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 bool must_forward_clk;
205 DECLARE_BITMAP(pending_map, WHEEL_SIZE);
206 struct hlist_head vectors[WHEEL_SIZE];
207 } ____cacheline_aligned;
209 static DEFINE_PER_CPU(struct timer_base, timer_bases[NR_BASES]);
211 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
212 unsigned int sysctl_timer_migration = 1;
214 void timers_update_migration(bool update_nohz)
216 bool on = sysctl_timer_migration && tick_nohz_active;
219 /* Avoid the loop, if nothing to update */
220 if (this_cpu_read(timer_bases[BASE_STD].migration_enabled) == on)
223 for_each_possible_cpu(cpu) {
224 per_cpu(timer_bases[BASE_STD].migration_enabled, cpu) = on;
225 per_cpu(timer_bases[BASE_DEF].migration_enabled, cpu) = on;
226 per_cpu(hrtimer_bases.migration_enabled, cpu) = on;
229 per_cpu(timer_bases[BASE_STD].nohz_active, cpu) = true;
230 per_cpu(timer_bases[BASE_DEF].nohz_active, cpu) = true;
231 per_cpu(hrtimer_bases.nohz_active, cpu) = true;
235 int timer_migration_handler(struct ctl_table *table, int write,
236 void __user *buffer, size_t *lenp,
239 static DEFINE_MUTEX(mutex);
243 ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
245 timers_update_migration(false);
246 mutex_unlock(&mutex);
251 static unsigned long round_jiffies_common(unsigned long j, int cpu,
255 unsigned long original = j;
258 * We don't want all cpus firing their timers at once hitting the
259 * same lock or cachelines, so we skew each extra cpu with an extra
260 * 3 jiffies. This 3 jiffies came originally from the mm/ code which
262 * The skew is done by adding 3*cpunr, then round, then subtract this
263 * extra offset again.
270 * If the target jiffie is just after a whole second (which can happen
271 * due to delays of the timer irq, long irq off times etc etc) then
272 * we should round down to the whole second, not up. Use 1/4th second
273 * as cutoff for this rounding as an extreme upper bound for this.
274 * But never round down if @force_up is set.
276 if (rem < HZ/4 && !force_up) /* round down */
281 /* now that we have rounded, subtract the extra skew again */
285 * Make sure j is still in the future. Otherwise return the
288 return time_is_after_jiffies(j) ? j : original;
292 * __round_jiffies - function to round jiffies to a full second
293 * @j: the time in (absolute) jiffies that should be rounded
294 * @cpu: the processor number on which the timeout will happen
296 * __round_jiffies() rounds an absolute time in the future (in jiffies)
297 * up or down to (approximately) full seconds. This is useful for timers
298 * for which the exact time they fire does not matter too much, as long as
299 * they fire approximately every X seconds.
301 * By rounding these timers to whole seconds, all such timers will fire
302 * at the same time, rather than at various times spread out. The goal
303 * of this is to have the CPU wake up less, which saves power.
305 * The exact rounding is skewed for each processor to avoid all
306 * processors firing at the exact same time, which could lead
307 * to lock contention or spurious cache line bouncing.
309 * The return value is the rounded version of the @j parameter.
311 unsigned long __round_jiffies(unsigned long j, int cpu)
313 return round_jiffies_common(j, cpu, false);
315 EXPORT_SYMBOL_GPL(__round_jiffies);
318 * __round_jiffies_relative - function to round jiffies to a full second
319 * @j: the time in (relative) jiffies that should be rounded
320 * @cpu: the processor number on which the timeout will happen
322 * __round_jiffies_relative() rounds a time delta in the future (in jiffies)
323 * up or down to (approximately) full seconds. This is useful for timers
324 * for which the exact time they fire does not matter too much, as long as
325 * they fire approximately every X seconds.
327 * By rounding these timers to whole seconds, all such timers will fire
328 * at the same time, rather than at various times spread out. The goal
329 * of this is to have the CPU wake up less, which saves power.
331 * The exact rounding is skewed for each processor to avoid all
332 * processors firing at the exact same time, which could lead
333 * to lock contention or spurious cache line bouncing.
335 * The return value is the rounded version of the @j parameter.
337 unsigned long __round_jiffies_relative(unsigned long j, int cpu)
339 unsigned long j0 = jiffies;
341 /* Use j0 because jiffies might change while we run */
342 return round_jiffies_common(j + j0, cpu, false) - j0;
344 EXPORT_SYMBOL_GPL(__round_jiffies_relative);
347 * round_jiffies - function to round jiffies to a full second
348 * @j: the time in (absolute) jiffies that should be rounded
350 * round_jiffies() rounds an absolute time in the future (in jiffies)
351 * up or down to (approximately) full seconds. This is useful for timers
352 * for which the exact time they fire does not matter too much, as long as
353 * they fire approximately every X seconds.
355 * By rounding these timers to whole seconds, all such timers will fire
356 * at the same time, rather than at various times spread out. The goal
357 * of this is to have the CPU wake up less, which saves power.
359 * The return value is the rounded version of the @j parameter.
361 unsigned long round_jiffies(unsigned long j)
363 return round_jiffies_common(j, raw_smp_processor_id(), false);
365 EXPORT_SYMBOL_GPL(round_jiffies);
368 * round_jiffies_relative - function to round jiffies to a full second
369 * @j: the time in (relative) jiffies that should be rounded
371 * round_jiffies_relative() rounds a time delta in the future (in jiffies)
372 * up or down to (approximately) full seconds. This is useful for timers
373 * for which the exact time they fire does not matter too much, as long as
374 * they fire approximately every X seconds.
376 * By rounding these timers to whole seconds, all such timers will fire
377 * at the same time, rather than at various times spread out. The goal
378 * of this is to have the CPU wake up less, which saves power.
380 * The return value is the rounded version of the @j parameter.
