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
;
252 #else /* CONFIG_SMP */
254 static inline struct hrtimer_clock_base
*
255 lock_hrtimer_base(const struct hrtimer
*timer
, unsigned long *flags
)
257 struct hrtimer_clock_base
*base
= timer
->base
;
259 raw_spin_lock_irqsave(&base
->cpu_base
->lock
, *flags
);
264 # define switch_hrtimer_base(t, b, p) (b)
266 #endif /* !CONFIG_SMP */
269 * Functions for the union type storage format of ktime_t which are
270 * too large for inlining:
272 #if BITS_PER_LONG < 64
273 # ifndef CONFIG_KTIME_SCALAR
275 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
277 * @nsec: the scalar nsec value to add
279 * Returns the sum of kt and nsec in ktime_t format
281 ktime_t
ktime_add_ns(const ktime_t kt
, u64 nsec
)
285 if (likely(nsec
< NSEC_PER_SEC
)) {
288 unsigned long rem
= do_div(nsec
, NSEC_PER_SEC
);
290 /* Make sure nsec fits into long */
291 if (unlikely(nsec
> KTIME_SEC_MAX
))
292 return (ktime_t
){ .tv64
= KTIME_MAX
};
294 tmp
= ktime_set((long)nsec
, rem
);
297 return ktime_add(kt
, tmp
);
300 EXPORT_SYMBOL_GPL(ktime_add_ns
);
303 * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable
305 * @nsec: the scalar nsec value to subtract
307 * Returns the subtraction of @nsec from @kt in ktime_t format
309 ktime_t
ktime_sub_ns(const ktime_t kt
, u64 nsec
)
313 if (likely(nsec
< NSEC_PER_SEC
)) {
316 unsigned long rem
= do_div(nsec
, NSEC_PER_SEC
);
318 tmp
= ktime_set((long)nsec
, rem
);
321 return ktime_sub(kt
, tmp
);
324 EXPORT_SYMBOL_GPL(ktime_sub_ns
);
325 # endif /* !CONFIG_KTIME_SCALAR */
328 * Divide a ktime value by a nanosecond value
330 u64
ktime_divns(const ktime_t kt
, s64 div
)
335 dclc
= ktime_to_ns(kt
);
336 /* Make sure the divisor is less than 2^32: */
342 do_div(dclc
, (unsigned long) div
);
346 #endif /* BITS_PER_LONG >= 64 */
349 * Add two ktime values and do a safety check for overflow:
351 ktime_t
ktime_add_safe(const ktime_t lhs
, const ktime_t rhs
)
353 ktime_t res
= ktime_add(lhs
, rhs
);
356 * We use KTIME_SEC_MAX here, the maximum timeout which we can
357 * return to user space in a timespec:
359 if (res
.tv64
< 0 || res
.tv64
< lhs
.tv64
|| res
.tv64
< rhs
.tv64
)
360 res
= ktime_set(KTIME_SEC_MAX
, 0);
365 EXPORT_SYMBOL_GPL(ktime_add_safe
);
367 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
369 static struct debug_obj_descr hrtimer_debug_descr
;
371 static void *hrtimer_debug_hint(void *addr
)
373 return ((struct hrtimer
*) addr
)->function
;
377 * fixup_init is called when:
378 * - an active object is initialized
380 static int hrtimer_fixup_init(void *addr
, enum debug_obj_state state
)
382 struct hrtimer
*timer
= addr
;
385 case ODEBUG_STATE_ACTIVE
:
386 hrtimer_cancel(timer
);
387 debug_object_init(timer
, &hrtimer_debug_descr
);
395 * fixup_activate is called when:
396 * - an active object is activated
397 * - an unknown object is activated (might be a statically initialized object)
399 static int hrtimer_fixup_activate(void *addr
, enum debug_obj_state state
)
403 case ODEBUG_STATE_NOTAVAILABLE
:
407 case ODEBUG_STATE_ACTIVE
:
416 * fixup_free is called when:
417 * - an active object is freed
419 static int hrtimer_fixup_free(void *addr
, enum debug_obj_state state
)
421 struct hrtimer
*timer
= addr
;
424 case ODEBUG_STATE_ACTIVE
:
425 hrtimer_cancel(timer
);
426 debug_object_free(timer
, &hrtimer_debug_descr
);
433 static struct debug_obj_descr hrtimer_debug_descr
= {
435 .debug_hint
= hrtimer_debug_hint
,
436 .fixup_init
= hrtimer_fixup_init
,
437 .fixup_activate
= hrtimer_fixup_activate
,
438 .fixup_free
= hrtimer_fixup_free
,
441 static inline void debug_hrtimer_init(struct hrtimer
*timer
)
443 debug_object_init(timer
, &hrtimer_debug_descr
);
446 static inline void debug_hrtimer_activate(struct hrtimer
*timer
)
448 debug_object_activate(timer
, &hrtimer_debug_descr
);
451 static inline void debug_hrtimer_deactivate(struct hrtimer
*timer
)
453 debug_object_deactivate(timer
, &hrtimer_debug_descr
);
456 static inline void debug_hrtimer_free(struct hrtimer
*timer
)
458 debug_object_free(timer
, &hrtimer_debug_descr
);
461 static void __hrtimer_init(struct hrtimer
*timer
, clockid_t clock_id
,
462 enum hrtimer_mode mode
);
464 void hrtimer_init_on_stack(struct hrtimer
*timer
, clockid_t clock_id
,
465 enum hrtimer_mode mode
)
467 debug_object_init_on_stack(timer
, &hrtimer_debug_descr
);
468 __hrtimer_init(timer
, clock_id
, mode
);
470 EXPORT_SYMBOL_GPL(hrtimer_init_on_stack
);
472 void destroy_hrtimer_on_stack(struct hrtimer
*timer
)
474 debug_object_free(timer
, &hrtimer_debug_descr
);
478 static inline void debug_hrtimer_init(struct hrtimer
*timer
) { }
479 static inline void debug_hrtimer_activate(struct hrtimer
*timer
) { }
480 static inline void debug_hrtimer_deactivate(struct hrtimer
*timer
) { }
484 debug_init(struct hrtimer
*timer
, clockid_t clockid
,
485 enum hrtimer_mode mode
)
487 debug_hrtimer_init(timer
);
488 trace_hrtimer_init(timer
, clockid
, mode
);
491 static inline void debug_activate(struct hrtimer
*timer
)
493 debug_hrtimer_activate(timer
);
494 trace_hrtimer_start(timer
);
497 static inline void debug_deactivate(struct hrtimer
*timer
)
499 debug_hrtimer_deactivate(timer
);
500 trace_hrtimer_cancel(timer
);
503 /* High resolution timer related functions */
504 #ifdef CONFIG_HIGH_RES_TIMERS
507 * High resolution timer enabled ?
