projects/linux·blob v4.4-rc4 · refs

1811 lines · 46.3 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/export.h>
36#include <linux/percpu.h>
37#include <linux/hrtimer.h>
38#include <linux/notifier.h>
39#include <linux/syscalls.h>
40#include <linux/kallsyms.h>
41#include <linux/interrupt.h>
42#include <linux/tick.h>
43#include <linux/seq_file.h>
44#include <linux/err.h>
45#include <linux/debugobjects.h>
46#include <linux/sched.h>
47#include <linux/sched/sysctl.h>
48#include <linux/sched/rt.h>
49#include <linux/sched/deadline.h>
50#include <linux/timer.h>
51#include <linux/freezer.h>
52
53#include <asm/uaccess.h>
54
55#include <trace/events/timer.h>
56
57#include "tick-internal.h"
58
59/*
60 * The timer bases:
61 *
62 * There are more clockids than hrtimer bases. Thus, we index
63 * into the timer bases by the hrtimer_base_type enum. When trying
64 * to reach a base using a clockid, hrtimer_clockid_to_base()
65 * is used to convert from clockid to the proper hrtimer_base_type.
66 */
67DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
68{
69 .lock = __RAW_SPIN_LOCK_UNLOCKED(hrtimer_bases.lock),
70 .seq = SEQCNT_ZERO(hrtimer_bases.seq),
71 .clock_base =
72 {
73 {
74 .index = HRTIMER_BASE_MONOTONIC,
75 .clockid = CLOCK_MONOTONIC,
76 .get_time = &ktime_get,
77 },
78 {
79 .index = HRTIMER_BASE_REALTIME,
80 .clockid = CLOCK_REALTIME,
81 .get_time = &ktime_get_real,
82 },
83 {
84 .index = HRTIMER_BASE_BOOTTIME,
85 .clockid = CLOCK_BOOTTIME,
86 .get_time = &ktime_get_boottime,
87 },
88 {
89 .index = HRTIMER_BASE_TAI,
90 .clockid = CLOCK_TAI,
91 .get_time = &ktime_get_clocktai,
92 },
93 }
94};
95
96static const int hrtimer_clock_to_base_table[MAX_CLOCKS] = {
97 [CLOCK_REALTIME] = HRTIMER_BASE_REALTIME,
98 [CLOCK_MONOTONIC] = HRTIMER_BASE_MONOTONIC,
99 [CLOCK_BOOTTIME] = HRTIMER_BASE_BOOTTIME,
100 [CLOCK_TAI] = HRTIMER_BASE_TAI,
101};
102
103static inline int hrtimer_clockid_to_base(clockid_t clock_id)
104{
105 return hrtimer_clock_to_base_table[clock_id];
106}
107
108/*
109 * Functions and macros which are different for UP/SMP systems are kept in a
110 * single place
111 */
112#ifdef CONFIG_SMP
113
114/*
115 * We require the migration_base for lock_hrtimer_base()/switch_hrtimer_base()
116 * such that hrtimer_callback_running() can unconditionally dereference
117 * timer->base->cpu_base
118 */
119static struct hrtimer_cpu_base migration_cpu_base = {
120 .seq = SEQCNT_ZERO(migration_cpu_base),
121 .clock_base = { { .cpu_base = &migration_cpu_base, }, },
122};
123
124#define migration_base migration_cpu_base.clock_base[0]
125
126/*
127 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
128 * means that all timers which are tied to this base via timer->base are
129 * locked, and the base itself is locked too.
130 *
131 * So __run_timers/migrate_timers can safely modify all timers which could
132 * be found on the lists/queues.
133 *
134 * When the timer's base is locked, and the timer removed from list, it is
135 * possible to set timer->base = &migration_base and drop the lock: the timer
136 * remains locked.
137 */
138static
139struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
140 unsigned long *flags)
141{
142 struct hrtimer_clock_base *base;
143
144 for (;;) {
145 base = timer->base;
146 if (likely(base != &migration_base)) {
147 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
148 if (likely(base == timer->base))
149 return base;
150 /* The timer has migrated to another CPU: */
151 raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
152 }
153 cpu_relax();
154 }
155}
156
157/*
158 * With HIGHRES=y we do not migrate the timer when it is expiring
159 * before the next event on the target cpu because we cannot reprogram
160 * the target cpu hardware and we would cause it to fire late.
161 *
162 * Called with cpu_base->lock of target cpu held.
163 */
164static int
165hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base)
166{
167#ifdef CONFIG_HIGH_RES_TIMERS
168 ktime_t expires;
169
170 if (!new_base->cpu_base->hres_active)
171 return 0;
172
173 expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset);
174 return expires.tv64 <= new_base->cpu_base->expires_next.tv64;
175#else
176 return 0;
177#endif
178}
179
180#if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
181static inline
182struct hrtimer_cpu_base *get_target_base(struct hrtimer_cpu_base *base,
183 int pinned)
184{
185 if (pinned || !base->migration_enabled)
186 return base;
187 return &per_cpu(hrtimer_bases, get_nohz_timer_target());
188}
189#else
190static inline
191struct hrtimer_cpu_base *get_target_base(struct hrtimer_cpu_base *base,
192 int pinned)
193{
194 return base;
195}
196#endif
197
198/*
199 * We switch the timer base to a power-optimized selected CPU target,
200 * if:
201 * - NO_HZ_COMMON is enabled
202 * - timer migration is enabled
203 * - the timer callback is not running
204 * - the timer is not the first expiring timer on the new target
205 *
206 * If one of the above requirements is not fulfilled we move the timer
207 * to the current CPU or leave it on the previously assigned CPU if
208 * the timer callback is currently running.
209 */
210static inline struct hrtimer_clock_base *
211switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
212 int pinned)
213{
214 struct hrtimer_cpu_base *new_cpu_base, *this_cpu_base;
215 struct hrtimer_clock_base *new_base;
216 int basenum = base->index;
217
218 this_cpu_base = this_cpu_ptr(&hrtimer_bases);
219 new_cpu_base = get_target_base(this_cpu_base, pinned);
220again:
221 new_base = &new_cpu_base->clock_base[basenum];
222
223 if (base != new_base) {
224 /*
225 * We are trying to move timer to new_base.
226 * However we can't change timer's base while it is running,
227 * so we keep it on the same CPU. No hassle vs. reprogramming
228 * the event source in the high resolution case. The softirq
229 * code will take care of this when the timer function has
230 * completed. There is no conflict as we hold the lock until
231 * the timer is enqueued.
