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 #define CREATE_TRACE_POINTS
53 #include <trace/events/timer.h>
55 u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
57 EXPORT_SYMBOL(jiffies_64);
60 * per-CPU timer vector definitions:
62 #define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6)
63 #define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8)
64 #define TVN_SIZE (1 << TVN_BITS)
65 #define TVR_SIZE (1 << TVR_BITS)
66 #define TVN_MASK (TVN_SIZE - 1)
67 #define TVR_MASK (TVR_SIZE - 1)
68 #define MAX_TVAL ((unsigned long)((1ULL << (TVR_BITS + 4*TVN_BITS)) - 1))
71 struct list_head vec[TVN_SIZE];
75 struct list_head vec[TVR_SIZE];
80 struct timer_list *running_timer;
81 unsigned long timer_jiffies;
82 unsigned long next_timer;
83 unsigned long active_timers;
89 } ____cacheline_aligned;
91 struct tvec_base boot_tvec_bases;
92 EXPORT_SYMBOL(boot_tvec_bases);
93 static DEFINE_PER_CPU(struct tvec_base *, tvec_bases) = &boot_tvec_bases;
95 /* Functions below help us manage 'deferrable' flag */
96 static inline unsigned int tbase_get_deferrable(struct tvec_base *base)
98 return ((unsigned int)(unsigned long)base & TIMER_DEFERRABLE);
101 static inline unsigned int tbase_get_irqsafe(struct tvec_base *base)
103 return ((unsigned int)(unsigned long)base & TIMER_IRQSAFE);
106 static inline struct tvec_base *tbase_get_base(struct tvec_base *base)
108 return ((struct tvec_base *)((unsigned long)base & ~TIMER_FLAG_MASK));
112 timer_set_base(struct timer_list *timer, struct tvec_base *new_base)
114 unsigned long flags = (unsigned long)timer->base & TIMER_FLAG_MASK;
116 timer->base = (struct tvec_base *)((unsigned long)(new_base) | flags);
119 static unsigned long round_jiffies_common(unsigned long j, int cpu,
123 unsigned long original = j;
126 * We don't want all cpus firing their timers at once hitting the
127 * same lock or cachelines, so we skew each extra cpu with an extra
128 * 3 jiffies. This 3 jiffies came originally from the mm/ code which
130 * The skew is done by adding 3*cpunr, then round, then subtract this
131 * extra offset again.
138 * If the target jiffie is just after a whole second (which can happen
139 * due to delays of the timer irq, long irq off times etc etc) then
140 * we should round down to the whole second, not up. Use 1/4th second
141 * as cutoff for this rounding as an extreme upper bound for this.
142 * But never round down if @force_up is set.
144 if (rem < HZ/4 && !force_up) /* round down */
149 /* now that we have rounded, subtract the extra skew again */
153 * Make sure j is still in the future. Otherwise return the
156 return time_is_after_jiffies(j) ? j : original;
160 * __round_jiffies - function to round jiffies to a full second
161 * @j: the time in (absolute) jiffies that should be rounded
162 * @cpu: the processor number on which the timeout will happen
164 * __round_jiffies() rounds an absolute time in the future (in jiffies)
165 * up or down to (approximately) full seconds. This is useful for timers
166 * for which the exact time they fire does not matter too much, as long as
167 * they fire approximately every X seconds.
169 * By rounding these timers to whole seconds, all such timers will fire
170 * at the same time, rather than at various times spread out. The goal
171 * of this is to have the CPU wake up less, which saves power.
173 * The exact rounding is skewed for each processor to avoid all
174 * processors firing at the exact same time, which could lead
175 * to lock contention or spurious cache line bouncing.
177 * The return value is the rounded version of the @j parameter.
179 unsigned long __round_jiffies(unsigned long j, int cpu)
181 return round_jiffies_common(j, cpu, false);
183 EXPORT_SYMBOL_GPL(__round_jiffies);
186 * __round_jiffies_relative - function to round jiffies to a full second
187 * @j: the time in (relative) jiffies that should be rounded
188 * @cpu: the processor number on which the timeout will happen
190 * __round_jiffies_relative() rounds a time delta in the future (in jiffies)
191 * up or down to (approximately) full seconds. This is useful for timers
192 * for which the exact time they fire does not matter too much, as long as
193 * they fire approximately every X seconds.
195 * By rounding these timers to whole seconds, all such timers will fire
196 * at the same time, rather than at various times spread out. The goal
197 * of this is to have the CPU wake up less, which saves power.
199 * The exact rounding is skewed for each processor to avoid all
200 * processors firing at the exact same time, which could lead
201 * to lock contention or spurious cache line bouncing.
203 * The return value is the rounded version of the @j parameter.
205 unsigned long __round_jiffies_relative(unsigned long j, int cpu)
207 unsigned long j0 = jiffies;
209 /* Use j0 because jiffies might change while we run */
210 return round_jiffies_common(j + j0, cpu, false) - j0;
212 EXPORT_SYMBOL_GPL(__round_jiffies_relative);
215 * round_jiffies - function to round jiffies to a full second
216 * @j: the time in (absolute) jiffies that should be rounded
218 * round_jiffies() rounds an absolute time in the future (in jiffies)
219 * up or down to (approximately) full seconds. This is useful for timers
220 * for which the exact time they fire does not matter too much, as long as
221 * they fire approximately every X seconds.
223 * By rounding these timers to whole seconds, all such timers will fire
224 * at the same time, rather than at various times spread out. The goal
225 * of this is to have the CPU wake up less, which saves power.
227 * The return value is the rounded version of the @j parameter.
229 unsigned long round_jiffies(unsigned long j)
231 return round_jiffies_common(j, raw_smp_processor_id(), false);
233 EXPORT_SYMBOL_GPL(round_jiffies);
236 * round_jiffies_relative - function to round jiffies to a full second
237 * @j: the time in (relative) jiffies that should be rounded
239 * round_jiffies_relative() rounds a time delta in the future (in jiffies)
240 * up or down to (approximately) full seconds. This is useful for timers
241 * for which the exact time they fire does not matter too much, as long as
242 * they fire approximately every X seconds.
