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/sched.h>
42 #include <asm/uaccess.h>
43 #include <asm/unistd.h>
44 #include <asm/div64.h>
45 #include <asm/timex.h>
48 u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
50 EXPORT_SYMBOL(jiffies_64);
53 * per-CPU timer vector definitions:
55 #define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6)
56 #define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8)
57 #define TVN_SIZE (1 << TVN_BITS)
58 #define TVR_SIZE (1 << TVR_BITS)
59 #define TVN_MASK (TVN_SIZE - 1)
60 #define TVR_MASK (TVR_SIZE - 1)
63 struct list_head vec[TVN_SIZE];
67 struct list_head vec[TVR_SIZE];
72 struct timer_list *running_timer;
73 unsigned long timer_jiffies;
79 } ____cacheline_aligned;
81 struct tvec_base boot_tvec_bases;
82 EXPORT_SYMBOL(boot_tvec_bases);
83 static DEFINE_PER_CPU(struct tvec_base *, tvec_bases) = &boot_tvec_bases;
86 * Note that all tvec_bases are 2 byte aligned and lower bit of
87 * base in timer_list is guaranteed to be zero. Use the LSB for
88 * the new flag to indicate whether the timer is deferrable
90 #define TBASE_DEFERRABLE_FLAG (0x1)
92 /* Functions below help us manage 'deferrable' flag */
93 static inline unsigned int tbase_get_deferrable(struct tvec_base *base)
95 return ((unsigned int)(unsigned long)base & TBASE_DEFERRABLE_FLAG);
98 static inline struct tvec_base *tbase_get_base(struct tvec_base *base)
100 return ((struct tvec_base *)((unsigned long)base & ~TBASE_DEFERRABLE_FLAG));
103 static inline void timer_set_deferrable(struct timer_list *timer)
105 timer->base = ((struct tvec_base *)((unsigned long)(timer->base) |
106 TBASE_DEFERRABLE_FLAG));
110 timer_set_base(struct timer_list *timer, struct tvec_base *new_base)
112 timer->base = (struct tvec_base *)((unsigned long)(new_base) |
113 tbase_get_deferrable(timer->base));
116 static unsigned long round_jiffies_common(unsigned long j, int cpu,
120 unsigned long original = j;
123 * We don't want all cpus firing their timers at once hitting the
124 * same lock or cachelines, so we skew each extra cpu with an extra
125 * 3 jiffies. This 3 jiffies came originally from the mm/ code which
127 * The skew is done by adding 3*cpunr, then round, then subtract this
128 * extra offset again.
135 * If the target jiffie is just after a whole second (which can happen
136 * due to delays of the timer irq, long irq off times etc etc) then
137 * we should round down to the whole second, not up. Use 1/4th second
138 * as cutoff for this rounding as an extreme upper bound for this.
139 * But never round down if @force_up is set.
141 if (rem < HZ/4 && !force_up) /* round down */
146 /* now that we have rounded, subtract the extra skew again */
149 if (j <= jiffies) /* rounding ate our timeout entirely; */
155 * __round_jiffies - function to round jiffies to a full second
156 * @j: the time in (absolute) jiffies that should be rounded
157 * @cpu: the processor number on which the timeout will happen
159 * __round_jiffies() rounds an absolute time in the future (in jiffies)
160 * up or down to (approximately) full seconds. This is useful for timers
161 * for which the exact time they fire does not matter too much, as long as
162 * they fire approximately every X seconds.
164 * By rounding these timers to whole seconds, all such timers will fire
165 * at the same time, rather than at various times spread out. The goal
166 * of this is to have the CPU wake up less, which saves power.
168 * The exact rounding is skewed for each processor to avoid all
169 * processors firing at the exact same time, which could lead
170 * to lock contention or spurious cache line bouncing.
172 * The return value is the rounded version of the @j parameter.
174 unsigned long __round_jiffies(unsigned long j, int cpu)
176 return round_jiffies_common(j, cpu, false);
178 EXPORT_SYMBOL_GPL(__round_jiffies);
181 * __round_jiffies_relative - function to round jiffies to a full second
182 * @j: the time in (relative) jiffies that should be rounded
183 * @cpu: the processor number on which the timeout will happen
185 * __round_jiffies_relative() rounds a time delta in the future (in jiffies)
186 * up or down to (approximately) full seconds. This is useful for timers
187 * for which the exact time they fire does not matter too much, as long as
188 * they fire approximately every X seconds.
190 * By rounding these timers to whole seconds, all such timers will fire
191 * at the same time, rather than at various times spread out. The goal
192 * of this is to have the CPU wake up less, which saves power.
194 * The exact rounding is skewed for each processor to avoid all
195 * processors firing at the exact same time, which could lead
196 * to lock contention or spurious cache line bouncing.
198 * The return value is the rounded version of the @j parameter.
200 unsigned long __round_jiffies_relative(unsigned long j, int cpu)
202 unsigned long j0 = jiffies;
204 /* Use j0 because jiffies might change while we run */
205 return round_jiffies_common(j + j0, cpu, false) - j0;
207 EXPORT_SYMBOL_GPL(__round_jiffies_relative);
210 * round_jiffies - function to round jiffies to a full second
211 * @j: the time in (absolute) jiffies that should be rounded
213 * round_jiffies() rounds an absolute time in the future (in jiffies)
214 * up or down to (approximately) full seconds. This is useful for timers
215 * for which the exact time they fire does not matter too much, as long as
216 * they fire approximately every X seconds.
218 * By rounding these timers to whole seconds, all such timers will fire
219 * at the same time, rather than at various times spread out. The goal
220 * of this is to have the CPU wake up less, which saves power.
