projects/linux·blob v4.15 · refs

1790 lines · 46.5 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/signal.h>
47#include <linux/sched/sysctl.h>
48#include <linux/sched/rt.h>
49#include <linux/sched/deadline.h>
50#include <linux/sched/nohz.h>
51#include <linux/sched/debug.h>
52#include <linux/timer.h>
53#include <linux/freezer.h>
54#include <linux/compat.h>
55
56#include <linux/uaccess.h>
57
58#include <trace/events/timer.h>
59
60#include "tick-internal.h"
61
62/*
63 * The timer bases:
64 *
65 * There are more clockids than hrtimer bases. Thus, we index
66 * into the timer bases by the hrtimer_base_type enum. When trying
67 * to reach a base using a clockid, hrtimer_clockid_to_base()
68 * is used to convert from clockid to the proper hrtimer_base_type.
69 */
70DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
71{
72 .lock = __RAW_SPIN_LOCK_UNLOCKED(hrtimer_bases.lock),
73 .seq = SEQCNT_ZERO(hrtimer_bases.seq),
74 .clock_base =
75 {
76 {
77 .index = HRTIMER_BASE_MONOTONIC,
78 .clockid = CLOCK_MONOTONIC,
79 .get_time = &ktime_get,
80 },
81 {
82 .index = HRTIMER_BASE_REALTIME,
83 .clockid = CLOCK_REALTIME,
84 .get_time = &ktime_get_real,
85 },
86 {
87 .index = HRTIMER_BASE_BOOTTIME,
88 .clockid = CLOCK_BOOTTIME,
89 .get_time = &ktime_get_boottime,
90 },
91 {
92 .index = HRTIMER_BASE_TAI,
93 .clockid = CLOCK_TAI,
94 .get_time = &ktime_get_clocktai,
95 },
96 }
97};
98
99static const int hrtimer_clock_to_base_table[MAX_CLOCKS] = {
100 /* Make sure we catch unsupported clockids */
101 [0 ... MAX_CLOCKS - 1] = HRTIMER_MAX_CLOCK_BASES,
102
103 [CLOCK_REALTIME] = HRTIMER_BASE_REALTIME,
104 [CLOCK_MONOTONIC] = HRTIMER_BASE_MONOTONIC,
105 [CLOCK_BOOTTIME] = HRTIMER_BASE_BOOTTIME,
106 [CLOCK_TAI] = HRTIMER_BASE_TAI,
107};
108
109/*
110 * Functions and macros which are different for UP/SMP systems are kept in a
111 * single place
112 */
113#ifdef CONFIG_SMP
114
115/*
116 * We require the migration_base for lock_hrtimer_base()/switch_hrtimer_base()
117 * such that hrtimer_callback_running() can unconditionally dereference
118 * timer->base->cpu_base
119 */
120static struct hrtimer_cpu_base migration_cpu_base = {
121 .seq = SEQCNT_ZERO(migration_cpu_base),
122 .clock_base = { { .cpu_base = &migration_cpu_base, }, },
123};
124
125#define migration_base migration_cpu_base.clock_base[0]
126
127/*
128 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
129 * means that all timers which are tied to this base via timer->base are
130 * locked, and the base itself is locked too.
131 *
132 * So __run_timers/migrate_timers can safely modify all timers which could
133 * be found on the lists/queues.
134 *
135 * When the timer's base is locked, and the timer removed from list, it is
136 * possible to set timer->base = &migration_base and drop the lock: the timer
137 * remains locked.
138 */
139static
140struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
141 unsigned long *flags)
142{
143 struct hrtimer_clock_base *base;
144
145 for (;;) {
146 base = timer->base;
147 if (likely(base != &migration_base)) {
148 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
149 if (likely(base == timer->base))
150 return base;
151 /* The timer has migrated to another CPU: */
152 raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
153 }
154 cpu_relax();
155 }
156}
157
158/*
159 * With HIGHRES=y we do not migrate the timer when it is expiring
160 * before the next event on the target cpu because we cannot reprogram
161 * the target cpu hardware and we would cause it to fire late.
162 *
163 * Called with cpu_base->lock of target cpu held.
164 */
165static int
166hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base)
167{
168#ifdef CONFIG_HIGH_RES_TIMERS
169 ktime_t expires;
170
171 if (!new_base->cpu_base->hres_active)
172 return 0;
173
174 expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset);
175 return expires <= new_base->cpu_base->expires_next;
176#else
177 return 0;
178#endif
179}
180
181#ifdef CONFIG_NO_HZ_COMMON
182static inline
183struct hrtimer_cpu_base *get_target_base(struct hrtimer_cpu_base *base,
184 int pinned)
185{
186 if (pinned || !base->migration_enabled)
187 return base;
188 return &per_cpu(hrtimer_bases, get_nohz_timer_target());
189}
190#else
191static inline
192struct hrtimer_cpu_base *get_target_base(struct hrtimer_cpu_base *base,
193 int pinned)
194{
195 return base;
196}
197#endif
198
199/*
200 * We switch the timer base to a power-optimized selected CPU target,
201 * if:
202 * - NO_HZ_COMMON is enabled
203 * - timer migration is enabled
204 * - the timer callback is not running
205 * - the timer is not the first expiring timer on the new target
206 *
207 * If one of the above requirements is not fulfilled we move the timer
208 * to the current CPU or leave it on the previously assigned CPU if
209 * the timer callback is currently running.
210 */
211static inline struct hrtimer_clock_base *
212switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
213 int pinned)
214{
215 struct hrtimer_cpu_base *new_cpu_base, *this_cpu_base;
216 struct hrtimer_clock_base *new_base;
217 int basenum = base->index;
218
219 this_cpu_base = this_cpu_ptr(&hrtimer_bases);
220 new_cpu_base = get_target_base(this_cpu_base, pinned);
221again:
222 new_base = &new_cpu_base->clock_base[basenum];
223
224 if (base != new_base) {
225 /*
226 * We are trying to move timer to new_base.
227 * However we can't change timer's base while it is running,
228 * so we keep it on the same CPU. No hassle vs. reprogramming
229 * the event source in the high resolution case. The softirq
230 * code will take care of this when the timer function has
231 * completed. There is no conflict as we hold the lock until
232 * the timer is enqueued.
