projects/linux·blob v2.6.30 · refs

1764 lines · 43.1 KiB

1/*
2 * linux/kernel/hrtimer.c
3 *
4 * Copyright(C) 2005-2006, Thomas Gleixner <[email protected]>
5 * Copyright(C) 2005-2007, Red Hat, Inc., Ingo Molnar
6 * Copyright(C) 2006-2007 Timesys Corp., Thomas Gleixner
7 *
8 * High-resolution kernel timers
9 *
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.
13 *
14 * These timers are currently used for:
15 * - itimers
16 * - POSIX timers
17 * - nanosleep
18 * - precise in-kernel timing
19 *
20 * Started by: Thomas Gleixner and Ingo Molnar
21 *
22 * Credits:
23 * based on kernel/timer.c
24 *
25 * Help, testing, suggestions, bugfixes, improvements were
26 * provided by:
27 *
28 * George Anzinger, Andrew Morton, Steven Rostedt, Roman Zippel
29 * et. al.
30 *
31 * For licencing details see kernel-base/COPYING
32 */
33
34#include <linux/cpu.h>
35#include <linux/module.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
47#include <asm/uaccess.h>
48
49/**
50 * ktime_get - get the monotonic time in ktime_t format
51 *
52 * returns the time in ktime_t format
53 */
54ktime_t ktime_get(void)
55{
56 struct timespec now;
57
58 ktime_get_ts(&now);
59
60 return timespec_to_ktime(now);
61}
62EXPORT_SYMBOL_GPL(ktime_get);
63
64/**
65 * ktime_get_real - get the real (wall-) time in ktime_t format
66 *
67 * returns the time in ktime_t format
68 */
69ktime_t ktime_get_real(void)
70{
71 struct timespec now;
72
73 getnstimeofday(&now);
74
75 return timespec_to_ktime(now);
76}
77
78EXPORT_SYMBOL_GPL(ktime_get_real);
79
80/*
81 * The timer bases:
82 *
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.
88 */
89DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
90{
91
92 .clock_base =
93 {
94 {
95 .index = CLOCK_REALTIME,
96 .get_time = &ktime_get_real,
97 .resolution = KTIME_LOW_RES,
98 },
99 {
100 .index = CLOCK_MONOTONIC,
101 .get_time = &ktime_get,
102 .resolution = KTIME_LOW_RES,
103 },
104 }
105};
106
107/**
108 * ktime_get_ts - get the monotonic clock in timespec format
109 * @ts: pointer to timespec variable
110 *
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.
114 */
115void ktime_get_ts(struct timespec *ts)
116{
117 struct timespec tomono;
118 unsigned long seq;
119
120 do {
121 seq = read_seqbegin(&xtime_lock);
122 getnstimeofday(ts);
123 tomono = wall_to_monotonic;
124
125 } while (read_seqretry(&xtime_lock, seq));
126
127 set_normalized_timespec(ts, ts->tv_sec + tomono.tv_sec,
128 ts->tv_nsec + tomono.tv_nsec);
129}
130EXPORT_SYMBOL_GPL(ktime_get_ts);
131
132/*
133 * Get the coarse grained time at the softirq based on xtime and
134 * wall_to_monotonic.
135 */
136static void hrtimer_get_softirq_time(struct hrtimer_cpu_base *base)
137{
138 ktime_t xtim, tomono;
139 struct timespec xts, tom;
140 unsigned long seq;
141
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));
147
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);
153}
154
155/*
156 * Functions and macros which are different for UP/SMP systems are kept in a
157 * single place
158 */
159#ifdef CONFIG_SMP
160
161/*
162 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
163 * means that all timers which are tied to this base via timer->base are
164 * locked, and the base itself is locked too.
165 *
166 * So __run_timers/migrate_timers can safely modify all timers which could
167 * be found on the lists/queues.
168 *
169 * When the timer's base is locked, and the timer removed from list, it is
170 * possible to set timer->base = NULL and drop the lock: the timer remains
171 * locked.
172 */
173static
174struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
175 unsigned long *flags)
176{
177 struct hrtimer_clock_base *base;
178
179 for (;;) {
180 base = timer->base;
181 if (likely(base != NULL)) {
182 spin_lock_irqsave(&base->cpu_base->lock, *flags);
183 if (likely(base == timer->base))
184 return base;
185 /* The timer has migrated to another CPU: */
186 spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
187 }
188 cpu_relax();
189 }
190}
191
192/*
193 * Switch the timer base to the current CPU when possible.
194 */
195static inline struct hrtimer_clock_base *
196switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base)
197{
198 struct hrtimer_clock_base *new_base;
199 struct hrtimer_cpu_base *new_cpu_base;
200
201 new_cpu_base = &__get_cpu_var(hrtimer_bases);
202 new_base = &new_cpu_base->clock_base[base->index];
203
204 if (base != new_base) {
205 /*
206 * We are trying to schedule the timer on the local CPU.
207 * However we can't change timer's base while it is running,
208 * so we keep it on the same CPU. No hassle vs. reprogramming
209 * the event source in the high resolution case. The softirq
210 * code will take care of this when the timer function has
211 * completed. There is no conflict as we hold the lock until
212 * the timer is enqueued.
