projects/linux·blob v2.6.39-rc1 · 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
86static inline int hrtimer_clockid_to_base(clockid_t clock_id)
87{
88 return hrtimer_clock_to_base_table[clock_id];
89}
90
91
92/*
93 * Get the coarse grained time at the softirq based on xtime and
94 * wall_to_monotonic.
95 */
96static void hrtimer_get_softirq_time(struct hrtimer_cpu_base *base)
97{
98 ktime_t xtim, mono, boot;
99 struct timespec xts, tom, slp;
100
101 get_xtime_and_monotonic_and_sleep_offset(&xts, &tom, &slp);
102
103 xtim = timespec_to_ktime(xts);
104 mono = ktime_add(xtim, timespec_to_ktime(tom));
105 boot = ktime_add(mono, timespec_to_ktime(slp));
106 base->clock_base[HRTIMER_BASE_REALTIME].softirq_time = xtim;
107 base->clock_base[HRTIMER_BASE_MONOTONIC].softirq_time = mono;
108 base->clock_base[HRTIMER_BASE_BOOTTIME].softirq_time = boot;
109}
110
111/*
112 * Functions and macros which are different for UP/SMP systems are kept in a
113 * single place
114 */
115#ifdef CONFIG_SMP
116
117/*
118 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
119 * means that all timers which are tied to this base via timer->base are
120 * locked, and the base itself is locked too.
121 *
122 * So __run_timers/migrate_timers can safely modify all timers which could
123 * be found on the lists/queues.
124 *
125 * When the timer's base is locked, and the timer removed from list, it is
126 * possible to set timer->base = NULL and drop the lock: the timer remains
127 * locked.
128 */
129static
130struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
131 unsigned long *flags)
132{
133 struct hrtimer_clock_base *base;
134
135 for (;;) {
136 base = timer->base;
137 if (likely(base != NULL)) {
138 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
139 if (likely(base == timer->base))
140 return base;
141 /* The timer has migrated to another CPU: */
142 raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
143 }
144 cpu_relax();
145 }
146}
147
148
149/*
150 * Get the preferred target CPU for NOHZ
151 */
152static int hrtimer_get_target(int this_cpu, int pinned)
153{
154#ifdef CONFIG_NO_HZ
155 if (!pinned && get_sysctl_timer_migration() && idle_cpu(this_cpu))
156 return get_nohz_timer_target();
157#endif
158 return this_cpu;
159}
160
161/*
162 * With HIGHRES=y we do not migrate the timer when it is expiring
163 * before the next event on the target cpu because we cannot reprogram
164 * the target cpu hardware and we would cause it to fire late.
165 *
166 * Called with cpu_base->lock of target cpu held.
167 */
168static int
169hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base)
170{
171#ifdef CONFIG_HIGH_RES_TIMERS
172 ktime_t expires;
173
174 if (!new_base->cpu_base->hres_active)
175 return 0;
176
177 expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset);
178 return expires.tv64 <= new_base->cpu_base->expires_next.tv64;
179#else
180 return 0;
181#endif
182}
183
184/*
185 * Switch the timer base to the current CPU when possible.
186 */
187static inline struct hrtimer_clock_base *
188switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
189 int pinned)
190{
191 struct hrtimer_clock_base *new_base;
192 struct hrtimer_cpu_base *new_cpu_base;
193 int this_cpu = smp_processor_id();
194 int cpu = hrtimer_get_target(this_cpu, pinned);
195 int basenum = hrtimer_clockid_to_base(base->index);
196
197again:
198 new_cpu_base = &per_cpu(hrtimer_bases, cpu);
199 new_base = &new_cpu_base->clock_base[basenum];
200
201 if (base != new_base) {
202 /*
203 * We are trying to move timer to new_base.
204 * However we can't change timer's base while it is running,
205 * so we keep it on the same CPU. No hassle vs. reprogramming
206 * the event source in the high resolution case. The softirq
207 * code will take care of this when the timer function has
208 * completed. There is no conflict as we hold the lock until
209 * the timer is enqueued.
210 */
211 if (unlikely(hrtimer_callback_running(timer)))
212 return base;
213
214 /* See the comment in lock_timer_base() */
215 timer->base = NULL;
216 raw_spin_unlock(&base->cpu_base->lock);
217 raw_spin_lock(&new_base->cpu_base->lock);
218
219 if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) {
220 cpu = this_cpu;
221 raw_spin_unlock(&new_base->cpu_base->lock);
222 raw_spin_lock(&base->cpu_base->lock);
223 timer->base = base;
224 goto again;
225 }
226 timer->base = new_base;
227 }
228 return new_base;
229}
230
231#else /* CONFIG_SMP */
232
233static inline struct hrtimer_clock_base *
234lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
235{
236 struct hrtimer_clock_base *base = timer->base;
237
238 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
239
240 return base;
241}
242
243# define switch_hrtimer_base(t, b, p) (b)
244
245#endif /* !CONFIG_SMP */
246
247/*
248 * Functions for the union type storage format of ktime_t which are
249 * too large for inlining:
250 */
251#if BITS_PER_LONG < 64
252# ifndef CONFIG_KTIME_SCALAR
253/**
254 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
255 * @kt: addend
256 * @nsec: the scalar nsec value to add
257 *
258 * Returns the sum of kt and nsec in ktime_t format
259 */
260ktime_t ktime_add_ns(const ktime_t kt, u64 nsec)
261{
262 ktime_t tmp;
263
264 if (likely(nsec < NSEC_PER_SEC)) {
265 tmp.tv64 = nsec;
266 } else {
267 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
268
269 tmp = ktime_set((long)nsec, rem);
270 }
271
272 return ktime_add(kt, tmp);
273}
274
275EXPORT_SYMBOL_GPL(ktime_add_ns);
276
277/**
278 * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable
