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>
47 #include <linux/uaccess.h>
48 #include <asm/unistd.h>
49 #include <asm/div64.h>
50 #include <asm/timex.h>
53 #include "tick-internal.h"
55 #define CREATE_TRACE_POINTS
56 #include <trace/events/timer.h>
58 __visible u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
60 EXPORT_SYMBOL(jiffies_64);
63 * The timer wheel has LVL_DEPTH array levels. Each level provides an array of
64 * LVL_SIZE buckets. Each level is driven by its own clock and therefor each
65 * level has a different granularity.
67 * The level granularity is: LVL_CLK_DIV ^ lvl
68 * The level clock frequency is: HZ / (LVL_CLK_DIV ^ level)
70 * The array level of a newly armed timer depends on the relative expiry
71 * time. The farther the expiry time is away the higher the array level and
72 * therefor the granularity becomes.
74 * Contrary to the original timer wheel implementation, which aims for 'exact'
75 * expiry of the timers, this implementation removes the need for recascading
76 * the timers into the lower array levels. The previous 'classic' timer wheel
77 * implementation of the kernel already violated the 'exact' expiry by adding
78 * slack to the expiry time to provide batched expiration. The granularity
79 * levels provide implicit batching.
81 * This is an optimization of the original timer wheel implementation for the
82 * majority of the timer wheel use cases: timeouts. The vast majority of
83 * timeout timers (networking, disk I/O ...) are canceled before expiry. If
84 * the timeout expires it indicates that normal operation is disturbed, so it
85 * does not matter much whether the timeout comes with a slight delay.
87 * The only exception to this are networking timers with a small expiry
88 * time. They rely on the granularity. Those fit into the first wheel level,
89 * which has HZ granularity.
91 * We don't have cascading anymore. timers with a expiry time above the
92 * capacity of the last wheel level are force expired at the maximum timeout
93 * value of the last wheel level. From data sampling we know that the maximum
94 * value observed is 5 days (network connection tracking), so this should not
97 * The currently chosen array constants values are a good compromise between
98 * array size and granularity.
100 * This results in the following granularity and range levels:
103 * Level Offset Granularity Range
104 * 0 0 1 ms 0 ms - 63 ms
105 * 1 64 8 ms 64 ms - 511 ms
106 * 2 128 64 ms 512 ms - 4095 ms (512ms - ~4s)
107 * 3 192 512 ms 4096 ms - 32767 ms (~4s - ~32s)
108 * 4 256 4096 ms (~4s) 32768 ms - 262143 ms (~32s - ~4m)
109 * 5 320 32768 ms (~32s) 262144 ms - 2097151 ms (~4m - ~34m)
110 * 6 384 262144 ms (~4m) 2097152 ms - 16777215 ms (~34m - ~4h)
111 * 7 448 2097152 ms (~34m) 16777216 ms - 134217727 ms (~4h - ~1d)
112 * 8 512 16777216 ms (~4h) 134217728 ms - 1073741822 ms (~1d - ~12d)
115 * Level Offset Granularity Range
116 * 0 0 3 ms 0 ms - 210 ms
117 * 1 64 26 ms 213 ms - 1703 ms (213ms - ~1s)
118 * 2 128 213 ms 1706 ms - 13650 ms (~1s - ~13s)
119 * 3 192 1706 ms (~1s) 13653 ms - 109223 ms (~13s - ~1m)
120 * 4 256 13653 ms (~13s) 109226 ms - 873810 ms (~1m - ~14m)
121 * 5 320 109226 ms (~1m) 873813 ms - 6990503 ms (~14m - ~1h)
122 * 6 384 873813 ms (~14m) 6990506 ms - 55924050 ms (~1h - ~15h)
123 * 7 448 6990506 ms (~1h) 55924053 ms - 447392423 ms (~15h - ~5d)
124 * 8 512 55924053 ms (~15h) 447392426 ms - 3579139406 ms (~5d - ~41d)
127 * Level Offset Granularity Range
128 * 0 0 4 ms 0 ms - 255 ms
129 * 1 64 32 ms 256 ms - 2047 ms (256ms - ~2s)
130 * 2 128 256 ms 2048 ms - 16383 ms (~2s - ~16s)
131 * 3 192 2048 ms (~2s) 16384 ms - 131071 ms (~16s - ~2m)
132 * 4 256 16384 ms (~16s) 131072 ms - 1048575 ms (~2m - ~17m)
133 * 5 320 131072 ms (~2m) 1048576 ms - 8388607 ms (~17m - ~2h)
134 * 6 384 1048576 ms (~17m) 8388608 ms - 67108863 ms (~2h - ~18h)
135 * 7 448 8388608 ms (~2h) 67108864 ms - 536870911 ms (~18h - ~6d)
136 * 8 512 67108864 ms (~18h) 536870912 ms - 4294967288 ms (~6d - ~49d)
139 * Level Offset Granularity Range
140 * 0 0 10 ms 0 ms - 630 ms
141 * 1 64 80 ms 640 ms - 5110 ms (640ms - ~5s)
142 * 2 128 640 ms 5120 ms - 40950 ms (~5s - ~40s)
143 * 3 192 5120 ms (~5s) 40960 ms - 327670 ms (~40s - ~5m)
144 * 4 256 40960 ms (~40s) 327680 ms - 2621430 ms (~5m - ~43m)
145 * 5 320 327680 ms (~5m) 2621440 ms - 20971510 ms (~43m - ~5h)
146 * 6 384 2621440 ms (~43m) 20971520 ms - 167772150 ms (~5h - ~1d)
147 * 7 448 20971520 ms (~5h) 167772160 ms - 1342177270 ms (~1d - ~15d)
150 /* Clock divisor for the next level */
151 #define LVL_CLK_SHIFT 3
152 #define LVL_CLK_DIV (1UL << LVL_CLK_SHIFT)
153 #define LVL_CLK_MASK (LVL_CLK_DIV - 1)
154 #define LVL_SHIFT(n) ((n) * LVL_CLK_SHIFT)
155 #define LVL_GRAN(n) (1UL << LVL_SHIFT(n))
158 * The time start value for each level to select the bucket at enqueue
161 #define LVL_START(n) ((LVL_SIZE - 1) << (((n) - 1) * LVL_CLK_SHIFT))
163 /* Size of each clock level */
165 #define LVL_SIZE (1UL << LVL_BITS)
166 #define LVL_MASK (LVL_SIZE - 1)
167 #define LVL_OFFS(n) ((n) * LVL_SIZE)
176 /* The cutoff (max. capacity of the wheel) */
177 #define WHEEL_TIMEOUT_CUTOFF (LVL_START(LVL_DEPTH))
178 #define WHEEL_TIMEOUT_MAX (WHEEL_TIMEOUT_CUTOFF - LVL_GRAN(LVL_DEPTH - 1))
181 * The resulting wheel size. If NOHZ is configured we allocate two
182 * wheels so we have a separate storage for the deferrable timers.