382 unsigned long round_jiffies_relative(unsigned long j)
384 return __round_jiffies_relative(j, raw_smp_processor_id());
386 EXPORT_SYMBOL_GPL(round_jiffies_relative);
389 * __round_jiffies_up - function to round jiffies up to a full second
390 * @j: the time in (absolute) jiffies that should be rounded
391 * @cpu: the processor number on which the timeout will happen
393 * This is the same as __round_jiffies() except that it will never
394 * round down. This is useful for timeouts for which the exact time
395 * of firing does not matter too much, as long as they don't fire too
398 unsigned long __round_jiffies_up(unsigned long j, int cpu)
400 return round_jiffies_common(j, cpu, true);
402 EXPORT_SYMBOL_GPL(__round_jiffies_up);
405 * __round_jiffies_up_relative - function to round jiffies up to a full second
406 * @j: the time in (relative) jiffies that should be rounded
407 * @cpu: the processor number on which the timeout will happen
409 * This is the same as __round_jiffies_relative() except that it will never
410 * round down. This is useful for timeouts for which the exact time
411 * of firing does not matter too much, as long as they don't fire too
414 unsigned long __round_jiffies_up_relative(unsigned long j, int cpu)
416 unsigned long j0 = jiffies;
418 /* Use j0 because jiffies might change while we run */
419 return round_jiffies_common(j + j0, cpu, true) - j0;
421 EXPORT_SYMBOL_GPL(__round_jiffies_up_relative);
424 * round_jiffies_up - function to round jiffies up to a full second
425 * @j: the time in (absolute) jiffies that should be rounded
427 * This is the same as round_jiffies() except that it will never
428 * round down. This is useful for timeouts for which the exact time
429 * of firing does not matter too much, as long as they don't fire too
432 unsigned long round_jiffies_up(unsigned long j)
434 return round_jiffies_common(j, raw_smp_processor_id(), true);
436 EXPORT_SYMBOL_GPL(round_jiffies_up);
439 * round_jiffies_up_relative - function to round jiffies up to a full second
440 * @j: the time in (relative) jiffies that should be rounded
442 * This is the same as round_jiffies_relative() except that it will never
443 * round down. This is useful for timeouts for which the exact time
444 * of firing does not matter too much, as long as they don't fire too
447 unsigned long round_jiffies_up_relative(unsigned long j)
449 return __round_jiffies_up_relative(j, raw_smp_processor_id());
451 EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
454 static inline unsigned int timer_get_idx(struct timer_list *timer)
456 return (timer->flags & TIMER_ARRAYMASK) >> TIMER_ARRAYSHIFT;
459 static inline void timer_set_idx(struct timer_list *timer, unsigned int idx)
461 timer->flags = (timer->flags & ~TIMER_ARRAYMASK) |
462 idx << TIMER_ARRAYSHIFT;
466 * Helper function to calculate the array index for a given expiry
469 static inline unsigned calc_index(unsigned expires, unsigned lvl)
471 expires = (expires + LVL_GRAN(lvl)) >> LVL_SHIFT(lvl);
472 return LVL_OFFS(lvl) + (expires & LVL_MASK);
475 static int calc_wheel_index(unsigned long expires, unsigned long clk)
477 unsigned long delta = expires - clk;
480 if (delta < LVL_START(1)) {
481 idx = calc_index(expires, 0);
482 } else if (delta < LVL_START(2)) {
483 idx = calc_index(expires, 1);
484 } else if (delta < LVL_START(3)) {
485 idx = calc_index(expires, 2);
486 } else if (delta < LVL_START(4)) {
487 idx = calc_index(expires, 3);
488 } else if (delta < LVL_START(5)) {
489 idx = calc_index(expires, 4);
490 } else if (delta < LVL_START(6)) {
491 idx = calc_index(expires, 5);
492 } else if (delta < LVL_START(7)) {
493 idx = calc_index(expires, 6);
494 } else if (LVL_DEPTH > 8 && delta < LVL_START(8)) {
495 idx = calc_index(expires, 7);
496 } else if ((long) delta < 0) {
497 idx = clk & LVL_MASK;
500 * Force expire obscene large timeouts to expire at the
501 * capacity limit of the wheel.
503 if (expires >= WHEEL_TIMEOUT_CUTOFF)
504 expires = WHEEL_TIMEOUT_MAX;
506 idx = calc_index(expires, LVL_DEPTH - 1);
512 * Enqueue the timer into the hash bucket, mark it pending in
513 * the bitmap and store the index in the timer flags.
515 static void enqueue_timer(struct timer_base *base, struct timer_list *timer,
518 hlist_add_head(&timer->entry, base->vectors + idx);
519 __set_bit(idx, base->pending_map);
520 timer_set_idx(timer, idx);
524 __internal_add_timer(struct timer_base *base, struct timer_list *timer)
528 idx = calc_wheel_index(timer->expires, base->clk);
529 enqueue_timer(base, timer, idx);
533 trigger_dyntick_cpu(struct timer_base *base, struct timer_list *timer)
535 if (!IS_ENABLED(CONFIG_NO_HZ_COMMON) || !base->nohz_active)
539 * TODO: This wants some optimizing similar to the code below, but we
540 * will do that when we switch from push to pull for deferrable timers.
