blob: 54d3912f8cadd497d09546ad8b9b1ba42369634d [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
383 if (unlikely(tbase_get_deferrable(timer->base)))
384 flag |= TIMER_STATS_FLAG_DEFERRABLE;
385
386 timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
387 timer->function, timer->start_comm, flag);
388}
389
390#else
391static void timer_stats_account_timer(struct timer_list *timer) {}
Ingo Molnar82f67cd2007-02-16 01:28:13 -0800392#endif
393
Thomas Gleixnerc6f3a972008-04-30 00:55:03 -0700394#ifdef CONFIG_DEBUG_OBJECTS_TIMERS
395
396static struct debug_obj_descr timer_debug_descr;
397
398/*
399 * fixup_init is called when:
400 * - an active object is initialized
401 */
402static int timer_fixup_init(void *addr, enum debug_obj_state state)
403{
404 struct timer_list *timer = addr;
405
406 switch (state) {
407 case ODEBUG_STATE_ACTIVE:
408 del_timer_sync(timer);
409 debug_object_init(timer, &timer_debug_descr);
410 return 1;
411 default:
412 return 0;
413 }
414}
415
416/*
417 * fixup_activate is called when:
418 * - an active object is activated
419 * - an unknown object is activated (might be a statically initialized object)
420 */
421static int timer_fixup_activate(void *addr, enum debug_obj_state state)
422{
423 struct timer_list *timer = addr;
424
425 switch (state) {
426
427 case ODEBUG_STATE_NOTAVAILABLE:
428 /*
429 * This is not really a fixup. The timer was
430 * statically initialized. We just make sure that it
431 * is tracked in the object tracker.
432 */
433 if (timer->entry.next == NULL &&
434 timer->entry.prev == TIMER_ENTRY_STATIC) {
435 debug_object_init(timer, &timer_debug_descr);
436 debug_object_activate(timer, &timer_debug_descr);
437 return 0;
438 } else {
439 WARN_ON_ONCE(1);
440 }
441 return 0;
442
443 case ODEBUG_STATE_ACTIVE:
444 WARN_ON(1);
445
446 default:
447 return 0;
448 }
449}
450
451/*
452 * fixup_free is called when:
453 * - an active object is freed
454 */
455static int timer_fixup_free(void *addr, enum debug_obj_state state)
456{
457 struct timer_list *timer = addr;
458
459 switch (state) {
460 case ODEBUG_STATE_ACTIVE:
461 del_timer_sync(timer);
462 debug_object_free(timer, &timer_debug_descr);
463 return 1;
464 default:
465 return 0;
466 }
467}
468
469static struct debug_obj_descr timer_debug_descr = {
470 .name = "timer_list",
471 .fixup_init = timer_fixup_init,
472 .fixup_activate = timer_fixup_activate,
473 .fixup_free = timer_fixup_free,
474};
475
476static inline void debug_timer_init(struct timer_list *timer)
477{
478 debug_object_init(timer, &timer_debug_descr);
479}
480
481static inline void debug_timer_activate(struct timer_list *timer)
482{
483 debug_object_activate(timer, &timer_debug_descr);
484}
485
486static inline void debug_timer_deactivate(struct timer_list *timer)
487{
488 debug_object_deactivate(timer, &timer_debug_descr);
489}
490
491static inline void debug_timer_free(struct timer_list *timer)
492{
493 debug_object_free(timer, &timer_debug_descr);
494}
495
Johannes Berg6f2b9b92009-01-29 16:03:20 +0100496static void __init_timer(struct timer_list *timer,
497 const char *name,
498 struct lock_class_key *key);
Thomas Gleixnerc6f3a972008-04-30 00:55:03 -0700499
Johannes Berg6f2b9b92009-01-29 16:03:20 +0100500void init_timer_on_stack_key(struct timer_list *timer,
501 const char *name,
502 struct lock_class_key *key)
Thomas Gleixnerc6f3a972008-04-30 00:55:03 -0700503{
504 debug_object_init_on_stack(timer, &timer_debug_descr);
Johannes Berg6f2b9b92009-01-29 16:03:20 +0100505 __init_timer(timer, name, key);
Thomas Gleixnerc6f3a972008-04-30 00:55:03 -0700506}
Johannes Berg6f2b9b92009-01-29 16:03:20 +0100507EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
Thomas Gleixnerc6f3a972008-04-30 00:55:03 -0700508
509void destroy_timer_on_stack(struct timer_list *timer)
510{
511 debug_object_free(timer, &timer_debug_descr);
512}
513EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
514
515#else
516static inline void debug_timer_init(struct timer_list *timer) { }
517static inline void debug_timer_activate(struct timer_list *timer) { }
518static inline void debug_timer_deactivate(struct timer_list *timer) { }
519#endif
520
Johannes Berg6f2b9b92009-01-29 16:03:20 +0100521static void __init_timer(struct timer_list *timer,
522 const char *name,
523 struct lock_class_key *key)
Thomas Gleixnerc6f3a972008-04-30 00:55:03 -0700524{
525 timer->entry.next = NULL;
526 timer->base = __raw_get_cpu_var(tvec_bases);
527#ifdef CONFIG_TIMER_STATS
528 timer->start_site = NULL;
529 timer->start_pid = -1;
530 memset(timer->start_comm, 0, TASK_COMM_LEN);
531#endif
Johannes Berg6f2b9b92009-01-29 16:03:20 +0100532 lockdep_init_map(&timer->lockdep_map, name, key, 0);
Thomas Gleixnerc6f3a972008-04-30 00:55:03 -0700533}
534
Rolf Eike Beer2aae4a12006-09-29 01:59:46 -0700535/**
Randy Dunlap633fe792009-04-01 17:47:23 -0700536 * init_timer_key - initialize a timer
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700537 * @timer: the timer to be initialized
Randy Dunlap633fe792009-04-01 17:47:23 -0700538 * @name: name of the timer
539 * @key: lockdep class key of the fake lock used for tracking timer
540 * sync lock dependencies
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700541 *
Randy Dunlap633fe792009-04-01 17:47:23 -0700542 * init_timer_key() must be done to a timer prior calling *any* of the
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700543 * other timer functions.
