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>
43 #include <asm/uaccess.h>
44 #include <asm/unistd.h>
45 #include <asm/div64.h>
46 #include <asm/timex.h>
49 #define CREATE_TRACE_POINTS
50 #include <trace/events/timer.h>
52 u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
54 EXPORT_SYMBOL(jiffies_64);
57 * per-CPU timer vector definitions:
59 #define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6)
60 #define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8)
61 #define TVN_SIZE (1 << TVN_BITS)
62 #define TVR_SIZE (1 << TVR_BITS)
63 #define TVN_MASK (TVN_SIZE - 1)
64 #define TVR_MASK (TVR_SIZE - 1)
67 struct list_head vec[TVN_SIZE];
71 struct list_head vec[TVR_SIZE];
76 struct timer_list *running_timer;
77 unsigned long timer_jiffies;
78 unsigned long next_timer;
84 } ____cacheline_aligned;
86 struct tvec_base boot_tvec_bases;
87 EXPORT_SYMBOL(boot_tvec_bases);
88 static DEFINE_PER_CPU(struct tvec_base *, tvec_bases) = &boot_tvec_bases;
91 * Note that all tvec_bases are 2 byte aligned and lower bit of
92 * base in timer_list is guaranteed to be zero. Use the LSB for
93 * the new flag to indicate whether the timer is deferrable
95 #define TBASE_DEFERRABLE_FLAG (0x1)
97 /* Functions below help us manage 'deferrable' flag */
98 static inline unsigned int tbase_get_deferrable(struct tvec_base *base)
100 return ((unsigned int)(unsigned long)base & TBASE_DEFERRABLE_FLAG);
103 static inline struct tvec_base *tbase_get_base(struct tvec_base *base)
105 return ((struct tvec_base *)((unsigned long)base & ~TBASE_DEFERRABLE_FLAG));
108 static inline void timer_set_deferrable(struct timer_list *timer)
110 timer->base = ((struct tvec_base *)((unsigned long)(timer->base) |
111 TBASE_DEFERRABLE_FLAG));
115 timer_set_base(struct timer_list *timer, struct tvec_base *new_base)
117 timer->base = (struct tvec_base *)((unsigned long)(new_base) |
118 tbase_get_deferrable(timer->base));
121 static unsigned long round_jiffies_common(unsigned long j, int cpu,
125 unsigned long original = j;
128 * We don't want all cpus firing their timers at once hitting the
129 * same lock or cachelines, so we skew each extra cpu with an extra
130 * 3 jiffies. This 3 jiffies came originally from the mm/ code which
132 * The skew is done by adding 3*cpunr, then round, then subtract this
133 * extra offset again.
140 * If the target jiffie is just after a whole second (which can happen
141 * due to delays of the timer irq, long irq off times etc etc) then
142 * we should round down to the whole second, not up. Use 1/4th second
143 * as cutoff for this rounding as an extreme upper bound for this.
144 * But never round down if @force_up is set.
146 if (rem < HZ/4 && !force_up) /* round down */
151 /* now that we have rounded, subtract the extra skew again */
154 if (j <= jiffies) /* rounding ate our timeout entirely; */
160 * __round_jiffies - function to round jiffies to a full second
161 * @j: the time in (absolute) jiffies that should be rounded
162 * @cpu: the processor number on which the timeout will happen
164 * __round_jiffies() rounds an absolute time in the future (in jiffies)
165 * up or down to (approximately) full seconds. This is useful for timers
166 * for which the exact time they fire does not matter too much, as long as
167 * they fire approximately every X seconds.
169 * By rounding these timers to whole seconds, all such timers will fire
170 * at the same time, rather than at various times spread out. The goal
171 * of this is to have the CPU wake up less, which saves power.
173 * The exact rounding is skewed for each processor to avoid all
174 * processors firing at the exact same time, which could lead
175 * to lock contention or spurious cache line bouncing.
177 * The return value is the rounded version of the @j parameter.
179 unsigned long __round_jiffies(unsigned long j, int cpu)
181 return round_jiffies_common(j, cpu, false);
183 EXPORT_SYMBOL_GPL(__round_jiffies);
186 * __round_jiffies_relative - function to round jiffies to a full second
187 * @j: the time in (relative) jiffies that should be rounded
188 * @cpu: the processor number on which the timeout will happen
190 * __round_jiffies_relative() rounds a time delta in the future (in jiffies)
191 * up or down to (approximately) full seconds. This is useful for timers
192 * for which the exact time they fire does not matter too much, as long as
193 * they fire approximately every X seconds.
195 * By rounding these timers to whole seconds, all such timers will fire
196 * at the same time, rather than at various times spread out. The goal
197 * of this is to have the CPU wake up less, which saves power.
199 * The exact rounding is skewed for each processor to avoid all
200 * processors firing at the exact same time, which could lead
201 * to lock contention or spurious cache line bouncing.
203 * The return value is the rounded version of the @j parameter.
205 unsigned long __round_jiffies_relative(unsigned long j, int cpu)
207 unsigned long j0 = jiffies;
209 /* Use j0 because jiffies might change while we run */
210 return round_jiffies_common(j + j0, cpu, false) - j0;
212 EXPORT_SYMBOL_GPL(__round_jiffies_relative);
215 * round_jiffies - function to round jiffies to a full second
216 * @j: the time in (absolute) jiffies that should be rounded
218 * round_jiffies() rounds an absolute time in the future (in jiffies)
219 * up or down to (approximately) full seconds. This is useful for timers
220 * for which the exact time they fire does not matter too much, as long as
221 * they fire approximately every X seconds.
223 * By rounding these timers to whole seconds, all such timers will fire
224 * at the same time, rather than at various times spread out. The goal
225 * of this is to have the CPU wake up less, which saves power.
227 * The return value is the rounded version of the @j parameter.
