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.h>
42 #include <linux/sched/sysctl.h>
43 #include <linux/slab.h>
44 #include <linux/compat.h>
46 #include <asm/uaccess.h>
47 #include <asm/unistd.h>
48 #include <asm/div64.h>
49 #include <asm/timex.h>
52 #include "tick-internal.h"
54 #define CREATE_TRACE_POINTS
55 #include <trace/events/timer.h>
57 __visible u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
59 EXPORT_SYMBOL(jiffies_64);
62 * The timer wheel has LVL_DEPTH array levels. Each level provides an array of
63 * LVL_SIZE buckets. Each level is driven by its own clock and therefor each
64 * level has a different granularity.
66 * The level granularity is: LVL_CLK_DIV ^ lvl
67 * The level clock frequency is: HZ / (LVL_CLK_DIV ^ level)
69 * The array level of a newly armed timer depends on the relative expiry
70 * time. The farther the expiry time is away the higher the array level and
71 * therefor the granularity becomes.
73 * Contrary to the original timer wheel implementation, which aims for 'exact'
74 * expiry of the timers, this implementation removes the need for recascading
75 * the timers into the lower array levels. The previous 'classic' timer wheel
76 * implementation of the kernel already violated the 'exact' expiry by adding
77 * slack to the expiry time to provide batched expiration. The granularity
78 * levels provide implicit batching.
80 * This is an optimization of the original timer wheel implementation for the
81 * majority of the timer wheel use cases: timeouts. The vast majority of
82 * timeout timers (networking, disk I/O ...) are canceled before expiry. If
83 * the timeout expires it indicates that normal operation is disturbed, so it
84 * does not matter much whether the timeout comes with a slight delay.
86 * The only exception to this are networking timers with a small expiry
87 * time. They rely on the granularity. Those fit into the first wheel level,
88 * which has HZ granularity.
90 * We don't have cascading anymore. timers with a expiry time above the
91 * capacity of the last wheel level are force expired at the maximum timeout
92 * value of the last wheel level. From data sampling we know that the maximum
93 * value observed is 5 days (network connection tracking), so this should not
96 * The currently chosen array constants values are a good compromise between
97 * array size and granularity.
99 * This results in the following granularity and range levels:
102 * Level Offset Granularity Range
103 * 0 0 1 ms 0 ms - 63 ms
104 * 1 64 8 ms 64 ms - 511 ms
105 * 2 128 64 ms 512 ms - 4095 ms (512ms - ~4s)
106 * 3 192 512 ms 4096 ms - 32767 ms (~4s - ~32s)
107 * 4 256 4096 ms (~4s) 32768 ms - 262143 ms (~32s - ~4m)
108 * 5 320 32768 ms (~32s) 262144 ms - 2097151 ms (~4m - ~34m)
109 * 6 384 262144 ms (~4m) 2097152 ms - 16777215 ms (~34m - ~4h)
110 * 7 448 2097152 ms (~34m) 16777216 ms - 134217727 ms (~4h - ~1d)
111 * 8 512 16777216 ms (~4h) 134217728 ms - 1073741822 ms (~1d - ~12d)
114 * Level Offset Granularity Range
115 * 0 0 3 ms 0 ms - 210 ms
116 * 1 64 26 ms 213 ms - 1703 ms (213ms - ~1s)
117 * 2 128 213 ms 1706 ms - 13650 ms (~1s - ~13s)
118 * 3 192 1706 ms (~1s) 13653 ms - 109223 ms (~13s - ~1m)
119 * 4 256 13653 ms (~13s) 109226 ms - 873810 ms (~1m - ~14m)
120 * 5 320 109226 ms (~1m) 873813 ms - 6990503 ms (~14m - ~1h)
121 * 6 384 873813 ms (~14m) 6990506 ms - 55924050 ms (~1h - ~15h)
122 * 7 448 6990506 ms (~1h) 55924053 ms - 447392423 ms (~15h - ~5d)
123 * 8 512 55924053 ms (~15h) 447392426 ms - 3579139406 ms (~5d - ~41d)
126 * Level Offset Granularity Range
127 * 0 0 4 ms 0 ms - 255 ms
128 * 1 64 32 ms 256 ms - 2047 ms (256ms - ~2s)
129 * 2 128 256 ms 2048 ms - 16383 ms (~2s - ~16s)
130 * 3 192 2048 ms (~2s) 16384 ms - 131071 ms (~16s - ~2m)
131 * 4 256 16384 ms (~16s) 131072 ms - 1048575 ms (~2m - ~17m)
132 * 5 320 131072 ms (~2m) 1048576 ms - 8388607 ms (~17m - ~2h)
133 * 6 384 1048576 ms (~17m) 8388608 ms - 67108863 ms (~2h - ~18h)
134 * 7 448 8388608 ms (~2h) 67108864 ms - 536870911 ms (~18h - ~6d)
135 * 8 512 67108864 ms (~18h) 536870912 ms - 4294967288 ms (~6d - ~49d)
138 * Level Offset Granularity Range
139 * 0 0 10 ms 0 ms - 630 ms
140 * 1 64 80 ms 640 ms - 5110 ms (640ms - ~5s)
141 * 2 128 640 ms 5120 ms - 40950 ms (~5s - ~40s)
142 * 3 192 5120 ms (~5s) 40960 ms - 327670 ms (~40s - ~5m)
143 * 4 256 40960 ms (~40s) 327680 ms - 2621430 ms (~5m - ~43m)
144 * 5 320 327680 ms (~5m) 2621440 ms - 20971510 ms (~43m - ~5h)
145 * 6 384 2621440 ms (~43m) 20971520 ms - 167772150 ms (~5h - ~1d)
146 * 7 448 20971520 ms (~5h) 167772160 ms - 1342177270 ms (~1d - ~15d)
149 /* Clock divisor for the next level */
150 #define LVL_CLK_SHIFT 3
151 #define LVL_CLK_DIV (1UL << LVL_CLK_SHIFT)
152 #define LVL_CLK_MASK (LVL_CLK_DIV - 1)
153 #define LVL_SHIFT(n) ((n) * LVL_CLK_SHIFT)
154 #define LVL_GRAN(n) (1UL << LVL_SHIFT(n))
157 * The time start value for each level to select the bucket at enqueue
160 #define LVL_START(n) ((LVL_SIZE - 1) << (((n) - 1) * LVL_CLK_SHIFT))
162 /* Size of each clock level */
164 #define LVL_SIZE (1UL << LVL_BITS)
165 #define LVL_MASK (LVL_SIZE - 1)
166 #define LVL_OFFS(n) ((n) * LVL_SIZE)
175 /* The cutoff (max. capacity of the wheel) */
176 #define WHEEL_TIMEOUT_CUTOFF (LVL_START(LVL_DEPTH))
177 #define WHEEL_TIMEOUT_MAX (WHEEL_TIMEOUT_CUTOFF - LVL_GRAN(LVL_DEPTH - 1))
180 * The resulting wheel size. If NOHZ is configured we allocate two
181 * wheels so we have a separate storage for the deferrable timers.
