projects/linux·blob v6.10 · refs

2374 lines · 66.2 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/sched/isolation.h>
42#include <linux/timer.h>
43#include <linux/freezer.h>
44#include <linux/compat.h>
45
46#include <linux/uaccess.h>
47
48#include <trace/events/timer.h>
49
50#include "tick-internal.h"
51
52/*
53 * Masks for selecting the soft and hard context timers from
54 * cpu_base->active
55 */
56#define MASK_SHIFT (HRTIMER_BASE_MONOTONIC_SOFT)
57#define HRTIMER_ACTIVE_HARD ((1U << MASK_SHIFT) - 1)
58#define HRTIMER_ACTIVE_SOFT (HRTIMER_ACTIVE_HARD << MASK_SHIFT)
59#define HRTIMER_ACTIVE_ALL (HRTIMER_ACTIVE_SOFT | HRTIMER_ACTIVE_HARD)
60
61/*
62 * The timer bases:
63 *
64 * There are more clockids than hrtimer bases. Thus, we index
65 * into the timer bases by the hrtimer_base_type enum. When trying
66 * to reach a base using a clockid, hrtimer_clockid_to_base()
67 * is used to convert from clockid to the proper hrtimer_base_type.
68 */
69DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
70{
71 .lock = __RAW_SPIN_LOCK_UNLOCKED(hrtimer_bases.lock),
72 .clock_base =
73 {
74 {
75 .index = HRTIMER_BASE_MONOTONIC,
76 .clockid = CLOCK_MONOTONIC,
77 .get_time = &ktime_get,
78 },
79 {
80 .index = HRTIMER_BASE_REALTIME,
81 .clockid = CLOCK_REALTIME,
82 .get_time = &ktime_get_real,
83 },
84 {
85 .index = HRTIMER_BASE_BOOTTIME,
86 .clockid = CLOCK_BOOTTIME,
87 .get_time = &ktime_get_boottime,
88 },
89 {
90 .index = HRTIMER_BASE_TAI,
91 .clockid = CLOCK_TAI,
92 .get_time = &ktime_get_clocktai,
93 },
94 {
95 .index = HRTIMER_BASE_MONOTONIC_SOFT,
96 .clockid = CLOCK_MONOTONIC,
97 .get_time = &ktime_get,
98 },
99 {
100 .index = HRTIMER_BASE_REALTIME_SOFT,
101 .clockid = CLOCK_REALTIME,
102 .get_time = &ktime_get_real,
103 },
104 {
105 .index = HRTIMER_BASE_BOOTTIME_SOFT,
106 .clockid = CLOCK_BOOTTIME,
107 .get_time = &ktime_get_boottime,
108 },
109 {
110 .index = HRTIMER_BASE_TAI_SOFT,
111 .clockid = CLOCK_TAI,
112 .get_time = &ktime_get_clocktai,
113 },
114 }
115};
116
117static const int hrtimer_clock_to_base_table[MAX_CLOCKS] = {
118 /* Make sure we catch unsupported clockids */
119 [0 ... MAX_CLOCKS - 1] = HRTIMER_MAX_CLOCK_BASES,
120
121 [CLOCK_REALTIME] = HRTIMER_BASE_REALTIME,
122 [CLOCK_MONOTONIC] = HRTIMER_BASE_MONOTONIC,
123 [CLOCK_BOOTTIME] = HRTIMER_BASE_BOOTTIME,
124 [CLOCK_TAI] = HRTIMER_BASE_TAI,
125};
126
127/*
128 * Functions and macros which are different for UP/SMP systems are kept in a
129 * single place
130 */
131#ifdef CONFIG_SMP
132
133/*
134 * We require the migration_base for lock_hrtimer_base()/switch_hrtimer_base()
135 * such that hrtimer_callback_running() can unconditionally dereference
136 * timer->base->cpu_base
137 */
138static struct hrtimer_cpu_base migration_cpu_base = {
139 .clock_base = { {
140 .cpu_base = &migration_cpu_base,
141 .seq = SEQCNT_RAW_SPINLOCK_ZERO(migration_cpu_base.seq,
142 &migration_cpu_base.lock),
143 }, },
144};
145
146#define migration_base migration_cpu_base.clock_base[0]
147
148static inline bool is_migration_base(struct hrtimer_clock_base *base)
149{
150 return base == &migration_base;
151}
152
153/*
154 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
155 * means that all timers which are tied to this base via timer->base are
156 * locked, and the base itself is locked too.
157 *
158 * So __run_timers/migrate_timers can safely modify all timers which could
159 * be found on the lists/queues.
160 *
161 * When the timer's base is locked, and the timer removed from list, it is
162 * possible to set timer->base = &migration_base and drop the lock: the timer
163 * remains locked.
164 */
165static
166struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
167 unsigned long *flags)
168 __acquires(&timer->base->lock)
169{
170 struct hrtimer_clock_base *base;
171
172 for (;;) {
173 base = READ_ONCE(timer->base);
174 if (likely(base != &migration_base)) {
175 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
176 if (likely(base == timer->base))
177 return base;
178 /* The timer has migrated to another CPU: */
179 raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
180 }
181 cpu_relax();
182 }
183}
184
185/*
186 * We do not migrate the timer when it is expiring before the next
187 * event on the target cpu. When high resolution is enabled, we cannot
188 * reprogram the target cpu hardware and we would cause it to fire
189 * late. To keep it simple, we handle the high resolution enabled and
190 * disabled case similar.
191 *
192 * Called with cpu_base->lock of target cpu held.
193 */
194static int
195hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base)
196{
197 ktime_t expires;
198
199 expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset);
200 return expires < new_base->cpu_base->expires_next;
201}
202
203static inline
204struct hrtimer_cpu_base *get_target_base(struct hrtimer_cpu_base *base,
205 int pinned)
206{
207#if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
208 if (static_branch_likely(&timers_migration_enabled) && !pinned)
209 return &per_cpu(hrtimer_bases, get_nohz_timer_target());
210#endif
211 return base;
212}
213
214/*
215 * We switch the timer base to a power-optimized selected CPU target,
216 * if:
217 * - NO_HZ_COMMON is enabled
218 * - timer migration is enabled
219 * - the timer callback is not running
220 * - the timer is not the first expiring timer on the new target
221 *
222 * If one of the above requirements is not fulfilled we move the timer
223 * to the current CPU or leave it on the previously assigned CPU if
224 * the timer callback is currently running.
225 */
226static inline struct hrtimer_clock_base *
227switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
228 int pinned)
229{
230 struct hrtimer_cpu_base *new_cpu_base, *this_cpu_base;
231 struct hrtimer_clock_base *new_base;
232 int basenum = base->index;
233
234 this_cpu_base = this_cpu_ptr(&hrtimer_bases);
235 new_cpu_base = get_target_base(this_cpu_base, pinned);
236again:
237 new_base = &new_cpu_base->clock_base[basenum];
238
239 if (base != new_base) {
240 /*
241 * We are trying to move timer to new_base.
242 * However we can't change timer's base while it is running,
243 * so we keep it on the same CPU. No hassle vs. reprogramming
244 * the event source in the high resolution case. The softirq
245 * code will take care of this when the timer function has
246 * completed. There is no conflict as we hold the lock until
247 * the timer is enqueued.
