blob: a3d25f4150190a7afcd36a0e1bb9c17e7df793cd [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/kernel/timer.c
3 *
john stultz85240702007-05-08 00:27:59 -07004 * Kernel internal timers, basic process system calls
Linus Torvalds1da177e2005-04-16 15:20:36 -07005 *
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>
Pavel Emelyanovb4888932007-10-18 23:40:14 -070029#include <linux/pid_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#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>
Adrian Bunk97a41e22006-01-08 01:02:17 -080037#include <linux/delay.h>
Thomas Gleixner79bf2bb2007-02-16 01:28:03 -080038#include <linux/tick.h>
Ingo Molnar82f67cd2007-02-16 01:28:13 -080039#include <linux/kallsyms.h>
Peter Zijlstra925d5192009-03-30 19:07:02 +020040#include <linux/perf_counter.h>
Arun R Bharadwajeea08f32009-04-16 12:16:41 +053041#include <linux/sched.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
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
Thomas Gleixnerecea8d12005-10-30 15:03:00 -080049u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
50
51EXPORT_SYMBOL(jiffies_64);
52
Linus Torvalds1da177e2005-04-16 15:20:36 -070053/*
54 * per-CPU timer vector definitions:
55 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070056#define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6)
57#define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8)
58#define TVN_SIZE (1 << TVN_BITS)
59#define TVR_SIZE (1 << TVR_BITS)
60#define TVN_MASK (TVN_SIZE - 1)
61#define TVR_MASK (TVR_SIZE - 1)
62
Pavel Macheka6fa8e52008-01-30 13:30:00 +010063struct tvec {
Linus Torvalds1da177e2005-04-16 15:20:36 -070064 struct list_head vec[TVN_SIZE];
Pavel Macheka6fa8e52008-01-30 13:30:00 +010065};
Linus Torvalds1da177e2005-04-16 15:20:36 -070066
Pavel Macheka6fa8e52008-01-30 13:30:00 +010067struct tvec_root {
Linus Torvalds1da177e2005-04-16 15:20:36 -070068 struct list_head vec[TVR_SIZE];
Pavel Macheka6fa8e52008-01-30 13:30:00 +010069};
Linus Torvalds1da177e2005-04-16 15:20:36 -070070
Pavel Macheka6fa8e52008-01-30 13:30:00 +010071struct tvec_base {
Oleg Nesterov3691c512006-03-31 02:30:30 -080072 spinlock_t lock;
73 struct timer_list *running_timer;
Linus Torvalds1da177e2005-04-16 15:20:36 -070074 unsigned long timer_jiffies;
Pavel Macheka6fa8e52008-01-30 13:30:00 +010075 struct tvec_root tv1;
76 struct tvec tv2;
77 struct tvec tv3;
78 struct tvec tv4;
79 struct tvec tv5;
Venki Pallipadi6e453a62007-05-08 00:27:44 -070080} ____cacheline_aligned;
Linus Torvalds1da177e2005-04-16 15:20:36 -070081
Pavel Macheka6fa8e52008-01-30 13:30:00 +010082struct tvec_base boot_tvec_bases;
Oleg Nesterov3691c512006-03-31 02:30:30 -080083EXPORT_SYMBOL(boot_tvec_bases);
Pavel Macheka6fa8e52008-01-30 13:30:00 +010084static DEFINE_PER_CPU(struct tvec_base *, tvec_bases) = &boot_tvec_bases;
Linus Torvalds1da177e2005-04-16 15:20:36 -070085
Venki Pallipadi6e453a62007-05-08 00:27:44 -070086/*
Pavel Macheka6fa8e52008-01-30 13:30:00 +010087 * Note that all tvec_bases are 2 byte aligned and lower bit of
Venki Pallipadi6e453a62007-05-08 00:27:44 -070088 * base in timer_list is guaranteed to be zero. Use the LSB for
89 * the new flag to indicate whether the timer is deferrable
90 */
91#define TBASE_DEFERRABLE_FLAG (0x1)
92
93/* Functions below help us manage 'deferrable' flag */
Pavel Macheka6fa8e52008-01-30 13:30:00 +010094static inline unsigned int tbase_get_deferrable(struct tvec_base *base)
Venki Pallipadi6e453a62007-05-08 00:27:44 -070095{
akpm@linux-foundation.orge9910842007-05-10 03:16:01 -070096 return ((unsigned int)(unsigned long)base & TBASE_DEFERRABLE_FLAG);
Venki Pallipadi6e453a62007-05-08 00:27:44 -070097}
98
Pavel Macheka6fa8e52008-01-30 13:30:00 +010099static inline struct tvec_base *tbase_get_base(struct tvec_base *base)
Venki Pallipadi6e453a62007-05-08 00:27:44 -0700100{
Pavel Macheka6fa8e52008-01-30 13:30:00 +0100101 return ((struct tvec_base *)((unsigned long)base & ~TBASE_DEFERRABLE_FLAG));
Venki Pallipadi6e453a62007-05-08 00:27:44 -0700102}
103
104static inline void timer_set_deferrable(struct timer_list *timer)
105{
Pavel Macheka6fa8e52008-01-30 13:30:00 +0100106 timer->base = ((struct tvec_base *)((unsigned long)(timer->base) |
Thomas Gleixner68194572007-07-19 01:49:16 -0700107 TBASE_DEFERRABLE_FLAG));
Venki Pallipadi6e453a62007-05-08 00:27:44 -0700108}
109
110static inline void
Pavel Macheka6fa8e52008-01-30 13:30:00 +0100111timer_set_base(struct timer_list *timer, struct tvec_base *new_base)
Venki Pallipadi6e453a62007-05-08 00:27:44 -0700112{
Pavel Macheka6fa8e52008-01-30 13:30:00 +0100113 timer->base = (struct tvec_base *)((unsigned long)(new_base) |
Thomas Gleixner68194572007-07-19 01:49:16 -0700114 tbase_get_deferrable(timer->base));
Venki Pallipadi6e453a62007-05-08 00:27:44 -0700115}
116
Alan Stern9c133c42008-11-06 08:42:48 +0100117static unsigned long round_jiffies_common(unsigned long j, int cpu,
118 bool force_up)
119{
120 int rem;
121 unsigned long original = j;
122
123 /*
124 * We don't want all cpus firing their timers at once hitting the
125 * same lock or cachelines, so we skew each extra cpu with an extra
126 * 3 jiffies. This 3 jiffies came originally from the mm/ code which
127 * already did this.
128 * The skew is done by adding 3*cpunr, then round, then subtract this
129 * extra offset again.
130 */
131 j += cpu * 3;
132
133 rem = j % HZ;
134
135 /*
136 * If the target jiffie is just after a whole second (which can happen
137 * due to delays of the timer irq, long irq off times etc etc) then
138 * we should round down to the whole second, not up. Use 1/4th second
139 * as cutoff for this rounding as an extreme upper bound for this.
140 * But never round down if @force_up is set.
141 */
142 if (rem < HZ/4 && !force_up) /* round down */
143 j = j - rem;
144 else /* round up */
145 j = j - rem + HZ;
146
147 /* now that we have rounded, subtract the extra skew again */
148 j -= cpu * 3;
149
150 if (j <= jiffies) /* rounding ate our timeout entirely; */
151 return original;
152 return j;
153}
154
Arjan van de Ven4c36a5d2006-12-10 02:21:24 -0800155/**
156 * __round_jiffies - function to round jiffies to a full second
157 * @j: the time in (absolute) jiffies that should be rounded
158 * @cpu: the processor number on which the timeout will happen
159 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800160 * __round_jiffies() rounds an absolute time in the future (in jiffies)
Arjan van de Ven4c36a5d2006-12-10 02:21:24 -0800161 * up or down to (approximately) full seconds. This is useful for timers
162 * for which the exact time they fire does not matter too much, as long as
163 * they fire approximately every X seconds.
164 *
165 * By rounding these timers to whole seconds, all such timers will fire
166 * at the same time, rather than at various times spread out. The goal
167 * of this is to have the CPU wake up less, which saves power.
168 *
169 * The exact rounding is skewed for each processor to avoid all
170 * processors firing at the exact same time, which could lead
171 * to lock contention or spurious cache line bouncing.
172 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800173 * The return value is the rounded version of the @j parameter.
Arjan van de Ven4c36a5d2006-12-10 02:21:24 -0800174 */
175unsigned long __round_jiffies(unsigned long j, int cpu)
176{
Alan Stern9c133c42008-11-06 08:42:48 +0100177 return round_jiffies_common(j, cpu, false);
Arjan van de Ven4c36a5d2006-12-10 02:21:24 -0800178}
179EXPORT_SYMBOL_GPL(__round_jiffies);
180
181/**
182 * __round_jiffies_relative - function to round jiffies to a full second
183 * @j: the time in (relative) jiffies that should be rounded
184 * @cpu: the processor number on which the timeout will happen
185 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800186 * __round_jiffies_relative() rounds a time delta in the future (in jiffies)
Arjan van de Ven4c36a5d2006-12-10 02:21:24 -0800187 * up or down to (approximately) full seconds. This is useful for timers
188 * for which the exact time they fire does not matter too much, as long as
189 * they fire approximately every X seconds.
