V4L/DVB: IR/imon: remove bad ir_input_dev use
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / kernel / timer.c
blobf1b8afe1ad86bd609f2bb63133ca1f0840aa86d3
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>
40 #include <linux/perf_event.h>
41 #include <linux/sched.h>
42 #include <linux/slab.h>
44 #include <asm/uaccess.h>
45 #include <asm/unistd.h>
46 #include <asm/div64.h>
47 #include <asm/timex.h>
48 #include <asm/io.h>
50 #define CREATE_TRACE_POINTS
51 #include <trace/events/timer.h>
53 u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
55 EXPORT_SYMBOL(jiffies_64);
58 * per-CPU timer vector definitions:
60 #define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6)
61 #define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8)
62 #define TVN_SIZE (1 << TVN_BITS)
63 #define TVR_SIZE (1 << TVR_BITS)
64 #define TVN_MASK (TVN_SIZE - 1)
65 #define TVR_MASK (TVR_SIZE - 1)
67 struct tvec {
68 struct list_head vec[TVN_SIZE];
71 struct tvec_root {
72 struct list_head vec[TVR_SIZE];
75 struct tvec_base {
76 spinlock_t lock;
77 struct timer_list *running_timer;
78 unsigned long timer_jiffies;
79 unsigned long next_timer;
80 struct tvec_root tv1;
81 struct tvec tv2;
82 struct tvec tv3;
83 struct tvec tv4;
84 struct tvec tv5;
85 } ____cacheline_aligned;
87 struct tvec_base boot_tvec_bases;
88 EXPORT_SYMBOL(boot_tvec_bases);
89 static DEFINE_PER_CPU(struct tvec_base *, tvec_bases) = &boot_tvec_bases;
92 * Note that all tvec_bases are 2 byte aligned and lower bit of
93 * base in timer_list is guaranteed to be zero. Use the LSB to
94 * indicate whether the timer is deferrable.
96 * A deferrable timer will work normally when the system is busy, but
97 * will not cause a CPU to come out of idle just to service it; instead,
98 * the timer will be serviced when the CPU eventually wakes up with a
99 * subsequent non-deferrable timer.
101 #define TBASE_DEFERRABLE_FLAG (0x1)
103 /* Functions below help us manage 'deferrable' flag */
104 static inline unsigned int tbase_get_deferrable(struct tvec_base *base)
106 return ((unsigned int)(unsigned long)base & TBASE_DEFERRABLE_FLAG);
109 static inline struct tvec_base *tbase_get_base(struct tvec_base *base)
111 return ((struct tvec_base *)((unsigned long)base & ~TBASE_DEFERRABLE_FLAG));
114 static inline void timer_set_deferrable(struct timer_list *timer)
116 timer->base = ((struct tvec_base *)((unsigned long)(timer->base) |
117 TBASE_DEFERRABLE_FLAG));
120 static inline void
121 timer_set_base(struct timer_list *timer, struct tvec_base *new_base)
123 timer->base = (struct tvec_base *)((unsigned long)(new_base) |
124 tbase_get_deferrable(timer->base));
127 static unsigned long round_jiffies_common(unsigned long j, int cpu,
128 bool force_up)
130 int rem;
131 unsigned long original = j;
134 * We don't want all cpus firing their timers at once hitting the
135 * same lock or cachelines, so we skew each extra cpu with an extra
136 * 3 jiffies. This 3 jiffies came originally from the mm/ code which
137 * already did this.
138 * The skew is done by adding 3*cpunr, then round, then subtract this
139 * extra offset again.
141 j += cpu * 3;
143 rem = j % HZ;
146 * If the target jiffie is just after a whole second (which can happen
147 * due to delays of the timer irq, long irq off times etc etc) then
148 * we should round down to the whole second, not up. Use 1/4th second
149 * as cutoff for this rounding as an extreme upper bound for this.
150 * But never round down if @force_up is set.
152 if (rem < HZ/4 && !force_up) /* round down */
153 j = j - rem;
154 else /* round up */
155 j = j - rem + HZ;
157 /* now that we have rounded, subtract the extra skew again */
158 j -= cpu * 3;
160 if (j <= jiffies) /* rounding ate our timeout entirely; */
161 return original;
162 return j;
166 * __round_jiffies - function to round jiffies to a full second
167 * @j: the time in (absolute) jiffies that should be rounded
168 * @cpu: the processor number on which the timeout will happen
170 * __round_jiffies() rounds an absolute time in the future (in jiffies)
171 * up or down to (approximately) full seconds. This is useful for timers
172 * for which the exact time they fire does not matter too much, as long as
173 * they fire approximately every X seconds.
175 * By rounding these timers to whole seconds, all such timers will fire
176 * at the same time, rather than at various times spread out. The goal
177 * of this is to have the CPU wake up less, which saves power.
179 * The exact rounding is skewed for each processor to avoid all
180 * processors firing at the exact same time, which could lead
181 * to lock contention or spurious cache line bouncing.
183 * The return value is the rounded version of the @j parameter.
185 unsigned long __round_jiffies(unsigned long j, int cpu)
187 return round_jiffies_common(j, cpu, false);
189 EXPORT_SYMBOL_GPL(__round_jiffies);
192 * __round_jiffies_relative - function to round jiffies to a full second
193 * @j: the time in (relative) jiffies that should be rounded
194 * @cpu: the processor number on which the timeout will happen
196 * __round_jiffies_relative() rounds a time delta in the future (in jiffies)
197 * up or down to (approximately) full seconds. This is useful for timers
198 * for which the exact time they fire does not matter too much, as long as
199 * they fire approximately every X seconds.
201 * By rounding these timers to whole seconds, all such timers will fire
202 * at the same time, rather than at various times spread out. The goal
203 * of this is to have the CPU wake up less, which saves power.
205 * The exact rounding is skewed for each processor to avoid all
206 * processors firing at the exact same time, which could lead
207 * to lock contention or spurious cache line bouncing.
209 * The return value is the rounded version of the @j parameter.
211 unsigned long __round_jiffies_relative(unsigned long j, int cpu)
213 unsigned long j0 = jiffies;
215 /* Use j0 because jiffies might change while we run */
216 return round_jiffies_common(j + j0, cpu, false) - j0;
218 EXPORT_SYMBOL_GPL(__round_jiffies_relative);
221 * round_jiffies - function to round jiffies to a full second
222 * @j: the time in (absolute) jiffies that should be rounded
224 * round_jiffies() rounds an absolute time in the future (in jiffies)
225 * up or down to (approximately) full seconds. This is useful for timers
226 * for which the exact time they fire does not matter too much, as long as
227 * they fire approximately every X seconds.