229 unsigned long round_jiffies(unsigned long j)
231 return round_jiffies_common(j, raw_smp_processor_id(), false);
233 EXPORT_SYMBOL_GPL(round_jiffies);
236 * round_jiffies_relative - function to round jiffies to a full second
237 * @j: the time in (relative) jiffies that should be rounded
239 * round_jiffies_relative() rounds a time delta in the future (in jiffies)
240 * up or down to (approximately) full seconds. This is useful for timers
241 * for which the exact time they fire does not matter too much, as long as
242 * they fire approximately every X seconds.
244 * By rounding these timers to whole seconds, all such timers will fire
245 * at the same time, rather than at various times spread out. The goal
246 * of this is to have the CPU wake up less, which saves power.
248 * The return value is the rounded version of the @j parameter.
250 unsigned long round_jiffies_relative(unsigned long j)
252 return __round_jiffies_relative(j, raw_smp_processor_id());
254 EXPORT_SYMBOL_GPL(round_jiffies_relative);
257 * __round_jiffies_up - function to round jiffies up to a full second
258 * @j: the time in (absolute) jiffies that should be rounded
259 * @cpu: the processor number on which the timeout will happen
261 * This is the same as __round_jiffies() except that it will never
262 * round down. This is useful for timeouts for which the exact time
263 * of firing does not matter too much, as long as they don't fire too
266 unsigned long __round_jiffies_up(unsigned long j, int cpu)
268 return round_jiffies_common(j, cpu, true);
270 EXPORT_SYMBOL_GPL(__round_jiffies_up);
273 * __round_jiffies_up_relative - function to round jiffies up to a full second
274 * @j: the time in (relative) jiffies that should be rounded
275 * @cpu: the processor number on which the timeout will happen
277 * This is the same as __round_jiffies_relative() except that it will never
278 * round down. This is useful for timeouts for which the exact time
279 * of firing does not matter too much, as long as they don't fire too
282 unsigned long __round_jiffies_up_relative(unsigned long j, int cpu)
284 unsigned long j0 = jiffies;
286 /* Use j0 because jiffies might change while we run */
287 return round_jiffies_common(j + j0, cpu, true) - j0;
289 EXPORT_SYMBOL_GPL(__round_jiffies_up_relative);
292 * round_jiffies_up - function to round jiffies up to a full second
293 * @j: the time in (absolute) jiffies that should be rounded
295 * This is the same as round_jiffies() except that it will never
296 * round down. This is useful for timeouts for which the exact time
297 * of firing does not matter too much, as long as they don't fire too
300 unsigned long round_jiffies_up(unsigned long j)
302 return round_jiffies_common(j, raw_smp_processor_id(), true);
304 EXPORT_SYMBOL_GPL(round_jiffies_up);
307 * round_jiffies_up_relative - function to round jiffies up to a full second
308 * @j: the time in (relative) jiffies that should be rounded
310 * This is the same as round_jiffies_relative() except that it will never
311 * round down. This is useful for timeouts for which the exact time
312 * of firing does not matter too much, as long as they don't fire too
315 unsigned long round_jiffies_up_relative(unsigned long j)
317 return __round_jiffies_up_relative(j, raw_smp_processor_id());
319 EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
322 * set_timer_slack - set the allowed slack for a timer
323 * @slack_hz: the amount of time (in jiffies) allowed for rounding
325 * Set the amount of time, in jiffies, that a certain timer has
326 * in terms of slack. By setting this value, the timer subsystem
327 * will schedule the actual timer somewhere between
328 * the time mod_timer() asks for, and that time plus the slack.
330 * By setting the slack to -1, a percentage of the delay is used
333 void set_timer_slack(struct timer_list *timer, int slack_hz)
335 timer->slack = slack_hz;
337 EXPORT_SYMBOL_GPL(set_timer_slack);
340 static inline void set_running_timer(struct tvec_base *base,
341 struct timer_list *timer)
344 base->running_timer = timer;
348 static void internal_add_timer(struct tvec_base *base, struct timer_list *timer)
350 unsigned long expires = timer->expires;
351 unsigned long idx = expires - base->timer_jiffies;
352 struct list_head *vec;
354 if (idx < TVR_SIZE) {
355 int i = expires & TVR_MASK;
356 vec = base->tv1.vec + i;
357 } else if (idx < 1 << (TVR_BITS + TVN_BITS)) {
358 int i = (expires >> TVR_BITS) & TVN_MASK;
359 vec = base->tv2.vec + i;
360 } else if (idx < 1 << (TVR_BITS + 2 * TVN_BITS)) {
361 int i = (expires >> (TVR_BITS + TVN_BITS)) & TVN_MASK;
362 vec = base->tv3.vec + i;
363 } else if (idx < 1 << (TVR_BITS + 3 * TVN_BITS)) {
364 int i = (expires >> (TVR_BITS + 2 * TVN_BITS)) & TVN_MASK;
365 vec = base->tv4.vec + i;
366 } else if ((signed long) idx < 0) {
368 * Can happen if you add a timer with expires == jiffies,
369 * or you set a timer to go off in the past
371 vec = base->tv1.vec + (base->timer_jiffies & TVR_MASK);
374 /* If the timeout is larger than 0xffffffff on 64-bit
375 * architectures then we use the maximum timeout:
377 if (idx > 0xffffffffUL) {
379 expires = idx + base->timer_jiffies;
381 i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK;
382 vec = base->tv5.vec + i;
387 list_add_tail(&timer->entry, vec);
390 #ifdef CONFIG_TIMER_STATS
391 void __timer_stats_timer_set_start_info(struct timer_list *timer, void *addr)
393 if (timer->start_site)
396 timer->start_site = addr;
397 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
398 timer->start_pid = current->pid;
401 static void timer_stats_account_timer(struct timer_list *timer)
403 unsigned int flag = 0;
405 if (likely(!timer->start_site))
407 if (unlikely(tbase_get_deferrable(timer->base)))
408 flag |= TIMER_STATS_FLAG_DEFERRABLE;
410 timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
411 timer->function, timer->start_comm, flag);
415 static void timer_stats_account_timer(struct timer_list *timer) {}
418 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
420 static struct debug_obj_descr timer_debug_descr;
423 * fixup_init is called when:
424 * - an active object is initialized
426 static int timer_fixup_init(void *addr, enum debug_obj_state state)
428 struct timer_list *timer = addr;
431 case ODEBUG_STATE_ACTIVE:
432 del_timer_sync(timer);
433 debug_object_init(timer, &timer_debug_descr);
441 * fixup_activate is called when:
442 * - an active object is activated
443 * - an unknown object is activated (might be a statically initialized object)
445 static int timer_fixup_activate(void *addr, enum debug_obj_state state)
447 struct timer_list *timer = addr;
451 case ODEBUG_STATE_NOTAVAILABLE:
453 * This is not really a fixup. The timer was
454 * statically initialized. We just make sure that it
455 * is tracked in the object tracker.