542 if (timer->flags & TIMER_DEFERRABLE) {
543 if (tick_nohz_full_cpu(base->cpu))
544 wake_up_nohz_cpu(base->cpu);
549 * We might have to IPI the remote CPU if the base is idle and the
550 * timer is not deferrable. If the other CPU is on the way to idle
551 * then it can't set base->is_idle as we hold the base lock:
556 /* Check whether this is the new first expiring timer: */
557 if (time_after_eq(timer->expires, base->next_expiry))
561 * Set the next expiry time and kick the CPU so it can reevaluate the
564 base->next_expiry = timer->expires;
565 wake_up_nohz_cpu(base->cpu);
569 internal_add_timer(struct timer_base *base, struct timer_list *timer)
571 __internal_add_timer(base, timer);
572 trigger_dyntick_cpu(base, timer);
575 #ifdef CONFIG_TIMER_STATS
576 void __timer_stats_timer_set_start_info(struct timer_list *timer, void *addr)
578 if (timer->start_site)
581 timer->start_site = addr;
582 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
583 timer->start_pid = current->pid;
586 static void timer_stats_account_timer(struct timer_list *timer)
591 * start_site can be concurrently reset by
592 * timer_stats_timer_clear_start_info()
594 site = READ_ONCE(timer->start_site);
598 timer_stats_update_stats(timer, timer->start_pid, site,
599 timer->function, timer->start_comm,
604 static void timer_stats_account_timer(struct timer_list *timer) {}
607 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
609 static struct debug_obj_descr timer_debug_descr;
611 static void *timer_debug_hint(void *addr)
613 return ((struct timer_list *) addr)->function;
616 static bool timer_is_static_object(void *addr)
618 struct timer_list *timer = addr;
620 return (timer->entry.pprev == NULL &&
621 timer->entry.next == TIMER_ENTRY_STATIC);
625 * fixup_init is called when:
626 * - an active object is initialized
628 static bool timer_fixup_init(void *addr, enum debug_obj_state state)
630 struct timer_list *timer = addr;
633 case ODEBUG_STATE_ACTIVE:
634 del_timer_sync(timer);
635 debug_object_init(timer, &timer_debug_descr);
642 /* Stub timer callback for improperly used timers. */
643 static void stub_timer(unsigned long data)
649 * fixup_activate is called when:
650 * - an active object is activated
651 * - an unknown non-static object is activated
653 static bool timer_fixup_activate(void *addr, enum debug_obj_state state)
655 struct timer_list *timer = addr;
658 case ODEBUG_STATE_NOTAVAILABLE:
659 setup_timer(timer, stub_timer, 0);
662 case ODEBUG_STATE_ACTIVE:
671 * fixup_free is called when:
672 * - an active object is freed
674 static bool timer_fixup_free(void *addr, enum debug_obj_state state)
676 struct timer_list *timer = addr;
679 case ODEBUG_STATE_ACTIVE:
680 del_timer_sync(timer);
681 debug_object_free(timer, &timer_debug_descr);
689 * fixup_assert_init is called when:
690 * - an untracked/uninit-ed object is found
692 static bool timer_fixup_assert_init(void *addr, enum debug_obj_state state)
694 struct timer_list *timer = addr;
697 case ODEBUG_STATE_NOTAVAILABLE:
698 setup_timer(timer, stub_timer, 0);
705 static struct debug_obj_descr timer_debug_descr = {
706 .name = "timer_list",
707 .debug_hint = timer_debug_hint,
708 .is_static_object = timer_is_static_object,
709 .fixup_init = timer_fixup_init,
710 .fixup_activate = timer_fixup_activate,
711 .fixup_free = timer_fixup_free,
712 .fixup_assert_init = timer_fixup_assert_init,
715 static inline void debug_timer_init(struct timer_list *timer)
717 debug_object_init(timer, &timer_debug_descr);
720 static inline void debug_timer_activate(struct timer_list *timer)
722 debug_object_activate(timer, &timer_debug_descr);
725 static inline void debug_timer_deactivate(struct timer_list *timer)
727 debug_object_deactivate(timer, &timer_debug_descr);
730 static inline void debug_timer_free(struct timer_list *timer)
732 debug_object_free(timer, &timer_debug_descr);
735 static inline void debug_timer_assert_init(struct timer_list *timer)
737 debug_object_assert_init(timer, &timer_debug_descr);
740 static void do_init_timer(struct timer_list *timer, unsigned int flags,
741 const char *name, struct lock_class_key *key);
743 void init_timer_on_stack_key(struct timer_list *timer, unsigned int flags,
744 const char *name, struct lock_class_key *key)
746 debug_object_init_on_stack(timer, &timer_debug_descr);
747 do_init_timer(timer, flags, name, key);
749 EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
751 void destroy_timer_on_stack(struct timer_list *timer)
753 debug_object_free(timer, &timer_debug_descr);
755 EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
758 static inline void debug_timer_init(struct timer_list *timer) { }
759 static inline void debug_timer_activate(struct timer_list *timer) { }
760 static inline void debug_timer_deactivate(struct timer_list *timer) { }
761 static inline void debug_timer_assert_init(struct timer_list *timer) { }
764 static inline void debug_init(struct timer_list *timer)
766 debug_timer_init(timer);
767 trace_timer_init(timer);
771 debug_activate(struct timer_list *timer, unsigned long expires)
773 debug_timer_activate(timer);
774 trace_timer_start(timer, expires, timer->flags);
777 static inline void debug_deactivate(struct timer_list *timer)
779 debug_timer_deactivate(timer);
780 trace_timer_cancel(timer);
783 static inline void debug_assert_init(struct timer_list *timer)
785 debug_timer_assert_init(timer);
788 static void do_init_timer(struct timer_list *timer, unsigned int flags,
789 const char *name, struct lock_class_key *key)
791 timer->entry.pprev = NULL;
792 timer->flags = flags | raw_smp_processor_id();
793 #ifdef CONFIG_TIMER_STATS
794 timer->start_site = NULL;
795 timer->start_pid = -1;
796 memset(timer->start_comm, 0, TASK_COMM_LEN);
798 lockdep_init_map(&timer->lockdep_map, name, key, 0);
802 * init_timer_key - initialize a timer
803 * @timer: the timer to be initialized
804 * @flags: timer flags
805 * @name: name of the timer
806 * @key: lockdep class key of the fake lock used for tracking timer
807 * sync lock dependencies
809 * init_timer_key() must be done to a timer prior calling *any* of the
810 * other timer functions.
812 void init_timer_key(struct timer_list *timer, unsigned int flags,
813 const char *name, struct lock_class_key *key)
816 do_init_timer(timer, flags, name, key);
818 EXPORT_SYMBOL(init_timer_key);
820 static inline void detach_timer(struct timer_list *timer, bool clear_pending)
822 struct hlist_node *entry = &timer->entry;
824 debug_deactivate(timer);
829 entry->next = LIST_POISON2;
832 static int detach_if_pending(struct timer_list *timer, struct timer_base *base,
835 unsigned idx = timer_get_idx(timer);
837 if (!timer_pending(timer))
840 if (hlist_is_singular_node(&timer->entry, base->vectors + idx))
841 __clear_bit(idx, base->pending_map);
843 detach_timer(timer, clear_pending);
847 static inline struct timer_base *get_timer_cpu_base(u32 tflags, u32 cpu)
849 struct timer_base *base = per_cpu_ptr(&timer_bases[BASE_STD], cpu);
852 * If the timer is deferrable and NO_HZ_COMMON is set then we need
853 * to use the deferrable base.
855 if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE))
856 base = per_cpu_ptr(&timer_bases[BASE_DEF], cpu);
860 static inline struct timer_base *get_timer_this_cpu_base(u32 tflags)
862 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
865 * If the timer is deferrable and NO_HZ_COMMON is set then we need
866 * to use the deferrable base.
