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 */
152 if (j <= jiffies) /* rounding ate our timeout entirely; */
158 * __round_jiffies - function to round jiffies to a full second
159 * @j: the time in (absolute) jiffies that should be rounded
160 * @cpu: the processor number on which the timeout will happen
162 * __round_jiffies() rounds an absolute time in the future (in jiffies)
163 * up or down to (approximately) full seconds. This is useful for timers
164 * for which the exact time they fire does not matter too much, as long as
165 * they fire approximately every X seconds.
167 * By rounding these timers to whole seconds, all such timers will fire
168 * at the same time, rather than at various times spread out. The goal
169 * of this is to have the CPU wake up less, which saves power.
171 * The exact rounding is skewed for each processor to avoid all
172 * processors firing at the exact same time, which could lead
173 * to lock contention or spurious cache line bouncing.
175 * The return value is the rounded version of the @j parameter.
177 unsigned long __round_jiffies(unsigned long j, int cpu)
179 return round_jiffies_common(j, cpu, false);
181 EXPORT_SYMBOL_GPL(__round_jiffies);
184 * __round_jiffies_relative - function to round jiffies to a full second
185 * @j: the time in (relative) jiffies that should be rounded
186 * @cpu: the processor number on which the timeout will happen
188 * __round_jiffies_relative() rounds a time delta in the future (in jiffies)
189 * up or down to (approximately) full seconds. This is useful for timers
190 * for which the exact time they fire does not matter too much, as long as
191 * they fire approximately every X seconds.
193 * By rounding these timers to whole seconds, all such timers will fire
194 * at the same time, rather than at various times spread out. The goal
195 * of this is to have the CPU wake up less, which saves power.
197 * The exact rounding is skewed for each processor to avoid all
198 * processors firing at the exact same time, which could lead
199 * to lock contention or spurious cache line bouncing.
201 * The return value is the rounded version of the @j parameter.
203 unsigned long __round_jiffies_relative(unsigned long j, int cpu)
205 unsigned long j0 = jiffies;
207 /* Use j0 because jiffies might change while we run */
208 return round_jiffies_common(j + j0, cpu, false) - j0;
210 EXPORT_SYMBOL_GPL(__round_jiffies_relative);
213 * round_jiffies - function to round jiffies to a full second
214 * @j: the time in (absolute) jiffies that should be rounded
216 * round_jiffies() rounds an absolute time in the future (in jiffies)
217 * up or down to (approximately) full seconds. This is useful for timers
218 * for which the exact time they fire does not matter too much, as long as
219 * they fire approximately every X seconds.
221 * By rounding these timers to whole seconds, all such timers will fire
222 * at the same time, rather than at various times spread out. The goal
223 * of this is to have the CPU wake up less, which saves power.
225 * The return value is the rounded version of the @j parameter.
227 unsigned long round_jiffies(unsigned long j)
229 return round_jiffies_common(j, raw_smp_processor_id(), false);
231 EXPORT_SYMBOL_GPL(round_jiffies);
234 * round_jiffies_relative - function to round jiffies to a full second
235 * @j: the time in (relative) jiffies that should be rounded
237 * round_jiffies_relative() rounds a time delta in the future (in jiffies)
238 * up or down to (approximately) full seconds. This is useful for timers
239 * for which the exact time they fire does not matter too much, as long as
240 * they fire approximately every X seconds.
242 * By rounding these timers to whole seconds, all such timers will fire
243 * at the same time, rather than at various times spread out. The goal
244 * of this is to have the CPU wake up less, which saves power.
246 * The return value is the rounded version of the @j parameter.
248 unsigned long round_jiffies_relative(unsigned long j)
250 return __round_jiffies_relative(j, raw_smp_processor_id());
252 EXPORT_SYMBOL_GPL(round_jiffies_relative);
255 * __round_jiffies_up - function to round jiffies up to a full second
256 * @j: the time in (absolute) jiffies that should be rounded
257 * @cpu: the processor number on which the timeout will happen
259 * This is the same as __round_jiffies() except that it will never
260 * round down. This is useful for timeouts for which the exact time
261 * of firing does not matter too much, as long as they don't fire too
264 unsigned long __round_jiffies_up(unsigned long j, int cpu)
266 return round_jiffies_common(j, cpu, true);
268 EXPORT_SYMBOL_GPL(__round_jiffies_up);
271 * __round_jiffies_up_relative - function to round jiffies up to a full second
272 * @j: the time in (relative) jiffies that should be rounded
273 * @cpu: the processor number on which the timeout will happen
275 * This is the same as __round_jiffies_relative() except that it will never
276 * round down. This is useful for timeouts for which the exact time
277 * of firing does not matter too much, as long as they don't fire too
280 unsigned long __round_jiffies_up_relative(unsigned long j, int cpu)
282 unsigned long j0 = jiffies;
284 /* Use j0 because jiffies might change while we run */
285 return round_jiffies_common(j + j0, cpu, true) - j0;
287 EXPORT_SYMBOL_GPL(__round_jiffies_up_relative);
290 * round_jiffies_up - function to round jiffies up to a full second
291 * @j: the time in (absolute) jiffies that should be rounded
293 * This is the same as round_jiffies() except that it will never
294 * round down. This is useful for timeouts for which the exact time
295 * of firing does not matter too much, as long as they don't fire too
298 unsigned long round_jiffies_up(unsigned long j)
300 return round_jiffies_common(j, raw_smp_processor_id(), true);
302 EXPORT_SYMBOL_GPL(round_jiffies_up);
305 * round_jiffies_up_relative - function to round jiffies up to a full second
306 * @j: the time in (relative) jiffies that should be rounded
308 * This is the same as round_jiffies_relative() except that it will never
309 * round down. This is useful for timeouts for which the exact time
310 * of firing does not matter too much, as long as they don't fire too
313 unsigned long round_jiffies_up_relative(unsigned long j)
315 return __round_jiffies_up_relative(j, raw_smp_processor_id());
317 EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
320 * set_timer_slack - set the allowed slack for a timer
321 * @timer: the timer to be modified
322 * @slack_hz: the amount of time (in jiffies) allowed for rounding
324 * Set the amount of time, in jiffies, that a certain timer has
325 * in terms of slack. By setting this value, the timer subsystem
326 * will schedule the actual timer somewhere between
327 * the time mod_timer() asks for, and that time plus the slack.
