Add a 00-INDEX file to Documentation/sysctl/
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / kernel / hrtimer.c
blobdc8a4451d79b15d64a37caf55c9dc4595c6ec5d2
1 /*
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:
15 * - itimers
16 * - POSIX timers
17 * - nanosleep
18 * - precise in-kernel timing
20 * Started by: Thomas Gleixner and Ingo Molnar
22 * Credits:
23 * based on kernel/timer.c
25 * Help, testing, suggestions, bugfixes, improvements were
26 * provided by:
28 * George Anzinger, Andrew Morton, Steven Rostedt, Roman Zippel
29 * et. al.
31 * For licencing details see kernel-base/COPYING
34 #include <linux/cpu.h>
35 #include <linux/irq.h>
36 #include <linux/module.h>
37 #include <linux/percpu.h>
38 #include <linux/hrtimer.h>
39 #include <linux/notifier.h>
40 #include <linux/syscalls.h>
41 #include <linux/kallsyms.h>
42 #include <linux/interrupt.h>
43 #include <linux/tick.h>
44 #include <linux/seq_file.h>
45 #include <linux/err.h>
47 #include <asm/uaccess.h>
49 /**
50 * ktime_get - get the monotonic time in ktime_t format
52 * returns the time in ktime_t format
54 ktime_t ktime_get(void)
56 struct timespec now;
58 ktime_get_ts(&now);
60 return timespec_to_ktime(now);
62 EXPORT_SYMBOL_GPL(ktime_get);
64 /**
65 * ktime_get_real - get the real (wall-) time in ktime_t format
67 * returns the time in ktime_t format
69 ktime_t ktime_get_real(void)
71 struct timespec now;
73 getnstimeofday(&now);
75 return timespec_to_ktime(now);
78 EXPORT_SYMBOL_GPL(ktime_get_real);
81 * The timer bases:
83 * Note: If we want to add new timer bases, we have to skip the two
84 * clock ids captured by the cpu-timers. We do this by holding empty
85 * entries rather than doing math adjustment of the clock ids.
86 * This ensures that we capture erroneous accesses to these clock ids
87 * rather than moving them into the range of valid clock id's.
89 DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
92 .clock_base =
95 .index = CLOCK_REALTIME,
96 .get_time = &ktime_get_real,
97 .resolution = KTIME_LOW_RES,
100 .index = CLOCK_MONOTONIC,
101 .get_time = &ktime_get,
102 .resolution = KTIME_LOW_RES,
108 * ktime_get_ts - get the monotonic clock in timespec format
109 * @ts: pointer to timespec variable
111 * The function calculates the monotonic clock from the realtime
112 * clock and the wall_to_monotonic offset and stores the result
113 * in normalized timespec format in the variable pointed to by @ts.
115 void ktime_get_ts(struct timespec *ts)
117 struct timespec tomono;
118 unsigned long seq;
120 do {
121 seq = read_seqbegin(&xtime_lock);
122 getnstimeofday(ts);
123 tomono = wall_to_monotonic;
125 } while (read_seqretry(&xtime_lock, seq));
127 set_normalized_timespec(ts, ts->tv_sec + tomono.tv_sec,
128 ts->tv_nsec + tomono.tv_nsec);
130 EXPORT_SYMBOL_GPL(ktime_get_ts);
133 * Get the coarse grained time at the softirq based on xtime and
134 * wall_to_monotonic.
136 static void hrtimer_get_softirq_time(struct hrtimer_cpu_base *base)
138 ktime_t xtim, tomono;
139 struct timespec xts, tom;
140 unsigned long seq;
142 do {
143 seq = read_seqbegin(&xtime_lock);
144 xts = current_kernel_time();
145 tom = wall_to_monotonic;
146 } while (read_seqretry(&xtime_lock, seq));
148 xtim = timespec_to_ktime(xts);
149 tomono = timespec_to_ktime(tom);
150 base->clock_base[CLOCK_REALTIME].softirq_time = xtim;
151 base->clock_base[CLOCK_MONOTONIC].softirq_time =
152 ktime_add(xtim, tomono);
156 * Helper function to check, whether the timer is running the callback
157 * function
159 static inline int hrtimer_callback_running(struct hrtimer *timer)
161 return timer->state & HRTIMER_STATE_CALLBACK;
165 * Functions and macros which are different for UP/SMP systems are kept in a
166 * single place
168 #ifdef CONFIG_SMP
171 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
172 * means that all timers which are tied to this base via timer->base are
173 * locked, and the base itself is locked too.
175 * So __run_timers/migrate_timers can safely modify all timers which could
176 * be found on the lists/queues.
178 * When the timer's base is locked, and the timer removed from list, it is
179 * possible to set timer->base = NULL and drop the lock: the timer remains
180 * locked.
182 static
183 struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
184 unsigned long *flags)
186 struct hrtimer_clock_base *base;
188 for (;;) {
189 base = timer->base;
190 if (likely(base != NULL)) {
191 spin_lock_irqsave(&base->cpu_base->lock, *flags);
192 if (likely(base == timer->base))
193 return base;
194 /* The timer has migrated to another CPU: */
195 spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
197 cpu_relax();
202 * Switch the timer base to the current CPU when possible.
