blob: 13dd64fe143db216a31e750d66bb29d07c932ba3 [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070040
41#include <asm/uaccess.h>
42#include <asm/unistd.h>
43#include <asm/div64.h>
44#include <asm/timex.h>
45#include <asm/io.h>
46
Thomas Gleixnerecea8d12005-10-30 15:03:00 -080047u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
48
49EXPORT_SYMBOL(jiffies_64);
50
Linus Torvalds1da177e2005-04-16 15:20:36 -070051/*
52 * per-CPU timer vector definitions:
53 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070054#define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6)
55#define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8)
56#define TVN_SIZE (1 << TVN_BITS)
57#define TVR_SIZE (1 << TVR_BITS)
58#define TVN_MASK (TVN_SIZE - 1)
59#define TVR_MASK (TVR_SIZE - 1)
60
Pavel Macheka6fa8e52008-01-30 13:30:00 +010061struct tvec {
Linus Torvalds1da177e2005-04-16 15:20:36 -070062 struct list_head vec[TVN_SIZE];
Pavel Macheka6fa8e52008-01-30 13:30:00 +010063};
Linus Torvalds1da177e2005-04-16 15:20:36 -070064
Pavel Macheka6fa8e52008-01-30 13:30:00 +010065struct tvec_root {
Linus Torvalds1da177e2005-04-16 15:20:36 -070066 struct list_head vec[TVR_SIZE];
Pavel Macheka6fa8e52008-01-30 13:30:00 +010067};
Linus Torvalds1da177e2005-04-16 15:20:36 -070068
Pavel Macheka6fa8e52008-01-30 13:30:00 +010069struct tvec_base {
Oleg Nesterov3691c512006-03-31 02:30:30 -080070 spinlock_t lock;
71 struct timer_list *running_timer;
Linus Torvalds1da177e2005-04-16 15:20:36 -070072 unsigned long timer_jiffies;
Pavel Macheka6fa8e52008-01-30 13:30:00 +010073 struct tvec_root tv1;
74 struct tvec tv2;
75 struct tvec tv3;
76 struct tvec tv4;
77 struct tvec tv5;
Venki Pallipadi6e453a62007-05-08 00:27:44 -070078} ____cacheline_aligned;
Linus Torvalds1da177e2005-04-16 15:20:36 -070079
Pavel Macheka6fa8e52008-01-30 13:30:00 +010080struct tvec_base boot_tvec_bases;
Oleg Nesterov3691c512006-03-31 02:30:30 -080081EXPORT_SYMBOL(boot_tvec_bases);
Pavel Macheka6fa8e52008-01-30 13:30:00 +010082static DEFINE_PER_CPU(struct tvec_base *, tvec_bases) = &boot_tvec_bases;
Linus Torvalds1da177e2005-04-16 15:20:36 -070083
Venki Pallipadi6e453a62007-05-08 00:27:44 -070084/*
Pavel Macheka6fa8e52008-01-30 13:30:00 +010085 * Note that all tvec_bases are 2 byte aligned and lower bit of
Venki Pallipadi6e453a62007-05-08 00:27:44 -070086 * base in timer_list is guaranteed to be zero. Use the LSB for
87 * the new flag to indicate whether the timer is deferrable
88 */
89#define TBASE_DEFERRABLE_FLAG (0x1)
90
91/* Functions below help us manage 'deferrable' flag */
Pavel Macheka6fa8e52008-01-30 13:30:00 +010092static inline unsigned int tbase_get_deferrable(struct tvec_base *base)
Venki Pallipadi6e453a62007-05-08 00:27:44 -070093{
akpm@linux-foundation.orge9910842007-05-10 03:16:01 -070094 return ((unsigned int)(unsigned long)base & TBASE_DEFERRABLE_FLAG);
Venki Pallipadi6e453a62007-05-08 00:27:44 -070095}
96
Pavel Macheka6fa8e52008-01-30 13:30:00 +010097static inline struct tvec_base *tbase_get_base(struct tvec_base *base)
Venki Pallipadi6e453a62007-05-08 00:27:44 -070098{
Pavel Macheka6fa8e52008-01-30 13:30:00 +010099 return ((struct tvec_base *)((unsigned long)base & ~TBASE_DEFERRABLE_FLAG));
Venki Pallipadi6e453a62007-05-08 00:27:44 -0700100}
101
102static inline void timer_set_deferrable(struct timer_list *timer)
103{
Pavel Macheka6fa8e52008-01-30 13:30:00 +0100104 timer->base = ((struct tvec_base *)((unsigned long)(timer->base) |
Thomas Gleixner68194572007-07-19 01:49:16 -0700105 TBASE_DEFERRABLE_FLAG));
Venki Pallipadi6e453a62007-05-08 00:27:44 -0700106}
107
108static inline void
Pavel Macheka6fa8e52008-01-30 13:30:00 +0100109timer_set_base(struct timer_list *timer, struct tvec_base *new_base)
Venki Pallipadi6e453a62007-05-08 00:27:44 -0700110{
Pavel Macheka6fa8e52008-01-30 13:30:00 +0100111 timer->base = (struct tvec_base *)((unsigned long)(new_base) |
Thomas Gleixner68194572007-07-19 01:49:16 -0700112 tbase_get_deferrable(timer->base));
Venki Pallipadi6e453a62007-05-08 00:27:44 -0700113}
114
Alan Stern9c133c42008-11-06 08:42:48 +0100115static unsigned long round_jiffies_common(unsigned long j, int cpu,
116 bool force_up)
117{
118 int rem;
119 unsigned long original = j;
120
121 /*
122 * We don't want all cpus firing their timers at once hitting the
123 * same lock or cachelines, so we skew each extra cpu with an extra
124 * 3 jiffies. This 3 jiffies came originally from the mm/ code which
125 * already did this.
126 * The skew is done by adding 3*cpunr, then round, then subtract this
127 * extra offset again.
128 */
129 j += cpu * 3;
130
131 rem = j % HZ;
132
133 /*
134 * If the target jiffie is just after a whole second (which can happen
135 * due to delays of the timer irq, long irq off times etc etc) then
136 * we should round down to the whole second, not up. Use 1/4th second
137 * as cutoff for this rounding as an extreme upper bound for this.
138 * But never round down if @force_up is set.
139 */
140 if (rem < HZ/4 && !force_up) /* round down */
141 j = j - rem;
142 else /* round up */
143 j = j - rem + HZ;
144
145 /* now that we have rounded, subtract the extra skew again */
146 j -= cpu * 3;
147
148 if (j <= jiffies) /* rounding ate our timeout entirely; */
149 return original;
150 return j;
151}
152
Arjan van de Ven4c36a5d2006-12-10 02:21:24 -0800153/**
154 * __round_jiffies - function to round jiffies to a full second
155 * @j: the time in (absolute) jiffies that should be rounded
156 * @cpu: the processor number on which the timeout will happen
157 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800158 * __round_jiffies() rounds an absolute time in the future (in jiffies)
Arjan van de Ven4c36a5d2006-12-10 02:21:24 -0800159 * up or down to (approximately) full seconds. This is useful for timers
160 * for which the exact time they fire does not matter too much, as long as
161 * they fire approximately every X seconds.
162 *
163 * By rounding these timers to whole seconds, all such timers will fire
164 * at the same time, rather than at various times spread out. The goal
165 * of this is to have the CPU wake up less, which saves power.
166 *
167 * The exact rounding is skewed for each processor to avoid all
168 * processors firing at the exact same time, which could lead
169 * to lock contention or spurious cache line bouncing.
170 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800171 * The return value is the rounded version of the @j parameter.
Arjan van de Ven4c36a5d2006-12-10 02:21:24 -0800172 */
173unsigned long __round_jiffies(unsigned long j, int cpu)
174{
Alan Stern9c133c42008-11-06 08:42:48 +0100175 return round_jiffies_common(j, cpu, false);
Arjan van de Ven4c36a5d2006-12-10 02:21:24 -0800176}
177EXPORT_SYMBOL_GPL(__round_jiffies);
178
179/**
180 * __round_jiffies_relative - function to round jiffies to a full second
181 * @j: the time in (relative) jiffies that should be rounded
182 * @cpu: the processor number on which the timeout will happen
183 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800184 * __round_jiffies_relative() rounds a time delta in the future (in jiffies)
Arjan van de Ven4c36a5d2006-12-10 02:21:24 -0800185 * up or down to (approximately) full seconds. This is useful for timers
186 * for which the exact time they fire does not matter too much, as long as
187 * they fire approximately every X seconds.
188 *
189 * By rounding these timers to whole seconds, all such timers will fire
190 * at the same time, rather than at various times spread out. The goal
191 * of this is to have the CPU wake up less, which saves power.