544 */
Johannes Berg6f2b9b92009-01-29 16:03:20 +0100545void init_timer_key(struct timer_list *timer,
546 const char *name,
547 struct lock_class_key *key)
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700548{
Thomas Gleixnerc6f3a972008-04-30 00:55:03 -0700549 debug_timer_init(timer);
Johannes Berg6f2b9b92009-01-29 16:03:20 +0100550 __init_timer(timer, name, key);
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700551}
Johannes Berg6f2b9b92009-01-29 16:03:20 +0100552EXPORT_SYMBOL(init_timer_key);
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700553
Johannes Berg6f2b9b92009-01-29 16:03:20 +0100554void init_timer_deferrable_key(struct timer_list *timer,
555 const char *name,
556 struct lock_class_key *key)
Venki Pallipadi6e453a62007-05-08 00:27:44 -0700557{
Johannes Berg6f2b9b92009-01-29 16:03:20 +0100558 init_timer_key(timer, name, key);
Venki Pallipadi6e453a62007-05-08 00:27:44 -0700559 timer_set_deferrable(timer);
560}
Johannes Berg6f2b9b92009-01-29 16:03:20 +0100561EXPORT_SYMBOL(init_timer_deferrable_key);
Venki Pallipadi6e453a62007-05-08 00:27:44 -0700562
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700563static inline void detach_timer(struct timer_list *timer,
Ingo Molnar82f67cd2007-02-16 01:28:13 -0800564 int clear_pending)
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700565{
566 struct list_head *entry = &timer->entry;
567
Thomas Gleixnerc6f3a972008-04-30 00:55:03 -0700568 debug_timer_deactivate(timer);
569
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700570 __list_del(entry->prev, entry->next);
571 if (clear_pending)
572 entry->next = NULL;
573 entry->prev = LIST_POISON2;
574}
575
576/*
Oleg Nesterov3691c512006-03-31 02:30:30 -0800577 * We are using hashed locking: holding per_cpu(tvec_bases).lock
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700578 * means that all timers which are tied to this base via timer->base are
579 * locked, and the base itself is locked too.
580 *
581 * So __run_timers/migrate_timers can safely modify all timers which could
582 * be found on ->tvX lists.
583 *
584 * When the timer's base is locked, and the timer removed from list, it is
585 * possible to set timer->base = NULL and drop the lock: the timer remains
586 * locked.
587 */
Pavel Macheka6fa8e52008-01-30 13:30:00 +0100588static struct tvec_base *lock_timer_base(struct timer_list *timer,
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700589 unsigned long *flags)
Josh Triplett89e7e3742006-09-29 01:59:36 -0700590 __acquires(timer->base->lock)
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700591{
Pavel Macheka6fa8e52008-01-30 13:30:00 +0100592 struct tvec_base *base;
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700593
594 for (;;) {
Pavel Macheka6fa8e52008-01-30 13:30:00 +0100595 struct tvec_base *prelock_base = timer->base;
Venki Pallipadi6e453a62007-05-08 00:27:44 -0700596 base = tbase_get_base(prelock_base);
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700597 if (likely(base != NULL)) {
598 spin_lock_irqsave(&base->lock, *flags);
Venki Pallipadi6e453a62007-05-08 00:27:44 -0700599 if (likely(prelock_base == timer->base))
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700600 return base;
601 /* The timer has migrated to another CPU */
602 spin_unlock_irqrestore(&base->lock, *flags);
603 }
604 cpu_relax();
605 }
606}
607
Ingo Molnar74019222009-02-18 12:23:29 +0100608static inline int
Arun R Bharadwaj597d0272009-04-16 12:13:26 +0530609__mod_timer(struct timer_list *timer, unsigned long expires,
610 bool pending_only, int pinned)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611{
Pavel Macheka6fa8e52008-01-30 13:30:00 +0100612 struct tvec_base *base, *new_base;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 unsigned long flags;
Arun R Bharadwajeea08f32009-04-16 12:16:41 +0530614 int ret = 0 , cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615
Ingo Molnar82f67cd2007-02-16 01:28:13 -0800616 timer_stats_timer_set_start_info(timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 BUG_ON(!timer->function);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700619 base = lock_timer_base(timer, &flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700621 if (timer_pending(timer)) {
622 detach_timer(timer, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 ret = 1;
Ingo Molnar74019222009-02-18 12:23:29 +0100624 } else {
625 if (pending_only)
626 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 }
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700628
Thomas Gleixnerc6f3a972008-04-30 00:55:03 -0700629 debug_timer_activate(timer);
630
Jan Beulicha4a61982006-03-24 03:15:54 -0800631 new_base = __get_cpu_var(tvec_bases);
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700632
Arun R Bharadwajeea08f32009-04-16 12:16:41 +0530633 cpu = smp_processor_id();
634
635#if defined(CONFIG_NO_HZ) && defined(CONFIG_SMP)
636 if (!pinned && get_sysctl_timer_migration() && idle_cpu(cpu)) {
637 int preferred_cpu = get_nohz_load_balancer();
638
639 if (preferred_cpu >= 0)
640 cpu = preferred_cpu;
641 }
642#endif
643 new_base = per_cpu(tvec_bases, cpu);
644
Oleg Nesterov3691c512006-03-31 02:30:30 -0800645 if (base != new_base) {
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700646 /*
647 * We are trying to schedule the timer on the local CPU.
648 * However we can't change timer's base while it is running,
649 * otherwise del_timer_sync() can't detect that the timer's
650 * handler yet has not finished. This also guarantees that
651 * the timer is serialized wrt itself.
652 */
Oleg Nesterova2c348f2006-03-31 02:30:31 -0800653 if (likely(base->running_timer != timer)) {
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700654 /* See the comment in lock_timer_base() */
Venki Pallipadi6e453a62007-05-08 00:27:44 -0700655 timer_set_base(timer, NULL);
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700656 spin_unlock(&base->lock);
Oleg Nesterova2c348f2006-03-31 02:30:31 -0800657 base = new_base;
658 spin_lock(&base->lock);
Venki Pallipadi6e453a62007-05-08 00:27:44 -0700659 timer_set_base(timer, base);
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700660 }
661 }
662
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 timer->expires = expires;
Oleg Nesterova2c348f2006-03-31 02:30:31 -0800664 internal_add_timer(base, timer);
Ingo Molnar74019222009-02-18 12:23:29 +0100665
666out_unlock:
Oleg Nesterova2c348f2006-03-31 02:30:31 -0800667 spin_unlock_irqrestore(&base->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668
669 return ret;
670}
671
Ingo Molnar74019222009-02-18 12:23:29 +0100672/**
673 * mod_timer_pending - modify a pending timer's timeout
674 * @timer: the pending timer to be modified
675 * @expires: new timeout in jiffies
676 *
677 * mod_timer_pending() is the same for pending timers as mod_timer(),
678 * but will not re-activate and modify already deleted timers.
679 *
680 * It is useful for unserialized use of timers.
681 */
682int mod_timer_pending(struct timer_list *timer, unsigned long expires)
683{
Arun R Bharadwaj597d0272009-04-16 12:13:26 +0530684 return __mod_timer(timer, expires, true, TIMER_NOT_PINNED);
Ingo Molnar74019222009-02-18 12:23:29 +0100685}
686EXPORT_SYMBOL(mod_timer_pending);
687
688/**
689 * mod_timer - modify a timer's timeout
690 * @timer: the timer to be modified
691 * @expires: new timeout in jiffies
692 *
693 * mod_timer() is a more efficient way to update the expire field of an
694 * active timer (if the timer is inactive it will be activated)
695 *
696 * mod_timer(timer, expires) is equivalent to:
697 *
698 * del_timer(timer); timer->expires = expires; add_timer(timer);
699 *
700 * Note that if there are multiple unserialized concurrent users of the
701 * same timer, then mod_timer() is the only safe way to modify the timeout,
702 * since add_timer() cannot modify an already running timer.
