Merge branch 'merge' of git://git.kernel.org/pub/scm/linux/kernel/git/paulus/powerpc
[linux-2.6/lfs.git] / kernel / timer.c
blobceacc6626572a7c2e2c345ba136bd456a267ba64
1 /*
2 * linux/kernel/timer.c
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>
27 #include <linux/mm.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>
45 #include <asm/io.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)
61 struct tvec {
62 struct list_head vec[TVN_SIZE];
65 struct tvec_root {
66 struct list_head vec[TVR_SIZE];
69 struct tvec_base {
70 spinlock_t lock;
71 struct timer_list *running_timer;
72 unsigned long timer_jiffies;
73 struct tvec_root tv1;
74 struct tvec tv2;
75 struct tvec tv3;
76 struct tvec tv4;
77 struct tvec tv5;
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));
108 static inline void
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));
116 * __round_jiffies - function to round jiffies to a full second
117 * @j: the time in (absolute) jiffies that should be rounded
118 * @cpu: the processor number on which the timeout will happen
120 * __round_jiffies() rounds an absolute time in the future (in jiffies)
121 * up or down to (approximately) full seconds. This is useful for timers
122 * for which the exact time they fire does not matter too much, as long as
123 * they fire approximately every X seconds.
125 * By rounding these timers to whole seconds, all such timers will fire
126 * at the same time, rather than at various times spread out. The goal
127 * of this is to have the CPU wake up less, which saves power.
129 * The exact rounding is skewed for each processor to avoid all
130 * processors firing at the exact same time, which could lead
131 * to lock contention or spurious cache line bouncing.
133 * The return value is the rounded version of the @j parameter.
135 unsigned long __round_jiffies(unsigned long j, int cpu)
137 int rem;
138 unsigned long original = j;
141 * We don't want all cpus firing their timers at once hitting the
142 * same lock or cachelines, so we skew each extra cpu with an extra
143 * 3 jiffies. This 3 jiffies came originally from the mm/ code which
144 * already did this.
145 * The skew is done by adding 3*cpunr, then round, then subtract this
146 * extra offset again.
148 j += cpu * 3;
150 rem = j % HZ;
153 * If the target jiffie is just after a whole second (which can happen
154 * due to delays of the timer irq, long irq off times etc etc) then
155 * we should round down to the whole second, not up. Use 1/4th second
156 * as cutoff for this rounding as an extreme upper bound for this.
158 if (rem < HZ/4) /* round down */
159 j = j - rem;
160 else /* round up */
161 j = j - rem + HZ;
163 /* now that we have rounded, subtract the extra skew again */
164 j -= cpu * 3;
166 if (j <= jiffies) /* rounding ate our timeout entirely; */
167 return original;
168 return j;
170 EXPORT_SYMBOL_GPL(__round_jiffies);
173 * __round_jiffies_relative - function to round jiffies to a full second
174 * @j: the time in (relative) jiffies that should be rounded
175 * @cpu: the processor number on which the timeout will happen
177 * __round_jiffies_relative() rounds a time delta in the future (in jiffies)
178 * up or down to (approximately) full seconds. This is useful for timers
179 * for which the exact time they fire does not matter too much, as long as
180 * they fire approximately every X seconds.
182 * By rounding these timers to whole seconds, all such timers will fire
183 * at the same time, rather than at various times spread out. The goal
184 * of this is to have the CPU wake up less, which saves power.
186 * The exact rounding is skewed for each processor to avoid all
187 * processors firing at the exact same time, which could lead
188 * to lock contention or spurious cache line bouncing.
190 * The return value is the rounded version of the @j parameter.
192 unsigned long __round_jiffies_relative(unsigned long j, int cpu)
195 * In theory the following code can skip a jiffy in case jiffies
196 * increments right between the addition and the later subtraction.
197 * However since the entire point of this function is to use approximate
198 * timeouts, it's entirely ok to not handle that.
200 return __round_jiffies(j + jiffies, cpu) - jiffies;
202 EXPORT_SYMBOL_GPL(__round_jiffies_relative);
205 * round_jiffies - function to round jiffies to a full second
206 * @j: the time in (absolute) jiffies that should be rounded
208 * round_jiffies() rounds an absolute time in the future (in jiffies)
209 * up or down to (approximately) full seconds. This is useful for timers
210 * for which the exact time they fire does not matter too much, as long as
211 * they fire approximately every X seconds.
213 * By rounding these timers to whole seconds, all such timers will fire
214 * at the same time, rather than at various times spread out. The goal
215 * of this is to have the CPU wake up less, which saves power.
217 * The return value is the rounded version of the @j parameter.
219 unsigned long round_jiffies(unsigned long j)
221 return __round_jiffies(j, raw_smp_processor_id());
223 EXPORT_SYMBOL_GPL(round_jiffies);
226 * round_jiffies_relative - function to round jiffies to a full second
227 * @j: the time in (relative) jiffies that should be rounded
229 * round_jiffies_relative() rounds a time delta in the future (in jiffies)
230 * up or down to (approximately) full seconds. This is useful for timers
231 * for which the exact time they fire does not matter too much, as long as
232 * they fire approximately every X seconds.
234 * By rounding these timers to whole seconds, all such timers will fire
235 * at the same time, rather than at various times spread out. The goal
236 * of this is to have the CPU wake up less, which saves power.
238 * The return value is the rounded version of the @j parameter.