329 * By setting the slack to -1, a percentage of the delay is used
332 void set_timer_slack(struct timer_list *timer, int slack_hz)
334 timer->slack = slack_hz;
336 EXPORT_SYMBOL_GPL(set_timer_slack);
339 __internal_add_timer(struct tvec_base *base, struct timer_list *timer)
341 unsigned long expires = timer->expires;
342 unsigned long idx = expires - base->timer_jiffies;
343 struct list_head *vec;
345 if (idx < TVR_SIZE) {
346 int i = expires & TVR_MASK;
347 vec = base->tv1.vec + i;
348 } else if (idx < 1 << (TVR_BITS + TVN_BITS)) {
349 int i = (expires >> TVR_BITS) & TVN_MASK;
350 vec = base->tv2.vec + i;
351 } else if (idx < 1 << (TVR_BITS + 2 * TVN_BITS)) {
352 int i = (expires >> (TVR_BITS + TVN_BITS)) & TVN_MASK;
353 vec = base->tv3.vec + i;
354 } else if (idx < 1 << (TVR_BITS + 3 * TVN_BITS)) {
355 int i = (expires >> (TVR_BITS + 2 * TVN_BITS)) & TVN_MASK;
356 vec = base->tv4.vec + i;
357 } else if ((signed long) idx < 0) {
359 * Can happen if you add a timer with expires == jiffies,
360 * or you set a timer to go off in the past
362 vec = base->tv1.vec + (base->timer_jiffies & TVR_MASK);
365 /* If the timeout is larger than MAX_TVAL (on 64-bit
366 * architectures or with CONFIG_BASE_SMALL=1) then we
367 * use the maximum timeout.
369 if (idx > MAX_TVAL) {
371 expires = idx + base->timer_jiffies;
373 i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK;
374 vec = base->tv5.vec + i;
379 list_add_tail(&timer->entry, vec);
382 static void internal_add_timer(struct tvec_base *base, struct timer_list *timer)
384 __internal_add_timer(base, timer);
386 * Update base->active_timers and base->next_timer
388 if (!tbase_get_deferrable(timer->base)) {
389 if (time_before(timer->expires, base->next_timer))
390 base->next_timer = timer->expires;
391 base->active_timers++;
395 #ifdef CONFIG_TIMER_STATS
396 void __timer_stats_timer_set_start_info(struct timer_list *timer, void *addr)
398 if (timer->start_site)
401 timer->start_site = addr;
402 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
403 timer->start_pid = current->pid;
406 static void timer_stats_account_timer(struct timer_list *timer)
408 unsigned int flag = 0;
410 if (likely(!timer->start_site))
412 if (unlikely(tbase_get_deferrable(timer->base)))
413 flag |= TIMER_STATS_FLAG_DEFERRABLE;
415 timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
416 timer->function, timer->start_comm, flag);
420 static void timer_stats_account_timer(struct timer_list *timer) {}
423 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
425 static struct debug_obj_descr timer_debug_descr;
427 static void *timer_debug_hint(void *addr)
429 return ((struct timer_list *) addr)->function;
433 * fixup_init is called when:
434 * - an active object is initialized
436 static int timer_fixup_init(void *addr, enum debug_obj_state state)
438 struct timer_list *timer = addr;
441 case ODEBUG_STATE_ACTIVE:
442 del_timer_sync(timer);
443 debug_object_init(timer, &timer_debug_descr);
450 /* Stub timer callback for improperly used timers. */
451 static void stub_timer(unsigned long data)
457 * fixup_activate is called when:
458 * - an active object is activated
459 * - an unknown object is activated (might be a statically initialized object)
461 static int timer_fixup_activate(void *addr, enum debug_obj_state state)
463 struct timer_list *timer = addr;
467 case ODEBUG_STATE_NOTAVAILABLE:
469 * This is not really a fixup. The timer was
470 * statically initialized. We just make sure that it
471 * is tracked in the object tracker.
473 if (timer->entry.next == NULL &&
474 timer->entry.prev == TIMER_ENTRY_STATIC) {
475 debug_object_init(timer, &timer_debug_descr);
476 debug_object_activate(timer, &timer_debug_descr);
479 setup_timer(timer, stub_timer, 0);
484 case ODEBUG_STATE_ACTIVE:
493 * fixup_free is called when:
494 * - an active object is freed
496 static int timer_fixup_free(void *addr, enum debug_obj_state state)
498 struct timer_list *timer = addr;
501 case ODEBUG_STATE_ACTIVE:
502 del_timer_sync(timer);
503 debug_object_free(timer, &timer_debug_descr);
511 * fixup_assert_init is called when:
512 * - an untracked/uninit-ed object is found
514 static int timer_fixup_assert_init(void *addr, enum debug_obj_state state)
516 struct timer_list *timer = addr;
519 case ODEBUG_STATE_NOTAVAILABLE:
520 if (timer->entry.prev == TIMER_ENTRY_STATIC) {
522 * This is not really a fixup. The timer was
523 * statically initialized. We just make sure that it
524 * is tracked in the object tracker.
