2 * linux/kernel/hrtimer.c
4 * Copyright(C) 2005, Thomas Gleixner <tglx@linutronix.de>
5 * Copyright(C) 2005, Red Hat, Inc., Ingo Molnar
7 * High-resolution kernel timers
9 * In contrast to the low-resolution timeout API implemented in
10 * kernel/timer.c, hrtimers provide finer resolution and accuracy
11 * depending on system configuration and capabilities.
13 * These timers are currently used for:
17 * - precise in-kernel timing
19 * Started by: Thomas Gleixner and Ingo Molnar
22 * based on kernel/timer.c
24 * Help, testing, suggestions, bugfixes, improvements were
27 * George Anzinger, Andrew Morton, Steven Rostedt, Roman Zippel
30 * For licencing details see kernel-base/COPYING
33 #include <linux/cpu.h>
34 #include <linux/module.h>
35 #include <linux/percpu.h>
36 #include <linux/hrtimer.h>
37 #include <linux/notifier.h>
38 #include <linux/syscalls.h>
39 #include <linux/interrupt.h>
41 #include <asm/uaccess.h>
44 * ktime_get - get the monotonic time in ktime_t format
46 * returns the time in ktime_t format
48 static ktime_t
ktime_get(void)
54 return timespec_to_ktime(now
);
58 * ktime_get_real - get the real (wall-) time in ktime_t format
60 * returns the time in ktime_t format
62 static ktime_t
ktime_get_real(void)
68 return timespec_to_ktime(now
);
74 * Note: If we want to add new timer bases, we have to skip the two
75 * clock ids captured by the cpu-timers. We do this by holding empty
76 * entries rather than doing math adjustment of the clock ids.
77 * This ensures that we capture erroneous accesses to these clock ids
78 * rather than moving them into the range of valid clock id's.
81 #define MAX_HRTIMER_BASES 2
83 static DEFINE_PER_CPU(struct hrtimer_base
, hrtimer_bases
[MAX_HRTIMER_BASES
]) =
86 .index
= CLOCK_REALTIME
,
87 .get_time
= &ktime_get_real
,
88 .resolution
= KTIME_REALTIME_RES
,
91 .index
= CLOCK_MONOTONIC
,
92 .get_time
= &ktime_get
,
93 .resolution
= KTIME_MONOTONIC_RES
,
98 * ktime_get_ts - get the monotonic clock in timespec format
100 * @ts: pointer to timespec variable
102 * The function calculates the monotonic clock from the realtime
103 * clock and the wall_to_monotonic offset and stores the result
104 * in normalized timespec format in the variable pointed to by ts.
106 void ktime_get_ts(struct timespec
*ts
)
108 struct timespec tomono
;
112 seq
= read_seqbegin(&xtime_lock
);
114 tomono
= wall_to_monotonic
;
116 } while (read_seqretry(&xtime_lock
, seq
));
118 set_normalized_timespec(ts
, ts
->tv_sec
+ tomono
.tv_sec
,
119 ts
->tv_nsec
+ tomono
.tv_nsec
);
121 EXPORT_SYMBOL_GPL(ktime_get_ts
);
124 * Functions and macros which are different for UP/SMP systems are kept in a
129 #define set_curr_timer(b, t) do { (b)->curr_timer = (t); } while (0)
132 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
133 * means that all timers which are tied to this base via timer->base are
134 * locked, and the base itself is locked too.
136 * So __run_timers/migrate_timers can safely modify all timers which could
137 * be found on the lists/queues.
139 * When the timer's base is locked, and the timer removed from list, it is
140 * possible to set timer->base = NULL and drop the lock: the timer remains
143 static struct hrtimer_base
*lock_hrtimer_base(const struct hrtimer
*timer
,
144 unsigned long *flags
)
146 struct hrtimer_base
*base
;
150 if (likely(base
!= NULL
)) {
151 spin_lock_irqsave(&base
->lock
, *flags
);
152 if (likely(base
== timer
->base
))
154 /* The timer has migrated to another CPU: */
155 spin_unlock_irqrestore(&base
->lock
, *flags
);
162 * Switch the timer base to the current CPU when possible.
