16  *  2000-10-05  Implemented scalable SMP per-CPU timer handling.
57 #include <trace/events/timer.h>
64  * The timer wheel has LVL_DEPTH array levels. Each level provides an array of
71  * The array level of a newly armed timer depends on the relative expiry
75  * Contrary to the original timer wheel implementation, which aims for 'exact'
77  * the timers into the lower array levels. The previous 'classic' timer wheel
82  * This is an optimization of the original timer wheel implementation for the
83  * majority of the timer wheel use cases: timeouts. The vast majority of
294 	 * due to delays of the timer irq, long irq off times etc etc) then  in round_jiffies_common()
477 static inline unsigned int timer_get_idx(struct timer_list *timer)  in timer_get_idx()  argument
479 	return (timer->flags & TIMER_ARRAYMASK) >> TIMER_ARRAYSHIFT;  in timer_get_idx()
482 static inline void timer_set_idx(struct timer_list *timer, unsigned int idx)  in timer_set_idx()  argument
484 	timer->flags = (timer->flags & ~TIMER_ARRAYMASK) |  in timer_set_idx()
497 	 * The timer wheel has to guarantee that a timer does not fire  in calc_index()
499 	 * - Timer is armed at the edge of a tick  in calc_index()
548 trigger_dyntick_cpu(struct timer_base *base, struct timer_list *timer)  in trigger_dyntick_cpu()  argument
557 	if (timer->flags & TIMER_DEFERRABLE) {  in trigger_dyntick_cpu()
565 	 * timer is not deferrable. If the other CPU is on the way to idle  in trigger_dyntick_cpu()
573  * Enqueue the timer into the hash bucket, mark it pending in
574  * the bitmap, store the index in the timer flags then wake up
577 static void enqueue_timer(struct timer_base *base, struct timer_list *timer,  in enqueue_timer()  argument
581 	hlist_add_head(&timer->entry, base->vectors + idx);  in enqueue_timer()
583 	timer_set_idx(timer, idx);  in enqueue_timer()
585 	trace_timer_start(timer, timer->expires, timer->flags);  in enqueue_timer()
588 	 * Check whether this is the new first expiring timer. The  in enqueue_timer()
589 	 * effective expiry time of the timer is required here  in enqueue_timer()
590 	 * (bucket_expiry) instead of timer->expires.  in enqueue_timer()
599 		trigger_dyntick_cpu(base, timer);  in enqueue_timer()
603 static void internal_add_timer(struct timer_base *base, struct timer_list *timer)  in internal_add_timer()  argument
608 	idx = calc_wheel_index(timer->expires, base->clk, &bucket_expiry);  in internal_add_timer()
609 	enqueue_timer(base, timer, idx, bucket_expiry);  in internal_add_timer()
623 	struct timer_list *timer = addr;  in timer_is_static_object()  local
625 	return (timer->entry.pprev == NULL &&  in timer_is_static_object()
626 		timer->entry.next == TIMER_ENTRY_STATIC);  in timer_is_static_object()
635 	struct timer_list *timer = addr;  in timer_fixup_init()  local
639 		del_timer_sync(timer);  in timer_fixup_init()
640 		debug_object_init(timer, &timer_debug_descr);  in timer_fixup_init()
647 /* Stub timer callback for improperly used timers. */
660 	struct timer_list *timer = addr;  in timer_fixup_activate()  local
664 		timer_setup(timer, stub_timer, 0);  in timer_fixup_activate()
681 	struct timer_list *timer = addr;  in timer_fixup_free()  local
685 		del_timer_sync(timer);  in timer_fixup_free()
686 		debug_object_free(timer, &timer_debug_descr);  in timer_fixup_free()
699 	struct timer_list *timer = addr;  in timer_fixup_assert_init()  local
703 		timer_setup(timer, stub_timer, 0);  in timer_fixup_assert_init()
720 static inline void debug_timer_init(struct timer_list *timer)  in debug_timer_init()  argument
722 	debug_object_init(timer, &timer_debug_descr);  in debug_timer_init()
725 static inline void debug_timer_activate(struct timer_list *timer)  in debug_timer_activate()  argument
727 	debug_object_activate(timer, &timer_debug_descr);  in debug_timer_activate()
730 static inline void debug_timer_deactivate(struct timer_list *timer)  in debug_timer_deactivate()  argument
