Loading...
Note: File does not exist in v3.15.
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright(C) 2005-2006, Thomas Gleixner <tglx@linutronix.de>
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{
168 struct hrtimer_clock_base *base;
169
170 for (;;) {
171 base = READ_ONCE(timer->base);
172 if (likely(base != &migration_base)) {
173 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
174 if (likely(base == timer->base))
175 return base;
176 /* The timer has migrated to another CPU: */
177 raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
178 }
179 cpu_relax();
180 }
181}
182
183/*
184 * We do not migrate the timer when it is expiring before the next
185 * event on the target cpu. When high resolution is enabled, we cannot
186 * reprogram the target cpu hardware and we would cause it to fire
187 * late. To keep it simple, we handle the high resolution enabled and
188 * disabled case similar.
189 *
190 * Called with cpu_base->lock of target cpu held.
191 */
192static int
193hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base)
194{
195 ktime_t expires;
196
197 expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset);
198 return expires < new_base->cpu_base->expires_next;
199}
200
201static inline
202struct hrtimer_cpu_base *get_target_base(struct hrtimer_cpu_base *base,
203 int pinned)
204{
205#if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
206 if (static_branch_likely(&timers_migration_enabled) && !pinned)
207 return &per_cpu(hrtimer_bases, get_nohz_timer_target());
208#endif
209 return base;
210}
211
212/*
213 * We switch the timer base to a power-optimized selected CPU target,
214 * if:
215 * - NO_HZ_COMMON is enabled
216 * - timer migration is enabled
217 * - the timer callback is not running
218 * - the timer is not the first expiring timer on the new target
219 *
220 * If one of the above requirements is not fulfilled we move the timer
221 * to the current CPU or leave it on the previously assigned CPU if
222 * the timer callback is currently running.
223 */
224static inline struct hrtimer_clock_base *
225switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
226 int pinned)
227{
228 struct hrtimer_cpu_base *new_cpu_base, *this_cpu_base;
229 struct hrtimer_clock_base *new_base;
230 int basenum = base->index;
231
232 this_cpu_base = this_cpu_ptr(&hrtimer_bases);
233 new_cpu_base = get_target_base(this_cpu_base, pinned);
234again:
235 new_base = &new_cpu_base->clock_base[basenum];
236
237 if (base != new_base) {
238 /*
239 * We are trying to move timer to new_base.
240 * However we can't change timer's base while it is running,
241 * so we keep it on the same CPU. No hassle vs. reprogramming
242 * the event source in the high resolution case. The softirq
243 * code will take care of this when the timer function has
244 * completed. There is no conflict as we hold the lock until
245 * the timer is enqueued.
246 */
247 if (unlikely(hrtimer_callback_running(timer)))
248 return base;
249
250 /* See the comment in lock_hrtimer_base() */
251 WRITE_ONCE(timer->base, &migration_base);
252 raw_spin_unlock(&base->cpu_base->lock);
253 raw_spin_lock(&new_base->cpu_base->lock);
254
255 if (new_cpu_base != this_cpu_base &&
256 hrtimer_check_target(timer, new_base)) {
257 raw_spin_unlock(&new_base->cpu_base->lock);
258 raw_spin_lock(&base->cpu_base->lock);
259 new_cpu_base = this_cpu_base;
260 WRITE_ONCE(timer->base, base);
261 goto again;
262 }
263 WRITE_ONCE(timer->base, new_base);
264 } else {
265 if (new_cpu_base != this_cpu_base &&
266 hrtimer_check_target(timer, new_base)) {
267 new_cpu_base = this_cpu_base;
268 goto again;
269 }
270 }
271 return new_base;
272}
273
274#else /* CONFIG_SMP */
275
276static inline bool is_migration_base(struct hrtimer_clock_base *base)
277{
278 return false;
279}
280
281static inline struct hrtimer_clock_base *
282lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
283{
284 struct hrtimer_clock_base *base = timer->base;
285
286 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
287
288 return base;
289}
290
291# define switch_hrtimer_base(t, b, p) (b)
292
293#endif /* !CONFIG_SMP */
294
295/*
296 * Functions for the union type storage format of ktime_t which are
297 * too large for inlining:
298 */
299#if BITS_PER_LONG < 64
300/*
301 * Divide a ktime value by a nanosecond value
302 */
303s64 __ktime_divns(const ktime_t kt, s64 div)
304{
305 int sft = 0;
306 s64 dclc;
307 u64 tmp;
308
309 dclc = ktime_to_ns(kt);
310 tmp = dclc < 0 ? -dclc : dclc;
311
312 /* Make sure the divisor is less than 2^32: */
313 while (div >> 32) {
314 sft++;
315 div >>= 1;
316 }
317 tmp >>= sft;
318 do_div(tmp, (u32) div);
319 return dclc < 0 ? -tmp : tmp;
320}
321EXPORT_SYMBOL_GPL(__ktime_divns);
322#endif /* BITS_PER_LONG >= 64 */
323
324/*
325 * Add two ktime values and do a safety check for overflow:
326 */
327ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
328{
329 ktime_t res = ktime_add_unsafe(lhs, rhs);
330
331 /*
332 * We use KTIME_SEC_MAX here, the maximum timeout which we can
333 * return to user space in a timespec:
334 */
335 if (res < 0 || res < lhs || res < rhs)
336 res = ktime_set(KTIME_SEC_MAX, 0);
337
338 return res;
339}
340
341EXPORT_SYMBOL_GPL(ktime_add_safe);
342
343#ifdef CONFIG_DEBUG_OBJECTS_TIMERS
344
345static const struct debug_obj_descr hrtimer_debug_descr;
346
347static void *hrtimer_debug_hint(void *addr)
348{
349 return ((struct hrtimer *) addr)->function;
350}
351
352/*
353 * fixup_init is called when:
354 * - an active object is initialized
355 */
356static bool hrtimer_fixup_init(void *addr, enum debug_obj_state state)
357{
358 struct hrtimer *timer = addr;
359
360 switch (state) {
361 case ODEBUG_STATE_ACTIVE:
362 hrtimer_cancel(timer);
363 debug_object_init(timer, &hrtimer_debug_descr);
364 return true;
365 default:
366 return false;
367 }
368}
369
370/*
371 * fixup_activate is called when:
372 * - an active object is activated
373 * - an unknown non-static object is activated
374 */
375static bool hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
376{
377 switch (state) {
378 case ODEBUG_STATE_ACTIVE:
379 WARN_ON(1);
380 fallthrough;
381 default:
382 return false;
383 }
384}
385
386/*
387 * fixup_free is called when:
388 * - an active object is freed
389 */
390static bool hrtimer_fixup_free(void *addr, enum debug_obj_state state)
391{
392 struct hrtimer *timer = addr;
393
394 switch (state) {
395 case ODEBUG_STATE_ACTIVE:
396 hrtimer_cancel(timer);
397 debug_object_free(timer, &hrtimer_debug_descr);
398 return true;
399 default:
400 return false;
401 }
402}
403
404static const struct debug_obj_descr hrtimer_debug_descr = {
405 .name = "hrtimer",
406 .debug_hint = hrtimer_debug_hint,
407 .fixup_init = hrtimer_fixup_init,
408 .fixup_activate = hrtimer_fixup_activate,
409 .fixup_free = hrtimer_fixup_free,
410};
411
412static inline void debug_hrtimer_init(struct hrtimer *timer)
413{
414 debug_object_init(timer, &hrtimer_debug_descr);
415}
416
417static inline void debug_hrtimer_activate(struct hrtimer *timer,
418 enum hrtimer_mode mode)
419{
420 debug_object_activate(timer, &hrtimer_debug_descr);
421}
422
423static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
424{
425 debug_object_deactivate(timer, &hrtimer_debug_descr);
426}
427
428static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
429 enum hrtimer_mode mode);
430
431void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
432 enum hrtimer_mode mode)
433{
434 debug_object_init_on_stack(timer, &hrtimer_debug_descr);
435 __hrtimer_init(timer, clock_id, mode);
436}
437EXPORT_SYMBOL_GPL(hrtimer_init_on_stack);
438
439static void __hrtimer_init_sleeper(struct hrtimer_sleeper *sl,
440 clockid_t clock_id, enum hrtimer_mode mode);
441
442void hrtimer_init_sleeper_on_stack(struct hrtimer_sleeper *sl,
443 clockid_t clock_id, enum hrtimer_mode mode)
444{
445 debug_object_init_on_stack(&sl->timer, &hrtimer_debug_descr);
446 __hrtimer_init_sleeper(sl, clock_id, mode);
447}
448EXPORT_SYMBOL_GPL(hrtimer_init_sleeper_on_stack);
449
450void destroy_hrtimer_on_stack(struct hrtimer *timer)
451{
452 debug_object_free(timer, &hrtimer_debug_descr);
453}
454EXPORT_SYMBOL_GPL(destroy_hrtimer_on_stack);
455
456#else
457
458static inline void debug_hrtimer_init(struct hrtimer *timer) { }
459static inline void debug_hrtimer_activate(struct hrtimer *timer,
460 enum hrtimer_mode mode) { }
461static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
462#endif
463
464static inline void
465debug_init(struct hrtimer *timer, clockid_t clockid,
466 enum hrtimer_mode mode)
467{
468 debug_hrtimer_init(timer);
469 trace_hrtimer_init(timer, clockid, mode);
470}
471
472static inline void debug_activate(struct hrtimer *timer,
473 enum hrtimer_mode mode)
474{
475 debug_hrtimer_activate(timer, mode);
476 trace_hrtimer_start(timer, mode);
477}
478
479static inline void debug_deactivate(struct hrtimer *timer)
480{
481 debug_hrtimer_deactivate(timer);
482 trace_hrtimer_cancel(timer);
483}
484
485static struct hrtimer_clock_base *
486__next_base(struct hrtimer_cpu_base *cpu_base, unsigned int *active)
487{
488 unsigned int idx;
489
490 if (!*active)
491 return NULL;
492
493 idx = __ffs(*active);
494 *active &= ~(1U << idx);
495
496 return &cpu_base->clock_base[idx];
497}
498
499#define for_each_active_base(base, cpu_base, active) \
500 while ((base = __next_base((cpu_base), &(active))))
501
502static ktime_t __hrtimer_next_event_base(struct hrtimer_cpu_base *cpu_base,
503 const struct hrtimer *exclude,
504 unsigned int active,
505 ktime_t expires_next)
506{
507 struct hrtimer_clock_base *base;
508 ktime_t expires;
509
510 for_each_active_base(base, cpu_base, active) {
511 struct timerqueue_node *next;
512 struct hrtimer *timer;
513
514 next = timerqueue_getnext(&base->active);
515 timer = container_of(next, struct hrtimer, node);
516 if (timer == exclude) {
517 /* Get to the next timer in the queue. */
518 next = timerqueue_iterate_next(next);
519 if (!next)
520 continue;
521
522 timer = container_of(next, struct hrtimer, node);
523 }
524 expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
525 if (expires < expires_next) {
526 expires_next = expires;
527
528 /* Skip cpu_base update if a timer is being excluded. */
529 if (exclude)
530 continue;
531
532 if (timer->is_soft)
533 cpu_base->softirq_next_timer = timer;
534 else
535 cpu_base->next_timer = timer;
536 }
537 }
538 /*
539 * clock_was_set() might have changed base->offset of any of
540 * the clock bases so the result might be negative. Fix it up
541 * to prevent a false positive in clockevents_program_event().
