Merge git://www.linux-watchdog.org/linux-watchdog
[linux-2.6/cjktty.git] / kernel / hrtimer.c
blobcc47812d3feb0c86cbcf8f03e6ee81340ef7c57c
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/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>
56 * The timer bases:
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 .clock_base =
69 .index = HRTIMER_BASE_MONOTONIC,
70 .clockid = CLOCK_MONOTONIC,
71 .get_time = &ktime_get,
72 .resolution = KTIME_LOW_RES,
75 .index = HRTIMER_BASE_REALTIME,
76 .clockid = CLOCK_REALTIME,
77 .get_time = &ktime_get_real,
78 .resolution = KTIME_LOW_RES,
81 .index = HRTIMER_BASE_BOOTTIME,
82 .clockid = CLOCK_BOOTTIME,
83 .get_time = &ktime_get_boottime,
84 .resolution = KTIME_LOW_RES,
89 static const int hrtimer_clock_to_base_table[MAX_CLOCKS] = {
90 [CLOCK_REALTIME] = HRTIMER_BASE_REALTIME,
91 [CLOCK_MONOTONIC] = HRTIMER_BASE_MONOTONIC,
92 [CLOCK_BOOTTIME] = HRTIMER_BASE_BOOTTIME,
95 static inline int hrtimer_clockid_to_base(clockid_t clock_id)
97 return hrtimer_clock_to_base_table[clock_id];
102 * Get the coarse grained time at the softirq based on xtime and
103 * wall_to_monotonic.
105 static void hrtimer_get_softirq_time(struct hrtimer_cpu_base *base)
107 ktime_t xtim, mono, boot;
108 struct timespec xts, tom, slp;
110 get_xtime_and_monotonic_and_sleep_offset(&xts, &tom, &slp);
112 xtim = timespec_to_ktime(xts);
113 mono = ktime_add(xtim, timespec_to_ktime(tom));
114 boot = ktime_add(mono, timespec_to_ktime(slp));
115 base->clock_base[HRTIMER_BASE_REALTIME].softirq_time = xtim;
116 base->clock_base[HRTIMER_BASE_MONOTONIC].softirq_time = mono;
117 base->clock_base[HRTIMER_BASE_BOOTTIME].softirq_time = boot;
121 * Functions and macros which are different for UP/SMP systems are kept in a
122 * single place
124 #ifdef CONFIG_SMP
127 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
128 * means that all timers which are tied to this base via timer->base are
129 * locked, and the base itself is locked too.
131 * So __run_timers/migrate_timers can safely modify all timers which could
132 * be found on the lists/queues.
134 * When the timer's base is locked, and the timer removed from list, it is
135 * possible to set timer->base = NULL and drop the lock: the timer remains
136 * locked.
138 static
139 struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
140 unsigned long *flags)
142 struct hrtimer_clock_base *base;
144 for (;;) {
145 base = timer->base;
146 if (likely(base != NULL)) {
147 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
148 if (likely(base == timer->base))
149 return base;
150 /* The timer has migrated to another CPU: */
151 raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
153 cpu_relax();
159 * Get the preferred target CPU for NOHZ
161 static int hrtimer_get_target(int this_cpu, int pinned)
163 #ifdef CONFIG_NO_HZ
164 if (!pinned && get_sysctl_timer_migration() && idle_cpu(this_cpu))
165 return get_nohz_timer_target();
166 #endif
167 return this_cpu;
171 * With HIGHRES=y we do not migrate the timer when it is expiring
172 * before the next event on the target cpu because we cannot reprogram
173 * the target cpu hardware and we would cause it to fire late.
175 * Called with cpu_base->lock of target cpu held.
177 static int
178 hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base)
180 #ifdef CONFIG_HIGH_RES_TIMERS
181 ktime_t expires;
183 if (!new_base->cpu_base->hres_active)
184 return 0;
186 expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset);
187 return expires.tv64 <= new_base->cpu_base->expires_next.tv64;
188 #else
189 return 0;
190 #endif
194 * Switch the timer base to the current CPU when possible.
196 static inline struct hrtimer_clock_base *
197 switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
198 int pinned)
200 struct hrtimer_clock_base *new_base;
201 struct hrtimer_cpu_base *new_cpu_base;
202 int this_cpu = smp_processor_id();
203 int cpu = hrtimer_get_target(this_cpu, pinned);
204 int basenum = base->index;
206 again:
207 new_cpu_base = &per_cpu(hrtimer_bases, cpu);
208 new_base = &new_cpu_base->clock_base[basenum];
210 if (base != new_base) {
212 * We are trying to move timer to new_base.
213 * However we can't change timer's base while it is running,
214 * so we keep it on the same CPU. No hassle vs. reprogramming
215 * the event source in the high resolution case. The softirq
216 * code will take care of this when the timer function has
217 * completed. There is no conflict as we hold the lock until
218 * the timer is enqueued.
