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/timer.h>
51 #include <asm/uaccess.h>
53 #include <trace/events/timer.h>
58 * There are more clockids then hrtimer bases. Thus, we index
59 * into the timer bases by the hrtimer_base_type enum. When trying
60 * to reach a base using a clockid, hrtimer_clockid_to_base()
61 * is used to convert from clockid to the proper hrtimer_base_type.
63 DEFINE_PER_CPU(struct hrtimer_cpu_base
, hrtimer_bases
) =
66 .lock
= __RAW_SPIN_LOCK_UNLOCKED(hrtimer_bases
.lock
),
70 .index
= HRTIMER_BASE_MONOTONIC
,
71 .clockid
= CLOCK_MONOTONIC
,
72 .get_time
= &ktime_get
,
73 .resolution
= KTIME_LOW_RES
,
76 .index
= HRTIMER_BASE_REALTIME
,
77 .clockid
= CLOCK_REALTIME
,
78 .get_time
= &ktime_get_real
,
79 .resolution
= KTIME_LOW_RES
,
82 .index
= HRTIMER_BASE_BOOTTIME
,
83 .clockid
= CLOCK_BOOTTIME
,
84 .get_time
= &ktime_get_boottime
,
85 .resolution
= KTIME_LOW_RES
,
88 .index
= HRTIMER_BASE_TAI
,
90 .get_time
= &ktime_get_clocktai
,
91 .resolution
= KTIME_LOW_RES
,
96 static const int hrtimer_clock_to_base_table
[MAX_CLOCKS
] = {
97 [CLOCK_REALTIME
] = HRTIMER_BASE_REALTIME
,
98 [CLOCK_MONOTONIC
] = HRTIMER_BASE_MONOTONIC
,
99 [CLOCK_BOOTTIME
] = HRTIMER_BASE_BOOTTIME
,
100 [CLOCK_TAI
] = HRTIMER_BASE_TAI
,
103 static inline int hrtimer_clockid_to_base(clockid_t clock_id
)
105 return hrtimer_clock_to_base_table
[clock_id
];
110 * Get the coarse grained time at the softirq based on xtime and
113 static void hrtimer_get_softirq_time(struct hrtimer_cpu_base
*base
)
115 ktime_t xtim
, mono
, boot
;
116 struct timespec xts
, tom
, slp
;
119 get_xtime_and_monotonic_and_sleep_offset(&xts
, &tom
, &slp
);
120 tai_offset
= timekeeping_get_tai_offset();
122 xtim
= timespec_to_ktime(xts
);
123 mono
= ktime_add(xtim
, timespec_to_ktime(tom
));
124 boot
= ktime_add(mono
, timespec_to_ktime(slp
));
125 base
->clock_base
[HRTIMER_BASE_REALTIME
].softirq_time
= xtim
;
126 base
->clock_base
[HRTIMER_BASE_MONOTONIC
].softirq_time
= mono
;
127 base
->clock_base
[HRTIMER_BASE_BOOTTIME
].softirq_time
= boot
;
128 base
->clock_base
[HRTIMER_BASE_TAI
].softirq_time
=
129 ktime_add(xtim
, ktime_set(tai_offset
, 0));
133 * Functions and macros which are different for UP/SMP systems are kept in a
139 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
140 * means that all timers which are tied to this base via timer->base are
141 * locked, and the base itself is locked too.
143 * So __run_timers/migrate_timers can safely modify all timers which could
144 * be found on the lists/queues.
146 * When the timer's base is locked, and the timer removed from list, it is
147 * possible to set timer->base = NULL and drop the lock: the timer remains
151 struct hrtimer_clock_base
*lock_hrtimer_base(const struct hrtimer
*timer
,
152 unsigned long *flags
)
154 struct hrtimer_clock_base
*base
;
158 if (likely(base
!= NULL
)) {
159 raw_spin_lock_irqsave(&base
->cpu_base
->lock
, *flags
);
160 if (likely(base
== timer
->base
))
162 /* The timer has migrated to another CPU: */
163 raw_spin_unlock_irqrestore(&base
->cpu_base
->lock
, *flags
);
171 * Get the preferred target CPU for NOHZ
173 static int hrtimer_get_target(int this_cpu
, int pinned
)
175 #ifdef CONFIG_NO_HZ_COMMON
176 if (!pinned
&& get_sysctl_timer_migration() && idle_cpu(this_cpu
))
177 return get_nohz_timer_target();
183 * With HIGHRES=y we do not migrate the timer when it is expiring
184 * before the next event on the target cpu because we cannot reprogram
185 * the target cpu hardware and we would cause it to fire late.
187 * Called with cpu_base->lock of target cpu held.
190 hrtimer_check_target(struct hrtimer
*timer
, struct hrtimer_clock_base
*new_base
)
192 #ifdef CONFIG_HIGH_RES_TIMERS
195 if (!new_base
->cpu_base
->hres_active
)
198 expires
= ktime_sub(hrtimer_get_expires(timer
), new_base
->offset
);
199 return expires
.tv64
<= new_base
->cpu_base
->expires_next
.tv64
;
206 * Switch the timer base to the current CPU when possible.
208 static inline struct hrtimer_clock_base
*
209 switch_hrtimer_base(struct hrtimer
*timer
, struct hrtimer_clock_base
*base
,
212 struct hrtimer_clock_base
*new_base
;
213 struct hrtimer_cpu_base
*new_cpu_base
;
214 int this_cpu
= smp_processor_id();
215 int cpu
= hrtimer_get_target(this_cpu
, pinned
);
216 int basenum
= base
->index
;
219 new_cpu_base
= &per_cpu(hrtimer_bases
, cpu
);
220 new_base
= &new_cpu_base
->clock_base
[basenum
];
222 if (base
!= new_base
) {
224 * We are trying to move timer to new_base.
225 * However we can't change timer's base while it is running,
226 * so we keep it on the same CPU. No hassle vs. reprogramming
227 * the event source in the high resolution case. The softirq
228 * code will take care of this when the timer function has
229 * completed. There is no conflict as we hold the lock until
230 * the timer is enqueued.