213 */
214 if (unlikely(hrtimer_callback_running(timer)))
215 return base;
216
217 /* See the comment in lock_timer_base() */
218 timer->base = NULL;
219 spin_unlock(&base->cpu_base->lock);
220 spin_lock(&new_base->cpu_base->lock);
221 timer->base = new_base;
222 }
223 return new_base;
224}
225
226#else /* CONFIG_SMP */
227
228static inline struct hrtimer_clock_base *
229lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
230{
231 struct hrtimer_clock_base *base = timer->base;
232
233 spin_lock_irqsave(&base->cpu_base->lock, *flags);
234
235 return base;
236}
237
238# define switch_hrtimer_base(t, b) (b)
239
240#endif /* !CONFIG_SMP */
241
242/*
243 * Functions for the union type storage format of ktime_t which are
244 * too large for inlining:
245 */
246#if BITS_PER_LONG < 64
247# ifndef CONFIG_KTIME_SCALAR
248/**
249 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
250 * @kt: addend
251 * @nsec: the scalar nsec value to add
252 *
253 * Returns the sum of kt and nsec in ktime_t format
254 */
255ktime_t ktime_add_ns(const ktime_t kt, u64 nsec)
256{
257 ktime_t tmp;
258
259 if (likely(nsec < NSEC_PER_SEC)) {
260 tmp.tv64 = nsec;
261 } else {
262 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
263
264 tmp = ktime_set((long)nsec, rem);
265 }
266
267 return ktime_add(kt, tmp);
268}
269
270EXPORT_SYMBOL_GPL(ktime_add_ns);
271
272/**
273 * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable
274 * @kt: minuend
275 * @nsec: the scalar nsec value to subtract
276 *
277 * Returns the subtraction of @nsec from @kt in ktime_t format
278 */
279ktime_t ktime_sub_ns(const ktime_t kt, u64 nsec)
280{
281 ktime_t tmp;
282
283 if (likely(nsec < NSEC_PER_SEC)) {
284 tmp.tv64 = nsec;
285 } else {
286 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
287
288 tmp = ktime_set((long)nsec, rem);
289 }
290
291 return ktime_sub(kt, tmp);
292}
293
294EXPORT_SYMBOL_GPL(ktime_sub_ns);
295# endif /* !CONFIG_KTIME_SCALAR */
296
297/*
298 * Divide a ktime value by a nanosecond value
299 */
300u64 ktime_divns(const ktime_t kt, s64 div)
301{
302 u64 dclc;
303 int sft = 0;
304
305 dclc = ktime_to_ns(kt);
306 /* Make sure the divisor is less than 2^32: */
307 while (div >> 32) {
308 sft++;
309 div >>= 1;
310 }
311 dclc >>= sft;
312 do_div(dclc, (unsigned long) div);
313
314 return dclc;
315}
316#endif /* BITS_PER_LONG >= 64 */
317
318/*
319 * Add two ktime values and do a safety check for overflow:
320 */
321ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
322{
323 ktime_t res = ktime_add(lhs, rhs);
324
325 /*
326 * We use KTIME_SEC_MAX here, the maximum timeout which we can
327 * return to user space in a timespec:
328 */
329 if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64)
330 res = ktime_set(KTIME_SEC_MAX, 0);
331
332 return res;
333}
334
335#ifdef CONFIG_DEBUG_OBJECTS_TIMERS
336
337static struct debug_obj_descr hrtimer_debug_descr;
338
339/*
340 * fixup_init is called when:
341 * - an active object is initialized
342 */
343static int hrtimer_fixup_init(void *addr, enum debug_obj_state state)
344{
345 struct hrtimer *timer = addr;
346
347 switch (state) {
348 case ODEBUG_STATE_ACTIVE:
349 hrtimer_cancel(timer);
350 debug_object_init(timer, &hrtimer_debug_descr);
351 return 1;
352 default:
353 return 0;
354 }
355}
356
357/*
358 * fixup_activate is called when:
359 * - an active object is activated
360 * - an unknown object is activated (might be a statically initialized object)
361 */
362static int hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
363{
364 switch (state) {
365
366 case ODEBUG_STATE_NOTAVAILABLE:
367 WARN_ON_ONCE(1);
368 return 0;
369
370 case ODEBUG_STATE_ACTIVE:
371 WARN_ON(1);
372
373 default:
374 return 0;
375 }
376}
377
378/*
379 * fixup_free is called when:
380 * - an active object is freed
381 */
382static int hrtimer_fixup_free(void *addr, enum debug_obj_state state)
383{
384 struct hrtimer *timer = addr;
385
386 switch (state) {
387 case ODEBUG_STATE_ACTIVE:
388 hrtimer_cancel(timer);
389 debug_object_free(timer, &hrtimer_debug_descr);
390 return 1;
391 default:
392 return 0;
393 }
394}
395
396static struct debug_obj_descr hrtimer_debug_descr = {
397 .name = "hrtimer",
398 .fixup_init = hrtimer_fixup_init,
399 .fixup_activate = hrtimer_fixup_activate,
400 .fixup_free = hrtimer_fixup_free,
401};
402
403static inline void debug_hrtimer_init(struct hrtimer *timer)
404{
405 debug_object_init(timer, &hrtimer_debug_descr);
406}
407
408static inline void debug_hrtimer_activate(struct hrtimer *timer)
409{
410 debug_object_activate(timer, &hrtimer_debug_descr);
411}
412
413static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
414{
415 debug_object_deactivate(timer, &hrtimer_debug_descr);
416}
417
418static inline void debug_hrtimer_free(struct hrtimer *timer)
419{
420 debug_object_free(timer, &hrtimer_debug_descr);
421}
422
423static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
424 enum hrtimer_mode mode);
425
426void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
427 enum hrtimer_mode mode)
428{
429 debug_object_init_on_stack(timer, &hrtimer_debug_descr);
430 __hrtimer_init(timer, clock_id, mode);
431}
432
433void destroy_hrtimer_on_stack(struct hrtimer *timer)
434{
435 debug_object_free(timer, &hrtimer_debug_descr);
436}
437
438#else
439static inline void debug_hrtimer_init(struct hrtimer *timer) { }
440static inline void debug_hrtimer_activate(struct hrtimer *timer) { }
441static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
442#endif
443
444/* High resolution timer related functions */
445#ifdef CONFIG_HIGH_RES_TIMERS
446
447/*
448 * High resolution timer enabled ?
449 */
450static int hrtimer_hres_enabled __read_mostly = 1;
451
452/*
453 * Enable / Disable high resolution mode
454 */
455static int __init setup_hrtimer_hres(char *str)
456{
457 if (!strcmp(str, "off"))
458 hrtimer_hres_enabled = 0;
459 else if (!strcmp(str, "on"))
460 hrtimer_hres_enabled = 1;
461 else
462 return 0;
463 return 1;
464}
465
466__setup("highres=", setup_hrtimer_hres);
467
468/*
469 * hrtimer_high_res_enabled - query, if the highres mode is enabled
470 */
471static inline int hrtimer_is_hres_enabled(void)
472{
473 return hrtimer_hres_enabled;
474}
475
476/*
477 * Is the high resolution mode active ?
478 */
479static inline int hrtimer_hres_active(void)
480{
481 return __get_cpu_var(hrtimer_bases).hres_active;
482}
483
484/*
485 * Reprogram the event source with checking both queues for the
486 * next event
487 * Called with interrupts disabled and base->lock held
488 */
489static void hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base)
490{
491 int i;
492 struct hrtimer_clock_base *base = cpu_base->clock_base;
493 ktime_t expires;
494
495 cpu_base->expires_next.tv64 = KTIME_MAX;
496
497 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
498 struct hrtimer *timer;
499
500 if (!base->first)
501 continue;
502 timer = rb_entry(base->first, struct hrtimer, node);
503 expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
504 /*
505 * clock_was_set() has changed base->offset so the
506 * result might be negative. Fix it up to prevent a
507 * false positive in clockevents_program_event()
508 */
509 if (expires.tv64 < 0)
510 expires.tv64 = 0;
511 if (expires.tv64 < cpu_base->expires_next.tv64)
512 cpu_base->expires_next = expires;
513 }
514
515 if (cpu_base->expires_next.tv64 != KTIME_MAX)
516 tick_program_event(cpu_base->expires_next, 1);
517}
518
519/*
520 * Shared reprogramming for clock_realtime and clock_monotonic
521 *
522 * When a timer is enqueued and expires earlier than the already enqueued
523 * timers, we have to check, whether it expires earlier than the timer for
524 * which the clock event device was armed.
