4 * Kernel internal timers
6 * Copyright (C) 1991, 1992 Linus Torvalds
8 * 1997-01-28 Modified by Finn Arne Gangstad to make timers scale better.
10 * 1997-09-10 Updated NTP code according to technical memorandum Jan '96
11 * "A Kernel Model for Precision Timekeeping" by Dave Mills
12 * 1998-12-24 Fixed a xtime SMP race (we need the xtime_lock rw spinlock to
13 * serialize accesses to xtime/lost_ticks).
14 * Copyright (C) 1998 Andrea Arcangeli
15 * 1999-03-10 Improved NTP compatibility by Ulrich Windl
16 * 2002-05-31 Move sys_sysinfo here and make its locking sane, Robert Love
17 * 2000-10-05 Implemented scalable SMP per-CPU timer handling.
18 * Copyright (C) 2000, 2001, 2002 Ingo Molnar
19 * Designed by David S. Miller, Alexey Kuznetsov and Ingo Molnar
22 #include <linux/kernel_stat.h>
23 #include <linux/export.h>
24 #include <linux/interrupt.h>
25 #include <linux/percpu.h>
26 #include <linux/init.h>
28 #include <linux/swap.h>
29 #include <linux/pid_namespace.h>
30 #include <linux/notifier.h>
31 #include <linux/thread_info.h>
32 #include <linux/time.h>
33 #include <linux/jiffies.h>
34 #include <linux/posix-timers.h>
35 #include <linux/cpu.h>
36 #include <linux/syscalls.h>
37 #include <linux/delay.h>
38 #include <linux/tick.h>
39 #include <linux/kallsyms.h>
40 #include <linux/irq_work.h>
41 #include <linux/sched/signal.h>
42 #include <linux/sched/sysctl.h>
43 #include <linux/sched/nohz.h>
44 #include <linux/sched/debug.h>
45 #include <linux/slab.h>
46 #include <linux/compat.h>
47 #include <linux/random.h>
49 #include <linux/uaccess.h>
50 #include <asm/unistd.h>
51 #include <asm/div64.h>
52 #include <asm/timex.h>
55 #include "tick-internal.h"
57 #define CREATE_TRACE_POINTS
58 #include <trace/events/timer.h>
60 __visible u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
62 EXPORT_SYMBOL(jiffies_64);
65 * The timer wheel has LVL_DEPTH array levels. Each level provides an array of
66 * LVL_SIZE buckets. Each level is driven by its own clock and therefor each
67 * level has a different granularity.
69 * The level granularity is: LVL_CLK_DIV ^ lvl
70 * The level clock frequency is: HZ / (LVL_CLK_DIV ^ level)
72 * The array level of a newly armed timer depends on the relative expiry
73 * time. The farther the expiry time is away the higher the array level and
74 * therefor the granularity becomes.
76 * Contrary to the original timer wheel implementation, which aims for 'exact'
77 * expiry of the timers, this implementation removes the need for recascading
78 * the timers into the lower array levels. The previous 'classic' timer wheel
79 * implementation of the kernel already violated the 'exact' expiry by adding
80 * slack to the expiry time to provide batched expiration. The granularity
81 * levels provide implicit batching.
83 * This is an optimization of the original timer wheel implementation for the
84 * majority of the timer wheel use cases: timeouts. The vast majority of
85 * timeout timers (networking, disk I/O ...) are canceled before expiry. If
86 * the timeout expires it indicates that normal operation is disturbed, so it
87 * does not matter much whether the timeout comes with a slight delay.
89 * The only exception to this are networking timers with a small expiry
90 * time. They rely on the granularity. Those fit into the first wheel level,
91 * which has HZ granularity.
93 * We don't have cascading anymore. timers with a expiry time above the
94 * capacity of the last wheel level are force expired at the maximum timeout
95 * value of the last wheel level. From data sampling we know that the maximum
96 * value observed is 5 days (network connection tracking), so this should not
99 * The currently chosen array constants values are a good compromise between
100 * array size and granularity.
102 * This results in the following granularity and range levels:
105 * Level Offset Granularity Range
106 * 0 0 1 ms 0 ms - 63 ms
107 * 1 64 8 ms 64 ms - 511 ms
108 * 2 128 64 ms 512 ms - 4095 ms (512ms - ~4s)
109 * 3 192 512 ms 4096 ms - 32767 ms (~4s - ~32s)
110 * 4 256 4096 ms (~4s) 32768 ms - 262143 ms (~32s - ~4m)
111 * 5 320 32768 ms (~32s) 262144 ms - 2097151 ms (~4m - ~34m)
112 * 6 384 262144 ms (~4m) 2097152 ms - 16777215 ms (~34m - ~4h)
113 * 7 448 2097152 ms (~34m) 16777216 ms - 134217727 ms (~4h - ~1d)
114 * 8 512 16777216 ms (~4h) 134217728 ms - 1073741822 ms (~1d - ~12d)
117 * Level Offset Granularity Range
118 * 0 0 3 ms 0 ms - 210 ms
119 * 1 64 26 ms 213 ms - 1703 ms (213ms - ~1s)
120 * 2 128 213 ms 1706 ms - 13650 ms (~1s - ~13s)
121 * 3 192 1706 ms (~1s) 13653 ms - 109223 ms (~13s - ~1m)
122 * 4 256 13653 ms (~13s) 109226 ms - 873810 ms (~1m - ~14m)
123 * 5 320 109226 ms (~1m) 873813 ms - 6990503 ms (~14m - ~1h)
124 * 6 384 873813 ms (~14m) 6990506 ms - 55924050 ms (~1h - ~15h)
125 * 7 448 6990506 ms (~1h) 55924053 ms - 447392423 ms (~15h - ~5d)
126 * 8 512 55924053 ms (~15h) 447392426 ms - 3579139406 ms (~5d - ~41d)
129 * Level Offset Granularity Range
130 * 0 0 4 ms 0 ms - 255 ms
131 * 1 64 32 ms 256 ms - 2047 ms (256ms - ~2s)
132 * 2 128 256 ms 2048 ms - 16383 ms (~2s - ~16s)
133 * 3 192 2048 ms (~2s) 16384 ms - 131071 ms (~16s - ~2m)
134 * 4 256 16384 ms (~16s) 131072 ms - 1048575 ms (~2m - ~17m)
135 * 5 320 131072 ms (~2m) 1048576 ms - 8388607 ms (~17m - ~2h)
136 * 6 384 1048576 ms (~17m) 8388608 ms - 67108863 ms (~2h - ~18h)
137 * 7 448 8388608 ms (~2h) 67108864 ms - 536870911 ms (~18h - ~6d)
138 * 8 512 67108864 ms (~18h) 536870912 ms - 4294967288 ms (~6d - ~49d)
141 * Level Offset Granularity Range
142 * 0 0 10 ms 0 ms - 630 ms
143 * 1 64 80 ms 640 ms - 5110 ms (640ms - ~5s)
144 * 2 128 640 ms 5120 ms - 40950 ms (~5s - ~40s)
145 * 3 192 5120 ms (~5s) 40960 ms - 327670 ms (~40s - ~5m)
146 * 4 256 40960 ms (~40s) 327680 ms - 2621430 ms (~5m - ~43m)
147 * 5 320 327680 ms (~5m) 2621440 ms - 20971510 ms (~43m - ~5h)
148 * 6 384 2621440 ms (~43m) 20971520 ms - 167772150 ms (~5h - ~1d)
149 * 7 448 20971520 ms (~5h) 167772160 ms - 1342177270 ms (~1d - ~15d)
152 /* Clock divisor for the next level */
153 #define LVL_CLK_SHIFT 3
154 #define LVL_CLK_DIV (1UL << LVL_CLK_SHIFT)
155 #define LVL_CLK_MASK (LVL_CLK_DIV - 1)
156 #define LVL_SHIFT(n) ((n) * LVL_CLK_SHIFT)
157 #define LVL_GRAN(n) (1UL << LVL_SHIFT(n))
160 * The time start value for each level to select the bucket at enqueue
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;
202 unsigned long next_expiry;
205 bool must_forward_clk;
