projects/linux·blob v4.7 · refs

1820 lines · 46.7 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 bool 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 true;
346 default:
347 return false;
348 }
349}
350
351/*
352 * fixup_activate is called when:
353 * - an active object is activated
354 * - an unknown non-static object is activated
355 */
356static bool hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
357{
358 switch (state) {
359 case ODEBUG_STATE_ACTIVE:
360 WARN_ON(1);
361
362 default:
363 return false;
364 }
365}
366
367/*
368 * fixup_free is called when:
369 * - an active object is freed
370 */
371static bool hrtimer_fixup_free(void *addr, enum debug_obj_state state)
372{
373 struct hrtimer *timer = addr;
374
375 switch (state) {
376 case ODEBUG_STATE_ACTIVE:
377 hrtimer_cancel(timer);
378 debug_object_free(timer, &hrtimer_debug_descr);
379 return true;
380 default:
381 return false;
382 }
383}
384
385static struct debug_obj_descr hrtimer_debug_descr = {
386 .name = "hrtimer",
387 .debug_hint = hrtimer_debug_hint,
388 .fixup_init = hrtimer_fixup_init,
389 .fixup_activate = hrtimer_fixup_activate,
390 .fixup_free = hrtimer_fixup_free,
391};
392
393static inline void debug_hrtimer_init(struct hrtimer *timer)
394{
395 debug_object_init(timer, &hrtimer_debug_descr);
396}
397
398static inline void debug_hrtimer_activate(struct hrtimer *timer)
399{
400 debug_object_activate(timer, &hrtimer_debug_descr);
401}
402
403static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
404{
405 debug_object_deactivate(timer, &hrtimer_debug_descr);
406}
407
408static inline void debug_hrtimer_free(struct hrtimer *timer)
409{
410 debug_object_free(timer, &hrtimer_debug_descr);
411}
412
413static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
414 enum hrtimer_mode mode);
415
416void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
417 enum hrtimer_mode mode)
418{
419 debug_object_init_on_stack(timer, &hrtimer_debug_descr);
420 __hrtimer_init(timer, clock_id, mode);
421}
422EXPORT_SYMBOL_GPL(hrtimer_init_on_stack);
423
424void destroy_hrtimer_on_stack(struct hrtimer *timer)
425{
426 debug_object_free(timer, &hrtimer_debug_descr);
427}
428EXPORT_SYMBOL_GPL(destroy_hrtimer_on_stack);
429
430#else
431static inline void debug_hrtimer_init(struct hrtimer *timer) { }
432static inline void debug_hrtimer_activate(struct hrtimer *timer) { }
433static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
434#endif
435
436static inline void
437debug_init(struct hrtimer *timer, clockid_t clockid,
438 enum hrtimer_mode mode)
439{
440 debug_hrtimer_init(timer);
441 trace_hrtimer_init(timer, clockid, mode);
442}
443
444static inline void debug_activate(struct hrtimer *timer)
445{
446 debug_hrtimer_activate(timer);
447 trace_hrtimer_start(timer);
448}
449
450static inline void debug_deactivate(struct hrtimer *timer)
451{
452 debug_hrtimer_deactivate(timer);
453 trace_hrtimer_cancel(timer);
454}
455
456#if defined(CONFIG_NO_HZ_COMMON) || defined(CONFIG_HIGH_RES_TIMERS)
457static inline void hrtimer_update_next_timer(struct hrtimer_cpu_base *cpu_base,
458 struct hrtimer *timer)
459{
460#ifdef CONFIG_HIGH_RES_TIMERS
461 cpu_base->next_timer = timer;
462#endif
463}
464
465static ktime_t __hrtimer_get_next_event(struct hrtimer_cpu_base *cpu_base)
466{
467 struct hrtimer_clock_base *base = cpu_base->clock_base;
468 ktime_t expires, expires_next = { .tv64 = KTIME_MAX };
469 unsigned int active = cpu_base->active_bases;
470
471 hrtimer_update_next_timer(cpu_base, NULL);
472 for (; active; base++, active >>= 1) {
473 struct timerqueue_node *next;
474 struct hrtimer *timer;
475
476 if (!(active & 0x01))
477 continue;
478
479 next = timerqueue_getnext(&base->active);
480 timer = container_of(next, struct hrtimer, node);
481 expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
482 if (expires.tv64 < expires_next.tv64) {
483 expires_next = expires;
484 hrtimer_update_next_timer(cpu_base, timer);
485 }
486 }
487 /*
488 * clock_was_set() might have changed base->offset of any of
489 * the clock bases so the result might be negative. Fix it up
490 * to prevent a false positive in clockevents_program_event().
491 */
492 if (expires_next.tv64 < 0)
493 expires_next.tv64 = 0;
494 return expires_next;
495}
496#endif
497
498static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base)
499{
500 ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset;
501 ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset;
502 ktime_t *offs_tai = &base->clock_base[HRTIMER_BASE_TAI].offset;
503
504 return ktime_get_update_offsets_now(&base->clock_was_set_seq,
505 offs_real, offs_boot, offs_tai);
506}
507
508/* High resolution timer related functions */
509#ifdef CONFIG_HIGH_RES_TIMERS
510
511/*
512 * High resolution timer enabled ?
513 */
514static bool hrtimer_hres_enabled __read_mostly = true;
515unsigned int hrtimer_resolution __read_mostly = LOW_RES_NSEC;
516EXPORT_SYMBOL_GPL(hrtimer_resolution);
517
518/*
519 * Enable / Disable high resolution mode
520 */
521static int __init setup_hrtimer_hres(char *str)
522{
523 return (kstrtobool(str, &hrtimer_hres_enabled) == 0);
524}
525
526__setup("highres=", setup_hrtimer_hres);
527
528/*
529 * hrtimer_high_res_enabled - query, if the highres mode is enabled
530 */
531static inline int hrtimer_is_hres_enabled(void)
532{
533 return hrtimer_hres_enabled;
534}
535
536/*
537 * Is the high resolution mode active ?