240 unsigned long round_jiffies_relative(unsigned long j)
242 return __round_jiffies_relative(j, raw_smp_processor_id());
244 EXPORT_SYMBOL_GPL(round_jiffies_relative);
247 static inline void set_running_timer(struct tvec_base *base,
248 struct timer_list *timer)
250 #ifdef CONFIG_SMP
251 base->running_timer = timer;
252 #endif
255 static void internal_add_timer(struct tvec_base *base, struct timer_list *timer)
257 unsigned long expires = timer->expires;
258 unsigned long idx = expires - base->timer_jiffies;
259 struct list_head *vec;
261 if (idx < TVR_SIZE) {
262 int i = expires & TVR_MASK;
263 vec = base->tv1.vec + i;
264 } else if (idx < 1 << (TVR_BITS + TVN_BITS)) {
265 int i = (expires >> TVR_BITS) & TVN_MASK;
266 vec = base->tv2.vec + i;
267 } else if (idx < 1 << (TVR_BITS + 2 * TVN_BITS)) {
268 int i = (expires >> (TVR_BITS + TVN_BITS)) & TVN_MASK;
269 vec = base->tv3.vec + i;
270 } else if (idx < 1 << (TVR_BITS + 3 * TVN_BITS)) {
271 int i = (expires >> (TVR_BITS + 2 * TVN_BITS)) & TVN_MASK;
272 vec = base->tv4.vec + i;
273 } else if ((signed long) idx < 0) {
275 * Can happen if you add a timer with expires == jiffies,
276 * or you set a timer to go off in the past
278 vec = base->tv1.vec + (base->timer_jiffies & TVR_MASK);
279 } else {
280 int i;
281 /* If the timeout is larger than 0xffffffff on 64-bit
282 * architectures then we use the maximum timeout:
284 if (idx > 0xffffffffUL) {
285 idx = 0xffffffffUL;
286 expires = idx + base->timer_jiffies;
288 i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK;
289 vec = base->tv5.vec + i;
292 * Timers are FIFO:
294 list_add_tail(&timer->entry, vec);
297 #ifdef CONFIG_TIMER_STATS
298 void __timer_stats_timer_set_start_info(struct timer_list *timer, void *addr)
300 if (timer->start_site)
301 return;
303 timer->start_site = addr;
304 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
305 timer->start_pid = current->pid;
308 static void timer_stats_account_timer(struct timer_list *timer)
310 unsigned int flag = 0;
312 if (unlikely(tbase_get_deferrable(timer->base)))
313 flag |= TIMER_STATS_FLAG_DEFERRABLE;
315 timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
316 timer->function, timer->start_comm, flag);
319 #else
320 static void timer_stats_account_timer(struct timer_list *timer) {}
321 #endif
323 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
325 static struct debug_obj_descr timer_debug_descr;
328 * fixup_init is called when:
329 * - an active object is initialized
331 static int timer_fixup_init(void *addr, enum debug_obj_state state)
333 struct timer_list *timer = addr;
335 switch (state) {
336 case ODEBUG_STATE_ACTIVE:
337 del_timer_sync(timer);
338 debug_object_init(timer, &timer_debug_descr);
339 return 1;
340 default:
341 return 0;
346 * fixup_activate is called when:
347 * - an active object is activated
348 * - an unknown object is activated (might be a statically initialized object)
350 static int timer_fixup_activate(void *addr, enum debug_obj_state state)
352 struct timer_list *timer = addr;
354 switch (state) {
356 case ODEBUG_STATE_NOTAVAILABLE:
358 * This is not really a fixup. The timer was
359 * statically initialized. We just make sure that it
360 * is tracked in the object tracker.
362 if (timer->entry.next == NULL &&
363 timer->entry.prev == TIMER_ENTRY_STATIC) {
364 debug_object_init(timer, &timer_debug_descr);
365 debug_object_activate(timer, &timer_debug_descr);
366 return 0;
367 } else {
368 WARN_ON_ONCE(1);
370 return 0;
372 case ODEBUG_STATE_ACTIVE:
373 WARN_ON(1);
375 default:
376 return 0;
381 * fixup_free is called when:
382 * - an active object is freed
384 static int timer_fixup_free(void *addr, enum debug_obj_state state)
386 struct timer_list *timer = addr;
388 switch (state) {
389 case ODEBUG_STATE_ACTIVE:
390 del_timer_sync(timer);
391 debug_object_free(timer, &timer_debug_descr);
392 return 1;
393 default:
394 return 0;
398 static struct debug_obj_descr timer_debug_descr = {
399 .name = "timer_list",
400 .fixup_init = timer_fixup_init,
401 .fixup_activate = timer_fixup_activate,
402 .fixup_free = timer_fixup_free,
405 static inline void debug_timer_init(struct timer_list *timer)
407 debug_object_init(timer, &timer_debug_descr);
410 static inline void debug_timer_activate(struct timer_list *timer)
412 debug_object_activate(timer, &timer_debug_descr);
415 static inline void debug_timer_deactivate(struct timer_list *timer)
417 debug_object_deactivate(timer, &timer_debug_descr);
420 static inline void debug_timer_free(struct timer_list *timer)
422 debug_object_free(timer, &timer_debug_descr);
425 static void __init_timer(struct timer_list *timer);
427 void init_timer_on_stack(struct timer_list *timer)
429 debug_object_init_on_stack(timer, &timer_debug_descr);
430 __init_timer(timer);
432 EXPORT_SYMBOL_GPL(init_timer_on_stack);
434 void destroy_timer_on_stack(struct timer_list *timer)
436 debug_object_free(timer, &timer_debug_descr);
438 EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
440 #else
441 static inline void debug_timer_init(struct timer_list *timer) { }
442 static inline void debug_timer_activate(struct timer_list *timer) { }
443 static inline void debug_timer_deactivate(struct timer_list *timer) { }
444 #endif
446 static void __init_timer(struct timer_list *timer)
448 timer->entry.next = NULL;
449 timer->base = __raw_get_cpu_var(tvec_bases);
450 #ifdef CONFIG_TIMER_STATS
451 timer->start_site = NULL;
452 timer->start_pid = -1;
453 memset(timer->start_comm, 0, TASK_COMM_LEN);
454 #endif
458 * init_timer - initialize a timer.