183 #define WHEEL_SIZE (LVL_SIZE * LVL_DEPTH)
185 #ifdef CONFIG_NO_HZ_COMMON
197 struct timer_list *running_timer;
199 unsigned long next_expiry;
201 bool migration_enabled;
204 DECLARE_BITMAP(pending_map, WHEEL_SIZE);
205 struct hlist_head vectors[WHEEL_SIZE];
206 } ____cacheline_aligned;
208 static DEFINE_PER_CPU(struct timer_base, timer_bases[NR_BASES]);
210 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
211 unsigned int sysctl_timer_migration = 1;
213 void timers_update_migration(bool update_nohz)
215 bool on = sysctl_timer_migration && tick_nohz_active;
218 /* Avoid the loop, if nothing to update */
219 if (this_cpu_read(timer_bases[BASE_STD].migration_enabled) == on)
222 for_each_possible_cpu(cpu) {
223 per_cpu(timer_bases[BASE_STD].migration_enabled, cpu) = on;
224 per_cpu(timer_bases[BASE_DEF].migration_enabled, cpu) = on;
225 per_cpu(hrtimer_bases.migration_enabled, cpu) = on;
228 per_cpu(timer_bases[BASE_STD].nohz_active, cpu) = true;
229 per_cpu(timer_bases[BASE_DEF].nohz_active, cpu) = true;
230 per_cpu(hrtimer_bases.nohz_active, cpu) = true;
234 int timer_migration_handler(struct ctl_table *table, int write,
235 void __user *buffer, size_t *lenp,
238 static DEFINE_MUTEX(mutex);
242 ret = proc_dointvec(table, write, buffer, lenp, ppos);
244 timers_update_migration(false);
245 mutex_unlock(&mutex);
250 static unsigned long round_jiffies_common(unsigned long j, int cpu,
254 unsigned long original = j;
257 * We don't want all cpus firing their timers at once hitting the
258 * same lock or cachelines, so we skew each extra cpu with an extra
259 * 3 jiffies. This 3 jiffies came originally from the mm/ code which
261 * The skew is done by adding 3*cpunr, then round, then subtract this
262 * extra offset again.
269 * If the target jiffie is just after a whole second (which can happen
270 * due to delays of the timer irq, long irq off times etc etc) then
271 * we should round down to the whole second, not up. Use 1/4th second
272 * as cutoff for this rounding as an extreme upper bound for this.
273 * But never round down if @force_up is set.
275 if (rem < HZ/4 && !force_up) /* round down */
280 /* now that we have rounded, subtract the extra skew again */
284 * Make sure j is still in the future. Otherwise return the
287 return time_is_after_jiffies(j) ? j : original;
291 * __round_jiffies - function to round jiffies to a full second
292 * @j: the time in (absolute) jiffies that should be rounded
293 * @cpu: the processor number on which the timeout will happen
295 * __round_jiffies() rounds an absolute time in the future (in jiffies)
296 * up or down to (approximately) full seconds. This is useful for timers
297 * for which the exact time they fire does not matter too much, as long as
298 * they fire approximately every X seconds.
300 * By rounding these timers to whole seconds, all such timers will fire
301 * at the same time, rather than at various times spread out. The goal
302 * of this is to have the CPU wake up less, which saves power.
304 * The exact rounding is skewed for each processor to avoid all
305 * processors firing at the exact same time, which could lead
306 * to lock contention or spurious cache line bouncing.
308 * The return value is the rounded version of the @j parameter.
310 unsigned long __round_jiffies(unsigned long j, int cpu)
312 return round_jiffies_common(j, cpu, false);
314 EXPORT_SYMBOL_GPL(__round_jiffies);
317 * __round_jiffies_relative - function to round jiffies to a full second
318 * @j: the time in (relative) jiffies that should be rounded
319 * @cpu: the processor number on which the timeout will happen
321 * __round_jiffies_relative() rounds a time delta in the future (in jiffies)
322 * up or down to (approximately) full seconds. This is useful for timers
323 * for which the exact time they fire does not matter too much, as long as
324 * they fire approximately every X seconds.
326 * By rounding these timers to whole seconds, all such timers will fire
327 * at the same time, rather than at various times spread out. The goal
328 * of this is to have the CPU wake up less, which saves power.
330 * The exact rounding is skewed for each processor to avoid all
331 * processors firing at the exact same time, which could lead
332 * to lock contention or spurious cache line bouncing.
334 * The return value is the rounded version of the @j parameter.
336 unsigned long __round_jiffies_relative(unsigned long j, int cpu)
338 unsigned long j0 = jiffies;
340 /* Use j0 because jiffies might change while we run */
341 return round_jiffies_common(j + j0, cpu, false) - j0;
343 EXPORT_SYMBOL_GPL(__round_jiffies_relative);
346 * round_jiffies - function to round jiffies to a full second
347 * @j: the time in (absolute) jiffies that should be rounded
349 * round_jiffies() rounds an absolute time in the future (in jiffies)
350 * up or down to (approximately) full seconds. This is useful for timers
351 * for which the exact time they fire does not matter too much, as long as
352 * they fire approximately every X seconds.
354 * By rounding these timers to whole seconds, all such timers will fire
355 * at the same time, rather than at various times spread out. The goal
356 * of this is to have the CPU wake up less, which saves power.
358 * The return value is the rounded version of the @j parameter.
