Loading...
Note: File does not exist in v3.15.
1/*
2 * linux/kernel/hrtimer.c
3 *
4 * Copyright(C) 2005-2006, Thomas Gleixner <tglx@linutronix.de>
5 * Copyright(C) 2005-2007, Red Hat, Inc., Ingo Molnar
6 * Copyright(C) 2006-2007 Timesys Corp., Thomas Gleixner
7 *
8 * High-resolution kernel timers
9 *
10 * In contrast to the low-resolution timeout API implemented in
11 * kernel/timer.c, hrtimers provide finer resolution and accuracy
12 * depending on system configuration and capabilities.
13 *
14 * These timers are currently used for:
15 * - itimers
16 * - POSIX timers
17 * - nanosleep
18 * - precise in-kernel timing
19 *
20 * Started by: Thomas Gleixner and Ingo Molnar
21 *
22 * Credits:
23 * based on kernel/timer.c
24 *
25 * Help, testing, suggestions, bugfixes, improvements were
26 * provided by:
27 *
28 * George Anzinger, Andrew Morton, Steven Rostedt, Roman Zippel
29 * et. al.
30 *
31 * For licencing details see kernel-base/COPYING
32 */
33
34#include <linux/cpu.h>
35#include <linux/export.h>
36#include <linux/percpu.h>
37#include <linux/hrtimer.h>
38#include <linux/notifier.h>
39#include <linux/syscalls.h>
40#include <linux/interrupt.h>
41#include <linux/tick.h>
42#include <linux/seq_file.h>
43#include <linux/err.h>
44#include <linux/debugobjects.h>
45#include <linux/sched/signal.h>
46#include <linux/sched/sysctl.h>
47#include <linux/sched/rt.h>
48#include <linux/sched/deadline.h>
49#include <linux/sched/nohz.h>
50#include <linux/sched/debug.h>
51#include <linux/timer.h>
52#include <linux/freezer.h>
53#include <linux/compat.h>
54
55#include <linux/uaccess.h>
56
57#include <trace/events/timer.h>
58
59#include "tick-internal.h"
60
61/*
62 * Masks for selecting the soft and hard context timers from
63 * cpu_base->active
64 */
65#define MASK_SHIFT (HRTIMER_BASE_MONOTONIC_SOFT)
66#define HRTIMER_ACTIVE_HARD ((1U << MASK_SHIFT) - 1)
67#define HRTIMER_ACTIVE_SOFT (HRTIMER_ACTIVE_HARD << MASK_SHIFT)
68#define HRTIMER_ACTIVE_ALL (HRTIMER_ACTIVE_SOFT | HRTIMER_ACTIVE_HARD)
69
70/*
71 * The timer bases:
72 *
73 * There are more clockids than hrtimer bases. Thus, we index
74 * into the timer bases by the hrtimer_base_type enum. When trying
75 * to reach a base using a clockid, hrtimer_clockid_to_base()
76 * is used to convert from clockid to the proper hrtimer_base_type.
77 */
78DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
79{
80 .lock = __RAW_SPIN_LOCK_UNLOCKED(hrtimer_bases.lock),
81 .clock_base =
82 {
83 {
84 .index = HRTIMER_BASE_MONOTONIC,
85 .clockid = CLOCK_MONOTONIC,
86 .get_time = &ktime_get,
87 },
88 {
89 .index = HRTIMER_BASE_REALTIME,
90 .clockid = CLOCK_REALTIME,
91 .get_time = &ktime_get_real,
92 },
93 {
94 .index = HRTIMER_BASE_BOOTTIME,
95 .clockid = CLOCK_BOOTTIME,
96 .get_time = &ktime_get_boottime,
97 },
98 {
99 .index = HRTIMER_BASE_TAI,
100 .clockid = CLOCK_TAI,
101 .get_time = &ktime_get_clocktai,
102 },
103 {
104 .index = HRTIMER_BASE_MONOTONIC_SOFT,
105 .clockid = CLOCK_MONOTONIC,
106 .get_time = &ktime_get,
107 },
108 {
109 .index = HRTIMER_BASE_REALTIME_SOFT,
110 .clockid = CLOCK_REALTIME,
111 .get_time = &ktime_get_real,
112 },
113 {
114 .index = HRTIMER_BASE_BOOTTIME_SOFT,
115 .clockid = CLOCK_BOOTTIME,
116 .get_time = &ktime_get_boottime,
117 },
118 {
119 .index = HRTIMER_BASE_TAI_SOFT,
120 .clockid = CLOCK_TAI,
121 .get_time = &ktime_get_clocktai,
122 },
123 }
124};
125
126static const int hrtimer_clock_to_base_table[MAX_CLOCKS] = {
127 /* Make sure we catch unsupported clockids */
128 [0 ... MAX_CLOCKS - 1] = HRTIMER_MAX_CLOCK_BASES,
129
130 [CLOCK_REALTIME] = HRTIMER_BASE_REALTIME,
131 [CLOCK_MONOTONIC] = HRTIMER_BASE_MONOTONIC,
132 [CLOCK_BOOTTIME] = HRTIMER_BASE_BOOTTIME,
133 [CLOCK_TAI] = HRTIMER_BASE_TAI,
134};
135
136/*
137 * Functions and macros which are different for UP/SMP systems are kept in a
138 * single place
139 */
140#ifdef CONFIG_SMP
141
142/*
143 * We require the migration_base for lock_hrtimer_base()/switch_hrtimer_base()
144 * such that hrtimer_callback_running() can unconditionally dereference
145 * timer->base->cpu_base
146 */
147static struct hrtimer_cpu_base migration_cpu_base = {
148 .clock_base = { { .cpu_base = &migration_cpu_base, }, },
149};
150
151#define migration_base migration_cpu_base.clock_base[0]
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{
169 struct hrtimer_clock_base *base;
170
171 for (;;) {
172 base = timer->base;
173 if (likely(base != &migration_base)) {
174 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
175 if (likely(base == timer->base))
176 return base;
177 /* The timer has migrated to another CPU: */
178 raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
179 }
180 cpu_relax();
181 }
182}
183
184/*
185 * We do not migrate the timer when it is expiring before the next
186 * event on the target cpu. When high resolution is enabled, we cannot
187 * reprogram the target cpu hardware and we would cause it to fire
188 * late. To keep it simple, we handle the high resolution enabled and
189 * disabled case similar.
190 *
191 * Called with cpu_base->lock of target cpu held.
192 */
193static int
194hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base)
195{
196 ktime_t expires;
197
198 expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset);
199 return expires < new_base->cpu_base->expires_next;
200}
201
202static inline
203struct hrtimer_cpu_base *get_target_base(struct hrtimer_cpu_base *base,
204 int pinned)
205{
206#if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
207 if (static_branch_likely(&timers_migration_enabled) && !pinned)
208 return &per_cpu(hrtimer_bases, get_nohz_timer_target());
209#endif
210 return base;
211}
212
213/*
214 * We switch the timer base to a power-optimized selected CPU target,
215 * if:
216 * - NO_HZ_COMMON is enabled
217 * - timer migration is enabled
218 * - the timer callback is not running
219 * - the timer is not the first expiring timer on the new target
220 *
221 * If one of the above requirements is not fulfilled we move the timer
222 * to the current CPU or leave it on the previously assigned CPU if
223 * the timer callback is currently running.
224 */
225static inline struct hrtimer_clock_base *
226switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
227 int pinned)
228{
229 struct hrtimer_cpu_base *new_cpu_base, *this_cpu_base;
230 struct hrtimer_clock_base *new_base;
231 int basenum = base->index;
232
233 this_cpu_base = this_cpu_ptr(&hrtimer_bases);
234 new_cpu_base = get_target_base(this_cpu_base, pinned);
235again:
236 new_base = &new_cpu_base->clock_base[basenum];
237
238 if (base != new_base) {
239 /*
240 * We are trying to move timer to new_base.
241 * However we can't change timer's base while it is running,
242 * so we keep it on the same CPU. No hassle vs. reprogramming
243 * the event source in the high resolution case. The softirq
244 * code will take care of this when the timer function has
245 * completed. There is no conflict as we hold the lock until
246 * the timer is enqueued.