279 * @kt: minuend
280 * @nsec: the scalar nsec value to subtract
281 *
282 * Returns the subtraction of @nsec from @kt in ktime_t format
283 */
284ktime_t ktime_sub_ns(const ktime_t kt, u64 nsec)
285{
286 ktime_t tmp;
287
288 if (likely(nsec < NSEC_PER_SEC)) {
289 tmp.tv64 = nsec;
290 } else {
291 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
292
293 tmp = ktime_set((long)nsec, rem);
294 }
295
296 return ktime_sub(kt, tmp);
297}
298
299EXPORT_SYMBOL_GPL(ktime_sub_ns);
300# endif /* !CONFIG_KTIME_SCALAR */
301
302/*
303 * Divide a ktime value by a nanosecond value
304 */
305u64 ktime_divns(const ktime_t kt, s64 div)
306{
307 u64 dclc;
308 int sft = 0;
309
310 dclc = ktime_to_ns(kt);
311 /* Make sure the divisor is less than 2^32: */
312 while (div >> 32) {
313 sft++;
314 div >>= 1;
315 }
316 dclc >>= sft;
317 do_div(dclc, (unsigned long) div);
318
319 return dclc;
320}
321#endif /* BITS_PER_LONG >= 64 */
322
323/*
324 * Add two ktime values and do a safety check for overflow:
325 */
326ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
327{
328 ktime_t res = ktime_add(lhs, rhs);
329
330 /*
331 * We use KTIME_SEC_MAX here, the maximum timeout which we can
332 * return to user space in a timespec:
333 */
334 if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64)
335 res = ktime_set(KTIME_SEC_MAX, 0);
336
337 return res;
338}
339
340EXPORT_SYMBOL_GPL(ktime_add_safe);
341
342#ifdef CONFIG_DEBUG_OBJECTS_TIMERS
343
344static struct debug_obj_descr hrtimer_debug_descr;
345
346static void *hrtimer_debug_hint(void *addr)
347{
348 return ((struct hrtimer *) addr)->function;
349}
350
351/*
352 * fixup_init is called when:
353 * - an active object is initialized
354 */
355static int hrtimer_fixup_init(void *addr, enum debug_obj_state state)
356{
357 struct hrtimer *timer = addr;
358
359 switch (state) {
360 case ODEBUG_STATE_ACTIVE:
361 hrtimer_cancel(timer);
362 debug_object_init(timer, &hrtimer_debug_descr);
363 return 1;
364 default:
365 return 0;
366 }
367}
368
369/*
370 * fixup_activate is called when:
371 * - an active object is activated
372 * - an unknown object is activated (might be a statically initialized object)
373 */
374static int hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
375{
376 switch (state) {
377
378 case ODEBUG_STATE_NOTAVAILABLE:
379 WARN_ON_ONCE(1);
380 return 0;
381
382 case ODEBUG_STATE_ACTIVE:
383 WARN_ON(1);
384
385 default:
386 return 0;
387 }
388}
389
390/*
391 * fixup_free is called when:
392 * - an active object is freed
393 */
394static int hrtimer_fixup_free(void *addr, enum debug_obj_state state)
395{
396 struct hrtimer *timer = addr;
397
398 switch (state) {
399 case ODEBUG_STATE_ACTIVE:
400 hrtimer_cancel(timer);
401 debug_object_free(timer, &hrtimer_debug_descr);
402 return 1;
403 default:
404 return 0;
405 }
406}
407
408static struct debug_obj_descr hrtimer_debug_descr = {
409 .name = "hrtimer",
410 .debug_hint = hrtimer_debug_hint,
411 .fixup_init = hrtimer_fixup_init,
412 .fixup_activate = hrtimer_fixup_activate,
413 .fixup_free = hrtimer_fixup_free,
414};
415
416static inline void debug_hrtimer_init(struct hrtimer *timer)
417{
418 debug_object_init(timer, &hrtimer_debug_descr);
419}
420
421static inline void debug_hrtimer_activate(struct hrtimer *timer)
422{
423 debug_object_activate(timer, &hrtimer_debug_descr);
424}
425
426static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
427{
428 debug_object_deactivate(timer, &hrtimer_debug_descr);
429}
430
431static inline void debug_hrtimer_free(struct hrtimer *timer)
432{
433 debug_object_free(timer, &hrtimer_debug_descr);
434}
435
436static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
437 enum hrtimer_mode mode);
438
439void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
440 enum hrtimer_mode mode)
441{
442 debug_object_init_on_stack(timer, &hrtimer_debug_descr);
443 __hrtimer_init(timer, clock_id, mode);
444}
445EXPORT_SYMBOL_GPL(hrtimer_init_on_stack);
446
447void destroy_hrtimer_on_stack(struct hrtimer *timer)
448{
449 debug_object_free(timer, &hrtimer_debug_descr);
450}
451
452#else
453static inline void debug_hrtimer_init(struct hrtimer *timer) { }
454static inline void debug_hrtimer_activate(struct hrtimer *timer) { }
455static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
456#endif
457
458static inline void
459debug_init(struct hrtimer *timer, clockid_t clockid,
460 enum hrtimer_mode mode)
461{
462 debug_hrtimer_init(timer);
463 trace_hrtimer_init(timer, clockid, mode);
464}
465
466static inline void debug_activate(struct hrtimer *timer)
467{
468 debug_hrtimer_activate(timer);
469 trace_hrtimer_start(timer);
470}
471
472static inline void debug_deactivate(struct hrtimer *timer)
473{
474 debug_hrtimer_deactivate(timer);
475 trace_hrtimer_cancel(timer);
476}
477
478/* High resolution timer related functions */
479#ifdef CONFIG_HIGH_RES_TIMERS
480
481/*
482 * High resolution timer enabled ?
483 */
484static int hrtimer_hres_enabled __read_mostly = 1;
485
486/*
487 * Enable / Disable high resolution mode
488 */
489static int __init setup_hrtimer_hres(char *str)
490{
491 if (!strcmp(str, "off"))
492 hrtimer_hres_enabled = 0;
493 else if (!strcmp(str, "on"))
494 hrtimer_hres_enabled = 1;
495 else
496 return 0;
497 return 1;
498}
499
500__setup("highres=", setup_hrtimer_hres);
501
502/*
503 * hrtimer_high_res_enabled - query, if the highres mode is enabled
504 */
505static inline int hrtimer_is_hres_enabled(void)
506{
507 return hrtimer_hres_enabled;
508}
509
510/*
511 * Is the high resolution mode active ?
