]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * linux/kernel/timer.c | |
3 | * | |
4 | * Kernel internal timers, basic process system calls | |
5 | * | |
6 | * Copyright (C) 1991, 1992 Linus Torvalds | |
7 | * | |
8 | * 1997-01-28 Modified by Finn Arne Gangstad to make timers scale better. | |
9 | * | |
10 | * 1997-09-10 Updated NTP code according to technical memorandum Jan '96 | |
11 | * "A Kernel Model for Precision Timekeeping" by Dave Mills | |
12 | * 1998-12-24 Fixed a xtime SMP race (we need the xtime_lock rw spinlock to | |
13 | * serialize accesses to xtime/lost_ticks). | |
14 | * Copyright (C) 1998 Andrea Arcangeli | |
15 | * 1999-03-10 Improved NTP compatibility by Ulrich Windl | |
16 | * 2002-05-31 Move sys_sysinfo here and make its locking sane, Robert Love | |
17 | * 2000-10-05 Implemented scalable SMP per-CPU timer handling. | |
18 | * Copyright (C) 2000, 2001, 2002 Ingo Molnar | |
19 | * Designed by David S. Miller, Alexey Kuznetsov and Ingo Molnar | |
20 | */ | |
21 | ||
22 | #include <linux/kernel_stat.h> | |
23 | #include <linux/module.h> | |
24 | #include <linux/interrupt.h> | |
25 | #include <linux/percpu.h> | |
26 | #include <linux/init.h> | |
27 | #include <linux/mm.h> | |
28 | #include <linux/swap.h> | |
29 | #include <linux/pid_namespace.h> | |
30 | #include <linux/notifier.h> | |
31 | #include <linux/thread_info.h> | |
32 | #include <linux/time.h> | |
33 | #include <linux/jiffies.h> | |
34 | #include <linux/posix-timers.h> | |
35 | #include <linux/cpu.h> | |
36 | #include <linux/syscalls.h> | |
37 | #include <linux/delay.h> | |
38 | #include <linux/tick.h> | |
39 | #include <linux/kallsyms.h> | |
40 | ||
41 | #include <asm/uaccess.h> | |
42 | #include <asm/unistd.h> | |
43 | #include <asm/div64.h> | |
44 | #include <asm/timex.h> | |
45 | #include <asm/io.h> | |
46 | ||
47 | u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES; | |
48 | ||
49 | EXPORT_SYMBOL(jiffies_64); | |
50 | ||
51 | /* | |
52 | * per-CPU timer vector definitions: | |
53 | */ | |
54 | #define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6) | |
55 | #define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8) | |
56 | #define TVN_SIZE (1 << TVN_BITS) | |
57 | #define TVR_SIZE (1 << TVR_BITS) | |
58 | #define TVN_MASK (TVN_SIZE - 1) | |
59 | #define TVR_MASK (TVR_SIZE - 1) | |
60 | ||
61 | struct tvec { | |
62 | struct list_head vec[TVN_SIZE]; | |
63 | }; | |
64 | ||
65 | struct tvec_root { | |
66 | struct list_head vec[TVR_SIZE]; | |
67 | }; | |
68 | ||
69 | struct tvec_base { | |
70 | spinlock_t lock; | |
71 | struct timer_list *running_timer; | |
72 | unsigned long timer_jiffies; | |
73 | struct tvec_root tv1; | |
74 | struct tvec tv2; | |
75 | struct tvec tv3; | |
76 | struct tvec tv4; | |
77 | struct tvec tv5; | |
78 | } ____cacheline_aligned; | |
79 | ||
80 | struct tvec_base boot_tvec_bases; | |
81 | EXPORT_SYMBOL(boot_tvec_bases); | |
82 | static DEFINE_PER_CPU(struct tvec_base *, tvec_bases) = &boot_tvec_bases; | |
83 | ||
84 | /* | |
85 | * Note that all tvec_bases are 2 byte aligned and lower bit of | |
86 | * base in timer_list is guaranteed to be zero. Use the LSB for | |
87 | * the new flag to indicate whether the timer is deferrable | |
88 | */ | |
89 | #define TBASE_DEFERRABLE_FLAG (0x1) | |
90 | ||
91 | /* Functions below help us manage 'deferrable' flag */ | |
92 | static inline unsigned int tbase_get_deferrable(struct tvec_base *base) | |
93 | { | |
94 | return ((unsigned int)(unsigned long)base & TBASE_DEFERRABLE_FLAG); | |
95 | } | |
96 | ||
97 | static inline struct tvec_base *tbase_get_base(struct tvec_base *base) | |
98 | { | |
99 | return ((struct tvec_base *)((unsigned long)base & ~TBASE_DEFERRABLE_FLAG)); | |
100 | } | |
101 | ||
102 | static inline void timer_set_deferrable(struct timer_list *timer) | |
103 | { | |
104 | timer->base = ((struct tvec_base *)((unsigned long)(timer->base) | | |
105 | TBASE_DEFERRABLE_FLAG)); | |
106 | } | |
107 | ||
108 | static inline void | |
109 | timer_set_base(struct timer_list *timer, struct tvec_base *new_base) | |
110 | { | |
111 | timer->base = (struct tvec_base *)((unsigned long)(new_base) | | |
112 | tbase_get_deferrable(timer->base)); | |
113 | } | |
114 | ||
115 | /** | |
116 | * __round_jiffies - function to round jiffies to a full second | |
117 | * @j: the time in (absolute) jiffies that should be rounded | |
118 | * @cpu: the processor number on which the timeout will happen | |
119 | * | |
120 | * __round_jiffies() rounds an absolute time in the future (in jiffies) | |
121 | * up or down to (approximately) full seconds. This is useful for timers | |
122 | * for which the exact time they fire does not matter too much, as long as | |
123 | * they fire approximately every X seconds. | |
124 | * | |
125 | * By rounding these timers to whole seconds, all such timers will fire | |
126 | * at the same time, rather than at various times spread out. The goal | |
127 | * of this is to have the CPU wake up less, which saves power. | |
128 | * | |
129 | * The exact rounding is skewed for each processor to avoid all | |
130 | * processors firing at the exact same time, which could lead | |
131 | * to lock contention or spurious cache line bouncing. | |
132 | * | |
133 | * The return value is the rounded version of the @j parameter. | |
134 | */ | |
135 | unsigned long __round_jiffies(unsigned long j, int cpu) | |
136 | { | |
137 | int rem; | |
138 | unsigned long original = j; | |
139 | ||
140 | /* | |
141 | * We don't want all cpus firing their timers at once hitting the | |
142 | * same lock or cachelines, so we skew each extra cpu with an extra | |
143 | * 3 jiffies. This 3 jiffies came originally from the mm/ code which | |
144 | * already did this. | |
145 | * The skew is done by adding 3*cpunr, then round, then subtract this | |
146 | * extra offset again. | |
147 | */ | |
148 | j += cpu * 3; | |
149 | ||
150 | rem = j % HZ; | |
151 | ||
152 | /* | |
153 | * If the target jiffie is just after a whole second (which can happen | |
154 | * due to delays of the timer irq, long irq off times etc etc) then | |
155 | * we should round down to the whole second, not up. Use 1/4th second | |
156 | * as cutoff for this rounding as an extreme upper bound for this. | |
157 | */ | |
158 | if (rem < HZ/4) /* round down */ | |
159 | j = j - rem; | |
160 | else /* round up */ | |
161 | j = j - rem + HZ; | |
162 | ||
163 | /* now that we have rounded, subtract the extra skew again */ | |
164 | j -= cpu * 3; | |
165 | ||
166 | if (j <= jiffies) /* rounding ate our timeout entirely; */ | |
167 | return original; | |
168 | return j; | |
169 | } | |
170 | EXPORT_SYMBOL_GPL(__round_jiffies); | |
171 | ||
172 | /** | |
