projects/linux·blob v4.16 · refs

1976 lines · 52.6 KiB

1/*
2 * linux/kernel/hrtimer.c
3 *
4 * Copyright(C) 2005-2006, Thomas Gleixner <[email protected]>
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 unsigned int active,
494 ktime_t expires_next)
495{
496 struct hrtimer_clock_base *base;
497 ktime_t expires;
498
499 for_each_active_base(base, cpu_base, active) {
500 struct timerqueue_node *next;
501 struct hrtimer *timer;
502
503 next = timerqueue_getnext(&base->active);
504 timer = container_of(next, struct hrtimer, node);
505 expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
506 if (expires < expires_next) {
507 expires_next = expires;
508 if (timer->is_soft)
509 cpu_base->softirq_next_timer = timer;
510 else
511 cpu_base->next_timer = timer;
512 }
513 }
514 /*
515 * clock_was_set() might have changed base->offset of any of
516 * the clock bases so the result might be negative. Fix it up
517 * to prevent a false positive in clockevents_program_event().
518 */
519 if (expires_next < 0)
520 expires_next = 0;
521 return expires_next;
522}
523
524/*
525 * Recomputes cpu_base::*next_timer and returns the earliest expires_next but
526 * does not set cpu_base::*expires_next, that is done by hrtimer_reprogram.
527 *
528 * When a softirq is pending, we can ignore the HRTIMER_ACTIVE_SOFT bases,
529 * those timers will get run whenever the softirq gets handled, at the end of
530 * hrtimer_run_softirq(), hrtimer_update_softirq_timer() will re-add these bases.
531 *
532 * Therefore softirq values are those from the HRTIMER_ACTIVE_SOFT clock bases.
533 * The !softirq values are the minima across HRTIMER_ACTIVE_ALL, unless an actual
534 * softirq is pending, in which case they're the minima of HRTIMER_ACTIVE_HARD.
535 *
536 * @active_mask must be one of:
537 * - HRTIMER_ACTIVE_ALL,
538 * - HRTIMER_ACTIVE_SOFT, or
539 * - HRTIMER_ACTIVE_HARD.
540 */
541static ktime_t
542__hrtimer_get_next_event(struct hrtimer_cpu_base *cpu_base, unsigned int active_mask)
543{
544 unsigned int active;
545 struct hrtimer *next_timer = NULL;
546 ktime_t expires_next = KTIME_MAX;
547
548 if (!cpu_base->softirq_activated && (active_mask & HRTIMER_ACTIVE_SOFT)) {
549 active = cpu_base->active_bases & HRTIMER_ACTIVE_SOFT;
550 cpu_base->softirq_next_timer = NULL;
551 expires_next = __hrtimer_next_event_base(cpu_base, active, KTIME_MAX);
552
553 next_timer = cpu_base->softirq_next_timer;
554 }
555
556 if (active_mask & HRTIMER_ACTIVE_HARD) {
557 active = cpu_base->active_bases & HRTIMER_ACTIVE_HARD;
558 cpu_base->next_timer = next_timer;
559 expires_next = __hrtimer_next_event_base(cpu_base, active, expires_next);
560 }
561
562 return expires_next;
563}
564
565static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base)
566{
567 ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset;
568 ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset;
569 ktime_t *offs_tai = &base->clock_base[HRTIMER_BASE_TAI].offset;
570
571 ktime_t now = ktime_get_update_offsets_now(&base->clock_was_set_seq,
572 offs_real, offs_boot, offs_tai);
573
574 base->clock_base[HRTIMER_BASE_REALTIME_SOFT].offset = *offs_real;
575 base->clock_base[HRTIMER_BASE_BOOTTIME_SOFT].offset = *offs_boot;
576 base->clock_base[HRTIMER_BASE_TAI_SOFT].offset = *offs_tai;
577
578 return now;
579}
580
581/*
582 * Is the high resolution mode active ?
583 */
584static inline int __hrtimer_hres_active(struct hrtimer_cpu_base *cpu_base)
585{
586 return IS_ENABLED(CONFIG_HIGH_RES_TIMERS) ?
587 cpu_base->hres_active : 0;
588}
589
590static inline int hrtimer_hres_active(void)
591{
592 return __hrtimer_hres_active(this_cpu_ptr(&hrtimer_bases));
593}
594
595/*
596 * Reprogram the event source with checking both queues for the
597 * next event
598 * Called with interrupts disabled and base->lock held
599 */
600static void
601hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
602{
603 ktime_t expires_next;
604
605 /*
606 * Find the current next expiration time.
607 */
608 expires_next = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_ALL);
609
610 if (cpu_base->next_timer && cpu_base->next_timer->is_soft) {
611 /*
612 * When the softirq is activated, hrtimer has to be
613 * programmed with the first hard hrtimer because soft
614 * timer interrupt could occur too late.
615 */
616 if (cpu_base->softirq_activated)
617 expires_next = __hrtimer_get_next_event(cpu_base,
618 HRTIMER_ACTIVE_HARD);
619 else
620 cpu_base->softirq_expires_next = expires_next;
621 }
622
623 if (skip_equal && expires_next == cpu_base->expires_next)
624 return;
625
626 cpu_base->expires_next = expires_next;
627
628 /*
629 * If hres is not active, hardware does not have to be
630 * reprogrammed yet.
631 *
632 * If a hang was detected in the last timer interrupt then we
633 * leave the hang delay active in the hardware. We want the
634 * system to make progress. That also prevents the following
635 * scenario:
636 * T1 expires 50ms from now
637 * T2 expires 5s from now
638 *
639 * T1 is removed, so this code is called and would reprogram
640 * the hardware to 5s from now. Any hrtimer_start after that
641 * will not reprogram the hardware due to hang_detected being
642 * set. So we'd effectivly block all timers until the T2 event
643 * fires.
644 */
645 if (!__hrtimer_hres_active(cpu_base) || cpu_base->hang_detected)
646 return;
647
648 tick_program_event(cpu_base->expires_next, 1);
649}
650
651/* High resolution timer related functions */
652#ifdef CONFIG_HIGH_RES_TIMERS
653
654/*
655 * High resolution timer enabled ?