512 */
513static inline int hrtimer_hres_active(void)
514{
515 return __this_cpu_read(hrtimer_bases.hres_active);
516}
517
518/*
519 * Reprogram the event source with checking both queues for the
520 * next event
521 * Called with interrupts disabled and base->lock held
522 */
523static void
524hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
525{
526 int i;
527 struct hrtimer_clock_base *base = cpu_base->clock_base;
528 ktime_t expires, expires_next;
529
530 expires_next.tv64 = KTIME_MAX;
531
532 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
533 struct hrtimer *timer;
534 struct timerqueue_node *next;
535
536 next = timerqueue_getnext(&base->active);
537 if (!next)
538 continue;
539 timer = container_of(next, struct hrtimer, node);
540
541 expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
542 /*
543 * clock_was_set() has changed base->offset so the
544 * result might be negative. Fix it up to prevent a
545 * false positive in clockevents_program_event()
546 */
547 if (expires.tv64 < 0)
548 expires.tv64 = 0;
549 if (expires.tv64 < expires_next.tv64)
550 expires_next = expires;
551 }
552
553 if (skip_equal && expires_next.tv64 == cpu_base->expires_next.tv64)
554 return;
555
556 cpu_base->expires_next.tv64 = expires_next.tv64;
557
558 if (cpu_base->expires_next.tv64 != KTIME_MAX)
559 tick_program_event(cpu_base->expires_next, 1);
560}
561
562/*
563 * Shared reprogramming for clock_realtime and clock_monotonic
564 *
565 * When a timer is enqueued and expires earlier than the already enqueued
566 * timers, we have to check, whether it expires earlier than the timer for
567 * which the clock event device was armed.
568 *
569 * Called with interrupts disabled and base->cpu_base.lock held
570 */
571static int hrtimer_reprogram(struct hrtimer *timer,
572 struct hrtimer_clock_base *base)
573{
574 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
575 ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
576 int res;
577
578 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
579
580 /*
581 * When the callback is running, we do not reprogram the clock event
582 * device. The timer callback is either running on a different CPU or
583 * the callback is executed in the hrtimer_interrupt context. The
584 * reprogramming is handled either by the softirq, which called the
585 * callback or at the end of the hrtimer_interrupt.
586 */
587 if (hrtimer_callback_running(timer))
588 return 0;
589
590 /*
591 * CLOCK_REALTIME timer might be requested with an absolute
592 * expiry time which is less than base->offset. Nothing wrong
593 * about that, just avoid to call into the tick code, which
594 * has now objections against negative expiry values.
595 */
596 if (expires.tv64 < 0)
597 return -ETIME;
598
599 if (expires.tv64 >= cpu_base->expires_next.tv64)
600 return 0;
601
602 /*
603 * If a hang was detected in the last timer interrupt then we
604 * do not schedule a timer which is earlier than the expiry
605 * which we enforced in the hang detection. We want the system
606 * to make progress.
607 */
608 if (cpu_base->hang_detected)
609 return 0;
610
611 /*
612 * Clockevents returns -ETIME, when the event was in the past.
613 */
614 res = tick_program_event(expires, 0);
615 if (!IS_ERR_VALUE(res))
616 cpu_base->expires_next = expires;
617 return res;
618}
619
620
621/*
622 * Retrigger next event is called after clock was set
623 *
624 * Called with interrupts disabled via on_each_cpu()
625 */
626static void retrigger_next_event(void *arg)
627{
628 struct hrtimer_cpu_base *base;
629 struct timespec realtime_offset, wtm, sleep;
630
631 if (!hrtimer_hres_active())
632 return;
633
634 get_xtime_and_monotonic_and_sleep_offset(&realtime_offset, &wtm,
635 &sleep);
636 set_normalized_timespec(&realtime_offset, -wtm.tv_sec, -wtm.tv_nsec);
637
638 base = &__get_cpu_var(hrtimer_bases);
639
640 /* Adjust CLOCK_REALTIME offset */
641 raw_spin_lock(&base->lock);
642 base->clock_base[HRTIMER_BASE_REALTIME].offset =
643 timespec_to_ktime(realtime_offset);
644 base->clock_base[HRTIMER_BASE_BOOTTIME].offset =
645 timespec_to_ktime(sleep);
646
647 hrtimer_force_reprogram(base, 0);
648 raw_spin_unlock(&base->lock);
649}
650
651/*
652 * Clock realtime was set
653 *
654 * Change the offset of the realtime clock vs. the monotonic
655 * clock.
656 *
657 * We might have to reprogram the high resolution timer interrupt. On
658 * SMP we call the architecture specific code to retrigger _all_ high
659 * resolution timer interrupts. On UP we just disable interrupts and
660 * call the high resolution interrupt code.
661 */
662void clock_was_set(void)
663{
664 /* Retrigger the CPU local events everywhere */
665 on_each_cpu(retrigger_next_event, NULL, 1);
666}
667
668/*
669 * During resume we might have to reprogram the high resolution timer
670 * interrupt (on the local CPU):
671 */
672void hres_timers_resume(void)
673{
674 WARN_ONCE(!irqs_disabled(),
675 KERN_INFO "hres_timers_resume() called with IRQs enabled!");
676
677 retrigger_next_event(NULL);
678}
679
680/*
681 * Initialize the high resolution related parts of cpu_base
682 */
683static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
684{
685 base->expires_next.tv64 = KTIME_MAX;
686 base->hres_active = 0;
687}
688
689/*
690 * When High resolution timers are active, try to reprogram. Note, that in case
691 * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry
692 * check happens. The timer gets enqueued into the rbtree. The reprogramming
693 * and expiry check is done in the hrtimer_interrupt or in the softirq.