248 */
249 if (unlikely(hrtimer_callback_running(timer)))
250 return base;
251
252 /* See the comment in lock_hrtimer_base() */
253 WRITE_ONCE(timer->base, &migration_base);
254 raw_spin_unlock(&base->cpu_base->lock);
255 raw_spin_lock(&new_base->cpu_base->lock);
256
257 if (new_cpu_base != this_cpu_base &&
258 hrtimer_check_target(timer, new_base)) {
259 raw_spin_unlock(&new_base->cpu_base->lock);
260 raw_spin_lock(&base->cpu_base->lock);
261 new_cpu_base = this_cpu_base;
262 WRITE_ONCE(timer->base, base);
263 goto again;
264 }
265 WRITE_ONCE(timer->base, new_base);
266 } else {
267 if (new_cpu_base != this_cpu_base &&
268 hrtimer_check_target(timer, new_base)) {
269 new_cpu_base = this_cpu_base;
270 goto again;
271 }
272 }
273 return new_base;
274}
275
276#else /* CONFIG_SMP */
277
278static inline bool is_migration_base(struct hrtimer_clock_base *base)
279{
280 return false;
281}
282
283static inline struct hrtimer_clock_base *
284lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
285 __acquires(&timer->base->cpu_base->lock)
286{
287 struct hrtimer_clock_base *base = timer->base;
288
289 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
290
291 return base;
292}
293
294# define switch_hrtimer_base(t, b, p) (b)
295
296#endif /* !CONFIG_SMP */
297
298/*
299 * Functions for the union type storage format of ktime_t which are
300 * too large for inlining:
301 */
302#if BITS_PER_LONG < 64
303/*
304 * Divide a ktime value by a nanosecond value
305 */
306s64 __ktime_divns(const ktime_t kt, s64 div)
307{
308 int sft = 0;
309 s64 dclc;
310 u64 tmp;
311
312 dclc = ktime_to_ns(kt);
313 tmp = dclc < 0 ? -dclc : dclc;
314
315 /* Make sure the divisor is less than 2^32: */
316 while (div >> 32) {
317 sft++;
318 div >>= 1;
319 }
320 tmp >>= sft;
321 do_div(tmp, (u32) div);
322 return dclc < 0 ? -tmp : tmp;
323}
324EXPORT_SYMBOL_GPL(__ktime_divns);
325#endif /* BITS_PER_LONG >= 64 */
326
327/*
328 * Add two ktime values and do a safety check for overflow:
329 */
330ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
331{
332 ktime_t res = ktime_add_unsafe(lhs, rhs);
333
334 /*
335 * We use KTIME_SEC_MAX here, the maximum timeout which we can
336 * return to user space in a timespec:
337 */
338 if (res < 0 || res < lhs || res < rhs)
339 res = ktime_set(KTIME_SEC_MAX, 0);
340
341 return res;
342}
343
344EXPORT_SYMBOL_GPL(ktime_add_safe);
345
346#ifdef CONFIG_DEBUG_OBJECTS_TIMERS
347
348static const struct debug_obj_descr hrtimer_debug_descr;
349
350static void *hrtimer_debug_hint(void *addr)
351{
352 return ((struct hrtimer *) addr)->function;
353}
354
355/*
356 * fixup_init is called when:
357 * - an active object is initialized
358 */
359static bool hrtimer_fixup_init(void *addr, enum debug_obj_state state)
360{
361 struct hrtimer *timer = addr;
362
363 switch (state) {
364 case ODEBUG_STATE_ACTIVE:
365 hrtimer_cancel(timer);
366 debug_object_init(timer, &hrtimer_debug_descr);
367 return true;
368 default:
369 return false;
370 }
371}
372
373/*
374 * fixup_activate is called when:
375 * - an active object is activated
376 * - an unknown non-static object is activated
377 */
378static bool hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
379{
380 switch (state) {
381 case ODEBUG_STATE_ACTIVE:
382 WARN_ON(1);
383 fallthrough;
384 default:
385 return false;
386 }
387}
388
389/*
390 * fixup_free is called when:
391 * - an active object is freed
392 */
393static bool hrtimer_fixup_free(void *addr, enum debug_obj_state state)
394{
395 struct hrtimer *timer = addr;
396
397 switch (state) {
398 case ODEBUG_STATE_ACTIVE:
399 hrtimer_cancel(timer);
400 debug_object_free(timer, &hrtimer_debug_descr);
401 return true;
402 default:
403 return false;
404 }
405}
406
407static const struct debug_obj_descr hrtimer_debug_descr = {
408 .name = "hrtimer",
409 .debug_hint = hrtimer_debug_hint,
410 .fixup_init = hrtimer_fixup_init,
411 .fixup_activate = hrtimer_fixup_activate,
412 .fixup_free = hrtimer_fixup_free,
413};
414
415static inline void debug_hrtimer_init(struct hrtimer *timer)
416{
417 debug_object_init(timer, &hrtimer_debug_descr);
418}
419
420static inline void debug_hrtimer_activate(struct hrtimer *timer,
421 enum hrtimer_mode mode)
422{
423 debug_object_activate(timer, &hrtimer_debug_descr);
424}
425
426static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
427{
428 debug_object_deactivate(timer, &hrtimer_debug_descr);
429}
430
431static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
432 enum hrtimer_mode mode);
433
434void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
435 enum hrtimer_mode mode)
436{
437 debug_object_init_on_stack(timer, &hrtimer_debug_descr);
438 __hrtimer_init(timer, clock_id, mode);
439}
440EXPORT_SYMBOL_GPL(hrtimer_init_on_stack);
441
442static void __hrtimer_init_sleeper(struct hrtimer_sleeper *sl,
443 clockid_t clock_id, enum hrtimer_mode mode);
444
445void hrtimer_init_sleeper_on_stack(struct hrtimer_sleeper *sl,
446 clockid_t clock_id, enum hrtimer_mode mode)
447{
448 debug_object_init_on_stack(&sl->timer, &hrtimer_debug_descr);
449 __hrtimer_init_sleeper(sl, clock_id, mode);
450}
451EXPORT_SYMBOL_GPL(hrtimer_init_sleeper_on_stack);
452
453void destroy_hrtimer_on_stack(struct hrtimer *timer)
454{
455 debug_object_free(timer, &hrtimer_debug_descr);
456}
457EXPORT_SYMBOL_GPL(destroy_hrtimer_on_stack);
458
459#else
460
461static inline void debug_hrtimer_init(struct hrtimer *timer) { }
462static inline void debug_hrtimer_activate(struct hrtimer *timer,
463 enum hrtimer_mode mode) { }
464static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
465#endif
466
467static inline void
468debug_init(struct hrtimer *timer, clockid_t clockid,
469 enum hrtimer_mode mode)
470{
471 debug_hrtimer_init(timer);
472 trace_hrtimer_init(timer, clockid, mode);
473}
474
475static inline void debug_activate(struct hrtimer *timer,
476 enum hrtimer_mode mode)
477{
478 debug_hrtimer_activate(timer, mode);
479 trace_hrtimer_start(timer, mode);
480}
481
482static inline void debug_deactivate(struct hrtimer *timer)
483{
484 debug_hrtimer_deactivate(timer);
485 trace_hrtimer_cancel(timer);
486}
487
488static struct hrtimer_clock_base *
489__next_base(struct hrtimer_cpu_base *cpu_base, unsigned int *active)
490{
491 unsigned int idx;
492
493 if (!*active)
494 return NULL;
495
496 idx = __ffs(*active);
497 *active &= ~(1U << idx);
498
499 return &cpu_base->clock_base[idx];
500}
501
502#define for_each_active_base(base, cpu_base, active) \
503 while ((base = __next_base((cpu_base), &(active))))
504
505static ktime_t __hrtimer_next_event_base(struct hrtimer_cpu_base *cpu_base,
506 const struct hrtimer *exclude,
507 unsigned int active,
508 ktime_t expires_next)
509{
510 struct hrtimer_clock_base *base;
511 ktime_t expires;
512
513 for_each_active_base(base, cpu_base, active) {
514 struct timerqueue_node *next;
515 struct hrtimer *timer;
516
517 next = timerqueue_getnext(&base->active);
518 timer = container_of(next, struct hrtimer, node);
519 if (timer == exclude) {
520 /* Get to the next timer in the queue. */
521 next = timerqueue_iterate_next(next);
522 if (!next)
523 continue;
524
525 timer = container_of(next, struct hrtimer, node);
526 }
527 expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
528 if (expires < expires_next) {
529 expires_next = expires;
530
531 /* Skip cpu_base update if a timer is being excluded. */
532 if (exclude)
533 continue;
534
535 if (timer->is_soft)
536 cpu_base->softirq_next_timer = timer;
537 else
538 cpu_base->next_timer = timer;
539 }
540 }
541 /*
542 * clock_was_set() might have changed base->offset of any of
543 * the clock bases so the result might be negative. Fix it up
544 * to prevent a false positive in clockevents_program_event().
545 */
546 if (expires_next < 0)
547 expires_next = 0;
548 return expires_next;
549}
550
551/*
552 * Recomputes cpu_base::*next_timer and returns the earliest expires_next
553 * but does not set cpu_base::*expires_next, that is done by
554 * hrtimer[_force]_reprogram and hrtimer_interrupt only. When updating
555 * cpu_base::*expires_next right away, reprogramming logic would no longer
556 * work.
557 *
558 * When a softirq is pending, we can ignore the HRTIMER_ACTIVE_SOFT bases,
559 * those timers will get run whenever the softirq gets handled, at the end of
560 * hrtimer_run_softirq(), hrtimer_update_softirq_timer() will re-add these bases.
561 *
562 * Therefore softirq values are those from the HRTIMER_ACTIVE_SOFT clock bases.
563 * The !softirq values are the minima across HRTIMER_ACTIVE_ALL, unless an actual
564 * softirq is pending, in which case they're the minima of HRTIMER_ACTIVE_HARD.
565 *
566 * @active_mask must be one of:
567 * - HRTIMER_ACTIVE_ALL,
568 * - HRTIMER_ACTIVE_SOFT, or
569 * - HRTIMER_ACTIVE_HARD.
570 */
571static ktime_t
572__hrtimer_get_next_event(struct hrtimer_cpu_base *cpu_base, unsigned int active_mask)
573{
574 unsigned int active;
575 struct hrtimer *next_timer = NULL;
576 ktime_t expires_next = KTIME_MAX;
577
578 if (!cpu_base->softirq_activated && (active_mask & HRTIMER_ACTIVE_SOFT)) {
579 active = cpu_base->active_bases & HRTIMER_ACTIVE_SOFT;
580 cpu_base->softirq_next_timer = NULL;
581 expires_next = __hrtimer_next_event_base(cpu_base, NULL,
582 active, KTIME_MAX);
583
584 next_timer = cpu_base->softirq_next_timer;
585 }
586
587 if (active_mask & HRTIMER_ACTIVE_HARD) {
588 active = cpu_base->active_bases & HRTIMER_ACTIVE_HARD;
589 cpu_base->next_timer = next_timer;
590 expires_next = __hrtimer_next_event_base(cpu_base, NULL, active,
591 expires_next);
592 }
593
594 return expires_next;
595}
596
597static ktime_t hrtimer_update_next_event(struct hrtimer_cpu_base *cpu_base)
598{
599 ktime_t expires_next, soft = KTIME_MAX;
600
601 /*
602 * If the soft interrupt has already been activated, ignore the
603 * soft bases. They will be handled in the already raised soft
604 * interrupt.
605 */
606 if (!cpu_base->softirq_activated) {
607 soft = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_SOFT);
608 /*
609 * Update the soft expiry time. clock_settime() might have
610 * affected it.
611 */
612 cpu_base->softirq_expires_next = soft;
613 }
614
615 expires_next = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_HARD);
616 /*
617 * If a softirq timer is expiring first, update cpu_base->next_timer
618 * and program the hardware with the soft expiry time.
619 */
620 if (expires_next > soft) {
621 cpu_base->next_timer = cpu_base->softirq_next_timer;
622 expires_next = soft;
623 }
624
625 return expires_next;
626}
627
628static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base)
629{
630 ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset;
631 ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset;
632 ktime_t *offs_tai = &base->clock_base[HRTIMER_BASE_TAI].offset;
633
634 ktime_t now = ktime_get_update_offsets_now(&base->clock_was_set_seq,
635 offs_real, offs_boot, offs_tai);
636
637 base->clock_base[HRTIMER_BASE_REALTIME_SOFT].offset = *offs_real;
638 base->clock_base[HRTIMER_BASE_BOOTTIME_SOFT].offset = *offs_boot;
639 base->clock_base[HRTIMER_BASE_TAI_SOFT].offset = *offs_tai;
640
641 return now;
642}
643
644/*
645 * Is the high resolution mode active ?
646 */
647static inline int hrtimer_hres_active(struct hrtimer_cpu_base *cpu_base)
648{
649 return IS_ENABLED(CONFIG_HIGH_RES_TIMERS) ?
650 cpu_base->hres_active : 0;
651}
652
653static void __hrtimer_reprogram(struct hrtimer_cpu_base *cpu_base,
654 struct hrtimer *next_timer,
655 ktime_t expires_next)
656{
657 cpu_base->expires_next = expires_next;
658
659 /*
660 * If hres is not active, hardware does not have to be
661 * reprogrammed yet.