360 unsigned long round_jiffies(unsigned long j)
362 return round_jiffies_common(j, raw_smp_processor_id(), false);
364 EXPORT_SYMBOL_GPL(round_jiffies);
367 * round_jiffies_relative - function to round jiffies to a full second
368 * @j: the time in (relative) jiffies that should be rounded
370 * round_jiffies_relative() rounds a time delta 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_relative(unsigned long j)
383 return __round_jiffies_relative(j, raw_smp_processor_id());
385 EXPORT_SYMBOL_GPL(round_jiffies_relative);
388 * __round_jiffies_up - function to round jiffies up to a full second
389 * @j: the time in (absolute) jiffies that should be rounded
390 * @cpu: the processor number on which the timeout will happen
392 * This is the same as __round_jiffies() except that it will never
393 * round down. This is useful for timeouts for which the exact time
394 * of firing does not matter too much, as long as they don't fire too
397 unsigned long __round_jiffies_up(unsigned long j, int cpu)
399 return round_jiffies_common(j, cpu, true);
401 EXPORT_SYMBOL_GPL(__round_jiffies_up);
404 * __round_jiffies_up_relative - function to round jiffies up to a full second
405 * @j: the time in (relative) jiffies that should be rounded
406 * @cpu: the processor number on which the timeout will happen
408 * This is the same as __round_jiffies_relative() except that it will never
409 * round down. This is useful for timeouts for which the exact time
410 * of firing does not matter too much, as long as they don't fire too
413 unsigned long __round_jiffies_up_relative(unsigned long j, int cpu)
415 unsigned long j0 = jiffies;
417 /* Use j0 because jiffies might change while we run */
418 return round_jiffies_common(j + j0, cpu, true) - j0;
420 EXPORT_SYMBOL_GPL(__round_jiffies_up_relative);
423 * round_jiffies_up - function to round jiffies up to a full second
424 * @j: the time in (absolute) jiffies that should be rounded
426 * This is the same as round_jiffies() except that it will never
427 * round down. This is useful for timeouts for which the exact time
428 * of firing does not matter too much, as long as they don't fire too
431 unsigned long round_jiffies_up(unsigned long j)
433 return round_jiffies_common(j, raw_smp_processor_id(), true);
435 EXPORT_SYMBOL_GPL(round_jiffies_up);
438 * round_jiffies_up_relative - function to round jiffies up to a full second
439 * @j: the time in (relative) jiffies that should be rounded
441 * This is the same as round_jiffies_relative() except that it will never
442 * round down. This is useful for timeouts for which the exact time
443 * of firing does not matter too much, as long as they don't fire too
446 unsigned long round_jiffies_up_relative(unsigned long j)
448 return __round_jiffies_up_relative(j, raw_smp_processor_id());
450 EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
453 static inline unsigned int timer_get_idx(struct timer_list *timer)
455 return (timer->flags & TIMER_ARRAYMASK) >> TIMER_ARRAYSHIFT;
458 static inline void timer_set_idx(struct timer_list *timer, unsigned int idx)
460 timer->flags = (timer->flags & ~TIMER_ARRAYMASK) |
461 idx << TIMER_ARRAYSHIFT;
465 * Helper function to calculate the array index for a given expiry
468 static inline unsigned calc_index(unsigned expires, unsigned lvl)
470 expires = (expires + LVL_GRAN(lvl)) >> LVL_SHIFT(lvl);
471 return LVL_OFFS(lvl) + (expires & LVL_MASK);
475 __internal_add_timer(struct timer_base *base, struct timer_list *timer)
477 unsigned long expires = timer->expires;
478 unsigned long delta = expires - base->clk;
479 struct hlist_head *vec;
482 if (delta < LVL_START(1)) {
483 idx = calc_index(expires, 0);
484 } else if (delta < LVL_START(2)) {
485 idx = calc_index(expires, 1);
486 } else if (delta < LVL_START(3)) {
487 idx = calc_index(expires, 2);
488 } else if (delta < LVL_START(4)) {
489 idx = calc_index(expires, 3);
490 } else if (delta < LVL_START(5)) {
491 idx = calc_index(expires, 4);
492 } else if (delta < LVL_START(6)) {
493 idx = calc_index(expires, 5);
494 } else if (delta < LVL_START(7)) {
495 idx = calc_index(expires, 6);
496 } else if (LVL_DEPTH > 8 && delta < LVL_START(8)) {
497 idx = calc_index(expires, 7);
498 } else if ((long) delta < 0) {
499 idx = base->clk & LVL_MASK;
502 * Force expire obscene large timeouts to expire at the
503 * capacity limit of the wheel.
505 if (expires >= WHEEL_TIMEOUT_CUTOFF)
506 expires = WHEEL_TIMEOUT_MAX;
508 idx = calc_index(expires, LVL_DEPTH - 1);
511 * Enqueue the timer into the array bucket, mark it pending in
512 * the bitmap and store the index in the timer flags.
514 vec = base->vectors + idx;
515 hlist_add_head(&timer->entry, vec);
516 __set_bit(idx, base->pending_map);
517 timer_set_idx(timer, idx);
520 static void internal_add_timer(struct timer_base *base, struct timer_list *timer)
522 __internal_add_timer(base, timer);
524 if (!IS_ENABLED(CONFIG_NO_HZ_COMMON) || !base->nohz_active)
528 * TODO: This wants some optimizing similar to the code below, but we
529 * will do that when we switch from push to pull for deferrable timers.
531 if (timer->flags & TIMER_DEFERRABLE) {
532 if (tick_nohz_full_cpu(base->cpu))
533 wake_up_nohz_cpu(base->cpu);
538 * We might have to IPI the remote CPU if the base is idle and the
539 * timer is not deferrable. If the other CPU is on the way to idle
540 * then it can't set base->is_idle as we hold the base lock:
545 /* Check whether this is the new first expiring timer: */
546 if (time_after_eq(timer->expires, base->next_expiry))
550 * Set the next expiry time and kick the CPU so it can reevaluate the
553 base->next_expiry = timer->expires;
554 wake_up_nohz_cpu(base->cpu);
557 #ifdef CONFIG_TIMER_STATS
558 void __timer_stats_timer_set_start_info(struct timer_list *timer, void *addr)
560 if (timer->start_site)
563 timer->start_site = addr;
564 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
565 timer->start_pid = current->pid;
568 static void timer_stats_account_timer(struct timer_list *timer)
573 * start_site can be concurrently reset by
574 * timer_stats_timer_clear_start_info()
576 site = READ_ONCE(timer->start_site);
580 timer_stats_update_stats(timer, timer->start_pid, site,
581 timer->function, timer->start_comm,
586 static void timer_stats_account_timer(struct timer_list *timer) {}
589 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
591 static struct debug_obj_descr timer_debug_descr;
593 static void *timer_debug_hint(void *addr)
595 return ((struct timer_list *) addr)->function;
598 static bool timer_is_static_object(void *addr)
600 struct timer_list *timer = addr;
602 return (timer->entry.pprev == NULL &&
603 timer->entry.next == TIMER_ENTRY_STATIC);
607 * fixup_init is called when:
608 * - an active object is initialized
610 static bool timer_fixup_init(void *addr, enum debug_obj_state state)
612 struct timer_list *timer = addr;
615 case ODEBUG_STATE_ACTIVE:
616 del_timer_sync(timer);
617 debug_object_init(timer, &timer_debug_descr);
624 /* Stub timer callback for improperly used timers. */
625 static void stub_timer(unsigned long data)