222 * The return value is the rounded version of the @j parameter.
224 unsigned long round_jiffies(unsigned long j)
226 return round_jiffies_common(j, raw_smp_processor_id(), false);
228 EXPORT_SYMBOL_GPL(round_jiffies);
231 * round_jiffies_relative - function to round jiffies to a full second
232 * @j: the time in (relative) jiffies that should be rounded
234 * round_jiffies_relative() rounds a time delta in the future (in jiffies)
235 * up or down to (approximately) full seconds. This is useful for timers
236 * for which the exact time they fire does not matter too much, as long as
237 * they fire approximately every X seconds.
239 * By rounding these timers to whole seconds, all such timers will fire
240 * at the same time, rather than at various times spread out. The goal
241 * of this is to have the CPU wake up less, which saves power.
243 * The return value is the rounded version of the @j parameter.
245 unsigned long round_jiffies_relative(unsigned long j)
247 return __round_jiffies_relative(j, raw_smp_processor_id());
249 EXPORT_SYMBOL_GPL(round_jiffies_relative);
252 * __round_jiffies_up - function to round jiffies up to a full second
253 * @j: the time in (absolute) jiffies that should be rounded
254 * @cpu: the processor number on which the timeout will happen
256 * This is the same as __round_jiffies() except that it will never
257 * round down. This is useful for timeouts for which the exact time
258 * of firing does not matter too much, as long as they don't fire too
261 unsigned long __round_jiffies_up(unsigned long j, int cpu)
263 return round_jiffies_common(j, cpu, true);
265 EXPORT_SYMBOL_GPL(__round_jiffies_up);
268 * __round_jiffies_up_relative - function to round jiffies up to a full second
269 * @j: the time in (relative) jiffies that should be rounded
270 * @cpu: the processor number on which the timeout will happen
272 * This is the same as __round_jiffies_relative() except that it will never
273 * round down. This is useful for timeouts for which the exact time
274 * of firing does not matter too much, as long as they don't fire too
277 unsigned long __round_jiffies_up_relative(unsigned long j, int cpu)
279 unsigned long j0 = jiffies;
281 /* Use j0 because jiffies might change while we run */
282 return round_jiffies_common(j + j0, cpu, true) - j0;
284 EXPORT_SYMBOL_GPL(__round_jiffies_up_relative);
287 * round_jiffies_up - function to round jiffies up to a full second
288 * @j: the time in (absolute) jiffies that should be rounded
290 * This is the same as round_jiffies() except that it will never
291 * round down. This is useful for timeouts for which the exact time
292 * of firing does not matter too much, as long as they don't fire too
295 unsigned long round_jiffies_up(unsigned long j)
297 return round_jiffies_common(j, raw_smp_processor_id(), true);
299 EXPORT_SYMBOL_GPL(round_jiffies_up);
302 * round_jiffies_up_relative - function to round jiffies up to a full second
303 * @j: the time in (relative) jiffies that should be rounded
305 * This is the same as round_jiffies_relative() except that it will never
306 * round down. This is useful for timeouts for which the exact time
307 * of firing does not matter too much, as long as they don't fire too
310 unsigned long round_jiffies_up_relative(unsigned long j)
312 return __round_jiffies_up_relative(j, raw_smp_processor_id());
314 EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
317 static inline void set_running_timer(struct tvec_base *base,
318 struct timer_list *timer)
321 base->running_timer = timer;
325 static void internal_add_timer(struct tvec_base *base, struct timer_list *timer)
327 unsigned long expires = timer->expires;
328 unsigned long idx = expires - base->timer_jiffies;
329 struct list_head *vec;
331 if (idx < TVR_SIZE) {
332 int i = expires & TVR_MASK;
333 vec = base->tv1.vec + i;
334 } else if (idx < 1 << (TVR_BITS + TVN_BITS)) {
335 int i = (expires >> TVR_BITS) & TVN_MASK;
336 vec = base->tv2.vec + i;
337 } else if (idx < 1 << (TVR_BITS + 2 * TVN_BITS)) {
338 int i = (expires >> (TVR_BITS + TVN_BITS)) & TVN_MASK;
339 vec = base->tv3.vec + i;
340 } else if (idx < 1 << (TVR_BITS + 3 * TVN_BITS)) {
341 int i = (expires >> (TVR_BITS + 2 * TVN_BITS)) & TVN_MASK;
342 vec = base->tv4.vec + i;
343 } else if ((signed long) idx < 0) {
345 * Can happen if you add a timer with expires == jiffies,
346 * or you set a timer to go off in the past
348 vec = base->tv1.vec + (base->timer_jiffies & TVR_MASK);
351 /* If the timeout is larger than 0xffffffff on 64-bit
352 * architectures then we use the maximum timeout:
354 if (idx > 0xffffffffUL) {
356 expires = idx + base->timer_jiffies;
358 i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK;
359 vec = base->tv5.vec + i;
364 list_add_tail(&timer->entry, vec);
367 #ifdef CONFIG_TIMER_STATS
368 void __timer_stats_timer_set_start_info(struct timer_list *timer, void *addr)
370 if (timer->start_site)
373 timer->start_site = addr;
374 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
375 timer->start_pid = current->pid;
378 static void timer_stats_account_timer(struct timer_list *timer)
380 unsigned int flag = 0;
382 if (unlikely(tbase_get_deferrable(timer->base)))
383 flag |= TIMER_STATS_FLAG_DEFERRABLE;
385 timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
386 timer->function, timer->start_comm, flag);
390 static void timer_stats_account_timer(struct timer_list *timer) {}
393 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
395 static struct debug_obj_descr timer_debug_descr;
398 * fixup_init is called when:
399 * - an active object is initialized
401 static int timer_fixup_init(void *addr, enum debug_obj_state state)
403 struct timer_list *timer = addr;
406 case ODEBUG_STATE_ACTIVE:
407 del_timer_sync(timer);
408 debug_object_init(timer, &timer_debug_descr);
416 * fixup_activate is called when:
417 * - an active object is activated
418 * - an unknown object is activated (might be a statically initialized object)
420 static int timer_fixup_activate(void *addr, enum debug_obj_state state)
422 struct timer_list *timer = addr;
426 case ODEBUG_STATE_NOTAVAILABLE:
428 * This is not really a fixup. The timer was
429 * statically initialized. We just make sure that it
430 * is tracked in the object tracker.