662 *
663 * If a hang was detected in the last timer interrupt then we
664 * leave the hang delay active in the hardware. We want the
665 * system to make progress. That also prevents the following
666 * scenario:
667 * T1 expires 50ms from now
668 * T2 expires 5s from now
669 *
670 * T1 is removed, so this code is called and would reprogram
671 * the hardware to 5s from now. Any hrtimer_start after that
672 * will not reprogram the hardware due to hang_detected being
673 * set. So we'd effectively block all timers until the T2 event
674 * fires.
675 */
676 if (!hrtimer_hres_active(cpu_base) || cpu_base->hang_detected)
677 return;
678
679 tick_program_event(expires_next, 1);
680}
681
682/*
683 * Reprogram the event source with checking both queues for the
684 * next event
685 * Called with interrupts disabled and base->lock held
686 */
687static void
688hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
689{
690 ktime_t expires_next;
691
692 expires_next = hrtimer_update_next_event(cpu_base);
693
694 if (skip_equal && expires_next == cpu_base->expires_next)
695 return;
696
697 __hrtimer_reprogram(cpu_base, cpu_base->next_timer, expires_next);
698}
699
700/* High resolution timer related functions */
701#ifdef CONFIG_HIGH_RES_TIMERS
702
703/*
704 * High resolution timer enabled ?
705 */
706static bool hrtimer_hres_enabled __read_mostly = true;
707unsigned int hrtimer_resolution __read_mostly = LOW_RES_NSEC;
708EXPORT_SYMBOL_GPL(hrtimer_resolution);
709
710/*
711 * Enable / Disable high resolution mode
712 */
713static int __init setup_hrtimer_hres(char *str)
714{
715 return (kstrtobool(str, &hrtimer_hres_enabled) == 0);
716}
717
718__setup("highres=", setup_hrtimer_hres);
719
720/*
721 * hrtimer_high_res_enabled - query, if the highres mode is enabled
722 */
723static inline int hrtimer_is_hres_enabled(void)
724{
725 return hrtimer_hres_enabled;
726}
727
728static void retrigger_next_event(void *arg);
729
730/*
731 * Switch to high resolution mode
732 */
733static void hrtimer_switch_to_hres(void)
734{
735 struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
736
737 if (tick_init_highres()) {
738 pr_warn("Could not switch to high resolution mode on CPU %u\n",
739 base->cpu);
740 return;
741 }
742 base->hres_active = 1;
743 hrtimer_resolution = HIGH_RES_NSEC;
744
745 tick_setup_sched_timer(true);
746 /* "Retrigger" the interrupt to get things going */
747 retrigger_next_event(NULL);
748}
749
750#else
751
752static inline int hrtimer_is_hres_enabled(void) { return 0; }
753static inline void hrtimer_switch_to_hres(void) { }
754
755#endif /* CONFIG_HIGH_RES_TIMERS */
756/*
757 * Retrigger next event is called after clock was set with interrupts
758 * disabled through an SMP function call or directly from low level
759 * resume code.
760 *
761 * This is only invoked when:
762 * - CONFIG_HIGH_RES_TIMERS is enabled.
763 * - CONFIG_NOHZ_COMMON is enabled
764 *
765 * For the other cases this function is empty and because the call sites
766 * are optimized out it vanishes as well, i.e. no need for lots of
767 * #ifdeffery.
768 */
769static void retrigger_next_event(void *arg)
770{
771 struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
772
773 /*
774 * When high resolution mode or nohz is active, then the offsets of
775 * CLOCK_REALTIME/TAI/BOOTTIME have to be updated. Otherwise the
776 * next tick will take care of that.
777 *
778 * If high resolution mode is active then the next expiring timer
779 * must be reevaluated and the clock event device reprogrammed if
780 * necessary.
781 *
782 * In the NOHZ case the update of the offset and the reevaluation
783 * of the next expiring timer is enough. The return from the SMP
784 * function call will take care of the reprogramming in case the
785 * CPU was in a NOHZ idle sleep.
786 */
787 if (!hrtimer_hres_active(base) && !tick_nohz_active)
788 return;
789
790 raw_spin_lock(&base->lock);
791 hrtimer_update_base(base);
792 if (hrtimer_hres_active(base))
793 hrtimer_force_reprogram(base, 0);
794 else
795 hrtimer_update_next_event(base);
796 raw_spin_unlock(&base->lock);
797}
798
799/*
800 * When a timer is enqueued and expires earlier than the already enqueued
801 * timers, we have to check, whether it expires earlier than the timer for
802 * which the clock event device was armed.
803 *
804 * Called with interrupts disabled and base->cpu_base.lock held
805 */
806static void hrtimer_reprogram(struct hrtimer *timer, bool reprogram)
807{
808 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
809 struct hrtimer_clock_base *base = timer->base;
810 ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
811
812 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
813
814 /*
815 * CLOCK_REALTIME timer might be requested with an absolute
816 * expiry time which is less than base->offset. Set it to 0.
817 */
818 if (expires < 0)
819 expires = 0;
820
821 if (timer->is_soft) {
822 /*
823 * soft hrtimer could be started on a remote CPU. In this
824 * case softirq_expires_next needs to be updated on the
825 * remote CPU. The soft hrtimer will not expire before the
826 * first hard hrtimer on the remote CPU -
827 * hrtimer_check_target() prevents this case.
828 */
829 struct hrtimer_cpu_base *timer_cpu_base = base->cpu_base;
830
831 if (timer_cpu_base->softirq_activated)
832 return;
833
834 if (!ktime_before(expires, timer_cpu_base->softirq_expires_next))
835 return;
836
837 timer_cpu_base->softirq_next_timer = timer;
838 timer_cpu_base->softirq_expires_next = expires;
839
840 if (!ktime_before(expires, timer_cpu_base->expires_next) ||
841 !reprogram)
842 return;
843 }
844
845 /*
846 * If the timer is not on the current cpu, we cannot reprogram
847 * the other cpus clock event device.
848 */
849 if (base->cpu_base != cpu_base)
850 return;
851
852 if (expires >= cpu_base->expires_next)
853 return;
854
855 /*
856 * If the hrtimer interrupt is running, then it will reevaluate the
857 * clock bases and reprogram the clock event device.
858 */
859 if (cpu_base->in_hrtirq)
860 return;
861
862 cpu_base->next_timer = timer;
863
864 __hrtimer_reprogram(cpu_base, timer, expires);
865}
866
867static bool update_needs_ipi(struct hrtimer_cpu_base *cpu_base,
868 unsigned int active)
869{
870 struct hrtimer_clock_base *base;
871 unsigned int seq;
872 ktime_t expires;
873
874 /*
875 * Update the base offsets unconditionally so the following
876 * checks whether the SMP function call is required works.
877 *
878 * The update is safe even when the remote CPU is in the hrtimer
879 * interrupt or the hrtimer soft interrupt and expiring affected
880 * bases. Either it will see the update before handling a base or
881 * it will see it when it finishes the processing and reevaluates
882 * the next expiring timer.
883 */
884 seq = cpu_base->clock_was_set_seq;
885 hrtimer_update_base(cpu_base);
886
887 /*
888 * If the sequence did not change over the update then the
889 * remote CPU already handled it.
890 */
891 if (seq == cpu_base->clock_was_set_seq)
892 return false;
893
894 /*
895 * If the remote CPU is currently handling an hrtimer interrupt, it
896 * will reevaluate the first expiring timer of all clock bases
897 * before reprogramming. Nothing to do here.
898 */
899 if (cpu_base->in_hrtirq)
900 return false;
901
902 /*
903 * Walk the affected clock bases and check whether the first expiring
904 * timer in a clock base is moving ahead of the first expiring timer of
905 * @cpu_base. If so, the IPI must be invoked because per CPU clock
906 * event devices cannot be remotely reprogrammed.
907 */
908 active &= cpu_base->active_bases;
909
910 for_each_active_base(base, cpu_base, active) {
911 struct timerqueue_node *next;
912
913 next = timerqueue_getnext(&base->active);
914 expires = ktime_sub(next->expires, base->offset);
915 if (expires < cpu_base->expires_next)
916 return true;
917
918 /* Extra check for softirq clock bases */
919 if (base->clockid < HRTIMER_BASE_MONOTONIC_SOFT)
920 continue;
921 if (cpu_base->softirq_activated)
922 continue;
923 if (expires < cpu_base->softirq_expires_next)
924 return true;
925 }
926 return false;
927}
928
929/*
930 * Clock was set. This might affect CLOCK_REALTIME, CLOCK_TAI and
931 * CLOCK_BOOTTIME (for late sleep time injection).
932 *
933 * This requires to update the offsets for these clocks
934 * vs. CLOCK_MONOTONIC. When high resolution timers are enabled, then this
935 * also requires to eventually reprogram the per CPU clock event devices
936 * when the change moves an affected timer ahead of the first expiring
937 * timer on that CPU. Obviously remote per CPU clock event devices cannot
938 * be reprogrammed. The other reason why an IPI has to be sent is when the
939 * system is in !HIGH_RES and NOHZ mode. The NOHZ mode updates the offsets
940 * in the tick, which obviously might be stopped, so this has to bring out
941 * the remote CPU which might sleep in idle to get this sorted.
942 */
943void clock_was_set(unsigned int bases)
944{
945 struct hrtimer_cpu_base *cpu_base = raw_cpu_ptr(&hrtimer_bases);
946 cpumask_var_t mask;
947 int cpu;
948
949 if (!hrtimer_hres_active(cpu_base) && !tick_nohz_active)
950 goto out_timerfd;
951
952 if (!zalloc_cpumask_var(&mask, GFP_KERNEL)) {
953 on_each_cpu(retrigger_next_event, NULL, 1);
954 goto out_timerfd;
955 }
956
957 /* Avoid interrupting CPUs if possible */
958 cpus_read_lock();
959 for_each_online_cpu(cpu) {
960 unsigned long flags;
961
962 cpu_base = &per_cpu(hrtimer_bases, cpu);
963 raw_spin_lock_irqsave(&cpu_base->lock, flags);
964
965 if (update_needs_ipi(cpu_base, bases))
966 cpumask_set_cpu(cpu, mask);
967
968 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
969 }
970
971 preempt_disable();
972 smp_call_function_many(mask, retrigger_next_event, NULL, 1);
973 preempt_enable();
974 cpus_read_unlock();
975 free_cpumask_var(mask);
976
977out_timerfd:
978 timerfd_clock_was_set();
979}
980
981static void clock_was_set_work(struct work_struct *work)
982{
983 clock_was_set(CLOCK_SET_WALL);
984}
985
986static DECLARE_WORK(hrtimer_work, clock_was_set_work);
987
988/*
989 * Called from timekeeping code to reprogram the hrtimer interrupt device
990 * on all cpus and to notify timerfd.