538 */
539static inline int __hrtimer_hres_active(struct hrtimer_cpu_base *cpu_base)
540{
541 return cpu_base->hres_active;
542}
543
544static inline int hrtimer_hres_active(void)
545{
546 return __hrtimer_hres_active(this_cpu_ptr(&hrtimer_bases));
547}
548
549/*
550 * Reprogram the event source with checking both queues for the
551 * next event
552 * Called with interrupts disabled and base->lock held
553 */
554static void
555hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
556{
557 ktime_t expires_next;
558
559 if (!cpu_base->hres_active)
560 return;
561
562 expires_next = __hrtimer_get_next_event(cpu_base);
563
564 if (skip_equal && expires_next.tv64 == cpu_base->expires_next.tv64)
565 return;
566
567 cpu_base->expires_next.tv64 = expires_next.tv64;
568
569 /*
570 * If a hang was detected in the last timer interrupt then we
571 * leave the hang delay active in the hardware. We want the
572 * system to make progress. That also prevents the following
573 * scenario:
574 * T1 expires 50ms from now
575 * T2 expires 5s from now
576 *
577 * T1 is removed, so this code is called and would reprogram
578 * the hardware to 5s from now. Any hrtimer_start after that
579 * will not reprogram the hardware due to hang_detected being
580 * set. So we'd effectivly block all timers until the T2 event
581 * fires.
582 */
583 if (cpu_base->hang_detected)
584 return;
585
586 tick_program_event(cpu_base->expires_next, 1);
587}
588
589/*
590 * When a timer is enqueued and expires earlier than the already enqueued
591 * timers, we have to check, whether it expires earlier than the timer for
592 * which the clock event device was armed.
593 *
594 * Called with interrupts disabled and base->cpu_base.lock held
595 */
596static void hrtimer_reprogram(struct hrtimer *timer,
597 struct hrtimer_clock_base *base)
598{
599 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
600 ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
601
602 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
603
604 /*
605 * If the timer is not on the current cpu, we cannot reprogram
606 * the other cpus clock event device.
607 */
608 if (base->cpu_base != cpu_base)
609 return;
610
611 /*
612 * If the hrtimer interrupt is running, then it will
613 * reevaluate the clock bases and reprogram the clock event
614 * device. The callbacks are always executed in hard interrupt
615 * context so we don't need an extra check for a running
616 * callback.
617 */
618 if (cpu_base->in_hrtirq)
619 return;
620
621 /*
622 * CLOCK_REALTIME timer might be requested with an absolute
623 * expiry time which is less than base->offset. Set it to 0.
624 */
625 if (expires.tv64 < 0)
626 expires.tv64 = 0;
627
628 if (expires.tv64 >= cpu_base->expires_next.tv64)
629 return;
630
631 /* Update the pointer to the next expiring timer */
632 cpu_base->next_timer = timer;
633
634 /*
635 * If a hang was detected in the last timer interrupt then we
636 * do not schedule a timer which is earlier than the expiry
637 * which we enforced in the hang detection. We want the system
638 * to make progress.
639 */
640 if (cpu_base->hang_detected)
641 return;
642
643 /*
644 * Program the timer hardware. We enforce the expiry for
645 * events which are already in the past.
646 */
647 cpu_base->expires_next = expires;
648 tick_program_event(expires, 1);
649}
650
651/*
652 * Initialize the high resolution related parts of cpu_base
653 */
654static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
655{
656 base->expires_next.tv64 = KTIME_MAX;
657 base->hres_active = 0;
658}
659
660/*
661 * Retrigger next event is called after clock was set
662 *
663 * Called with interrupts disabled via on_each_cpu()
664 */
665static void retrigger_next_event(void *arg)
666{
667 struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
668
669 if (!base->hres_active)
670 return;
671
672 raw_spin_lock(&base->lock);
673 hrtimer_update_base(base);
674 hrtimer_force_reprogram(base, 0);
675 raw_spin_unlock(&base->lock);
676}
677
678/*
679 * Switch to high resolution mode
680 */
681static void hrtimer_switch_to_hres(void)
682{
683 struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
684
685 if (tick_init_highres()) {
686 printk(KERN_WARNING "Could not switch to high resolution "
687 "mode on CPU %d\n", base->cpu);
688 return;
689 }
690 base->hres_active = 1;
691 hrtimer_resolution = HIGH_RES_NSEC;
692
693 tick_setup_sched_timer();
694 /* "Retrigger" the interrupt to get things going */
695 retrigger_next_event(NULL);
696}
697
698static void clock_was_set_work(struct work_struct *work)
699{
700 clock_was_set();
701}
702
703static DECLARE_WORK(hrtimer_work, clock_was_set_work);
704
705/*
706 * Called from timekeeping and resume code to reprogramm the hrtimer
707 * interrupt device on all cpus.
708 */
709void clock_was_set_delayed(void)
710{
711 schedule_work(&hrtimer_work);
712}
713
714#else
715
716static inline int __hrtimer_hres_active(struct hrtimer_cpu_base *b) { return 0; }
717static inline int hrtimer_hres_active(void) { return 0; }
718static inline int hrtimer_is_hres_enabled(void) { return 0; }
719static inline void hrtimer_switch_to_hres(void) { }
720static inline void
721hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { }
722static inline int hrtimer_reprogram(struct hrtimer *timer,
723 struct hrtimer_clock_base *base)
724{
725 return 0;
726}
727static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { }
728static inline void retrigger_next_event(void *arg) { }
729
730#endif /* CONFIG_HIGH_RES_TIMERS */
731
732/*
733 * Clock realtime was set
734 *
735 * Change the offset of the realtime clock vs. the monotonic
736 * clock.
737 *
738 * We might have to reprogram the high resolution timer interrupt. On
739 * SMP we call the architecture specific code to retrigger _all_ high
740 * resolution timer interrupts. On UP we just disable interrupts and
741 * call the high resolution interrupt code.