457 if (timer->entry.next == NULL &&
458 timer->entry.prev == TIMER_ENTRY_STATIC) {
459 debug_object_init(timer, &timer_debug_descr);
460 debug_object_activate(timer, &timer_debug_descr);
467 case ODEBUG_STATE_ACTIVE:
476 * fixup_free is called when:
477 * - an active object is freed
479 static int timer_fixup_free(void *addr, enum debug_obj_state state)
481 struct timer_list *timer = addr;
484 case ODEBUG_STATE_ACTIVE:
485 del_timer_sync(timer);
486 debug_object_free(timer, &timer_debug_descr);
493 static struct debug_obj_descr timer_debug_descr = {
494 .name = "timer_list",
495 .fixup_init = timer_fixup_init,
496 .fixup_activate = timer_fixup_activate,
497 .fixup_free = timer_fixup_free,
500 static inline void debug_timer_init(struct timer_list *timer)
502 debug_object_init(timer, &timer_debug_descr);
505 static inline void debug_timer_activate(struct timer_list *timer)
507 debug_object_activate(timer, &timer_debug_descr);
510 static inline void debug_timer_deactivate(struct timer_list *timer)
512 debug_object_deactivate(timer, &timer_debug_descr);
515 static inline void debug_timer_free(struct timer_list *timer)
517 debug_object_free(timer, &timer_debug_descr);
520 static void __init_timer(struct timer_list *timer,
522 struct lock_class_key *key);
524 void init_timer_on_stack_key(struct timer_list *timer,
526 struct lock_class_key *key)
528 debug_object_init_on_stack(timer, &timer_debug_descr);
529 __init_timer(timer, name, key);
531 EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
533 void destroy_timer_on_stack(struct timer_list *timer)
535 debug_object_free(timer, &timer_debug_descr);
537 EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
540 static inline void debug_timer_init(struct timer_list *timer) { }
541 static inline void debug_timer_activate(struct timer_list *timer) { }
542 static inline void debug_timer_deactivate(struct timer_list *timer) { }
545 static inline void debug_init(struct timer_list *timer)
547 debug_timer_init(timer);
548 trace_timer_init(timer);
552 debug_activate(struct timer_list *timer, unsigned long expires)
554 debug_timer_activate(timer);
555 trace_timer_start(timer, expires);
558 static inline void debug_deactivate(struct timer_list *timer)
560 debug_timer_deactivate(timer);
561 trace_timer_cancel(timer);
564 static void __init_timer(struct timer_list *timer,
566 struct lock_class_key *key)
568 timer->entry.next = NULL;
569 timer->base = __raw_get_cpu_var(tvec_bases);
571 #ifdef CONFIG_TIMER_STATS
572 timer->start_site = NULL;
573 timer->start_pid = -1;
574 memset(timer->start_comm, 0, TASK_COMM_LEN);
576 lockdep_init_map(&timer->lockdep_map, name, key, 0);
580 * init_timer_key - initialize a timer
581 * @timer: the timer to be initialized
582 * @name: name of the timer
583 * @key: lockdep class key of the fake lock used for tracking timer
584 * sync lock dependencies
586 * init_timer_key() must be done to a timer prior calling *any* of the
587 * other timer functions.
589 void init_timer_key(struct timer_list *timer,
591 struct lock_class_key *key)
594 __init_timer(timer, name, key);
596 EXPORT_SYMBOL(init_timer_key);
598 void init_timer_deferrable_key(struct timer_list *timer,
600 struct lock_class_key *key)
602 init_timer_key(timer, name, key);
603 timer_set_deferrable(timer);
605 EXPORT_SYMBOL(init_timer_deferrable_key);
607 static inline void detach_timer(struct timer_list *timer,
610 struct list_head *entry = &timer->entry;
612 debug_deactivate(timer);
614 __list_del(entry->prev, entry->next);
617 entry->prev = LIST_POISON2;
621 * We are using hashed locking: holding per_cpu(tvec_bases).lock
622 * means that all timers which are tied to this base via timer->base are
623 * locked, and the base itself is locked too.
625 * So __run_timers/migrate_timers can safely modify all timers which could
626 * be found on ->tvX lists.
