4 * Kernel internal timers, basic process system calls
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/module.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/perf_event.h>
41 #include <linux/sched.h>
42 #include <linux/slab.h>
44 #include <asm/uaccess.h>
45 #include <asm/unistd.h>
46 #include <asm/div64.h>
47 #include <asm/timex.h>
50 #define CREATE_TRACE_POINTS
51 #include <trace/events/timer.h>
53 u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
55 EXPORT_SYMBOL(jiffies_64);
58 * per-CPU timer vector definitions:
60 #define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6)
61 #define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8)
62 #define TVN_SIZE (1 << TVN_BITS)
63 #define TVR_SIZE (1 << TVR_BITS)
64 #define TVN_MASK (TVN_SIZE - 1)
65 #define TVR_MASK (TVR_SIZE - 1)
68 struct list_head vec[TVN_SIZE];
72 struct list_head vec[TVR_SIZE];
77 struct timer_list *running_timer;
78 unsigned long timer_jiffies;
79 unsigned long next_timer;
85 } ____cacheline_aligned;
87 struct tvec_base boot_tvec_bases;
88 EXPORT_SYMBOL(boot_tvec_bases);
89 static DEFINE_PER_CPU(struct tvec_base *, tvec_bases) = &boot_tvec_bases;
92 * Note that all tvec_bases are 2 byte aligned and lower bit of
93 * base in timer_list is guaranteed to be zero. Use the LSB for
94 * the new flag to indicate whether the timer is deferrable
96 #define TBASE_DEFERRABLE_FLAG (0x1)
98 /* Functions below help us manage 'deferrable' flag */
99 static inline unsigned int tbase_get_deferrable(struct tvec_base *base)
101 return ((unsigned int)(unsigned long)base & TBASE_DEFERRABLE_FLAG);
104 static inline struct tvec_base *tbase_get_base(struct tvec_base *base)
106 return ((struct tvec_base *)((unsigned long)base & ~TBASE_DEFERRABLE_FLAG));
109 static inline void timer_set_deferrable(struct timer_list *timer)
111 timer->base = ((struct tvec_base *)((unsigned long)(timer->base) |
112 TBASE_DEFERRABLE_FLAG));
116 timer_set_base(struct timer_list *timer, struct tvec_base *new_base)
118 timer->base = (struct tvec_base *)((unsigned long)(new_base) |
119 tbase_get_deferrable(timer->base));
122 static unsigned long round_jiffies_common(unsigned long j, int cpu,
126 unsigned long original = j;
129 * We don't want all cpus firing their timers at once hitting the
130 * same lock or cachelines, so we skew each extra cpu with an extra
131 * 3 jiffies. This 3 jiffies came originally from the mm/ code which
133 * The skew is done by adding 3*cpunr, then round, then subtract this
134 * extra offset again.
141 * If the target jiffie is just after a whole second (which can happen
142 * due to delays of the timer irq, long irq off times etc etc) then
143 * we should round down to the whole second, not up. Use 1/4th second
144 * as cutoff for this rounding as an extreme upper bound for this.
145 * But never round down if @force_up is set.
147 if (rem < HZ/4 && !force_up) /* round down */
152 /* now that we have rounded, subtract the extra skew again */
155 if (j <= jiffies) /* rounding ate our timeout entirely; */
161 * __round_jiffies - function to round jiffies to a full second
162 * @j: the time in (absolute) jiffies that should be rounded
163 * @cpu: the processor number on which the timeout will happen
165 * __round_jiffies() rounds an absolute time in the future (in jiffies)
166 * up or down to (approximately) full seconds. This is useful for timers
167 * for which the exact time they fire does not matter too much, as long as
168 * they fire approximately every X seconds.
170 * By rounding these timers to whole seconds, all such timers will fire
171 * at the same time, rather than at various times spread out. The goal
172 * of this is to have the CPU wake up less, which saves power.
174 * The exact rounding is skewed for each processor to avoid all
175 * processors firing at the exact same time, which could lead
176 * to lock contention or spurious cache line bouncing.
178 * The return value is the rounded version of the @j parameter.
180 unsigned long __round_jiffies(unsigned long j, int cpu)
182 return round_jiffies_common(j, cpu, false);
184 EXPORT_SYMBOL_GPL(__round_jiffies);
187 * __round_jiffies_relative - function to round jiffies to a full second
188 * @j: the time in (relative) jiffies that should be rounded
189 * @cpu: the processor number on which the timeout will happen
191 * __round_jiffies_relative() rounds a time delta in the future (in jiffies)
192 * up or down to (approximately) full seconds. This is useful for timers
193 * for which the exact time they fire does not matter too much, as long as
194 * they fire approximately every X seconds.
196 * By rounding these timers to whole seconds, all such timers will fire
197 * at the same time, rather than at various times spread out. The goal
198 * of this is to have the CPU wake up less, which saves power.
200 * The exact rounding is skewed for each processor to avoid all
201 * processors firing at the exact same time, which could lead
202 * to lock contention or spurious cache line bouncing.
204 * The return value is the rounded version of the @j parameter.
206 unsigned long __round_jiffies_relative(unsigned long j, int cpu)
208 unsigned long j0 = jiffies;
210 /* Use j0 because jiffies might change while we run */
211 return round_jiffies_common(j + j0, cpu, false) - j0;
213 EXPORT_SYMBOL_GPL(__round_jiffies_relative);
216 * round_jiffies - function to round jiffies to a full second
217 * @j: the time in (absolute) jiffies that should be rounded
219 * round_jiffies() rounds an absolute time in the future (in jiffies)
220 * up or down to (approximately) full seconds. This is useful for timers
221 * for which the exact time they fire does not matter too much, as long as
222 * they fire approximately every X seconds.
224 * By rounding these timers to whole seconds, all such timers will fire
225 * at the same time, rather than at various times spread out. The goal
226 * of this is to have the CPU wake up less, which saves power.
