projects/linux·blob v2.6.33-rc2 · refs

1684 lines · 44.4 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/module.h>
24#include <linux/interrupt.h>
25#include <linux/percpu.h>
26#include <linux/init.h>
27#include <linux/mm.h>
28#include <linux/swap.h>
29#include <linux/pid_namespace.h>
30#include <linux/notifier.h>
31#include <linux/thread_info.h>
32#include <linux/time.h>
33#include <linux/jiffies.h>
34#include <linux/posix-timers.h>
35#include <linux/cpu.h>
36#include <linux/syscalls.h>
37#include <linux/delay.h>
38#include <linux/tick.h>
39#include <linux/kallsyms.h>
40#include <linux/perf_event.h>
41#include <linux/sched.h>
42
43#include <asm/uaccess.h>
44#include <asm/unistd.h>
45#include <asm/div64.h>
46#include <asm/timex.h>
47#include <asm/io.h>
48
49#define CREATE_TRACE_POINTS
50#include <trace/events/timer.h>
51
52u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
53
54EXPORT_SYMBOL(jiffies_64);
55
56/*
57 * per-CPU timer vector definitions:
58 */
59#define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6)
60#define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8)
61#define TVN_SIZE (1 << TVN_BITS)
62#define TVR_SIZE (1 << TVR_BITS)
63#define TVN_MASK (TVN_SIZE - 1)
64#define TVR_MASK (TVR_SIZE - 1)
65
66struct tvec {
67 struct list_head vec[TVN_SIZE];
68};
69
70struct tvec_root {
71 struct list_head vec[TVR_SIZE];
72};
73
74struct tvec_base {
75 spinlock_t lock;
76 struct timer_list *running_timer;
77 unsigned long timer_jiffies;
78 unsigned long next_timer;
79 struct tvec_root tv1;
80 struct tvec tv2;
81 struct tvec tv3;
82 struct tvec tv4;
83 struct tvec tv5;
84} ____cacheline_aligned;
85
86struct tvec_base boot_tvec_bases;
87EXPORT_SYMBOL(boot_tvec_bases);
88static DEFINE_PER_CPU(struct tvec_base *, tvec_bases) = &boot_tvec_bases;
89
90/*
91 * Note that all tvec_bases are 2 byte aligned and lower bit of
92 * base in timer_list is guaranteed to be zero. Use the LSB for
93 * the new flag to indicate whether the timer is deferrable
94 */
95#define TBASE_DEFERRABLE_FLAG (0x1)
96
97/* Functions below help us manage 'deferrable' flag */
98static inline unsigned int tbase_get_deferrable(struct tvec_base *base)
99{
100 return ((unsigned int)(unsigned long)base & TBASE_DEFERRABLE_FLAG);
101}
102
103static inline struct tvec_base *tbase_get_base(struct tvec_base *base)
104{
105 return ((struct tvec_base *)((unsigned long)base & ~TBASE_DEFERRABLE_FLAG));
106}
107
108static inline void timer_set_deferrable(struct timer_list *timer)
109{
110 timer->base = ((struct tvec_base *)((unsigned long)(timer->base) |
111 TBASE_DEFERRABLE_FLAG));
112}
113
114static inline void
115timer_set_base(struct timer_list *timer, struct tvec_base *new_base)
116{
117 timer->base = (struct tvec_base *)((unsigned long)(new_base) |
118 tbase_get_deferrable(timer->base));
119}
120
121static unsigned long round_jiffies_common(unsigned long j, int cpu,
122 bool force_up)
123{
124 int rem;
125 unsigned long original = j;
126
127 /*
128 * We don't want all cpus firing their timers at once hitting the
129 * same lock or cachelines, so we skew each extra cpu with an extra
130 * 3 jiffies. This 3 jiffies came originally from the mm/ code which
131 * already did this.
132 * The skew is done by adding 3*cpunr, then round, then subtract this
133 * extra offset again.
134 */
135 j += cpu * 3;
136
137 rem = j % HZ;
138
139 /*
140 * If the target jiffie is just after a whole second (which can happen
141 * due to delays of the timer irq, long irq off times etc etc) then
142 * we should round down to the whole second, not up. Use 1/4th second
143 * as cutoff for this rounding as an extreme upper bound for this.
144 * But never round down if @force_up is set.
145 */
146 if (rem < HZ/4 && !force_up) /* round down */
147 j = j - rem;
148 else /* round up */
149 j = j - rem + HZ;
150
151 /* now that we have rounded, subtract the extra skew again */
152 j -= cpu * 3;
153
154 if (j <= jiffies) /* rounding ate our timeout entirely; */
155 return original;
156 return j;
157}
158
159/**
160 * __round_jiffies - function to round jiffies to a full second
161 * @j: the time in (absolute) jiffies that should be rounded
162 * @cpu: the processor number on which the timeout will happen
163 *
164 * __round_jiffies() rounds an absolute time in the future (in jiffies)
165 * up or down to (approximately) full seconds. This is useful for timers
166 * for which the exact time they fire does not matter too much, as long as
167 * they fire approximately every X seconds.
168 *
169 * By rounding these timers to whole seconds, all such timers will fire
170 * at the same time, rather than at various times spread out. The goal
171 * of this is to have the CPU wake up less, which saves power.
172 *
173 * The exact rounding is skewed for each processor to avoid all
174 * processors firing at the exact same time, which could lead
175 * to lock contention or spurious cache line bouncing.
176 *
177 * The return value is the rounded version of the @j parameter.
178 */
179unsigned long __round_jiffies(unsigned long j, int cpu)
180{
181 return round_jiffies_common(j, cpu, false);
182}
183EXPORT_SYMBOL_GPL(__round_jiffies);
184
185/**
186 * __round_jiffies_relative - function to round jiffies to a full second
187 * @j: the time in (relative) jiffies that should be rounded
188 * @cpu: the processor number on which the timeout will happen
189 *
190 * __round_jiffies_relative() rounds a time delta in the future (in jiffies)
191 * up or down to (approximately) full seconds. This is useful for timers
192 * for which the exact time they fire does not matter too much, as long as
193 * they fire approximately every X seconds.
194 *
195 * By rounding these timers to whole seconds, all such timers will fire
196 * at the same time, rather than at various times spread out. The goal
197 * of this is to have the CPU wake up less, which saves power.
198 *
199 * The exact rounding is skewed for each processor to avoid all
200 * processors firing at the exact same time, which could lead
201 * to lock contention or spurious cache line bouncing.
202 *
203 * The return value is the rounded version of the @j parameter.
204 */
205unsigned long __round_jiffies_relative(unsigned long j, int cpu)
206{
207 unsigned long j0 = jiffies;
208
209 /* Use j0 because jiffies might change while we run */
210 return round_jiffies_common(j + j0, cpu, false) - j0;
211}
212EXPORT_SYMBOL_GPL(__round_jiffies_relative);
213
214/**
215 * round_jiffies - function to round jiffies to a full second
216 * @j: the time in (absolute) jiffies that should be rounded
217 *
218 * round_jiffies() rounds an absolute time in the future (in jiffies)
219 * up or down to (approximately) full seconds. This is useful for timers
220 * for which the exact time they fire does not matter too much, as long as
221 * they fire approximately every X seconds.
