Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ieee1394...
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / kernel / timer.c
blobfd6198692b57b16e47132d1eb5183bafd98b050e
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/irq_work.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;
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 = TBASE_MAKE_DEFERRED(timer->base);
107 static inline void
108 timer_set_base(struct timer_list *timer, struct tvec_base *new_base)
110 timer->base = (struct tvec_base *)((unsigned long)(new_base) |
111 tbase_get_deferrable(timer->base));
114 static unsigned long round_jiffies_common(unsigned long j, int cpu,
115 bool force_up)
117 int rem;
118 unsigned long original = j;
121 * We don't want all cpus firing their timers at once hitting the
122 * same lock or cachelines, so we skew each extra cpu with an extra
123 * 3 jiffies. This 3 jiffies came originally from the mm/ code which
124 * already did this.
125 * The skew is done by adding 3*cpunr, then round, then subtract this
126 * extra offset again.
128 j += cpu * 3;
130 rem = j % HZ;
133 * If the target jiffie is just after a whole second (which can happen
134 * due to delays of the timer irq, long irq off times etc etc) then
135 * we should round down to the whole second, not up. Use 1/4th second
136 * as cutoff for this rounding as an extreme upper bound for this.
137 * But never round down if @force_up is set.
139 if (rem < HZ/4 && !force_up) /* round down */
140 j = j - rem;
141 else /* round up */
142 j = j - rem + HZ;
144 /* now that we have rounded, subtract the extra skew again */
145 j -= cpu * 3;
147 if (j <= jiffies) /* rounding ate our timeout entirely; */
148 return original;
149 return j;
153 * __round_jiffies - function to round jiffies to a full second
154 * @j: the time in (absolute) jiffies that should be rounded
155 * @cpu: the processor number on which the timeout will happen
157 * __round_jiffies() rounds an absolute time in the future (in jiffies)
158 * up or down to (approximately) full seconds. This is useful for timers
159 * for which the exact time they fire does not matter too much, as long as
160 * they fire approximately every X seconds.
162 * By rounding these timers to whole seconds, all such timers will fire
163 * at the same time, rather than at various times spread out. The goal
164 * of this is to have the CPU wake up less, which saves power.
166 * The exact rounding is skewed for each processor to avoid all
167 * processors firing at the exact same time, which could lead
168 * to lock contention or spurious cache line bouncing.
170 * The return value is the rounded version of the @j parameter.
172 unsigned long __round_jiffies(unsigned long j, int cpu)
174 return round_jiffies_common(j, cpu, false);
176 EXPORT_SYMBOL_GPL(__round_jiffies);
179 * __round_jiffies_relative - function to round jiffies to a full second
180 * @j: the time in (relative) jiffies that should be rounded
181 * @cpu: the processor number on which the timeout will happen
183 * __round_jiffies_relative() rounds a time delta in the future (in jiffies)
184 * up or down to (approximately) full seconds. This is useful for timers
185 * for which the exact time they fire does not matter too much, as long as
186 * they fire approximately every X seconds.
188 * By rounding these timers to whole seconds, all such timers will fire
189 * at the same time, rather than at various times spread out. The goal
190 * of this is to have the CPU wake up less, which saves power.
192 * The exact rounding is skewed for each processor to avoid all
193 * processors firing at the exact same time, which could lead
194 * to lock contention or spurious cache line bouncing.
196 * The return value is the rounded version of the @j parameter.
198 unsigned long __round_jiffies_relative(unsigned long j, int cpu)
200 unsigned long j0 = jiffies;
202 /* Use j0 because jiffies might change while we run */
203 return round_jiffies_common(j + j0, cpu, false) - j0;
205 EXPORT_SYMBOL_GPL(__round_jiffies_relative);
208 * round_jiffies - function to round jiffies to a full second
209 * @j: the time in (absolute) jiffies that should be rounded
211 * round_jiffies() rounds an absolute time in the future (in jiffies)
212 * up or down to (approximately) full seconds. This is useful for timers
213 * for which the exact time they fire does not matter too much, as long as
214 * they fire approximately every X seconds.
216 * By rounding these timers to whole seconds, all such timers will fire
217 * at the same time, rather than at various times spread out. The goal
218 * of this is to have the CPU wake up less, which saves power.
220 * The return value is the rounded version of the @j parameter.
222 unsigned long round_jiffies(unsigned long j)
224 return round_jiffies_common(j, raw_smp_processor_id(), false);
226 EXPORT_SYMBOL_GPL(round_jiffies);
229 * round_jiffies_relative - function to round jiffies to a full second
230 * @j: the time in (relative) jiffies that should be rounded
232 * round_jiffies_relative() rounds a time delta in the future (in jiffies)
233 * up or down to (approximately) full seconds. This is useful for timers
234 * for which the exact time they fire does not matter too much, as long as
235 * they fire approximately every X seconds.
237 * By rounding these timers to whole seconds, all such timers will fire
238 * at the same time, rather than at various times spread out. The goal
239 * of this is to have the CPU wake up less, which saves power.
241 * The return value is the rounded version of the @j parameter.
243 unsigned long round_jiffies_relative(unsigned long j)
245 return __round_jiffies_relative(j, raw_smp_processor_id());
247 EXPORT_SYMBOL_GPL(round_jiffies_relative);
250 * __round_jiffies_up - function to round jiffies up to a full second
251 * @j: the time in (absolute) jiffies that should be rounded
252 * @cpu: the processor number on which the timeout will happen
254 * This is the same as __round_jiffies() except that it will never
255 * round down. This is useful for timeouts for which the exact time
256 * of firing does not matter too much, as long as they don't fire too
257 * early.
259 unsigned long __round_jiffies_up(unsigned long j, int cpu)
261 return round_jiffies_common(j, cpu, true);
263 EXPORT_SYMBOL_GPL(__round_jiffies_up);
266 * __round_jiffies_up_relative - function to round jiffies up to a full second
267 * @j: the time in (relative) jiffies that should be rounded
268 * @cpu: the processor number on which the timeout will happen
270 * This is the same as __round_jiffies_relative() except that it will never
271 * round down. This is useful for timeouts for which the exact time
272 * of firing does not matter too much, as long as they don't fire too
273 * early.