228 * The return value is the rounded version of the @j parameter.
230 unsigned long round_jiffies(unsigned long j)
232 return round_jiffies_common(j, raw_smp_processor_id(), false);
234 EXPORT_SYMBOL_GPL(round_jiffies);
237 * round_jiffies_relative - function to round jiffies to a full second
238 * @j: the time in (relative) jiffies that should be rounded
240 * round_jiffies_relative() rounds a time delta in the future (in jiffies)
241 * up or down to (approximately) full seconds. This is useful for timers
242 * for which the exact time they fire does not matter too much, as long as
243 * they fire approximately every X seconds.
245 * By rounding these timers to whole seconds, all such timers will fire
246 * at the same time, rather than at various times spread out. The goal
247 * of this is to have the CPU wake up less, which saves power.
249 * The return value is the rounded version of the @j parameter.
251 unsigned long round_jiffies_relative(unsigned long j)
253 return __round_jiffies_relative(j, raw_smp_processor_id());
255 EXPORT_SYMBOL_GPL(round_jiffies_relative);
258 * __round_jiffies_up - function to round jiffies up to a full second
259 * @j: the time in (absolute) jiffies that should be rounded
260 * @cpu: the processor number on which the timeout will happen
262 * This is the same as __round_jiffies() except that it will never
263 * round down. This is useful for timeouts for which the exact time
264 * of firing does not matter too much, as long as they don't fire too
267 unsigned long __round_jiffies_up(unsigned long j, int cpu)
269 return round_jiffies_common(j, cpu, true);
271 EXPORT_SYMBOL_GPL(__round_jiffies_up);
274 * __round_jiffies_up_relative - function to round jiffies up to a full second
275 * @j: the time in (relative) jiffies that should be rounded
276 * @cpu: the processor number on which the timeout will happen
278 * This is the same as __round_jiffies_relative() except that it will never
279 * round down. This is useful for timeouts for which the exact time
280 * of firing does not matter too much, as long as they don't fire too
283 unsigned long __round_jiffies_up_relative(unsigned long j, int cpu)
285 unsigned long j0 = jiffies;
287 /* Use j0 because jiffies might change while we run */
288 return round_jiffies_common(j + j0, cpu, true) - j0;
290 EXPORT_SYMBOL_GPL(__round_jiffies_up_relative);
293 * round_jiffies_up - function to round jiffies up to a full second
294 * @j: the time in (absolute) jiffies that should be rounded
296 * This is the same as round_jiffies() except that it will never
297 * round down. This is useful for timeouts for which the exact time
298 * of firing does not matter too much, as long as they don't fire too
301 unsigned long round_jiffies_up(unsigned long j)
303 return round_jiffies_common(j, raw_smp_processor_id(), true);
305 EXPORT_SYMBOL_GPL(round_jiffies_up);
308 * round_jiffies_up_relative - function to round jiffies up to a full second
309 * @j: the time in (relative) jiffies that should be rounded
311 * This is the same as round_jiffies_relative() except that it will never
312 * round down. This is useful for timeouts for which the exact time
313 * of firing does not matter too much, as long as they don't fire too
316 unsigned long round_jiffies_up_relative(unsigned long j)
318 return __round_jiffies_up_relative(j, raw_smp_processor_id());
320 EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
323 * set_timer_slack - set the allowed slack for a timer
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 static inline void set_running_timer(struct tvec_base *base,
342 struct timer_list *timer)
345 base->running_timer = timer;
349 static void internal_add_timer(struct tvec_base *base, struct timer_list *timer)
351 unsigned long expires = timer->expires;
352 unsigned long idx = expires - base->timer_jiffies;
353 struct list_head *vec;
355 if (idx < TVR_SIZE) {
356 int i = expires & TVR_MASK;
357 vec = base->tv1.vec + i;
358 } else if (idx < 1 << (TVR_BITS + TVN_BITS)) {
359 int i = (expires >> TVR_BITS) & TVN_MASK;
360 vec = base->tv2.vec + i;
361 } else if (idx < 1 << (TVR_BITS + 2 * TVN_BITS)) {
362 int i = (expires >> (TVR_BITS + TVN_BITS)) & TVN_MASK;
363 vec = base->tv3.vec + i;
364 } else if (idx < 1 << (TVR_BITS + 3 * TVN_BITS)) {
365 int i = (expires >> (TVR_BITS + 2 * TVN_BITS)) & TVN_MASK;
366 vec = base->tv4.vec + i;
367 } else if ((signed long) idx < 0) {
369 * Can happen if you add a timer with expires == jiffies,
370 * or you set a timer to go off in the past
372 vec = base->tv1.vec + (base->timer_jiffies & TVR_MASK);
375 /* If the timeout is larger than 0xffffffff on 64-bit
376 * architectures then we use the maximum timeout:
378 if (idx > 0xffffffffUL) {
380 expires = idx + base->timer_jiffies;
382 i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK;
383 vec = base->tv5.vec + i;
388 list_add_tail(&timer->entry, vec);
391 #ifdef CONFIG_TIMER_STATS
392 void __timer_stats_timer_set_start_info(struct timer_list *timer, void *addr)
394 if (timer->start_site)
397 timer->start_site = addr;
398 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
399 timer->start_pid = current->pid;
402 static void timer_stats_account_timer(struct timer_list *timer)
404 unsigned int flag = 0;
406 if (likely(!timer->start_site))
408 if (unlikely(tbase_get_deferrable(timer->base)))
409 flag |= TIMER_STATS_FLAG_DEFERRABLE;
411 timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
412 timer->function, timer->start_comm, flag);
416 static void timer_stats_account_timer(struct timer_list *timer) {}
419 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
421 static struct debug_obj_descr timer_debug_descr;
424 * fixup_init is called when:
425 * - an active object is initialized
427 static int timer_fixup_init(void *addr, enum debug_obj_state state)
429 struct timer_list *timer = addr;
432 case ODEBUG_STATE_ACTIVE:
433 del_timer_sync(timer);
434 debug_object_init(timer, &timer_debug_descr);
442 * fixup_activate is called when:
443 * - an active object is activated
444 * - an unknown object is activated (might be a statically initialized object)
446 static int timer_fixup_activate(void *addr, enum debug_obj_state state)
448 struct timer_list *timer = addr;
452 case ODEBUG_STATE_NOTAVAILABLE:
454 * This is not really a fixup. The timer was
455 * statically initialized. We just make sure that it
456 * is tracked in the object tracker.