192 *
193 * The exact rounding is skewed for each processor to avoid all
194 * processors firing at the exact same time, which could lead
195 * to lock contention or spurious cache line bouncing.
196 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800197 * The return value is the rounded version of the @j parameter.
Arjan van de Ven4c36a5d2006-12-10 02:21:24 -0800198 */
199unsigned long __round_jiffies_relative(unsigned long j, int cpu)
200{
Alan Stern9c133c42008-11-06 08:42:48 +0100201 unsigned long j0 = jiffies;
202
203 /* Use j0 because jiffies might change while we run */
204 return round_jiffies_common(j + j0, cpu, false) - j0;
Arjan van de Ven4c36a5d2006-12-10 02:21:24 -0800205}
206EXPORT_SYMBOL_GPL(__round_jiffies_relative);
207
208/**
209 * round_jiffies - function to round jiffies to a full second
210 * @j: the time in (absolute) jiffies that should be rounded
211 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800212 * round_jiffies() rounds an absolute time in the future (in jiffies)
Arjan van de Ven4c36a5d2006-12-10 02:21:24 -0800213 * up or down to (approximately) full seconds. This is useful for timers
214 * for which the exact time they fire does not matter too much, as long as
215 * they fire approximately every X seconds.
216 *
217 * By rounding these timers to whole seconds, all such timers will fire
218 * at the same time, rather than at various times spread out. The goal
219 * of this is to have the CPU wake up less, which saves power.
220 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800221 * The return value is the rounded version of the @j parameter.
Arjan van de Ven4c36a5d2006-12-10 02:21:24 -0800222 */
223unsigned long round_jiffies(unsigned long j)
224{
Alan Stern9c133c42008-11-06 08:42:48 +0100225 return round_jiffies_common(j, raw_smp_processor_id(), false);
Arjan van de Ven4c36a5d2006-12-10 02:21:24 -0800226}
227EXPORT_SYMBOL_GPL(round_jiffies);
228
229/**
230 * round_jiffies_relative - function to round jiffies to a full second
231 * @j: the time in (relative) jiffies that should be rounded
232 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800233 * round_jiffies_relative() rounds a time delta in the future (in jiffies)
Arjan van de Ven4c36a5d2006-12-10 02:21:24 -0800234 * up or down to (approximately) full seconds. This is useful for timers
235 * for which the exact time they fire does not matter too much, as long as
236 * they fire approximately every X seconds.
237 *
238 * By rounding these timers to whole seconds, all such timers will fire
239 * at the same time, rather than at various times spread out. The goal
240 * of this is to have the CPU wake up less, which saves power.
241 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800242 * The return value is the rounded version of the @j parameter.
Arjan van de Ven4c36a5d2006-12-10 02:21:24 -0800243 */
244unsigned long round_jiffies_relative(unsigned long j)
245{
246 return __round_jiffies_relative(j, raw_smp_processor_id());
247}
248EXPORT_SYMBOL_GPL(round_jiffies_relative);
249
Alan Stern9c133c42008-11-06 08:42:48 +0100250/**
251 * __round_jiffies_up - function to round jiffies up to a full second
252 * @j: the time in (absolute) jiffies that should be rounded
253 * @cpu: the processor number on which the timeout will happen
254 *
255 * This is the same as __round_jiffies() except that it will never
256 * round down. This is useful for timeouts for which the exact time
257 * of firing does not matter too much, as long as they don't fire too
258 * early.
259 */
260unsigned long __round_jiffies_up(unsigned long j, int cpu)
261{
262 return round_jiffies_common(j, cpu, true);
263}
264EXPORT_SYMBOL_GPL(__round_jiffies_up);
265
266/**
267 * __round_jiffies_up_relative - function to round jiffies up to a full second
268 * @j: the time in (relative) jiffies that should be rounded
269 * @cpu: the processor number on which the timeout will happen
270 *
271 * This is the same as __round_jiffies_relative() except that it will never
272 * round down. This is useful for timeouts for which the exact time
273 * of firing does not matter too much, as long as they don't fire too
274 * early.
275 */
276unsigned long __round_jiffies_up_relative(unsigned long j, int cpu)
277{
278 unsigned long j0 = jiffies;
279
280 /* Use j0 because jiffies might change while we run */
281 return round_jiffies_common(j + j0, cpu, true) - j0;
282}
283EXPORT_SYMBOL_GPL(__round_jiffies_up_relative);
284
285/**
286 * round_jiffies_up - function to round jiffies up to a full second
287 * @j: the time in (absolute) jiffies that should be rounded
288 *
289 * This is the same as round_jiffies() except that it will never
290 * round down. This is useful for timeouts for which the exact time
291 * of firing does not matter too much, as long as they don't fire too
292 * early.
293 */
294unsigned long round_jiffies_up(unsigned long j)
295{
296 return round_jiffies_common(j, raw_smp_processor_id(), true);
297}
298EXPORT_SYMBOL_GPL(round_jiffies_up);
299
300/**
301 * round_jiffies_up_relative - function to round jiffies up to a full second
302 * @j: the time in (relative) jiffies that should be rounded
303 *
304 * This is the same as round_jiffies_relative() except that it will never
305 * round down. This is useful for timeouts for which the exact time
306 * of firing does not matter too much, as long as they don't fire too
307 * early.
308 */
309unsigned long round_jiffies_up_relative(unsigned long j)
310{
311 return __round_jiffies_up_relative(j, raw_smp_processor_id());
312}
313EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
314
Arjan van de Ven4c36a5d2006-12-10 02:21:24 -0800315
Pavel Macheka6fa8e52008-01-30 13:30:00 +0100316static inline void set_running_timer(struct tvec_base *base,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 struct timer_list *timer)
318{
319#ifdef CONFIG_SMP
Oleg Nesterov3691c512006-03-31 02:30:30 -0800320 base->running_timer = timer;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321#endif
322}
323
Pavel Macheka6fa8e52008-01-30 13:30:00 +0100324static void internal_add_timer(struct tvec_base *base, struct timer_list *timer)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325{
326 unsigned long expires = timer->expires;
327 unsigned long idx = expires - base->timer_jiffies;
328 struct list_head *vec;
329
330 if (idx < TVR_SIZE) {
331 int i = expires & TVR_MASK;
332 vec = base->tv1.vec + i;
333 } else if (idx < 1 << (TVR_BITS + TVN_BITS)) {
334 int i = (expires >> TVR_BITS) & TVN_MASK;
335 vec = base->tv2.vec + i;
336 } else if (idx < 1 << (TVR_BITS + 2 * TVN_BITS)) {
337 int i = (expires >> (TVR_BITS + TVN_BITS)) & TVN_MASK;
338 vec = base->tv3.vec + i;
339 } else if (idx < 1 << (TVR_BITS + 3 * TVN_BITS)) {
340 int i = (expires >> (TVR_BITS + 2 * TVN_BITS)) & TVN_MASK;
341 vec = base->tv4.vec + i;
342 } else if ((signed long) idx < 0) {
343 /*
344 * Can happen if you add a timer with expires == jiffies,
345 * or you set a timer to go off in the past
346 */
347 vec = base->tv1.vec + (base->timer_jiffies & TVR_MASK);
348 } else {
349 int i;
350 /* If the timeout is larger than 0xffffffff on 64-bit
351 * architectures then we use the maximum timeout:
352 */
353 if (idx > 0xffffffffUL) {
354 idx = 0xffffffffUL;
355 expires = idx + base->timer_jiffies;
356 }
357 i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK;
358 vec = base->tv5.vec + i;
359 }
360 /*
361 * Timers are FIFO:
362 */
363 list_add_tail(&timer->entry, vec);
364}
365
Ingo Molnar82f67cd2007-02-16 01:28:13 -0800366#ifdef CONFIG_TIMER_STATS
367void __timer_stats_timer_set_start_info(struct timer_list *timer, void *addr)
368{
369 if (timer->start_site)
370 return;
371
372 timer->start_site = addr;
373 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
374 timer->start_pid = current->pid;
375}
Venki Pallipadic5c061b82007-07-15 23:40:30 -0700376
377static void timer_stats_account_timer(struct timer_list *timer)
378{
379 unsigned int flag = 0;
380
381 if (unlikely(tbase_get_deferrable(timer->base)))
382 flag |= TIMER_STATS_FLAG_DEFERRABLE;
383
384 timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
385 timer->function, timer->start_comm, flag);
386}
387
388#else
389static void timer_stats_account_timer(struct timer_list *timer) {}
Ingo Molnar82f67cd2007-02-16 01:28:13 -0800390#endif
391
Thomas Gleixnerc6f3a972008-04-30 00:55:03 -0700392#ifdef CONFIG_DEBUG_OBJECTS_TIMERS
393
394static struct debug_obj_descr timer_debug_descr;
395
396/*
397 * fixup_init is called when:
398 * - an active object is initialized
399 */
400static int timer_fixup_init(void *addr, enum debug_obj_state state)
401{
402 struct timer_list *timer = addr;
403
404 switch (state) {
405 case ODEBUG_STATE_ACTIVE:
406 del_timer_sync(timer);
407 debug_object_init(timer, &timer_debug_descr);
408 return 1;
409 default:
410 return 0;
411 }
412}
413
414/*
415 * fixup_activate is called when:
416 * - an active object is activated
417 * - an unknown object is activated (might be a statically initialized object)
418 */
419static int timer_fixup_activate(void *addr, enum debug_obj_state state)
420{
421 struct timer_list *timer = addr;
422
423 switch (state) {
424
425 case ODEBUG_STATE_NOTAVAILABLE:
426 /*
427 * This is not really a fixup. The timer was
428 * statically initialized. We just make sure that it
429 * is tracked in the object tracker.
