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>
48 #include <linux/uaccess.h>
49 #include <asm/unistd.h>
50 #include <asm/div64.h>
51 #include <asm/timex.h>
54 #include "tick-internal.h"
56 #define CREATE_TRACE_POINTS
57 #include <trace/events/timer.h>
59 __visible u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
61 EXPORT_SYMBOL(jiffies_64);
64 * The timer wheel has LVL_DEPTH array levels. Each level provides an array of
65 * LVL_SIZE buckets. Each level is driven by its own clock and therefor each
66 * level has a different granularity.
68 * The level granularity is: LVL_CLK_DIV ^ lvl
69 * The level clock frequency is: HZ / (LVL_CLK_DIV ^ level)
71 * The array level of a newly armed timer depends on the relative expiry
72 * time. The farther the expiry time is away the higher the array level and
73 * therefor the granularity becomes.
75 * Contrary to the original timer wheel implementation, which aims for 'exact'
76 * expiry of the timers, this implementation removes the need for recascading
77 * the timers into the lower array levels. The previous 'classic' timer wheel
78 * implementation of the kernel already violated the 'exact' expiry by adding
79 * slack to the expiry time to provide batched expiration. The granularity
80 * levels provide implicit batching.
82 * This is an optimization of the original timer wheel implementation for the
83 * majority of the timer wheel use cases: timeouts. The vast majority of
84 * timeout timers (networking, disk I/O ...) are canceled before expiry. If
85 * the timeout expires it indicates that normal operation is disturbed, so it
86 * does not matter much whether the timeout comes with a slight delay.
88 * The only exception to this are networking timers with a small expiry
89 * time. They rely on the granularity. Those fit into the first wheel level,
90 * which has HZ granularity.
92 * We don't have cascading anymore. timers with a expiry time above the
93 * capacity of the last wheel level are force expired at the maximum timeout
94 * value of the last wheel level. From data sampling we know that the maximum
95 * value observed is 5 days (network connection tracking), so this should not
98 * The currently chosen array constants values are a good compromise between
99 * array size and granularity.
101 * This results in the following granularity and range levels:
104 * Level Offset Granularity Range
105 * 0 0 1 ms 0 ms - 63 ms
106 * 1 64 8 ms 64 ms - 511 ms
107 * 2 128 64 ms 512 ms - 4095 ms (512ms - ~4s)
108 * 3 192 512 ms 4096 ms - 32767 ms (~4s - ~32s)
109 * 4 256 4096 ms (~4s) 32768 ms - 262143 ms (~32s - ~4m)
110 * 5 320 32768 ms (~32s) 262144 ms - 2097151 ms (~4m - ~34m)
111 * 6 384 262144 ms (~4m) 2097152 ms - 16777215 ms (~34m - ~4h)
112 * 7 448 2097152 ms (~34m) 16777216 ms - 134217727 ms (~4h - ~1d)
113 * 8 512 16777216 ms (~4h) 134217728 ms - 1073741822 ms (~1d - ~12d)
116 * Level Offset Granularity Range
117 * 0 0 3 ms 0 ms - 210 ms
118 * 1 64 26 ms 213 ms - 1703 ms (213ms - ~1s)
119 * 2 128 213 ms 1706 ms - 13650 ms (~1s - ~13s)
120 * 3 192 1706 ms (~1s) 13653 ms - 109223 ms (~13s - ~1m)
121 * 4 256 13653 ms (~13s) 109226 ms - 873810 ms (~1m - ~14m)
122 * 5 320 109226 ms (~1m) 873813 ms - 6990503 ms (~14m - ~1h)
123 * 6 384 873813 ms (~14m) 6990506 ms - 55924050 ms (~1h - ~15h)
124 * 7 448 6990506 ms (~1h) 55924053 ms - 447392423 ms (~15h - ~5d)
125 * 8 512 55924053 ms (~15h) 447392426 ms - 3579139406 ms (~5d - ~41d)
128 * Level Offset Granularity Range
129 * 0 0 4 ms 0 ms - 255 ms
130 * 1 64 32 ms 256 ms - 2047 ms (256ms - ~2s)
131 * 2 128 256 ms 2048 ms - 16383 ms (~2s - ~16s)
132 * 3 192 2048 ms (~2s) 16384 ms - 131071 ms (~16s - ~2m)
133 * 4 256 16384 ms (~16s) 131072 ms - 1048575 ms (~2m - ~17m)
134 * 5 320 131072 ms (~2m) 1048576 ms - 8388607 ms (~17m - ~2h)
135 * 6 384 1048576 ms (~17m) 8388608 ms - 67108863 ms (~2h - ~18h)
136 * 7 448 8388608 ms (~2h) 67108864 ms - 536870911 ms (~18h - ~6d)
137 * 8 512 67108864 ms (~18h) 536870912 ms - 4294967288 ms (~6d - ~49d)
140 * Level Offset Granularity Range
141 * 0 0 10 ms 0 ms - 630 ms
142 * 1 64 80 ms 640 ms - 5110 ms (640ms - ~5s)
143 * 2 128 640 ms 5120 ms - 40950 ms (~5s - ~40s)
144 * 3 192 5120 ms (~5s) 40960 ms - 327670 ms (~40s - ~5m)
145 * 4 256 40960 ms (~40s) 327680 ms - 2621430 ms (~5m - ~43m)
146 * 5 320 327680 ms (~5m) 2621440 ms - 20971510 ms (~43m - ~5h)
147 * 6 384 2621440 ms (~43m) 20971520 ms - 167772150 ms (~5h - ~1d)
148 * 7 448 20971520 ms (~5h) 167772160 ms - 1342177270 ms (~1d - ~15d)
151 /* Clock divisor for the next level */
152 #define LVL_CLK_SHIFT 3
153 #define LVL_CLK_DIV (1UL << LVL_CLK_SHIFT)
154 #define LVL_CLK_MASK (LVL_CLK_DIV - 1)
155 #define LVL_SHIFT(n) ((n) * LVL_CLK_SHIFT)
156 #define LVL_GRAN(n) (1UL << LVL_SHIFT(n))
159 * The time start value for each level to select the bucket at enqueue
160 * time. We start from the last possible delta of the previous level
161 * so that we can later add an extra LVL_GRAN(n) to n (see calc_index()).
163 #define LVL_START(n) ((LVL_SIZE - 1) << (((n) - 1) * LVL_CLK_SHIFT))
165 /* Size of each clock level */
167 #define LVL_SIZE (1UL << LVL_BITS)
168 #define LVL_MASK (LVL_SIZE - 1)
169 #define LVL_OFFS(n) ((n) * LVL_SIZE)
178 /* The cutoff (max. capacity of the wheel) */
179 #define WHEEL_TIMEOUT_CUTOFF (LVL_START(LVL_DEPTH))
180 #define WHEEL_TIMEOUT_MAX (WHEEL_TIMEOUT_CUTOFF - LVL_GRAN(LVL_DEPTH - 1))
183 * The resulting wheel size. If NOHZ is configured we allocate two
184 * wheels so we have a separate storage for the deferrable timers.