190 *
191 * By rounding these timers to whole seconds, all such timers will fire
192 * at the same time, rather than at various times spread out. The goal
193 * of this is to have the CPU wake up less, which saves power.
194 *
195 * The exact rounding is skewed for each processor to avoid all
196 * processors firing at the exact same time, which could lead
197 * to lock contention or spurious cache line bouncing.
198 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800199 * The return value is the rounded version of the @j parameter.
Arjan van de Ven4c36a5d2006-12-10 02:21:24 -0800200 */
201unsigned long __round_jiffies_relative(unsigned long j, int cpu)
202{
Alan Stern9c133c42008-11-06 08:42:48 +0100203 unsigned long j0 = jiffies;
204
205 /* Use j0 because jiffies might change while we run */
206 return round_jiffies_common(j + j0, cpu, false) - j0;
Arjan van de Ven4c36a5d2006-12-10 02:21:24 -0800207}
208EXPORT_SYMBOL_GPL(__round_jiffies_relative);
209
210/**
211 * round_jiffies - function to round jiffies to a full second
212 * @j: the time in (absolute) jiffies that should be rounded
213 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800214 * round_jiffies() rounds an absolute time in the future (in jiffies)
Arjan van de Ven4c36a5d2006-12-10 02:21:24 -0800215 * up or down to (approximately) full seconds. This is useful for timers
216 * for which the exact time they fire does not matter too much, as long as
217 * they fire approximately every X seconds.
218 *
219 * By rounding these timers to whole seconds, all such timers will fire
220 * at the same time, rather than at various times spread out. The goal
221 * of this is to have the CPU wake up less, which saves power.
222 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800223 * The return value is the rounded version of the @j parameter.
Arjan van de Ven4c36a5d2006-12-10 02:21:24 -0800224 */
225unsigned long round_jiffies(unsigned long j)
226{
Alan Stern9c133c42008-11-06 08:42:48 +0100227 return round_jiffies_common(j, raw_smp_processor_id(), false);
Arjan van de Ven4c36a5d2006-12-10 02:21:24 -0800228}
229EXPORT_SYMBOL_GPL(round_jiffies);
230
231/**
232 * round_jiffies_relative - function to round jiffies to a full second
233 * @j: the time in (relative) jiffies that should be rounded
234 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800235 * round_jiffies_relative() rounds a time delta in the future (in jiffies)
Arjan van de Ven4c36a5d2006-12-10 02:21:24 -0800236 * up or down to (approximately) full seconds. This is useful for timers
237 * for which the exact time they fire does not matter too much, as long as
238 * they fire approximately every X seconds.
239 *
240 * By rounding these timers to whole seconds, all such timers will fire
241 * at the same time, rather than at various times spread out. The goal
242 * of this is to have the CPU wake up less, which saves power.
243 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800244 * The return value is the rounded version of the @j parameter.
Arjan van de Ven4c36a5d2006-12-10 02:21:24 -0800245 */
246unsigned long round_jiffies_relative(unsigned long j)
247{
248 return __round_jiffies_relative(j, raw_smp_processor_id());
249}
250EXPORT_SYMBOL_GPL(round_jiffies_relative);
251
Alan Stern9c133c42008-11-06 08:42:48 +0100252/**
253 * __round_jiffies_up - function to round jiffies up to a full second
254 * @j: the time in (absolute) jiffies that should be rounded
255 * @cpu: the processor number on which the timeout will happen
256 *
257 * This is the same as __round_jiffies() except that it will never
258 * round down. This is useful for timeouts for which the exact time
259 * of firing does not matter too much, as long as they don't fire too
260 * early.
261 */
262unsigned long __round_jiffies_up(unsigned long j, int cpu)
263{
264 return round_jiffies_common(j, cpu, true);
265}
266EXPORT_SYMBOL_GPL(__round_jiffies_up);
267
268/**
269 * __round_jiffies_up_relative - function to round jiffies up to a full second
270 * @j: the time in (relative) jiffies that should be rounded
271 * @cpu: the processor number on which the timeout will happen
272 *
273 * This is the same as __round_jiffies_relative() except that it will never
274 * round down. This is useful for timeouts for which the exact time
275 * of firing does not matter too much, as long as they don't fire too
276 * early.
277 */
278unsigned long __round_jiffies_up_relative(unsigned long j, int cpu)
279{
280 unsigned long j0 = jiffies;
281
282 /* Use j0 because jiffies might change while we run */
283 return round_jiffies_common(j + j0, cpu, true) - j0;
284}
285EXPORT_SYMBOL_GPL(__round_jiffies_up_relative);
286
287/**
288 * round_jiffies_up - function to round jiffies up to a full second
289 * @j: the time in (absolute) jiffies that should be rounded
290 *
291 * This is the same as round_jiffies() except that it will never
292 * round down. This is useful for timeouts for which the exact time
293 * of firing does not matter too much, as long as they don't fire too
294 * early.
295 */
296unsigned long round_jiffies_up(unsigned long j)
297{
298 return round_jiffies_common(j, raw_smp_processor_id(), true);
299}
300EXPORT_SYMBOL_GPL(round_jiffies_up);
301
302/**
303 * round_jiffies_up_relative - function to round jiffies up to a full second
304 * @j: the time in (relative) jiffies that should be rounded
305 *
306 * This is the same as round_jiffies_relative() except that it will never
307 * round down. This is useful for timeouts for which the exact time
308 * of firing does not matter too much, as long as they don't fire too
309 * early.
310 */
311unsigned long round_jiffies_up_relative(unsigned long j)
312{
313 return __round_jiffies_up_relative(j, raw_smp_processor_id());
314}
315EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
316
Arjan van de Ven4c36a5d2006-12-10 02:21:24 -0800317
Pavel Macheka6fa8e52008-01-30 13:30:00 +0100318static inline void set_running_timer(struct tvec_base *base,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319 struct timer_list *timer)
320{
321#ifdef CONFIG_SMP
Oleg Nesterov3691c512006-03-31 02:30:30 -0800322 base->running_timer = timer;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323#endif
324}
325
Pavel Macheka6fa8e52008-01-30 13:30:00 +0100326static void internal_add_timer(struct tvec_base *base, struct timer_list *timer)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327{
328 unsigned long expires = timer->expires;
329 unsigned long idx = expires - base->timer_jiffies;
330 struct list_head *vec;
331
332 if (idx < TVR_SIZE) {
333 int i = expires & TVR_MASK;
334 vec = base->tv1.vec + i;
335 } else if (idx < 1 << (TVR_BITS + TVN_BITS)) {
336 int i = (expires >> TVR_BITS) & TVN_MASK;
337 vec = base->tv2.vec + i;
338 } else if (idx < 1 << (TVR_BITS + 2 * TVN_BITS)) {
339 int i = (expires >> (TVR_BITS + TVN_BITS)) & TVN_MASK;
340 vec = base->tv3.vec + i;
341 } else if (idx < 1 << (TVR_BITS + 3 * TVN_BITS)) {
342 int i = (expires >> (TVR_BITS + 2 * TVN_BITS)) & TVN_MASK;
343 vec = base->tv4.vec + i;
344 } else if ((signed long) idx < 0) {
345 /*
346 * Can happen if you add a timer with expires == jiffies,
347 * or you set a timer to go off in the past
348 */
349 vec = base->tv1.vec + (base->timer_jiffies & TVR_MASK);
350 } else {
351 int i;
352 /* If the timeout is larger than 0xffffffff on 64-bit
353 * architectures then we use the maximum timeout:
354 */
355 if (idx > 0xffffffffUL) {
356 idx = 0xffffffffUL;
357 expires = idx + base->timer_jiffies;
358 }
359 i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK;
360 vec = base->tv5.vec + i;
361 }
362 /*
363 * Timers are FIFO:
364 */
365 list_add_tail(&timer->entry, vec);
366}
367
Ingo Molnar82f67cd2007-02-16 01:28:13 -0800368#ifdef CONFIG_TIMER_STATS
369void __timer_stats_timer_set_start_info(struct timer_list *timer, void *addr)
370{
371 if (timer->start_site)
372 return;
373
374 timer->start_site = addr;
375 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
376 timer->start_pid = current->pid;
377}
Venki Pallipadic5c061b82007-07-15 23:40:30 -0700378
379static void timer_stats_account_timer(struct timer_list *timer)
380{
381 unsigned int flag = 0;
382
Heiko Carstens507e1232009-06-23 17:38:15 +0200383 if (likely(!timer->start_site))
384 return;
Venki Pallipadic5c061b82007-07-15 23:40:30 -0700385 if (unlikely(tbase_get_deferrable(timer->base)))
386 flag |= TIMER_STATS_FLAG_DEFERRABLE;
387
388 timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
389 timer->function, timer->start_comm, flag);
390}
391
392#else
393static void timer_stats_account_timer(struct timer_list *timer) {}
Ingo Molnar82f67cd2007-02-16 01:28:13 -0800394#endif
395
Thomas Gleixnerc6f3a972008-04-30 00:55:03 -0700396#ifdef CONFIG_DEBUG_OBJECTS_TIMERS
397
398static struct debug_obj_descr timer_debug_descr;
399
400/*
401 * fixup_init is called when:
402 * - an active object is initialized
403 */
404static int timer_fixup_init(void *addr, enum debug_obj_state state)
405{
406 struct timer_list *timer = addr;
407
408 switch (state) {
409 case ODEBUG_STATE_ACTIVE:
410 del_timer_sync(timer);
411 debug_object_init(timer, &timer_debug_descr);
412 return 1;
413 default:
414 return 0;
415 }
416}
417
418/*
419 * fixup_activate is called when:
420 * - an active object is activated
421 * - an unknown object is activated (might be a statically initialized object)
422 */
423static int timer_fixup_activate(void *addr, enum debug_obj_state state)
424{
425 struct timer_list *timer = addr;
426
427 switch (state) {
428
429 case ODEBUG_STATE_NOTAVAILABLE:
430 /*
431 * This is not really a fixup. The timer was
432 * statically initialized. We just make sure that it
433 * is tracked in the object tracker.
