Merge commit 'v3.3-rc1' into stable/for-linus-fixes-3.3
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / kernel / timer.c
bloba297ffcf888eb20093b0609485e03edbd85cd214
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/export.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;
430 /* Stub timer callback for improperly used timers. */
431 static void stub_timer(unsigned long data)
433 WARN_ON(1);
437 * fixup_activate is called when:
438 * - an active object is activated
439 * - an unknown object is activated (might be a statically initialized object)
441 static int timer_fixup_activate(void *addr, enum debug_obj_state state)
443 struct timer_list *timer = addr;
445 switch (state) {
447 case ODEBUG_STATE_NOTAVAILABLE:
449 * This is not really a fixup. The timer was
450 * statically initialized. We just make sure that it
451 * is tracked in the object tracker.
453 if (timer->entry.next == NULL &&
454 timer->entry.prev == TIMER_ENTRY_STATIC) {
455 debug_object_init(timer, &timer_debug_descr);
456 debug_object_activate(timer, &timer_debug_descr);
457 return 0;
458 } else {
459 setup_timer(timer, stub_timer, 0);
460 return 1;
462 return 0;
464 case ODEBUG_STATE_ACTIVE:
465 WARN_ON(1);
467 default:
468 return 0;
473 * fixup_free is called when:
474 * - an active object is freed
476 static int timer_fixup_free(void *addr, enum debug_obj_state state)
478 struct timer_list *timer = addr;
480 switch (state) {
481 case ODEBUG_STATE_ACTIVE:
482 del_timer_sync(timer);
483 debug_object_free(timer, &timer_debug_descr);
484 return 1;
485 default:
486 return 0;
491 * fixup_assert_init is called when:
492 * - an untracked/uninit-ed object is found
494 static int timer_fixup_assert_init(void *addr, enum debug_obj_state state)
496 struct timer_list *timer = addr;
498 switch (state) {
499 case ODEBUG_STATE_NOTAVAILABLE:
500 if (timer->entry.prev == TIMER_ENTRY_STATIC) {
502 * This is not really a fixup. The timer was
503 * statically initialized. We just make sure that it
504 * is tracked in the object tracker.
506 debug_object_init(timer, &timer_debug_descr);
507 return 0;
508 } else {
509 setup_timer(timer, stub_timer, 0);
510 return 1;
512 default:
513 return 0;
517 static struct debug_obj_descr timer_debug_descr = {
518 .name = "timer_list",
519 .debug_hint = timer_debug_hint,
520 .fixup_init = timer_fixup_init,
521 .fixup_activate = timer_fixup_activate,
522 .fixup_free = timer_fixup_free,
523 .fixup_assert_init = timer_fixup_assert_init,
526 static inline void debug_timer_init(struct timer_list *timer)
528 debug_object_init(timer, &timer_debug_descr);
531 static inline void debug_timer_activate(struct timer_list *timer)
533 debug_object_activate(timer, &timer_debug_descr);
536 static inline void debug_timer_deactivate(struct timer_list *timer)
538 debug_object_deactivate(timer, &timer_debug_descr);
541 static inline void debug_timer_free(struct timer_list *timer)
543 debug_object_free(timer, &timer_debug_descr);
546 static inline void debug_timer_assert_init(struct timer_list *timer)
548 debug_object_assert_init(timer, &timer_debug_descr);
551 static void __init_timer(struct timer_list *timer,
552 const char *name,
553 struct lock_class_key *key);
555 void init_timer_on_stack_key(struct timer_list *timer,
556 const char *name,
557 struct lock_class_key *key)
559 debug_object_init_on_stack(timer, &timer_debug_descr);
560 __init_timer(timer, name, key);
562 EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
564 void destroy_timer_on_stack(struct timer_list *timer)
566 debug_object_free(timer, &timer_debug_descr);
568 EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
570 #else
571 static inline void debug_timer_init(struct timer_list *timer) { }
572 static inline void debug_timer_activate(struct timer_list *timer) { }
573 static inline void debug_timer_deactivate(struct timer_list *timer) { }
574 static inline void debug_timer_assert_init(struct timer_list *timer) { }
575 #endif
577 static inline void debug_init(struct timer_list *timer)
579 debug_timer_init(timer);
580 trace_timer_init(timer);
583 static inline void
584 debug_activate(struct timer_list *timer, unsigned long expires)
586 debug_timer_activate(timer);
587 trace_timer_start(timer, expires);
590 static inline void debug_deactivate(struct timer_list *timer)
592 debug_timer_deactivate(timer);
593 trace_timer_cancel(timer);
596 static inline void debug_assert_init(struct timer_list *timer)
598 debug_timer_assert_init(timer);
601 static void __init_timer(struct timer_list *timer,
602 const char *name,
603 struct lock_class_key *key)
605 timer->entry.next = NULL;
606 timer->base = __raw_get_cpu_var(tvec_bases);
607 timer->slack = -1;
608 #ifdef CONFIG_TIMER_STATS
609 timer->start_site = NULL;
610 timer->start_pid = -1;
611 memset(timer->start_comm, 0, TASK_COMM_LEN);
612 #endif
613 lockdep_init_map(&timer->lockdep_map, name, key, 0);
616 void setup_deferrable_timer_on_stack_key(struct timer_list *timer,
617 const char *name,
618 struct lock_class_key *key,
619 void (*function)(unsigned long),
620 unsigned long data)
622 timer->function = function;
623 timer->data = data;
624 init_timer_on_stack_key(timer, name, key);
625 timer_set_deferrable(timer);
627 EXPORT_SYMBOL_GPL(setup_deferrable_timer_on_stack_key);
630 * init_timer_key - initialize a timer
631 * @timer: the timer to be initialized
632 * @name: name of the timer
633 * @key: lockdep class key of the fake lock used for tracking timer
634 * sync lock dependencies
636 * init_timer_key() must be done to a timer prior calling *any* of the
637 * other timer functions.