204 static inline struct hrtimer_clock_base *
205 switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base)
207 struct hrtimer_clock_base *new_base;
208 struct hrtimer_cpu_base *new_cpu_base;
210 new_cpu_base = &__get_cpu_var(hrtimer_bases);
211 new_base = &new_cpu_base->clock_base[base->index];
213 if (base != new_base) {
215 * We are trying to schedule the timer on the local CPU.
216 * However we can't change timer's base while it is running,
217 * so we keep it on the same CPU. No hassle vs. reprogramming
218 * the event source in the high resolution case. The softirq
219 * code will take care of this when the timer function has
220 * completed. There is no conflict as we hold the lock until
221 * the timer is enqueued.
223 if (unlikely(hrtimer_callback_running(timer)))
224 return base;
226 /* See the comment in lock_timer_base() */
227 timer->base = NULL;
228 spin_unlock(&base->cpu_base->lock);
229 spin_lock(&new_base->cpu_base->lock);
230 timer->base = new_base;
232 return new_base;
235 #else /* CONFIG_SMP */
237 static inline struct hrtimer_clock_base *
238 lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
240 struct hrtimer_clock_base *base = timer->base;
242 spin_lock_irqsave(&base->cpu_base->lock, *flags);
244 return base;
247 # define switch_hrtimer_base(t, b) (b)
249 #endif /* !CONFIG_SMP */
252 * Functions for the union type storage format of ktime_t which are
253 * too large for inlining:
255 #if BITS_PER_LONG < 64
256 # ifndef CONFIG_KTIME_SCALAR
258 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
259 * @kt: addend
260 * @nsec: the scalar nsec value to add
262 * Returns the sum of kt and nsec in ktime_t format
264 ktime_t ktime_add_ns(const ktime_t kt, u64 nsec)
266 ktime_t tmp;
268 if (likely(nsec < NSEC_PER_SEC)) {
269 tmp.tv64 = nsec;
270 } else {
271 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
273 tmp = ktime_set((long)nsec, rem);
276 return ktime_add(kt, tmp);
279 EXPORT_SYMBOL_GPL(ktime_add_ns);
282 * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable
283 * @kt: minuend
284 * @nsec: the scalar nsec value to subtract
286 * Returns the subtraction of @nsec from @kt in ktime_t format
288 ktime_t ktime_sub_ns(const ktime_t kt, u64 nsec)
290 ktime_t tmp;
292 if (likely(nsec < NSEC_PER_SEC)) {
293 tmp.tv64 = nsec;
294 } else {
295 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
297 tmp = ktime_set((long)nsec, rem);
300 return ktime_sub(kt, tmp);
303 EXPORT_SYMBOL_GPL(ktime_sub_ns);
304 # endif /* !CONFIG_KTIME_SCALAR */
307 * Divide a ktime value by a nanosecond value
309 unsigned long ktime_divns(const ktime_t kt, s64 div)
311 u64 dclc, inc, dns;
312 int sft = 0;
314 dclc = dns = ktime_to_ns(kt);
315 inc = div;
316 /* Make sure the divisor is less than 2^32: */
317 while (div >> 32) {
318 sft++;
319 div >>= 1;
321 dclc >>= sft;
322 do_div(dclc, (unsigned long) div);
324 return (unsigned long) dclc;
326 #endif /* BITS_PER_LONG >= 64 */
328 /* High resolution timer related functions */
329 #ifdef CONFIG_HIGH_RES_TIMERS
332 * High resolution timer enabled ?
334 static int hrtimer_hres_enabled __read_mostly = 1;
337 * Enable / Disable high resolution mode
339 static int __init setup_hrtimer_hres(char *str)
341 if (!strcmp(str, "off"))
342 hrtimer_hres_enabled = 0;
343 else if (!strcmp(str, "on"))
344 hrtimer_hres_enabled = 1;
345 else
346 return 0;
347 return 1;
350 __setup("highres=", setup_hrtimer_hres);
353 * hrtimer_high_res_enabled - query, if the highres mode is enabled
355 static inline int hrtimer_is_hres_enabled(void)
357 return hrtimer_hres_enabled;
361 * Is the high resolution mode active ?
363 static inline int hrtimer_hres_active(void)
365 return __get_cpu_var(hrtimer_bases).hres_active;
369 * Reprogram the event source with checking both queues for the
370 * next event
371 * Called with interrupts disabled and base->lock held
373 static void hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base)
375 int i;
376 struct hrtimer_clock_base *base = cpu_base->clock_base;
377 ktime_t expires;
379 cpu_base->expires_next.tv64 = KTIME_MAX;
381 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
382 struct hrtimer *timer;
384 if (!base->first)
385 continue;
386 timer = rb_entry(base->first, struct hrtimer, node);
387 expires = ktime_sub(timer->expires, base->offset);
388 if (expires.tv64 < cpu_base->expires_next.tv64)
389 cpu_base->expires_next = expires;
392 if (cpu_base->expires_next.tv64 != KTIME_MAX)
393 tick_program_event(cpu_base->expires_next, 1);
397 * Shared reprogramming for clock_realtime and clock_monotonic
399 * When a timer is enqueued and expires earlier than the already enqueued
400 * timers, we have to check, whether it expires earlier than the timer for
401 * which the clock event device was armed.