247 */
248 if (unlikely(hrtimer_callback_running(timer)))
249 return base;
250
251 /* See the comment in lock_hrtimer_base() */
252 timer->base = &migration_base;
253 raw_spin_unlock(&base->cpu_base->lock);
254 raw_spin_lock(&new_base->cpu_base->lock);
255
256 if (new_cpu_base != this_cpu_base &&
257 hrtimer_check_target(timer, new_base)) {
258 raw_spin_unlock(&new_base->cpu_base->lock);
259 raw_spin_lock(&base->cpu_base->lock);
260 new_cpu_base = this_cpu_base;
261 timer->base = base;
262 goto again;
263 }
264 timer->base = new_base;
265 } else {
266 if (new_cpu_base != this_cpu_base &&
267 hrtimer_check_target(timer, new_base)) {
268 new_cpu_base = this_cpu_base;
269 goto again;
270 }
271 }
272 return new_base;
273}
274
275#else /* CONFIG_SMP */
276
277static inline struct hrtimer_clock_base *
278lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
279{
280 struct hrtimer_clock_base *base = timer->base;
281
282 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
283
284 return base;
285}
286
287# define switch_hrtimer_base(t, b, p) (b)
288
289#endif /* !CONFIG_SMP */
290
291/*
292 * Functions for the union type storage format of ktime_t which are
293 * too large for inlining:
294 */
295#if BITS_PER_LONG < 64
296/*
297 * Divide a ktime value by a nanosecond value
298 */
299s64 __ktime_divns(const ktime_t kt, s64 div)
300{
301 int sft = 0;
302 s64 dclc;
303 u64 tmp;
304
305 dclc = ktime_to_ns(kt);
306 tmp = dclc < 0 ? -dclc : dclc;
307
308 /* Make sure the divisor is less than 2^32: */
309 while (div >> 32) {
310 sft++;
311 div >>= 1;
312 }
313 tmp >>= sft;
314 do_div(tmp, (unsigned long) div);
315 return dclc < 0 ? -tmp : tmp;
316}
317EXPORT_SYMBOL_GPL(__ktime_divns);
318#endif /* BITS_PER_LONG >= 64 */
319
320/*
321 * Add two ktime values and do a safety check for overflow:
322 */
323ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
324{
325 ktime_t res = ktime_add_unsafe(lhs, rhs);
326
327 /*
328 * We use KTIME_SEC_MAX here, the maximum timeout which we can
329 * return to user space in a timespec:
330 */
331 if (res < 0 || res < lhs || res < rhs)
332 res = ktime_set(KTIME_SEC_MAX, 0);
333
334 return res;
335}
336
337EXPORT_SYMBOL_GPL(ktime_add_safe);
338
339#ifdef CONFIG_DEBUG_OBJECTS_TIMERS
340
341static struct debug_obj_descr hrtimer_debug_descr;
342
343static void *hrtimer_debug_hint(void *addr)
344{
345 return ((struct hrtimer *) addr)->function;
346}
347
348/*
349 * fixup_init is called when:
350 * - an active object is initialized
351 */
352static bool hrtimer_fixup_init(void *addr, enum debug_obj_state state)
353{
354 struct hrtimer *timer = addr;
355
356 switch (state) {
357 case ODEBUG_STATE_ACTIVE:
358 hrtimer_cancel(timer);
359 debug_object_init(timer, &hrtimer_debug_descr);
360 return true;
361 default:
362 return false;
363 }
364}
365
366/*
367 * fixup_activate is called when:
368 * - an active object is activated
369 * - an unknown non-static object is activated
370 */
371static bool hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
372{
373 switch (state) {
374 case ODEBUG_STATE_ACTIVE:
375 WARN_ON(1);
376
377 default:
378 return false;
379 }
380}
381
382/*
383 * fixup_free is called when:
384 * - an active object is freed
385 */
386static bool hrtimer_fixup_free(void *addr, enum debug_obj_state state)
387{
388 struct hrtimer *timer = addr;
389
390 switch (state) {
391 case ODEBUG_STATE_ACTIVE:
392 hrtimer_cancel(timer);
393 debug_object_free(timer, &hrtimer_debug_descr);
394 return true;
395 default:
396 return false;
397 }
398}
399
400static struct debug_obj_descr hrtimer_debug_descr = {
401 .name = "hrtimer",
402 .debug_hint = hrtimer_debug_hint,
403 .fixup_init = hrtimer_fixup_init,
404 .fixup_activate = hrtimer_fixup_activate,
405 .fixup_free = hrtimer_fixup_free,
406};
407
408static inline void debug_hrtimer_init(struct hrtimer *timer)
409{
410 debug_object_init(timer, &hrtimer_debug_descr);
411}
412
413static inline void debug_hrtimer_activate(struct hrtimer *timer,
414 enum hrtimer_mode mode)
415{
416 debug_object_activate(timer, &hrtimer_debug_descr);
417}
418
419static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
420{
421 debug_object_deactivate(timer, &hrtimer_debug_descr);
422}
423
424static inline void debug_hrtimer_free(struct hrtimer *timer)
425{
426 debug_object_free(timer, &hrtimer_debug_descr);
427}
428
429static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
430 enum hrtimer_mode mode);
431
432void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
433 enum hrtimer_mode mode)
434{
435 debug_object_init_on_stack(timer, &hrtimer_debug_descr);
436 __hrtimer_init(timer, clock_id, mode);
437}
438EXPORT_SYMBOL_GPL(hrtimer_init_on_stack);
439
440void destroy_hrtimer_on_stack(struct hrtimer *timer)
441{
442 debug_object_free(timer, &hrtimer_debug_descr);
443}
444EXPORT_SYMBOL_GPL(destroy_hrtimer_on_stack);
445
446#else
447
448static inline void debug_hrtimer_init(struct hrtimer *timer) { }
449static inline void debug_hrtimer_activate(struct hrtimer *timer,
450 enum hrtimer_mode mode) { }
451static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
452#endif
453
454static inline void
455debug_init(struct hrtimer *timer, clockid_t clockid,
456 enum hrtimer_mode mode)
457{
458 debug_hrtimer_init(timer);
459 trace_hrtimer_init(timer, clockid, mode);
460}
461
462static inline void debug_activate(struct hrtimer *timer,
463 enum hrtimer_mode mode)
464{
465 debug_hrtimer_activate(timer, mode);
466 trace_hrtimer_start(timer, mode);
467}
468
469static inline void debug_deactivate(struct hrtimer *timer)
470{
471 debug_hrtimer_deactivate(timer);
472 trace_hrtimer_cancel(timer);
473}
474
475static struct hrtimer_clock_base *
476__next_base(struct hrtimer_cpu_base *cpu_base, unsigned int *active)
477{
478 unsigned int idx;
479
480 if (!*active)
481 return NULL;
482
483 idx = __ffs(*active);
484 *active &= ~(1U << idx);
485
486 return &cpu_base->clock_base[idx];
487}
488
489#define for_each_active_base(base, cpu_base, active) \
490 while ((base = __next_base((cpu_base), &(active))))
491
492static ktime_t __hrtimer_next_event_base(struct hrtimer_cpu_base *cpu_base,
493 const struct hrtimer *exclude,
494 unsigned int active,
495 ktime_t expires_next)
496{
497 struct hrtimer_clock_base *base;
498 ktime_t expires;
499
500 for_each_active_base(base, cpu_base, active) {
501 struct timerqueue_node *next;
502 struct hrtimer *timer;
503
504 next = timerqueue_getnext(&base->active);
505 timer = container_of(next, struct hrtimer, node);
506 if (timer == exclude) {
507 /* Get to the next timer in the queue. */
508 next = timerqueue_iterate_next(next);
509 if (!next)
510 continue;
511
512 timer = container_of(next, struct hrtimer, node);
513 }
514 expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
515 if (expires < expires_next) {
516 expires_next = expires;
517
518 /* Skip cpu_base update if a timer is being excluded. */
519 if (exclude)
520 continue;
521
522 if (timer->is_soft)
523 cpu_base->softirq_next_timer = timer;
524 else
525 cpu_base->next_timer = timer;
526 }
527 }
528 /*
529 * clock_was_set() might have changed base->offset of any of
530 * the clock bases so the result might be negative. Fix it up
531 * to prevent a false positive in clockevents_program_event().
532 */
533 if (expires_next < 0)
534 expires_next = 0;
535 return expires_next;
536}
537
538/*
539 * Recomputes cpu_base::*next_timer and returns the earliest expires_next but
540 * does not set cpu_base::*expires_next, that is done by hrtimer_reprogram.