991 */
992void clock_was_set_delayed(void)
993{
994 schedule_work(&hrtimer_work);
995}
996
997/*
998 * Called during resume either directly from via timekeeping_resume()
999 * or in the case of s2idle from tick_unfreeze() to ensure that the
1000 * hrtimers are up to date.
1001 */
1002void hrtimers_resume_local(void)
1003{
1004 lockdep_assert_irqs_disabled();
1005 /* Retrigger on the local CPU */
1006 retrigger_next_event(NULL);
1007}
1008
1009/*
1010 * Counterpart to lock_hrtimer_base above:
1011 */
1012static inline
1013void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
1014 __releases(&timer->base->cpu_base->lock)
1015{
1016 raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
1017}
1018
1019/**
1020 * hrtimer_forward() - forward the timer expiry
1021 * @timer: hrtimer to forward
1022 * @now: forward past this time
1023 * @interval: the interval to forward
1024 *
1025 * Forward the timer expiry so it will expire in the future.
1026 *
1027 * .. note::
1028 * This only updates the timer expiry value and does not requeue the timer.
1029 *
1030 * There is also a variant of the function hrtimer_forward_now().
1031 *
1032 * Context: Can be safely called from the callback function of @timer. If called
1033 * from other contexts @timer must neither be enqueued nor running the
1034 * callback and the caller needs to take care of serialization.
1035 *
1036 * Return: The number of overruns are returned.
1037 */
1038u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
1039{
1040 u64 orun = 1;
1041 ktime_t delta;
1042
1043 delta = ktime_sub(now, hrtimer_get_expires(timer));
1044
1045 if (delta < 0)
1046 return 0;
1047
1048 if (WARN_ON(timer->state & HRTIMER_STATE_ENQUEUED))
1049 return 0;
1050
1051 if (interval < hrtimer_resolution)
1052 interval = hrtimer_resolution;
1053
1054 if (unlikely(delta >= interval)) {
1055 s64 incr = ktime_to_ns(interval);
1056
1057 orun = ktime_divns(delta, incr);
1058 hrtimer_add_expires_ns(timer, incr * orun);
1059 if (hrtimer_get_expires_tv64(timer) > now)
1060 return orun;
1061 /*
1062 * This (and the ktime_add() below) is the
1063 * correction for exact:
1064 */
1065 orun++;
1066 }
1067 hrtimer_add_expires(timer, interval);
1068
1069 return orun;
1070}
1071EXPORT_SYMBOL_GPL(hrtimer_forward);
1072
1073/*
1074 * enqueue_hrtimer - internal function to (re)start a timer
1075 *
1076 * The timer is inserted in expiry order. Insertion into the
1077 * red black tree is O(log(n)). Must hold the base lock.
1078 *
1079 * Returns 1 when the new timer is the leftmost timer in the tree.
1080 */
1081static int enqueue_hrtimer(struct hrtimer *timer,
1082 struct hrtimer_clock_base *base,
1083 enum hrtimer_mode mode)
1084{
1085 debug_activate(timer, mode);
1086 WARN_ON_ONCE(!base->cpu_base->online);
1087
1088 base->cpu_base->active_bases |= 1 << base->index;
1089
1090 /* Pairs with the lockless read in hrtimer_is_queued() */
1091 WRITE_ONCE(timer->state, HRTIMER_STATE_ENQUEUED);
1092
1093 return timerqueue_add(&base->active, &timer->node);
1094}
1095
1096/*
1097 * __remove_hrtimer - internal function to remove a timer
1098 *
1099 * Caller must hold the base lock.
1100 *
1101 * High resolution timer mode reprograms the clock event device when the
1102 * timer is the one which expires next. The caller can disable this by setting
1103 * reprogram to zero. This is useful, when the context does a reprogramming
1104 * anyway (e.g. timer interrupt)
1105 */
1106static void __remove_hrtimer(struct hrtimer *timer,
1107 struct hrtimer_clock_base *base,
1108 u8 newstate, int reprogram)
1109{
1110 struct hrtimer_cpu_base *cpu_base = base->cpu_base;
1111 u8 state = timer->state;
1112
1113 /* Pairs with the lockless read in hrtimer_is_queued() */
1114 WRITE_ONCE(timer->state, newstate);
1115 if (!(state & HRTIMER_STATE_ENQUEUED))
1116 return;
1117
1118 if (!timerqueue_del(&base->active, &timer->node))
1119 cpu_base->active_bases &= ~(1 << base->index);
1120
1121 /*
1122 * Note: If reprogram is false we do not update
1123 * cpu_base->next_timer. This happens when we remove the first
1124 * timer on a remote cpu. No harm as we never dereference
1125 * cpu_base->next_timer. So the worst thing what can happen is
1126 * an superfluous call to hrtimer_force_reprogram() on the
1127 * remote cpu later on if the same timer gets enqueued again.
1128 */
1129 if (reprogram && timer == cpu_base->next_timer)
1130 hrtimer_force_reprogram(cpu_base, 1);
1131}
1132
1133/*
1134 * remove hrtimer, called with base lock held
1135 */
1136static inline int
1137remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base,
1138 bool restart, bool keep_local)
1139{
1140 u8 state = timer->state;
1141
1142 if (state & HRTIMER_STATE_ENQUEUED) {
1143 bool reprogram;
1144
1145 /*
1146 * Remove the timer and force reprogramming when high
1147 * resolution mode is active and the timer is on the current
1148 * CPU. If we remove a timer on another CPU, reprogramming is
1149 * skipped. The interrupt event on this CPU is fired and
1150 * reprogramming happens in the interrupt handler. This is a
1151 * rare case and less expensive than a smp call.
1152 */
1153 debug_deactivate(timer);
1154 reprogram = base->cpu_base == this_cpu_ptr(&hrtimer_bases);
1155
1156 /*
1157 * If the timer is not restarted then reprogramming is
1158 * required if the timer is local. If it is local and about
1159 * to be restarted, avoid programming it twice (on removal
1160 * and a moment later when it's requeued).
1161 */
1162 if (!restart)
1163 state = HRTIMER_STATE_INACTIVE;
1164 else
1165 reprogram &= !keep_local;
1166
1167 __remove_hrtimer(timer, base, state, reprogram);
1168 return 1;
1169 }
1170 return 0;
1171}
1172
1173static inline ktime_t hrtimer_update_lowres(struct hrtimer *timer, ktime_t tim,
1174 const enum hrtimer_mode mode)
1175{
1176#ifdef CONFIG_TIME_LOW_RES
1177 /*
1178 * CONFIG_TIME_LOW_RES indicates that the system has no way to return
1179 * granular time values. For relative timers we add hrtimer_resolution
1180 * (i.e. one jiffie) to prevent short timeouts.
1181 */
1182 timer->is_rel = mode & HRTIMER_MODE_REL;
1183 if (timer->is_rel)
1184 tim = ktime_add_safe(tim, hrtimer_resolution);
1185#endif
1186 return tim;
1187}
1188
1189static void
1190hrtimer_update_softirq_timer(struct hrtimer_cpu_base *cpu_base, bool reprogram)
1191{
1192 ktime_t expires;
1193
1194 /*
1195 * Find the next SOFT expiration.
1196 */
1197 expires = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_SOFT);
1198
1199 /*
1200 * reprogramming needs to be triggered, even if the next soft
1201 * hrtimer expires at the same time than the next hard
1202 * hrtimer. cpu_base->softirq_expires_next needs to be updated!
1203 */
1204 if (expires == KTIME_MAX)
1205 return;
1206
1207 /*
1208 * cpu_base->*next_timer is recomputed by __hrtimer_get_next_event()
1209 * cpu_base->*expires_next is only set by hrtimer_reprogram()
1210 */
1211 hrtimer_reprogram(cpu_base->softirq_next_timer, reprogram);
1212}
1213
1214static int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1215 u64 delta_ns, const enum hrtimer_mode mode,
1216 struct hrtimer_clock_base *base)
1217{
1218 struct hrtimer_clock_base *new_base;
1219 bool force_local, first;
1220
1221 /*
1222 * If the timer is on the local cpu base and is the first expiring
1223 * timer then this might end up reprogramming the hardware twice
1224 * (on removal and on enqueue). To avoid that by prevent the
1225 * reprogram on removal, keep the timer local to the current CPU
1226 * and enforce reprogramming after it is queued no matter whether
1227 * it is the new first expiring timer again or not.
1228 */
1229 force_local = base->cpu_base == this_cpu_ptr(&hrtimer_bases);
1230 force_local &= base->cpu_base->next_timer == timer;
1231
1232 /*
1233 * Remove an active timer from the queue. In case it is not queued
1234 * on the current CPU, make sure that remove_hrtimer() updates the
1235 * remote data correctly.
1236 *
1237 * If it's on the current CPU and the first expiring timer, then
1238 * skip reprogramming, keep the timer local and enforce
1239 * reprogramming later if it was the first expiring timer. This
1240 * avoids programming the underlying clock event twice (once at
1241 * removal and once after enqueue).
1242 */
1243 remove_hrtimer(timer, base, true, force_local);
1244
1245 if (mode & HRTIMER_MODE_REL)
1246 tim = ktime_add_safe(tim, base->get_time());
1247
1248 tim = hrtimer_update_lowres(timer, tim, mode);
1249
1250 hrtimer_set_expires_range_ns(timer, tim, delta_ns);
1251
1252 /* Switch the timer base, if necessary: */
1253 if (!force_local) {
1254 new_base = switch_hrtimer_base(timer, base,
1255 mode & HRTIMER_MODE_PINNED);
1256 } else {
1257 new_base = base;
1258 }
1259
1260 first = enqueue_hrtimer(timer, new_base, mode);
1261 if (!force_local)
1262 return first;
1263
1264 /*
1265 * Timer was forced to stay on the current CPU to avoid
1266 * reprogramming on removal and enqueue. Force reprogram the
1267 * hardware by evaluating the new first expiring timer.