275 unsigned long __round_jiffies_up_relative(unsigned long j, int cpu)
277 unsigned long j0 = jiffies;
279 /* Use j0 because jiffies might change while we run */
280 return round_jiffies_common(j + j0, cpu, true) - j0;
282 EXPORT_SYMBOL_GPL(__round_jiffies_up_relative);
285 * round_jiffies_up - function to round jiffies up to a full second
286 * @j: the time in (absolute) jiffies that should be rounded
288 * This is the same as round_jiffies() except that it will never
289 * round down. This is useful for timeouts for which the exact time
290 * of firing does not matter too much, as long as they don't fire too
291 * early.
293 unsigned long round_jiffies_up(unsigned long j)
295 return round_jiffies_common(j, raw_smp_processor_id(), true);
297 EXPORT_SYMBOL_GPL(round_jiffies_up);
300 * round_jiffies_up_relative - function to round jiffies up to a full second
301 * @j: the time in (relative) jiffies that should be rounded
303 * This is the same as round_jiffies_relative() except that it will never
304 * round down. This is useful for timeouts for which the exact time
305 * of firing does not matter too much, as long as they don't fire too
306 * early.
308 unsigned long round_jiffies_up_relative(unsigned long j)
310 return __round_jiffies_up_relative(j, raw_smp_processor_id());
312 EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
315 * set_timer_slack - set the allowed slack for a timer
316 * @timer: the timer to be modified
317 * @slack_hz: the amount of time (in jiffies) allowed for rounding
319 * Set the amount of time, in jiffies, that a certain timer has
320 * in terms of slack. By setting this value, the timer subsystem
321 * will schedule the actual timer somewhere between
322 * the time mod_timer() asks for, and that time plus the slack.
324 * By setting the slack to -1, a percentage of the delay is used
325 * instead.
327 void set_timer_slack(struct timer_list *timer, int slack_hz)
329 timer->slack = slack_hz;
331 EXPORT_SYMBOL_GPL(set_timer_slack);
333 static void internal_add_timer(struct tvec_base *base, struct timer_list *timer)
335 unsigned long expires = timer->expires;
336 unsigned long idx = expires - base->timer_jiffies;
337 struct list_head *vec;
339 if (idx < TVR_SIZE) {
340 int i = expires & TVR_MASK;
341 vec = base->tv1.vec + i;
342 } else if (idx < 1 << (TVR_BITS + TVN_BITS)) {
343 int i = (expires >> TVR_BITS) & TVN_MASK;
344 vec = base->tv2.vec + i;
345 } else if (idx < 1 << (TVR_BITS + 2 * TVN_BITS)) {
346 int i = (expires >> (TVR_BITS + TVN_BITS)) & TVN_MASK;
347 vec = base->tv3.vec + i;
348 } else if (idx < 1 << (TVR_BITS + 3 * TVN_BITS)) {
349 int i = (expires >> (TVR_BITS + 2 * TVN_BITS)) & TVN_MASK;
350 vec = base->tv4.vec + i;
351 } else if ((signed long) idx < 0) {
353 * Can happen if you add a timer with expires == jiffies,
354 * or you set a timer to go off in the past
356 vec = base->tv1.vec + (base->timer_jiffies & TVR_MASK);
357 } else {
358 int i;
359 /* If the timeout is larger than 0xffffffff on 64-bit
360 * architectures then we use the maximum timeout:
362 if (idx > 0xffffffffUL) {
363 idx = 0xffffffffUL;
364 expires = idx + base->timer_jiffies;
366 i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK;
367 vec = base->tv5.vec + i;
370 * Timers are FIFO:
372 list_add_tail(&timer->entry, vec);
375 #ifdef CONFIG_TIMER_STATS
376 void __timer_stats_timer_set_start_info(struct timer_list *timer, void *addr)
378 if (timer->start_site)
379 return;
381 timer->start_site = addr;
382 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
383 timer->start_pid = current->pid;
386 static void timer_stats_account_timer(struct timer_list *timer)
388 unsigned int flag = 0;
390 if (likely(!timer->start_site))
391 return;
392 if (unlikely(tbase_get_deferrable(timer->base)))
393 flag |= TIMER_STATS_FLAG_DEFERRABLE;
395 timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
396 timer->function, timer->start_comm, flag);
399 #else
400 static void timer_stats_account_timer(struct timer_list *timer) {}
401 #endif
403 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
405 static struct debug_obj_descr timer_debug_descr;
407 static void *timer_debug_hint(void *addr)
409 return ((struct timer_list *) addr)->function;
413 * fixup_init is called when:
414 * - an active object is initialized
416 static int timer_fixup_init(void *addr, enum debug_obj_state state)
418 struct timer_list *timer = addr;
420 switch (state) {
421 case ODEBUG_STATE_ACTIVE:
422 del_timer_sync(timer);
423 debug_object_init(timer, &timer_debug_descr);
424 return 1;
425 default:
426 return 0;
431 * fixup_activate is called when:
432 * - an active object is activated
433 * - an unknown object is activated (might be a statically initialized object)
435 static int timer_fixup_activate(void *addr, enum debug_obj_state state)
437 struct timer_list *timer = addr;
439 switch (state) {
441 case ODEBUG_STATE_NOTAVAILABLE:
443 * This is not really a fixup. The timer was
444 * statically initialized. We just make sure that it
445 * is tracked in the object tracker.
