]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - kernel/timer.c
ACPI: video: DMI workaround broken Acer 5315 BIOS enabling display brightness
[mirror_ubuntu-artful-kernel.git] / kernel / timer.c
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
41 #include <asm/uaccess.h>
42 #include <asm/unistd.h>
43 #include <asm/div64.h>
44 #include <asm/timex.h>
45 #include <asm/io.h>
46
47 u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
48
49 EXPORT_SYMBOL(jiffies_64);
50
51 /*
52 * per-CPU timer vector definitions:
53 */
54 #define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6)
55 #define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8)
56 #define TVN_SIZE (1 << TVN_BITS)
57 #define TVR_SIZE (1 << TVR_BITS)
58 #define TVN_MASK (TVN_SIZE - 1)
59 #define TVR_MASK (TVR_SIZE - 1)
60
61 struct tvec {
62 struct list_head vec[TVN_SIZE];
63 };
64
65 struct tvec_root {
66 struct list_head vec[TVR_SIZE];
67 };
68
69 struct tvec_base {
70 spinlock_t lock;
71 struct timer_list *running_timer;
72 unsigned long timer_jiffies;
73 struct tvec_root tv1;
74 struct tvec tv2;
75 struct tvec tv3;
76 struct tvec tv4;
77 struct tvec tv5;
78 } ____cacheline_aligned;
79
80 struct tvec_base boot_tvec_bases;
81 EXPORT_SYMBOL(boot_tvec_bases);
82 static DEFINE_PER_CPU(struct tvec_base *, tvec_bases) = &boot_tvec_bases;
83
84 /*
85 * Note that all tvec_bases are 2 byte aligned and lower bit of
86 * base in timer_list is guaranteed to be zero. Use the LSB for
87 * the new flag to indicate whether the timer is deferrable
88 */
89 #define TBASE_DEFERRABLE_FLAG (0x1)
90
91 /* Functions below help us manage 'deferrable' flag */
92 static inline unsigned int tbase_get_deferrable(struct tvec_base *base)
93 {
94 return ((unsigned int)(unsigned long)base & TBASE_DEFERRABLE_FLAG);
95 }
96
97 static 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
102 static inline void timer_set_deferrable(struct timer_list *timer)
103 {
104 timer->base = ((struct tvec_base *)((unsigned long)(timer->base) |
105 TBASE_DEFERRABLE_FLAG));
106 }
107
108 static inline void
109 timer_set_base(struct timer_list *timer, struct tvec_base *new_base)
110 {
111 timer->base = (struct tvec_base *)((unsigned long)(new_base) |
112 tbase_get_deferrable(timer->base));
113 }
114
115 static unsigned long round_jiffies_common(unsigned long j, int cpu,
116 bool force_up)
117 {
118 int rem;
119 unsigned long original = j;
120
121 /*
122 * We don't want all cpus firing their timers at once hitting the
123 * same lock or cachelines, so we skew each extra cpu with an extra
124 * 3 jiffies. This 3 jiffies came originally from the mm/ code which
125 * already did this.
126 * The skew is done by adding 3*cpunr, then round, then subtract this
127 * extra offset again.
128 */
129 j += cpu * 3;
130
131 rem = j % HZ;
132
133 /*
134 * If the target jiffie is just after a whole second (which can happen
135 * due to delays of the timer irq, long irq off times etc etc) then
136 * we should round down to the whole second, not up. Use 1/4th second
137 * as cutoff for this rounding as an extreme upper bound for this.
138 * But never round down if @force_up is set.
139 */
140 if (rem < HZ/4 && !force_up) /* round down */
141 j = j - rem;
142 else /* round up */
143 j = j - rem + HZ;
144
145 /* now that we have rounded, subtract the extra skew again */
146 j -= cpu * 3;
147
148 if (j <= jiffies) /* rounding ate our timeout entirely; */
149 return original;
150 return j;
151 }
152
153 /**
154 * __round_jiffies - function to round jiffies to a full second
155 * @j: the time in (absolute) jiffies that should be rounded
156 * @cpu: the processor number on which the timeout will happen
157 *
158 * __round_jiffies() rounds an absolute time in the future (in jiffies)
159 * up or down to (approximately) full seconds. This is useful for timers
160 * for which the exact time they fire does not matter too much, as long as
161 * they fire approximately every X seconds.
162 *
163 * By rounding these timers to whole seconds, all such timers will fire
164 * at the same time, rather than at various times spread out. The goal
165 * of this is to have the CPU wake up less, which saves power.
166 *
167 * The exact rounding is skewed for each processor to avoid all
168 * processors firing at the exact same time, which could lead
169 * to lock contention or spurious cache line bouncing.
170 *
171 * The return value is the rounded version of the @j parameter.
172 */
173 unsigned long __round_jiffies(unsigned long j, int cpu)
174 {
175 return round_jiffies_common(j, cpu, false);
176 }
177 EXPORT_SYMBOL_GPL(__round_jiffies);
178
179 /**
180 * __round_jiffies_relative - function to round jiffies to a full second
181 * @j: the time in (relative) jiffies that should be rounded
182 * @cpu: the processor number on which the timeout will happen
183 *
184 * __round_jiffies_relative() rounds a time delta in the future (in jiffies)
185 * up or down to (approximately) full seconds. This is useful for timers
186 * for which the exact time they fire does not matter too much, as long as
187 * they fire approximately every X seconds.
188 *
189 * By rounding these timers to whole seconds, all such timers will fire
190 * at the same time, rather than at various times spread out. The goal
191 * of this is to have the CPU wake up less, which saves power.
192 *
193 * The exact rounding is skewed for each processor to avoid all
194 * processors firing at the exact same time, which could lead
195 * to lock contention or spurious cache line bouncing.
196 *
197 * The return value is the rounded version of the @j parameter.
198 */
199 unsigned long __round_jiffies_relative(unsigned long j, int cpu)
200 {
201 unsigned long j0 = jiffies;
202
203 /* Use j0 because jiffies might change while we run */
204 return round_jiffies_common(j + j0, cpu, false) - j0;
205 }
206 EXPORT_SYMBOL_GPL(__round_jiffies_relative);
207
208 /**
209 * round_jiffies - function to round jiffies to a full second
210 * @j: the time in (absolute) jiffies that should be rounded
211 *
212 * round_jiffies() rounds an absolute time in the future (in jiffies)
213 * up or down to (approximately) full seconds. This is useful for timers
214 * for which the exact time they fire does not matter too much, as long as
215 * they fire approximately every X seconds.
216 *
217 * By rounding these timers to whole seconds, all such timers will fire
218 * at the same time, rather than at various times spread out. The goal
219 * of this is to have the CPU wake up less, which saves power.
220 *
221 * The return value is the rounded version of the @j parameter.