458 if (timer->entry.next == NULL &&
459 timer->entry.prev == TIMER_ENTRY_STATIC) {
460 debug_object_init(timer, &timer_debug_descr);
461 debug_object_activate(timer, &timer_debug_descr);
468 case ODEBUG_STATE_ACTIVE:
477 * fixup_free is called when:
478 * - an active object is freed
480 static int timer_fixup_free(void *addr, enum debug_obj_state state)
482 struct timer_list *timer = addr;
485 case ODEBUG_STATE_ACTIVE:
486 del_timer_sync(timer);
487 debug_object_free(timer, &timer_debug_descr);
494 static struct debug_obj_descr timer_debug_descr = {
495 .name = "timer_list",
496 .fixup_init = timer_fixup_init,
497 .fixup_activate = timer_fixup_activate,
498 .fixup_free = timer_fixup_free,
501 static inline void debug_timer_init(struct timer_list *timer)
503 debug_object_init(timer, &timer_debug_descr);
506 static inline void debug_timer_activate(struct timer_list *timer)
508 debug_object_activate(timer, &timer_debug_descr);
511 static inline void debug_timer_deactivate(struct timer_list *timer)
513 debug_object_deactivate(timer, &timer_debug_descr);
516 static inline void debug_timer_free(struct timer_list *timer)
518 debug_object_free(timer, &timer_debug_descr);
521 static void __init_timer(struct timer_list *timer,
523 struct lock_class_key *key);
525 void init_timer_on_stack_key(struct timer_list *timer,
527 struct lock_class_key *key)
529 debug_object_init_on_stack(timer, &timer_debug_descr);
530 __init_timer(timer, name, key);
532 EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
534 void destroy_timer_on_stack(struct timer_list *timer)
536 debug_object_free(timer, &timer_debug_descr);
538 EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
541 static inline void debug_timer_init(struct timer_list *timer) { }
542 static inline void debug_timer_activate(struct timer_list *timer) { }
543 static inline void debug_timer_deactivate(struct timer_list *timer) { }
546 static inline void debug_init(struct timer_list *timer)
548 debug_timer_init(timer);
549 trace_timer_init(timer);
553 debug_activate(struct timer_list *timer, unsigned long expires)
555 debug_timer_activate(timer);
556 trace_timer_start(timer, expires);
559 static inline void debug_deactivate(struct timer_list *timer)
561 debug_timer_deactivate(timer);
562 trace_timer_cancel(timer);
565 static void __init_timer(struct timer_list *timer,
567 struct lock_class_key *key)
569 timer->entry.next = NULL;
570 timer->base = __raw_get_cpu_var(tvec_bases);
572 #ifdef CONFIG_TIMER_STATS
573 timer->start_site = NULL;
574 timer->start_pid = -1;
575 memset(timer->start_comm, 0, TASK_COMM_LEN);
577 lockdep_init_map(&timer->lockdep_map, name, key, 0);
580 void setup_deferrable_timer_on_stack_key(struct timer_list *timer,
582 struct lock_class_key *key,
583 void (*function)(unsigned long),
586 timer->function = function;
588 init_timer_on_stack_key(timer, name, key);
589 timer_set_deferrable(timer);
591 EXPORT_SYMBOL_GPL(setup_deferrable_timer_on_stack_key);
594 * init_timer_key - initialize a timer
595 * @timer: the timer to be initialized
596 * @name: name of the timer
597 * @key: lockdep class key of the fake lock used for tracking timer
598 * sync lock dependencies
600 * init_timer_key() must be done to a timer prior calling *any* of the
601 * other timer functions.
603 void init_timer_key(struct timer_list *timer,
605 struct lock_class_key *key)
608 __init_timer(timer, name, key);
610 EXPORT_SYMBOL(init_timer_key);
612 void init_timer_deferrable_key(struct timer_list *timer,
614 struct lock_class_key *key)
616 init_timer_key(timer, name, key);
617 timer_set_deferrable(timer);
619 EXPORT_SYMBOL(init_timer_deferrable_key);
621 static inline void detach_timer(struct timer_list *timer,
624 struct list_head *entry = &timer->entry;
626 debug_deactivate(timer);
628 __list_del(entry->prev, entry->next);
631 entry->prev = LIST_POISON2;
635 * We are using hashed locking: holding per_cpu(tvec_bases).lock
636 * means that all timers which are tied to this base via timer->base are
637 * locked, and the base itself is locked too.
639 * So __run_timers/migrate_timers can safely modify all timers which could
640 * be found on ->tvX lists.
