projects/linux·blob v6.7 · refs

2391 lines · 66.4 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 = { {
139 .cpu_base = &migration_cpu_base,
140 .seq = SEQCNT_RAW_SPINLOCK_ZERO(migration_cpu_base.seq,
141 &migration_cpu_base.lock),
142 }, },
143};
144
145#define migration_base migration_cpu_base.clock_base[0]
146
147static inline bool is_migration_base(struct hrtimer_clock_base *base)
148{
149 return base == &migration_base;
150}
151
152/*
153 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
154 * means that all timers which are tied to this base via timer->base are
155 * locked, and the base itself is locked too.
156 *
157 * So __run_timers/migrate_timers can safely modify all timers which could
158 * be found on the lists/queues.
159 *
160 * When the timer's base is locked, and the timer removed from list, it is
161 * possible to set timer->base = &migration_base and drop the lock: the timer
162 * remains locked.
163 */
164static
165struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
166 unsigned long *flags)
167 __acquires(&timer->base->lock)
168{
169 struct hrtimer_clock_base *base;
170
171 for (;;) {
172 base = READ_ONCE(timer->base);
173 if (likely(base != &migration_base)) {
174 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
175 if (likely(base == timer->base))
176 return base;
177 /* The timer has migrated to another CPU: */
178 raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
179 }
180 cpu_relax();
181 }
182}
183
184/*
185 * We do not migrate the timer when it is expiring before the next
186 * event on the target cpu. When high resolution is enabled, we cannot
187 * reprogram the target cpu hardware and we would cause it to fire
188 * late. To keep it simple, we handle the high resolution enabled and
189 * disabled case similar.
190 *
191 * Called with cpu_base->lock of target cpu held.
192 */
193static int
194hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base)
195{
196 ktime_t expires;
197
198 expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset);
199 return expires < new_base->cpu_base->expires_next;
200}
201
202static inline
203struct hrtimer_cpu_base *get_target_base(struct hrtimer_cpu_base *base,
204 int pinned)
205{
206#if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
207 if (static_branch_likely(&timers_migration_enabled) && !pinned)
208 return &per_cpu(hrtimer_bases, get_nohz_timer_target());
209#endif
210 return base;
211}
212
213/*
214 * We switch the timer base to a power-optimized selected CPU target,
215 * if:
216 * - NO_HZ_COMMON is enabled
217 * - timer migration is enabled
218 * - the timer callback is not running
219 * - the timer is not the first expiring timer on the new target
220 *
221 * If one of the above requirements is not fulfilled we move the timer
222 * to the current CPU or leave it on the previously assigned CPU if
223 * the timer callback is currently running.
224 */
225static inline struct hrtimer_clock_base *
226switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
227 int pinned)
228{
229 struct hrtimer_cpu_base *new_cpu_base, *this_cpu_base;
230 struct hrtimer_clock_base *new_base;
231 int basenum = base->index;
232
233 this_cpu_base = this_cpu_ptr(&hrtimer_bases);
234 new_cpu_base = get_target_base(this_cpu_base, pinned);
235again:
236 new_base = &new_cpu_base->clock_base[basenum];
237
238 if (base != new_base) {
239 /*
240 * We are trying to move timer to new_base.
241 * However we can't change timer's base while it is running,
242 * so we keep it on the same CPU. No hassle vs. reprogramming
243 * the event source in the high resolution case. The softirq
244 * code will take care of this when the timer function has
245 * completed. There is no conflict as we hold the lock until
246 * the timer is enqueued.
247 */
248 if (unlikely(hrtimer_callback_running(timer)))
249 return base;
250
251 /* See the comment in lock_hrtimer_base() */
252 WRITE_ONCE(timer->base, &migration_base);
253 raw_spin_unlock(&base->cpu_base->lock);
254 raw_spin_lock(&new_base->cpu_base->lock);
255
256 if (new_cpu_base != this_cpu_base &&
257 hrtimer_check_target(timer, new_base)) {
258 raw_spin_unlock(&new_base->cpu_base->lock);
259 raw_spin_lock(&base->cpu_base->lock);
260 new_cpu_base = this_cpu_base;
261 WRITE_ONCE(timer->base, base);
262 goto again;
263 }
264 WRITE_ONCE(timer->base, new_base);
265 } else {
266 if (new_cpu_base != this_cpu_base &&
267 hrtimer_check_target(timer, new_base)) {
268 new_cpu_base = this_cpu_base;
269 goto again;
270 }
271 }
272 return new_base;
273}
274
275#else /* CONFIG_SMP */
276
277static inline bool is_migration_base(struct hrtimer_clock_base *base)
278{
279 return false;
280}
281
282static inline struct hrtimer_clock_base *
283lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
284 __acquires(&timer->base->cpu_base->lock)
285{
286 struct hrtimer_clock_base *base = timer->base;
287
288 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
289
290 return base;
291}
292
293# define switch_hrtimer_base(t, b, p) (b)
294
295#endif /* !CONFIG_SMP */
296
297/*
298 * Functions for the union type storage format of ktime_t which are
299 * too large for inlining:
300 */
301#if BITS_PER_LONG < 64
302/*
303 * Divide a ktime value by a nanosecond value
304 */
305s64 __ktime_divns(const ktime_t kt, s64 div)
306{
307 int sft = 0;
308 s64 dclc;
309 u64 tmp;
310
311 dclc = ktime_to_ns(kt);
312 tmp = dclc < 0 ? -dclc : dclc;
313
314 /* Make sure the divisor is less than 2^32: */
315 while (div >> 32) {
316 sft++;
317 div >>= 1;
318 }
319 tmp >>= sft;
320 do_div(tmp, (u32) div);
321 return dclc < 0 ? -tmp : tmp;
322}
323EXPORT_SYMBOL_GPL(__ktime_divns);
324#endif /* BITS_PER_LONG >= 64 */
325
326/*
327 * Add two ktime values and do a safety check for overflow:
328 */
329ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
330{
331 ktime_t res = ktime_add_unsafe(lhs, rhs);
332
333 /*
334 * We use KTIME_SEC_MAX here, the maximum timeout which we can
335 * return to user space in a timespec:
336 */
337 if (res < 0 || res < lhs || res < rhs)
338 res = ktime_set(KTIME_SEC_MAX, 0);
339
340 return res;
341}
342
343EXPORT_SYMBOL_GPL(ktime_add_safe);
344
345#ifdef CONFIG_DEBUG_OBJECTS_TIMERS
346
347static const struct debug_obj_descr hrtimer_debug_descr;
348
349static void *hrtimer_debug_hint(void *addr)
350{
351 return ((struct hrtimer *) addr)->function;
352}
353
354/*
355 * fixup_init is called when:
356 * - an active object is initialized
357 */
358static bool hrtimer_fixup_init(void *addr, enum debug_obj_state state)
359{
360 struct hrtimer *timer = addr;
361
362 switch (state) {
363 case ODEBUG_STATE_ACTIVE:
364 hrtimer_cancel(timer);
365 debug_object_init(timer, &hrtimer_debug_descr);
366 return true;
367 default:
368 return false;
369 }
370}
371
372/*
373 * fixup_activate is called when:
374 * - an active object is activated
375 * - an unknown non-static object is activated
376 */
377static bool hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
378{
379 switch (state) {
380 case ODEBUG_STATE_ACTIVE:
381 WARN_ON(1);
382 fallthrough;
383 default:
384 return false;
385 }
386}
387
388/*
389 * fixup_free is called when:
390 * - an active object is freed
391 */
392static bool hrtimer_fixup_free(void *addr, enum debug_obj_state state)
393{
394 struct hrtimer *timer = addr;
395
396 switch (state) {
397 case ODEBUG_STATE_ACTIVE:
398 hrtimer_cancel(timer);
399 debug_object_free(timer, &hrtimer_debug_descr);
400 return true;
401 default:
402 return false;
403 }
404}
405
406static const struct debug_obj_descr hrtimer_debug_descr = {
407 .name = "hrtimer",
408 .debug_hint = hrtimer_debug_hint,
409 .fixup_init = hrtimer_fixup_init,
410 .fixup_activate = hrtimer_fixup_activate,
411 .fixup_free = hrtimer_fixup_free,
412};
413
414static inline void debug_hrtimer_init(struct hrtimer *timer)
415{
416 debug_object_init(timer, &hrtimer_debug_descr);
417}
418
419static inline void debug_hrtimer_activate(struct hrtimer *timer,
420 enum hrtimer_mode mode)
421{
422 debug_object_activate(timer, &hrtimer_debug_descr);
423}
424
425static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
426{
427 debug_object_deactivate(timer, &hrtimer_debug_descr);
428}
429
430static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
431 enum hrtimer_mode mode);
432
433void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
434 enum hrtimer_mode mode)
435{
436 debug_object_init_on_stack(timer, &hrtimer_debug_descr);
437 __hrtimer_init(timer, clock_id, mode);
438}
439EXPORT_SYMBOL_GPL(hrtimer_init_on_stack);
440
441static void __hrtimer_init_sleeper(struct hrtimer_sleeper *sl,
442 clockid_t clock_id, enum hrtimer_mode mode);
443
444void hrtimer_init_sleeper_on_stack(struct hrtimer_sleeper *sl,
445 clockid_t clock_id, enum hrtimer_mode mode)
446{
447 debug_object_init_on_stack(&sl->timer, &hrtimer_debug_descr);
448 __hrtimer_init_sleeper(sl, clock_id, mode);
449}
450EXPORT_SYMBOL_GPL(hrtimer_init_sleeper_on_stack);
451
452void destroy_hrtimer_on_stack(struct hrtimer *timer)
453{
454 debug_object_free(timer, &hrtimer_debug_descr);
455}
456EXPORT_SYMBOL_GPL(destroy_hrtimer_on_stack);
457
458#else
459
460static inline void debug_hrtimer_init(struct hrtimer *timer) { }
461static inline void debug_hrtimer_activate(struct hrtimer *timer,
462 enum hrtimer_mode mode) { }
463static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
464#endif
465
466static inline void
467debug_init(struct hrtimer *timer, clockid_t clockid,
468 enum hrtimer_mode mode)
469{
470 debug_hrtimer_init(timer);
471 trace_hrtimer_init(timer, clockid, mode);
472}
473
474static inline void debug_activate(struct hrtimer *timer,
475 enum hrtimer_mode mode)
476{
477 debug_hrtimer_activate(timer, mode);
478 trace_hrtimer_start(timer, mode);
479}
480
481static inline void debug_deactivate(struct hrtimer *timer)
482{
483 debug_hrtimer_deactivate(timer);
484 trace_hrtimer_cancel(timer);
485}
486
487static struct hrtimer_clock_base *
488__next_base(struct hrtimer_cpu_base *cpu_base, unsigned int *active)
489{
490 unsigned int idx;
491
492 if (!*active)
493 return NULL;
494
495 idx = __ffs(*active);
496 *active &= ~(1U << idx);
497
498 return &cpu_base->clock_base[idx];
499}
500
501#define for_each_active_base(base, cpu_base, active) \
502 while ((base = __next_base((cpu_base), &(active))))
503
504static ktime_t __hrtimer_next_event_base(struct hrtimer_cpu_base *cpu_base,
505 const struct hrtimer *exclude,
506 unsigned int active,
507 ktime_t expires_next)
508{
509 struct hrtimer_clock_base *base;
510 ktime_t expires;
511
512 for_each_active_base(base, cpu_base, active) {
513 struct timerqueue_node *next;
514 struct hrtimer *timer;
515
516 next = timerqueue_getnext(&base->active);
517 timer = container_of(next, struct hrtimer, node);
518 if (timer == exclude) {
519 /* Get to the next timer in the queue. */
520 next = timerqueue_iterate_next(next);
521 if (!next)
522 continue;
523
524 timer = container_of(next, struct hrtimer, node);
525 }
526 expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
527 if (expires < expires_next) {
528 expires_next = expires;
529
530 /* Skip cpu_base update if a timer is being excluded. */
531 if (exclude)
532 continue;
533
534 if (timer->is_soft)
535 cpu_base->softirq_next_timer = timer;
536 else
537 cpu_base->next_timer = timer;
538 }
539 }
540 /*
541 * clock_was_set() might have changed base->offset of any of
542 * the clock bases so the result might be negative. Fix it up
543 * to prevent a false positive in clockevents_program_event().
