1 // SPDX-License-Identifier: GPL-2.0
3 * Kernel internal timers
5 * Copyright (C) 1991, 1992 Linus Torvalds
7 * 1997-01-28 Modified by Finn Arne Gangstad to make timers scale better.
9 * 1997-09-10 Updated NTP code according to technical memorandum Jan '96
10 * "A Kernel Model for Precision Timekeeping" by Dave Mills
11 * 1998-12-24 Fixed a xtime SMP race (we need the xtime_lock rw spinlock to
12 * serialize accesses to xtime/lost_ticks).
13 * Copyright (C) 1998 Andrea Arcangeli
14 * 1999-03-10 Improved NTP compatibility by Ulrich Windl
15 * 2002-05-31 Move sys_sysinfo here and make its locking sane, Robert Love
16 * 2000-10-05 Implemented scalable SMP per-CPU timer handling.
17 * Copyright (C) 2000, 2001, 2002 Ingo Molnar
18 * Designed by David S. Miller, Alexey Kuznetsov and Ingo Molnar
21 #include <linux/kernel_stat.h>
22 #include <linux/export.h>
23 #include <linux/interrupt.h>
24 #include <linux/percpu.h>
25 #include <linux/init.h>
27 #include <linux/swap.h>
28 #include <linux/pid_namespace.h>
29 #include <linux/notifier.h>
30 #include <linux/thread_info.h>
31 #include <linux/time.h>
32 #include <linux/jiffies.h>
33 #include <linux/posix-timers.h>
34 #include <linux/cpu.h>
35 #include <linux/syscalls.h>
36 #include <linux/delay.h>
37 #include <linux/tick.h>
38 #include <linux/kallsyms.h>
39 #include <linux/irq_work.h>
40 #include <linux/sched/signal.h>
41 #include <linux/sched/sysctl.h>
42 #include <linux/sched/nohz.h>
43 #include <linux/sched/debug.h>
44 #include <linux/slab.h>
45 #include <linux/compat.h>
46 #include <linux/random.h>
47 #include <linux/sysctl.h>
49 #include <linux/uaccess.h>
50 #include <asm/unistd.h>
51 #include <asm/div64.h>
52 #include <asm/timex.h>
55 #include "tick-internal.h"
57 #define CREATE_TRACE_POINTS
58 #include <trace/events/timer.h>
60 __visible u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
62 EXPORT_SYMBOL(jiffies_64);
65 * The timer wheel has LVL_DEPTH array levels. Each level provides an array of
66 * LVL_SIZE buckets. Each level is driven by its own clock and therefor each
67 * level has a different granularity.
69 * The level granularity is: LVL_CLK_DIV ^ lvl
70 * The level clock frequency is: HZ / (LVL_CLK_DIV ^ level)
72 * The array level of a newly armed timer depends on the relative expiry
73 * time. The farther the expiry time is away the higher the array level and
74 * therefor the granularity becomes.
76 * Contrary to the original timer wheel implementation, which aims for 'exact'
77 * expiry of the timers, this implementation removes the need for recascading
78 * the timers into the lower array levels. The previous 'classic' timer wheel
79 * implementation of the kernel already violated the 'exact' expiry by adding
80 * slack to the expiry time to provide batched expiration. The granularity
81 * levels provide implicit batching.
83 * This is an optimization of the original timer wheel implementation for the
84 * majority of the timer wheel use cases: timeouts. The vast majority of
85 * timeout timers (networking, disk I/O ...) are canceled before expiry. If
86 * the timeout expires it indicates that normal operation is disturbed, so it
87 * does not matter much whether the timeout comes with a slight delay.
89 * The only exception to this are networking timers with a small expiry
90 * time. They rely on the granularity. Those fit into the first wheel level,
91 * which has HZ granularity.
93 * We don't have cascading anymore. timers with a expiry time above the
94 * capacity of the last wheel level are force expired at the maximum timeout
95 * value of the last wheel level. From data sampling we know that the maximum
96 * value observed is 5 days (network connection tracking), so this should not
99 * The currently chosen array constants values are a good compromise between
100 * array size and granularity.
102 * This results in the following granularity and range levels:
105 * Level Offset Granularity Range
106 * 0 0 1 ms 0 ms - 63 ms
107 * 1 64 8 ms 64 ms - 511 ms
108 * 2 128 64 ms 512 ms - 4095 ms (512ms - ~4s)
109 * 3 192 512 ms 4096 ms - 32767 ms (~4s - ~32s)
110 * 4 256 4096 ms (~4s) 32768 ms - 262143 ms (~32s - ~4m)
111 * 5 320 32768 ms (~32s) 262144 ms - 2097151 ms (~4m - ~34m)
112 * 6 384 262144 ms (~4m) 2097152 ms - 16777215 ms (~34m - ~4h)
113 * 7 448 2097152 ms (~34m) 16777216 ms - 134217727 ms (~4h - ~1d)
114 * 8 512 16777216 ms (~4h) 134217728 ms - 1073741822 ms (~1d - ~12d)
117 * Level Offset Granularity Range
118 * 0 0 3 ms 0 ms - 210 ms
119 * 1 64 26 ms 213 ms - 1703 ms (213ms - ~1s)
120 * 2 128 213 ms 1706 ms - 13650 ms (~1s - ~13s)
121 * 3 192 1706 ms (~1s) 13653 ms - 109223 ms (~13s - ~1m)
122 * 4 256 13653 ms (~13s) 109226 ms - 873810 ms (~1m - ~14m)
123 * 5 320 109226 ms (~1m) 873813 ms - 6990503 ms (~14m - ~1h)
124 * 6 384 873813 ms (~14m) 6990506 ms - 55924050 ms (~1h - ~15h)
125 * 7 448 6990506 ms (~1h) 55924053 ms - 447392423 ms (~15h - ~5d)
126 * 8 512 55924053 ms (~15h) 447392426 ms - 3579139406 ms (~5d - ~41d)
129 * Level Offset Granularity Range
130 * 0 0 4 ms 0 ms - 255 ms
131 * 1 64 32 ms 256 ms - 2047 ms (256ms - ~2s)
132 * 2 128 256 ms 2048 ms - 16383 ms (~2s - ~16s)
133 * 3 192 2048 ms (~2s) 16384 ms - 131071 ms (~16s - ~2m)
134 * 4 256 16384 ms (~16s) 131072 ms - 1048575 ms (~2m - ~17m)
135 * 5 320 131072 ms (~2m) 1048576 ms - 8388607 ms (~17m - ~2h)
136 * 6 384 1048576 ms (~17m) 8388608 ms - 67108863 ms (~2h - ~18h)
137 * 7 448 8388608 ms (~2h) 67108864 ms - 536870911 ms (~18h - ~6d)
138 * 8 512 67108864 ms (~18h) 536870912 ms - 4294967288 ms (~6d - ~49d)
141 * Level Offset Granularity Range
142 * 0 0 10 ms 0 ms - 630 ms
143 * 1 64 80 ms 640 ms - 5110 ms (640ms - ~5s)
144 * 2 128 640 ms 5120 ms - 40950 ms (~5s - ~40s)
145 * 3 192 5120 ms (~5s) 40960 ms - 327670 ms (~40s - ~5m)
146 * 4 256 40960 ms (~40s) 327680 ms - 2621430 ms (~5m - ~43m)
147 * 5 320 327680 ms (~5m) 2621440 ms - 20971510 ms (~43m - ~5h)
148 * 6 384 2621440 ms (~43m) 20971520 ms - 167772150 ms (~5h - ~1d)
149 * 7 448 20971520 ms (~5h) 167772160 ms - 1342177270 ms (~1d - ~15d)
152 /* Clock divisor for the next level */
153 #define LVL_CLK_SHIFT 3
154 #define LVL_CLK_DIV (1UL << LVL_CLK_SHIFT)
155 #define LVL_CLK_MASK (LVL_CLK_DIV - 1)
156 #define LVL_SHIFT(n) ((n) * LVL_CLK_SHIFT)
157 #define LVL_GRAN(n) (1UL << LVL_SHIFT(n))
160 * The time start value for each level to select the bucket at enqueue
161 * time. We start from the last possible delta of the previous level
162 * so that we can later add an extra LVL_GRAN(n) to n (see calc_index()).