642 * When the timer's base is locked, and the timer removed from list, it is
643 * possible to set timer->base = NULL and drop the lock: the timer remains
646 static struct tvec_base *lock_timer_base(struct timer_list *timer,
647 unsigned long *flags)
648 __acquires(timer->base->lock)
650 struct tvec_base *base;
653 struct tvec_base *prelock_base = timer->base;
654 base = tbase_get_base(prelock_base);
655 if (likely(base != NULL)) {
656 spin_lock_irqsave(&base->lock, *flags);
657 if (likely(prelock_base == timer->base))
659 /* The timer has migrated to another CPU */
660 spin_unlock_irqrestore(&base->lock, *flags);
667 __mod_timer(struct timer_list *timer, unsigned long expires,
668 bool pending_only, int pinned)
670 struct tvec_base *base, *new_base;
674 timer_stats_timer_set_start_info(timer);
675 BUG_ON(!timer->function);
677 base = lock_timer_base(timer, &flags);
679 if (timer_pending(timer)) {
680 detach_timer(timer, 0);
681 if (timer->expires == base->next_timer &&
682 !tbase_get_deferrable(timer->base))
683 base->next_timer = base->timer_jiffies;
690 debug_activate(timer, expires);
692 cpu = smp_processor_id();
694 #if defined(CONFIG_NO_HZ) && defined(CONFIG_SMP)
695 if (!pinned && get_sysctl_timer_migration() && idle_cpu(cpu))
696 cpu = get_nohz_timer_target();
698 new_base = per_cpu(tvec_bases, cpu);
700 if (base != new_base) {
702 * We are trying to schedule the timer on the local CPU.
703 * However we can't change timer's base while it is running,
704 * otherwise del_timer_sync() can't detect that the timer's
705 * handler yet has not finished. This also guarantees that
706 * the timer is serialized wrt itself.
708 if (likely(base->running_timer != timer)) {
709 /* See the comment in lock_timer_base() */
710 timer_set_base(timer, NULL);
711 spin_unlock(&base->lock);
713 spin_lock(&base->lock);
714 timer_set_base(timer, base);
718 timer->expires = expires;
719 if (time_before(timer->expires, base->next_timer) &&
720 !tbase_get_deferrable(timer->base))
721 base->next_timer = timer->expires;
722 internal_add_timer(base, timer);
725 spin_unlock_irqrestore(&base->lock, flags);
731 * mod_timer_pending - modify a pending timer's timeout
732 * @timer: the pending timer to be modified
733 * @expires: new timeout in jiffies
735 * mod_timer_pending() is the same for pending timers as mod_timer(),
736 * but will not re-activate and modify already deleted timers.
738 * It is useful for unserialized use of timers.
740 int mod_timer_pending(struct timer_list *timer, unsigned long expires)
742 return __mod_timer(timer, expires, true, TIMER_NOT_PINNED);
744 EXPORT_SYMBOL(mod_timer_pending);
747 * Decide where to put the timer while taking the slack into account
750 * 1) calculate the maximum (absolute) time
751 * 2) calculate the highest bit where the expires and new max are different
752 * 3) use this bit to make a mask
753 * 4) use the bitmask to round down the maximum time, so that all last
757 unsigned long apply_slack(struct timer_list *timer, unsigned long expires)
759 unsigned long expires_limit, mask;
762 expires_limit = expires;
764 if (timer->slack >= 0) {
765 expires_limit = expires + timer->slack;
767 unsigned long now = jiffies;
769 /* No slack, if already expired else auto slack 0.4% */
770 if (time_after(expires, now))
771 expires_limit = expires + (expires - now)/256;
773 mask = expires ^ expires_limit;
777 bit = find_last_bit(&mask, BITS_PER_LONG);
779 mask = (1 << bit) - 1;
781 expires_limit = expires_limit & ~(mask);
783 return expires_limit;
787 * mod_timer - modify a timer's timeout
788 * @timer: the timer to be modified
789 * @expires: new timeout in jiffies
791 * mod_timer() is a more efficient way to update the expire field of an
792 * active timer (if the timer is inactive it will be activated)
794 * mod_timer(timer, expires) is equivalent to:
796 * del_timer(timer); timer->expires = expires; add_timer(timer);
798 * Note that if there are multiple unserialized concurrent users of the
799 * same timer, then mod_timer() is the only safe way to modify the timeout,
800 * since add_timer() cannot modify an already running timer.
802 * The function returns whether it has modified a pending timer or not.
803 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
804 * active timer returns 1.)
806 int mod_timer(struct timer_list *timer, unsigned long expires)
809 * This is a common optimization triggered by the
810 * networking code - if the timer is re-modified
811 * to be the same thing then just return:
813 if (timer_pending(timer) && timer->expires == expires)
816 expires = apply_slack(timer, expires);
818 return __mod_timer(timer, expires, false, TIMER_NOT_PINNED);
820 EXPORT_SYMBOL(mod_timer);
823 * mod_timer_pinned - modify a timer's timeout
824 * @timer: the timer to be modified
825 * @expires: new timeout in jiffies
827 * mod_timer_pinned() is a way to update the expire field of an
828 * active timer (if the timer is inactive it will be activated)
829 * and not allow the timer to be migrated to a different CPU.
831 * mod_timer_pinned(timer, expires) is equivalent to:
833 * del_timer(timer); timer->expires = expires; add_timer(timer);
835 int mod_timer_pinned(struct timer_list *timer, unsigned long expires)
837 if (timer->expires == expires && timer_pending(timer))
840 return __mod_timer(timer, expires, false, TIMER_PINNED);
842 EXPORT_SYMBOL(mod_timer_pinned);
845 * add_timer - start a timer
846 * @timer: the timer to be added
848 * The kernel will do a ->function(->data) callback from the
849 * timer interrupt at the ->expires point in the future. The
850 * current time is 'jiffies'.
852 * The timer's ->expires, ->function (and if the handler uses it, ->data)
853 * fields must be set prior calling this function.
855 * Timers with an ->expires field in the past will be executed in the next
858 void add_timer(struct timer_list *timer)
860 BUG_ON(timer_pending(timer));
861 mod_timer(timer, timer->expires);
863 EXPORT_SYMBOL(add_timer);
866 * add_timer_on - start a timer on a particular CPU
867 * @timer: the timer to be added
868 * @cpu: the CPU to start it on
870 * This is not very scalable on SMP. Double adds are not possible.
