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"
56 #include "timer_migration.h"
58 #define CREATE_TRACE_POINTS
59 #include <trace/events/timer.h>
61 __visible u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
63 EXPORT_SYMBOL(jiffies_64);
66 * The timer wheel has LVL_DEPTH array levels. Each level provides an array of
67 * LVL_SIZE buckets. Each level is driven by its own clock and therefor each
68 * level has a different granularity.
70 * The level granularity is: LVL_CLK_DIV ^ lvl
71 * The level clock frequency is: HZ / (LVL_CLK_DIV ^ level)
73 * The array level of a newly armed timer depends on the relative expiry
74 * time. The farther the expiry time is away the higher the array level and
75 * therefor the granularity becomes.
77 * Contrary to the original timer wheel implementation, which aims for 'exact'
78 * expiry of the timers, this implementation removes the need for recascading
79 * the timers into the lower array levels. The previous 'classic' timer wheel
80 * implementation of the kernel already violated the 'exact' expiry by adding
81 * slack to the expiry time to provide batched expiration. The granularity
82 * levels provide implicit batching.
84 * This is an optimization of the original timer wheel implementation for the
85 * majority of the timer wheel use cases: timeouts. The vast majority of
86 * timeout timers (networking, disk I/O ...) are canceled before expiry. If
87 * the timeout expires it indicates that normal operation is disturbed, so it
88 * does not matter much whether the timeout comes with a slight delay.
90 * The only exception to this are networking timers with a small expiry
91 * time. They rely on the granularity. Those fit into the first wheel level,
92 * which has HZ granularity.
94 * We don't have cascading anymore. timers with a expiry time above the
95 * capacity of the last wheel level are force expired at the maximum timeout
96 * value of the last wheel level. From data sampling we know that the maximum
97 * value observed is 5 days (network connection tracking), so this should not
100 * The currently chosen array constants values are a good compromise between
101 * array size and granularity.
103 * This results in the following granularity and range levels:
106 * Level Offset Granularity Range
107 * 0 0 1 ms 0 ms - 63 ms
108 * 1 64 8 ms 64 ms - 511 ms
109 * 2 128 64 ms 512 ms - 4095 ms (512ms - ~4s)
110 * 3 192 512 ms 4096 ms - 32767 ms (~4s - ~32s)
111 * 4 256 4096 ms (~4s) 32768 ms - 262143 ms (~32s - ~4m)
112 * 5 320 32768 ms (~32s) 262144 ms - 2097151 ms (~4m - ~34m)
113 * 6 384 262144 ms (~4m) 2097152 ms - 16777215 ms (~34m - ~4h)
114 * 7 448 2097152 ms (~34m) 16777216 ms - 134217727 ms (~4h - ~1d)
115 * 8 512 16777216 ms (~4h) 134217728 ms - 1073741822 ms (~1d - ~12d)
118 * Level Offset Granularity Range
119 * 0 0 3 ms 0 ms - 210 ms
120 * 1 64 26 ms 213 ms - 1703 ms (213ms - ~1s)
121 * 2 128 213 ms 1706 ms - 13650 ms (~1s - ~13s)
122 * 3 192 1706 ms (~1s) 13653 ms - 109223 ms (~13s - ~1m)
123 * 4 256 13653 ms (~13s) 109226 ms - 873810 ms (~1m - ~14m)
124 * 5 320 109226 ms (~1m) 873813 ms - 6990503 ms (~14m - ~1h)
125 * 6 384 873813 ms (~14m) 6990506 ms - 55924050 ms (~1h - ~15h)
126 * 7 448 6990506 ms (~1h) 55924053 ms - 447392423 ms (~15h - ~5d)
127 * 8 512 55924053 ms (~15h) 447392426 ms - 3579139406 ms (~5d - ~41d)
130 * Level Offset Granularity Range
131 * 0 0 4 ms 0 ms - 255 ms
132 * 1 64 32 ms 256 ms - 2047 ms (256ms - ~2s)
133 * 2 128 256 ms 2048 ms - 16383 ms (~2s - ~16s)
134 * 3 192 2048 ms (~2s) 16384 ms - 131071 ms (~16s - ~2m)
135 * 4 256 16384 ms (~16s) 131072 ms - 1048575 ms (~2m - ~17m)
136 * 5 320 131072 ms (~2m) 1048576 ms - 8388607 ms (~17m - ~2h)
137 * 6 384 1048576 ms (~17m) 8388608 ms - 67108863 ms (~2h - ~18h)
138 * 7 448 8388608 ms (~2h) 67108864 ms - 536870911 ms (~18h - ~6d)
139 * 8 512 67108864 ms (~18h) 536870912 ms - 4294967288 ms (~6d - ~49d)
142 * Level Offset Granularity Range
143 * 0 0 10 ms 0 ms - 630 ms
144 * 1 64 80 ms 640 ms - 5110 ms (640ms - ~5s)
145 * 2 128 640 ms 5120 ms - 40950 ms (~5s - ~40s)
146 * 3 192 5120 ms (~5s) 40960 ms - 327670 ms (~40s - ~5m)
147 * 4 256 40960 ms (~40s) 327680 ms - 2621430 ms (~5m - ~43m)
148 * 5 320 327680 ms (~5m) 2621440 ms - 20971510 ms (~43m - ~5h)
149 * 6 384 2621440 ms (~43m) 20971520 ms - 167772150 ms (~5h - ~1d)
150 * 7 448 20971520 ms (~5h) 167772160 ms - 1342177270 ms (~1d - ~15d)
153 /* Clock divisor for the next level */
154 #define LVL_CLK_SHIFT 3
155 #define LVL_CLK_DIV (1UL << LVL_CLK_SHIFT)
156 #define LVL_CLK_MASK (LVL_CLK_DIV - 1)
157 #define LVL_SHIFT(n) ((n) * LVL_CLK_SHIFT)
158 #define LVL_GRAN(n) (1UL << LVL_SHIFT(n))
161 * The time start value for each level to select the bucket at enqueue
162 * time. We start from the last possible delta of the previous level
163 * so that we can later add an extra LVL_GRAN(n) to n (see calc_index()).
165 #define LVL_START(n) ((LVL_SIZE - 1) << (((n) - 1) * LVL_CLK_SHIFT))
167 /* Size of each clock level */
169 #define LVL_SIZE (1UL << LVL_BITS)
170 #define LVL_MASK (LVL_SIZE - 1)
171 #define LVL_OFFS(n) ((n) * LVL_SIZE)
180 /* The cutoff (max. capacity of the wheel) */
181 #define WHEEL_TIMEOUT_CUTOFF (LVL_START(LVL_DEPTH))
182 #define WHEEL_TIMEOUT_MAX (WHEEL_TIMEOUT_CUTOFF - LVL_GRAN(LVL_DEPTH - 1))
185 * The resulting wheel size. If NOHZ is configured we allocate two
186 * wheels so we have a separate storage for the deferrable timers.
188 #define WHEEL_SIZE (LVL_SIZE * LVL_DEPTH)
190 #ifdef CONFIG_NO_HZ_COMMON
192 * If multiple bases need to be locked, use the base ordering for lock
193 * nesting, i.e. lowest number first.
196 # define BASE_LOCAL 0
197 # define BASE_GLOBAL 1
201 # define BASE_LOCAL 0
202 # define BASE_GLOBAL 0
207 * struct timer_base - Per CPU timer base (number of base depends on config)
208 * @lock: Lock protecting the timer_base
209 * @running_timer: When expiring timers, the lock is dropped. To make
210 * sure not to race agains deleting/modifying a
211 * currently running timer, the pointer is set to the
212 * timer, which expires at the moment. If no timer is
213 * running, the pointer is NULL.
214 * @expiry_lock: PREEMPT_RT only: Lock is taken in softirq around
215 * timer expiry callback execution and when trying to
216 * delete a running timer and it wasn't successful in
217 * the first glance. It prevents priority inversion
218 * when callback was preempted on a remote CPU and a
219 * caller tries to delete the running timer. It also
220 * prevents a life lock, when the task which tries to
221 * delete a timer preempted the softirq thread which
222 * is running the timer callback function.
223 * @timer_waiters: PREEMPT_RT only: Tells, if there is a waiter
224 * waiting for the end of the timer callback function
226 * @clk: clock of the timer base; is updated before enqueue
227 * of a timer; during expiry, it is 1 offset ahead of
228 * jiffies to avoid endless requeuing to current
230 * @next_expiry: expiry value of the first timer; it is updated when
231 * finding the next timer and during enqueue; the
232 * value is not valid, when next_expiry_recalc is set
233 * @cpu: Number of CPU the timer base belongs to
234 * @next_expiry_recalc: States, whether a recalculation of next_expiry is
235 * required. Value is set true, when a timer was
237 * @is_idle: Is set, when timer_base is idle. It is triggered by NOHZ
238 * code. This state is only used in standard
239 * base. Deferrable timers, which are enqueued remotely
240 * never wake up an idle CPU. So no matter of supporting it
242 * @timers_pending: Is set, when a timer is pending in the base. It is only
243 * reliable when next_expiry_recalc is not set.
244 * @pending_map: bitmap of the timer wheel; each bit reflects a
245 * bucket of the wheel. When a bit is set, at least a
246 * single timer is enqueued in the related bucket.
247 * @vectors: Array of lists; Each array member reflects a bucket
248 * of the timer wheel. The list contains all timers
249 * which are enqueued into a specific bucket.