184 #define WHEEL_SIZE (LVL_SIZE * LVL_DEPTH)
186 #ifdef CONFIG_NO_HZ_COMMON
198 struct timer_list *running_timer;
199 #ifdef CONFIG_PREEMPT_RT
200 spinlock_t expiry_lock;
201 atomic_t timer_waiters;
204 unsigned long next_expiry;
207 bool must_forward_clk;
208 DECLARE_BITMAP(pending_map, WHEEL_SIZE);
209 struct hlist_head vectors[WHEEL_SIZE];
210 } ____cacheline_aligned;
212 static DEFINE_PER_CPU(struct timer_base, timer_bases[NR_BASES]);
214 #ifdef CONFIG_NO_HZ_COMMON
216 static DEFINE_STATIC_KEY_FALSE(timers_nohz_active);
217 static DEFINE_MUTEX(timer_keys_mutex);
219 static void timer_update_keys(struct work_struct *work);
220 static DECLARE_WORK(timer_update_work, timer_update_keys);
223 unsigned int sysctl_timer_migration = 1;
225 DEFINE_STATIC_KEY_FALSE(timers_migration_enabled);
227 static void timers_update_migration(void)
229 if (sysctl_timer_migration && tick_nohz_active)
230 static_branch_enable(&timers_migration_enabled);
232 static_branch_disable(&timers_migration_enabled);
235 static inline void timers_update_migration(void) { }
236 #endif /* !CONFIG_SMP */
238 static void timer_update_keys(struct work_struct *work)
240 mutex_lock(&timer_keys_mutex);
241 timers_update_migration();
242 static_branch_enable(&timers_nohz_active);
243 mutex_unlock(&timer_keys_mutex);
246 void timers_update_nohz(void)
248 schedule_work(&timer_update_work);
251 int timer_migration_handler(struct ctl_table *table, int write,
252 void __user *buffer, size_t *lenp,
257 mutex_lock(&timer_keys_mutex);
258 ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
260 timers_update_migration();
261 mutex_unlock(&timer_keys_mutex);
265 static inline bool is_timers_nohz_active(void)
267 return static_branch_unlikely(&timers_nohz_active);
270 static inline bool is_timers_nohz_active(void) { return false; }
271 #endif /* NO_HZ_COMMON */
273 static unsigned long round_jiffies_common(unsigned long j, int cpu,
277 unsigned long original = j;
280 * We don't want all cpus firing their timers at once hitting the
281 * same lock or cachelines, so we skew each extra cpu with an extra
282 * 3 jiffies. This 3 jiffies came originally from the mm/ code which
284 * The skew is done by adding 3*cpunr, then round, then subtract this
285 * extra offset again.
292 * If the target jiffie is just after a whole second (which can happen
293 * due to delays of the timer irq, long irq off times etc etc) then
294 * we should round down to the whole second, not up. Use 1/4th second
295 * as cutoff for this rounding as an extreme upper bound for this.
296 * But never round down if @force_up is set.
298 if (rem < HZ/4 && !force_up) /* round down */
303 /* now that we have rounded, subtract the extra skew again */
307 * Make sure j is still in the future. Otherwise return the
310 return time_is_after_jiffies(j) ? j : original;
314 * __round_jiffies - function to round jiffies to a full second
315 * @j: the time in (absolute) jiffies that should be rounded
316 * @cpu: the processor number on which the timeout will happen
318 * __round_jiffies() rounds an absolute time in the future (in jiffies)
319 * up or down to (approximately) full seconds. This is useful for timers
320 * for which the exact time they fire does not matter too much, as long as
321 * they fire approximately every X seconds.
323 * By rounding these timers to whole seconds, all such timers will fire
324 * at the same time, rather than at various times spread out. The goal
325 * of this is to have the CPU wake up less, which saves power.
327 * The exact rounding is skewed for each processor to avoid all
328 * processors firing at the exact same time, which could lead
329 * to lock contention or spurious cache line bouncing.
331 * The return value is the rounded version of the @j parameter.
333 unsigned long __round_jiffies(unsigned long j, int cpu)
335 return round_jiffies_common(j, cpu, false);
337 EXPORT_SYMBOL_GPL(__round_jiffies);
340 * __round_jiffies_relative - function to round jiffies to a full second
341 * @j: the time in (relative) jiffies that should be rounded
342 * @cpu: the processor number on which the timeout will happen
344 * __round_jiffies_relative() rounds a time delta in the future (in jiffies)
345 * up or down to (approximately) full seconds. This is useful for timers
346 * for which the exact time they fire does not matter too much, as long as
347 * they fire approximately every X seconds.
349 * By rounding these timers to whole seconds, all such timers will fire
350 * at the same time, rather than at various times spread out. The goal
351 * of this is to have the CPU wake up less, which saves power.
353 * The exact rounding is skewed for each processor to avoid all
354 * processors firing at the exact same time, which could lead
355 * to lock contention or spurious cache line bouncing.
357 * The return value is the rounded version of the @j parameter.
359 unsigned long __round_jiffies_relative(unsigned long j, int cpu)
361 unsigned long j0 = jiffies;
363 /* Use j0 because jiffies might change while we run */
364 return round_jiffies_common(j + j0, cpu, false) - j0;
366 EXPORT_SYMBOL_GPL(__round_jiffies_relative);
369 * round_jiffies - function to round jiffies to a full second
370 * @j: the time in (absolute) jiffies that should be rounded
372 * round_jiffies() rounds an absolute time in the future (in jiffies)
373 * up or down to (approximately) full seconds. This is useful for timers
374 * for which the exact time they fire does not matter too much, as long as
375 * they fire approximately every X seconds.
377 * By rounding these timers to whole seconds, all such timers will fire
378 * at the same time, rather than at various times spread out. The goal
379 * of this is to have the CPU wake up less, which saves power.
381 * The return value is the rounded version of the @j parameter.
383 unsigned long round_jiffies(unsigned long j)
385 return round_jiffies_common(j, raw_smp_processor_id(), false);
387 EXPORT_SYMBOL_GPL(round_jiffies);
390 * round_jiffies_relative - function to round jiffies to a full second
391 * @j: the time in (relative) jiffies that should be rounded
393 * round_jiffies_relative() rounds a time delta in the future (in jiffies)
394 * up or down to (approximately) full seconds. This is useful for timers
395 * for which the exact time they fire does not matter too much, as long as
396 * they fire approximately every X seconds.
398 * By rounding these timers to whole seconds, all such timers will fire
399 * at the same time, rather than at various times spread out. The goal
400 * of this is to have the CPU wake up less, which saves power.
402 * The return value is the rounded version of the @j parameter.