244 * By rounding these timers to whole seconds, all such timers will fire
245 * at the same time, rather than at various times spread out. The goal
246 * of this is to have the CPU wake up less, which saves power.
248 * The return value is the rounded version of the @j parameter.
250 unsigned long round_jiffies_relative(unsigned long j)
252 return __round_jiffies_relative(j, raw_smp_processor_id());
254 EXPORT_SYMBOL_GPL(round_jiffies_relative);
257 * __round_jiffies_up - function to round jiffies up to a full second
258 * @j: the time in (absolute) jiffies that should be rounded
259 * @cpu: the processor number on which the timeout will happen
261 * This is the same as __round_jiffies() except that it will never
262 * round down. This is useful for timeouts for which the exact time
263 * of firing does not matter too much, as long as they don't fire too
266 unsigned long __round_jiffies_up(unsigned long j, int cpu)
268 return round_jiffies_common(j, cpu, true);
270 EXPORT_SYMBOL_GPL(__round_jiffies_up);
273 * __round_jiffies_up_relative - function to round jiffies up to a full second
274 * @j: the time in (relative) jiffies that should be rounded
275 * @cpu: the processor number on which the timeout will happen
277 * This is the same as __round_jiffies_relative() except that it will never
278 * round down. This is useful for timeouts for which the exact time
279 * of firing does not matter too much, as long as they don't fire too
282 unsigned long __round_jiffies_up_relative(unsigned long j, int cpu)
284 unsigned long j0 = jiffies;
286 /* Use j0 because jiffies might change while we run */
287 return round_jiffies_common(j + j0, cpu, true) - j0;
289 EXPORT_SYMBOL_GPL(__round_jiffies_up_relative);
292 * round_jiffies_up - function to round jiffies up to a full second
293 * @j: the time in (absolute) jiffies that should be rounded
295 * This is the same as round_jiffies() except that it will never
296 * round down. This is useful for timeouts for which the exact time
297 * of firing does not matter too much, as long as they don't fire too
300 unsigned long round_jiffies_up(unsigned long j)
302 return round_jiffies_common(j, raw_smp_processor_id(), true);
304 EXPORT_SYMBOL_GPL(round_jiffies_up);
307 * round_jiffies_up_relative - function to round jiffies up to a full second
308 * @j: the time in (relative) jiffies that should be rounded
310 * This is the same as round_jiffies_relative() except that it will never
311 * round down. This is useful for timeouts for which the exact time
312 * of firing does not matter too much, as long as they don't fire too
315 unsigned long round_jiffies_up_relative(unsigned long j)
317 return __round_jiffies_up_relative(j, raw_smp_processor_id());
319 EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
322 * set_timer_slack - set the allowed slack for a timer
323 * @timer: the timer to be modified
324 * @slack_hz: the amount of time (in jiffies) allowed for rounding
326 * Set the amount of time, in jiffies, that a certain timer has
327 * in terms of slack. By setting this value, the timer subsystem
328 * will schedule the actual timer somewhere between
329 * the time mod_timer() asks for, and that time plus the slack.
331 * By setting the slack to -1, a percentage of the delay is used
334 void set_timer_slack(struct timer_list *timer, int slack_hz)
336 timer->slack = slack_hz;
338 EXPORT_SYMBOL_GPL(set_timer_slack);
341 __internal_add_timer(struct tvec_base *base, struct timer_list *timer)
343 unsigned long expires = timer->expires;
344 unsigned long idx = expires - base->timer_jiffies;
345 struct list_head *vec;
347 if (idx < TVR_SIZE) {
348 int i = expires & TVR_MASK;
349 vec = base->tv1.vec + i;
350 } else if (idx < 1 << (TVR_BITS + TVN_BITS)) {
351 int i = (expires >> TVR_BITS) & TVN_MASK;
352 vec = base->tv2.vec + i;
353 } else if (idx < 1 << (TVR_BITS + 2 * TVN_BITS)) {
354 int i = (expires >> (TVR_BITS + TVN_BITS)) & TVN_MASK;
355 vec = base->tv3.vec + i;
356 } else if (idx < 1 << (TVR_BITS + 3 * TVN_BITS)) {
357 int i = (expires >> (TVR_BITS + 2 * TVN_BITS)) & TVN_MASK;
358 vec = base->tv4.vec + i;
359 } else if ((signed long) idx < 0) {
361 * Can happen if you add a timer with expires == jiffies,
362 * or you set a timer to go off in the past
364 vec = base->tv1.vec + (base->timer_jiffies & TVR_MASK);
367 /* If the timeout is larger than MAX_TVAL (on 64-bit
368 * architectures or with CONFIG_BASE_SMALL=1) then we
369 * use the maximum timeout.
