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;
204 bool migration_enabled;
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 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
215 unsigned int sysctl_timer_migration = 1;
217 void timers_update_migration(bool update_nohz)
219 bool on = sysctl_timer_migration && tick_nohz_active;
222 /* Avoid the loop, if nothing to update */
223 if (this_cpu_read(timer_bases[BASE_STD].migration_enabled) == on)
226 for_each_possible_cpu(cpu) {
227 per_cpu(timer_bases[BASE_STD].migration_enabled, cpu) = on;
228 per_cpu(timer_bases[BASE_DEF].migration_enabled, cpu) = on;
229 per_cpu(hrtimer_bases.migration_enabled, cpu) = on;
232 per_cpu(timer_bases[BASE_STD].nohz_active, cpu) = true;
233 per_cpu(timer_bases[BASE_DEF].nohz_active, cpu) = true;
234 per_cpu(hrtimer_bases.nohz_active, cpu) = true;
238 int timer_migration_handler(struct ctl_table *table, int write,
239 void __user *buffer, size_t *lenp,
242 static DEFINE_MUTEX(mutex);
246 ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
248 timers_update_migration(false);
249 mutex_unlock(&mutex);
254 static unsigned long round_jiffies_common(unsigned long j, int cpu,
258 unsigned long original = j;
261 * We don't want all cpus firing their timers at once hitting the
262 * same lock or cachelines, so we skew each extra cpu with an extra
263 * 3 jiffies. This 3 jiffies came originally from the mm/ code which
265 * The skew is done by adding 3*cpunr, then round, then subtract this
266 * extra offset again.
273 * If the target jiffie is just after a whole second (which can happen
274 * due to delays of the timer irq, long irq off times etc etc) then
275 * we should round down to the whole second, not up. Use 1/4th second
276 * as cutoff for this rounding as an extreme upper bound for this.
277 * But never round down if @force_up is set.
279 if (rem < HZ/4 && !force_up) /* round down */
284 /* now that we have rounded, subtract the extra skew again */
288 * Make sure j is still in the future. Otherwise return the
291 return time_is_after_jiffies(j) ? j : original;
295 * __round_jiffies - function to round jiffies to a full second
296 * @j: the time in (absolute) jiffies that should be rounded
297 * @cpu: the processor number on which the timeout will happen
299 * __round_jiffies() rounds an absolute time in the future (in jiffies)
300 * up or down to (approximately) full seconds. This is useful for timers
301 * for which the exact time they fire does not matter too much, as long as
302 * they fire approximately every X seconds.
304 * By rounding these timers to whole seconds, all such timers will fire
305 * at the same time, rather than at various times spread out. The goal
306 * of this is to have the CPU wake up less, which saves power.
308 * The exact rounding is skewed for each processor to avoid all
309 * processors firing at the exact same time, which could lead
310 * to lock contention or spurious cache line bouncing.
312 * The return value is the rounded version of the @j parameter.
314 unsigned long __round_jiffies(unsigned long j, int cpu)
316 return round_jiffies_common(j, cpu, false);
318 EXPORT_SYMBOL_GPL(__round_jiffies);
321 * __round_jiffies_relative - function to round jiffies to a full second
322 * @j: the time in (relative) jiffies that should be rounded
323 * @cpu: the processor number on which the timeout will happen
325 * __round_jiffies_relative() rounds a time delta in the future (in jiffies)
326 * up or down to (approximately) full seconds. This is useful for timers
327 * for which the exact time they fire does not matter too much, as long as
328 * they fire approximately every X seconds.
330 * By rounding these timers to whole seconds, all such timers will fire
331 * at the same time, rather than at various times spread out. The goal
332 * of this is to have the CPU wake up less, which saves power.
334 * The exact rounding is skewed for each processor to avoid all
335 * processors firing at the exact same time, which could lead
336 * to lock contention or spurious cache line bouncing.
338 * The return value is the rounded version of the @j parameter.
340 unsigned long __round_jiffies_relative(unsigned long j, int cpu)
342 unsigned long j0 = jiffies;
344 /* Use j0 because jiffies might change while we run */
345 return round_jiffies_common(j + j0, cpu, false) - j0;
347 EXPORT_SYMBOL_GPL(__round_jiffies_relative);
350 * round_jiffies - function to round jiffies to a full second
351 * @j: the time in (absolute) jiffies that should be rounded
353 * round_jiffies() rounds an absolute time in the future (in jiffies)
354 * up or down to (approximately) full seconds. This is useful for timers
355 * for which the exact time they fire does not matter too much, as long as
356 * they fire approximately every X seconds.
358 * By rounding these timers to whole seconds, all such timers will fire
359 * at the same time, rather than at various times spread out. The goal
360 * of this is to have the CPU wake up less, which saves power.
362 * The return value is the rounded version of the @j parameter.
364 unsigned long round_jiffies(unsigned long j)
366 return round_jiffies_common(j, raw_smp_processor_id(), false);
368 EXPORT_SYMBOL_GPL(round_jiffies);
371 * round_jiffies_relative - function to round jiffies to a full second
372 * @j: the time in (relative) jiffies that should be rounded
374 * round_jiffies_relative() rounds a time delta in the future (in jiffies)
375 * up or down to (approximately) full seconds. This is useful for timers
376 * for which the exact time they fire does not matter too much, as long as
377 * they fire approximately every X seconds.
379 * By rounding these timers to whole seconds, all such timers will fire
380 * at the same time, rather than at various times spread out. The goal
381 * of this is to have the CPU wake up less, which saves power.
383 * The return value is the rounded version of the @j parameter.
