projects/linux·blob v7.0 · refs

2340 lines · 65.3 KiB

1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright(C) 2005-2006, Linutronix GmbH, Thomas Gleixner <[email protected]>
4 * Copyright(C) 2005-2007, Red Hat, Inc., Ingo Molnar
5 * Copyright(C) 2006-2007 Timesys Corp., Thomas Gleixner
6 *
7 * High-resolution kernel timers
8 *
9 * In contrast to the low-resolution timeout API, aka timer wheel,
10 * hrtimers provide finer resolution and accuracy depending on system
11 * configuration and capabilities.
12 *
13 * Started by: Thomas Gleixner and Ingo Molnar
14 *
15 * Credits:
16 * Based on the original timer wheel code
17 *
18 * Help, testing, suggestions, bugfixes, improvements were
19 * provided by:
20 *
21 * George Anzinger, Andrew Morton, Steven Rostedt, Roman Zippel
22 * et. al.
23 */
24
25#include <linux/cpu.h>
26#include <linux/export.h>
27#include <linux/percpu.h>
28#include <linux/hrtimer.h>
29#include <linux/notifier.h>
30#include <linux/syscalls.h>
31#include <linux/interrupt.h>
32#include <linux/tick.h>
33#include <linux/err.h>
34#include <linux/debugobjects.h>
35#include <linux/sched/signal.h>
36#include <linux/sched/sysctl.h>
37#include <linux/sched/rt.h>
38#include <linux/sched/deadline.h>
39#include <linux/sched/nohz.h>
40#include <linux/sched/debug.h>
41#include <linux/sched/isolation.h>
42#include <linux/timer.h>
43#include <linux/freezer.h>
44#include <linux/compat.h>
45
46#include <linux/uaccess.h>
47
48#include <trace/events/timer.h>
49
50#include "tick-internal.h"
51
52/*
53 * The resolution of the clocks. The resolution value is returned in
54 * the clock_getres() system call to give application programmers an
55 * idea of the (in)accuracy of timers. Timer values are rounded up to
56 * this resolution values.
57 */
58#define HIGH_RES_NSEC 1
59
60/*
61 * Masks for selecting the soft and hard context timers from
62 * cpu_base->active
63 */
64#define MASK_SHIFT (HRTIMER_BASE_MONOTONIC_SOFT)
65#define HRTIMER_ACTIVE_HARD ((1U << MASK_SHIFT) - 1)
66#define HRTIMER_ACTIVE_SOFT (HRTIMER_ACTIVE_HARD << MASK_SHIFT)
67#define HRTIMER_ACTIVE_ALL (HRTIMER_ACTIVE_SOFT | HRTIMER_ACTIVE_HARD)
68
69static void retrigger_next_event(void *arg);
70static ktime_t __hrtimer_cb_get_time(clockid_t clock_id);
71
72/*
73 * The timer bases:
74 *
75 * There are more clockids than hrtimer bases. Thus, we index
76 * into the timer bases by the hrtimer_base_type enum. When trying
77 * to reach a base using a clockid, hrtimer_clockid_to_base()
78 * is used to convert from clockid to the proper hrtimer_base_type.
79 */
80DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
81{
82 .lock = __RAW_SPIN_LOCK_UNLOCKED(hrtimer_bases.lock),
83 .clock_base =
84 {
85 {
86 .index = HRTIMER_BASE_MONOTONIC,
87 .clockid = CLOCK_MONOTONIC,
88 },
89 {
90 .index = HRTIMER_BASE_REALTIME,
91 .clockid = CLOCK_REALTIME,
92 },
93 {
94 .index = HRTIMER_BASE_BOOTTIME,
95 .clockid = CLOCK_BOOTTIME,
96 },
97 {
98 .index = HRTIMER_BASE_TAI,
99 .clockid = CLOCK_TAI,
100 },
101 {
102 .index = HRTIMER_BASE_MONOTONIC_SOFT,
103 .clockid = CLOCK_MONOTONIC,
104 },
105 {
106 .index = HRTIMER_BASE_REALTIME_SOFT,
107 .clockid = CLOCK_REALTIME,
108 },
109 {
110 .index = HRTIMER_BASE_BOOTTIME_SOFT,
111 .clockid = CLOCK_BOOTTIME,
112 },
113 {
114 .index = HRTIMER_BASE_TAI_SOFT,
115 .clockid = CLOCK_TAI,
116 },
117 },
118 .csd = CSD_INIT(retrigger_next_event, NULL)
119};
120
121static inline bool hrtimer_base_is_online(struct hrtimer_cpu_base *base)
122{
123 if (!IS_ENABLED(CONFIG_HOTPLUG_CPU))
124 return true;
125 else
126 return likely(base->online);
127}
128
129/*
130 * Functions and macros which are different for UP/SMP systems are kept in a
131 * single place
132 */
133#ifdef CONFIG_SMP
134
135/*
136 * We require the migration_base for lock_hrtimer_base()/switch_hrtimer_base()
137 * such that hrtimer_callback_running() can unconditionally dereference
138 * timer->base->cpu_base
139 */
140static struct hrtimer_cpu_base migration_cpu_base = {
141 .clock_base = { {
142 .cpu_base = &migration_cpu_base,
143 .seq = SEQCNT_RAW_SPINLOCK_ZERO(migration_cpu_base.seq,
144 &migration_cpu_base.lock),
145 }, },
146};
147
148#define migration_base migration_cpu_base.clock_base[0]
149
150/*
151 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
152 * means that all timers which are tied to this base via timer->base are
153 * locked, and the base itself is locked too.
154 *
155 * So __run_timers/migrate_timers can safely modify all timers which could
156 * be found on the lists/queues.
157 *
158 * When the timer's base is locked, and the timer removed from list, it is
159 * possible to set timer->base = &migration_base and drop the lock: the timer
160 * remains locked.
161 */
162static
163struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
164 unsigned long *flags)
165 __acquires(&timer->base->lock)
166{
167 struct hrtimer_clock_base *base;
168
169 for (;;) {
170 base = READ_ONCE(timer->base);
171 if (likely(base != &migration_base)) {
172 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
173 if (likely(base == timer->base))
174 return base;
175 /* The timer has migrated to another CPU: */
176 raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
177 }
178 cpu_relax();
179 }
180}
181
182/*
183 * Check if the elected target is suitable considering its next
184 * event and the hotplug state of the current CPU.
185 *
186 * If the elected target is remote and its next event is after the timer
187 * to queue, then a remote reprogram is necessary. However there is no
188 * guarantee the IPI handling the operation would arrive in time to meet
189 * the high resolution deadline. In this case the local CPU becomes a
190 * preferred target, unless it is offline.
191 *
192 * High and low resolution modes are handled the same way for simplicity.
193 *
194 * Called with cpu_base->lock of target cpu held.
195 */
196static bool hrtimer_suitable_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base,
197 struct hrtimer_cpu_base *new_cpu_base,
198 struct hrtimer_cpu_base *this_cpu_base)
199{
200 ktime_t expires;
201
202 /*
203 * The local CPU clockevent can be reprogrammed. Also get_target_base()
204 * guarantees it is online.
205 */
206 if (new_cpu_base == this_cpu_base)
207 return true;
208
209 /*
210 * The offline local CPU can't be the default target if the
211 * next remote target event is after this timer. Keep the
212 * elected new base. An IPI will be issued to reprogram
213 * it as a last resort.
214 */
215 if (!hrtimer_base_is_online(this_cpu_base))
216 return true;
217
218 expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset);
219
220 return expires >= new_base->cpu_base->expires_next;
221}
222
223static inline struct hrtimer_cpu_base *get_target_base(struct hrtimer_cpu_base *base, int pinned)
224{
225 if (!hrtimer_base_is_online(base)) {
226 int cpu = cpumask_any_and(cpu_online_mask, housekeeping_cpumask(HK_TYPE_TIMER));
227
228 return &per_cpu(hrtimer_bases, cpu);
229 }
230
231#if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
232 if (static_branch_likely(&timers_migration_enabled) && !pinned)
233 return &per_cpu(hrtimer_bases, get_nohz_timer_target());
234#endif
235 return base;
236}
237
238/*
239 * We switch the timer base to a power-optimized selected CPU target,
240 * if:
241 * - NO_HZ_COMMON is enabled
242 * - timer migration is enabled
243 * - the timer callback is not running
244 * - the timer is not the first expiring timer on the new target
245 *
246 * If one of the above requirements is not fulfilled we move the timer
247 * to the current CPU or leave it on the previously assigned CPU if
248 * the timer callback is currently running.
249 */
250static inline struct hrtimer_clock_base *
251switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
252 int pinned)
253{
254 struct hrtimer_cpu_base *new_cpu_base, *this_cpu_base;
255 struct hrtimer_clock_base *new_base;
256 int basenum = base->index;
257
258 this_cpu_base = this_cpu_ptr(&hrtimer_bases);
259 new_cpu_base = get_target_base(this_cpu_base, pinned);
260again:
261 new_base = &new_cpu_base->clock_base[basenum];
262
263 if (base != new_base) {
264 /*
265 * We are trying to move timer to new_base.
266 * However we can't change timer's base while it is running,
267 * so we keep it on the same CPU. No hassle vs. reprogramming
268 * the event source in the high resolution case. The softirq
269 * code will take care of this when the timer function has
270 * completed. There is no conflict as we hold the lock until
271 * the timer is enqueued.
