Root/
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 | #include <linux/irq_work.h> |
41 | #include <linux/sched.h> |
42 | #include <linux/slab.h> |
43 | |
44 | #include <asm/uaccess.h> |
45 | #include <asm/unistd.h> |
46 | #include <asm/div64.h> |
47 | #include <asm/timex.h> |
48 | #include <asm/io.h> |
49 | |
50 | #define CREATE_TRACE_POINTS |
51 | #include <trace/events/timer.h> |
52 | |
53 | u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES; |
54 | |
55 | EXPORT_SYMBOL(jiffies_64); |
56 | |
57 | /* |
58 | * per-CPU timer vector definitions: |
59 | */ |
60 | #define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6) |
61 | #define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8) |
62 | #define TVN_SIZE (1 << TVN_BITS) |
63 | #define TVR_SIZE (1 << TVR_BITS) |
64 | #define TVN_MASK (TVN_SIZE - 1) |
65 | #define TVR_MASK (TVR_SIZE - 1) |
66 | |
67 | struct tvec { |
68 | struct list_head vec[TVN_SIZE]; |
69 | }; |
70 | |
71 | struct tvec_root { |
72 | struct list_head vec[TVR_SIZE]; |
73 | }; |
74 | |
75 | struct tvec_base { |
76 | spinlock_t lock; |
77 | struct timer_list *running_timer; |
78 | unsigned long timer_jiffies; |
79 | unsigned long next_timer; |
80 | struct tvec_root tv1; |
81 | struct tvec tv2; |
82 | struct tvec tv3; |
83 | struct tvec tv4; |
84 | struct tvec tv5; |
85 | } ____cacheline_aligned; |
86 | |
87 | struct tvec_base boot_tvec_bases; |
88 | EXPORT_SYMBOL(boot_tvec_bases); |
89 | static DEFINE_PER_CPU(struct tvec_base *, tvec_bases) = &boot_tvec_bases; |
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 = TBASE_MAKE_DEFERRED(timer->base); |
105 | } |
106 | |
107 | static inline void |
108 | timer_set_base(struct timer_list *timer, struct tvec_base *new_base) |
109 | { |
110 | timer->base = (struct tvec_base *)((unsigned long)(new_base) | |
111 | tbase_get_deferrable(timer->base)); |
112 | } |
113 | |
114 | static unsigned long round_jiffies_common(unsigned long j, int cpu, |
115 | bool force_up) |
116 | { |
117 | int rem; |
118 | unsigned long original = j; |
119 | |
120 | /* |
121 | * We don't want all cpus firing their timers at once hitting the |
122 | * same lock or cachelines, so we skew each extra cpu with an extra |
123 | * 3 jiffies. This 3 jiffies came originally from the mm/ code which |
124 | * already did this. |
125 | * The skew is done by adding 3*cpunr, then round, then subtract this |
126 | * extra offset again. |
127 | */ |
128 | j += cpu * 3; |
129 | |
130 | rem = j % HZ; |
131 | |
132 | /* |
133 | * If the target jiffie is just after a whole second (which can happen |
134 | * due to delays of the timer irq, long irq off times etc etc) then |
135 | * we should round down to the whole second, not up. Use 1/4th second |
136 | * as cutoff for this rounding as an extreme upper bound for this. |
137 | * But never round down if @force_up is set. |
138 | */ |
139 | if (rem < HZ/4 && !force_up) /* round down */ |
140 | j = j - rem; |
141 | else /* round up */ |
142 | j = j - rem + HZ; |
143 | |
144 | /* now that we have rounded, subtract the extra skew again */ |
145 | j -= cpu * 3; |
146 | |
147 | if (j <= jiffies) /* rounding ate our timeout entirely; */ |
148 | return original; |
149 | return j; |
150 | } |
151 | |
152 | /** |
153 | * __round_jiffies - function to round jiffies to a full second |
154 | * @j: the time in (absolute) jiffies that should be rounded |
155 | * @cpu: the processor number on which the timeout will happen |
156 | * |
157 | * __round_jiffies() rounds an absolute time in the future (in jiffies) |
158 | * up or down to (approximately) full seconds. This is useful for timers |
159 | * for which the exact time they fire does not matter too much, as long as |
160 | * they fire approximately every X seconds. |
161 | * |
162 | * By rounding these timers to whole seconds, all such timers will fire |
163 | * at the same time, rather than at various times spread out. The goal |
164 | * of this is to have the CPU wake up less, which saves power. |
165 | * |
166 | * The exact rounding is skewed for each processor to avoid all |
167 | * processors firing at the exact same time, which could lead |
168 | * to lock contention or spurious cache line bouncing. |
169 | * |
170 | * The return value is the rounded version of the @j parameter. |
171 | */ |
172 | unsigned long __round_jiffies(unsigned long j, int cpu) |
173 | { |
174 | return round_jiffies_common(j, cpu, false); |
175 | } |
176 | EXPORT_SYMBOL_GPL(__round_jiffies); |
177 | |
178 | /** |
179 | * __round_jiffies_relative - function to round jiffies to a full second |
180 | * @j: the time in (relative) jiffies that should be rounded |
181 | * @cpu: the processor number on which the timeout will happen |
182 | * |
183 | * __round_jiffies_relative() rounds a time delta in the future (in jiffies) |
184 | * up or down to (approximately) full seconds. This is useful for timers |
185 | * for which the exact time they fire does not matter too much, as long as |
186 | * they fire approximately every X seconds. |
187 | * |
188 | * By rounding these timers to whole seconds, all such timers will fire |
189 | * at the same time, rather than at various times spread out. The goal |
190 | * of this is to have the CPU wake up less, which saves power. |
191 | * |
192 | * The exact rounding is skewed for each processor to avoid all |
193 | * processors firing at the exact same time, which could lead |
194 | * to lock contention or spurious cache line bouncing. |
195 | * |
196 | * The return value is the rounded version of the @j parameter. |
197 | */ |
198 | unsigned long __round_jiffies_relative(unsigned long j, int cpu) |
199 | { |
200 | unsigned long j0 = jiffies; |
201 | |
202 | /* Use j0 because jiffies might change while we run */ |
203 | return round_jiffies_common(j + j0, cpu, false) - j0; |
204 | } |
205 | EXPORT_SYMBOL_GPL(__round_jiffies_relative); |
206 | |
207 | /** |
208 | * round_jiffies - function to round jiffies to a full second |
209 | * @j: the time in (absolute) jiffies that should be rounded |
210 | * |
211 | * round_jiffies() rounds an absolute time in the future (in jiffies) |
212 | * up or down to (approximately) full seconds. This is useful for timers |
213 | * for which the exact time they fire does not matter too much, as long as |
214 | * they fire approximately every X seconds. |
215 | * |
216 | * By rounding these timers to whole seconds, all such timers will fire |
217 | * at the same time, rather than at various times spread out. The goal |
218 | * of this is to have the CPU wake up less, which saves power. |
219 | * |
220 | * The return value is the rounded version of the @j parameter. |
221 | */ |
222 | unsigned long round_jiffies(unsigned long j) |
223 | { |
224 | return round_jiffies_common(j, raw_smp_processor_id(), false); |
225 | } |
226 | EXPORT_SYMBOL_GPL(round_jiffies); |
227 | |
228 | /** |
229 | * round_jiffies_relative - function to round jiffies to a full second |
230 | * @j: the time in (relative) jiffies that should be rounded |
231 | * |
232 | * round_jiffies_relative() rounds a time delta in the future (in jiffies) |
233 | * up or down to (approximately) full seconds. This is useful for timers |
234 | * for which the exact time they fire does not matter too much, as long as |
235 | * they fire approximately every X seconds. |
236 | * |
237 | * By rounding these timers to whole seconds, all such timers will fire |