656 */
657static bool hrtimer_hres_enabled __read_mostly = true;
658unsigned int hrtimer_resolution __read_mostly = LOW_RES_NSEC;
659EXPORT_SYMBOL_GPL(hrtimer_resolution);
660
661/*
662 * Enable / Disable high resolution mode
663 */
664static int __init setup_hrtimer_hres(char *str)
665{
666 return (kstrtobool(str, &hrtimer_hres_enabled) == 0);
667}
668
669__setup("highres=", setup_hrtimer_hres);
670
671/*
672 * hrtimer_high_res_enabled - query, if the highres mode is enabled
673 */
674static inline int hrtimer_is_hres_enabled(void)
675{
676 return hrtimer_hres_enabled;
677}
678
679/*
680 * Retrigger next event is called after clock was set
681 *
682 * Called with interrupts disabled via on_each_cpu()
683 */
684static void retrigger_next_event(void *arg)
685{
686 struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
687
688 if (!__hrtimer_hres_active(base))
689 return;
690
691 raw_spin_lock(&base->lock);
692 hrtimer_update_base(base);
693 hrtimer_force_reprogram(base, 0);
694 raw_spin_unlock(&base->lock);
695}
696
697/*
698 * Switch to high resolution mode
699 */
700static void hrtimer_switch_to_hres(void)
701{
702 struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
703
704 if (tick_init_highres()) {
705 printk(KERN_WARNING "Could not switch to high resolution "
706 "mode on CPU %d\n", base->cpu);
707 return;
708 }
709 base->hres_active = 1;
710 hrtimer_resolution = HIGH_RES_NSEC;
711
712 tick_setup_sched_timer();
713 /* "Retrigger" the interrupt to get things going */
714 retrigger_next_event(NULL);
715}
716
717static void clock_was_set_work(struct work_struct *work)
718{
719 clock_was_set();
720}
721
722static DECLARE_WORK(hrtimer_work, clock_was_set_work);
723
724/*
725 * Called from timekeeping and resume code to reprogram the hrtimer
726 * interrupt device on all cpus.
727 */
728void clock_was_set_delayed(void)
729{
730 schedule_work(&hrtimer_work);
731}
732
733#else
734
735static inline int hrtimer_is_hres_enabled(void) { return 0; }
736static inline void hrtimer_switch_to_hres(void) { }
737static inline void retrigger_next_event(void *arg) { }
738
739#endif /* CONFIG_HIGH_RES_TIMERS */
740
741/*
742 * When a timer is enqueued and expires earlier than the already enqueued
743 * timers, we have to check, whether it expires earlier than the timer for
744 * which the clock event device was armed.
745 *
746 * Called with interrupts disabled and base->cpu_base.lock held
747 */
748static void hrtimer_reprogram(struct hrtimer *timer, bool reprogram)
749{
750 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
751 struct hrtimer_clock_base *base = timer->base;
752 ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
753
754 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
755
756 /*
757 * CLOCK_REALTIME timer might be requested with an absolute
758 * expiry time which is less than base->offset. Set it to 0.
759 */
760 if (expires < 0)
761 expires = 0;
762
763 if (timer->is_soft) {
764 /*
765 * soft hrtimer could be started on a remote CPU. In this
766 * case softirq_expires_next needs to be updated on the
767 * remote CPU. The soft hrtimer will not expire before the
768 * first hard hrtimer on the remote CPU -
769 * hrtimer_check_target() prevents this case.
770 */
771 struct hrtimer_cpu_base *timer_cpu_base = base->cpu_base;
772
773 if (timer_cpu_base->softirq_activated)
774 return;
775
776 if (!ktime_before(expires, timer_cpu_base->softirq_expires_next))
777 return;
778
779 timer_cpu_base->softirq_next_timer = timer;
780 timer_cpu_base->softirq_expires_next = expires;
781
782 if (!ktime_before(expires, timer_cpu_base->expires_next) ||
783 !reprogram)
784 return;
785 }
786
787 /*
788 * If the timer is not on the current cpu, we cannot reprogram
789 * the other cpus clock event device.
790 */
791 if (base->cpu_base != cpu_base)
792 return;
793
794 /*
795 * If the hrtimer interrupt is running, then it will
796 * reevaluate the clock bases and reprogram the clock event
797 * device. The callbacks are always executed in hard interrupt
798 * context so we don't need an extra check for a running
799 * callback.
800 */
801 if (cpu_base->in_hrtirq)
802 return;
803
804 if (expires >= cpu_base->expires_next)
805 return;
806
807 /* Update the pointer to the next expiring timer */
808 cpu_base->next_timer = timer;
809 cpu_base->expires_next = expires;
810
811 /*
812 * If hres is not active, hardware does not have to be
813 * programmed yet.
814 *
815 * If a hang was detected in the last timer interrupt then we
816 * do not schedule a timer which is earlier than the expiry
817 * which we enforced in the hang detection. We want the system
818 * to make progress.
819 */
820 if (!__hrtimer_hres_active(cpu_base) || cpu_base->hang_detected)
821 return;
822
823 /*
824 * Program the timer hardware. We enforce the expiry for
825 * events which are already in the past.
826 */
827 tick_program_event(expires, 1);
828}
829
830/*
831 * Clock realtime was set
832 *
833 * Change the offset of the realtime clock vs. the monotonic
834 * clock.
835 *
836 * We might have to reprogram the high resolution timer interrupt. On
837 * SMP we call the architecture specific code to retrigger _all_ high
838 * resolution timer interrupts. On UP we just disable interrupts and
839 * call the high resolution interrupt code.
840 */
841void clock_was_set(void)
842{
843#ifdef CONFIG_HIGH_RES_TIMERS
844 /* Retrigger the CPU local events everywhere */
845 on_each_cpu(retrigger_next_event, NULL, 1);
846#endif
847 timerfd_clock_was_set();
848}
849
850/*
851 * During resume we might have to reprogram the high resolution timer
852 * interrupt on all online CPUs. However, all other CPUs will be
853 * stopped with IRQs interrupts disabled so the clock_was_set() call
854 * must be deferred.
855 */
856void hrtimers_resume(void)
857{
858 lockdep_assert_irqs_disabled();
859 /* Retrigger on the local CPU */
860 retrigger_next_event(NULL);
861 /* And schedule a retrigger for all others */
862 clock_was_set_delayed();
863}
864
865/*
866 * Counterpart to lock_hrtimer_base above:
867 */
868static inline
869void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
870{
871 raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
872}
873
874/**
875 * hrtimer_forward - forward the timer expiry
876 * @timer: hrtimer to forward
877 * @now: forward past this time
878 * @interval: the interval to forward
879 *
880 * Forward the timer expiry so it will expire in the future.