206 DECLARE_BITMAP(pending_map, WHEEL_SIZE);
207 struct hlist_head vectors[WHEEL_SIZE];
208 } ____cacheline_aligned;
210 static DEFINE_PER_CPU(struct timer_base, timer_bases[NR_BASES]);
212 #ifdef CONFIG_NO_HZ_COMMON
214 static DEFINE_STATIC_KEY_FALSE(timers_nohz_active);
215 static DEFINE_MUTEX(timer_keys_mutex);
217 static void timer_update_keys(struct work_struct *work);
218 static DECLARE_WORK(timer_update_work, timer_update_keys);
221 unsigned int sysctl_timer_migration = 1;
223 DEFINE_STATIC_KEY_FALSE(timers_migration_enabled);
225 static void timers_update_migration(void)
227 if (sysctl_timer_migration && tick_nohz_active)
228 static_branch_enable(&timers_migration_enabled);
230 static_branch_disable(&timers_migration_enabled);
233 static inline void timers_update_migration(void) { }
234 #endif /* !CONFIG_SMP */
236 static void timer_update_keys(struct work_struct *work)
238 mutex_lock(&timer_keys_mutex);
239 timers_update_migration();
240 static_branch_enable(&timers_nohz_active);
241 mutex_unlock(&timer_keys_mutex);
244 void timers_update_nohz(void)
246 schedule_work(&timer_update_work);
249 int timer_migration_handler(struct ctl_table *table, int write,
250 void __user *buffer, size_t *lenp,
255 mutex_lock(&timer_keys_mutex);
256 ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
258 timers_update_migration();
259 mutex_unlock(&timer_keys_mutex);
263 static inline bool is_timers_nohz_active(void)
265 return static_branch_unlikely(&timers_nohz_active);
268 static inline bool is_timers_nohz_active(void) { return false; }
269 #endif /* NO_HZ_COMMON */
271 static unsigned long round_jiffies_common(unsigned long j, int cpu,
275 unsigned long original = j;
278 * We don't want all cpus firing their timers at once hitting the
279 * same lock or cachelines, so we skew each extra cpu with an extra
280 * 3 jiffies. This 3 jiffies came originally from the mm/ code which
282 * The skew is done by adding 3*cpunr, then round, then subtract this
283 * extra offset again.
290 * If the target jiffie is just after a whole second (which can happen
291 * due to delays of the timer irq, long irq off times etc etc) then
292 * we should round down to the whole second, not up. Use 1/4th second
293 * as cutoff for this rounding as an extreme upper bound for this.
294 * But never round down if @force_up is set.
296 if (rem < HZ/4 && !force_up) /* round down */
301 /* now that we have rounded, subtract the extra skew again */
305 * Make sure j is still in the future. Otherwise return the
308 return time_is_after_jiffies(j) ? j : original;
312 * __round_jiffies - function to round jiffies to a full second
313 * @j: the time in (absolute) jiffies that should be rounded
314 * @cpu: the processor number on which the timeout will happen
316 * __round_jiffies() rounds an absolute time in the future (in jiffies)
317 * up or down to (approximately) full seconds. This is useful for timers
318 * for which the exact time they fire does not matter too much, as long as
319 * they fire approximately every X seconds.
321 * By rounding these timers to whole seconds, all such timers will fire
322 * at the same time, rather than at various times spread out. The goal
323 * of this is to have the CPU wake up less, which saves power.
325 * The exact rounding is skewed for each processor to avoid all
326 * processors firing at the exact same time, which could lead
327 * to lock contention or spurious cache line bouncing.
329 * The return value is the rounded version of the @j parameter.
331 unsigned long __round_jiffies(unsigned long j, int cpu)
333 return round_jiffies_common(j, cpu, false);
335 EXPORT_SYMBOL_GPL(__round_jiffies);
338 * __round_jiffies_relative - function to round jiffies to a full second
339 * @j: the time in (relative) jiffies that should be rounded
340 * @cpu: the processor number on which the timeout will happen
342 * __round_jiffies_relative() rounds a time delta in the future (in jiffies)
343 * up or down to (approximately) full seconds. This is useful for timers
344 * for which the exact time they fire does not matter too much, as long as
345 * they fire approximately every X seconds.
347 * By rounding these timers to whole seconds, all such timers will fire
348 * at the same time, rather than at various times spread out. The goal
349 * of this is to have the CPU wake up less, which saves power.
351 * The exact rounding is skewed for each processor to avoid all
352 * processors firing at the exact same time, which could lead
353 * to lock contention or spurious cache line bouncing.
355 * The return value is the rounded version of the @j parameter.
357 unsigned long __round_jiffies_relative(unsigned long j, int cpu)
359 unsigned long j0 = jiffies;
361 /* Use j0 because jiffies might change while we run */
362 return round_jiffies_common(j + j0, cpu, false) - j0;
364 EXPORT_SYMBOL_GPL(__round_jiffies_relative);
367 * round_jiffies - function to round jiffies to a full second
368 * @j: the time in (absolute) jiffies that should be rounded
370 * round_jiffies() rounds an absolute time in the future (in jiffies)
371 * up or down to (approximately) full seconds. This is useful for timers
372 * for which the exact time they fire does not matter too much, as long as
373 * they fire approximately every X seconds.
375 * By rounding these timers to whole seconds, all such timers will fire
376 * at the same time, rather than at various times spread out. The goal
377 * of this is to have the CPU wake up less, which saves power.
379 * The return value is the rounded version of the @j parameter.
381 unsigned long round_jiffies(unsigned long j)
383 return round_jiffies_common(j, raw_smp_processor_id(), false);
385 EXPORT_SYMBOL_GPL(round_jiffies);
388 * round_jiffies_relative - function to round jiffies to a full second
389 * @j: the time in (relative) jiffies that should be rounded
391 * round_jiffies_relative() rounds a time delta in the future (in jiffies)
392 * up or down to (approximately) full seconds. This is useful for timers
393 * for which the exact time they fire does not matter too much, as long as
394 * they fire approximately every X seconds.
396 * By rounding these timers to whole seconds, all such timers will fire
397 * at the same time, rather than at various times spread out. The goal
398 * of this is to have the CPU wake up less, which saves power.
400 * The return value is the rounded version of the @j parameter.