173 | * __round_jiffies_relative - function to round jiffies to a full second | |
174 | * @j: the time in (relative) jiffies that should be rounded | |
175 | * @cpu: the processor number on which the timeout will happen | |
176 | * | |
177 | * __round_jiffies_relative() rounds a time delta in the future (in jiffies) | |
178 | * up or down to (approximately) full seconds. This is useful for timers | |
179 | * for which the exact time they fire does not matter too much, as long as | |
180 | * they fire approximately every X seconds. | |
181 | * | |
182 | * By rounding these timers to whole seconds, all such timers will fire | |
183 | * at the same time, rather than at various times spread out. The goal | |
184 | * of this is to have the CPU wake up less, which saves power. | |
185 | * | |
186 | * The exact rounding is skewed for each processor to avoid all | |
187 | * processors firing at the exact same time, which could lead | |
188 | * to lock contention or spurious cache line bouncing. | |
189 | * | |
190 | * The return value is the rounded version of the @j parameter. | |
191 | */ | |
192 | unsigned long __round_jiffies_relative(unsigned long j, int cpu) | |
193 | { | |
194 | /* | |
195 | * In theory the following code can skip a jiffy in case jiffies | |
196 | * increments right between the addition and the later subtraction. | |
197 | * However since the entire point of this function is to use approximate | |
198 | * timeouts, it's entirely ok to not handle that. | |
199 | */ | |
200 | return __round_jiffies(j + jiffies, cpu) - jiffies; | |
201 | } | |
202 | EXPORT_SYMBOL_GPL(__round_jiffies_relative); | |
203 | ||
204 | /** | |
205 | * round_jiffies - function to round jiffies to a full second | |
206 | * @j: the time in (absolute) jiffies that should be rounded | |
207 | * | |
208 | * round_jiffies() rounds an absolute time in the future (in jiffies) | |
209 | * up or down to (approximately) full seconds. This is useful for timers | |
210 | * for which the exact time they fire does not matter too much, as long as | |
211 | * they fire approximately every X seconds. | |
212 | * | |
213 | * By rounding these timers to whole seconds, all such timers will fire | |
214 | * at the same time, rather than at various times spread out. The goal | |
215 | * of this is to have the CPU wake up less, which saves power. | |
216 | * | |
217 | * The return value is the rounded version of the @j parameter. | |
218 | */ | |
219 | unsigned long round_jiffies(unsigned long j) | |
220 | { | |
221 | return __round_jiffies(j, raw_smp_processor_id()); | |
222 | } | |
223 | EXPORT_SYMBOL_GPL(round_jiffies); | |
224 | ||
225 | /** | |
226 | * round_jiffies_relative - function to round jiffies to a full second | |
227 | * @j: the time in (relative) jiffies that should be rounded | |
228 | * | |
229 | * round_jiffies_relative() rounds a time delta in the future (in jiffies) | |
230 | * up or down to (approximately) full seconds. This is useful for timers | |
231 | * for which the exact time they fire does not matter too much, as long as | |
232 | * they fire approximately every X seconds. | |
233 | * | |
234 | * By rounding these timers to whole seconds, all such timers will fire | |
235 | * at the same time, rather than at various times spread out. The goal | |
236 | * of this is to have the CPU wake up less, which saves power. | |
237 | * | |
238 | * The return value is the rounded version of the @j parameter. | |
239 | */ | |
240 | unsigned long round_jiffies_relative(unsigned long j) | |
241 | { | |
242 | return __round_jiffies_relative(j, raw_smp_processor_id()); | |
243 | } | |
244 | EXPORT_SYMBOL_GPL(round_jiffies_relative); | |
245 | ||
246 | ||
247 | static inline void set_running_timer(struct tvec_base *base, | |
248 | struct timer_list *timer) | |
249 | { | |
250 | #ifdef CONFIG_SMP | |
251 | base->running_timer = timer; | |
252 | #endif | |
253 | } | |
254 | ||
255 | static void internal_add_timer(struct tvec_base *base, struct timer_list *timer) | |
256 | { | |
257 | unsigned long expires = timer->expires; | |
258 | unsigned long idx = expires - base->timer_jiffies; | |
259 | struct list_head *vec; | |
260 | ||
261 | if (idx < TVR_SIZE) { | |
262 | int i = expires & TVR_MASK; | |
263 | vec = base->tv1.vec + i; | |
264 | } else if (idx < 1 << (TVR_BITS + TVN_BITS)) { | |
265 | int i = (expires >> TVR_BITS) & TVN_MASK; | |
266 | vec = base->tv2.vec + i; | |
267 | } else if (idx < 1 << (TVR_BITS + 2 * TVN_BITS)) { | |
268 | int i = (expires >> (TVR_BITS + TVN_BITS)) & TVN_MASK; | |
269 | vec = base->tv3.vec + i; | |
270 | } else if (idx < 1 << (TVR_BITS + 3 * TVN_BITS)) { | |
271 | int i = (expires >> (TVR_BITS + 2 * TVN_BITS)) & TVN_MASK; | |
272 | vec = base->tv4.vec + i; | |
273 | } else if ((signed long) idx < 0) { | |
274 | /* | |
275 | * Can happen if you add a timer with expires == jiffies, | |
276 | * or you set a timer to go off in the past | |
277 | */ | |
278 | vec = base->tv1.vec + (base->timer_jiffies & TVR_MASK); | |
279 | } else { | |
280 | int i; | |
281 | /* If the timeout is larger than 0xffffffff on 64-bit | |
282 | * architectures then we use the maximum timeout: | |
283 | */ | |
284 | if (idx > 0xffffffffUL) { | |
285 | idx = 0xffffffffUL; | |
286 | expires = idx + base->timer_jiffies; | |
287 | } | |
288 | i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK; | |
289 | vec = base->tv5.vec + i; | |
290 | } | |
291 | /* | |
292 | * Timers are FIFO: | |
293 | */ | |
294 | list_add_tail(&timer->entry, vec); | |
295 | } | |
296 | ||
297 | #ifdef CONFIG_TIMER_STATS | |
298 | void __timer_stats_timer_set_start_info(struct timer_list *timer, void *addr) | |
299 | { | |
300 | if (timer->start_site) | |
301 | return; | |
302 | ||
303 | timer->start_site = addr; | |
304 | memcpy(timer->start_comm, current->comm, TASK_COMM_LEN); | |
305 | timer->start_pid = current->pid; | |
306 | } | |
307 | ||
308 | static void timer_stats_account_timer(struct timer_list *timer) | |
309 | { | |
310 | unsigned int flag = 0; | |
311 | ||
312 | if (unlikely(tbase_get_deferrable(timer->base))) | |
313 | flag |= TIMER_STATS_FLAG_DEFERRABLE; | |
314 | ||
315 | timer_stats_update_stats(timer, timer->start_pid, timer->start_site, | |
316 | timer->function, timer->start_comm, flag); | |
317 | } | |
318 | ||
319 | #else | |
320 | static void timer_stats_account_timer(struct timer_list *timer) {} | |
321 | #endif | |
322 | ||
323 | #ifdef CONFIG_DEBUG_OBJECTS_TIMERS | |
324 | ||
325 | static struct debug_obj_descr timer_debug_descr; | |
326 | ||
327 | /* | |
328 | * fixup_init is called when: | |
329 | * - an active object is initialized | |
330 | */ | |
331 | static int timer_fixup_init(void *addr, enum debug_obj_state state) | |