222 */
223 unsigned long round_jiffies(unsigned long j)
224 {
225 return round_jiffies_common(j, raw_smp_processor_id(), false);
226 }
227 EXPORT_SYMBOL_GPL(round_jiffies);
228
229 /**
230 * round_jiffies_relative - function to round jiffies to a full second
231 * @j: the time in (relative) jiffies that should be rounded
232 *
233 * round_jiffies_relative() rounds a time delta in the future (in jiffies)
234 * up or down to (approximately) full seconds. This is useful for timers
235 * for which the exact time they fire does not matter too much, as long as
236 * they fire approximately every X seconds.
237 *
238 * By rounding these timers to whole seconds, all such timers will fire
239 * at the same time, rather than at various times spread out. The goal
240 * of this is to have the CPU wake up less, which saves power.
241 *
242 * The return value is the rounded version of the @j parameter.
243 */
244 unsigned long round_jiffies_relative(unsigned long j)
245 {
246 return __round_jiffies_relative(j, raw_smp_processor_id());
247 }
248 EXPORT_SYMBOL_GPL(round_jiffies_relative);
249
250 /**
251 * __round_jiffies_up - function to round jiffies up to a full second
252 * @j: the time in (absolute) jiffies that should be rounded
253 * @cpu: the processor number on which the timeout will happen
254 *
255 * This is the same as __round_jiffies() except that it will never
256 * round down. This is useful for timeouts for which the exact time
257 * of firing does not matter too much, as long as they don't fire too
258 * early.
259 */
260 unsigned long __round_jiffies_up(unsigned long j, int cpu)
261 {
262 return round_jiffies_common(j, cpu, true);
263 }
264 EXPORT_SYMBOL_GPL(__round_jiffies_up);
265
266 /**
267 * __round_jiffies_up_relative - function to round jiffies up to a full second
268 * @j: the time in (relative) jiffies that should be rounded
269 * @cpu: the processor number on which the timeout will happen
270 *
271 * This is the same as __round_jiffies_relative() except that it will never
272 * round down. This is useful for timeouts for which the exact time
273 * of firing does not matter too much, as long as they don't fire too
274 * early.
275 */
276 unsigned long __round_jiffies_up_relative(unsigned long j, int cpu)
277 {
278 unsigned long j0 = jiffies;
279
280 /* Use j0 because jiffies might change while we run */
281 return round_jiffies_common(j + j0, cpu, true) - j0;
282 }
283 EXPORT_SYMBOL_GPL(__round_jiffies_up_relative);
284
285 /**
286 * round_jiffies_up - function to round jiffies up to a full second
287 * @j: the time in (absolute) jiffies that should be rounded
288 *
289 * This is the same as round_jiffies() except that it will never
290 * round down. This is useful for timeouts for which the exact time
291 * of firing does not matter too much, as long as they don't fire too
292 * early.
293 */
294 unsigned long round_jiffies_up(unsigned long j)
295 {
296 return round_jiffies_common(j, raw_smp_processor_id(), true);
297 }
298 EXPORT_SYMBOL_GPL(round_jiffies_up);
299
300 /**
301 * round_jiffies_up_relative - function to round jiffies up to a full second
302 * @j: the time in (relative) jiffies that should be rounded
303 *
304 * This is the same as round_jiffies_relative() except that it will never
305 * round down. This is useful for timeouts for which the exact time
306 * of firing does not matter too much, as long as they don't fire too
307 * early.
308 */
309 unsigned long round_jiffies_up_relative(unsigned long j)
310 {
311 return __round_jiffies_up_relative(j, raw_smp_processor_id());
312 }
313 EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
314
315
316 static inline void set_running_timer(struct tvec_base *base,
317 struct timer_list *timer)
318 {
319 #ifdef CONFIG_SMP
320 base->running_timer = timer;
321 #endif
322 }
323
324 static void internal_add_timer(struct tvec_base *base, struct timer_list *timer)
325 {
326 unsigned long expires = timer->expires;
327 unsigned long idx = expires - base->timer_jiffies;
328 struct list_head *vec;
329
330 if (idx < TVR_SIZE) {
331 int i = expires & TVR_MASK;
332 vec = base->tv1.vec + i;
333 } else if (idx < 1 << (TVR_BITS + TVN_BITS)) {
334 int i = (expires >> TVR_BITS) & TVN_MASK;
335 vec = base->tv2.vec + i;
336 } else if (idx < 1 << (TVR_BITS + 2 * TVN_BITS)) {
337 int i = (expires >> (TVR_BITS + TVN_BITS)) & TVN_MASK;
338 vec = base->tv3.vec + i;
339 } else if (idx < 1 << (TVR_BITS + 3 * TVN_BITS)) {
340 int i = (expires >> (TVR_BITS + 2 * TVN_BITS)) & TVN_MASK;
341 vec = base->tv4.vec + i;
342 } else if ((signed long) idx < 0) {
343 /*
344 * Can happen if you add a timer with expires == jiffies,
345 * or you set a timer to go off in the past
346 */
347 vec = base->tv1.vec + (base->timer_jiffies & TVR_MASK);
348 } else {
349 int i;
350 /* If the timeout is larger than 0xffffffff on 64-bit
351 * architectures then we use the maximum timeout:
352 */
353 if (idx > 0xffffffffUL) {
354 idx = 0xffffffffUL;
355 expires = idx + base->timer_jiffies;
356 }
357 i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK;
358 vec = base->tv5.vec + i;
359 }
360 /*
361 * Timers are FIFO:
362 */
363 list_add_tail(&timer->entry, vec);
364 }
365
366 #ifdef CONFIG_TIMER_STATS
367 void __timer_stats_timer_set_start_info(struct timer_list *timer, void *addr)
368 {
369 if (timer->start_site)
370 return;
371
372 timer->start_site = addr;
373 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
374 timer->start_pid = current->pid;
375 }
376
377 static void timer_stats_account_timer(struct timer_list *timer)
378 {
379 unsigned int flag = 0;
380
381 if (unlikely(tbase_get_deferrable(timer->base)))
382 flag |= TIMER_STATS_FLAG_DEFERRABLE;
383
384 timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
385 timer->function, timer->start_comm, flag);
386 }
387
388 #else
389 static void timer_stats_account_timer(struct timer_list *timer) {}
390 #endif
391
392 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
393
394 static struct debug_obj_descr timer_debug_descr;
395
396 /*
397 * fixup_init is called when:
398 * - an active object is initialized
399 */
400 static int timer_fixup_init(void *addr, enum debug_obj_state state)
401 {
402 struct timer_list *timer = addr;
403
404 switch (state) {
405 case ODEBUG_STATE_ACTIVE:
406 del_timer_sync(timer);
407 debug_object_init(timer, &timer_debug_descr);
408 return 1;
409 default:
410 return 0;
411 }
412 }
413
414 /*
415 * fixup_activate is called when:
416 * - an active object is activated
417 * - an unknown object is activated (might be a statically initialized object)
418 */
419 static int timer_fixup_activate(void *addr, enum debug_obj_state state)
420 {
421 struct timer_list *timer = addr;
422
423 switch (state) {
424
425 case ODEBUG_STATE_NOTAVAILABLE:
426 /*
427 * This is not really a fixup. The timer was
428 * statically initialized. We just make sure that it
429 * is tracked in the object tracker.
