16  *  2000-10-05  Implemented scalable SMP per-CPU timer handling.
58 #include <trace/events/timer.h>
65  * The timer wheel has LVL_DEPTH array levels. Each level provides an array of
72  * The array level of a newly armed timer depends on the relative expiry
76  * Contrary to the original timer wheel implementation, which aims for 'exact'
78  * the timers into the lower array levels. The previous 'classic' timer wheel
83  * This is an optimization of the original timer wheel implementation for the
84  * majority of the timer wheel use cases: timeouts. The vast majority of
318 	 * due to delays of the timer irq, long irq off times etc etc) then  in round_jiffies_common()
501 static inline unsigned int timer_get_idx(struct timer_list *timer)  in timer_get_idx()  argument
503 	return (timer->flags & TIMER_ARRAYMASK) >> TIMER_ARRAYSHIFT;  in timer_get_idx()
506 static inline void timer_set_idx(struct timer_list *timer, unsigned int idx)  in timer_set_idx()  argument
508 	timer->flags = (timer->flags & ~TIMER_ARRAYMASK) |  in timer_set_idx()
521 	 * The timer wheel has to guarantee that a timer does not fire  in calc_index()
523 	 * - Timer is armed at the edge of a tick  in calc_index()
572 trigger_dyntick_cpu(struct timer_base *base, struct timer_list *timer)  in trigger_dyntick_cpu()  argument
581 	if (timer->flags & TIMER_DEFERRABLE) {  in trigger_dyntick_cpu()
589 	 * timer is not deferrable. If the other CPU is on the way to idle  in trigger_dyntick_cpu()
597  * Enqueue the timer into the hash bucket, mark it pending in
598  * the bitmap, store the index in the timer flags then wake up
601 static void enqueue_timer(struct timer_base *base, struct timer_list *timer,  in enqueue_timer()  argument
605 	hlist_add_head(&timer->entry, base->vectors + idx);  in enqueue_timer()
607 	timer_set_idx(timer, idx);  in enqueue_timer()
609 	trace_timer_start(timer, timer->expires, timer->flags);  in enqueue_timer()
612 	 * Check whether this is the new first expiring timer. The  in enqueue_timer()
613 	 * effective expiry time of the timer is required here  in enqueue_timer()
614 	 * (bucket_expiry) instead of timer->expires.  in enqueue_timer()
624 		trigger_dyntick_cpu(base, timer);  in enqueue_timer()
628 static void internal_add_timer(struct timer_base *base, struct timer_list *timer)  in internal_add_timer()  argument
633 	idx = calc_wheel_index(timer->expires, base->clk, &bucket_expiry);  in internal_add_timer()
634 	enqueue_timer(base, timer, idx, bucket_expiry);  in internal_add_timer()
655 		   struct delayed_work, timer, work.func),
657 		   struct kthread_delayed_work, timer, work.func),
662 	struct timer_list *timer = addr;  in timer_debug_hint()  local
666 		if (timer_hints[i].function == timer->function) {  in timer_debug_hint()
673 	return timer->function;  in timer_debug_hint()
678 	struct timer_list *timer = addr;  in timer_is_static_object()  local
680 	return (timer->entry.pprev == NULL &&  in timer_is_static_object()
681 		timer->entry.next == TIMER_ENTRY_STATIC);  in timer_is_static_object()
690 	struct timer_list *timer = addr;  in timer_fixup_init()  local
694 		del_timer_sync(timer);  in timer_fixup_init()
695 		debug_object_init(timer, &timer_debug_descr);  in timer_fixup_init()
702 /* Stub timer callback for improperly used timers. */
715 	struct timer_list *timer = addr;  in timer_fixup_activate()  local
719 		timer_setup(timer, stub_timer, 0);  in timer_fixup_activate()
736 	struct timer_list *timer = addr;  in timer_fixup_free()  local
740 		del_timer_sync(timer);  in timer_fixup_free()
741 		debug_object_free(timer, &timer_debug_descr);  in timer_fixup_free()
754 	struct timer_list *timer = addr;  in timer_fixup_assert_init()  local
758 		timer_setup(timer, stub_timer, 0);  in timer_fixup_assert_init()
775 static inline void debug_timer_init(struct timer_list *timer)  in debug_timer_init()  argument
777 	debug_object_init(timer, &timer_debug_descr);  in debug_timer_init()
780 static inline void debug_timer_activate(struct timer_list *timer)  in debug_timer_activate()  argument
