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;
200 bool migration_enabled;
202 DECLARE_BITMAP(pending_map, WHEEL_SIZE);
203 struct hlist_head vectors[WHEEL_SIZE];
204 } ____cacheline_aligned;
206 static DEFINE_PER_CPU(struct timer_base, timer_bases[NR_BASES]);
208 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
209 unsigned int sysctl_timer_migration = 1;
211 void timers_update_migration(bool update_nohz)
213 bool on = sysctl_timer_migration && tick_nohz_active;
216 /* Avoid the loop, if nothing to update */
217 if (this_cpu_read(timer_bases[BASE_STD].migration_enabled) == on)
220 for_each_possible_cpu(cpu) {
221 per_cpu(timer_bases[BASE_STD].migration_enabled, cpu) = on;
222 per_cpu(timer_bases[BASE_DEF].migration_enabled, cpu) = on;
223 per_cpu(hrtimer_bases.migration_enabled, cpu) = on;
226 per_cpu(timer_bases[BASE_STD].nohz_active, cpu) = true;
227 per_cpu(timer_bases[BASE_DEF].nohz_active, cpu) = true;
228 per_cpu(hrtimer_bases.nohz_active, cpu) = true;
232 int timer_migration_handler(struct ctl_table *table, int write,
233 void __user *buffer, size_t *lenp,
236 static DEFINE_MUTEX(mutex);
240 ret = proc_dointvec(table, write, buffer, lenp, ppos);
242 timers_update_migration(false);
243 mutex_unlock(&mutex);
248 static unsigned long round_jiffies_common(unsigned long j, int cpu,
252 unsigned long original = j;
255 * We don't want all cpus firing their timers at once hitting the
256 * same lock or cachelines, so we skew each extra cpu with an extra
257 * 3 jiffies. This 3 jiffies came originally from the mm/ code which
259 * The skew is done by adding 3*cpunr, then round, then subtract this
260 * extra offset again.
267 * If the target jiffie is just after a whole second (which can happen
268 * due to delays of the timer irq, long irq off times etc etc) then
269 * we should round down to the whole second, not up. Use 1/4th second
270 * as cutoff for this rounding as an extreme upper bound for this.
271 * But never round down if @force_up is set.
273 if (rem < HZ/4 && !force_up) /* round down */
278 /* now that we have rounded, subtract the extra skew again */
282 * Make sure j is still in the future. Otherwise return the
285 return time_is_after_jiffies(j) ? j : original;
289 * __round_jiffies - function to round jiffies to a full second
290 * @j: the time in (absolute) jiffies that should be rounded
291 * @cpu: the processor number on which the timeout will happen
293 * __round_jiffies() rounds an absolute time in the future (in jiffies)
294 * up or down to (approximately) full seconds. This is useful for timers
295 * for which the exact time they fire does not matter too much, as long as
296 * they fire approximately every X seconds.
298 * By rounding these timers to whole seconds, all such timers will fire
299 * at the same time, rather than at various times spread out. The goal
300 * of this is to have the CPU wake up less, which saves power.
302 * The exact rounding is skewed for each processor to avoid all
303 * processors firing at the exact same time, which could lead
304 * to lock contention or spurious cache line bouncing.
306 * The return value is the rounded version of the @j parameter.
308 unsigned long __round_jiffies(unsigned long j, int cpu)
310 return round_jiffies_common(j, cpu, false);
312 EXPORT_SYMBOL_GPL(__round_jiffies);
315 * __round_jiffies_relative - function to round jiffies to a full second
316 * @j: the time in (relative) jiffies that should be rounded
317 * @cpu: the processor number on which the timeout will happen
319 * __round_jiffies_relative() rounds a time delta in the future (in jiffies)
320 * up or down to (approximately) full seconds. This is useful for timers
321 * for which the exact time they fire does not matter too much, as long as
322 * they fire approximately every X seconds.
324 * By rounding these timers to whole seconds, all such timers will fire
325 * at the same time, rather than at various times spread out. The goal
326 * of this is to have the CPU wake up less, which saves power.
328 * The exact rounding is skewed for each processor to avoid all
329 * processors firing at the exact same time, which could lead
330 * to lock contention or spurious cache line bouncing.
332 * The return value is the rounded version of the @j parameter.
334 unsigned long __round_jiffies_relative(unsigned long j, int cpu)
336 unsigned long j0 = jiffies;
338 /* Use j0 because jiffies might change while we run */
339 return round_jiffies_common(j + j0, cpu, false) - j0;
341 EXPORT_SYMBOL_GPL(__round_jiffies_relative);
344 * round_jiffies - function to round jiffies to a full second
345 * @j: the time in (absolute) jiffies that should be rounded
347 * round_jiffies() rounds an absolute time in the future (in jiffies)
348 * up or down to (approximately) full seconds. This is useful for timers
349 * for which the exact time they fire does not matter too much, as long as
350 * they fire approximately every X seconds.
352 * By rounding these timers to whole seconds, all such timers will fire
353 * at the same time, rather than at various times spread out. The goal
354 * of this is to have the CPU wake up less, which saves power.
356 * The return value is the rounded version of the @j parameter.
358 unsigned long round_jiffies(unsigned long j)
360 return round_jiffies_common(j, raw_smp_processor_id(), false);
362 EXPORT_SYMBOL_GPL(round_jiffies);
365 * round_jiffies_relative - function to round jiffies to a full second
366 * @j: the time in (relative) jiffies that should be rounded
368 * round_jiffies_relative() rounds a time delta in the future (in jiffies)
369 * up or down to (approximately) full seconds. This is useful for timers
370 * for which the exact time they fire does not matter too much, as long as
371 * they fire approximately every X seconds.