164 #define LVL_START(n) ((LVL_SIZE - 1) << (((n) - 1) * LVL_CLK_SHIFT))
166 /* Size of each clock level */
168 #define LVL_SIZE (1UL << LVL_BITS)
169 #define LVL_MASK (LVL_SIZE - 1)
170 #define LVL_OFFS(n) ((n) * LVL_SIZE)
179 /* The cutoff (max. capacity of the wheel) */
180 #define WHEEL_TIMEOUT_CUTOFF (LVL_START(LVL_DEPTH))
181 #define WHEEL_TIMEOUT_MAX (WHEEL_TIMEOUT_CUTOFF - LVL_GRAN(LVL_DEPTH - 1))
184 * The resulting wheel size. If NOHZ is configured we allocate two
185 * wheels so we have a separate storage for the deferrable timers.
187 #define WHEEL_SIZE (LVL_SIZE * LVL_DEPTH)
189 #ifdef CONFIG_NO_HZ_COMMON
201 struct timer_list *running_timer;
202 #ifdef CONFIG_PREEMPT_RT
203 spinlock_t expiry_lock;
204 atomic_t timer_waiters;
207 unsigned long next_expiry;
209 bool next_expiry_recalc;
212 DECLARE_BITMAP(pending_map, WHEEL_SIZE);
213 struct hlist_head vectors[WHEEL_SIZE];
214 } ____cacheline_aligned;
216 static DEFINE_PER_CPU(struct timer_base, timer_bases[NR_BASES]);
218 #ifdef CONFIG_NO_HZ_COMMON
220 static DEFINE_STATIC_KEY_FALSE(timers_nohz_active);
221 static DEFINE_MUTEX(timer_keys_mutex);
223 static void timer_update_keys(struct work_struct *work);
224 static DECLARE_WORK(timer_update_work, timer_update_keys);
227 static unsigned int sysctl_timer_migration = 1;
229 DEFINE_STATIC_KEY_FALSE(timers_migration_enabled);
231 static void timers_update_migration(void)
233 if (sysctl_timer_migration && tick_nohz_active)
234 static_branch_enable(&timers_migration_enabled);
236 static_branch_disable(&timers_migration_enabled);
240 static int timer_migration_handler(struct ctl_table *table, int write,
241 void *buffer, size_t *lenp, loff_t *ppos)
245 mutex_lock(&timer_keys_mutex);
246 ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
248 timers_update_migration();
249 mutex_unlock(&timer_keys_mutex);
253 static struct ctl_table timer_sysctl[] = {
255 .procname = "timer_migration",
256 .data = &sysctl_timer_migration,
257 .maxlen = sizeof(unsigned int),
259 .proc_handler = timer_migration_handler,
260 .extra1 = SYSCTL_ZERO,
261 .extra2 = SYSCTL_ONE,
266 static int __init timer_sysctl_init(void)
268 register_sysctl("kernel", timer_sysctl);
271 device_initcall(timer_sysctl_init);
272 #endif /* CONFIG_SYSCTL */
273 #else /* CONFIG_SMP */
274 static inline void timers_update_migration(void) { }
275 #endif /* !CONFIG_SMP */
277 static void timer_update_keys(struct work_struct *work)
279 mutex_lock(&timer_keys_mutex);
280 timers_update_migration();
281 static_branch_enable(&timers_nohz_active);
282 mutex_unlock(&timer_keys_mutex);
285 void timers_update_nohz(void)
287 schedule_work(&timer_update_work);
290 static inline bool is_timers_nohz_active(void)
292 return static_branch_unlikely(&timers_nohz_active);
295 static inline bool is_timers_nohz_active(void) { return false; }
296 #endif /* NO_HZ_COMMON */
298 static unsigned long round_jiffies_common(unsigned long j, int cpu,
302 unsigned long original = j;
305 * We don't want all cpus firing their timers at once hitting the
306 * same lock or cachelines, so we skew each extra cpu with an extra
307 * 3 jiffies. This 3 jiffies came originally from the mm/ code which
309 * The skew is done by adding 3*cpunr, then round, then subtract this
310 * extra offset again.
317 * If the target jiffie is just after a whole second (which can happen
318 * due to delays of the timer irq, long irq off times etc etc) then
319 * we should round down to the whole second, not up. Use 1/4th second
320 * as cutoff for this rounding as an extreme upper bound for this.
321 * But never round down if @force_up is set.
323 if (rem < HZ/4 && !force_up) /* round down */
328 /* now that we have rounded, subtract the extra skew again */
332 * Make sure j is still in the future. Otherwise return the
335 return time_is_after_jiffies(j) ? j : original;
339 * __round_jiffies - function to round jiffies to a full second
340 * @j: the time in (absolute) jiffies that should be rounded
341 * @cpu: the processor number on which the timeout will happen
343 * __round_jiffies() rounds an absolute time in the future (in jiffies)
344 * up or down to (approximately) full seconds. This is useful for timers
345 * for which the exact time they fire does not matter too much, as long as
346 * they fire approximately every X seconds.
348 * By rounding these timers to whole seconds, all such timers will fire
349 * at the same time, rather than at various times spread out. The goal
350 * of this is to have the CPU wake up less, which saves power.
352 * The exact rounding is skewed for each processor to avoid all
353 * processors firing at the exact same time, which could lead
354 * to lock contention or spurious cache line bouncing.
356 * The return value is the rounded version of the @j parameter.
358 unsigned long __round_jiffies(unsigned long j, int cpu)
360 return round_jiffies_common(j, cpu, false);
362 EXPORT_SYMBOL_GPL(__round_jiffies);
365 * __round_jiffies_relative - function to round jiffies to a full second
366 * @j: the time in (relative) jiffies that should be rounded
367 * @cpu: the processor number on which the timeout will happen
369 * __round_jiffies_relative() rounds a time delta in the future (in jiffies)
370 * up or down to (approximately) full seconds. This is useful for timers
371 * for which the exact time they fire does not matter too much, as long as
372 * they fire approximately every X seconds.
374 * By rounding these timers to whole seconds, all such timers will fire
375 * at the same time, rather than at various times spread out. The goal
376 * of this is to have the CPU wake up less, which saves power.
378 * The exact rounding is skewed for each processor to avoid all
379 * processors firing at the exact same time, which could lead
380 * to lock contention or spurious cache line bouncing.
382 * The return value is the rounded version of the @j parameter.
384 unsigned long __round_jiffies_relative(unsigned long j, int cpu)
386 unsigned long j0 = jiffies;
388 /* Use j0 because jiffies might change while we run */
389 return round_jiffies_common(j + j0, cpu, false) - j0;
391 EXPORT_SYMBOL_GPL(__round_jiffies_relative);
394 * round_jiffies - function to round jiffies to a full second
395 * @j: the time in (absolute) jiffies that should be rounded
397 * round_jiffies() rounds an absolute time in the future (in jiffies)
398 * up or down to (approximately) full seconds. This is useful for timers
399 * for which the exact time they fire does not matter too much, as long as
400 * they fire approximately every X seconds.
402 * By rounding these timers to whole seconds, all such timers will fire
403 * at the same time, rather than at various times spread out. The goal
404 * of this is to have the CPU wake up less, which saves power.
406 * The return value is the rounded version of the @j parameter.
408 unsigned long round_jiffies(unsigned long j)
410 return round_jiffies_common(j, raw_smp_processor_id(), false);
412 EXPORT_SYMBOL_GPL(round_jiffies);
415 * round_jiffies_relative - function to round jiffies to a full second
416 * @j: the time in (relative) jiffies that should be rounded
418 * round_jiffies_relative() rounds a time delta in the future (in jiffies)
419 * up or down to (approximately) full seconds. This is useful for timers
420 * for which the exact time they fire does not matter too much, as long as
421 * they fire approximately every X seconds.
423 * By rounding these timers to whole seconds, all such timers will fire
424 * at the same time, rather than at various times spread out. The goal
425 * of this is to have the CPU wake up less, which saves power.
427 * The return value is the rounded version of the @j parameter.