403 * Called with interrupts disabled and base->cpu_base.lock held
405 static int hrtimer_reprogram(struct hrtimer *timer,
406 struct hrtimer_clock_base *base)
408 ktime_t *expires_next = &__get_cpu_var(hrtimer_bases).expires_next;
409 ktime_t expires = ktime_sub(timer->expires, base->offset);
410 int res;
413 * When the callback is running, we do not reprogram the clock event
414 * device. The timer callback is either running on a different CPU or
415 * the callback is executed in the hrtimer_interupt context. The
416 * reprogramming is handled either by the softirq, which called the
417 * callback or at the end of the hrtimer_interrupt.
419 if (hrtimer_callback_running(timer))
420 return 0;
422 if (expires.tv64 >= expires_next->tv64)
423 return 0;
426 * Clockevents returns -ETIME, when the event was in the past.
428 res = tick_program_event(expires, 0);
429 if (!IS_ERR_VALUE(res))
430 *expires_next = expires;
431 return res;
436 * Retrigger next event is called after clock was set
438 * Called with interrupts disabled via on_each_cpu()
440 static void retrigger_next_event(void *arg)
442 struct hrtimer_cpu_base *base;
443 struct timespec realtime_offset;
444 unsigned long seq;
446 if (!hrtimer_hres_active())
447 return;
449 do {
450 seq = read_seqbegin(&xtime_lock);
451 set_normalized_timespec(&realtime_offset,
452 -wall_to_monotonic.tv_sec,
453 -wall_to_monotonic.tv_nsec);
454 } while (read_seqretry(&xtime_lock, seq));
456 base = &__get_cpu_var(hrtimer_bases);
458 /* Adjust CLOCK_REALTIME offset */
459 spin_lock(&base->lock);
460 base->clock_base[CLOCK_REALTIME].offset =
461 timespec_to_ktime(realtime_offset);
463 hrtimer_force_reprogram(base);
464 spin_unlock(&base->lock);
468 * Clock realtime was set
470 * Change the offset of the realtime clock vs. the monotonic
471 * clock.
473 * We might have to reprogram the high resolution timer interrupt. On
474 * SMP we call the architecture specific code to retrigger _all_ high
475 * resolution timer interrupts. On UP we just disable interrupts and
476 * call the high resolution interrupt code.
478 void clock_was_set(void)
480 /* Retrigger the CPU local events everywhere */
481 on_each_cpu(retrigger_next_event, NULL, 0, 1);
485 * During resume we might have to reprogram the high resolution timer
486 * interrupt (on the local CPU):
488 void hres_timers_resume(void)
490 WARN_ON_ONCE(num_online_cpus() > 1);
492 /* Retrigger the CPU local events: */
493 retrigger_next_event(NULL);
497 * Check, whether the timer is on the callback pending list
499 static inline int hrtimer_cb_pending(const struct hrtimer *timer)
501 return timer->state & HRTIMER_STATE_PENDING;
505 * Remove a timer from the callback pending list
507 static inline void hrtimer_remove_cb_pending(struct hrtimer *timer)
509 list_del_init(&timer->cb_entry);
513 * Initialize the high resolution related parts of cpu_base
515 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
517 base->expires_next.tv64 = KTIME_MAX;
518 base->hres_active = 0;
519 INIT_LIST_HEAD(&base->cb_pending);
523 * Initialize the high resolution related parts of a hrtimer
525 static inline void hrtimer_init_timer_hres(struct hrtimer *timer)
527 INIT_LIST_HEAD(&timer->cb_entry);
531 * When High resolution timers are active, try to reprogram. Note, that in case
532 * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry
533 * check happens. The timer gets enqueued into the rbtree. The reprogramming
534 * and expiry check is done in the hrtimer_interrupt or in the softirq.
536 static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
537 struct hrtimer_clock_base *base)
539 if (base->cpu_base->hres_active && hrtimer_reprogram(timer, base)) {
541 /* Timer is expired, act upon the callback mode */
542 switch(timer->cb_mode) {
543 case HRTIMER_CB_IRQSAFE_NO_RESTART:
545 * We can call the callback from here. No restart
546 * happens, so no danger of recursion
548 BUG_ON(timer->function(timer) != HRTIMER_NORESTART);
549 return 1;
550 case HRTIMER_CB_IRQSAFE_NO_SOFTIRQ:
552 * This is solely for the sched tick emulation with
553 * dynamic tick support to ensure that we do not
554 * restart the tick right on the edge and end up with
555 * the tick timer in the softirq ! The calling site
556 * takes care of this.
558 return 1;
559 case HRTIMER_CB_IRQSAFE:
560 case HRTIMER_CB_SOFTIRQ:
562 * Move everything else into the softirq pending list !