404 unsigned long round_jiffies_relative(unsigned long j)
406 return __round_jiffies_relative(j, raw_smp_processor_id());
408 EXPORT_SYMBOL_GPL(round_jiffies_relative);
411 * __round_jiffies_up - function to round jiffies up to a full second
412 * @j: the time in (absolute) jiffies that should be rounded
413 * @cpu: the processor number on which the timeout will happen
415 * This is the same as __round_jiffies() except that it will never
416 * round down. This is useful for timeouts for which the exact time
417 * of firing does not matter too much, as long as they don't fire too
420 unsigned long __round_jiffies_up(unsigned long j, int cpu)
422 return round_jiffies_common(j, cpu, true);
424 EXPORT_SYMBOL_GPL(__round_jiffies_up);
427 * __round_jiffies_up_relative - function to round jiffies up to a full second
428 * @j: the time in (relative) jiffies that should be rounded
429 * @cpu: the processor number on which the timeout will happen
431 * This is the same as __round_jiffies_relative() except that it will never
432 * round down. This is useful for timeouts for which the exact time
433 * of firing does not matter too much, as long as they don't fire too
436 unsigned long __round_jiffies_up_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, true) - j0;
443 EXPORT_SYMBOL_GPL(__round_jiffies_up_relative);
446 * round_jiffies_up - function to round jiffies up to a full second
447 * @j: the time in (absolute) jiffies that should be rounded
449 * This is the same as round_jiffies() except that it will never
450 * round down. This is useful for timeouts for which the exact time
451 * of firing does not matter too much, as long as they don't fire too
454 unsigned long round_jiffies_up(unsigned long j)
456 return round_jiffies_common(j, raw_smp_processor_id(), true);
458 EXPORT_SYMBOL_GPL(round_jiffies_up);
461 * round_jiffies_up_relative - function to round jiffies up to a full second
462 * @j: the time in (relative) jiffies that should be rounded
464 * This is the same as round_jiffies_relative() except that it will never
465 * round down. This is useful for timeouts for which the exact time
466 * of firing does not matter too much, as long as they don't fire too
469 unsigned long round_jiffies_up_relative(unsigned long j)
471 return __round_jiffies_up_relative(j, raw_smp_processor_id());
473 EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
476 static inline unsigned int timer_get_idx(struct timer_list *timer)
478 return (timer->flags & TIMER_ARRAYMASK) >> TIMER_ARRAYSHIFT;
481 static inline void timer_set_idx(struct timer_list *timer, unsigned int idx)
483 timer->flags = (timer->flags & ~TIMER_ARRAYMASK) |
484 idx << TIMER_ARRAYSHIFT;
488 * Helper function to calculate the array index for a given expiry
491 static inline unsigned calc_index(unsigned expires, unsigned lvl)
493 expires = (expires + LVL_GRAN(lvl)) >> LVL_SHIFT(lvl);
494 return LVL_OFFS(lvl) + (expires & LVL_MASK);
497 static int calc_wheel_index(unsigned long expires, unsigned long clk)
499 unsigned long delta = expires - clk;
502 if (delta < LVL_START(1)) {
503 idx = calc_index(expires, 0);
504 } else if (delta < LVL_START(2)) {
505 idx = calc_index(expires, 1);
506 } else if (delta < LVL_START(3)) {
507 idx = calc_index(expires, 2);
508 } else if (delta < LVL_START(4)) {
509 idx = calc_index(expires, 3);
510 } else if (delta < LVL_START(5)) {
511 idx = calc_index(expires, 4);
512 } else if (delta < LVL_START(6)) {
513 idx = calc_index(expires, 5);
514 } else if (delta < LVL_START(7)) {
515 idx = calc_index(expires, 6);
516 } else if (LVL_DEPTH > 8 && delta < LVL_START(8)) {
517 idx = calc_index(expires, 7);
518 } else if ((long) delta < 0) {
519 idx = clk & LVL_MASK;
522 * Force expire obscene large timeouts to expire at the
523 * capacity limit of the wheel.
525 if (expires >= WHEEL_TIMEOUT_CUTOFF)
526 expires = WHEEL_TIMEOUT_MAX;
528 idx = calc_index(expires, LVL_DEPTH - 1);
534 * Enqueue the timer into the hash bucket, mark it pending in
535 * the bitmap and store the index in the timer flags.
537 static void enqueue_timer(struct timer_base *base, struct timer_list *timer,
540 hlist_add_head(&timer->entry, base->vectors + idx);
541 __set_bit(idx, base->pending_map);
542 timer_set_idx(timer, idx);
544 trace_timer_start(timer, timer->expires, timer->flags);
548 __internal_add_timer(struct timer_base *base, struct timer_list *timer)
552 idx = calc_wheel_index(timer->expires, base->clk);
553 enqueue_timer(base, timer, idx);
557 trigger_dyntick_cpu(struct timer_base *base, struct timer_list *timer)
559 if (!is_timers_nohz_active())
563 * TODO: This wants some optimizing similar to the code below, but we
564 * will do that when we switch from push to pull for deferrable timers.
566 if (timer->flags & TIMER_DEFERRABLE) {
567 if (tick_nohz_full_cpu(base->cpu))
568 wake_up_nohz_cpu(base->cpu);
573 * We might have to IPI the remote CPU if the base is idle and the
574 * timer is not deferrable. If the other CPU is on the way to idle
575 * then it can't set base->is_idle as we hold the base lock:
580 /* Check whether this is the new first expiring timer: */
581 if (time_after_eq(timer->expires, base->next_expiry))
585 * Set the next expiry time and kick the CPU so it can reevaluate the
588 base->next_expiry = timer->expires;
589 wake_up_nohz_cpu(base->cpu);
593 internal_add_timer(struct timer_base *base, struct timer_list *timer)
595 __internal_add_timer(base, timer);
596 trigger_dyntick_cpu(base, timer);
599 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
601 static struct debug_obj_descr timer_debug_descr;
603 static void *timer_debug_hint(void *addr)
605 return ((struct timer_list *) addr)->function;
608 static bool timer_is_static_object(void *addr)
610 struct timer_list *timer = addr;
612 return (timer->entry.pprev == NULL &&
613 timer->entry.next == TIMER_ENTRY_STATIC);
617 * fixup_init is called when:
618 * - an active object is initialized
620 static bool timer_fixup_init(void *addr, enum debug_obj_state state)
622 struct timer_list *timer = addr;
625 case ODEBUG_STATE_ACTIVE:
626 del_timer_sync(timer);
627 debug_object_init(timer, &timer_debug_descr);
634 /* Stub timer callback for improperly used timers. */
635 static void stub_timer(struct timer_list *unused)
641 * fixup_activate is called when:
642 * - an active object is activated
643 * - an unknown non-static object is activated
645 static bool timer_fixup_activate(void *addr, enum debug_obj_state state)
647 struct timer_list *timer = addr;
650 case ODEBUG_STATE_NOTAVAILABLE:
651 timer_setup(timer, stub_timer, 0);
654 case ODEBUG_STATE_ACTIVE:
663 * fixup_free is called when:
664 * - an active object is freed
666 static bool timer_fixup_free(void *addr, enum debug_obj_state state)
668 struct timer_list *timer = addr;
671 case ODEBUG_STATE_ACTIVE:
672 del_timer_sync(timer);
673 debug_object_free(timer, &timer_debug_descr);
681 * fixup_assert_init is called when:
682 * - an untracked/uninit-ed object is found
684 static bool timer_fixup_assert_init(void *addr, enum debug_obj_state state)
686 struct timer_list *timer = addr;
689 case ODEBUG_STATE_NOTAVAILABLE:
690 timer_setup(timer, stub_timer, 0);
697 static struct debug_obj_descr timer_debug_descr = {
698 .name = "timer_list",
699 .debug_hint = timer_debug_hint,
700 .is_static_object = timer_is_static_object,
701 .fixup_init = timer_fixup_init,
702 .fixup_activate = timer_fixup_activate,
703 .fixup_free = timer_fixup_free,
704 .fixup_assert_init = timer_fixup_assert_init,
707 static inline void debug_timer_init(struct timer_list *timer)
709 debug_object_init(timer, &timer_debug_descr);
712 static inline void debug_timer_activate(struct timer_list *timer)
714 debug_object_activate(timer, &timer_debug_descr);
717 static inline void debug_timer_deactivate(struct timer_list *timer)
719 debug_object_deactivate(timer, &timer_debug_descr);
722 static inline void debug_timer_free(struct timer_list *timer)
724 debug_object_free(timer, &timer_debug_descr);
727 static inline void debug_timer_assert_init(struct timer_list *timer)
729 debug_object_assert_init(timer, &timer_debug_descr);
732 static void do_init_timer(struct timer_list *timer,
733 void (*func)(struct timer_list *),
735 const char *name, struct lock_class_key *key);
737 void init_timer_on_stack_key(struct timer_list *timer,
738 void (*func)(struct timer_list *),
740 const char *name, struct lock_class_key *key)
742 debug_object_init_on_stack(timer, &timer_debug_descr);
743 do_init_timer(timer, func, flags, name, key);
745 EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
747 void destroy_timer_on_stack(struct timer_list *timer)
749 debug_object_free(timer, &timer_debug_descr);
751 EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
754 static inline void debug_timer_init(struct timer_list *timer) { }
755 static inline void debug_timer_activate(struct timer_list *timer) { }
756 static inline void debug_timer_deactivate(struct timer_list *timer) { }
757 static inline void debug_timer_assert_init(struct timer_list *timer) { }
760 static inline void debug_init(struct timer_list *timer)
762 debug_timer_init(timer);
763 trace_timer_init(timer);
766 static inline void debug_deactivate(struct timer_list *timer)
768 debug_timer_deactivate(timer);
769 trace_timer_cancel(timer);
772 static inline void debug_assert_init(struct timer_list *timer)
774 debug_timer_assert_init(timer);
777 static void do_init_timer(struct timer_list *timer,
778 void (*func)(struct timer_list *),
780 const char *name, struct lock_class_key *key)
782 timer->entry.pprev = NULL;
783 timer->function = func;
784 timer->flags = flags | raw_smp_processor_id();
785 lockdep_init_map(&timer->lockdep_map, name, key, 0);
789 * init_timer_key - initialize a timer
790 * @timer: the timer to be initialized
791 * @func: timer callback function
792 * @flags: timer flags
793 * @name: name of the timer
794 * @key: lockdep class key of the fake lock used for tracking timer
795 * sync lock dependencies
797 * init_timer_key() must be done to a timer prior calling *any* of the
798 * other timer functions.