432 if (timer->entry.next == NULL &&
433 timer->entry.prev == TIMER_ENTRY_STATIC) {
434 debug_object_init(timer, &timer_debug_descr);
435 debug_object_activate(timer, &timer_debug_descr);
442 case ODEBUG_STATE_ACTIVE:
451 * fixup_free is called when:
452 * - an active object is freed
454 static int timer_fixup_free(void *addr, enum debug_obj_state state)
456 struct timer_list *timer = addr;
459 case ODEBUG_STATE_ACTIVE:
460 del_timer_sync(timer);
461 debug_object_free(timer, &timer_debug_descr);
468 static struct debug_obj_descr timer_debug_descr = {
469 .name = "timer_list",
470 .fixup_init = timer_fixup_init,
471 .fixup_activate = timer_fixup_activate,
472 .fixup_free = timer_fixup_free,
475 static inline void debug_timer_init(struct timer_list *timer)
477 debug_object_init(timer, &timer_debug_descr);
480 static inline void debug_timer_activate(struct timer_list *timer)
482 debug_object_activate(timer, &timer_debug_descr);
485 static inline void debug_timer_deactivate(struct timer_list *timer)
487 debug_object_deactivate(timer, &timer_debug_descr);
490 static inline void debug_timer_free(struct timer_list *timer)
492 debug_object_free(timer, &timer_debug_descr);
495 static void __init_timer(struct timer_list *timer,
497 struct lock_class_key *key);
499 void init_timer_on_stack_key(struct timer_list *timer,
501 struct lock_class_key *key)
503 debug_object_init_on_stack(timer, &timer_debug_descr);
504 __init_timer(timer, name, key);
506 EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
508 void destroy_timer_on_stack(struct timer_list *timer)
510 debug_object_free(timer, &timer_debug_descr);
512 EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
515 static inline void debug_timer_init(struct timer_list *timer) { }
516 static inline void debug_timer_activate(struct timer_list *timer) { }
517 static inline void debug_timer_deactivate(struct timer_list *timer) { }
520 static void __init_timer(struct timer_list *timer,
522 struct lock_class_key *key)
524 timer->entry.next = NULL;
525 timer->base = __raw_get_cpu_var(tvec_bases);
526 #ifdef CONFIG_TIMER_STATS
527 timer->start_site = NULL;
528 timer->start_pid = -1;
529 memset(timer->start_comm, 0, TASK_COMM_LEN);
531 lockdep_init_map(&timer->lockdep_map, name, key, 0);
535 * init_timer_key - initialize a timer
536 * @timer: the timer to be initialized
537 * @name: name of the timer
538 * @key: lockdep class key of the fake lock used for tracking timer
539 * sync lock dependencies
541 * init_timer_key() must be done to a timer prior calling *any* of the
542 * other timer functions.
544 void init_timer_key(struct timer_list *timer,
546 struct lock_class_key *key)
548 debug_timer_init(timer);
549 __init_timer(timer, name, key);
551 EXPORT_SYMBOL(init_timer_key);
553 void init_timer_deferrable_key(struct timer_list *timer,
555 struct lock_class_key *key)
557 init_timer_key(timer, name, key);
558 timer_set_deferrable(timer);
560 EXPORT_SYMBOL(init_timer_deferrable_key);
562 static inline void detach_timer(struct timer_list *timer,
565 struct list_head *entry = &timer->entry;
567 debug_timer_deactivate(timer);
569 __list_del(entry->prev, entry->next);
572 entry->prev = LIST_POISON2;
576 * We are using hashed locking: holding per_cpu(tvec_bases).lock
577 * means that all timers which are tied to this base via timer->base are
578 * locked, and the base itself is locked too.
580 * So __run_timers/migrate_timers can safely modify all timers which could
581 * be found on ->tvX lists.
583 * When the timer's base is locked, and the timer removed from list, it is
584 * possible to set timer->base = NULL and drop the lock: the timer remains
587 static struct tvec_base *lock_timer_base(struct timer_list *timer,
588 unsigned long *flags)
589 __acquires(timer->base->lock)
591 struct tvec_base *base;
594 struct tvec_base *prelock_base = timer->base;
595 base = tbase_get_base(prelock_base);
596 if (likely(base != NULL)) {
597 spin_lock_irqsave(&base->lock, *flags);
598 if (likely(prelock_base == timer->base))
600 /* The timer has migrated to another CPU */
601 spin_unlock_irqrestore(&base->lock, *flags);
608 __mod_timer(struct timer_list *timer, unsigned long expires,
609 bool pending_only, int pinned)
611 struct tvec_base *base, *new_base;
615 timer_stats_timer_set_start_info(timer);
616 BUG_ON(!timer->function);
618 base = lock_timer_base(timer, &flags);
620 if (timer_pending(timer)) {
621 detach_timer(timer, 0);
628 debug_timer_activate(timer);
630 new_base = __get_cpu_var(tvec_bases);
632 cpu = smp_processor_id();
634 #if defined(CONFIG_NO_HZ) && defined(CONFIG_SMP)
635 if (!pinned && get_sysctl_timer_migration() && idle_cpu(cpu)) {
636 int preferred_cpu = get_nohz_load_balancer();
638 if (preferred_cpu >= 0)
642 new_base = per_cpu(tvec_bases, cpu);
644 if (base != new_base) {
646 * We are trying to schedule the timer on the local CPU.