232 */
233 if (unlikely(hrtimer_callback_running(timer)))
234 return base;
235
236 /* See the comment in lock_hrtimer_base() */
237 timer->base = &migration_base;
238 raw_spin_unlock(&base->cpu_base->lock);
239 raw_spin_lock(&new_base->cpu_base->lock);
240
241 if (new_cpu_base != this_cpu_base &&
242 hrtimer_check_target(timer, new_base)) {
243 raw_spin_unlock(&new_base->cpu_base->lock);
244 raw_spin_lock(&base->cpu_base->lock);
245 new_cpu_base = this_cpu_base;
246 timer->base = base;
247 goto again;
248 }
249 timer->base = new_base;
250 } else {
251 if (new_cpu_base != this_cpu_base &&
252 hrtimer_check_target(timer, new_base)) {
253 new_cpu_base = this_cpu_base;
254 goto again;
255 }
256 }
257 return new_base;
258}
259
260#else /* CONFIG_SMP */
261
262static inline struct hrtimer_clock_base *
263lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
264{
265 struct hrtimer_clock_base *base = timer->base;
266
267 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
268
269 return base;
270}
271
272# define switch_hrtimer_base(t, b, p) (b)
273
274#endif /* !CONFIG_SMP */
275
276/*
277 * Functions for the union type storage format of ktime_t which are
278 * too large for inlining:
279 */
280#if BITS_PER_LONG < 64
281/*
282 * Divide a ktime value by a nanosecond value
283 */
284s64 __ktime_divns(const ktime_t kt, s64 div)
285{
286 int sft = 0;
287 s64 dclc;
288 u64 tmp;
289
290 dclc = ktime_to_ns(kt);
291 tmp = dclc < 0 ? -dclc : dclc;
292
293 /* Make sure the divisor is less than 2^32: */
294 while (div >> 32) {
295 sft++;
296 div >>= 1;
297 }
298 tmp >>= sft;
299 do_div(tmp, (unsigned long) div);
300 return dclc < 0 ? -tmp : tmp;
301}
302EXPORT_SYMBOL_GPL(__ktime_divns);
303#endif /* BITS_PER_LONG >= 64 */
304
305/*
306 * Add two ktime values and do a safety check for overflow:
307 */
308ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
309{
310 ktime_t res = ktime_add(lhs, rhs);
311
312 /*
313 * We use KTIME_SEC_MAX here, the maximum timeout which we can
314 * return to user space in a timespec:
315 */
316 if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64)
317 res = ktime_set(KTIME_SEC_MAX, 0);
318
319 return res;
320}
321
322EXPORT_SYMBOL_GPL(ktime_add_safe);
323
324#ifdef CONFIG_DEBUG_OBJECTS_TIMERS
325
326static struct debug_obj_descr hrtimer_debug_descr;
327
328static void *hrtimer_debug_hint(void *addr)
329{
330 return ((struct hrtimer *) addr)->function;
331}
332
333/*
334 * fixup_init is called when:
335 * - an active object is initialized
336 */
337static int hrtimer_fixup_init(void *addr, enum debug_obj_state state)
338{
339 struct hrtimer *timer = addr;
340
341 switch (state) {
342 case ODEBUG_STATE_ACTIVE:
343 hrtimer_cancel(timer);
344 debug_object_init(timer, &hrtimer_debug_descr);
345 return 1;
346 default:
347 return 0;
348 }
349}
350
351/*
352 * fixup_activate is called when:
353 * - an active object is activated
354 * - an unknown object is activated (might be a statically initialized object)
355 */
356static int hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
357{
358 switch (state) {
359
360 case ODEBUG_STATE_NOTAVAILABLE:
361 WARN_ON_ONCE(1);
362 return 0;
363
364 case ODEBUG_STATE_ACTIVE:
365 WARN_ON(1);
366
367 default:
368 return 0;
369 }
370}
371
372/*
373 * fixup_free is called when:
374 * - an active object is freed
375 */
376static int hrtimer_fixup_free(void *addr, enum debug_obj_state state)
377{
378 struct hrtimer *timer = addr;
379
380 switch (state) {
381 case ODEBUG_STATE_ACTIVE:
382 hrtimer_cancel(timer);
383 debug_object_free(timer, &hrtimer_debug_descr);
384 return 1;
385 default:
386 return 0;
387 }
388}
389
390static struct debug_obj_descr hrtimer_debug_descr = {
391 .name = "hrtimer",
392 .debug_hint = hrtimer_debug_hint,
393 .fixup_init = hrtimer_fixup_init,
394 .fixup_activate = hrtimer_fixup_activate,
395 .fixup_free = hrtimer_fixup_free,
396};
397
398static inline void debug_hrtimer_init(struct hrtimer *timer)
399{
400 debug_object_init(timer, &hrtimer_debug_descr);
401}
402
403static inline void debug_hrtimer_activate(struct hrtimer *timer)
404{
405 debug_object_activate(timer, &hrtimer_debug_descr);
406}
407
408static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
409{
410 debug_object_deactivate(timer, &hrtimer_debug_descr);
411}
412
413static inline void debug_hrtimer_free(struct hrtimer *timer)
414{
415 debug_object_free(timer, &hrtimer_debug_descr);
416}
417
418static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
419 enum hrtimer_mode mode);
420
421void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
422 enum hrtimer_mode mode)
423{
424 debug_object_init_on_stack(timer, &hrtimer_debug_descr);
425 __hrtimer_init(timer, clock_id, mode);
426}
427EXPORT_SYMBOL_GPL(hrtimer_init_on_stack);
428
429void destroy_hrtimer_on_stack(struct hrtimer *timer)
430{
431 debug_object_free(timer, &hrtimer_debug_descr);
432}
433
434#else
435static inline void debug_hrtimer_init(struct hrtimer *timer) { }
436static inline void debug_hrtimer_activate(struct hrtimer *timer) { }
437static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
438#endif
439
440static inline void
441debug_init(struct hrtimer *timer, clockid_t clockid,
442 enum hrtimer_mode mode)
443{
444 debug_hrtimer_init(timer);
445 trace_hrtimer_init(timer, clockid, mode);
446}
447
448static inline void debug_activate(struct hrtimer *timer)
449{
450 debug_hrtimer_activate(timer);
451 trace_hrtimer_start(timer);
452}
453
454static inline void debug_deactivate(struct hrtimer *timer)
455{
456 debug_hrtimer_deactivate(timer);
457 trace_hrtimer_cancel(timer);
458}
459
460#if defined(CONFIG_NO_HZ_COMMON) || defined(CONFIG_HIGH_RES_TIMERS)
461static inline void hrtimer_update_next_timer(struct hrtimer_cpu_base *cpu_base,
462 struct hrtimer *timer)
463{
464#ifdef CONFIG_HIGH_RES_TIMERS
465 cpu_base->next_timer = timer;
466#endif
467}
468
469static ktime_t __hrtimer_get_next_event(struct hrtimer_cpu_base *cpu_base)
470{
471 struct hrtimer_clock_base *base = cpu_base->clock_base;
472 ktime_t expires, expires_next = { .tv64 = KTIME_MAX };
473 unsigned int active = cpu_base->active_bases;
474
475 hrtimer_update_next_timer(cpu_base, NULL);
476 for (; active; base++, active >>= 1) {
477 struct timerqueue_node *next;
478 struct hrtimer *timer;
479
480 if (!(active & 0x01))
481 continue;
482
483 next = timerqueue_getnext(&base->active);
484 timer = container_of(next, struct hrtimer, node);
485 expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
486 if (expires.tv64 < expires_next.tv64) {
487 expires_next = expires;
488 hrtimer_update_next_timer(cpu_base, timer);
489 }
490 }
491 /*
492 * clock_was_set() might have changed base->offset of any of
493 * the clock bases so the result might be negative. Fix it up
494 * to prevent a false positive in clockevents_program_event().