694 */
695static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
696 struct hrtimer_clock_base *base,
697 int wakeup)
698{
699 if (base->cpu_base->hres_active && hrtimer_reprogram(timer, base)) {
700 if (wakeup) {
701 raw_spin_unlock(&base->cpu_base->lock);
702 raise_softirq_irqoff(HRTIMER_SOFTIRQ);
703 raw_spin_lock(&base->cpu_base->lock);
704 } else
705 __raise_softirq_irqoff(HRTIMER_SOFTIRQ);
706
707 return 1;
708 }
709
710 return 0;
711}
712
713/*
714 * Switch to high resolution mode
715 */
716static int hrtimer_switch_to_hres(void)
717{
718 int cpu = smp_processor_id();
719 struct hrtimer_cpu_base *base = &per_cpu(hrtimer_bases, cpu);
720 unsigned long flags;
721
722 if (base->hres_active)
723 return 1;
724
725 local_irq_save(flags);
726
727 if (tick_init_highres()) {
728 local_irq_restore(flags);
729 printk(KERN_WARNING "Could not switch to high resolution "
730 "mode on CPU %d\n", cpu);
731 return 0;
732 }
733 base->hres_active = 1;
734 base->clock_base[HRTIMER_BASE_REALTIME].resolution = KTIME_HIGH_RES;
735 base->clock_base[HRTIMER_BASE_MONOTONIC].resolution = KTIME_HIGH_RES;
736 base->clock_base[HRTIMER_BASE_BOOTTIME].resolution = KTIME_HIGH_RES;
737
738 tick_setup_sched_timer();
739
740 /* "Retrigger" the interrupt to get things going */
741 retrigger_next_event(NULL);
742 local_irq_restore(flags);
743 return 1;
744}
745
746#else
747
748static inline int hrtimer_hres_active(void) { return 0; }
749static inline int hrtimer_is_hres_enabled(void) { return 0; }
750static inline int hrtimer_switch_to_hres(void) { return 0; }
751static inline void
752hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { }
753static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
754 struct hrtimer_clock_base *base,
755 int wakeup)
756{
757 return 0;
758}
759static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { }
760
761#endif /* CONFIG_HIGH_RES_TIMERS */
762
763static inline void timer_stats_hrtimer_set_start_info(struct hrtimer *timer)
764{
765#ifdef CONFIG_TIMER_STATS
766 if (timer->start_site)
767 return;
768 timer->start_site = __builtin_return_address(0);
769 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
770 timer->start_pid = current->pid;
771#endif
772}
773
774static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer *timer)
775{
776#ifdef CONFIG_TIMER_STATS
777 timer->start_site = NULL;
778#endif
779}
780
781static inline void timer_stats_account_hrtimer(struct hrtimer *timer)
782{
783#ifdef CONFIG_TIMER_STATS
784 if (likely(!timer_stats_active))
785 return;
786 timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
787 timer->function, timer->start_comm, 0);
788#endif
789}
790
791/*
792 * Counterpart to lock_hrtimer_base above:
793 */
794static inline
795void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
796{
797 raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
798}
799
800/**
801 * hrtimer_forward - forward the timer expiry
802 * @timer: hrtimer to forward
803 * @now: forward past this time
804 * @interval: the interval to forward
805 *
806 * Forward the timer expiry so it will expire in the future.
807 * Returns the number of overruns.
808 */
809u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
810{
811 u64 orun = 1;
812 ktime_t delta;
813
814 delta = ktime_sub(now, hrtimer_get_expires(timer));
815
816 if (delta.tv64 < 0)
817 return 0;
818
819 if (interval.tv64 < timer->base->resolution.tv64)
820 interval.tv64 = timer->base->resolution.tv64;
821
822 if (unlikely(delta.tv64 >= interval.tv64)) {
823 s64 incr = ktime_to_ns(interval);
824
825 orun = ktime_divns(delta, incr);
826 hrtimer_add_expires_ns(timer, incr * orun);
827 if (hrtimer_get_expires_tv64(timer) > now.tv64)
828 return orun;
829 /*
830 * This (and the ktime_add() below) is the
831 * correction for exact:
832 */
833 orun++;
834 }
835 hrtimer_add_expires(timer, interval);
836
837 return orun;
838}
839EXPORT_SYMBOL_GPL(hrtimer_forward);
840
841/*
842 * enqueue_hrtimer - internal function to (re)start a timer
843 *
844 * The timer is inserted in expiry order. Insertion into the
845 * red black tree is O(log(n)). Must hold the base lock.
846 *
847 * Returns 1 when the new timer is the leftmost timer in the tree.
848 */
849static int enqueue_hrtimer(struct hrtimer *timer,
850 struct hrtimer_clock_base *base)
851{
852 debug_activate(timer);
853
854 timerqueue_add(&base->active, &timer->node);
855
856 /*
857 * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
858 * state of a possibly running callback.
859 */
860 timer->state |= HRTIMER_STATE_ENQUEUED;
861
862 return (&timer->node == base->active.next);
863}
864
865/*
866 * __remove_hrtimer - internal function to remove a timer
867 *
868 * Caller must hold the base lock.
869 *
870 * High resolution timer mode reprograms the clock event device when the
871 * timer is the one which expires next. The caller can disable this by setting
872 * reprogram to zero. This is useful, when the context does a reprogramming
873 * anyway (e.g. timer interrupt)
874 */
875static void __remove_hrtimer(struct hrtimer *timer,
876 struct hrtimer_clock_base *base,
877 unsigned long newstate, int reprogram)
878{
879 if (!(timer->state & HRTIMER_STATE_ENQUEUED))
880 goto out;
881
882 if (&timer->node == timerqueue_getnext(&base->active)) {
883#ifdef CONFIG_HIGH_RES_TIMERS
884 /* Reprogram the clock event device. if enabled */
885 if (reprogram && hrtimer_hres_active()) {
886 ktime_t expires;
887
888 expires = ktime_sub(hrtimer_get_expires(timer),
889 base->offset);
890 if (base->cpu_base->expires_next.tv64 == expires.tv64)
891 hrtimer_force_reprogram(base->cpu_base, 1);
892 }
893#endif
894 }
895 timerqueue_del(&base->active, &timer->node);
896out:
897 timer->state = newstate;
898}
899
900/*
901 * remove hrtimer, called with base lock held
902 */
903static inline int
904remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base)
905{
906 if (hrtimer_is_queued(timer)) {
907 unsigned long state;
908 int reprogram;
909
910 /*
911 * Remove the timer and force reprogramming when high
912 * resolution mode is active and the timer is on the current
913 * CPU. If we remove a timer on another CPU, reprogramming is
914 * skipped. The interrupt event on this CPU is fired and
915 * reprogramming happens in the interrupt handler. This is a
916 * rare case and less expensive than a smp call.
917 */
918 debug_deactivate(timer);
919 timer_stats_hrtimer_clear_start_info(timer);
920 reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases);
921 /*
922 * We must preserve the CALLBACK state flag here,
923 * otherwise we could move the timer base in
924 * switch_hrtimer_base.