429 unsigned long round_jiffies_relative(unsigned long j)
431 return __round_jiffies_relative(j, raw_smp_processor_id());
433 EXPORT_SYMBOL_GPL(round_jiffies_relative);
436 * __round_jiffies_up - function to round jiffies up to a full second
437 * @j: the time in (absolute) jiffies that should be rounded
438 * @cpu: the processor number on which the timeout will happen
440 * This is the same as __round_jiffies() except that it will never
441 * round down. This is useful for timeouts for which the exact time
442 * of firing does not matter too much, as long as they don't fire too
445 unsigned long __round_jiffies_up(unsigned long j, int cpu)
447 return round_jiffies_common(j, cpu, true);
449 EXPORT_SYMBOL_GPL(__round_jiffies_up);
452 * __round_jiffies_up_relative - function to round jiffies up to a full second
453 * @j: the time in (relative) jiffies that should be rounded
454 * @cpu: the processor number on which the timeout will happen
456 * This is the same as __round_jiffies_relative() except that it will never
457 * round down. This is useful for timeouts for which the exact time
458 * of firing does not matter too much, as long as they don't fire too
461 unsigned long __round_jiffies_up_relative(unsigned long j, int cpu)
463 unsigned long j0 = jiffies;
465 /* Use j0 because jiffies might change while we run */
466 return round_jiffies_common(j + j0, cpu, true) - j0;
468 EXPORT_SYMBOL_GPL(__round_jiffies_up_relative);
471 * round_jiffies_up - function to round jiffies up to a full second
472 * @j: the time in (absolute) jiffies that should be rounded
474 * This is the same as round_jiffies() except that it will never
475 * round down. This is useful for timeouts for which the exact time
476 * of firing does not matter too much, as long as they don't fire too
479 unsigned long round_jiffies_up(unsigned long j)
481 return round_jiffies_common(j, raw_smp_processor_id(), true);
483 EXPORT_SYMBOL_GPL(round_jiffies_up);
486 * round_jiffies_up_relative - function to round jiffies up to a full second
487 * @j: the time in (relative) jiffies that should be rounded
489 * This is the same as round_jiffies_relative() except that it will never
490 * round down. This is useful for timeouts for which the exact time
491 * of firing does not matter too much, as long as they don't fire too
494 unsigned long round_jiffies_up_relative(unsigned long j)
496 return __round_jiffies_up_relative(j, raw_smp_processor_id());
498 EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
501 static inline unsigned int timer_get_idx(struct timer_list *timer)
503 return (timer->flags & TIMER_ARRAYMASK) >> TIMER_ARRAYSHIFT;
506 static inline void timer_set_idx(struct timer_list *timer, unsigned int idx)
508 timer->flags = (timer->flags & ~TIMER_ARRAYMASK) |
509 idx << TIMER_ARRAYSHIFT;
513 * Helper function to calculate the array index for a given expiry
516 static inline unsigned calc_index(unsigned long expires, unsigned lvl,
517 unsigned long *bucket_expiry)
521 * The timer wheel has to guarantee that a timer does not fire
522 * early. Early expiry can happen due to:
523 * - Timer is armed at the edge of a tick
524 * - Truncation of the expiry time in the outer wheel levels
526 * Round up with level granularity to prevent this.
528 expires = (expires >> LVL_SHIFT(lvl)) + 1;
529 *bucket_expiry = expires << LVL_SHIFT(lvl);
530 return LVL_OFFS(lvl) + (expires & LVL_MASK);
533 static int calc_wheel_index(unsigned long expires, unsigned long clk,
534 unsigned long *bucket_expiry)
536 unsigned long delta = expires - clk;
539 if (delta < LVL_START(1)) {
540 idx = calc_index(expires, 0, bucket_expiry);
541 } else if (delta < LVL_START(2)) {
542 idx = calc_index(expires, 1, bucket_expiry);
543 } else if (delta < LVL_START(3)) {
544 idx = calc_index(expires, 2, bucket_expiry);
545 } else if (delta < LVL_START(4)) {
546 idx = calc_index(expires, 3, bucket_expiry);
547 } else if (delta < LVL_START(5)) {
548 idx = calc_index(expires, 4, bucket_expiry);
549 } else if (delta < LVL_START(6)) {
550 idx = calc_index(expires, 5, bucket_expiry);
551 } else if (delta < LVL_START(7)) {
552 idx = calc_index(expires, 6, bucket_expiry);
553 } else if (LVL_DEPTH > 8 && delta < LVL_START(8)) {
554 idx = calc_index(expires, 7, bucket_expiry);
555 } else if ((long) delta < 0) {
556 idx = clk & LVL_MASK;
557 *bucket_expiry = clk;
560 * Force expire obscene large timeouts to expire at the
561 * capacity limit of the wheel.
563 if (delta >= WHEEL_TIMEOUT_CUTOFF)
564 expires = clk + WHEEL_TIMEOUT_MAX;
566 idx = calc_index(expires, LVL_DEPTH - 1, bucket_expiry);
572 trigger_dyntick_cpu(struct timer_base *base, struct timer_list *timer)
574 if (!is_timers_nohz_active())
578 * TODO: This wants some optimizing similar to the code below, but we
579 * will do that when we switch from push to pull for deferrable timers.
581 if (timer->flags & TIMER_DEFERRABLE) {
582 if (tick_nohz_full_cpu(base->cpu))
583 wake_up_nohz_cpu(base->cpu);
588 * We might have to IPI the remote CPU if the base is idle and the
589 * timer is not deferrable. If the other CPU is on the way to idle
590 * then it can't set base->is_idle as we hold the base lock:
593 wake_up_nohz_cpu(base->cpu);
597 * Enqueue the timer into the hash bucket, mark it pending in
598 * the bitmap, store the index in the timer flags then wake up
599 * the target CPU if needed.
601 static void enqueue_timer(struct timer_base *base, struct timer_list *timer,
602 unsigned int idx, unsigned long bucket_expiry)
605 hlist_add_head(&timer->entry, base->vectors + idx);
606 __set_bit(idx, base->pending_map);
607 timer_set_idx(timer, idx);
609 trace_timer_start(timer, timer->expires, timer->flags);
612 * Check whether this is the new first expiring timer. The
613 * effective expiry time of the timer is required here
614 * (bucket_expiry) instead of timer->expires.
