projects/linux·blob v3.2-rc4 · refs

1792 lines · 47.8 KiB

1/*
2 * linux/kernel/timer.c
3 *
4 * Kernel internal timers, basic process system calls
5 *
6 * Copyright (C) 1991, 1992 Linus Torvalds
7 *
8 * 1997-01-28 Modified by Finn Arne Gangstad to make timers scale better.
9 *
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
20 */
21
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>
43
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>
49
50#define CREATE_TRACE_POINTS
51#include <trace/events/timer.h>
52
53u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
54
55EXPORT_SYMBOL(jiffies_64);
56
57/*
58 * per-CPU timer vector definitions:
59 */
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)
66
67struct tvec {
68 struct list_head vec[TVN_SIZE];
69};
70
71struct tvec_root {
72 struct list_head vec[TVR_SIZE];
73};
74
75struct 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;
86
87struct tvec_base boot_tvec_bases;
88EXPORT_SYMBOL(boot_tvec_bases);
89static DEFINE_PER_CPU(struct tvec_base *, tvec_bases) = &boot_tvec_bases;
90
91/* Functions below help us manage 'deferrable' flag */
92static inline unsigned int tbase_get_deferrable(struct tvec_base *base)
93{
94 return ((unsigned int)(unsigned long)base & TBASE_DEFERRABLE_FLAG);
95}
96
97static inline struct tvec_base *tbase_get_base(struct tvec_base *base)
98{
99 return ((struct tvec_base *)((unsigned long)base & ~TBASE_DEFERRABLE_FLAG));
100}
101
102static inline void timer_set_deferrable(struct timer_list *timer)
103{
104 timer->base = TBASE_MAKE_DEFERRED(timer->base);
105}
106
107static inline void
108timer_set_base(struct timer_list *timer, struct tvec_base *new_base)
109{
110 timer->base = (struct tvec_base *)((unsigned long)(new_base) |
111 tbase_get_deferrable(timer->base));
112}
113
114static unsigned long round_jiffies_common(unsigned long j, int cpu,
115 bool force_up)
116{
117 int rem;
118 unsigned long original = j;
119
120 /*
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.
127 */
128 j += cpu * 3;
129
130 rem = j % HZ;
131
132 /*
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.
138 */
139 if (rem < HZ/4 && !force_up) /* round down */
140 j = j - rem;
141 else /* round up */
142 j = j - rem + HZ;
143
144 /* now that we have rounded, subtract the extra skew again */
145 j -= cpu * 3;
146
147 if (j <= jiffies) /* rounding ate our timeout entirely; */
148 return original;
149 return j;
150}
151
152/**
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
156 *
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.
161 *
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.
165 *
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.
169 *
170 * The return value is the rounded version of the @j parameter.
171 */
172unsigned long __round_jiffies(unsigned long j, int cpu)
173{
174 return round_jiffies_common(j, cpu, false);
175}
176EXPORT_SYMBOL_GPL(__round_jiffies);
177
178/**
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
182 *
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.
187 *
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.
191 *
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.
195 *
196 * The return value is the rounded version of the @j parameter.
197 */
198unsigned long __round_jiffies_relative(unsigned long j, int cpu)
199{
200 unsigned long j0 = jiffies;
201
202 /* Use j0 because jiffies might change while we run */
203 return round_jiffies_common(j + j0, cpu, false) - j0;
204}
205EXPORT_SYMBOL_GPL(__round_jiffies_relative);
206
207/**
208 * round_jiffies - function to round jiffies to a full second
209 * @j: the time in (absolute) jiffies that should be rounded
210 *
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.
215 *
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.
219 *
220 * The return value is the rounded version of the @j parameter.
221 */
222unsigned long round_jiffies(unsigned long j)
223{
224 return round_jiffies_common(j, raw_smp_processor_id(), false);
225}
226EXPORT_SYMBOL_GPL(round_jiffies);
227
228/**
229 * round_jiffies_relative - function to round jiffies to a full second
230 * @j: the time in (relative) jiffies that should be rounded
231 *
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.
236 *
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.
240 *
241 * The return value is the rounded version of the @j parameter.
242 */
243unsigned long round_jiffies_relative(unsigned long j)
244{
245 return __round_jiffies_relative(j, raw_smp_processor_id());
246}
247EXPORT_SYMBOL_GPL(round_jiffies_relative);
248
249/**
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
253 *
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.
258 */
259unsigned long __round_jiffies_up(unsigned long j, int cpu)
260{
261 return round_jiffies_common(j, cpu, true);
262}
263EXPORT_SYMBOL_GPL(__round_jiffies_up);
264
265/**
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
269 *
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.
274 */
275unsigned long __round_jiffies_up_relative(unsigned long j, int cpu)
276{
277 unsigned long j0 = jiffies;
278
279 /* Use j0 because jiffies might change while we run */
280 return round_jiffies_common(j + j0, cpu, true) - j0;
281}
282EXPORT_SYMBOL_GPL(__round_jiffies_up_relative);
283
284/**
285 * round_jiffies_up - function to round jiffies up to a full second
286 * @j: the time in (absolute) jiffies that should be rounded
287 *
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.
292 */
293unsigned long round_jiffies_up(unsigned long j)
294{
295 return round_jiffies_common(j, raw_smp_processor_id(), true);
296}
297EXPORT_SYMBOL_GPL(round_jiffies_up);
298
299/**
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
302 *
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.
307 */
308unsigned long round_jiffies_up_relative(unsigned long j)
309{
310 return __round_jiffies_up_relative(j, raw_smp_processor_id());
311}
312EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
313
314/**
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
318 *
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.
323 *
324 * By setting the slack to -1, a percentage of the delay is used
325 * instead.