459 * @timer: the timer to be initialized
461 * init_timer() must be done to a timer prior calling *any* of the
462 * other timer functions.
464 void init_timer(struct timer_list *timer)
466 debug_timer_init(timer);
467 __init_timer(timer);
469 EXPORT_SYMBOL(init_timer);
471 void init_timer_deferrable(struct timer_list *timer)
473 init_timer(timer);
474 timer_set_deferrable(timer);
476 EXPORT_SYMBOL(init_timer_deferrable);
478 static inline void detach_timer(struct timer_list *timer,
479 int clear_pending)
481 struct list_head *entry = &timer->entry;
483 debug_timer_deactivate(timer);
485 __list_del(entry->prev, entry->next);
486 if (clear_pending)
487 entry->next = NULL;
488 entry->prev = LIST_POISON2;
492 * We are using hashed locking: holding per_cpu(tvec_bases).lock
493 * means that all timers which are tied to this base via timer->base are
494 * locked, and the base itself is locked too.
496 * So __run_timers/migrate_timers can safely modify all timers which could
497 * be found on ->tvX lists.
499 * When the timer's base is locked, and the timer removed from list, it is
500 * possible to set timer->base = NULL and drop the lock: the timer remains
501 * locked.
503 static struct tvec_base *lock_timer_base(struct timer_list *timer,
504 unsigned long *flags)
505 __acquires(timer->base->lock)
507 struct tvec_base *base;
509 for (;;) {
510 struct tvec_base *prelock_base = timer->base;
511 base = tbase_get_base(prelock_base);
512 if (likely(base != NULL)) {
513 spin_lock_irqsave(&base->lock, *flags);
514 if (likely(prelock_base == timer->base))
515 return base;
516 /* The timer has migrated to another CPU */
517 spin_unlock_irqrestore(&base->lock, *flags);
519 cpu_relax();
523 int __mod_timer(struct timer_list *timer, unsigned long expires)
525 struct tvec_base *base, *new_base;
526 unsigned long flags;
527 int ret = 0;
529 timer_stats_timer_set_start_info(timer);
530 BUG_ON(!timer->function);
532 base = lock_timer_base(timer, &flags);
534 if (timer_pending(timer)) {
535 detach_timer(timer, 0);
536 ret = 1;
539 debug_timer_activate(timer);
541 new_base = __get_cpu_var(tvec_bases);
543 if (base != new_base) {
545 * We are trying to schedule the timer on the local CPU.
546 * However we can't change timer's base while it is running,
547 * otherwise del_timer_sync() can't detect that the timer's
548 * handler yet has not finished. This also guarantees that
549 * the timer is serialized wrt itself.
551 if (likely(base->running_timer != timer)) {
552 /* See the comment in lock_timer_base() */
553 timer_set_base(timer, NULL);
554 spin_unlock(&base->lock);
555 base = new_base;
556 spin_lock(&base->lock);
557 timer_set_base(timer, base);
561 timer->expires = expires;
562 internal_add_timer(base, timer);
563 spin_unlock_irqrestore(&base->lock, flags);
565 return ret;
568 EXPORT_SYMBOL(__mod_timer);
571 * add_timer_on - start a timer on a particular CPU
572 * @timer: the timer to be added
573 * @cpu: the CPU to start it on
575 * This is not very scalable on SMP. Double adds are not possible.
577 void add_timer_on(struct timer_list *timer, int cpu)
579 struct tvec_base *base = per_cpu(tvec_bases, cpu);
580 unsigned long flags;
582 timer_stats_timer_set_start_info(timer);
583 BUG_ON(timer_pending(timer) || !timer->function);
584 spin_lock_irqsave(&base->lock, flags);
585 timer_set_base(timer, base);
586 debug_timer_activate(timer);
587 internal_add_timer(base, timer);
589 * Check whether the other CPU is idle and needs to be
590 * triggered to reevaluate the timer wheel when nohz is
591 * active. We are protected against the other CPU fiddling
592 * with the timer by holding the timer base lock. This also
593 * makes sure that a CPU on the way to idle can not evaluate
594 * the timer wheel.
596 wake_up_idle_cpu(cpu);
597 spin_unlock_irqrestore(&base->lock, flags);
601 * mod_timer - modify a timer's timeout
602 * @timer: the timer to be modified
603 * @expires: new timeout in jiffies
605 * mod_timer() is a more efficient way to update the expire field of an
606 * active timer (if the timer is inactive it will be activated)
608 * mod_timer(timer, expires) is equivalent to:
610 * del_timer(timer); timer->expires = expires; add_timer(timer);
612 * Note that if there are multiple unserialized concurrent users of the
613 * same timer, then mod_timer() is the only safe way to modify the timeout,
614 * since add_timer() cannot modify an already running timer.