800 void init_timer_key(struct timer_list *timer,
801 void (*func)(struct timer_list *), unsigned int flags,
802 const char *name, struct lock_class_key *key)
805 do_init_timer(timer, func, flags, name, key);
807 EXPORT_SYMBOL(init_timer_key);
809 static inline void detach_timer(struct timer_list *timer, bool clear_pending)
811 struct hlist_node *entry = &timer->entry;
813 debug_deactivate(timer);
818 entry->next = LIST_POISON2;
821 static int detach_if_pending(struct timer_list *timer, struct timer_base *base,
824 unsigned idx = timer_get_idx(timer);
826 if (!timer_pending(timer))
829 if (hlist_is_singular_node(&timer->entry, base->vectors + idx))
830 __clear_bit(idx, base->pending_map);
832 detach_timer(timer, clear_pending);
836 static inline struct timer_base *get_timer_cpu_base(u32 tflags, u32 cpu)
838 struct timer_base *base = per_cpu_ptr(&timer_bases[BASE_STD], cpu);
841 * If the timer is deferrable and NO_HZ_COMMON is set then we need
842 * to use the deferrable base.
844 if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE))
845 base = per_cpu_ptr(&timer_bases[BASE_DEF], cpu);
849 static inline struct timer_base *get_timer_this_cpu_base(u32 tflags)
851 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
854 * If the timer is deferrable and NO_HZ_COMMON is set then we need
855 * to use the deferrable base.
857 if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE))
858 base = this_cpu_ptr(&timer_bases[BASE_DEF]);
862 static inline struct timer_base *get_timer_base(u32 tflags)
864 return get_timer_cpu_base(tflags, tflags & TIMER_CPUMASK);
867 static inline struct timer_base *
868 get_target_base(struct timer_base *base, unsigned tflags)
870 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
871 if (static_branch_likely(&timers_migration_enabled) &&
872 !(tflags & TIMER_PINNED))
873 return get_timer_cpu_base(tflags, get_nohz_timer_target());
875 return get_timer_this_cpu_base(tflags);
878 static inline void forward_timer_base(struct timer_base *base)
880 #ifdef CONFIG_NO_HZ_COMMON
884 * We only forward the base when we are idle or have just come out of
885 * idle (must_forward_clk logic), and have a delta between base clock
886 * and jiffies. In the common case, run_timers will take care of it.
888 if (likely(!base->must_forward_clk))
891 jnow = READ_ONCE(jiffies);
892 base->must_forward_clk = base->is_idle;
893 if ((long)(jnow - base->clk) < 2)
897 * If the next expiry value is > jiffies, then we fast forward to
898 * jiffies otherwise we forward to the next expiry value.
900 if (time_after(base->next_expiry, jnow))
903 base->clk = base->next_expiry;
909 * We are using hashed locking: Holding per_cpu(timer_bases[x]).lock means
910 * that all timers which are tied to this base are locked, and the base itself
913 * So __run_timers/migrate_timers can safely modify all timers which could
914 * be found in the base->vectors array.
916 * When a timer is migrating then the TIMER_MIGRATING flag is set and we need
917 * to wait until the migration is done.
919 static struct timer_base *lock_timer_base(struct timer_list *timer,
920 unsigned long *flags)
921 __acquires(timer->base->lock)
924 struct timer_base *base;
928 * We need to use READ_ONCE() here, otherwise the compiler
929 * might re-read @tf between the check for TIMER_MIGRATING
932 tf = READ_ONCE(timer->flags);
934 if (!(tf & TIMER_MIGRATING)) {
935 base = get_timer_base(tf);
936 raw_spin_lock_irqsave(&base->lock, *flags);
937 if (timer->flags == tf)
939 raw_spin_unlock_irqrestore(&base->lock, *flags);
945 #define MOD_TIMER_PENDING_ONLY 0x01
946 #define MOD_TIMER_REDUCE 0x02
947 #define MOD_TIMER_NOTPENDING 0x04
950 __mod_timer(struct timer_list *timer, unsigned long expires, unsigned int options)
952 struct timer_base *base, *new_base;
953 unsigned int idx = UINT_MAX;
954 unsigned long clk = 0, flags;
957 BUG_ON(!timer->function);
960 * This is a common optimization triggered by the networking code - if
961 * the timer is re-modified to have the same timeout or ends up in the
962 * same array bucket then just return:
964 if (!(options & MOD_TIMER_NOTPENDING) && timer_pending(timer)) {
966 * The downside of this optimization is that it can result in
967 * larger granularity than you would get from adding a new
968 * timer with this expiry.
970 long diff = timer->expires - expires;
974 if (options & MOD_TIMER_REDUCE && diff <= 0)
978 * We lock timer base and calculate the bucket index right
979 * here. If the timer ends up in the same bucket, then we
980 * just update the expiry time and avoid the whole
981 * dequeue/enqueue dance.
983 base = lock_timer_base(timer, &flags);
984 forward_timer_base(base);
986 if (timer_pending(timer) && (options & MOD_TIMER_REDUCE) &&
987 time_before_eq(timer->expires, expires)) {
993 idx = calc_wheel_index(expires, clk);
996 * Retrieve and compare the array index of the pending
997 * timer. If it matches set the expiry to the new value so a
998 * subsequent call will exit in the expires check above.
1000 if (idx == timer_get_idx(timer)) {
1001 if (!(options & MOD_TIMER_REDUCE))
1002 timer->expires = expires;
1003 else if (time_after(timer->expires, expires))
1004 timer->expires = expires;
1009 base = lock_timer_base(timer, &flags);
1010 forward_timer_base(base);
1013 ret = detach_if_pending(timer, base, false);
1014 if (!ret && (options & MOD_TIMER_PENDING_ONLY))
1017 new_base = get_target_base(base, timer->flags);
1019 if (base != new_base) {
1021 * We are trying to schedule the timer on the new base.
1022 * However we can't change timer's base while it is running,
1023 * otherwise del_timer_sync() can't detect that the timer's
1024 * handler yet has not finished. This also guarantees that the
1025 * timer is serialized wrt itself.
1027 if (likely(base->running_timer != timer)) {
1028 /* See the comment in lock_timer_base() */
1029 timer->flags |= TIMER_MIGRATING;
1031 raw_spin_unlock(&base->lock);
1033 raw_spin_lock(&base->lock);
1034 WRITE_ONCE(timer->flags,
1035 (timer->flags & ~TIMER_BASEMASK) | base->cpu);
1036 forward_timer_base(base);
1040 debug_timer_activate(timer);
1042 timer->expires = expires;
1044 * If 'idx' was calculated above and the base time did not advance
1045 * between calculating 'idx' and possibly switching the base, only
1046 * enqueue_timer() and trigger_dyntick_cpu() is required. Otherwise
1047 * we need to (re)calculate the wheel index via
1048 * internal_add_timer().