186 #define WHEEL_SIZE (LVL_SIZE * LVL_DEPTH)
188 #ifdef CONFIG_NO_HZ_COMMON
200 struct timer_list *running_timer;
201 #ifdef CONFIG_PREEMPT_RT
202 spinlock_t expiry_lock;
203 atomic_t timer_waiters;
206 unsigned long next_expiry;
208 bool next_expiry_recalc;
211 DECLARE_BITMAP(pending_map, WHEEL_SIZE);
212 struct hlist_head vectors[WHEEL_SIZE];
213 } ____cacheline_aligned;
215 static DEFINE_PER_CPU(struct timer_base, timer_bases[NR_BASES]);
217 #ifdef CONFIG_NO_HZ_COMMON
219 static DEFINE_STATIC_KEY_FALSE(timers_nohz_active);
220 static DEFINE_MUTEX(timer_keys_mutex);
222 static void timer_update_keys(struct work_struct *work);
223 static DECLARE_WORK(timer_update_work, timer_update_keys);
226 unsigned int sysctl_timer_migration = 1;
228 DEFINE_STATIC_KEY_FALSE(timers_migration_enabled);
230 static void timers_update_migration(void)
232 if (sysctl_timer_migration && tick_nohz_active)
233 static_branch_enable(&timers_migration_enabled);
235 static_branch_disable(&timers_migration_enabled);
238 static inline void timers_update_migration(void) { }
239 #endif /* !CONFIG_SMP */
241 static void timer_update_keys(struct work_struct *work)
243 mutex_lock(&timer_keys_mutex);
244 timers_update_migration();
245 static_branch_enable(&timers_nohz_active);
246 mutex_unlock(&timer_keys_mutex);
249 void timers_update_nohz(void)
251 schedule_work(&timer_update_work);
254 int timer_migration_handler(struct ctl_table *table, int write,
255 void *buffer, size_t *lenp, loff_t *ppos)
259 mutex_lock(&timer_keys_mutex);
260 ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
262 timers_update_migration();
263 mutex_unlock(&timer_keys_mutex);
267 static inline bool is_timers_nohz_active(void)
269 return static_branch_unlikely(&timers_nohz_active);
272 static inline bool is_timers_nohz_active(void) { return false; }
273 #endif /* NO_HZ_COMMON */
275 static unsigned long round_jiffies_common(unsigned long j, int cpu,
279 unsigned long original = j;
282 * We don't want all cpus firing their timers at once hitting the
283 * same lock or cachelines, so we skew each extra cpu with an extra
284 * 3 jiffies. This 3 jiffies came originally from the mm/ code which
286 * The skew is done by adding 3*cpunr, then round, then subtract this
287 * extra offset again.
294 * If the target jiffie is just after a whole second (which can happen
295 * due to delays of the timer irq, long irq off times etc etc) then
296 * we should round down to the whole second, not up. Use 1/4th second
297 * as cutoff for this rounding as an extreme upper bound for this.
298 * But never round down if @force_up is set.
300 if (rem < HZ/4 && !force_up) /* round down */
305 /* now that we have rounded, subtract the extra skew again */
309 * Make sure j is still in the future. Otherwise return the
312 return time_is_after_jiffies(j) ? j : original;
316 * __round_jiffies - function to round jiffies to a full second
317 * @j: the time in (absolute) jiffies that should be rounded
318 * @cpu: the processor number on which the timeout will happen
320 * __round_jiffies() rounds an absolute time in the future (in jiffies)
321 * up or down to (approximately) full seconds. This is useful for timers
322 * for which the exact time they fire does not matter too much, as long as
323 * they fire approximately every X seconds.
325 * By rounding these timers to whole seconds, all such timers will fire
326 * at the same time, rather than at various times spread out. The goal
327 * of this is to have the CPU wake up less, which saves power.
329 * The exact rounding is skewed for each processor to avoid all
330 * processors firing at the exact same time, which could lead
331 * to lock contention or spurious cache line bouncing.
333 * The return value is the rounded version of the @j parameter.
335 unsigned long __round_jiffies(unsigned long j, int cpu)
337 return round_jiffies_common(j, cpu, false);
339 EXPORT_SYMBOL_GPL(__round_jiffies);
342 * __round_jiffies_relative - function to round jiffies to a full second
343 * @j: the time in (relative) jiffies that should be rounded
344 * @cpu: the processor number on which the timeout will happen
346 * __round_jiffies_relative() rounds a time delta in the future (in jiffies)
347 * up or down to (approximately) full seconds. This is useful for timers
348 * for which the exact time they fire does not matter too much, as long as
349 * they fire approximately every X seconds.
351 * By rounding these timers to whole seconds, all such timers will fire
352 * at the same time, rather than at various times spread out. The goal
353 * of this is to have the CPU wake up less, which saves power.
355 * The exact rounding is skewed for each processor to avoid all
356 * processors firing at the exact same time, which could lead
357 * to lock contention or spurious cache line bouncing.
359 * The return value is the rounded version of the @j parameter.
361 unsigned long __round_jiffies_relative(unsigned long j, int cpu)
363 unsigned long j0 = jiffies;
365 /* Use j0 because jiffies might change while we run */
366 return round_jiffies_common(j + j0, cpu, false) - j0;
368 EXPORT_SYMBOL_GPL(__round_jiffies_relative);
371 * round_jiffies - function to round jiffies to a full second
372 * @j: the time in (absolute) jiffies that should be rounded
374 * round_jiffies() rounds an absolute time in the future (in jiffies)
375 * up or down to (approximately) full seconds. This is useful for timers
376 * for which the exact time they fire does not matter too much, as long as
377 * they fire approximately every X seconds.
379 * By rounding these timers to whole seconds, all such timers will fire
380 * at the same time, rather than at various times spread out. The goal
381 * of this is to have the CPU wake up less, which saves power.
383 * The return value is the rounded version of the @j parameter.
385 unsigned long round_jiffies(unsigned long j)
387 return round_jiffies_common(j, raw_smp_processor_id(), false);
389 EXPORT_SYMBOL_GPL(round_jiffies);
392 * round_jiffies_relative - function to round jiffies to a full second
393 * @j: the time in (relative) jiffies that should be rounded
395 * round_jiffies_relative() rounds a time delta 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 return value is the rounded version of the @j parameter.
406 unsigned long round_jiffies_relative(unsigned long j)
408 return __round_jiffies_relative(j, raw_smp_processor_id());
410 EXPORT_SYMBOL_GPL(round_jiffies_relative);
413 * __round_jiffies_up - function to round jiffies up to a full second
414 * @j: the time in (absolute) jiffies that should be rounded
415 * @cpu: the processor number on which the timeout will happen
417 * This is the same as __round_jiffies() except that it will never
418 * round down. This is useful for timeouts for which the exact time
419 * of firing does not matter too much, as long as they don't fire too
422 unsigned long __round_jiffies_up(unsigned long j, int cpu)
424 return round_jiffies_common(j, cpu, true);
426 EXPORT_SYMBOL_GPL(__round_jiffies_up);
429 * __round_jiffies_up_relative - function to round jiffies up to a full second
430 * @j: the time in (relative) jiffies that should be rounded
431 * @cpu: the processor number on which the timeout will happen
433 * This is the same as __round_jiffies_relative() except that it will never
434 * round down. This is useful for timeouts for which the exact time
435 * of firing does not matter too much, as long as they don't fire too
438 unsigned long __round_jiffies_up_relative(unsigned long j, int cpu)
440 unsigned long j0 = jiffies;
442 /* Use j0 because jiffies might change while we run */
443 return round_jiffies_common(j + j0, cpu, true) - j0;
445 EXPORT_SYMBOL_GPL(__round_jiffies_up_relative);
448 * round_jiffies_up - function to round jiffies up to a full second
449 * @j: the time in (absolute) jiffies that should be rounded
451 * This is the same as round_jiffies() except that it will never
452 * round down. This is useful for timeouts for which the exact time
453 * of firing does not matter too much, as long as they don't fire too
456 unsigned long round_jiffies_up(unsigned long j)
458 return round_jiffies_common(j, raw_smp_processor_id(), true);
460 EXPORT_SYMBOL_GPL(round_jiffies_up);
463 * round_jiffies_up_relative - function to round jiffies up to a full second
464 * @j: the time in (relative) jiffies that should be rounded
466 * This is the same as round_jiffies_relative() except that it will never
467 * round down. This is useful for timeouts for which the exact time
468 * of firing does not matter too much, as long as they don't fire too
471 unsigned long round_jiffies_up_relative(unsigned long j)
473 return __round_jiffies_up_relative(j, raw_smp_processor_id());
475 EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
478 static inline unsigned int timer_get_idx(struct timer_list *timer)
480 return (timer->flags & TIMER_ARRAYMASK) >> TIMER_ARRAYSHIFT;
483 static inline void timer_set_idx(struct timer_list *timer, unsigned int idx)
485 timer->flags = (timer->flags & ~TIMER_ARRAYMASK) |
486 idx << TIMER_ARRAYSHIFT;
490 * Helper function to calculate the array index for a given expiry
493 static inline unsigned calc_index(unsigned long expires, unsigned lvl,
494 unsigned long *bucket_expiry)
498 * The timer wheel has to guarantee that a timer does not fire
499 * early. Early expiry can happen due to:
500 * - Timer is armed at the edge of a tick
501 * - Truncation of the expiry time in the outer wheel levels
503 * Round up with level granularity to prevent this.
