projects/linux·blob v2.6.39 · refs

1851 lines · 46.1 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#include <linux/sched.h>
47#include <linux/timer.h>
48
49#include <asm/uaccess.h>
50
51#include <trace/events/timer.h>
52
53/*
54 * The timer bases:
55 *
56 * There are more clockids then hrtimer bases. Thus, we index
57 * into the timer bases by the hrtimer_base_type enum. When trying
58 * to reach a base using a clockid, hrtimer_clockid_to_base()
59 * is used to convert from clockid to the proper hrtimer_base_type.
60 */
61DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
62{
63
64 .clock_base =
65 {
66 {
67 .index = CLOCK_REALTIME,
68 .get_time = &ktime_get_real,
69 .resolution = KTIME_LOW_RES,
70 },
71 {
72 .index = CLOCK_MONOTONIC,
73 .get_time = &ktime_get,
74 .resolution = KTIME_LOW_RES,
75 },
76 {
77 .index = CLOCK_BOOTTIME,
78 .get_time = &ktime_get_boottime,
79 .resolution = KTIME_LOW_RES,
80 },
81 }
82};
83
84static int hrtimer_clock_to_base_table[MAX_CLOCKS] = {
85 [CLOCK_REALTIME] = HRTIMER_BASE_REALTIME,
86 [CLOCK_MONOTONIC] = HRTIMER_BASE_MONOTONIC,
87 [CLOCK_BOOTTIME] = HRTIMER_BASE_BOOTTIME,
88};
89
90static inline int hrtimer_clockid_to_base(clockid_t clock_id)
91{
92 return hrtimer_clock_to_base_table[clock_id];
93}
94
95
96/*
97 * Get the coarse grained time at the softirq based on xtime and
98 * wall_to_monotonic.
99 */
100static void hrtimer_get_softirq_time(struct hrtimer_cpu_base *base)
101{
102 ktime_t xtim, mono, boot;
103 struct timespec xts, tom, slp;
104
105 get_xtime_and_monotonic_and_sleep_offset(&xts, &tom, &slp);
106
107 xtim = timespec_to_ktime(xts);
108 mono = ktime_add(xtim, timespec_to_ktime(tom));
109 boot = ktime_add(mono, timespec_to_ktime(slp));
110 base->clock_base[HRTIMER_BASE_REALTIME].softirq_time = xtim;
111 base->clock_base[HRTIMER_BASE_MONOTONIC].softirq_time = mono;
112 base->clock_base[HRTIMER_BASE_BOOTTIME].softirq_time = boot;
113}
114
115/*
116 * Functions and macros which are different for UP/SMP systems are kept in a
117 * single place
118 */
119#ifdef CONFIG_SMP
120
121/*
122 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
123 * means that all timers which are tied to this base via timer->base are
124 * locked, and the base itself is locked too.
125 *
126 * So __run_timers/migrate_timers can safely modify all timers which could
127 * be found on the lists/queues.
128 *
129 * When the timer's base is locked, and the timer removed from list, it is
130 * possible to set timer->base = NULL and drop the lock: the timer remains
131 * locked.
132 */
133static
134struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
135 unsigned long *flags)
136{
137 struct hrtimer_clock_base *base;
138
139 for (;;) {
140 base = timer->base;
141 if (likely(base != NULL)) {
142 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
143 if (likely(base == timer->base))
144 return base;
145 /* The timer has migrated to another CPU: */
146 raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
147 }
148 cpu_relax();
149 }
150}
151
152
153/*
154 * Get the preferred target CPU for NOHZ
155 */
156static int hrtimer_get_target(int this_cpu, int pinned)
157{
158#ifdef CONFIG_NO_HZ
159 if (!pinned && get_sysctl_timer_migration() && idle_cpu(this_cpu))
160 return get_nohz_timer_target();
161#endif
162 return this_cpu;
163}
164
165/*
166 * With HIGHRES=y we do not migrate the timer when it is expiring
167 * before the next event on the target cpu because we cannot reprogram
168 * the target cpu hardware and we would cause it to fire late.
169 *
170 * Called with cpu_base->lock of target cpu held.
171 */
172static int
173hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base)
174{
175#ifdef CONFIG_HIGH_RES_TIMERS
176 ktime_t expires;
177
178 if (!new_base->cpu_base->hres_active)
179 return 0;
180
181 expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset);
182 return expires.tv64 <= new_base->cpu_base->expires_next.tv64;
183#else
184 return 0;
185#endif
186}
187
188/*
189 * Switch the timer base to the current CPU when possible.
190 */
191static inline struct hrtimer_clock_base *
192switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
193 int pinned)
194{
195 struct hrtimer_clock_base *new_base;
196 struct hrtimer_cpu_base *new_cpu_base;
197 int this_cpu = smp_processor_id();
198 int cpu = hrtimer_get_target(this_cpu, pinned);
199 int basenum = hrtimer_clockid_to_base(base->index);
200
201again:
202 new_cpu_base = &per_cpu(hrtimer_bases, cpu);
203 new_base = &new_cpu_base->clock_base[basenum];
204
205 if (base != new_base) {
206 /*
207 * We are trying to move timer to new_base.
208 * However we can't change timer's base while it is running,
209 * so we keep it on the same CPU. No hassle vs. reprogramming
210 * the event source in the high resolution case. The softirq
211 * code will take care of this when the timer function has
212 * completed. There is no conflict as we hold the lock until
213 * the timer is enqueued.
