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>
41 #include <asm/uaccess.h>
42 #include <asm/unistd.h>
43 #include <asm/div64.h>
44 #include <asm/timex.h>
47 u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
49 EXPORT_SYMBOL(jiffies_64);
52 * per-CPU timer vector definitions:
54 #define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6)
55 #define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8)
56 #define TVN_SIZE (1 << TVN_BITS)
57 #define TVR_SIZE (1 << TVR_BITS)
58 #define TVN_MASK (TVN_SIZE - 1)
59 #define TVR_MASK (TVR_SIZE - 1)
62 struct list_head vec[TVN_SIZE];
66 struct list_head vec[TVR_SIZE];
71 struct timer_list *running_timer;
72 unsigned long timer_jiffies;
78 } ____cacheline_aligned;
80 struct tvec_base boot_tvec_bases;
81 EXPORT_SYMBOL(boot_tvec_bases);
82 static DEFINE_PER_CPU(struct tvec_base *, tvec_bases) = &boot_tvec_bases;
85 * Note that all tvec_bases are 2 byte aligned and lower bit of
86 * base in timer_list is guaranteed to be zero. Use the LSB for
87 * the new flag to indicate whether the timer is deferrable
89 #define TBASE_DEFERRABLE_FLAG (0x1)
91 /* Functions below help us manage 'deferrable' flag */
92 static inline unsigned int tbase_get_deferrable(struct tvec_base *base)
94 return ((unsigned int)(unsigned long)base & TBASE_DEFERRABLE_FLAG);
97 static inline struct tvec_base *tbase_get_base(struct tvec_base *base)
99 return ((struct tvec_base *)((unsigned long)base & ~TBASE_DEFERRABLE_FLAG));
102 static inline void timer_set_deferrable(struct timer_list *timer)
104 timer->base = ((struct tvec_base *)((unsigned long)(timer->base) |
105 TBASE_DEFERRABLE_FLAG));
109 timer_set_base(struct timer_list *timer, struct tvec_base *new_base)
111 timer->base = (struct tvec_base *)((unsigned long)(new_base) |
112 tbase_get_deferrable(timer->base));
115 static unsigned long round_jiffies_common(unsigned long j, int cpu,
119 unsigned long original = j;
122 * We don't want all cpus firing their timers at once hitting the
123 * same lock or cachelines, so we skew each extra cpu with an extra
124 * 3 jiffies. This 3 jiffies came originally from the mm/ code which
126 * The skew is done by adding 3*cpunr, then round, then subtract this
127 * extra offset again.
134 * If the target jiffie is just after a whole second (which can happen
135 * due to delays of the timer irq, long irq off times etc etc) then
136 * we should round down to the whole second, not up. Use 1/4th second
137 * as cutoff for this rounding as an extreme upper bound for this.
138 * But never round down if @force_up is set.
140 if (rem < HZ/4 && !force_up) /* round down */
145 /* now that we have rounded, subtract the extra skew again */
148 if (j <= jiffies) /* rounding ate our timeout entirely; */
154 * __round_jiffies - function to round jiffies to a full second
155 * @j: the time in (absolute) jiffies that should be rounded
156 * @cpu: the processor number on which the timeout will happen
158 * __round_jiffies() rounds an absolute time in the future (in jiffies)
159 * up or down to (approximately) full seconds. This is useful for timers
160 * for which the exact time they fire does not matter too much, as long as
161 * they fire approximately every X seconds.
163 * By rounding these timers to whole seconds, all such timers will fire
164 * at the same time, rather than at various times spread out. The goal
165 * of this is to have the CPU wake up less, which saves power.
167 * The exact rounding is skewed for each processor to avoid all
168 * processors firing at the exact same time, which could lead
169 * to lock contention or spurious cache line bouncing.
171 * The return value is the rounded version of the @j parameter.
173 unsigned long __round_jiffies(unsigned long j, int cpu)
175 return round_jiffies_common(j, cpu, false);
177 EXPORT_SYMBOL_GPL(__round_jiffies);
180 * __round_jiffies_relative - function to round jiffies to a full second
181 * @j: the time in (relative) jiffies that should be rounded
182 * @cpu: the processor number on which the timeout will happen
184 * __round_jiffies_relative() rounds a time delta in the future (in jiffies)
185 * up or down to (approximately) full seconds. This is useful for timers
186 * for which the exact time they fire does not matter too much, as long as
187 * they fire approximately every X seconds.
189 * By rounding these timers to whole seconds, all such timers will fire
190 * at the same time, rather than at various times spread out. The goal
191 * of this is to have the CPU wake up less, which saves power.
193 * The exact rounding is skewed for each processor to avoid all
194 * processors firing at the exact same time, which could lead
195 * to lock contention or spurious cache line bouncing.
197 * The return value is the rounded version of the @j parameter.
199 unsigned long __round_jiffies_relative(unsigned long j, int cpu)
201 unsigned long j0 = jiffies;
203 /* Use j0 because jiffies might change while we run */
204 return round_jiffies_common(j + j0, cpu, false) - j0;
206 EXPORT_SYMBOL_GPL(__round_jiffies_relative);
209 * round_jiffies - function to round jiffies to a full second
210 * @j: the time in (absolute) jiffies that should be rounded
212 * round_jiffies() rounds an absolute time in the future (in jiffies)
213 * up or down to (approximately) full seconds. This is useful for timers
214 * for which the exact time they fire does not matter too much, as long as
215 * they fire approximately every X seconds.