233 */
234 if (unlikely(hrtimer_callback_running(timer)))
235 return base;
236
237 /* See the comment in lock_hrtimer_base() */
238 timer->base = &migration_base;
239 raw_spin_unlock(&base->cpu_base->lock);
240 raw_spin_lock(&new_base->cpu_base->lock);
241
242 if (new_cpu_base != this_cpu_base &&
243 hrtimer_check_target(timer, new_base)) {
244 raw_spin_unlock(&new_base->cpu_base->lock);
245 raw_spin_lock(&base->cpu_base->lock);
246 new_cpu_base = this_cpu_base;
247 timer->base = base;
248 goto again;
249 }
250 timer->base = new_base;
251 } else {
252 if (new_cpu_base != this_cpu_base &&
253 hrtimer_check_target(timer, new_base)) {
254 new_cpu_base = this_cpu_base;
255 goto again;
256 }
257 }
258 return new_base;
259}
260
261#else /* CONFIG_SMP */
262
263static inline struct hrtimer_clock_base *
264lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
265{
266 struct hrtimer_clock_base *base = timer->base;
267
268 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
269
270 return base;
271}
272
273# define switch_hrtimer_base(t, b, p) (b)
274
275#endif /* !CONFIG_SMP */
276
277/*
278 * Functions for the union type storage format of ktime_t which are
279 * too large for inlining:
280 */
281#if BITS_PER_LONG < 64
282/*
283 * Divide a ktime value by a nanosecond value
284 */
285s64 __ktime_divns(const ktime_t kt, s64 div)
286{
287 int sft = 0;
288 s64 dclc;
289 u64 tmp;
290
291 dclc = ktime_to_ns(kt);
292 tmp = dclc < 0 ? -dclc : dclc;
293
294 /* Make sure the divisor is less than 2^32: */
295 while (div >> 32) {
296 sft++;
297 div >>= 1;
298 }
299 tmp >>= sft;
300 do_div(tmp, (unsigned long) div);
301 return dclc < 0 ? -tmp : tmp;
302}
303EXPORT_SYMBOL_GPL(__ktime_divns);
304#endif /* BITS_PER_LONG >= 64 */
305
306/*
307 * Add two ktime values and do a safety check for overflow:
308 */
309ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
310{
311 ktime_t res = ktime_add_unsafe(lhs, rhs);
312
313 /*
314 * We use KTIME_SEC_MAX here, the maximum timeout which we can
315 * return to user space in a timespec:
316 */
317 if (res < 0 || res < lhs || res < rhs)
318 res = ktime_set(KTIME_SEC_MAX, 0);
319
320 return res;
321}
322
323EXPORT_SYMBOL_GPL(ktime_add_safe);
324
325#ifdef CONFIG_DEBUG_OBJECTS_TIMERS
326
327static struct debug_obj_descr hrtimer_debug_descr;
328
329static void *hrtimer_debug_hint(void *addr)
330{
331 return ((struct hrtimer *) addr)->function;
332}
333
334/*
335 * fixup_init is called when:
336 * - an active object is initialized
337 */
338static bool hrtimer_fixup_init(void *addr, enum debug_obj_state state)
339{
340 struct hrtimer *timer = addr;
341
342 switch (state) {
343 case ODEBUG_STATE_ACTIVE:
344 hrtimer_cancel(timer);
345 debug_object_init(timer, &hrtimer_debug_descr);
346 return true;
347 default:
348 return false;
349 }
350}
351
352/*
353 * fixup_activate is called when:
354 * - an active object is activated
355 * - an unknown non-static object is activated
356 */
357static bool hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
358{
359 switch (state) {
360 case ODEBUG_STATE_ACTIVE:
361 WARN_ON(1);
362
363 default:
364 return false;
365 }
366}
367
368/*
369 * fixup_free is called when:
370 * - an active object is freed
371 */
372static bool hrtimer_fixup_free(void *addr, enum debug_obj_state state)
373{
374 struct hrtimer *timer = addr;
375
376 switch (state) {
377 case ODEBUG_STATE_ACTIVE:
378 hrtimer_cancel(timer);
379 debug_object_free(timer, &hrtimer_debug_descr);
380 return true;
381 default:
382 return false;
383 }
384}
385
386static struct debug_obj_descr hrtimer_debug_descr = {
387 .name = "hrtimer",
388 .debug_hint = hrtimer_debug_hint,
389 .fixup_init = hrtimer_fixup_init,
390 .fixup_activate = hrtimer_fixup_activate,
391 .fixup_free = hrtimer_fixup_free,
392};
393
394static inline void debug_hrtimer_init(struct hrtimer *timer)
395{
396 debug_object_init(timer, &hrtimer_debug_descr);
397}
398
399static inline void debug_hrtimer_activate(struct hrtimer *timer)
400{
401 debug_object_activate(timer, &hrtimer_debug_descr);
402}
403
404static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
405{
406 debug_object_deactivate(timer, &hrtimer_debug_descr);
407}
408
409static inline void debug_hrtimer_free(struct hrtimer *timer)
410{
411 debug_object_free(timer, &hrtimer_debug_descr);
412}
413
414static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
415 enum hrtimer_mode mode);
416
417void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
418 enum hrtimer_mode mode)
419{
420 debug_object_init_on_stack(timer, &hrtimer_debug_descr);
421 __hrtimer_init(timer, clock_id, mode);
422}
423EXPORT_SYMBOL_GPL(hrtimer_init_on_stack);
424
425void destroy_hrtimer_on_stack(struct hrtimer *timer)
426{
427 debug_object_free(timer, &hrtimer_debug_descr);
428}
429EXPORT_SYMBOL_GPL(destroy_hrtimer_on_stack);
430
431#else
432static inline void debug_hrtimer_init(struct hrtimer *timer) { }
433static inline void debug_hrtimer_activate(struct hrtimer *timer) { }
434static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
435#endif
436
437static inline void
438debug_init(struct hrtimer *timer, clockid_t clockid,
439 enum hrtimer_mode mode)
440{
441 debug_hrtimer_init(timer);
442 trace_hrtimer_init(timer, clockid, mode);
443}
444
445static inline void debug_activate(struct hrtimer *timer)
446{
447 debug_hrtimer_activate(timer);
448 trace_hrtimer_start(timer);
449}
450
451static inline void debug_deactivate(struct hrtimer *timer)
452{
453 debug_hrtimer_deactivate(timer);
454 trace_hrtimer_cancel(timer);
455}
456
457#if defined(CONFIG_NO_HZ_COMMON) || defined(CONFIG_HIGH_RES_TIMERS)
458static inline void hrtimer_update_next_timer(struct hrtimer_cpu_base *cpu_base,
459 struct hrtimer *timer)
460{
461#ifdef CONFIG_HIGH_RES_TIMERS
462 cpu_base->next_timer = timer;
463#endif
464}
465
466static ktime_t __hrtimer_get_next_event(struct hrtimer_cpu_base *cpu_base)
467{
468 struct hrtimer_clock_base *base = cpu_base->clock_base;
469 unsigned int active = cpu_base->active_bases;
470 ktime_t expires, expires_next = KTIME_MAX;
471
472 hrtimer_update_next_timer(cpu_base, NULL);
473 for (; active; base++, active >>= 1) {
474 struct timerqueue_node *next;
475 struct hrtimer *timer;
476
477 if (!(active & 0x01))
478 continue;
479
480 next = timerqueue_getnext(&base->active);
481 timer = container_of(next, struct hrtimer, node);
482 expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
483 if (expires < expires_next) {
484 expires_next = expires;
485 hrtimer_update_next_timer(cpu_base, timer);
486 }
487 }
488 /*
489 * clock_was_set() might have changed base->offset of any of
490 * the clock bases so the result might be negative. Fix it up
491 * to prevent a false positive in clockevents_program_event().