253 struct timer_list *running_timer;
254 #ifdef CONFIG_PREEMPT_RT
255 spinlock_t expiry_lock;
256 atomic_t timer_waiters;
259 unsigned long next_expiry;
261 bool next_expiry_recalc;
264 DECLARE_BITMAP(pending_map, WHEEL_SIZE);
265 struct hlist_head vectors[WHEEL_SIZE];
266 } ____cacheline_aligned;
268 static DEFINE_PER_CPU(struct timer_base, timer_bases[NR_BASES]);
270 #ifdef CONFIG_NO_HZ_COMMON
272 static DEFINE_STATIC_KEY_FALSE(timers_nohz_active);
273 static DEFINE_MUTEX(timer_keys_mutex);
275 static void timer_update_keys(struct work_struct *work);
276 static DECLARE_WORK(timer_update_work, timer_update_keys);
279 static unsigned int sysctl_timer_migration = 1;
281 DEFINE_STATIC_KEY_FALSE(timers_migration_enabled);
283 static void timers_update_migration(void)
285 if (sysctl_timer_migration && tick_nohz_active)
286 static_branch_enable(&timers_migration_enabled);
288 static_branch_disable(&timers_migration_enabled);
292 static int timer_migration_handler(struct ctl_table *table, int write,
293 void *buffer, size_t *lenp, loff_t *ppos)
297 mutex_lock(&timer_keys_mutex);
298 ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
300 timers_update_migration();
301 mutex_unlock(&timer_keys_mutex);
305 static struct ctl_table timer_sysctl[] = {
307 .procname = "timer_migration",
308 .data = &sysctl_timer_migration,
309 .maxlen = sizeof(unsigned int),
311 .proc_handler = timer_migration_handler,
312 .extra1 = SYSCTL_ZERO,
313 .extra2 = SYSCTL_ONE,
318 static int __init timer_sysctl_init(void)
320 register_sysctl("kernel", timer_sysctl);
323 device_initcall(timer_sysctl_init);
324 #endif /* CONFIG_SYSCTL */
325 #else /* CONFIG_SMP */
326 static inline void timers_update_migration(void) { }
327 #endif /* !CONFIG_SMP */
329 static void timer_update_keys(struct work_struct *work)
331 mutex_lock(&timer_keys_mutex);
332 timers_update_migration();
333 static_branch_enable(&timers_nohz_active);
334 mutex_unlock(&timer_keys_mutex);
337 void timers_update_nohz(void)
339 schedule_work(&timer_update_work);
342 static inline bool is_timers_nohz_active(void)
344 return static_branch_unlikely(&timers_nohz_active);
347 static inline bool is_timers_nohz_active(void) { return false; }
348 #endif /* NO_HZ_COMMON */
350 static unsigned long round_jiffies_common(unsigned long j, int cpu,
354 unsigned long original = j;
357 * We don't want all cpus firing their timers at once hitting the
358 * same lock or cachelines, so we skew each extra cpu with an extra
359 * 3 jiffies. This 3 jiffies came originally from the mm/ code which
361 * The skew is done by adding 3*cpunr, then round, then subtract this
362 * extra offset again.
369 * If the target jiffie is just after a whole second (which can happen
370 * due to delays of the timer irq, long irq off times etc etc) then
371 * we should round down to the whole second, not up. Use 1/4th second
372 * as cutoff for this rounding as an extreme upper bound for this.
373 * But never round down if @force_up is set.
375 if (rem < HZ/4 && !force_up) /* round down */
380 /* now that we have rounded, subtract the extra skew again */
384 * Make sure j is still in the future. Otherwise return the
387 return time_is_after_jiffies(j) ? j : original;
391 * __round_jiffies - function to round jiffies to a full second
392 * @j: the time in (absolute) jiffies that should be rounded
393 * @cpu: the processor number on which the timeout will happen
395 * __round_jiffies() rounds an absolute time in the future (in jiffies)
396 * up or down to (approximately) full seconds. This is useful for timers
397 * for which the exact time they fire does not matter too much, as long as
398 * they fire approximately every X seconds.
400 * By rounding these timers to whole seconds, all such timers will fire
401 * at the same time, rather than at various times spread out. The goal
402 * of this is to have the CPU wake up less, which saves power.
404 * The exact rounding is skewed for each processor to avoid all
405 * processors firing at the exact same time, which could lead
406 * to lock contention or spurious cache line bouncing.
408 * The return value is the rounded version of the @j parameter.
410 unsigned long __round_jiffies(unsigned long j, int cpu)
412 return round_jiffies_common(j, cpu, false);
414 EXPORT_SYMBOL_GPL(__round_jiffies);
417 * __round_jiffies_relative - function to round jiffies to a full second
418 * @j: the time in (relative) jiffies that should be rounded
419 * @cpu: the processor number on which the timeout will happen
421 * __round_jiffies_relative() rounds a time delta in the future (in jiffies)
422 * up or down to (approximately) full seconds. This is useful for timers
423 * for which the exact time they fire does not matter too much, as long as
424 * they fire approximately every X seconds.
426 * By rounding these timers to whole seconds, all such timers will fire
427 * at the same time, rather than at various times spread out. The goal
428 * of this is to have the CPU wake up less, which saves power.
430 * The exact rounding is skewed for each processor to avoid all
431 * processors firing at the exact same time, which could lead
432 * to lock contention or spurious cache line bouncing.
434 * The return value is the rounded version of the @j parameter.
436 unsigned long __round_jiffies_relative(unsigned long j, int cpu)
438 unsigned long j0 = jiffies;
440 /* Use j0 because jiffies might change while we run */
441 return round_jiffies_common(j + j0, cpu, false) - j0;
443 EXPORT_SYMBOL_GPL(__round_jiffies_relative);
446 * round_jiffies - function to round jiffies to a full second
447 * @j: the time in (absolute) jiffies that should be rounded
449 * round_jiffies() rounds an absolute time in the future (in jiffies)
450 * up or down to (approximately) full seconds. This is useful for timers
451 * for which the exact time they fire does not matter too much, as long as
452 * they fire approximately every X seconds.
454 * By rounding these timers to whole seconds, all such timers will fire
455 * at the same time, rather than at various times spread out. The goal
456 * of this is to have the CPU wake up less, which saves power.
458 * The return value is the rounded version of the @j parameter.
460 unsigned long round_jiffies(unsigned long j)
462 return round_jiffies_common(j, raw_smp_processor_id(), false);
464 EXPORT_SYMBOL_GPL(round_jiffies);
467 * round_jiffies_relative - function to round jiffies to a full second
468 * @j: the time in (relative) jiffies that should be rounded
470 * round_jiffies_relative() rounds a time delta in the future (in jiffies)
471 * up or down to (approximately) full seconds. This is useful for timers
472 * for which the exact time they fire does not matter too much, as long as
473 * they fire approximately every X seconds.
475 * By rounding these timers to whole seconds, all such timers will fire
476 * at the same time, rather than at various times spread out. The goal
477 * of this is to have the CPU wake up less, which saves power.
479 * The return value is the rounded version of the @j parameter.
481 unsigned long round_jiffies_relative(unsigned long j)
483 return __round_jiffies_relative(j, raw_smp_processor_id());
485 EXPORT_SYMBOL_GPL(round_jiffies_relative);
488 * __round_jiffies_up - function to round jiffies up to a full second
489 * @j: the time in (absolute) jiffies that should be rounded
490 * @cpu: the processor number on which the timeout will happen
492 * This is the same as __round_jiffies() except that it will never
493 * round down. This is useful for timeouts for which the exact time
494 * of firing does not matter too much, as long as they don't fire too
497 unsigned long __round_jiffies_up(unsigned long j, int cpu)
499 return round_jiffies_common(j, cpu, true);
501 EXPORT_SYMBOL_GPL(__round_jiffies_up);
504 * __round_jiffies_up_relative - function to round jiffies up to a full second
505 * @j: the time in (relative) jiffies that should be rounded
506 * @cpu: the processor number on which the timeout will happen
508 * This is the same as __round_jiffies_relative() except that it will never
509 * round down. This is useful for timeouts for which the exact time
510 * of firing does not matter too much, as long as they don't fire too
513 unsigned long __round_jiffies_up_relative(unsigned long j, int cpu)
515 unsigned long j0 = jiffies;
517 /* Use j0 because jiffies might change while we run */
518 return round_jiffies_common(j + j0, cpu, true) - j0;
520 EXPORT_SYMBOL_GPL(__round_jiffies_up_relative);
523 * round_jiffies_up - function to round jiffies up to a full second
524 * @j: the time in (absolute) jiffies that should be rounded
526 * This is the same as round_jiffies() except that it will never
527 * round down. This is useful for timeouts for which the exact time
528 * of firing does not matter too much, as long as they don't fire too
531 unsigned long round_jiffies_up(unsigned long j)
533 return round_jiffies_common(j, raw_smp_processor_id(), true);
535 EXPORT_SYMBOL_GPL(round_jiffies_up);
538 * round_jiffies_up_relative - function to round jiffies up to a full second
539 * @j: the time in (relative) jiffies that should be rounded
541 * This is the same as round_jiffies_relative() except that it will never
542 * round down. This is useful for timeouts for which the exact time
543 * of firing does not matter too much, as long as they don't fire too
546 unsigned long round_jiffies_up_relative(unsigned long j)
548 return __round_jiffies_up_relative(j, raw_smp_processor_id());
550 EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
553 static inline unsigned int timer_get_idx(struct timer_list *timer)
555 return (timer->flags & TIMER_ARRAYMASK) >> TIMER_ARRAYSHIFT;
558 static inline void timer_set_idx(struct timer_list *timer, unsigned int idx)
560 timer->flags = (timer->flags & ~TIMER_ARRAYMASK) |
561 idx << TIMER_ARRAYSHIFT;
565 * Helper function to calculate the array index for a given expiry
568 static inline unsigned calc_index(unsigned long expires, unsigned lvl,
569 unsigned long *bucket_expiry)
573 * The timer wheel has to guarantee that a timer does not fire
574 * early. Early expiry can happen due to:
575 * - Timer is armed at the edge of a tick
576 * - Truncation of the expiry time in the outer wheel levels
578 * Round up with level granularity to prevent this.
580 expires = (expires >> LVL_SHIFT(lvl)) + 1;
581 *bucket_expiry = expires << LVL_SHIFT(lvl);
582 return LVL_OFFS(lvl) + (expires & LVL_MASK);
585 static int calc_wheel_index(unsigned long expires, unsigned long clk,
586 unsigned long *bucket_expiry)
588 unsigned long delta = expires - clk;
591 if (delta < LVL_START(1)) {
592 idx = calc_index(expires, 0, bucket_expiry);
593 } else if (delta < LVL_START(2)) {
594 idx = calc_index(expires, 1, bucket_expiry);
595 } else if (delta < LVL_START(3)) {
596 idx = calc_index(expires, 2, bucket_expiry);
597 } else if (delta < LVL_START(4)) {
598 idx = calc_index(expires, 3, bucket_expiry);
599 } else if (delta < LVL_START(5)) {
600 idx = calc_index(expires, 4, bucket_expiry);
601 } else if (delta < LVL_START(6)) {
602 idx = calc_index(expires, 5, bucket_expiry);
603 } else if (delta < LVL_START(7)) {
604 idx = calc_index(expires, 6, bucket_expiry);
605 } else if (LVL_DEPTH > 8 && delta < LVL_START(8)) {
606 idx = calc_index(expires, 7, bucket_expiry);
607 } else if ((long) delta < 0) {
608 idx = clk & LVL_MASK;
609 *bucket_expiry = clk;
612 * Force expire obscene large timeouts to expire at the
613 * capacity limit of the wheel.
615 if (delta >= WHEEL_TIMEOUT_CUTOFF)
616 expires = clk + WHEEL_TIMEOUT_MAX;
618 idx = calc_index(expires, LVL_DEPTH - 1, bucket_expiry);
624 trigger_dyntick_cpu(struct timer_base *base, struct timer_list *timer)
627 * Deferrable timers do not prevent the CPU from entering dynticks and
628 * are not taken into account on the idle/nohz_full path. An IPI when a
629 * new deferrable timer is enqueued will wake up the remote CPU but
630 * nothing will be done with the deferrable timer base. Therefore skip
631 * the remote IPI for deferrable timers completely.
