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>
60 * There are more clockids then hrtimer bases. Thus, we index
61 * into the timer bases by the hrtimer_base_type enum. When trying
62 * to reach a base using a clockid, hrtimer_clockid_to_base()
63 * is used to convert from clockid to the proper hrtimer_base_type.
65 DEFINE_PER_CPU(struct hrtimer_cpu_base
, hrtimer_bases
) =
68 .lock
= __RAW_SPIN_LOCK_UNLOCKED(hrtimer_bases
.lock
),
72 .index
= HRTIMER_BASE_MONOTONIC
,
73 .clockid
= CLOCK_MONOTONIC
,
74 .get_time
= &ktime_get
,
75 .resolution
= KTIME_LOW_RES
,
78 .index
= HRTIMER_BASE_REALTIME
,
79 .clockid
= CLOCK_REALTIME
,
80 .get_time
= &ktime_get_real
,
81 .resolution
= KTIME_LOW_RES
,
84 .index
= HRTIMER_BASE_BOOTTIME
,
85 .clockid
= CLOCK_BOOTTIME
,
86 .get_time
= &ktime_get_boottime
,
87 .resolution
= KTIME_LOW_RES
,
90 .index
= HRTIMER_BASE_TAI
,
92 .get_time
= &ktime_get_clocktai
,
93 .resolution
= KTIME_LOW_RES
,
98 static const int hrtimer_clock_to_base_table
[MAX_CLOCKS
] = {
99 [CLOCK_REALTIME
] = HRTIMER_BASE_REALTIME
,
100 [CLOCK_MONOTONIC
] = HRTIMER_BASE_MONOTONIC
,
101 [CLOCK_BOOTTIME
] = HRTIMER_BASE_BOOTTIME
,
102 [CLOCK_TAI
] = HRTIMER_BASE_TAI
,
105 static inline int hrtimer_clockid_to_base(clockid_t clock_id
)
107 return hrtimer_clock_to_base_table
[clock_id
];
112 * Get the coarse grained time at the softirq based on xtime and
115 static void hrtimer_get_softirq_time(struct hrtimer_cpu_base
*base
)
117 ktime_t xtim
, mono
, boot
;
118 struct timespec xts
, tom
, slp
;
121 get_xtime_and_monotonic_and_sleep_offset(&xts
, &tom
, &slp
);
122 tai_offset
= timekeeping_get_tai_offset();
124 xtim
= timespec_to_ktime(xts
);
125 mono
= ktime_add(xtim
, timespec_to_ktime(tom
));
126 boot
= ktime_add(mono
, timespec_to_ktime(slp
));
127 base
->clock_base
[HRTIMER_BASE_REALTIME
].softirq_time
= xtim
;
128 base
->clock_base
[HRTIMER_BASE_MONOTONIC
].softirq_time
= mono
;
129 base
->clock_base
[HRTIMER_BASE_BOOTTIME
].softirq_time
= boot
;
130 base
->clock_base
[HRTIMER_BASE_TAI
].softirq_time
=
131 ktime_add(xtim
, ktime_set(tai_offset
, 0));
135 * Functions and macros which are different for UP/SMP systems are kept in a
141 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
142 * means that all timers which are tied to this base via timer->base are
143 * locked, and the base itself is locked too.
145 * So __run_timers/migrate_timers can safely modify all timers which could
146 * be found on the lists/queues.
148 * When the timer's base is locked, and the timer removed from list, it is
149 * possible to set timer->base = NULL and drop the lock: the timer remains
153 struct hrtimer_clock_base
*lock_hrtimer_base(const struct hrtimer
*timer
,
154 unsigned long *flags
)
156 struct hrtimer_clock_base
*base
;
160 if (likely(base
!= NULL
)) {
161 raw_spin_lock_irqsave(&base
->cpu_base
->lock
, *flags
);
162 if (likely(base
== timer
->base
))
164 /* The timer has migrated to another CPU: */
165 raw_spin_unlock_irqrestore(&base
->cpu_base
->lock
, *flags
);
172 * With HIGHRES=y we do not migrate the timer when it is expiring
173 * before the next event on the target cpu because we cannot reprogram
174 * the target cpu hardware and we would cause it to fire late.
176 * Called with cpu_base->lock of target cpu held.
179 hrtimer_check_target(struct hrtimer
*timer
, struct hrtimer_clock_base
*new_base
)
181 #ifdef CONFIG_HIGH_RES_TIMERS
184 if (!new_base
->cpu_base
->hres_active
)
187 expires
= ktime_sub(hrtimer_get_expires(timer
), new_base
->offset
);
188 return expires
.tv64
<= new_base
->cpu_base
->expires_next
.tv64
;
195 * Switch the timer base to the current CPU when possible.
197 static inline struct hrtimer_clock_base
*
198 switch_hrtimer_base(struct hrtimer
*timer
, struct hrtimer_clock_base
*base
,
201 struct hrtimer_clock_base
*new_base
;
202 struct hrtimer_cpu_base
*new_cpu_base
;
203 int this_cpu
= smp_processor_id();
204 int cpu
= get_nohz_timer_target(pinned
);
205 int basenum
= base
->index
;
208 new_cpu_base
= &per_cpu(hrtimer_bases
, cpu
);
209 new_base
= &new_cpu_base
->clock_base
[basenum
];
211 if (base
!= new_base
) {
213 * We are trying to move timer to new_base.
214 * However we can't change timer's base while it is running,
215 * so we keep it on the same CPU. No hassle vs. reprogramming
216 * the event source in the high resolution case. The softirq
217 * code will take care of this when the timer function has
218 * completed. There is no conflict as we hold the lock until
219 * the timer is enqueued.
