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 void init_timer_on_stack(struct timer_list *timer)
498 debug_object_init_on_stack(timer, &timer_debug_descr);
501 EXPORT_SYMBOL_GPL(init_timer_on_stack);
503 void destroy_timer_on_stack(struct timer_list *timer)
505 debug_object_free(timer, &timer_debug_descr);
507 EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
510 static inline void debug_timer_init(struct timer_list *timer) { }
511 static inline void debug_timer_activate(struct timer_list *timer) { }
512 static inline void debug_timer_deactivate(struct timer_list *timer) { }
515 static void __init_timer(struct timer_list *timer)
517 timer->entry.next = NULL;
518 timer->base = __raw_get_cpu_var(tvec_bases);
519 #ifdef CONFIG_TIMER_STATS
520 timer->start_site = NULL;
521 timer->start_pid = -1;
522 memset(timer->start_comm, 0, TASK_COMM_LEN);
527 * init_timer_key - initialize a timer
528 * @timer: the timer to be initialized
529 * @name: name of the timer
530 * @key: lockdep class key of the fake lock used for tracking timer
531 * sync lock dependencies
533 * init_timer_key() must be done to a timer prior calling *any* of the
534 * other timer functions.
536 void init_timer(struct timer_list *timer)
538 debug_timer_init(timer);
541 EXPORT_SYMBOL(init_timer);
543 void init_timer_deferrable(struct timer_list *timer)
546 timer_set_deferrable(timer);
548 EXPORT_SYMBOL(init_timer_deferrable);
550 static inline void detach_timer(struct timer_list *timer,
553 struct list_head *entry = &timer->entry;
555 debug_timer_deactivate(timer);
557 __list_del(entry->prev, entry->next);
560 entry->prev = LIST_POISON2;
564 * We are using hashed locking: holding per_cpu(tvec_bases).lock
565 * means that all timers which are tied to this base via timer->base are
566 * locked, and the base itself is locked too.
568 * So __run_timers/migrate_timers can safely modify all timers which could
569 * be found on ->tvX lists.
571 * When the timer's base is locked, and the timer removed from list, it is
572 * possible to set timer->base = NULL and drop the lock: the timer remains
575 static struct tvec_base *lock_timer_base(struct timer_list *timer,
576 unsigned long *flags)
577 __acquires(timer->base->lock)
579 struct tvec_base *base;
582 struct tvec_base *prelock_base = timer->base;
583 base = tbase_get_base(prelock_base);
584 if (likely(base != NULL)) {
585 spin_lock_irqsave(&base->lock, *flags);
586 if (likely(prelock_base == timer->base))
588 /* The timer has migrated to another CPU */
589 spin_unlock_irqrestore(&base->lock, *flags);
596 __mod_timer(struct timer_list *timer, unsigned long expires, bool pending_only)
598 struct tvec_base *base, *new_base;
604 timer_stats_timer_set_start_info(timer);
605 BUG_ON(!timer->function);
607 base = lock_timer_base(timer, &flags);
609 if (timer_pending(timer)) {
610 detach_timer(timer, 0);
617 debug_timer_activate(timer);
619 new_base = __get_cpu_var(tvec_bases);
621 if (base != new_base) {
623 * We are trying to schedule the timer on the local CPU.
624 * However we can't change timer's base while it is running,
625 * otherwise del_timer_sync() can't detect that the timer's
626 * handler yet has not finished. This also guarantees that
627 * the timer is serialized wrt itself.
629 if (likely(base->running_timer != timer)) {
630 /* See the comment in lock_timer_base() */
631 timer_set_base(timer, NULL);
632 spin_unlock(&base->lock);
634 spin_lock(&base->lock);
635 timer_set_base(timer, base);
639 timer->expires = expires;
640 internal_add_timer(base, timer);
643 spin_unlock_irqrestore(&base->lock, flags);
649 * mod_timer_pending - modify a pending timer's timeout
650 * @timer: the pending timer to be modified
651 * @expires: new timeout in jiffies
653 * mod_timer_pending() is the same for pending timers as mod_timer(),
654 * but will not re-activate and modify already deleted timers.
656 * It is useful for unserialized use of timers.
