base->offset of any of
490 * the clock bases so the result might be negative. Fix it up
491 * to prevent a false positive in clockevents_program_event().
493 if (expires_next
< 0)
499 static inline ktime_t
hrtimer_update_base(struct hrtimer_cpu_base
*base
)
501 ktime_t
*offs_real
= &base
->clock_base
[HRTIMER_BASE_REALTIME
].offset
;
502 ktime_t
*offs_boot
= &base
->clock_base
[HRTIMER_BASE_BOOTTIME
].offset
;
503 ktime_t
*offs_tai
= &base
->clock_base
[HRTIMER_BASE_TAI
].offset
;
505 return ktime_get_update_offsets_now(&base
->clock_was_set_seq
,
506 offs_real
, offs_boot
, offs_tai
);
509 /* High resolution timer related functions */
510 #ifdef CONFIG_HIGH_RES_TIMERS
513 * High resolution timer enabled ?
515 static bool hrtimer_hres_enabled __read_mostly
= true;
516 unsigned int hrtimer_resolution __read_mostly
= LOW_RES_NSEC
;
517 EXPORT_SYMBOL_GPL(hrtimer_resolution
);
520 * Enable / Disable high resolution mode
522 static int __init
setup_hrtimer_hres(char *str
)
524 return (kstrtobool(str
, &hrtimer_hres_enabled
) == 0);
527 __setup("highres=", setup_hrtimer_hres
);
530 * hrtimer_high_res_enabled - query, if the highres mode is enabled
532 static inline int hrtimer_is_hres_enabled(void)
534 return hrtimer_hres_enabled
;
538 * Is the high resolution mode active ?
540 static inline int __hrtimer_hres_active(struct hrtimer_cpu_base
*cpu_base
)
542 return cpu_base
->hres_active
;
545 static inline int hrtimer_hres_active(void)
547 return __hrtimer_hres_active(this_cpu_ptr(&hrtimer_bases
));
551 * Reprogram the event source with checking both queues for the
553 * Called with interrupts disabled and base->lock held
556 hrtimer_force_reprogram(struct hrtimer_cpu_base
*cpu_base
, int skip_equal
)
558 ktime_t expires_next
;
560 if (!cpu_base
->hres_active
)
563 expires_next
= __hrtimer_get_next_event(cpu_base
);
565 if (skip_equal
&& expires_next
== cpu_base
->expires_next
)
568 cpu_base
->expires_next
= expires_next
;
571 * If a hang was detected in the last timer interrupt then we
572 * leave the hang delay active in the hardware. We want the
573 * system to make progress. That also prevents the following
575 * T1 expires 50ms from now
576 * T2 expires 5s from now
578 * T1 is removed, so this code is called and would reprogram
579 * the hardware to 5s from now. Any hrtimer_start after that
580 * will not reprogram the hardware due to hang_detected being
581 * set. So we'd effectivly block all timers until the T2 event
584 if (cpu_base
->hang_detected
)
587 tick_program_event(cpu_base
->expires_next
, 1);
591 * When a timer is enqueued and expires earlier than the already enqueued
592 * timers, we have to check, whether it expires earlier than the timer for
593 * which the clock event device was armed.
595 * Called with interrupts disabled and base->cpu_base.lock held
597 static void hrtimer_reprogram(struct hrtimer
*timer
,
598 struct hrtimer_clock_base
*base
)
600 struct hrtimer_cpu_base
*cpu_base
= this_cpu_ptr(&hrtimer_bases
);
601 ktime_t expires
= ktime_sub(hrtimer_get_expires(timer
), base
->offset
);
603 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer
) < 0);
606 * If the timer is not on the current cpu, we cannot reprogram
607 * the other cpus clock event device.
609 if (base
->cpu_base
!= cpu_base
)
613 * If the hrtimer interrupt is running, then it will
614 * reevaluate the clock bases and reprogram the clock event
615 * device. The callbacks are always executed in hard interrupt
616 * context so we don't need an extra check for a running
619 if (cpu_base
->in_hrtirq
)
623 * CLOCK_REALTIME timer might be requested with an absolute
624 * expiry time which is less than base->offset. Set it to 0.
629 if (expires
>= cpu_base
->expires_next
)
632 /* Update the pointer to the next expiring timer */
633 cpu_base
->next_timer
= timer
;
636 * If a hang was detected in the last timer interrupt then we
637 * do not schedule a timer which is earlier than the expiry
638 * which we enforced in the hang detection. We want the system
641 if (cpu_base
->hang_detected
)
645 * Program the timer hardware. We enforce the expiry for
646 * events which are already in the past.