782 	debug_object_activate(timer, &timer_debug_descr);  in debug_timer_activate()
785 static inline void debug_timer_deactivate(struct timer_list *timer)  in debug_timer_deactivate()  argument
787 	debug_object_deactivate(timer, &timer_debug_descr);  in debug_timer_deactivate()
790 static inline void debug_timer_assert_init(struct timer_list *timer)  in debug_timer_assert_init()  argument
792 	debug_object_assert_init(timer, &timer_debug_descr);  in debug_timer_assert_init()
795 static void do_init_timer(struct timer_list *timer,
800 void init_timer_on_stack_key(struct timer_list *timer,  in init_timer_on_stack_key()  argument
805 	debug_object_init_on_stack(timer, &timer_debug_descr);  in init_timer_on_stack_key()
806 	do_init_timer(timer, func, flags, name, key);  in init_timer_on_stack_key()
810 void destroy_timer_on_stack(struct timer_list *timer)  in destroy_timer_on_stack()  argument
812 	debug_object_free(timer, &timer_debug_descr);  in destroy_timer_on_stack()
817 static inline void debug_timer_init(struct timer_list *timer) { }  in debug_timer_init()  argument
818 static inline void debug_timer_activate(struct timer_list *timer) { }  in debug_timer_activate()  argument
819 static inline void debug_timer_deactivate(struct timer_list *timer) { }  in debug_timer_deactivate()  argument
820 static inline void debug_timer_assert_init(struct timer_list *timer) { }  in debug_timer_assert_init()  argument
823 static inline void debug_init(struct timer_list *timer)  in debug_init()  argument
825 	debug_timer_init(timer);  in debug_init()
826 	trace_timer_init(timer);  in debug_init()
829 static inline void debug_deactivate(struct timer_list *timer)  in debug_deactivate()  argument
831 	debug_timer_deactivate(timer);  in debug_deactivate()
832 	trace_timer_cancel(timer);  in debug_deactivate()
835 static inline void debug_assert_init(struct timer_list *timer)  in debug_assert_init()  argument
837 	debug_timer_assert_init(timer);  in debug_assert_init()
840 static void do_init_timer(struct timer_list *timer,  in do_init_timer()  argument
845 	timer->entry.pprev = NULL;  in do_init_timer()
846 	timer->function = func;  in do_init_timer()
849 	timer->flags = flags | raw_smp_processor_id();  in do_init_timer()
850 	lockdep_init_map(&timer->lockdep_map, name, key, 0);  in do_init_timer()
854  * init_timer_key - initialize a timer
855  * @timer: the timer to be initialized
856  * @func: timer callback function
857  * @flags: timer flags
858  * @name: name of the timer
859  * @key: lockdep class key of the fake lock used for tracking timer
862  * init_timer_key() must be done to a timer prior calling *any* of the
863  * other timer functions.
865 void init_timer_key(struct timer_list *timer,  in init_timer_key()  argument
869 	debug_init(timer);  in init_timer_key()
870 	do_init_timer(timer, func, flags, name, key);  in init_timer_key()
874 static inline void detach_timer(struct timer_list *timer, bool clear_pending)  in detach_timer()  argument
876 	struct hlist_node *entry = &timer->entry;  in detach_timer()
878 	debug_deactivate(timer);  in detach_timer()
886 static int detach_if_pending(struct timer_list *timer, struct timer_base *base,  in detach_if_pending()  argument
889 	unsigned idx = timer_get_idx(timer);  in detach_if_pending()
891 	if (!timer_pending(timer))  in detach_if_pending()
894 	if (hlist_is_singular_node(&timer->entry, base->vectors + idx)) {  in detach_if_pending()
899 	detach_timer(timer, clear_pending);  in detach_if_pending()
908 	 * If the timer is deferrable and NO_HZ_COMMON is set then we need  in get_timer_cpu_base()
921 	 * If the timer is deferrable and NO_HZ_COMMON is set then we need  in get_timer_this_cpu_base()
979  * When a timer is migrating then the TIMER_MIGRATING flag is set and we need
982 static struct timer_base *lock_timer_base(struct timer_list *timer,  in lock_timer_base()  argument
984 	__acquires(timer->base->lock)  in lock_timer_base()
995 		tf = READ_ONCE(timer->flags);  in lock_timer_base()