542 */
543 if (expires_next < 0)
544 expires_next = 0;
545 return expires_next;
546}
547
548/*
549 * Recomputes cpu_base::*next_timer and returns the earliest expires_next
550 * but does not set cpu_base::*expires_next, that is done by
551 * hrtimer[_force]_reprogram and hrtimer_interrupt only. When updating
552 * cpu_base::*expires_next right away, reprogramming logic would no longer
553 * work.
554 *
555 * When a softirq is pending, we can ignore the HRTIMER_ACTIVE_SOFT bases,
556 * those timers will get run whenever the softirq gets handled, at the end of
557 * hrtimer_run_softirq(), hrtimer_update_softirq_timer() will re-add these bases.
558 *
559 * Therefore softirq values are those from the HRTIMER_ACTIVE_SOFT clock bases.
560 * The !softirq values are the minima across HRTIMER_ACTIVE_ALL, unless an actual
561 * softirq is pending, in which case they're the minima of HRTIMER_ACTIVE_HARD.
562 *
563 * @active_mask must be one of:
564 * - HRTIMER_ACTIVE_ALL,
565 * - HRTIMER_ACTIVE_SOFT, or
566 * - HRTIMER_ACTIVE_HARD.
567 */
568static ktime_t
569__hrtimer_get_next_event(struct hrtimer_cpu_base *cpu_base, unsigned int active_mask)
570{
571 unsigned int active;
572 struct hrtimer *next_timer = NULL;
573 ktime_t expires_next = KTIME_MAX;
574
575 if (!cpu_base->softirq_activated && (active_mask & HRTIMER_ACTIVE_SOFT)) {
576 active = cpu_base->active_bases & HRTIMER_ACTIVE_SOFT;
577 cpu_base->softirq_next_timer = NULL;
578 expires_next = __hrtimer_next_event_base(cpu_base, NULL,
579 active, KTIME_MAX);
580
581 next_timer = cpu_base->softirq_next_timer;
582 }
583
584 if (active_mask & HRTIMER_ACTIVE_HARD) {
585 active = cpu_base->active_bases & HRTIMER_ACTIVE_HARD;
586 cpu_base->next_timer = next_timer;
587 expires_next = __hrtimer_next_event_base(cpu_base, NULL, active,
588 expires_next);
589 }
590
591 return expires_next;
592}
593
594static ktime_t hrtimer_update_next_event(struct hrtimer_cpu_base *cpu_base)
595{
596 ktime_t expires_next, soft = KTIME_MAX;
597
598 /*
599 * If the soft interrupt has already been activated, ignore the
600 * soft bases. They will be handled in the already raised soft
601 * interrupt.
602 */
603 if (!cpu_base->softirq_activated) {
604 soft = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_SOFT);
605 /*
606 * Update the soft expiry time. clock_settime() might have
607 * affected it.
608 */
609 cpu_base->softirq_expires_next = soft;
610 }
611
612 expires_next = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_HARD);
613 /*
614 * If a softirq timer is expiring first, update cpu_base->next_timer
615 * and program the hardware with the soft expiry time.
616 */
617 if (expires_next > soft) {
618 cpu_base->next_timer = cpu_base->softirq_next_timer;
619 expires_next = soft;
620 }
621
622 return expires_next;
623}
624
625static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base)
626{
627 ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset;
628 ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset;
629 ktime_t *offs_tai = &base->clock_base[HRTIMER_BASE_TAI].offset;
630
631 ktime_t now = ktime_get_update_offsets_now(&base->clock_was_set_seq,
632 offs_real, offs_boot, offs_tai);
633
634 base->clock_base[HRTIMER_BASE_REALTIME_SOFT].offset = *offs_real;
635 base->clock_base[HRTIMER_BASE_BOOTTIME_SOFT].offset = *offs_boot;
636 base->clock_base[HRTIMER_BASE_TAI_SOFT].offset = *offs_tai;
637
638 return now;
639}
640
641/*
642 * Is the high resolution mode active ?
643 */
644static inline int __hrtimer_hres_active(struct hrtimer_cpu_base *cpu_base)
645{
646 return IS_ENABLED(CONFIG_HIGH_RES_TIMERS) ?
647 cpu_base->hres_active : 0;
648}
649
650static inline int hrtimer_hres_active(void)
651{
652 return __hrtimer_hres_active(this_cpu_ptr(&hrtimer_bases));
653}
654
655/*
656 * Reprogram the event source with checking both queues for the
657 * next event
658 * Called with interrupts disabled and base->lock held
659 */
660static void
661hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
662{
663 ktime_t expires_next;
664
665 expires_next = hrtimer_update_next_event(cpu_base);
666
667 if (skip_equal && expires_next == cpu_base->expires_next)
668 return;
669
670 cpu_base->expires_next = expires_next;
671
672 /*
673 * If hres is not active, hardware does not have to be
674 * reprogrammed yet.
675 *
676 * If a hang was detected in the last timer interrupt then we
677 * leave the hang delay active in the hardware. We want the
678 * system to make progress. That also prevents the following
679 * scenario:
680 * T1 expires 50ms from now
681 * T2 expires 5s from now
682 *
683 * T1 is removed, so this code is called and would reprogram
684 * the hardware to 5s from now. Any hrtimer_start after that
685 * will not reprogram the hardware due to hang_detected being
686 * set. So we'd effectively block all timers until the T2 event
687 * fires.
688 */
689 if (!__hrtimer_hres_active(cpu_base) || cpu_base->hang_detected)
690 return;
691
692 tick_program_event(cpu_base->expires_next, 1);
693}
694
695/* High resolution timer related functions */
696#ifdef CONFIG_HIGH_RES_TIMERS
697
698/*
699 * High resolution timer enabled ?
700 */
701static bool hrtimer_hres_enabled __read_mostly = true;
702unsigned int hrtimer_resolution __read_mostly = LOW_RES_NSEC;
703EXPORT_SYMBOL_GPL(hrtimer_resolution);
704
705/*
706 * Enable / Disable high resolution mode
707 */
708static int __init setup_hrtimer_hres(char *str)
709{
710 return (kstrtobool(str, &hrtimer_hres_enabled) == 0);
711}
712
713__setup("highres=", setup_hrtimer_hres);
714
715/*
716 * hrtimer_high_res_enabled - query, if the highres mode is enabled
717 */
718static inline int hrtimer_is_hres_enabled(void)
719{
720 return hrtimer_hres_enabled;
721}
722
723/*
724 * Retrigger next event is called after clock was set
725 *
726 * Called with interrupts disabled via on_each_cpu()
727 */
728static void retrigger_next_event(void *arg)
729{
730 struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
731
732 if (!__hrtimer_hres_active(base))
733 return;
734
735 raw_spin_lock(&base->lock);
736 hrtimer_update_base(base);
737 hrtimer_force_reprogram(base, 0);
738 raw_spin_unlock(&base->lock);
739}
740
741/*
742 * Switch to high resolution mode
743 */
744static void hrtimer_switch_to_hres(void)
745{
746 struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
747
748 if (tick_init_highres()) {
749 pr_warn("Could not switch to high resolution mode on CPU %u\n",
750 base->cpu);
751 return;
752 }
753 base->hres_active = 1;
754 hrtimer_resolution = HIGH_RES_NSEC;
755
756 tick_setup_sched_timer();
757 /* "Retrigger" the interrupt to get things going */
758 retrigger_next_event(NULL);
759}
760
761#else
762
763static inline int hrtimer_is_hres_enabled(void) { return 0; }
764static inline void hrtimer_switch_to_hres(void) { }
765static inline void retrigger_next_event(void *arg) { }
766
767#endif /* CONFIG_HIGH_RES_TIMERS */
768
769/*
770 * When a timer is enqueued and expires earlier than the already enqueued
771 * timers, we have to check, whether it expires earlier than the timer for
772 * which the clock event device was armed.