544 */
545 if (expires_next < 0)
546 expires_next = 0;
547 return expires_next;
548}
549
550/*
551 * Recomputes cpu_base::*next_timer and returns the earliest expires_next
552 * but does not set cpu_base::*expires_next, that is done by
553 * hrtimer[_force]_reprogram and hrtimer_interrupt only. When updating
554 * cpu_base::*expires_next right away, reprogramming logic would no longer
555 * work.
556 *
557 * When a softirq is pending, we can ignore the HRTIMER_ACTIVE_SOFT bases,
558 * those timers will get run whenever the softirq gets handled, at the end of
559 * hrtimer_run_softirq(), hrtimer_update_softirq_timer() will re-add these bases.
560 *
561 * Therefore softirq values are those from the HRTIMER_ACTIVE_SOFT clock bases.
562 * The !softirq values are the minima across HRTIMER_ACTIVE_ALL, unless an actual
563 * softirq is pending, in which case they're the minima of HRTIMER_ACTIVE_HARD.
564 *
565 * @active_mask must be one of:
566 * - HRTIMER_ACTIVE_ALL,
567 * - HRTIMER_ACTIVE_SOFT, or
568 * - HRTIMER_ACTIVE_HARD.
569 */
570static ktime_t
571__hrtimer_get_next_event(struct hrtimer_cpu_base *cpu_base, unsigned int active_mask)
572{
573 unsigned int active;
574 struct hrtimer *next_timer = NULL;
575 ktime_t expires_next = KTIME_MAX;
576
577 if (!cpu_base->softirq_activated && (active_mask & HRTIMER_ACTIVE_SOFT)) {
578 active = cpu_base->active_bases & HRTIMER_ACTIVE_SOFT;
579 cpu_base->softirq_next_timer = NULL;
580 expires_next = __hrtimer_next_event_base(cpu_base, NULL,
581 active, KTIME_MAX);
582
583 next_timer = cpu_base->softirq_next_timer;
584 }
585
586 if (active_mask & HRTIMER_ACTIVE_HARD) {
587 active = cpu_base->active_bases & HRTIMER_ACTIVE_HARD;
588 cpu_base->next_timer = next_timer;
589 expires_next = __hrtimer_next_event_base(cpu_base, NULL, active,
590 expires_next);
591 }
592
593 return expires_next;
594}
595
596static ktime_t hrtimer_update_next_event(struct hrtimer_cpu_base *cpu_base)
597{
598 ktime_t expires_next, soft = KTIME_MAX;
599
600 /*
601 * If the soft interrupt has already been activated, ignore the
602 * soft bases. They will be handled in the already raised soft
603 * interrupt.
604 */
605 if (!cpu_base->softirq_activated) {
606 soft = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_SOFT);
607 /*
608 * Update the soft expiry time. clock_settime() might have
609 * affected it.
610 */
611 cpu_base->softirq_expires_next = soft;
612 }
613
614 expires_next = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_HARD);
615 /*
616 * If a softirq timer is expiring first, update cpu_base->next_timer
617 * and program the hardware with the soft expiry time.
618 */
619 if (expires_next > soft) {
620 cpu_base->next_timer = cpu_base->softirq_next_timer;
621 expires_next = soft;
622 }
623
624 return expires_next;
625}
626
627static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base)
628{
629 ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset;
630 ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset;
631 ktime_t *offs_tai = &base->clock_base[HRTIMER_BASE_TAI].offset;
632
633 ktime_t now = ktime_get_update_offsets_now(&base->clock_was_set_seq,
634 offs_real, offs_boot, offs_tai);
635
636 base->clock_base[HRTIMER_BASE_REALTIME_SOFT].offset = *offs_real;
637 base->clock_base[HRTIMER_BASE_BOOTTIME_SOFT].offset = *offs_boot;
638 base->clock_base[HRTIMER_BASE_TAI_SOFT].offset = *offs_tai;
639
640 return now;
641}
642
643/*
644 * Is the high resolution mode active ?
645 */
646static inline int __hrtimer_hres_active(struct hrtimer_cpu_base *cpu_base)
647{
648 return IS_ENABLED(CONFIG_HIGH_RES_TIMERS) ?
649 cpu_base->hres_active : 0;
650}
651
652static inline int hrtimer_hres_active(void)
653{
654 return __hrtimer_hres_active(this_cpu_ptr(&hrtimer_bases));
655}
656
657static void __hrtimer_reprogram(struct hrtimer_cpu_base *cpu_base,
658 struct hrtimer *next_timer,
659 ktime_t expires_next)
660{
661 cpu_base->expires_next = expires_next;
662
663 /*
664 * If hres is not active, hardware does not have to be
665 * reprogrammed yet.
666 *
667 * If a hang was detected in the last timer interrupt then we
668 * leave the hang delay active in the hardware. We want the
669 * system to make progress. That also prevents the following
670 * scenario:
671 * T1 expires 50ms from now
672 * T2 expires 5s from now
673 *
674 * T1 is removed, so this code is called and would reprogram
675 * the hardware to 5s from now. Any hrtimer_start after that
676 * will not reprogram the hardware due to hang_detected being
677 * set. So we'd effectively block all timers until the T2 event
678 * fires.
679 */
680 if (!__hrtimer_hres_active(cpu_base) || cpu_base->hang_detected)
681 return;
682
683 tick_program_event(expires_next, 1);
684}
685
686/*
687 * Reprogram the event source with checking both queues for the
688 * next event
689 * Called with interrupts disabled and base->lock held
690 */
691static void
692hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
693{
694 ktime_t expires_next;
695
696 expires_next = hrtimer_update_next_event(cpu_base);
697
698 if (skip_equal && expires_next == cpu_base->expires_next)
699 return;
700
701 __hrtimer_reprogram(cpu_base, cpu_base->next_timer, expires_next);
702}
703
704/* High resolution timer related functions */
705#ifdef CONFIG_HIGH_RES_TIMERS
706
707/*
708 * High resolution timer enabled ?
709 */
710static bool hrtimer_hres_enabled __read_mostly = true;
711unsigned int hrtimer_resolution __read_mostly = LOW_RES_NSEC;
712EXPORT_SYMBOL_GPL(hrtimer_resolution);
713
714/*
715 * Enable / Disable high resolution mode
716 */
717static int __init setup_hrtimer_hres(char *str)
718{
719 return (kstrtobool(str, &hrtimer_hres_enabled) == 0);
720}
721
722__setup("highres=", setup_hrtimer_hres);
723
724/*
725 * hrtimer_high_res_enabled - query, if the highres mode is enabled
726 */
727static inline int hrtimer_is_hres_enabled(void)
728{
729 return hrtimer_hres_enabled;
730}
731
732static void retrigger_next_event(void *arg);
733
734/*
735 * Switch to high resolution mode
736 */
737static void hrtimer_switch_to_hres(void)
738{
739 struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
740
741 if (tick_init_highres()) {
742 pr_warn("Could not switch to high resolution mode on CPU %u\n",
743 base->cpu);
744 return;
745 }
746 base->hres_active = 1;
747 hrtimer_resolution = HIGH_RES_NSEC;
748
749 tick_setup_sched_timer();
750 /* "Retrigger" the interrupt to get things going */
751 retrigger_next_event(NULL);
752}
753
754#else
755
756static inline int hrtimer_is_hres_enabled(void) { return 0; }
757static inline void hrtimer_switch_to_hres(void) { }
758
759#endif /* CONFIG_HIGH_RES_TIMERS */
760/*
761 * Retrigger next event is called after clock was set with interrupts
762 * disabled through an SMP function call or directly from low level
763 * resume code.