492 */
493 if (expires_next < 0)
494 expires_next = 0;
495 return expires_next;
496}
497#endif
498
499static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base)
500{
501 ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset;
502 ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset;
503 ktime_t *offs_tai = &base->clock_base[HRTIMER_BASE_TAI].offset;
504
505 return ktime_get_update_offsets_now(&base->clock_was_set_seq,
506 offs_real, offs_boot, offs_tai);
507}
508
509/* High resolution timer related functions */
510#ifdef CONFIG_HIGH_RES_TIMERS
511
512/*
513 * High resolution timer enabled ?
514 */
515static bool hrtimer_hres_enabled __read_mostly = true;
516unsigned int hrtimer_resolution __read_mostly = LOW_RES_NSEC;
517EXPORT_SYMBOL_GPL(hrtimer_resolution);
518
519/*
520 * Enable / Disable high resolution mode
521 */
522static int __init setup_hrtimer_hres(char *str)
523{
524 return (kstrtobool(str, &hrtimer_hres_enabled) == 0);
525}
526
527__setup("highres=", setup_hrtimer_hres);
528
529/*
530 * hrtimer_high_res_enabled - query, if the highres mode is enabled
531 */
532static inline int hrtimer_is_hres_enabled(void)
533{
534 return hrtimer_hres_enabled;
535}
536
537/*
538 * Is the high resolution mode active ?
539 */
540static inline int __hrtimer_hres_active(struct hrtimer_cpu_base *cpu_base)
541{
542 return cpu_base->hres_active;
543}
544
545static inline int hrtimer_hres_active(void)
546{
547 return __hrtimer_hres_active(this_cpu_ptr(&hrtimer_bases));
548}
549
550/*
551 * Reprogram the event source with checking both queues for the
552 * next event
553 * Called with interrupts disabled and base->lock held
554 */
555static void
556hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
557{
558 ktime_t expires_next;
559
560 if (!cpu_base->hres_active)
561 return;
562
563 expires_next = __hrtimer_get_next_event(cpu_base);
564
565 if (skip_equal && expires_next == cpu_base->expires_next)
566 return;
567
568 cpu_base->expires_next = expires_next;
569
570 /*
571 * If a hang was detected in the last timer interrupt then we
572 * leave the hang delay active in the hardware. We want the
573 * system to make progress. That also prevents the following
574 * scenario:
575 * T1 expires 50ms from now
576 * T2 expires 5s from now
577 *
578 * T1 is removed, so this code is called and would reprogram
579 * the hardware to 5s from now. Any hrtimer_start after that
580 * will not reprogram the hardware due to hang_detected being
581 * set. So we'd effectivly block all timers until the T2 event
582 * fires.
583 */
584 if (cpu_base->hang_detected)
585 return;
586
587 tick_program_event(cpu_base->expires_next, 1);
588}
589
590/*
591 * When a timer is enqueued and expires earlier than the already enqueued
592 * timers, we have to check, whether it expires earlier than the timer for
593 * which the clock event device was armed.
594 *
595 * Called with interrupts disabled and base->cpu_base.lock held
596 */
597static void hrtimer_reprogram(struct hrtimer *timer,
598 struct hrtimer_clock_base *base)
599{
600 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
601 ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
602
603 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
604
605 /*
606 * If the timer is not on the current cpu, we cannot reprogram
607 * the other cpus clock event device.
608 */
609 if (base->cpu_base != cpu_base)
610 return;
611
612 /*
613 * If the hrtimer interrupt is running, then it will
614 * reevaluate the clock bases and reprogram the clock event
615 * device. The callbacks are always executed in hard interrupt
616 * context so we don't need an extra check for a running
617 * callback.
618 */
619 if (cpu_base->in_hrtirq)
620 return;
621
622 /*
623 * CLOCK_REALTIME timer might be requested with an absolute
624 * expiry time which is less than base->offset. Set it to 0.
625 */
626 if (expires < 0)
627 expires = 0;
628
629 if (expires >= cpu_base->expires_next)
630 return;
631
632 /* Update the pointer to the next expiring timer */
633 cpu_base->next_timer = timer;
634
635 /*
636 * If a hang was detected in the last timer interrupt then we
637 * do not schedule a timer which is earlier than the expiry
638 * which we enforced in the hang detection. We want the system
639 * to make progress.
640 */
641 if (cpu_base->hang_detected)
642 return;
643
644 /*
645 * Program the timer hardware. We enforce the expiry for
646 * events which are already in the past.
647 */
648 cpu_base->expires_next = expires;
649 tick_program_event(expires, 1);
650}
651
652/*
653 * Initialize the high resolution related parts of cpu_base
654 */
655static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
656{
657 base->expires_next = KTIME_MAX;
658 base->hang_detected = 0;
659 base->hres_active = 0;
660 base->next_timer = NULL;
661}
662
663/*
664 * Retrigger next event is called after clock was set
665 *
666 * Called with interrupts disabled via on_each_cpu()
667 */
668static void retrigger_next_event(void *arg)
669{
670 struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
671
672 if (!base->hres_active)
673 return;
674
675 raw_spin_lock(&base->lock);
676 hrtimer_update_base(base);
677 hrtimer_force_reprogram(base, 0);
678 raw_spin_unlock(&base->lock);
679}
680
681/*
682 * Switch to high resolution mode
683 */
684static void hrtimer_switch_to_hres(void)
685{
686 struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
687
688 if (tick_init_highres()) {
689 printk(KERN_WARNING "Could not switch to high resolution "
690 "mode on CPU %d\n", base->cpu);
691 return;
692 }
693 base->hres_active = 1;
694 hrtimer_resolution = HIGH_RES_NSEC;
695
696 tick_setup_sched_timer();
697 /* "Retrigger" the interrupt to get things going */
698 retrigger_next_event(NULL);
699}
700
701static void clock_was_set_work(struct work_struct *work)
702{
703 clock_was_set();
704}
705
706static DECLARE_WORK(hrtimer_work, clock_was_set_work);
707
708/*
709 * Called from timekeeping and resume code to reprogram the hrtimer
710 * interrupt device on all cpus.
711 */
712void clock_was_set_delayed(void)
713{
714 schedule_work(&hrtimer_work);
715}
716
717#else
718
719static inline int __hrtimer_hres_active(struct hrtimer_cpu_base *b) { return 0; }
720static inline int hrtimer_hres_active(void) { return 0; }
721static inline int hrtimer_is_hres_enabled(void) { return 0; }
722static inline void hrtimer_switch_to_hres(void) { }
723static inline void
724hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { }
725static inline int hrtimer_reprogram(struct hrtimer *timer,
726 struct hrtimer_clock_base *base)
727{
728 return 0;
729}
730static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { }
731static inline void retrigger_next_event(void *arg) { }
732
733#endif /* CONFIG_HIGH_RES_TIMERS */
734
735/*
736 * Clock realtime was set
737 *
738 * Change the offset of the realtime clock vs. the monotonic
739 * clock.