881 * Returns the number of overruns.
882 *
883 * Can be safely called from the callback function of @timer. If
884 * called from other contexts @timer must neither be enqueued nor
885 * running the callback and the caller needs to take care of
886 * serialization.
887 *
888 * Note: This only updates the timer expiry value and does not requeue
889 * the timer.
890 */
891u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
892{
893 u64 orun = 1;
894 ktime_t delta;
895
896 delta = ktime_sub(now, hrtimer_get_expires(timer));
897
898 if (delta < 0)
899 return 0;
900
901 if (WARN_ON(timer->state & HRTIMER_STATE_ENQUEUED))
902 return 0;
903
904 if (interval < hrtimer_resolution)
905 interval = hrtimer_resolution;
906
907 if (unlikely(delta >= interval)) {
908 s64 incr = ktime_to_ns(interval);
909
910 orun = ktime_divns(delta, incr);
911 hrtimer_add_expires_ns(timer, incr * orun);
912 if (hrtimer_get_expires_tv64(timer) > now)
913 return orun;
914 /*
915 * This (and the ktime_add() below) is the
916 * correction for exact:
917 */
918 orun++;
919 }
920 hrtimer_add_expires(timer, interval);
921
922 return orun;
923}
924EXPORT_SYMBOL_GPL(hrtimer_forward);
925
926/*
927 * enqueue_hrtimer - internal function to (re)start a timer
928 *
929 * The timer is inserted in expiry order. Insertion into the
930 * red black tree is O(log(n)). Must hold the base lock.
931 *
932 * Returns 1 when the new timer is the leftmost timer in the tree.
933 */
934static int enqueue_hrtimer(struct hrtimer *timer,
935 struct hrtimer_clock_base *base,
936 enum hrtimer_mode mode)
937{
938 debug_activate(timer, mode);
939
940 base->cpu_base->active_bases |= 1 << base->index;
941
942 timer->state = HRTIMER_STATE_ENQUEUED;
943
944 return timerqueue_add(&base->active, &timer->node);
945}
946
947/*
948 * __remove_hrtimer - internal function to remove a timer
949 *
950 * Caller must hold the base lock.
951 *
952 * High resolution timer mode reprograms the clock event device when the
953 * timer is the one which expires next. The caller can disable this by setting
954 * reprogram to zero. This is useful, when the context does a reprogramming
955 * anyway (e.g. timer interrupt)
956 */
957static void __remove_hrtimer(struct hrtimer *timer,
958 struct hrtimer_clock_base *base,
959 u8 newstate, int reprogram)
960{
961 struct hrtimer_cpu_base *cpu_base = base->cpu_base;
962 u8 state = timer->state;
963
964 timer->state = newstate;
965 if (!(state & HRTIMER_STATE_ENQUEUED))
966 return;
967
968 if (!timerqueue_del(&base->active, &timer->node))
969 cpu_base->active_bases &= ~(1 << base->index);
970
971 /*
972 * Note: If reprogram is false we do not update
973 * cpu_base->next_timer. This happens when we remove the first
974 * timer on a remote cpu. No harm as we never dereference
975 * cpu_base->next_timer. So the worst thing what can happen is
976 * an superflous call to hrtimer_force_reprogram() on the
977 * remote cpu later on if the same timer gets enqueued again.
978 */
979 if (reprogram && timer == cpu_base->next_timer)
980 hrtimer_force_reprogram(cpu_base, 1);
981}
982
983/*
984 * remove hrtimer, called with base lock held
985 */
986static inline int
987remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base, bool restart)
988{
989 if (hrtimer_is_queued(timer)) {
990 u8 state = timer->state;
991 int reprogram;
992
993 /*
994 * Remove the timer and force reprogramming when high
995 * resolution mode is active and the timer is on the current
996 * CPU. If we remove a timer on another CPU, reprogramming is
997 * skipped. The interrupt event on this CPU is fired and
998 * reprogramming happens in the interrupt handler. This is a
999 * rare case and less expensive than a smp call.
1000 */
1001 debug_deactivate(timer);
1002 reprogram = base->cpu_base == this_cpu_ptr(&hrtimer_bases);
1003
1004 if (!restart)
1005 state = HRTIMER_STATE_INACTIVE;
1006
1007 __remove_hrtimer(timer, base, state, reprogram);
1008 return 1;
1009 }
1010 return 0;
1011}
1012
1013static inline ktime_t hrtimer_update_lowres(struct hrtimer *timer, ktime_t tim,
1014 const enum hrtimer_mode mode)
1015{
1016#ifdef CONFIG_TIME_LOW_RES
1017 /*
1018 * CONFIG_TIME_LOW_RES indicates that the system has no way to return
1019 * granular time values. For relative timers we add hrtimer_resolution
1020 * (i.e. one jiffie) to prevent short timeouts.
1021 */
1022 timer->is_rel = mode & HRTIMER_MODE_REL;
1023 if (timer->is_rel)
1024 tim = ktime_add_safe(tim, hrtimer_resolution);
1025#endif
1026 return tim;
1027}
1028
1029static void
1030hrtimer_update_softirq_timer(struct hrtimer_cpu_base *cpu_base, bool reprogram)
1031{
1032 ktime_t expires;
1033
1034 /*
1035 * Find the next SOFT expiration.
1036 */
1037 expires = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_SOFT);
1038
1039 /*
1040 * reprogramming needs to be triggered, even if the next soft
1041 * hrtimer expires at the same time than the next hard
1042 * hrtimer. cpu_base->softirq_expires_next needs to be updated!