214 */
215 if (unlikely(hrtimer_callback_running(timer)))
216 return base;
217
218 /* See the comment in lock_timer_base() */
219 timer->base = NULL;
220 raw_spin_unlock(&base->cpu_base->lock);
221 raw_spin_lock(&new_base->cpu_base->lock);
222
223 if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) {
224 cpu = this_cpu;
225 raw_spin_unlock(&new_base->cpu_base->lock);
226 raw_spin_lock(&base->cpu_base->lock);
227 timer->base = base;
228 goto again;
229 }
230 timer->base = new_base;
231 }
232 return new_base;
233}
234
235#else /* CONFIG_SMP */
236
237static inline struct hrtimer_clock_base *
238lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
239{
240 struct hrtimer_clock_base *base = timer->base;
241
242 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
243
244 return base;
245}
246
247# define switch_hrtimer_base(t, b, p) (b)
248
249#endif /* !CONFIG_SMP */
250
251/*
252 * Functions for the union type storage format of ktime_t which are
253 * too large for inlining:
254 */
255#if BITS_PER_LONG < 64
256# ifndef CONFIG_KTIME_SCALAR
257/**
258 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
259 * @kt: addend
260 * @nsec: the scalar nsec value to add
261 *
262 * Returns the sum of kt and nsec in ktime_t format
263 */
264ktime_t ktime_add_ns(const ktime_t kt, u64 nsec)
265{
266 ktime_t tmp;
267
268 if (likely(nsec < NSEC_PER_SEC)) {
269 tmp.tv64 = nsec;
270 } else {
271 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
272
273 tmp = ktime_set((long)nsec, rem);
274 }
275
276 return ktime_add(kt, tmp);
277}
278
279EXPORT_SYMBOL_GPL(ktime_add_ns);
280
281/**
282 * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable
283 * @kt: minuend
284 * @nsec: the scalar nsec value to subtract
285 *
286 * Returns the subtraction of @nsec from @kt in ktime_t format
287 */
288ktime_t ktime_sub_ns(const ktime_t kt, u64 nsec)
289{
290 ktime_t tmp;
291
292 if (likely(nsec < NSEC_PER_SEC)) {
293 tmp.tv64 = nsec;
294 } else {
295 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
296
297 tmp = ktime_set((long)nsec, rem);
298 }
299
300 return ktime_sub(kt, tmp);
301}
302
303EXPORT_SYMBOL_GPL(ktime_sub_ns);
304# endif /* !CONFIG_KTIME_SCALAR */
305
306/*
307 * Divide a ktime value by a nanosecond value
308 */
309u64 ktime_divns(const ktime_t kt, s64 div)
310{
311 u64 dclc;
312 int sft = 0;
313
314 dclc = ktime_to_ns(kt);
315 /* Make sure the divisor is less than 2^32: */
316 while (div >> 32) {
317 sft++;
318 div >>= 1;
319 }
320 dclc >>= sft;
321 do_div(dclc, (unsigned long) div);
322
323 return dclc;
324}
325#endif /* BITS_PER_LONG >= 64 */
326
327/*
328 * Add two ktime values and do a safety check for overflow:
329 */
330ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
331{
332 ktime_t res = ktime_add(lhs, rhs);
333
334 /*
335 * We use KTIME_SEC_MAX here, the maximum timeout which we can
336 * return to user space in a timespec:
337 */
338 if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64)
339 res = ktime_set(KTIME_SEC_MAX, 0);
340
341 return res;
342}
343
344EXPORT_SYMBOL_GPL(ktime_add_safe);
345
346#ifdef CONFIG_DEBUG_OBJECTS_TIMERS
347
348static struct debug_obj_descr hrtimer_debug_descr;
349
350static void *hrtimer_debug_hint(void *addr)
351{
352 return ((struct hrtimer *) addr)->function;
353}
354
355/*
356 * fixup_init is called when:
357 * - an active object is initialized
358 */
359static int hrtimer_fixup_init(void *addr, enum debug_obj_state state)
360{
361 struct hrtimer *timer = addr;
362
363 switch (state) {
364 case ODEBUG_STATE_ACTIVE:
365 hrtimer_cancel(timer);
366 debug_object_init(timer, &hrtimer_debug_descr);
367 return 1;
368 default:
369 return 0;
370 }
371}
372
373/*
374 * fixup_activate is called when:
375 * - an active object is activated
376 * - an unknown object is activated (might be a statically initialized object)
377 */
378static int hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
379{
380 switch (state) {
381
382 case ODEBUG_STATE_NOTAVAILABLE:
383 WARN_ON_ONCE(1);
384 return 0;
385
386 case ODEBUG_STATE_ACTIVE:
387 WARN_ON(1);
388
389 default:
390 return 0;
391 }
392}
393
394/*
395 * fixup_free is called when:
396 * - an active object is freed
397 */
398static int hrtimer_fixup_free(void *addr, enum debug_obj_state state)
399{
400 struct hrtimer *timer = addr;
401
402 switch (state) {
403 case ODEBUG_STATE_ACTIVE:
404 hrtimer_cancel(timer);
405 debug_object_free(timer, &hrtimer_debug_descr);
406 return 1;
407 default:
408 return 0;
409 }
410}
411
412static struct debug_obj_descr hrtimer_debug_descr = {
413 .name = "hrtimer",
414 .debug_hint = hrtimer_debug_hint,
415 .fixup_init = hrtimer_fixup_init,
416 .fixup_activate = hrtimer_fixup_activate,
417 .fixup_free = hrtimer_fixup_free,
418};
419
420static inline void debug_hrtimer_init(struct hrtimer *timer)
421{
422 debug_object_init(timer, &hrtimer_debug_descr);
423}
424
425static inline void debug_hrtimer_activate(struct hrtimer *timer)
426{
427 debug_object_activate(timer, &hrtimer_debug_descr);
428}
429
430static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
431{
432 debug_object_deactivate(timer, &hrtimer_debug_descr);
433}
434
435static inline void debug_hrtimer_free(struct hrtimer *timer)
436{
437 debug_object_free(timer, &hrtimer_debug_descr);
438}
439
440static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
441 enum hrtimer_mode mode);
442
443void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
444 enum hrtimer_mode mode)
445{
446 debug_object_init_on_stack(timer, &hrtimer_debug_descr);
447 __hrtimer_init(timer, clock_id, mode);
448}
449EXPORT_SYMBOL_GPL(hrtimer_init_on_stack);
450
451void destroy_hrtimer_on_stack(struct hrtimer *timer)
452{
453 debug_object_free(timer, &hrtimer_debug_descr);
454}
455
456#else
457static inline void debug_hrtimer_init(struct hrtimer *timer) { }
458static inline void debug_hrtimer_activate(struct hrtimer *timer) { }
459static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
460#endif
461
462static inline void
463debug_init(struct hrtimer *timer, clockid_t clockid,
464 enum hrtimer_mode mode)
465{
466 debug_hrtimer_init(timer);
467 trace_hrtimer_init(timer, clockid, mode);
468}
469
470static inline void debug_activate(struct hrtimer *timer)
471{
472 debug_hrtimer_activate(timer);
473 trace_hrtimer_start(timer);
474}
475
476static inline void debug_deactivate(struct hrtimer *timer)
477{
478 debug_hrtimer_deactivate(timer);
479 trace_hrtimer_cancel(timer);
480}
481
482/* High resolution timer related functions */
483#ifdef CONFIG_HIGH_RES_TIMERS
484
485/*
486 * High resolution timer enabled ?
487 */
488static int hrtimer_hres_enabled __read_mostly = 1;
489
490/*
491 * Enable / Disable high resolution mode
492 */
493static int __init setup_hrtimer_hres(char *str)
494{
495 if (!strcmp(str, "off"))
496 hrtimer_hres_enabled = 0;
497 else if (!strcmp(str, "on"))
498 hrtimer_hres_enabled = 1;
499 else
500 return 0;
501 return 1;
502}
503
504__setup("highres=", setup_hrtimer_hres);
505
506/*
507 * hrtimer_high_res_enabled - query, if the highres mode is enabled
508 */
509static inline int hrtimer_is_hres_enabled(void)
510{
511 return hrtimer_hres_enabled;
512}
513
514/*
515 * Is the high resolution mode active ?