430 */
431 if (timer->entry.next == NULL &&
432 timer->entry.prev == TIMER_ENTRY_STATIC) {
433 debug_object_init(timer, &timer_debug_descr);
434 debug_object_activate(timer, &timer_debug_descr);
435 return 0;
436 } else {
437 WARN_ON_ONCE(1);
438 }
439 return 0;
440
441 case ODEBUG_STATE_ACTIVE:
442 WARN_ON(1);
443
444 default:
445 return 0;
446 }
447 }
448
449 /*
450 * fixup_free is called when:
451 * - an active object is freed
452 */
453 static int timer_fixup_free(void *addr, enum debug_obj_state state)
454 {
455 struct timer_list *timer = addr;
456
457 switch (state) {
458 case ODEBUG_STATE_ACTIVE:
459 del_timer_sync(timer);
460 debug_object_free(timer, &timer_debug_descr);
461 return 1;
462 default:
463 return 0;
464 }
465 }
466
467 static struct debug_obj_descr timer_debug_descr = {
468 .name = "timer_list",
469 .fixup_init = timer_fixup_init,
470 .fixup_activate = timer_fixup_activate,
471 .fixup_free = timer_fixup_free,
472 };
473
474 static inline void debug_timer_init(struct timer_list *timer)
475 {
476 debug_object_init(timer, &timer_debug_descr);
477 }
478
479 static inline void debug_timer_activate(struct timer_list *timer)
480 {
481 debug_object_activate(timer, &timer_debug_descr);
482 }
483
484 static inline void debug_timer_deactivate(struct timer_list *timer)
485 {
486 debug_object_deactivate(timer, &timer_debug_descr);
487 }
488
489 static inline void debug_timer_free(struct timer_list *timer)
490 {
491 debug_object_free(timer, &timer_debug_descr);
492 }
493
494 static void __init_timer(struct timer_list *timer,
495 const char *name,
496 struct lock_class_key *key);
497
498 void init_timer_on_stack_key(struct timer_list *timer,
499 const char *name,
500 struct lock_class_key *key)
501 {
502 debug_object_init_on_stack(timer, &timer_debug_descr);
503 __init_timer(timer, name, key);
504 }
505 EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
506
507 void destroy_timer_on_stack(struct timer_list *timer)
508 {
509 debug_object_free(timer, &timer_debug_descr);
510 }
511 EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
512
513 #else
514 static inline void debug_timer_init(struct timer_list *timer) { }
515 static inline void debug_timer_activate(struct timer_list *timer) { }
516 static inline void debug_timer_deactivate(struct timer_list *timer) { }
517 #endif
518
519 static void __init_timer(struct timer_list *timer,
520 const char *name,
521 struct lock_class_key *key)
522 {
523 timer->entry.next = NULL;
524 timer->base = __raw_get_cpu_var(tvec_bases);
525 #ifdef CONFIG_TIMER_STATS
526 timer->start_site = NULL;
527 timer->start_pid = -1;
528 memset(timer->start_comm, 0, TASK_COMM_LEN);
529 #endif
530 lockdep_init_map(&timer->lockdep_map, name, key, 0);
531 }
532
533 /**
534 * init_timer_key - initialize a timer
535 * @timer: the timer to be initialized
536 * @name: name of the timer
537 * @key: lockdep class key of the fake lock used for tracking timer
538 * sync lock dependencies
539 *
540 * init_timer_key() must be done to a timer prior calling *any* of the
541 * other timer functions.
542 */
543 void init_timer_key(struct timer_list *timer,
544 const char *name,
545 struct lock_class_key *key)
546 {
547 debug_timer_init(timer);
548 __init_timer(timer, name, key);
549 }
550 EXPORT_SYMBOL(init_timer_key);
551
552 void init_timer_deferrable_key(struct timer_list *timer,
553 const char *name,
554 struct lock_class_key *key)
555 {
556 init_timer_key(timer, name, key);
557 timer_set_deferrable(timer);
558 }
559 EXPORT_SYMBOL(init_timer_deferrable_key);
560
561 static inline void detach_timer(struct timer_list *timer,
562 int clear_pending)
563 {
564 struct list_head *entry = &timer->entry;
565
566 debug_timer_deactivate(timer);
567
568 __list_del(entry->prev, entry->next);
569 if (clear_pending)
570 entry->next = NULL;
571 entry->prev = LIST_POISON2;
572 }
573
574 /*
575 * We are using hashed locking: holding per_cpu(tvec_bases).lock
576 * means that all timers which are tied to this base via timer->base are
577 * locked, and the base itself is locked too.
578 *
579 * So __run_timers/migrate_timers can safely modify all timers which could
580 * be found on ->tvX lists.
581 *
582 * When the timer's base is locked, and the timer removed from list, it is
583 * possible to set timer->base = NULL and drop the lock: the timer remains
584 * locked.
585 */
586 static struct tvec_base *lock_timer_base(struct timer_list *timer,
587 unsigned long *flags)
588 __acquires(timer->base->lock)
589 {
590 struct tvec_base *base;
591
592 for (;;) {
593 struct tvec_base *prelock_base = timer->base;
594 base = tbase_get_base(prelock_base);
595 if (likely(base != NULL)) {
596 spin_lock_irqsave(&base->lock, *flags);
597 if (likely(prelock_base == timer->base))
598 return base;
599 /* The timer has migrated to another CPU */
600 spin_unlock_irqrestore(&base->lock, *flags);
601 }
602 cpu_relax();
603 }
604 }
605
606 static inline int
607 __mod_timer(struct timer_list *timer, unsigned long expires, bool pending_only)
608 {
609 struct tvec_base *base, *new_base;
610 unsigned long flags;
611 int ret;
612
613 ret = 0;
614
615 timer_stats_timer_set_start_info(timer);
616 BUG_ON(!timer->function);
617
618 base = lock_timer_base(timer, &flags);
619
620 if (timer_pending(timer)) {
621 detach_timer(timer, 0);
622 ret = 1;
623 } else {
624 if (pending_only)
625 goto out_unlock;
626 }
627
628 debug_timer_activate(timer);
629
630 new_base = __get_cpu_var(tvec_bases);
631
632 if (base != new_base) {
633 /*
634 * We are trying to schedule the timer on the local CPU.