616 * The function returns whether it has modified a pending timer or not.
617 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
618 * active timer returns 1.)
620 int mod_timer(struct timer_list *timer, unsigned long expires)
622 BUG_ON(!timer->function);
624 timer_stats_timer_set_start_info(timer);
626 * This is a common optimization triggered by the
627 * networking code - if the timer is re-modified
628 * to be the same thing then just return:
630 if (timer->expires == expires && timer_pending(timer))
631 return 1;
633 return __mod_timer(timer, expires);
636 EXPORT_SYMBOL(mod_timer);
639 * del_timer - deactive a timer.
640 * @timer: the timer to be deactivated
642 * del_timer() deactivates a timer - this works on both active and inactive
643 * timers.
645 * The function returns whether it has deactivated a pending timer or not.
646 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
647 * active timer returns 1.)
649 int del_timer(struct timer_list *timer)
651 struct tvec_base *base;
652 unsigned long flags;
653 int ret = 0;
655 timer_stats_timer_clear_start_info(timer);
656 if (timer_pending(timer)) {
657 base = lock_timer_base(timer, &flags);
658 if (timer_pending(timer)) {
659 detach_timer(timer, 1);
660 ret = 1;
662 spin_unlock_irqrestore(&base->lock, flags);
665 return ret;
668 EXPORT_SYMBOL(del_timer);
670 #ifdef CONFIG_SMP
672 * try_to_del_timer_sync - Try to deactivate a timer
673 * @timer: timer do del
675 * This function tries to deactivate a timer. Upon successful (ret >= 0)
676 * exit the timer is not queued and the handler is not running on any CPU.
678 * It must not be called from interrupt contexts.
680 int try_to_del_timer_sync(struct timer_list *timer)
682 struct tvec_base *base;
683 unsigned long flags;
684 int ret = -1;
686 base = lock_timer_base(timer, &flags);
688 if (base->running_timer == timer)
689 goto out;
691 ret = 0;
692 if (timer_pending(timer)) {
693 detach_timer(timer, 1);
694 ret = 1;
696 out:
697 spin_unlock_irqrestore(&base->lock, flags);
699 return ret;
702 EXPORT_SYMBOL(try_to_del_timer_sync);
705 * del_timer_sync - deactivate a timer and wait for the handler to finish.
706 * @timer: the timer to be deactivated
708 * This function only differs from del_timer() on SMP: besides deactivating
709 * the timer it also makes sure the handler has finished executing on other
710 * CPUs.
712 * Synchronization rules: Callers must prevent restarting of the timer,
713 * otherwise this function is meaningless. It must not be called from
714 * interrupt contexts. The caller must not hold locks which would prevent
715 * completion of the timer's handler. The timer's handler must not call
716 * add_timer_on(). Upon exit the timer is not queued and the handler is
717 * not running on any CPU.
719 * The function returns whether it has deactivated a pending timer or not.
721 int del_timer_sync(struct timer_list *timer)
723 for (;;) {
724 int ret = try_to_del_timer_sync(timer);
725 if (ret >= 0)
726 return ret;
727 cpu_relax();
731 EXPORT_SYMBOL(del_timer_sync);
732 #endif
734 static int cascade(struct tvec_base *base, struct tvec *tv, int index)
736 /* cascade all the timers from tv up one level */
737 struct timer_list *timer, *tmp;
738 struct list_head tv_list;
740 list_replace_init(tv->vec + index, &tv_list);
743 * We are removing _all_ timers from the list, so we
744 * don't have to detach them individually.
746 list_for_each_entry_safe(timer, tmp, &tv_list, entry) {
747 BUG_ON(tbase_get_base(timer->base) != base);
748 internal_add_timer(base, timer);
751 return index;
754 #define INDEX(N) ((base->timer_jiffies >> (TVR_BITS + (N) * TVN_BITS)) & TVN_MASK)
757 * __run_timers - run all expired timers (if any) on this CPU.
758 * @base: the timer vector to be processed.
760 * This function cascades all vectors and executes all expired timer
761 * vectors.
763 static inline void __run_timers(struct tvec_base *base)
765 struct timer_list *timer;
767 spin_lock_irq(&base->lock);
768 while (time_after_eq(jiffies, base->timer_jiffies)) {
769 struct list_head work_list;
770 struct list_head *head = &work_list;
771 int index = base->timer_jiffies & TVR_MASK;
774 * Cascade timers:
776 if (!index &&
777 (!cascade(base, &base->tv2, INDEX(0))) &&
778 (!cascade(base, &base->tv3, INDEX(1))) &&
779 !cascade(base, &base->tv4, INDEX(2)))
780 cascade(base, &base->tv5, INDEX(3));
781 ++base->timer_jiffies;
782 list_replace_init(base->tv1.vec + index, &work_list);
783 while (!list_empty(head)) {
784 void (*fn)(unsigned long);
785 unsigned long data;
787 timer = list_first_entry(head, struct timer_list,entry);
788 fn = timer->function;
789 data = timer->data;
791 timer_stats_account_timer(timer);
793 set_running_timer(base, timer);
794 detach_timer(timer, 1);
795 spin_unlock_irq(&base->lock);
797 int preempt_count = preempt_count();
798 fn(data);
799 if (preempt_count != preempt_count()) {
800 printk(KERN_ERR "huh, entered %p "
801 "with preempt_count %08x, exited"
802 " with %08x?\n",
803 fn, preempt_count,
804 preempt_count());
805 BUG();
808 spin_lock_irq(&base->lock);
811 set_running_timer(base, NULL);
812 spin_unlock_irq(&base->lock);
815 #if defined(CONFIG_NO_IDLE_HZ) || defined(CONFIG_NO_HZ)
817 * Find out when the next timer event is due to happen. This
818 * is used on S/390 to stop all activity when a cpus is idle.