232 if (unlikely(hrtimer_callback_running(timer
)))
235 /* See the comment in lock_timer_base() */
237 raw_spin_unlock(&base
->cpu_base
->lock
);
238 raw_spin_lock(&new_base
->cpu_base
->lock
);
240 if (cpu
!= this_cpu
&& hrtimer_check_target(timer
, new_base
)) {
242 raw_spin_unlock(&new_base
->cpu_base
->lock
);
243 raw_spin_lock(&base
->cpu_base
->lock
);
247 timer
->base
= new_base
;
249 if (cpu
!= this_cpu
&& hrtimer_check_target(timer
, new_base
)) {
257 #else /* CONFIG_SMP */
259 static inline struct hrtimer_clock_base
*
260 lock_hrtimer_base(const struct hrtimer
*timer
, unsigned long *flags
)
262 struct hrtimer_clock_base
*base
= timer
->base
;
264 raw_spin_lock_irqsave(&base
->cpu_base
->lock
, *flags
);
269 # define switch_hrtimer_base(t, b, p) (b)
271 #endif /* !CONFIG_SMP */
274 * Functions for the union type storage format of ktime_t which are
275 * too large for inlining:
277 #if BITS_PER_LONG < 64
278 # ifndef CONFIG_KTIME_SCALAR
280 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
282 * @nsec: the scalar nsec value to add
284 * Returns the sum of kt and nsec in ktime_t format
286 ktime_t
ktime_add_ns(const ktime_t kt
, u64 nsec
)
290 if (likely(nsec
< NSEC_PER_SEC
)) {
293 unsigned long rem
= do_div(nsec
, NSEC_PER_SEC
);
295 /* Make sure nsec fits into long */
296 if (unlikely(nsec
> KTIME_SEC_MAX
))
297 return (ktime_t
){ .tv64
= KTIME_MAX
};
299 tmp
= ktime_set((long)nsec
, rem
);
302 return ktime_add(kt
, tmp
);
305 EXPORT_SYMBOL_GPL(ktime_add_ns
);
308 * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable
310 * @nsec: the scalar nsec value to subtract
312 * Returns the subtraction of @nsec from @kt in ktime_t format
314 ktime_t
ktime_sub_ns(const ktime_t kt
, u64 nsec
)
318 if (likely(nsec
< NSEC_PER_SEC
)) {
321 unsigned long rem
= do_div(nsec
, NSEC_PER_SEC
);
323 tmp
= ktime_set((long)nsec
, rem
);
326 return ktime_sub(kt
, tmp
);
329 EXPORT_SYMBOL_GPL(ktime_sub_ns
);
330 # endif /* !CONFIG_KTIME_SCALAR */
333 * Divide a ktime value by a nanosecond value
335 u64
ktime_divns(const ktime_t kt
, s64 div
)
340 dclc
= ktime_to_ns(kt
);
341 /* Make sure the divisor is less than 2^32: */
347 do_div(dclc
, (unsigned long) div
);
351 #endif /* BITS_PER_LONG >= 64 */
354 * Add two ktime values and do a safety check for overflow:
356 ktime_t
ktime_add_safe(const ktime_t lhs
, const ktime_t rhs
)
358 ktime_t res
= ktime_add(lhs
, rhs
);
361 * We use KTIME_SEC_MAX here, the maximum timeout which we can
362 * return to user space in a timespec:
364 if (res
.tv64
< 0 || res
.tv64
< lhs
.tv64
|| res
.tv64
< rhs
.tv64
)
365 res
= ktime_set(KTIME_SEC_MAX
, 0);
370 EXPORT_SYMBOL_GPL(ktime_add_safe
);
372 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
374 static struct debug_obj_descr hrtimer_debug_descr
;
376 static void *hrtimer_debug_hint(void *addr
)
378 return ((struct hrtimer
*) addr
)->function
;
382 * fixup_init is called when:
383 * - an active object is initialized
385 static int hrtimer_fixup_init(void *addr
, enum debug_obj_state state
)
387 struct hrtimer
*timer
= addr
;
390 case ODEBUG_STATE_ACTIVE
:
391 hrtimer_cancel(timer
);
392 debug_object_init(timer
, &hrtimer_debug_descr
);
400 * fixup_activate is called when:
401 * - an active object is activated
402 * - an unknown object is activated (might be a statically initialized object)
404 static int hrtimer_fixup_activate(void *addr
, enum debug_obj_state state
)
408 case ODEBUG_STATE_NOTAVAILABLE
:
412 case ODEBUG_STATE_ACTIVE
:
421 * fixup_free is called when:
422 * - an active object is freed
424 static int hrtimer_fixup_free(void *addr
, enum debug_obj_state state
)
426 struct hrtimer
*timer
= addr
;
429 case ODEBUG_STATE_ACTIVE
:
430 hrtimer_cancel(timer
);
431 debug_object_free(timer
, &hrtimer_debug_descr
);
438 static struct debug_obj_descr hrtimer_debug_descr
= {
440 .debug_hint
= hrtimer_debug_hint
,
441 .fixup_init
= hrtimer_fixup_init
,
442 .fixup_activate
= hrtimer_fixup_activate
,
443 .fixup_free
= hrtimer_fixup_free
,
446 static inline void debug_hrtimer_init(struct hrtimer
*timer
)
448 debug_object_init(timer
, &hrtimer_debug_descr
);
451 static inline void debug_hrtimer_activate(struct hrtimer
*timer
)
453 debug_object_activate(timer
, &hrtimer_debug_descr
);
456 static inline void debug_hrtimer_deactivate(struct hrtimer
*timer
)
458 debug_object_deactivate(timer
, &hrtimer_debug_descr
);
461 static inline void debug_hrtimer_free(struct hrtimer
*timer
)
463 debug_object_free(timer
, &hrtimer_debug_descr
);
466 static void __hrtimer_init(struct hrtimer
*timer
, clockid_t clock_id
,
467 enum hrtimer_mode mode
);
469 void hrtimer_init_on_stack(struct hrtimer
*timer
, clockid_t clock_id
,
470 enum hrtimer_mode mode
)
472 debug_object_init_on_stack(timer
, &hrtimer_debug_descr
);
473 __hrtimer_init(timer
, clock_id
, mode
);
475 EXPORT_SYMBOL_GPL(hrtimer_init_on_stack
);
477 void destroy_hrtimer_on_stack(struct hrtimer
*timer
)
479 debug_object_free(timer
, &hrtimer_debug_descr
);
483 static inline void debug_hrtimer_init(struct hrtimer
*timer
) { }
484 static inline void debug_hrtimer_activate(struct hrtimer
*timer
) { }
485 static inline void debug_hrtimer_deactivate(struct hrtimer
*timer
) { }
489 debug_init(struct hrtimer
*timer
, clockid_t clockid
,
490 enum hrtimer_mode mode
)
492 debug_hrtimer_init(timer
);
493 trace_hrtimer_init(timer
, clockid
, mode
);
496 static inline void debug_activate(struct hrtimer
*timer
)
498 debug_hrtimer_activate(timer
);
499 trace_hrtimer_start(timer
);
502 static inline void debug_deactivate(struct hrtimer
*timer
)
504 debug_hrtimer_deactivate(timer
);
505 trace_hrtimer_cancel(timer
);
508 /* High resolution timer related functions */
509 #ifdef CONFIG_HIGH_RES_TIMERS
512 * High resolution timer enabled ?
514 static int hrtimer_hres_enabled __read_mostly
= 1;
517 * Enable / Disable high resolution mode
519 static int __init
setup_hrtimer_hres(char *str
)
521 if (!strcmp(str
, "off"))
522 hrtimer_hres_enabled
= 0;
523 else if (!strcmp(str
, "on"))
524 hrtimer_hres_enabled
= 1;
530 __setup("highres=", setup_hrtimer_hres
);
533 * hrtimer_high_res_enabled - query, if the highres mode is enabled
535 static inline int hrtimer_is_hres_enabled(void)
537 return hrtimer_hres_enabled
;
541 * Is the high resolution mode active ?