773 *
774 * Called with interrupts disabled and base->cpu_base.lock held
775 */
776static void hrtimer_reprogram(struct hrtimer *timer, bool reprogram)
777{
778 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
779 struct hrtimer_clock_base *base = timer->base;
780 ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
781
782 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
783
784 /*
785 * CLOCK_REALTIME timer might be requested with an absolute
786 * expiry time which is less than base->offset. Set it to 0.
787 */
788 if (expires < 0)
789 expires = 0;
790
791 if (timer->is_soft) {
792 /*
793 * soft hrtimer could be started on a remote CPU. In this
794 * case softirq_expires_next needs to be updated on the
795 * remote CPU. The soft hrtimer will not expire before the
796 * first hard hrtimer on the remote CPU -
797 * hrtimer_check_target() prevents this case.
798 */
799 struct hrtimer_cpu_base *timer_cpu_base = base->cpu_base;
800
801 if (timer_cpu_base->softirq_activated)
802 return;
803
804 if (!ktime_before(expires, timer_cpu_base->softirq_expires_next))
805 return;
806
807 timer_cpu_base->softirq_next_timer = timer;
808 timer_cpu_base->softirq_expires_next = expires;
809
810 if (!ktime_before(expires, timer_cpu_base->expires_next) ||
811 !reprogram)
812 return;
813 }
814
815 /*
816 * If the timer is not on the current cpu, we cannot reprogram
817 * the other cpus clock event device.
818 */
819 if (base->cpu_base != cpu_base)
820 return;
821
822 /*
823 * If the hrtimer interrupt is running, then it will
824 * reevaluate the clock bases and reprogram the clock event
825 * device. The callbacks are always executed in hard interrupt
826 * context so we don't need an extra check for a running
827 * callback.
828 */
829 if (cpu_base->in_hrtirq)
830 return;
831
832 if (expires >= cpu_base->expires_next)
833 return;
834
835 /* Update the pointer to the next expiring timer */
836 cpu_base->next_timer = timer;
837 cpu_base->expires_next = expires;
838
839 /*
840 * If hres is not active, hardware does not have to be
841 * programmed yet.
842 *
843 * If a hang was detected in the last timer interrupt then we
844 * do not schedule a timer which is earlier than the expiry
845 * which we enforced in the hang detection. We want the system
846 * to make progress.
847 */
848 if (!__hrtimer_hres_active(cpu_base) || cpu_base->hang_detected)
849 return;
850
851 /*
852 * Program the timer hardware. We enforce the expiry for
853 * events which are already in the past.
854 */
855 tick_program_event(expires, 1);
856}
857
858/*
859 * Clock realtime was set
860 *
861 * Change the offset of the realtime clock vs. the monotonic
862 * clock.
863 *
864 * We might have to reprogram the high resolution timer interrupt. On
865 * SMP we call the architecture specific code to retrigger _all_ high
866 * resolution timer interrupts. On UP we just disable interrupts and
867 * call the high resolution interrupt code.
868 */
869void clock_was_set(void)
870{
871#ifdef CONFIG_HIGH_RES_TIMERS
872 /* Retrigger the CPU local events everywhere */
873 on_each_cpu(retrigger_next_event, NULL, 1);
874#endif
875 timerfd_clock_was_set();
876}
877
878static void clock_was_set_work(struct work_struct *work)
879{
880 clock_was_set();
881}
882
883static DECLARE_WORK(hrtimer_work, clock_was_set_work);
884
885/*
886 * Called from timekeeping and resume code to reprogram the hrtimer
887 * interrupt device on all cpus and to notify timerfd.
888 */
889void clock_was_set_delayed(void)
890{
891 schedule_work(&hrtimer_work);
892}
893
894/*
895 * During resume we might have to reprogram the high resolution timer
896 * interrupt on all online CPUs. However, all other CPUs will be
897 * stopped with IRQs interrupts disabled so the clock_was_set() call
898 * must be deferred.
899 */
900void hrtimers_resume(void)
901{
902 lockdep_assert_irqs_disabled();
903 /* Retrigger on the local CPU */
904 retrigger_next_event(NULL);
905 /* And schedule a retrigger for all others */
906 clock_was_set_delayed();
907}
908
909/*
910 * Counterpart to lock_hrtimer_base above:
911 */
912static inline
913void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
914{
915 raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
916}
917
918/**
919 * hrtimer_forward - forward the timer expiry
920 * @timer: hrtimer to forward
921 * @now: forward past this time
922 * @interval: the interval to forward
923 *
924 * Forward the timer expiry so it will expire in the future.
925 * Returns the number of overruns.
926 *
927 * Can be safely called from the callback function of @timer. If
928 * called from other contexts @timer must neither be enqueued nor
929 * running the callback and the caller needs to take care of
930 * serialization.
931 *
932 * Note: This only updates the timer expiry value and does not requeue
933 * the timer.
934 */
935u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
936{
937 u64 orun = 1;
938 ktime_t delta;
939
940 delta = ktime_sub(now, hrtimer_get_expires(timer));
941
942 if (delta < 0)
943 return 0;
944
945 if (WARN_ON(timer->state & HRTIMER_STATE_ENQUEUED))
946 return 0;
947
948 if (interval < hrtimer_resolution)
949 interval = hrtimer_resolution;
950
951 if (unlikely(delta >= interval)) {
952 s64 incr = ktime_to_ns(interval);
953
954 orun = ktime_divns(delta, incr);
955 hrtimer_add_expires_ns(timer, incr * orun);
956 if (hrtimer_get_expires_tv64(timer) > now)
957 return orun;
958 /*
959 * This (and the ktime_add() below) is the
960 * correction for exact:
961 */
962 orun++;
963 }
964 hrtimer_add_expires(timer, interval);
965
966 return orun;
967}
968EXPORT_SYMBOL_GPL(hrtimer_forward);
969
970/*
971 * enqueue_hrtimer - internal function to (re)start a timer
972 *
973 * The timer is inserted in expiry order. Insertion into the
974 * red black tree is O(log(n)). Must hold the base lock.
975 *
976 * Returns 1 when the new timer is the leftmost timer in the tree.
977 */
978static int enqueue_hrtimer(struct hrtimer *timer,
979 struct hrtimer_clock_base *base,
980 enum hrtimer_mode mode)
981{
982 debug_activate(timer, mode);
983
984 base->cpu_base->active_bases |= 1 << base->index;
985
986 /* Pairs with the lockless read in hrtimer_is_queued() */
987 WRITE_ONCE(timer->state, HRTIMER_STATE_ENQUEUED);
988
989 return timerqueue_add(&base->active, &timer->node);
990}
991
992/*
993 * __remove_hrtimer - internal function to remove a timer
994 *
995 * Caller must hold the base lock.
996 *
997 * High resolution timer mode reprograms the clock event device when the
998 * timer is the one which expires next. The caller can disable this by setting
999 * reprogram to zero. This is useful, when the context does a reprogramming
1000 * anyway (e.g. timer interrupt)
1001 */
1002static void __remove_hrtimer(struct hrtimer *timer,
1003 struct hrtimer_clock_base *base,
1004 u8 newstate, int reprogram)
1005{
1006 struct hrtimer_cpu_base *cpu_base = base->cpu_base;
1007 u8 state = timer->state;
1008
1009 /* Pairs with the lockless read in hrtimer_is_queued() */
1010 WRITE_ONCE(timer->state, newstate);
1011 if (!(state & HRTIMER_STATE_ENQUEUED))
1012 return;
1013
1014 if (!timerqueue_del(&base->active, &timer->node))
1015 cpu_base->active_bases &= ~(1 << base->index);
1016
1017 /*
1018 * Note: If reprogram is false we do not update
1019 * cpu_base->next_timer. This happens when we remove the first
1020 * timer on a remote cpu. No harm as we never dereference
1021 * cpu_base->next_timer. So the worst thing what can happen is
1022 * an superfluous call to hrtimer_force_reprogram() on the
1023 * remote cpu later on if the same timer gets enqueued again.