272 */
273 if (unlikely(hrtimer_callback_running(timer)))
274 return base;
275
276 /* See the comment in lock_hrtimer_base() */
277 WRITE_ONCE(timer->base, &migration_base);
278 raw_spin_unlock(&base->cpu_base->lock);
279 raw_spin_lock(&new_base->cpu_base->lock);
280
281 if (!hrtimer_suitable_target(timer, new_base, new_cpu_base,
282 this_cpu_base)) {
283 raw_spin_unlock(&new_base->cpu_base->lock);
284 raw_spin_lock(&base->cpu_base->lock);
285 new_cpu_base = this_cpu_base;
286 WRITE_ONCE(timer->base, base);
287 goto again;
288 }
289 WRITE_ONCE(timer->base, new_base);
290 } else {
291 if (!hrtimer_suitable_target(timer, new_base, new_cpu_base, this_cpu_base)) {
292 new_cpu_base = this_cpu_base;
293 goto again;
294 }
295 }
296 return new_base;
297}
298
299#else /* CONFIG_SMP */
300
301static inline struct hrtimer_clock_base *
302lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
303 __acquires(&timer->base->cpu_base->lock)
304{
305 struct hrtimer_clock_base *base = timer->base;
306
307 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
308
309 return base;
310}
311
312# define switch_hrtimer_base(t, b, p) (b)
313
314#endif /* !CONFIG_SMP */
315
316/*
317 * Functions for the union type storage format of ktime_t which are
318 * too large for inlining:
319 */
320#if BITS_PER_LONG < 64
321/*
322 * Divide a ktime value by a nanosecond value
323 */
324s64 __ktime_divns(const ktime_t kt, s64 div)
325{
326 int sft = 0;
327 s64 dclc;
328 u64 tmp;
329
330 dclc = ktime_to_ns(kt);
331 tmp = dclc < 0 ? -dclc : dclc;
332
333 /* Make sure the divisor is less than 2^32: */
334 while (div >> 32) {
335 sft++;
336 div >>= 1;
337 }
338 tmp >>= sft;
339 do_div(tmp, (u32) div);
340 return dclc < 0 ? -tmp : tmp;
341}
342EXPORT_SYMBOL_GPL(__ktime_divns);
343#endif /* BITS_PER_LONG >= 64 */
344
345/*
346 * Add two ktime values and do a safety check for overflow:
347 */
348ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
349{
350 ktime_t res = ktime_add_unsafe(lhs, rhs);
351
352 /*
353 * We use KTIME_SEC_MAX here, the maximum timeout which we can
354 * return to user space in a timespec:
355 */
356 if (res < 0 || res < lhs || res < rhs)
357 res = ktime_set(KTIME_SEC_MAX, 0);
358
359 return res;
360}
361
362EXPORT_SYMBOL_GPL(ktime_add_safe);
363
364#ifdef CONFIG_DEBUG_OBJECTS_TIMERS
365
366static const struct debug_obj_descr hrtimer_debug_descr;
367
368static void *hrtimer_debug_hint(void *addr)
369{
370 return ACCESS_PRIVATE((struct hrtimer *)addr, function);
371}
372
373/*
374 * fixup_init is called when:
375 * - an active object is initialized
376 */
377static bool hrtimer_fixup_init(void *addr, enum debug_obj_state state)
378{
379 struct hrtimer *timer = addr;
380
381 switch (state) {
382 case ODEBUG_STATE_ACTIVE:
383 hrtimer_cancel(timer);
384 debug_object_init(timer, &hrtimer_debug_descr);
385 return true;
386 default:
387 return false;
388 }
389}
390
391/*
392 * fixup_activate is called when:
393 * - an active object is activated
394 * - an unknown non-static object is activated
395 */
396static bool hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
397{
398 switch (state) {
399 case ODEBUG_STATE_ACTIVE:
400 WARN_ON(1);
401 fallthrough;
402 default:
403 return false;
404 }
405}
406
407/*
408 * fixup_free is called when:
409 * - an active object is freed
410 */
411static bool hrtimer_fixup_free(void *addr, enum debug_obj_state state)
412{
413 struct hrtimer *timer = addr;
414
415 switch (state) {
416 case ODEBUG_STATE_ACTIVE:
417 hrtimer_cancel(timer);
418 debug_object_free(timer, &hrtimer_debug_descr);
419 return true;
420 default:
421 return false;
422 }
423}
424
425static const struct debug_obj_descr hrtimer_debug_descr = {
426 .name = "hrtimer",
427 .debug_hint = hrtimer_debug_hint,
428 .fixup_init = hrtimer_fixup_init,
429 .fixup_activate = hrtimer_fixup_activate,
430 .fixup_free = hrtimer_fixup_free,
431};
432
433static inline void debug_hrtimer_init(struct hrtimer *timer)
434{
435 debug_object_init(timer, &hrtimer_debug_descr);
436}
437
438static inline void debug_hrtimer_init_on_stack(struct hrtimer *timer)
439{
440 debug_object_init_on_stack(timer, &hrtimer_debug_descr);
441}
442
443static inline void debug_hrtimer_activate(struct hrtimer *timer,
444 enum hrtimer_mode mode)
445{
446 debug_object_activate(timer, &hrtimer_debug_descr);
447}
448
449static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
450{
451 debug_object_deactivate(timer, &hrtimer_debug_descr);
452}
453
454void destroy_hrtimer_on_stack(struct hrtimer *timer)
455{
456 debug_object_free(timer, &hrtimer_debug_descr);
457}
458EXPORT_SYMBOL_GPL(destroy_hrtimer_on_stack);
459
460#else
461
462static inline void debug_hrtimer_init(struct hrtimer *timer) { }
463static inline void debug_hrtimer_init_on_stack(struct hrtimer *timer) { }
464static inline void debug_hrtimer_activate(struct hrtimer *timer,
465 enum hrtimer_mode mode) { }
466static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
467#endif
468
469static inline void debug_setup(struct hrtimer *timer, clockid_t clockid, enum hrtimer_mode mode)
470{
471 debug_hrtimer_init(timer);
472 trace_hrtimer_setup(timer, clockid, mode);
473}
474
475static inline void debug_setup_on_stack(struct hrtimer *timer, clockid_t clockid,
476 enum hrtimer_mode mode)
477{
478 debug_hrtimer_init_on_stack(timer);
479 trace_hrtimer_setup(timer, clockid, mode);
480}
481
482static inline void debug_activate(struct hrtimer *timer,
483 enum hrtimer_mode mode)
484{
485 debug_hrtimer_activate(timer, mode);
486 trace_hrtimer_start(timer, mode);
487}
488
489static inline void debug_deactivate(struct hrtimer *timer)
490{
491 debug_hrtimer_deactivate(timer);
492 trace_hrtimer_cancel(timer);
493}
494
495static struct hrtimer_clock_base *
496__next_base(struct hrtimer_cpu_base *cpu_base, unsigned int *active)
497{
498 unsigned int idx;
499
500 if (!*active)
501 return NULL;
502
503 idx = __ffs(*active);
504 *active &= ~(1U << idx);
505
506 return &cpu_base->clock_base[idx];
507}
508
509#define for_each_active_base(base, cpu_base, active) \
510 while ((base = __next_base((cpu_base), &(active))))
511
512static ktime_t __hrtimer_next_event_base(struct hrtimer_cpu_base *cpu_base,
513 const struct hrtimer *exclude,
514 unsigned int active,
515 ktime_t expires_next)
516{
517 struct hrtimer_clock_base *base;
518 ktime_t expires;
519
520 for_each_active_base(base, cpu_base, active) {
521 struct timerqueue_node *next;
522 struct hrtimer *timer;
523
524 next = timerqueue_getnext(&base->active);
525 timer = container_of(next, struct hrtimer, node);
526 if (timer == exclude) {
527 /* Get to the next timer in the queue. */
528 next = timerqueue_iterate_next(next);
529 if (!next)
530 continue;
531
532 timer = container_of(next, struct hrtimer, node);
533 }
534 expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
535 if (expires < expires_next) {
536 expires_next = expires;
537
538 /* Skip cpu_base update if a timer is being excluded. */
539 if (exclude)
540 continue;
541
542 if (timer->is_soft)
543 cpu_base->softirq_next_timer = timer;
544 else
545 cpu_base->next_timer = timer;
546 }
547 }
548 /*
549 * clock_was_set() might have changed base->offset of any of
550 * the clock bases so the result might be negative. Fix it up
551 * to prevent a false positive in clockevents_program_event().
552 */
553 if (expires_next < 0)
554 expires_next = 0;
555 return expires_next;
556}
557
558/*
559 * Recomputes cpu_base::*next_timer and returns the earliest expires_next
560 * but does not set cpu_base::*expires_next, that is done by
561 * hrtimer[_force]_reprogram and hrtimer_interrupt only. When updating
562 * cpu_base::*expires_next right away, reprogramming logic would no longer
563 * work.
564 *
565 * When a softirq is pending, we can ignore the HRTIMER_ACTIVE_SOFT bases,
566 * those timers will get run whenever the softirq gets handled, at the end of
567 * hrtimer_run_softirq(), hrtimer_update_softirq_timer() will re-add these bases.
568 *
569 * Therefore softirq values are those from the HRTIMER_ACTIVE_SOFT clock bases.
570 * The !softirq values are the minima across HRTIMER_ACTIVE_ALL, unless an actual
571 * softirq is pending, in which case they're the minima of HRTIMER_ACTIVE_HARD.
572 *
573 * @active_mask must be one of:
574 * - HRTIMER_ACTIVE_ALL,
575 * - HRTIMER_ACTIVE_SOFT, or
576 * - HRTIMER_ACTIVE_HARD.
577 */
578static ktime_t
579__hrtimer_get_next_event(struct hrtimer_cpu_base *cpu_base, unsigned int active_mask)
580{
581 unsigned int active;
582 struct hrtimer *next_timer = NULL;
583 ktime_t expires_next = KTIME_MAX;
584
585 if (!cpu_base->softirq_activated && (active_mask & HRTIMER_ACTIVE_SOFT)) {
586 active = cpu_base->active_bases & HRTIMER_ACTIVE_SOFT;
587 cpu_base->softirq_next_timer = NULL;
588 expires_next = __hrtimer_next_event_base(cpu_base, NULL,
589 active, KTIME_MAX);
590
591 next_timer = cpu_base->softirq_next_timer;
592 }
593
594 if (active_mask & HRTIMER_ACTIVE_HARD) {
595 active = cpu_base->active_bases & HRTIMER_ACTIVE_HARD;
596 cpu_base->next_timer = next_timer;
597 expires_next = __hrtimer_next_event_base(cpu_base, NULL, active,
598 expires_next);
599 }
600
601 return expires_next;
602}
603
604static ktime_t hrtimer_update_next_event(struct hrtimer_cpu_base *cpu_base)
605{
606 ktime_t expires_next, soft = KTIME_MAX;
607
608 /*
609 * If the soft interrupt has already been activated, ignore the
610 * soft bases. They will be handled in the already raised soft
611 * interrupt.
612 */
613 if (!cpu_base->softirq_activated) {
614 soft = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_SOFT);
615 /*
616 * Update the soft expiry time. clock_settime() might have
617 * affected it.
618 */
619 cpu_base->softirq_expires_next = soft;
620 }
621
622 expires_next = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_HARD);
623 /*
624 * If a softirq timer is expiring first, update cpu_base->next_timer
625 * and program the hardware with the soft expiry time.