509 static int hrtimer_hres_enabled __read_mostly
= 1;
512 * Enable / Disable high resolution mode
514 static int __init
setup_hrtimer_hres(char *str
)
516 if (!strcmp(str
, "off"))
517 hrtimer_hres_enabled
= 0;
518 else if (!strcmp(str
, "on"))
519 hrtimer_hres_enabled
= 1;
525 __setup("highres=", setup_hrtimer_hres
);
528 * hrtimer_high_res_enabled - query, if the highres mode is enabled
530 static inline int hrtimer_is_hres_enabled(void)
532 return hrtimer_hres_enabled
;
536 * Is the high resolution mode active ?
538 static inline int hrtimer_hres_active(void)
540 return __this_cpu_read(hrtimer_bases
.hres_active
);
544 * Reprogram the event source with checking both queues for the
546 * Called with interrupts disabled and base->lock held
549 hrtimer_force_reprogram(struct hrtimer_cpu_base
*cpu_base
, int skip_equal
)
552 struct hrtimer_clock_base
*base
= cpu_base
->clock_base
;
553 ktime_t expires
, expires_next
;
555 expires_next
.tv64
= KTIME_MAX
;
557 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++, base
++) {
558 struct hrtimer
*timer
;
559 struct timerqueue_node
*next
;
561 next
= timerqueue_getnext(&base
->active
);
564 timer
= container_of(next
, struct hrtimer
, node
);
566 expires
= ktime_sub(hrtimer_get_expires(timer
), base
->offset
);
568 * clock_was_set() has changed base->offset so the
569 * result might be negative. Fix it up to prevent a
570 * false positive in clockevents_program_event()
572 if (expires
.tv64
< 0)
574 if (expires
.tv64
< expires_next
.tv64
)
575 expires_next
= expires
;
578 if (skip_equal
&& expires_next
.tv64
== cpu_base
->expires_next
.tv64
)
581 cpu_base
->expires_next
.tv64
= expires_next
.tv64
;
583 if (cpu_base
->expires_next
.tv64
!= KTIME_MAX
)
584 tick_program_event(cpu_base
->expires_next
, 1);
588 * Shared reprogramming for clock_realtime and clock_monotonic
590 * When a timer is enqueued and expires earlier than the already enqueued
591 * timers, we have to check, whether it expires earlier than the timer for
592 * which the clock event device was armed.
594 * Called with interrupts disabled and base->cpu_base.lock held
596 static int hrtimer_reprogram(struct hrtimer
*timer
,
597 struct hrtimer_clock_base
*base
)
599 struct hrtimer_cpu_base
*cpu_base
= &__get_cpu_var(hrtimer_bases
);
600 ktime_t expires
= ktime_sub(hrtimer_get_expires(timer
), base
->offset
);
603 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer
) < 0);
606 * When the callback is running, we do not reprogram the clock event
607 * device. The timer callback is either running on a different CPU or
608 * the callback is executed in the hrtimer_interrupt context. The
609 * reprogramming is handled either by the softirq, which called the
610 * callback or at the end of the hrtimer_interrupt.
612 if (hrtimer_callback_running(timer
))
616 * CLOCK_REALTIME timer might be requested with an absolute
617 * expiry time which is less than base->offset. Nothing wrong
618 * about that, just avoid to call into the tick code, which
619 * has now objections against negative expiry values.
621 if (expires
.tv64
< 0)
624 if (expires
.tv64
>= cpu_base
->expires_next
.tv64
)
628 * If a hang was detected in the last timer interrupt then we
629 * do not schedule a timer which is earlier than the expiry
630 * which we enforced in the hang detection. We want the system
633 if (cpu_base
->hang_detected
)
637 * Clockevents returns -ETIME, when the event was in the past.
639 res
= tick_program_event(expires
, 0);
640 if (!IS_ERR_VALUE(res
))
641 cpu_base
->expires_next
= expires
;
646 * Initialize the high resolution related parts of cpu_base
648 static inline void hrtimer_init_hres(struct hrtimer_cpu_base
*base
)
650 base
->expires_next
.tv64
= KTIME_MAX
;
651 base
->hres_active
= 0;
655 * When High resolution timers are active, try to reprogram. Note, that in case
656 * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry
657 * check happens. The timer gets enqueued into the rbtree. The reprogramming
658 * and expiry check is done in the hrtimer_interrupt or in the softirq.
660 static inline int hrtimer_enqueue_reprogram(struct hrtimer
*timer
,
661 struct hrtimer_clock_base
*base
)
663 return base
->cpu_base
->hres_active
&& hrtimer_reprogram(timer
, base
);
666 static inline ktime_t
hrtimer_update_base(struct hrtimer_cpu_base
*base
)
668 ktime_t
*offs_real
= &base
->clock_base
[HRTIMER_BASE_REALTIME
].offset
;
669 ktime_t
*offs_boot
= &base
->clock_base
[HRTIMER_BASE_BOOTTIME
].offset
;
670 ktime_t
*offs_tai
= &base
->clock_base
[HRTIMER_BASE_TAI
].offset
;
672 return ktime_get_update_offsets(offs_real
, offs_boot
, offs_tai
);
676 * Retrigger next event is called after clock was set
678 * Called with interrupts disabled via on_each_cpu()
680 static void retrigger_next_event(void *arg
)
682 struct hrtimer_cpu_base
*base
= &__get_cpu_var(hrtimer_bases
);
684 if (!hrtimer_hres_active())
687 raw_spin_lock(&base
->lock
);
688 hrtimer_update_base(base
);
689 hrtimer_force_reprogram(base
, 0);
690 raw_spin_unlock(&base
->lock
);
694 * Switch to high resolution mode
696 static int hrtimer_switch_to_hres(void)
698 int i
, cpu
= smp_processor_id();
699 struct hrtimer_cpu_base
*base
= &per_cpu(hrtimer_bases
, cpu
);
702 if (base
->hres_active
)
705 local_irq_save(flags
);
707 if (tick_init_highres()) {
708 local_irq_restore(flags
);
709 printk(KERN_WARNING
"Could not switch to high resolution "
710 "mode on CPU %d\n", cpu
);
713 base
->hres_active
= 1;
714 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++)
715 base
->clock_base
[i
].resolution
= KTIME_HIGH_RES
;
717 tick_setup_sched_timer();
718 /* "Retrigger" the interrupt to get things going */
719 retrigger_next_event(NULL
);
720 local_irq_restore(flags
);
724 static void clock_was_set_work(struct work_struct
*work
)
729 static DECLARE_WORK(hrtimer_work
, clock_was_set_work
);
732 * Called from timekeeping and resume code to reprogramm the hrtimer
733 * interrupt device on all cpus.