220 if (unlikely(hrtimer_callback_running(timer)))
221 return base;
223 /* See the comment in lock_timer_base() */
224 timer->base = NULL;
225 raw_spin_unlock(&base->cpu_base->lock);
226 raw_spin_lock(&new_base->cpu_base->lock);
228 if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) {
229 cpu = this_cpu;
230 raw_spin_unlock(&new_base->cpu_base->lock);
231 raw_spin_lock(&base->cpu_base->lock);
232 timer->base = base;
233 goto again;
235 timer->base = new_base;
237 return new_base;
240 #else /* CONFIG_SMP */
242 static inline struct hrtimer_clock_base *
243 lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
245 struct hrtimer_clock_base *base = timer->base;
247 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
249 return base;
252 # define switch_hrtimer_base(t, b, p) (b)
254 #endif /* !CONFIG_SMP */
257 * Functions for the union type storage format of ktime_t which are
258 * too large for inlining:
260 #if BITS_PER_LONG < 64
261 # ifndef CONFIG_KTIME_SCALAR
263 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
264 * @kt: addend
265 * @nsec: the scalar nsec value to add
267 * Returns the sum of kt and nsec in ktime_t format
269 ktime_t ktime_add_ns(const ktime_t kt, u64 nsec)
271 ktime_t tmp;
273 if (likely(nsec < NSEC_PER_SEC)) {
274 tmp.tv64 = nsec;
275 } else {
276 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
278 tmp = ktime_set((long)nsec, rem);
281 return ktime_add(kt, tmp);
284 EXPORT_SYMBOL_GPL(ktime_add_ns);
287 * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable
288 * @kt: minuend
289 * @nsec: the scalar nsec value to subtract
291 * Returns the subtraction of @nsec from @kt in ktime_t format
293 ktime_t ktime_sub_ns(const ktime_t kt, u64 nsec)
295 ktime_t tmp;
297 if (likely(nsec < NSEC_PER_SEC)) {
298 tmp.tv64 = nsec;
299 } else {
300 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
302 tmp = ktime_set((long)nsec, rem);
305 return ktime_sub(kt, tmp);
308 EXPORT_SYMBOL_GPL(ktime_sub_ns);
309 # endif /* !CONFIG_KTIME_SCALAR */
312 * Divide a ktime value by a nanosecond value
314 u64 ktime_divns(const ktime_t kt, s64 div)
316 u64 dclc;
317 int sft = 0;
319 dclc = ktime_to_ns(kt);
320 /* Make sure the divisor is less than 2^32: */
321 while (div >> 32) {
322 sft++;
323 div >>= 1;
325 dclc >>= sft;
326 do_div(dclc, (unsigned long) div);
328 return dclc;
330 #endif /* BITS_PER_LONG >= 64 */
333 * Add two ktime values and do a safety check for overflow:
335 ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
337 ktime_t res = ktime_add(lhs, rhs);
340 * We use KTIME_SEC_MAX here, the maximum timeout which we can
341 * return to user space in a timespec:
343 if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64)
344 res = ktime_set(KTIME_SEC_MAX, 0);
346 return res;
349 EXPORT_SYMBOL_GPL(ktime_add_safe);
351 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
353 static struct debug_obj_descr hrtimer_debug_descr;
355 static void *hrtimer_debug_hint(void *addr)
357 return ((struct hrtimer *) addr)->function;
361 * fixup_init is called when:
362 * - an active object is initialized
364 static int hrtimer_fixup_init(void *addr, enum debug_obj_state state)
366 struct hrtimer *timer = addr;
368 switch (state) {
369 case ODEBUG_STATE_ACTIVE:
370 hrtimer_cancel(timer);
371 debug_object_init(timer, &hrtimer_debug_descr);
372 return 1;
373 default:
374 return 0;
379 * fixup_activate is called when:
380 * - an active object is activated
381 * - an unknown object is activated (might be a statically initialized object)
383 static int hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
385 switch (state) {
387 case ODEBUG_STATE_NOTAVAILABLE:
388 WARN_ON_ONCE(1);
389 return 0;
391 case ODEBUG_STATE_ACTIVE:
392 WARN_ON(1);
394 default:
395 return 0;
400 * fixup_free is called when:
401 * - an active object is freed
403 static int hrtimer_fixup_free(void *addr, enum debug_obj_state state)
405 struct hrtimer *timer = addr;
407 switch (state) {
408 case ODEBUG_STATE_ACTIVE:
409 hrtimer_cancel(timer);
410 debug_object_free(timer, &hrtimer_debug_descr);
411 return 1;
412 default:
413 return 0;
417 static struct debug_obj_descr hrtimer_debug_descr = {
418 .name = "hrtimer",
419 .debug_hint = hrtimer_debug_hint,
420 .fixup_init = hrtimer_fixup_init,
421 .fixup_activate = hrtimer_fixup_activate,
422 .fixup_free = hrtimer_fixup_free,
425 static inline void debug_hrtimer_init(struct hrtimer *timer)
427 debug_object_init(timer, &hrtimer_debug_descr);
430 static inline void debug_hrtimer_activate(struct hrtimer *timer)
432 debug_object_activate(timer, &hrtimer_debug_descr);
435 static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
437 debug_object_deactivate(timer, &hrtimer_debug_descr);
440 static inline void debug_hrtimer_free(struct hrtimer *timer)
442 debug_object_free(timer, &hrtimer_debug_descr);
445 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
446 enum hrtimer_mode mode);
448 void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
449 enum hrtimer_mode mode)
451 debug_object_init_on_stack(timer, &hrtimer_debug_descr);
452 __hrtimer_init(timer, clock_id, mode);
454 EXPORT_SYMBOL_GPL(hrtimer_init_on_stack);
456 void destroy_hrtimer_on_stack(struct hrtimer *timer)
458 debug_object_free(timer, &hrtimer_debug_descr);
461 #else
462 static inline void debug_hrtimer_init(struct hrtimer *timer) { }
463 static inline void debug_hrtimer_activate(struct hrtimer *timer) { }
464 static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
465 #endif
467 static inline void
468 debug_init(struct hrtimer *timer, clockid_t clockid,
469 enum hrtimer_mode mode)
471 debug_hrtimer_init(timer);
472 trace_hrtimer_init(timer, clockid, mode);
475 static inline void debug_activate(struct hrtimer *timer)
477 debug_hrtimer_activate(timer);
478 trace_hrtimer_start(timer);
481 static inline void debug_deactivate(struct hrtimer *timer)
483 debug_hrtimer_deactivate(timer);
484 trace_hrtimer_cancel(timer);
487 /* High resolution timer related functions */
488 #ifdef CONFIG_HIGH_RES_TIMERS
491 * High resolution timer enabled ?
493 static int hrtimer_hres_enabled __read_mostly = 1;
496 * Enable / Disable high resolution mode
498 static int __init setup_hrtimer_hres(char *str)
500 if (!strcmp(str, "off"))
501 hrtimer_hres_enabled = 0;
502 else if (!strcmp(str, "on"))
503 hrtimer_hres_enabled = 1;
504 else
505 return 0;
506 return 1;
509 __setup("highres=", setup_hrtimer_hres);
512 * hrtimer_high_res_enabled - query, if the highres mode is enabled
514 static inline int hrtimer_is_hres_enabled(void)
516 return hrtimer_hres_enabled;
520 * Is the high resolution mode active ?