222 *
223 * By rounding these timers to whole seconds, all such timers will fire
224 * at the same time, rather than at various times spread out. The goal
225 * of this is to have the CPU wake up less, which saves power.
226 *
227 * The return value is the rounded version of the @j parameter.
228 */
229unsigned long round_jiffies(unsigned long j)
230{
231 return round_jiffies_common(j, raw_smp_processor_id(), false);
232}
233EXPORT_SYMBOL_GPL(round_jiffies);
234
235/**
236 * round_jiffies_relative - function to round jiffies to a full second
237 * @j: the time in (relative) jiffies that should be rounded
238 *
239 * round_jiffies_relative() rounds a time delta in the future (in jiffies)
240 * up or down to (approximately) full seconds. This is useful for timers
241 * for which the exact time they fire does not matter too much, as long as
242 * they fire approximately every X seconds.
243 *
244 * By rounding these timers to whole seconds, all such timers will fire
245 * at the same time, rather than at various times spread out. The goal
246 * of this is to have the CPU wake up less, which saves power.
247 *
248 * The return value is the rounded version of the @j parameter.
249 */
250unsigned long round_jiffies_relative(unsigned long j)
251{
252 return __round_jiffies_relative(j, raw_smp_processor_id());
253}
254EXPORT_SYMBOL_GPL(round_jiffies_relative);
255
256/**
257 * __round_jiffies_up - function to round jiffies up to a full second
258 * @j: the time in (absolute) jiffies that should be rounded
259 * @cpu: the processor number on which the timeout will happen
260 *
261 * This is the same as __round_jiffies() except that it will never
262 * round down. This is useful for timeouts for which the exact time
263 * of firing does not matter too much, as long as they don't fire too
264 * early.
265 */
266unsigned long __round_jiffies_up(unsigned long j, int cpu)
267{
268 return round_jiffies_common(j, cpu, true);
269}
270EXPORT_SYMBOL_GPL(__round_jiffies_up);
271
272/**
273 * __round_jiffies_up_relative - function to round jiffies up to a full second
274 * @j: the time in (relative) jiffies that should be rounded
275 * @cpu: the processor number on which the timeout will happen
276 *
277 * This is the same as __round_jiffies_relative() except that it will never
278 * round down. This is useful for timeouts for which the exact time
279 * of firing does not matter too much, as long as they don't fire too
280 * early.
281 */
282unsigned long __round_jiffies_up_relative(unsigned long j, int cpu)
283{
284 unsigned long j0 = jiffies;
285
286 /* Use j0 because jiffies might change while we run */
287 return round_jiffies_common(j + j0, cpu, true) - j0;
288}
289EXPORT_SYMBOL_GPL(__round_jiffies_up_relative);
290
291/**
292 * round_jiffies_up - function to round jiffies up to a full second
293 * @j: the time in (absolute) jiffies that should be rounded
294 *
295 * This is the same as round_jiffies() except that it will never
296 * round down. This is useful for timeouts for which the exact time
297 * of firing does not matter too much, as long as they don't fire too
298 * early.
299 */
300unsigned long round_jiffies_up(unsigned long j)
301{
302 return round_jiffies_common(j, raw_smp_processor_id(), true);
303}
304EXPORT_SYMBOL_GPL(round_jiffies_up);
305
306/**
307 * round_jiffies_up_relative - function to round jiffies up to a full second
308 * @j: the time in (relative) jiffies that should be rounded
309 *
310 * This is the same as round_jiffies_relative() except that it will never
311 * round down. This is useful for timeouts for which the exact time
312 * of firing does not matter too much, as long as they don't fire too
313 * early.
314 */
315unsigned long round_jiffies_up_relative(unsigned long j)
316{
317 return __round_jiffies_up_relative(j, raw_smp_processor_id());
318}
319EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
320
321
322static inline void set_running_timer(struct tvec_base *base,
323 struct timer_list *timer)
324{
325#ifdef CONFIG_SMP
326 base->running_timer = timer;
327#endif
328}
329
330static void internal_add_timer(struct tvec_base *base, struct timer_list *timer)
331{
332 unsigned long expires = timer->expires;
333 unsigned long idx = expires - base->timer_jiffies;
334 struct list_head *vec;
335
336 if (idx < TVR_SIZE) {
337 int i = expires & TVR_MASK;
338 vec = base->tv1.vec + i;
339 } else if (idx < 1 << (TVR_BITS + TVN_BITS)) {
340 int i = (expires >> TVR_BITS) & TVN_MASK;
341 vec = base->tv2.vec + i;
342 } else if (idx < 1 << (TVR_BITS + 2 * TVN_BITS)) {
343 int i = (expires >> (TVR_BITS + TVN_BITS)) & TVN_MASK;
344 vec = base->tv3.vec + i;
345 } else if (idx < 1 << (TVR_BITS + 3 * TVN_BITS)) {
346 int i = (expires >> (TVR_BITS + 2 * TVN_BITS)) & TVN_MASK;
347 vec = base->tv4.vec + i;
348 } else if ((signed long) idx < 0) {
349 /*
350 * Can happen if you add a timer with expires == jiffies,
351 * or you set a timer to go off in the past
352 */
353 vec = base->tv1.vec + (base->timer_jiffies & TVR_MASK);
354 } else {
355 int i;
356 /* If the timeout is larger than 0xffffffff on 64-bit
357 * architectures then we use the maximum timeout:
358 */
359 if (idx > 0xffffffffUL) {
360 idx = 0xffffffffUL;
361 expires = idx + base->timer_jiffies;
362 }
363 i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK;
364 vec = base->tv5.vec + i;
365 }
366 /*
367 * Timers are FIFO:
368 */
369 list_add_tail(&timer->entry, vec);
370}
371
372#ifdef CONFIG_TIMER_STATS
373void __timer_stats_timer_set_start_info(struct timer_list *timer, void *addr)
374{
375 if (timer->start_site)
376 return;
377
378 timer->start_site = addr;
379 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
380 timer->start_pid = current->pid;
381}
382
383static void timer_stats_account_timer(struct timer_list *timer)
384{
385 unsigned int flag = 0;
386
387 if (likely(!timer->start_site))
388 return;
389 if (unlikely(tbase_get_deferrable(timer->base)))
390 flag |= TIMER_STATS_FLAG_DEFERRABLE;
391
392 timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
393 timer->function, timer->start_comm, flag);
394}
395
396#else
397static void timer_stats_account_timer(struct timer_list *timer) {}
398#endif
399
400#ifdef CONFIG_DEBUG_OBJECTS_TIMERS
401
402static struct debug_obj_descr timer_debug_descr;
403
404/*
405 * fixup_init is called when:
406 * - an active object is initialized
407 */
408static int timer_fixup_init(void *addr, enum debug_obj_state state)
409{
410 struct timer_list *timer = addr;
411
412 switch (state) {
413 case ODEBUG_STATE_ACTIVE:
414 del_timer_sync(timer);
415 debug_object_init(timer, &timer_debug_descr);
416 return 1;
417 default:
418 return 0;
419 }
420}
421
422/*
423 * fixup_activate is called when:
424 * - an active object is activated
425 * - an unknown object is activated (might be a statically initialized object)
426 */
427static int timer_fixup_activate(void *addr, enum debug_obj_state state)
428{
429 struct timer_list *timer = addr;
430
431 switch (state) {
432
433 case ODEBUG_STATE_NOTAVAILABLE:
434 /*
435 * This is not really a fixup. The timer was
436 * statically initialized. We just make sure that it
437 * is tracked in the object tracker.
