projects/linux·blob v5.7 · refs

2218 lines · 60.7 KiB

1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright(C) 2005-2006, Thomas Gleixner <[email protected]>
4 * Copyright(C) 2005-2007, Red Hat, Inc., Ingo Molnar
5 * Copyright(C) 2006-2007 Timesys Corp., Thomas Gleixner
6 *
7 * High-resolution kernel timers
8 *
9 * In contrast to the low-resolution timeout API, aka timer wheel,
10 * hrtimers provide finer resolution and accuracy depending on system
11 * configuration and capabilities.
12 *
13 * Started by: Thomas Gleixner and Ingo Molnar
14 *
15 * Credits:
16 * Based on the original timer wheel code
17 *
18 * Help, testing, suggestions, bugfixes, improvements were
19 * provided by:
20 *
21 * George Anzinger, Andrew Morton, Steven Rostedt, Roman Zippel
22 * et. al.
23 */
24
25#include <linux/cpu.h>
26#include <linux/export.h>
27#include <linux/percpu.h>
28#include <linux/hrtimer.h>
29#include <linux/notifier.h>
30#include <linux/syscalls.h>
31#include <linux/interrupt.h>
32#include <linux/tick.h>
33#include <linux/err.h>
34#include <linux/debugobjects.h>
35#include <linux/sched/signal.h>
36#include <linux/sched/sysctl.h>
37#include <linux/sched/rt.h>
38#include <linux/sched/deadline.h>
39#include <linux/sched/nohz.h>
40#include <linux/sched/debug.h>
41#include <linux/timer.h>
42#include <linux/freezer.h>
43#include <linux/compat.h>
44
45#include <linux/uaccess.h>
46
47#include <trace/events/timer.h>
48
49#include "tick-internal.h"
50
51/*
52 * Masks for selecting the soft and hard context timers from
53 * cpu_base->active
54 */
55#define MASK_SHIFT (HRTIMER_BASE_MONOTONIC_SOFT)
56#define HRTIMER_ACTIVE_HARD ((1U << MASK_SHIFT) - 1)
57#define HRTIMER_ACTIVE_SOFT (HRTIMER_ACTIVE_HARD << MASK_SHIFT)
58#define HRTIMER_ACTIVE_ALL (HRTIMER_ACTIVE_SOFT | HRTIMER_ACTIVE_HARD)
59
60/*
61 * The timer bases:
62 *
63 * There are more clockids than hrtimer bases. Thus, we index
64 * into the timer bases by the hrtimer_base_type enum. When trying
65 * to reach a base using a clockid, hrtimer_clockid_to_base()
66 * is used to convert from clockid to the proper hrtimer_base_type.
67 */
68DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
69{
70 .lock = __RAW_SPIN_LOCK_UNLOCKED(hrtimer_bases.lock),
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 .index = HRTIMER_BASE_MONOTONIC_SOFT,
95 .clockid = CLOCK_MONOTONIC,
96 .get_time = &ktime_get,
97 },
98 {
99 .index = HRTIMER_BASE_REALTIME_SOFT,
100 .clockid = CLOCK_REALTIME,
101 .get_time = &ktime_get_real,
102 },
103 {
104 .index = HRTIMER_BASE_BOOTTIME_SOFT,
105 .clockid = CLOCK_BOOTTIME,
106 .get_time = &ktime_get_boottime,
107 },
108 {
109 .index = HRTIMER_BASE_TAI_SOFT,
110 .clockid = CLOCK_TAI,
111 .get_time = &ktime_get_clocktai,
112 },
113 }
114};
115
116static const int hrtimer_clock_to_base_table[MAX_CLOCKS] = {
117 /* Make sure we catch unsupported clockids */
118 [0 ... MAX_CLOCKS - 1] = HRTIMER_MAX_CLOCK_BASES,
119
120 [CLOCK_REALTIME] = HRTIMER_BASE_REALTIME,
121 [CLOCK_MONOTONIC] = HRTIMER_BASE_MONOTONIC,
122 [CLOCK_BOOTTIME] = HRTIMER_BASE_BOOTTIME,
123 [CLOCK_TAI] = HRTIMER_BASE_TAI,
124};
125
126/*
127 * Functions and macros which are different for UP/SMP systems are kept in a
128 * single place
129 */
130#ifdef CONFIG_SMP
131
132/*
133 * We require the migration_base for lock_hrtimer_base()/switch_hrtimer_base()
134 * such that hrtimer_callback_running() can unconditionally dereference
135 * timer->base->cpu_base
136 */
137static struct hrtimer_cpu_base migration_cpu_base = {
138 .clock_base = { { .cpu_base = &migration_cpu_base, }, },
139};
140
141#define migration_base migration_cpu_base.clock_base[0]
142
143static inline bool is_migration_base(struct hrtimer_clock_base *base)
144{
145 return base == &migration_base;
146}
147
148/*
149 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
150 * means that all timers which are tied to this base via timer->base are
151 * locked, and the base itself is locked too.
152 *
153 * So __run_timers/migrate_timers can safely modify all timers which could
154 * be found on the lists/queues.
155 *
156 * When the timer's base is locked, and the timer removed from list, it is
157 * possible to set timer->base = &migration_base and drop the lock: the timer
158 * remains locked.
159 */
160static
161struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
162 unsigned long *flags)
163{
164 struct hrtimer_clock_base *base;
165
166 for (;;) {
167 base = READ_ONCE(timer->base);
168 if (likely(base != &migration_base)) {
169 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
170 if (likely(base == timer->base))
171 return base;
172 /* The timer has migrated to another CPU: */
173 raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
174 }
175 cpu_relax();
176 }
177}
178
179/*
180 * We do not migrate the timer when it is expiring before the next
181 * event on the target cpu. When high resolution is enabled, we cannot
182 * reprogram the target cpu hardware and we would cause it to fire
183 * late. To keep it simple, we handle the high resolution enabled and
184 * disabled case similar.
185 *
186 * Called with cpu_base->lock of target cpu held.
187 */
188static int
189hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base)
190{
191 ktime_t expires;
192
193 expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset);
194 return expires < new_base->cpu_base->expires_next;
195}
196
197static inline
198struct hrtimer_cpu_base *get_target_base(struct hrtimer_cpu_base *base,
199 int pinned)
200{
201#if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
202 if (static_branch_likely(&timers_migration_enabled) && !pinned)
203 return &per_cpu(hrtimer_bases, get_nohz_timer_target());
204#endif
205 return base;
206}
207
208/*
209 * We switch the timer base to a power-optimized selected CPU target,
210 * if:
211 * - NO_HZ_COMMON is enabled
212 * - timer migration is enabled
213 * - the timer callback is not running
214 * - the timer is not the first expiring timer on the new target
215 *
216 * If one of the above requirements is not fulfilled we move the timer
217 * to the current CPU or leave it on the previously assigned CPU if
218 * the timer callback is currently running.
219 */
220static inline struct hrtimer_clock_base *
221switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
222 int pinned)
223{
224 struct hrtimer_cpu_base *new_cpu_base, *this_cpu_base;
225 struct hrtimer_clock_base *new_base;
226 int basenum = base->index;
227
228 this_cpu_base = this_cpu_ptr(&hrtimer_bases);
229 new_cpu_base = get_target_base(this_cpu_base, pinned);
230again:
231 new_base = &new_cpu_base->clock_base[basenum];
232
233 if (base != new_base) {
234 /*
235 * We are trying to move timer to new_base.
236 * However we can't change timer's base while it is running,
237 * so we keep it on the same CPU. No hassle vs. reprogramming
238 * the event source in the high resolution case. The softirq
239 * code will take care of this when the timer function has
240 * completed. There is no conflict as we hold the lock until
241 * the timer is enqueued.