238 | * at the same time, rather than at various times spread out. The goal |
239 | * of this is to have the CPU wake up less, which saves power. |
240 | * |
241 | * The return value is the rounded version of the @j parameter. |
242 | */ |
243 | unsigned long round_jiffies_relative(unsigned long j) |
244 | { |
245 | return __round_jiffies_relative(j, raw_smp_processor_id()); |
246 | } |
247 | EXPORT_SYMBOL_GPL(round_jiffies_relative); |
248 | |
249 | /** |
250 | * __round_jiffies_up - function to round jiffies up to a full second |
251 | * @j: the time in (absolute) jiffies that should be rounded |
252 | * @cpu: the processor number on which the timeout will happen |
253 | * |
254 | * This is the same as __round_jiffies() except that it will never |
255 | * round down. This is useful for timeouts for which the exact time |
256 | * of firing does not matter too much, as long as they don't fire too |
257 | * early. |
258 | */ |
259 | unsigned long __round_jiffies_up(unsigned long j, int cpu) |
260 | { |
261 | return round_jiffies_common(j, cpu, true); |
262 | } |
263 | EXPORT_SYMBOL_GPL(__round_jiffies_up); |
264 | |
265 | /** |
266 | * __round_jiffies_up_relative - function to round jiffies up to a full second |
267 | * @j: the time in (relative) jiffies that should be rounded |
268 | * @cpu: the processor number on which the timeout will happen |
269 | * |
270 | * This is the same as __round_jiffies_relative() except that it will never |
271 | * round down. This is useful for timeouts for which the exact time |
272 | * of firing does not matter too much, as long as they don't fire too |
273 | * early. |
274 | */ |
275 | unsigned long __round_jiffies_up_relative(unsigned long j, int cpu) |
276 | { |
277 | unsigned long j0 = jiffies; |
278 | |
279 | /* Use j0 because jiffies might change while we run */ |
280 | return round_jiffies_common(j + j0, cpu, true) - j0; |
281 | } |
282 | EXPORT_SYMBOL_GPL(__round_jiffies_up_relative); |
283 | |
284 | /** |
285 | * round_jiffies_up - function to round jiffies up to a full second |
286 | * @j: the time in (absolute) jiffies that should be rounded |
287 | * |
288 | * This is the same as round_jiffies() except that it will never |
289 | * round down. This is useful for timeouts for which the exact time |
290 | * of firing does not matter too much, as long as they don't fire too |
291 | * early. |
292 | */ |
293 | unsigned long round_jiffies_up(unsigned long j) |
294 | { |
295 | return round_jiffies_common(j, raw_smp_processor_id(), true); |
296 | } |
297 | EXPORT_SYMBOL_GPL(round_jiffies_up); |
298 | |
299 | /** |
300 | * round_jiffies_up_relative - function to round jiffies up to a full second |
301 | * @j: the time in (relative) jiffies that should be rounded |
302 | * |
303 | * This is the same as round_jiffies_relative() except that it will never |
304 | * round down. This is useful for timeouts for which the exact time |
305 | * of firing does not matter too much, as long as they don't fire too |
306 | * early. |
307 | */ |
308 | unsigned long round_jiffies_up_relative(unsigned long j) |
309 | { |
310 | return __round_jiffies_up_relative(j, raw_smp_processor_id()); |
311 | } |
312 | EXPORT_SYMBOL_GPL(round_jiffies_up_relative); |
313 | |
314 | /** |
315 | * set_timer_slack - set the allowed slack for a timer |
316 | * @timer: the timer to be modified |
317 | * @slack_hz: the amount of time (in jiffies) allowed for rounding |
318 | * |
319 | * Set the amount of time, in jiffies, that a certain timer has |
320 | * in terms of slack. By setting this value, the timer subsystem |
321 | * will schedule the actual timer somewhere between |
322 | * the time mod_timer() asks for, and that time plus the slack. |
323 | * |
324 | * By setting the slack to -1, a percentage of the delay is used |
325 | * instead. |
326 | */ |
327 | void set_timer_slack(struct timer_list *timer, int slack_hz) |
328 | { |
329 | timer->slack = slack_hz; |
330 | } |
331 | EXPORT_SYMBOL_GPL(set_timer_slack); |
332 | |
333 | static void internal_add_timer(struct tvec_base *base, struct timer_list *timer) |
334 | { |
335 | unsigned long expires = timer->expires; |
336 | unsigned long idx = expires - base->timer_jiffies; |
337 | struct list_head *vec; |
338 | |
339 | if (idx < TVR_SIZE) { |
340 | int i = expires & TVR_MASK; |
341 | vec = base->tv1.vec + i; |
342 | } else if (idx < 1 << (TVR_BITS + TVN_BITS)) { |
343 | int i = (expires >> TVR_BITS) & TVN_MASK; |
344 | vec = base->tv2.vec + i; |
345 | } else if (idx < 1 << (TVR_BITS + 2 * TVN_BITS)) { |
346 | int i = (expires >> (TVR_BITS + TVN_BITS)) & TVN_MASK; |
347 | vec = base->tv3.vec + i; |
348 | } else if (idx < 1 << (TVR_BITS + 3 * TVN_BITS)) { |
349 | int i = (expires >> (TVR_BITS + 2 * TVN_BITS)) & TVN_MASK; |
350 | vec = base->tv4.vec + i; |
351 | } else if ((signed long) idx < 0) { |
352 | /* |
353 | * Can happen if you add a timer with expires == jiffies, |
354 | * or you set a timer to go off in the past |
355 | */ |
356 | vec = base->tv1.vec + (base->timer_jiffies & TVR_MASK); |
357 | } else { |
358 | int i; |
359 | /* If the timeout is larger than 0xffffffff on 64-bit |
360 | * architectures then we use the maximum timeout: |
361 | */ |
362 | if (idx > 0xffffffffUL) { |
363 | idx = 0xffffffffUL; |
364 | expires = idx + base->timer_jiffies; |
365 | } |
366 | i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK; |
367 | vec = base->tv5.vec + i; |
368 | } |
369 | /* |
370 | * Timers are FIFO: |
371 | */ |
372 | list_add_tail(&timer->entry, vec); |
373 | } |
374 | |
375 | #ifdef CONFIG_TIMER_STATS |
376 | void __timer_stats_timer_set_start_info(struct timer_list *timer, void *addr) |
377 | { |
378 | if (timer->start_site) |
379 | return; |
380 | |
381 | timer->start_site = addr; |
382 | memcpy(timer->start_comm, current->comm, TASK_COMM_LEN); |
383 | timer->start_pid = current->pid; |
384 | } |
385 | |
386 | static void timer_stats_account_timer(struct timer_list *timer) |
387 | { |
388 | unsigned int flag = 0; |
389 | |
390 | if (likely(!timer->start_site)) |
391 | return; |
392 | if (unlikely(tbase_get_deferrable(timer->base))) |
393 | flag |= TIMER_STATS_FLAG_DEFERRABLE; |
394 | |
395 | timer_stats_update_stats(timer, timer->start_pid, timer->start_site, |
396 | timer->function, timer->start_comm, flag); |
397 | } |
398 | |
399 | #else |
400 | static void timer_stats_account_timer(struct timer_list *timer) {} |
401 | #endif |
402 | |
403 | #ifdef CONFIG_DEBUG_OBJECTS_TIMERS |
404 | |
405 | static struct debug_obj_descr timer_debug_descr; |
406 | |
407 | /* |
408 | * fixup_init is called when: |
409 | * - an active object is initialized |
410 | */ |
411 | static int timer_fixup_init(void *addr, enum debug_obj_state state) |
412 | { |
413 | struct timer_list *timer = addr; |
414 | |
415 | switch (state) { |
416 | case ODEBUG_STATE_ACTIVE: |
417 | del_timer_sync(timer); |
418 | debug_object_init(timer, &timer_debug_descr); |
419 | return 1; |
420 | default: |
421 | return 0; |
422 | } |
423 | } |
424 | |
425 | /* |
426 | * fixup_activate is called when: |
427 | * - an active object is activated |
428 | * - an unknown object is activated (might be a statically initialized object) |
429 | */ |
430 | static int timer_fixup_activate(void *addr, enum debug_obj_state state) |
431 | { |
432 | struct timer_list *timer = addr; |
433 | |