735 void clock_was_set_delayed(void)
737 schedule_work(&hrtimer_work
);
742 static inline int hrtimer_hres_active(void) { return 0; }
743 static inline int hrtimer_is_hres_enabled(void) { return 0; }
744 static inline int hrtimer_switch_to_hres(void) { return 0; }
746 hrtimer_force_reprogram(struct hrtimer_cpu_base
*base
, int skip_equal
) { }
747 static inline int hrtimer_enqueue_reprogram(struct hrtimer
*timer
,
748 struct hrtimer_clock_base
*base
)
752 static inline void hrtimer_init_hres(struct hrtimer_cpu_base
*base
) { }
753 static inline void retrigger_next_event(void *arg
) { }
755 #endif /* CONFIG_HIGH_RES_TIMERS */
758 * Clock realtime was set
760 * Change the offset of the realtime clock vs. the monotonic
763 * We might have to reprogram the high resolution timer interrupt. On
764 * SMP we call the architecture specific code to retrigger _all_ high
765 * resolution timer interrupts. On UP we just disable interrupts and
766 * call the high resolution interrupt code.
768 void clock_was_set(void)
770 #ifdef CONFIG_HIGH_RES_TIMERS
771 /* Retrigger the CPU local events everywhere */
772 on_each_cpu(retrigger_next_event
, NULL
, 1);
774 timerfd_clock_was_set();
778 * During resume we might have to reprogram the high resolution timer
779 * interrupt (on the local CPU):
781 void hrtimers_resume(void)
783 WARN_ONCE(!irqs_disabled(),
784 KERN_INFO
"hrtimers_resume() called with IRQs enabled!");
786 /* Retrigger on the local CPU */
787 retrigger_next_event(NULL
);
788 /* And schedule a retrigger for all others */
789 clock_was_set_delayed();
792 static inline void timer_stats_hrtimer_set_start_info(struct hrtimer
*timer
)
794 #ifdef CONFIG_TIMER_STATS
795 if (timer
->start_site
)
797 timer
->start_site
= __builtin_return_address(0);
798 memcpy(timer
->start_comm
, current
->comm
, TASK_COMM_LEN
);
799 timer
->start_pid
= current
->pid
;
803 static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer
*timer
)
805 #ifdef CONFIG_TIMER_STATS
806 timer
->start_site
= NULL
;
810 static inline void timer_stats_account_hrtimer(struct hrtimer
*timer
)
812 #ifdef CONFIG_TIMER_STATS
813 if (likely(!timer_stats_active
))
815 timer_stats_update_stats(timer
, timer
->start_pid
, timer
->start_site
,
816 timer
->function
, timer
->start_comm
, 0);
821 * Counterpart to lock_hrtimer_base above:
824 void unlock_hrtimer_base(const struct hrtimer
*timer
, unsigned long *flags
)
826 raw_spin_unlock_irqrestore(&timer
->base
->cpu_base
->lock
, *flags
);
830 * hrtimer_forward - forward the timer expiry
831 * @timer: hrtimer to forward
832 * @now: forward past this time
833 * @interval: the interval to forward
835 * Forward the timer expiry so it will expire in the future.
836 * Returns the number of overruns.
838 u64
hrtimer_forward(struct hrtimer
*timer
, ktime_t now
, ktime_t interval
)
843 delta
= ktime_sub(now
, hrtimer_get_expires(timer
));
848 if (interval
.tv64
< timer
->base
->resolution
.tv64
)
849 interval
.tv64
= timer
->base
->resolution
.tv64
;
851 if (unlikely(delta
.tv64
>= interval
.tv64
)) {
852 s64 incr
= ktime_to_ns(interval
);
854 orun
= ktime_divns(delta
, incr
);
855 hrtimer_add_expires_ns(timer
, incr
* orun
);
856 if (hrtimer_get_expires_tv64(timer
) > now
.tv64
)
859 * This (and the ktime_add() below) is the
860 * correction for exact:
864 hrtimer_add_expires(timer
, interval
);
868 EXPORT_SYMBOL_GPL(hrtimer_forward
);
871 * enqueue_hrtimer - internal function to (re)start a timer
873 * The timer is inserted in expiry order. Insertion into the
874 * red black tree is O(log(n)). Must hold the base lock.
876 * Returns 1 when the new timer is the leftmost timer in the tree.
878 static int enqueue_hrtimer(struct hrtimer
*timer
,
879 struct hrtimer_clock_base
*base
)
881 debug_activate(timer
);
883 timerqueue_add(&base
->active
, &timer
->node
);
884 base
->cpu_base
->active_bases
|= 1 << base
->index
;
887 * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
888 * state of a possibly running callback.
890 timer
->state
|= HRTIMER_STATE_ENQUEUED
;
892 return (&timer
->node
== base
->active
.next
);
896 * __remove_hrtimer - internal function to remove a timer
898 * Caller must hold the base lock.