430 */
431 if (timer->entry.next == NULL &&
432 timer->entry.prev == TIMER_ENTRY_STATIC) {
433 debug_object_init(timer, &timer_debug_descr);
434 debug_object_activate(timer, &timer_debug_descr);
435 return 0;
436 } else {
437 WARN_ON_ONCE(1);
438 }
439 return 0;
440
441 case ODEBUG_STATE_ACTIVE:
442 WARN_ON(1);
443
444 default:
445 return 0;
446 }
447}
448
449/*
450 * fixup_free is called when:
451 * - an active object is freed
452 */
453static int timer_fixup_free(void *addr, enum debug_obj_state state)
454{
455 struct timer_list *timer = addr;
456
457 switch (state) {
458 case ODEBUG_STATE_ACTIVE:
459 del_timer_sync(timer);
460 debug_object_free(timer, &timer_debug_descr);
461 return 1;
462 default:
463 return 0;
464 }
465}
466
467static struct debug_obj_descr timer_debug_descr = {
468 .name = "timer_list",
469 .fixup_init = timer_fixup_init,
470 .fixup_activate = timer_fixup_activate,
471 .fixup_free = timer_fixup_free,
472};
473
474static inline void debug_timer_init(struct timer_list *timer)
475{
476 debug_object_init(timer, &timer_debug_descr);
477}
478
479static inline void debug_timer_activate(struct timer_list *timer)
480{
481 debug_object_activate(timer, &timer_debug_descr);
482}
483
484static inline void debug_timer_deactivate(struct timer_list *timer)
485{
486 debug_object_deactivate(timer, &timer_debug_descr);
487}
488
489static inline void debug_timer_free(struct timer_list *timer)
490{
491 debug_object_free(timer, &timer_debug_descr);
492}
493
494static void __init_timer(struct timer_list *timer);
495
496void init_timer_on_stack(struct timer_list *timer)
497{
498 debug_object_init_on_stack(timer, &timer_debug_descr);
499 __init_timer(timer);
500}
501EXPORT_SYMBOL_GPL(init_timer_on_stack);
502
503void destroy_timer_on_stack(struct timer_list *timer)
504{
505 debug_object_free(timer, &timer_debug_descr);
506}
507EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
508
509#else
510static inline void debug_timer_init(struct timer_list *timer) { }
511static inline void debug_timer_activate(struct timer_list *timer) { }
512static inline void debug_timer_deactivate(struct timer_list *timer) { }
513#endif
514
515static void __init_timer(struct timer_list *timer)
516{
517 timer->entry.next = NULL;
518 timer->base = __raw_get_cpu_var(tvec_bases);
519#ifdef CONFIG_TIMER_STATS
520 timer->start_site = NULL;
521 timer->start_pid = -1;
522 memset(timer->start_comm, 0, TASK_COMM_LEN);
523#endif
524}
525
Rolf Eike Beer2aae4a12006-09-29 01:59:46 -0700526/**
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700527 * init_timer - initialize a timer.
528 * @timer: the timer to be initialized
529 *
530 * init_timer() must be done to a timer prior calling *any* of the
531 * other timer functions.
532 */
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -0800533void init_timer(struct timer_list *timer)
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700534{
Thomas Gleixnerc6f3a972008-04-30 00:55:03 -0700535 debug_timer_init(timer);
536 __init_timer(timer);
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700537}
538EXPORT_SYMBOL(init_timer);
539
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -0800540void init_timer_deferrable(struct timer_list *timer)
Venki Pallipadi6e453a62007-05-08 00:27:44 -0700541{
542 init_timer(timer);
543 timer_set_deferrable(timer);
544}
545EXPORT_SYMBOL(init_timer_deferrable);
546
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700547static inline void detach_timer(struct timer_list *timer,
Ingo Molnar82f67cd2007-02-16 01:28:13 -0800548 int clear_pending)
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700549{
550 struct list_head *entry = &timer->entry;
551
Thomas Gleixnerc6f3a972008-04-30 00:55:03 -0700552 debug_timer_deactivate(timer);
553
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700554 __list_del(entry->prev, entry->next);
555 if (clear_pending)
556 entry->next = NULL;
557 entry->prev = LIST_POISON2;
558}
559
560/*
Oleg Nesterov3691c512006-03-31 02:30:30 -0800561 * We are using hashed locking: holding per_cpu(tvec_bases).lock
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700562 * means that all timers which are tied to this base via timer->base are
563 * locked, and the base itself is locked too.
564 *
565 * So __run_timers/migrate_timers can safely modify all timers which could
566 * be found on ->tvX lists.
567 *
568 * When the timer's base is locked, and the timer removed from list, it is
569 * possible to set timer->base = NULL and drop the lock: the timer remains
570 * locked.
571 */
Pavel Macheka6fa8e52008-01-30 13:30:00 +0100572static struct tvec_base *lock_timer_base(struct timer_list *timer,
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700573 unsigned long *flags)
Josh Triplett89e7e3742006-09-29 01:59:36 -0700574 __acquires(timer->base->lock)
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700575{
Pavel Macheka6fa8e52008-01-30 13:30:00 +0100576 struct tvec_base *base;
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700577
578 for (;;) {
Pavel Macheka6fa8e52008-01-30 13:30:00 +0100579 struct tvec_base *prelock_base = timer->base;
Venki Pallipadi6e453a62007-05-08 00:27:44 -0700580 base = tbase_get_base(prelock_base);
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700581 if (likely(base != NULL)) {
582 spin_lock_irqsave(&base->lock, *flags);
Venki Pallipadi6e453a62007-05-08 00:27:44 -0700583 if (likely(prelock_base == timer->base))
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700584 return base;
585 /* The timer has migrated to another CPU */
586 spin_unlock_irqrestore(&base->lock, *flags);
587 }
588 cpu_relax();
589 }
590}
591
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592int __mod_timer(struct timer_list *timer, unsigned long expires)
593{
Pavel Macheka6fa8e52008-01-30 13:30:00 +0100594 struct tvec_base *base, *new_base;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 unsigned long flags;
596 int ret = 0;
597
Ingo Molnar82f67cd2007-02-16 01:28:13 -0800598 timer_stats_timer_set_start_info(timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 BUG_ON(!timer->function);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700601 base = lock_timer_base(timer, &flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700603 if (timer_pending(timer)) {
604 detach_timer(timer, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 ret = 1;
606 }
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700607
Thomas Gleixnerc6f3a972008-04-30 00:55:03 -0700608 debug_timer_activate(timer);
609
Jan Beulicha4a61982006-03-24 03:15:54 -0800610 new_base = __get_cpu_var(tvec_bases);
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700611
Oleg Nesterov3691c512006-03-31 02:30:30 -0800612 if (base != new_base) {
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700613 /*
614 * We are trying to schedule the timer on the local CPU.
615 * However we can't change timer's base while it is running,
616 * otherwise del_timer_sync() can't detect that the timer's
617 * handler yet has not finished. This also guarantees that
618 * the timer is serialized wrt itself.
619 */
Oleg Nesterova2c348f2006-03-31 02:30:31 -0800620 if (likely(base->running_timer != timer)) {
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700621 /* See the comment in lock_timer_base() */
Venki Pallipadi6e453a62007-05-08 00:27:44 -0700622 timer_set_base(timer, NULL);
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700623 spin_unlock(&base->lock);
Oleg Nesterova2c348f2006-03-31 02:30:31 -0800624 base = new_base;
625 spin_lock(&base->lock);
Venki Pallipadi6e453a62007-05-08 00:27:44 -0700626 timer_set_base(timer, base);
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700627 }
628 }
629
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 timer->expires = expires;
Oleg Nesterova2c348f2006-03-31 02:30:31 -0800631 internal_add_timer(base, timer);
632 spin_unlock_irqrestore(&base->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633
634 return ret;
635}
636
637EXPORT_SYMBOL(__mod_timer);
638
Rolf Eike Beer2aae4a12006-09-29 01:59:46 -0700639/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 * add_timer_on - start a timer on a particular CPU
641 * @timer: the timer to be added
642 * @cpu: the CPU to start it on
643 *
644 * This is not very scalable on SMP. Double adds are not possible.