647 * However we can't change timer's base while it is running,
648 * otherwise del_timer_sync() can't detect that the timer's
649 * handler yet has not finished. This also guarantees that
650 * the timer is serialized wrt itself.
652 if (likely(base->running_timer != timer)) {
653 /* See the comment in lock_timer_base() */
654 timer_set_base(timer, NULL);
655 spin_unlock(&base->lock);
657 spin_lock(&base->lock);
658 timer_set_base(timer, base);
662 timer->expires = expires;
663 internal_add_timer(base, timer);
666 spin_unlock_irqrestore(&base->lock, flags);
672 * mod_timer_pending - modify a pending timer's timeout
673 * @timer: the pending timer to be modified
674 * @expires: new timeout in jiffies
676 * mod_timer_pending() is the same for pending timers as mod_timer(),
677 * but will not re-activate and modify already deleted timers.
679 * It is useful for unserialized use of timers.
681 int mod_timer_pending(struct timer_list *timer, unsigned long expires)
683 return __mod_timer(timer, expires, true, TIMER_NOT_PINNED);
685 EXPORT_SYMBOL(mod_timer_pending);
688 * mod_timer - modify a timer's timeout
689 * @timer: the timer to be modified
690 * @expires: new timeout in jiffies
692 * mod_timer() is a more efficient way to update the expire field of an
693 * active timer (if the timer is inactive it will be activated)
695 * mod_timer(timer, expires) is equivalent to:
697 * del_timer(timer); timer->expires = expires; add_timer(timer);
699 * Note that if there are multiple unserialized concurrent users of the
700 * same timer, then mod_timer() is the only safe way to modify the timeout,
701 * since add_timer() cannot modify an already running timer.
703 * The function returns whether it has modified a pending timer or not.
704 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
705 * active timer returns 1.)
707 int mod_timer(struct timer_list *timer, unsigned long expires)
710 * This is a common optimization triggered by the
711 * networking code - if the timer is re-modified
712 * to be the same thing then just return:
714 if (timer->expires == expires && timer_pending(timer))
717 return __mod_timer(timer, expires, false, TIMER_NOT_PINNED);
719 EXPORT_SYMBOL(mod_timer);
722 * mod_timer_pinned - modify a timer's timeout
723 * @timer: the timer to be modified
724 * @expires: new timeout in jiffies
726 * mod_timer_pinned() is a way to update the expire field of an
727 * active timer (if the timer is inactive it will be activated)
728 * and not allow the timer to be migrated to a different CPU.
730 * mod_timer_pinned(timer, expires) is equivalent to:
732 * del_timer(timer); timer->expires = expires; add_timer(timer);
734 int mod_timer_pinned(struct timer_list *timer, unsigned long expires)
736 if (timer->expires == expires && timer_pending(timer))
739 return __mod_timer(timer, expires, false, TIMER_PINNED);
741 EXPORT_SYMBOL(mod_timer_pinned);
744 * add_timer - start a timer
745 * @timer: the timer to be added
747 * The kernel will do a ->function(->data) callback from the
748 * timer interrupt at the ->expires point in the future. The
749 * current time is 'jiffies'.
751 * The timer's ->expires, ->function (and if the handler uses it, ->data)
752 * fields must be set prior calling this function.
754 * Timers with an ->expires field in the past will be executed in the next
757 void add_timer(struct timer_list *timer)
759 BUG_ON(timer_pending(timer));
760 mod_timer(timer, timer->expires);
762 EXPORT_SYMBOL(add_timer);
765 * add_timer_on - start a timer on a particular CPU
766 * @timer: the timer to be added
767 * @cpu: the CPU to start it on
769 * This is not very scalable on SMP. Double adds are not possible.
771 void add_timer_on(struct timer_list *timer, int cpu)
773 struct tvec_base *base = per_cpu(tvec_bases, cpu);
776 timer_stats_timer_set_start_info(timer);
777 BUG_ON(timer_pending(timer) || !timer->function);
778 spin_lock_irqsave(&base->lock, flags);
779 timer_set_base(timer, base);
780 debug_timer_activate(timer);
781 internal_add_timer(base, timer);
783 * Check whether the other CPU is idle and needs to be
784 * triggered to reevaluate the timer wheel when nohz is
785 * active. We are protected against the other CPU fiddling
786 * with the timer by holding the timer base lock. This also
787 * makes sure that a CPU on the way to idle can not evaluate
790 wake_up_idle_cpu(cpu);
791 spin_unlock_irqrestore(&base->lock, flags);
795 * del_timer - deactive a timer.
796 * @timer: the timer to be deactivated
798 * del_timer() deactivates a timer - this works on both active and inactive
801 * The function returns whether it has deactivated a pending timer or not.
802 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
803 * active timer returns 1.)
805 int del_timer(struct timer_list *timer)
807 struct tvec_base *base;
811 timer_stats_timer_clear_start_info(timer);
812 if (timer_pending(timer)) {
813 base = lock_timer_base(timer, &flags);
814 if (timer_pending(timer)) {
815 detach_timer(timer, 1);
818 spin_unlock_irqrestore(&base->lock, flags);
823 EXPORT_SYMBOL(del_timer);
827 * try_to_del_timer_sync - Try to deactivate a timer
828 * @timer: timer do del
830 * This function tries to deactivate a timer. Upon successful (ret >= 0)
831 * exit the timer is not queued and the handler is not running on any CPU.