925 */
926 state = timer->state & HRTIMER_STATE_CALLBACK;
927 __remove_hrtimer(timer, base, state, reprogram);
928 return 1;
929 }
930 return 0;
931}
932
933int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
934 unsigned long delta_ns, const enum hrtimer_mode mode,
935 int wakeup)
936{
937 struct hrtimer_clock_base *base, *new_base;
938 unsigned long flags;
939 int ret, leftmost;
940
941 base = lock_hrtimer_base(timer, &flags);
942
943 /* Remove an active timer from the queue: */
944 ret = remove_hrtimer(timer, base);
945
946 /* Switch the timer base, if necessary: */
947 new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED);
948
949 if (mode & HRTIMER_MODE_REL) {
950 tim = ktime_add_safe(tim, new_base->get_time());
951 /*
952 * CONFIG_TIME_LOW_RES is a temporary way for architectures
953 * to signal that they simply return xtime in
954 * do_gettimeoffset(). In this case we want to round up by
955 * resolution when starting a relative timer, to avoid short
956 * timeouts. This will go away with the GTOD framework.
957 */
958#ifdef CONFIG_TIME_LOW_RES
959 tim = ktime_add_safe(tim, base->resolution);
960#endif
961 }
962
963 hrtimer_set_expires_range_ns(timer, tim, delta_ns);
964
965 timer_stats_hrtimer_set_start_info(timer);
966
967 leftmost = enqueue_hrtimer(timer, new_base);
968
969 /*
970 * Only allow reprogramming if the new base is on this CPU.
971 * (it might still be on another CPU if the timer was pending)
972 *
973 * XXX send_remote_softirq() ?
974 */
975 if (leftmost && new_base->cpu_base == &__get_cpu_var(hrtimer_bases))
976 hrtimer_enqueue_reprogram(timer, new_base, wakeup);
977
978 unlock_hrtimer_base(timer, &flags);
979
980 return ret;
981}
982
983/**
984 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
985 * @timer: the timer to be added
986 * @tim: expiry time
987 * @delta_ns: "slack" range for the timer
988 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
989 *
990 * Returns:
991 * 0 on success
992 * 1 when the timer was active
993 */
994int hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
995 unsigned long delta_ns, const enum hrtimer_mode mode)
996{
997 return __hrtimer_start_range_ns(timer, tim, delta_ns, mode, 1);
998}
999EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
1000
1001/**
1002 * hrtimer_start - (re)start an hrtimer on the current CPU
1003 * @timer: the timer to be added
1004 * @tim: expiry time
1005 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
1006 *
1007 * Returns:
1008 * 0 on success
1009 * 1 when the timer was active
1010 */
1011int
1012hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode)
1013{
1014 return __hrtimer_start_range_ns(timer, tim, 0, mode, 1);
1015}
1016EXPORT_SYMBOL_GPL(hrtimer_start);
1017
1018
1019/**
1020 * hrtimer_try_to_cancel - try to deactivate a timer
1021 * @timer: hrtimer to stop
1022 *
1023 * Returns:
1024 * 0 when the timer was not active
1025 * 1 when the timer was active
1026 * -1 when the timer is currently excuting the callback function and
1027 * cannot be stopped
1028 */
1029int hrtimer_try_to_cancel(struct hrtimer *timer)
1030{
1031 struct hrtimer_clock_base *base;
1032 unsigned long flags;
1033 int ret = -1;
1034
1035 base = lock_hrtimer_base(timer, &flags);
1036
1037 if (!hrtimer_callback_running(timer))
1038 ret = remove_hrtimer(timer, base);
1039
1040 unlock_hrtimer_base(timer, &flags);
1041
1042 return ret;
1043
1044}
1045EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1046
1047/**
1048 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1049 * @timer: the timer to be cancelled
1050 *
1051 * Returns:
1052 * 0 when the timer was not active
1053 * 1 when the timer was active
1054 */
1055int hrtimer_cancel(struct hrtimer *timer)
1056{
1057 for (;;) {
1058 int ret = hrtimer_try_to_cancel(timer);
1059
1060 if (ret >= 0)
1061 return ret;
1062 cpu_relax();
1063 }
1064}
1065EXPORT_SYMBOL_GPL(hrtimer_cancel);
1066
1067/**
1068 * hrtimer_get_remaining - get remaining time for the timer
1069 * @timer: the timer to read
1070 */
1071ktime_t hrtimer_get_remaining(const struct hrtimer *timer)
1072{
1073 unsigned long flags;
1074 ktime_t rem;
1075
1076 lock_hrtimer_base(timer, &flags);
1077 rem = hrtimer_expires_remaining(timer);
1078 unlock_hrtimer_base(timer, &flags);
1079
1080 return rem;
1081}
1082EXPORT_SYMBOL_GPL(hrtimer_get_remaining);
1083
1084#ifdef CONFIG_NO_HZ
1085/**
1086 * hrtimer_get_next_event - get the time until next expiry event
1087 *
1088 * Returns the delta to the next expiry event or KTIME_MAX if no timer
1089 * is pending.
1090 */
1091ktime_t hrtimer_get_next_event(void)
1092{
1093 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1094 struct hrtimer_clock_base *base = cpu_base->clock_base;
1095 ktime_t delta, mindelta = { .tv64 = KTIME_MAX };
1096 unsigned long flags;
1097 int i;
1098
1099 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1100
1101 if (!hrtimer_hres_active()) {
1102 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
1103 struct hrtimer *timer;
1104 struct timerqueue_node *next;
1105
1106 next = timerqueue_getnext(&base->active);
1107 if (!next)
1108 continue;
1109
1110 timer = container_of(next, struct hrtimer, node);
1111 delta.tv64 = hrtimer_get_expires_tv64(timer);
1112 delta = ktime_sub(delta, base->get_time());
1113 if (delta.tv64 < mindelta.tv64)
1114 mindelta.tv64 = delta.tv64;
1115 }
1116 }
1117
1118 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1119
1120 if (mindelta.tv64 < 0)
1121 mindelta.tv64 = 0;
1122 return mindelta;
1123}
1124#endif
1125
1126static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1127 enum hrtimer_mode mode)
1128{
1129 struct hrtimer_cpu_base *cpu_base;
1130 int base;
1131
1132 memset(timer, 0, sizeof(struct hrtimer));
1133
1134 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1135
1136 if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
1137 clock_id = CLOCK_MONOTONIC;
1138
1139 base = hrtimer_clockid_to_base(clock_id);
1140 timer->base = &cpu_base->clock_base[base];
1141 timerqueue_init(&timer->node);
1142
1143#ifdef CONFIG_TIMER_STATS
1144 timer->start_site = NULL;
1145 timer->start_pid = -1;
1146 memset(timer->start_comm, 0, TASK_COMM_LEN);
1147#endif
1148}
1149
1150/**
1151 * hrtimer_init - initialize a timer to the given clock
1152 * @timer: the timer to be initialized
1153 * @clock_id: the clock to be used
1154 * @mode: timer mode abs/rel
1155 */
1156void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1157 enum hrtimer_mode mode)
1158{
1159 debug_init(timer, clock_id, mode);
1160 __hrtimer_init(timer, clock_id, mode);
1161}
1162EXPORT_SYMBOL_GPL(hrtimer_init);
1163
1164/**
1165 * hrtimer_get_res - get the timer resolution for a clock
1166 * @which_clock: which clock to query
1167 * @tp: pointer to timespec variable to store the resolution
1168 *
1169 * Store the resolution of the clock selected by @which_clock in the
1170 * variable pointed to by @tp.