438 */
439 if (timer->entry.next == NULL &&
440 timer->entry.prev == TIMER_ENTRY_STATIC) {
441 debug_object_init(timer, &timer_debug_descr);
442 debug_object_activate(timer, &timer_debug_descr);
443 return 0;
444 } else {
445 WARN_ON_ONCE(1);
446 }
447 return 0;
448
449 case ODEBUG_STATE_ACTIVE:
450 WARN_ON(1);
451
452 default:
453 return 0;
454 }
455}
456
457/*
458 * fixup_free is called when:
459 * - an active object is freed
460 */
461static int timer_fixup_free(void *addr, enum debug_obj_state state)
462{
463 struct timer_list *timer = addr;
464
465 switch (state) {
466 case ODEBUG_STATE_ACTIVE:
467 del_timer_sync(timer);
468 debug_object_free(timer, &timer_debug_descr);
469 return 1;
470 default:
471 return 0;
472 }
473}
474
475static struct debug_obj_descr timer_debug_descr = {
476 .name = "timer_list",
477 .fixup_init = timer_fixup_init,
478 .fixup_activate = timer_fixup_activate,
479 .fixup_free = timer_fixup_free,
480};
481
482static inline void debug_timer_init(struct timer_list *timer)
483{
484 debug_object_init(timer, &timer_debug_descr);
485}
486
487static inline void debug_timer_activate(struct timer_list *timer)
488{
489 debug_object_activate(timer, &timer_debug_descr);
490}
491
492static inline void debug_timer_deactivate(struct timer_list *timer)
493{
494 debug_object_deactivate(timer, &timer_debug_descr);
495}
496
497static inline void debug_timer_free(struct timer_list *timer)
498{
499 debug_object_free(timer, &timer_debug_descr);
500}
501
502static void __init_timer(struct timer_list *timer,
503 const char *name,
504 struct lock_class_key *key);
505
506void init_timer_on_stack_key(struct timer_list *timer,
507 const char *name,
508 struct lock_class_key *key)
509{
510 debug_object_init_on_stack(timer, &timer_debug_descr);
511 __init_timer(timer, name, key);
512}
513EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
514
515void destroy_timer_on_stack(struct timer_list *timer)
516{
517 debug_object_free(timer, &timer_debug_descr);
518}
519EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
520
521#else
522static inline void debug_timer_init(struct timer_list *timer) { }
523static inline void debug_timer_activate(struct timer_list *timer) { }
524static inline void debug_timer_deactivate(struct timer_list *timer) { }
525#endif
526
527static inline void debug_init(struct timer_list *timer)
528{
529 debug_timer_init(timer);
530 trace_timer_init(timer);
531}
532
533static inline void
534debug_activate(struct timer_list *timer, unsigned long expires)
535{
536 debug_timer_activate(timer);
537 trace_timer_start(timer, expires);
538}
539
540static inline void debug_deactivate(struct timer_list *timer)
541{
542 debug_timer_deactivate(timer);
543 trace_timer_cancel(timer);
544}
545
546static void __init_timer(struct timer_list *timer,
547 const char *name,
548 struct lock_class_key *key)
549{
550 timer->entry.next = NULL;
551 timer->base = __raw_get_cpu_var(tvec_bases);
552#ifdef CONFIG_TIMER_STATS
553 timer->start_site = NULL;
554 timer->start_pid = -1;
555 memset(timer->start_comm, 0, TASK_COMM_LEN);
556#endif
557 lockdep_init_map(&timer->lockdep_map, name, key, 0);
558}
559
560/**
561 * init_timer_key - initialize a timer
562 * @timer: the timer to be initialized
563 * @name: name of the timer
564 * @key: lockdep class key of the fake lock used for tracking timer
565 * sync lock dependencies
566 *
567 * init_timer_key() must be done to a timer prior calling *any* of the
568 * other timer functions.
569 */
570void init_timer_key(struct timer_list *timer,
571 const char *name,
572 struct lock_class_key *key)
573{
574 debug_init(timer);
575 __init_timer(timer, name, key);
576}
577EXPORT_SYMBOL(init_timer_key);
578
579void init_timer_deferrable_key(struct timer_list *timer,
580 const char *name,
581 struct lock_class_key *key)
582{
583 init_timer_key(timer, name, key);
584 timer_set_deferrable(timer);
585}
586EXPORT_SYMBOL(init_timer_deferrable_key);
587
588static inline void detach_timer(struct timer_list *timer,
589 int clear_pending)
590{
591 struct list_head *entry = &timer->entry;
592
593 debug_deactivate(timer);
594
595 __list_del(entry->prev, entry->next);
596 if (clear_pending)
597 entry->next = NULL;
598 entry->prev = LIST_POISON2;
599}
600
601/*
602 * We are using hashed locking: holding per_cpu(tvec_bases).lock
603 * means that all timers which are tied to this base via timer->base are
604 * locked, and the base itself is locked too.
605 *
606 * So __run_timers/migrate_timers can safely modify all timers which could
607 * be found on ->tvX lists.
608 *
609 * When the timer's base is locked, and the timer removed from list, it is
610 * possible to set timer->base = NULL and drop the lock: the timer remains
611 * locked.