541 *
542 * When a softirq is pending, we can ignore the HRTIMER_ACTIVE_SOFT bases,
543 * those timers will get run whenever the softirq gets handled, at the end of
544 * hrtimer_run_softirq(), hrtimer_update_softirq_timer() will re-add these bases.
545 *
546 * Therefore softirq values are those from the HRTIMER_ACTIVE_SOFT clock bases.
547 * The !softirq values are the minima across HRTIMER_ACTIVE_ALL, unless an actual
548 * softirq is pending, in which case they're the minima of HRTIMER_ACTIVE_HARD.
549 *
550 * @active_mask must be one of:
551 * - HRTIMER_ACTIVE_ALL,
552 * - HRTIMER_ACTIVE_SOFT, or
553 * - HRTIMER_ACTIVE_HARD.
554 */
555static ktime_t
556__hrtimer_get_next_event(struct hrtimer_cpu_base *cpu_base, unsigned int active_mask)
557{
558 unsigned int active;
559 struct hrtimer *next_timer = NULL;
560 ktime_t expires_next = KTIME_MAX;
561
562 if (!cpu_base->softirq_activated && (active_mask & HRTIMER_ACTIVE_SOFT)) {
563 active = cpu_base->active_bases & HRTIMER_ACTIVE_SOFT;
564 cpu_base->softirq_next_timer = NULL;
565 expires_next = __hrtimer_next_event_base(cpu_base, NULL,
566 active, KTIME_MAX);
567
568 next_timer = cpu_base->softirq_next_timer;
569 }
570
571 if (active_mask & HRTIMER_ACTIVE_HARD) {
572 active = cpu_base->active_bases & HRTIMER_ACTIVE_HARD;
573 cpu_base->next_timer = next_timer;
574 expires_next = __hrtimer_next_event_base(cpu_base, NULL, active,
575 expires_next);
576 }
577
578 return expires_next;
579}
580
581static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base)
582{
583 ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset;
584 ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset;
585 ktime_t *offs_tai = &base->clock_base[HRTIMER_BASE_TAI].offset;
586
587 ktime_t now = ktime_get_update_offsets_now(&base->clock_was_set_seq,
588 offs_real, offs_boot, offs_tai);
589
590 base->clock_base[HRTIMER_BASE_REALTIME_SOFT].offset = *offs_real;
591 base->clock_base[HRTIMER_BASE_BOOTTIME_SOFT].offset = *offs_boot;
592 base->clock_base[HRTIMER_BASE_TAI_SOFT].offset = *offs_tai;
593
594 return now;
595}
596
597/*
598 * Is the high resolution mode active ?
599 */
600static inline int __hrtimer_hres_active(struct hrtimer_cpu_base *cpu_base)
601{
602 return IS_ENABLED(CONFIG_HIGH_RES_TIMERS) ?
603 cpu_base->hres_active : 0;
604}
605
606static inline int hrtimer_hres_active(void)
607{
608 return __hrtimer_hres_active(this_cpu_ptr(&hrtimer_bases));
609}
610
611/*
612 * Reprogram the event source with checking both queues for the
613 * next event
614 * Called with interrupts disabled and base->lock held
615 */
616static void
617hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
618{
619 ktime_t expires_next;
620
621 /*
622 * Find the current next expiration time.
623 */
624 expires_next = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_ALL);
625
626 if (cpu_base->next_timer && cpu_base->next_timer->is_soft) {
627 /*
628 * When the softirq is activated, hrtimer has to be
629 * programmed with the first hard hrtimer because soft
630 * timer interrupt could occur too late.
631 */
632 if (cpu_base->softirq_activated)
633 expires_next = __hrtimer_get_next_event(cpu_base,
634 HRTIMER_ACTIVE_HARD);
635 else
636 cpu_base->softirq_expires_next = expires_next;
637 }
638
639 if (skip_equal && expires_next == cpu_base->expires_next)
640 return;
641
642 cpu_base->expires_next = expires_next;
643
644 /*
645 * If hres is not active, hardware does not have to be
646 * reprogrammed yet.
647 *
648 * If a hang was detected in the last timer interrupt then we
649 * leave the hang delay active in the hardware. We want the
650 * system to make progress. That also prevents the following
651 * scenario:
652 * T1 expires 50ms from now
653 * T2 expires 5s from now
654 *
655 * T1 is removed, so this code is called and would reprogram
656 * the hardware to 5s from now. Any hrtimer_start after that
657 * will not reprogram the hardware due to hang_detected being
658 * set. So we'd effectivly block all timers until the T2 event
659 * fires.
660 */
661 if (!__hrtimer_hres_active(cpu_base) || cpu_base->hang_detected)
662 return;
663
664 tick_program_event(cpu_base->expires_next, 1);
665}
666
667/* High resolution timer related functions */
668#ifdef CONFIG_HIGH_RES_TIMERS
669
670/*
671 * High resolution timer enabled ?
672 */
673static bool hrtimer_hres_enabled __read_mostly = true;
674unsigned int hrtimer_resolution __read_mostly = LOW_RES_NSEC;
675EXPORT_SYMBOL_GPL(hrtimer_resolution);
676
677/*
678 * Enable / Disable high resolution mode
679 */
680static int __init setup_hrtimer_hres(char *str)
681{
682 return (kstrtobool(str, &hrtimer_hres_enabled) == 0);
683}
684
685__setup("highres=", setup_hrtimer_hres);
686
687/*
688 * hrtimer_high_res_enabled - query, if the highres mode is enabled
689 */
690static inline int hrtimer_is_hres_enabled(void)
691{
692 return hrtimer_hres_enabled;
693}
694
695/*
696 * Retrigger next event is called after clock was set
697 *
698 * Called with interrupts disabled via on_each_cpu()
699 */
700static void retrigger_next_event(void *arg)
701{
702 struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
703
704 if (!__hrtimer_hres_active(base))
705 return;
706
707 raw_spin_lock(&base->lock);
708 hrtimer_update_base(base);
709 hrtimer_force_reprogram(base, 0);
710 raw_spin_unlock(&base->lock);
711}
712
713/*
714 * Switch to high resolution mode
715 */
716static void hrtimer_switch_to_hres(void)
717{
718 struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
719
720 if (tick_init_highres()) {
721 printk(KERN_WARNING "Could not switch to high resolution "
722 "mode on CPU %d\n", base->cpu);
723 return;
724 }
725 base->hres_active = 1;
726 hrtimer_resolution = HIGH_RES_NSEC;
727
728 tick_setup_sched_timer();
729 /* "Retrigger" the interrupt to get things going */
730 retrigger_next_event(NULL);
731}
732
733static void clock_was_set_work(struct work_struct *work)
734{
735 clock_was_set();
736}
737
738static DECLARE_WORK(hrtimer_work, clock_was_set_work);
739
740/*
741 * Called from timekeeping and resume code to reprogram the hrtimer
742 * interrupt device on all cpus.
743 */
744void clock_was_set_delayed(void)
745{
746 schedule_work(&hrtimer_work);
747}
748
749#else
750
751static inline int hrtimer_is_hres_enabled(void) { return 0; }
752static inline void hrtimer_switch_to_hres(void) { }
753static inline void retrigger_next_event(void *arg) { }
754
755#endif /* CONFIG_HIGH_RES_TIMERS */
756
757/*
758 * When a timer is enqueued and expires earlier than the already enqueued
759 * timers, we have to check, whether it expires earlier than the timer for
760 * which the clock event device was armed.
761 *
762 * Called with interrupts disabled and base->cpu_base.lock held
763 */
764static void hrtimer_reprogram(struct hrtimer *timer, bool reprogram)
765{
766 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
767 struct hrtimer_clock_base *base = timer->base;
768 ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
769
770 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
771
772 /*
773 * CLOCK_REALTIME timer might be requested with an absolute
774 * expiry time which is less than base->offset. Set it to 0.
775 */
776 if (expires < 0)
777 expires = 0;
778
779 if (timer->is_soft) {
780 /*
781 * soft hrtimer could be started on a remote CPU. In this
782 * case softirq_expires_next needs to be updated on the
783 * remote CPU. The soft hrtimer will not expire before the
784 * first hard hrtimer on the remote CPU -
785 * hrtimer_check_target() prevents this case.