703 *
704 * The function returns whether it has modified a pending timer or not.
705 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
706 * active timer returns 1.)
707 */
708int mod_timer(struct timer_list *timer, unsigned long expires)
709{
710 /*
711 * This is a common optimization triggered by the
712 * networking code - if the timer is re-modified
713 * to be the same thing then just return:
714 */
715 if (timer->expires == expires && timer_pending(timer))
716 return 1;
717
Arun R Bharadwaj597d0272009-04-16 12:13:26 +0530718 return __mod_timer(timer, expires, false, TIMER_NOT_PINNED);
Ingo Molnar74019222009-02-18 12:23:29 +0100719}
720EXPORT_SYMBOL(mod_timer);
721
722/**
Arun R Bharadwaj597d0272009-04-16 12:13:26 +0530723 * mod_timer_pinned - modify a timer's timeout
724 * @timer: the timer to be modified
725 * @expires: new timeout in jiffies
726 *
727 * mod_timer_pinned() is a way to update the expire field of an
728 * active timer (if the timer is inactive it will be activated)
729 * and not allow the timer to be migrated to a different CPU.
730 *
731 * mod_timer_pinned(timer, expires) is equivalent to:
732 *
733 * del_timer(timer); timer->expires = expires; add_timer(timer);
734 */
735int mod_timer_pinned(struct timer_list *timer, unsigned long expires)
736{
737 if (timer->expires == expires && timer_pending(timer))
738 return 1;
739
740 return __mod_timer(timer, expires, false, TIMER_PINNED);
741}
742EXPORT_SYMBOL(mod_timer_pinned);
743
744/**
Ingo Molnar74019222009-02-18 12:23:29 +0100745 * add_timer - start a timer
746 * @timer: the timer to be added
747 *
748 * The kernel will do a ->function(->data) callback from the
749 * timer interrupt at the ->expires point in the future. The
750 * current time is 'jiffies'.
751 *
752 * The timer's ->expires, ->function (and if the handler uses it, ->data)
753 * fields must be set prior calling this function.
754 *
755 * Timers with an ->expires field in the past will be executed in the next
756 * timer tick.
757 */
758void add_timer(struct timer_list *timer)
759{
760 BUG_ON(timer_pending(timer));
761 mod_timer(timer, timer->expires);
762}
763EXPORT_SYMBOL(add_timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764
Rolf Eike Beer2aae4a12006-09-29 01:59:46 -0700765/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 * add_timer_on - start a timer on a particular CPU
767 * @timer: the timer to be added
768 * @cpu: the CPU to start it on
769 *
770 * This is not very scalable on SMP. Double adds are not possible.
771 */
772void add_timer_on(struct timer_list *timer, int cpu)
773{
Pavel Macheka6fa8e52008-01-30 13:30:00 +0100774 struct tvec_base *base = per_cpu(tvec_bases, cpu);
Thomas Gleixner68194572007-07-19 01:49:16 -0700775 unsigned long flags;
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700776
Ingo Molnar82f67cd2007-02-16 01:28:13 -0800777 timer_stats_timer_set_start_info(timer);
Thomas Gleixner68194572007-07-19 01:49:16 -0700778 BUG_ON(timer_pending(timer) || !timer->function);
Oleg Nesterov3691c512006-03-31 02:30:30 -0800779 spin_lock_irqsave(&base->lock, flags);
Venki Pallipadi6e453a62007-05-08 00:27:44 -0700780 timer_set_base(timer, base);
Thomas Gleixnerc6f3a972008-04-30 00:55:03 -0700781 debug_timer_activate(timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 internal_add_timer(base, timer);
Thomas Gleixner06d83082008-03-22 09:20:24 +0100783 /*
784 * Check whether the other CPU is idle and needs to be
785 * triggered to reevaluate the timer wheel when nohz is
786 * active. We are protected against the other CPU fiddling
787 * with the timer by holding the timer base lock. This also
788 * makes sure that a CPU on the way to idle can not evaluate
789 * the timer wheel.
790 */
791 wake_up_idle_cpu(cpu);
Oleg Nesterov3691c512006-03-31 02:30:30 -0800792 spin_unlock_irqrestore(&base->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793}
Andi Kleena9862e02009-05-19 22:49:07 +0200794EXPORT_SYMBOL_GPL(add_timer_on);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795
Rolf Eike Beer2aae4a12006-09-29 01:59:46 -0700796/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 * del_timer - deactive a timer.
798 * @timer: the timer to be deactivated
799 *
800 * del_timer() deactivates a timer - this works on both active and inactive
801 * timers.
802 *
803 * The function returns whether it has deactivated a pending timer or not.
804 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
805 * active timer returns 1.)
806 */
807int del_timer(struct timer_list *timer)
808{
Pavel Macheka6fa8e52008-01-30 13:30:00 +0100809 struct tvec_base *base;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 unsigned long flags;
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700811 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812
Ingo Molnar82f67cd2007-02-16 01:28:13 -0800813 timer_stats_timer_clear_start_info(timer);
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700814 if (timer_pending(timer)) {
815 base = lock_timer_base(timer, &flags);
816 if (timer_pending(timer)) {
817 detach_timer(timer, 1);
818 ret = 1;
819 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 spin_unlock_irqrestore(&base->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700823 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825EXPORT_SYMBOL(del_timer);
826
827#ifdef CONFIG_SMP
Rolf Eike Beer2aae4a12006-09-29 01:59:46 -0700828/**
829 * try_to_del_timer_sync - Try to deactivate a timer
830 * @timer: timer do del
831 *
Oleg Nesterovfd450b72005-06-23 00:08:59 -0700832 * This function tries to deactivate a timer. Upon successful (ret >= 0)
833 * exit the timer is not queued and the handler is not running on any CPU.
834 *
835 * It must not be called from interrupt contexts.
836 */
837int try_to_del_timer_sync(struct timer_list *timer)
838{
Pavel Macheka6fa8e52008-01-30 13:30:00 +0100839 struct tvec_base *base;
Oleg Nesterovfd450b72005-06-23 00:08:59 -0700840 unsigned long flags;
841 int ret = -1;
842
843 base = lock_timer_base(timer, &flags);
844
845 if (base->running_timer == timer)
846 goto out;
847
848 ret = 0;
849 if (timer_pending(timer)) {
850 detach_timer(timer, 1);
851 ret = 1;
852 }
853out:
854 spin_unlock_irqrestore(&base->lock, flags);
855
856 return ret;
857}
David Howellse19dff12007-04-26 15:46:56 -0700858EXPORT_SYMBOL(try_to_del_timer_sync);
859
Rolf Eike Beer2aae4a12006-09-29 01:59:46 -0700860/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 * del_timer_sync - deactivate a timer and wait for the handler to finish.
862 * @timer: the timer to be deactivated
863 *
864 * This function only differs from del_timer() on SMP: besides deactivating
865 * the timer it also makes sure the handler has finished executing on other
866 * CPUs.