326 */
327void set_timer_slack(struct timer_list *timer, int slack_hz)
328{
329 timer->slack = slack_hz;
330}
331EXPORT_SYMBOL_GPL(set_timer_slack);
332
333static void internal_add_timer(struct tvec_base *base, struct timer_list *timer)
334{
335 unsigned long expires = timer->expires;
336 unsigned long idx = expires - base->timer_jiffies;
337 struct list_head *vec;
338
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) {
352 /*
353 * Can happen if you add a timer with expires == jiffies,
354 * or you set a timer to go off in the past
355 */
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:
361 */
362 if (idx > 0xffffffffUL) {
363 idx = 0xffffffffUL;
364 expires = idx + base->timer_jiffies;
365 }
366 i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK;
367 vec = base->tv5.vec + i;
368 }
369 /*
370 * Timers are FIFO:
371 */
372 list_add_tail(&timer->entry, vec);
373}
374
375#ifdef CONFIG_TIMER_STATS
376void __timer_stats_timer_set_start_info(struct timer_list *timer, void *addr)
377{
378 if (timer->start_site)
379 return;
380
381 timer->start_site = addr;
382 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
383 timer->start_pid = current->pid;
384}
385
386static void timer_stats_account_timer(struct timer_list *timer)
387{
388 unsigned int flag = 0;
389
390 if (likely(!timer->start_site))
391 return;
392 if (unlikely(tbase_get_deferrable(timer->base)))
393 flag |= TIMER_STATS_FLAG_DEFERRABLE;
394
395 timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
396 timer->function, timer->start_comm, flag);
397}
398
399#else
400static void timer_stats_account_timer(struct timer_list *timer) {}
401#endif
402
403#ifdef CONFIG_DEBUG_OBJECTS_TIMERS
404
405static struct debug_obj_descr timer_debug_descr;
406
407static void *timer_debug_hint(void *addr)
408{
409 return ((struct timer_list *) addr)->function;
410}
411
412/*
413 * fixup_init is called when:
414 * - an active object is initialized
415 */
416static int timer_fixup_init(void *addr, enum debug_obj_state state)
417{
418 struct timer_list *timer = addr;
419
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;
427 }
428}
429
430/*
431 * fixup_activate is called when:
432 * - an active object is activated
433 * - an unknown object is activated (might be a statically initialized object)
434 */
435static int timer_fixup_activate(void *addr, enum debug_obj_state state)
436{
437 struct timer_list *timer = addr;
438
439 switch (state) {
440
441 case ODEBUG_STATE_NOTAVAILABLE:
442 /*
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.
446 */
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);
454 }
455 return 0;
456
457 case ODEBUG_STATE_ACTIVE:
458 WARN_ON(1);
459
460 default:
461 return 0;
462 }
463}
464
465/*
466 * fixup_free is called when:
467 * - an active object is freed
468 */
469static int timer_fixup_free(void *addr, enum debug_obj_state state)
470{
471 struct timer_list *timer = addr;
472
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;
480 }
481}
482
483static 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,
489};
490
491static inline void debug_timer_init(struct timer_list *timer)
492{
493 debug_object_init(timer, &timer_debug_descr);
494}
495
496static inline void debug_timer_activate(struct timer_list *timer)
497{
498 debug_object_activate(timer, &timer_debug_descr);
499}
500
501static inline void debug_timer_deactivate(struct timer_list *timer)
502{
503 debug_object_deactivate(timer, &timer_debug_descr);
504}
505
506static inline void debug_timer_free(struct timer_list *timer)
507{
508 debug_object_free(timer, &timer_debug_descr);
509}
510
511static void __init_timer(struct timer_list *timer,
512 const char *name,
513 struct lock_class_key *key);
514
515void init_timer_on_stack_key(struct timer_list *timer,
516 const char *name,
517 struct lock_class_key *key)
518{
519 debug_object_init_on_stack(timer, &timer_debug_descr);
520 __init_timer(timer, name, key);
521}
522EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
523
524void destroy_timer_on_stack(struct timer_list *timer)
525{
526 debug_object_free(timer, &timer_debug_descr);
527}
528EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
529
530#else
531static inline void debug_timer_init(struct timer_list *timer) { }
532static inline void debug_timer_activate(struct timer_list *timer) { }
533static inline void debug_timer_deactivate(struct timer_list *timer) { }
534#endif
535
536static inline void debug_init(struct timer_list *timer)
537{
538 debug_timer_init(timer);
539 trace_timer_init(timer);
540}
541
542static inline void
543debug_activate(struct timer_list *timer, unsigned long expires)
544{
545 debug_timer_activate(timer);
546 trace_timer_start(timer, expires);
547}
548
549static inline void debug_deactivate(struct timer_list *timer)
550{
551 debug_timer_deactivate(timer);
552 trace_timer_cancel(timer);
553}
554
555static void __init_timer(struct timer_list *timer,
556 const char *name,
557 struct lock_class_key *key)
558{
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);
568}
569
570void 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)
575{
576 timer->function = function;
577 timer->data = data;
578 init_timer_on_stack_key(timer, name, key);
579 timer_set_deferrable(timer);
580}
581EXPORT_SYMBOL_GPL(setup_deferrable_timer_on_stack_key);
582
583/**
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
589 *
590 * init_timer_key() must be done to a timer prior calling *any* of the
591 * other timer functions.
592 */
593void init_timer_key(struct timer_list *timer,
594 const char *name,
595 struct lock_class_key *key)
596{
597 debug_init(timer);
598 __init_timer(timer, name, key);
599}
600EXPORT_SYMBOL(init_timer_key);
601
602void init_timer_deferrable_key(struct timer_list *timer,
603 const char *name,
604 struct lock_class_key *key)
605{
606 init_timer_key(timer, name, key);
607 timer_set_deferrable(timer);
608}
609EXPORT_SYMBOL(init_timer_deferrable_key);
610
611static inline void detach_timer(struct timer_list *timer,
612 int clear_pending)
613{
614 struct list_head *entry = &timer->entry;
615
616 debug_deactivate(timer);
617
618 __list_del(entry->prev, entry->next);
619 if (clear_pending)
620 entry->next = NULL;
621 entry->prev = LIST_POISON2;
622}
623
624/*
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.
628 *
629 * So __run_timers/migrate_timers can safely modify all timers which could
630 * be found on ->tvX lists.
631 *
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.