434 | switch (state) { |
435 | |
436 | case ODEBUG_STATE_NOTAVAILABLE: |
437 | /* |
438 | * This is not really a fixup. The timer was |
439 | * statically initialized. We just make sure that it |
440 | * is tracked in the object tracker. |
441 | */ |
442 | if (timer->entry.next == NULL && |
443 | timer->entry.prev == TIMER_ENTRY_STATIC) { |
444 | debug_object_init(timer, &timer_debug_descr); |
445 | debug_object_activate(timer, &timer_debug_descr); |
446 | return 0; |
447 | } else { |
448 | WARN_ON_ONCE(1); |
449 | } |
450 | return 0; |
451 | |
452 | case ODEBUG_STATE_ACTIVE: |
453 | WARN_ON(1); |
454 | |
455 | default: |
456 | return 0; |
457 | } |
458 | } |
459 | |
460 | /* |
461 | * fixup_free is called when: |
462 | * - an active object is freed |
463 | */ |
464 | static int timer_fixup_free(void *addr, enum debug_obj_state state) |
465 | { |
466 | struct timer_list *timer = addr; |
467 | |
468 | switch (state) { |
469 | case ODEBUG_STATE_ACTIVE: |
470 | del_timer_sync(timer); |
471 | debug_object_free(timer, &timer_debug_descr); |
472 | return 1; |
473 | default: |
474 | return 0; |
475 | } |
476 | } |
477 | |
478 | static struct debug_obj_descr timer_debug_descr = { |
479 | .name = "timer_list", |
480 | .fixup_init = timer_fixup_init, |
481 | .fixup_activate = timer_fixup_activate, |
482 | .fixup_free = timer_fixup_free, |
483 | }; |
484 | |
485 | static inline void debug_timer_init(struct timer_list *timer) |
486 | { |
487 | debug_object_init(timer, &timer_debug_descr); |
488 | } |
489 | |
490 | static inline void debug_timer_activate(struct timer_list *timer) |
491 | { |
492 | debug_object_activate(timer, &timer_debug_descr); |
493 | } |
494 | |
495 | static inline void debug_timer_deactivate(struct timer_list *timer) |
496 | { |
497 | debug_object_deactivate(timer, &timer_debug_descr); |
498 | } |
499 | |
500 | static inline void debug_timer_free(struct timer_list *timer) |
501 | { |
502 | debug_object_free(timer, &timer_debug_descr); |
503 | } |
504 | |
505 | static void __init_timer(struct timer_list *timer, |
506 | const char *name, |
507 | struct lock_class_key *key); |
508 | |
509 | void init_timer_on_stack_key(struct timer_list *timer, |
510 | const char *name, |
511 | struct lock_class_key *key) |
512 | { |
513 | debug_object_init_on_stack(timer, &timer_debug_descr); |
514 | __init_timer(timer, name, key); |
515 | } |
516 | EXPORT_SYMBOL_GPL(init_timer_on_stack_key); |
517 | |
518 | void destroy_timer_on_stack(struct timer_list *timer) |
519 | { |
520 | debug_object_free(timer, &timer_debug_descr); |
521 | } |
522 | EXPORT_SYMBOL_GPL(destroy_timer_on_stack); |
523 | |
524 | #else |
525 | static inline void debug_timer_init(struct timer_list *timer) { } |
526 | static inline void debug_timer_activate(struct timer_list *timer) { } |
527 | static inline void debug_timer_deactivate(struct timer_list *timer) { } |
528 | #endif |
529 | |
530 | static inline void debug_init(struct timer_list *timer) |
531 | { |
532 | debug_timer_init(timer); |
533 | trace_timer_init(timer); |
534 | } |
535 | |
536 | static inline void |
537 | debug_activate(struct timer_list *timer, unsigned long expires) |
538 | { |
539 | debug_timer_activate(timer); |
540 | trace_timer_start(timer, expires); |
541 | } |
542 | |
543 | static inline void debug_deactivate(struct timer_list *timer) |
544 | { |
545 | debug_timer_deactivate(timer); |
546 | trace_timer_cancel(timer); |
547 | } |
548 | |
549 | static void __init_timer(struct timer_list *timer, |
550 | const char *name, |
551 | struct lock_class_key *key) |
552 | { |
553 | timer->entry.next = NULL; |
554 | timer->base = __raw_get_cpu_var(tvec_bases); |
555 | timer->slack = -1; |
556 | #ifdef CONFIG_TIMER_STATS |
557 | timer->start_site = NULL; |
558 | timer->start_pid = -1; |
559 | memset(timer->start_comm, 0, TASK_COMM_LEN); |
560 | #endif |
561 | lockdep_init_map(&timer->lockdep_map, name, key, 0); |
562 | } |
563 | |
564 | void setup_deferrable_timer_on_stack_key(struct timer_list *timer, |
565 | const char *name, |
566 | struct lock_class_key *key, |
567 | void (*function)(unsigned long), |
568 | unsigned long data) |
569 | { |
570 | timer->function = function; |
571 | timer->data = data; |
572 | init_timer_on_stack_key(timer, name, key); |
573 | timer_set_deferrable(timer); |
574 | } |
575 | EXPORT_SYMBOL_GPL(setup_deferrable_timer_on_stack_key); |
576 | |
577 | /** |
578 | * init_timer_key - initialize a timer |
579 | * @timer: the timer to be initialized |
580 | * @name: name of the timer |
581 | * @key: lockdep class key of the fake lock used for tracking timer |
582 | * sync lock dependencies |
583 | * |
584 | * init_timer_key() must be done to a timer prior calling *any* of the |
585 | * other timer functions. |
586 | */ |
587 | void init_timer_key(struct timer_list *timer, |
588 | const char *name, |
589 | struct lock_class_key *key) |
590 | { |
591 | debug_init(timer); |
592 | __init_timer(timer, name, key); |
593 | } |
594 | EXPORT_SYMBOL(init_timer_key); |
595 | |
596 | void init_timer_deferrable_key(struct timer_list *timer, |
597 | const char *name, |
598 | struct lock_class_key *key) |
599 | { |
600 | init_timer_key(timer, name, key); |
601 | timer_set_deferrable(timer); |
602 | } |
603 | EXPORT_SYMBOL(init_timer_deferrable_key); |
604 | |
605 | static inline void detach_timer(struct timer_list *timer, |
606 | int clear_pending) |
607 | { |
608 | struct list_head *entry = &timer->entry; |
609 | |
610 | debug_deactivate(timer); |
611 | |
612 | __list_del(entry->prev, entry->next); |
613 | if (clear_pending) |
614 | entry->next = NULL; |
615 | entry->prev = LIST_POISON2; |
616 | } |
617 | |
618 | /* |
619 | * We are using hashed locking: holding per_cpu(tvec_bases).lock |
620 | * means that all timers which are tied to this base via timer->base are |
621 | * locked, and the base itself is locked too. |
622 | * |
623 | * So __run_timers/migrate_timers can safely modify all timers which could |
624 | * be found on ->tvX lists. |
625 | * |
626 | * When the timer's base is locked, and the timer removed from list, it is |
627 | * possible to set timer->base = NULL and drop the lock: the timer remains |
628 | * locked. |
629 | */ |
630 | static struct tvec_base *lock_timer_base(struct timer_list *timer, |
631 | unsigned long *flags) |
632 | __acquires(timer->base->lock) |
633 | { |
634 | struct tvec_base *base; |
635 | |
636 | for (;;) { |
637 | struct tvec_base *prelock_base = timer->base; |
638 | base = tbase_get_base(prelock_base); |
639 | if (likely(base != NULL)) { |
640 | spin_lock_irqsave(&base->lock, *flags); |
641 | if (likely(prelock_base == timer->base)) |
642 | return base; |
643 | /* The timer has migrated to another CPU */ |
644 | spin_unlock_irqrestore(&base->lock, *flags); |
645 | } |
646 | cpu_relax(); |
647 | } |
648 | } |
649 | |
650 | static inline int |
651 | __mod_timer(struct timer_list *timer, unsigned long expires, |
652 | bool pending_only, int pinned) |
653 | { |
654 | struct tvec_base *base, *new_base; |
655 | unsigned long flags; |
656 | int ret = 0 , cpu; |
657 | |
658 | timer_stats_timer_set_start_info(timer); |
659 | BUG_ON(!timer->function); |
660 | |
661 | base = lock_timer_base(timer, &flags); |
662 | |
663 | if (timer_pending(timer)) { |
664 | detach_timer(timer, 0); |
665 | if (timer->expires == base->next_timer && |
666 | !tbase_get_deferrable(timer->base)) |