229 * By rounding these timers to whole seconds, all such timers will fire
230 * at the same time, rather than at various times spread out. The goal
231 * of this is to have the CPU wake up less, which saves power.
233 * The return value is the rounded version of the @j parameter.
235 unsigned long round_jiffies(unsigned long j)
237 return round_jiffies_common(j, raw_smp_processor_id(), false);
239 EXPORT_SYMBOL_GPL(round_jiffies);
242 * round_jiffies_relative - function to round jiffies to a full second
243 * @j: the time in (relative) jiffies that should be rounded
245 * round_jiffies_relative() rounds a time delta in the future (in jiffies)
246 * up or down to (approximately) full seconds. This is useful for timers
247 * for which the exact time they fire does not matter too much, as long as
248 * they fire approximately every X seconds.
250 * By rounding these timers to whole seconds, all such timers will fire
251 * at the same time, rather than at various times spread out. The goal
252 * of this is to have the CPU wake up less, which saves power.
254 * The return value is the rounded version of the @j parameter.
256 unsigned long round_jiffies_relative(unsigned long j)
258 return __round_jiffies_relative(j, raw_smp_processor_id());
260 EXPORT_SYMBOL_GPL(round_jiffies_relative);
263 * __round_jiffies_up - function to round jiffies up to a full second
264 * @j: the time in (absolute) jiffies that should be rounded
265 * @cpu: the processor number on which the timeout will happen
267 * This is the same as __round_jiffies() except that it will never
268 * round down. This is useful for timeouts for which the exact time
269 * of firing does not matter too much, as long as they don't fire too
270 * early.
272 unsigned long __round_jiffies_up(unsigned long j, int cpu)
274 return round_jiffies_common(j, cpu, true);
276 EXPORT_SYMBOL_GPL(__round_jiffies_up);
279 * __round_jiffies_up_relative - function to round jiffies up to a full second
280 * @j: the time in (relative) jiffies that should be rounded
281 * @cpu: the processor number on which the timeout will happen
283 * This is the same as __round_jiffies_relative() except that it will never
284 * round down. This is useful for timeouts for which the exact time
285 * of firing does not matter too much, as long as they don't fire too
286 * early.
288 unsigned long __round_jiffies_up_relative(unsigned long j, int cpu)
290 unsigned long j0 = jiffies;
292 /* Use j0 because jiffies might change while we run */
293 return round_jiffies_common(j + j0, cpu, true) - j0;
295 EXPORT_SYMBOL_GPL(__round_jiffies_up_relative);
298 * round_jiffies_up - function to round jiffies up to a full second
299 * @j: the time in (absolute) jiffies that should be rounded
301 * This is the same as round_jiffies() except that it will never
302 * round down. This is useful for timeouts for which the exact time
303 * of firing does not matter too much, as long as they don't fire too
304 * early.
306 unsigned long round_jiffies_up(unsigned long j)
308 return round_jiffies_common(j, raw_smp_processor_id(), true);
310 EXPORT_SYMBOL_GPL(round_jiffies_up);
313 * round_jiffies_up_relative - function to round jiffies up to a full second
314 * @j: the time in (relative) jiffies that should be rounded
316 * This is the same as round_jiffies_relative() except that it will never
317 * round down. This is useful for timeouts for which the exact time
318 * of firing does not matter too much, as long as they don't fire too
319 * early.
321 unsigned long round_jiffies_up_relative(unsigned long j)
323 return __round_jiffies_up_relative(j, raw_smp_processor_id());
325 EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
328 * set_timer_slack - set the allowed slack for a timer
329 * @slack_hz: the amount of time (in jiffies) allowed for rounding
331 * Set the amount of time, in jiffies, that a certain timer has
332 * in terms of slack. By setting this value, the timer subsystem
333 * will schedule the actual timer somewhere between
334 * the time mod_timer() asks for, and that time plus the slack.
336 * By setting the slack to -1, a percentage of the delay is used
337 * instead.
339 void set_timer_slack(struct timer_list *timer, int slack_hz)
341 timer->slack = slack_hz;
343 EXPORT_SYMBOL_GPL(set_timer_slack);
346 static inline void set_running_timer(struct tvec_base *base,
347 struct timer_list *timer)
349 #ifdef CONFIG_SMP
350 base->running_timer = timer;
351 #endif
354 static void internal_add_timer(struct tvec_base *base, struct timer_list *timer)
356 unsigned long expires = timer->expires;
357 unsigned long idx = expires - base->timer_jiffies;
358 struct list_head *vec;
360 if (idx < TVR_SIZE) {
361 int i = expires & TVR_MASK;
362 vec = base->tv1.vec + i;
363 } else if (idx < 1 << (TVR_BITS + TVN_BITS)) {
364 int i = (expires >> TVR_BITS) & TVN_MASK;
365 vec = base->tv2.vec + i;
366 } else if (idx < 1 << (TVR_BITS + 2 * TVN_BITS)) {
367 int i = (expires >> (TVR_BITS + TVN_BITS)) & TVN_MASK;
368 vec = base->tv3.vec + i;
369 } else if (idx < 1 << (TVR_BITS + 3 * TVN_BITS)) {
370 int i = (expires >> (TVR_BITS + 2 * TVN_BITS)) & TVN_MASK;
371 vec = base->tv4.vec + i;
372 } else if ((signed long) idx < 0) {
374 * Can happen if you add a timer with expires == jiffies,
375 * or you set a timer to go off in the past
377 vec = base->tv1.vec + (base->timer_jiffies & TVR_MASK);
378 } else {
379 int i;
380 /* If the timeout is larger than 0xffffffff on 64-bit
381 * architectures then we use the maximum timeout:
383 if (idx > 0xffffffffUL) {
384 idx = 0xffffffffUL;
385 expires = idx + base->timer_jiffies;
387 i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK;
388 vec = base->tv5.vec + i;
391 * Timers are FIFO:
393 list_add_tail(&timer->entry, vec);
396 #ifdef CONFIG_TIMER_STATS
397 void __timer_stats_timer_set_start_info(struct timer_list *timer, void *addr)
399 if (timer->start_site)
400 return;
402 timer->start_site = addr;
403 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
404 timer->start_pid = current->pid;
407 static void timer_stats_account_timer(struct timer_list *timer)
409 unsigned int flag = 0;
411 if (likely(!timer->start_site))
412 return;
413 if (unlikely(tbase_get_deferrable(timer->base)))
414 flag |= TIMER_STATS_FLAG_DEFERRABLE;
416 timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
417 timer->function, timer->start_comm, flag);
420 #else
421 static void timer_stats_account_timer(struct timer_list *timer) {}
422 #endif
424 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
426 static struct debug_obj_descr timer_debug_descr;
429 * fixup_init is called when:
430 * - an active object is initialized
432 static int timer_fixup_init(void *addr, enum debug_obj_state state)
434 struct timer_list *timer = addr;
436 switch (state) {
437 case ODEBUG_STATE_ACTIVE:
438 del_timer_sync(timer);
439 debug_object_init(timer, &timer_debug_descr);
440 return 1;
441 default:
442 return 0;
447 * fixup_activate is called when:
448 * - an active object is activated
449 * - an unknown object is activated (might be a statically initialized object)
451 static int timer_fixup_activate(void *addr, enum debug_obj_state state)
453 struct timer_list *timer = addr;
455 switch (state) {
457 case ODEBUG_STATE_NOTAVAILABLE:
459 * This is not really a fixup. The timer was
460 * statically initialized. We just make sure that it
461 * is tracked in the object tracker.