616 if (time_before(bucket_expiry, base->next_expiry)) {
618 * Set the next expiry time and kick the CPU so it
619 * can reevaluate the wheel:
621 base->next_expiry = bucket_expiry;
622 base->timers_pending = true;
623 base->next_expiry_recalc = false;
624 trigger_dyntick_cpu(base, timer);
628 static void internal_add_timer(struct timer_base *base, struct timer_list *timer)
630 unsigned long bucket_expiry;
633 idx = calc_wheel_index(timer->expires, base->clk, &bucket_expiry);
634 enqueue_timer(base, timer, idx, bucket_expiry);
637 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
639 static const struct debug_obj_descr timer_debug_descr;
642 void (*function)(struct timer_list *t);
646 #define TIMER_HINT(fn, container, timr, hintfn) \
649 .offset = offsetof(container, hintfn) - \
650 offsetof(container, timr) \
653 static const struct timer_hint timer_hints[] = {
654 TIMER_HINT(delayed_work_timer_fn,
655 struct delayed_work, timer, work.func),
656 TIMER_HINT(kthread_delayed_work_timer_fn,
657 struct kthread_delayed_work, timer, work.func),
660 static void *timer_debug_hint(void *addr)
662 struct timer_list *timer = addr;
665 for (i = 0; i < ARRAY_SIZE(timer_hints); i++) {
666 if (timer_hints[i].function == timer->function) {
667 void (**fn)(void) = addr + timer_hints[i].offset;
673 return timer->function;
676 static bool timer_is_static_object(void *addr)
678 struct timer_list *timer = addr;
680 return (timer->entry.pprev == NULL &&
681 timer->entry.next == TIMER_ENTRY_STATIC);
685 * fixup_init is called when:
686 * - an active object is initialized
688 static bool timer_fixup_init(void *addr, enum debug_obj_state state)
690 struct timer_list *timer = addr;
693 case ODEBUG_STATE_ACTIVE:
694 del_timer_sync(timer);
695 debug_object_init(timer, &timer_debug_descr);
702 /* Stub timer callback for improperly used timers. */
703 static void stub_timer(struct timer_list *unused)
709 * fixup_activate is called when:
710 * - an active object is activated
711 * - an unknown non-static object is activated
713 static bool timer_fixup_activate(void *addr, enum debug_obj_state state)
715 struct timer_list *timer = addr;
718 case ODEBUG_STATE_NOTAVAILABLE:
719 timer_setup(timer, stub_timer, 0);
722 case ODEBUG_STATE_ACTIVE:
731 * fixup_free is called when:
732 * - an active object is freed
734 static bool timer_fixup_free(void *addr, enum debug_obj_state state)
736 struct timer_list *timer = addr;
739 case ODEBUG_STATE_ACTIVE:
740 del_timer_sync(timer);
741 debug_object_free(timer, &timer_debug_descr);
749 * fixup_assert_init is called when:
750 * - an untracked/uninit-ed object is found
752 static bool timer_fixup_assert_init(void *addr, enum debug_obj_state state)
754 struct timer_list *timer = addr;
757 case ODEBUG_STATE_NOTAVAILABLE:
758 timer_setup(timer, stub_timer, 0);
765 static const struct debug_obj_descr timer_debug_descr = {
766 .name = "timer_list",
767 .debug_hint = timer_debug_hint,
768 .is_static_object = timer_is_static_object,
769 .fixup_init = timer_fixup_init,
770 .fixup_activate = timer_fixup_activate,
771 .fixup_free = timer_fixup_free,
772 .fixup_assert_init = timer_fixup_assert_init,
775 static inline void debug_timer_init(struct timer_list *timer)
777 debug_object_init(timer, &timer_debug_descr);
780 static inline void debug_timer_activate(struct timer_list *timer)
782 debug_object_activate(timer, &timer_debug_descr);
785 static inline void debug_timer_deactivate(struct timer_list *timer)
787 debug_object_deactivate(timer, &timer_debug_descr);
790 static inline void debug_timer_assert_init(struct timer_list *timer)
792 debug_object_assert_init(timer, &timer_debug_descr);
795 static void do_init_timer(struct timer_list *timer,
796 void (*func)(struct timer_list *),
798 const char *name, struct lock_class_key *key);
800 void init_timer_on_stack_key(struct timer_list *timer,
801 void (*func)(struct timer_list *),
803 const char *name, struct lock_class_key *key)
805 debug_object_init_on_stack(timer, &timer_debug_descr);
806 do_init_timer(timer, func, flags, name, key);
808 EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
810 void destroy_timer_on_stack(struct timer_list *timer)
812 debug_object_free(timer, &timer_debug_descr);
814 EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
817 static inline void debug_timer_init(struct timer_list *timer) { }
818 static inline void debug_timer_activate(struct timer_list *timer) { }
819 static inline void debug_timer_deactivate(struct timer_list *timer) { }
820 static inline void debug_timer_assert_init(struct timer_list *timer) { }
823 static inline void debug_init(struct timer_list *timer)
825 debug_timer_init(timer);
826 trace_timer_init(timer);
829 static inline void debug_deactivate(struct timer_list *timer)
831 debug_timer_deactivate(timer);
832 trace_timer_cancel(timer);
835 static inline void debug_assert_init(struct timer_list *timer)
837 debug_timer_assert_init(timer);
840 static void do_init_timer(struct timer_list *timer,
841 void (*func)(struct timer_list *),
843 const char *name, struct lock_class_key *key)
845 timer->entry.pprev = NULL;
846 timer->function = func;
847 if (WARN_ON_ONCE(flags & ~TIMER_INIT_FLAGS))
848 flags &= TIMER_INIT_FLAGS;
849 timer->flags = flags | raw_smp_processor_id();
850 lockdep_init_map(&timer->lockdep_map, name, key, 0);
854 * init_timer_key - initialize a timer
855 * @timer: the timer to be initialized
856 * @func: timer callback function
857 * @flags: timer flags
858 * @name: name of the timer
859 * @key: lockdep class key of the fake lock used for tracking timer
860 * sync lock dependencies
862 * init_timer_key() must be done to a timer prior calling *any* of the
863 * other timer functions.
865 void init_timer_key(struct timer_list *timer,
866 void (*func)(struct timer_list *), unsigned int flags,
867 const char *name, struct lock_class_key *key)
870 do_init_timer(timer, func, flags, name, key);
872 EXPORT_SYMBOL(init_timer_key);
874 static inline void detach_timer(struct timer_list *timer, bool clear_pending)
876 struct hlist_node *entry = &timer->entry;
878 debug_deactivate(timer);
883 entry->next = LIST_POISON2;
886 static int detach_if_pending(struct timer_list *timer, struct timer_base *base,
889 unsigned idx = timer_get_idx(timer);
891 if (!timer_pending(timer))
894 if (hlist_is_singular_node(&timer->entry, base->vectors + idx)) {
895 __clear_bit(idx, base->pending_map);
896 base->next_expiry_recalc = true;
899 detach_timer(timer, clear_pending);
903 static inline struct timer_base *get_timer_cpu_base(u32 tflags, u32 cpu)
905 struct timer_base *base = per_cpu_ptr(&timer_bases[BASE_STD], cpu);
908 * If the timer is deferrable and NO_HZ_COMMON is set then we need
909 * to use the deferrable base.
911 if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE))
912 base = per_cpu_ptr(&timer_bases[BASE_DEF], cpu);
916 static inline struct timer_base *get_timer_this_cpu_base(u32 tflags)
918 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
921 * If the timer is deferrable and NO_HZ_COMMON is set then we need
922 * to use the deferrable base.
924 if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE))
925 base = this_cpu_ptr(&timer_bases[BASE_DEF]);
929 static inline struct timer_base *get_timer_base(u32 tflags)
931 return get_timer_cpu_base(tflags, tflags & TIMER_CPUMASK);
934 static inline struct timer_base *
935 get_target_base(struct timer_base *base, unsigned tflags)
937 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
938 if (static_branch_likely(&timers_migration_enabled) &&
939 !(tflags & TIMER_PINNED))
940 return get_timer_cpu_base(tflags, get_nohz_timer_target());
942 return get_timer_this_cpu_base(tflags);
945 static inline void forward_timer_base(struct timer_base *base)
947 unsigned long jnow = READ_ONCE(jiffies);
950 * No need to forward if we are close enough below jiffies.
951 * Also while executing timers, base->clk is 1 offset ahead
952 * of jiffies to avoid endless requeuing to current jiffies.
954 if ((long)(jnow - base->clk) < 1)
958 * If the next expiry value is > jiffies, then we fast forward to
959 * jiffies otherwise we forward to the next expiry value.
961 if (time_after(base->next_expiry, jnow)) {
964 if (WARN_ON_ONCE(time_before(base->next_expiry, base->clk)))
966 base->clk = base->next_expiry;
972 * We are using hashed locking: Holding per_cpu(timer_bases[x]).lock means
973 * that all timers which are tied to this base are locked, and the base itself
976 * So __run_timers/migrate_timers can safely modify all timers which could
977 * be found in the base->vectors array.
979 * When a timer is migrating then the TIMER_MIGRATING flag is set and we need
980 * to wait until the migration is done.
982 static struct timer_base *lock_timer_base(struct timer_list *timer,
983 unsigned long *flags)
984 __acquires(timer->base->lock)
987 struct timer_base *base;
991 * We need to use READ_ONCE() here, otherwise the compiler
992 * might re-read @tf between the check for TIMER_MIGRATING
995 tf = READ_ONCE(timer->flags);
997 if (!(tf & TIMER_MIGRATING)) {
998 base = get_timer_base(tf);
999 raw_spin_lock_irqsave(&base->lock, *flags);
1000 if (timer->flags == tf)
1002 raw_spin_unlock_irqrestore(&base->lock, *flags);
1008 #define MOD_TIMER_PENDING_ONLY 0x01
1009 #define MOD_TIMER_REDUCE 0x02
1010 #define MOD_TIMER_NOTPENDING 0x04
1013 __mod_timer(struct timer_list *timer, unsigned long expires, unsigned int options)
1015 unsigned long clk = 0, flags, bucket_expiry;
1016 struct timer_base *base, *new_base;
1017 unsigned int idx = UINT_MAX;
1020 BUG_ON(!timer->function);
1023 * This is a common optimization triggered by the networking code - if
1024 * the timer is re-modified to have the same timeout or ends up in the
1025 * same array bucket then just return:
1027 if (!(options & MOD_TIMER_NOTPENDING) && timer_pending(timer)) {
1029 * The downside of this optimization is that it can result in
1030 * larger granularity than you would get from adding a new
1031 * timer with this expiry.