658 int mod_timer_pending(struct timer_list *timer, unsigned long expires)
660 return __mod_timer(timer, expires, true);
662 EXPORT_SYMBOL(mod_timer_pending);
665 * mod_timer - modify a timer's timeout
666 * @timer: the timer to be modified
667 * @expires: new timeout in jiffies
669 * mod_timer() is a more efficient way to update the expire field of an
670 * active timer (if the timer is inactive it will be activated)
672 * mod_timer(timer, expires) is equivalent to:
674 * del_timer(timer); timer->expires = expires; add_timer(timer);
676 * Note that if there are multiple unserialized concurrent users of the
677 * same timer, then mod_timer() is the only safe way to modify the timeout,
678 * since add_timer() cannot modify an already running timer.
680 * The function returns whether it has modified a pending timer or not.
681 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
682 * active timer returns 1.)
684 int mod_timer(struct timer_list *timer, unsigned long expires)
687 * This is a common optimization triggered by the
688 * networking code - if the timer is re-modified
689 * to be the same thing then just return:
691 if (timer->expires == expires && timer_pending(timer))
694 return __mod_timer(timer, expires, false);
696 EXPORT_SYMBOL(mod_timer);
699 * add_timer - start a timer
700 * @timer: the timer to be added
702 * The kernel will do a ->function(->data) callback from the
703 * timer interrupt at the ->expires point in the future. The
704 * current time is 'jiffies'.
706 * The timer's ->expires, ->function (and if the handler uses it, ->data)
707 * fields must be set prior calling this function.
709 * Timers with an ->expires field in the past will be executed in the next
712 void add_timer(struct timer_list *timer)
714 BUG_ON(timer_pending(timer));
715 mod_timer(timer, timer->expires);
717 EXPORT_SYMBOL(add_timer);
720 * add_timer_on - start a timer on a particular CPU
721 * @timer: the timer to be added
722 * @cpu: the CPU to start it on
724 * This is not very scalable on SMP. Double adds are not possible.
726 void add_timer_on(struct timer_list *timer, int cpu)
728 struct tvec_base *base = per_cpu(tvec_bases, cpu);
731 timer_stats_timer_set_start_info(timer);
732 BUG_ON(timer_pending(timer) || !timer->function);
733 spin_lock_irqsave(&base->lock, flags);
734 timer_set_base(timer, base);
735 debug_timer_activate(timer);
736 internal_add_timer(base, timer);
738 * Check whether the other CPU is idle and needs to be
739 * triggered to reevaluate the timer wheel when nohz is
740 * active. We are protected against the other CPU fiddling
741 * with the timer by holding the timer base lock. This also
742 * makes sure that a CPU on the way to idle can not evaluate
745 wake_up_idle_cpu(cpu);
746 spin_unlock_irqrestore(&base->lock, flags);
750 * del_timer - deactive a timer.
751 * @timer: the timer to be deactivated
753 * del_timer() deactivates a timer - this works on both active and inactive
756 * The function returns whether it has deactivated a pending timer or not.
757 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
758 * active timer returns 1.)
760 int del_timer(struct timer_list *timer)
762 struct tvec_base *base;
766 timer_stats_timer_clear_start_info(timer);
767 if (timer_pending(timer)) {
768 base = lock_timer_base(timer, &flags);
769 if (timer_pending(timer)) {
770 detach_timer(timer, 1);
773 spin_unlock_irqrestore(&base->lock, flags);
778 EXPORT_SYMBOL(del_timer);
782 * try_to_del_timer_sync - Try to deactivate a timer
783 * @timer: timer do del
785 * This function tries to deactivate a timer. Upon successful (ret >= 0)
786 * exit the timer is not queued and the handler is not running on any CPU.
788 * It must not be called from interrupt contexts.
790 int try_to_del_timer_sync(struct timer_list *timer)
792 struct tvec_base *base;
796 base = lock_timer_base(timer, &flags);
798 if (base->running_timer == timer)
802 if (timer_pending(timer)) {
803 detach_timer(timer, 1);
807 spin_unlock_irqrestore(&base->lock, flags);
811 EXPORT_SYMBOL(try_to_del_timer_sync);
814 * del_timer_sync - deactivate a timer and wait for the handler to finish.