1043 */
1044 if (expires == KTIME_MAX)
1045 return;
1046
1047 /*
1048 * cpu_base->*next_timer is recomputed by __hrtimer_get_next_event()
1049 * cpu_base->*expires_next is only set by hrtimer_reprogram()
1050 */
1051 hrtimer_reprogram(cpu_base->softirq_next_timer, reprogram);
1052}
1053
1054static int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1055 u64 delta_ns, const enum hrtimer_mode mode,
1056 struct hrtimer_clock_base *base)
1057{
1058 struct hrtimer_clock_base *new_base;
1059
1060 /* Remove an active timer from the queue: */
1061 remove_hrtimer(timer, base, true);
1062
1063 if (mode & HRTIMER_MODE_REL)
1064 tim = ktime_add_safe(tim, base->get_time());
1065
1066 tim = hrtimer_update_lowres(timer, tim, mode);
1067
1068 hrtimer_set_expires_range_ns(timer, tim, delta_ns);
1069
1070 /* Switch the timer base, if necessary: */
1071 new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED);
1072
1073 return enqueue_hrtimer(timer, new_base, mode);
1074}
1075
1076/**
1077 * hrtimer_start_range_ns - (re)start an hrtimer
1078 * @timer: the timer to be added
1079 * @tim: expiry time
1080 * @delta_ns: "slack" range for the timer
1081 * @mode: timer mode: absolute (HRTIMER_MODE_ABS) or
1082 * relative (HRTIMER_MODE_REL), and pinned (HRTIMER_MODE_PINNED);
1083 * softirq based mode is considered for debug purpose only!
1084 */
1085void hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1086 u64 delta_ns, const enum hrtimer_mode mode)
1087{
1088 struct hrtimer_clock_base *base;
1089 unsigned long flags;
1090
1091 /*
1092 * Check whether the HRTIMER_MODE_SOFT bit and hrtimer.is_soft
1093 * match.
1094 */
1095 WARN_ON_ONCE(!(mode & HRTIMER_MODE_SOFT) ^ !timer->is_soft);
1096
1097 base = lock_hrtimer_base(timer, &flags);
1098
1099 if (__hrtimer_start_range_ns(timer, tim, delta_ns, mode, base))
1100 hrtimer_reprogram(timer, true);
1101
1102 unlock_hrtimer_base(timer, &flags);
1103}
1104EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
1105
1106/**
1107 * hrtimer_try_to_cancel - try to deactivate a timer
1108 * @timer: hrtimer to stop
1109 *
1110 * Returns:
1111 * 0 when the timer was not active
1112 * 1 when the timer was active
1113 * -1 when the timer is currently executing the callback function and
1114 * cannot be stopped
1115 */
1116int hrtimer_try_to_cancel(struct hrtimer *timer)
1117{
1118 struct hrtimer_clock_base *base;
1119 unsigned long flags;
1120 int ret = -1;
1121
1122 /*
1123 * Check lockless first. If the timer is not active (neither
1124 * enqueued nor running the callback, nothing to do here. The
1125 * base lock does not serialize against a concurrent enqueue,
1126 * so we can avoid taking it.
1127 */
1128 if (!hrtimer_active(timer))
1129 return 0;
1130
1131 base = lock_hrtimer_base(timer, &flags);
1132
1133 if (!hrtimer_callback_running(timer))
1134 ret = remove_hrtimer(timer, base, false);
1135
1136 unlock_hrtimer_base(timer, &flags);
1137
1138 return ret;
1139
1140}
1141EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1142
1143/**
1144 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1145 * @timer: the timer to be cancelled
1146 *
1147 * Returns:
1148 * 0 when the timer was not active
1149 * 1 when the timer was active
1150 */
1151int hrtimer_cancel(struct hrtimer *timer)
1152{
1153 for (;;) {
1154 int ret = hrtimer_try_to_cancel(timer);
1155
1156 if (ret >= 0)
1157 return ret;
1158 cpu_relax();
1159 }
1160}
1161EXPORT_SYMBOL_GPL(hrtimer_cancel);
1162
1163/**
1164 * hrtimer_get_remaining - get remaining time for the timer
1165 * @timer: the timer to read
1166 * @adjust: adjust relative timers when CONFIG_TIME_LOW_RES=y
1167 */
1168ktime_t __hrtimer_get_remaining(const struct hrtimer *timer, bool adjust)
1169{
1170 unsigned long flags;
1171 ktime_t rem;
1172
1173 lock_hrtimer_base(timer, &flags);
1174 if (IS_ENABLED(CONFIG_TIME_LOW_RES) && adjust)
1175 rem = hrtimer_expires_remaining_adjusted(timer);
1176 else
1177 rem = hrtimer_expires_remaining(timer);
1178 unlock_hrtimer_base(timer, &flags);
1179
1180 return rem;
1181}
1182EXPORT_SYMBOL_GPL(__hrtimer_get_remaining);
1183
1184#ifdef CONFIG_NO_HZ_COMMON
1185/**
1186 * hrtimer_get_next_event - get the time until next expiry event
1187 *
1188 * Returns the next expiry time or KTIME_MAX if no timer is pending.
1189 */
1190u64 hrtimer_get_next_event(void)
1191{
1192 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1193 u64 expires = KTIME_MAX;
1194 unsigned long flags;
1195
1196 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1197
1198 if (!__hrtimer_hres_active(cpu_base))
1199 expires = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_ALL);
1200
1201 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1202
1203 return expires;
1204}
1205#endif
1206
1207static inline int hrtimer_clockid_to_base(clockid_t clock_id)
1208{
1209 if (likely(clock_id < MAX_CLOCKS)) {
1210 int base = hrtimer_clock_to_base_table[clock_id];
1211
1212 if (likely(base != HRTIMER_MAX_CLOCK_BASES))
1213 return base;
1214 }
1215 WARN(1, "Invalid clockid %d. Using MONOTONIC\n", clock_id);
1216 return HRTIMER_BASE_MONOTONIC;
1217}
1218
1219static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1220 enum hrtimer_mode mode)
1221{
1222 bool softtimer = !!(mode & HRTIMER_MODE_SOFT);
1223 int base = softtimer ? HRTIMER_MAX_CLOCK_BASES / 2 : 0;
1224 struct hrtimer_cpu_base *cpu_base;
1225
1226 memset(timer, 0, sizeof(struct hrtimer));
1227
1228 cpu_base = raw_cpu_ptr(&hrtimer_bases);
1229
1230 /*
1231 * POSIX magic: Relative CLOCK_REALTIME timers are not affected by
1232 * clock modifications, so they needs to become CLOCK_MONOTONIC to
1233 * ensure POSIX compliance.