628 * When the timer's base is locked, and the timer removed from list, it is
629 * possible to set timer->base = NULL and drop the lock: the timer remains
632 static struct tvec_base *lock_timer_base(struct timer_list *timer,
633 unsigned long *flags)
634 __acquires(timer->base->lock)
636 struct tvec_base *base;
639 struct tvec_base *prelock_base = timer->base;
640 base = tbase_get_base(prelock_base);
641 if (likely(base != NULL)) {
642 spin_lock_irqsave(&base->lock, *flags);
643 if (likely(prelock_base == timer->base))
645 /* The timer has migrated to another CPU */
646 spin_unlock_irqrestore(&base->lock, *flags);
653 __mod_timer(struct timer_list *timer, unsigned long expires,
654 bool pending_only, int pinned)
656 struct tvec_base *base, *new_base;
660 timer_stats_timer_set_start_info(timer);
661 BUG_ON(!timer->function);
663 base = lock_timer_base(timer, &flags);
665 if (timer_pending(timer)) {
666 detach_timer(timer, 0);
667 if (timer->expires == base->next_timer &&
668 !tbase_get_deferrable(timer->base))
669 base->next_timer = base->timer_jiffies;
676 debug_activate(timer, expires);
678 cpu = smp_processor_id();
680 #if defined(CONFIG_NO_HZ) && defined(CONFIG_SMP)
681 if (!pinned && get_sysctl_timer_migration() && idle_cpu(cpu)) {
682 int preferred_cpu = get_nohz_load_balancer();
684 if (preferred_cpu >= 0)
688 new_base = per_cpu(tvec_bases, cpu);
690 if (base != new_base) {
692 * We are trying to schedule the timer on the local CPU.
693 * However we can't change timer's base while it is running,
694 * otherwise del_timer_sync() can't detect that the timer's
695 * handler yet has not finished. This also guarantees that
696 * the timer is serialized wrt itself.
698 if (likely(base->running_timer != timer)) {
699 /* See the comment in lock_timer_base() */
700 timer_set_base(timer, NULL);
701 spin_unlock(&base->lock);
703 spin_lock(&base->lock);
704 timer_set_base(timer, base);
708 timer->expires = expires;
709 if (time_before(timer->expires, base->next_timer) &&
710 !tbase_get_deferrable(timer->base))
711 base->next_timer = timer->expires;
712 internal_add_timer(base, timer);
715 spin_unlock_irqrestore(&base->lock, flags);
721 * mod_timer_pending - modify a pending timer's timeout
722 * @timer: the pending timer to be modified
723 * @expires: new timeout in jiffies
725 * mod_timer_pending() is the same for pending timers as mod_timer(),
726 * but will not re-activate and modify already deleted timers.
728 * It is useful for unserialized use of timers.
730 int mod_timer_pending(struct timer_list *timer, unsigned long expires)
732 return __mod_timer(timer, expires, true, TIMER_NOT_PINNED);
734 EXPORT_SYMBOL(mod_timer_pending);
737 * Decide where to put the timer while taking the slack into account
740 * 1) calculate the maximum (absolute) time
741 * 2) calculate the highest bit where the expires and new max are different
742 * 3) use this bit to make a mask
743 * 4) use the bitmask to round down the maximum time, so that all last
747 unsigned long apply_slack(struct timer_list *timer, unsigned long expires)
749 unsigned long expires_limit, mask;
752 expires_limit = expires + timer->slack;
754 if (timer->slack < 0) /* auto slack: use 0.4% */
755 expires_limit = expires + (expires - jiffies)/256;
757 mask = expires ^ expires_limit;
762 bit = find_last_bit(&mask, BITS_PER_LONG);
764 mask = (1 << bit) - 1;
766 expires_limit = expires_limit & ~(mask);
768 return expires_limit;
772 * mod_timer - modify a timer's timeout
773 * @timer: the timer to be modified
774 * @expires: new timeout in jiffies
776 * mod_timer() is a more efficient way to update the expire field of an
777 * active timer (if the timer is inactive it will be activated)
779 * mod_timer(timer, expires) is equivalent to:
781 * del_timer(timer); timer->expires = expires; add_timer(timer);
783 * Note that if there are multiple unserialized concurrent users of the
784 * same timer, then mod_timer() is the only safe way to modify the timeout,
785 * since add_timer() cannot modify an already running timer.
787 * The function returns whether it has modified a pending timer or not.
788 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
789 * active timer returns 1.)
791 int mod_timer(struct timer_list *timer, unsigned long expires)
794 * This is a common optimization triggered by the
795 * networking code - if the timer is re-modified
796 * to be the same thing then just return:
798 if (timer_pending(timer) && timer->expires == expires)
801 expires = apply_slack(timer, expires);
803 return __mod_timer(timer, expires, false, TIMER_NOT_PINNED);
805 EXPORT_SYMBOL(mod_timer);
808 * mod_timer_pinned - modify a timer's timeout
809 * @timer: the timer to be modified
810 * @expires: new timeout in jiffies
812 * mod_timer_pinned() is a way to update the expire field of an
813 * active timer (if the timer is inactive it will be activated)
814 * and not allow the timer to be migrated to a different CPU.
816 * mod_timer_pinned(timer, expires) is equivalent to:
818 * del_timer(timer); timer->expires = expires; add_timer(timer);
820 int mod_timer_pinned(struct timer_list *timer, unsigned long expires)
822 if (timer->expires == expires && timer_pending(timer))
825 return __mod_timer(timer, expires, false, TIMER_PINNED);
827 EXPORT_SYMBOL(mod_timer_pinned);
830 * add_timer - start a timer
831 * @timer: the timer to be added
833 * The kernel will do a ->function(->data) callback from the
834 * timer interrupt at the ->expires point in the future. The
835 * current time is 'jiffies'.
837 * The timer's ->expires, ->function (and if the handler uses it, ->data)
838 * fields must be set prior calling this function.
840 * Timers with an ->expires field in the past will be executed in the next
843 void add_timer(struct timer_list *timer)
845 BUG_ON(timer_pending(timer));
846 mod_timer(timer, timer->expires);
848 EXPORT_SYMBOL(add_timer);
851 * add_timer_on - start a timer on a particular CPU
852 * @timer: the timer to be added
853 * @cpu: the CPU to start it on
855 * This is not very scalable on SMP. Double adds are not possible.