819 * This functions needs to be called disabled.
821 static unsigned long __next_timer_interrupt(struct tvec_base *base)
823 unsigned long timer_jiffies = base->timer_jiffies;
824 unsigned long expires = timer_jiffies + NEXT_TIMER_MAX_DELTA;
825 int index, slot, array, found = 0;
826 struct timer_list *nte;
827 struct tvec *varray[4];
829 /* Look for timer events in tv1. */
830 index = slot = timer_jiffies & TVR_MASK;
831 do {
832 list_for_each_entry(nte, base->tv1.vec + slot, entry) {
833 if (tbase_get_deferrable(nte->base))
834 continue;
836 found = 1;
837 expires = nte->expires;
838 /* Look at the cascade bucket(s)? */
839 if (!index || slot < index)
840 goto cascade;
841 return expires;
843 slot = (slot + 1) & TVR_MASK;
844 } while (slot != index);
846 cascade:
847 /* Calculate the next cascade event */
848 if (index)
849 timer_jiffies += TVR_SIZE - index;
850 timer_jiffies >>= TVR_BITS;
852 /* Check tv2-tv5. */
853 varray[0] = &base->tv2;
854 varray[1] = &base->tv3;
855 varray[2] = &base->tv4;
856 varray[3] = &base->tv5;
858 for (array = 0; array < 4; array++) {
859 struct tvec *varp = varray[array];
861 index = slot = timer_jiffies & TVN_MASK;
862 do {
863 list_for_each_entry(nte, varp->vec + slot, entry) {
864 found = 1;
865 if (time_before(nte->expires, expires))
866 expires = nte->expires;
869 * Do we still search for the first timer or are
870 * we looking up the cascade buckets ?
872 if (found) {
873 /* Look at the cascade bucket(s)? */
874 if (!index || slot < index)
875 break;
876 return expires;
878 slot = (slot + 1) & TVN_MASK;
879 } while (slot != index);
881 if (index)
882 timer_jiffies += TVN_SIZE - index;
883 timer_jiffies >>= TVN_BITS;
885 return expires;
889 * Check, if the next hrtimer event is before the next timer wheel
890 * event:
892 static unsigned long cmp_next_hrtimer_event(unsigned long now,
893 unsigned long expires)
895 ktime_t hr_delta = hrtimer_get_next_event();
896 struct timespec tsdelta;
897 unsigned long delta;
899 if (hr_delta.tv64 == KTIME_MAX)
900 return expires;
903 * Expired timer available, let it expire in the next tick
905 if (hr_delta.tv64 <= 0)
906 return now + 1;
908 tsdelta = ktime_to_timespec(hr_delta);
909 delta = timespec_to_jiffies(&tsdelta);
912 * Limit the delta to the max value, which is checked in
913 * tick_nohz_stop_sched_tick():
915 if (delta > NEXT_TIMER_MAX_DELTA)
916 delta = NEXT_TIMER_MAX_DELTA;
919 * Take rounding errors in to account and make sure, that it
920 * expires in the next tick. Otherwise we go into an endless
921 * ping pong due to tick_nohz_stop_sched_tick() retriggering
922 * the timer softirq
924 if (delta < 1)
925 delta = 1;
926 now += delta;
927 if (time_before(now, expires))
928 return now;
929 return expires;
933 * get_next_timer_interrupt - return the jiffy of the next pending timer
934 * @now: current time (in jiffies)
936 unsigned long get_next_timer_interrupt(unsigned long now)
938 struct tvec_base *base = __get_cpu_var(tvec_bases);
939 unsigned long expires;
941 spin_lock(&base->lock);
942 expires = __next_timer_interrupt(base);
943 spin_unlock(&base->lock);
945 if (time_before_eq(expires, now))
946 return now;
948 return cmp_next_hrtimer_event(now, expires);
951 #ifdef CONFIG_NO_IDLE_HZ
952 unsigned long next_timer_interrupt(void)
954 return get_next_timer_interrupt(jiffies);
956 #endif
958 #endif
960 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
961 void account_process_tick(struct task_struct *p, int user_tick)
963 cputime_t one_jiffy = jiffies_to_cputime(1);
965 if (user_tick) {
966 account_user_time(p, one_jiffy);
967 account_user_time_scaled(p, cputime_to_scaled(one_jiffy));
968 } else {
969 account_system_time(p, HARDIRQ_OFFSET, one_jiffy);
970 account_system_time_scaled(p, cputime_to_scaled(one_jiffy));
973 #endif
976 * Called from the timer interrupt handler to charge one tick to the current
977 * process. user_tick is 1 if the tick is user time, 0 for system.