525 *
526 * Called with interrupts disabled and base->cpu_base.lock held
527 */
528static int hrtimer_reprogram(struct hrtimer *timer,
529 struct hrtimer_clock_base *base)
530{
531 ktime_t *expires_next = &__get_cpu_var(hrtimer_bases).expires_next;
532 ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
533 int res;
534
535 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
536
537 /*
538 * When the callback is running, we do not reprogram the clock event
539 * device. The timer callback is either running on a different CPU or
540 * the callback is executed in the hrtimer_interrupt context. The
541 * reprogramming is handled either by the softirq, which called the
542 * callback or at the end of the hrtimer_interrupt.
543 */
544 if (hrtimer_callback_running(timer))
545 return 0;
546
547 /*
548 * CLOCK_REALTIME timer might be requested with an absolute
549 * expiry time which is less than base->offset. Nothing wrong
550 * about that, just avoid to call into the tick code, which
551 * has now objections against negative expiry values.
552 */
553 if (expires.tv64 < 0)
554 return -ETIME;
555
556 if (expires.tv64 >= expires_next->tv64)
557 return 0;
558
559 /*
560 * Clockevents returns -ETIME, when the event was in the past.
561 */
562 res = tick_program_event(expires, 0);
563 if (!IS_ERR_VALUE(res))
564 *expires_next = expires;
565 return res;
566}
567
568
569/*
570 * Retrigger next event is called after clock was set
571 *
572 * Called with interrupts disabled via on_each_cpu()
573 */
574static void retrigger_next_event(void *arg)
575{
576 struct hrtimer_cpu_base *base;
577 struct timespec realtime_offset;
578 unsigned long seq;
579
580 if (!hrtimer_hres_active())
581 return;
582
583 do {
584 seq = read_seqbegin(&xtime_lock);
585 set_normalized_timespec(&realtime_offset,
586 -wall_to_monotonic.tv_sec,
587 -wall_to_monotonic.tv_nsec);
588 } while (read_seqretry(&xtime_lock, seq));
589
590 base = &__get_cpu_var(hrtimer_bases);
591
592 /* Adjust CLOCK_REALTIME offset */
593 spin_lock(&base->lock);
594 base->clock_base[CLOCK_REALTIME].offset =
595 timespec_to_ktime(realtime_offset);
596
597 hrtimer_force_reprogram(base);
598 spin_unlock(&base->lock);
599}
600
601/*
602 * Clock realtime was set
603 *
604 * Change the offset of the realtime clock vs. the monotonic
605 * clock.
606 *
607 * We might have to reprogram the high resolution timer interrupt. On
608 * SMP we call the architecture specific code to retrigger _all_ high
609 * resolution timer interrupts. On UP we just disable interrupts and
610 * call the high resolution interrupt code.
611 */
612void clock_was_set(void)
613{
614 /* Retrigger the CPU local events everywhere */
615 on_each_cpu(retrigger_next_event, NULL, 1);
616}
617
618/*
619 * During resume we might have to reprogram the high resolution timer
620 * interrupt (on the local CPU):
621 */
622void hres_timers_resume(void)
623{
624 WARN_ONCE(!irqs_disabled(),
625 KERN_INFO "hres_timers_resume() called with IRQs enabled!");
626
627 retrigger_next_event(NULL);
628}
629
630/*
631 * Initialize the high resolution related parts of cpu_base
632 */
633static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
634{
635 base->expires_next.tv64 = KTIME_MAX;
636 base->hres_active = 0;
637}
638
639/*
640 * Initialize the high resolution related parts of a hrtimer
641 */
642static inline void hrtimer_init_timer_hres(struct hrtimer *timer)
643{
644}
645
646
647/*
648 * When High resolution timers are active, try to reprogram. Note, that in case
649 * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry
650 * check happens. The timer gets enqueued into the rbtree. The reprogramming
651 * and expiry check is done in the hrtimer_interrupt or in the softirq.
652 */
653static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
654 struct hrtimer_clock_base *base,
655 int wakeup)
656{
657 if (base->cpu_base->hres_active && hrtimer_reprogram(timer, base)) {
658 if (wakeup) {
659 spin_unlock(&base->cpu_base->lock);
660 raise_softirq_irqoff(HRTIMER_SOFTIRQ);
661 spin_lock(&base->cpu_base->lock);
662 } else
663 __raise_softirq_irqoff(HRTIMER_SOFTIRQ);
664
665 return 1;
666 }
667
668 return 0;
669}
670
671/*
672 * Switch to high resolution mode
673 */
674static int hrtimer_switch_to_hres(void)
675{
676 int cpu = smp_processor_id();
677 struct hrtimer_cpu_base *base = &per_cpu(hrtimer_bases, cpu);
678 unsigned long flags;
679
680 if (base->hres_active)
681 return 1;
682
683 local_irq_save(flags);
684
685 if (tick_init_highres()) {
686 local_irq_restore(flags);
687 printk(KERN_WARNING "Could not switch to high resolution "
688 "mode on CPU %d\n", cpu);
689 return 0;
690 }
691 base->hres_active = 1;
692 base->clock_base[CLOCK_REALTIME].resolution = KTIME_HIGH_RES;
693 base->clock_base[CLOCK_MONOTONIC].resolution = KTIME_HIGH_RES;
694
695 tick_setup_sched_timer();
696
697 /* "Retrigger" the interrupt to get things going */
698 retrigger_next_event(NULL);
699 local_irq_restore(flags);
700 printk(KERN_DEBUG "Switched to high resolution mode on CPU %d\n",
701 smp_processor_id());
702 return 1;
703}
704
705#else
706
707static inline int hrtimer_hres_active(void) { return 0; }
708static inline int hrtimer_is_hres_enabled(void) { return 0; }
709static inline int hrtimer_switch_to_hres(void) { return 0; }
710static inline void hrtimer_force_reprogram(struct hrtimer_cpu_base *base) { }
711static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
712 struct hrtimer_clock_base *base,
713 int wakeup)
714{
715 return 0;
716}
717static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { }
718static inline void hrtimer_init_timer_hres(struct hrtimer *timer) { }
719
720#endif /* CONFIG_HIGH_RES_TIMERS */
721
722#ifdef CONFIG_TIMER_STATS
723void __timer_stats_hrtimer_set_start_info(struct hrtimer *timer, void *addr)
724{
725 if (timer->start_site)
726 return;
727
728 timer->start_site = addr;
729 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
730 timer->start_pid = current->pid;
731}
732#endif
733
734/*
735 * Counterpart to lock_hrtimer_base above:
736 */
737static inline
738void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
739{
740 spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
741}
742
743/**
744 * hrtimer_forward - forward the timer expiry
745 * @timer: hrtimer to forward
746 * @now: forward past this time
747 * @interval: the interval to forward
748 *
749 * Forward the timer expiry so it will expire in the future.