1268 */
1269 hrtimer_force_reprogram(new_base->cpu_base, 1);
1270 return 0;
1271}
1272
1273/**
1274 * hrtimer_start_range_ns - (re)start an hrtimer
1275 * @timer: the timer to be added
1276 * @tim: expiry time
1277 * @delta_ns: "slack" range for the timer
1278 * @mode: timer mode: absolute (HRTIMER_MODE_ABS) or
1279 * relative (HRTIMER_MODE_REL), and pinned (HRTIMER_MODE_PINNED);
1280 * softirq based mode is considered for debug purpose only!
1281 */
1282void hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1283 u64 delta_ns, const enum hrtimer_mode mode)
1284{
1285 struct hrtimer_clock_base *base;
1286 unsigned long flags;
1287
1288 if (WARN_ON_ONCE(!timer->function))
1289 return;
1290 /*
1291 * Check whether the HRTIMER_MODE_SOFT bit and hrtimer.is_soft
1292 * match on CONFIG_PREEMPT_RT = n. With PREEMPT_RT check the hard
1293 * expiry mode because unmarked timers are moved to softirq expiry.
1294 */
1295 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
1296 WARN_ON_ONCE(!(mode & HRTIMER_MODE_SOFT) ^ !timer->is_soft);
1297 else
1298 WARN_ON_ONCE(!(mode & HRTIMER_MODE_HARD) ^ !timer->is_hard);
1299
1300 base = lock_hrtimer_base(timer, &flags);
1301
1302 if (__hrtimer_start_range_ns(timer, tim, delta_ns, mode, base))
1303 hrtimer_reprogram(timer, true);
1304
1305 unlock_hrtimer_base(timer, &flags);
1306}
1307EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
1308
1309/**
1310 * hrtimer_try_to_cancel - try to deactivate a timer
1311 * @timer: hrtimer to stop
1312 *
1313 * Returns:
1314 *
1315 * * 0 when the timer was not active
1316 * * 1 when the timer was active
1317 * * -1 when the timer is currently executing the callback function and
1318 * cannot be stopped
1319 */
1320int hrtimer_try_to_cancel(struct hrtimer *timer)
1321{
1322 struct hrtimer_clock_base *base;
1323 unsigned long flags;
1324 int ret = -1;
1325
1326 /*
1327 * Check lockless first. If the timer is not active (neither
1328 * enqueued nor running the callback, nothing to do here. The
1329 * base lock does not serialize against a concurrent enqueue,
1330 * so we can avoid taking it.
1331 */
1332 if (!hrtimer_active(timer))
1333 return 0;
1334
1335 base = lock_hrtimer_base(timer, &flags);
1336
1337 if (!hrtimer_callback_running(timer))
1338 ret = remove_hrtimer(timer, base, false, false);
1339
1340 unlock_hrtimer_base(timer, &flags);
1341
1342 return ret;
1343
1344}
1345EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1346
1347#ifdef CONFIG_PREEMPT_RT
1348static void hrtimer_cpu_base_init_expiry_lock(struct hrtimer_cpu_base *base)
1349{
1350 spin_lock_init(&base->softirq_expiry_lock);
1351}
1352
1353static void hrtimer_cpu_base_lock_expiry(struct hrtimer_cpu_base *base)
1354{
1355 spin_lock(&base->softirq_expiry_lock);
1356}
1357
1358static void hrtimer_cpu_base_unlock_expiry(struct hrtimer_cpu_base *base)
1359{
1360 spin_unlock(&base->softirq_expiry_lock);
1361}
1362
1363/*
1364 * The counterpart to hrtimer_cancel_wait_running().
1365 *
1366 * If there is a waiter for cpu_base->expiry_lock, then it was waiting for
1367 * the timer callback to finish. Drop expiry_lock and reacquire it. That
1368 * allows the waiter to acquire the lock and make progress.
1369 */
1370static void hrtimer_sync_wait_running(struct hrtimer_cpu_base *cpu_base,
1371 unsigned long flags)
1372{
1373 if (atomic_read(&cpu_base->timer_waiters)) {
1374 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1375 spin_unlock(&cpu_base->softirq_expiry_lock);
1376 spin_lock(&cpu_base->softirq_expiry_lock);
1377 raw_spin_lock_irq(&cpu_base->lock);
1378 }
1379}
1380
1381/*
1382 * This function is called on PREEMPT_RT kernels when the fast path
1383 * deletion of a timer failed because the timer callback function was
1384 * running.
1385 *
1386 * This prevents priority inversion: if the soft irq thread is preempted
1387 * in the middle of a timer callback, then calling del_timer_sync() can
1388 * lead to two issues:
1389 *
1390 * - If the caller is on a remote CPU then it has to spin wait for the timer
1391 * handler to complete. This can result in unbound priority inversion.
1392 *
1393 * - If the caller originates from the task which preempted the timer
1394 * handler on the same CPU, then spin waiting for the timer handler to
1395 * complete is never going to end.
1396 */
1397void hrtimer_cancel_wait_running(const struct hrtimer *timer)
1398{
1399 /* Lockless read. Prevent the compiler from reloading it below */
1400 struct hrtimer_clock_base *base = READ_ONCE(timer->base);
1401
1402 /*
1403 * Just relax if the timer expires in hard interrupt context or if
1404 * it is currently on the migration base.
1405 */
1406 if (!timer->is_soft || is_migration_base(base)) {
1407 cpu_relax();
1408 return;
1409 }
1410
1411 /*
1412 * Mark the base as contended and grab the expiry lock, which is
1413 * held by the softirq across the timer callback. Drop the lock
1414 * immediately so the softirq can expire the next timer. In theory
1415 * the timer could already be running again, but that's more than
1416 * unlikely and just causes another wait loop.
1417 */
1418 atomic_inc(&base->cpu_base->timer_waiters);
1419 spin_lock_bh(&base->cpu_base->softirq_expiry_lock);
1420 atomic_dec(&base->cpu_base->timer_waiters);
1421 spin_unlock_bh(&base->cpu_base->softirq_expiry_lock);
1422}
1423#else
1424static inline void
1425hrtimer_cpu_base_init_expiry_lock(struct hrtimer_cpu_base *base) { }
1426static inline void
1427hrtimer_cpu_base_lock_expiry(struct hrtimer_cpu_base *base) { }
1428static inline void
1429hrtimer_cpu_base_unlock_expiry(struct hrtimer_cpu_base *base) { }
1430static inline void hrtimer_sync_wait_running(struct hrtimer_cpu_base *base,
1431 unsigned long flags) { }
1432#endif
1433
1434/**
1435 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1436 * @timer: the timer to be cancelled
1437 *
1438 * Returns:
1439 * 0 when the timer was not active
1440 * 1 when the timer was active
1441 */
1442int hrtimer_cancel(struct hrtimer *timer)
1443{
1444 int ret;
1445
1446 do {
1447 ret = hrtimer_try_to_cancel(timer);
1448
1449 if (ret < 0)
1450 hrtimer_cancel_wait_running(timer);
1451 } while (ret < 0);
1452 return ret;
1453}
1454EXPORT_SYMBOL_GPL(hrtimer_cancel);
1455
1456/**
1457 * __hrtimer_get_remaining - get remaining time for the timer
1458 * @timer: the timer to read
1459 * @adjust: adjust relative timers when CONFIG_TIME_LOW_RES=y
1460 */
1461ktime_t __hrtimer_get_remaining(const struct hrtimer *timer, bool adjust)
1462{
1463 unsigned long flags;
1464 ktime_t rem;
1465
1466 lock_hrtimer_base(timer, &flags);
1467 if (IS_ENABLED(CONFIG_TIME_LOW_RES) && adjust)
1468 rem = hrtimer_expires_remaining_adjusted(timer);
1469 else
1470 rem = hrtimer_expires_remaining(timer);
1471 unlock_hrtimer_base(timer, &flags);
1472
1473 return rem;
1474}
1475EXPORT_SYMBOL_GPL(__hrtimer_get_remaining);
1476
1477#ifdef CONFIG_NO_HZ_COMMON
1478/**
1479 * hrtimer_get_next_event - get the time until next expiry event
1480 *
1481 * Returns the next expiry time or KTIME_MAX if no timer is pending.
1482 */
1483u64 hrtimer_get_next_event(void)
1484{
1485 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1486 u64 expires = KTIME_MAX;
1487 unsigned long flags;
1488
1489 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1490
1491 if (!hrtimer_hres_active(cpu_base))
1492 expires = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_ALL);
1493
1494 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1495
1496 return expires;
1497}
1498
1499/**
1500 * hrtimer_next_event_without - time until next expiry event w/o one timer
1501 * @exclude: timer to exclude
1502 *
1503 * Returns the next expiry time over all timers except for the @exclude one or
1504 * KTIME_MAX if none of them is pending.
1505 */
1506u64 hrtimer_next_event_without(const struct hrtimer *exclude)
1507{
1508 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1509 u64 expires = KTIME_MAX;
1510 unsigned long flags;
1511
1512 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1513
1514 if (hrtimer_hres_active(cpu_base)) {
1515 unsigned int active;
1516
1517 if (!cpu_base->softirq_activated) {
1518 active = cpu_base->active_bases & HRTIMER_ACTIVE_SOFT;
1519 expires = __hrtimer_next_event_base(cpu_base, exclude,
1520 active, KTIME_MAX);
1521 }
1522 active = cpu_base->active_bases & HRTIMER_ACTIVE_HARD;
1523 expires = __hrtimer_next_event_base(cpu_base, exclude, active,
1524 expires);
1525 }
1526
1527 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1528
1529 return expires;
1530}
1531#endif
1532
1533static inline int hrtimer_clockid_to_base(clockid_t clock_id)
1534{
1535 if (likely(clock_id < MAX_CLOCKS)) {
1536 int base = hrtimer_clock_to_base_table[clock_id];
1537
1538 if (likely(base != HRTIMER_MAX_CLOCK_BASES))
1539 return base;
1540 }
1541 WARN(1, "Invalid clockid %d. Using MONOTONIC\n", clock_id);
1542 return HRTIMER_BASE_MONOTONIC;
1543}
1544
1545static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1546 enum hrtimer_mode mode)
1547{
1548 bool softtimer = !!(mode & HRTIMER_MODE_SOFT);
1549 struct hrtimer_cpu_base *cpu_base;
1550 int base;
1551
1552 /*
1553 * On PREEMPT_RT enabled kernels hrtimers which are not explicitly
1554 * marked for hard interrupt expiry mode are moved into soft
1555 * interrupt context for latency reasons and because the callbacks
1556 * can invoke functions which might sleep on RT, e.g. spin_lock().