612 */
613static struct tvec_base *lock_timer_base(struct timer_list *timer,
614 unsigned long *flags)
615 __acquires(timer->base->lock)
616{
617 struct tvec_base *base;
618
619 for (;;) {
620 struct tvec_base *prelock_base = timer->base;
621 base = tbase_get_base(prelock_base);
622 if (likely(base != NULL)) {
623 spin_lock_irqsave(&base->lock, *flags);
624 if (likely(prelock_base == timer->base))
625 return base;
626 /* The timer has migrated to another CPU */
627 spin_unlock_irqrestore(&base->lock, *flags);
628 }
629 cpu_relax();
630 }
631}
632
633static inline int
634__mod_timer(struct timer_list *timer, unsigned long expires,
635 bool pending_only, int pinned)
636{
637 struct tvec_base *base, *new_base;
638 unsigned long flags;
639 int ret = 0 , cpu;
640
641 timer_stats_timer_set_start_info(timer);
642 BUG_ON(!timer->function);
643
644 base = lock_timer_base(timer, &flags);
645
646 if (timer_pending(timer)) {
647 detach_timer(timer, 0);
648 if (timer->expires == base->next_timer &&
649 !tbase_get_deferrable(timer->base))
650 base->next_timer = base->timer_jiffies;
651 ret = 1;
652 } else {
653 if (pending_only)
654 goto out_unlock;
655 }
656
657 debug_activate(timer, expires);
658
659 cpu = smp_processor_id();
660
661#if defined(CONFIG_NO_HZ) && defined(CONFIG_SMP)
662 if (!pinned && get_sysctl_timer_migration() && idle_cpu(cpu)) {
663 int preferred_cpu = get_nohz_load_balancer();
664
665 if (preferred_cpu >= 0)
666 cpu = preferred_cpu;
667 }
668#endif
669 new_base = per_cpu(tvec_bases, cpu);
670
671 if (base != new_base) {
672 /*
673 * We are trying to schedule the timer on the local CPU.
674 * However we can't change timer's base while it is running,
675 * otherwise del_timer_sync() can't detect that the timer's
676 * handler yet has not finished. This also guarantees that
677 * the timer is serialized wrt itself.
678 */
679 if (likely(base->running_timer != timer)) {
680 /* See the comment in lock_timer_base() */
681 timer_set_base(timer, NULL);
682 spin_unlock(&base->lock);
683 base = new_base;
684 spin_lock(&base->lock);
685 timer_set_base(timer, base);
686 }
687 }
688
689 timer->expires = expires;
690 if (time_before(timer->expires, base->next_timer) &&
691 !tbase_get_deferrable(timer->base))
692 base->next_timer = timer->expires;
693 internal_add_timer(base, timer);
694
695out_unlock:
696 spin_unlock_irqrestore(&base->lock, flags);
697
698 return ret;
699}
700
701/**
702 * mod_timer_pending - modify a pending timer's timeout
703 * @timer: the pending timer to be modified
704 * @expires: new timeout in jiffies
705 *
706 * mod_timer_pending() is the same for pending timers as mod_timer(),
707 * but will not re-activate and modify already deleted timers.
708 *
709 * It is useful for unserialized use of timers.
710 */
711int mod_timer_pending(struct timer_list *timer, unsigned long expires)
712{
713 return __mod_timer(timer, expires, true, TIMER_NOT_PINNED);
714}
715EXPORT_SYMBOL(mod_timer_pending);
716
717/**
718 * mod_timer - modify a timer's timeout
719 * @timer: the timer to be modified
720 * @expires: new timeout in jiffies
721 *
722 * mod_timer() is a more efficient way to update the expire field of an
723 * active timer (if the timer is inactive it will be activated)
724 *
725 * mod_timer(timer, expires) is equivalent to:
726 *
727 * del_timer(timer); timer->expires = expires; add_timer(timer);
728 *
729 * Note that if there are multiple unserialized concurrent users of the
730 * same timer, then mod_timer() is the only safe way to modify the timeout,
731 * since add_timer() cannot modify an already running timer.
732 *
733 * The function returns whether it has modified a pending timer or not.
734 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
735 * active timer returns 1.)
736 */
737int mod_timer(struct timer_list *timer, unsigned long expires)
738{
739 /*
740 * This is a common optimization triggered by the
741 * networking code - if the timer is re-modified
742 * to be the same thing then just return:
743 */
744 if (timer_pending(timer) && timer->expires == expires)
745 return 1;
746
747 return __mod_timer(timer, expires, false, TIMER_NOT_PINNED);
748}
749EXPORT_SYMBOL(mod_timer);
750
751/**
752 * mod_timer_pinned - modify a timer's timeout
753 * @timer: the timer to be modified
754 * @expires: new timeout in jiffies
755 *
756 * mod_timer_pinned() is a way to update the expire field of an
757 * active timer (if the timer is inactive it will be activated)
758 * and not allow the timer to be migrated to a different CPU.
759 *
760 * mod_timer_pinned(timer, expires) is equivalent to:
761 *
762 * del_timer(timer); timer->expires = expires; add_timer(timer);
763 */
764int mod_timer_pinned(struct timer_list *timer, unsigned long expires)
765{
766 if (timer->expires == expires && timer_pending(timer))
767 return 1;
768
769 return __mod_timer(timer, expires, false, TIMER_PINNED);
770}
771EXPORT_SYMBOL(mod_timer_pinned);
772
773/**
774 * add_timer - start a timer
775 * @timer: the timer to be added
776 *
777 * The kernel will do a ->function(->data) callback from the
778 * timer interrupt at the ->expires point in the future. The
779 * current time is 'jiffies'.
780 *
781 * The timer's ->expires, ->function (and if the handler uses it, ->data)
782 * fields must be set prior calling this function.
783 *
784 * Timers with an ->expires field in the past will be executed in the next
785 * timer tick.
786 */
787void add_timer(struct timer_list *timer)
788{
789 BUG_ON(timer_pending(timer));
790 mod_timer(timer, timer->expires);
791}
792EXPORT_SYMBOL(add_timer);
793
794/**
795 * add_timer_on - start a timer on a particular CPU
796 * @timer: the timer to be added
797 * @cpu: the CPU to start it on
798 *
799 * This is not very scalable on SMP. Double adds are not possible.
800 */
801void add_timer_on(struct timer_list *timer, int cpu)
802{
803 struct tvec_base *base = per_cpu(tvec_bases, cpu);
804 unsigned long flags;
805
806 timer_stats_timer_set_start_info(timer);
807 BUG_ON(timer_pending(timer) || !timer->function);
808 spin_lock_irqsave(&base->lock, flags);
809 timer_set_base(timer, base);
810 debug_activate(timer, timer->expires);
811 if (time_before(timer->expires, base->next_timer) &&
812 !tbase_get_deferrable(timer->base))
813 base->next_timer = timer->expires;
814 internal_add_timer(base, timer);
815 /*
816 * Check whether the other CPU is idle and needs to be
817 * triggered to reevaluate the timer wheel when nohz is
818 * active. We are protected against the other CPU fiddling
819 * with the timer by holding the timer base lock. This also
820 * makes sure that a CPU on the way to idle can not evaluate
821 * the timer wheel.
822 */
823 wake_up_idle_cpu(cpu);
824 spin_unlock_irqrestore(&base->lock, flags);
825}
826EXPORT_SYMBOL_GPL(add_timer_on);
827
828/**
829 * del_timer - deactive a timer.
830 * @timer: the timer to be deactivated
831 *
832 * del_timer() deactivates a timer - this works on both active and inactive
833 * timers.