633 if (!is_timers_nohz_active() || timer->flags & TIMER_DEFERRABLE)
637 * We might have to IPI the remote CPU if the base is idle and the
638 * timer is pinned. If it is a non pinned timer, it is only queued
639 * on the remote CPU, when timer was running during queueing. Then
640 * everything is handled by remote CPU anyway. If the other CPU is
641 * on the way to idle then it can't set base->is_idle as we hold
645 WARN_ON_ONCE(!(timer->flags & TIMER_PINNED));
646 wake_up_nohz_cpu(base->cpu);
651 * Enqueue the timer into the hash bucket, mark it pending in
652 * the bitmap, store the index in the timer flags then wake up
653 * the target CPU if needed.
655 static void enqueue_timer(struct timer_base *base, struct timer_list *timer,
656 unsigned int idx, unsigned long bucket_expiry)
659 hlist_add_head(&timer->entry, base->vectors + idx);
660 __set_bit(idx, base->pending_map);
661 timer_set_idx(timer, idx);
663 trace_timer_start(timer, bucket_expiry);
666 * Check whether this is the new first expiring timer. The
667 * effective expiry time of the timer is required here
668 * (bucket_expiry) instead of timer->expires.
670 if (time_before(bucket_expiry, base->next_expiry)) {
672 * Set the next expiry time and kick the CPU so it
673 * can reevaluate the wheel:
675 base->next_expiry = bucket_expiry;
676 base->timers_pending = true;
677 base->next_expiry_recalc = false;
678 trigger_dyntick_cpu(base, timer);
682 static void internal_add_timer(struct timer_base *base, struct timer_list *timer)
684 unsigned long bucket_expiry;
687 idx = calc_wheel_index(timer->expires, base->clk, &bucket_expiry);
688 enqueue_timer(base, timer, idx, bucket_expiry);
691 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
693 static const struct debug_obj_descr timer_debug_descr;
696 void (*function)(struct timer_list *t);
700 #define TIMER_HINT(fn, container, timr, hintfn) \
703 .offset = offsetof(container, hintfn) - \
704 offsetof(container, timr) \
707 static const struct timer_hint timer_hints[] = {
708 TIMER_HINT(delayed_work_timer_fn,
709 struct delayed_work, timer, work.func),
710 TIMER_HINT(kthread_delayed_work_timer_fn,
711 struct kthread_delayed_work, timer, work.func),
714 static void *timer_debug_hint(void *addr)
716 struct timer_list *timer = addr;
719 for (i = 0; i < ARRAY_SIZE(timer_hints); i++) {
720 if (timer_hints[i].function == timer->function) {
721 void (**fn)(void) = addr + timer_hints[i].offset;
727 return timer->function;
730 static bool timer_is_static_object(void *addr)
732 struct timer_list *timer = addr;
734 return (timer->entry.pprev == NULL &&
735 timer->entry.next == TIMER_ENTRY_STATIC);
739 * fixup_init is called when:
740 * - an active object is initialized
742 static bool timer_fixup_init(void *addr, enum debug_obj_state state)
744 struct timer_list *timer = addr;
747 case ODEBUG_STATE_ACTIVE:
748 del_timer_sync(timer);
749 debug_object_init(timer, &timer_debug_descr);
756 /* Stub timer callback for improperly used timers. */
757 static void stub_timer(struct timer_list *unused)
763 * fixup_activate is called when:
764 * - an active object is activated
765 * - an unknown non-static object is activated
767 static bool timer_fixup_activate(void *addr, enum debug_obj_state state)
769 struct timer_list *timer = addr;
772 case ODEBUG_STATE_NOTAVAILABLE:
773 timer_setup(timer, stub_timer, 0);
776 case ODEBUG_STATE_ACTIVE:
785 * fixup_free is called when:
786 * - an active object is freed
788 static bool timer_fixup_free(void *addr, enum debug_obj_state state)
790 struct timer_list *timer = addr;
793 case ODEBUG_STATE_ACTIVE:
794 del_timer_sync(timer);
795 debug_object_free(timer, &timer_debug_descr);
803 * fixup_assert_init is called when:
804 * - an untracked/uninit-ed object is found
806 static bool timer_fixup_assert_init(void *addr, enum debug_obj_state state)
808 struct timer_list *timer = addr;
811 case ODEBUG_STATE_NOTAVAILABLE:
812 timer_setup(timer, stub_timer, 0);
819 static const struct debug_obj_descr timer_debug_descr = {
820 .name = "timer_list",
821 .debug_hint = timer_debug_hint,
822 .is_static_object = timer_is_static_object,
823 .fixup_init = timer_fixup_init,
824 .fixup_activate = timer_fixup_activate,
825 .fixup_free = timer_fixup_free,
826 .fixup_assert_init = timer_fixup_assert_init,
829 static inline void debug_timer_init(struct timer_list *timer)
831 debug_object_init(timer, &timer_debug_descr);
834 static inline void debug_timer_activate(struct timer_list *timer)
836 debug_object_activate(timer, &timer_debug_descr);
839 static inline void debug_timer_deactivate(struct timer_list *timer)
841 debug_object_deactivate(timer, &timer_debug_descr);
844 static inline void debug_timer_assert_init(struct timer_list *timer)
846 debug_object_assert_init(timer, &timer_debug_descr);
849 static void do_init_timer(struct timer_list *timer,
850 void (*func)(struct timer_list *),
852 const char *name, struct lock_class_key *key);
854 void init_timer_on_stack_key(struct timer_list *timer,
855 void (*func)(struct timer_list *),
857 const char *name, struct lock_class_key *key)
859 debug_object_init_on_stack(timer, &timer_debug_descr);
860 do_init_timer(timer, func, flags, name, key);
862 EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
864 void destroy_timer_on_stack(struct timer_list *timer)
866 debug_object_free(timer, &timer_debug_descr);
868 EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
871 static inline void debug_timer_init(struct timer_list *timer) { }
872 static inline void debug_timer_activate(struct timer_list *timer) { }
873 static inline void debug_timer_deactivate(struct timer_list *timer) { }
874 static inline void debug_timer_assert_init(struct timer_list *timer) { }
877 static inline void debug_init(struct timer_list *timer)
879 debug_timer_init(timer);
880 trace_timer_init(timer);
883 static inline void debug_deactivate(struct timer_list *timer)
885 debug_timer_deactivate(timer);
886 trace_timer_cancel(timer);
889 static inline void debug_assert_init(struct timer_list *timer)
891 debug_timer_assert_init(timer);
894 static void do_init_timer(struct timer_list *timer,
895 void (*func)(struct timer_list *),
897 const char *name, struct lock_class_key *key)
899 timer->entry.pprev = NULL;
900 timer->function = func;
901 if (WARN_ON_ONCE(flags & ~TIMER_INIT_FLAGS))
902 flags &= TIMER_INIT_FLAGS;
903 timer->flags = flags | raw_smp_processor_id();
904 lockdep_init_map(&timer->lockdep_map, name, key, 0);
908 * init_timer_key - initialize a timer
909 * @timer: the timer to be initialized
910 * @func: timer callback function
911 * @flags: timer flags
912 * @name: name of the timer
913 * @key: lockdep class key of the fake lock used for tracking timer
914 * sync lock dependencies
916 * init_timer_key() must be done to a timer prior calling *any* of the
917 * other timer functions.
919 void init_timer_key(struct timer_list *timer,
920 void (*func)(struct timer_list *), unsigned int flags,
921 const char *name, struct lock_class_key *key)
924 do_init_timer(timer, func, flags, name, key);
926 EXPORT_SYMBOL(init_timer_key);
928 static inline void detach_timer(struct timer_list *timer, bool clear_pending)
930 struct hlist_node *entry = &timer->entry;
932 debug_deactivate(timer);
937 entry->next = LIST_POISON2;
940 static int detach_if_pending(struct timer_list *timer, struct timer_base *base,
943 unsigned idx = timer_get_idx(timer);
945 if (!timer_pending(timer))
948 if (hlist_is_singular_node(&timer->entry, base->vectors + idx)) {
949 __clear_bit(idx, base->pending_map);
950 base->next_expiry_recalc = true;
953 detach_timer(timer, clear_pending);
957 static inline struct timer_base *get_timer_cpu_base(u32 tflags, u32 cpu)
959 int index = tflags & TIMER_PINNED ? BASE_LOCAL : BASE_GLOBAL;
960 struct timer_base *base;
962 base = per_cpu_ptr(&timer_bases[index], cpu);
965 * If the timer is deferrable and NO_HZ_COMMON is set then we need
966 * to use the deferrable base.
968 if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE))
969 base = per_cpu_ptr(&timer_bases[BASE_DEF], cpu);
973 static inline struct timer_base *get_timer_this_cpu_base(u32 tflags)
975 int index = tflags & TIMER_PINNED ? BASE_LOCAL : BASE_GLOBAL;
976 struct timer_base *base;
978 base = this_cpu_ptr(&timer_bases[index]);
981 * If the timer is deferrable and NO_HZ_COMMON is set then we need
982 * to use the deferrable base.
984 if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE))
985 base = this_cpu_ptr(&timer_bases[BASE_DEF]);
989 static inline struct timer_base *get_timer_base(u32 tflags)
991 return get_timer_cpu_base(tflags, tflags & TIMER_CPUMASK);
994 static inline void __forward_timer_base(struct timer_base *base,
998 * Check whether we can forward the base. We can only do that when
999 * @basej is past base->clk otherwise we might rewind base->clk.
1001 if (time_before_eq(basej, base->clk))
1005 * If the next expiry value is > jiffies, then we fast forward to
1006 * jiffies otherwise we forward to the next expiry value.
1008 if (time_after(base->next_expiry, basej)) {
1011 if (WARN_ON_ONCE(time_before(base->next_expiry, base->clk)))
1013 base->clk = base->next_expiry;
1018 static inline void forward_timer_base(struct timer_base *base)
1020 __forward_timer_base(base, READ_ONCE(jiffies));
1024 * We are using hashed locking: Holding per_cpu(timer_bases[x]).lock means
1025 * that all timers which are tied to this base are locked, and the base itself
1028 * So __run_timers/migrate_timers can safely modify all timers which could
1029 * be found in the base->vectors array.
1031 * When a timer is migrating then the TIMER_MIGRATING flag is set and we need
1032 * to wait until the migration is done.