732 	debug_object_deactivate(timer, &timer_debug_descr);  in debug_timer_deactivate()
735 static inline void debug_timer_assert_init(struct timer_list *timer)  in debug_timer_assert_init()  argument
737 	debug_object_assert_init(timer, &timer_debug_descr);  in debug_timer_assert_init()
740 static void do_init_timer(struct timer_list *timer,
745 void init_timer_on_stack_key(struct timer_list *timer,  in init_timer_on_stack_key()  argument
750 	debug_object_init_on_stack(timer, &timer_debug_descr);  in init_timer_on_stack_key()
751 	do_init_timer(timer, func, flags, name, key);  in init_timer_on_stack_key()
755 void destroy_timer_on_stack(struct timer_list *timer)  in destroy_timer_on_stack()  argument
757 	debug_object_free(timer, &timer_debug_descr);  in destroy_timer_on_stack()
762 static inline void debug_timer_init(struct timer_list *timer) { }  in debug_timer_init()  argument
763 static inline void debug_timer_activate(struct timer_list *timer) { }  in debug_timer_activate()  argument
764 static inline void debug_timer_deactivate(struct timer_list *timer) { }  in debug_timer_deactivate()  argument
765 static inline void debug_timer_assert_init(struct timer_list *timer) { }  in debug_timer_assert_init()  argument
768 static inline void debug_init(struct timer_list *timer)  in debug_init()  argument
770 	debug_timer_init(timer);  in debug_init()
771 	trace_timer_init(timer);  in debug_init()
774 static inline void debug_deactivate(struct timer_list *timer)  in debug_deactivate()  argument
776 	debug_timer_deactivate(timer);  in debug_deactivate()
777 	trace_timer_cancel(timer);  in debug_deactivate()
780 static inline void debug_assert_init(struct timer_list *timer)  in debug_assert_init()  argument
782 	debug_timer_assert_init(timer);  in debug_assert_init()
785 static void do_init_timer(struct timer_list *timer,  in do_init_timer()  argument
790 	timer->entry.pprev = NULL;  in do_init_timer()
791 	timer->function = func;  in do_init_timer()
794 	timer->flags = flags | raw_smp_processor_id();  in do_init_timer()
795 	lockdep_init_map(&timer->lockdep_map, name, key, 0);  in do_init_timer()
799  * init_timer_key - initialize a timer
800  * @timer: the timer to be initialized
801  * @func: timer callback function
802  * @flags: timer flags
803  * @name: name of the timer
804  * @key: lockdep class key of the fake lock used for tracking timer
807  * init_timer_key() must be done to a timer prior calling *any* of the
808  * other timer functions.
810 void init_timer_key(struct timer_list *timer,  in init_timer_key()  argument
814 	debug_init(timer);  in init_timer_key()
815 	do_init_timer(timer, func, flags, name, key);  in init_timer_key()
819 static inline void detach_timer(struct timer_list *timer, bool clear_pending)  in detach_timer()  argument
821 	struct hlist_node *entry = &timer->entry;  in detach_timer()
823 	debug_deactivate(timer);  in detach_timer()
831 static int detach_if_pending(struct timer_list *timer, struct timer_base *base,  in detach_if_pending()  argument
834 	unsigned idx = timer_get_idx(timer);  in detach_if_pending()
836 	if (!timer_pending(timer))  in detach_if_pending()
839 	if (hlist_is_singular_node(&timer->entry, base->vectors + idx)) {  in detach_if_pending()
844 	detach_timer(timer, clear_pending);  in detach_if_pending()
853 	 * If the timer is deferrable and NO_HZ_COMMON is set then we need  in get_timer_cpu_base()
866 	 * If the timer is deferrable and NO_HZ_COMMON is set then we need  in get_timer_this_cpu_base()
924  * When a timer is migrating then the TIMER_MIGRATING flag is set and we need
927 static struct timer_base *lock_timer_base(struct timer_list *timer,  in lock_timer_base()  argument
929 	__acquires(timer->base->lock)  in lock_timer_base()
940 		tf = READ_ONCE(timer->flags);  in lock_timer_base()
945 			if (timer->flags == tf)  in lock_timer_base()