242 */
243 if (unlikely(hrtimer_callback_running(timer)))
244 return base;
245
246 /* See the comment in lock_hrtimer_base() */
247 WRITE_ONCE(timer->base, &migration_base);
248 raw_spin_unlock(&base->cpu_base->lock);
249 raw_spin_lock(&new_base->cpu_base->lock);
250
251 if (new_cpu_base != this_cpu_base &&
252 hrtimer_check_target(timer, new_base)) {
253 raw_spin_unlock(&new_base->cpu_base->lock);
254 raw_spin_lock(&base->cpu_base->lock);
255 new_cpu_base = this_cpu_base;
256 WRITE_ONCE(timer->base, base);
257 goto again;
258 }
259 WRITE_ONCE(timer->base, new_base);
260 } else {
261 if (new_cpu_base != this_cpu_base &&
262 hrtimer_check_target(timer, new_base)) {
263 new_cpu_base = this_cpu_base;
264 goto again;
265 }
266 }
267 return new_base;
268}
269
270#else /* CONFIG_SMP */
271
272static inline bool is_migration_base(struct hrtimer_clock_base *base)
273{
274 return false;
275}
276
277static inline struct hrtimer_clock_base *
278lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
279{
280 struct hrtimer_clock_base *base = timer->base;
281
282 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
283
284 return base;
285}
286
287# define switch_hrtimer_base(t, b, p) (b)
288
289#endif /* !CONFIG_SMP */
290
291/*
292 * Functions for the union type storage format of ktime_t which are
293 * too large for inlining:
294 */
295#if BITS_PER_LONG < 64
296/*
297 * Divide a ktime value by a nanosecond value
298 */
299s64 __ktime_divns(const ktime_t kt, s64 div)
300{
301 int sft = 0;
302 s64 dclc;
303 u64 tmp;
304
305 dclc = ktime_to_ns(kt);
306 tmp = dclc < 0 ? -dclc : dclc;
307
308 /* Make sure the divisor is less than 2^32: */
309 while (div >> 32) {
310 sft++;
311 div >>= 1;
312 }
313 tmp >>= sft;
314 do_div(tmp, (u32) div);
315 return dclc < 0 ? -tmp : tmp;
316}
317EXPORT_SYMBOL_GPL(__ktime_divns);
318#endif /* BITS_PER_LONG >= 64 */
319
320/*
321 * Add two ktime values and do a safety check for overflow:
322 */
323ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
324{
325 ktime_t res = ktime_add_unsafe(lhs, rhs);
326
327 /*
328 * We use KTIME_SEC_MAX here, the maximum timeout which we can
329 * return to user space in a timespec:
330 */
331 if (res < 0 || res < lhs || res < rhs)
332 res = ktime_set(KTIME_SEC_MAX, 0);
333
334 return res;
335}
336
337EXPORT_SYMBOL_GPL(ktime_add_safe);
338
339#ifdef CONFIG_DEBUG_OBJECTS_TIMERS
340
341static struct debug_obj_descr hrtimer_debug_descr;
342
343static void *hrtimer_debug_hint(void *addr)
344{
345 return ((struct hrtimer *) addr)->function;
346}
347
348/*
349 * fixup_init is called when:
350 * - an active object is initialized
351 */
352static bool hrtimer_fixup_init(void *addr, enum debug_obj_state state)
353{
354 struct hrtimer *timer = addr;
355
356 switch (state) {
357 case ODEBUG_STATE_ACTIVE:
358 hrtimer_cancel(timer);
359 debug_object_init(timer, &hrtimer_debug_descr);
360 return true;
361 default:
362 return false;
363 }
364}
365
366/*
367 * fixup_activate is called when:
368 * - an active object is activated
369 * - an unknown non-static object is activated
370 */
371static bool hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
372{
373 switch (state) {
374 case ODEBUG_STATE_ACTIVE:
375 WARN_ON(1);
376 /* fall through */
377 default:
378 return false;
379 }
380}
381
382/*
383 * fixup_free is called when:
384 * - an active object is freed
385 */
386static bool hrtimer_fixup_free(void *addr, enum debug_obj_state state)
387{
388 struct hrtimer *timer = addr;
389
390 switch (state) {
391 case ODEBUG_STATE_ACTIVE:
392 hrtimer_cancel(timer);
393 debug_object_free(timer, &hrtimer_debug_descr);
394 return true;
395 default:
396 return false;
397 }
398}
399
400static struct debug_obj_descr hrtimer_debug_descr = {
401 .name = "hrtimer",
402 .debug_hint = hrtimer_debug_hint,
403 .fixup_init = hrtimer_fixup_init,
404 .fixup_activate = hrtimer_fixup_activate,
405 .fixup_free = hrtimer_fixup_free,
406};
407
408static inline void debug_hrtimer_init(struct hrtimer *timer)
409{
410 debug_object_init(timer, &hrtimer_debug_descr);
411}
412
413static inline void debug_hrtimer_activate(struct hrtimer *timer,
414 enum hrtimer_mode mode)
415{
416 debug_object_activate(timer, &hrtimer_debug_descr);
417}
418
419static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
420{
421 debug_object_deactivate(timer, &hrtimer_debug_descr);
422}
423
424static inline void debug_hrtimer_free(struct hrtimer *timer)
425{
426 debug_object_free(timer, &hrtimer_debug_descr);
427}
428
429static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
430 enum hrtimer_mode mode);
431
432void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
433 enum hrtimer_mode mode)
434{
435 debug_object_init_on_stack(timer, &hrtimer_debug_descr);
436 __hrtimer_init(timer, clock_id, mode);
437}
438EXPORT_SYMBOL_GPL(hrtimer_init_on_stack);
439
440static void __hrtimer_init_sleeper(struct hrtimer_sleeper *sl,
441 clockid_t clock_id, enum hrtimer_mode mode);
442
443void hrtimer_init_sleeper_on_stack(struct hrtimer_sleeper *sl,
444 clockid_t clock_id, enum hrtimer_mode mode)
445{
446 debug_object_init_on_stack(&sl->timer, &hrtimer_debug_descr);
447 __hrtimer_init_sleeper(sl, clock_id, mode);
448}
449EXPORT_SYMBOL_GPL(hrtimer_init_sleeper_on_stack);
450
451void destroy_hrtimer_on_stack(struct hrtimer *timer)
452{
453 debug_object_free(timer, &hrtimer_debug_descr);
454}
455EXPORT_SYMBOL_GPL(destroy_hrtimer_on_stack);
456
457#else
458
459static inline void debug_hrtimer_init(struct hrtimer *timer) { }
460static inline void debug_hrtimer_activate(struct hrtimer *timer,
461 enum hrtimer_mode mode) { }
462static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
463#endif
464
465static inline void
466debug_init(struct hrtimer *timer, clockid_t clockid,
467 enum hrtimer_mode mode)
468{
469 debug_hrtimer_init(timer);
470 trace_hrtimer_init(timer, clockid, mode);
471}
472
473static inline void debug_activate(struct hrtimer *timer,
474 enum hrtimer_mode mode)
475{
476 debug_hrtimer_activate(timer, mode);
477 trace_hrtimer_start(timer, mode);
478}
479
480static inline void debug_deactivate(struct hrtimer *timer)
481{
482 debug_hrtimer_deactivate(timer);
483 trace_hrtimer_cancel(timer);
484}
485
486static struct hrtimer_clock_base *
487__next_base(struct hrtimer_cpu_base *cpu_base, unsigned int *active)
488{
489 unsigned int idx;
490
491 if (!*active)
492 return NULL;
493
494 idx = __ffs(*active);
495 *active &= ~(1U << idx);
496
497 return &cpu_base->clock_base[idx];
498}
499
500#define for_each_active_base(base, cpu_base, active) \
501 while ((base = __next_base((cpu_base), &(active))))
502
503static ktime_t __hrtimer_next_event_base(struct hrtimer_cpu_base *cpu_base,
504 const struct hrtimer *exclude,
505 unsigned int active,
506 ktime_t expires_next)
507{
508 struct hrtimer_clock_base *base;
509 ktime_t expires;
510
511 for_each_active_base(base, cpu_base, active) {
512 struct timerqueue_node *next;
513 struct hrtimer *timer;
514
515 next = timerqueue_getnext(&base->active);
516 timer = container_of(next, struct hrtimer, node);
517 if (timer == exclude) {
518 /* Get to the next timer in the queue. */
519 next = timerqueue_iterate_next(next);
520 if (!next)
521 continue;
522
523 timer = container_of(next, struct hrtimer, node);
524 }
525 expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
526 if (expires < expires_next) {
527 expires_next = expires;
528
529 /* Skip cpu_base update if a timer is being excluded. */
530 if (exclude)
531 continue;
532
533 if (timer->is_soft)
534 cpu_base->softirq_next_timer = timer;
535 else
536 cpu_base->next_timer = timer;
537 }
538 }
539 /*
540 * clock_was_set() might have changed base->offset of any of
541 * the clock bases so the result might be negative. Fix it up
542 * to prevent a false positive in clockevents_program_event().
543 */
544 if (expires_next < 0)
545 expires_next = 0;
546 return expires_next;
547}
548
549/*
550 * Recomputes cpu_base::*next_timer and returns the earliest expires_next but
551 * does not set cpu_base::*expires_next, that is done by hrtimer_reprogram.
552 *
553 * When a softirq is pending, we can ignore the HRTIMER_ACTIVE_SOFT bases,
554 * those timers will get run whenever the softirq gets handled, at the end of
555 * hrtimer_run_softirq(), hrtimer_update_softirq_timer() will re-add these bases.
556 *
557 * Therefore softirq values are those from the HRTIMER_ACTIVE_SOFT clock bases.
558 * The !softirq values are the minima across HRTIMER_ACTIVE_ALL, unless an actual
559 * softirq is pending, in which case they're the minima of HRTIMER_ACTIVE_HARD.
560 *
561 * @active_mask must be one of:
562 * - HRTIMER_ACTIVE_ALL,
563 * - HRTIMER_ACTIVE_SOFT, or
564 * - HRTIMER_ACTIVE_HARD.
565 */
566static ktime_t
567__hrtimer_get_next_event(struct hrtimer_cpu_base *cpu_base, unsigned int active_mask)
568{
569 unsigned int active;
570 struct hrtimer *next_timer = NULL;
571 ktime_t expires_next = KTIME_MAX;
572
573 if (!cpu_base->softirq_activated && (active_mask & HRTIMER_ACTIVE_SOFT)) {
574 active = cpu_base->active_bases & HRTIMER_ACTIVE_SOFT;
575 cpu_base->softirq_next_timer = NULL;
576 expires_next = __hrtimer_next_event_base(cpu_base, NULL,
577 active, KTIME_MAX);
578
579 next_timer = cpu_base->softirq_next_timer;
580 }
581
582 if (active_mask & HRTIMER_ACTIVE_HARD) {
583 active = cpu_base->active_bases & HRTIMER_ACTIVE_HARD;
584 cpu_base->next_timer = next_timer;
585 expires_next = __hrtimer_next_event_base(cpu_base, NULL, active,
586 expires_next);
587 }
588
589 return expires_next;
590}
591
592static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base)
593{
594 ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset;
595 ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset;
596 ktime_t *offs_tai = &base->clock_base[HRTIMER_BASE_TAI].offset;
597
598 ktime_t now = ktime_get_update_offsets_now(&base->clock_was_set_seq,
599 offs_real, offs_boot, offs_tai);
600
601 base->clock_base[HRTIMER_BASE_REALTIME_SOFT].offset = *offs_real;
602 base->clock_base[HRTIMER_BASE_BOOTTIME_SOFT].offset = *offs_boot;
603 base->clock_base[HRTIMER_BASE_TAI_SOFT].offset = *offs_tai;
604
605 return now;
606}
607
608/*
609 * Is the high resolution mode active ?