834 *
835 * The function returns whether it has deactivated a pending timer or not.
836 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
837 * active timer returns 1.)
838 */
839int del_timer(struct timer_list *timer)
840{
841 struct tvec_base *base;
842 unsigned long flags;
843 int ret = 0;
844
845 timer_stats_timer_clear_start_info(timer);
846 if (timer_pending(timer)) {
847 base = lock_timer_base(timer, &flags);
848 if (timer_pending(timer)) {
849 detach_timer(timer, 1);
850 if (timer->expires == base->next_timer &&
851 !tbase_get_deferrable(timer->base))
852 base->next_timer = base->timer_jiffies;
853 ret = 1;
854 }
855 spin_unlock_irqrestore(&base->lock, flags);
856 }
857
858 return ret;
859}
860EXPORT_SYMBOL(del_timer);
861
862#ifdef CONFIG_SMP
863/**
864 * try_to_del_timer_sync - Try to deactivate a timer
865 * @timer: timer do del
866 *
867 * This function tries to deactivate a timer. Upon successful (ret >= 0)
868 * exit the timer is not queued and the handler is not running on any CPU.
869 *
870 * It must not be called from interrupt contexts.
871 */
872int try_to_del_timer_sync(struct timer_list *timer)
873{
874 struct tvec_base *base;
875 unsigned long flags;
876 int ret = -1;
877
878 base = lock_timer_base(timer, &flags);
879
880 if (base->running_timer == timer)
881 goto out;
882
883 ret = 0;
884 if (timer_pending(timer)) {
885 detach_timer(timer, 1);
886 if (timer->expires == base->next_timer &&
887 !tbase_get_deferrable(timer->base))
888 base->next_timer = base->timer_jiffies;
889 ret = 1;
890 }
891out:
892 spin_unlock_irqrestore(&base->lock, flags);
893
894 return ret;
895}
896EXPORT_SYMBOL(try_to_del_timer_sync);
897
898/**
899 * del_timer_sync - deactivate a timer and wait for the handler to finish.
900 * @timer: the timer to be deactivated
901 *
902 * This function only differs from del_timer() on SMP: besides deactivating
903 * the timer it also makes sure the handler has finished executing on other
904 * CPUs.
905 *
906 * Synchronization rules: Callers must prevent restarting of the timer,
907 * otherwise this function is meaningless. It must not be called from
908 * interrupt contexts. The caller must not hold locks which would prevent
909 * completion of the timer's handler. The timer's handler must not call
910 * add_timer_on(). Upon exit the timer is not queued and the handler is
911 * not running on any CPU.
912 *
913 * The function returns whether it has deactivated a pending timer or not.
914 */
915int del_timer_sync(struct timer_list *timer)
916{
917#ifdef CONFIG_LOCKDEP
918 unsigned long flags;
919
920 local_irq_save(flags);
921 lock_map_acquire(&timer->lockdep_map);
922 lock_map_release(&timer->lockdep_map);
923 local_irq_restore(flags);
924#endif
925
926 for (;;) {
927 int ret = try_to_del_timer_sync(timer);
928 if (ret >= 0)
929 return ret;
930 cpu_relax();
931 }
932}
933EXPORT_SYMBOL(del_timer_sync);
934#endif
935
936static int cascade(struct tvec_base *base, struct tvec *tv, int index)
937{
938 /* cascade all the timers from tv up one level */
939 struct timer_list *timer, *tmp;
940 struct list_head tv_list;
941
942 list_replace_init(tv->vec + index, &tv_list);
943
944 /*
945 * We are removing _all_ timers from the list, so we
946 * don't have to detach them individually.
947 */
948 list_for_each_entry_safe(timer, tmp, &tv_list, entry) {
949 BUG_ON(tbase_get_base(timer->base) != base);
950 internal_add_timer(base, timer);
951 }
952
953 return index;
954}
955
956#define INDEX(N) ((base->timer_jiffies >> (TVR_BITS + (N) * TVN_BITS)) & TVN_MASK)
957
958/**
959 * __run_timers - run all expired timers (if any) on this CPU.
960 * @base: the timer vector to be processed.
961 *
962 * This function cascades all vectors and executes all expired timer
963 * vectors.
964 */
965static inline void __run_timers(struct tvec_base *base)
966{
967 struct timer_list *timer;
968
969 spin_lock_irq(&base->lock);
970 while (time_after_eq(jiffies, base->timer_jiffies)) {
971 struct list_head work_list;
972 struct list_head *head = &work_list;
973 int index = base->timer_jiffies & TVR_MASK;
974
975 /*
976 * Cascade timers:
977 */
978 if (!index &&
979 (!cascade(base, &base->tv2, INDEX(0))) &&
980 (!cascade(base, &base->tv3, INDEX(1))) &&
981 !cascade(base, &base->tv4, INDEX(2)))
982 cascade(base, &base->tv5, INDEX(3));
983 ++base->timer_jiffies;
984 list_replace_init(base->tv1.vec + index, &work_list);
985 while (!list_empty(head)) {
986 void (*fn)(unsigned long);
987 unsigned long data;
988
989 timer = list_first_entry(head, struct timer_list,entry);
990 fn = timer->function;
991 data = timer->data;
992
993 timer_stats_account_timer(timer);
994
995 set_running_timer(base, timer);
996 detach_timer(timer, 1);
997
998 spin_unlock_irq(&base->lock);
999 {
1000 int preempt_count = preempt_count();
1001
1002#ifdef CONFIG_LOCKDEP
1003 /*
1004 * It is permissible to free the timer from
1005 * inside the function that is called from
1006 * it, this we need to take into account for
1007 * lockdep too. To avoid bogus "held lock
1008 * freed" warnings as well as problems when
1009 * looking into timer->lockdep_map, make a
1010 * copy and use that here.
1011 */
1012 struct lockdep_map lockdep_map =
1013 timer->lockdep_map;
1014#endif
1015 /*
1016 * Couple the lock chain with the lock chain at
1017 * del_timer_sync() by acquiring the lock_map
1018 * around the fn() call here and in
1019 * del_timer_sync().
1020 */
1021 lock_map_acquire(&lockdep_map);
1022
1023 trace_timer_expire_entry(timer);
1024 fn(data);
1025 trace_timer_expire_exit(timer);
1026
1027 lock_map_release(&lockdep_map);
1028
1029 if (preempt_count != preempt_count()) {
1030 printk(KERN_ERR "huh, entered %p "
1031 "with preempt_count %08x, exited"
1032 " with %08x?\n",
1033 fn, preempt_count,
1034 preempt_count());
1035 BUG();
1036 }
1037 }
1038 spin_lock_irq(&base->lock);
1039 }
1040 }
1041 set_running_timer(base, NULL);
1042 spin_unlock_irq(&base->lock);
1043}
1044
1045#ifdef CONFIG_NO_HZ
1046/*
1047 * Find out when the next timer event is due to happen. This
1048 * is used on S/390 to stop all activity when a CPU is idle.