373 * By rounding these timers to whole seconds, all such timers will fire
374 * at the same time, rather than at various times spread out. The goal
375 * of this is to have the CPU wake up less, which saves power.
377 * The return value is the rounded version of the @j parameter.
379 unsigned long round_jiffies_relative(unsigned long j)
381 return __round_jiffies_relative(j, raw_smp_processor_id());
383 EXPORT_SYMBOL_GPL(round_jiffies_relative);
386 * __round_jiffies_up - function to round jiffies up to a full second
387 * @j: the time in (absolute) jiffies that should be rounded
388 * @cpu: the processor number on which the timeout will happen
390 * This is the same as __round_jiffies() except that it will never
391 * round down. This is useful for timeouts for which the exact time
392 * of firing does not matter too much, as long as they don't fire too
395 unsigned long __round_jiffies_up(unsigned long j, int cpu)
397 return round_jiffies_common(j, cpu, true);
399 EXPORT_SYMBOL_GPL(__round_jiffies_up);
402 * __round_jiffies_up_relative - function to round jiffies up to a full second
403 * @j: the time in (relative) jiffies that should be rounded
404 * @cpu: the processor number on which the timeout will happen
406 * This is the same as __round_jiffies_relative() except that it will never
407 * round down. This is useful for timeouts for which the exact time
408 * of firing does not matter too much, as long as they don't fire too
411 unsigned long __round_jiffies_up_relative(unsigned long j, int cpu)
413 unsigned long j0 = jiffies;
415 /* Use j0 because jiffies might change while we run */
416 return round_jiffies_common(j + j0, cpu, true) - j0;
418 EXPORT_SYMBOL_GPL(__round_jiffies_up_relative);
421 * round_jiffies_up - function to round jiffies up to a full second
422 * @j: the time in (absolute) jiffies that should be rounded
424 * This is the same as round_jiffies() except that it will never
425 * round down. This is useful for timeouts for which the exact time
426 * of firing does not matter too much, as long as they don't fire too
429 unsigned long round_jiffies_up(unsigned long j)
431 return round_jiffies_common(j, raw_smp_processor_id(), true);
433 EXPORT_SYMBOL_GPL(round_jiffies_up);
436 * round_jiffies_up_relative - function to round jiffies up to a full second
437 * @j: the time in (relative) jiffies that should be rounded
439 * This is the same as round_jiffies_relative() except that it will never
440 * round down. This is useful for timeouts for which the exact time
441 * of firing does not matter too much, as long as they don't fire too
444 unsigned long round_jiffies_up_relative(unsigned long j)
446 return __round_jiffies_up_relative(j, raw_smp_processor_id());
448 EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
451 * set_timer_slack - set the allowed slack for a timer
452 * @timer: the timer to be modified
453 * @slack_hz: the amount of time (in jiffies) allowed for rounding
455 * Set the amount of time, in jiffies, that a certain timer has
456 * in terms of slack. By setting this value, the timer subsystem
457 * will schedule the actual timer somewhere between
458 * the time mod_timer() asks for, and that time plus the slack.
460 * By setting the slack to -1, a percentage of the delay is used
463 void set_timer_slack(struct timer_list *timer, int slack_hz)
465 timer->slack = slack_hz;
467 EXPORT_SYMBOL_GPL(set_timer_slack);
469 static inline unsigned int timer_get_idx(struct timer_list *timer)
471 return (timer->flags & TIMER_ARRAYMASK) >> TIMER_ARRAYSHIFT;
474 static inline void timer_set_idx(struct timer_list *timer, unsigned int idx)
476 timer->flags = (timer->flags & ~TIMER_ARRAYMASK) |
477 idx << TIMER_ARRAYSHIFT;
481 * Helper function to calculate the array index for a given expiry
484 static inline unsigned calc_index(unsigned expires, unsigned lvl)
486 expires = (expires + LVL_GRAN(lvl)) >> LVL_SHIFT(lvl);
487 return LVL_OFFS(lvl) + (expires & LVL_MASK);
491 __internal_add_timer(struct timer_base *base, struct timer_list *timer)
493 unsigned long expires = timer->expires;
494 unsigned long delta = expires - base->clk;
495 struct hlist_head *vec;
498 if (delta < LVL_START(1)) {
499 idx = calc_index(expires, 0);
500 } else if (delta < LVL_START(2)) {
501 idx = calc_index(expires, 1);
502 } else if (delta < LVL_START(3)) {
503 idx = calc_index(expires, 2);
504 } else if (delta < LVL_START(4)) {
505 idx = calc_index(expires, 3);
506 } else if (delta < LVL_START(5)) {
507 idx = calc_index(expires, 4);
508 } else if (delta < LVL_START(6)) {
509 idx = calc_index(expires, 5);
510 } else if (delta < LVL_START(7)) {
511 idx = calc_index(expires, 6);
512 } else if (LVL_DEPTH > 8 && delta < LVL_START(8)) {
513 idx = calc_index(expires, 7);
514 } else if ((long) delta < 0) {
515 idx = base->clk & LVL_MASK;
518 * Force expire obscene large timeouts to expire at the
519 * capacity limit of the wheel.
521 if (expires >= WHEEL_TIMEOUT_CUTOFF)
522 expires = WHEEL_TIMEOUT_MAX;
524 idx = calc_index(expires, LVL_DEPTH - 1);
527 * Enqueue the timer into the array bucket, mark it pending in
528 * the bitmap and store the index in the timer flags.