402 unsigned long round_jiffies_relative(unsigned long j)
404 return __round_jiffies_relative(j, raw_smp_processor_id());
406 EXPORT_SYMBOL_GPL(round_jiffies_relative);
409 * __round_jiffies_up - function to round jiffies up to a full second
410 * @j: the time in (absolute) jiffies that should be rounded
411 * @cpu: the processor number on which the timeout will happen
413 * This is the same as __round_jiffies() except that it will never
414 * round down. This is useful for timeouts for which the exact time
415 * of firing does not matter too much, as long as they don't fire too
418 unsigned long __round_jiffies_up(unsigned long j, int cpu)
420 return round_jiffies_common(j, cpu, true);
422 EXPORT_SYMBOL_GPL(__round_jiffies_up);
425 * __round_jiffies_up_relative - function to round jiffies up to a full second
426 * @j: the time in (relative) jiffies that should be rounded
427 * @cpu: the processor number on which the timeout will happen
429 * This is the same as __round_jiffies_relative() except that it will never
430 * round down. This is useful for timeouts for which the exact time
431 * of firing does not matter too much, as long as they don't fire too
434 unsigned long __round_jiffies_up_relative(unsigned long j, int cpu)
436 unsigned long j0 = jiffies;
438 /* Use j0 because jiffies might change while we run */
439 return round_jiffies_common(j + j0, cpu, true) - j0;
441 EXPORT_SYMBOL_GPL(__round_jiffies_up_relative);
444 * round_jiffies_up - function to round jiffies up to a full second
445 * @j: the time in (absolute) jiffies that should be rounded
447 * This is the same as round_jiffies() except that it will never
448 * round down. This is useful for timeouts for which the exact time
449 * of firing does not matter too much, as long as they don't fire too
452 unsigned long round_jiffies_up(unsigned long j)
454 return round_jiffies_common(j, raw_smp_processor_id(), true);
456 EXPORT_SYMBOL_GPL(round_jiffies_up);
459 * round_jiffies_up_relative - function to round jiffies up to a full second
460 * @j: the time in (relative) jiffies that should be rounded
462 * This is the same as round_jiffies_relative() except that it will never
463 * round down. This is useful for timeouts for which the exact time
464 * of firing does not matter too much, as long as they don't fire too
467 unsigned long round_jiffies_up_relative(unsigned long j)
469 return __round_jiffies_up_relative(j, raw_smp_processor_id());
471 EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
474 static inline unsigned int timer_get_idx(struct timer_list *timer)
476 return (timer->flags & TIMER_ARRAYMASK) >> TIMER_ARRAYSHIFT;
479 static inline void timer_set_idx(struct timer_list *timer, unsigned int idx)
481 timer->flags = (timer->flags & ~TIMER_ARRAYMASK) |
482 idx << TIMER_ARRAYSHIFT;
486 * Helper function to calculate the array index for a given expiry
489 static inline unsigned calc_index(unsigned expires, unsigned lvl)
491 expires = (expires + LVL_GRAN(lvl)) >> LVL_SHIFT(lvl);
492 return LVL_OFFS(lvl) + (expires & LVL_MASK);
495 static int calc_wheel_index(unsigned long expires, unsigned long clk)
497 unsigned long delta = expires - clk;
500 if (delta < LVL_START(1)) {
501 idx = calc_index(expires, 0);
502 } else if (delta < LVL_START(2)) {
503 idx = calc_index(expires, 1);
504 } else if (delta < LVL_START(3)) {
505 idx = calc_index(expires, 2);
506 } else if (delta < LVL_START(4)) {
507 idx = calc_index(expires, 3);
508 } else if (delta < LVL_START(5)) {
509 idx = calc_index(expires, 4);
510 } else if (delta < LVL_START(6)) {
511 idx = calc_index(expires, 5);
512 } else if (delta < LVL_START(7)) {
513 idx = calc_index(expires, 6);
514 } else if (LVL_DEPTH > 8 && delta < LVL_START(8)) {
515 idx = calc_index(expires, 7);
516 } else if ((long) delta < 0) {
517 idx = clk & LVL_MASK;
520 * Force expire obscene large timeouts to expire at the
521 * capacity limit of the wheel.
523 if (delta >= WHEEL_TIMEOUT_CUTOFF)
524 expires = clk + WHEEL_TIMEOUT_MAX;
526 idx = calc_index(expires, LVL_DEPTH - 1);
532 * Enqueue the timer into the hash bucket, mark it pending in
533 * the bitmap and store the index in the timer flags.
535 static void enqueue_timer(struct timer_base *base, struct timer_list *timer,
538 hlist_add_head(&timer->entry, base->vectors + idx);
539 __set_bit(idx, base->pending_map);
540 timer_set_idx(timer, idx);
544 __internal_add_timer(struct timer_base *base, struct timer_list *timer)
548 idx = calc_wheel_index(timer->expires, base->clk);
549 enqueue_timer(base, timer, idx);
553 trigger_dyntick_cpu(struct timer_base *base, struct timer_list *timer)
555 if (!is_timers_nohz_active())
559 * TODO: This wants some optimizing similar to the code below, but we
560 * will do that when we switch from push to pull for deferrable timers.
562 if (timer->flags & TIMER_DEFERRABLE) {
563 if (tick_nohz_full_cpu(base->cpu))
564 wake_up_nohz_cpu(base->cpu);
569 * We might have to IPI the remote CPU if the base is idle and the
570 * timer is not deferrable. If the other CPU is on the way to idle
571 * then it can't set base->is_idle as we hold the base lock:
576 /* Check whether this is the new first expiring timer: */
577 if (time_after_eq(timer->expires, base->next_expiry))
581 * Set the next expiry time and kick the CPU so it can reevaluate the
584 if (time_before(timer->expires, base->clk)) {
586 * Prevent from forward_timer_base() moving the base->clk
589 base->next_expiry = base->clk;
591 base->next_expiry = timer->expires;
593 wake_up_nohz_cpu(base->cpu);
597 internal_add_timer(struct timer_base *base, struct timer_list *timer)
599 __internal_add_timer(base, timer);
600 trigger_dyntick_cpu(base, timer);
603 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
605 static struct debug_obj_descr timer_debug_descr;
607 static void *timer_debug_hint(void *addr)
609 return ((struct timer_list *) addr)->function;
612 static bool timer_is_static_object(void *addr)
614 struct timer_list *timer = addr;
616 return (timer->entry.pprev == NULL &&
617 timer->entry.next == TIMER_ENTRY_STATIC);
621 * fixup_init is called when:
622 * - an active object is initialized
624 static bool timer_fixup_init(void *addr, enum debug_obj_state state)
626 struct timer_list *timer = addr;
629 case ODEBUG_STATE_ACTIVE:
630 del_timer_sync(timer);
631 debug_object_init(timer, &timer_debug_descr);
638 /* Stub timer callback for improperly used timers. */
639 static void stub_timer(struct timer_list *unused)
645 * fixup_activate is called when:
646 * - an active object is activated
647 * - an unknown non-static object is activated
649 static bool timer_fixup_activate(void *addr, enum debug_obj_state state)
651 struct timer_list *timer = addr;
654 case ODEBUG_STATE_NOTAVAILABLE:
655 timer_setup(timer, stub_timer, 0);
658 case ODEBUG_STATE_ACTIVE:
667 * fixup_free is called when:
668 * - an active object is freed
670 static bool timer_fixup_free(void *addr, enum debug_obj_state state)
672 struct timer_list *timer = addr;
675 case ODEBUG_STATE_ACTIVE:
676 del_timer_sync(timer);
677 debug_object_free(timer, &timer_debug_descr);
685 * fixup_assert_init is called when:
686 * - an untracked/uninit-ed object is found
688 static bool timer_fixup_assert_init(void *addr, enum debug_obj_state state)
690 struct timer_list *timer = addr;
693 case ODEBUG_STATE_NOTAVAILABLE:
694 timer_setup(timer, stub_timer, 0);