610 */
611static inline int __hrtimer_hres_active(struct hrtimer_cpu_base *cpu_base)
612{
613 return IS_ENABLED(CONFIG_HIGH_RES_TIMERS) ?
614 cpu_base->hres_active : 0;
615}
616
617static inline int hrtimer_hres_active(void)
618{
619 return __hrtimer_hres_active(this_cpu_ptr(&hrtimer_bases));
620}
621
622/*
623 * Reprogram the event source with checking both queues for the
624 * next event
625 * Called with interrupts disabled and base->lock held
626 */
627static void
628hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
629{
630 ktime_t expires_next;
631
632 /*
633 * Find the current next expiration time.
634 */
635 expires_next = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_ALL);
636
637 if (cpu_base->next_timer && cpu_base->next_timer->is_soft) {
638 /*
639 * When the softirq is activated, hrtimer has to be
640 * programmed with the first hard hrtimer because soft
641 * timer interrupt could occur too late.
642 */
643 if (cpu_base->softirq_activated)
644 expires_next = __hrtimer_get_next_event(cpu_base,
645 HRTIMER_ACTIVE_HARD);
646 else
647 cpu_base->softirq_expires_next = expires_next;
648 }
649
650 if (skip_equal && expires_next == cpu_base->expires_next)
651 return;
652
653 cpu_base->expires_next = expires_next;
654
655 /*
656 * If hres is not active, hardware does not have to be
657 * reprogrammed yet.
658 *
659 * If a hang was detected in the last timer interrupt then we
660 * leave the hang delay active in the hardware. We want the
661 * system to make progress. That also prevents the following
662 * scenario:
663 * T1 expires 50ms from now
664 * T2 expires 5s from now
665 *
666 * T1 is removed, so this code is called and would reprogram
667 * the hardware to 5s from now. Any hrtimer_start after that
668 * will not reprogram the hardware due to hang_detected being
669 * set. So we'd effectivly block all timers until the T2 event
670 * fires.
671 */
672 if (!__hrtimer_hres_active(cpu_base) || cpu_base->hang_detected)
673 return;
674
675 tick_program_event(cpu_base->expires_next, 1);
676}
677
678/* High resolution timer related functions */
679#ifdef CONFIG_HIGH_RES_TIMERS
680
681/*
682 * High resolution timer enabled ?
683 */
684static bool hrtimer_hres_enabled __read_mostly = true;
685unsigned int hrtimer_resolution __read_mostly = LOW_RES_NSEC;
686EXPORT_SYMBOL_GPL(hrtimer_resolution);
687
688/*
689 * Enable / Disable high resolution mode
690 */
691static int __init setup_hrtimer_hres(char *str)
692{
693 return (kstrtobool(str, &hrtimer_hres_enabled) == 0);
694}
695
696__setup("highres=", setup_hrtimer_hres);
697
698/*
699 * hrtimer_high_res_enabled - query, if the highres mode is enabled
700 */
701static inline int hrtimer_is_hres_enabled(void)
702{
703 return hrtimer_hres_enabled;
704}
705
706/*
707 * Retrigger next event is called after clock was set
708 *
709 * Called with interrupts disabled via on_each_cpu()
710 */
711static void retrigger_next_event(void *arg)
712{
713 struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
714
715 if (!__hrtimer_hres_active(base))
716 return;
717
718 raw_spin_lock(&base->lock);
719 hrtimer_update_base(base);
720 hrtimer_force_reprogram(base, 0);
721 raw_spin_unlock(&base->lock);
722}
723
724/*
725 * Switch to high resolution mode
726 */
727static void hrtimer_switch_to_hres(void)
728{
729 struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
730
731 if (tick_init_highres()) {
732 pr_warn("Could not switch to high resolution mode on CPU %u\n",
733 base->cpu);
734 return;
735 }
736 base->hres_active = 1;
737 hrtimer_resolution = HIGH_RES_NSEC;
738
739 tick_setup_sched_timer();
740 /* "Retrigger" the interrupt to get things going */
741 retrigger_next_event(NULL);
742}
743
744static void clock_was_set_work(struct work_struct *work)
745{
746 clock_was_set();
747}
748
749static DECLARE_WORK(hrtimer_work, clock_was_set_work);
750
751/*
752 * Called from timekeeping and resume code to reprogram the hrtimer
753 * interrupt device on all cpus.
754 */
755void clock_was_set_delayed(void)
756{
757 schedule_work(&hrtimer_work);
758}
759
760#else
761
762static inline int hrtimer_is_hres_enabled(void) { return 0; }
763static inline void hrtimer_switch_to_hres(void) { }
764static inline void retrigger_next_event(void *arg) { }
765
766#endif /* CONFIG_HIGH_RES_TIMERS */
767
768/*
769 * When a timer is enqueued and expires earlier than the already enqueued
770 * timers, we have to check, whether it expires earlier than the timer for
771 * which the clock event device was armed.
772 *
773 * Called with interrupts disabled and base->cpu_base.lock held
774 */
775static void hrtimer_reprogram(struct hrtimer *timer, bool reprogram)
776{
777 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
778 struct hrtimer_clock_base *base = timer->base;
779 ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
780
781 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
782
783 /*
784 * CLOCK_REALTIME timer might be requested with an absolute
785 * expiry time which is less than base->offset. Set it to 0.
786 */
787 if (expires < 0)
788 expires = 0;
789
790 if (timer->is_soft) {
791 /*
792 * soft hrtimer could be started on a remote CPU. In this
793 * case softirq_expires_next needs to be updated on the
794 * remote CPU. The soft hrtimer will not expire before the
795 * first hard hrtimer on the remote CPU -
796 * hrtimer_check_target() prevents this case.
797 */
798 struct hrtimer_cpu_base *timer_cpu_base = base->cpu_base;
799
800 if (timer_cpu_base->softirq_activated)
801 return;
802
803 if (!ktime_before(expires, timer_cpu_base->softirq_expires_next))
804 return;
805
806 timer_cpu_base->softirq_next_timer = timer;
807 timer_cpu_base->softirq_expires_next = expires;
808
809 if (!ktime_before(expires, timer_cpu_base->expires_next) ||
810 !reprogram)
811 return;
812 }
813
814 /*
815 * If the timer is not on the current cpu, we cannot reprogram
816 * the other cpus clock event device.
817 */
818 if (base->cpu_base != cpu_base)
819 return;
820
821 /*
822 * If the hrtimer interrupt is running, then it will
823 * reevaluate the clock bases and reprogram the clock event
824 * device. The callbacks are always executed in hard interrupt
825 * context so we don't need an extra check for a running
826 * callback.
827 */
828 if (cpu_base->in_hrtirq)
829 return;
830
831 if (expires >= cpu_base->expires_next)
832 return;
833
834 /* Update the pointer to the next expiring timer */
835 cpu_base->next_timer = timer;
836 cpu_base->expires_next = expires;
837
838 /*
839 * If hres is not active, hardware does not have to be
840 * programmed yet.
841 *
842 * If a hang was detected in the last timer interrupt then we
843 * do not schedule a timer which is earlier than the expiry
844 * which we enforced in the hang detection. We want the system
845 * to make progress.
846 */
847 if (!__hrtimer_hres_active(cpu_base) || cpu_base->hang_detected)
848 return;
849
850 /*
851 * Program the timer hardware. We enforce the expiry for
852 * events which are already in the past.
853 */
854 tick_program_event(expires, 1);
855}
856
857/*
858 * Clock realtime was set
859 *
860 * Change the offset of the realtime clock vs. the monotonic
861 * clock.
862 *
863 * We might have to reprogram the high resolution timer interrupt. On
864 * SMP we call the architecture specific code to retrigger _all_ high
865 * resolution timer interrupts. On UP we just disable interrupts and
866 * call the high resolution interrupt code.
867 */
868void clock_was_set(void)
869{
870#ifdef CONFIG_HIGH_RES_TIMERS
871 /* Retrigger the CPU local events everywhere */
872 on_each_cpu(retrigger_next_event, NULL, 1);
873#endif
874 timerfd_clock_was_set();
875}
876
877/*
878 * During resume we might have to reprogram the high resolution timer
879 * interrupt on all online CPUs. However, all other CPUs will be
880 * stopped with IRQs interrupts disabled so the clock_was_set() call
881 * must be deferred.
882 */
883void hrtimers_resume(void)
884{
885 lockdep_assert_irqs_disabled();
886 /* Retrigger on the local CPU */
887 retrigger_next_event(NULL);
888 /* And schedule a retrigger for all others */
889 clock_was_set_delayed();
890}
891
892/*
893 * Counterpart to lock_hrtimer_base above:
894 */
895static inline
896void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
897{
898 raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
899}
900
901/**
902 * hrtimer_forward - forward the timer expiry
903 * @timer: hrtimer to forward
904 * @now: forward past this time
905 * @interval: the interval to forward
906 *
907 * Forward the timer expiry so it will expire in the future.