564 list_add_tail(&timer->cb_entry,
565 &base->cpu_base->cb_pending);
566 timer->state = HRTIMER_STATE_PENDING;
567 raise_softirq(HRTIMER_SOFTIRQ);
568 return 1;
569 default:
570 BUG();
573 return 0;
577 * Switch to high resolution mode
579 static int hrtimer_switch_to_hres(void)
581 int cpu = smp_processor_id();
582 struct hrtimer_cpu_base *base = &per_cpu(hrtimer_bases, cpu);
583 unsigned long flags;
585 if (base->hres_active)
586 return 1;
588 local_irq_save(flags);
590 if (tick_init_highres()) {
591 local_irq_restore(flags);
592 printk(KERN_WARNING "Could not switch to high resolution "
593 "mode on CPU %d\n", cpu);
594 return 0;
596 base->hres_active = 1;
597 base->clock_base[CLOCK_REALTIME].resolution = KTIME_HIGH_RES;
598 base->clock_base[CLOCK_MONOTONIC].resolution = KTIME_HIGH_RES;
600 tick_setup_sched_timer();
602 /* "Retrigger" the interrupt to get things going */
603 retrigger_next_event(NULL);
604 local_irq_restore(flags);
605 printk(KERN_INFO "Switched to high resolution mode on CPU %d\n",
606 smp_processor_id());
607 return 1;
610 #else
612 static inline int hrtimer_hres_active(void) { return 0; }
613 static inline int hrtimer_is_hres_enabled(void) { return 0; }
614 static inline int hrtimer_switch_to_hres(void) { return 0; }
615 static inline void hrtimer_force_reprogram(struct hrtimer_cpu_base *base) { }
616 static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
617 struct hrtimer_clock_base *base)
619 return 0;
621 static inline int hrtimer_cb_pending(struct hrtimer *timer) { return 0; }
622 static inline void hrtimer_remove_cb_pending(struct hrtimer *timer) { }
623 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { }
624 static inline void hrtimer_init_timer_hres(struct hrtimer *timer) { }
626 #endif /* CONFIG_HIGH_RES_TIMERS */
628 #ifdef CONFIG_TIMER_STATS
629 void __timer_stats_hrtimer_set_start_info(struct hrtimer *timer, void *addr)
631 if (timer->start_site)
632 return;
634 timer->start_site = addr;
635 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
636 timer->start_pid = current->pid;
638 #endif
641 * Counterpart to lock_timer_base above:
643 static inline
644 void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
646 spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
650 * hrtimer_forward - forward the timer expiry
651 * @timer: hrtimer to forward
652 * @now: forward past this time
653 * @interval: the interval to forward
655 * Forward the timer expiry so it will expire in the future.
656 * Returns the number of overruns.
658 unsigned long
659 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
661 unsigned long orun = 1;
662 ktime_t delta;
664 delta = ktime_sub(now, timer->expires);
666 if (delta.tv64 < 0)
667 return 0;
669 if (interval.tv64 < timer->base->resolution.tv64)
670 interval.tv64 = timer->base->resolution.tv64;
672 if (unlikely(delta.tv64 >= interval.tv64)) {
673 s64 incr = ktime_to_ns(interval);
675 orun = ktime_divns(delta, incr);
676 timer->expires = ktime_add_ns(timer->expires, incr * orun);
677 if (timer->expires.tv64 > now.tv64)
678 return orun;
680 * This (and the ktime_add() below) is the
681 * correction for exact:
683 orun++;
685 timer->expires = ktime_add(timer->expires, interval);
687 * Make sure, that the result did not wrap with a very large
688 * interval.
690 if (timer->expires.tv64 < 0)
691 timer->expires = ktime_set(KTIME_SEC_MAX, 0);
693 return orun;
695 EXPORT_SYMBOL_GPL(hrtimer_forward);
698 * enqueue_hrtimer - internal function to (re)start a timer
700 * The timer is inserted in expiry order. Insertion into the
701 * red black tree is O(log(n)). Must hold the base lock.
703 static void enqueue_hrtimer(struct hrtimer *timer,
704 struct hrtimer_clock_base *base, int reprogram)
706 struct rb_node **link = &base->active.rb_node;
707 struct rb_node *parent = NULL;
708 struct hrtimer *entry;
709 int leftmost = 1;
712 * Find the right place in the rbtree:
714 while (*link) {
715 parent = *link;
716 entry = rb_entry(parent, struct hrtimer, node);
718 * We dont care about collisions. Nodes with
719 * the same expiry time stay together.
721 if (timer->expires.tv64 < entry->expires.tv64) {
722 link = &(*link)->rb_left;
723 } else {
724 link = &(*link)->rb_right;
725 leftmost = 0;
730 * Insert the timer to the rbtree and check whether it
731 * replaces the first pending timer
733 if (leftmost) {
735 * Reprogram the clock event device. When the timer is already
736 * expired hrtimer_enqueue_reprogram has either called the
737 * callback or added it to the pending list and raised the
738 * softirq.
740 * This is a NOP for !HIGHRES
742 if (reprogram && hrtimer_enqueue_reprogram(timer, base))
743 return;
745 base->first = &timer->node;
748 rb_link_node(&timer->node, parent, link);
749 rb_insert_color(&timer->node, &base->active);
751 * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
752 * state of a possibly running callback.
754 timer->state |= HRTIMER_STATE_ENQUEUED;
758 * __remove_hrtimer - internal function to remove a timer
760 * Caller must hold the base lock.