522 static inline int hrtimer_hres_active(void)
524 return __this_cpu_read(hrtimer_bases.hres_active);
528 * Reprogram the event source with checking both queues for the
529 * next event
530 * Called with interrupts disabled and base->lock held
532 static void
533 hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
535 int i;
536 struct hrtimer_clock_base *base = cpu_base->clock_base;
537 ktime_t expires, expires_next;
539 expires_next.tv64 = KTIME_MAX;
541 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
542 struct hrtimer *timer;
543 struct timerqueue_node *next;
545 next = timerqueue_getnext(&base->active);
546 if (!next)
547 continue;
548 timer = container_of(next, struct hrtimer, node);
550 expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
552 * clock_was_set() has changed base->offset so the
553 * result might be negative. Fix it up to prevent a
554 * false positive in clockevents_program_event()
556 if (expires.tv64 < 0)
557 expires.tv64 = 0;
558 if (expires.tv64 < expires_next.tv64)
559 expires_next = expires;
562 if (skip_equal && expires_next.tv64 == cpu_base->expires_next.tv64)
563 return;
565 cpu_base->expires_next.tv64 = expires_next.tv64;
567 if (cpu_base->expires_next.tv64 != KTIME_MAX)
568 tick_program_event(cpu_base->expires_next, 1);
572 * Shared reprogramming for clock_realtime and clock_monotonic
574 * When a timer is enqueued and expires earlier than the already enqueued
575 * timers, we have to check, whether it expires earlier than the timer for
576 * which the clock event device was armed.
578 * Called with interrupts disabled and base->cpu_base.lock held
580 static int hrtimer_reprogram(struct hrtimer *timer,
581 struct hrtimer_clock_base *base)
583 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
584 ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
585 int res;
587 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
590 * When the callback is running, we do not reprogram the clock event
591 * device. The timer callback is either running on a different CPU or
592 * the callback is executed in the hrtimer_interrupt context. The
593 * reprogramming is handled either by the softirq, which called the
594 * callback or at the end of the hrtimer_interrupt.
596 if (hrtimer_callback_running(timer))
597 return 0;
600 * CLOCK_REALTIME timer might be requested with an absolute
601 * expiry time which is less than base->offset. Nothing wrong
602 * about that, just avoid to call into the tick code, which
603 * has now objections against negative expiry values.
605 if (expires.tv64 < 0)
606 return -ETIME;
608 if (expires.tv64 >= cpu_base->expires_next.tv64)
609 return 0;
612 * If a hang was detected in the last timer interrupt then we
613 * do not schedule a timer which is earlier than the expiry
614 * which we enforced in the hang detection. We want the system
615 * to make progress.
617 if (cpu_base->hang_detected)
618 return 0;
621 * Clockevents returns -ETIME, when the event was in the past.
623 res = tick_program_event(expires, 0);
624 if (!IS_ERR_VALUE(res))
625 cpu_base->expires_next = expires;
626 return res;
630 * Initialize the high resolution related parts of cpu_base
632 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
634 base->expires_next.tv64 = KTIME_MAX;
635 base->hres_active = 0;
639 * When High resolution timers are active, try to reprogram. Note, that in case
640 * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry
641 * check happens. The timer gets enqueued into the rbtree. The reprogramming
642 * and expiry check is done in the hrtimer_interrupt or in the softirq.
644 static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
645 struct hrtimer_clock_base *base)
647 return base->cpu_base->hres_active && hrtimer_reprogram(timer, base);
650 static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base)
652 ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset;
653 ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset;
655 return ktime_get_update_offsets(offs_real, offs_boot);
659 * Retrigger next event is called after clock was set
661 * Called with interrupts disabled via on_each_cpu()
663 static void retrigger_next_event(void *arg)
665 struct hrtimer_cpu_base *base = &__get_cpu_var(hrtimer_bases);
667 if (!hrtimer_hres_active())
668 return;
670 raw_spin_lock(&base->lock);
671 hrtimer_update_base(base);
672 hrtimer_force_reprogram(base, 0);
673 raw_spin_unlock(&base->lock);
677 * Switch to high resolution mode
679 static int hrtimer_switch_to_hres(void)
681 int i, cpu = smp_processor_id();
682 struct hrtimer_cpu_base *base = &per_cpu(hrtimer_bases, cpu);
683 unsigned long flags;
685 if (base->hres_active)
686 return 1;
688 local_irq_save(flags);
690 if (tick_init_highres()) {
691 local_irq_restore(flags);
692 printk(KERN_WARNING "Could not switch to high resolution "
693 "mode on CPU %d\n", cpu);
694 return 0;
696 base->hres_active = 1;
697 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
698 base->clock_base[i].resolution = KTIME_HIGH_RES;
700 tick_setup_sched_timer();
701 /* "Retrigger" the interrupt to get things going */
702 retrigger_next_event(NULL);
703 local_irq_restore(flags);
704 return 1;
708 * Called from timekeeping code to reprogramm the hrtimer interrupt
709 * device. If called from the timer interrupt context we defer it to
710 * softirq context.
712 void clock_was_set_delayed(void)
714 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
716 cpu_base->clock_was_set = 1;
717 __raise_softirq_irqoff(HRTIMER_SOFTIRQ);
720 #else
722 static inline int hrtimer_hres_active(void) { return 0; }
723 static inline int hrtimer_is_hres_enabled(void) { return 0; }
724 static inline int hrtimer_switch_to_hres(void) { return 0; }
725 static inline void
726 hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { }
727 static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
728 struct hrtimer_clock_base *base)
730 return 0;
732 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { }
733 static inline void retrigger_next_event(void *arg) { }
735 #endif /* CONFIG_HIGH_RES_TIMERS */
738 * Clock realtime was set
740 * Change the offset of the realtime clock vs. the monotonic
741 * clock.
743 * We might have to reprogram the high resolution timer interrupt. On
744 * SMP we call the architecture specific code to retrigger _all_ high
745 * resolution timer interrupts. On UP we just disable interrupts and
746 * call the high resolution interrupt code.