667 | base->next_timer = base->timer_jiffies; |
668 | ret = 1; |
669 | } else { |
670 | if (pending_only) |
671 | goto out_unlock; |
672 | } |
673 | |
674 | debug_activate(timer, expires); |
675 | |
676 | cpu = smp_processor_id(); |
677 | |
678 | #if defined(CONFIG_NO_HZ) && defined(CONFIG_SMP) |
679 | if (!pinned && get_sysctl_timer_migration() && idle_cpu(cpu)) |
680 | cpu = get_nohz_timer_target(); |
681 | #endif |
682 | new_base = per_cpu(tvec_bases, cpu); |
683 | |
684 | if (base != new_base) { |
685 | /* |
686 | * We are trying to schedule the timer on the local CPU. |
687 | * However we can't change timer's base while it is running, |
688 | * otherwise del_timer_sync() can't detect that the timer's |
689 | * handler yet has not finished. This also guarantees that |
690 | * the timer is serialized wrt itself. |
691 | */ |
692 | if (likely(base->running_timer != timer)) { |
693 | /* See the comment in lock_timer_base() */ |
694 | timer_set_base(timer, NULL); |
695 | spin_unlock(&base->lock); |
696 | base = new_base; |
697 | spin_lock(&base->lock); |
698 | timer_set_base(timer, base); |
699 | } |
700 | } |
701 | |
702 | timer->expires = expires; |
703 | if (time_before(timer->expires, base->next_timer) && |
704 | !tbase_get_deferrable(timer->base)) |
705 | base->next_timer = timer->expires; |
706 | internal_add_timer(base, timer); |
707 | |
708 | out_unlock: |
709 | spin_unlock_irqrestore(&base->lock, flags); |
710 | |
711 | return ret; |
712 | } |
713 | |
714 | /** |
715 | * mod_timer_pending - modify a pending timer's timeout |
716 | * @timer: the pending timer to be modified |
717 | * @expires: new timeout in jiffies |
718 | * |
719 | * mod_timer_pending() is the same for pending timers as mod_timer(), |
720 | * but will not re-activate and modify already deleted timers. |
721 | * |
722 | * It is useful for unserialized use of timers. |
723 | */ |
724 | int mod_timer_pending(struct timer_list *timer, unsigned long expires) |
725 | { |
726 | return __mod_timer(timer, expires, true, TIMER_NOT_PINNED); |
727 | } |
728 | EXPORT_SYMBOL(mod_timer_pending); |
729 | |
730 | /* |
731 | * Decide where to put the timer while taking the slack into account |
732 | * |
733 | * Algorithm: |
734 | * 1) calculate the maximum (absolute) time |
735 | * 2) calculate the highest bit where the expires and new max are different |
736 | * 3) use this bit to make a mask |
737 | * 4) use the bitmask to round down the maximum time, so that all last |
738 | * bits are zeros |
739 | */ |
740 | static inline |
741 | unsigned long apply_slack(struct timer_list *timer, unsigned long expires) |
742 | { |
743 | unsigned long expires_limit, mask; |
744 | int bit; |
745 | |
746 | expires_limit = expires; |
747 | |
748 | if (timer->slack >= 0) { |
749 | expires_limit = expires + timer->slack; |
750 | } else { |
751 | unsigned long now = jiffies; |
752 | |
753 | /* No slack, if already expired else auto slack 0.4% */ |
754 | if (time_after(expires, now)) |
755 | expires_limit = expires + (expires - now)/256; |
756 | } |
757 | mask = expires ^ expires_limit; |
758 | if (mask == 0) |
759 | return expires; |
760 | |
761 | bit = find_last_bit(&mask, BITS_PER_LONG); |
762 | |
763 | mask = (1 << bit) - 1; |
764 | |
765 | expires_limit = expires_limit & ~(mask); |
766 | |
767 | return expires_limit; |
768 | } |
769 | |
770 | /** |
771 | * mod_timer - modify a timer's timeout |
772 | * @timer: the timer to be modified |
773 | * @expires: new timeout in jiffies |
774 | * |
775 | * mod_timer() is a more efficient way to update the expire field of an |
776 | * active timer (if the timer is inactive it will be activated) |
777 | * |
778 | * mod_timer(timer, expires) is equivalent to: |
779 | * |
780 | * del_timer(timer); timer->expires = expires; add_timer(timer); |
781 | * |
782 | * Note that if there are multiple unserialized concurrent users of the |
783 | * same timer, then mod_timer() is the only safe way to modify the timeout, |
784 | * since add_timer() cannot modify an already running timer. |
785 | * |
786 | * The function returns whether it has modified a pending timer or not. |
787 | * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an |
788 | * active timer returns 1.) |
789 | */ |
790 | int mod_timer(struct timer_list *timer, unsigned long expires) |
791 | { |
792 | /* |
793 | * This is a common optimization triggered by the |
794 | * networking code - if the timer is re-modified |
795 | * to be the same thing then just return: |
796 | */ |
797 | if (timer_pending(timer) && timer->expires == expires) |
798 | return 1; |
799 | |
800 | expires = apply_slack(timer, expires); |
801 | |
802 | return __mod_timer(timer, expires, false, TIMER_NOT_PINNED); |
803 | } |
804 | EXPORT_SYMBOL(mod_timer); |
805 | |
806 | /** |
807 | * mod_timer_pinned - modify a timer's timeout |
808 | * @timer: the timer to be modified |
809 | * @expires: new timeout in jiffies |
810 | * |
811 | * mod_timer_pinned() is a way to update the expire field of an |
812 | * active timer (if the timer is inactive it will be activated) |
813 | * and not allow the timer to be migrated to a different CPU. |
814 | * |
815 | * mod_timer_pinned(timer, expires) is equivalent to: |
816 | * |
817 | * del_timer(timer); timer->expires = expires; add_timer(timer); |
818 | */ |
819 | int mod_timer_pinned(struct timer_list *timer, unsigned long expires) |
820 | { |
821 | if (timer->expires == expires && timer_pending(timer)) |
822 | return 1; |
823 | |
824 | return __mod_timer(timer, expires, false, TIMER_PINNED); |
825 | } |
826 | EXPORT_SYMBOL(mod_timer_pinned); |
827 | |
828 | /** |
829 | * add_timer - start a timer |
830 | * @timer: the timer to be added |
831 | * |
832 | * The kernel will do a ->function(->data) callback from the |
833 | * timer interrupt at the ->expires point in the future. The |
834 | * current time is 'jiffies'. |
835 | * |
836 | * The timer's ->expires, ->function (and if the handler uses it, ->data) |
837 | * fields must be set prior calling this function. |
838 | * |
839 | * Timers with an ->expires field in the past will be executed in the next |
840 | * timer tick. |
841 | */ |
842 | void add_timer(struct timer_list *timer) |
843 | { |
844 | BUG_ON(timer_pending(timer)); |
845 | mod_timer(timer, timer->expires); |
846 | } |
847 | EXPORT_SYMBOL(add_timer); |
848 | |
849 | /** |
850 | * add_timer_on - start a timer on a particular CPU |
851 | * @timer: the timer to be added |
852 | * @cpu: the CPU to start it on |
853 | * |
854 | * This is not very scalable on SMP. Double adds are not possible. |
855 | */ |
856 | void add_timer_on(struct timer_list *timer, int cpu) |
857 | { |
858 | struct tvec_base *base = per_cpu(tvec_bases, cpu); |
859 | unsigned long flags; |
860 | |
861 | timer_stats_timer_set_start_info(timer); |
862 | BUG_ON(timer_pending(timer) || !timer->function); |
863 | spin_lock_irqsave(&base->lock, flags); |
864 | timer_set_base(timer, base); |
865 | debug_activate(timer, timer->expires); |
866 | if (time_before(timer->expires, base->next_timer) && |
867 | !tbase_get_deferrable(timer->base)) |
868 | base->next_timer = timer->expires; |
869 | internal_add_timer(base, timer); |
870 | /* |
871 | * Check whether the other CPU is idle and needs to be |
872 | * triggered to reevaluate the timer wheel when nohz is |
873 | * active. We are protected against the other CPU fiddling |