221 if (unlikely(hrtimer_callback_running(timer
)))
224 /* See the comment in lock_timer_base() */
226 raw_spin_unlock(&base
->cpu_base
->lock
);
227 raw_spin_lock(&new_base
->cpu_base
->lock
);
229 if (cpu
!= this_cpu
&& hrtimer_check_target(timer
, new_base
)) {
231 raw_spin_unlock(&new_base
->cpu_base
->lock
);
232 raw_spin_lock(&base
->cpu_base
->lock
);
236 timer
->base
= new_base
;
238 if (cpu
!= this_cpu
&& hrtimer_check_target(timer
, new_base
)) {
246 #else /* CONFIG_SMP */
248 static inline struct hrtimer_clock_base
*
249 lock_hrtimer_base(const struct hrtimer
*timer
, unsigned long *flags
)
251 struct hrtimer_clock_base
*base
= timer
->base
;
253 raw_spin_lock_irqsave(&base
->cpu_base
->lock
, *flags
);
258 # define switch_hrtimer_base(t, b, p) (b)
260 #endif /* !CONFIG_SMP */
263 * Functions for the union type storage format of ktime_t which are
264 * too large for inlining:
266 #if BITS_PER_LONG < 64
267 # ifndef CONFIG_KTIME_SCALAR
269 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
271 * @nsec: the scalar nsec value to add
273 * Returns the sum of kt and nsec in ktime_t format
275 ktime_t
ktime_add_ns(const ktime_t kt
, u64 nsec
)
279 if (likely(nsec
< NSEC_PER_SEC
)) {
282 unsigned long rem
= do_div(nsec
, NSEC_PER_SEC
);
284 /* Make sure nsec fits into long */
285 if (unlikely(nsec
> KTIME_SEC_MAX
))
286 return (ktime_t
){ .tv64
= KTIME_MAX
};
288 tmp
= ktime_set((long)nsec
, rem
);
291 return ktime_add(kt
, tmp
);
294 EXPORT_SYMBOL_GPL(ktime_add_ns
);
297 * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable
299 * @nsec: the scalar nsec value to subtract
301 * Returns the subtraction of @nsec from @kt in ktime_t format
303 ktime_t
ktime_sub_ns(const ktime_t kt
, u64 nsec
)
307 if (likely(nsec
< NSEC_PER_SEC
)) {
310 unsigned long rem
= do_div(nsec
, NSEC_PER_SEC
);
312 tmp
= ktime_set((long)nsec
, rem
);
315 return ktime_sub(kt
, tmp
);
318 EXPORT_SYMBOL_GPL(ktime_sub_ns
);
319 # endif /* !CONFIG_KTIME_SCALAR */
322 * Divide a ktime value by a nanosecond value
324 u64
ktime_divns(const ktime_t kt
, s64 div
)
329 dclc
= ktime_to_ns(kt
);
330 /* Make sure the divisor is less than 2^32: */
336 do_div(dclc
, (unsigned long) div
);
340 #endif /* BITS_PER_LONG >= 64 */
343 * Add two ktime values and do a safety check for overflow:
345 ktime_t
ktime_add_safe(const ktime_t lhs
, const ktime_t rhs
)
347 ktime_t res
= ktime_add(lhs
, rhs
);
350 * We use KTIME_SEC_MAX here, the maximum timeout which we can
351 * return to user space in a timespec:
353 if (res
.tv64
< 0 || res
.tv64
< lhs
.tv64
|| res
.tv64
< rhs
.tv64
)
354 res
= ktime_set(KTIME_SEC_MAX
, 0);
359 EXPORT_SYMBOL_GPL(ktime_add_safe
);
361 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
363 static struct debug_obj_descr hrtimer_debug_descr
;
365 static void *hrtimer_debug_hint(void *addr
)
367 return ((struct hrtimer
*) addr
)->function
;
371 * fixup_init is called when:
372 * - an active object is initialized
374 static int hrtimer_fixup_init(void *addr
, enum debug_obj_state state
)
376 struct hrtimer
*timer
= addr
;
379 case ODEBUG_STATE_ACTIVE
:
380 hrtimer_cancel(timer
);
381 debug_object_init(timer
, &hrtimer_debug_descr
);
389 * fixup_activate is called when:
390 * - an active object is activated
391 * - an unknown object is activated (might be a statically initialized object)
393 static int hrtimer_fixup_activate(void *addr
, enum debug_obj_state state
)
397 case ODEBUG_STATE_NOTAVAILABLE
:
401 case ODEBUG_STATE_ACTIVE
:
410 * fixup_free is called when:
411 * - an active object is freed
413 static int hrtimer_fixup_free(void *addr
, enum debug_obj_state state
)
415 struct hrtimer
*timer
= addr
;
418 case ODEBUG_STATE_ACTIVE
:
419 hrtimer_cancel(timer
);
420 debug_object_free(timer
, &hrtimer_debug_descr
);
427 static struct debug_obj_descr hrtimer_debug_descr
= {
429 .debug_hint
= hrtimer_debug_hint
,
430 .fixup_init
= hrtimer_fixup_init
,
431 .fixup_activate
= hrtimer_fixup_activate
,
432 .fixup_free
= hrtimer_fixup_free
,
435 static inline void debug_hrtimer_init(struct hrtimer
*timer
)
437 debug_object_init(timer
, &hrtimer_debug_descr
);
440 static inline void debug_hrtimer_activate(struct hrtimer
*timer
)
442 debug_object_activate(timer
, &hrtimer_debug_descr
);
445 static inline void debug_hrtimer_deactivate(struct hrtimer
*timer
)
447 debug_object_deactivate(timer
, &hrtimer_debug_descr
);
450 static inline void debug_hrtimer_free(struct hrtimer
*timer
)
452 debug_object_free(timer
, &hrtimer_debug_descr
);
455 static void __hrtimer_init(struct hrtimer
*timer
, clockid_t clock_id
,
456 enum hrtimer_mode mode
);
458 void hrtimer_init_on_stack(struct hrtimer
*timer
, clockid_t clock_id
,
459 enum hrtimer_mode mode
)
461 debug_object_init_on_stack(timer
, &hrtimer_debug_descr
);
462 __hrtimer_init(timer
, clock_id
, mode
);
464 EXPORT_SYMBOL_GPL(hrtimer_init_on_stack
);
466 void destroy_hrtimer_on_stack(struct hrtimer
*timer
)
468 debug_object_free(timer
, &hrtimer_debug_descr
);
472 static inline void debug_hrtimer_init(struct hrtimer
*timer
) { }
473 static inline void debug_hrtimer_activate(struct hrtimer
*timer
) { }
474 static inline void debug_hrtimer_deactivate(struct hrtimer
*timer
) { }
478 debug_init(struct hrtimer
*timer
, clockid_t clockid
,
479 enum hrtimer_mode mode
)
481 debug_hrtimer_init(timer
);
482 trace_hrtimer_init(timer
, clockid
, mode
);
485 static inline void debug_activate(struct hrtimer
*timer
)
487 debug_hrtimer_activate(timer
);
488 trace_hrtimer_start(timer
);
491 static inline void debug_deactivate(struct hrtimer
*timer
)
493 debug_hrtimer_deactivate(timer
);
494 trace_hrtimer_cancel(timer
);
497 /* High resolution timer related functions */
498 #ifdef CONFIG_HIGH_RES_TIMERS
501 * High resolution timer enabled ?
503 static int hrtimer_hres_enabled __read_mostly
= 1;
506 * Enable / Disable high resolution mode
508 static int __init
setup_hrtimer_hres(char *str
)
510 if (!strcmp(str
, "off"))
511 hrtimer_hres_enabled
= 0;
512 else if (!strcmp(str
, "on"))
513 hrtimer_hres_enabled
= 1;
519 __setup("highres=", setup_hrtimer_hres
);
522 * hrtimer_high_res_enabled - query, if the highres mode is enabled
524 static inline int hrtimer_is_hres_enabled(void)
526 return hrtimer_hres_enabled
;
530 * Is the high resolution mode active ?