447 if (timer->entry.next == NULL &&
448 timer->entry.prev == TIMER_ENTRY_STATIC) {
449 debug_object_init(timer, &timer_debug_descr);
450 debug_object_activate(timer, &timer_debug_descr);
451 return 0;
452 } else {
453 WARN_ON_ONCE(1);
455 return 0;
457 case ODEBUG_STATE_ACTIVE:
458 WARN_ON(1);
460 default:
461 return 0;
466 * fixup_free is called when:
467 * - an active object is freed
469 static int timer_fixup_free(void *addr, enum debug_obj_state state)
471 struct timer_list *timer = addr;
473 switch (state) {
474 case ODEBUG_STATE_ACTIVE:
475 del_timer_sync(timer);
476 debug_object_free(timer, &timer_debug_descr);
477 return 1;
478 default:
479 return 0;
483 static struct debug_obj_descr timer_debug_descr = {
484 .name = "timer_list",
485 .debug_hint = timer_debug_hint,
486 .fixup_init = timer_fixup_init,
487 .fixup_activate = timer_fixup_activate,
488 .fixup_free = timer_fixup_free,
491 static inline void debug_timer_init(struct timer_list *timer)
493 debug_object_init(timer, &timer_debug_descr);
496 static inline void debug_timer_activate(struct timer_list *timer)
498 debug_object_activate(timer, &timer_debug_descr);
501 static inline void debug_timer_deactivate(struct timer_list *timer)
503 debug_object_deactivate(timer, &timer_debug_descr);
506 static inline void debug_timer_free(struct timer_list *timer)
508 debug_object_free(timer, &timer_debug_descr);
511 static void __init_timer(struct timer_list *timer,
512 const char *name,
513 struct lock_class_key *key);
515 void init_timer_on_stack_key(struct timer_list *timer,
516 const char *name,
517 struct lock_class_key *key)
519 debug_object_init_on_stack(timer, &timer_debug_descr);
520 __init_timer(timer, name, key);
522 EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
524 void destroy_timer_on_stack(struct timer_list *timer)
526 debug_object_free(timer, &timer_debug_descr);
528 EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
530 #else
531 static inline void debug_timer_init(struct timer_list *timer) { }
532 static inline void debug_timer_activate(struct timer_list *timer) { }
533 static inline void debug_timer_deactivate(struct timer_list *timer) { }
534 #endif
536 static inline void debug_init(struct timer_list *timer)
538 debug_timer_init(timer);
539 trace_timer_init(timer);
542 static inline void
543 debug_activate(struct timer_list *timer, unsigned long expires)
545 debug_timer_activate(timer);
546 trace_timer_start(timer, expires);
549 static inline void debug_deactivate(struct timer_list *timer)
551 debug_timer_deactivate(timer);
552 trace_timer_cancel(timer);
555 static void __init_timer(struct timer_list *timer,
556 const char *name,
557 struct lock_class_key *key)
559 timer->entry.next = NULL;
560 timer->base = __raw_get_cpu_var(tvec_bases);
561 timer->slack = -1;
562 #ifdef CONFIG_TIMER_STATS
563 timer->start_site = NULL;
564 timer->start_pid = -1;
565 memset(timer->start_comm, 0, TASK_COMM_LEN);
566 #endif
567 lockdep_init_map(&timer->lockdep_map, name, key, 0);
570 void setup_deferrable_timer_on_stack_key(struct timer_list *timer,
571 const char *name,
572 struct lock_class_key *key,
573 void (*function)(unsigned long),
574 unsigned long data)
576 timer->function = function;
577 timer->data = data;
578 init_timer_on_stack_key(timer, name, key);
579 timer_set_deferrable(timer);
581 EXPORT_SYMBOL_GPL(setup_deferrable_timer_on_stack_key);
584 * init_timer_key - initialize a timer
585 * @timer: the timer to be initialized
586 * @name: name of the timer
587 * @key: lockdep class key of the fake lock used for tracking timer
588 * sync lock dependencies
590 * init_timer_key() must be done to a timer prior calling *any* of the
591 * other timer functions.
593 void init_timer_key(struct timer_list *timer,
594 const char *name,
595 struct lock_class_key *key)
597 debug_init(timer);
598 __init_timer(timer, name, key);
600 EXPORT_SYMBOL(init_timer_key);
602 void init_timer_deferrable_key(struct timer_list *timer,
603 const char *name,
604 struct lock_class_key *key)
606 init_timer_key(timer, name, key);
607 timer_set_deferrable(timer);
609 EXPORT_SYMBOL(init_timer_deferrable_key);
611 static inline void detach_timer(struct timer_list *timer,
612 int clear_pending)
614 struct list_head *entry = &timer->entry;
616 debug_deactivate(timer);
618 __list_del(entry->prev, entry->next);
619 if (clear_pending)
620 entry->next = NULL;
621 entry->prev = LIST_POISON2;
625 * We are using hashed locking: holding per_cpu(tvec_bases).lock
626 * means that all timers which are tied to this base via timer->base are
627 * locked, and the base itself is locked too.
629 * So __run_timers/migrate_timers can safely modify all timers which could
630 * be found on ->tvX lists.
632 * When the timer's base is locked, and the timer removed from list, it is
633 * possible to set timer->base = NULL and drop the lock: the timer remains
634 * locked.
636 static struct tvec_base *lock_timer_base(struct timer_list *timer,
637 unsigned long *flags)
638 __acquires(timer->base->lock)
640 struct tvec_base *base;
642 for (;;) {
643 struct tvec_base *prelock_base = timer->base;
644 base = tbase_get_base(prelock_base);
645 if (likely(base != NULL)) {
646 spin_lock_irqsave(&base->lock, *flags);
647 if (likely(prelock_base == timer->base))
648 return base;
649 /* The timer has migrated to another CPU */
650 spin_unlock_irqrestore(&base->lock, *flags);
652 cpu_relax();
656 static inline int
657 __mod_timer(struct timer_list *timer, unsigned long expires,
658 bool pending_only, int pinned)
660 struct tvec_base *base, *new_base;
661 unsigned long flags;
662 int ret = 0 , cpu;
664 timer_stats_timer_set_start_info(timer);
665 BUG_ON(!timer->function);
667 base = lock_timer_base(timer, &flags);
669 if (timer_pending(timer)) {
670 detach_timer(timer, 0);
671 if (timer->expires == base->next_timer &&
672 !tbase_get_deferrable(timer->base))
673 base->next_timer = base->timer_jiffies;
674 ret = 1;
675 } else {
676 if (pending_only)
677 goto out_unlock;
680 debug_activate(timer, expires);
682 cpu = smp_processor_id();
684 #if defined(CONFIG_NO_HZ) && defined(CONFIG_SMP)
685 if (!pinned && get_sysctl_timer_migration() && idle_cpu(cpu))
686 cpu = get_nohz_timer_target();
687 #endif
688 new_base = per_cpu(tvec_bases, cpu);
690 if (base != new_base) {
692 * We are trying to schedule the timer on the local CPU.
693 * However we can't change timer's base while it is running,
694 * otherwise del_timer_sync() can't detect that the timer's
695 * handler yet has not finished. This also guarantees that
696 * the timer is serialized wrt itself.
698 if (likely(base->running_timer != timer)) {
699 /* See the comment in lock_timer_base() */
700 timer_set_base(timer, NULL);
701 spin_unlock(&base->lock);
702 base = new_base;
703 spin_lock(&base->lock);
704 timer_set_base(timer, base);
708 timer->expires = expires;
709 if (time_before(timer->expires, base->next_timer) &&
710 !tbase_get_deferrable(timer->base))
711 base->next_timer = timer->expires;
712 internal_add_timer(base, timer);
714 out_unlock:
715 spin_unlock_irqrestore(&base->lock, flags);
717 return ret;
721 * mod_timer_pending - modify a pending timer's timeout
722 * @timer: the pending timer to be modified
723 * @expires: new timeout in jiffies
725 * mod_timer_pending() is the same for pending timers as mod_timer(),
726 * but will not re-activate and modify already deleted timers.