631 * fixup_activate is called when:
632 * - an active object is activated
633 * - an unknown non-static object is activated
635 static bool timer_fixup_activate(void *addr, enum debug_obj_state state)
637 struct timer_list *timer = addr;
640 case ODEBUG_STATE_NOTAVAILABLE:
641 setup_timer(timer, stub_timer, 0);
644 case ODEBUG_STATE_ACTIVE:
653 * fixup_free is called when:
654 * - an active object is freed
656 static bool timer_fixup_free(void *addr, enum debug_obj_state state)
658 struct timer_list *timer = addr;
661 case ODEBUG_STATE_ACTIVE:
662 del_timer_sync(timer);
663 debug_object_free(timer, &timer_debug_descr);
671 * fixup_assert_init is called when:
672 * - an untracked/uninit-ed object is found
674 static bool timer_fixup_assert_init(void *addr, enum debug_obj_state state)
676 struct timer_list *timer = addr;
679 case ODEBUG_STATE_NOTAVAILABLE:
680 setup_timer(timer, stub_timer, 0);
687 static struct debug_obj_descr timer_debug_descr = {
688 .name = "timer_list",
689 .debug_hint = timer_debug_hint,
690 .is_static_object = timer_is_static_object,
691 .fixup_init = timer_fixup_init,
692 .fixup_activate = timer_fixup_activate,
693 .fixup_free = timer_fixup_free,
694 .fixup_assert_init = timer_fixup_assert_init,
697 static inline void debug_timer_init(struct timer_list *timer)
699 debug_object_init(timer, &timer_debug_descr);
702 static inline void debug_timer_activate(struct timer_list *timer)
704 debug_object_activate(timer, &timer_debug_descr);
707 static inline void debug_timer_deactivate(struct timer_list *timer)
709 debug_object_deactivate(timer, &timer_debug_descr);
712 static inline void debug_timer_free(struct timer_list *timer)
714 debug_object_free(timer, &timer_debug_descr);
717 static inline void debug_timer_assert_init(struct timer_list *timer)
719 debug_object_assert_init(timer, &timer_debug_descr);
722 static void do_init_timer(struct timer_list *timer, unsigned int flags,
723 const char *name, struct lock_class_key *key);
725 void init_timer_on_stack_key(struct timer_list *timer, unsigned int flags,
726 const char *name, struct lock_class_key *key)
728 debug_object_init_on_stack(timer, &timer_debug_descr);
729 do_init_timer(timer, flags, name, key);
731 EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
733 void destroy_timer_on_stack(struct timer_list *timer)
735 debug_object_free(timer, &timer_debug_descr);
737 EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
740 static inline void debug_timer_init(struct timer_list *timer) { }
741 static inline void debug_timer_activate(struct timer_list *timer) { }
742 static inline void debug_timer_deactivate(struct timer_list *timer) { }
743 static inline void debug_timer_assert_init(struct timer_list *timer) { }
746 static inline void debug_init(struct timer_list *timer)
748 debug_timer_init(timer);
749 trace_timer_init(timer);
753 debug_activate(struct timer_list *timer, unsigned long expires)
755 debug_timer_activate(timer);
756 trace_timer_start(timer, expires, timer->flags);
759 static inline void debug_deactivate(struct timer_list *timer)
761 debug_timer_deactivate(timer);
762 trace_timer_cancel(timer);
765 static inline void debug_assert_init(struct timer_list *timer)
767 debug_timer_assert_init(timer);
770 static void do_init_timer(struct timer_list *timer, unsigned int flags,
771 const char *name, struct lock_class_key *key)
773 timer->entry.pprev = NULL;
774 timer->flags = flags | raw_smp_processor_id();
775 #ifdef CONFIG_TIMER_STATS
776 timer->start_site = NULL;
777 timer->start_pid = -1;
778 memset(timer->start_comm, 0, TASK_COMM_LEN);
780 lockdep_init_map(&timer->lockdep_map, name, key, 0);
784 * init_timer_key - initialize a timer
785 * @timer: the timer to be initialized
786 * @flags: timer flags
787 * @name: name of the timer
788 * @key: lockdep class key of the fake lock used for tracking timer
789 * sync lock dependencies
791 * init_timer_key() must be done to a timer prior calling *any* of the
792 * other timer functions.
794 void init_timer_key(struct timer_list *timer, unsigned int flags,
795 const char *name, struct lock_class_key *key)
798 do_init_timer(timer, flags, name, key);
800 EXPORT_SYMBOL(init_timer_key);
802 static inline void detach_timer(struct timer_list *timer, bool clear_pending)
804 struct hlist_node *entry = &timer->entry;
806 debug_deactivate(timer);
811 entry->next = LIST_POISON2;
814 static int detach_if_pending(struct timer_list *timer, struct timer_base *base,
817 unsigned idx = timer_get_idx(timer);
819 if (!timer_pending(timer))
822 if (hlist_is_singular_node(&timer->entry, base->vectors + idx))
823 __clear_bit(idx, base->pending_map);
825 detach_timer(timer, clear_pending);
829 static inline struct timer_base *get_timer_cpu_base(u32 tflags, u32 cpu)
831 struct timer_base *base = per_cpu_ptr(&timer_bases[BASE_STD], cpu);
834 * If the timer is deferrable and nohz is active then we need to use
835 * the deferrable base.
837 if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && base->nohz_active &&
838 (tflags & TIMER_DEFERRABLE))
839 base = per_cpu_ptr(&timer_bases[BASE_DEF], cpu);
843 static inline struct timer_base *get_timer_this_cpu_base(u32 tflags)
845 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
848 * If the timer is deferrable and nohz is active then we need to use
849 * the deferrable base.
851 if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && base->nohz_active &&
852 (tflags & TIMER_DEFERRABLE))
853 base = this_cpu_ptr(&timer_bases[BASE_DEF]);
857 static inline struct timer_base *get_timer_base(u32 tflags)
859 return get_timer_cpu_base(tflags, tflags & TIMER_CPUMASK);
862 #ifdef CONFIG_NO_HZ_COMMON
863 static inline struct timer_base *
864 __get_target_base(struct timer_base *base, unsigned tflags)
867 if ((tflags & TIMER_PINNED) || !base->migration_enabled)
868 return get_timer_this_cpu_base(tflags);
869 return get_timer_cpu_base(tflags, get_nohz_timer_target());
871 return get_timer_this_cpu_base(tflags);
875 static inline void forward_timer_base(struct timer_base *base)
878 * We only forward the base when it's idle and we have a delta between
879 * base clock and jiffies.
881 if (!base->is_idle || (long) (jiffies - base->clk) < 2)
885 * If the next expiry value is > jiffies, then we fast forward to
886 * jiffies otherwise we forward to the next expiry value.
888 if (time_after(base->next_expiry, jiffies))
891 base->clk = base->next_expiry;
894 static inline struct timer_base *
895 __get_target_base(struct timer_base *base, unsigned tflags)
897 return get_timer_this_cpu_base(tflags);
900 static inline void forward_timer_base(struct timer_base *base) { }
903 static inline struct timer_base *
904 get_target_base(struct timer_base *base, unsigned tflags)
906 struct timer_base *target = __get_target_base(base, tflags);
908 forward_timer_base(target);
913 * We are using hashed locking: Holding per_cpu(timer_bases[x]).lock means
914 * that all timers which are tied to this base are locked, and the base itself
917 * So __run_timers/migrate_timers can safely modify all timers which could
918 * be found in the base->vectors array.