543 static inline int hrtimer_hres_active(void)
545 return __this_cpu_read(hrtimer_bases
.hres_active
);
549 * Reprogram the event source with checking both queues for the
551 * Called with interrupts disabled and base->lock held
554 hrtimer_force_reprogram(struct hrtimer_cpu_base
*cpu_base
, int skip_equal
)
557 struct hrtimer_clock_base
*base
= cpu_base
->clock_base
;
558 ktime_t expires
, expires_next
;
560 expires_next
.tv64
= KTIME_MAX
;
562 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++, base
++) {
563 struct hrtimer
*timer
;
564 struct timerqueue_node
*next
;
566 next
= timerqueue_getnext(&base
->active
);
569 timer
= container_of(next
, struct hrtimer
, node
);
571 expires
= ktime_sub(hrtimer_get_expires(timer
), base
->offset
);
573 * clock_was_set() has changed base->offset so the
574 * result might be negative. Fix it up to prevent a
575 * false positive in clockevents_program_event()
577 if (expires
.tv64
< 0)
579 if (expires
.tv64
< expires_next
.tv64
)
580 expires_next
= expires
;
583 if (skip_equal
&& expires_next
.tv64
== cpu_base
->expires_next
.tv64
)
586 cpu_base
->expires_next
.tv64
= expires_next
.tv64
;
589 * If a hang was detected in the last timer interrupt then we
590 * leave the hang delay active in the hardware. We want the
591 * system to make progress. That also prevents the following
593 * T1 expires 50ms from now
594 * T2 expires 5s from now
596 * T1 is removed, so this code is called and would reprogram
597 * the hardware to 5s from now. Any hrtimer_start after that
598 * will not reprogram the hardware due to hang_detected being
599 * set. So we'd effectivly block all timers until the T2 event
602 if (cpu_base
->hang_detected
)
605 if (cpu_base
->expires_next
.tv64
!= KTIME_MAX
)
606 tick_program_event(cpu_base
->expires_next
, 1);
610 * Shared reprogramming for clock_realtime and clock_monotonic
612 * When a timer is enqueued and expires earlier than the already enqueued
613 * timers, we have to check, whether it expires earlier than the timer for
614 * which the clock event device was armed.
616 * Called with interrupts disabled and base->cpu_base.lock held
618 static int hrtimer_reprogram(struct hrtimer
*timer
,
619 struct hrtimer_clock_base
*base
)
621 struct hrtimer_cpu_base
*cpu_base
= &__get_cpu_var(hrtimer_bases
);
622 ktime_t expires
= ktime_sub(hrtimer_get_expires(timer
), base
->offset
);
625 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer
) < 0);
628 * When the callback is running, we do not reprogram the clock event
629 * device. The timer callback is either running on a different CPU or
630 * the callback is executed in the hrtimer_interrupt context. The
631 * reprogramming is handled either by the softirq, which called the
632 * callback or at the end of the hrtimer_interrupt.
634 if (hrtimer_callback_running(timer
))
638 * CLOCK_REALTIME timer might be requested with an absolute
639 * expiry time which is less than base->offset. Nothing wrong
640 * about that, just avoid to call into the tick code, which
641 * has now objections against negative expiry values.
643 if (expires
.tv64
< 0)
646 if (expires
.tv64
>= cpu_base
->expires_next
.tv64
)
650 * If a hang was detected in the last timer interrupt then we
651 * do not schedule a timer which is earlier than the expiry
652 * which we enforced in the hang detection. We want the system
655 if (cpu_base
->hang_detected
)
659 * Clockevents returns -ETIME, when the event was in the past.
661 res
= tick_program_event(expires
, 0);
662 if (!IS_ERR_VALUE(res
))
663 cpu_base
->expires_next
= expires
;
668 * Initialize the high resolution related parts of cpu_base
670 static inline void hrtimer_init_hres(struct hrtimer_cpu_base
*base
)
672 base
->expires_next
.tv64
= KTIME_MAX
;
673 base
->hres_active
= 0;
677 * When High resolution timers are active, try to reprogram. Note, that in case
678 * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry
679 * check happens. The timer gets enqueued into the rbtree. The reprogramming
680 * and expiry check is done in the hrtimer_interrupt or in the softirq.
682 static inline int hrtimer_enqueue_reprogram(struct hrtimer
*timer
,
683 struct hrtimer_clock_base
*base
)
685 return base
->cpu_base
->hres_active
&& hrtimer_reprogram(timer
, base
);
688 static inline ktime_t
hrtimer_update_base(struct hrtimer_cpu_base
*base
)
690 ktime_t
*offs_real
= &base
->clock_base
[HRTIMER_BASE_REALTIME
].offset
;
691 ktime_t
*offs_boot
= &base
->clock_base
[HRTIMER_BASE_BOOTTIME
].offset
;
692 ktime_t
*offs_tai
= &base
->clock_base
[HRTIMER_BASE_TAI
].offset
;
694 return ktime_get_update_offsets(offs_real
, offs_boot
, offs_tai
);
698 * Retrigger next event is called after clock was set
700 * Called with interrupts disabled via on_each_cpu()
702 static void retrigger_next_event(void *arg
)
704 struct hrtimer_cpu_base
*base
= &__get_cpu_var(hrtimer_bases
);
706 if (!hrtimer_hres_active())
709 raw_spin_lock(&base
->lock
);
710 hrtimer_update_base(base
);
711 hrtimer_force_reprogram(base
, 0);
712 raw_spin_unlock(&base
->lock
);
716 * Switch to high resolution mode
718 static int hrtimer_switch_to_hres(void)
720 int i
, cpu
= smp_processor_id();
721 struct hrtimer_cpu_base
*base
= &per_cpu(hrtimer_bases
, cpu
);
724 if (base
->hres_active
)
727 local_irq_save(flags
);
729 if (tick_init_highres()) {
730 local_irq_restore(flags
);
731 printk(KERN_WARNING
"Could not switch to high resolution "
732 "mode on CPU %d\n", cpu
);
735 base
->hres_active
= 1;
736 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++)
737 base
->clock_base
[i
].resolution
= KTIME_HIGH_RES
;
739 tick_setup_sched_timer();
740 /* "Retrigger" the interrupt to get things going */
741 retrigger_next_event(NULL
);
742 local_irq_restore(flags
);
746 static void clock_was_set_work(struct work_struct
*work
)
751 static DECLARE_WORK(hrtimer_work
, clock_was_set_work
);
754 * Called from timekeeping and resume code to reprogramm the hrtimer
755 * interrupt device on all cpus.
757 void clock_was_set_delayed(void)
759 schedule_work(&hrtimer_work
);
764 static inline int hrtimer_hres_active(void) { return 0; }
765 static inline int hrtimer_is_hres_enabled(void) { return 0; }
766 static inline int hrtimer_switch_to_hres(void) { return 0; }
768 hrtimer_force_reprogram(struct hrtimer_cpu_base
*base
, int skip_equal
) { }
769 static inline int hrtimer_enqueue_reprogram(struct hrtimer
*timer
,
770 struct hrtimer_clock_base
*base
)
774 static inline void hrtimer_init_hres(struct hrtimer_cpu_base
*base
) { }
775 static inline void retrigger_next_event(void *arg
) { }
777 #endif /* CONFIG_HIGH_RES_TIMERS */
780 * Clock realtime was set
782 * Change the offset of the realtime clock vs. the monotonic
785 * We might have to reprogram the high resolution timer interrupt. On
786 * SMP we call the architecture specific code to retrigger _all_ high
787 * resolution timer interrupts. On UP we just disable interrupts and
788 * call the high resolution interrupt code.