371 if (idx > MAX_TVAL) {
373 expires = idx + base->timer_jiffies;
375 i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK;
376 vec = base->tv5.vec + i;
381 list_add_tail(&timer->entry, vec);
384 static void internal_add_timer(struct tvec_base *base, struct timer_list *timer)
386 __internal_add_timer(base, timer);
388 * Update base->active_timers and base->next_timer
390 if (!tbase_get_deferrable(timer->base)) {
391 if (time_before(timer->expires, base->next_timer))
392 base->next_timer = timer->expires;
393 base->active_timers++;
397 #ifdef CONFIG_TIMER_STATS
398 void __timer_stats_timer_set_start_info(struct timer_list *timer, void *addr)
400 if (timer->start_site)
403 timer->start_site = addr;
404 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
405 timer->start_pid = current->pid;
408 static void timer_stats_account_timer(struct timer_list *timer)
410 unsigned int flag = 0;
412 if (likely(!timer->start_site))
414 if (unlikely(tbase_get_deferrable(timer->base)))
415 flag |= TIMER_STATS_FLAG_DEFERRABLE;
417 timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
418 timer->function, timer->start_comm, flag);
422 static void timer_stats_account_timer(struct timer_list *timer) {}
425 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
427 static struct debug_obj_descr timer_debug_descr;
429 static void *timer_debug_hint(void *addr)
431 return ((struct timer_list *) addr)->function;
435 * fixup_init is called when:
436 * - an active object is initialized
438 static int timer_fixup_init(void *addr, enum debug_obj_state state)
440 struct timer_list *timer = addr;
443 case ODEBUG_STATE_ACTIVE:
444 del_timer_sync(timer);
445 debug_object_init(timer, &timer_debug_descr);
452 /* Stub timer callback for improperly used timers. */
453 static void stub_timer(unsigned long data)
459 * fixup_activate is called when:
460 * - an active object is activated
461 * - an unknown object is activated (might be a statically initialized object)
463 static int timer_fixup_activate(void *addr, enum debug_obj_state state)
465 struct timer_list *timer = addr;
469 case ODEBUG_STATE_NOTAVAILABLE:
471 * This is not really a fixup. The timer was
472 * statically initialized. We just make sure that it
473 * is tracked in the object tracker.
475 if (timer->entry.next == NULL &&
476 timer->entry.prev == TIMER_ENTRY_STATIC) {
477 debug_object_init(timer, &timer_debug_descr);
478 debug_object_activate(timer, &timer_debug_descr);
481 setup_timer(timer, stub_timer, 0);
486 case ODEBUG_STATE_ACTIVE:
495 * fixup_free is called when:
496 * - an active object is freed
498 static int timer_fixup_free(void *addr, enum debug_obj_state state)
500 struct timer_list *timer = addr;
503 case ODEBUG_STATE_ACTIVE:
504 del_timer_sync(timer);
505 debug_object_free(timer, &timer_debug_descr);
513 * fixup_assert_init is called when:
514 * - an untracked/uninit-ed object is found
516 static int timer_fixup_assert_init(void *addr, enum debug_obj_state state)
518 struct timer_list *timer = addr;
521 case ODEBUG_STATE_NOTAVAILABLE:
522 if (timer->entry.prev == TIMER_ENTRY_STATIC) {
524 * This is not really a fixup. The timer was
525 * statically initialized. We just make sure that it
526 * is tracked in the object tracker.
528 debug_object_init(timer, &timer_debug_descr);
531 setup_timer(timer, stub_timer, 0);
539 static struct debug_obj_descr timer_debug_descr = {
540 .name = "timer_list",
541 .debug_hint = timer_debug_hint,
542 .fixup_init = timer_fixup_init,
543 .fixup_activate = timer_fixup_activate,
544 .fixup_free = timer_fixup_free,
545 .fixup_assert_init = timer_fixup_assert_init,
548 static inline void debug_timer_init(struct timer_list *timer)
550 debug_object_init(timer, &timer_debug_descr);
553 static inline void debug_timer_activate(struct timer_list *timer)
555 debug_object_activate(timer, &timer_debug_descr);
558 static inline void debug_timer_deactivate(struct timer_list *timer)
560 debug_object_deactivate(timer, &timer_debug_descr);
563 static inline void debug_timer_free(struct timer_list *timer)
565 debug_object_free(timer, &timer_debug_descr);
568 static inline void debug_timer_assert_init(struct timer_list *timer)
570 debug_object_assert_init(timer, &timer_debug_descr);
573 static void do_init_timer(struct timer_list *timer, unsigned int flags,
574 const char *name, struct lock_class_key *key);
576 void init_timer_on_stack_key(struct timer_list *timer, unsigned int flags,
577 const char *name, struct lock_class_key *key)
579 debug_object_init_on_stack(timer, &timer_debug_descr);
580 do_init_timer(timer, flags, name, key);
582 EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
584 void destroy_timer_on_stack(struct timer_list *timer)
586 debug_object_free(timer, &timer_debug_descr);
588 EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
591 static inline void debug_timer_init(struct timer_list *timer) { }
592 static inline void debug_timer_activate(struct timer_list *timer) { }
593 static inline void debug_timer_deactivate(struct timer_list *timer) { }
594 static inline void debug_timer_assert_init(struct timer_list *timer) { }
597 static inline void debug_init(struct timer_list *timer)
599 debug_timer_init(timer);
600 trace_timer_init(timer);
604 debug_activate(struct timer_list *timer, unsigned long expires)
606 debug_timer_activate(timer);
607 trace_timer_start(timer, expires);
610 static inline void debug_deactivate(struct timer_list *timer)
612 debug_timer_deactivate(timer);
613 trace_timer_cancel(timer);
616 static inline void debug_assert_init(struct timer_list *timer)
618 debug_timer_assert_init(timer);
621 static void do_init_timer(struct timer_list *timer, unsigned int flags,
622 const char *name, struct lock_class_key *key)
624 struct tvec_base *base = __raw_get_cpu_var(tvec_bases);
626 timer->entry.next = NULL;
627 timer->base = (void *)((unsigned long)base | flags);
629 #ifdef CONFIG_TIMER_STATS
630 timer->start_site = NULL;
631 timer->start_pid = -1;
632 memset(timer->start_comm, 0, TASK_COMM_LEN);
634 lockdep_init_map(&timer->lockdep_map, name, key, 0);
638 * init_timer_key - initialize a timer
639 * @timer: the timer to be initialized
640 * @flags: timer flags
641 * @name: name of the timer
642 * @key: lockdep class key of the fake lock used for tracking timer
643 * sync lock dependencies
645 * init_timer_key() must be done to a timer prior calling *any* of the
646 * other timer functions.