908 * Returns the number of overruns.
909 *
910 * Can be safely called from the callback function of @timer. If
911 * called from other contexts @timer must neither be enqueued nor
912 * running the callback and the caller needs to take care of
913 * serialization.
914 *
915 * Note: This only updates the timer expiry value and does not requeue
916 * the timer.
917 */
918u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
919{
920 u64 orun = 1;
921 ktime_t delta;
922
923 delta = ktime_sub(now, hrtimer_get_expires(timer));
924
925 if (delta < 0)
926 return 0;
927
928 if (WARN_ON(timer->state & HRTIMER_STATE_ENQUEUED))
929 return 0;
930
931 if (interval < hrtimer_resolution)
932 interval = hrtimer_resolution;
933
934 if (unlikely(delta >= interval)) {
935 s64 incr = ktime_to_ns(interval);
936
937 orun = ktime_divns(delta, incr);
938 hrtimer_add_expires_ns(timer, incr * orun);
939 if (hrtimer_get_expires_tv64(timer) > now)
940 return orun;
941 /*
942 * This (and the ktime_add() below) is the
943 * correction for exact:
944 */
945 orun++;
946 }
947 hrtimer_add_expires(timer, interval);
948
949 return orun;
950}
951EXPORT_SYMBOL_GPL(hrtimer_forward);
952
953/*
954 * enqueue_hrtimer - internal function to (re)start a timer
955 *
956 * The timer is inserted in expiry order. Insertion into the
957 * red black tree is O(log(n)). Must hold the base lock.
958 *
959 * Returns 1 when the new timer is the leftmost timer in the tree.
960 */
961static int enqueue_hrtimer(struct hrtimer *timer,
962 struct hrtimer_clock_base *base,
963 enum hrtimer_mode mode)
964{
965 debug_activate(timer, mode);
966
967 base->cpu_base->active_bases |= 1 << base->index;
968
969 /* Pairs with the lockless read in hrtimer_is_queued() */
970 WRITE_ONCE(timer->state, HRTIMER_STATE_ENQUEUED);
971
972 return timerqueue_add(&base->active, &timer->node);
973}
974
975/*
976 * __remove_hrtimer - internal function to remove a timer
977 *
978 * Caller must hold the base lock.
979 *
980 * High resolution timer mode reprograms the clock event device when the
981 * timer is the one which expires next. The caller can disable this by setting
982 * reprogram to zero. This is useful, when the context does a reprogramming
983 * anyway (e.g. timer interrupt)
984 */
985static void __remove_hrtimer(struct hrtimer *timer,
986 struct hrtimer_clock_base *base,
987 u8 newstate, int reprogram)
988{
989 struct hrtimer_cpu_base *cpu_base = base->cpu_base;
990 u8 state = timer->state;
991
992 /* Pairs with the lockless read in hrtimer_is_queued() */
993 WRITE_ONCE(timer->state, newstate);
994 if (!(state & HRTIMER_STATE_ENQUEUED))
995 return;
996
997 if (!timerqueue_del(&base->active, &timer->node))
998 cpu_base->active_bases &= ~(1 << base->index);
999
1000 /*
1001 * Note: If reprogram is false we do not update
1002 * cpu_base->next_timer. This happens when we remove the first
1003 * timer on a remote cpu. No harm as we never dereference
1004 * cpu_base->next_timer. So the worst thing what can happen is
1005 * an superflous call to hrtimer_force_reprogram() on the
1006 * remote cpu later on if the same timer gets enqueued again.
1007 */
1008 if (reprogram && timer == cpu_base->next_timer)
1009 hrtimer_force_reprogram(cpu_base, 1);
1010}
1011
1012/*
1013 * remove hrtimer, called with base lock held
1014 */
1015static inline int
1016remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base, bool restart)
1017{
1018 u8 state = timer->state;
1019
1020 if (state & HRTIMER_STATE_ENQUEUED) {
1021 int reprogram;
1022
1023 /*
1024 * Remove the timer and force reprogramming when high
1025 * resolution mode is active and the timer is on the current
1026 * CPU. If we remove a timer on another CPU, reprogramming is
1027 * skipped. The interrupt event on this CPU is fired and
1028 * reprogramming happens in the interrupt handler. This is a
1029 * rare case and less expensive than a smp call.
1030 */
1031 debug_deactivate(timer);
1032 reprogram = base->cpu_base == this_cpu_ptr(&hrtimer_bases);
1033
1034 if (!restart)
1035 state = HRTIMER_STATE_INACTIVE;
1036
1037 __remove_hrtimer(timer, base, state, reprogram);
1038 return 1;
1039 }
1040 return 0;
1041}
1042
1043static inline ktime_t hrtimer_update_lowres(struct hrtimer *timer, ktime_t tim,
1044 const enum hrtimer_mode mode)
1045{
1046#ifdef CONFIG_TIME_LOW_RES
1047 /*
1048 * CONFIG_TIME_LOW_RES indicates that the system has no way to return
1049 * granular time values. For relative timers we add hrtimer_resolution
1050 * (i.e. one jiffie) to prevent short timeouts.
1051 */
1052 timer->is_rel = mode & HRTIMER_MODE_REL;
1053 if (timer->is_rel)
1054 tim = ktime_add_safe(tim, hrtimer_resolution);
1055#endif
1056 return tim;
1057}
1058
1059static void
1060hrtimer_update_softirq_timer(struct hrtimer_cpu_base *cpu_base, bool reprogram)
1061{
1062 ktime_t expires;
1063
1064 /*
1065 * Find the next SOFT expiration.
1066 */
1067 expires = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_SOFT);
1068
1069 /*
1070 * reprogramming needs to be triggered, even if the next soft
1071 * hrtimer expires at the same time than the next hard
1072 * hrtimer. cpu_base->softirq_expires_next needs to be updated!
1073 */
1074 if (expires == KTIME_MAX)
1075 return;
1076
1077 /*
1078 * cpu_base->*next_timer is recomputed by __hrtimer_get_next_event()
1079 * cpu_base->*expires_next is only set by hrtimer_reprogram()
1080 */
1081 hrtimer_reprogram(cpu_base->softirq_next_timer, reprogram);
1082}
1083
1084static int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1085 u64 delta_ns, const enum hrtimer_mode mode,
1086 struct hrtimer_clock_base *base)
1087{
1088 struct hrtimer_clock_base *new_base;
1089
1090 /* Remove an active timer from the queue: */
1091 remove_hrtimer(timer, base, true);
1092
1093 if (mode & HRTIMER_MODE_REL)
1094 tim = ktime_add_safe(tim, base->get_time());
1095
1096 tim = hrtimer_update_lowres(timer, tim, mode);
1097
1098 hrtimer_set_expires_range_ns(timer, tim, delta_ns);
1099
1100 /* Switch the timer base, if necessary: */
1101 new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED);
1102
1103 return enqueue_hrtimer(timer, new_base, mode);
1104}
1105
1106/**
1107 * hrtimer_start_range_ns - (re)start an hrtimer
1108 * @timer: the timer to be added
1109 * @tim: expiry time
1110 * @delta_ns: "slack" range for the timer
1111 * @mode: timer mode: absolute (HRTIMER_MODE_ABS) or
1112 * relative (HRTIMER_MODE_REL), and pinned (HRTIMER_MODE_PINNED);
1113 * softirq based mode is considered for debug purpose only!
1114 */
1115void hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1116 u64 delta_ns, const enum hrtimer_mode mode)
1117{
1118 struct hrtimer_clock_base *base;
1119 unsigned long flags;
1120
1121 /*
1122 * Check whether the HRTIMER_MODE_SOFT bit and hrtimer.is_soft
1123 * match on CONFIG_PREEMPT_RT = n. With PREEMPT_RT check the hard
1124 * expiry mode because unmarked timers are moved to softirq expiry.
1125 */
1126 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
1127 WARN_ON_ONCE(!(mode & HRTIMER_MODE_SOFT) ^ !timer->is_soft);
1128 else
1129 WARN_ON_ONCE(!(mode & HRTIMER_MODE_HARD) ^ !timer->is_hard);
1130
1131 base = lock_hrtimer_base(timer, &flags);
1132
1133 if (__hrtimer_start_range_ns(timer, tim, delta_ns, mode, base))
1134 hrtimer_reprogram(timer, true);
1135
1136 unlock_hrtimer_base(timer, &flags);
1137}
1138EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
1139
1140/**
1141 * hrtimer_try_to_cancel - try to deactivate a timer
1142 * @timer: hrtimer to stop
1143 *
1144 * Returns:
1145 *
1146 * * 0 when the timer was not active
1147 * * 1 when the timer was active
1148 * * -1 when the timer is currently executing the callback function and
1149 * cannot be stopped
1150 */
1151int hrtimer_try_to_cancel(struct hrtimer *timer)
1152{
1153 struct hrtimer_clock_base *base;
1154 unsigned long flags;
1155 int ret = -1;
1156
1157 /*
1158 * Check lockless first. If the timer is not active (neither
1159 * enqueued nor running the callback, nothing to do here. The
1160 * base lock does not serialize against a concurrent enqueue,
1161 * so we can avoid taking it.