790 void clock_was_set(void)
792 #ifdef CONFIG_HIGH_RES_TIMERS
793 /* Retrigger the CPU local events everywhere */
794 on_each_cpu(retrigger_next_event
, NULL
, 1);
796 timerfd_clock_was_set();
800 * During resume we might have to reprogram the high resolution timer
801 * interrupt (on the local CPU):
803 void hrtimers_resume(void)
805 WARN_ONCE(!irqs_disabled(),
806 KERN_INFO
"hrtimers_resume() called with IRQs enabled!");
808 /* Retrigger on the local CPU */
809 retrigger_next_event(NULL
);
810 /* And schedule a retrigger for all others */
811 clock_was_set_delayed();
814 static inline void timer_stats_hrtimer_set_start_info(struct hrtimer
*timer
)
816 #ifdef CONFIG_TIMER_STATS
817 if (timer
->start_site
)
819 timer
->start_site
= __builtin_return_address(0);
820 memcpy(timer
->start_comm
, current
->comm
, TASK_COMM_LEN
);
821 timer
->start_pid
= current
->pid
;
825 static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer
*timer
)
827 #ifdef CONFIG_TIMER_STATS
828 timer
->start_site
= NULL
;
832 static inline void timer_stats_account_hrtimer(struct hrtimer
*timer
)
834 #ifdef CONFIG_TIMER_STATS
835 if (likely(!timer_stats_active
))
837 timer_stats_update_stats(timer
, timer
->start_pid
, timer
->start_site
,
838 timer
->function
, timer
->start_comm
, 0);
843 * Counterpart to lock_hrtimer_base above:
846 void unlock_hrtimer_base(const struct hrtimer
*timer
, unsigned long *flags
)
848 raw_spin_unlock_irqrestore(&timer
->base
->cpu_base
->lock
, *flags
);
852 * hrtimer_forward - forward the timer expiry
853 * @timer: hrtimer to forward
854 * @now: forward past this time
855 * @interval: the interval to forward
857 * Forward the timer expiry so it will expire in the future.
858 * Returns the number of overruns.
860 u64
hrtimer_forward(struct hrtimer
*timer
, ktime_t now
, ktime_t interval
)
865 delta
= ktime_sub(now
, hrtimer_get_expires(timer
));
870 if (interval
.tv64
< timer
->base
->resolution
.tv64
)
871 interval
.tv64
= timer
->base
->resolution
.tv64
;
873 if (unlikely(delta
.tv64
>= interval
.tv64
)) {
874 s64 incr
= ktime_to_ns(interval
);
876 orun
= ktime_divns(delta
, incr
);
877 hrtimer_add_expires_ns(timer
, incr
* orun
);
878 if (hrtimer_get_expires_tv64(timer
) > now
.tv64
)
881 * This (and the ktime_add() below) is the
882 * correction for exact:
886 hrtimer_add_expires(timer
, interval
);
890 EXPORT_SYMBOL_GPL(hrtimer_forward
);
893 * enqueue_hrtimer - internal function to (re)start a timer
895 * The timer is inserted in expiry order. Insertion into the
896 * red black tree is O(log(n)). Must hold the base lock.
898 * Returns 1 when the new timer is the leftmost timer in the tree.
900 static int enqueue_hrtimer(struct hrtimer
*timer
,
901 struct hrtimer_clock_base
*base
)
903 debug_activate(timer
);
905 timerqueue_add(&base
->active
, &timer
->node
);
906 base
->cpu_base
->active_bases
|= 1 << base
->index
;
909 * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
910 * state of a possibly running callback.
912 timer
->state
|= HRTIMER_STATE_ENQUEUED
;
914 return (&timer
->node
== base
->active
.next
);
918 * __remove_hrtimer - internal function to remove a timer
920 * Caller must hold the base lock.
922 * High resolution timer mode reprograms the clock event device when the
923 * timer is the one which expires next. The caller can disable this by setting
924 * reprogram to zero. This is useful, when the context does a reprogramming
925 * anyway (e.g. timer interrupt)
927 static void __remove_hrtimer(struct hrtimer
*timer
,
928 struct hrtimer_clock_base
*base
,
929 unsigned long newstate
, int reprogram
)
931 struct timerqueue_node
*next_timer
;
932 if (!(timer
->state
& HRTIMER_STATE_ENQUEUED
))
935 next_timer
= timerqueue_getnext(&base
->active
);
936 timerqueue_del(&base
->active
, &timer
->node
);
937 if (&timer
->node
== next_timer
) {
938 #ifdef CONFIG_HIGH_RES_TIMERS
939 /* Reprogram the clock event device. if enabled */
940 if (reprogram
&& hrtimer_hres_active()) {
943 expires
= ktime_sub(hrtimer_get_expires(timer
),
945 if (base
->cpu_base
->expires_next
.tv64
== expires
.tv64
)
946 hrtimer_force_reprogram(base
->cpu_base
, 1);
950 if (!timerqueue_getnext(&base
->active
))
951 base
->cpu_base
->active_bases
&= ~(1 << base
->index
);
953 timer
->state
= newstate
;
957 * remove hrtimer, called with base lock held
960 remove_hrtimer(struct hrtimer
*timer
, struct hrtimer_clock_base
*base
)
962 if (hrtimer_is_queued(timer
)) {
967 * Remove the timer and force reprogramming when high
968 * resolution mode is active and the timer is on the current
969 * CPU. If we remove a timer on another CPU, reprogramming is
970 * skipped. The interrupt event on this CPU is fired and
971 * reprogramming happens in the interrupt handler. This is a
972 * rare case and less expensive than a smp call.
974 debug_deactivate(timer
);
975 timer_stats_hrtimer_clear_start_info(timer
);
976 reprogram
= base
->cpu_base
== &__get_cpu_var(hrtimer_bases
);
978 * We must preserve the CALLBACK state flag here,
979 * otherwise we could move the timer base in
980 * switch_hrtimer_base.
982 state
= timer
->state
& HRTIMER_STATE_CALLBACK
;
983 __remove_hrtimer(timer
, base
, state
, reprogram
);
989 int __hrtimer_start_range_ns(struct hrtimer
*timer
, ktime_t tim
,
990 unsigned long delta_ns
, const enum hrtimer_mode mode
,
993 struct hrtimer_clock_base
*base
, *new_base
;
997 base
= lock_hrtimer_base(timer
, &flags
);
999 /* Remove an active timer from the queue: */
1000 ret
= remove_hrtimer(timer
, base
);
1002 if (mode
& HRTIMER_MODE_REL
) {
1003 tim
= ktime_add_safe(tim
, base
->get_time());
1005 * CONFIG_TIME_LOW_RES is a temporary way for architectures
1006 * to signal that they simply return xtime in
1007 * do_gettimeoffset(). In this case we want to round up by
1008 * resolution when starting a relative timer, to avoid short
1009 * timeouts. This will go away with the GTOD framework.