874 | * with the timer by holding the timer base lock. This also |
875 | * makes sure that a CPU on the way to idle can not evaluate |
876 | * the timer wheel. |
877 | */ |
878 | wake_up_idle_cpu(cpu); |
879 | spin_unlock_irqrestore(&base->lock, flags); |
880 | } |
881 | EXPORT_SYMBOL_GPL(add_timer_on); |
882 | |
883 | /** |
884 | * del_timer - deactive a timer. |
885 | * @timer: the timer to be deactivated |
886 | * |
887 | * del_timer() deactivates a timer - this works on both active and inactive |
888 | * timers. |
889 | * |
890 | * The function returns whether it has deactivated a pending timer or not. |
891 | * (ie. del_timer() of an inactive timer returns 0, del_timer() of an |
892 | * active timer returns 1.) |
893 | */ |
894 | int del_timer(struct timer_list *timer) |
895 | { |
896 | struct tvec_base *base; |
897 | unsigned long flags; |
898 | int ret = 0; |
899 | |
900 | timer_stats_timer_clear_start_info(timer); |
901 | if (timer_pending(timer)) { |
902 | base = lock_timer_base(timer, &flags); |
903 | if (timer_pending(timer)) { |
904 | detach_timer(timer, 1); |
905 | if (timer->expires == base->next_timer && |
906 | !tbase_get_deferrable(timer->base)) |
907 | base->next_timer = base->timer_jiffies; |
908 | ret = 1; |
909 | } |
910 | spin_unlock_irqrestore(&base->lock, flags); |
911 | } |
912 | |
913 | return ret; |
914 | } |
915 | EXPORT_SYMBOL(del_timer); |
916 | |
917 | /** |
918 | * try_to_del_timer_sync - Try to deactivate a timer |
919 | * @timer: timer do del |
920 | * |
921 | * This function tries to deactivate a timer. Upon successful (ret >= 0) |
922 | * exit the timer is not queued and the handler is not running on any CPU. |
923 | */ |
924 | int try_to_del_timer_sync(struct timer_list *timer) |
925 | { |
926 | struct tvec_base *base; |
927 | unsigned long flags; |
928 | int ret = -1; |
929 | |
930 | base = lock_timer_base(timer, &flags); |
931 | |
932 | if (base->running_timer == timer) |
933 | goto out; |
934 | |
935 | timer_stats_timer_clear_start_info(timer); |
936 | ret = 0; |
937 | if (timer_pending(timer)) { |
938 | detach_timer(timer, 1); |
939 | if (timer->expires == base->next_timer && |
940 | !tbase_get_deferrable(timer->base)) |
941 | base->next_timer = base->timer_jiffies; |
942 | ret = 1; |
943 | } |
944 | out: |
945 | spin_unlock_irqrestore(&base->lock, flags); |
946 | |
947 | return ret; |
948 | } |
949 | EXPORT_SYMBOL(try_to_del_timer_sync); |
950 | |
951 | #ifdef CONFIG_SMP |
952 | /** |
953 | * del_timer_sync - deactivate a timer and wait for the handler to finish. |
954 | * @timer: the timer to be deactivated |
955 | * |
956 | * This function only differs from del_timer() on SMP: besides deactivating |
957 | * the timer it also makes sure the handler has finished executing on other |
958 | * CPUs. |
959 | * |
960 | * Synchronization rules: Callers must prevent restarting of the timer, |
961 | * otherwise this function is meaningless. It must not be called from |
962 | * interrupt contexts. The caller must not hold locks which would prevent |
963 | * completion of the timer's handler. The timer's handler must not call |
964 | * add_timer_on(). Upon exit the timer is not queued and the handler is |
965 | * not running on any CPU. |
966 | * |
967 | * The function returns whether it has deactivated a pending timer or not. |
968 | */ |
969 | int del_timer_sync(struct timer_list *timer) |
970 | { |
971 | #ifdef CONFIG_LOCKDEP |
972 | unsigned long flags; |
973 | |
974 | local_irq_save(flags); |
975 | lock_map_acquire(&timer->lockdep_map); |
976 | lock_map_release(&timer->lockdep_map); |
977 | local_irq_restore(flags); |
978 | #endif |
979 | /* |
980 | * don't use it in hardirq context, because it |
981 | * could lead to deadlock. |
982 | */ |
983 | WARN_ON(in_irq()); |
984 | for (;;) { |
985 | int ret = try_to_del_timer_sync(timer); |
986 | if (ret >= 0) |
987 | return ret; |
988 | cpu_relax(); |
989 | } |
990 | } |
991 | EXPORT_SYMBOL(del_timer_sync); |
992 | #endif |
993 | |
994 | static int cascade(struct tvec_base *base, struct tvec *tv, int index) |
995 | { |
996 | /* cascade all the timers from tv up one level */ |
997 | struct timer_list *timer, *tmp; |
998 | struct list_head tv_list; |
999 | |
1000 | list_replace_init(tv->vec + index, &tv_list); |
1001 | |
1002 | /* |
1003 | * We are removing _all_ timers from the list, so we |
1004 | * don't have to detach them individually. |
1005 | */ |
1006 | list_for_each_entry_safe(timer, tmp, &tv_list, entry) { |
1007 | BUG_ON(tbase_get_base(timer->base) != base); |
1008 | internal_add_timer(base, timer); |
1009 | } |
1010 | |
1011 | return index; |
1012 | } |
1013 | |
1014 | static void call_timer_fn(struct timer_list *timer, void (*fn)(unsigned long), |
1015 | unsigned long data) |
1016 | { |
1017 | int preempt_count = preempt_count(); |
1018 | |
1019 | #ifdef CONFIG_LOCKDEP |
1020 | /* |
1021 | * It is permissible to free the timer from inside the |
1022 | * function that is called from it, this we need to take into |
1023 | * account for lockdep too. To avoid bogus "held lock freed" |
1024 | * warnings as well as problems when looking into |
1025 | * timer->lockdep_map, make a copy and use that here. |
1026 | */ |
1027 | struct lockdep_map lockdep_map = timer->lockdep_map; |
1028 | #endif |
1029 | /* |
1030 | * Couple the lock chain with the lock chain at |
1031 | * del_timer_sync() by acquiring the lock_map around the fn() |
1032 | * call here and in del_timer_sync(). |
1033 | */ |
1034 | lock_map_acquire(&lockdep_map); |
1035 | |
1036 | trace_timer_expire_entry(timer); |
1037 | fn(data); |
1038 | trace_timer_expire_exit(timer); |
1039 | |
1040 | lock_map_release(&lockdep_map); |
1041 | |
1042 | if (preempt_count != preempt_count()) { |
1043 | WARN_ONCE(1, "timer: %pF preempt leak: %08x -> %08x\n", |
1044 | fn, preempt_count, preempt_count()); |
1045 | /* |
1046 | * Restore the preempt count. That gives us a decent |
1047 | * chance to survive and extract information. If the |
1048 | * callback kept a lock held, bad luck, but not worse |
1049 | * than the BUG() we had. |
1050 | */ |
1051 | preempt_count() = preempt_count; |
1052 | } |
1053 | } |
1054 | |
1055 | #define INDEX(N) ((base->timer_jiffies >> (TVR_BITS + (N) * TVN_BITS)) & TVN_MASK) |
1056 | |
1057 | /** |
1058 | * __run_timers - run all expired timers (if any) on this CPU. |
1059 | * @base: the timer vector to be processed. |
1060 | * |
1061 | * This function cascades all vectors and executes all expired timer |
1062 | * vectors. |
1063 | */ |
1064 | static inline void __run_timers(struct tvec_base *base) |
1065 | { |
1066 | struct timer_list *timer; |
1067 | |
1068 | spin_lock_irq(&base->lock); |
1069 | while (time_after_eq(jiffies, base->timer_jiffies)) { |
1070 | struct list_head work_list; |
1071 | struct list_head *head = &work_list; |
1072 | int index = base->timer_jiffies & TVR_MASK; |
1073 | |
1074 | /* |
1075 | * Cascade timers: |
1076 | */ |
1077 | if (!index && |
1078 | (!cascade(base, &base->tv2, INDEX(0))) && |
1079 | (!cascade(base, &base->tv3, INDEX(1))) && |
1080 | !cascade(base, &base->tv4, INDEX(2))) |
1081 | cascade(base, &base->tv5, INDEX(3)); |
1082 | ++base->timer_jiffies; |
1083 | list_replace_init(base->tv1.vec + index, &work_list); |
1084 | while (!list_empty(head)) { |
1085 | void (*fn)(unsigned long); |
1086 | unsigned long data; |
1087 | |