1234 */
1235 if (clock_id == CLOCK_REALTIME && mode & HRTIMER_MODE_REL)
1236 clock_id = CLOCK_MONOTONIC;
1237
1238 base += hrtimer_clockid_to_base(clock_id);
1239 timer->is_soft = softtimer;
1240 timer->base = &cpu_base->clock_base[base];
1241 timerqueue_init(&timer->node);
1242}
1243
1244/**
1245 * hrtimer_init - initialize a timer to the given clock
1246 * @timer: the timer to be initialized
1247 * @clock_id: the clock to be used
1248 * @mode: The modes which are relevant for intitialization:
1249 * HRTIMER_MODE_ABS, HRTIMER_MODE_REL, HRTIMER_MODE_ABS_SOFT,
1250 * HRTIMER_MODE_REL_SOFT
1251 *
1252 * The PINNED variants of the above can be handed in,
1253 * but the PINNED bit is ignored as pinning happens
1254 * when the hrtimer is started
1255 */
1256void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1257 enum hrtimer_mode mode)
1258{
1259 debug_init(timer, clock_id, mode);
1260 __hrtimer_init(timer, clock_id, mode);
1261}
1262EXPORT_SYMBOL_GPL(hrtimer_init);
1263
1264/*
1265 * A timer is active, when it is enqueued into the rbtree or the
1266 * callback function is running or it's in the state of being migrated
1267 * to another cpu.
1268 *
1269 * It is important for this function to not return a false negative.
1270 */
1271bool hrtimer_active(const struct hrtimer *timer)
1272{
1273 struct hrtimer_clock_base *base;
1274 unsigned int seq;
1275
1276 do {
1277 base = READ_ONCE(timer->base);
1278 seq = raw_read_seqcount_begin(&base->seq);
1279
1280 if (timer->state != HRTIMER_STATE_INACTIVE ||
1281 base->running == timer)
1282 return true;
1283
1284 } while (read_seqcount_retry(&base->seq, seq) ||
1285 base != READ_ONCE(timer->base));
1286
1287 return false;
1288}
1289EXPORT_SYMBOL_GPL(hrtimer_active);
1290
1291/*
1292 * The write_seqcount_barrier()s in __run_hrtimer() split the thing into 3
1293 * distinct sections:
1294 *
1295 * - queued: the timer is queued
1296 * - callback: the timer is being ran
1297 * - post: the timer is inactive or (re)queued
1298 *
1299 * On the read side we ensure we observe timer->state and cpu_base->running
1300 * from the same section, if anything changed while we looked at it, we retry.
1301 * This includes timer->base changing because sequence numbers alone are
1302 * insufficient for that.
1303 *
1304 * The sequence numbers are required because otherwise we could still observe
1305 * a false negative if the read side got smeared over multiple consequtive
1306 * __run_hrtimer() invocations.
1307 */
1308
1309static void __run_hrtimer(struct hrtimer_cpu_base *cpu_base,
1310 struct hrtimer_clock_base *base,
1311 struct hrtimer *timer, ktime_t *now,
1312 unsigned long flags)
1313{
1314 enum hrtimer_restart (*fn)(struct hrtimer *);
1315 int restart;
1316
1317 lockdep_assert_held(&cpu_base->lock);
1318
1319 debug_deactivate(timer);
1320 base->running = timer;
1321
1322 /*
1323 * Separate the ->running assignment from the ->state assignment.
1324 *
1325 * As with a regular write barrier, this ensures the read side in
1326 * hrtimer_active() cannot observe base->running == NULL &&
1327 * timer->state == INACTIVE.
1328 */
1329 raw_write_seqcount_barrier(&base->seq);
1330
1331 __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE, 0);
1332 fn = timer->function;
1333
1334 /*
1335 * Clear the 'is relative' flag for the TIME_LOW_RES case. If the
1336 * timer is restarted with a period then it becomes an absolute
1337 * timer. If its not restarted it does not matter.
1338 */
1339 if (IS_ENABLED(CONFIG_TIME_LOW_RES))
1340 timer->is_rel = false;
1341
1342 /*
1343 * The timer is marked as running in the CPU base, so it is
1344 * protected against migration to a different CPU even if the lock
1345 * is dropped.
1346 */
1347 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1348 trace_hrtimer_expire_entry(timer, now);
1349 restart = fn(timer);
1350 trace_hrtimer_expire_exit(timer);
1351 raw_spin_lock_irq(&cpu_base->lock);
1352
1353 /*
1354 * Note: We clear the running state after enqueue_hrtimer and
1355 * we do not reprogram the event hardware. Happens either in
1356 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1357 *
1358 * Note: Because we dropped the cpu_base->lock above,
1359 * hrtimer_start_range_ns() can have popped in and enqueued the timer
1360 * for us already.
1361 */
1362 if (restart != HRTIMER_NORESTART &&
1363 !(timer->state & HRTIMER_STATE_ENQUEUED))
1364 enqueue_hrtimer(timer, base, HRTIMER_MODE_ABS);
1365
1366 /*
1367 * Separate the ->running assignment from the ->state assignment.
1368 *
1369 * As with a regular write barrier, this ensures the read side in
1370 * hrtimer_active() cannot observe base->running.timer == NULL &&
1371 * timer->state == INACTIVE.
1372 */
1373 raw_write_seqcount_barrier(&base->seq);
1374
1375 WARN_ON_ONCE(base->running != timer);
1376 base->running = NULL;
1377}
1378
1379static void __hrtimer_run_queues(struct hrtimer_cpu_base *cpu_base, ktime_t now,
1380 unsigned long flags, unsigned int active_mask)
1381{
1382 struct hrtimer_clock_base *base;
1383 unsigned int active = cpu_base->active_bases & active_mask;
1384
1385 for_each_active_base(base, cpu_base, active) {
1386 struct timerqueue_node *node;
1387 ktime_t basenow;
1388
1389 basenow = ktime_add(now, base->offset);
1390
1391 while ((node = timerqueue_getnext(&base->active))) {
1392 struct hrtimer *timer;
1393
1394 timer = container_of(node, struct hrtimer, node);
1395
1396 /*
1397 * The immediate goal for using the softexpires is
1398 * minimizing wakeups, not running timers at the
1399 * earliest interrupt after their soft expiration.
1400 * This allows us to avoid using a Priority Search
1401 * Tree, which can answer a stabbing querry for
1402 * overlapping intervals and instead use the simple
1403 * BST we already have.
1404 * We don't add extra wakeups by delaying timers that
1405 * are right-of a not yet expired timer, because that
1406 * timer will have to trigger a wakeup anyway.