526 debug_object_init(timer, &timer_debug_descr);
529 setup_timer(timer, stub_timer, 0);
537 static struct debug_obj_descr timer_debug_descr = {
538 .name = "timer_list",
539 .debug_hint = timer_debug_hint,
540 .fixup_init = timer_fixup_init,
541 .fixup_activate = timer_fixup_activate,
542 .fixup_free = timer_fixup_free,
543 .fixup_assert_init = timer_fixup_assert_init,
546 static inline void debug_timer_init(struct timer_list *timer)
548 debug_object_init(timer, &timer_debug_descr);
551 static inline void debug_timer_activate(struct timer_list *timer)
553 debug_object_activate(timer, &timer_debug_descr);
556 static inline void debug_timer_deactivate(struct timer_list *timer)
558 debug_object_deactivate(timer, &timer_debug_descr);
561 static inline void debug_timer_free(struct timer_list *timer)
563 debug_object_free(timer, &timer_debug_descr);
566 static inline void debug_timer_assert_init(struct timer_list *timer)
568 debug_object_assert_init(timer, &timer_debug_descr);
571 static void do_init_timer(struct timer_list *timer, unsigned int flags,
572 const char *name, struct lock_class_key *key);
574 void init_timer_on_stack_key(struct timer_list *timer, unsigned int flags,
575 const char *name, struct lock_class_key *key)
577 debug_object_init_on_stack(timer, &timer_debug_descr);
578 do_init_timer(timer, flags, name, key);
580 EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
582 void destroy_timer_on_stack(struct timer_list *timer)
584 debug_object_free(timer, &timer_debug_descr);
586 EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
589 static inline void debug_timer_init(struct timer_list *timer) { }
590 static inline void debug_timer_activate(struct timer_list *timer) { }
591 static inline void debug_timer_deactivate(struct timer_list *timer) { }
592 static inline void debug_timer_assert_init(struct timer_list *timer) { }
595 static inline void debug_init(struct timer_list *timer)
597 debug_timer_init(timer);
598 trace_timer_init(timer);
602 debug_activate(struct timer_list *timer, unsigned long expires)
604 debug_timer_activate(timer);
605 trace_timer_start(timer, expires);
608 static inline void debug_deactivate(struct timer_list *timer)
610 debug_timer_deactivate(timer);
611 trace_timer_cancel(timer);
614 static inline void debug_assert_init(struct timer_list *timer)
616 debug_timer_assert_init(timer);
619 static void do_init_timer(struct timer_list *timer, unsigned int flags,
620 const char *name, struct lock_class_key *key)
622 struct tvec_base *base = __raw_get_cpu_var(tvec_bases);
624 timer->entry.next = NULL;
625 timer->base = (void *)((unsigned long)base | flags);
627 #ifdef CONFIG_TIMER_STATS
628 timer->start_site = NULL;
629 timer->start_pid = -1;
630 memset(timer->start_comm, 0, TASK_COMM_LEN);
632 lockdep_init_map(&timer->lockdep_map, name, key, 0);
636 * init_timer_key - initialize a timer
637 * @timer: the timer to be initialized
638 * @flags: timer flags
639 * @name: name of the timer
640 * @key: lockdep class key of the fake lock used for tracking timer
641 * sync lock dependencies
643 * init_timer_key() must be done to a timer prior calling *any* of the
644 * other timer functions.
646 void init_timer_key(struct timer_list *timer, unsigned int flags,
647 const char *name, struct lock_class_key *key)
650 do_init_timer(timer, flags, name, key);
652 EXPORT_SYMBOL(init_timer_key);
654 static inline void detach_timer(struct timer_list *timer, bool clear_pending)
656 struct list_head *entry = &timer->entry;
658 debug_deactivate(timer);
660 __list_del(entry->prev, entry->next);
663 entry->prev = LIST_POISON2;
667 detach_expired_timer(struct timer_list *timer, struct tvec_base *base)
669 detach_timer(timer, true);
670 if (!tbase_get_deferrable(timer->base))
671 base->active_timers--;
674 static int detach_if_pending(struct timer_list *timer, struct tvec_base *base,
677 if (!timer_pending(timer))
680 detach_timer(timer, clear_pending);
681 if (!tbase_get_deferrable(timer->base)) {
682 base->active_timers--;
683 if (timer->expires == base->next_timer)
684 base->next_timer = base->timer_jiffies;
690 * We are using hashed locking: holding per_cpu(tvec_bases).lock
691 * means that all timers which are tied to this base via timer->base are
692 * locked, and the base itself is locked too.
694 * So __run_timers/migrate_timers can safely modify all timers which could
695 * be found on ->tvX lists.