1088 | timer = list_first_entry(head, struct timer_list,entry); |
1089 | fn = timer->function; |
1090 | data = timer->data; |
1091 | |
1092 | timer_stats_account_timer(timer); |
1093 | |
1094 | base->running_timer = timer; |
1095 | detach_timer(timer, 1); |
1096 | |
1097 | spin_unlock_irq(&base->lock); |
1098 | call_timer_fn(timer, fn, data); |
1099 | spin_lock_irq(&base->lock); |
1100 | } |
1101 | } |
1102 | base->running_timer = NULL; |
1103 | spin_unlock_irq(&base->lock); |
1104 | } |
1105 | |
1106 | #ifdef CONFIG_NO_HZ |
1107 | /* |
1108 | * Find out when the next timer event is due to happen. This |
1109 | * is used on S/390 to stop all activity when a CPU is idle. |
1110 | * This function needs to be called with interrupts disabled. |
1111 | */ |
1112 | static unsigned long __next_timer_interrupt(struct tvec_base *base) |
1113 | { |
1114 | unsigned long timer_jiffies = base->timer_jiffies; |
1115 | unsigned long expires = timer_jiffies + NEXT_TIMER_MAX_DELTA; |
1116 | int index, slot, array, found = 0; |
1117 | struct timer_list *nte; |
1118 | struct tvec *varray[4]; |
1119 | |
1120 | /* Look for timer events in tv1. */ |
1121 | index = slot = timer_jiffies & TVR_MASK; |
1122 | do { |
1123 | list_for_each_entry(nte, base->tv1.vec + slot, entry) { |
1124 | if (tbase_get_deferrable(nte->base)) |
1125 | continue; |
1126 | |
1127 | found = 1; |
1128 | expires = nte->expires; |
1129 | /* Look at the cascade bucket(s)? */ |
1130 | if (!index || slot < index) |
1131 | goto cascade; |
1132 | return expires; |
1133 | } |
1134 | slot = (slot + 1) & TVR_MASK; |
1135 | } while (slot != index); |
1136 | |
1137 | cascade: |
1138 | /* Calculate the next cascade event */ |
1139 | if (index) |
1140 | timer_jiffies += TVR_SIZE - index; |
1141 | timer_jiffies >>= TVR_BITS; |
1142 | |
1143 | /* Check tv2-tv5. */ |
1144 | varray[0] = &base->tv2; |
1145 | varray[1] = &base->tv3; |
1146 | varray[2] = &base->tv4; |
1147 | varray[3] = &base->tv5; |
1148 | |
1149 | for (array = 0; array < 4; array++) { |
1150 | struct tvec *varp = varray[array]; |
1151 | |
1152 | index = slot = timer_jiffies & TVN_MASK; |
1153 | do { |
1154 | list_for_each_entry(nte, varp->vec + slot, entry) { |
1155 | if (tbase_get_deferrable(nte->base)) |
1156 | continue; |
1157 | |
1158 | found = 1; |
1159 | if (time_before(nte->expires, expires)) |
1160 | expires = nte->expires; |
1161 | } |
1162 | /* |
1163 | * Do we still search for the first timer or are |
1164 | * we looking up the cascade buckets ? |
1165 | */ |
1166 | if (found) { |
1167 | /* Look at the cascade bucket(s)? */ |
1168 | if (!index || slot < index) |
1169 | break; |
1170 | return expires; |
1171 | } |
1172 | slot = (slot + 1) & TVN_MASK; |
1173 | } while (slot != index); |
1174 | |
1175 | if (index) |
1176 | timer_jiffies += TVN_SIZE - index; |
1177 | timer_jiffies >>= TVN_BITS; |
1178 | } |
1179 | return expires; |
1180 | } |
1181 | |
1182 | /* |
1183 | * Check, if the next hrtimer event is before the next timer wheel |
1184 | * event: |
1185 | */ |
1186 | static unsigned long cmp_next_hrtimer_event(unsigned long now, |
1187 | unsigned long expires) |
1188 | { |
1189 | ktime_t hr_delta = hrtimer_get_next_event(); |
1190 | struct timespec tsdelta; |
1191 | unsigned long delta; |
1192 | |
1193 | if (hr_delta.tv64 == KTIME_MAX) |
1194 | return expires; |
1195 | |
1196 | /* |
1197 | * Expired timer available, let it expire in the next tick |
1198 | */ |
1199 | if (hr_delta.tv64 <= 0) |
1200 | return now + 1; |
1201 | |
1202 | tsdelta = ktime_to_timespec(hr_delta); |
1203 | delta = timespec_to_jiffies(&tsdelta); |
1204 | |
1205 | /* |
1206 | * Limit the delta to the max value, which is checked in |
1207 | * tick_nohz_stop_sched_tick(): |
1208 | */ |
1209 | if (delta > NEXT_TIMER_MAX_DELTA) |
1210 | delta = NEXT_TIMER_MAX_DELTA; |
1211 | |
1212 | /* |
1213 | * Take rounding errors in to account and make sure, that it |
1214 | * expires in the next tick. Otherwise we go into an endless |
1215 | * ping pong due to tick_nohz_stop_sched_tick() retriggering |
1216 | * the timer softirq |
1217 | */ |
1218 | if (delta < 1) |
1219 | delta = 1; |
1220 | now += delta; |
1221 | if (time_before(now, expires)) |
1222 | return now; |
1223 | return expires; |
1224 | } |
1225 | |
1226 | /** |
1227 | * get_next_timer_interrupt - return the jiffy of the next pending timer |
1228 | * @now: current time (in jiffies) |
1229 | */ |
1230 | unsigned long get_next_timer_interrupt(unsigned long now) |
1231 | { |
1232 | struct tvec_base *base = __this_cpu_read(tvec_bases); |
1233 | unsigned long expires; |
1234 | |
1235 | /* |
1236 | * Pretend that there is no timer pending if the cpu is offline. |
1237 | * Possible pending timers will be migrated later to an active cpu. |
1238 | */ |
1239 | if (cpu_is_offline(smp_processor_id())) |
1240 | return now + NEXT_TIMER_MAX_DELTA; |
1241 | spin_lock(&base->lock); |
1242 | if (time_before_eq(base->next_timer, base->timer_jiffies)) |
1243 | base->next_timer = __next_timer_interrupt(base); |
1244 | expires = base->next_timer; |
1245 | spin_unlock(&base->lock); |
1246 | |
1247 | if (time_before_eq(expires, now)) |
1248 | return now; |
1249 | |
1250 | return cmp_next_hrtimer_event(now, expires); |
1251 | } |
1252 | #endif |
1253 | |
1254 | /* |
1255 | * Called from the timer interrupt handler to charge one tick to the current |
1256 | * process. user_tick is 1 if the tick is user time, 0 for system. |
1257 | */ |
1258 | void update_process_times(int user_tick) |
1259 | { |
1260 | struct task_struct *p = current; |
1261 | int cpu = smp_processor_id(); |
1262 | |
1263 | /* Note: this timer irq context must be accounted for as well. */ |
1264 | account_process_tick(p, user_tick); |
1265 | run_local_timers(); |
1266 | rcu_check_callbacks(cpu, user_tick); |
1267 | printk_tick(); |
1268 | #ifdef CONFIG_IRQ_WORK |
1269 | if (in_irq()) |
1270 | irq_work_run(); |
1271 | #endif |
1272 | scheduler_tick(); |
1273 | run_posix_cpu_timers(p); |
1274 | } |
1275 | |
1276 | /* |
1277 | * This function runs timers and the timer-tq in bottom half context. |
1278 | */ |
1279 | static void run_timer_softirq(struct softirq_action *h) |
1280 | { |
1281 | struct tvec_base *base = __this_cpu_read(tvec_bases); |
1282 | |
1283 | hrtimer_run_pending(); |
1284 | |
1285 | if (time_after_eq(jiffies, base->timer_jiffies)) |
1286 | __run_timers(base); |
1287 | } |
1288 | |
1289 | /* |
1290 | * Called by the local, per-CPU timer interrupt on SMP. |
1291 | */ |
1292 | void run_local_timers(void) |
1293 | { |
1294 | hrtimer_run_queues(); |
1295 | raise_softirq(TIMER_SOFTIRQ); |
1296 | } |
1297 | |
1298 | /* |
1299 | * The 64-bit jiffies value is not atomic - you MUST NOT read it |
1300 | * without sampling the sequence number in xtime_lock. |
1301 | * jiffies is defined in the linker script... |
1302 | */ |
1303 | |
1304 | void do_timer(unsigned long ticks) |
1305 | { |
1306 | jiffies_64 += ticks; |
1307 | update_wall_time(); |
1308 | calc_global_load(ticks); |
1309 | } |
1310 | |
1311 | #ifdef __ARCH_WANT_SYS_ALARM |
1312 | |
1313 | /* |
1314 | * For backwards compatibility? This can be done in libc so Alpha |
1315 | * and all newer ports shouldn't need it. |
1316 | */ |
1317 | SYSCALL_DEFINE1(alarm, unsigned int, seconds) |
1318 | { |
1319 | return alarm_setitimer(seconds); |
1320 | } |
1321 | |
1322 | #endif |
1323 | |
1324 | #ifndef __alpha__ |
1325 | |
1326 | /* |