635 */
636static struct tvec_base *lock_timer_base(struct timer_list *timer,
637 unsigned long *flags)
638 __acquires(timer->base->lock)
639{
640 struct tvec_base *base;
641
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);
651 }
652 cpu_relax();
653 }
654}
655
656static inline int
657__mod_timer(struct timer_list *timer, unsigned long expires,
658 bool pending_only, int pinned)
659{
660 struct tvec_base *base, *new_base;
661 unsigned long flags;
662 int ret = 0 , cpu;
663
664 timer_stats_timer_set_start_info(timer);
665 BUG_ON(!timer->function);
666
667 base = lock_timer_base(timer, &flags);
668
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;
678 }
679
680 debug_activate(timer, expires);
681
682 cpu = smp_processor_id();
683
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);
689
690 if (base != new_base) {
691 /*
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.
697 */
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);
705 }
706 }
707
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);
713
714out_unlock:
715 spin_unlock_irqrestore(&base->lock, flags);
716
717 return ret;
718}
719
720/**
721 * mod_timer_pending - modify a pending timer's timeout
722 * @timer: the pending timer to be modified
723 * @expires: new timeout in jiffies
724 *
725 * mod_timer_pending() is the same for pending timers as mod_timer(),
726 * but will not re-activate and modify already deleted timers.
727 *
728 * It is useful for unserialized use of timers.
729 */
730int mod_timer_pending(struct timer_list *timer, unsigned long expires)
731{
732 return __mod_timer(timer, expires, true, TIMER_NOT_PINNED);
733}
734EXPORT_SYMBOL(mod_timer_pending);
735
736/*
737 * Decide where to put the timer while taking the slack into account
738 *
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
745 */
746static inline
747unsigned long apply_slack(struct timer_list *timer, unsigned long expires)
748{
749 unsigned long expires_limit, mask;
750 int bit;
751
752 if (timer->slack >= 0) {
753 expires_limit = expires + timer->slack;
754 } else {
755 long delta = expires - jiffies;
756
757 if (delta < 256)
758 return expires;
759
760 expires_limit = expires + delta / 256;
761 }
762 mask = expires ^ expires_limit;
763 if (mask == 0)
764 return expires;
765
766 bit = find_last_bit(&mask, BITS_PER_LONG);
767
768 mask = (1 << bit) - 1;
769
770 expires_limit = expires_limit & ~(mask);
771
772 return expires_limit;
773}
774
775/**
776 * mod_timer - modify a timer's timeout
777 * @timer: the timer to be modified
778 * @expires: new timeout in jiffies
779 *
780 * mod_timer() is a more efficient way to update the expire field of an
781 * active timer (if the timer is inactive it will be activated)
782 *
783 * mod_timer(timer, expires) is equivalent to:
784 *
785 * del_timer(timer); timer->expires = expires; add_timer(timer);
786 *
787 * Note that if there are multiple unserialized concurrent users of the
788 * same timer, then mod_timer() is the only safe way to modify the timeout,
789 * since add_timer() cannot modify an already running timer.
790 *
791 * The function returns whether it has modified a pending timer or not.
792 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
793 * active timer returns 1.)
794 */
795int mod_timer(struct timer_list *timer, unsigned long expires)
796{
797 expires = apply_slack(timer, expires);
798
799 /*
800 * This is a common optimization triggered by the
801 * networking code - if the timer is re-modified
802 * to be the same thing then just return:
803 */
804 if (timer_pending(timer) && timer->expires == expires)
805 return 1;
806
807 return __mod_timer(timer, expires, false, TIMER_NOT_PINNED);
808}
809EXPORT_SYMBOL(mod_timer);
810
811/**
812 * mod_timer_pinned - modify a timer's timeout
813 * @timer: the timer to be modified
814 * @expires: new timeout in jiffies
815 *
816 * mod_timer_pinned() is a way to update the expire field of an
817 * active timer (if the timer is inactive it will be activated)
818 * and not allow the timer to be migrated to a different CPU.
819 *
820 * mod_timer_pinned(timer, expires) is equivalent to:
821 *
822 * del_timer(timer); timer->expires = expires; add_timer(timer);
823 */
824int mod_timer_pinned(struct timer_list *timer, unsigned long expires)
825{
826 if (timer->expires == expires && timer_pending(timer))
827 return 1;
828
829 return __mod_timer(timer, expires, false, TIMER_PINNED);
830}
831EXPORT_SYMBOL(mod_timer_pinned);
832
833/**
834 * add_timer - start a timer
835 * @timer: the timer to be added
836 *
837 * The kernel will do a ->function(->data) callback from the
838 * timer interrupt at the ->expires point in the future. The
839 * current time is 'jiffies'.
840 *
841 * The timer's ->expires, ->function (and if the handler uses it, ->data)
842 * fields must be set prior calling this function.
843 *
844 * Timers with an ->expires field in the past will be executed in the next
845 * timer tick.
846 */
847void add_timer(struct timer_list *timer)
848{
849 BUG_ON(timer_pending(timer));
850 mod_timer(timer, timer->expires);
851}
852EXPORT_SYMBOL(add_timer);
853
854/**
855 * add_timer_on - start a timer on a particular CPU
856 * @timer: the timer to be added
857 * @cpu: the CPU to start it on
858 *
859 * This is not very scalable on SMP. Double adds are not possible.
860 */
861void add_timer_on(struct timer_list *timer, int cpu)
862{
863 struct tvec_base *base = per_cpu(tvec_bases, cpu);
864 unsigned long flags;
865
866 timer_stats_timer_set_start_info(timer);
867 BUG_ON(timer_pending(timer) || !timer->function);
868 spin_lock_irqsave(&base->lock, flags);
869 timer_set_base(timer, base);
870 debug_activate(timer, timer->expires);
871 if (time_before(timer->expires, base->next_timer) &&
872 !tbase_get_deferrable(timer->base))
873 base->next_timer = timer->expires;
874 internal_add_timer(base, timer);
875 /*
876 * Check whether the other CPU is idle and needs to be
877 * triggered to reevaluate the timer wheel when nohz is
878 * active. We are protected against the other CPU fiddling
879 * with the timer by holding the timer base lock. This also
880 * makes sure that a CPU on the way to idle can not evaluate
881 * the timer wheel.