648 cpu_base
->expires_next
= expires
;
649 tick_program_event(expires
, 1);
653 * Initialize the high resolution related parts of cpu_base
655 static inline void hrtimer_init_hres(struct hrtimer_cpu_base
*base
)
657 base
->expires_next
= KTIME_MAX
;
658 base
->hang_detected
= 0;
659 base
->hres_active
= 0;
660 base
->next_timer
= NULL
;
664 * Retrigger next event is called after clock was set
666 * Called with interrupts disabled via on_each_cpu()
668 static void retrigger_next_event(void *arg
)
670 struct hrtimer_cpu_base
*base
= this_cpu_ptr(&hrtimer_bases
);
672 if (!base
->hres_active
)
675 raw_spin_lock(&base
->lock
);
676 hrtimer_update_base(base
);
677 hrtimer_force_reprogram(base
, 0);
678 raw_spin_unlock(&base
->lock
);
682 * Switch to high resolution mode
684 static void hrtimer_switch_to_hres(void)
686 struct hrtimer_cpu_base
*base
= this_cpu_ptr(&hrtimer_bases
);
688 if (tick_init_highres()) {
689 printk(KERN_WARNING
"Could not switch to high resolution "
690 "mode on CPU %d\n", base
->cpu
);
693 base
->hres_active
= 1;
694 hrtimer_resolution
= HIGH_RES_NSEC
;
696 tick_setup_sched_timer();
697 /* "Retrigger" the interrupt to get things going */
698 retrigger_next_event(NULL
);
701 static void clock_was_set_work(struct work_struct
*work
)
706 static DECLARE_WORK(hrtimer_work
, clock_was_set_work
);
709 * Called from timekeeping and resume code to reprogram the hrtimer
710 * interrupt device on all cpus.
712 void clock_was_set_delayed(void)
714 schedule_work(&hrtimer_work
);
719 static inline int __hrtimer_hres_active(struct hrtimer_cpu_base
*b
) { return 0; }
720 static inline int hrtimer_hres_active(void) { return 0; }
721 static inline int hrtimer_is_hres_enabled(void) { return 0; }
722 static inline void hrtimer_switch_to_hres(void) { }
724 hrtimer_force_reprogram(struct hrtimer_cpu_base
*base
, int skip_equal
) { }
725 static inline int hrtimer_reprogram(struct hrtimer
*timer
,
726 struct hrtimer_clock_base
*base
)
730 static inline void hrtimer_init_hres(struct hrtimer_cpu_base
*base
) { }
731 static inline void retrigger_next_event(void *arg
) { }
733 #endif /* CONFIG_HIGH_RES_TIMERS */
736 * Clock realtime was set
738 * Change the offset of the realtime clock vs. the monotonic
741 * We might have to reprogram the high resolution timer interrupt. On
742 * SMP we call the architecture specific code to retrigger _all_ high
743 * resolution timer interrupts. On UP we just disable interrupts and
744 * call the high resolution interrupt code.
746 void clock_was_set(void)
748 #ifdef CONFIG_HIGH_RES_TIMERS
749 /* Retrigger the CPU local events everywhere */
750 on_each_cpu(retrigger_next_event
, NULL
, 1);
752 timerfd_clock_was_set();
756 * During resume we might have to reprogram the high resolution timer
757 * interrupt on all online CPUs. However, all other CPUs will be
758 * stopped with IRQs interrupts disabled so the clock_was_set() call
761 void hrtimers_resume(void)
763 WARN_ONCE(!irqs_disabled(),
764 KERN_INFO
"hrtimers_resume() called with IRQs enabled!");
766 /* Retrigger on the local CPU */
767 retrigger_next_event(NULL
);
768 /* And schedule a retrigger for all others */
769 clock_was_set_delayed();
773 * Counterpart to lock_hrtimer_base above:
776 void unlock_hrtimer_base(const struct hrtimer
*timer
, unsigned long *flags
)
778 raw_spin_unlock_irqrestore(&timer
->base
->cpu_base
->lock
, *flags
);
782 * hrtimer_forward - forward the timer expiry
783 * @timer: hrtimer to forward
784 * @now: forward past this time
785 * @interval: the interval to forward
787 * Forward the timer expiry so it will expire in the future.
788 * Returns the number of overruns.
790 * Can be safely called from the callback function of @timer. If
791 * called from other contexts @timer must neither be enqueued nor
792 * running the callback and the caller needs to take care of
795 * Note: This only updates the timer expiry value and does not requeue
798 u64
hrtimer_forward(struct hrtimer
*timer
, ktime_t now
, ktime_t interval
)
803 delta
= ktime_sub(now
, hrtimer_get_expires(timer
));
808 if (WARN_ON(timer
->state
& HRTIMER_STATE_ENQUEUED
))
811 if (interval
< hrtimer_resolution
)
812 interval
= hrtimer_resolution
;
814 if (unlikely(delta
>= interval
)) {
815 s64 incr
= ktime_to_ns(interval
);
817 orun
= ktime_divns(delta
, incr
);
818 hrtimer_add_expires_ns(timer
, incr
* orun
);
819 if (hrtimer_get_expires_tv64(timer
) > now
)
822 * This (and the ktime_add() below) is the
823 * correction for exact:
827 hrtimer_add_expires(timer
, interval
);
831 EXPORT_SYMBOL_GPL(hrtimer_forward
);
834 * enqueue_hrtimer - internal function to (re)start a timer
836 * The timer is inserted in expiry order. Insertion into the
837 * red black tree is O(log(n)). Must hold the base lock.
839 * Returns 1 when the new timer is the leftmost timer in the tree.