867 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800868 * Synchronization rules: Callers must prevent restarting of the timer,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 * otherwise this function is meaningless. It must not be called from
870 * interrupt contexts. The caller must not hold locks which would prevent
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700871 * completion of the timer's handler. The timer's handler must not call
872 * add_timer_on(). Upon exit the timer is not queued and the handler is
873 * not running on any CPU.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 *
875 * The function returns whether it has deactivated a pending timer or not.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 */
877int del_timer_sync(struct timer_list *timer)
878{
Johannes Berg6f2b9b92009-01-29 16:03:20 +0100879#ifdef CONFIG_LOCKDEP
880 unsigned long flags;
881
882 local_irq_save(flags);
883 lock_map_acquire(&timer->lockdep_map);
884 lock_map_release(&timer->lockdep_map);
885 local_irq_restore(flags);
886#endif
887
Oleg Nesterovfd450b72005-06-23 00:08:59 -0700888 for (;;) {
889 int ret = try_to_del_timer_sync(timer);
890 if (ret >= 0)
891 return ret;
Andrew Mortona0009652006-07-14 00:24:06 -0700892 cpu_relax();
Oleg Nesterovfd450b72005-06-23 00:08:59 -0700893 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894}
895EXPORT_SYMBOL(del_timer_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896#endif
897
Pavel Macheka6fa8e52008-01-30 13:30:00 +0100898static int cascade(struct tvec_base *base, struct tvec *tv, int index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899{
900 /* cascade all the timers from tv up one level */
Porpoise3439dd82006-06-23 02:05:56 -0700901 struct timer_list *timer, *tmp;
902 struct list_head tv_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903
Porpoise3439dd82006-06-23 02:05:56 -0700904 list_replace_init(tv->vec + index, &tv_list);
905
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 /*
Porpoise3439dd82006-06-23 02:05:56 -0700907 * We are removing _all_ timers from the list, so we
908 * don't have to detach them individually.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 */
Porpoise3439dd82006-06-23 02:05:56 -0700910 list_for_each_entry_safe(timer, tmp, &tv_list, entry) {
Venki Pallipadi6e453a62007-05-08 00:27:44 -0700911 BUG_ON(tbase_get_base(timer->base) != base);
Porpoise3439dd82006-06-23 02:05:56 -0700912 internal_add_timer(base, timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914
915 return index;
916}
917
Rolf Eike Beer2aae4a12006-09-29 01:59:46 -0700918#define INDEX(N) ((base->timer_jiffies >> (TVR_BITS + (N) * TVN_BITS)) & TVN_MASK)
919
920/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 * __run_timers - run all expired timers (if any) on this CPU.
922 * @base: the timer vector to be processed.
923 *
924 * This function cascades all vectors and executes all expired timer
925 * vectors.
926 */
Pavel Macheka6fa8e52008-01-30 13:30:00 +0100927static inline void __run_timers(struct tvec_base *base)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928{
929 struct timer_list *timer;
930
Oleg Nesterov3691c512006-03-31 02:30:30 -0800931 spin_lock_irq(&base->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 while (time_after_eq(jiffies, base->timer_jiffies)) {
Oleg Nesterov626ab0e2006-06-23 02:05:55 -0700933 struct list_head work_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 struct list_head *head = &work_list;
Thomas Gleixner68194572007-07-19 01:49:16 -0700935 int index = base->timer_jiffies & TVR_MASK;
Oleg Nesterov626ab0e2006-06-23 02:05:55 -0700936
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 /*
938 * Cascade timers:
939 */
940 if (!index &&
941 (!cascade(base, &base->tv2, INDEX(0))) &&
942 (!cascade(base, &base->tv3, INDEX(1))) &&
943 !cascade(base, &base->tv4, INDEX(2)))
944 cascade(base, &base->tv5, INDEX(3));
Oleg Nesterov626ab0e2006-06-23 02:05:55 -0700945 ++base->timer_jiffies;
946 list_replace_init(base->tv1.vec + index, &work_list);
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700947 while (!list_empty(head)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 void (*fn)(unsigned long);
949 unsigned long data;
950
Pavel Emelianovb5e61812007-05-08 00:30:19 -0700951 timer = list_first_entry(head, struct timer_list,entry);
Thomas Gleixner68194572007-07-19 01:49:16 -0700952 fn = timer->function;
953 data = timer->data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954
Ingo Molnar82f67cd2007-02-16 01:28:13 -0800955 timer_stats_account_timer(timer);
956
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 set_running_timer(base, timer);
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700958 detach_timer(timer, 1);
Johannes Berg6f2b9b92009-01-29 16:03:20 +0100959
Oleg Nesterov3691c512006-03-31 02:30:30 -0800960 spin_unlock_irq(&base->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 {
Jesper Juhlbe5b4fb2005-06-23 00:09:09 -0700962 int preempt_count = preempt_count();
Johannes Berg6f2b9b92009-01-29 16:03:20 +0100963
964#ifdef CONFIG_LOCKDEP
965 /*
966 * It is permissible to free the timer from
967 * inside the function that is called from
968 * it, this we need to take into account for
969 * lockdep too. To avoid bogus "held lock
970 * freed" warnings as well as problems when
971 * looking into timer->lockdep_map, make a
972 * copy and use that here.
973 */
974 struct lockdep_map lockdep_map =
975 timer->lockdep_map;
976#endif
977 /*
978 * Couple the lock chain with the lock chain at
979 * del_timer_sync() by acquiring the lock_map
980 * around the fn() call here and in
981 * del_timer_sync().
982 */
983 lock_map_acquire(&lockdep_map);
984
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 fn(data);
Johannes Berg6f2b9b92009-01-29 16:03:20 +0100986
987 lock_map_release(&lockdep_map);
988
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 if (preempt_count != preempt_count()) {
Pavel Machek4c9dc642008-01-30 13:30:00 +0100990 printk(KERN_ERR "huh, entered %p "
Jesper Juhlbe5b4fb2005-06-23 00:09:09 -0700991 "with preempt_count %08x, exited"
992 " with %08x?\n",
993 fn, preempt_count,
994 preempt_count());
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 BUG();
996 }
997 }
Oleg Nesterov3691c512006-03-31 02:30:30 -0800998 spin_lock_irq(&base->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 }
1000 }
1001 set_running_timer(base, NULL);
Oleg Nesterov3691c512006-03-31 02:30:30 -08001002 spin_unlock_irq(&base->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003}
1004
Russell Kingee9c5782008-04-20 13:59:33 +01001005#ifdef CONFIG_NO_HZ
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006/*
1007 * Find out when the next timer event is due to happen. This
1008 * is used on S/390 to stop all activity when a cpus is idle.
1009 * This functions needs to be called disabled.