495 */
496 if (expires_next.tv64 < 0)
497 expires_next.tv64 = 0;
498 return expires_next;
499}
500#endif
501
502static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base)
503{
504 ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset;
505 ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset;
506 ktime_t *offs_tai = &base->clock_base[HRTIMER_BASE_TAI].offset;
507
508 return ktime_get_update_offsets_now(&base->clock_was_set_seq,
509 offs_real, offs_boot, offs_tai);
510}
511
512/* High resolution timer related functions */
513#ifdef CONFIG_HIGH_RES_TIMERS
514
515/*
516 * High resolution timer enabled ?
517 */
518static int hrtimer_hres_enabled __read_mostly = 1;
519unsigned int hrtimer_resolution __read_mostly = LOW_RES_NSEC;
520EXPORT_SYMBOL_GPL(hrtimer_resolution);
521
522/*
523 * Enable / Disable high resolution mode
524 */
525static int __init setup_hrtimer_hres(char *str)
526{
527 if (!strcmp(str, "off"))
528 hrtimer_hres_enabled = 0;
529 else if (!strcmp(str, "on"))
530 hrtimer_hres_enabled = 1;
531 else
532 return 0;
533 return 1;
534}
535
536__setup("highres=", setup_hrtimer_hres);
537
538/*
539 * hrtimer_high_res_enabled - query, if the highres mode is enabled
540 */
541static inline int hrtimer_is_hres_enabled(void)
542{
543 return hrtimer_hres_enabled;
544}
545
546/*
547 * Is the high resolution mode active ?
548 */
549static inline int __hrtimer_hres_active(struct hrtimer_cpu_base *cpu_base)
550{
551 return cpu_base->hres_active;
552}
553
554static inline int hrtimer_hres_active(void)
555{
556 return __hrtimer_hres_active(this_cpu_ptr(&hrtimer_bases));
557}
558
559/*
560 * Reprogram the event source with checking both queues for the
561 * next event
562 * Called with interrupts disabled and base->lock held
563 */
564static void
565hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
566{
567 ktime_t expires_next;
568
569 if (!cpu_base->hres_active)
570 return;
571
572 expires_next = __hrtimer_get_next_event(cpu_base);
573
574 if (skip_equal && expires_next.tv64 == cpu_base->expires_next.tv64)
575 return;
576
577 cpu_base->expires_next.tv64 = expires_next.tv64;
578
579 /*
580 * If a hang was detected in the last timer interrupt then we
581 * leave the hang delay active in the hardware. We want the
582 * system to make progress. That also prevents the following
583 * scenario:
584 * T1 expires 50ms from now
585 * T2 expires 5s from now
586 *
587 * T1 is removed, so this code is called and would reprogram
588 * the hardware to 5s from now. Any hrtimer_start after that
589 * will not reprogram the hardware due to hang_detected being
590 * set. So we'd effectivly block all timers until the T2 event
591 * fires.
592 */
593 if (cpu_base->hang_detected)
594 return;
595
596 tick_program_event(cpu_base->expires_next, 1);
597}
598
599/*
600 * When a timer is enqueued and expires earlier than the already enqueued
601 * timers, we have to check, whether it expires earlier than the timer for
602 * which the clock event device was armed.
603 *
604 * Called with interrupts disabled and base->cpu_base.lock held
605 */
606static void hrtimer_reprogram(struct hrtimer *timer,
607 struct hrtimer_clock_base *base)
608{
609 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
610 ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
611
612 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
613
614 /*
615 * If the timer is not on the current cpu, we cannot reprogram
616 * the other cpus clock event device.
617 */
618 if (base->cpu_base != cpu_base)
619 return;
620
621 /*
622 * If the hrtimer interrupt is running, then it will
623 * reevaluate the clock bases and reprogram the clock event
624 * device. The callbacks are always executed in hard interrupt
625 * context so we don't need an extra check for a running
626 * callback.
627 */
628 if (cpu_base->in_hrtirq)
629 return;
630
631 /*
632 * CLOCK_REALTIME timer might be requested with an absolute
633 * expiry time which is less than base->offset. Set it to 0.
634 */
635 if (expires.tv64 < 0)
636 expires.tv64 = 0;
637
638 if (expires.tv64 >= cpu_base->expires_next.tv64)
639 return;
640
641 /* Update the pointer to the next expiring timer */
642 cpu_base->next_timer = timer;
643
644 /*
645 * If a hang was detected in the last timer interrupt then we
646 * do not schedule a timer which is earlier than the expiry
647 * which we enforced in the hang detection. We want the system
648 * to make progress.
649 */
650 if (cpu_base->hang_detected)
651 return;
652
653 /*
654 * Program the timer hardware. We enforce the expiry for
655 * events which are already in the past.
656 */
657 cpu_base->expires_next = expires;
658 tick_program_event(expires, 1);
659}
660
661/*
662 * Initialize the high resolution related parts of cpu_base
663 */
664static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
665{
666 base->expires_next.tv64 = KTIME_MAX;
667 base->hres_active = 0;
668}
669
670/*
671 * Retrigger next event is called after clock was set
672 *
673 * Called with interrupts disabled via on_each_cpu()
674 */
675static void retrigger_next_event(void *arg)
676{
677 struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
678
679 if (!base->hres_active)
680 return;
681
682 raw_spin_lock(&base->lock);
683 hrtimer_update_base(base);
684 hrtimer_force_reprogram(base, 0);
685 raw_spin_unlock(&base->lock);
686}
687
688/*
689 * Switch to high resolution mode
690 */
691static void hrtimer_switch_to_hres(void)
692{
693 struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
694
695 if (tick_init_highres()) {
696 printk(KERN_WARNING "Could not switch to high resolution "
697 "mode on CPU %d\n", base->cpu);
698 return;
699 }
700 base->hres_active = 1;
701 hrtimer_resolution = HIGH_RES_NSEC;
702
703 tick_setup_sched_timer();
704 /* "Retrigger" the interrupt to get things going */
705 retrigger_next_event(NULL);
706}
707
708static void clock_was_set_work(struct work_struct *work)
709{
710 clock_was_set();
711}
712
713static DECLARE_WORK(hrtimer_work, clock_was_set_work);
714
715/*
716 * Called from timekeeping and resume code to reprogramm the hrtimer
717 * interrupt device on all cpus.