1407 */
1408 if (basenow < hrtimer_get_softexpires_tv64(timer))
1409 break;
1410
1411 __run_hrtimer(cpu_base, base, timer, &basenow, flags);
1412 }
1413 }
1414}
1415
1416static __latent_entropy void hrtimer_run_softirq(struct softirq_action *h)
1417{
1418 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1419 unsigned long flags;
1420 ktime_t now;
1421
1422 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1423
1424 now = hrtimer_update_base(cpu_base);
1425 __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_SOFT);
1426
1427 cpu_base->softirq_activated = 0;
1428 hrtimer_update_softirq_timer(cpu_base, true);
1429
1430 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1431}
1432
1433#ifdef CONFIG_HIGH_RES_TIMERS
1434
1435/*
1436 * High resolution timer interrupt
1437 * Called with interrupts disabled
1438 */
1439void hrtimer_interrupt(struct clock_event_device *dev)
1440{
1441 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1442 ktime_t expires_next, now, entry_time, delta;
1443 unsigned long flags;
1444 int retries = 0;
1445
1446 BUG_ON(!cpu_base->hres_active);
1447 cpu_base->nr_events++;
1448 dev->next_event = KTIME_MAX;
1449
1450 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1451 entry_time = now = hrtimer_update_base(cpu_base);
1452retry:
1453 cpu_base->in_hrtirq = 1;
1454 /*
1455 * We set expires_next to KTIME_MAX here with cpu_base->lock
1456 * held to prevent that a timer is enqueued in our queue via
1457 * the migration code. This does not affect enqueueing of
1458 * timers which run their callback and need to be requeued on
1459 * this CPU.
1460 */
1461 cpu_base->expires_next = KTIME_MAX;
1462
1463 if (!ktime_before(now, cpu_base->softirq_expires_next)) {
1464 cpu_base->softirq_expires_next = KTIME_MAX;
1465 cpu_base->softirq_activated = 1;
1466 raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1467 }
1468
1469 __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_HARD);
1470
1471 /* Reevaluate the clock bases for the next expiry */
1472 expires_next = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_ALL);
1473 /*
1474 * Store the new expiry value so the migration code can verify
1475 * against it.
1476 */
1477 cpu_base->expires_next = expires_next;
1478 cpu_base->in_hrtirq = 0;
1479 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1480
1481 /* Reprogramming necessary ? */
1482 if (!tick_program_event(expires_next, 0)) {
1483 cpu_base->hang_detected = 0;
1484 return;
1485 }
1486
1487 /*
1488 * The next timer was already expired due to:
1489 * - tracing
1490 * - long lasting callbacks
1491 * - being scheduled away when running in a VM
1492 *
1493 * We need to prevent that we loop forever in the hrtimer
1494 * interrupt routine. We give it 3 attempts to avoid
1495 * overreacting on some spurious event.
1496 *
1497 * Acquire base lock for updating the offsets and retrieving
1498 * the current time.
1499 */
1500 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1501 now = hrtimer_update_base(cpu_base);
1502 cpu_base->nr_retries++;
1503 if (++retries < 3)
1504 goto retry;
1505 /*
1506 * Give the system a chance to do something else than looping
1507 * here. We stored the entry time, so we know exactly how long
1508 * we spent here. We schedule the next event this amount of
1509 * time away.
1510 */
1511 cpu_base->nr_hangs++;
1512 cpu_base->hang_detected = 1;
1513 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1514
1515 delta = ktime_sub(now, entry_time);
1516 if ((unsigned int)delta > cpu_base->max_hang_time)
1517 cpu_base->max_hang_time = (unsigned int) delta;
1518 /*
1519 * Limit it to a sensible value as we enforce a longer
1520 * delay. Give the CPU at least 100ms to catch up.
1521 */
1522 if (delta > 100 * NSEC_PER_MSEC)
1523 expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1524 else
1525 expires_next = ktime_add(now, delta);
1526 tick_program_event(expires_next, 1);
1527 printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n",
1528 ktime_to_ns(delta));
1529}
1530
1531/* called with interrupts disabled */
1532static inline void __hrtimer_peek_ahead_timers(void)
1533{
1534 struct tick_device *td;
1535
1536 if (!hrtimer_hres_active())
1537 return;
1538
1539 td = this_cpu_ptr(&tick_cpu_device);
1540 if (td && td->evtdev)
1541 hrtimer_interrupt(td->evtdev);
1542}
1543
1544#else /* CONFIG_HIGH_RES_TIMERS */
1545
1546static inline void __hrtimer_peek_ahead_timers(void) { }
1547
1548#endif /* !CONFIG_HIGH_RES_TIMERS */
1549
1550/*
1551 * Called from run_local_timers in hardirq context every jiffy
1552 */
1553void hrtimer_run_queues(void)
1554{
1555 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1556 unsigned long flags;
1557 ktime_t now;
1558
1559 if (__hrtimer_hres_active(cpu_base))
1560 return;
1561
1562 /*
1563 * This _is_ ugly: We have to check periodically, whether we
1564 * can switch to highres and / or nohz mode. The clocksource
1565 * switch happens with xtime_lock held. Notification from
1566 * there only sets the check bit in the tick_oneshot code,
1567 * otherwise we might deadlock vs. xtime_lock.