762 * High resolution timer mode reprograms the clock event device when the
763 * timer is the one which expires next. The caller can disable this by setting
764 * reprogram to zero. This is useful, when the context does a reprogramming
765 * anyway (e.g. timer interrupt)
767 static void __remove_hrtimer(struct hrtimer *timer,
768 struct hrtimer_clock_base *base,
769 unsigned long newstate, int reprogram)
771 /* High res. callback list. NOP for !HIGHRES */
772 if (hrtimer_cb_pending(timer))
773 hrtimer_remove_cb_pending(timer);
774 else {
776 * Remove the timer from the rbtree and replace the
777 * first entry pointer if necessary.
779 if (base->first == &timer->node) {
780 base->first = rb_next(&timer->node);
781 /* Reprogram the clock event device. if enabled */
782 if (reprogram && hrtimer_hres_active())
783 hrtimer_force_reprogram(base->cpu_base);
785 rb_erase(&timer->node, &base->active);
787 timer->state = newstate;
791 * remove hrtimer, called with base lock held
793 static inline int
794 remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base)
796 if (hrtimer_is_queued(timer)) {
797 int reprogram;
800 * Remove the timer and force reprogramming when high
801 * resolution mode is active and the timer is on the current
802 * CPU. If we remove a timer on another CPU, reprogramming is
803 * skipped. The interrupt event on this CPU is fired and
804 * reprogramming happens in the interrupt handler. This is a
805 * rare case and less expensive than a smp call.
807 timer_stats_hrtimer_clear_start_info(timer);
808 reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases);
809 __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE,
810 reprogram);
811 return 1;
813 return 0;
817 * hrtimer_start - (re)start an relative timer on the current CPU
818 * @timer: the timer to be added
819 * @tim: expiry time
820 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
822 * Returns:
823 * 0 on success
824 * 1 when the timer was active
827 hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode)
829 struct hrtimer_clock_base *base, *new_base;
830 unsigned long flags;
831 int ret;
833 base = lock_hrtimer_base(timer, &flags);
835 /* Remove an active timer from the queue: */
836 ret = remove_hrtimer(timer, base);
838 /* Switch the timer base, if necessary: */
839 new_base = switch_hrtimer_base(timer, base);
841 if (mode == HRTIMER_MODE_REL) {
842 tim = ktime_add(tim, new_base->get_time());
844 * CONFIG_TIME_LOW_RES is a temporary way for architectures
845 * to signal that they simply return xtime in
846 * do_gettimeoffset(). In this case we want to round up by
847 * resolution when starting a relative timer, to avoid short
848 * timeouts. This will go away with the GTOD framework.
850 #ifdef CONFIG_TIME_LOW_RES
851 tim = ktime_add(tim, base->resolution);
852 #endif
854 timer->expires = tim;
856 timer_stats_hrtimer_set_start_info(timer);
859 * Only allow reprogramming if the new base is on this CPU.
860 * (it might still be on another CPU if the timer was pending)
862 enqueue_hrtimer(timer, new_base,
863 new_base->cpu_base == &__get_cpu_var(hrtimer_bases));
865 unlock_hrtimer_base(timer, &flags);
867 return ret;
869 EXPORT_SYMBOL_GPL(hrtimer_start);
872 * hrtimer_try_to_cancel - try to deactivate a timer
873 * @timer: hrtimer to stop
875 * Returns:
876 * 0 when the timer was not active
877 * 1 when the timer was active
878 * -1 when the timer is currently excuting the callback function and
879 * cannot be stopped
881 int hrtimer_try_to_cancel(struct hrtimer *timer)
883 struct hrtimer_clock_base *base;
884 unsigned long flags;
885 int ret = -1;
887 base = lock_hrtimer_base(timer, &flags);
889 if (!hrtimer_callback_running(timer))
890 ret = remove_hrtimer(timer, base);
892 unlock_hrtimer_base(timer, &flags);
894 return ret;
897 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
900 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
901 * @timer: the timer to be cancelled
903 * Returns:
904 * 0 when the timer was not active
905 * 1 when the timer was active
907 int hrtimer_cancel(struct hrtimer *timer)
909 for (;;) {
910 int ret = hrtimer_try_to_cancel(timer);
912 if (ret >= 0)
913 return ret;
914 cpu_relax();
917 EXPORT_SYMBOL_GPL(hrtimer_cancel);
920 * hrtimer_get_remaining - get remaining time for the timer
921 * @timer: the timer to read
923 ktime_t hrtimer_get_remaining(const struct hrtimer *timer)
925 struct hrtimer_clock_base *base;
926 unsigned long flags;
927 ktime_t rem;
929 base = lock_hrtimer_base(timer, &flags);
930 rem = ktime_sub(timer->expires, base->get_time());
931 unlock_hrtimer_base(timer, &flags);
933 return rem;
935 EXPORT_SYMBOL_GPL(hrtimer_get_remaining);
937 #if defined(CONFIG_NO_IDLE_HZ) || defined(CONFIG_NO_HZ)
939 * hrtimer_get_next_event - get the time until next expiry event
941 * Returns the delta to the next expiry event or KTIME_MAX if no timer
942 * is pending.