857 void add_timer_on(struct timer_list *timer, int cpu)
859 struct tvec_base *base = per_cpu(tvec_bases, cpu);
862 timer_stats_timer_set_start_info(timer);
863 BUG_ON(timer_pending(timer) || !timer->function);
864 spin_lock_irqsave(&base->lock, flags);
865 timer_set_base(timer, base);
866 debug_activate(timer, timer->expires);
867 if (time_before(timer->expires, base->next_timer) &&
868 !tbase_get_deferrable(timer->base))
869 base->next_timer = timer->expires;
870 internal_add_timer(base, timer);
872 * Check whether the other CPU is idle and needs to be
873 * triggered to reevaluate the timer wheel when nohz is
874 * active. We are protected against the other CPU fiddling
875 * with the timer by holding the timer base lock. This also
876 * makes sure that a CPU on the way to idle can not evaluate
879 wake_up_idle_cpu(cpu);
880 spin_unlock_irqrestore(&base->lock, flags);
882 EXPORT_SYMBOL_GPL(add_timer_on);
885 * del_timer - deactive a timer.
886 * @timer: the timer to be deactivated
888 * del_timer() deactivates a timer - this works on both active and inactive
891 * The function returns whether it has deactivated a pending timer or not.
892 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
893 * active timer returns 1.)
895 int del_timer(struct timer_list *timer)
897 struct tvec_base *base;
901 timer_stats_timer_clear_start_info(timer);
902 if (timer_pending(timer)) {
903 base = lock_timer_base(timer, &flags);
904 if (timer_pending(timer)) {
905 detach_timer(timer, 1);
906 if (timer->expires == base->next_timer &&
907 !tbase_get_deferrable(timer->base))
908 base->next_timer = base->timer_jiffies;
911 spin_unlock_irqrestore(&base->lock, flags);
916 EXPORT_SYMBOL(del_timer);
920 * try_to_del_timer_sync - Try to deactivate a timer
921 * @timer: timer do del
923 * This function tries to deactivate a timer. Upon successful (ret >= 0)
924 * exit the timer is not queued and the handler is not running on any CPU.
926 * It must not be called from interrupt contexts.
928 int try_to_del_timer_sync(struct timer_list *timer)
930 struct tvec_base *base;
934 base = lock_timer_base(timer, &flags);
936 if (base->running_timer == timer)
940 if (timer_pending(timer)) {
941 detach_timer(timer, 1);
942 if (timer->expires == base->next_timer &&
943 !tbase_get_deferrable(timer->base))
944 base->next_timer = base->timer_jiffies;
948 spin_unlock_irqrestore(&base->lock, flags);
952 EXPORT_SYMBOL(try_to_del_timer_sync);
955 * del_timer_sync - deactivate a timer and wait for the handler to finish.
956 * @timer: the timer to be deactivated
958 * This function only differs from del_timer() on SMP: besides deactivating
959 * the timer it also makes sure the handler has finished executing on other
962 * Synchronization rules: Callers must prevent restarting of the timer,
963 * otherwise this function is meaningless. It must not be called from
964 * interrupt contexts. The caller must not hold locks which would prevent
965 * completion of the timer's handler. The timer's handler must not call
966 * add_timer_on(). Upon exit the timer is not queued and the handler is
967 * not running on any CPU.
969 * The function returns whether it has deactivated a pending timer or not.
971 int del_timer_sync(struct timer_list *timer)
973 #ifdef CONFIG_LOCKDEP
976 local_irq_save(flags);
977 lock_map_acquire(&timer->lockdep_map);
978 lock_map_release(&timer->lockdep_map);
979 local_irq_restore(flags);
983 int ret = try_to_del_timer_sync(timer);
989 EXPORT_SYMBOL(del_timer_sync);
992 static int cascade(struct tvec_base *base, struct tvec *tv, int index)
994 /* cascade all the timers from tv up one level */
995 struct timer_list *timer, *tmp;
996 struct list_head tv_list;
998 list_replace_init(tv->vec + index, &tv_list);
1001 * We are removing _all_ timers from the list, so we
1002 * don't have to detach them individually.
1004 list_for_each_entry_safe(timer, tmp, &tv_list, entry) {
1005 BUG_ON(tbase_get_base(timer->base) != base);
1006 internal_add_timer(base, timer);
1012 static void call_timer_fn(struct timer_list *timer, void (*fn)(unsigned long),
1015 int preempt_count = preempt_count();
1017 #ifdef CONFIG_LOCKDEP
1019 * It is permissible to free the timer from inside the
1020 * function that is called from it, this we need to take into
1021 * account for lockdep too. To avoid bogus "held lock freed"
1022 * warnings as well as problems when looking into
1023 * timer->lockdep_map, make a copy and use that here.
1025 struct lockdep_map lockdep_map = timer->lockdep_map;
1028 * Couple the lock chain with the lock chain at
1029 * del_timer_sync() by acquiring the lock_map around the fn()
1030 * call here and in del_timer_sync().
1032 lock_map_acquire(&lockdep_map);
1034 trace_timer_expire_entry(timer);
1036 trace_timer_expire_exit(timer);
1038 lock_map_release(&lockdep_map);
1040 if (preempt_count != preempt_count()) {
1041 WARN_ONCE(1, "timer: %pF preempt leak: %08x -> %08x\n",
1042 fn, preempt_count, preempt_count());
1044 * Restore the preempt count. That gives us a decent
1045 * chance to survive and extract information. If the
1046 * callback kept a lock held, bad luck, but not worse
1047 * than the BUG() we had.
1049 preempt_count() = preempt_count;
1053 #define INDEX(N) ((base->timer_jiffies >> (TVR_BITS + (N) * TVN_BITS)) & TVN_MASK)
1056 * __run_timers - run all expired timers (if any) on this CPU.
1057 * @base: the timer vector to be processed.