748 void clock_was_set(void)
750 #ifdef CONFIG_HIGH_RES_TIMERS
751 /* Retrigger the CPU local events everywhere */
752 on_each_cpu(retrigger_next_event, NULL, 1);
753 #endif
754 timerfd_clock_was_set();
758 * During resume we might have to reprogram the high resolution timer
759 * interrupt (on the local CPU):
761 void hrtimers_resume(void)
763 WARN_ONCE(!irqs_disabled(),
764 KERN_INFO "hrtimers_resume() called with IRQs enabled!");
766 retrigger_next_event(NULL);
767 timerfd_clock_was_set();
770 static inline void timer_stats_hrtimer_set_start_info(struct hrtimer *timer)
772 #ifdef CONFIG_TIMER_STATS
773 if (timer->start_site)
774 return;
775 timer->start_site = __builtin_return_address(0);
776 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
777 timer->start_pid = current->pid;
778 #endif
781 static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer *timer)
783 #ifdef CONFIG_TIMER_STATS
784 timer->start_site = NULL;
785 #endif
788 static inline void timer_stats_account_hrtimer(struct hrtimer *timer)
790 #ifdef CONFIG_TIMER_STATS
791 if (likely(!timer_stats_active))
792 return;
793 timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
794 timer->function, timer->start_comm, 0);
795 #endif
799 * Counterpart to lock_hrtimer_base above:
801 static inline
802 void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
804 raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
808 * hrtimer_forward - forward the timer expiry
809 * @timer: hrtimer to forward
810 * @now: forward past this time
811 * @interval: the interval to forward
813 * Forward the timer expiry so it will expire in the future.
814 * Returns the number of overruns.
816 u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
818 u64 orun = 1;
819 ktime_t delta;
821 delta = ktime_sub(now, hrtimer_get_expires(timer));
823 if (delta.tv64 < 0)
824 return 0;
826 if (interval.tv64 < timer->base->resolution.tv64)
827 interval.tv64 = timer->base->resolution.tv64;
829 if (unlikely(delta.tv64 >= interval.tv64)) {
830 s64 incr = ktime_to_ns(interval);
832 orun = ktime_divns(delta, incr);
833 hrtimer_add_expires_ns(timer, incr * orun);
834 if (hrtimer_get_expires_tv64(timer) > now.tv64)
835 return orun;
837 * This (and the ktime_add() below) is the
838 * correction for exact:
840 orun++;
842 hrtimer_add_expires(timer, interval);
844 return orun;
846 EXPORT_SYMBOL_GPL(hrtimer_forward);
849 * enqueue_hrtimer - internal function to (re)start a timer
851 * The timer is inserted in expiry order. Insertion into the
852 * red black tree is O(log(n)). Must hold the base lock.
854 * Returns 1 when the new timer is the leftmost timer in the tree.
856 static int enqueue_hrtimer(struct hrtimer *timer,
857 struct hrtimer_clock_base *base)
859 debug_activate(timer);
861 timerqueue_add(&base->active, &timer->node);
862 base->cpu_base->active_bases |= 1 << base->index;
865 * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
866 * state of a possibly running callback.
868 timer->state |= HRTIMER_STATE_ENQUEUED;
870 return (&timer->node == base->active.next);
874 * __remove_hrtimer - internal function to remove a timer
876 * Caller must hold the base lock.
878 * High resolution timer mode reprograms the clock event device when the
879 * timer is the one which expires next. The caller can disable this by setting
880 * reprogram to zero. This is useful, when the context does a reprogramming
881 * anyway (e.g. timer interrupt)
883 static void __remove_hrtimer(struct hrtimer *timer,
884 struct hrtimer_clock_base *base,
885 unsigned long newstate, int reprogram)
887 struct timerqueue_node *next_timer;
888 if (!(timer->state & HRTIMER_STATE_ENQUEUED))
889 goto out;
891 next_timer = timerqueue_getnext(&base->active);
892 timerqueue_del(&base->active, &timer->node);
893 if (&timer->node == next_timer) {
894 #ifdef CONFIG_HIGH_RES_TIMERS
895 /* Reprogram the clock event device. if enabled */
896 if (reprogram && hrtimer_hres_active()) {
897 ktime_t expires;
899 expires = ktime_sub(hrtimer_get_expires(timer),
900 base->offset);
901 if (base->cpu_base->expires_next.tv64 == expires.tv64)
902 hrtimer_force_reprogram(base->cpu_base, 1);
904 #endif
906 if (!timerqueue_getnext(&base->active))
907 base->cpu_base->active_bases &= ~(1 << base->index);
908 out:
909 timer->state = newstate;
913 * remove hrtimer, called with base lock held
915 static inline int
916 remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base)
918 if (hrtimer_is_queued(timer)) {
919 unsigned long state;
920 int reprogram;
923 * Remove the timer and force reprogramming when high
924 * resolution mode is active and the timer is on the current
925 * CPU. If we remove a timer on another CPU, reprogramming is
926 * skipped. The interrupt event on this CPU is fired and
927 * reprogramming happens in the interrupt handler. This is a
928 * rare case and less expensive than a smp call.
930 debug_deactivate(timer);
931 timer_stats_hrtimer_clear_start_info(timer);
932 reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases);
934 * We must preserve the CALLBACK state flag here,
935 * otherwise we could move the timer base in
936 * switch_hrtimer_base.
938 state = timer->state & HRTIMER_STATE_CALLBACK;
939 __remove_hrtimer(timer, base, state, reprogram);
940 return 1;
942 return 0;
945 int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
946 unsigned long delta_ns, const enum hrtimer_mode mode,
947 int wakeup)
949 struct hrtimer_clock_base *base, *new_base;
950 unsigned long flags;
951 int ret, leftmost;
953 base = lock_hrtimer_base(timer, &flags);
955 /* Remove an active timer from the queue: */
956 ret = remove_hrtimer(timer, base);
958 /* Switch the timer base, if necessary: */
959 new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED);
961 if (mode & HRTIMER_MODE_REL) {
962 tim = ktime_add_safe(tim, new_base->get_time());
964 * CONFIG_TIME_LOW_RES is a temporary way for architectures
965 * to signal that they simply return xtime in
966 * do_gettimeoffset(). In this case we want to round up by
967 * resolution when starting a relative timer, to avoid short
968 * timeouts. This will go away with the GTOD framework.
970 #ifdef CONFIG_TIME_LOW_RES
971 tim = ktime_add_safe(tim, base->resolution);
972 #endif
975 hrtimer_set_expires_range_ns(timer, tim, delta_ns);
977 timer_stats_hrtimer_set_start_info(timer);
979 leftmost = enqueue_hrtimer(timer, new_base);
982 * Only allow reprogramming if the new base is on this CPU.
983 * (it might still be on another CPU if the timer was pending)
985 * XXX send_remote_softirq() ?
987 if (leftmost && new_base->cpu_base == &__get_cpu_var(hrtimer_bases)
988 && hrtimer_enqueue_reprogram(timer, new_base)) {
989 if (wakeup) {
991 * We need to drop cpu_base->lock to avoid a
992 * lock ordering issue vs. rq->lock.