944 ktime_t hrtimer_get_next_event(void)
946 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
947 struct hrtimer_clock_base *base = cpu_base->clock_base;
948 ktime_t delta, mindelta = { .tv64 = KTIME_MAX };
949 unsigned long flags;
950 int i;
952 spin_lock_irqsave(&cpu_base->lock, flags);
954 if (!hrtimer_hres_active()) {
955 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
956 struct hrtimer *timer;
958 if (!base->first)
959 continue;
961 timer = rb_entry(base->first, struct hrtimer, node);
962 delta.tv64 = timer->expires.tv64;
963 delta = ktime_sub(delta, base->get_time());
964 if (delta.tv64 < mindelta.tv64)
965 mindelta.tv64 = delta.tv64;
969 spin_unlock_irqrestore(&cpu_base->lock, flags);
971 if (mindelta.tv64 < 0)
972 mindelta.tv64 = 0;
973 return mindelta;
975 #endif
978 * hrtimer_init - initialize a timer to the given clock
979 * @timer: the timer to be initialized
980 * @clock_id: the clock to be used
981 * @mode: timer mode abs/rel
983 void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
984 enum hrtimer_mode mode)
986 struct hrtimer_cpu_base *cpu_base;
988 memset(timer, 0, sizeof(struct hrtimer));
990 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
992 if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
993 clock_id = CLOCK_MONOTONIC;
995 timer->base = &cpu_base->clock_base[clock_id];
996 hrtimer_init_timer_hres(timer);
998 #ifdef CONFIG_TIMER_STATS
999 timer->start_site = NULL;
1000 timer->start_pid = -1;
1001 memset(timer->start_comm, 0, TASK_COMM_LEN);
1002 #endif
1004 EXPORT_SYMBOL_GPL(hrtimer_init);
1007 * hrtimer_get_res - get the timer resolution for a clock
1008 * @which_clock: which clock to query
1009 * @tp: pointer to timespec variable to store the resolution
1011 * Store the resolution of the clock selected by @which_clock in the
1012 * variable pointed to by @tp.
1014 int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp)
1016 struct hrtimer_cpu_base *cpu_base;
1018 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1019 *tp = ktime_to_timespec(cpu_base->clock_base[which_clock].resolution);
1021 return 0;
1023 EXPORT_SYMBOL_GPL(hrtimer_get_res);
1025 #ifdef CONFIG_HIGH_RES_TIMERS
1028 * High resolution timer interrupt
1029 * Called with interrupts disabled
1031 void hrtimer_interrupt(struct clock_event_device *dev)
1033 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1034 struct hrtimer_clock_base *base;
1035 ktime_t expires_next, now;
1036 int i, raise = 0;
1038 BUG_ON(!cpu_base->hres_active);
1039 cpu_base->nr_events++;
1040 dev->next_event.tv64 = KTIME_MAX;
1042 retry:
1043 now = ktime_get();
1045 expires_next.tv64 = KTIME_MAX;
1047 base = cpu_base->clock_base;
1049 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1050 ktime_t basenow;
1051 struct rb_node *node;
1053 spin_lock(&cpu_base->lock);
1055 basenow = ktime_add(now, base->offset);
1057 while ((node = base->first)) {
1058 struct hrtimer *timer;
1060 timer = rb_entry(node, struct hrtimer, node);
1062 if (basenow.tv64 < timer->expires.tv64) {
1063 ktime_t expires;
1065 expires = ktime_sub(timer->expires,
1066 base->offset);
1067 if (expires.tv64 < expires_next.tv64)
1068 expires_next = expires;
1069 break;
1072 /* Move softirq callbacks to the pending list */
1073 if (timer->cb_mode == HRTIMER_CB_SOFTIRQ) {
1074 __remove_hrtimer(timer, base,
1075 HRTIMER_STATE_PENDING, 0);
1076 list_add_tail(&timer->cb_entry,
1077 &base->cpu_base->cb_pending);
1078 raise = 1;
1079 continue;
1082 __remove_hrtimer(timer, base,
1083 HRTIMER_STATE_CALLBACK, 0);
1084 timer_stats_account_hrtimer(timer);
1087 * Note: We clear the CALLBACK bit after
1088 * enqueue_hrtimer to avoid reprogramming of
1089 * the event hardware. This happens at the end
1090 * of this function anyway.
1092 if (timer->function(timer) != HRTIMER_NORESTART) {
1093 BUG_ON(timer->state != HRTIMER_STATE_CALLBACK);
1094 enqueue_hrtimer(timer, base, 0);
1096 timer->state &= ~HRTIMER_STATE_CALLBACK;
1098 spin_unlock(&cpu_base->lock);
1099 base++;
1102 cpu_base->expires_next = expires_next;
1104 /* Reprogramming necessary ? */
1105 if (expires_next.tv64 != KTIME_MAX) {
1106 if (tick_program_event(expires_next, 0))
1107 goto retry;
1110 /* Raise softirq ? */
1111 if (raise)
1112 raise_softirq(HRTIMER_SOFTIRQ);
1115 static void run_hrtimer_softirq(struct softirq_action *h)
1117 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1119 spin_lock_irq(&cpu_base->lock);
1121 while (!list_empty(&cpu_base->cb_pending)) {
1122 enum hrtimer_restart (*fn)(struct hrtimer *);
1123 struct hrtimer *timer;
1124 int restart;
1126 timer = list_entry(cpu_base->cb_pending.next,
1127 struct hrtimer, cb_entry);
1129 timer_stats_account_hrtimer(timer);
1131 fn = timer->function;
1132 __remove_hrtimer(timer, timer->base, HRTIMER_STATE_CALLBACK, 0);
1133 spin_unlock_irq(&cpu_base->lock);
1135 restart = fn(timer);
1137 spin_lock_irq(&cpu_base->lock);
1139 timer->state &= ~HRTIMER_STATE_CALLBACK;
1140 if (restart == HRTIMER_RESTART) {
1141 BUG_ON(hrtimer_active(timer));
1143 * Enqueue the timer, allow reprogramming of the event
1144 * device
1146 enqueue_hrtimer(timer, timer->base, 1);
1147 } else if (hrtimer_active(timer)) {
1149 * If the timer was rearmed on another CPU, reprogram
1150 * the event device.