841 static int enqueue_hrtimer(struct hrtimer
*timer
,
842 struct hrtimer_clock_base
*base
)
844 debug_activate(timer
);
846 base
->cpu_base
->active_bases
|= 1 << base
->index
;
848 timer
->state
= HRTIMER_STATE_ENQUEUED
;
850 return timerqueue_add(&base
->active
, &timer
->node
);
854 * __remove_hrtimer - internal function to remove a timer
856 * Caller must hold the base lock.
858 * High resolution timer mode reprograms the clock event device when the
859 * timer is the one which expires next. The caller can disable this by setting
860 * reprogram to zero. This is useful, when the context does a reprogramming
861 * anyway (e.g. timer interrupt)
863 static void __remove_hrtimer(struct hrtimer
*timer
,
864 struct hrtimer_clock_base
*base
,
865 u8 newstate
, int reprogram
)
867 struct hrtimer_cpu_base
*cpu_base
= base
->cpu_base
;
868 u8 state
= timer
->state
;
870 timer
->state
= newstate
;
871 if (!(state
& HRTIMER_STATE_ENQUEUED
))
874 if (!timerqueue_del(&base
->active
, &timer
->node
))
875 cpu_base
->active_bases
&= ~(1 << base
->index
);
877 #ifdef CONFIG_HIGH_RES_TIMERS
879 * Note: If reprogram is false we do not update
880 * cpu_base->next_timer. This happens when we remove the first
881 * timer on a remote cpu. No harm as we never dereference
882 * cpu_base->next_timer. So the worst thing what can happen is
883 * an superflous call to hrtimer_force_reprogram() on the
884 * remote cpu later on if the same timer gets enqueued again.
886 if (reprogram
&& timer
== cpu_base
->next_timer
)
887 hrtimer_force_reprogram(cpu_base
, 1);
892 * remove hrtimer, called with base lock held
895 remove_hrtimer(struct hrtimer
*timer
, struct hrtimer_clock_base
*base
, bool restart
)
897 if (hrtimer_is_queued(timer
)) {
898 u8 state
= timer
->state
;
902 * Remove the timer and force reprogramming when high
903 * resolution mode is active and the timer is on the current
904 * CPU. If we remove a timer on another CPU, reprogramming is
905 * skipped. The interrupt event on this CPU is fired and
906 * reprogramming happens in the interrupt handler. This is a
907 * rare case and less expensive than a smp call.
909 debug_deactivate(timer
);
910 reprogram
= base
->cpu_base
== this_cpu_ptr(&hrtimer_bases
);
913 state
= HRTIMER_STATE_INACTIVE
;
915 __remove_hrtimer(timer
, base
, state
, reprogram
);
921 static inline ktime_t
hrtimer_update_lowres(struct hrtimer
*timer
, ktime_t tim
,
922 const enum hrtimer_mode mode
)
924 #ifdef CONFIG_TIME_LOW_RES
926 * CONFIG_TIME_LOW_RES indicates that the system has no way to return
927 * granular time values. For relative timers we add hrtimer_resolution
928 * (i.e. one jiffie) to prevent short timeouts.
930 timer
->is_rel
= mode
& HRTIMER_MODE_REL
;
932 tim
= ktime_add_safe(tim
, hrtimer_resolution
);
938 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
939 * @timer: the timer to be added
941 * @delta_ns: "slack" range for the timer
942 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
943 * relative (HRTIMER_MODE_REL)
945 void hrtimer_start_range_ns(struct hrtimer
*timer
, ktime_t tim
,
946 u64 delta_ns
, const enum hrtimer_mode mode
)
948 struct hrtimer_clock_base
*base
, *new_base
;
952 base
= lock_hrtimer_base(timer
, &flags
);
954 /* Remove an active timer from the queue: */
955 remove_hrtimer(timer
, base
, true);
957 if (mode
& HRTIMER_MODE_REL
)
958 tim
= ktime_add_safe(tim
, base
->get_time());
960 tim
= hrtimer_update_lowres(timer
, tim
, mode
);
962 hrtimer_set_expires_range_ns(timer
, tim
, delta_ns
);
964 /* Switch the timer base, if necessary: */
965 new_base
= switch_hrtimer_base(timer
, base
, mode
& HRTIMER_MODE_PINNED
);
967 leftmost
= enqueue_hrtimer(timer
, new_base
);
971 if (!hrtimer_is_hres_active(timer
)) {
973 * Kick to reschedule the next tick to handle the new timer
974 * on dynticks target.
976 if (new_base
->cpu_base
->nohz_active
)
977 wake_up_nohz_cpu(new_base
->cpu_base
->cpu
);
979 hrtimer_reprogram(timer
, new_base
);
982 unlock_hrtimer_base(timer
, &flags
);
984 EXPORT_SYMBOL_GPL(hrtimer_start_range_ns
);
987 * hrtimer_try_to_cancel - try to deactivate a timer
988 * @timer: hrtimer to stop
991 * 0 when the timer was not active
992 * 1 when the timer was active
993 * -1 when the timer is currently executing the callback function and
996 int hrtimer_try_to_cancel(struct hrtimer
*timer
)
998 struct hrtimer_clock_base
*base
;
1003 * Check lockless first. If the timer is not active (neither
1004 * enqueued nor running the callback, nothing to do here. The
1005 * base lock does not serialize against a concurrent enqueue,
1006 * so we can avoid taking it.