994 raw_spin_unlock(&new_base->cpu_base->lock);
995 raise_softirq_irqoff(HRTIMER_SOFTIRQ);
996 local_irq_restore(flags);
997 return ret;
998 } else {
999 __raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1003 unlock_hrtimer_base(timer, &flags);
1005 return ret;
1009 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
1010 * @timer: the timer to be added
1011 * @tim: expiry time
1012 * @delta_ns: "slack" range for the timer
1013 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
1015 * Returns:
1016 * 0 on success
1017 * 1 when the timer was active
1019 int hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1020 unsigned long delta_ns, const enum hrtimer_mode mode)
1022 return __hrtimer_start_range_ns(timer, tim, delta_ns, mode, 1);
1024 EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
1027 * hrtimer_start - (re)start an hrtimer on the current CPU
1028 * @timer: the timer to be added
1029 * @tim: expiry time
1030 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
1032 * Returns:
1033 * 0 on success
1034 * 1 when the timer was active
1037 hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode)
1039 return __hrtimer_start_range_ns(timer, tim, 0, mode, 1);
1041 EXPORT_SYMBOL_GPL(hrtimer_start);
1045 * hrtimer_try_to_cancel - try to deactivate a timer
1046 * @timer: hrtimer to stop
1048 * Returns:
1049 * 0 when the timer was not active
1050 * 1 when the timer was active
1051 * -1 when the timer is currently excuting the callback function and
1052 * cannot be stopped
1054 int hrtimer_try_to_cancel(struct hrtimer *timer)
1056 struct hrtimer_clock_base *base;
1057 unsigned long flags;
1058 int ret = -1;
1060 base = lock_hrtimer_base(timer, &flags);
1062 if (!hrtimer_callback_running(timer))
1063 ret = remove_hrtimer(timer, base);
1065 unlock_hrtimer_base(timer, &flags);
1067 return ret;
1070 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1073 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1074 * @timer: the timer to be cancelled
1076 * Returns:
1077 * 0 when the timer was not active
1078 * 1 when the timer was active
1080 int hrtimer_cancel(struct hrtimer *timer)
1082 for (;;) {
1083 int ret = hrtimer_try_to_cancel(timer);
1085 if (ret >= 0)
1086 return ret;
1087 cpu_relax();
1090 EXPORT_SYMBOL_GPL(hrtimer_cancel);
1093 * hrtimer_get_remaining - get remaining time for the timer
1094 * @timer: the timer to read
1096 ktime_t hrtimer_get_remaining(const struct hrtimer *timer)
1098 unsigned long flags;
1099 ktime_t rem;
1101 lock_hrtimer_base(timer, &flags);
1102 rem = hrtimer_expires_remaining(timer);
1103 unlock_hrtimer_base(timer, &flags);
1105 return rem;
1107 EXPORT_SYMBOL_GPL(hrtimer_get_remaining);
1109 #ifdef CONFIG_NO_HZ
1111 * hrtimer_get_next_event - get the time until next expiry event
1113 * Returns the delta to the next expiry event or KTIME_MAX if no timer
1114 * is pending.
1116 ktime_t hrtimer_get_next_event(void)
1118 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1119 struct hrtimer_clock_base *base = cpu_base->clock_base;
1120 ktime_t delta, mindelta = { .tv64 = KTIME_MAX };
1121 unsigned long flags;
1122 int i;
1124 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1126 if (!hrtimer_hres_active()) {
1127 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
1128 struct hrtimer *timer;
1129 struct timerqueue_node *next;
1131 next = timerqueue_getnext(&base->active);
1132 if (!next)
1133 continue;
1135 timer = container_of(next, struct hrtimer, node);
1136 delta.tv64 = hrtimer_get_expires_tv64(timer);
1137 delta = ktime_sub(delta, base->get_time());
1138 if (delta.tv64 < mindelta.tv64)
1139 mindelta.tv64 = delta.tv64;
1143 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1145 if (mindelta.tv64 < 0)
1146 mindelta.tv64 = 0;
1147 return mindelta;
1149 #endif
1151 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1152 enum hrtimer_mode mode)
1154 struct hrtimer_cpu_base *cpu_base;
1155 int base;
1157 memset(timer, 0, sizeof(struct hrtimer));
1159 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1161 if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
1162 clock_id = CLOCK_MONOTONIC;
1164 base = hrtimer_clockid_to_base(clock_id);
1165 timer->base = &cpu_base->clock_base[base];
1166 timerqueue_init(&timer->node);
1168 #ifdef CONFIG_TIMER_STATS
1169 timer->start_site = NULL;
1170 timer->start_pid = -1;
1171 memset(timer->start_comm, 0, TASK_COMM_LEN);
1172 #endif
1176 * hrtimer_init - initialize a timer to the given clock
1177 * @timer: the timer to be initialized
1178 * @clock_id: the clock to be used
1179 * @mode: timer mode abs/rel
1181 void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1182 enum hrtimer_mode mode)
1184 debug_init(timer, clock_id, mode);
1185 __hrtimer_init(timer, clock_id, mode);
1187 EXPORT_SYMBOL_GPL(hrtimer_init);
1190 * hrtimer_get_res - get the timer resolution for a clock
1191 * @which_clock: which clock to query
1192 * @tp: pointer to timespec variable to store the resolution
1194 * Store the resolution of the clock selected by @which_clock in the
1195 * variable pointed to by @tp.
1197 int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp)
1199 struct hrtimer_cpu_base *cpu_base;
1200 int base = hrtimer_clockid_to_base(which_clock);
1202 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1203 *tp = ktime_to_timespec(cpu_base->clock_base[base].resolution);
1205 return 0;
1207 EXPORT_SYMBOL_GPL(hrtimer_get_res);
1209 static void __run_hrtimer(struct hrtimer *timer, ktime_t *now)
1211 struct hrtimer_clock_base *base = timer->base;
1212 struct hrtimer_cpu_base *cpu_base = base->cpu_base;
1213 enum hrtimer_restart (*fn)(struct hrtimer *);
1214 int restart;
1216 WARN_ON(!irqs_disabled());
1218 debug_deactivate(timer);
1219 __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0);
1220 timer_stats_account_hrtimer(timer);
1221 fn = timer->function;
1224 * Because we run timers from hardirq context, there is no chance
1225 * they get migrated to another cpu, therefore its safe to unlock
1226 * the timer base.