530 vec = base->vectors + idx;
531 hlist_add_head(&timer->entry, vec);
532 __set_bit(idx, base->pending_map);
533 timer_set_idx(timer, idx);
536 static void internal_add_timer(struct timer_base *base, struct timer_list *timer)
538 __internal_add_timer(base, timer);
541 * Check whether the other CPU is in dynticks mode and needs
542 * to be triggered to reevaluate the timer wheel. We are
543 * protected against the other CPU fiddling with the timer by
544 * holding the timer base lock. This also makes sure that a
545 * CPU on the way to stop its tick can not evaluate the timer
548 * Spare the IPI for deferrable timers on idle targets though.
549 * The next busy ticks will take care of it. Except full dynticks
550 * require special care against races with idle_cpu(), lets deal
553 if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && base->nohz_active) {
554 if (!(timer->flags & TIMER_DEFERRABLE) ||
555 tick_nohz_full_cpu(base->cpu))
556 wake_up_nohz_cpu(base->cpu);
560 #ifdef CONFIG_TIMER_STATS
561 void __timer_stats_timer_set_start_info(struct timer_list *timer, void *addr)
563 if (timer->start_site)
566 timer->start_site = addr;
567 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
568 timer->start_pid = current->pid;
571 static void timer_stats_account_timer(struct timer_list *timer)
576 * start_site can be concurrently reset by
577 * timer_stats_timer_clear_start_info()
579 site = READ_ONCE(timer->start_site);
583 timer_stats_update_stats(timer, timer->start_pid, site,
584 timer->function, timer->start_comm,
589 static void timer_stats_account_timer(struct timer_list *timer) {}
592 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
594 static struct debug_obj_descr timer_debug_descr;
596 static void *timer_debug_hint(void *addr)
598 return ((struct timer_list *) addr)->function;
601 static bool timer_is_static_object(void *addr)
603 struct timer_list *timer = addr;
605 return (timer->entry.pprev == NULL &&
606 timer->entry.next == TIMER_ENTRY_STATIC);
610 * fixup_init is called when:
611 * - an active object is initialized
613 static bool timer_fixup_init(void *addr, enum debug_obj_state state)
615 struct timer_list *timer = addr;
618 case ODEBUG_STATE_ACTIVE:
619 del_timer_sync(timer);
620 debug_object_init(timer, &timer_debug_descr);
627 /* Stub timer callback for improperly used timers. */
628 static void stub_timer(unsigned long data)
634 * fixup_activate is called when:
635 * - an active object is activated
636 * - an unknown non-static object is activated
638 static bool timer_fixup_activate(void *addr, enum debug_obj_state state)
640 struct timer_list *timer = addr;
643 case ODEBUG_STATE_NOTAVAILABLE:
644 setup_timer(timer, stub_timer, 0);
647 case ODEBUG_STATE_ACTIVE:
656 * fixup_free is called when:
657 * - an active object is freed
659 static bool timer_fixup_free(void *addr, enum debug_obj_state state)
661 struct timer_list *timer = addr;
664 case ODEBUG_STATE_ACTIVE:
665 del_timer_sync(timer);
666 debug_object_free(timer, &timer_debug_descr);
674 * fixup_assert_init is called when:
675 * - an untracked/uninit-ed object is found
677 static bool timer_fixup_assert_init(void *addr, enum debug_obj_state state)
679 struct timer_list *timer = addr;
682 case ODEBUG_STATE_NOTAVAILABLE:
683 setup_timer(timer, stub_timer, 0);
690 static struct debug_obj_descr timer_debug_descr = {
691 .name = "timer_list",
692 .debug_hint = timer_debug_hint,
693 .is_static_object = timer_is_static_object,
694 .fixup_init = timer_fixup_init,
695 .fixup_activate = timer_fixup_activate,
696 .fixup_free = timer_fixup_free,
697 .fixup_assert_init = timer_fixup_assert_init,
700 static inline void debug_timer_init(struct timer_list *timer)
702 debug_object_init(timer, &timer_debug_descr);
705 static inline void debug_timer_activate(struct timer_list *timer)
707 debug_object_activate(timer, &timer_debug_descr);
710 static inline void debug_timer_deactivate(struct timer_list *timer)
712 debug_object_deactivate(timer, &timer_debug_descr);
715 static inline void debug_timer_free(struct timer_list *timer)
717 debug_object_free(timer, &timer_debug_descr);
720 static inline void debug_timer_assert_init(struct timer_list *timer)
722 debug_object_assert_init(timer, &timer_debug_descr);
725 static void do_init_timer(struct timer_list *timer, unsigned int flags,
726 const char *name, struct lock_class_key *key);
728 void init_timer_on_stack_key(struct timer_list *timer, unsigned int flags,
729 const char *name, struct lock_class_key *key)
731 debug_object_init_on_stack(timer, &timer_debug_descr);
732 do_init_timer(timer, flags, name, key);
734 EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
736 void destroy_timer_on_stack(struct timer_list *timer)
738 debug_object_free(timer, &timer_debug_descr);
740 EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
743 static inline void debug_timer_init(struct timer_list *timer) { }
744 static inline void debug_timer_activate(struct timer_list *timer) { }
745 static inline void debug_timer_deactivate(struct timer_list *timer) { }
746 static inline void debug_timer_assert_init(struct timer_list *timer) { }
749 static inline void debug_init(struct timer_list *timer)
751 debug_timer_init(timer);
752 trace_timer_init(timer);
756 debug_activate(struct timer_list *timer, unsigned long expires)
758 debug_timer_activate(timer);
759 trace_timer_start(timer, expires, timer->flags);
762 static inline void debug_deactivate(struct timer_list *timer)
764 debug_timer_deactivate(timer);
765 trace_timer_cancel(timer);
768 static inline void debug_assert_init(struct timer_list *timer)
770 debug_timer_assert_init(timer);
773 static void do_init_timer(struct timer_list *timer, unsigned int flags,
774 const char *name, struct lock_class_key *key)
776 timer->entry.pprev = NULL;
777 timer->flags = flags | raw_smp_processor_id();
779 #ifdef CONFIG_TIMER_STATS
780 timer->start_site = NULL;
781 timer->start_pid = -1;
782 memset(timer->start_comm, 0, TASK_COMM_LEN);
784 lockdep_init_map(&timer->lockdep_map, name, key, 0);
788 * init_timer_key - initialize a timer
789 * @timer: the timer to be initialized
790 * @flags: timer flags
791 * @name: name of the timer
792 * @key: lockdep class key of the fake lock used for tracking timer
793 * sync lock dependencies
795 * init_timer_key() must be done to a timer prior calling *any* of the
796 * other timer functions.