626 */
627 if (expires_next > soft) {
628 cpu_base->next_timer = cpu_base->softirq_next_timer;
629 expires_next = soft;
630 }
631
632 return expires_next;
633}
634
635static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base)
636{
637 ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset;
638 ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset;
639 ktime_t *offs_tai = &base->clock_base[HRTIMER_BASE_TAI].offset;
640
641 ktime_t now = ktime_get_update_offsets_now(&base->clock_was_set_seq,
642 offs_real, offs_boot, offs_tai);
643
644 base->clock_base[HRTIMER_BASE_REALTIME_SOFT].offset = *offs_real;
645 base->clock_base[HRTIMER_BASE_BOOTTIME_SOFT].offset = *offs_boot;
646 base->clock_base[HRTIMER_BASE_TAI_SOFT].offset = *offs_tai;
647
648 return now;
649}
650
651/*
652 * Is the high resolution mode active ?
653 */
654static inline int hrtimer_hres_active(struct hrtimer_cpu_base *cpu_base)
655{
656 return IS_ENABLED(CONFIG_HIGH_RES_TIMERS) ?
657 cpu_base->hres_active : 0;
658}
659
660static void __hrtimer_reprogram(struct hrtimer_cpu_base *cpu_base,
661 struct hrtimer *next_timer,
662 ktime_t expires_next)
663{
664 cpu_base->expires_next = expires_next;
665
666 /*
667 * If hres is not active, hardware does not have to be
668 * reprogrammed yet.
669 *
670 * If a hang was detected in the last timer interrupt then we
671 * leave the hang delay active in the hardware. We want the
672 * system to make progress. That also prevents the following
673 * scenario:
674 * T1 expires 50ms from now
675 * T2 expires 5s from now
676 *
677 * T1 is removed, so this code is called and would reprogram
678 * the hardware to 5s from now. Any hrtimer_start after that
679 * will not reprogram the hardware due to hang_detected being
680 * set. So we'd effectively block all timers until the T2 event
681 * fires.
682 */
683 if (!hrtimer_hres_active(cpu_base) || cpu_base->hang_detected)
684 return;
685
686 tick_program_event(expires_next, 1);
687}
688
689/*
690 * Reprogram the event source with checking both queues for the
691 * next event
692 * Called with interrupts disabled and base->lock held
693 */
694static void
695hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
696{
697 ktime_t expires_next;
698
699 expires_next = hrtimer_update_next_event(cpu_base);
700
701 if (skip_equal && expires_next == cpu_base->expires_next)
702 return;
703
704 __hrtimer_reprogram(cpu_base, cpu_base->next_timer, expires_next);
705}
706
707/* High resolution timer related functions */
708#ifdef CONFIG_HIGH_RES_TIMERS
709
710/*
711 * High resolution timer enabled ?
712 */
713static bool hrtimer_hres_enabled __read_mostly = true;
714unsigned int hrtimer_resolution __read_mostly = LOW_RES_NSEC;
715EXPORT_SYMBOL_GPL(hrtimer_resolution);
716
717/*
718 * Enable / Disable high resolution mode
719 */
720static int __init setup_hrtimer_hres(char *str)
721{
722 return (kstrtobool(str, &hrtimer_hres_enabled) == 0);
723}
724
725__setup("highres=", setup_hrtimer_hres);
726
727/*
728 * hrtimer_high_res_enabled - query, if the highres mode is enabled
729 */
730static inline int hrtimer_is_hres_enabled(void)
731{
732 return hrtimer_hres_enabled;
733}
734
735/*
736 * Switch to high resolution mode
737 */
738static void hrtimer_switch_to_hres(void)
739{
740 struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
741
742 if (tick_init_highres()) {
743 pr_warn("Could not switch to high resolution mode on CPU %u\n",
744 base->cpu);
745 return;
746 }
747 base->hres_active = 1;
748 hrtimer_resolution = HIGH_RES_NSEC;
749
750 tick_setup_sched_timer(true);
751 /* "Retrigger" the interrupt to get things going */
752 retrigger_next_event(NULL);
753}
754
755#else
756
757static inline int hrtimer_is_hres_enabled(void) { return 0; }
758static inline void hrtimer_switch_to_hres(void) { }
759
760#endif /* CONFIG_HIGH_RES_TIMERS */
761/*
762 * Retrigger next event is called after clock was set with interrupts
763 * disabled through an SMP function call or directly from low level
764 * resume code.
765 *
766 * This is only invoked when:
767 * - CONFIG_HIGH_RES_TIMERS is enabled.
768 * - CONFIG_NOHZ_COMMON is enabled
769 *
770 * For the other cases this function is empty and because the call sites
771 * are optimized out it vanishes as well, i.e. no need for lots of
772 * #ifdeffery.
773 */
774static void retrigger_next_event(void *arg)
775{
776 struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
777
778 /*
779 * When high resolution mode or nohz is active, then the offsets of
780 * CLOCK_REALTIME/TAI/BOOTTIME have to be updated. Otherwise the
781 * next tick will take care of that.
782 *
783 * If high resolution mode is active then the next expiring timer
784 * must be reevaluated and the clock event device reprogrammed if
785 * necessary.
786 *
787 * In the NOHZ case the update of the offset and the reevaluation
788 * of the next expiring timer is enough. The return from the SMP
789 * function call will take care of the reprogramming in case the
790 * CPU was in a NOHZ idle sleep.
791 *
792 * In periodic low resolution mode, the next softirq expiration
793 * must also be updated.
794 */
795 raw_spin_lock(&base->lock);
796 hrtimer_update_base(base);
797 if (hrtimer_hres_active(base))
798 hrtimer_force_reprogram(base, 0);
799 else
800 hrtimer_update_next_event(base);
801 raw_spin_unlock(&base->lock);
802}
803
804/*
805 * When a timer is enqueued and expires earlier than the already enqueued
806 * timers, we have to check, whether it expires earlier than the timer for
807 * which the clock event device was armed.
808 *
809 * Called with interrupts disabled and base->cpu_base.lock held
810 */
811static void hrtimer_reprogram(struct hrtimer *timer, bool reprogram)
812{
813 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
814 struct hrtimer_clock_base *base = timer->base;
815 ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
816
817 WARN_ON_ONCE(hrtimer_get_expires(timer) < 0);
818
819 /*
820 * CLOCK_REALTIME timer might be requested with an absolute
821 * expiry time which is less than base->offset. Set it to 0.
822 */
823 if (expires < 0)
824 expires = 0;
825
826 if (timer->is_soft) {
827 /*
828 * soft hrtimer could be started on a remote CPU. In this
829 * case softirq_expires_next needs to be updated on the
830 * remote CPU. The soft hrtimer will not expire before the
831 * first hard hrtimer on the remote CPU -
832 * hrtimer_check_target() prevents this case.
833 */
834 struct hrtimer_cpu_base *timer_cpu_base = base->cpu_base;
835
836 if (timer_cpu_base->softirq_activated)
837 return;
838
839 if (!ktime_before(expires, timer_cpu_base->softirq_expires_next))
840 return;
841
842 timer_cpu_base->softirq_next_timer = timer;
843 timer_cpu_base->softirq_expires_next = expires;
844
845 if (!ktime_before(expires, timer_cpu_base->expires_next) ||
846 !reprogram)
847 return;
848 }
849
850 /*
851 * If the timer is not on the current cpu, we cannot reprogram
852 * the other cpus clock event device.
853 */
854 if (base->cpu_base != cpu_base)
855 return;
856
857 if (expires >= cpu_base->expires_next)
858 return;
859
860 /*
861 * If the hrtimer interrupt is running, then it will reevaluate the
862 * clock bases and reprogram the clock event device.
863 */
864 if (cpu_base->in_hrtirq)
865 return;
866
867 cpu_base->next_timer = timer;
868
869 __hrtimer_reprogram(cpu_base, timer, expires);
870}
871
872static bool update_needs_ipi(struct hrtimer_cpu_base *cpu_base,
873 unsigned int active)
874{
875 struct hrtimer_clock_base *base;
876 unsigned int seq;
877 ktime_t expires;
878
879 /*
880 * Update the base offsets unconditionally so the following
881 * checks whether the SMP function call is required works.
882 *
883 * The update is safe even when the remote CPU is in the hrtimer
884 * interrupt or the hrtimer soft interrupt and expiring affected
885 * bases. Either it will see the update before handling a base or
886 * it will see it when it finishes the processing and reevaluates
887 * the next expiring timer.
888 */
889 seq = cpu_base->clock_was_set_seq;
890 hrtimer_update_base(cpu_base);
891
892 /*
893 * If the sequence did not change over the update then the
894 * remote CPU already handled it.
895 */
896 if (seq == cpu_base->clock_was_set_seq)
897 return false;
898
899 /*
900 * If the remote CPU is currently handling an hrtimer interrupt, it
901 * will reevaluate the first expiring timer of all clock bases
902 * before reprogramming. Nothing to do here.
903 */
904 if (cpu_base->in_hrtirq)
905 return false;
906
907 /*
908 * Walk the affected clock bases and check whether the first expiring
909 * timer in a clock base is moving ahead of the first expiring timer of
910 * @cpu_base. If so, the IPI must be invoked because per CPU clock
911 * event devices cannot be remotely reprogrammed.
912 */
913 active &= cpu_base->active_bases;
914
915 for_each_active_base(base, cpu_base, active) {
916 struct timerqueue_node *next;
917
918 next = timerqueue_getnext(&base->active);
919 expires = ktime_sub(next->expires, base->offset);
920 if (expires < cpu_base->expires_next)
921 return true;
922
923 /* Extra check for softirq clock bases */
924 if (base->index < HRTIMER_BASE_MONOTONIC_SOFT)
925 continue;
926 if (cpu_base->softirq_activated)
927 continue;
928 if (expires < cpu_base->softirq_expires_next)
929 return true;
930 }
931 return false;
932}
933
934/*
935 * Clock was set. This might affect CLOCK_REALTIME, CLOCK_TAI and
936 * CLOCK_BOOTTIME (for late sleep time injection).
937 *
938 * This requires to update the offsets for these clocks
939 * vs. CLOCK_MONOTONIC. When high resolution timers are enabled, then this
940 * also requires to eventually reprogram the per CPU clock event devices
941 * when the change moves an affected timer ahead of the first expiring
942 * timer on that CPU. Obviously remote per CPU clock event devices cannot
943 * be reprogrammed. The other reason why an IPI has to be sent is when the
944 * system is in !HIGH_RES and NOHZ mode. The NOHZ mode updates the offsets
945 * in the tick, which obviously might be stopped, so this has to bring out
946 * the remote CPU which might sleep in idle to get this sorted.