1050 if (idx != UINT_MAX && clk == base->clk) {
1051 enqueue_timer(base, timer, idx);
1052 trigger_dyntick_cpu(base, timer);
1054 internal_add_timer(base, timer);
1058 raw_spin_unlock_irqrestore(&base->lock, flags);
1064 * mod_timer_pending - modify a pending timer's timeout
1065 * @timer: the pending timer to be modified
1066 * @expires: new timeout in jiffies
1068 * mod_timer_pending() is the same for pending timers as mod_timer(),
1069 * but will not re-activate and modify already deleted timers.
1071 * It is useful for unserialized use of timers.
1073 int mod_timer_pending(struct timer_list *timer, unsigned long expires)
1075 return __mod_timer(timer, expires, MOD_TIMER_PENDING_ONLY);
1077 EXPORT_SYMBOL(mod_timer_pending);
1080 * mod_timer - modify a timer's timeout
1081 * @timer: the timer to be modified
1082 * @expires: new timeout in jiffies
1084 * mod_timer() is a more efficient way to update the expire field of an
1085 * active timer (if the timer is inactive it will be activated)
1087 * mod_timer(timer, expires) is equivalent to:
1089 * del_timer(timer); timer->expires = expires; add_timer(timer);
1091 * Note that if there are multiple unserialized concurrent users of the
1092 * same timer, then mod_timer() is the only safe way to modify the timeout,
1093 * since add_timer() cannot modify an already running timer.
1095 * The function returns whether it has modified a pending timer or not.
1096 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
1097 * active timer returns 1.)
1099 int mod_timer(struct timer_list *timer, unsigned long expires)
1101 return __mod_timer(timer, expires, 0);
1103 EXPORT_SYMBOL(mod_timer);
1106 * timer_reduce - Modify a timer's timeout if it would reduce the timeout
1107 * @timer: The timer to be modified
1108 * @expires: New timeout in jiffies
1110 * timer_reduce() is very similar to mod_timer(), except that it will only
1111 * modify a running timer if that would reduce the expiration time (it will
1112 * start a timer that isn't running).
1114 int timer_reduce(struct timer_list *timer, unsigned long expires)
1116 return __mod_timer(timer, expires, MOD_TIMER_REDUCE);
1118 EXPORT_SYMBOL(timer_reduce);
1121 * add_timer - start a timer
1122 * @timer: the timer to be added
1124 * The kernel will do a ->function(@timer) callback from the
1125 * timer interrupt at the ->expires point in the future. The
1126 * current time is 'jiffies'.
1128 * The timer's ->expires, ->function fields must be set prior calling this
1131 * Timers with an ->expires field in the past will be executed in the next
1134 void add_timer(struct timer_list *timer)
1136 BUG_ON(timer_pending(timer));
1137 __mod_timer(timer, timer->expires, MOD_TIMER_NOTPENDING);
1139 EXPORT_SYMBOL(add_timer);
1142 * add_timer_on - start a timer on a particular CPU
1143 * @timer: the timer to be added
1144 * @cpu: the CPU to start it on
1146 * This is not very scalable on SMP. Double adds are not possible.
1148 void add_timer_on(struct timer_list *timer, int cpu)
1150 struct timer_base *new_base, *base;
1151 unsigned long flags;
1153 BUG_ON(timer_pending(timer) || !timer->function);
1155 new_base = get_timer_cpu_base(timer->flags, cpu);
1158 * If @timer was on a different CPU, it should be migrated with the
1159 * old base locked to prevent other operations proceeding with the
1160 * wrong base locked. See lock_timer_base().
1162 base = lock_timer_base(timer, &flags);
1163 if (base != new_base) {
1164 timer->flags |= TIMER_MIGRATING;
1166 raw_spin_unlock(&base->lock);
1168 raw_spin_lock(&base->lock);
1169 WRITE_ONCE(timer->flags,
1170 (timer->flags & ~TIMER_BASEMASK) | cpu);
1172 forward_timer_base(base);
1174 debug_timer_activate(timer);
1175 internal_add_timer(base, timer);
1176 raw_spin_unlock_irqrestore(&base->lock, flags);
1178 EXPORT_SYMBOL_GPL(add_timer_on);
1181 * del_timer - deactivate a timer.
1182 * @timer: the timer to be deactivated
1184 * del_timer() deactivates a timer - this works on both active and inactive
1187 * The function returns whether it has deactivated a pending timer or not.
1188 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
1189 * active timer returns 1.)
1191 int del_timer(struct timer_list *timer)
1193 struct timer_base *base;
1194 unsigned long flags;
1197 debug_assert_init(timer);
1199 if (timer_pending(timer)) {
1200 base = lock_timer_base(timer, &flags);
1201 ret = detach_if_pending(timer, base, true);
1202 raw_spin_unlock_irqrestore(&base->lock, flags);
1207 EXPORT_SYMBOL(del_timer);
1210 * try_to_del_timer_sync - Try to deactivate a timer
1211 * @timer: timer to delete
1213 * This function tries to deactivate a timer. Upon successful (ret >= 0)
1214 * exit the timer is not queued and the handler is not running on any CPU.
1216 int try_to_del_timer_sync(struct timer_list *timer)
1218 struct timer_base *base;
1219 unsigned long flags;
1222 debug_assert_init(timer);
1224 base = lock_timer_base(timer, &flags);
1226 if (base->running_timer != timer)
1227 ret = detach_if_pending(timer, base, true);
1229 raw_spin_unlock_irqrestore(&base->lock, flags);
1233 EXPORT_SYMBOL(try_to_del_timer_sync);
1235 #ifdef CONFIG_PREEMPT_RT
1236 static __init void timer_base_init_expiry_lock(struct timer_base *base)
1238 spin_lock_init(&base->expiry_lock);
1241 static inline void timer_base_lock_expiry(struct timer_base *base)
1243 spin_lock(&base->expiry_lock);
1246 static inline void timer_base_unlock_expiry(struct timer_base *base)
1248 spin_unlock(&base->expiry_lock);
1252 * The counterpart to del_timer_wait_running().
1254 * If there is a waiter for base->expiry_lock, then it was waiting for the
1255 * timer callback to finish. Drop expiry_lock and reaquire it. That allows
1256 * the waiter to acquire the lock and make progress.
1258 static void timer_sync_wait_running(struct timer_base *base)
1260 if (atomic_read(&base->timer_waiters)) {
1261 spin_unlock(&base->expiry_lock);
1262 spin_lock(&base->expiry_lock);
1267 * This function is called on PREEMPT_RT kernels when the fast path
1268 * deletion of a timer failed because the timer callback function was
1271 * This prevents priority inversion, if the softirq thread on a remote CPU
1272 * got preempted, and it prevents a life lock when the task which tries to
1273 * delete a timer preempted the softirq thread running the timer callback
1276 static void del_timer_wait_running(struct timer_list *timer)
1280 tf = READ_ONCE(timer->flags);
1281 if (!(tf & TIMER_MIGRATING)) {
1282 struct timer_base *base = get_timer_base(tf);
1285 * Mark the base as contended and grab the expiry lock,
1286 * which is held by the softirq across the timer
1287 * callback. Drop the lock immediately so the softirq can
1288 * expire the next timer. In theory the timer could already
1289 * be running again, but that's more than unlikely and just
1290 * causes another wait loop.