718 */
719void clock_was_set_delayed(void)
720{
721 schedule_work(&hrtimer_work);
722}
723
724#else
725
726static inline int __hrtimer_hres_active(struct hrtimer_cpu_base *b) { return 0; }
727static inline int hrtimer_hres_active(void) { return 0; }
728static inline int hrtimer_is_hres_enabled(void) { return 0; }
729static inline void hrtimer_switch_to_hres(void) { }
730static inline void
731hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { }
732static inline int hrtimer_reprogram(struct hrtimer *timer,
733 struct hrtimer_clock_base *base)
734{
735 return 0;
736}
737static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { }
738static inline void retrigger_next_event(void *arg) { }
739
740#endif /* CONFIG_HIGH_RES_TIMERS */
741
742/*
743 * Clock realtime was set
744 *
745 * Change the offset of the realtime clock vs. the monotonic
746 * clock.
747 *
748 * We might have to reprogram the high resolution timer interrupt. On
749 * SMP we call the architecture specific code to retrigger _all_ high
750 * resolution timer interrupts. On UP we just disable interrupts and
751 * call the high resolution interrupt code.
752 */
753void clock_was_set(void)
754{
755#ifdef CONFIG_HIGH_RES_TIMERS
756 /* Retrigger the CPU local events everywhere */
757 on_each_cpu(retrigger_next_event, NULL, 1);
758#endif
759 timerfd_clock_was_set();
760}
761
762/*
763 * During resume we might have to reprogram the high resolution timer
764 * interrupt on all online CPUs. However, all other CPUs will be
765 * stopped with IRQs interrupts disabled so the clock_was_set() call
766 * must be deferred.
767 */
768void hrtimers_resume(void)
769{
770 WARN_ONCE(!irqs_disabled(),
771 KERN_INFO "hrtimers_resume() called with IRQs enabled!");
772
773 /* Retrigger on the local CPU */
774 retrigger_next_event(NULL);
775 /* And schedule a retrigger for all others */
776 clock_was_set_delayed();
777}
778
779static inline void timer_stats_hrtimer_set_start_info(struct hrtimer *timer)
780{
781#ifdef CONFIG_TIMER_STATS
782 if (timer->start_site)
783 return;
784 timer->start_site = __builtin_return_address(0);
785 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
786 timer->start_pid = current->pid;
787#endif
788}
789
790static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer *timer)
791{
792#ifdef CONFIG_TIMER_STATS
793 timer->start_site = NULL;
794#endif
795}
796
797static inline void timer_stats_account_hrtimer(struct hrtimer *timer)
798{
799#ifdef CONFIG_TIMER_STATS
800 if (likely(!timer_stats_active))
801 return;
802 timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
803 timer->function, timer->start_comm, 0);
804#endif
805}
806
807/*
808 * Counterpart to lock_hrtimer_base above:
809 */
810static inline
811void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
812{
813 raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
814}
815
816/**
817 * hrtimer_forward - forward the timer expiry
818 * @timer: hrtimer to forward
819 * @now: forward past this time
820 * @interval: the interval to forward
821 *
822 * Forward the timer expiry so it will expire in the future.
823 * Returns the number of overruns.
824 *
825 * Can be safely called from the callback function of @timer. If
826 * called from other contexts @timer must neither be enqueued nor
827 * running the callback and the caller needs to take care of
828 * serialization.
829 *
830 * Note: This only updates the timer expiry value and does not requeue
831 * the timer.
832 */
833u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
834{
835 u64 orun = 1;
836 ktime_t delta;
837
838 delta = ktime_sub(now, hrtimer_get_expires(timer));
839
840 if (delta.tv64 < 0)
841 return 0;
842
843 if (WARN_ON(timer->state & HRTIMER_STATE_ENQUEUED))
844 return 0;
845
846 if (interval.tv64 < hrtimer_resolution)
847 interval.tv64 = hrtimer_resolution;
848
849 if (unlikely(delta.tv64 >= interval.tv64)) {
850 s64 incr = ktime_to_ns(interval);
851
852 orun = ktime_divns(delta, incr);
853 hrtimer_add_expires_ns(timer, incr * orun);
854 if (hrtimer_get_expires_tv64(timer) > now.tv64)
855 return orun;
856 /*
857 * This (and the ktime_add() below) is the
858 * correction for exact:
859 */
860 orun++;
861 }
862 hrtimer_add_expires(timer, interval);
863
864 return orun;
865}
866EXPORT_SYMBOL_GPL(hrtimer_forward);
867
868/*
869 * enqueue_hrtimer - internal function to (re)start a timer
870 *
871 * The timer is inserted in expiry order. Insertion into the
872 * red black tree is O(log(n)). Must hold the base lock.
873 *
874 * Returns 1 when the new timer is the leftmost timer in the tree.
875 */
876static int enqueue_hrtimer(struct hrtimer *timer,
877 struct hrtimer_clock_base *base)
878{
879 debug_activate(timer);
880
881 base->cpu_base->active_bases |= 1 << base->index;
882
883 timer->state = HRTIMER_STATE_ENQUEUED;
884
885 return timerqueue_add(&base->active, &timer->node);
886}
887
888/*
889 * __remove_hrtimer - internal function to remove a timer
890 *
891 * Caller must hold the base lock.
892 *
893 * High resolution timer mode reprograms the clock event device when the
894 * timer is the one which expires next. The caller can disable this by setting
895 * reprogram to zero. This is useful, when the context does a reprogramming
896 * anyway (e.g. timer interrupt)
897 */
898static void __remove_hrtimer(struct hrtimer *timer,
899 struct hrtimer_clock_base *base,
900 unsigned long newstate, int reprogram)
901{
902 struct hrtimer_cpu_base *cpu_base = base->cpu_base;
903 unsigned int state = timer->state;
904
905 timer->state = newstate;
906 if (!(state & HRTIMER_STATE_ENQUEUED))
907 return;
908
909 if (!timerqueue_del(&base->active, &timer->node))
910 cpu_base->active_bases &= ~(1 << base->index);
911
912#ifdef CONFIG_HIGH_RES_TIMERS
913 /*
914 * Note: If reprogram is false we do not update
915 * cpu_base->next_timer. This happens when we remove the first
916 * timer on a remote cpu. No harm as we never dereference
917 * cpu_base->next_timer. So the worst thing what can happen is
918 * an superflous call to hrtimer_force_reprogram() on the
919 * remote cpu later on if the same timer gets enqueued again.
920 */
921 if (reprogram && timer == cpu_base->next_timer)
922 hrtimer_force_reprogram(cpu_base, 1);
923#endif
924}
925
926/*
927 * remove hrtimer, called with base lock held
928 */
929static inline int
930remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base, bool restart)
931{
932 if (hrtimer_is_queued(timer)) {
933 unsigned long state = timer->state;
934 int reprogram;
935
936 /*
937 * Remove the timer and force reprogramming when high
938 * resolution mode is active and the timer is on the current
939 * CPU. If we remove a timer on another CPU, reprogramming is
940 * skipped. The interrupt event on this CPU is fired and
941 * reprogramming happens in the interrupt handler. This is a
942 * rare case and less expensive than a smp call.