1557 */
1558 if (IS_ENABLED(CONFIG_PREEMPT_RT) && !(mode & HRTIMER_MODE_HARD))
1559 softtimer = true;
1560
1561 memset(timer, 0, sizeof(struct hrtimer));
1562
1563 cpu_base = raw_cpu_ptr(&hrtimer_bases);
1564
1565 /*
1566 * POSIX magic: Relative CLOCK_REALTIME timers are not affected by
1567 * clock modifications, so they needs to become CLOCK_MONOTONIC to
1568 * ensure POSIX compliance.
1569 */
1570 if (clock_id == CLOCK_REALTIME && mode & HRTIMER_MODE_REL)
1571 clock_id = CLOCK_MONOTONIC;
1572
1573 base = softtimer ? HRTIMER_MAX_CLOCK_BASES / 2 : 0;
1574 base += hrtimer_clockid_to_base(clock_id);
1575 timer->is_soft = softtimer;
1576 timer->is_hard = !!(mode & HRTIMER_MODE_HARD);
1577 timer->base = &cpu_base->clock_base[base];
1578 timerqueue_init(&timer->node);
1579}
1580
1581/**
1582 * hrtimer_init - initialize a timer to the given clock
1583 * @timer: the timer to be initialized
1584 * @clock_id: the clock to be used
1585 * @mode: The modes which are relevant for initialization:
1586 * HRTIMER_MODE_ABS, HRTIMER_MODE_REL, HRTIMER_MODE_ABS_SOFT,
1587 * HRTIMER_MODE_REL_SOFT
1588 *
1589 * The PINNED variants of the above can be handed in,
1590 * but the PINNED bit is ignored as pinning happens
1591 * when the hrtimer is started
1592 */
1593void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1594 enum hrtimer_mode mode)
1595{
1596 debug_init(timer, clock_id, mode);
1597 __hrtimer_init(timer, clock_id, mode);
1598}
1599EXPORT_SYMBOL_GPL(hrtimer_init);
1600
1601/*
1602 * A timer is active, when it is enqueued into the rbtree or the
1603 * callback function is running or it's in the state of being migrated
1604 * to another cpu.
1605 *
1606 * It is important for this function to not return a false negative.
1607 */
1608bool hrtimer_active(const struct hrtimer *timer)
1609{
1610 struct hrtimer_clock_base *base;
1611 unsigned int seq;
1612
1613 do {
1614 base = READ_ONCE(timer->base);
1615 seq = raw_read_seqcount_begin(&base->seq);
1616
1617 if (timer->state != HRTIMER_STATE_INACTIVE ||
1618 base->running == timer)
1619 return true;
1620
1621 } while (read_seqcount_retry(&base->seq, seq) ||
1622 base != READ_ONCE(timer->base));
1623
1624 return false;
1625}
1626EXPORT_SYMBOL_GPL(hrtimer_active);
1627
1628/*
1629 * The write_seqcount_barrier()s in __run_hrtimer() split the thing into 3
1630 * distinct sections:
1631 *
1632 * - queued: the timer is queued
1633 * - callback: the timer is being ran
1634 * - post: the timer is inactive or (re)queued
1635 *
1636 * On the read side we ensure we observe timer->state and cpu_base->running
1637 * from the same section, if anything changed while we looked at it, we retry.
1638 * This includes timer->base changing because sequence numbers alone are
1639 * insufficient for that.
1640 *
1641 * The sequence numbers are required because otherwise we could still observe
1642 * a false negative if the read side got smeared over multiple consecutive
1643 * __run_hrtimer() invocations.
1644 */
1645
1646static void __run_hrtimer(struct hrtimer_cpu_base *cpu_base,
1647 struct hrtimer_clock_base *base,
1648 struct hrtimer *timer, ktime_t *now,
1649 unsigned long flags) __must_hold(&cpu_base->lock)
1650{
1651 enum hrtimer_restart (*fn)(struct hrtimer *);
1652 bool expires_in_hardirq;
1653 int restart;
1654
1655 lockdep_assert_held(&cpu_base->lock);
1656
1657 debug_deactivate(timer);
1658 base->running = timer;
1659
1660 /*
1661 * Separate the ->running assignment from the ->state assignment.
1662 *
1663 * As with a regular write barrier, this ensures the read side in
1664 * hrtimer_active() cannot observe base->running == NULL &&
1665 * timer->state == INACTIVE.
1666 */
1667 raw_write_seqcount_barrier(&base->seq);
1668
1669 __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE, 0);
1670 fn = timer->function;
1671
1672 /*
1673 * Clear the 'is relative' flag for the TIME_LOW_RES case. If the
1674 * timer is restarted with a period then it becomes an absolute
1675 * timer. If its not restarted it does not matter.
1676 */
1677 if (IS_ENABLED(CONFIG_TIME_LOW_RES))
1678 timer->is_rel = false;
1679
1680 /*
1681 * The timer is marked as running in the CPU base, so it is
1682 * protected against migration to a different CPU even if the lock
1683 * is dropped.
1684 */
1685 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1686 trace_hrtimer_expire_entry(timer, now);
1687 expires_in_hardirq = lockdep_hrtimer_enter(timer);
1688
1689 restart = fn(timer);
1690
1691 lockdep_hrtimer_exit(expires_in_hardirq);
1692 trace_hrtimer_expire_exit(timer);
1693 raw_spin_lock_irq(&cpu_base->lock);
1694
1695 /*
1696 * Note: We clear the running state after enqueue_hrtimer and
1697 * we do not reprogram the event hardware. Happens either in
1698 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1699 *
1700 * Note: Because we dropped the cpu_base->lock above,
1701 * hrtimer_start_range_ns() can have popped in and enqueued the timer
1702 * for us already.
1703 */
1704 if (restart != HRTIMER_NORESTART &&
1705 !(timer->state & HRTIMER_STATE_ENQUEUED))
1706 enqueue_hrtimer(timer, base, HRTIMER_MODE_ABS);
1707
1708 /*
1709 * Separate the ->running assignment from the ->state assignment.
1710 *
1711 * As with a regular write barrier, this ensures the read side in
1712 * hrtimer_active() cannot observe base->running.timer == NULL &&
1713 * timer->state == INACTIVE.
1714 */
1715 raw_write_seqcount_barrier(&base->seq);
1716
1717 WARN_ON_ONCE(base->running != timer);
1718 base->running = NULL;
1719}
1720
1721static void __hrtimer_run_queues(struct hrtimer_cpu_base *cpu_base, ktime_t now,
1722 unsigned long flags, unsigned int active_mask)
1723{
1724 struct hrtimer_clock_base *base;
1725 unsigned int active = cpu_base->active_bases & active_mask;
1726
1727 for_each_active_base(base, cpu_base, active) {
1728 struct timerqueue_node *node;
1729 ktime_t basenow;
1730
1731 basenow = ktime_add(now, base->offset);
1732
1733 while ((node = timerqueue_getnext(&base->active))) {
1734 struct hrtimer *timer;
1735
1736 timer = container_of(node, struct hrtimer, node);
1737
1738 /*
1739 * The immediate goal for using the softexpires is
1740 * minimizing wakeups, not running timers at the
1741 * earliest interrupt after their soft expiration.
1742 * This allows us to avoid using a Priority Search
1743 * Tree, which can answer a stabbing query for
1744 * overlapping intervals and instead use the simple
1745 * BST we already have.
1746 * We don't add extra wakeups by delaying timers that
1747 * are right-of a not yet expired timer, because that
1748 * timer will have to trigger a wakeup anyway.
1749 */
1750 if (basenow < hrtimer_get_softexpires_tv64(timer))
1751 break;
1752
1753 __run_hrtimer(cpu_base, base, timer, &basenow, flags);
1754 if (active_mask == HRTIMER_ACTIVE_SOFT)
1755 hrtimer_sync_wait_running(cpu_base, flags);
1756 }
1757 }
1758}
1759
1760static __latent_entropy void hrtimer_run_softirq(struct softirq_action *h)
1761{
1762 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1763 unsigned long flags;
1764 ktime_t now;
1765
1766 hrtimer_cpu_base_lock_expiry(cpu_base);
1767 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1768
1769 now = hrtimer_update_base(cpu_base);
1770 __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_SOFT);
1771
1772 cpu_base->softirq_activated = 0;
1773 hrtimer_update_softirq_timer(cpu_base, true);
1774
1775 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1776 hrtimer_cpu_base_unlock_expiry(cpu_base);
1777}
1778
1779#ifdef CONFIG_HIGH_RES_TIMERS
1780
1781/*
1782 * High resolution timer interrupt
1783 * Called with interrupts disabled
1784 */
1785void hrtimer_interrupt(struct clock_event_device *dev)
1786{
1787 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1788 ktime_t expires_next, now, entry_time, delta;
1789 unsigned long flags;
1790 int retries = 0;
1791
1792 BUG_ON(!cpu_base->hres_active);
1793 cpu_base->nr_events++;
1794 dev->next_event = KTIME_MAX;
1795
1796 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1797 entry_time = now = hrtimer_update_base(cpu_base);
1798retry:
1799 cpu_base->in_hrtirq = 1;
1800 /*
1801 * We set expires_next to KTIME_MAX here with cpu_base->lock
1802 * held to prevent that a timer is enqueued in our queue via
1803 * the migration code. This does not affect enqueueing of
1804 * timers which run their callback and need to be requeued on
1805 * this CPU.