701 static struct debug_obj_descr timer_debug_descr = {
702 .name = "timer_list",
703 .debug_hint = timer_debug_hint,
704 .is_static_object = timer_is_static_object,
705 .fixup_init = timer_fixup_init,
706 .fixup_activate = timer_fixup_activate,
707 .fixup_free = timer_fixup_free,
708 .fixup_assert_init = timer_fixup_assert_init,
711 static inline void debug_timer_init(struct timer_list *timer)
713 debug_object_init(timer, &timer_debug_descr);
716 static inline void debug_timer_activate(struct timer_list *timer)
718 debug_object_activate(timer, &timer_debug_descr);
721 static inline void debug_timer_deactivate(struct timer_list *timer)
723 debug_object_deactivate(timer, &timer_debug_descr);
726 static inline void debug_timer_free(struct timer_list *timer)
728 debug_object_free(timer, &timer_debug_descr);
731 static inline void debug_timer_assert_init(struct timer_list *timer)
733 debug_object_assert_init(timer, &timer_debug_descr);
736 static void do_init_timer(struct timer_list *timer,
737 void (*func)(struct timer_list *),
739 const char *name, struct lock_class_key *key);
741 void init_timer_on_stack_key(struct timer_list *timer,
742 void (*func)(struct timer_list *),
744 const char *name, struct lock_class_key *key)
746 debug_object_init_on_stack(timer, &timer_debug_descr);
747 do_init_timer(timer, func, flags, name, key);
749 EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
751 void destroy_timer_on_stack(struct timer_list *timer)
753 debug_object_free(timer, &timer_debug_descr);
755 EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
758 static inline void debug_timer_init(struct timer_list *timer) { }
759 static inline void debug_timer_activate(struct timer_list *timer) { }
760 static inline void debug_timer_deactivate(struct timer_list *timer) { }
761 static inline void debug_timer_assert_init(struct timer_list *timer) { }
764 static inline void debug_init(struct timer_list *timer)
766 debug_timer_init(timer);
767 trace_timer_init(timer);
771 debug_activate(struct timer_list *timer, unsigned long expires)
773 debug_timer_activate(timer);
774 trace_timer_start(timer, expires, timer->flags);
777 static inline void debug_deactivate(struct timer_list *timer)
779 debug_timer_deactivate(timer);
780 trace_timer_cancel(timer);
783 static inline void debug_assert_init(struct timer_list *timer)
785 debug_timer_assert_init(timer);
788 static void do_init_timer(struct timer_list *timer,
789 void (*func)(struct timer_list *),
791 const char *name, struct lock_class_key *key)
793 timer->entry.pprev = NULL;
794 timer->function = func;
795 timer->flags = flags | raw_smp_processor_id();
796 lockdep_init_map(&timer->lockdep_map, name, key, 0);
800 * init_timer_key - initialize a timer
801 * @timer: the timer to be initialized
802 * @func: timer callback function
803 * @flags: timer flags
804 * @name: name of the timer
805 * @key: lockdep class key of the fake lock used for tracking timer
806 * sync lock dependencies
808 * init_timer_key() must be done to a timer prior calling *any* of the
809 * other timer functions.
811 void init_timer_key(struct timer_list *timer,
812 void (*func)(struct timer_list *), unsigned int flags,
813 const char *name, struct lock_class_key *key)
816 do_init_timer(timer, func, flags, name, key);
818 EXPORT_SYMBOL(init_timer_key);
820 static inline void detach_timer(struct timer_list *timer, bool clear_pending)
822 struct hlist_node *entry = &timer->entry;
824 debug_deactivate(timer);
829 entry->next = LIST_POISON2;
832 static int detach_if_pending(struct timer_list *timer, struct timer_base *base,
835 unsigned idx = timer_get_idx(timer);
837 if (!timer_pending(timer))
840 if (hlist_is_singular_node(&timer->entry, base->vectors + idx))
841 __clear_bit(idx, base->pending_map);
843 detach_timer(timer, clear_pending);
847 static inline struct timer_base *get_timer_cpu_base(u32 tflags, u32 cpu)
849 struct timer_base *base = per_cpu_ptr(&timer_bases[BASE_STD], cpu);
852 * If the timer is deferrable and NO_HZ_COMMON is set then we need
853 * to use the deferrable base.
855 if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE))
856 base = per_cpu_ptr(&timer_bases[BASE_DEF], cpu);
860 static inline struct timer_base *get_timer_this_cpu_base(u32 tflags)
862 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
865 * If the timer is deferrable and NO_HZ_COMMON is set then we need
866 * to use the deferrable base.
868 if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE))
869 base = this_cpu_ptr(&timer_bases[BASE_DEF]);
873 static inline struct timer_base *get_timer_base(u32 tflags)
875 return get_timer_cpu_base(tflags, tflags & TIMER_CPUMASK);
878 static inline struct timer_base *
879 get_target_base(struct timer_base *base, unsigned tflags)
881 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
882 if (static_branch_likely(&timers_migration_enabled) &&
883 !(tflags & TIMER_PINNED))
884 return get_timer_cpu_base(tflags, get_nohz_timer_target());
886 return get_timer_this_cpu_base(tflags);
889 static inline void forward_timer_base(struct timer_base *base)
891 #ifdef CONFIG_NO_HZ_COMMON
895 * We only forward the base when we are idle or have just come out of
896 * idle (must_forward_clk logic), and have a delta between base clock
897 * and jiffies. In the common case, run_timers will take care of it.
899 if (likely(!base->must_forward_clk))
902 jnow = READ_ONCE(jiffies);
903 base->must_forward_clk = base->is_idle;
904 if ((long)(jnow - base->clk) < 2)
908 * If the next expiry value is > jiffies, then we fast forward to
909 * jiffies otherwise we forward to the next expiry value.
911 if (time_after(base->next_expiry, jnow)) {
914 if (WARN_ON_ONCE(time_before(base->next_expiry, base->clk)))
916 base->clk = base->next_expiry;
923 * We are using hashed locking: Holding per_cpu(timer_bases[x]).lock means
924 * that all timers which are tied to this base are locked, and the base itself
927 * So __run_timers/migrate_timers can safely modify all timers which could
928 * be found in the base->vectors array.
930 * When a timer is migrating then the TIMER_MIGRATING flag is set and we need
931 * to wait until the migration is done.
933 static struct timer_base *lock_timer_base(struct timer_list *timer,
934 unsigned long *flags)
935 __acquires(timer->base->lock)
938 struct timer_base *base;
942 * We need to use READ_ONCE() here, otherwise the compiler
943 * might re-read @tf between the check for TIMER_MIGRATING
946 tf = READ_ONCE(timer->flags);
948 if (!(tf & TIMER_MIGRATING)) {
949 base = get_timer_base(tf);
950 raw_spin_lock_irqsave(&base->lock, *flags);
951 if (timer->flags == tf)
953 raw_spin_unlock_irqrestore(&base->lock, *flags);
959 #define MOD_TIMER_PENDING_ONLY 0x01
960 #define MOD_TIMER_REDUCE 0x02
963 __mod_timer(struct timer_list *timer, unsigned long expires, unsigned int options)
965 struct timer_base *base, *new_base;
966 unsigned int idx = UINT_MAX;
967 unsigned long clk = 0, flags;
970 BUG_ON(!timer->function);
973 * This is a common optimization triggered by the networking code - if
974 * the timer is re-modified to have the same timeout or ends up in the
975 * same array bucket then just return:
977 if (timer_pending(timer)) {
979 * The downside of this optimization is that it can result in
980 * larger granularity than you would get from adding a new
981 * timer with this expiry.