868 if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE))
869 base = this_cpu_ptr(&timer_bases[BASE_DEF]);
873 static inline struct timer_base *get_timer_base(u32 tflags)
875 return get_timer_cpu_base(tflags, tflags & TIMER_CPUMASK);
878 #ifdef CONFIG_NO_HZ_COMMON
879 static inline struct timer_base *
880 get_target_base(struct timer_base *base, unsigned tflags)
883 if ((tflags & TIMER_PINNED) || !base->migration_enabled)
884 return get_timer_this_cpu_base(tflags);
885 return get_timer_cpu_base(tflags, get_nohz_timer_target());
887 return get_timer_this_cpu_base(tflags);
891 static inline void forward_timer_base(struct timer_base *base)
896 * We only forward the base when we are idle or have just come out of
897 * idle (must_forward_clk logic), and have a delta between base clock
898 * and jiffies. In the common case, run_timers will take care of it.
900 if (likely(!base->must_forward_clk))
903 jnow = READ_ONCE(jiffies);
904 base->must_forward_clk = base->is_idle;
905 if ((long)(jnow - base->clk) < 2)
909 * If the next expiry value is > jiffies, then we fast forward to
910 * jiffies otherwise we forward to the next expiry value.
912 if (time_after(base->next_expiry, jnow))
915 base->clk = base->next_expiry;
918 static inline struct timer_base *
919 get_target_base(struct timer_base *base, unsigned tflags)
921 return get_timer_this_cpu_base(tflags);
924 static inline void forward_timer_base(struct timer_base *base) { }
929 * We are using hashed locking: Holding per_cpu(timer_bases[x]).lock means
930 * that all timers which are tied to this base are locked, and the base itself
933 * So __run_timers/migrate_timers can safely modify all timers which could
934 * be found in the base->vectors array.
936 * When a timer is migrating then the TIMER_MIGRATING flag is set and we need
937 * to wait until the migration is done.
939 static struct timer_base *lock_timer_base(struct timer_list *timer,
940 unsigned long *flags)
941 __acquires(timer->base->lock)
944 struct timer_base *base;
948 * We need to use READ_ONCE() here, otherwise the compiler
949 * might re-read @tf between the check for TIMER_MIGRATING
952 tf = READ_ONCE(timer->flags);
954 if (!(tf & TIMER_MIGRATING)) {
955 base = get_timer_base(tf);
956 spin_lock_irqsave(&base->lock, *flags);
957 if (timer->flags == tf)
959 spin_unlock_irqrestore(&base->lock, *flags);
966 __mod_timer(struct timer_list *timer, unsigned long expires, bool pending_only)
968 struct timer_base *base, *new_base;
969 unsigned int idx = UINT_MAX;
970 unsigned long clk = 0, flags;
973 BUG_ON(!timer->function);
976 * This is a common optimization triggered by the networking code - if
977 * the timer is re-modified to have the same timeout or ends up in the
978 * same array bucket then just return:
980 if (timer_pending(timer)) {
982 * The downside of this optimization is that it can result in
983 * larger granularity than you would get from adding a new
984 * timer with this expiry.
986 if (timer->expires == expires)
990 * We lock timer base and calculate the bucket index right
991 * here. If the timer ends up in the same bucket, then we
992 * just update the expiry time and avoid the whole
993 * dequeue/enqueue dance.
995 base = lock_timer_base(timer, &flags);
996 forward_timer_base(base);
999 idx = calc_wheel_index(expires, clk);
1002 * Retrieve and compare the array index of the pending
1003 * timer. If it matches set the expiry to the new value so a
1004 * subsequent call will exit in the expires check above.
1006 if (idx == timer_get_idx(timer)) {
1007 timer->expires = expires;
1012 base = lock_timer_base(timer, &flags);
1013 forward_timer_base(base);
1016 timer_stats_timer_set_start_info(timer);
1018 ret = detach_if_pending(timer, base, false);
1019 if (!ret && pending_only)
1022 new_base = get_target_base(base, timer->flags);
1024 if (base != new_base) {
1026 * We are trying to schedule the timer on the new base.
1027 * However we can't change timer's base while it is running,
1028 * otherwise del_timer_sync() can't detect that the timer's
1029 * handler yet has not finished. This also guarantees that the
1030 * timer is serialized wrt itself.
1032 if (likely(base->running_timer != timer)) {
1033 /* See the comment in lock_timer_base() */
1034 timer->flags |= TIMER_MIGRATING;
1036 spin_unlock(&base->lock);
1038 spin_lock(&base->lock);
1039 WRITE_ONCE(timer->flags,
1040 (timer->flags & ~TIMER_BASEMASK) | base->cpu);
1041 forward_timer_base(base);
1045 debug_activate(timer, expires);
1047 timer->expires = expires;
1049 * If 'idx' was calculated above and the base time did not advance
1050 * between calculating 'idx' and possibly switching the base, only
1051 * enqueue_timer() and trigger_dyntick_cpu() is required. Otherwise
1052 * we need to (re)calculate the wheel index via
1053 * internal_add_timer().
1055 if (idx != UINT_MAX && clk == base->clk) {
1056 enqueue_timer(base, timer, idx);
1057 trigger_dyntick_cpu(base, timer);
1059 internal_add_timer(base, timer);
1063 spin_unlock_irqrestore(&base->lock, flags);
1069 * mod_timer_pending - modify a pending timer's timeout
1070 * @timer: the pending timer to be modified
1071 * @expires: new timeout in jiffies
1073 * mod_timer_pending() is the same for pending timers as mod_timer(),
1074 * but will not re-activate and modify already deleted timers.
1076 * It is useful for unserialized use of timers.
1078 int mod_timer_pending(struct timer_list *timer, unsigned long expires)
1080 return __mod_timer(timer, expires, true);
1082 EXPORT_SYMBOL(mod_timer_pending);
1085 * mod_timer - modify a timer's timeout
1086 * @timer: the timer to be modified
1087 * @expires: new timeout in jiffies
1089 * mod_timer() is a more efficient way to update the expire field of an
1090 * active timer (if the timer is inactive it will be activated)
1092 * mod_timer(timer, expires) is equivalent to:
1094 * del_timer(timer); timer->expires = expires; add_timer(timer);
1096 * Note that if there are multiple unserialized concurrent users of the
1097 * same timer, then mod_timer() is the only safe way to modify the timeout,
1098 * since add_timer() cannot modify an already running timer.