434 */
435 if (timer->entry.next == NULL &&
436 timer->entry.prev == TIMER_ENTRY_STATIC) {
437 debug_object_init(timer, &timer_debug_descr);
438 debug_object_activate(timer, &timer_debug_descr);
439 return 0;
440 } else {
441 WARN_ON_ONCE(1);
442 }
443 return 0;
444
445 case ODEBUG_STATE_ACTIVE:
446 WARN_ON(1);
447
448 default:
449 return 0;
450 }
451}
452
453/*
454 * fixup_free is called when:
455 * - an active object is freed
456 */
457static int timer_fixup_free(void *addr, enum debug_obj_state state)
458{
459 struct timer_list *timer = addr;
460
461 switch (state) {
462 case ODEBUG_STATE_ACTIVE:
463 del_timer_sync(timer);
464 debug_object_free(timer, &timer_debug_descr);
465 return 1;
466 default:
467 return 0;
468 }
469}
470
471static struct debug_obj_descr timer_debug_descr = {
472 .name = "timer_list",
473 .fixup_init = timer_fixup_init,
474 .fixup_activate = timer_fixup_activate,
475 .fixup_free = timer_fixup_free,
476};
477
478static inline void debug_timer_init(struct timer_list *timer)
479{
480 debug_object_init(timer, &timer_debug_descr);
481}
482
483static inline void debug_timer_activate(struct timer_list *timer)
484{
485 debug_object_activate(timer, &timer_debug_descr);
486}
487
488static inline void debug_timer_deactivate(struct timer_list *timer)
489{
490 debug_object_deactivate(timer, &timer_debug_descr);
491}
492
493static inline void debug_timer_free(struct timer_list *timer)
494{
495 debug_object_free(timer, &timer_debug_descr);
496}
497
Johannes Berg6f2b9b92009-01-29 16:03:20 +0100498static void __init_timer(struct timer_list *timer,
499 const char *name,
500 struct lock_class_key *key);
Thomas Gleixnerc6f3a972008-04-30 00:55:03 -0700501
Johannes Berg6f2b9b92009-01-29 16:03:20 +0100502void init_timer_on_stack_key(struct timer_list *timer,
503 const char *name,
504 struct lock_class_key *key)
Thomas Gleixnerc6f3a972008-04-30 00:55:03 -0700505{
506 debug_object_init_on_stack(timer, &timer_debug_descr);
Johannes Berg6f2b9b92009-01-29 16:03:20 +0100507 __init_timer(timer, name, key);
Thomas Gleixnerc6f3a972008-04-30 00:55:03 -0700508}
Johannes Berg6f2b9b92009-01-29 16:03:20 +0100509EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
Thomas Gleixnerc6f3a972008-04-30 00:55:03 -0700510
511void destroy_timer_on_stack(struct timer_list *timer)
512{
513 debug_object_free(timer, &timer_debug_descr);
514}
515EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
516
517#else
518static inline void debug_timer_init(struct timer_list *timer) { }
519static inline void debug_timer_activate(struct timer_list *timer) { }
520static inline void debug_timer_deactivate(struct timer_list *timer) { }
521#endif
522
Johannes Berg6f2b9b92009-01-29 16:03:20 +0100523static void __init_timer(struct timer_list *timer,
524 const char *name,
525 struct lock_class_key *key)
Thomas Gleixnerc6f3a972008-04-30 00:55:03 -0700526{
527 timer->entry.next = NULL;
528 timer->base = __raw_get_cpu_var(tvec_bases);
529#ifdef CONFIG_TIMER_STATS
530 timer->start_site = NULL;
531 timer->start_pid = -1;
532 memset(timer->start_comm, 0, TASK_COMM_LEN);
533#endif
Johannes Berg6f2b9b92009-01-29 16:03:20 +0100534 lockdep_init_map(&timer->lockdep_map, name, key, 0);
Thomas Gleixnerc6f3a972008-04-30 00:55:03 -0700535}
536
Rolf Eike Beer2aae4a12006-09-29 01:59:46 -0700537/**
Randy Dunlap633fe792009-04-01 17:47:23 -0700538 * init_timer_key - initialize a timer
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700539 * @timer: the timer to be initialized
Randy Dunlap633fe792009-04-01 17:47:23 -0700540 * @name: name of the timer
541 * @key: lockdep class key of the fake lock used for tracking timer
542 * sync lock dependencies
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700543 *
Randy Dunlap633fe792009-04-01 17:47:23 -0700544 * init_timer_key() must be done to a timer prior calling *any* of the
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700545 * other timer functions.
546 */
Johannes Berg6f2b9b92009-01-29 16:03:20 +0100547void init_timer_key(struct timer_list *timer,
548 const char *name,
549 struct lock_class_key *key)
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700550{
Thomas Gleixnerc6f3a972008-04-30 00:55:03 -0700551 debug_timer_init(timer);
Johannes Berg6f2b9b92009-01-29 16:03:20 +0100552 __init_timer(timer, name, key);
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700553}
Johannes Berg6f2b9b92009-01-29 16:03:20 +0100554EXPORT_SYMBOL(init_timer_key);
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700555
Johannes Berg6f2b9b92009-01-29 16:03:20 +0100556void init_timer_deferrable_key(struct timer_list *timer,
557 const char *name,
558 struct lock_class_key *key)
Venki Pallipadi6e453a62007-05-08 00:27:44 -0700559{
Johannes Berg6f2b9b92009-01-29 16:03:20 +0100560 init_timer_key(timer, name, key);
Venki Pallipadi6e453a62007-05-08 00:27:44 -0700561 timer_set_deferrable(timer);
562}
Johannes Berg6f2b9b92009-01-29 16:03:20 +0100563EXPORT_SYMBOL(init_timer_deferrable_key);
Venki Pallipadi6e453a62007-05-08 00:27:44 -0700564
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700565static inline void detach_timer(struct timer_list *timer,
Ingo Molnar82f67cd2007-02-16 01:28:13 -0800566 int clear_pending)
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700567{
568 struct list_head *entry = &timer->entry;
569
Thomas Gleixnerc6f3a972008-04-30 00:55:03 -0700570 debug_timer_deactivate(timer);
571
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700572 __list_del(entry->prev, entry->next);
573 if (clear_pending)
574 entry->next = NULL;
575 entry->prev = LIST_POISON2;
576}
577
578/*
Oleg Nesterov3691c512006-03-31 02:30:30 -0800579 * We are using hashed locking: holding per_cpu(tvec_bases).lock
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700580 * means that all timers which are tied to this base via timer->base are
581 * locked, and the base itself is locked too.
582 *
583 * So __run_timers/migrate_timers can safely modify all timers which could
584 * be found on ->tvX lists.
585 *
586 * When the timer's base is locked, and the timer removed from list, it is
587 * possible to set timer->base = NULL and drop the lock: the timer remains
588 * locked.
589 */
Pavel Macheka6fa8e52008-01-30 13:30:00 +0100590static struct tvec_base *lock_timer_base(struct timer_list *timer,
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700591 unsigned long *flags)
Josh Triplett89e7e3742006-09-29 01:59:36 -0700592 __acquires(timer->base->lock)
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700593{
Pavel Macheka6fa8e52008-01-30 13:30:00 +0100594 struct tvec_base *base;
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700595
596 for (;;) {
Pavel Macheka6fa8e52008-01-30 13:30:00 +0100597 struct tvec_base *prelock_base = timer->base;
Venki Pallipadi6e453a62007-05-08 00:27:44 -0700598 base = tbase_get_base(prelock_base);
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700599 if (likely(base != NULL)) {
600 spin_lock_irqsave(&base->lock, *flags);
Venki Pallipadi6e453a62007-05-08 00:27:44 -0700601 if (likely(prelock_base == timer->base))
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700602 return base;
603 /* The timer has migrated to another CPU */
604 spin_unlock_irqrestore(&base->lock, *flags);
605 }
606 cpu_relax();
607 }
608}
609
Ingo Molnar74019222009-02-18 12:23:29 +0100610static inline int
Arun R Bharadwaj597d0272009-04-16 12:13:26 +0530611__mod_timer(struct timer_list *timer, unsigned long expires,
612 bool pending_only, int pinned)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613{
Pavel Macheka6fa8e52008-01-30 13:30:00 +0100614 struct tvec_base *base, *new_base;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 unsigned long flags;
Arun R Bharadwajeea08f32009-04-16 12:16:41 +0530616 int ret = 0 , cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617
Ingo Molnar82f67cd2007-02-16 01:28:13 -0800618 timer_stats_timer_set_start_info(timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 BUG_ON(!timer->function);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700621 base = lock_timer_base(timer, &flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700623 if (timer_pending(timer)) {
624 detach_timer(timer, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 ret = 1;
Ingo Molnar74019222009-02-18 12:23:29 +0100626 } else {
627 if (pending_only)
628 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 }
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700630
Thomas Gleixnerc6f3a972008-04-30 00:55:03 -0700631 debug_timer_activate(timer);
632
Jan Beulicha4a61982006-03-24 03:15:54 -0800633 new_base = __get_cpu_var(tvec_bases);
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700634
Arun R Bharadwajeea08f32009-04-16 12:16:41 +0530635 cpu = smp_processor_id();
636
637#if defined(CONFIG_NO_HZ) && defined(CONFIG_SMP)
638 if (!pinned && get_sysctl_timer_migration() && idle_cpu(cpu)) {
639 int preferred_cpu = get_nohz_load_balancer();
640
641 if (preferred_cpu >= 0)
642 cpu = preferred_cpu;
643 }
644#endif
645 new_base = per_cpu(tvec_bases, cpu);
646
Oleg Nesterov3691c512006-03-31 02:30:30 -0800647 if (base != new_base) {
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700648 /*
649 * We are trying to schedule the timer on the local CPU.