728 * It is useful for unserialized use of timers.
730 int mod_timer_pending(struct timer_list *timer, unsigned long expires)
732 return __mod_timer(timer, expires, true, TIMER_NOT_PINNED);
734 EXPORT_SYMBOL(mod_timer_pending);
737 * Decide where to put the timer while taking the slack into account
739 * Algorithm:
740 * 1) calculate the maximum (absolute) time
741 * 2) calculate the highest bit where the expires and new max are different
742 * 3) use this bit to make a mask
743 * 4) use the bitmask to round down the maximum time, so that all last
744 * bits are zeros
746 static inline
747 unsigned long apply_slack(struct timer_list *timer, unsigned long expires)
749 unsigned long expires_limit, mask;
750 int bit;
752 expires_limit = expires;
754 if (timer->slack >= 0) {
755 expires_limit = expires + timer->slack;
756 } else {
757 unsigned long now = jiffies;
759 /* No slack, if already expired else auto slack 0.4% */
760 if (time_after(expires, now))
761 expires_limit = expires + (expires - now)/256;
763 mask = expires ^ expires_limit;
764 if (mask == 0)
765 return expires;
767 bit = find_last_bit(&mask, BITS_PER_LONG);
769 mask = (1 << bit) - 1;
771 expires_limit = expires_limit & ~(mask);
773 return expires_limit;
777 * mod_timer - modify a timer's timeout
778 * @timer: the timer to be modified
779 * @expires: new timeout in jiffies
781 * mod_timer() is a more efficient way to update the expire field of an
782 * active timer (if the timer is inactive it will be activated)
784 * mod_timer(timer, expires) is equivalent to:
786 * del_timer(timer); timer->expires = expires; add_timer(timer);
788 * Note that if there are multiple unserialized concurrent users of the
789 * same timer, then mod_timer() is the only safe way to modify the timeout,
790 * since add_timer() cannot modify an already running timer.
792 * The function returns whether it has modified a pending timer or not.
793 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
794 * active timer returns 1.)
796 int mod_timer(struct timer_list *timer, unsigned long expires)
799 * This is a common optimization triggered by the
800 * networking code - if the timer is re-modified
801 * to be the same thing then just return:
803 if (timer_pending(timer) && timer->expires == expires)
804 return 1;
806 expires = apply_slack(timer, expires);
808 return __mod_timer(timer, expires, false, TIMER_NOT_PINNED);
810 EXPORT_SYMBOL(mod_timer);
813 * mod_timer_pinned - modify a timer's timeout
814 * @timer: the timer to be modified
815 * @expires: new timeout in jiffies
817 * mod_timer_pinned() is a way to update the expire field of an
818 * active timer (if the timer is inactive it will be activated)
819 * and not allow the timer to be migrated to a different CPU.
821 * mod_timer_pinned(timer, expires) is equivalent to:
823 * del_timer(timer); timer->expires = expires; add_timer(timer);
825 int mod_timer_pinned(struct timer_list *timer, unsigned long expires)
827 if (timer->expires == expires && timer_pending(timer))
828 return 1;
830 return __mod_timer(timer, expires, false, TIMER_PINNED);
832 EXPORT_SYMBOL(mod_timer_pinned);
835 * add_timer - start a timer
836 * @timer: the timer to be added
838 * The kernel will do a ->function(->data) callback from the
839 * timer interrupt at the ->expires point in the future. The
840 * current time is 'jiffies'.
842 * The timer's ->expires, ->function (and if the handler uses it, ->data)
843 * fields must be set prior calling this function.
845 * Timers with an ->expires field in the past will be executed in the next
846 * timer tick.
848 void add_timer(struct timer_list *timer)
850 BUG_ON(timer_pending(timer));
851 mod_timer(timer, timer->expires);
853 EXPORT_SYMBOL(add_timer);
856 * add_timer_on - start a timer on a particular CPU
857 * @timer: the timer to be added
858 * @cpu: the CPU to start it on
860 * This is not very scalable on SMP. Double adds are not possible.
862 void add_timer_on(struct timer_list *timer, int cpu)
864 struct tvec_base *base = per_cpu(tvec_bases, cpu);
865 unsigned long flags;
867 timer_stats_timer_set_start_info(timer);
868 BUG_ON(timer_pending(timer) || !timer->function);
869 spin_lock_irqsave(&base->lock, flags);
870 timer_set_base(timer, base);
871 debug_activate(timer, timer->expires);
872 if (time_before(timer->expires, base->next_timer) &&
873 !tbase_get_deferrable(timer->base))
874 base->next_timer = timer->expires;
875 internal_add_timer(base, timer);
877 * Check whether the other CPU is idle and needs to be
878 * triggered to reevaluate the timer wheel when nohz is
879 * active. We are protected against the other CPU fiddling
880 * with the timer by holding the timer base lock. This also
881 * makes sure that a CPU on the way to idle can not evaluate
882 * the timer wheel.
884 wake_up_idle_cpu(cpu);
885 spin_unlock_irqrestore(&base->lock, flags);
887 EXPORT_SYMBOL_GPL(add_timer_on);
890 * del_timer - deactive a timer.
891 * @timer: the timer to be deactivated
893 * del_timer() deactivates a timer - this works on both active and inactive
894 * timers.
896 * The function returns whether it has deactivated a pending timer or not.
897 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
898 * active timer returns 1.)
900 int del_timer(struct timer_list *timer)
902 struct tvec_base *base;
903 unsigned long flags;
904 int ret = 0;
906 timer_stats_timer_clear_start_info(timer);
907 if (timer_pending(timer)) {
908 base = lock_timer_base(timer, &flags);
909 if (timer_pending(timer)) {
910 detach_timer(timer, 1);
911 if (timer->expires == base->next_timer &&
912 !tbase_get_deferrable(timer->base))
913 base->next_timer = base->timer_jiffies;
914 ret = 1;
916 spin_unlock_irqrestore(&base->lock, flags);
919 return ret;
921 EXPORT_SYMBOL(del_timer);
924 * try_to_del_timer_sync - Try to deactivate a timer
925 * @timer: timer do del
927 * This function tries to deactivate a timer. Upon successful (ret >= 0)
928 * exit the timer is not queued and the handler is not running on any CPU.