639 void init_timer_key(struct timer_list *timer,
640 const char *name,
641 struct lock_class_key *key)
643 debug_init(timer);
644 __init_timer(timer, name, key);
646 EXPORT_SYMBOL(init_timer_key);
648 void init_timer_deferrable_key(struct timer_list *timer,
649 const char *name,
650 struct lock_class_key *key)
652 init_timer_key(timer, name, key);
653 timer_set_deferrable(timer);
655 EXPORT_SYMBOL(init_timer_deferrable_key);
657 static inline void detach_timer(struct timer_list *timer,
658 int clear_pending)
660 struct list_head *entry = &timer->entry;
662 debug_deactivate(timer);
664 __list_del(entry->prev, entry->next);
665 if (clear_pending)
666 entry->next = NULL;
667 entry->prev = LIST_POISON2;
671 * We are using hashed locking: holding per_cpu(tvec_bases).lock
672 * means that all timers which are tied to this base via timer->base are
673 * locked, and the base itself is locked too.
675 * So __run_timers/migrate_timers can safely modify all timers which could
676 * be found on ->tvX lists.
678 * When the timer's base is locked, and the timer removed from list, it is
679 * possible to set timer->base = NULL and drop the lock: the timer remains
680 * locked.
682 static struct tvec_base *lock_timer_base(struct timer_list *timer,
683 unsigned long *flags)
684 __acquires(timer->base->lock)
686 struct tvec_base *base;
688 for (;;) {
689 struct tvec_base *prelock_base = timer->base;
690 base = tbase_get_base(prelock_base);
691 if (likely(base != NULL)) {
692 spin_lock_irqsave(&base->lock, *flags);
693 if (likely(prelock_base == timer->base))
694 return base;
695 /* The timer has migrated to another CPU */
696 spin_unlock_irqrestore(&base->lock, *flags);
698 cpu_relax();
702 static inline int
703 __mod_timer(struct timer_list *timer, unsigned long expires,
704 bool pending_only, int pinned)
706 struct tvec_base *base, *new_base;
707 unsigned long flags;
708 int ret = 0 , cpu;
710 timer_stats_timer_set_start_info(timer);
711 BUG_ON(!timer->function);
713 base = lock_timer_base(timer, &flags);
715 if (timer_pending(timer)) {
716 detach_timer(timer, 0);
717 if (timer->expires == base->next_timer &&
718 !tbase_get_deferrable(timer->base))
719 base->next_timer = base->timer_jiffies;
720 ret = 1;
721 } else {
722 if (pending_only)
723 goto out_unlock;
726 debug_activate(timer, expires);
728 cpu = smp_processor_id();
730 #if defined(CONFIG_NO_HZ) && defined(CONFIG_SMP)
731 if (!pinned && get_sysctl_timer_migration() && idle_cpu(cpu))
732 cpu = get_nohz_timer_target();
733 #endif
734 new_base = per_cpu(tvec_bases, cpu);
736 if (base != new_base) {
738 * We are trying to schedule the timer on the local CPU.
739 * However we can't change timer's base while it is running,
740 * otherwise del_timer_sync() can't detect that the timer's
741 * handler yet has not finished. This also guarantees that
742 * the timer is serialized wrt itself.
744 if (likely(base->running_timer != timer)) {
745 /* See the comment in lock_timer_base() */
746 timer_set_base(timer, NULL);
747 spin_unlock(&base->lock);
748 base = new_base;
749 spin_lock(&base->lock);
750 timer_set_base(timer, base);
754 timer->expires = expires;
755 if (time_before(timer->expires, base->next_timer) &&
756 !tbase_get_deferrable(timer->base))
757 base->next_timer = timer->expires;
758 internal_add_timer(base, timer);
760 out_unlock:
761 spin_unlock_irqrestore(&base->lock, flags);
763 return ret;
767 * mod_timer_pending - modify a pending timer's timeout
768 * @timer: the pending timer to be modified
769 * @expires: new timeout in jiffies
771 * mod_timer_pending() is the same for pending timers as mod_timer(),
772 * but will not re-activate and modify already deleted timers.
774 * It is useful for unserialized use of timers.