786 */
787 struct hrtimer_cpu_base *timer_cpu_base = base->cpu_base;
788
789 if (timer_cpu_base->softirq_activated)
790 return;
791
792 if (!ktime_before(expires, timer_cpu_base->softirq_expires_next))
793 return;
794
795 timer_cpu_base->softirq_next_timer = timer;
796 timer_cpu_base->softirq_expires_next = expires;
797
798 if (!ktime_before(expires, timer_cpu_base->expires_next) ||
799 !reprogram)
800 return;
801 }
802
803 /*
804 * If the timer is not on the current cpu, we cannot reprogram
805 * the other cpus clock event device.
806 */
807 if (base->cpu_base != cpu_base)
808 return;
809
810 /*
811 * If the hrtimer interrupt is running, then it will
812 * reevaluate the clock bases and reprogram the clock event
813 * device. The callbacks are always executed in hard interrupt
814 * context so we don't need an extra check for a running
815 * callback.
816 */
817 if (cpu_base->in_hrtirq)
818 return;
819
820 if (expires >= cpu_base->expires_next)
821 return;
822
823 /* Update the pointer to the next expiring timer */
824 cpu_base->next_timer = timer;
825 cpu_base->expires_next = expires;
826
827 /*
828 * If hres is not active, hardware does not have to be
829 * programmed yet.
830 *
831 * If a hang was detected in the last timer interrupt then we
832 * do not schedule a timer which is earlier than the expiry
833 * which we enforced in the hang detection. We want the system
834 * to make progress.
835 */
836 if (!__hrtimer_hres_active(cpu_base) || cpu_base->hang_detected)
837 return;
838
839 /*
840 * Program the timer hardware. We enforce the expiry for
841 * events which are already in the past.
842 */
843 tick_program_event(expires, 1);
844}
845
846/*
847 * Clock realtime was set
848 *
849 * Change the offset of the realtime clock vs. the monotonic
850 * clock.
851 *
852 * We might have to reprogram the high resolution timer interrupt. On
853 * SMP we call the architecture specific code to retrigger _all_ high
854 * resolution timer interrupts. On UP we just disable interrupts and
855 * call the high resolution interrupt code.
856 */
857void clock_was_set(void)
858{
859#ifdef CONFIG_HIGH_RES_TIMERS
860 /* Retrigger the CPU local events everywhere */
861 on_each_cpu(retrigger_next_event, NULL, 1);
862#endif
863 timerfd_clock_was_set();
864}
865
866/*
867 * During resume we might have to reprogram the high resolution timer
868 * interrupt on all online CPUs. However, all other CPUs will be
869 * stopped with IRQs interrupts disabled so the clock_was_set() call
870 * must be deferred.
871 */
872void hrtimers_resume(void)
873{
874 lockdep_assert_irqs_disabled();
875 /* Retrigger on the local CPU */
876 retrigger_next_event(NULL);
877 /* And schedule a retrigger for all others */
878 clock_was_set_delayed();
879}
880
881/*
882 * Counterpart to lock_hrtimer_base above:
883 */
884static inline
885void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
886{
887 raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
888}
889
890/**
891 * hrtimer_forward - forward the timer expiry
892 * @timer: hrtimer to forward
893 * @now: forward past this time
894 * @interval: the interval to forward
895 *
896 * Forward the timer expiry so it will expire in the future.
897 * Returns the number of overruns.
898 *
899 * Can be safely called from the callback function of @timer. If
900 * called from other contexts @timer must neither be enqueued nor
901 * running the callback and the caller needs to take care of
902 * serialization.
903 *
904 * Note: This only updates the timer expiry value and does not requeue
905 * the timer.
906 */
907u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
908{
909 u64 orun = 1;
910 ktime_t delta;
911
912 delta = ktime_sub(now, hrtimer_get_expires(timer));
913
914 if (delta < 0)
915 return 0;
916
917 if (WARN_ON(timer->state & HRTIMER_STATE_ENQUEUED))
918 return 0;
919
920 if (interval < hrtimer_resolution)
921 interval = hrtimer_resolution;
922
923 if (unlikely(delta >= interval)) {
924 s64 incr = ktime_to_ns(interval);
925
926 orun = ktime_divns(delta, incr);
927 hrtimer_add_expires_ns(timer, incr * orun);
928 if (hrtimer_get_expires_tv64(timer) > now)
929 return orun;
930 /*
931 * This (and the ktime_add() below) is the
932 * correction for exact:
933 */
934 orun++;
935 }
936 hrtimer_add_expires(timer, interval);
937
938 return orun;
939}
940EXPORT_SYMBOL_GPL(hrtimer_forward);
941
942/*
943 * enqueue_hrtimer - internal function to (re)start a timer
944 *
945 * The timer is inserted in expiry order. Insertion into the
946 * red black tree is O(log(n)). Must hold the base lock.
947 *
948 * Returns 1 when the new timer is the leftmost timer in the tree.
949 */
950static int enqueue_hrtimer(struct hrtimer *timer,
951 struct hrtimer_clock_base *base,
952 enum hrtimer_mode mode)
953{
954 debug_activate(timer, mode);
955
956 base->cpu_base->active_bases |= 1 << base->index;
957
958 timer->state = HRTIMER_STATE_ENQUEUED;
959
960 return timerqueue_add(&base->active, &timer->node);
961}
962
963/*
964 * __remove_hrtimer - internal function to remove a timer
965 *
966 * Caller must hold the base lock.
967 *
968 * High resolution timer mode reprograms the clock event device when the
969 * timer is the one which expires next. The caller can disable this by setting
970 * reprogram to zero. This is useful, when the context does a reprogramming
971 * anyway (e.g. timer interrupt)
972 */
973static void __remove_hrtimer(struct hrtimer *timer,
974 struct hrtimer_clock_base *base,
975 u8 newstate, int reprogram)
976{
977 struct hrtimer_cpu_base *cpu_base = base->cpu_base;
978 u8 state = timer->state;
979
980 timer->state = newstate;
981 if (!(state & HRTIMER_STATE_ENQUEUED))
982 return;
983
984 if (!timerqueue_del(&base->active, &timer->node))
985 cpu_base->active_bases &= ~(1 << base->index);
986
987 /*
988 * Note: If reprogram is false we do not update
989 * cpu_base->next_timer. This happens when we remove the first
990 * timer on a remote cpu. No harm as we never dereference
991 * cpu_base->next_timer. So the worst thing what can happen is
992 * an superflous call to hrtimer_force_reprogram() on the
993 * remote cpu later on if the same timer gets enqueued again.
994 */
995 if (reprogram && timer == cpu_base->next_timer)
996 hrtimer_force_reprogram(cpu_base, 1);
997}
998
999/*
1000 * remove hrtimer, called with base lock held
1001 */
1002static inline int
1003remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base, bool restart)
1004{
1005 if (hrtimer_is_queued(timer)) {
1006 u8 state = timer->state;
1007 int reprogram;
1008
1009 /*
1010 * Remove the timer and force reprogramming when high
1011 * resolution mode is active and the timer is on the current
1012 * CPU. If we remove a timer on another CPU, reprogramming is
1013 * skipped. The interrupt event on this CPU is fired and
1014 * reprogramming happens in the interrupt handler. This is a
1015 * rare case and less expensive than a smp call.
1016 */
1017 debug_deactivate(timer);
1018 reprogram = base->cpu_base == this_cpu_ptr(&hrtimer_bases);
1019
1020 if (!restart)
1021 state = HRTIMER_STATE_INACTIVE;
1022
1023 __remove_hrtimer(timer, base, state, reprogram);
1024 return 1;
1025 }
1026 return 0;
1027}
1028
1029static inline ktime_t hrtimer_update_lowres(struct hrtimer *timer, ktime_t tim,
1030 const enum hrtimer_mode mode)
1031{
1032#ifdef CONFIG_TIME_LOW_RES
1033 /*
1034 * CONFIG_TIME_LOW_RES indicates that the system has no way to return
1035 * granular time values. For relative timers we add hrtimer_resolution
1036 * (i.e. one jiffie) to prevent short timeouts.
1037 */
1038 timer->is_rel = mode & HRTIMER_MODE_REL;
1039 if (timer->is_rel)
1040 tim = ktime_add_safe(tim, hrtimer_resolution);
1041#endif
1042 return tim;
1043}
1044
1045static void
1046hrtimer_update_softirq_timer(struct hrtimer_cpu_base *cpu_base, bool reprogram)
1047{
1048 ktime_t expires;
1049
1050 /*
1051 * Find the next SOFT expiration.