900 * High resolution timer mode reprograms the clock event device when the
901 * timer is the one which expires next. The caller can disable this by setting
902 * reprogram to zero. This is useful, when the context does a reprogramming
903 * anyway (e.g. timer interrupt)
905 static void __remove_hrtimer(struct hrtimer
*timer
,
906 struct hrtimer_clock_base
*base
,
907 unsigned long newstate
, int reprogram
)
909 struct timerqueue_node
*next_timer
;
910 if (!(timer
->state
& HRTIMER_STATE_ENQUEUED
))
913 next_timer
= timerqueue_getnext(&base
->active
);
914 timerqueue_del(&base
->active
, &timer
->node
);
915 if (&timer
->node
== next_timer
) {
916 #ifdef CONFIG_HIGH_RES_TIMERS
917 /* Reprogram the clock event device. if enabled */
918 if (reprogram
&& hrtimer_hres_active()) {
921 expires
= ktime_sub(hrtimer_get_expires(timer
),
923 if (base
->cpu_base
->expires_next
.tv64
== expires
.tv64
)
924 hrtimer_force_reprogram(base
->cpu_base
, 1);
928 if (!timerqueue_getnext(&base
->active
))
929 base
->cpu_base
->active_bases
&= ~(1 << base
->index
);
931 timer
->state
= newstate
;
935 * remove hrtimer, called with base lock held
938 remove_hrtimer(struct hrtimer
*timer
, struct hrtimer_clock_base
*base
)
940 if (hrtimer_is_queued(timer
)) {
945 * Remove the timer and force reprogramming when high
946 * resolution mode is active and the timer is on the current
947 * CPU. If we remove a timer on another CPU, reprogramming is
948 * skipped. The interrupt event on this CPU is fired and
949 * reprogramming happens in the interrupt handler. This is a
950 * rare case and less expensive than a smp call.
952 debug_deactivate(timer
);
953 timer_stats_hrtimer_clear_start_info(timer
);
954 reprogram
= base
->cpu_base
== &__get_cpu_var(hrtimer_bases
);
956 * We must preserve the CALLBACK state flag here,
957 * otherwise we could move the timer base in
958 * switch_hrtimer_base.
960 state
= timer
->state
& HRTIMER_STATE_CALLBACK
;
961 __remove_hrtimer(timer
, base
, state
, reprogram
);
967 int __hrtimer_start_range_ns(struct hrtimer
*timer
, ktime_t tim
,
968 unsigned long delta_ns
, const enum hrtimer_mode mode
,
971 struct hrtimer_clock_base
*base
, *new_base
;
975 base
= lock_hrtimer_base(timer
, &flags
);
977 /* Remove an active timer from the queue: */
978 ret
= remove_hrtimer(timer
, base
);
980 /* Switch the timer base, if necessary: */
981 new_base
= switch_hrtimer_base(timer
, base
, mode
& HRTIMER_MODE_PINNED
);
983 if (mode
& HRTIMER_MODE_REL
) {
984 tim
= ktime_add_safe(tim
, new_base
->get_time());
986 * CONFIG_TIME_LOW_RES is a temporary way for architectures
987 * to signal that they simply return xtime in
988 * do_gettimeoffset(). In this case we want to round up by
989 * resolution when starting a relative timer, to avoid short
990 * timeouts. This will go away with the GTOD framework.
992 #ifdef CONFIG_TIME_LOW_RES
993 tim
= ktime_add_safe(tim
, base
->resolution
);
997 hrtimer_set_expires_range_ns(timer
, tim
, delta_ns
);
999 timer_stats_hrtimer_set_start_info(timer
);
1001 leftmost
= enqueue_hrtimer(timer
, new_base
);
1004 * Only allow reprogramming if the new base is on this CPU.
1005 * (it might still be on another CPU if the timer was pending)
1007 * XXX send_remote_softirq() ?
1009 if (leftmost
&& new_base
->cpu_base
== &__get_cpu_var(hrtimer_bases
)
1010 && hrtimer_enqueue_reprogram(timer
, new_base
)) {
1013 * We need to drop cpu_base->lock to avoid a
1014 * lock ordering issue vs. rq->lock.
1016 raw_spin_unlock(&new_base
->cpu_base
->lock
);
1017 raise_softirq_irqoff(HRTIMER_SOFTIRQ
);
1018 local_irq_restore(flags
);
1021 __raise_softirq_irqoff(HRTIMER_SOFTIRQ
);
1025 unlock_hrtimer_base(timer
, &flags
);
1031 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
1032 * @timer: the timer to be added
1034 * @delta_ns: "slack" range for the timer
1035 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
1036 * relative (HRTIMER_MODE_REL)
1040 * 1 when the timer was active
1042 int hrtimer_start_range_ns(struct hrtimer
*timer
, ktime_t tim
,
1043 unsigned long delta_ns
, const enum hrtimer_mode mode
)
1045 return __hrtimer_start_range_ns(timer
, tim
, delta_ns
, mode
, 1);
1047 EXPORT_SYMBOL_GPL(hrtimer_start_range_ns
);
1050 * hrtimer_start - (re)start an hrtimer on the current CPU
1051 * @timer: the timer to be added
1053 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
1054 * relative (HRTIMER_MODE_REL)
1058 * 1 when the timer was active
1061 hrtimer_start(struct hrtimer
*timer
, ktime_t tim
, const enum hrtimer_mode mode
)
1063 return __hrtimer_start_range_ns(timer
, tim
, 0, mode
, 1);
1065 EXPORT_SYMBOL_GPL(hrtimer_start
);
1069 * hrtimer_try_to_cancel - try to deactivate a timer
1070 * @timer: hrtimer to stop
1073 * 0 when the timer was not active
1074 * 1 when the timer was active
1075 * -1 when the timer is currently excuting the callback function and
1078 int hrtimer_try_to_cancel(struct hrtimer
*timer
)
1080 struct hrtimer_clock_base
*base
;
1081 unsigned long flags
;
1084 base
= lock_hrtimer_base(timer
, &flags
);
1086 if (!hrtimer_callback_running(timer
))
1087 ret
= remove_hrtimer(timer
, base
);
1089 unlock_hrtimer_base(timer
, &flags
);
1094 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel
);
1097 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1098 * @timer: the timer to be cancelled
1101 * 0 when the timer was not active
1102 * 1 when the timer was active
1104 int hrtimer_cancel(struct hrtimer
*timer
)
1107 int ret
= hrtimer_try_to_cancel(timer
);
1114 EXPORT_SYMBOL_GPL(hrtimer_cancel
);
1117 * hrtimer_get_remaining - get remaining time for the timer
1118 * @timer: the timer to read