217 * By rounding these timers to whole seconds, all such timers will fire
218 * at the same time, rather than at various times spread out. The goal
219 * of this is to have the CPU wake up less, which saves power.
221 * The return value is the rounded version of the @j parameter.
223 unsigned long round_jiffies(unsigned long j)
225 return round_jiffies_common(j, raw_smp_processor_id(), false);
227 EXPORT_SYMBOL_GPL(round_jiffies);
230 * round_jiffies_relative - function to round jiffies to a full second
231 * @j: the time in (relative) jiffies that should be rounded
233 * round_jiffies_relative() rounds a time delta in the future (in jiffies)
234 * up or down to (approximately) full seconds. This is useful for timers
235 * for which the exact time they fire does not matter too much, as long as
236 * they fire approximately every X seconds.
238 * By rounding these timers to whole seconds, all such timers will fire
239 * at the same time, rather than at various times spread out. The goal
240 * of this is to have the CPU wake up less, which saves power.
242 * The return value is the rounded version of the @j parameter.
244 unsigned long round_jiffies_relative(unsigned long j)
246 return __round_jiffies_relative(j, raw_smp_processor_id());
248 EXPORT_SYMBOL_GPL(round_jiffies_relative);
251 * __round_jiffies_up - function to round jiffies up to a full second
252 * @j: the time in (absolute) jiffies that should be rounded
253 * @cpu: the processor number on which the timeout will happen
255 * This is the same as __round_jiffies() except that it will never
256 * round down. This is useful for timeouts for which the exact time
257 * of firing does not matter too much, as long as they don't fire too
260 unsigned long __round_jiffies_up(unsigned long j, int cpu)
262 return round_jiffies_common(j, cpu, true);
264 EXPORT_SYMBOL_GPL(__round_jiffies_up);
267 * __round_jiffies_up_relative - function to round jiffies up to a full second
268 * @j: the time in (relative) jiffies that should be rounded
269 * @cpu: the processor number on which the timeout will happen
271 * This is the same as __round_jiffies_relative() except that it will never
272 * round down. This is useful for timeouts for which the exact time
273 * of firing does not matter too much, as long as they don't fire too
276 unsigned long __round_jiffies_up_relative(unsigned long j, int cpu)
278 unsigned long j0 = jiffies;
280 /* Use j0 because jiffies might change while we run */
281 return round_jiffies_common(j + j0, cpu, true) - j0;
283 EXPORT_SYMBOL_GPL(__round_jiffies_up_relative);
286 * round_jiffies_up - function to round jiffies up to a full second
287 * @j: the time in (absolute) jiffies that should be rounded
289 * This is the same as round_jiffies() except that it will never
290 * round down. This is useful for timeouts for which the exact time
291 * of firing does not matter too much, as long as they don't fire too
294 unsigned long round_jiffies_up(unsigned long j)
296 return round_jiffies_common(j, raw_smp_processor_id(), true);
298 EXPORT_SYMBOL_GPL(round_jiffies_up);
301 * round_jiffies_up_relative - function to round jiffies up to a full second
302 * @j: the time in (relative) jiffies that should be rounded
304 * This is the same as round_jiffies_relative() except that it will never
305 * round down. This is useful for timeouts for which the exact time
306 * of firing does not matter too much, as long as they don't fire too
309 unsigned long round_jiffies_up_relative(unsigned long j)
311 return __round_jiffies_up_relative(j, raw_smp_processor_id());
313 EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
316 static inline void set_running_timer(struct tvec_base *base,
317 struct timer_list *timer)
320 base->running_timer = timer;
324 static void internal_add_timer(struct tvec_base *base, struct timer_list *timer)
326 unsigned long expires = timer->expires;
327 unsigned long idx = expires - base->timer_jiffies;
328 struct list_head *vec;
330 if (idx < TVR_SIZE) {
331 int i = expires & TVR_MASK;
332 vec = base->tv1.vec + i;
333 } else if (idx < 1 << (TVR_BITS + TVN_BITS)) {
334 int i = (expires >> TVR_BITS) & TVN_MASK;
335 vec = base->tv2.vec + i;
336 } else if (idx < 1 << (TVR_BITS + 2 * TVN_BITS)) {
337 int i = (expires >> (TVR_BITS + TVN_BITS)) & TVN_MASK;
338 vec = base->tv3.vec + i;
339 } else if (idx < 1 << (TVR_BITS + 3 * TVN_BITS)) {
340 int i = (expires >> (TVR_BITS + 2 * TVN_BITS)) & TVN_MASK;
341 vec = base->tv4.vec + i;
342 } else if ((signed long) idx < 0) {
344 * Can happen if you add a timer with expires == jiffies,
345 * or you set a timer to go off in the past
347 vec = base->tv1.vec + (base->timer_jiffies & TVR_MASK);
350 /* If the timeout is larger than 0xffffffff on 64-bit
351 * architectures then we use the maximum timeout:
353 if (idx > 0xffffffffUL) {
355 expires = idx + base->timer_jiffies;
357 i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK;
358 vec = base->tv5.vec + i;
363 list_add_tail(&timer->entry, vec);
366 #ifdef CONFIG_TIMER_STATS
367 void __timer_stats_timer_set_start_info(struct timer_list *timer, void *addr)
369 if (timer->start_site)
372 timer->start_site = addr;
373 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
374 timer->start_pid = current->pid;
377 static void timer_stats_account_timer(struct timer_list *timer)
379 unsigned int flag = 0;
381 if (unlikely(tbase_get_deferrable(timer->base)))
382 flag |= TIMER_STATS_FLAG_DEFERRABLE;
384 timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
385 timer->function, timer->start_comm, flag);
389 static void timer_stats_account_timer(struct timer_list *timer) {}
392 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
394 static struct debug_obj_descr timer_debug_descr;
397 * fixup_init is called when:
398 * - an active object is initialized
400 static int timer_fixup_init(void *addr, enum debug_obj_state state)
402 struct timer_list *timer = addr;
405 case ODEBUG_STATE_ACTIVE:
406 del_timer_sync(timer);
407 debug_object_init(timer, &timer_debug_descr);
415 * fixup_activate is called when:
416 * - an active object is activated
417 * - an unknown object is activated (might be a statically initialized object)
419 static int timer_fixup_activate(void *addr, enum debug_obj_state state)
421 struct timer_list *timer = addr;
425 case ODEBUG_STATE_NOTAVAILABLE:
427 * This is not really a fixup. The timer was
428 * statically initialized. We just make sure that it
429 * is tracked in the object tracker.