764 *
765 * This is only invoked when:
766 * - CONFIG_HIGH_RES_TIMERS is enabled.
767 * - CONFIG_NOHZ_COMMON is enabled
768 *
769 * For the other cases this function is empty and because the call sites
770 * are optimized out it vanishes as well, i.e. no need for lots of
771 * #ifdeffery.
772 */
773static void retrigger_next_event(void *arg)
774{
775 struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
776
777 /*
778 * When high resolution mode or nohz is active, then the offsets of
779 * CLOCK_REALTIME/TAI/BOOTTIME have to be updated. Otherwise the
780 * next tick will take care of that.
781 *
782 * If high resolution mode is active then the next expiring timer
783 * must be reevaluated and the clock event device reprogrammed if
784 * necessary.
785 *
786 * In the NOHZ case the update of the offset and the reevaluation
787 * of the next expiring timer is enough. The return from the SMP
788 * function call will take care of the reprogramming in case the
789 * CPU was in a NOHZ idle sleep.
790 */
791 if (!__hrtimer_hres_active(base) && !tick_nohz_active)
792 return;
793
794 raw_spin_lock(&base->lock);
795 hrtimer_update_base(base);
796 if (__hrtimer_hres_active(base))
797 hrtimer_force_reprogram(base, 0);
798 else
799 hrtimer_update_next_event(base);
800 raw_spin_unlock(&base->lock);
801}
802
803/*
804 * When a timer is enqueued and expires earlier than the already enqueued
805 * timers, we have to check, whether it expires earlier than the timer for
806 * which the clock event device was armed.
807 *
808 * Called with interrupts disabled and base->cpu_base.lock held
809 */
810static void hrtimer_reprogram(struct hrtimer *timer, bool reprogram)
811{
812 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
813 struct hrtimer_clock_base *base = timer->base;
814 ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
815
816 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
817
818 /*
819 * CLOCK_REALTIME timer might be requested with an absolute
820 * expiry time which is less than base->offset. Set it to 0.
821 */
822 if (expires < 0)
823 expires = 0;
824
825 if (timer->is_soft) {
826 /*
827 * soft hrtimer could be started on a remote CPU. In this
828 * case softirq_expires_next needs to be updated on the
829 * remote CPU. The soft hrtimer will not expire before the
830 * first hard hrtimer on the remote CPU -
831 * hrtimer_check_target() prevents this case.
832 */
833 struct hrtimer_cpu_base *timer_cpu_base = base->cpu_base;
834
835 if (timer_cpu_base->softirq_activated)
836 return;
837
838 if (!ktime_before(expires, timer_cpu_base->softirq_expires_next))
839 return;
840
841 timer_cpu_base->softirq_next_timer = timer;
842 timer_cpu_base->softirq_expires_next = expires;
843
844 if (!ktime_before(expires, timer_cpu_base->expires_next) ||
845 !reprogram)
846 return;
847 }
848
849 /*
850 * If the timer is not on the current cpu, we cannot reprogram
851 * the other cpus clock event device.
852 */
853 if (base->cpu_base != cpu_base)
854 return;
855
856 if (expires >= cpu_base->expires_next)
857 return;
858
859 /*
860 * If the hrtimer interrupt is running, then it will reevaluate the
861 * clock bases and reprogram the clock event device.
862 */
863 if (cpu_base->in_hrtirq)
864 return;
865
866 cpu_base->next_timer = timer;
867
868 __hrtimer_reprogram(cpu_base, timer, expires);
869}
870
871static bool update_needs_ipi(struct hrtimer_cpu_base *cpu_base,
872 unsigned int active)
873{
874 struct hrtimer_clock_base *base;
875 unsigned int seq;
876 ktime_t expires;
877
878 /*
879 * Update the base offsets unconditionally so the following
880 * checks whether the SMP function call is required works.
881 *
882 * The update is safe even when the remote CPU is in the hrtimer
883 * interrupt or the hrtimer soft interrupt and expiring affected
884 * bases. Either it will see the update before handling a base or
885 * it will see it when it finishes the processing and reevaluates
886 * the next expiring timer.
887 */
888 seq = cpu_base->clock_was_set_seq;
889 hrtimer_update_base(cpu_base);
890
891 /*
892 * If the sequence did not change over the update then the
893 * remote CPU already handled it.
894 */
895 if (seq == cpu_base->clock_was_set_seq)
896 return false;
897
898 /*
899 * If the remote CPU is currently handling an hrtimer interrupt, it
900 * will reevaluate the first expiring timer of all clock bases
901 * before reprogramming. Nothing to do here.
902 */
903 if (cpu_base->in_hrtirq)
904 return false;
905
906 /*
907 * Walk the affected clock bases and check whether the first expiring
908 * timer in a clock base is moving ahead of the first expiring timer of
909 * @cpu_base. If so, the IPI must be invoked because per CPU clock
910 * event devices cannot be remotely reprogrammed.
911 */
912 active &= cpu_base->active_bases;
913
914 for_each_active_base(base, cpu_base, active) {
915 struct timerqueue_node *next;
916
917 next = timerqueue_getnext(&base->active);
918 expires = ktime_sub(next->expires, base->offset);
919 if (expires < cpu_base->expires_next)
920 return true;
921
922 /* Extra check for softirq clock bases */
923 if (base->clockid < HRTIMER_BASE_MONOTONIC_SOFT)
924 continue;
925 if (cpu_base->softirq_activated)
926 continue;
927 if (expires < cpu_base->softirq_expires_next)
928 return true;
929 }
930 return false;
931}
932
933/*
934 * Clock was set. This might affect CLOCK_REALTIME, CLOCK_TAI and
935 * CLOCK_BOOTTIME (for late sleep time injection).
936 *
937 * This requires to update the offsets for these clocks
938 * vs. CLOCK_MONOTONIC. When high resolution timers are enabled, then this
939 * also requires to eventually reprogram the per CPU clock event devices
940 * when the change moves an affected timer ahead of the first expiring
941 * timer on that CPU. Obviously remote per CPU clock event devices cannot
942 * be reprogrammed. The other reason why an IPI has to be sent is when the
943 * system is in !HIGH_RES and NOHZ mode. The NOHZ mode updates the offsets
944 * in the tick, which obviously might be stopped, so this has to bring out
945 * the remote CPU which might sleep in idle to get this sorted.
946 */
947void clock_was_set(unsigned int bases)
948{
949 struct hrtimer_cpu_base *cpu_base = raw_cpu_ptr(&hrtimer_bases);
950 cpumask_var_t mask;
951 int cpu;
952
953 if (!__hrtimer_hres_active(cpu_base) && !tick_nohz_active)
954 goto out_timerfd;
955
956 if (!zalloc_cpumask_var(&mask, GFP_KERNEL)) {
957 on_each_cpu(retrigger_next_event, NULL, 1);
958 goto out_timerfd;
959 }
960
961 /* Avoid interrupting CPUs if possible */
962 cpus_read_lock();
963 for_each_online_cpu(cpu) {
964 unsigned long flags;
965
966 cpu_base = &per_cpu(hrtimer_bases, cpu);
967 raw_spin_lock_irqsave(&cpu_base->lock, flags);
968
969 if (update_needs_ipi(cpu_base, bases))
970 cpumask_set_cpu(cpu, mask);
971
972 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
973 }
974
975 preempt_disable();
976 smp_call_function_many(mask, retrigger_next_event, NULL, 1);
977 preempt_enable();
978 cpus_read_unlock();
979 free_cpumask_var(mask);
980
981out_timerfd:
982 timerfd_clock_was_set();
983}
984
985static void clock_was_set_work(struct work_struct *work)
986{
987 clock_was_set(CLOCK_SET_WALL);
988}
989
990static DECLARE_WORK(hrtimer_work, clock_was_set_work);
991
992/*
993 * Called from timekeeping code to reprogram the hrtimer interrupt device
994 * on all cpus and to notify timerfd.
995 */
996void clock_was_set_delayed(void)
997{
998 schedule_work(&hrtimer_work);
999}
1000
1001/*
1002 * Called during resume either directly from via timekeeping_resume()
1003 * or in the case of s2idle from tick_unfreeze() to ensure that the
1004 * hrtimers are up to date.
1005 */
1006void hrtimers_resume_local(void)
1007{
1008 lockdep_assert_irqs_disabled();
1009 /* Retrigger on the local CPU */
1010 retrigger_next_event(NULL);
1011}
1012
1013/*
1014 * Counterpart to lock_hrtimer_base above:
1015 */
1016static inline
1017void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
1018 __releases(&timer->base->cpu_base->lock)
1019{
1020 raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
1021}
1022
1023/**
1024 * hrtimer_forward - forward the timer expiry
1025 * @timer: hrtimer to forward
1026 * @now: forward past this time
1027 * @interval: the interval to forward
1028 *
1029 * Forward the timer expiry so it will expire in the future.
1030 * Returns the number of overruns.
1031 *
1032 * Can be safely called from the callback function of @timer. If
1033 * called from other contexts @timer must neither be enqueued nor
1034 * running the callback and the caller needs to take care of
1035 * serialization.
1036 *
1037 * Note: This only updates the timer expiry value and does not requeue
1038 * the timer.