635 * However we can't change timer's base while it is running,
636 * otherwise del_timer_sync() can't detect that the timer's
637 * handler yet has not finished. This also guarantees that
638 * the timer is serialized wrt itself.
639 */
640 if (likely(base->running_timer != timer)) {
641 /* See the comment in lock_timer_base() */
642 timer_set_base(timer, NULL);
643 spin_unlock(&base->lock);
644 base = new_base;
645 spin_lock(&base->lock);
646 timer_set_base(timer, base);
647 }
648 }
649
650 timer->expires = expires;
651 internal_add_timer(base, timer);
652
653 out_unlock:
654 spin_unlock_irqrestore(&base->lock, flags);
655
656 return ret;
657 }
658
659 /**
660 * mod_timer_pending - modify a pending timer's timeout
661 * @timer: the pending timer to be modified
662 * @expires: new timeout in jiffies
663 *
664 * mod_timer_pending() is the same for pending timers as mod_timer(),
665 * but will not re-activate and modify already deleted timers.
666 *
667 * It is useful for unserialized use of timers.
668 */
669 int mod_timer_pending(struct timer_list *timer, unsigned long expires)
670 {
671 return __mod_timer(timer, expires, true);
672 }
673 EXPORT_SYMBOL(mod_timer_pending);
674
675 /**
676 * mod_timer - modify a timer's timeout
677 * @timer: the timer to be modified
678 * @expires: new timeout in jiffies
679 *
680 * mod_timer() is a more efficient way to update the expire field of an
681 * active timer (if the timer is inactive it will be activated)
682 *
683 * mod_timer(timer, expires) is equivalent to:
684 *
685 * del_timer(timer); timer->expires = expires; add_timer(timer);
686 *
687 * Note that if there are multiple unserialized concurrent users of the
688 * same timer, then mod_timer() is the only safe way to modify the timeout,
689 * since add_timer() cannot modify an already running timer.
690 *
691 * The function returns whether it has modified a pending timer or not.
692 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
693 * active timer returns 1.)
694 */
695 int mod_timer(struct timer_list *timer, unsigned long expires)
696 {
697 /*
698 * This is a common optimization triggered by the
699 * networking code - if the timer is re-modified
700 * to be the same thing then just return:
701 */
702 if (timer->expires == expires && timer_pending(timer))
703 return 1;
704
705 return __mod_timer(timer, expires, false);
706 }
707 EXPORT_SYMBOL(mod_timer);
708
709 /**
710 * add_timer - start a timer
711 * @timer: the timer to be added
712 *
713 * The kernel will do a ->function(->data) callback from the
714 * timer interrupt at the ->expires point in the future. The
715 * current time is 'jiffies'.
716 *
717 * The timer's ->expires, ->function (and if the handler uses it, ->data)
718 * fields must be set prior calling this function.
719 *
720 * Timers with an ->expires field in the past will be executed in the next
721 * timer tick.
722 */
723 void add_timer(struct timer_list *timer)
724 {
725 BUG_ON(timer_pending(timer));
726 mod_timer(timer, timer->expires);
727 }
728 EXPORT_SYMBOL(add_timer);
729
730 /**
731 * add_timer_on - start a timer on a particular CPU
732 * @timer: the timer to be added
733 * @cpu: the CPU to start it on
734 *
735 * This is not very scalable on SMP. Double adds are not possible.
736 */
737 void add_timer_on(struct timer_list *timer, int cpu)
738 {
739 struct tvec_base *base = per_cpu(tvec_bases, cpu);
740 unsigned long flags;
741
742 timer_stats_timer_set_start_info(timer);
743 BUG_ON(timer_pending(timer) || !timer->function);
744 spin_lock_irqsave(&base->lock, flags);
745 timer_set_base(timer, base);
746 debug_timer_activate(timer);
747 internal_add_timer(base, timer);
748 /*
749 * Check whether the other CPU is idle and needs to be
750 * triggered to reevaluate the timer wheel when nohz is
751 * active. We are protected against the other CPU fiddling
752 * with the timer by holding the timer base lock. This also
753 * makes sure that a CPU on the way to idle can not evaluate
754 * the timer wheel.
755 */
756 wake_up_idle_cpu(cpu);
757 spin_unlock_irqrestore(&base->lock, flags);
758 }
759
760 /**
761 * del_timer - deactive a timer.
762 * @timer: the timer to be deactivated
763 *
764 * del_timer() deactivates a timer - this works on both active and inactive
765 * timers.
766 *
767 * The function returns whether it has deactivated a pending timer or not.
768 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
769 * active timer returns 1.)
770 */
771 int del_timer(struct timer_list *timer)
772 {
773 struct tvec_base *base;
774 unsigned long flags;
775 int ret = 0;
776
777 timer_stats_timer_clear_start_info(timer);
778 if (timer_pending(timer)) {
779 base = lock_timer_base(timer, &flags);
780 if (timer_pending(timer)) {
781 detach_timer(timer, 1);
782 ret = 1;
783 }
784 spin_unlock_irqrestore(&base->lock, flags);
785 }
786
787 return ret;
788 }
789 EXPORT_SYMBOL(del_timer);
790
791 #ifdef CONFIG_SMP
792 /**
793 * try_to_del_timer_sync - Try to deactivate a timer
794 * @timer: timer do del
795 *
796 * This function tries to deactivate a timer. Upon successful (ret >= 0)
797 * exit the timer is not queued and the handler is not running on any CPU.
798 *
799 * It must not be called from interrupt contexts.
800 */
801 int try_to_del_timer_sync(struct timer_list *timer)
802 {
803 struct tvec_base *base;
804 unsigned long flags;
805 int ret = -1;
806
807 base = lock_timer_base(timer, &flags);
808
809 if (base->running_timer == timer)
810 goto out;
811
812 ret = 0;
813 if (timer_pending(timer)) {
814 detach_timer(timer, 1);
815 ret = 1;
816 }
817 out:
818 spin_unlock_irqrestore(&base->lock, flags);
819
820 return ret;
821 }
822 EXPORT_SYMBOL(try_to_del_timer_sync);
823
824 /**
825 * del_timer_sync - deactivate a timer and wait for the handler to finish.
826 * @timer: the timer to be deactivated
827 *
828 * This function only differs from del_timer() on SMP: besides deactivating
829 * the timer it also makes sure the handler has finished executing on other
830 * CPUs.
831 *
832 * Synchronization rules: Callers must prevent restarting of the timer,
833 * otherwise this function is meaningless. It must not be called from
834 * interrupt contexts. The caller must not hold locks which would prevent
835 * completion of the timer's handler. The timer's handler must not call
836 * add_timer_on(). Upon exit the timer is not queued and the handler is
837 * not running on any CPU.
838 *
839 * The function returns whether it has deactivated a pending timer or not.