815 * @timer: the timer to be deactivated
817 * This function only differs from del_timer() on SMP: besides deactivating
818 * the timer it also makes sure the handler has finished executing on other
821 * Synchronization rules: Callers must prevent restarting of the timer,
822 * otherwise this function is meaningless. It must not be called from
823 * interrupt contexts. The caller must not hold locks which would prevent
824 * completion of the timer's handler. The timer's handler must not call
825 * add_timer_on(). Upon exit the timer is not queued and the handler is
826 * not running on any CPU.
828 * The function returns whether it has deactivated a pending timer or not.
830 int del_timer_sync(struct timer_list *timer)
833 int ret = try_to_del_timer_sync(timer);
839 EXPORT_SYMBOL(del_timer_sync);
842 static int cascade(struct tvec_base *base, struct tvec *tv, int index)
844 /* cascade all the timers from tv up one level */
845 struct timer_list *timer, *tmp;
846 struct list_head tv_list;
848 list_replace_init(tv->vec + index, &tv_list);
851 * We are removing _all_ timers from the list, so we
852 * don't have to detach them individually.
854 list_for_each_entry_safe(timer, tmp, &tv_list, entry) {
855 BUG_ON(tbase_get_base(timer->base) != base);
856 internal_add_timer(base, timer);
862 #define INDEX(N) ((base->timer_jiffies >> (TVR_BITS + (N) * TVN_BITS)) & TVN_MASK)
865 * __run_timers - run all expired timers (if any) on this CPU.
866 * @base: the timer vector to be processed.
868 * This function cascades all vectors and executes all expired timer
871 static inline void __run_timers(struct tvec_base *base)
873 struct timer_list *timer;
875 spin_lock_irq(&base->lock);
876 while (time_after_eq(jiffies, base->timer_jiffies)) {
877 struct list_head work_list;
878 struct list_head *head = &work_list;
879 int index = base->timer_jiffies & TVR_MASK;
885 (!cascade(base, &base->tv2, INDEX(0))) &&
886 (!cascade(base, &base->tv3, INDEX(1))) &&
887 !cascade(base, &base->tv4, INDEX(2)))
888 cascade(base, &base->tv5, INDEX(3));
889 ++base->timer_jiffies;
890 list_replace_init(base->tv1.vec + index, &work_list);
891 while (!list_empty(head)) {
892 void (*fn)(unsigned long);
895 timer = list_first_entry(head, struct timer_list,entry);
896 fn = timer->function;
899 timer_stats_account_timer(timer);
901 set_running_timer(base, timer);
902 detach_timer(timer, 1);
903 spin_unlock_irq(&base->lock);
905 int preempt_count = preempt_count();
907 if (preempt_count != preempt_count()) {
908 printk(KERN_ERR "huh, entered %p "
909 "with preempt_count %08x, exited"
916 spin_lock_irq(&base->lock);
919 set_running_timer(base, NULL);
920 spin_unlock_irq(&base->lock);
925 * Find out when the next timer event is due to happen. This
926 * is used on S/390 to stop all activity when a cpus is idle.
927 * This functions needs to be called disabled.
929 static unsigned long __next_timer_interrupt(struct tvec_base *base)
931 unsigned long timer_jiffies = base->timer_jiffies;
932 unsigned long expires = timer_jiffies + NEXT_TIMER_MAX_DELTA;
933 int index, slot, array, found = 0;
934 struct timer_list *nte;
935 struct tvec *varray[4];
937 /* Look for timer events in tv1. */
938 index = slot = timer_jiffies & TVR_MASK;
940 list_for_each_entry(nte, base->tv1.vec + slot, entry) {
941 if (tbase_get_deferrable(nte->base))
945 expires = nte->expires;
946 /* Look at the cascade bucket(s)? */
947 if (!index || slot < index)
951 slot = (slot + 1) & TVR_MASK;
952 } while (slot != index);
955 /* Calculate the next cascade event */
957 timer_jiffies += TVR_SIZE - index;
958 timer_jiffies >>= TVR_BITS;
961 varray[0] = &base->tv2;
962 varray[1] = &base->tv3;
963 varray[2] = &base->tv4;
964 varray[3] = &base->tv5;
966 for (array = 0; array < 4; array++) {
967 struct tvec *varp = varray[array];
969 index = slot = timer_jiffies & TVN_MASK;
971 list_for_each_entry(nte, varp->vec + slot, entry) {
973 if (time_before(nte->expires, expires))
974 expires = nte->expires;
977 * Do we still search for the first timer or are
978 * we looking up the cascade buckets ?