882 */
883 wake_up_idle_cpu(cpu);
884 spin_unlock_irqrestore(&base->lock, flags);
885}
886EXPORT_SYMBOL_GPL(add_timer_on);
887
888/**
889 * del_timer - deactive a timer.
890 * @timer: the timer to be deactivated
891 *
892 * del_timer() deactivates a timer - this works on both active and inactive
893 * timers.
894 *
895 * The function returns whether it has deactivated a pending timer or not.
896 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
897 * active timer returns 1.)
898 */
899int del_timer(struct timer_list *timer)
900{
901 struct tvec_base *base;
902 unsigned long flags;
903 int ret = 0;
904
905 timer_stats_timer_clear_start_info(timer);
906 if (timer_pending(timer)) {
907 base = lock_timer_base(timer, &flags);
908 if (timer_pending(timer)) {
909 detach_timer(timer, 1);
910 if (timer->expires == base->next_timer &&
911 !tbase_get_deferrable(timer->base))
912 base->next_timer = base->timer_jiffies;
913 ret = 1;
914 }
915 spin_unlock_irqrestore(&base->lock, flags);
916 }
917
918 return ret;
919}
920EXPORT_SYMBOL(del_timer);
921
922/**
923 * try_to_del_timer_sync - Try to deactivate a timer
924 * @timer: timer do del
925 *
926 * This function tries to deactivate a timer. Upon successful (ret >= 0)
927 * exit the timer is not queued and the handler is not running on any CPU.
928 */
929int try_to_del_timer_sync(struct timer_list *timer)
930{
931 struct tvec_base *base;
932 unsigned long flags;
933 int ret = -1;
934
935 base = lock_timer_base(timer, &flags);
936
937 if (base->running_timer == timer)
938 goto out;
939
940 timer_stats_timer_clear_start_info(timer);
941 ret = 0;
942 if (timer_pending(timer)) {
943 detach_timer(timer, 1);
944 if (timer->expires == base->next_timer &&
945 !tbase_get_deferrable(timer->base))
946 base->next_timer = base->timer_jiffies;
947 ret = 1;
948 }
949out:
950 spin_unlock_irqrestore(&base->lock, flags);
951
952 return ret;
953}
954EXPORT_SYMBOL(try_to_del_timer_sync);
955
956#ifdef CONFIG_SMP
957/**
958 * del_timer_sync - deactivate a timer and wait for the handler to finish.
959 * @timer: the timer to be deactivated
960 *
961 * This function only differs from del_timer() on SMP: besides deactivating
962 * the timer it also makes sure the handler has finished executing on other
963 * CPUs.
964 *
965 * Synchronization rules: Callers must prevent restarting of the timer,
966 * otherwise this function is meaningless. It must not be called from
967 * interrupt contexts. The caller must not hold locks which would prevent
968 * completion of the timer's handler. The timer's handler must not call
969 * add_timer_on(). Upon exit the timer is not queued and the handler is
970 * not running on any CPU.
971 *
972 * Note: You must not hold locks that are held in interrupt context
973 * while calling this function. Even if the lock has nothing to do
974 * with the timer in question. Here's why:
975 *
976 * CPU0 CPU1
977 * ---- ----
978 * <SOFTIRQ>
979 * call_timer_fn();
980 * base->running_timer = mytimer;
981 * spin_lock_irq(somelock);
982 * <IRQ>
983 * spin_lock(somelock);
984 * del_timer_sync(mytimer);
985 * while (base->running_timer == mytimer);
986 *
987 * Now del_timer_sync() will never return and never release somelock.
988 * The interrupt on the other CPU is waiting to grab somelock but
989 * it has interrupted the softirq that CPU0 is waiting to finish.
990 *
991 * The function returns whether it has deactivated a pending timer or not.
992 */
993int del_timer_sync(struct timer_list *timer)
994{
995#ifdef CONFIG_LOCKDEP
996 unsigned long flags;
997
998 /*
999 * If lockdep gives a backtrace here, please reference
1000 * the synchronization rules above.
1001 */
1002 local_irq_save(flags);
1003 lock_map_acquire(&timer->lockdep_map);
1004 lock_map_release(&timer->lockdep_map);
1005 local_irq_restore(flags);
1006#endif
1007 /*
1008 * don't use it in hardirq context, because it
1009 * could lead to deadlock.
1010 */
1011 WARN_ON(in_irq());
1012 for (;;) {
1013 int ret = try_to_del_timer_sync(timer);
1014 if (ret >= 0)
1015 return ret;
1016 cpu_relax();
1017 }
1018}
1019EXPORT_SYMBOL(del_timer_sync);
1020#endif
1021
1022static int cascade(struct tvec_base *base, struct tvec *tv, int index)
1023{
1024 /* cascade all the timers from tv up one level */
1025 struct timer_list *timer, *tmp;
1026 struct list_head tv_list;
1027
1028 list_replace_init(tv->vec + index, &tv_list);
1029
1030 /*
1031 * We are removing _all_ timers from the list, so we
1032 * don't have to detach them individually.
1033 */
1034 list_for_each_entry_safe(timer, tmp, &tv_list, entry) {
1035 BUG_ON(tbase_get_base(timer->base) != base);
1036 internal_add_timer(base, timer);
1037 }
1038
1039 return index;
1040}
1041
1042static void call_timer_fn(struct timer_list *timer, void (*fn)(unsigned long),
1043 unsigned long data)
1044{
1045 int preempt_count = preempt_count();
1046
1047#ifdef CONFIG_LOCKDEP
1048 /*
1049 * It is permissible to free the timer from inside the
1050 * function that is called from it, this we need to take into
1051 * account for lockdep too. To avoid bogus "held lock freed"
1052 * warnings as well as problems when looking into
1053 * timer->lockdep_map, make a copy and use that here.
1054 */
1055 struct lockdep_map lockdep_map = timer->lockdep_map;
1056#endif
1057 /*
1058 * Couple the lock chain with the lock chain at
1059 * del_timer_sync() by acquiring the lock_map around the fn()
1060 * call here and in del_timer_sync().