1011 #ifdef CONFIG_TIME_LOW_RES
1012 tim
= ktime_add_safe(tim
, base
->resolution
);
1016 hrtimer_set_expires_range_ns(timer
, tim
, delta_ns
);
1018 /* Switch the timer base, if necessary: */
1019 new_base
= switch_hrtimer_base(timer
, base
, mode
& HRTIMER_MODE_PINNED
);
1021 timer_stats_hrtimer_set_start_info(timer
);
1023 leftmost
= enqueue_hrtimer(timer
, new_base
);
1026 * Only allow reprogramming if the new base is on this CPU.
1027 * (it might still be on another CPU if the timer was pending)
1029 * XXX send_remote_softirq() ?
1031 if (leftmost
&& new_base
->cpu_base
== &__get_cpu_var(hrtimer_bases
)
1032 && hrtimer_enqueue_reprogram(timer
, new_base
)) {
1035 * We need to drop cpu_base->lock to avoid a
1036 * lock ordering issue vs. rq->lock.
1038 raw_spin_unlock(&new_base
->cpu_base
->lock
);
1039 raise_softirq_irqoff(HRTIMER_SOFTIRQ
);
1040 local_irq_restore(flags
);
1043 __raise_softirq_irqoff(HRTIMER_SOFTIRQ
);
1047 unlock_hrtimer_base(timer
, &flags
);
1053 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
1054 * @timer: the timer to be added
1056 * @delta_ns: "slack" range for the timer
1057 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
1058 * relative (HRTIMER_MODE_REL)
1062 * 1 when the timer was active
1064 int hrtimer_start_range_ns(struct hrtimer
*timer
, ktime_t tim
,
1065 unsigned long delta_ns
, const enum hrtimer_mode mode
)
1067 return __hrtimer_start_range_ns(timer
, tim
, delta_ns
, mode
, 1);
1069 EXPORT_SYMBOL_GPL(hrtimer_start_range_ns
);
1072 * hrtimer_start - (re)start an hrtimer on the current CPU
1073 * @timer: the timer to be added
1075 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
1076 * relative (HRTIMER_MODE_REL)
1080 * 1 when the timer was active
1083 hrtimer_start(struct hrtimer
*timer
, ktime_t tim
, const enum hrtimer_mode mode
)
1085 return __hrtimer_start_range_ns(timer
, tim
, 0, mode
, 1);
1087 EXPORT_SYMBOL_GPL(hrtimer_start
);
1091 * hrtimer_try_to_cancel - try to deactivate a timer
1092 * @timer: hrtimer to stop
1095 * 0 when the timer was not active
1096 * 1 when the timer was active
1097 * -1 when the timer is currently excuting the callback function and
1100 int hrtimer_try_to_cancel(struct hrtimer
*timer
)
1102 struct hrtimer_clock_base
*base
;
1103 unsigned long flags
;
1106 base
= lock_hrtimer_base(timer
, &flags
);
1108 if (!hrtimer_callback_running(timer
))
1109 ret
= remove_hrtimer(timer
, base
);
1111 unlock_hrtimer_base(timer
, &flags
);
1116 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel
);
1119 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1120 * @timer: the timer to be cancelled
1123 * 0 when the timer was not active
1124 * 1 when the timer was active
1126 int hrtimer_cancel(struct hrtimer
*timer
)
1129 int ret
= hrtimer_try_to_cancel(timer
);
1136 EXPORT_SYMBOL_GPL(hrtimer_cancel
);
1139 * hrtimer_get_remaining - get remaining time for the timer
1140 * @timer: the timer to read
1142 ktime_t
hrtimer_get_remaining(const struct hrtimer
*timer
)
1144 unsigned long flags
;
1147 lock_hrtimer_base(timer
, &flags
);
1148 rem
= hrtimer_expires_remaining(timer
);
1149 unlock_hrtimer_base(timer
, &flags
);
1153 EXPORT_SYMBOL_GPL(hrtimer_get_remaining
);
1155 #ifdef CONFIG_NO_HZ_COMMON
1157 * hrtimer_get_next_event - get the time until next expiry event
1159 * Returns the delta to the next expiry event or KTIME_MAX if no timer
1162 ktime_t
hrtimer_get_next_event(void)
1164 struct hrtimer_cpu_base
*cpu_base
= &__get_cpu_var(hrtimer_bases
);
1165 struct hrtimer_clock_base
*base
= cpu_base
->clock_base
;
1166 ktime_t delta
, mindelta
= { .tv64
= KTIME_MAX
};
1167 unsigned long flags
;
1170 raw_spin_lock_irqsave(&cpu_base
->lock
, flags
);
1172 if (!hrtimer_hres_active()) {
1173 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++, base
++) {
1174 struct hrtimer
*timer
;
1175 struct timerqueue_node
*next
;
1177 next
= timerqueue_getnext(&base
->active
);
1181 timer
= container_of(next
, struct hrtimer
, node
);
1182 delta
.tv64
= hrtimer_get_expires_tv64(timer
);
1183 delta
= ktime_sub(delta
, base
->get_time());
1184 if (delta
.tv64
< mindelta
.tv64
)
1185 mindelta
.tv64
= delta
.tv64
;
1189 raw_spin_unlock_irqrestore(&cpu_base
->lock
, flags
);
1191 if (mindelta
.tv64
< 0)
1197 static void __hrtimer_init(struct hrtimer
*timer
, clockid_t clock_id
,
1198 enum hrtimer_mode mode
)
1200 struct hrtimer_cpu_base
*cpu_base
;
1203 memset(timer
, 0, sizeof(struct hrtimer
));
1205 cpu_base
= &__raw_get_cpu_var(hrtimer_bases
);
1207 if (clock_id
== CLOCK_REALTIME
&& mode
!= HRTIMER_MODE_ABS
)
1208 clock_id
= CLOCK_MONOTONIC
;
1210 base
= hrtimer_clockid_to_base(clock_id
);
1211 timer
->base
= &cpu_base
->clock_base
[base
];
1212 timerqueue_init(&timer
->node
);
1214 #ifdef CONFIG_TIMER_STATS
1215 timer
->start_site
= NULL
;
1216 timer
->start_pid
= -1;
1217 memset(timer
->start_comm
, 0, TASK_COMM_LEN
);
1222 * hrtimer_init - initialize a timer to the given clock
1223 * @timer: the timer to be initialized
1224 * @clock_id: the clock to be used
1225 * @mode: timer mode abs/rel
1227 void hrtimer_init(struct hrtimer
*timer
, clockid_t clock_id
,
1228 enum hrtimer_mode mode
)
1230 debug_init(timer
, clock_id
, mode
);
1231 __hrtimer_init(timer
, clock_id
, mode
);
1233 EXPORT_SYMBOL_GPL(hrtimer_init
);
1236 * hrtimer_get_res - get the timer resolution for a clock
1237 * @which_clock: which clock to query
1238 * @tp: pointer to timespec variable to store the resolution
1240 * Store the resolution of the clock selected by @which_clock in the
1241 * variable pointed to by @tp.