648 void init_timer_key(struct timer_list *timer, unsigned int flags,
649 const char *name, struct lock_class_key *key)
652 do_init_timer(timer, flags, name, key);
654 EXPORT_SYMBOL(init_timer_key);
656 static inline void detach_timer(struct timer_list *timer, bool clear_pending)
658 struct list_head *entry = &timer->entry;
660 debug_deactivate(timer);
662 __list_del(entry->prev, entry->next);
665 entry->prev = LIST_POISON2;
669 detach_expired_timer(struct timer_list *timer, struct tvec_base *base)
671 detach_timer(timer, true);
672 if (!tbase_get_deferrable(timer->base))
673 base->active_timers--;
676 static int detach_if_pending(struct timer_list *timer, struct tvec_base *base,
679 if (!timer_pending(timer))
682 detach_timer(timer, clear_pending);
683 if (!tbase_get_deferrable(timer->base)) {
684 base->active_timers--;
685 if (timer->expires == base->next_timer)
686 base->next_timer = base->timer_jiffies;
692 * We are using hashed locking: holding per_cpu(tvec_bases).lock
693 * means that all timers which are tied to this base via timer->base are
694 * locked, and the base itself is locked too.
696 * So __run_timers/migrate_timers can safely modify all timers which could
697 * be found on ->tvX lists.
699 * When the timer's base is locked, and the timer removed from list, it is
700 * possible to set timer->base = NULL and drop the lock: the timer remains
703 static struct tvec_base *lock_timer_base(struct timer_list *timer,
704 unsigned long *flags)
705 __acquires(timer->base->lock)
707 struct tvec_base *base;
710 struct tvec_base *prelock_base = timer->base;
711 base = tbase_get_base(prelock_base);
712 if (likely(base != NULL)) {
713 spin_lock_irqsave(&base->lock, *flags);
714 if (likely(prelock_base == timer->base))
716 /* The timer has migrated to another CPU */
717 spin_unlock_irqrestore(&base->lock, *flags);
724 __mod_timer(struct timer_list *timer, unsigned long expires,
725 bool pending_only, int pinned)
727 struct tvec_base *base, *new_base;
731 timer_stats_timer_set_start_info(timer);
732 BUG_ON(!timer->function);
734 base = lock_timer_base(timer, &flags);
736 ret = detach_if_pending(timer, base, false);
737 if (!ret && pending_only)
740 debug_activate(timer, expires);
742 cpu = smp_processor_id();
744 #if defined(CONFIG_NO_HZ_COMMON) && defined(CONFIG_SMP)
745 if (!pinned && get_sysctl_timer_migration() && idle_cpu(cpu))
746 cpu = get_nohz_timer_target();
748 new_base = per_cpu(tvec_bases, cpu);
750 if (base != new_base) {
752 * We are trying to schedule the timer on the local CPU.
753 * However we can't change timer's base while it is running,
754 * otherwise del_timer_sync() can't detect that the timer's
755 * handler yet has not finished. This also guarantees that
756 * the timer is serialized wrt itself.
758 if (likely(base->running_timer != timer)) {
759 /* See the comment in lock_timer_base() */
760 timer_set_base(timer, NULL);
761 spin_unlock(&base->lock);
763 spin_lock(&base->lock);
764 timer_set_base(timer, base);
768 timer->expires = expires;
769 internal_add_timer(base, timer);
772 spin_unlock_irqrestore(&base->lock, flags);
778 * mod_timer_pending - modify a pending timer's timeout
779 * @timer: the pending timer to be modified
780 * @expires: new timeout in jiffies
782 * mod_timer_pending() is the same for pending timers as mod_timer(),
783 * but will not re-activate and modify already deleted timers.
785 * It is useful for unserialized use of timers.
787 int mod_timer_pending(struct timer_list *timer, unsigned long expires)
789 return __mod_timer(timer, expires, true, TIMER_NOT_PINNED);
791 EXPORT_SYMBOL(mod_timer_pending);
794 * Decide where to put the timer while taking the slack into account
797 * 1) calculate the maximum (absolute) time
798 * 2) calculate the highest bit where the expires and new max are different
799 * 3) use this bit to make a mask
800 * 4) use the bitmask to round down the maximum time, so that all last
804 unsigned long apply_slack(struct timer_list *timer, unsigned long expires)
806 unsigned long expires_limit, mask;
809 if (timer->slack >= 0) {
810 expires_limit = expires + timer->slack;
812 long delta = expires - jiffies;
817 expires_limit = expires + delta / 256;
819 mask = expires ^ expires_limit;
823 bit = find_last_bit(&mask, BITS_PER_LONG);
825 mask = (1 << bit) - 1;
827 expires_limit = expires_limit & ~(mask);
829 return expires_limit;
833 * mod_timer - modify a timer's timeout
834 * @timer: the timer to be modified
835 * @expires: new timeout in jiffies
837 * mod_timer() is a more efficient way to update the expire field of an
838 * active timer (if the timer is inactive it will be activated)
840 * mod_timer(timer, expires) is equivalent to:
842 * del_timer(timer); timer->expires = expires; add_timer(timer);
844 * Note that if there are multiple unserialized concurrent users of the
845 * same timer, then mod_timer() is the only safe way to modify the timeout,
846 * since add_timer() cannot modify an already running timer.
848 * The function returns whether it has modified a pending timer or not.
849 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
850 * active timer returns 1.)
852 int mod_timer(struct timer_list *timer, unsigned long expires)
854 expires = apply_slack(timer, expires);
857 * This is a common optimization triggered by the
858 * networking code - if the timer is re-modified
859 * to be the same thing then just return:
861 if (timer_pending(timer) && timer->expires == expires)
864 return __mod_timer(timer, expires, false, TIMER_NOT_PINNED);
866 EXPORT_SYMBOL(mod_timer);
869 * mod_timer_pinned - modify a timer's timeout
870 * @timer: the timer to be modified
871 * @expires: new timeout in jiffies
873 * mod_timer_pinned() is a way to update the expire field of an
874 * active timer (if the timer is inactive it will be activated)
875 * and to ensure that the timer is scheduled on the current CPU.