697 * When the timer's base is locked, and the timer removed from list, it is
698 * possible to set timer->base = NULL and drop the lock: the timer remains
701 static struct tvec_base *lock_timer_base(struct timer_list *timer,
702 unsigned long *flags)
703 __acquires(timer->base->lock)
705 struct tvec_base *base;
708 struct tvec_base *prelock_base = timer->base;
709 base = tbase_get_base(prelock_base);
710 if (likely(base != NULL)) {
711 spin_lock_irqsave(&base->lock, *flags);
712 if (likely(prelock_base == timer->base))
714 /* The timer has migrated to another CPU */
715 spin_unlock_irqrestore(&base->lock, *flags);
722 __mod_timer(struct timer_list *timer, unsigned long expires,
723 bool pending_only, int pinned)
725 struct tvec_base *base, *new_base;
729 timer_stats_timer_set_start_info(timer);
730 BUG_ON(!timer->function);
732 base = lock_timer_base(timer, &flags);
734 ret = detach_if_pending(timer, base, false);
735 if (!ret && pending_only)
738 debug_activate(timer, expires);
740 cpu = smp_processor_id();
742 #if defined(CONFIG_NO_HZ) && defined(CONFIG_SMP)
743 if (!pinned && get_sysctl_timer_migration() && idle_cpu(cpu))
744 cpu = get_nohz_timer_target();
746 new_base = per_cpu(tvec_bases, cpu);
748 if (base != new_base) {
750 * We are trying to schedule the timer on the local CPU.
751 * However we can't change timer's base while it is running,
752 * otherwise del_timer_sync() can't detect that the timer's
753 * handler yet has not finished. This also guarantees that
754 * the timer is serialized wrt itself.
756 if (likely(base->running_timer != timer)) {
757 /* See the comment in lock_timer_base() */
758 timer_set_base(timer, NULL);
759 spin_unlock(&base->lock);
761 spin_lock(&base->lock);
762 timer_set_base(timer, base);
766 timer->expires = expires;
767 internal_add_timer(base, timer);
770 spin_unlock_irqrestore(&base->lock, flags);
776 * mod_timer_pending - modify a pending timer's timeout
777 * @timer: the pending timer to be modified
778 * @expires: new timeout in jiffies
780 * mod_timer_pending() is the same for pending timers as mod_timer(),
781 * but will not re-activate and modify already deleted timers.
783 * It is useful for unserialized use of timers.
785 int mod_timer_pending(struct timer_list *timer, unsigned long expires)
787 return __mod_timer(timer, expires, true, TIMER_NOT_PINNED);
789 EXPORT_SYMBOL(mod_timer_pending);
792 * Decide where to put the timer while taking the slack into account
795 * 1) calculate the maximum (absolute) time
796 * 2) calculate the highest bit where the expires and new max are different
797 * 3) use this bit to make a mask
798 * 4) use the bitmask to round down the maximum time, so that all last
802 unsigned long apply_slack(struct timer_list *timer, unsigned long expires)
804 unsigned long expires_limit, mask;
807 if (timer->slack >= 0) {
808 expires_limit = expires + timer->slack;
810 long delta = expires - jiffies;
815 expires_limit = expires + delta / 256;
817 mask = expires ^ expires_limit;
821 bit = find_last_bit(&mask, BITS_PER_LONG);
823 mask = (1 << bit) - 1;
825 expires_limit = expires_limit & ~(mask);
827 return expires_limit;
831 * mod_timer - modify a timer's timeout
832 * @timer: the timer to be modified
833 * @expires: new timeout in jiffies
835 * mod_timer() is a more efficient way to update the expire field of an
836 * active timer (if the timer is inactive it will be activated)
838 * mod_timer(timer, expires) is equivalent to:
840 * del_timer(timer); timer->expires = expires; add_timer(timer);
842 * Note that if there are multiple unserialized concurrent users of the
843 * same timer, then mod_timer() is the only safe way to modify the timeout,
844 * since add_timer() cannot modify an already running timer.
846 * The function returns whether it has modified a pending timer or not.
847 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
848 * active timer returns 1.)
850 int mod_timer(struct timer_list *timer, unsigned long expires)
852 expires = apply_slack(timer, expires);
855 * This is a common optimization triggered by the
856 * networking code - if the timer is re-modified
857 * to be the same thing then just return:
859 if (timer_pending(timer) && timer->expires == expires)
862 return __mod_timer(timer, expires, false, TIMER_NOT_PINNED);
864 EXPORT_SYMBOL(mod_timer);
867 * mod_timer_pinned - modify a timer's timeout
868 * @timer: the timer to be modified
869 * @expires: new timeout in jiffies
871 * mod_timer_pinned() is a way to update the expire field of an
872 * active timer (if the timer is inactive it will be activated)
873 * and to ensure that the timer is scheduled on the current CPU.
875 * Note that this does not prevent the timer from being migrated
876 * when the current CPU goes offline. If this is a problem for
877 * you, use CPU-hotplug notifiers to handle it correctly, for
878 * example, cancelling the timer when the corresponding CPU goes
881 * mod_timer_pinned(timer, expires) is equivalent to:
883 * del_timer(timer); timer->expires = expires; add_timer(timer);
885 int mod_timer_pinned(struct timer_list *timer, unsigned long expires)
887 if (timer->expires == expires && timer_pending(timer))
890 return __mod_timer(timer, expires, false, TIMER_PINNED);
892 EXPORT_SYMBOL(mod_timer_pinned);
895 * add_timer - start a timer
896 * @timer: the timer to be added
898 * The kernel will do a ->function(->data) callback from the
899 * timer interrupt at the ->expires point in the future. The
900 * current time is 'jiffies'.
902 * The timer's ->expires, ->function (and if the handler uses it, ->data)
903 * fields must be set prior calling this function.
905 * Timers with an ->expires field in the past will be executed in the next
908 void add_timer(struct timer_list *timer)
910 BUG_ON(timer_pending(timer));
911 mod_timer(timer, timer->expires);
913 EXPORT_SYMBOL(add_timer);
916 * add_timer_on - start a timer on a particular CPU
917 * @timer: the timer to be added
918 * @cpu: the CPU to start it on
920 * This is not very scalable on SMP. Double adds are not possible.