833 * It must not be called from interrupt contexts.
835 int try_to_del_timer_sync(struct timer_list *timer)
837 struct tvec_base *base;
841 base = lock_timer_base(timer, &flags);
843 if (base->running_timer == timer)
847 if (timer_pending(timer)) {
848 detach_timer(timer, 1);
852 spin_unlock_irqrestore(&base->lock, flags);
856 EXPORT_SYMBOL(try_to_del_timer_sync);
859 * del_timer_sync - deactivate a timer and wait for the handler to finish.
860 * @timer: the timer to be deactivated
862 * This function only differs from del_timer() on SMP: besides deactivating
863 * the timer it also makes sure the handler has finished executing on other
866 * Synchronization rules: Callers must prevent restarting of the timer,
867 * otherwise this function is meaningless. It must not be called from
868 * interrupt contexts. The caller must not hold locks which would prevent
869 * completion of the timer's handler. The timer's handler must not call
870 * add_timer_on(). Upon exit the timer is not queued and the handler is
871 * not running on any CPU.
873 * The function returns whether it has deactivated a pending timer or not.
875 int del_timer_sync(struct timer_list *timer)
877 #ifdef CONFIG_LOCKDEP
880 local_irq_save(flags);
881 lock_map_acquire(&timer->lockdep_map);
882 lock_map_release(&timer->lockdep_map);
883 local_irq_restore(flags);
887 int ret = try_to_del_timer_sync(timer);
893 EXPORT_SYMBOL(del_timer_sync);
896 static int cascade(struct tvec_base *base, struct tvec *tv, int index)
898 /* cascade all the timers from tv up one level */
899 struct timer_list *timer, *tmp;
900 struct list_head tv_list;
902 list_replace_init(tv->vec + index, &tv_list);
905 * We are removing _all_ timers from the list, so we
906 * don't have to detach them individually.
908 list_for_each_entry_safe(timer, tmp, &tv_list, entry) {
909 BUG_ON(tbase_get_base(timer->base) != base);
910 internal_add_timer(base, timer);
916 #define INDEX(N) ((base->timer_jiffies >> (TVR_BITS + (N) * TVN_BITS)) & TVN_MASK)
919 * __run_timers - run all expired timers (if any) on this CPU.
920 * @base: the timer vector to be processed.
922 * This function cascades all vectors and executes all expired timer
925 static inline void __run_timers(struct tvec_base *base)
927 struct timer_list *timer;
929 spin_lock_irq(&base->lock);
930 while (time_after_eq(jiffies, base->timer_jiffies)) {
931 struct list_head work_list;
932 struct list_head *head = &work_list;
933 int index = base->timer_jiffies & TVR_MASK;
939 (!cascade(base, &base->tv2, INDEX(0))) &&
940 (!cascade(base, &base->tv3, INDEX(1))) &&
941 !cascade(base, &base->tv4, INDEX(2)))
942 cascade(base, &base->tv5, INDEX(3));
943 ++base->timer_jiffies;
944 list_replace_init(base->tv1.vec + index, &work_list);
945 while (!list_empty(head)) {
946 void (*fn)(unsigned long);
949 timer = list_first_entry(head, struct timer_list,entry);
950 fn = timer->function;
953 timer_stats_account_timer(timer);
955 set_running_timer(base, timer);
956 detach_timer(timer, 1);
958 spin_unlock_irq(&base->lock);
960 int preempt_count = preempt_count();
962 #ifdef CONFIG_LOCKDEP
964 * It is permissible to free the timer from
965 * inside the function that is called from
966 * it, this we need to take into account for
967 * lockdep too. To avoid bogus "held lock
968 * freed" warnings as well as problems when
969 * looking into timer->lockdep_map, make a
970 * copy and use that here.
972 struct lockdep_map lockdep_map =
976 * Couple the lock chain with the lock chain at
977 * del_timer_sync() by acquiring the lock_map
978 * around the fn() call here and in
981 lock_map_acquire(&lockdep_map);
985 lock_map_release(&lockdep_map);
987 if (preempt_count != preempt_count()) {
988 printk(KERN_ERR "huh, entered %p "
989 "with preempt_count %08x, exited"
996 spin_lock_irq(&base->lock);
999 set_running_timer(base, NULL);
1000 spin_unlock_irq(&base->lock);
1005 * Find out when the next timer event is due to happen. This
1006 * is used on S/390 to stop all activity when a cpus is idle.
1007 * This functions needs to be called disabled.
1009 static unsigned long __next_timer_interrupt(struct tvec_base *base)
1011 unsigned long timer_jiffies = base->timer_jiffies;
1012 unsigned long expires = timer_jiffies + NEXT_TIMER_MAX_DELTA;
1013 int index, slot, array, found = 0;
1014 struct timer_list *nte;
1015 struct tvec *varray[4];
1017 /* Look for timer events in tv1. */
1018 index = slot = timer_jiffies & TVR_MASK;
1020 list_for_each_entry(nte, base->tv1.vec + slot, entry) {
1021 if (tbase_get_deferrable(nte->base))
1025 expires = nte->expires;
1026 /* Look at the cascade bucket(s)? */
1027 if (!index || slot < index)
1031 slot = (slot + 1) & TVR_MASK;
1032 } while (slot != index);
1035 /* Calculate the next cascade event */
1037 timer_jiffies += TVR_SIZE - index;
1038 timer_jiffies >>= TVR_BITS;
1040 /* Check tv2-tv5. */
1041 varray[0] = &base->tv2;
1042 varray[1] = &base->tv3;
1043 varray[2] = &base->tv4;
1044 varray[3] = &base->tv5;
1046 for (array = 0; array < 4; array++) {
1047 struct tvec *varp = varray[array];
1049 index = slot = timer_jiffies & TVN_MASK;
1051 list_for_each_entry(nte, varp->vec + slot, entry) {
1052 if (tbase_get_deferrable(nte->base))
1056 if (time_before(nte->expires, expires))
1057 expires = nte->expires;
1060 * Do we still search for the first timer or are
1061 * we looking up the cascade buckets ?