1061 */
1062 lock_map_acquire(&lockdep_map);
1063
1064 trace_timer_expire_entry(timer);
1065 fn(data);
1066 trace_timer_expire_exit(timer);
1067
1068 lock_map_release(&lockdep_map);
1069
1070 if (preempt_count != preempt_count()) {
1071 WARN_ONCE(1, "timer: %pF preempt leak: %08x -> %08x\n",
1072 fn, preempt_count, preempt_count());
1073 /*
1074 * Restore the preempt count. That gives us a decent
1075 * chance to survive and extract information. If the
1076 * callback kept a lock held, bad luck, but not worse
1077 * than the BUG() we had.
1078 */
1079 preempt_count() = preempt_count;
1080 }
1081}
1082
1083#define INDEX(N) ((base->timer_jiffies >> (TVR_BITS + (N) * TVN_BITS)) & TVN_MASK)
1084
1085/**
1086 * __run_timers - run all expired timers (if any) on this CPU.
1087 * @base: the timer vector to be processed.
1088 *
1089 * This function cascades all vectors and executes all expired timer
1090 * vectors.
1091 */
1092static inline void __run_timers(struct tvec_base *base)
1093{
1094 struct timer_list *timer;
1095
1096 spin_lock_irq(&base->lock);
1097 while (time_after_eq(jiffies, base->timer_jiffies)) {
1098 struct list_head work_list;
1099 struct list_head *head = &work_list;
1100 int index = base->timer_jiffies & TVR_MASK;
1101
1102 /*
1103 * Cascade timers:
1104 */
1105 if (!index &&
1106 (!cascade(base, &base->tv2, INDEX(0))) &&
1107 (!cascade(base, &base->tv3, INDEX(1))) &&
1108 !cascade(base, &base->tv4, INDEX(2)))
1109 cascade(base, &base->tv5, INDEX(3));
1110 ++base->timer_jiffies;
1111 list_replace_init(base->tv1.vec + index, &work_list);
1112 while (!list_empty(head)) {
1113 void (*fn)(unsigned long);
1114 unsigned long data;
1115
1116 timer = list_first_entry(head, struct timer_list,entry);
1117 fn = timer->function;
1118 data = timer->data;
1119
1120 timer_stats_account_timer(timer);
1121
1122 base->running_timer = timer;
1123 detach_timer(timer, 1);
1124
1125 spin_unlock_irq(&base->lock);
1126 call_timer_fn(timer, fn, data);
1127 spin_lock_irq(&base->lock);
1128 }
1129 }
1130 base->running_timer = NULL;
1131 spin_unlock_irq(&base->lock);
1132}
1133
1134#ifdef CONFIG_NO_HZ
1135/*
1136 * Find out when the next timer event is due to happen. This
1137 * is used on S/390 to stop all activity when a CPU is idle.
1138 * This function needs to be called with interrupts disabled.
1139 */
1140static unsigned long __next_timer_interrupt(struct tvec_base *base)
1141{
1142 unsigned long timer_jiffies = base->timer_jiffies;
1143 unsigned long expires = timer_jiffies + NEXT_TIMER_MAX_DELTA;
1144 int index, slot, array, found = 0;
1145 struct timer_list *nte;
1146 struct tvec *varray[4];
1147
1148 /* Look for timer events in tv1. */
1149 index = slot = timer_jiffies & TVR_MASK;
1150 do {
1151 list_for_each_entry(nte, base->tv1.vec + slot, entry) {
1152 if (tbase_get_deferrable(nte->base))
1153 continue;
1154
1155 found = 1;
1156 expires = nte->expires;
1157 /* Look at the cascade bucket(s)? */
1158 if (!index || slot < index)
1159 goto cascade;
1160 return expires;
1161 }
1162 slot = (slot + 1) & TVR_MASK;
1163 } while (slot != index);
1164
1165cascade:
1166 /* Calculate the next cascade event */
1167 if (index)
1168 timer_jiffies += TVR_SIZE - index;
1169 timer_jiffies >>= TVR_BITS;
1170
1171 /* Check tv2-tv5. */
1172 varray[0] = &base->tv2;
1173 varray[1] = &base->tv3;
1174 varray[2] = &base->tv4;
1175 varray[3] = &base->tv5;
1176
1177 for (array = 0; array < 4; array++) {
1178 struct tvec *varp = varray[array];
1179
1180 index = slot = timer_jiffies & TVN_MASK;
1181 do {
1182 list_for_each_entry(nte, varp->vec + slot, entry) {
1183 if (tbase_get_deferrable(nte->base))
1184 continue;
1185
1186 found = 1;
1187 if (time_before(nte->expires, expires))
1188 expires = nte->expires;
1189 }
1190 /*
1191 * Do we still search for the first timer or are
1192 * we looking up the cascade buckets ?
1193 */
1194 if (found) {
1195 /* Look at the cascade bucket(s)? */
1196 if (!index || slot < index)
1197 break;
1198 return expires;
1199 }
1200 slot = (slot + 1) & TVN_MASK;
1201 } while (slot != index);
1202
1203 if (index)
1204 timer_jiffies += TVN_SIZE - index;
1205 timer_jiffies >>= TVN_BITS;
1206 }
1207 return expires;
1208}
1209
1210/*
1211 * Check, if the next hrtimer event is before the next timer wheel
1212 * event:
1213 */
1214static unsigned long cmp_next_hrtimer_event(unsigned long now,
1215 unsigned long expires)
1216{
1217 ktime_t hr_delta = hrtimer_get_next_event();
1218 struct timespec tsdelta;
1219 unsigned long delta;
1220
1221 if (hr_delta.tv64 == KTIME_MAX)
1222 return expires;
1223
1224 /*
1225 * Expired timer available, let it expire in the next tick
1226 */
1227 if (hr_delta.tv64 <= 0)
1228 return now + 1;
1229
1230 tsdelta = ktime_to_timespec(hr_delta);
1231 delta = timespec_to_jiffies(&tsdelta);
1232
1233 /*
1234 * Limit the delta to the max value, which is checked in
1235 * tick_nohz_stop_sched_tick():
1236 */
1237 if (delta > NEXT_TIMER_MAX_DELTA)
1238 delta = NEXT_TIMER_MAX_DELTA;
1239
1240 /*
1241 * Take rounding errors in to account and make sure, that it
1242 * expires in the next tick. Otherwise we go into an endless
1243 * ping pong due to tick_nohz_stop_sched_tick() retriggering
1244 * the timer softirq
1245 */
1246 if (delta < 1)
1247 delta = 1;
1248 now += delta;
1249 if (time_before(now, expires))
1250 return now;
1251 return expires;
1252}
1253
1254/**
1255 * get_next_timer_interrupt - return the jiffy of the next pending timer
1256 * @now: current time (in jiffies)
1257 */
1258unsigned long get_next_timer_interrupt(unsigned long now)
1259{
1260 struct tvec_base *base = __this_cpu_read(tvec_bases);
1261 unsigned long expires;
1262
1263 /*
1264 * Pretend that there is no timer pending if the cpu is offline.