947 */
948void clock_was_set(unsigned int bases)
949{
950 struct hrtimer_cpu_base *cpu_base = raw_cpu_ptr(&hrtimer_bases);
951 cpumask_var_t mask;
952 int cpu;
953
954 if (!hrtimer_hres_active(cpu_base) && !tick_nohz_is_active())
955 goto out_timerfd;
956
957 if (!zalloc_cpumask_var(&mask, GFP_KERNEL)) {
958 on_each_cpu(retrigger_next_event, NULL, 1);
959 goto out_timerfd;
960 }
961
962 /* Avoid interrupting CPUs if possible */
963 cpus_read_lock();
964 for_each_online_cpu(cpu) {
965 unsigned long flags;
966
967 cpu_base = &per_cpu(hrtimer_bases, cpu);
968 raw_spin_lock_irqsave(&cpu_base->lock, flags);
969
970 if (update_needs_ipi(cpu_base, bases))
971 cpumask_set_cpu(cpu, mask);
972
973 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
974 }
975
976 preempt_disable();
977 smp_call_function_many(mask, retrigger_next_event, NULL, 1);
978 preempt_enable();
979 cpus_read_unlock();
980 free_cpumask_var(mask);
981
982out_timerfd:
983 timerfd_clock_was_set();
984}
985
986static void clock_was_set_work(struct work_struct *work)
987{
988 clock_was_set(CLOCK_SET_WALL);
989}
990
991static DECLARE_WORK(hrtimer_work, clock_was_set_work);
992
993/*
994 * Called from timekeeping code to reprogram the hrtimer interrupt device
995 * on all cpus and to notify timerfd.
996 */
997void clock_was_set_delayed(void)
998{
999 schedule_work(&hrtimer_work);
1000}
1001
1002/*
1003 * Called during resume either directly from via timekeeping_resume()
1004 * or in the case of s2idle from tick_unfreeze() to ensure that the
1005 * hrtimers are up to date.
1006 */
1007void hrtimers_resume_local(void)
1008{
1009 lockdep_assert_irqs_disabled();
1010 /* Retrigger on the local CPU */
1011 retrigger_next_event(NULL);
1012}
1013
1014/*
1015 * Counterpart to lock_hrtimer_base above:
1016 */
1017static inline
1018void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
1019 __releases(&timer->base->cpu_base->lock)
1020{
1021 raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
1022}
1023
1024/**
1025 * hrtimer_forward() - forward the timer expiry
1026 * @timer: hrtimer to forward
1027 * @now: forward past this time
1028 * @interval: the interval to forward
1029 *
1030 * Forward the timer expiry so it will expire in the future.
1031 *
1032 * .. note::
1033 * This only updates the timer expiry value and does not requeue the timer.
1034 *
1035 * There is also a variant of the function hrtimer_forward_now().
1036 *
1037 * Context: Can be safely called from the callback function of @timer. If called
1038 * from other contexts @timer must neither be enqueued nor running the
1039 * callback and the caller needs to take care of serialization.
1040 *
1041 * Return: The number of overruns are returned.
1042 */
1043u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
1044{
1045 u64 orun = 1;
1046 ktime_t delta;
1047
1048 delta = ktime_sub(now, hrtimer_get_expires(timer));
1049
1050 if (delta < 0)
1051 return 0;
1052
1053 if (WARN_ON(timer->state & HRTIMER_STATE_ENQUEUED))
1054 return 0;
1055
1056 if (interval < hrtimer_resolution)
1057 interval = hrtimer_resolution;
1058
1059 if (unlikely(delta >= interval)) {
1060 s64 incr = ktime_to_ns(interval);
1061
1062 orun = ktime_divns(delta, incr);
1063 hrtimer_add_expires_ns(timer, incr * orun);
1064 if (hrtimer_get_expires(timer) > now)
1065 return orun;
1066 /*
1067 * This (and the ktime_add() below) is the
1068 * correction for exact:
1069 */
1070 orun++;
1071 }
1072 hrtimer_add_expires(timer, interval);
1073
1074 return orun;
1075}
1076EXPORT_SYMBOL_GPL(hrtimer_forward);
1077
1078/*
1079 * enqueue_hrtimer - internal function to (re)start a timer
1080 *
1081 * The timer is inserted in expiry order. Insertion into the
1082 * red black tree is O(log(n)). Must hold the base lock.
1083 *
1084 * Returns true when the new timer is the leftmost timer in the tree.
1085 */
1086static bool enqueue_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base,
1087 enum hrtimer_mode mode)
1088{
1089 debug_activate(timer, mode);
1090 WARN_ON_ONCE(!base->cpu_base->online);
1091
1092 base->cpu_base->active_bases |= 1 << base->index;
1093
1094 /* Pairs with the lockless read in hrtimer_is_queued() */
1095 WRITE_ONCE(timer->state, HRTIMER_STATE_ENQUEUED);
1096
1097 return timerqueue_add(&base->active, &timer->node);
1098}
1099
1100/*
1101 * __remove_hrtimer - internal function to remove a timer
1102 *
1103 * Caller must hold the base lock.
1104 *
1105 * High resolution timer mode reprograms the clock event device when the
1106 * timer is the one which expires next. The caller can disable this by setting
1107 * reprogram to zero. This is useful, when the context does a reprogramming
1108 * anyway (e.g. timer interrupt)
1109 */
1110static void __remove_hrtimer(struct hrtimer *timer,
1111 struct hrtimer_clock_base *base,
1112 u8 newstate, int reprogram)
1113{
1114 struct hrtimer_cpu_base *cpu_base = base->cpu_base;
1115 u8 state = timer->state;
1116
1117 /* Pairs with the lockless read in hrtimer_is_queued() */
1118 WRITE_ONCE(timer->state, newstate);
1119 if (!(state & HRTIMER_STATE_ENQUEUED))
1120 return;
1121
1122 if (!timerqueue_del(&base->active, &timer->node))
1123 cpu_base->active_bases &= ~(1 << base->index);
1124
1125 /*
1126 * Note: If reprogram is false we do not update
1127 * cpu_base->next_timer. This happens when we remove the first
1128 * timer on a remote cpu. No harm as we never dereference
1129 * cpu_base->next_timer. So the worst thing what can happen is
1130 * an superfluous call to hrtimer_force_reprogram() on the
1131 * remote cpu later on if the same timer gets enqueued again.
1132 */
1133 if (reprogram && timer == cpu_base->next_timer)
1134 hrtimer_force_reprogram(cpu_base, 1);
1135}
1136
1137/*
1138 * remove hrtimer, called with base lock held
1139 */
1140static inline int
1141remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base,
1142 bool restart, bool keep_local)
1143{
1144 u8 state = timer->state;
1145
1146 if (state & HRTIMER_STATE_ENQUEUED) {
1147 bool reprogram;
1148
1149 /*
1150 * Remove the timer and force reprogramming when high
1151 * resolution mode is active and the timer is on the current
1152 * CPU. If we remove a timer on another CPU, reprogramming is
1153 * skipped. The interrupt event on this CPU is fired and
1154 * reprogramming happens in the interrupt handler. This is a
1155 * rare case and less expensive than a smp call.
1156 */
1157 debug_deactivate(timer);
1158 reprogram = base->cpu_base == this_cpu_ptr(&hrtimer_bases);
1159
1160 /*
1161 * If the timer is not restarted then reprogramming is
1162 * required if the timer is local. If it is local and about
1163 * to be restarted, avoid programming it twice (on removal
1164 * and a moment later when it's requeued).
1165 */
1166 if (!restart)
1167 state = HRTIMER_STATE_INACTIVE;
1168 else
1169 reprogram &= !keep_local;
1170
1171 __remove_hrtimer(timer, base, state, reprogram);
1172 return 1;
1173 }
1174 return 0;
1175}
1176
1177static inline ktime_t hrtimer_update_lowres(struct hrtimer *timer, ktime_t tim,
1178 const enum hrtimer_mode mode)
1179{
1180#ifdef CONFIG_TIME_LOW_RES
1181 /*
1182 * CONFIG_TIME_LOW_RES indicates that the system has no way to return
1183 * granular time values. For relative timers we add hrtimer_resolution
1184 * (i.e. one jiffy) to prevent short timeouts.
1185 */
1186 timer->is_rel = mode & HRTIMER_MODE_REL;
1187 if (timer->is_rel)
1188 tim = ktime_add_safe(tim, hrtimer_resolution);
1189#endif
1190 return tim;
1191}
1192
1193static void
1194hrtimer_update_softirq_timer(struct hrtimer_cpu_base *cpu_base, bool reprogram)
1195{
1196 ktime_t expires;
1197
1198 /*
1199 * Find the next SOFT expiration.
1200 */
1201 expires = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_SOFT);
1202
1203 /*
1204 * reprogramming needs to be triggered, even if the next soft
1205 * hrtimer expires at the same time than the next hard
1206 * hrtimer. cpu_base->softirq_expires_next needs to be updated!
1207 */
1208 if (expires == KTIME_MAX)
1209 return;
1210
1211 /*
1212 * cpu_base->*next_timer is recomputed by __hrtimer_get_next_event()
1213 * cpu_base->*expires_next is only set by hrtimer_reprogram()
1214 */
1215 hrtimer_reprogram(cpu_base->softirq_next_timer, reprogram);
1216}
1217
1218static int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1219 u64 delta_ns, const enum hrtimer_mode mode,
1220 struct hrtimer_clock_base *base)
1221{
1222 struct hrtimer_cpu_base *this_cpu_base = this_cpu_ptr(&hrtimer_bases);
1223 struct hrtimer_clock_base *new_base;
1224 bool force_local, first;
1225
1226 /*
1227 * If the timer is on the local cpu base and is the first expiring
1228 * timer then this might end up reprogramming the hardware twice
1229 * (on removal and on enqueue). To avoid that by prevent the
1230 * reprogram on removal, keep the timer local to the current CPU
1231 * and enforce reprogramming after it is queued no matter whether
1232 * it is the new first expiring timer again or not.
1233 */
1234 force_local = base->cpu_base == this_cpu_base;
1235 force_local &= base->cpu_base->next_timer == timer;
1236
1237 /*
1238 * Don't force local queuing if this enqueue happens on a unplugged
1239 * CPU after hrtimer_cpu_dying() has been invoked.