922 void add_timer_on(struct timer_list *timer, int cpu)
924 struct tvec_base *base = per_cpu(tvec_bases, cpu);
927 timer_stats_timer_set_start_info(timer);
928 BUG_ON(timer_pending(timer) || !timer->function);
929 spin_lock_irqsave(&base->lock, flags);
930 timer_set_base(timer, base);
931 debug_activate(timer, timer->expires);
932 internal_add_timer(base, timer);
934 * Check whether the other CPU is idle and needs to be
935 * triggered to reevaluate the timer wheel when nohz is
936 * active. We are protected against the other CPU fiddling
937 * with the timer by holding the timer base lock. This also
938 * makes sure that a CPU on the way to idle can not evaluate
941 wake_up_idle_cpu(cpu);
942 spin_unlock_irqrestore(&base->lock, flags);
944 EXPORT_SYMBOL_GPL(add_timer_on);
947 * del_timer - deactive a timer.
948 * @timer: the timer to be deactivated
950 * del_timer() deactivates a timer - this works on both active and inactive
953 * The function returns whether it has deactivated a pending timer or not.
954 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
955 * active timer returns 1.)
957 int del_timer(struct timer_list *timer)
959 struct tvec_base *base;
963 debug_assert_init(timer);
965 timer_stats_timer_clear_start_info(timer);
966 if (timer_pending(timer)) {
967 base = lock_timer_base(timer, &flags);
968 ret = detach_if_pending(timer, base, true);
969 spin_unlock_irqrestore(&base->lock, flags);
974 EXPORT_SYMBOL(del_timer);
977 * try_to_del_timer_sync - Try to deactivate a timer
978 * @timer: timer do del
980 * This function tries to deactivate a timer. Upon successful (ret >= 0)
981 * exit the timer is not queued and the handler is not running on any CPU.
983 int try_to_del_timer_sync(struct timer_list *timer)
985 struct tvec_base *base;
989 debug_assert_init(timer);
991 base = lock_timer_base(timer, &flags);
993 if (base->running_timer != timer) {
994 timer_stats_timer_clear_start_info(timer);
995 ret = detach_if_pending(timer, base, true);
997 spin_unlock_irqrestore(&base->lock, flags);
1001 EXPORT_SYMBOL(try_to_del_timer_sync);
1005 * del_timer_sync - deactivate a timer and wait for the handler to finish.
1006 * @timer: the timer to be deactivated
1008 * This function only differs from del_timer() on SMP: besides deactivating
1009 * the timer it also makes sure the handler has finished executing on other
1012 * Synchronization rules: Callers must prevent restarting of the timer,
1013 * otherwise this function is meaningless. It must not be called from
1014 * interrupt contexts unless the timer is an irqsafe one. The caller must
1015 * not hold locks which would prevent completion of the timer's
1016 * handler. The timer's handler must not call add_timer_on(). Upon exit the
1017 * timer is not queued and the handler is not running on any CPU.
1019 * Note: For !irqsafe timers, you must not hold locks that are held in
1020 * interrupt context while calling this function. Even if the lock has
1021 * nothing to do with the timer in question. Here's why:
1027 * base->running_timer = mytimer;
1028 * spin_lock_irq(somelock);
1030 * spin_lock(somelock);
1031 * del_timer_sync(mytimer);
1032 * while (base->running_timer == mytimer);
1034 * Now del_timer_sync() will never return and never release somelock.
1035 * The interrupt on the other CPU is waiting to grab somelock but
1036 * it has interrupted the softirq that CPU0 is waiting to finish.
1038 * The function returns whether it has deactivated a pending timer or not.
1040 int del_timer_sync(struct timer_list *timer)
1042 #ifdef CONFIG_LOCKDEP
1043 unsigned long flags;
1046 * If lockdep gives a backtrace here, please reference
1047 * the synchronization rules above.
1049 local_irq_save(flags);
1050 lock_map_acquire(&timer->lockdep_map);
1051 lock_map_release(&timer->lockdep_map);
1052 local_irq_restore(flags);
1055 * don't use it in hardirq context, because it
1056 * could lead to deadlock.
1058 WARN_ON(in_irq() && !tbase_get_irqsafe(timer->base));
1060 int ret = try_to_del_timer_sync(timer);
1066 EXPORT_SYMBOL(del_timer_sync);
1069 static int cascade(struct tvec_base *base, struct tvec *tv, int index)
1071 /* cascade all the timers from tv up one level */
1072 struct timer_list *timer, *tmp;
1073 struct list_head tv_list;
1075 list_replace_init(tv->vec + index, &tv_list);
1078 * We are removing _all_ timers from the list, so we
1079 * don't have to detach them individually.
1081 list_for_each_entry_safe(timer, tmp, &tv_list, entry) {
1082 BUG_ON(tbase_get_base(timer->base) != base);
1083 /* No accounting, while moving them */
1084 __internal_add_timer(base, timer);
1090 static void call_timer_fn(struct timer_list *timer, void (*fn)(unsigned long),
1093 int preempt_count = preempt_count();
1095 #ifdef CONFIG_LOCKDEP
1097 * It is permissible to free the timer from inside the
1098 * function that is called from it, this we need to take into
1099 * account for lockdep too. To avoid bogus "held lock freed"
1100 * warnings as well as problems when looking into
1101 * timer->lockdep_map, make a copy and use that here.
1103 struct lockdep_map lockdep_map;
1105 lockdep_copy_map(&lockdep_map, &timer->lockdep_map);
1108 * Couple the lock chain with the lock chain at
1109 * del_timer_sync() by acquiring the lock_map around the fn()
1110 * call here and in del_timer_sync().