505 expires = (expires + LVL_GRAN(lvl)) >> LVL_SHIFT(lvl);
506 *bucket_expiry = expires << LVL_SHIFT(lvl);
507 return LVL_OFFS(lvl) + (expires & LVL_MASK);
510 static int calc_wheel_index(unsigned long expires, unsigned long clk,
511 unsigned long *bucket_expiry)
513 unsigned long delta = expires - clk;
516 if (delta < LVL_START(1)) {
517 idx = calc_index(expires, 0, bucket_expiry);
518 } else if (delta < LVL_START(2)) {
519 idx = calc_index(expires, 1, bucket_expiry);
520 } else if (delta < LVL_START(3)) {
521 idx = calc_index(expires, 2, bucket_expiry);
522 } else if (delta < LVL_START(4)) {
523 idx = calc_index(expires, 3, bucket_expiry);
524 } else if (delta < LVL_START(5)) {
525 idx = calc_index(expires, 4, bucket_expiry);
526 } else if (delta < LVL_START(6)) {
527 idx = calc_index(expires, 5, bucket_expiry);
528 } else if (delta < LVL_START(7)) {
529 idx = calc_index(expires, 6, bucket_expiry);
530 } else if (LVL_DEPTH > 8 && delta < LVL_START(8)) {
531 idx = calc_index(expires, 7, bucket_expiry);
532 } else if ((long) delta < 0) {
533 idx = clk & LVL_MASK;
534 *bucket_expiry = clk;
537 * Force expire obscene large timeouts to expire at the
538 * capacity limit of the wheel.
540 if (delta >= WHEEL_TIMEOUT_CUTOFF)
541 expires = clk + WHEEL_TIMEOUT_MAX;
543 idx = calc_index(expires, LVL_DEPTH - 1, bucket_expiry);
549 trigger_dyntick_cpu(struct timer_base *base, struct timer_list *timer)
551 if (!is_timers_nohz_active())
555 * TODO: This wants some optimizing similar to the code below, but we
556 * will do that when we switch from push to pull for deferrable timers.
558 if (timer->flags & TIMER_DEFERRABLE) {
559 if (tick_nohz_full_cpu(base->cpu))
560 wake_up_nohz_cpu(base->cpu);
565 * We might have to IPI the remote CPU if the base is idle and the
566 * timer is not deferrable. If the other CPU is on the way to idle
567 * then it can't set base->is_idle as we hold the base lock:
570 wake_up_nohz_cpu(base->cpu);
574 * Enqueue the timer into the hash bucket, mark it pending in
575 * the bitmap, store the index in the timer flags then wake up
576 * the target CPU if needed.
578 static void enqueue_timer(struct timer_base *base, struct timer_list *timer,
579 unsigned int idx, unsigned long bucket_expiry)
582 hlist_add_head(&timer->entry, base->vectors + idx);
583 __set_bit(idx, base->pending_map);
584 timer_set_idx(timer, idx);
586 trace_timer_start(timer, timer->expires, timer->flags);
589 * Check whether this is the new first expiring timer. The
590 * effective expiry time of the timer is required here
591 * (bucket_expiry) instead of timer->expires.
593 if (time_before(bucket_expiry, base->next_expiry)) {
595 * Set the next expiry time and kick the CPU so it
596 * can reevaluate the wheel:
598 base->next_expiry = bucket_expiry;
599 base->timers_pending = true;
600 base->next_expiry_recalc = false;
601 trigger_dyntick_cpu(base, timer);
605 static void internal_add_timer(struct timer_base *base, struct timer_list *timer)
607 unsigned long bucket_expiry;
610 idx = calc_wheel_index(timer->expires, base->clk, &bucket_expiry);
611 enqueue_timer(base, timer, idx, bucket_expiry);
614 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
616 static const struct debug_obj_descr timer_debug_descr;
618 static void *timer_debug_hint(void *addr)
620 return ((struct timer_list *) addr)->function;
623 static bool timer_is_static_object(void *addr)
625 struct timer_list *timer = addr;
627 return (timer->entry.pprev == NULL &&
628 timer->entry.next == TIMER_ENTRY_STATIC);
632 * fixup_init is called when:
633 * - an active object is initialized
635 static bool timer_fixup_init(void *addr, enum debug_obj_state state)
637 struct timer_list *timer = addr;
640 case ODEBUG_STATE_ACTIVE:
641 del_timer_sync(timer);
642 debug_object_init(timer, &timer_debug_descr);
649 /* Stub timer callback for improperly used timers. */
650 static void stub_timer(struct timer_list *unused)
656 * fixup_activate is called when:
657 * - an active object is activated
658 * - an unknown non-static object is activated
660 static bool timer_fixup_activate(void *addr, enum debug_obj_state state)
662 struct timer_list *timer = addr;
665 case ODEBUG_STATE_NOTAVAILABLE:
666 timer_setup(timer, stub_timer, 0);
669 case ODEBUG_STATE_ACTIVE:
678 * fixup_free is called when:
679 * - an active object is freed
681 static bool timer_fixup_free(void *addr, enum debug_obj_state state)
683 struct timer_list *timer = addr;
686 case ODEBUG_STATE_ACTIVE:
687 del_timer_sync(timer);
688 debug_object_free(timer, &timer_debug_descr);
696 * fixup_assert_init is called when:
697 * - an untracked/uninit-ed object is found
699 static bool timer_fixup_assert_init(void *addr, enum debug_obj_state state)
701 struct timer_list *timer = addr;
704 case ODEBUG_STATE_NOTAVAILABLE:
705 timer_setup(timer, stub_timer, 0);
712 static const struct debug_obj_descr timer_debug_descr = {
713 .name = "timer_list",
714 .debug_hint = timer_debug_hint,
715 .is_static_object = timer_is_static_object,
716 .fixup_init = timer_fixup_init,
717 .fixup_activate = timer_fixup_activate,
718 .fixup_free = timer_fixup_free,
719 .fixup_assert_init = timer_fixup_assert_init,
722 static inline void debug_timer_init(struct timer_list *timer)
724 debug_object_init(timer, &timer_debug_descr);
727 static inline void debug_timer_activate(struct timer_list *timer)
729 debug_object_activate(timer, &timer_debug_descr);
732 static inline void debug_timer_deactivate(struct timer_list *timer)
734 debug_object_deactivate(timer, &timer_debug_descr);
737 static inline void debug_timer_assert_init(struct timer_list *timer)
739 debug_object_assert_init(timer, &timer_debug_descr);
742 static void do_init_timer(struct timer_list *timer,
743 void (*func)(struct timer_list *),
745 const char *name, struct lock_class_key *key);
747 void init_timer_on_stack_key(struct timer_list *timer,
748 void (*func)(struct timer_list *),
750 const char *name, struct lock_class_key *key)
752 debug_object_init_on_stack(timer, &timer_debug_descr);
753 do_init_timer(timer, func, flags, name, key);
755 EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
757 void destroy_timer_on_stack(struct timer_list *timer)
759 debug_object_free(timer, &timer_debug_descr);
761 EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
764 static inline void debug_timer_init(struct timer_list *timer) { }
765 static inline void debug_timer_activate(struct timer_list *timer) { }
766 static inline void debug_timer_deactivate(struct timer_list *timer) { }
767 static inline void debug_timer_assert_init(struct timer_list *timer) { }
770 static inline void debug_init(struct timer_list *timer)
772 debug_timer_init(timer);
773 trace_timer_init(timer);
776 static inline void debug_deactivate(struct timer_list *timer)
778 debug_timer_deactivate(timer);
779 trace_timer_cancel(timer);
782 static inline void debug_assert_init(struct timer_list *timer)
784 debug_timer_assert_init(timer);
787 static void do_init_timer(struct timer_list *timer,
788 void (*func)(struct timer_list *),
790 const char *name, struct lock_class_key *key)
792 timer->entry.pprev = NULL;
793 timer->function = func;
794 if (WARN_ON_ONCE(flags & ~TIMER_INIT_FLAGS))
795 flags &= TIMER_INIT_FLAGS;
796 timer->flags = flags | raw_smp_processor_id();
797 lockdep_init_map(&timer->lockdep_map, name, key, 0);
801 * init_timer_key - initialize a timer
802 * @timer: the timer to be initialized
803 * @func: timer callback function
804 * @flags: timer flags
805 * @name: name of the timer
806 * @key: lockdep class key of the fake lock used for tracking timer
807 * sync lock dependencies
809 * init_timer_key() must be done to a timer prior calling *any* of the
810 * other timer functions.
