Loading...
Note: File does not exist in v3.15.
1/*
2 * linux/kernel/hrtimer.c
3 *
4 * Copyright(C) 2005-2006, Thomas Gleixner <tglx@linutronix.de>
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 bool hrtimer_hres_enabled __read_mostly = true;
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 return (kstrtobool(str, &hrtimer_hres_enabled) == 0);
528}
529
530__setup("highres=", setup_hrtimer_hres);
531
532/*
533 * hrtimer_high_res_enabled - query, if the highres mode is enabled
534 */
535static inline int hrtimer_is_hres_enabled(void)
536{
537 return hrtimer_hres_enabled;
538}
539
540/*
541 * Is the high resolution mode active ?
542 */
543static inline int __hrtimer_hres_active(struct hrtimer_cpu_base *cpu_base)
544{
545 return cpu_base->hres_active;
546}
547
548static inline int hrtimer_hres_active(void)
549{
550 return __hrtimer_hres_active(this_cpu_ptr(&hrtimer_bases));
551}
552
553/*
554 * Reprogram the event source with checking both queues for the
555 * next event
556 * Called with interrupts disabled and base->lock held
557 */
558static void
559hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
560{
561 ktime_t expires_next;
562
563 if (!cpu_base->hres_active)
564 return;
565
566 expires_next = __hrtimer_get_next_event(cpu_base);
567
568 if (skip_equal && expires_next.tv64 == cpu_base->expires_next.tv64)
569 return;
570
571 cpu_base->expires_next.tv64 = expires_next.tv64;
572
573 /*
574 * If a hang was detected in the last timer interrupt then we
575 * leave the hang delay active in the hardware. We want the
576 * system to make progress. That also prevents the following
577 * scenario:
578 * T1 expires 50ms from now
579 * T2 expires 5s from now
580 *
581 * T1 is removed, so this code is called and would reprogram
582 * the hardware to 5s from now. Any hrtimer_start after that
583 * will not reprogram the hardware due to hang_detected being
584 * set. So we'd effectivly block all timers until the T2 event
585 * fires.
586 */
587 if (cpu_base->hang_detected)
588 return;
589
590 tick_program_event(cpu_base->expires_next, 1);
591}
592
593/*
594 * When a timer is enqueued and expires earlier than the already enqueued
595 * timers, we have to check, whether it expires earlier than the timer for
596 * which the clock event device was armed.
597 *
598 * Called with interrupts disabled and base->cpu_base.lock held
599 */
600static void hrtimer_reprogram(struct hrtimer *timer,
601 struct hrtimer_clock_base *base)
602{
603 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
604 ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
605
606 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
607
608 /*
609 * If the timer is not on the current cpu, we cannot reprogram
610 * the other cpus clock event device.
611 */
612 if (base->cpu_base != cpu_base)
613 return;
614
615 /*
616 * If the hrtimer interrupt is running, then it will
617 * reevaluate the clock bases and reprogram the clock event
618 * device. The callbacks are always executed in hard interrupt
619 * context so we don't need an extra check for a running
620 * callback.
621 */
622 if (cpu_base->in_hrtirq)
623 return;
624
625 /*
626 * CLOCK_REALTIME timer might be requested with an absolute
627 * expiry time which is less than base->offset. Set it to 0.
628 */
629 if (expires.tv64 < 0)
630 expires.tv64 = 0;
631
632 if (expires.tv64 >= cpu_base->expires_next.tv64)
633 return;
634
635 /* Update the pointer to the next expiring timer */
636 cpu_base->next_timer = timer;
637
638 /*
639 * If a hang was detected in the last timer interrupt then we
640 * do not schedule a timer which is earlier than the expiry
641 * which we enforced in the hang detection. We want the system
642 * to make progress.
643 */
644 if (cpu_base->hang_detected)
645 return;
646
647 /*
648 * Program the timer hardware. We enforce the expiry for
649 * events which are already in the past.
650 */
651 cpu_base->expires_next = expires;
652 tick_program_event(expires, 1);
653}
654
655/*
656 * Initialize the high resolution related parts of cpu_base
657 */
658static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
659{
660 base->expires_next.tv64 = KTIME_MAX;
661 base->hres_active = 0;
662}
663
664/*
665 * Retrigger next event is called after clock was set
666 *
667 * Called with interrupts disabled via on_each_cpu()
668 */
669static void retrigger_next_event(void *arg)
670{
671 struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
672
673 if (!base->hres_active)
674 return;
675
676 raw_spin_lock(&base->lock);
677 hrtimer_update_base(base);
678 hrtimer_force_reprogram(base, 0);
679 raw_spin_unlock(&base->lock);
680}
681
682/*
683 * Switch to high resolution mode
684 */
685static void hrtimer_switch_to_hres(void)
686{
687 struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
688
689 if (tick_init_highres()) {
690 printk(KERN_WARNING "Could not switch to high resolution "
691 "mode on CPU %d\n", base->cpu);
692 return;
693 }
694 base->hres_active = 1;
695 hrtimer_resolution = HIGH_RES_NSEC;
696
697 tick_setup_sched_timer();
698 /* "Retrigger" the interrupt to get things going */
699 retrigger_next_event(NULL);
700}
701
702static void clock_was_set_work(struct work_struct *work)
703{
704 clock_was_set();
705}
706
707static DECLARE_WORK(hrtimer_work, clock_was_set_work);
708
709/*
710 * Called from timekeeping and resume code to reprogramm the hrtimer
711 * interrupt device on all cpus.