943 */
944 debug_deactivate(timer);
945 timer_stats_hrtimer_clear_start_info(timer);
946 reprogram = base->cpu_base == this_cpu_ptr(&hrtimer_bases);
947
948 if (!restart)
949 state = HRTIMER_STATE_INACTIVE;
950
951 __remove_hrtimer(timer, base, state, reprogram);
952 return 1;
953 }
954 return 0;
955}
956
957/**
958 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
959 * @timer: the timer to be added
960 * @tim: expiry time
961 * @delta_ns: "slack" range for the timer
962 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
963 * relative (HRTIMER_MODE_REL)
964 */
965void hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
966 unsigned long delta_ns, const enum hrtimer_mode mode)
967{
968 struct hrtimer_clock_base *base, *new_base;
969 unsigned long flags;
970 int leftmost;
971
972 base = lock_hrtimer_base(timer, &flags);
973
974 /* Remove an active timer from the queue: */
975 remove_hrtimer(timer, base, true);
976
977 if (mode & HRTIMER_MODE_REL) {
978 tim = ktime_add_safe(tim, base->get_time());
979 /*
980 * CONFIG_TIME_LOW_RES is a temporary way for architectures
981 * to signal that they simply return xtime in
982 * do_gettimeoffset(). In this case we want to round up by
983 * resolution when starting a relative timer, to avoid short
984 * timeouts. This will go away with the GTOD framework.
985 */
986#ifdef CONFIG_TIME_LOW_RES
987 tim = ktime_add_safe(tim, ktime_set(0, hrtimer_resolution));
988#endif
989 }
990
991 hrtimer_set_expires_range_ns(timer, tim, delta_ns);
992
993 /* Switch the timer base, if necessary: */
994 new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED);
995
996 timer_stats_hrtimer_set_start_info(timer);
997
998 leftmost = enqueue_hrtimer(timer, new_base);
999 if (!leftmost)
1000 goto unlock;
1001
1002 if (!hrtimer_is_hres_active(timer)) {
1003 /*
1004 * Kick to reschedule the next tick to handle the new timer
1005 * on dynticks target.
1006 */
1007 if (new_base->cpu_base->nohz_active)
1008 wake_up_nohz_cpu(new_base->cpu_base->cpu);
1009 } else {
1010 hrtimer_reprogram(timer, new_base);
1011 }
1012unlock:
1013 unlock_hrtimer_base(timer, &flags);
1014}
1015EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
1016
1017/**
1018 * hrtimer_try_to_cancel - try to deactivate a timer
1019 * @timer: hrtimer to stop
1020 *
1021 * Returns:
1022 * 0 when the timer was not active
1023 * 1 when the timer was active
1024 * -1 when the timer is currently excuting the callback function and
1025 * cannot be stopped
1026 */
1027int hrtimer_try_to_cancel(struct hrtimer *timer)
1028{
1029 struct hrtimer_clock_base *base;
1030 unsigned long flags;
1031 int ret = -1;
1032
1033 /*
1034 * Check lockless first. If the timer is not active (neither
1035 * enqueued nor running the callback, nothing to do here. The
1036 * base lock does not serialize against a concurrent enqueue,
1037 * so we can avoid taking it.
1038 */
1039 if (!hrtimer_active(timer))
1040 return 0;
1041
1042 base = lock_hrtimer_base(timer, &flags);
1043
1044 if (!hrtimer_callback_running(timer))
1045 ret = remove_hrtimer(timer, base, false);
1046
1047 unlock_hrtimer_base(timer, &flags);
1048
1049 return ret;
1050
1051}
1052EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1053
1054/**
1055 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1056 * @timer: the timer to be cancelled
1057 *
1058 * Returns:
1059 * 0 when the timer was not active
1060 * 1 when the timer was active
1061 */
1062int hrtimer_cancel(struct hrtimer *timer)
1063{
1064 for (;;) {
1065 int ret = hrtimer_try_to_cancel(timer);
1066
1067 if (ret >= 0)
1068 return ret;
1069 cpu_relax();
1070 }
1071}
1072EXPORT_SYMBOL_GPL(hrtimer_cancel);
1073
1074/**
1075 * hrtimer_get_remaining - get remaining time for the timer
1076 * @timer: the timer to read
1077 */
1078ktime_t hrtimer_get_remaining(const struct hrtimer *timer)
1079{
1080 unsigned long flags;
1081 ktime_t rem;
1082
1083 lock_hrtimer_base(timer, &flags);
1084 rem = hrtimer_expires_remaining(timer);
1085 unlock_hrtimer_base(timer, &flags);
1086
1087 return rem;
1088}
1089EXPORT_SYMBOL_GPL(hrtimer_get_remaining);
1090
1091#ifdef CONFIG_NO_HZ_COMMON
1092/**
1093 * hrtimer_get_next_event - get the time until next expiry event
1094 *
1095 * Returns the next expiry time or KTIME_MAX if no timer is pending.
1096 */
1097u64 hrtimer_get_next_event(void)
1098{
1099 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1100 u64 expires = KTIME_MAX;
1101 unsigned long flags;
1102
1103 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1104
1105 if (!__hrtimer_hres_active(cpu_base))
1106 expires = __hrtimer_get_next_event(cpu_base).tv64;
1107
1108 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1109
1110 return expires;
1111}
1112#endif
1113
1114static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1115 enum hrtimer_mode mode)
1116{
1117 struct hrtimer_cpu_base *cpu_base;
1118 int base;
1119
1120 memset(timer, 0, sizeof(struct hrtimer));
1121
1122 cpu_base = raw_cpu_ptr(&hrtimer_bases);
1123
1124 if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
1125 clock_id = CLOCK_MONOTONIC;
1126
1127 base = hrtimer_clockid_to_base(clock_id);
1128 timer->base = &cpu_base->clock_base[base];
1129 timerqueue_init(&timer->node);
1130
1131#ifdef CONFIG_TIMER_STATS
1132 timer->start_site = NULL;
1133 timer->start_pid = -1;
1134 memset(timer->start_comm, 0, TASK_COMM_LEN);
1135#endif
1136}
1137
1138/**
1139 * hrtimer_init - initialize a timer to the given clock
1140 * @timer: the timer to be initialized
1141 * @clock_id: the clock to be used
1142 * @mode: timer mode abs/rel
1143 */
1144void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1145 enum hrtimer_mode mode)
1146{
1147 debug_init(timer, clock_id, mode);
1148 __hrtimer_init(timer, clock_id, mode);
1149}
1150EXPORT_SYMBOL_GPL(hrtimer_init);
1151
1152/*
1153 * A timer is active, when it is enqueued into the rbtree or the
1154 * callback function is running or it's in the state of being migrated
1155 * to another cpu.
1156 *
1157 * It is important for this function to not return a false negative.