1228 raw_spin_unlock(&cpu_base->lock);
1229 trace_hrtimer_expire_entry(timer, now);
1230 restart = fn(timer);
1231 trace_hrtimer_expire_exit(timer);
1232 raw_spin_lock(&cpu_base->lock);
1235 * Note: We clear the CALLBACK bit after enqueue_hrtimer and
1236 * we do not reprogramm the event hardware. Happens either in
1237 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1239 if (restart != HRTIMER_NORESTART) {
1240 BUG_ON(timer->state != HRTIMER_STATE_CALLBACK);
1241 enqueue_hrtimer(timer, base);
1244 WARN_ON_ONCE(!(timer->state & HRTIMER_STATE_CALLBACK));
1246 timer->state &= ~HRTIMER_STATE_CALLBACK;
1249 #ifdef CONFIG_HIGH_RES_TIMERS
1252 * High resolution timer interrupt
1253 * Called with interrupts disabled
1255 void hrtimer_interrupt(struct clock_event_device *dev)
1257 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1258 ktime_t expires_next, now, entry_time, delta;
1259 int i, retries = 0;
1261 BUG_ON(!cpu_base->hres_active);
1262 cpu_base->nr_events++;
1263 dev->next_event.tv64 = KTIME_MAX;
1265 raw_spin_lock(&cpu_base->lock);
1266 entry_time = now = hrtimer_update_base(cpu_base);
1267 retry:
1268 expires_next.tv64 = KTIME_MAX;
1270 * We set expires_next to KTIME_MAX here with cpu_base->lock
1271 * held to prevent that a timer is enqueued in our queue via
1272 * the migration code. This does not affect enqueueing of
1273 * timers which run their callback and need to be requeued on
1274 * this CPU.
1276 cpu_base->expires_next.tv64 = KTIME_MAX;
1278 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1279 struct hrtimer_clock_base *base;
1280 struct timerqueue_node *node;
1281 ktime_t basenow;
1283 if (!(cpu_base->active_bases & (1 << i)))
1284 continue;
1286 base = cpu_base->clock_base + i;
1287 basenow = ktime_add(now, base->offset);
1289 while ((node = timerqueue_getnext(&base->active))) {
1290 struct hrtimer *timer;
1292 timer = container_of(node, struct hrtimer, node);
1295 * The immediate goal for using the softexpires is
1296 * minimizing wakeups, not running timers at the
1297 * earliest interrupt after their soft expiration.
1298 * This allows us to avoid using a Priority Search
1299 * Tree, which can answer a stabbing querry for
1300 * overlapping intervals and instead use the simple
1301 * BST we already have.
1302 * We don't add extra wakeups by delaying timers that
1303 * are right-of a not yet expired timer, because that
1304 * timer will have to trigger a wakeup anyway.
1307 if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer)) {
1308 ktime_t expires;
1310 expires = ktime_sub(hrtimer_get_expires(timer),
1311 base->offset);
1312 if (expires.tv64 < expires_next.tv64)
1313 expires_next = expires;
1314 break;
1317 __run_hrtimer(timer, &basenow);
1322 * Store the new expiry value so the migration code can verify
1323 * against it.
1325 cpu_base->expires_next = expires_next;
1326 raw_spin_unlock(&cpu_base->lock);
1328 /* Reprogramming necessary ? */
1329 if (expires_next.tv64 == KTIME_MAX ||
1330 !tick_program_event(expires_next, 0)) {
1331 cpu_base->hang_detected = 0;
1332 return;
1336 * The next timer was already expired due to:
1337 * - tracing
1338 * - long lasting callbacks
1339 * - being scheduled away when running in a VM
1341 * We need to prevent that we loop forever in the hrtimer
1342 * interrupt routine. We give it 3 attempts to avoid
1343 * overreacting on some spurious event.
1345 * Acquire base lock for updating the offsets and retrieving
1346 * the current time.
1348 raw_spin_lock(&cpu_base->lock);
1349 now = hrtimer_update_base(cpu_base);
1350 cpu_base->nr_retries++;
1351 if (++retries < 3)
1352 goto retry;
1354 * Give the system a chance to do something else than looping
1355 * here. We stored the entry time, so we know exactly how long
1356 * we spent here. We schedule the next event this amount of
1357 * time away.
1359 cpu_base->nr_hangs++;
1360 cpu_base->hang_detected = 1;
1361 raw_spin_unlock(&cpu_base->lock);
1362 delta = ktime_sub(now, entry_time);
1363 if (delta.tv64 > cpu_base->max_hang_time.tv64)
1364 cpu_base->max_hang_time = delta;
1366 * Limit it to a sensible value as we enforce a longer
1367 * delay. Give the CPU at least 100ms to catch up.
1369 if (delta.tv64 > 100 * NSEC_PER_MSEC)
1370 expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1371 else
1372 expires_next = ktime_add(now, delta);
1373 tick_program_event(expires_next, 1);
1374 printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n",
1375 ktime_to_ns(delta));
1379 * local version of hrtimer_peek_ahead_timers() called with interrupts
1380 * disabled.
1382 static void __hrtimer_peek_ahead_timers(void)
1384 struct tick_device *td;
1386 if (!hrtimer_hres_active())
1387 return;
1389 td = &__get_cpu_var(tick_cpu_device);
1390 if (td && td->evtdev)
1391 hrtimer_interrupt(td->evtdev);
1395 * hrtimer_peek_ahead_timers -- run soft-expired timers now
1397 * hrtimer_peek_ahead_timers will peek at the timer queue of
1398 * the current cpu and check if there are any timers for which
1399 * the soft expires time has passed. If any such timers exist,
1400 * they are run immediately and then removed from the timer queue.
1403 void hrtimer_peek_ahead_timers(void)
1405 unsigned long flags;
1407 local_irq_save(flags);
1408 __hrtimer_peek_ahead_timers();
1409 local_irq_restore(flags);
1412 static void run_hrtimer_softirq(struct softirq_action *h)
1414 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1416 if (cpu_base->clock_was_set) {
1417 cpu_base->clock_was_set = 0;
1418 clock_was_set();
1421 hrtimer_peek_ahead_timers();
1424 #else /* CONFIG_HIGH_RES_TIMERS */
1426 static inline void __hrtimer_peek_ahead_timers(void) { }
1428 #endif /* !CONFIG_HIGH_RES_TIMERS */
1431 * Called from timer softirq every jiffy, expire hrtimers:
1433 * For HRT its the fall back code to run the softirq in the timer
1434 * softirq context in case the hrtimer initialization failed or has
1435 * not been done yet.