650 * However we can't change timer's base while it is running,
651 * otherwise del_timer_sync() can't detect that the timer's
652 * handler yet has not finished. This also guarantees that
653 * the timer is serialized wrt itself.
654 */
Oleg Nesterova2c348f2006-03-31 02:30:31 -0800655 if (likely(base->running_timer != timer)) {
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700656 /* See the comment in lock_timer_base() */
Venki Pallipadi6e453a62007-05-08 00:27:44 -0700657 timer_set_base(timer, NULL);
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700658 spin_unlock(&base->lock);
Oleg Nesterova2c348f2006-03-31 02:30:31 -0800659 base = new_base;
660 spin_lock(&base->lock);
Venki Pallipadi6e453a62007-05-08 00:27:44 -0700661 timer_set_base(timer, base);
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700662 }
663 }
664
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 timer->expires = expires;
Oleg Nesterova2c348f2006-03-31 02:30:31 -0800666 internal_add_timer(base, timer);
Ingo Molnar74019222009-02-18 12:23:29 +0100667
668out_unlock:
Oleg Nesterova2c348f2006-03-31 02:30:31 -0800669 spin_unlock_irqrestore(&base->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670
671 return ret;
672}
673
Ingo Molnar74019222009-02-18 12:23:29 +0100674/**
675 * mod_timer_pending - modify a pending timer's timeout
676 * @timer: the pending timer to be modified
677 * @expires: new timeout in jiffies
678 *
679 * mod_timer_pending() is the same for pending timers as mod_timer(),
680 * but will not re-activate and modify already deleted timers.
681 *
682 * It is useful for unserialized use of timers.
683 */
684int mod_timer_pending(struct timer_list *timer, unsigned long expires)
685{
Arun R Bharadwaj597d0272009-04-16 12:13:26 +0530686 return __mod_timer(timer, expires, true, TIMER_NOT_PINNED);
Ingo Molnar74019222009-02-18 12:23:29 +0100687}
688EXPORT_SYMBOL(mod_timer_pending);
689
690/**
691 * mod_timer - modify a timer's timeout
692 * @timer: the timer to be modified
693 * @expires: new timeout in jiffies
694 *
695 * mod_timer() is a more efficient way to update the expire field of an
696 * active timer (if the timer is inactive it will be activated)
697 *
698 * mod_timer(timer, expires) is equivalent to:
699 *
700 * del_timer(timer); timer->expires = expires; add_timer(timer);
701 *
702 * Note that if there are multiple unserialized concurrent users of the
703 * same timer, then mod_timer() is the only safe way to modify the timeout,
704 * since add_timer() cannot modify an already running timer.
705 *
706 * The function returns whether it has modified a pending timer or not.
707 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
708 * active timer returns 1.)
709 */
710int mod_timer(struct timer_list *timer, unsigned long expires)
711{
712 /*
713 * This is a common optimization triggered by the
714 * networking code - if the timer is re-modified
715 * to be the same thing then just return:
716 */
Pavel Roskin48411582009-07-18 16:46:02 -0400717 if (timer_pending(timer) && timer->expires == expires)
Ingo Molnar74019222009-02-18 12:23:29 +0100718 return 1;
719
Arun R Bharadwaj597d0272009-04-16 12:13:26 +0530720 return __mod_timer(timer, expires, false, TIMER_NOT_PINNED);
Ingo Molnar74019222009-02-18 12:23:29 +0100721}
722EXPORT_SYMBOL(mod_timer);
723
724/**
Arun R Bharadwaj597d0272009-04-16 12:13:26 +0530725 * mod_timer_pinned - modify a timer's timeout
726 * @timer: the timer to be modified
727 * @expires: new timeout in jiffies
728 *
729 * mod_timer_pinned() is a way to update the expire field of an
730 * active timer (if the timer is inactive it will be activated)
731 * and not allow the timer to be migrated to a different CPU.
732 *
733 * mod_timer_pinned(timer, expires) is equivalent to:
734 *
735 * del_timer(timer); timer->expires = expires; add_timer(timer);
736 */
737int mod_timer_pinned(struct timer_list *timer, unsigned long expires)
738{
739 if (timer->expires == expires && timer_pending(timer))
740 return 1;
741
742 return __mod_timer(timer, expires, false, TIMER_PINNED);
743}
744EXPORT_SYMBOL(mod_timer_pinned);
745
746/**
Ingo Molnar74019222009-02-18 12:23:29 +0100747 * add_timer - start a timer
748 * @timer: the timer to be added
749 *
750 * The kernel will do a ->function(->data) callback from the
751 * timer interrupt at the ->expires point in the future. The
752 * current time is 'jiffies'.
753 *
754 * The timer's ->expires, ->function (and if the handler uses it, ->data)
755 * fields must be set prior calling this function.
756 *
757 * Timers with an ->expires field in the past will be executed in the next
758 * timer tick.
759 */
760void add_timer(struct timer_list *timer)
761{
762 BUG_ON(timer_pending(timer));
763 mod_timer(timer, timer->expires);
764}
765EXPORT_SYMBOL(add_timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766
Rolf Eike Beer2aae4a12006-09-29 01:59:46 -0700767/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 * add_timer_on - start a timer on a particular CPU
769 * @timer: the timer to be added
770 * @cpu: the CPU to start it on
771 *
772 * This is not very scalable on SMP. Double adds are not possible.
773 */
774void add_timer_on(struct timer_list *timer, int cpu)
775{
Pavel Macheka6fa8e52008-01-30 13:30:00 +0100776 struct tvec_base *base = per_cpu(tvec_bases, cpu);
Thomas Gleixner68194572007-07-19 01:49:16 -0700777 unsigned long flags;
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700778
Ingo Molnar82f67cd2007-02-16 01:28:13 -0800779 timer_stats_timer_set_start_info(timer);
Thomas Gleixner68194572007-07-19 01:49:16 -0700780 BUG_ON(timer_pending(timer) || !timer->function);
Oleg Nesterov3691c512006-03-31 02:30:30 -0800781 spin_lock_irqsave(&base->lock, flags);
Venki Pallipadi6e453a62007-05-08 00:27:44 -0700782 timer_set_base(timer, base);
Thomas Gleixnerc6f3a972008-04-30 00:55:03 -0700783 debug_timer_activate(timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 internal_add_timer(base, timer);
Thomas Gleixner06d83082008-03-22 09:20:24 +0100785 /*
786 * Check whether the other CPU is idle and needs to be
787 * triggered to reevaluate the timer wheel when nohz is
788 * active. We are protected against the other CPU fiddling
789 * with the timer by holding the timer base lock. This also
790 * makes sure that a CPU on the way to idle can not evaluate
791 * the timer wheel.
792 */
793 wake_up_idle_cpu(cpu);
Oleg Nesterov3691c512006-03-31 02:30:30 -0800794 spin_unlock_irqrestore(&base->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795}
Andi Kleena9862e02009-05-19 22:49:07 +0200796EXPORT_SYMBOL_GPL(add_timer_on);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797
Rolf Eike Beer2aae4a12006-09-29 01:59:46 -0700798/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 * del_timer - deactive a timer.
800 * @timer: the timer to be deactivated
801 *
802 * del_timer() deactivates a timer - this works on both active and inactive
803 * timers.
804 *
805 * The function returns whether it has deactivated a pending timer or not.
806 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
807 * active timer returns 1.)
808 */
809int del_timer(struct timer_list *timer)
810{
Pavel Macheka6fa8e52008-01-30 13:30:00 +0100811 struct tvec_base *base;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 unsigned long flags;
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700813 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814
Ingo Molnar82f67cd2007-02-16 01:28:13 -0800815 timer_stats_timer_clear_start_info(timer);
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700816 if (timer_pending(timer)) {
817 base = lock_timer_base(timer, &flags);
818 if (timer_pending(timer)) {
819 detach_timer(timer, 1);
820 ret = 1;
821 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 spin_unlock_irqrestore(&base->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700825 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827EXPORT_SYMBOL(del_timer);
828
829#ifdef CONFIG_SMP
Rolf Eike Beer2aae4a12006-09-29 01:59:46 -0700830/**
831 * try_to_del_timer_sync - Try to deactivate a timer
832 * @timer: timer do del
833 *
Oleg Nesterovfd450b72005-06-23 00:08:59 -0700834 * This function tries to deactivate a timer. Upon successful (ret >= 0)
835 * exit the timer is not queued and the handler is not running on any CPU.