385 unsigned long round_jiffies_relative(unsigned long j)
387 return __round_jiffies_relative(j, raw_smp_processor_id());
389 EXPORT_SYMBOL_GPL(round_jiffies_relative);
392 * __round_jiffies_up - function to round jiffies up to a full second
393 * @j: the time in (absolute) jiffies that should be rounded
394 * @cpu: the processor number on which the timeout will happen
396 * This is the same as __round_jiffies() except that it will never
397 * round down. This is useful for timeouts for which the exact time
398 * of firing does not matter too much, as long as they don't fire too
401 unsigned long __round_jiffies_up(unsigned long j, int cpu)
403 return round_jiffies_common(j, cpu, true);
405 EXPORT_SYMBOL_GPL(__round_jiffies_up);
408 * __round_jiffies_up_relative - function to round jiffies up to a full second
409 * @j: the time in (relative) jiffies that should be rounded
410 * @cpu: the processor number on which the timeout will happen
412 * This is the same as __round_jiffies_relative() except that it will never
413 * round down. This is useful for timeouts for which the exact time
414 * of firing does not matter too much, as long as they don't fire too
417 unsigned long __round_jiffies_up_relative(unsigned long j, int cpu)
419 unsigned long j0 = jiffies;
421 /* Use j0 because jiffies might change while we run */
422 return round_jiffies_common(j + j0, cpu, true) - j0;
424 EXPORT_SYMBOL_GPL(__round_jiffies_up_relative);
427 * round_jiffies_up - function to round jiffies up to a full second
428 * @j: the time in (absolute) jiffies that should be rounded
430 * This is the same as round_jiffies() except that it will never
431 * round down. This is useful for timeouts for which the exact time
432 * of firing does not matter too much, as long as they don't fire too
435 unsigned long round_jiffies_up(unsigned long j)
437 return round_jiffies_common(j, raw_smp_processor_id(), true);
439 EXPORT_SYMBOL_GPL(round_jiffies_up);
442 * round_jiffies_up_relative - function to round jiffies up to a full second
443 * @j: the time in (relative) jiffies that should be rounded
445 * This is the same as round_jiffies_relative() except that it will never
446 * round down. This is useful for timeouts for which the exact time
447 * of firing does not matter too much, as long as they don't fire too
450 unsigned long round_jiffies_up_relative(unsigned long j)
452 return __round_jiffies_up_relative(j, raw_smp_processor_id());
454 EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
457 static inline unsigned int timer_get_idx(struct timer_list *timer)
459 return (timer->flags & TIMER_ARRAYMASK) >> TIMER_ARRAYSHIFT;
462 static inline void timer_set_idx(struct timer_list *timer, unsigned int idx)
464 timer->flags = (timer->flags & ~TIMER_ARRAYMASK) |
465 idx << TIMER_ARRAYSHIFT;
469 * Helper function to calculate the array index for a given expiry
472 static inline unsigned calc_index(unsigned expires, unsigned lvl)
474 expires = (expires + LVL_GRAN(lvl)) >> LVL_SHIFT(lvl);
475 return LVL_OFFS(lvl) + (expires & LVL_MASK);
478 static int calc_wheel_index(unsigned long expires, unsigned long clk)
480 unsigned long delta = expires - clk;
483 if (delta < LVL_START(1)) {
484 idx = calc_index(expires, 0);
485 } else if (delta < LVL_START(2)) {
486 idx = calc_index(expires, 1);
487 } else if (delta < LVL_START(3)) {
488 idx = calc_index(expires, 2);
489 } else if (delta < LVL_START(4)) {
490 idx = calc_index(expires, 3);
491 } else if (delta < LVL_START(5)) {
492 idx = calc_index(expires, 4);
493 } else if (delta < LVL_START(6)) {
494 idx = calc_index(expires, 5);
495 } else if (delta < LVL_START(7)) {
496 idx = calc_index(expires, 6);
497 } else if (LVL_DEPTH > 8 && delta < LVL_START(8)) {
498 idx = calc_index(expires, 7);
499 } else if ((long) delta < 0) {
500 idx = clk & LVL_MASK;
503 * Force expire obscene large timeouts to expire at the
504 * capacity limit of the wheel.
506 if (delta >= WHEEL_TIMEOUT_CUTOFF)
507 expires = clk + WHEEL_TIMEOUT_MAX;
509 idx = calc_index(expires, LVL_DEPTH - 1);
515 * Enqueue the timer into the hash bucket, mark it pending in
516 * the bitmap and store the index in the timer flags.
518 static void enqueue_timer(struct timer_base *base, struct timer_list *timer,
521 hlist_add_head(&timer->entry, base->vectors + idx);
522 __set_bit(idx, base->pending_map);
523 timer_set_idx(timer, idx);
527 __internal_add_timer(struct timer_base *base, struct timer_list *timer)
531 idx = calc_wheel_index(timer->expires, base->clk);
532 enqueue_timer(base, timer, idx);
536 trigger_dyntick_cpu(struct timer_base *base, struct timer_list *timer)
538 if (!IS_ENABLED(CONFIG_NO_HZ_COMMON) || !base->nohz_active)
542 * TODO: This wants some optimizing similar to the code below, but we
543 * will do that when we switch from push to pull for deferrable timers.