1112 lock_map_acquire(&lockdep_map);
1114 trace_timer_expire_entry(timer);
1116 trace_timer_expire_exit(timer);
1118 lock_map_release(&lockdep_map);
1120 if (preempt_count != preempt_count()) {
1121 WARN_ONCE(1, "timer: %pF preempt leak: %08x -> %08x\n",
1122 fn, preempt_count, preempt_count());
1124 * Restore the preempt count. That gives us a decent
1125 * chance to survive and extract information. If the
1126 * callback kept a lock held, bad luck, but not worse
1127 * than the BUG() we had.
1129 preempt_count() = preempt_count;
1133 #define INDEX(N) ((base->timer_jiffies >> (TVR_BITS + (N) * TVN_BITS)) & TVN_MASK)
1136 * __run_timers - run all expired timers (if any) on this CPU.
1137 * @base: the timer vector to be processed.
1139 * This function cascades all vectors and executes all expired timer
1142 static inline void __run_timers(struct tvec_base *base)
1144 struct timer_list *timer;
1146 spin_lock_irq(&base->lock);
1147 while (time_after_eq(jiffies, base->timer_jiffies)) {
1148 struct list_head work_list;
1149 struct list_head *head = &work_list;
1150 int index = base->timer_jiffies & TVR_MASK;
1156 (!cascade(base, &base->tv2, INDEX(0))) &&
1157 (!cascade(base, &base->tv3, INDEX(1))) &&
1158 !cascade(base, &base->tv4, INDEX(2)))
1159 cascade(base, &base->tv5, INDEX(3));
1160 ++base->timer_jiffies;
1161 list_replace_init(base->tv1.vec + index, &work_list);
1162 while (!list_empty(head)) {
1163 void (*fn)(unsigned long);
1167 timer = list_first_entry(head, struct timer_list,entry);
1168 fn = timer->function;
1170 irqsafe = tbase_get_irqsafe(timer->base);
1172 timer_stats_account_timer(timer);
1174 base->running_timer = timer;
1175 detach_expired_timer(timer, base);
1178 spin_unlock(&base->lock);
1179 call_timer_fn(timer, fn, data);
1180 spin_lock(&base->lock);
1182 spin_unlock_irq(&base->lock);
1183 call_timer_fn(timer, fn, data);
1184 spin_lock_irq(&base->lock);
1188 base->running_timer = NULL;
1189 spin_unlock_irq(&base->lock);
1194 * Find out when the next timer event is due to happen. This
1195 * is used on S/390 to stop all activity when a CPU is idle.
1196 * This function needs to be called with interrupts disabled.
1198 static unsigned long __next_timer_interrupt(struct tvec_base *base)
1200 unsigned long timer_jiffies = base->timer_jiffies;
1201 unsigned long expires = timer_jiffies + NEXT_TIMER_MAX_DELTA;
1202 int index, slot, array, found = 0;
1203 struct timer_list *nte;
1204 struct tvec *varray[4];
1206 /* Look for timer events in tv1. */
1207 index = slot = timer_jiffies & TVR_MASK;
1209 list_for_each_entry(nte, base->tv1.vec + slot, entry) {
1210 if (tbase_get_deferrable(nte->base))
1214 expires = nte->expires;
1215 /* Look at the cascade bucket(s)? */
1216 if (!index || slot < index)
1220 slot = (slot + 1) & TVR_MASK;
1221 } while (slot != index);
1224 /* Calculate the next cascade event */
1226 timer_jiffies += TVR_SIZE - index;
1227 timer_jiffies >>= TVR_BITS;
1229 /* Check tv2-tv5. */
1230 varray[0] = &base->tv2;
1231 varray[1] = &base->tv3;
1232 varray[2] = &base->tv4;
1233 varray[3] = &base->tv5;
1235 for (array = 0; array < 4; array++) {
1236 struct tvec *varp = varray[array];
1238 index = slot = timer_jiffies & TVN_MASK;
1240 list_for_each_entry(nte, varp->vec + slot, entry) {
1241 if (tbase_get_deferrable(nte->base))
1245 if (time_before(nte->expires, expires))
1246 expires = nte->expires;
1249 * Do we still search for the first timer or are
1250 * we looking up the cascade buckets ?
1253 /* Look at the cascade bucket(s)? */
1254 if (!index || slot < index)
1258 slot = (slot + 1) & TVN_MASK;
1259 } while (slot != index);
1262 timer_jiffies += TVN_SIZE - index;
1263 timer_jiffies >>= TVN_BITS;
1269 * Check, if the next hrtimer event is before the next timer wheel
1272 static unsigned long cmp_next_hrtimer_event(unsigned long now,
1273 unsigned long expires)
1275 ktime_t hr_delta = hrtimer_get_next_event();
1276 struct timespec tsdelta;
1277 unsigned long delta;
1279 if (hr_delta.tv64 == KTIME_MAX)
1283 * Expired timer available, let it expire in the next tick
1285 if (hr_delta.tv64 <= 0)
1288 tsdelta = ktime_to_timespec(hr_delta);
1289 delta = timespec_to_jiffies(&tsdelta);
1292 * Limit the delta to the max value, which is checked in
1293 * tick_nohz_stop_sched_tick():
1295 if (delta > NEXT_TIMER_MAX_DELTA)
1296 delta = NEXT_TIMER_MAX_DELTA;
1299 * Take rounding errors in to account and make sure, that it
1300 * expires in the next tick. Otherwise we go into an endless
1301 * ping pong due to tick_nohz_stop_sched_tick() retriggering
1307 if (time_before(now, expires))
1313 * get_next_timer_interrupt - return the jiffy of the next pending timer
1314 * @now: current time (in jiffies)
1316 unsigned long get_next_timer_interrupt(unsigned long now)
1318 struct tvec_base *base = __this_cpu_read(tvec_bases);
1319 unsigned long expires = now + NEXT_TIMER_MAX_DELTA;
1322 * Pretend that there is no timer pending if the cpu is offline.