1039 */
1040u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
1041{
1042 u64 orun = 1;
1043 ktime_t delta;
1044
1045 delta = ktime_sub(now, hrtimer_get_expires(timer));
1046
1047 if (delta < 0)
1048 return 0;
1049
1050 if (WARN_ON(timer->state & HRTIMER_STATE_ENQUEUED))
1051 return 0;
1052
1053 if (interval < hrtimer_resolution)
1054 interval = hrtimer_resolution;
1055
1056 if (unlikely(delta >= interval)) {
1057 s64 incr = ktime_to_ns(interval);
1058
1059 orun = ktime_divns(delta, incr);
1060 hrtimer_add_expires_ns(timer, incr * orun);
1061 if (hrtimer_get_expires_tv64(timer) > now)
1062 return orun;
1063 /*
1064 * This (and the ktime_add() below) is the
1065 * correction for exact:
1066 */
1067 orun++;
1068 }
1069 hrtimer_add_expires(timer, interval);
1070
1071 return orun;
1072}
1073EXPORT_SYMBOL_GPL(hrtimer_forward);
1074
1075/*
1076 * enqueue_hrtimer - internal function to (re)start a timer
1077 *
1078 * The timer is inserted in expiry order. Insertion into the
1079 * red black tree is O(log(n)). Must hold the base lock.
1080 *
1081 * Returns 1 when the new timer is the leftmost timer in the tree.
1082 */
1083static int enqueue_hrtimer(struct hrtimer *timer,
1084 struct hrtimer_clock_base *base,
1085 enum hrtimer_mode mode)
1086{
1087 debug_activate(timer, mode);
1088
1089 base->cpu_base->active_bases |= 1 << base->index;
1090
1091 /* Pairs with the lockless read in hrtimer_is_queued() */
1092 WRITE_ONCE(timer->state, HRTIMER_STATE_ENQUEUED);
1093
1094 return timerqueue_add(&base->active, &timer->node);
1095}
1096
1097/*
1098 * __remove_hrtimer - internal function to remove a timer
1099 *
1100 * Caller must hold the base lock.
1101 *
1102 * High resolution timer mode reprograms the clock event device when the
1103 * timer is the one which expires next. The caller can disable this by setting
1104 * reprogram to zero. This is useful, when the context does a reprogramming
1105 * anyway (e.g. timer interrupt)
1106 */
1107static void __remove_hrtimer(struct hrtimer *timer,
1108 struct hrtimer_clock_base *base,
1109 u8 newstate, int reprogram)
1110{
1111 struct hrtimer_cpu_base *cpu_base = base->cpu_base;
1112 u8 state = timer->state;
1113
1114 /* Pairs with the lockless read in hrtimer_is_queued() */
1115 WRITE_ONCE(timer->state, newstate);
1116 if (!(state & HRTIMER_STATE_ENQUEUED))
1117 return;
1118
1119 if (!timerqueue_del(&base->active, &timer->node))
1120 cpu_base->active_bases &= ~(1 << base->index);
1121
1122 /*
1123 * Note: If reprogram is false we do not update
1124 * cpu_base->next_timer. This happens when we remove the first
1125 * timer on a remote cpu. No harm as we never dereference
1126 * cpu_base->next_timer. So the worst thing what can happen is
1127 * an superfluous call to hrtimer_force_reprogram() on the
1128 * remote cpu later on if the same timer gets enqueued again.
1129 */
1130 if (reprogram && timer == cpu_base->next_timer)
1131 hrtimer_force_reprogram(cpu_base, 1);
1132}
1133
1134/*
1135 * remove hrtimer, called with base lock held
1136 */
1137static inline int
1138remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base,
1139 bool restart, bool keep_local)
1140{
1141 u8 state = timer->state;
1142
1143 if (state & HRTIMER_STATE_ENQUEUED) {
1144 bool reprogram;
1145
1146 /*
1147 * Remove the timer and force reprogramming when high
1148 * resolution mode is active and the timer is on the current
1149 * CPU. If we remove a timer on another CPU, reprogramming is
1150 * skipped. The interrupt event on this CPU is fired and
1151 * reprogramming happens in the interrupt handler. This is a
1152 * rare case and less expensive than a smp call.
1153 */
1154 debug_deactivate(timer);
1155 reprogram = base->cpu_base == this_cpu_ptr(&hrtimer_bases);
1156
1157 /*
1158 * If the timer is not restarted then reprogramming is
1159 * required if the timer is local. If it is local and about
1160 * to be restarted, avoid programming it twice (on removal
1161 * and a moment later when it's requeued).
1162 */
1163 if (!restart)
1164 state = HRTIMER_STATE_INACTIVE;
1165 else
1166 reprogram &= !keep_local;
1167
1168 __remove_hrtimer(timer, base, state, reprogram);
1169 return 1;
1170 }
1171 return 0;
1172}
1173
1174static inline ktime_t hrtimer_update_lowres(struct hrtimer *timer, ktime_t tim,
1175 const enum hrtimer_mode mode)
1176{
1177#ifdef CONFIG_TIME_LOW_RES
1178 /*
1179 * CONFIG_TIME_LOW_RES indicates that the system has no way to return
1180 * granular time values. For relative timers we add hrtimer_resolution
1181 * (i.e. one jiffie) to prevent short timeouts.
1182 */
1183 timer->is_rel = mode & HRTIMER_MODE_REL;
1184 if (timer->is_rel)
1185 tim = ktime_add_safe(tim, hrtimer_resolution);
1186#endif
1187 return tim;
1188}
1189
1190static void
1191hrtimer_update_softirq_timer(struct hrtimer_cpu_base *cpu_base, bool reprogram)
1192{
1193 ktime_t expires;
1194
1195 /*
1196 * Find the next SOFT expiration.
1197 */
1198 expires = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_SOFT);
1199
1200 /*
1201 * reprogramming needs to be triggered, even if the next soft
1202 * hrtimer expires at the same time than the next hard
1203 * hrtimer. cpu_base->softirq_expires_next needs to be updated!
1204 */
1205 if (expires == KTIME_MAX)
1206 return;
1207
1208 /*
1209 * cpu_base->*next_timer is recomputed by __hrtimer_get_next_event()
1210 * cpu_base->*expires_next is only set by hrtimer_reprogram()
1211 */
1212 hrtimer_reprogram(cpu_base->softirq_next_timer, reprogram);
1213}
1214
1215static int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1216 u64 delta_ns, const enum hrtimer_mode mode,
1217 struct hrtimer_clock_base *base)
1218{
1219 struct hrtimer_clock_base *new_base;
1220 bool force_local, first;
1221
1222 /*
1223 * If the timer is on the local cpu base and is the first expiring
1224 * timer then this might end up reprogramming the hardware twice
1225 * (on removal and on enqueue). To avoid that by prevent the
1226 * reprogram on removal, keep the timer local to the current CPU
1227 * and enforce reprogramming after it is queued no matter whether
1228 * it is the new first expiring timer again or not.
1229 */
1230 force_local = base->cpu_base == this_cpu_ptr(&hrtimer_bases);
1231 force_local &= base->cpu_base->next_timer == timer;
1232
1233 /*
1234 * Remove an active timer from the queue. In case it is not queued
1235 * on the current CPU, make sure that remove_hrtimer() updates the
1236 * remote data correctly.
1237 *
1238 * If it's on the current CPU and the first expiring timer, then
1239 * skip reprogramming, keep the timer local and enforce
1240 * reprogramming later if it was the first expiring timer. This
1241 * avoids programming the underlying clock event twice (once at
1242 * removal and once after enqueue).
1243 */
1244 remove_hrtimer(timer, base, true, force_local);
1245
1246 if (mode & HRTIMER_MODE_REL)
1247 tim = ktime_add_safe(tim, base->get_time());
1248
1249 tim = hrtimer_update_lowres(timer, tim, mode);
1250
1251 hrtimer_set_expires_range_ns(timer, tim, delta_ns);
1252
1253 /* Switch the timer base, if necessary: */
1254 if (!force_local) {
1255 new_base = switch_hrtimer_base(timer, base,
1256 mode & HRTIMER_MODE_PINNED);
1257 } else {
1258 new_base = base;
1259 }
1260
1261 first = enqueue_hrtimer(timer, new_base, mode);
1262 if (!force_local)
1263 return first;
1264
1265 /*
1266 * Timer was forced to stay on the current CPU to avoid
1267 * reprogramming on removal and enqueue. Force reprogram the
1268 * hardware by evaluating the new first expiring timer.
1269 */
1270 hrtimer_force_reprogram(new_base->cpu_base, 1);
1271 return 0;
1272}
1273
1274/**
1275 * hrtimer_start_range_ns - (re)start an hrtimer
1276 * @timer: the timer to be added
1277 * @tim: expiry time
1278 * @delta_ns: "slack" range for the timer
1279 * @mode: timer mode: absolute (HRTIMER_MODE_ABS) or
1280 * relative (HRTIMER_MODE_REL), and pinned (HRTIMER_MODE_PINNED);
1281 * softirq based mode is considered for debug purpose only!
1282 */
1283void hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1284 u64 delta_ns, const enum hrtimer_mode mode)
1285{
1286 struct hrtimer_clock_base *base;
1287 unsigned long flags;
1288
1289 /*
1290 * Check whether the HRTIMER_MODE_SOFT bit and hrtimer.is_soft
1291 * match on CONFIG_PREEMPT_RT = n. With PREEMPT_RT check the hard
1292 * expiry mode because unmarked timers are moved to softirq expiry.
1293 */
1294 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
1295 WARN_ON_ONCE(!(mode & HRTIMER_MODE_SOFT) ^ !timer->is_soft);
1296 else
1297 WARN_ON_ONCE(!(mode & HRTIMER_MODE_HARD) ^ !timer->is_hard);
1298
1299 base = lock_hrtimer_base(timer, &flags);
1300
1301 if (__hrtimer_start_range_ns(timer, tim, delta_ns, mode, base))
1302 hrtimer_reprogram(timer, true);
1303
1304 unlock_hrtimer_base(timer, &flags);
1305}
1306EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
1307
1308/**
1309 * hrtimer_try_to_cancel - try to deactivate a timer
1310 * @timer: hrtimer to stop
1311 *
1312 * Returns:
1313 *
1314 * * 0 when the timer was not active
1315 * * 1 when the timer was active
1316 * * -1 when the timer is currently executing the callback function and
1317 * cannot be stopped
1318 */
1319int hrtimer_try_to_cancel(struct hrtimer *timer)
1320{
1321 struct hrtimer_clock_base *base;
1322 unsigned long flags;
1323 int ret = -1;
1324
1325 /*
1326 * Check lockless first. If the timer is not active (neither
1327 * enqueued nor running the callback, nothing to do here. The
1328 * base lock does not serialize against a concurrent enqueue,
1329 * so we can avoid taking it.