981 /* Look at the cascade bucket(s)? */
982 if (!index || slot < index)
986 slot = (slot + 1) & TVN_MASK;
987 } while (slot != index);
990 timer_jiffies += TVN_SIZE - index;
991 timer_jiffies >>= TVN_BITS;
997 * Check, if the next hrtimer event is before the next timer wheel
1000 static unsigned long cmp_next_hrtimer_event(unsigned long now,
1001 unsigned long expires)
1003 ktime_t hr_delta = hrtimer_get_next_event();
1004 struct timespec tsdelta;
1005 unsigned long delta;
1007 if (hr_delta.tv64 == KTIME_MAX)
1011 * Expired timer available, let it expire in the next tick
1013 if (hr_delta.tv64 <= 0)
1016 tsdelta = ktime_to_timespec(hr_delta);
1017 delta = timespec_to_jiffies(&tsdelta);
1020 * Limit the delta to the max value, which is checked in
1021 * tick_nohz_stop_sched_tick():
1023 if (delta > NEXT_TIMER_MAX_DELTA)
1024 delta = NEXT_TIMER_MAX_DELTA;
1027 * Take rounding errors in to account and make sure, that it
1028 * expires in the next tick. Otherwise we go into an endless
1029 * ping pong due to tick_nohz_stop_sched_tick() retriggering
1035 if (time_before(now, expires))
1041 * get_next_timer_interrupt - return the jiffy of the next pending timer
1042 * @now: current time (in jiffies)
1044 unsigned long get_next_timer_interrupt(unsigned long now)
1046 struct tvec_base *base = __get_cpu_var(tvec_bases);
1047 unsigned long expires;
1049 spin_lock(&base->lock);
1050 expires = __next_timer_interrupt(base);
1051 spin_unlock(&base->lock);
1053 if (time_before_eq(expires, now))
1056 return cmp_next_hrtimer_event(now, expires);
1061 * Called from the timer interrupt handler to charge one tick to the current
1062 * process. user_tick is 1 if the tick is user time, 0 for system.
1064 void update_process_times(int user_tick)
1066 struct task_struct *p = current;
1067 int cpu = smp_processor_id();
1069 /* Note: this timer irq context must be accounted for as well. */
1070 account_process_tick(p, user_tick);
1072 if (rcu_pending(cpu))
1073 rcu_check_callbacks(cpu, user_tick);
1076 run_posix_cpu_timers(p);
1080 * Nr of active tasks - counted in fixed-point numbers
1082 static unsigned long count_active_tasks(void)
1084 return nr_active() * FIXED_1;
1088 * Hmm.. Changed this, as the GNU make sources (load.c) seems to
1089 * imply that avenrun[] is the standard name for this kind of thing.
1090 * Nothing else seems to be standardized: the fractional size etc
1091 * all seem to differ on different machines.
1093 * Requires xtime_lock to access.
1095 unsigned long avenrun[3];
1097 EXPORT_SYMBOL(avenrun);
1100 * calc_load - given tick count, update the avenrun load estimates.
1101 * This is called while holding a write_lock on xtime_lock.
1103 static inline void calc_load(unsigned long ticks)
1105 unsigned long active_tasks; /* fixed-point */
1106 static int count = LOAD_FREQ;
1109 if (unlikely(count < 0)) {
1110 active_tasks = count_active_tasks();
1112 CALC_LOAD(avenrun[0], EXP_1, active_tasks);
1113 CALC_LOAD(avenrun[1], EXP_5, active_tasks);
1114 CALC_LOAD(avenrun[2], EXP_15, active_tasks);
1116 } while (count < 0);
1121 * This function runs timers and the timer-tq in bottom half context.
1123 static void run_timer_softirq(struct softirq_action *h)
1125 struct tvec_base *base = __get_cpu_var(tvec_bases);
1127 hrtimer_run_pending();
1129 if (time_after_eq(jiffies, base->timer_jiffies))
1134 * Called by the local, per-CPU timer interrupt on SMP.