750 * Returns the number of overruns.
751 */
752u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
753{
754 u64 orun = 1;
755 ktime_t delta;
756
757 delta = ktime_sub(now, hrtimer_get_expires(timer));
758
759 if (delta.tv64 < 0)
760 return 0;
761
762 if (interval.tv64 < timer->base->resolution.tv64)
763 interval.tv64 = timer->base->resolution.tv64;
764
765 if (unlikely(delta.tv64 >= interval.tv64)) {
766 s64 incr = ktime_to_ns(interval);
767
768 orun = ktime_divns(delta, incr);
769 hrtimer_add_expires_ns(timer, incr * orun);
770 if (hrtimer_get_expires_tv64(timer) > now.tv64)
771 return orun;
772 /*
773 * This (and the ktime_add() below) is the
774 * correction for exact:
775 */
776 orun++;
777 }
778 hrtimer_add_expires(timer, interval);
779
780 return orun;
781}
782EXPORT_SYMBOL_GPL(hrtimer_forward);
783
784/*
785 * enqueue_hrtimer - internal function to (re)start a timer
786 *
787 * The timer is inserted in expiry order. Insertion into the
788 * red black tree is O(log(n)). Must hold the base lock.
789 *
790 * Returns 1 when the new timer is the leftmost timer in the tree.
791 */
792static int enqueue_hrtimer(struct hrtimer *timer,
793 struct hrtimer_clock_base *base)
794{
795 struct rb_node **link = &base->active.rb_node;
796 struct rb_node *parent = NULL;
797 struct hrtimer *entry;
798 int leftmost = 1;
799
800 debug_hrtimer_activate(timer);
801
802 /*
803 * Find the right place in the rbtree:
804 */
805 while (*link) {
806 parent = *link;
807 entry = rb_entry(parent, struct hrtimer, node);
808 /*
809 * We dont care about collisions. Nodes with
810 * the same expiry time stay together.
811 */
812 if (hrtimer_get_expires_tv64(timer) <
813 hrtimer_get_expires_tv64(entry)) {
814 link = &(*link)->rb_left;
815 } else {
816 link = &(*link)->rb_right;
817 leftmost = 0;
818 }
819 }
820
821 /*
822 * Insert the timer to the rbtree and check whether it
823 * replaces the first pending timer
824 */
825 if (leftmost)
826 base->first = &timer->node;
827
828 rb_link_node(&timer->node, parent, link);
829 rb_insert_color(&timer->node, &base->active);
830 /*
831 * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
832 * state of a possibly running callback.
833 */
834 timer->state |= HRTIMER_STATE_ENQUEUED;
835
836 return leftmost;
837}
838
839/*
840 * __remove_hrtimer - internal function to remove a timer
841 *
842 * Caller must hold the base lock.
843 *
844 * High resolution timer mode reprograms the clock event device when the
845 * timer is the one which expires next. The caller can disable this by setting
846 * reprogram to zero. This is useful, when the context does a reprogramming
847 * anyway (e.g. timer interrupt)
848 */
849static void __remove_hrtimer(struct hrtimer *timer,
850 struct hrtimer_clock_base *base,
851 unsigned long newstate, int reprogram)
852{
853 if (timer->state & HRTIMER_STATE_ENQUEUED) {
854 /*
855 * Remove the timer from the rbtree and replace the
856 * first entry pointer if necessary.
857 */
858 if (base->first == &timer->node) {
859 base->first = rb_next(&timer->node);
860 /* Reprogram the clock event device. if enabled */
861 if (reprogram && hrtimer_hres_active())
862 hrtimer_force_reprogram(base->cpu_base);
863 }
864 rb_erase(&timer->node, &base->active);
865 }
866 timer->state = newstate;
867}
868
869/*
870 * remove hrtimer, called with base lock held
871 */
872static inline int
873remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base)
874{
875 if (hrtimer_is_queued(timer)) {
876 int reprogram;
877
878 /*
879 * Remove the timer and force reprogramming when high
880 * resolution mode is active and the timer is on the current
881 * CPU. If we remove a timer on another CPU, reprogramming is
882 * skipped. The interrupt event on this CPU is fired and
883 * reprogramming happens in the interrupt handler. This is a
884 * rare case and less expensive than a smp call.
885 */
886 debug_hrtimer_deactivate(timer);
887 timer_stats_hrtimer_clear_start_info(timer);
888 reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases);
889 __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE,
890 reprogram);
891 return 1;
892 }
893 return 0;
894}
895
896int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
897 unsigned long delta_ns, const enum hrtimer_mode mode,
898 int wakeup)
899{
900 struct hrtimer_clock_base *base, *new_base;
901 unsigned long flags;
902 int ret, leftmost;
903
904 base = lock_hrtimer_base(timer, &flags);
905
906 /* Remove an active timer from the queue: */
907 ret = remove_hrtimer(timer, base);
908
909 /* Switch the timer base, if necessary: */
910 new_base = switch_hrtimer_base(timer, base);
911
912 if (mode == HRTIMER_MODE_REL) {
913 tim = ktime_add_safe(tim, new_base->get_time());
914 /*
915 * CONFIG_TIME_LOW_RES is a temporary way for architectures
916 * to signal that they simply return xtime in
917 * do_gettimeoffset(). In this case we want to round up by
918 * resolution when starting a relative timer, to avoid short
919 * timeouts. This will go away with the GTOD framework.
920 */
921#ifdef CONFIG_TIME_LOW_RES
922 tim = ktime_add_safe(tim, base->resolution);
923#endif
924 }
925
926 hrtimer_set_expires_range_ns(timer, tim, delta_ns);
927
928 timer_stats_hrtimer_set_start_info(timer);
929
930 leftmost = enqueue_hrtimer(timer, new_base);
931
932 /*
933 * Only allow reprogramming if the new base is on this CPU.