1292 atomic_inc(&base->timer_waiters);
1293 spin_lock_bh(&base->expiry_lock);
1294 atomic_dec(&base->timer_waiters);
1295 spin_unlock_bh(&base->expiry_lock);
1299 static inline void timer_base_init_expiry_lock(struct timer_base *base) { }
1300 static inline void timer_base_lock_expiry(struct timer_base *base) { }
1301 static inline void timer_base_unlock_expiry(struct timer_base *base) { }
1302 static inline void timer_sync_wait_running(struct timer_base *base) { }
1303 static inline void del_timer_wait_running(struct timer_list *timer) { }
1306 #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT_RT)
1308 * del_timer_sync - deactivate a timer and wait for the handler to finish.
1309 * @timer: the timer to be deactivated
1311 * This function only differs from del_timer() on SMP: besides deactivating
1312 * the timer it also makes sure the handler has finished executing on other
1315 * Synchronization rules: Callers must prevent restarting of the timer,
1316 * otherwise this function is meaningless. It must not be called from
1317 * interrupt contexts unless the timer is an irqsafe one. The caller must
1318 * not hold locks which would prevent completion of the timer's
1319 * handler. The timer's handler must not call add_timer_on(). Upon exit the
1320 * timer is not queued and the handler is not running on any CPU.
1322 * Note: For !irqsafe timers, you must not hold locks that are held in
1323 * interrupt context while calling this function. Even if the lock has
1324 * nothing to do with the timer in question. Here's why::
1330 * base->running_timer = mytimer;
1331 * spin_lock_irq(somelock);
1333 * spin_lock(somelock);
1334 * del_timer_sync(mytimer);
1335 * while (base->running_timer == mytimer);
1337 * Now del_timer_sync() will never return and never release somelock.
1338 * The interrupt on the other CPU is waiting to grab somelock but
1339 * it has interrupted the softirq that CPU0 is waiting to finish.
1341 * The function returns whether it has deactivated a pending timer or not.
1343 int del_timer_sync(struct timer_list *timer)
1347 #ifdef CONFIG_LOCKDEP
1348 unsigned long flags;
1351 * If lockdep gives a backtrace here, please reference
1352 * the synchronization rules above.
1354 local_irq_save(flags);
1355 lock_map_acquire(&timer->lockdep_map);
1356 lock_map_release(&timer->lockdep_map);
1357 local_irq_restore(flags);
1360 * don't use it in hardirq context, because it
1361 * could lead to deadlock.
1363 WARN_ON(in_irq() && !(timer->flags & TIMER_IRQSAFE));
1366 ret = try_to_del_timer_sync(timer);
1368 if (unlikely(ret < 0)) {
1369 del_timer_wait_running(timer);
1376 EXPORT_SYMBOL(del_timer_sync);
1379 static void call_timer_fn(struct timer_list *timer,
1380 void (*fn)(struct timer_list *),
1381 unsigned long baseclk)
1383 int count = preempt_count();
1385 #ifdef CONFIG_LOCKDEP
1387 * It is permissible to free the timer from inside the
1388 * function that is called from it, this we need to take into
1389 * account for lockdep too. To avoid bogus "held lock freed"
1390 * warnings as well as problems when looking into
1391 * timer->lockdep_map, make a copy and use that here.
1393 struct lockdep_map lockdep_map;
1395 lockdep_copy_map(&lockdep_map, &timer->lockdep_map);
1398 * Couple the lock chain with the lock chain at
1399 * del_timer_sync() by acquiring the lock_map around the fn()
1400 * call here and in del_timer_sync().
1402 lock_map_acquire(&lockdep_map);
1404 trace_timer_expire_entry(timer, baseclk);
1406 trace_timer_expire_exit(timer);
1408 lock_map_release(&lockdep_map);
1410 if (count != preempt_count()) {
1411 WARN_ONCE(1, "timer: %pS preempt leak: %08x -> %08x\n",
1412 fn, count, preempt_count());
1414 * Restore the preempt count. That gives us a decent
1415 * chance to survive and extract information. If the
1416 * callback kept a lock held, bad luck, but not worse
1417 * than the BUG() we had.
1419 preempt_count_set(count);
1423 static void expire_timers(struct timer_base *base, struct hlist_head *head)
1426 * This value is required only for tracing. base->clk was
1427 * incremented directly before expire_timers was called. But expiry
1428 * is related to the old base->clk value.
1430 unsigned long baseclk = base->clk - 1;
1432 while (!hlist_empty(head)) {
1433 struct timer_list *timer;
1434 void (*fn)(struct timer_list *);
1436 timer = hlist_entry(head->first, struct timer_list, entry);
1438 base->running_timer = timer;
1439 detach_timer(timer, true);
1441 fn = timer->function;
1443 if (timer->flags & TIMER_IRQSAFE) {
1444 raw_spin_unlock(&base->lock);
1445 call_timer_fn(timer, fn, baseclk);
1446 base->running_timer = NULL;
1447 raw_spin_lock(&base->lock);
1449 raw_spin_unlock_irq(&base->lock);
1450 call_timer_fn(timer, fn, baseclk);
1451 base->running_timer = NULL;
1452 timer_sync_wait_running(base);
1453 raw_spin_lock_irq(&base->lock);
1458 static int __collect_expired_timers(struct timer_base *base,
1459 struct hlist_head *heads)
1461 unsigned long clk = base->clk;
1462 struct hlist_head *vec;
1466 for (i = 0; i < LVL_DEPTH; i++) {
1467 idx = (clk & LVL_MASK) + i * LVL_SIZE;
1469 if (__test_and_clear_bit(idx, base->pending_map)) {
1470 vec = base->vectors + idx;
1471 hlist_move_list(vec, heads++);
1474 /* Is it time to look at the next level? */
1475 if (clk & LVL_CLK_MASK)
1477 /* Shift clock for the next level granularity */
1478 clk >>= LVL_CLK_SHIFT;
1483 #ifdef CONFIG_NO_HZ_COMMON
1485 * Find the next pending bucket of a level. Search from level start (@offset)
1486 * + @clk upwards and if nothing there, search from start of the level
1487 * (@offset) up to @offset + clk.
1489 static int next_pending_bucket(struct timer_base *base, unsigned offset,
1492 unsigned pos, start = offset + clk;
1493 unsigned end = offset + LVL_SIZE;
1495 pos = find_next_bit(base->pending_map, end, start);
1499 pos = find_next_bit(base->pending_map, start, offset);
1500 return pos < start ? pos + LVL_SIZE - start : -1;
1504 * Search the first expiring timer in the various clock levels. Caller must
1507 static unsigned long __next_timer_interrupt(struct timer_base *base)
1509 unsigned long clk, next, adj;
1510 unsigned lvl, offset = 0;
1512 next = base->clk + NEXT_TIMER_MAX_DELTA;
1514 for (lvl = 0; lvl < LVL_DEPTH; lvl++, offset += LVL_SIZE) {
1515 int pos = next_pending_bucket(base, offset, clk & LVL_MASK);
1518 unsigned long tmp = clk + (unsigned long) pos;
1520 tmp <<= LVL_SHIFT(lvl);
1521 if (time_before(tmp, next))
1525 * Clock for the next level. If the current level clock lower
1526 * bits are zero, we look at the next level as is. If not we
1527 * need to advance it by one because that's going to be the
1528 * next expiring bucket in that level. base->clk is the next
1529 * expiring jiffie. So in case of:
1531 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1534 * we have to look at all levels @index 0. With
1536 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1539 * LVL0 has the next expiring bucket @index 2. The upper
1540 * levels have the next expiring bucket @index 1.
1542 * In case that the propagation wraps the next level the same
1545 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1548 * So after looking at LVL0 we get:
1550 * LVL5 LVL4 LVL3 LVL2 LVL1
1553 * So no propagation from LVL1 to LVL2 because that happened
1554 * with the add already, but then we need to propagate further
1555 * from LVL2 to LVL3.