742 */
743void clock_was_set(void)
744{
745#ifdef CONFIG_HIGH_RES_TIMERS
746 /* Retrigger the CPU local events everywhere */
747 on_each_cpu(retrigger_next_event, NULL, 1);
748#endif
749 timerfd_clock_was_set();
750}
751
752/*
753 * During resume we might have to reprogram the high resolution timer
754 * interrupt on all online CPUs. However, all other CPUs will be
755 * stopped with IRQs interrupts disabled so the clock_was_set() call
756 * must be deferred.
757 */
758void hrtimers_resume(void)
759{
760 WARN_ONCE(!irqs_disabled(),
761 KERN_INFO "hrtimers_resume() called with IRQs enabled!");
762
763 /* Retrigger on the local CPU */
764 retrigger_next_event(NULL);
765 /* And schedule a retrigger for all others */
766 clock_was_set_delayed();
767}
768
769static inline void timer_stats_hrtimer_set_start_info(struct hrtimer *timer)
770{
771#ifdef CONFIG_TIMER_STATS
772 if (timer->start_site)
773 return;
774 timer->start_site = __builtin_return_address(0);
775 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
776 timer->start_pid = current->pid;
777#endif
778}
779
780static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer *timer)
781{
782#ifdef CONFIG_TIMER_STATS
783 timer->start_site = NULL;
784#endif
785}
786
787static inline void timer_stats_account_hrtimer(struct hrtimer *timer)
788{
789#ifdef CONFIG_TIMER_STATS
790 if (likely(!timer_stats_active))
791 return;
792 timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
793 timer->function, timer->start_comm, 0);
794#endif
795}
796
797/*
798 * Counterpart to lock_hrtimer_base above:
799 */
800static inline
801void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
802{
803 raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
804}
805
806/**
807 * hrtimer_forward - forward the timer expiry
808 * @timer: hrtimer to forward
809 * @now: forward past this time
810 * @interval: the interval to forward
811 *
812 * Forward the timer expiry so it will expire in the future.
813 * Returns the number of overruns.
814 *
815 * Can be safely called from the callback function of @timer. If
816 * called from other contexts @timer must neither be enqueued nor
817 * running the callback and the caller needs to take care of
818 * serialization.
819 *
820 * Note: This only updates the timer expiry value and does not requeue
821 * the timer.
822 */
823u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
824{
825 u64 orun = 1;
826 ktime_t delta;
827
828 delta = ktime_sub(now, hrtimer_get_expires(timer));
829
830 if (delta.tv64 < 0)
831 return 0;
832
833 if (WARN_ON(timer->state & HRTIMER_STATE_ENQUEUED))
834 return 0;
835
836 if (interval.tv64 < hrtimer_resolution)
837 interval.tv64 = hrtimer_resolution;
838
839 if (unlikely(delta.tv64 >= interval.tv64)) {
840 s64 incr = ktime_to_ns(interval);
841
842 orun = ktime_divns(delta, incr);
843 hrtimer_add_expires_ns(timer, incr * orun);
844 if (hrtimer_get_expires_tv64(timer) > now.tv64)
845 return orun;
846 /*
847 * This (and the ktime_add() below) is the
848 * correction for exact:
849 */
850 orun++;
851 }
852 hrtimer_add_expires(timer, interval);
853
854 return orun;
855}
856EXPORT_SYMBOL_GPL(hrtimer_forward);
857
858/*
859 * enqueue_hrtimer - internal function to (re)start a timer
860 *
861 * The timer is inserted in expiry order. Insertion into the
862 * red black tree is O(log(n)). Must hold the base lock.
863 *
864 * Returns 1 when the new timer is the leftmost timer in the tree.
865 */
866static int enqueue_hrtimer(struct hrtimer *timer,
867 struct hrtimer_clock_base *base)
868{
869 debug_activate(timer);
870
871 base->cpu_base->active_bases |= 1 << base->index;
872
873 timer->state = HRTIMER_STATE_ENQUEUED;
874
875 return timerqueue_add(&base->active, &timer->node);
876}
877
878/*
879 * __remove_hrtimer - internal function to remove a timer
880 *
881 * Caller must hold the base lock.
882 *
883 * High resolution timer mode reprograms the clock event device when the
884 * timer is the one which expires next. The caller can disable this by setting
885 * reprogram to zero. This is useful, when the context does a reprogramming
886 * anyway (e.g. timer interrupt)
887 */
888static void __remove_hrtimer(struct hrtimer *timer,
889 struct hrtimer_clock_base *base,
890 u8 newstate, int reprogram)
891{
892 struct hrtimer_cpu_base *cpu_base = base->cpu_base;
893 u8 state = timer->state;
894
895 timer->state = newstate;
896 if (!(state & HRTIMER_STATE_ENQUEUED))
897 return;
898
899 if (!timerqueue_del(&base->active, &timer->node))
900 cpu_base->active_bases &= ~(1 << base->index);
901
902#ifdef CONFIG_HIGH_RES_TIMERS
903 /*
904 * Note: If reprogram is false we do not update
905 * cpu_base->next_timer. This happens when we remove the first
906 * timer on a remote cpu. No harm as we never dereference
907 * cpu_base->next_timer. So the worst thing what can happen is
908 * an superflous call to hrtimer_force_reprogram() on the
909 * remote cpu later on if the same timer gets enqueued again.
910 */
911 if (reprogram && timer == cpu_base->next_timer)
912 hrtimer_force_reprogram(cpu_base, 1);
913#endif
914}
915
916/*
917 * remove hrtimer, called with base lock held
918 */
919static inline int
920remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base, bool restart)
921{
922 if (hrtimer_is_queued(timer)) {
923 u8 state = timer->state;
924 int reprogram;
925
926 /*
927 * Remove the timer and force reprogramming when high
928 * resolution mode is active and the timer is on the current
929 * CPU. If we remove a timer on another CPU, reprogramming is
930 * skipped. The interrupt event on this CPU is fired and
931 * reprogramming happens in the interrupt handler. This is a
932 * rare case and less expensive than a smp call.