1033 long diff = timer->expires - expires;
1037 if (options & MOD_TIMER_REDUCE && diff <= 0)
1041 * We lock timer base and calculate the bucket index right
1042 * here. If the timer ends up in the same bucket, then we
1043 * just update the expiry time and avoid the whole
1044 * dequeue/enqueue dance.
1046 base = lock_timer_base(timer, &flags);
1047 forward_timer_base(base);
1049 if (timer_pending(timer) && (options & MOD_TIMER_REDUCE) &&
1050 time_before_eq(timer->expires, expires)) {
1056 idx = calc_wheel_index(expires, clk, &bucket_expiry);
1059 * Retrieve and compare the array index of the pending
1060 * timer. If it matches set the expiry to the new value so a
1061 * subsequent call will exit in the expires check above.
1063 if (idx == timer_get_idx(timer)) {
1064 if (!(options & MOD_TIMER_REDUCE))
1065 timer->expires = expires;
1066 else if (time_after(timer->expires, expires))
1067 timer->expires = expires;
1072 base = lock_timer_base(timer, &flags);
1073 forward_timer_base(base);
1076 ret = detach_if_pending(timer, base, false);
1077 if (!ret && (options & MOD_TIMER_PENDING_ONLY))
1080 new_base = get_target_base(base, timer->flags);
1082 if (base != new_base) {
1084 * We are trying to schedule the timer on the new base.
1085 * However we can't change timer's base while it is running,
1086 * otherwise del_timer_sync() can't detect that the timer's
1087 * handler yet has not finished. This also guarantees that the
1088 * timer is serialized wrt itself.
1090 if (likely(base->running_timer != timer)) {
1091 /* See the comment in lock_timer_base() */
1092 timer->flags |= TIMER_MIGRATING;
1094 raw_spin_unlock(&base->lock);
1096 raw_spin_lock(&base->lock);
1097 WRITE_ONCE(timer->flags,
1098 (timer->flags & ~TIMER_BASEMASK) | base->cpu);
1099 forward_timer_base(base);
1103 debug_timer_activate(timer);
1105 timer->expires = expires;
1107 * If 'idx' was calculated above and the base time did not advance
1108 * between calculating 'idx' and possibly switching the base, only
1109 * enqueue_timer() is required. Otherwise we need to (re)calculate
1110 * the wheel index via internal_add_timer().
1112 if (idx != UINT_MAX && clk == base->clk)
1113 enqueue_timer(base, timer, idx, bucket_expiry);
1115 internal_add_timer(base, timer);
1118 raw_spin_unlock_irqrestore(&base->lock, flags);
1124 * mod_timer_pending - modify a pending timer's timeout
1125 * @timer: the pending timer to be modified
1126 * @expires: new timeout in jiffies
1128 * mod_timer_pending() is the same for pending timers as mod_timer(),
1129 * but will not re-activate and modify already deleted timers.
1131 * It is useful for unserialized use of timers.
1133 int mod_timer_pending(struct timer_list *timer, unsigned long expires)
1135 return __mod_timer(timer, expires, MOD_TIMER_PENDING_ONLY);
1137 EXPORT_SYMBOL(mod_timer_pending);
1140 * mod_timer - modify a timer's timeout
1141 * @timer: the timer to be modified
1142 * @expires: new timeout in jiffies
1144 * mod_timer() is a more efficient way to update the expire field of an
1145 * active timer (if the timer is inactive it will be activated)
1147 * mod_timer(timer, expires) is equivalent to:
1149 * del_timer(timer); timer->expires = expires; add_timer(timer);
1151 * Note that if there are multiple unserialized concurrent users of the
1152 * same timer, then mod_timer() is the only safe way to modify the timeout,
1153 * since add_timer() cannot modify an already running timer.
1155 * The function returns whether it has modified a pending timer or not.
1156 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
1157 * active timer returns 1.)
1159 int mod_timer(struct timer_list *timer, unsigned long expires)
1161 return __mod_timer(timer, expires, 0);
1163 EXPORT_SYMBOL(mod_timer);
1166 * timer_reduce - Modify a timer's timeout if it would reduce the timeout
1167 * @timer: The timer to be modified
1168 * @expires: New timeout in jiffies
1170 * timer_reduce() is very similar to mod_timer(), except that it will only
1171 * modify a running timer if that would reduce the expiration time (it will
1172 * start a timer that isn't running).
1174 int timer_reduce(struct timer_list *timer, unsigned long expires)
1176 return __mod_timer(timer, expires, MOD_TIMER_REDUCE);
1178 EXPORT_SYMBOL(timer_reduce);
1181 * add_timer - start a timer
1182 * @timer: the timer to be added
1184 * The kernel will do a ->function(@timer) callback from the
1185 * timer interrupt at the ->expires point in the future. The
1186 * current time is 'jiffies'.
1188 * The timer's ->expires, ->function fields must be set prior calling this
1191 * Timers with an ->expires field in the past will be executed in the next
1194 void add_timer(struct timer_list *timer)
1196 BUG_ON(timer_pending(timer));
1197 __mod_timer(timer, timer->expires, MOD_TIMER_NOTPENDING);
1199 EXPORT_SYMBOL(add_timer);
1202 * add_timer_on - start a timer on a particular CPU
1203 * @timer: the timer to be added
1204 * @cpu: the CPU to start it on
1206 * This is not very scalable on SMP. Double adds are not possible.
1208 void add_timer_on(struct timer_list *timer, int cpu)
1210 struct timer_base *new_base, *base;
1211 unsigned long flags;
1213 BUG_ON(timer_pending(timer) || !timer->function);
1215 new_base = get_timer_cpu_base(timer->flags, cpu);
1218 * If @timer was on a different CPU, it should be migrated with the
1219 * old base locked to prevent other operations proceeding with the
1220 * wrong base locked. See lock_timer_base().
1222 base = lock_timer_base(timer, &flags);
1223 if (base != new_base) {
1224 timer->flags |= TIMER_MIGRATING;
1226 raw_spin_unlock(&base->lock);
1228 raw_spin_lock(&base->lock);
1229 WRITE_ONCE(timer->flags,
1230 (timer->flags & ~TIMER_BASEMASK) | cpu);
1232 forward_timer_base(base);
1234 debug_timer_activate(timer);
1235 internal_add_timer(base, timer);
1236 raw_spin_unlock_irqrestore(&base->lock, flags);
1238 EXPORT_SYMBOL_GPL(add_timer_on);
1241 * del_timer - deactivate a timer.
1242 * @timer: the timer to be deactivated
1244 * del_timer() deactivates a timer - this works on both active and inactive
1247 * The function returns whether it has deactivated a pending timer or not.
1248 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
1249 * active timer returns 1.)
1251 int del_timer(struct timer_list *timer)
1253 struct timer_base *base;
1254 unsigned long flags;
1257 debug_assert_init(timer);
1259 if (timer_pending(timer)) {
1260 base = lock_timer_base(timer, &flags);
1261 ret = detach_if_pending(timer, base, true);
1262 raw_spin_unlock_irqrestore(&base->lock, flags);
1267 EXPORT_SYMBOL(del_timer);
1270 * try_to_del_timer_sync - Try to deactivate a timer
1271 * @timer: timer to delete
1273 * This function tries to deactivate a timer. Upon successful (ret >= 0)
1274 * exit the timer is not queued and the handler is not running on any CPU.
1276 int try_to_del_timer_sync(struct timer_list *timer)
1278 struct timer_base *base;
1279 unsigned long flags;
1282 debug_assert_init(timer);
1284 base = lock_timer_base(timer, &flags);
1286 if (base->running_timer != timer)
1287 ret = detach_if_pending(timer, base, true);
1289 raw_spin_unlock_irqrestore(&base->lock, flags);
1293 EXPORT_SYMBOL(try_to_del_timer_sync);
1295 #ifdef CONFIG_PREEMPT_RT
1296 static __init void timer_base_init_expiry_lock(struct timer_base *base)
1298 spin_lock_init(&base->expiry_lock);
1301 static inline void timer_base_lock_expiry(struct timer_base *base)
1303 spin_lock(&base->expiry_lock);
1306 static inline void timer_base_unlock_expiry(struct timer_base *base)
1308 spin_unlock(&base->expiry_lock);
1312 * The counterpart to del_timer_wait_running().