983 long diff = timer->expires - expires;
987 if (options & MOD_TIMER_REDUCE && diff <= 0)
991 * We lock timer base and calculate the bucket index right
992 * here. If the timer ends up in the same bucket, then we
993 * just update the expiry time and avoid the whole
994 * dequeue/enqueue dance.
996 base = lock_timer_base(timer, &flags);
997 forward_timer_base(base);
999 if (timer_pending(timer) && (options & MOD_TIMER_REDUCE) &&
1000 time_before_eq(timer->expires, expires)) {
1006 idx = calc_wheel_index(expires, clk);
1009 * Retrieve and compare the array index of the pending
1010 * timer. If it matches set the expiry to the new value so a
1011 * subsequent call will exit in the expires check above.
1013 if (idx == timer_get_idx(timer)) {
1014 if (!(options & MOD_TIMER_REDUCE))
1015 timer->expires = expires;
1016 else if (time_after(timer->expires, expires))
1017 timer->expires = expires;
1022 base = lock_timer_base(timer, &flags);
1023 forward_timer_base(base);
1026 ret = detach_if_pending(timer, base, false);
1027 if (!ret && (options & MOD_TIMER_PENDING_ONLY))
1030 new_base = get_target_base(base, timer->flags);
1032 if (base != new_base) {
1034 * We are trying to schedule the timer on the new base.
1035 * However we can't change timer's base while it is running,
1036 * otherwise del_timer_sync() can't detect that the timer's
1037 * handler yet has not finished. This also guarantees that the
1038 * timer is serialized wrt itself.
1040 if (likely(base->running_timer != timer)) {
1041 /* See the comment in lock_timer_base() */
1042 timer->flags |= TIMER_MIGRATING;
1044 raw_spin_unlock(&base->lock);
1046 raw_spin_lock(&base->lock);
1047 WRITE_ONCE(timer->flags,
1048 (timer->flags & ~TIMER_BASEMASK) | base->cpu);
1049 forward_timer_base(base);
1053 debug_activate(timer, expires);
1055 timer->expires = expires;
1057 * If 'idx' was calculated above and the base time did not advance
1058 * between calculating 'idx' and possibly switching the base, only
1059 * enqueue_timer() and trigger_dyntick_cpu() is required. Otherwise
1060 * we need to (re)calculate the wheel index via
1061 * internal_add_timer().
1063 if (idx != UINT_MAX && clk == base->clk) {
1064 enqueue_timer(base, timer, idx);
1065 trigger_dyntick_cpu(base, timer);
1067 internal_add_timer(base, timer);
1071 raw_spin_unlock_irqrestore(&base->lock, flags);
1077 * mod_timer_pending - modify a pending timer's timeout
1078 * @timer: the pending timer to be modified
1079 * @expires: new timeout in jiffies
1081 * mod_timer_pending() is the same for pending timers as mod_timer(),
1082 * but will not re-activate and modify already deleted timers.
1084 * It is useful for unserialized use of timers.
1086 int mod_timer_pending(struct timer_list *timer, unsigned long expires)
1088 return __mod_timer(timer, expires, MOD_TIMER_PENDING_ONLY);
1090 EXPORT_SYMBOL(mod_timer_pending);
1093 * mod_timer - modify a timer's timeout
1094 * @timer: the timer to be modified
1095 * @expires: new timeout in jiffies
1097 * mod_timer() is a more efficient way to update the expire field of an
1098 * active timer (if the timer is inactive it will be activated)
1100 * mod_timer(timer, expires) is equivalent to:
1102 * del_timer(timer); timer->expires = expires; add_timer(timer);
1104 * Note that if there are multiple unserialized concurrent users of the
1105 * same timer, then mod_timer() is the only safe way to modify the timeout,
1106 * since add_timer() cannot modify an already running timer.
1108 * The function returns whether it has modified a pending timer or not.
1109 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
1110 * active timer returns 1.)
1112 int mod_timer(struct timer_list *timer, unsigned long expires)
1114 return __mod_timer(timer, expires, 0);
1116 EXPORT_SYMBOL(mod_timer);
1119 * timer_reduce - Modify a timer's timeout if it would reduce the timeout
1120 * @timer: The timer to be modified
1121 * @expires: New timeout in jiffies
1123 * timer_reduce() is very similar to mod_timer(), except that it will only
1124 * modify a running timer if that would reduce the expiration time (it will
1125 * start a timer that isn't running).
1127 int timer_reduce(struct timer_list *timer, unsigned long expires)
1129 return __mod_timer(timer, expires, MOD_TIMER_REDUCE);
1131 EXPORT_SYMBOL(timer_reduce);
1134 * add_timer - start a timer
1135 * @timer: the timer to be added
1137 * The kernel will do a ->function(@timer) callback from the
1138 * timer interrupt at the ->expires point in the future. The
1139 * current time is 'jiffies'.
1141 * The timer's ->expires, ->function fields must be set prior calling this
1144 * Timers with an ->expires field in the past will be executed in the next
1147 void add_timer(struct timer_list *timer)
1149 BUG_ON(timer_pending(timer));
1150 mod_timer(timer, timer->expires);
1152 EXPORT_SYMBOL(add_timer);
1155 * add_timer_on - start a timer on a particular CPU
1156 * @timer: the timer to be added
1157 * @cpu: the CPU to start it on
1159 * This is not very scalable on SMP. Double adds are not possible.
1161 void add_timer_on(struct timer_list *timer, int cpu)
1163 struct timer_base *new_base, *base;
1164 unsigned long flags;
1166 BUG_ON(timer_pending(timer) || !timer->function);
1168 new_base = get_timer_cpu_base(timer->flags, cpu);
1171 * If @timer was on a different CPU, it should be migrated with the
1172 * old base locked to prevent other operations proceeding with the
1173 * wrong base locked. See lock_timer_base().
1175 base = lock_timer_base(timer, &flags);
1176 if (base != new_base) {
1177 timer->flags |= TIMER_MIGRATING;
1179 raw_spin_unlock(&base->lock);
1181 raw_spin_lock(&base->lock);
1182 WRITE_ONCE(timer->flags,
1183 (timer->flags & ~TIMER_BASEMASK) | cpu);
1185 forward_timer_base(base);
1187 debug_activate(timer, timer->expires);
1188 internal_add_timer(base, timer);
1189 raw_spin_unlock_irqrestore(&base->lock, flags);
1191 EXPORT_SYMBOL_GPL(add_timer_on);
1194 * del_timer - deactivate a timer.
1195 * @timer: the timer to be deactivated
1197 * del_timer() deactivates a timer - this works on both active and inactive
1200 * The function returns whether it has deactivated a pending timer or not.
1201 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
1202 * active timer returns 1.)
1204 int del_timer(struct timer_list *timer)
1206 struct timer_base *base;
1207 unsigned long flags;
1210 debug_assert_init(timer);
1212 if (timer_pending(timer)) {
1213 base = lock_timer_base(timer, &flags);
1214 ret = detach_if_pending(timer, base, true);
1215 raw_spin_unlock_irqrestore(&base->lock, flags);
1220 EXPORT_SYMBOL(del_timer);
1223 * try_to_del_timer_sync - Try to deactivate a timer
1224 * @timer: timer to delete
1226 * This function tries to deactivate a timer. Upon successful (ret >= 0)
1227 * exit the timer is not queued and the handler is not running on any CPU.