545 if (timer->flags & TIMER_DEFERRABLE) {
546 if (tick_nohz_full_cpu(base->cpu))
547 wake_up_nohz_cpu(base->cpu);
552 * We might have to IPI the remote CPU if the base is idle and the
553 * timer is not deferrable. If the other CPU is on the way to idle
554 * then it can't set base->is_idle as we hold the base lock:
559 /* Check whether this is the new first expiring timer: */
560 if (time_after_eq(timer->expires, base->next_expiry))
564 * Set the next expiry time and kick the CPU so it can reevaluate the
567 base->next_expiry = timer->expires;
568 wake_up_nohz_cpu(base->cpu);
572 internal_add_timer(struct timer_base *base, struct timer_list *timer)
574 __internal_add_timer(base, timer);
575 trigger_dyntick_cpu(base, timer);
578 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
580 static struct debug_obj_descr timer_debug_descr;
582 static void *timer_debug_hint(void *addr)
584 return ((struct timer_list *) addr)->function;
587 static bool timer_is_static_object(void *addr)
589 struct timer_list *timer = addr;
591 return (timer->entry.pprev == NULL &&
592 timer->entry.next == TIMER_ENTRY_STATIC);
596 * fixup_init is called when:
597 * - an active object is initialized
599 static bool timer_fixup_init(void *addr, enum debug_obj_state state)
601 struct timer_list *timer = addr;
604 case ODEBUG_STATE_ACTIVE:
605 del_timer_sync(timer);
606 debug_object_init(timer, &timer_debug_descr);
613 /* Stub timer callback for improperly used timers. */
614 static void stub_timer(unsigned long data)
620 * fixup_activate is called when:
621 * - an active object is activated
622 * - an unknown non-static object is activated
624 static bool timer_fixup_activate(void *addr, enum debug_obj_state state)
626 struct timer_list *timer = addr;
629 case ODEBUG_STATE_NOTAVAILABLE:
630 setup_timer(timer, stub_timer, 0);
633 case ODEBUG_STATE_ACTIVE:
642 * fixup_free is called when:
643 * - an active object is freed
645 static bool timer_fixup_free(void *addr, enum debug_obj_state state)
647 struct timer_list *timer = addr;
650 case ODEBUG_STATE_ACTIVE:
651 del_timer_sync(timer);
652 debug_object_free(timer, &timer_debug_descr);
660 * fixup_assert_init is called when:
661 * - an untracked/uninit-ed object is found
663 static bool timer_fixup_assert_init(void *addr, enum debug_obj_state state)
665 struct timer_list *timer = addr;
668 case ODEBUG_STATE_NOTAVAILABLE:
669 setup_timer(timer, stub_timer, 0);
676 static struct debug_obj_descr timer_debug_descr = {
677 .name = "timer_list",
678 .debug_hint = timer_debug_hint,
679 .is_static_object = timer_is_static_object,
680 .fixup_init = timer_fixup_init,
681 .fixup_activate = timer_fixup_activate,
682 .fixup_free = timer_fixup_free,
683 .fixup_assert_init = timer_fixup_assert_init,
686 static inline void debug_timer_init(struct timer_list *timer)
688 debug_object_init(timer, &timer_debug_descr);
691 static inline void debug_timer_activate(struct timer_list *timer)
693 debug_object_activate(timer, &timer_debug_descr);
696 static inline void debug_timer_deactivate(struct timer_list *timer)
698 debug_object_deactivate(timer, &timer_debug_descr);
701 static inline void debug_timer_free(struct timer_list *timer)
703 debug_object_free(timer, &timer_debug_descr);
706 static inline void debug_timer_assert_init(struct timer_list *timer)
708 debug_object_assert_init(timer, &timer_debug_descr);
711 static void do_init_timer(struct timer_list *timer, unsigned int flags,
712 const char *name, struct lock_class_key *key);
714 void init_timer_on_stack_key(struct timer_list *timer, unsigned int flags,
715 const char *name, struct lock_class_key *key)
717 debug_object_init_on_stack(timer, &timer_debug_descr);
718 do_init_timer(timer, flags, name, key);
720 EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
722 void destroy_timer_on_stack(struct timer_list *timer)
724 debug_object_free(timer, &timer_debug_descr);
726 EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
729 static inline void debug_timer_init(struct timer_list *timer) { }
730 static inline void debug_timer_activate(struct timer_list *timer) { }
731 static inline void debug_timer_deactivate(struct timer_list *timer) { }
732 static inline void debug_timer_assert_init(struct timer_list *timer) { }
735 static inline void debug_init(struct timer_list *timer)
737 debug_timer_init(timer);
738 trace_timer_init(timer);
742 debug_activate(struct timer_list *timer, unsigned long expires)
744 debug_timer_activate(timer);
745 trace_timer_start(timer, expires, timer->flags);
748 static inline void debug_deactivate(struct timer_list *timer)
750 debug_timer_deactivate(timer);
751 trace_timer_cancel(timer);
754 static inline void debug_assert_init(struct timer_list *timer)
756 debug_timer_assert_init(timer);
759 static void do_init_timer(struct timer_list *timer, unsigned int flags,
760 const char *name, struct lock_class_key *key)
762 timer->entry.pprev = NULL;
763 timer->flags = flags | raw_smp_processor_id();
764 lockdep_init_map(&timer->lockdep_map, name, key, 0);
768 * init_timer_key - initialize a timer
769 * @timer: the timer to be initialized
770 * @flags: timer flags
771 * @name: name of the timer
772 * @key: lockdep class key of the fake lock used for tracking timer
773 * sync lock dependencies
775 * init_timer_key() must be done to a timer prior calling *any* of the
776 * other timer functions.
778 void init_timer_key(struct timer_list *timer, unsigned int flags,
779 const char *name, struct lock_class_key *key)
782 do_init_timer(timer, flags, name, key);
784 EXPORT_SYMBOL(init_timer_key);
786 static inline void detach_timer(struct timer_list *timer, bool clear_pending)
788 struct hlist_node *entry = &timer->entry;
790 debug_deactivate(timer);
795 entry->next = LIST_POISON2;
798 static int detach_if_pending(struct timer_list *timer, struct timer_base *base,
801 unsigned idx = timer_get_idx(timer);
803 if (!timer_pending(timer))
806 if (hlist_is_singular_node(&timer->entry, base->vectors + idx))
807 __clear_bit(idx, base->pending_map);
809 detach_timer(timer, clear_pending);
813 static inline struct timer_base *get_timer_cpu_base(u32 tflags, u32 cpu)
815 struct timer_base *base = per_cpu_ptr(&timer_bases[BASE_STD], cpu);
818 * If the timer is deferrable and NO_HZ_COMMON is set then we need
819 * to use the deferrable base.
821 if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE))
822 base = per_cpu_ptr(&timer_bases[BASE_DEF], cpu);
826 static inline struct timer_base *get_timer_this_cpu_base(u32 tflags)
828 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
831 * If the timer is deferrable and NO_HZ_COMMON is set then we need
832 * to use the deferrable base.
834 if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE))
835 base = this_cpu_ptr(&timer_bases[BASE_DEF]);
839 static inline struct timer_base *get_timer_base(u32 tflags)
841 return get_timer_cpu_base(tflags, tflags & TIMER_CPUMASK);
844 #ifdef CONFIG_NO_HZ_COMMON
845 static inline struct timer_base *
846 get_target_base(struct timer_base *base, unsigned tflags)
849 if ((tflags & TIMER_PINNED) || !base->migration_enabled)
850 return get_timer_this_cpu_base(tflags);
851 return get_timer_cpu_base(tflags, get_nohz_timer_target());
853 return get_timer_this_cpu_base(tflags);
857 static inline void forward_timer_base(struct timer_base *base)
862 * We only forward the base when we are idle or have just come out of
863 * idle (must_forward_clk logic), and have a delta between base clock
864 * and jiffies. In the common case, run_timers will take care of it.