712 */
713void clock_was_set_delayed(void)
714{
715 schedule_work(&hrtimer_work);
716}
717
718#else
719
720static inline int __hrtimer_hres_active(struct hrtimer_cpu_base *b) { return 0; }
721static inline int hrtimer_hres_active(void) { return 0; }
722static inline int hrtimer_is_hres_enabled(void) { return 0; }
723static inline void hrtimer_switch_to_hres(void) { }
724static inline void
725hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { }
726static inline int hrtimer_reprogram(struct hrtimer *timer,
727 struct hrtimer_clock_base *base)
728{
729 return 0;
730}
731static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { }
732static inline void retrigger_next_event(void *arg) { }
733
734#endif /* CONFIG_HIGH_RES_TIMERS */
735
736/*
737 * Clock realtime was set
738 *
739 * Change the offset of the realtime clock vs. the monotonic
740 * clock.
741 *
742 * We might have to reprogram the high resolution timer interrupt. On
743 * SMP we call the architecture specific code to retrigger _all_ high
744 * resolution timer interrupts. On UP we just disable interrupts and
745 * call the high resolution interrupt code.
746 */
747void clock_was_set(void)
748{
749#ifdef CONFIG_HIGH_RES_TIMERS
750 /* Retrigger the CPU local events everywhere */
751 on_each_cpu(retrigger_next_event, NULL, 1);
752#endif
753 timerfd_clock_was_set();
754}
755
756/*
757 * During resume we might have to reprogram the high resolution timer
758 * interrupt on all online CPUs. However, all other CPUs will be
759 * stopped with IRQs interrupts disabled so the clock_was_set() call
760 * must be deferred.
761 */
762void hrtimers_resume(void)
763{
764 WARN_ONCE(!irqs_disabled(),
765 KERN_INFO "hrtimers_resume() called with IRQs enabled!");
766
767 /* Retrigger on the local CPU */
768 retrigger_next_event(NULL);
769 /* And schedule a retrigger for all others */
770 clock_was_set_delayed();
771}
772
773static inline void timer_stats_hrtimer_set_start_info(struct hrtimer *timer)
774{
775#ifdef CONFIG_TIMER_STATS
776 if (timer->start_site)
777 return;
778 timer->start_site = __builtin_return_address(0);
779 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
780 timer->start_pid = current->pid;
781#endif
782}
783
784static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer *timer)
785{
786#ifdef CONFIG_TIMER_STATS
787 timer->start_site = NULL;
788#endif
789}
790
791static inline void timer_stats_account_hrtimer(struct hrtimer *timer)
792{
793#ifdef CONFIG_TIMER_STATS
794 if (likely(!timer_stats_active))
795 return;
796 timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
797 timer->function, timer->start_comm, 0);
798#endif
799}
800
801/*
802 * Counterpart to lock_hrtimer_base above:
803 */
804static inline
805void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
806{
807 raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
808}
809
810/**
811 * hrtimer_forward - forward the timer expiry
812 * @timer: hrtimer to forward
813 * @now: forward past this time
814 * @interval: the interval to forward
815 *
816 * Forward the timer expiry so it will expire in the future.
817 * Returns the number of overruns.
818 *
819 * Can be safely called from the callback function of @timer. If
820 * called from other contexts @timer must neither be enqueued nor
821 * running the callback and the caller needs to take care of
822 * serialization.
823 *
824 * Note: This only updates the timer expiry value and does not requeue
825 * the timer.
826 */
827u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
828{
829 u64 orun = 1;
830 ktime_t delta;
831
832 delta = ktime_sub(now, hrtimer_get_expires(timer));
833
834 if (delta.tv64 < 0)
835 return 0;
836
837 if (WARN_ON(timer->state & HRTIMER_STATE_ENQUEUED))
838 return 0;
839
840 if (interval.tv64 < hrtimer_resolution)
841 interval.tv64 = hrtimer_resolution;
842
843 if (unlikely(delta.tv64 >= interval.tv64)) {
844 s64 incr = ktime_to_ns(interval);
845
846 orun = ktime_divns(delta, incr);
847 hrtimer_add_expires_ns(timer, incr * orun);
848 if (hrtimer_get_expires_tv64(timer) > now.tv64)
849 return orun;
850 /*
851 * This (and the ktime_add() below) is the
852 * correction for exact:
853 */
854 orun++;
855 }
856 hrtimer_add_expires(timer, interval);
857
858 return orun;
859}
860EXPORT_SYMBOL_GPL(hrtimer_forward);
861
862/*
863 * enqueue_hrtimer - internal function to (re)start a timer
864 *
865 * The timer is inserted in expiry order. Insertion into the
866 * red black tree is O(log(n)). Must hold the base lock.
867 *
868 * Returns 1 when the new timer is the leftmost timer in the tree.
869 */
870static int enqueue_hrtimer(struct hrtimer *timer,
871 struct hrtimer_clock_base *base)
872{
873 debug_activate(timer);
874
875 base->cpu_base->active_bases |= 1 << base->index;
876
877 timer->state = HRTIMER_STATE_ENQUEUED;
878
879 return timerqueue_add(&base->active, &timer->node);
880}
881
882/*
883 * __remove_hrtimer - internal function to remove a timer
884 *
885 * Caller must hold the base lock.