463 if (timer->entry.next == NULL &&
464 timer->entry.prev == TIMER_ENTRY_STATIC) {
465 debug_object_init(timer, &timer_debug_descr);
466 debug_object_activate(timer, &timer_debug_descr);
467 return 0;
468 } else {
469 WARN_ON_ONCE(1);
471 return 0;
473 case ODEBUG_STATE_ACTIVE:
474 WARN_ON(1);
476 default:
477 return 0;
482 * fixup_free is called when:
483 * - an active object is freed
485 static int timer_fixup_free(void *addr, enum debug_obj_state state)
487 struct timer_list *timer = addr;
489 switch (state) {
490 case ODEBUG_STATE_ACTIVE:
491 del_timer_sync(timer);
492 debug_object_free(timer, &timer_debug_descr);
493 return 1;
494 default:
495 return 0;
499 static struct debug_obj_descr timer_debug_descr = {
500 .name = "timer_list",
501 .fixup_init = timer_fixup_init,
502 .fixup_activate = timer_fixup_activate,
503 .fixup_free = timer_fixup_free,
506 static inline void debug_timer_init(struct timer_list *timer)
508 debug_object_init(timer, &timer_debug_descr);
511 static inline void debug_timer_activate(struct timer_list *timer)
513 debug_object_activate(timer, &timer_debug_descr);
516 static inline void debug_timer_deactivate(struct timer_list *timer)
518 debug_object_deactivate(timer, &timer_debug_descr);
521 static inline void debug_timer_free(struct timer_list *timer)
523 debug_object_free(timer, &timer_debug_descr);
526 static void __init_timer(struct timer_list *timer,
527 const char *name,
528 struct lock_class_key *key);
530 void init_timer_on_stack_key(struct timer_list *timer,
531 const char *name,
532 struct lock_class_key *key)
534 debug_object_init_on_stack(timer, &timer_debug_descr);
535 __init_timer(timer, name, key);
537 EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
539 void destroy_timer_on_stack(struct timer_list *timer)
541 debug_object_free(timer, &timer_debug_descr);
543 EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
545 #else
546 static inline void debug_timer_init(struct timer_list *timer) { }
547 static inline void debug_timer_activate(struct timer_list *timer) { }
548 static inline void debug_timer_deactivate(struct timer_list *timer) { }
549 #endif
551 static inline void debug_init(struct timer_list *timer)
553 debug_timer_init(timer);
554 trace_timer_init(timer);
557 static inline void
558 debug_activate(struct timer_list *timer, unsigned long expires)
560 debug_timer_activate(timer);
561 trace_timer_start(timer, expires);
564 static inline void debug_deactivate(struct timer_list *timer)
566 debug_timer_deactivate(timer);
567 trace_timer_cancel(timer);
570 static void __init_timer(struct timer_list *timer,
571 const char *name,
572 struct lock_class_key *key)
574 timer->entry.next = NULL;
575 timer->base = __raw_get_cpu_var(tvec_bases);
576 timer->slack = -1;
577 #ifdef CONFIG_TIMER_STATS
578 timer->start_site = NULL;
579 timer->start_pid = -1;
580 memset(timer->start_comm, 0, TASK_COMM_LEN);
581 #endif
582 lockdep_init_map(&timer->lockdep_map, name, key, 0);
585 void setup_deferrable_timer_on_stack_key(struct timer_list *timer,
586 const char *name,
587 struct lock_class_key *key,
588 void (*function)(unsigned long),
589 unsigned long data)
591 timer->function = function;
592 timer->data = data;
593 init_timer_on_stack_key(timer, name, key);
594 timer_set_deferrable(timer);
596 EXPORT_SYMBOL_GPL(setup_deferrable_timer_on_stack_key);
599 * init_timer_key - initialize a timer
600 * @timer: the timer to be initialized
601 * @name: name of the timer
602 * @key: lockdep class key of the fake lock used for tracking timer
603 * sync lock dependencies
605 * init_timer_key() must be done to a timer prior calling *any* of the
606 * other timer functions.
608 void init_timer_key(struct timer_list *timer,
609 const char *name,
610 struct lock_class_key *key)
612 debug_init(timer);
613 __init_timer(timer, name, key);
615 EXPORT_SYMBOL(init_timer_key);
617 void init_timer_deferrable_key(struct timer_list *timer,
618 const char *name,
619 struct lock_class_key *key)
621 init_timer_key(timer, name, key);
622 timer_set_deferrable(timer);
624 EXPORT_SYMBOL(init_timer_deferrable_key);
626 static inline void detach_timer(struct timer_list *timer,
627 int clear_pending)
629 struct list_head *entry = &timer->entry;
631 debug_deactivate(timer);
633 __list_del(entry->prev, entry->next);
634 if (clear_pending)
635 entry->next = NULL;
636 entry->prev = LIST_POISON2;
640 * We are using hashed locking: holding per_cpu(tvec_bases).lock
641 * means that all timers which are tied to this base via timer->base are
642 * locked, and the base itself is locked too.
644 * So __run_timers/migrate_timers can safely modify all timers which could
645 * be found on ->tvX lists.
647 * When the timer's base is locked, and the timer removed from list, it is
648 * possible to set timer->base = NULL and drop the lock: the timer remains
649 * locked.