1158 */
1159bool hrtimer_active(const struct hrtimer *timer)
1160{
1161 struct hrtimer_cpu_base *cpu_base;
1162 unsigned int seq;
1163
1164 do {
1165 cpu_base = READ_ONCE(timer->base->cpu_base);
1166 seq = raw_read_seqcount_begin(&cpu_base->seq);
1167
1168 if (timer->state != HRTIMER_STATE_INACTIVE ||
1169 cpu_base->running == timer)
1170 return true;
1171
1172 } while (read_seqcount_retry(&cpu_base->seq, seq) ||
1173 cpu_base != READ_ONCE(timer->base->cpu_base));
1174
1175 return false;
1176}
1177EXPORT_SYMBOL_GPL(hrtimer_active);
1178
1179/*
1180 * The write_seqcount_barrier()s in __run_hrtimer() split the thing into 3
1181 * distinct sections:
1182 *
1183 * - queued: the timer is queued
1184 * - callback: the timer is being ran
1185 * - post: the timer is inactive or (re)queued
1186 *
1187 * On the read side we ensure we observe timer->state and cpu_base->running
1188 * from the same section, if anything changed while we looked at it, we retry.
1189 * This includes timer->base changing because sequence numbers alone are
1190 * insufficient for that.
1191 *
1192 * The sequence numbers are required because otherwise we could still observe
1193 * a false negative if the read side got smeared over multiple consequtive
1194 * __run_hrtimer() invocations.
1195 */
1196
1197static void __run_hrtimer(struct hrtimer_cpu_base *cpu_base,
1198 struct hrtimer_clock_base *base,
1199 struct hrtimer *timer, ktime_t *now)
1200{
1201 enum hrtimer_restart (*fn)(struct hrtimer *);
1202 int restart;
1203
1204 lockdep_assert_held(&cpu_base->lock);
1205
1206 debug_deactivate(timer);
1207 cpu_base->running = timer;
1208
1209 /*
1210 * Separate the ->running assignment from the ->state assignment.
1211 *
1212 * As with a regular write barrier, this ensures the read side in
1213 * hrtimer_active() cannot observe cpu_base->running == NULL &&
1214 * timer->state == INACTIVE.
1215 */
1216 raw_write_seqcount_barrier(&cpu_base->seq);
1217
1218 __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE, 0);
1219 timer_stats_account_hrtimer(timer);
1220 fn = timer->function;
1221
1222 /*
1223 * Because we run timers from hardirq context, there is no chance
1224 * they get migrated to another cpu, therefore its safe to unlock
1225 * the timer base.
1226 */
1227 raw_spin_unlock(&cpu_base->lock);
1228 trace_hrtimer_expire_entry(timer, now);
1229 restart = fn(timer);
1230 trace_hrtimer_expire_exit(timer);
1231 raw_spin_lock(&cpu_base->lock);
1232
1233 /*
1234 * Note: We clear the running state after enqueue_hrtimer and
1235 * we do not reprogramm the event hardware. Happens either in
1236 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1237 *
1238 * Note: Because we dropped the cpu_base->lock above,
1239 * hrtimer_start_range_ns() can have popped in and enqueued the timer
1240 * for us already.
1241 */
1242 if (restart != HRTIMER_NORESTART &&
1243 !(timer->state & HRTIMER_STATE_ENQUEUED))
1244 enqueue_hrtimer(timer, base);
1245
1246 /*
1247 * Separate the ->running assignment from the ->state assignment.
1248 *
1249 * As with a regular write barrier, this ensures the read side in
1250 * hrtimer_active() cannot observe cpu_base->running == NULL &&
1251 * timer->state == INACTIVE.
1252 */
1253 raw_write_seqcount_barrier(&cpu_base->seq);
1254
1255 WARN_ON_ONCE(cpu_base->running != timer);
1256 cpu_base->running = NULL;
1257}
1258
1259static void __hrtimer_run_queues(struct hrtimer_cpu_base *cpu_base, ktime_t now)
1260{
1261 struct hrtimer_clock_base *base = cpu_base->clock_base;
1262 unsigned int active = cpu_base->active_bases;
1263
1264 for (; active; base++, active >>= 1) {
1265 struct timerqueue_node *node;
1266 ktime_t basenow;
1267
1268 if (!(active & 0x01))
1269 continue;
1270
1271 basenow = ktime_add(now, base->offset);
1272
1273 while ((node = timerqueue_getnext(&base->active))) {
1274 struct hrtimer *timer;
1275
1276 timer = container_of(node, struct hrtimer, node);
1277
1278 /*
1279 * The immediate goal for using the softexpires is
1280 * minimizing wakeups, not running timers at the
1281 * earliest interrupt after their soft expiration.
1282 * This allows us to avoid using a Priority Search
1283 * Tree, which can answer a stabbing querry for
1284 * overlapping intervals and instead use the simple
1285 * BST we already have.
1286 * We don't add extra wakeups by delaying timers that
1287 * are right-of a not yet expired timer, because that
1288 * timer will have to trigger a wakeup anyway.
1289 */
1290 if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer))
1291 break;
1292
1293 __run_hrtimer(cpu_base, base, timer, &basenow);
1294 }
1295 }
1296}
1297
1298#ifdef CONFIG_HIGH_RES_TIMERS
1299
1300/*
1301 * High resolution timer interrupt
1302 * Called with interrupts disabled
1303 */
1304void hrtimer_interrupt(struct clock_event_device *dev)
1305{
1306 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1307 ktime_t expires_next, now, entry_time, delta;
1308 int retries = 0;
1309
1310 BUG_ON(!cpu_base->hres_active);
1311 cpu_base->nr_events++;
1312 dev->next_event.tv64 = KTIME_MAX;
1313
1314 raw_spin_lock(&cpu_base->lock);
1315 entry_time = now = hrtimer_update_base(cpu_base);
1316retry:
1317 cpu_base->in_hrtirq = 1;
1318 /*
1319 * We set expires_next to KTIME_MAX here with cpu_base->lock
1320 * held to prevent that a timer is enqueued in our queue via
1321 * the migration code. This does not affect enqueueing of
1322 * timers which run their callback and need to be requeued on
1323 * this CPU.
1324 */
1325 cpu_base->expires_next.tv64 = KTIME_MAX;
1326
1327 __hrtimer_run_queues(cpu_base, now);
1328
1329 /* Reevaluate the clock bases for the next expiry */
1330 expires_next = __hrtimer_get_next_event(cpu_base);
1331 /*
1332 * Store the new expiry value so the migration code can verify
1333 * against it.
1334 */
1335 cpu_base->expires_next = expires_next;
1336 cpu_base->in_hrtirq = 0;
1337 raw_spin_unlock(&cpu_base->lock);
1338
1339 /* Reprogramming necessary ? */
1340 if (!tick_program_event(expires_next, 0)) {
1341 cpu_base->hang_detected = 0;
1342 return;
1343 }
1344
1345 /*
1346 * The next timer was already expired due to:
1347 * - tracing
1348 * - long lasting callbacks
1349 * - being scheduled away when running in a VM
1350 *
1351 * We need to prevent that we loop forever in the hrtimer
1352 * interrupt routine. We give it 3 attempts to avoid
1353 * overreacting on some spurious event.
1354 *
1355 * Acquire base lock for updating the offsets and retrieving
1356 * the current time.