934 * (it might still be on another CPU if the timer was pending)
935 *
936 * XXX send_remote_softirq() ?
937 */
938 if (leftmost && new_base->cpu_base == &__get_cpu_var(hrtimer_bases))
939 hrtimer_enqueue_reprogram(timer, new_base, wakeup);
940
941 unlock_hrtimer_base(timer, &flags);
942
943 return ret;
944}
945
946/**
947 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
948 * @timer: the timer to be added
949 * @tim: expiry time
950 * @delta_ns: "slack" range for the timer
951 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
952 *
953 * Returns:
954 * 0 on success
955 * 1 when the timer was active
956 */
957int hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
958 unsigned long delta_ns, const enum hrtimer_mode mode)
959{
960 return __hrtimer_start_range_ns(timer, tim, delta_ns, mode, 1);
961}
962EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
963
964/**
965 * hrtimer_start - (re)start an hrtimer on the current CPU
966 * @timer: the timer to be added
967 * @tim: expiry time
968 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
969 *
970 * Returns:
971 * 0 on success
972 * 1 when the timer was active
973 */
974int
975hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode)
976{
977 return __hrtimer_start_range_ns(timer, tim, 0, mode, 1);
978}
979EXPORT_SYMBOL_GPL(hrtimer_start);
980
981
982/**
983 * hrtimer_try_to_cancel - try to deactivate a timer
984 * @timer: hrtimer to stop
985 *
986 * Returns:
987 * 0 when the timer was not active
988 * 1 when the timer was active
989 * -1 when the timer is currently excuting the callback function and
990 * cannot be stopped
991 */
992int hrtimer_try_to_cancel(struct hrtimer *timer)
993{
994 struct hrtimer_clock_base *base;
995 unsigned long flags;
996 int ret = -1;
997
998 base = lock_hrtimer_base(timer, &flags);
999
1000 if (!hrtimer_callback_running(timer))
1001 ret = remove_hrtimer(timer, base);
1002
1003 unlock_hrtimer_base(timer, &flags);
1004
1005 return ret;
1006
1007}
1008EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1009
1010/**
1011 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1012 * @timer: the timer to be cancelled
1013 *
1014 * Returns:
1015 * 0 when the timer was not active
1016 * 1 when the timer was active
1017 */
1018int hrtimer_cancel(struct hrtimer *timer)
1019{
1020 for (;;) {
1021 int ret = hrtimer_try_to_cancel(timer);
1022
1023 if (ret >= 0)
1024 return ret;
1025 cpu_relax();
1026 }
1027}
1028EXPORT_SYMBOL_GPL(hrtimer_cancel);
1029
1030/**
1031 * hrtimer_get_remaining - get remaining time for the timer
1032 * @timer: the timer to read
1033 */
1034ktime_t hrtimer_get_remaining(const struct hrtimer *timer)
1035{
1036 struct hrtimer_clock_base *base;
1037 unsigned long flags;
1038 ktime_t rem;
1039
1040 base = lock_hrtimer_base(timer, &flags);
1041 rem = hrtimer_expires_remaining(timer);
1042 unlock_hrtimer_base(timer, &flags);
1043
1044 return rem;
1045}
1046EXPORT_SYMBOL_GPL(hrtimer_get_remaining);
1047
1048#ifdef CONFIG_NO_HZ
1049/**
1050 * hrtimer_get_next_event - get the time until next expiry event
1051 *
1052 * Returns the delta to the next expiry event or KTIME_MAX if no timer
1053 * is pending.
1054 */
1055ktime_t hrtimer_get_next_event(void)
1056{
1057 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1058 struct hrtimer_clock_base *base = cpu_base->clock_base;
1059 ktime_t delta, mindelta = { .tv64 = KTIME_MAX };
1060 unsigned long flags;
1061 int i;
1062
1063 spin_lock_irqsave(&cpu_base->lock, flags);
1064
1065 if (!hrtimer_hres_active()) {
1066 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
1067 struct hrtimer *timer;
1068
1069 if (!base->first)
1070 continue;
1071
1072 timer = rb_entry(base->first, struct hrtimer, node);
1073 delta.tv64 = hrtimer_get_expires_tv64(timer);
1074 delta = ktime_sub(delta, base->get_time());
1075 if (delta.tv64 < mindelta.tv64)
1076 mindelta.tv64 = delta.tv64;
1077 }
1078 }
1079
1080 spin_unlock_irqrestore(&cpu_base->lock, flags);
1081
1082 if (mindelta.tv64 < 0)
1083 mindelta.tv64 = 0;
1084 return mindelta;
1085}
1086#endif
1087
1088static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1089 enum hrtimer_mode mode)
1090{
1091 struct hrtimer_cpu_base *cpu_base;
1092
1093 memset(timer, 0, sizeof(struct hrtimer));
1094
1095 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1096
1097 if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
1098 clock_id = CLOCK_MONOTONIC;
1099
1100 timer->base = &cpu_base->clock_base[clock_id];
1101 INIT_LIST_HEAD(&timer->cb_entry);
1102 hrtimer_init_timer_hres(timer);
1103
1104#ifdef CONFIG_TIMER_STATS
1105 timer->start_site = NULL;
1106 timer->start_pid = -1;
1107 memset(timer->start_comm, 0, TASK_COMM_LEN);
1108#endif
1109}
1110
1111/**
1112 * hrtimer_init - initialize a timer to the given clock
1113 * @timer: the timer to be initialized
1114 * @clock_id: the clock to be used
1115 * @mode: timer mode abs/rel
1116 */
1117void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1118 enum hrtimer_mode mode)
1119{
1120 debug_hrtimer_init(timer);
1121 __hrtimer_init(timer, clock_id, mode);
1122}
1123EXPORT_SYMBOL_GPL(hrtimer_init);
1124
1125/**
1126 * hrtimer_get_res - get the timer resolution for a clock
1127 * @which_clock: which clock to query
1128 * @tp: pointer to timespec variable to store the resolution
1129 *
1130 * Store the resolution of the clock selected by @which_clock in the
1131 * variable pointed to by @tp.