1100 * The function returns whether it has modified a pending timer or not.
1101 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
1102 * active timer returns 1.)
1104 int mod_timer(struct timer_list *timer, unsigned long expires)
1106 return __mod_timer(timer, expires, false);
1108 EXPORT_SYMBOL(mod_timer);
1111 * add_timer - start a timer
1112 * @timer: the timer to be added
1114 * The kernel will do a ->function(->data) callback from the
1115 * timer interrupt at the ->expires point in the future. The
1116 * current time is 'jiffies'.
1118 * The timer's ->expires, ->function (and if the handler uses it, ->data)
1119 * fields must be set prior calling this function.
1121 * Timers with an ->expires field in the past will be executed in the next
1124 void add_timer(struct timer_list *timer)
1126 BUG_ON(timer_pending(timer));
1127 mod_timer(timer, timer->expires);
1129 EXPORT_SYMBOL(add_timer);
1132 * add_timer_on - start a timer on a particular CPU
1133 * @timer: the timer to be added
1134 * @cpu: the CPU to start it on
1136 * This is not very scalable on SMP. Double adds are not possible.
1138 void add_timer_on(struct timer_list *timer, int cpu)
1140 struct timer_base *new_base, *base;
1141 unsigned long flags;
1143 timer_stats_timer_set_start_info(timer);
1144 BUG_ON(timer_pending(timer) || !timer->function);
1146 new_base = get_timer_cpu_base(timer->flags, cpu);
1149 * If @timer was on a different CPU, it should be migrated with the
1150 * old base locked to prevent other operations proceeding with the
1151 * wrong base locked. See lock_timer_base().
1153 base = lock_timer_base(timer, &flags);
1154 if (base != new_base) {
1155 timer->flags |= TIMER_MIGRATING;
1157 spin_unlock(&base->lock);
1159 spin_lock(&base->lock);
1160 WRITE_ONCE(timer->flags,
1161 (timer->flags & ~TIMER_BASEMASK) | cpu);
1163 forward_timer_base(base);
1165 debug_activate(timer, timer->expires);
1166 internal_add_timer(base, timer);
1167 spin_unlock_irqrestore(&base->lock, flags);
1169 EXPORT_SYMBOL_GPL(add_timer_on);
1172 * del_timer - deactive a timer.
1173 * @timer: the timer to be deactivated
1175 * del_timer() deactivates a timer - this works on both active and inactive
1178 * The function returns whether it has deactivated a pending timer or not.
1179 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
1180 * active timer returns 1.)
1182 int del_timer(struct timer_list *timer)
1184 struct timer_base *base;
1185 unsigned long flags;
1188 debug_assert_init(timer);
1190 timer_stats_timer_clear_start_info(timer);
1191 if (timer_pending(timer)) {
1192 base = lock_timer_base(timer, &flags);
1193 ret = detach_if_pending(timer, base, true);
1194 spin_unlock_irqrestore(&base->lock, flags);
1199 EXPORT_SYMBOL(del_timer);
1202 * try_to_del_timer_sync - Try to deactivate a timer
1203 * @timer: timer do del
1205 * This function tries to deactivate a timer. Upon successful (ret >= 0)
1206 * exit the timer is not queued and the handler is not running on any CPU.
1208 int try_to_del_timer_sync(struct timer_list *timer)
1210 struct timer_base *base;
1211 unsigned long flags;
1214 debug_assert_init(timer);
1216 base = lock_timer_base(timer, &flags);
1218 if (base->running_timer != timer) {
1219 timer_stats_timer_clear_start_info(timer);
1220 ret = detach_if_pending(timer, base, true);
1222 spin_unlock_irqrestore(&base->lock, flags);
1226 EXPORT_SYMBOL(try_to_del_timer_sync);
1230 * del_timer_sync - deactivate a timer and wait for the handler to finish.
1231 * @timer: the timer to be deactivated
1233 * This function only differs from del_timer() on SMP: besides deactivating
1234 * the timer it also makes sure the handler has finished executing on other
1237 * Synchronization rules: Callers must prevent restarting of the timer,
1238 * otherwise this function is meaningless. It must not be called from
1239 * interrupt contexts unless the timer is an irqsafe one. The caller must
1240 * not hold locks which would prevent completion of the timer's
1241 * handler. The timer's handler must not call add_timer_on(). Upon exit the
1242 * timer is not queued and the handler is not running on any CPU.
1244 * Note: For !irqsafe timers, you must not hold locks that are held in
1245 * interrupt context while calling this function. Even if the lock has
1246 * nothing to do with the timer in question. Here's why:
1252 * base->running_timer = mytimer;
1253 * spin_lock_irq(somelock);
1255 * spin_lock(somelock);
1256 * del_timer_sync(mytimer);
1257 * while (base->running_timer == mytimer);
1259 * Now del_timer_sync() will never return and never release somelock.
1260 * The interrupt on the other CPU is waiting to grab somelock but
1261 * it has interrupted the softirq that CPU0 is waiting to finish.
1263 * The function returns whether it has deactivated a pending timer or not.
1265 int del_timer_sync(struct timer_list *timer)
1267 #ifdef CONFIG_LOCKDEP
1268 unsigned long flags;
1271 * If lockdep gives a backtrace here, please reference
1272 * the synchronization rules above.
1274 local_irq_save(flags);
1275 lock_map_acquire(&timer->lockdep_map);
1276 lock_map_release(&timer->lockdep_map);
1277 local_irq_restore(flags);
1280 * don't use it in hardirq context, because it
1281 * could lead to deadlock.
1283 WARN_ON(in_irq() && !(timer->flags & TIMER_IRQSAFE));
1285 int ret = try_to_del_timer_sync(timer);
1291 EXPORT_SYMBOL(del_timer_sync);
1294 static void call_timer_fn(struct timer_list *timer, void (*fn)(unsigned long),
1297 int count = preempt_count();
1299 #ifdef CONFIG_LOCKDEP
1301 * It is permissible to free the timer from inside the
1302 * function that is called from it, this we need to take into
1303 * account for lockdep too. To avoid bogus "held lock freed"
1304 * warnings as well as problems when looking into
1305 * timer->lockdep_map, make a copy and use that here.