877 * Note that this does not prevent the timer from being migrated
878 * when the current CPU goes offline. If this is a problem for
879 * you, use CPU-hotplug notifiers to handle it correctly, for
880 * example, cancelling the timer when the corresponding CPU goes
883 * mod_timer_pinned(timer, expires) is equivalent to:
885 * del_timer(timer); timer->expires = expires; add_timer(timer);
887 int mod_timer_pinned(struct timer_list *timer, unsigned long expires)
889 if (timer->expires == expires && timer_pending(timer))
892 return __mod_timer(timer, expires, false, TIMER_PINNED);
894 EXPORT_SYMBOL(mod_timer_pinned);
897 * add_timer - start a timer
898 * @timer: the timer to be added
900 * The kernel will do a ->function(->data) callback from the
901 * timer interrupt at the ->expires point in the future. The
902 * current time is 'jiffies'.
904 * The timer's ->expires, ->function (and if the handler uses it, ->data)
905 * fields must be set prior calling this function.
907 * Timers with an ->expires field in the past will be executed in the next
910 void add_timer(struct timer_list *timer)
912 BUG_ON(timer_pending(timer));
913 mod_timer(timer, timer->expires);
915 EXPORT_SYMBOL(add_timer);
918 * add_timer_on - start a timer on a particular CPU
919 * @timer: the timer to be added
920 * @cpu: the CPU to start it on
922 * This is not very scalable on SMP. Double adds are not possible.
924 void add_timer_on(struct timer_list *timer, int cpu)
926 struct tvec_base *base = per_cpu(tvec_bases, cpu);
929 timer_stats_timer_set_start_info(timer);
930 BUG_ON(timer_pending(timer) || !timer->function);
931 spin_lock_irqsave(&base->lock, flags);
932 timer_set_base(timer, base);
933 debug_activate(timer, timer->expires);
934 internal_add_timer(base, timer);
936 * Check whether the other CPU is in dynticks mode and needs
937 * to be triggered to reevaluate the timer wheel.
938 * We are protected against the other CPU fiddling
939 * with the timer by holding the timer base lock. This also
940 * makes sure that a CPU on the way to stop its tick can not
941 * evaluate the timer wheel.
943 wake_up_nohz_cpu(cpu);
944 spin_unlock_irqrestore(&base->lock, flags);
946 EXPORT_SYMBOL_GPL(add_timer_on);
949 * del_timer - deactive a timer.
950 * @timer: the timer to be deactivated
952 * del_timer() deactivates a timer - this works on both active and inactive
955 * The function returns whether it has deactivated a pending timer or not.
956 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
957 * active timer returns 1.)
959 int del_timer(struct timer_list *timer)
961 struct tvec_base *base;
965 debug_assert_init(timer);
967 timer_stats_timer_clear_start_info(timer);
968 if (timer_pending(timer)) {
969 base = lock_timer_base(timer, &flags);
970 ret = detach_if_pending(timer, base, true);
971 spin_unlock_irqrestore(&base->lock, flags);
976 EXPORT_SYMBOL(del_timer);
979 * try_to_del_timer_sync - Try to deactivate a timer
980 * @timer: timer do del
982 * This function tries to deactivate a timer. Upon successful (ret >= 0)
983 * exit the timer is not queued and the handler is not running on any CPU.
985 int try_to_del_timer_sync(struct timer_list *timer)
987 struct tvec_base *base;
991 debug_assert_init(timer);
993 base = lock_timer_base(timer, &flags);
995 if (base->running_timer != timer) {
996 timer_stats_timer_clear_start_info(timer);
997 ret = detach_if_pending(timer, base, true);
999 spin_unlock_irqrestore(&base->lock, flags);
1003 EXPORT_SYMBOL(try_to_del_timer_sync);
1007 * del_timer_sync - deactivate a timer and wait for the handler to finish.
1008 * @timer: the timer to be deactivated
1010 * This function only differs from del_timer() on SMP: besides deactivating
1011 * the timer it also makes sure the handler has finished executing on other
1014 * Synchronization rules: Callers must prevent restarting of the timer,
1015 * otherwise this function is meaningless. It must not be called from
1016 * interrupt contexts unless the timer is an irqsafe one. The caller must
1017 * not hold locks which would prevent completion of the timer's
1018 * handler. The timer's handler must not call add_timer_on(). Upon exit the
1019 * timer is not queued and the handler is not running on any CPU.
1021 * Note: For !irqsafe timers, you must not hold locks that are held in
1022 * interrupt context while calling this function. Even if the lock has
1023 * nothing to do with the timer in question. Here's why:
1029 * base->running_timer = mytimer;
1030 * spin_lock_irq(somelock);
1032 * spin_lock(somelock);
1033 * del_timer_sync(mytimer);
1034 * while (base->running_timer == mytimer);
1036 * Now del_timer_sync() will never return and never release somelock.
1037 * The interrupt on the other CPU is waiting to grab somelock but
1038 * it has interrupted the softirq that CPU0 is waiting to finish.
1040 * The function returns whether it has deactivated a pending timer or not.
1042 int del_timer_sync(struct timer_list *timer)
1044 #ifdef CONFIG_LOCKDEP
1045 unsigned long flags;
1048 * If lockdep gives a backtrace here, please reference
1049 * the synchronization rules above.
1051 local_irq_save(flags);
1052 lock_map_acquire(&timer->lockdep_map);
1053 lock_map_release(&timer->lockdep_map);
1054 local_irq_restore(flags);
1057 * don't use it in hardirq context, because it
1058 * could lead to deadlock.
1060 WARN_ON(in_irq() && !tbase_get_irqsafe(timer->base));
1062 int ret = try_to_del_timer_sync(timer);
1068 EXPORT_SYMBOL(del_timer_sync);
1071 static int cascade(struct tvec_base *base, struct tvec *tv, int index)
1073 /* cascade all the timers from tv up one level */
1074 struct timer_list *timer, *tmp;
1075 struct list_head tv_list;
1077 list_replace_init(tv->vec + index, &tv_list);
1080 * We are removing _all_ timers from the list, so we
1081 * don't have to detach them individually.