1162 */
1163 if (!hrtimer_active(timer))
1164 return 0;
1165
1166 base = lock_hrtimer_base(timer, &flags);
1167
1168 if (!hrtimer_callback_running(timer))
1169 ret = remove_hrtimer(timer, base, false);
1170
1171 unlock_hrtimer_base(timer, &flags);
1172
1173 return ret;
1174
1175}
1176EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1177
1178#ifdef CONFIG_PREEMPT_RT
1179static void hrtimer_cpu_base_init_expiry_lock(struct hrtimer_cpu_base *base)
1180{
1181 spin_lock_init(&base->softirq_expiry_lock);
1182}
1183
1184static void hrtimer_cpu_base_lock_expiry(struct hrtimer_cpu_base *base)
1185{
1186 spin_lock(&base->softirq_expiry_lock);
1187}
1188
1189static void hrtimer_cpu_base_unlock_expiry(struct hrtimer_cpu_base *base)
1190{
1191 spin_unlock(&base->softirq_expiry_lock);
1192}
1193
1194/*
1195 * The counterpart to hrtimer_cancel_wait_running().
1196 *
1197 * If there is a waiter for cpu_base->expiry_lock, then it was waiting for
1198 * the timer callback to finish. Drop expiry_lock and reaquire it. That
1199 * allows the waiter to acquire the lock and make progress.
1200 */
1201static void hrtimer_sync_wait_running(struct hrtimer_cpu_base *cpu_base,
1202 unsigned long flags)
1203{
1204 if (atomic_read(&cpu_base->timer_waiters)) {
1205 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1206 spin_unlock(&cpu_base->softirq_expiry_lock);
1207 spin_lock(&cpu_base->softirq_expiry_lock);
1208 raw_spin_lock_irq(&cpu_base->lock);
1209 }
1210}
1211
1212/*
1213 * This function is called on PREEMPT_RT kernels when the fast path
1214 * deletion of a timer failed because the timer callback function was
1215 * running.
1216 *
1217 * This prevents priority inversion: if the soft irq thread is preempted
1218 * in the middle of a timer callback, then calling del_timer_sync() can
1219 * lead to two issues:
1220 *
1221 * - If the caller is on a remote CPU then it has to spin wait for the timer
1222 * handler to complete. This can result in unbound priority inversion.
1223 *
1224 * - If the caller originates from the task which preempted the timer
1225 * handler on the same CPU, then spin waiting for the timer handler to
1226 * complete is never going to end.
1227 */
1228void hrtimer_cancel_wait_running(const struct hrtimer *timer)
1229{
1230 /* Lockless read. Prevent the compiler from reloading it below */
1231 struct hrtimer_clock_base *base = READ_ONCE(timer->base);
1232
1233 /*
1234 * Just relax if the timer expires in hard interrupt context or if
1235 * it is currently on the migration base.
1236 */
1237 if (!timer->is_soft || is_migration_base(base)) {
1238 cpu_relax();
1239 return;
1240 }
1241
1242 /*
1243 * Mark the base as contended and grab the expiry lock, which is
1244 * held by the softirq across the timer callback. Drop the lock
1245 * immediately so the softirq can expire the next timer. In theory
1246 * the timer could already be running again, but that's more than
1247 * unlikely and just causes another wait loop.
1248 */
1249 atomic_inc(&base->cpu_base->timer_waiters);
1250 spin_lock_bh(&base->cpu_base->softirq_expiry_lock);
1251 atomic_dec(&base->cpu_base->timer_waiters);
1252 spin_unlock_bh(&base->cpu_base->softirq_expiry_lock);
1253}
1254#else
1255static inline void
1256hrtimer_cpu_base_init_expiry_lock(struct hrtimer_cpu_base *base) { }
1257static inline void
1258hrtimer_cpu_base_lock_expiry(struct hrtimer_cpu_base *base) { }
1259static inline void
1260hrtimer_cpu_base_unlock_expiry(struct hrtimer_cpu_base *base) { }
1261static inline void hrtimer_sync_wait_running(struct hrtimer_cpu_base *base,
1262 unsigned long flags) { }
1263#endif
1264
1265/**
1266 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1267 * @timer: the timer to be cancelled
1268 *
1269 * Returns:
1270 * 0 when the timer was not active
1271 * 1 when the timer was active
1272 */
1273int hrtimer_cancel(struct hrtimer *timer)
1274{
1275 int ret;
1276
1277 do {
1278 ret = hrtimer_try_to_cancel(timer);
1279
1280 if (ret < 0)
1281 hrtimer_cancel_wait_running(timer);
1282 } while (ret < 0);
1283 return ret;
1284}
1285EXPORT_SYMBOL_GPL(hrtimer_cancel);
1286
1287/**
1288 * hrtimer_get_remaining - get remaining time for the timer
1289 * @timer: the timer to read
1290 * @adjust: adjust relative timers when CONFIG_TIME_LOW_RES=y
1291 */
1292ktime_t __hrtimer_get_remaining(const struct hrtimer *timer, bool adjust)
1293{
1294 unsigned long flags;
1295 ktime_t rem;
1296
1297 lock_hrtimer_base(timer, &flags);
1298 if (IS_ENABLED(CONFIG_TIME_LOW_RES) && adjust)
1299 rem = hrtimer_expires_remaining_adjusted(timer);
1300 else
1301 rem = hrtimer_expires_remaining(timer);
1302 unlock_hrtimer_base(timer, &flags);
1303
1304 return rem;
1305}
1306EXPORT_SYMBOL_GPL(__hrtimer_get_remaining);
1307
1308#ifdef CONFIG_NO_HZ_COMMON
1309/**
1310 * hrtimer_get_next_event - get the time until next expiry event
1311 *
1312 * Returns the next expiry time or KTIME_MAX if no timer is pending.
1313 */
1314u64 hrtimer_get_next_event(void)
1315{
1316 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1317 u64 expires = KTIME_MAX;
1318 unsigned long flags;
1319
1320 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1321
1322 if (!__hrtimer_hres_active(cpu_base))
1323 expires = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_ALL);
1324
1325 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1326
1327 return expires;
1328}
1329
1330/**
1331 * hrtimer_next_event_without - time until next expiry event w/o one timer
1332 * @exclude: timer to exclude
1333 *
1334 * Returns the next expiry time over all timers except for the @exclude one or
1335 * KTIME_MAX if none of them is pending.
1336 */
1337u64 hrtimer_next_event_without(const struct hrtimer *exclude)
1338{
1339 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1340 u64 expires = KTIME_MAX;
1341 unsigned long flags;
1342
1343 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1344
1345 if (__hrtimer_hres_active(cpu_base)) {
1346 unsigned int active;
1347
1348 if (!cpu_base->softirq_activated) {
1349 active = cpu_base->active_bases & HRTIMER_ACTIVE_SOFT;
1350 expires = __hrtimer_next_event_base(cpu_base, exclude,
1351 active, KTIME_MAX);
1352 }
1353 active = cpu_base->active_bases & HRTIMER_ACTIVE_HARD;
1354 expires = __hrtimer_next_event_base(cpu_base, exclude, active,
1355 expires);
1356 }
1357
1358 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1359
1360 return expires;
1361}
1362#endif
1363
1364static inline int hrtimer_clockid_to_base(clockid_t clock_id)
1365{
1366 if (likely(clock_id < MAX_CLOCKS)) {
1367 int base = hrtimer_clock_to_base_table[clock_id];
1368
1369 if (likely(base != HRTIMER_MAX_CLOCK_BASES))
1370 return base;
1371 }
1372 WARN(1, "Invalid clockid %d. Using MONOTONIC\n", clock_id);
1373 return HRTIMER_BASE_MONOTONIC;
1374}
1375
1376static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1377 enum hrtimer_mode mode)
1378{
1379 bool softtimer = !!(mode & HRTIMER_MODE_SOFT);
1380 struct hrtimer_cpu_base *cpu_base;
1381 int base;
1382
1383 /*
1384 * On PREEMPT_RT enabled kernels hrtimers which are not explicitely
1385 * marked for hard interrupt expiry mode are moved into soft
1386 * interrupt context for latency reasons and because the callbacks
1387 * can invoke functions which might sleep on RT, e.g. spin_lock().
1388 */
1389 if (IS_ENABLED(CONFIG_PREEMPT_RT) && !(mode & HRTIMER_MODE_HARD))
1390 softtimer = true;
1391
1392 memset(timer, 0, sizeof(struct hrtimer));
1393
1394 cpu_base = raw_cpu_ptr(&hrtimer_bases);
1395
1396 /*
1397 * POSIX magic: Relative CLOCK_REALTIME timers are not affected by
1398 * clock modifications, so they needs to become CLOCK_MONOTONIC to
1399 * ensure POSIX compliance.
1400 */
1401 if (clock_id == CLOCK_REALTIME && mode & HRTIMER_MODE_REL)
1402 clock_id = CLOCK_MONOTONIC;
1403
1404 base = softtimer ? HRTIMER_MAX_CLOCK_BASES / 2 : 0;
1405 base += hrtimer_clockid_to_base(clock_id);
1406 timer->is_soft = softtimer;
1407 timer->is_hard = !!(mode & HRTIMER_MODE_HARD);
1408 timer->base = &cpu_base->clock_base[base];
1409 timerqueue_init(&timer->node);
1410}
1411
1412/**
1413 * hrtimer_init - initialize a timer to the given clock
1414 * @timer: the timer to be initialized
1415 * @clock_id: the clock to be used
1416 * @mode: The modes which are relevant for intitialization:
1417 * HRTIMER_MODE_ABS, HRTIMER_MODE_REL, HRTIMER_MODE_ABS_SOFT,
1418 * HRTIMER_MODE_REL_SOFT
1419 *
1420 * The PINNED variants of the above can be handed in,
1421 * but the PINNED bit is ignored as pinning happens
1422 * when the hrtimer is started
1423 */
1424void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1425 enum hrtimer_mode mode)
1426{
1427 debug_init(timer, clock_id, mode);
1428 __hrtimer_init(timer, clock_id, mode);
1429}
1430EXPORT_SYMBOL_GPL(hrtimer_init);
1431
1432/*
1433 * A timer is active, when it is enqueued into the rbtree or the
1434 * callback function is running or it's in the state of being migrated
1435 * to another cpu.