886 *
887 * High resolution timer mode reprograms the clock event device when the
888 * timer is the one which expires next. The caller can disable this by setting
889 * reprogram to zero. This is useful, when the context does a reprogramming
890 * anyway (e.g. timer interrupt)
891 */
892static void __remove_hrtimer(struct hrtimer *timer,
893 struct hrtimer_clock_base *base,
894 u8 newstate, int reprogram)
895{
896 struct hrtimer_cpu_base *cpu_base = base->cpu_base;
897 u8 state = timer->state;
898
899 timer->state = newstate;
900 if (!(state & HRTIMER_STATE_ENQUEUED))
901 return;
902
903 if (!timerqueue_del(&base->active, &timer->node))
904 cpu_base->active_bases &= ~(1 << base->index);
905
906#ifdef CONFIG_HIGH_RES_TIMERS
907 /*
908 * Note: If reprogram is false we do not update
909 * cpu_base->next_timer. This happens when we remove the first
910 * timer on a remote cpu. No harm as we never dereference
911 * cpu_base->next_timer. So the worst thing what can happen is
912 * an superflous call to hrtimer_force_reprogram() on the
913 * remote cpu later on if the same timer gets enqueued again.
914 */
915 if (reprogram && timer == cpu_base->next_timer)
916 hrtimer_force_reprogram(cpu_base, 1);
917#endif
918}
919
920/*
921 * remove hrtimer, called with base lock held
922 */
923static inline int
924remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base, bool restart)
925{
926 if (hrtimer_is_queued(timer)) {
927 u8 state = timer->state;
928 int reprogram;
929
930 /*
931 * Remove the timer and force reprogramming when high
932 * resolution mode is active and the timer is on the current
933 * CPU. If we remove a timer on another CPU, reprogramming is
934 * skipped. The interrupt event on this CPU is fired and
935 * reprogramming happens in the interrupt handler. This is a
936 * rare case and less expensive than a smp call.
937 */
938 debug_deactivate(timer);
939 timer_stats_hrtimer_clear_start_info(timer);
940 reprogram = base->cpu_base == this_cpu_ptr(&hrtimer_bases);
941
942 if (!restart)
943 state = HRTIMER_STATE_INACTIVE;
944
945 __remove_hrtimer(timer, base, state, reprogram);
946 return 1;
947 }
948 return 0;
949}
950
951static inline ktime_t hrtimer_update_lowres(struct hrtimer *timer, ktime_t tim,
952 const enum hrtimer_mode mode)
953{
954#ifdef CONFIG_TIME_LOW_RES
955 /*
956 * CONFIG_TIME_LOW_RES indicates that the system has no way to return
957 * granular time values. For relative timers we add hrtimer_resolution
958 * (i.e. one jiffie) to prevent short timeouts.
959 */
960 timer->is_rel = mode & HRTIMER_MODE_REL;
961 if (timer->is_rel)
962 tim = ktime_add_safe(tim, ktime_set(0, hrtimer_resolution));
963#endif
964 return tim;
965}
966
967/**
968 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
969 * @timer: the timer to be added
970 * @tim: expiry time
971 * @delta_ns: "slack" range for the timer
972 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
973 * relative (HRTIMER_MODE_REL)
974 */
975void hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
976 u64 delta_ns, const enum hrtimer_mode mode)
977{
978 struct hrtimer_clock_base *base, *new_base;
979 unsigned long flags;
980 int leftmost;
981
982 base = lock_hrtimer_base(timer, &flags);
983
984 /* Remove an active timer from the queue: */
985 remove_hrtimer(timer, base, true);
986
987 if (mode & HRTIMER_MODE_REL)
988 tim = ktime_add_safe(tim, base->get_time());
989
990 tim = hrtimer_update_lowres(timer, tim, mode);
991
992 hrtimer_set_expires_range_ns(timer, tim, delta_ns);
993
994 /* Switch the timer base, if necessary: */
995 new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED);
996
997 timer_stats_hrtimer_set_start_info(timer);
998
999 leftmost = enqueue_hrtimer(timer, new_base);
1000 if (!leftmost)
1001 goto unlock;
1002
1003 if (!hrtimer_is_hres_active(timer)) {
1004 /*
1005 * Kick to reschedule the next tick to handle the new timer
1006 * on dynticks target.
1007 */
1008 if (new_base->cpu_base->nohz_active)
1009 wake_up_nohz_cpu(new_base->cpu_base->cpu);
1010 } else {
1011 hrtimer_reprogram(timer, new_base);
1012 }
1013unlock:
1014 unlock_hrtimer_base(timer, &flags);
1015}
1016EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
1017
1018/**
1019 * hrtimer_try_to_cancel - try to deactivate a timer
1020 * @timer: hrtimer to stop
1021 *
1022 * Returns:
1023 * 0 when the timer was not active
1024 * 1 when the timer was active
1025 * -1 when the timer is currently excuting the callback function and
1026 * cannot be stopped
1027 */
1028int hrtimer_try_to_cancel(struct hrtimer *timer)
1029{
1030 struct hrtimer_clock_base *base;
1031 unsigned long flags;
1032 int ret = -1;
1033
1034 /*
1035 * Check lockless first. If the timer is not active (neither
1036 * enqueued nor running the callback, nothing to do here. The
1037 * base lock does not serialize against a concurrent enqueue,
1038 * so we can avoid taking it.