1136 void run_local_timers(void)
1138 hrtimer_run_queues();
1139 raise_softirq(TIMER_SOFTIRQ);
1144 * Called by the timer interrupt. xtime_lock must already be taken
1147 static inline void update_times(unsigned long ticks)
1154 * The 64-bit jiffies value is not atomic - you MUST NOT read it
1155 * without sampling the sequence number in xtime_lock.
1156 * jiffies is defined in the linker script...
1159 void do_timer(unsigned long ticks)
1161 jiffies_64 += ticks;
1162 update_times(ticks);
1165 #ifdef __ARCH_WANT_SYS_ALARM
1168 * For backwards compatibility? This can be done in libc so Alpha
1169 * and all newer ports shouldn't need it.
1171 SYSCALL_DEFINE1(alarm, unsigned int, seconds)
1173 return alarm_setitimer(seconds);
1181 * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this
1182 * should be moved into arch/i386 instead?
1186 * sys_getpid - return the thread group id of the current process
1188 * Note, despite the name, this returns the tgid not the pid. The tgid and
1189 * the pid are identical unless CLONE_THREAD was specified on clone() in
1190 * which case the tgid is the same in all threads of the same group.
1192 * This is SMP safe as current->tgid does not change.
1194 SYSCALL_DEFINE0(getpid)
1196 return task_tgid_vnr(current);
1200 * Accessing ->real_parent is not SMP-safe, it could
1201 * change from under us. However, we can use a stale
1202 * value of ->real_parent under rcu_read_lock(), see
1203 * release_task()->call_rcu(delayed_put_task_struct).
1205 SYSCALL_DEFINE0(getppid)
1210 pid = task_tgid_vnr(current->real_parent);
1216 SYSCALL_DEFINE0(getuid)
1218 /* Only we change this so SMP safe */
1219 return current_uid();
1222 SYSCALL_DEFINE0(geteuid)
1224 /* Only we change this so SMP safe */
1225 return current_euid();
1228 SYSCALL_DEFINE0(getgid)
1230 /* Only we change this so SMP safe */
1231 return current_gid();
1234 SYSCALL_DEFINE0(getegid)
1236 /* Only we change this so SMP safe */
1237 return current_egid();
1242 static void process_timeout(unsigned long __data)
1244 wake_up_process((struct task_struct *)__data);
1248 * schedule_timeout - sleep until timeout
1249 * @timeout: timeout value in jiffies
1251 * Make the current task sleep until @timeout jiffies have
1252 * elapsed. The routine will return immediately unless
1253 * the current task state has been set (see set_current_state()).
1255 * You can set the task state as follows -
1257 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1258 * pass before the routine returns. The routine will return 0
1260 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1261 * delivered to the current task. In this case the remaining time
1262 * in jiffies will be returned, or 0 if the timer expired in time
1264 * The current task state is guaranteed to be TASK_RUNNING when this
1267 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1268 * the CPU away without a bound on the timeout. In this case the return
1269 * value will be %MAX_SCHEDULE_TIMEOUT.
1271 * In all cases the return value is guaranteed to be non-negative.
1273 signed long __sched schedule_timeout(signed long timeout)
1275 struct timer_list timer;
1276 unsigned long expire;
1280 case MAX_SCHEDULE_TIMEOUT:
1282 * These two special cases are useful to be comfortable
1283 * in the caller. Nothing more. We could take
1284 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1285 * but I' d like to return a valid offset (>=0) to allow
1286 * the caller to do everything it want with the retval.
1292 * Another bit of PARANOID. Note that the retval will be
1293 * 0 since no piece of kernel is supposed to do a check
1294 * for a negative retval of schedule_timeout() (since it
1295 * should never happens anyway). You just have the printk()
1296 * that will tell you if something is gone wrong and where.