1052 */
1053 expires = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_SOFT);
1054
1055 /*
1056 * reprogramming needs to be triggered, even if the next soft
1057 * hrtimer expires at the same time than the next hard
1058 * hrtimer. cpu_base->softirq_expires_next needs to be updated!
1059 */
1060 if (expires == KTIME_MAX)
1061 return;
1062
1063 /*
1064 * cpu_base->*next_timer is recomputed by __hrtimer_get_next_event()
1065 * cpu_base->*expires_next is only set by hrtimer_reprogram()
1066 */
1067 hrtimer_reprogram(cpu_base->softirq_next_timer, reprogram);
1068}
1069
1070static int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1071 u64 delta_ns, const enum hrtimer_mode mode,
1072 struct hrtimer_clock_base *base)
1073{
1074 struct hrtimer_clock_base *new_base;
1075
1076 /* Remove an active timer from the queue: */
1077 remove_hrtimer(timer, base, true);
1078
1079 if (mode & HRTIMER_MODE_REL)
1080 tim = ktime_add_safe(tim, base->get_time());
1081
1082 tim = hrtimer_update_lowres(timer, tim, mode);
1083
1084 hrtimer_set_expires_range_ns(timer, tim, delta_ns);
1085
1086 /* Switch the timer base, if necessary: */
1087 new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED);
1088
1089 return enqueue_hrtimer(timer, new_base, mode);
1090}
1091
1092/**
1093 * hrtimer_start_range_ns - (re)start an hrtimer
1094 * @timer: the timer to be added
1095 * @tim: expiry time
1096 * @delta_ns: "slack" range for the timer
1097 * @mode: timer mode: absolute (HRTIMER_MODE_ABS) or
1098 * relative (HRTIMER_MODE_REL), and pinned (HRTIMER_MODE_PINNED);
1099 * softirq based mode is considered for debug purpose only!
1100 */
1101void hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1102 u64 delta_ns, const enum hrtimer_mode mode)
1103{
1104 struct hrtimer_clock_base *base;
1105 unsigned long flags;
1106
1107 /*
1108 * Check whether the HRTIMER_MODE_SOFT bit and hrtimer.is_soft
1109 * match.
1110 */
1111 WARN_ON_ONCE(!(mode & HRTIMER_MODE_SOFT) ^ !timer->is_soft);
1112
1113 base = lock_hrtimer_base(timer, &flags);
1114
1115 if (__hrtimer_start_range_ns(timer, tim, delta_ns, mode, base))
1116 hrtimer_reprogram(timer, true);
1117
1118 unlock_hrtimer_base(timer, &flags);
1119}
1120EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
1121
1122/**
1123 * hrtimer_try_to_cancel - try to deactivate a timer
1124 * @timer: hrtimer to stop
1125 *
1126 * Returns:
1127 * 0 when the timer was not active
1128 * 1 when the timer was active
1129 * -1 when the timer is currently executing the callback function and
1130 * cannot be stopped
1131 */
1132int hrtimer_try_to_cancel(struct hrtimer *timer)
1133{
1134 struct hrtimer_clock_base *base;
1135 unsigned long flags;
1136 int ret = -1;
1137
1138 /*
1139 * Check lockless first. If the timer is not active (neither
1140 * enqueued nor running the callback, nothing to do here. The
1141 * base lock does not serialize against a concurrent enqueue,
1142 * so we can avoid taking it.
1143 */
1144 if (!hrtimer_active(timer))
1145 return 0;
1146
1147 base = lock_hrtimer_base(timer, &flags);
1148
1149 if (!hrtimer_callback_running(timer))
1150 ret = remove_hrtimer(timer, base, false);
1151
1152 unlock_hrtimer_base(timer, &flags);
1153
1154 return ret;
1155
1156}
1157EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1158
1159/**
1160 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1161 * @timer: the timer to be cancelled
1162 *
1163 * Returns:
1164 * 0 when the timer was not active
1165 * 1 when the timer was active
1166 */
1167int hrtimer_cancel(struct hrtimer *timer)
1168{
1169 for (;;) {
1170 int ret = hrtimer_try_to_cancel(timer);
1171
1172 if (ret >= 0)
1173 return ret;
1174 cpu_relax();
1175 }
1176}
1177EXPORT_SYMBOL_GPL(hrtimer_cancel);
1178
1179/**
1180 * hrtimer_get_remaining - get remaining time for the timer
1181 * @timer: the timer to read
1182 * @adjust: adjust relative timers when CONFIG_TIME_LOW_RES=y
1183 */
1184ktime_t __hrtimer_get_remaining(const struct hrtimer *timer, bool adjust)
1185{
1186 unsigned long flags;
1187 ktime_t rem;
1188
1189 lock_hrtimer_base(timer, &flags);
1190 if (IS_ENABLED(CONFIG_TIME_LOW_RES) && adjust)
1191 rem = hrtimer_expires_remaining_adjusted(timer);
1192 else
1193 rem = hrtimer_expires_remaining(timer);
1194 unlock_hrtimer_base(timer, &flags);
1195
1196 return rem;
1197}
1198EXPORT_SYMBOL_GPL(__hrtimer_get_remaining);
1199
1200#ifdef CONFIG_NO_HZ_COMMON
1201/**
1202 * hrtimer_get_next_event - get the time until next expiry event
1203 *
1204 * Returns the next expiry time or KTIME_MAX if no timer is pending.
1205 */
1206u64 hrtimer_get_next_event(void)
1207{
1208 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1209 u64 expires = KTIME_MAX;
1210 unsigned long flags;
1211
1212 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1213
1214 if (!__hrtimer_hres_active(cpu_base))
1215 expires = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_ALL);
1216
1217 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1218
1219 return expires;
1220}
1221
1222/**
1223 * hrtimer_next_event_without - time until next expiry event w/o one timer
1224 * @exclude: timer to exclude
1225 *
1226 * Returns the next expiry time over all timers except for the @exclude one or
1227 * KTIME_MAX if none of them is pending.
1228 */
1229u64 hrtimer_next_event_without(const struct hrtimer *exclude)
1230{
1231 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1232 u64 expires = KTIME_MAX;
1233 unsigned long flags;
1234
1235 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1236
1237 if (__hrtimer_hres_active(cpu_base)) {
1238 unsigned int active;
1239
1240 if (!cpu_base->softirq_activated) {
1241 active = cpu_base->active_bases & HRTIMER_ACTIVE_SOFT;
1242 expires = __hrtimer_next_event_base(cpu_base, exclude,
1243 active, KTIME_MAX);
1244 }
1245 active = cpu_base->active_bases & HRTIMER_ACTIVE_HARD;
1246 expires = __hrtimer_next_event_base(cpu_base, exclude, active,
1247 expires);
1248 }
1249
1250 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1251
1252 return expires;
1253}
1254#endif
1255
1256static inline int hrtimer_clockid_to_base(clockid_t clock_id)
1257{
1258 if (likely(clock_id < MAX_CLOCKS)) {
1259 int base = hrtimer_clock_to_base_table[clock_id];
1260
1261 if (likely(base != HRTIMER_MAX_CLOCK_BASES))
1262 return base;
1263 }
1264 WARN(1, "Invalid clockid %d. Using MONOTONIC\n", clock_id);
1265 return HRTIMER_BASE_MONOTONIC;
1266}
1267
1268static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1269 enum hrtimer_mode mode)
1270{
1271 bool softtimer = !!(mode & HRTIMER_MODE_SOFT);
1272 int base = softtimer ? HRTIMER_MAX_CLOCK_BASES / 2 : 0;
1273 struct hrtimer_cpu_base *cpu_base;
1274
1275 memset(timer, 0, sizeof(struct hrtimer));
1276
1277 cpu_base = raw_cpu_ptr(&hrtimer_bases);
1278
1279 /*
1280 * POSIX magic: Relative CLOCK_REALTIME timers are not affected by
1281 * clock modifications, so they needs to become CLOCK_MONOTONIC to
1282 * ensure POSIX compliance.