645 */
646void add_timer_on(struct timer_list *timer, int cpu)
647{
Pavel Macheka6fa8e52008-01-30 13:30:00 +0100648 struct tvec_base *base = per_cpu(tvec_bases, cpu);
Thomas Gleixner68194572007-07-19 01:49:16 -0700649 unsigned long flags;
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700650
Ingo Molnar82f67cd2007-02-16 01:28:13 -0800651 timer_stats_timer_set_start_info(timer);
Thomas Gleixner68194572007-07-19 01:49:16 -0700652 BUG_ON(timer_pending(timer) || !timer->function);
Oleg Nesterov3691c512006-03-31 02:30:30 -0800653 spin_lock_irqsave(&base->lock, flags);
Venki Pallipadi6e453a62007-05-08 00:27:44 -0700654 timer_set_base(timer, base);
Thomas Gleixnerc6f3a972008-04-30 00:55:03 -0700655 debug_timer_activate(timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 internal_add_timer(base, timer);
Thomas Gleixner06d83082008-03-22 09:20:24 +0100657 /*
658 * Check whether the other CPU is idle and needs to be
659 * triggered to reevaluate the timer wheel when nohz is
660 * active. We are protected against the other CPU fiddling
661 * with the timer by holding the timer base lock. This also
662 * makes sure that a CPU on the way to idle can not evaluate
663 * the timer wheel.
664 */
665 wake_up_idle_cpu(cpu);
Oleg Nesterov3691c512006-03-31 02:30:30 -0800666 spin_unlock_irqrestore(&base->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667}
668
Rolf Eike Beer2aae4a12006-09-29 01:59:46 -0700669/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 * mod_timer - modify a timer's timeout
671 * @timer: the timer to be modified
Rolf Eike Beer2aae4a12006-09-29 01:59:46 -0700672 * @expires: new timeout in jiffies
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800674 * mod_timer() is a more efficient way to update the expire field of an
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 * active timer (if the timer is inactive it will be activated)
676 *
677 * mod_timer(timer, expires) is equivalent to:
678 *
679 * del_timer(timer); timer->expires = expires; add_timer(timer);
680 *
681 * Note that if there are multiple unserialized concurrent users of the
682 * same timer, then mod_timer() is the only safe way to modify the timeout,
683 * since add_timer() cannot modify an already running timer.
684 *
685 * The function returns whether it has modified a pending timer or not.
686 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
687 * active timer returns 1.)
688 */
689int mod_timer(struct timer_list *timer, unsigned long expires)
690{
691 BUG_ON(!timer->function);
692
Ingo Molnar82f67cd2007-02-16 01:28:13 -0800693 timer_stats_timer_set_start_info(timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 /*
695 * This is a common optimization triggered by the
696 * networking code - if the timer is re-modified
697 * to be the same thing then just return:
698 */
699 if (timer->expires == expires && timer_pending(timer))
700 return 1;
701
702 return __mod_timer(timer, expires);
703}
704
705EXPORT_SYMBOL(mod_timer);
706
Rolf Eike Beer2aae4a12006-09-29 01:59:46 -0700707/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 * del_timer - deactive a timer.
709 * @timer: the timer to be deactivated
710 *
711 * del_timer() deactivates a timer - this works on both active and inactive
712 * timers.
713 *
714 * The function returns whether it has deactivated a pending timer or not.
715 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
716 * active timer returns 1.)
717 */
718int del_timer(struct timer_list *timer)
719{
Pavel Macheka6fa8e52008-01-30 13:30:00 +0100720 struct tvec_base *base;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 unsigned long flags;
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700722 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723
Ingo Molnar82f67cd2007-02-16 01:28:13 -0800724 timer_stats_timer_clear_start_info(timer);
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700725 if (timer_pending(timer)) {
726 base = lock_timer_base(timer, &flags);
727 if (timer_pending(timer)) {
728 detach_timer(timer, 1);
729 ret = 1;
730 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 spin_unlock_irqrestore(&base->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700734 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735}
736
737EXPORT_SYMBOL(del_timer);
738
739#ifdef CONFIG_SMP
Rolf Eike Beer2aae4a12006-09-29 01:59:46 -0700740/**
741 * try_to_del_timer_sync - Try to deactivate a timer
742 * @timer: timer do del
743 *
Oleg Nesterovfd450b72005-06-23 00:08:59 -0700744 * This function tries to deactivate a timer. Upon successful (ret >= 0)
745 * exit the timer is not queued and the handler is not running on any CPU.
746 *
747 * It must not be called from interrupt contexts.
748 */
749int try_to_del_timer_sync(struct timer_list *timer)
750{
Pavel Macheka6fa8e52008-01-30 13:30:00 +0100751 struct tvec_base *base;
Oleg Nesterovfd450b72005-06-23 00:08:59 -0700752 unsigned long flags;
753 int ret = -1;
754
755 base = lock_timer_base(timer, &flags);
756
757 if (base->running_timer == timer)
758 goto out;
759
760 ret = 0;
761 if (timer_pending(timer)) {
762 detach_timer(timer, 1);
763 ret = 1;
764 }
765out:
766 spin_unlock_irqrestore(&base->lock, flags);
767
768 return ret;
769}
770
David Howellse19dff12007-04-26 15:46:56 -0700771EXPORT_SYMBOL(try_to_del_timer_sync);
772
Rolf Eike Beer2aae4a12006-09-29 01:59:46 -0700773/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 * del_timer_sync - deactivate a timer and wait for the handler to finish.
775 * @timer: the timer to be deactivated
776 *
777 * This function only differs from del_timer() on SMP: besides deactivating
778 * the timer it also makes sure the handler has finished executing on other
779 * CPUs.
780 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800781 * Synchronization rules: Callers must prevent restarting of the timer,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 * otherwise this function is meaningless. It must not be called from
783 * interrupt contexts. The caller must not hold locks which would prevent
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700784 * completion of the timer's handler. The timer's handler must not call
785 * add_timer_on(). Upon exit the timer is not queued and the handler is
786 * not running on any CPU.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 *
788 * The function returns whether it has deactivated a pending timer or not.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 */
790int del_timer_sync(struct timer_list *timer)
791{
Oleg Nesterovfd450b72005-06-23 00:08:59 -0700792 for (;;) {
793 int ret = try_to_del_timer_sync(timer);
794 if (ret >= 0)
795 return ret;
Andrew Mortona0009652006-07-14 00:24:06 -0700796 cpu_relax();
Oleg Nesterovfd450b72005-06-23 00:08:59 -0700797 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798}
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700799
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800EXPORT_SYMBOL(del_timer_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801#endif
802
Pavel Macheka6fa8e52008-01-30 13:30:00 +0100803static int cascade(struct tvec_base *base, struct tvec *tv, int index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804{
805 /* cascade all the timers from tv up one level */
Porpoise3439dd82006-06-23 02:05:56 -0700806 struct timer_list *timer, *tmp;
807 struct list_head tv_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808
Porpoise3439dd82006-06-23 02:05:56 -0700809 list_replace_init(tv->vec + index, &tv_list);
810
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 /*
Porpoise3439dd82006-06-23 02:05:56 -0700812 * We are removing _all_ timers from the list, so we
813 * don't have to detach them individually.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 */
Porpoise3439dd82006-06-23 02:05:56 -0700815 list_for_each_entry_safe(timer, tmp, &tv_list, entry) {
Venki Pallipadi6e453a62007-05-08 00:27:44 -0700816 BUG_ON(tbase_get_base(timer->base) != base);
Porpoise3439dd82006-06-23 02:05:56 -0700817 internal_add_timer(base, timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819
820 return index;
821}
822
Rolf Eike Beer2aae4a12006-09-29 01:59:46 -0700823#define INDEX(N) ((base->timer_jiffies >> (TVR_BITS + (N) * TVN_BITS)) & TVN_MASK)
824
825/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 * __run_timers - run all expired timers (if any) on this CPU.
827 * @base: the timer vector to be processed.
828 *
829 * This function cascades all vectors and executes all expired timer
830 * vectors.