840 */
841 int del_timer_sync(struct timer_list *timer)
842 {
843 #ifdef CONFIG_LOCKDEP
844 unsigned long flags;
845
846 local_irq_save(flags);
847 lock_map_acquire(&timer->lockdep_map);
848 lock_map_release(&timer->lockdep_map);
849 local_irq_restore(flags);
850 #endif
851
852 for (;;) {
853 int ret = try_to_del_timer_sync(timer);
854 if (ret >= 0)
855 return ret;
856 cpu_relax();
857 }
858 }
859 EXPORT_SYMBOL(del_timer_sync);
860 #endif
861
862 static int cascade(struct tvec_base *base, struct tvec *tv, int index)
863 {
864 /* cascade all the timers from tv up one level */
865 struct timer_list *timer, *tmp;
866 struct list_head tv_list;
867
868 list_replace_init(tv->vec + index, &tv_list);
869
870 /*
871 * We are removing _all_ timers from the list, so we
872 * don't have to detach them individually.
873 */
874 list_for_each_entry_safe(timer, tmp, &tv_list, entry) {
875 BUG_ON(tbase_get_base(timer->base) != base);
876 internal_add_timer(base, timer);
877 }
878
879 return index;
880 }
881
882 #define INDEX(N) ((base->timer_jiffies >> (TVR_BITS + (N) * TVN_BITS)) & TVN_MASK)
883
884 /**
885 * __run_timers - run all expired timers (if any) on this CPU.
886 * @base: the timer vector to be processed.
887 *
888 * This function cascades all vectors and executes all expired timer
889 * vectors.
890 */
891 static inline void __run_timers(struct tvec_base *base)
892 {
893 struct timer_list *timer;
894
895 spin_lock_irq(&base->lock);
896 while (time_after_eq(jiffies, base->timer_jiffies)) {
897 struct list_head work_list;
898 struct list_head *head = &work_list;
899 int index = base->timer_jiffies & TVR_MASK;
900
901 /*
902 * Cascade timers:
903 */
904 if (!index &&
905 (!cascade(base, &base->tv2, INDEX(0))) &&
906 (!cascade(base, &base->tv3, INDEX(1))) &&
907 !cascade(base, &base->tv4, INDEX(2)))
908 cascade(base, &base->tv5, INDEX(3));
909 ++base->timer_jiffies;
910 list_replace_init(base->tv1.vec + index, &work_list);
911 while (!list_empty(head)) {
912 void (*fn)(unsigned long);
913 unsigned long data;
914
915 timer = list_first_entry(head, struct timer_list,entry);
916 fn = timer->function;
917 data = timer->data;
918
919 timer_stats_account_timer(timer);
920
921 set_running_timer(base, timer);
922 detach_timer(timer, 1);
923
924 spin_unlock_irq(&base->lock);
925 {
926 int preempt_count = preempt_count();
927
928 #ifdef CONFIG_LOCKDEP
929 /*
930 * It is permissible to free the timer from
931 * inside the function that is called from
932 * it, this we need to take into account for
933 * lockdep too. To avoid bogus "held lock
934 * freed" warnings as well as problems when
935 * looking into timer->lockdep_map, make a
936 * copy and use that here.
937 */
938 struct lockdep_map lockdep_map =
939 timer->lockdep_map;
940 #endif
941 /*
942 * Couple the lock chain with the lock chain at
943 * del_timer_sync() by acquiring the lock_map
944 * around the fn() call here and in
945 * del_timer_sync().
946 */
947 lock_map_acquire(&lockdep_map);
948
949 fn(data);
950
951 lock_map_release(&lockdep_map);
952
953 if (preempt_count != preempt_count()) {
954 printk(KERN_ERR "huh, entered %p "
955 "with preempt_count %08x, exited"
956 " with %08x?\n",
957 fn, preempt_count,
958 preempt_count());
959 BUG();
960 }
961 }
962 spin_lock_irq(&base->lock);
963 }
964 }
965 set_running_timer(base, NULL);
966 spin_unlock_irq(&base->lock);
967 }
968
969 #ifdef CONFIG_NO_HZ
970 /*
971 * Find out when the next timer event is due to happen. This
972 * is used on S/390 to stop all activity when a cpus is idle.
973 * This functions needs to be called disabled.
974 */
975 static unsigned long __next_timer_interrupt(struct tvec_base *base)
976 {
977 unsigned long timer_jiffies = base->timer_jiffies;
978 unsigned long expires = timer_jiffies + NEXT_TIMER_MAX_DELTA;
979 int index, slot, array, found = 0;
980 struct timer_list *nte;
981 struct tvec *varray[4];
982
983 /* Look for timer events in tv1. */
984 index = slot = timer_jiffies & TVR_MASK;
985 do {
986 list_for_each_entry(nte, base->tv1.vec + slot, entry) {
987 if (tbase_get_deferrable(nte->base))
988 continue;
989
990 found = 1;
991 expires = nte->expires;
992 /* Look at the cascade bucket(s)? */
993 if (!index || slot < index)
994 goto cascade;
995 return expires;
996 }
997 slot = (slot + 1) & TVR_MASK;
998 } while (slot != index);
999
1000 cascade:
1001 /* Calculate the next cascade event */
1002 if (index)
1003 timer_jiffies += TVR_SIZE - index;
1004 timer_jiffies >>= TVR_BITS;
1005
1006 /* Check tv2-tv5. */
1007 varray[0] = &base->tv2;
1008 varray[1] = &base->tv3;
1009 varray[2] = &base->tv4;
1010 varray[3] = &base->tv5;
1011
1012 for (array = 0; array < 4; array++) {
1013 struct tvec *varp = varray[array];
1014
1015 index = slot = timer_jiffies & TVN_MASK;
1016 do {
1017 list_for_each_entry(nte, varp->vec + slot, entry) {
1018 found = 1;
1019 if (time_before(nte->expires, expires))
1020 expires = nte->expires;
1021 }
1022 /*
1023 * Do we still search for the first timer or are
1024 * we looking up the cascade buckets ?