798 void init_timer_key(struct timer_list *timer, unsigned int flags,
799 const char *name, struct lock_class_key *key)
802 do_init_timer(timer, flags, name, key);
804 EXPORT_SYMBOL(init_timer_key);
806 static inline void detach_timer(struct timer_list *timer, bool clear_pending)
808 struct hlist_node *entry = &timer->entry;
810 debug_deactivate(timer);
815 entry->next = LIST_POISON2;
818 static int detach_if_pending(struct timer_list *timer, struct timer_base *base,
821 unsigned idx = timer_get_idx(timer);
823 if (!timer_pending(timer))
826 if (hlist_is_singular_node(&timer->entry, base->vectors + idx))
827 __clear_bit(idx, base->pending_map);
829 detach_timer(timer, clear_pending);
833 static inline struct timer_base *get_timer_cpu_base(u32 tflags, u32 cpu)
835 struct timer_base *base = per_cpu_ptr(&timer_bases[BASE_STD], cpu);
838 * If the timer is deferrable and nohz is active then we need to use
839 * the deferrable base.
841 if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && base->nohz_active &&
842 (tflags & TIMER_DEFERRABLE))
843 base = per_cpu_ptr(&timer_bases[BASE_DEF], cpu);
847 static inline struct timer_base *get_timer_this_cpu_base(u32 tflags)
849 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
852 * If the timer is deferrable and nohz is active then we need to use
853 * the deferrable base.
855 if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && base->nohz_active &&
856 (tflags & TIMER_DEFERRABLE))
857 base = this_cpu_ptr(&timer_bases[BASE_DEF]);
861 static inline struct timer_base *get_timer_base(u32 tflags)
863 return get_timer_cpu_base(tflags, tflags & TIMER_CPUMASK);
866 static inline struct timer_base *get_target_base(struct timer_base *base,
869 #if defined(CONFIG_NO_HZ_COMMON) && defined(CONFIG_SMP)
870 if ((tflags & TIMER_PINNED) || !base->migration_enabled)
871 return get_timer_this_cpu_base(tflags);
872 return get_timer_cpu_base(tflags, get_nohz_timer_target());
874 return get_timer_this_cpu_base(tflags);
879 * We are using hashed locking: Holding per_cpu(timer_bases[x]).lock means
880 * that all timers which are tied to this base are locked, and the base itself
883 * So __run_timers/migrate_timers can safely modify all timers which could
884 * be found in the base->vectors array.
886 * When a timer is migrating then the TIMER_MIGRATING flag is set and we need
887 * to wait until the migration is done.
889 static struct timer_base *lock_timer_base(struct timer_list *timer,
890 unsigned long *flags)
891 __acquires(timer->base->lock)
894 struct timer_base *base;
895 u32 tf = timer->flags;
897 if (!(tf & TIMER_MIGRATING)) {
898 base = get_timer_base(tf);
899 spin_lock_irqsave(&base->lock, *flags);
900 if (timer->flags == tf)
902 spin_unlock_irqrestore(&base->lock, *flags);
909 __mod_timer(struct timer_list *timer, unsigned long expires, bool pending_only)
911 struct timer_base *base, *new_base;
916 * TODO: Calculate the array bucket of the timer right here w/o
917 * holding the base lock. This allows to check not only
918 * timer->expires == expires below, but also whether the timer
919 * ends up in the same bucket. If we really need to requeue
920 * the timer then we check whether base->clk have
921 * advanced between here and locking the timer base. If
922 * jiffies advanced we have to recalc the array bucket with the
927 * This is a common optimization triggered by the
928 * networking code - if the timer is re-modified
929 * to be the same thing then just return:
931 if (timer_pending(timer)) {
932 if (timer->expires == expires)
936 timer_stats_timer_set_start_info(timer);
937 BUG_ON(!timer->function);
939 base = lock_timer_base(timer, &flags);
941 ret = detach_if_pending(timer, base, false);
942 if (!ret && pending_only)
945 debug_activate(timer, expires);
947 new_base = get_target_base(base, timer->flags);
949 if (base != new_base) {
951 * We are trying to schedule the timer on the new base.
952 * However we can't change timer's base while it is running,
953 * otherwise del_timer_sync() can't detect that the timer's
954 * handler yet has not finished. This also guarantees that the
955 * timer is serialized wrt itself.
957 if (likely(base->running_timer != timer)) {
958 /* See the comment in lock_timer_base() */
959 timer->flags |= TIMER_MIGRATING;
961 spin_unlock(&base->lock);
963 spin_lock(&base->lock);
964 WRITE_ONCE(timer->flags,
965 (timer->flags & ~TIMER_BASEMASK) | base->cpu);
969 timer->expires = expires;
970 internal_add_timer(base, timer);
973 spin_unlock_irqrestore(&base->lock, flags);
979 * mod_timer_pending - modify a pending timer's timeout
980 * @timer: the pending timer to be modified
981 * @expires: new timeout in jiffies
983 * mod_timer_pending() is the same for pending timers as mod_timer(),
984 * but will not re-activate and modify already deleted timers.