920 * When a timer is migrating then the TIMER_MIGRATING flag is set and we need
921 * to wait until the migration is done.
923 static struct timer_base *lock_timer_base(struct timer_list *timer,
924 unsigned long *flags)
925 __acquires(timer->base->lock)
928 struct timer_base *base;
929 u32 tf = timer->flags;
931 if (!(tf & TIMER_MIGRATING)) {
932 base = get_timer_base(tf);
933 spin_lock_irqsave(&base->lock, *flags);
934 if (timer->flags == tf)
936 spin_unlock_irqrestore(&base->lock, *flags);
943 __mod_timer(struct timer_list *timer, unsigned long expires, bool pending_only)
945 struct timer_base *base, *new_base;
950 * TODO: Calculate the array bucket of the timer right here w/o
951 * holding the base lock. This allows to check not only
952 * timer->expires == expires below, but also whether the timer
953 * ends up in the same bucket. If we really need to requeue
954 * the timer then we check whether base->clk have
955 * advanced between here and locking the timer base. If
956 * jiffies advanced we have to recalc the array bucket with the
961 * This is a common optimization triggered by the
962 * networking code - if the timer is re-modified
963 * to be the same thing then just return:
965 if (timer_pending(timer)) {
966 if (timer->expires == expires)
970 timer_stats_timer_set_start_info(timer);
971 BUG_ON(!timer->function);
973 base = lock_timer_base(timer, &flags);
975 ret = detach_if_pending(timer, base, false);
976 if (!ret && pending_only)
979 debug_activate(timer, expires);
981 new_base = get_target_base(base, timer->flags);
983 if (base != new_base) {
985 * We are trying to schedule the timer on the new base.
986 * However we can't change timer's base while it is running,
987 * otherwise del_timer_sync() can't detect that the timer's
988 * handler yet has not finished. This also guarantees that the
989 * timer is serialized wrt itself.
991 if (likely(base->running_timer != timer)) {
992 /* See the comment in lock_timer_base() */
993 timer->flags |= TIMER_MIGRATING;
995 spin_unlock(&base->lock);
997 spin_lock(&base->lock);
998 WRITE_ONCE(timer->flags,
999 (timer->flags & ~TIMER_BASEMASK) | base->cpu);
1003 timer->expires = expires;
1004 internal_add_timer(base, timer);
1007 spin_unlock_irqrestore(&base->lock, flags);
1013 * mod_timer_pending - modify a pending timer's timeout
1014 * @timer: the pending timer to be modified
1015 * @expires: new timeout in jiffies
1017 * mod_timer_pending() is the same for pending timers as mod_timer(),
1018 * but will not re-activate and modify already deleted timers.
1020 * It is useful for unserialized use of timers.
1022 int mod_timer_pending(struct timer_list *timer, unsigned long expires)
1024 return __mod_timer(timer, expires, true);
1026 EXPORT_SYMBOL(mod_timer_pending);
1029 * mod_timer - modify a timer's timeout
1030 * @timer: the timer to be modified
1031 * @expires: new timeout in jiffies
1033 * mod_timer() is a more efficient way to update the expire field of an
1034 * active timer (if the timer is inactive it will be activated)
1036 * mod_timer(timer, expires) is equivalent to:
1038 * del_timer(timer); timer->expires = expires; add_timer(timer);
1040 * Note that if there are multiple unserialized concurrent users of the
1041 * same timer, then mod_timer() is the only safe way to modify the timeout,
1042 * since add_timer() cannot modify an already running timer.
1044 * The function returns whether it has modified a pending timer or not.
1045 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
1046 * active timer returns 1.)
1048 int mod_timer(struct timer_list *timer, unsigned long expires)
1050 return __mod_timer(timer, expires, false);
1052 EXPORT_SYMBOL(mod_timer);
1055 * add_timer - start a timer
1056 * @timer: the timer to be added
1058 * The kernel will do a ->function(->data) callback from the
1059 * timer interrupt at the ->expires point in the future. The
1060 * current time is 'jiffies'.
1062 * The timer's ->expires, ->function (and if the handler uses it, ->data)
1063 * fields must be set prior calling this function.
1065 * Timers with an ->expires field in the past will be executed in the next
1068 void add_timer(struct timer_list *timer)
1070 BUG_ON(timer_pending(timer));
1071 mod_timer(timer, timer->expires);
1073 EXPORT_SYMBOL(add_timer);
1076 * add_timer_on - start a timer on a particular CPU
1077 * @timer: the timer to be added
1078 * @cpu: the CPU to start it on
1080 * This is not very scalable on SMP. Double adds are not possible.
1082 void add_timer_on(struct timer_list *timer, int cpu)
1084 struct timer_base *new_base, *base;
1085 unsigned long flags;
1087 timer_stats_timer_set_start_info(timer);
1088 BUG_ON(timer_pending(timer) || !timer->function);
1090 new_base = get_timer_cpu_base(timer->flags, cpu);
1093 * If @timer was on a different CPU, it should be migrated with the
1094 * old base locked to prevent other operations proceeding with the
1095 * wrong base locked. See lock_timer_base().
1097 base = lock_timer_base(timer, &flags);
1098 if (base != new_base) {
1099 timer->flags |= TIMER_MIGRATING;
1101 spin_unlock(&base->lock);
1103 spin_lock(&base->lock);
1104 WRITE_ONCE(timer->flags,
1105 (timer->flags & ~TIMER_BASEMASK) | cpu);
1108 debug_activate(timer, timer->expires);
1109 internal_add_timer(base, timer);
1110 spin_unlock_irqrestore(&base->lock, flags);
1112 EXPORT_SYMBOL_GPL(add_timer_on);
1115 * del_timer - deactive a timer.
1116 * @timer: the timer to be deactivated
1118 * del_timer() deactivates a timer - this works on both active and inactive
1121 * The function returns whether it has deactivated a pending timer or not.
1122 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
1123 * active timer returns 1.)
1125 int del_timer(struct timer_list *timer)
1127 struct timer_base *base;
1128 unsigned long flags;
1131 debug_assert_init(timer);
1133 timer_stats_timer_clear_start_info(timer);
1134 if (timer_pending(timer)) {
1135 base = lock_timer_base(timer, &flags);
1136 ret = detach_if_pending(timer, base, true);
1137 spin_unlock_irqrestore(&base->lock, flags);
1142 EXPORT_SYMBOL(del_timer);
1145 * try_to_del_timer_sync - Try to deactivate a timer
1146 * @timer: timer do del
1148 * This function tries to deactivate a timer. Upon successful (ret >= 0)
1149 * exit the timer is not queued and the handler is not running on any CPU.