1171 */
1172int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp)
1173{
1174 struct hrtimer_cpu_base *cpu_base;
1175 int base = hrtimer_clockid_to_base(which_clock);
1176
1177 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1178 *tp = ktime_to_timespec(cpu_base->clock_base[base].resolution);
1179
1180 return 0;
1181}
1182EXPORT_SYMBOL_GPL(hrtimer_get_res);
1183
1184static void __run_hrtimer(struct hrtimer *timer, ktime_t *now)
1185{
1186 struct hrtimer_clock_base *base = timer->base;
1187 struct hrtimer_cpu_base *cpu_base = base->cpu_base;
1188 enum hrtimer_restart (*fn)(struct hrtimer *);
1189 int restart;
1190
1191 WARN_ON(!irqs_disabled());
1192
1193 debug_deactivate(timer);
1194 __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0);
1195 timer_stats_account_hrtimer(timer);
1196 fn = timer->function;
1197
1198 /*
1199 * Because we run timers from hardirq context, there is no chance
1200 * they get migrated to another cpu, therefore its safe to unlock
1201 * the timer base.
1202 */
1203 raw_spin_unlock(&cpu_base->lock);
1204 trace_hrtimer_expire_entry(timer, now);
1205 restart = fn(timer);
1206 trace_hrtimer_expire_exit(timer);
1207 raw_spin_lock(&cpu_base->lock);
1208
1209 /*
1210 * Note: We clear the CALLBACK bit after enqueue_hrtimer and
1211 * we do not reprogramm the event hardware. Happens either in
1212 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1213 */
1214 if (restart != HRTIMER_NORESTART) {
1215 BUG_ON(timer->state != HRTIMER_STATE_CALLBACK);
1216 enqueue_hrtimer(timer, base);
1217 }
1218
1219 WARN_ON_ONCE(!(timer->state & HRTIMER_STATE_CALLBACK));
1220
1221 timer->state &= ~HRTIMER_STATE_CALLBACK;
1222}
1223
1224#ifdef CONFIG_HIGH_RES_TIMERS
1225
1226/*
1227 * High resolution timer interrupt
1228 * Called with interrupts disabled
1229 */
1230void hrtimer_interrupt(struct clock_event_device *dev)
1231{
1232 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1233 struct hrtimer_clock_base *base;
1234 ktime_t expires_next, now, entry_time, delta;
1235 int i, retries = 0;
1236
1237 BUG_ON(!cpu_base->hres_active);
1238 cpu_base->nr_events++;
1239 dev->next_event.tv64 = KTIME_MAX;
1240
1241 entry_time = now = ktime_get();
1242retry:
1243 expires_next.tv64 = KTIME_MAX;
1244
1245 raw_spin_lock(&cpu_base->lock);
1246 /*
1247 * We set expires_next to KTIME_MAX here with cpu_base->lock
1248 * held to prevent that a timer is enqueued in our queue via
1249 * the migration code. This does not affect enqueueing of
1250 * timers which run their callback and need to be requeued on
1251 * this CPU.
1252 */
1253 cpu_base->expires_next.tv64 = KTIME_MAX;
1254
1255 base = cpu_base->clock_base;
1256
1257 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1258 ktime_t basenow;
1259 struct timerqueue_node *node;
1260
1261 basenow = ktime_add(now, base->offset);
1262
1263 while ((node = timerqueue_getnext(&base->active))) {
1264 struct hrtimer *timer;
1265
1266 timer = container_of(node, struct hrtimer, node);
1267
1268 /*
1269 * The immediate goal for using the softexpires is
1270 * minimizing wakeups, not running timers at the
1271 * earliest interrupt after their soft expiration.
1272 * This allows us to avoid using a Priority Search
1273 * Tree, which can answer a stabbing querry for
1274 * overlapping intervals and instead use the simple
1275 * BST we already have.
1276 * We don't add extra wakeups by delaying timers that
1277 * are right-of a not yet expired timer, because that
1278 * timer will have to trigger a wakeup anyway.
1279 */
1280
1281 if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer)) {
1282 ktime_t expires;
1283
1284 expires = ktime_sub(hrtimer_get_expires(timer),
1285 base->offset);
1286 if (expires.tv64 < expires_next.tv64)
1287 expires_next = expires;
1288 break;
1289 }
1290
1291 __run_hrtimer(timer, &basenow);
1292 }
1293 base++;
1294 }
1295
1296 /*
1297 * Store the new expiry value so the migration code can verify
1298 * against it.
1299 */
1300 cpu_base->expires_next = expires_next;
1301 raw_spin_unlock(&cpu_base->lock);
1302
1303 /* Reprogramming necessary ? */
1304 if (expires_next.tv64 == KTIME_MAX ||
1305 !tick_program_event(expires_next, 0)) {
1306 cpu_base->hang_detected = 0;
1307 return;
1308 }
1309
1310 /*
1311 * The next timer was already expired due to:
1312 * - tracing
1313 * - long lasting callbacks
1314 * - being scheduled away when running in a VM
1315 *
1316 * We need to prevent that we loop forever in the hrtimer
1317 * interrupt routine. We give it 3 attempts to avoid
1318 * overreacting on some spurious event.
1319 */
1320 now = ktime_get();
1321 cpu_base->nr_retries++;
1322 if (++retries < 3)
1323 goto retry;
1324 /*
1325 * Give the system a chance to do something else than looping
1326 * here. We stored the entry time, so we know exactly how long
1327 * we spent here. We schedule the next event this amount of
1328 * time away.