532 static inline int hrtimer_hres_active(void)
534 return __this_cpu_read(hrtimer_bases
.hres_active
);
538 * Reprogram the event source with checking both queues for the
540 * Called with interrupts disabled and base->lock held
543 hrtimer_force_reprogram(struct hrtimer_cpu_base
*cpu_base
, int skip_equal
)
546 struct hrtimer_clock_base
*base
= cpu_base
->clock_base
;
547 ktime_t expires
, expires_next
;
549 expires_next
.tv64
= KTIME_MAX
;
551 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++, base
++) {
552 struct hrtimer
*timer
;
553 struct timerqueue_node
*next
;
555 next
= timerqueue_getnext(&base
->active
);
558 timer
= container_of(next
, struct hrtimer
, node
);
560 expires
= ktime_sub(hrtimer_get_expires(timer
), base
->offset
);
562 * clock_was_set() has changed base->offset so the
563 * result might be negative. Fix it up to prevent a
564 * false positive in clockevents_program_event()
566 if (expires
.tv64
< 0)
568 if (expires
.tv64
< expires_next
.tv64
)
569 expires_next
= expires
;
572 if (skip_equal
&& expires_next
.tv64
== cpu_base
->expires_next
.tv64
)
575 cpu_base
->expires_next
.tv64
= expires_next
.tv64
;
578 * If a hang was detected in the last timer interrupt then we
579 * leave the hang delay active in the hardware. We want the
580 * system to make progress. That also prevents the following
582 * T1 expires 50ms from now
583 * T2 expires 5s from now
585 * T1 is removed, so this code is called and would reprogram
586 * the hardware to 5s from now. Any hrtimer_start after that
587 * will not reprogram the hardware due to hang_detected being
588 * set. So we'd effectivly block all timers until the T2 event
591 if (cpu_base
->hang_detected
)
594 if (cpu_base
->expires_next
.tv64
!= KTIME_MAX
)
595 tick_program_event(cpu_base
->expires_next
, 1);
599 * Shared reprogramming for clock_realtime and clock_monotonic
601 * When a timer is enqueued and expires earlier than the already enqueued
602 * timers, we have to check, whether it expires earlier than the timer for
603 * which the clock event device was armed.
605 * Called with interrupts disabled and base->cpu_base.lock held
607 static int hrtimer_reprogram(struct hrtimer
*timer
,
608 struct hrtimer_clock_base
*base
)
610 struct hrtimer_cpu_base
*cpu_base
= &__get_cpu_var(hrtimer_bases
);
611 ktime_t expires
= ktime_sub(hrtimer_get_expires(timer
), base
->offset
);
614 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer
) < 0);
617 * When the callback is running, we do not reprogram the clock event
618 * device. The timer callback is either running on a different CPU or
619 * the callback is executed in the hrtimer_interrupt context. The
620 * reprogramming is handled either by the softirq, which called the
621 * callback or at the end of the hrtimer_interrupt.
623 if (hrtimer_callback_running(timer
))
627 * CLOCK_REALTIME timer might be requested with an absolute
628 * expiry time which is less than base->offset. Nothing wrong
629 * about that, just avoid to call into the tick code, which
630 * has now objections against negative expiry values.
632 if (expires
.tv64
< 0)
635 if (expires
.tv64
>= cpu_base
->expires_next
.tv64
)
639 * If a hang was detected in the last timer interrupt then we
640 * do not schedule a timer which is earlier than the expiry
641 * which we enforced in the hang detection. We want the system
644 if (cpu_base
->hang_detected
)
648 * Clockevents returns -ETIME, when the event was in the past.
650 res
= tick_program_event(expires
, 0);
651 if (!IS_ERR_VALUE(res
))
652 cpu_base
->expires_next
= expires
;
657 * Initialize the high resolution related parts of cpu_base
659 static inline void hrtimer_init_hres(struct hrtimer_cpu_base
*base
)
661 base
->expires_next
.tv64
= KTIME_MAX
;
662 base
->hres_active
= 0;
666 * When High resolution timers are active, try to reprogram. Note, that in case
667 * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry
668 * check happens. The timer gets enqueued into the rbtree. The reprogramming
669 * and expiry check is done in the hrtimer_interrupt or in the softirq.
671 static inline int hrtimer_enqueue_reprogram(struct hrtimer
*timer
,
672 struct hrtimer_clock_base
*base
)
674 return base
->cpu_base
->hres_active
&& hrtimer_reprogram(timer
, base
);
677 static inline ktime_t
hrtimer_update_base(struct hrtimer_cpu_base
*base
)
679 ktime_t
*offs_real
= &base
->clock_base
[HRTIMER_BASE_REALTIME
].offset
;
680 ktime_t
*offs_boot
= &base
->clock_base
[HRTIMER_BASE_BOOTTIME
].offset
;
681 ktime_t
*offs_tai
= &base
->clock_base
[HRTIMER_BASE_TAI
].offset
;
683 return ktime_get_update_offsets(offs_real
, offs_boot
, offs_tai
);
687 * Retrigger next event is called after clock was set
689 * Called with interrupts disabled via on_each_cpu()
691 static void retrigger_next_event(void *arg
)
693 struct hrtimer_cpu_base
*base
= &__get_cpu_var(hrtimer_bases
);
695 if (!hrtimer_hres_active())
698 raw_spin_lock(&base
->lock
);
699 hrtimer_update_base(base
);
700 hrtimer_force_reprogram(base
, 0);
701 raw_spin_unlock(&base
->lock
);
705 * Switch to high resolution mode
707 static int hrtimer_switch_to_hres(void)
709 int i
, cpu
= smp_processor_id();
710 struct hrtimer_cpu_base
*base
= &per_cpu(hrtimer_bases
, cpu
);
713 if (base
->hres_active
)
716 local_irq_save(flags
);
718 if (tick_init_highres()) {
719 local_irq_restore(flags
);
720 printk(KERN_WARNING
"Could not switch to high resolution "
721 "mode on CPU %d\n", cpu
);
724 base
->hres_active
= 1;
725 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++)
726 base
->clock_base
[i
].resolution
= KTIME_HIGH_RES
;
728 tick_setup_sched_timer();
729 /* "Retrigger" the interrupt to get things going */
730 retrigger_next_event(NULL
);
731 local_irq_restore(flags
);
735 static void clock_was_set_work(struct work_struct
*work
)
740 static DECLARE_WORK(hrtimer_work
, clock_was_set_work
);
743 * Called from timekeeping and resume code to reprogramm the hrtimer
744 * interrupt device on all cpus.
746 void clock_was_set_delayed(void)
748 schedule_work(&hrtimer_work
);
753 static inline int hrtimer_hres_active(void) { return 0; }
754 static inline int hrtimer_is_hres_enabled(void) { return 0; }
755 static inline int hrtimer_switch_to_hres(void) { return 0; }
757 hrtimer_force_reprogram(struct hrtimer_cpu_base
*base
, int skip_equal
) { }
758 static inline int hrtimer_enqueue_reprogram(struct hrtimer
*timer
,
759 struct hrtimer_clock_base
*base
)
763 static inline void hrtimer_init_hres(struct hrtimer_cpu_base
*base
) { }
764 static inline void retrigger_next_event(void *arg
) { }
766 #endif /* CONFIG_HIGH_RES_TIMERS */
769 * Clock realtime was set
771 * Change the offset of the realtime clock vs. the monotonic
774 * We might have to reprogram the high resolution timer interrupt. On
775 * SMP we call the architecture specific code to retrigger _all_ high
776 * resolution timer interrupts. On UP we just disable interrupts and
777 * call the high resolution interrupt code.