1307 struct lockdep_map lockdep_map;
1309 lockdep_copy_map(&lockdep_map, &timer->lockdep_map);
1312 * Couple the lock chain with the lock chain at
1313 * del_timer_sync() by acquiring the lock_map around the fn()
1314 * call here and in del_timer_sync().
1316 lock_map_acquire(&lockdep_map);
1318 trace_timer_expire_entry(timer);
1320 trace_timer_expire_exit(timer);
1322 lock_map_release(&lockdep_map);
1324 if (count != preempt_count()) {
1325 WARN_ONCE(1, "timer: %pF preempt leak: %08x -> %08x\n",
1326 fn, count, preempt_count());
1328 * Restore the preempt count. That gives us a decent
1329 * chance to survive and extract information. If the
1330 * callback kept a lock held, bad luck, but not worse
1331 * than the BUG() we had.
1333 preempt_count_set(count);
1337 static void expire_timers(struct timer_base *base, struct hlist_head *head)
1339 while (!hlist_empty(head)) {
1340 struct timer_list *timer;
1341 void (*fn)(unsigned long);
1344 timer = hlist_entry(head->first, struct timer_list, entry);
1345 timer_stats_account_timer(timer);
1347 base->running_timer = timer;
1348 detach_timer(timer, true);
1350 fn = timer->function;
1353 if (timer->flags & TIMER_IRQSAFE) {
1354 spin_unlock(&base->lock);
1355 call_timer_fn(timer, fn, data);
1356 spin_lock(&base->lock);
1358 spin_unlock_irq(&base->lock);
1359 call_timer_fn(timer, fn, data);
1360 spin_lock_irq(&base->lock);
1365 static int __collect_expired_timers(struct timer_base *base,
1366 struct hlist_head *heads)
1368 unsigned long clk = base->clk;
1369 struct hlist_head *vec;
1373 for (i = 0; i < LVL_DEPTH; i++) {
1374 idx = (clk & LVL_MASK) + i * LVL_SIZE;
1376 if (__test_and_clear_bit(idx, base->pending_map)) {
1377 vec = base->vectors + idx;
1378 hlist_move_list(vec, heads++);
1381 /* Is it time to look at the next level? */
1382 if (clk & LVL_CLK_MASK)
1384 /* Shift clock for the next level granularity */
1385 clk >>= LVL_CLK_SHIFT;
1390 #ifdef CONFIG_NO_HZ_COMMON
1392 * Find the next pending bucket of a level. Search from level start (@offset)
1393 * + @clk upwards and if nothing there, search from start of the level
1394 * (@offset) up to @offset + clk.
1396 static int next_pending_bucket(struct timer_base *base, unsigned offset,
1399 unsigned pos, start = offset + clk;
1400 unsigned end = offset + LVL_SIZE;
1402 pos = find_next_bit(base->pending_map, end, start);
1406 pos = find_next_bit(base->pending_map, start, offset);
1407 return pos < start ? pos + LVL_SIZE - start : -1;
1411 * Search the first expiring timer in the various clock levels. Caller must
1414 static unsigned long __next_timer_interrupt(struct timer_base *base)
1416 unsigned long clk, next, adj;
1417 unsigned lvl, offset = 0;
1419 next = base->clk + NEXT_TIMER_MAX_DELTA;
1421 for (lvl = 0; lvl < LVL_DEPTH; lvl++, offset += LVL_SIZE) {
1422 int pos = next_pending_bucket(base, offset, clk & LVL_MASK);
1425 unsigned long tmp = clk + (unsigned long) pos;
1427 tmp <<= LVL_SHIFT(lvl);
1428 if (time_before(tmp, next))
1432 * Clock for the next level. If the current level clock lower
1433 * bits are zero, we look at the next level as is. If not we
1434 * need to advance it by one because that's going to be the
1435 * next expiring bucket in that level. base->clk is the next
1436 * expiring jiffie. So in case of:
1438 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1441 * we have to look at all levels @index 0. With
1443 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1446 * LVL0 has the next expiring bucket @index 2. The upper
1447 * levels have the next expiring bucket @index 1.
1449 * In case that the propagation wraps the next level the same
1452 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1455 * So after looking at LVL0 we get:
1457 * LVL5 LVL4 LVL3 LVL2 LVL1
1460 * So no propagation from LVL1 to LVL2 because that happened
1461 * with the add already, but then we need to propagate further
1462 * from LVL2 to LVL3.
1464 * So the simple check whether the lower bits of the current
1465 * level are 0 or not is sufficient for all cases.
1467 adj = clk & LVL_CLK_MASK ? 1 : 0;
1468 clk >>= LVL_CLK_SHIFT;
1475 * Check, if the next hrtimer event is before the next timer wheel
1478 static u64 cmp_next_hrtimer_event(u64 basem, u64 expires)
1480 u64 nextevt = hrtimer_get_next_event();
1483 * If high resolution timers are enabled
1484 * hrtimer_get_next_event() returns KTIME_MAX.
1486 if (expires <= nextevt)
1490 * If the next timer is already expired, return the tick base
1491 * time so the tick is fired immediately.
1493 if (nextevt <= basem)
1497 * Round up to the next jiffie. High resolution timers are
1498 * off, so the hrtimers are expired in the tick and we need to
1499 * make sure that this tick really expires the timer to avoid
1500 * a ping pong of the nohz stop code.
1502 * Use DIV_ROUND_UP_ULL to prevent gcc calling __divdi3
1504 return DIV_ROUND_UP_ULL(nextevt, TICK_NSEC) * TICK_NSEC;
1508 * get_next_timer_interrupt - return the time (clock mono) of the next timer
1509 * @basej: base time jiffies
1510 * @basem: base time clock monotonic
1512 * Returns the tick aligned clock monotonic time of the next pending
1513 * timer or KTIME_MAX if no timer is pending.
1515 u64 get_next_timer_interrupt(unsigned long basej, u64 basem)
1517 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1518 u64 expires = KTIME_MAX;
1519 unsigned long nextevt;
1523 * Pretend that there is no timer pending if the cpu is offline.
1524 * Possible pending timers will be migrated later to an active cpu.
1526 if (cpu_is_offline(smp_processor_id()))
1529 spin_lock(&base->lock);
1530 nextevt = __next_timer_interrupt(base);
1531 is_max_delta = (nextevt == base->clk + NEXT_TIMER_MAX_DELTA);
1532 base->next_expiry = nextevt;
1534 * We have a fresh next event. Check whether we can forward the
1535 * base. We can only do that when @basej is past base->clk
1536 * otherwise we might rewind base->clk.