516 */
517static inline int hrtimer_hres_active(void)
518{
519 return __this_cpu_read(hrtimer_bases.hres_active);
520}
521
522/*
523 * Reprogram the event source with checking both queues for the
524 * next event
525 * Called with interrupts disabled and base->lock held
526 */
527static void
528hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
529{
530 int i;
531 struct hrtimer_clock_base *base = cpu_base->clock_base;
532 ktime_t expires, expires_next;
533
534 expires_next.tv64 = KTIME_MAX;
535
536 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
537 struct hrtimer *timer;
538 struct timerqueue_node *next;
539
540 next = timerqueue_getnext(&base->active);
541 if (!next)
542 continue;
543 timer = container_of(next, struct hrtimer, node);
544
545 expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
546 /*
547 * clock_was_set() has changed base->offset so the
548 * result might be negative. Fix it up to prevent a
549 * false positive in clockevents_program_event()
550 */
551 if (expires.tv64 < 0)
552 expires.tv64 = 0;
553 if (expires.tv64 < expires_next.tv64)
554 expires_next = expires;
555 }
556
557 if (skip_equal && expires_next.tv64 == cpu_base->expires_next.tv64)
558 return;
559
560 cpu_base->expires_next.tv64 = expires_next.tv64;
561
562 if (cpu_base->expires_next.tv64 != KTIME_MAX)
563 tick_program_event(cpu_base->expires_next, 1);
564}
565
566/*
567 * Shared reprogramming for clock_realtime and clock_monotonic
568 *
569 * When a timer is enqueued and expires earlier than the already enqueued
570 * timers, we have to check, whether it expires earlier than the timer for
571 * which the clock event device was armed.
572 *
573 * Called with interrupts disabled and base->cpu_base.lock held
574 */
575static int hrtimer_reprogram(struct hrtimer *timer,
576 struct hrtimer_clock_base *base)
577{
578 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
579 ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
580 int res;
581
582 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
583
584 /*
585 * When the callback is running, we do not reprogram the clock event
586 * device. The timer callback is either running on a different CPU or
587 * the callback is executed in the hrtimer_interrupt context. The
588 * reprogramming is handled either by the softirq, which called the
589 * callback or at the end of the hrtimer_interrupt.
590 */
591 if (hrtimer_callback_running(timer))
592 return 0;
593
594 /*
595 * CLOCK_REALTIME timer might be requested with an absolute
596 * expiry time which is less than base->offset. Nothing wrong
597 * about that, just avoid to call into the tick code, which
598 * has now objections against negative expiry values.
599 */
600 if (expires.tv64 < 0)
601 return -ETIME;
602
603 if (expires.tv64 >= cpu_base->expires_next.tv64)
604 return 0;
605
606 /*
607 * If a hang was detected in the last timer interrupt then we
608 * do not schedule a timer which is earlier than the expiry
609 * which we enforced in the hang detection. We want the system
610 * to make progress.
611 */
612 if (cpu_base->hang_detected)
613 return 0;
614
615 /*
616 * Clockevents returns -ETIME, when the event was in the past.
617 */
618 res = tick_program_event(expires, 0);
619 if (!IS_ERR_VALUE(res))
620 cpu_base->expires_next = expires;
621 return res;
622}
623
624
625/*
626 * Retrigger next event is called after clock was set
627 *
628 * Called with interrupts disabled via on_each_cpu()
629 */
630static void retrigger_next_event(void *arg)
631{
632 struct hrtimer_cpu_base *base;
633 struct timespec realtime_offset, wtm, sleep;
634
635 if (!hrtimer_hres_active())
636 return;
637
638 get_xtime_and_monotonic_and_sleep_offset(&realtime_offset, &wtm,
639 &sleep);
640 set_normalized_timespec(&realtime_offset, -wtm.tv_sec, -wtm.tv_nsec);
641
642 base = &__get_cpu_var(hrtimer_bases);
643
644 /* Adjust CLOCK_REALTIME offset */
645 raw_spin_lock(&base->lock);
646 base->clock_base[HRTIMER_BASE_REALTIME].offset =
647 timespec_to_ktime(realtime_offset);
648 base->clock_base[HRTIMER_BASE_BOOTTIME].offset =
649 timespec_to_ktime(sleep);
650
651 hrtimer_force_reprogram(base, 0);
652 raw_spin_unlock(&base->lock);
653}
654
655/*
656 * Clock realtime was set
657 *
658 * Change the offset of the realtime clock vs. the monotonic
659 * clock.
660 *
661 * We might have to reprogram the high resolution timer interrupt. On
662 * SMP we call the architecture specific code to retrigger _all_ high
663 * resolution timer interrupts. On UP we just disable interrupts and
664 * call the high resolution interrupt code.
665 */
666void clock_was_set(void)
667{
668 /* Retrigger the CPU local events everywhere */
669 on_each_cpu(retrigger_next_event, NULL, 1);
670}
671
672/*
673 * During resume we might have to reprogram the high resolution timer
674 * interrupt (on the local CPU):
675 */
676void hres_timers_resume(void)
677{
678 WARN_ONCE(!irqs_disabled(),
679 KERN_INFO "hres_timers_resume() called with IRQs enabled!");
680
681 retrigger_next_event(NULL);
682}
683
684/*
685 * Initialize the high resolution related parts of cpu_base
686 */
687static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
688{
689 base->expires_next.tv64 = KTIME_MAX;
690 base->hres_active = 0;
691}
692
693/*
694 * When High resolution timers are active, try to reprogram. Note, that in case
695 * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry
696 * check happens. The timer gets enqueued into the rbtree. The reprogramming
697 * and expiry check is done in the hrtimer_interrupt or in the softirq.