1243 int hrtimer_get_res(const clockid_t which_clock
, struct timespec
*tp
)
1245 struct hrtimer_cpu_base
*cpu_base
;
1246 int base
= hrtimer_clockid_to_base(which_clock
);
1248 cpu_base
= &__raw_get_cpu_var(hrtimer_bases
);
1249 *tp
= ktime_to_timespec(cpu_base
->clock_base
[base
].resolution
);
1253 EXPORT_SYMBOL_GPL(hrtimer_get_res
);
1255 static void __run_hrtimer(struct hrtimer
*timer
, ktime_t
*now
)
1257 struct hrtimer_clock_base
*base
= timer
->base
;
1258 struct hrtimer_cpu_base
*cpu_base
= base
->cpu_base
;
1259 enum hrtimer_restart (*fn
)(struct hrtimer
*);
1262 WARN_ON(!irqs_disabled());
1264 debug_deactivate(timer
);
1265 __remove_hrtimer(timer
, base
, HRTIMER_STATE_CALLBACK
, 0);
1266 timer_stats_account_hrtimer(timer
);
1267 fn
= timer
->function
;
1270 * Because we run timers from hardirq context, there is no chance
1271 * they get migrated to another cpu, therefore its safe to unlock
1274 raw_spin_unlock(&cpu_base
->lock
);
1275 trace_hrtimer_expire_entry(timer
, now
);
1276 restart
= fn(timer
);
1277 trace_hrtimer_expire_exit(timer
);
1278 raw_spin_lock(&cpu_base
->lock
);
1281 * Note: We clear the CALLBACK bit after enqueue_hrtimer and
1282 * we do not reprogramm the event hardware. Happens either in
1283 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1285 if (restart
!= HRTIMER_NORESTART
) {
1286 BUG_ON(timer
->state
!= HRTIMER_STATE_CALLBACK
);
1287 enqueue_hrtimer(timer
, base
);
1290 WARN_ON_ONCE(!(timer
->state
& HRTIMER_STATE_CALLBACK
));
1292 timer
->state
&= ~HRTIMER_STATE_CALLBACK
;
1295 #ifdef CONFIG_HIGH_RES_TIMERS
1298 * High resolution timer interrupt
1299 * Called with interrupts disabled
1301 void hrtimer_interrupt(struct clock_event_device
*dev
)
1303 struct hrtimer_cpu_base
*cpu_base
= &__get_cpu_var(hrtimer_bases
);
1304 ktime_t expires_next
, now
, entry_time
, delta
;
1307 BUG_ON(!cpu_base
->hres_active
);
1308 cpu_base
->nr_events
++;
1309 dev
->next_event
.tv64
= KTIME_MAX
;
1311 raw_spin_lock(&cpu_base
->lock
);
1312 entry_time
= now
= hrtimer_update_base(cpu_base
);
1314 expires_next
.tv64
= KTIME_MAX
;
1316 * We set expires_next to KTIME_MAX here with cpu_base->lock
1317 * held to prevent that a timer is enqueued in our queue via
1318 * the migration code. This does not affect enqueueing of
1319 * timers which run their callback and need to be requeued on
1322 cpu_base
->expires_next
.tv64
= KTIME_MAX
;
1324 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++) {
1325 struct hrtimer_clock_base
*base
;
1326 struct timerqueue_node
*node
;
1329 if (!(cpu_base
->active_bases
& (1 << i
)))
1332 base
= cpu_base
->clock_base
+ i
;
1333 basenow
= ktime_add(now
, base
->offset
);
1335 while ((node
= timerqueue_getnext(&base
->active
))) {
1336 struct hrtimer
*timer
;
1338 timer
= container_of(node
, struct hrtimer
, node
);
1341 * The immediate goal for using the softexpires is
1342 * minimizing wakeups, not running timers at the
1343 * earliest interrupt after their soft expiration.
1344 * This allows us to avoid using a Priority Search
1345 * Tree, which can answer a stabbing querry for
1346 * overlapping intervals and instead use the simple
1347 * BST we already have.
1348 * We don't add extra wakeups by delaying timers that
1349 * are right-of a not yet expired timer, because that
1350 * timer will have to trigger a wakeup anyway.
1353 if (basenow
.tv64
< hrtimer_get_softexpires_tv64(timer
)) {
1356 expires
= ktime_sub(hrtimer_get_expires(timer
),
1358 if (expires
.tv64
< 0)
1359 expires
.tv64
= KTIME_MAX
;
1360 if (expires
.tv64
< expires_next
.tv64
)
1361 expires_next
= expires
;
1365 __run_hrtimer(timer
, &basenow
);
1370 * Store the new expiry value so the migration code can verify
1373 cpu_base
->expires_next
= expires_next
;
1374 raw_spin_unlock(&cpu_base
->lock
);
1376 /* Reprogramming necessary ? */
1377 if (expires_next
.tv64
== KTIME_MAX
||
1378 !tick_program_event(expires_next
, 0)) {
1379 cpu_base
->hang_detected
= 0;
1384 * The next timer was already expired due to:
1386 * - long lasting callbacks
1387 * - being scheduled away when running in a VM
1389 * We need to prevent that we loop forever in the hrtimer
1390 * interrupt routine. We give it 3 attempts to avoid
1391 * overreacting on some spurious event.
1393 * Acquire base lock for updating the offsets and retrieving
1396 raw_spin_lock(&cpu_base
->lock
);
1397 now
= hrtimer_update_base(cpu_base
);
1398 cpu_base
->nr_retries
++;
1402 * Give the system a chance to do something else than looping
1403 * here. We stored the entry time, so we know exactly how long
1404 * we spent here. We schedule the next event this amount of
1407 cpu_base
->nr_hangs
++;
1408 cpu_base
->hang_detected
= 1;
1409 raw_spin_unlock(&cpu_base
->lock
);
1410 delta
= ktime_sub(now
, entry_time
);
1411 if (delta
.tv64
> cpu_base
->max_hang_time
.tv64
)
1412 cpu_base
->max_hang_time
= delta
;
1414 * Limit it to a sensible value as we enforce a longer
1415 * delay. Give the CPU at least 100ms to catch up.
1417 if (delta
.tv64
> 100 * NSEC_PER_MSEC
)
1418 expires_next
= ktime_add_ns(now
, 100 * NSEC_PER_MSEC
);
1420 expires_next
= ktime_add(now
, delta
);
1421 tick_program_event(expires_next
, 1);
1422 printk_once(KERN_WARNING
"hrtimer: interrupt took %llu ns\n",
1423 ktime_to_ns(delta
));
1427 * local version of hrtimer_peek_ahead_timers() called with interrupts
1430 static void __hrtimer_peek_ahead_timers(void)
1432 struct tick_device
*td
;
1434 if (!hrtimer_hres_active())
1437 td
= &__get_cpu_var(tick_cpu_device
);
1438 if (td
&& td
->evtdev
)
1439 hrtimer_interrupt(td
->evtdev
);
1443 * hrtimer_peek_ahead_timers -- run soft-expired timers now
1445 * hrtimer_peek_ahead_timers will peek at the timer queue of
1446 * the current cpu and check if there are any timers for which
1447 * the soft expires time has passed. If any such timers exist,
1448 * they are run immediately and then removed from the timer queue.