1240 */
1241 force_local &= this_cpu_base->online;
1242
1243 /*
1244 * Remove an active timer from the queue. In case it is not queued
1245 * on the current CPU, make sure that remove_hrtimer() updates the
1246 * remote data correctly.
1247 *
1248 * If it's on the current CPU and the first expiring timer, then
1249 * skip reprogramming, keep the timer local and enforce
1250 * reprogramming later if it was the first expiring timer. This
1251 * avoids programming the underlying clock event twice (once at
1252 * removal and once after enqueue).
1253 */
1254 remove_hrtimer(timer, base, true, force_local);
1255
1256 if (mode & HRTIMER_MODE_REL)
1257 tim = ktime_add_safe(tim, __hrtimer_cb_get_time(base->clockid));
1258
1259 tim = hrtimer_update_lowres(timer, tim, mode);
1260
1261 hrtimer_set_expires_range_ns(timer, tim, delta_ns);
1262
1263 /* Switch the timer base, if necessary: */
1264 if (!force_local) {
1265 new_base = switch_hrtimer_base(timer, base,
1266 mode & HRTIMER_MODE_PINNED);
1267 } else {
1268 new_base = base;
1269 }
1270
1271 first = enqueue_hrtimer(timer, new_base, mode);
1272 if (!force_local) {
1273 /*
1274 * If the current CPU base is online, then the timer is
1275 * never queued on a remote CPU if it would be the first
1276 * expiring timer there.
1277 */
1278 if (hrtimer_base_is_online(this_cpu_base))
1279 return first;
1280
1281 /*
1282 * Timer was enqueued remote because the current base is
1283 * already offline. If the timer is the first to expire,
1284 * kick the remote CPU to reprogram the clock event.
1285 */
1286 if (first) {
1287 struct hrtimer_cpu_base *new_cpu_base = new_base->cpu_base;
1288
1289 smp_call_function_single_async(new_cpu_base->cpu, &new_cpu_base->csd);
1290 }
1291 return 0;
1292 }
1293
1294 /*
1295 * Timer was forced to stay on the current CPU to avoid
1296 * reprogramming on removal and enqueue. Force reprogram the
1297 * hardware by evaluating the new first expiring timer.
1298 */
1299 hrtimer_force_reprogram(new_base->cpu_base, 1);
1300 return 0;
1301}
1302
1303/**
1304 * hrtimer_start_range_ns - (re)start an hrtimer
1305 * @timer: the timer to be added
1306 * @tim: expiry time
1307 * @delta_ns: "slack" range for the timer
1308 * @mode: timer mode: absolute (HRTIMER_MODE_ABS) or
1309 * relative (HRTIMER_MODE_REL), and pinned (HRTIMER_MODE_PINNED);
1310 * softirq based mode is considered for debug purpose only!
1311 */
1312void hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1313 u64 delta_ns, const enum hrtimer_mode mode)
1314{
1315 struct hrtimer_clock_base *base;
1316 unsigned long flags;
1317
1318 /*
1319 * Check whether the HRTIMER_MODE_SOFT bit and hrtimer.is_soft
1320 * match on CONFIG_PREEMPT_RT = n. With PREEMPT_RT check the hard
1321 * expiry mode because unmarked timers are moved to softirq expiry.
1322 */
1323 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
1324 WARN_ON_ONCE(!(mode & HRTIMER_MODE_SOFT) ^ !timer->is_soft);
1325 else
1326 WARN_ON_ONCE(!(mode & HRTIMER_MODE_HARD) ^ !timer->is_hard);
1327
1328 base = lock_hrtimer_base(timer, &flags);
1329
1330 if (__hrtimer_start_range_ns(timer, tim, delta_ns, mode, base))
1331 hrtimer_reprogram(timer, true);
1332
1333 unlock_hrtimer_base(timer, &flags);
1334}
1335EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
1336
1337/**
1338 * hrtimer_try_to_cancel - try to deactivate a timer
1339 * @timer: hrtimer to stop
1340 *
1341 * Returns:
1342 *
1343 * * 0 when the timer was not active
1344 * * 1 when the timer was active
1345 * * -1 when the timer is currently executing the callback function and
1346 * cannot be stopped
1347 */
1348int hrtimer_try_to_cancel(struct hrtimer *timer)
1349{
1350 struct hrtimer_clock_base *base;
1351 unsigned long flags;
1352 int ret = -1;
1353
1354 /*
1355 * Check lockless first. If the timer is not active (neither
1356 * enqueued nor running the callback, nothing to do here. The
1357 * base lock does not serialize against a concurrent enqueue,
1358 * so we can avoid taking it.
1359 */
1360 if (!hrtimer_active(timer))
1361 return 0;
1362
1363 base = lock_hrtimer_base(timer, &flags);
1364
1365 if (!hrtimer_callback_running(timer))
1366 ret = remove_hrtimer(timer, base, false, false);
1367
1368 unlock_hrtimer_base(timer, &flags);
1369
1370 return ret;
1371
1372}
1373EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1374
1375#ifdef CONFIG_PREEMPT_RT
1376static void hrtimer_cpu_base_init_expiry_lock(struct hrtimer_cpu_base *base)
1377{
1378 spin_lock_init(&base->softirq_expiry_lock);
1379}
1380
1381static void hrtimer_cpu_base_lock_expiry(struct hrtimer_cpu_base *base)
1382 __acquires(&base->softirq_expiry_lock)
1383{
1384 spin_lock(&base->softirq_expiry_lock);
1385}
1386
1387static void hrtimer_cpu_base_unlock_expiry(struct hrtimer_cpu_base *base)
1388 __releases(&base->softirq_expiry_lock)
1389{
1390 spin_unlock(&base->softirq_expiry_lock);
1391}
1392
1393/*
1394 * The counterpart to hrtimer_cancel_wait_running().
1395 *
1396 * If there is a waiter for cpu_base->expiry_lock, then it was waiting for
1397 * the timer callback to finish. Drop expiry_lock and reacquire it. That
1398 * allows the waiter to acquire the lock and make progress.
1399 */
1400static void hrtimer_sync_wait_running(struct hrtimer_cpu_base *cpu_base,
1401 unsigned long flags)
1402{
1403 if (atomic_read(&cpu_base->timer_waiters)) {
1404 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1405 spin_unlock(&cpu_base->softirq_expiry_lock);
1406 spin_lock(&cpu_base->softirq_expiry_lock);
1407 raw_spin_lock_irq(&cpu_base->lock);
1408 }
1409}
1410
1411#ifdef CONFIG_SMP
1412static __always_inline bool is_migration_base(struct hrtimer_clock_base *base)
1413{
1414 return base == &migration_base;
1415}
1416#else
1417static __always_inline bool is_migration_base(struct hrtimer_clock_base *base)
1418{
1419 return false;
1420}
1421#endif
1422
1423/*
1424 * This function is called on PREEMPT_RT kernels when the fast path
1425 * deletion of a timer failed because the timer callback function was
1426 * running.
1427 *
1428 * This prevents priority inversion: if the soft irq thread is preempted
1429 * in the middle of a timer callback, then calling hrtimer_cancel() can
1430 * lead to two issues:
1431 *
1432 * - If the caller is on a remote CPU then it has to spin wait for the timer
1433 * handler to complete. This can result in unbound priority inversion.
1434 *
1435 * - If the caller originates from the task which preempted the timer
1436 * handler on the same CPU, then spin waiting for the timer handler to
1437 * complete is never going to end.
1438 */
1439void hrtimer_cancel_wait_running(const struct hrtimer *timer)
1440{
1441 /* Lockless read. Prevent the compiler from reloading it below */
1442 struct hrtimer_clock_base *base = READ_ONCE(timer->base);
1443
1444 /*
1445 * Just relax if the timer expires in hard interrupt context or if
1446 * it is currently on the migration base.
1447 */
1448 if (!timer->is_soft || is_migration_base(base)) {
1449 cpu_relax();
1450 return;
1451 }
1452
1453 /*
1454 * Mark the base as contended and grab the expiry lock, which is
1455 * held by the softirq across the timer callback. Drop the lock
1456 * immediately so the softirq can expire the next timer. In theory
1457 * the timer could already be running again, but that's more than
1458 * unlikely and just causes another wait loop.
1459 */
1460 atomic_inc(&base->cpu_base->timer_waiters);
1461 spin_lock_bh(&base->cpu_base->softirq_expiry_lock);
1462 atomic_dec(&base->cpu_base->timer_waiters);
1463 spin_unlock_bh(&base->cpu_base->softirq_expiry_lock);
1464}
1465#else
1466static inline void
1467hrtimer_cpu_base_init_expiry_lock(struct hrtimer_cpu_base *base) { }
1468static inline void
1469hrtimer_cpu_base_lock_expiry(struct hrtimer_cpu_base *base) { }
1470static inline void
1471hrtimer_cpu_base_unlock_expiry(struct hrtimer_cpu_base *base) { }
1472static inline void hrtimer_sync_wait_running(struct hrtimer_cpu_base *base,
1473 unsigned long flags) { }
1474#endif
1475
1476/**
1477 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1478 * @timer: the timer to be cancelled
1479 *
1480 * Returns:
1481 * 0 when the timer was not active
1482 * 1 when the timer was active
1483 */
1484int hrtimer_cancel(struct hrtimer *timer)
1485{
1486 int ret;
1487
1488 do {
1489 ret = hrtimer_try_to_cancel(timer);
1490
1491 if (ret < 0)
1492 hrtimer_cancel_wait_running(timer);
1493 } while (ret < 0);
1494 return ret;
1495}
1496EXPORT_SYMBOL_GPL(hrtimer_cancel);
1497
1498/**
1499 * __hrtimer_get_remaining - get remaining time for the timer
1500 * @timer: the timer to read
1501 * @adjust: adjust relative timers when CONFIG_TIME_LOW_RES=y
1502 */
1503ktime_t __hrtimer_get_remaining(const struct hrtimer *timer, bool adjust)
1504{
1505 unsigned long flags;
1506 ktime_t rem;
1507
1508 lock_hrtimer_base(timer, &flags);
1509 if (IS_ENABLED(CONFIG_TIME_LOW_RES) && adjust)
1510 rem = hrtimer_expires_remaining_adjusted(timer);
1511 else
1512 rem = hrtimer_expires_remaining(timer);
1513 unlock_hrtimer_base(timer, &flags);
1514
1515 return rem;
1516}
1517EXPORT_SYMBOL_GPL(__hrtimer_get_remaining);
1518
1519#ifdef CONFIG_NO_HZ_COMMON
1520/**
1521 * hrtimer_get_next_event - get the time until next expiry event
1522 *
1523 * Returns the next expiry time or KTIME_MAX if no timer is pending.