1120 ktime_t
hrtimer_get_remaining(const struct hrtimer
*timer
)
1122 unsigned long flags
;
1125 lock_hrtimer_base(timer
, &flags
);
1126 rem
= hrtimer_expires_remaining(timer
);
1127 unlock_hrtimer_base(timer
, &flags
);
1131 EXPORT_SYMBOL_GPL(hrtimer_get_remaining
);
1133 #ifdef CONFIG_NO_HZ_COMMON
1135 * hrtimer_get_next_event - get the time until next expiry event
1137 * Returns the delta to the next expiry event or KTIME_MAX if no timer
1140 ktime_t
hrtimer_get_next_event(void)
1142 struct hrtimer_cpu_base
*cpu_base
= &__get_cpu_var(hrtimer_bases
);
1143 struct hrtimer_clock_base
*base
= cpu_base
->clock_base
;
1144 ktime_t delta
, mindelta
= { .tv64
= KTIME_MAX
};
1145 unsigned long flags
;
1148 raw_spin_lock_irqsave(&cpu_base
->lock
, flags
);
1150 if (!hrtimer_hres_active()) {
1151 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++, base
++) {
1152 struct hrtimer
*timer
;
1153 struct timerqueue_node
*next
;
1155 next
= timerqueue_getnext(&base
->active
);
1159 timer
= container_of(next
, struct hrtimer
, node
);
1160 delta
.tv64
= hrtimer_get_expires_tv64(timer
);
1161 delta
= ktime_sub(delta
, base
->get_time());
1162 if (delta
.tv64
< mindelta
.tv64
)
1163 mindelta
.tv64
= delta
.tv64
;
1167 raw_spin_unlock_irqrestore(&cpu_base
->lock
, flags
);
1169 if (mindelta
.tv64
< 0)
1175 static void __hrtimer_init(struct hrtimer
*timer
, clockid_t clock_id
,
1176 enum hrtimer_mode mode
)
1178 struct hrtimer_cpu_base
*cpu_base
;
1181 memset(timer
, 0, sizeof(struct hrtimer
));
1183 cpu_base
= &__raw_get_cpu_var(hrtimer_bases
);
1185 if (clock_id
== CLOCK_REALTIME
&& mode
!= HRTIMER_MODE_ABS
)
1186 clock_id
= CLOCK_MONOTONIC
;
1188 base
= hrtimer_clockid_to_base(clock_id
);
1189 timer
->base
= &cpu_base
->clock_base
[base
];
1190 timerqueue_init(&timer
->node
);
1192 #ifdef CONFIG_TIMER_STATS
1193 timer
->start_site
= NULL
;
1194 timer
->start_pid
= -1;
1195 memset(timer
->start_comm
, 0, TASK_COMM_LEN
);
1200 * hrtimer_init - initialize a timer to the given clock
1201 * @timer: the timer to be initialized
1202 * @clock_id: the clock to be used
1203 * @mode: timer mode abs/rel
1205 void hrtimer_init(struct hrtimer
*timer
, clockid_t clock_id
,
1206 enum hrtimer_mode mode
)
1208 debug_init(timer
, clock_id
, mode
);
1209 __hrtimer_init(timer
, clock_id
, mode
);
1211 EXPORT_SYMBOL_GPL(hrtimer_init
);
1214 * hrtimer_get_res - get the timer resolution for a clock
1215 * @which_clock: which clock to query
1216 * @tp: pointer to timespec variable to store the resolution
1218 * Store the resolution of the clock selected by @which_clock in the
1219 * variable pointed to by @tp.
1221 int hrtimer_get_res(const clockid_t which_clock
, struct timespec
*tp
)
1223 struct hrtimer_cpu_base
*cpu_base
;
1224 int base
= hrtimer_clockid_to_base(which_clock
);
1226 cpu_base
= &__raw_get_cpu_var(hrtimer_bases
);
1227 *tp
= ktime_to_timespec(cpu_base
->clock_base
[base
].resolution
);
1231 EXPORT_SYMBOL_GPL(hrtimer_get_res
);
1233 static void __run_hrtimer(struct hrtimer
*timer
, ktime_t
*now
)
1235 struct hrtimer_clock_base
*base
= timer
->base
;
1236 struct hrtimer_cpu_base
*cpu_base
= base
->cpu_base
;
1237 enum hrtimer_restart (*fn
)(struct hrtimer
*);
1240 WARN_ON(!irqs_disabled());
1242 debug_deactivate(timer
);
1243 __remove_hrtimer(timer
, base
, HRTIMER_STATE_CALLBACK
, 0);
1244 timer_stats_account_hrtimer(timer
);
1245 fn
= timer
->function
;
1248 * Because we run timers from hardirq context, there is no chance
1249 * they get migrated to another cpu, therefore its safe to unlock
1252 raw_spin_unlock(&cpu_base
->lock
);
1253 trace_hrtimer_expire_entry(timer
, now
);
1254 restart
= fn(timer
);
1255 trace_hrtimer_expire_exit(timer
);
1256 raw_spin_lock(&cpu_base
->lock
);
1259 * Note: We clear the CALLBACK bit after enqueue_hrtimer and
1260 * we do not reprogramm the event hardware. Happens either in
1261 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1263 if (restart
!= HRTIMER_NORESTART
) {
1264 BUG_ON(timer
->state
!= HRTIMER_STATE_CALLBACK
);
1265 enqueue_hrtimer(timer
, base
);
1268 WARN_ON_ONCE(!(timer
->state
& HRTIMER_STATE_CALLBACK
));
1270 timer
->state
&= ~HRTIMER_STATE_CALLBACK
;
1273 #ifdef CONFIG_HIGH_RES_TIMERS
1276 * High resolution timer interrupt
1277 * Called with interrupts disabled
1279 void hrtimer_interrupt(struct clock_event_device
*dev
)
1281 struct hrtimer_cpu_base
*cpu_base
= &__get_cpu_var(hrtimer_bases
);
1282 ktime_t expires_next
, now
, entry_time
, delta
;
1285 BUG_ON(!cpu_base
->hres_active
);
1286 cpu_base
->nr_events
++;
1287 dev
->next_event
.tv64
= KTIME_MAX
;
1289 raw_spin_lock(&cpu_base
->lock
);
1290 entry_time
= now
= hrtimer_update_base(cpu_base
);
1292 expires_next
.tv64
= KTIME_MAX
;
1294 * We set expires_next to KTIME_MAX here with cpu_base->lock
1295 * held to prevent that a timer is enqueued in our queue via
1296 * the migration code. This does not affect enqueueing of
1297 * timers which run their callback and need to be requeued on
1300 cpu_base
->expires_next
.tv64
= KTIME_MAX
;
1302 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++) {
1303 struct hrtimer_clock_base
*base
;
1304 struct timerqueue_node
*node
;
1307 if (!(cpu_base
->active_bases
& (1 << i
)))
1310 base
= cpu_base
->clock_base
+ i
;
1311 basenow
= ktime_add(now
, base
->offset
);
1313 while ((node
= timerqueue_getnext(&base
->active
))) {
1314 struct hrtimer
*timer
;
1316 timer
= container_of(node
, struct hrtimer
, node
);
1319 * The immediate goal for using the softexpires is
1320 * minimizing wakeups, not running timers at the
1321 * earliest interrupt after their soft expiration.