1451 void hrtimer_peek_ahead_timers(void)
1453 unsigned long flags
;
1455 local_irq_save(flags
);
1456 __hrtimer_peek_ahead_timers();
1457 local_irq_restore(flags
);
1460 static void run_hrtimer_softirq(struct softirq_action
*h
)
1462 hrtimer_peek_ahead_timers();
1465 #else /* CONFIG_HIGH_RES_TIMERS */
1467 static inline void __hrtimer_peek_ahead_timers(void) { }
1469 #endif /* !CONFIG_HIGH_RES_TIMERS */
1472 * Called from timer softirq every jiffy, expire hrtimers:
1474 * For HRT its the fall back code to run the softirq in the timer
1475 * softirq context in case the hrtimer initialization failed or has
1476 * not been done yet.
1478 void hrtimer_run_pending(void)
1480 if (hrtimer_hres_active())
1484 * This _is_ ugly: We have to check in the softirq context,
1485 * whether we can switch to highres and / or nohz mode. The
1486 * clocksource switch happens in the timer interrupt with
1487 * xtime_lock held. Notification from there only sets the
1488 * check bit in the tick_oneshot code, otherwise we might
1489 * deadlock vs. xtime_lock.
1491 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1492 hrtimer_switch_to_hres();
1496 * Called from hardirq context every jiffy
1498 void hrtimer_run_queues(void)
1500 struct timerqueue_node
*node
;
1501 struct hrtimer_cpu_base
*cpu_base
= &__get_cpu_var(hrtimer_bases
);
1502 struct hrtimer_clock_base
*base
;
1503 int index
, gettime
= 1;
1505 if (hrtimer_hres_active())
1508 for (index
= 0; index
< HRTIMER_MAX_CLOCK_BASES
; index
++) {
1509 base
= &cpu_base
->clock_base
[index
];
1510 if (!timerqueue_getnext(&base
->active
))
1514 hrtimer_get_softirq_time(cpu_base
);
1518 raw_spin_lock(&cpu_base
->lock
);
1520 while ((node
= timerqueue_getnext(&base
->active
))) {
1521 struct hrtimer
*timer
;
1523 timer
= container_of(node
, struct hrtimer
, node
);
1524 if (base
->softirq_time
.tv64
<=
1525 hrtimer_get_expires_tv64(timer
))
1528 __run_hrtimer(timer
, &base
->softirq_time
);
1530 raw_spin_unlock(&cpu_base
->lock
);
1535 * Sleep related functions:
1537 static enum hrtimer_restart
hrtimer_wakeup(struct hrtimer
*timer
)
1539 struct hrtimer_sleeper
*t
=
1540 container_of(timer
, struct hrtimer_sleeper
, timer
);
1541 struct task_struct
*task
= t
->task
;
1545 wake_up_process(task
);
1547 return HRTIMER_NORESTART
;
1550 void hrtimer_init_sleeper(struct hrtimer_sleeper
*sl
, struct task_struct
*task
)
1552 sl
->timer
.function
= hrtimer_wakeup
;
1555 EXPORT_SYMBOL_GPL(hrtimer_init_sleeper
);
1557 static int __sched
do_nanosleep(struct hrtimer_sleeper
*t
, enum hrtimer_mode mode
)
1559 hrtimer_init_sleeper(t
, current
);
1562 set_current_state(TASK_INTERRUPTIBLE
);
1563 hrtimer_start_expires(&t
->timer
, mode
);
1564 if (!hrtimer_active(&t
->timer
))
1567 if (likely(t
->task
))
1570 hrtimer_cancel(&t
->timer
);
1571 mode
= HRTIMER_MODE_ABS
;
1573 } while (t
->task
&& !signal_pending(current
));
1575 __set_current_state(TASK_RUNNING
);
1577 return t
->task
== NULL
;
1580 static int update_rmtp(struct hrtimer
*timer
, struct timespec __user
*rmtp
)
1582 struct timespec rmt
;
1585 rem
= hrtimer_expires_remaining(timer
);
1588 rmt
= ktime_to_timespec(rem
);
1590 if (copy_to_user(rmtp
, &rmt
, sizeof(*rmtp
)))
1596 long __sched
hrtimer_nanosleep_restart(struct restart_block
*restart
)
1598 struct hrtimer_sleeper t
;
1599 struct timespec __user
*rmtp
;
1602 hrtimer_init_on_stack(&t
.timer
, restart
->nanosleep
.clockid
,
1604 hrtimer_set_expires_tv64(&t
.timer
, restart
->nanosleep
.expires
);
1606 if (do_nanosleep(&t
, HRTIMER_MODE_ABS
))
1609 rmtp
= restart
->nanosleep
.rmtp
;
1611 ret
= update_rmtp(&t
.timer
, rmtp
);
1616 /* The other values in restart are already filled in */
1617 ret
= -ERESTART_RESTARTBLOCK
;
1619 destroy_hrtimer_on_stack(&t
.timer
);
1623 long hrtimer_nanosleep(struct timespec
*rqtp
, struct timespec __user
*rmtp
,
1624 const enum hrtimer_mode mode
, const clockid_t clockid
)
1626 struct restart_block
*restart
;
1627 struct hrtimer_sleeper t
;
1629 unsigned long slack
;
1631 slack
= current
->timer_slack_ns
;
1632 if (rt_task(current
))
1635 hrtimer_init_on_stack(&t
.timer
, clockid
, mode
);
1636 hrtimer_set_expires_range_ns(&t
.timer
, timespec_to_ktime(*rqtp
), slack
);
1637 if (do_nanosleep(&t
, mode
))
1640 /* Absolute timers do not update the rmtp value and restart: */
1641 if (mode
== HRTIMER_MODE_ABS
) {
1642 ret
= -ERESTARTNOHAND
;
1647 ret
= update_rmtp(&t
.timer
, rmtp
);
1652 restart
= ¤t_thread_info()->restart_block
;
1653 restart
->fn
= hrtimer_nanosleep_restart
;
1654 restart
->nanosleep
.clockid
= t
.timer
.base
->clockid
;
1655 restart
->nanosleep
.rmtp
= rmtp
;
1656 restart
->nanosleep
.expires
= hrtimer_get_expires_tv64(&t
.timer
);
1658 ret
= -ERESTART_RESTARTBLOCK
;
1660 destroy_hrtimer_on_stack(&t
.timer
);
1664 SYSCALL_DEFINE2(nanosleep
, struct timespec __user
*, rqtp
,
1665 struct timespec __user
*, rmtp
)
1669 if (copy_from_user(&tu
, rqtp
, sizeof(tu
)))
1672 if (!timespec_valid(&tu
))
1675 return hrtimer_nanosleep(&tu
, rmtp
, HRTIMER_MODE_REL
, CLOCK_MONOTONIC
);
1679 * Functions related to boot-time initialization:
1681 static void __cpuinit
init_hrtimers_cpu(int cpu
)
1683 struct hrtimer_cpu_base
*cpu_base
= &per_cpu(hrtimer_bases
, cpu
);
1686 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++) {
1687 cpu_base
->clock_base
[i
].cpu_base
= cpu_base
;
1688 timerqueue_init_head(&cpu_base
->clock_base
[i
].active
);
1691 hrtimer_init_hres(cpu_base
);
1694 #ifdef CONFIG_HOTPLUG_CPU
1696 static void migrate_hrtimer_list(struct hrtimer_clock_base
*old_base
,
1697 struct hrtimer_clock_base
*new_base
)
1699 struct hrtimer
*timer
;
1700 struct timerqueue_node
*node
;
1702 while ((node
= timerqueue_getnext(&old_base
->active
))) {
1703 timer
= container_of(node
, struct hrtimer
, node
);
1704 BUG_ON(hrtimer_callback_running(timer
));
1705 debug_deactivate(timer
);
1708 * Mark it as STATE_MIGRATE not INACTIVE otherwise the
1709 * timer could be seen as !active and just vanish away
1710 * under us on another CPU
1712 __remove_hrtimer(timer
, old_base
, HRTIMER_STATE_MIGRATE
, 0);
1713 timer
->base
= new_base
;
1715 * Enqueue the timers on the new cpu. This does not
1716 * reprogram the event device in case the timer
1717 * expires before the earliest on this CPU, but we run
1718 * hrtimer_interrupt after we migrated everything to
1719 * sort out already expired timers and reprogram the
1722 enqueue_hrtimer(timer
, new_base
);
1724 /* Clear the migration state bit */
1725 timer
->state
&= ~HRTIMER_STATE_MIGRATE
;
1729 static void migrate_hrtimers(int scpu
)
1731 struct hrtimer_cpu_base
*old_base
, *new_base
;
1734 BUG_ON(cpu_online(scpu
));
1735 tick_cancel_sched_timer(scpu
);
1737 local_irq_disable();
1738 old_base
= &per_cpu(hrtimer_bases
, scpu
);
1739 new_base
= &__get_cpu_var(hrtimer_bases
);
1741 * The caller is globally serialized and nobody else
1742 * takes two locks at once, deadlock is not possible.