1083 list_for_each_entry_safe(timer, tmp, &tv_list, entry) {
1084 BUG_ON(tbase_get_base(timer->base) != base);
1085 /* No accounting, while moving them */
1086 __internal_add_timer(base, timer);
1092 static void call_timer_fn(struct timer_list *timer, void (*fn)(unsigned long),
1095 int count = preempt_count();
1097 #ifdef CONFIG_LOCKDEP
1099 * It is permissible to free the timer from inside the
1100 * function that is called from it, this we need to take into
1101 * account for lockdep too. To avoid bogus "held lock freed"
1102 * warnings as well as problems when looking into
1103 * timer->lockdep_map, make a copy and use that here.
1105 struct lockdep_map lockdep_map;
1107 lockdep_copy_map(&lockdep_map, &timer->lockdep_map);
1110 * Couple the lock chain with the lock chain at
1111 * del_timer_sync() by acquiring the lock_map around the fn()
1112 * call here and in del_timer_sync().
1114 lock_map_acquire(&lockdep_map);
1116 trace_timer_expire_entry(timer);
1118 trace_timer_expire_exit(timer);
1120 lock_map_release(&lockdep_map);
1122 if (count != preempt_count()) {
1123 WARN_ONCE(1, "timer: %pF preempt leak: %08x -> %08x\n",
1124 fn, count, preempt_count());
1126 * Restore the preempt count. That gives us a decent
1127 * chance to survive and extract information. If the
1128 * callback kept a lock held, bad luck, but not worse
1129 * than the BUG() we had.
1131 preempt_count_set(count);
1135 #define INDEX(N) ((base->timer_jiffies >> (TVR_BITS + (N) * TVN_BITS)) & TVN_MASK)
1138 * __run_timers - run all expired timers (if any) on this CPU.
1139 * @base: the timer vector to be processed.
1141 * This function cascades all vectors and executes all expired timer
1144 static inline void __run_timers(struct tvec_base *base)
1146 struct timer_list *timer;
1148 spin_lock_irq(&base->lock);
1149 while (time_after_eq(jiffies, base->timer_jiffies)) {
1150 struct list_head work_list;
1151 struct list_head *head = &work_list;
1152 int index = base->timer_jiffies & TVR_MASK;
1158 (!cascade(base, &base->tv2, INDEX(0))) &&
1159 (!cascade(base, &base->tv3, INDEX(1))) &&
1160 !cascade(base, &base->tv4, INDEX(2)))
1161 cascade(base, &base->tv5, INDEX(3));
1162 ++base->timer_jiffies;
1163 list_replace_init(base->tv1.vec + index, &work_list);
1164 while (!list_empty(head)) {
1165 void (*fn)(unsigned long);
1169 timer = list_first_entry(head, struct timer_list,entry);
1170 fn = timer->function;
1172 irqsafe = tbase_get_irqsafe(timer->base);
1174 timer_stats_account_timer(timer);
1176 base->running_timer = timer;
1177 detach_expired_timer(timer, base);
1180 spin_unlock(&base->lock);
1181 call_timer_fn(timer, fn, data);
1182 spin_lock(&base->lock);
1184 spin_unlock_irq(&base->lock);
1185 call_timer_fn(timer, fn, data);
1186 spin_lock_irq(&base->lock);
1190 base->running_timer = NULL;
1191 spin_unlock_irq(&base->lock);
1194 #ifdef CONFIG_NO_HZ_COMMON
1196 * Find out when the next timer event is due to happen. This
1197 * is used on S/390 to stop all activity when a CPU is idle.
1198 * This function needs to be called with interrupts disabled.
1200 static unsigned long __next_timer_interrupt(struct tvec_base *base)
1202 unsigned long timer_jiffies = base->timer_jiffies;
1203 unsigned long expires = timer_jiffies + NEXT_TIMER_MAX_DELTA;
1204 int index, slot, array, found = 0;
1205 struct timer_list *nte;
1206 struct tvec *varray[4];
1208 /* Look for timer events in tv1. */
1209 index = slot = timer_jiffies & TVR_MASK;
1211 list_for_each_entry(nte, base->tv1.vec + slot, entry) {
1212 if (tbase_get_deferrable(nte->base))
1216 expires = nte->expires;
1217 /* Look at the cascade bucket(s)? */
1218 if (!index || slot < index)
1222 slot = (slot + 1) & TVR_MASK;
1223 } while (slot != index);
1226 /* Calculate the next cascade event */
1228 timer_jiffies += TVR_SIZE - index;
1229 timer_jiffies >>= TVR_BITS;
1231 /* Check tv2-tv5. */
1232 varray[0] = &base->tv2;
1233 varray[1] = &base->tv3;
1234 varray[2] = &base->tv4;
1235 varray[3] = &base->tv5;
1237 for (array = 0; array < 4; array++) {
1238 struct tvec *varp = varray[array];
1240 index = slot = timer_jiffies & TVN_MASK;
1242 list_for_each_entry(nte, varp->vec + slot, entry) {
1243 if (tbase_get_deferrable(nte->base))
1247 if (time_before(nte->expires, expires))
1248 expires = nte->expires;
1251 * Do we still search for the first timer or are
1252 * we looking up the cascade buckets ?