958 __mod_timer(struct timer_list *timer, unsigned long expires, unsigned int options)  in __mod_timer()  argument
965 	BUG_ON(!timer->function);  in __mod_timer()
969 	 * the timer is re-modified to have the same timeout or ends up in the  in __mod_timer()
972 	if (!(options & MOD_TIMER_NOTPENDING) && timer_pending(timer)) {  in __mod_timer()
976 		 * timer with this expiry.  in __mod_timer()
978 		long diff = timer->expires - expires;  in __mod_timer()
986 		 * We lock timer base and calculate the bucket index right  in __mod_timer()
987 		 * here. If the timer ends up in the same bucket, then we  in __mod_timer()
991 		base = lock_timer_base(timer, &flags);  in __mod_timer()
994 		if (timer_pending(timer) && (options & MOD_TIMER_REDUCE) &&  in __mod_timer()
995 		    time_before_eq(timer->expires, expires)) {  in __mod_timer()
1005 		 * timer. If it matches set the expiry to the new value so a  in __mod_timer()
1008 		if (idx == timer_get_idx(timer)) {  in __mod_timer()
1010 				timer->expires = expires;  in __mod_timer()
1011 			else if (time_after(timer->expires, expires))  in __mod_timer()
1012 				timer->expires = expires;  in __mod_timer()
1017 		base = lock_timer_base(timer, &flags);  in __mod_timer()
1021 	ret = detach_if_pending(timer, base, false);  in __mod_timer()
1025 	new_base = get_target_base(base, timer->flags);  in __mod_timer()
1029 		 * We are trying to schedule the timer on the new base.  in __mod_timer()
1030 		 * However we can't change timer's base while it is running,  in __mod_timer()
1031 		 * otherwise del_timer_sync() can't detect that the timer's  in __mod_timer()
1033 		 * timer is serialized wrt itself.  in __mod_timer()
1035 		if (likely(base->running_timer != timer)) {  in __mod_timer()
1037 			timer->flags |= TIMER_MIGRATING;  in __mod_timer()
1042 			WRITE_ONCE(timer->flags,  in __mod_timer()
1043 				   (timer->flags & ~TIMER_BASEMASK) | base->cpu);  in __mod_timer()
1048 	debug_timer_activate(timer);  in __mod_timer()
1050 	timer->expires = expires;  in __mod_timer()
1058 		enqueue_timer(base, timer, idx, bucket_expiry);  in __mod_timer()
1060 		internal_add_timer(base, timer);  in __mod_timer()
1069  * mod_timer_pending - modify a pending timer's timeout
1070  * @timer: the pending timer to be modified
1078 int mod_timer_pending(struct timer_list *timer, unsigned long expires)  in mod_timer_pending()  argument
1080 	return __mod_timer(timer, expires, MOD_TIMER_PENDING_ONLY);  in mod_timer_pending()
1085  * mod_timer - modify a timer's timeout
1086  * @timer: the timer to be modified
1090  * active timer (if the timer is inactive it will be activated)
1092  * mod_timer(timer, expires) is equivalent to:
1094  *     del_timer(timer); timer->expires = expires; add_timer(timer);
1097  * same timer, then mod_timer() is the only safe way to modify the timeout,
1098  * since add_timer() cannot modify an already running timer.
1100  * The function returns whether it has modified a pending timer or not.
1101  * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
1102  * active timer returns 1.)
1104 int mod_timer(struct timer_list *timer, unsigned long expires)  in mod_timer()  argument
1106 	return __mod_timer(timer, expires, 0);  in mod_timer()
1111  * timer_reduce - Modify a timer's timeout if it would reduce the timeout
1112  * @timer:	The timer to be modified
1116  * modify a running timer if that would reduce the expiration time (it will
1117  * start a timer that isn't running).
1119 int timer_reduce(struct timer_list *timer, unsigned long expires)  in timer_reduce()  argument
1121 	return __mod_timer(timer, expires, MOD_TIMER_REDUCE);  in timer_reduce()
1126  * add_timer - start a timer
1127  * @timer: the timer to be added
1129  * The kernel will do a ->function(@timer) callback from the
1130  * timer interrupt at the ->expires point in the future. The
1133  * The timer's ->expires, ->function fields must be set prior calling this
1137  * timer tick.