776 int mod_timer_pending(struct timer_list *timer, unsigned long expires)
778 return __mod_timer(timer, expires, true, TIMER_NOT_PINNED);
780 EXPORT_SYMBOL(mod_timer_pending);
783 * Decide where to put the timer while taking the slack into account
785 * Algorithm:
786 * 1) calculate the maximum (absolute) time
787 * 2) calculate the highest bit where the expires and new max are different
788 * 3) use this bit to make a mask
789 * 4) use the bitmask to round down the maximum time, so that all last
790 * bits are zeros
792 static inline
793 unsigned long apply_slack(struct timer_list *timer, unsigned long expires)
795 unsigned long expires_limit, mask;
796 int bit;
798 if (timer->slack >= 0) {
799 expires_limit = expires + timer->slack;
800 } else {
801 long delta = expires - jiffies;
803 if (delta < 256)
804 return expires;
806 expires_limit = expires + delta / 256;
808 mask = expires ^ expires_limit;
809 if (mask == 0)
810 return expires;
812 bit = find_last_bit(&mask, BITS_PER_LONG);
814 mask = (1 << bit) - 1;
816 expires_limit = expires_limit & ~(mask);
818 return expires_limit;
822 * mod_timer - modify a timer's timeout
823 * @timer: the timer to be modified
824 * @expires: new timeout in jiffies
826 * mod_timer() is a more efficient way to update the expire field of an
827 * active timer (if the timer is inactive it will be activated)
829 * mod_timer(timer, expires) is equivalent to:
831 * del_timer(timer); timer->expires = expires; add_timer(timer);
833 * Note that if there are multiple unserialized concurrent users of the
834 * same timer, then mod_timer() is the only safe way to modify the timeout,
835 * since add_timer() cannot modify an already running timer.
837 * The function returns whether it has modified a pending timer or not.
838 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
839 * active timer returns 1.)
841 int mod_timer(struct timer_list *timer, unsigned long expires)
843 expires = apply_slack(timer, expires);
846 * This is a common optimization triggered by the
847 * networking code - if the timer is re-modified
848 * to be the same thing then just return:
850 if (timer_pending(timer) && timer->expires == expires)
851 return 1;
853 return __mod_timer(timer, expires, false, TIMER_NOT_PINNED);
855 EXPORT_SYMBOL(mod_timer);
858 * mod_timer_pinned - modify a timer's timeout
859 * @timer: the timer to be modified
860 * @expires: new timeout in jiffies
862 * mod_timer_pinned() is a way to update the expire field of an
863 * active timer (if the timer is inactive it will be activated)
864 * and not allow the timer to be migrated to a different CPU.
866 * mod_timer_pinned(timer, expires) is equivalent to:
868 * del_timer(timer); timer->expires = expires; add_timer(timer);
870 int mod_timer_pinned(struct timer_list *timer, unsigned long expires)
872 if (timer->expires == expires && timer_pending(timer))
873 return 1;
875 return __mod_timer(timer, expires, false, TIMER_PINNED);
877 EXPORT_SYMBOL(mod_timer_pinned);
880 * add_timer - start a timer
881 * @timer: the timer to be added
883 * The kernel will do a ->function(->data) callback from the
884 * timer interrupt at the ->expires point in the future. The
885 * current time is 'jiffies'.
887 * The timer's ->expires, ->function (and if the handler uses it, ->data)
888 * fields must be set prior calling this function.
890 * Timers with an ->expires field in the past will be executed in the next
891 * timer tick.
893 void add_timer(struct timer_list *timer)
895 BUG_ON(timer_pending(timer));
896 mod_timer(timer, timer->expires);
898 EXPORT_SYMBOL(add_timer);
901 * add_timer_on - start a timer on a particular CPU
902 * @timer: the timer to be added
903 * @cpu: the CPU to start it on
905 * This is not very scalable on SMP. Double adds are not possible.
907 void add_timer_on(struct timer_list *timer, int cpu)
909 struct tvec_base *base = per_cpu(tvec_bases, cpu);
910 unsigned long flags;
912 timer_stats_timer_set_start_info(timer);
913 BUG_ON(timer_pending(timer) || !timer->function);
914 spin_lock_irqsave(&base->lock, flags);
915 timer_set_base(timer, base);
916 debug_activate(timer, timer->expires);
917 if (time_before(timer->expires, base->next_timer) &&
918 !tbase_get_deferrable(timer->base))
919 base->next_timer = timer->expires;
920 internal_add_timer(base, timer);
922 * Check whether the other CPU is idle and needs to be
923 * triggered to reevaluate the timer wheel when nohz is
924 * active. We are protected against the other CPU fiddling
925 * with the timer by holding the timer base lock. This also
926 * makes sure that a CPU on the way to idle can not evaluate
927 * the timer wheel.
929 wake_up_idle_cpu(cpu);
930 spin_unlock_irqrestore(&base->lock, flags);
932 EXPORT_SYMBOL_GPL(add_timer_on);
935 * del_timer - deactive a timer.
936 * @timer: the timer to be deactivated
938 * del_timer() deactivates a timer - this works on both active and inactive
939 * timers.
941 * The function returns whether it has deactivated a pending timer or not.
942 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
943 * active timer returns 1.)
945 int del_timer(struct timer_list *timer)
947 struct tvec_base *base;
948 unsigned long flags;
949 int ret = 0;
951 debug_assert_init(timer);
953 timer_stats_timer_clear_start_info(timer);
954 if (timer_pending(timer)) {
955 base = lock_timer_base(timer, &flags);
956 if (timer_pending(timer)) {
957 detach_timer(timer, 1);
958 if (timer->expires == base->next_timer &&
959 !tbase_get_deferrable(timer->base))
960 base->next_timer = base->timer_jiffies;
961 ret = 1;
963 spin_unlock_irqrestore(&base->lock, flags);
966 return ret;
968 EXPORT_SYMBOL(del_timer);
971 * try_to_del_timer_sync - Try to deactivate a timer
972 * @timer: timer do del
974 * This function tries to deactivate a timer. Upon successful (ret >= 0)
975 * exit the timer is not queued and the handler is not running on any CPU.