1568 */
1569 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) {
1570 hrtimer_switch_to_hres();
1571 return;
1572 }
1573
1574 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1575 now = hrtimer_update_base(cpu_base);
1576
1577 if (!ktime_before(now, cpu_base->softirq_expires_next)) {
1578 cpu_base->softirq_expires_next = KTIME_MAX;
1579 cpu_base->softirq_activated = 1;
1580 raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1581 }
1582
1583 __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_HARD);
1584 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1585}
1586
1587/*
1588 * Sleep related functions:
1589 */
1590static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1591{
1592 struct hrtimer_sleeper *t =
1593 container_of(timer, struct hrtimer_sleeper, timer);
1594 struct task_struct *task = t->task;
1595
1596 t->task = NULL;
1597 if (task)
1598 wake_up_process(task);
1599
1600 return HRTIMER_NORESTART;
1601}
1602
1603void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1604{
1605 sl->timer.function = hrtimer_wakeup;
1606 sl->task = task;
1607}
1608EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
1609
1610int nanosleep_copyout(struct restart_block *restart, struct timespec64 *ts)
1611{
1612 switch(restart->nanosleep.type) {
1613#ifdef CONFIG_COMPAT
1614 case TT_COMPAT:
1615 if (compat_put_timespec64(ts, restart->nanosleep.compat_rmtp))
1616 return -EFAULT;
1617 break;
1618#endif
1619 case TT_NATIVE:
1620 if (put_timespec64(ts, restart->nanosleep.rmtp))
1621 return -EFAULT;
1622 break;
1623 default:
1624 BUG();
1625 }
1626 return -ERESTART_RESTARTBLOCK;
1627}
1628
1629static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1630{
1631 struct restart_block *restart;
1632
1633 hrtimer_init_sleeper(t, current);
1634
1635 do {
1636 set_current_state(TASK_INTERRUPTIBLE);
1637 hrtimer_start_expires(&t->timer, mode);
1638
1639 if (likely(t->task))
1640 freezable_schedule();
1641
1642 hrtimer_cancel(&t->timer);
1643 mode = HRTIMER_MODE_ABS;
1644
1645 } while (t->task && !signal_pending(current));
1646
1647 __set_current_state(TASK_RUNNING);
1648
1649 if (!t->task)
1650 return 0;
1651
1652 restart = &current->restart_block;
1653 if (restart->nanosleep.type != TT_NONE) {
1654 ktime_t rem = hrtimer_expires_remaining(&t->timer);
1655 struct timespec64 rmt;
1656
1657 if (rem <= 0)
1658 return 0;
1659 rmt = ktime_to_timespec64(rem);
1660
1661 return nanosleep_copyout(restart, &rmt);
1662 }
1663 return -ERESTART_RESTARTBLOCK;
1664}
1665
1666static long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1667{
1668 struct hrtimer_sleeper t;
1669 int ret;
1670
1671 hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid,
1672 HRTIMER_MODE_ABS);
1673 hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
1674
1675 ret = do_nanosleep(&t, HRTIMER_MODE_ABS);
1676 destroy_hrtimer_on_stack(&t.timer);
1677 return ret;
1678}
1679
1680long hrtimer_nanosleep(const struct timespec64 *rqtp,
1681 const enum hrtimer_mode mode, const clockid_t clockid)
1682{
1683 struct restart_block *restart;
1684 struct hrtimer_sleeper t;
1685 int ret = 0;
1686 u64 slack;
1687
1688 slack = current->timer_slack_ns;
1689 if (dl_task(current) || rt_task(current))
1690 slack = 0;
1691
1692 hrtimer_init_on_stack(&t.timer, clockid, mode);
1693 hrtimer_set_expires_range_ns(&t.timer, timespec64_to_ktime(*rqtp), slack);
1694 ret = do_nanosleep(&t, mode);
1695 if (ret != -ERESTART_RESTARTBLOCK)
1696 goto out;
1697
1698 /* Absolute timers do not update the rmtp value and restart: */
1699 if (mode == HRTIMER_MODE_ABS) {
1700 ret = -ERESTARTNOHAND;
1701 goto out;
1702 }
1703
1704 restart = &current->restart_block;
1705 restart->fn = hrtimer_nanosleep_restart;
1706 restart->nanosleep.clockid = t.timer.base->clockid;
1707 restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
1708out:
1709 destroy_hrtimer_on_stack(&t.timer);
1710 return ret;
1711}
1712
1713SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp,
1714 struct timespec __user *, rmtp)
1715{
1716 struct timespec64 tu;
1717
1718 if (get_timespec64(&tu, rqtp))
1719 return -EFAULT;
1720
1721 if (!timespec64_valid(&tu))
1722 return -EINVAL;
1723
1724 current->restart_block.nanosleep.type = rmtp ? TT_NATIVE : TT_NONE;
1725 current->restart_block.nanosleep.rmtp = rmtp;
1726 return hrtimer_nanosleep(&tu, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1727}
1728
1729#ifdef CONFIG_COMPAT
1730
1731COMPAT_SYSCALL_DEFINE2(nanosleep, struct compat_timespec __user *, rqtp,
1732 struct compat_timespec __user *, rmtp)
1733{
1734 struct timespec64 tu;
1735
1736 if (compat_get_timespec64(&tu, rqtp))
1737 return -EFAULT;
1738
1739 if (!timespec64_valid(&tu))
1740 return -EINVAL;
1741
1742 current->restart_block.nanosleep.type = rmtp ? TT_COMPAT : TT_NONE;
1743 current->restart_block.nanosleep.compat_rmtp = rmtp;
1744 return hrtimer_nanosleep(&tu, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1745}
1746#endif
1747
1748/*
1749 * Functions related to boot-time initialization:
1750 */
1751int hrtimers_prepare_cpu(unsigned int cpu)
1752{
1753 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1754 int i;
1755
1756 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1757 cpu_base->clock_base[i].cpu_base = cpu_base;
1758 timerqueue_init_head(&cpu_base->clock_base[i].active);
1759 }
1760
1761 cpu_base->cpu = cpu;
1762 cpu_base->active_bases = 0;
1763 cpu_base->hres_active = 0;
1764 cpu_base->hang_detected = 0;
1765 cpu_base->next_timer = NULL;
1766 cpu_base->softirq_next_timer = NULL;
1767 cpu_base->expires_next = KTIME_MAX;
1768 cpu_base->softirq_expires_next = KTIME_MAX;
1769 return 0;
1770}
1771
1772#ifdef CONFIG_HOTPLUG_CPU
1773
1774static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1775 struct hrtimer_clock_base *new_base)
1776{
1777 struct hrtimer *timer;
1778 struct timerqueue_node *node;
1779
1780 while ((node = timerqueue_getnext(&old_base->active))) {
1781 timer = container_of(node, struct hrtimer, node);
1782 BUG_ON(hrtimer_callback_running(timer));
1783 debug_deactivate(timer);
1784
1785 /*
1786 * Mark it as ENQUEUED not INACTIVE otherwise the
1787 * timer could be seen as !active and just vanish away
1788 * under us on another CPU
1789 */
1790 __remove_hrtimer(timer, old_base, HRTIMER_STATE_ENQUEUED, 0);
1791 timer->base = new_base;
1792 /*
1793 * Enqueue the timers on the new cpu. This does not
1794 * reprogram the event device in case the timer
1795 * expires before the earliest on this CPU, but we run
1796 * hrtimer_interrupt after we migrated everything to
1797 * sort out already expired timers and reprogram the
1798 * event device.