332 | { | |
333 | struct timer_list *timer = addr; | |
334 | ||
335 | switch (state) { | |
336 | case ODEBUG_STATE_ACTIVE: | |
337 | del_timer_sync(timer); | |
338 | debug_object_init(timer, &timer_debug_descr); | |
339 | return 1; | |
340 | default: | |
341 | return 0; | |
342 | } | |
343 | } | |
344 | ||
345 | /* | |
346 | * fixup_activate is called when: | |
347 | * - an active object is activated | |
348 | * - an unknown object is activated (might be a statically initialized object) | |
349 | */ | |
350 | static int timer_fixup_activate(void *addr, enum debug_obj_state state) | |
351 | { | |
352 | struct timer_list *timer = addr; | |
353 | ||
354 | switch (state) { | |
355 | ||
356 | case ODEBUG_STATE_NOTAVAILABLE: | |
357 | /* | |
358 | * This is not really a fixup. The timer was | |
359 | * statically initialized. We just make sure that it | |
360 | * is tracked in the object tracker. | |
361 | */ | |
362 | if (timer->entry.next == NULL && | |
363 | timer->entry.prev == TIMER_ENTRY_STATIC) { | |
364 | debug_object_init(timer, &timer_debug_descr); | |
365 | debug_object_activate(timer, &timer_debug_descr); | |
366 | return 0; | |
367 | } else { | |
368 | WARN_ON_ONCE(1); | |
369 | } | |
370 | return 0; | |
371 | ||
372 | case ODEBUG_STATE_ACTIVE: | |
373 | WARN_ON(1); | |
374 | ||
375 | default: | |
376 | return 0; | |
377 | } | |
378 | } | |
379 | ||
380 | /* | |
381 | * fixup_free is called when: | |
382 | * - an active object is freed | |
383 | */ | |
384 | static int timer_fixup_free(void *addr, enum debug_obj_state state) | |
385 | { | |
386 | struct timer_list *timer = addr; | |
387 | ||
388 | switch (state) { | |
389 | case ODEBUG_STATE_ACTIVE: | |
390 | del_timer_sync(timer); | |
391 | debug_object_free(timer, &timer_debug_descr); | |
392 | return 1; | |
393 | default: | |
394 | return 0; | |
395 | } | |
396 | } | |
397 | ||
398 | static struct debug_obj_descr timer_debug_descr = { | |
399 | .name = "timer_list", | |
400 | .fixup_init = timer_fixup_init, | |
401 | .fixup_activate = timer_fixup_activate, | |
402 | .fixup_free = timer_fixup_free, | |
403 | }; | |
404 | ||
405 | static inline void debug_timer_init(struct timer_list *timer) | |
406 | { | |
407 | debug_object_init(timer, &timer_debug_descr); | |
408 | } | |
409 | ||
410 | static inline void debug_timer_activate(struct timer_list *timer) | |
411 | { | |
412 | debug_object_activate(timer, &timer_debug_descr); | |
413 | } | |
414 | ||
415 | static inline void debug_timer_deactivate(struct timer_list *timer) | |
416 | { | |
417 | debug_object_deactivate(timer, &timer_debug_descr); | |
418 | } | |
419 | ||
420 | static inline void debug_timer_free(struct timer_list *timer) | |
421 | { | |
422 | debug_object_free(timer, &timer_debug_descr); | |
423 | } | |
424 | ||
425 | static void __init_timer(struct timer_list *timer); | |
426 | ||
427 | void init_timer_on_stack(struct timer_list *timer) | |
428 | { | |
429 | debug_object_init_on_stack(timer, &timer_debug_descr); | |
430 | __init_timer(timer); | |
431 | } | |
432 | EXPORT_SYMBOL_GPL(init_timer_on_stack); | |
433 | ||
434 | void destroy_timer_on_stack(struct timer_list *timer) | |
435 | { | |
436 | debug_object_free(timer, &timer_debug_descr); | |
437 | } | |
438 | EXPORT_SYMBOL_GPL(destroy_timer_on_stack); | |
439 | ||
440 | #else | |
441 | static inline void debug_timer_init(struct timer_list *timer) { } | |
442 | static inline void debug_timer_activate(struct timer_list *timer) { } | |
443 | static inline void debug_timer_deactivate(struct timer_list *timer) { } | |
444 | #endif | |
445 | ||
446 | static void __init_timer(struct timer_list *timer) | |
447 | { | |
448 | timer->entry.next = NULL; | |
449 | timer->base = __raw_get_cpu_var(tvec_bases); | |
450 | #ifdef CONFIG_TIMER_STATS | |
451 | timer->start_site = NULL; | |
452 | timer->start_pid = -1; | |
453 | memset(timer->start_comm, 0, TASK_COMM_LEN); | |
454 | #endif | |
455 | } | |
456 | ||
457 | /** | |
458 | * init_timer - initialize a timer. | |
459 | * @timer: the timer to be initialized | |
460 | * | |
461 | * init_timer() must be done to a timer prior calling *any* of the | |
462 | * other timer functions. | |
463 | */ | |
464 | void init_timer(struct timer_list *timer) | |
465 | { | |
466 | debug_timer_init(timer); | |
467 | __init_timer(timer); | |
468 | } | |
469 | EXPORT_SYMBOL(init_timer); | |
470 | ||
471 | void init_timer_deferrable(struct timer_list *timer) | |
472 | { | |
473 | init_timer(timer); | |
474 | timer_set_deferrable(timer); | |
475 | } | |
476 | EXPORT_SYMBOL(init_timer_deferrable); | |
477 | ||
478 | static inline void detach_timer(struct timer_list *timer, | |
479 | int clear_pending) | |
480 | { | |
481 | struct list_head *entry = &timer->entry; | |
482 | ||
483 | debug_timer_deactivate(timer); | |
484 | ||
485 | __list_del(entry->prev, entry->next); | |
486 | if (clear_pending) | |
487 | entry->next = NULL; | |
488 | entry->prev = LIST_POISON2; | |
489 | } | |
490 | ||
491 | /* | |
492 | * We are using hashed locking: holding per_cpu(tvec_bases).lock | |
493 | * means that all timers which are tied to this base via timer->base are | |
494 | * locked, and the base itself is locked too. | |
495 | * | |
496 | * So __run_timers/migrate_timers can safely modify all timers which could | |
497 | * be found on ->tvX lists. | |
498 | * | |
499 | * When the timer's base is locked, and the timer removed from list, it is | |
500 | * possible to set timer->base = NULL and drop the lock: the timer remains | |
501 | * locked. | |
502 | */ | |
503 | static struct tvec_base *lock_timer_base(struct timer_list *timer, | |
504 | unsigned long *flags) | |
505 | __acquires(timer->base->lock) | |
506 | { | |
507 | struct tvec_base *base; | |
508 | ||
509 | for (;;) { | |
510 | struct tvec_base *prelock_base = timer->base; | |
511 | base = tbase_get_base(prelock_base); | |
512 | if (likely(base != NULL)) { | |
513 | spin_lock_irqsave(&base->lock, *flags); | |
514 | if (likely(prelock_base == timer->base)) | |
515 | return base; | |
516 | /* The timer has migrated to another CPU */ | |
517 | spin_unlock_irqrestore(&base->lock, *flags); | |
518 | } | |
519 | cpu_relax(); | |
520 | } | |
521 | } | |
522 | ||
523 | int __mod_timer(struct timer_list *timer, unsigned long expires) | |
524 | { | |
525 | struct tvec_base *base, *new_base; | |
526 | unsigned long flags; | |
527 | int ret = 0; | |
528 | ||
529 | timer_stats_timer_set_start_info(timer); | |
530 | BUG_ON(!timer->function); | |
531 | ||
532 | base = lock_timer_base(timer, &flags); | |
533 | ||
534 | if (timer_pending(timer)) { | |
535 | detach_timer(timer, 0); | |
536 | ret = 1; | |
537 | } | |
538 | ||
539 | debug_timer_activate(timer); | |