431 if (timer->entry.next == NULL &&
432 timer->entry.prev == TIMER_ENTRY_STATIC) {
433 debug_object_init(timer, &timer_debug_descr);
434 debug_object_activate(timer, &timer_debug_descr);
441 case ODEBUG_STATE_ACTIVE:
450 * fixup_free is called when:
451 * - an active object is freed
453 static int timer_fixup_free(void *addr, enum debug_obj_state state)
455 struct timer_list *timer = addr;
458 case ODEBUG_STATE_ACTIVE:
459 del_timer_sync(timer);
460 debug_object_free(timer, &timer_debug_descr);
467 static struct debug_obj_descr timer_debug_descr = {
468 .name = "timer_list",
469 .fixup_init = timer_fixup_init,
470 .fixup_activate = timer_fixup_activate,
471 .fixup_free = timer_fixup_free,
474 static inline void debug_timer_init(struct timer_list *timer)
476 debug_object_init(timer, &timer_debug_descr);
479 static inline void debug_timer_activate(struct timer_list *timer)
481 debug_object_activate(timer, &timer_debug_descr);
484 static inline void debug_timer_deactivate(struct timer_list *timer)
486 debug_object_deactivate(timer, &timer_debug_descr);
489 static inline void debug_timer_free(struct timer_list *timer)
491 debug_object_free(timer, &timer_debug_descr);
494 static void __init_timer(struct timer_list *timer,
496 struct lock_class_key *key);
498 void init_timer_on_stack_key(struct timer_list *timer,
500 struct lock_class_key *key)
502 debug_object_init_on_stack(timer, &timer_debug_descr);
503 __init_timer(timer, name, key);
505 EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
507 void destroy_timer_on_stack(struct timer_list *timer)
509 debug_object_free(timer, &timer_debug_descr);
511 EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
514 static inline void debug_timer_init(struct timer_list *timer) { }
515 static inline void debug_timer_activate(struct timer_list *timer) { }
516 static inline void debug_timer_deactivate(struct timer_list *timer) { }
519 static void __init_timer(struct timer_list *timer,
521 struct lock_class_key *key)
523 timer->entry.next = NULL;
524 timer->base = __raw_get_cpu_var(tvec_bases);
525 #ifdef CONFIG_TIMER_STATS
526 timer->start_site = NULL;
527 timer->start_pid = -1;
528 memset(timer->start_comm, 0, TASK_COMM_LEN);
530 lockdep_init_map(&timer->lockdep_map, name, key, 0);
534 * init_timer - initialize a timer.
535 * @timer: the timer to be initialized
537 * init_timer() must be done to a timer prior calling *any* of the
538 * other timer functions.
540 void init_timer_key(struct timer_list *timer,
542 struct lock_class_key *key)
544 debug_timer_init(timer);
545 __init_timer(timer, name, key);
547 EXPORT_SYMBOL(init_timer_key);
549 void init_timer_deferrable_key(struct timer_list *timer,
551 struct lock_class_key *key)
553 init_timer_key(timer, name, key);
554 timer_set_deferrable(timer);
556 EXPORT_SYMBOL(init_timer_deferrable_key);
558 static inline void detach_timer(struct timer_list *timer,
561 struct list_head *entry = &timer->entry;
563 debug_timer_deactivate(timer);
565 __list_del(entry->prev, entry->next);
568 entry->prev = LIST_POISON2;
572 * We are using hashed locking: holding per_cpu(tvec_bases).lock
573 * means that all timers which are tied to this base via timer->base are
574 * locked, and the base itself is locked too.
576 * So __run_timers/migrate_timers can safely modify all timers which could
577 * be found on ->tvX lists.
579 * When the timer's base is locked, and the timer removed from list, it is
580 * possible to set timer->base = NULL and drop the lock: the timer remains
583 static struct tvec_base *lock_timer_base(struct timer_list *timer,
584 unsigned long *flags)
585 __acquires(timer->base->lock)
587 struct tvec_base *base;
590 struct tvec_base *prelock_base = timer->base;
591 base = tbase_get_base(prelock_base);
592 if (likely(base != NULL)) {
593 spin_lock_irqsave(&base->lock, *flags);
594 if (likely(prelock_base == timer->base))
596 /* The timer has migrated to another CPU */
597 spin_unlock_irqrestore(&base->lock, *flags);
604 __mod_timer(struct timer_list *timer, unsigned long expires, bool pending_only)
606 struct tvec_base *base, *new_base;
612 timer_stats_timer_set_start_info(timer);
613 BUG_ON(!timer->function);
615 base = lock_timer_base(timer, &flags);
617 if (timer_pending(timer)) {
618 detach_timer(timer, 0);
625 debug_timer_activate(timer);
627 new_base = __get_cpu_var(tvec_bases);
629 if (base != new_base) {
631 * We are trying to schedule the timer on the local CPU.