1299 printk(KERN_ERR "schedule_timeout: wrong timeout "
1300 "value %lx\n", timeout);
1302 current->state = TASK_RUNNING;
1307 expire = timeout + jiffies;
1309 setup_timer_on_stack(&timer, process_timeout, (unsigned long)current);
1310 __mod_timer(&timer, expire, false);
1312 del_singleshot_timer_sync(&timer);
1314 /* Remove the timer from the object tracker */
1315 destroy_timer_on_stack(&timer);
1317 timeout = expire - jiffies;
1320 return timeout < 0 ? 0 : timeout;
1322 EXPORT_SYMBOL(schedule_timeout);
1325 * We can use __set_current_state() here because schedule_timeout() calls
1326 * schedule() unconditionally.
1328 signed long __sched schedule_timeout_interruptible(signed long timeout)
1330 __set_current_state(TASK_INTERRUPTIBLE);
1331 return schedule_timeout(timeout);
1333 EXPORT_SYMBOL(schedule_timeout_interruptible);
1335 signed long __sched schedule_timeout_killable(signed long timeout)
1337 __set_current_state(TASK_KILLABLE);
1338 return schedule_timeout(timeout);
1340 EXPORT_SYMBOL(schedule_timeout_killable);
1342 signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1344 __set_current_state(TASK_UNINTERRUPTIBLE);
1345 return schedule_timeout(timeout);
1347 EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1349 /* Thread ID - the internal kernel "pid" */
1350 SYSCALL_DEFINE0(gettid)
1352 return task_pid_vnr(current);
1356 * do_sysinfo - fill in sysinfo struct
1357 * @info: pointer to buffer to fill
1359 int do_sysinfo(struct sysinfo *info)
1361 unsigned long mem_total, sav_total;
1362 unsigned int mem_unit, bitcount;
1365 memset(info, 0, sizeof(struct sysinfo));
1369 seq = read_seqbegin(&xtime_lock);
1372 * This is annoying. The below is the same thing
1373 * posix_get_clock_monotonic() does, but it wants to
1374 * take the lock which we want to cover the loads stuff
1378 getnstimeofday(&tp);
1379 tp.tv_sec += wall_to_monotonic.tv_sec;
1380 tp.tv_nsec += wall_to_monotonic.tv_nsec;
1381 monotonic_to_bootbased(&tp);
1382 if (tp.tv_nsec - NSEC_PER_SEC >= 0) {
1383 tp.tv_nsec = tp.tv_nsec - NSEC_PER_SEC;
1386 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
1388 info->loads[0] = avenrun[0] << (SI_LOAD_SHIFT - FSHIFT);
1389 info->loads[1] = avenrun[1] << (SI_LOAD_SHIFT - FSHIFT);
1390 info->loads[2] = avenrun[2] << (SI_LOAD_SHIFT - FSHIFT);
1392 info->procs = nr_threads;
1393 } while (read_seqretry(&xtime_lock, seq));
1399 * If the sum of all the available memory (i.e. ram + swap)
1400 * is less than can be stored in a 32 bit unsigned long then
1401 * we can be binary compatible with 2.2.x kernels. If not,
1402 * well, in that case 2.2.x was broken anyways...
1404 * -Erik Andersen <andersee@debian.org>
1407 mem_total = info->totalram + info->totalswap;
1408 if (mem_total < info->totalram || mem_total < info->totalswap)
1411 mem_unit = info->mem_unit;
1412 while (mem_unit > 1) {
1415 sav_total = mem_total;
1417 if (mem_total < sav_total)
1422 * If mem_total did not overflow, multiply all memory values by
1423 * info->mem_unit and set it to 1. This leaves things compatible
1424 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1429 info->totalram <<= bitcount;
1430 info->freeram <<= bitcount;
1431 info->sharedram <<= bitcount;
1432 info->bufferram <<= bitcount;
1433 info->totalswap <<= bitcount;
1434 info->freeswap <<= bitcount;
1435 info->totalhigh <<= bitcount;
1436 info->freehigh <<= bitcount;
1442 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
1448 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
1454 static int __cpuinit init_timers_cpu(int cpu)
1457 struct tvec_base *base;
1458 static char __cpuinitdata tvec_base_done[NR_CPUS];
1460 if (!tvec_base_done[cpu]) {
1461 static char boot_done;
1465 * The APs use this path later in boot
1467 base = kmalloc_node(sizeof(*base),
1468 GFP_KERNEL | __GFP_ZERO,
1473 /* Make sure that tvec_base is 2 byte aligned */
1474 if (tbase_get_deferrable(base)) {
1479 per_cpu(tvec_bases, cpu) = base;
1482 * This is for the boot CPU - we use compile-time
1483 * static initialisation because per-cpu memory isn't
1484 * ready yet and because the memory allocators are not
1485 * initialised either.