698 */
699static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
700 struct hrtimer_clock_base *base,
701 int wakeup)
702{
703 if (base->cpu_base->hres_active && hrtimer_reprogram(timer, base)) {
704 if (wakeup) {
705 raw_spin_unlock(&base->cpu_base->lock);
706 raise_softirq_irqoff(HRTIMER_SOFTIRQ);
707 raw_spin_lock(&base->cpu_base->lock);
708 } else
709 __raise_softirq_irqoff(HRTIMER_SOFTIRQ);
710
711 return 1;
712 }
713
714 return 0;
715}
716
717/*
718 * Switch to high resolution mode
719 */
720static int hrtimer_switch_to_hres(void)
721{
722 int cpu = smp_processor_id();
723 struct hrtimer_cpu_base *base = &per_cpu(hrtimer_bases, cpu);
724 unsigned long flags;
725
726 if (base->hres_active)
727 return 1;
728
729 local_irq_save(flags);
730
731 if (tick_init_highres()) {
732 local_irq_restore(flags);
733 printk(KERN_WARNING "Could not switch to high resolution "
734 "mode on CPU %d\n", cpu);
735 return 0;
736 }
737 base->hres_active = 1;
738 base->clock_base[HRTIMER_BASE_REALTIME].resolution = KTIME_HIGH_RES;
739 base->clock_base[HRTIMER_BASE_MONOTONIC].resolution = KTIME_HIGH_RES;
740 base->clock_base[HRTIMER_BASE_BOOTTIME].resolution = KTIME_HIGH_RES;
741
742 tick_setup_sched_timer();
743
744 /* "Retrigger" the interrupt to get things going */
745 retrigger_next_event(NULL);
746 local_irq_restore(flags);
747 return 1;
748}
749
750#else
751
752static inline int hrtimer_hres_active(void) { return 0; }
753static inline int hrtimer_is_hres_enabled(void) { return 0; }
754static inline int hrtimer_switch_to_hres(void) { return 0; }
755static inline void
756hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { }
757static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
758 struct hrtimer_clock_base *base,
759 int wakeup)
760{
761 return 0;
762}
763static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { }
764
765#endif /* CONFIG_HIGH_RES_TIMERS */
766
767static inline void timer_stats_hrtimer_set_start_info(struct hrtimer *timer)
768{
769#ifdef CONFIG_TIMER_STATS
770 if (timer->start_site)
771 return;
772 timer->start_site = __builtin_return_address(0);
773 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
774 timer->start_pid = current->pid;
775#endif
776}
777
778static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer *timer)
779{
780#ifdef CONFIG_TIMER_STATS
781 timer->start_site = NULL;
782#endif
783}
784
785static inline void timer_stats_account_hrtimer(struct hrtimer *timer)
786{
787#ifdef CONFIG_TIMER_STATS
788 if (likely(!timer_stats_active))
789 return;
790 timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
791 timer->function, timer->start_comm, 0);
792#endif
793}
794
795/*
796 * Counterpart to lock_hrtimer_base above:
797 */
798static inline
799void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
800{
801 raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
802}
803
804/**
805 * hrtimer_forward - forward the timer expiry
806 * @timer: hrtimer to forward
807 * @now: forward past this time
808 * @interval: the interval to forward
809 *
810 * Forward the timer expiry so it will expire in the future.
811 * Returns the number of overruns.
812 */
813u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
814{
815 u64 orun = 1;
816 ktime_t delta;
817
818 delta = ktime_sub(now, hrtimer_get_expires(timer));
819
820 if (delta.tv64 < 0)
821 return 0;
822
823 if (interval.tv64 < timer->base->resolution.tv64)
824 interval.tv64 = timer->base->resolution.tv64;
825
826 if (unlikely(delta.tv64 >= interval.tv64)) {
827 s64 incr = ktime_to_ns(interval);
828
829 orun = ktime_divns(delta, incr);
830 hrtimer_add_expires_ns(timer, incr * orun);
831 if (hrtimer_get_expires_tv64(timer) > now.tv64)
832 return orun;
833 /*
834 * This (and the ktime_add() below) is the
835 * correction for exact:
836 */
837 orun++;
838 }
839 hrtimer_add_expires(timer, interval);
840
841 return orun;
842}
843EXPORT_SYMBOL_GPL(hrtimer_forward);
844
845/*
846 * enqueue_hrtimer - internal function to (re)start a timer
847 *
848 * The timer is inserted in expiry order. Insertion into the
849 * red black tree is O(log(n)). Must hold the base lock.
850 *
851 * Returns 1 when the new timer is the leftmost timer in the tree.
852 */
853static int enqueue_hrtimer(struct hrtimer *timer,
854 struct hrtimer_clock_base *base)
855{
856 debug_activate(timer);
857
858 timerqueue_add(&base->active, &timer->node);
859
860 /*
861 * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
862 * state of a possibly running callback.
863 */
864 timer->state |= HRTIMER_STATE_ENQUEUED;
865
866 return (&timer->node == base->active.next);
867}
868
869/*
870 * __remove_hrtimer - internal function to remove a timer
871 *
872 * Caller must hold the base lock.
873 *
874 * High resolution timer mode reprograms the clock event device when the
875 * timer is the one which expires next. The caller can disable this by setting
876 * reprogram to zero. This is useful, when the context does a reprogramming
877 * anyway (e.g. timer interrupt)
878 */
879static void __remove_hrtimer(struct hrtimer *timer,
880 struct hrtimer_clock_base *base,
881 unsigned long newstate, int reprogram)
882{
883 if (!(timer->state & HRTIMER_STATE_ENQUEUED))
884 goto out;
885
886 if (&timer->node == timerqueue_getnext(&base->active)) {
887#ifdef CONFIG_HIGH_RES_TIMERS
888 /* Reprogram the clock event device. if enabled */
889 if (reprogram && hrtimer_hres_active()) {
890 ktime_t expires;
891
892 expires = ktime_sub(hrtimer_get_expires(timer),
893 base->offset);
894 if (base->cpu_base->expires_next.tv64 == expires.tv64)
895 hrtimer_force_reprogram(base->cpu_base, 1);
896 }
897#endif
898 }
899 timerqueue_del(&base->active, &timer->node);
900out:
901 timer->state = newstate;
902}
903
904/*
905 * remove hrtimer, called with base lock held
906 */
907static inline int
908remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base)
909{
910 if (hrtimer_is_queued(timer)) {
911 unsigned long state;
912 int reprogram;
913
914 /*
915 * Remove the timer and force reprogramming when high
916 * resolution mode is active and the timer is on the current
917 * CPU. If we remove a timer on another CPU, reprogramming is
918 * skipped. The interrupt event on this CPU is fired and
919 * reprogramming happens in the interrupt handler. This is a
920 * rare case and less expensive than a smp call.
921 */
922 debug_deactivate(timer);
923 timer_stats_hrtimer_clear_start_info(timer);
924 reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases);
925 /*
926 * We must preserve the CALLBACK state flag here,
927 * otherwise we could move the timer base in
928 * switch_hrtimer_base.
929 */
930 state = timer->state & HRTIMER_STATE_CALLBACK;
931 __remove_hrtimer(timer, base, state, reprogram);
932 return 1;
933 }
934 return 0;
935}
936
937int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
938 unsigned long delta_ns, const enum hrtimer_mode mode,
939 int wakeup)
940{
941 struct hrtimer_clock_base *base, *new_base;
942 unsigned long flags;
943 int ret, leftmost;
944
945 base = lock_hrtimer_base(timer, &flags);
946
947 /* Remove an active timer from the queue: */
948 ret = remove_hrtimer(timer, base);
949
950 /* Switch the timer base, if necessary: */
951 new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED);
952
953 if (mode & HRTIMER_MODE_REL) {
954 tim = ktime_add_safe(tim, new_base->get_time());
955 /*
956 * CONFIG_TIME_LOW_RES is a temporary way for architectures
957 * to signal that they simply return xtime in
958 * do_gettimeoffset(). In this case we want to round up by
959 * resolution when starting a relative timer, to avoid short
960 * timeouts. This will go away with the GTOD framework.
961 */
962#ifdef CONFIG_TIME_LOW_RES
963 tim = ktime_add_safe(tim, base->resolution);
964#endif
965 }
966
967 hrtimer_set_expires_range_ns(timer, tim, delta_ns);
968
969 timer_stats_hrtimer_set_start_info(timer);
970
971 leftmost = enqueue_hrtimer(timer, new_base);
972
973 /*
974 * Only allow reprogramming if the new base is on this CPU.