977 int try_to_del_timer_sync(struct timer_list *timer)
979 struct tvec_base *base;
980 unsigned long flags;
981 int ret = -1;
983 debug_assert_init(timer);
985 base = lock_timer_base(timer, &flags);
987 if (base->running_timer == timer)
988 goto out;
990 timer_stats_timer_clear_start_info(timer);
991 ret = 0;
992 if (timer_pending(timer)) {
993 detach_timer(timer, 1);
994 if (timer->expires == base->next_timer &&
995 !tbase_get_deferrable(timer->base))
996 base->next_timer = base->timer_jiffies;
997 ret = 1;
999 out:
1000 spin_unlock_irqrestore(&base->lock, flags);
1002 return ret;
1004 EXPORT_SYMBOL(try_to_del_timer_sync);
1006 #ifdef CONFIG_SMP
1008 * del_timer_sync - deactivate a timer and wait for the handler to finish.
1009 * @timer: the timer to be deactivated
1011 * This function only differs from del_timer() on SMP: besides deactivating
1012 * the timer it also makes sure the handler has finished executing on other
1013 * CPUs.
1015 * Synchronization rules: Callers must prevent restarting of the timer,
1016 * otherwise this function is meaningless. It must not be called from
1017 * interrupt contexts. The caller must not hold locks which would prevent
1018 * completion of the timer's handler. The timer's handler must not call
1019 * add_timer_on(). Upon exit the timer is not queued and the handler is
1020 * not running on any CPU.
1022 * Note: You must not hold locks that are held in interrupt context
1023 * while calling this function. Even if the lock has nothing to do
1024 * with the timer in question. Here's why:
1026 * CPU0 CPU1
1027 * ---- ----
1028 * <SOFTIRQ>
1029 * call_timer_fn();
1030 * base->running_timer = mytimer;
1031 * spin_lock_irq(somelock);
1032 * <IRQ>
1033 * spin_lock(somelock);
1034 * del_timer_sync(mytimer);
1035 * while (base->running_timer == mytimer);
1037 * Now del_timer_sync() will never return and never release somelock.
1038 * The interrupt on the other CPU is waiting to grab somelock but
1039 * it has interrupted the softirq that CPU0 is waiting to finish.
1041 * The function returns whether it has deactivated a pending timer or not.
1043 int del_timer_sync(struct timer_list *timer)
1045 #ifdef CONFIG_LOCKDEP
1046 unsigned long flags;
1049 * If lockdep gives a backtrace here, please reference
1050 * the synchronization rules above.
1052 local_irq_save(flags);
1053 lock_map_acquire(&timer->lockdep_map);
1054 lock_map_release(&timer->lockdep_map);
1055 local_irq_restore(flags);
1056 #endif
1058 * don't use it in hardirq context, because it
1059 * could lead to deadlock.
1061 WARN_ON(in_irq());
1062 for (;;) {
1063 int ret = try_to_del_timer_sync(timer);
1064 if (ret >= 0)
1065 return ret;
1066 cpu_relax();
1069 EXPORT_SYMBOL(del_timer_sync);
1070 #endif
1072 static int cascade(struct tvec_base *base, struct tvec *tv, int index)
1074 /* cascade all the timers from tv up one level */
1075 struct timer_list *timer, *tmp;
1076 struct list_head tv_list;
1078 list_replace_init(tv->vec + index, &tv_list);
1081 * We are removing _all_ timers from the list, so we
1082 * don't have to detach them individually.
1084 list_for_each_entry_safe(timer, tmp, &tv_list, entry) {
1085 BUG_ON(tbase_get_base(timer->base) != base);
1086 internal_add_timer(base, timer);
1089 return index;
1092 static void call_timer_fn(struct timer_list *timer, void (*fn)(unsigned long),
1093 unsigned long data)
1095 int preempt_count = preempt_count();
1097 #ifdef CONFIG_LOCKDEP
1099 * It is permissible to free the timer from inside the
1100 * function that is called from it, this we need to take into
1101 * account for lockdep too. To avoid bogus "held lock freed"
1102 * warnings as well as problems when looking into
1103 * timer->lockdep_map, make a copy and use that here.
1105 struct lockdep_map lockdep_map = timer->lockdep_map;
1106 #endif
1108 * Couple the lock chain with the lock chain at
1109 * del_timer_sync() by acquiring the lock_map around the fn()
1110 * call here and in del_timer_sync().
1112 lock_map_acquire(&lockdep_map);
1114 trace_timer_expire_entry(timer);
1115 fn(data);
1116 trace_timer_expire_exit(timer);
1118 lock_map_release(&lockdep_map);
1120 if (preempt_count != preempt_count()) {
1121 WARN_ONCE(1, "timer: %pF preempt leak: %08x -> %08x\n",
1122 fn, preempt_count, preempt_count());
1124 * Restore the preempt count. That gives us a decent
1125 * chance to survive and extract information. If the
1126 * callback kept a lock held, bad luck, but not worse
1127 * than the BUG() we had.