979 void update_process_times(int user_tick)
981 struct task_struct *p = current;
982 int cpu = smp_processor_id();
984 /* Note: this timer irq context must be accounted for as well. */
985 account_process_tick(p, user_tick);
986 run_local_timers();
987 if (rcu_pending(cpu))
988 rcu_check_callbacks(cpu, user_tick);
989 scheduler_tick();
990 run_posix_cpu_timers(p);
994 * Nr of active tasks - counted in fixed-point numbers
996 static unsigned long count_active_tasks(void)
998 return nr_active() * FIXED_1;
1002 * Hmm.. Changed this, as the GNU make sources (load.c) seems to
1003 * imply that avenrun[] is the standard name for this kind of thing.
1004 * Nothing else seems to be standardized: the fractional size etc
1005 * all seem to differ on different machines.
1007 * Requires xtime_lock to access.
1009 unsigned long avenrun[3];
1011 EXPORT_SYMBOL(avenrun);
1014 * calc_load - given tick count, update the avenrun load estimates.
1015 * This is called while holding a write_lock on xtime_lock.
1017 static inline void calc_load(unsigned long ticks)
1019 unsigned long active_tasks; /* fixed-point */
1020 static int count = LOAD_FREQ;
1022 count -= ticks;
1023 if (unlikely(count < 0)) {
1024 active_tasks = count_active_tasks();
1025 do {
1026 CALC_LOAD(avenrun[0], EXP_1, active_tasks);
1027 CALC_LOAD(avenrun[1], EXP_5, active_tasks);
1028 CALC_LOAD(avenrun[2], EXP_15, active_tasks);
1029 count += LOAD_FREQ;
1030 } while (count < 0);
1035 * This function runs timers and the timer-tq in bottom half context.
1037 static void run_timer_softirq(struct softirq_action *h)
1039 struct tvec_base *base = __get_cpu_var(tvec_bases);
1041 hrtimer_run_pending();
1043 if (time_after_eq(jiffies, base->timer_jiffies))
1044 __run_timers(base);
1048 * Called by the local, per-CPU timer interrupt on SMP.
1050 void run_local_timers(void)
1052 hrtimer_run_queues();
1053 raise_softirq(TIMER_SOFTIRQ);
1054 softlockup_tick();
1058 * Called by the timer interrupt. xtime_lock must already be taken
1059 * by the timer IRQ!
1061 static inline void update_times(unsigned long ticks)
1063 update_wall_time();
1064 calc_load(ticks);
1068 * The 64-bit jiffies value is not atomic - you MUST NOT read it
1069 * without sampling the sequence number in xtime_lock.
1070 * jiffies is defined in the linker script...
1073 void do_timer(unsigned long ticks)
1075 jiffies_64 += ticks;
1076 update_times(ticks);
1079 #ifdef __ARCH_WANT_SYS_ALARM
1082 * For backwards compatibility? This can be done in libc so Alpha
1083 * and all newer ports shouldn't need it.
1085 asmlinkage unsigned long sys_alarm(unsigned int seconds)
1087 return alarm_setitimer(seconds);
1090 #endif
1092 #ifndef __alpha__
1095 * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this
1096 * should be moved into arch/i386 instead?
1100 * sys_getpid - return the thread group id of the current process
1102 * Note, despite the name, this returns the tgid not the pid. The tgid and
1103 * the pid are identical unless CLONE_THREAD was specified on clone() in
1104 * which case the tgid is the same in all threads of the same group.
1106 * This is SMP safe as current->tgid does not change.
1108 asmlinkage long sys_getpid(void)
1110 return task_tgid_vnr(current);
1114 * Accessing ->real_parent is not SMP-safe, it could
1115 * change from under us. However, we can use a stale
1116 * value of ->real_parent under rcu_read_lock(), see
1117 * release_task()->call_rcu(delayed_put_task_struct).
1119 asmlinkage long sys_getppid(void)
1121 int pid;
1123 rcu_read_lock();
1124 pid = task_tgid_vnr(current->real_parent);
1125 rcu_read_unlock();
1127 return pid;
1130 asmlinkage long sys_getuid(void)
1132 /* Only we change this so SMP safe */
1133 return current->uid;
1136 asmlinkage long sys_geteuid(void)
1138 /* Only we change this so SMP safe */
1139 return current->euid;
1142 asmlinkage long sys_getgid(void)
1144 /* Only we change this so SMP safe */
1145 return current->gid;
1148 asmlinkage long sys_getegid(void)
1150 /* Only we change this so SMP safe */
1151 return current->egid;
1154 #endif
1156 static void process_timeout(unsigned long __data)
1158 wake_up_process((struct task_struct *)__data);
1162 * schedule_timeout - sleep until timeout
1163 * @timeout: timeout value in jiffies
1165 * Make the current task sleep until @timeout jiffies have
1166 * elapsed. The routine will return immediately unless
1167 * the current task state has been set (see set_current_state()).
1169 * You can set the task state as follows -
1171 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1172 * pass before the routine returns. The routine will return 0
1174 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1175 * delivered to the current task. In this case the remaining time
1176 * in jiffies will be returned, or 0 if the timer expired in time
1178 * The current task state is guaranteed to be TASK_RUNNING when this
1179 * routine returns.
1181 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1182 * the CPU away without a bound on the timeout. In this case the return
1183 * value will be %MAX_SCHEDULE_TIMEOUT.
1185 * In all cases the return value is guaranteed to be non-negative.