930 int try_to_del_timer_sync(struct timer_list *timer)
932 struct tvec_base *base;
933 unsigned long flags;
934 int ret = -1;
936 base = lock_timer_base(timer, &flags);
938 if (base->running_timer == timer)
939 goto out;
941 timer_stats_timer_clear_start_info(timer);
942 ret = 0;
943 if (timer_pending(timer)) {
944 detach_timer(timer, 1);
945 if (timer->expires == base->next_timer &&
946 !tbase_get_deferrable(timer->base))
947 base->next_timer = base->timer_jiffies;
948 ret = 1;
950 out:
951 spin_unlock_irqrestore(&base->lock, flags);
953 return ret;
955 EXPORT_SYMBOL(try_to_del_timer_sync);
957 #ifdef CONFIG_SMP
959 * del_timer_sync - deactivate a timer and wait for the handler to finish.
960 * @timer: the timer to be deactivated
962 * This function only differs from del_timer() on SMP: besides deactivating
963 * the timer it also makes sure the handler has finished executing on other
964 * CPUs.
966 * Synchronization rules: Callers must prevent restarting of the timer,
967 * otherwise this function is meaningless. It must not be called from
968 * interrupt contexts. The caller must not hold locks which would prevent
969 * completion of the timer's handler. The timer's handler must not call
970 * add_timer_on(). Upon exit the timer is not queued and the handler is
971 * not running on any CPU.
973 * Note: You must not hold locks that are held in interrupt context
974 * while calling this function. Even if the lock has nothing to do
975 * with the timer in question. Here's why:
977 * CPU0 CPU1
978 * ---- ----
979 * <SOFTIRQ>
980 * call_timer_fn();
981 * base->running_timer = mytimer;
982 * spin_lock_irq(somelock);
983 * <IRQ>
984 * spin_lock(somelock);
985 * del_timer_sync(mytimer);
986 * while (base->running_timer == mytimer);
988 * Now del_timer_sync() will never return and never release somelock.
989 * The interrupt on the other CPU is waiting to grab somelock but
990 * it has interrupted the softirq that CPU0 is waiting to finish.
992 * The function returns whether it has deactivated a pending timer or not.
994 int del_timer_sync(struct timer_list *timer)
996 #ifdef CONFIG_LOCKDEP
997 unsigned long flags;
1000 * If lockdep gives a backtrace here, please reference
1001 * the synchronization rules above.
1003 local_irq_save(flags);
1004 lock_map_acquire(&timer->lockdep_map);
1005 lock_map_release(&timer->lockdep_map);
1006 local_irq_restore(flags);
1007 #endif
1009 * don't use it in hardirq context, because it
1010 * could lead to deadlock.
1012 WARN_ON(in_irq());
1013 for (;;) {
1014 int ret = try_to_del_timer_sync(timer);
1015 if (ret >= 0)
1016 return ret;
1017 cpu_relax();
1020 EXPORT_SYMBOL(del_timer_sync);
1021 #endif
1023 static int cascade(struct tvec_base *base, struct tvec *tv, int index)
1025 /* cascade all the timers from tv up one level */
1026 struct timer_list *timer, *tmp;
1027 struct list_head tv_list;
1029 list_replace_init(tv->vec + index, &tv_list);
1032 * We are removing _all_ timers from the list, so we
1033 * don't have to detach them individually.
1035 list_for_each_entry_safe(timer, tmp, &tv_list, entry) {
1036 BUG_ON(tbase_get_base(timer->base) != base);
1037 internal_add_timer(base, timer);
1040 return index;
1043 static void call_timer_fn(struct timer_list *timer, void (*fn)(unsigned long),
1044 unsigned long data)
1046 int preempt_count = preempt_count();
1048 #ifdef CONFIG_LOCKDEP
1050 * It is permissible to free the timer from inside the
1051 * function that is called from it, this we need to take into
1052 * account for lockdep too. To avoid bogus "held lock freed"
1053 * warnings as well as problems when looking into
1054 * timer->lockdep_map, make a copy and use that here.
1056 struct lockdep_map lockdep_map = timer->lockdep_map;
1057 #endif
1059 * Couple the lock chain with the lock chain at
1060 * del_timer_sync() by acquiring the lock_map around the fn()
1061 * call here and in del_timer_sync().
1063 lock_map_acquire(&lockdep_map);
1065 trace_timer_expire_entry(timer);
1066 fn(data);
1067 trace_timer_expire_exit(timer);
1069 lock_map_release(&lockdep_map);
1071 if (preempt_count != preempt_count()) {
1072 WARN_ONCE(1, "timer: %pF preempt leak: %08x -> %08x\n",
1073 fn, preempt_count, preempt_count());
1075 * Restore the preempt count. That gives us a decent
1076 * chance to survive and extract information. If the
1077 * callback kept a lock held, bad luck, but not worse
1078 * than the BUG() we had.
1080 preempt_count() = preempt_count;
1084 #define INDEX(N) ((base->timer_jiffies >> (TVR_BITS + (N) * TVN_BITS)) & TVN_MASK)
1087 * __run_timers - run all expired timers (if any) on this CPU.
1088 * @base: the timer vector to be processed.
1090 * This function cascades all vectors and executes all expired timer
1091 * vectors.
1093 static inline void __run_timers(struct tvec_base *base)
1095 struct timer_list *timer;
1097 spin_lock_irq(&base->lock);
1098 while (time_after_eq(jiffies, base->timer_jiffies)) {
1099 struct list_head work_list;
1100 struct list_head *head = &work_list;
1101 int index = base->timer_jiffies & TVR_MASK;
1104 * Cascade timers:
1106 if (!index &&
1107 (!cascade(base, &base->tv2, INDEX(0))) &&
1108 (!cascade(base, &base->tv3, INDEX(1))) &&
1109 !cascade(base, &base->tv4, INDEX(2)))
1110 cascade(base, &base->tv5, INDEX(3));
1111 ++base->timer_jiffies;
1112 list_replace_init(base->tv1.vec + index, &work_list);
1113 while (!list_empty(head)) {
1114 void (*fn)(unsigned long);
1115 unsigned long data;
1117 timer = list_first_entry(head, struct timer_list,entry);
1118 fn = timer->function;
1119 data = timer->data;
1121 timer_stats_account_timer(timer);
1123 base->running_timer = timer;
1124 detach_timer(timer, 1);
1126 spin_unlock_irq(&base->lock);
1127 call_timer_fn(timer, fn, data);
1128 spin_lock_irq(&base->lock);
1131 base->running_timer = NULL;
1132 spin_unlock_irq(&base->lock);
1135 #ifdef CONFIG_NO_HZ
1137 * Find out when the next timer event is due to happen. This
1138 * is used on S/390 to stop all activity when a CPU is idle.