1524 */
1525u64 hrtimer_get_next_event(void)
1526{
1527 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1528 u64 expires = KTIME_MAX;
1529 unsigned long flags;
1530
1531 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1532
1533 if (!hrtimer_hres_active(cpu_base))
1534 expires = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_ALL);
1535
1536 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1537
1538 return expires;
1539}
1540
1541/**
1542 * hrtimer_next_event_without - time until next expiry event w/o one timer
1543 * @exclude: timer to exclude
1544 *
1545 * Returns the next expiry time over all timers except for the @exclude one or
1546 * KTIME_MAX if none of them is pending.
1547 */
1548u64 hrtimer_next_event_without(const struct hrtimer *exclude)
1549{
1550 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1551 u64 expires = KTIME_MAX;
1552 unsigned long flags;
1553
1554 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1555
1556 if (hrtimer_hres_active(cpu_base)) {
1557 unsigned int active;
1558
1559 if (!cpu_base->softirq_activated) {
1560 active = cpu_base->active_bases & HRTIMER_ACTIVE_SOFT;
1561 expires = __hrtimer_next_event_base(cpu_base, exclude,
1562 active, KTIME_MAX);
1563 }
1564 active = cpu_base->active_bases & HRTIMER_ACTIVE_HARD;
1565 expires = __hrtimer_next_event_base(cpu_base, exclude, active,
1566 expires);
1567 }
1568
1569 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1570
1571 return expires;
1572}
1573#endif
1574
1575static inline int hrtimer_clockid_to_base(clockid_t clock_id)
1576{
1577 switch (clock_id) {
1578 case CLOCK_MONOTONIC:
1579 return HRTIMER_BASE_MONOTONIC;
1580 case CLOCK_REALTIME:
1581 return HRTIMER_BASE_REALTIME;
1582 case CLOCK_BOOTTIME:
1583 return HRTIMER_BASE_BOOTTIME;
1584 case CLOCK_TAI:
1585 return HRTIMER_BASE_TAI;
1586 default:
1587 WARN(1, "Invalid clockid %d. Using MONOTONIC\n", clock_id);
1588 return HRTIMER_BASE_MONOTONIC;
1589 }
1590}
1591
1592static ktime_t __hrtimer_cb_get_time(clockid_t clock_id)
1593{
1594 switch (clock_id) {
1595 case CLOCK_MONOTONIC:
1596 return ktime_get();
1597 case CLOCK_REALTIME:
1598 return ktime_get_real();
1599 case CLOCK_BOOTTIME:
1600 return ktime_get_boottime();
1601 case CLOCK_TAI:
1602 return ktime_get_clocktai();
1603 default:
1604 WARN(1, "Invalid clockid %d. Using MONOTONIC\n", clock_id);
1605 return ktime_get();
1606 }
1607}
1608
1609ktime_t hrtimer_cb_get_time(const struct hrtimer *timer)
1610{
1611 return __hrtimer_cb_get_time(timer->base->clockid);
1612}
1613EXPORT_SYMBOL_GPL(hrtimer_cb_get_time);
1614
1615static void __hrtimer_setup(struct hrtimer *timer,
1616 enum hrtimer_restart (*function)(struct hrtimer *),
1617 clockid_t clock_id, enum hrtimer_mode mode)
1618{
1619 bool softtimer = !!(mode & HRTIMER_MODE_SOFT);
1620 struct hrtimer_cpu_base *cpu_base;
1621 int base;
1622
1623 /*
1624 * On PREEMPT_RT enabled kernels hrtimers which are not explicitly
1625 * marked for hard interrupt expiry mode are moved into soft
1626 * interrupt context for latency reasons and because the callbacks
1627 * can invoke functions which might sleep on RT, e.g. spin_lock().
1628 */
1629 if (IS_ENABLED(CONFIG_PREEMPT_RT) && !(mode & HRTIMER_MODE_HARD))
1630 softtimer = true;
1631
1632 memset(timer, 0, sizeof(struct hrtimer));
1633
1634 cpu_base = raw_cpu_ptr(&hrtimer_bases);
1635
1636 /*
1637 * POSIX magic: Relative CLOCK_REALTIME timers are not affected by
1638 * clock modifications, so they needs to become CLOCK_MONOTONIC to
1639 * ensure POSIX compliance.
1640 */
1641 if (clock_id == CLOCK_REALTIME && mode & HRTIMER_MODE_REL)
1642 clock_id = CLOCK_MONOTONIC;
1643
1644 base = softtimer ? HRTIMER_MAX_CLOCK_BASES / 2 : 0;
1645 base += hrtimer_clockid_to_base(clock_id);
1646 timer->is_soft = softtimer;
1647 timer->is_hard = !!(mode & HRTIMER_MODE_HARD);
1648 timer->base = &cpu_base->clock_base[base];
1649 timerqueue_init(&timer->node);
1650
1651 if (WARN_ON_ONCE(!function))
1652 ACCESS_PRIVATE(timer, function) = hrtimer_dummy_timeout;
1653 else
1654 ACCESS_PRIVATE(timer, function) = function;
1655}
1656
1657/**
1658 * hrtimer_setup - initialize a timer to the given clock
1659 * @timer: the timer to be initialized
1660 * @function: the callback function
1661 * @clock_id: the clock to be used
1662 * @mode: The modes which are relevant for initialization:
1663 * HRTIMER_MODE_ABS, HRTIMER_MODE_REL, HRTIMER_MODE_ABS_SOFT,
1664 * HRTIMER_MODE_REL_SOFT
1665 *
1666 * The PINNED variants of the above can be handed in,
1667 * but the PINNED bit is ignored as pinning happens
1668 * when the hrtimer is started
1669 */
1670void hrtimer_setup(struct hrtimer *timer, enum hrtimer_restart (*function)(struct hrtimer *),
1671 clockid_t clock_id, enum hrtimer_mode mode)
1672{
1673 debug_setup(timer, clock_id, mode);
1674 __hrtimer_setup(timer, function, clock_id, mode);
1675}
1676EXPORT_SYMBOL_GPL(hrtimer_setup);
1677
1678/**
1679 * hrtimer_setup_on_stack - initialize a timer on stack memory
1680 * @timer: The timer to be initialized
1681 * @function: the callback function
1682 * @clock_id: The clock to be used
1683 * @mode: The timer mode
1684 *
1685 * Similar to hrtimer_setup(), except that this one must be used if struct hrtimer is in stack
1686 * memory.
1687 */
1688void hrtimer_setup_on_stack(struct hrtimer *timer,
1689 enum hrtimer_restart (*function)(struct hrtimer *),
1690 clockid_t clock_id, enum hrtimer_mode mode)
1691{
1692 debug_setup_on_stack(timer, clock_id, mode);
1693 __hrtimer_setup(timer, function, clock_id, mode);
1694}
1695EXPORT_SYMBOL_GPL(hrtimer_setup_on_stack);
1696
1697/*
1698 * A timer is active, when it is enqueued into the rbtree or the
1699 * callback function is running or it's in the state of being migrated
1700 * to another cpu.
1701 *
1702 * It is important for this function to not return a false negative.
1703 */
1704bool hrtimer_active(const struct hrtimer *timer)
1705{
1706 struct hrtimer_clock_base *base;
1707 unsigned int seq;
1708
1709 do {
1710 base = READ_ONCE(timer->base);
1711 seq = raw_read_seqcount_begin(&base->seq);
1712
1713 if (timer->state != HRTIMER_STATE_INACTIVE ||
1714 base->running == timer)
1715 return true;
1716
1717 } while (read_seqcount_retry(&base->seq, seq) ||
1718 base != READ_ONCE(timer->base));
1719
1720 return false;
1721}
1722EXPORT_SYMBOL_GPL(hrtimer_active);
1723
1724/*
1725 * The write_seqcount_barrier()s in __run_hrtimer() split the thing into 3
1726 * distinct sections:
1727 *
1728 * - queued: the timer is queued
1729 * - callback: the timer is being ran
1730 * - post: the timer is inactive or (re)queued
1731 *
1732 * On the read side we ensure we observe timer->state and cpu_base->running
1733 * from the same section, if anything changed while we looked at it, we retry.
1734 * This includes timer->base changing because sequence numbers alone are
1735 * insufficient for that.
1736 *
1737 * The sequence numbers are required because otherwise we could still observe
1738 * a false negative if the read side got smeared over multiple consecutive
1739 * __run_hrtimer() invocations.
1740 */
1741
1742static void __run_hrtimer(struct hrtimer_cpu_base *cpu_base,
1743 struct hrtimer_clock_base *base,
1744 struct hrtimer *timer, ktime_t *now,
1745 unsigned long flags) __must_hold(&cpu_base->lock)
1746{
1747 enum hrtimer_restart (*fn)(struct hrtimer *);
1748 bool expires_in_hardirq;
1749 int restart;
1750
1751 lockdep_assert_held(&cpu_base->lock);
1752
1753 debug_hrtimer_deactivate(timer);
1754 base->running = timer;
1755
1756 /*
1757 * Separate the ->running assignment from the ->state assignment.
1758 *
1759 * As with a regular write barrier, this ensures the read side in
1760 * hrtimer_active() cannot observe base->running == NULL &&
1761 * timer->state == INACTIVE.
1762 */
1763 raw_write_seqcount_barrier(&base->seq);
1764
1765 __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE, 0);
1766 fn = ACCESS_PRIVATE(timer, function);
1767
1768 /*
1769 * Clear the 'is relative' flag for the TIME_LOW_RES case. If the
1770 * timer is restarted with a period then it becomes an absolute
1771 * timer. If its not restarted it does not matter.
1772 */
1773 if (IS_ENABLED(CONFIG_TIME_LOW_RES))
1774 timer->is_rel = false;
1775
1776 /*
1777 * The timer is marked as running in the CPU base, so it is
1778 * protected against migration to a different CPU even if the lock
1779 * is dropped.