933 */
934 debug_deactivate(timer);
935 timer_stats_hrtimer_clear_start_info(timer);
936 reprogram = base->cpu_base == this_cpu_ptr(&hrtimer_bases);
937
938 if (!restart)
939 state = HRTIMER_STATE_INACTIVE;
940
941 __remove_hrtimer(timer, base, state, reprogram);
942 return 1;
943 }
944 return 0;
945}
946
947static inline ktime_t hrtimer_update_lowres(struct hrtimer *timer, ktime_t tim,
948 const enum hrtimer_mode mode)
949{
950#ifdef CONFIG_TIME_LOW_RES
951 /*
952 * CONFIG_TIME_LOW_RES indicates that the system has no way to return
953 * granular time values. For relative timers we add hrtimer_resolution
954 * (i.e. one jiffie) to prevent short timeouts.
955 */
956 timer->is_rel = mode & HRTIMER_MODE_REL;
957 if (timer->is_rel)
958 tim = ktime_add_safe(tim, ktime_set(0, hrtimer_resolution));
959#endif
960 return tim;
961}
962
963/**
964 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
965 * @timer: the timer to be added
966 * @tim: expiry time
967 * @delta_ns: "slack" range for the timer
968 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
969 * relative (HRTIMER_MODE_REL)
970 */
971void hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
972 u64 delta_ns, const enum hrtimer_mode mode)
973{
974 struct hrtimer_clock_base *base, *new_base;
975 unsigned long flags;
976 int leftmost;
977
978 base = lock_hrtimer_base(timer, &flags);
979
980 /* Remove an active timer from the queue: */
981 remove_hrtimer(timer, base, true);
982
983 if (mode & HRTIMER_MODE_REL)
984 tim = ktime_add_safe(tim, base->get_time());
985
986 tim = hrtimer_update_lowres(timer, tim, mode);
987
988 hrtimer_set_expires_range_ns(timer, tim, delta_ns);
989
990 /* Switch the timer base, if necessary: */
991 new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED);
992
993 timer_stats_hrtimer_set_start_info(timer);
994
995 leftmost = enqueue_hrtimer(timer, new_base);
996 if (!leftmost)
997 goto unlock;
998
999 if (!hrtimer_is_hres_active(timer)) {
1000 /*
1001 * Kick to reschedule the next tick to handle the new timer
1002 * on dynticks target.
1003 */
1004 if (new_base->cpu_base->nohz_active)
1005 wake_up_nohz_cpu(new_base->cpu_base->cpu);
1006 } else {
1007 hrtimer_reprogram(timer, new_base);
1008 }
1009unlock:
1010 unlock_hrtimer_base(timer, &flags);
1011}
1012EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
1013
1014/**
1015 * hrtimer_try_to_cancel - try to deactivate a timer
1016 * @timer: hrtimer to stop
1017 *
1018 * Returns:
1019 * 0 when the timer was not active
1020 * 1 when the timer was active
1021 * -1 when the timer is currently excuting the callback function and
1022 * cannot be stopped
1023 */
1024int hrtimer_try_to_cancel(struct hrtimer *timer)
1025{
1026 struct hrtimer_clock_base *base;
1027 unsigned long flags;
1028 int ret = -1;
1029
1030 /*
1031 * Check lockless first. If the timer is not active (neither
1032 * enqueued nor running the callback, nothing to do here. The
1033 * base lock does not serialize against a concurrent enqueue,
1034 * so we can avoid taking it.
1035 */
1036 if (!hrtimer_active(timer))
1037 return 0;
1038
1039 base = lock_hrtimer_base(timer, &flags);
1040
1041 if (!hrtimer_callback_running(timer))
1042 ret = remove_hrtimer(timer, base, false);
1043
1044 unlock_hrtimer_base(timer, &flags);
1045
1046 return ret;
1047
1048}
1049EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1050
1051/**
1052 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1053 * @timer: the timer to be cancelled
1054 *
1055 * Returns:
1056 * 0 when the timer was not active
1057 * 1 when the timer was active
1058 */
1059int hrtimer_cancel(struct hrtimer *timer)
1060{
1061 for (;;) {
1062 int ret = hrtimer_try_to_cancel(timer);
1063
1064 if (ret >= 0)
1065 return ret;
1066 cpu_relax();
1067 }
1068}
1069EXPORT_SYMBOL_GPL(hrtimer_cancel);
1070
1071/**
1072 * hrtimer_get_remaining - get remaining time for the timer
1073 * @timer: the timer to read
1074 * @adjust: adjust relative timers when CONFIG_TIME_LOW_RES=y
1075 */
1076ktime_t __hrtimer_get_remaining(const struct hrtimer *timer, bool adjust)
1077{
1078 unsigned long flags;
1079 ktime_t rem;
1080
1081 lock_hrtimer_base(timer, &flags);
1082 if (IS_ENABLED(CONFIG_TIME_LOW_RES) && adjust)
1083 rem = hrtimer_expires_remaining_adjusted(timer);
1084 else
1085 rem = hrtimer_expires_remaining(timer);
1086 unlock_hrtimer_base(timer, &flags);
1087
1088 return rem;
1089}
1090EXPORT_SYMBOL_GPL(__hrtimer_get_remaining);
1091
1092#ifdef CONFIG_NO_HZ_COMMON
1093/**
1094 * hrtimer_get_next_event - get the time until next expiry event
1095 *
1096 * Returns the next expiry time or KTIME_MAX if no timer is pending.