740 *
741 * We might have to reprogram the high resolution timer interrupt. On
742 * SMP we call the architecture specific code to retrigger _all_ high
743 * resolution timer interrupts. On UP we just disable interrupts and
744 * call the high resolution interrupt code.
745 */
746void clock_was_set(void)
747{
748#ifdef CONFIG_HIGH_RES_TIMERS
749 /* Retrigger the CPU local events everywhere */
750 on_each_cpu(retrigger_next_event, NULL, 1);
751#endif
752 timerfd_clock_was_set();
753}
754
755/*
756 * During resume we might have to reprogram the high resolution timer
757 * interrupt on all online CPUs. However, all other CPUs will be
758 * stopped with IRQs interrupts disabled so the clock_was_set() call
759 * must be deferred.
760 */
761void hrtimers_resume(void)
762{
763 lockdep_assert_irqs_disabled();
764 /* Retrigger on the local CPU */
765 retrigger_next_event(NULL);
766 /* And schedule a retrigger for all others */
767 clock_was_set_delayed();
768}
769
770/*
771 * Counterpart to lock_hrtimer_base above:
772 */
773static inline
774void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
775{
776 raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
777}
778
779/**
780 * hrtimer_forward - forward the timer expiry
781 * @timer: hrtimer to forward
782 * @now: forward past this time
783 * @interval: the interval to forward
784 *
785 * Forward the timer expiry so it will expire in the future.
786 * Returns the number of overruns.
787 *
788 * Can be safely called from the callback function of @timer. If
789 * called from other contexts @timer must neither be enqueued nor
790 * running the callback and the caller needs to take care of
791 * serialization.
792 *
793 * Note: This only updates the timer expiry value and does not requeue
794 * the timer.
795 */
796u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
797{
798 u64 orun = 1;
799 ktime_t delta;
800
801 delta = ktime_sub(now, hrtimer_get_expires(timer));
802
803 if (delta < 0)
804 return 0;
805
806 if (WARN_ON(timer->state & HRTIMER_STATE_ENQUEUED))
807 return 0;
808
809 if (interval < hrtimer_resolution)
810 interval = hrtimer_resolution;
811
812 if (unlikely(delta >= interval)) {
813 s64 incr = ktime_to_ns(interval);
814
815 orun = ktime_divns(delta, incr);
816 hrtimer_add_expires_ns(timer, incr * orun);
817 if (hrtimer_get_expires_tv64(timer) > now)
818 return orun;
819 /*
820 * This (and the ktime_add() below) is the
821 * correction for exact:
822 */
823 orun++;
824 }
825 hrtimer_add_expires(timer, interval);
826
827 return orun;
828}
829EXPORT_SYMBOL_GPL(hrtimer_forward);
830
831/*
832 * enqueue_hrtimer - internal function to (re)start a timer
833 *
834 * The timer is inserted in expiry order. Insertion into the
835 * red black tree is O(log(n)). Must hold the base lock.
836 *
837 * Returns 1 when the new timer is the leftmost timer in the tree.
838 */
839static int enqueue_hrtimer(struct hrtimer *timer,
840 struct hrtimer_clock_base *base)
841{
842 debug_activate(timer);
843
844 base->cpu_base->active_bases |= 1 << base->index;
845
846 timer->state = HRTIMER_STATE_ENQUEUED;
847
848 return timerqueue_add(&base->active, &timer->node);
849}
850
851/*
852 * __remove_hrtimer - internal function to remove a timer
853 *
854 * Caller must hold the base lock.
855 *
856 * High resolution timer mode reprograms the clock event device when the
857 * timer is the one which expires next. The caller can disable this by setting
858 * reprogram to zero. This is useful, when the context does a reprogramming
859 * anyway (e.g. timer interrupt)
860 */
861static void __remove_hrtimer(struct hrtimer *timer,
862 struct hrtimer_clock_base *base,
863 u8 newstate, int reprogram)
864{
865 struct hrtimer_cpu_base *cpu_base = base->cpu_base;
866 u8 state = timer->state;
867
868 timer->state = newstate;
869 if (!(state & HRTIMER_STATE_ENQUEUED))
870 return;
871
872 if (!timerqueue_del(&base->active, &timer->node))
873 cpu_base->active_bases &= ~(1 << base->index);
874
875#ifdef CONFIG_HIGH_RES_TIMERS
876 /*
877 * Note: If reprogram is false we do not update
878 * cpu_base->next_timer. This happens when we remove the first
879 * timer on a remote cpu. No harm as we never dereference
880 * cpu_base->next_timer. So the worst thing what can happen is
881 * an superflous call to hrtimer_force_reprogram() on the
882 * remote cpu later on if the same timer gets enqueued again.
883 */
884 if (reprogram && timer == cpu_base->next_timer)
885 hrtimer_force_reprogram(cpu_base, 1);
886#endif
887}
888
889/*
890 * remove hrtimer, called with base lock held
891 */
892static inline int
893remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base, bool restart)
894{
895 if (hrtimer_is_queued(timer)) {
896 u8 state = timer->state;
897 int reprogram;
898
899 /*
900 * Remove the timer and force reprogramming when high
901 * resolution mode is active and the timer is on the current
902 * CPU. If we remove a timer on another CPU, reprogramming is
903 * skipped. The interrupt event on this CPU is fired and
904 * reprogramming happens in the interrupt handler. This is a
905 * rare case and less expensive than a smp call.
906 */
907 debug_deactivate(timer);
908 reprogram = base->cpu_base == this_cpu_ptr(&hrtimer_bases);
909
910 if (!restart)
911 state = HRTIMER_STATE_INACTIVE;
912
913 __remove_hrtimer(timer, base, state, reprogram);
914 return 1;
915 }
916 return 0;
917}
918
919static inline ktime_t hrtimer_update_lowres(struct hrtimer *timer, ktime_t tim,
920 const enum hrtimer_mode mode)
921{
922#ifdef CONFIG_TIME_LOW_RES
923 /*
924 * CONFIG_TIME_LOW_RES indicates that the system has no way to return
925 * granular time values. For relative timers we add hrtimer_resolution
926 * (i.e. one jiffie) to prevent short timeouts.
927 */
928 timer->is_rel = mode & HRTIMER_MODE_REL;
929 if (timer->is_rel)
930 tim = ktime_add_safe(tim, hrtimer_resolution);
931#endif
932 return tim;
933}
934
935/**
936 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
937 * @timer: the timer to be added
938 * @tim: expiry time
939 * @delta_ns: "slack" range for the timer
940 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
941 * relative (HRTIMER_MODE_REL)
942 */
943void hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
944 u64 delta_ns, const enum hrtimer_mode mode)
945{
946 struct hrtimer_clock_base *base, *new_base;
947 unsigned long flags;
948 int leftmost;
949
950 base = lock_hrtimer_base(timer, &flags);
951
952 /* Remove an active timer from the queue: */
953 remove_hrtimer(timer, base, true);
954
955 if (mode & HRTIMER_MODE_REL)
956 tim = ktime_add_safe(tim, base->get_time());
957
958 tim = hrtimer_update_lowres(timer, tim, mode);
959
960 hrtimer_set_expires_range_ns(timer, tim, delta_ns);
961
962 /* Switch the timer base, if necessary: */
963 new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED);
964
965 leftmost = enqueue_hrtimer(timer, new_base);
966 if (!leftmost)
967 goto unlock;
968
969 if (!hrtimer_is_hres_active(timer)) {
970 /*
971 * Kick to reschedule the next tick to handle the new timer
972 * on dynticks target.