1059 * This function cascades all vectors and executes all expired timer
1062 static inline void __run_timers(struct tvec_base *base)
1064 struct timer_list *timer;
1066 spin_lock_irq(&base->lock);
1067 while (time_after_eq(jiffies, base->timer_jiffies)) {
1068 struct list_head work_list;
1069 struct list_head *head = &work_list;
1070 int index = base->timer_jiffies & TVR_MASK;
1076 (!cascade(base, &base->tv2, INDEX(0))) &&
1077 (!cascade(base, &base->tv3, INDEX(1))) &&
1078 !cascade(base, &base->tv4, INDEX(2)))
1079 cascade(base, &base->tv5, INDEX(3));
1080 ++base->timer_jiffies;
1081 list_replace_init(base->tv1.vec + index, &work_list);
1082 while (!list_empty(head)) {
1083 void (*fn)(unsigned long);
1086 timer = list_first_entry(head, struct timer_list,entry);
1087 fn = timer->function;
1090 timer_stats_account_timer(timer);
1092 set_running_timer(base, timer);
1093 detach_timer(timer, 1);
1095 spin_unlock_irq(&base->lock);
1096 call_timer_fn(timer, fn, data);
1097 spin_lock_irq(&base->lock);
1100 set_running_timer(base, NULL);
1101 spin_unlock_irq(&base->lock);
1106 * Find out when the next timer event is due to happen. This
1107 * is used on S/390 to stop all activity when a CPU is idle.
1108 * This function needs to be called with interrupts disabled.
1110 static unsigned long __next_timer_interrupt(struct tvec_base *base)
1112 unsigned long timer_jiffies = base->timer_jiffies;
1113 unsigned long expires = timer_jiffies + NEXT_TIMER_MAX_DELTA;
1114 int index, slot, array, found = 0;
1115 struct timer_list *nte;
1116 struct tvec *varray[4];
1118 /* Look for timer events in tv1. */
1119 index = slot = timer_jiffies & TVR_MASK;
1121 list_for_each_entry(nte, base->tv1.vec + slot, entry) {
1122 if (tbase_get_deferrable(nte->base))
1126 expires = nte->expires;
1127 /* Look at the cascade bucket(s)? */
1128 if (!index || slot < index)
1132 slot = (slot + 1) & TVR_MASK;
1133 } while (slot != index);
1136 /* Calculate the next cascade event */
1138 timer_jiffies += TVR_SIZE - index;
1139 timer_jiffies >>= TVR_BITS;
1141 /* Check tv2-tv5. */
1142 varray[0] = &base->tv2;
1143 varray[1] = &base->tv3;
1144 varray[2] = &base->tv4;
1145 varray[3] = &base->tv5;
1147 for (array = 0; array < 4; array++) {
1148 struct tvec *varp = varray[array];
1150 index = slot = timer_jiffies & TVN_MASK;
1152 list_for_each_entry(nte, varp->vec + slot, entry) {
1153 if (tbase_get_deferrable(nte->base))
1157 if (time_before(nte->expires, expires))
1158 expires = nte->expires;
1161 * Do we still search for the first timer or are
1162 * we looking up the cascade buckets ?
1165 /* Look at the cascade bucket(s)? */
1166 if (!index || slot < index)
1170 slot = (slot + 1) & TVN_MASK;
1171 } while (slot != index);
1174 timer_jiffies += TVN_SIZE - index;
1175 timer_jiffies >>= TVN_BITS;
1181 * Check, if the next hrtimer event is before the next timer wheel
1184 static unsigned long cmp_next_hrtimer_event(unsigned long now,
1185 unsigned long expires)
1187 ktime_t hr_delta = hrtimer_get_next_event();
1188 struct timespec tsdelta;
1189 unsigned long delta;
1191 if (hr_delta.tv64 == KTIME_MAX)
1195 * Expired timer available, let it expire in the next tick
1197 if (hr_delta.tv64 <= 0)
1200 tsdelta = ktime_to_timespec(hr_delta);
1201 delta = timespec_to_jiffies(&tsdelta);
1204 * Limit the delta to the max value, which is checked in
1205 * tick_nohz_stop_sched_tick():
1207 if (delta > NEXT_TIMER_MAX_DELTA)
1208 delta = NEXT_TIMER_MAX_DELTA;
1211 * Take rounding errors in to account and make sure, that it
1212 * expires in the next tick. Otherwise we go into an endless
1213 * ping pong due to tick_nohz_stop_sched_tick() retriggering
1219 if (time_before(now, expires))
1225 * get_next_timer_interrupt - return the jiffy of the next pending timer
1226 * @now: current time (in jiffies)
1228 unsigned long get_next_timer_interrupt(unsigned long now)
1230 struct tvec_base *base = __get_cpu_var(tvec_bases);
1231 unsigned long expires;
1233 spin_lock(&base->lock);
1234 if (time_before_eq(base->next_timer, base->timer_jiffies))
1235 base->next_timer = __next_timer_interrupt(base);
1236 expires = base->next_timer;
1237 spin_unlock(&base->lock);
1239 if (time_before_eq(expires, now))
1242 return cmp_next_hrtimer_event(now, expires);
1247 * Called from the timer interrupt handler to charge one tick to the current
1248 * process. user_tick is 1 if the tick is user time, 0 for system.
1250 void update_process_times(int user_tick)
1252 struct task_struct *p = current;
1253 int cpu = smp_processor_id();
1255 /* Note: this timer irq context must be accounted for as well. */
1256 account_process_tick(p, user_tick);
1258 rcu_check_callbacks(cpu, user_tick);
1260 perf_event_do_pending();
1262 run_posix_cpu_timers(p);
1266 * This function runs timers and the timer-tq in bottom half context.
1268 static void run_timer_softirq(struct softirq_action *h)
1270 struct tvec_base *base = __get_cpu_var(tvec_bases);
1272 hrtimer_run_pending();
1274 if (time_after_eq(jiffies, base->timer_jiffies))
1279 * Called by the local, per-CPU timer interrupt on SMP.