1255 /* Look at the cascade bucket(s)? */
1256 if (!index || slot < index)
1260 slot = (slot + 1) & TVN_MASK;
1261 } while (slot != index);
1264 timer_jiffies += TVN_SIZE - index;
1265 timer_jiffies >>= TVN_BITS;
1271 * Check, if the next hrtimer event is before the next timer wheel
1274 static unsigned long cmp_next_hrtimer_event(unsigned long now,
1275 unsigned long expires)
1277 ktime_t hr_delta = hrtimer_get_next_event();
1278 struct timespec tsdelta;
1279 unsigned long delta;
1281 if (hr_delta.tv64 == KTIME_MAX)
1285 * Expired timer available, let it expire in the next tick
1287 if (hr_delta.tv64 <= 0)
1290 tsdelta = ktime_to_timespec(hr_delta);
1291 delta = timespec_to_jiffies(&tsdelta);
1294 * Limit the delta to the max value, which is checked in
1295 * tick_nohz_stop_sched_tick():
1297 if (delta > NEXT_TIMER_MAX_DELTA)
1298 delta = NEXT_TIMER_MAX_DELTA;
1301 * Take rounding errors in to account and make sure, that it
1302 * expires in the next tick. Otherwise we go into an endless
1303 * ping pong due to tick_nohz_stop_sched_tick() retriggering
1309 if (time_before(now, expires))
1315 * get_next_timer_interrupt - return the jiffy of the next pending timer
1316 * @now: current time (in jiffies)
1318 unsigned long get_next_timer_interrupt(unsigned long now)
1320 struct tvec_base *base = __this_cpu_read(tvec_bases);
1321 unsigned long expires = now + NEXT_TIMER_MAX_DELTA;
1324 * Pretend that there is no timer pending if the cpu is offline.
1325 * Possible pending timers will be migrated later to an active cpu.
1327 if (cpu_is_offline(smp_processor_id()))
1330 spin_lock(&base->lock);
1331 if (base->active_timers) {
1332 if (time_before_eq(base->next_timer, base->timer_jiffies))
1333 base->next_timer = __next_timer_interrupt(base);
1334 expires = base->next_timer;
1336 spin_unlock(&base->lock);
1338 if (time_before_eq(expires, now))
1341 return cmp_next_hrtimer_event(now, expires);
1346 * Called from the timer interrupt handler to charge one tick to the current
1347 * process. user_tick is 1 if the tick is user time, 0 for system.
1349 void update_process_times(int user_tick)
1351 struct task_struct *p = current;
1352 int cpu = smp_processor_id();
1354 /* Note: this timer irq context must be accounted for as well. */
1355 account_process_tick(p, user_tick);
1357 rcu_check_callbacks(cpu, user_tick);
1358 #ifdef CONFIG_IRQ_WORK
1363 run_posix_cpu_timers(p);
1367 * This function runs timers and the timer-tq in bottom half context.
1369 static void run_timer_softirq(struct softirq_action *h)
1371 struct tvec_base *base = __this_cpu_read(tvec_bases);
1373 hrtimer_run_pending();
1375 if (time_after_eq(jiffies, base->timer_jiffies))
1380 * Called by the local, per-CPU timer interrupt on SMP.
1382 void run_local_timers(void)
1384 hrtimer_run_queues();
1385 raise_softirq(TIMER_SOFTIRQ);
1388 #ifdef __ARCH_WANT_SYS_ALARM
1391 * For backwards compatibility? This can be done in libc so Alpha
1392 * and all newer ports shouldn't need it.
1394 SYSCALL_DEFINE1(alarm, unsigned int, seconds)
1396 return alarm_setitimer(seconds);
1401 static void process_timeout(unsigned long __data)
1403 wake_up_process((struct task_struct *)__data);
1407 * schedule_timeout - sleep until timeout
1408 * @timeout: timeout value in jiffies
1410 * Make the current task sleep until @timeout jiffies have
1411 * elapsed. The routine will return immediately unless
1412 * the current task state has been set (see set_current_state()).
1414 * You can set the task state as follows -
1416 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1417 * pass before the routine returns. The routine will return 0
1419 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1420 * delivered to the current task. In this case the remaining time
1421 * in jiffies will be returned, or 0 if the timer expired in time
1423 * The current task state is guaranteed to be TASK_RUNNING when this
1426 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1427 * the CPU away without a bound on the timeout. In this case the return
1428 * value will be %MAX_SCHEDULE_TIMEOUT.
1430 * In all cases the return value is guaranteed to be non-negative.
1432 signed long __sched schedule_timeout(signed long timeout)
1434 struct timer_list timer;
1435 unsigned long expire;
1439 case MAX_SCHEDULE_TIMEOUT:
1441 * These two special cases are useful to be comfortable
1442 * in the caller. Nothing more. We could take
1443 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1444 * but I' d like to return a valid offset (>=0) to allow
1445 * the caller to do everything it want with the retval.
1451 * Another bit of PARANOID. Note that the retval will be
1452 * 0 since no piece of kernel is supposed to do a check
1453 * for a negative retval of schedule_timeout() (since it
1454 * should never happens anyway). You just have the printk()
1455 * that will tell you if something is gone wrong and where.
1458 printk(KERN_ERR "schedule_timeout: wrong timeout "
1459 "value %lx\n", timeout);
1461 current->state = TASK_RUNNING;
1466 expire = timeout + jiffies;
1468 setup_timer_on_stack(&timer, process_timeout, (unsigned long)current);
1469 __mod_timer(&timer, expire, false, TIMER_NOT_PINNED);
1471 del_singleshot_timer_sync(&timer);
1473 /* Remove the timer from the object tracker */
1474 destroy_timer_on_stack(&timer);
1476 timeout = expire - jiffies;
1479 return timeout < 0 ? 0 : timeout;
1481 EXPORT_SYMBOL(schedule_timeout);
1484 * We can use __set_current_state() here because schedule_timeout() calls
1485 * schedule() unconditionally.