1780 */
1781 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1782 trace_hrtimer_expire_entry(timer, now);
1783 expires_in_hardirq = lockdep_hrtimer_enter(timer);
1784
1785 restart = fn(timer);
1786
1787 lockdep_hrtimer_exit(expires_in_hardirq);
1788 trace_hrtimer_expire_exit(timer);
1789 raw_spin_lock_irq(&cpu_base->lock);
1790
1791 /*
1792 * Note: We clear the running state after enqueue_hrtimer and
1793 * we do not reprogram the event hardware. Happens either in
1794 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1795 *
1796 * Note: Because we dropped the cpu_base->lock above,
1797 * hrtimer_start_range_ns() can have popped in and enqueued the timer
1798 * for us already.
1799 */
1800 if (restart != HRTIMER_NORESTART &&
1801 !(timer->state & HRTIMER_STATE_ENQUEUED))
1802 enqueue_hrtimer(timer, base, HRTIMER_MODE_ABS);
1803
1804 /*
1805 * Separate the ->running assignment from the ->state assignment.
1806 *
1807 * As with a regular write barrier, this ensures the read side in
1808 * hrtimer_active() cannot observe base->running.timer == NULL &&
1809 * timer->state == INACTIVE.
1810 */
1811 raw_write_seqcount_barrier(&base->seq);
1812
1813 WARN_ON_ONCE(base->running != timer);
1814 base->running = NULL;
1815}
1816
1817static void __hrtimer_run_queues(struct hrtimer_cpu_base *cpu_base, ktime_t now,
1818 unsigned long flags, unsigned int active_mask)
1819{
1820 struct hrtimer_clock_base *base;
1821 unsigned int active = cpu_base->active_bases & active_mask;
1822
1823 for_each_active_base(base, cpu_base, active) {
1824 struct timerqueue_node *node;
1825 ktime_t basenow;
1826
1827 basenow = ktime_add(now, base->offset);
1828
1829 while ((node = timerqueue_getnext(&base->active))) {
1830 struct hrtimer *timer;
1831
1832 timer = container_of(node, struct hrtimer, node);
1833
1834 /*
1835 * The immediate goal for using the softexpires is
1836 * minimizing wakeups, not running timers at the
1837 * earliest interrupt after their soft expiration.
1838 * This allows us to avoid using a Priority Search
1839 * Tree, which can answer a stabbing query for
1840 * overlapping intervals and instead use the simple
1841 * BST we already have.
1842 * We don't add extra wakeups by delaying timers that
1843 * are right-of a not yet expired timer, because that
1844 * timer will have to trigger a wakeup anyway.
1845 */
1846 if (basenow < hrtimer_get_softexpires(timer))
1847 break;
1848
1849 __run_hrtimer(cpu_base, base, timer, &basenow, flags);
1850 if (active_mask == HRTIMER_ACTIVE_SOFT)
1851 hrtimer_sync_wait_running(cpu_base, flags);
1852 }
1853 }
1854}
1855
1856static __latent_entropy void hrtimer_run_softirq(void)
1857{
1858 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1859 unsigned long flags;
1860 ktime_t now;
1861
1862 hrtimer_cpu_base_lock_expiry(cpu_base);
1863 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1864
1865 now = hrtimer_update_base(cpu_base);
1866 __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_SOFT);
1867
1868 cpu_base->softirq_activated = 0;
1869 hrtimer_update_softirq_timer(cpu_base, true);
1870
1871 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1872 hrtimer_cpu_base_unlock_expiry(cpu_base);
1873}
1874
1875#ifdef CONFIG_HIGH_RES_TIMERS
1876
1877/*
1878 * High resolution timer interrupt
1879 * Called with interrupts disabled
1880 */
1881void hrtimer_interrupt(struct clock_event_device *dev)
1882{
1883 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1884 ktime_t expires_next, now, entry_time, delta;
1885 unsigned long flags;
1886 int retries = 0;
1887
1888 BUG_ON(!cpu_base->hres_active);
1889 cpu_base->nr_events++;
1890 dev->next_event = KTIME_MAX;
1891 dev->next_event_forced = 0;
1892
1893 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1894 entry_time = now = hrtimer_update_base(cpu_base);
1895retry:
1896 cpu_base->in_hrtirq = 1;
1897 /*
1898 * We set expires_next to KTIME_MAX here with cpu_base->lock
1899 * held to prevent that a timer is enqueued in our queue via
1900 * the migration code. This does not affect enqueueing of
1901 * timers which run their callback and need to be requeued on
1902 * this CPU.
1903 */
1904 cpu_base->expires_next = KTIME_MAX;
1905
1906 if (!ktime_before(now, cpu_base->softirq_expires_next)) {
1907 cpu_base->softirq_expires_next = KTIME_MAX;
1908 cpu_base->softirq_activated = 1;
1909 raise_timer_softirq(HRTIMER_SOFTIRQ);
1910 }
1911
1912 __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_HARD);
1913
1914 /* Reevaluate the clock bases for the [soft] next expiry */
1915 expires_next = hrtimer_update_next_event(cpu_base);
1916 /*
1917 * Store the new expiry value so the migration code can verify
1918 * against it.
1919 */
1920 cpu_base->expires_next = expires_next;
1921 cpu_base->in_hrtirq = 0;
1922 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1923
1924 /* Reprogramming necessary ? */
1925 if (!tick_program_event(expires_next, 0)) {
1926 cpu_base->hang_detected = 0;
1927 return;
1928 }
1929
1930 /*
1931 * The next timer was already expired due to:
1932 * - tracing
1933 * - long lasting callbacks
1934 * - being scheduled away when running in a VM
1935 *
1936 * We need to prevent that we loop forever in the hrtimer
1937 * interrupt routine. We give it 3 attempts to avoid
1938 * overreacting on some spurious event.
1939 *
1940 * Acquire base lock for updating the offsets and retrieving
1941 * the current time.
1942 */
1943 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1944 now = hrtimer_update_base(cpu_base);
1945 cpu_base->nr_retries++;
1946 if (++retries < 3)
1947 goto retry;
1948 /*
1949 * Give the system a chance to do something else than looping
1950 * here. We stored the entry time, so we know exactly how long
1951 * we spent here. We schedule the next event this amount of
1952 * time away.
1953 */
1954 cpu_base->nr_hangs++;
1955 cpu_base->hang_detected = 1;
1956 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1957
1958 delta = ktime_sub(now, entry_time);
1959 if ((unsigned int)delta > cpu_base->max_hang_time)
1960 cpu_base->max_hang_time = (unsigned int) delta;
1961 /*
1962 * Limit it to a sensible value as we enforce a longer
1963 * delay. Give the CPU at least 100ms to catch up.
1964 */
1965 if (delta > 100 * NSEC_PER_MSEC)
1966 expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1967 else
1968 expires_next = ktime_add(now, delta);
1969 tick_program_event(expires_next, 1);
1970 pr_warn_once("hrtimer: interrupt took %llu ns\n", ktime_to_ns(delta));
1971}
1972#endif /* !CONFIG_HIGH_RES_TIMERS */
1973
1974/*
1975 * Called from run_local_timers in hardirq context every jiffy
1976 */
1977void hrtimer_run_queues(void)
1978{
1979 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1980 unsigned long flags;
1981 ktime_t now;
1982
1983 if (hrtimer_hres_active(cpu_base))
1984 return;
1985
1986 /*
1987 * This _is_ ugly: We have to check periodically, whether we
1988 * can switch to highres and / or nohz mode. The clocksource
1989 * switch happens with xtime_lock held. Notification from
1990 * there only sets the check bit in the tick_oneshot code,
1991 * otherwise we might deadlock vs. xtime_lock.
1992 */
1993 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) {
1994 hrtimer_switch_to_hres();
1995 return;
1996 }
1997
1998 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1999 now = hrtimer_update_base(cpu_base);
2000
2001 if (!ktime_before(now, cpu_base->softirq_expires_next)) {
2002 cpu_base->softirq_expires_next = KTIME_MAX;
2003 cpu_base->softirq_activated = 1;
2004 raise_timer_softirq(HRTIMER_SOFTIRQ);
2005 }
2006
2007 __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_HARD);
2008 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
2009}
2010
2011/*
2012 * Sleep related functions:
2013 */
2014static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
2015{
2016 struct hrtimer_sleeper *t =
2017 container_of(timer, struct hrtimer_sleeper, timer);
2018 struct task_struct *task = t->task;
2019
2020 t->task = NULL;
2021 if (task)
2022 wake_up_process(task);
2023
2024 return HRTIMER_NORESTART;
2025}
2026
2027/**
2028 * hrtimer_sleeper_start_expires - Start a hrtimer sleeper timer
2029 * @sl: sleeper to be started
2030 * @mode: timer mode abs/rel
2031 *
2032 * Wrapper around hrtimer_start_expires() for hrtimer_sleeper based timers
2033 * to allow PREEMPT_RT to tweak the delivery mode (soft/hardirq context)
2034 */
2035void hrtimer_sleeper_start_expires(struct hrtimer_sleeper *sl,
2036 enum hrtimer_mode mode)
2037{
2038 /*
2039 * Make the enqueue delivery mode check work on RT. If the sleeper
2040 * was initialized for hard interrupt delivery, force the mode bit.
2041 * This is a special case for hrtimer_sleepers because
2042 * __hrtimer_setup_sleeper() determines the delivery mode on RT so the
2043 * fiddling with this decision is avoided at the call sites.
2044 */
2045 if (IS_ENABLED(CONFIG_PREEMPT_RT) && sl->timer.is_hard)
2046 mode |= HRTIMER_MODE_HARD;
2047
2048 hrtimer_start_expires(&sl->timer, mode);
2049}
2050EXPORT_SYMBOL_GPL(hrtimer_sleeper_start_expires);
2051
2052static void __hrtimer_setup_sleeper(struct hrtimer_sleeper *sl,
2053 clockid_t clock_id, enum hrtimer_mode mode)
2054{
2055 /*
2056 * On PREEMPT_RT enabled kernels hrtimers which are not explicitly
2057 * marked for hard interrupt expiry mode are moved into soft
2058 * interrupt context either for latency reasons or because the
2059 * hrtimer callback takes regular spinlocks or invokes other
2060 * functions which are not suitable for hard interrupt context on
2061 * PREEMPT_RT.
2062 *
2063 * The hrtimer_sleeper callback is RT compatible in hard interrupt
2064 * context, but there is a latency concern: Untrusted userspace can
2065 * spawn many threads which arm timers for the same expiry time on
2066 * the same CPU. That causes a latency spike due to the wakeup of
2067 * a gazillion threads.