1357 */
1358 raw_spin_lock(&cpu_base->lock);
1359 now = hrtimer_update_base(cpu_base);
1360 cpu_base->nr_retries++;
1361 if (++retries < 3)
1362 goto retry;
1363 /*
1364 * Give the system a chance to do something else than looping
1365 * here. We stored the entry time, so we know exactly how long
1366 * we spent here. We schedule the next event this amount of
1367 * time away.
1368 */
1369 cpu_base->nr_hangs++;
1370 cpu_base->hang_detected = 1;
1371 raw_spin_unlock(&cpu_base->lock);
1372 delta = ktime_sub(now, entry_time);
1373 if ((unsigned int)delta.tv64 > cpu_base->max_hang_time)
1374 cpu_base->max_hang_time = (unsigned int) delta.tv64;
1375 /*
1376 * Limit it to a sensible value as we enforce a longer
1377 * delay. Give the CPU at least 100ms to catch up.
1378 */
1379 if (delta.tv64 > 100 * NSEC_PER_MSEC)
1380 expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1381 else
1382 expires_next = ktime_add(now, delta);
1383 tick_program_event(expires_next, 1);
1384 printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n",
1385 ktime_to_ns(delta));
1386}
1387
1388/*
1389 * local version of hrtimer_peek_ahead_timers() called with interrupts
1390 * disabled.
1391 */
1392static inline void __hrtimer_peek_ahead_timers(void)
1393{
1394 struct tick_device *td;
1395
1396 if (!hrtimer_hres_active())
1397 return;
1398
1399 td = this_cpu_ptr(&tick_cpu_device);
1400 if (td && td->evtdev)
1401 hrtimer_interrupt(td->evtdev);
1402}
1403
1404#else /* CONFIG_HIGH_RES_TIMERS */
1405
1406static inline void __hrtimer_peek_ahead_timers(void) { }
1407
1408#endif /* !CONFIG_HIGH_RES_TIMERS */
1409
1410/*
1411 * Called from run_local_timers in hardirq context every jiffy
1412 */
1413void hrtimer_run_queues(void)
1414{
1415 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1416 ktime_t now;
1417
1418 if (__hrtimer_hres_active(cpu_base))
1419 return;
1420
1421 /*
1422 * This _is_ ugly: We have to check periodically, whether we
1423 * can switch to highres and / or nohz mode. The clocksource
1424 * switch happens with xtime_lock held. Notification from
1425 * there only sets the check bit in the tick_oneshot code,
1426 * otherwise we might deadlock vs. xtime_lock.
1427 */
1428 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) {
1429 hrtimer_switch_to_hres();
1430 return;
1431 }
1432
1433 raw_spin_lock(&cpu_base->lock);
1434 now = hrtimer_update_base(cpu_base);
1435 __hrtimer_run_queues(cpu_base, now);
1436 raw_spin_unlock(&cpu_base->lock);
1437}
1438
1439/*
1440 * Sleep related functions:
1441 */
1442static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1443{
1444 struct hrtimer_sleeper *t =
1445 container_of(timer, struct hrtimer_sleeper, timer);
1446 struct task_struct *task = t->task;
1447
1448 t->task = NULL;
1449 if (task)
1450 wake_up_process(task);
1451
1452 return HRTIMER_NORESTART;
1453}
1454
1455void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1456{
1457 sl->timer.function = hrtimer_wakeup;
1458 sl->task = task;
1459}
1460EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
1461
1462static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1463{
1464 hrtimer_init_sleeper(t, current);
1465
1466 do {
1467 set_current_state(TASK_INTERRUPTIBLE);
1468 hrtimer_start_expires(&t->timer, mode);
1469
1470 if (likely(t->task))
1471 freezable_schedule();
1472
1473 hrtimer_cancel(&t->timer);
1474 mode = HRTIMER_MODE_ABS;
1475
1476 } while (t->task && !signal_pending(current));
1477
1478 __set_current_state(TASK_RUNNING);
1479
1480 return t->task == NULL;
1481}
1482
1483static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp)
1484{
1485 struct timespec rmt;
1486 ktime_t rem;
1487
1488 rem = hrtimer_expires_remaining(timer);
1489 if (rem.tv64 <= 0)
1490 return 0;
1491 rmt = ktime_to_timespec(rem);
1492
1493 if (copy_to_user(rmtp, &rmt, sizeof(*rmtp)))
1494 return -EFAULT;
1495
1496 return 1;
1497}
1498
1499long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1500{
1501 struct hrtimer_sleeper t;
1502 struct timespec __user *rmtp;
1503 int ret = 0;
1504
1505 hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid,
1506 HRTIMER_MODE_ABS);
1507 hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
1508
1509 if (do_nanosleep(&t, HRTIMER_MODE_ABS))
1510 goto out;
1511
1512 rmtp = restart->nanosleep.rmtp;
1513 if (rmtp) {
1514 ret = update_rmtp(&t.timer, rmtp);
1515 if (ret <= 0)
1516 goto out;
1517 }
1518
1519 /* The other values in restart are already filled in */
1520 ret = -ERESTART_RESTARTBLOCK;
1521out:
1522 destroy_hrtimer_on_stack(&t.timer);
1523 return ret;
1524}
1525
1526long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
1527 const enum hrtimer_mode mode, const clockid_t clockid)
1528{
1529 struct restart_block *restart;
1530 struct hrtimer_sleeper t;
1531 int ret = 0;
1532 unsigned long slack;
1533
1534 slack = current->timer_slack_ns;
1535 if (dl_task(current) || rt_task(current))
1536 slack = 0;
1537
1538 hrtimer_init_on_stack(&t.timer, clockid, mode);
1539 hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack);
1540 if (do_nanosleep(&t, mode))
1541 goto out;
1542
1543 /* Absolute timers do not update the rmtp value and restart: */
1544 if (mode == HRTIMER_MODE_ABS) {
1545 ret = -ERESTARTNOHAND;
1546 goto out;
1547 }
1548
1549 if (rmtp) {
1550 ret = update_rmtp(&t.timer, rmtp);
1551 if (ret <= 0)
1552 goto out;
1553 }
1554
1555 restart = &current->restart_block;
1556 restart->fn = hrtimer_nanosleep_restart;
1557 restart->nanosleep.clockid = t.timer.base->clockid;
1558 restart->nanosleep.rmtp = rmtp;
1559 restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
1560
1561 ret = -ERESTART_RESTARTBLOCK;
1562out:
1563 destroy_hrtimer_on_stack(&t.timer);
1564 return ret;
1565}
1566
1567SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp,
1568 struct timespec __user *, rmtp)
1569{
1570 struct timespec tu;
1571
1572 if (copy_from_user(&tu, rqtp, sizeof(tu)))
1573 return -EFAULT;
1574
1575 if (!timespec_valid(&tu))
1576 return -EINVAL;
1577
1578 return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1579}
1580
1581/*
1582 * Functions related to boot-time initialization:
1583 */
1584static void init_hrtimers_cpu(int cpu)
1585{
1586 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1587 int i;
1588
1589 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1590 cpu_base->clock_base[i].cpu_base = cpu_base;
1591 timerqueue_init_head(&cpu_base->clock_base[i].active);
1592 }
1593
1594 cpu_base->cpu = cpu;
1595 hrtimer_init_hres(cpu_base);
1596}
1597
1598#ifdef CONFIG_HOTPLUG_CPU
1599
1600static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1601 struct hrtimer_clock_base *new_base)
1602{
1603 struct hrtimer *timer;
1604 struct timerqueue_node *node;
1605
1606 while ((node = timerqueue_getnext(&old_base->active))) {
1607 timer = container_of(node, struct hrtimer, node);
1608 BUG_ON(hrtimer_callback_running(timer));
1609 debug_deactivate(timer);
1610
1611 /*
1612 * Mark it as ENQUEUED not INACTIVE otherwise the
1613 * timer could be seen as !active and just vanish away
1614 * under us on another CPU
1615 */
1616 __remove_hrtimer(timer, old_base, HRTIMER_STATE_ENQUEUED, 0);
1617 timer->base = new_base;
1618 /*
1619 * Enqueue the timers on the new cpu. This does not
1620 * reprogram the event device in case the timer
1621 * expires before the earliest on this CPU, but we run
1622 * hrtimer_interrupt after we migrated everything to
1623 * sort out already expired timers and reprogram the
1624 * event device.