651 static struct tvec_base *lock_timer_base(struct timer_list *timer,
652 unsigned long *flags)
653 __acquires(timer->base->lock)
655 struct tvec_base *base;
657 for (;;) {
658 struct tvec_base *prelock_base = timer->base;
659 base = tbase_get_base(prelock_base);
660 if (likely(base != NULL)) {
661 spin_lock_irqsave(&base->lock, *flags);
662 if (likely(prelock_base == timer->base))
663 return base;
664 /* The timer has migrated to another CPU */
665 spin_unlock_irqrestore(&base->lock, *flags);
667 cpu_relax();
671 static inline int
672 __mod_timer(struct timer_list *timer, unsigned long expires,
673 bool pending_only, int pinned)
675 struct tvec_base *base, *new_base;
676 unsigned long flags;
677 int ret = 0 , cpu;
679 timer_stats_timer_set_start_info(timer);
680 BUG_ON(!timer->function);
682 base = lock_timer_base(timer, &flags);
684 if (timer_pending(timer)) {
685 detach_timer(timer, 0);
686 if (timer->expires == base->next_timer &&
687 !tbase_get_deferrable(timer->base))
688 base->next_timer = base->timer_jiffies;
689 ret = 1;
690 } else {
691 if (pending_only)
692 goto out_unlock;
695 debug_activate(timer, expires);
697 cpu = smp_processor_id();
699 #if defined(CONFIG_NO_HZ) && defined(CONFIG_SMP)
700 if (!pinned && get_sysctl_timer_migration() && idle_cpu(cpu))
701 cpu = get_nohz_timer_target();
702 #endif
703 new_base = per_cpu(tvec_bases, cpu);
705 if (base != new_base) {
707 * We are trying to schedule the timer on the local CPU.
708 * However we can't change timer's base while it is running,
709 * otherwise del_timer_sync() can't detect that the timer's
710 * handler yet has not finished. This also guarantees that
711 * the timer is serialized wrt itself.
713 if (likely(base->running_timer != timer)) {
714 /* See the comment in lock_timer_base() */
715 timer_set_base(timer, NULL);
716 spin_unlock(&base->lock);
717 base = new_base;
718 spin_lock(&base->lock);
719 timer_set_base(timer, base);
723 timer->expires = expires;
724 if (time_before(timer->expires, base->next_timer) &&
725 !tbase_get_deferrable(timer->base))
726 base->next_timer = timer->expires;
727 internal_add_timer(base, timer);
729 out_unlock:
730 spin_unlock_irqrestore(&base->lock, flags);
732 return ret;
736 * mod_timer_pending - modify a pending timer's timeout
737 * @timer: the pending timer to be modified
738 * @expires: new timeout in jiffies
740 * mod_timer_pending() is the same for pending timers as mod_timer(),
741 * but will not re-activate and modify already deleted timers.
743 * It is useful for unserialized use of timers.
745 int mod_timer_pending(struct timer_list *timer, unsigned long expires)
747 return __mod_timer(timer, expires, true, TIMER_NOT_PINNED);
749 EXPORT_SYMBOL(mod_timer_pending);
752 * Decide where to put the timer while taking the slack into account
754 * Algorithm:
755 * 1) calculate the maximum (absolute) time
756 * 2) calculate the highest bit where the expires and new max are different
757 * 3) use this bit to make a mask
758 * 4) use the bitmask to round down the maximum time, so that all last
759 * bits are zeros
761 static inline
762 unsigned long apply_slack(struct timer_list *timer, unsigned long expires)
764 unsigned long expires_limit, mask;
765 int bit;
767 expires_limit = expires;
769 if (timer->slack >= 0) {
770 expires_limit = expires + timer->slack;
771 } else {
772 unsigned long now = jiffies;
774 /* No slack, if already expired else auto slack 0.4% */
775 if (time_after(expires, now))
776 expires_limit = expires + (expires - now)/256;
778 mask = expires ^ expires_limit;
779 if (mask == 0)
780 return expires;
782 bit = find_last_bit(&mask, BITS_PER_LONG);
784 mask = (1 << bit) - 1;
786 expires_limit = expires_limit & ~(mask);
788 return expires_limit;
792 * mod_timer - modify a timer's timeout
793 * @timer: the timer to be modified
794 * @expires: new timeout in jiffies
796 * mod_timer() is a more efficient way to update the expire field of an
797 * active timer (if the timer is inactive it will be activated)
799 * mod_timer(timer, expires) is equivalent to:
801 * del_timer(timer); timer->expires = expires; add_timer(timer);
803 * Note that if there are multiple unserialized concurrent users of the
804 * same timer, then mod_timer() is the only safe way to modify the timeout,
805 * since add_timer() cannot modify an already running timer.
807 * The function returns whether it has modified a pending timer or not.
808 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
809 * active timer returns 1.)
811 int mod_timer(struct timer_list *timer, unsigned long expires)
814 * This is a common optimization triggered by the
815 * networking code - if the timer is re-modified
816 * to be the same thing then just return:
818 if (timer_pending(timer) && timer->expires == expires)
819 return 1;
821 expires = apply_slack(timer, expires);
823 return __mod_timer(timer, expires, false, TIMER_NOT_PINNED);
825 EXPORT_SYMBOL(mod_timer);
828 * mod_timer_pinned - modify a timer's timeout
829 * @timer: the timer to be modified
830 * @expires: new timeout in jiffies
832 * mod_timer_pinned() is a way to update the expire field of an
833 * active timer (if the timer is inactive it will be activated)
834 * and not allow the timer to be migrated to a different CPU.
836 * mod_timer_pinned(timer, expires) is equivalent to:
838 * del_timer(timer); timer->expires = expires; add_timer(timer);
840 int mod_timer_pinned(struct timer_list *timer, unsigned long expires)
842 if (timer->expires == expires && timer_pending(timer))
843 return 1;
845 return __mod_timer(timer, expires, false, TIMER_PINNED);
847 EXPORT_SYMBOL(mod_timer_pinned);
850 * add_timer - start a timer
851 * @timer: the timer to be added
853 * The kernel will do a ->function(->data) callback from the
854 * timer interrupt at the ->expires point in the future. The
855 * current time is 'jiffies'.
857 * The timer's ->expires, ->function (and if the handler uses it, ->data)
858 * fields must be set prior calling this function.
860 * Timers with an ->expires field in the past will be executed in the next
861 * timer tick.
863 void add_timer(struct timer_list *timer)
865 BUG_ON(timer_pending(timer));
866 mod_timer(timer, timer->expires);
868 EXPORT_SYMBOL(add_timer);
871 * add_timer_on - start a timer on a particular CPU
872 * @timer: the timer to be added
873 * @cpu: the CPU to start it on
875 * This is not very scalable on SMP. Double adds are not possible.
877 void add_timer_on(struct timer_list *timer, int cpu)
879 struct tvec_base *base = per_cpu(tvec_bases, cpu);
880 unsigned long flags;
882 timer_stats_timer_set_start_info(timer);
883 BUG_ON(timer_pending(timer) || !timer->function);
884 spin_lock_irqsave(&base->lock, flags);
885 timer_set_base(timer, base);
886 debug_activate(timer, timer->expires);
887 if (time_before(timer->expires, base->next_timer) &&
888 !tbase_get_deferrable(timer->base))
889 base->next_timer = timer->expires;
890 internal_add_timer(base, timer);
892 * Check whether the other CPU is idle and needs to be
893 * triggered to reevaluate the timer wheel when nohz is
894 * active. We are protected against the other CPU fiddling
895 * with the timer by holding the timer base lock. This also
896 * makes sure that a CPU on the way to idle can not evaluate
897 * the timer wheel.