1314 * If there is a waiter for base->expiry_lock, then it was waiting for the
1315 * timer callback to finish. Drop expiry_lock and reacquire it. That allows
1316 * the waiter to acquire the lock and make progress.
1318 static void timer_sync_wait_running(struct timer_base *base)
1320 if (atomic_read(&base->timer_waiters)) {
1321 raw_spin_unlock_irq(&base->lock);
1322 spin_unlock(&base->expiry_lock);
1323 spin_lock(&base->expiry_lock);
1324 raw_spin_lock_irq(&base->lock);
1329 * This function is called on PREEMPT_RT kernels when the fast path
1330 * deletion of a timer failed because the timer callback function was
1333 * This prevents priority inversion, if the softirq thread on a remote CPU
1334 * got preempted, and it prevents a life lock when the task which tries to
1335 * delete a timer preempted the softirq thread running the timer callback
1338 static void del_timer_wait_running(struct timer_list *timer)
1342 tf = READ_ONCE(timer->flags);
1343 if (!(tf & (TIMER_MIGRATING | TIMER_IRQSAFE))) {
1344 struct timer_base *base = get_timer_base(tf);
1347 * Mark the base as contended and grab the expiry lock,
1348 * which is held by the softirq across the timer
1349 * callback. Drop the lock immediately so the softirq can
1350 * expire the next timer. In theory the timer could already
1351 * be running again, but that's more than unlikely and just
1352 * causes another wait loop.
1354 atomic_inc(&base->timer_waiters);
1355 spin_lock_bh(&base->expiry_lock);
1356 atomic_dec(&base->timer_waiters);
1357 spin_unlock_bh(&base->expiry_lock);
1361 static inline void timer_base_init_expiry_lock(struct timer_base *base) { }
1362 static inline void timer_base_lock_expiry(struct timer_base *base) { }
1363 static inline void timer_base_unlock_expiry(struct timer_base *base) { }
1364 static inline void timer_sync_wait_running(struct timer_base *base) { }
1365 static inline void del_timer_wait_running(struct timer_list *timer) { }
1368 #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT_RT)
1370 * del_timer_sync - deactivate a timer and wait for the handler to finish.
1371 * @timer: the timer to be deactivated
1373 * This function only differs from del_timer() on SMP: besides deactivating
1374 * the timer it also makes sure the handler has finished executing on other
1377 * Synchronization rules: Callers must prevent restarting of the timer,
1378 * otherwise this function is meaningless. It must not be called from
1379 * interrupt contexts unless the timer is an irqsafe one. The caller must
1380 * not hold locks which would prevent completion of the timer's
1381 * handler. The timer's handler must not call add_timer_on(). Upon exit the
1382 * timer is not queued and the handler is not running on any CPU.
1384 * Note: For !irqsafe timers, you must not hold locks that are held in
1385 * interrupt context while calling this function. Even if the lock has
1386 * nothing to do with the timer in question. Here's why::
1392 * base->running_timer = mytimer;
1393 * spin_lock_irq(somelock);
1395 * spin_lock(somelock);
1396 * del_timer_sync(mytimer);
1397 * while (base->running_timer == mytimer);
1399 * Now del_timer_sync() will never return and never release somelock.
1400 * The interrupt on the other CPU is waiting to grab somelock but
1401 * it has interrupted the softirq that CPU0 is waiting to finish.
1403 * The function returns whether it has deactivated a pending timer or not.
1405 int del_timer_sync(struct timer_list *timer)
1409 #ifdef CONFIG_LOCKDEP
1410 unsigned long flags;
1413 * If lockdep gives a backtrace here, please reference
1414 * the synchronization rules above.
1416 local_irq_save(flags);
1417 lock_map_acquire(&timer->lockdep_map);
1418 lock_map_release(&timer->lockdep_map);
1419 local_irq_restore(flags);
1422 * don't use it in hardirq context, because it
1423 * could lead to deadlock.
1425 WARN_ON(in_irq() && !(timer->flags & TIMER_IRQSAFE));
1428 * Must be able to sleep on PREEMPT_RT because of the slowpath in
1429 * del_timer_wait_running().
1431 if (IS_ENABLED(CONFIG_PREEMPT_RT) && !(timer->flags & TIMER_IRQSAFE))
1432 lockdep_assert_preemption_enabled();
1435 ret = try_to_del_timer_sync(timer);
1437 if (unlikely(ret < 0)) {
1438 del_timer_wait_running(timer);
1445 EXPORT_SYMBOL(del_timer_sync);
1448 static void call_timer_fn(struct timer_list *timer,
1449 void (*fn)(struct timer_list *),
1450 unsigned long baseclk)
1452 int count = preempt_count();
1454 #ifdef CONFIG_LOCKDEP
1456 * It is permissible to free the timer from inside the
1457 * function that is called from it, this we need to take into
1458 * account for lockdep too. To avoid bogus "held lock freed"
1459 * warnings as well as problems when looking into
1460 * timer->lockdep_map, make a copy and use that here.
1462 struct lockdep_map lockdep_map;
1464 lockdep_copy_map(&lockdep_map, &timer->lockdep_map);
1467 * Couple the lock chain with the lock chain at
1468 * del_timer_sync() by acquiring the lock_map around the fn()
1469 * call here and in del_timer_sync().
1471 lock_map_acquire(&lockdep_map);
1473 trace_timer_expire_entry(timer, baseclk);
1475 trace_timer_expire_exit(timer);
1477 lock_map_release(&lockdep_map);
1479 if (count != preempt_count()) {
1480 WARN_ONCE(1, "timer: %pS preempt leak: %08x -> %08x\n",
1481 fn, count, preempt_count());
1483 * Restore the preempt count. That gives us a decent
1484 * chance to survive and extract information. If the
1485 * callback kept a lock held, bad luck, but not worse
1486 * than the BUG() we had.
1488 preempt_count_set(count);
1492 static void expire_timers(struct timer_base *base, struct hlist_head *head)
1495 * This value is required only for tracing. base->clk was
1496 * incremented directly before expire_timers was called. But expiry
1497 * is related to the old base->clk value.
1499 unsigned long baseclk = base->clk - 1;
1501 while (!hlist_empty(head)) {
1502 struct timer_list *timer;
1503 void (*fn)(struct timer_list *);
1505 timer = hlist_entry(head->first, struct timer_list, entry);
1507 base->running_timer = timer;
1508 detach_timer(timer, true);
1510 fn = timer->function;
1512 if (timer->flags & TIMER_IRQSAFE) {
1513 raw_spin_unlock(&base->lock);
1514 call_timer_fn(timer, fn, baseclk);
1515 raw_spin_lock(&base->lock);
1516 base->running_timer = NULL;
1518 raw_spin_unlock_irq(&base->lock);
1519 call_timer_fn(timer, fn, baseclk);
1520 raw_spin_lock_irq(&base->lock);
1521 base->running_timer = NULL;
1522 timer_sync_wait_running(base);
1527 static int collect_expired_timers(struct timer_base *base,
1528 struct hlist_head *heads)
1530 unsigned long clk = base->clk = base->next_expiry;
1531 struct hlist_head *vec;
1535 for (i = 0; i < LVL_DEPTH; i++) {
1536 idx = (clk & LVL_MASK) + i * LVL_SIZE;
1538 if (__test_and_clear_bit(idx, base->pending_map)) {
1539 vec = base->vectors + idx;
1540 hlist_move_list(vec, heads++);
1543 /* Is it time to look at the next level? */
1544 if (clk & LVL_CLK_MASK)
1546 /* Shift clock for the next level granularity */
1547 clk >>= LVL_CLK_SHIFT;
1553 * Find the next pending bucket of a level. Search from level start (@offset)
1554 * + @clk upwards and if nothing there, search from start of the level
1555 * (@offset) up to @offset + clk.