632 * However we can't change timer's base while it is running,
633 * otherwise del_timer_sync() can't detect that the timer's
634 * handler yet has not finished. This also guarantees that
635 * the timer is serialized wrt itself.
637 if (likely(base->running_timer != timer)) {
638 /* See the comment in lock_timer_base() */
639 timer_set_base(timer, NULL);
640 spin_unlock(&base->lock);
642 spin_lock(&base->lock);
643 timer_set_base(timer, base);
647 timer->expires = expires;
648 internal_add_timer(base, timer);
651 spin_unlock_irqrestore(&base->lock, flags);
657 * mod_timer_pending - modify a pending timer's timeout
658 * @timer: the pending timer to be modified
659 * @expires: new timeout in jiffies
661 * mod_timer_pending() is the same for pending timers as mod_timer(),
662 * but will not re-activate and modify already deleted timers.
664 * It is useful for unserialized use of timers.
666 int mod_timer_pending(struct timer_list *timer, unsigned long expires)
668 return __mod_timer(timer, expires, true);
670 EXPORT_SYMBOL(mod_timer_pending);
673 * mod_timer - modify a timer's timeout
674 * @timer: the timer to be modified
675 * @expires: new timeout in jiffies
677 * mod_timer() is a more efficient way to update the expire field of an
678 * active timer (if the timer is inactive it will be activated)
680 * mod_timer(timer, expires) is equivalent to:
682 * del_timer(timer); timer->expires = expires; add_timer(timer);
684 * Note that if there are multiple unserialized concurrent users of the
685 * same timer, then mod_timer() is the only safe way to modify the timeout,
686 * since add_timer() cannot modify an already running timer.
688 * The function returns whether it has modified a pending timer or not.
689 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
690 * active timer returns 1.)
692 int mod_timer(struct timer_list *timer, unsigned long expires)
695 * This is a common optimization triggered by the
696 * networking code - if the timer is re-modified
697 * to be the same thing then just return:
699 if (timer->expires == expires && timer_pending(timer))
702 return __mod_timer(timer, expires, false);
704 EXPORT_SYMBOL(mod_timer);
707 * add_timer - start a timer
708 * @timer: the timer to be added
710 * The kernel will do a ->function(->data) callback from the
711 * timer interrupt at the ->expires point in the future. The
712 * current time is 'jiffies'.
714 * The timer's ->expires, ->function (and if the handler uses it, ->data)
715 * fields must be set prior calling this function.
717 * Timers with an ->expires field in the past will be executed in the next
720 void add_timer(struct timer_list *timer)
722 BUG_ON(timer_pending(timer));
723 mod_timer(timer, timer->expires);
725 EXPORT_SYMBOL(add_timer);
728 * add_timer_on - start a timer on a particular CPU
729 * @timer: the timer to be added
730 * @cpu: the CPU to start it on
732 * This is not very scalable on SMP. Double adds are not possible.
734 void add_timer_on(struct timer_list *timer, int cpu)
736 struct tvec_base *base = per_cpu(tvec_bases, cpu);
739 timer_stats_timer_set_start_info(timer);
740 BUG_ON(timer_pending(timer) || !timer->function);
741 spin_lock_irqsave(&base->lock, flags);
742 timer_set_base(timer, base);
743 debug_timer_activate(timer);
744 internal_add_timer(base, timer);
746 * Check whether the other CPU is idle and needs to be
747 * triggered to reevaluate the timer wheel when nohz is
748 * active. We are protected against the other CPU fiddling
749 * with the timer by holding the timer base lock. This also
750 * makes sure that a CPU on the way to idle can not evaluate
753 wake_up_idle_cpu(cpu);
754 spin_unlock_irqrestore(&base->lock, flags);
758 * del_timer - deactive a timer.
759 * @timer: the timer to be deactivated
761 * del_timer() deactivates a timer - this works on both active and inactive
764 * The function returns whether it has deactivated a pending timer or not.
765 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
766 * active timer returns 1.)
768 int del_timer(struct timer_list *timer)
770 struct tvec_base *base;
774 timer_stats_timer_clear_start_info(timer);
775 if (timer_pending(timer)) {
776 base = lock_timer_base(timer, &flags);
777 if (timer_pending(timer)) {
778 detach_timer(timer, 1);
781 spin_unlock_irqrestore(&base->lock, flags);
786 EXPORT_SYMBOL(del_timer);
790 * try_to_del_timer_sync - Try to deactivate a timer
791 * @timer: timer do del
793 * This function tries to deactivate a timer. Upon successful (ret >= 0)
794 * exit the timer is not queued and the handler is not running on any CPU.
796 * It must not be called from interrupt contexts.
798 int try_to_del_timer_sync(struct timer_list *timer)
800 struct tvec_base *base;
804 base = lock_timer_base(timer, &flags);
806 if (base->running_timer == timer)
810 if (timer_pending(timer)) {
811 detach_timer(timer, 1);
815 spin_unlock_irqrestore(&base->lock, flags);
819 EXPORT_SYMBOL(try_to_del_timer_sync);
822 * del_timer_sync - deactivate a timer and wait for the handler to finish.