1322 * This allows us to avoid using a Priority Search
1323 * Tree, which can answer a stabbing querry for
1324 * overlapping intervals and instead use the simple
1325 * BST we already have.
1326 * We don't add extra wakeups by delaying timers that
1327 * are right-of a not yet expired timer, because that
1328 * timer will have to trigger a wakeup anyway.
1331 if (basenow
.tv64
< hrtimer_get_softexpires_tv64(timer
)) {
1334 expires
= ktime_sub(hrtimer_get_expires(timer
),
1336 if (expires
.tv64
< 0)
1337 expires
.tv64
= KTIME_MAX
;
1338 if (expires
.tv64
< expires_next
.tv64
)
1339 expires_next
= expires
;
1343 __run_hrtimer(timer
, &basenow
);
1348 * Store the new expiry value so the migration code can verify
1351 cpu_base
->expires_next
= expires_next
;
1352 raw_spin_unlock(&cpu_base
->lock
);
1354 /* Reprogramming necessary ? */
1355 if (expires_next
.tv64
== KTIME_MAX
||
1356 !tick_program_event(expires_next
, 0)) {
1357 cpu_base
->hang_detected
= 0;
1362 * The next timer was already expired due to:
1364 * - long lasting callbacks
1365 * - being scheduled away when running in a VM
1367 * We need to prevent that we loop forever in the hrtimer
1368 * interrupt routine. We give it 3 attempts to avoid
1369 * overreacting on some spurious event.
1371 * Acquire base lock for updating the offsets and retrieving
1374 raw_spin_lock(&cpu_base
->lock
);
1375 now
= hrtimer_update_base(cpu_base
);
1376 cpu_base
->nr_retries
++;
1380 * Give the system a chance to do something else than looping
1381 * here. We stored the entry time, so we know exactly how long
1382 * we spent here. We schedule the next event this amount of
1385 cpu_base
->nr_hangs
++;
1386 cpu_base
->hang_detected
= 1;
1387 raw_spin_unlock(&cpu_base
->lock
);
1388 delta
= ktime_sub(now
, entry_time
);
1389 if (delta
.tv64
> cpu_base
->max_hang_time
.tv64
)
1390 cpu_base
->max_hang_time
= delta
;
1392 * Limit it to a sensible value as we enforce a longer
1393 * delay. Give the CPU at least 100ms to catch up.
1395 if (delta
.tv64
> 100 * NSEC_PER_MSEC
)
1396 expires_next
= ktime_add_ns(now
, 100 * NSEC_PER_MSEC
);
1398 expires_next
= ktime_add(now
, delta
);
1399 tick_program_event(expires_next
, 1);
1400 printk_once(KERN_WARNING
"hrtimer: interrupt took %llu ns\n",
1401 ktime_to_ns(delta
));
1405 * local version of hrtimer_peek_ahead_timers() called with interrupts
1408 static void __hrtimer_peek_ahead_timers(void)
1410 struct tick_device
*td
;
1412 if (!hrtimer_hres_active())
1415 td
= &__get_cpu_var(tick_cpu_device
);
1416 if (td
&& td
->evtdev
)
1417 hrtimer_interrupt(td
->evtdev
);
1421 * hrtimer_peek_ahead_timers -- run soft-expired timers now
1423 * hrtimer_peek_ahead_timers will peek at the timer queue of
1424 * the current cpu and check if there are any timers for which
1425 * the soft expires time has passed. If any such timers exist,
1426 * they are run immediately and then removed from the timer queue.
1429 void hrtimer_peek_ahead_timers(void)
1431 unsigned long flags
;
1433 local_irq_save(flags
);
1434 __hrtimer_peek_ahead_timers();
1435 local_irq_restore(flags
);
1438 static void run_hrtimer_softirq(struct softirq_action
*h
)
1440 hrtimer_peek_ahead_timers();
1443 #else /* CONFIG_HIGH_RES_TIMERS */
1445 static inline void __hrtimer_peek_ahead_timers(void) { }
1447 #endif /* !CONFIG_HIGH_RES_TIMERS */
1450 * Called from timer softirq every jiffy, expire hrtimers:
1452 * For HRT its the fall back code to run the softirq in the timer
1453 * softirq context in case the hrtimer initialization failed or has
1454 * not been done yet.
1456 void hrtimer_run_pending(void)
1458 if (hrtimer_hres_active())
1462 * This _is_ ugly: We have to check in the softirq context,
1463 * whether we can switch to highres and / or nohz mode. The
1464 * clocksource switch happens in the timer interrupt with
1465 * xtime_lock held. Notification from there only sets the
1466 * check bit in the tick_oneshot code, otherwise we might
1467 * deadlock vs. xtime_lock.