1129 preempt_count() = preempt_count;
1133 #define INDEX(N) ((base->timer_jiffies >> (TVR_BITS + (N) * TVN_BITS)) & TVN_MASK)
1136 * __run_timers - run all expired timers (if any) on this CPU.
1137 * @base: the timer vector to be processed.
1139 * This function cascades all vectors and executes all expired timer
1140 * vectors.
1142 static inline void __run_timers(struct tvec_base *base)
1144 struct timer_list *timer;
1146 spin_lock_irq(&base->lock);
1147 while (time_after_eq(jiffies, base->timer_jiffies)) {
1148 struct list_head work_list;
1149 struct list_head *head = &work_list;
1150 int index = base->timer_jiffies & TVR_MASK;
1153 * Cascade timers:
1155 if (!index &&
1156 (!cascade(base, &base->tv2, INDEX(0))) &&
1157 (!cascade(base, &base->tv3, INDEX(1))) &&
1158 !cascade(base, &base->tv4, INDEX(2)))
1159 cascade(base, &base->tv5, INDEX(3));
1160 ++base->timer_jiffies;
1161 list_replace_init(base->tv1.vec + index, &work_list);
1162 while (!list_empty(head)) {
1163 void (*fn)(unsigned long);
1164 unsigned long data;
1166 timer = list_first_entry(head, struct timer_list,entry);
1167 fn = timer->function;
1168 data = timer->data;
1170 timer_stats_account_timer(timer);
1172 base->running_timer = timer;
1173 detach_timer(timer, 1);
1175 spin_unlock_irq(&base->lock);
1176 call_timer_fn(timer, fn, data);
1177 spin_lock_irq(&base->lock);
1180 base->running_timer = NULL;
1181 spin_unlock_irq(&base->lock);
1184 #ifdef CONFIG_NO_HZ
1186 * Find out when the next timer event is due to happen. This
1187 * is used on S/390 to stop all activity when a CPU is idle.
1188 * This function needs to be called with interrupts disabled.
1190 static unsigned long __next_timer_interrupt(struct tvec_base *base)
1192 unsigned long timer_jiffies = base->timer_jiffies;
1193 unsigned long expires = timer_jiffies + NEXT_TIMER_MAX_DELTA;
1194 int index, slot, array, found = 0;
1195 struct timer_list *nte;
1196 struct tvec *varray[4];
1198 /* Look for timer events in tv1. */
1199 index = slot = timer_jiffies & TVR_MASK;
1200 do {
1201 list_for_each_entry(nte, base->tv1.vec + slot, entry) {
1202 if (tbase_get_deferrable(nte->base))
1203 continue;
1205 found = 1;
1206 expires = nte->expires;
1207 /* Look at the cascade bucket(s)? */
1208 if (!index || slot < index)
1209 goto cascade;
1210 return expires;
1212 slot = (slot + 1) & TVR_MASK;
1213 } while (slot != index);
1215 cascade:
1216 /* Calculate the next cascade event */
1217 if (index)
1218 timer_jiffies += TVR_SIZE - index;
1219 timer_jiffies >>= TVR_BITS;
1221 /* Check tv2-tv5. */
1222 varray[0] = &base->tv2;
1223 varray[1] = &base->tv3;
1224 varray[2] = &base->tv4;
1225 varray[3] = &base->tv5;
1227 for (array = 0; array < 4; array++) {
1228 struct tvec *varp = varray[array];
1230 index = slot = timer_jiffies & TVN_MASK;
1231 do {
1232 list_for_each_entry(nte, varp->vec + slot, entry) {
1233 if (tbase_get_deferrable(nte->base))
1234 continue;
1236 found = 1;
1237 if (time_before(nte->expires, expires))
1238 expires = nte->expires;
1241 * Do we still search for the first timer or are
1242 * we looking up the cascade buckets ?
1244 if (found) {
1245 /* Look at the cascade bucket(s)? */
1246 if (!index || slot < index)
1247 break;
1248 return expires;
1250 slot = (slot + 1) & TVN_MASK;
1251 } while (slot != index);
1253 if (index)
1254 timer_jiffies += TVN_SIZE - index;
1255 timer_jiffies >>= TVN_BITS;
1257 return expires;
1261 * Check, if the next hrtimer event is before the next timer wheel
1262 * event:
1264 static unsigned long cmp_next_hrtimer_event(unsigned long now,
1265 unsigned long expires)
1267 ktime_t hr_delta = hrtimer_get_next_event();
1268 struct timespec tsdelta;
1269 unsigned long delta;
1271 if (hr_delta.tv64 == KTIME_MAX)
1272 return expires;
1275 * Expired timer available, let it expire in the next tick
1277 if (hr_delta.tv64 <= 0)
1278 return now + 1;
1280 tsdelta = ktime_to_timespec(hr_delta);
1281 delta = timespec_to_jiffies(&tsdelta);
1284 * Limit the delta to the max value, which is checked in
1285 * tick_nohz_stop_sched_tick():
1287 if (delta > NEXT_TIMER_MAX_DELTA)
1288 delta = NEXT_TIMER_MAX_DELTA;
1291 * Take rounding errors in to account and make sure, that it
1292 * expires in the next tick. Otherwise we go into an endless
1293 * ping pong due to tick_nohz_stop_sched_tick() retriggering
1294 * the timer softirq
1296 if (delta < 1)
1297 delta = 1;
1298 now += delta;
1299 if (time_before(now, expires))
1300 return now;
1301 return expires;
1305 * get_next_timer_interrupt - return the jiffy of the next pending timer
1306 * @now: current time (in jiffies)
1308 unsigned long get_next_timer_interrupt(unsigned long now)
1310 struct tvec_base *base = __this_cpu_read(tvec_bases);
1311 unsigned long expires;
1314 * Pretend that there is no timer pending if the cpu is offline.