1132 */
1133int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp)
1134{
1135 struct hrtimer_cpu_base *cpu_base;
1136
1137 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1138 *tp = ktime_to_timespec(cpu_base->clock_base[which_clock].resolution);
1139
1140 return 0;
1141}
1142EXPORT_SYMBOL_GPL(hrtimer_get_res);
1143
1144static void __run_hrtimer(struct hrtimer *timer)
1145{
1146 struct hrtimer_clock_base *base = timer->base;
1147 struct hrtimer_cpu_base *cpu_base = base->cpu_base;
1148 enum hrtimer_restart (*fn)(struct hrtimer *);
1149 int restart;
1150
1151 WARN_ON(!irqs_disabled());
1152
1153 debug_hrtimer_deactivate(timer);
1154 __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0);
1155 timer_stats_account_hrtimer(timer);
1156 fn = timer->function;
1157
1158 /*
1159 * Because we run timers from hardirq context, there is no chance
1160 * they get migrated to another cpu, therefore its safe to unlock
1161 * the timer base.
1162 */
1163 spin_unlock(&cpu_base->lock);
1164 restart = fn(timer);
1165 spin_lock(&cpu_base->lock);
1166
1167 /*
1168 * Note: We clear the CALLBACK bit after enqueue_hrtimer and
1169 * we do not reprogramm the event hardware. Happens either in
1170 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1171 */
1172 if (restart != HRTIMER_NORESTART) {
1173 BUG_ON(timer->state != HRTIMER_STATE_CALLBACK);
1174 enqueue_hrtimer(timer, base);
1175 }
1176 timer->state &= ~HRTIMER_STATE_CALLBACK;
1177}
1178
1179#ifdef CONFIG_HIGH_RES_TIMERS
1180
1181static int force_clock_reprogram;
1182
1183/*
1184 * After 5 iteration's attempts, we consider that hrtimer_interrupt()
1185 * is hanging, which could happen with something that slows the interrupt
1186 * such as the tracing. Then we force the clock reprogramming for each future
1187 * hrtimer interrupts to avoid infinite loops and use the min_delta_ns
1188 * threshold that we will overwrite.
1189 * The next tick event will be scheduled to 3 times we currently spend on
1190 * hrtimer_interrupt(). This gives a good compromise, the cpus will spend
1191 * 1/4 of their time to process the hrtimer interrupts. This is enough to
1192 * let it running without serious starvation.
1193 */
1194
1195static inline void
1196hrtimer_interrupt_hanging(struct clock_event_device *dev,
1197 ktime_t try_time)
1198{
1199 force_clock_reprogram = 1;
1200 dev->min_delta_ns = (unsigned long)try_time.tv64 * 3;
1201 printk(KERN_WARNING "hrtimer: interrupt too slow, "
1202 "forcing clock min delta to %lu ns\n", dev->min_delta_ns);
1203}
1204/*
1205 * High resolution timer interrupt
1206 * Called with interrupts disabled
1207 */
1208void hrtimer_interrupt(struct clock_event_device *dev)
1209{
1210 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1211 struct hrtimer_clock_base *base;
1212 ktime_t expires_next, now;
1213 int nr_retries = 0;
1214 int i;
1215
1216 BUG_ON(!cpu_base->hres_active);
1217 cpu_base->nr_events++;
1218 dev->next_event.tv64 = KTIME_MAX;
1219
1220 retry:
1221 /* 5 retries is enough to notice a hang */
1222 if (!(++nr_retries % 5))
1223 hrtimer_interrupt_hanging(dev, ktime_sub(ktime_get(), now));
1224
1225 now = ktime_get();
1226
1227 expires_next.tv64 = KTIME_MAX;
1228
1229 base = cpu_base->clock_base;
1230
1231 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1232 ktime_t basenow;
1233 struct rb_node *node;
1234
1235 spin_lock(&cpu_base->lock);
1236
1237 basenow = ktime_add(now, base->offset);
1238
1239 while ((node = base->first)) {
1240 struct hrtimer *timer;
1241
1242 timer = rb_entry(node, struct hrtimer, node);
1243
1244 /*
1245 * The immediate goal for using the softexpires is
1246 * minimizing wakeups, not running timers at the
1247 * earliest interrupt after their soft expiration.
1248 * This allows us to avoid using a Priority Search
1249 * Tree, which can answer a stabbing querry for
1250 * overlapping intervals and instead use the simple
1251 * BST we already have.
1252 * We don't add extra wakeups by delaying timers that
1253 * are right-of a not yet expired timer, because that
1254 * timer will have to trigger a wakeup anyway.
1255 */
1256
1257 if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer)) {
1258 ktime_t expires;
1259
1260 expires = ktime_sub(hrtimer_get_expires(timer),
1261 base->offset);
1262 if (expires.tv64 < expires_next.tv64)
1263 expires_next = expires;
1264 break;
1265 }
1266
1267 __run_hrtimer(timer);
1268 }
1269 spin_unlock(&cpu_base->lock);
1270 base++;
1271 }
1272
1273 cpu_base->expires_next = expires_next;
1274
1275 /* Reprogramming necessary ? */
1276 if (expires_next.tv64 != KTIME_MAX) {
1277 if (tick_program_event(expires_next, force_clock_reprogram))
1278 goto retry;
1279 }
1280}
1281
1282/*
1283 * local version of hrtimer_peek_ahead_timers() called with interrupts
1284 * disabled.
1285 */
1286static void __hrtimer_peek_ahead_timers(void)
1287{
1288 struct tick_device *td;
1289
1290 if (!hrtimer_hres_active())
1291 return;
1292
1293 td = &__get_cpu_var(tick_cpu_device);
1294 if (td && td->evtdev)
1295 hrtimer_interrupt(td->evtdev);
1296}
1297
1298/**
1299 * hrtimer_peek_ahead_timers -- run soft-expired timers now
1300 *
1301 * hrtimer_peek_ahead_timers will peek at the timer queue of
1302 * the current cpu and check if there are any timers for which
1303 * the soft expires time has passed. If any such timers exist,
1304 * they are run immediately and then removed from the timer queue.
1305 *
1306 */
1307void hrtimer_peek_ahead_timers(void)
1308{
1309 unsigned long flags;
1310
1311 local_irq_save(flags);
1312 __hrtimer_peek_ahead_timers();
1313 local_irq_restore(flags);
1314}
1315
1316static void run_hrtimer_softirq(struct softirq_action *h)
1317{
1318 hrtimer_peek_ahead_timers();
1319}
1320
1321#else /* CONFIG_HIGH_RES_TIMERS */
1322
1323static inline void __hrtimer_peek_ahead_timers(void) { }
1324
1325#endif /* !CONFIG_HIGH_RES_TIMERS */
1326
1327/*
1328 * Called from timer softirq every jiffy, expire hrtimers:
1329 *
1330 * For HRT its the fall back code to run the softirq in the timer
1331 * softirq context in case the hrtimer initialization failed or has
1332 * not been done yet.