779 void clock_was_set(void)
781 #ifdef CONFIG_HIGH_RES_TIMERS
782 /* Retrigger the CPU local events everywhere */
783 on_each_cpu(retrigger_next_event
, NULL
, 1);
785 timerfd_clock_was_set();
789 * During resume we might have to reprogram the high resolution timer
790 * interrupt on all online CPUs. However, all other CPUs will be
791 * stopped with IRQs interrupts disabled so the clock_was_set() call
794 void hrtimers_resume(void)
796 WARN_ONCE(!irqs_disabled(),
797 KERN_INFO
"hrtimers_resume() called with IRQs enabled!");
799 /* Retrigger on the local CPU */
800 retrigger_next_event(NULL
);
801 /* And schedule a retrigger for all others */
802 clock_was_set_delayed();
805 static inline void timer_stats_hrtimer_set_start_info(struct hrtimer
*timer
)
807 #ifdef CONFIG_TIMER_STATS
808 if (timer
->start_site
)
810 timer
->start_site
= __builtin_return_address(0);
811 memcpy(timer
->start_comm
, current
->comm
, TASK_COMM_LEN
);
812 timer
->start_pid
= current
->pid
;
816 static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer
*timer
)
818 #ifdef CONFIG_TIMER_STATS
819 timer
->start_site
= NULL
;
823 static inline void timer_stats_account_hrtimer(struct hrtimer
*timer
)
825 #ifdef CONFIG_TIMER_STATS
826 if (likely(!timer_stats_active
))
828 timer_stats_update_stats(timer
, timer
->start_pid
, timer
->start_site
,
829 timer
->function
, timer
->start_comm
, 0);
834 * Counterpart to lock_hrtimer_base above:
837 void unlock_hrtimer_base(const struct hrtimer
*timer
, unsigned long *flags
)
839 raw_spin_unlock_irqrestore(&timer
->base
->cpu_base
->lock
, *flags
);
843 * hrtimer_forward - forward the timer expiry
844 * @timer: hrtimer to forward
845 * @now: forward past this time
846 * @interval: the interval to forward
848 * Forward the timer expiry so it will expire in the future.
849 * Returns the number of overruns.
851 u64
hrtimer_forward(struct hrtimer
*timer
, ktime_t now
, ktime_t interval
)
856 delta
= ktime_sub(now
, hrtimer_get_expires(timer
));
861 if (interval
.tv64
< timer
->base
->resolution
.tv64
)
862 interval
.tv64
= timer
->base
->resolution
.tv64
;
864 if (unlikely(delta
.tv64
>= interval
.tv64
)) {
865 s64 incr
= ktime_to_ns(interval
);
867 orun
= ktime_divns(delta
, incr
);
868 hrtimer_add_expires_ns(timer
, incr
* orun
);
869 if (hrtimer_get_expires_tv64(timer
) > now
.tv64
)
872 * This (and the ktime_add() below) is the
873 * correction for exact:
877 hrtimer_add_expires(timer
, interval
);
881 EXPORT_SYMBOL_GPL(hrtimer_forward
);
884 * enqueue_hrtimer - internal function to (re)start a timer
886 * The timer is inserted in expiry order. Insertion into the
887 * red black tree is O(log(n)). Must hold the base lock.
889 * Returns 1 when the new timer is the leftmost timer in the tree.
891 static int enqueue_hrtimer(struct hrtimer
*timer
,
892 struct hrtimer_clock_base
*base
)
894 debug_activate(timer
);
896 timerqueue_add(&base
->active
, &timer
->node
);
897 base
->cpu_base
->active_bases
|= 1 << base
->index
;
900 * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
901 * state of a possibly running callback.
903 timer
->state
|= HRTIMER_STATE_ENQUEUED
;
905 return (&timer
->node
== base
->active
.next
);
909 * __remove_hrtimer - internal function to remove a timer
911 * Caller must hold the base lock.
913 * High resolution timer mode reprograms the clock event device when the
914 * timer is the one which expires next. The caller can disable this by setting
915 * reprogram to zero. This is useful, when the context does a reprogramming
916 * anyway (e.g. timer interrupt)
918 static void __remove_hrtimer(struct hrtimer
*timer
,
919 struct hrtimer_clock_base
*base
,
920 unsigned long newstate
, int reprogram
)
922 struct timerqueue_node
*next_timer
;
923 if (!(timer
->state
& HRTIMER_STATE_ENQUEUED
))
926 next_timer
= timerqueue_getnext(&base
->active
);
927 timerqueue_del(&base
->active
, &timer
->node
);
928 if (&timer
->node
== next_timer
) {
929 #ifdef CONFIG_HIGH_RES_TIMERS
930 /* Reprogram the clock event device. if enabled */
931 if (reprogram
&& hrtimer_hres_active()) {
934 expires
= ktime_sub(hrtimer_get_expires(timer
),
936 if (base
->cpu_base
->expires_next
.tv64
== expires
.tv64
)
937 hrtimer_force_reprogram(base
->cpu_base
, 1);
941 if (!timerqueue_getnext(&base
->active
))
942 base
->cpu_base
->active_bases
&= ~(1 << base
->index
);
944 timer
->state
= newstate
;
948 * remove hrtimer, called with base lock held
951 remove_hrtimer(struct hrtimer
*timer
, struct hrtimer_clock_base
*base
)
953 if (hrtimer_is_queued(timer
)) {
958 * Remove the timer and force reprogramming when high
959 * resolution mode is active and the timer is on the current
960 * CPU. If we remove a timer on another CPU, reprogramming is
961 * skipped. The interrupt event on this CPU is fired and
962 * reprogramming happens in the interrupt handler. This is a
963 * rare case and less expensive than a smp call.
965 debug_deactivate(timer
);
966 timer_stats_hrtimer_clear_start_info(timer
);
967 reprogram
= base
->cpu_base
== &__get_cpu_var(hrtimer_bases
);
969 * We must preserve the CALLBACK state flag here,
970 * otherwise we could move the timer base in
971 * switch_hrtimer_base.
973 state
= timer
->state
& HRTIMER_STATE_CALLBACK
;
974 __remove_hrtimer(timer
, base
, state
, reprogram
);
980 int __hrtimer_start_range_ns(struct hrtimer
*timer
, ktime_t tim
,
981 unsigned long delta_ns
, const enum hrtimer_mode mode
,
984 struct hrtimer_clock_base
*base
, *new_base
;
988 base
= lock_hrtimer_base(timer
, &flags
);
990 /* Remove an active timer from the queue: */
991 ret
= remove_hrtimer(timer
, base
);
993 if (mode
& HRTIMER_MODE_REL
) {
994 tim
= ktime_add_safe(tim
, base
->get_time());
996 * CONFIG_TIME_LOW_RES is a temporary way for architectures
997 * to signal that they simply return xtime in
998 * do_gettimeoffset(). In this case we want to round up by
999 * resolution when starting a relative timer, to avoid short
1000 * timeouts. This will go away with the GTOD framework.