1097 */
1098u64 hrtimer_get_next_event(void)
1099{
1100 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1101 u64 expires = KTIME_MAX;
1102 unsigned long flags;
1103
1104 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1105
1106 if (!__hrtimer_hres_active(cpu_base))
1107 expires = __hrtimer_get_next_event(cpu_base).tv64;
1108
1109 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1110
1111 return expires;
1112}
1113#endif
1114
1115static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1116 enum hrtimer_mode mode)
1117{
1118 struct hrtimer_cpu_base *cpu_base;
1119 int base;
1120
1121 memset(timer, 0, sizeof(struct hrtimer));
1122
1123 cpu_base = raw_cpu_ptr(&hrtimer_bases);
1124
1125 if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
1126 clock_id = CLOCK_MONOTONIC;
1127
1128 base = hrtimer_clockid_to_base(clock_id);
1129 timer->base = &cpu_base->clock_base[base];
1130 timerqueue_init(&timer->node);
1131
1132#ifdef CONFIG_TIMER_STATS
1133 timer->start_site = NULL;
1134 timer->start_pid = -1;
1135 memset(timer->start_comm, 0, TASK_COMM_LEN);
1136#endif
1137}
1138
1139/**
1140 * hrtimer_init - initialize a timer to the given clock
1141 * @timer: the timer to be initialized
1142 * @clock_id: the clock to be used
1143 * @mode: timer mode abs/rel
1144 */
1145void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1146 enum hrtimer_mode mode)
1147{
1148 debug_init(timer, clock_id, mode);
1149 __hrtimer_init(timer, clock_id, mode);
1150}
1151EXPORT_SYMBOL_GPL(hrtimer_init);
1152
1153/*
1154 * A timer is active, when it is enqueued into the rbtree or the
1155 * callback function is running or it's in the state of being migrated
1156 * to another cpu.
1157 *
1158 * It is important for this function to not return a false negative.
1159 */
1160bool hrtimer_active(const struct hrtimer *timer)
1161{
1162 struct hrtimer_cpu_base *cpu_base;
1163 unsigned int seq;
1164
1165 do {
1166 cpu_base = READ_ONCE(timer->base->cpu_base);
1167 seq = raw_read_seqcount_begin(&cpu_base->seq);
1168
1169 if (timer->state != HRTIMER_STATE_INACTIVE ||
1170 cpu_base->running == timer)
1171 return true;
1172
1173 } while (read_seqcount_retry(&cpu_base->seq, seq) ||
1174 cpu_base != READ_ONCE(timer->base->cpu_base));
1175
1176 return false;
1177}
1178EXPORT_SYMBOL_GPL(hrtimer_active);
1179
1180/*
1181 * The write_seqcount_barrier()s in __run_hrtimer() split the thing into 3
1182 * distinct sections:
1183 *
1184 * - queued: the timer is queued
1185 * - callback: the timer is being ran
1186 * - post: the timer is inactive or (re)queued
1187 *
1188 * On the read side we ensure we observe timer->state and cpu_base->running
1189 * from the same section, if anything changed while we looked at it, we retry.
1190 * This includes timer->base changing because sequence numbers alone are
1191 * insufficient for that.
1192 *
1193 * The sequence numbers are required because otherwise we could still observe
1194 * a false negative if the read side got smeared over multiple consequtive
1195 * __run_hrtimer() invocations.
1196 */
1197
1198static void __run_hrtimer(struct hrtimer_cpu_base *cpu_base,
1199 struct hrtimer_clock_base *base,
1200 struct hrtimer *timer, ktime_t *now)
1201{
1202 enum hrtimer_restart (*fn)(struct hrtimer *);
1203 int restart;
1204
1205 lockdep_assert_held(&cpu_base->lock);
1206
1207 debug_deactivate(timer);
1208 cpu_base->running = timer;
1209
1210 /*
1211 * Separate the ->running assignment from the ->state assignment.
1212 *
1213 * As with a regular write barrier, this ensures the read side in
1214 * hrtimer_active() cannot observe cpu_base->running == NULL &&
1215 * timer->state == INACTIVE.
1216 */
1217 raw_write_seqcount_barrier(&cpu_base->seq);
1218
1219 __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE, 0);
1220 timer_stats_account_hrtimer(timer);
1221 fn = timer->function;
1222
1223 /*
1224 * Clear the 'is relative' flag for the TIME_LOW_RES case. If the
1225 * timer is restarted with a period then it becomes an absolute
1226 * timer. If its not restarted it does not matter.
1227 */
1228 if (IS_ENABLED(CONFIG_TIME_LOW_RES))
1229 timer->is_rel = false;
1230
1231 /*
1232 * Because we run timers from hardirq context, there is no chance
1233 * they get migrated to another cpu, therefore its safe to unlock
1234 * the timer base.
1235 */
1236 raw_spin_unlock(&cpu_base->lock);
1237 trace_hrtimer_expire_entry(timer, now);
1238 restart = fn(timer);
1239 trace_hrtimer_expire_exit(timer);
1240 raw_spin_lock(&cpu_base->lock);
1241
1242 /*
1243 * Note: We clear the running state after enqueue_hrtimer and
1244 * we do not reprogramm the event hardware. Happens either in
1245 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1246 *
1247 * Note: Because we dropped the cpu_base->lock above,
1248 * hrtimer_start_range_ns() can have popped in and enqueued the timer
1249 * for us already.
1250 */
1251 if (restart != HRTIMER_NORESTART &&
1252 !(timer->state & HRTIMER_STATE_ENQUEUED))
1253 enqueue_hrtimer(timer, base);
1254
1255 /*
1256 * Separate the ->running assignment from the ->state assignment.
1257 *
1258 * As with a regular write barrier, this ensures the read side in
1259 * hrtimer_active() cannot observe cpu_base->running == NULL &&
1260 * timer->state == INACTIVE.