1557 static int next_pending_bucket(struct timer_base *base, unsigned offset,
1560 unsigned pos, start = offset + clk;
1561 unsigned end = offset + LVL_SIZE;
1563 pos = find_next_bit(base->pending_map, end, start);
1567 pos = find_next_bit(base->pending_map, start, offset);
1568 return pos < start ? pos + LVL_SIZE - start : -1;
1572 * Search the first expiring timer in the various clock levels. Caller must
1575 static unsigned long __next_timer_interrupt(struct timer_base *base)
1577 unsigned long clk, next, adj;
1578 unsigned lvl, offset = 0;
1580 next = base->clk + NEXT_TIMER_MAX_DELTA;
1582 for (lvl = 0; lvl < LVL_DEPTH; lvl++, offset += LVL_SIZE) {
1583 int pos = next_pending_bucket(base, offset, clk & LVL_MASK);
1584 unsigned long lvl_clk = clk & LVL_CLK_MASK;
1587 unsigned long tmp = clk + (unsigned long) pos;
1589 tmp <<= LVL_SHIFT(lvl);
1590 if (time_before(tmp, next))
1594 * If the next expiration happens before we reach
1595 * the next level, no need to check further.
1597 if (pos <= ((LVL_CLK_DIV - lvl_clk) & LVL_CLK_MASK))
1601 * Clock for the next level. If the current level clock lower
1602 * bits are zero, we look at the next level as is. If not we
1603 * need to advance it by one because that's going to be the
1604 * next expiring bucket in that level. base->clk is the next
1605 * expiring jiffie. So in case of:
1607 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1610 * we have to look at all levels @index 0. With
1612 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1615 * LVL0 has the next expiring bucket @index 2. The upper
1616 * levels have the next expiring bucket @index 1.
1618 * In case that the propagation wraps the next level the same
1621 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1624 * So after looking at LVL0 we get:
1626 * LVL5 LVL4 LVL3 LVL2 LVL1
1629 * So no propagation from LVL1 to LVL2 because that happened
1630 * with the add already, but then we need to propagate further
1631 * from LVL2 to LVL3.
1633 * So the simple check whether the lower bits of the current
1634 * level are 0 or not is sufficient for all cases.
1636 adj = lvl_clk ? 1 : 0;
1637 clk >>= LVL_CLK_SHIFT;
1641 base->next_expiry_recalc = false;
1642 base->timers_pending = !(next == base->clk + NEXT_TIMER_MAX_DELTA);
1647 #ifdef CONFIG_NO_HZ_COMMON
1649 * Check, if the next hrtimer event is before the next timer wheel
1652 static u64 cmp_next_hrtimer_event(u64 basem, u64 expires)
1654 u64 nextevt = hrtimer_get_next_event();
1657 * If high resolution timers are enabled
1658 * hrtimer_get_next_event() returns KTIME_MAX.
1660 if (expires <= nextevt)
1664 * If the next timer is already expired, return the tick base
1665 * time so the tick is fired immediately.
1667 if (nextevt <= basem)
1671 * Round up to the next jiffie. High resolution timers are
1672 * off, so the hrtimers are expired in the tick and we need to
1673 * make sure that this tick really expires the timer to avoid
1674 * a ping pong of the nohz stop code.
1676 * Use DIV_ROUND_UP_ULL to prevent gcc calling __divdi3
1678 return DIV_ROUND_UP_ULL(nextevt, TICK_NSEC) * TICK_NSEC;
1682 * get_next_timer_interrupt - return the time (clock mono) of the next timer
1683 * @basej: base time jiffies
1684 * @basem: base time clock monotonic
1686 * Returns the tick aligned clock monotonic time of the next pending
1687 * timer or KTIME_MAX if no timer is pending.
1689 u64 get_next_timer_interrupt(unsigned long basej, u64 basem)
1691 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1692 u64 expires = KTIME_MAX;
1693 unsigned long nextevt;
1696 * Pretend that there is no timer pending if the cpu is offline.
1697 * Possible pending timers will be migrated later to an active cpu.
1699 if (cpu_is_offline(smp_processor_id()))
1702 raw_spin_lock(&base->lock);
1703 if (base->next_expiry_recalc)
1704 base->next_expiry = __next_timer_interrupt(base);
1705 nextevt = base->next_expiry;
1708 * We have a fresh next event. Check whether we can forward the
1709 * base. We can only do that when @basej is past base->clk
1710 * otherwise we might rewind base->clk.
1712 if (time_after(basej, base->clk)) {
1713 if (time_after(nextevt, basej))
1715 else if (time_after(nextevt, base->clk))
1716 base->clk = nextevt;
1719 if (time_before_eq(nextevt, basej)) {
1721 base->is_idle = false;
1723 if (base->timers_pending)
1724 expires = basem + (u64)(nextevt - basej) * TICK_NSEC;
1726 * If we expect to sleep more than a tick, mark the base idle.
1727 * Also the tick is stopped so any added timer must forward
1728 * the base clk itself to keep granularity small. This idle
1729 * logic is only maintained for the BASE_STD base, deferrable
1730 * timers may still see large granularity skew (by design).
1732 if ((expires - basem) > TICK_NSEC)
1733 base->is_idle = true;
1735 raw_spin_unlock(&base->lock);
1737 return cmp_next_hrtimer_event(basem, expires);
1741 * timer_clear_idle - Clear the idle state of the timer base
1743 * Called with interrupts disabled
1745 void timer_clear_idle(void)
1747 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1750 * We do this unlocked. The worst outcome is a remote enqueue sending
1751 * a pointless IPI, but taking the lock would just make the window for
1752 * sending the IPI a few instructions smaller for the cost of taking
1753 * the lock in the exit from idle path.
1755 base->is_idle = false;
1760 * __run_timers - run all expired timers (if any) on this CPU.
1761 * @base: the timer vector to be processed.
1763 static inline void __run_timers(struct timer_base *base)
1765 struct hlist_head heads[LVL_DEPTH];
1768 if (time_before(jiffies, base->next_expiry))
1771 timer_base_lock_expiry(base);
1772 raw_spin_lock_irq(&base->lock);
1774 while (time_after_eq(jiffies, base->clk) &&
1775 time_after_eq(jiffies, base->next_expiry)) {
1776 levels = collect_expired_timers(base, heads);
1778 * The two possible reasons for not finding any expired
1779 * timer at this clk are that all matching timers have been
1780 * dequeued or no timer has been queued since
1781 * base::next_expiry was set to base::clk +
1782 * NEXT_TIMER_MAX_DELTA.
1784 WARN_ON_ONCE(!levels && !base->next_expiry_recalc
1785 && base->timers_pending);
1787 base->next_expiry = __next_timer_interrupt(base);
1790 expire_timers(base, heads + levels);
1792 raw_spin_unlock_irq(&base->lock);
1793 timer_base_unlock_expiry(base);
1797 * This function runs timers and the timer-tq in bottom half context.
1799 static __latent_entropy void run_timer_softirq(struct softirq_action *h)
1801 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1804 if (IS_ENABLED(CONFIG_NO_HZ_COMMON))
1805 __run_timers(this_cpu_ptr(&timer_bases[BASE_DEF]));
1809 * Called by the local, per-CPU timer interrupt on SMP.
1811 static void run_local_timers(void)
1813 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1815 hrtimer_run_queues();
1816 /* Raise the softirq only if required. */
1817 if (time_before(jiffies, base->next_expiry)) {
1818 if (!IS_ENABLED(CONFIG_NO_HZ_COMMON))
1820 /* CPU is awake, so check the deferrable base. */
1822 if (time_before(jiffies, base->next_expiry))
1825 raise_softirq(TIMER_SOFTIRQ);
1829 * Called from the timer interrupt handler to charge one tick to the current
1830 * process. user_tick is 1 if the tick is user time, 0 for system.
1832 void update_process_times(int user_tick)
1834 struct task_struct *p = current;
1836 /* Note: this timer irq context must be accounted for as well. */
1837 account_process_tick(p, user_tick);
1839 rcu_sched_clock_irq(user_tick);
1840 #ifdef CONFIG_IRQ_WORK
1845 if (IS_ENABLED(CONFIG_POSIX_TIMERS))
1846 run_posix_cpu_timers();
1850 * Since schedule_timeout()'s timer is defined on the stack, it must store
1851 * the target task on the stack as well.