1327 | * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this |
1328 | * should be moved into arch/i386 instead? |
1329 | */ |
1330 | |
1331 | /** |
1332 | * sys_getpid - return the thread group id of the current process |
1333 | * |
1334 | * Note, despite the name, this returns the tgid not the pid. The tgid and |
1335 | * the pid are identical unless CLONE_THREAD was specified on clone() in |
1336 | * which case the tgid is the same in all threads of the same group. |
1337 | * |
1338 | * This is SMP safe as current->tgid does not change. |
1339 | */ |
1340 | SYSCALL_DEFINE0(getpid) |
1341 | { |
1342 | return task_tgid_vnr(current); |
1343 | } |
1344 | |
1345 | /* |
1346 | * Accessing ->real_parent is not SMP-safe, it could |
1347 | * change from under us. However, we can use a stale |
1348 | * value of ->real_parent under rcu_read_lock(), see |
1349 | * release_task()->call_rcu(delayed_put_task_struct). |
1350 | */ |
1351 | SYSCALL_DEFINE0(getppid) |
1352 | { |
1353 | int pid; |
1354 | |
1355 | rcu_read_lock(); |
1356 | pid = task_tgid_vnr(current->real_parent); |
1357 | rcu_read_unlock(); |
1358 | |
1359 | return pid; |
1360 | } |
1361 | |
1362 | SYSCALL_DEFINE0(getuid) |
1363 | { |
1364 | /* Only we change this so SMP safe */ |
1365 | return current_uid(); |
1366 | } |
1367 | |
1368 | SYSCALL_DEFINE0(geteuid) |
1369 | { |
1370 | /* Only we change this so SMP safe */ |
1371 | return current_euid(); |
1372 | } |
1373 | |
1374 | SYSCALL_DEFINE0(getgid) |
1375 | { |
1376 | /* Only we change this so SMP safe */ |
1377 | return current_gid(); |
1378 | } |
1379 | |
1380 | SYSCALL_DEFINE0(getegid) |
1381 | { |
1382 | /* Only we change this so SMP safe */ |
1383 | return current_egid(); |
1384 | } |
1385 | |
1386 | #endif |
1387 | |
1388 | static void process_timeout(unsigned long __data) |
1389 | { |
1390 | wake_up_process((struct task_struct *)__data); |
1391 | } |
1392 | |
1393 | /** |
1394 | * schedule_timeout - sleep until timeout |
1395 | * @timeout: timeout value in jiffies |
1396 | * |
1397 | * Make the current task sleep until @timeout jiffies have |
1398 | * elapsed. The routine will return immediately unless |
1399 | * the current task state has been set (see set_current_state()). |
1400 | * |
1401 | * You can set the task state as follows - |
1402 | * |
1403 | * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to |
1404 | * pass before the routine returns. The routine will return 0 |
1405 | * |
1406 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is |
1407 | * delivered to the current task. In this case the remaining time |
1408 | * in jiffies will be returned, or 0 if the timer expired in time |
1409 | * |
1410 | * The current task state is guaranteed to be TASK_RUNNING when this |
1411 | * routine returns. |
1412 | * |
1413 | * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule |
1414 | * the CPU away without a bound on the timeout. In this case the return |
1415 | * value will be %MAX_SCHEDULE_TIMEOUT. |
1416 | * |
1417 | * In all cases the return value is guaranteed to be non-negative. |
1418 | */ |
1419 | signed long __sched schedule_timeout(signed long timeout) |
1420 | { |
1421 | struct timer_list timer; |
1422 | unsigned long expire; |
1423 | |
1424 | switch (timeout) |
1425 | { |
1426 | case MAX_SCHEDULE_TIMEOUT: |
1427 | /* |
1428 | * These two special cases are useful to be comfortable |
1429 | * in the caller. Nothing more. We could take |
1430 | * MAX_SCHEDULE_TIMEOUT from one of the negative value |
1431 | * but I' d like to return a valid offset (>=0) to allow |
1432 | * the caller to do everything it want with the retval. |
1433 | */ |
1434 | schedule(); |
1435 | goto out; |
1436 | default: |
1437 | /* |
1438 | * Another bit of PARANOID. Note that the retval will be |
1439 | * 0 since no piece of kernel is supposed to do a check |
1440 | * for a negative retval of schedule_timeout() (since it |
1441 | * should never happens anyway). You just have the printk() |
1442 | * that will tell you if something is gone wrong and where. |
1443 | */ |
1444 | if (timeout < 0) { |
1445 | printk(KERN_ERR "schedule_timeout: wrong timeout " |
1446 | "value %lx\n", timeout); |
1447 | dump_stack(); |
1448 | current->state = TASK_RUNNING; |
1449 | goto out; |
1450 | } |
1451 | } |
1452 | |
1453 | expire = timeout + jiffies; |
1454 | |
1455 | setup_timer_on_stack(&timer, process_timeout, (unsigned long)current); |
1456 | __mod_timer(&timer, expire, false, TIMER_NOT_PINNED); |
1457 | schedule(); |
1458 | del_singleshot_timer_sync(&timer); |
1459 | |
1460 | /* Remove the timer from the object tracker */ |
1461 | destroy_timer_on_stack(&timer); |
1462 | |
1463 | timeout = expire - jiffies; |
1464 | |
1465 | out: |
1466 | return timeout < 0 ? 0 : timeout; |
1467 | } |
1468 | EXPORT_SYMBOL(schedule_timeout); |
1469 | |
1470 | /* |
1471 | * We can use __set_current_state() here because schedule_timeout() calls |
1472 | * schedule() unconditionally. |
1473 | */ |
1474 | signed long __sched schedule_timeout_interruptible(signed long timeout) |
1475 | { |
1476 | __set_current_state(TASK_INTERRUPTIBLE); |
1477 | return schedule_timeout(timeout); |
1478 | } |
1479 | EXPORT_SYMBOL(schedule_timeout_interruptible); |
1480 | |
1481 | signed long __sched schedule_timeout_killable(signed long timeout) |
1482 | { |
1483 | __set_current_state(TASK_KILLABLE); |
1484 | return schedule_timeout(timeout); |
1485 | } |
1486 | EXPORT_SYMBOL(schedule_timeout_killable); |
1487 | |
1488 | signed long __sched schedule_timeout_uninterruptible(signed long timeout) |
1489 | { |
1490 | __set_current_state(TASK_UNINTERRUPTIBLE); |
1491 | return schedule_timeout(timeout); |
1492 | } |
1493 | EXPORT_SYMBOL(schedule_timeout_uninterruptible); |
1494 | |
1495 | /* Thread ID - the internal kernel "pid" */ |
1496 | SYSCALL_DEFINE0(gettid) |
1497 | { |
1498 | return task_pid_vnr(current); |
1499 | } |
1500 | |
1501 | /** |
1502 | * do_sysinfo - fill in sysinfo struct |
1503 | * @info: pointer to buffer to fill |
1504 | */ |
1505 | int do_sysinfo(struct sysinfo *info) |
1506 | { |
1507 | unsigned long mem_total, sav_total; |
1508 | unsigned int mem_unit, bitcount; |
1509 | struct timespec tp; |
1510 | |
1511 | memset(info, 0, sizeof(struct sysinfo)); |
1512 | |
1513 | ktime_get_ts(&tp); |
1514 | monotonic_to_bootbased(&tp); |
1515 | info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0); |
1516 | |
1517 | get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT); |
1518 | |
1519 | info->procs = nr_threads; |
1520 | |
1521 | si_meminfo(info); |
1522 | si_swapinfo(info); |
1523 | |
1524 | /* |
1525 | * If the sum of all the available memory (i.e. ram + swap) |
1526 | * is less than can be stored in a 32 bit unsigned long then |
1527 | * we can be binary compatible with 2.2.x kernels. If not, |
1528 | * well, in that case 2.2.x was broken anyways... |
1529 | * |
1530 | * -Erik Andersen <andersee@debian.org> |
1531 | */ |
1532 | |
1533 | mem_total = info->totalram + info->totalswap; |
1534 | if (mem_total < info->totalram || mem_total < info->totalswap) |
1535 | goto out; |
1536 | bitcount = 0; |
1537 | mem_unit = info->mem_unit; |
1538 | while (mem_unit > 1) { |
1539 | bitcount++; |
1540 | mem_unit >>= 1; |
1541 | sav_total = mem_total; |
1542 | mem_total <<= 1; |
1543 | if (mem_total < sav_total) |
1544 | goto out; |
1545 | } |
1546 | |
1547 | /* |