1261 */
1262 raw_write_seqcount_barrier(&cpu_base->seq);
1263
1264 WARN_ON_ONCE(cpu_base->running != timer);
1265 cpu_base->running = NULL;
1266}
1267
1268static void __hrtimer_run_queues(struct hrtimer_cpu_base *cpu_base, ktime_t now)
1269{
1270 struct hrtimer_clock_base *base = cpu_base->clock_base;
1271 unsigned int active = cpu_base->active_bases;
1272
1273 for (; active; base++, active >>= 1) {
1274 struct timerqueue_node *node;
1275 ktime_t basenow;
1276
1277 if (!(active & 0x01))
1278 continue;
1279
1280 basenow = ktime_add(now, base->offset);
1281
1282 while ((node = timerqueue_getnext(&base->active))) {
1283 struct hrtimer *timer;
1284
1285 timer = container_of(node, struct hrtimer, node);
1286
1287 /*
1288 * The immediate goal for using the softexpires is
1289 * minimizing wakeups, not running timers at the
1290 * earliest interrupt after their soft expiration.
1291 * This allows us to avoid using a Priority Search
1292 * Tree, which can answer a stabbing querry for
1293 * overlapping intervals and instead use the simple
1294 * BST we already have.
1295 * We don't add extra wakeups by delaying timers that
1296 * are right-of a not yet expired timer, because that
1297 * timer will have to trigger a wakeup anyway.
1298 */
1299 if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer))
1300 break;
1301
1302 __run_hrtimer(cpu_base, base, timer, &basenow);
1303 }
1304 }
1305}
1306
1307#ifdef CONFIG_HIGH_RES_TIMERS
1308
1309/*
1310 * High resolution timer interrupt
1311 * Called with interrupts disabled
1312 */
1313void hrtimer_interrupt(struct clock_event_device *dev)
1314{
1315 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1316 ktime_t expires_next, now, entry_time, delta;
1317 int retries = 0;
1318
1319 BUG_ON(!cpu_base->hres_active);
1320 cpu_base->nr_events++;
1321 dev->next_event.tv64 = KTIME_MAX;
1322
1323 raw_spin_lock(&cpu_base->lock);
1324 entry_time = now = hrtimer_update_base(cpu_base);
1325retry:
1326 cpu_base->in_hrtirq = 1;
1327 /*
1328 * We set expires_next to KTIME_MAX here with cpu_base->lock
1329 * held to prevent that a timer is enqueued in our queue via
1330 * the migration code. This does not affect enqueueing of
1331 * timers which run their callback and need to be requeued on
1332 * this CPU.
1333 */
1334 cpu_base->expires_next.tv64 = KTIME_MAX;
1335
1336 __hrtimer_run_queues(cpu_base, now);
1337
1338 /* Reevaluate the clock bases for the next expiry */
1339 expires_next = __hrtimer_get_next_event(cpu_base);
1340 /*
1341 * Store the new expiry value so the migration code can verify
1342 * against it.
1343 */
1344 cpu_base->expires_next = expires_next;
1345 cpu_base->in_hrtirq = 0;
1346 raw_spin_unlock(&cpu_base->lock);
1347
1348 /* Reprogramming necessary ? */
1349 if (!tick_program_event(expires_next, 0)) {
1350 cpu_base->hang_detected = 0;
1351 return;
1352 }
1353
1354 /*
1355 * The next timer was already expired due to:
1356 * - tracing
1357 * - long lasting callbacks
1358 * - being scheduled away when running in a VM
1359 *
1360 * We need to prevent that we loop forever in the hrtimer
1361 * interrupt routine. We give it 3 attempts to avoid
1362 * overreacting on some spurious event.
1363 *
1364 * Acquire base lock for updating the offsets and retrieving
1365 * the current time.
1366 */
1367 raw_spin_lock(&cpu_base->lock);
1368 now = hrtimer_update_base(cpu_base);
1369 cpu_base->nr_retries++;
1370 if (++retries < 3)
1371 goto retry;
1372 /*
1373 * Give the system a chance to do something else than looping
1374 * here. We stored the entry time, so we know exactly how long
1375 * we spent here. We schedule the next event this amount of
1376 * time away.
1377 */
1378 cpu_base->nr_hangs++;
1379 cpu_base->hang_detected = 1;
1380 raw_spin_unlock(&cpu_base->lock);
1381 delta = ktime_sub(now, entry_time);
1382 if ((unsigned int)delta.tv64 > cpu_base->max_hang_time)
1383 cpu_base->max_hang_time = (unsigned int) delta.tv64;
1384 /*
1385 * Limit it to a sensible value as we enforce a longer
1386 * delay. Give the CPU at least 100ms to catch up.
1387 */
1388 if (delta.tv64 > 100 * NSEC_PER_MSEC)
1389 expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1390 else
1391 expires_next = ktime_add(now, delta);
1392 tick_program_event(expires_next, 1);
1393 printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n",
1394 ktime_to_ns(delta));
1395}
1396
1397/*
1398 * local version of hrtimer_peek_ahead_timers() called with interrupts
1399 * disabled.
1400 */
1401static inline void __hrtimer_peek_ahead_timers(void)
1402{
1403 struct tick_device *td;
1404
1405 if (!hrtimer_hres_active())
1406 return;
1407
1408 td = this_cpu_ptr(&tick_cpu_device);
1409 if (td && td->evtdev)
1410 hrtimer_interrupt(td->evtdev);
1411}
1412
1413#else /* CONFIG_HIGH_RES_TIMERS */
1414
1415static inline void __hrtimer_peek_ahead_timers(void) { }
1416
1417#endif /* !CONFIG_HIGH_RES_TIMERS */
1418
1419/*
1420 * Called from run_local_timers in hardirq context every jiffy
1421 */
1422void hrtimer_run_queues(void)
1423{
1424 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1425 ktime_t now;
1426
1427 if (__hrtimer_hres_active(cpu_base))
1428 return;
1429
1430 /*
1431 * This _is_ ugly: We have to check periodically, whether we
1432 * can switch to highres and / or nohz mode. The clocksource
1433 * switch happens with xtime_lock held. Notification from
1434 * there only sets the check bit in the tick_oneshot code,
1435 * otherwise we might deadlock vs. xtime_lock.