1229 int try_to_del_timer_sync(struct timer_list *timer)
1231 struct timer_base *base;
1232 unsigned long flags;
1235 debug_assert_init(timer);
1237 base = lock_timer_base(timer, &flags);
1239 if (base->running_timer != timer)
1240 ret = detach_if_pending(timer, base, true);
1242 raw_spin_unlock_irqrestore(&base->lock, flags);
1246 EXPORT_SYMBOL(try_to_del_timer_sync);
1250 * del_timer_sync - deactivate a timer and wait for the handler to finish.
1251 * @timer: the timer to be deactivated
1253 * This function only differs from del_timer() on SMP: besides deactivating
1254 * the timer it also makes sure the handler has finished executing on other
1257 * Synchronization rules: Callers must prevent restarting of the timer,
1258 * otherwise this function is meaningless. It must not be called from
1259 * interrupt contexts unless the timer is an irqsafe one. The caller must
1260 * not hold locks which would prevent completion of the timer's
1261 * handler. The timer's handler must not call add_timer_on(). Upon exit the
1262 * timer is not queued and the handler is not running on any CPU.
1264 * Note: For !irqsafe timers, you must not hold locks that are held in
1265 * interrupt context while calling this function. Even if the lock has
1266 * nothing to do with the timer in question. Here's why::
1272 * base->running_timer = mytimer;
1273 * spin_lock_irq(somelock);
1275 * spin_lock(somelock);
1276 * del_timer_sync(mytimer);
1277 * while (base->running_timer == mytimer);
1279 * Now del_timer_sync() will never return and never release somelock.
1280 * The interrupt on the other CPU is waiting to grab somelock but
1281 * it has interrupted the softirq that CPU0 is waiting to finish.
1283 * The function returns whether it has deactivated a pending timer or not.
1285 int del_timer_sync(struct timer_list *timer)
1287 #ifdef CONFIG_LOCKDEP
1288 unsigned long flags;
1291 * If lockdep gives a backtrace here, please reference
1292 * the synchronization rules above.
1294 local_irq_save(flags);
1295 lock_map_acquire(&timer->lockdep_map);
1296 lock_map_release(&timer->lockdep_map);
1297 local_irq_restore(flags);
1300 * don't use it in hardirq context, because it
1301 * could lead to deadlock.
1303 WARN_ON(in_irq() && !(timer->flags & TIMER_IRQSAFE));
1305 int ret = try_to_del_timer_sync(timer);
1311 EXPORT_SYMBOL(del_timer_sync);
1314 static void call_timer_fn(struct timer_list *timer, void (*fn)(struct timer_list *))
1316 int count = preempt_count();
1318 #ifdef CONFIG_LOCKDEP
1320 * It is permissible to free the timer from inside the
1321 * function that is called from it, this we need to take into
1322 * account for lockdep too. To avoid bogus "held lock freed"
1323 * warnings as well as problems when looking into
1324 * timer->lockdep_map, make a copy and use that here.
1326 struct lockdep_map lockdep_map;
1328 lockdep_copy_map(&lockdep_map, &timer->lockdep_map);
1331 * Couple the lock chain with the lock chain at
1332 * del_timer_sync() by acquiring the lock_map around the fn()
1333 * call here and in del_timer_sync().
1335 lock_map_acquire(&lockdep_map);
1337 trace_timer_expire_entry(timer);
1339 trace_timer_expire_exit(timer);
1341 lock_map_release(&lockdep_map);
1343 if (count != preempt_count()) {
1344 WARN_ONCE(1, "timer: %pF preempt leak: %08x -> %08x\n",
1345 fn, count, preempt_count());
1347 * Restore the preempt count. That gives us a decent
1348 * chance to survive and extract information. If the
1349 * callback kept a lock held, bad luck, but not worse
1350 * than the BUG() we had.
1352 preempt_count_set(count);
1356 static void expire_timers(struct timer_base *base, struct hlist_head *head)
1358 while (!hlist_empty(head)) {
1359 struct timer_list *timer;
1360 void (*fn)(struct timer_list *);
1362 timer = hlist_entry(head->first, struct timer_list, entry);
1364 base->running_timer = timer;
1365 detach_timer(timer, true);
1367 fn = timer->function;
1369 if (timer->flags & TIMER_IRQSAFE) {
1370 raw_spin_unlock(&base->lock);
1371 call_timer_fn(timer, fn);
1372 raw_spin_lock(&base->lock);
1374 raw_spin_unlock_irq(&base->lock);
1375 call_timer_fn(timer, fn);
1376 raw_spin_lock_irq(&base->lock);
1381 static int __collect_expired_timers(struct timer_base *base,
1382 struct hlist_head *heads)
1384 unsigned long clk = base->clk;
1385 struct hlist_head *vec;
1389 for (i = 0; i < LVL_DEPTH; i++) {
1390 idx = (clk & LVL_MASK) + i * LVL_SIZE;
1392 if (__test_and_clear_bit(idx, base->pending_map)) {
1393 vec = base->vectors + idx;
1394 hlist_move_list(vec, heads++);
1397 /* Is it time to look at the next level? */
1398 if (clk & LVL_CLK_MASK)
1400 /* Shift clock for the next level granularity */
1401 clk >>= LVL_CLK_SHIFT;
1406 #ifdef CONFIG_NO_HZ_COMMON
1408 * Find the next pending bucket of a level. Search from level start (@offset)
1409 * + @clk upwards and if nothing there, search from start of the level
1410 * (@offset) up to @offset + clk.
1412 static int next_pending_bucket(struct timer_base *base, unsigned offset,
1415 unsigned pos, start = offset + clk;
1416 unsigned end = offset + LVL_SIZE;
1418 pos = find_next_bit(base->pending_map, end, start);
1422 pos = find_next_bit(base->pending_map, start, offset);
1423 return pos < start ? pos + LVL_SIZE - start : -1;
1427 * Search the first expiring timer in the various clock levels. Caller must
1430 static unsigned long __next_timer_interrupt(struct timer_base *base)
1432 unsigned long clk, next, adj;
1433 unsigned lvl, offset = 0;
1435 next = base->clk + NEXT_TIMER_MAX_DELTA;
1437 for (lvl = 0; lvl < LVL_DEPTH; lvl++, offset += LVL_SIZE) {
1438 int pos = next_pending_bucket(base, offset, clk & LVL_MASK);
1441 unsigned long tmp = clk + (unsigned long) pos;
1443 tmp <<= LVL_SHIFT(lvl);
1444 if (time_before(tmp, next))
1448 * Clock for the next level. If the current level clock lower
1449 * bits are zero, we look at the next level as is. If not we
1450 * need to advance it by one because that's going to be the
1451 * next expiring bucket in that level. base->clk is the next
1452 * expiring jiffie. So in case of:
1454 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1457 * we have to look at all levels @index 0. With
1459 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1462 * LVL0 has the next expiring bucket @index 2. The upper
1463 * levels have the next expiring bucket @index 1.
1465 * In case that the propagation wraps the next level the same
1468 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1471 * So after looking at LVL0 we get:
1473 * LVL5 LVL4 LVL3 LVL2 LVL1
1476 * So no propagation from LVL1 to LVL2 because that happened
1477 * with the add already, but then we need to propagate further
1478 * from LVL2 to LVL3.
1480 * So the simple check whether the lower bits of the current
1481 * level are 0 or not is sufficient for all cases.