866 if (likely(!base->must_forward_clk))
869 jnow = READ_ONCE(jiffies);
870 base->must_forward_clk = base->is_idle;
871 if ((long)(jnow - base->clk) < 2)
875 * If the next expiry value is > jiffies, then we fast forward to
876 * jiffies otherwise we forward to the next expiry value.
878 if (time_after(base->next_expiry, jnow))
881 base->clk = base->next_expiry;
884 static inline struct timer_base *
885 get_target_base(struct timer_base *base, unsigned tflags)
887 return get_timer_this_cpu_base(tflags);
890 static inline void forward_timer_base(struct timer_base *base) { }
895 * We are using hashed locking: Holding per_cpu(timer_bases[x]).lock means
896 * that all timers which are tied to this base are locked, and the base itself
899 * So __run_timers/migrate_timers can safely modify all timers which could
900 * be found in the base->vectors array.
902 * When a timer is migrating then the TIMER_MIGRATING flag is set and we need
903 * to wait until the migration is done.
905 static struct timer_base *lock_timer_base(struct timer_list *timer,
906 unsigned long *flags)
907 __acquires(timer->base->lock)
910 struct timer_base *base;
914 * We need to use READ_ONCE() here, otherwise the compiler
915 * might re-read @tf between the check for TIMER_MIGRATING
918 tf = READ_ONCE(timer->flags);
920 if (!(tf & TIMER_MIGRATING)) {
921 base = get_timer_base(tf);
922 raw_spin_lock_irqsave(&base->lock, *flags);
923 if (timer->flags == tf)
925 raw_spin_unlock_irqrestore(&base->lock, *flags);
932 __mod_timer(struct timer_list *timer, unsigned long expires, bool pending_only)
934 struct timer_base *base, *new_base;
935 unsigned int idx = UINT_MAX;
936 unsigned long clk = 0, flags;
939 BUG_ON(!timer->function);
942 * This is a common optimization triggered by the networking code - if
943 * the timer is re-modified to have the same timeout or ends up in the
944 * same array bucket then just return:
946 if (timer_pending(timer)) {
948 * The downside of this optimization is that it can result in
949 * larger granularity than you would get from adding a new
950 * timer with this expiry.
952 if (timer->expires == expires)
956 * We lock timer base and calculate the bucket index right
957 * here. If the timer ends up in the same bucket, then we
958 * just update the expiry time and avoid the whole
959 * dequeue/enqueue dance.
961 base = lock_timer_base(timer, &flags);
962 forward_timer_base(base);
965 idx = calc_wheel_index(expires, clk);
968 * Retrieve and compare the array index of the pending
969 * timer. If it matches set the expiry to the new value so a
970 * subsequent call will exit in the expires check above.
972 if (idx == timer_get_idx(timer)) {
973 timer->expires = expires;
978 base = lock_timer_base(timer, &flags);
979 forward_timer_base(base);
982 ret = detach_if_pending(timer, base, false);
983 if (!ret && pending_only)
986 new_base = get_target_base(base, timer->flags);
988 if (base != new_base) {
990 * We are trying to schedule the timer on the new base.
991 * However we can't change timer's base while it is running,
992 * otherwise del_timer_sync() can't detect that the timer's
993 * handler yet has not finished. This also guarantees that the
994 * timer is serialized wrt itself.
996 if (likely(base->running_timer != timer)) {
997 /* See the comment in lock_timer_base() */
998 timer->flags |= TIMER_MIGRATING;
1000 raw_spin_unlock(&base->lock);
1002 raw_spin_lock(&base->lock);
1003 WRITE_ONCE(timer->flags,
1004 (timer->flags & ~TIMER_BASEMASK) | base->cpu);
1005 forward_timer_base(base);
1009 debug_activate(timer, expires);
1011 timer->expires = expires;
1013 * If 'idx' was calculated above and the base time did not advance
1014 * between calculating 'idx' and possibly switching the base, only
1015 * enqueue_timer() and trigger_dyntick_cpu() is required. Otherwise
1016 * we need to (re)calculate the wheel index via
1017 * internal_add_timer().
1019 if (idx != UINT_MAX && clk == base->clk) {
1020 enqueue_timer(base, timer, idx);
1021 trigger_dyntick_cpu(base, timer);
1023 internal_add_timer(base, timer);
1027 raw_spin_unlock_irqrestore(&base->lock, flags);
1033 * mod_timer_pending - modify a pending timer's timeout
1034 * @timer: the pending timer to be modified
1035 * @expires: new timeout in jiffies
1037 * mod_timer_pending() is the same for pending timers as mod_timer(),
1038 * but will not re-activate and modify already deleted timers.
1040 * It is useful for unserialized use of timers.
1042 int mod_timer_pending(struct timer_list *timer, unsigned long expires)
1044 return __mod_timer(timer, expires, true);
1046 EXPORT_SYMBOL(mod_timer_pending);
1049 * mod_timer - modify a timer's timeout
1050 * @timer: the timer to be modified
1051 * @expires: new timeout in jiffies
1053 * mod_timer() is a more efficient way to update the expire field of an
1054 * active timer (if the timer is inactive it will be activated)
1056 * mod_timer(timer, expires) is equivalent to:
1058 * del_timer(timer); timer->expires = expires; add_timer(timer);
1060 * Note that if there are multiple unserialized concurrent users of the
1061 * same timer, then mod_timer() is the only safe way to modify the timeout,
1062 * since add_timer() cannot modify an already running timer.
1064 * The function returns whether it has modified a pending timer or not.
1065 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
1066 * active timer returns 1.)
1068 int mod_timer(struct timer_list *timer, unsigned long expires)
1070 return __mod_timer(timer, expires, false);
1072 EXPORT_SYMBOL(mod_timer);
1075 * add_timer - start a timer
1076 * @timer: the timer to be added
1078 * The kernel will do a ->function(->data) callback from the
1079 * timer interrupt at the ->expires point in the future. The
1080 * current time is 'jiffies'.