1283 */
1284 if (clock_id == CLOCK_REALTIME && mode & HRTIMER_MODE_REL)
1285 clock_id = CLOCK_MONOTONIC;
1286
1287 base += hrtimer_clockid_to_base(clock_id);
1288 timer->is_soft = softtimer;
1289 timer->base = &cpu_base->clock_base[base];
1290 timerqueue_init(&timer->node);
1291}
1292
1293/**
1294 * hrtimer_init - initialize a timer to the given clock
1295 * @timer: the timer to be initialized
1296 * @clock_id: the clock to be used
1297 * @mode: The modes which are relevant for intitialization:
1298 * HRTIMER_MODE_ABS, HRTIMER_MODE_REL, HRTIMER_MODE_ABS_SOFT,
1299 * HRTIMER_MODE_REL_SOFT
1300 *
1301 * The PINNED variants of the above can be handed in,
1302 * but the PINNED bit is ignored as pinning happens
1303 * when the hrtimer is started
1304 */
1305void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1306 enum hrtimer_mode mode)
1307{
1308 debug_init(timer, clock_id, mode);
1309 __hrtimer_init(timer, clock_id, mode);
1310}
1311EXPORT_SYMBOL_GPL(hrtimer_init);
1312
1313/*
1314 * A timer is active, when it is enqueued into the rbtree or the
1315 * callback function is running or it's in the state of being migrated
1316 * to another cpu.
1317 *
1318 * It is important for this function to not return a false negative.
1319 */
1320bool hrtimer_active(const struct hrtimer *timer)
1321{
1322 struct hrtimer_clock_base *base;
1323 unsigned int seq;
1324
1325 do {
1326 base = READ_ONCE(timer->base);
1327 seq = raw_read_seqcount_begin(&base->seq);
1328
1329 if (timer->state != HRTIMER_STATE_INACTIVE ||
1330 base->running == timer)
1331 return true;
1332
1333 } while (read_seqcount_retry(&base->seq, seq) ||
1334 base != READ_ONCE(timer->base));
1335
1336 return false;
1337}
1338EXPORT_SYMBOL_GPL(hrtimer_active);
1339
1340/*
1341 * The write_seqcount_barrier()s in __run_hrtimer() split the thing into 3
1342 * distinct sections:
1343 *
1344 * - queued: the timer is queued
1345 * - callback: the timer is being ran
1346 * - post: the timer is inactive or (re)queued
1347 *
1348 * On the read side we ensure we observe timer->state and cpu_base->running
1349 * from the same section, if anything changed while we looked at it, we retry.
1350 * This includes timer->base changing because sequence numbers alone are
1351 * insufficient for that.
1352 *
1353 * The sequence numbers are required because otherwise we could still observe
1354 * a false negative if the read side got smeared over multiple consequtive
1355 * __run_hrtimer() invocations.
1356 */
1357
1358static void __run_hrtimer(struct hrtimer_cpu_base *cpu_base,
1359 struct hrtimer_clock_base *base,
1360 struct hrtimer *timer, ktime_t *now,
1361 unsigned long flags)
1362{
1363 enum hrtimer_restart (*fn)(struct hrtimer *);
1364 int restart;
1365
1366 lockdep_assert_held(&cpu_base->lock);
1367
1368 debug_deactivate(timer);
1369 base->running = timer;
1370
1371 /*
1372 * Separate the ->running assignment from the ->state assignment.
1373 *
1374 * As with a regular write barrier, this ensures the read side in
1375 * hrtimer_active() cannot observe base->running == NULL &&
1376 * timer->state == INACTIVE.
1377 */
1378 raw_write_seqcount_barrier(&base->seq);
1379
1380 __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE, 0);
1381 fn = timer->function;
1382
1383 /*
1384 * Clear the 'is relative' flag for the TIME_LOW_RES case. If the
1385 * timer is restarted with a period then it becomes an absolute
1386 * timer. If its not restarted it does not matter.
1387 */
1388 if (IS_ENABLED(CONFIG_TIME_LOW_RES))
1389 timer->is_rel = false;
1390
1391 /*
1392 * The timer is marked as running in the CPU base, so it is
1393 * protected against migration to a different CPU even if the lock
1394 * is dropped.
1395 */
1396 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1397 trace_hrtimer_expire_entry(timer, now);
1398 restart = fn(timer);
1399 trace_hrtimer_expire_exit(timer);
1400 raw_spin_lock_irq(&cpu_base->lock);
1401
1402 /*
1403 * Note: We clear the running state after enqueue_hrtimer and
1404 * we do not reprogram the event hardware. Happens either in
1405 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1406 *
1407 * Note: Because we dropped the cpu_base->lock above,
1408 * hrtimer_start_range_ns() can have popped in and enqueued the timer
1409 * for us already.
1410 */
1411 if (restart != HRTIMER_NORESTART &&
1412 !(timer->state & HRTIMER_STATE_ENQUEUED))
1413 enqueue_hrtimer(timer, base, HRTIMER_MODE_ABS);
1414
1415 /*
1416 * Separate the ->running assignment from the ->state assignment.
1417 *
1418 * As with a regular write barrier, this ensures the read side in
1419 * hrtimer_active() cannot observe base->running.timer == NULL &&
1420 * timer->state == INACTIVE.
1421 */
1422 raw_write_seqcount_barrier(&base->seq);
1423
1424 WARN_ON_ONCE(base->running != timer);
1425 base->running = NULL;
1426}
1427
1428static void __hrtimer_run_queues(struct hrtimer_cpu_base *cpu_base, ktime_t now,
1429 unsigned long flags, unsigned int active_mask)
1430{
1431 struct hrtimer_clock_base *base;
1432 unsigned int active = cpu_base->active_bases & active_mask;
1433
1434 for_each_active_base(base, cpu_base, active) {
1435 struct timerqueue_node *node;
1436 ktime_t basenow;
1437
1438 basenow = ktime_add(now, base->offset);
1439
1440 while ((node = timerqueue_getnext(&base->active))) {
1441 struct hrtimer *timer;
1442
1443 timer = container_of(node, struct hrtimer, node);
1444
1445 /*
1446 * The immediate goal for using the softexpires is
1447 * minimizing wakeups, not running timers at the
1448 * earliest interrupt after their soft expiration.
1449 * This allows us to avoid using a Priority Search
1450 * Tree, which can answer a stabbing querry for
1451 * overlapping intervals and instead use the simple
1452 * BST we already have.
1453 * We don't add extra wakeups by delaying timers that
1454 * are right-of a not yet expired timer, because that
1455 * timer will have to trigger a wakeup anyway.
1456 */
1457 if (basenow < hrtimer_get_softexpires_tv64(timer))
1458 break;
1459
1460 __run_hrtimer(cpu_base, base, timer, &basenow, flags);
1461 }
1462 }
1463}
1464
1465static __latent_entropy void hrtimer_run_softirq(struct softirq_action *h)
1466{
1467 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1468 unsigned long flags;
1469 ktime_t now;
1470
1471 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1472
1473 now = hrtimer_update_base(cpu_base);
1474 __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_SOFT);
1475
1476 cpu_base->softirq_activated = 0;
1477 hrtimer_update_softirq_timer(cpu_base, true);
1478
1479 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1480}
1481
1482#ifdef CONFIG_HIGH_RES_TIMERS
1483
1484/*
1485 * High resolution timer interrupt
1486 * Called with interrupts disabled
1487 */
1488void hrtimer_interrupt(struct clock_event_device *dev)
1489{
1490 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1491 ktime_t expires_next, now, entry_time, delta;
1492 unsigned long flags;
1493 int retries = 0;
1494
1495 BUG_ON(!cpu_base->hres_active);
1496 cpu_base->nr_events++;
1497 dev->next_event = KTIME_MAX;
1498
1499 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1500 entry_time = now = hrtimer_update_base(cpu_base);
1501retry:
1502 cpu_base->in_hrtirq = 1;
1503 /*
1504 * We set expires_next to KTIME_MAX here with cpu_base->lock
1505 * held to prevent that a timer is enqueued in our queue via
1506 * the migration code. This does not affect enqueueing of
1507 * timers which run their callback and need to be requeued on
1508 * this CPU.
1509 */
1510 cpu_base->expires_next = KTIME_MAX;
1511
1512 if (!ktime_before(now, cpu_base->softirq_expires_next)) {
1513 cpu_base->softirq_expires_next = KTIME_MAX;
1514 cpu_base->softirq_activated = 1;
1515 raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1516 }
1517
1518 __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_HARD);
1519
1520 /* Reevaluate the clock bases for the next expiry */
1521 expires_next = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_ALL);
1522 /*
1523 * Store the new expiry value so the migration code can verify
1524 * against it.