1024 */
1025 if (reprogram && timer == cpu_base->next_timer)
1026 hrtimer_force_reprogram(cpu_base, 1);
1027}
1028
1029/*
1030 * remove hrtimer, called with base lock held
1031 */
1032static inline int
1033remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base,
1034 bool restart, bool keep_local)
1035{
1036 u8 state = timer->state;
1037
1038 if (state & HRTIMER_STATE_ENQUEUED) {
1039 bool reprogram;
1040
1041 /*
1042 * Remove the timer and force reprogramming when high
1043 * resolution mode is active and the timer is on the current
1044 * CPU. If we remove a timer on another CPU, reprogramming is
1045 * skipped. The interrupt event on this CPU is fired and
1046 * reprogramming happens in the interrupt handler. This is a
1047 * rare case and less expensive than a smp call.
1048 */
1049 debug_deactivate(timer);
1050 reprogram = base->cpu_base == this_cpu_ptr(&hrtimer_bases);
1051
1052 /*
1053 * If the timer is not restarted then reprogramming is
1054 * required if the timer is local. If it is local and about
1055 * to be restarted, avoid programming it twice (on removal
1056 * and a moment later when it's requeued).
1057 */
1058 if (!restart)
1059 state = HRTIMER_STATE_INACTIVE;
1060 else
1061 reprogram &= !keep_local;
1062
1063 __remove_hrtimer(timer, base, state, reprogram);
1064 return 1;
1065 }
1066 return 0;
1067}
1068
1069static inline ktime_t hrtimer_update_lowres(struct hrtimer *timer, ktime_t tim,
1070 const enum hrtimer_mode mode)
1071{
1072#ifdef CONFIG_TIME_LOW_RES
1073 /*
1074 * CONFIG_TIME_LOW_RES indicates that the system has no way to return
1075 * granular time values. For relative timers we add hrtimer_resolution
1076 * (i.e. one jiffie) to prevent short timeouts.
1077 */
1078 timer->is_rel = mode & HRTIMER_MODE_REL;
1079 if (timer->is_rel)
1080 tim = ktime_add_safe(tim, hrtimer_resolution);
1081#endif
1082 return tim;
1083}
1084
1085static void
1086hrtimer_update_softirq_timer(struct hrtimer_cpu_base *cpu_base, bool reprogram)
1087{
1088 ktime_t expires;
1089
1090 /*
1091 * Find the next SOFT expiration.
1092 */
1093 expires = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_SOFT);
1094
1095 /*
1096 * reprogramming needs to be triggered, even if the next soft
1097 * hrtimer expires at the same time than the next hard
1098 * hrtimer. cpu_base->softirq_expires_next needs to be updated!
1099 */
1100 if (expires == KTIME_MAX)
1101 return;
1102
1103 /*
1104 * cpu_base->*next_timer is recomputed by __hrtimer_get_next_event()
1105 * cpu_base->*expires_next is only set by hrtimer_reprogram()
1106 */
1107 hrtimer_reprogram(cpu_base->softirq_next_timer, reprogram);
1108}
1109
1110static int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1111 u64 delta_ns, const enum hrtimer_mode mode,
1112 struct hrtimer_clock_base *base)
1113{
1114 struct hrtimer_clock_base *new_base;
1115 bool force_local, first;
1116
1117 /*
1118 * If the timer is on the local cpu base and is the first expiring
1119 * timer then this might end up reprogramming the hardware twice
1120 * (on removal and on enqueue). To avoid that by prevent the
1121 * reprogram on removal, keep the timer local to the current CPU
1122 * and enforce reprogramming after it is queued no matter whether
1123 * it is the new first expiring timer again or not.
1124 */
1125 force_local = base->cpu_base == this_cpu_ptr(&hrtimer_bases);
1126 force_local &= base->cpu_base->next_timer == timer;
1127
1128 /*
1129 * Remove an active timer from the queue. In case it is not queued
1130 * on the current CPU, make sure that remove_hrtimer() updates the
1131 * remote data correctly.
1132 *
1133 * If it's on the current CPU and the first expiring timer, then
1134 * skip reprogramming, keep the timer local and enforce
1135 * reprogramming later if it was the first expiring timer. This
1136 * avoids programming the underlying clock event twice (once at
1137 * removal and once after enqueue).
1138 */
1139 remove_hrtimer(timer, base, true, force_local);
1140
1141 if (mode & HRTIMER_MODE_REL)
1142 tim = ktime_add_safe(tim, base->get_time());
1143
1144 tim = hrtimer_update_lowres(timer, tim, mode);
1145
1146 hrtimer_set_expires_range_ns(timer, tim, delta_ns);
1147
1148 /* Switch the timer base, if necessary: */
1149 if (!force_local) {
1150 new_base = switch_hrtimer_base(timer, base,
1151 mode & HRTIMER_MODE_PINNED);
1152 } else {
1153 new_base = base;
1154 }
1155
1156 first = enqueue_hrtimer(timer, new_base, mode);
1157 if (!force_local)
1158 return first;
1159
1160 /*
1161 * Timer was forced to stay on the current CPU to avoid
1162 * reprogramming on removal and enqueue. Force reprogram the
1163 * hardware by evaluating the new first expiring timer.
1164 */
1165 hrtimer_force_reprogram(new_base->cpu_base, 1);
1166 return 0;
1167}
1168
1169/**
1170 * hrtimer_start_range_ns - (re)start an hrtimer
1171 * @timer: the timer to be added
1172 * @tim: expiry time
1173 * @delta_ns: "slack" range for the timer
1174 * @mode: timer mode: absolute (HRTIMER_MODE_ABS) or
1175 * relative (HRTIMER_MODE_REL), and pinned (HRTIMER_MODE_PINNED);
1176 * softirq based mode is considered for debug purpose only!
1177 */
1178void hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1179 u64 delta_ns, const enum hrtimer_mode mode)
1180{
1181 struct hrtimer_clock_base *base;
1182 unsigned long flags;
1183
1184 /*
1185 * Check whether the HRTIMER_MODE_SOFT bit and hrtimer.is_soft
1186 * match on CONFIG_PREEMPT_RT = n. With PREEMPT_RT check the hard
1187 * expiry mode because unmarked timers are moved to softirq expiry.
1188 */
1189 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
1190 WARN_ON_ONCE(!(mode & HRTIMER_MODE_SOFT) ^ !timer->is_soft);
1191 else
1192 WARN_ON_ONCE(!(mode & HRTIMER_MODE_HARD) ^ !timer->is_hard);
1193
1194 base = lock_hrtimer_base(timer, &flags);
1195
1196 if (__hrtimer_start_range_ns(timer, tim, delta_ns, mode, base))
1197 hrtimer_reprogram(timer, true);
1198
1199 unlock_hrtimer_base(timer, &flags);
1200}
1201EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
1202
1203/**
1204 * hrtimer_try_to_cancel - try to deactivate a timer
1205 * @timer: hrtimer to stop
1206 *
1207 * Returns:
1208 *
1209 * * 0 when the timer was not active
1210 * * 1 when the timer was active
1211 * * -1 when the timer is currently executing the callback function and
1212 * cannot be stopped
1213 */
1214int hrtimer_try_to_cancel(struct hrtimer *timer)
1215{
1216 struct hrtimer_clock_base *base;
1217 unsigned long flags;
1218 int ret = -1;
1219
1220 /*
1221 * Check lockless first. If the timer is not active (neither
1222 * enqueued nor running the callback, nothing to do here. The
1223 * base lock does not serialize against a concurrent enqueue,
1224 * so we can avoid taking it.
1225 */
1226 if (!hrtimer_active(timer))
1227 return 0;
1228
1229 base = lock_hrtimer_base(timer, &flags);
1230
1231 if (!hrtimer_callback_running(timer))
1232 ret = remove_hrtimer(timer, base, false, false);
1233
1234 unlock_hrtimer_base(timer, &flags);
1235
1236 return ret;
1237
1238}
1239EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1240
1241#ifdef CONFIG_PREEMPT_RT
1242static void hrtimer_cpu_base_init_expiry_lock(struct hrtimer_cpu_base *base)
1243{
1244 spin_lock_init(&base->softirq_expiry_lock);
1245}
1246
1247static void hrtimer_cpu_base_lock_expiry(struct hrtimer_cpu_base *base)
1248{
1249 spin_lock(&base->softirq_expiry_lock);
1250}
1251
1252static void hrtimer_cpu_base_unlock_expiry(struct hrtimer_cpu_base *base)
1253{
1254 spin_unlock(&base->softirq_expiry_lock);
1255}
1256
1257/*
1258 * The counterpart to hrtimer_cancel_wait_running().
1259 *
1260 * If there is a waiter for cpu_base->expiry_lock, then it was waiting for
1261 * the timer callback to finish. Drop expiry_lock and reacquire it. That
1262 * allows the waiter to acquire the lock and make progress.
1263 */
1264static void hrtimer_sync_wait_running(struct hrtimer_cpu_base *cpu_base,
1265 unsigned long flags)
1266{
1267 if (atomic_read(&cpu_base->timer_waiters)) {
1268 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1269 spin_unlock(&cpu_base->softirq_expiry_lock);
1270 spin_lock(&cpu_base->softirq_expiry_lock);
1271 raw_spin_lock_irq(&cpu_base->lock);
1272 }
1273}
1274
1275/*
1276 * This function is called on PREEMPT_RT kernels when the fast path
1277 * deletion of a timer failed because the timer callback function was
1278 * running.