836 *
837 * It must not be called from interrupt contexts.
838 */
839int try_to_del_timer_sync(struct timer_list *timer)
840{
Pavel Macheka6fa8e52008-01-30 13:30:00 +0100841 struct tvec_base *base;
Oleg Nesterovfd450b72005-06-23 00:08:59 -0700842 unsigned long flags;
843 int ret = -1;
844
845 base = lock_timer_base(timer, &flags);
846
847 if (base->running_timer == timer)
848 goto out;
849
850 ret = 0;
851 if (timer_pending(timer)) {
852 detach_timer(timer, 1);
853 ret = 1;
854 }
855out:
856 spin_unlock_irqrestore(&base->lock, flags);
857
858 return ret;
859}
David Howellse19dff12007-04-26 15:46:56 -0700860EXPORT_SYMBOL(try_to_del_timer_sync);
861
Rolf Eike Beer2aae4a12006-09-29 01:59:46 -0700862/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 * del_timer_sync - deactivate a timer and wait for the handler to finish.
864 * @timer: the timer to be deactivated
865 *
866 * This function only differs from del_timer() on SMP: besides deactivating
867 * the timer it also makes sure the handler has finished executing on other
868 * CPUs.
869 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800870 * Synchronization rules: Callers must prevent restarting of the timer,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 * otherwise this function is meaningless. It must not be called from
872 * interrupt contexts. The caller must not hold locks which would prevent
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700873 * completion of the timer's handler. The timer's handler must not call
874 * add_timer_on(). Upon exit the timer is not queued and the handler is
875 * not running on any CPU.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 *
877 * The function returns whether it has deactivated a pending timer or not.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 */
879int del_timer_sync(struct timer_list *timer)
880{
Johannes Berg6f2b9b92009-01-29 16:03:20 +0100881#ifdef CONFIG_LOCKDEP
882 unsigned long flags;
883
884 local_irq_save(flags);
885 lock_map_acquire(&timer->lockdep_map);
886 lock_map_release(&timer->lockdep_map);
887 local_irq_restore(flags);
888#endif
889
Oleg Nesterovfd450b72005-06-23 00:08:59 -0700890 for (;;) {
891 int ret = try_to_del_timer_sync(timer);
892 if (ret >= 0)
893 return ret;
Andrew Mortona0009652006-07-14 00:24:06 -0700894 cpu_relax();
Oleg Nesterovfd450b72005-06-23 00:08:59 -0700895 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896}
897EXPORT_SYMBOL(del_timer_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898#endif
899
Pavel Macheka6fa8e52008-01-30 13:30:00 +0100900static int cascade(struct tvec_base *base, struct tvec *tv, int index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901{
902 /* cascade all the timers from tv up one level */
Porpoise3439dd82006-06-23 02:05:56 -0700903 struct timer_list *timer, *tmp;
904 struct list_head tv_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905
Porpoise3439dd82006-06-23 02:05:56 -0700906 list_replace_init(tv->vec + index, &tv_list);
907
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 /*
Porpoise3439dd82006-06-23 02:05:56 -0700909 * We are removing _all_ timers from the list, so we
910 * don't have to detach them individually.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 */
Porpoise3439dd82006-06-23 02:05:56 -0700912 list_for_each_entry_safe(timer, tmp, &tv_list, entry) {
Venki Pallipadi6e453a62007-05-08 00:27:44 -0700913 BUG_ON(tbase_get_base(timer->base) != base);
Porpoise3439dd82006-06-23 02:05:56 -0700914 internal_add_timer(base, timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916
917 return index;
918}
919
Rolf Eike Beer2aae4a12006-09-29 01:59:46 -0700920#define INDEX(N) ((base->timer_jiffies >> (TVR_BITS + (N) * TVN_BITS)) & TVN_MASK)
921
922/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 * __run_timers - run all expired timers (if any) on this CPU.
924 * @base: the timer vector to be processed.
925 *
926 * This function cascades all vectors and executes all expired timer
927 * vectors.
928 */
Pavel Macheka6fa8e52008-01-30 13:30:00 +0100929static inline void __run_timers(struct tvec_base *base)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930{
931 struct timer_list *timer;
932
Oleg Nesterov3691c512006-03-31 02:30:30 -0800933 spin_lock_irq(&base->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 while (time_after_eq(jiffies, base->timer_jiffies)) {
Oleg Nesterov626ab0e2006-06-23 02:05:55 -0700935 struct list_head work_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 struct list_head *head = &work_list;
Thomas Gleixner68194572007-07-19 01:49:16 -0700937 int index = base->timer_jiffies & TVR_MASK;
Oleg Nesterov626ab0e2006-06-23 02:05:55 -0700938
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 /*
940 * Cascade timers:
941 */
942 if (!index &&
943 (!cascade(base, &base->tv2, INDEX(0))) &&
944 (!cascade(base, &base->tv3, INDEX(1))) &&
945 !cascade(base, &base->tv4, INDEX(2)))
946 cascade(base, &base->tv5, INDEX(3));
Oleg Nesterov626ab0e2006-06-23 02:05:55 -0700947 ++base->timer_jiffies;
948 list_replace_init(base->tv1.vec + index, &work_list);
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700949 while (!list_empty(head)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 void (*fn)(unsigned long);
951 unsigned long data;
952
Pavel Emelianovb5e61812007-05-08 00:30:19 -0700953 timer = list_first_entry(head, struct timer_list,entry);
Thomas Gleixner68194572007-07-19 01:49:16 -0700954 fn = timer->function;
955 data = timer->data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956
Ingo Molnar82f67cd2007-02-16 01:28:13 -0800957 timer_stats_account_timer(timer);
958
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 set_running_timer(base, timer);
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700960 detach_timer(timer, 1);
Johannes Berg6f2b9b92009-01-29 16:03:20 +0100961
Oleg Nesterov3691c512006-03-31 02:30:30 -0800962 spin_unlock_irq(&base->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 {
Jesper Juhlbe5b4fb2005-06-23 00:09:09 -0700964 int preempt_count = preempt_count();
Johannes Berg6f2b9b92009-01-29 16:03:20 +0100965
966#ifdef CONFIG_LOCKDEP
967 /*
968 * It is permissible to free the timer from
969 * inside the function that is called from
970 * it, this we need to take into account for
971 * lockdep too. To avoid bogus "held lock
972 * freed" warnings as well as problems when
973 * looking into timer->lockdep_map, make a
974 * copy and use that here.
975 */
976 struct lockdep_map lockdep_map =
977 timer->lockdep_map;
978#endif
979 /*
980 * Couple the lock chain with the lock chain at
981 * del_timer_sync() by acquiring the lock_map
982 * around the fn() call here and in
983 * del_timer_sync().
984 */
985 lock_map_acquire(&lockdep_map);
986
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 fn(data);
Johannes Berg6f2b9b92009-01-29 16:03:20 +0100988
989 lock_map_release(&lockdep_map);
990
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 if (preempt_count != preempt_count()) {
Pavel Machek4c9dc642008-01-30 13:30:00 +0100992 printk(KERN_ERR "huh, entered %p "
Jesper Juhlbe5b4fb2005-06-23 00:09:09 -0700993 "with preempt_count %08x, exited"
994 " with %08x?\n",
995 fn, preempt_count,
996 preempt_count());
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 BUG();
998 }
999 }
Oleg Nesterov3691c512006-03-31 02:30:30 -08001000 spin_lock_irq(&base->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 }
1002 }
1003 set_running_timer(base, NULL);
Oleg Nesterov3691c512006-03-31 02:30:30 -08001004 spin_unlock_irq(&base->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005}
1006
Russell Kingee9c5782008-04-20 13:59:33 +01001007#ifdef CONFIG_NO_HZ
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008/*
1009 * Find out when the next timer event is due to happen. This
1010 * is used on S/390 to stop all activity when a cpus is idle.
1011 * This functions needs to be called disabled.