1330 */
1331 if (!hrtimer_active(timer))
1332 return 0;
1333
1334 base = lock_hrtimer_base(timer, &flags);
1335
1336 if (!hrtimer_callback_running(timer))
1337 ret = remove_hrtimer(timer, base, false, false);
1338
1339 unlock_hrtimer_base(timer, &flags);
1340
1341 return ret;
1342
1343}
1344EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1345
1346#ifdef CONFIG_PREEMPT_RT
1347static void hrtimer_cpu_base_init_expiry_lock(struct hrtimer_cpu_base *base)
1348{
1349 spin_lock_init(&base->softirq_expiry_lock);
1350}
1351
1352static void hrtimer_cpu_base_lock_expiry(struct hrtimer_cpu_base *base)
1353{
1354 spin_lock(&base->softirq_expiry_lock);
1355}
1356
1357static void hrtimer_cpu_base_unlock_expiry(struct hrtimer_cpu_base *base)
1358{
1359 spin_unlock(&base->softirq_expiry_lock);
1360}
1361
1362/*
1363 * The counterpart to hrtimer_cancel_wait_running().
1364 *
1365 * If there is a waiter for cpu_base->expiry_lock, then it was waiting for
1366 * the timer callback to finish. Drop expiry_lock and reacquire it. That
1367 * allows the waiter to acquire the lock and make progress.
1368 */
1369static void hrtimer_sync_wait_running(struct hrtimer_cpu_base *cpu_base,
1370 unsigned long flags)
1371{
1372 if (atomic_read(&cpu_base->timer_waiters)) {
1373 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1374 spin_unlock(&cpu_base->softirq_expiry_lock);
1375 spin_lock(&cpu_base->softirq_expiry_lock);
1376 raw_spin_lock_irq(&cpu_base->lock);
1377 }
1378}
1379
1380/*
1381 * This function is called on PREEMPT_RT kernels when the fast path
1382 * deletion of a timer failed because the timer callback function was
1383 * running.
1384 *
1385 * This prevents priority inversion: if the soft irq thread is preempted
1386 * in the middle of a timer callback, then calling del_timer_sync() can
1387 * lead to two issues:
1388 *
1389 * - If the caller is on a remote CPU then it has to spin wait for the timer
1390 * handler to complete. This can result in unbound priority inversion.
1391 *
1392 * - If the caller originates from the task which preempted the timer
1393 * handler on the same CPU, then spin waiting for the timer handler to
1394 * complete is never going to end.
1395 */
1396void hrtimer_cancel_wait_running(const struct hrtimer *timer)
1397{
1398 /* Lockless read. Prevent the compiler from reloading it below */
1399 struct hrtimer_clock_base *base = READ_ONCE(timer->base);
1400
1401 /*
1402 * Just relax if the timer expires in hard interrupt context or if
1403 * it is currently on the migration base.
1404 */
1405 if (!timer->is_soft || is_migration_base(base)) {
1406 cpu_relax();
1407 return;
1408 }
1409
1410 /*
1411 * Mark the base as contended and grab the expiry lock, which is
1412 * held by the softirq across the timer callback. Drop the lock
1413 * immediately so the softirq can expire the next timer. In theory
1414 * the timer could already be running again, but that's more than
1415 * unlikely and just causes another wait loop.
1416 */
1417 atomic_inc(&base->cpu_base->timer_waiters);
1418 spin_lock_bh(&base->cpu_base->softirq_expiry_lock);
1419 atomic_dec(&base->cpu_base->timer_waiters);
1420 spin_unlock_bh(&base->cpu_base->softirq_expiry_lock);
1421}
1422#else
1423static inline void
1424hrtimer_cpu_base_init_expiry_lock(struct hrtimer_cpu_base *base) { }
1425static inline void
1426hrtimer_cpu_base_lock_expiry(struct hrtimer_cpu_base *base) { }
1427static inline void
1428hrtimer_cpu_base_unlock_expiry(struct hrtimer_cpu_base *base) { }
1429static inline void hrtimer_sync_wait_running(struct hrtimer_cpu_base *base,
1430 unsigned long flags) { }
1431#endif
1432
1433/**
1434 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1435 * @timer: the timer to be cancelled
1436 *
1437 * Returns:
1438 * 0 when the timer was not active
1439 * 1 when the timer was active
1440 */
1441int hrtimer_cancel(struct hrtimer *timer)
1442{
1443 int ret;
1444
1445 do {
1446 ret = hrtimer_try_to_cancel(timer);
1447
1448 if (ret < 0)
1449 hrtimer_cancel_wait_running(timer);
1450 } while (ret < 0);
1451 return ret;
1452}
1453EXPORT_SYMBOL_GPL(hrtimer_cancel);
1454
1455/**
1456 * __hrtimer_get_remaining - get remaining time for the timer
1457 * @timer: the timer to read
1458 * @adjust: adjust relative timers when CONFIG_TIME_LOW_RES=y
1459 */
1460ktime_t __hrtimer_get_remaining(const struct hrtimer *timer, bool adjust)
1461{
1462 unsigned long flags;
1463 ktime_t rem;
1464
1465 lock_hrtimer_base(timer, &flags);
1466 if (IS_ENABLED(CONFIG_TIME_LOW_RES) && adjust)
1467 rem = hrtimer_expires_remaining_adjusted(timer);
1468 else
1469 rem = hrtimer_expires_remaining(timer);
1470 unlock_hrtimer_base(timer, &flags);
1471
1472 return rem;
1473}
1474EXPORT_SYMBOL_GPL(__hrtimer_get_remaining);
1475
1476#ifdef CONFIG_NO_HZ_COMMON
1477/**
1478 * hrtimer_get_next_event - get the time until next expiry event
1479 *
1480 * Returns the next expiry time or KTIME_MAX if no timer is pending.
1481 */
1482u64 hrtimer_get_next_event(void)
1483{
1484 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1485 u64 expires = KTIME_MAX;
1486 unsigned long flags;
1487
1488 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1489
1490 if (!__hrtimer_hres_active(cpu_base))
1491 expires = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_ALL);
1492
1493 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1494
1495 return expires;
1496}
1497
1498/**
1499 * hrtimer_next_event_without - time until next expiry event w/o one timer
1500 * @exclude: timer to exclude
1501 *
1502 * Returns the next expiry time over all timers except for the @exclude one or
1503 * KTIME_MAX if none of them is pending.
1504 */
1505u64 hrtimer_next_event_without(const struct hrtimer *exclude)
1506{
1507 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1508 u64 expires = KTIME_MAX;
1509 unsigned long flags;
1510
1511 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1512
1513 if (__hrtimer_hres_active(cpu_base)) {
1514 unsigned int active;
1515
1516 if (!cpu_base->softirq_activated) {
1517 active = cpu_base->active_bases & HRTIMER_ACTIVE_SOFT;
1518 expires = __hrtimer_next_event_base(cpu_base, exclude,
1519 active, KTIME_MAX);
1520 }
1521 active = cpu_base->active_bases & HRTIMER_ACTIVE_HARD;
1522 expires = __hrtimer_next_event_base(cpu_base, exclude, active,
1523 expires);
1524 }
1525
1526 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1527
1528 return expires;
1529}
1530#endif
1531
1532static inline int hrtimer_clockid_to_base(clockid_t clock_id)
1533{
1534 if (likely(clock_id < MAX_CLOCKS)) {
1535 int base = hrtimer_clock_to_base_table[clock_id];
1536
1537 if (likely(base != HRTIMER_MAX_CLOCK_BASES))
1538 return base;
1539 }
1540 WARN(1, "Invalid clockid %d. Using MONOTONIC\n", clock_id);
1541 return HRTIMER_BASE_MONOTONIC;
1542}
1543
1544static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1545 enum hrtimer_mode mode)
1546{
1547 bool softtimer = !!(mode & HRTIMER_MODE_SOFT);
1548 struct hrtimer_cpu_base *cpu_base;
1549 int base;
1550
1551 /*
1552 * On PREEMPT_RT enabled kernels hrtimers which are not explicitly
1553 * marked for hard interrupt expiry mode are moved into soft
1554 * interrupt context for latency reasons and because the callbacks
1555 * can invoke functions which might sleep on RT, e.g. spin_lock().
1556 */
1557 if (IS_ENABLED(CONFIG_PREEMPT_RT) && !(mode & HRTIMER_MODE_HARD))
1558 softtimer = true;
1559
1560 memset(timer, 0, sizeof(struct hrtimer));
1561
1562 cpu_base = raw_cpu_ptr(&hrtimer_bases);
1563
1564 /*
1565 * POSIX magic: Relative CLOCK_REALTIME timers are not affected by
1566 * clock modifications, so they needs to become CLOCK_MONOTONIC to
1567 * ensure POSIX compliance.