1279 *
1280 * This prevents priority inversion: if the soft irq thread is preempted
1281 * in the middle of a timer callback, then calling del_timer_sync() can
1282 * lead to two issues:
1283 *
1284 * - If the caller is on a remote CPU then it has to spin wait for the timer
1285 * handler to complete. This can result in unbound priority inversion.
1286 *
1287 * - If the caller originates from the task which preempted the timer
1288 * handler on the same CPU, then spin waiting for the timer handler to
1289 * complete is never going to end.
1290 */
1291void hrtimer_cancel_wait_running(const struct hrtimer *timer)
1292{
1293 /* Lockless read. Prevent the compiler from reloading it below */
1294 struct hrtimer_clock_base *base = READ_ONCE(timer->base);
1295
1296 /*
1297 * Just relax if the timer expires in hard interrupt context or if
1298 * it is currently on the migration base.
1299 */
1300 if (!timer->is_soft || is_migration_base(base)) {
1301 cpu_relax();
1302 return;
1303 }
1304
1305 /*
1306 * Mark the base as contended and grab the expiry lock, which is
1307 * held by the softirq across the timer callback. Drop the lock
1308 * immediately so the softirq can expire the next timer. In theory
1309 * the timer could already be running again, but that's more than
1310 * unlikely and just causes another wait loop.
1311 */
1312 atomic_inc(&base->cpu_base->timer_waiters);
1313 spin_lock_bh(&base->cpu_base->softirq_expiry_lock);
1314 atomic_dec(&base->cpu_base->timer_waiters);
1315 spin_unlock_bh(&base->cpu_base->softirq_expiry_lock);
1316}
1317#else
1318static inline void
1319hrtimer_cpu_base_init_expiry_lock(struct hrtimer_cpu_base *base) { }
1320static inline void
1321hrtimer_cpu_base_lock_expiry(struct hrtimer_cpu_base *base) { }
1322static inline void
1323hrtimer_cpu_base_unlock_expiry(struct hrtimer_cpu_base *base) { }
1324static inline void hrtimer_sync_wait_running(struct hrtimer_cpu_base *base,
1325 unsigned long flags) { }
1326#endif
1327
1328/**
1329 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1330 * @timer: the timer to be cancelled
1331 *
1332 * Returns:
1333 * 0 when the timer was not active
1334 * 1 when the timer was active
1335 */
1336int hrtimer_cancel(struct hrtimer *timer)
1337{
1338 int ret;
1339
1340 do {
1341 ret = hrtimer_try_to_cancel(timer);
1342
1343 if (ret < 0)
1344 hrtimer_cancel_wait_running(timer);
1345 } while (ret < 0);
1346 return ret;
1347}
1348EXPORT_SYMBOL_GPL(hrtimer_cancel);
1349
1350/**
1351 * __hrtimer_get_remaining - get remaining time for the timer
1352 * @timer: the timer to read
1353 * @adjust: adjust relative timers when CONFIG_TIME_LOW_RES=y
1354 */
1355ktime_t __hrtimer_get_remaining(const struct hrtimer *timer, bool adjust)
1356{
1357 unsigned long flags;
1358 ktime_t rem;
1359
1360 lock_hrtimer_base(timer, &flags);
1361 if (IS_ENABLED(CONFIG_TIME_LOW_RES) && adjust)
1362 rem = hrtimer_expires_remaining_adjusted(timer);
1363 else
1364 rem = hrtimer_expires_remaining(timer);
1365 unlock_hrtimer_base(timer, &flags);
1366
1367 return rem;
1368}
1369EXPORT_SYMBOL_GPL(__hrtimer_get_remaining);
1370
1371#ifdef CONFIG_NO_HZ_COMMON
1372/**
1373 * hrtimer_get_next_event - get the time until next expiry event
1374 *
1375 * Returns the next expiry time or KTIME_MAX if no timer is pending.
1376 */
1377u64 hrtimer_get_next_event(void)
1378{
1379 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1380 u64 expires = KTIME_MAX;
1381 unsigned long flags;
1382
1383 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1384
1385 if (!__hrtimer_hres_active(cpu_base))
1386 expires = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_ALL);
1387
1388 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1389
1390 return expires;
1391}
1392
1393/**
1394 * hrtimer_next_event_without - time until next expiry event w/o one timer
1395 * @exclude: timer to exclude
1396 *
1397 * Returns the next expiry time over all timers except for the @exclude one or
1398 * KTIME_MAX if none of them is pending.
1399 */
1400u64 hrtimer_next_event_without(const struct hrtimer *exclude)
1401{
1402 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1403 u64 expires = KTIME_MAX;
1404 unsigned long flags;
1405
1406 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1407
1408 if (__hrtimer_hres_active(cpu_base)) {
1409 unsigned int active;
1410
1411 if (!cpu_base->softirq_activated) {
1412 active = cpu_base->active_bases & HRTIMER_ACTIVE_SOFT;
1413 expires = __hrtimer_next_event_base(cpu_base, exclude,
1414 active, KTIME_MAX);
1415 }
1416 active = cpu_base->active_bases & HRTIMER_ACTIVE_HARD;
1417 expires = __hrtimer_next_event_base(cpu_base, exclude, active,
1418 expires);
1419 }
1420
1421 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1422
1423 return expires;
1424}
1425#endif
1426
1427static inline int hrtimer_clockid_to_base(clockid_t clock_id)
1428{
1429 if (likely(clock_id < MAX_CLOCKS)) {
1430 int base = hrtimer_clock_to_base_table[clock_id];
1431
1432 if (likely(base != HRTIMER_MAX_CLOCK_BASES))
1433 return base;
1434 }
1435 WARN(1, "Invalid clockid %d. Using MONOTONIC\n", clock_id);
1436 return HRTIMER_BASE_MONOTONIC;
1437}
1438
1439static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1440 enum hrtimer_mode mode)
1441{
1442 bool softtimer = !!(mode & HRTIMER_MODE_SOFT);
1443 struct hrtimer_cpu_base *cpu_base;
1444 int base;
1445
1446 /*
1447 * On PREEMPT_RT enabled kernels hrtimers which are not explicitly
1448 * marked for hard interrupt expiry mode are moved into soft
1449 * interrupt context for latency reasons and because the callbacks
1450 * can invoke functions which might sleep on RT, e.g. spin_lock().
1451 */
1452 if (IS_ENABLED(CONFIG_PREEMPT_RT) && !(mode & HRTIMER_MODE_HARD))
1453 softtimer = true;
1454
1455 memset(timer, 0, sizeof(struct hrtimer));
1456
1457 cpu_base = raw_cpu_ptr(&hrtimer_bases);
1458
1459 /*
1460 * POSIX magic: Relative CLOCK_REALTIME timers are not affected by
1461 * clock modifications, so they needs to become CLOCK_MONOTONIC to
1462 * ensure POSIX compliance.
1463 */
1464 if (clock_id == CLOCK_REALTIME && mode & HRTIMER_MODE_REL)
1465 clock_id = CLOCK_MONOTONIC;
1466
1467 base = softtimer ? HRTIMER_MAX_CLOCK_BASES / 2 : 0;
1468 base += hrtimer_clockid_to_base(clock_id);
1469 timer->is_soft = softtimer;
1470 timer->is_hard = !!(mode & HRTIMER_MODE_HARD);
1471 timer->base = &cpu_base->clock_base[base];
1472 timerqueue_init(&timer->node);
1473}
1474
1475/**
1476 * hrtimer_init - initialize a timer to the given clock
1477 * @timer: the timer to be initialized
1478 * @clock_id: the clock to be used
1479 * @mode: The modes which are relevant for initialization:
1480 * HRTIMER_MODE_ABS, HRTIMER_MODE_REL, HRTIMER_MODE_ABS_SOFT,
1481 * HRTIMER_MODE_REL_SOFT
1482 *
1483 * The PINNED variants of the above can be handed in,
1484 * but the PINNED bit is ignored as pinning happens
1485 * when the hrtimer is started
1486 */
1487void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1488 enum hrtimer_mode mode)
1489{
1490 debug_init(timer, clock_id, mode);
1491 __hrtimer_init(timer, clock_id, mode);
1492}
1493EXPORT_SYMBOL_GPL(hrtimer_init);
1494
1495/*
1496 * A timer is active, when it is enqueued into the rbtree or the
1497 * callback function is running or it's in the state of being migrated
1498 * to another cpu.
1499 *
1500 * It is important for this function to not return a false negative.
1501 */
1502bool hrtimer_active(const struct hrtimer *timer)
1503{
1504 struct hrtimer_clock_base *base;
1505 unsigned int seq;
1506
1507 do {
1508 base = READ_ONCE(timer->base);
1509 seq = raw_read_seqcount_begin(&base->seq);
1510
1511 if (timer->state != HRTIMER_STATE_INACTIVE ||
1512 base->running == timer)
1513 return true;
1514
1515 } while (read_seqcount_retry(&base->seq, seq) ||
1516 base != READ_ONCE(timer->base));
1517
1518 return false;
1519}
1520EXPORT_SYMBOL_GPL(hrtimer_active);
1521
1522/*
1523 * The write_seqcount_barrier()s in __run_hrtimer() split the thing into 3
1524 * distinct sections:
1525 *
1526 * - queued: the timer is queued
1527 * - callback: the timer is being ran
1528 * - post: the timer is inactive or (re)queued
1529 *
1530 * On the read side we ensure we observe timer->state and cpu_base->running
1531 * from the same section, if anything changed while we looked at it, we retry.