1002 #ifdef CONFIG_TIME_LOW_RES
1003 tim
= ktime_add_safe(tim
, base
->resolution
);
1007 hrtimer_set_expires_range_ns(timer
, tim
, delta_ns
);
1009 /* Switch the timer base, if necessary: */
1010 new_base
= switch_hrtimer_base(timer
, base
, mode
& HRTIMER_MODE_PINNED
);
1012 timer_stats_hrtimer_set_start_info(timer
);
1014 leftmost
= enqueue_hrtimer(timer
, new_base
);
1017 * Only allow reprogramming if the new base is on this CPU.
1018 * (it might still be on another CPU if the timer was pending)
1020 * XXX send_remote_softirq() ?
1022 if (leftmost
&& new_base
->cpu_base
== &__get_cpu_var(hrtimer_bases
)
1023 && hrtimer_enqueue_reprogram(timer
, new_base
)) {
1026 * We need to drop cpu_base->lock to avoid a
1027 * lock ordering issue vs. rq->lock.
1029 raw_spin_unlock(&new_base
->cpu_base
->lock
);
1030 raise_softirq_irqoff(HRTIMER_SOFTIRQ
);
1031 local_irq_restore(flags
);
1034 __raise_softirq_irqoff(HRTIMER_SOFTIRQ
);
1038 unlock_hrtimer_base(timer
, &flags
);
1042 EXPORT_SYMBOL_GPL(__hrtimer_start_range_ns
);
1045 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
1046 * @timer: the timer to be added
1048 * @delta_ns: "slack" range for the timer
1049 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
1050 * relative (HRTIMER_MODE_REL)
1054 * 1 when the timer was active
1056 int hrtimer_start_range_ns(struct hrtimer
*timer
, ktime_t tim
,
1057 unsigned long delta_ns
, const enum hrtimer_mode mode
)
1059 return __hrtimer_start_range_ns(timer
, tim
, delta_ns
, mode
, 1);
1061 EXPORT_SYMBOL_GPL(hrtimer_start_range_ns
);
1064 * hrtimer_start - (re)start an hrtimer on the current CPU
1065 * @timer: the timer to be added
1067 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
1068 * relative (HRTIMER_MODE_REL)
1072 * 1 when the timer was active
1075 hrtimer_start(struct hrtimer
*timer
, ktime_t tim
, const enum hrtimer_mode mode
)
1077 return __hrtimer_start_range_ns(timer
, tim
, 0, mode
, 1);
1079 EXPORT_SYMBOL_GPL(hrtimer_start
);
1083 * hrtimer_try_to_cancel - try to deactivate a timer
1084 * @timer: hrtimer to stop
1087 * 0 when the timer was not active
1088 * 1 when the timer was active
1089 * -1 when the timer is currently excuting the callback function and
1092 int hrtimer_try_to_cancel(struct hrtimer
*timer
)
1094 struct hrtimer_clock_base
*base
;
1095 unsigned long flags
;
1098 base
= lock_hrtimer_base(timer
, &flags
);
1100 if (!hrtimer_callback_running(timer
))
1101 ret
= remove_hrtimer(timer
, base
);
1103 unlock_hrtimer_base(timer
, &flags
);
1108 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel
);
1111 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1112 * @timer: the timer to be cancelled
1115 * 0 when the timer was not active
1116 * 1 when the timer was active
1118 int hrtimer_cancel(struct hrtimer
*timer
)
1121 int ret
= hrtimer_try_to_cancel(timer
);
1128 EXPORT_SYMBOL_GPL(hrtimer_cancel
);
1131 * hrtimer_get_remaining - get remaining time for the timer
1132 * @timer: the timer to read
1134 ktime_t
hrtimer_get_remaining(const struct hrtimer
*timer
)
1136 unsigned long flags
;
1139 lock_hrtimer_base(timer
, &flags
);
1140 rem
= hrtimer_expires_remaining(timer
);
1141 unlock_hrtimer_base(timer
, &flags
);
1145 EXPORT_SYMBOL_GPL(hrtimer_get_remaining
);
1147 #ifdef CONFIG_NO_HZ_COMMON
1149 * hrtimer_get_next_event - get the time until next expiry event
1151 * Returns the delta to the next expiry event or KTIME_MAX if no timer
1154 ktime_t
hrtimer_get_next_event(void)
1156 struct hrtimer_cpu_base
*cpu_base
= &__get_cpu_var(hrtimer_bases
);
1157 struct hrtimer_clock_base
*base
= cpu_base
->clock_base
;
1158 ktime_t delta
, mindelta
= { .tv64
= KTIME_MAX
};
1159 unsigned long flags
;
1162 raw_spin_lock_irqsave(&cpu_base
->lock
, flags
);
1164 if (!hrtimer_hres_active()) {
1165 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++, base
++) {
1166 struct hrtimer
*timer
;
1167 struct timerqueue_node
*next
;
1169 next
= timerqueue_getnext(&base
->active
);
1173 timer
= container_of(next
, struct hrtimer
, node
);
1174 delta
.tv64
= hrtimer_get_expires_tv64(timer
);
1175 delta
= ktime_sub(delta
, base
->get_time());
1176 if (delta
.tv64
< mindelta
.tv64
)
1177 mindelta
.tv64
= delta
.tv64
;
1181 raw_spin_unlock_irqrestore(&cpu_base
->lock
, flags
);
1183 if (mindelta
.tv64
< 0)
1189 static void __hrtimer_init(struct hrtimer
*timer
, clockid_t clock_id
,
1190 enum hrtimer_mode mode
)
1192 struct hrtimer_cpu_base
*cpu_base
;
1195 memset(timer
, 0, sizeof(struct hrtimer
));
1197 cpu_base
= &__raw_get_cpu_var(hrtimer_bases
);
1199 if (clock_id
== CLOCK_REALTIME
&& mode
!= HRTIMER_MODE_ABS
)
1200 clock_id
= CLOCK_MONOTONIC
;
1202 base
= hrtimer_clockid_to_base(clock_id
);
1203 timer
->base
= &cpu_base
->clock_base
[base
];
1204 timerqueue_init(&timer
->node
);
1206 #ifdef CONFIG_TIMER_STATS
1207 timer
->start_site
= NULL
;
1208 timer
->start_pid
= -1;
1209 memset(timer
->start_comm
, 0, TASK_COMM_LEN
);
1214 * hrtimer_init - initialize a timer to the given clock
1215 * @timer: the timer to be initialized
1216 * @clock_id: the clock to be used
1217 * @mode: timer mode abs/rel
1219 void hrtimer_init(struct hrtimer
*timer
, clockid_t clock_id
,
1220 enum hrtimer_mode mode
)
1222 debug_init(timer
, clock_id
, mode
);
1223 __hrtimer_init(timer
, clock_id
, mode
);
1225 EXPORT_SYMBOL_GPL(hrtimer_init
);
1228 * hrtimer_get_res - get the timer resolution for a clock
1229 * @which_clock: which clock to query
1230 * @tp: pointer to timespec variable to store the resolution
1232 * Store the resolution of the clock selected by @which_clock in the
1233 * variable pointed to by @tp.