831 */
Pavel Macheka6fa8e52008-01-30 13:30:00 +0100832static inline void __run_timers(struct tvec_base *base)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833{
834 struct timer_list *timer;
835
Oleg Nesterov3691c512006-03-31 02:30:30 -0800836 spin_lock_irq(&base->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 while (time_after_eq(jiffies, base->timer_jiffies)) {
Oleg Nesterov626ab0e2006-06-23 02:05:55 -0700838 struct list_head work_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 struct list_head *head = &work_list;
Thomas Gleixner68194572007-07-19 01:49:16 -0700840 int index = base->timer_jiffies & TVR_MASK;
Oleg Nesterov626ab0e2006-06-23 02:05:55 -0700841
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 /*
843 * Cascade timers:
844 */
845 if (!index &&
846 (!cascade(base, &base->tv2, INDEX(0))) &&
847 (!cascade(base, &base->tv3, INDEX(1))) &&
848 !cascade(base, &base->tv4, INDEX(2)))
849 cascade(base, &base->tv5, INDEX(3));
Oleg Nesterov626ab0e2006-06-23 02:05:55 -0700850 ++base->timer_jiffies;
851 list_replace_init(base->tv1.vec + index, &work_list);
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700852 while (!list_empty(head)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 void (*fn)(unsigned long);
854 unsigned long data;
855
Pavel Emelianovb5e61812007-05-08 00:30:19 -0700856 timer = list_first_entry(head, struct timer_list,entry);
Thomas Gleixner68194572007-07-19 01:49:16 -0700857 fn = timer->function;
858 data = timer->data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859
Ingo Molnar82f67cd2007-02-16 01:28:13 -0800860 timer_stats_account_timer(timer);
861
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 set_running_timer(base, timer);
Oleg Nesterov55c888d2005-06-23 00:08:56 -0700863 detach_timer(timer, 1);
Oleg Nesterov3691c512006-03-31 02:30:30 -0800864 spin_unlock_irq(&base->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 {
Jesper Juhlbe5b4fb2005-06-23 00:09:09 -0700866 int preempt_count = preempt_count();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 fn(data);
868 if (preempt_count != preempt_count()) {
Pavel Machek4c9dc642008-01-30 13:30:00 +0100869 printk(KERN_ERR "huh, entered %p "
Jesper Juhlbe5b4fb2005-06-23 00:09:09 -0700870 "with preempt_count %08x, exited"
871 " with %08x?\n",
872 fn, preempt_count,
873 preempt_count());
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 BUG();
875 }
876 }
Oleg Nesterov3691c512006-03-31 02:30:30 -0800877 spin_lock_irq(&base->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 }
879 }
880 set_running_timer(base, NULL);
Oleg Nesterov3691c512006-03-31 02:30:30 -0800881 spin_unlock_irq(&base->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882}
883
Russell Kingee9c5782008-04-20 13:59:33 +0100884#ifdef CONFIG_NO_HZ
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885/*
886 * Find out when the next timer event is due to happen. This
887 * is used on S/390 to stop all activity when a cpus is idle.
888 * This functions needs to be called disabled.
889 */
Pavel Macheka6fa8e52008-01-30 13:30:00 +0100890static unsigned long __next_timer_interrupt(struct tvec_base *base)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891{
Thomas Gleixner1cfd6842007-02-16 01:27:46 -0800892 unsigned long timer_jiffies = base->timer_jiffies;
Thomas Gleixnereaad0842007-05-29 23:47:39 +0200893 unsigned long expires = timer_jiffies + NEXT_TIMER_MAX_DELTA;
Thomas Gleixner1cfd6842007-02-16 01:27:46 -0800894 int index, slot, array, found = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 struct timer_list *nte;
Pavel Macheka6fa8e52008-01-30 13:30:00 +0100896 struct tvec *varray[4];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897
898 /* Look for timer events in tv1. */
Thomas Gleixner1cfd6842007-02-16 01:27:46 -0800899 index = slot = timer_jiffies & TVR_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 do {
Thomas Gleixner1cfd6842007-02-16 01:27:46 -0800901 list_for_each_entry(nte, base->tv1.vec + slot, entry) {
Thomas Gleixner68194572007-07-19 01:49:16 -0700902 if (tbase_get_deferrable(nte->base))
903 continue;
Venki Pallipadi6e453a62007-05-08 00:27:44 -0700904
Thomas Gleixner1cfd6842007-02-16 01:27:46 -0800905 found = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 expires = nte->expires;
Thomas Gleixner1cfd6842007-02-16 01:27:46 -0800907 /* Look at the cascade bucket(s)? */
908 if (!index || slot < index)
909 goto cascade;
910 return expires;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 }
Thomas Gleixner1cfd6842007-02-16 01:27:46 -0800912 slot = (slot + 1) & TVR_MASK;
913 } while (slot != index);
914
915cascade:
916 /* Calculate the next cascade event */
917 if (index)
918 timer_jiffies += TVR_SIZE - index;
919 timer_jiffies >>= TVR_BITS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920
921 /* Check tv2-tv5. */
922 varray[0] = &base->tv2;
923 varray[1] = &base->tv3;
924 varray[2] = &base->tv4;
925 varray[3] = &base->tv5;
Thomas Gleixner1cfd6842007-02-16 01:27:46 -0800926
927 for (array = 0; array < 4; array++) {
Pavel Macheka6fa8e52008-01-30 13:30:00 +0100928 struct tvec *varp = varray[array];
Thomas Gleixner1cfd6842007-02-16 01:27:46 -0800929
930 index = slot = timer_jiffies & TVN_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 do {
Thomas Gleixner1cfd6842007-02-16 01:27:46 -0800932 list_for_each_entry(nte, varp->vec + slot, entry) {
933 found = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 if (time_before(nte->expires, expires))
935 expires = nte->expires;
Thomas Gleixner1cfd6842007-02-16 01:27:46 -0800936 }
937 /*
938 * Do we still search for the first timer or are
939 * we looking up the cascade buckets ?
940 */
941 if (found) {
942 /* Look at the cascade bucket(s)? */
943 if (!index || slot < index)
944 break;
945 return expires;
946 }
947 slot = (slot + 1) & TVN_MASK;
948 } while (slot != index);
949
950 if (index)
951 timer_jiffies += TVN_SIZE - index;
952 timer_jiffies >>= TVN_BITS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 }
Thomas Gleixner1cfd6842007-02-16 01:27:46 -0800954 return expires;
955}
956
957/*
958 * Check, if the next hrtimer event is before the next timer wheel
959 * event:
960 */
961static unsigned long cmp_next_hrtimer_event(unsigned long now,
962 unsigned long expires)
963{
964 ktime_t hr_delta = hrtimer_get_next_event();
965 struct timespec tsdelta;
Thomas Gleixner9501b6c2007-03-25 14:31:17 +0200966 unsigned long delta;
Thomas Gleixner1cfd6842007-02-16 01:27:46 -0800967
968 if (hr_delta.tv64 == KTIME_MAX)
969 return expires;
970
Thomas Gleixner9501b6c2007-03-25 14:31:17 +0200971 /*
972 * Expired timer available, let it expire in the next tick
973 */
974 if (hr_delta.tv64 <= 0)
975 return now + 1;
Thomas Gleixner1cfd6842007-02-16 01:27:46 -0800976
977 tsdelta = ktime_to_timespec(hr_delta);
Thomas Gleixner9501b6c2007-03-25 14:31:17 +0200978 delta = timespec_to_jiffies(&tsdelta);
Thomas Gleixnereaad0842007-05-29 23:47:39 +0200979
980 /*
981 * Limit the delta to the max value, which is checked in
982 * tick_nohz_stop_sched_tick():
983 */
984 if (delta > NEXT_TIMER_MAX_DELTA)
985 delta = NEXT_TIMER_MAX_DELTA;
986
Thomas Gleixner9501b6c2007-03-25 14:31:17 +0200987 /*
988 * Take rounding errors in to account and make sure, that it
989 * expires in the next tick. Otherwise we go into an endless
990 * ping pong due to tick_nohz_stop_sched_tick() retriggering
991 * the timer softirq
992 */
993 if (delta < 1)
994 delta = 1;
995 now += delta;
Thomas Gleixner1cfd6842007-02-16 01:27:46 -0800996 if (time_before(now, expires))
997 return now;
998 return expires;
999}
1000
1001/**
Li Zefan8dce39c2007-11-05 14:51:10 -08001002 * get_next_timer_interrupt - return the jiffy of the next pending timer
Randy Dunlap05fb6bf2007-02-28 20:12:13 -08001003 * @now: current time (in jiffies)
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001004 */
Thomas Gleixnerfd064b92007-02-16 01:27:47 -08001005unsigned long get_next_timer_interrupt(unsigned long now)
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001006{
Pavel Macheka6fa8e52008-01-30 13:30:00 +01001007 struct tvec_base *base = __get_cpu_var(tvec_bases);
Thomas Gleixnerfd064b92007-02-16 01:27:47 -08001008 unsigned long expires;
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001009
1010 spin_lock(&base->lock);
1011 expires = __next_timer_interrupt(base);
Oleg Nesterov3691c512006-03-31 02:30:30 -08001012 spin_unlock(&base->lock);
Tony Lindgren69239742006-03-06 15:42:45 -08001013
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001014 if (time_before_eq(expires, now))
1015 return now;
Zachary Amsden0662b712006-05-20 15:00:24 -07001016
Thomas Gleixner1cfd6842007-02-16 01:27:46 -08001017 return cmp_next_hrtimer_event(now, expires);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018}
1019#endif
1020
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021/*
Daniel Walker5b4db0c2007-10-18 03:06:11 -07001022 * Called from the timer interrupt handler to charge one tick to the current
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 * process. user_tick is 1 if the tick is user time, 0 for system.