1806 */
1807 cpu_base->expires_next = KTIME_MAX;
1808
1809 if (!ktime_before(now, cpu_base->softirq_expires_next)) {
1810 cpu_base->softirq_expires_next = KTIME_MAX;
1811 cpu_base->softirq_activated = 1;
1812 raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1813 }
1814
1815 __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_HARD);
1816
1817 /* Reevaluate the clock bases for the [soft] next expiry */
1818 expires_next = hrtimer_update_next_event(cpu_base);
1819 /*
1820 * Store the new expiry value so the migration code can verify
1821 * against it.
1822 */
1823 cpu_base->expires_next = expires_next;
1824 cpu_base->in_hrtirq = 0;
1825 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1826
1827 /* Reprogramming necessary ? */
1828 if (!tick_program_event(expires_next, 0)) {
1829 cpu_base->hang_detected = 0;
1830 return;
1831 }
1832
1833 /*
1834 * The next timer was already expired due to:
1835 * - tracing
1836 * - long lasting callbacks
1837 * - being scheduled away when running in a VM
1838 *
1839 * We need to prevent that we loop forever in the hrtimer
1840 * interrupt routine. We give it 3 attempts to avoid
1841 * overreacting on some spurious event.
1842 *
1843 * Acquire base lock for updating the offsets and retrieving
1844 * the current time.
1845 */
1846 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1847 now = hrtimer_update_base(cpu_base);
1848 cpu_base->nr_retries++;
1849 if (++retries < 3)
1850 goto retry;
1851 /*
1852 * Give the system a chance to do something else than looping
1853 * here. We stored the entry time, so we know exactly how long
1854 * we spent here. We schedule the next event this amount of
1855 * time away.
1856 */
1857 cpu_base->nr_hangs++;
1858 cpu_base->hang_detected = 1;
1859 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1860
1861 delta = ktime_sub(now, entry_time);
1862 if ((unsigned int)delta > cpu_base->max_hang_time)
1863 cpu_base->max_hang_time = (unsigned int) delta;
1864 /*
1865 * Limit it to a sensible value as we enforce a longer
1866 * delay. Give the CPU at least 100ms to catch up.
1867 */
1868 if (delta > 100 * NSEC_PER_MSEC)
1869 expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1870 else
1871 expires_next = ktime_add(now, delta);
1872 tick_program_event(expires_next, 1);
1873 pr_warn_once("hrtimer: interrupt took %llu ns\n", ktime_to_ns(delta));
1874}
1875#endif /* !CONFIG_HIGH_RES_TIMERS */
1876
1877/*
1878 * Called from run_local_timers in hardirq context every jiffy
1879 */
1880void hrtimer_run_queues(void)
1881{
1882 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1883 unsigned long flags;
1884 ktime_t now;
1885
1886 if (hrtimer_hres_active(cpu_base))
1887 return;
1888
1889 /*
1890 * This _is_ ugly: We have to check periodically, whether we
1891 * can switch to highres and / or nohz mode. The clocksource
1892 * switch happens with xtime_lock held. Notification from
1893 * there only sets the check bit in the tick_oneshot code,
1894 * otherwise we might deadlock vs. xtime_lock.
1895 */
1896 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) {
1897 hrtimer_switch_to_hres();
1898 return;
1899 }
1900
1901 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1902 now = hrtimer_update_base(cpu_base);
1903
1904 if (!ktime_before(now, cpu_base->softirq_expires_next)) {
1905 cpu_base->softirq_expires_next = KTIME_MAX;
1906 cpu_base->softirq_activated = 1;
1907 raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1908 }
1909
1910 __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_HARD);
1911 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1912}
1913
1914/*
1915 * Sleep related functions:
1916 */
1917static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1918{
1919 struct hrtimer_sleeper *t =
1920 container_of(timer, struct hrtimer_sleeper, timer);
1921 struct task_struct *task = t->task;
1922
1923 t->task = NULL;
1924 if (task)
1925 wake_up_process(task);
1926
1927 return HRTIMER_NORESTART;
1928}
1929
1930/**
1931 * hrtimer_sleeper_start_expires - Start a hrtimer sleeper timer
1932 * @sl: sleeper to be started
1933 * @mode: timer mode abs/rel
1934 *
1935 * Wrapper around hrtimer_start_expires() for hrtimer_sleeper based timers
1936 * to allow PREEMPT_RT to tweak the delivery mode (soft/hardirq context)
1937 */
1938void hrtimer_sleeper_start_expires(struct hrtimer_sleeper *sl,
1939 enum hrtimer_mode mode)
1940{
1941 /*
1942 * Make the enqueue delivery mode check work on RT. If the sleeper
1943 * was initialized for hard interrupt delivery, force the mode bit.
1944 * This is a special case for hrtimer_sleepers because
1945 * hrtimer_init_sleeper() determines the delivery mode on RT so the
1946 * fiddling with this decision is avoided at the call sites.
1947 */
1948 if (IS_ENABLED(CONFIG_PREEMPT_RT) && sl->timer.is_hard)
1949 mode |= HRTIMER_MODE_HARD;
1950
1951 hrtimer_start_expires(&sl->timer, mode);
1952}
1953EXPORT_SYMBOL_GPL(hrtimer_sleeper_start_expires);
1954
1955static void __hrtimer_init_sleeper(struct hrtimer_sleeper *sl,
1956 clockid_t clock_id, enum hrtimer_mode mode)
1957{
1958 /*
1959 * On PREEMPT_RT enabled kernels hrtimers which are not explicitly
1960 * marked for hard interrupt expiry mode are moved into soft
1961 * interrupt context either for latency reasons or because the
1962 * hrtimer callback takes regular spinlocks or invokes other
1963 * functions which are not suitable for hard interrupt context on
1964 * PREEMPT_RT.
1965 *
1966 * The hrtimer_sleeper callback is RT compatible in hard interrupt
1967 * context, but there is a latency concern: Untrusted userspace can
1968 * spawn many threads which arm timers for the same expiry time on
1969 * the same CPU. That causes a latency spike due to the wakeup of
1970 * a gazillion threads.
1971 *
1972 * OTOH, privileged real-time user space applications rely on the
1973 * low latency of hard interrupt wakeups. If the current task is in
1974 * a real-time scheduling class, mark the mode for hard interrupt
1975 * expiry.
1976 */
1977 if (IS_ENABLED(CONFIG_PREEMPT_RT)) {
1978 if (task_is_realtime(current) && !(mode & HRTIMER_MODE_SOFT))
1979 mode |= HRTIMER_MODE_HARD;
1980 }
1981
1982 __hrtimer_init(&sl->timer, clock_id, mode);
1983 sl->timer.function = hrtimer_wakeup;
1984 sl->task = current;
1985}
1986
1987/**
1988 * hrtimer_init_sleeper - initialize sleeper to the given clock
1989 * @sl: sleeper to be initialized
1990 * @clock_id: the clock to be used
1991 * @mode: timer mode abs/rel
1992 */
1993void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, clockid_t clock_id,
1994 enum hrtimer_mode mode)
1995{
1996 debug_init(&sl->timer, clock_id, mode);
1997 __hrtimer_init_sleeper(sl, clock_id, mode);
1998
1999}
2000EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
2001
2002int nanosleep_copyout(struct restart_block *restart, struct timespec64 *ts)
2003{
2004 switch(restart->nanosleep.type) {
2005#ifdef CONFIG_COMPAT_32BIT_TIME
2006 case TT_COMPAT:
2007 if (put_old_timespec32(ts, restart->nanosleep.compat_rmtp))
2008 return -EFAULT;
2009 break;
2010#endif
2011 case TT_NATIVE:
2012 if (put_timespec64(ts, restart->nanosleep.rmtp))
2013 return -EFAULT;
2014 break;
2015 default:
2016 BUG();
2017 }
2018 return -ERESTART_RESTARTBLOCK;
2019}
2020
2021static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
2022{
2023 struct restart_block *restart;
2024
2025 do {
2026 set_current_state(TASK_INTERRUPTIBLE|TASK_FREEZABLE);
2027 hrtimer_sleeper_start_expires(t, mode);
2028
2029 if (likely(t->task))
2030 schedule();
2031
2032 hrtimer_cancel(&t->timer);
2033 mode = HRTIMER_MODE_ABS;
2034
2035 } while (t->task && !signal_pending(current));
2036
2037 __set_current_state(TASK_RUNNING);
2038
2039 if (!t->task)
2040 return 0;
2041
2042 restart = &current->restart_block;
2043 if (restart->nanosleep.type != TT_NONE) {
2044 ktime_t rem = hrtimer_expires_remaining(&t->timer);
2045 struct timespec64 rmt;
2046
2047 if (rem <= 0)
2048 return 0;
2049 rmt = ktime_to_timespec64(rem);
2050
2051 return nanosleep_copyout(restart, &rmt);
2052 }
2053 return -ERESTART_RESTARTBLOCK;
2054}
2055
2056static long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
2057{
2058 struct hrtimer_sleeper t;
2059 int ret;
2060
2061 hrtimer_init_sleeper_on_stack(&t, restart->nanosleep.clockid,
2062 HRTIMER_MODE_ABS);
2063 hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
2064 ret = do_nanosleep(&t, HRTIMER_MODE_ABS);
2065 destroy_hrtimer_on_stack(&t.timer);
2066 return ret;
2067}
2068
2069long hrtimer_nanosleep(ktime_t rqtp, const enum hrtimer_mode mode,
2070 const clockid_t clockid)
2071{
2072 struct restart_block *restart;
2073 struct hrtimer_sleeper t;
2074 int ret = 0;
2075 u64 slack;
2076
2077 slack = current->timer_slack_ns;
2078 if (rt_task(current))
2079 slack = 0;
2080
2081 hrtimer_init_sleeper_on_stack(&t, clockid, mode);
2082 hrtimer_set_expires_range_ns(&t.timer, rqtp, slack);
2083 ret = do_nanosleep(&t, mode);
2084 if (ret != -ERESTART_RESTARTBLOCK)
2085 goto out;
2086
2087 /* Absolute timers do not update the rmtp value and restart: */
2088 if (mode == HRTIMER_MODE_ABS) {
2089 ret = -ERESTARTNOHAND;
2090 goto out;
2091 }
2092