1799 */
1800 enqueue_hrtimer(timer, new_base, HRTIMER_MODE_ABS);
1801 }
1802}
1803
1804int hrtimers_dead_cpu(unsigned int scpu)
1805{
1806 struct hrtimer_cpu_base *old_base, *new_base;
1807 int i;
1808
1809 BUG_ON(cpu_online(scpu));
1810 tick_cancel_sched_timer(scpu);
1811
1812 /*
1813 * this BH disable ensures that raise_softirq_irqoff() does
1814 * not wakeup ksoftirqd (and acquire the pi-lock) while
1815 * holding the cpu_base lock
1816 */
1817 local_bh_disable();
1818 local_irq_disable();
1819 old_base = &per_cpu(hrtimer_bases, scpu);
1820 new_base = this_cpu_ptr(&hrtimer_bases);
1821 /*
1822 * The caller is globally serialized and nobody else
1823 * takes two locks at once, deadlock is not possible.
1824 */
1825 raw_spin_lock(&new_base->lock);
1826 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1827
1828 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1829 migrate_hrtimer_list(&old_base->clock_base[i],
1830 &new_base->clock_base[i]);
1831 }
1832
1833 /*
1834 * The migration might have changed the first expiring softirq
1835 * timer on this CPU. Update it.
1836 */
1837 hrtimer_update_softirq_timer(new_base, false);
1838
1839 raw_spin_unlock(&old_base->lock);
1840 raw_spin_unlock(&new_base->lock);
1841
1842 /* Check, if we got expired work to do */
1843 __hrtimer_peek_ahead_timers();
1844 local_irq_enable();
1845 local_bh_enable();
1846 return 0;
1847}
1848
1849#endif /* CONFIG_HOTPLUG_CPU */
1850
1851void __init hrtimers_init(void)
1852{
1853 hrtimers_prepare_cpu(smp_processor_id());
1854 open_softirq(HRTIMER_SOFTIRQ, hrtimer_run_softirq);
1855}
1856
1857/**
1858 * schedule_hrtimeout_range_clock - sleep until timeout
1859 * @expires: timeout value (ktime_t)
1860 * @delta: slack in expires timeout (ktime_t)
1861 * @mode: timer mode
1862 * @clock_id: timer clock to be used
1863 */
1864int __sched
1865schedule_hrtimeout_range_clock(ktime_t *expires, u64 delta,
1866 const enum hrtimer_mode mode, clockid_t clock_id)
1867{
1868 struct hrtimer_sleeper t;
1869
1870 /*
1871 * Optimize when a zero timeout value is given. It does not
1872 * matter whether this is an absolute or a relative time.
1873 */
1874 if (expires && *expires == 0) {
1875 __set_current_state(TASK_RUNNING);
1876 return 0;
1877 }
1878
1879 /*
1880 * A NULL parameter means "infinite"
1881 */
1882 if (!expires) {
1883 schedule();
1884 return -EINTR;
1885 }
1886
1887 hrtimer_init_on_stack(&t.timer, clock_id, mode);
1888 hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
1889
1890 hrtimer_init_sleeper(&t, current);
1891
1892 hrtimer_start_expires(&t.timer, mode);
1893
1894 if (likely(t.task))
1895 schedule();
1896
1897 hrtimer_cancel(&t.timer);
1898 destroy_hrtimer_on_stack(&t.timer);
1899
1900 __set_current_state(TASK_RUNNING);
1901
1902 return !t.task ? 0 : -EINTR;
1903}
1904
1905/**
1906 * schedule_hrtimeout_range - sleep until timeout
1907 * @expires: timeout value (ktime_t)
1908 * @delta: slack in expires timeout (ktime_t)
1909 * @mode: timer mode
1910 *
1911 * Make the current task sleep until the given expiry time has
1912 * elapsed. The routine will return immediately unless
1913 * the current task state has been set (see set_current_state()).
1914 *
1915 * The @delta argument gives the kernel the freedom to schedule the
1916 * actual wakeup to a time that is both power and performance friendly.
1917 * The kernel give the normal best effort behavior for "@expires+@delta",
1918 * but may decide to fire the timer earlier, but no earlier than @expires.
1919 *
1920 * You can set the task state as follows -
1921 *
1922 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1923 * pass before the routine returns unless the current task is explicitly
1924 * woken up, (e.g. by wake_up_process()).
1925 *
1926 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1927 * delivered to the current task or the current task is explicitly woken
1928 * up.
1929 *
1930 * The current task state is guaranteed to be TASK_RUNNING when this
1931 * routine returns.
1932 *
1933 * Returns 0 when the timer has expired. If the task was woken before the
1934 * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or
1935 * by an explicit wakeup, it returns -EINTR.
1936 */
1937int __sched schedule_hrtimeout_range(ktime_t *expires, u64 delta,
1938 const enum hrtimer_mode mode)
1939{
1940 return schedule_hrtimeout_range_clock(expires, delta, mode,
1941 CLOCK_MONOTONIC);
1942}
1943EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
1944
1945/**
1946 * schedule_hrtimeout - sleep until timeout
1947 * @expires: timeout value (ktime_t)
1948 * @mode: timer mode
1949 *
1950 * Make the current task sleep until the given expiry time has
1951 * elapsed. The routine will return immediately unless
1952 * the current task state has been set (see set_current_state()).
1953 *
1954 * You can set the task state as follows -
1955 *
1956 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1957 * pass before the routine returns unless the current task is explicitly
1958 * woken up, (e.g. by wake_up_process()).
1959 *
1960 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1961 * delivered to the current task or the current task is explicitly woken
1962 * up.
1963 *
1964 * The current task state is guaranteed to be TASK_RUNNING when this
1965 * routine returns.
1966 *
1967 * Returns 0 when the timer has expired. If the task was woken before the
1968 * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or
1969 * by an explicit wakeup, it returns -EINTR.
1970 */
1971int __sched schedule_hrtimeout(ktime_t *expires,
1972 const enum hrtimer_mode mode)
1973{
1974 return schedule_hrtimeout_range(expires, 0, mode);
1975}
1976EXPORT_SYMBOL_GPL(schedule_hrtimeout);