1436 */
1437 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) {
1438 hrtimer_switch_to_hres();
1439 return;
1440 }
1441
1442 raw_spin_lock(&cpu_base->lock);
1443 now = hrtimer_update_base(cpu_base);
1444 __hrtimer_run_queues(cpu_base, now);
1445 raw_spin_unlock(&cpu_base->lock);
1446}
1447
1448/*
1449 * Sleep related functions:
1450 */
1451static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1452{
1453 struct hrtimer_sleeper *t =
1454 container_of(timer, struct hrtimer_sleeper, timer);
1455 struct task_struct *task = t->task;
1456
1457 t->task = NULL;
1458 if (task)
1459 wake_up_process(task);
1460
1461 return HRTIMER_NORESTART;
1462}
1463
1464void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1465{
1466 sl->timer.function = hrtimer_wakeup;
1467 sl->task = task;
1468}
1469EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
1470
1471static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1472{
1473 hrtimer_init_sleeper(t, current);
1474
1475 do {
1476 set_current_state(TASK_INTERRUPTIBLE);
1477 hrtimer_start_expires(&t->timer, mode);
1478
1479 if (likely(t->task))
1480 freezable_schedule();
1481
1482 hrtimer_cancel(&t->timer);
1483 mode = HRTIMER_MODE_ABS;
1484
1485 } while (t->task && !signal_pending(current));
1486
1487 __set_current_state(TASK_RUNNING);
1488
1489 return t->task == NULL;
1490}
1491
1492static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp)
1493{
1494 struct timespec rmt;
1495 ktime_t rem;
1496
1497 rem = hrtimer_expires_remaining(timer);
1498 if (rem.tv64 <= 0)
1499 return 0;
1500 rmt = ktime_to_timespec(rem);
1501
1502 if (copy_to_user(rmtp, &rmt, sizeof(*rmtp)))
1503 return -EFAULT;
1504
1505 return 1;
1506}
1507
1508long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1509{
1510 struct hrtimer_sleeper t;
1511 struct timespec __user *rmtp;
1512 int ret = 0;
1513
1514 hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid,
1515 HRTIMER_MODE_ABS);
1516 hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
1517
1518 if (do_nanosleep(&t, HRTIMER_MODE_ABS))
1519 goto out;
1520
1521 rmtp = restart->nanosleep.rmtp;
1522 if (rmtp) {
1523 ret = update_rmtp(&t.timer, rmtp);
1524 if (ret <= 0)
1525 goto out;
1526 }
1527
1528 /* The other values in restart are already filled in */
1529 ret = -ERESTART_RESTARTBLOCK;
1530out:
1531 destroy_hrtimer_on_stack(&t.timer);
1532 return ret;
1533}
1534
1535long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
1536 const enum hrtimer_mode mode, const clockid_t clockid)
1537{
1538 struct restart_block *restart;
1539 struct hrtimer_sleeper t;
1540 int ret = 0;
1541 u64 slack;
1542
1543 slack = current->timer_slack_ns;
1544 if (dl_task(current) || rt_task(current))
1545 slack = 0;
1546
1547 hrtimer_init_on_stack(&t.timer, clockid, mode);
1548 hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack);
1549 if (do_nanosleep(&t, mode))
1550 goto out;
1551
1552 /* Absolute timers do not update the rmtp value and restart: */
1553 if (mode == HRTIMER_MODE_ABS) {
1554 ret = -ERESTARTNOHAND;
1555 goto out;
1556 }
1557
1558 if (rmtp) {
1559 ret = update_rmtp(&t.timer, rmtp);
1560 if (ret <= 0)
1561 goto out;
1562 }
1563
1564 restart = &current->restart_block;
1565 restart->fn = hrtimer_nanosleep_restart;
1566 restart->nanosleep.clockid = t.timer.base->clockid;
1567 restart->nanosleep.rmtp = rmtp;
1568 restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
1569
1570 ret = -ERESTART_RESTARTBLOCK;
1571out:
1572 destroy_hrtimer_on_stack(&t.timer);
1573 return ret;
1574}
1575
1576SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp,
1577 struct timespec __user *, rmtp)
1578{
1579 struct timespec tu;
1580
1581 if (copy_from_user(&tu, rqtp, sizeof(tu)))
1582 return -EFAULT;
1583
1584 if (!timespec_valid(&tu))
1585 return -EINVAL;
1586
1587 return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1588}
1589
1590/*
1591 * Functions related to boot-time initialization:
1592 */
1593static void init_hrtimers_cpu(int cpu)
1594{
1595 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1596 int i;
1597
1598 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1599 cpu_base->clock_base[i].cpu_base = cpu_base;
1600 timerqueue_init_head(&cpu_base->clock_base[i].active);
1601 }
1602
1603 cpu_base->cpu = cpu;
1604 hrtimer_init_hres(cpu_base);
1605}
1606
1607#ifdef CONFIG_HOTPLUG_CPU
1608
1609static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1610 struct hrtimer_clock_base *new_base)
1611{
1612 struct hrtimer *timer;
1613 struct timerqueue_node *node;
1614
1615 while ((node = timerqueue_getnext(&old_base->active))) {
1616 timer = container_of(node, struct hrtimer, node);
1617 BUG_ON(hrtimer_callback_running(timer));
1618 debug_deactivate(timer);
1619
1620 /*
1621 * Mark it as ENQUEUED not INACTIVE otherwise the
1622 * timer could be seen as !active and just vanish away
1623 * under us on another CPU
1624 */
1625 __remove_hrtimer(timer, old_base, HRTIMER_STATE_ENQUEUED, 0);
1626 timer->base = new_base;
1627 /*
1628 * Enqueue the timers on the new cpu. This does not
1629 * reprogram the event device in case the timer
1630 * expires before the earliest on this CPU, but we run
1631 * hrtimer_interrupt after we migrated everything to
1632 * sort out already expired timers and reprogram the
1633 * event device.