1538 if (time_after(basej, base->clk)) {
1539 if (time_after(nextevt, basej))
1541 else if (time_after(nextevt, base->clk))
1542 base->clk = nextevt;
1545 if (time_before_eq(nextevt, basej)) {
1547 base->is_idle = false;
1550 expires = basem + (u64)(nextevt - basej) * TICK_NSEC;
1552 * If we expect to sleep more than a tick, mark the base idle.
1553 * Also the tick is stopped so any added timer must forward
1554 * the base clk itself to keep granularity small. This idle
1555 * logic is only maintained for the BASE_STD base, deferrable
1556 * timers may still see large granularity skew (by design).
1558 if ((expires - basem) > TICK_NSEC) {
1559 base->must_forward_clk = true;
1560 base->is_idle = true;
1563 spin_unlock(&base->lock);
1565 return cmp_next_hrtimer_event(basem, expires);
1569 * timer_clear_idle - Clear the idle state of the timer base
1571 * Called with interrupts disabled
1573 void timer_clear_idle(void)
1575 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1578 * We do this unlocked. The worst outcome is a remote enqueue sending
1579 * a pointless IPI, but taking the lock would just make the window for
1580 * sending the IPI a few instructions smaller for the cost of taking
1581 * the lock in the exit from idle path.
1583 base->is_idle = false;
1586 static int collect_expired_timers(struct timer_base *base,
1587 struct hlist_head *heads)
1590 * NOHZ optimization. After a long idle sleep we need to forward the
1591 * base to current jiffies. Avoid a loop by searching the bitfield for
1592 * the next expiring timer.
1594 if ((long)(jiffies - base->clk) > 2) {
1595 unsigned long next = __next_timer_interrupt(base);
1598 * If the next timer is ahead of time forward to current
1599 * jiffies, otherwise forward to the next expiry time:
1601 if (time_after(next, jiffies)) {
1602 /* The call site will increment clock! */
1603 base->clk = jiffies - 1;
1608 return __collect_expired_timers(base, heads);
1611 static inline int collect_expired_timers(struct timer_base *base,
1612 struct hlist_head *heads)
1614 return __collect_expired_timers(base, heads);
1619 * Called from the timer interrupt handler to charge one tick to the current
1620 * process. user_tick is 1 if the tick is user time, 0 for system.
1622 void update_process_times(int user_tick)
1624 struct task_struct *p = current;
1626 /* Note: this timer irq context must be accounted for as well. */
1627 account_process_tick(p, user_tick);
1629 rcu_check_callbacks(user_tick);
1630 #ifdef CONFIG_IRQ_WORK
1635 run_posix_cpu_timers(p);
1639 * __run_timers - run all expired timers (if any) on this CPU.
1640 * @base: the timer vector to be processed.
1642 static inline void __run_timers(struct timer_base *base)
1644 struct hlist_head heads[LVL_DEPTH];
1647 if (!time_after_eq(jiffies, base->clk))
1650 spin_lock_irq(&base->lock);
1653 * timer_base::must_forward_clk must be cleared before running
1654 * timers so that any timer functions that call mod_timer() will
1655 * not try to forward the base. Idle tracking / clock forwarding
1656 * logic is only used with BASE_STD timers.
1658 * The must_forward_clk flag is cleared unconditionally also for
1659 * the deferrable base. The deferrable base is not affected by idle
1660 * tracking and never forwarded, so clearing the flag is a NOOP.
1662 * The fact that the deferrable base is never forwarded can cause
1663 * large variations in granularity for deferrable timers, but they
1664 * can be deferred for long periods due to idle anyway.
1666 base->must_forward_clk = false;
1668 while (time_after_eq(jiffies, base->clk)) {
1670 levels = collect_expired_timers(base, heads);
1674 expire_timers(base, heads + levels);
1676 base->running_timer = NULL;
1677 spin_unlock_irq(&base->lock);
1681 * This function runs timers and the timer-tq in bottom half context.
1683 static __latent_entropy void run_timer_softirq(struct softirq_action *h)
1685 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1688 if (IS_ENABLED(CONFIG_NO_HZ_COMMON))
1689 __run_timers(this_cpu_ptr(&timer_bases[BASE_DEF]));
1693 * Called by the local, per-CPU timer interrupt on SMP.
1695 void run_local_timers(void)
1697 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1699 hrtimer_run_queues();
1700 /* Raise the softirq only if required. */
1701 if (time_before(jiffies, base->clk)) {
1702 if (!IS_ENABLED(CONFIG_NO_HZ_COMMON))
1704 /* CPU is awake, so check the deferrable base. */
1706 if (time_before(jiffies, base->clk))
1709 raise_softirq(TIMER_SOFTIRQ);
1712 #ifdef __ARCH_WANT_SYS_ALARM
1715 * For backwards compatibility? This can be done in libc so Alpha
1716 * and all newer ports shouldn't need it.
1718 SYSCALL_DEFINE1(alarm, unsigned int, seconds)
1720 return alarm_setitimer(seconds);
1725 static void process_timeout(unsigned long __data)
1727 wake_up_process((struct task_struct *)__data);
1731 * schedule_timeout - sleep until timeout
1732 * @timeout: timeout value in jiffies
1734 * Make the current task sleep until @timeout jiffies have
1735 * elapsed. The routine will return immediately unless
1736 * the current task state has been set (see set_current_state()).
1738 * You can set the task state as follows -
1740 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1741 * pass before the routine returns. The routine will return 0
1743 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1744 * delivered to the current task. In this case the remaining time
1745 * in jiffies will be returned, or 0 if the timer expired in time
1747 * The current task state is guaranteed to be TASK_RUNNING when this
1750 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1751 * the CPU away without a bound on the timeout. In this case the return
1752 * value will be %MAX_SCHEDULE_TIMEOUT.
1754 * In all cases the return value is guaranteed to be non-negative.
1756 signed long __sched schedule_timeout(signed long timeout)
1758 struct timer_list timer;
1759 unsigned long expire;
1763 case MAX_SCHEDULE_TIMEOUT:
1765 * These two special cases are useful to be comfortable
1766 * in the caller. Nothing more. We could take
1767 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1768 * but I' d like to return a valid offset (>=0) to allow
1769 * the caller to do everything it want with the retval.