812 void init_timer_key(struct timer_list *timer,
813 void (*func)(struct timer_list *), unsigned int flags,
814 const char *name, struct lock_class_key *key)
817 do_init_timer(timer, func, flags, name, key);
819 EXPORT_SYMBOL(init_timer_key);
821 static inline void detach_timer(struct timer_list *timer, bool clear_pending)
823 struct hlist_node *entry = &timer->entry;
825 debug_deactivate(timer);
830 entry->next = LIST_POISON2;
833 static int detach_if_pending(struct timer_list *timer, struct timer_base *base,
836 unsigned idx = timer_get_idx(timer);
838 if (!timer_pending(timer))
841 if (hlist_is_singular_node(&timer->entry, base->vectors + idx)) {
842 __clear_bit(idx, base->pending_map);
843 base->next_expiry_recalc = true;
846 detach_timer(timer, clear_pending);
850 static inline struct timer_base *get_timer_cpu_base(u32 tflags, u32 cpu)
852 struct timer_base *base = per_cpu_ptr(&timer_bases[BASE_STD], cpu);
855 * If the timer is deferrable and NO_HZ_COMMON is set then we need
856 * to use the deferrable base.
858 if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE))
859 base = per_cpu_ptr(&timer_bases[BASE_DEF], cpu);
863 static inline struct timer_base *get_timer_this_cpu_base(u32 tflags)
865 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
868 * If the timer is deferrable and NO_HZ_COMMON is set then we need
869 * to use the deferrable base.
871 if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE))
872 base = this_cpu_ptr(&timer_bases[BASE_DEF]);
876 static inline struct timer_base *get_timer_base(u32 tflags)
878 return get_timer_cpu_base(tflags, tflags & TIMER_CPUMASK);
881 static inline struct timer_base *
882 get_target_base(struct timer_base *base, unsigned tflags)
884 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
885 if (static_branch_likely(&timers_migration_enabled) &&
886 !(tflags & TIMER_PINNED))
887 return get_timer_cpu_base(tflags, get_nohz_timer_target());
889 return get_timer_this_cpu_base(tflags);
892 static inline void forward_timer_base(struct timer_base *base)
894 unsigned long jnow = READ_ONCE(jiffies);
897 * No need to forward if we are close enough below jiffies.
898 * Also while executing timers, base->clk is 1 offset ahead
899 * of jiffies to avoid endless requeuing to current jffies.
901 if ((long)(jnow - base->clk) < 1)
905 * If the next expiry value is > jiffies, then we fast forward to
906 * jiffies otherwise we forward to the next expiry value.
908 if (time_after(base->next_expiry, jnow)) {
911 if (WARN_ON_ONCE(time_before(base->next_expiry, base->clk)))
913 base->clk = base->next_expiry;
919 * We are using hashed locking: Holding per_cpu(timer_bases[x]).lock means
920 * that all timers which are tied to this base are locked, and the base itself
923 * So __run_timers/migrate_timers can safely modify all timers which could
924 * be found in the base->vectors array.
926 * When a timer is migrating then the TIMER_MIGRATING flag is set and we need
927 * to wait until the migration is done.
929 static struct timer_base *lock_timer_base(struct timer_list *timer,
930 unsigned long *flags)
931 __acquires(timer->base->lock)
934 struct timer_base *base;
938 * We need to use READ_ONCE() here, otherwise the compiler
939 * might re-read @tf between the check for TIMER_MIGRATING
942 tf = READ_ONCE(timer->flags);
944 if (!(tf & TIMER_MIGRATING)) {
945 base = get_timer_base(tf);
946 raw_spin_lock_irqsave(&base->lock, *flags);
947 if (timer->flags == tf)
949 raw_spin_unlock_irqrestore(&base->lock, *flags);
955 #define MOD_TIMER_PENDING_ONLY 0x01
956 #define MOD_TIMER_REDUCE 0x02
957 #define MOD_TIMER_NOTPENDING 0x04
960 __mod_timer(struct timer_list *timer, unsigned long expires, unsigned int options)
962 unsigned long clk = 0, flags, bucket_expiry;
963 struct timer_base *base, *new_base;
964 unsigned int idx = UINT_MAX;
967 BUG_ON(!timer->function);
970 * This is a common optimization triggered by the networking code - if
971 * the timer is re-modified to have the same timeout or ends up in the
972 * same array bucket then just return:
974 if (!(options & MOD_TIMER_NOTPENDING) && timer_pending(timer)) {
976 * The downside of this optimization is that it can result in
977 * larger granularity than you would get from adding a new
978 * timer with this expiry.
980 long diff = timer->expires - expires;
984 if (options & MOD_TIMER_REDUCE && diff <= 0)
988 * We lock timer base and calculate the bucket index right
989 * here. If the timer ends up in the same bucket, then we
990 * just update the expiry time and avoid the whole
991 * dequeue/enqueue dance.
993 base = lock_timer_base(timer, &flags);
994 forward_timer_base(base);
996 if (timer_pending(timer) && (options & MOD_TIMER_REDUCE) &&
997 time_before_eq(timer->expires, expires)) {
1003 idx = calc_wheel_index(expires, clk, &bucket_expiry);
1006 * Retrieve and compare the array index of the pending
1007 * timer. If it matches set the expiry to the new value so a
1008 * subsequent call will exit in the expires check above.
1010 if (idx == timer_get_idx(timer)) {
1011 if (!(options & MOD_TIMER_REDUCE))
1012 timer->expires = expires;
1013 else if (time_after(timer->expires, expires))
1014 timer->expires = expires;
1019 base = lock_timer_base(timer, &flags);
1020 forward_timer_base(base);
1023 ret = detach_if_pending(timer, base, false);
1024 if (!ret && (options & MOD_TIMER_PENDING_ONLY))
1027 new_base = get_target_base(base, timer->flags);
1029 if (base != new_base) {
1031 * We are trying to schedule the timer on the new base.
1032 * However we can't change timer's base while it is running,
1033 * otherwise timer_delete_sync() can't detect that the timer's
1034 * handler yet has not finished. This also guarantees that the
1035 * timer is serialized wrt itself.
1037 if (likely(base->running_timer != timer)) {
1038 /* See the comment in lock_timer_base() */
1039 timer->flags |= TIMER_MIGRATING;
1041 raw_spin_unlock(&base->lock);
1043 raw_spin_lock(&base->lock);
1044 WRITE_ONCE(timer->flags,
1045 (timer->flags & ~TIMER_BASEMASK) | base->cpu);
1046 forward_timer_base(base);
1050 debug_timer_activate(timer);
1052 timer->expires = expires;
1054 * If 'idx' was calculated above and the base time did not advance
1055 * between calculating 'idx' and possibly switching the base, only
1056 * enqueue_timer() is required. Otherwise we need to (re)calculate
1057 * the wheel index via internal_add_timer().