1483 adj = clk & LVL_CLK_MASK ? 1 : 0;
1484 clk >>= LVL_CLK_SHIFT;
1491 * Check, if the next hrtimer event is before the next timer wheel
1494 static u64 cmp_next_hrtimer_event(u64 basem, u64 expires)
1496 u64 nextevt = hrtimer_get_next_event();
1499 * If high resolution timers are enabled
1500 * hrtimer_get_next_event() returns KTIME_MAX.
1502 if (expires <= nextevt)
1506 * If the next timer is already expired, return the tick base
1507 * time so the tick is fired immediately.
1509 if (nextevt <= basem)
1513 * Round up to the next jiffie. High resolution timers are
1514 * off, so the hrtimers are expired in the tick and we need to
1515 * make sure that this tick really expires the timer to avoid
1516 * a ping pong of the nohz stop code.
1518 * Use DIV_ROUND_UP_ULL to prevent gcc calling __divdi3
1520 return DIV_ROUND_UP_ULL(nextevt, TICK_NSEC) * TICK_NSEC;
1524 * get_next_timer_interrupt - return the time (clock mono) of the next timer
1525 * @basej: base time jiffies
1526 * @basem: base time clock monotonic
1528 * Returns the tick aligned clock monotonic time of the next pending
1529 * timer or KTIME_MAX if no timer is pending.
1531 u64 get_next_timer_interrupt(unsigned long basej, u64 basem)
1533 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1534 u64 expires = KTIME_MAX;
1535 unsigned long nextevt;
1539 * Pretend that there is no timer pending if the cpu is offline.
1540 * Possible pending timers will be migrated later to an active cpu.
1542 if (cpu_is_offline(smp_processor_id()))
1545 raw_spin_lock(&base->lock);
1546 nextevt = __next_timer_interrupt(base);
1547 is_max_delta = (nextevt == base->clk + NEXT_TIMER_MAX_DELTA);
1548 base->next_expiry = nextevt;
1550 * We have a fresh next event. Check whether we can forward the
1551 * base. We can only do that when @basej is past base->clk
1552 * otherwise we might rewind base->clk.
1554 if (time_after(basej, base->clk)) {
1555 if (time_after(nextevt, basej))
1557 else if (time_after(nextevt, base->clk))
1558 base->clk = nextevt;
1561 if (time_before_eq(nextevt, basej)) {
1563 base->is_idle = false;
1566 expires = basem + (u64)(nextevt - basej) * TICK_NSEC;
1568 * If we expect to sleep more than a tick, mark the base idle.
1569 * Also the tick is stopped so any added timer must forward
1570 * the base clk itself to keep granularity small. This idle
1571 * logic is only maintained for the BASE_STD base, deferrable
1572 * timers may still see large granularity skew (by design).
1574 if ((expires - basem) > TICK_NSEC) {
1575 base->must_forward_clk = true;
1576 base->is_idle = true;
1579 raw_spin_unlock(&base->lock);
1581 return cmp_next_hrtimer_event(basem, expires);
1585 * timer_clear_idle - Clear the idle state of the timer base
1587 * Called with interrupts disabled
1589 void timer_clear_idle(void)
1591 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1594 * We do this unlocked. The worst outcome is a remote enqueue sending
1595 * a pointless IPI, but taking the lock would just make the window for
1596 * sending the IPI a few instructions smaller for the cost of taking
1597 * the lock in the exit from idle path.
1599 base->is_idle = false;
1602 static int collect_expired_timers(struct timer_base *base,
1603 struct hlist_head *heads)
1605 unsigned long now = READ_ONCE(jiffies);
1608 * NOHZ optimization. After a long idle sleep we need to forward the
1609 * base to current jiffies. Avoid a loop by searching the bitfield for
1610 * the next expiring timer.
1612 if ((long)(now - base->clk) > 2) {
1613 unsigned long next = __next_timer_interrupt(base);
1616 * If the next timer is ahead of time forward to current
1617 * jiffies, otherwise forward to the next expiry time:
1619 if (time_after(next, now)) {
1621 * The call site will increment base->clk and then
1622 * terminate the expiry loop immediately.
1629 return __collect_expired_timers(base, heads);
1632 static inline int collect_expired_timers(struct timer_base *base,
1633 struct hlist_head *heads)
1635 return __collect_expired_timers(base, heads);
1640 * Called from the timer interrupt handler to charge one tick to the current
1641 * process. user_tick is 1 if the tick is user time, 0 for system.
1643 void update_process_times(int user_tick)
1645 struct task_struct *p = current;
1647 /* Note: this timer irq context must be accounted for as well. */
1648 account_process_tick(p, user_tick);
1650 rcu_check_callbacks(user_tick);
1651 #ifdef CONFIG_IRQ_WORK
1656 if (IS_ENABLED(CONFIG_POSIX_TIMERS))
1657 run_posix_cpu_timers(p);
1661 * __run_timers - run all expired timers (if any) on this CPU.
1662 * @base: the timer vector to be processed.
1664 static inline void __run_timers(struct timer_base *base)
1666 struct hlist_head heads[LVL_DEPTH];
1669 if (!time_after_eq(jiffies, base->clk))
1672 raw_spin_lock_irq(&base->lock);
1675 * timer_base::must_forward_clk must be cleared before running
1676 * timers so that any timer functions that call mod_timer() will
1677 * not try to forward the base. Idle tracking / clock forwarding
1678 * logic is only used with BASE_STD timers.
1680 * The must_forward_clk flag is cleared unconditionally also for
1681 * the deferrable base. The deferrable base is not affected by idle
1682 * tracking and never forwarded, so clearing the flag is a NOOP.
1684 * The fact that the deferrable base is never forwarded can cause
1685 * large variations in granularity for deferrable timers, but they
1686 * can be deferred for long periods due to idle anyway.
1688 base->must_forward_clk = false;
1690 while (time_after_eq(jiffies, base->clk)) {
1692 levels = collect_expired_timers(base, heads);
1696 expire_timers(base, heads + levels);
1698 base->running_timer = NULL;
1699 raw_spin_unlock_irq(&base->lock);
1703 * This function runs timers and the timer-tq in bottom half context.
1705 static __latent_entropy void run_timer_softirq(struct softirq_action *h)
1707 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1710 if (IS_ENABLED(CONFIG_NO_HZ_COMMON))
1711 __run_timers(this_cpu_ptr(&timer_bases[BASE_DEF]));
1715 * Called by the local, per-CPU timer interrupt on SMP.
1717 void run_local_timers(void)
1719 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1721 hrtimer_run_queues();
1722 /* Raise the softirq only if required. */
1723 if (time_before(jiffies, base->clk)) {
1724 if (!IS_ENABLED(CONFIG_NO_HZ_COMMON))
1726 /* CPU is awake, so check the deferrable base. */
1728 if (time_before(jiffies, base->clk))
1731 raise_softirq(TIMER_SOFTIRQ);
1735 * Since schedule_timeout()'s timer is defined on the stack, it must store
1736 * the target task on the stack as well.
1738 struct process_timer {
1739 struct timer_list timer;
1740 struct task_struct *task;
1743 static void process_timeout(struct timer_list *t)
1745 struct process_timer *timeout = from_timer(timeout, t, timer);
1747 wake_up_process(timeout->task);
1751 * schedule_timeout - sleep until timeout
1752 * @timeout: timeout value in jiffies
1754 * Make the current task sleep until @timeout jiffies have
1755 * elapsed. The routine will return immediately unless
1756 * the current task state has been set (see set_current_state()).