1151 int try_to_del_timer_sync(struct timer_list *timer)
1153 struct timer_base *base;
1154 unsigned long flags;
1157 debug_assert_init(timer);
1159 base = lock_timer_base(timer, &flags);
1161 if (base->running_timer != timer) {
1162 timer_stats_timer_clear_start_info(timer);
1163 ret = detach_if_pending(timer, base, true);
1165 spin_unlock_irqrestore(&base->lock, flags);
1169 EXPORT_SYMBOL(try_to_del_timer_sync);
1173 * del_timer_sync - deactivate a timer and wait for the handler to finish.
1174 * @timer: the timer to be deactivated
1176 * This function only differs from del_timer() on SMP: besides deactivating
1177 * the timer it also makes sure the handler has finished executing on other
1180 * Synchronization rules: Callers must prevent restarting of the timer,
1181 * otherwise this function is meaningless. It must not be called from
1182 * interrupt contexts unless the timer is an irqsafe one. The caller must
1183 * not hold locks which would prevent completion of the timer's
1184 * handler. The timer's handler must not call add_timer_on(). Upon exit the
1185 * timer is not queued and the handler is not running on any CPU.
1187 * Note: For !irqsafe timers, you must not hold locks that are held in
1188 * interrupt context while calling this function. Even if the lock has
1189 * nothing to do with the timer in question. Here's why:
1195 * base->running_timer = mytimer;
1196 * spin_lock_irq(somelock);
1198 * spin_lock(somelock);
1199 * del_timer_sync(mytimer);
1200 * while (base->running_timer == mytimer);
1202 * Now del_timer_sync() will never return and never release somelock.
1203 * The interrupt on the other CPU is waiting to grab somelock but
1204 * it has interrupted the softirq that CPU0 is waiting to finish.
1206 * The function returns whether it has deactivated a pending timer or not.
1208 int del_timer_sync(struct timer_list *timer)
1210 #ifdef CONFIG_LOCKDEP
1211 unsigned long flags;
1214 * If lockdep gives a backtrace here, please reference
1215 * the synchronization rules above.
1217 local_irq_save(flags);
1218 lock_map_acquire(&timer->lockdep_map);
1219 lock_map_release(&timer->lockdep_map);
1220 local_irq_restore(flags);
1223 * don't use it in hardirq context, because it
1224 * could lead to deadlock.
1226 WARN_ON(in_irq() && !(timer->flags & TIMER_IRQSAFE));
1228 int ret = try_to_del_timer_sync(timer);
1234 EXPORT_SYMBOL(del_timer_sync);
1237 static void call_timer_fn(struct timer_list *timer, void (*fn)(unsigned long),
1240 int count = preempt_count();
1242 #ifdef CONFIG_LOCKDEP
1244 * It is permissible to free the timer from inside the
1245 * function that is called from it, this we need to take into
1246 * account for lockdep too. To avoid bogus "held lock freed"
1247 * warnings as well as problems when looking into
1248 * timer->lockdep_map, make a copy and use that here.
1250 struct lockdep_map lockdep_map;
1252 lockdep_copy_map(&lockdep_map, &timer->lockdep_map);
1255 * Couple the lock chain with the lock chain at
1256 * del_timer_sync() by acquiring the lock_map around the fn()
1257 * call here and in del_timer_sync().
1259 lock_map_acquire(&lockdep_map);
1261 trace_timer_expire_entry(timer);
1263 trace_timer_expire_exit(timer);
1265 lock_map_release(&lockdep_map);
1267 if (count != preempt_count()) {
1268 WARN_ONCE(1, "timer: %pF preempt leak: %08x -> %08x\n",
1269 fn, count, preempt_count());
1271 * Restore the preempt count. That gives us a decent
1272 * chance to survive and extract information. If the
1273 * callback kept a lock held, bad luck, but not worse
1274 * than the BUG() we had.
1276 preempt_count_set(count);
1280 static void expire_timers(struct timer_base *base, struct hlist_head *head)
1282 while (!hlist_empty(head)) {
1283 struct timer_list *timer;
1284 void (*fn)(unsigned long);
1287 timer = hlist_entry(head->first, struct timer_list, entry);
1288 timer_stats_account_timer(timer);
1290 base->running_timer = timer;
1291 detach_timer(timer, true);
1293 fn = timer->function;
1296 if (timer->flags & TIMER_IRQSAFE) {
1297 spin_unlock(&base->lock);
1298 call_timer_fn(timer, fn, data);
1299 spin_lock(&base->lock);
1301 spin_unlock_irq(&base->lock);
1302 call_timer_fn(timer, fn, data);
1303 spin_lock_irq(&base->lock);
1308 static int __collect_expired_timers(struct timer_base *base,
1309 struct hlist_head *heads)
1311 unsigned long clk = base->clk;
1312 struct hlist_head *vec;
1316 for (i = 0; i < LVL_DEPTH; i++) {
1317 idx = (clk & LVL_MASK) + i * LVL_SIZE;
1319 if (__test_and_clear_bit(idx, base->pending_map)) {
1320 vec = base->vectors + idx;
1321 hlist_move_list(vec, heads++);
1324 /* Is it time to look at the next level? */
1325 if (clk & LVL_CLK_MASK)
1327 /* Shift clock for the next level granularity */
1328 clk >>= LVL_CLK_SHIFT;
1333 #ifdef CONFIG_NO_HZ_COMMON
1335 * Find the next pending bucket of a level. Search from level start (@offset)
1336 * + @clk upwards and if nothing there, search from start of the level
1337 * (@offset) up to @offset + clk.
1339 static int next_pending_bucket(struct timer_base *base, unsigned offset,
1342 unsigned pos, start = offset + clk;
1343 unsigned end = offset + LVL_SIZE;
1345 pos = find_next_bit(base->pending_map, end, start);
1349 pos = find_next_bit(base->pending_map, start, offset);
1350 return pos < start ? pos + LVL_SIZE - start : -1;
1354 * Search the first expiring timer in the various clock levels. Caller must
1357 static unsigned long __next_timer_interrupt(struct timer_base *base)
1359 unsigned long clk, next, adj;
1360 unsigned lvl, offset = 0;
1362 next = base->clk + NEXT_TIMER_MAX_DELTA;
1364 for (lvl = 0; lvl < LVL_DEPTH; lvl++, offset += LVL_SIZE) {
1365 int pos = next_pending_bucket(base, offset, clk & LVL_MASK);
1368 unsigned long tmp = clk + (unsigned long) pos;
1370 tmp <<= LVL_SHIFT(lvl);
1371 if (time_before(tmp, next))
1375 * Clock for the next level. If the current level clock lower
1376 * bits are zero, we look at the next level as is. If not we
1377 * need to advance it by one because that's going to be the
1378 * next expiring bucket in that level. base->clk is the next
1379 * expiring jiffie. So in case of:
1381 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1384 * we have to look at all levels @index 0. With
1386 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1389 * LVL0 has the next expiring bucket @index 2. The upper
1390 * levels have the next expiring bucket @index 1.