1235 int hrtimer_get_res(const clockid_t which_clock
, struct timespec
*tp
)
1237 struct hrtimer_cpu_base
*cpu_base
;
1238 int base
= hrtimer_clockid_to_base(which_clock
);
1240 cpu_base
= &__raw_get_cpu_var(hrtimer_bases
);
1241 *tp
= ktime_to_timespec(cpu_base
->clock_base
[base
].resolution
);
1245 EXPORT_SYMBOL_GPL(hrtimer_get_res
);
1247 static void __run_hrtimer(struct hrtimer
*timer
, ktime_t
*now
)
1249 struct hrtimer_clock_base
*base
= timer
->base
;
1250 struct hrtimer_cpu_base
*cpu_base
= base
->cpu_base
;
1251 enum hrtimer_restart (*fn
)(struct hrtimer
*);
1254 WARN_ON(!irqs_disabled());
1256 debug_deactivate(timer
);
1257 __remove_hrtimer(timer
, base
, HRTIMER_STATE_CALLBACK
, 0);
1258 timer_stats_account_hrtimer(timer
);
1259 fn
= timer
->function
;
1262 * Because we run timers from hardirq context, there is no chance
1263 * they get migrated to another cpu, therefore its safe to unlock
1266 raw_spin_unlock(&cpu_base
->lock
);
1267 trace_hrtimer_expire_entry(timer
, now
);
1268 restart
= fn(timer
);
1269 trace_hrtimer_expire_exit(timer
);
1270 raw_spin_lock(&cpu_base
->lock
);
1273 * Note: We clear the CALLBACK bit after enqueue_hrtimer and
1274 * we do not reprogramm the event hardware. Happens either in
1275 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1277 if (restart
!= HRTIMER_NORESTART
) {
1278 BUG_ON(timer
->state
!= HRTIMER_STATE_CALLBACK
);
1279 enqueue_hrtimer(timer
, base
);
1282 WARN_ON_ONCE(!(timer
->state
& HRTIMER_STATE_CALLBACK
));
1284 timer
->state
&= ~HRTIMER_STATE_CALLBACK
;
1287 #ifdef CONFIG_HIGH_RES_TIMERS
1290 * High resolution timer interrupt
1291 * Called with interrupts disabled
1293 void hrtimer_interrupt(struct clock_event_device
*dev
)
1295 struct hrtimer_cpu_base
*cpu_base
= &__get_cpu_var(hrtimer_bases
);
1296 ktime_t expires_next
, now
, entry_time
, delta
;
1299 BUG_ON(!cpu_base
->hres_active
);
1300 cpu_base
->nr_events
++;
1301 dev
->next_event
.tv64
= KTIME_MAX
;
1303 raw_spin_lock(&cpu_base
->lock
);
1304 entry_time
= now
= hrtimer_update_base(cpu_base
);
1306 expires_next
.tv64
= KTIME_MAX
;
1308 * We set expires_next to KTIME_MAX here with cpu_base->lock
1309 * held to prevent that a timer is enqueued in our queue via
1310 * the migration code. This does not affect enqueueing of
1311 * timers which run their callback and need to be requeued on
1314 cpu_base
->expires_next
.tv64
= KTIME_MAX
;
1316 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++) {
1317 struct hrtimer_clock_base
*base
;
1318 struct timerqueue_node
*node
;
1321 if (!(cpu_base
->active_bases
& (1 << i
)))
1324 base
= cpu_base
->clock_base
+ i
;
1325 basenow
= ktime_add(now
, base
->offset
);
1327 while ((node
= timerqueue_getnext(&base
->active
))) {
1328 struct hrtimer
*timer
;
1330 timer
= container_of(node
, struct hrtimer
, node
);
1333 * The immediate goal for using the softexpires is
1334 * minimizing wakeups, not running timers at the
1335 * earliest interrupt after their soft expiration.
1336 * This allows us to avoid using a Priority Search
1337 * Tree, which can answer a stabbing querry for
1338 * overlapping intervals and instead use the simple
1339 * BST we already have.
1340 * We don't add extra wakeups by delaying timers that
1341 * are right-of a not yet expired timer, because that
1342 * timer will have to trigger a wakeup anyway.
1345 if (basenow
.tv64
< hrtimer_get_softexpires_tv64(timer
)) {
1348 expires
= ktime_sub(hrtimer_get_expires(timer
),
1350 if (expires
.tv64
< 0)
1351 expires
.tv64
= KTIME_MAX
;
1352 if (expires
.tv64
< expires_next
.tv64
)
1353 expires_next
= expires
;
1357 __run_hrtimer(timer
, &basenow
);
1362 * Store the new expiry value so the migration code can verify
1365 cpu_base
->expires_next
= expires_next
;
1366 raw_spin_unlock(&cpu_base
->lock
);
1368 /* Reprogramming necessary ? */
1369 if (expires_next
.tv64
== KTIME_MAX
||
1370 !tick_program_event(expires_next
, 0)) {
1371 cpu_base
->hang_detected
= 0;
1376 * The next timer was already expired due to:
1378 * - long lasting callbacks
1379 * - being scheduled away when running in a VM
1381 * We need to prevent that we loop forever in the hrtimer
1382 * interrupt routine. We give it 3 attempts to avoid
1383 * overreacting on some spurious event.
1385 * Acquire base lock for updating the offsets and retrieving
1388 raw_spin_lock(&cpu_base
->lock
);
1389 now
= hrtimer_update_base(cpu_base
);
1390 cpu_base
->nr_retries
++;
1394 * Give the system a chance to do something else than looping
1395 * here. We stored the entry time, so we know exactly how long
1396 * we spent here. We schedule the next event this amount of
1399 cpu_base
->nr_hangs
++;
1400 cpu_base
->hang_detected
= 1;
1401 raw_spin_unlock(&cpu_base
->lock
);
1402 delta
= ktime_sub(now
, entry_time
);
1403 if (delta
.tv64
> cpu_base
->max_hang_time
.tv64
)
1404 cpu_base
->max_hang_time
= delta
;
1406 * Limit it to a sensible value as we enforce a longer
1407 * delay. Give the CPU at least 100ms to catch up.
1409 if (delta
.tv64
> 100 * NSEC_PER_MSEC
)
1410 expires_next
= ktime_add_ns(now
, 100 * NSEC_PER_MSEC
);
1412 expires_next
= ktime_add(now
, delta
);
1413 tick_program_event(expires_next
, 1);
1414 printk_once(KERN_WARNING
"hrtimer: interrupt took %llu ns\n",
1415 ktime_to_ns(delta
));
1419 * local version of hrtimer_peek_ahead_timers() called with interrupts
1422 static void __hrtimer_peek_ahead_timers(void)
1424 struct tick_device
*td
;
1426 if (!hrtimer_hres_active())
1429 td
= &__get_cpu_var(tick_cpu_device
);
1430 if (td
&& td
->evtdev
)
1431 hrtimer_interrupt(td
->evtdev
);
1435 * hrtimer_peek_ahead_timers -- run soft-expired timers now
1437 * hrtimer_peek_ahead_timers will peek at the timer queue of
1438 * the current cpu and check if there are any timers for which
1439 * the soft expires time has passed. If any such timers exist,
1440 * they are run immediately and then removed from the timer queue.