1039 */
1040 if (!hrtimer_active(timer))
1041 return 0;
1042
1043 base = lock_hrtimer_base(timer, &flags);
1044
1045 if (!hrtimer_callback_running(timer))
1046 ret = remove_hrtimer(timer, base, false);
1047
1048 unlock_hrtimer_base(timer, &flags);
1049
1050 return ret;
1051
1052}
1053EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1054
1055/**
1056 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1057 * @timer: the timer to be cancelled
1058 *
1059 * Returns:
1060 * 0 when the timer was not active
1061 * 1 when the timer was active
1062 */
1063int hrtimer_cancel(struct hrtimer *timer)
1064{
1065 for (;;) {
1066 int ret = hrtimer_try_to_cancel(timer);
1067
1068 if (ret >= 0)
1069 return ret;
1070 cpu_relax();
1071 }
1072}
1073EXPORT_SYMBOL_GPL(hrtimer_cancel);
1074
1075/**
1076 * hrtimer_get_remaining - get remaining time for the timer
1077 * @timer: the timer to read
1078 * @adjust: adjust relative timers when CONFIG_TIME_LOW_RES=y
1079 */
1080ktime_t __hrtimer_get_remaining(const struct hrtimer *timer, bool adjust)
1081{
1082 unsigned long flags;
1083 ktime_t rem;
1084
1085 lock_hrtimer_base(timer, &flags);
1086 if (IS_ENABLED(CONFIG_TIME_LOW_RES) && adjust)
1087 rem = hrtimer_expires_remaining_adjusted(timer);
1088 else
1089 rem = hrtimer_expires_remaining(timer);
1090 unlock_hrtimer_base(timer, &flags);
1091
1092 return rem;
1093}
1094EXPORT_SYMBOL_GPL(__hrtimer_get_remaining);
1095
1096#ifdef CONFIG_NO_HZ_COMMON
1097/**
1098 * hrtimer_get_next_event - get the time until next expiry event
1099 *
1100 * Returns the next expiry time or KTIME_MAX if no timer is pending.
1101 */
1102u64 hrtimer_get_next_event(void)
1103{
1104 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1105 u64 expires = KTIME_MAX;
1106 unsigned long flags;
1107
1108 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1109
1110 if (!__hrtimer_hres_active(cpu_base))
1111 expires = __hrtimer_get_next_event(cpu_base).tv64;
1112
1113 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1114
1115 return expires;
1116}
1117#endif
1118
1119static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1120 enum hrtimer_mode mode)
1121{
1122 struct hrtimer_cpu_base *cpu_base;
1123 int base;
1124
1125 memset(timer, 0, sizeof(struct hrtimer));
1126
1127 cpu_base = raw_cpu_ptr(&hrtimer_bases);
1128
1129 if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
1130 clock_id = CLOCK_MONOTONIC;
1131
1132 base = hrtimer_clockid_to_base(clock_id);
1133 timer->base = &cpu_base->clock_base[base];
1134 timerqueue_init(&timer->node);
1135
1136#ifdef CONFIG_TIMER_STATS
1137 timer->start_site = NULL;
1138 timer->start_pid = -1;
1139 memset(timer->start_comm, 0, TASK_COMM_LEN);
1140#endif
1141}
1142
1143/**
1144 * hrtimer_init - initialize a timer to the given clock
1145 * @timer: the timer to be initialized
1146 * @clock_id: the clock to be used
1147 * @mode: timer mode abs/rel
1148 */
1149void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1150 enum hrtimer_mode mode)
1151{
1152 debug_init(timer, clock_id, mode);
1153 __hrtimer_init(timer, clock_id, mode);
1154}
1155EXPORT_SYMBOL_GPL(hrtimer_init);
1156
1157/*
1158 * A timer is active, when it is enqueued into the rbtree or the
1159 * callback function is running or it's in the state of being migrated
1160 * to another cpu.
1161 *
1162 * It is important for this function to not return a false negative.
1163 */
1164bool hrtimer_active(const struct hrtimer *timer)
1165{
1166 struct hrtimer_cpu_base *cpu_base;
1167 unsigned int seq;
1168
1169 do {
1170 cpu_base = READ_ONCE(timer->base->cpu_base);
1171 seq = raw_read_seqcount_begin(&cpu_base->seq);
1172
1173 if (timer->state != HRTIMER_STATE_INACTIVE ||
1174 cpu_base->running == timer)
1175 return true;
1176
1177 } while (read_seqcount_retry(&cpu_base->seq, seq) ||
1178 cpu_base != READ_ONCE(timer->base->cpu_base));
1179
1180 return false;
1181}
1182EXPORT_SYMBOL_GPL(hrtimer_active);
1183
1184/*
1185 * The write_seqcount_barrier()s in __run_hrtimer() split the thing into 3
1186 * distinct sections:
1187 *
1188 * - queued: the timer is queued
1189 * - callback: the timer is being ran
1190 * - post: the timer is inactive or (re)queued
1191 *
1192 * On the read side we ensure we observe timer->state and cpu_base->running
1193 * from the same section, if anything changed while we looked at it, we retry.