872 void add_timer_on(struct timer_list *timer, int cpu)
874 struct tvec_base *base = per_cpu(tvec_bases, cpu);
877 timer_stats_timer_set_start_info(timer);
878 BUG_ON(timer_pending(timer) || !timer->function);
879 spin_lock_irqsave(&base->lock, flags);
880 timer_set_base(timer, base);
881 debug_activate(timer, timer->expires);
882 if (time_before(timer->expires, base->next_timer) &&
883 !tbase_get_deferrable(timer->base))
884 base->next_timer = timer->expires;
885 internal_add_timer(base, timer);
887 * Check whether the other CPU is idle and needs to be
888 * triggered to reevaluate the timer wheel when nohz is
889 * active. We are protected against the other CPU fiddling
890 * with the timer by holding the timer base lock. This also
891 * makes sure that a CPU on the way to idle can not evaluate
894 wake_up_idle_cpu(cpu);
895 spin_unlock_irqrestore(&base->lock, flags);
897 EXPORT_SYMBOL_GPL(add_timer_on);
900 * del_timer - deactive a timer.
901 * @timer: the timer to be deactivated
903 * del_timer() deactivates a timer - this works on both active and inactive
906 * The function returns whether it has deactivated a pending timer or not.
907 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
908 * active timer returns 1.)
910 int del_timer(struct timer_list *timer)
912 struct tvec_base *base;
916 timer_stats_timer_clear_start_info(timer);
917 if (timer_pending(timer)) {
918 base = lock_timer_base(timer, &flags);
919 if (timer_pending(timer)) {
920 detach_timer(timer, 1);
921 if (timer->expires == base->next_timer &&
922 !tbase_get_deferrable(timer->base))
923 base->next_timer = base->timer_jiffies;
926 spin_unlock_irqrestore(&base->lock, flags);
931 EXPORT_SYMBOL(del_timer);
935 * try_to_del_timer_sync - Try to deactivate a timer
936 * @timer: timer do del
938 * This function tries to deactivate a timer. Upon successful (ret >= 0)
939 * exit the timer is not queued and the handler is not running on any CPU.
941 * It must not be called from interrupt contexts.
943 int try_to_del_timer_sync(struct timer_list *timer)
945 struct tvec_base *base;
949 base = lock_timer_base(timer, &flags);
951 if (base->running_timer == timer)
954 timer_stats_timer_clear_start_info(timer);
956 if (timer_pending(timer)) {
957 detach_timer(timer, 1);
958 if (timer->expires == base->next_timer &&
959 !tbase_get_deferrable(timer->base))
960 base->next_timer = base->timer_jiffies;
964 spin_unlock_irqrestore(&base->lock, flags);
968 EXPORT_SYMBOL(try_to_del_timer_sync);
971 * del_timer_sync - deactivate a timer and wait for the handler to finish.
972 * @timer: the timer to be deactivated
974 * This function only differs from del_timer() on SMP: besides deactivating
975 * the timer it also makes sure the handler has finished executing on other
978 * Synchronization rules: Callers must prevent restarting of the timer,
979 * otherwise this function is meaningless. It must not be called from
980 * interrupt contexts. The caller must not hold locks which would prevent
981 * completion of the timer's handler. The timer's handler must not call
982 * add_timer_on(). Upon exit the timer is not queued and the handler is
983 * not running on any CPU.
985 * The function returns whether it has deactivated a pending timer or not.
987 int del_timer_sync(struct timer_list *timer)
989 #ifdef CONFIG_LOCKDEP
992 local_irq_save(flags);
993 lock_map_acquire(&timer->lockdep_map);
994 lock_map_release(&timer->lockdep_map);
995 local_irq_restore(flags);
999 int ret = try_to_del_timer_sync(timer);
1005 EXPORT_SYMBOL(del_timer_sync);
1008 static int cascade(struct tvec_base *base, struct tvec *tv, int index)
1010 /* cascade all the timers from tv up one level */
1011 struct timer_list *timer, *tmp;
1012 struct list_head tv_list;
1014 list_replace_init(tv->vec + index, &tv_list);
1017 * We are removing _all_ timers from the list, so we
1018 * don't have to detach them individually.
1020 list_for_each_entry_safe(timer, tmp, &tv_list, entry) {
1021 BUG_ON(tbase_get_base(timer->base) != base);
1022 internal_add_timer(base, timer);
1028 static void call_timer_fn(struct timer_list *timer, void (*fn)(unsigned long),
1031 int preempt_count = preempt_count();
1033 #ifdef CONFIG_LOCKDEP
1035 * It is permissible to free the timer from inside the
1036 * function that is called from it, this we need to take into
1037 * account for lockdep too. To avoid bogus "held lock freed"
1038 * warnings as well as problems when looking into
1039 * timer->lockdep_map, make a copy and use that here.
1041 struct lockdep_map lockdep_map = timer->lockdep_map;
1044 * Couple the lock chain with the lock chain at
1045 * del_timer_sync() by acquiring the lock_map around the fn()
1046 * call here and in del_timer_sync().
1048 lock_map_acquire(&lockdep_map);
1050 trace_timer_expire_entry(timer);
1052 trace_timer_expire_exit(timer);
1054 lock_map_release(&lockdep_map);
1056 if (preempt_count != preempt_count()) {
1057 WARN_ONCE(1, "timer: %pF preempt leak: %08x -> %08x\n",
1058 fn, preempt_count, preempt_count());
1060 * Restore the preempt count. That gives us a decent
1061 * chance to survive and extract information. If the
1062 * callback kept a lock held, bad luck, but not worse
1063 * than the BUG() we had.
1065 preempt_count() = preempt_count;
1069 #define INDEX(N) ((base->timer_jiffies >> (TVR_BITS + (N) * TVN_BITS)) & TVN_MASK)
1072 * __run_timers - run all expired timers (if any) on this CPU.
1073 * @base: the timer vector to be processed.