1059 if (idx != UINT_MAX && clk == base->clk)
1060 enqueue_timer(base, timer, idx, bucket_expiry);
1062 internal_add_timer(base, timer);
1065 raw_spin_unlock_irqrestore(&base->lock, flags);
1071 * mod_timer_pending - Modify a pending timer's timeout
1072 * @timer: The pending timer to be modified
1073 * @expires: New absolute timeout in jiffies
1075 * mod_timer_pending() is the same for pending timers as mod_timer(), but
1076 * will not activate inactive timers.
1079 * * %0 - The timer was inactive and not modified
1080 * * %1 - The timer was active and requeued to expire at @expires
1082 int mod_timer_pending(struct timer_list *timer, unsigned long expires)
1084 return __mod_timer(timer, expires, MOD_TIMER_PENDING_ONLY);
1086 EXPORT_SYMBOL(mod_timer_pending);
1089 * mod_timer - Modify a timer's timeout
1090 * @timer: The timer to be modified
1091 * @expires: New absolute timeout in jiffies
1093 * mod_timer(timer, expires) is equivalent to:
1095 * del_timer(timer); timer->expires = expires; add_timer(timer);
1097 * mod_timer() is more efficient than the above open coded sequence. In
1098 * case that the timer is inactive, the del_timer() part is a NOP. The
1099 * timer is in any case activated with the new expiry time @expires.
1101 * Note that if there are multiple unserialized concurrent users of the
1102 * same timer, then mod_timer() is the only safe way to modify the timeout,
1103 * since add_timer() cannot modify an already running timer.
1106 * * %0 - The timer was inactive and started
1107 * * %1 - The timer was active and requeued to expire at @expires or
1108 * the timer was active and not modified because @expires did
1109 * not change the effective expiry time
1111 int mod_timer(struct timer_list *timer, unsigned long expires)
1113 return __mod_timer(timer, expires, 0);
1115 EXPORT_SYMBOL(mod_timer);
1118 * timer_reduce - Modify a timer's timeout if it would reduce the timeout
1119 * @timer: The timer to be modified
1120 * @expires: New absolute timeout in jiffies
1122 * timer_reduce() is very similar to mod_timer(), except that it will only
1123 * modify an enqueued timer if that would reduce the expiration time. If
1124 * @timer is not enqueued it starts the timer.
1127 * * %0 - The timer was inactive and started
1128 * * %1 - The timer was active and requeued to expire at @expires or
1129 * the timer was active and not modified because @expires
1130 * did not change the effective expiry time such that the
1131 * timer would expire earlier than already scheduled
1133 int timer_reduce(struct timer_list *timer, unsigned long expires)
1135 return __mod_timer(timer, expires, MOD_TIMER_REDUCE);
1137 EXPORT_SYMBOL(timer_reduce);
1140 * add_timer - Start a timer
1141 * @timer: The timer to be started
1143 * Start @timer to expire at @timer->expires in the future. @timer->expires
1144 * is the absolute expiry time measured in 'jiffies'. When the timer expires
1145 * timer->function(timer) will be invoked from soft interrupt context.
1147 * The @timer->expires and @timer->function fields must be set prior
1148 * to calling this function.
1150 * If @timer->expires is already in the past @timer will be queued to
1151 * expire at the next timer tick.
1153 * This can only operate on an inactive timer. Attempts to invoke this on
1154 * an active timer are rejected with a warning.
1156 void add_timer(struct timer_list *timer)
1158 BUG_ON(timer_pending(timer));
1159 __mod_timer(timer, timer->expires, MOD_TIMER_NOTPENDING);
1161 EXPORT_SYMBOL(add_timer);
1164 * add_timer_on - Start a timer on a particular CPU
1165 * @timer: The timer to be started
1166 * @cpu: The CPU to start it on
1168 * Same as add_timer() except that it starts the timer on the given CPU.
1170 * See add_timer() for further details.
1172 void add_timer_on(struct timer_list *timer, int cpu)
1174 struct timer_base *new_base, *base;
1175 unsigned long flags;
1177 BUG_ON(timer_pending(timer) || !timer->function);
1179 new_base = get_timer_cpu_base(timer->flags, cpu);
1182 * If @timer was on a different CPU, it should be migrated with the
1183 * old base locked to prevent other operations proceeding with the
1184 * wrong base locked. See lock_timer_base().
1186 base = lock_timer_base(timer, &flags);
1187 if (base != new_base) {
1188 timer->flags |= TIMER_MIGRATING;
1190 raw_spin_unlock(&base->lock);
1192 raw_spin_lock(&base->lock);
1193 WRITE_ONCE(timer->flags,
1194 (timer->flags & ~TIMER_BASEMASK) | cpu);
1196 forward_timer_base(base);
1198 debug_timer_activate(timer);
1199 internal_add_timer(base, timer);
1200 raw_spin_unlock_irqrestore(&base->lock, flags);
1202 EXPORT_SYMBOL_GPL(add_timer_on);
1205 * del_timer - Deactivate a timer.
1206 * @timer: The timer to be deactivated
1208 * The function only deactivates a pending timer, but contrary to
1209 * timer_delete_sync() it does not take into account whether the timer's
1210 * callback function is concurrently executed on a different CPU or not.
1211 * It neither prevents rearming of the timer. If @timer can be rearmed
1212 * concurrently then the return value of this function is meaningless.
1215 * * %0 - The timer was not pending
1216 * * %1 - The timer was pending and deactivated
1218 int del_timer(struct timer_list *timer)
1220 struct timer_base *base;
1221 unsigned long flags;
1224 debug_assert_init(timer);
1226 if (timer_pending(timer)) {
1227 base = lock_timer_base(timer, &flags);
1228 ret = detach_if_pending(timer, base, true);
1229 raw_spin_unlock_irqrestore(&base->lock, flags);
1234 EXPORT_SYMBOL(del_timer);
1237 * try_to_del_timer_sync - Try to deactivate a timer
1238 * @timer: Timer to deactivate
1240 * This function tries to deactivate a timer. On success the timer is not
1241 * queued and the timer callback function is not running on any CPU.
1243 * This function does not guarantee that the timer cannot be rearmed right
1244 * after dropping the base lock. That needs to be prevented by the calling
1245 * code if necessary.
1248 * * %0 - The timer was not pending
1249 * * %1 - The timer was pending and deactivated
1250 * * %-1 - The timer callback function is running on a different CPU
1252 int try_to_del_timer_sync(struct timer_list *timer)
1254 struct timer_base *base;
1255 unsigned long flags;
1258 debug_assert_init(timer);
1260 base = lock_timer_base(timer, &flags);
1262 if (base->running_timer != timer)
1263 ret = detach_if_pending(timer, base, true);
1265 raw_spin_unlock_irqrestore(&base->lock, flags);
1269 EXPORT_SYMBOL(try_to_del_timer_sync);
1271 #ifdef CONFIG_PREEMPT_RT
1272 static __init void timer_base_init_expiry_lock(struct timer_base *base)
1274 spin_lock_init(&base->expiry_lock);
1277 static inline void timer_base_lock_expiry(struct timer_base *base)
1279 spin_lock(&base->expiry_lock);
1282 static inline void timer_base_unlock_expiry(struct timer_base *base)
1284 spin_unlock(&base->expiry_lock);
1288 * The counterpart to del_timer_wait_running().
1290 * If there is a waiter for base->expiry_lock, then it was waiting for the
1291 * timer callback to finish. Drop expiry_lock and reaquire it. That allows
1292 * the waiter to acquire the lock and make progress.