1392 * In case that the propagation wraps the next level the same
1395 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1398 * So after looking at LVL0 we get:
1400 * LVL5 LVL4 LVL3 LVL2 LVL1
1403 * So no propagation from LVL1 to LVL2 because that happened
1404 * with the add already, but then we need to propagate further
1405 * from LVL2 to LVL3.
1407 * So the simple check whether the lower bits of the current
1408 * level are 0 or not is sufficient for all cases.
1410 adj = clk & LVL_CLK_MASK ? 1 : 0;
1411 clk >>= LVL_CLK_SHIFT;
1418 * Check, if the next hrtimer event is before the next timer wheel
1421 static u64 cmp_next_hrtimer_event(u64 basem, u64 expires)
1423 u64 nextevt = hrtimer_get_next_event();
1426 * If high resolution timers are enabled
1427 * hrtimer_get_next_event() returns KTIME_MAX.
1429 if (expires <= nextevt)
1433 * If the next timer is already expired, return the tick base
1434 * time so the tick is fired immediately.
1436 if (nextevt <= basem)
1440 * Round up to the next jiffie. High resolution timers are
1441 * off, so the hrtimers are expired in the tick and we need to
1442 * make sure that this tick really expires the timer to avoid
1443 * a ping pong of the nohz stop code.
1445 * Use DIV_ROUND_UP_ULL to prevent gcc calling __divdi3
1447 return DIV_ROUND_UP_ULL(nextevt, TICK_NSEC) * TICK_NSEC;
1451 * get_next_timer_interrupt - return the time (clock mono) of the next timer
1452 * @basej: base time jiffies
1453 * @basem: base time clock monotonic
1455 * Returns the tick aligned clock monotonic time of the next pending
1456 * timer or KTIME_MAX if no timer is pending.
1458 u64 get_next_timer_interrupt(unsigned long basej, u64 basem)
1460 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1461 u64 expires = KTIME_MAX;
1462 unsigned long nextevt;
1465 * Pretend that there is no timer pending if the cpu is offline.
1466 * Possible pending timers will be migrated later to an active cpu.
1468 if (cpu_is_offline(smp_processor_id()))
1471 spin_lock(&base->lock);
1472 nextevt = __next_timer_interrupt(base);
1473 base->next_expiry = nextevt;
1475 * We have a fresh next event. Check whether we can forward the base:
1477 if (time_after(nextevt, jiffies))
1478 base->clk = jiffies;
1479 else if (time_after(nextevt, base->clk))
1480 base->clk = nextevt;
1482 if (time_before_eq(nextevt, basej)) {
1484 base->is_idle = false;
1486 expires = basem + (nextevt - basej) * TICK_NSEC;
1488 * If we expect to sleep more than a tick, mark the base idle:
1490 if ((expires - basem) > TICK_NSEC)
1491 base->is_idle = true;
1493 spin_unlock(&base->lock);
1495 return cmp_next_hrtimer_event(basem, expires);
1499 * timer_clear_idle - Clear the idle state of the timer base
1501 * Called with interrupts disabled
1503 void timer_clear_idle(void)
1505 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1508 * We do this unlocked. The worst outcome is a remote enqueue sending
1509 * a pointless IPI, but taking the lock would just make the window for
1510 * sending the IPI a few instructions smaller for the cost of taking
1511 * the lock in the exit from idle path.
1513 base->is_idle = false;
1516 static int collect_expired_timers(struct timer_base *base,
1517 struct hlist_head *heads)
1520 * NOHZ optimization. After a long idle sleep we need to forward the
1521 * base to current jiffies. Avoid a loop by searching the bitfield for
1522 * the next expiring timer.
1524 if ((long)(jiffies - base->clk) > 2) {
1525 unsigned long next = __next_timer_interrupt(base);
1528 * If the next timer is ahead of time forward to current
1529 * jiffies, otherwise forward to the next expiry time:
1531 if (time_after(next, jiffies)) {
1532 /* The call site will increment clock! */
1533 base->clk = jiffies - 1;
1538 return __collect_expired_timers(base, heads);
1541 static inline int collect_expired_timers(struct timer_base *base,
1542 struct hlist_head *heads)
1544 return __collect_expired_timers(base, heads);
1549 * Called from the timer interrupt handler to charge one tick to the current
1550 * process. user_tick is 1 if the tick is user time, 0 for system.
1552 void update_process_times(int user_tick)
1554 struct task_struct *p = current;
1556 /* Note: this timer irq context must be accounted for as well. */
1557 account_process_tick(p, user_tick);
1559 rcu_check_callbacks(user_tick);
1560 #ifdef CONFIG_IRQ_WORK
1565 run_posix_cpu_timers(p);
1569 * __run_timers - run all expired timers (if any) on this CPU.
1570 * @base: the timer vector to be processed.
1572 static inline void __run_timers(struct timer_base *base)
1574 struct hlist_head heads[LVL_DEPTH];
1577 if (!time_after_eq(jiffies, base->clk))
1580 spin_lock_irq(&base->lock);
1582 while (time_after_eq(jiffies, base->clk)) {
1584 levels = collect_expired_timers(base, heads);
1588 expire_timers(base, heads + levels);
1590 base->running_timer = NULL;
1591 spin_unlock_irq(&base->lock);
1595 * This function runs timers and the timer-tq in bottom half context.
1597 static void run_timer_softirq(struct softirq_action *h)
1599 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1602 if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && base->nohz_active)
1603 __run_timers(this_cpu_ptr(&timer_bases[BASE_DEF]));
1607 * Called by the local, per-CPU timer interrupt on SMP.
1609 void run_local_timers(void)
1611 hrtimer_run_queues();
1612 raise_softirq(TIMER_SOFTIRQ);
1615 #ifdef __ARCH_WANT_SYS_ALARM
1618 * For backwards compatibility? This can be done in libc so Alpha
1619 * and all newer ports shouldn't need it.
1621 SYSCALL_DEFINE1(alarm, unsigned int, seconds)
1623 return alarm_setitimer(seconds);
1628 static void process_timeout(unsigned long __data)
1630 wake_up_process((struct task_struct *)__data);
1634 * schedule_timeout - sleep until timeout
1635 * @timeout: timeout value in jiffies
1637 * Make the current task sleep until @timeout jiffies have
1638 * elapsed. The routine will return immediately unless
1639 * the current task state has been set (see set_current_state()).