1075 * This function cascades all vectors and executes all expired timer
1078 static inline void __run_timers(struct tvec_base *base)
1080 struct timer_list *timer;
1082 spin_lock_irq(&base->lock);
1083 while (time_after_eq(jiffies, base->timer_jiffies)) {
1084 struct list_head work_list;
1085 struct list_head *head = &work_list;
1086 int index = base->timer_jiffies & TVR_MASK;
1092 (!cascade(base, &base->tv2, INDEX(0))) &&
1093 (!cascade(base, &base->tv3, INDEX(1))) &&
1094 !cascade(base, &base->tv4, INDEX(2)))
1095 cascade(base, &base->tv5, INDEX(3));
1096 ++base->timer_jiffies;
1097 list_replace_init(base->tv1.vec + index, &work_list);
1098 while (!list_empty(head)) {
1099 void (*fn)(unsigned long);
1102 timer = list_first_entry(head, struct timer_list,entry);
1103 fn = timer->function;
1106 timer_stats_account_timer(timer);
1108 set_running_timer(base, timer);
1109 detach_timer(timer, 1);
1111 spin_unlock_irq(&base->lock);
1112 call_timer_fn(timer, fn, data);
1113 spin_lock_irq(&base->lock);
1116 set_running_timer(base, NULL);
1117 spin_unlock_irq(&base->lock);
1122 * Find out when the next timer event is due to happen. This
1123 * is used on S/390 to stop all activity when a CPU is idle.
1124 * This function needs to be called with interrupts disabled.
1126 static unsigned long __next_timer_interrupt(struct tvec_base *base)
1128 unsigned long timer_jiffies = base->timer_jiffies;
1129 unsigned long expires = timer_jiffies + NEXT_TIMER_MAX_DELTA;
1130 int index, slot, array, found = 0;
1131 struct timer_list *nte;
1132 struct tvec *varray[4];
1134 /* Look for timer events in tv1. */
1135 index = slot = timer_jiffies & TVR_MASK;
1137 list_for_each_entry(nte, base->tv1.vec + slot, entry) {
1138 if (tbase_get_deferrable(nte->base))
1142 expires = nte->expires;
1143 /* Look at the cascade bucket(s)? */
1144 if (!index || slot < index)
1148 slot = (slot + 1) & TVR_MASK;
1149 } while (slot != index);
1152 /* Calculate the next cascade event */
1154 timer_jiffies += TVR_SIZE - index;
1155 timer_jiffies >>= TVR_BITS;
1157 /* Check tv2-tv5. */
1158 varray[0] = &base->tv2;
1159 varray[1] = &base->tv3;
1160 varray[2] = &base->tv4;
1161 varray[3] = &base->tv5;
1163 for (array = 0; array < 4; array++) {
1164 struct tvec *varp = varray[array];
1166 index = slot = timer_jiffies & TVN_MASK;
1168 list_for_each_entry(nte, varp->vec + slot, entry) {
1169 if (tbase_get_deferrable(nte->base))
1173 if (time_before(nte->expires, expires))
1174 expires = nte->expires;
1177 * Do we still search for the first timer or are
1178 * we looking up the cascade buckets ?
1181 /* Look at the cascade bucket(s)? */
1182 if (!index || slot < index)
1186 slot = (slot + 1) & TVN_MASK;
1187 } while (slot != index);
1190 timer_jiffies += TVN_SIZE - index;
1191 timer_jiffies >>= TVN_BITS;
1197 * Check, if the next hrtimer event is before the next timer wheel
1200 static unsigned long cmp_next_hrtimer_event(unsigned long now,
1201 unsigned long expires)
1203 ktime_t hr_delta = hrtimer_get_next_event();
1204 struct timespec tsdelta;
1205 unsigned long delta;
1207 if (hr_delta.tv64 == KTIME_MAX)
1211 * Expired timer available, let it expire in the next tick
1213 if (hr_delta.tv64 <= 0)
1216 tsdelta = ktime_to_timespec(hr_delta);
1217 delta = timespec_to_jiffies(&tsdelta);
1220 * Limit the delta to the max value, which is checked in
1221 * tick_nohz_stop_sched_tick():
1223 if (delta > NEXT_TIMER_MAX_DELTA)
1224 delta = NEXT_TIMER_MAX_DELTA;
1227 * Take rounding errors in to account and make sure, that it
1228 * expires in the next tick. Otherwise we go into an endless
1229 * ping pong due to tick_nohz_stop_sched_tick() retriggering
1235 if (time_before(now, expires))
1241 * get_next_timer_interrupt - return the jiffy of the next pending timer
1242 * @now: current time (in jiffies)
1244 unsigned long get_next_timer_interrupt(unsigned long now)
1246 struct tvec_base *base = __get_cpu_var(tvec_bases);
1247 unsigned long expires;
1249 spin_lock(&base->lock);
1250 if (time_before_eq(base->next_timer, base->timer_jiffies))
1251 base->next_timer = __next_timer_interrupt(base);
1252 expires = base->next_timer;
1253 spin_unlock(&base->lock);
1255 if (time_before_eq(expires, now))
1258 return cmp_next_hrtimer_event(now, expires);
1263 * Called from the timer interrupt handler to charge one tick to the current
1264 * process. user_tick is 1 if the tick is user time, 0 for system.
1266 void update_process_times(int user_tick)
1268 struct task_struct *p = current;
1269 int cpu = smp_processor_id();
1271 /* Note: this timer irq context must be accounted for as well. */
1272 account_process_tick(p, user_tick);
1274 rcu_check_callbacks(cpu, user_tick);
1276 perf_event_do_pending();
1278 run_posix_cpu_timers(p);
1282 * This function runs timers and the timer-tq in bottom half context.
1284 static void run_timer_softirq(struct softirq_action *h)
1286 struct tvec_base *base = __get_cpu_var(tvec_bases);
1288 hrtimer_run_pending();
1290 if (time_after_eq(jiffies, base->timer_jiffies))
1295 * Called by the local, per-CPU timer interrupt on SMP.