1436 *
1437 * It is important for this function to not return a false negative.
1438 */
1439bool hrtimer_active(const struct hrtimer *timer)
1440{
1441 struct hrtimer_clock_base *base;
1442 unsigned int seq;
1443
1444 do {
1445 base = READ_ONCE(timer->base);
1446 seq = raw_read_seqcount_begin(&base->seq);
1447
1448 if (timer->state != HRTIMER_STATE_INACTIVE ||
1449 base->running == timer)
1450 return true;
1451
1452 } while (read_seqcount_retry(&base->seq, seq) ||
1453 base != READ_ONCE(timer->base));
1454
1455 return false;
1456}
1457EXPORT_SYMBOL_GPL(hrtimer_active);
1458
1459/*
1460 * The write_seqcount_barrier()s in __run_hrtimer() split the thing into 3
1461 * distinct sections:
1462 *
1463 * - queued: the timer is queued
1464 * - callback: the timer is being ran
1465 * - post: the timer is inactive or (re)queued
1466 *
1467 * On the read side we ensure we observe timer->state and cpu_base->running
1468 * from the same section, if anything changed while we looked at it, we retry.
1469 * This includes timer->base changing because sequence numbers alone are
1470 * insufficient for that.
1471 *
1472 * The sequence numbers are required because otherwise we could still observe
1473 * a false negative if the read side got smeared over multiple consequtive
1474 * __run_hrtimer() invocations.
1475 */
1476
1477static void __run_hrtimer(struct hrtimer_cpu_base *cpu_base,
1478 struct hrtimer_clock_base *base,
1479 struct hrtimer *timer, ktime_t *now,
1480 unsigned long flags) __must_hold(&cpu_base->lock)
1481{
1482 enum hrtimer_restart (*fn)(struct hrtimer *);
1483 bool expires_in_hardirq;
1484 int restart;
1485
1486 lockdep_assert_held(&cpu_base->lock);
1487
1488 debug_deactivate(timer);
1489 base->running = timer;
1490
1491 /*
1492 * Separate the ->running assignment from the ->state assignment.
1493 *
1494 * As with a regular write barrier, this ensures the read side in
1495 * hrtimer_active() cannot observe base->running == NULL &&
1496 * timer->state == INACTIVE.
1497 */
1498 raw_write_seqcount_barrier(&base->seq);
1499
1500 __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE, 0);
1501 fn = timer->function;
1502
1503 /*
1504 * Clear the 'is relative' flag for the TIME_LOW_RES case. If the
1505 * timer is restarted with a period then it becomes an absolute
1506 * timer. If its not restarted it does not matter.
1507 */
1508 if (IS_ENABLED(CONFIG_TIME_LOW_RES))
1509 timer->is_rel = false;
1510
1511 /*
1512 * The timer is marked as running in the CPU base, so it is
1513 * protected against migration to a different CPU even if the lock
1514 * is dropped.
1515 */
1516 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1517 trace_hrtimer_expire_entry(timer, now);
1518 expires_in_hardirq = lockdep_hrtimer_enter(timer);
1519
1520 restart = fn(timer);
1521
1522 lockdep_hrtimer_exit(expires_in_hardirq);
1523 trace_hrtimer_expire_exit(timer);
1524 raw_spin_lock_irq(&cpu_base->lock);
1525
1526 /*
1527 * Note: We clear the running state after enqueue_hrtimer and
1528 * we do not reprogram the event hardware. Happens either in
1529 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1530 *
1531 * Note: Because we dropped the cpu_base->lock above,
1532 * hrtimer_start_range_ns() can have popped in and enqueued the timer
1533 * for us already.
1534 */
1535 if (restart != HRTIMER_NORESTART &&
1536 !(timer->state & HRTIMER_STATE_ENQUEUED))
1537 enqueue_hrtimer(timer, base, HRTIMER_MODE_ABS);
1538
1539 /*
1540 * Separate the ->running assignment from the ->state assignment.
1541 *
1542 * As with a regular write barrier, this ensures the read side in
1543 * hrtimer_active() cannot observe base->running.timer == NULL &&
1544 * timer->state == INACTIVE.
1545 */
1546 raw_write_seqcount_barrier(&base->seq);
1547
1548 WARN_ON_ONCE(base->running != timer);
1549 base->running = NULL;
1550}
1551
1552static void __hrtimer_run_queues(struct hrtimer_cpu_base *cpu_base, ktime_t now,
1553 unsigned long flags, unsigned int active_mask)
1554{
1555 struct hrtimer_clock_base *base;
1556 unsigned int active = cpu_base->active_bases & active_mask;
1557
1558 for_each_active_base(base, cpu_base, active) {
1559 struct timerqueue_node *node;
1560 ktime_t basenow;
1561
1562 basenow = ktime_add(now, base->offset);
1563
1564 while ((node = timerqueue_getnext(&base->active))) {
1565 struct hrtimer *timer;
1566
1567 timer = container_of(node, struct hrtimer, node);
1568
1569 /*
1570 * The immediate goal for using the softexpires is
1571 * minimizing wakeups, not running timers at the
1572 * earliest interrupt after their soft expiration.
1573 * This allows us to avoid using a Priority Search
1574 * Tree, which can answer a stabbing querry for
1575 * overlapping intervals and instead use the simple
1576 * BST we already have.
1577 * We don't add extra wakeups by delaying timers that
1578 * are right-of a not yet expired timer, because that
1579 * timer will have to trigger a wakeup anyway.
1580 */
1581 if (basenow < hrtimer_get_softexpires_tv64(timer))
1582 break;
1583
1584 __run_hrtimer(cpu_base, base, timer, &basenow, flags);
1585 if (active_mask == HRTIMER_ACTIVE_SOFT)
1586 hrtimer_sync_wait_running(cpu_base, flags);
1587 }
1588 }
1589}
1590
1591static __latent_entropy void hrtimer_run_softirq(struct softirq_action *h)
1592{
1593 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1594 unsigned long flags;
1595 ktime_t now;
1596
1597 hrtimer_cpu_base_lock_expiry(cpu_base);
1598 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1599
1600 now = hrtimer_update_base(cpu_base);
1601 __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_SOFT);
1602
1603 cpu_base->softirq_activated = 0;
1604 hrtimer_update_softirq_timer(cpu_base, true);
1605
1606 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1607 hrtimer_cpu_base_unlock_expiry(cpu_base);
1608}
1609
1610#ifdef CONFIG_HIGH_RES_TIMERS
1611
1612/*
1613 * High resolution timer interrupt
1614 * Called with interrupts disabled
1615 */
1616void hrtimer_interrupt(struct clock_event_device *dev)
1617{
1618 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1619 ktime_t expires_next, now, entry_time, delta;
1620 unsigned long flags;
1621 int retries = 0;
1622
1623 BUG_ON(!cpu_base->hres_active);
1624 cpu_base->nr_events++;
1625 dev->next_event = KTIME_MAX;
1626
1627 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1628 entry_time = now = hrtimer_update_base(cpu_base);
1629retry:
1630 cpu_base->in_hrtirq = 1;
1631 /*
1632 * We set expires_next to KTIME_MAX here with cpu_base->lock
1633 * held to prevent that a timer is enqueued in our queue via
1634 * the migration code. This does not affect enqueueing of
1635 * timers which run their callback and need to be requeued on
1636 * this CPU.
1637 */
1638 cpu_base->expires_next = KTIME_MAX;
1639
1640 if (!ktime_before(now, cpu_base->softirq_expires_next)) {
1641 cpu_base->softirq_expires_next = KTIME_MAX;
1642 cpu_base->softirq_activated = 1;
1643 raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1644 }
1645
1646 __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_HARD);
1647
1648 /* Reevaluate the clock bases for the next expiry */
1649 expires_next = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_ALL);
1650 /*
1651 * Store the new expiry value so the migration code can verify
1652 * against it.
1653 */
1654 cpu_base->expires_next = expires_next;
1655 cpu_base->in_hrtirq = 0;
1656 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1657
1658 /* Reprogramming necessary ? */
1659 if (!tick_program_event(expires_next, 0)) {
1660 cpu_base->hang_detected = 0;
1661 return;
1662 }
1663
1664 /*
1665 * The next timer was already expired due to:
1666 * - tracing
1667 * - long lasting callbacks
1668 * - being scheduled away when running in a VM
1669 *
1670 * We need to prevent that we loop forever in the hrtimer
1671 * interrupt routine. We give it 3 attempts to avoid
1672 * overreacting on some spurious event.
1673 *
1674 * Acquire base lock for updating the offsets and retrieving
1675 * the current time.
1676 */
1677 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1678 now = hrtimer_update_base(cpu_base);
1679 cpu_base->nr_retries++;
1680 if (++retries < 3)
1681 goto retry;
1682 /*
1683 * Give the system a chance to do something else than looping
1684 * here. We stored the entry time, so we know exactly how long
1685 * we spent here. We schedule the next event this amount of
1686 * time away.