823 * @timer: the timer to be deactivated
825 * This function only differs from del_timer() on SMP: besides deactivating
826 * the timer it also makes sure the handler has finished executing on other
829 * Synchronization rules: Callers must prevent restarting of the timer,
830 * otherwise this function is meaningless. It must not be called from
831 * interrupt contexts. The caller must not hold locks which would prevent
832 * completion of the timer's handler. The timer's handler must not call
833 * add_timer_on(). Upon exit the timer is not queued and the handler is
834 * not running on any CPU.
836 * The function returns whether it has deactivated a pending timer or not.
838 int del_timer_sync(struct timer_list *timer)
840 #ifdef CONFIG_LOCKDEP
843 local_irq_save(flags);
844 lock_map_acquire(&timer->lockdep_map);
845 lock_map_release(&timer->lockdep_map);
846 local_irq_restore(flags);
850 int ret = try_to_del_timer_sync(timer);
856 EXPORT_SYMBOL(del_timer_sync);
859 static int cascade(struct tvec_base *base, struct tvec *tv, int index)
861 /* cascade all the timers from tv up one level */
862 struct timer_list *timer, *tmp;
863 struct list_head tv_list;
865 list_replace_init(tv->vec + index, &tv_list);
868 * We are removing _all_ timers from the list, so we
869 * don't have to detach them individually.
871 list_for_each_entry_safe(timer, tmp, &tv_list, entry) {
872 BUG_ON(tbase_get_base(timer->base) != base);
873 internal_add_timer(base, timer);
879 #define INDEX(N) ((base->timer_jiffies >> (TVR_BITS + (N) * TVN_BITS)) & TVN_MASK)
882 * __run_timers - run all expired timers (if any) on this CPU.
883 * @base: the timer vector to be processed.
885 * This function cascades all vectors and executes all expired timer
888 static inline void __run_timers(struct tvec_base *base)
890 struct timer_list *timer;
892 spin_lock_irq(&base->lock);
893 while (time_after_eq(jiffies, base->timer_jiffies)) {
894 struct list_head work_list;
895 struct list_head *head = &work_list;
896 int index = base->timer_jiffies & TVR_MASK;
902 (!cascade(base, &base->tv2, INDEX(0))) &&
903 (!cascade(base, &base->tv3, INDEX(1))) &&
904 !cascade(base, &base->tv4, INDEX(2)))
905 cascade(base, &base->tv5, INDEX(3));
906 ++base->timer_jiffies;
907 list_replace_init(base->tv1.vec + index, &work_list);
908 while (!list_empty(head)) {
909 void (*fn)(unsigned long);
912 timer = list_first_entry(head, struct timer_list,entry);
913 fn = timer->function;
916 timer_stats_account_timer(timer);
918 set_running_timer(base, timer);
919 detach_timer(timer, 1);
921 spin_unlock_irq(&base->lock);
923 int preempt_count = preempt_count();
925 #ifdef CONFIG_LOCKDEP
927 * It is permissible to free the timer from
928 * inside the function that is called from
929 * it, this we need to take into account for
930 * lockdep too. To avoid bogus "held lock
931 * freed" warnings as well as problems when
932 * looking into timer->lockdep_map, make a
933 * copy and use that here.
935 struct lockdep_map lockdep_map =
939 * Couple the lock chain with the lock chain at
940 * del_timer_sync() by acquiring the lock_map
941 * around the fn() call here and in
944 lock_map_acquire(&lockdep_map);
948 lock_map_release(&lockdep_map);
950 if (preempt_count != preempt_count()) {
951 printk(KERN_ERR "huh, entered %p "
952 "with preempt_count %08x, exited"
959 spin_lock_irq(&base->lock);
962 set_running_timer(base, NULL);
963 spin_unlock_irq(&base->lock);
968 * Find out when the next timer event is due to happen. This
969 * is used on S/390 to stop all activity when a cpus is idle.
970 * This functions needs to be called disabled.
972 static unsigned long __next_timer_interrupt(struct tvec_base *base)
974 unsigned long timer_jiffies = base->timer_jiffies;
975 unsigned long expires = timer_jiffies + NEXT_TIMER_MAX_DELTA;
976 int index, slot, array, found = 0;
977 struct timer_list *nte;
978 struct tvec *varray[4];
980 /* Look for timer events in tv1. */
981 index = slot = timer_jiffies & TVR_MASK;
983 list_for_each_entry(nte, base->tv1.vec + slot, entry) {
984 if (tbase_get_deferrable(nte->base))
988 expires = nte->expires;
989 /* Look at the cascade bucket(s)? */
990 if (!index || slot < index)
994 slot = (slot + 1) & TVR_MASK;
995 } while (slot != index);
998 /* Calculate the next cascade event */
1000 timer_jiffies += TVR_SIZE - index;
1001 timer_jiffies >>= TVR_BITS;
1003 /* Check tv2-tv5. */
1004 varray[0] = &base->tv2;
1005 varray[1] = &base->tv3;
1006 varray[2] = &base->tv4;
1007 varray[3] = &base->tv5;
1009 for (array = 0; array < 4; array++) {
1010 struct tvec *varp = varray[array];
1012 index = slot = timer_jiffies & TVN_MASK;
1014 list_for_each_entry(nte, varp->vec + slot, entry) {
1016 if (time_before(nte->expires, expires))
1017 expires = nte->expires;
1020 * Do we still search for the first timer or are
1021 * we looking up the cascade buckets ?