164 static inline struct hrtimer_base
*
165 switch_hrtimer_base(struct hrtimer
*timer
, struct hrtimer_base
*base
)
167 struct hrtimer_base
*new_base
;
169 new_base
= &__get_cpu_var(hrtimer_bases
[base
->index
]);
171 if (base
!= new_base
) {
173 * We are trying to schedule the timer on the local CPU.
174 * However we can't change timer's base while it is running,
175 * so we keep it on the same CPU. No hassle vs. reprogramming
176 * the event source in the high resolution case. The softirq
177 * code will take care of this when the timer function has
178 * completed. There is no conflict as we hold the lock until
179 * the timer is enqueued.
181 if (unlikely(base
->curr_timer
== timer
))
184 /* See the comment in lock_timer_base() */
186 spin_unlock(&base
->lock
);
187 spin_lock(&new_base
->lock
);
188 timer
->base
= new_base
;
193 #else /* CONFIG_SMP */
195 #define set_curr_timer(b, t) do { } while (0)
197 static inline struct hrtimer_base
*
198 lock_hrtimer_base(const struct hrtimer
*timer
, unsigned long *flags
)
200 struct hrtimer_base
*base
= timer
->base
;
202 spin_lock_irqsave(&base
->lock
, *flags
);
207 #define switch_hrtimer_base(t, b) (b)
209 #endif /* !CONFIG_SMP */
212 * Functions for the union type storage format of ktime_t which are
213 * too large for inlining:
215 #if BITS_PER_LONG < 64
216 # ifndef CONFIG_KTIME_SCALAR
218 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
221 * @nsec: the scalar nsec value to add
223 * Returns the sum of kt and nsec in ktime_t format
225 ktime_t
ktime_add_ns(const ktime_t kt
, u64 nsec
)
229 if (likely(nsec
< NSEC_PER_SEC
)) {
232 unsigned long rem
= do_div(nsec
, NSEC_PER_SEC
);
234 tmp
= ktime_set((long)nsec
, rem
);
237 return ktime_add(kt
, tmp
);
240 #else /* CONFIG_KTIME_SCALAR */
242 # endif /* !CONFIG_KTIME_SCALAR */
245 * Divide a ktime value by a nanosecond value
247 static unsigned long ktime_divns(const ktime_t kt
, nsec_t div
)
252 dclc
= dns
= ktime_to_ns(kt
);
254 /* Make sure the divisor is less than 2^32: */
260 do_div(dclc
, (unsigned long) div
);
262 return (unsigned long) dclc
;
265 #else /* BITS_PER_LONG < 64 */
266 # define ktime_divns(kt, div) (unsigned long)((kt).tv64 / (div))
267 #endif /* BITS_PER_LONG >= 64 */
270 * Counterpart to lock_timer_base above:
273 void unlock_hrtimer_base(const struct hrtimer
*timer
, unsigned long *flags
)
275 spin_unlock_irqrestore(&timer
->base
->lock
, *flags
);
279 * hrtimer_forward - forward the timer expiry
281 * @timer: hrtimer to forward
282 * @interval: the interval to forward
284 * Forward the timer expiry so it will expire in the future.
285 * Returns the number of overruns.