1034 static struct timer_base *lock_timer_base(struct timer_list *timer,
1035 unsigned long *flags)
1036 __acquires(timer->base->lock)
1039 struct timer_base *base;
1043 * We need to use READ_ONCE() here, otherwise the compiler
1044 * might re-read @tf between the check for TIMER_MIGRATING
1047 tf = READ_ONCE(timer->flags);
1049 if (!(tf & TIMER_MIGRATING)) {
1050 base = get_timer_base(tf);
1051 raw_spin_lock_irqsave(&base->lock, *flags);
1052 if (timer->flags == tf)
1054 raw_spin_unlock_irqrestore(&base->lock, *flags);
1060 #define MOD_TIMER_PENDING_ONLY 0x01
1061 #define MOD_TIMER_REDUCE 0x02
1062 #define MOD_TIMER_NOTPENDING 0x04
1065 __mod_timer(struct timer_list *timer, unsigned long expires, unsigned int options)
1067 unsigned long clk = 0, flags, bucket_expiry;
1068 struct timer_base *base, *new_base;
1069 unsigned int idx = UINT_MAX;
1072 debug_assert_init(timer);
1075 * This is a common optimization triggered by the networking code - if
1076 * the timer is re-modified to have the same timeout or ends up in the
1077 * same array bucket then just return:
1079 if (!(options & MOD_TIMER_NOTPENDING) && timer_pending(timer)) {
1081 * The downside of this optimization is that it can result in
1082 * larger granularity than you would get from adding a new
1083 * timer with this expiry.
1085 long diff = timer->expires - expires;
1089 if (options & MOD_TIMER_REDUCE && diff <= 0)
1093 * We lock timer base and calculate the bucket index right
1094 * here. If the timer ends up in the same bucket, then we
1095 * just update the expiry time and avoid the whole
1096 * dequeue/enqueue dance.
1098 base = lock_timer_base(timer, &flags);
1100 * Has @timer been shutdown? This needs to be evaluated
1101 * while holding base lock to prevent a race against the
1104 if (!timer->function)
1107 forward_timer_base(base);
1109 if (timer_pending(timer) && (options & MOD_TIMER_REDUCE) &&
1110 time_before_eq(timer->expires, expires)) {
1116 idx = calc_wheel_index(expires, clk, &bucket_expiry);
1119 * Retrieve and compare the array index of the pending
1120 * timer. If it matches set the expiry to the new value so a
1121 * subsequent call will exit in the expires check above.
1123 if (idx == timer_get_idx(timer)) {
1124 if (!(options & MOD_TIMER_REDUCE))
1125 timer->expires = expires;
1126 else if (time_after(timer->expires, expires))
1127 timer->expires = expires;
1132 base = lock_timer_base(timer, &flags);
1134 * Has @timer been shutdown? This needs to be evaluated
1135 * while holding base lock to prevent a race against the
1138 if (!timer->function)
1141 forward_timer_base(base);
1144 ret = detach_if_pending(timer, base, false);
1145 if (!ret && (options & MOD_TIMER_PENDING_ONLY))
1148 new_base = get_timer_this_cpu_base(timer->flags);
1150 if (base != new_base) {
1152 * We are trying to schedule the timer on the new base.
1153 * However we can't change timer's base while it is running,
1154 * otherwise timer_delete_sync() can't detect that the timer's
1155 * handler yet has not finished. This also guarantees that the
1156 * timer is serialized wrt itself.
1158 if (likely(base->running_timer != timer)) {
1159 /* See the comment in lock_timer_base() */
1160 timer->flags |= TIMER_MIGRATING;
1162 raw_spin_unlock(&base->lock);
1164 raw_spin_lock(&base->lock);
1165 WRITE_ONCE(timer->flags,
1166 (timer->flags & ~TIMER_BASEMASK) | base->cpu);
1167 forward_timer_base(base);
1171 debug_timer_activate(timer);
1173 timer->expires = expires;
1175 * If 'idx' was calculated above and the base time did not advance
1176 * between calculating 'idx' and possibly switching the base, only
1177 * enqueue_timer() is required. Otherwise we need to (re)calculate
1178 * the wheel index via internal_add_timer().
1180 if (idx != UINT_MAX && clk == base->clk)
1181 enqueue_timer(base, timer, idx, bucket_expiry);
1183 internal_add_timer(base, timer);
1186 raw_spin_unlock_irqrestore(&base->lock, flags);
1192 * mod_timer_pending - Modify a pending timer's timeout
1193 * @timer: The pending timer to be modified
1194 * @expires: New absolute timeout in jiffies
1196 * mod_timer_pending() is the same for pending timers as mod_timer(), but
1197 * will not activate inactive timers.
1199 * If @timer->function == NULL then the start operation is silently
1203 * * %0 - The timer was inactive and not modified or was in
1204 * shutdown state and the operation was discarded
1205 * * %1 - The timer was active and requeued to expire at @expires
1207 int mod_timer_pending(struct timer_list *timer, unsigned long expires)
1209 return __mod_timer(timer, expires, MOD_TIMER_PENDING_ONLY);
1211 EXPORT_SYMBOL(mod_timer_pending);
1214 * mod_timer - Modify a timer's timeout
1215 * @timer: The timer to be modified
1216 * @expires: New absolute timeout in jiffies
1218 * mod_timer(timer, expires) is equivalent to:
1220 * del_timer(timer); timer->expires = expires; add_timer(timer);
1222 * mod_timer() is more efficient than the above open coded sequence. In
1223 * case that the timer is inactive, the del_timer() part is a NOP. The
1224 * timer is in any case activated with the new expiry time @expires.
1226 * Note that if there are multiple unserialized concurrent users of the
1227 * same timer, then mod_timer() is the only safe way to modify the timeout,
1228 * since add_timer() cannot modify an already running timer.
1230 * If @timer->function == NULL then the start operation is silently
1231 * discarded. In this case the return value is 0 and meaningless.
1234 * * %0 - The timer was inactive and started or was in shutdown
1235 * state and the operation was discarded
1236 * * %1 - The timer was active and requeued to expire at @expires or
1237 * the timer was active and not modified because @expires did
1238 * not change the effective expiry time
1240 int mod_timer(struct timer_list *timer, unsigned long expires)
1242 return __mod_timer(timer, expires, 0);
1244 EXPORT_SYMBOL(mod_timer);
1247 * timer_reduce - Modify a timer's timeout if it would reduce the timeout
1248 * @timer: The timer to be modified
1249 * @expires: New absolute timeout in jiffies
1251 * timer_reduce() is very similar to mod_timer(), except that it will only
1252 * modify an enqueued timer if that would reduce the expiration time. If
1253 * @timer is not enqueued it starts the timer.
1255 * If @timer->function == NULL then the start operation is silently
1259 * * %0 - The timer was inactive and started or was in shutdown
1260 * state and the operation was discarded
1261 * * %1 - The timer was active and requeued to expire at @expires or
1262 * the timer was active and not modified because @expires
1263 * did not change the effective expiry time such that the
1264 * timer would expire earlier than already scheduled
1266 int timer_reduce(struct timer_list *timer, unsigned long expires)
1268 return __mod_timer(timer, expires, MOD_TIMER_REDUCE);
1270 EXPORT_SYMBOL(timer_reduce);
1273 * add_timer - Start a timer
1274 * @timer: The timer to be started
1276 * Start @timer to expire at @timer->expires in the future. @timer->expires
1277 * is the absolute expiry time measured in 'jiffies'. When the timer expires
1278 * timer->function(timer) will be invoked from soft interrupt context.
1280 * The @timer->expires and @timer->function fields must be set prior
1281 * to calling this function.
1283 * If @timer->function == NULL then the start operation is silently
1286 * If @timer->expires is already in the past @timer will be queued to
1287 * expire at the next timer tick.
1289 * This can only operate on an inactive timer. Attempts to invoke this on
1290 * an active timer are rejected with a warning.
1292 void add_timer(struct timer_list *timer)
1294 if (WARN_ON_ONCE(timer_pending(timer)))
1296 __mod_timer(timer, timer->expires, MOD_TIMER_NOTPENDING);
1298 EXPORT_SYMBOL(add_timer);
1301 * add_timer_local() - Start a timer on the local CPU
1302 * @timer: The timer to be started
1304 * Same as add_timer() except that the timer flag TIMER_PINNED is set.
1306 * See add_timer() for further details.
1308 void add_timer_local(struct timer_list *timer)
1310 if (WARN_ON_ONCE(timer_pending(timer)))
1312 timer->flags |= TIMER_PINNED;
1313 __mod_timer(timer, timer->expires, MOD_TIMER_NOTPENDING);
1315 EXPORT_SYMBOL(add_timer_local);
1318 * add_timer_global() - Start a timer without TIMER_PINNED flag set
1319 * @timer: The timer to be started
1321 * Same as add_timer() except that the timer flag TIMER_PINNED is unset.
1323 * See add_timer() for further details.
1325 void add_timer_global(struct timer_list *timer)
1327 if (WARN_ON_ONCE(timer_pending(timer)))
1329 timer->flags &= ~TIMER_PINNED;
1330 __mod_timer(timer, timer->expires, MOD_TIMER_NOTPENDING);
1332 EXPORT_SYMBOL(add_timer_global);
1335 * add_timer_on - Start a timer on a particular CPU
1336 * @timer: The timer to be started
1337 * @cpu: The CPU to start it on
1339 * Same as add_timer() except that it starts the timer on the given CPU and
1340 * the TIMER_PINNED flag is set. When timer shouldn't be a pinned timer in
1341 * the next round, add_timer_global() should be used instead as it unsets
1342 * the TIMER_PINNED flag.
1344 * See add_timer() for further details.
1346 void add_timer_on(struct timer_list *timer, int cpu)
1348 struct timer_base *new_base, *base;
1349 unsigned long flags;
1351 debug_assert_init(timer);
1353 if (WARN_ON_ONCE(timer_pending(timer)))
1356 /* Make sure timer flags have TIMER_PINNED flag set */
1357 timer->flags |= TIMER_PINNED;
1359 new_base = get_timer_cpu_base(timer->flags, cpu);
1362 * If @timer was on a different CPU, it should be migrated with the
1363 * old base locked to prevent other operations proceeding with the
1364 * wrong base locked. See lock_timer_base().
1366 base = lock_timer_base(timer, &flags);
1368 * Has @timer been shutdown? This needs to be evaluated while
1369 * holding base lock to prevent a race against the shutdown code.
1371 if (!timer->function)
1374 if (base != new_base) {
1375 timer->flags |= TIMER_MIGRATING;
1377 raw_spin_unlock(&base->lock);
1379 raw_spin_lock(&base->lock);
1380 WRITE_ONCE(timer->flags,
1381 (timer->flags & ~TIMER_BASEMASK) | cpu);
1383 forward_timer_base(base);
1385 debug_timer_activate(timer);
1386 internal_add_timer(base, timer);
1388 raw_spin_unlock_irqrestore(&base->lock, flags);
1390 EXPORT_SYMBOL_GPL(add_timer_on);
1393 * __timer_delete - Internal function: Deactivate a timer
1394 * @timer: The timer to be deactivated
1395 * @shutdown: If true, this indicates that the timer is about to be
1396 * shutdown permanently.