1641 * You can set the task state as follows -
1643 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1644 * pass before the routine returns. The routine will return 0
1646 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1647 * delivered to the current task. In this case the remaining time
1648 * in jiffies will be returned, or 0 if the timer expired in time
1650 * The current task state is guaranteed to be TASK_RUNNING when this
1653 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1654 * the CPU away without a bound on the timeout. In this case the return
1655 * value will be %MAX_SCHEDULE_TIMEOUT.
1657 * In all cases the return value is guaranteed to be non-negative.
1659 signed long __sched schedule_timeout(signed long timeout)
1661 struct timer_list timer;
1662 unsigned long expire;
1666 case MAX_SCHEDULE_TIMEOUT:
1668 * These two special cases are useful to be comfortable
1669 * in the caller. Nothing more. We could take
1670 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1671 * but I' d like to return a valid offset (>=0) to allow
1672 * the caller to do everything it want with the retval.
1678 * Another bit of PARANOID. Note that the retval will be
1679 * 0 since no piece of kernel is supposed to do a check
1680 * for a negative retval of schedule_timeout() (since it
1681 * should never happens anyway). You just have the printk()
1682 * that will tell you if something is gone wrong and where.
1685 printk(KERN_ERR "schedule_timeout: wrong timeout "
1686 "value %lx\n", timeout);
1688 current->state = TASK_RUNNING;
1693 expire = timeout + jiffies;
1695 setup_timer_on_stack(&timer, process_timeout, (unsigned long)current);
1696 __mod_timer(&timer, expire, false);
1698 del_singleshot_timer_sync(&timer);
1700 /* Remove the timer from the object tracker */
1701 destroy_timer_on_stack(&timer);
1703 timeout = expire - jiffies;
1706 return timeout < 0 ? 0 : timeout;
1708 EXPORT_SYMBOL(schedule_timeout);
1711 * We can use __set_current_state() here because schedule_timeout() calls
1712 * schedule() unconditionally.
1714 signed long __sched schedule_timeout_interruptible(signed long timeout)
1716 __set_current_state(TASK_INTERRUPTIBLE);
1717 return schedule_timeout(timeout);
1719 EXPORT_SYMBOL(schedule_timeout_interruptible);
1721 signed long __sched schedule_timeout_killable(signed long timeout)
1723 __set_current_state(TASK_KILLABLE);
1724 return schedule_timeout(timeout);
1726 EXPORT_SYMBOL(schedule_timeout_killable);
1728 signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1730 __set_current_state(TASK_UNINTERRUPTIBLE);
1731 return schedule_timeout(timeout);
1733 EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1736 * Like schedule_timeout_uninterruptible(), except this task will not contribute
1739 signed long __sched schedule_timeout_idle(signed long timeout)
1741 __set_current_state(TASK_IDLE);
1742 return schedule_timeout(timeout);
1744 EXPORT_SYMBOL(schedule_timeout_idle);
1746 #ifdef CONFIG_HOTPLUG_CPU
1747 static void migrate_timer_list(struct timer_base *new_base, struct hlist_head *head)
1749 struct timer_list *timer;
1750 int cpu = new_base->cpu;
1752 while (!hlist_empty(head)) {
1753 timer = hlist_entry(head->first, struct timer_list, entry);
1754 detach_timer(timer, false);
1755 timer->flags = (timer->flags & ~TIMER_BASEMASK) | cpu;
1756 internal_add_timer(new_base, timer);
1760 static void migrate_timers(int cpu)
1762 struct timer_base *old_base;
1763 struct timer_base *new_base;
1766 BUG_ON(cpu_online(cpu));
1768 for (b = 0; b < NR_BASES; b++) {
1769 old_base = per_cpu_ptr(&timer_bases[b], cpu);
1770 new_base = get_cpu_ptr(&timer_bases[b]);
1772 * The caller is globally serialized and nobody else
1773 * takes two locks at once, deadlock is not possible.
1775 spin_lock_irq(&new_base->lock);
1776 spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1778 BUG_ON(old_base->running_timer);
1780 for (i = 0; i < WHEEL_SIZE; i++)
1781 migrate_timer_list(new_base, old_base->vectors + i);
1783 spin_unlock(&old_base->lock);
1784 spin_unlock_irq(&new_base->lock);
1785 put_cpu_ptr(&timer_bases);
1789 static int timer_cpu_notify(struct notifier_block *self,
1790 unsigned long action, void *hcpu)
1794 case CPU_DEAD_FROZEN:
1795 migrate_timers((long)hcpu);
1804 static inline void timer_register_cpu_notifier(void)
1806 cpu_notifier(timer_cpu_notify, 0);
1809 static inline void timer_register_cpu_notifier(void) { }
1810 #endif /* CONFIG_HOTPLUG_CPU */
1812 static void __init init_timer_cpu(int cpu)
1814 struct timer_base *base;
1817 for (i = 0; i < NR_BASES; i++) {
1818 base = per_cpu_ptr(&timer_bases[i], cpu);
1820 spin_lock_init(&base->lock);
1821 base->clk = jiffies;
1825 static void __init init_timer_cpus(void)
1829 for_each_possible_cpu(cpu)
1830 init_timer_cpu(cpu);
1833 void __init init_timers(void)
1837 timer_register_cpu_notifier();
1838 open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
1842 * msleep - sleep safely even with waitqueue interruptions
1843 * @msecs: Time in milliseconds to sleep for
1845 void msleep(unsigned int msecs)
1847 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1850 timeout = schedule_timeout_uninterruptible(timeout);
1853 EXPORT_SYMBOL(msleep);
1856 * msleep_interruptible - sleep waiting for signals
1857 * @msecs: Time in milliseconds to sleep for
1859 unsigned long msleep_interruptible(unsigned int msecs)
1861 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1863 while (timeout && !signal_pending(current))
1864 timeout = schedule_timeout_interruptible(timeout);
1865 return jiffies_to_msecs(timeout);
1868 EXPORT_SYMBOL(msleep_interruptible);
1870 static void __sched do_usleep_range(unsigned long min, unsigned long max)
1875 kmin = ktime_set(0, min * NSEC_PER_USEC);
1876 delta = (u64)(max - min) * NSEC_PER_USEC;
1877 schedule_hrtimeout_range(&kmin, delta, HRTIMER_MODE_REL);
1881 * usleep_range - Drop in replacement for udelay where wakeup is flexible
1882 * @min: Minimum time in usecs to sleep
1883 * @max: Maximum time in usecs to sleep
1885 void __sched usleep_range(unsigned long min, unsigned long max)
1887 __set_current_state(TASK_UNINTERRUPTIBLE);
1888 do_usleep_range(min, max);
1890 EXPORT_SYMBOL(usleep_range);