1139 * This function needs to be called with interrupts disabled.
1141 static unsigned long __next_timer_interrupt(struct tvec_base *base)
1143 unsigned long timer_jiffies = base->timer_jiffies;
1144 unsigned long expires = timer_jiffies + NEXT_TIMER_MAX_DELTA;
1145 int index, slot, array, found = 0;
1146 struct timer_list *nte;
1147 struct tvec *varray[4];
1149 /* Look for timer events in tv1. */
1150 index = slot = timer_jiffies & TVR_MASK;
1151 do {
1152 list_for_each_entry(nte, base->tv1.vec + slot, entry) {
1153 if (tbase_get_deferrable(nte->base))
1154 continue;
1156 found = 1;
1157 expires = nte->expires;
1158 /* Look at the cascade bucket(s)? */
1159 if (!index || slot < index)
1160 goto cascade;
1161 return expires;
1163 slot = (slot + 1) & TVR_MASK;
1164 } while (slot != index);
1166 cascade:
1167 /* Calculate the next cascade event */
1168 if (index)
1169 timer_jiffies += TVR_SIZE - index;
1170 timer_jiffies >>= TVR_BITS;
1172 /* Check tv2-tv5. */
1173 varray[0] = &base->tv2;
1174 varray[1] = &base->tv3;
1175 varray[2] = &base->tv4;
1176 varray[3] = &base->tv5;
1178 for (array = 0; array < 4; array++) {
1179 struct tvec *varp = varray[array];
1181 index = slot = timer_jiffies & TVN_MASK;
1182 do {
1183 list_for_each_entry(nte, varp->vec + slot, entry) {
1184 if (tbase_get_deferrable(nte->base))
1185 continue;
1187 found = 1;
1188 if (time_before(nte->expires, expires))
1189 expires = nte->expires;
1192 * Do we still search for the first timer or are
1193 * we looking up the cascade buckets ?
1195 if (found) {
1196 /* Look at the cascade bucket(s)? */
1197 if (!index || slot < index)
1198 break;
1199 return expires;
1201 slot = (slot + 1) & TVN_MASK;
1202 } while (slot != index);
1204 if (index)
1205 timer_jiffies += TVN_SIZE - index;
1206 timer_jiffies >>= TVN_BITS;
1208 return expires;
1212 * Check, if the next hrtimer event is before the next timer wheel
1213 * event:
1215 static unsigned long cmp_next_hrtimer_event(unsigned long now,
1216 unsigned long expires)
1218 ktime_t hr_delta = hrtimer_get_next_event();
1219 struct timespec tsdelta;
1220 unsigned long delta;
1222 if (hr_delta.tv64 == KTIME_MAX)
1223 return expires;
1226 * Expired timer available, let it expire in the next tick
1228 if (hr_delta.tv64 <= 0)
1229 return now + 1;
1231 tsdelta = ktime_to_timespec(hr_delta);
1232 delta = timespec_to_jiffies(&tsdelta);
1235 * Limit the delta to the max value, which is checked in
1236 * tick_nohz_stop_sched_tick():
1238 if (delta > NEXT_TIMER_MAX_DELTA)
1239 delta = NEXT_TIMER_MAX_DELTA;
1242 * Take rounding errors in to account and make sure, that it
1243 * expires in the next tick. Otherwise we go into an endless
1244 * ping pong due to tick_nohz_stop_sched_tick() retriggering
1245 * the timer softirq
1247 if (delta < 1)
1248 delta = 1;
1249 now += delta;
1250 if (time_before(now, expires))
1251 return now;
1252 return expires;
1256 * get_next_timer_interrupt - return the jiffy of the next pending timer
1257 * @now: current time (in jiffies)
1259 unsigned long get_next_timer_interrupt(unsigned long now)
1261 struct tvec_base *base = __this_cpu_read(tvec_bases);
1262 unsigned long expires;
1265 * Pretend that there is no timer pending if the cpu is offline.
1266 * Possible pending timers will be migrated later to an active cpu.
1268 if (cpu_is_offline(smp_processor_id()))
1269 return now + NEXT_TIMER_MAX_DELTA;
1270 spin_lock(&base->lock);
1271 if (time_before_eq(base->next_timer, base->timer_jiffies))
1272 base->next_timer = __next_timer_interrupt(base);
1273 expires = base->next_timer;
1274 spin_unlock(&base->lock);
1276 if (time_before_eq(expires, now))
1277 return now;
1279 return cmp_next_hrtimer_event(now, expires);
1281 #endif
1284 * Called from the timer interrupt handler to charge one tick to the current
1285 * process. user_tick is 1 if the tick is user time, 0 for system.
1287 void update_process_times(int user_tick)
1289 struct task_struct *p = current;
1290 int cpu = smp_processor_id();
1292 /* Note: this timer irq context must be accounted for as well. */
1293 account_process_tick(p, user_tick);
1294 run_local_timers();
1295 rcu_check_callbacks(cpu, user_tick);
1296 printk_tick();
1297 #ifdef CONFIG_IRQ_WORK
1298 if (in_irq())
1299 irq_work_run();
1300 #endif
1301 scheduler_tick();
1302 run_posix_cpu_timers(p);
1306 * This function runs timers and the timer-tq in bottom half context.
1308 static void run_timer_softirq(struct softirq_action *h)
1310 struct tvec_base *base = __this_cpu_read(tvec_bases);
1312 hrtimer_run_pending();
1314 if (time_after_eq(jiffies, base->timer_jiffies))
1315 __run_timers(base);
1319 * Called by the local, per-CPU timer interrupt on SMP.
1321 void run_local_timers(void)
1323 hrtimer_run_queues();
1324 raise_softirq(TIMER_SOFTIRQ);
1327 #ifdef __ARCH_WANT_SYS_ALARM
1330 * For backwards compatibility? This can be done in libc so Alpha
1331 * and all newer ports shouldn't need it.
1333 SYSCALL_DEFINE1(alarm, unsigned int, seconds)
1335 return alarm_setitimer(seconds);
1338 #endif
1340 #ifndef __alpha__
1343 * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this
1344 * should be moved into arch/i386 instead?
1348 * sys_getpid - return the thread group id of the current process
1350 * Note, despite the name, this returns the tgid not the pid. The tgid and
1351 * the pid are identical unless CLONE_THREAD was specified on clone() in
1352 * which case the tgid is the same in all threads of the same group.