1025 */
1026 if (found) {
1027 /* Look at the cascade bucket(s)? */
1028 if (!index || slot < index)
1029 break;
1030 return expires;
1031 }
1032 slot = (slot + 1) & TVN_MASK;
1033 } while (slot != index);
1034
1035 if (index)
1036 timer_jiffies += TVN_SIZE - index;
1037 timer_jiffies >>= TVN_BITS;
1038 }
1039 return expires;
1040 }
1041
1042 /*
1043 * Check, if the next hrtimer event is before the next timer wheel
1044 * event:
1045 */
1046 static unsigned long cmp_next_hrtimer_event(unsigned long now,
1047 unsigned long expires)
1048 {
1049 ktime_t hr_delta = hrtimer_get_next_event();
1050 struct timespec tsdelta;
1051 unsigned long delta;
1052
1053 if (hr_delta.tv64 == KTIME_MAX)
1054 return expires;
1055
1056 /*
1057 * Expired timer available, let it expire in the next tick
1058 */
1059 if (hr_delta.tv64 <= 0)
1060 return now + 1;
1061
1062 tsdelta = ktime_to_timespec(hr_delta);
1063 delta = timespec_to_jiffies(&tsdelta);
1064
1065 /*
1066 * Limit the delta to the max value, which is checked in
1067 * tick_nohz_stop_sched_tick():
1068 */
1069 if (delta > NEXT_TIMER_MAX_DELTA)
1070 delta = NEXT_TIMER_MAX_DELTA;
1071
1072 /*
1073 * Take rounding errors in to account and make sure, that it
1074 * expires in the next tick. Otherwise we go into an endless
1075 * ping pong due to tick_nohz_stop_sched_tick() retriggering
1076 * the timer softirq
1077 */
1078 if (delta < 1)
1079 delta = 1;
1080 now += delta;
1081 if (time_before(now, expires))
1082 return now;
1083 return expires;
1084 }
1085
1086 /**
1087 * get_next_timer_interrupt - return the jiffy of the next pending timer
1088 * @now: current time (in jiffies)
1089 */
1090 unsigned long get_next_timer_interrupt(unsigned long now)
1091 {
1092 struct tvec_base *base = __get_cpu_var(tvec_bases);
1093 unsigned long expires;
1094
1095 spin_lock(&base->lock);
1096 expires = __next_timer_interrupt(base);
1097 spin_unlock(&base->lock);
1098
1099 if (time_before_eq(expires, now))
1100 return now;
1101
1102 return cmp_next_hrtimer_event(now, expires);
1103 }
1104 #endif
1105
1106 /*
1107 * Called from the timer interrupt handler to charge one tick to the current
1108 * process. user_tick is 1 if the tick is user time, 0 for system.
1109 */
1110 void update_process_times(int user_tick)
1111 {
1112 struct task_struct *p = current;
1113 int cpu = smp_processor_id();
1114
1115 /* Note: this timer irq context must be accounted for as well. */
1116 account_process_tick(p, user_tick);
1117 run_local_timers();
1118 if (rcu_pending(cpu))
1119 rcu_check_callbacks(cpu, user_tick);
1120 printk_tick();
1121 scheduler_tick();
1122 run_posix_cpu_timers(p);
1123 }
1124
1125 /*
1126 * Nr of active tasks - counted in fixed-point numbers
1127 */
1128 static unsigned long count_active_tasks(void)
1129 {
1130 return nr_active() * FIXED_1;
1131 }
1132
1133 /*
1134 * Hmm.. Changed this, as the GNU make sources (load.c) seems to
1135 * imply that avenrun[] is the standard name for this kind of thing.
1136 * Nothing else seems to be standardized: the fractional size etc
1137 * all seem to differ on different machines.
1138 *
1139 * Requires xtime_lock to access.
1140 */
1141 unsigned long avenrun[3];
1142
1143 EXPORT_SYMBOL(avenrun);
1144
1145 /*
1146 * calc_load - given tick count, update the avenrun load estimates.
1147 * This is called while holding a write_lock on xtime_lock.
1148 */
1149 static inline void calc_load(unsigned long ticks)
1150 {
1151 unsigned long active_tasks; /* fixed-point */
1152 static int count = LOAD_FREQ;
1153
1154 count -= ticks;
1155 if (unlikely(count < 0)) {
1156 active_tasks = count_active_tasks();
1157 do {
1158 CALC_LOAD(avenrun[0], EXP_1, active_tasks);
1159 CALC_LOAD(avenrun[1], EXP_5, active_tasks);
1160 CALC_LOAD(avenrun[2], EXP_15, active_tasks);
1161 count += LOAD_FREQ;
1162 } while (count < 0);
1163 }
1164 }
1165
1166 /*
1167 * This function runs timers and the timer-tq in bottom half context.
1168 */
1169 static void run_timer_softirq(struct softirq_action *h)
1170 {
1171 struct tvec_base *base = __get_cpu_var(tvec_bases);
1172
1173 hrtimer_run_pending();
1174
1175 if (time_after_eq(jiffies, base->timer_jiffies))
1176 __run_timers(base);
1177 }
1178
1179 /*
1180 * Called by the local, per-CPU timer interrupt on SMP.
1181 */
1182 void run_local_timers(void)
1183 {
1184 hrtimer_run_queues();
1185 raise_softirq(TIMER_SOFTIRQ);
1186 softlockup_tick();
1187 }
1188
1189 /*
1190 * Called by the timer interrupt. xtime_lock must already be taken
1191 * by the timer IRQ!
1192 */
1193 static inline void update_times(unsigned long ticks)
1194 {
1195 update_wall_time();
1196 calc_load(ticks);
1197 }
1198
1199 /*
1200 * The 64-bit jiffies value is not atomic - you MUST NOT read it
1201 * without sampling the sequence number in xtime_lock.
1202 * jiffies is defined in the linker script...
1203 */
1204
1205 void do_timer(unsigned long ticks)
1206 {
1207 jiffies_64 += ticks;
1208 update_times(ticks);
1209 }
1210
1211 #ifdef __ARCH_WANT_SYS_ALARM
1212
1213 /*
1214 * For backwards compatibility? This can be done in libc so Alpha
1215 * and all newer ports shouldn't need it.
1216 */
1217 SYSCALL_DEFINE1(alarm, unsigned int, seconds)
1218 {
1219 return alarm_setitimer(seconds);
1220 }
1221
1222 #endif
1223
1224 #ifndef __alpha__
1225
1226 /*
1227 * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this
1228 * should be moved into arch/i386 instead?
1229 */
1230
1231 /**
1232 * sys_getpid - return the thread group id of the current process
1233 *
1234 * Note, despite the name, this returns the tgid not the pid. The tgid and
1235 * the pid are identical unless CLONE_THREAD was specified on clone() in
1236 * which case the tgid is the same in all threads of the same group.
1237 *
1238 * This is SMP safe as current->tgid does not change.
1239 */
1240 SYSCALL_DEFINE0(getpid)
1241 {
1242 return task_tgid_vnr(current);
1243 }
1244
1245 /*
1246 * Accessing ->real_parent is not SMP-safe, it could
1247 * change from under us. However, we can use a stale
1248 * value of ->real_parent under rcu_read_lock(), see
1249 * release_task()->call_rcu(delayed_put_task_struct).