1194 * This includes timer->base changing because sequence numbers alone are
1195 * insufficient for that.
1196 *
1197 * The sequence numbers are required because otherwise we could still observe
1198 * a false negative if the read side got smeared over multiple consequtive
1199 * __run_hrtimer() invocations.
1200 */
1201
1202static void __run_hrtimer(struct hrtimer_cpu_base *cpu_base,
1203 struct hrtimer_clock_base *base,
1204 struct hrtimer *timer, ktime_t *now)
1205{
1206 enum hrtimer_restart (*fn)(struct hrtimer *);
1207 int restart;
1208
1209 lockdep_assert_held(&cpu_base->lock);
1210
1211 debug_deactivate(timer);
1212 cpu_base->running = timer;
1213
1214 /*
1215 * Separate the ->running assignment from the ->state assignment.
1216 *
1217 * As with a regular write barrier, this ensures the read side in
1218 * hrtimer_active() cannot observe cpu_base->running == NULL &&
1219 * timer->state == INACTIVE.
1220 */
1221 raw_write_seqcount_barrier(&cpu_base->seq);
1222
1223 __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE, 0);
1224 timer_stats_account_hrtimer(timer);
1225 fn = timer->function;
1226
1227 /*
1228 * Clear the 'is relative' flag for the TIME_LOW_RES case. If the
1229 * timer is restarted with a period then it becomes an absolute
1230 * timer. If its not restarted it does not matter.
1231 */
1232 if (IS_ENABLED(CONFIG_TIME_LOW_RES))
1233 timer->is_rel = false;
1234
1235 /*
1236 * Because we run timers from hardirq context, there is no chance
1237 * they get migrated to another cpu, therefore its safe to unlock
1238 * the timer base.
1239 */
1240 raw_spin_unlock(&cpu_base->lock);
1241 trace_hrtimer_expire_entry(timer, now);
1242 restart = fn(timer);
1243 trace_hrtimer_expire_exit(timer);
1244 raw_spin_lock(&cpu_base->lock);
1245
1246 /*
1247 * Note: We clear the running state after enqueue_hrtimer and
1248 * we do not reprogramm the event hardware. Happens either in
1249 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1250 *
1251 * Note: Because we dropped the cpu_base->lock above,
1252 * hrtimer_start_range_ns() can have popped in and enqueued the timer
1253 * for us already.
1254 */
1255 if (restart != HRTIMER_NORESTART &&
1256 !(timer->state & HRTIMER_STATE_ENQUEUED))
1257 enqueue_hrtimer(timer, base);
1258
1259 /*
1260 * Separate the ->running assignment from the ->state assignment.
1261 *
1262 * As with a regular write barrier, this ensures the read side in
1263 * hrtimer_active() cannot observe cpu_base->running == NULL &&
1264 * timer->state == INACTIVE.
1265 */
1266 raw_write_seqcount_barrier(&cpu_base->seq);
1267
1268 WARN_ON_ONCE(cpu_base->running != timer);
1269 cpu_base->running = NULL;
1270}
1271
1272static void __hrtimer_run_queues(struct hrtimer_cpu_base *cpu_base, ktime_t now)
1273{
1274 struct hrtimer_clock_base *base = cpu_base->clock_base;
1275 unsigned int active = cpu_base->active_bases;
1276
1277 for (; active; base++, active >>= 1) {
1278 struct timerqueue_node *node;
1279 ktime_t basenow;
1280
1281 if (!(active & 0x01))
1282 continue;
1283
1284 basenow = ktime_add(now, base->offset);
1285
1286 while ((node = timerqueue_getnext(&base->active))) {
1287 struct hrtimer *timer;
1288
1289 timer = container_of(node, struct hrtimer, node);
1290
1291 /*
1292 * The immediate goal for using the softexpires is
1293 * minimizing wakeups, not running timers at the
1294 * earliest interrupt after their soft expiration.
1295 * This allows us to avoid using a Priority Search
1296 * Tree, which can answer a stabbing querry for
1297 * overlapping intervals and instead use the simple
1298 * BST we already have.
1299 * We don't add extra wakeups by delaying timers that
1300 * are right-of a not yet expired timer, because that
1301 * timer will have to trigger a wakeup anyway.
1302 */
1303 if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer))
1304 break;
1305
1306 __run_hrtimer(cpu_base, base, timer, &basenow);
1307 }
1308 }
1309}
1310
1311#ifdef CONFIG_HIGH_RES_TIMERS
1312
1313/*
1314 * High resolution timer interrupt
1315 * Called with interrupts disabled
1316 */
1317void hrtimer_interrupt(struct clock_event_device *dev)
1318{
1319 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1320 ktime_t expires_next, now, entry_time, delta;
1321 int retries = 0;
1322
1323 BUG_ON(!cpu_base->hres_active);
1324 cpu_base->nr_events++;
1325 dev->next_event.tv64 = KTIME_MAX;
1326
1327 raw_spin_lock(&cpu_base->lock);
1328 entry_time = now = hrtimer_update_base(cpu_base);
1329retry:
1330 cpu_base->in_hrtirq = 1;
1331 /*
1332 * We set expires_next to KTIME_MAX here with cpu_base->lock
1333 * held to prevent that a timer is enqueued in our queue via
1334 * the migration code. This does not affect enqueueing of
1335 * timers which run their callback and need to be requeued on
1336 * this CPU.