540 | ||
541 | new_base = __get_cpu_var(tvec_bases); | |
542 | ||
543 | if (base != new_base) { | |
544 | /* | |
545 | * We are trying to schedule the timer on the local CPU. | |
546 | * However we can't change timer's base while it is running, | |
547 | * otherwise del_timer_sync() can't detect that the timer's | |
548 | * handler yet has not finished. This also guarantees that | |
549 | * the timer is serialized wrt itself. | |
550 | */ | |
551 | if (likely(base->running_timer != timer)) { | |
552 | /* See the comment in lock_timer_base() */ | |
553 | timer_set_base(timer, NULL); | |
554 | spin_unlock(&base->lock); | |
555 | base = new_base; | |
556 | spin_lock(&base->lock); | |
557 | timer_set_base(timer, base); | |
558 | } | |
559 | } | |
560 | ||
561 | timer->expires = expires; | |
562 | internal_add_timer(base, timer); | |
563 | spin_unlock_irqrestore(&base->lock, flags); | |
564 | ||
565 | return ret; | |
566 | } | |
567 | ||
568 | EXPORT_SYMBOL(__mod_timer); | |
569 | ||
570 | /** | |
571 | * add_timer_on - start a timer on a particular CPU | |
572 | * @timer: the timer to be added | |
573 | * @cpu: the CPU to start it on | |
574 | * | |
575 | * This is not very scalable on SMP. Double adds are not possible. | |
576 | */ | |
577 | void add_timer_on(struct timer_list *timer, int cpu) | |
578 | { | |
579 | struct tvec_base *base = per_cpu(tvec_bases, cpu); | |
580 | unsigned long flags; | |
581 | ||
582 | timer_stats_timer_set_start_info(timer); | |
583 | BUG_ON(timer_pending(timer) || !timer->function); | |
584 | spin_lock_irqsave(&base->lock, flags); | |
585 | timer_set_base(timer, base); | |
586 | debug_timer_activate(timer); | |
587 | internal_add_timer(base, timer); | |
588 | /* | |
589 | * Check whether the other CPU is idle and needs to be | |
590 | * triggered to reevaluate the timer wheel when nohz is | |
591 | * active. We are protected against the other CPU fiddling | |
592 | * with the timer by holding the timer base lock. This also | |
593 | * makes sure that a CPU on the way to idle can not evaluate | |
594 | * the timer wheel. | |
595 | */ | |
596 | wake_up_idle_cpu(cpu); | |
597 | spin_unlock_irqrestore(&base->lock, flags); | |
598 | } | |
599 | ||
600 | /** | |
601 | * mod_timer - modify a timer's timeout | |
602 | * @timer: the timer to be modified | |
603 | * @expires: new timeout in jiffies | |
604 | * | |
605 | * mod_timer() is a more efficient way to update the expire field of an | |
606 | * active timer (if the timer is inactive it will be activated) | |
607 | * | |
608 | * mod_timer(timer, expires) is equivalent to: | |
609 | * | |
610 | * del_timer(timer); timer->expires = expires; add_timer(timer); | |
611 | * | |
612 | * Note that if there are multiple unserialized concurrent users of the | |
613 | * same timer, then mod_timer() is the only safe way to modify the timeout, | |
614 | * since add_timer() cannot modify an already running timer. | |
615 | * | |
616 | * The function returns whether it has modified a pending timer or not. | |
617 | * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an | |
618 | * active timer returns 1.) | |
619 | */ | |
620 | int mod_timer(struct timer_list *timer, unsigned long expires) | |
621 | { | |
622 | BUG_ON(!timer->function); | |
623 | ||
624 | timer_stats_timer_set_start_info(timer); | |
625 | /* | |
626 | * This is a common optimization triggered by the | |
627 | * networking code - if the timer is re-modified | |
628 | * to be the same thing then just return: | |
629 | */ | |
630 | if (timer->expires == expires && timer_pending(timer)) | |
631 | return 1; | |
632 | ||
633 | return __mod_timer(timer, expires); | |
634 | } | |
635 | ||
636 | EXPORT_SYMBOL(mod_timer); | |
637 | ||
638 | /** | |
639 | * del_timer - deactive a timer. | |
640 | * @timer: the timer to be deactivated | |
641 | * | |
642 | * del_timer() deactivates a timer - this works on both active and inactive | |
643 | * timers. | |
644 | * | |
645 | * The function returns whether it has deactivated a pending timer or not. | |
646 | * (ie. del_timer() of an inactive timer returns 0, del_timer() of an | |
647 | * active timer returns 1.) | |
648 | */ | |
649 | int del_timer(struct timer_list *timer) | |
650 | { | |
651 | struct tvec_base *base; | |
652 | unsigned long flags; | |
653 | int ret = 0; | |
654 | ||
655 | timer_stats_timer_clear_start_info(timer); | |
656 | if (timer_pending(timer)) { | |
657 | base = lock_timer_base(timer, &flags); | |
658 | if (timer_pending(timer)) { | |
659 | detach_timer(timer, 1); | |
660 | ret = 1; | |
661 | } | |
662 | spin_unlock_irqrestore(&base->lock, flags); | |
663 | } | |
664 | ||
665 | return ret; | |
666 | } | |
667 | ||
668 | EXPORT_SYMBOL(del_timer); | |
669 | ||
670 | #ifdef CONFIG_SMP | |
671 | /** | |
672 | * try_to_del_timer_sync - Try to deactivate a timer | |
673 | * @timer: timer do del | |
674 | * | |
675 | * This function tries to deactivate a timer. Upon successful (ret >= 0) | |
676 | * exit the timer is not queued and the handler is not running on any CPU. | |
677 | * | |
678 | * It must not be called from interrupt contexts. | |
679 | */ | |
680 | int try_to_del_timer_sync(struct timer_list *timer) | |
681 | { | |
682 | struct tvec_base *base; | |
683 | unsigned long flags; | |
684 | int ret = -1; | |
685 | ||
686 | base = lock_timer_base(timer, &flags); | |
687 | ||
688 | if (base->running_timer == timer) | |
689 | goto out; | |
690 | ||
691 | ret = 0; | |
692 | if (timer_pending(timer)) { | |
693 | detach_timer(timer, 1); | |
694 | ret = 1; | |
695 | } | |
696 | out: | |
697 | spin_unlock_irqrestore(&base->lock, flags); | |
698 | ||
699 | return ret; | |
700 | } | |
701 | ||
702 | EXPORT_SYMBOL(try_to_del_timer_sync); | |
703 | ||
704 | /** | |
705 | * del_timer_sync - deactivate a timer and wait for the handler to finish. | |
706 | * @timer: the timer to be deactivated | |
707 | * | |
708 | * This function only differs from del_timer() on SMP: besides deactivating | |
709 | * the timer it also makes sure the handler has finished executing on other | |
710 | * CPUs. | |
711 | * | |
712 | * Synchronization rules: Callers must prevent restarting of the timer, | |
713 | * otherwise this function is meaningless. It must not be called from | |
714 | * interrupt contexts. The caller must not hold locks which would prevent | |
715 | * completion of the timer's handler. The timer's handler must not call | |
716 | * add_timer_on(). Upon exit the timer is not queued and the handler is | |
717 | * not running on any CPU. | |
718 | * | |
719 | * The function returns whether it has deactivated a pending timer or not. | |
720 | */ | |
721 | int del_timer_sync(struct timer_list *timer) | |
722 | { | |