1323 * Possible pending timers will be migrated later to an active cpu.
1325 if (cpu_is_offline(smp_processor_id()))
1328 spin_lock(&base->lock);
1329 if (base->active_timers) {
1330 if (time_before_eq(base->next_timer, base->timer_jiffies))
1331 base->next_timer = __next_timer_interrupt(base);
1332 expires = base->next_timer;
1334 spin_unlock(&base->lock);
1336 if (time_before_eq(expires, now))
1339 return cmp_next_hrtimer_event(now, expires);
1344 * Called from the timer interrupt handler to charge one tick to the current
1345 * process. user_tick is 1 if the tick is user time, 0 for system.
1347 void update_process_times(int user_tick)
1349 struct task_struct *p = current;
1350 int cpu = smp_processor_id();
1352 /* Note: this timer irq context must be accounted for as well. */
1353 account_process_tick(p, user_tick);
1355 rcu_check_callbacks(cpu, user_tick);
1356 #ifdef CONFIG_IRQ_WORK
1361 run_posix_cpu_timers(p);
1365 * This function runs timers and the timer-tq in bottom half context.
1367 static void run_timer_softirq(struct softirq_action *h)
1369 struct tvec_base *base = __this_cpu_read(tvec_bases);
1371 hrtimer_run_pending();
1373 if (time_after_eq(jiffies, base->timer_jiffies))
1378 * Called by the local, per-CPU timer interrupt on SMP.
1380 void run_local_timers(void)
1382 hrtimer_run_queues();
1383 raise_softirq(TIMER_SOFTIRQ);
1386 #ifdef __ARCH_WANT_SYS_ALARM
1389 * For backwards compatibility? This can be done in libc so Alpha
1390 * and all newer ports shouldn't need it.
1392 SYSCALL_DEFINE1(alarm, unsigned int, seconds)
1394 return alarm_setitimer(seconds);
1399 static void process_timeout(unsigned long __data)
1401 wake_up_process((struct task_struct *)__data);
1405 * schedule_timeout - sleep until timeout
1406 * @timeout: timeout value in jiffies
1408 * Make the current task sleep until @timeout jiffies have
1409 * elapsed. The routine will return immediately unless
1410 * the current task state has been set (see set_current_state()).
1412 * You can set the task state as follows -
1414 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1415 * pass before the routine returns. The routine will return 0
1417 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1418 * delivered to the current task. In this case the remaining time
1419 * in jiffies will be returned, or 0 if the timer expired in time
1421 * The current task state is guaranteed to be TASK_RUNNING when this
1424 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1425 * the CPU away without a bound on the timeout. In this case the return
1426 * value will be %MAX_SCHEDULE_TIMEOUT.
1428 * In all cases the return value is guaranteed to be non-negative.
1430 signed long __sched schedule_timeout(signed long timeout)
1432 struct timer_list timer;
1433 unsigned long expire;
1437 case MAX_SCHEDULE_TIMEOUT:
1439 * These two special cases are useful to be comfortable
1440 * in the caller. Nothing more. We could take
1441 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1442 * but I' d like to return a valid offset (>=0) to allow
1443 * the caller to do everything it want with the retval.
1449 * Another bit of PARANOID. Note that the retval will be
1450 * 0 since no piece of kernel is supposed to do a check
1451 * for a negative retval of schedule_timeout() (since it
1452 * should never happens anyway). You just have the printk()
1453 * that will tell you if something is gone wrong and where.
1456 printk(KERN_ERR "schedule_timeout: wrong timeout "
1457 "value %lx\n", timeout);
1459 current->state = TASK_RUNNING;
1464 expire = timeout + jiffies;
1466 setup_timer_on_stack(&timer, process_timeout, (unsigned long)current);
1467 __mod_timer(&timer, expire, false, TIMER_NOT_PINNED);
1469 del_singleshot_timer_sync(&timer);
1471 /* Remove the timer from the object tracker */
1472 destroy_timer_on_stack(&timer);
1474 timeout = expire - jiffies;
1477 return timeout < 0 ? 0 : timeout;
1479 EXPORT_SYMBOL(schedule_timeout);
1482 * We can use __set_current_state() here because schedule_timeout() calls
1483 * schedule() unconditionally.
1485 signed long __sched schedule_timeout_interruptible(signed long timeout)
1487 __set_current_state(TASK_INTERRUPTIBLE);
1488 return schedule_timeout(timeout);
1490 EXPORT_SYMBOL(schedule_timeout_interruptible);
1492 signed long __sched schedule_timeout_killable(signed long timeout)
1494 __set_current_state(TASK_KILLABLE);
1495 return schedule_timeout(timeout);
1497 EXPORT_SYMBOL(schedule_timeout_killable);
1499 signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1501 __set_current_state(TASK_UNINTERRUPTIBLE);
1502 return schedule_timeout(timeout);
1504 EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1506 static int __cpuinit init_timers_cpu(int cpu)
1509 struct tvec_base *base;
1510 static char __cpuinitdata tvec_base_done[NR_CPUS];
1512 if (!tvec_base_done[cpu]) {
1513 static char boot_done;
1517 * The APs use this path later in boot
1519 base = kmalloc_node(sizeof(*base),
1520 GFP_KERNEL | __GFP_ZERO,
1525 /* Make sure that tvec_base is 2 byte aligned */
1526 if (tbase_get_deferrable(base)) {
1531 per_cpu(tvec_bases, cpu) = base;
1534 * This is for the boot CPU - we use compile-time
1535 * static initialisation because per-cpu memory isn't
1536 * ready yet and because the memory allocators are not
1537 * initialised either.