1557 * So the simple check whether the lower bits of the current
1558 * level are 0 or not is sufficient for all cases.
1560 adj = clk & LVL_CLK_MASK ? 1 : 0;
1561 clk >>= LVL_CLK_SHIFT;
1568 * Check, if the next hrtimer event is before the next timer wheel
1571 static u64 cmp_next_hrtimer_event(u64 basem, u64 expires)
1573 u64 nextevt = hrtimer_get_next_event();
1576 * If high resolution timers are enabled
1577 * hrtimer_get_next_event() returns KTIME_MAX.
1579 if (expires <= nextevt)
1583 * If the next timer is already expired, return the tick base
1584 * time so the tick is fired immediately.
1586 if (nextevt <= basem)
1590 * Round up to the next jiffie. High resolution timers are
1591 * off, so the hrtimers are expired in the tick and we need to
1592 * make sure that this tick really expires the timer to avoid
1593 * a ping pong of the nohz stop code.
1595 * Use DIV_ROUND_UP_ULL to prevent gcc calling __divdi3
1597 return DIV_ROUND_UP_ULL(nextevt, TICK_NSEC) * TICK_NSEC;
1601 * get_next_timer_interrupt - return the time (clock mono) of the next timer
1602 * @basej: base time jiffies
1603 * @basem: base time clock monotonic
1605 * Returns the tick aligned clock monotonic time of the next pending
1606 * timer or KTIME_MAX if no timer is pending.
1608 u64 get_next_timer_interrupt(unsigned long basej, u64 basem)
1610 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1611 u64 expires = KTIME_MAX;
1612 unsigned long nextevt;
1616 * Pretend that there is no timer pending if the cpu is offline.
1617 * Possible pending timers will be migrated later to an active cpu.
1619 if (cpu_is_offline(smp_processor_id()))
1622 raw_spin_lock(&base->lock);
1623 nextevt = __next_timer_interrupt(base);
1624 is_max_delta = (nextevt == base->clk + NEXT_TIMER_MAX_DELTA);
1625 base->next_expiry = nextevt;
1627 * We have a fresh next event. Check whether we can forward the
1628 * base. We can only do that when @basej is past base->clk
1629 * otherwise we might rewind base->clk.
1631 if (time_after(basej, base->clk)) {
1632 if (time_after(nextevt, basej))
1634 else if (time_after(nextevt, base->clk))
1635 base->clk = nextevt;
1638 if (time_before_eq(nextevt, basej)) {
1640 base->is_idle = false;
1643 expires = basem + (u64)(nextevt - basej) * TICK_NSEC;
1645 * If we expect to sleep more than a tick, mark the base idle.
1646 * Also the tick is stopped so any added timer must forward
1647 * the base clk itself to keep granularity small. This idle
1648 * logic is only maintained for the BASE_STD base, deferrable
1649 * timers may still see large granularity skew (by design).
1651 if ((expires - basem) > TICK_NSEC) {
1652 base->must_forward_clk = true;
1653 base->is_idle = true;
1656 raw_spin_unlock(&base->lock);
1658 return cmp_next_hrtimer_event(basem, expires);
1662 * timer_clear_idle - Clear the idle state of the timer base
1664 * Called with interrupts disabled
1666 void timer_clear_idle(void)
1668 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1671 * We do this unlocked. The worst outcome is a remote enqueue sending
1672 * a pointless IPI, but taking the lock would just make the window for
1673 * sending the IPI a few instructions smaller for the cost of taking
1674 * the lock in the exit from idle path.
1676 base->is_idle = false;
1679 static int collect_expired_timers(struct timer_base *base,
1680 struct hlist_head *heads)
1682 unsigned long now = READ_ONCE(jiffies);
1685 * NOHZ optimization. After a long idle sleep we need to forward the
1686 * base to current jiffies. Avoid a loop by searching the bitfield for
1687 * the next expiring timer.
1689 if ((long)(now - base->clk) > 2) {
1690 unsigned long next = __next_timer_interrupt(base);
1693 * If the next timer is ahead of time forward to current
1694 * jiffies, otherwise forward to the next expiry time:
1696 if (time_after(next, now)) {
1698 * The call site will increment base->clk and then
1699 * terminate the expiry loop immediately.
1706 return __collect_expired_timers(base, heads);
1709 static inline int collect_expired_timers(struct timer_base *base,
1710 struct hlist_head *heads)
1712 return __collect_expired_timers(base, heads);
1717 * Called from the timer interrupt handler to charge one tick to the current
1718 * process. user_tick is 1 if the tick is user time, 0 for system.
1720 void update_process_times(int user_tick)
1722 struct task_struct *p = current;
1724 /* Note: this timer irq context must be accounted for as well. */
1725 account_process_tick(p, user_tick);
1727 rcu_sched_clock_irq(user_tick);
1728 #ifdef CONFIG_IRQ_WORK
1733 if (IS_ENABLED(CONFIG_POSIX_TIMERS))
1734 run_posix_cpu_timers();
1738 * __run_timers - run all expired timers (if any) on this CPU.
1739 * @base: the timer vector to be processed.
1741 static inline void __run_timers(struct timer_base *base)
1743 struct hlist_head heads[LVL_DEPTH];
1746 if (!time_after_eq(jiffies, base->clk))
1749 timer_base_lock_expiry(base);
1750 raw_spin_lock_irq(&base->lock);
1753 * timer_base::must_forward_clk must be cleared before running
1754 * timers so that any timer functions that call mod_timer() will
1755 * not try to forward the base. Idle tracking / clock forwarding
1756 * logic is only used with BASE_STD timers.
1758 * The must_forward_clk flag is cleared unconditionally also for
1759 * the deferrable base. The deferrable base is not affected by idle
1760 * tracking and never forwarded, so clearing the flag is a NOOP.
1762 * The fact that the deferrable base is never forwarded can cause
1763 * large variations in granularity for deferrable timers, but they
1764 * can be deferred for long periods due to idle anyway.
1766 base->must_forward_clk = false;
1768 while (time_after_eq(jiffies, base->clk)) {
1770 levels = collect_expired_timers(base, heads);
1774 expire_timers(base, heads + levels);
1776 raw_spin_unlock_irq(&base->lock);
1777 timer_base_unlock_expiry(base);
1781 * This function runs timers and the timer-tq in bottom half context.
1783 static __latent_entropy void run_timer_softirq(struct softirq_action *h)
1785 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1788 if (IS_ENABLED(CONFIG_NO_HZ_COMMON))
1789 __run_timers(this_cpu_ptr(&timer_bases[BASE_DEF]));
1793 * Called by the local, per-CPU timer interrupt on SMP.
1795 void run_local_timers(void)
1797 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1799 hrtimer_run_queues();
1800 /* Raise the softirq only if required. */
1801 if (time_before(jiffies, base->clk)) {
1802 if (!IS_ENABLED(CONFIG_NO_HZ_COMMON))
1804 /* CPU is awake, so check the deferrable base. */
1806 if (time_before(jiffies, base->clk))
1809 raise_softirq(TIMER_SOFTIRQ);
1813 * Since schedule_timeout()'s timer is defined on the stack, it must store
1814 * the target task on the stack as well.
1816 struct process_timer {
1817 struct timer_list timer;
1818 struct task_struct *task;
1821 static void process_timeout(struct timer_list *t)
1823 struct process_timer *timeout = from_timer(timeout, t, timer);
1825 wake_up_process(timeout->task);
1829 * schedule_timeout - sleep until timeout
1830 * @timeout: timeout value in jiffies
1832 * Make the current task sleep until @timeout jiffies have elapsed.