1012 */
Pavel Macheka6fa8e52008-01-30 13:30:00 +01001013static unsigned long __next_timer_interrupt(struct tvec_base *base)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014{
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001015 unsigned long timer_jiffies = base->timer_jiffies;
Thomas Gleixnereaad0842007-05-29 23:47:39 +02001016 unsigned long expires = timer_jiffies + NEXT_TIMER_MAX_DELTA;
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001017 int index, slot, array, found = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 struct timer_list *nte;
Pavel Macheka6fa8e52008-01-30 13:30:00 +01001019 struct tvec *varray[4];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020
1021 /* Look for timer events in tv1. */
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001022 index = slot = timer_jiffies & TVR_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 do {
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001024 list_for_each_entry(nte, base->tv1.vec + slot, entry) {
Thomas Gleixner68194572007-07-19 01:49:16 -07001025 if (tbase_get_deferrable(nte->base))
1026 continue;
Venki Pallipadi6e453a62007-05-08 00:27:44 -07001027
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001028 found = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 expires = nte->expires;
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001030 /* Look at the cascade bucket(s)? */
1031 if (!index || slot < index)
1032 goto cascade;
1033 return expires;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034 }
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001035 slot = (slot + 1) & TVR_MASK;
1036 } while (slot != index);
1037
1038cascade:
1039 /* Calculate the next cascade event */
1040 if (index)
1041 timer_jiffies += TVR_SIZE - index;
1042 timer_jiffies >>= TVR_BITS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043
1044 /* Check tv2-tv5. */
1045 varray[0] = &base->tv2;
1046 varray[1] = &base->tv3;
1047 varray[2] = &base->tv4;
1048 varray[3] = &base->tv5;
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001049
1050 for (array = 0; array < 4; array++) {
Pavel Macheka6fa8e52008-01-30 13:30:00 +01001051 struct tvec *varp = varray[array];
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001052
1053 index = slot = timer_jiffies & TVN_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 do {
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001055 list_for_each_entry(nte, varp->vec + slot, entry) {
Jon Huntera0419882009-05-01 13:10:23 -07001056 if (tbase_get_deferrable(nte->base))
1057 continue;
1058
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001059 found = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 if (time_before(nte->expires, expires))
1061 expires = nte->expires;
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001062 }
1063 /*
1064 * Do we still search for the first timer or are
1065 * we looking up the cascade buckets ?
1066 */
1067 if (found) {
1068 /* Look at the cascade bucket(s)? */
1069 if (!index || slot < index)
1070 break;
1071 return expires;
1072 }
1073 slot = (slot + 1) & TVN_MASK;
1074 } while (slot != index);
1075
1076 if (index)
1077 timer_jiffies += TVN_SIZE - index;
1078 timer_jiffies >>= TVN_BITS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 }
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001080 return expires;
1081}
1082
1083/*
1084 * Check, if the next hrtimer event is before the next timer wheel
1085 * event:
1086 */
1087static unsigned long cmp_next_hrtimer_event(unsigned long now,
1088 unsigned long expires)
1089{
1090 ktime_t hr_delta = hrtimer_get_next_event();
1091 struct timespec tsdelta;
Thomas Gleixner9501b6c2007-03-25 14:31:17 +02001092 unsigned long delta;
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001093
1094 if (hr_delta.tv64 == KTIME_MAX)
1095 return expires;
1096
Thomas Gleixner9501b6c2007-03-25 14:31:17 +02001097 /*
1098 * Expired timer available, let it expire in the next tick
1099 */
1100 if (hr_delta.tv64 <= 0)
1101 return now + 1;
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001102
1103 tsdelta = ktime_to_timespec(hr_delta);
Thomas Gleixner9501b6c2007-03-25 14:31:17 +02001104 delta = timespec_to_jiffies(&tsdelta);
Thomas Gleixnereaad0842007-05-29 23:47:39 +02001105
1106 /*
1107 * Limit the delta to the max value, which is checked in
1108 * tick_nohz_stop_sched_tick():
1109 */
1110 if (delta > NEXT_TIMER_MAX_DELTA)
1111 delta = NEXT_TIMER_MAX_DELTA;
1112
Thomas Gleixner9501b6c2007-03-25 14:31:17 +02001113 /*
1114 * Take rounding errors in to account and make sure, that it
1115 * expires in the next tick. Otherwise we go into an endless
1116 * ping pong due to tick_nohz_stop_sched_tick() retriggering
1117 * the timer softirq
1118 */
1119 if (delta < 1)
1120 delta = 1;
1121 now += delta;
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001122 if (time_before(now, expires))
1123 return now;
1124 return expires;
1125}
1126
1127/**
Li Zefan8dce39c2007-11-05 14:51:10 -08001128 * get_next_timer_interrupt - return the jiffy of the next pending timer
Randy Dunlap05fb6bf2007-02-28 20:12:13 -08001129 * @now: current time (in jiffies)
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001130 */
Thomas Gleixnerfd064b92007-02-16 01:27:47 -08001131unsigned long get_next_timer_interrupt(unsigned long now)
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001132{
Pavel Macheka6fa8e52008-01-30 13:30:00 +01001133 struct tvec_base *base = __get_cpu_var(tvec_bases);
Thomas Gleixnerfd064b92007-02-16 01:27:47 -08001134 unsigned long expires;
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001135
1136 spin_lock(&base->lock);
1137 expires = __next_timer_interrupt(base);
Oleg Nesterov3691c512006-03-31 02:30:30 -08001138 spin_unlock(&base->lock);
Tony Lindgren69239742006-03-06 15:42:45 -08001139
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001140 if (time_before_eq(expires, now))
1141 return now;
Zachary Amsden0662b712006-05-20 15:00:24 -07001142
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001143 return cmp_next_hrtimer_event(now, expires);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144}
1145#endif
1146
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147/*
Daniel Walker5b4db0c2007-10-18 03:06:11 -07001148 * Called from the timer interrupt handler to charge one tick to the current
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 * process. user_tick is 1 if the tick is user time, 0 for system.
1150 */
1151void update_process_times(int user_tick)
1152{
1153 struct task_struct *p = current;
1154 int cpu = smp_processor_id();
1155
1156 /* Note: this timer irq context must be accounted for as well. */
Paul Mackerrasfa13a5a2007-11-09 22:39:38 +01001157 account_process_tick(p, user_tick);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 run_local_timers();
Paul E. McKenneya1572292009-08-22 13:56:51 -07001159 rcu_check_callbacks(cpu, user_tick);
Peter Zijlstrab845b512008-08-08 21:47:09 +02001160 printk_tick();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 scheduler_tick();
Thomas Gleixner68194572007-07-19 01:49:16 -07001162 run_posix_cpu_timers(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163}
1164
1165/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 * This function runs timers and the timer-tq in bottom half context.
1167 */
1168static void run_timer_softirq(struct softirq_action *h)
1169{
Pavel Macheka6fa8e52008-01-30 13:30:00 +01001170 struct tvec_base *base = __get_cpu_var(tvec_bases);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171
Peter Zijlstra925d5192009-03-30 19:07:02 +02001172 perf_counter_do_pending();
1173
Peter Zijlstrad3d74452008-01-25 21:08:31 +01001174 hrtimer_run_pending();
Ingo Molnar82f67cd2007-02-16 01:28:13 -08001175
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 if (time_after_eq(jiffies, base->timer_jiffies))
1177 __run_timers(base);
1178}
1179
1180/*
1181 * Called by the local, per-CPU timer interrupt on SMP.
1182 */
1183void run_local_timers(void)
1184{
Peter Zijlstrad3d74452008-01-25 21:08:31 +01001185 hrtimer_run_queues();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 raise_softirq(TIMER_SOFTIRQ);
Ingo Molnar6687a972006-03-24 03:18:41 -08001187 softlockup_tick();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188}
1189
1190/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191 * The 64-bit jiffies value is not atomic - you MUST NOT read it
1192 * without sampling the sequence number in xtime_lock.
1193 * jiffies is defined in the linker script...
1194 */
1195
Atsushi Nemoto3171a032006-09-29 02:00:32 -07001196void do_timer(unsigned long ticks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197{
Atsushi Nemoto3171a032006-09-29 02:00:32 -07001198 jiffies_64 += ticks;
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02001199 update_wall_time();
1200 calc_global_load();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201}
1202
1203#ifdef __ARCH_WANT_SYS_ALARM
1204
1205/*
1206 * For backwards compatibility? This can be done in libc so Alpha
1207 * and all newer ports shouldn't need it.
1208 */
Heiko Carstens58fd3aa2009-01-14 14:14:03 +01001209SYSCALL_DEFINE1(alarm, unsigned int, seconds)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210{
Thomas Gleixnerc08b8a42006-03-25 03:06:33 -08001211 return alarm_setitimer(seconds);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212}
1213
1214#endif
1215
1216#ifndef __alpha__
1217
1218/*
1219 * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this
1220 * should be moved into arch/i386 instead?
1221 */
1222
1223/**
1224 * sys_getpid - return the thread group id of the current process
1225 *
1226 * Note, despite the name, this returns the tgid not the pid. The tgid and
1227 * the pid are identical unless CLONE_THREAD was specified on clone() in
1228 * which case the tgid is the same in all threads of the same group.
1229 *
1230 * This is SMP safe as current->tgid does not change.