1532 * This includes timer->base changing because sequence numbers alone are
1533 * insufficient for that.
1534 *
1535 * The sequence numbers are required because otherwise we could still observe
1536 * a false negative if the read side got smeared over multiple consecutive
1537 * __run_hrtimer() invocations.
1538 */
1539
1540static void __run_hrtimer(struct hrtimer_cpu_base *cpu_base,
1541 struct hrtimer_clock_base *base,
1542 struct hrtimer *timer, ktime_t *now,
1543 unsigned long flags) __must_hold(&cpu_base->lock)
1544{
1545 enum hrtimer_restart (*fn)(struct hrtimer *);
1546 bool expires_in_hardirq;
1547 int restart;
1548
1549 lockdep_assert_held(&cpu_base->lock);
1550
1551 debug_deactivate(timer);
1552 base->running = timer;
1553
1554 /*
1555 * Separate the ->running assignment from the ->state assignment.
1556 *
1557 * As with a regular write barrier, this ensures the read side in
1558 * hrtimer_active() cannot observe base->running == NULL &&
1559 * timer->state == INACTIVE.
1560 */
1561 raw_write_seqcount_barrier(&base->seq);
1562
1563 __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE, 0);
1564 fn = timer->function;
1565
1566 /*
1567 * Clear the 'is relative' flag for the TIME_LOW_RES case. If the
1568 * timer is restarted with a period then it becomes an absolute
1569 * timer. If its not restarted it does not matter.
1570 */
1571 if (IS_ENABLED(CONFIG_TIME_LOW_RES))
1572 timer->is_rel = false;
1573
1574 /*
1575 * The timer is marked as running in the CPU base, so it is
1576 * protected against migration to a different CPU even if the lock
1577 * is dropped.
1578 */
1579 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1580 trace_hrtimer_expire_entry(timer, now);
1581 expires_in_hardirq = lockdep_hrtimer_enter(timer);
1582
1583 restart = fn(timer);
1584
1585 lockdep_hrtimer_exit(expires_in_hardirq);
1586 trace_hrtimer_expire_exit(timer);
1587 raw_spin_lock_irq(&cpu_base->lock);
1588
1589 /*
1590 * Note: We clear the running state after enqueue_hrtimer and
1591 * we do not reprogram the event hardware. Happens either in
1592 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1593 *
1594 * Note: Because we dropped the cpu_base->lock above,
1595 * hrtimer_start_range_ns() can have popped in and enqueued the timer
1596 * for us already.
1597 */
1598 if (restart != HRTIMER_NORESTART &&
1599 !(timer->state & HRTIMER_STATE_ENQUEUED))
1600 enqueue_hrtimer(timer, base, HRTIMER_MODE_ABS);
1601
1602 /*
1603 * Separate the ->running assignment from the ->state assignment.
1604 *
1605 * As with a regular write barrier, this ensures the read side in
1606 * hrtimer_active() cannot observe base->running.timer == NULL &&
1607 * timer->state == INACTIVE.
1608 */
1609 raw_write_seqcount_barrier(&base->seq);
1610
1611 WARN_ON_ONCE(base->running != timer);
1612 base->running = NULL;
1613}
1614
1615static void __hrtimer_run_queues(struct hrtimer_cpu_base *cpu_base, ktime_t now,
1616 unsigned long flags, unsigned int active_mask)
1617{
1618 struct hrtimer_clock_base *base;
1619 unsigned int active = cpu_base->active_bases & active_mask;
1620
1621 for_each_active_base(base, cpu_base, active) {
1622 struct timerqueue_node *node;
1623 ktime_t basenow;
1624
1625 basenow = ktime_add(now, base->offset);
1626
1627 while ((node = timerqueue_getnext(&base->active))) {
1628 struct hrtimer *timer;
1629
1630 timer = container_of(node, struct hrtimer, node);
1631
1632 /*
1633 * The immediate goal for using the softexpires is
1634 * minimizing wakeups, not running timers at the
1635 * earliest interrupt after their soft expiration.
1636 * This allows us to avoid using a Priority Search
1637 * Tree, which can answer a stabbing query for
1638 * overlapping intervals and instead use the simple
1639 * BST we already have.
1640 * We don't add extra wakeups by delaying timers that
1641 * are right-of a not yet expired timer, because that
1642 * timer will have to trigger a wakeup anyway.
1643 */
1644 if (basenow < hrtimer_get_softexpires_tv64(timer))
1645 break;
1646
1647 __run_hrtimer(cpu_base, base, timer, &basenow, flags);
1648 if (active_mask == HRTIMER_ACTIVE_SOFT)
1649 hrtimer_sync_wait_running(cpu_base, flags);
1650 }
1651 }
1652}
1653
1654static __latent_entropy void hrtimer_run_softirq(struct softirq_action *h)
1655{
1656 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1657 unsigned long flags;
1658 ktime_t now;
1659
1660 hrtimer_cpu_base_lock_expiry(cpu_base);
1661 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1662
1663 now = hrtimer_update_base(cpu_base);
1664 __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_SOFT);
1665
1666 cpu_base->softirq_activated = 0;
1667 hrtimer_update_softirq_timer(cpu_base, true);
1668
1669 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1670 hrtimer_cpu_base_unlock_expiry(cpu_base);
1671}
1672
1673#ifdef CONFIG_HIGH_RES_TIMERS
1674
1675/*
1676 * High resolution timer interrupt
1677 * Called with interrupts disabled
1678 */
1679void hrtimer_interrupt(struct clock_event_device *dev)
1680{
1681 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1682 ktime_t expires_next, now, entry_time, delta;
1683 unsigned long flags;
1684 int retries = 0;
1685
1686 BUG_ON(!cpu_base->hres_active);
1687 cpu_base->nr_events++;
1688 dev->next_event = KTIME_MAX;
1689
1690 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1691 entry_time = now = hrtimer_update_base(cpu_base);
1692retry:
1693 cpu_base->in_hrtirq = 1;
1694 /*
1695 * We set expires_next to KTIME_MAX here with cpu_base->lock
1696 * held to prevent that a timer is enqueued in our queue via
1697 * the migration code. This does not affect enqueueing of
1698 * timers which run their callback and need to be requeued on
1699 * this CPU.
1700 */
1701 cpu_base->expires_next = KTIME_MAX;
1702
1703 if (!ktime_before(now, cpu_base->softirq_expires_next)) {
1704 cpu_base->softirq_expires_next = KTIME_MAX;
1705 cpu_base->softirq_activated = 1;
1706 raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1707 }
1708
1709 __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_HARD);
1710
1711 /* Reevaluate the clock bases for the [soft] next expiry */
1712 expires_next = hrtimer_update_next_event(cpu_base);
1713 /*
1714 * Store the new expiry value so the migration code can verify
1715 * against it.
1716 */
1717 cpu_base->expires_next = expires_next;
1718 cpu_base->in_hrtirq = 0;
1719 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1720
1721 /* Reprogramming necessary ? */
1722 if (!tick_program_event(expires_next, 0)) {
1723 cpu_base->hang_detected = 0;
1724 return;
1725 }
1726
1727 /*
1728 * The next timer was already expired due to:
1729 * - tracing
1730 * - long lasting callbacks
1731 * - being scheduled away when running in a VM
1732 *
1733 * We need to prevent that we loop forever in the hrtimer
1734 * interrupt routine. We give it 3 attempts to avoid
1735 * overreacting on some spurious event.
1736 *
1737 * Acquire base lock for updating the offsets and retrieving
1738 * the current time.
1739 */
1740 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1741 now = hrtimer_update_base(cpu_base);
1742 cpu_base->nr_retries++;
1743 if (++retries < 3)
1744 goto retry;
1745 /*
1746 * Give the system a chance to do something else than looping
1747 * here. We stored the entry time, so we know exactly how long
1748 * we spent here. We schedule the next event this amount of
1749 * time away.
1750 */
1751 cpu_base->nr_hangs++;
1752 cpu_base->hang_detected = 1;
1753 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1754
1755 delta = ktime_sub(now, entry_time);
1756 if ((unsigned int)delta > cpu_base->max_hang_time)
1757 cpu_base->max_hang_time = (unsigned int) delta;
1758 /*
1759 * Limit it to a sensible value as we enforce a longer
1760 * delay. Give the CPU at least 100ms to catch up.