1758 * You can set the task state as follows -
1760 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1761 * pass before the routine returns unless the current task is explicitly
1762 * woken up, (e.g. by wake_up_process())".
1764 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1765 * delivered to the current task or the current task is explicitly woken
1768 * The current task state is guaranteed to be TASK_RUNNING when this
1771 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1772 * the CPU away without a bound on the timeout. In this case the return
1773 * value will be %MAX_SCHEDULE_TIMEOUT.
1775 * Returns 0 when the timer has expired otherwise the remaining time in
1776 * jiffies will be returned. In all cases the return value is guaranteed
1777 * to be non-negative.
1779 signed long __sched schedule_timeout(signed long timeout)
1781 struct process_timer timer;
1782 unsigned long expire;
1786 case MAX_SCHEDULE_TIMEOUT:
1788 * These two special cases are useful to be comfortable
1789 * in the caller. Nothing more. We could take
1790 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1791 * but I' d like to return a valid offset (>=0) to allow
1792 * the caller to do everything it want with the retval.
1798 * Another bit of PARANOID. Note that the retval will be
1799 * 0 since no piece of kernel is supposed to do a check
1800 * for a negative retval of schedule_timeout() (since it
1801 * should never happens anyway). You just have the printk()
1802 * that will tell you if something is gone wrong and where.
1805 printk(KERN_ERR "schedule_timeout: wrong timeout "
1806 "value %lx\n", timeout);
1808 current->state = TASK_RUNNING;
1813 expire = timeout + jiffies;
1815 timer.task = current;
1816 timer_setup_on_stack(&timer.timer, process_timeout, 0);
1817 __mod_timer(&timer.timer, expire, 0);
1819 del_singleshot_timer_sync(&timer.timer);
1821 /* Remove the timer from the object tracker */
1822 destroy_timer_on_stack(&timer.timer);
1824 timeout = expire - jiffies;
1827 return timeout < 0 ? 0 : timeout;
1829 EXPORT_SYMBOL(schedule_timeout);
1832 * We can use __set_current_state() here because schedule_timeout() calls
1833 * schedule() unconditionally.
1835 signed long __sched schedule_timeout_interruptible(signed long timeout)
1837 __set_current_state(TASK_INTERRUPTIBLE);
1838 return schedule_timeout(timeout);
1840 EXPORT_SYMBOL(schedule_timeout_interruptible);
1842 signed long __sched schedule_timeout_killable(signed long timeout)
1844 __set_current_state(TASK_KILLABLE);
1845 return schedule_timeout(timeout);
1847 EXPORT_SYMBOL(schedule_timeout_killable);
1849 signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1851 __set_current_state(TASK_UNINTERRUPTIBLE);
1852 return schedule_timeout(timeout);
1854 EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1857 * Like schedule_timeout_uninterruptible(), except this task will not contribute
1860 signed long __sched schedule_timeout_idle(signed long timeout)
1862 __set_current_state(TASK_IDLE);
1863 return schedule_timeout(timeout);
1865 EXPORT_SYMBOL(schedule_timeout_idle);
1867 #ifdef CONFIG_HOTPLUG_CPU
1868 static void migrate_timer_list(struct timer_base *new_base, struct hlist_head *head)
1870 struct timer_list *timer;
1871 int cpu = new_base->cpu;
1873 while (!hlist_empty(head)) {
1874 timer = hlist_entry(head->first, struct timer_list, entry);
1875 detach_timer(timer, false);
1876 timer->flags = (timer->flags & ~TIMER_BASEMASK) | cpu;
1877 internal_add_timer(new_base, timer);
1881 int timers_prepare_cpu(unsigned int cpu)
1883 struct timer_base *base;
1886 for (b = 0; b < NR_BASES; b++) {
1887 base = per_cpu_ptr(&timer_bases[b], cpu);
1888 base->clk = jiffies;
1889 base->next_expiry = base->clk + NEXT_TIMER_MAX_DELTA;
1890 base->is_idle = false;
1891 base->must_forward_clk = true;
1896 int timers_dead_cpu(unsigned int cpu)
1898 struct timer_base *old_base;
1899 struct timer_base *new_base;
1902 BUG_ON(cpu_online(cpu));
1904 for (b = 0; b < NR_BASES; b++) {
1905 old_base = per_cpu_ptr(&timer_bases[b], cpu);
1906 new_base = get_cpu_ptr(&timer_bases[b]);
1908 * The caller is globally serialized and nobody else
1909 * takes two locks at once, deadlock is not possible.
1911 raw_spin_lock_irq(&new_base->lock);
1912 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1915 * The current CPUs base clock might be stale. Update it
1916 * before moving the timers over.
1918 forward_timer_base(new_base);
1920 BUG_ON(old_base->running_timer);
1922 for (i = 0; i < WHEEL_SIZE; i++)
1923 migrate_timer_list(new_base, old_base->vectors + i);
1925 raw_spin_unlock(&old_base->lock);
1926 raw_spin_unlock_irq(&new_base->lock);
1927 put_cpu_ptr(&timer_bases);
1932 #endif /* CONFIG_HOTPLUG_CPU */
1934 static void __init init_timer_cpu(int cpu)
1936 struct timer_base *base;
1939 for (i = 0; i < NR_BASES; i++) {
1940 base = per_cpu_ptr(&timer_bases[i], cpu);
1942 raw_spin_lock_init(&base->lock);
1943 base->clk = jiffies;
1947 static void __init init_timer_cpus(void)
1951 for_each_possible_cpu(cpu)
1952 init_timer_cpu(cpu);
1955 void __init init_timers(void)
1958 open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
1962 * msleep - sleep safely even with waitqueue interruptions
1963 * @msecs: Time in milliseconds to sleep for
1965 void msleep(unsigned int msecs)
1967 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1970 timeout = schedule_timeout_uninterruptible(timeout);
1973 EXPORT_SYMBOL(msleep);
1976 * msleep_interruptible - sleep waiting for signals
1977 * @msecs: Time in milliseconds to sleep for
1979 unsigned long msleep_interruptible(unsigned int msecs)
1981 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1983 while (timeout && !signal_pending(current))
1984 timeout = schedule_timeout_interruptible(timeout);
1985 return jiffies_to_msecs(timeout);
1988 EXPORT_SYMBOL(msleep_interruptible);
1991 * usleep_range - Sleep for an approximate time
1992 * @min: Minimum time in usecs to sleep
1993 * @max: Maximum time in usecs to sleep
1995 * In non-atomic context where the exact wakeup time is flexible, use
1996 * usleep_range() instead of udelay(). The sleep improves responsiveness
1997 * by avoiding the CPU-hogging busy-wait of udelay(), and the range reduces
1998 * power usage by allowing hrtimers to take advantage of an already-
1999 * scheduled interrupt instead of scheduling a new one just for this sleep.
2001 void __sched usleep_range(unsigned long min, unsigned long max)
2003 ktime_t exp = ktime_add_us(ktime_get(), min);
2004 u64 delta = (u64)(max - min) * NSEC_PER_USEC;
2007 __set_current_state(TASK_UNINTERRUPTIBLE);
2008 /* Do not return before the requested sleep time has elapsed */
2009 if (!schedule_hrtimeout_range(&exp, delta, HRTIMER_MODE_ABS))
2013 EXPORT_SYMBOL(usleep_range);