1152 if (timer->base->first == &timer->node)
1153 hrtimer_reprogram(timer, timer->base);
1156 spin_unlock_irq(&cpu_base->lock);
1159 #endif /* CONFIG_HIGH_RES_TIMERS */
1162 * Expire the per base hrtimer-queue:
1164 static inline void run_hrtimer_queue(struct hrtimer_cpu_base *cpu_base,
1165 int index)
1167 struct rb_node *node;
1168 struct hrtimer_clock_base *base = &cpu_base->clock_base[index];
1170 if (!base->first)
1171 return;
1173 if (base->get_softirq_time)
1174 base->softirq_time = base->get_softirq_time();
1176 spin_lock_irq(&cpu_base->lock);
1178 while ((node = base->first)) {
1179 struct hrtimer *timer;
1180 enum hrtimer_restart (*fn)(struct hrtimer *);
1181 int restart;
1183 timer = rb_entry(node, struct hrtimer, node);
1184 if (base->softirq_time.tv64 <= timer->expires.tv64)
1185 break;
1187 #ifdef CONFIG_HIGH_RES_TIMERS
1188 WARN_ON_ONCE(timer->cb_mode == HRTIMER_CB_IRQSAFE_NO_SOFTIRQ);
1189 #endif
1190 timer_stats_account_hrtimer(timer);
1192 fn = timer->function;
1193 __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0);
1194 spin_unlock_irq(&cpu_base->lock);
1196 restart = fn(timer);
1198 spin_lock_irq(&cpu_base->lock);
1200 timer->state &= ~HRTIMER_STATE_CALLBACK;
1201 if (restart != HRTIMER_NORESTART) {
1202 BUG_ON(hrtimer_active(timer));
1203 enqueue_hrtimer(timer, base, 0);
1206 spin_unlock_irq(&cpu_base->lock);
1210 * Called from timer softirq every jiffy, expire hrtimers:
1212 * For HRT its the fall back code to run the softirq in the timer
1213 * softirq context in case the hrtimer initialization failed or has
1214 * not been done yet.
1216 void hrtimer_run_queues(void)
1218 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1219 int i;
1221 if (hrtimer_hres_active())
1222 return;
1225 * This _is_ ugly: We have to check in the softirq context,
1226 * whether we can switch to highres and / or nohz mode. The
1227 * clocksource switch happens in the timer interrupt with
1228 * xtime_lock held. Notification from there only sets the
1229 * check bit in the tick_oneshot code, otherwise we might
1230 * deadlock vs. xtime_lock.
1232 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1233 if (hrtimer_switch_to_hres())
1234 return;
1236 hrtimer_get_softirq_time(cpu_base);
1238 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
1239 run_hrtimer_queue(cpu_base, i);
1243 * Sleep related functions:
1245 static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1247 struct hrtimer_sleeper *t =
1248 container_of(timer, struct hrtimer_sleeper, timer);
1249 struct task_struct *task = t->task;
1251 t->task = NULL;
1252 if (task)
1253 wake_up_process(task);
1255 return HRTIMER_NORESTART;
1258 void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1260 sl->timer.function = hrtimer_wakeup;
1261 sl->task = task;
1262 #ifdef CONFIG_HIGH_RES_TIMERS
1263 sl->timer.cb_mode = HRTIMER_CB_IRQSAFE_NO_RESTART;
1264 #endif
1267 static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1269 hrtimer_init_sleeper(t, current);
1271 do {
1272 set_current_state(TASK_INTERRUPTIBLE);
1273 hrtimer_start(&t->timer, t->timer.expires, mode);
1275 if (likely(t->task))
1276 schedule();
1278 hrtimer_cancel(&t->timer);
1279 mode = HRTIMER_MODE_ABS;
1281 } while (t->task && !signal_pending(current));
1283 return t->task == NULL;
1286 long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1288 struct hrtimer_sleeper t;
1289 struct timespec __user *rmtp;
1290 struct timespec tu;
1291 ktime_t time;
1293 restart->fn = do_no_restart_syscall;
1295 hrtimer_init(&t.timer, restart->arg0, HRTIMER_MODE_ABS);
1296 t.timer.expires.tv64 = ((u64)restart->arg3 << 32) | (u64) restart->arg2;
1298 if (do_nanosleep(&t, HRTIMER_MODE_ABS))
1299 return 0;
1301 rmtp = (struct timespec __user *) restart->arg1;
1302 if (rmtp) {
1303 time = ktime_sub(t.timer.expires, t.timer.base->get_time());
1304 if (time.tv64 <= 0)
1305 return 0;
1306 tu = ktime_to_timespec(time);
1307 if (copy_to_user(rmtp, &tu, sizeof(tu)))
1308 return -EFAULT;
1311 restart->fn = hrtimer_nanosleep_restart;