1761 */
1762 if (delta > 100 * NSEC_PER_MSEC)
1763 expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1764 else
1765 expires_next = ktime_add(now, delta);
1766 tick_program_event(expires_next, 1);
1767 pr_warn_once("hrtimer: interrupt took %llu ns\n", ktime_to_ns(delta));
1768}
1769
1770/* called with interrupts disabled */
1771static inline void __hrtimer_peek_ahead_timers(void)
1772{
1773 struct tick_device *td;
1774
1775 if (!hrtimer_hres_active())
1776 return;
1777
1778 td = this_cpu_ptr(&tick_cpu_device);
1779 if (td && td->evtdev)
1780 hrtimer_interrupt(td->evtdev);
1781}
1782
1783#else /* CONFIG_HIGH_RES_TIMERS */
1784
1785static inline void __hrtimer_peek_ahead_timers(void) { }
1786
1787#endif /* !CONFIG_HIGH_RES_TIMERS */
1788
1789/*
1790 * Called from run_local_timers in hardirq context every jiffy
1791 */
1792void hrtimer_run_queues(void)
1793{
1794 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1795 unsigned long flags;
1796 ktime_t now;
1797
1798 if (__hrtimer_hres_active(cpu_base))
1799 return;
1800
1801 /*
1802 * This _is_ ugly: We have to check periodically, whether we
1803 * can switch to highres and / or nohz mode. The clocksource
1804 * switch happens with xtime_lock held. Notification from
1805 * there only sets the check bit in the tick_oneshot code,
1806 * otherwise we might deadlock vs. xtime_lock.
1807 */
1808 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) {
1809 hrtimer_switch_to_hres();
1810 return;
1811 }
1812
1813 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1814 now = hrtimer_update_base(cpu_base);
1815
1816 if (!ktime_before(now, cpu_base->softirq_expires_next)) {
1817 cpu_base->softirq_expires_next = KTIME_MAX;
1818 cpu_base->softirq_activated = 1;
1819 raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1820 }
1821
1822 __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_HARD);
1823 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1824}
1825
1826/*
1827 * Sleep related functions:
1828 */
1829static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1830{
1831 struct hrtimer_sleeper *t =
1832 container_of(timer, struct hrtimer_sleeper, timer);
1833 struct task_struct *task = t->task;
1834
1835 t->task = NULL;
1836 if (task)
1837 wake_up_process(task);
1838
1839 return HRTIMER_NORESTART;
1840}
1841
1842/**
1843 * hrtimer_sleeper_start_expires - Start a hrtimer sleeper timer
1844 * @sl: sleeper to be started
1845 * @mode: timer mode abs/rel
1846 *
1847 * Wrapper around hrtimer_start_expires() for hrtimer_sleeper based timers
1848 * to allow PREEMPT_RT to tweak the delivery mode (soft/hardirq context)
1849 */
1850void hrtimer_sleeper_start_expires(struct hrtimer_sleeper *sl,
1851 enum hrtimer_mode mode)
1852{
1853 /*
1854 * Make the enqueue delivery mode check work on RT. If the sleeper
1855 * was initialized for hard interrupt delivery, force the mode bit.
1856 * This is a special case for hrtimer_sleepers because
1857 * hrtimer_init_sleeper() determines the delivery mode on RT so the
1858 * fiddling with this decision is avoided at the call sites.
1859 */
1860 if (IS_ENABLED(CONFIG_PREEMPT_RT) && sl->timer.is_hard)
1861 mode |= HRTIMER_MODE_HARD;
1862
1863 hrtimer_start_expires(&sl->timer, mode);
1864}
1865EXPORT_SYMBOL_GPL(hrtimer_sleeper_start_expires);
1866
1867static void __hrtimer_init_sleeper(struct hrtimer_sleeper *sl,
1868 clockid_t clock_id, enum hrtimer_mode mode)
1869{
1870 /*
1871 * On PREEMPT_RT enabled kernels hrtimers which are not explicitly
1872 * marked for hard interrupt expiry mode are moved into soft
1873 * interrupt context either for latency reasons or because the
1874 * hrtimer callback takes regular spinlocks or invokes other
1875 * functions which are not suitable for hard interrupt context on
1876 * PREEMPT_RT.
1877 *
1878 * The hrtimer_sleeper callback is RT compatible in hard interrupt
1879 * context, but there is a latency concern: Untrusted userspace can
1880 * spawn many threads which arm timers for the same expiry time on
1881 * the same CPU. That causes a latency spike due to the wakeup of
1882 * a gazillion threads.
1883 *
1884 * OTOH, privileged real-time user space applications rely on the
1885 * low latency of hard interrupt wakeups. If the current task is in
1886 * a real-time scheduling class, mark the mode for hard interrupt
1887 * expiry.
1888 */
1889 if (IS_ENABLED(CONFIG_PREEMPT_RT)) {
1890 if (task_is_realtime(current) && !(mode & HRTIMER_MODE_SOFT))
1891 mode |= HRTIMER_MODE_HARD;
1892 }
1893
1894 __hrtimer_init(&sl->timer, clock_id, mode);
1895 sl->timer.function = hrtimer_wakeup;
1896 sl->task = current;
1897}
1898
1899/**
1900 * hrtimer_init_sleeper - initialize sleeper to the given clock
1901 * @sl: sleeper to be initialized
1902 * @clock_id: the clock to be used
1903 * @mode: timer mode abs/rel
1904 */
1905void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, clockid_t clock_id,
1906 enum hrtimer_mode mode)
1907{
1908 debug_init(&sl->timer, clock_id, mode);
1909 __hrtimer_init_sleeper(sl, clock_id, mode);
1910
1911}
1912EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
1913
1914int nanosleep_copyout(struct restart_block *restart, struct timespec64 *ts)
1915{
1916 switch(restart->nanosleep.type) {
1917#ifdef CONFIG_COMPAT_32BIT_TIME
1918 case TT_COMPAT:
1919 if (put_old_timespec32(ts, restart->nanosleep.compat_rmtp))
1920 return -EFAULT;
1921 break;
1922#endif
1923 case TT_NATIVE:
1924 if (put_timespec64(ts, restart->nanosleep.rmtp))
1925 return -EFAULT;
1926 break;
1927 default:
1928 BUG();
1929 }
1930 return -ERESTART_RESTARTBLOCK;
1931}
1932
1933static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1934{
1935 struct restart_block *restart;
1936
1937 do {
1938 set_current_state(TASK_INTERRUPTIBLE);
1939 hrtimer_sleeper_start_expires(t, mode);
1940
1941 if (likely(t->task))
1942 freezable_schedule();
1943
1944 hrtimer_cancel(&t->timer);
1945 mode = HRTIMER_MODE_ABS;
1946
1947 } while (t->task && !signal_pending(current));
1948
1949 __set_current_state(TASK_RUNNING);
1950
1951 if (!t->task)
1952 return 0;
1953
1954 restart = ¤t->restart_block;
1955 if (restart->nanosleep.type != TT_NONE) {
1956 ktime_t rem = hrtimer_expires_remaining(&t->timer);
1957 struct timespec64 rmt;
1958
1959 if (rem <= 0)
1960 return 0;
1961 rmt = ktime_to_timespec64(rem);
1962
1963 return nanosleep_copyout(restart, &rmt);
1964 }
1965 return -ERESTART_RESTARTBLOCK;
1966}
1967
1968static long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1969{
1970 struct hrtimer_sleeper t;
1971 int ret;
1972
1973 hrtimer_init_sleeper_on_stack(&t, restart->nanosleep.clockid,
1974 HRTIMER_MODE_ABS);
1975 hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
1976 ret = do_nanosleep(&t, HRTIMER_MODE_ABS);
1977 destroy_hrtimer_on_stack(&t.timer);
1978 return ret;
1979}
1980
1981long hrtimer_nanosleep(ktime_t rqtp, const enum hrtimer_mode mode,
1982 const clockid_t clockid)
1983{
1984 struct restart_block *restart;
1985 struct hrtimer_sleeper t;
1986 int ret = 0;
1987 u64 slack;
1988
1989 slack = current->timer_slack_ns;
1990 if (dl_task(current) || rt_task(current))
1991 slack = 0;
1992
1993 hrtimer_init_sleeper_on_stack(&t, clockid, mode);
1994 hrtimer_set_expires_range_ns(&t.timer, rqtp, slack);
1995 ret = do_nanosleep(&t, mode);
1996 if (ret != -ERESTART_RESTARTBLOCK)
1997 goto out;
1998
1999 /* Absolute timers do not update the rmtp value and restart: */
2000 if (mode == HRTIMER_MODE_ABS) {
2001 ret = -ERESTARTNOHAND;
2002 goto out;
2003 }
2004
2005 restart = ¤t->restart_block;
2006 restart->nanosleep.clockid = t.timer.base->clockid;
2007 restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
2008 set_restart_fn(restart, hrtimer_nanosleep_restart);
2009out:
2010 destroy_hrtimer_on_stack(&t.timer);
2011 return ret;
2012}
2013
2014#ifdef CONFIG_64BIT
2015
2016SYSCALL_DEFINE2(nanosleep, struct __kernel_timespec __user *, rqtp,
2017 struct __kernel_timespec __user *, rmtp)
2018{
2019 struct timespec64 tu;
2020
2021 if (get_timespec64(&tu, rqtp))
2022 return -EFAULT;
2023
2024 if (!timespec64_valid(&tu))
2025 return -EINVAL;
2026
2027 current->restart_block.nanosleep.type = rmtp ? TT_NATIVE : TT_NONE;
2028 current->restart_block.nanosleep.rmtp = rmtp;