1548 | * If mem_total did not overflow, multiply all memory values by |
1549 | * info->mem_unit and set it to 1. This leaves things compatible |
1550 | * with 2.2.x, and also retains compatibility with earlier 2.4.x |
1551 | * kernels... |
1552 | */ |
1553 | |
1554 | info->mem_unit = 1; |
1555 | info->totalram <<= bitcount; |
1556 | info->freeram <<= bitcount; |
1557 | info->sharedram <<= bitcount; |
1558 | info->bufferram <<= bitcount; |
1559 | info->totalswap <<= bitcount; |
1560 | info->freeswap <<= bitcount; |
1561 | info->totalhigh <<= bitcount; |
1562 | info->freehigh <<= bitcount; |
1563 | |
1564 | out: |
1565 | return 0; |
1566 | } |
1567 | |
1568 | SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info) |
1569 | { |
1570 | struct sysinfo val; |
1571 | |
1572 | do_sysinfo(&val); |
1573 | |
1574 | if (copy_to_user(info, &val, sizeof(struct sysinfo))) |
1575 | return -EFAULT; |
1576 | |
1577 | return 0; |
1578 | } |
1579 | |
1580 | static int __cpuinit init_timers_cpu(int cpu) |
1581 | { |
1582 | int j; |
1583 | struct tvec_base *base; |
1584 | static char __cpuinitdata tvec_base_done[NR_CPUS]; |
1585 | |
1586 | if (!tvec_base_done[cpu]) { |
1587 | static char boot_done; |
1588 | |
1589 | if (boot_done) { |
1590 | /* |
1591 | * The APs use this path later in boot |
1592 | */ |
1593 | base = kmalloc_node(sizeof(*base), |
1594 | GFP_KERNEL | __GFP_ZERO, |
1595 | cpu_to_node(cpu)); |
1596 | if (!base) |
1597 | return -ENOMEM; |
1598 | |
1599 | /* Make sure that tvec_base is 2 byte aligned */ |
1600 | if (tbase_get_deferrable(base)) { |
1601 | WARN_ON(1); |
1602 | kfree(base); |
1603 | return -ENOMEM; |
1604 | } |
1605 | per_cpu(tvec_bases, cpu) = base; |
1606 | } else { |
1607 | /* |
1608 | * This is for the boot CPU - we use compile-time |
1609 | * static initialisation because per-cpu memory isn't |
1610 | * ready yet and because the memory allocators are not |
1611 | * initialised either. |
1612 | */ |
1613 | boot_done = 1; |
1614 | base = &boot_tvec_bases; |
1615 | } |
1616 | tvec_base_done[cpu] = 1; |
1617 | } else { |
1618 | base = per_cpu(tvec_bases, cpu); |
1619 | } |
1620 | |
1621 | spin_lock_init(&base->lock); |
1622 | |
1623 | for (j = 0; j < TVN_SIZE; j++) { |
1624 | INIT_LIST_HEAD(base->tv5.vec + j); |
1625 | INIT_LIST_HEAD(base->tv4.vec + j); |
1626 | INIT_LIST_HEAD(base->tv3.vec + j); |
1627 | INIT_LIST_HEAD(base->tv2.vec + j); |
1628 | } |
1629 | for (j = 0; j < TVR_SIZE; j++) |
1630 | INIT_LIST_HEAD(base->tv1.vec + j); |
1631 | |
1632 | base->timer_jiffies = jiffies; |
1633 | base->next_timer = base->timer_jiffies; |
1634 | return 0; |
1635 | } |
1636 | |
1637 | #ifdef CONFIG_HOTPLUG_CPU |
1638 | static void migrate_timer_list(struct tvec_base *new_base, struct list_head *head) |
1639 | { |
1640 | struct timer_list *timer; |
1641 | |
1642 | while (!list_empty(head)) { |
1643 | timer = list_first_entry(head, struct timer_list, entry); |
1644 | detach_timer(timer, 0); |
1645 | timer_set_base(timer, new_base); |
1646 | if (time_before(timer->expires, new_base->next_timer) && |
1647 | !tbase_get_deferrable(timer->base)) |
1648 | new_base->next_timer = timer->expires; |
1649 | internal_add_timer(new_base, timer); |
1650 | } |
1651 | } |
1652 | |
1653 | static void __cpuinit migrate_timers(int cpu) |
1654 | { |
1655 | struct tvec_base *old_base; |
1656 | struct tvec_base *new_base; |
1657 | int i; |
1658 | |
1659 | BUG_ON(cpu_online(cpu)); |
1660 | old_base = per_cpu(tvec_bases, cpu); |
1661 | new_base = get_cpu_var(tvec_bases); |
1662 | /* |
1663 | * The caller is globally serialized and nobody else |
1664 | * takes two locks at once, deadlock is not possible. |
1665 | */ |
1666 | spin_lock_irq(&new_base->lock); |
1667 | spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING); |
1668 | |
1669 | BUG_ON(old_base->running_timer); |
1670 | |
1671 | for (i = 0; i < TVR_SIZE; i++) |
1672 | migrate_timer_list(new_base, old_base->tv1.vec + i); |
1673 | for (i = 0; i < TVN_SIZE; i++) { |
1674 | migrate_timer_list(new_base, old_base->tv2.vec + i); |
1675 | migrate_timer_list(new_base, old_base->tv3.vec + i); |
1676 | migrate_timer_list(new_base, old_base->tv4.vec + i); |
1677 | migrate_timer_list(new_base, old_base->tv5.vec + i); |
1678 | } |
1679 | |
1680 | spin_unlock(&old_base->lock); |
1681 | spin_unlock_irq(&new_base->lock); |
1682 | put_cpu_var(tvec_bases); |
1683 | } |
1684 | #endif /* CONFIG_HOTPLUG_CPU */ |
1685 | |
1686 | static int __cpuinit timer_cpu_notify(struct notifier_block *self, |
1687 | unsigned long action, void *hcpu) |
1688 | { |
1689 | long cpu = (long)hcpu; |
1690 | int err; |
1691 | |
1692 | switch(action) { |
1693 | case CPU_UP_PREPARE: |
1694 | case CPU_UP_PREPARE_FROZEN: |
1695 | err = init_timers_cpu(cpu); |
1696 | if (err < 0) |
1697 | return notifier_from_errno(err); |
1698 | break; |
1699 | #ifdef CONFIG_HOTPLUG_CPU |
1700 | case CPU_DEAD: |
1701 | case CPU_DEAD_FROZEN: |
1702 | migrate_timers(cpu); |
1703 | break; |
1704 | #endif |
1705 | default: |
1706 | break; |
1707 | } |
1708 | return NOTIFY_OK; |
1709 | } |
1710 | |
1711 | static struct notifier_block __cpuinitdata timers_nb = { |
1712 | .notifier_call = timer_cpu_notify, |
1713 | }; |
1714 | |
1715 | |
1716 | void __init init_timers(void) |
1717 | { |
1718 | int err = timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE, |
1719 | (void *)(long)smp_processor_id()); |
1720 | |
1721 | init_timer_stats(); |
1722 | |
1723 | BUG_ON(err != NOTIFY_OK); |
1724 | register_cpu_notifier(&timers_nb); |
1725 | open_softirq(TIMER_SOFTIRQ, run_timer_softirq); |
1726 | } |
1727 | |
1728 | /** |
1729 | * msleep - sleep safely even with waitqueue interruptions |
1730 | * @msecs: Time in milliseconds to sleep for |
1731 | */ |
1732 | void msleep(unsigned int msecs) |
1733 | { |
1734 | unsigned long timeout = msecs_to_jiffies(msecs) + 1; |
1735 | |
1736 | while (timeout) |
1737 | timeout = schedule_timeout_uninterruptible(timeout); |
1738 | } |
1739 | |
1740 | EXPORT_SYMBOL(msleep); |
1741 | |
1742 | /** |
1743 | * msleep_interruptible - sleep waiting for signals |
1744 | * @msecs: Time in milliseconds to sleep for |
1745 | */ |
1746 | unsigned long msleep_interruptible(unsigned int msecs) |
1747 | { |
1748 | unsigned long timeout = msecs_to_jiffies(msecs) + 1; |
1749 | |
1750 | while (timeout && !signal_pending(current)) |
1751 | timeout = schedule_timeout_interruptible(timeout); |
1752 | return jiffies_to_msecs(timeout); |
1753 | } |
1754 | |
1755 | EXPORT_SYMBOL(msleep_interruptible); |
1756 | |
1757 | static int __sched do_usleep_range(unsigned long min, unsigned long max) |
1758 | { |
1759 | ktime_t kmin; |
1760 | unsigned long delta; |
1761 | |
1762 | kmin = ktime_set(0, min * NSEC_PER_USEC); |
1763 | delta = (max - min) * NSEC_PER_USEC; |
1764 | return schedule_hrtimeout_range(&kmin, delta, HRTIMER_MODE_REL); |
1765 | } |
1766 | |
1767 | /** |
1768 | * usleep_range - Drop in replacement for udelay where wakeup is flexible |
1769 | * @min: Minimum time in usecs to sleep |
1770 | * @max: Maximum time in usecs to sleep |
1771 | */ |
1772 | void usleep_range(unsigned long min, unsigned long max) |
1773 | { |
1774 | __set_current_state(TASK_UNINTERRUPTIBLE); |
1775 | do_usleep_range(min, max); |
1776 | } |
1777 | EXPORT_SYMBOL(usleep_range); |
1778 |
Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master
Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9