1082 * The timer's ->expires, ->function (and if the handler uses it, ->data)
1083 * fields must be set prior calling this function.
1085 * Timers with an ->expires field in the past will be executed in the next
1088 void add_timer(struct timer_list *timer)
1090 BUG_ON(timer_pending(timer));
1091 mod_timer(timer, timer->expires);
1093 EXPORT_SYMBOL(add_timer);
1096 * add_timer_on - start a timer on a particular CPU
1097 * @timer: the timer to be added
1098 * @cpu: the CPU to start it on
1100 * This is not very scalable on SMP. Double adds are not possible.
1102 void add_timer_on(struct timer_list *timer, int cpu)
1104 struct timer_base *new_base, *base;
1105 unsigned long flags;
1107 BUG_ON(timer_pending(timer) || !timer->function);
1109 new_base = get_timer_cpu_base(timer->flags, cpu);
1112 * If @timer was on a different CPU, it should be migrated with the
1113 * old base locked to prevent other operations proceeding with the
1114 * wrong base locked. See lock_timer_base().
1116 base = lock_timer_base(timer, &flags);
1117 if (base != new_base) {
1118 timer->flags |= TIMER_MIGRATING;
1120 raw_spin_unlock(&base->lock);
1122 raw_spin_lock(&base->lock);
1123 WRITE_ONCE(timer->flags,
1124 (timer->flags & ~TIMER_BASEMASK) | cpu);
1126 forward_timer_base(base);
1128 debug_activate(timer, timer->expires);
1129 internal_add_timer(base, timer);
1130 raw_spin_unlock_irqrestore(&base->lock, flags);
1132 EXPORT_SYMBOL_GPL(add_timer_on);
1135 * del_timer - deactivate a timer.
1136 * @timer: the timer to be deactivated
1138 * del_timer() deactivates a timer - this works on both active and inactive
1141 * The function returns whether it has deactivated a pending timer or not.
1142 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
1143 * active timer returns 1.)
1145 int del_timer(struct timer_list *timer)
1147 struct timer_base *base;
1148 unsigned long flags;
1151 debug_assert_init(timer);
1153 if (timer_pending(timer)) {
1154 base = lock_timer_base(timer, &flags);
1155 ret = detach_if_pending(timer, base, true);
1156 raw_spin_unlock_irqrestore(&base->lock, flags);
1161 EXPORT_SYMBOL(del_timer);
1164 * try_to_del_timer_sync - Try to deactivate a timer
1165 * @timer: timer to delete
1167 * This function tries to deactivate a timer. Upon successful (ret >= 0)
1168 * exit the timer is not queued and the handler is not running on any CPU.
1170 int try_to_del_timer_sync(struct timer_list *timer)
1172 struct timer_base *base;
1173 unsigned long flags;
1176 debug_assert_init(timer);
1178 base = lock_timer_base(timer, &flags);
1180 if (base->running_timer != timer)
1181 ret = detach_if_pending(timer, base, true);
1183 raw_spin_unlock_irqrestore(&base->lock, flags);
1187 EXPORT_SYMBOL(try_to_del_timer_sync);
1191 * del_timer_sync - deactivate a timer and wait for the handler to finish.
1192 * @timer: the timer to be deactivated
1194 * This function only differs from del_timer() on SMP: besides deactivating
1195 * the timer it also makes sure the handler has finished executing on other
1198 * Synchronization rules: Callers must prevent restarting of the timer,
1199 * otherwise this function is meaningless. It must not be called from
1200 * interrupt contexts unless the timer is an irqsafe one. The caller must
1201 * not hold locks which would prevent completion of the timer's
1202 * handler. The timer's handler must not call add_timer_on(). Upon exit the
1203 * timer is not queued and the handler is not running on any CPU.
1205 * Note: For !irqsafe timers, you must not hold locks that are held in
1206 * interrupt context while calling this function. Even if the lock has
1207 * nothing to do with the timer in question. Here's why:
1213 * base->running_timer = mytimer;
1214 * spin_lock_irq(somelock);
1216 * spin_lock(somelock);
1217 * del_timer_sync(mytimer);
1218 * while (base->running_timer == mytimer);
1220 * Now del_timer_sync() will never return and never release somelock.
1221 * The interrupt on the other CPU is waiting to grab somelock but
1222 * it has interrupted the softirq that CPU0 is waiting to finish.
1224 * The function returns whether it has deactivated a pending timer or not.
1226 int del_timer_sync(struct timer_list *timer)
1228 #ifdef CONFIG_LOCKDEP
1229 unsigned long flags;
1232 * If lockdep gives a backtrace here, please reference
1233 * the synchronization rules above.
1235 local_irq_save(flags);
1236 lock_map_acquire(&timer->lockdep_map);
1237 lock_map_release(&timer->lockdep_map);
1238 local_irq_restore(flags);
1241 * don't use it in hardirq context, because it
1242 * could lead to deadlock.
1244 WARN_ON(in_irq() && !(timer->flags & TIMER_IRQSAFE));
1246 int ret = try_to_del_timer_sync(timer);
1252 EXPORT_SYMBOL(del_timer_sync);
1255 static void call_timer_fn(struct timer_list *timer, void (*fn)(unsigned long),
1258 int count = preempt_count();
1260 #ifdef CONFIG_LOCKDEP
1262 * It is permissible to free the timer from inside the
1263 * function that is called from it, this we need to take into
1264 * account for lockdep too. To avoid bogus "held lock freed"
1265 * warnings as well as problems when looking into
1266 * timer->lockdep_map, make a copy and use that here.
1268 struct lockdep_map lockdep_map;
1270 lockdep_copy_map(&lockdep_map, &timer->lockdep_map);
1273 * Couple the lock chain with the lock chain at
1274 * del_timer_sync() by acquiring the lock_map around the fn()
1275 * call here and in del_timer_sync().