1294 static void timer_sync_wait_running(struct timer_base *base)
1296 if (atomic_read(&base->timer_waiters)) {
1297 raw_spin_unlock_irq(&base->lock);
1298 spin_unlock(&base->expiry_lock);
1299 spin_lock(&base->expiry_lock);
1300 raw_spin_lock_irq(&base->lock);
1305 * This function is called on PREEMPT_RT kernels when the fast path
1306 * deletion of a timer failed because the timer callback function was
1309 * This prevents priority inversion, if the softirq thread on a remote CPU
1310 * got preempted, and it prevents a life lock when the task which tries to
1311 * delete a timer preempted the softirq thread running the timer callback
1314 static void del_timer_wait_running(struct timer_list *timer)
1318 tf = READ_ONCE(timer->flags);
1319 if (!(tf & TIMER_MIGRATING)) {
1320 struct timer_base *base = get_timer_base(tf);
1323 * Mark the base as contended and grab the expiry lock,
1324 * which is held by the softirq across the timer
1325 * callback. Drop the lock immediately so the softirq can
1326 * expire the next timer. In theory the timer could already
1327 * be running again, but that's more than unlikely and just
1328 * causes another wait loop.
1330 atomic_inc(&base->timer_waiters);
1331 spin_lock_bh(&base->expiry_lock);
1332 atomic_dec(&base->timer_waiters);
1333 spin_unlock_bh(&base->expiry_lock);
1337 static inline void timer_base_init_expiry_lock(struct timer_base *base) { }
1338 static inline void timer_base_lock_expiry(struct timer_base *base) { }
1339 static inline void timer_base_unlock_expiry(struct timer_base *base) { }
1340 static inline void timer_sync_wait_running(struct timer_base *base) { }
1341 static inline void del_timer_wait_running(struct timer_list *timer) { }
1345 * timer_delete_sync - Deactivate a timer and wait for the handler to finish.
1346 * @timer: The timer to be deactivated
1348 * Synchronization rules: Callers must prevent restarting of the timer,
1349 * otherwise this function is meaningless. It must not be called from
1350 * interrupt contexts unless the timer is an irqsafe one. The caller must
1351 * not hold locks which would prevent completion of the timer's callback
1352 * function. The timer's handler must not call add_timer_on(). Upon exit
1353 * the timer is not queued and the handler is not running on any CPU.
1355 * For !irqsafe timers, the caller must not hold locks that are held in
1356 * interrupt context. Even if the lock has nothing to do with the timer in
1357 * question. Here's why::
1363 * base->running_timer = mytimer;
1364 * spin_lock_irq(somelock);
1366 * spin_lock(somelock);
1367 * timer_delete_sync(mytimer);
1368 * while (base->running_timer == mytimer);
1370 * Now timer_delete_sync() will never return and never release somelock.
1371 * The interrupt on the other CPU is waiting to grab somelock but it has
1372 * interrupted the softirq that CPU0 is waiting to finish.
1374 * This function cannot guarantee that the timer is not rearmed again by
1375 * some concurrent or preempting code, right after it dropped the base
1376 * lock. If there is the possibility of a concurrent rearm then the return
1377 * value of the function is meaningless.
1380 * * %0 - The timer was not pending
1381 * * %1 - The timer was pending and deactivated
1383 int timer_delete_sync(struct timer_list *timer)
1387 #ifdef CONFIG_LOCKDEP
1388 unsigned long flags;
1391 * If lockdep gives a backtrace here, please reference
1392 * the synchronization rules above.
1394 local_irq_save(flags);
1395 lock_map_acquire(&timer->lockdep_map);
1396 lock_map_release(&timer->lockdep_map);
1397 local_irq_restore(flags);
1400 * don't use it in hardirq context, because it
1401 * could lead to deadlock.
1403 WARN_ON(in_irq() && !(timer->flags & TIMER_IRQSAFE));
1406 ret = try_to_del_timer_sync(timer);
1408 if (unlikely(ret < 0)) {
1409 del_timer_wait_running(timer);
1416 EXPORT_SYMBOL(timer_delete_sync);
1418 static void call_timer_fn(struct timer_list *timer,
1419 void (*fn)(struct timer_list *),
1420 unsigned long baseclk)
1422 int count = preempt_count();
1424 #ifdef CONFIG_LOCKDEP
1426 * It is permissible to free the timer from inside the
1427 * function that is called from it, this we need to take into
1428 * account for lockdep too. To avoid bogus "held lock freed"
1429 * warnings as well as problems when looking into
1430 * timer->lockdep_map, make a copy and use that here.
1432 struct lockdep_map lockdep_map;
1434 lockdep_copy_map(&lockdep_map, &timer->lockdep_map);
1437 * Couple the lock chain with the lock chain at
1438 * timer_delete_sync() by acquiring the lock_map around the fn()
1439 * call here and in timer_delete_sync().
1441 lock_map_acquire(&lockdep_map);
1443 trace_timer_expire_entry(timer, baseclk);
1445 trace_timer_expire_exit(timer);
1447 lock_map_release(&lockdep_map);
1449 if (count != preempt_count()) {
1450 WARN_ONCE(1, "timer: %pS preempt leak: %08x -> %08x\n",
1451 fn, count, preempt_count());
1453 * Restore the preempt count. That gives us a decent
1454 * chance to survive and extract information. If the
1455 * callback kept a lock held, bad luck, but not worse
1456 * than the BUG() we had.
1458 preempt_count_set(count);
1462 static void expire_timers(struct timer_base *base, struct hlist_head *head)
1465 * This value is required only for tracing. base->clk was
1466 * incremented directly before expire_timers was called. But expiry
1467 * is related to the old base->clk value.
1469 unsigned long baseclk = base->clk - 1;
1471 while (!hlist_empty(head)) {
1472 struct timer_list *timer;
1473 void (*fn)(struct timer_list *);
1475 timer = hlist_entry(head->first, struct timer_list, entry);
1477 base->running_timer = timer;
1478 detach_timer(timer, true);
1480 fn = timer->function;
1482 if (timer->flags & TIMER_IRQSAFE) {
1483 raw_spin_unlock(&base->lock);
1484 call_timer_fn(timer, fn, baseclk);
1485 raw_spin_lock(&base->lock);
1486 base->running_timer = NULL;
1488 raw_spin_unlock_irq(&base->lock);
1489 call_timer_fn(timer, fn, baseclk);
1490 raw_spin_lock_irq(&base->lock);
1491 base->running_timer = NULL;
1492 timer_sync_wait_running(base);
1497 static int collect_expired_timers(struct timer_base *base,
1498 struct hlist_head *heads)
1500 unsigned long clk = base->clk = base->next_expiry;
1501 struct hlist_head *vec;
1505 for (i = 0; i < LVL_DEPTH; i++) {
1506 idx = (clk & LVL_MASK) + i * LVL_SIZE;
1508 if (__test_and_clear_bit(idx, base->pending_map)) {
1509 vec = base->vectors + idx;
1510 hlist_move_list(vec, heads++);
1513 /* Is it time to look at the next level? */
1514 if (clk & LVL_CLK_MASK)
1516 /* Shift clock for the next level granularity */
1517 clk >>= LVL_CLK_SHIFT;
1523 * Find the next pending bucket of a level. Search from level start (@offset)
1524 * + @clk upwards and if nothing there, search from start of the level
1525 * (@offset) up to @offset + clk.
1527 static int next_pending_bucket(struct timer_base *base, unsigned offset,
1530 unsigned pos, start = offset + clk;
1531 unsigned end = offset + LVL_SIZE;
1533 pos = find_next_bit(base->pending_map, end, start);
1537 pos = find_next_bit(base->pending_map, start, offset);
1538 return pos < start ? pos + LVL_SIZE - start : -1;
1542 * Search the first expiring timer in the various clock levels. Caller must
1545 static unsigned long __next_timer_interrupt(struct timer_base *base)
1547 unsigned long clk, next, adj;
1548 unsigned lvl, offset = 0;
1550 next = base->clk + NEXT_TIMER_MAX_DELTA;
1552 for (lvl = 0; lvl < LVL_DEPTH; lvl++, offset += LVL_SIZE) {
1553 int pos = next_pending_bucket(base, offset, clk & LVL_MASK);
1554 unsigned long lvl_clk = clk & LVL_CLK_MASK;
1557 unsigned long tmp = clk + (unsigned long) pos;
1559 tmp <<= LVL_SHIFT(lvl);
1560 if (time_before(tmp, next))
1564 * If the next expiration happens before we reach
1565 * the next level, no need to check further.