1064 /* Look at the cascade bucket(s)? */
1065 if (!index || slot < index)
1069 slot = (slot + 1) & TVN_MASK;
1070 } while (slot != index);
1073 timer_jiffies += TVN_SIZE - index;
1074 timer_jiffies >>= TVN_BITS;
1080 * Check, if the next hrtimer event is before the next timer wheel
1083 static unsigned long cmp_next_hrtimer_event(unsigned long now,
1084 unsigned long expires)
1086 ktime_t hr_delta = hrtimer_get_next_event();
1087 struct timespec tsdelta;
1088 unsigned long delta;
1090 if (hr_delta.tv64 == KTIME_MAX)
1094 * Expired timer available, let it expire in the next tick
1096 if (hr_delta.tv64 <= 0)
1099 tsdelta = ktime_to_timespec(hr_delta);
1100 delta = timespec_to_jiffies(&tsdelta);
1103 * Limit the delta to the max value, which is checked in
1104 * tick_nohz_stop_sched_tick():
1106 if (delta > NEXT_TIMER_MAX_DELTA)
1107 delta = NEXT_TIMER_MAX_DELTA;
1110 * Take rounding errors in to account and make sure, that it
1111 * expires in the next tick. Otherwise we go into an endless
1112 * ping pong due to tick_nohz_stop_sched_tick() retriggering
1118 if (time_before(now, expires))
1124 * get_next_timer_interrupt - return the jiffy of the next pending timer
1125 * @now: current time (in jiffies)
1127 unsigned long get_next_timer_interrupt(unsigned long now)
1129 struct tvec_base *base = __get_cpu_var(tvec_bases);
1130 unsigned long expires;
1132 spin_lock(&base->lock);
1133 expires = __next_timer_interrupt(base);
1134 spin_unlock(&base->lock);
1136 if (time_before_eq(expires, now))
1139 return cmp_next_hrtimer_event(now, expires);
1144 * Called from the timer interrupt handler to charge one tick to the current
1145 * process. user_tick is 1 if the tick is user time, 0 for system.
1147 void update_process_times(int user_tick)
1149 struct task_struct *p = current;
1150 int cpu = smp_processor_id();
1152 /* Note: this timer irq context must be accounted for as well. */
1153 account_process_tick(p, user_tick);
1155 if (rcu_pending(cpu))
1156 rcu_check_callbacks(cpu, user_tick);
1159 run_posix_cpu_timers(p);
1163 * Nr of active tasks - counted in fixed-point numbers
1165 static unsigned long count_active_tasks(void)
1167 return nr_active() * FIXED_1;
1171 * Hmm.. Changed this, as the GNU make sources (load.c) seems to
1172 * imply that avenrun[] is the standard name for this kind of thing.
1173 * Nothing else seems to be standardized: the fractional size etc
1174 * all seem to differ on different machines.
1176 * Requires xtime_lock to access.
1178 unsigned long avenrun[3];
1180 EXPORT_SYMBOL(avenrun);
1183 * calc_load - given tick count, update the avenrun load estimates.
1184 * This is called while holding a write_lock on xtime_lock.
1186 static inline void calc_load(unsigned long ticks)
1188 unsigned long active_tasks; /* fixed-point */
1189 static int count = LOAD_FREQ;
1192 if (unlikely(count < 0)) {
1193 active_tasks = count_active_tasks();
1195 CALC_LOAD(avenrun[0], EXP_1, active_tasks);
1196 CALC_LOAD(avenrun[1], EXP_5, active_tasks);
1197 CALC_LOAD(avenrun[2], EXP_15, active_tasks);
1199 } while (count < 0);
1204 * This function runs timers and the timer-tq in bottom half context.
1206 static void run_timer_softirq(struct softirq_action *h)
1208 struct tvec_base *base = __get_cpu_var(tvec_bases);
1210 hrtimer_run_pending();
1212 if (time_after_eq(jiffies, base->timer_jiffies))
1217 * Called by the local, per-CPU timer interrupt on SMP.
1219 void run_local_timers(void)
1221 hrtimer_run_queues();
1222 raise_softirq(TIMER_SOFTIRQ);
1227 * Called by the timer interrupt. xtime_lock must already be taken
1230 static inline void update_times(unsigned long ticks)
1237 * The 64-bit jiffies value is not atomic - you MUST NOT read it
1238 * without sampling the sequence number in xtime_lock.
1239 * jiffies is defined in the linker script...
1242 void do_timer(unsigned long ticks)
1244 jiffies_64 += ticks;
1245 update_times(ticks);
1248 #ifdef __ARCH_WANT_SYS_ALARM
1251 * For backwards compatibility? This can be done in libc so Alpha
1252 * and all newer ports shouldn't need it.
1254 SYSCALL_DEFINE1(alarm, unsigned int, seconds)
1256 return alarm_setitimer(seconds);
1264 * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this
1265 * should be moved into arch/i386 instead?
1269 * sys_getpid - return the thread group id of the current process
1271 * Note, despite the name, this returns the tgid not the pid. The tgid and
1272 * the pid are identical unless CLONE_THREAD was specified on clone() in
1273 * which case the tgid is the same in all threads of the same group.