1000 			if (timer->flags == tf)  in lock_timer_base()
1013 __mod_timer(struct timer_list *timer, unsigned long expires, unsigned int options)  in __mod_timer()  argument
1020 	BUG_ON(!timer->function);  in __mod_timer()
1024 	 * the timer is re-modified to have the same timeout or ends up in the  in __mod_timer()
1027 	if (!(options & MOD_TIMER_NOTPENDING) && timer_pending(timer)) {  in __mod_timer()
1031 		 * timer with this expiry.  in __mod_timer()
1033 		long diff = timer->expires - expires;  in __mod_timer()
1041 		 * We lock timer base and calculate the bucket index right  in __mod_timer()
1042 		 * here. If the timer ends up in the same bucket, then we  in __mod_timer()
1046 		base = lock_timer_base(timer, &flags);  in __mod_timer()
1049 		if (timer_pending(timer) && (options & MOD_TIMER_REDUCE) &&  in __mod_timer()
1050 		    time_before_eq(timer->expires, expires)) {  in __mod_timer()
1060 		 * timer. If it matches set the expiry to the new value so a  in __mod_timer()
1063 		if (idx == timer_get_idx(timer)) {  in __mod_timer()
1065 				timer->expires = expires;  in __mod_timer()
1066 			else if (time_after(timer->expires, expires))  in __mod_timer()
1067 				timer->expires = expires;  in __mod_timer()
1072 		base = lock_timer_base(timer, &flags);  in __mod_timer()
1076 	ret = detach_if_pending(timer, base, false);  in __mod_timer()
1080 	new_base = get_target_base(base, timer->flags);  in __mod_timer()
1084 		 * We are trying to schedule the timer on the new base.  in __mod_timer()
1085 		 * However we can't change timer's base while it is running,  in __mod_timer()
1086 		 * otherwise del_timer_sync() can't detect that the timer's  in __mod_timer()
1088 		 * timer is serialized wrt itself.  in __mod_timer()
1090 		if (likely(base->running_timer != timer)) {  in __mod_timer()
1092 			timer->flags |= TIMER_MIGRATING;  in __mod_timer()
1097 			WRITE_ONCE(timer->flags,  in __mod_timer()
1098 				   (timer->flags & ~TIMER_BASEMASK) | base->cpu);  in __mod_timer()
1103 	debug_timer_activate(timer);  in __mod_timer()
1105 	timer->expires = expires;  in __mod_timer()
1113 		enqueue_timer(base, timer, idx, bucket_expiry);  in __mod_timer()
1115 		internal_add_timer(base, timer);  in __mod_timer()
1124  * mod_timer_pending - modify a pending timer's timeout
1125  * @timer: the pending timer to be modified
1133 int mod_timer_pending(struct timer_list *timer, unsigned long expires)  in mod_timer_pending()  argument
1135 	return __mod_timer(timer, expires, MOD_TIMER_PENDING_ONLY);  in mod_timer_pending()
1140  * mod_timer - modify a timer's timeout
1141  * @timer: the timer to be modified
1145  * active timer (if the timer is inactive it will be activated)
1147  * mod_timer(timer, expires) is equivalent to:
1149  *     del_timer(timer); timer->expires = expires; add_timer(timer);
1152  * same timer, then mod_timer() is the only safe way to modify the timeout,
1153  * since add_timer() cannot modify an already running timer.
1155  * The function returns whether it has modified a pending timer or not.
1156  * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
1157  * active timer returns 1.)
1159 int mod_timer(struct timer_list *timer, unsigned long expires)  in mod_timer()  argument
1161 	return __mod_timer(timer, expires, 0);  in mod_timer()
1166  * timer_reduce - Modify a timer's timeout if it would reduce the timeout
1167  * @timer:	The timer to be modified
1171  * modify a running timer if that would reduce the expiration time (it will
1172  * start a timer that isn't running).
1174 int timer_reduce(struct timer_list *timer, unsigned long expires)  in timer_reduce()  argument
1176 	return __mod_timer(timer, expires, MOD_TIMER_REDUCE);  in timer_reduce()
1181  * add_timer - start a timer
1182  * @timer: the timer to be added
1184  * The kernel will do a ->function(@timer) callback from the
1185  * timer interrupt at the ->expires point in the future. The
1188  * The timer's ->expires, ->function fields must be set prior calling this
1192  * timer tick.