1231 */
Heiko Carstens58fd3aa2009-01-14 14:14:03 +01001232SYSCALL_DEFINE0(getpid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233{
Pavel Emelyanovb4888932007-10-18 23:40:14 -07001234 return task_tgid_vnr(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235}
1236
1237/*
Kirill Korotaev6997a6f2006-08-13 23:24:23 -07001238 * Accessing ->real_parent is not SMP-safe, it could
1239 * change from under us. However, we can use a stale
1240 * value of ->real_parent under rcu_read_lock(), see
1241 * release_task()->call_rcu(delayed_put_task_struct).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242 */
Heiko Carstensdbf040d2009-01-14 14:14:04 +01001243SYSCALL_DEFINE0(getppid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244{
1245 int pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246
Kirill Korotaev6997a6f2006-08-13 23:24:23 -07001247 rcu_read_lock();
Pavel Emelyanov6c5f3e72008-02-08 04:19:20 -08001248 pid = task_tgid_vnr(current->real_parent);
Kirill Korotaev6997a6f2006-08-13 23:24:23 -07001249 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 return pid;
1252}
1253
Heiko Carstensdbf040d2009-01-14 14:14:04 +01001254SYSCALL_DEFINE0(getuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255{
1256 /* Only we change this so SMP safe */
David Howells76aac0e2008-11-14 10:39:12 +11001257 return current_uid();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258}
1259
Heiko Carstensdbf040d2009-01-14 14:14:04 +01001260SYSCALL_DEFINE0(geteuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261{
1262 /* Only we change this so SMP safe */
David Howells76aac0e2008-11-14 10:39:12 +11001263 return current_euid();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264}
1265
Heiko Carstensdbf040d2009-01-14 14:14:04 +01001266SYSCALL_DEFINE0(getgid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267{
1268 /* Only we change this so SMP safe */
David Howells76aac0e2008-11-14 10:39:12 +11001269 return current_gid();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270}
1271
Heiko Carstensdbf040d2009-01-14 14:14:04 +01001272SYSCALL_DEFINE0(getegid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273{
1274 /* Only we change this so SMP safe */
David Howells76aac0e2008-11-14 10:39:12 +11001275 return current_egid();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276}
1277
1278#endif
1279
1280static void process_timeout(unsigned long __data)
1281{
Ingo Molnar36c8b582006-07-03 00:25:41 -07001282 wake_up_process((struct task_struct *)__data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283}
1284
1285/**
1286 * schedule_timeout - sleep until timeout
1287 * @timeout: timeout value in jiffies
1288 *
1289 * Make the current task sleep until @timeout jiffies have
1290 * elapsed. The routine will return immediately unless
1291 * the current task state has been set (see set_current_state()).
1292 *
1293 * You can set the task state as follows -
1294 *
1295 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1296 * pass before the routine returns. The routine will return 0
1297 *
1298 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1299 * delivered to the current task. In this case the remaining time
1300 * in jiffies will be returned, or 0 if the timer expired in time
1301 *
1302 * The current task state is guaranteed to be TASK_RUNNING when this
1303 * routine returns.
1304 *
1305 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1306 * the CPU away without a bound on the timeout. In this case the return
1307 * value will be %MAX_SCHEDULE_TIMEOUT.
1308 *
1309 * In all cases the return value is guaranteed to be non-negative.
1310 */
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -08001311signed long __sched schedule_timeout(signed long timeout)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312{
1313 struct timer_list timer;
1314 unsigned long expire;
1315
1316 switch (timeout)
1317 {
1318 case MAX_SCHEDULE_TIMEOUT:
1319 /*
1320 * These two special cases are useful to be comfortable
1321 * in the caller. Nothing more. We could take
1322 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1323 * but I' d like to return a valid offset (>=0) to allow
1324 * the caller to do everything it want with the retval.
1325 */
1326 schedule();
1327 goto out;
1328 default:
1329 /*
1330 * Another bit of PARANOID. Note that the retval will be
1331 * 0 since no piece of kernel is supposed to do a check
1332 * for a negative retval of schedule_timeout() (since it
1333 * should never happens anyway). You just have the printk()
1334 * that will tell you if something is gone wrong and where.
1335 */
Andrew Morton5b149bc2006-12-22 01:10:14 -08001336 if (timeout < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 printk(KERN_ERR "schedule_timeout: wrong timeout "
Andrew Morton5b149bc2006-12-22 01:10:14 -08001338 "value %lx\n", timeout);
1339 dump_stack();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 current->state = TASK_RUNNING;
1341 goto out;
1342 }
1343 }
1344
1345 expire = timeout + jiffies;
1346
Thomas Gleixnerc6f3a972008-04-30 00:55:03 -07001347 setup_timer_on_stack(&timer, process_timeout, (unsigned long)current);
Arun R Bharadwaj597d0272009-04-16 12:13:26 +05301348 __mod_timer(&timer, expire, false, TIMER_NOT_PINNED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 schedule();
1350 del_singleshot_timer_sync(&timer);
1351
Thomas Gleixnerc6f3a972008-04-30 00:55:03 -07001352 /* Remove the timer from the object tracker */
1353 destroy_timer_on_stack(&timer);
1354
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 timeout = expire - jiffies;
1356
1357 out:
1358 return timeout < 0 ? 0 : timeout;
1359}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360EXPORT_SYMBOL(schedule_timeout);
1361
Andrew Morton8a1c1752005-09-13 01:25:15 -07001362/*
1363 * We can use __set_current_state() here because schedule_timeout() calls
1364 * schedule() unconditionally.
1365 */
Nishanth Aravamudan64ed93a2005-09-10 00:27:21 -07001366signed long __sched schedule_timeout_interruptible(signed long timeout)
1367{
Andrew Mortona5a0d522005-10-30 15:01:42 -08001368 __set_current_state(TASK_INTERRUPTIBLE);
1369 return schedule_timeout(timeout);
Nishanth Aravamudan64ed93a2005-09-10 00:27:21 -07001370}
1371EXPORT_SYMBOL(schedule_timeout_interruptible);
1372
Matthew Wilcox294d5cc2007-12-06 11:59:46 -05001373signed long __sched schedule_timeout_killable(signed long timeout)
1374{
1375 __set_current_state(TASK_KILLABLE);
1376 return schedule_timeout(timeout);
1377}
1378EXPORT_SYMBOL(schedule_timeout_killable);
1379
Nishanth Aravamudan64ed93a2005-09-10 00:27:21 -07001380signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1381{
Andrew Mortona5a0d522005-10-30 15:01:42 -08001382 __set_current_state(TASK_UNINTERRUPTIBLE);
1383 return schedule_timeout(timeout);
Nishanth Aravamudan64ed93a2005-09-10 00:27:21 -07001384}
1385EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1386
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387/* Thread ID - the internal kernel "pid" */
Heiko Carstens58fd3aa2009-01-14 14:14:03 +01001388SYSCALL_DEFINE0(gettid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389{
Pavel Emelyanovb4888932007-10-18 23:40:14 -07001390 return task_pid_vnr(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391}
1392
Rolf Eike Beer2aae4a12006-09-29 01:59:46 -07001393/**
Kyle McMartind4d23ad2007-02-10 01:46:00 -08001394 * do_sysinfo - fill in sysinfo struct
Rolf Eike Beer2aae4a12006-09-29 01:59:46 -07001395 * @info: pointer to buffer to fill
Thomas Gleixner68194572007-07-19 01:49:16 -07001396 */
Kyle McMartind4d23ad2007-02-10 01:46:00 -08001397int do_sysinfo(struct sysinfo *info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 unsigned long mem_total, sav_total;
1400 unsigned int mem_unit, bitcount;
Thomas Gleixner2d024942009-05-02 20:08:52 +02001401 struct timespec tp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402
Kyle McMartind4d23ad2007-02-10 01:46:00 -08001403 memset(info, 0, sizeof(struct sysinfo));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404
Thomas Gleixner2d024942009-05-02 20:08:52 +02001405 ktime_get_ts(&tp);
1406 monotonic_to_bootbased(&tp);
1407 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408
Thomas Gleixner2d024942009-05-02 20:08:52 +02001409 get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410
Thomas Gleixner2d024942009-05-02 20:08:52 +02001411 info->procs = nr_threads;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412
Kyle McMartind4d23ad2007-02-10 01:46:00 -08001413 si_meminfo(info);
1414 si_swapinfo(info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415
1416 /*
1417 * If the sum of all the available memory (i.e. ram + swap)
1418 * is less than can be stored in a 32 bit unsigned long then
1419 * we can be binary compatible with 2.2.x kernels. If not,
1420 * well, in that case 2.2.x was broken anyways...
1421 *
1422 * -Erik Andersen <andersee@debian.org>
1423 */
1424
Kyle McMartind4d23ad2007-02-10 01:46:00 -08001425 mem_total = info->totalram + info->totalswap;
1426 if (mem_total < info->totalram || mem_total < info->totalswap)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 goto out;
1428 bitcount = 0;
Kyle McMartind4d23ad2007-02-10 01:46:00 -08001429 mem_unit = info->mem_unit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 while (mem_unit > 1) {
1431 bitcount++;
1432 mem_unit >>= 1;
1433 sav_total = mem_total;
1434 mem_total <<= 1;
1435 if (mem_total < sav_total)
1436 goto out;
1437 }
1438
1439 /*
1440 * If mem_total did not overflow, multiply all memory values by
Kyle McMartind4d23ad2007-02-10 01:46:00 -08001441 * info->mem_unit and set it to 1. This leaves things compatible
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1443 * kernels...