1250 */
1251 SYSCALL_DEFINE0(getppid)
1252 {
1253 int pid;
1254
1255 rcu_read_lock();
1256 pid = task_tgid_vnr(current->real_parent);
1257 rcu_read_unlock();
1258
1259 return pid;
1260 }
1261
1262 SYSCALL_DEFINE0(getuid)
1263 {
1264 /* Only we change this so SMP safe */
1265 return current_uid();
1266 }
1267
1268 SYSCALL_DEFINE0(geteuid)
1269 {
1270 /* Only we change this so SMP safe */
1271 return current_euid();
1272 }
1273
1274 SYSCALL_DEFINE0(getgid)
1275 {
1276 /* Only we change this so SMP safe */
1277 return current_gid();
1278 }
1279
1280 SYSCALL_DEFINE0(getegid)
1281 {
1282 /* Only we change this so SMP safe */
1283 return current_egid();
1284 }
1285
1286 #endif
1287
1288 static void process_timeout(unsigned long __data)
1289 {
1290 wake_up_process((struct task_struct *)__data);
1291 }
1292
1293 /**
1294 * schedule_timeout - sleep until timeout
1295 * @timeout: timeout value in jiffies
1296 *
1297 * Make the current task sleep until @timeout jiffies have
1298 * elapsed. The routine will return immediately unless
1299 * the current task state has been set (see set_current_state()).
1300 *
1301 * You can set the task state as follows -
1302 *
1303 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1304 * pass before the routine returns. The routine will return 0
1305 *
1306 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1307 * delivered to the current task. In this case the remaining time
1308 * in jiffies will be returned, or 0 if the timer expired in time
1309 *
1310 * The current task state is guaranteed to be TASK_RUNNING when this
1311 * routine returns.
1312 *
1313 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1314 * the CPU away without a bound on the timeout. In this case the return
1315 * value will be %MAX_SCHEDULE_TIMEOUT.
1316 *
1317 * In all cases the return value is guaranteed to be non-negative.
1318 */
1319 signed long __sched schedule_timeout(signed long timeout)
1320 {
1321 struct timer_list timer;
1322 unsigned long expire;
1323
1324 switch (timeout)
1325 {
1326 case MAX_SCHEDULE_TIMEOUT:
1327 /*
1328 * These two special cases are useful to be comfortable
1329 * in the caller. Nothing more. We could take
1330 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1331 * but I' d like to return a valid offset (>=0) to allow
1332 * the caller to do everything it want with the retval.
1333 */
1334 schedule();
1335 goto out;
1336 default:
1337 /*
1338 * Another bit of PARANOID. Note that the retval will be
1339 * 0 since no piece of kernel is supposed to do a check
1340 * for a negative retval of schedule_timeout() (since it
1341 * should never happens anyway). You just have the printk()
1342 * that will tell you if something is gone wrong and where.
1343 */
1344 if (timeout < 0) {
1345 printk(KERN_ERR "schedule_timeout: wrong timeout "
1346 "value %lx\n", timeout);
1347 dump_stack();
1348 current->state = TASK_RUNNING;
1349 goto out;
1350 }
1351 }
1352
1353 expire = timeout + jiffies;
1354
1355 setup_timer_on_stack(&timer, process_timeout, (unsigned long)current);
1356 __mod_timer(&timer, expire, false);
1357 schedule();
1358 del_singleshot_timer_sync(&timer);
1359
1360 /* Remove the timer from the object tracker */
1361 destroy_timer_on_stack(&timer);
1362
1363 timeout = expire - jiffies;
1364
1365 out:
1366 return timeout < 0 ? 0 : timeout;
1367 }
1368 EXPORT_SYMBOL(schedule_timeout);
1369
1370 /*
1371 * We can use __set_current_state() here because schedule_timeout() calls
1372 * schedule() unconditionally.
1373 */
1374 signed long __sched schedule_timeout_interruptible(signed long timeout)
1375 {
1376 __set_current_state(TASK_INTERRUPTIBLE);
1377 return schedule_timeout(timeout);
1378 }
1379 EXPORT_SYMBOL(schedule_timeout_interruptible);
1380
1381 signed long __sched schedule_timeout_killable(signed long timeout)
1382 {
1383 __set_current_state(TASK_KILLABLE);
1384 return schedule_timeout(timeout);
1385 }
1386 EXPORT_SYMBOL(schedule_timeout_killable);
1387
1388 signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1389 {
1390 __set_current_state(TASK_UNINTERRUPTIBLE);
1391 return schedule_timeout(timeout);
1392 }
1393 EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1394
1395 /* Thread ID - the internal kernel "pid" */
1396 SYSCALL_DEFINE0(gettid)
1397 {
1398 return task_pid_vnr(current);
1399 }
1400
1401 /**
1402 * do_sysinfo - fill in sysinfo struct
1403 * @info: pointer to buffer to fill
1404 */
1405 int do_sysinfo(struct sysinfo *info)
1406 {
1407 unsigned long mem_total, sav_total;
1408 unsigned int mem_unit, bitcount;
1409 unsigned long seq;
1410
1411 memset(info, 0, sizeof(struct sysinfo));
1412
1413 do {
1414 struct timespec tp;
1415 seq = read_seqbegin(&xtime_lock);
1416
1417 /*
1418 * This is annoying. The below is the same thing
1419 * posix_get_clock_monotonic() does, but it wants to
1420 * take the lock which we want to cover the loads stuff
1421 * too.
1422 */
1423
1424 getnstimeofday(&tp);
1425 tp.tv_sec += wall_to_monotonic.tv_sec;
1426 tp.tv_nsec += wall_to_monotonic.tv_nsec;
1427 monotonic_to_bootbased(&tp);
1428 if (tp.tv_nsec - NSEC_PER_SEC >= 0) {
1429 tp.tv_nsec = tp.tv_nsec - NSEC_PER_SEC;
1430 tp.tv_sec++;
1431 }
1432 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
1433
1434 info->loads[0] = avenrun[0] << (SI_LOAD_SHIFT - FSHIFT);
1435 info->loads[1] = avenrun[1] << (SI_LOAD_SHIFT - FSHIFT);
1436 info->loads[2] = avenrun[2] << (SI_LOAD_SHIFT - FSHIFT);
1437
1438 info->procs = nr_threads;
1439 } while (read_seqretry(&xtime_lock, seq));
1440
1441 si_meminfo(info);
1442 si_swapinfo(info);
1443
1444 /*
1445 * If the sum of all the available memory (i.e. ram + swap)
1446 * is less than can be stored in a 32 bit unsigned long then
1447 * we can be binary compatible with 2.2.x kernels. If not,
1448 * well, in that case 2.2.x was broken anyways...
1449 *
1450 * -Erik Andersen <andersee@debian.org>
1451 */
1452
1453 mem_total = info->totalram + info->totalswap;
1454 if (mem_total < info->totalram || mem_total < info->totalswap)
1455 goto out;
1456 bitcount = 0;
1457 mem_unit = info->mem_unit;
1458 while (mem_unit > 1) {
1459 bitcount++;
1460 mem_unit >>= 1;
1461 sav_total = mem_total;
1462 mem_total <<= 1;
1463 if (mem_total < sav_total)
1464 goto out;
1465 }
1466
1467 /*
1468 * If mem_total did not overflow, multiply all memory values by
1469 * info->mem_unit and set it to 1. This leaves things compatible
1470 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1471 * kernels...