1297 void run_local_timers(void)
1299 hrtimer_run_queues();
1300 raise_softirq(TIMER_SOFTIRQ);
1304 * The 64-bit jiffies value is not atomic - you MUST NOT read it
1305 * without sampling the sequence number in xtime_lock.
1306 * jiffies is defined in the linker script...
1309 void do_timer(unsigned long ticks)
1311 jiffies_64 += ticks;
1316 #ifdef __ARCH_WANT_SYS_ALARM
1319 * For backwards compatibility? This can be done in libc so Alpha
1320 * and all newer ports shouldn't need it.
1322 SYSCALL_DEFINE1(alarm, unsigned int, seconds)
1324 return alarm_setitimer(seconds);
1332 * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this
1333 * should be moved into arch/i386 instead?
1337 * sys_getpid - return the thread group id of the current process
1339 * Note, despite the name, this returns the tgid not the pid. The tgid and
1340 * the pid are identical unless CLONE_THREAD was specified on clone() in
1341 * which case the tgid is the same in all threads of the same group.
1343 * This is SMP safe as current->tgid does not change.
1345 SYSCALL_DEFINE0(getpid)
1347 return task_tgid_vnr(current);
1351 * Accessing ->real_parent is not SMP-safe, it could
1352 * change from under us. However, we can use a stale
1353 * value of ->real_parent under rcu_read_lock(), see
1354 * release_task()->call_rcu(delayed_put_task_struct).
1356 SYSCALL_DEFINE0(getppid)
1361 pid = task_tgid_vnr(current->real_parent);
1367 SYSCALL_DEFINE0(getuid)
1369 /* Only we change this so SMP safe */
1370 return current_uid();
1373 SYSCALL_DEFINE0(geteuid)
1375 /* Only we change this so SMP safe */
1376 return current_euid();
1379 SYSCALL_DEFINE0(getgid)
1381 /* Only we change this so SMP safe */
1382 return current_gid();
1385 SYSCALL_DEFINE0(getegid)
1387 /* Only we change this so SMP safe */
1388 return current_egid();
1393 static void process_timeout(unsigned long __data)
1395 wake_up_process((struct task_struct *)__data);
1399 * schedule_timeout - sleep until timeout
1400 * @timeout: timeout value in jiffies
1402 * Make the current task sleep until @timeout jiffies have
1403 * elapsed. The routine will return immediately unless
1404 * the current task state has been set (see set_current_state()).
1406 * You can set the task state as follows -
1408 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1409 * pass before the routine returns. The routine will return 0
1411 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1412 * delivered to the current task. In this case the remaining time
1413 * in jiffies will be returned, or 0 if the timer expired in time
1415 * The current task state is guaranteed to be TASK_RUNNING when this
1418 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1419 * the CPU away without a bound on the timeout. In this case the return
1420 * value will be %MAX_SCHEDULE_TIMEOUT.
1422 * In all cases the return value is guaranteed to be non-negative.
1424 signed long __sched schedule_timeout(signed long timeout)
1426 struct timer_list timer;
1427 unsigned long expire;
1431 case MAX_SCHEDULE_TIMEOUT:
1433 * These two special cases are useful to be comfortable
1434 * in the caller. Nothing more. We could take
1435 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1436 * but I' d like to return a valid offset (>=0) to allow
1437 * the caller to do everything it want with the retval.
1443 * Another bit of PARANOID. Note that the retval will be
1444 * 0 since no piece of kernel is supposed to do a check
1445 * for a negative retval of schedule_timeout() (since it
1446 * should never happens anyway). You just have the printk()
1447 * that will tell you if something is gone wrong and where.
1450 printk(KERN_ERR "schedule_timeout: wrong timeout "
1451 "value %lx\n", timeout);
1453 current->state = TASK_RUNNING;
1458 expire = timeout + jiffies;
1460 setup_timer_on_stack(&timer, process_timeout, (unsigned long)current);
1461 __mod_timer(&timer, expire, false, TIMER_NOT_PINNED);
1463 del_singleshot_timer_sync(&timer);
1465 /* Remove the timer from the object tracker */
1466 destroy_timer_on_stack(&timer);
1468 timeout = expire - jiffies;
1471 return timeout < 0 ? 0 : timeout;
1473 EXPORT_SYMBOL(schedule_timeout);
1476 * We can use __set_current_state() here because schedule_timeout() calls
1477 * schedule() unconditionally.
1479 signed long __sched schedule_timeout_interruptible(signed long timeout)
1481 __set_current_state(TASK_INTERRUPTIBLE);
1482 return schedule_timeout(timeout);
1484 EXPORT_SYMBOL(schedule_timeout_interruptible);
1486 signed long __sched schedule_timeout_killable(signed long timeout)
1488 __set_current_state(TASK_KILLABLE);
1489 return schedule_timeout(timeout);
1491 EXPORT_SYMBOL(schedule_timeout_killable);
1493 signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1495 __set_current_state(TASK_UNINTERRUPTIBLE);
1496 return schedule_timeout(timeout);
1498 EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1500 /* Thread ID - the internal kernel "pid" */
1501 SYSCALL_DEFINE0(gettid)
1503 return task_pid_vnr(current);
1507 * do_sysinfo - fill in sysinfo struct
1508 * @info: pointer to buffer to fill
1510 int do_sysinfo(struct sysinfo *info)
1512 unsigned long mem_total, sav_total;
1513 unsigned int mem_unit, bitcount;
1516 memset(info, 0, sizeof(struct sysinfo));
1519 monotonic_to_bootbased(&tp);
1520 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
1522 get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
1524 info->procs = nr_threads;
1530 * If the sum of all the available memory (i.e. ram + swap)
1531 * is less than can be stored in a 32 bit unsigned long then
1532 * we can be binary compatible with 2.2.x kernels. If not,
1533 * well, in that case 2.2.x was broken anyways...