1265 * Possible pending timers will be migrated later to an active cpu.
1266 */
1267 if (cpu_is_offline(smp_processor_id()))
1268 return now + NEXT_TIMER_MAX_DELTA;
1269 spin_lock(&base->lock);
1270 if (time_before_eq(base->next_timer, base->timer_jiffies))
1271 base->next_timer = __next_timer_interrupt(base);
1272 expires = base->next_timer;
1273 spin_unlock(&base->lock);
1274
1275 if (time_before_eq(expires, now))
1276 return now;
1277
1278 return cmp_next_hrtimer_event(now, expires);
1279}
1280#endif
1281
1282/*
1283 * Called from the timer interrupt handler to charge one tick to the current
1284 * process. user_tick is 1 if the tick is user time, 0 for system.
1285 */
1286void update_process_times(int user_tick)
1287{
1288 struct task_struct *p = current;
1289 int cpu = smp_processor_id();
1290
1291 /* Note: this timer irq context must be accounted for as well. */
1292 account_process_tick(p, user_tick);
1293 run_local_timers();
1294 rcu_check_callbacks(cpu, user_tick);
1295 printk_tick();
1296#ifdef CONFIG_IRQ_WORK
1297 if (in_irq())
1298 irq_work_run();
1299#endif
1300 scheduler_tick();
1301 run_posix_cpu_timers(p);
1302}
1303
1304/*
1305 * This function runs timers and the timer-tq in bottom half context.
1306 */
1307static void run_timer_softirq(struct softirq_action *h)
1308{
1309 struct tvec_base *base = __this_cpu_read(tvec_bases);
1310
1311 hrtimer_run_pending();
1312
1313 if (time_after_eq(jiffies, base->timer_jiffies))
1314 __run_timers(base);
1315}
1316
1317/*
1318 * Called by the local, per-CPU timer interrupt on SMP.
1319 */
1320void run_local_timers(void)
1321{
1322 hrtimer_run_queues();
1323 raise_softirq(TIMER_SOFTIRQ);
1324}
1325
1326#ifdef __ARCH_WANT_SYS_ALARM
1327
1328/*
1329 * For backwards compatibility? This can be done in libc so Alpha
1330 * and all newer ports shouldn't need it.
1331 */
1332SYSCALL_DEFINE1(alarm, unsigned int, seconds)
1333{
1334 return alarm_setitimer(seconds);
1335}
1336
1337#endif
1338
1339#ifndef __alpha__
1340
1341/*
1342 * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this
1343 * should be moved into arch/i386 instead?
1344 */
1345
1346/**
1347 * sys_getpid - return the thread group id of the current process
1348 *
1349 * Note, despite the name, this returns the tgid not the pid. The tgid and
1350 * the pid are identical unless CLONE_THREAD was specified on clone() in
1351 * which case the tgid is the same in all threads of the same group.
1352 *
1353 * This is SMP safe as current->tgid does not change.
1354 */
1355SYSCALL_DEFINE0(getpid)
1356{
1357 return task_tgid_vnr(current);
1358}
1359
1360/*
1361 * Accessing ->real_parent is not SMP-safe, it could
1362 * change from under us. However, we can use a stale
1363 * value of ->real_parent under rcu_read_lock(), see
1364 * release_task()->call_rcu(delayed_put_task_struct).
1365 */
1366SYSCALL_DEFINE0(getppid)
1367{
1368 int pid;
1369
1370 rcu_read_lock();
1371 pid = task_tgid_vnr(current->real_parent);
1372 rcu_read_unlock();
1373
1374 return pid;
1375}
1376
1377SYSCALL_DEFINE0(getuid)
1378{
1379 /* Only we change this so SMP safe */
1380 return current_uid();
1381}
1382
1383SYSCALL_DEFINE0(geteuid)
1384{
1385 /* Only we change this so SMP safe */
1386 return current_euid();
1387}
1388
1389SYSCALL_DEFINE0(getgid)
1390{
1391 /* Only we change this so SMP safe */
1392 return current_gid();
1393}
1394
1395SYSCALL_DEFINE0(getegid)
1396{
1397 /* Only we change this so SMP safe */
1398 return current_egid();
1399}
1400
1401#endif
1402
1403static void process_timeout(unsigned long __data)
1404{
1405 wake_up_process((struct task_struct *)__data);
1406}
1407
1408/**
1409 * schedule_timeout - sleep until timeout
1410 * @timeout: timeout value in jiffies
1411 *
1412 * Make the current task sleep until @timeout jiffies have
1413 * elapsed. The routine will return immediately unless
1414 * the current task state has been set (see set_current_state()).
1415 *
1416 * You can set the task state as follows -
1417 *
1418 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1419 * pass before the routine returns. The routine will return 0
1420 *
1421 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1422 * delivered to the current task. In this case the remaining time
1423 * in jiffies will be returned, or 0 if the timer expired in time
1424 *
1425 * The current task state is guaranteed to be TASK_RUNNING when this
1426 * routine returns.
1427 *
1428 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1429 * the CPU away without a bound on the timeout. In this case the return
1430 * value will be %MAX_SCHEDULE_TIMEOUT.