1024 */
1025void update_process_times(int user_tick)
1026{
1027 struct task_struct *p = current;
1028 int cpu = smp_processor_id();
1029
1030 /* Note: this timer irq context must be accounted for as well. */
Paul Mackerrasfa13a5a2007-11-09 22:39:38 +01001031 account_process_tick(p, user_tick);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 run_local_timers();
1033 if (rcu_pending(cpu))
1034 rcu_check_callbacks(cpu, user_tick);
Peter Zijlstrab845b512008-08-08 21:47:09 +02001035 printk_tick();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 scheduler_tick();
Thomas Gleixner68194572007-07-19 01:49:16 -07001037 run_posix_cpu_timers(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038}
1039
1040/*
1041 * Nr of active tasks - counted in fixed-point numbers
1042 */
1043static unsigned long count_active_tasks(void)
1044{
Jack Steinerdb1b1fe2006-03-31 02:31:21 -08001045 return nr_active() * FIXED_1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046}
1047
1048/*
1049 * Hmm.. Changed this, as the GNU make sources (load.c) seems to
1050 * imply that avenrun[] is the standard name for this kind of thing.
1051 * Nothing else seems to be standardized: the fractional size etc
1052 * all seem to differ on different machines.
1053 *
1054 * Requires xtime_lock to access.
1055 */
1056unsigned long avenrun[3];
1057
1058EXPORT_SYMBOL(avenrun);
1059
1060/*
1061 * calc_load - given tick count, update the avenrun load estimates.
1062 * This is called while holding a write_lock on xtime_lock.
1063 */
1064static inline void calc_load(unsigned long ticks)
1065{
1066 unsigned long active_tasks; /* fixed-point */
1067 static int count = LOAD_FREQ;
1068
Eric Dumazetcd7175e2006-12-13 00:35:45 -08001069 count -= ticks;
1070 if (unlikely(count < 0)) {
1071 active_tasks = count_active_tasks();
1072 do {
1073 CALC_LOAD(avenrun[0], EXP_1, active_tasks);
1074 CALC_LOAD(avenrun[1], EXP_5, active_tasks);
1075 CALC_LOAD(avenrun[2], EXP_15, active_tasks);
1076 count += LOAD_FREQ;
1077 } while (count < 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 }
1079}
1080
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 * This function runs timers and the timer-tq in bottom half context.
1083 */
1084static void run_timer_softirq(struct softirq_action *h)
1085{
Pavel Macheka6fa8e52008-01-30 13:30:00 +01001086 struct tvec_base *base = __get_cpu_var(tvec_bases);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087
Peter Zijlstrad3d74452008-01-25 21:08:31 +01001088 hrtimer_run_pending();
Ingo Molnar82f67cd2007-02-16 01:28:13 -08001089
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090 if (time_after_eq(jiffies, base->timer_jiffies))
1091 __run_timers(base);
1092}
1093
1094/*
1095 * Called by the local, per-CPU timer interrupt on SMP.
1096 */
1097void run_local_timers(void)
1098{
Peter Zijlstrad3d74452008-01-25 21:08:31 +01001099 hrtimer_run_queues();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 raise_softirq(TIMER_SOFTIRQ);
Ingo Molnar6687a972006-03-24 03:18:41 -08001101 softlockup_tick();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102}
1103
1104/*
1105 * Called by the timer interrupt. xtime_lock must already be taken
1106 * by the timer IRQ!
1107 */
Atsushi Nemoto3171a032006-09-29 02:00:32 -07001108static inline void update_times(unsigned long ticks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109{
john stultzad596172006-06-26 00:25:06 -07001110 update_wall_time();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111 calc_load(ticks);
1112}
Thomas Gleixner68194572007-07-19 01:49:16 -07001113
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114/*
1115 * The 64-bit jiffies value is not atomic - you MUST NOT read it
1116 * without sampling the sequence number in xtime_lock.
1117 * jiffies is defined in the linker script...
1118 */
1119
Atsushi Nemoto3171a032006-09-29 02:00:32 -07001120void do_timer(unsigned long ticks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121{
Atsushi Nemoto3171a032006-09-29 02:00:32 -07001122 jiffies_64 += ticks;
1123 update_times(ticks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124}
1125
1126#ifdef __ARCH_WANT_SYS_ALARM
1127
1128/*
1129 * For backwards compatibility? This can be done in libc so Alpha
1130 * and all newer ports shouldn't need it.
1131 */
Heiko Carstens58fd3aa2009-01-14 14:14:03 +01001132SYSCALL_DEFINE1(alarm, unsigned int, seconds)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133{
Thomas Gleixnerc08b8a42006-03-25 03:06:33 -08001134 return alarm_setitimer(seconds);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135}
1136
1137#endif
1138
1139#ifndef __alpha__
1140
1141/*
1142 * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this
1143 * should be moved into arch/i386 instead?
1144 */
1145
1146/**
1147 * sys_getpid - return the thread group id of the current process
1148 *
1149 * Note, despite the name, this returns the tgid not the pid. The tgid and
1150 * the pid are identical unless CLONE_THREAD was specified on clone() in
1151 * which case the tgid is the same in all threads of the same group.
1152 *
1153 * This is SMP safe as current->tgid does not change.
1154 */
Heiko Carstens58fd3aa2009-01-14 14:14:03 +01001155SYSCALL_DEFINE0(getpid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156{
Pavel Emelyanovb4888932007-10-18 23:40:14 -07001157 return task_tgid_vnr(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158}
1159
1160/*
Kirill Korotaev6997a6f2006-08-13 23:24:23 -07001161 * Accessing ->real_parent is not SMP-safe, it could
1162 * change from under us. However, we can use a stale
1163 * value of ->real_parent under rcu_read_lock(), see
1164 * release_task()->call_rcu(delayed_put_task_struct).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 */
Heiko Carstensdbf040d2009-01-14 14:14:04 +01001166SYSCALL_DEFINE0(getppid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167{
1168 int pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169
Kirill Korotaev6997a6f2006-08-13 23:24:23 -07001170 rcu_read_lock();
Pavel Emelyanov6c5f3e72008-02-08 04:19:20 -08001171 pid = task_tgid_vnr(current->real_parent);
Kirill Korotaev6997a6f2006-08-13 23:24:23 -07001172 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174 return pid;
1175}
1176
Heiko Carstensdbf040d2009-01-14 14:14:04 +01001177SYSCALL_DEFINE0(getuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178{
1179 /* Only we change this so SMP safe */
David Howells76aac0e2008-11-14 10:39:12 +11001180 return current_uid();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181}
1182
Heiko Carstensdbf040d2009-01-14 14:14:04 +01001183SYSCALL_DEFINE0(geteuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184{
1185 /* Only we change this so SMP safe */
David Howells76aac0e2008-11-14 10:39:12 +11001186 return current_euid();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187}
1188
Heiko Carstensdbf040d2009-01-14 14:14:04 +01001189SYSCALL_DEFINE0(getgid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190{
1191 /* Only we change this so SMP safe */
David Howells76aac0e2008-11-14 10:39:12 +11001192 return current_gid();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193}
1194
Heiko Carstensdbf040d2009-01-14 14:14:04 +01001195SYSCALL_DEFINE0(getegid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196{
1197 /* Only we change this so SMP safe */
David Howells76aac0e2008-11-14 10:39:12 +11001198 return current_egid();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199}
1200
1201#endif
1202
1203static void process_timeout(unsigned long __data)
1204{
Ingo Molnar36c8b582006-07-03 00:25:41 -07001205 wake_up_process((struct task_struct *)__data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206}
1207
1208/**
1209 * schedule_timeout - sleep until timeout
1210 * @timeout: timeout value in jiffies
1211 *
1212 * Make the current task sleep until @timeout jiffies have
1213 * elapsed. The routine will return immediately unless
1214 * the current task state has been set (see set_current_state()).