1398 * If @shutdown is true then @timer->function is set to NULL under the
1399 * timer base lock which prevents further rearming of the time. In that
1400 * case any attempt to rearm @timer after this function returns will be
1404 * * %0 - The timer was not pending
1405 * * %1 - The timer was pending and deactivated
1407 static int __timer_delete(struct timer_list *timer, bool shutdown)
1409 struct timer_base *base;
1410 unsigned long flags;
1413 debug_assert_init(timer);
1416 * If @shutdown is set then the lock has to be taken whether the
1417 * timer is pending or not to protect against a concurrent rearm
1418 * which might hit between the lockless pending check and the lock
1419 * aquisition. By taking the lock it is ensured that such a newly
1420 * enqueued timer is dequeued and cannot end up with
1421 * timer->function == NULL in the expiry code.
1423 * If timer->function is currently executed, then this makes sure
1424 * that the callback cannot requeue the timer.
1426 if (timer_pending(timer) || shutdown) {
1427 base = lock_timer_base(timer, &flags);
1428 ret = detach_if_pending(timer, base, true);
1430 timer->function = NULL;
1431 raw_spin_unlock_irqrestore(&base->lock, flags);
1438 * timer_delete - Deactivate a timer
1439 * @timer: The timer to be deactivated
1441 * The function only deactivates a pending timer, but contrary to
1442 * timer_delete_sync() it does not take into account whether the timer's
1443 * callback function is concurrently executed on a different CPU or not.
1444 * It neither prevents rearming of the timer. If @timer can be rearmed
1445 * concurrently then the return value of this function is meaningless.
1448 * * %0 - The timer was not pending
1449 * * %1 - The timer was pending and deactivated
1451 int timer_delete(struct timer_list *timer)
1453 return __timer_delete(timer, false);
1455 EXPORT_SYMBOL(timer_delete);
1458 * timer_shutdown - Deactivate a timer and prevent rearming
1459 * @timer: The timer to be deactivated
1461 * The function does not wait for an eventually running timer callback on a
1462 * different CPU but it prevents rearming of the timer. Any attempt to arm
1463 * @timer after this function returns will be silently ignored.
1465 * This function is useful for teardown code and should only be used when
1466 * timer_shutdown_sync() cannot be invoked due to locking or context constraints.
1469 * * %0 - The timer was not pending
1470 * * %1 - The timer was pending
1472 int timer_shutdown(struct timer_list *timer)
1474 return __timer_delete(timer, true);
1476 EXPORT_SYMBOL_GPL(timer_shutdown);
1479 * __try_to_del_timer_sync - Internal function: Try to deactivate a timer
1480 * @timer: Timer to deactivate
1481 * @shutdown: If true, this indicates that the timer is about to be
1482 * shutdown permanently.
1484 * If @shutdown is true then @timer->function is set to NULL under the
1485 * timer base lock which prevents further rearming of the timer. Any
1486 * attempt to rearm @timer after this function returns will be silently
1489 * This function cannot guarantee that the timer cannot be rearmed
1490 * right after dropping the base lock if @shutdown is false. That
1491 * needs to be prevented by the calling code if necessary.
1494 * * %0 - The timer was not pending
1495 * * %1 - The timer was pending and deactivated
1496 * * %-1 - The timer callback function is running on a different CPU
1498 static int __try_to_del_timer_sync(struct timer_list *timer, bool shutdown)
1500 struct timer_base *base;
1501 unsigned long flags;
1504 debug_assert_init(timer);
1506 base = lock_timer_base(timer, &flags);
1508 if (base->running_timer != timer)
1509 ret = detach_if_pending(timer, base, true);
1511 timer->function = NULL;
1513 raw_spin_unlock_irqrestore(&base->lock, flags);
1519 * try_to_del_timer_sync - Try to deactivate a timer
1520 * @timer: Timer to deactivate
1522 * This function tries to deactivate a timer. On success the timer is not
1523 * queued and the timer callback function is not running on any CPU.
1525 * This function does not guarantee that the timer cannot be rearmed right
1526 * after dropping the base lock. That needs to be prevented by the calling
1527 * code if necessary.
1530 * * %0 - The timer was not pending
1531 * * %1 - The timer was pending and deactivated
1532 * * %-1 - The timer callback function is running on a different CPU
1534 int try_to_del_timer_sync(struct timer_list *timer)
1536 return __try_to_del_timer_sync(timer, false);
1538 EXPORT_SYMBOL(try_to_del_timer_sync);
1540 #ifdef CONFIG_PREEMPT_RT
1541 static __init void timer_base_init_expiry_lock(struct timer_base *base)
1543 spin_lock_init(&base->expiry_lock);
1546 static inline void timer_base_lock_expiry(struct timer_base *base)
1548 spin_lock(&base->expiry_lock);
1551 static inline void timer_base_unlock_expiry(struct timer_base *base)
1553 spin_unlock(&base->expiry_lock);
1557 * The counterpart to del_timer_wait_running().
1559 * If there is a waiter for base->expiry_lock, then it was waiting for the
1560 * timer callback to finish. Drop expiry_lock and reacquire it. That allows
1561 * the waiter to acquire the lock and make progress.
1563 static void timer_sync_wait_running(struct timer_base *base)
1565 if (atomic_read(&base->timer_waiters)) {
1566 raw_spin_unlock_irq(&base->lock);
1567 spin_unlock(&base->expiry_lock);
1568 spin_lock(&base->expiry_lock);
1569 raw_spin_lock_irq(&base->lock);
1574 * This function is called on PREEMPT_RT kernels when the fast path
1575 * deletion of a timer failed because the timer callback function was
1578 * This prevents priority inversion, if the softirq thread on a remote CPU
1579 * got preempted, and it prevents a life lock when the task which tries to
1580 * delete a timer preempted the softirq thread running the timer callback
1583 static void del_timer_wait_running(struct timer_list *timer)
1587 tf = READ_ONCE(timer->flags);
1588 if (!(tf & (TIMER_MIGRATING | TIMER_IRQSAFE))) {
1589 struct timer_base *base = get_timer_base(tf);
1592 * Mark the base as contended and grab the expiry lock,
1593 * which is held by the softirq across the timer
1594 * callback. Drop the lock immediately so the softirq can
1595 * expire the next timer. In theory the timer could already
1596 * be running again, but that's more than unlikely and just
1597 * causes another wait loop.
1599 atomic_inc(&base->timer_waiters);
1600 spin_lock_bh(&base->expiry_lock);
1601 atomic_dec(&base->timer_waiters);
1602 spin_unlock_bh(&base->expiry_lock);
1606 static inline void timer_base_init_expiry_lock(struct timer_base *base) { }
1607 static inline void timer_base_lock_expiry(struct timer_base *base) { }
1608 static inline void timer_base_unlock_expiry(struct timer_base *base) { }
1609 static inline void timer_sync_wait_running(struct timer_base *base) { }
1610 static inline void del_timer_wait_running(struct timer_list *timer) { }
1614 * __timer_delete_sync - Internal function: Deactivate a timer and wait
1615 * for the handler to finish.
1616 * @timer: The timer to be deactivated
1617 * @shutdown: If true, @timer->function will be set to NULL under the
1618 * timer base lock which prevents rearming of @timer
1620 * If @shutdown is not set the timer can be rearmed later. If the timer can
1621 * be rearmed concurrently, i.e. after dropping the base lock then the
1622 * return value is meaningless.
1624 * If @shutdown is set then @timer->function is set to NULL under timer
1625 * base lock which prevents rearming of the timer. Any attempt to rearm
1626 * a shutdown timer is silently ignored.
1628 * If the timer should be reused after shutdown it has to be initialized
1632 * * %0 - The timer was not pending
1633 * * %1 - The timer was pending and deactivated
1635 static int __timer_delete_sync(struct timer_list *timer, bool shutdown)
1639 #ifdef CONFIG_LOCKDEP
1640 unsigned long flags;
1643 * If lockdep gives a backtrace here, please reference
1644 * the synchronization rules above.
1646 local_irq_save(flags);
1647 lock_map_acquire(&timer->lockdep_map);
1648 lock_map_release(&timer->lockdep_map);
1649 local_irq_restore(flags);
1652 * don't use it in hardirq context, because it
1653 * could lead to deadlock.
1655 WARN_ON(in_hardirq() && !(timer->flags & TIMER_IRQSAFE));
1658 * Must be able to sleep on PREEMPT_RT because of the slowpath in
1659 * del_timer_wait_running().
1661 if (IS_ENABLED(CONFIG_PREEMPT_RT) && !(timer->flags & TIMER_IRQSAFE))
1662 lockdep_assert_preemption_enabled();
1665 ret = __try_to_del_timer_sync(timer, shutdown);
1667 if (unlikely(ret < 0)) {
1668 del_timer_wait_running(timer);
1677 * timer_delete_sync - Deactivate a timer and wait for the handler to finish.
1678 * @timer: The timer to be deactivated
1680 * Synchronization rules: Callers must prevent restarting of the timer,
1681 * otherwise this function is meaningless. It must not be called from
1682 * interrupt contexts unless the timer is an irqsafe one. The caller must
1683 * not hold locks which would prevent completion of the timer's callback
1684 * function. The timer's handler must not call add_timer_on(). Upon exit
1685 * the timer is not queued and the handler is not running on any CPU.
1687 * For !irqsafe timers, the caller must not hold locks that are held in
1688 * interrupt context. Even if the lock has nothing to do with the timer in
1689 * question. Here's why::
1695 * base->running_timer = mytimer;
1696 * spin_lock_irq(somelock);
1698 * spin_lock(somelock);
1699 * timer_delete_sync(mytimer);
1700 * while (base->running_timer == mytimer);
1702 * Now timer_delete_sync() will never return and never release somelock.
1703 * The interrupt on the other CPU is waiting to grab somelock but it has
1704 * interrupted the softirq that CPU0 is waiting to finish.
1706 * This function cannot guarantee that the timer is not rearmed again by
1707 * some concurrent or preempting code, right after it dropped the base
1708 * lock. If there is the possibility of a concurrent rearm then the return
1709 * value of the function is meaningless.
1711 * If such a guarantee is needed, e.g. for teardown situations then use
1712 * timer_shutdown_sync() instead.