1187 signed long __sched schedule_timeout(signed long timeout)
1189 struct timer_list timer;
1190 unsigned long expire;
1192 switch (timeout)
1194 case MAX_SCHEDULE_TIMEOUT:
1196 * These two special cases are useful to be comfortable
1197 * in the caller. Nothing more. We could take
1198 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1199 * but I' d like to return a valid offset (>=0) to allow
1200 * the caller to do everything it want with the retval.
1202 schedule();
1203 goto out;
1204 default:
1206 * Another bit of PARANOID. Note that the retval will be
1207 * 0 since no piece of kernel is supposed to do a check
1208 * for a negative retval of schedule_timeout() (since it
1209 * should never happens anyway). You just have the printk()
1210 * that will tell you if something is gone wrong and where.
1212 if (timeout < 0) {
1213 printk(KERN_ERR "schedule_timeout: wrong timeout "
1214 "value %lx\n", timeout);
1215 dump_stack();
1216 current->state = TASK_RUNNING;
1217 goto out;
1221 expire = timeout + jiffies;
1223 setup_timer_on_stack(&timer, process_timeout, (unsigned long)current);
1224 __mod_timer(&timer, expire);
1225 schedule();
1226 del_singleshot_timer_sync(&timer);
1228 /* Remove the timer from the object tracker */
1229 destroy_timer_on_stack(&timer);
1231 timeout = expire - jiffies;
1233 out:
1234 return timeout < 0 ? 0 : timeout;
1236 EXPORT_SYMBOL(schedule_timeout);
1239 * We can use __set_current_state() here because schedule_timeout() calls
1240 * schedule() unconditionally.
1242 signed long __sched schedule_timeout_interruptible(signed long timeout)
1244 __set_current_state(TASK_INTERRUPTIBLE);
1245 return schedule_timeout(timeout);
1247 EXPORT_SYMBOL(schedule_timeout_interruptible);
1249 signed long __sched schedule_timeout_killable(signed long timeout)
1251 __set_current_state(TASK_KILLABLE);
1252 return schedule_timeout(timeout);
1254 EXPORT_SYMBOL(schedule_timeout_killable);
1256 signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1258 __set_current_state(TASK_UNINTERRUPTIBLE);
1259 return schedule_timeout(timeout);
1261 EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1263 /* Thread ID - the internal kernel "pid" */
1264 asmlinkage long sys_gettid(void)
1266 return task_pid_vnr(current);
1270 * do_sysinfo - fill in sysinfo struct
1271 * @info: pointer to buffer to fill
1273 int do_sysinfo(struct sysinfo *info)
1275 unsigned long mem_total, sav_total;
1276 unsigned int mem_unit, bitcount;
1277 unsigned long seq;
1279 memset(info, 0, sizeof(struct sysinfo));
1281 do {
1282 struct timespec tp;
1283 seq = read_seqbegin(&xtime_lock);
1286 * This is annoying. The below is the same thing
1287 * posix_get_clock_monotonic() does, but it wants to
1288 * take the lock which we want to cover the loads stuff
1289 * too.
1292 getnstimeofday(&tp);
1293 tp.tv_sec += wall_to_monotonic.tv_sec;
1294 tp.tv_nsec += wall_to_monotonic.tv_nsec;
1295 monotonic_to_bootbased(&tp);
1296 if (tp.tv_nsec - NSEC_PER_SEC >= 0) {
1297 tp.tv_nsec = tp.tv_nsec - NSEC_PER_SEC;
1298 tp.tv_sec++;
1300 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
1302 info->loads[0] = avenrun[0] << (SI_LOAD_SHIFT - FSHIFT);
1303 info->loads[1] = avenrun[1] << (SI_LOAD_SHIFT - FSHIFT);
1304 info->loads[2] = avenrun[2] << (SI_LOAD_SHIFT - FSHIFT);
1306 info->procs = nr_threads;
1307 } while (read_seqretry(&xtime_lock, seq));
1309 si_meminfo(info);
1310 si_swapinfo(info);
1313 * If the sum of all the available memory (i.e. ram + swap)
1314 * is less than can be stored in a 32 bit unsigned long then
1315 * we can be binary compatible with 2.2.x kernels. If not,
1316 * well, in that case 2.2.x was broken anyways...
1318 * -Erik Andersen <andersee@debian.org>
1321 mem_total = info->totalram + info->totalswap;
1322 if (mem_total < info->totalram || mem_total < info->totalswap)
1323 goto out;
1324 bitcount = 0;
1325 mem_unit = info->mem_unit;
1326 while (mem_unit > 1) {
1327 bitcount++;
1328 mem_unit >>= 1;
1329 sav_total = mem_total;
1330 mem_total <<= 1;
1331 if (mem_total < sav_total)
1332 goto out;
1336 * If mem_total did not overflow, multiply all memory values by
1337 * info->mem_unit and set it to 1. This leaves things compatible
1338 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1339 * kernels...