1215 *
1216 * You can set the task state as follows -
1217 *
1218 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1219 * pass before the routine returns. The routine will return 0
1220 *
1221 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1222 * delivered to the current task. In this case the remaining time
1223 * in jiffies will be returned, or 0 if the timer expired in time
1224 *
1225 * The current task state is guaranteed to be TASK_RUNNING when this
1226 * routine returns.
1227 *
1228 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1229 * the CPU away without a bound on the timeout. In this case the return
1230 * value will be %MAX_SCHEDULE_TIMEOUT.
1231 *
1232 * In all cases the return value is guaranteed to be non-negative.
1233 */
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -08001234signed long __sched schedule_timeout(signed long timeout)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235{
1236 struct timer_list timer;
1237 unsigned long expire;
1238
1239 switch (timeout)
1240 {
1241 case MAX_SCHEDULE_TIMEOUT:
1242 /*
1243 * These two special cases are useful to be comfortable
1244 * in the caller. Nothing more. We could take
1245 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1246 * but I' d like to return a valid offset (>=0) to allow
1247 * the caller to do everything it want with the retval.
1248 */
1249 schedule();
1250 goto out;
1251 default:
1252 /*
1253 * Another bit of PARANOID. Note that the retval will be
1254 * 0 since no piece of kernel is supposed to do a check
1255 * for a negative retval of schedule_timeout() (since it
1256 * should never happens anyway). You just have the printk()
1257 * that will tell you if something is gone wrong and where.
1258 */
Andrew Morton5b149bc2006-12-22 01:10:14 -08001259 if (timeout < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 printk(KERN_ERR "schedule_timeout: wrong timeout "
Andrew Morton5b149bc2006-12-22 01:10:14 -08001261 "value %lx\n", timeout);
1262 dump_stack();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 current->state = TASK_RUNNING;
1264 goto out;
1265 }
1266 }
1267
1268 expire = timeout + jiffies;
1269
Thomas Gleixnerc6f3a972008-04-30 00:55:03 -07001270 setup_timer_on_stack(&timer, process_timeout, (unsigned long)current);
Oleg Nesterova8db2db2005-10-30 15:01:38 -08001271 __mod_timer(&timer, expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272 schedule();
1273 del_singleshot_timer_sync(&timer);
1274
Thomas Gleixnerc6f3a972008-04-30 00:55:03 -07001275 /* Remove the timer from the object tracker */
1276 destroy_timer_on_stack(&timer);
1277
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 timeout = expire - jiffies;
1279
1280 out:
1281 return timeout < 0 ? 0 : timeout;
1282}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283EXPORT_SYMBOL(schedule_timeout);
1284
Andrew Morton8a1c1752005-09-13 01:25:15 -07001285/*
1286 * We can use __set_current_state() here because schedule_timeout() calls
1287 * schedule() unconditionally.
1288 */
Nishanth Aravamudan64ed93a2005-09-10 00:27:21 -07001289signed long __sched schedule_timeout_interruptible(signed long timeout)
1290{
Andrew Mortona5a0d522005-10-30 15:01:42 -08001291 __set_current_state(TASK_INTERRUPTIBLE);
1292 return schedule_timeout(timeout);
Nishanth Aravamudan64ed93a2005-09-10 00:27:21 -07001293}
1294EXPORT_SYMBOL(schedule_timeout_interruptible);
1295
Matthew Wilcox294d5cc2007-12-06 11:59:46 -05001296signed long __sched schedule_timeout_killable(signed long timeout)
1297{
1298 __set_current_state(TASK_KILLABLE);
1299 return schedule_timeout(timeout);
1300}
1301EXPORT_SYMBOL(schedule_timeout_killable);
1302
Nishanth Aravamudan64ed93a2005-09-10 00:27:21 -07001303signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1304{
Andrew Mortona5a0d522005-10-30 15:01:42 -08001305 __set_current_state(TASK_UNINTERRUPTIBLE);
1306 return schedule_timeout(timeout);
Nishanth Aravamudan64ed93a2005-09-10 00:27:21 -07001307}
1308EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1309
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310/* Thread ID - the internal kernel "pid" */
Heiko Carstens58fd3aa2009-01-14 14:14:03 +01001311SYSCALL_DEFINE0(gettid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312{
Pavel Emelyanovb4888932007-10-18 23:40:14 -07001313 return task_pid_vnr(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314}
1315
Rolf Eike Beer2aae4a12006-09-29 01:59:46 -07001316/**
Kyle McMartind4d23ad2007-02-10 01:46:00 -08001317 * do_sysinfo - fill in sysinfo struct
Rolf Eike Beer2aae4a12006-09-29 01:59:46 -07001318 * @info: pointer to buffer to fill
Thomas Gleixner68194572007-07-19 01:49:16 -07001319 */
Kyle McMartind4d23ad2007-02-10 01:46:00 -08001320int do_sysinfo(struct sysinfo *info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 unsigned long mem_total, sav_total;
1323 unsigned int mem_unit, bitcount;
1324 unsigned long seq;
1325
Kyle McMartind4d23ad2007-02-10 01:46:00 -08001326 memset(info, 0, sizeof(struct sysinfo));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327
1328 do {
1329 struct timespec tp;
1330 seq = read_seqbegin(&xtime_lock);
1331
1332 /*
1333 * This is annoying. The below is the same thing
1334 * posix_get_clock_monotonic() does, but it wants to
1335 * take the lock which we want to cover the loads stuff
1336 * too.
1337 */
1338
1339 getnstimeofday(&tp);
1340 tp.tv_sec += wall_to_monotonic.tv_sec;
1341 tp.tv_nsec += wall_to_monotonic.tv_nsec;
Tomas Janousekd6214142007-07-15 23:39:42 -07001342 monotonic_to_bootbased(&tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 if (tp.tv_nsec - NSEC_PER_SEC >= 0) {
1344 tp.tv_nsec = tp.tv_nsec - NSEC_PER_SEC;
1345 tp.tv_sec++;
1346 }
Kyle McMartind4d23ad2007-02-10 01:46:00 -08001347 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348
Kyle McMartind4d23ad2007-02-10 01:46:00 -08001349 info->loads[0] = avenrun[0] << (SI_LOAD_SHIFT - FSHIFT);
1350 info->loads[1] = avenrun[1] << (SI_LOAD_SHIFT - FSHIFT);
1351 info->loads[2] = avenrun[2] << (SI_LOAD_SHIFT - FSHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352
Kyle McMartind4d23ad2007-02-10 01:46:00 -08001353 info->procs = nr_threads;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 } while (read_seqretry(&xtime_lock, seq));
1355
Kyle McMartind4d23ad2007-02-10 01:46:00 -08001356 si_meminfo(info);
1357 si_swapinfo(info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358
1359 /*
1360 * If the sum of all the available memory (i.e. ram + swap)
1361 * is less than can be stored in a 32 bit unsigned long then
1362 * we can be binary compatible with 2.2.x kernels. If not,
1363 * well, in that case 2.2.x was broken anyways...
1364 *
1365 * -Erik Andersen <andersee@debian.org>
1366 */
1367
Kyle McMartind4d23ad2007-02-10 01:46:00 -08001368 mem_total = info->totalram + info->totalswap;
1369 if (mem_total < info->totalram || mem_total < info->totalswap)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 goto out;
1371 bitcount = 0;
Kyle McMartind4d23ad2007-02-10 01:46:00 -08001372 mem_unit = info->mem_unit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 while (mem_unit > 1) {
1374 bitcount++;
1375 mem_unit >>= 1;
1376 sav_total = mem_total;
1377 mem_total <<= 1;
1378 if (mem_total < sav_total)
1379 goto out;
1380 }
1381
1382 /*
1383 * If mem_total did not overflow, multiply all memory values by
Kyle McMartind4d23ad2007-02-10 01:46:00 -08001384 * info->mem_unit and set it to 1. This leaves things compatible
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1386 * kernels...