1194 void add_timer(struct timer_list *timer)  in add_timer()  argument
1196 	BUG_ON(timer_pending(timer));  in add_timer()
1197 	__mod_timer(timer, timer->expires, MOD_TIMER_NOTPENDING);  in add_timer()
1202  * add_timer_on - start a timer on a particular CPU
1203  * @timer: the timer to be added
1208 void add_timer_on(struct timer_list *timer, int cpu)  in add_timer_on()  argument
1213 	BUG_ON(timer_pending(timer) || !timer->function);  in add_timer_on()
1215 	new_base = get_timer_cpu_base(timer->flags, cpu);  in add_timer_on()
1218 	 * If @timer was on a different CPU, it should be migrated with the  in add_timer_on()
1222 	base = lock_timer_base(timer, &flags);  in add_timer_on()
1224 		timer->flags |= TIMER_MIGRATING;  in add_timer_on()
1229 		WRITE_ONCE(timer->flags,  in add_timer_on()
1230 			   (timer->flags & ~TIMER_BASEMASK) | cpu);  in add_timer_on()
1234 	debug_timer_activate(timer);  in add_timer_on()
1235 	internal_add_timer(base, timer);  in add_timer_on()
1241  * del_timer - deactivate a timer.
1242  * @timer: the timer to be deactivated
1244  * del_timer() deactivates a timer - this works on both active and inactive
1247  * The function returns whether it has deactivated a pending timer or not.
1248  * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
1249  * active timer returns 1.)
1251 int del_timer(struct timer_list *timer)  in del_timer()  argument
1257 	debug_assert_init(timer);  in del_timer()
1259 	if (timer_pending(timer)) {  in del_timer()
1260 		base = lock_timer_base(timer, &flags);  in del_timer()
1261 		ret = detach_if_pending(timer, base, true);  in del_timer()
1270  * try_to_del_timer_sync - Try to deactivate a timer
1271  * @timer: timer to delete
1273  * This function tries to deactivate a timer. Upon successful (ret >= 0)
1274  * exit the timer is not queued and the handler is not running on any CPU.
1276 int try_to_del_timer_sync(struct timer_list *timer)  in try_to_del_timer_sync()  argument
1282 	debug_assert_init(timer);  in try_to_del_timer_sync()
1284 	base = lock_timer_base(timer, &flags);  in try_to_del_timer_sync()
1286 	if (base->running_timer != timer)  in try_to_del_timer_sync()
1287 		ret = detach_if_pending(timer, base, true);  in try_to_del_timer_sync()
1315  * timer callback to finish. Drop expiry_lock and reacquire it. That allows
1330  * deletion of a timer failed because the timer callback function was
1335  * delete a timer preempted the softirq thread running the timer callback
1338 static void del_timer_wait_running(struct timer_list *timer)  in del_timer_wait_running()  argument
1342 	tf = READ_ONCE(timer->flags);  in del_timer_wait_running()
1348 		 * which is held by the softirq across the timer  in del_timer_wait_running()
1350 		 * expire the next timer. In theory the timer could already  in del_timer_wait_running()
1365 static inline void del_timer_wait_running(struct timer_list *timer) { }  in del_timer_wait_running()  argument
1370  * del_timer_sync - deactivate a timer and wait for the handler to finish.
1371  * @timer: the timer to be deactivated
1374  * the timer it also makes sure the handler has finished executing on other
1377  * Synchronization rules: Callers must prevent restarting of the timer,
1379  * interrupt contexts unless the timer is an irqsafe one. The caller must
1380  * not hold locks which would prevent completion of the timer's
1381  * handler. The timer's handler must not call add_timer_on(). Upon exit the
1382  * timer is not queued and the handler is not running on any CPU.
1386  *   nothing to do with the timer in question.  Here's why::
1403  * The function returns whether it has deactivated a pending timer or not.