1715 * * %0 - The timer was not pending
1716 * * %1 - The timer was pending and deactivated
1718 int timer_delete_sync(struct timer_list *timer)
1720 return __timer_delete_sync(timer, false);
1722 EXPORT_SYMBOL(timer_delete_sync);
1725 * timer_shutdown_sync - Shutdown a timer and prevent rearming
1726 * @timer: The timer to be shutdown
1728 * When the function returns it is guaranteed that:
1729 * - @timer is not queued
1730 * - The callback function of @timer is not running
1731 * - @timer cannot be enqueued again. Any attempt to rearm
1732 * @timer is silently ignored.
1734 * See timer_delete_sync() for synchronization rules.
1736 * This function is useful for final teardown of an infrastructure where
1737 * the timer is subject to a circular dependency problem.
1739 * A common pattern for this is a timer and a workqueue where the timer can
1740 * schedule work and work can arm the timer. On shutdown the workqueue must
1741 * be destroyed and the timer must be prevented from rearming. Unless the
1742 * code has conditionals like 'if (mything->in_shutdown)' to prevent that
1743 * there is no way to get this correct with timer_delete_sync().
1745 * timer_shutdown_sync() is solving the problem. The correct ordering of
1746 * calls in this case is:
1748 * timer_shutdown_sync(&mything->timer);
1749 * workqueue_destroy(&mything->workqueue);
1751 * After this 'mything' can be safely freed.
1753 * This obviously implies that the timer is not required to be functional
1754 * for the rest of the shutdown operation.
1757 * * %0 - The timer was not pending
1758 * * %1 - The timer was pending
1760 int timer_shutdown_sync(struct timer_list *timer)
1762 return __timer_delete_sync(timer, true);
1764 EXPORT_SYMBOL_GPL(timer_shutdown_sync);
1766 static void call_timer_fn(struct timer_list *timer,
1767 void (*fn)(struct timer_list *),
1768 unsigned long baseclk)
1770 int count = preempt_count();
1772 #ifdef CONFIG_LOCKDEP
1774 * It is permissible to free the timer from inside the
1775 * function that is called from it, this we need to take into
1776 * account for lockdep too. To avoid bogus "held lock freed"
1777 * warnings as well as problems when looking into
1778 * timer->lockdep_map, make a copy and use that here.
1780 struct lockdep_map lockdep_map;
1782 lockdep_copy_map(&lockdep_map, &timer->lockdep_map);
1785 * Couple the lock chain with the lock chain at
1786 * timer_delete_sync() by acquiring the lock_map around the fn()
1787 * call here and in timer_delete_sync().
1789 lock_map_acquire(&lockdep_map);
1791 trace_timer_expire_entry(timer, baseclk);
1793 trace_timer_expire_exit(timer);
1795 lock_map_release(&lockdep_map);
1797 if (count != preempt_count()) {
1798 WARN_ONCE(1, "timer: %pS preempt leak: %08x -> %08x\n",
1799 fn, count, preempt_count());
1801 * Restore the preempt count. That gives us a decent
1802 * chance to survive and extract information. If the
1803 * callback kept a lock held, bad luck, but not worse
1804 * than the BUG() we had.
1806 preempt_count_set(count);
1810 static void expire_timers(struct timer_base *base, struct hlist_head *head)
1813 * This value is required only for tracing. base->clk was
1814 * incremented directly before expire_timers was called. But expiry
1815 * is related to the old base->clk value.
1817 unsigned long baseclk = base->clk - 1;
1819 while (!hlist_empty(head)) {
1820 struct timer_list *timer;
1821 void (*fn)(struct timer_list *);
1823 timer = hlist_entry(head->first, struct timer_list, entry);
1825 base->running_timer = timer;
1826 detach_timer(timer, true);
1828 fn = timer->function;
1830 if (WARN_ON_ONCE(!fn)) {
1831 /* Should never happen. Emphasis on should! */
1832 base->running_timer = NULL;
1836 if (timer->flags & TIMER_IRQSAFE) {
1837 raw_spin_unlock(&base->lock);
1838 call_timer_fn(timer, fn, baseclk);
1839 raw_spin_lock(&base->lock);
1840 base->running_timer = NULL;
1842 raw_spin_unlock_irq(&base->lock);
1843 call_timer_fn(timer, fn, baseclk);
1844 raw_spin_lock_irq(&base->lock);
1845 base->running_timer = NULL;
1846 timer_sync_wait_running(base);
1851 static int collect_expired_timers(struct timer_base *base,
1852 struct hlist_head *heads)
1854 unsigned long clk = base->clk = base->next_expiry;
1855 struct hlist_head *vec;
1859 for (i = 0; i < LVL_DEPTH; i++) {
1860 idx = (clk & LVL_MASK) + i * LVL_SIZE;
1862 if (__test_and_clear_bit(idx, base->pending_map)) {
1863 vec = base->vectors + idx;
1864 hlist_move_list(vec, heads++);
1867 /* Is it time to look at the next level? */
1868 if (clk & LVL_CLK_MASK)
1870 /* Shift clock for the next level granularity */
1871 clk >>= LVL_CLK_SHIFT;
1877 * Find the next pending bucket of a level. Search from level start (@offset)
1878 * + @clk upwards and if nothing there, search from start of the level
1879 * (@offset) up to @offset + clk.
1881 static int next_pending_bucket(struct timer_base *base, unsigned offset,
1884 unsigned pos, start = offset + clk;
1885 unsigned end = offset + LVL_SIZE;
1887 pos = find_next_bit(base->pending_map, end, start);
1891 pos = find_next_bit(base->pending_map, start, offset);
1892 return pos < start ? pos + LVL_SIZE - start : -1;
1896 * Search the first expiring timer in the various clock levels. Caller must
1899 * Store next expiry time in base->next_expiry.
1901 static void next_expiry_recalc(struct timer_base *base)
1903 unsigned long clk, next, adj;
1904 unsigned lvl, offset = 0;
1906 next = base->clk + NEXT_TIMER_MAX_DELTA;
1908 for (lvl = 0; lvl < LVL_DEPTH; lvl++, offset += LVL_SIZE) {
1909 int pos = next_pending_bucket(base, offset, clk & LVL_MASK);
1910 unsigned long lvl_clk = clk & LVL_CLK_MASK;
1913 unsigned long tmp = clk + (unsigned long) pos;
1915 tmp <<= LVL_SHIFT(lvl);
1916 if (time_before(tmp, next))
1920 * If the next expiration happens before we reach
1921 * the next level, no need to check further.
1923 if (pos <= ((LVL_CLK_DIV - lvl_clk) & LVL_CLK_MASK))
1927 * Clock for the next level. If the current level clock lower
1928 * bits are zero, we look at the next level as is. If not we
1929 * need to advance it by one because that's going to be the
1930 * next expiring bucket in that level. base->clk is the next
1931 * expiring jiffie. So in case of:
1933 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1936 * we have to look at all levels @index 0. With
1938 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1941 * LVL0 has the next expiring bucket @index 2. The upper
1942 * levels have the next expiring bucket @index 1.
1944 * In case that the propagation wraps the next level the same
1947 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1950 * So after looking at LVL0 we get:
1952 * LVL5 LVL4 LVL3 LVL2 LVL1
1955 * So no propagation from LVL1 to LVL2 because that happened
1956 * with the add already, but then we need to propagate further
1957 * from LVL2 to LVL3.
1959 * So the simple check whether the lower bits of the current
1960 * level are 0 or not is sufficient for all cases.
1962 adj = lvl_clk ? 1 : 0;
1963 clk >>= LVL_CLK_SHIFT;
1967 base->next_expiry = next;
1968 base->next_expiry_recalc = false;
1969 base->timers_pending = !(next == base->clk + NEXT_TIMER_MAX_DELTA);
1972 #ifdef CONFIG_NO_HZ_COMMON
1974 * Check, if the next hrtimer event is before the next timer wheel
1977 static u64 cmp_next_hrtimer_event(u64 basem, u64 expires)
1979 u64 nextevt = hrtimer_get_next_event();
1982 * If high resolution timers are enabled
1983 * hrtimer_get_next_event() returns KTIME_MAX.
1985 if (expires <= nextevt)
1989 * If the next timer is already expired, return the tick base
1990 * time so the tick is fired immediately.
1992 if (nextevt <= basem)
1996 * Round up to the next jiffie. High resolution timers are
1997 * off, so the hrtimers are expired in the tick and we need to
1998 * make sure that this tick really expires the timer to avoid
1999 * a ping pong of the nohz stop code.
2001 * Use DIV_ROUND_UP_ULL to prevent gcc calling __divdi3
2003 return DIV_ROUND_UP_ULL(nextevt, TICK_NSEC) * TICK_NSEC;
2006 static unsigned long next_timer_interrupt(struct timer_base *base,
2007 unsigned long basej)
2009 if (base->next_expiry_recalc)
2010 next_expiry_recalc(base);
2013 * Move next_expiry for the empty base into the future to prevent an
2014 * unnecessary raise of the timer softirq when the next_expiry value
2015 * will be reached even if there is no timer pending.
2017 * This update is also required to make timer_base::next_expiry values
2018 * easy comparable to find out which base holds the first pending timer.
2020 if (!base->timers_pending)
2021 base->next_expiry = basej + NEXT_TIMER_MAX_DELTA;
2023 return base->next_expiry;
2026 static unsigned long fetch_next_timer_interrupt(unsigned long basej, u64 basem,
2027 struct timer_base *base_local,
2028 struct timer_base *base_global,
2029 struct timer_events *tevt)
2031 unsigned long nextevt, nextevt_local, nextevt_global;
2034 nextevt_local = next_timer_interrupt(base_local, basej);
2035 nextevt_global = next_timer_interrupt(base_global, basej);
2037 local_first = time_before_eq(nextevt_local, nextevt_global);
2039 nextevt = local_first ? nextevt_local : nextevt_global;
2042 * If the @nextevt is at max. one tick away, use @nextevt and store
2043 * it in the local expiry value. The next global event is irrelevant in
2044 * this case and can be left as KTIME_MAX.
2046 if (time_before_eq(nextevt, basej + 1)) {
2047 /* If we missed a tick already, force 0 delta */
2048 if (time_before(nextevt, basej))
2050 tevt->local = basem + (u64)(nextevt - basej) * TICK_NSEC;
2053 * This is required for the remote check only but it doesn't
2054 * hurt, when it is done for both call sites:
2056 * * The remote callers will only take care of the global timers
2057 * as local timers will be handled by CPU itself. When not
2058 * updating tevt->global with the already missed first global
2059 * timer, it is possible that it will be missed completely.
2061 * * The local callers will ignore the tevt->global anyway, when
2062 * nextevt is max. one tick away.
2065 tevt->global = tevt->local;
2070 * Update tevt.* values:
2072 * If the local queue expires first, then the global event can be
2073 * ignored. If the global queue is empty, nothing to do either.