1315 * Possible pending timers will be migrated later to an active cpu.
1317 if (cpu_is_offline(smp_processor_id()))
1318 return now + NEXT_TIMER_MAX_DELTA;
1319 spin_lock(&base->lock);
1320 if (time_before_eq(base->next_timer, base->timer_jiffies))
1321 base->next_timer = __next_timer_interrupt(base);
1322 expires = base->next_timer;
1323 spin_unlock(&base->lock);
1325 if (time_before_eq(expires, now))
1326 return now;
1328 return cmp_next_hrtimer_event(now, expires);
1330 #endif
1333 * Called from the timer interrupt handler to charge one tick to the current
1334 * process. user_tick is 1 if the tick is user time, 0 for system.
1336 void update_process_times(int user_tick)
1338 struct task_struct *p = current;
1339 int cpu = smp_processor_id();
1341 /* Note: this timer irq context must be accounted for as well. */
1342 account_process_tick(p, user_tick);
1343 run_local_timers();
1344 rcu_check_callbacks(cpu, user_tick);
1345 printk_tick();
1346 #ifdef CONFIG_IRQ_WORK
1347 if (in_irq())
1348 irq_work_run();
1349 #endif
1350 scheduler_tick();
1351 run_posix_cpu_timers(p);
1355 * This function runs timers and the timer-tq in bottom half context.
1357 static void run_timer_softirq(struct softirq_action *h)
1359 struct tvec_base *base = __this_cpu_read(tvec_bases);
1361 hrtimer_run_pending();
1363 if (time_after_eq(jiffies, base->timer_jiffies))
1364 __run_timers(base);
1368 * Called by the local, per-CPU timer interrupt on SMP.
1370 void run_local_timers(void)
1372 hrtimer_run_queues();
1373 raise_softirq(TIMER_SOFTIRQ);
1376 #ifdef __ARCH_WANT_SYS_ALARM
1379 * For backwards compatibility? This can be done in libc so Alpha
1380 * and all newer ports shouldn't need it.
1382 SYSCALL_DEFINE1(alarm, unsigned int, seconds)
1384 return alarm_setitimer(seconds);
1387 #endif
1389 #ifndef __alpha__
1392 * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this
1393 * should be moved into arch/i386 instead?
1397 * sys_getpid - return the thread group id of the current process
1399 * Note, despite the name, this returns the tgid not the pid. The tgid and
1400 * the pid are identical unless CLONE_THREAD was specified on clone() in
1401 * which case the tgid is the same in all threads of the same group.
1403 * This is SMP safe as current->tgid does not change.
1405 SYSCALL_DEFINE0(getpid)
1407 return task_tgid_vnr(current);
1411 * Accessing ->real_parent is not SMP-safe, it could
1412 * change from under us. However, we can use a stale
1413 * value of ->real_parent under rcu_read_lock(), see
1414 * release_task()->call_rcu(delayed_put_task_struct).
1416 SYSCALL_DEFINE0(getppid)
1418 int pid;
1420 rcu_read_lock();
1421 pid = task_tgid_vnr(rcu_dereference(current->real_parent));
1422 rcu_read_unlock();
1424 return pid;
1427 SYSCALL_DEFINE0(getuid)
1429 /* Only we change this so SMP safe */
1430 return current_uid();
1433 SYSCALL_DEFINE0(geteuid)
1435 /* Only we change this so SMP safe */
1436 return current_euid();
1439 SYSCALL_DEFINE0(getgid)
1441 /* Only we change this so SMP safe */
1442 return current_gid();
1445 SYSCALL_DEFINE0(getegid)
1447 /* Only we change this so SMP safe */
1448 return current_egid();
1451 #endif
1453 static void process_timeout(unsigned long __data)
1455 wake_up_process((struct task_struct *)__data);
1459 * schedule_timeout - sleep until timeout
1460 * @timeout: timeout value in jiffies
1462 * Make the current task sleep until @timeout jiffies have
1463 * elapsed. The routine will return immediately unless
1464 * the current task state has been set (see set_current_state()).
1466 * You can set the task state as follows -
1468 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1469 * pass before the routine returns. The routine will return 0
1471 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1472 * delivered to the current task. In this case the remaining time
1473 * in jiffies will be returned, or 0 if the timer expired in time
1475 * The current task state is guaranteed to be TASK_RUNNING when this
1476 * routine returns.
1478 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1479 * the CPU away without a bound on the timeout. In this case the return
1480 * value will be %MAX_SCHEDULE_TIMEOUT.
1482 * In all cases the return value is guaranteed to be non-negative.
1484 signed long __sched schedule_timeout(signed long timeout)
1486 struct timer_list timer;
1487 unsigned long expire;
1489 switch (timeout)
1491 case MAX_SCHEDULE_TIMEOUT:
1493 * These two special cases are useful to be comfortable
1494 * in the caller. Nothing more. We could take
1495 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1496 * but I' d like to return a valid offset (>=0) to allow
1497 * the caller to do everything it want with the retval.