1313 /* The other values in restart are already filled in */
1314 return -ERESTART_RESTARTBLOCK;
1317 long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
1318 const enum hrtimer_mode mode, const clockid_t clockid)
1320 struct restart_block *restart;
1321 struct hrtimer_sleeper t;
1322 struct timespec tu;
1323 ktime_t rem;
1325 hrtimer_init(&t.timer, clockid, mode);
1326 t.timer.expires = timespec_to_ktime(*rqtp);
1327 if (do_nanosleep(&t, mode))
1328 return 0;
1330 /* Absolute timers do not update the rmtp value and restart: */
1331 if (mode == HRTIMER_MODE_ABS)
1332 return -ERESTARTNOHAND;
1334 if (rmtp) {
1335 rem = ktime_sub(t.timer.expires, t.timer.base->get_time());
1336 if (rem.tv64 <= 0)
1337 return 0;
1338 tu = ktime_to_timespec(rem);
1339 if (copy_to_user(rmtp, &tu, sizeof(tu)))
1340 return -EFAULT;
1343 restart = &current_thread_info()->restart_block;
1344 restart->fn = hrtimer_nanosleep_restart;
1345 restart->arg0 = (unsigned long) t.timer.base->index;
1346 restart->arg1 = (unsigned long) rmtp;
1347 restart->arg2 = t.timer.expires.tv64 & 0xFFFFFFFF;
1348 restart->arg3 = t.timer.expires.tv64 >> 32;
1350 return -ERESTART_RESTARTBLOCK;
1353 asmlinkage long
1354 sys_nanosleep(struct timespec __user *rqtp, struct timespec __user *rmtp)
1356 struct timespec tu;
1358 if (copy_from_user(&tu, rqtp, sizeof(tu)))
1359 return -EFAULT;
1361 if (!timespec_valid(&tu))
1362 return -EINVAL;
1364 return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1368 * Functions related to boot-time initialization:
1370 static void __devinit init_hrtimers_cpu(int cpu)
1372 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1373 int i;
1375 spin_lock_init(&cpu_base->lock);
1376 lockdep_set_class(&cpu_base->lock, &cpu_base->lock_key);
1378 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
1379 cpu_base->clock_base[i].cpu_base = cpu_base;
1381 hrtimer_init_hres(cpu_base);
1384 #ifdef CONFIG_HOTPLUG_CPU
1386 static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1387 struct hrtimer_clock_base *new_base)
1389 struct hrtimer *timer;
1390 struct rb_node *node;
1392 while ((node = rb_first(&old_base->active))) {
1393 timer = rb_entry(node, struct hrtimer, node);
1394 BUG_ON(hrtimer_callback_running(timer));
1395 __remove_hrtimer(timer, old_base, HRTIMER_STATE_INACTIVE, 0);
1396 timer->base = new_base;
1398 * Enqueue the timer. Allow reprogramming of the event device
1400 enqueue_hrtimer(timer, new_base, 1);
1404 static void migrate_hrtimers(int cpu)
1406 struct hrtimer_cpu_base *old_base, *new_base;
1407 int i;
1409 BUG_ON(cpu_online(cpu));
1410 old_base = &per_cpu(hrtimer_bases, cpu);
1411 new_base = &get_cpu_var(hrtimer_bases);
1413 tick_cancel_sched_timer(cpu);
1415 local_irq_disable();
1416 double_spin_lock(&new_base->lock, &old_base->lock,
1417 smp_processor_id() < cpu);
1419 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1420 migrate_hrtimer_list(&old_base->clock_base[i],
1421 &new_base->clock_base[i]);
1424 double_spin_unlock(&new_base->lock, &old_base->lock,
1425 smp_processor_id() < cpu);
1426 local_irq_enable();
1427 put_cpu_var(hrtimer_bases);
1429 #endif /* CONFIG_HOTPLUG_CPU */
1431 static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self,
1432 unsigned long action, void *hcpu)
1434 unsigned int cpu = (long)hcpu;
1436 switch (action) {
1438 case CPU_UP_PREPARE:
1439 case CPU_UP_PREPARE_FROZEN:
1440 init_hrtimers_cpu(cpu);
1441 break;
1443 #ifdef CONFIG_HOTPLUG_CPU
1444 case CPU_DEAD:
1445 case CPU_DEAD_FROZEN:
1446 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &cpu);
1447 migrate_hrtimers(cpu);
1448 break;
1449 #endif
1451 default:
1452 break;
1455 return NOTIFY_OK;
1458 static struct notifier_block __cpuinitdata hrtimers_nb = {
1459 .notifier_call = hrtimer_cpu_notify,
1462 void __init hrtimers_init(void)
1464 hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
1465 (void *)(long)smp_processor_id());
1466 register_cpu_notifier(&hrtimers_nb);
1467 #ifdef CONFIG_HIGH_RES_TIMERS
1468 open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq, NULL);
1469 #endif