1008 if (!hrtimer_active(timer
))
1011 base
= lock_hrtimer_base(timer
, &flags
);
1013 if (!hrtimer_callback_running(timer
))
1014 ret
= remove_hrtimer(timer
, base
, false);
1016 unlock_hrtimer_base(timer
, &flags
);
1021 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel
);
1024 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1025 * @timer: the timer to be cancelled
1028 * 0 when the timer was not active
1029 * 1 when the timer was active
1031 int hrtimer_cancel(struct hrtimer
*timer
)
1034 int ret
= hrtimer_try_to_cancel(timer
);
1041 EXPORT_SYMBOL_GPL(hrtimer_cancel
);
1044 * hrtimer_get_remaining - get remaining time for the timer
1045 * @timer: the timer to read
1046 * @adjust: adjust relative timers when CONFIG_TIME_LOW_RES=y
1048 ktime_t
__hrtimer_get_remaining(const struct hrtimer
*timer
, bool adjust
)
1050 unsigned long flags
;
1053 lock_hrtimer_base(timer
, &flags
);
1054 if (IS_ENABLED(CONFIG_TIME_LOW_RES
) && adjust
)
1055 rem
= hrtimer_expires_remaining_adjusted(timer
);
1057 rem
= hrtimer_expires_remaining(timer
);
1058 unlock_hrtimer_base(timer
, &flags
);
1062 EXPORT_SYMBOL_GPL(__hrtimer_get_remaining
);
1064 #ifdef CONFIG_NO_HZ_COMMON
1066 * hrtimer_get_next_event - get the time until next expiry event
1068 * Returns the next expiry time or KTIME_MAX if no timer is pending.
1070 u64
hrtimer_get_next_event(void)
1072 struct hrtimer_cpu_base
*cpu_base
= this_cpu_ptr(&hrtimer_bases
);
1073 u64 expires
= KTIME_MAX
;
1074 unsigned long flags
;
1076 raw_spin_lock_irqsave(&cpu_base
->lock
, flags
);
1078 if (!__hrtimer_hres_active(cpu_base
))
1079 expires
= __hrtimer_get_next_event(cpu_base
);
1081 raw_spin_unlock_irqrestore(&cpu_base
->lock
, flags
);
1087 static inline int hrtimer_clockid_to_base(clockid_t clock_id
)
1089 if (likely(clock_id
< MAX_CLOCKS
)) {
1090 int base
= hrtimer_clock_to_base_table
[clock_id
];
1092 if (likely(base
!= HRTIMER_MAX_CLOCK_BASES
))
1095 WARN(1, "Invalid clockid %d. Using MONOTONIC\n", clock_id
);
1096 return HRTIMER_BASE_MONOTONIC
;
1099 static void __hrtimer_init(struct hrtimer
*timer
, clockid_t clock_id
,
1100 enum hrtimer_mode mode
)
1102 struct hrtimer_cpu_base
*cpu_base
;
1105 memset(timer
, 0, sizeof(struct hrtimer
));
1107 cpu_base
= raw_cpu_ptr(&hrtimer_bases
);
1109 if (clock_id
== CLOCK_REALTIME
&& mode
!= HRTIMER_MODE_ABS
)
1110 clock_id
= CLOCK_MONOTONIC
;
1112 base
= hrtimer_clockid_to_base(clock_id
);
1113 timer
->base
= &cpu_base
->clock_base
[base
];
1114 timerqueue_init(&timer
->node
);
1118 * hrtimer_init - initialize a timer to the given clock
1119 * @timer: the timer to be initialized
1120 * @clock_id: the clock to be used
1121 * @mode: timer mode abs/rel
1123 void hrtimer_init(struct hrtimer
*timer
, clockid_t clock_id
,
1124 enum hrtimer_mode mode
)
1126 debug_init(timer
, clock_id
, mode
);
1127 __hrtimer_init(timer
, clock_id
, mode
);
1129 EXPORT_SYMBOL_GPL(hrtimer_init
);
1132 * A timer is active, when it is enqueued into the rbtree or the
1133 * callback function is running or it's in the state of being migrated
1136 * It is important for this function to not return a false negative.
1138 bool hrtimer_active(const struct hrtimer
*timer
)
1140 struct hrtimer_cpu_base
*cpu_base
;
1144 cpu_base
= READ_ONCE(timer
->base
->cpu_base
);
1145 seq
= raw_read_seqcount_begin(&cpu_base
->seq
);
1147 if (timer
->state
!= HRTIMER_STATE_INACTIVE
||
1148 cpu_base
->running
== timer
)
1151 } while (read_seqcount_retry(&cpu_base
->seq
, seq
) ||
1152 cpu_base
!= READ_ONCE(timer
->base
->cpu_base
));
1156 EXPORT_SYMBOL_GPL(hrtimer_active
);
1159 * The write_seqcount_barrier()s in __run_hrtimer() split the thing into 3
1160 * distinct sections:
1162 * - queued: the timer is queued
1163 * - callback: the timer is being ran
1164 * - post: the timer is inactive or (re)queued
1166 * On the read side we ensure we observe timer->state and cpu_base->running
1167 * from the same section, if anything changed while we looked at it, we retry.