1333 */
1334void hrtimer_run_pending(void)
1335{
1336 if (hrtimer_hres_active())
1337 return;
1338
1339 /*
1340 * This _is_ ugly: We have to check in the softirq context,
1341 * whether we can switch to highres and / or nohz mode. The
1342 * clocksource switch happens in the timer interrupt with
1343 * xtime_lock held. Notification from there only sets the
1344 * check bit in the tick_oneshot code, otherwise we might
1345 * deadlock vs. xtime_lock.
1346 */
1347 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1348 hrtimer_switch_to_hres();
1349}
1350
1351/*
1352 * Called from hardirq context every jiffy
1353 */
1354void hrtimer_run_queues(void)
1355{
1356 struct rb_node *node;
1357 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1358 struct hrtimer_clock_base *base;
1359 int index, gettime = 1;
1360
1361 if (hrtimer_hres_active())
1362 return;
1363
1364 for (index = 0; index < HRTIMER_MAX_CLOCK_BASES; index++) {
1365 base = &cpu_base->clock_base[index];
1366
1367 if (!base->first)
1368 continue;
1369
1370 if (gettime) {
1371 hrtimer_get_softirq_time(cpu_base);
1372 gettime = 0;
1373 }
1374
1375 spin_lock(&cpu_base->lock);
1376
1377 while ((node = base->first)) {
1378 struct hrtimer *timer;
1379
1380 timer = rb_entry(node, struct hrtimer, node);
1381 if (base->softirq_time.tv64 <=
1382 hrtimer_get_expires_tv64(timer))
1383 break;
1384
1385 __run_hrtimer(timer);
1386 }
1387 spin_unlock(&cpu_base->lock);
1388 }
1389}
1390
1391/*
1392 * Sleep related functions:
1393 */
1394static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1395{
1396 struct hrtimer_sleeper *t =
1397 container_of(timer, struct hrtimer_sleeper, timer);
1398 struct task_struct *task = t->task;
1399
1400 t->task = NULL;
1401 if (task)
1402 wake_up_process(task);
1403
1404 return HRTIMER_NORESTART;
1405}
1406
1407void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1408{
1409 sl->timer.function = hrtimer_wakeup;
1410 sl->task = task;
1411}
1412
1413static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1414{
1415 hrtimer_init_sleeper(t, current);
1416
1417 do {
1418 set_current_state(TASK_INTERRUPTIBLE);
1419 hrtimer_start_expires(&t->timer, mode);
1420 if (!hrtimer_active(&t->timer))
1421 t->task = NULL;
1422
1423 if (likely(t->task))
1424 schedule();
1425
1426 hrtimer_cancel(&t->timer);
1427 mode = HRTIMER_MODE_ABS;
1428
1429 } while (t->task && !signal_pending(current));
1430
1431 __set_current_state(TASK_RUNNING);
1432
1433 return t->task == NULL;
1434}
1435
1436static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp)
1437{
1438 struct timespec rmt;
1439 ktime_t rem;
1440
1441 rem = hrtimer_expires_remaining(timer);
1442 if (rem.tv64 <= 0)
1443 return 0;
1444 rmt = ktime_to_timespec(rem);
1445
1446 if (copy_to_user(rmtp, &rmt, sizeof(*rmtp)))
1447 return -EFAULT;
1448
1449 return 1;
1450}
1451
1452long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1453{
1454 struct hrtimer_sleeper t;
1455 struct timespec __user *rmtp;
1456 int ret = 0;
1457
1458 hrtimer_init_on_stack(&t.timer, restart->nanosleep.index,
1459 HRTIMER_MODE_ABS);
1460 hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
1461
1462 if (do_nanosleep(&t, HRTIMER_MODE_ABS))
1463 goto out;
1464
1465 rmtp = restart->nanosleep.rmtp;
1466 if (rmtp) {
1467 ret = update_rmtp(&t.timer, rmtp);
1468 if (ret <= 0)
1469 goto out;
1470 }
1471
1472 /* The other values in restart are already filled in */
1473 ret = -ERESTART_RESTARTBLOCK;
1474out:
1475 destroy_hrtimer_on_stack(&t.timer);
1476 return ret;
1477}
1478
1479long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
1480 const enum hrtimer_mode mode, const clockid_t clockid)
1481{
1482 struct restart_block *restart;
1483 struct hrtimer_sleeper t;
1484 int ret = 0;
1485 unsigned long slack;
1486
1487 slack = current->timer_slack_ns;
1488 if (rt_task(current))
1489 slack = 0;
1490
1491 hrtimer_init_on_stack(&t.timer, clockid, mode);
1492 hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack);
1493 if (do_nanosleep(&t, mode))
1494 goto out;
1495
1496 /* Absolute timers do not update the rmtp value and restart: */
1497 if (mode == HRTIMER_MODE_ABS) {
1498 ret = -ERESTARTNOHAND;
1499 goto out;
1500 }
1501
1502 if (rmtp) {
1503 ret = update_rmtp(&t.timer, rmtp);
1504 if (ret <= 0)
1505 goto out;
1506 }
1507
1508 restart = &current_thread_info()->restart_block;
1509 restart->fn = hrtimer_nanosleep_restart;
1510 restart->nanosleep.index = t.timer.base->index;
1511 restart->nanosleep.rmtp = rmtp;
1512 restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
1513
1514 ret = -ERESTART_RESTARTBLOCK;
1515out:
1516 destroy_hrtimer_on_stack(&t.timer);
1517 return ret;
1518}
1519
1520SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp,
1521 struct timespec __user *, rmtp)
1522{
1523 struct timespec tu;
1524
1525 if (copy_from_user(&tu, rqtp, sizeof(tu)))
1526 return -EFAULT;
1527
1528 if (!timespec_valid(&tu))
1529 return -EINVAL;
1530
1531 return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1532}
1533
1534/*
1535 * Functions related to boot-time initialization:
1536 */
1537static void __cpuinit init_hrtimers_cpu(int cpu)
1538{
1539 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1540 int i;
1541
1542 spin_lock_init(&cpu_base->lock);
1543
1544 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
1545 cpu_base->clock_base[i].cpu_base = cpu_base;
1546
1547 hrtimer_init_hres(cpu_base);
1548}
1549
1550#ifdef CONFIG_HOTPLUG_CPU
1551
1552static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1553 struct hrtimer_clock_base *new_base)
1554{
1555 struct hrtimer *timer;
1556 struct rb_node *node;
1557
1558 while ((node = rb_first(&old_base->active))) {
1559 timer = rb_entry(node, struct hrtimer, node);
1560 BUG_ON(hrtimer_callback_running(timer));
1561 debug_hrtimer_deactivate(timer);
1562
1563 /*
1564 * Mark it as STATE_MIGRATE not INACTIVE otherwise the
1565 * timer could be seen as !active and just vanish away
1566 * under us on another CPU
1567 */
1568 __remove_hrtimer(timer, old_base, HRTIMER_STATE_MIGRATE, 0);
1569 timer->base = new_base;
1570 /*
1571 * Enqueue the timers on the new cpu. This does not
1572 * reprogram the event device in case the timer
1573 * expires before the earliest on this CPU, but we run
1574 * hrtimer_interrupt after we migrated everything to
1575 * sort out already expired timers and reprogram the
1576 * event device.