2093 restart = &current->restart_block;
2094 restart->nanosleep.clockid = t.timer.base->clockid;
2095 restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
2096 set_restart_fn(restart, hrtimer_nanosleep_restart);
2097out:
2098 destroy_hrtimer_on_stack(&t.timer);
2099 return ret;
2100}
2101
2102#ifdef CONFIG_64BIT
2103
2104SYSCALL_DEFINE2(nanosleep, struct __kernel_timespec __user *, rqtp,
2105 struct __kernel_timespec __user *, rmtp)
2106{
2107 struct timespec64 tu;
2108
2109 if (get_timespec64(&tu, rqtp))
2110 return -EFAULT;
2111
2112 if (!timespec64_valid(&tu))
2113 return -EINVAL;
2114
2115 current->restart_block.fn = do_no_restart_syscall;
2116 current->restart_block.nanosleep.type = rmtp ? TT_NATIVE : TT_NONE;
2117 current->restart_block.nanosleep.rmtp = rmtp;
2118 return hrtimer_nanosleep(timespec64_to_ktime(tu), HRTIMER_MODE_REL,
2119 CLOCK_MONOTONIC);
2120}
2121
2122#endif
2123
2124#ifdef CONFIG_COMPAT_32BIT_TIME
2125
2126SYSCALL_DEFINE2(nanosleep_time32, struct old_timespec32 __user *, rqtp,
2127 struct old_timespec32 __user *, rmtp)
2128{
2129 struct timespec64 tu;
2130
2131 if (get_old_timespec32(&tu, rqtp))
2132 return -EFAULT;
2133
2134 if (!timespec64_valid(&tu))
2135 return -EINVAL;
2136
2137 current->restart_block.fn = do_no_restart_syscall;
2138 current->restart_block.nanosleep.type = rmtp ? TT_COMPAT : TT_NONE;
2139 current->restart_block.nanosleep.compat_rmtp = rmtp;
2140 return hrtimer_nanosleep(timespec64_to_ktime(tu), HRTIMER_MODE_REL,
2141 CLOCK_MONOTONIC);
2142}
2143#endif
2144
2145/*
2146 * Functions related to boot-time initialization:
2147 */
2148int hrtimers_prepare_cpu(unsigned int cpu)
2149{
2150 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
2151 int i;
2152
2153 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
2154 struct hrtimer_clock_base *clock_b = &cpu_base->clock_base[i];
2155
2156 clock_b->cpu_base = cpu_base;
2157 seqcount_raw_spinlock_init(&clock_b->seq, &cpu_base->lock);
2158 timerqueue_init_head(&clock_b->active);
2159 }
2160
2161 cpu_base->cpu = cpu;
2162 cpu_base->active_bases = 0;
2163 cpu_base->hres_active = 0;
2164 cpu_base->hang_detected = 0;
2165 cpu_base->next_timer = NULL;
2166 cpu_base->softirq_next_timer = NULL;
2167 cpu_base->expires_next = KTIME_MAX;
2168 cpu_base->softirq_expires_next = KTIME_MAX;
2169 cpu_base->online = 1;
2170 hrtimer_cpu_base_init_expiry_lock(cpu_base);
2171 return 0;
2172}
2173
2174#ifdef CONFIG_HOTPLUG_CPU
2175
2176static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
2177 struct hrtimer_clock_base *new_base)
2178{
2179 struct hrtimer *timer;
2180 struct timerqueue_node *node;
2181
2182 while ((node = timerqueue_getnext(&old_base->active))) {
2183 timer = container_of(node, struct hrtimer, node);
2184 BUG_ON(hrtimer_callback_running(timer));
2185 debug_deactivate(timer);
2186
2187 /*
2188 * Mark it as ENQUEUED not INACTIVE otherwise the
2189 * timer could be seen as !active and just vanish away
2190 * under us on another CPU
2191 */
2192 __remove_hrtimer(timer, old_base, HRTIMER_STATE_ENQUEUED, 0);
2193 timer->base = new_base;
2194 /*
2195 * Enqueue the timers on the new cpu. This does not
2196 * reprogram the event device in case the timer
2197 * expires before the earliest on this CPU, but we run
2198 * hrtimer_interrupt after we migrated everything to
2199 * sort out already expired timers and reprogram the
2200 * event device.
2201 */
2202 enqueue_hrtimer(timer, new_base, HRTIMER_MODE_ABS);
2203 }
2204}
2205
2206int hrtimers_cpu_dying(unsigned int dying_cpu)
2207{
2208 int i, ncpu = cpumask_any_and(cpu_active_mask, housekeeping_cpumask(HK_TYPE_TIMER));
2209 struct hrtimer_cpu_base *old_base, *new_base;
2210
2211 old_base = this_cpu_ptr(&hrtimer_bases);
2212 new_base = &per_cpu(hrtimer_bases, ncpu);
2213
2214 /*
2215 * The caller is globally serialized and nobody else
2216 * takes two locks at once, deadlock is not possible.
2217 */
2218 raw_spin_lock(&old_base->lock);
2219 raw_spin_lock_nested(&new_base->lock, SINGLE_DEPTH_NESTING);
2220
2221 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
2222 migrate_hrtimer_list(&old_base->clock_base[i],
2223 &new_base->clock_base[i]);
2224 }
2225
2226 /*
2227 * The migration might have changed the first expiring softirq
2228 * timer on this CPU. Update it.
2229 */
2230 __hrtimer_get_next_event(new_base, HRTIMER_ACTIVE_SOFT);
2231 /* Tell the other CPU to retrigger the next event */
2232 smp_call_function_single(ncpu, retrigger_next_event, NULL, 0);
2233
2234 raw_spin_unlock(&new_base->lock);
2235 old_base->online = 0;
2236 raw_spin_unlock(&old_base->lock);
2237
2238 return 0;
2239}
2240
2241#endif /* CONFIG_HOTPLUG_CPU */
2242
2243void __init hrtimers_init(void)
2244{
2245 hrtimers_prepare_cpu(smp_processor_id());
2246 open_softirq(HRTIMER_SOFTIRQ, hrtimer_run_softirq);
2247}
2248
2249/**
2250 * schedule_hrtimeout_range_clock - sleep until timeout
2251 * @expires: timeout value (ktime_t)
2252 * @delta: slack in expires timeout (ktime_t) for SCHED_OTHER tasks
2253 * @mode: timer mode
2254 * @clock_id: timer clock to be used
2255 */
2256int __sched
2257schedule_hrtimeout_range_clock(ktime_t *expires, u64 delta,
2258 const enum hrtimer_mode mode, clockid_t clock_id)
2259{
2260 struct hrtimer_sleeper t;
2261
2262 /*
2263 * Optimize when a zero timeout value is given. It does not
2264 * matter whether this is an absolute or a relative time.
2265 */
2266 if (expires && *expires == 0) {
2267 __set_current_state(TASK_RUNNING);
2268 return 0;
2269 }
2270
2271 /*
2272 * A NULL parameter means "infinite"
2273 */
2274 if (!expires) {
2275 schedule();
2276 return -EINTR;
2277 }
2278
2279 /*
2280 * Override any slack passed by the user if under
2281 * rt contraints.
2282 */
2283 if (rt_task(current))
2284 delta = 0;
2285
2286 hrtimer_init_sleeper_on_stack(&t, clock_id, mode);
2287 hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
2288 hrtimer_sleeper_start_expires(&t, mode);
2289
2290 if (likely(t.task))
2291 schedule();
2292
2293 hrtimer_cancel(&t.timer);
2294 destroy_hrtimer_on_stack(&t.timer);
2295
2296 __set_current_state(TASK_RUNNING);
2297
2298 return !t.task ? 0 : -EINTR;
2299}
2300EXPORT_SYMBOL_GPL(schedule_hrtimeout_range_clock);
2301
2302/**
2303 * schedule_hrtimeout_range - sleep until timeout
2304 * @expires: timeout value (ktime_t)
2305 * @delta: slack in expires timeout (ktime_t) for SCHED_OTHER tasks
2306 * @mode: timer mode
2307 *
2308 * Make the current task sleep until the given expiry time has
2309 * elapsed. The routine will return immediately unless
2310 * the current task state has been set (see set_current_state()).
2311 *
2312 * The @delta argument gives the kernel the freedom to schedule the
2313 * actual wakeup to a time that is both power and performance friendly
2314 * for regular (non RT/DL) tasks.
2315 * The kernel give the normal best effort behavior for "@expires+@delta",
2316 * but may decide to fire the timer earlier, but no earlier than @expires.
2317 *
2318 * You can set the task state as follows -
2319 *
2320 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
2321 * pass before the routine returns unless the current task is explicitly
2322 * woken up, (e.g. by wake_up_process()).
2323 *
2324 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
2325 * delivered to the current task or the current task is explicitly woken
2326 * up.
2327 *
2328 * The current task state is guaranteed to be TASK_RUNNING when this
2329 * routine returns.
2330 *
2331 * Returns 0 when the timer has expired. If the task was woken before the
2332 * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or
2333 * by an explicit wakeup, it returns -EINTR.
2334 */
2335int __sched schedule_hrtimeout_range(ktime_t *expires, u64 delta,
2336 const enum hrtimer_mode mode)
2337{
2338 return schedule_hrtimeout_range_clock(expires, delta, mode,
2339 CLOCK_MONOTONIC);
2340}
2341EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
2342
2343/**
2344 * schedule_hrtimeout - sleep until timeout
2345 * @expires: timeout value (ktime_t)
2346 * @mode: timer mode
2347 *
2348 * Make the current task sleep until the given expiry time has
2349 * elapsed. The routine will return immediately unless
2350 * the current task state has been set (see set_current_state()).
2351 *
2352 * You can set the task state as follows -
2353 *
2354 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
2355 * pass before the routine returns unless the current task is explicitly
2356 * woken up, (e.g. by wake_up_process()).
2357 *
2358 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
2359 * delivered to the current task or the current task is explicitly woken
2360 * up.
2361 *
2362 * The current task state is guaranteed to be TASK_RUNNING when this
2363 * routine returns.
2364 *
2365 * Returns 0 when the timer has expired. If the task was woken before the
2366 * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or
2367 * by an explicit wakeup, it returns -EINTR.
2368 */
2369int __sched schedule_hrtimeout(ktime_t *expires,
2370 const enum hrtimer_mode mode)
2371{
2372 return schedule_hrtimeout_range(expires, 0, mode);
2373}
2374EXPORT_SYMBOL_GPL(schedule_hrtimeout);