1567 if (pos <= ((LVL_CLK_DIV - lvl_clk) & LVL_CLK_MASK))
1571 * Clock for the next level. If the current level clock lower
1572 * bits are zero, we look at the next level as is. If not we
1573 * need to advance it by one because that's going to be the
1574 * next expiring bucket in that level. base->clk is the next
1575 * expiring jiffie. So in case of:
1577 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1580 * we have to look at all levels @index 0. With
1582 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1585 * LVL0 has the next expiring bucket @index 2. The upper
1586 * levels have the next expiring bucket @index 1.
1588 * In case that the propagation wraps the next level the same
1591 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1594 * So after looking at LVL0 we get:
1596 * LVL5 LVL4 LVL3 LVL2 LVL1
1599 * So no propagation from LVL1 to LVL2 because that happened
1600 * with the add already, but then we need to propagate further
1601 * from LVL2 to LVL3.
1603 * So the simple check whether the lower bits of the current
1604 * level are 0 or not is sufficient for all cases.
1606 adj = lvl_clk ? 1 : 0;
1607 clk >>= LVL_CLK_SHIFT;
1611 base->next_expiry_recalc = false;
1612 base->timers_pending = !(next == base->clk + NEXT_TIMER_MAX_DELTA);
1617 #ifdef CONFIG_NO_HZ_COMMON
1619 * Check, if the next hrtimer event is before the next timer wheel
1622 static u64 cmp_next_hrtimer_event(u64 basem, u64 expires)
1624 u64 nextevt = hrtimer_get_next_event();
1627 * If high resolution timers are enabled
1628 * hrtimer_get_next_event() returns KTIME_MAX.
1630 if (expires <= nextevt)
1634 * If the next timer is already expired, return the tick base
1635 * time so the tick is fired immediately.
1637 if (nextevt <= basem)
1641 * Round up to the next jiffie. High resolution timers are
1642 * off, so the hrtimers are expired in the tick and we need to
1643 * make sure that this tick really expires the timer to avoid
1644 * a ping pong of the nohz stop code.
1646 * Use DIV_ROUND_UP_ULL to prevent gcc calling __divdi3
1648 return DIV_ROUND_UP_ULL(nextevt, TICK_NSEC) * TICK_NSEC;
1652 * get_next_timer_interrupt - return the time (clock mono) of the next timer
1653 * @basej: base time jiffies
1654 * @basem: base time clock monotonic
1656 * Returns the tick aligned clock monotonic time of the next pending
1657 * timer or KTIME_MAX if no timer is pending.
1659 u64 get_next_timer_interrupt(unsigned long basej, u64 basem)
1661 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1662 u64 expires = KTIME_MAX;
1663 unsigned long nextevt;
1666 * Pretend that there is no timer pending if the cpu is offline.
1667 * Possible pending timers will be migrated later to an active cpu.
1669 if (cpu_is_offline(smp_processor_id()))
1672 raw_spin_lock(&base->lock);
1673 if (base->next_expiry_recalc)
1674 base->next_expiry = __next_timer_interrupt(base);
1675 nextevt = base->next_expiry;
1678 * We have a fresh next event. Check whether we can forward the
1679 * base. We can only do that when @basej is past base->clk
1680 * otherwise we might rewind base->clk.
1682 if (time_after(basej, base->clk)) {
1683 if (time_after(nextevt, basej))
1685 else if (time_after(nextevt, base->clk))
1686 base->clk = nextevt;
1689 if (time_before_eq(nextevt, basej)) {
1691 base->is_idle = false;
1693 if (base->timers_pending)
1694 expires = basem + (u64)(nextevt - basej) * TICK_NSEC;
1696 * If we expect to sleep more than a tick, mark the base idle.
1697 * Also the tick is stopped so any added timer must forward
1698 * the base clk itself to keep granularity small. This idle
1699 * logic is only maintained for the BASE_STD base, deferrable
1700 * timers may still see large granularity skew (by design).
1702 if ((expires - basem) > TICK_NSEC)
1703 base->is_idle = true;
1705 raw_spin_unlock(&base->lock);
1707 return cmp_next_hrtimer_event(basem, expires);
1711 * timer_clear_idle - Clear the idle state of the timer base
1713 * Called with interrupts disabled
1715 void timer_clear_idle(void)
1717 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1720 * We do this unlocked. The worst outcome is a remote enqueue sending
1721 * a pointless IPI, but taking the lock would just make the window for
1722 * sending the IPI a few instructions smaller for the cost of taking
1723 * the lock in the exit from idle path.
1725 base->is_idle = false;
1730 * Called from the timer interrupt handler to charge one tick to the current
1731 * process. user_tick is 1 if the tick is user time, 0 for system.
1733 void update_process_times(int user_tick)
1735 struct task_struct *p = current;
1737 PRANDOM_ADD_NOISE(jiffies, user_tick, p, 0);
1739 /* Note: this timer irq context must be accounted for as well. */
1740 account_process_tick(p, user_tick);
1742 rcu_sched_clock_irq(user_tick);
1743 #ifdef CONFIG_IRQ_WORK
1748 if (IS_ENABLED(CONFIG_POSIX_TIMERS))
1749 run_posix_cpu_timers();
1753 * __run_timers - run all expired timers (if any) on this CPU.
1754 * @base: the timer vector to be processed.
1756 static inline void __run_timers(struct timer_base *base)
1758 struct hlist_head heads[LVL_DEPTH];
1761 if (time_before(jiffies, base->next_expiry))
1764 timer_base_lock_expiry(base);
1765 raw_spin_lock_irq(&base->lock);
1767 while (time_after_eq(jiffies, base->clk) &&
1768 time_after_eq(jiffies, base->next_expiry)) {
1769 levels = collect_expired_timers(base, heads);
1771 * The two possible reasons for not finding any expired
1772 * timer at this clk are that all matching timers have been
1773 * dequeued or no timer has been queued since
1774 * base::next_expiry was set to base::clk +
1775 * NEXT_TIMER_MAX_DELTA.
1777 WARN_ON_ONCE(!levels && !base->next_expiry_recalc
1778 && base->timers_pending);
1780 base->next_expiry = __next_timer_interrupt(base);
1783 expire_timers(base, heads + levels);
1785 raw_spin_unlock_irq(&base->lock);
1786 timer_base_unlock_expiry(base);
1790 * This function runs timers and the timer-tq in bottom half context.
1792 static __latent_entropy void run_timer_softirq(struct softirq_action *h)
1794 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1797 if (IS_ENABLED(CONFIG_NO_HZ_COMMON))
1798 __run_timers(this_cpu_ptr(&timer_bases[BASE_DEF]));
1802 * Called by the local, per-CPU timer interrupt on SMP.
1804 void run_local_timers(void)
1806 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1808 hrtimer_run_queues();
1809 /* Raise the softirq only if required. */
1810 if (time_before(jiffies, base->next_expiry)) {
1811 if (!IS_ENABLED(CONFIG_NO_HZ_COMMON))
1813 /* CPU is awake, so check the deferrable base. */
1815 if (time_before(jiffies, base->next_expiry))
1818 raise_softirq(TIMER_SOFTIRQ);
1822 * Since schedule_timeout()'s timer is defined on the stack, it must store
1823 * the target task on the stack as well.
1825 struct process_timer {
1826 struct timer_list timer;
1827 struct task_struct *task;
1830 static void process_timeout(struct timer_list *t)
1832 struct process_timer *timeout = from_timer(timeout, t, timer);
1834 wake_up_process(timeout->task);
1838 * schedule_timeout - sleep until timeout
1839 * @timeout: timeout value in jiffies
1841 * Make the current task sleep until @timeout jiffies have elapsed.