973 */
974 if (new_base->cpu_base->nohz_active)
975 wake_up_nohz_cpu(new_base->cpu_base->cpu);
976 } else {
977 hrtimer_reprogram(timer, new_base);
978 }
979unlock:
980 unlock_hrtimer_base(timer, &flags);
981}
982EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
983
984/**
985 * hrtimer_try_to_cancel - try to deactivate a timer
986 * @timer: hrtimer to stop
987 *
988 * Returns:
989 * 0 when the timer was not active
990 * 1 when the timer was active
991 * -1 when the timer is currently executing the callback function and
992 * cannot be stopped
993 */
994int hrtimer_try_to_cancel(struct hrtimer *timer)
995{
996 struct hrtimer_clock_base *base;
997 unsigned long flags;
998 int ret = -1;
999
1000 /*
1001 * Check lockless first. If the timer is not active (neither
1002 * enqueued nor running the callback, nothing to do here. The
1003 * base lock does not serialize against a concurrent enqueue,
1004 * so we can avoid taking it.
1005 */
1006 if (!hrtimer_active(timer))
1007 return 0;
1008
1009 base = lock_hrtimer_base(timer, &flags);
1010
1011 if (!hrtimer_callback_running(timer))
1012 ret = remove_hrtimer(timer, base, false);
1013
1014 unlock_hrtimer_base(timer, &flags);
1015
1016 return ret;
1017
1018}
1019EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1020
1021/**
1022 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1023 * @timer: the timer to be cancelled
1024 *
1025 * Returns:
1026 * 0 when the timer was not active
1027 * 1 when the timer was active
1028 */
1029int hrtimer_cancel(struct hrtimer *timer)
1030{
1031 for (;;) {
1032 int ret = hrtimer_try_to_cancel(timer);
1033
1034 if (ret >= 0)
1035 return ret;
1036 cpu_relax();
1037 }
1038}
1039EXPORT_SYMBOL_GPL(hrtimer_cancel);
1040
1041/**
1042 * hrtimer_get_remaining - get remaining time for the timer
1043 * @timer: the timer to read
1044 * @adjust: adjust relative timers when CONFIG_TIME_LOW_RES=y
1045 */
1046ktime_t __hrtimer_get_remaining(const struct hrtimer *timer, bool adjust)
1047{
1048 unsigned long flags;
1049 ktime_t rem;
1050
1051 lock_hrtimer_base(timer, &flags);
1052 if (IS_ENABLED(CONFIG_TIME_LOW_RES) && adjust)
1053 rem = hrtimer_expires_remaining_adjusted(timer);
1054 else
1055 rem = hrtimer_expires_remaining(timer);
1056 unlock_hrtimer_base(timer, &flags);
1057
1058 return rem;
1059}
1060EXPORT_SYMBOL_GPL(__hrtimer_get_remaining);
1061
1062#ifdef CONFIG_NO_HZ_COMMON
1063/**
1064 * hrtimer_get_next_event - get the time until next expiry event
1065 *
1066 * Returns the next expiry time or KTIME_MAX if no timer is pending.
1067 */
1068u64 hrtimer_get_next_event(void)
1069{
1070 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1071 u64 expires = KTIME_MAX;
1072 unsigned long flags;
1073
1074 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1075
1076 if (!__hrtimer_hres_active(cpu_base))
1077 expires = __hrtimer_get_next_event(cpu_base);
1078
1079 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1080
1081 return expires;
1082}
1083#endif
1084
1085static inline int hrtimer_clockid_to_base(clockid_t clock_id)
1086{
1087 if (likely(clock_id < MAX_CLOCKS)) {
1088 int base = hrtimer_clock_to_base_table[clock_id];
1089
1090 if (likely(base != HRTIMER_MAX_CLOCK_BASES))
1091 return base;
1092 }
1093 WARN(1, "Invalid clockid %d. Using MONOTONIC\n", clock_id);
1094 return HRTIMER_BASE_MONOTONIC;
1095}
1096
1097static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1098 enum hrtimer_mode mode)
1099{
1100 struct hrtimer_cpu_base *cpu_base;
1101 int base;
1102
1103 memset(timer, 0, sizeof(struct hrtimer));
1104
1105 cpu_base = raw_cpu_ptr(&hrtimer_bases);
1106
1107 if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
1108 clock_id = CLOCK_MONOTONIC;
1109
1110 base = hrtimer_clockid_to_base(clock_id);
1111 timer->base = &cpu_base->clock_base[base];
1112 timerqueue_init(&timer->node);
1113}
1114
1115/**
1116 * hrtimer_init - initialize a timer to the given clock
1117 * @timer: the timer to be initialized
1118 * @clock_id: the clock to be used
1119 * @mode: timer mode abs/rel
1120 */
1121void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1122 enum hrtimer_mode mode)
1123{
1124 debug_init(timer, clock_id, mode);
1125 __hrtimer_init(timer, clock_id, mode);
1126}
1127EXPORT_SYMBOL_GPL(hrtimer_init);
1128
1129/*
1130 * A timer is active, when it is enqueued into the rbtree or the
1131 * callback function is running or it's in the state of being migrated
1132 * to another cpu.
1133 *
1134 * It is important for this function to not return a false negative.
1135 */
1136bool hrtimer_active(const struct hrtimer *timer)
1137{
1138 struct hrtimer_cpu_base *cpu_base;
1139 unsigned int seq;
1140
1141 do {
1142 cpu_base = READ_ONCE(timer->base->cpu_base);
1143 seq = raw_read_seqcount_begin(&cpu_base->seq);
1144
1145 if (timer->state != HRTIMER_STATE_INACTIVE ||
1146 cpu_base->running == timer)
1147 return true;
1148
1149 } while (read_seqcount_retry(&cpu_base->seq, seq) ||
1150 cpu_base != READ_ONCE(timer->base->cpu_base));
1151
1152 return false;
1153}
1154EXPORT_SYMBOL_GPL(hrtimer_active);
1155
1156/*
1157 * The write_seqcount_barrier()s in __run_hrtimer() split the thing into 3
1158 * distinct sections:
1159 *
1160 * - queued: the timer is queued
1161 * - callback: the timer is being ran
1162 * - post: the timer is inactive or (re)queued
1163 *
1164 * On the read side we ensure we observe timer->state and cpu_base->running
1165 * from the same section, if anything changed while we looked at it, we retry.
1166 * This includes timer->base changing because sequence numbers alone are
1167 * insufficient for that.
1168 *
1169 * The sequence numbers are required because otherwise we could still observe
1170 * a false negative if the read side got smeared over multiple consequtive
1171 * __run_hrtimer() invocations.