1329 */
1330 cpu_base->nr_hangs++;
1331 cpu_base->hang_detected = 1;
1332 delta = ktime_sub(now, entry_time);
1333 if (delta.tv64 > cpu_base->max_hang_time.tv64)
1334 cpu_base->max_hang_time = delta;
1335 /*
1336 * Limit it to a sensible value as we enforce a longer
1337 * delay. Give the CPU at least 100ms to catch up.
1338 */
1339 if (delta.tv64 > 100 * NSEC_PER_MSEC)
1340 expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1341 else
1342 expires_next = ktime_add(now, delta);
1343 tick_program_event(expires_next, 1);
1344 printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n",
1345 ktime_to_ns(delta));
1346}
1347
1348/*
1349 * local version of hrtimer_peek_ahead_timers() called with interrupts
1350 * disabled.
1351 */
1352static void __hrtimer_peek_ahead_timers(void)
1353{
1354 struct tick_device *td;
1355
1356 if (!hrtimer_hres_active())
1357 return;
1358
1359 td = &__get_cpu_var(tick_cpu_device);
1360 if (td && td->evtdev)
1361 hrtimer_interrupt(td->evtdev);
1362}
1363
1364/**
1365 * hrtimer_peek_ahead_timers -- run soft-expired timers now
1366 *
1367 * hrtimer_peek_ahead_timers will peek at the timer queue of
1368 * the current cpu and check if there are any timers for which
1369 * the soft expires time has passed. If any such timers exist,
1370 * they are run immediately and then removed from the timer queue.
1371 *
1372 */
1373void hrtimer_peek_ahead_timers(void)
1374{
1375 unsigned long flags;
1376
1377 local_irq_save(flags);
1378 __hrtimer_peek_ahead_timers();
1379 local_irq_restore(flags);
1380}
1381
1382static void run_hrtimer_softirq(struct softirq_action *h)
1383{
1384 hrtimer_peek_ahead_timers();
1385}
1386
1387#else /* CONFIG_HIGH_RES_TIMERS */
1388
1389static inline void __hrtimer_peek_ahead_timers(void) { }
1390
1391#endif /* !CONFIG_HIGH_RES_TIMERS */
1392
1393/*
1394 * Called from timer softirq every jiffy, expire hrtimers:
1395 *
1396 * For HRT its the fall back code to run the softirq in the timer
1397 * softirq context in case the hrtimer initialization failed or has
1398 * not been done yet.
1399 */
1400void hrtimer_run_pending(void)
1401{
1402 if (hrtimer_hres_active())
1403 return;
1404
1405 /*
1406 * This _is_ ugly: We have to check in the softirq context,
1407 * whether we can switch to highres and / or nohz mode. The
1408 * clocksource switch happens in the timer interrupt with
1409 * xtime_lock held. Notification from there only sets the
1410 * check bit in the tick_oneshot code, otherwise we might
1411 * deadlock vs. xtime_lock.
1412 */
1413 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1414 hrtimer_switch_to_hres();
1415}
1416
1417/*
1418 * Called from hardirq context every jiffy
1419 */
1420void hrtimer_run_queues(void)
1421{
1422 struct timerqueue_node *node;
1423 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1424 struct hrtimer_clock_base *base;
1425 int index, gettime = 1;
1426
1427 if (hrtimer_hres_active())
1428 return;
1429
1430 for (index = 0; index < HRTIMER_MAX_CLOCK_BASES; index++) {
1431 base = &cpu_base->clock_base[index];
1432 if (!timerqueue_getnext(&base->active))
1433 continue;
1434
1435 if (gettime) {
1436 hrtimer_get_softirq_time(cpu_base);
1437 gettime = 0;
1438 }
1439
1440 raw_spin_lock(&cpu_base->lock);
1441
1442 while ((node = timerqueue_getnext(&base->active))) {
1443 struct hrtimer *timer;
1444
1445 timer = container_of(node, struct hrtimer, node);
1446 if (base->softirq_time.tv64 <=
1447 hrtimer_get_expires_tv64(timer))
1448 break;
1449
1450 __run_hrtimer(timer, &base->softirq_time);
1451 }
1452 raw_spin_unlock(&cpu_base->lock);
1453 }
1454}
1455
1456/*
1457 * Sleep related functions:
1458 */
1459static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1460{
1461 struct hrtimer_sleeper *t =
1462 container_of(timer, struct hrtimer_sleeper, timer);
1463 struct task_struct *task = t->task;
1464
1465 t->task = NULL;
1466 if (task)
1467 wake_up_process(task);
1468
1469 return HRTIMER_NORESTART;
1470}
1471
1472void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1473{
1474 sl->timer.function = hrtimer_wakeup;
1475 sl->task = task;
1476}
1477EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
1478
1479static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1480{
1481 hrtimer_init_sleeper(t, current);
1482
1483 do {
1484 set_current_state(TASK_INTERRUPTIBLE);
1485 hrtimer_start_expires(&t->timer, mode);
1486 if (!hrtimer_active(&t->timer))
1487 t->task = NULL;
1488
1489 if (likely(t->task))
1490 schedule();
1491
1492 hrtimer_cancel(&t->timer);
1493 mode = HRTIMER_MODE_ABS;
1494
1495 } while (t->task && !signal_pending(current));
1496
1497 __set_current_state(TASK_RUNNING);
1498
1499 return t->task == NULL;
1500}
1501
1502static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp)
1503{
1504 struct timespec rmt;
1505 ktime_t rem;
1506
1507 rem = hrtimer_expires_remaining(timer);
1508 if (rem.tv64 <= 0)
1509 return 0;
1510 rmt = ktime_to_timespec(rem);
1511
1512 if (copy_to_user(rmtp, &rmt, sizeof(*rmtp)))
1513 return -EFAULT;
1514
1515 return 1;
1516}
1517
1518long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1519{
1520 struct hrtimer_sleeper t;
1521 struct timespec __user *rmtp;
1522 int ret = 0;
1523
1524 hrtimer_init_on_stack(&t.timer, restart->nanosleep.index,
1525 HRTIMER_MODE_ABS);
1526 hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
1527
1528 if (do_nanosleep(&t, HRTIMER_MODE_ABS))
1529 goto out;
1530
1531 rmtp = restart->nanosleep.rmtp;