1049 * This function needs to be called with interrupts disabled.
1050 */
1051static unsigned long __next_timer_interrupt(struct tvec_base *base)
1052{
1053 unsigned long timer_jiffies = base->timer_jiffies;
1054 unsigned long expires = timer_jiffies + NEXT_TIMER_MAX_DELTA;
1055 int index, slot, array, found = 0;
1056 struct timer_list *nte;
1057 struct tvec *varray[4];
1058
1059 /* Look for timer events in tv1. */
1060 index = slot = timer_jiffies & TVR_MASK;
1061 do {
1062 list_for_each_entry(nte, base->tv1.vec + slot, entry) {
1063 if (tbase_get_deferrable(nte->base))
1064 continue;
1065
1066 found = 1;
1067 expires = nte->expires;
1068 /* Look at the cascade bucket(s)? */
1069 if (!index || slot < index)
1070 goto cascade;
1071 return expires;
1072 }
1073 slot = (slot + 1) & TVR_MASK;
1074 } while (slot != index);
1075
1076cascade:
1077 /* Calculate the next cascade event */
1078 if (index)
1079 timer_jiffies += TVR_SIZE - index;
1080 timer_jiffies >>= TVR_BITS;
1081
1082 /* Check tv2-tv5. */
1083 varray[0] = &base->tv2;
1084 varray[1] = &base->tv3;
1085 varray[2] = &base->tv4;
1086 varray[3] = &base->tv5;
1087
1088 for (array = 0; array < 4; array++) {
1089 struct tvec *varp = varray[array];
1090
1091 index = slot = timer_jiffies & TVN_MASK;
1092 do {
1093 list_for_each_entry(nte, varp->vec + slot, entry) {
1094 if (tbase_get_deferrable(nte->base))
1095 continue;
1096
1097 found = 1;
1098 if (time_before(nte->expires, expires))
1099 expires = nte->expires;
1100 }
1101 /*
1102 * Do we still search for the first timer or are
1103 * we looking up the cascade buckets ?
1104 */
1105 if (found) {
1106 /* Look at the cascade bucket(s)? */
1107 if (!index || slot < index)
1108 break;
1109 return expires;
1110 }
1111 slot = (slot + 1) & TVN_MASK;
1112 } while (slot != index);
1113
1114 if (index)
1115 timer_jiffies += TVN_SIZE - index;
1116 timer_jiffies >>= TVN_BITS;
1117 }
1118 return expires;
1119}
1120
1121/*
1122 * Check, if the next hrtimer event is before the next timer wheel
1123 * event:
1124 */
1125static unsigned long cmp_next_hrtimer_event(unsigned long now,
1126 unsigned long expires)
1127{
1128 ktime_t hr_delta = hrtimer_get_next_event();
1129 struct timespec tsdelta;
1130 unsigned long delta;
1131
1132 if (hr_delta.tv64 == KTIME_MAX)
1133 return expires;
1134
1135 /*
1136 * Expired timer available, let it expire in the next tick
1137 */
1138 if (hr_delta.tv64 <= 0)
1139 return now + 1;
1140
1141 tsdelta = ktime_to_timespec(hr_delta);
1142 delta = timespec_to_jiffies(&tsdelta);
1143
1144 /*
1145 * Limit the delta to the max value, which is checked in
1146 * tick_nohz_stop_sched_tick():
1147 */
1148 if (delta > NEXT_TIMER_MAX_DELTA)
1149 delta = NEXT_TIMER_MAX_DELTA;
1150
1151 /*
1152 * Take rounding errors in to account and make sure, that it
1153 * expires in the next tick. Otherwise we go into an endless
1154 * ping pong due to tick_nohz_stop_sched_tick() retriggering
1155 * the timer softirq
1156 */
1157 if (delta < 1)
1158 delta = 1;
1159 now += delta;
1160 if (time_before(now, expires))
1161 return now;
1162 return expires;
1163}
1164
1165/**
1166 * get_next_timer_interrupt - return the jiffy of the next pending timer
1167 * @now: current time (in jiffies)
1168 */
1169unsigned long get_next_timer_interrupt(unsigned long now)
1170{
1171 struct tvec_base *base = __get_cpu_var(tvec_bases);
1172 unsigned long expires;
1173
1174 spin_lock(&base->lock);
1175 if (time_before_eq(base->next_timer, base->timer_jiffies))
1176 base->next_timer = __next_timer_interrupt(base);
1177 expires = base->next_timer;
1178 spin_unlock(&base->lock);
1179
1180 if (time_before_eq(expires, now))
1181 return now;
1182
1183 return cmp_next_hrtimer_event(now, expires);
1184}
1185#endif
1186
1187/*
1188 * Called from the timer interrupt handler to charge one tick to the current
1189 * process. user_tick is 1 if the tick is user time, 0 for system.
1190 */
1191void update_process_times(int user_tick)
1192{
1193 struct task_struct *p = current;
1194 int cpu = smp_processor_id();
1195
1196 /* Note: this timer irq context must be accounted for as well. */
1197 account_process_tick(p, user_tick);
1198 run_local_timers();
1199 rcu_check_callbacks(cpu, user_tick);
1200 printk_tick();
1201 scheduler_tick();
1202 run_posix_cpu_timers(p);
1203}
1204
1205/*
1206 * This function runs timers and the timer-tq in bottom half context.
1207 */
1208static void run_timer_softirq(struct softirq_action *h)
1209{
1210 struct tvec_base *base = __get_cpu_var(tvec_bases);
1211
1212 perf_event_do_pending();
1213
1214 hrtimer_run_pending();
1215
1216 if (time_after_eq(jiffies, base->timer_jiffies))
1217 __run_timers(base);
1218}
1219
1220/*
1221 * Called by the local, per-CPU timer interrupt on SMP.
1222 */
1223void run_local_timers(void)
1224{
1225 hrtimer_run_queues();
1226 raise_softirq(TIMER_SOFTIRQ);
1227 softlockup_tick();
1228}
1229
1230/*
1231 * The 64-bit jiffies value is not atomic - you MUST NOT read it
1232 * without sampling the sequence number in xtime_lock.
1233 * jiffies is defined in the linker script...
1234 */
1235
1236void do_timer(unsigned long ticks)
1237{
1238 jiffies_64 += ticks;
1239 update_wall_time();
1240 calc_global_load();
1241}
1242
1243#ifdef __ARCH_WANT_SYS_ALARM
1244
1245/*
1246 * For backwards compatibility? This can be done in libc so Alpha
1247 * and all newer ports shouldn't need it.
1248 */
1249SYSCALL_DEFINE1(alarm, unsigned int, seconds)
1250{
1251 return alarm_setitimer(seconds);
1252}
1253
1254#endif
1255
1256#ifndef __alpha__
1257
1258/*
1259 * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this
1260 * should be moved into arch/i386 instead?