1744 raw_spin_lock(&new_base
->lock
);
1745 raw_spin_lock_nested(&old_base
->lock
, SINGLE_DEPTH_NESTING
);
1747 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++) {
1748 migrate_hrtimer_list(&old_base
->clock_base
[i
],
1749 &new_base
->clock_base
[i
]);
1752 raw_spin_unlock(&old_base
->lock
);
1753 raw_spin_unlock(&new_base
->lock
);
1755 /* Check, if we got expired work to do */
1756 __hrtimer_peek_ahead_timers();
1760 #endif /* CONFIG_HOTPLUG_CPU */
1762 static int __cpuinit
hrtimer_cpu_notify(struct notifier_block
*self
,
1763 unsigned long action
, void *hcpu
)
1765 int scpu
= (long)hcpu
;
1769 case CPU_UP_PREPARE
:
1770 case CPU_UP_PREPARE_FROZEN
:
1771 init_hrtimers_cpu(scpu
);
1774 #ifdef CONFIG_HOTPLUG_CPU
1776 case CPU_DYING_FROZEN
:
1777 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DYING
, &scpu
);
1780 case CPU_DEAD_FROZEN
:
1782 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD
, &scpu
);
1783 migrate_hrtimers(scpu
);
1795 static struct notifier_block __cpuinitdata hrtimers_nb
= {
1796 .notifier_call
= hrtimer_cpu_notify
,
1799 void __init
hrtimers_init(void)
1801 hrtimer_cpu_notify(&hrtimers_nb
, (unsigned long)CPU_UP_PREPARE
,
1802 (void *)(long)smp_processor_id());
1803 register_cpu_notifier(&hrtimers_nb
);
1804 #ifdef CONFIG_HIGH_RES_TIMERS
1805 open_softirq(HRTIMER_SOFTIRQ
, run_hrtimer_softirq
);
1810 * schedule_hrtimeout_range_clock - sleep until timeout
1811 * @expires: timeout value (ktime_t)
1812 * @delta: slack in expires timeout (ktime_t)
1813 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1814 * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME
1817 schedule_hrtimeout_range_clock(ktime_t
*expires
, unsigned long delta
,
1818 const enum hrtimer_mode mode
, int clock
)
1820 struct hrtimer_sleeper t
;
1823 * Optimize when a zero timeout value is given. It does not
1824 * matter whether this is an absolute or a relative time.
1826 if (expires
&& !expires
->tv64
) {
1827 __set_current_state(TASK_RUNNING
);
1832 * A NULL parameter means "infinite"
1836 __set_current_state(TASK_RUNNING
);
1840 hrtimer_init_on_stack(&t
.timer
, clock
, mode
);
1841 hrtimer_set_expires_range_ns(&t
.timer
, *expires
, delta
);
1843 hrtimer_init_sleeper(&t
, current
);
1845 hrtimer_start_expires(&t
.timer
, mode
);
1846 if (!hrtimer_active(&t
.timer
))
1852 hrtimer_cancel(&t
.timer
);
1853 destroy_hrtimer_on_stack(&t
.timer
);
1855 __set_current_state(TASK_RUNNING
);
1857 return !t
.task
? 0 : -EINTR
;
1861 * schedule_hrtimeout_range - sleep until timeout
1862 * @expires: timeout value (ktime_t)
1863 * @delta: slack in expires timeout (ktime_t)
1864 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1866 * Make the current task sleep until the given expiry time has
1867 * elapsed. The routine will return immediately unless
1868 * the current task state has been set (see set_current_state()).
1870 * The @delta argument gives the kernel the freedom to schedule the
1871 * actual wakeup to a time that is both power and performance friendly.
1872 * The kernel give the normal best effort behavior for "@expires+@delta",
1873 * but may decide to fire the timer earlier, but no earlier than @expires.
1875 * You can set the task state as follows -
1877 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1878 * pass before the routine returns.
1880 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1881 * delivered to the current task.
1883 * The current task state is guaranteed to be TASK_RUNNING when this
1886 * Returns 0 when the timer has expired otherwise -EINTR
1888 int __sched
schedule_hrtimeout_range(ktime_t
*expires
, unsigned long delta
,
1889 const enum hrtimer_mode mode
)
1891 return schedule_hrtimeout_range_clock(expires
, delta
, mode
,
1894 EXPORT_SYMBOL_GPL(schedule_hrtimeout_range
);
1897 * schedule_hrtimeout - sleep until timeout
1898 * @expires: timeout value (ktime_t)
1899 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1901 * Make the current task sleep until the given expiry time has
1902 * elapsed. The routine will return immediately unless
1903 * the current task state has been set (see set_current_state()).
1905 * You can set the task state as follows -
1907 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1908 * pass before the routine returns.
1910 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1911 * delivered to the current task.
1913 * The current task state is guaranteed to be TASK_RUNNING when this
1916 * Returns 0 when the timer has expired otherwise -EINTR
1918 int __sched
schedule_hrtimeout(ktime_t
*expires
,
1919 const enum hrtimer_mode mode
)
1921 return schedule_hrtimeout_range(expires
, 0, mode
);
1923 EXPORT_SYMBOL_GPL(schedule_hrtimeout
);