1437 void hrtimer_run_pending(void)
1439 if (hrtimer_hres_active())
1440 return;
1443 * This _is_ ugly: We have to check in the softirq context,
1444 * whether we can switch to highres and / or nohz mode. The
1445 * clocksource switch happens in the timer interrupt with
1446 * xtime_lock held. Notification from there only sets the
1447 * check bit in the tick_oneshot code, otherwise we might
1448 * deadlock vs. xtime_lock.
1450 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1451 hrtimer_switch_to_hres();
1455 * Called from hardirq context every jiffy
1457 void hrtimer_run_queues(void)
1459 struct timerqueue_node *node;
1460 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1461 struct hrtimer_clock_base *base;
1462 int index, gettime = 1;
1464 if (hrtimer_hres_active())
1465 return;
1467 for (index = 0; index < HRTIMER_MAX_CLOCK_BASES; index++) {
1468 base = &cpu_base->clock_base[index];
1469 if (!timerqueue_getnext(&base->active))
1470 continue;
1472 if (gettime) {
1473 hrtimer_get_softirq_time(cpu_base);
1474 gettime = 0;
1477 raw_spin_lock(&cpu_base->lock);
1479 while ((node = timerqueue_getnext(&base->active))) {
1480 struct hrtimer *timer;
1482 timer = container_of(node, struct hrtimer, node);
1483 if (base->softirq_time.tv64 <=
1484 hrtimer_get_expires_tv64(timer))
1485 break;
1487 __run_hrtimer(timer, &base->softirq_time);
1489 raw_spin_unlock(&cpu_base->lock);
1494 * Sleep related functions:
1496 static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1498 struct hrtimer_sleeper *t =
1499 container_of(timer, struct hrtimer_sleeper, timer);
1500 struct task_struct *task = t->task;
1502 t->task = NULL;
1503 if (task)
1504 wake_up_process(task);
1506 return HRTIMER_NORESTART;
1509 void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1511 sl->timer.function = hrtimer_wakeup;
1512 sl->task = task;
1514 EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
1516 static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1518 hrtimer_init_sleeper(t, current);
1520 do {
1521 set_current_state(TASK_INTERRUPTIBLE);
1522 hrtimer_start_expires(&t->timer, mode);
1523 if (!hrtimer_active(&t->timer))
1524 t->task = NULL;
1526 if (likely(t->task))
1527 schedule();
1529 hrtimer_cancel(&t->timer);
1530 mode = HRTIMER_MODE_ABS;
1532 } while (t->task && !signal_pending(current));
1534 __set_current_state(TASK_RUNNING);
1536 return t->task == NULL;
1539 static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp)
1541 struct timespec rmt;
1542 ktime_t rem;
1544 rem = hrtimer_expires_remaining(timer);
1545 if (rem.tv64 <= 0)
1546 return 0;
1547 rmt = ktime_to_timespec(rem);
1549 if (copy_to_user(rmtp, &rmt, sizeof(*rmtp)))
1550 return -EFAULT;
1552 return 1;
1555 long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1557 struct hrtimer_sleeper t;
1558 struct timespec __user *rmtp;
1559 int ret = 0;
1561 hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid,
1562 HRTIMER_MODE_ABS);
1563 hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
1565 if (do_nanosleep(&t, HRTIMER_MODE_ABS))
1566 goto out;
1568 rmtp = restart->nanosleep.rmtp;
1569 if (rmtp) {
1570 ret = update_rmtp(&t.timer, rmtp);
1571 if (ret <= 0)
1572 goto out;
1575 /* The other values in restart are already filled in */
1576 ret = -ERESTART_RESTARTBLOCK;
1577 out:
1578 destroy_hrtimer_on_stack(&t.timer);
1579 return ret;
1582 long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
1583 const enum hrtimer_mode mode, const clockid_t clockid)
1585 struct restart_block *restart;
1586 struct hrtimer_sleeper t;
1587 int ret = 0;
1588 unsigned long slack;
1590 slack = current->timer_slack_ns;
1591 if (rt_task(current))
1592 slack = 0;
1594 hrtimer_init_on_stack(&t.timer, clockid, mode);
1595 hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack);
1596 if (do_nanosleep(&t, mode))
1597 goto out;
1599 /* Absolute timers do not update the rmtp value and restart: */
1600 if (mode == HRTIMER_MODE_ABS) {
1601 ret = -ERESTARTNOHAND;
1602 goto out;
1605 if (rmtp) {
1606 ret = update_rmtp(&t.timer, rmtp);
1607 if (ret <= 0)
1608 goto out;
1611 restart = &current_thread_info()->restart_block;
1612 restart->fn = hrtimer_nanosleep_restart;
1613 restart->nanosleep.clockid = t.timer.base->clockid;
1614 restart->nanosleep.rmtp = rmtp;
1615 restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
1617 ret = -ERESTART_RESTARTBLOCK;
1618 out:
1619 destroy_hrtimer_on_stack(&t.timer);
1620 return ret;
1623 SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp,
1624 struct timespec __user *, rmtp)
1626 struct timespec tu;
1628 if (copy_from_user(&tu, rqtp, sizeof(tu)))
1629 return -EFAULT;
1631 if (!timespec_valid(&tu))
1632 return -EINVAL;
1634 return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1638 * Functions related to boot-time initialization:
1640 static void __cpuinit init_hrtimers_cpu(int cpu)
1642 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1643 int i;
1645 raw_spin_lock_init(&cpu_base->lock);
1647 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1648 cpu_base->clock_base[i].cpu_base = cpu_base;
1649 timerqueue_init_head(&cpu_base->clock_base[i].active);
1652 hrtimer_init_hres(cpu_base);
1655 #ifdef CONFIG_HOTPLUG_CPU
1657 static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1658 struct hrtimer_clock_base *new_base)
1660 struct hrtimer *timer;
1661 struct timerqueue_node *node;
1663 while ((node = timerqueue_getnext(&old_base->active))) {
1664 timer = container_of(node, struct hrtimer, node);
1665 BUG_ON(hrtimer_callback_running(timer));
1666 debug_deactivate(timer);
1669 * Mark it as STATE_MIGRATE not INACTIVE otherwise the
1670 * timer could be seen as !active and just vanish away
1671 * under us on another CPU
1673 __remove_hrtimer(timer, old_base, HRTIMER_STATE_MIGRATE, 0);
1674 timer->base = new_base;
1676 * Enqueue the timers on the new cpu. This does not
1677 * reprogram the event device in case the timer
1678 * expires before the earliest on this CPU, but we run
1679 * hrtimer_interrupt after we migrated everything to
1680 * sort out already expired timers and reprogram the
1681 * event device.