1405 int del_timer_sync(struct timer_list *timer)  in del_timer_sync()  argument
1417 	lock_map_acquire(&timer->lockdep_map);  in del_timer_sync()
1418 	lock_map_release(&timer->lockdep_map);  in del_timer_sync()
1425 	WARN_ON(in_irq() && !(timer->flags & TIMER_IRQSAFE));  in del_timer_sync()
1431 	if (IS_ENABLED(CONFIG_PREEMPT_RT) && !(timer->flags & TIMER_IRQSAFE))  in del_timer_sync()
1435 		ret = try_to_del_timer_sync(timer);  in del_timer_sync()
1438 			del_timer_wait_running(timer);  in del_timer_sync()
1448 static void call_timer_fn(struct timer_list *timer,  in call_timer_fn()  argument
1456 	 * It is permissible to free the timer from inside the  in call_timer_fn()
1460 	 * timer->lockdep_map, make a copy and use that here.  in call_timer_fn()
1464 	lockdep_copy_map(&lockdep_map, &timer->lockdep_map);  in call_timer_fn()
1473 	trace_timer_expire_entry(timer, baseclk);  in call_timer_fn()
1474 	fn(timer);  in call_timer_fn()
1475 	trace_timer_expire_exit(timer);  in call_timer_fn()
1480 		WARN_ONCE(1, "timer: %pS preempt leak: %08x -> %08x\n",  in call_timer_fn()
1502 		struct timer_list *timer;  in expire_timers()  local
1505 		timer = hlist_entry(head->first, struct timer_list, entry);  in expire_timers()
1507 		base->running_timer = timer;  in expire_timers()
1508 		detach_timer(timer, true);  in expire_timers()
1510 		fn = timer->function;  in expire_timers()
1512 		if (timer->flags & TIMER_IRQSAFE) {  in expire_timers()
1514 			call_timer_fn(timer, fn, baseclk);  in expire_timers()
1519 			call_timer_fn(timer, fn, baseclk);  in expire_timers()
1572  * Search the first expiring timer in the various clock levels. Caller must
1649  * Check, if the next hrtimer event is before the next timer wheel
1664 	 * If the next timer is already expired, return the tick base  in cmp_next_hrtimer_event()
1673 	 * make sure that this tick really expires the timer to avoid  in cmp_next_hrtimer_event()
1682  * get_next_timer_interrupt - return the time (clock mono) of the next timer
1687  * timer or KTIME_MAX if no timer is pending.
1696 	 * Pretend that there is no timer pending if the cpu is offline.  in get_next_timer_interrupt()
1727 		 * Also the tick is stopped so any added timer must forward  in get_next_timer_interrupt()
1741  * timer_clear_idle - Clear the idle state of the timer base
1761  * @base: the timer vector to be processed.
1779 		 * timer at this clk are that all matching timers have been  in __run_timers()
1780 		 * dequeued or no timer has been queued since  in __run_timers()
1797  * This function runs timers and the timer-tq in bottom half context.
1809  * Called by the local, per-CPU timer interrupt on SMP.
1829  * Called from the timer interrupt handler to charge one tick to the current
1836 	/* Note: this timer irq context must be accounted for as well. */  in update_process_times()
1850  * Since schedule_timeout()'s timer is defined on the stack, it must store
1854 	struct timer_list timer;  member
1860 	struct process_timer *timeout = from_timer(timeout, t, timer);  in process_timeout()
1892  * Returns 0 when the timer has expired otherwise the remaining time in
1898 	struct process_timer timer;  in schedule_timeout()  local
1932 	timer.task = current;  in schedule_timeout()
1933 	timer_setup_on_stack(&timer.timer, process_timeout, 0);  in schedule_timeout()
1934 	__mod_timer(&timer.timer, expire, MOD_TIMER_NOTPENDING);  in schedule_timeout()
1936 	del_singleshot_timer_sync(&timer.timer);  in schedule_timeout()
1938 	/* Remove the timer from the object tracker */  in schedule_timeout()
1939 	destroy_timer_on_stack(&timer.timer);  in schedule_timeout()
1987 	struct timer_list *timer;  in migrate_timer_list()  local
1991 		timer = hlist_entry(head->first, struct timer_list, entry);  in migrate_timer_list()
1992 		detach_timer(timer, false);  in migrate_timer_list()
1993 		timer->flags = (timer->flags & ~TIMER_BASEMASK) | cpu;  in migrate_timer_list()
1994 		internal_add_timer(new_base, timer);  in migrate_timer_list()