1634 */
1635 enqueue_hrtimer(timer, new_base);
1636 }
1637}
1638
1639static void migrate_hrtimers(int scpu)
1640{
1641 struct hrtimer_cpu_base *old_base, *new_base;
1642 int i;
1643
1644 BUG_ON(cpu_online(scpu));
1645 tick_cancel_sched_timer(scpu);
1646
1647 local_irq_disable();
1648 old_base = &per_cpu(hrtimer_bases, scpu);
1649 new_base = this_cpu_ptr(&hrtimer_bases);
1650 /*
1651 * The caller is globally serialized and nobody else
1652 * takes two locks at once, deadlock is not possible.
1653 */
1654 raw_spin_lock(&new_base->lock);
1655 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1656
1657 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1658 migrate_hrtimer_list(&old_base->clock_base[i],
1659 &new_base->clock_base[i]);
1660 }
1661
1662 raw_spin_unlock(&old_base->lock);
1663 raw_spin_unlock(&new_base->lock);
1664
1665 /* Check, if we got expired work to do */
1666 __hrtimer_peek_ahead_timers();
1667 local_irq_enable();
1668}
1669
1670#endif /* CONFIG_HOTPLUG_CPU */
1671
1672static int hrtimer_cpu_notify(struct notifier_block *self,
1673 unsigned long action, void *hcpu)
1674{
1675 int scpu = (long)hcpu;
1676
1677 switch (action) {
1678
1679 case CPU_UP_PREPARE:
1680 case CPU_UP_PREPARE_FROZEN:
1681 init_hrtimers_cpu(scpu);
1682 break;
1683
1684#ifdef CONFIG_HOTPLUG_CPU
1685 case CPU_DEAD:
1686 case CPU_DEAD_FROZEN:
1687 migrate_hrtimers(scpu);
1688 break;
1689#endif
1690
1691 default:
1692 break;
1693 }
1694
1695 return NOTIFY_OK;
1696}
1697
1698static struct notifier_block hrtimers_nb = {
1699 .notifier_call = hrtimer_cpu_notify,
1700};
1701
1702void __init hrtimers_init(void)
1703{
1704 hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
1705 (void *)(long)smp_processor_id());
1706 register_cpu_notifier(&hrtimers_nb);
1707}
1708
1709/**
1710 * schedule_hrtimeout_range_clock - sleep until timeout
1711 * @expires: timeout value (ktime_t)
1712 * @delta: slack in expires timeout (ktime_t)
1713 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1714 * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME
1715 */
1716int __sched
1717schedule_hrtimeout_range_clock(ktime_t *expires, u64 delta,
1718 const enum hrtimer_mode mode, int clock)
1719{
1720 struct hrtimer_sleeper t;
1721
1722 /*
1723 * Optimize when a zero timeout value is given. It does not
1724 * matter whether this is an absolute or a relative time.
1725 */
1726 if (expires && !expires->tv64) {
1727 __set_current_state(TASK_RUNNING);
1728 return 0;
1729 }
1730
1731 /*
1732 * A NULL parameter means "infinite"
1733 */
1734 if (!expires) {
1735 schedule();
1736 return -EINTR;
1737 }
1738
1739 hrtimer_init_on_stack(&t.timer, clock, mode);
1740 hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
1741
1742 hrtimer_init_sleeper(&t, current);
1743
1744 hrtimer_start_expires(&t.timer, mode);
1745
1746 if (likely(t.task))
1747 schedule();
1748
1749 hrtimer_cancel(&t.timer);
1750 destroy_hrtimer_on_stack(&t.timer);
1751
1752 __set_current_state(TASK_RUNNING);
1753
1754 return !t.task ? 0 : -EINTR;
1755}
1756
1757/**
1758 * schedule_hrtimeout_range - sleep until timeout
1759 * @expires: timeout value (ktime_t)
1760 * @delta: slack in expires timeout (ktime_t)
1761 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1762 *
1763 * Make the current task sleep until the given expiry time has
1764 * elapsed. The routine will return immediately unless
1765 * the current task state has been set (see set_current_state()).
1766 *
1767 * The @delta argument gives the kernel the freedom to schedule the
1768 * actual wakeup to a time that is both power and performance friendly.
1769 * The kernel give the normal best effort behavior for "@expires+@delta",
1770 * but may decide to fire the timer earlier, but no earlier than @expires.
1771 *
1772 * You can set the task state as follows -
1773 *
1774 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1775 * pass before the routine returns.
1776 *
1777 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1778 * delivered to the current task.
1779 *
1780 * The current task state is guaranteed to be TASK_RUNNING when this
1781 * routine returns.
1782 *
1783 * Returns 0 when the timer has expired otherwise -EINTR
1784 */
1785int __sched schedule_hrtimeout_range(ktime_t *expires, u64 delta,
1786 const enum hrtimer_mode mode)
1787{
1788 return schedule_hrtimeout_range_clock(expires, delta, mode,
1789 CLOCK_MONOTONIC);
1790}
1791EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
1792
1793/**
1794 * schedule_hrtimeout - sleep until timeout
1795 * @expires: timeout value (ktime_t)
1796 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1797 *
1798 * Make the current task sleep until the given expiry time has
1799 * elapsed. The routine will return immediately unless
1800 * the current task state has been set (see set_current_state()).
1801 *
1802 * You can set the task state as follows -
1803 *
1804 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1805 * pass before the routine returns.
1806 *
1807 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1808 * delivered to the current task.
1809 *
1810 * The current task state is guaranteed to be TASK_RUNNING when this
1811 * routine returns.
1812 *
1813 * Returns 0 when the timer has expired otherwise -EINTR
1814 */
1815int __sched schedule_hrtimeout(ktime_t *expires,
1816 const enum hrtimer_mode mode)
1817{
1818 return schedule_hrtimeout_range(expires, 0, mode);
1819}
1820EXPORT_SYMBOL_GPL(schedule_hrtimeout);