1469 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1470 hrtimer_switch_to_hres();
1474 * Called from hardirq context every jiffy
1476 void hrtimer_run_queues(void)
1478 struct timerqueue_node
*node
;
1479 struct hrtimer_cpu_base
*cpu_base
= &__get_cpu_var(hrtimer_bases
);
1480 struct hrtimer_clock_base
*base
;
1481 int index
, gettime
= 1;
1483 if (hrtimer_hres_active())
1486 for (index
= 0; index
< HRTIMER_MAX_CLOCK_BASES
; index
++) {
1487 base
= &cpu_base
->clock_base
[index
];
1488 if (!timerqueue_getnext(&base
->active
))
1492 hrtimer_get_softirq_time(cpu_base
);
1496 raw_spin_lock(&cpu_base
->lock
);
1498 while ((node
= timerqueue_getnext(&base
->active
))) {
1499 struct hrtimer
*timer
;
1501 timer
= container_of(node
, struct hrtimer
, node
);
1502 if (base
->softirq_time
.tv64
<=
1503 hrtimer_get_expires_tv64(timer
))
1506 __run_hrtimer(timer
, &base
->softirq_time
);
1508 raw_spin_unlock(&cpu_base
->lock
);
1513 * Sleep related functions:
1515 static enum hrtimer_restart
hrtimer_wakeup(struct hrtimer
*timer
)
1517 struct hrtimer_sleeper
*t
=
1518 container_of(timer
, struct hrtimer_sleeper
, timer
);
1519 struct task_struct
*task
= t
->task
;
1523 wake_up_process(task
);
1525 return HRTIMER_NORESTART
;
1528 void hrtimer_init_sleeper(struct hrtimer_sleeper
*sl
, struct task_struct
*task
)
1530 sl
->timer
.function
= hrtimer_wakeup
;
1533 EXPORT_SYMBOL_GPL(hrtimer_init_sleeper
);
1535 static int __sched
do_nanosleep(struct hrtimer_sleeper
*t
, enum hrtimer_mode mode
)
1537 hrtimer_init_sleeper(t
, current
);
1540 set_current_state(TASK_INTERRUPTIBLE
);
1541 hrtimer_start_expires(&t
->timer
, mode
);
1542 if (!hrtimer_active(&t
->timer
))
1545 if (likely(t
->task
))
1548 hrtimer_cancel(&t
->timer
);
1549 mode
= HRTIMER_MODE_ABS
;
1551 } while (t
->task
&& !signal_pending(current
));
1553 __set_current_state(TASK_RUNNING
);
1555 return t
->task
== NULL
;
1558 static int update_rmtp(struct hrtimer
*timer
, struct timespec __user
*rmtp
)
1560 struct timespec rmt
;
1563 rem
= hrtimer_expires_remaining(timer
);
1566 rmt
= ktime_to_timespec(rem
);
1568 if (copy_to_user(rmtp
, &rmt
, sizeof(*rmtp
)))
1574 long __sched
hrtimer_nanosleep_restart(struct restart_block
*restart
)
1576 struct hrtimer_sleeper t
;
1577 struct timespec __user
*rmtp
;
1580 hrtimer_init_on_stack(&t
.timer
, restart
->nanosleep
.clockid
,
1582 hrtimer_set_expires_tv64(&t
.timer
, restart
->nanosleep
.expires
);
1584 if (do_nanosleep(&t
, HRTIMER_MODE_ABS
))
1587 rmtp
= restart
->nanosleep
.rmtp
;
1589 ret
= update_rmtp(&t
.timer
, rmtp
);
1594 /* The other values in restart are already filled in */
1595 ret
= -ERESTART_RESTARTBLOCK
;
1597 destroy_hrtimer_on_stack(&t
.timer
);
1601 long hrtimer_nanosleep(struct timespec
*rqtp
, struct timespec __user
*rmtp
,
1602 const enum hrtimer_mode mode
, const clockid_t clockid
)
1604 struct restart_block
*restart
;
1605 struct hrtimer_sleeper t
;
1607 unsigned long slack
;
1609 slack
= current
->timer_slack_ns
;
1610 if (rt_task(current
))
1613 hrtimer_init_on_stack(&t
.timer
, clockid
, mode
);
1614 hrtimer_set_expires_range_ns(&t
.timer
, timespec_to_ktime(*rqtp
), slack
);
1615 if (do_nanosleep(&t
, mode
))
1618 /* Absolute timers do not update the rmtp value and restart: */
1619 if (mode
== HRTIMER_MODE_ABS
) {
1620 ret
= -ERESTARTNOHAND
;
1625 ret
= update_rmtp(&t
.timer
, rmtp
);
1630 restart
= ¤t_thread_info()->restart_block
;
1631 restart
->fn
= hrtimer_nanosleep_restart
;
1632 restart
->nanosleep
.clockid
= t
.timer
.base
->clockid
;
1633 restart
->nanosleep
.rmtp
= rmtp
;
1634 restart
->nanosleep
.expires
= hrtimer_get_expires_tv64(&t
.timer
);
1636 ret
= -ERESTART_RESTARTBLOCK
;
1638 destroy_hrtimer_on_stack(&t
.timer
);
1642 SYSCALL_DEFINE2(nanosleep
, struct timespec __user
*, rqtp
,
1643 struct timespec __user
*, rmtp
)
1647 if (copy_from_user(&tu
, rqtp
, sizeof(tu
)))
1650 if (!timespec_valid(&tu
))
1653 return hrtimer_nanosleep(&tu
, rmtp
, HRTIMER_MODE_REL
, CLOCK_MONOTONIC
);
1657 * Functions related to boot-time initialization:
1659 static void __cpuinit
init_hrtimers_cpu(int cpu
)
1661 struct hrtimer_cpu_base
*cpu_base
= &per_cpu(hrtimer_bases
, cpu
);
1664 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++) {
1665 cpu_base
->clock_base
[i
].cpu_base
= cpu_base
;
1666 timerqueue_init_head(&cpu_base
->clock_base
[i
].active
);
1669 hrtimer_init_hres(cpu_base
);
1672 #ifdef CONFIG_HOTPLUG_CPU
1674 static void migrate_hrtimer_list(struct hrtimer_clock_base
*old_base
,
1675 struct hrtimer_clock_base
*new_base
)
1677 struct hrtimer
*timer
;
1678 struct timerqueue_node
*node
;
1680 while ((node
= timerqueue_getnext(&old_base
->active
))) {
1681 timer
= container_of(node
, struct hrtimer
, node
);
1682 BUG_ON(hrtimer_callback_running(timer
));
1683 debug_deactivate(timer
);
1686 * Mark it as STATE_MIGRATE not INACTIVE otherwise the
1687 * timer could be seen as !active and just vanish away
1688 * under us on another CPU
1690 __remove_hrtimer(timer
, old_base
, HRTIMER_STATE_MIGRATE
, 0);
1691 timer
->base
= new_base
;
1693 * Enqueue the timers on the new cpu. This does not
1694 * reprogram the event device in case the timer
1695 * expires before the earliest on this CPU, but we run
1696 * hrtimer_interrupt after we migrated everything to
1697 * sort out already expired timers and reprogram the
1700 enqueue_hrtimer(timer
, new_base
);
1702 /* Clear the migration state bit */
1703 timer
->state
&= ~HRTIMER_STATE_MIGRATE
;
1707 static void migrate_hrtimers(int scpu
)
1709 struct hrtimer_cpu_base
*old_base
, *new_base
;
1712 BUG_ON(cpu_online(scpu
));
1713 tick_cancel_sched_timer(scpu
);
1715 local_irq_disable();
1716 old_base
= &per_cpu(hrtimer_bases
, scpu
);
1717 new_base
= &__get_cpu_var(hrtimer_bases
);
1719 * The caller is globally serialized and nobody else
1720 * takes two locks at once, deadlock is not possible.