1775 * Another bit of PARANOID. Note that the retval will be
1776 * 0 since no piece of kernel is supposed to do a check
1777 * for a negative retval of schedule_timeout() (since it
1778 * should never happens anyway). You just have the printk()
1779 * that will tell you if something is gone wrong and where.
1782 printk(KERN_ERR "schedule_timeout: wrong timeout "
1783 "value %lx\n", timeout);
1785 current->state = TASK_RUNNING;
1790 expire = timeout + jiffies;
1792 setup_timer_on_stack(&timer, process_timeout, (unsigned long)current);
1793 __mod_timer(&timer, expire, false);
1795 del_singleshot_timer_sync(&timer);
1797 /* Remove the timer from the object tracker */
1798 destroy_timer_on_stack(&timer);
1800 timeout = expire - jiffies;
1803 return timeout < 0 ? 0 : timeout;
1805 EXPORT_SYMBOL(schedule_timeout);
1808 * We can use __set_current_state() here because schedule_timeout() calls
1809 * schedule() unconditionally.
1811 signed long __sched schedule_timeout_interruptible(signed long timeout)
1813 __set_current_state(TASK_INTERRUPTIBLE);
1814 return schedule_timeout(timeout);
1816 EXPORT_SYMBOL(schedule_timeout_interruptible);
1818 signed long __sched schedule_timeout_killable(signed long timeout)
1820 __set_current_state(TASK_KILLABLE);
1821 return schedule_timeout(timeout);
1823 EXPORT_SYMBOL(schedule_timeout_killable);
1825 signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1827 __set_current_state(TASK_UNINTERRUPTIBLE);
1828 return schedule_timeout(timeout);
1830 EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1833 * Like schedule_timeout_uninterruptible(), except this task will not contribute
1836 signed long __sched schedule_timeout_idle(signed long timeout)
1838 __set_current_state(TASK_IDLE);
1839 return schedule_timeout(timeout);
1841 EXPORT_SYMBOL(schedule_timeout_idle);
1843 #ifdef CONFIG_HOTPLUG_CPU
1844 static void migrate_timer_list(struct timer_base *new_base, struct hlist_head *head)
1846 struct timer_list *timer;
1847 int cpu = new_base->cpu;
1849 while (!hlist_empty(head)) {
1850 timer = hlist_entry(head->first, struct timer_list, entry);
1851 detach_timer(timer, false);
1852 timer->flags = (timer->flags & ~TIMER_BASEMASK) | cpu;
1853 internal_add_timer(new_base, timer);
1857 int timers_prepare_cpu(unsigned int cpu)
1859 struct timer_base *base;
1862 for (b = 0; b < NR_BASES; b++) {
1863 base = per_cpu_ptr(&timer_bases[b], cpu);
1864 base->clk = jiffies;
1865 base->next_expiry = base->clk + NEXT_TIMER_MAX_DELTA;
1866 base->is_idle = false;
1867 base->must_forward_clk = true;
1872 int timers_dead_cpu(unsigned int cpu)
1874 struct timer_base *old_base;
1875 struct timer_base *new_base;
1878 BUG_ON(cpu_online(cpu));
1880 for (b = 0; b < NR_BASES; b++) {
1881 old_base = per_cpu_ptr(&timer_bases[b], cpu);
1882 new_base = get_cpu_ptr(&timer_bases[b]);
1884 * The caller is globally serialized and nobody else
1885 * takes two locks at once, deadlock is not possible.
1887 spin_lock_irq(&new_base->lock);
1888 spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1891 * The current CPUs base clock might be stale. Update it
1892 * before moving the timers over.
1894 forward_timer_base(new_base);
1896 BUG_ON(old_base->running_timer);
1898 for (i = 0; i < WHEEL_SIZE; i++)
1899 migrate_timer_list(new_base, old_base->vectors + i);
1901 spin_unlock(&old_base->lock);
1902 spin_unlock_irq(&new_base->lock);
1903 put_cpu_ptr(&timer_bases);
1908 #endif /* CONFIG_HOTPLUG_CPU */
1910 static void __init init_timer_cpu(int cpu)
1912 struct timer_base *base;
1915 for (i = 0; i < NR_BASES; i++) {
1916 base = per_cpu_ptr(&timer_bases[i], cpu);
1918 spin_lock_init(&base->lock);
1919 base->clk = jiffies;
1923 static void __init init_timer_cpus(void)
1927 for_each_possible_cpu(cpu)
1928 init_timer_cpu(cpu);
1931 void __init init_timers(void)
1935 open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
1939 * msleep - sleep safely even with waitqueue interruptions
1940 * @msecs: Time in milliseconds to sleep for
1942 void msleep(unsigned int msecs)
1944 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1947 timeout = schedule_timeout_uninterruptible(timeout);
1950 EXPORT_SYMBOL(msleep);
1953 * msleep_interruptible - sleep waiting for signals
1954 * @msecs: Time in milliseconds to sleep for
1956 unsigned long msleep_interruptible(unsigned int msecs)
1958 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1960 while (timeout && !signal_pending(current))
1961 timeout = schedule_timeout_interruptible(timeout);
1962 return jiffies_to_msecs(timeout);
1965 EXPORT_SYMBOL(msleep_interruptible);
1967 static void __sched do_usleep_range(unsigned long min, unsigned long max)
1972 kmin = ktime_set(0, min * NSEC_PER_USEC);
1973 delta = (u64)(max - min) * NSEC_PER_USEC;
1974 schedule_hrtimeout_range(&kmin, delta, HRTIMER_MODE_REL);
1978 * usleep_range - Sleep for an approximate time
1979 * @min: Minimum time in usecs to sleep
1980 * @max: Maximum time in usecs to sleep
1982 * In non-atomic context where the exact wakeup time is flexible, use
1983 * usleep_range() instead of udelay(). The sleep improves responsiveness
1984 * by avoiding the CPU-hogging busy-wait of udelay(), and the range reduces
1985 * power usage by allowing hrtimers to take advantage of an already-
1986 * scheduled interrupt instead of scheduling a new one just for this sleep.
1988 void __sched usleep_range(unsigned long min, unsigned long max)
1990 __set_current_state(TASK_UNINTERRUPTIBLE);
1991 do_usleep_range(min, max);
1993 EXPORT_SYMBOL(usleep_range);