1354 * This is SMP safe as current->tgid does not change.
1356 SYSCALL_DEFINE0(getpid)
1358 return task_tgid_vnr(current);
1362 * Accessing ->real_parent is not SMP-safe, it could
1363 * change from under us. However, we can use a stale
1364 * value of ->real_parent under rcu_read_lock(), see
1365 * release_task()->call_rcu(delayed_put_task_struct).
1367 SYSCALL_DEFINE0(getppid)
1369 int pid;
1371 rcu_read_lock();
1372 pid = task_tgid_vnr(current->real_parent);
1373 rcu_read_unlock();
1375 return pid;
1378 SYSCALL_DEFINE0(getuid)
1380 /* Only we change this so SMP safe */
1381 return current_uid();
1384 SYSCALL_DEFINE0(geteuid)
1386 /* Only we change this so SMP safe */
1387 return current_euid();
1390 SYSCALL_DEFINE0(getgid)
1392 /* Only we change this so SMP safe */
1393 return current_gid();
1396 SYSCALL_DEFINE0(getegid)
1398 /* Only we change this so SMP safe */
1399 return current_egid();
1402 #endif
1404 static void process_timeout(unsigned long __data)
1406 wake_up_process((struct task_struct *)__data);
1410 * schedule_timeout - sleep until timeout
1411 * @timeout: timeout value in jiffies
1413 * Make the current task sleep until @timeout jiffies have
1414 * elapsed. The routine will return immediately unless
1415 * the current task state has been set (see set_current_state()).
1417 * You can set the task state as follows -
1419 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1420 * pass before the routine returns. The routine will return 0
1422 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1423 * delivered to the current task. In this case the remaining time
1424 * in jiffies will be returned, or 0 if the timer expired in time
1426 * The current task state is guaranteed to be TASK_RUNNING when this
1427 * routine returns.
1429 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1430 * the CPU away without a bound on the timeout. In this case the return
1431 * value will be %MAX_SCHEDULE_TIMEOUT.
1433 * In all cases the return value is guaranteed to be non-negative.
1435 signed long __sched schedule_timeout(signed long timeout)
1437 struct timer_list timer;
1438 unsigned long expire;
1440 switch (timeout)
1442 case MAX_SCHEDULE_TIMEOUT:
1444 * These two special cases are useful to be comfortable
1445 * in the caller. Nothing more. We could take
1446 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1447 * but I' d like to return a valid offset (>=0) to allow
1448 * the caller to do everything it want with the retval.
1450 schedule();
1451 goto out;
1452 default:
1454 * Another bit of PARANOID. Note that the retval will be
1455 * 0 since no piece of kernel is supposed to do a check
1456 * for a negative retval of schedule_timeout() (since it
1457 * should never happens anyway). You just have the printk()
1458 * that will tell you if something is gone wrong and where.
1460 if (timeout < 0) {
1461 printk(KERN_ERR "schedule_timeout: wrong timeout "
1462 "value %lx\n", timeout);
1463 dump_stack();
1464 current->state = TASK_RUNNING;
1465 goto out;
1469 expire = timeout + jiffies;
1471 setup_timer_on_stack(&timer, process_timeout, (unsigned long)current);
1472 __mod_timer(&timer, expire, false, TIMER_NOT_PINNED);
1473 schedule();
1474 del_singleshot_timer_sync(&timer);
1476 /* Remove the timer from the object tracker */
1477 destroy_timer_on_stack(&timer);
1479 timeout = expire - jiffies;
1481 out:
1482 return timeout < 0 ? 0 : timeout;
1484 EXPORT_SYMBOL(schedule_timeout);
1487 * We can use __set_current_state() here because schedule_timeout() calls
1488 * schedule() unconditionally.
1490 signed long __sched schedule_timeout_interruptible(signed long timeout)
1492 __set_current_state(TASK_INTERRUPTIBLE);
1493 return schedule_timeout(timeout);
1495 EXPORT_SYMBOL(schedule_timeout_interruptible);
1497 signed long __sched schedule_timeout_killable(signed long timeout)
1499 __set_current_state(TASK_KILLABLE);
1500 return schedule_timeout(timeout);
1502 EXPORT_SYMBOL(schedule_timeout_killable);
1504 signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1506 __set_current_state(TASK_UNINTERRUPTIBLE);
1507 return schedule_timeout(timeout);
1509 EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1511 /* Thread ID - the internal kernel "pid" */
1512 SYSCALL_DEFINE0(gettid)
1514 return task_pid_vnr(current);
1518 * do_sysinfo - fill in sysinfo struct
1519 * @info: pointer to buffer to fill
1521 int do_sysinfo(struct sysinfo *info)
1523 unsigned long mem_total, sav_total;
1524 unsigned int mem_unit, bitcount;
1525 struct timespec tp;
1527 memset(info, 0, sizeof(struct sysinfo));
1529 ktime_get_ts(&tp);
1530 monotonic_to_bootbased(&tp);
1531 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
1533 get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
1535 info->procs = nr_threads;
1537 si_meminfo(info);
1538 si_swapinfo(info);
1541 * If the sum of all the available memory (i.e. ram + swap)
1542 * is less than can be stored in a 32 bit unsigned long then
1543 * we can be binary compatible with 2.2.x kernels. If not,
1544 * well, in that case 2.2.x was broken anyways...
1546 * -Erik Andersen <andersee@debian.org>
1549 mem_total = info->totalram + info->totalswap;
1550 if (mem_total < info->totalram || mem_total < info->totalswap)
1551 goto out;
1552 bitcount = 0;
1553 mem_unit = info->mem_unit;
1554 while (mem_unit > 1) {
1555 bitcount++;
1556 mem_unit >>= 1;
1557 sav_total = mem_total;
1558 mem_total <<= 1;
1559 if (mem_total < sav_total)
1560 goto out;
1564 * If mem_total did not overflow, multiply all memory values by
1565 * info->mem_unit and set it to 1. This leaves things compatible
1566 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1567 * kernels...