288 hrtimer_forward(struct hrtimer
*timer
, ktime_t interval
)
290 unsigned long orun
= 1;
293 now
= timer
->base
->get_time();
295 delta
= ktime_sub(now
, timer
->expires
);
300 if (interval
.tv64
< timer
->base
->resolution
.tv64
)
301 interval
.tv64
= timer
->base
->resolution
.tv64
;
303 if (unlikely(delta
.tv64
>= interval
.tv64
)) {
304 nsec_t incr
= ktime_to_ns(interval
);
306 orun
= ktime_divns(delta
, incr
);
307 timer
->expires
= ktime_add_ns(timer
->expires
, incr
* orun
);
308 if (timer
->expires
.tv64
> now
.tv64
)
311 * This (and the ktime_add() below) is the
312 * correction for exact:
316 timer
->expires
= ktime_add(timer
->expires
, interval
);
318 * Make sure, that the result did not wrap with a very large
321 if (timer
->expires
.tv64
< 0)
322 timer
->expires
= ktime_set(KTIME_SEC_MAX
, 0);
328 * enqueue_hrtimer - internal function to (re)start a timer
330 * The timer is inserted in expiry order. Insertion into the
331 * red black tree is O(log(n)). Must hold the base lock.
333 static void enqueue_hrtimer(struct hrtimer
*timer
, struct hrtimer_base
*base
)
335 struct rb_node
**link
= &base
->active
.rb_node
;
336 struct rb_node
*parent
= NULL
;
337 struct hrtimer
*entry
;
340 * Find the right place in the rbtree:
344 entry
= rb_entry(parent
, struct hrtimer
, node
);
346 * We dont care about collisions. Nodes with
347 * the same expiry time stay together.
349 if (timer
->expires
.tv64
< entry
->expires
.tv64
)
350 link
= &(*link
)->rb_left
;
352 link
= &(*link
)->rb_right
;
356 * Insert the timer to the rbtree and check whether it
357 * replaces the first pending timer
359 rb_link_node(&timer
->node
, parent
, link
);
360 rb_insert_color(&timer
->node
, &base
->active
);
362 timer
->state
= HRTIMER_PENDING
;
364 if (!base
->first
|| timer
->expires
.tv64
<
365 rb_entry(base
->first
, struct hrtimer
, node
)->expires
.tv64
)
366 base
->first
= &timer
->node
;
370 * __remove_hrtimer - internal function to remove a timer
372 * Caller must hold the base lock.
374 static void __remove_hrtimer(struct hrtimer
*timer
, struct hrtimer_base
*base
)
377 * Remove the timer from the rbtree and replace the
378 * first entry pointer if necessary.
380 if (base
->first
== &timer
->node
)
381 base
->first
= rb_next(&timer
->node
);
382 rb_erase(&timer
->node
, &base
->active
);
386 * remove hrtimer, called with base lock held
389 remove_hrtimer(struct hrtimer
*timer
, struct hrtimer_base
*base
)
391 if (hrtimer_active(timer
)) {
392 __remove_hrtimer(timer
, base
);
393 timer
->state
= HRTIMER_INACTIVE
;
400 * hrtimer_start - (re)start an relative timer on the current CPU
402 * @timer: the timer to be added
404 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
408 * 1 when the timer was active
411 hrtimer_start(struct hrtimer
*timer
, ktime_t tim
, const enum hrtimer_mode mode
)
413 struct hrtimer_base
*base
, *new_base
;
417 base
= lock_hrtimer_base(timer
, &flags
);
419 /* Remove an active timer from the queue: */
420 ret
= remove_hrtimer(timer
, base
);
422 /* Switch the timer base, if necessary: */
423 new_base
= switch_hrtimer_base(timer
, base
);
425 if (mode
== HRTIMER_REL
) {
426 tim
= ktime_add(tim
, new_base
->get_time());
428 * CONFIG_TIME_LOW_RES is a temporary way for architectures
429 * to signal that they simply return xtime in
430 * do_gettimeoffset(). In this case we want to round up by
431 * resolution when starting a relative timer, to avoid short
432 * timeouts. This will go away with the GTOD framework.