1337 */
1338 cpu_base->expires_next.tv64 = KTIME_MAX;
1339
1340 __hrtimer_run_queues(cpu_base, now);
1341
1342 /* Reevaluate the clock bases for the next expiry */
1343 expires_next = __hrtimer_get_next_event(cpu_base);
1344 /*
1345 * Store the new expiry value so the migration code can verify
1346 * against it.
1347 */
1348 cpu_base->expires_next = expires_next;
1349 cpu_base->in_hrtirq = 0;
1350 raw_spin_unlock(&cpu_base->lock);
1351
1352 /* Reprogramming necessary ? */
1353 if (!tick_program_event(expires_next, 0)) {
1354 cpu_base->hang_detected = 0;
1355 return;
1356 }
1357
1358 /*
1359 * The next timer was already expired due to:
1360 * - tracing
1361 * - long lasting callbacks
1362 * - being scheduled away when running in a VM
1363 *
1364 * We need to prevent that we loop forever in the hrtimer
1365 * interrupt routine. We give it 3 attempts to avoid
1366 * overreacting on some spurious event.
1367 *
1368 * Acquire base lock for updating the offsets and retrieving
1369 * the current time.
1370 */
1371 raw_spin_lock(&cpu_base->lock);
1372 now = hrtimer_update_base(cpu_base);
1373 cpu_base->nr_retries++;
1374 if (++retries < 3)
1375 goto retry;
1376 /*
1377 * Give the system a chance to do something else than looping
1378 * here. We stored the entry time, so we know exactly how long
1379 * we spent here. We schedule the next event this amount of
1380 * time away.
1381 */
1382 cpu_base->nr_hangs++;
1383 cpu_base->hang_detected = 1;
1384 raw_spin_unlock(&cpu_base->lock);
1385 delta = ktime_sub(now, entry_time);
1386 if ((unsigned int)delta.tv64 > cpu_base->max_hang_time)
1387 cpu_base->max_hang_time = (unsigned int) delta.tv64;
1388 /*
1389 * Limit it to a sensible value as we enforce a longer
1390 * delay. Give the CPU at least 100ms to catch up.
1391 */
1392 if (delta.tv64 > 100 * NSEC_PER_MSEC)
1393 expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1394 else
1395 expires_next = ktime_add(now, delta);
1396 tick_program_event(expires_next, 1);
1397 printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n",
1398 ktime_to_ns(delta));
1399}
1400
1401/*
1402 * local version of hrtimer_peek_ahead_timers() called with interrupts
1403 * disabled.
1404 */
1405static inline void __hrtimer_peek_ahead_timers(void)
1406{
1407 struct tick_device *td;
1408
1409 if (!hrtimer_hres_active())
1410 return;
1411
1412 td = this_cpu_ptr(&tick_cpu_device);
1413 if (td && td->evtdev)
1414 hrtimer_interrupt(td->evtdev);
1415}
1416
1417#else /* CONFIG_HIGH_RES_TIMERS */
1418
1419static inline void __hrtimer_peek_ahead_timers(void) { }
1420
1421#endif /* !CONFIG_HIGH_RES_TIMERS */
1422
1423/*
1424 * Called from run_local_timers in hardirq context every jiffy
1425 */
1426void hrtimer_run_queues(void)
1427{
1428 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1429 ktime_t now;
1430
1431 if (__hrtimer_hres_active(cpu_base))
1432 return;
1433
1434 /*
1435 * This _is_ ugly: We have to check periodically, whether we
1436 * can switch to highres and / or nohz mode. The clocksource
1437 * switch happens with xtime_lock held. Notification from
1438 * there only sets the check bit in the tick_oneshot code,
1439 * otherwise we might deadlock vs. xtime_lock.