1139 void add_timer(struct timer_list *timer)  in add_timer()  argument
1141 	BUG_ON(timer_pending(timer));  in add_timer()
1142 	__mod_timer(timer, timer->expires, MOD_TIMER_NOTPENDING);  in add_timer()
1147  * add_timer_on - start a timer on a particular CPU
1148  * @timer: the timer to be added
1153 void add_timer_on(struct timer_list *timer, int cpu)  in add_timer_on()  argument
1158 	BUG_ON(timer_pending(timer) || !timer->function);  in add_timer_on()
1160 	new_base = get_timer_cpu_base(timer->flags, cpu);  in add_timer_on()
1163 	 * If @timer was on a different CPU, it should be migrated with the  in add_timer_on()
1167 	base = lock_timer_base(timer, &flags);  in add_timer_on()
1169 		timer->flags |= TIMER_MIGRATING;  in add_timer_on()
1174 		WRITE_ONCE(timer->flags,  in add_timer_on()
1175 			   (timer->flags & ~TIMER_BASEMASK) | cpu);  in add_timer_on()
1179 	debug_timer_activate(timer);  in add_timer_on()
1180 	internal_add_timer(base, timer);  in add_timer_on()
1186  * del_timer - deactivate a timer.
1187  * @timer: the timer to be deactivated
1189  * del_timer() deactivates a timer - this works on both active and inactive
1192  * The function returns whether it has deactivated a pending timer or not.
1193  * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
1194  * active timer returns 1.)
1196 int del_timer(struct timer_list *timer)  in del_timer()  argument
1202 	debug_assert_init(timer);  in del_timer()
1204 	if (timer_pending(timer)) {  in del_timer()
1205 		base = lock_timer_base(timer, &flags);  in del_timer()
1206 		ret = detach_if_pending(timer, base, true);  in del_timer()
1215  * try_to_del_timer_sync - Try to deactivate a timer
1216  * @timer: timer to delete
1218  * This function tries to deactivate a timer. Upon successful (ret >= 0)
1219  * exit the timer is not queued and the handler is not running on any CPU.
1221 int try_to_del_timer_sync(struct timer_list *timer)  in try_to_del_timer_sync()  argument
1227 	debug_assert_init(timer);  in try_to_del_timer_sync()
1229 	base = lock_timer_base(timer, &flags);  in try_to_del_timer_sync()
1231 	if (base->running_timer != timer)  in try_to_del_timer_sync()
1232 		ret = detach_if_pending(timer, base, true);  in try_to_del_timer_sync()
1260  * timer callback to finish. Drop expiry_lock and reaquire it. That allows
1273  * deletion of a timer failed because the timer callback function was
1278  * delete a timer preempted the softirq thread running the timer callback
1281 static void del_timer_wait_running(struct timer_list *timer)  in del_timer_wait_running()  argument
1285 	tf = READ_ONCE(timer->flags);  in del_timer_wait_running()
1291 		 * which is held by the softirq across the timer  in del_timer_wait_running()
1293 		 * expire the next timer. In theory the timer could already  in del_timer_wait_running()
1308 static inline void del_timer_wait_running(struct timer_list *timer) { }  in del_timer_wait_running()  argument
1313  * del_timer_sync - deactivate a timer and wait for the handler to finish.
1314  * @timer: the timer to be deactivated
1317  * the timer it also makes sure the handler has finished executing on other
1320  * Synchronization rules: Callers must prevent restarting of the timer,
1322  * interrupt contexts unless the timer is an irqsafe one. The caller must
1323  * not hold locks which would prevent completion of the timer's
1324  * handler. The timer's handler must not call add_timer_on(). Upon exit the
1325  * timer is not queued and the handler is not running on any CPU.
1329  *   nothing to do with the timer in question.  Here's why::
1346  * The function returns whether it has deactivated a pending timer or not.