899 wake_up_idle_cpu(cpu);
900 spin_unlock_irqrestore(&base->lock, flags);
902 EXPORT_SYMBOL_GPL(add_timer_on);
905 * del_timer - deactive a timer.
906 * @timer: the timer to be deactivated
908 * del_timer() deactivates a timer - this works on both active and inactive
909 * timers.
911 * The function returns whether it has deactivated a pending timer or not.
912 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
913 * active timer returns 1.)
915 int del_timer(struct timer_list *timer)
917 struct tvec_base *base;
918 unsigned long flags;
919 int ret = 0;
921 timer_stats_timer_clear_start_info(timer);
922 if (timer_pending(timer)) {
923 base = lock_timer_base(timer, &flags);
924 if (timer_pending(timer)) {
925 detach_timer(timer, 1);
926 if (timer->expires == base->next_timer &&
927 !tbase_get_deferrable(timer->base))
928 base->next_timer = base->timer_jiffies;
929 ret = 1;
931 spin_unlock_irqrestore(&base->lock, flags);
934 return ret;
936 EXPORT_SYMBOL(del_timer);
938 #ifdef CONFIG_SMP
940 * try_to_del_timer_sync - Try to deactivate a timer
941 * @timer: timer do del
943 * This function tries to deactivate a timer. Upon successful (ret >= 0)
944 * exit the timer is not queued and the handler is not running on any CPU.
946 * It must not be called from interrupt contexts.
948 int try_to_del_timer_sync(struct timer_list *timer)
950 struct tvec_base *base;
951 unsigned long flags;
952 int ret = -1;
954 base = lock_timer_base(timer, &flags);
956 if (base->running_timer == timer)
957 goto out;
959 timer_stats_timer_clear_start_info(timer);
960 ret = 0;
961 if (timer_pending(timer)) {
962 detach_timer(timer, 1);
963 if (timer->expires == base->next_timer &&
964 !tbase_get_deferrable(timer->base))
965 base->next_timer = base->timer_jiffies;
966 ret = 1;
968 out:
969 spin_unlock_irqrestore(&base->lock, flags);
971 return ret;
973 EXPORT_SYMBOL(try_to_del_timer_sync);
976 * del_timer_sync - deactivate a timer and wait for the handler to finish.
977 * @timer: the timer to be deactivated
979 * This function only differs from del_timer() on SMP: besides deactivating
980 * the timer it also makes sure the handler has finished executing on other
981 * CPUs.
983 * Synchronization rules: Callers must prevent restarting of the timer,
984 * otherwise this function is meaningless. It must not be called from
985 * interrupt contexts. The caller must not hold locks which would prevent
986 * completion of the timer's handler. The timer's handler must not call
987 * add_timer_on(). Upon exit the timer is not queued and the handler is
988 * not running on any CPU.
990 * The function returns whether it has deactivated a pending timer or not.
992 int del_timer_sync(struct timer_list *timer)
994 #ifdef CONFIG_LOCKDEP
995 unsigned long flags;
997 local_irq_save(flags);
998 lock_map_acquire(&timer->lockdep_map);
999 lock_map_release(&timer->lockdep_map);
1000 local_irq_restore(flags);
1001 #endif
1003 for (;;) {
1004 int ret = try_to_del_timer_sync(timer);
1005 if (ret >= 0)
1006 return ret;
1007 cpu_relax();
1010 EXPORT_SYMBOL(del_timer_sync);
1011 #endif
1013 static int cascade(struct tvec_base *base, struct tvec *tv, int index)
1015 /* cascade all the timers from tv up one level */
1016 struct timer_list *timer, *tmp;
1017 struct list_head tv_list;
1019 list_replace_init(tv->vec + index, &tv_list);
1022 * We are removing _all_ timers from the list, so we
1023 * don't have to detach them individually.
1025 list_for_each_entry_safe(timer, tmp, &tv_list, entry) {
1026 BUG_ON(tbase_get_base(timer->base) != base);
1027 internal_add_timer(base, timer);
1030 return index;
1033 static void call_timer_fn(struct timer_list *timer, void (*fn)(unsigned long),
1034 unsigned long data)
1036 int preempt_count = preempt_count();
1038 #ifdef CONFIG_LOCKDEP
1040 * It is permissible to free the timer from inside the
1041 * function that is called from it, this we need to take into
1042 * account for lockdep too. To avoid bogus "held lock freed"
1043 * warnings as well as problems when looking into
1044 * timer->lockdep_map, make a copy and use that here.
1046 struct lockdep_map lockdep_map = timer->lockdep_map;
1047 #endif
1049 * Couple the lock chain with the lock chain at
1050 * del_timer_sync() by acquiring the lock_map around the fn()
1051 * call here and in del_timer_sync().
1053 lock_map_acquire(&lockdep_map);
1055 trace_timer_expire_entry(timer);
1056 fn(data);
1057 trace_timer_expire_exit(timer);
1059 lock_map_release(&lockdep_map);
1061 if (preempt_count != preempt_count()) {
1062 WARN_ONCE(1, "timer: %pF preempt leak: %08x -> %08x\n",
1063 fn, preempt_count, preempt_count());
1065 * Restore the preempt count. That gives us a decent
1066 * chance to survive and extract information. If the
1067 * callback kept a lock held, bad luck, but not worse
1068 * than the BUG() we had.
1070 preempt_count() = preempt_count;
1074 #define INDEX(N) ((base->timer_jiffies >> (TVR_BITS + (N) * TVN_BITS)) & TVN_MASK)
1077 * __run_timers - run all expired timers (if any) on this CPU.
1078 * @base: the timer vector to be processed.
1080 * This function cascades all vectors and executes all expired timer
1081 * vectors.