1342 info->mem_unit = 1;
1343 info->totalram <<= bitcount;
1344 info->freeram <<= bitcount;
1345 info->sharedram <<= bitcount;
1346 info->bufferram <<= bitcount;
1347 info->totalswap <<= bitcount;
1348 info->freeswap <<= bitcount;
1349 info->totalhigh <<= bitcount;
1350 info->freehigh <<= bitcount;
1352 out:
1353 return 0;
1356 asmlinkage long sys_sysinfo(struct sysinfo __user *info)
1358 struct sysinfo val;
1360 do_sysinfo(&val);
1362 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
1363 return -EFAULT;
1365 return 0;
1368 static int __cpuinit init_timers_cpu(int cpu)
1370 int j;
1371 struct tvec_base *base;
1372 static char __cpuinitdata tvec_base_done[NR_CPUS];
1374 if (!tvec_base_done[cpu]) {
1375 static char boot_done;
1377 if (boot_done) {
1379 * The APs use this path later in boot
1381 base = kmalloc_node(sizeof(*base),
1382 GFP_KERNEL | __GFP_ZERO,
1383 cpu_to_node(cpu));
1384 if (!base)
1385 return -ENOMEM;
1387 /* Make sure that tvec_base is 2 byte aligned */
1388 if (tbase_get_deferrable(base)) {
1389 WARN_ON(1);
1390 kfree(base);
1391 return -ENOMEM;
1393 per_cpu(tvec_bases, cpu) = base;
1394 } else {
1396 * This is for the boot CPU - we use compile-time
1397 * static initialisation because per-cpu memory isn't
1398 * ready yet and because the memory allocators are not
1399 * initialised either.
1401 boot_done = 1;
1402 base = &boot_tvec_bases;
1404 tvec_base_done[cpu] = 1;
1405 } else {
1406 base = per_cpu(tvec_bases, cpu);
1409 spin_lock_init(&base->lock);
1411 for (j = 0; j < TVN_SIZE; j++) {
1412 INIT_LIST_HEAD(base->tv5.vec + j);
1413 INIT_LIST_HEAD(base->tv4.vec + j);
1414 INIT_LIST_HEAD(base->tv3.vec + j);
1415 INIT_LIST_HEAD(base->tv2.vec + j);
1417 for (j = 0; j < TVR_SIZE; j++)
1418 INIT_LIST_HEAD(base->tv1.vec + j);
1420 base->timer_jiffies = jiffies;
1421 return 0;
1424 #ifdef CONFIG_HOTPLUG_CPU
1425 static void migrate_timer_list(struct tvec_base *new_base, struct list_head *head)
1427 struct timer_list *timer;
1429 while (!list_empty(head)) {
1430 timer = list_first_entry(head, struct timer_list, entry);
1431 detach_timer(timer, 0);
1432 timer_set_base(timer, new_base);
1433 internal_add_timer(new_base, timer);
1437 static void __cpuinit migrate_timers(int cpu)
1439 struct tvec_base *old_base;
1440 struct tvec_base *new_base;
1441 int i;
1443 BUG_ON(cpu_online(cpu));
1444 old_base = per_cpu(tvec_bases, cpu);
1445 new_base = get_cpu_var(tvec_bases);
1447 local_irq_disable();
1448 spin_lock(&new_base->lock);
1449 spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1451 BUG_ON(old_base->running_timer);
1453 for (i = 0; i < TVR_SIZE; i++)
1454 migrate_timer_list(new_base, old_base->tv1.vec + i);
1455 for (i = 0; i < TVN_SIZE; i++) {
1456 migrate_timer_list(new_base, old_base->tv2.vec + i);
1457 migrate_timer_list(new_base, old_base->tv3.vec + i);
1458 migrate_timer_list(new_base, old_base->tv4.vec + i);
1459 migrate_timer_list(new_base, old_base->tv5.vec + i);
1462 spin_unlock(&old_base->lock);
1463 spin_unlock(&new_base->lock);
1464 local_irq_enable();
1465 put_cpu_var(tvec_bases);
1467 #endif /* CONFIG_HOTPLUG_CPU */
1469 static int __cpuinit timer_cpu_notify(struct notifier_block *self,
1470 unsigned long action, void *hcpu)
1472 long cpu = (long)hcpu;
1473 switch(action) {
1474 case CPU_UP_PREPARE:
1475 case CPU_UP_PREPARE_FROZEN:
1476 if (init_timers_cpu(cpu) < 0)
1477 return NOTIFY_BAD;
1478 break;
1479 #ifdef CONFIG_HOTPLUG_CPU
1480 case CPU_DEAD:
1481 case CPU_DEAD_FROZEN:
1482 migrate_timers(cpu);
1483 break;
1484 #endif
1485 default:
1486 break;
1488 return NOTIFY_OK;
1491 static struct notifier_block __cpuinitdata timers_nb = {
1492 .notifier_call = timer_cpu_notify,
1496 void __init init_timers(void)
1498 int err = timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE,
1499 (void *)(long)smp_processor_id());
1501 init_timer_stats();
1503 BUG_ON(err == NOTIFY_BAD);
1504 register_cpu_notifier(&timers_nb);
1505 open_softirq(TIMER_SOFTIRQ, run_timer_softirq, NULL);
1509 * msleep - sleep safely even with waitqueue interruptions
1510 * @msecs: Time in milliseconds to sleep for
1512 void msleep(unsigned int msecs)
1514 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1516 while (timeout)
1517 timeout = schedule_timeout_uninterruptible(timeout);
1520 EXPORT_SYMBOL(msleep);
1523 * msleep_interruptible - sleep waiting for signals
1524 * @msecs: Time in milliseconds to sleep for
1526 unsigned long msleep_interruptible(unsigned int msecs)
1528 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1530 while (timeout && !signal_pending(current))
1531 timeout = schedule_timeout_interruptible(timeout);
1532 return jiffies_to_msecs(timeout);
1535 EXPORT_SYMBOL(msleep_interruptible);