2068 *
2069 * OTOH, privileged real-time user space applications rely on the
2070 * low latency of hard interrupt wakeups. If the current task is in
2071 * a real-time scheduling class, mark the mode for hard interrupt
2072 * expiry.
2073 */
2074 if (IS_ENABLED(CONFIG_PREEMPT_RT)) {
2075 if (rt_or_dl_task_policy(current) && !(mode & HRTIMER_MODE_SOFT))
2076 mode |= HRTIMER_MODE_HARD;
2077 }
2078
2079 __hrtimer_setup(&sl->timer, hrtimer_wakeup, clock_id, mode);
2080 sl->task = current;
2081}
2082
2083/**
2084 * hrtimer_setup_sleeper_on_stack - initialize a sleeper in stack memory
2085 * @sl: sleeper to be initialized
2086 * @clock_id: the clock to be used
2087 * @mode: timer mode abs/rel
2088 */
2089void hrtimer_setup_sleeper_on_stack(struct hrtimer_sleeper *sl,
2090 clockid_t clock_id, enum hrtimer_mode mode)
2091{
2092 debug_setup_on_stack(&sl->timer, clock_id, mode);
2093 __hrtimer_setup_sleeper(sl, clock_id, mode);
2094}
2095EXPORT_SYMBOL_GPL(hrtimer_setup_sleeper_on_stack);
2096
2097int nanosleep_copyout(struct restart_block *restart, struct timespec64 *ts)
2098{
2099 switch(restart->nanosleep.type) {
2100#ifdef CONFIG_COMPAT_32BIT_TIME
2101 case TT_COMPAT:
2102 if (put_old_timespec32(ts, restart->nanosleep.compat_rmtp))
2103 return -EFAULT;
2104 break;
2105#endif
2106 case TT_NATIVE:
2107 if (put_timespec64(ts, restart->nanosleep.rmtp))
2108 return -EFAULT;
2109 break;
2110 default:
2111 BUG();
2112 }
2113 return -ERESTART_RESTARTBLOCK;
2114}
2115
2116static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
2117{
2118 struct restart_block *restart;
2119
2120 do {
2121 set_current_state(TASK_INTERRUPTIBLE|TASK_FREEZABLE);
2122 hrtimer_sleeper_start_expires(t, mode);
2123
2124 if (likely(t->task))
2125 schedule();
2126
2127 hrtimer_cancel(&t->timer);
2128 mode = HRTIMER_MODE_ABS;
2129
2130 } while (t->task && !signal_pending(current));
2131
2132 __set_current_state(TASK_RUNNING);
2133
2134 if (!t->task)
2135 return 0;
2136
2137 restart = &current->restart_block;
2138 if (restart->nanosleep.type != TT_NONE) {
2139 ktime_t rem = hrtimer_expires_remaining(&t->timer);
2140 struct timespec64 rmt;
2141
2142 if (rem <= 0)
2143 return 0;
2144 rmt = ktime_to_timespec64(rem);
2145
2146 return nanosleep_copyout(restart, &rmt);
2147 }
2148 return -ERESTART_RESTARTBLOCK;
2149}
2150
2151static long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
2152{
2153 struct hrtimer_sleeper t;
2154 int ret;
2155
2156 hrtimer_setup_sleeper_on_stack(&t, restart->nanosleep.clockid, HRTIMER_MODE_ABS);
2157 hrtimer_set_expires(&t.timer, restart->nanosleep.expires);
2158 ret = do_nanosleep(&t, HRTIMER_MODE_ABS);
2159 destroy_hrtimer_on_stack(&t.timer);
2160 return ret;
2161}
2162
2163long hrtimer_nanosleep(ktime_t rqtp, const enum hrtimer_mode mode,
2164 const clockid_t clockid)
2165{
2166 struct restart_block *restart;
2167 struct hrtimer_sleeper t;
2168 int ret = 0;
2169
2170 hrtimer_setup_sleeper_on_stack(&t, clockid, mode);
2171 hrtimer_set_expires_range_ns(&t.timer, rqtp, current->timer_slack_ns);
2172 ret = do_nanosleep(&t, mode);
2173 if (ret != -ERESTART_RESTARTBLOCK)
2174 goto out;
2175
2176 /* Absolute timers do not update the rmtp value and restart: */
2177 if (mode == HRTIMER_MODE_ABS) {
2178 ret = -ERESTARTNOHAND;
2179 goto out;
2180 }
2181
2182 restart = &current->restart_block;
2183 restart->nanosleep.clockid = t.timer.base->clockid;
2184 restart->nanosleep.expires = hrtimer_get_expires(&t.timer);
2185 set_restart_fn(restart, hrtimer_nanosleep_restart);
2186out:
2187 destroy_hrtimer_on_stack(&t.timer);
2188 return ret;
2189}
2190
2191#ifdef CONFIG_64BIT
2192
2193SYSCALL_DEFINE2(nanosleep, struct __kernel_timespec __user *, rqtp,
2194 struct __kernel_timespec __user *, rmtp)
2195{
2196 struct timespec64 tu;
2197
2198 if (get_timespec64(&tu, rqtp))
2199 return -EFAULT;
2200
2201 if (!timespec64_valid(&tu))
2202 return -EINVAL;
2203
2204 current->restart_block.fn = do_no_restart_syscall;
2205 current->restart_block.nanosleep.type = rmtp ? TT_NATIVE : TT_NONE;
2206 current->restart_block.nanosleep.rmtp = rmtp;
2207 return hrtimer_nanosleep(timespec64_to_ktime(tu), HRTIMER_MODE_REL,
2208 CLOCK_MONOTONIC);
2209}
2210
2211#endif
2212
2213#ifdef CONFIG_COMPAT_32BIT_TIME
2214
2215SYSCALL_DEFINE2(nanosleep_time32, struct old_timespec32 __user *, rqtp,
2216 struct old_timespec32 __user *, rmtp)
2217{
2218 struct timespec64 tu;
2219
2220 if (get_old_timespec32(&tu, rqtp))
2221 return -EFAULT;
2222
2223 if (!timespec64_valid(&tu))
2224 return -EINVAL;
2225
2226 current->restart_block.fn = do_no_restart_syscall;
2227 current->restart_block.nanosleep.type = rmtp ? TT_COMPAT : TT_NONE;
2228 current->restart_block.nanosleep.compat_rmtp = rmtp;
2229 return hrtimer_nanosleep(timespec64_to_ktime(tu), HRTIMER_MODE_REL,
2230 CLOCK_MONOTONIC);
2231}
2232#endif
2233
2234/*
2235 * Functions related to boot-time initialization:
2236 */
2237int hrtimers_prepare_cpu(unsigned int cpu)
2238{
2239 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
2240 int i;
2241
2242 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
2243 struct hrtimer_clock_base *clock_b = &cpu_base->clock_base[i];
2244
2245 clock_b->cpu_base = cpu_base;
2246 seqcount_raw_spinlock_init(&clock_b->seq, &cpu_base->lock);
2247 timerqueue_init_head(&clock_b->active);
2248 }
2249
2250 cpu_base->cpu = cpu;
2251 hrtimer_cpu_base_init_expiry_lock(cpu_base);
2252 return 0;
2253}
2254
2255int hrtimers_cpu_starting(unsigned int cpu)
2256{
2257 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
2258
2259 /* Clear out any left over state from a CPU down operation */
2260 cpu_base->active_bases = 0;
2261 cpu_base->hres_active = 0;
2262 cpu_base->hang_detected = 0;
2263 cpu_base->next_timer = NULL;
2264 cpu_base->softirq_next_timer = NULL;
2265 cpu_base->expires_next = KTIME_MAX;
2266 cpu_base->softirq_expires_next = KTIME_MAX;
2267 cpu_base->online = 1;
2268 return 0;
2269}
2270
2271#ifdef CONFIG_HOTPLUG_CPU
2272
2273static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
2274 struct hrtimer_clock_base *new_base)
2275{
2276 struct hrtimer *timer;
2277 struct timerqueue_node *node;
2278
2279 while ((node = timerqueue_getnext(&old_base->active))) {
2280 timer = container_of(node, struct hrtimer, node);
2281 BUG_ON(hrtimer_callback_running(timer));
2282 debug_deactivate(timer);
2283
2284 /*
2285 * Mark it as ENQUEUED not INACTIVE otherwise the
2286 * timer could be seen as !active and just vanish away
2287 * under us on another CPU
2288 */
2289 __remove_hrtimer(timer, old_base, HRTIMER_STATE_ENQUEUED, 0);
2290 timer->base = new_base;
2291 /*
2292 * Enqueue the timers on the new cpu. This does not
2293 * reprogram the event device in case the timer
2294 * expires before the earliest on this CPU, but we run
2295 * hrtimer_interrupt after we migrated everything to
2296 * sort out already expired timers and reprogram the
2297 * event device.
2298 */
2299 enqueue_hrtimer(timer, new_base, HRTIMER_MODE_ABS);
2300 }
2301}
2302
2303int hrtimers_cpu_dying(unsigned int dying_cpu)
2304{
2305 int i, ncpu = cpumask_any_and(cpu_active_mask, housekeeping_cpumask(HK_TYPE_TIMER));
2306 struct hrtimer_cpu_base *old_base, *new_base;
2307
2308 old_base = this_cpu_ptr(&hrtimer_bases);
2309 new_base = &per_cpu(hrtimer_bases, ncpu);
2310
2311 /*
2312 * The caller is globally serialized and nobody else
2313 * takes two locks at once, deadlock is not possible.
2314 */
2315 raw_spin_lock(&old_base->lock);
2316 raw_spin_lock_nested(&new_base->lock, SINGLE_DEPTH_NESTING);
2317
2318 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
2319 migrate_hrtimer_list(&old_base->clock_base[i],
2320 &new_base->clock_base[i]);
2321 }
2322
2323 /* Tell the other CPU to retrigger the next event */
2324 smp_call_function_single(ncpu, retrigger_next_event, NULL, 0);
2325
2326 raw_spin_unlock(&new_base->lock);
2327 old_base->online = 0;
2328 raw_spin_unlock(&old_base->lock);
2329
2330 return 0;
2331}
2332
2333#endif /* CONFIG_HOTPLUG_CPU */
2334
2335void __init hrtimers_init(void)
2336{
2337 hrtimers_prepare_cpu(smp_processor_id());
2338 hrtimers_cpu_starting(smp_processor_id());
2339 open_softirq(HRTIMER_SOFTIRQ, hrtimer_run_softirq);
2340}