723 | for (;;) { | |
724 | int ret = try_to_del_timer_sync(timer); | |
725 | if (ret >= 0) | |
726 | return ret; | |
727 | cpu_relax(); | |
728 | } | |
729 | } | |
730 | ||
731 | EXPORT_SYMBOL(del_timer_sync); | |
732 | #endif | |
733 | ||
734 | static int cascade(struct tvec_base *base, struct tvec *tv, int index) | |
735 | { | |
736 | /* cascade all the timers from tv up one level */ | |
737 | struct timer_list *timer, *tmp; | |
738 | struct list_head tv_list; | |
739 | ||
740 | list_replace_init(tv->vec + index, &tv_list); | |
741 | ||
742 | /* | |
743 | * We are removing _all_ timers from the list, so we | |
744 | * don't have to detach them individually. | |
745 | */ | |
746 | list_for_each_entry_safe(timer, tmp, &tv_list, entry) { | |
747 | BUG_ON(tbase_get_base(timer->base) != base); | |
748 | internal_add_timer(base, timer); | |
749 | } | |
750 | ||
751 | return index; | |
752 | } | |
753 | ||
754 | #define INDEX(N) ((base->timer_jiffies >> (TVR_BITS + (N) * TVN_BITS)) & TVN_MASK) | |
755 | ||
756 | /** | |
757 | * __run_timers - run all expired timers (if any) on this CPU. | |
758 | * @base: the timer vector to be processed. | |
759 | * | |
760 | * This function cascades all vectors and executes all expired timer | |
761 | * vectors. | |
762 | */ | |
763 | static inline void __run_timers(struct tvec_base *base) | |
764 | { | |
765 | struct timer_list *timer; | |
766 | ||
767 | spin_lock_irq(&base->lock); | |
768 | while (time_after_eq(jiffies, base->timer_jiffies)) { | |
769 | struct list_head work_list; | |
770 | struct list_head *head = &work_list; | |
771 | int index = base->timer_jiffies & TVR_MASK; | |
772 | ||
773 | /* | |
774 | * Cascade timers: | |
775 | */ | |
776 | if (!index && | |
777 | (!cascade(base, &base->tv2, INDEX(0))) && | |
778 | (!cascade(base, &base->tv3, INDEX(1))) && | |
779 | !cascade(base, &base->tv4, INDEX(2))) | |
780 | cascade(base, &base->tv5, INDEX(3)); | |
781 | ++base->timer_jiffies; | |
782 | list_replace_init(base->tv1.vec + index, &work_list); | |
783 | while (!list_empty(head)) { | |
784 | void (*fn)(unsigned long); | |
785 | unsigned long data; | |
786 | ||
787 | timer = list_first_entry(head, struct timer_list,entry); | |
788 | fn = timer->function; | |
789 | data = timer->data; | |
790 | ||
791 | timer_stats_account_timer(timer); | |
792 | ||
793 | set_running_timer(base, timer); | |
794 | detach_timer(timer, 1); | |
795 | spin_unlock_irq(&base->lock); | |
796 | { | |
797 | int preempt_count = preempt_count(); | |
798 | fn(data); | |
799 | if (preempt_count != preempt_count()) { | |
800 | printk(KERN_ERR "huh, entered %p " | |
801 | "with preempt_count %08x, exited" | |
802 | " with %08x?\n", | |
803 | fn, preempt_count, | |
804 | preempt_count()); | |
805 | BUG(); | |
806 | } | |
807 | } | |
808 | spin_lock_irq(&base->lock); | |
809 | } | |
810 | } | |
811 | set_running_timer(base, NULL); | |
812 | spin_unlock_irq(&base->lock); | |
813 | } | |
814 | ||
815 | #ifdef CONFIG_NO_HZ | |
816 | /* | |
817 | * Find out when the next timer event is due to happen. This | |
818 | * is used on S/390 to stop all activity when a cpus is idle. | |
819 | * This functions needs to be called disabled. | |
820 | */ | |
821 | static unsigned long __next_timer_interrupt(struct tvec_base *base) | |
822 | { | |
823 | unsigned long timer_jiffies = base->timer_jiffies; | |
824 | unsigned long expires = timer_jiffies + NEXT_TIMER_MAX_DELTA; | |
825 | int index, slot, array, found = 0; | |
826 | struct timer_list *nte; | |
827 | struct tvec *varray[4]; | |
828 | ||
829 | /* Look for timer events in tv1. */ | |
830 | index = slot = timer_jiffies & TVR_MASK; | |
831 | do { | |
832 | list_for_each_entry(nte, base->tv1.vec + slot, entry) { | |
833 | if (tbase_get_deferrable(nte->base)) | |
834 | continue; | |
835 | ||
836 | found = 1; | |
837 | expires = nte->expires; | |
838 | /* Look at the cascade bucket(s)? */ | |
839 | if (!index || slot < index) | |
840 | goto cascade; | |
841 | return expires; | |
842 | } | |
843 | slot = (slot + 1) & TVR_MASK; | |
844 | } while (slot != index); | |
845 | ||
846 | cascade: | |
847 | /* Calculate the next cascade event */ | |
848 | if (index) | |
849 | timer_jiffies += TVR_SIZE - index; | |
850 | timer_jiffies >>= TVR_BITS; | |
851 | ||
852 | /* Check tv2-tv5. */ | |
853 | varray[0] = &base->tv2; | |
854 | varray[1] = &base->tv3; | |
855 | varray[2] = &base->tv4; | |
856 | varray[3] = &base->tv5; | |
857 | ||
858 | for (array = 0; array < 4; array++) { | |
859 | struct tvec *varp = varray[array]; | |
860 | ||
861 | index = slot = timer_jiffies & TVN_MASK; | |
862 | do { | |
863 | list_for_each_entry(nte, varp->vec + slot, entry) { | |
864 | found = 1; | |
865 | if (time_before(nte->expires, expires)) | |
866 | expires = nte->expires; | |
867 | } | |
868 | /* | |
869 | * Do we still search for the first timer or are | |
870 | * we looking up the cascade buckets ? | |
871 | */ | |
872 | if (found) { | |
873 | /* Look at the cascade bucket(s)? */ | |
874 | if (!index || slot < index) | |
875 | break; | |
876 | return expires; | |
877 | } | |
878 | slot = (slot + 1) & TVN_MASK; | |
879 | } while (slot != index); | |
880 | ||
881 | if (index) | |
882 | timer_jiffies += TVN_SIZE - index; | |
883 | timer_jiffies >>= TVN_BITS; | |
884 | } | |
885 | return expires; | |
886 | } | |
887 | ||
888 | /* | |
889 | * Check, if the next hrtimer event is before the next timer wheel | |
890 | * event: | |
891 | */ | |
892 | static unsigned long cmp_next_hrtimer_event(unsigned long now, | |
893 | unsigned long expires) | |
894 | { | |
895 | ktime_t hr_delta = hrtimer_get_next_event(); | |
896 | struct timespec tsdelta; | |
897 | unsigned long delta; | |
898 | ||
899 | if (hr_delta.tv64 == KTIME_MAX) | |
900 | return expires; | |
901 | ||
902 | /* | |
903 | * Expired timer available, let it expire in the next tick | |
904 | */ | |
905 | if (hr_delta.tv64 <= 0) | |
906 | return now + 1; | |
907 | ||
908 | tsdelta = ktime_to_timespec(hr_delta); | |
909 | delta = timespec_to_jiffies(&tsdelta); | |
910 | ||
911 | /* | |
912 | * Limit the delta to the max value, which is checked in | |
913 | * tick_nohz_stop_sched_tick(): | |
914 | */ | |
915 | if (delta > NEXT_TIMER_MAX_DELTA) | |
916 | delta = NEXT_TIMER_MAX_DELTA; | |
917 | ||
918 | /* | |
919 | * Take rounding errors in to account and make sure, that it | |
920 | * expires in the next tick. Otherwise we go into an endless | |
921 | * ping pong due to tick_nohz_stop_sched_tick() retriggering | |
922 | * the timer softirq | |
923 | */ | |
924 | if (delta < 1) | |
925 | delta = 1; | |
926 | now += delta; | |
927 | if (time_before(now, expires)) | |
928 | return now; | |
929 | return expires; | |
930 | } | |
931 | ||