1168 * This includes timer->base changing because sequence numbers alone are
1169 * insufficient for that.
1171 * The sequence numbers are required because otherwise we could still observe
1172 * a false negative if the read side got smeared over multiple consequtive
1173 * __run_hrtimer() invocations.
1176 static void __run_hrtimer(struct hrtimer_cpu_base
*cpu_base
,
1177 struct hrtimer_clock_base
*base
,
1178 struct hrtimer
*timer
, ktime_t
*now
)
1180 enum hrtimer_restart (*fn
)(struct hrtimer
*);
1183 lockdep_assert_held(&cpu_base
->lock
);
1185 debug_deactivate(timer
);
1186 cpu_base
->running
= timer
;
1189 * Separate the ->running assignment from the ->state assignment.
1191 * As with a regular write barrier, this ensures the read side in
1192 * hrtimer_active() cannot observe cpu_base->running == NULL &&
1193 * timer->state == INACTIVE.
1195 raw_write_seqcount_barrier(&cpu_base
->seq
);
1197 __remove_hrtimer(timer
, base
, HRTIMER_STATE_INACTIVE
, 0);
1198 fn
= timer
->function
;
1201 * Clear the 'is relative' flag for the TIME_LOW_RES case. If the
1202 * timer is restarted with a period then it becomes an absolute
1203 * timer. If its not restarted it does not matter.
1205 if (IS_ENABLED(CONFIG_TIME_LOW_RES
))
1206 timer
->is_rel
= false;
1209 * Because we run timers from hardirq context, there is no chance
1210 * they get migrated to another cpu, therefore its safe to unlock
1213 raw_spin_unlock(&cpu_base
->lock
);
1214 trace_hrtimer_expire_entry(timer
, now
);
1215 restart
= fn(timer
);
1216 trace_hrtimer_expire_exit(timer
);
1217 raw_spin_lock(&cpu_base
->lock
);
1220 * Note: We clear the running state after enqueue_hrtimer and
1221 * we do not reprogram the event hardware. Happens either in
1222 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1224 * Note: Because we dropped the cpu_base->lock above,
1225 * hrtimer_start_range_ns() can have popped in and enqueued the timer
1228 if (restart
!= HRTIMER_NORESTART
&&
1229 !(timer
->state
& HRTIMER_STATE_ENQUEUED
))
1230 enqueue_hrtimer(timer
, base
);
1233 * Separate the ->running assignment from the ->state assignment.
1235 * As with a regular write barrier, this ensures the read side in
1236 * hrtimer_active() cannot observe cpu_base->running == NULL &&
1237 * timer->state == INACTIVE.
1239 raw_write_seqcount_barrier(&cpu_base
->seq
);
1241 WARN_ON_ONCE(cpu_base
->running
!= timer
);
1242 cpu_base
->running
= NULL
;
1245 static void __hrtimer_run_queues(struct hrtimer_cpu_base
*cpu_base
, ktime_t now
)
1247 struct hrtimer_clock_base
*base
= cpu_base
->clock_base
;
1248 unsigned int active
= cpu_base
->active_bases
;
1250 for (; active
; base
++, active
>>= 1) {
1251 struct timerqueue_node
*node
;
1254 if (!(active
& 0x01))
1257 basenow
= ktime_add(now
, base
->offset
);
1259 while ((node
= timerqueue_getnext(&base
->active
))) {
1260 struct hrtimer
*timer
;
1262 timer
= container_of(node
, struct hrtimer
, node
);
1265 * The immediate goal for using the softexpires is
1266 * minimizing wakeups, not running timers at the
1267 * earliest interrupt after their soft expiration.
1268 * This allows us to avoid using a Priority Search
1269 * Tree, which can answer a stabbing querry for
1270 * overlapping intervals and instead use the simple
1271 * BST we already have.
1272 * We don't add extra wakeups by delaying timers that
1273 * are right-of a not yet expired timer, because that
1274 * timer will have to trigger a wakeup anyway.
1276 if (basenow
< hrtimer_get_softexpires_tv64(timer
))
1279 __run_hrtimer(cpu_base
, base
, timer
, &basenow
);
1284 #ifdef CONFIG_HIGH_RES_TIMERS
1287 * High resolution timer interrupt
1288 * Called with interrupts disabled
1290 void hrtimer_interrupt(struct clock_event_device
*dev
)
1292 struct hrtimer_cpu_base
*cpu_base
= this_cpu_ptr(&hrtimer_bases
);
1293 ktime_t expires_next
, now
, entry_time
, delta
;
1296 BUG_ON(!cpu_base
->hres_active
);
1297 cpu_base
->nr_events
++;
1298 dev
->next_event
= KTIME_MAX
;
1300 raw_spin_lock(&cpu_base
->lock
);
1301 entry_time
= now
= hrtimer_update_base(cpu_base
);
1303 cpu_base
->in_hrtirq
= 1;
1305 * We set expires_next to KTIME_MAX here with cpu_base->lock
1306 * held to prevent that a timer is enqueued in our queue via
1307 * the migration code. This does not affect enqueueing of
1308 * timers which run their callback and need to be requeued on
1311 cpu_base
->expires_next
= KTIME_MAX
;
1313 __hrtimer_run_queues(cpu_base
, now
);
1315 /* Reevaluate the clock bases for the next expiry */
1316 expires_next
= __hrtimer_get_next_event(cpu_base
);
1318 * Store the new expiry value so the migration code can verify
1321 cpu_base
->expires_next
= expires_next
;
1322 cpu_base
->in_hrtirq
= 0;
1323 raw_spin_unlock(&cpu_base
->lock
);
1325 /* Reprogramming necessary ? */
1326 if (!tick_program_event(expires_next
, 0)) {
1327 cpu_base
->hang_detected
= 0;
1332 * The next timer was already expired due to:
1334 * - long lasting callbacks
1335 * - being scheduled away when running in a VM
1337 * We need to prevent that we loop forever in the hrtimer
1338 * interrupt routine. We give it 3 attempts to avoid
1339 * overreacting on some spurious event.