1488 base = &boot_tvec_bases;
1490 tvec_base_done[cpu] = 1;
1492 base = per_cpu(tvec_bases, cpu);
1495 spin_lock_init(&base->lock);
1497 for (j = 0; j < TVN_SIZE; j++) {
1498 INIT_LIST_HEAD(base->tv5.vec + j);
1499 INIT_LIST_HEAD(base->tv4.vec + j);
1500 INIT_LIST_HEAD(base->tv3.vec + j);
1501 INIT_LIST_HEAD(base->tv2.vec + j);
1503 for (j = 0; j < TVR_SIZE; j++)
1504 INIT_LIST_HEAD(base->tv1.vec + j);
1506 base->timer_jiffies = jiffies;
1510 #ifdef CONFIG_HOTPLUG_CPU
1511 static void migrate_timer_list(struct tvec_base *new_base, struct list_head *head)
1513 struct timer_list *timer;
1515 while (!list_empty(head)) {
1516 timer = list_first_entry(head, struct timer_list, entry);
1517 detach_timer(timer, 0);
1518 timer_set_base(timer, new_base);
1519 internal_add_timer(new_base, timer);
1523 static void __cpuinit migrate_timers(int cpu)
1525 struct tvec_base *old_base;
1526 struct tvec_base *new_base;
1529 BUG_ON(cpu_online(cpu));
1530 old_base = per_cpu(tvec_bases, cpu);
1531 new_base = get_cpu_var(tvec_bases);
1533 * The caller is globally serialized and nobody else
1534 * takes two locks at once, deadlock is not possible.
1536 spin_lock_irq(&new_base->lock);
1537 spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1539 BUG_ON(old_base->running_timer);
1541 for (i = 0; i < TVR_SIZE; i++)
1542 migrate_timer_list(new_base, old_base->tv1.vec + i);
1543 for (i = 0; i < TVN_SIZE; i++) {
1544 migrate_timer_list(new_base, old_base->tv2.vec + i);
1545 migrate_timer_list(new_base, old_base->tv3.vec + i);
1546 migrate_timer_list(new_base, old_base->tv4.vec + i);
1547 migrate_timer_list(new_base, old_base->tv5.vec + i);
1550 spin_unlock(&old_base->lock);
1551 spin_unlock_irq(&new_base->lock);
1552 put_cpu_var(tvec_bases);
1554 #endif /* CONFIG_HOTPLUG_CPU */
1556 static int __cpuinit timer_cpu_notify(struct notifier_block *self,
1557 unsigned long action, void *hcpu)
1559 long cpu = (long)hcpu;
1561 case CPU_UP_PREPARE:
1562 case CPU_UP_PREPARE_FROZEN:
1563 if (init_timers_cpu(cpu) < 0)
1566 #ifdef CONFIG_HOTPLUG_CPU
1568 case CPU_DEAD_FROZEN:
1569 migrate_timers(cpu);
1578 static struct notifier_block __cpuinitdata timers_nb = {
1579 .notifier_call = timer_cpu_notify,
1583 void __init init_timers(void)
1585 int err = timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE,
1586 (void *)(long)smp_processor_id());
1590 BUG_ON(err == NOTIFY_BAD);
1591 register_cpu_notifier(&timers_nb);
1592 open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
1596 * msleep - sleep safely even with waitqueue interruptions
1597 * @msecs: Time in milliseconds to sleep for
1599 void msleep(unsigned int msecs)
1601 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1604 timeout = schedule_timeout_uninterruptible(timeout);
1607 EXPORT_SYMBOL(msleep);
1610 * msleep_interruptible - sleep waiting for signals
1611 * @msecs: Time in milliseconds to sleep for
1613 unsigned long msleep_interruptible(unsigned int msecs)
1615 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1617 while (timeout && !signal_pending(current))
1618 timeout = schedule_timeout_interruptible(timeout);
1619 return jiffies_to_msecs(timeout);
1622 EXPORT_SYMBOL(msleep_interruptible);