932 | /** | |
933 | * get_next_timer_interrupt - return the jiffy of the next pending timer | |
934 | * @now: current time (in jiffies) | |
935 | */ | |
936 | unsigned long get_next_timer_interrupt(unsigned long now) | |
937 | { | |
938 | struct tvec_base *base = __get_cpu_var(tvec_bases); | |
939 | unsigned long expires; | |
940 | ||
941 | spin_lock(&base->lock); | |
942 | expires = __next_timer_interrupt(base); | |
943 | spin_unlock(&base->lock); | |
944 | ||
945 | if (time_before_eq(expires, now)) | |
946 | return now; | |
947 | ||
948 | return cmp_next_hrtimer_event(now, expires); | |
949 | } | |
950 | #endif | |
951 | ||
952 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING | |
953 | void account_process_tick(struct task_struct *p, int user_tick) | |
954 | { | |
955 | cputime_t one_jiffy = jiffies_to_cputime(1); | |
956 | ||
957 | if (user_tick) { | |
958 | account_user_time(p, one_jiffy); | |
959 | account_user_time_scaled(p, cputime_to_scaled(one_jiffy)); | |
960 | } else { | |
961 | account_system_time(p, HARDIRQ_OFFSET, one_jiffy); | |
962 | account_system_time_scaled(p, cputime_to_scaled(one_jiffy)); | |
963 | } | |
964 | } | |
965 | #endif | |
966 | ||
967 | /* | |
968 | * Called from the timer interrupt handler to charge one tick to the current | |
969 | * process. user_tick is 1 if the tick is user time, 0 for system. | |
970 | */ | |
971 | void update_process_times(int user_tick) | |
972 | { | |
973 | struct task_struct *p = current; | |
974 | int cpu = smp_processor_id(); | |
975 | ||
976 | /* Note: this timer irq context must be accounted for as well. */ | |
977 | account_process_tick(p, user_tick); | |
978 | run_local_timers(); | |
979 | if (rcu_pending(cpu)) | |
980 | rcu_check_callbacks(cpu, user_tick); | |
981 | scheduler_tick(); | |
982 | run_posix_cpu_timers(p); | |
983 | } | |
984 | ||
985 | /* | |
986 | * Nr of active tasks - counted in fixed-point numbers | |
987 | */ | |
988 | static unsigned long count_active_tasks(void) | |
989 | { | |
990 | return nr_active() * FIXED_1; | |
991 | } | |
992 | ||
993 | /* | |
994 | * Hmm.. Changed this, as the GNU make sources (load.c) seems to | |
995 | * imply that avenrun[] is the standard name for this kind of thing. | |
996 | * Nothing else seems to be standardized: the fractional size etc | |
997 | * all seem to differ on different machines. | |
998 | * | |
999 | * Requires xtime_lock to access. | |
1000 | */ | |
1001 | unsigned long avenrun[3]; | |
1002 | ||
1003 | EXPORT_SYMBOL(avenrun); | |
1004 | ||
1005 | /* | |
1006 | * calc_load - given tick count, update the avenrun load estimates. | |
1007 | * This is called while holding a write_lock on xtime_lock. | |
1008 | */ | |
1009 | static inline void calc_load(unsigned long ticks) | |
1010 | { | |
1011 | unsigned long active_tasks; /* fixed-point */ | |
1012 | static int count = LOAD_FREQ; | |
1013 | ||
1014 | count -= ticks; | |
1015 | if (unlikely(count < 0)) { | |
1016 | active_tasks = count_active_tasks(); | |
1017 | do { | |
1018 | CALC_LOAD(avenrun[0], EXP_1, active_tasks); | |
1019 | CALC_LOAD(avenrun[1], EXP_5, active_tasks); | |
1020 | CALC_LOAD(avenrun[2], EXP_15, active_tasks); | |
1021 | count += LOAD_FREQ; | |
1022 | } while (count < 0); | |
1023 | } | |
1024 | } | |
1025 | ||
1026 | /* | |
1027 | * This function runs timers and the timer-tq in bottom half context. | |
1028 | */ | |
1029 | static void run_timer_softirq(struct softirq_action *h) | |
1030 | { | |
1031 | struct tvec_base *base = __get_cpu_var(tvec_bases); | |
1032 | ||
1033 | hrtimer_run_pending(); | |
1034 | ||
1035 | if (time_after_eq(jiffies, base->timer_jiffies)) | |
1036 | __run_timers(base); | |
1037 | } | |
1038 | ||
1039 | /* | |
1040 | * Called by the local, per-CPU timer interrupt on SMP. | |
1041 | */ | |
1042 | void run_local_timers(void) | |
1043 | { | |
1044 | hrtimer_run_queues(); | |
1045 | raise_softirq(TIMER_SOFTIRQ); | |
1046 | softlockup_tick(); | |
1047 | } | |
1048 | ||
1049 | /* | |
1050 | * Called by the timer interrupt. xtime_lock must already be taken | |
1051 | * by the timer IRQ! | |
1052 | */ | |
1053 | static inline void update_times(unsigned long ticks) | |
1054 | { | |
1055 | update_wall_time(); | |
1056 | calc_load(ticks); | |
1057 | } | |
1058 | ||
1059 | /* | |
1060 | * The 64-bit jiffies value is not atomic - you MUST NOT read it | |
1061 | * without sampling the sequence number in xtime_lock. | |
1062 | * jiffies is defined in the linker script... | |
1063 | */ | |
1064 | ||
1065 | void do_timer(unsigned long ticks) | |
1066 | { | |
1067 | jiffies_64 += ticks; | |
1068 | update_times(ticks); | |
1069 | } | |
1070 | ||
1071 | #ifdef __ARCH_WANT_SYS_ALARM | |
1072 | ||
1073 | /* | |
1074 | * For backwards compatibility? This can be done in libc so Alpha | |
1075 | * and all newer ports shouldn't need it. | |
1076 | */ | |
1077 | asmlinkage unsigned long sys_alarm(unsigned int seconds) | |
1078 | { | |
1079 | return alarm_setitimer(seconds); | |
1080 | } | |
1081 | ||
1082 | #endif | |
1083 | ||
1084 | #ifndef __alpha__ | |
1085 | ||
1086 | /* | |
1087 | * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this | |
1088 | * should be moved into arch/i386 instead? | |
1089 | */ | |
1090 | ||
1091 | /** | |
1092 | * sys_getpid - return the thread group id of the current process | |
1093 | * | |
1094 | * Note, despite the name, this returns the tgid not the pid. The tgid and | |
1095 | * the pid are identical unless CLONE_THREAD was specified on clone() in | |
1096 | * which case the tgid is the same in all threads of the same group. | |
1097 | * | |
1098 | * This is SMP safe as current->tgid does not change. | |
1099 | */ | |
1100 | asmlinkage long sys_getpid(void) | |
1101 | { | |
1102 | return task_tgid_vnr(current); | |
1103 | } | |
1104 | ||
1105 | /* | |
1106 | * Accessing ->real_parent is not SMP-safe, it could | |
1107 | * change from under us. However, we can use a stale | |
1108 | * value of ->real_parent under rcu_read_lock(), see | |
1109 | * release_task()->call_rcu(delayed_put_task_struct). | |
1110 | */ | |
1111 | asmlinkage long sys_getppid(void) | |
1112 | { | |
1113 | int pid; | |
1114 | ||
1115 | rcu_read_lock(); | |
1116 | pid = task_tgid_vnr(current->real_parent); | |
1117 | rcu_read_unlock(); | |
1118 | ||
1119 | return pid; | |
1120 | } | |
1121 | ||
1122 | asmlinkage long sys_getuid(void) | |
1123 | { | |
1124 | /* Only we change this so SMP safe */ | |
1125 | return current->uid; | |
1126 | } | |
1127 | ||
1128 | asmlinkage long sys_geteuid(void) | |
1129 | { | |
1130 | /* Only we change this so SMP safe */ | |
1131 | return current->euid; | |
1132 | } | |
1133 | ||
1134 | asmlinkage long sys_getgid(void) | |
1135 | { | |