975 * (it might still be on another CPU if the timer was pending)
976 *
977 * XXX send_remote_softirq() ?
978 */
979 if (leftmost && new_base->cpu_base == &__get_cpu_var(hrtimer_bases))
980 hrtimer_enqueue_reprogram(timer, new_base, wakeup);
981
982 unlock_hrtimer_base(timer, &flags);
983
984 return ret;
985}
986
987/**
988 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
989 * @timer: the timer to be added
990 * @tim: expiry time
991 * @delta_ns: "slack" range for the timer
992 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
993 *
994 * Returns:
995 * 0 on success
996 * 1 when the timer was active
997 */
998int hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
999 unsigned long delta_ns, const enum hrtimer_mode mode)
1000{
1001 return __hrtimer_start_range_ns(timer, tim, delta_ns, mode, 1);
1002}
1003EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
1004
1005/**
1006 * hrtimer_start - (re)start an hrtimer on the current CPU
1007 * @timer: the timer to be added
1008 * @tim: expiry time
1009 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
1010 *
1011 * Returns:
1012 * 0 on success
1013 * 1 when the timer was active
1014 */
1015int
1016hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode)
1017{
1018 return __hrtimer_start_range_ns(timer, tim, 0, mode, 1);
1019}
1020EXPORT_SYMBOL_GPL(hrtimer_start);
1021
1022
1023/**
1024 * hrtimer_try_to_cancel - try to deactivate a timer
1025 * @timer: hrtimer to stop
1026 *
1027 * Returns:
1028 * 0 when the timer was not active
1029 * 1 when the timer was active
1030 * -1 when the timer is currently excuting the callback function and
1031 * cannot be stopped
1032 */
1033int hrtimer_try_to_cancel(struct hrtimer *timer)
1034{
1035 struct hrtimer_clock_base *base;
1036 unsigned long flags;
1037 int ret = -1;
1038
1039 base = lock_hrtimer_base(timer, &flags);
1040
1041 if (!hrtimer_callback_running(timer))
1042 ret = remove_hrtimer(timer, base);
1043
1044 unlock_hrtimer_base(timer, &flags);
1045
1046 return ret;
1047
1048}
1049EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1050
1051/**
1052 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1053 * @timer: the timer to be cancelled
1054 *
1055 * Returns:
1056 * 0 when the timer was not active
1057 * 1 when the timer was active
1058 */
1059int hrtimer_cancel(struct hrtimer *timer)
1060{
1061 for (;;) {
1062 int ret = hrtimer_try_to_cancel(timer);
1063
1064 if (ret >= 0)
1065 return ret;
1066 cpu_relax();
1067 }
1068}
1069EXPORT_SYMBOL_GPL(hrtimer_cancel);
1070
1071/**
1072 * hrtimer_get_remaining - get remaining time for the timer
1073 * @timer: the timer to read
1074 */
1075ktime_t hrtimer_get_remaining(const struct hrtimer *timer)
1076{
1077 unsigned long flags;
1078 ktime_t rem;
1079
1080 lock_hrtimer_base(timer, &flags);
1081 rem = hrtimer_expires_remaining(timer);
1082 unlock_hrtimer_base(timer, &flags);
1083
1084 return rem;
1085}
1086EXPORT_SYMBOL_GPL(hrtimer_get_remaining);
1087
1088#ifdef CONFIG_NO_HZ
1089/**
1090 * hrtimer_get_next_event - get the time until next expiry event
1091 *
1092 * Returns the delta to the next expiry event or KTIME_MAX if no timer
1093 * is pending.
1094 */
1095ktime_t hrtimer_get_next_event(void)
1096{
1097 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1098 struct hrtimer_clock_base *base = cpu_base->clock_base;
1099 ktime_t delta, mindelta = { .tv64 = KTIME_MAX };
1100 unsigned long flags;
1101 int i;
1102
1103 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1104
1105 if (!hrtimer_hres_active()) {
1106 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
1107 struct hrtimer *timer;
1108 struct timerqueue_node *next;
1109
1110 next = timerqueue_getnext(&base->active);
1111 if (!next)
1112 continue;
1113
1114 timer = container_of(next, struct hrtimer, node);
1115 delta.tv64 = hrtimer_get_expires_tv64(timer);
1116 delta = ktime_sub(delta, base->get_time());
1117 if (delta.tv64 < mindelta.tv64)
1118 mindelta.tv64 = delta.tv64;
1119 }
1120 }
1121
1122 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1123
1124 if (mindelta.tv64 < 0)
1125 mindelta.tv64 = 0;
1126 return mindelta;
1127}
1128#endif
1129
1130static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1131 enum hrtimer_mode mode)
1132{
1133 struct hrtimer_cpu_base *cpu_base;
1134 int base;
1135
1136 memset(timer, 0, sizeof(struct hrtimer));
1137
1138 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1139
1140 if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
1141 clock_id = CLOCK_MONOTONIC;
1142
1143 base = hrtimer_clockid_to_base(clock_id);
1144 timer->base = &cpu_base->clock_base[base];
1145 timerqueue_init(&timer->node);
1146
1147#ifdef CONFIG_TIMER_STATS
1148 timer->start_site = NULL;
1149 timer->start_pid = -1;
1150 memset(timer->start_comm, 0, TASK_COMM_LEN);
1151#endif
1152}
1153
1154/**
1155 * hrtimer_init - initialize a timer to the given clock
1156 * @timer: the timer to be initialized
1157 * @clock_id: the clock to be used
1158 * @mode: timer mode abs/rel
1159 */
1160void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1161 enum hrtimer_mode mode)
1162{
1163 debug_init(timer, clock_id, mode);
1164 __hrtimer_init(timer, clock_id, mode);
1165}
1166EXPORT_SYMBOL_GPL(hrtimer_init);
1167
1168/**
1169 * hrtimer_get_res - get the timer resolution for a clock
1170 * @which_clock: which clock to query
1171 * @tp: pointer to timespec variable to store the resolution
1172 *
1173 * Store the resolution of the clock selected by @which_clock in the
1174 * variable pointed to by @tp.
1175 */
1176int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp)
1177{
1178 struct hrtimer_cpu_base *cpu_base;
1179 int base = hrtimer_clockid_to_base(which_clock);
1180
1181 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1182 *tp = ktime_to_timespec(cpu_base->clock_base[base].resolution);
1183
1184 return 0;
1185}
1186EXPORT_SYMBOL_GPL(hrtimer_get_res);
1187
1188static void __run_hrtimer(struct hrtimer *timer, ktime_t *now)
1189{
1190 struct hrtimer_clock_base *base = timer->base;
1191 struct hrtimer_cpu_base *cpu_base = base->cpu_base;
1192 enum hrtimer_restart (*fn)(struct hrtimer *);
1193 int restart;
1194
1195 WARN_ON(!irqs_disabled());
1196
1197 debug_deactivate(timer);
1198 __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0);
1199 timer_stats_account_hrtimer(timer);
1200 fn = timer->function;
1201
1202 /*
1203 * Because we run timers from hardirq context, there is no chance
1204 * they get migrated to another cpu, therefore its safe to unlock
1205 * the timer base.