1525 */
1526 cpu_base->expires_next = expires_next;
1527 cpu_base->in_hrtirq = 0;
1528 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1529
1530 /* Reprogramming necessary ? */
1531 if (!tick_program_event(expires_next, 0)) {
1532 cpu_base->hang_detected = 0;
1533 return;
1534 }
1535
1536 /*
1537 * The next timer was already expired due to:
1538 * - tracing
1539 * - long lasting callbacks
1540 * - being scheduled away when running in a VM
1541 *
1542 * We need to prevent that we loop forever in the hrtimer
1543 * interrupt routine. We give it 3 attempts to avoid
1544 * overreacting on some spurious event.
1545 *
1546 * Acquire base lock for updating the offsets and retrieving
1547 * the current time.
1548 */
1549 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1550 now = hrtimer_update_base(cpu_base);
1551 cpu_base->nr_retries++;
1552 if (++retries < 3)
1553 goto retry;
1554 /*
1555 * Give the system a chance to do something else than looping
1556 * here. We stored the entry time, so we know exactly how long
1557 * we spent here. We schedule the next event this amount of
1558 * time away.
1559 */
1560 cpu_base->nr_hangs++;
1561 cpu_base->hang_detected = 1;
1562 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1563
1564 delta = ktime_sub(now, entry_time);
1565 if ((unsigned int)delta > cpu_base->max_hang_time)
1566 cpu_base->max_hang_time = (unsigned int) delta;
1567 /*
1568 * Limit it to a sensible value as we enforce a longer
1569 * delay. Give the CPU at least 100ms to catch up.
1570 */
1571 if (delta > 100 * NSEC_PER_MSEC)
1572 expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1573 else
1574 expires_next = ktime_add(now, delta);
1575 tick_program_event(expires_next, 1);
1576 printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n",
1577 ktime_to_ns(delta));
1578}
1579
1580/* called with interrupts disabled */
1581static inline void __hrtimer_peek_ahead_timers(void)
1582{
1583 struct tick_device *td;
1584
1585 if (!hrtimer_hres_active())
1586 return;
1587
1588 td = this_cpu_ptr(&tick_cpu_device);
1589 if (td && td->evtdev)
1590 hrtimer_interrupt(td->evtdev);
1591}
1592
1593#else /* CONFIG_HIGH_RES_TIMERS */
1594
1595static inline void __hrtimer_peek_ahead_timers(void) { }
1596
1597#endif /* !CONFIG_HIGH_RES_TIMERS */
1598
1599/*
1600 * Called from run_local_timers in hardirq context every jiffy
1601 */
1602void hrtimer_run_queues(void)
1603{
1604 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1605 unsigned long flags;
1606 ktime_t now;
1607
1608 if (__hrtimer_hres_active(cpu_base))
1609 return;
1610
1611 /*
1612 * This _is_ ugly: We have to check periodically, whether we
1613 * can switch to highres and / or nohz mode. The clocksource
1614 * switch happens with xtime_lock held. Notification from
1615 * there only sets the check bit in the tick_oneshot code,
1616 * otherwise we might deadlock vs. xtime_lock.
1617 */
1618 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) {
1619 hrtimer_switch_to_hres();
1620 return;
1621 }
1622
1623 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1624 now = hrtimer_update_base(cpu_base);
1625
1626 if (!ktime_before(now, cpu_base->softirq_expires_next)) {
1627 cpu_base->softirq_expires_next = KTIME_MAX;
1628 cpu_base->softirq_activated = 1;
1629 raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1630 }
1631
1632 __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_HARD);
1633 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1634}
1635
1636/*
1637 * Sleep related functions:
1638 */
1639static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1640{
1641 struct hrtimer_sleeper *t =
1642 container_of(timer, struct hrtimer_sleeper, timer);
1643 struct task_struct *task = t->task;
1644
1645 t->task = NULL;
1646 if (task)
1647 wake_up_process(task);
1648
1649 return HRTIMER_NORESTART;
1650}
1651
1652void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1653{
1654 sl->timer.function = hrtimer_wakeup;
1655 sl->task = task;
1656}
1657EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
1658
1659int nanosleep_copyout(struct restart_block *restart, struct timespec64 *ts)
1660{
1661 switch(restart->nanosleep.type) {
1662#ifdef CONFIG_COMPAT
1663 case TT_COMPAT:
1664 if (compat_put_timespec64(ts, restart->nanosleep.compat_rmtp))
1665 return -EFAULT;
1666 break;
1667#endif
1668 case TT_NATIVE:
1669 if (put_timespec64(ts, restart->nanosleep.rmtp))
1670 return -EFAULT;
1671 break;
1672 default:
1673 BUG();
1674 }
1675 return -ERESTART_RESTARTBLOCK;
1676}
1677
1678static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1679{
1680 struct restart_block *restart;
1681
1682 hrtimer_init_sleeper(t, current);
1683
1684 do {
1685 set_current_state(TASK_INTERRUPTIBLE);
1686 hrtimer_start_expires(&t->timer, mode);
1687
1688 if (likely(t->task))
1689 freezable_schedule();
1690
1691 hrtimer_cancel(&t->timer);
1692 mode = HRTIMER_MODE_ABS;
1693
1694 } while (t->task && !signal_pending(current));
1695
1696 __set_current_state(TASK_RUNNING);
1697
1698 if (!t->task)
1699 return 0;
1700
1701 restart = ¤t->restart_block;
1702 if (restart->nanosleep.type != TT_NONE) {
1703 ktime_t rem = hrtimer_expires_remaining(&t->timer);
1704 struct timespec64 rmt;
1705
1706 if (rem <= 0)
1707 return 0;
1708 rmt = ktime_to_timespec64(rem);
1709
1710 return nanosleep_copyout(restart, &rmt);
1711 }
1712 return -ERESTART_RESTARTBLOCK;
1713}
1714
1715static long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1716{
1717 struct hrtimer_sleeper t;
1718 int ret;
1719
1720 hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid,
1721 HRTIMER_MODE_ABS);
1722 hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
1723
1724 ret = do_nanosleep(&t, HRTIMER_MODE_ABS);
1725 destroy_hrtimer_on_stack(&t.timer);
1726 return ret;
1727}
1728
1729long hrtimer_nanosleep(const struct timespec64 *rqtp,
1730 const enum hrtimer_mode mode, const clockid_t clockid)
1731{
1732 struct restart_block *restart;
1733 struct hrtimer_sleeper t;
1734 int ret = 0;
1735 u64 slack;
1736
1737 slack = current->timer_slack_ns;
1738 if (dl_task(current) || rt_task(current))
1739 slack = 0;
1740
1741 hrtimer_init_on_stack(&t.timer, clockid, mode);
1742 hrtimer_set_expires_range_ns(&t.timer, timespec64_to_ktime(*rqtp), slack);
1743 ret = do_nanosleep(&t, mode);
1744 if (ret != -ERESTART_RESTARTBLOCK)
1745 goto out;
1746
1747 /* Absolute timers do not update the rmtp value and restart: */
1748 if (mode == HRTIMER_MODE_ABS) {
1749 ret = -ERESTARTNOHAND;
1750 goto out;
1751 }
1752
1753 restart = ¤t->restart_block;
1754 restart->fn = hrtimer_nanosleep_restart;
1755 restart->nanosleep.clockid = t.timer.base->clockid;
1756 restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
1757out:
1758 destroy_hrtimer_on_stack(&t.timer);
1759 return ret;
1760}
1761
1762SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp,
1763 struct timespec __user *, rmtp)
1764{
1765 struct timespec64 tu;
1766
1767 if (get_timespec64(&tu, rqtp))
1768 return -EFAULT;
1769
1770 if (!timespec64_valid(&tu))
1771 return -EINVAL;
1772
1773 current->restart_block.nanosleep.type = rmtp ? TT_NATIVE : TT_NONE;
1774 current->restart_block.nanosleep.rmtp = rmtp;
1775 return hrtimer_nanosleep(&tu, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1776}
1777
1778#ifdef CONFIG_COMPAT
1779
1780COMPAT_SYSCALL_DEFINE2(nanosleep, struct compat_timespec __user *, rqtp,
1781 struct compat_timespec __user *, rmtp)
1782{
1783 struct timespec64 tu;
1784
1785 if (compat_get_timespec64(&tu, rqtp))
1786 return -EFAULT;
1787
1788 if (!timespec64_valid(&tu))
1789 return -EINVAL;
1790
1791 current->restart_block.nanosleep.type = rmtp ? TT_COMPAT : TT_NONE;
1792 current->restart_block.nanosleep.compat_rmtp = rmtp;
1793 return hrtimer_nanosleep(&tu, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1794}
1795#endif
1796
1797/*
1798 * Functions related to boot-time initialization:
1799 */
1800int hrtimers_prepare_cpu(unsigned int cpu)
1801{
1802 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1803 int i;
1804
1805 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1806 cpu_base->clock_base[i].cpu_base = cpu_base;
1807 timerqueue_init_head(&cpu_base->clock_base[i].active);
1808 }
1809
1810 cpu_base->cpu = cpu;
1811 cpu_base->active_bases = 0;
1812 cpu_base->hres_active = 0;
1813 cpu_base->hang_detected = 0;
1814 cpu_base->next_timer = NULL;
1815 cpu_base->softirq_next_timer = NULL;
1816 cpu_base->expires_next = KTIME_MAX;
1817 cpu_base->softirq_expires_next = KTIME_MAX;
1818 return 0;
1819}
1820
1821#ifdef CONFIG_HOTPLUG_CPU
1822
1823static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1824 struct hrtimer_clock_base *new_base)
1825{
1826 struct hrtimer *timer;
1827 struct timerqueue_node *node;
1828
1829 while ((node = timerqueue_getnext(&old_base->active))) {
1830 timer = container_of(node, struct hrtimer, node);
1831 BUG_ON(hrtimer_callback_running(timer));
1832 debug_deactivate(timer);
1833
1834 /*
1835 * Mark it as ENQUEUED not INACTIVE otherwise the
1836 * timer could be seen as !active and just vanish away
1837 * under us on another CPU
1838 */
1839 __remove_hrtimer(timer, old_base, HRTIMER_STATE_ENQUEUED, 0);
1840 timer->base = new_base;
1841 /*
1842 * Enqueue the timers on the new cpu. This does not
1843 * reprogram the event device in case the timer
1844 * expires before the earliest on this CPU, but we run
1845 * hrtimer_interrupt after we migrated everything to
1846 * sort out already expired timers and reprogram the
1847 * event device.