1499 schedule();
1500 goto out;
1501 default:
1503 * Another bit of PARANOID. Note that the retval will be
1504 * 0 since no piece of kernel is supposed to do a check
1505 * for a negative retval of schedule_timeout() (since it
1506 * should never happens anyway). You just have the printk()
1507 * that will tell you if something is gone wrong and where.
1509 if (timeout < 0) {
1510 printk(KERN_ERR "schedule_timeout: wrong timeout "
1511 "value %lx\n", timeout);
1512 dump_stack();
1513 current->state = TASK_RUNNING;
1514 goto out;
1518 expire = timeout + jiffies;
1520 setup_timer_on_stack(&timer, process_timeout, (unsigned long)current);
1521 __mod_timer(&timer, expire, false, TIMER_NOT_PINNED);
1522 schedule();
1523 del_singleshot_timer_sync(&timer);
1525 /* Remove the timer from the object tracker */
1526 destroy_timer_on_stack(&timer);
1528 timeout = expire - jiffies;
1530 out:
1531 return timeout < 0 ? 0 : timeout;
1533 EXPORT_SYMBOL(schedule_timeout);
1536 * We can use __set_current_state() here because schedule_timeout() calls
1537 * schedule() unconditionally.
1539 signed long __sched schedule_timeout_interruptible(signed long timeout)
1541 __set_current_state(TASK_INTERRUPTIBLE);
1542 return schedule_timeout(timeout);
1544 EXPORT_SYMBOL(schedule_timeout_interruptible);
1546 signed long __sched schedule_timeout_killable(signed long timeout)
1548 __set_current_state(TASK_KILLABLE);
1549 return schedule_timeout(timeout);
1551 EXPORT_SYMBOL(schedule_timeout_killable);
1553 signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1555 __set_current_state(TASK_UNINTERRUPTIBLE);
1556 return schedule_timeout(timeout);
1558 EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1560 /* Thread ID - the internal kernel "pid" */
1561 SYSCALL_DEFINE0(gettid)
1563 return task_pid_vnr(current);
1567 * do_sysinfo - fill in sysinfo struct
1568 * @info: pointer to buffer to fill
1570 int do_sysinfo(struct sysinfo *info)
1572 unsigned long mem_total, sav_total;
1573 unsigned int mem_unit, bitcount;
1574 struct timespec tp;
1576 memset(info, 0, sizeof(struct sysinfo));
1578 ktime_get_ts(&tp);
1579 monotonic_to_bootbased(&tp);
1580 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
1582 get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
1584 info->procs = nr_threads;
1586 si_meminfo(info);
1587 si_swapinfo(info);
1590 * If the sum of all the available memory (i.e. ram + swap)
1591 * is less than can be stored in a 32 bit unsigned long then
1592 * we can be binary compatible with 2.2.x kernels. If not,
1593 * well, in that case 2.2.x was broken anyways...
1595 * -Erik Andersen <andersee@debian.org>
1598 mem_total = info->totalram + info->totalswap;
1599 if (mem_total < info->totalram || mem_total < info->totalswap)
1600 goto out;
1601 bitcount = 0;
1602 mem_unit = info->mem_unit;
1603 while (mem_unit > 1) {
1604 bitcount++;
1605 mem_unit >>= 1;
1606 sav_total = mem_total;
1607 mem_total <<= 1;
1608 if (mem_total < sav_total)
1609 goto out;
1613 * If mem_total did not overflow, multiply all memory values by
1614 * info->mem_unit and set it to 1. This leaves things compatible
1615 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1616 * kernels...
1619 info->mem_unit = 1;
1620 info->totalram <<= bitcount;
1621 info->freeram <<= bitcount;
1622 info->sharedram <<= bitcount;
1623 info->bufferram <<= bitcount;
1624 info->totalswap <<= bitcount;
1625 info->freeswap <<= bitcount;
1626 info->totalhigh <<= bitcount;
1627 info->freehigh <<= bitcount;
1629 out:
1630 return 0;
1633 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
1635 struct sysinfo val;
1637 do_sysinfo(&val);
1639 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
1640 return -EFAULT;
1642 return 0;
1645 static int __cpuinit init_timers_cpu(int cpu)
1647 int j;
1648 struct tvec_base *base;
1649 static char __cpuinitdata tvec_base_done[NR_CPUS];
1651 if (!tvec_base_done[cpu]) {
1652 static char boot_done;
1654 if (boot_done) {
1656 * The APs use this path later in boot
1658 base = kmalloc_node(sizeof(*base),
1659 GFP_KERNEL | __GFP_ZERO,
1660 cpu_to_node(cpu));
1661 if (!base)
1662 return -ENOMEM;
1664 /* Make sure that tvec_base is 2 byte aligned */
1665 if (tbase_get_deferrable(base)) {
1666 WARN_ON(1);
1667 kfree(base);
1668 return -ENOMEM;
1670 per_cpu(tvec_bases, cpu) = base;
1671 } else {
1673 * This is for the boot CPU - we use compile-time
1674 * static initialisation because per-cpu memory isn't
1675 * ready yet and because the memory allocators are not
1676 * initialised either.