1341 * Acquire base lock for updating the offsets and retrieving
1344 raw_spin_lock(&cpu_base
->lock
);
1345 now
= hrtimer_update_base(cpu_base
);
1346 cpu_base
->nr_retries
++;
1350 * Give the system a chance to do something else than looping
1351 * here. We stored the entry time, so we know exactly how long
1352 * we spent here. We schedule the next event this amount of
1355 cpu_base
->nr_hangs
++;
1356 cpu_base
->hang_detected
= 1;
1357 raw_spin_unlock(&cpu_base
->lock
);
1358 delta
= ktime_sub(now
, entry_time
);
1359 if ((unsigned int)delta
> cpu_base
->max_hang_time
)
1360 cpu_base
->max_hang_time
= (unsigned int) delta
;
1362 * Limit it to a sensible value as we enforce a longer
1363 * delay. Give the CPU at least 100ms to catch up.
1365 if (delta
> 100 * NSEC_PER_MSEC
)
1366 expires_next
= ktime_add_ns(now
, 100 * NSEC_PER_MSEC
);
1368 expires_next
= ktime_add(now
, delta
);
1369 tick_program_event(expires_next
, 1);
1370 printk_once(KERN_WARNING
"hrtimer: interrupt took %llu ns\n",
1371 ktime_to_ns(delta
));
1374 /* called with interrupts disabled */
1375 static inline void __hrtimer_peek_ahead_timers(void)
1377 struct tick_device
*td
;
1379 if (!hrtimer_hres_active())
1382 td
= this_cpu_ptr(&tick_cpu_device
);
1383 if (td
&& td
->evtdev
)
1384 hrtimer_interrupt(td
->evtdev
);
1387 #else /* CONFIG_HIGH_RES_TIMERS */
1389 static inline void __hrtimer_peek_ahead_timers(void) { }
1391 #endif /* !CONFIG_HIGH_RES_TIMERS */
1394 * Called from run_local_timers in hardirq context every jiffy
1396 void hrtimer_run_queues(void)
1398 struct hrtimer_cpu_base
*cpu_base
= this_cpu_ptr(&hrtimer_bases
);
1401 if (__hrtimer_hres_active(cpu_base
))
1405 * This _is_ ugly: We have to check periodically, whether we
1406 * can switch to highres and / or nohz mode. The clocksource
1407 * switch happens with xtime_lock held. Notification from
1408 * there only sets the check bit in the tick_oneshot code,
1409 * otherwise we might deadlock vs. xtime_lock.
1411 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) {
1412 hrtimer_switch_to_hres();
1416 raw_spin_lock(&cpu_base
->lock
);
1417 now
= hrtimer_update_base(cpu_base
);
1418 __hrtimer_run_queues(cpu_base
, now
);
1419 raw_spin_unlock(&cpu_base
->lock
);
1423 * Sleep related functions:
1425 static enum hrtimer_restart
hrtimer_wakeup(struct hrtimer
*timer
)
1427 struct hrtimer_sleeper
*t
=
1428 container_of(timer
, struct hrtimer_sleeper
, timer
);
1429 struct task_struct
*task
= t
->task
;
1433 wake_up_process(task
);
1435 return HRTIMER_NORESTART
;
1438 void hrtimer_init_sleeper(struct hrtimer_sleeper
*sl
, struct task_struct
*task
)
1440 sl
->timer
.function
= hrtimer_wakeup
;
1443 EXPORT_SYMBOL_GPL(hrtimer_init_sleeper
);
1445 int nanosleep_copyout(struct restart_block
*restart
, struct timespec64
*ts
)
1447 switch(restart
->nanosleep
.type
) {
1448 #ifdef CONFIG_COMPAT
1450 if (compat_put_timespec64(ts
, restart
->nanosleep
.compat_rmtp
))
1455 if (put_timespec64(ts
, restart
->nanosleep
.rmtp
))
1461 return -ERESTART_RESTARTBLOCK
;
1464 static int __sched
do_nanosleep(struct hrtimer_sleeper
*t
, enum hrtimer_mode mode
)
1466 struct restart_block
*restart
;
1468 hrtimer_init_sleeper(t
, current
);
1471 set_current_state(TASK_INTERRUPTIBLE
);
1472 hrtimer_start_expires(&t
->timer
, mode
);
1474 if (likely(t
->task
))
1475 freezable_schedule();
1477 hrtimer_cancel(&t
->timer
);
1478 mode
= HRTIMER_MODE_ABS
;
1480 } while (t
->task
&& !signal_pending(current
));
1482 __set_current_state(TASK_RUNNING
);
1487 restart
= ¤t
->restart_block
;
1488 if (restart
->nanosleep
.type
!= TT_NONE
) {
1489 ktime_t rem
= hrtimer_expires_remaining(&t
->timer
);
1490 struct timespec64 rmt
;
1494 rmt
= ktime_to_timespec64(rem
);
1496 return nanosleep_copyout(restart
, &rmt
);
1498 return -ERESTART_RESTARTBLOCK
;
1501 static long __sched
hrtimer_nanosleep_restart(struct restart_block
*restart
)