1348 int del_timer_sync(struct timer_list *timer)  in del_timer_sync()  argument
1360 	lock_map_acquire(&timer->lockdep_map);  in del_timer_sync()
1361 	lock_map_release(&timer->lockdep_map);  in del_timer_sync()
1368 	WARN_ON(in_irq() && !(timer->flags & TIMER_IRQSAFE));  in del_timer_sync()
1371 		ret = try_to_del_timer_sync(timer);  in del_timer_sync()
1374 			del_timer_wait_running(timer);  in del_timer_sync()
1384 static void call_timer_fn(struct timer_list *timer,  in call_timer_fn()  argument
1392 	 * It is permissible to free the timer from inside the  in call_timer_fn()
1396 	 * timer->lockdep_map, make a copy and use that here.  in call_timer_fn()
1400 	lockdep_copy_map(&lockdep_map, &timer->lockdep_map);  in call_timer_fn()
1409 	trace_timer_expire_entry(timer, baseclk);  in call_timer_fn()
1410 	fn(timer);  in call_timer_fn()
1411 	trace_timer_expire_exit(timer);  in call_timer_fn()
1416 		WARN_ONCE(1, "timer: %pS preempt leak: %08x -> %08x\n",  in call_timer_fn()
1438 		struct timer_list *timer;  in expire_timers()  local
1441 		timer = hlist_entry(head->first, struct timer_list, entry);  in expire_timers()
1443 		base->running_timer = timer;  in expire_timers()
1444 		detach_timer(timer, true);  in expire_timers()
1446 		fn = timer->function;  in expire_timers()
1448 		if (timer->flags & TIMER_IRQSAFE) {  in expire_timers()
1450 			call_timer_fn(timer, fn, baseclk);  in expire_timers()
1455 			call_timer_fn(timer, fn, baseclk);  in expire_timers()
1508  * Search the first expiring timer in the various clock levels. Caller must
1584  * Check, if the next hrtimer event is before the next timer wheel
1599 	 * If the next timer is already expired, return the tick base  in cmp_next_hrtimer_event()
1608 	 * make sure that this tick really expires the timer to avoid  in cmp_next_hrtimer_event()
1617  * get_next_timer_interrupt - return the time (clock mono) of the next timer
1622  * timer or KTIME_MAX if no timer is pending.
1632 	 * Pretend that there is no timer pending if the cpu is offline.  in get_next_timer_interrupt()
1664 		 * Also the tick is stopped so any added timer must forward  in get_next_timer_interrupt()
1678  * timer_clear_idle - Clear the idle state of the timer base
1697  * Called from the timer interrupt handler to charge one tick to the current
1706 	/* Note: this timer irq context must be accounted for as well. */  in update_process_times()
1721  * @base: the timer vector to be processed.
1739 		 * timer at this clk is that all matching timers have been  in __run_timers()
1754  * This function runs timers and the timer-tq in bottom half context.
1766  * Called by the local, per-CPU timer interrupt on SMP.
1786  * Since schedule_timeout()'s timer is defined on the stack, it must store
1790 	struct timer_list timer;  member
1796 	struct process_timer *timeout = from_timer(timeout, t, timer);  in process_timeout()
1828  * Returns 0 when the timer has expired otherwise the remaining time in
1834 	struct process_timer timer;  in schedule_timeout()  local
1868 	timer.task = current;  in schedule_timeout()
1869 	timer_setup_on_stack(&timer.timer, process_timeout, 0);  in schedule_timeout()
1870 	__mod_timer(&timer.timer, expire, MOD_TIMER_NOTPENDING);  in schedule_timeout()
1872 	del_singleshot_timer_sync(&timer.timer);  in schedule_timeout()
1874 	/* Remove the timer from the object tracker */  in schedule_timeout()
1875 	destroy_timer_on_stack(&timer.timer);  in schedule_timeout()
1923 	struct timer_list *timer;  in migrate_timer_list()  local
1927 		timer = hlist_entry(head->first, struct timer_list, entry);  in migrate_timer_list()
1928 		detach_timer(timer, false);  in migrate_timer_list()
1929 		timer->flags = (timer->flags & ~TIMER_BASEMASK) | cpu;  in migrate_timer_list()
1930 		internal_add_timer(new_base, timer);  in migrate_timer_list()