1472 */
1473
1474 info->mem_unit = 1;
1475 info->totalram <<= bitcount;
1476 info->freeram <<= bitcount;
1477 info->sharedram <<= bitcount;
1478 info->bufferram <<= bitcount;
1479 info->totalswap <<= bitcount;
1480 info->freeswap <<= bitcount;
1481 info->totalhigh <<= bitcount;
1482 info->freehigh <<= bitcount;
1483
1484 out:
1485 return 0;
1486 }
1487
1488 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
1489 {
1490 struct sysinfo val;
1491
1492 do_sysinfo(&val);
1493
1494 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
1495 return -EFAULT;
1496
1497 return 0;
1498 }
1499
1500 static int __cpuinit init_timers_cpu(int cpu)
1501 {
1502 int j;
1503 struct tvec_base *base;
1504 static char __cpuinitdata tvec_base_done[NR_CPUS];
1505
1506 if (!tvec_base_done[cpu]) {
1507 static char boot_done;
1508
1509 if (boot_done) {
1510 /*
1511 * The APs use this path later in boot
1512 */
1513 base = kmalloc_node(sizeof(*base),
1514 GFP_KERNEL | __GFP_ZERO,
1515 cpu_to_node(cpu));
1516 if (!base)
1517 return -ENOMEM;
1518
1519 /* Make sure that tvec_base is 2 byte aligned */
1520 if (tbase_get_deferrable(base)) {
1521 WARN_ON(1);
1522 kfree(base);
1523 return -ENOMEM;
1524 }
1525 per_cpu(tvec_bases, cpu) = base;
1526 } else {
1527 /*
1528 * This is for the boot CPU - we use compile-time
1529 * static initialisation because per-cpu memory isn't
1530 * ready yet and because the memory allocators are not
1531 * initialised either.
1532 */
1533 boot_done = 1;
1534 base = &boot_tvec_bases;
1535 }
1536 tvec_base_done[cpu] = 1;
1537 } else {
1538 base = per_cpu(tvec_bases, cpu);
1539 }
1540
1541 spin_lock_init(&base->lock);
1542
1543 for (j = 0; j < TVN_SIZE; j++) {
1544 INIT_LIST_HEAD(base->tv5.vec + j);
1545 INIT_LIST_HEAD(base->tv4.vec + j);
1546 INIT_LIST_HEAD(base->tv3.vec + j);
1547 INIT_LIST_HEAD(base->tv2.vec + j);
1548 }
1549 for (j = 0; j < TVR_SIZE; j++)
1550 INIT_LIST_HEAD(base->tv1.vec + j);
1551
1552 base->timer_jiffies = jiffies;
1553 return 0;
1554 }
1555
1556 #ifdef CONFIG_HOTPLUG_CPU
1557 static void migrate_timer_list(struct tvec_base *new_base, struct list_head *head)
1558 {
1559 struct timer_list *timer;
1560
1561 while (!list_empty(head)) {
1562 timer = list_first_entry(head, struct timer_list, entry);
1563 detach_timer(timer, 0);
1564 timer_set_base(timer, new_base);
1565 internal_add_timer(new_base, timer);
1566 }
1567 }
1568
1569 static void __cpuinit migrate_timers(int cpu)
1570 {
1571 struct tvec_base *old_base;
1572 struct tvec_base *new_base;
1573 int i;
1574
1575 BUG_ON(cpu_online(cpu));
1576 old_base = per_cpu(tvec_bases, cpu);
1577 new_base = get_cpu_var(tvec_bases);
1578 /*
1579 * The caller is globally serialized and nobody else
1580 * takes two locks at once, deadlock is not possible.
1581 */
1582 spin_lock_irq(&new_base->lock);
1583 spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1584
1585 BUG_ON(old_base->running_timer);
1586
1587 for (i = 0; i < TVR_SIZE; i++)
1588 migrate_timer_list(new_base, old_base->tv1.vec + i);
1589 for (i = 0; i < TVN_SIZE; i++) {
1590 migrate_timer_list(new_base, old_base->tv2.vec + i);
1591 migrate_timer_list(new_base, old_base->tv3.vec + i);
1592 migrate_timer_list(new_base, old_base->tv4.vec + i);
1593 migrate_timer_list(new_base, old_base->tv5.vec + i);
1594 }
1595
1596 spin_unlock(&old_base->lock);
1597 spin_unlock_irq(&new_base->lock);
1598 put_cpu_var(tvec_bases);
1599 }
1600 #endif /* CONFIG_HOTPLUG_CPU */
1601
1602 static int __cpuinit timer_cpu_notify(struct notifier_block *self,
1603 unsigned long action, void *hcpu)
1604 {
1605 long cpu = (long)hcpu;
1606 switch(action) {
1607 case CPU_UP_PREPARE:
1608 case CPU_UP_PREPARE_FROZEN:
1609 if (init_timers_cpu(cpu) < 0)
1610 return NOTIFY_BAD;
1611 break;
1612 #ifdef CONFIG_HOTPLUG_CPU
1613 case CPU_DEAD:
1614 case CPU_DEAD_FROZEN:
1615 migrate_timers(cpu);
1616 break;
1617 #endif
1618 default:
1619 break;
1620 }
1621 return NOTIFY_OK;
1622 }
1623
1624 static struct notifier_block __cpuinitdata timers_nb = {
1625 .notifier_call = timer_cpu_notify,
1626 };
1627
1628
1629 void __init init_timers(void)
1630 {
1631 int err = timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE,
1632 (void *)(long)smp_processor_id());
1633
1634 init_timer_stats();
1635
1636 BUG_ON(err == NOTIFY_BAD);
1637 register_cpu_notifier(&timers_nb);
1638 open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
1639 }
1640
1641 /**
1642 * msleep - sleep safely even with waitqueue interruptions
1643 * @msecs: Time in milliseconds to sleep for
1644 */
1645 void msleep(unsigned int msecs)
1646 {
1647 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1648
1649 while (timeout)
1650 timeout = schedule_timeout_uninterruptible(timeout);
1651 }
1652
1653 EXPORT_SYMBOL(msleep);
1654
1655 /**
1656 * msleep_interruptible - sleep waiting for signals
1657 * @msecs: Time in milliseconds to sleep for
1658 */
1659 unsigned long msleep_interruptible(unsigned int msecs)
1660 {
1661 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1662
1663 while (timeout && !signal_pending(current))
1664 timeout = schedule_timeout_interruptible(timeout);
1665 return jiffies_to_msecs(timeout);
1666 }
1667
1668 EXPORT_SYMBOL(msleep_interruptible);