1503 struct hrtimer_sleeper t
;
1506 hrtimer_init_on_stack(&t
.timer
, restart
->nanosleep
.clockid
,
1508 hrtimer_set_expires_tv64(&t
.timer
, restart
->nanosleep
.expires
);
1510 ret
= do_nanosleep(&t
, HRTIMER_MODE_ABS
);
1511 destroy_hrtimer_on_stack(&t
.timer
);
1515 long hrtimer_nanosleep(const struct timespec64
*rqtp
,
1516 const enum hrtimer_mode mode
, const clockid_t clockid
)
1518 struct restart_block
*restart
;
1519 struct hrtimer_sleeper t
;
1523 slack
= current
->timer_slack_ns
;
1524 if (dl_task(current
) || rt_task(current
))
1527 hrtimer_init_on_stack(&t
.timer
, clockid
, mode
);
1528 hrtimer_set_expires_range_ns(&t
.timer
, timespec64_to_ktime(*rqtp
), slack
);
1529 ret
= do_nanosleep(&t
, mode
);
1530 if (ret
!= -ERESTART_RESTARTBLOCK
)
1533 /* Absolute timers do not update the rmtp value and restart: */
1534 if (mode
== HRTIMER_MODE_ABS
) {
1535 ret
= -ERESTARTNOHAND
;
1539 restart
= ¤t
->restart_block
;
1540 restart
->fn
= hrtimer_nanosleep_restart
;
1541 restart
->nanosleep
.clockid
= t
.timer
.base
->clockid
;
1542 restart
->nanosleep
.expires
= hrtimer_get_expires_tv64(&t
.timer
);
1544 destroy_hrtimer_on_stack(&t
.timer
);
1548 SYSCALL_DEFINE2(nanosleep
, struct timespec __user
*, rqtp
,
1549 struct timespec __user
*, rmtp
)
1551 struct timespec64 tu
;
1553 if (get_timespec64(&tu
, rqtp
))
1556 if (!timespec64_valid(&tu
))
1559 current
->restart_block
.nanosleep
.type
= rmtp
? TT_NATIVE
: TT_NONE
;
1560 current
->restart_block
.nanosleep
.rmtp
= rmtp
;
1561 return hrtimer_nanosleep(&tu
, HRTIMER_MODE_REL
, CLOCK_MONOTONIC
);
1564 #ifdef CONFIG_COMPAT
1566 COMPAT_SYSCALL_DEFINE2(nanosleep
, struct compat_timespec __user
*, rqtp
,
1567 struct compat_timespec __user
*, rmtp
)
1569 struct timespec64 tu
;
1571 if (compat_get_timespec64(&tu
, rqtp
))
1574 if (!timespec64_valid(&tu
))
1577 current
->restart_block
.nanosleep
.type
= rmtp
? TT_COMPAT
: TT_NONE
;
1578 current
->restart_block
.nanosleep
.compat_rmtp
= rmtp
;
1579 return hrtimer_nanosleep(&tu
, HRTIMER_MODE_REL
, CLOCK_MONOTONIC
);
1584 * Functions related to boot-time initialization:
1586 int hrtimers_prepare_cpu(unsigned int cpu
)
1588 struct hrtimer_cpu_base
*cpu_base
= &per_cpu(hrtimer_bases
, cpu
);
1591 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++) {
1592 cpu_base
->clock_base
[i
].cpu_base
= cpu_base
;
1593 timerqueue_init_head(&cpu_base
->clock_base
[i
].active
);
1596 cpu_base
->active_bases
= 0;
1597 cpu_base
->cpu
= cpu
;
1598 hrtimer_init_hres(cpu_base
);
1602 #ifdef CONFIG_HOTPLUG_CPU
1604 static void migrate_hrtimer_list(struct hrtimer_clock_base
*old_base
,
1605 struct hrtimer_clock_base
*new_base
)
1607 struct hrtimer
*timer
;
1608 struct timerqueue_node
*node
;
1610 while ((node
= timerqueue_getnext(&old_base
->active
))) {
1611 timer
= container_of(node
, struct hrtimer
, node
);
1612 BUG_ON(hrtimer_callback_running(timer
));
1613 debug_deactivate(timer
);
1616 * Mark it as ENQUEUED not INACTIVE otherwise the
1617 * timer could be seen as !active and just vanish away
1618 * under us on another CPU
1620 __remove_hrtimer(timer
, old_base
, HRTIMER_STATE_ENQUEUED
, 0);
1621 timer
->base
= new_base
;
1623 * Enqueue the timers on the new cpu. This does not
1624 * reprogram the event device in case the timer
1625 * expires before the earliest on this CPU, but we run
1626 * hrtimer_interrupt after we migrated everything to
1627 * sort out already expired timers and reprogram the
1630 enqueue_hrtimer(timer
, new_base
);
1634 int hrtimers_dead_cpu(unsigned int scpu
)
1636 struct hrtimer_cpu_base
*old_base
, *new_base
;
1639 BUG_ON(cpu_online(scpu
));
1640 tick_cancel_sched_timer(scpu
);
1642 local_irq_disable();
1643 old_base
= &per_cpu(hrtimer_bases
, scpu
);
1644 new_base
= this_cpu_ptr(&hrtimer_bases
);
1646 * The caller is globally serialized and nobody else
1647 * takes two locks at once, deadlock is not possible.