1010 */
Pavel Macheka6fa8e52008-01-30 13:30:00 +01001011static unsigned long __next_timer_interrupt(struct tvec_base *base)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012{
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001013 unsigned long timer_jiffies = base->timer_jiffies;
Thomas Gleixnereaad0842007-05-29 23:47:39 +02001014 unsigned long expires = timer_jiffies + NEXT_TIMER_MAX_DELTA;
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001015 int index, slot, array, found = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016 struct timer_list *nte;
Pavel Macheka6fa8e52008-01-30 13:30:00 +01001017 struct tvec *varray[4];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018
1019 /* Look for timer events in tv1. */
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001020 index = slot = timer_jiffies & TVR_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 do {
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001022 list_for_each_entry(nte, base->tv1.vec + slot, entry) {
Thomas Gleixner68194572007-07-19 01:49:16 -07001023 if (tbase_get_deferrable(nte->base))
1024 continue;
Venki Pallipadi6e453a62007-05-08 00:27:44 -07001025
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001026 found = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 expires = nte->expires;
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001028 /* Look at the cascade bucket(s)? */
1029 if (!index || slot < index)
1030 goto cascade;
1031 return expires;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 }
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001033 slot = (slot + 1) & TVR_MASK;
1034 } while (slot != index);
1035
1036cascade:
1037 /* Calculate the next cascade event */
1038 if (index)
1039 timer_jiffies += TVR_SIZE - index;
1040 timer_jiffies >>= TVR_BITS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041
1042 /* Check tv2-tv5. */
1043 varray[0] = &base->tv2;
1044 varray[1] = &base->tv3;
1045 varray[2] = &base->tv4;
1046 varray[3] = &base->tv5;
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001047
1048 for (array = 0; array < 4; array++) {
Pavel Macheka6fa8e52008-01-30 13:30:00 +01001049 struct tvec *varp = varray[array];
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001050
1051 index = slot = timer_jiffies & TVN_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 do {
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001053 list_for_each_entry(nte, varp->vec + slot, entry) {
Jon Huntera0419882009-05-01 13:10:23 -07001054 if (tbase_get_deferrable(nte->base))
1055 continue;
1056
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001057 found = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 if (time_before(nte->expires, expires))
1059 expires = nte->expires;
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001060 }
1061 /*
1062 * Do we still search for the first timer or are
1063 * we looking up the cascade buckets ?
1064 */
1065 if (found) {
1066 /* Look at the cascade bucket(s)? */
1067 if (!index || slot < index)
1068 break;
1069 return expires;
1070 }
1071 slot = (slot + 1) & TVN_MASK;
1072 } while (slot != index);
1073
1074 if (index)
1075 timer_jiffies += TVN_SIZE - index;
1076 timer_jiffies >>= TVN_BITS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077 }
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001078 return expires;
1079}
1080
1081/*
1082 * Check, if the next hrtimer event is before the next timer wheel
1083 * event:
1084 */
1085static unsigned long cmp_next_hrtimer_event(unsigned long now,
1086 unsigned long expires)
1087{
1088 ktime_t hr_delta = hrtimer_get_next_event();
1089 struct timespec tsdelta;
Thomas Gleixner9501b6c2007-03-25 14:31:17 +02001090 unsigned long delta;
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001091
1092 if (hr_delta.tv64 == KTIME_MAX)
1093 return expires;
1094
Thomas Gleixner9501b6c2007-03-25 14:31:17 +02001095 /*
1096 * Expired timer available, let it expire in the next tick
1097 */
1098 if (hr_delta.tv64 <= 0)
1099 return now + 1;
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001100
1101 tsdelta = ktime_to_timespec(hr_delta);
Thomas Gleixner9501b6c2007-03-25 14:31:17 +02001102 delta = timespec_to_jiffies(&tsdelta);
Thomas Gleixnereaad0842007-05-29 23:47:39 +02001103
1104 /*
1105 * Limit the delta to the max value, which is checked in
1106 * tick_nohz_stop_sched_tick():
1107 */
1108 if (delta > NEXT_TIMER_MAX_DELTA)
1109 delta = NEXT_TIMER_MAX_DELTA;
1110
Thomas Gleixner9501b6c2007-03-25 14:31:17 +02001111 /*
1112 * Take rounding errors in to account and make sure, that it
1113 * expires in the next tick. Otherwise we go into an endless
1114 * ping pong due to tick_nohz_stop_sched_tick() retriggering
1115 * the timer softirq
1116 */
1117 if (delta < 1)
1118 delta = 1;
1119 now += delta;
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001120 if (time_before(now, expires))
1121 return now;
1122 return expires;
1123}
1124
1125/**
Li Zefan8dce39c2007-11-05 14:51:10 -08001126 * get_next_timer_interrupt - return the jiffy of the next pending timer
Randy Dunlap05fb6bf2007-02-28 20:12:13 -08001127 * @now: current time (in jiffies)
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001128 */
Thomas Gleixnerfd064b92007-02-16 01:27:47 -08001129unsigned long get_next_timer_interrupt(unsigned long now)
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001130{
Pavel Macheka6fa8e52008-01-30 13:30:00 +01001131 struct tvec_base *base = __get_cpu_var(tvec_bases);
Thomas Gleixnerfd064b92007-02-16 01:27:47 -08001132 unsigned long expires;
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001133
1134 spin_lock(&base->lock);
1135 expires = __next_timer_interrupt(base);
Oleg Nesterov3691c512006-03-31 02:30:30 -08001136 spin_unlock(&base->lock);
Tony Lindgren69239742006-03-06 15:42:45 -08001137
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001138 if (time_before_eq(expires, now))
1139 return now;
Zachary Amsden0662b712006-05-20 15:00:24 -07001140
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001141 return cmp_next_hrtimer_event(now, expires);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142}
1143#endif
1144
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145/*
Daniel Walker5b4db0c2007-10-18 03:06:11 -07001146 * Called from the timer interrupt handler to charge one tick to the current
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 * process. user_tick is 1 if the tick is user time, 0 for system.
1148 */
1149void update_process_times(int user_tick)
1150{
1151 struct task_struct *p = current;
1152 int cpu = smp_processor_id();
1153
1154 /* Note: this timer irq context must be accounted for as well. */
Paul Mackerrasfa13a5a2007-11-09 22:39:38 +01001155 account_process_tick(p, user_tick);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 run_local_timers();
1157 if (rcu_pending(cpu))
1158 rcu_check_callbacks(cpu, user_tick);
Peter Zijlstrab845b512008-08-08 21:47:09 +02001159 printk_tick();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 scheduler_tick();
Thomas Gleixner68194572007-07-19 01:49:16 -07001161 run_posix_cpu_timers(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162}
1163
1164/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 * This function runs timers and the timer-tq in bottom half context.
1166 */
1167static void run_timer_softirq(struct softirq_action *h)
1168{
Pavel Macheka6fa8e52008-01-30 13:30:00 +01001169 struct tvec_base *base = __get_cpu_var(tvec_bases);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170
Peter Zijlstra925d5192009-03-30 19:07:02 +02001171 perf_counter_do_pending();
1172
Peter Zijlstrad3d74452008-01-25 21:08:31 +01001173 hrtimer_run_pending();
Ingo Molnar82f67cd2007-02-16 01:28:13 -08001174
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 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 */
1182void run_local_timers(void)
1183{
Peter Zijlstrad3d74452008-01-25 21:08:31 +01001184 hrtimer_run_queues();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185 raise_softirq(TIMER_SOFTIRQ);
Ingo Molnar6687a972006-03-24 03:18:41 -08001186 softlockup_tick();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187}
1188
1189/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 * The 64-bit jiffies value is not atomic - you MUST NOT read it
1191 * without sampling the sequence number in xtime_lock.