1083 static inline void __run_timers(struct tvec_base *base)
1085 struct timer_list *timer;
1087 spin_lock_irq(&base->lock);
1088 while (time_after_eq(jiffies, base->timer_jiffies)) {
1089 struct list_head work_list;
1090 struct list_head *head = &work_list;
1091 int index = base->timer_jiffies & TVR_MASK;
1094 * Cascade timers:
1096 if (!index &&
1097 (!cascade(base, &base->tv2, INDEX(0))) &&
1098 (!cascade(base, &base->tv3, INDEX(1))) &&
1099 !cascade(base, &base->tv4, INDEX(2)))
1100 cascade(base, &base->tv5, INDEX(3));
1101 ++base->timer_jiffies;
1102 list_replace_init(base->tv1.vec + index, &work_list);
1103 while (!list_empty(head)) {
1104 void (*fn)(unsigned long);
1105 unsigned long data;
1107 timer = list_first_entry(head, struct timer_list,entry);
1108 fn = timer->function;
1109 data = timer->data;
1111 timer_stats_account_timer(timer);
1113 set_running_timer(base, timer);
1114 detach_timer(timer, 1);
1116 spin_unlock_irq(&base->lock);
1117 call_timer_fn(timer, fn, data);
1118 spin_lock_irq(&base->lock);
1121 set_running_timer(base, NULL);
1122 spin_unlock_irq(&base->lock);
1125 #ifdef CONFIG_NO_HZ
1127 * Find out when the next timer event is due to happen. This
1128 * is used on S/390 to stop all activity when a CPU is idle.
1129 * This function needs to be called with interrupts disabled.
1131 static unsigned long __next_timer_interrupt(struct tvec_base *base)
1133 unsigned long timer_jiffies = base->timer_jiffies;
1134 unsigned long expires = timer_jiffies + NEXT_TIMER_MAX_DELTA;
1135 int index, slot, array, found = 0;
1136 struct timer_list *nte;
1137 struct tvec *varray[4];
1139 /* Look for timer events in tv1. */
1140 index = slot = timer_jiffies & TVR_MASK;
1141 do {
1142 list_for_each_entry(nte, base->tv1.vec + slot, entry) {
1143 if (tbase_get_deferrable(nte->base))
1144 continue;
1146 found = 1;
1147 expires = nte->expires;
1148 /* Look at the cascade bucket(s)? */
1149 if (!index || slot < index)
1150 goto cascade;
1151 return expires;
1153 slot = (slot + 1) & TVR_MASK;
1154 } while (slot != index);
1156 cascade:
1157 /* Calculate the next cascade event */
1158 if (index)
1159 timer_jiffies += TVR_SIZE - index;
1160 timer_jiffies >>= TVR_BITS;
1162 /* Check tv2-tv5. */
1163 varray[0] = &base->tv2;
1164 varray[1] = &base->tv3;
1165 varray[2] = &base->tv4;
1166 varray[3] = &base->tv5;
1168 for (array = 0; array < 4; array++) {
1169 struct tvec *varp = varray[array];
1171 index = slot = timer_jiffies & TVN_MASK;
1172 do {
1173 list_for_each_entry(nte, varp->vec + slot, entry) {
1174 if (tbase_get_deferrable(nte->base))
1175 continue;
1177 found = 1;
1178 if (time_before(nte->expires, expires))
1179 expires = nte->expires;
1182 * Do we still search for the first timer or are
1183 * we looking up the cascade buckets ?
1185 if (found) {
1186 /* Look at the cascade bucket(s)? */
1187 if (!index || slot < index)
1188 break;
1189 return expires;
1191 slot = (slot + 1) & TVN_MASK;
1192 } while (slot != index);
1194 if (index)
1195 timer_jiffies += TVN_SIZE - index;
1196 timer_jiffies >>= TVN_BITS;
1198 return expires;
1202 * Check, if the next hrtimer event is before the next timer wheel
1203 * event:
1205 static unsigned long cmp_next_hrtimer_event(unsigned long now,
1206 unsigned long expires)
1208 ktime_t hr_delta = hrtimer_get_next_event();
1209 struct timespec tsdelta;
1210 unsigned long delta;
1212 if (hr_delta.tv64 == KTIME_MAX)
1213 return expires;
1216 * Expired timer available, let it expire in the next tick
1218 if (hr_delta.tv64 <= 0)
1219 return now + 1;
1221 tsdelta = ktime_to_timespec(hr_delta);
1222 delta = timespec_to_jiffies(&tsdelta);
1225 * Limit the delta to the max value, which is checked in
1226 * tick_nohz_stop_sched_tick():
1228 if (delta > NEXT_TIMER_MAX_DELTA)
1229 delta = NEXT_TIMER_MAX_DELTA;
1232 * Take rounding errors in to account and make sure, that it
1233 * expires in the next tick. Otherwise we go into an endless
1234 * ping pong due to tick_nohz_stop_sched_tick() retriggering
1235 * the timer softirq
1237 if (delta < 1)
1238 delta = 1;
1239 now += delta;
1240 if (time_before(now, expires))
1241 return now;
1242 return expires;
1246 * get_next_timer_interrupt - return the jiffy of the next pending timer
1247 * @now: current time (in jiffies)
1249 unsigned long get_next_timer_interrupt(unsigned long now)
1251 struct tvec_base *base = __get_cpu_var(tvec_bases);
1252 unsigned long expires;
1254 spin_lock(&base->lock);
1255 if (time_before_eq(base->next_timer, base->timer_jiffies))
1256 base->next_timer = __next_timer_interrupt(base);
1257 expires = base->next_timer;
1258 spin_unlock(&base->lock);
1260 if (time_before_eq(expires, now))
1261 return now;
1263 return cmp_next_hrtimer_event(now, expires);
1265 #endif
1268 * Called from the timer interrupt handler to charge one tick to the current
1269 * process. user_tick is 1 if the tick is user time, 0 for system.
1271 void update_process_times(int user_tick)
1273 struct task_struct *p = current;
1274 int cpu = smp_processor_id();
1276 /* Note: this timer irq context must be accounted for as well. */
1277 account_process_tick(p, user_tick);
1278 run_local_timers();
1279 rcu_check_callbacks(cpu, user_tick);
1280 printk_tick();
1281 perf_event_do_pending();
1282 scheduler_tick();
1283 run_posix_cpu_timers(p);
1287 * This function runs timers and the timer-tq in bottom half context.
1289 static void run_timer_softirq(struct softirq_action *h)
1291 struct tvec_base *base = __get_cpu_var(tvec_bases);
1293 hrtimer_run_pending();
1295 if (time_after_eq(jiffies, base->timer_jiffies))
1296 __run_timers(base);
1300 * Called by the local, per-CPU timer interrupt on SMP.
1302 void run_local_timers(void)
1304 hrtimer_run_queues();
1305 raise_softirq(TIMER_SOFTIRQ);
1309 * The 64-bit jiffies value is not atomic - you MUST NOT read it
1310 * without sampling the sequence number in xtime_lock.
1311 * jiffies is defined in the linker script...
1314 void do_timer(unsigned long ticks)
1316 jiffies_64 += ticks;
1317 update_wall_time();
1318 calc_global_load();
1321 #ifdef __ARCH_WANT_SYS_ALARM
1324 * For backwards compatibility? This can be done in libc so Alpha
1325 * and all newer ports shouldn't need it.