986 * It is useful for unserialized use of timers.
988 int mod_timer_pending(struct timer_list *timer, unsigned long expires)
990 return __mod_timer(timer, expires, true);
992 EXPORT_SYMBOL(mod_timer_pending);
995 * mod_timer - modify a timer's timeout
996 * @timer: the timer to be modified
997 * @expires: new timeout in jiffies
999 * mod_timer() is a more efficient way to update the expire field of an
1000 * active timer (if the timer is inactive it will be activated)
1002 * mod_timer(timer, expires) is equivalent to:
1004 * del_timer(timer); timer->expires = expires; add_timer(timer);
1006 * Note that if there are multiple unserialized concurrent users of the
1007 * same timer, then mod_timer() is the only safe way to modify the timeout,
1008 * since add_timer() cannot modify an already running timer.
1010 * The function returns whether it has modified a pending timer or not.
1011 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
1012 * active timer returns 1.)
1014 int mod_timer(struct timer_list *timer, unsigned long expires)
1016 return __mod_timer(timer, expires, false);
1018 EXPORT_SYMBOL(mod_timer);
1021 * add_timer - start a timer
1022 * @timer: the timer to be added
1024 * The kernel will do a ->function(->data) callback from the
1025 * timer interrupt at the ->expires point in the future. The
1026 * current time is 'jiffies'.
1028 * The timer's ->expires, ->function (and if the handler uses it, ->data)
1029 * fields must be set prior calling this function.
1031 * Timers with an ->expires field in the past will be executed in the next
1034 void add_timer(struct timer_list *timer)
1036 BUG_ON(timer_pending(timer));
1037 mod_timer(timer, timer->expires);
1039 EXPORT_SYMBOL(add_timer);
1042 * add_timer_on - start a timer on a particular CPU
1043 * @timer: the timer to be added
1044 * @cpu: the CPU to start it on
1046 * This is not very scalable on SMP. Double adds are not possible.
1048 void add_timer_on(struct timer_list *timer, int cpu)
1050 struct timer_base *new_base, *base;
1051 unsigned long flags;
1053 timer_stats_timer_set_start_info(timer);
1054 BUG_ON(timer_pending(timer) || !timer->function);
1056 new_base = get_timer_cpu_base(timer->flags, cpu);
1059 * If @timer was on a different CPU, it should be migrated with the
1060 * old base locked to prevent other operations proceeding with the
1061 * wrong base locked. See lock_timer_base().
1063 base = lock_timer_base(timer, &flags);
1064 if (base != new_base) {
1065 timer->flags |= TIMER_MIGRATING;
1067 spin_unlock(&base->lock);
1069 spin_lock(&base->lock);
1070 WRITE_ONCE(timer->flags,
1071 (timer->flags & ~TIMER_BASEMASK) | cpu);
1074 debug_activate(timer, timer->expires);
1075 internal_add_timer(base, timer);
1076 spin_unlock_irqrestore(&base->lock, flags);
1078 EXPORT_SYMBOL_GPL(add_timer_on);
1081 * del_timer - deactive a timer.
1082 * @timer: the timer to be deactivated
1084 * del_timer() deactivates a timer - this works on both active and inactive
1087 * The function returns whether it has deactivated a pending timer or not.
1088 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
1089 * active timer returns 1.)
1091 int del_timer(struct timer_list *timer)
1093 struct timer_base *base;
1094 unsigned long flags;
1097 debug_assert_init(timer);
1099 timer_stats_timer_clear_start_info(timer);
1100 if (timer_pending(timer)) {
1101 base = lock_timer_base(timer, &flags);
1102 ret = detach_if_pending(timer, base, true);
1103 spin_unlock_irqrestore(&base->lock, flags);
1108 EXPORT_SYMBOL(del_timer);
1111 * try_to_del_timer_sync - Try to deactivate a timer
1112 * @timer: timer do del
1114 * This function tries to deactivate a timer. Upon successful (ret >= 0)
1115 * exit the timer is not queued and the handler is not running on any CPU.
1117 int try_to_del_timer_sync(struct timer_list *timer)
1119 struct timer_base *base;
1120 unsigned long flags;
1123 debug_assert_init(timer);
1125 base = lock_timer_base(timer, &flags);
1127 if (base->running_timer != timer) {
1128 timer_stats_timer_clear_start_info(timer);
1129 ret = detach_if_pending(timer, base, true);
1131 spin_unlock_irqrestore(&base->lock, flags);
1135 EXPORT_SYMBOL(try_to_del_timer_sync);
1139 * del_timer_sync - deactivate a timer and wait for the handler to finish.
1140 * @timer: the timer to be deactivated
1142 * This function only differs from del_timer() on SMP: besides deactivating
1143 * the timer it also makes sure the handler has finished executing on other
1146 * Synchronization rules: Callers must prevent restarting of the timer,
1147 * otherwise this function is meaningless. It must not be called from
1148 * interrupt contexts unless the timer is an irqsafe one. The caller must
1149 * not hold locks which would prevent completion of the timer's
1150 * handler. The timer's handler must not call add_timer_on(). Upon exit the
1151 * timer is not queued and the handler is not running on any CPU.
1153 * Note: For !irqsafe timers, you must not hold locks that are held in
1154 * interrupt context while calling this function. Even if the lock has
1155 * nothing to do with the timer in question. Here's why:
1161 * base->running_timer = mytimer;
1162 * spin_lock_irq(somelock);
1164 * spin_lock(somelock);
1165 * del_timer_sync(mytimer);
1166 * while (base->running_timer == mytimer);
1168 * Now del_timer_sync() will never return and never release somelock.