434 #ifdef CONFIG_TIME_LOW_RES
435 tim
= ktime_add(tim
, base
->resolution
);
438 timer
->expires
= tim
;
440 enqueue_hrtimer(timer
, new_base
);
442 unlock_hrtimer_base(timer
, &flags
);
448 * hrtimer_try_to_cancel - try to deactivate a timer
450 * @timer: hrtimer to stop
453 * 0 when the timer was not active
454 * 1 when the timer was active
455 * -1 when the timer is currently excuting the callback function and
458 int hrtimer_try_to_cancel(struct hrtimer
*timer
)
460 struct hrtimer_base
*base
;
464 base
= lock_hrtimer_base(timer
, &flags
);
466 if (base
->curr_timer
!= timer
)
467 ret
= remove_hrtimer(timer
, base
);
469 unlock_hrtimer_base(timer
, &flags
);
476 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
478 * @timer: the timer to be cancelled
481 * 0 when the timer was not active
482 * 1 when the timer was active
484 int hrtimer_cancel(struct hrtimer
*timer
)
487 int ret
= hrtimer_try_to_cancel(timer
);
495 * hrtimer_get_remaining - get remaining time for the timer
497 * @timer: the timer to read
499 ktime_t
hrtimer_get_remaining(const struct hrtimer
*timer
)
501 struct hrtimer_base
*base
;
505 base
= lock_hrtimer_base(timer
, &flags
);
506 rem
= ktime_sub(timer
->expires
, timer
->base
->get_time());
507 unlock_hrtimer_base(timer
, &flags
);
512 #ifdef CONFIG_NO_IDLE_HZ
514 * hrtimer_get_next_event - get the time until next expiry event
516 * Returns the delta to the next expiry event or KTIME_MAX if no timer
519 ktime_t
hrtimer_get_next_event(void)
521 struct hrtimer_base
*base
= __get_cpu_var(hrtimer_bases
);
522 ktime_t delta
, mindelta
= { .tv64
= KTIME_MAX
};
526 for (i
= 0; i
< MAX_HRTIMER_BASES
; i
++, base
++) {
527 struct hrtimer
*timer
;
529 spin_lock_irqsave(&base
->lock
, flags
);
531 spin_unlock_irqrestore(&base
->lock
, flags
);
534 timer
= rb_entry(base
->first
, struct hrtimer
, node
);
535 delta
.tv64
= timer
->expires
.tv64
;
536 spin_unlock_irqrestore(&base
->lock
, flags
);
537 delta
= ktime_sub(delta
, base
->get_time());
538 if (delta
.tv64
< mindelta
.tv64
)
539 mindelta
.tv64
= delta
.tv64
;
541 if (mindelta
.tv64
< 0)
548 * hrtimer_init - initialize a timer to the given clock
550 * @timer: the timer to be initialized
551 * @clock_id: the clock to be used
552 * @mode: timer mode abs/rel
554 void hrtimer_init(struct hrtimer
*timer
, clockid_t clock_id
,
555 enum hrtimer_mode mode
)
557 struct hrtimer_base
*bases
;
559 memset(timer
, 0, sizeof(struct hrtimer
));
561 bases
= per_cpu(hrtimer_bases
, raw_smp_processor_id());
563 if (clock_id
== CLOCK_REALTIME
&& mode
!= HRTIMER_ABS
)
564 clock_id
= CLOCK_MONOTONIC
;
566 timer
->base
= &bases
[clock_id
];
570 * hrtimer_get_res - get the timer resolution for a clock
572 * @which_clock: which clock to query
573 * @tp: pointer to timespec variable to store the resolution
575 * Store the resolution of the clock selected by which_clock in the
576 * variable pointed to by tp.