1649 raw_spin_lock(&new_base
->lock
);
1650 raw_spin_lock_nested(&old_base
->lock
, SINGLE_DEPTH_NESTING
);
1652 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++) {
1653 migrate_hrtimer_list(&old_base
->clock_base
[i
],
1654 &new_base
->clock_base
[i
]);
1657 raw_spin_unlock(&old_base
->lock
);
1658 raw_spin_unlock(&new_base
->lock
);
1660 /* Check, if we got expired work to do */
1661 __hrtimer_peek_ahead_timers();
1666 #endif /* CONFIG_HOTPLUG_CPU */
1668 void __init
hrtimers_init(void)
1670 hrtimers_prepare_cpu(smp_processor_id());
1674 * schedule_hrtimeout_range_clock - sleep until timeout
1675 * @expires: timeout value (ktime_t)
1676 * @delta: slack in expires timeout (ktime_t)
1677 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1678 * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME
1681 schedule_hrtimeout_range_clock(ktime_t
*expires
, u64 delta
,
1682 const enum hrtimer_mode mode
, int clock
)
1684 struct hrtimer_sleeper t
;
1687 * Optimize when a zero timeout value is given. It does not
1688 * matter whether this is an absolute or a relative time.
1690 if (expires
&& *expires
== 0) {
1691 __set_current_state(TASK_RUNNING
);
1696 * A NULL parameter means "infinite"
1703 hrtimer_init_on_stack(&t
.timer
, clock
, mode
);
1704 hrtimer_set_expires_range_ns(&t
.timer
, *expires
, delta
);
1706 hrtimer_init_sleeper(&t
, current
);
1708 hrtimer_start_expires(&t
.timer
, mode
);
1713 hrtimer_cancel(&t
.timer
);
1714 destroy_hrtimer_on_stack(&t
.timer
);
1716 __set_current_state(TASK_RUNNING
);
1718 return !t
.task
? 0 : -EINTR
;
1722 * schedule_hrtimeout_range - sleep until timeout
1723 * @expires: timeout value (ktime_t)
1724 * @delta: slack in expires timeout (ktime_t)
1725 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1727 * Make the current task sleep until the given expiry time has
1728 * elapsed. The routine will return immediately unless
1729 * the current task state has been set (see set_current_state()).
1731 * The @delta argument gives the kernel the freedom to schedule the
1732 * actual wakeup to a time that is both power and performance friendly.
1733 * The kernel give the normal best effort behavior for "@expires+@delta",
1734 * but may decide to fire the timer earlier, but no earlier than @expires.
1736 * You can set the task state as follows -
1738 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1739 * pass before the routine returns unless the current task is explicitly
1740 * woken up, (e.g. by wake_up_process()).
1742 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1743 * delivered to the current task or the current task is explicitly woken
1746 * The current task state is guaranteed to be TASK_RUNNING when this
1749 * Returns 0 when the timer has expired. If the task was woken before the
1750 * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or
1751 * by an explicit wakeup, it returns -EINTR.
1753 int __sched
schedule_hrtimeout_range(ktime_t
*expires
, u64 delta
,
1754 const enum hrtimer_mode mode
)
1756 return schedule_hrtimeout_range_clock(expires
, delta
, mode
,
1759 EXPORT_SYMBOL_GPL(schedule_hrtimeout_range
);
1762 * schedule_hrtimeout - sleep until timeout
1763 * @expires: timeout value (ktime_t)
1764 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1766 * Make the current task sleep until the given expiry time has
1767 * elapsed. The routine will return immediately unless
1768 * the current task state has been set (see set_current_state()).
1770 * You can set the task state as follows -
1772 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1773 * pass before the routine returns unless the current task is explicitly
1774 * woken up, (e.g. by wake_up_process()).
1776 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1777 * delivered to the current task or the current task is explicitly woken
1780 * The current task state is guaranteed to be TASK_RUNNING when this
1783 * Returns 0 when the timer has expired. If the task was woken before the
1784 * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or
1785 * by an explicit wakeup, it returns -EINTR.
1787 int __sched
schedule_hrtimeout(ktime_t
*expires
,
1788 const enum hrtimer_mode mode
)
1790 return schedule_hrtimeout_range(expires
, 0, mode
);
1792 EXPORT_SYMBOL_GPL(schedule_hrtimeout
);