1169 * The interrupt on the other CPU is waiting to grab somelock but
1170 * it has interrupted the softirq that CPU0 is waiting to finish.
1172 * The function returns whether it has deactivated a pending timer or not.
1174 int del_timer_sync(struct timer_list *timer)
1176 #ifdef CONFIG_LOCKDEP
1177 unsigned long flags;
1180 * If lockdep gives a backtrace here, please reference
1181 * the synchronization rules above.
1183 local_irq_save(flags);
1184 lock_map_acquire(&timer->lockdep_map);
1185 lock_map_release(&timer->lockdep_map);
1186 local_irq_restore(flags);
1189 * don't use it in hardirq context, because it
1190 * could lead to deadlock.
1192 WARN_ON(in_irq() && !(timer->flags & TIMER_IRQSAFE));
1194 int ret = try_to_del_timer_sync(timer);
1200 EXPORT_SYMBOL(del_timer_sync);
1203 static void call_timer_fn(struct timer_list *timer, void (*fn)(unsigned long),
1206 int count = preempt_count();
1208 #ifdef CONFIG_LOCKDEP
1210 * It is permissible to free the timer from inside the
1211 * function that is called from it, this we need to take into
1212 * account for lockdep too. To avoid bogus "held lock freed"
1213 * warnings as well as problems when looking into
1214 * timer->lockdep_map, make a copy and use that here.
1216 struct lockdep_map lockdep_map;
1218 lockdep_copy_map(&lockdep_map, &timer->lockdep_map);
1221 * Couple the lock chain with the lock chain at
1222 * del_timer_sync() by acquiring the lock_map around the fn()
1223 * call here and in del_timer_sync().
1225 lock_map_acquire(&lockdep_map);
1227 trace_timer_expire_entry(timer);
1229 trace_timer_expire_exit(timer);
1231 lock_map_release(&lockdep_map);
1233 if (count != preempt_count()) {
1234 WARN_ONCE(1, "timer: %pF preempt leak: %08x -> %08x\n",
1235 fn, count, preempt_count());
1237 * Restore the preempt count. That gives us a decent
1238 * chance to survive and extract information. If the
1239 * callback kept a lock held, bad luck, but not worse
1240 * than the BUG() we had.
1242 preempt_count_set(count);
1246 static void expire_timers(struct timer_base *base, struct hlist_head *head)
1248 while (!hlist_empty(head)) {
1249 struct timer_list *timer;
1250 void (*fn)(unsigned long);
1253 timer = hlist_entry(head->first, struct timer_list, entry);
1254 timer_stats_account_timer(timer);
1256 base->running_timer = timer;
1257 detach_timer(timer, true);
1259 fn = timer->function;
1262 if (timer->flags & TIMER_IRQSAFE) {
1263 spin_unlock(&base->lock);
1264 call_timer_fn(timer, fn, data);
1265 spin_lock(&base->lock);
1267 spin_unlock_irq(&base->lock);
1268 call_timer_fn(timer, fn, data);
1269 spin_lock_irq(&base->lock);
1274 static int collect_expired_timers(struct timer_base *base,
1275 struct hlist_head *heads)
1277 unsigned long clk = base->clk;
1278 struct hlist_head *vec;
1282 for (i = 0; i < LVL_DEPTH; i++) {
1283 idx = (clk & LVL_MASK) + i * LVL_SIZE;
1285 if (__test_and_clear_bit(idx, base->pending_map)) {
1286 vec = base->vectors + idx;
1287 hlist_move_list(vec, heads++);
1290 /* Is it time to look at the next level? */
1291 if (clk & LVL_CLK_MASK)
1293 /* Shift clock for the next level granularity */
1294 clk >>= LVL_CLK_SHIFT;
1300 * __run_timers - run all expired timers (if any) on this CPU.
1301 * @base: the timer vector to be processed.
1303 static inline void __run_timers(struct timer_base *base)
1305 struct hlist_head heads[LVL_DEPTH];
1308 if (!time_after_eq(jiffies, base->clk))
1311 spin_lock_irq(&base->lock);
1313 while (time_after_eq(jiffies, base->clk)) {
1315 levels = collect_expired_timers(base, heads);
1319 expire_timers(base, heads + levels);
1321 base->running_timer = NULL;
1322 spin_unlock_irq(&base->lock);
1325 #ifdef CONFIG_NO_HZ_COMMON
1327 * Find the next pending bucket of a level. Search from @offset + @clk upwards
1328 * and if nothing there, search from start of the level (@offset) up to
1331 static int next_pending_bucket(struct timer_base *base, unsigned offset,
1334 unsigned pos, start = offset + clk;
1335 unsigned end = offset + LVL_SIZE;
1337 pos = find_next_bit(base->pending_map, end, start);
1341 pos = find_next_bit(base->pending_map, start, offset);
1342 return pos < start ? pos + LVL_SIZE - start : -1;
1346 * Search the first expiring timer in the various clock levels.
1348 static unsigned long __next_timer_interrupt(struct timer_base *base)
1350 unsigned long clk, next, adj;
1351 unsigned lvl, offset = 0;
1353 spin_lock(&base->lock);
1354 next = base->clk + NEXT_TIMER_MAX_DELTA;
1356 for (lvl = 0; lvl < LVL_DEPTH; lvl++, offset += LVL_SIZE) {
1357 int pos = next_pending_bucket(base, offset, clk & LVL_MASK);
1360 unsigned long tmp = clk + (unsigned long) pos;
1362 tmp <<= LVL_SHIFT(lvl);
1363 if (time_before(tmp, next))
1367 * Clock for the next level. If the current level clock lower
1368 * bits are zero, we look at the next level as is. If not we
1369 * need to advance it by one because that's going to be the
1370 * next expiring bucket in that level. base->clk is the next
1371 * expiring jiffie. So in case of:
1373 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1376 * we have to look at all levels @index 0. With
1378 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1381 * LVL0 has the next expiring bucket @index 2. The upper
1382 * levels have the next expiring bucket @index 1.