1172 */
1173
1174static void __run_hrtimer(struct hrtimer_cpu_base *cpu_base,
1175 struct hrtimer_clock_base *base,
1176 struct hrtimer *timer, ktime_t *now)
1177{
1178 enum hrtimer_restart (*fn)(struct hrtimer *);
1179 int restart;
1180
1181 lockdep_assert_held(&cpu_base->lock);
1182
1183 debug_deactivate(timer);
1184 cpu_base->running = timer;
1185
1186 /*
1187 * Separate the ->running assignment from the ->state assignment.
1188 *
1189 * As with a regular write barrier, this ensures the read side in
1190 * hrtimer_active() cannot observe cpu_base->running == NULL &&
1191 * timer->state == INACTIVE.
1192 */
1193 raw_write_seqcount_barrier(&cpu_base->seq);
1194
1195 __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE, 0);
1196 fn = timer->function;
1197
1198 /*
1199 * Clear the 'is relative' flag for the TIME_LOW_RES case. If the
1200 * timer is restarted with a period then it becomes an absolute
1201 * timer. If its not restarted it does not matter.
1202 */
1203 if (IS_ENABLED(CONFIG_TIME_LOW_RES))
1204 timer->is_rel = false;
1205
1206 /*
1207 * Because we run timers from hardirq context, there is no chance
1208 * they get migrated to another cpu, therefore its safe to unlock
1209 * the timer base.
1210 */
1211 raw_spin_unlock(&cpu_base->lock);
1212 trace_hrtimer_expire_entry(timer, now);
1213 restart = fn(timer);
1214 trace_hrtimer_expire_exit(timer);
1215 raw_spin_lock(&cpu_base->lock);
1216
1217 /*
1218 * Note: We clear the running state after enqueue_hrtimer and
1219 * we do not reprogram the event hardware. Happens either in
1220 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1221 *
1222 * Note: Because we dropped the cpu_base->lock above,
1223 * hrtimer_start_range_ns() can have popped in and enqueued the timer
1224 * for us already.
1225 */
1226 if (restart != HRTIMER_NORESTART &&
1227 !(timer->state & HRTIMER_STATE_ENQUEUED))
1228 enqueue_hrtimer(timer, base);
1229
1230 /*
1231 * Separate the ->running assignment from the ->state assignment.
1232 *
1233 * As with a regular write barrier, this ensures the read side in
1234 * hrtimer_active() cannot observe cpu_base->running == NULL &&
1235 * timer->state == INACTIVE.
1236 */
1237 raw_write_seqcount_barrier(&cpu_base->seq);
1238
1239 WARN_ON_ONCE(cpu_base->running != timer);
1240 cpu_base->running = NULL;
1241}
1242
1243static void __hrtimer_run_queues(struct hrtimer_cpu_base *cpu_base, ktime_t now)
1244{
1245 struct hrtimer_clock_base *base = cpu_base->clock_base;
1246 unsigned int active = cpu_base->active_bases;
1247
1248 for (; active; base++, active >>= 1) {
1249 struct timerqueue_node *node;
1250 ktime_t basenow;
1251
1252 if (!(active & 0x01))
1253 continue;
1254
1255 basenow = ktime_add(now, base->offset);
1256
1257 while ((node = timerqueue_getnext(&base->active))) {
1258 struct hrtimer *timer;
1259
1260 timer = container_of(node, struct hrtimer, node);
1261
1262 /*
1263 * The immediate goal for using the softexpires is
1264 * minimizing wakeups, not running timers at the
1265 * earliest interrupt after their soft expiration.
1266 * This allows us to avoid using a Priority Search
1267 * Tree, which can answer a stabbing querry for
1268 * overlapping intervals and instead use the simple
1269 * BST we already have.
1270 * We don't add extra wakeups by delaying timers that
1271 * are right-of a not yet expired timer, because that
1272 * timer will have to trigger a wakeup anyway.
1273 */
1274 if (basenow < hrtimer_get_softexpires_tv64(timer))
1275 break;
1276
1277 __run_hrtimer(cpu_base, base, timer, &basenow);
1278 }
1279 }
1280}
1281
1282#ifdef CONFIG_HIGH_RES_TIMERS
1283
1284/*
1285 * High resolution timer interrupt
1286 * Called with interrupts disabled
1287 */
1288void hrtimer_interrupt(struct clock_event_device *dev)
1289{
1290 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1291 ktime_t expires_next, now, entry_time, delta;
1292 int retries = 0;
1293
1294 BUG_ON(!cpu_base->hres_active);
1295 cpu_base->nr_events++;
1296 dev->next_event = KTIME_MAX;
1297
1298 raw_spin_lock(&cpu_base->lock);
1299 entry_time = now = hrtimer_update_base(cpu_base);
1300retry:
1301 cpu_base->in_hrtirq = 1;
1302 /*
1303 * We set expires_next to KTIME_MAX here with cpu_base->lock
1304 * held to prevent that a timer is enqueued in our queue via
1305 * the migration code. This does not affect enqueueing of
1306 * timers which run their callback and need to be requeued on
1307 * this CPU.
1308 */
1309 cpu_base->expires_next = KTIME_MAX;
1310
1311 __hrtimer_run_queues(cpu_base, now);
1312
1313 /* Reevaluate the clock bases for the next expiry */
1314 expires_next = __hrtimer_get_next_event(cpu_base);
1315 /*
1316 * Store the new expiry value so the migration code can verify
1317 * against it.
1318 */
1319 cpu_base->expires_next = expires_next;
1320 cpu_base->in_hrtirq = 0;
1321 raw_spin_unlock(&cpu_base->lock);
1322
1323 /* Reprogramming necessary ? */
1324 if (!tick_program_event(expires_next, 0)) {
1325 cpu_base->hang_detected = 0;
1326 return;
1327 }
1328
1329 /*
1330 * The next timer was already expired due to:
1331 * - tracing
1332 * - long lasting callbacks
1333 * - being scheduled away when running in a VM
1334 *
1335 * We need to prevent that we loop forever in the hrtimer
1336 * interrupt routine. We give it 3 attempts to avoid
1337 * overreacting on some spurious event.
1338 *
1339 * Acquire base lock for updating the offsets and retrieving
1340 * the current time.
1341 */
1342 raw_spin_lock(&cpu_base->lock);
1343 now = hrtimer_update_base(cpu_base);
1344 cpu_base->nr_retries++;
1345 if (++retries < 3)
1346 goto retry;
1347 /*
1348 * Give the system a chance to do something else than looping
1349 * here. We stored the entry time, so we know exactly how long
1350 * we spent here. We schedule the next event this amount of
1351 * time away.
1352 */
1353 cpu_base->nr_hangs++;
1354 cpu_base->hang_detected = 1;
1355 raw_spin_unlock(&cpu_base->lock);
1356 delta = ktime_sub(now, entry_time);
1357 if ((unsigned int)delta > cpu_base->max_hang_time)
1358 cpu_base->max_hang_time = (unsigned int) delta;
1359 /*
1360 * Limit it to a sensible value as we enforce a longer
1361 * delay. Give the CPU at least 100ms to catch up.