1136 | /* Only we change this so SMP safe */ | |
1137 | return current->gid; | |
1138 | } | |
1139 | ||
1140 | asmlinkage long sys_getegid(void) | |
1141 | { | |
1142 | /* Only we change this so SMP safe */ | |
1143 | return current->egid; | |
1144 | } | |
1145 | ||
1146 | #endif | |
1147 | ||
1148 | static void process_timeout(unsigned long __data) | |
1149 | { | |
1150 | wake_up_process((struct task_struct *)__data); | |
1151 | } | |
1152 | ||
1153 | /** | |
1154 | * schedule_timeout - sleep until timeout | |
1155 | * @timeout: timeout value in jiffies | |
1156 | * | |
1157 | * Make the current task sleep until @timeout jiffies have | |
1158 | * elapsed. The routine will return immediately unless | |
1159 | * the current task state has been set (see set_current_state()). | |
1160 | * | |
1161 | * You can set the task state as follows - | |
1162 | * | |
1163 | * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to | |
1164 | * pass before the routine returns. The routine will return 0 | |
1165 | * | |
1166 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
1167 | * delivered to the current task. In this case the remaining time | |
1168 | * in jiffies will be returned, or 0 if the timer expired in time | |
1169 | * | |
1170 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1171 | * routine returns. | |
1172 | * | |
1173 | * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule | |
1174 | * the CPU away without a bound on the timeout. In this case the return | |
1175 | * value will be %MAX_SCHEDULE_TIMEOUT. | |
1176 | * | |
1177 | * In all cases the return value is guaranteed to be non-negative. | |
1178 | */ | |
1179 | signed long __sched schedule_timeout(signed long timeout) | |
1180 | { | |
1181 | struct timer_list timer; | |
1182 | unsigned long expire; | |
1183 | ||
1184 | switch (timeout) | |
1185 | { | |
1186 | case MAX_SCHEDULE_TIMEOUT: | |
1187 | /* | |
1188 | * These two special cases are useful to be comfortable | |
1189 | * in the caller. Nothing more. We could take | |
1190 | * MAX_SCHEDULE_TIMEOUT from one of the negative value | |
1191 | * but I' d like to return a valid offset (>=0) to allow | |
1192 | * the caller to do everything it want with the retval. | |
1193 | */ | |
1194 | schedule(); | |
1195 | goto out; | |
1196 | default: | |
1197 | /* | |
1198 | * Another bit of PARANOID. Note that the retval will be | |
1199 | * 0 since no piece of kernel is supposed to do a check | |
1200 | * for a negative retval of schedule_timeout() (since it | |
1201 | * should never happens anyway). You just have the printk() | |
1202 | * that will tell you if something is gone wrong and where. | |
1203 | */ | |
1204 | if (timeout < 0) { | |
1205 | printk(KERN_ERR "schedule_timeout: wrong timeout " | |
1206 | "value %lx\n", timeout); | |
1207 | dump_stack(); | |
1208 | current->state = TASK_RUNNING; | |
1209 | goto out; | |
1210 | } | |
1211 | } | |
1212 | ||
1213 | expire = timeout + jiffies; | |
1214 | ||
1215 | setup_timer_on_stack(&timer, process_timeout, (unsigned long)current); | |
1216 | __mod_timer(&timer, expire); | |
1217 | schedule(); | |
1218 | del_singleshot_timer_sync(&timer); | |
1219 | ||
1220 | /* Remove the timer from the object tracker */ | |
1221 | destroy_timer_on_stack(&timer); | |
1222 | ||
1223 | timeout = expire - jiffies; | |
1224 | ||
1225 | out: | |
1226 | return timeout < 0 ? 0 : timeout; | |
1227 | } | |
1228 | EXPORT_SYMBOL(schedule_timeout); | |
1229 | ||
1230 | /* | |
1231 | * We can use __set_current_state() here because schedule_timeout() calls | |
1232 | * schedule() unconditionally. | |
1233 | */ | |
1234 | signed long __sched schedule_timeout_interruptible(signed long timeout) | |
1235 | { | |
1236 | __set_current_state(TASK_INTERRUPTIBLE); | |
1237 | return schedule_timeout(timeout); | |
1238 | } | |
1239 | EXPORT_SYMBOL(schedule_timeout_interruptible); | |
1240 | ||
1241 | signed long __sched schedule_timeout_killable(signed long timeout) | |
1242 | { | |
1243 | __set_current_state(TASK_KILLABLE); | |
1244 | return schedule_timeout(timeout); | |
1245 | } | |
1246 | EXPORT_SYMBOL(schedule_timeout_killable); | |
1247 | ||
1248 | signed long __sched schedule_timeout_uninterruptible(signed long timeout) | |
1249 | { | |
1250 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
1251 | return schedule_timeout(timeout); | |
1252 | } | |
1253 | EXPORT_SYMBOL(schedule_timeout_uninterruptible); | |
1254 | ||
1255 | /* Thread ID - the internal kernel "pid" */ | |
1256 | asmlinkage long sys_gettid(void) | |
1257 | { | |
1258 | return task_pid_vnr(current); | |
1259 | } | |
1260 | ||
1261 | /** | |
1262 | * do_sysinfo - fill in sysinfo struct | |
1263 | * @info: pointer to buffer to fill | |
1264 | */ | |
1265 | int do_sysinfo(struct sysinfo *info) | |
1266 | { | |
1267 | unsigned long mem_total, sav_total; | |
1268 | unsigned int mem_unit, bitcount; | |
1269 | unsigned long seq; | |
1270 | ||
1271 | memset(info, 0, sizeof(struct sysinfo)); | |
1272 | ||
1273 | do { | |
1274 | struct timespec tp; | |
1275 | seq = read_seqbegin(&xtime_lock); | |
1276 | ||
1277 | /* | |
1278 | * This is annoying. The below is the same thing | |
1279 | * posix_get_clock_monotonic() does, but it wants to | |
1280 | * take the lock which we want to cover the loads stuff | |
1281 | * too. | |
1282 | */ | |
1283 | ||
1284 | getnstimeofday(&tp); | |
1285 | tp.tv_sec += wall_to_monotonic.tv_sec; | |
1286 | tp.tv_nsec += wall_to_monotonic.tv_nsec; | |
1287 | monotonic_to_bootbased(&tp); | |
1288 | if (tp.tv_nsec - NSEC_PER_SEC >= 0) { | |
1289 | tp.tv_nsec = tp.tv_nsec - NSEC_PER_SEC; | |
1290 | tp.tv_sec++; | |
1291 | } | |
1292 | info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0); | |
1293 | ||
1294 | info->loads[0] = avenrun[0] << (SI_LOAD_SHIFT - FSHIFT); | |
1295 | info->loads[1] = avenrun[1] << (SI_LOAD_SHIFT - FSHIFT); | |
1296 | info->loads[2] = avenrun[2] << (SI_LOAD_SHIFT - FSHIFT); | |
1297 | ||
1298 | info->procs = nr_threads; | |
1299 | } while (read_seqretry(&xtime_lock, seq)); | |
1300 | ||
1301 | si_meminfo(info); | |
1302 | si_swapinfo(info); | |
1303 | ||
1304 | /* | |
1305 | * If the sum of all the available memory (i.e. ram + swap) | |
1306 | * is less than can be stored in a 32 bit unsigned long then | |
1307 | * we can be binary compatible with 2.2.x kernels. If not, | |
1308 | * well, in that case 2.2.x was broken anyways... | |
1309 | * | |
1310 | * -Erik Andersen <andersee@debian.org> | |
1311 | */ | |
1312 | ||
1313 | mem_total = info->totalram + info->totalswap; | |
1314 | if (mem_total < info->totalram || mem_total < info->totalswap) | |
1315 | goto out; | |
1316 | bitcount = 0; | |
1317 | mem_unit = info->mem_unit; | |
1318 | while (mem_unit > 1) { | |