1842 * The function behavior depends on the current task state
1843 * (see also set_current_state() description):
1845 * %TASK_RUNNING - the scheduler is called, but the task does not sleep
1846 * at all. That happens because sched_submit_work() does nothing for
1847 * tasks in %TASK_RUNNING state.
1849 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1850 * pass before the routine returns unless the current task is explicitly
1851 * woken up, (e.g. by wake_up_process()).
1853 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1854 * delivered to the current task or the current task is explicitly woken
1857 * The current task state is guaranteed to be %TASK_RUNNING when this
1860 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1861 * the CPU away without a bound on the timeout. In this case the return
1862 * value will be %MAX_SCHEDULE_TIMEOUT.
1864 * Returns 0 when the timer has expired otherwise the remaining time in
1865 * jiffies will be returned. In all cases the return value is guaranteed
1866 * to be non-negative.
1868 signed long __sched schedule_timeout(signed long timeout)
1870 struct process_timer timer;
1871 unsigned long expire;
1875 case MAX_SCHEDULE_TIMEOUT:
1877 * These two special cases are useful to be comfortable
1878 * in the caller. Nothing more. We could take
1879 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1880 * but I' d like to return a valid offset (>=0) to allow
1881 * the caller to do everything it want with the retval.
1887 * Another bit of PARANOID. Note that the retval will be
1888 * 0 since no piece of kernel is supposed to do a check
1889 * for a negative retval of schedule_timeout() (since it
1890 * should never happens anyway). You just have the printk()
1891 * that will tell you if something is gone wrong and where.
1894 printk(KERN_ERR "schedule_timeout: wrong timeout "
1895 "value %lx\n", timeout);
1897 current->state = TASK_RUNNING;
1902 expire = timeout + jiffies;
1904 timer.task = current;
1905 timer_setup_on_stack(&timer.timer, process_timeout, 0);
1906 __mod_timer(&timer.timer, expire, MOD_TIMER_NOTPENDING);
1908 del_singleshot_timer_sync(&timer.timer);
1910 /* Remove the timer from the object tracker */
1911 destroy_timer_on_stack(&timer.timer);
1913 timeout = expire - jiffies;
1916 return timeout < 0 ? 0 : timeout;
1918 EXPORT_SYMBOL(schedule_timeout);
1921 * We can use __set_current_state() here because schedule_timeout() calls
1922 * schedule() unconditionally.
1924 signed long __sched schedule_timeout_interruptible(signed long timeout)
1926 __set_current_state(TASK_INTERRUPTIBLE);
1927 return schedule_timeout(timeout);
1929 EXPORT_SYMBOL(schedule_timeout_interruptible);
1931 signed long __sched schedule_timeout_killable(signed long timeout)
1933 __set_current_state(TASK_KILLABLE);
1934 return schedule_timeout(timeout);
1936 EXPORT_SYMBOL(schedule_timeout_killable);
1938 signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1940 __set_current_state(TASK_UNINTERRUPTIBLE);
1941 return schedule_timeout(timeout);
1943 EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1946 * Like schedule_timeout_uninterruptible(), except this task will not contribute
1949 signed long __sched schedule_timeout_idle(signed long timeout)
1951 __set_current_state(TASK_IDLE);
1952 return schedule_timeout(timeout);
1954 EXPORT_SYMBOL(schedule_timeout_idle);
1956 #ifdef CONFIG_HOTPLUG_CPU
1957 static void migrate_timer_list(struct timer_base *new_base, struct hlist_head *head)
1959 struct timer_list *timer;
1960 int cpu = new_base->cpu;
1962 while (!hlist_empty(head)) {
1963 timer = hlist_entry(head->first, struct timer_list, entry);
1964 detach_timer(timer, false);
1965 timer->flags = (timer->flags & ~TIMER_BASEMASK) | cpu;
1966 internal_add_timer(new_base, timer);
1970 int timers_prepare_cpu(unsigned int cpu)
1972 struct timer_base *base;
1975 for (b = 0; b < NR_BASES; b++) {
1976 base = per_cpu_ptr(&timer_bases[b], cpu);
1977 base->clk = jiffies;
1978 base->next_expiry = base->clk + NEXT_TIMER_MAX_DELTA;
1979 base->timers_pending = false;
1980 base->is_idle = false;
1985 int timers_dead_cpu(unsigned int cpu)
1987 struct timer_base *old_base;
1988 struct timer_base *new_base;
1991 BUG_ON(cpu_online(cpu));
1993 for (b = 0; b < NR_BASES; b++) {
1994 old_base = per_cpu_ptr(&timer_bases[b], cpu);
1995 new_base = get_cpu_ptr(&timer_bases[b]);
1997 * The caller is globally serialized and nobody else
1998 * takes two locks at once, deadlock is not possible.
2000 raw_spin_lock_irq(&new_base->lock);
2001 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
2004 * The current CPUs base clock might be stale. Update it
2005 * before moving the timers over.
2007 forward_timer_base(new_base);
2009 BUG_ON(old_base->running_timer);
2011 for (i = 0; i < WHEEL_SIZE; i++)
2012 migrate_timer_list(new_base, old_base->vectors + i);
2014 raw_spin_unlock(&old_base->lock);
2015 raw_spin_unlock_irq(&new_base->lock);
2016 put_cpu_ptr(&timer_bases);
2021 #endif /* CONFIG_HOTPLUG_CPU */
2023 static void __init init_timer_cpu(int cpu)
2025 struct timer_base *base;
2028 for (i = 0; i < NR_BASES; i++) {
2029 base = per_cpu_ptr(&timer_bases[i], cpu);
2031 raw_spin_lock_init(&base->lock);
2032 base->clk = jiffies;
2033 base->next_expiry = base->clk + NEXT_TIMER_MAX_DELTA;
2034 timer_base_init_expiry_lock(base);
2038 static void __init init_timer_cpus(void)
2042 for_each_possible_cpu(cpu)
2043 init_timer_cpu(cpu);
2046 void __init init_timers(void)
2049 posix_cputimers_init_work();
2050 open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
2054 * msleep - sleep safely even with waitqueue interruptions
2055 * @msecs: Time in milliseconds to sleep for
2057 void msleep(unsigned int msecs)
2059 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
2062 timeout = schedule_timeout_uninterruptible(timeout);
2065 EXPORT_SYMBOL(msleep);
2068 * msleep_interruptible - sleep waiting for signals
2069 * @msecs: Time in milliseconds to sleep for
2071 unsigned long msleep_interruptible(unsigned int msecs)
2073 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
2075 while (timeout && !signal_pending(current))
2076 timeout = schedule_timeout_interruptible(timeout);
2077 return jiffies_to_msecs(timeout);
2080 EXPORT_SYMBOL(msleep_interruptible);
2083 * usleep_range - Sleep for an approximate time
2084 * @min: Minimum time in usecs to sleep
2085 * @max: Maximum time in usecs to sleep
2087 * In non-atomic context where the exact wakeup time is flexible, use
2088 * usleep_range() instead of udelay(). The sleep improves responsiveness
2089 * by avoiding the CPU-hogging busy-wait of udelay(), and the range reduces
2090 * power usage by allowing hrtimers to take advantage of an already-
2091 * scheduled interrupt instead of scheduling a new one just for this sleep.
2093 void __sched usleep_range(unsigned long min, unsigned long max)
2095 ktime_t exp = ktime_add_us(ktime_get(), min);
2096 u64 delta = (u64)(max - min) * NSEC_PER_USEC;
2099 __set_current_state(TASK_UNINTERRUPTIBLE);
2100 /* Do not return before the requested sleep time has elapsed */
2101 if (!schedule_hrtimeout_range(&exp, delta, HRTIMER_MODE_ABS))
2105 EXPORT_SYMBOL(usleep_range);