1277 lock_map_acquire(&lockdep_map);
1279 trace_timer_expire_entry(timer);
1281 trace_timer_expire_exit(timer);
1283 lock_map_release(&lockdep_map);
1285 if (count != preempt_count()) {
1286 WARN_ONCE(1, "timer: %pF preempt leak: %08x -> %08x\n",
1287 fn, count, preempt_count());
1289 * Restore the preempt count. That gives us a decent
1290 * chance to survive and extract information. If the
1291 * callback kept a lock held, bad luck, but not worse
1292 * than the BUG() we had.
1294 preempt_count_set(count);
1298 static void expire_timers(struct timer_base *base, struct hlist_head *head)
1300 while (!hlist_empty(head)) {
1301 struct timer_list *timer;
1302 void (*fn)(unsigned long);
1305 timer = hlist_entry(head->first, struct timer_list, entry);
1307 base->running_timer = timer;
1308 detach_timer(timer, true);
1310 fn = timer->function;
1313 if (timer->flags & TIMER_IRQSAFE) {
1314 raw_spin_unlock(&base->lock);
1315 call_timer_fn(timer, fn, data);
1316 raw_spin_lock(&base->lock);
1318 raw_spin_unlock_irq(&base->lock);
1319 call_timer_fn(timer, fn, data);
1320 raw_spin_lock_irq(&base->lock);
1325 static int __collect_expired_timers(struct timer_base *base,
1326 struct hlist_head *heads)
1328 unsigned long clk = base->clk;
1329 struct hlist_head *vec;
1333 for (i = 0; i < LVL_DEPTH; i++) {
1334 idx = (clk & LVL_MASK) + i * LVL_SIZE;
1336 if (__test_and_clear_bit(idx, base->pending_map)) {
1337 vec = base->vectors + idx;
1338 hlist_move_list(vec, heads++);
1341 /* Is it time to look at the next level? */
1342 if (clk & LVL_CLK_MASK)
1344 /* Shift clock for the next level granularity */
1345 clk >>= LVL_CLK_SHIFT;
1350 #ifdef CONFIG_NO_HZ_COMMON
1352 * Find the next pending bucket of a level. Search from level start (@offset)
1353 * + @clk upwards and if nothing there, search from start of the level
1354 * (@offset) up to @offset + clk.
1356 static int next_pending_bucket(struct timer_base *base, unsigned offset,
1359 unsigned pos, start = offset + clk;
1360 unsigned end = offset + LVL_SIZE;
1362 pos = find_next_bit(base->pending_map, end, start);
1366 pos = find_next_bit(base->pending_map, start, offset);
1367 return pos < start ? pos + LVL_SIZE - start : -1;
1371 * Search the first expiring timer in the various clock levels. Caller must
1374 static unsigned long __next_timer_interrupt(struct timer_base *base)
1376 unsigned long clk, next, adj;
1377 unsigned lvl, offset = 0;
1379 next = base->clk + NEXT_TIMER_MAX_DELTA;
1381 for (lvl = 0; lvl < LVL_DEPTH; lvl++, offset += LVL_SIZE) {
1382 int pos = next_pending_bucket(base, offset, clk & LVL_MASK);
1385 unsigned long tmp = clk + (unsigned long) pos;
1387 tmp <<= LVL_SHIFT(lvl);
1388 if (time_before(tmp, next))
1392 * Clock for the next level. If the current level clock lower
1393 * bits are zero, we look at the next level as is. If not we
1394 * need to advance it by one because that's going to be the
1395 * next expiring bucket in that level. base->clk is the next
1396 * expiring jiffie. So in case of:
1398 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1401 * we have to look at all levels @index 0. With
1403 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1406 * LVL0 has the next expiring bucket @index 2. The upper
1407 * levels have the next expiring bucket @index 1.
1409 * In case that the propagation wraps the next level the same
1412 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1415 * So after looking at LVL0 we get:
1417 * LVL5 LVL4 LVL3 LVL2 LVL1
1420 * So no propagation from LVL1 to LVL2 because that happened
1421 * with the add already, but then we need to propagate further
1422 * from LVL2 to LVL3.
1424 * So the simple check whether the lower bits of the current
1425 * level are 0 or not is sufficient for all cases.
1427 adj = clk & LVL_CLK_MASK ? 1 : 0;
1428 clk >>= LVL_CLK_SHIFT;
1435 * Check, if the next hrtimer event is before the next timer wheel
1438 static u64 cmp_next_hrtimer_event(u64 basem, u64 expires)
1440 u64 nextevt = hrtimer_get_next_event();
1443 * If high resolution timers are enabled
1444 * hrtimer_get_next_event() returns KTIME_MAX.
1446 if (expires <= nextevt)
1450 * If the next timer is already expired, return the tick base
1451 * time so the tick is fired immediately.
1453 if (nextevt <= basem)
1457 * Round up to the next jiffie. High resolution timers are
1458 * off, so the hrtimers are expired in the tick and we need to
1459 * make sure that this tick really expires the timer to avoid
1460 * a ping pong of the nohz stop code.
1462 * Use DIV_ROUND_UP_ULL to prevent gcc calling __divdi3
1464 return DIV_ROUND_UP_ULL(nextevt, TICK_NSEC) * TICK_NSEC;
1468 * get_next_timer_interrupt - return the time (clock mono) of the next timer
1469 * @basej: base time jiffies
1470 * @basem: base time clock monotonic
1472 * Returns the tick aligned clock monotonic time of the next pending
1473 * timer or KTIME_MAX if no timer is pending.
1475 u64 get_next_timer_interrupt(unsigned long basej, u64 basem)
1477 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1478 u64 expires = KTIME_MAX;
1479 unsigned long nextevt;
1483 * Pretend that there is no timer pending if the cpu is offline.
1484 * Possible pending timers will be migrated later to an active cpu.
1486 if (cpu_is_offline(smp_processor_id()))
1489 raw_spin_lock(&base->lock);
1490 nextevt = __next_timer_interrupt(base);
1491 is_max_delta = (nextevt == base->clk + NEXT_TIMER_MAX_DELTA);
1492 base->next_expiry = nextevt;
1494 * We have a fresh next event. Check whether we can forward the
1495 * base. We can only do that when @basej is past base->clk
1496 * otherwise we might rewind base->clk.