1722 raw_spin_lock(&new_base
->lock
);
1723 raw_spin_lock_nested(&old_base
->lock
, SINGLE_DEPTH_NESTING
);
1725 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++) {
1726 migrate_hrtimer_list(&old_base
->clock_base
[i
],
1727 &new_base
->clock_base
[i
]);
1730 raw_spin_unlock(&old_base
->lock
);
1731 raw_spin_unlock(&new_base
->lock
);
1733 /* Check, if we got expired work to do */
1734 __hrtimer_peek_ahead_timers();
1738 #endif /* CONFIG_HOTPLUG_CPU */
1740 static int __cpuinit
hrtimer_cpu_notify(struct notifier_block
*self
,
1741 unsigned long action
, void *hcpu
)
1743 int scpu
= (long)hcpu
;
1747 case CPU_UP_PREPARE
:
1748 case CPU_UP_PREPARE_FROZEN
:
1749 init_hrtimers_cpu(scpu
);
1752 #ifdef CONFIG_HOTPLUG_CPU
1754 case CPU_DYING_FROZEN
:
1755 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DYING
, &scpu
);
1758 case CPU_DEAD_FROZEN
:
1760 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD
, &scpu
);
1761 migrate_hrtimers(scpu
);
1773 static struct notifier_block __cpuinitdata hrtimers_nb
= {
1774 .notifier_call
= hrtimer_cpu_notify
,
1777 void __init
hrtimers_init(void)
1779 hrtimer_cpu_notify(&hrtimers_nb
, (unsigned long)CPU_UP_PREPARE
,
1780 (void *)(long)smp_processor_id());
1781 register_cpu_notifier(&hrtimers_nb
);
1782 #ifdef CONFIG_HIGH_RES_TIMERS
1783 open_softirq(HRTIMER_SOFTIRQ
, run_hrtimer_softirq
);
1788 * schedule_hrtimeout_range_clock - sleep until timeout
1789 * @expires: timeout value (ktime_t)
1790 * @delta: slack in expires timeout (ktime_t)
1791 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1792 * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME
1795 schedule_hrtimeout_range_clock(ktime_t
*expires
, unsigned long delta
,
1796 const enum hrtimer_mode mode
, int clock
)
1798 struct hrtimer_sleeper t
;
1801 * Optimize when a zero timeout value is given. It does not
1802 * matter whether this is an absolute or a relative time.
1804 if (expires
&& !expires
->tv64
) {
1805 __set_current_state(TASK_RUNNING
);
1810 * A NULL parameter means "infinite"
1814 __set_current_state(TASK_RUNNING
);
1818 hrtimer_init_on_stack(&t
.timer
, clock
, mode
);
1819 hrtimer_set_expires_range_ns(&t
.timer
, *expires
, delta
);
1821 hrtimer_init_sleeper(&t
, current
);
1823 hrtimer_start_expires(&t
.timer
, mode
);
1824 if (!hrtimer_active(&t
.timer
))
1830 hrtimer_cancel(&t
.timer
);
1831 destroy_hrtimer_on_stack(&t
.timer
);
1833 __set_current_state(TASK_RUNNING
);
1835 return !t
.task
? 0 : -EINTR
;
1839 * schedule_hrtimeout_range - sleep until timeout
1840 * @expires: timeout value (ktime_t)
1841 * @delta: slack in expires timeout (ktime_t)
1842 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1844 * Make the current task sleep until the given expiry time has
1845 * elapsed. The routine will return immediately unless
1846 * the current task state has been set (see set_current_state()).
1848 * The @delta argument gives the kernel the freedom to schedule the
1849 * actual wakeup to a time that is both power and performance friendly.
1850 * The kernel give the normal best effort behavior for "@expires+@delta",
1851 * but may decide to fire the timer earlier, but no earlier than @expires.
1853 * You can set the task state as follows -
1855 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1856 * pass before the routine returns.
1858 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1859 * delivered to the current task.
1861 * The current task state is guaranteed to be TASK_RUNNING when this
1864 * Returns 0 when the timer has expired otherwise -EINTR
1866 int __sched
schedule_hrtimeout_range(ktime_t
*expires
, unsigned long delta
,
1867 const enum hrtimer_mode mode
)
1869 return schedule_hrtimeout_range_clock(expires
, delta
, mode
,
1872 EXPORT_SYMBOL_GPL(schedule_hrtimeout_range
);
1875 * schedule_hrtimeout - sleep until timeout
1876 * @expires: timeout value (ktime_t)
1877 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1879 * Make the current task sleep until the given expiry time has
1880 * elapsed. The routine will return immediately unless
1881 * the current task state has been set (see set_current_state()).
1883 * You can set the task state as follows -
1885 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1886 * pass before the routine returns.
1888 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1889 * delivered to the current task.
1891 * The current task state is guaranteed to be TASK_RUNNING when this
1894 * Returns 0 when the timer has expired otherwise -EINTR
1896 int __sched
schedule_hrtimeout(ktime_t
*expires
,
1897 const enum hrtimer_mode mode
)
1899 return schedule_hrtimeout_range(expires
, 0, mode
);
1901 EXPORT_SYMBOL_GPL(schedule_hrtimeout
);