1570 info->mem_unit = 1;
1571 info->totalram <<= bitcount;
1572 info->freeram <<= bitcount;
1573 info->sharedram <<= bitcount;
1574 info->bufferram <<= bitcount;
1575 info->totalswap <<= bitcount;
1576 info->freeswap <<= bitcount;
1577 info->totalhigh <<= bitcount;
1578 info->freehigh <<= bitcount;
1580 out:
1581 return 0;
1584 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
1586 struct sysinfo val;
1588 do_sysinfo(&val);
1590 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
1591 return -EFAULT;
1593 return 0;
1596 static int __cpuinit init_timers_cpu(int cpu)
1598 int j;
1599 struct tvec_base *base;
1600 static char __cpuinitdata tvec_base_done[NR_CPUS];
1602 if (!tvec_base_done[cpu]) {
1603 static char boot_done;
1605 if (boot_done) {
1607 * The APs use this path later in boot
1609 base = kmalloc_node(sizeof(*base),
1610 GFP_KERNEL | __GFP_ZERO,
1611 cpu_to_node(cpu));
1612 if (!base)
1613 return -ENOMEM;
1615 /* Make sure that tvec_base is 2 byte aligned */
1616 if (tbase_get_deferrable(base)) {
1617 WARN_ON(1);
1618 kfree(base);
1619 return -ENOMEM;
1621 per_cpu(tvec_bases, cpu) = base;
1622 } else {
1624 * This is for the boot CPU - we use compile-time
1625 * static initialisation because per-cpu memory isn't
1626 * ready yet and because the memory allocators are not
1627 * initialised either.
1629 boot_done = 1;
1630 base = &boot_tvec_bases;
1632 tvec_base_done[cpu] = 1;
1633 } else {
1634 base = per_cpu(tvec_bases, cpu);
1637 spin_lock_init(&base->lock);
1639 for (j = 0; j < TVN_SIZE; j++) {
1640 INIT_LIST_HEAD(base->tv5.vec + j);
1641 INIT_LIST_HEAD(base->tv4.vec + j);
1642 INIT_LIST_HEAD(base->tv3.vec + j);
1643 INIT_LIST_HEAD(base->tv2.vec + j);
1645 for (j = 0; j < TVR_SIZE; j++)
1646 INIT_LIST_HEAD(base->tv1.vec + j);
1648 base->timer_jiffies = jiffies;
1649 base->next_timer = base->timer_jiffies;
1650 return 0;
1653 #ifdef CONFIG_HOTPLUG_CPU
1654 static void migrate_timer_list(struct tvec_base *new_base, struct list_head *head)
1656 struct timer_list *timer;
1658 while (!list_empty(head)) {
1659 timer = list_first_entry(head, struct timer_list, entry);
1660 detach_timer(timer, 0);
1661 timer_set_base(timer, new_base);
1662 if (time_before(timer->expires, new_base->next_timer) &&
1663 !tbase_get_deferrable(timer->base))
1664 new_base->next_timer = timer->expires;
1665 internal_add_timer(new_base, timer);
1669 static void __cpuinit migrate_timers(int cpu)
1671 struct tvec_base *old_base;
1672 struct tvec_base *new_base;
1673 int i;
1675 BUG_ON(cpu_online(cpu));
1676 old_base = per_cpu(tvec_bases, cpu);
1677 new_base = get_cpu_var(tvec_bases);
1679 * The caller is globally serialized and nobody else
1680 * takes two locks at once, deadlock is not possible.
1682 spin_lock_irq(&new_base->lock);
1683 spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1685 BUG_ON(old_base->running_timer);
1687 for (i = 0; i < TVR_SIZE; i++)
1688 migrate_timer_list(new_base, old_base->tv1.vec + i);
1689 for (i = 0; i < TVN_SIZE; i++) {
1690 migrate_timer_list(new_base, old_base->tv2.vec + i);
1691 migrate_timer_list(new_base, old_base->tv3.vec + i);
1692 migrate_timer_list(new_base, old_base->tv4.vec + i);
1693 migrate_timer_list(new_base, old_base->tv5.vec + i);
1696 spin_unlock(&old_base->lock);
1697 spin_unlock_irq(&new_base->lock);
1698 put_cpu_var(tvec_bases);
1700 #endif /* CONFIG_HOTPLUG_CPU */
1702 static int __cpuinit timer_cpu_notify(struct notifier_block *self,
1703 unsigned long action, void *hcpu)
1705 long cpu = (long)hcpu;
1706 int err;
1708 switch(action) {
1709 case CPU_UP_PREPARE:
1710 case CPU_UP_PREPARE_FROZEN:
1711 err = init_timers_cpu(cpu);
1712 if (err < 0)
1713 return notifier_from_errno(err);
1714 break;
1715 #ifdef CONFIG_HOTPLUG_CPU
1716 case CPU_DEAD:
1717 case CPU_DEAD_FROZEN:
1718 migrate_timers(cpu);
1719 break;
1720 #endif
1721 default:
1722 break;
1724 return NOTIFY_OK;
1727 static struct notifier_block __cpuinitdata timers_nb = {
1728 .notifier_call = timer_cpu_notify,
1732 void __init init_timers(void)
1734 int err = timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE,
1735 (void *)(long)smp_processor_id());
1737 init_timer_stats();
1739 BUG_ON(err != NOTIFY_OK);
1740 register_cpu_notifier(&timers_nb);
1741 open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
1745 * msleep - sleep safely even with waitqueue interruptions
1746 * @msecs: Time in milliseconds to sleep for
1748 void msleep(unsigned int msecs)
1750 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1752 while (timeout)
1753 timeout = schedule_timeout_uninterruptible(timeout);
1756 EXPORT_SYMBOL(msleep);
1759 * msleep_interruptible - sleep waiting for signals
1760 * @msecs: Time in milliseconds to sleep for
1762 unsigned long msleep_interruptible(unsigned int msecs)
1764 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1766 while (timeout && !signal_pending(current))
1767 timeout = schedule_timeout_interruptible(timeout);
1768 return jiffies_to_msecs(timeout);
1771 EXPORT_SYMBOL(msleep_interruptible);
1773 static int __sched do_usleep_range(unsigned long min, unsigned long max)
1775 ktime_t kmin;
1776 unsigned long delta;
1778 kmin = ktime_set(0, min * NSEC_PER_USEC);
1779 delta = (max - min) * NSEC_PER_USEC;
1780 return schedule_hrtimeout_range(&kmin, delta, HRTIMER_MODE_REL);
1784 * usleep_range - Drop in replacement for udelay where wakeup is flexible
1785 * @min: Minimum time in usecs to sleep
1786 * @max: Maximum time in usecs to sleep
1788 void usleep_range(unsigned long min, unsigned long max)
1790 __set_current_state(TASK_UNINTERRUPTIBLE);
1791 do_usleep_range(min, max);
1793 EXPORT_SYMBOL(usleep_range);