1327 SYSCALL_DEFINE1(alarm, unsigned int, seconds)
1329 return alarm_setitimer(seconds);
1332 #endif
1334 #ifndef __alpha__
1337 * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this
1338 * should be moved into arch/i386 instead?
1342 * sys_getpid - return the thread group id of the current process
1344 * Note, despite the name, this returns the tgid not the pid. The tgid and
1345 * the pid are identical unless CLONE_THREAD was specified on clone() in
1346 * which case the tgid is the same in all threads of the same group.
1348 * This is SMP safe as current->tgid does not change.
1350 SYSCALL_DEFINE0(getpid)
1352 return task_tgid_vnr(current);
1356 * Accessing ->real_parent is not SMP-safe, it could
1357 * change from under us. However, we can use a stale
1358 * value of ->real_parent under rcu_read_lock(), see
1359 * release_task()->call_rcu(delayed_put_task_struct).
1361 SYSCALL_DEFINE0(getppid)
1363 int pid;
1365 rcu_read_lock();
1366 pid = task_tgid_vnr(current->real_parent);
1367 rcu_read_unlock();
1369 return pid;
1372 SYSCALL_DEFINE0(getuid)
1374 /* Only we change this so SMP safe */
1375 return current_uid();
1378 SYSCALL_DEFINE0(geteuid)
1380 /* Only we change this so SMP safe */
1381 return current_euid();
1384 SYSCALL_DEFINE0(getgid)
1386 /* Only we change this so SMP safe */
1387 return current_gid();
1390 SYSCALL_DEFINE0(getegid)
1392 /* Only we change this so SMP safe */
1393 return current_egid();
1396 #endif
1398 static void process_timeout(unsigned long __data)
1400 wake_up_process((struct task_struct *)__data);
1404 * schedule_timeout - sleep until timeout
1405 * @timeout: timeout value in jiffies
1407 * Make the current task sleep until @timeout jiffies have
1408 * elapsed. The routine will return immediately unless
1409 * the current task state has been set (see set_current_state()).
1411 * You can set the task state as follows -
1413 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1414 * pass before the routine returns. The routine will return 0
1416 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1417 * delivered to the current task. In this case the remaining time
1418 * in jiffies will be returned, or 0 if the timer expired in time
1420 * The current task state is guaranteed to be TASK_RUNNING when this
1421 * routine returns.
1423 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1424 * the CPU away without a bound on the timeout. In this case the return
1425 * value will be %MAX_SCHEDULE_TIMEOUT.
1427 * In all cases the return value is guaranteed to be non-negative.
1429 signed long __sched schedule_timeout(signed long timeout)
1431 struct timer_list timer;
1432 unsigned long expire;
1434 switch (timeout)
1436 case MAX_SCHEDULE_TIMEOUT:
1438 * These two special cases are useful to be comfortable
1439 * in the caller. Nothing more. We could take
1440 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1441 * but I' d like to return a valid offset (>=0) to allow
1442 * the caller to do everything it want with the retval.
1444 schedule();
1445 goto out;
1446 default:
1448 * Another bit of PARANOID. Note that the retval will be
1449 * 0 since no piece of kernel is supposed to do a check
1450 * for a negative retval of schedule_timeout() (since it
1451 * should never happens anyway). You just have the printk()
1452 * that will tell you if something is gone wrong and where.
1454 if (timeout < 0) {
1455 printk(KERN_ERR "schedule_timeout: wrong timeout "
1456 "value %lx\n", timeout);
1457 dump_stack();
1458 current->state = TASK_RUNNING;
1459 goto out;
1463 expire = timeout + jiffies;
1465 setup_timer_on_stack(&timer, process_timeout, (unsigned long)current);
1466 __mod_timer(&timer, expire, false, TIMER_NOT_PINNED);
1467 schedule();
1468 del_singleshot_timer_sync(&timer);
1470 /* Remove the timer from the object tracker */
1471 destroy_timer_on_stack(&timer);
1473 timeout = expire - jiffies;
1475 out:
1476 return timeout < 0 ? 0 : timeout;
1478 EXPORT_SYMBOL(schedule_timeout);
1481 * We can use __set_current_state() here because schedule_timeout() calls
1482 * schedule() unconditionally.
1484 signed long __sched schedule_timeout_interruptible(signed long timeout)
1486 __set_current_state(TASK_INTERRUPTIBLE);
1487 return schedule_timeout(timeout);
1489 EXPORT_SYMBOL(schedule_timeout_interruptible);
1491 signed long __sched schedule_timeout_killable(signed long timeout)
1493 __set_current_state(TASK_KILLABLE);
1494 return schedule_timeout(timeout);
1496 EXPORT_SYMBOL(schedule_timeout_killable);
1498 signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1500 __set_current_state(TASK_UNINTERRUPTIBLE);
1501 return schedule_timeout(timeout);
1503 EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1505 /* Thread ID - the internal kernel "pid" */
1506 SYSCALL_DEFINE0(gettid)
1508 return task_pid_vnr(current);
1512 * do_sysinfo - fill in sysinfo struct
1513 * @info: pointer to buffer to fill
1515 int do_sysinfo(struct sysinfo *info)
1517 unsigned long mem_total, sav_total;
1518 unsigned int mem_unit, bitcount;
1519 struct timespec tp;
1521 memset(info, 0, sizeof(struct sysinfo));
1523 ktime_get_ts(&tp);
1524 monotonic_to_bootbased(&tp);
1525 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
1527 get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
1529 info->procs = nr_threads;
1531 si_meminfo(info);
1532 si_swapinfo(info);
1535 * If the sum of all the available memory (i.e. ram + swap)
1536 * is less than can be stored in a 32 bit unsigned long then
1537 * we can be binary compatible with 2.2.x kernels. If not,
1538 * well, in that case 2.2.x was broken anyways...
1540 * -Erik Andersen <andersee@debian.org>
1543 mem_total = info->totalram + info->totalswap;
1544 if (mem_total < info->totalram || mem_total < info->totalswap)
1545 goto out;
1546 bitcount = 0;
1547 mem_unit = info->mem_unit;
1548 while (mem_unit > 1) {
1549 bitcount++;
1550 mem_unit >>= 1;
1551 sav_total = mem_total;
1552 mem_total <<= 1;
1553 if (mem_total < sav_total)
1554 goto out;
1558 * If mem_total did not overflow, multiply all memory values by
1559 * info->mem_unit and set it to 1. This leaves things compatible
1560 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1561 * kernels...