1853 struct process_timer {
1854 struct timer_list timer;
1855 struct task_struct *task;
1858 static void process_timeout(struct timer_list *t)
1860 struct process_timer *timeout = from_timer(timeout, t, timer);
1862 wake_up_process(timeout->task);
1866 * schedule_timeout - sleep until timeout
1867 * @timeout: timeout value in jiffies
1869 * Make the current task sleep until @timeout jiffies have elapsed.
1870 * The function behavior depends on the current task state
1871 * (see also set_current_state() description):
1873 * %TASK_RUNNING - the scheduler is called, but the task does not sleep
1874 * at all. That happens because sched_submit_work() does nothing for
1875 * tasks in %TASK_RUNNING state.
1877 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1878 * pass before the routine returns unless the current task is explicitly
1879 * woken up, (e.g. by wake_up_process()).
1881 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1882 * delivered to the current task or the current task is explicitly woken
1885 * The current task state is guaranteed to be %TASK_RUNNING when this
1888 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1889 * the CPU away without a bound on the timeout. In this case the return
1890 * value will be %MAX_SCHEDULE_TIMEOUT.
1892 * Returns 0 when the timer has expired otherwise the remaining time in
1893 * jiffies will be returned. In all cases the return value is guaranteed
1894 * to be non-negative.
1896 signed long __sched schedule_timeout(signed long timeout)
1898 struct process_timer timer;
1899 unsigned long expire;
1903 case MAX_SCHEDULE_TIMEOUT:
1905 * These two special cases are useful to be comfortable
1906 * in the caller. Nothing more. We could take
1907 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1908 * but I' d like to return a valid offset (>=0) to allow
1909 * the caller to do everything it want with the retval.
1915 * Another bit of PARANOID. Note that the retval will be
1916 * 0 since no piece of kernel is supposed to do a check
1917 * for a negative retval of schedule_timeout() (since it
1918 * should never happens anyway). You just have the printk()
1919 * that will tell you if something is gone wrong and where.
1922 printk(KERN_ERR "schedule_timeout: wrong timeout "
1923 "value %lx\n", timeout);
1925 __set_current_state(TASK_RUNNING);
1930 expire = timeout + jiffies;
1932 timer.task = current;
1933 timer_setup_on_stack(&timer.timer, process_timeout, 0);
1934 __mod_timer(&timer.timer, expire, MOD_TIMER_NOTPENDING);
1936 del_singleshot_timer_sync(&timer.timer);
1938 /* Remove the timer from the object tracker */
1939 destroy_timer_on_stack(&timer.timer);
1941 timeout = expire - jiffies;
1944 return timeout < 0 ? 0 : timeout;
1946 EXPORT_SYMBOL(schedule_timeout);
1949 * We can use __set_current_state() here because schedule_timeout() calls
1950 * schedule() unconditionally.
1952 signed long __sched schedule_timeout_interruptible(signed long timeout)
1954 __set_current_state(TASK_INTERRUPTIBLE);
1955 return schedule_timeout(timeout);
1957 EXPORT_SYMBOL(schedule_timeout_interruptible);
1959 signed long __sched schedule_timeout_killable(signed long timeout)
1961 __set_current_state(TASK_KILLABLE);
1962 return schedule_timeout(timeout);
1964 EXPORT_SYMBOL(schedule_timeout_killable);
1966 signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1968 __set_current_state(TASK_UNINTERRUPTIBLE);
1969 return schedule_timeout(timeout);
1971 EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1974 * Like schedule_timeout_uninterruptible(), except this task will not contribute
1977 signed long __sched schedule_timeout_idle(signed long timeout)
1979 __set_current_state(TASK_IDLE);
1980 return schedule_timeout(timeout);
1982 EXPORT_SYMBOL(schedule_timeout_idle);
1984 #ifdef CONFIG_HOTPLUG_CPU
1985 static void migrate_timer_list(struct timer_base *new_base, struct hlist_head *head)
1987 struct timer_list *timer;
1988 int cpu = new_base->cpu;
1990 while (!hlist_empty(head)) {
1991 timer = hlist_entry(head->first, struct timer_list, entry);
1992 detach_timer(timer, false);
1993 timer->flags = (timer->flags & ~TIMER_BASEMASK) | cpu;
1994 internal_add_timer(new_base, timer);
1998 int timers_prepare_cpu(unsigned int cpu)
2000 struct timer_base *base;
2003 for (b = 0; b < NR_BASES; b++) {
2004 base = per_cpu_ptr(&timer_bases[b], cpu);
2005 base->clk = jiffies;
2006 base->next_expiry = base->clk + NEXT_TIMER_MAX_DELTA;
2007 base->next_expiry_recalc = false;
2008 base->timers_pending = false;
2009 base->is_idle = false;
2014 int timers_dead_cpu(unsigned int cpu)
2016 struct timer_base *old_base;
2017 struct timer_base *new_base;
2020 BUG_ON(cpu_online(cpu));
2022 for (b = 0; b < NR_BASES; b++) {
2023 old_base = per_cpu_ptr(&timer_bases[b], cpu);
2024 new_base = get_cpu_ptr(&timer_bases[b]);
2026 * The caller is globally serialized and nobody else
2027 * takes two locks at once, deadlock is not possible.
2029 raw_spin_lock_irq(&new_base->lock);
2030 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
2033 * The current CPUs base clock might be stale. Update it
2034 * before moving the timers over.
2036 forward_timer_base(new_base);
2038 BUG_ON(old_base->running_timer);
2040 for (i = 0; i < WHEEL_SIZE; i++)
2041 migrate_timer_list(new_base, old_base->vectors + i);
2043 raw_spin_unlock(&old_base->lock);
2044 raw_spin_unlock_irq(&new_base->lock);
2045 put_cpu_ptr(&timer_bases);
2050 #endif /* CONFIG_HOTPLUG_CPU */
2052 static void __init init_timer_cpu(int cpu)
2054 struct timer_base *base;
2057 for (i = 0; i < NR_BASES; i++) {
2058 base = per_cpu_ptr(&timer_bases[i], cpu);
2060 raw_spin_lock_init(&base->lock);
2061 base->clk = jiffies;
2062 base->next_expiry = base->clk + NEXT_TIMER_MAX_DELTA;
2063 timer_base_init_expiry_lock(base);
2067 static void __init init_timer_cpus(void)
2071 for_each_possible_cpu(cpu)
2072 init_timer_cpu(cpu);
2075 void __init init_timers(void)
2078 posix_cputimers_init_work();
2079 open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
2083 * msleep - sleep safely even with waitqueue interruptions
2084 * @msecs: Time in milliseconds to sleep for
2086 void msleep(unsigned int msecs)
2088 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
2091 timeout = schedule_timeout_uninterruptible(timeout);
2094 EXPORT_SYMBOL(msleep);
2097 * msleep_interruptible - sleep waiting for signals
2098 * @msecs: Time in milliseconds to sleep for
2100 unsigned long msleep_interruptible(unsigned int msecs)
2102 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
2104 while (timeout && !signal_pending(current))
2105 timeout = schedule_timeout_interruptible(timeout);
2106 return jiffies_to_msecs(timeout);
2109 EXPORT_SYMBOL(msleep_interruptible);
2112 * usleep_range_state - Sleep for an approximate time in a given state
2113 * @min: Minimum time in usecs to sleep
2114 * @max: Maximum time in usecs to sleep
2115 * @state: State of the current task that will be while sleeping
2117 * In non-atomic context where the exact wakeup time is flexible, use
2118 * usleep_range_state() instead of udelay(). The sleep improves responsiveness
2119 * by avoiding the CPU-hogging busy-wait of udelay(), and the range reduces
2120 * power usage by allowing hrtimers to take advantage of an already-
2121 * scheduled interrupt instead of scheduling a new one just for this sleep.
2123 void __sched usleep_range_state(unsigned long min, unsigned long max,
2126 ktime_t exp = ktime_add_us(ktime_get(), min);
2127 u64 delta = (u64)(max - min) * NSEC_PER_USEC;
2130 __set_current_state(state);
2131 /* Do not return before the requested sleep time has elapsed */
2132 if (!schedule_hrtimeout_range(&exp, delta, HRTIMER_MODE_ABS))
2136 EXPORT_SYMBOL(usleep_range_state);