1384 * In case that the propagation wraps the next level the same
1387 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1390 * So after looking at LVL0 we get:
1392 * LVL5 LVL4 LVL3 LVL2 LVL1
1395 * So no propagation from LVL1 to LVL2 because that happened
1396 * with the add already, but then we need to propagate further
1397 * from LVL2 to LVL3.
1399 * So the simple check whether the lower bits of the current
1400 * level are 0 or not is sufficient for all cases.
1402 adj = clk & LVL_CLK_MASK ? 1 : 0;
1403 clk >>= LVL_CLK_SHIFT;
1406 spin_unlock(&base->lock);
1411 * Check, if the next hrtimer event is before the next timer wheel
1414 static u64 cmp_next_hrtimer_event(u64 basem, u64 expires)
1416 u64 nextevt = hrtimer_get_next_event();
1419 * If high resolution timers are enabled
1420 * hrtimer_get_next_event() returns KTIME_MAX.
1422 if (expires <= nextevt)
1426 * If the next timer is already expired, return the tick base
1427 * time so the tick is fired immediately.
1429 if (nextevt <= basem)
1433 * Round up to the next jiffie. High resolution timers are
1434 * off, so the hrtimers are expired in the tick and we need to
1435 * make sure that this tick really expires the timer to avoid
1436 * a ping pong of the nohz stop code.
1438 * Use DIV_ROUND_UP_ULL to prevent gcc calling __divdi3
1440 return DIV_ROUND_UP_ULL(nextevt, TICK_NSEC) * TICK_NSEC;
1444 * get_next_timer_interrupt - return the time (clock mono) of the next timer
1445 * @basej: base time jiffies
1446 * @basem: base time clock monotonic
1448 * Returns the tick aligned clock monotonic time of the next pending
1449 * timer or KTIME_MAX if no timer is pending.
1451 u64 get_next_timer_interrupt(unsigned long basej, u64 basem)
1453 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1454 u64 expires = KTIME_MAX;
1455 unsigned long nextevt;
1458 * Pretend that there is no timer pending if the cpu is offline.
1459 * Possible pending timers will be migrated later to an active cpu.
1461 if (cpu_is_offline(smp_processor_id()))
1464 nextevt = __next_timer_interrupt(base);
1465 if (time_before_eq(nextevt, basej))
1468 expires = basem + (nextevt - basej) * TICK_NSEC;
1470 return cmp_next_hrtimer_event(basem, expires);
1475 * Called from the timer interrupt handler to charge one tick to the current
1476 * process. user_tick is 1 if the tick is user time, 0 for system.
1478 void update_process_times(int user_tick)
1480 struct task_struct *p = current;
1482 /* Note: this timer irq context must be accounted for as well. */
1483 account_process_tick(p, user_tick);
1485 rcu_check_callbacks(user_tick);
1486 #ifdef CONFIG_IRQ_WORK
1491 run_posix_cpu_timers(p);
1495 * This function runs timers and the timer-tq in bottom half context.
1497 static void run_timer_softirq(struct softirq_action *h)
1499 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1502 if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && base->nohz_active)
1503 __run_timers(this_cpu_ptr(&timer_bases[BASE_DEF]));
1507 * Called by the local, per-CPU timer interrupt on SMP.
1509 void run_local_timers(void)
1511 hrtimer_run_queues();
1512 raise_softirq(TIMER_SOFTIRQ);
1515 #ifdef __ARCH_WANT_SYS_ALARM
1518 * For backwards compatibility? This can be done in libc so Alpha
1519 * and all newer ports shouldn't need it.
1521 SYSCALL_DEFINE1(alarm, unsigned int, seconds)
1523 return alarm_setitimer(seconds);
1528 static void process_timeout(unsigned long __data)
1530 wake_up_process((struct task_struct *)__data);
1534 * schedule_timeout - sleep until timeout
1535 * @timeout: timeout value in jiffies
1537 * Make the current task sleep until @timeout jiffies have
1538 * elapsed. The routine will return immediately unless
1539 * the current task state has been set (see set_current_state()).
1541 * You can set the task state as follows -
1543 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1544 * pass before the routine returns. The routine will return 0
1546 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1547 * delivered to the current task. In this case the remaining time
1548 * in jiffies will be returned, or 0 if the timer expired in time
1550 * The current task state is guaranteed to be TASK_RUNNING when this
1553 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1554 * the CPU away without a bound on the timeout. In this case the return
1555 * value will be %MAX_SCHEDULE_TIMEOUT.
1557 * In all cases the return value is guaranteed to be non-negative.
1559 signed long __sched schedule_timeout(signed long timeout)
1561 struct timer_list timer;
1562 unsigned long expire;
1566 case MAX_SCHEDULE_TIMEOUT:
1568 * These two special cases are useful to be comfortable
1569 * in the caller. Nothing more. We could take
1570 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1571 * but I' d like to return a valid offset (>=0) to allow
1572 * the caller to do everything it want with the retval.
1578 * Another bit of PARANOID. Note that the retval will be
1579 * 0 since no piece of kernel is supposed to do a check
1580 * for a negative retval of schedule_timeout() (since it
1581 * should never happens anyway). You just have the printk()
1582 * that will tell you if something is gone wrong and where.