1431 *
1432 * In all cases the return value is guaranteed to be non-negative.
1433 */
1434signed long __sched schedule_timeout(signed long timeout)
1435{
1436 struct timer_list timer;
1437 unsigned long expire;
1438
1439 switch (timeout)
1440 {
1441 case MAX_SCHEDULE_TIMEOUT:
1442 /*
1443 * These two special cases are useful to be comfortable
1444 * in the caller. Nothing more. We could take
1445 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1446 * but I' d like to return a valid offset (>=0) to allow
1447 * the caller to do everything it want with the retval.
1448 */
1449 schedule();
1450 goto out;
1451 default:
1452 /*
1453 * Another bit of PARANOID. Note that the retval will be
1454 * 0 since no piece of kernel is supposed to do a check
1455 * for a negative retval of schedule_timeout() (since it
1456 * should never happens anyway). You just have the printk()
1457 * that will tell you if something is gone wrong and where.
1458 */
1459 if (timeout < 0) {
1460 printk(KERN_ERR "schedule_timeout: wrong timeout "
1461 "value %lx\n", timeout);
1462 dump_stack();
1463 current->state = TASK_RUNNING;
1464 goto out;
1465 }
1466 }
1467
1468 expire = timeout + jiffies;
1469
1470 setup_timer_on_stack(&timer, process_timeout, (unsigned long)current);
1471 __mod_timer(&timer, expire, false, TIMER_NOT_PINNED);
1472 schedule();
1473 del_singleshot_timer_sync(&timer);
1474
1475 /* Remove the timer from the object tracker */
1476 destroy_timer_on_stack(&timer);
1477
1478 timeout = expire - jiffies;
1479
1480 out:
1481 return timeout < 0 ? 0 : timeout;
1482}
1483EXPORT_SYMBOL(schedule_timeout);
1484
1485/*
1486 * We can use __set_current_state() here because schedule_timeout() calls
1487 * schedule() unconditionally.
1488 */
1489signed long __sched schedule_timeout_interruptible(signed long timeout)
1490{
1491 __set_current_state(TASK_INTERRUPTIBLE);
1492 return schedule_timeout(timeout);
1493}
1494EXPORT_SYMBOL(schedule_timeout_interruptible);
1495
1496signed long __sched schedule_timeout_killable(signed long timeout)
1497{
1498 __set_current_state(TASK_KILLABLE);
1499 return schedule_timeout(timeout);
1500}
1501EXPORT_SYMBOL(schedule_timeout_killable);
1502
1503signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1504{
1505 __set_current_state(TASK_UNINTERRUPTIBLE);
1506 return schedule_timeout(timeout);
1507}
1508EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1509
1510/* Thread ID - the internal kernel "pid" */
1511SYSCALL_DEFINE0(gettid)
1512{
1513 return task_pid_vnr(current);
1514}
1515
1516/**
1517 * do_sysinfo - fill in sysinfo struct
1518 * @info: pointer to buffer to fill
1519 */
1520int do_sysinfo(struct sysinfo *info)
1521{
1522 unsigned long mem_total, sav_total;
1523 unsigned int mem_unit, bitcount;
1524 struct timespec tp;
1525
1526 memset(info, 0, sizeof(struct sysinfo));
1527
1528 ktime_get_ts(&tp);
1529 monotonic_to_bootbased(&tp);
1530 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
1531
1532 get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
1533
1534 info->procs = nr_threads;
1535
1536 si_meminfo(info);
1537 si_swapinfo(info);
1538
1539 /*
1540 * If the sum of all the available memory (i.e. ram + swap)
1541 * is less than can be stored in a 32 bit unsigned long then
1542 * we can be binary compatible with 2.2.x kernels. If not,
1543 * well, in that case 2.2.x was broken anyways...
1544 *
1545 * -Erik Andersen <[email protected]>
1546 */
1547
1548 mem_total = info->totalram + info->totalswap;
1549 if (mem_total < info->totalram || mem_total < info->totalswap)
1550 goto out;
1551 bitcount = 0;
1552 mem_unit = info->mem_unit;
1553 while (mem_unit > 1) {
1554 bitcount++;
1555 mem_unit >>= 1;
1556 sav_total = mem_total;
1557 mem_total <<= 1;
1558 if (mem_total < sav_total)
1559 goto out;
1560 }
1561
1562 /*
1563 * If mem_total did not overflow, multiply all memory values by
1564 * info->mem_unit and set it to 1. This leaves things compatible
1565 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1566 * kernels...
1567 */
1568
1569 info->mem_unit = 1;
1570 info->totalram <<= bitcount;
1571 info->freeram <<= bitcount;
1572 info->sharedram <<= bitcount;
1573 info->bufferram <<= bitcount;
1574 info->totalswap <<= bitcount;
1575 info->freeswap <<= bitcount;
1576 info->totalhigh <<= bitcount;
1577 info->freehigh <<= bitcount;
1578
1579out:
1580 return 0;
1581}
1582
1583SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
1584{
1585 struct sysinfo val;
1586
1587 do_sysinfo(&val);
1588
1589 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
1590 return -EFAULT;
1591
1592 return 0;
1593}
1594
1595static int __cpuinit init_timers_cpu(int cpu)
1596{
1597 int j;
1598 struct tvec_base *base;
1599 static char __cpuinitdata tvec_base_done[NR_CPUS];
1600
1601 if (!tvec_base_done[cpu]) {
1602 static char boot_done;
1603
1604 if (boot_done) {
1605 /*
1606 * The APs use this path later in boot
1607 */
1608 base = kmalloc_node(sizeof(*base),
1609 GFP_KERNEL | __GFP_ZERO,
1610 cpu_to_node(cpu));
1611 if (!base)
1612 return -ENOMEM;
1613
1614 /* Make sure that tvec_base is 2 byte aligned */
1615 if (tbase_get_deferrable(base)) {
1616 WARN_ON(1);
1617 kfree(base);
1618 return -ENOMEM;
1619 }
1620 per_cpu(tvec_bases, cpu) = base;
1621 } else {
1622 /*
1623 * This is for the boot CPU - we use compile-time
1624 * static initialisation because per-cpu memory isn't
1625 * ready yet and because the memory allocators are not
1626 * initialised either.