1192 * jiffies is defined in the linker script...
1193 */
1194
Atsushi Nemoto3171a032006-09-29 02:00:32 -07001195void do_timer(unsigned long ticks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196{
Atsushi Nemoto3171a032006-09-29 02:00:32 -07001197 jiffies_64 += ticks;
Thomas Gleixnerdce48a82009-04-11 10:43:41 +02001198 update_wall_time();
1199 calc_global_load();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200}
1201
1202#ifdef __ARCH_WANT_SYS_ALARM
1203
1204/*
1205 * For backwards compatibility? This can be done in libc so Alpha
1206 * and all newer ports shouldn't need it.
1207 */
Heiko Carstens58fd3aa2009-01-14 14:14:03 +01001208SYSCALL_DEFINE1(alarm, unsigned int, seconds)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209{
Thomas Gleixnerc08b8a42006-03-25 03:06:33 -08001210 return alarm_setitimer(seconds);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211}
1212
1213#endif
1214
1215#ifndef __alpha__
1216
1217/*
1218 * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this
1219 * should be moved into arch/i386 instead?
1220 */
1221
1222/**
1223 * sys_getpid - return the thread group id of the current process
1224 *
1225 * Note, despite the name, this returns the tgid not the pid. The tgid and
1226 * the pid are identical unless CLONE_THREAD was specified on clone() in
1227 * which case the tgid is the same in all threads of the same group.
1228 *
1229 * This is SMP safe as current->tgid does not change.
1230 */
Heiko Carstens58fd3aa2009-01-14 14:14:03 +01001231SYSCALL_DEFINE0(getpid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232{
Pavel Emelyanovb4888932007-10-18 23:40:14 -07001233 return task_tgid_vnr(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234}
1235
1236/*
Kirill Korotaev6997a6f2006-08-13 23:24:23 -07001237 * Accessing ->real_parent is not SMP-safe, it could
1238 * change from under us. However, we can use a stale
1239 * value of ->real_parent under rcu_read_lock(), see
1240 * release_task()->call_rcu(delayed_put_task_struct).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 */
Heiko Carstensdbf040d2009-01-14 14:14:04 +01001242SYSCALL_DEFINE0(getppid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243{
1244 int pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245
Kirill Korotaev6997a6f2006-08-13 23:24:23 -07001246 rcu_read_lock();
Pavel Emelyanov6c5f3e72008-02-08 04:19:20 -08001247 pid = task_tgid_vnr(current->real_parent);
Kirill Korotaev6997a6f2006-08-13 23:24:23 -07001248 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 return pid;
1251}
1252
Heiko Carstensdbf040d2009-01-14 14:14:04 +01001253SYSCALL_DEFINE0(getuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254{
1255 /* Only we change this so SMP safe */
David Howells76aac0e2008-11-14 10:39:12 +11001256 return current_uid();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257}
1258
Heiko Carstensdbf040d2009-01-14 14:14:04 +01001259SYSCALL_DEFINE0(geteuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260{
1261 /* Only we change this so SMP safe */
David Howells76aac0e2008-11-14 10:39:12 +11001262 return current_euid();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263}
1264
Heiko Carstensdbf040d2009-01-14 14:14:04 +01001265SYSCALL_DEFINE0(getgid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266{
1267 /* Only we change this so SMP safe */
David Howells76aac0e2008-11-14 10:39:12 +11001268 return current_gid();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269}
1270
Heiko Carstensdbf040d2009-01-14 14:14:04 +01001271SYSCALL_DEFINE0(getegid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272{
1273 /* Only we change this so SMP safe */
David Howells76aac0e2008-11-14 10:39:12 +11001274 return current_egid();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275}
1276
1277#endif
1278
1279static void process_timeout(unsigned long __data)
1280{
Ingo Molnar36c8b582006-07-03 00:25:41 -07001281 wake_up_process((struct task_struct *)__data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282}
1283
1284/**
1285 * schedule_timeout - sleep until timeout
1286 * @timeout: timeout value in jiffies
1287 *
1288 * Make the current task sleep until @timeout jiffies have
1289 * elapsed. The routine will return immediately unless
1290 * the current task state has been set (see set_current_state()).
1291 *
1292 * You can set the task state as follows -
1293 *
1294 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1295 * pass before the routine returns. The routine will return 0
1296 *
1297 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1298 * delivered to the current task. In this case the remaining time
1299 * in jiffies will be returned, or 0 if the timer expired in time
1300 *
1301 * The current task state is guaranteed to be TASK_RUNNING when this
1302 * routine returns.
1303 *
1304 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1305 * the CPU away without a bound on the timeout. In this case the return
1306 * value will be %MAX_SCHEDULE_TIMEOUT.
1307 *
1308 * In all cases the return value is guaranteed to be non-negative.
1309 */
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -08001310signed long __sched schedule_timeout(signed long timeout)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311{
1312 struct timer_list timer;
1313 unsigned long expire;
1314
1315 switch (timeout)
1316 {
1317 case MAX_SCHEDULE_TIMEOUT:
1318 /*
1319 * These two special cases are useful to be comfortable
1320 * in the caller. Nothing more. We could take
1321 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1322 * but I' d like to return a valid offset (>=0) to allow
1323 * the caller to do everything it want with the retval.
1324 */
1325 schedule();
1326 goto out;
1327 default:
1328 /*
1329 * Another bit of PARANOID. Note that the retval will be
1330 * 0 since no piece of kernel is supposed to do a check
1331 * for a negative retval of schedule_timeout() (since it
1332 * should never happens anyway). You just have the printk()
1333 * that will tell you if something is gone wrong and where.
1334 */
Andrew Morton5b149bc2006-12-22 01:10:14 -08001335 if (timeout < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 printk(KERN_ERR "schedule_timeout: wrong timeout "
Andrew Morton5b149bc2006-12-22 01:10:14 -08001337 "value %lx\n", timeout);
1338 dump_stack();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 current->state = TASK_RUNNING;
1340 goto out;
1341 }
1342 }
1343
1344 expire = timeout + jiffies;
1345
Thomas Gleixnerc6f3a972008-04-30 00:55:03 -07001346 setup_timer_on_stack(&timer, process_timeout, (unsigned long)current);
Arun R Bharadwaj597d0272009-04-16 12:13:26 +05301347 __mod_timer(&timer, expire, false, TIMER_NOT_PINNED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 schedule();
1349 del_singleshot_timer_sync(&timer);
1350
Thomas Gleixnerc6f3a972008-04-30 00:55:03 -07001351 /* Remove the timer from the object tracker */
1352 destroy_timer_on_stack(&timer);
1353
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 timeout = expire - jiffies;
1355
1356 out:
1357 return timeout < 0 ? 0 : timeout;
1358}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359EXPORT_SYMBOL(schedule_timeout);
1360
Andrew Morton8a1c1752005-09-13 01:25:15 -07001361/*
1362 * We can use __set_current_state() here because schedule_timeout() calls
1363 * schedule() unconditionally.