1387 */
1388
Kyle McMartind4d23ad2007-02-10 01:46:00 -08001389 info->mem_unit = 1;
1390 info->totalram <<= bitcount;
1391 info->freeram <<= bitcount;
1392 info->sharedram <<= bitcount;
1393 info->bufferram <<= bitcount;
1394 info->totalswap <<= bitcount;
1395 info->freeswap <<= bitcount;
1396 info->totalhigh <<= bitcount;
1397 info->freehigh <<= bitcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398
Kyle McMartind4d23ad2007-02-10 01:46:00 -08001399out:
1400 return 0;
1401}
1402
Heiko Carstens1e7bfb22009-01-14 14:14:29 +01001403SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
Kyle McMartind4d23ad2007-02-10 01:46:00 -08001404{
1405 struct sysinfo val;
1406
1407 do_sysinfo(&val);
1408
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
1410 return -EFAULT;
1411
1412 return 0;
1413}
1414
Adrian Bunkb4be6252007-12-18 18:05:58 +01001415static int __cpuinit init_timers_cpu(int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416{
1417 int j;
Pavel Macheka6fa8e52008-01-30 13:30:00 +01001418 struct tvec_base *base;
Adrian Bunkb4be6252007-12-18 18:05:58 +01001419 static char __cpuinitdata tvec_base_done[NR_CPUS];
Oleg Nesterov55c888d2005-06-23 00:08:56 -07001420
Andrew Mortonba6edfc2006-04-10 22:53:58 -07001421 if (!tvec_base_done[cpu]) {
Jan Beulicha4a61982006-03-24 03:15:54 -08001422 static char boot_done;
1423
Jan Beulicha4a61982006-03-24 03:15:54 -08001424 if (boot_done) {
Andrew Mortonba6edfc2006-04-10 22:53:58 -07001425 /*
1426 * The APs use this path later in boot
1427 */
Christoph Lameter94f60302007-07-17 04:03:29 -07001428 base = kmalloc_node(sizeof(*base),
1429 GFP_KERNEL | __GFP_ZERO,
Jan Beulicha4a61982006-03-24 03:15:54 -08001430 cpu_to_node(cpu));
1431 if (!base)
1432 return -ENOMEM;
Venki Pallipadi6e453a62007-05-08 00:27:44 -07001433
1434 /* Make sure that tvec_base is 2 byte aligned */
1435 if (tbase_get_deferrable(base)) {
1436 WARN_ON(1);
1437 kfree(base);
1438 return -ENOMEM;
1439 }
Andrew Mortonba6edfc2006-04-10 22:53:58 -07001440 per_cpu(tvec_bases, cpu) = base;
Jan Beulicha4a61982006-03-24 03:15:54 -08001441 } else {
Andrew Mortonba6edfc2006-04-10 22:53:58 -07001442 /*
1443 * This is for the boot CPU - we use compile-time
1444 * static initialisation because per-cpu memory isn't
1445 * ready yet and because the memory allocators are not
1446 * initialised either.
1447 */
Jan Beulicha4a61982006-03-24 03:15:54 -08001448 boot_done = 1;
Andrew Mortonba6edfc2006-04-10 22:53:58 -07001449 base = &boot_tvec_bases;
Jan Beulicha4a61982006-03-24 03:15:54 -08001450 }
Andrew Mortonba6edfc2006-04-10 22:53:58 -07001451 tvec_base_done[cpu] = 1;
1452 } else {
1453 base = per_cpu(tvec_bases, cpu);
Jan Beulicha4a61982006-03-24 03:15:54 -08001454 }
Andrew Mortonba6edfc2006-04-10 22:53:58 -07001455
Oleg Nesterov3691c512006-03-31 02:30:30 -08001456 spin_lock_init(&base->lock);
Ingo Molnard730e882006-07-03 00:25:10 -07001457
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 for (j = 0; j < TVN_SIZE; j++) {
1459 INIT_LIST_HEAD(base->tv5.vec + j);
1460 INIT_LIST_HEAD(base->tv4.vec + j);
1461 INIT_LIST_HEAD(base->tv3.vec + j);
1462 INIT_LIST_HEAD(base->tv2.vec + j);
1463 }
1464 for (j = 0; j < TVR_SIZE; j++)
1465 INIT_LIST_HEAD(base->tv1.vec + j);
1466
1467 base->timer_jiffies = jiffies;
Jan Beulicha4a61982006-03-24 03:15:54 -08001468 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469}
1470
1471#ifdef CONFIG_HOTPLUG_CPU
Pavel Macheka6fa8e52008-01-30 13:30:00 +01001472static void migrate_timer_list(struct tvec_base *new_base, struct list_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473{
1474 struct timer_list *timer;
1475
1476 while (!list_empty(head)) {
Pavel Emelianovb5e61812007-05-08 00:30:19 -07001477 timer = list_first_entry(head, struct timer_list, entry);
Oleg Nesterov55c888d2005-06-23 00:08:56 -07001478 detach_timer(timer, 0);
Venki Pallipadi6e453a62007-05-08 00:27:44 -07001479 timer_set_base(timer, new_base);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 internal_add_timer(new_base, timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482}
1483
Randy Dunlap48ccf3d2008-01-21 17:18:25 -08001484static void __cpuinit migrate_timers(int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485{
Pavel Macheka6fa8e52008-01-30 13:30:00 +01001486 struct tvec_base *old_base;
1487 struct tvec_base *new_base;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 int i;
1489
1490 BUG_ON(cpu_online(cpu));
Jan Beulicha4a61982006-03-24 03:15:54 -08001491 old_base = per_cpu(tvec_bases, cpu);
1492 new_base = get_cpu_var(tvec_bases);
Oleg Nesterovd82f0b02008-08-20 16:46:04 -07001493 /*
1494 * The caller is globally serialized and nobody else
1495 * takes two locks at once, deadlock is not possible.
1496 */
1497 spin_lock_irq(&new_base->lock);
Oleg Nesterov0d180402008-04-04 20:54:10 +02001498 spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499
Oleg Nesterov3691c512006-03-31 02:30:30 -08001500 BUG_ON(old_base->running_timer);
1501
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 for (i = 0; i < TVR_SIZE; i++)
Oleg Nesterov55c888d2005-06-23 00:08:56 -07001503 migrate_timer_list(new_base, old_base->tv1.vec + i);
1504 for (i = 0; i < TVN_SIZE; i++) {
1505 migrate_timer_list(new_base, old_base->tv2.vec + i);
1506 migrate_timer_list(new_base, old_base->tv3.vec + i);
1507 migrate_timer_list(new_base, old_base->tv4.vec + i);
1508 migrate_timer_list(new_base, old_base->tv5.vec + i);
1509 }
1510
Oleg Nesterov0d180402008-04-04 20:54:10 +02001511 spin_unlock(&old_base->lock);
Oleg Nesterovd82f0b02008-08-20 16:46:04 -07001512 spin_unlock_irq(&new_base->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513 put_cpu_var(tvec_bases);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514}
1515#endif /* CONFIG_HOTPLUG_CPU */
1516
Chandra Seetharaman8c78f302006-07-30 03:03:35 -07001517static int __cpuinit timer_cpu_notify(struct notifier_block *self,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518 unsigned long action, void *hcpu)
1519{
1520 long cpu = (long)hcpu;
1521 switch(action) {
1522 case CPU_UP_PREPARE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07001523 case CPU_UP_PREPARE_FROZEN:
Jan Beulicha4a61982006-03-24 03:15:54 -08001524 if (init_timers_cpu(cpu) < 0)
1525 return NOTIFY_BAD;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 break;
1527#ifdef CONFIG_HOTPLUG_CPU
1528 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07001529 case CPU_DEAD_FROZEN:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 migrate_timers(cpu);
1531 break;
1532#endif
1533 default:
1534 break;
1535 }
1536 return NOTIFY_OK;
1537}
1538
Chandra Seetharaman8c78f302006-07-30 03:03:35 -07001539static struct notifier_block __cpuinitdata timers_nb = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540 .notifier_call = timer_cpu_notify,
1541};
1542
1543
1544void __init init_timers(void)
1545{
Akinobu Mita07dccf32006-09-29 02:00:22 -07001546 int err = timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547 (void *)(long)smp_processor_id());
Akinobu Mita07dccf32006-09-29 02:00:22 -07001548
Ingo Molnar82f67cd2007-02-16 01:28:13 -08001549 init_timer_stats();
1550
Akinobu Mita07dccf32006-09-29 02:00:22 -07001551 BUG_ON(err == NOTIFY_BAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 register_cpu_notifier(&timers_nb);
Carlos R. Mafra962cf362008-05-15 11:15:37 -03001553 open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554}
1555
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556/**
1557 * msleep - sleep safely even with waitqueue interruptions
1558 * @msecs: Time in milliseconds to sleep for
1559 */
1560void msleep(unsigned int msecs)
1561{
1562 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1563
Nishanth Aravamudan75bcc8c2005-09-10 00:27:24 -07001564 while (timeout)
1565 timeout = schedule_timeout_uninterruptible(timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566}
1567
1568EXPORT_SYMBOL(msleep);
1569
1570/**
Domen Puncer96ec3ef2005-06-25 14:58:43 -07001571 * msleep_interruptible - sleep waiting for signals
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572 * @msecs: Time in milliseconds to sleep for
1573 */
1574unsigned long msleep_interruptible(unsigned int msecs)
1575{
1576 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1577
Nishanth Aravamudan75bcc8c2005-09-10 00:27:24 -07001578 while (timeout && !signal_pending(current))
1579 timeout = schedule_timeout_interruptible(timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 return jiffies_to_msecs(timeout);
1581}
1582
1583EXPORT_SYMBOL(msleep_interruptible);