1206 */
1207 raw_spin_unlock(&cpu_base->lock);
1208 trace_hrtimer_expire_entry(timer, now);
1209 restart = fn(timer);
1210 trace_hrtimer_expire_exit(timer);
1211 raw_spin_lock(&cpu_base->lock);
1212
1213 /*
1214 * Note: We clear the CALLBACK bit after enqueue_hrtimer and
1215 * we do not reprogramm the event hardware. Happens either in
1216 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1217 */
1218 if (restart != HRTIMER_NORESTART) {
1219 BUG_ON(timer->state != HRTIMER_STATE_CALLBACK);
1220 enqueue_hrtimer(timer, base);
1221 }
1222
1223 WARN_ON_ONCE(!(timer->state & HRTIMER_STATE_CALLBACK));
1224
1225 timer->state &= ~HRTIMER_STATE_CALLBACK;
1226}
1227
1228#ifdef CONFIG_HIGH_RES_TIMERS
1229
1230/*
1231 * High resolution timer interrupt
1232 * Called with interrupts disabled
1233 */
1234void hrtimer_interrupt(struct clock_event_device *dev)
1235{
1236 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1237 struct hrtimer_clock_base *base;
1238 ktime_t expires_next, now, entry_time, delta;
1239 int i, retries = 0;
1240
1241 BUG_ON(!cpu_base->hres_active);
1242 cpu_base->nr_events++;
1243 dev->next_event.tv64 = KTIME_MAX;
1244
1245 entry_time = now = ktime_get();
1246retry:
1247 expires_next.tv64 = KTIME_MAX;
1248
1249 raw_spin_lock(&cpu_base->lock);
1250 /*
1251 * We set expires_next to KTIME_MAX here with cpu_base->lock
1252 * held to prevent that a timer is enqueued in our queue via
1253 * the migration code. This does not affect enqueueing of
1254 * timers which run their callback and need to be requeued on
1255 * this CPU.
1256 */
1257 cpu_base->expires_next.tv64 = KTIME_MAX;
1258
1259 base = cpu_base->clock_base;
1260
1261 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1262 ktime_t basenow;
1263 struct timerqueue_node *node;
1264
1265 basenow = ktime_add(now, base->offset);
1266
1267 while ((node = timerqueue_getnext(&base->active))) {
1268 struct hrtimer *timer;
1269
1270 timer = container_of(node, struct hrtimer, node);
1271
1272 /*
1273 * The immediate goal for using the softexpires is
1274 * minimizing wakeups, not running timers at the
1275 * earliest interrupt after their soft expiration.
1276 * This allows us to avoid using a Priority Search
1277 * Tree, which can answer a stabbing querry for
1278 * overlapping intervals and instead use the simple
1279 * BST we already have.
1280 * We don't add extra wakeups by delaying timers that
1281 * are right-of a not yet expired timer, because that
1282 * timer will have to trigger a wakeup anyway.
1283 */
1284
1285 if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer)) {
1286 ktime_t expires;
1287
1288 expires = ktime_sub(hrtimer_get_expires(timer),
1289 base->offset);
1290 if (expires.tv64 < expires_next.tv64)
1291 expires_next = expires;
1292 break;
1293 }
1294
1295 __run_hrtimer(timer, &basenow);
1296 }
1297 base++;
1298 }
1299
1300 /*
1301 * Store the new expiry value so the migration code can verify
1302 * against it.
1303 */
1304 cpu_base->expires_next = expires_next;
1305 raw_spin_unlock(&cpu_base->lock);
1306
1307 /* Reprogramming necessary ? */
1308 if (expires_next.tv64 == KTIME_MAX ||
1309 !tick_program_event(expires_next, 0)) {
1310 cpu_base->hang_detected = 0;
1311 return;
1312 }
1313
1314 /*
1315 * The next timer was already expired due to:
1316 * - tracing
1317 * - long lasting callbacks
1318 * - being scheduled away when running in a VM
1319 *
1320 * We need to prevent that we loop forever in the hrtimer
1321 * interrupt routine. We give it 3 attempts to avoid
1322 * overreacting on some spurious event.
1323 */
1324 now = ktime_get();
1325 cpu_base->nr_retries++;
1326 if (++retries < 3)
1327 goto retry;
1328 /*
1329 * Give the system a chance to do something else than looping
1330 * here. We stored the entry time, so we know exactly how long
1331 * we spent here. We schedule the next event this amount of
1332 * time away.
1333 */
1334 cpu_base->nr_hangs++;
1335 cpu_base->hang_detected = 1;
1336 delta = ktime_sub(now, entry_time);
1337 if (delta.tv64 > cpu_base->max_hang_time.tv64)
1338 cpu_base->max_hang_time = delta;
1339 /*
1340 * Limit it to a sensible value as we enforce a longer
1341 * delay. Give the CPU at least 100ms to catch up.
1342 */
1343 if (delta.tv64 > 100 * NSEC_PER_MSEC)
1344 expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1345 else
1346 expires_next = ktime_add(now, delta);
1347 tick_program_event(expires_next, 1);
1348 printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n",
1349 ktime_to_ns(delta));
1350}
1351
1352/*
1353 * local version of hrtimer_peek_ahead_timers() called with interrupts
1354 * disabled.
1355 */
1356static void __hrtimer_peek_ahead_timers(void)
1357{
1358 struct tick_device *td;
1359
1360 if (!hrtimer_hres_active())
1361 return;
1362
1363 td = &__get_cpu_var(tick_cpu_device);
1364 if (td && td->evtdev)
1365 hrtimer_interrupt(td->evtdev);
1366}
1367
1368/**
1369 * hrtimer_peek_ahead_timers -- run soft-expired timers now
1370 *
1371 * hrtimer_peek_ahead_timers will peek at the timer queue of
1372 * the current cpu and check if there are any timers for which
1373 * the soft expires time has passed. If any such timers exist,
1374 * they are run immediately and then removed from the timer queue.
1375 *
1376 */
1377void hrtimer_peek_ahead_timers(void)
1378{
1379 unsigned long flags;
1380
1381 local_irq_save(flags);
1382 __hrtimer_peek_ahead_timers();
1383 local_irq_restore(flags);
1384}
1385
1386static void run_hrtimer_softirq(struct softirq_action *h)
1387{
1388 hrtimer_peek_ahead_timers();
1389}
1390
1391#else /* CONFIG_HIGH_RES_TIMERS */
1392
1393static inline void __hrtimer_peek_ahead_timers(void) { }
1394
1395#endif /* !CONFIG_HIGH_RES_TIMERS */
1396
1397/*
1398 * Called from timer softirq every jiffy, expire hrtimers:
1399 *
1400 * For HRT its the fall back code to run the softirq in the timer
1401 * softirq context in case the hrtimer initialization failed or has
1402 * not been done yet.