1683 enqueue_hrtimer(timer, new_base);
1685 /* Clear the migration state bit */
1686 timer->state &= ~HRTIMER_STATE_MIGRATE;
1690 static void migrate_hrtimers(int scpu)
1692 struct hrtimer_cpu_base *old_base, *new_base;
1693 int i;
1695 BUG_ON(cpu_online(scpu));
1696 tick_cancel_sched_timer(scpu);
1698 local_irq_disable();
1699 old_base = &per_cpu(hrtimer_bases, scpu);
1700 new_base = &__get_cpu_var(hrtimer_bases);
1702 * The caller is globally serialized and nobody else
1703 * takes two locks at once, deadlock is not possible.
1705 raw_spin_lock(&new_base->lock);
1706 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1708 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1709 migrate_hrtimer_list(&old_base->clock_base[i],
1710 &new_base->clock_base[i]);
1713 raw_spin_unlock(&old_base->lock);
1714 raw_spin_unlock(&new_base->lock);
1716 /* Check, if we got expired work to do */
1717 __hrtimer_peek_ahead_timers();
1718 local_irq_enable();
1721 #endif /* CONFIG_HOTPLUG_CPU */
1723 static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self,
1724 unsigned long action, void *hcpu)
1726 int scpu = (long)hcpu;
1728 switch (action) {
1730 case CPU_UP_PREPARE:
1731 case CPU_UP_PREPARE_FROZEN:
1732 init_hrtimers_cpu(scpu);
1733 break;
1735 #ifdef CONFIG_HOTPLUG_CPU
1736 case CPU_DYING:
1737 case CPU_DYING_FROZEN:
1738 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DYING, &scpu);
1739 break;
1740 case CPU_DEAD:
1741 case CPU_DEAD_FROZEN:
1743 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &scpu);
1744 migrate_hrtimers(scpu);
1745 break;
1747 #endif
1749 default:
1750 break;
1753 return NOTIFY_OK;
1756 static struct notifier_block __cpuinitdata hrtimers_nb = {
1757 .notifier_call = hrtimer_cpu_notify,
1760 void __init hrtimers_init(void)
1762 hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
1763 (void *)(long)smp_processor_id());
1764 register_cpu_notifier(&hrtimers_nb);
1765 #ifdef CONFIG_HIGH_RES_TIMERS
1766 open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq);
1767 #endif
1771 * schedule_hrtimeout_range_clock - sleep until timeout
1772 * @expires: timeout value (ktime_t)
1773 * @delta: slack in expires timeout (ktime_t)
1774 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1775 * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME
1777 int __sched
1778 schedule_hrtimeout_range_clock(ktime_t *expires, unsigned long delta,
1779 const enum hrtimer_mode mode, int clock)
1781 struct hrtimer_sleeper t;
1784 * Optimize when a zero timeout value is given. It does not
1785 * matter whether this is an absolute or a relative time.
1787 if (expires && !expires->tv64) {
1788 __set_current_state(TASK_RUNNING);
1789 return 0;
1793 * A NULL parameter means "infinite"
1795 if (!expires) {
1796 schedule();
1797 __set_current_state(TASK_RUNNING);
1798 return -EINTR;
1801 hrtimer_init_on_stack(&t.timer, clock, mode);
1802 hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
1804 hrtimer_init_sleeper(&t, current);
1806 hrtimer_start_expires(&t.timer, mode);
1807 if (!hrtimer_active(&t.timer))
1808 t.task = NULL;
1810 if (likely(t.task))
1811 schedule();
1813 hrtimer_cancel(&t.timer);
1814 destroy_hrtimer_on_stack(&t.timer);
1816 __set_current_state(TASK_RUNNING);
1818 return !t.task ? 0 : -EINTR;
1822 * schedule_hrtimeout_range - sleep until timeout
1823 * @expires: timeout value (ktime_t)
1824 * @delta: slack in expires timeout (ktime_t)
1825 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1827 * Make the current task sleep until the given expiry time has
1828 * elapsed. The routine will return immediately unless
1829 * the current task state has been set (see set_current_state()).
1831 * The @delta argument gives the kernel the freedom to schedule the
1832 * actual wakeup to a time that is both power and performance friendly.
1833 * The kernel give the normal best effort behavior for "@expires+@delta",
1834 * but may decide to fire the timer earlier, but no earlier than @expires.
1836 * You can set the task state as follows -
1838 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1839 * pass before the routine returns.
1841 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1842 * delivered to the current task.
1844 * The current task state is guaranteed to be TASK_RUNNING when this
1845 * routine returns.
1847 * Returns 0 when the timer has expired otherwise -EINTR
1849 int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta,
1850 const enum hrtimer_mode mode)
1852 return schedule_hrtimeout_range_clock(expires, delta, mode,
1853 CLOCK_MONOTONIC);
1855 EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
1858 * schedule_hrtimeout - sleep until timeout
1859 * @expires: timeout value (ktime_t)
1860 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1862 * Make the current task sleep until the given expiry time has
1863 * elapsed. The routine will return immediately unless
1864 * the current task state has been set (see set_current_state()).
1866 * You can set the task state as follows -
1868 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1869 * pass before the routine returns.
1871 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1872 * delivered to the current task.
1874 * The current task state is guaranteed to be TASK_RUNNING when this
1875 * routine returns.
1877 * Returns 0 when the timer has expired otherwise -EINTR
1879 int __sched schedule_hrtimeout(ktime_t *expires,
1880 const enum hrtimer_mode mode)
1882 return schedule_hrtimeout_range(expires, 0, mode);
1884 EXPORT_SYMBOL_GPL(schedule_hrtimeout);