1319 | bitcount++; | |
1320 | mem_unit >>= 1; | |
1321 | sav_total = mem_total; | |
1322 | mem_total <<= 1; | |
1323 | if (mem_total < sav_total) | |
1324 | goto out; | |
1325 | } | |
1326 | ||
1327 | /* | |
1328 | * If mem_total did not overflow, multiply all memory values by | |
1329 | * info->mem_unit and set it to 1. This leaves things compatible | |
1330 | * with 2.2.x, and also retains compatibility with earlier 2.4.x | |
1331 | * kernels... | |
1332 | */ | |
1333 | ||
1334 | info->mem_unit = 1; | |
1335 | info->totalram <<= bitcount; | |
1336 | info->freeram <<= bitcount; | |
1337 | info->sharedram <<= bitcount; | |
1338 | info->bufferram <<= bitcount; | |
1339 | info->totalswap <<= bitcount; | |
1340 | info->freeswap <<= bitcount; | |
1341 | info->totalhigh <<= bitcount; | |
1342 | info->freehigh <<= bitcount; | |
1343 | ||
1344 | out: | |
1345 | return 0; | |
1346 | } | |
1347 | ||
1348 | asmlinkage long sys_sysinfo(struct sysinfo __user *info) | |
1349 | { | |
1350 | struct sysinfo val; | |
1351 | ||
1352 | do_sysinfo(&val); | |
1353 | ||
1354 | if (copy_to_user(info, &val, sizeof(struct sysinfo))) | |
1355 | return -EFAULT; | |
1356 | ||
1357 | return 0; | |
1358 | } | |
1359 | ||
1360 | static int __cpuinit init_timers_cpu(int cpu) | |
1361 | { | |
1362 | int j; | |
1363 | struct tvec_base *base; | |
1364 | static char __cpuinitdata tvec_base_done[NR_CPUS]; | |
1365 | ||
1366 | if (!tvec_base_done[cpu]) { | |
1367 | static char boot_done; | |
1368 | ||
1369 | if (boot_done) { | |
1370 | /* | |
1371 | * The APs use this path later in boot | |
1372 | */ | |
1373 | base = kmalloc_node(sizeof(*base), | |
1374 | GFP_KERNEL | __GFP_ZERO, | |
1375 | cpu_to_node(cpu)); | |
1376 | if (!base) | |
1377 | return -ENOMEM; | |
1378 | ||
1379 | /* Make sure that tvec_base is 2 byte aligned */ | |
1380 | if (tbase_get_deferrable(base)) { | |
1381 | WARN_ON(1); | |
1382 | kfree(base); | |
1383 | return -ENOMEM; | |
1384 | } | |
1385 | per_cpu(tvec_bases, cpu) = base; | |
1386 | } else { | |
1387 | /* | |
1388 | * This is for the boot CPU - we use compile-time | |
1389 | * static initialisation because per-cpu memory isn't | |
1390 | * ready yet and because the memory allocators are not | |
1391 | * initialised either. | |
1392 | */ | |
1393 | boot_done = 1; | |
1394 | base = &boot_tvec_bases; | |
1395 | } | |
1396 | tvec_base_done[cpu] = 1; | |
1397 | } else { | |
1398 | base = per_cpu(tvec_bases, cpu); | |
1399 | } | |
1400 | ||
1401 | spin_lock_init(&base->lock); | |
1402 | ||
1403 | for (j = 0; j < TVN_SIZE; j++) { | |
1404 | INIT_LIST_HEAD(base->tv5.vec + j); | |
1405 | INIT_LIST_HEAD(base->tv4.vec + j); | |
1406 | INIT_LIST_HEAD(base->tv3.vec + j); | |
1407 | INIT_LIST_HEAD(base->tv2.vec + j); | |
1408 | } | |
1409 | for (j = 0; j < TVR_SIZE; j++) | |
1410 | INIT_LIST_HEAD(base->tv1.vec + j); | |
1411 | ||
1412 | base->timer_jiffies = jiffies; | |
1413 | return 0; | |
1414 | } | |
1415 | ||
1416 | #ifdef CONFIG_HOTPLUG_CPU | |
1417 | static void migrate_timer_list(struct tvec_base *new_base, struct list_head *head) | |
1418 | { | |
1419 | struct timer_list *timer; | |
1420 | ||
1421 | while (!list_empty(head)) { | |
1422 | timer = list_first_entry(head, struct timer_list, entry); | |
1423 | detach_timer(timer, 0); | |
1424 | timer_set_base(timer, new_base); | |
1425 | internal_add_timer(new_base, timer); | |
1426 | } | |
1427 | } | |
1428 | ||
1429 | static void __cpuinit migrate_timers(int cpu) | |
1430 | { | |
1431 | struct tvec_base *old_base; | |
1432 | struct tvec_base *new_base; | |
1433 | int i; | |
1434 | ||
1435 | BUG_ON(cpu_online(cpu)); | |
1436 | old_base = per_cpu(tvec_bases, cpu); | |
1437 | new_base = get_cpu_var(tvec_bases); | |
1438 | ||
1439 | local_irq_disable(); | |
1440 | spin_lock(&new_base->lock); | |
1441 | spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING); | |
1442 | ||
1443 | BUG_ON(old_base->running_timer); | |
1444 | ||
1445 | for (i = 0; i < TVR_SIZE; i++) | |
1446 | migrate_timer_list(new_base, old_base->tv1.vec + i); | |
1447 | for (i = 0; i < TVN_SIZE; i++) { | |
1448 | migrate_timer_list(new_base, old_base->tv2.vec + i); | |
1449 | migrate_timer_list(new_base, old_base->tv3.vec + i); | |
1450 | migrate_timer_list(new_base, old_base->tv4.vec + i); | |
1451 | migrate_timer_list(new_base, old_base->tv5.vec + i); | |
1452 | } | |
1453 | ||
1454 | spin_unlock(&old_base->lock); | |
1455 | spin_unlock(&new_base->lock); | |
1456 | local_irq_enable(); | |
1457 | put_cpu_var(tvec_bases); | |
1458 | } | |
1459 | #endif /* CONFIG_HOTPLUG_CPU */ | |
1460 | ||
1461 | static int __cpuinit timer_cpu_notify(struct notifier_block *self, | |
1462 | unsigned long action, void *hcpu) | |
1463 | { | |
1464 | long cpu = (long)hcpu; | |
1465 | switch(action) { | |
1466 | case CPU_UP_PREPARE: | |
1467 | case CPU_UP_PREPARE_FROZEN: | |
1468 | if (init_timers_cpu(cpu) < 0) | |
1469 | return NOTIFY_BAD; | |
1470 | break; | |
1471 | #ifdef CONFIG_HOTPLUG_CPU | |
1472 | case CPU_DEAD: | |
1473 | case CPU_DEAD_FROZEN: | |
1474 | migrate_timers(cpu); | |
1475 | break; | |
1476 | #endif | |
1477 | default: | |
1478 | break; | |
1479 | } | |
1480 | return NOTIFY_OK; | |
1481 | } | |
1482 | ||
1483 | static struct notifier_block __cpuinitdata timers_nb = { | |
1484 | .notifier_call = timer_cpu_notify, | |
1485 | }; | |
1486 | ||
1487 | ||
1488 | void __init init_timers(void) | |
1489 | { | |
1490 | int err = timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE, | |
1491 | (void *)(long)smp_processor_id()); | |
1492 | ||
1493 | init_timer_stats(); | |
1494 | ||
1495 | BUG_ON(err == NOTIFY_BAD); | |
1496 | register_cpu_notifier(&timers_nb); | |
1497 | open_softirq(TIMER_SOFTIRQ, run_timer_softirq); | |
1498 | } | |
1499 | ||
1500 | /** | |
1501 | * msleep - sleep safely even with waitqueue interruptions | |
1502 | * @msecs: Time in milliseconds to sleep for | |
1503 | */ | |
1504 | void msleep(unsigned int msecs) | |
1505 | { | |
1506 | unsigned long timeout = msecs_to_jiffies(msecs) + 1; | |
1507 | ||
1508 | while (timeout) | |
1509 | timeout = schedule_timeout_uninterruptible(timeout); | |
1510 | } | |
1511 | ||
1512 | EXPORT_SYMBOL(msleep); | |
1513 | ||
1514 | /** | |
1515 | * msleep_interruptible - sleep waiting for signals | |
1516 | * @msecs: Time in milliseconds to sleep for | |
1517 | */ | |
1518 | unsigned long msleep_interruptible(unsigned int msecs) | |
1519 | { | |
1520 | unsigned long timeout = msecs_to_jiffies(msecs) + 1; | |
1521 | ||
1522 | while (timeout && !signal_pending(current)) | |
1523 | timeout = schedule_timeout_interruptible(timeout); | |
1524 | return jiffies_to_msecs(timeout); | |
1525 | } | |
1526 | ||
1527 | EXPORT_SYMBOL(msleep_interruptible); |