1403 */
1404void hrtimer_run_pending(void)
1405{
1406 if (hrtimer_hres_active())
1407 return;
1408
1409 /*
1410 * This _is_ ugly: We have to check in the softirq context,
1411 * whether we can switch to highres and / or nohz mode. The
1412 * clocksource switch happens in the timer interrupt with
1413 * xtime_lock held. Notification from there only sets the
1414 * check bit in the tick_oneshot code, otherwise we might
1415 * deadlock vs. xtime_lock.
1416 */
1417 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1418 hrtimer_switch_to_hres();
1419}
1420
1421/*
1422 * Called from hardirq context every jiffy
1423 */
1424void hrtimer_run_queues(void)
1425{
1426 struct timerqueue_node *node;
1427 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1428 struct hrtimer_clock_base *base;
1429 int index, gettime = 1;
1430
1431 if (hrtimer_hres_active())
1432 return;
1433
1434 for (index = 0; index < HRTIMER_MAX_CLOCK_BASES; index++) {
1435 base = &cpu_base->clock_base[index];
1436 if (!timerqueue_getnext(&base->active))
1437 continue;
1438
1439 if (gettime) {
1440 hrtimer_get_softirq_time(cpu_base);
1441 gettime = 0;
1442 }
1443
1444 raw_spin_lock(&cpu_base->lock);
1445
1446 while ((node = timerqueue_getnext(&base->active))) {
1447 struct hrtimer *timer;
1448
1449 timer = container_of(node, struct hrtimer, node);
1450 if (base->softirq_time.tv64 <=
1451 hrtimer_get_expires_tv64(timer))
1452 break;
1453
1454 __run_hrtimer(timer, &base->softirq_time);
1455 }
1456 raw_spin_unlock(&cpu_base->lock);
1457 }
1458}
1459
1460/*
1461 * Sleep related functions:
1462 */
1463static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1464{
1465 struct hrtimer_sleeper *t =
1466 container_of(timer, struct hrtimer_sleeper, timer);
1467 struct task_struct *task = t->task;
1468
1469 t->task = NULL;
1470 if (task)
1471 wake_up_process(task);
1472
1473 return HRTIMER_NORESTART;
1474}
1475
1476void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1477{
1478 sl->timer.function = hrtimer_wakeup;
1479 sl->task = task;
1480}
1481EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
1482
1483static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1484{
1485 hrtimer_init_sleeper(t, current);
1486
1487 do {
1488 set_current_state(TASK_INTERRUPTIBLE);
1489 hrtimer_start_expires(&t->timer, mode);
1490 if (!hrtimer_active(&t->timer))
1491 t->task = NULL;
1492
1493 if (likely(t->task))
1494 schedule();
1495
1496 hrtimer_cancel(&t->timer);
1497 mode = HRTIMER_MODE_ABS;
1498
1499 } while (t->task && !signal_pending(current));
1500
1501 __set_current_state(TASK_RUNNING);
1502
1503 return t->task == NULL;
1504}
1505
1506static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp)
1507{
1508 struct timespec rmt;
1509 ktime_t rem;
1510
1511 rem = hrtimer_expires_remaining(timer);
1512 if (rem.tv64 <= 0)
1513 return 0;
1514 rmt = ktime_to_timespec(rem);
1515
1516 if (copy_to_user(rmtp, &rmt, sizeof(*rmtp)))
1517 return -EFAULT;
1518
1519 return 1;
1520}
1521
1522long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1523{
1524 struct hrtimer_sleeper t;
1525 struct timespec __user *rmtp;
1526 int ret = 0;
1527
1528 hrtimer_init_on_stack(&t.timer, restart->nanosleep.index,
1529 HRTIMER_MODE_ABS);
1530 hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
1531
1532 if (do_nanosleep(&t, HRTIMER_MODE_ABS))
1533 goto out;
1534
1535 rmtp = restart->nanosleep.rmtp;
1536 if (rmtp) {
1537 ret = update_rmtp(&t.timer, rmtp);
1538 if (ret <= 0)
1539 goto out;
1540 }
1541
1542 /* The other values in restart are already filled in */
1543 ret = -ERESTART_RESTARTBLOCK;
1544out:
1545 destroy_hrtimer_on_stack(&t.timer);
1546 return ret;
1547}
1548
1549long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
1550 const enum hrtimer_mode mode, const clockid_t clockid)
1551{
1552 struct restart_block *restart;
1553 struct hrtimer_sleeper t;
1554 int ret = 0;
1555 unsigned long slack;
1556
1557 slack = current->timer_slack_ns;
1558 if (rt_task(current))
1559 slack = 0;
1560
1561 hrtimer_init_on_stack(&t.timer, clockid, mode);
1562 hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack);
1563 if (do_nanosleep(&t, mode))
1564 goto out;
1565
1566 /* Absolute timers do not update the rmtp value and restart: */
1567 if (mode == HRTIMER_MODE_ABS) {
1568 ret = -ERESTARTNOHAND;
1569 goto out;
1570 }
1571
1572 if (rmtp) {
1573 ret = update_rmtp(&t.timer, rmtp);
1574 if (ret <= 0)
1575 goto out;
1576 }
1577
1578 restart = &current_thread_info()->restart_block;
1579 restart->fn = hrtimer_nanosleep_restart;
1580 restart->nanosleep.index = t.timer.base->index;
1581 restart->nanosleep.rmtp = rmtp;
1582 restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
1583
1584 ret = -ERESTART_RESTARTBLOCK;
1585out:
1586 destroy_hrtimer_on_stack(&t.timer);
1587 return ret;
1588}
1589
1590SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp,
1591 struct timespec __user *, rmtp)
1592{
1593 struct timespec tu;
1594
1595 if (copy_from_user(&tu, rqtp, sizeof(tu)))
1596 return -EFAULT;
1597
1598 if (!timespec_valid(&tu))
1599 return -EINVAL;
1600
1601 return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1602}
1603
1604/*
1605 * Functions related to boot-time initialization:
1606 */
1607static void __cpuinit init_hrtimers_cpu(int cpu)
1608{
1609 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1610 int i;
1611
1612 raw_spin_lock_init(&cpu_base->lock);
1613
1614 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1615 cpu_base->clock_base[i].cpu_base = cpu_base;
1616 timerqueue_init_head(&cpu_base->clock_base[i].active);
1617 }
1618
1619 hrtimer_init_hres(cpu_base);
1620}
1621
1622#ifdef CONFIG_HOTPLUG_CPU
1623
1624static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1625 struct hrtimer_clock_base *new_base)
1626{
1627 struct hrtimer *timer;
1628 struct timerqueue_node *node;
1629
1630 while ((node = timerqueue_getnext(&old_base->active))) {
1631 timer = container_of(node, struct hrtimer, node);
1632 BUG_ON(hrtimer_callback_running(timer));
1633 debug_deactivate(timer);
1634
1635 /*
1636 * Mark it as STATE_MIGRATE not INACTIVE otherwise the
1637 * timer could be seen as !active and just vanish away
1638 * under us on another CPU
1639 */
1640 __remove_hrtimer(timer, old_base, HRTIMER_STATE_MIGRATE, 0);
1641 timer->base = new_base;
1642 /*
1643 * Enqueue the timers on the new cpu. This does not
1644 * reprogram the event device in case the timer
1645 * expires before the earliest on this CPU, but we run
1646 * hrtimer_interrupt after we migrated everything to
1647 * sort out already expired timers and reprogram the
1648 * event device.