1281 void run_local_timers(void)
1283 hrtimer_run_queues();
1284 raise_softirq(TIMER_SOFTIRQ);
1289 * The 64-bit jiffies value is not atomic - you MUST NOT read it
1290 * without sampling the sequence number in xtime_lock.
1291 * jiffies is defined in the linker script...
1294 void do_timer(unsigned long ticks)
1296 jiffies_64 += ticks;
1301 #ifdef __ARCH_WANT_SYS_ALARM
1304 * For backwards compatibility? This can be done in libc so Alpha
1305 * and all newer ports shouldn't need it.
1307 SYSCALL_DEFINE1(alarm, unsigned int, seconds)
1309 return alarm_setitimer(seconds);
1317 * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this
1318 * should be moved into arch/i386 instead?
1322 * sys_getpid - return the thread group id of the current process
1324 * Note, despite the name, this returns the tgid not the pid. The tgid and
1325 * the pid are identical unless CLONE_THREAD was specified on clone() in
1326 * which case the tgid is the same in all threads of the same group.
1328 * This is SMP safe as current->tgid does not change.
1330 SYSCALL_DEFINE0(getpid)
1332 return task_tgid_vnr(current);
1336 * Accessing ->real_parent is not SMP-safe, it could
1337 * change from under us. However, we can use a stale
1338 * value of ->real_parent under rcu_read_lock(), see
1339 * release_task()->call_rcu(delayed_put_task_struct).
1341 SYSCALL_DEFINE0(getppid)
1346 pid = task_tgid_vnr(current->real_parent);
1352 SYSCALL_DEFINE0(getuid)
1354 /* Only we change this so SMP safe */
1355 return current_uid();
1358 SYSCALL_DEFINE0(geteuid)
1360 /* Only we change this so SMP safe */
1361 return current_euid();
1364 SYSCALL_DEFINE0(getgid)
1366 /* Only we change this so SMP safe */
1367 return current_gid();
1370 SYSCALL_DEFINE0(getegid)
1372 /* Only we change this so SMP safe */
1373 return current_egid();
1378 static void process_timeout(unsigned long __data)
1380 wake_up_process((struct task_struct *)__data);
1384 * schedule_timeout - sleep until timeout
1385 * @timeout: timeout value in jiffies
1387 * Make the current task sleep until @timeout jiffies have
1388 * elapsed. The routine will return immediately unless
1389 * the current task state has been set (see set_current_state()).
1391 * You can set the task state as follows -
1393 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1394 * pass before the routine returns. The routine will return 0
1396 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1397 * delivered to the current task. In this case the remaining time
1398 * in jiffies will be returned, or 0 if the timer expired in time
1400 * The current task state is guaranteed to be TASK_RUNNING when this
1403 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1404 * the CPU away without a bound on the timeout. In this case the return
1405 * value will be %MAX_SCHEDULE_TIMEOUT.
1407 * In all cases the return value is guaranteed to be non-negative.
1409 signed long __sched schedule_timeout(signed long timeout)
1411 struct timer_list timer;
1412 unsigned long expire;
1416 case MAX_SCHEDULE_TIMEOUT:
1418 * These two special cases are useful to be comfortable
1419 * in the caller. Nothing more. We could take
1420 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1421 * but I' d like to return a valid offset (>=0) to allow
1422 * the caller to do everything it want with the retval.
1428 * Another bit of PARANOID. Note that the retval will be
1429 * 0 since no piece of kernel is supposed to do a check
1430 * for a negative retval of schedule_timeout() (since it
1431 * should never happens anyway). You just have the printk()
1432 * that will tell you if something is gone wrong and where.
1435 printk(KERN_ERR "schedule_timeout: wrong timeout "
1436 "value %lx\n", timeout);
1438 current->state = TASK_RUNNING;
1443 expire = timeout + jiffies;
1445 setup_timer_on_stack(&timer, process_timeout, (unsigned long)current);
1446 __mod_timer(&timer, expire, false, TIMER_NOT_PINNED);
1448 del_singleshot_timer_sync(&timer);
1450 /* Remove the timer from the object tracker */
1451 destroy_timer_on_stack(&timer);
1453 timeout = expire - jiffies;
1456 return timeout < 0 ? 0 : timeout;
1458 EXPORT_SYMBOL(schedule_timeout);
1461 * We can use __set_current_state() here because schedule_timeout() calls
1462 * schedule() unconditionally.
1464 signed long __sched schedule_timeout_interruptible(signed long timeout)
1466 __set_current_state(TASK_INTERRUPTIBLE);
1467 return schedule_timeout(timeout);
1469 EXPORT_SYMBOL(schedule_timeout_interruptible);
1471 signed long __sched schedule_timeout_killable(signed long timeout)
1473 __set_current_state(TASK_KILLABLE);
1474 return schedule_timeout(timeout);
1476 EXPORT_SYMBOL(schedule_timeout_killable);
1478 signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1480 __set_current_state(TASK_UNINTERRUPTIBLE);
1481 return schedule_timeout(timeout);
1483 EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1485 /* Thread ID - the internal kernel "pid" */
1486 SYSCALL_DEFINE0(gettid)
1488 return task_pid_vnr(current);
1492 * do_sysinfo - fill in sysinfo struct
1493 * @info: pointer to buffer to fill
1495 int do_sysinfo(struct sysinfo *info)
1497 unsigned long mem_total, sav_total;
1498 unsigned int mem_unit, bitcount;
1501 memset(info, 0, sizeof(struct sysinfo));
1504 monotonic_to_bootbased(&tp);
1505 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
1507 get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
1509 info->procs = nr_threads;
1515 * If the sum of all the available memory (i.e. ram + swap)
1516 * is less than can be stored in a 32 bit unsigned long then
1517 * we can be binary compatible with 2.2.x kernels. If not,
1518 * well, in that case 2.2.x was broken anyways...