2075 if (!local_first && base_global->timers_pending)
2076 tevt->global = basem + (u64)(nextevt_global - basej) * TICK_NSEC;
2078 if (base_local->timers_pending)
2079 tevt->local = basem + (u64)(nextevt_local - basej) * TICK_NSEC;
2086 * fetch_next_timer_interrupt_remote() - Store next timers into @tevt
2087 * @basej: base time jiffies
2088 * @basem: base time clock monotonic
2089 * @tevt: Pointer to the storage for the expiry values
2092 * Stores the next pending local and global timer expiry values in the
2093 * struct pointed to by @tevt. If a queue is empty the corresponding
2094 * field is set to KTIME_MAX. If local event expires before global
2095 * event, global event is set to KTIME_MAX as well.
2097 * Caller needs to make sure timer base locks are held (use
2098 * timer_lock_remote_bases() for this purpose).
2100 void fetch_next_timer_interrupt_remote(unsigned long basej, u64 basem,
2101 struct timer_events *tevt,
2104 struct timer_base *base_local, *base_global;
2106 /* Preset local / global events */
2107 tevt->local = tevt->global = KTIME_MAX;
2109 base_local = per_cpu_ptr(&timer_bases[BASE_LOCAL], cpu);
2110 base_global = per_cpu_ptr(&timer_bases[BASE_GLOBAL], cpu);
2112 lockdep_assert_held(&base_local->lock);
2113 lockdep_assert_held(&base_global->lock);
2115 fetch_next_timer_interrupt(basej, basem, base_local, base_global, tevt);
2119 * timer_unlock_remote_bases - unlock timer bases of cpu
2122 * Unlocks the remote timer bases.
2124 void timer_unlock_remote_bases(unsigned int cpu)
2125 __releases(timer_bases[BASE_LOCAL]->lock)
2126 __releases(timer_bases[BASE_GLOBAL]->lock)
2128 struct timer_base *base_local, *base_global;
2130 base_local = per_cpu_ptr(&timer_bases[BASE_LOCAL], cpu);
2131 base_global = per_cpu_ptr(&timer_bases[BASE_GLOBAL], cpu);
2133 raw_spin_unlock(&base_global->lock);
2134 raw_spin_unlock(&base_local->lock);
2138 * timer_lock_remote_bases - lock timer bases of cpu
2141 * Locks the remote timer bases.
2143 void timer_lock_remote_bases(unsigned int cpu)
2144 __acquires(timer_bases[BASE_LOCAL]->lock)
2145 __acquires(timer_bases[BASE_GLOBAL]->lock)
2147 struct timer_base *base_local, *base_global;
2149 base_local = per_cpu_ptr(&timer_bases[BASE_LOCAL], cpu);
2150 base_global = per_cpu_ptr(&timer_bases[BASE_GLOBAL], cpu);
2152 lockdep_assert_irqs_disabled();
2154 raw_spin_lock(&base_local->lock);
2155 raw_spin_lock_nested(&base_global->lock, SINGLE_DEPTH_NESTING);
2159 * timer_base_is_idle() - Return whether timer base is set idle
2161 * Returns value of local timer base is_idle value.
2163 bool timer_base_is_idle(void)
2165 return __this_cpu_read(timer_bases[BASE_LOCAL].is_idle);
2168 static void __run_timer_base(struct timer_base *base);
2171 * timer_expire_remote() - expire global timers of cpu
2174 * Expire timers of global base of remote CPU.
2176 void timer_expire_remote(unsigned int cpu)
2178 struct timer_base *base = per_cpu_ptr(&timer_bases[BASE_GLOBAL], cpu);
2180 __run_timer_base(base);
2183 static void timer_use_tmigr(unsigned long basej, u64 basem,
2184 unsigned long *nextevt, bool *tick_stop_path,
2185 bool timer_base_idle, struct timer_events *tevt)
2189 if (timer_base_idle)
2190 next_tmigr = tmigr_cpu_new_timer(tevt->global);
2191 else if (tick_stop_path)
2192 next_tmigr = tmigr_cpu_deactivate(tevt->global);
2194 next_tmigr = tmigr_quick_check(tevt->global);
2197 * If the CPU is the last going idle in timer migration hierarchy, make
2198 * sure the CPU will wake up in time to handle remote timers.
2199 * next_tmigr == KTIME_MAX if other CPUs are still active.
2201 if (next_tmigr < tevt->local) {
2204 /* If we missed a tick already, force 0 delta */
2205 if (next_tmigr < basem)
2208 tmp = div_u64(next_tmigr - basem, TICK_NSEC);
2210 *nextevt = basej + (unsigned long)tmp;
2211 tevt->local = next_tmigr;
2215 static void timer_use_tmigr(unsigned long basej, u64 basem,
2216 unsigned long *nextevt, bool *tick_stop_path,
2217 bool timer_base_idle, struct timer_events *tevt)
2220 * Make sure first event is written into tevt->local to not miss a
2221 * timer on !SMP systems.
2223 tevt->local = min_t(u64, tevt->local, tevt->global);
2225 # endif /* CONFIG_SMP */
2227 static inline u64 __get_next_timer_interrupt(unsigned long basej, u64 basem,
2230 struct timer_events tevt = { .local = KTIME_MAX, .global = KTIME_MAX };
2231 struct timer_base *base_local, *base_global;
2232 unsigned long nextevt;
2233 bool idle_is_possible;
2236 * When the CPU is offline, the tick is cancelled and nothing is supposed
2237 * to try to stop it.
2239 if (WARN_ON_ONCE(cpu_is_offline(smp_processor_id()))) {
2245 base_local = this_cpu_ptr(&timer_bases[BASE_LOCAL]);
2246 base_global = this_cpu_ptr(&timer_bases[BASE_GLOBAL]);
2248 raw_spin_lock(&base_local->lock);
2249 raw_spin_lock_nested(&base_global->lock, SINGLE_DEPTH_NESTING);
2251 nextevt = fetch_next_timer_interrupt(basej, basem, base_local,
2252 base_global, &tevt);
2255 * If the next event is only one jiffie ahead there is no need to call
2256 * timer migration hierarchy related functions. The value for the next
2257 * global timer in @tevt struct equals then KTIME_MAX. This is also
2258 * true, when the timer base is idle.
2260 * The proper timer migration hierarchy function depends on the callsite
2261 * and whether timer base is idle or not. @nextevt will be updated when
2262 * this CPU needs to handle the first timer migration hierarchy
2263 * event. See timer_use_tmigr() for detailed information.
2265 idle_is_possible = time_after(nextevt, basej + 1);
2266 if (idle_is_possible)
2267 timer_use_tmigr(basej, basem, &nextevt, idle,
2268 base_local->is_idle, &tevt);
2271 * We have a fresh next event. Check whether we can forward the
2274 __forward_timer_base(base_local, basej);
2275 __forward_timer_base(base_global, basej);
2278 * Set base->is_idle only when caller is timer_base_try_to_set_idle()
2282 * Bases are idle if the next event is more than a tick
2283 * away. Caution: @nextevt could have changed by enqueueing a
2284 * global timer into timer migration hierarchy. Therefore a new
2285 * check is required here.
2287 * If the base is marked idle then any timer add operation must
2288 * forward the base clk itself to keep granularity small. This
2289 * idle logic is only maintained for the BASE_LOCAL and
2290 * BASE_GLOBAL base, deferrable timers may still see large
2291 * granularity skew (by design).
2293 if (!base_local->is_idle && time_after(nextevt, basej + 1)) {
2294 base_local->is_idle = true;
2295 trace_timer_base_idle(true, base_local->cpu);
2297 *idle = base_local->is_idle;
2300 * When timer base is not set idle, undo the effect of
2301 * tmigr_cpu_deactivate() to prevent inconsitent states - active
2302 * timer base but inactive timer migration hierarchy.
2304 * When timer base was already marked idle, nothing will be
2307 if (!base_local->is_idle && idle_is_possible)
2308 tmigr_cpu_activate();
2311 raw_spin_unlock(&base_global->lock);
2312 raw_spin_unlock(&base_local->lock);
2314 return cmp_next_hrtimer_event(basem, tevt.local);
2318 * get_next_timer_interrupt() - return the time (clock mono) of the next timer
2319 * @basej: base time jiffies
2320 * @basem: base time clock monotonic
2322 * Returns the tick aligned clock monotonic time of the next pending timer or
2323 * KTIME_MAX if no timer is pending. If timer of global base was queued into
2324 * timer migration hierarchy, first global timer is not taken into account. If
2325 * it was the last CPU of timer migration hierarchy going idle, first global
2326 * event is taken into account.
2328 u64 get_next_timer_interrupt(unsigned long basej, u64 basem)
2330 return __get_next_timer_interrupt(basej, basem, NULL);
2334 * timer_base_try_to_set_idle() - Try to set the idle state of the timer bases
2335 * @basej: base time jiffies
2336 * @basem: base time clock monotonic
2337 * @idle: pointer to store the value of timer_base->is_idle on return;
2338 * *idle contains the information whether tick was already stopped
2340 * Returns the tick aligned clock monotonic time of the next pending timer or
2341 * KTIME_MAX if no timer is pending. When tick was already stopped KTIME_MAX is
2344 u64 timer_base_try_to_set_idle(unsigned long basej, u64 basem, bool *idle)
2349 return __get_next_timer_interrupt(basej, basem, idle);
2353 * timer_clear_idle - Clear the idle state of the timer base
2355 * Called with interrupts disabled
2357 void timer_clear_idle(void)
2360 * We do this unlocked. The worst outcome is a remote pinned timer
2361 * enqueue sending a pointless IPI, but taking the lock would just
2362 * make the window for sending the IPI a few instructions smaller
2363 * for the cost of taking the lock in the exit from idle
2364 * path. Required for BASE_LOCAL only.
2366 __this_cpu_write(timer_bases[BASE_LOCAL].is_idle, false);
2367 trace_timer_base_idle(false, smp_processor_id());
2369 /* Activate without holding the timer_base->lock */
2370 tmigr_cpu_activate();
2375 * __run_timers - run all expired timers (if any) on this CPU.
2376 * @base: the timer vector to be processed.
2378 static inline void __run_timers(struct timer_base *base)
2380 struct hlist_head heads[LVL_DEPTH];
2383 lockdep_assert_held(&base->lock);
2385 if (base->running_timer)
2388 while (time_after_eq(jiffies, base->clk) &&
2389 time_after_eq(jiffies, base->next_expiry)) {
2390 levels = collect_expired_timers(base, heads);
2392 * The two possible reasons for not finding any expired
2393 * timer at this clk are that all matching timers have been
2394 * dequeued or no timer has been queued since
2395 * base::next_expiry was set to base::clk +
2396 * NEXT_TIMER_MAX_DELTA.