1275 * This is SMP safe as current->tgid does not change.
1277 SYSCALL_DEFINE0(getpid)
1279 return task_tgid_vnr(current);
1283 * Accessing ->real_parent is not SMP-safe, it could
1284 * change from under us. However, we can use a stale
1285 * value of ->real_parent under rcu_read_lock(), see
1286 * release_task()->call_rcu(delayed_put_task_struct).
1288 SYSCALL_DEFINE0(getppid)
1293 pid = task_tgid_vnr(current->real_parent);
1299 SYSCALL_DEFINE0(getuid)
1301 /* Only we change this so SMP safe */
1302 return current_uid();
1305 SYSCALL_DEFINE0(geteuid)
1307 /* Only we change this so SMP safe */
1308 return current_euid();
1311 SYSCALL_DEFINE0(getgid)
1313 /* Only we change this so SMP safe */
1314 return current_gid();
1317 SYSCALL_DEFINE0(getegid)
1319 /* Only we change this so SMP safe */
1320 return current_egid();
1325 static void process_timeout(unsigned long __data)
1327 wake_up_process((struct task_struct *)__data);
1331 * schedule_timeout - sleep until timeout
1332 * @timeout: timeout value in jiffies
1334 * Make the current task sleep until @timeout jiffies have
1335 * elapsed. The routine will return immediately unless
1336 * the current task state has been set (see set_current_state()).
1338 * You can set the task state as follows -
1340 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1341 * pass before the routine returns. The routine will return 0
1343 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1344 * delivered to the current task. In this case the remaining time
1345 * in jiffies will be returned, or 0 if the timer expired in time
1347 * The current task state is guaranteed to be TASK_RUNNING when this
1350 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1351 * the CPU away without a bound on the timeout. In this case the return
1352 * value will be %MAX_SCHEDULE_TIMEOUT.
1354 * In all cases the return value is guaranteed to be non-negative.
1356 signed long __sched schedule_timeout(signed long timeout)
1358 struct timer_list timer;
1359 unsigned long expire;
1363 case MAX_SCHEDULE_TIMEOUT:
1365 * These two special cases are useful to be comfortable
1366 * in the caller. Nothing more. We could take
1367 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1368 * but I' d like to return a valid offset (>=0) to allow
1369 * the caller to do everything it want with the retval.
1375 * Another bit of PARANOID. Note that the retval will be
1376 * 0 since no piece of kernel is supposed to do a check
1377 * for a negative retval of schedule_timeout() (since it
1378 * should never happens anyway). You just have the printk()
1379 * that will tell you if something is gone wrong and where.
1382 printk(KERN_ERR "schedule_timeout: wrong timeout "
1383 "value %lx\n", timeout);
1385 current->state = TASK_RUNNING;
1390 expire = timeout + jiffies;
1392 setup_timer_on_stack(&timer, process_timeout, (unsigned long)current);
1393 __mod_timer(&timer, expire, false, TIMER_NOT_PINNED);
1395 del_singleshot_timer_sync(&timer);
1397 /* Remove the timer from the object tracker */
1398 destroy_timer_on_stack(&timer);
1400 timeout = expire - jiffies;
1403 return timeout < 0 ? 0 : timeout;
1405 EXPORT_SYMBOL(schedule_timeout);
1408 * We can use __set_current_state() here because schedule_timeout() calls
1409 * schedule() unconditionally.
1411 signed long __sched schedule_timeout_interruptible(signed long timeout)
1413 __set_current_state(TASK_INTERRUPTIBLE);
1414 return schedule_timeout(timeout);
1416 EXPORT_SYMBOL(schedule_timeout_interruptible);
1418 signed long __sched schedule_timeout_killable(signed long timeout)
1420 __set_current_state(TASK_KILLABLE);
1421 return schedule_timeout(timeout);
1423 EXPORT_SYMBOL(schedule_timeout_killable);
1425 signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1427 __set_current_state(TASK_UNINTERRUPTIBLE);
1428 return schedule_timeout(timeout);
1430 EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1432 /* Thread ID - the internal kernel "pid" */
1433 SYSCALL_DEFINE0(gettid)
1435 return task_pid_vnr(current);
1439 * do_sysinfo - fill in sysinfo struct
1440 * @info: pointer to buffer to fill
1442 int do_sysinfo(struct sysinfo *info)
1444 unsigned long mem_total, sav_total;
1445 unsigned int mem_unit, bitcount;
1448 memset(info, 0, sizeof(struct sysinfo));
1452 seq = read_seqbegin(&xtime_lock);
1455 * This is annoying. The below is the same thing
1456 * posix_get_clock_monotonic() does, but it wants to
1457 * take the lock which we want to cover the loads stuff
1461 getnstimeofday(&tp);
1462 tp.tv_sec += wall_to_monotonic.tv_sec;
1463 tp.tv_nsec += wall_to_monotonic.tv_nsec;
1464 monotonic_to_bootbased(&tp);
1465 if (tp.tv_nsec - NSEC_PER_SEC >= 0) {
1466 tp.tv_nsec = tp.tv_nsec - NSEC_PER_SEC;
1469 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
1471 info->loads[0] = avenrun[0] << (SI_LOAD_SHIFT - FSHIFT);
1472 info->loads[1] = avenrun[1] << (SI_LOAD_SHIFT - FSHIFT);
1473 info->loads[2] = avenrun[2] << (SI_LOAD_SHIFT - FSHIFT);
1475 info->procs = nr_threads;
1476 } while (read_seqretry(&xtime_lock, seq));
1482 * If the sum of all the available memory (i.e. ram + swap)
1483 * is less than can be stored in a 32 bit unsigned long then
1484 * we can be binary compatible with 2.2.x kernels. If not,
1485 * well, in that case 2.2.x was broken anyways...