1487 signed long __sched schedule_timeout_interruptible(signed long timeout)
1489 __set_current_state(TASK_INTERRUPTIBLE);
1490 return schedule_timeout(timeout);
1492 EXPORT_SYMBOL(schedule_timeout_interruptible);
1494 signed long __sched schedule_timeout_killable(signed long timeout)
1496 __set_current_state(TASK_KILLABLE);
1497 return schedule_timeout(timeout);
1499 EXPORT_SYMBOL(schedule_timeout_killable);
1501 signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1503 __set_current_state(TASK_UNINTERRUPTIBLE);
1504 return schedule_timeout(timeout);
1506 EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1508 static int init_timers_cpu(int cpu)
1511 struct tvec_base *base;
1512 static char tvec_base_done[NR_CPUS];
1514 if (!tvec_base_done[cpu]) {
1515 static char boot_done;
1519 * The APs use this path later in boot
1521 base = kzalloc_node(sizeof(*base), GFP_KERNEL,
1526 /* Make sure that tvec_base is 2 byte aligned */
1527 if (tbase_get_deferrable(base)) {
1532 per_cpu(tvec_bases, cpu) = base;
1535 * This is for the boot CPU - we use compile-time
1536 * static initialisation because per-cpu memory isn't
1537 * ready yet and because the memory allocators are not
1538 * initialised either.
1541 base = &boot_tvec_bases;
1543 spin_lock_init(&base->lock);
1544 tvec_base_done[cpu] = 1;
1546 base = per_cpu(tvec_bases, cpu);
1550 for (j = 0; j < TVN_SIZE; j++) {
1551 INIT_LIST_HEAD(base->tv5.vec + j);
1552 INIT_LIST_HEAD(base->tv4.vec + j);
1553 INIT_LIST_HEAD(base->tv3.vec + j);
1554 INIT_LIST_HEAD(base->tv2.vec + j);
1556 for (j = 0; j < TVR_SIZE; j++)
1557 INIT_LIST_HEAD(base->tv1.vec + j);
1559 base->timer_jiffies = jiffies;
1560 base->next_timer = base->timer_jiffies;
1561 base->active_timers = 0;
1565 #ifdef CONFIG_HOTPLUG_CPU
1566 static void migrate_timer_list(struct tvec_base *new_base, struct list_head *head)
1568 struct timer_list *timer;
1570 while (!list_empty(head)) {
1571 timer = list_first_entry(head, struct timer_list, entry);
1572 /* We ignore the accounting on the dying cpu */
1573 detach_timer(timer, false);
1574 timer_set_base(timer, new_base);
1575 internal_add_timer(new_base, timer);
1579 static void migrate_timers(int cpu)
1581 struct tvec_base *old_base;
1582 struct tvec_base *new_base;
1585 BUG_ON(cpu_online(cpu));
1586 old_base = per_cpu(tvec_bases, cpu);
1587 new_base = get_cpu_var(tvec_bases);
1589 * The caller is globally serialized and nobody else
1590 * takes two locks at once, deadlock is not possible.
1592 spin_lock_irq(&new_base->lock);
1593 spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1595 BUG_ON(old_base->running_timer);
1597 for (i = 0; i < TVR_SIZE; i++)
1598 migrate_timer_list(new_base, old_base->tv1.vec + i);
1599 for (i = 0; i < TVN_SIZE; i++) {
1600 migrate_timer_list(new_base, old_base->tv2.vec + i);
1601 migrate_timer_list(new_base, old_base->tv3.vec + i);
1602 migrate_timer_list(new_base, old_base->tv4.vec + i);
1603 migrate_timer_list(new_base, old_base->tv5.vec + i);
1606 spin_unlock(&old_base->lock);
1607 spin_unlock_irq(&new_base->lock);
1608 put_cpu_var(tvec_bases);
1610 #endif /* CONFIG_HOTPLUG_CPU */
1612 static int timer_cpu_notify(struct notifier_block *self,
1613 unsigned long action, void *hcpu)
1615 long cpu = (long)hcpu;
1619 case CPU_UP_PREPARE:
1620 case CPU_UP_PREPARE_FROZEN:
1621 err = init_timers_cpu(cpu);
1623 return notifier_from_errno(err);
1625 #ifdef CONFIG_HOTPLUG_CPU
1627 case CPU_DEAD_FROZEN:
1628 migrate_timers(cpu);
1637 static struct notifier_block timers_nb = {
1638 .notifier_call = timer_cpu_notify,
1642 void __init init_timers(void)
1646 /* ensure there are enough low bits for flags in timer->base pointer */
1647 BUILD_BUG_ON(__alignof__(struct tvec_base) & TIMER_FLAG_MASK);
1649 err = timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE,
1650 (void *)(long)smp_processor_id());
1653 BUG_ON(err != NOTIFY_OK);
1654 register_cpu_notifier(&timers_nb);
1655 open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
1659 * msleep - sleep safely even with waitqueue interruptions
1660 * @msecs: Time in milliseconds to sleep for
1662 void msleep(unsigned int msecs)
1664 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1667 timeout = schedule_timeout_uninterruptible(timeout);
1670 EXPORT_SYMBOL(msleep);
1673 * msleep_interruptible - sleep waiting for signals
1674 * @msecs: Time in milliseconds to sleep for
1676 unsigned long msleep_interruptible(unsigned int msecs)
1678 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1680 while (timeout && !signal_pending(current))
1681 timeout = schedule_timeout_interruptible(timeout);
1682 return jiffies_to_msecs(timeout);
1685 EXPORT_SYMBOL(msleep_interruptible);
1687 static int __sched do_usleep_range(unsigned long min, unsigned long max)
1690 unsigned long delta;
1692 kmin = ktime_set(0, min * NSEC_PER_USEC);
1693 delta = (max - min) * NSEC_PER_USEC;
1694 return schedule_hrtimeout_range(&kmin, delta, HRTIMER_MODE_REL);
1698 * usleep_range - Drop in replacement for udelay where wakeup is flexible
1699 * @min: Minimum time in usecs to sleep
1700 * @max: Maximum time in usecs to sleep
1702 void usleep_range(unsigned long min, unsigned long max)
1704 __set_current_state(TASK_UNINTERRUPTIBLE);
1705 do_usleep_range(min, max);
1707 EXPORT_SYMBOL(usleep_range);