1687 */
1688 cpu_base->nr_hangs++;
1689 cpu_base->hang_detected = 1;
1690 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1691
1692 delta = ktime_sub(now, entry_time);
1693 if ((unsigned int)delta > cpu_base->max_hang_time)
1694 cpu_base->max_hang_time = (unsigned int) delta;
1695 /*
1696 * Limit it to a sensible value as we enforce a longer
1697 * delay. Give the CPU at least 100ms to catch up.
1698 */
1699 if (delta > 100 * NSEC_PER_MSEC)
1700 expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1701 else
1702 expires_next = ktime_add(now, delta);
1703 tick_program_event(expires_next, 1);
1704 pr_warn_once("hrtimer: interrupt took %llu ns\n", ktime_to_ns(delta));
1705}
1706
1707/* called with interrupts disabled */
1708static inline void __hrtimer_peek_ahead_timers(void)
1709{
1710 struct tick_device *td;
1711
1712 if (!hrtimer_hres_active())
1713 return;
1714
1715 td = this_cpu_ptr(&tick_cpu_device);
1716 if (td && td->evtdev)
1717 hrtimer_interrupt(td->evtdev);
1718}
1719
1720#else /* CONFIG_HIGH_RES_TIMERS */
1721
1722static inline void __hrtimer_peek_ahead_timers(void) { }
1723
1724#endif /* !CONFIG_HIGH_RES_TIMERS */
1725
1726/*
1727 * Called from run_local_timers in hardirq context every jiffy
1728 */
1729void hrtimer_run_queues(void)
1730{
1731 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1732 unsigned long flags;
1733 ktime_t now;
1734
1735 if (__hrtimer_hres_active(cpu_base))
1736 return;
1737
1738 /*
1739 * This _is_ ugly: We have to check periodically, whether we
1740 * can switch to highres and / or nohz mode. The clocksource
1741 * switch happens with xtime_lock held. Notification from
1742 * there only sets the check bit in the tick_oneshot code,
1743 * otherwise we might deadlock vs. xtime_lock.
1744 */
1745 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) {
1746 hrtimer_switch_to_hres();
1747 return;
1748 }
1749
1750 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1751 now = hrtimer_update_base(cpu_base);
1752
1753 if (!ktime_before(now, cpu_base->softirq_expires_next)) {
1754 cpu_base->softirq_expires_next = KTIME_MAX;
1755 cpu_base->softirq_activated = 1;
1756 raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1757 }
1758
1759 __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_HARD);
1760 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1761}
1762
1763/*
1764 * Sleep related functions:
1765 */
1766static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1767{
1768 struct hrtimer_sleeper *t =
1769 container_of(timer, struct hrtimer_sleeper, timer);
1770 struct task_struct *task = t->task;
1771
1772 t->task = NULL;
1773 if (task)
1774 wake_up_process(task);
1775
1776 return HRTIMER_NORESTART;
1777}
1778
1779/**
1780 * hrtimer_sleeper_start_expires - Start a hrtimer sleeper timer
1781 * @sl: sleeper to be started
1782 * @mode: timer mode abs/rel
1783 *
1784 * Wrapper around hrtimer_start_expires() for hrtimer_sleeper based timers
1785 * to allow PREEMPT_RT to tweak the delivery mode (soft/hardirq context)
1786 */
1787void hrtimer_sleeper_start_expires(struct hrtimer_sleeper *sl,
1788 enum hrtimer_mode mode)
1789{
1790 /*
1791 * Make the enqueue delivery mode check work on RT. If the sleeper
1792 * was initialized for hard interrupt delivery, force the mode bit.
1793 * This is a special case for hrtimer_sleepers because
1794 * hrtimer_init_sleeper() determines the delivery mode on RT so the
1795 * fiddling with this decision is avoided at the call sites.
1796 */
1797 if (IS_ENABLED(CONFIG_PREEMPT_RT) && sl->timer.is_hard)
1798 mode |= HRTIMER_MODE_HARD;
1799
1800 hrtimer_start_expires(&sl->timer, mode);
1801}
1802EXPORT_SYMBOL_GPL(hrtimer_sleeper_start_expires);
1803
1804static void __hrtimer_init_sleeper(struct hrtimer_sleeper *sl,
1805 clockid_t clock_id, enum hrtimer_mode mode)
1806{
1807 /*
1808 * On PREEMPT_RT enabled kernels hrtimers which are not explicitely
1809 * marked for hard interrupt expiry mode are moved into soft
1810 * interrupt context either for latency reasons or because the
1811 * hrtimer callback takes regular spinlocks or invokes other
1812 * functions which are not suitable for hard interrupt context on
1813 * PREEMPT_RT.
1814 *
1815 * The hrtimer_sleeper callback is RT compatible in hard interrupt
1816 * context, but there is a latency concern: Untrusted userspace can
1817 * spawn many threads which arm timers for the same expiry time on
1818 * the same CPU. That causes a latency spike due to the wakeup of
1819 * a gazillion threads.
1820 *
1821 * OTOH, priviledged real-time user space applications rely on the
1822 * low latency of hard interrupt wakeups. If the current task is in
1823 * a real-time scheduling class, mark the mode for hard interrupt
1824 * expiry.
1825 */
1826 if (IS_ENABLED(CONFIG_PREEMPT_RT)) {
1827 if (task_is_realtime(current) && !(mode & HRTIMER_MODE_SOFT))
1828 mode |= HRTIMER_MODE_HARD;
1829 }
1830
1831 __hrtimer_init(&sl->timer, clock_id, mode);
1832 sl->timer.function = hrtimer_wakeup;
1833 sl->task = current;
1834}
1835
1836/**
1837 * hrtimer_init_sleeper - initialize sleeper to the given clock
1838 * @sl: sleeper to be initialized
1839 * @clock_id: the clock to be used
1840 * @mode: timer mode abs/rel
1841 */
1842void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, clockid_t clock_id,
1843 enum hrtimer_mode mode)
1844{
1845 debug_init(&sl->timer, clock_id, mode);
1846 __hrtimer_init_sleeper(sl, clock_id, mode);
1847
1848}
1849EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
1850
1851int nanosleep_copyout(struct restart_block *restart, struct timespec64 *ts)
1852{
1853 switch(restart->nanosleep.type) {
1854#ifdef CONFIG_COMPAT_32BIT_TIME
1855 case TT_COMPAT:
1856 if (put_old_timespec32(ts, restart->nanosleep.compat_rmtp))
1857 return -EFAULT;
1858 break;
1859#endif
1860 case TT_NATIVE:
1861 if (put_timespec64(ts, restart->nanosleep.rmtp))
1862 return -EFAULT;
1863 break;
1864 default:
1865 BUG();
1866 }
1867 return -ERESTART_RESTARTBLOCK;
1868}
1869
1870static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1871{
1872 struct restart_block *restart;
1873
1874 do {
1875 set_current_state(TASK_INTERRUPTIBLE);
1876 hrtimer_sleeper_start_expires(t, mode);
1877
1878 if (likely(t->task))
1879 freezable_schedule();
1880
1881 hrtimer_cancel(&t->timer);
1882 mode = HRTIMER_MODE_ABS;
1883
1884 } while (t->task && !signal_pending(current));
1885
1886 __set_current_state(TASK_RUNNING);
1887
1888 if (!t->task)
1889 return 0;
1890
1891 restart = &current->restart_block;
1892 if (restart->nanosleep.type != TT_NONE) {
1893 ktime_t rem = hrtimer_expires_remaining(&t->timer);
1894 struct timespec64 rmt;
1895
1896 if (rem <= 0)
1897 return 0;
1898 rmt = ktime_to_timespec64(rem);
1899
1900 return nanosleep_copyout(restart, &rmt);
1901 }
1902 return -ERESTART_RESTARTBLOCK;
1903}
1904
1905static long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1906{
1907 struct hrtimer_sleeper t;
1908 int ret;
1909
1910 hrtimer_init_sleeper_on_stack(&t, restart->nanosleep.clockid,
1911 HRTIMER_MODE_ABS);
1912 hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
1913 ret = do_nanosleep(&t, HRTIMER_MODE_ABS);
1914 destroy_hrtimer_on_stack(&t.timer);
1915 return ret;
1916}
1917
1918long hrtimer_nanosleep(ktime_t rqtp, const enum hrtimer_mode mode,
1919 const clockid_t clockid)
1920{
1921 struct restart_block *restart;
1922 struct hrtimer_sleeper t;
1923 int ret = 0;
1924 u64 slack;
1925
1926 slack = current->timer_slack_ns;
1927 if (dl_task(current) || rt_task(current))
1928 slack = 0;
1929
1930 hrtimer_init_sleeper_on_stack(&t, clockid, mode);
1931 hrtimer_set_expires_range_ns(&t.timer, rqtp, slack);
1932 ret = do_nanosleep(&t, mode);
1933 if (ret != -ERESTART_RESTARTBLOCK)
1934 goto out;
1935
1936 /* Absolute timers do not update the rmtp value and restart: */
1937 if (mode == HRTIMER_MODE_ABS) {
1938 ret = -ERESTARTNOHAND;
1939 goto out;
1940 }
1941
1942 restart = &current->restart_block;
1943 restart->fn = hrtimer_nanosleep_restart;
1944 restart->nanosleep.clockid = t.timer.base->clockid;
1945 restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
1946out:
1947 destroy_hrtimer_on_stack(&t.timer);
1948 return ret;
1949}
1950
1951#ifdef CONFIG_64BIT
1952
1953SYSCALL_DEFINE2(nanosleep, struct __kernel_timespec __user *, rqtp,
1954 struct __kernel_timespec __user *, rmtp)
1955{