1362 */
1363 if (delta > 100 * NSEC_PER_MSEC)
1364 expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1365 else
1366 expires_next = ktime_add(now, delta);
1367 tick_program_event(expires_next, 1);
1368 printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n",
1369 ktime_to_ns(delta));
1370}
1371
1372/* called with interrupts disabled */
1373static inline void __hrtimer_peek_ahead_timers(void)
1374{
1375 struct tick_device *td;
1376
1377 if (!hrtimer_hres_active())
1378 return;
1379
1380 td = this_cpu_ptr(&tick_cpu_device);
1381 if (td && td->evtdev)
1382 hrtimer_interrupt(td->evtdev);
1383}
1384
1385#else /* CONFIG_HIGH_RES_TIMERS */
1386
1387static inline void __hrtimer_peek_ahead_timers(void) { }
1388
1389#endif /* !CONFIG_HIGH_RES_TIMERS */
1390
1391/*
1392 * Called from run_local_timers in hardirq context every jiffy
1393 */
1394void hrtimer_run_queues(void)
1395{
1396 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1397 ktime_t now;
1398
1399 if (__hrtimer_hres_active(cpu_base))
1400 return;
1401
1402 /*
1403 * This _is_ ugly: We have to check periodically, whether we
1404 * can switch to highres and / or nohz mode. The clocksource
1405 * switch happens with xtime_lock held. Notification from
1406 * there only sets the check bit in the tick_oneshot code,
1407 * otherwise we might deadlock vs. xtime_lock.
1408 */
1409 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) {
1410 hrtimer_switch_to_hres();
1411 return;
1412 }
1413
1414 raw_spin_lock(&cpu_base->lock);
1415 now = hrtimer_update_base(cpu_base);
1416 __hrtimer_run_queues(cpu_base, now);
1417 raw_spin_unlock(&cpu_base->lock);
1418}
1419
1420/*
1421 * Sleep related functions:
1422 */
1423static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1424{
1425 struct hrtimer_sleeper *t =
1426 container_of(timer, struct hrtimer_sleeper, timer);
1427 struct task_struct *task = t->task;
1428
1429 t->task = NULL;
1430 if (task)
1431 wake_up_process(task);
1432
1433 return HRTIMER_NORESTART;
1434}
1435
1436void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1437{
1438 sl->timer.function = hrtimer_wakeup;
1439 sl->task = task;
1440}
1441EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
1442
1443int nanosleep_copyout(struct restart_block *restart, struct timespec64 *ts)
1444{
1445 switch(restart->nanosleep.type) {
1446#ifdef CONFIG_COMPAT
1447 case TT_COMPAT:
1448 if (compat_put_timespec64(ts, restart->nanosleep.compat_rmtp))
1449 return -EFAULT;
1450 break;
1451#endif
1452 case TT_NATIVE:
1453 if (put_timespec64(ts, restart->nanosleep.rmtp))
1454 return -EFAULT;
1455 break;
1456 default:
1457 BUG();
1458 }
1459 return -ERESTART_RESTARTBLOCK;
1460}
1461
1462static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1463{
1464 struct restart_block *restart;
1465
1466 hrtimer_init_sleeper(t, current);
1467
1468 do {
1469 set_current_state(TASK_INTERRUPTIBLE);
1470 hrtimer_start_expires(&t->timer, mode);
1471
1472 if (likely(t->task))
1473 freezable_schedule();
1474
1475 hrtimer_cancel(&t->timer);
1476 mode = HRTIMER_MODE_ABS;
1477
1478 } while (t->task && !signal_pending(current));
1479
1480 __set_current_state(TASK_RUNNING);
1481
1482 if (!t->task)
1483 return 0;
1484
1485 restart = &current->restart_block;
1486 if (restart->nanosleep.type != TT_NONE) {
1487 ktime_t rem = hrtimer_expires_remaining(&t->timer);
1488 struct timespec64 rmt;
1489
1490 if (rem <= 0)
1491 return 0;
1492 rmt = ktime_to_timespec64(rem);
1493
1494 return nanosleep_copyout(restart, &rmt);
1495 }
1496 return -ERESTART_RESTARTBLOCK;
1497}
1498
1499static long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1500{
1501 struct hrtimer_sleeper t;
1502 int ret;
1503
1504 hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid,
1505 HRTIMER_MODE_ABS);
1506 hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
1507
1508 ret = do_nanosleep(&t, HRTIMER_MODE_ABS);
1509 destroy_hrtimer_on_stack(&t.timer);
1510 return ret;
1511}
1512
1513long hrtimer_nanosleep(const struct timespec64 *rqtp,
1514 const enum hrtimer_mode mode, const clockid_t clockid)
1515{
1516 struct restart_block *restart;
1517 struct hrtimer_sleeper t;
1518 int ret = 0;
1519 u64 slack;
1520
1521 slack = current->timer_slack_ns;
1522 if (dl_task(current) || rt_task(current))
1523 slack = 0;
1524
1525 hrtimer_init_on_stack(&t.timer, clockid, mode);
1526 hrtimer_set_expires_range_ns(&t.timer, timespec64_to_ktime(*rqtp), slack);
1527 ret = do_nanosleep(&t, mode);
1528 if (ret != -ERESTART_RESTARTBLOCK)
1529 goto out;
1530
1531 /* Absolute timers do not update the rmtp value and restart: */
1532 if (mode == HRTIMER_MODE_ABS) {
1533 ret = -ERESTARTNOHAND;
1534 goto out;
1535 }
1536
1537 restart = &current->restart_block;
1538 restart->fn = hrtimer_nanosleep_restart;
1539 restart->nanosleep.clockid = t.timer.base->clockid;
1540 restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
1541out:
1542 destroy_hrtimer_on_stack(&t.timer);
1543 return ret;
1544}
1545
1546SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp,
1547 struct timespec __user *, rmtp)
1548{
1549 struct timespec64 tu;
1550
1551 if (get_timespec64(&tu, rqtp))
1552 return -EFAULT;
1553
1554 if (!timespec64_valid(&tu))
1555 return -EINVAL;
1556
1557 current->restart_block.nanosleep.type = rmtp ? TT_NATIVE : TT_NONE;
1558 current->restart_block.nanosleep.rmtp = rmtp;
1559 return hrtimer_nanosleep(&tu, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1560}
1561
1562#ifdef CONFIG_COMPAT
1563
1564COMPAT_SYSCALL_DEFINE2(nanosleep, struct compat_timespec __user *, rqtp,
1565 struct compat_timespec __user *, rmtp)
1566{
1567 struct timespec64 tu;
1568
1569 if (compat_get_timespec64(&tu, rqtp))
1570 return -EFAULT;
1571
1572 if (!timespec64_valid(&tu))
1573 return -EINVAL;
1574
1575 current->restart_block.nanosleep.type = rmtp ? TT_COMPAT : TT_NONE;
1576 current->restart_block.nanosleep.compat_rmtp = rmtp;
1577 return hrtimer_nanosleep(&tu, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1578}
1579#endif
1580
1581/*
1582 * Functions related to boot-time initialization:
1583 */
1584int hrtimers_prepare_cpu(unsigned int cpu)
1585{
1586 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1587 int i;
1588
1589 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1590 cpu_base->clock_base[i].cpu_base = cpu_base;
1591 timerqueue_init_head(&cpu_base->clock_base[i].active);
1592 }
1593
1594 cpu_base->active_bases = 0;
1595 cpu_base->cpu = cpu;
1596 hrtimer_init_hres(cpu_base);
1597 return 0;
1598}
1599
1600#ifdef CONFIG_HOTPLUG_CPU
1601
1602static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1603 struct hrtimer_clock_base *new_base)
1604{
1605 struct hrtimer *timer;
1606 struct timerqueue_node *node;
1607
1608 while ((node = timerqueue_getnext(&old_base->active))) {
1609 timer = container_of(node, struct hrtimer, node);
1610 BUG_ON(hrtimer_callback_running(timer));
1611 debug_deactivate(timer);
1612
1613 /*
1614 * Mark it as ENQUEUED not INACTIVE otherwise the
1615 * timer could be seen as !active and just vanish away
1616 * under us on another CPU
1617 */
1618 __remove_hrtimer(timer, old_base, HRTIMER_STATE_ENQUEUED, 0);
1619 timer->base = new_base;
1620 /*
1621 * Enqueue the timers on the new cpu. This does not
1622 * reprogram the event device in case the timer
1623 * expires before the earliest on this CPU, but we run
1624 * hrtimer_interrupt after we migrated everything to
1625 * sort out already expired timers and reprogram the
1626 * event device.