1540 base = &boot_tvec_bases;
1542 tvec_base_done[cpu] = 1;
1544 base = per_cpu(tvec_bases, cpu);
1547 spin_lock_init(&base->lock);
1549 for (j = 0; j < TVN_SIZE; j++) {
1550 INIT_LIST_HEAD(base->tv5.vec + j);
1551 INIT_LIST_HEAD(base->tv4.vec + j);
1552 INIT_LIST_HEAD(base->tv3.vec + j);
1553 INIT_LIST_HEAD(base->tv2.vec + j);
1555 for (j = 0; j < TVR_SIZE; j++)
1556 INIT_LIST_HEAD(base->tv1.vec + j);
1558 base->timer_jiffies = jiffies;
1559 base->next_timer = base->timer_jiffies;
1560 base->active_timers = 0;
1564 #ifdef CONFIG_HOTPLUG_CPU
1565 static void migrate_timer_list(struct tvec_base *new_base, struct list_head *head)
1567 struct timer_list *timer;
1569 while (!list_empty(head)) {
1570 timer = list_first_entry(head, struct timer_list, entry);
1571 /* We ignore the accounting on the dying cpu */
1572 detach_timer(timer, false);
1573 timer_set_base(timer, new_base);
1574 internal_add_timer(new_base, timer);
1578 static void __cpuinit migrate_timers(int cpu)
1580 struct tvec_base *old_base;
1581 struct tvec_base *new_base;
1584 BUG_ON(cpu_online(cpu));
1585 old_base = per_cpu(tvec_bases, cpu);
1586 new_base = get_cpu_var(tvec_bases);
1588 * The caller is globally serialized and nobody else
1589 * takes two locks at once, deadlock is not possible.
1591 spin_lock_irq(&new_base->lock);
1592 spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1594 BUG_ON(old_base->running_timer);
1596 for (i = 0; i < TVR_SIZE; i++)
1597 migrate_timer_list(new_base, old_base->tv1.vec + i);
1598 for (i = 0; i < TVN_SIZE; i++) {
1599 migrate_timer_list(new_base, old_base->tv2.vec + i);
1600 migrate_timer_list(new_base, old_base->tv3.vec + i);
1601 migrate_timer_list(new_base, old_base->tv4.vec + i);
1602 migrate_timer_list(new_base, old_base->tv5.vec + i);
1605 spin_unlock(&old_base->lock);
1606 spin_unlock_irq(&new_base->lock);
1607 put_cpu_var(tvec_bases);
1609 #endif /* CONFIG_HOTPLUG_CPU */
1611 static int __cpuinit timer_cpu_notify(struct notifier_block *self,
1612 unsigned long action, void *hcpu)
1614 long cpu = (long)hcpu;
1618 case CPU_UP_PREPARE:
1619 case CPU_UP_PREPARE_FROZEN:
1620 err = init_timers_cpu(cpu);
1622 return notifier_from_errno(err);
1624 #ifdef CONFIG_HOTPLUG_CPU
1626 case CPU_DEAD_FROZEN:
1627 migrate_timers(cpu);
1636 static struct notifier_block __cpuinitdata timers_nb = {
1637 .notifier_call = timer_cpu_notify,
1641 void __init init_timers(void)
1645 /* ensure there are enough low bits for flags in timer->base pointer */
1646 BUILD_BUG_ON(__alignof__(struct tvec_base) & TIMER_FLAG_MASK);
1648 err = timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE,
1649 (void *)(long)smp_processor_id());
1652 BUG_ON(err != NOTIFY_OK);
1653 register_cpu_notifier(&timers_nb);
1654 open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
1658 * msleep - sleep safely even with waitqueue interruptions
1659 * @msecs: Time in milliseconds to sleep for
1661 void msleep(unsigned int msecs)
1663 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1666 timeout = schedule_timeout_uninterruptible(timeout);
1669 EXPORT_SYMBOL(msleep);
1672 * msleep_interruptible - sleep waiting for signals
1673 * @msecs: Time in milliseconds to sleep for
1675 unsigned long msleep_interruptible(unsigned int msecs)
1677 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1679 while (timeout && !signal_pending(current))
1680 timeout = schedule_timeout_interruptible(timeout);
1681 return jiffies_to_msecs(timeout);
1684 EXPORT_SYMBOL(msleep_interruptible);
1686 static int __sched do_usleep_range(unsigned long min, unsigned long max)
1689 unsigned long delta;
1691 kmin = ktime_set(0, min * NSEC_PER_USEC);
1692 delta = (max - min) * NSEC_PER_USEC;
1693 return schedule_hrtimeout_range(&kmin, delta, HRTIMER_MODE_REL);
1697 * usleep_range - Drop in replacement for udelay where wakeup is flexible
1698 * @min: Minimum time in usecs to sleep
1699 * @max: Maximum time in usecs to sleep
1701 void usleep_range(unsigned long min, unsigned long max)
1703 __set_current_state(TASK_UNINTERRUPTIBLE);
1704 do_usleep_range(min, max);
1706 EXPORT_SYMBOL(usleep_range);