1498 if (time_after(basej, base->clk)) {
1499 if (time_after(nextevt, basej))
1501 else if (time_after(nextevt, base->clk))
1502 base->clk = nextevt;
1505 if (time_before_eq(nextevt, basej)) {
1507 base->is_idle = false;
1510 expires = basem + (u64)(nextevt - basej) * TICK_NSEC;
1512 * If we expect to sleep more than a tick, mark the base idle.
1513 * Also the tick is stopped so any added timer must forward
1514 * the base clk itself to keep granularity small. This idle
1515 * logic is only maintained for the BASE_STD base, deferrable
1516 * timers may still see large granularity skew (by design).
1518 if ((expires - basem) > TICK_NSEC) {
1519 base->must_forward_clk = true;
1520 base->is_idle = true;
1523 raw_spin_unlock(&base->lock);
1525 return cmp_next_hrtimer_event(basem, expires);
1529 * timer_clear_idle - Clear the idle state of the timer base
1531 * Called with interrupts disabled
1533 void timer_clear_idle(void)
1535 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1538 * We do this unlocked. The worst outcome is a remote enqueue sending
1539 * a pointless IPI, but taking the lock would just make the window for
1540 * sending the IPI a few instructions smaller for the cost of taking
1541 * the lock in the exit from idle path.
1543 base->is_idle = false;
1546 static int collect_expired_timers(struct timer_base *base,
1547 struct hlist_head *heads)
1549 unsigned long now = READ_ONCE(jiffies);
1552 * NOHZ optimization. After a long idle sleep we need to forward the
1553 * base to current jiffies. Avoid a loop by searching the bitfield for
1554 * the next expiring timer.
1556 if ((long)(now - base->clk) > 2) {
1557 unsigned long next = __next_timer_interrupt(base);
1560 * If the next timer is ahead of time forward to current
1561 * jiffies, otherwise forward to the next expiry time:
1563 if (time_after(next, now)) {
1564 /* The call site will increment clock! */
1565 base->clk = now - 1;
1570 return __collect_expired_timers(base, heads);
1573 static inline int collect_expired_timers(struct timer_base *base,
1574 struct hlist_head *heads)
1576 return __collect_expired_timers(base, heads);
1581 * Called from the timer interrupt handler to charge one tick to the current
1582 * process. user_tick is 1 if the tick is user time, 0 for system.
1584 void update_process_times(int user_tick)
1586 struct task_struct *p = current;
1588 /* Note: this timer irq context must be accounted for as well. */
1589 account_process_tick(p, user_tick);
1591 rcu_check_callbacks(user_tick);
1592 #ifdef CONFIG_IRQ_WORK
1597 if (IS_ENABLED(CONFIG_POSIX_TIMERS))
1598 run_posix_cpu_timers(p);
1602 * __run_timers - run all expired timers (if any) on this CPU.
1603 * @base: the timer vector to be processed.
1605 static inline void __run_timers(struct timer_base *base)
1607 struct hlist_head heads[LVL_DEPTH];
1610 if (!time_after_eq(jiffies, base->clk))
1613 raw_spin_lock_irq(&base->lock);
1616 * timer_base::must_forward_clk must be cleared before running
1617 * timers so that any timer functions that call mod_timer() will
1618 * not try to forward the base. Idle tracking / clock forwarding
1619 * logic is only used with BASE_STD timers.
1621 * The must_forward_clk flag is cleared unconditionally also for
1622 * the deferrable base. The deferrable base is not affected by idle
1623 * tracking and never forwarded, so clearing the flag is a NOOP.
1625 * The fact that the deferrable base is never forwarded can cause
1626 * large variations in granularity for deferrable timers, but they
1627 * can be deferred for long periods due to idle anyway.
1629 base->must_forward_clk = false;
1631 while (time_after_eq(jiffies, base->clk)) {
1633 levels = collect_expired_timers(base, heads);
1637 expire_timers(base, heads + levels);
1639 base->running_timer = NULL;
1640 raw_spin_unlock_irq(&base->lock);
1644 * This function runs timers and the timer-tq in bottom half context.
1646 static __latent_entropy void run_timer_softirq(struct softirq_action *h)
1648 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1651 if (IS_ENABLED(CONFIG_NO_HZ_COMMON))
1652 __run_timers(this_cpu_ptr(&timer_bases[BASE_DEF]));
1656 * Called by the local, per-CPU timer interrupt on SMP.
1658 void run_local_timers(void)
1660 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1662 hrtimer_run_queues();
1663 /* Raise the softirq only if required. */
1664 if (time_before(jiffies, base->clk)) {
1665 if (!IS_ENABLED(CONFIG_NO_HZ_COMMON))
1667 /* CPU is awake, so check the deferrable base. */
1669 if (time_before(jiffies, base->clk))
1672 raise_softirq(TIMER_SOFTIRQ);
1675 static void process_timeout(unsigned long __data)
1677 wake_up_process((struct task_struct *)__data);
1681 * schedule_timeout - sleep until timeout
1682 * @timeout: timeout value in jiffies
1684 * Make the current task sleep until @timeout jiffies have
1685 * elapsed. The routine will return immediately unless
1686 * the current task state has been set (see set_current_state()).
1688 * You can set the task state as follows -
1690 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1691 * pass before the routine returns unless the current task is explicitly
1692 * woken up, (e.g. by wake_up_process())".
1694 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1695 * delivered to the current task or the current task is explicitly woken
1698 * The current task state is guaranteed to be TASK_RUNNING when this
1701 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1702 * the CPU away without a bound on the timeout. In this case the return
1703 * value will be %MAX_SCHEDULE_TIMEOUT.
1705 * Returns 0 when the timer has expired otherwise the remaining time in
1706 * jiffies will be returned. In all cases the return value is guaranteed
1707 * to be non-negative.
1709 signed long __sched schedule_timeout(signed long timeout)
1711 struct timer_list timer;
1712 unsigned long expire;
1716 case MAX_SCHEDULE_TIMEOUT:
1718 * These two special cases are useful to be comfortable
1719 * in the caller. Nothing more. We could take
1720 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1721 * but I' d like to return a valid offset (>=0) to allow
1722 * the caller to do everything it want with the retval.