1577 */
1578 enqueue_hrtimer(timer, new_base);
1579
1580 /* Clear the migration state bit */
1581 timer->state &= ~HRTIMER_STATE_MIGRATE;
1582 }
1583}
1584
1585static void migrate_hrtimers(int scpu)
1586{
1587 struct hrtimer_cpu_base *old_base, *new_base;
1588 int i;
1589
1590 BUG_ON(cpu_online(scpu));
1591 tick_cancel_sched_timer(scpu);
1592
1593 local_irq_disable();
1594 old_base = &per_cpu(hrtimer_bases, scpu);
1595 new_base = &__get_cpu_var(hrtimer_bases);
1596 /*
1597 * The caller is globally serialized and nobody else
1598 * takes two locks at once, deadlock is not possible.
1599 */
1600 spin_lock(&new_base->lock);
1601 spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1602
1603 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1604 migrate_hrtimer_list(&old_base->clock_base[i],
1605 &new_base->clock_base[i]);
1606 }
1607
1608 spin_unlock(&old_base->lock);
1609 spin_unlock(&new_base->lock);
1610
1611 /* Check, if we got expired work to do */
1612 __hrtimer_peek_ahead_timers();
1613 local_irq_enable();
1614}
1615
1616#endif /* CONFIG_HOTPLUG_CPU */
1617
1618static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self,
1619 unsigned long action, void *hcpu)
1620{
1621 int scpu = (long)hcpu;
1622
1623 switch (action) {
1624
1625 case CPU_UP_PREPARE:
1626 case CPU_UP_PREPARE_FROZEN:
1627 init_hrtimers_cpu(scpu);
1628 break;
1629
1630#ifdef CONFIG_HOTPLUG_CPU
1631 case CPU_DYING:
1632 case CPU_DYING_FROZEN:
1633 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DYING, &scpu);
1634 break;
1635 case CPU_DEAD:
1636 case CPU_DEAD_FROZEN:
1637 {
1638 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &scpu);
1639 migrate_hrtimers(scpu);
1640 break;
1641 }
1642#endif
1643
1644 default:
1645 break;
1646 }
1647
1648 return NOTIFY_OK;
1649}
1650
1651static struct notifier_block __cpuinitdata hrtimers_nb = {
1652 .notifier_call = hrtimer_cpu_notify,
1653};
1654
1655void __init hrtimers_init(void)
1656{
1657 hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
1658 (void *)(long)smp_processor_id());
1659 register_cpu_notifier(&hrtimers_nb);
1660#ifdef CONFIG_HIGH_RES_TIMERS
1661 open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq);
1662#endif
1663}
1664
1665/**
1666 * schedule_hrtimeout_range - sleep until timeout
1667 * @expires: timeout value (ktime_t)
1668 * @delta: slack in expires timeout (ktime_t)
1669 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1670 *
1671 * Make the current task sleep until the given expiry time has
1672 * elapsed. The routine will return immediately unless
1673 * the current task state has been set (see set_current_state()).
1674 *
1675 * The @delta argument gives the kernel the freedom to schedule the
1676 * actual wakeup to a time that is both power and performance friendly.
1677 * The kernel give the normal best effort behavior for "@expires+@delta",
1678 * but may decide to fire the timer earlier, but no earlier than @expires.
1679 *
1680 * You can set the task state as follows -
1681 *
1682 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1683 * pass before the routine returns.
1684 *
1685 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1686 * delivered to the current task.
1687 *
1688 * The current task state is guaranteed to be TASK_RUNNING when this
1689 * routine returns.
1690 *
1691 * Returns 0 when the timer has expired otherwise -EINTR
1692 */
1693int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta,
1694 const enum hrtimer_mode mode)
1695{
1696 struct hrtimer_sleeper t;
1697
1698 /*
1699 * Optimize when a zero timeout value is given. It does not
1700 * matter whether this is an absolute or a relative time.
1701 */
1702 if (expires && !expires->tv64) {
1703 __set_current_state(TASK_RUNNING);
1704 return 0;
1705 }
1706
1707 /*
1708 * A NULL parameter means "inifinte"
1709 */
1710 if (!expires) {
1711 schedule();
1712 __set_current_state(TASK_RUNNING);
1713 return -EINTR;
1714 }
1715
1716 hrtimer_init_on_stack(&t.timer, CLOCK_MONOTONIC, mode);
1717 hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
1718
1719 hrtimer_init_sleeper(&t, current);
1720
1721 hrtimer_start_expires(&t.timer, mode);
1722 if (!hrtimer_active(&t.timer))
1723 t.task = NULL;
1724
1725 if (likely(t.task))
1726 schedule();
1727
1728 hrtimer_cancel(&t.timer);
1729 destroy_hrtimer_on_stack(&t.timer);
1730
1731 __set_current_state(TASK_RUNNING);
1732
1733 return !t.task ? 0 : -EINTR;
1734}
1735EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
1736
1737/**
1738 * schedule_hrtimeout - sleep until timeout
1739 * @expires: timeout value (ktime_t)
1740 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1741 *
1742 * Make the current task sleep until the given expiry time has
1743 * elapsed. The routine will return immediately unless
1744 * the current task state has been set (see set_current_state()).
1745 *
1746 * You can set the task state as follows -
1747 *
1748 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1749 * pass before the routine returns.
1750 *
1751 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1752 * delivered to the current task.
1753 *
1754 * The current task state is guaranteed to be TASK_RUNNING when this
1755 * routine returns.
1756 *
1757 * Returns 0 when the timer has expired otherwise -EINTR
1758 */
1759int __sched schedule_hrtimeout(ktime_t *expires,
1760 const enum hrtimer_mode mode)
1761{
1762 return schedule_hrtimeout_range(expires, 0, mode);
1763}
1764EXPORT_SYMBOL_GPL(schedule_hrtimeout);