1444 */
1445
Kyle McMartind4d23ad2007-02-10 01:46:00 -08001446 info->mem_unit = 1;
1447 info->totalram <<= bitcount;
1448 info->freeram <<= bitcount;
1449 info->sharedram <<= bitcount;
1450 info->bufferram <<= bitcount;
1451 info->totalswap <<= bitcount;
1452 info->freeswap <<= bitcount;
1453 info->totalhigh <<= bitcount;
1454 info->freehigh <<= bitcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455
Kyle McMartind4d23ad2007-02-10 01:46:00 -08001456out:
1457 return 0;
1458}
1459
Heiko Carstens1e7bfb22009-01-14 14:14:29 +01001460SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
Kyle McMartind4d23ad2007-02-10 01:46:00 -08001461{
1462 struct sysinfo val;
1463
1464 do_sysinfo(&val);
1465
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
1467 return -EFAULT;
1468
1469 return 0;
1470}
1471
Adrian Bunkb4be6252007-12-18 18:05:58 +01001472static int __cpuinit init_timers_cpu(int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473{
1474 int j;
Pavel Macheka6fa8e52008-01-30 13:30:00 +01001475 struct tvec_base *base;
Adrian Bunkb4be6252007-12-18 18:05:58 +01001476 static char __cpuinitdata tvec_base_done[NR_CPUS];
Oleg Nesterov55c888d2005-06-23 00:08:56 -07001477
Andrew Mortonba6edfc2006-04-10 22:53:58 -07001478 if (!tvec_base_done[cpu]) {
Jan Beulicha4a61982006-03-24 03:15:54 -08001479 static char boot_done;
1480
Jan Beulicha4a61982006-03-24 03:15:54 -08001481 if (boot_done) {
Andrew Mortonba6edfc2006-04-10 22:53:58 -07001482 /*
1483 * The APs use this path later in boot
1484 */
Christoph Lameter94f60302007-07-17 04:03:29 -07001485 base = kmalloc_node(sizeof(*base),
1486 GFP_KERNEL | __GFP_ZERO,
Jan Beulicha4a61982006-03-24 03:15:54 -08001487 cpu_to_node(cpu));
1488 if (!base)
1489 return -ENOMEM;
Venki Pallipadi6e453a62007-05-08 00:27:44 -07001490
1491 /* Make sure that tvec_base is 2 byte aligned */
1492 if (tbase_get_deferrable(base)) {
1493 WARN_ON(1);
1494 kfree(base);
1495 return -ENOMEM;
1496 }
Andrew Mortonba6edfc2006-04-10 22:53:58 -07001497 per_cpu(tvec_bases, cpu) = base;
Jan Beulicha4a61982006-03-24 03:15:54 -08001498 } else {
Andrew Mortonba6edfc2006-04-10 22:53:58 -07001499 /*
1500 * This is for the boot CPU - we use compile-time
1501 * static initialisation because per-cpu memory isn't
1502 * ready yet and because the memory allocators are not
1503 * initialised either.
1504 */
Jan Beulicha4a61982006-03-24 03:15:54 -08001505 boot_done = 1;
Andrew Mortonba6edfc2006-04-10 22:53:58 -07001506 base = &boot_tvec_bases;
Jan Beulicha4a61982006-03-24 03:15:54 -08001507 }
Andrew Mortonba6edfc2006-04-10 22:53:58 -07001508 tvec_base_done[cpu] = 1;
1509 } else {
1510 base = per_cpu(tvec_bases, cpu);
Jan Beulicha4a61982006-03-24 03:15:54 -08001511 }
Andrew Mortonba6edfc2006-04-10 22:53:58 -07001512
Oleg Nesterov3691c512006-03-31 02:30:30 -08001513 spin_lock_init(&base->lock);
Ingo Molnard730e882006-07-03 00:25:10 -07001514
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515 for (j = 0; j < TVN_SIZE; j++) {
1516 INIT_LIST_HEAD(base->tv5.vec + j);
1517 INIT_LIST_HEAD(base->tv4.vec + j);
1518 INIT_LIST_HEAD(base->tv3.vec + j);
1519 INIT_LIST_HEAD(base->tv2.vec + j);
1520 }
1521 for (j = 0; j < TVR_SIZE; j++)
1522 INIT_LIST_HEAD(base->tv1.vec + j);
1523
1524 base->timer_jiffies = jiffies;
Jan Beulicha4a61982006-03-24 03:15:54 -08001525 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526}
1527
1528#ifdef CONFIG_HOTPLUG_CPU
Pavel Macheka6fa8e52008-01-30 13:30:00 +01001529static void migrate_timer_list(struct tvec_base *new_base, struct list_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530{
1531 struct timer_list *timer;
1532
1533 while (!list_empty(head)) {
Pavel Emelianovb5e61812007-05-08 00:30:19 -07001534 timer = list_first_entry(head, struct timer_list, entry);
Oleg Nesterov55c888d2005-06-23 00:08:56 -07001535 detach_timer(timer, 0);
Venki Pallipadi6e453a62007-05-08 00:27:44 -07001536 timer_set_base(timer, new_base);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537 internal_add_timer(new_base, timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539}
1540
Randy Dunlap48ccf3d2008-01-21 17:18:25 -08001541static void __cpuinit migrate_timers(int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542{
Pavel Macheka6fa8e52008-01-30 13:30:00 +01001543 struct tvec_base *old_base;
1544 struct tvec_base *new_base;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545 int i;
1546
1547 BUG_ON(cpu_online(cpu));
Jan Beulicha4a61982006-03-24 03:15:54 -08001548 old_base = per_cpu(tvec_bases, cpu);
1549 new_base = get_cpu_var(tvec_bases);
Oleg Nesterovd82f0b02008-08-20 16:46:04 -07001550 /*
1551 * The caller is globally serialized and nobody else
1552 * takes two locks at once, deadlock is not possible.
1553 */
1554 spin_lock_irq(&new_base->lock);
Oleg Nesterov0d180402008-04-04 20:54:10 +02001555 spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556
Oleg Nesterov3691c512006-03-31 02:30:30 -08001557 BUG_ON(old_base->running_timer);
1558
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559 for (i = 0; i < TVR_SIZE; i++)
Oleg Nesterov55c888d2005-06-23 00:08:56 -07001560 migrate_timer_list(new_base, old_base->tv1.vec + i);
1561 for (i = 0; i < TVN_SIZE; i++) {
1562 migrate_timer_list(new_base, old_base->tv2.vec + i);
1563 migrate_timer_list(new_base, old_base->tv3.vec + i);
1564 migrate_timer_list(new_base, old_base->tv4.vec + i);
1565 migrate_timer_list(new_base, old_base->tv5.vec + i);
1566 }
1567
Oleg Nesterov0d180402008-04-04 20:54:10 +02001568 spin_unlock(&old_base->lock);
Oleg Nesterovd82f0b02008-08-20 16:46:04 -07001569 spin_unlock_irq(&new_base->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 put_cpu_var(tvec_bases);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571}
1572#endif /* CONFIG_HOTPLUG_CPU */
1573
Chandra Seetharaman8c78f302006-07-30 03:03:35 -07001574static int __cpuinit timer_cpu_notify(struct notifier_block *self,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575 unsigned long action, void *hcpu)
1576{
1577 long cpu = (long)hcpu;
1578 switch(action) {
1579 case CPU_UP_PREPARE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07001580 case CPU_UP_PREPARE_FROZEN:
Jan Beulicha4a61982006-03-24 03:15:54 -08001581 if (init_timers_cpu(cpu) < 0)
1582 return NOTIFY_BAD;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583 break;
1584#ifdef CONFIG_HOTPLUG_CPU
1585 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07001586 case CPU_DEAD_FROZEN:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587 migrate_timers(cpu);
1588 break;
1589#endif
1590 default:
1591 break;
1592 }
1593 return NOTIFY_OK;
1594}
1595
Chandra Seetharaman8c78f302006-07-30 03:03:35 -07001596static struct notifier_block __cpuinitdata timers_nb = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597 .notifier_call = timer_cpu_notify,
1598};
1599
1600
1601void __init init_timers(void)
1602{
Akinobu Mita07dccf32006-09-29 02:00:22 -07001603 int err = timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 (void *)(long)smp_processor_id());
Akinobu Mita07dccf32006-09-29 02:00:22 -07001605
Ingo Molnar82f67cd2007-02-16 01:28:13 -08001606 init_timer_stats();
1607
Akinobu Mita07dccf32006-09-29 02:00:22 -07001608 BUG_ON(err == NOTIFY_BAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609 register_cpu_notifier(&timers_nb);
Carlos R. Mafra962cf362008-05-15 11:15:37 -03001610 open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611}
1612
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613/**
1614 * msleep - sleep safely even with waitqueue interruptions
1615 * @msecs: Time in milliseconds to sleep for
1616 */
1617void msleep(unsigned int msecs)
1618{
1619 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1620
Nishanth Aravamudan75bcc8c2005-09-10 00:27:24 -07001621 while (timeout)
1622 timeout = schedule_timeout_uninterruptible(timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623}
1624
1625EXPORT_SYMBOL(msleep);
1626
1627/**
Domen Puncer96ec3ef2005-06-25 14:58:43 -07001628 * msleep_interruptible - sleep waiting for signals
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629 * @msecs: Time in milliseconds to sleep for
1630 */
1631unsigned long msleep_interruptible(unsigned int msecs)
1632{
1633 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1634
Nishanth Aravamudan75bcc8c2005-09-10 00:27:24 -07001635 while (timeout && !signal_pending(current))
1636 timeout = schedule_timeout_interruptible(timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 return jiffies_to_msecs(timeout);
1638}
1639
1640EXPORT_SYMBOL(msleep_interruptible);