1024 /* Look at the cascade bucket(s)? */
1025 if (!index || slot < index)
1029 slot = (slot + 1) & TVN_MASK;
1030 } while (slot != index);
1033 timer_jiffies += TVN_SIZE - index;
1034 timer_jiffies >>= TVN_BITS;
1040 * Check, if the next hrtimer event is before the next timer wheel
1043 static unsigned long cmp_next_hrtimer_event(unsigned long now,
1044 unsigned long expires)
1046 ktime_t hr_delta = hrtimer_get_next_event();
1047 struct timespec tsdelta;
1048 unsigned long delta;
1050 if (hr_delta.tv64 == KTIME_MAX)
1054 * Expired timer available, let it expire in the next tick
1056 if (hr_delta.tv64 <= 0)
1059 tsdelta = ktime_to_timespec(hr_delta);
1060 delta = timespec_to_jiffies(&tsdelta);
1063 * Limit the delta to the max value, which is checked in
1064 * tick_nohz_stop_sched_tick():
1066 if (delta > NEXT_TIMER_MAX_DELTA)
1067 delta = NEXT_TIMER_MAX_DELTA;
1070 * Take rounding errors in to account and make sure, that it
1071 * expires in the next tick. Otherwise we go into an endless
1072 * ping pong due to tick_nohz_stop_sched_tick() retriggering
1078 if (time_before(now, expires))
1084 * get_next_timer_interrupt - return the jiffy of the next pending timer
1085 * @now: current time (in jiffies)
1087 unsigned long get_next_timer_interrupt(unsigned long now)
1089 struct tvec_base *base = __get_cpu_var(tvec_bases);
1090 unsigned long expires;
1092 spin_lock(&base->lock);
1093 expires = __next_timer_interrupt(base);
1094 spin_unlock(&base->lock);
1096 if (time_before_eq(expires, now))
1099 return cmp_next_hrtimer_event(now, expires);
1104 * Called from the timer interrupt handler to charge one tick to the current
1105 * process. user_tick is 1 if the tick is user time, 0 for system.
1107 void update_process_times(int user_tick)
1109 struct task_struct *p = current;
1110 int cpu = smp_processor_id();
1112 /* Note: this timer irq context must be accounted for as well. */
1113 account_process_tick(p, user_tick);
1115 if (rcu_pending(cpu))
1116 rcu_check_callbacks(cpu, user_tick);
1119 run_posix_cpu_timers(p);
1123 * Nr of active tasks - counted in fixed-point numbers
1125 static unsigned long count_active_tasks(void)
1127 return nr_active() * FIXED_1;
1131 * Hmm.. Changed this, as the GNU make sources (load.c) seems to
1132 * imply that avenrun[] is the standard name for this kind of thing.
1133 * Nothing else seems to be standardized: the fractional size etc
1134 * all seem to differ on different machines.
1136 * Requires xtime_lock to access.
1138 unsigned long avenrun[3];
1140 EXPORT_SYMBOL(avenrun);
1143 * calc_load - given tick count, update the avenrun load estimates.
1144 * This is called while holding a write_lock on xtime_lock.
1146 static inline void calc_load(unsigned long ticks)
1148 unsigned long active_tasks; /* fixed-point */
1149 static int count = LOAD_FREQ;
1152 if (unlikely(count < 0)) {
1153 active_tasks = count_active_tasks();
1155 CALC_LOAD(avenrun[0], EXP_1, active_tasks);
1156 CALC_LOAD(avenrun[1], EXP_5, active_tasks);
1157 CALC_LOAD(avenrun[2], EXP_15, active_tasks);
1159 } while (count < 0);
1164 * This function runs timers and the timer-tq in bottom half context.
1166 static void run_timer_softirq(struct softirq_action *h)
1168 struct tvec_base *base = __get_cpu_var(tvec_bases);
1170 hrtimer_run_pending();
1172 if (time_after_eq(jiffies, base->timer_jiffies))
1177 * Called by the local, per-CPU timer interrupt on SMP.
1179 void run_local_timers(void)
1181 hrtimer_run_queues();
1182 raise_softirq(TIMER_SOFTIRQ);
1187 * Called by the timer interrupt. xtime_lock must already be taken
1190 static inline void update_times(unsigned long ticks)
1197 * The 64-bit jiffies value is not atomic - you MUST NOT read it
1198 * without sampling the sequence number in xtime_lock.
1199 * jiffies is defined in the linker script...
1202 void do_timer(unsigned long ticks)
1204 jiffies_64 += ticks;
1205 update_times(ticks);
1208 #ifdef __ARCH_WANT_SYS_ALARM
1211 * For backwards compatibility? This can be done in libc so Alpha
1212 * and all newer ports shouldn't need it.
1214 SYSCALL_DEFINE1(alarm, unsigned int, seconds)
1216 return alarm_setitimer(seconds);
1224 * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this
1225 * should be moved into arch/i386 instead?
1229 * sys_getpid - return the thread group id of the current process
1231 * Note, despite the name, this returns the tgid not the pid. The tgid and
1232 * the pid are identical unless CLONE_THREAD was specified on clone() in
1233 * which case the tgid is the same in all threads of the same group.
1235 * This is SMP safe as current->tgid does not change.
1237 SYSCALL_DEFINE0(getpid)
1239 return task_tgid_vnr(current);
1243 * Accessing ->real_parent is not SMP-safe, it could
1244 * change from under us. However, we can use a stale
1245 * value of ->real_parent under rcu_read_lock(), see
1246 * release_task()->call_rcu(delayed_put_task_struct).