1443 void hrtimer_peek_ahead_timers(void)
1445 unsigned long flags
;
1447 local_irq_save(flags
);
1448 __hrtimer_peek_ahead_timers();
1449 local_irq_restore(flags
);
1452 static void run_hrtimer_softirq(struct softirq_action
*h
)
1454 hrtimer_peek_ahead_timers();
1457 #else /* CONFIG_HIGH_RES_TIMERS */
1459 static inline void __hrtimer_peek_ahead_timers(void) { }
1461 #endif /* !CONFIG_HIGH_RES_TIMERS */
1464 * Called from timer softirq every jiffy, expire hrtimers:
1466 * For HRT its the fall back code to run the softirq in the timer
1467 * softirq context in case the hrtimer initialization failed or has
1468 * not been done yet.
1470 void hrtimer_run_pending(void)
1472 if (hrtimer_hres_active())
1476 * This _is_ ugly: We have to check in the softirq context,
1477 * whether we can switch to highres and / or nohz mode. The
1478 * clocksource switch happens in the timer interrupt with
1479 * xtime_lock held. Notification from there only sets the
1480 * check bit in the tick_oneshot code, otherwise we might
1481 * deadlock vs. xtime_lock.
1483 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1484 hrtimer_switch_to_hres();
1488 * Called from hardirq context every jiffy
1490 void hrtimer_run_queues(void)
1492 struct timerqueue_node
*node
;
1493 struct hrtimer_cpu_base
*cpu_base
= &__get_cpu_var(hrtimer_bases
);
1494 struct hrtimer_clock_base
*base
;
1495 int index
, gettime
= 1;
1497 if (hrtimer_hres_active())
1500 for (index
= 0; index
< HRTIMER_MAX_CLOCK_BASES
; index
++) {
1501 base
= &cpu_base
->clock_base
[index
];
1502 if (!timerqueue_getnext(&base
->active
))
1506 hrtimer_get_softirq_time(cpu_base
);
1510 raw_spin_lock(&cpu_base
->lock
);
1512 while ((node
= timerqueue_getnext(&base
->active
))) {
1513 struct hrtimer
*timer
;
1515 timer
= container_of(node
, struct hrtimer
, node
);
1516 if (base
->softirq_time
.tv64
<=
1517 hrtimer_get_expires_tv64(timer
))
1520 __run_hrtimer(timer
, &base
->softirq_time
);
1522 raw_spin_unlock(&cpu_base
->lock
);
1527 * Sleep related functions:
1529 static enum hrtimer_restart
hrtimer_wakeup(struct hrtimer
*timer
)
1531 struct hrtimer_sleeper
*t
=
1532 container_of(timer
, struct hrtimer_sleeper
, timer
);
1533 struct task_struct
*task
= t
->task
;
1537 wake_up_process(task
);
1539 return HRTIMER_NORESTART
;
1542 void hrtimer_init_sleeper(struct hrtimer_sleeper
*sl
, struct task_struct
*task
)
1544 sl
->timer
.function
= hrtimer_wakeup
;
1547 EXPORT_SYMBOL_GPL(hrtimer_init_sleeper
);
1549 static int __sched
do_nanosleep(struct hrtimer_sleeper
*t
, enum hrtimer_mode mode
)
1551 hrtimer_init_sleeper(t
, current
);
1554 set_current_state(TASK_INTERRUPTIBLE
);
1555 hrtimer_start_expires(&t
->timer
, mode
);
1556 if (!hrtimer_active(&t
->timer
))
1559 if (likely(t
->task
))
1560 freezable_schedule();
1562 hrtimer_cancel(&t
->timer
);
1563 mode
= HRTIMER_MODE_ABS
;
1565 } while (t
->task
&& !signal_pending(current
));
1567 __set_current_state(TASK_RUNNING
);
1569 return t
->task
== NULL
;
1572 static int update_rmtp(struct hrtimer
*timer
, struct timespec __user
*rmtp
)
1574 struct timespec rmt
;
1577 rem
= hrtimer_expires_remaining(timer
);
1580 rmt
= ktime_to_timespec(rem
);
1582 if (copy_to_user(rmtp
, &rmt
, sizeof(*rmtp
)))
1588 long __sched
hrtimer_nanosleep_restart(struct restart_block
*restart
)
1590 struct hrtimer_sleeper t
;
1591 struct timespec __user
*rmtp
;
1594 hrtimer_init_on_stack(&t
.timer
, restart
->nanosleep
.clockid
,
1596 hrtimer_set_expires_tv64(&t
.timer
, restart
->nanosleep
.expires
);
1598 if (do_nanosleep(&t
, HRTIMER_MODE_ABS
))
1601 rmtp
= restart
->nanosleep
.rmtp
;
1603 ret
= update_rmtp(&t
.timer
, rmtp
);
1608 /* The other values in restart are already filled in */
1609 ret
= -ERESTART_RESTARTBLOCK
;
1611 destroy_hrtimer_on_stack(&t
.timer
);
1615 long hrtimer_nanosleep(struct timespec
*rqtp
, struct timespec __user
*rmtp
,
1616 const enum hrtimer_mode mode
, const clockid_t clockid
)
1618 struct restart_block
*restart
;
1619 struct hrtimer_sleeper t
;
1621 unsigned long slack
;
1623 slack
= current
->timer_slack_ns
;
1624 if (dl_task(current
) || rt_task(current
))
1627 hrtimer_init_on_stack(&t
.timer
, clockid
, mode
);
1628 hrtimer_set_expires_range_ns(&t
.timer
, timespec_to_ktime(*rqtp
), slack
);
1629 if (do_nanosleep(&t
, mode
))
1632 /* Absolute timers do not update the rmtp value and restart: */
1633 if (mode
== HRTIMER_MODE_ABS
) {
1634 ret
= -ERESTARTNOHAND
;
1639 ret
= update_rmtp(&t
.timer
, rmtp
);
1644 restart
= ¤t_thread_info()->restart_block
;
1645 restart
->fn
= hrtimer_nanosleep_restart
;
1646 restart
->nanosleep
.clockid
= t
.timer
.base
->clockid
;
1647 restart
->nanosleep
.rmtp
= rmtp
;
1648 restart
->nanosleep
.expires
= hrtimer_get_expires_tv64(&t
.timer
);
1650 ret
= -ERESTART_RESTARTBLOCK
;
1652 destroy_hrtimer_on_stack(&t
.timer
);
1656 SYSCALL_DEFINE2(nanosleep
, struct timespec __user
*, rqtp
,
1657 struct timespec __user
*, rmtp
)
1661 if (copy_from_user(&tu
, rqtp
, sizeof(tu
)))
1664 if (!timespec_valid(&tu
))
1667 return hrtimer_nanosleep(&tu
, rmtp
, HRTIMER_MODE_REL
, CLOCK_MONOTONIC
);
1671 * Functions related to boot-time initialization:
1673 static void init_hrtimers_cpu(int cpu
)
1675 struct hrtimer_cpu_base
*cpu_base
= &per_cpu(hrtimer_bases
, cpu
);
1678 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++) {
1679 cpu_base
->clock_base
[i
].cpu_base
= cpu_base
;
1680 timerqueue_init_head(&cpu_base
->clock_base
[i
].active
);
1683 hrtimer_init_hres(cpu_base
);
1686 #ifdef CONFIG_HOTPLUG_CPU
1688 static void migrate_hrtimer_list(struct hrtimer_clock_base
*old_base
,
1689 struct hrtimer_clock_base
*new_base
)
1691 struct hrtimer
*timer
;
1692 struct timerqueue_node
*node
;
1694 while ((node
= timerqueue_getnext(&old_base
->active
))) {
1695 timer
= container_of(node
, struct hrtimer
, node
);
1696 BUG_ON(hrtimer_callback_running(timer
));
1697 debug_deactivate(timer
);
1700 * Mark it as STATE_MIGRATE not INACTIVE otherwise the
1701 * timer could be seen as !active and just vanish away
1702 * under us on another CPU
1704 __remove_hrtimer(timer
, old_base
, HRTIMER_STATE_MIGRATE
, 0);
1705 timer
->base
= new_base
;
1707 * Enqueue the timers on the new cpu. This does not
1708 * reprogram the event device in case the timer
1709 * expires before the earliest on this CPU, but we run
1710 * hrtimer_interrupt after we migrated everything to
1711 * sort out already expired timers and reprogram the
1714 enqueue_hrtimer(timer
, new_base
);
1716 /* Clear the migration state bit */
1717 timer
->state
&= ~HRTIMER_STATE_MIGRATE
;
1721 static void migrate_hrtimers(int scpu
)
1723 struct hrtimer_cpu_base
*old_base
, *new_base
;
1726 BUG_ON(cpu_online(scpu
));
1727 tick_cancel_sched_timer(scpu
);
1729 local_irq_disable();
1730 old_base
= &per_cpu(hrtimer_bases
, scpu
);
1731 new_base
= &__get_cpu_var(hrtimer_bases
);
1733 * The caller is globally serialized and nobody else
1734 * takes two locks at once, deadlock is not possible.