1728 * Another bit of PARANOID. Note that the retval will be
1729 * 0 since no piece of kernel is supposed to do a check
1730 * for a negative retval of schedule_timeout() (since it
1731 * should never happens anyway). You just have the printk()
1732 * that will tell you if something is gone wrong and where.
1735 printk(KERN_ERR "schedule_timeout: wrong timeout "
1736 "value %lx\n", timeout);
1738 current->state = TASK_RUNNING;
1743 expire = timeout + jiffies;
1745 setup_timer_on_stack(&timer, process_timeout, (unsigned long)current);
1746 __mod_timer(&timer, expire, false);
1748 del_singleshot_timer_sync(&timer);
1750 /* Remove the timer from the object tracker */
1751 destroy_timer_on_stack(&timer);
1753 timeout = expire - jiffies;
1756 return timeout < 0 ? 0 : timeout;
1758 EXPORT_SYMBOL(schedule_timeout);
1761 * We can use __set_current_state() here because schedule_timeout() calls
1762 * schedule() unconditionally.
1764 signed long __sched schedule_timeout_interruptible(signed long timeout)
1766 __set_current_state(TASK_INTERRUPTIBLE);
1767 return schedule_timeout(timeout);
1769 EXPORT_SYMBOL(schedule_timeout_interruptible);
1771 signed long __sched schedule_timeout_killable(signed long timeout)
1773 __set_current_state(TASK_KILLABLE);
1774 return schedule_timeout(timeout);
1776 EXPORT_SYMBOL(schedule_timeout_killable);
1778 signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1780 __set_current_state(TASK_UNINTERRUPTIBLE);
1781 return schedule_timeout(timeout);
1783 EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1786 * Like schedule_timeout_uninterruptible(), except this task will not contribute
1789 signed long __sched schedule_timeout_idle(signed long timeout)
1791 __set_current_state(TASK_IDLE);
1792 return schedule_timeout(timeout);
1794 EXPORT_SYMBOL(schedule_timeout_idle);
1796 #ifdef CONFIG_HOTPLUG_CPU
1797 static void migrate_timer_list(struct timer_base *new_base, struct hlist_head *head)
1799 struct timer_list *timer;
1800 int cpu = new_base->cpu;
1802 while (!hlist_empty(head)) {
1803 timer = hlist_entry(head->first, struct timer_list, entry);
1804 detach_timer(timer, false);
1805 timer->flags = (timer->flags & ~TIMER_BASEMASK) | cpu;
1806 internal_add_timer(new_base, timer);
1810 int timers_prepare_cpu(unsigned int cpu)
1812 struct timer_base *base;
1815 for (b = 0; b < NR_BASES; b++) {
1816 base = per_cpu_ptr(&timer_bases[b], cpu);
1817 base->clk = jiffies;
1818 base->next_expiry = base->clk + NEXT_TIMER_MAX_DELTA;
1819 base->is_idle = false;
1820 base->must_forward_clk = true;
1825 int timers_dead_cpu(unsigned int cpu)
1827 struct timer_base *old_base;
1828 struct timer_base *new_base;
1831 BUG_ON(cpu_online(cpu));
1833 for (b = 0; b < NR_BASES; b++) {
1834 old_base = per_cpu_ptr(&timer_bases[b], cpu);
1835 new_base = get_cpu_ptr(&timer_bases[b]);
1837 * The caller is globally serialized and nobody else
1838 * takes two locks at once, deadlock is not possible.
1840 raw_spin_lock_irq(&new_base->lock);
1841 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1844 * The current CPUs base clock might be stale. Update it
1845 * before moving the timers over.
1847 forward_timer_base(new_base);
1849 BUG_ON(old_base->running_timer);
1851 for (i = 0; i < WHEEL_SIZE; i++)
1852 migrate_timer_list(new_base, old_base->vectors + i);
1854 raw_spin_unlock(&old_base->lock);
1855 raw_spin_unlock_irq(&new_base->lock);
1856 put_cpu_ptr(&timer_bases);
1861 #endif /* CONFIG_HOTPLUG_CPU */
1863 static void __init init_timer_cpu(int cpu)
1865 struct timer_base *base;
1868 for (i = 0; i < NR_BASES; i++) {
1869 base = per_cpu_ptr(&timer_bases[i], cpu);
1871 raw_spin_lock_init(&base->lock);
1872 base->clk = jiffies;
1876 static void __init init_timer_cpus(void)
1880 for_each_possible_cpu(cpu)
1881 init_timer_cpu(cpu);
1884 void __init init_timers(void)
1887 open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
1891 * msleep - sleep safely even with waitqueue interruptions
1892 * @msecs: Time in milliseconds to sleep for
1894 void msleep(unsigned int msecs)
1896 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1899 timeout = schedule_timeout_uninterruptible(timeout);
1902 EXPORT_SYMBOL(msleep);
1905 * msleep_interruptible - sleep waiting for signals
1906 * @msecs: Time in milliseconds to sleep for
1908 unsigned long msleep_interruptible(unsigned int msecs)
1910 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1912 while (timeout && !signal_pending(current))
1913 timeout = schedule_timeout_interruptible(timeout);
1914 return jiffies_to_msecs(timeout);
1917 EXPORT_SYMBOL(msleep_interruptible);
1920 * usleep_range - Sleep for an approximate time
1921 * @min: Minimum time in usecs to sleep
1922 * @max: Maximum time in usecs to sleep
1924 * In non-atomic context where the exact wakeup time is flexible, use
1925 * usleep_range() instead of udelay(). The sleep improves responsiveness
1926 * by avoiding the CPU-hogging busy-wait of udelay(), and the range reduces
1927 * power usage by allowing hrtimers to take advantage of an already-
1928 * scheduled interrupt instead of scheduling a new one just for this sleep.
1930 void __sched usleep_range(unsigned long min, unsigned long max)
1932 ktime_t exp = ktime_add_us(ktime_get(), min);
1933 u64 delta = (u64)(max - min) * NSEC_PER_USEC;
1936 __set_current_state(TASK_UNINTERRUPTIBLE);
1937 /* Do not return before the requested sleep time has elapsed */
1938 if (!schedule_hrtimeout_range(&exp, delta, HRTIMER_MODE_ABS))
1942 EXPORT_SYMBOL(usleep_range);