1261 */
1262
1263/**
1264 * sys_getpid - return the thread group id of the current process
1265 *
1266 * Note, despite the name, this returns the tgid not the pid. The tgid and
1267 * the pid are identical unless CLONE_THREAD was specified on clone() in
1268 * which case the tgid is the same in all threads of the same group.
1269 *
1270 * This is SMP safe as current->tgid does not change.
1271 */
1272SYSCALL_DEFINE0(getpid)
1273{
1274 return task_tgid_vnr(current);
1275}
1276
1277/*
1278 * Accessing ->real_parent is not SMP-safe, it could
1279 * change from under us. However, we can use a stale
1280 * value of ->real_parent under rcu_read_lock(), see
1281 * release_task()->call_rcu(delayed_put_task_struct).
1282 */
1283SYSCALL_DEFINE0(getppid)
1284{
1285 int pid;
1286
1287 rcu_read_lock();
1288 pid = task_tgid_vnr(current->real_parent);
1289 rcu_read_unlock();
1290
1291 return pid;
1292}
1293
1294SYSCALL_DEFINE0(getuid)
1295{
1296 /* Only we change this so SMP safe */
1297 return current_uid();
1298}
1299
1300SYSCALL_DEFINE0(geteuid)
1301{
1302 /* Only we change this so SMP safe */
1303 return current_euid();
1304}
1305
1306SYSCALL_DEFINE0(getgid)
1307{
1308 /* Only we change this so SMP safe */
1309 return current_gid();
1310}
1311
1312SYSCALL_DEFINE0(getegid)
1313{
1314 /* Only we change this so SMP safe */
1315 return current_egid();
1316}
1317
1318#endif
1319
1320static void process_timeout(unsigned long __data)
1321{
1322 wake_up_process((struct task_struct *)__data);
1323}
1324
1325/**
1326 * schedule_timeout - sleep until timeout
1327 * @timeout: timeout value in jiffies
1328 *
1329 * Make the current task sleep until @timeout jiffies have
1330 * elapsed. The routine will return immediately unless
1331 * the current task state has been set (see set_current_state()).
1332 *
1333 * You can set the task state as follows -
1334 *
1335 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1336 * pass before the routine returns. The routine will return 0
1337 *
1338 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1339 * delivered to the current task. In this case the remaining time
1340 * in jiffies will be returned, or 0 if the timer expired in time
1341 *
1342 * The current task state is guaranteed to be TASK_RUNNING when this
1343 * routine returns.
1344 *
1345 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1346 * the CPU away without a bound on the timeout. In this case the return
1347 * value will be %MAX_SCHEDULE_TIMEOUT.
1348 *
1349 * In all cases the return value is guaranteed to be non-negative.
1350 */
1351signed long __sched schedule_timeout(signed long timeout)
1352{
1353 struct timer_list timer;
1354 unsigned long expire;
1355
1356 switch (timeout)
1357 {
1358 case MAX_SCHEDULE_TIMEOUT:
1359 /*
1360 * These two special cases are useful to be comfortable
1361 * in the caller. Nothing more. We could take
1362 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1363 * but I' d like to return a valid offset (>=0) to allow
1364 * the caller to do everything it want with the retval.
1365 */
1366 schedule();
1367 goto out;
1368 default:
1369 /*
1370 * Another bit of PARANOID. Note that the retval will be
1371 * 0 since no piece of kernel is supposed to do a check
1372 * for a negative retval of schedule_timeout() (since it
1373 * should never happens anyway). You just have the printk()
1374 * that will tell you if something is gone wrong and where.
1375 */
1376 if (timeout < 0) {
1377 printk(KERN_ERR "schedule_timeout: wrong timeout "
1378 "value %lx\n", timeout);
1379 dump_stack();
1380 current->state = TASK_RUNNING;
1381 goto out;
1382 }
1383 }
1384
1385 expire = timeout + jiffies;
1386
1387 setup_timer_on_stack(&timer, process_timeout, (unsigned long)current);
1388 __mod_timer(&timer, expire, false, TIMER_NOT_PINNED);
1389 schedule();
1390 del_singleshot_timer_sync(&timer);
1391
1392 /* Remove the timer from the object tracker */
1393 destroy_timer_on_stack(&timer);
1394
1395 timeout = expire - jiffies;
1396
1397 out:
1398 return timeout < 0 ? 0 : timeout;
1399}
1400EXPORT_SYMBOL(schedule_timeout);
1401
1402/*
1403 * We can use __set_current_state() here because schedule_timeout() calls
1404 * schedule() unconditionally.
1405 */
1406signed long __sched schedule_timeout_interruptible(signed long timeout)
1407{
1408 __set_current_state(TASK_INTERRUPTIBLE);
1409 return schedule_timeout(timeout);
1410}
1411EXPORT_SYMBOL(schedule_timeout_interruptible);
1412
1413signed long __sched schedule_timeout_killable(signed long timeout)
1414{
1415 __set_current_state(TASK_KILLABLE);
1416 return schedule_timeout(timeout);
1417}
1418EXPORT_SYMBOL(schedule_timeout_killable);
1419
1420signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1421{
1422 __set_current_state(TASK_UNINTERRUPTIBLE);
1423 return schedule_timeout(timeout);
1424}
1425EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1426
1427/* Thread ID - the internal kernel "pid" */
1428SYSCALL_DEFINE0(gettid)
1429{
1430 return task_pid_vnr(current);
1431}
1432
1433/**
1434 * do_sysinfo - fill in sysinfo struct
1435 * @info: pointer to buffer to fill
1436 */
1437int do_sysinfo(struct sysinfo *info)
1438{
1439 unsigned long mem_total, sav_total;
1440 unsigned int mem_unit, bitcount;
1441 struct timespec tp;
1442
1443 memset(info, 0, sizeof(struct sysinfo));
1444
1445 ktime_get_ts(&tp);
1446 monotonic_to_bootbased(&tp);
1447 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
1448
1449 get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
1450
1451 info->procs = nr_threads;
1452
1453 si_meminfo(info);
1454 si_swapinfo(info);
1455
1456 /*
1457 * If the sum of all the available memory (i.e. ram + swap)
1458 * is less than can be stored in a 32 bit unsigned long then
1459 * we can be binary compatible with 2.2.x kernels. If not,
1460 * well, in that case 2.2.x was broken anyways...
1461 *
1462 * -Erik Andersen <[email protected]>
1463 */
1464
1465 mem_total = info->totalram + info->totalswap;
1466 if (mem_total < info->totalram || mem_total < info->totalswap)
1467 goto out;
1468 bitcount = 0;
1469 mem_unit = info->mem_unit;
1470 while (mem_unit > 1) {
1471 bitcount++;
1472 mem_unit >>= 1;
1473 sav_total = mem_total;
1474 mem_total <<= 1;
1475 if (mem_total < sav_total)
1476 goto out;
1477 }
1478
1479 /*
1480 * If mem_total did not overflow, multiply all memory values by
1481 * info->mem_unit and set it to 1. This leaves things compatible
1482 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1483 * kernels...