1625 */
1626 enqueue_hrtimer(timer, new_base);
1627 }
1628}
1629
1630static void migrate_hrtimers(int scpu)
1631{
1632 struct hrtimer_cpu_base *old_base, *new_base;
1633 int i;
1634
1635 BUG_ON(cpu_online(scpu));
1636 tick_cancel_sched_timer(scpu);
1637
1638 local_irq_disable();
1639 old_base = &per_cpu(hrtimer_bases, scpu);
1640 new_base = this_cpu_ptr(&hrtimer_bases);
1641 /*
1642 * The caller is globally serialized and nobody else
1643 * takes two locks at once, deadlock is not possible.
1644 */
1645 raw_spin_lock(&new_base->lock);
1646 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1647
1648 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1649 migrate_hrtimer_list(&old_base->clock_base[i],
1650 &new_base->clock_base[i]);
1651 }
1652
1653 raw_spin_unlock(&old_base->lock);
1654 raw_spin_unlock(&new_base->lock);
1655
1656 /* Check, if we got expired work to do */
1657 __hrtimer_peek_ahead_timers();
1658 local_irq_enable();
1659}
1660
1661#endif /* CONFIG_HOTPLUG_CPU */
1662
1663static int hrtimer_cpu_notify(struct notifier_block *self,
1664 unsigned long action, void *hcpu)
1665{
1666 int scpu = (long)hcpu;
1667
1668 switch (action) {
1669
1670 case CPU_UP_PREPARE:
1671 case CPU_UP_PREPARE_FROZEN:
1672 init_hrtimers_cpu(scpu);
1673 break;
1674
1675#ifdef CONFIG_HOTPLUG_CPU
1676 case CPU_DEAD:
1677 case CPU_DEAD_FROZEN:
1678 migrate_hrtimers(scpu);
1679 break;
1680#endif
1681
1682 default:
1683 break;
1684 }
1685
1686 return NOTIFY_OK;
1687}
1688
1689static struct notifier_block hrtimers_nb = {
1690 .notifier_call = hrtimer_cpu_notify,
1691};
1692
1693void __init hrtimers_init(void)
1694{
1695 hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
1696 (void *)(long)smp_processor_id());
1697 register_cpu_notifier(&hrtimers_nb);
1698}
1699
1700/**
1701 * schedule_hrtimeout_range_clock - sleep until timeout
1702 * @expires: timeout value (ktime_t)
1703 * @delta: slack in expires timeout (ktime_t)
1704 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1705 * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME
1706 */
1707int __sched
1708schedule_hrtimeout_range_clock(ktime_t *expires, unsigned long delta,
1709 const enum hrtimer_mode mode, int clock)
1710{
1711 struct hrtimer_sleeper t;
1712
1713 /*
1714 * Optimize when a zero timeout value is given. It does not
1715 * matter whether this is an absolute or a relative time.
1716 */
1717 if (expires && !expires->tv64) {
1718 __set_current_state(TASK_RUNNING);
1719 return 0;
1720 }
1721
1722 /*
1723 * A NULL parameter means "infinite"
1724 */
1725 if (!expires) {
1726 schedule();
1727 return -EINTR;
1728 }
1729
1730 hrtimer_init_on_stack(&t.timer, clock, mode);
1731 hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
1732
1733 hrtimer_init_sleeper(&t, current);
1734
1735 hrtimer_start_expires(&t.timer, mode);
1736
1737 if (likely(t.task))
1738 schedule();
1739
1740 hrtimer_cancel(&t.timer);
1741 destroy_hrtimer_on_stack(&t.timer);
1742
1743 __set_current_state(TASK_RUNNING);
1744
1745 return !t.task ? 0 : -EINTR;
1746}
1747
1748/**
1749 * schedule_hrtimeout_range - sleep until timeout
1750 * @expires: timeout value (ktime_t)
1751 * @delta: slack in expires timeout (ktime_t)
1752 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1753 *
1754 * Make the current task sleep until the given expiry time has
1755 * elapsed. The routine will return immediately unless
1756 * the current task state has been set (see set_current_state()).
1757 *
1758 * The @delta argument gives the kernel the freedom to schedule the
1759 * actual wakeup to a time that is both power and performance friendly.
1760 * The kernel give the normal best effort behavior for "@expires+@delta",
1761 * but may decide to fire the timer earlier, but no earlier than @expires.
1762 *
1763 * You can set the task state as follows -
1764 *
1765 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1766 * pass before the routine returns.
1767 *
1768 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1769 * delivered to the current task.
1770 *
1771 * The current task state is guaranteed to be TASK_RUNNING when this
1772 * routine returns.
1773 *
1774 * Returns 0 when the timer has expired otherwise -EINTR
1775 */
1776int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta,
1777 const enum hrtimer_mode mode)
1778{
1779 return schedule_hrtimeout_range_clock(expires, delta, mode,
1780 CLOCK_MONOTONIC);
1781}
1782EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
1783
1784/**
1785 * schedule_hrtimeout - sleep until timeout
1786 * @expires: timeout value (ktime_t)
1787 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1788 *
1789 * Make the current task sleep until the given expiry time has
1790 * elapsed. The routine will return immediately unless
1791 * the current task state has been set (see set_current_state()).
1792 *
1793 * You can set the task state as follows -
1794 *
1795 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1796 * pass before the routine returns.
1797 *
1798 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1799 * delivered to the current task.
1800 *
1801 * The current task state is guaranteed to be TASK_RUNNING when this
1802 * routine returns.
1803 *
1804 * Returns 0 when the timer has expired otherwise -EINTR
1805 */
1806int __sched schedule_hrtimeout(ktime_t *expires,
1807 const enum hrtimer_mode mode)
1808{
1809 return schedule_hrtimeout_range(expires, 0, mode);
1810}
1811EXPORT_SYMBOL_GPL(schedule_hrtimeout);