2398 WARN_ON_ONCE(!levels && !base->next_expiry_recalc
2399 && base->timers_pending);
2401 * While executing timers, base->clk is set 1 offset ahead of
2402 * jiffies to avoid endless requeuing to current jiffies.
2405 next_expiry_recalc(base);
2408 expire_timers(base, heads + levels);
2412 static void __run_timer_base(struct timer_base *base)
2414 if (time_before(jiffies, base->next_expiry))
2417 timer_base_lock_expiry(base);
2418 raw_spin_lock_irq(&base->lock);
2420 raw_spin_unlock_irq(&base->lock);
2421 timer_base_unlock_expiry(base);
2424 static void run_timer_base(int index)
2426 struct timer_base *base = this_cpu_ptr(&timer_bases[index]);
2428 __run_timer_base(base);
2432 * This function runs timers and the timer-tq in bottom half context.
2434 static __latent_entropy void run_timer_softirq(struct softirq_action *h)
2436 run_timer_base(BASE_LOCAL);
2437 if (IS_ENABLED(CONFIG_NO_HZ_COMMON)) {
2438 run_timer_base(BASE_GLOBAL);
2439 run_timer_base(BASE_DEF);
2441 if (is_timers_nohz_active())
2442 tmigr_handle_remote();
2447 * Called by the local, per-CPU timer interrupt on SMP.
2449 static void run_local_timers(void)
2451 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_LOCAL]);
2453 hrtimer_run_queues();
2455 for (int i = 0; i < NR_BASES; i++, base++) {
2456 /* Raise the softirq only if required. */
2457 if (time_after_eq(jiffies, base->next_expiry) ||
2458 (i == BASE_DEF && tmigr_requires_handle_remote())) {
2459 raise_softirq(TIMER_SOFTIRQ);
2466 * Called from the timer interrupt handler to charge one tick to the current
2467 * process. user_tick is 1 if the tick is user time, 0 for system.
2469 void update_process_times(int user_tick)
2471 struct task_struct *p = current;
2473 /* Note: this timer irq context must be accounted for as well. */
2474 account_process_tick(p, user_tick);
2476 rcu_sched_clock_irq(user_tick);
2477 #ifdef CONFIG_IRQ_WORK
2482 if (IS_ENABLED(CONFIG_POSIX_TIMERS))
2483 run_posix_cpu_timers();
2487 * Since schedule_timeout()'s timer is defined on the stack, it must store
2488 * the target task on the stack as well.
2490 struct process_timer {
2491 struct timer_list timer;
2492 struct task_struct *task;
2495 static void process_timeout(struct timer_list *t)
2497 struct process_timer *timeout = from_timer(timeout, t, timer);
2499 wake_up_process(timeout->task);
2503 * schedule_timeout - sleep until timeout
2504 * @timeout: timeout value in jiffies
2506 * Make the current task sleep until @timeout jiffies have elapsed.
2507 * The function behavior depends on the current task state
2508 * (see also set_current_state() description):
2510 * %TASK_RUNNING - the scheduler is called, but the task does not sleep
2511 * at all. That happens because sched_submit_work() does nothing for
2512 * tasks in %TASK_RUNNING state.
2514 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
2515 * pass before the routine returns unless the current task is explicitly
2516 * woken up, (e.g. by wake_up_process()).
2518 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
2519 * delivered to the current task or the current task is explicitly woken
2522 * The current task state is guaranteed to be %TASK_RUNNING when this
2525 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
2526 * the CPU away without a bound on the timeout. In this case the return
2527 * value will be %MAX_SCHEDULE_TIMEOUT.
2529 * Returns 0 when the timer has expired otherwise the remaining time in
2530 * jiffies will be returned. In all cases the return value is guaranteed
2531 * to be non-negative.
2533 signed long __sched schedule_timeout(signed long timeout)
2535 struct process_timer timer;
2536 unsigned long expire;
2540 case MAX_SCHEDULE_TIMEOUT:
2542 * These two special cases are useful to be comfortable
2543 * in the caller. Nothing more. We could take
2544 * MAX_SCHEDULE_TIMEOUT from one of the negative value
2545 * but I' d like to return a valid offset (>=0) to allow
2546 * the caller to do everything it want with the retval.
2552 * Another bit of PARANOID. Note that the retval will be
2553 * 0 since no piece of kernel is supposed to do a check
2554 * for a negative retval of schedule_timeout() (since it
2555 * should never happens anyway). You just have the printk()
2556 * that will tell you if something is gone wrong and where.
2559 printk(KERN_ERR "schedule_timeout: wrong timeout "
2560 "value %lx\n", timeout);
2562 __set_current_state(TASK_RUNNING);
2567 expire = timeout + jiffies;
2569 timer.task = current;
2570 timer_setup_on_stack(&timer.timer, process_timeout, 0);
2571 __mod_timer(&timer.timer, expire, MOD_TIMER_NOTPENDING);
2573 del_timer_sync(&timer.timer);
2575 /* Remove the timer from the object tracker */
2576 destroy_timer_on_stack(&timer.timer);
2578 timeout = expire - jiffies;
2581 return timeout < 0 ? 0 : timeout;
2583 EXPORT_SYMBOL(schedule_timeout);
2586 * We can use __set_current_state() here because schedule_timeout() calls
2587 * schedule() unconditionally.
2589 signed long __sched schedule_timeout_interruptible(signed long timeout)
2591 __set_current_state(TASK_INTERRUPTIBLE);
2592 return schedule_timeout(timeout);
2594 EXPORT_SYMBOL(schedule_timeout_interruptible);
2596 signed long __sched schedule_timeout_killable(signed long timeout)
2598 __set_current_state(TASK_KILLABLE);
2599 return schedule_timeout(timeout);
2601 EXPORT_SYMBOL(schedule_timeout_killable);
2603 signed long __sched schedule_timeout_uninterruptible(signed long timeout)
2605 __set_current_state(TASK_UNINTERRUPTIBLE);
2606 return schedule_timeout(timeout);
2608 EXPORT_SYMBOL(schedule_timeout_uninterruptible);
2611 * Like schedule_timeout_uninterruptible(), except this task will not contribute
2614 signed long __sched schedule_timeout_idle(signed long timeout)
2616 __set_current_state(TASK_IDLE);
2617 return schedule_timeout(timeout);
2619 EXPORT_SYMBOL(schedule_timeout_idle);
2621 #ifdef CONFIG_HOTPLUG_CPU
2622 static void migrate_timer_list(struct timer_base *new_base, struct hlist_head *head)
2624 struct timer_list *timer;
2625 int cpu = new_base->cpu;
2627 while (!hlist_empty(head)) {
2628 timer = hlist_entry(head->first, struct timer_list, entry);
2629 detach_timer(timer, false);
2630 timer->flags = (timer->flags & ~TIMER_BASEMASK) | cpu;
2631 internal_add_timer(new_base, timer);
2635 int timers_prepare_cpu(unsigned int cpu)
2637 struct timer_base *base;
2640 for (b = 0; b < NR_BASES; b++) {
2641 base = per_cpu_ptr(&timer_bases[b], cpu);
2642 base->clk = jiffies;
2643 base->next_expiry = base->clk + NEXT_TIMER_MAX_DELTA;
2644 base->next_expiry_recalc = false;
2645 base->timers_pending = false;
2646 base->is_idle = false;
2651 int timers_dead_cpu(unsigned int cpu)
2653 struct timer_base *old_base;
2654 struct timer_base *new_base;
2657 for (b = 0; b < NR_BASES; b++) {
2658 old_base = per_cpu_ptr(&timer_bases[b], cpu);
2659 new_base = get_cpu_ptr(&timer_bases[b]);
2661 * The caller is globally serialized and nobody else
2662 * takes two locks at once, deadlock is not possible.
2664 raw_spin_lock_irq(&new_base->lock);
2665 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
2668 * The current CPUs base clock might be stale. Update it
2669 * before moving the timers over.
2671 forward_timer_base(new_base);
2673 WARN_ON_ONCE(old_base->running_timer);
2674 old_base->running_timer = NULL;
2676 for (i = 0; i < WHEEL_SIZE; i++)
2677 migrate_timer_list(new_base, old_base->vectors + i);
2679 raw_spin_unlock(&old_base->lock);
2680 raw_spin_unlock_irq(&new_base->lock);
2681 put_cpu_ptr(&timer_bases);
2686 #endif /* CONFIG_HOTPLUG_CPU */
2688 static void __init init_timer_cpu(int cpu)
2690 struct timer_base *base;
2693 for (i = 0; i < NR_BASES; i++) {
2694 base = per_cpu_ptr(&timer_bases[i], cpu);
2696 raw_spin_lock_init(&base->lock);
2697 base->clk = jiffies;
2698 base->next_expiry = base->clk + NEXT_TIMER_MAX_DELTA;
2699 timer_base_init_expiry_lock(base);
2703 static void __init init_timer_cpus(void)
2707 for_each_possible_cpu(cpu)
2708 init_timer_cpu(cpu);
2711 void __init init_timers(void)
2714 posix_cputimers_init_work();
2715 open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
2719 * msleep - sleep safely even with waitqueue interruptions
2720 * @msecs: Time in milliseconds to sleep for
2722 void msleep(unsigned int msecs)
2724 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
2727 timeout = schedule_timeout_uninterruptible(timeout);
2730 EXPORT_SYMBOL(msleep);
2733 * msleep_interruptible - sleep waiting for signals
2734 * @msecs: Time in milliseconds to sleep for
2736 unsigned long msleep_interruptible(unsigned int msecs)
2738 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
2740 while (timeout && !signal_pending(current))
2741 timeout = schedule_timeout_interruptible(timeout);
2742 return jiffies_to_msecs(timeout);
2745 EXPORT_SYMBOL(msleep_interruptible);
2748 * usleep_range_state - Sleep for an approximate time in a given state
2749 * @min: Minimum time in usecs to sleep
2750 * @max: Maximum time in usecs to sleep
2751 * @state: State of the current task that will be while sleeping
2753 * In non-atomic context where the exact wakeup time is flexible, use
2754 * usleep_range_state() instead of udelay(). The sleep improves responsiveness
2755 * by avoiding the CPU-hogging busy-wait of udelay(), and the range reduces
2756 * power usage by allowing hrtimers to take advantage of an already-
2757 * scheduled interrupt instead of scheduling a new one just for this sleep.
2759 void __sched usleep_range_state(unsigned long min, unsigned long max,
2762 ktime_t exp = ktime_add_us(ktime_get(), min);
2763 u64 delta = (u64)(max - min) * NSEC_PER_USEC;
2766 __set_current_state(state);
2767 /* Do not return before the requested sleep time has elapsed */
2768 if (!schedule_hrtimeout_range(&exp, delta, HRTIMER_MODE_ABS))
2772 EXPORT_SYMBOL(usleep_range_state);