1568 */
1569 if (clock_id == CLOCK_REALTIME && mode & HRTIMER_MODE_REL)
1570 clock_id = CLOCK_MONOTONIC;
1571
1572 base = softtimer ? HRTIMER_MAX_CLOCK_BASES / 2 : 0;
1573 base += hrtimer_clockid_to_base(clock_id);
1574 timer->is_soft = softtimer;
1575 timer->is_hard = !!(mode & HRTIMER_MODE_HARD);
1576 timer->base = &cpu_base->clock_base[base];
1577 timerqueue_init(&timer->node);
1578}
1579
1580/**
1581 * hrtimer_init - initialize a timer to the given clock
1582 * @timer: the timer to be initialized
1583 * @clock_id: the clock to be used
1584 * @mode: The modes which are relevant for initialization:
1585 * HRTIMER_MODE_ABS, HRTIMER_MODE_REL, HRTIMER_MODE_ABS_SOFT,
1586 * HRTIMER_MODE_REL_SOFT
1587 *
1588 * The PINNED variants of the above can be handed in,
1589 * but the PINNED bit is ignored as pinning happens
1590 * when the hrtimer is started
1591 */
1592void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1593 enum hrtimer_mode mode)
1594{
1595 debug_init(timer, clock_id, mode);
1596 __hrtimer_init(timer, clock_id, mode);
1597}
1598EXPORT_SYMBOL_GPL(hrtimer_init);
1599
1600/*
1601 * A timer is active, when it is enqueued into the rbtree or the
1602 * callback function is running or it's in the state of being migrated
1603 * to another cpu.
1604 *
1605 * It is important for this function to not return a false negative.
1606 */
1607bool hrtimer_active(const struct hrtimer *timer)
1608{
1609 struct hrtimer_clock_base *base;
1610 unsigned int seq;
1611
1612 do {
1613 base = READ_ONCE(timer->base);
1614 seq = raw_read_seqcount_begin(&base->seq);
1615
1616 if (timer->state != HRTIMER_STATE_INACTIVE ||
1617 base->running == timer)
1618 return true;
1619
1620 } while (read_seqcount_retry(&base->seq, seq) ||
1621 base != READ_ONCE(timer->base));
1622
1623 return false;
1624}
1625EXPORT_SYMBOL_GPL(hrtimer_active);
1626
1627/*
1628 * The write_seqcount_barrier()s in __run_hrtimer() split the thing into 3
1629 * distinct sections:
1630 *
1631 * - queued: the timer is queued
1632 * - callback: the timer is being ran
1633 * - post: the timer is inactive or (re)queued
1634 *
1635 * On the read side we ensure we observe timer->state and cpu_base->running
1636 * from the same section, if anything changed while we looked at it, we retry.
1637 * This includes timer->base changing because sequence numbers alone are
1638 * insufficient for that.
1639 *
1640 * The sequence numbers are required because otherwise we could still observe
1641 * a false negative if the read side got smeared over multiple consecutive
1642 * __run_hrtimer() invocations.
1643 */
1644
1645static void __run_hrtimer(struct hrtimer_cpu_base *cpu_base,
1646 struct hrtimer_clock_base *base,
1647 struct hrtimer *timer, ktime_t *now,
1648 unsigned long flags) __must_hold(&cpu_base->lock)
1649{
1650 enum hrtimer_restart (*fn)(struct hrtimer *);
1651 bool expires_in_hardirq;
1652 int restart;
1653
1654 lockdep_assert_held(&cpu_base->lock);
1655
1656 debug_deactivate(timer);
1657 base->running = timer;
1658
1659 /*
1660 * Separate the ->running assignment from the ->state assignment.
1661 *
1662 * As with a regular write barrier, this ensures the read side in
1663 * hrtimer_active() cannot observe base->running == NULL &&
1664 * timer->state == INACTIVE.
1665 */
1666 raw_write_seqcount_barrier(&base->seq);
1667
1668 __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE, 0);
1669 fn = timer->function;
1670
1671 /*
1672 * Clear the 'is relative' flag for the TIME_LOW_RES case. If the
1673 * timer is restarted with a period then it becomes an absolute
1674 * timer. If its not restarted it does not matter.
1675 */
1676 if (IS_ENABLED(CONFIG_TIME_LOW_RES))
1677 timer->is_rel = false;
1678
1679 /*
1680 * The timer is marked as running in the CPU base, so it is
1681 * protected against migration to a different CPU even if the lock
1682 * is dropped.
1683 */
1684 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1685 trace_hrtimer_expire_entry(timer, now);
1686 expires_in_hardirq = lockdep_hrtimer_enter(timer);
1687
1688 restart = fn(timer);
1689
1690 lockdep_hrtimer_exit(expires_in_hardirq);
1691 trace_hrtimer_expire_exit(timer);
1692 raw_spin_lock_irq(&cpu_base->lock);
1693
1694 /*
1695 * Note: We clear the running state after enqueue_hrtimer and
1696 * we do not reprogram the event hardware. Happens either in
1697 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1698 *
1699 * Note: Because we dropped the cpu_base->lock above,
1700 * hrtimer_start_range_ns() can have popped in and enqueued the timer
1701 * for us already.
1702 */
1703 if (restart != HRTIMER_NORESTART &&
1704 !(timer->state & HRTIMER_STATE_ENQUEUED))
1705 enqueue_hrtimer(timer, base, HRTIMER_MODE_ABS);
1706
1707 /*
1708 * Separate the ->running assignment from the ->state assignment.
1709 *
1710 * As with a regular write barrier, this ensures the read side in
1711 * hrtimer_active() cannot observe base->running.timer == NULL &&
1712 * timer->state == INACTIVE.
1713 */
1714 raw_write_seqcount_barrier(&base->seq);
1715
1716 WARN_ON_ONCE(base->running != timer);
1717 base->running = NULL;
1718}
1719
1720static void __hrtimer_run_queues(struct hrtimer_cpu_base *cpu_base, ktime_t now,
1721 unsigned long flags, unsigned int active_mask)
1722{
1723 struct hrtimer_clock_base *base;
1724 unsigned int active = cpu_base->active_bases & active_mask;
1725
1726 for_each_active_base(base, cpu_base, active) {
1727 struct timerqueue_node *node;
1728 ktime_t basenow;
1729
1730 basenow = ktime_add(now, base->offset);
1731
1732 while ((node = timerqueue_getnext(&base->active))) {
1733 struct hrtimer *timer;
1734
1735 timer = container_of(node, struct hrtimer, node);
1736
1737 /*
1738 * The immediate goal for using the softexpires is
1739 * minimizing wakeups, not running timers at the
1740 * earliest interrupt after their soft expiration.
1741 * This allows us to avoid using a Priority Search
1742 * Tree, which can answer a stabbing query for
1743 * overlapping intervals and instead use the simple
1744 * BST we already have.
1745 * We don't add extra wakeups by delaying timers that
1746 * are right-of a not yet expired timer, because that
1747 * timer will have to trigger a wakeup anyway.
1748 */
1749 if (basenow < hrtimer_get_softexpires_tv64(timer))
1750 break;
1751
1752 __run_hrtimer(cpu_base, base, timer, &basenow, flags);
1753 if (active_mask == HRTIMER_ACTIVE_SOFT)
1754 hrtimer_sync_wait_running(cpu_base, flags);
1755 }
1756 }
1757}
1758
1759static __latent_entropy void hrtimer_run_softirq(struct softirq_action *h)
1760{
1761 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1762 unsigned long flags;
1763 ktime_t now;
1764
1765 hrtimer_cpu_base_lock_expiry(cpu_base);
1766 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1767
1768 now = hrtimer_update_base(cpu_base);
1769 __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_SOFT);
1770
1771 cpu_base->softirq_activated = 0;
1772 hrtimer_update_softirq_timer(cpu_base, true);
1773
1774 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1775 hrtimer_cpu_base_unlock_expiry(cpu_base);
1776}
1777
1778#ifdef CONFIG_HIGH_RES_TIMERS
1779
1780/*
1781 * High resolution timer interrupt
1782 * Called with interrupts disabled
1783 */
1784void hrtimer_interrupt(struct clock_event_device *dev)
1785{
1786 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1787 ktime_t expires_next, now, entry_time, delta;
1788 unsigned long flags;
1789 int retries = 0;
1790
1791 BUG_ON(!cpu_base->hres_active);
1792 cpu_base->nr_events++;
1793 dev->next_event = KTIME_MAX;
1794
1795 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1796 entry_time = now = hrtimer_update_base(cpu_base);
1797retry:
1798 cpu_base->in_hrtirq = 1;
1799 /*
1800 * We set expires_next to KTIME_MAX here with cpu_base->lock
1801 * held to prevent that a timer is enqueued in our queue via
1802 * the migration code. This does not affect enqueueing of
1803 * timers which run their callback and need to be requeued on
1804 * this CPU.
1805 */
1806 cpu_base->expires_next = KTIME_MAX;
1807
1808 if (!ktime_before(now, cpu_base->softirq_expires_next)) {
1809 cpu_base->softirq_expires_next = KTIME_MAX;
1810 cpu_base->softirq_activated = 1;
1811 raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1812 }
1813
1814 __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_HARD);
1815
1816 /* Reevaluate the clock bases for the [soft] next expiry */
1817 expires_next = hrtimer_update_next_event(cpu_base);
1818 /*
1819 * Store the new expiry value so the migration code can verify
1820 * against it.
1821 */
1822 cpu_base->expires_next = expires_next;
1823 cpu_base->in_hrtirq = 0;
1824 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1825
1826 /* Reprogramming necessary ? */
1827 if (!tick_program_event(expires_next, 0)) {
1828 cpu_base->hang_detected = 0;
1829 return;
1830 }
1831
1832 /*
1833 * The next timer was already expired due to:
1834 * - tracing
1835 * - long lasting callbacks
1836 * - being scheduled away when running in a VM
1837 *
1838 * We need to prevent that we loop forever in the hrtimer
1839 * interrupt routine. We give it 3 attempts to avoid
1840 * overreacting on some spurious event.
1841 *
1842 * Acquire base lock for updating the offsets and retrieving
1843 * the current time.
1844 */
1845 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1846 now = hrtimer_update_base(cpu_base);
1847 cpu_base->nr_retries++;
1848 if (++retries < 3)
1849 goto retry;
1850 /*
1851 * Give the system a chance to do something else than looping
1852 * here. We stored the entry time, so we know exactly how long
1853 * we spent here. We schedule the next event this amount of
1854 * time away.