1736 raw_spin_lock(&new_base
->lock
);
1737 raw_spin_lock_nested(&old_base
->lock
, SINGLE_DEPTH_NESTING
);
1739 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++) {
1740 migrate_hrtimer_list(&old_base
->clock_base
[i
],
1741 &new_base
->clock_base
[i
]);
1744 raw_spin_unlock(&old_base
->lock
);
1745 raw_spin_unlock(&new_base
->lock
);
1747 /* Check, if we got expired work to do */
1748 __hrtimer_peek_ahead_timers();
1752 #endif /* CONFIG_HOTPLUG_CPU */
1754 static int hrtimer_cpu_notify(struct notifier_block
*self
,
1755 unsigned long action
, void *hcpu
)
1757 int scpu
= (long)hcpu
;
1761 case CPU_UP_PREPARE
:
1762 case CPU_UP_PREPARE_FROZEN
:
1763 init_hrtimers_cpu(scpu
);
1766 #ifdef CONFIG_HOTPLUG_CPU
1768 case CPU_DYING_FROZEN
:
1769 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DYING
, &scpu
);
1772 case CPU_DEAD_FROZEN
:
1774 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD
, &scpu
);
1775 migrate_hrtimers(scpu
);
1787 static struct notifier_block hrtimers_nb
= {
1788 .notifier_call
= hrtimer_cpu_notify
,
1791 void __init
hrtimers_init(void)
1793 hrtimer_cpu_notify(&hrtimers_nb
, (unsigned long)CPU_UP_PREPARE
,
1794 (void *)(long)smp_processor_id());
1795 register_cpu_notifier(&hrtimers_nb
);
1796 #ifdef CONFIG_HIGH_RES_TIMERS
1797 open_softirq(HRTIMER_SOFTIRQ
, run_hrtimer_softirq
);
1802 * schedule_hrtimeout_range_clock - sleep until timeout
1803 * @expires: timeout value (ktime_t)
1804 * @delta: slack in expires timeout (ktime_t)
1805 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1806 * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME
1809 schedule_hrtimeout_range_clock(ktime_t
*expires
, unsigned long delta
,
1810 const enum hrtimer_mode mode
, int clock
)
1812 struct hrtimer_sleeper t
;
1815 * Optimize when a zero timeout value is given. It does not
1816 * matter whether this is an absolute or a relative time.
1818 if (expires
&& !expires
->tv64
) {
1819 __set_current_state(TASK_RUNNING
);
1824 * A NULL parameter means "infinite"
1828 __set_current_state(TASK_RUNNING
);
1832 hrtimer_init_on_stack(&t
.timer
, clock
, mode
);
1833 hrtimer_set_expires_range_ns(&t
.timer
, *expires
, delta
);
1835 hrtimer_init_sleeper(&t
, current
);
1837 hrtimer_start_expires(&t
.timer
, mode
);
1838 if (!hrtimer_active(&t
.timer
))
1844 hrtimer_cancel(&t
.timer
);
1845 destroy_hrtimer_on_stack(&t
.timer
);
1847 __set_current_state(TASK_RUNNING
);
1849 return !t
.task
? 0 : -EINTR
;
1853 * schedule_hrtimeout_range - sleep until timeout
1854 * @expires: timeout value (ktime_t)
1855 * @delta: slack in expires timeout (ktime_t)
1856 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1858 * Make the current task sleep until the given expiry time has
1859 * elapsed. The routine will return immediately unless
1860 * the current task state has been set (see set_current_state()).
1862 * The @delta argument gives the kernel the freedom to schedule the
1863 * actual wakeup to a time that is both power and performance friendly.
1864 * The kernel give the normal best effort behavior for "@expires+@delta",
1865 * but may decide to fire the timer earlier, but no earlier than @expires.
1867 * You can set the task state as follows -
1869 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1870 * pass before the routine returns.
1872 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1873 * delivered to the current task.
1875 * The current task state is guaranteed to be TASK_RUNNING when this
1878 * Returns 0 when the timer has expired otherwise -EINTR
1880 int __sched
schedule_hrtimeout_range(ktime_t
*expires
, unsigned long delta
,
1881 const enum hrtimer_mode mode
)
1883 return schedule_hrtimeout_range_clock(expires
, delta
, mode
,
1886 EXPORT_SYMBOL_GPL(schedule_hrtimeout_range
);
1889 * schedule_hrtimeout - sleep until timeout
1890 * @expires: timeout value (ktime_t)
1891 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1893 * Make the current task sleep until the given expiry time has
1894 * elapsed. The routine will return immediately unless
1895 * the current task state has been set (see set_current_state()).
1897 * You can set the task state as follows -
1899 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1900 * pass before the routine returns.
1902 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1903 * delivered to the current task.
1905 * The current task state is guaranteed to be TASK_RUNNING when this
1908 * Returns 0 when the timer has expired otherwise -EINTR
1910 int __sched
schedule_hrtimeout(ktime_t
*expires
,
1911 const enum hrtimer_mode mode
)
1913 return schedule_hrtimeout_range(expires
, 0, mode
);
1915 EXPORT_SYMBOL_GPL(schedule_hrtimeout
);