1627 */
1628 boot_done = 1;
1629 base = &boot_tvec_bases;
1630 }
1631 tvec_base_done[cpu] = 1;
1632 } else {
1633 base = per_cpu(tvec_bases, cpu);
1634 }
1635
1636 spin_lock_init(&base->lock);
1637
1638 for (j = 0; j < TVN_SIZE; j++) {
1639 INIT_LIST_HEAD(base->tv5.vec + j);
1640 INIT_LIST_HEAD(base->tv4.vec + j);
1641 INIT_LIST_HEAD(base->tv3.vec + j);
1642 INIT_LIST_HEAD(base->tv2.vec + j);
1643 }
1644 for (j = 0; j < TVR_SIZE; j++)
1645 INIT_LIST_HEAD(base->tv1.vec + j);
1646
1647 base->timer_jiffies = jiffies;
1648 base->next_timer = base->timer_jiffies;
1649 return 0;
1650}
1651
1652#ifdef CONFIG_HOTPLUG_CPU
1653static void migrate_timer_list(struct tvec_base *new_base, struct list_head *head)
1654{
1655 struct timer_list *timer;
1656
1657 while (!list_empty(head)) {
1658 timer = list_first_entry(head, struct timer_list, entry);
1659 detach_timer(timer, 0);
1660 timer_set_base(timer, new_base);
1661 if (time_before(timer->expires, new_base->next_timer) &&
1662 !tbase_get_deferrable(timer->base))
1663 new_base->next_timer = timer->expires;
1664 internal_add_timer(new_base, timer);
1665 }
1666}
1667
1668static void __cpuinit migrate_timers(int cpu)
1669{
1670 struct tvec_base *old_base;
1671 struct tvec_base *new_base;
1672 int i;
1673
1674 BUG_ON(cpu_online(cpu));
1675 old_base = per_cpu(tvec_bases, cpu);
1676 new_base = get_cpu_var(tvec_bases);
1677 /*
1678 * The caller is globally serialized and nobody else
1679 * takes two locks at once, deadlock is not possible.
1680 */
1681 spin_lock_irq(&new_base->lock);
1682 spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1683
1684 BUG_ON(old_base->running_timer);
1685
1686 for (i = 0; i < TVR_SIZE; i++)
1687 migrate_timer_list(new_base, old_base->tv1.vec + i);
1688 for (i = 0; i < TVN_SIZE; i++) {
1689 migrate_timer_list(new_base, old_base->tv2.vec + i);
1690 migrate_timer_list(new_base, old_base->tv3.vec + i);
1691 migrate_timer_list(new_base, old_base->tv4.vec + i);
1692 migrate_timer_list(new_base, old_base->tv5.vec + i);
1693 }
1694
1695 spin_unlock(&old_base->lock);
1696 spin_unlock_irq(&new_base->lock);
1697 put_cpu_var(tvec_bases);
1698}
1699#endif /* CONFIG_HOTPLUG_CPU */
1700
1701static int __cpuinit timer_cpu_notify(struct notifier_block *self,
1702 unsigned long action, void *hcpu)
1703{
1704 long cpu = (long)hcpu;
1705 int err;
1706
1707 switch(action) {
1708 case CPU_UP_PREPARE:
1709 case CPU_UP_PREPARE_FROZEN:
1710 err = init_timers_cpu(cpu);
1711 if (err < 0)
1712 return notifier_from_errno(err);
1713 break;
1714#ifdef CONFIG_HOTPLUG_CPU
1715 case CPU_DEAD:
1716 case CPU_DEAD_FROZEN:
1717 migrate_timers(cpu);
1718 break;
1719#endif
1720 default:
1721 break;
1722 }
1723 return NOTIFY_OK;
1724}
1725
1726static struct notifier_block __cpuinitdata timers_nb = {
1727 .notifier_call = timer_cpu_notify,
1728};
1729
1730
1731void __init init_timers(void)
1732{
1733 int err = timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE,
1734 (void *)(long)smp_processor_id());
1735
1736 init_timer_stats();
1737
1738 BUG_ON(err != NOTIFY_OK);
1739 register_cpu_notifier(&timers_nb);
1740 open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
1741}
1742
1743/**
1744 * msleep - sleep safely even with waitqueue interruptions
1745 * @msecs: Time in milliseconds to sleep for
1746 */
1747void msleep(unsigned int msecs)
1748{
1749 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1750
1751 while (timeout)
1752 timeout = schedule_timeout_uninterruptible(timeout);
1753}
1754
1755EXPORT_SYMBOL(msleep);
1756
1757/**
1758 * msleep_interruptible - sleep waiting for signals
1759 * @msecs: Time in milliseconds to sleep for
1760 */
1761unsigned long msleep_interruptible(unsigned int msecs)
1762{
1763 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1764
1765 while (timeout && !signal_pending(current))
1766 timeout = schedule_timeout_interruptible(timeout);
1767 return jiffies_to_msecs(timeout);
1768}
1769
1770EXPORT_SYMBOL(msleep_interruptible);
1771
1772static int __sched do_usleep_range(unsigned long min, unsigned long max)
1773{
1774 ktime_t kmin;
1775 unsigned long delta;
1776
1777 kmin = ktime_set(0, min * NSEC_PER_USEC);
1778 delta = (max - min) * NSEC_PER_USEC;
1779 return schedule_hrtimeout_range(&kmin, delta, HRTIMER_MODE_REL);
1780}
1781
1782/**
1783 * usleep_range - Drop in replacement for udelay where wakeup is flexible
1784 * @min: Minimum time in usecs to sleep
1785 * @max: Maximum time in usecs to sleep
1786 */
1787void usleep_range(unsigned long min, unsigned long max)
1788{
1789 __set_current_state(TASK_UNINTERRUPTIBLE);
1790 do_usleep_range(min, max);
1791}
1792EXPORT_SYMBOL(usleep_range);