1627 */
1628 enqueue_hrtimer(timer, new_base);
1629 }
1630}
1631
1632int hrtimers_dead_cpu(unsigned int scpu)
1633{
1634 struct hrtimer_cpu_base *old_base, *new_base;
1635 int i;
1636
1637 BUG_ON(cpu_online(scpu));
1638 tick_cancel_sched_timer(scpu);
1639
1640 local_irq_disable();
1641 old_base = &per_cpu(hrtimer_bases, scpu);
1642 new_base = this_cpu_ptr(&hrtimer_bases);
1643 /*
1644 * The caller is globally serialized and nobody else
1645 * takes two locks at once, deadlock is not possible.
1646 */
1647 raw_spin_lock(&new_base->lock);
1648 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1649
1650 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1651 migrate_hrtimer_list(&old_base->clock_base[i],
1652 &new_base->clock_base[i]);
1653 }
1654
1655 raw_spin_unlock(&old_base->lock);
1656 raw_spin_unlock(&new_base->lock);
1657
1658 /* Check, if we got expired work to do */
1659 __hrtimer_peek_ahead_timers();
1660 local_irq_enable();
1661 return 0;
1662}
1663
1664#endif /* CONFIG_HOTPLUG_CPU */
1665
1666void __init hrtimers_init(void)
1667{
1668 hrtimers_prepare_cpu(smp_processor_id());
1669}
1670
1671/**
1672 * schedule_hrtimeout_range_clock - sleep until timeout
1673 * @expires: timeout value (ktime_t)
1674 * @delta: slack in expires timeout (ktime_t)
1675 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1676 * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME
1677 */
1678int __sched
1679schedule_hrtimeout_range_clock(ktime_t *expires, u64 delta,
1680 const enum hrtimer_mode mode, int clock)
1681{
1682 struct hrtimer_sleeper t;
1683
1684 /*
1685 * Optimize when a zero timeout value is given. It does not
1686 * matter whether this is an absolute or a relative time.
1687 */
1688 if (expires && *expires == 0) {
1689 __set_current_state(TASK_RUNNING);
1690 return 0;
1691 }
1692
1693 /*
1694 * A NULL parameter means "infinite"
1695 */
1696 if (!expires) {
1697 schedule();
1698 return -EINTR;
1699 }
1700
1701 hrtimer_init_on_stack(&t.timer, clock, mode);
1702 hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
1703
1704 hrtimer_init_sleeper(&t, current);
1705
1706 hrtimer_start_expires(&t.timer, mode);
1707
1708 if (likely(t.task))
1709 schedule();
1710
1711 hrtimer_cancel(&t.timer);
1712 destroy_hrtimer_on_stack(&t.timer);
1713
1714 __set_current_state(TASK_RUNNING);
1715
1716 return !t.task ? 0 : -EINTR;
1717}
1718
1719/**
1720 * schedule_hrtimeout_range - sleep until timeout
1721 * @expires: timeout value (ktime_t)
1722 * @delta: slack in expires timeout (ktime_t)
1723 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1724 *
1725 * Make the current task sleep until the given expiry time has
1726 * elapsed. The routine will return immediately unless
1727 * the current task state has been set (see set_current_state()).
1728 *
1729 * The @delta argument gives the kernel the freedom to schedule the
1730 * actual wakeup to a time that is both power and performance friendly.
1731 * The kernel give the normal best effort behavior for "@expires+@delta",
1732 * but may decide to fire the timer earlier, but no earlier than @expires.
1733 *
1734 * You can set the task state as follows -
1735 *
1736 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1737 * pass before the routine returns unless the current task is explicitly
1738 * woken up, (e.g. by wake_up_process()).
1739 *
1740 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1741 * delivered to the current task or the current task is explicitly woken
1742 * up.
1743 *
1744 * The current task state is guaranteed to be TASK_RUNNING when this
1745 * routine returns.
1746 *
1747 * Returns 0 when the timer has expired. If the task was woken before the
1748 * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or
1749 * by an explicit wakeup, it returns -EINTR.
1750 */
1751int __sched schedule_hrtimeout_range(ktime_t *expires, u64 delta,
1752 const enum hrtimer_mode mode)
1753{
1754 return schedule_hrtimeout_range_clock(expires, delta, mode,
1755 CLOCK_MONOTONIC);
1756}
1757EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
1758
1759/**
1760 * schedule_hrtimeout - sleep until timeout
1761 * @expires: timeout value (ktime_t)
1762 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1763 *
1764 * Make the current task sleep until the given expiry time has
1765 * elapsed. The routine will return immediately unless
1766 * the current task state has been set (see set_current_state()).
1767 *
1768 * You can set the task state as follows -
1769 *
1770 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1771 * pass before the routine returns unless the current task is explicitly
1772 * woken up, (e.g. by wake_up_process()).
1773 *
1774 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1775 * delivered to the current task or the current task is explicitly woken
1776 * up.
1777 *
1778 * The current task state is guaranteed to be TASK_RUNNING when this
1779 * routine returns.
1780 *
1781 * Returns 0 when the timer has expired. If the task was woken before the
1782 * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or
1783 * by an explicit wakeup, it returns -EINTR.
1784 */
1785int __sched schedule_hrtimeout(ktime_t *expires,
1786 const enum hrtimer_mode mode)
1787{
1788 return schedule_hrtimeout_range(expires, 0, mode);
1789}
1790EXPORT_SYMBOL_GPL(schedule_hrtimeout);