578 int hrtimer_get_res(const clockid_t which_clock
, struct timespec
*tp
)
580 struct hrtimer_base
*bases
;
582 bases
= per_cpu(hrtimer_bases
, raw_smp_processor_id());
583 *tp
= ktime_to_timespec(bases
[which_clock
].resolution
);
589 * Expire the per base hrtimer-queue:
591 static inline void run_hrtimer_queue(struct hrtimer_base
*base
)
593 ktime_t now
= base
->get_time();
594 struct rb_node
*node
;
596 spin_lock_irq(&base
->lock
);
598 while ((node
= base
->first
)) {
599 struct hrtimer
*timer
;
604 timer
= rb_entry(node
, struct hrtimer
, node
);
605 if (now
.tv64
<= timer
->expires
.tv64
)
608 fn
= timer
->function
;
610 set_curr_timer(base
, timer
);
611 timer
->state
= HRTIMER_RUNNING
;
612 __remove_hrtimer(timer
, base
);
613 spin_unlock_irq(&base
->lock
);
616 * fn == NULL is special case for the simplest timer
617 * variant - wake up process and do not restart:
620 wake_up_process(data
);
621 restart
= HRTIMER_NORESTART
;
625 spin_lock_irq(&base
->lock
);
627 /* Another CPU has added back the timer */
628 if (timer
->state
!= HRTIMER_RUNNING
)
631 if (restart
== HRTIMER_RESTART
)
632 enqueue_hrtimer(timer
, base
);
634 timer
->state
= HRTIMER_EXPIRED
;
636 set_curr_timer(base
, NULL
);
637 spin_unlock_irq(&base
->lock
);
641 * Called from timer softirq every jiffy, expire hrtimers:
643 void hrtimer_run_queues(void)
645 struct hrtimer_base
*base
= __get_cpu_var(hrtimer_bases
);
648 for (i
= 0; i
< MAX_HRTIMER_BASES
; i
++)
649 run_hrtimer_queue(&base
[i
]);
653 * Sleep related functions:
657 * schedule_hrtimer - sleep until timeout
659 * @timer: hrtimer variable initialized with the correct clock base
660 * @mode: timeout value is abs/rel
662 * Make the current task sleep until @timeout is
665 * You can set the task state as follows -
667 * %TASK_UNINTERRUPTIBLE - at least @timeout is guaranteed to
668 * pass before the routine returns. The routine will return 0
670 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
671 * delivered to the current task. In this case the remaining time
674 * The current task state is guaranteed to be TASK_RUNNING when this
677 static ktime_t __sched
678 schedule_hrtimer(struct hrtimer
*timer
, const enum hrtimer_mode mode
)
680 /* fn stays NULL, meaning single-shot wakeup: */
681 timer
->data
= current
;
683 hrtimer_start(timer
, timer
->expires
, mode
);
686 hrtimer_cancel(timer
);
688 /* Return the remaining time: */
689 if (timer
->state
!= HRTIMER_EXPIRED
)
690 return ktime_sub(timer
->expires
, timer
->base
->get_time());
692 return (ktime_t
) {.tv64
= 0 };
695 static inline ktime_t __sched
696 schedule_hrtimer_interruptible(struct hrtimer
*timer
,
697 const enum hrtimer_mode mode
)
699 set_current_state(TASK_INTERRUPTIBLE
);
701 return schedule_hrtimer(timer
, mode
);
704 static long __sched
nanosleep_restart(struct restart_block
*restart
)
706 struct timespec __user
*rmtp
;
708 void *rfn_save
= restart
->fn
;
709 struct hrtimer timer
;
712 restart
->fn
= do_no_restart_syscall
;
714 hrtimer_init(&timer
, (clockid_t
) restart
->arg3
, HRTIMER_ABS
);
716 timer
.expires
.tv64
= ((u64
)restart
->arg1
<< 32) | (u64
) restart
->arg0
;
718 rem
= schedule_hrtimer_interruptible(&timer