1484 */
1485
1486 info->mem_unit = 1;
1487 info->totalram <<= bitcount;
1488 info->freeram <<= bitcount;
1489 info->sharedram <<= bitcount;
1490 info->bufferram <<= bitcount;
1491 info->totalswap <<= bitcount;
1492 info->freeswap <<= bitcount;
1493 info->totalhigh <<= bitcount;
1494 info->freehigh <<= bitcount;
1495
1496out:
1497 return 0;
1498}
1499
1500SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
1501{
1502 struct sysinfo val;
1503
1504 do_sysinfo(&val);
1505
1506 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
1507 return -EFAULT;
1508
1509 return 0;
1510}
1511
1512static int __cpuinit init_timers_cpu(int cpu)
1513{
1514 int j;
1515 struct tvec_base *base;
1516 static char __cpuinitdata tvec_base_done[NR_CPUS];
1517
1518 if (!tvec_base_done[cpu]) {
1519 static char boot_done;
1520
1521 if (boot_done) {
1522 /*
1523 * The APs use this path later in boot
1524 */
1525 base = kmalloc_node(sizeof(*base),
1526 GFP_KERNEL | __GFP_ZERO,
1527 cpu_to_node(cpu));
1528 if (!base)
1529 return -ENOMEM;
1530
1531 /* Make sure that tvec_base is 2 byte aligned */
1532 if (tbase_get_deferrable(base)) {
1533 WARN_ON(1);
1534 kfree(base);
1535 return -ENOMEM;
1536 }
1537 per_cpu(tvec_bases, cpu) = base;
1538 } else {
1539 /*
1540 * This is for the boot CPU - we use compile-time
1541 * static initialisation because per-cpu memory isn't
1542 * ready yet and because the memory allocators are not
1543 * initialised either.
1544 */
1545 boot_done = 1;
1546 base = &boot_tvec_bases;
1547 }
1548 tvec_base_done[cpu] = 1;
1549 } else {
1550 base = per_cpu(tvec_bases, cpu);
1551 }
1552
1553 spin_lock_init(&base->lock);
1554
1555 for (j = 0; j < TVN_SIZE; j++) {
1556 INIT_LIST_HEAD(base->tv5.vec + j);
1557 INIT_LIST_HEAD(base->tv4.vec + j);
1558 INIT_LIST_HEAD(base->tv3.vec + j);
1559 INIT_LIST_HEAD(base->tv2.vec + j);
1560 }
1561 for (j = 0; j < TVR_SIZE; j++)
1562 INIT_LIST_HEAD(base->tv1.vec + j);
1563
1564 base->timer_jiffies = jiffies;
1565 base->next_timer = base->timer_jiffies;
1566 return 0;
1567}
1568
1569#ifdef CONFIG_HOTPLUG_CPU
1570static void migrate_timer_list(struct tvec_base *new_base, struct list_head *head)
1571{
1572 struct timer_list *timer;
1573
1574 while (!list_empty(head)) {
1575 timer = list_first_entry(head, struct timer_list, entry);
1576 detach_timer(timer, 0);
1577 timer_set_base(timer, new_base);
1578 if (time_before(timer->expires, new_base->next_timer) &&
1579 !tbase_get_deferrable(timer->base))
1580 new_base->next_timer = timer->expires;
1581 internal_add_timer(new_base, timer);
1582 }
1583}
1584
1585static void __cpuinit migrate_timers(int cpu)
1586{
1587 struct tvec_base *old_base;
1588 struct tvec_base *new_base;
1589 int i;
1590
1591 BUG_ON(cpu_online(cpu));
1592 old_base = per_cpu(tvec_bases, cpu);
1593 new_base = get_cpu_var(tvec_bases);
1594 /*
1595 * The caller is globally serialized and nobody else
1596 * takes two locks at once, deadlock is not possible.
1597 */
1598 spin_lock_irq(&new_base->lock);
1599 spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1600
1601 BUG_ON(old_base->running_timer);
1602
1603 for (i = 0; i < TVR_SIZE; i++)
1604 migrate_timer_list(new_base, old_base->tv1.vec + i);
1605 for (i = 0; i < TVN_SIZE; i++) {
1606 migrate_timer_list(new_base, old_base->tv2.vec + i);
1607 migrate_timer_list(new_base, old_base->tv3.vec + i);
1608 migrate_timer_list(new_base, old_base->tv4.vec + i);
1609 migrate_timer_list(new_base, old_base->tv5.vec + i);
1610 }
1611
1612 spin_unlock(&old_base->lock);
1613 spin_unlock_irq(&new_base->lock);
1614 put_cpu_var(tvec_bases);
1615}
1616#endif /* CONFIG_HOTPLUG_CPU */
1617
1618static int __cpuinit timer_cpu_notify(struct notifier_block *self,
1619 unsigned long action, void *hcpu)
1620{
1621 long cpu = (long)hcpu;
1622 switch(action) {
1623 case CPU_UP_PREPARE:
1624 case CPU_UP_PREPARE_FROZEN:
1625 if (init_timers_cpu(cpu) < 0)
1626 return NOTIFY_BAD;
1627 break;
1628#ifdef CONFIG_HOTPLUG_CPU
1629 case CPU_DEAD:
1630 case CPU_DEAD_FROZEN:
1631 migrate_timers(cpu);
1632 break;
1633#endif
1634 default:
1635 break;
1636 }
1637 return NOTIFY_OK;
1638}
1639
1640static struct notifier_block __cpuinitdata timers_nb = {
1641 .notifier_call = timer_cpu_notify,
1642};
1643
1644
1645void __init init_timers(void)
1646{
1647 int err = timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE,
1648 (void *)(long)smp_processor_id());
1649
1650 init_timer_stats();
1651
1652 BUG_ON(err == NOTIFY_BAD);
1653 register_cpu_notifier(&timers_nb);
1654 open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
1655}
1656
1657/**
1658 * msleep - sleep safely even with waitqueue interruptions
1659 * @msecs: Time in milliseconds to sleep for
1660 */
1661void msleep(unsigned int msecs)
1662{
1663 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1664
1665 while (timeout)
1666 timeout = schedule_timeout_uninterruptible(timeout);
1667}
1668
1669EXPORT_SYMBOL(msleep);
1670
1671/**
1672 * msleep_interruptible - sleep waiting for signals
1673 * @msecs: Time in milliseconds to sleep for
1674 */
1675unsigned long msleep_interruptible(unsigned int msecs)
1676{
1677 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1678
1679 while (timeout && !signal_pending(current))
1680 timeout = schedule_timeout_interruptible(timeout);
1681 return jiffies_to_msecs(timeout);
1682}
1683
1684EXPORT_SYMBOL(msleep_interruptible);