1855 */
1856 cpu_base->nr_hangs++;
1857 cpu_base->hang_detected = 1;
1858 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1859
1860 delta = ktime_sub(now, entry_time);
1861 if ((unsigned int)delta > cpu_base->max_hang_time)
1862 cpu_base->max_hang_time = (unsigned int) delta;
1863 /*
1864 * Limit it to a sensible value as we enforce a longer
1865 * delay. Give the CPU at least 100ms to catch up.
1866 */
1867 if (delta > 100 * NSEC_PER_MSEC)
1868 expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1869 else
1870 expires_next = ktime_add(now, delta);
1871 tick_program_event(expires_next, 1);
1872 pr_warn_once("hrtimer: interrupt took %llu ns\n", ktime_to_ns(delta));
1873}
1874
1875/* called with interrupts disabled */
1876static inline void __hrtimer_peek_ahead_timers(void)
1877{
1878 struct tick_device *td;
1879
1880 if (!hrtimer_hres_active())
1881 return;
1882
1883 td = this_cpu_ptr(&tick_cpu_device);
1884 if (td && td->evtdev)
1885 hrtimer_interrupt(td->evtdev);
1886}
1887
1888#else /* CONFIG_HIGH_RES_TIMERS */
1889
1890static inline void __hrtimer_peek_ahead_timers(void) { }
1891
1892#endif /* !CONFIG_HIGH_RES_TIMERS */
1893
1894/*
1895 * Called from run_local_timers in hardirq context every jiffy
1896 */
1897void hrtimer_run_queues(void)
1898{
1899 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1900 unsigned long flags;
1901 ktime_t now;
1902
1903 if (__hrtimer_hres_active(cpu_base))
1904 return;
1905
1906 /*
1907 * This _is_ ugly: We have to check periodically, whether we
1908 * can switch to highres and / or nohz mode. The clocksource
1909 * switch happens with xtime_lock held. Notification from
1910 * there only sets the check bit in the tick_oneshot code,
1911 * otherwise we might deadlock vs. xtime_lock.
1912 */
1913 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) {
1914 hrtimer_switch_to_hres();
1915 return;
1916 }
1917
1918 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1919 now = hrtimer_update_base(cpu_base);
1920
1921 if (!ktime_before(now, cpu_base->softirq_expires_next)) {
1922 cpu_base->softirq_expires_next = KTIME_MAX;
1923 cpu_base->softirq_activated = 1;
1924 raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1925 }
1926
1927 __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_HARD);
1928 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1929}
1930
1931/*
1932 * Sleep related functions:
1933 */
1934static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1935{
1936 struct hrtimer_sleeper *t =
1937 container_of(timer, struct hrtimer_sleeper, timer);
1938 struct task_struct *task = t->task;
1939
1940 t->task = NULL;
1941 if (task)
1942 wake_up_process(task);
1943
1944 return HRTIMER_NORESTART;
1945}
1946
1947/**
1948 * hrtimer_sleeper_start_expires - Start a hrtimer sleeper timer
1949 * @sl: sleeper to be started
1950 * @mode: timer mode abs/rel
1951 *
1952 * Wrapper around hrtimer_start_expires() for hrtimer_sleeper based timers
1953 * to allow PREEMPT_RT to tweak the delivery mode (soft/hardirq context)
1954 */
1955void hrtimer_sleeper_start_expires(struct hrtimer_sleeper *sl,
1956 enum hrtimer_mode mode)
1957{
1958 /*
1959 * Make the enqueue delivery mode check work on RT. If the sleeper
1960 * was initialized for hard interrupt delivery, force the mode bit.
1961 * This is a special case for hrtimer_sleepers because
1962 * hrtimer_init_sleeper() determines the delivery mode on RT so the
1963 * fiddling with this decision is avoided at the call sites.
1964 */
1965 if (IS_ENABLED(CONFIG_PREEMPT_RT) && sl->timer.is_hard)
1966 mode |= HRTIMER_MODE_HARD;
1967
1968 hrtimer_start_expires(&sl->timer, mode);
1969}
1970EXPORT_SYMBOL_GPL(hrtimer_sleeper_start_expires);
1971
1972static void __hrtimer_init_sleeper(struct hrtimer_sleeper *sl,
1973 clockid_t clock_id, enum hrtimer_mode mode)
1974{
1975 /*
1976 * On PREEMPT_RT enabled kernels hrtimers which are not explicitly
1977 * marked for hard interrupt expiry mode are moved into soft
1978 * interrupt context either for latency reasons or because the
1979 * hrtimer callback takes regular spinlocks or invokes other
1980 * functions which are not suitable for hard interrupt context on
1981 * PREEMPT_RT.
1982 *
1983 * The hrtimer_sleeper callback is RT compatible in hard interrupt
1984 * context, but there is a latency concern: Untrusted userspace can
1985 * spawn many threads which arm timers for the same expiry time on
1986 * the same CPU. That causes a latency spike due to the wakeup of
1987 * a gazillion threads.
1988 *
1989 * OTOH, privileged real-time user space applications rely on the
1990 * low latency of hard interrupt wakeups. If the current task is in
1991 * a real-time scheduling class, mark the mode for hard interrupt
1992 * expiry.
1993 */
1994 if (IS_ENABLED(CONFIG_PREEMPT_RT)) {
1995 if (task_is_realtime(current) && !(mode & HRTIMER_MODE_SOFT))
1996 mode |= HRTIMER_MODE_HARD;
1997 }
1998
1999 __hrtimer_init(&sl->timer, clock_id, mode);
2000 sl->timer.function = hrtimer_wakeup;
2001 sl->task = current;
2002}
2003
2004/**
2005 * hrtimer_init_sleeper - initialize sleeper to the given clock
2006 * @sl: sleeper to be initialized
2007 * @clock_id: the clock to be used
2008 * @mode: timer mode abs/rel
2009 */
2010void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, clockid_t clock_id,
2011 enum hrtimer_mode mode)
2012{
2013 debug_init(&sl->timer, clock_id, mode);
2014 __hrtimer_init_sleeper(sl, clock_id, mode);
2015
2016}
2017EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
2018
2019int nanosleep_copyout(struct restart_block *restart, struct timespec64 *ts)
2020{
2021 switch(restart->nanosleep.type) {
2022#ifdef CONFIG_COMPAT_32BIT_TIME
2023 case TT_COMPAT:
2024 if (put_old_timespec32(ts, restart->nanosleep.compat_rmtp))
2025 return -EFAULT;
2026 break;
2027#endif
2028 case TT_NATIVE:
2029 if (put_timespec64(ts, restart->nanosleep.rmtp))
2030 return -EFAULT;
2031 break;
2032 default:
2033 BUG();
2034 }
2035 return -ERESTART_RESTARTBLOCK;
2036}
2037
2038static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
2039{
2040 struct restart_block *restart;
2041
2042 do {
2043 set_current_state(TASK_INTERRUPTIBLE|TASK_FREEZABLE);
2044 hrtimer_sleeper_start_expires(t, mode);
2045
2046 if (likely(t->task))
2047 schedule();
2048
2049 hrtimer_cancel(&t->timer);
2050 mode = HRTIMER_MODE_ABS;
2051
2052 } while (t->task && !signal_pending(current));
2053
2054 __set_current_state(TASK_RUNNING);
2055
2056 if (!t->task)
2057 return 0;
2058
2059 restart = &current->restart_block;
2060 if (restart->nanosleep.type != TT_NONE) {
2061 ktime_t rem = hrtimer_expires_remaining(&t->timer);
2062 struct timespec64 rmt;
2063
2064 if (rem <= 0)
2065 return 0;
2066 rmt = ktime_to_timespec64(rem);
2067
2068 return nanosleep_copyout(restart, &rmt);
2069 }
2070 return -ERESTART_RESTARTBLOCK;
2071}
2072
2073static long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
2074{
2075 struct hrtimer_sleeper t;
2076 int ret;
2077
2078 hrtimer_init_sleeper_on_stack(&t, restart->nanosleep.clockid,
2079 HRTIMER_MODE_ABS);
2080 hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
2081 ret = do_nanosleep(&t, HRTIMER_MODE_ABS);
2082 destroy_hrtimer_on_stack(&t.timer);
2083 return ret;
2084}
2085
2086long hrtimer_nanosleep(ktime_t rqtp, const enum hrtimer_mode mode,
2087 const clockid_t clockid)
2088{
2089 struct restart_block *restart;
2090 struct hrtimer_sleeper t;
2091 int ret = 0;
2092 u64 slack;
2093
2094 slack = current->timer_slack_ns;
2095 if (rt_task(current))
2096 slack = 0;
2097
2098 hrtimer_init_sleeper_on_stack(&t, clockid, mode);
2099 hrtimer_set_expires_range_ns(&t.timer, rqtp, slack);
2100 ret = do_nanosleep(&t, mode);
2101 if (ret != -ERESTART_RESTARTBLOCK)
2102 goto out;
2103
2104 /* Absolute timers do not update the rmtp value and restart: */
2105 if (mode == HRTIMER_MODE_ABS) {
2106 ret = -ERESTARTNOHAND;
2107 goto out;
2108 }
2109
2110 restart = &current->restart_block;
2111 restart->nanosleep.clockid = t.timer.base->clockid;
2112 restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
2113 set_restart_fn(restart, hrtimer_nanosleep_restart);
2114out:
2115 destroy_hrtimer_on_stack(&t.timer);
2116 return ret;
2117}
2118
2119#ifdef CONFIG_64BIT
2120
2121SYSCALL_DEFINE2(nanosleep, struct __kernel_timespec __user *, rqtp,
2122 struct __kernel_timespec __user *, rmtp)
2123{
2124 struct timespec64 tu;
2125
2126 if (get_timespec64(&tu, rqtp))
2127 return -EFAULT;
2128
2129 if (!timespec64_valid(&tu))
2130 return -EINVAL;
2131