1532 if (rmtp) {
1533 ret = update_rmtp(&t.timer, rmtp);
1534 if (ret <= 0)
1535 goto out;
1536 }
1537
1538 /* The other values in restart are already filled in */
1539 ret = -ERESTART_RESTARTBLOCK;
1540out:
1541 destroy_hrtimer_on_stack(&t.timer);
1542 return ret;
1543}
1544
1545long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
1546 const enum hrtimer_mode mode, const clockid_t clockid)
1547{
1548 struct restart_block *restart;
1549 struct hrtimer_sleeper t;
1550 int ret = 0;
1551 unsigned long slack;
1552
1553 slack = current->timer_slack_ns;
1554 if (rt_task(current))
1555 slack = 0;
1556
1557 hrtimer_init_on_stack(&t.timer, clockid, mode);
1558 hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack);
1559 if (do_nanosleep(&t, mode))
1560 goto out;
1561
1562 /* Absolute timers do not update the rmtp value and restart: */
1563 if (mode == HRTIMER_MODE_ABS) {
1564 ret = -ERESTARTNOHAND;
1565 goto out;
1566 }
1567
1568 if (rmtp) {
1569 ret = update_rmtp(&t.timer, rmtp);
1570 if (ret <= 0)
1571 goto out;
1572 }
1573
1574 restart = &current_thread_info()->restart_block;
1575 restart->fn = hrtimer_nanosleep_restart;
1576 restart->nanosleep.index = t.timer.base->index;
1577 restart->nanosleep.rmtp = rmtp;
1578 restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
1579
1580 ret = -ERESTART_RESTARTBLOCK;
1581out:
1582 destroy_hrtimer_on_stack(&t.timer);
1583 return ret;
1584}
1585
1586SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp,
1587 struct timespec __user *, rmtp)
1588{
1589 struct timespec tu;
1590
1591 if (copy_from_user(&tu, rqtp, sizeof(tu)))
1592 return -EFAULT;
1593
1594 if (!timespec_valid(&tu))
1595 return -EINVAL;
1596
1597 return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1598}
1599
1600/*
1601 * Functions related to boot-time initialization:
1602 */
1603static void __cpuinit init_hrtimers_cpu(int cpu)
1604{
1605 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1606 int i;
1607
1608 raw_spin_lock_init(&cpu_base->lock);
1609
1610 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1611 cpu_base->clock_base[i].cpu_base = cpu_base;
1612 timerqueue_init_head(&cpu_base->clock_base[i].active);
1613 }
1614
1615 hrtimer_init_hres(cpu_base);
1616}
1617
1618#ifdef CONFIG_HOTPLUG_CPU
1619
1620static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1621 struct hrtimer_clock_base *new_base)
1622{
1623 struct hrtimer *timer;
1624 struct timerqueue_node *node;
1625
1626 while ((node = timerqueue_getnext(&old_base->active))) {
1627 timer = container_of(node, struct hrtimer, node);
1628 BUG_ON(hrtimer_callback_running(timer));
1629 debug_deactivate(timer);
1630
1631 /*
1632 * Mark it as STATE_MIGRATE not INACTIVE otherwise the
1633 * timer could be seen as !active and just vanish away
1634 * under us on another CPU
1635 */
1636 __remove_hrtimer(timer, old_base, HRTIMER_STATE_MIGRATE, 0);
1637 timer->base = new_base;
1638 /*
1639 * Enqueue the timers on the new cpu. This does not
1640 * reprogram the event device in case the timer
1641 * expires before the earliest on this CPU, but we run
1642 * hrtimer_interrupt after we migrated everything to
1643 * sort out already expired timers and reprogram the
1644 * event device.
1645 */
1646 enqueue_hrtimer(timer, new_base);
1647
1648 /* Clear the migration state bit */
1649 timer->state &= ~HRTIMER_STATE_MIGRATE;
1650 }
1651}
1652
1653static void migrate_hrtimers(int scpu)
1654{
1655 struct hrtimer_cpu_base *old_base, *new_base;
1656 int i;
1657
1658 BUG_ON(cpu_online(scpu));
1659 tick_cancel_sched_timer(scpu);
1660
1661 local_irq_disable();
1662 old_base = &per_cpu(hrtimer_bases, scpu);
1663 new_base = &__get_cpu_var(hrtimer_bases);
1664 /*
1665 * The caller is globally serialized and nobody else
1666 * takes two locks at once, deadlock is not possible.
1667 */
1668 raw_spin_lock(&new_base->lock);
1669 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1670
1671 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1672 migrate_hrtimer_list(&old_base->clock_base[i],
1673 &new_base->clock_base[i]);
1674 }
1675
1676 raw_spin_unlock(&old_base->lock);
1677 raw_spin_unlock(&new_base->lock);
1678
1679 /* Check, if we got expired work to do */
1680 __hrtimer_peek_ahead_timers();
1681 local_irq_enable();
1682}
1683
1684#endif /* CONFIG_HOTPLUG_CPU */
1685
1686static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self,
1687 unsigned long action, void *hcpu)
1688{
1689 int scpu = (long)hcpu;
1690
1691 switch (action) {
1692
1693 case CPU_UP_PREPARE:
1694 case CPU_UP_PREPARE_FROZEN:
1695 init_hrtimers_cpu(scpu);
1696 break;
1697
1698#ifdef CONFIG_HOTPLUG_CPU
1699 case CPU_DYING:
1700 case CPU_DYING_FROZEN:
1701 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DYING, &scpu);
1702 break;
1703 case CPU_DEAD:
1704 case CPU_DEAD_FROZEN:
1705 {
1706 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &scpu);
1707 migrate_hrtimers(scpu);
1708 break;
1709 }
1710#endif
1711
1712 default:
1713 break;
1714 }
1715
1716 return NOTIFY_OK;
1717}
1718
1719static struct notifier_block __cpuinitdata hrtimers_nb = {
1720 .notifier_call = hrtimer_cpu_notify,
1721};
1722
1723void __init hrtimers_init(void)
1724{
1725 hrtimer_clock_to_base_table[CLOCK_REALTIME] = HRTIMER_BASE_REALTIME;
1726 hrtimer_clock_to_base_table[CLOCK_MONOTONIC] = HRTIMER_BASE_MONOTONIC;
1727 hrtimer_clock_to_base_table[CLOCK_BOOTTIME] = HRTIMER_BASE_BOOTTIME;
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);