1956 struct timespec64 tu;
1957
1958 if (get_timespec64(&tu, rqtp))
1959 return -EFAULT;
1960
1961 if (!timespec64_valid(&tu))
1962 return -EINVAL;
1963
1964 current->restart_block.nanosleep.type = rmtp ? TT_NATIVE : TT_NONE;
1965 current->restart_block.nanosleep.rmtp = rmtp;
1966 return hrtimer_nanosleep(timespec64_to_ktime(tu), HRTIMER_MODE_REL,
1967 CLOCK_MONOTONIC);
1968}
1969
1970#endif
1971
1972#ifdef CONFIG_COMPAT_32BIT_TIME
1973
1974SYSCALL_DEFINE2(nanosleep_time32, struct old_timespec32 __user *, rqtp,
1975 struct old_timespec32 __user *, rmtp)
1976{
1977 struct timespec64 tu;
1978
1979 if (get_old_timespec32(&tu, rqtp))
1980 return -EFAULT;
1981
1982 if (!timespec64_valid(&tu))
1983 return -EINVAL;
1984
1985 current->restart_block.nanosleep.type = rmtp ? TT_COMPAT : TT_NONE;
1986 current->restart_block.nanosleep.compat_rmtp = rmtp;
1987 return hrtimer_nanosleep(timespec64_to_ktime(tu), HRTIMER_MODE_REL,
1988 CLOCK_MONOTONIC);
1989}
1990#endif
1991
1992/*
1993 * Functions related to boot-time initialization:
1994 */
1995int hrtimers_prepare_cpu(unsigned int cpu)
1996{
1997 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1998 int i;
1999
2000 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
2001 cpu_base->clock_base[i].cpu_base = cpu_base;
2002 timerqueue_init_head(&cpu_base->clock_base[i].active);
2003 }
2004
2005 cpu_base->cpu = cpu;
2006 cpu_base->active_bases = 0;
2007 cpu_base->hres_active = 0;
2008 cpu_base->hang_detected = 0;
2009 cpu_base->next_timer = NULL;
2010 cpu_base->softirq_next_timer = NULL;
2011 cpu_base->expires_next = KTIME_MAX;
2012 cpu_base->softirq_expires_next = KTIME_MAX;
2013 hrtimer_cpu_base_init_expiry_lock(cpu_base);
2014 return 0;
2015}
2016
2017#ifdef CONFIG_HOTPLUG_CPU
2018
2019static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
2020 struct hrtimer_clock_base *new_base)
2021{
2022 struct hrtimer *timer;
2023 struct timerqueue_node *node;
2024
2025 while ((node = timerqueue_getnext(&old_base->active))) {
2026 timer = container_of(node, struct hrtimer, node);
2027 BUG_ON(hrtimer_callback_running(timer));
2028 debug_deactivate(timer);
2029
2030 /*
2031 * Mark it as ENQUEUED not INACTIVE otherwise the
2032 * timer could be seen as !active and just vanish away
2033 * under us on another CPU
2034 */
2035 __remove_hrtimer(timer, old_base, HRTIMER_STATE_ENQUEUED, 0);
2036 timer->base = new_base;
2037 /*
2038 * Enqueue the timers on the new cpu. This does not
2039 * reprogram the event device in case the timer
2040 * expires before the earliest on this CPU, but we run
2041 * hrtimer_interrupt after we migrated everything to
2042 * sort out already expired timers and reprogram the
2043 * event device.
2044 */
2045 enqueue_hrtimer(timer, new_base, HRTIMER_MODE_ABS);
2046 }
2047}
2048
2049int hrtimers_dead_cpu(unsigned int scpu)
2050{
2051 struct hrtimer_cpu_base *old_base, *new_base;
2052 int i;
2053
2054 BUG_ON(cpu_online(scpu));
2055 tick_cancel_sched_timer(scpu);
2056
2057 /*
2058 * this BH disable ensures that raise_softirq_irqoff() does
2059 * not wakeup ksoftirqd (and acquire the pi-lock) while
2060 * holding the cpu_base lock
2061 */
2062 local_bh_disable();
2063 local_irq_disable();
2064 old_base = &per_cpu(hrtimer_bases, scpu);
2065 new_base = this_cpu_ptr(&hrtimer_bases);
2066 /*
2067 * The caller is globally serialized and nobody else
2068 * takes two locks at once, deadlock is not possible.
2069 */
2070 raw_spin_lock(&new_base->lock);
2071 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
2072
2073 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
2074 migrate_hrtimer_list(&old_base->clock_base[i],
2075 &new_base->clock_base[i]);
2076 }
2077
2078 /*
2079 * The migration might have changed the first expiring softirq
2080 * timer on this CPU. Update it.
2081 */
2082 hrtimer_update_softirq_timer(new_base, false);
2083
2084 raw_spin_unlock(&old_base->lock);
2085 raw_spin_unlock(&new_base->lock);
2086
2087 /* Check, if we got expired work to do */
2088 __hrtimer_peek_ahead_timers();
2089 local_irq_enable();
2090 local_bh_enable();
2091 return 0;
2092}
2093
2094#endif /* CONFIG_HOTPLUG_CPU */
2095
2096void __init hrtimers_init(void)
2097{
2098 hrtimers_prepare_cpu(smp_processor_id());
2099 open_softirq(HRTIMER_SOFTIRQ, hrtimer_run_softirq);
2100}
2101
2102/**
2103 * schedule_hrtimeout_range_clock - sleep until timeout
2104 * @expires: timeout value (ktime_t)
2105 * @delta: slack in expires timeout (ktime_t)
2106 * @mode: timer mode
2107 * @clock_id: timer clock to be used
2108 */
2109int __sched
2110schedule_hrtimeout_range_clock(ktime_t *expires, u64 delta,
2111 const enum hrtimer_mode mode, clockid_t clock_id)
2112{
2113 struct hrtimer_sleeper t;
2114
2115 /*
2116 * Optimize when a zero timeout value is given. It does not
2117 * matter whether this is an absolute or a relative time.
2118 */
2119 if (expires && *expires == 0) {
2120 __set_current_state(TASK_RUNNING);
2121 return 0;
2122 }
2123
2124 /*
2125 * A NULL parameter means "infinite"
2126 */
2127 if (!expires) {
2128 schedule();
2129 return -EINTR;
2130 }
2131
2132 hrtimer_init_sleeper_on_stack(&t, clock_id, mode);
2133 hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
2134 hrtimer_sleeper_start_expires(&t, mode);
2135
2136 if (likely(t.task))
2137 schedule();
2138
2139 hrtimer_cancel(&t.timer);
2140 destroy_hrtimer_on_stack(&t.timer);
2141
2142 __set_current_state(TASK_RUNNING);
2143
2144 return !t.task ? 0 : -EINTR;
2145}
2146
2147/**
2148 * schedule_hrtimeout_range - sleep until timeout
2149 * @expires: timeout value (ktime_t)
2150 * @delta: slack in expires timeout (ktime_t)
2151 * @mode: timer mode
2152 *
2153 * Make the current task sleep until the given expiry time has
2154 * elapsed. The routine will return immediately unless
2155 * the current task state has been set (see set_current_state()).
2156 *
2157 * The @delta argument gives the kernel the freedom to schedule the
2158 * actual wakeup to a time that is both power and performance friendly.
2159 * The kernel give the normal best effort behavior for "@expires+@delta",
2160 * but may decide to fire the timer earlier, but no earlier than @expires.
2161 *
2162 * You can set the task state as follows -
2163 *
2164 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
2165 * pass before the routine returns unless the current task is explicitly
2166 * woken up, (e.g. by wake_up_process()).
2167 *
2168 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
2169 * delivered to the current task or the current task is explicitly woken
2170 * up.
2171 *
2172 * The current task state is guaranteed to be TASK_RUNNING when this
2173 * routine returns.
2174 *
2175 * Returns 0 when the timer has expired. If the task was woken before the
2176 * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or
2177 * by an explicit wakeup, it returns -EINTR.
2178 */
2179int __sched schedule_hrtimeout_range(ktime_t *expires, u64 delta,
2180 const enum hrtimer_mode mode)
2181{
2182 return schedule_hrtimeout_range_clock(expires, delta, mode,
2183 CLOCK_MONOTONIC);
2184}
2185EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
2186
2187/**
2188 * schedule_hrtimeout - sleep until timeout
2189 * @expires: timeout value (ktime_t)
2190 * @mode: timer mode
2191 *
2192 * Make the current task sleep until the given expiry time has
2193 * elapsed. The routine will return immediately unless
2194 * the current task state has been set (see set_current_state()).
2195 *
2196 * You can set the task state as follows -
2197 *
2198 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
2199 * pass before the routine returns unless the current task is explicitly
2200 * woken up, (e.g. by wake_up_process()).
2201 *
2202 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
2203 * delivered to the current task or the current task is explicitly woken
2204 * up.
2205 *
2206 * The current task state is guaranteed to be TASK_RUNNING when this
2207 * routine returns.
2208 *
2209 * Returns 0 when the timer has expired. If the task was woken before the
2210 * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or
2211 * by an explicit wakeup, it returns -EINTR.
2212 */
2213int __sched schedule_hrtimeout(ktime_t *expires,
2214 const enum hrtimer_mode mode)
2215{
2216 return schedule_hrtimeout_range(expires, 0, mode);
2217}
2218EXPORT_SYMBOL_GPL(schedule_hrtimeout);