1585 printk(KERN_ERR "schedule_timeout: wrong timeout "
1586 "value %lx\n", timeout);
1588 current->state = TASK_RUNNING;
1593 expire = timeout + jiffies;
1595 setup_timer_on_stack(&timer, process_timeout, (unsigned long)current);
1596 __mod_timer(&timer, expire, false);
1598 del_singleshot_timer_sync(&timer);
1600 /* Remove the timer from the object tracker */
1601 destroy_timer_on_stack(&timer);
1603 timeout = expire - jiffies;
1606 return timeout < 0 ? 0 : timeout;
1608 EXPORT_SYMBOL(schedule_timeout);
1611 * We can use __set_current_state() here because schedule_timeout() calls
1612 * schedule() unconditionally.
1614 signed long __sched schedule_timeout_interruptible(signed long timeout)
1616 __set_current_state(TASK_INTERRUPTIBLE);
1617 return schedule_timeout(timeout);
1619 EXPORT_SYMBOL(schedule_timeout_interruptible);
1621 signed long __sched schedule_timeout_killable(signed long timeout)
1623 __set_current_state(TASK_KILLABLE);
1624 return schedule_timeout(timeout);
1626 EXPORT_SYMBOL(schedule_timeout_killable);
1628 signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1630 __set_current_state(TASK_UNINTERRUPTIBLE);
1631 return schedule_timeout(timeout);
1633 EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1636 * Like schedule_timeout_uninterruptible(), except this task will not contribute
1639 signed long __sched schedule_timeout_idle(signed long timeout)
1641 __set_current_state(TASK_IDLE);
1642 return schedule_timeout(timeout);
1644 EXPORT_SYMBOL(schedule_timeout_idle);
1646 #ifdef CONFIG_HOTPLUG_CPU
1647 static void migrate_timer_list(struct timer_base *new_base, struct hlist_head *head)
1649 struct timer_list *timer;
1650 int cpu = new_base->cpu;
1652 while (!hlist_empty(head)) {
1653 timer = hlist_entry(head->first, struct timer_list, entry);
1654 detach_timer(timer, false);
1655 timer->flags = (timer->flags & ~TIMER_BASEMASK) | cpu;
1656 internal_add_timer(new_base, timer);
1660 static void migrate_timers(int cpu)
1662 struct timer_base *old_base;
1663 struct timer_base *new_base;
1666 BUG_ON(cpu_online(cpu));
1668 for (b = 0; b < NR_BASES; b++) {
1669 old_base = per_cpu_ptr(&timer_bases[b], cpu);
1670 new_base = get_cpu_ptr(&timer_bases[b]);
1672 * The caller is globally serialized and nobody else
1673 * takes two locks at once, deadlock is not possible.
1675 spin_lock_irq(&new_base->lock);
1676 spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1678 BUG_ON(old_base->running_timer);
1680 for (i = 0; i < WHEEL_SIZE; i++)
1681 migrate_timer_list(new_base, old_base->vectors + i);
1683 spin_unlock(&old_base->lock);
1684 spin_unlock_irq(&new_base->lock);
1685 put_cpu_ptr(&timer_bases);
1689 static int timer_cpu_notify(struct notifier_block *self,
1690 unsigned long action, void *hcpu)
1694 case CPU_DEAD_FROZEN:
1695 migrate_timers((long)hcpu);
1704 static inline void timer_register_cpu_notifier(void)
1706 cpu_notifier(timer_cpu_notify, 0);
1709 static inline void timer_register_cpu_notifier(void) { }
1710 #endif /* CONFIG_HOTPLUG_CPU */
1712 static void __init init_timer_cpu(int cpu)
1714 struct timer_base *base;
1717 for (i = 0; i < NR_BASES; i++) {
1718 base = per_cpu_ptr(&timer_bases[i], cpu);
1720 spin_lock_init(&base->lock);
1721 base->clk = jiffies;
1725 static void __init init_timer_cpus(void)
1729 for_each_possible_cpu(cpu)
1730 init_timer_cpu(cpu);
1733 void __init init_timers(void)
1737 timer_register_cpu_notifier();
1738 open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
1742 * msleep - sleep safely even with waitqueue interruptions
1743 * @msecs: Time in milliseconds to sleep for
1745 void msleep(unsigned int msecs)
1747 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1750 timeout = schedule_timeout_uninterruptible(timeout);
1753 EXPORT_SYMBOL(msleep);
1756 * msleep_interruptible - sleep waiting for signals
1757 * @msecs: Time in milliseconds to sleep for
1759 unsigned long msleep_interruptible(unsigned int msecs)
1761 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1763 while (timeout && !signal_pending(current))
1764 timeout = schedule_timeout_interruptible(timeout);
1765 return jiffies_to_msecs(timeout);
1768 EXPORT_SYMBOL(msleep_interruptible);
1770 static void __sched do_usleep_range(unsigned long min, unsigned long max)
1775 kmin = ktime_set(0, min * NSEC_PER_USEC);
1776 delta = (u64)(max - min) * NSEC_PER_USEC;
1777 schedule_hrtimeout_range(&kmin, delta, HRTIMER_MODE_REL);
1781 * usleep_range - Drop in replacement for udelay where wakeup is flexible
1782 * @min: Minimum time in usecs to sleep
1783 * @max: Maximum time in usecs to sleep
1785 void __sched usleep_range(unsigned long min, unsigned long max)
1787 __set_current_state(TASK_UNINTERRUPTIBLE);
1788 do_usleep_range(min, max);
1790 EXPORT_SYMBOL(usleep_range);