1364 */
Nishanth Aravamudan64ed93a2005-09-10 00:27:21 -07001365signed long __sched schedule_timeout_interruptible(signed long timeout)
1366{
Andrew Mortona5a0d522005-10-30 15:01:42 -08001367 __set_current_state(TASK_INTERRUPTIBLE);
1368 return schedule_timeout(timeout);
Nishanth Aravamudan64ed93a2005-09-10 00:27:21 -07001369}
1370EXPORT_SYMBOL(schedule_timeout_interruptible);
1371
Matthew Wilcox294d5cc2007-12-06 11:59:46 -05001372signed long __sched schedule_timeout_killable(signed long timeout)
1373{
1374 __set_current_state(TASK_KILLABLE);
1375 return schedule_timeout(timeout);
1376}
1377EXPORT_SYMBOL(schedule_timeout_killable);
1378
Nishanth Aravamudan64ed93a2005-09-10 00:27:21 -07001379signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1380{
Andrew Mortona5a0d522005-10-30 15:01:42 -08001381 __set_current_state(TASK_UNINTERRUPTIBLE);
1382 return schedule_timeout(timeout);
Nishanth Aravamudan64ed93a2005-09-10 00:27:21 -07001383}
1384EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1385
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386/* Thread ID - the internal kernel "pid" */
Heiko Carstens58fd3aa2009-01-14 14:14:03 +01001387SYSCALL_DEFINE0(gettid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388{
Pavel Emelyanovb4888932007-10-18 23:40:14 -07001389 return task_pid_vnr(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390}
1391
Rolf Eike Beer2aae4a12006-09-29 01:59:46 -07001392/**
Kyle McMartind4d23ad2007-02-10 01:46:00 -08001393 * do_sysinfo - fill in sysinfo struct
Rolf Eike Beer2aae4a12006-09-29 01:59:46 -07001394 * @info: pointer to buffer to fill
Thomas Gleixner68194572007-07-19 01:49:16 -07001395 */
Kyle McMartind4d23ad2007-02-10 01:46:00 -08001396int do_sysinfo(struct sysinfo *info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 unsigned long mem_total, sav_total;
1399 unsigned int mem_unit, bitcount;
Thomas Gleixner2d024942009-05-02 20:08:52 +02001400 struct timespec tp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401
Kyle McMartind4d23ad2007-02-10 01:46:00 -08001402 memset(info, 0, sizeof(struct sysinfo));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403
Thomas Gleixner2d024942009-05-02 20:08:52 +02001404 ktime_get_ts(&tp);
1405 monotonic_to_bootbased(&tp);
1406 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407
Thomas Gleixner2d024942009-05-02 20:08:52 +02001408 get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409
Thomas Gleixner2d024942009-05-02 20:08:52 +02001410 info->procs = nr_threads;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411
Kyle McMartind4d23ad2007-02-10 01:46:00 -08001412 si_meminfo(info);
1413 si_swapinfo(info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414
1415 /*
1416 * If the sum of all the available memory (i.e. ram + swap)
1417 * is less than can be stored in a 32 bit unsigned long then
1418 * we can be binary compatible with 2.2.x kernels. If not,
1419 * well, in that case 2.2.x was broken anyways...
1420 *
1421 * -Erik Andersen <andersee@debian.org>
1422 */
1423
Kyle McMartind4d23ad2007-02-10 01:46:00 -08001424 mem_total = info->totalram + info->totalswap;
1425 if (mem_total < info->totalram || mem_total < info->totalswap)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 goto out;
1427 bitcount = 0;
Kyle McMartind4d23ad2007-02-10 01:46:00 -08001428 mem_unit = info->mem_unit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 while (mem_unit > 1) {
1430 bitcount++;
1431 mem_unit >>= 1;
1432 sav_total = mem_total;
1433 mem_total <<= 1;
1434 if (mem_total < sav_total)
1435 goto out;
1436 }
1437
1438 /*
1439 * If mem_total did not overflow, multiply all memory values by
Kyle McMartind4d23ad2007-02-10 01:46:00 -08001440 * info->mem_unit and set it to 1. This leaves things compatible
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1442 * kernels...
1443 */
1444
Kyle McMartind4d23ad2007-02-10 01:46:00 -08001445 info->mem_unit = 1;
1446 info->totalram <<= bitcount;
1447 info->freeram <<= bitcount;
1448 info->sharedram <<= bitcount;
1449 info->bufferram <<= bitcount;
1450 info->totalswap <<= bitcount;
1451 info->freeswap <<= bitcount;
1452 info->totalhigh <<= bitcount;
1453 info->freehigh <<= bitcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454
Kyle McMartind4d23ad2007-02-10 01:46:00 -08001455out:
1456 return 0;
1457}
1458
Heiko Carstens1e7bfb22009-01-14 14:14:29 +01001459SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
Kyle McMartind4d23ad2007-02-10 01:46:00 -08001460{
1461 struct sysinfo val;
1462
1463 do_sysinfo(&val);
1464
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
1466 return -EFAULT;
1467
1468 return 0;
1469}
1470
Adrian Bunkb4be6252007-12-18 18:05:58 +01001471static int __cpuinit init_timers_cpu(int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472{
1473 int j;
Pavel Macheka6fa8e52008-01-30 13:30:00 +01001474 struct tvec_base *base;
Adrian Bunkb4be6252007-12-18 18:05:58 +01001475 static char __cpuinitdata tvec_base_done[NR_CPUS];
Oleg Nesterov55c888d2005-06-23 00:08:56 -07001476
Andrew Mortonba6edfc2006-04-10 22:53:58 -07001477 if (!tvec_base_done[cpu]) {
Jan Beulicha4a61982006-03-24 03:15:54 -08001478 static char boot_done;
1479
Jan Beulicha4a61982006-03-24 03:15:54 -08001480 if (boot_done) {
Andrew Mortonba6edfc2006-04-10 22:53:58 -07001481 /*
1482 * The APs use this path later in boot
1483 */
Christoph Lameter94f60302007-07-17 04:03:29 -07001484 base = kmalloc_node(sizeof(*base),
1485 GFP_KERNEL | __GFP_ZERO,
Jan Beulicha4a61982006-03-24 03:15:54 -08001486 cpu_to_node(cpu));
1487 if (!base)
1488 return -ENOMEM;
Venki Pallipadi6e453a62007-05-08 00:27:44 -07001489
1490 /* Make sure that tvec_base is 2 byte aligned */
1491 if (tbase_get_deferrable(base)) {
1492 WARN_ON(1);
1493 kfree(base);
1494 return -ENOMEM;
1495 }
Andrew Mortonba6edfc2006-04-10 22:53:58 -07001496 per_cpu(tvec_bases, cpu) = base;
Jan Beulicha4a61982006-03-24 03:15:54 -08001497 } else {
Andrew Mortonba6edfc2006-04-10 22:53:58 -07001498 /*
1499 * This is for the boot CPU - we use compile-time
1500 * static initialisation because per-cpu memory isn't
1501 * ready yet and because the memory allocators are not
1502 * initialised either.