1248 SYSCALL_DEFINE0(getppid)
1253 pid = task_tgid_vnr(current->real_parent);
1259 SYSCALL_DEFINE0(getuid)
1261 /* Only we change this so SMP safe */
1262 return current_uid();
1265 SYSCALL_DEFINE0(geteuid)
1267 /* Only we change this so SMP safe */
1268 return current_euid();
1271 SYSCALL_DEFINE0(getgid)
1273 /* Only we change this so SMP safe */
1274 return current_gid();
1277 SYSCALL_DEFINE0(getegid)
1279 /* Only we change this so SMP safe */
1280 return current_egid();
1285 static void process_timeout(unsigned long __data)
1287 wake_up_process((struct task_struct *)__data);
1291 * schedule_timeout - sleep until timeout
1292 * @timeout: timeout value in jiffies
1294 * Make the current task sleep until @timeout jiffies have
1295 * elapsed. The routine will return immediately unless
1296 * the current task state has been set (see set_current_state()).
1298 * You can set the task state as follows -
1300 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1301 * pass before the routine returns. The routine will return 0
1303 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1304 * delivered to the current task. In this case the remaining time
1305 * in jiffies will be returned, or 0 if the timer expired in time
1307 * The current task state is guaranteed to be TASK_RUNNING when this
1310 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1311 * the CPU away without a bound on the timeout. In this case the return
1312 * value will be %MAX_SCHEDULE_TIMEOUT.
1314 * In all cases the return value is guaranteed to be non-negative.
1316 signed long __sched schedule_timeout(signed long timeout)
1318 struct timer_list timer;
1319 unsigned long expire;
1323 case MAX_SCHEDULE_TIMEOUT:
1325 * These two special cases are useful to be comfortable
1326 * in the caller. Nothing more. We could take
1327 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1328 * but I' d like to return a valid offset (>=0) to allow
1329 * the caller to do everything it want with the retval.
1335 * Another bit of PARANOID. Note that the retval will be
1336 * 0 since no piece of kernel is supposed to do a check
1337 * for a negative retval of schedule_timeout() (since it
1338 * should never happens anyway). You just have the printk()
1339 * that will tell you if something is gone wrong and where.
1342 printk(KERN_ERR "schedule_timeout: wrong timeout "
1343 "value %lx\n", timeout);
1345 current->state = TASK_RUNNING;
1350 expire = timeout + jiffies;
1352 setup_timer_on_stack(&timer, process_timeout, (unsigned long)current);
1353 __mod_timer(&timer, expire, false);
1355 del_singleshot_timer_sync(&timer);
1357 /* Remove the timer from the object tracker */
1358 destroy_timer_on_stack(&timer);
1360 timeout = expire - jiffies;
1363 return timeout < 0 ? 0 : timeout;
1365 EXPORT_SYMBOL(schedule_timeout);
1368 * We can use __set_current_state() here because schedule_timeout() calls
1369 * schedule() unconditionally.
1371 signed long __sched schedule_timeout_interruptible(signed long timeout)
1373 __set_current_state(TASK_INTERRUPTIBLE);
1374 return schedule_timeout(timeout);
1376 EXPORT_SYMBOL(schedule_timeout_interruptible);
1378 signed long __sched schedule_timeout_killable(signed long timeout)
1380 __set_current_state(TASK_KILLABLE);
1381 return schedule_timeout(timeout);
1383 EXPORT_SYMBOL(schedule_timeout_killable);
1385 signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1387 __set_current_state(TASK_UNINTERRUPTIBLE);
1388 return schedule_timeout(timeout);
1390 EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1392 /* Thread ID - the internal kernel "pid" */
1393 SYSCALL_DEFINE0(gettid)
1395 return task_pid_vnr(current);
1399 * do_sysinfo - fill in sysinfo struct
1400 * @info: pointer to buffer to fill
1402 int do_sysinfo(struct sysinfo *info)
1404 unsigned long mem_total, sav_total;
1405 unsigned int mem_unit, bitcount;
1408 memset(info, 0, sizeof(struct sysinfo));
1412 seq = read_seqbegin(&xtime_lock);
1415 * This is annoying. The below is the same thing
1416 * posix_get_clock_monotonic() does, but it wants to
1417 * take the lock which we want to cover the loads stuff
1421 getnstimeofday(&tp);
1422 tp.tv_sec += wall_to_monotonic.tv_sec;
1423 tp.tv_nsec += wall_to_monotonic.tv_nsec;
1424 monotonic_to_bootbased(&tp);
1425 if (tp.tv_nsec - NSEC_PER_SEC >= 0) {
1426 tp.tv_nsec = tp.tv_nsec - NSEC_PER_SEC;
1429 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
1431 info->loads[0] = avenrun[0] << (SI_LOAD_SHIFT - FSHIFT);
1432 info->loads[1] = avenrun[1] << (SI_LOAD_SHIFT - FSHIFT);
1433 info->loads[2] = avenrun[2] << (SI_LOAD_SHIFT - FSHIFT);
1435 info->procs = nr_threads;
1436 } while (read_seqretry(&xtime_lock, seq));
1442 * If the sum of all the available memory (i.e. ram + swap)
1443 * is less than can be stored in a 32 bit unsigned long then
1444 * we can be binary compatible with 2.2.x kernels. If not,
1445 * well, in that case 2.2.x was broken anyways...