1848 */
1849 enqueue_hrtimer(timer, new_base, HRTIMER_MODE_ABS);
1850 }
1851}
1852
1853int hrtimers_dead_cpu(unsigned int scpu)
1854{
1855 struct hrtimer_cpu_base *old_base, *new_base;
1856 int i;
1857
1858 BUG_ON(cpu_online(scpu));
1859 tick_cancel_sched_timer(scpu);
1860
1861 /*
1862 * this BH disable ensures that raise_softirq_irqoff() does
1863 * not wakeup ksoftirqd (and acquire the pi-lock) while
1864 * holding the cpu_base lock
1865 */
1866 local_bh_disable();
1867 local_irq_disable();
1868 old_base = &per_cpu(hrtimer_bases, scpu);
1869 new_base = this_cpu_ptr(&hrtimer_bases);
1870 /*
1871 * The caller is globally serialized and nobody else
1872 * takes two locks at once, deadlock is not possible.
1873 */
1874 raw_spin_lock(&new_base->lock);
1875 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1876
1877 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1878 migrate_hrtimer_list(&old_base->clock_base[i],
1879 &new_base->clock_base[i]);
1880 }
1881
1882 /*
1883 * The migration might have changed the first expiring softirq
1884 * timer on this CPU. Update it.
1885 */
1886 hrtimer_update_softirq_timer(new_base, false);
1887
1888 raw_spin_unlock(&old_base->lock);
1889 raw_spin_unlock(&new_base->lock);
1890
1891 /* Check, if we got expired work to do */
1892 __hrtimer_peek_ahead_timers();
1893 local_irq_enable();
1894 local_bh_enable();
1895 return 0;
1896}
1897
1898#endif /* CONFIG_HOTPLUG_CPU */
1899
1900void __init hrtimers_init(void)
1901{
1902 hrtimers_prepare_cpu(smp_processor_id());
1903 open_softirq(HRTIMER_SOFTIRQ, hrtimer_run_softirq);
1904}
1905
1906/**
1907 * schedule_hrtimeout_range_clock - sleep until timeout
1908 * @expires: timeout value (ktime_t)
1909 * @delta: slack in expires timeout (ktime_t)
1910 * @mode: timer mode
1911 * @clock_id: timer clock to be used
1912 */
1913int __sched
1914schedule_hrtimeout_range_clock(ktime_t *expires, u64 delta,
1915 const enum hrtimer_mode mode, clockid_t clock_id)
1916{
1917 struct hrtimer_sleeper t;
1918
1919 /*
1920 * Optimize when a zero timeout value is given. It does not
1921 * matter whether this is an absolute or a relative time.
1922 */
1923 if (expires && *expires == 0) {
1924 __set_current_state(TASK_RUNNING);
1925 return 0;
1926 }
1927
1928 /*
1929 * A NULL parameter means "infinite"
1930 */
1931 if (!expires) {
1932 schedule();
1933 return -EINTR;
1934 }
1935
1936 hrtimer_init_on_stack(&t.timer, clock_id, mode);
1937 hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
1938
1939 hrtimer_init_sleeper(&t, current);
1940
1941 hrtimer_start_expires(&t.timer, mode);
1942
1943 if (likely(t.task))
1944 schedule();
1945
1946 hrtimer_cancel(&t.timer);
1947 destroy_hrtimer_on_stack(&t.timer);
1948
1949 __set_current_state(TASK_RUNNING);
1950
1951 return !t.task ? 0 : -EINTR;
1952}
1953
1954/**
1955 * schedule_hrtimeout_range - sleep until timeout
1956 * @expires: timeout value (ktime_t)
1957 * @delta: slack in expires timeout (ktime_t)
1958 * @mode: timer mode
1959 *
1960 * Make the current task sleep until the given expiry time has
1961 * elapsed. The routine will return immediately unless
1962 * the current task state has been set (see set_current_state()).
1963 *
1964 * The @delta argument gives the kernel the freedom to schedule the
1965 * actual wakeup to a time that is both power and performance friendly.
1966 * The kernel give the normal best effort behavior for "@expires+@delta",
1967 * but may decide to fire the timer earlier, but no earlier than @expires.
1968 *
1969 * You can set the task state as follows -
1970 *
1971 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1972 * pass before the routine returns unless the current task is explicitly
1973 * woken up, (e.g. by wake_up_process()).
1974 *
1975 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1976 * delivered to the current task or the current task is explicitly woken
1977 * up.
1978 *
1979 * The current task state is guaranteed to be TASK_RUNNING when this
1980 * routine returns.
1981 *
1982 * Returns 0 when the timer has expired. If the task was woken before the
1983 * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or
1984 * by an explicit wakeup, it returns -EINTR.
1985 */
1986int __sched schedule_hrtimeout_range(ktime_t *expires, u64 delta,
1987 const enum hrtimer_mode mode)
1988{
1989 return schedule_hrtimeout_range_clock(expires, delta, mode,
1990 CLOCK_MONOTONIC);
1991}
1992EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
1993
1994/**
1995 * schedule_hrtimeout - sleep until timeout
1996 * @expires: timeout value (ktime_t)
1997 * @mode: timer mode
1998 *
1999 * Make the current task sleep until the given expiry time has
2000 * elapsed. The routine will return immediately unless
2001 * the current task state has been set (see set_current_state()).
2002 *
2003 * You can set the task state as follows -
2004 *
2005 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
2006 * pass before the routine returns unless the current task is explicitly
2007 * woken up, (e.g. by wake_up_process()).
2008 *
2009 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
2010 * delivered to the current task or the current task is explicitly woken
2011 * up.
2012 *
2013 * The current task state is guaranteed to be TASK_RUNNING when this
2014 * routine returns.
2015 *
2016 * Returns 0 when the timer has expired. If the task was woken before the
2017 * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or
2018 * by an explicit wakeup, it returns -EINTR.
2019 */
2020int __sched schedule_hrtimeout(ktime_t *expires,
2021 const enum hrtimer_mode mode)
2022{
2023 return schedule_hrtimeout_range(expires, 0, mode);
2024}
2025EXPORT_SYMBOL_GPL(schedule_hrtimeout);