1678 boot_done = 1;
1679 base = &boot_tvec_bases;
1681 tvec_base_done[cpu] = 1;
1682 } else {
1683 base = per_cpu(tvec_bases, cpu);
1686 spin_lock_init(&base->lock);
1688 for (j = 0; j < TVN_SIZE; j++) {
1689 INIT_LIST_HEAD(base->tv5.vec + j);
1690 INIT_LIST_HEAD(base->tv4.vec + j);
1691 INIT_LIST_HEAD(base->tv3.vec + j);
1692 INIT_LIST_HEAD(base->tv2.vec + j);
1694 for (j = 0; j < TVR_SIZE; j++)
1695 INIT_LIST_HEAD(base->tv1.vec + j);
1697 base->timer_jiffies = jiffies;
1698 base->next_timer = base->timer_jiffies;
1699 return 0;
1702 #ifdef CONFIG_HOTPLUG_CPU
1703 static void migrate_timer_list(struct tvec_base *new_base, struct list_head *head)
1705 struct timer_list *timer;
1707 while (!list_empty(head)) {
1708 timer = list_first_entry(head, struct timer_list, entry);
1709 detach_timer(timer, 0);
1710 timer_set_base(timer, new_base);
1711 if (time_before(timer->expires, new_base->next_timer) &&
1712 !tbase_get_deferrable(timer->base))
1713 new_base->next_timer = timer->expires;
1714 internal_add_timer(new_base, timer);
1718 static void __cpuinit migrate_timers(int cpu)
1720 struct tvec_base *old_base;
1721 struct tvec_base *new_base;
1722 int i;
1724 BUG_ON(cpu_online(cpu));
1725 old_base = per_cpu(tvec_bases, cpu);
1726 new_base = get_cpu_var(tvec_bases);
1728 * The caller is globally serialized and nobody else
1729 * takes two locks at once, deadlock is not possible.
1731 spin_lock_irq(&new_base->lock);
1732 spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1734 BUG_ON(old_base->running_timer);
1736 for (i = 0; i < TVR_SIZE; i++)
1737 migrate_timer_list(new_base, old_base->tv1.vec + i);
1738 for (i = 0; i < TVN_SIZE; i++) {
1739 migrate_timer_list(new_base, old_base->tv2.vec + i);
1740 migrate_timer_list(new_base, old_base->tv3.vec + i);
1741 migrate_timer_list(new_base, old_base->tv4.vec + i);
1742 migrate_timer_list(new_base, old_base->tv5.vec + i);
1745 spin_unlock(&old_base->lock);
1746 spin_unlock_irq(&new_base->lock);
1747 put_cpu_var(tvec_bases);
1749 #endif /* CONFIG_HOTPLUG_CPU */
1751 static int __cpuinit timer_cpu_notify(struct notifier_block *self,
1752 unsigned long action, void *hcpu)
1754 long cpu = (long)hcpu;
1755 int err;
1757 switch(action) {
1758 case CPU_UP_PREPARE:
1759 case CPU_UP_PREPARE_FROZEN:
1760 err = init_timers_cpu(cpu);
1761 if (err < 0)
1762 return notifier_from_errno(err);
1763 break;
1764 #ifdef CONFIG_HOTPLUG_CPU
1765 case CPU_DEAD:
1766 case CPU_DEAD_FROZEN:
1767 migrate_timers(cpu);
1768 break;
1769 #endif
1770 default:
1771 break;
1773 return NOTIFY_OK;
1776 static struct notifier_block __cpuinitdata timers_nb = {
1777 .notifier_call = timer_cpu_notify,
1781 void __init init_timers(void)
1783 int err = timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE,
1784 (void *)(long)smp_processor_id());
1786 init_timer_stats();
1788 BUG_ON(err != NOTIFY_OK);
1789 register_cpu_notifier(&timers_nb);
1790 open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
1794 * msleep - sleep safely even with waitqueue interruptions
1795 * @msecs: Time in milliseconds to sleep for
1797 void msleep(unsigned int msecs)
1799 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1801 while (timeout)
1802 timeout = schedule_timeout_uninterruptible(timeout);
1805 EXPORT_SYMBOL(msleep);
1808 * msleep_interruptible - sleep waiting for signals
1809 * @msecs: Time in milliseconds to sleep for
1811 unsigned long msleep_interruptible(unsigned int msecs)
1813 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1815 while (timeout && !signal_pending(current))
1816 timeout = schedule_timeout_interruptible(timeout);
1817 return jiffies_to_msecs(timeout);
1820 EXPORT_SYMBOL(msleep_interruptible);
1822 static int __sched do_usleep_range(unsigned long min, unsigned long max)
1824 ktime_t kmin;
1825 unsigned long delta;
1827 kmin = ktime_set(0, min * NSEC_PER_USEC);
1828 delta = (max - min) * NSEC_PER_USEC;
1829 return schedule_hrtimeout_range(&kmin, delta, HRTIMER_MODE_REL);
1833 * usleep_range - Drop in replacement for udelay where wakeup is flexible
1834 * @min: Minimum time in usecs to sleep
1835 * @max: Maximum time in usecs to sleep
1837 void usleep_range(unsigned long min, unsigned long max)
1839 __set_current_state(TASK_UNINTERRUPTIBLE);
1840 do_usleep_range(min, max);
1842 EXPORT_SYMBOL(usleep_range);