2132 current->restart_block.fn = do_no_restart_syscall;
2133 current->restart_block.nanosleep.type = rmtp ? TT_NATIVE : TT_NONE;
2134 current->restart_block.nanosleep.rmtp = rmtp;
2135 return hrtimer_nanosleep(timespec64_to_ktime(tu), HRTIMER_MODE_REL,
2136 CLOCK_MONOTONIC);
2137}
2138
2139#endif
2140
2141#ifdef CONFIG_COMPAT_32BIT_TIME
2142
2143SYSCALL_DEFINE2(nanosleep_time32, struct old_timespec32 __user *, rqtp,
2144 struct old_timespec32 __user *, rmtp)
2145{
2146 struct timespec64 tu;
2147
2148 if (get_old_timespec32(&tu, rqtp))
2149 return -EFAULT;
2150
2151 if (!timespec64_valid(&tu))
2152 return -EINVAL;
2153
2154 current->restart_block.fn = do_no_restart_syscall;
2155 current->restart_block.nanosleep.type = rmtp ? TT_COMPAT : TT_NONE;
2156 current->restart_block.nanosleep.compat_rmtp = rmtp;
2157 return hrtimer_nanosleep(timespec64_to_ktime(tu), HRTIMER_MODE_REL,
2158 CLOCK_MONOTONIC);
2159}
2160#endif
2161
2162/*
2163 * Functions related to boot-time initialization:
2164 */
2165int hrtimers_prepare_cpu(unsigned int cpu)
2166{
2167 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
2168 int i;
2169
2170 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
2171 struct hrtimer_clock_base *clock_b = &cpu_base->clock_base[i];
2172
2173 clock_b->cpu_base = cpu_base;
2174 seqcount_raw_spinlock_init(&clock_b->seq, &cpu_base->lock);
2175 timerqueue_init_head(&clock_b->active);
2176 }
2177
2178 cpu_base->cpu = cpu;
2179 cpu_base->active_bases = 0;
2180 cpu_base->hres_active = 0;
2181 cpu_base->hang_detected = 0;
2182 cpu_base->next_timer = NULL;
2183 cpu_base->softirq_next_timer = NULL;
2184 cpu_base->expires_next = KTIME_MAX;
2185 cpu_base->softirq_expires_next = KTIME_MAX;
2186 hrtimer_cpu_base_init_expiry_lock(cpu_base);
2187 return 0;
2188}
2189
2190#ifdef CONFIG_HOTPLUG_CPU
2191
2192static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
2193 struct hrtimer_clock_base *new_base)
2194{
2195 struct hrtimer *timer;
2196 struct timerqueue_node *node;
2197
2198 while ((node = timerqueue_getnext(&old_base->active))) {
2199 timer = container_of(node, struct hrtimer, node);
2200 BUG_ON(hrtimer_callback_running(timer));
2201 debug_deactivate(timer);
2202
2203 /*
2204 * Mark it as ENQUEUED not INACTIVE otherwise the
2205 * timer could be seen as !active and just vanish away
2206 * under us on another CPU
2207 */
2208 __remove_hrtimer(timer, old_base, HRTIMER_STATE_ENQUEUED, 0);
2209 timer->base = new_base;
2210 /*
2211 * Enqueue the timers on the new cpu. This does not
2212 * reprogram the event device in case the timer
2213 * expires before the earliest on this CPU, but we run
2214 * hrtimer_interrupt after we migrated everything to
2215 * sort out already expired timers and reprogram the
2216 * event device.
2217 */
2218 enqueue_hrtimer(timer, new_base, HRTIMER_MODE_ABS);
2219 }
2220}
2221
2222int hrtimers_cpu_dying(unsigned int dying_cpu)
2223{
2224 struct hrtimer_cpu_base *old_base, *new_base;
2225 int i, ncpu = cpumask_first(cpu_active_mask);
2226
2227 tick_cancel_sched_timer(dying_cpu);
2228
2229 old_base = this_cpu_ptr(&hrtimer_bases);
2230 new_base = &per_cpu(hrtimer_bases, ncpu);
2231
2232 /*
2233 * The caller is globally serialized and nobody else
2234 * takes two locks at once, deadlock is not possible.
2235 */
2236 raw_spin_lock(&old_base->lock);
2237 raw_spin_lock_nested(&new_base->lock, SINGLE_DEPTH_NESTING);
2238
2239 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
2240 migrate_hrtimer_list(&old_base->clock_base[i],
2241 &new_base->clock_base[i]);
2242 }
2243
2244 /*
2245 * The migration might have changed the first expiring softirq
2246 * timer on this CPU. Update it.
2247 */
2248 __hrtimer_get_next_event(new_base, HRTIMER_ACTIVE_SOFT);
2249 /* Tell the other CPU to retrigger the next event */
2250 smp_call_function_single(ncpu, retrigger_next_event, NULL, 0);
2251
2252 raw_spin_unlock(&new_base->lock);
2253 raw_spin_unlock(&old_base->lock);
2254
2255 return 0;
2256}
2257
2258#endif /* CONFIG_HOTPLUG_CPU */
2259
2260void __init hrtimers_init(void)
2261{
2262 hrtimers_prepare_cpu(smp_processor_id());
2263 open_softirq(HRTIMER_SOFTIRQ, hrtimer_run_softirq);
2264}
2265
2266/**
2267 * schedule_hrtimeout_range_clock - sleep until timeout
2268 * @expires: timeout value (ktime_t)
2269 * @delta: slack in expires timeout (ktime_t) for SCHED_OTHER tasks
2270 * @mode: timer mode
2271 * @clock_id: timer clock to be used
2272 */
2273int __sched
2274schedule_hrtimeout_range_clock(ktime_t *expires, u64 delta,
2275 const enum hrtimer_mode mode, clockid_t clock_id)
2276{
2277 struct hrtimer_sleeper t;
2278
2279 /*
2280 * Optimize when a zero timeout value is given. It does not
2281 * matter whether this is an absolute or a relative time.
2282 */
2283 if (expires && *expires == 0) {
2284 __set_current_state(TASK_RUNNING);
2285 return 0;
2286 }
2287
2288 /*
2289 * A NULL parameter means "infinite"
2290 */
2291 if (!expires) {
2292 schedule();
2293 return -EINTR;
2294 }
2295
2296 /*
2297 * Override any slack passed by the user if under
2298 * rt contraints.
2299 */
2300 if (rt_task(current))
2301 delta = 0;
2302
2303 hrtimer_init_sleeper_on_stack(&t, clock_id, mode);
2304 hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
2305 hrtimer_sleeper_start_expires(&t, mode);
2306
2307 if (likely(t.task))
2308 schedule();
2309
2310 hrtimer_cancel(&t.timer);
2311 destroy_hrtimer_on_stack(&t.timer);
2312
2313 __set_current_state(TASK_RUNNING);
2314
2315 return !t.task ? 0 : -EINTR;
2316}
2317EXPORT_SYMBOL_GPL(schedule_hrtimeout_range_clock);
2318
2319/**
2320 * schedule_hrtimeout_range - sleep until timeout
2321 * @expires: timeout value (ktime_t)
2322 * @delta: slack in expires timeout (ktime_t) for SCHED_OTHER tasks
2323 * @mode: timer mode
2324 *
2325 * Make the current task sleep until the given expiry time has
2326 * elapsed. The routine will return immediately unless
2327 * the current task state has been set (see set_current_state()).
2328 *
2329 * The @delta argument gives the kernel the freedom to schedule the
2330 * actual wakeup to a time that is both power and performance friendly
2331 * for regular (non RT/DL) tasks.
2332 * The kernel give the normal best effort behavior for "@expires+@delta",
2333 * but may decide to fire the timer earlier, but no earlier than @expires.
2334 *
2335 * You can set the task state as follows -
2336 *
2337 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
2338 * pass before the routine returns unless the current task is explicitly
2339 * woken up, (e.g. by wake_up_process()).
2340 *
2341 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
2342 * delivered to the current task or the current task is explicitly woken
2343 * up.
2344 *
2345 * The current task state is guaranteed to be TASK_RUNNING when this
2346 * routine returns.
2347 *
2348 * Returns 0 when the timer has expired. If the task was woken before the
2349 * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or
2350 * by an explicit wakeup, it returns -EINTR.
2351 */
2352int __sched schedule_hrtimeout_range(ktime_t *expires, u64 delta,
2353 const enum hrtimer_mode mode)
2354{
2355 return schedule_hrtimeout_range_clock(expires, delta, mode,
2356 CLOCK_MONOTONIC);
2357}
2358EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
2359
2360/**
2361 * schedule_hrtimeout - sleep until timeout
2362 * @expires: timeout value (ktime_t)
2363 * @mode: timer mode
2364 *
2365 * Make the current task sleep until the given expiry time has
2366 * elapsed. The routine will return immediately unless
2367 * the current task state has been set (see set_current_state()).
2368 *
2369 * You can set the task state as follows -
2370 *
2371 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
2372 * pass before the routine returns unless the current task is explicitly
2373 * woken up, (e.g. by wake_up_process()).
2374 *
2375 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
2376 * delivered to the current task or the current task is explicitly woken
2377 * up.
2378 *
2379 * The current task state is guaranteed to be TASK_RUNNING when this
2380 * routine returns.
2381 *
2382 * Returns 0 when the timer has expired. If the task was woken before the
2383 * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or
2384 * by an explicit wakeup, it returns -EINTR.
2385 */
2386int __sched schedule_hrtimeout(ktime_t *expires,
2387 const enum hrtimer_mode mode)
2388{
2389 return schedule_hrtimeout_range(expires, 0, mode);
2390}
2391EXPORT_SYMBOL_GPL(schedule_hrtimeout);