1440 */
1441 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) {
1442 hrtimer_switch_to_hres();
1443 return;
1444 }
1445
1446 raw_spin_lock(&cpu_base->lock);
1447 now = hrtimer_update_base(cpu_base);
1448 __hrtimer_run_queues(cpu_base, now);
1449 raw_spin_unlock(&cpu_base->lock);
1450}
1451
1452/*
1453 * Sleep related functions:
1454 */
1455static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1456{
1457 struct hrtimer_sleeper *t =
1458 container_of(timer, struct hrtimer_sleeper, timer);
1459 struct task_struct *task = t->task;
1460
1461 t->task = NULL;
1462 if (task)
1463 wake_up_process(task);
1464
1465 return HRTIMER_NORESTART;
1466}
1467
1468void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1469{
1470 sl->timer.function = hrtimer_wakeup;
1471 sl->task = task;
1472}
1473EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
1474
1475static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1476{
1477 hrtimer_init_sleeper(t, current);
1478
1479 do {
1480 set_current_state(TASK_INTERRUPTIBLE);
1481 hrtimer_start_expires(&t->timer, mode);
1482
1483 if (likely(t->task))
1484 freezable_schedule();
1485
1486 hrtimer_cancel(&t->timer);
1487 mode = HRTIMER_MODE_ABS;
1488
1489 } while (t->task && !signal_pending(current));
1490
1491 __set_current_state(TASK_RUNNING);
1492
1493 return t->task == NULL;
1494}
1495
1496static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp)
1497{
1498 struct timespec rmt;
1499 ktime_t rem;
1500
1501 rem = hrtimer_expires_remaining(timer);
1502 if (rem.tv64 <= 0)
1503 return 0;
1504 rmt = ktime_to_timespec(rem);
1505
1506 if (copy_to_user(rmtp, &rmt, sizeof(*rmtp)))
1507 return -EFAULT;
1508
1509 return 1;
1510}
1511
1512long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1513{
1514 struct hrtimer_sleeper t;
1515 struct timespec __user *rmtp;
1516 int ret = 0;
1517
1518 hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid,
1519 HRTIMER_MODE_ABS);
1520 hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
1521
1522 if (do_nanosleep(&t, HRTIMER_MODE_ABS))
1523 goto out;
1524
1525 rmtp = restart->nanosleep.rmtp;
1526 if (rmtp) {
1527 ret = update_rmtp(&t.timer, rmtp);
1528 if (ret <= 0)
1529 goto out;
1530 }
1531
1532 /* The other values in restart are already filled in */
1533 ret = -ERESTART_RESTARTBLOCK;
1534out:
1535 destroy_hrtimer_on_stack(&t.timer);
1536 return ret;
1537}
1538
1539long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
1540 const enum hrtimer_mode mode, const clockid_t clockid)
1541{
1542 struct restart_block *restart;
1543 struct hrtimer_sleeper t;
1544 int ret = 0;
1545 u64 slack;
1546
1547 slack = current->timer_slack_ns;
1548 if (dl_task(current) || rt_task(current))
1549 slack = 0;
1550
1551 hrtimer_init_on_stack(&t.timer, clockid, mode);
1552 hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack);
1553 if (do_nanosleep(&t, mode))
1554 goto out;
1555
1556 /* Absolute timers do not update the rmtp value and restart: */
1557 if (mode == HRTIMER_MODE_ABS) {
1558 ret = -ERESTARTNOHAND;
1559 goto out;
1560 }
1561
1562 if (rmtp) {
1563 ret = update_rmtp(&t.timer, rmtp);
1564 if (ret <= 0)
1565 goto out;
1566 }
1567
1568 restart = ¤t->restart_block;
1569 restart->fn = hrtimer_nanosleep_restart;
1570 restart->nanosleep.clockid = t.timer.base->clockid;
1571 restart->nanosleep.rmtp = rmtp;
1572 restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
1573
1574 ret = -ERESTART_RESTARTBLOCK;
1575out:
1576 destroy_hrtimer_on_stack(&t.timer);
1577 return ret;
1578}
1579
1580SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp,
1581 struct timespec __user *, rmtp)
1582{
1583 struct timespec tu;
1584
1585 if (copy_from_user(&tu, rqtp, sizeof(tu)))
1586 return -EFAULT;
1587
1588 if (!timespec_valid(&tu))
1589 return -EINVAL;
1590
1591 return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1592}
1593
1594/*
1595 * Functions related to boot-time initialization:
1596 */
1597static void init_hrtimers_cpu(int cpu)
1598{
1599 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1600 int i;
1601
1602 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1603 cpu_base->clock_base[i].cpu_base = cpu_base;
1604 timerqueue_init_head(&cpu_base->clock_base[i].active);
1605 }
1606
1607 cpu_base->cpu = cpu;
1608 hrtimer_init_hres(cpu_base);
1609}
1610
1611#ifdef CONFIG_HOTPLUG_CPU
1612
1613static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1614 struct hrtimer_clock_base *new_base)
1615{
1616 struct hrtimer *timer;
1617 struct timerqueue_node *node;
1618
1619 while ((node = timerqueue_getnext(&old_base->active))) {
1620 timer = container_of(node, struct hrtimer, node);
1621 BUG_ON(hrtimer_callback_running(timer));
1622 debug_deactivate(timer);
1623
1624 /*
1625 * Mark it as ENQUEUED not INACTIVE otherwise the
1626 * timer could be seen as !active and just vanish away
1627 * under us on another CPU
1628 */
1629 __remove_hrtimer(timer, old_base, HRTIMER_STATE_ENQUEUED, 0);
1630 timer->base = new_base;
1631 /*
1632 * Enqueue the timers on the new cpu. This does not
1633 * reprogram the event device in case the timer
1634 * expires before the earliest on this CPU, but we run
1635 * hrtimer_interrupt after we migrated everything to
1636 * sort out already expired timers and reprogram the
1637 * event device.
1638 */
1639 enqueue_hrtimer(timer, new_base);
1640 }
1641}
1642
1643static void migrate_hrtimers(int scpu)
1644{
1645 struct hrtimer_cpu_base *old_base, *new_base;
1646 int i;
1647
1648 BUG_ON(cpu_online(scpu));
1649 tick_cancel_sched_timer(scpu);
1650
1651 local_irq_disable();
1652 old_base = &per_cpu(hrtimer_bases, scpu);
1653 new_base = this_cpu_ptr(&hrtimer_bases);
1654 /*
1655 * The caller is globally serialized and nobody else
1656 * takes two locks at once, deadlock is not possible.