, HRTIMER_ABS
);
723 rmtp
= (struct timespec __user
*) restart
->arg2
;
724 tu
= ktime_to_timespec(rem
);
725 if (rmtp
&& copy_to_user(rmtp
, &tu
, sizeof(tu
)))
728 restart
->fn
= rfn_save
;
730 /* The other values in restart are already filled in */
731 return -ERESTART_RESTARTBLOCK
;
734 long hrtimer_nanosleep(struct timespec
*rqtp
, struct timespec __user
*rmtp
,
735 const enum hrtimer_mode mode
, const clockid_t clockid
)
737 struct restart_block
*restart
;
738 struct hrtimer timer
;
742 hrtimer_init(&timer
, clockid
, mode
);
744 timer
.expires
= timespec_to_ktime(*rqtp
);
746 rem
= schedule_hrtimer_interruptible(&timer
, mode
);
750 /* Absolute timers do not update the rmtp value and restart: */
751 if (mode
== HRTIMER_ABS
)
752 return -ERESTARTNOHAND
;
754 tu
= ktime_to_timespec(rem
);
756 if (rmtp
&& copy_to_user(rmtp
, &tu
, sizeof(tu
)))
759 restart
= ¤t_thread_info()->restart_block
;
760 restart
->fn
= nanosleep_restart
;
761 restart
->arg0
= timer
.expires
.tv64
& 0xFFFFFFFF;
762 restart
->arg1
= timer
.expires
.tv64
>> 32;
763 restart
->arg2
= (unsigned long) rmtp
;
764 restart
->arg3
= (unsigned long) timer
.base
->index
;
766 return -ERESTART_RESTARTBLOCK
;
770 sys_nanosleep(struct timespec __user
*rqtp
, struct timespec __user
*rmtp
)
774 if (copy_from_user(&tu
, rqtp
, sizeof(tu
)))
777 if (!timespec_valid(&tu
))
780 return hrtimer_nanosleep(&tu
, rmtp
, HRTIMER_REL
, CLOCK_MONOTONIC
);
784 * Functions related to boot-time initialization:
786 static void __devinit
init_hrtimers_cpu(int cpu
)
788 struct hrtimer_base
*base
= per_cpu(hrtimer_bases
, cpu
);
791 for (i
= 0; i
< MAX_HRTIMER_BASES
; i
++, base
++)
792 spin_lock_init(&base
->lock
);
795 #ifdef CONFIG_HOTPLUG_CPU
797 static void migrate_hrtimer_list(struct hrtimer_base
*old_base
,
798 struct hrtimer_base
*new_base
)
800 struct hrtimer
*timer
;
801 struct rb_node
*node
;
803 while ((node
= rb_first(&old_base
->active
))) {
804 timer
= rb_entry(node
, struct hrtimer
, node
);
805 __remove_hrtimer(timer
, old_base
);
806 timer
->base
= new_base
;
807 enqueue_hrtimer(timer
, new_base
);
811 static void migrate_hrtimers(int cpu
)
813 struct hrtimer_base
*old_base
, *new_base
;
816 BUG_ON(cpu_online(cpu
));
817 old_base
= per_cpu(hrtimer_bases
, cpu
);
818 new_base
= get_cpu_var(hrtimer_bases
);
822 for (i
= 0; i
< MAX_HRTIMER_BASES
; i
++) {
824 spin_lock(&new_base
->lock
);
825 spin_lock(&old_base
->lock
);
827 BUG_ON(old_base
->curr_timer
);
829 migrate_hrtimer_list(old_base
, new_base
);
831 spin_unlock(&old_base
->lock
);
832 spin_unlock(&new_base
->lock
);
838 put_cpu_var(hrtimer_bases
);
840 #endif /* CONFIG_HOTPLUG_CPU */
842 static int __devinit
hrtimer_cpu_notify(struct notifier_block
*self
,
843 unsigned long action
, void *hcpu
)
845 long cpu
= (long)hcpu
;
850 init_hrtimers_cpu(cpu
);
853 #ifdef CONFIG_HOTPLUG_CPU
855 migrate_hrtimers(cpu
);
866 static struct notifier_block __devinitdata hrtimers_nb
= {
867 .notifier_call
= hrtimer_cpu_notify
,
870 void __init
hrtimers_init(void)
872 hrtimer_cpu_notify(&hrtimers_nb
, (unsigned long)CPU_UP_PREPARE
,
873 (void *)(long)smp_processor_id());
874 register_cpu_notifier(&hrtimers_nb
);