1503 */
Jan Beulicha4a61982006-03-24 03:15:54 -08001504 boot_done = 1;
Andrew Mortonba6edfc2006-04-10 22:53:58 -07001505 base = &boot_tvec_bases;
Jan Beulicha4a61982006-03-24 03:15:54 -08001506 }
Andrew Mortonba6edfc2006-04-10 22:53:58 -07001507 tvec_base_done[cpu] = 1;
1508 } else {
1509 base = per_cpu(tvec_bases, cpu);
Jan Beulicha4a61982006-03-24 03:15:54 -08001510 }
Andrew Mortonba6edfc2006-04-10 22:53:58 -07001511
Oleg Nesterov3691c512006-03-31 02:30:30 -08001512 spin_lock_init(&base->lock);
Ingo Molnard730e882006-07-03 00:25:10 -07001513
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514 for (j = 0; j < TVN_SIZE; j++) {
1515 INIT_LIST_HEAD(base->tv5.vec + j);
1516 INIT_LIST_HEAD(base->tv4.vec + j);
1517 INIT_LIST_HEAD(base->tv3.vec + j);
1518 INIT_LIST_HEAD(base->tv2.vec + j);
1519 }
1520 for (j = 0; j < TVR_SIZE; j++)
1521 INIT_LIST_HEAD(base->tv1.vec + j);
1522
1523 base->timer_jiffies = jiffies;
Jan Beulicha4a61982006-03-24 03:15:54 -08001524 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525}
1526
1527#ifdef CONFIG_HOTPLUG_CPU
Pavel Macheka6fa8e52008-01-30 13:30:00 +01001528static void migrate_timer_list(struct tvec_base *new_base, struct list_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529{
1530 struct timer_list *timer;
1531
1532 while (!list_empty(head)) {
Pavel Emelianovb5e61812007-05-08 00:30:19 -07001533 timer = list_first_entry(head, struct timer_list, entry);
Oleg Nesterov55c888d2005-06-23 00:08:56 -07001534 detach_timer(timer, 0);
Venki Pallipadi6e453a62007-05-08 00:27:44 -07001535 timer_set_base(timer, new_base);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536 internal_add_timer(new_base, timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538}
1539
Randy Dunlap48ccf3d2008-01-21 17:18:25 -08001540static void __cpuinit migrate_timers(int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541{
Pavel Macheka6fa8e52008-01-30 13:30:00 +01001542 struct tvec_base *old_base;
1543 struct tvec_base *new_base;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 int i;
1545
1546 BUG_ON(cpu_online(cpu));
Jan Beulicha4a61982006-03-24 03:15:54 -08001547 old_base = per_cpu(tvec_bases, cpu);
1548 new_base = get_cpu_var(tvec_bases);
Oleg Nesterovd82f0b02008-08-20 16:46:04 -07001549 /*
1550 * The caller is globally serialized and nobody else
1551 * takes two locks at once, deadlock is not possible.
1552 */
1553 spin_lock_irq(&new_base->lock);
Oleg Nesterov0d180402008-04-04 20:54:10 +02001554 spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555
Oleg Nesterov3691c512006-03-31 02:30:30 -08001556 BUG_ON(old_base->running_timer);
1557
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 for (i = 0; i < TVR_SIZE; i++)
Oleg Nesterov55c888d2005-06-23 00:08:56 -07001559 migrate_timer_list(new_base, old_base->tv1.vec + i);
1560 for (i = 0; i < TVN_SIZE; i++) {
1561 migrate_timer_list(new_base, old_base->tv2.vec + i);
1562 migrate_timer_list(new_base, old_base->tv3.vec + i);
1563 migrate_timer_list(new_base, old_base->tv4.vec + i);
1564 migrate_timer_list(new_base, old_base->tv5.vec + i);
1565 }
1566
Oleg Nesterov0d180402008-04-04 20:54:10 +02001567 spin_unlock(&old_base->lock);
Oleg Nesterovd82f0b02008-08-20 16:46:04 -07001568 spin_unlock_irq(&new_base->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 put_cpu_var(tvec_bases);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570}
1571#endif /* CONFIG_HOTPLUG_CPU */
1572
Chandra Seetharaman8c78f302006-07-30 03:03:35 -07001573static int __cpuinit timer_cpu_notify(struct notifier_block *self,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574 unsigned long action, void *hcpu)
1575{
1576 long cpu = (long)hcpu;
1577 switch(action) {
1578 case CPU_UP_PREPARE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07001579 case CPU_UP_PREPARE_FROZEN:
Jan Beulicha4a61982006-03-24 03:15:54 -08001580 if (init_timers_cpu(cpu) < 0)
1581 return NOTIFY_BAD;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 break;
1583#ifdef CONFIG_HOTPLUG_CPU
1584 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07001585 case CPU_DEAD_FROZEN:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586 migrate_timers(cpu);
1587 break;
1588#endif
1589 default:
1590 break;
1591 }
1592 return NOTIFY_OK;
1593}
1594
Chandra Seetharaman8c78f302006-07-30 03:03:35 -07001595static struct notifier_block __cpuinitdata timers_nb = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 .notifier_call = timer_cpu_notify,
1597};
1598
1599
1600void __init init_timers(void)
1601{
Akinobu Mita07dccf32006-09-29 02:00:22 -07001602 int err = timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603 (void *)(long)smp_processor_id());
Akinobu Mita07dccf32006-09-29 02:00:22 -07001604
Ingo Molnar82f67cd2007-02-16 01:28:13 -08001605 init_timer_stats();
1606
Akinobu Mita07dccf32006-09-29 02:00:22 -07001607 BUG_ON(err == NOTIFY_BAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 register_cpu_notifier(&timers_nb);
Carlos R. Mafra962cf362008-05-15 11:15:37 -03001609 open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610}
1611
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612/**
1613 * msleep - sleep safely even with waitqueue interruptions
1614 * @msecs: Time in milliseconds to sleep for
1615 */
1616void msleep(unsigned int msecs)
1617{
1618 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1619
Nishanth Aravamudan75bcc8c2005-09-10 00:27:24 -07001620 while (timeout)
1621 timeout = schedule_timeout_uninterruptible(timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622}
1623
1624EXPORT_SYMBOL(msleep);
1625
1626/**
Domen Puncer96ec3ef2005-06-25 14:58:43 -07001627 * msleep_interruptible - sleep waiting for signals
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628 * @msecs: Time in milliseconds to sleep for
1629 */
1630unsigned long msleep_interruptible(unsigned int msecs)
1631{
1632 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1633
Nishanth Aravamudan75bcc8c2005-09-10 00:27:24 -07001634 while (timeout && !signal_pending(current))
1635 timeout = schedule_timeout_interruptible(timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 return jiffies_to_msecs(timeout);
1637}
1638
1639EXPORT_SYMBOL(msleep_interruptible);