1657 */
1658 raw_spin_lock(&new_base->lock);
1659 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1660
1661 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1662 migrate_hrtimer_list(&old_base->clock_base[i],
1663 &new_base->clock_base[i]);
1664 }
1665
1666 raw_spin_unlock(&old_base->lock);
1667 raw_spin_unlock(&new_base->lock);
1668
1669 /* Check, if we got expired work to do */
1670 __hrtimer_peek_ahead_timers();
1671 local_irq_enable();
1672}
1673
1674#endif /* CONFIG_HOTPLUG_CPU */
1675
1676static int hrtimer_cpu_notify(struct notifier_block *self,
1677 unsigned long action, void *hcpu)
1678{
1679 int scpu = (long)hcpu;
1680
1681 switch (action) {
1682
1683 case CPU_UP_PREPARE:
1684 case CPU_UP_PREPARE_FROZEN:
1685 init_hrtimers_cpu(scpu);
1686 break;
1687
1688#ifdef CONFIG_HOTPLUG_CPU
1689 case CPU_DEAD:
1690 case CPU_DEAD_FROZEN:
1691 migrate_hrtimers(scpu);
1692 break;
1693#endif
1694
1695 default:
1696 break;
1697 }
1698
1699 return NOTIFY_OK;
1700}
1701
1702static struct notifier_block hrtimers_nb = {
1703 .notifier_call = hrtimer_cpu_notify,
1704};
1705
1706void __init hrtimers_init(void)
1707{
1708 hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
1709 (void *)(long)smp_processor_id());
1710 register_cpu_notifier(&hrtimers_nb);
1711}
1712
1713/**
1714 * schedule_hrtimeout_range_clock - sleep until timeout
1715 * @expires: timeout value (ktime_t)
1716 * @delta: slack in expires timeout (ktime_t)
1717 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1718 * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME
1719 */
1720int __sched
1721schedule_hrtimeout_range_clock(ktime_t *expires, u64 delta,
1722 const enum hrtimer_mode mode, int clock)
1723{
1724 struct hrtimer_sleeper t;
1725
1726 /*
1727 * Optimize when a zero timeout value is given. It does not
1728 * matter whether this is an absolute or a relative time.
1729 */
1730 if (expires && !expires->tv64) {
1731 __set_current_state(TASK_RUNNING);
1732 return 0;
1733 }
1734
1735 /*
1736 * A NULL parameter means "infinite"
1737 */
1738 if (!expires) {
1739 schedule();
1740 return -EINTR;
1741 }
1742
1743 hrtimer_init_on_stack(&t.timer, clock, mode);
1744 hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
1745
1746 hrtimer_init_sleeper(&t, current);
1747
1748 hrtimer_start_expires(&t.timer, mode);
1749
1750 if (likely(t.task))
1751 schedule();
1752
1753 hrtimer_cancel(&t.timer);
1754 destroy_hrtimer_on_stack(&t.timer);
1755
1756 __set_current_state(TASK_RUNNING);
1757
1758 return !t.task ? 0 : -EINTR;
1759}
1760
1761/**
1762 * schedule_hrtimeout_range - sleep until timeout
1763 * @expires: timeout value (ktime_t)
1764 * @delta: slack in expires timeout (ktime_t)
1765 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1766 *
1767 * Make the current task sleep until the given expiry time has
1768 * elapsed. The routine will return immediately unless
1769 * the current task state has been set (see set_current_state()).
1770 *
1771 * The @delta argument gives the kernel the freedom to schedule the
1772 * actual wakeup to a time that is both power and performance friendly.
1773 * The kernel give the normal best effort behavior for "@expires+@delta",
1774 * but may decide to fire the timer earlier, but no earlier than @expires.
1775 *
1776 * You can set the task state as follows -
1777 *
1778 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1779 * pass before the routine returns.
1780 *
1781 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1782 * delivered to the current task.
1783 *
1784 * The current task state is guaranteed to be TASK_RUNNING when this
1785 * routine returns.
1786 *
1787 * Returns 0 when the timer has expired otherwise -EINTR
1788 */
1789int __sched schedule_hrtimeout_range(ktime_t *expires, u64 delta,
1790 const enum hrtimer_mode mode)
1791{
1792 return schedule_hrtimeout_range_clock(expires, delta, mode,
1793 CLOCK_MONOTONIC);
1794}
1795EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
1796
1797/**
1798 * schedule_hrtimeout - sleep until timeout
1799 * @expires: timeout value (ktime_t)
1800 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1801 *
1802 * Make the current task sleep until the given expiry time has
1803 * elapsed. The routine will return immediately unless
1804 * the current task state has been set (see set_current_state()).
1805 *
1806 * You can set the task state as follows -
1807 *
1808 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1809 * pass before the routine returns.
1810 *
1811 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1812 * delivered to the current task.
1813 *
1814 * The current task state is guaranteed to be TASK_RUNNING when this
1815 * routine returns.
1816 *
1817 * Returns 0 when the timer has expired otherwise -EINTR
1818 */
1819int __sched schedule_hrtimeout(ktime_t *expires,
1820 const enum hrtimer_mode mode)
1821{
1822 return schedule_hrtimeout_range(expires, 0, mode);
1823}
1824EXPORT_SYMBOL_GPL(schedule_hrtimeout);