2029 return hrtimer_nanosleep(timespec64_to_ktime(tu), HRTIMER_MODE_REL,
2030 CLOCK_MONOTONIC);
2031}
2032
2033#endif
2034
2035#ifdef CONFIG_COMPAT_32BIT_TIME
2036
2037SYSCALL_DEFINE2(nanosleep_time32, struct old_timespec32 __user *, rqtp,
2038 struct old_timespec32 __user *, rmtp)
2039{
2040 struct timespec64 tu;
2041
2042 if (get_old_timespec32(&tu, rqtp))
2043 return -EFAULT;
2044
2045 if (!timespec64_valid(&tu))
2046 return -EINVAL;
2047
2048 current->restart_block.nanosleep.type = rmtp ? TT_COMPAT : TT_NONE;
2049 current->restart_block.nanosleep.compat_rmtp = rmtp;
2050 return hrtimer_nanosleep(timespec64_to_ktime(tu), HRTIMER_MODE_REL,
2051 CLOCK_MONOTONIC);
2052}
2053#endif
2054
2055/*
2056 * Functions related to boot-time initialization:
2057 */
2058int hrtimers_prepare_cpu(unsigned int cpu)
2059{
2060 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
2061 int i;
2062
2063 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
2064 struct hrtimer_clock_base *clock_b = &cpu_base->clock_base[i];
2065
2066 clock_b->cpu_base = cpu_base;
2067 seqcount_raw_spinlock_init(&clock_b->seq, &cpu_base->lock);
2068 timerqueue_init_head(&clock_b->active);
2069 }
2070
2071 cpu_base->cpu = cpu;
2072 cpu_base->active_bases = 0;
2073 cpu_base->hres_active = 0;
2074 cpu_base->hang_detected = 0;
2075 cpu_base->next_timer = NULL;
2076 cpu_base->softirq_next_timer = NULL;
2077 cpu_base->expires_next = KTIME_MAX;
2078 cpu_base->softirq_expires_next = KTIME_MAX;
2079 hrtimer_cpu_base_init_expiry_lock(cpu_base);
2080 return 0;
2081}
2082
2083#ifdef CONFIG_HOTPLUG_CPU
2084
2085static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
2086 struct hrtimer_clock_base *new_base)
2087{
2088 struct hrtimer *timer;
2089 struct timerqueue_node *node;
2090
2091 while ((node = timerqueue_getnext(&old_base->active))) {
2092 timer = container_of(node, struct hrtimer, node);
2093 BUG_ON(hrtimer_callback_running(timer));
2094 debug_deactivate(timer);
2095
2096 /*
2097 * Mark it as ENQUEUED not INACTIVE otherwise the
2098 * timer could be seen as !active and just vanish away
2099 * under us on another CPU
2100 */
2101 __remove_hrtimer(timer, old_base, HRTIMER_STATE_ENQUEUED, 0);
2102 timer->base = new_base;
2103 /*
2104 * Enqueue the timers on the new cpu. This does not
2105 * reprogram the event device in case the timer
2106 * expires before the earliest on this CPU, but we run
2107 * hrtimer_interrupt after we migrated everything to
2108 * sort out already expired timers and reprogram the
2109 * event device.
2110 */
2111 enqueue_hrtimer(timer, new_base, HRTIMER_MODE_ABS);
2112 }
2113}
2114
2115int hrtimers_dead_cpu(unsigned int scpu)
2116{
2117 struct hrtimer_cpu_base *old_base, *new_base;
2118 int i;
2119
2120 BUG_ON(cpu_online(scpu));
2121 tick_cancel_sched_timer(scpu);
2122
2123 /*
2124 * this BH disable ensures that raise_softirq_irqoff() does
2125 * not wakeup ksoftirqd (and acquire the pi-lock) while
2126 * holding the cpu_base lock
2127 */
2128 local_bh_disable();
2129 local_irq_disable();
2130 old_base = &per_cpu(hrtimer_bases, scpu);
2131 new_base = this_cpu_ptr(&hrtimer_bases);
2132 /*
2133 * The caller is globally serialized and nobody else
2134 * takes two locks at once, deadlock is not possible.
2135 */
2136 raw_spin_lock(&new_base->lock);
2137 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
2138
2139 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
2140 migrate_hrtimer_list(&old_base->clock_base[i],
2141 &new_base->clock_base[i]);
2142 }
2143
2144 /*
2145 * The migration might have changed the first expiring softirq
2146 * timer on this CPU. Update it.
2147 */
2148 hrtimer_update_softirq_timer(new_base, false);
2149
2150 raw_spin_unlock(&old_base->lock);
2151 raw_spin_unlock(&new_base->lock);
2152
2153 /* Check, if we got expired work to do */
2154 __hrtimer_peek_ahead_timers();
2155 local_irq_enable();
2156 local_bh_enable();
2157 return 0;
2158}
2159
2160#endif /* CONFIG_HOTPLUG_CPU */
2161
2162void __init hrtimers_init(void)
2163{
2164 hrtimers_prepare_cpu(smp_processor_id());
2165 open_softirq(HRTIMER_SOFTIRQ, hrtimer_run_softirq);
2166}
2167
2168/**
2169 * schedule_hrtimeout_range_clock - sleep until timeout
2170 * @expires: timeout value (ktime_t)
2171 * @delta: slack in expires timeout (ktime_t)
2172 * @mode: timer mode
2173 * @clock_id: timer clock to be used
2174 */
2175int __sched
2176schedule_hrtimeout_range_clock(ktime_t *expires, u64 delta,
2177 const enum hrtimer_mode mode, clockid_t clock_id)
2178{
2179 struct hrtimer_sleeper t;
2180
2181 /*
2182 * Optimize when a zero timeout value is given. It does not
2183 * matter whether this is an absolute or a relative time.
2184 */
2185 if (expires && *expires == 0) {
2186 __set_current_state(TASK_RUNNING);
2187 return 0;
2188 }
2189
2190 /*
2191 * A NULL parameter means "infinite"
2192 */
2193 if (!expires) {
2194 schedule();
2195 return -EINTR;
2196 }
2197
2198 hrtimer_init_sleeper_on_stack(&t, clock_id, mode);
2199 hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
2200 hrtimer_sleeper_start_expires(&t, mode);
2201
2202 if (likely(t.task))
2203 schedule();
2204
2205 hrtimer_cancel(&t.timer);
2206 destroy_hrtimer_on_stack(&t.timer);
2207
2208 __set_current_state(TASK_RUNNING);
2209
2210 return !t.task ? 0 : -EINTR;
2211}
2212
2213/**
2214 * schedule_hrtimeout_range - sleep until timeout
2215 * @expires: timeout value (ktime_t)
2216 * @delta: slack in expires timeout (ktime_t)
2217 * @mode: timer mode
2218 *
2219 * Make the current task sleep until the given expiry time has
2220 * elapsed. The routine will return immediately unless
2221 * the current task state has been set (see set_current_state()).
2222 *
2223 * The @delta argument gives the kernel the freedom to schedule the
2224 * actual wakeup to a time that is both power and performance friendly.
2225 * The kernel give the normal best effort behavior for "@expires+@delta",
2226 * but may decide to fire the timer earlier, but no earlier than @expires.
2227 *
2228 * You can set the task state as follows -
2229 *
2230 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
2231 * pass before the routine returns unless the current task is explicitly
2232 * woken up, (e.g. by wake_up_process()).
2233 *
2234 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
2235 * delivered to the current task or the current task is explicitly woken
2236 * up.
2237 *
2238 * The current task state is guaranteed to be TASK_RUNNING when this
2239 * routine returns.
2240 *
2241 * Returns 0 when the timer has expired. If the task was woken before the
2242 * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or
2243 * by an explicit wakeup, it returns -EINTR.
2244 */
2245int __sched schedule_hrtimeout_range(ktime_t *expires, u64 delta,
2246 const enum hrtimer_mode mode)
2247{
2248 return schedule_hrtimeout_range_clock(expires, delta, mode,
2249 CLOCK_MONOTONIC);
2250}
2251EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
2252
2253/**
2254 * schedule_hrtimeout - sleep until timeout
2255 * @expires: timeout value (ktime_t)
2256 * @mode: timer mode
2257 *
2258 * Make the current task sleep until the given expiry time has
2259 * elapsed. The routine will return immediately unless
2260 * the current task state has been set (see set_current_state()).
2261 *
2262 * You can set the task state as follows -
2263 *
2264 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
2265 * pass before the routine returns unless the current task is explicitly
2266 * woken up, (e.g. by wake_up_process()).
2267 *
2268 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
2269 * delivered to the current task or the current task is explicitly woken
2270 * up.
2271 *
2272 * The current task state is guaranteed to be TASK_RUNNING when this
2273 * routine returns.
2274 *
2275 * Returns 0 when the timer has expired. If the task was woken before the
2276 * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or
2277 * by an explicit wakeup, it returns -EINTR.
2278 */
2279int __sched schedule_hrtimeout(ktime_t *expires,
2280 const enum hrtimer_mode mode)
2281{
2282 return schedule_hrtimeout_range(expires, 0, mode);
2283}
2284EXPORT_SYMBOL_GPL(schedule_hrtimeout);