1487 * -Erik Andersen <andersee@debian.org>
1490 mem_total = info->totalram + info->totalswap;
1491 if (mem_total < info->totalram || mem_total < info->totalswap)
1494 mem_unit = info->mem_unit;
1495 while (mem_unit > 1) {
1498 sav_total = mem_total;
1500 if (mem_total < sav_total)
1505 * If mem_total did not overflow, multiply all memory values by
1506 * info->mem_unit and set it to 1. This leaves things compatible
1507 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1512 info->totalram <<= bitcount;
1513 info->freeram <<= bitcount;
1514 info->sharedram <<= bitcount;
1515 info->bufferram <<= bitcount;
1516 info->totalswap <<= bitcount;
1517 info->freeswap <<= bitcount;
1518 info->totalhigh <<= bitcount;
1519 info->freehigh <<= bitcount;
1525 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
1531 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
1537 static int __cpuinit init_timers_cpu(int cpu)
1540 struct tvec_base *base;
1541 static char __cpuinitdata tvec_base_done[NR_CPUS];
1543 if (!tvec_base_done[cpu]) {
1544 static char boot_done;
1548 * The APs use this path later in boot
1550 base = kmalloc_node(sizeof(*base),
1551 GFP_KERNEL | __GFP_ZERO,
1556 /* Make sure that tvec_base is 2 byte aligned */
1557 if (tbase_get_deferrable(base)) {
1562 per_cpu(tvec_bases, cpu) = base;
1565 * This is for the boot CPU - we use compile-time
1566 * static initialisation because per-cpu memory isn't
1567 * ready yet and because the memory allocators are not
1568 * initialised either.
1571 base = &boot_tvec_bases;
1573 tvec_base_done[cpu] = 1;
1575 base = per_cpu(tvec_bases, cpu);
1578 spin_lock_init(&base->lock);
1580 for (j = 0; j < TVN_SIZE; j++) {
1581 INIT_LIST_HEAD(base->tv5.vec + j);
1582 INIT_LIST_HEAD(base->tv4.vec + j);
1583 INIT_LIST_HEAD(base->tv3.vec + j);
1584 INIT_LIST_HEAD(base->tv2.vec + j);
1586 for (j = 0; j < TVR_SIZE; j++)
1587 INIT_LIST_HEAD(base->tv1.vec + j);
1589 base->timer_jiffies = jiffies;
1593 #ifdef CONFIG_HOTPLUG_CPU
1594 static void migrate_timer_list(struct tvec_base *new_base, struct list_head *head)
1596 struct timer_list *timer;
1598 while (!list_empty(head)) {
1599 timer = list_first_entry(head, struct timer_list, entry);
1600 detach_timer(timer, 0);
1601 timer_set_base(timer, new_base);
1602 internal_add_timer(new_base, timer);
1606 static void __cpuinit migrate_timers(int cpu)
1608 struct tvec_base *old_base;
1609 struct tvec_base *new_base;
1612 BUG_ON(cpu_online(cpu));
1613 old_base = per_cpu(tvec_bases, cpu);
1614 new_base = get_cpu_var(tvec_bases);
1616 * The caller is globally serialized and nobody else
1617 * takes two locks at once, deadlock is not possible.
1619 spin_lock_irq(&new_base->lock);
1620 spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1622 BUG_ON(old_base->running_timer);
1624 for (i = 0; i < TVR_SIZE; i++)
1625 migrate_timer_list(new_base, old_base->tv1.vec + i);
1626 for (i = 0; i < TVN_SIZE; i++) {
1627 migrate_timer_list(new_base, old_base->tv2.vec + i);
1628 migrate_timer_list(new_base, old_base->tv3.vec + i);
1629 migrate_timer_list(new_base, old_base->tv4.vec + i);
1630 migrate_timer_list(new_base, old_base->tv5.vec + i);
1633 spin_unlock(&old_base->lock);
1634 spin_unlock_irq(&new_base->lock);
1635 put_cpu_var(tvec_bases);
1637 #endif /* CONFIG_HOTPLUG_CPU */
1639 static int __cpuinit timer_cpu_notify(struct notifier_block *self,
1640 unsigned long action, void *hcpu)
1642 long cpu = (long)hcpu;
1644 case CPU_UP_PREPARE:
1645 case CPU_UP_PREPARE_FROZEN:
1646 if (init_timers_cpu(cpu) < 0)
1649 #ifdef CONFIG_HOTPLUG_CPU
1651 case CPU_DEAD_FROZEN:
1652 migrate_timers(cpu);
1661 static struct notifier_block __cpuinitdata timers_nb = {
1662 .notifier_call = timer_cpu_notify,
1666 void __init init_timers(void)
1668 int err = timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE,
1669 (void *)(long)smp_processor_id());
1673 BUG_ON(err == NOTIFY_BAD);
1674 register_cpu_notifier(&timers_nb);
1675 open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
1679 * msleep - sleep safely even with waitqueue interruptions
1680 * @msecs: Time in milliseconds to sleep for
1682 void msleep(unsigned int msecs)
1684 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1687 timeout = schedule_timeout_uninterruptible(timeout);
1690 EXPORT_SYMBOL(msleep);
1693 * msleep_interruptible - sleep waiting for signals
1694 * @msecs: Time in milliseconds to sleep for
1696 unsigned long msleep_interruptible(unsigned int msecs)
1698 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1700 while (timeout && !signal_pending(current))
1701 timeout = schedule_timeout_interruptible(timeout);
1702 return jiffies_to_msecs(timeout);
1705 EXPORT_SYMBOL(msleep_interruptible);