1447 * -Erik Andersen <andersee@debian.org>
1450 mem_total = info->totalram + info->totalswap;
1451 if (mem_total < info->totalram || mem_total < info->totalswap)
1454 mem_unit = info->mem_unit;
1455 while (mem_unit > 1) {
1458 sav_total = mem_total;
1460 if (mem_total < sav_total)
1465 * If mem_total did not overflow, multiply all memory values by
1466 * info->mem_unit and set it to 1. This leaves things compatible
1467 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1472 info->totalram <<= bitcount;
1473 info->freeram <<= bitcount;
1474 info->sharedram <<= bitcount;
1475 info->bufferram <<= bitcount;
1476 info->totalswap <<= bitcount;
1477 info->freeswap <<= bitcount;
1478 info->totalhigh <<= bitcount;
1479 info->freehigh <<= bitcount;
1485 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
1491 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
1497 static int __cpuinit init_timers_cpu(int cpu)
1500 struct tvec_base *base;
1501 static char __cpuinitdata tvec_base_done[NR_CPUS];
1503 if (!tvec_base_done[cpu]) {
1504 static char boot_done;
1508 * The APs use this path later in boot
1510 base = kmalloc_node(sizeof(*base),
1511 GFP_KERNEL | __GFP_ZERO,
1516 /* Make sure that tvec_base is 2 byte aligned */
1517 if (tbase_get_deferrable(base)) {
1522 per_cpu(tvec_bases, cpu) = base;
1525 * This is for the boot CPU - we use compile-time
1526 * static initialisation because per-cpu memory isn't
1527 * ready yet and because the memory allocators are not
1528 * initialised either.
1531 base = &boot_tvec_bases;
1533 tvec_base_done[cpu] = 1;
1535 base = per_cpu(tvec_bases, cpu);
1538 spin_lock_init(&base->lock);
1540 for (j = 0; j < TVN_SIZE; j++) {
1541 INIT_LIST_HEAD(base->tv5.vec + j);
1542 INIT_LIST_HEAD(base->tv4.vec + j);
1543 INIT_LIST_HEAD(base->tv3.vec + j);
1544 INIT_LIST_HEAD(base->tv2.vec + j);
1546 for (j = 0; j < TVR_SIZE; j++)
1547 INIT_LIST_HEAD(base->tv1.vec + j);
1549 base->timer_jiffies = jiffies;
1553 #ifdef CONFIG_HOTPLUG_CPU
1554 static void migrate_timer_list(struct tvec_base *new_base, struct list_head *head)
1556 struct timer_list *timer;
1558 while (!list_empty(head)) {
1559 timer = list_first_entry(head, struct timer_list, entry);
1560 detach_timer(timer, 0);
1561 timer_set_base(timer, new_base);
1562 internal_add_timer(new_base, timer);
1566 static void __cpuinit migrate_timers(int cpu)
1568 struct tvec_base *old_base;
1569 struct tvec_base *new_base;
1572 BUG_ON(cpu_online(cpu));
1573 old_base = per_cpu(tvec_bases, cpu);
1574 new_base = get_cpu_var(tvec_bases);
1576 * The caller is globally serialized and nobody else
1577 * takes two locks at once, deadlock is not possible.
1579 spin_lock_irq(&new_base->lock);
1580 spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1582 BUG_ON(old_base->running_timer);
1584 for (i = 0; i < TVR_SIZE; i++)
1585 migrate_timer_list(new_base, old_base->tv1.vec + i);
1586 for (i = 0; i < TVN_SIZE; i++) {
1587 migrate_timer_list(new_base, old_base->tv2.vec + i);
1588 migrate_timer_list(new_base, old_base->tv3.vec + i);
1589 migrate_timer_list(new_base, old_base->tv4.vec + i);
1590 migrate_timer_list(new_base, old_base->tv5.vec + i);
1593 spin_unlock(&old_base->lock);
1594 spin_unlock_irq(&new_base->lock);
1595 put_cpu_var(tvec_bases);
1597 #endif /* CONFIG_HOTPLUG_CPU */
1599 static int __cpuinit timer_cpu_notify(struct notifier_block *self,
1600 unsigned long action, void *hcpu)
1602 long cpu = (long)hcpu;
1604 case CPU_UP_PREPARE:
1605 case CPU_UP_PREPARE_FROZEN:
1606 if (init_timers_cpu(cpu) < 0)
1609 #ifdef CONFIG_HOTPLUG_CPU
1611 case CPU_DEAD_FROZEN:
1612 migrate_timers(cpu);
1621 static struct notifier_block __cpuinitdata timers_nb = {
1622 .notifier_call = timer_cpu_notify,
1626 void __init init_timers(void)
1628 int err = timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE,
1629 (void *)(long)smp_processor_id());
1633 BUG_ON(err == NOTIFY_BAD);
1634 register_cpu_notifier(&timers_nb);
1635 open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
1639 * msleep - sleep safely even with waitqueue interruptions
1640 * @msecs: Time in milliseconds to sleep for
1642 void msleep(unsigned int msecs)
1644 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1647 timeout = schedule_timeout_uninterruptible(timeout);
1650 EXPORT_SYMBOL(msleep);
1653 * msleep_interruptible - sleep waiting for signals
1654 * @msecs: Time in milliseconds to sleep for
1656 unsigned long msleep_interruptible(unsigned int msecs)
1658 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1660 while (timeout && !signal_pending(current))
1661 timeout = schedule_timeout_interruptible(timeout);
1662 return jiffies_to_msecs(timeout);
1665 EXPORT_SYMBOL(msleep_interruptible);