1535 * -Erik Andersen <andersee@debian.org>
1538 mem_total = info->totalram + info->totalswap;
1539 if (mem_total < info->totalram || mem_total < info->totalswap)
1542 mem_unit = info->mem_unit;
1543 while (mem_unit > 1) {
1546 sav_total = mem_total;
1548 if (mem_total < sav_total)
1553 * If mem_total did not overflow, multiply all memory values by
1554 * info->mem_unit and set it to 1. This leaves things compatible
1555 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1560 info->totalram <<= bitcount;
1561 info->freeram <<= bitcount;
1562 info->sharedram <<= bitcount;
1563 info->bufferram <<= bitcount;
1564 info->totalswap <<= bitcount;
1565 info->freeswap <<= bitcount;
1566 info->totalhigh <<= bitcount;
1567 info->freehigh <<= bitcount;
1573 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
1579 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
1585 static int __cpuinit init_timers_cpu(int cpu)
1588 struct tvec_base *base;
1589 static char __cpuinitdata tvec_base_done[NR_CPUS];
1591 if (!tvec_base_done[cpu]) {
1592 static char boot_done;
1596 * The APs use this path later in boot
1598 base = kmalloc_node(sizeof(*base),
1599 GFP_KERNEL | __GFP_ZERO,
1604 /* Make sure that tvec_base is 2 byte aligned */
1605 if (tbase_get_deferrable(base)) {
1610 per_cpu(tvec_bases, cpu) = base;
1613 * This is for the boot CPU - we use compile-time
1614 * static initialisation because per-cpu memory isn't
1615 * ready yet and because the memory allocators are not
1616 * initialised either.
1619 base = &boot_tvec_bases;
1621 tvec_base_done[cpu] = 1;
1623 base = per_cpu(tvec_bases, cpu);
1626 spin_lock_init(&base->lock);
1628 for (j = 0; j < TVN_SIZE; j++) {
1629 INIT_LIST_HEAD(base->tv5.vec + j);
1630 INIT_LIST_HEAD(base->tv4.vec + j);
1631 INIT_LIST_HEAD(base->tv3.vec + j);
1632 INIT_LIST_HEAD(base->tv2.vec + j);
1634 for (j = 0; j < TVR_SIZE; j++)
1635 INIT_LIST_HEAD(base->tv1.vec + j);
1637 base->timer_jiffies = jiffies;
1638 base->next_timer = base->timer_jiffies;
1642 #ifdef CONFIG_HOTPLUG_CPU
1643 static void migrate_timer_list(struct tvec_base *new_base, struct list_head *head)
1645 struct timer_list *timer;
1647 while (!list_empty(head)) {
1648 timer = list_first_entry(head, struct timer_list, entry);
1649 detach_timer(timer, 0);
1650 timer_set_base(timer, new_base);
1651 if (time_before(timer->expires, new_base->next_timer) &&
1652 !tbase_get_deferrable(timer->base))
1653 new_base->next_timer = timer->expires;
1654 internal_add_timer(new_base, timer);
1658 static void __cpuinit migrate_timers(int cpu)
1660 struct tvec_base *old_base;
1661 struct tvec_base *new_base;
1664 BUG_ON(cpu_online(cpu));
1665 old_base = per_cpu(tvec_bases, cpu);
1666 new_base = get_cpu_var(tvec_bases);
1668 * The caller is globally serialized and nobody else
1669 * takes two locks at once, deadlock is not possible.
1671 spin_lock_irq(&new_base->lock);
1672 spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1674 BUG_ON(old_base->running_timer);
1676 for (i = 0; i < TVR_SIZE; i++)
1677 migrate_timer_list(new_base, old_base->tv1.vec + i);
1678 for (i = 0; i < TVN_SIZE; i++) {
1679 migrate_timer_list(new_base, old_base->tv2.vec + i);
1680 migrate_timer_list(new_base, old_base->tv3.vec + i);
1681 migrate_timer_list(new_base, old_base->tv4.vec + i);
1682 migrate_timer_list(new_base, old_base->tv5.vec + i);
1685 spin_unlock(&old_base->lock);
1686 spin_unlock_irq(&new_base->lock);
1687 put_cpu_var(tvec_bases);
1689 #endif /* CONFIG_HOTPLUG_CPU */
1691 static int __cpuinit timer_cpu_notify(struct notifier_block *self,
1692 unsigned long action, void *hcpu)
1694 long cpu = (long)hcpu;
1698 case CPU_UP_PREPARE:
1699 case CPU_UP_PREPARE_FROZEN:
1700 err = init_timers_cpu(cpu);
1702 return notifier_from_errno(err);
1704 #ifdef CONFIG_HOTPLUG_CPU
1706 case CPU_DEAD_FROZEN:
1707 migrate_timers(cpu);
1716 static struct notifier_block __cpuinitdata timers_nb = {
1717 .notifier_call = timer_cpu_notify,
1721 void __init init_timers(void)
1723 int err = timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE,
1724 (void *)(long)smp_processor_id());
1728 BUG_ON(err != NOTIFY_OK);
1729 register_cpu_notifier(&timers_nb);
1730 open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
1734 * msleep - sleep safely even with waitqueue interruptions
1735 * @msecs: Time in milliseconds to sleep for
1737 void msleep(unsigned int msecs)
1739 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1742 timeout = schedule_timeout_uninterruptible(timeout);
1745 EXPORT_SYMBOL(msleep);
1748 * msleep_interruptible - sleep waiting for signals
1749 * @msecs: Time in milliseconds to sleep for
1751 unsigned long msleep_interruptible(unsigned int msecs)
1753 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1755 while (timeout && !signal_pending(current))
1756 timeout = schedule_timeout_interruptible(timeout);
1757 return jiffies_to_msecs(timeout);
1760 EXPORT_SYMBOL(msleep_interruptible);