1649 */
1650 enqueue_hrtimer(timer, new_base);
1651
1652 /* Clear the migration state bit */
1653 timer->state &= ~HRTIMER_STATE_MIGRATE;
1654 }
1655}
1656
1657static void migrate_hrtimers(int scpu)
1658{
1659 struct hrtimer_cpu_base *old_base, *new_base;
1660 int i;
1661
1662 BUG_ON(cpu_online(scpu));
1663 tick_cancel_sched_timer(scpu);
1664
1665 local_irq_disable();
1666 old_base = &per_cpu(hrtimer_bases, scpu);
1667 new_base = &__get_cpu_var(hrtimer_bases);
1668 /*
1669 * The caller is globally serialized and nobody else
1670 * takes two locks at once, deadlock is not possible.
1671 */
1672 raw_spin_lock(&new_base->lock);
1673 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1674
1675 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1676 migrate_hrtimer_list(&old_base->clock_base[i],
1677 &new_base->clock_base[i]);
1678 }
1679
1680 raw_spin_unlock(&old_base->lock);
1681 raw_spin_unlock(&new_base->lock);
1682
1683 /* Check, if we got expired work to do */
1684 __hrtimer_peek_ahead_timers();
1685 local_irq_enable();
1686}
1687
1688#endif /* CONFIG_HOTPLUG_CPU */
1689
1690static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self,
1691 unsigned long action, void *hcpu)
1692{
1693 int scpu = (long)hcpu;
1694
1695 switch (action) {
1696
1697 case CPU_UP_PREPARE:
1698 case CPU_UP_PREPARE_FROZEN:
1699 init_hrtimers_cpu(scpu);
1700 break;
1701
1702#ifdef CONFIG_HOTPLUG_CPU
1703 case CPU_DYING:
1704 case CPU_DYING_FROZEN:
1705 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DYING, &scpu);
1706 break;
1707 case CPU_DEAD:
1708 case CPU_DEAD_FROZEN:
1709 {
1710 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &scpu);
1711 migrate_hrtimers(scpu);
1712 break;
1713 }
1714#endif
1715
1716 default:
1717 break;
1718 }
1719
1720 return NOTIFY_OK;
1721}
1722
1723static struct notifier_block __cpuinitdata hrtimers_nb = {
1724 .notifier_call = hrtimer_cpu_notify,
1725};
1726
1727void __init hrtimers_init(void)
1728{
1729 hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
1730 (void *)(long)smp_processor_id());
1731 register_cpu_notifier(&hrtimers_nb);
1732#ifdef CONFIG_HIGH_RES_TIMERS
1733 open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq);
1734#endif
1735}
1736
1737/**
1738 * schedule_hrtimeout_range_clock - sleep until timeout
1739 * @expires: timeout value (ktime_t)
1740 * @delta: slack in expires timeout (ktime_t)
1741 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1742 * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME
1743 */
1744int __sched
1745schedule_hrtimeout_range_clock(ktime_t *expires, unsigned long delta,
1746 const enum hrtimer_mode mode, int clock)
1747{
1748 struct hrtimer_sleeper t;
1749
1750 /*
1751 * Optimize when a zero timeout value is given. It does not
1752 * matter whether this is an absolute or a relative time.
1753 */
1754 if (expires && !expires->tv64) {
1755 __set_current_state(TASK_RUNNING);
1756 return 0;
1757 }
1758
1759 /*
1760 * A NULL parameter means "infinite"
1761 */
1762 if (!expires) {
1763 schedule();
1764 __set_current_state(TASK_RUNNING);
1765 return -EINTR;
1766 }
1767
1768 hrtimer_init_on_stack(&t.timer, clock, mode);
1769 hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
1770
1771 hrtimer_init_sleeper(&t, current);
1772
1773 hrtimer_start_expires(&t.timer, mode);
1774 if (!hrtimer_active(&t.timer))
1775 t.task = NULL;
1776
1777 if (likely(t.task))
1778 schedule();
1779
1780 hrtimer_cancel(&t.timer);
1781 destroy_hrtimer_on_stack(&t.timer);
1782
1783 __set_current_state(TASK_RUNNING);
1784
1785 return !t.task ? 0 : -EINTR;
1786}
1787
1788/**
1789 * schedule_hrtimeout_range - sleep until timeout
1790 * @expires: timeout value (ktime_t)
1791 * @delta: slack in expires timeout (ktime_t)
1792 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1793 *
1794 * Make the current task sleep until the given expiry time has
1795 * elapsed. The routine will return immediately unless
1796 * the current task state has been set (see set_current_state()).
1797 *
1798 * The @delta argument gives the kernel the freedom to schedule the
1799 * actual wakeup to a time that is both power and performance friendly.
1800 * The kernel give the normal best effort behavior for "@expires+@delta",
1801 * but may decide to fire the timer earlier, but no earlier than @expires.
1802 *
1803 * You can set the task state as follows -
1804 *
1805 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1806 * pass before the routine returns.
1807 *
1808 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1809 * delivered to the current task.
1810 *
1811 * The current task state is guaranteed to be TASK_RUNNING when this
1812 * routine returns.
1813 *
1814 * Returns 0 when the timer has expired otherwise -EINTR
1815 */
1816int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta,
1817 const enum hrtimer_mode mode)
1818{
1819 return schedule_hrtimeout_range_clock(expires, delta, mode,
1820 CLOCK_MONOTONIC);
1821}
1822EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
1823
1824/**
1825 * schedule_hrtimeout - sleep until timeout
1826 * @expires: timeout value (ktime_t)
1827 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1828 *
1829 * Make the current task sleep until the given expiry time has
1830 * elapsed. The routine will return immediately unless
1831 * the current task state has been set (see set_current_state()).
1832 *
1833 * You can set the task state as follows -
1834 *
1835 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1836 * pass before the routine returns.
1837 *
1838 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1839 * delivered to the current task.
1840 *
1841 * The current task state is guaranteed to be TASK_RUNNING when this
1842 * routine returns.
1843 *
1844 * Returns 0 when the timer has expired otherwise -EINTR
1845 */
1846int __sched schedule_hrtimeout(ktime_t *expires,
1847 const enum hrtimer_mode mode)
1848{
1849 return schedule_hrtimeout_range(expires, 0, mode);
1850}
1851EXPORT_SYMBOL_GPL(schedule_hrtimeout);