1520 * -Erik Andersen <andersee@debian.org>
1523 mem_total = info->totalram + info->totalswap;
1524 if (mem_total < info->totalram || mem_total < info->totalswap)
1527 mem_unit = info->mem_unit;
1528 while (mem_unit > 1) {
1531 sav_total = mem_total;
1533 if (mem_total < sav_total)
1538 * If mem_total did not overflow, multiply all memory values by
1539 * info->mem_unit and set it to 1. This leaves things compatible
1540 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1545 info->totalram <<= bitcount;
1546 info->freeram <<= bitcount;
1547 info->sharedram <<= bitcount;
1548 info->bufferram <<= bitcount;
1549 info->totalswap <<= bitcount;
1550 info->freeswap <<= bitcount;
1551 info->totalhigh <<= bitcount;
1552 info->freehigh <<= bitcount;
1558 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
1564 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
1570 static int __cpuinit init_timers_cpu(int cpu)
1573 struct tvec_base *base;
1574 static char __cpuinitdata tvec_base_done[NR_CPUS];
1576 if (!tvec_base_done[cpu]) {
1577 static char boot_done;
1581 * The APs use this path later in boot
1583 base = kmalloc_node(sizeof(*base),
1584 GFP_KERNEL | __GFP_ZERO,
1589 /* Make sure that tvec_base is 2 byte aligned */
1590 if (tbase_get_deferrable(base)) {
1595 per_cpu(tvec_bases, cpu) = base;
1598 * This is for the boot CPU - we use compile-time
1599 * static initialisation because per-cpu memory isn't
1600 * ready yet and because the memory allocators are not
1601 * initialised either.
1604 base = &boot_tvec_bases;
1606 tvec_base_done[cpu] = 1;
1608 base = per_cpu(tvec_bases, cpu);
1611 spin_lock_init(&base->lock);
1613 for (j = 0; j < TVN_SIZE; j++) {
1614 INIT_LIST_HEAD(base->tv5.vec + j);
1615 INIT_LIST_HEAD(base->tv4.vec + j);
1616 INIT_LIST_HEAD(base->tv3.vec + j);
1617 INIT_LIST_HEAD(base->tv2.vec + j);
1619 for (j = 0; j < TVR_SIZE; j++)
1620 INIT_LIST_HEAD(base->tv1.vec + j);
1622 base->timer_jiffies = jiffies;
1623 base->next_timer = base->timer_jiffies;
1627 #ifdef CONFIG_HOTPLUG_CPU
1628 static void migrate_timer_list(struct tvec_base *new_base, struct list_head *head)
1630 struct timer_list *timer;
1632 while (!list_empty(head)) {
1633 timer = list_first_entry(head, struct timer_list, entry);
1634 detach_timer(timer, 0);
1635 timer_set_base(timer, new_base);
1636 if (time_before(timer->expires, new_base->next_timer) &&
1637 !tbase_get_deferrable(timer->base))
1638 new_base->next_timer = timer->expires;
1639 internal_add_timer(new_base, timer);
1643 static void __cpuinit migrate_timers(int cpu)
1645 struct tvec_base *old_base;
1646 struct tvec_base *new_base;
1649 BUG_ON(cpu_online(cpu));
1650 old_base = per_cpu(tvec_bases, cpu);
1651 new_base = get_cpu_var(tvec_bases);
1653 * The caller is globally serialized and nobody else
1654 * takes two locks at once, deadlock is not possible.
1656 spin_lock_irq(&new_base->lock);
1657 spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1659 BUG_ON(old_base->running_timer);
1661 for (i = 0; i < TVR_SIZE; i++)
1662 migrate_timer_list(new_base, old_base->tv1.vec + i);
1663 for (i = 0; i < TVN_SIZE; i++) {
1664 migrate_timer_list(new_base, old_base->tv2.vec + i);
1665 migrate_timer_list(new_base, old_base->tv3.vec + i);
1666 migrate_timer_list(new_base, old_base->tv4.vec + i);
1667 migrate_timer_list(new_base, old_base->tv5.vec + i);
1670 spin_unlock(&old_base->lock);
1671 spin_unlock_irq(&new_base->lock);
1672 put_cpu_var(tvec_bases);
1674 #endif /* CONFIG_HOTPLUG_CPU */
1676 static int __cpuinit timer_cpu_notify(struct notifier_block *self,
1677 unsigned long action, void *hcpu)
1679 long cpu = (long)hcpu;
1681 case CPU_UP_PREPARE:
1682 case CPU_UP_PREPARE_FROZEN:
1683 if (init_timers_cpu(cpu) < 0)
1686 #ifdef CONFIG_HOTPLUG_CPU
1688 case CPU_DEAD_FROZEN:
1689 migrate_timers(cpu);
1698 static struct notifier_block __cpuinitdata timers_nb = {
1699 .notifier_call = timer_cpu_notify,
1703 void __init init_timers(void)
1705 int err = timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE,
1706 (void *)(long)smp_processor_id());
1710 BUG_ON(err == NOTIFY_BAD);
1711 register_cpu_notifier(&timers_nb);
1712 open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
1716 * msleep - sleep safely even with waitqueue interruptions
1717 * @msecs: Time in milliseconds to sleep for
1719 void msleep(unsigned int msecs)
1721 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1724 timeout = schedule_timeout_uninterruptible(timeout);
1727 EXPORT_SYMBOL(msleep);
1730 * msleep_interruptible - sleep waiting for signals
1731 * @msecs: Time in milliseconds to sleep for
1733 unsigned long msleep_interruptible(unsigned int msecs)
1735 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1737 while (timeout && !signal_pending(current))
1738 timeout = schedule_timeout_interruptible(timeout);
1739 return jiffies_to_msecs(timeout);
1742 EXPORT_SYMBOL(msleep_interruptible);