1564 info->mem_unit = 1;
1565 info->totalram <<= bitcount;
1566 info->freeram <<= bitcount;
1567 info->sharedram <<= bitcount;
1568 info->bufferram <<= bitcount;
1569 info->totalswap <<= bitcount;
1570 info->freeswap <<= bitcount;
1571 info->totalhigh <<= bitcount;
1572 info->freehigh <<= bitcount;
1574 out:
1575 return 0;
1578 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
1580 struct sysinfo val;
1582 do_sysinfo(&val);
1584 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
1585 return -EFAULT;
1587 return 0;
1590 static int __cpuinit init_timers_cpu(int cpu)
1592 int j;
1593 struct tvec_base *base;
1594 static char __cpuinitdata tvec_base_done[NR_CPUS];
1596 if (!tvec_base_done[cpu]) {
1597 static char boot_done;
1599 if (boot_done) {
1601 * The APs use this path later in boot
1603 base = kmalloc_node(sizeof(*base),
1604 GFP_KERNEL | __GFP_ZERO,
1605 cpu_to_node(cpu));
1606 if (!base)
1607 return -ENOMEM;
1609 /* Make sure that tvec_base is 2 byte aligned */
1610 if (tbase_get_deferrable(base)) {
1611 WARN_ON(1);
1612 kfree(base);
1613 return -ENOMEM;
1615 per_cpu(tvec_bases, cpu) = base;
1616 } else {
1618 * This is for the boot CPU - we use compile-time
1619 * static initialisation because per-cpu memory isn't
1620 * ready yet and because the memory allocators are not
1621 * initialised either.
1623 boot_done = 1;
1624 base = &boot_tvec_bases;
1626 tvec_base_done[cpu] = 1;
1627 } else {
1628 base = per_cpu(tvec_bases, cpu);
1631 spin_lock_init(&base->lock);
1633 for (j = 0; j < TVN_SIZE; j++) {
1634 INIT_LIST_HEAD(base->tv5.vec + j);
1635 INIT_LIST_HEAD(base->tv4.vec + j);
1636 INIT_LIST_HEAD(base->tv3.vec + j);
1637 INIT_LIST_HEAD(base->tv2.vec + j);
1639 for (j = 0; j < TVR_SIZE; j++)
1640 INIT_LIST_HEAD(base->tv1.vec + j);
1642 base->timer_jiffies = jiffies;
1643 base->next_timer = base->timer_jiffies;
1644 return 0;
1647 #ifdef CONFIG_HOTPLUG_CPU
1648 static void migrate_timer_list(struct tvec_base *new_base, struct list_head *head)
1650 struct timer_list *timer;
1652 while (!list_empty(head)) {
1653 timer = list_first_entry(head, struct timer_list, entry);
1654 detach_timer(timer, 0);
1655 timer_set_base(timer, new_base);
1656 if (time_before(timer->expires, new_base->next_timer) &&
1657 !tbase_get_deferrable(timer->base))
1658 new_base->next_timer = timer->expires;
1659 internal_add_timer(new_base, timer);
1663 static void __cpuinit migrate_timers(int cpu)
1665 struct tvec_base *old_base;
1666 struct tvec_base *new_base;
1667 int i;
1669 BUG_ON(cpu_online(cpu));
1670 old_base = per_cpu(tvec_bases, cpu);
1671 new_base = get_cpu_var(tvec_bases);
1673 * The caller is globally serialized and nobody else
1674 * takes two locks at once, deadlock is not possible.
1676 spin_lock_irq(&new_base->lock);
1677 spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1679 BUG_ON(old_base->running_timer);
1681 for (i = 0; i < TVR_SIZE; i++)
1682 migrate_timer_list(new_base, old_base->tv1.vec + i);
1683 for (i = 0; i < TVN_SIZE; i++) {
1684 migrate_timer_list(new_base, old_base->tv2.vec + i);
1685 migrate_timer_list(new_base, old_base->tv3.vec + i);
1686 migrate_timer_list(new_base, old_base->tv4.vec + i);
1687 migrate_timer_list(new_base, old_base->tv5.vec + i);
1690 spin_unlock(&old_base->lock);
1691 spin_unlock_irq(&new_base->lock);
1692 put_cpu_var(tvec_bases);
1694 #endif /* CONFIG_HOTPLUG_CPU */
1696 static int __cpuinit timer_cpu_notify(struct notifier_block *self,
1697 unsigned long action, void *hcpu)
1699 long cpu = (long)hcpu;
1700 int err;
1702 switch(action) {
1703 case CPU_UP_PREPARE:
1704 case CPU_UP_PREPARE_FROZEN:
1705 err = init_timers_cpu(cpu);
1706 if (err < 0)
1707 return notifier_from_errno(err);
1708 break;
1709 #ifdef CONFIG_HOTPLUG_CPU
1710 case CPU_DEAD:
1711 case CPU_DEAD_FROZEN:
1712 migrate_timers(cpu);
1713 break;
1714 #endif
1715 default:
1716 break;
1718 return NOTIFY_OK;
1721 static struct notifier_block __cpuinitdata timers_nb = {
1722 .notifier_call = timer_cpu_notify,
1726 void __init init_timers(void)
1728 int err = timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE,
1729 (void *)(long)smp_processor_id());
1731 init_timer_stats();
1733 BUG_ON(err != NOTIFY_OK);
1734 register_cpu_notifier(&timers_nb);
1735 open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
1739 * msleep - sleep safely even with waitqueue interruptions
1740 * @msecs: Time in milliseconds to sleep for
1742 void msleep(unsigned int msecs)
1744 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1746 while (timeout)
1747 timeout = schedule_timeout_uninterruptible(timeout);
1750 EXPORT_SYMBOL(msleep);
1753 * msleep_interruptible - sleep waiting for signals
1754 * @msecs: Time in milliseconds to sleep for
1756 unsigned long msleep_interruptible(unsigned int msecs)
1758 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1760 while (timeout && !signal_pending(current))
1761 timeout = schedule_timeout_interruptible(timeout);
1762 return jiffies_to_msecs(timeout);
1765 EXPORT_SYMBOL(msleep_interruptible);
1767 static int __sched do_usleep_range(unsigned long min, unsigned long max)
1769 ktime_t kmin;
1770 unsigned long delta;
1772 kmin = ktime_set(0, min * NSEC_PER_USEC);
1773 delta = (max - min) * NSEC_PER_USEC;
1774 return schedule_hrtimeout_range(&kmin, delta, HRTIMER_MODE_REL);
1778 * usleep_range - Drop in replacement for udelay where wakeup is flexible
1779 * @min: Minimum time in usecs to sleep
1780 * @max: Maximum time in usecs to sleep
1782 void usleep_range(unsigned long min, unsigned long max)
1784 __set_current_state(TASK_UNINTERRUPTIBLE);
1785 do_usleep_range(min, max);
1787 EXPORT_SYMBOL(usleep_range);