1833 * The function behavior depends on the current task state
1834 * (see also set_current_state() description):
1836 * %TASK_RUNNING - the scheduler is called, but the task does not sleep
1837 * at all. That happens because sched_submit_work() does nothing for
1838 * tasks in %TASK_RUNNING state.
1840 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1841 * pass before the routine returns unless the current task is explicitly
1842 * woken up, (e.g. by wake_up_process()).
1844 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1845 * delivered to the current task or the current task is explicitly woken
1848 * The current task state is guaranteed to be %TASK_RUNNING when this
1851 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1852 * the CPU away without a bound on the timeout. In this case the return
1853 * value will be %MAX_SCHEDULE_TIMEOUT.
1855 * Returns 0 when the timer has expired otherwise the remaining time in
1856 * jiffies will be returned. In all cases the return value is guaranteed
1857 * to be non-negative.
1859 signed long __sched schedule_timeout(signed long timeout)
1861 struct process_timer timer;
1862 unsigned long expire;
1866 case MAX_SCHEDULE_TIMEOUT:
1868 * These two special cases are useful to be comfortable
1869 * in the caller. Nothing more. We could take
1870 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1871 * but I' d like to return a valid offset (>=0) to allow
1872 * the caller to do everything it want with the retval.
1878 * Another bit of PARANOID. Note that the retval will be
1879 * 0 since no piece of kernel is supposed to do a check
1880 * for a negative retval of schedule_timeout() (since it
1881 * should never happens anyway). You just have the printk()
1882 * that will tell you if something is gone wrong and where.
1885 printk(KERN_ERR "schedule_timeout: wrong timeout "
1886 "value %lx\n", timeout);
1888 current->state = TASK_RUNNING;
1893 expire = timeout + jiffies;
1895 timer.task = current;
1896 timer_setup_on_stack(&timer.timer, process_timeout, 0);
1897 __mod_timer(&timer.timer, expire, MOD_TIMER_NOTPENDING);
1899 del_singleshot_timer_sync(&timer.timer);
1901 /* Remove the timer from the object tracker */
1902 destroy_timer_on_stack(&timer.timer);
1904 timeout = expire - jiffies;
1907 return timeout < 0 ? 0 : timeout;
1909 EXPORT_SYMBOL(schedule_timeout);
1912 * We can use __set_current_state() here because schedule_timeout() calls
1913 * schedule() unconditionally.
1915 signed long __sched schedule_timeout_interruptible(signed long timeout)
1917 __set_current_state(TASK_INTERRUPTIBLE);
1918 return schedule_timeout(timeout);
1920 EXPORT_SYMBOL(schedule_timeout_interruptible);
1922 signed long __sched schedule_timeout_killable(signed long timeout)
1924 __set_current_state(TASK_KILLABLE);
1925 return schedule_timeout(timeout);
1927 EXPORT_SYMBOL(schedule_timeout_killable);
1929 signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1931 __set_current_state(TASK_UNINTERRUPTIBLE);
1932 return schedule_timeout(timeout);
1934 EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1937 * Like schedule_timeout_uninterruptible(), except this task will not contribute
1940 signed long __sched schedule_timeout_idle(signed long timeout)
1942 __set_current_state(TASK_IDLE);
1943 return schedule_timeout(timeout);
1945 EXPORT_SYMBOL(schedule_timeout_idle);
1947 #ifdef CONFIG_HOTPLUG_CPU
1948 static void migrate_timer_list(struct timer_base *new_base, struct hlist_head *head)
1950 struct timer_list *timer;
1951 int cpu = new_base->cpu;
1953 while (!hlist_empty(head)) {
1954 timer = hlist_entry(head->first, struct timer_list, entry);
1955 detach_timer(timer, false);
1956 timer->flags = (timer->flags & ~TIMER_BASEMASK) | cpu;
1957 internal_add_timer(new_base, timer);
1961 int timers_prepare_cpu(unsigned int cpu)
1963 struct timer_base *base;
1966 for (b = 0; b < NR_BASES; b++) {
1967 base = per_cpu_ptr(&timer_bases[b], cpu);
1968 base->clk = jiffies;
1969 base->next_expiry = base->clk + NEXT_TIMER_MAX_DELTA;
1970 base->is_idle = false;
1971 base->must_forward_clk = true;
1976 int timers_dead_cpu(unsigned int cpu)
1978 struct timer_base *old_base;
1979 struct timer_base *new_base;
1982 BUG_ON(cpu_online(cpu));
1984 for (b = 0; b < NR_BASES; b++) {
1985 old_base = per_cpu_ptr(&timer_bases[b], cpu);
1986 new_base = get_cpu_ptr(&timer_bases[b]);
1988 * The caller is globally serialized and nobody else
1989 * takes two locks at once, deadlock is not possible.
1991 raw_spin_lock_irq(&new_base->lock);
1992 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1995 * The current CPUs base clock might be stale. Update it
1996 * before moving the timers over.
1998 forward_timer_base(new_base);
2000 BUG_ON(old_base->running_timer);
2002 for (i = 0; i < WHEEL_SIZE; i++)
2003 migrate_timer_list(new_base, old_base->vectors + i);
2005 raw_spin_unlock(&old_base->lock);
2006 raw_spin_unlock_irq(&new_base->lock);
2007 put_cpu_ptr(&timer_bases);
2012 #endif /* CONFIG_HOTPLUG_CPU */
2014 static void __init init_timer_cpu(int cpu)
2016 struct timer_base *base;
2019 for (i = 0; i < NR_BASES; i++) {
2020 base = per_cpu_ptr(&timer_bases[i], cpu);
2022 raw_spin_lock_init(&base->lock);
2023 base->clk = jiffies;
2024 timer_base_init_expiry_lock(base);
2028 static void __init init_timer_cpus(void)
2032 for_each_possible_cpu(cpu)
2033 init_timer_cpu(cpu);
2036 void __init init_timers(void)
2039 open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
2043 * msleep - sleep safely even with waitqueue interruptions
2044 * @msecs: Time in milliseconds to sleep for
2046 void msleep(unsigned int msecs)
2048 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
2051 timeout = schedule_timeout_uninterruptible(timeout);
2054 EXPORT_SYMBOL(msleep);
2057 * msleep_interruptible - sleep waiting for signals
2058 * @msecs: Time in milliseconds to sleep for
2060 unsigned long msleep_interruptible(unsigned int msecs)
2062 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
2064 while (timeout && !signal_pending(current))
2065 timeout = schedule_timeout_interruptible(timeout);
2066 return jiffies_to_msecs(timeout);
2069 EXPORT_SYMBOL(msleep_interruptible);
2072 * usleep_range - Sleep for an approximate time
2073 * @min: Minimum time in usecs to sleep
2074 * @max: Maximum time in usecs to sleep
2076 * In non-atomic context where the exact wakeup time is flexible, use
2077 * usleep_range() instead of udelay(). The sleep improves responsiveness
2078 * by avoiding the CPU-hogging busy-wait of udelay(), and the range reduces
2079 * power usage by allowing hrtimers to take advantage of an already-
2080 * scheduled interrupt instead of scheduling a new one just for this sleep.
2082 void __sched usleep_range(unsigned long min, unsigned long max)
2084 ktime_t exp = ktime_add_us(ktime_get(), min);
2085 u64 delta = (u64)(max - min) * NSEC_PER_USEC;
2088 __set_current_state(TASK_UNINTERRUPTIBLE);
2089 /* Do not return before the requested sleep time has elapsed */
2090 if (!schedule_hrtimeout_range(&exp, delta, HRTIMER_MODE_ABS))
2094 EXPORT_SYMBOL(usleep_range);