Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/kernel/timer.c | |
3 | * | |
8524070b | 4 | * Kernel internal timers, basic process system calls |
1da177e4 LT |
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> | |
b488893a | 29 | #include <linux/pid_namespace.h> |
1da177e4 LT |
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> | |
97a41e26 | 37 | #include <linux/delay.h> |
79bf2bb3 | 38 | #include <linux/tick.h> |
82f67cd9 | 39 | #include <linux/kallsyms.h> |
925d519a | 40 | #include <linux/perf_counter.h> |
eea08f32 | 41 | #include <linux/sched.h> |
1da177e4 LT |
42 | |
43 | #include <asm/uaccess.h> | |
44 | #include <asm/unistd.h> | |
45 | #include <asm/div64.h> | |
46 | #include <asm/timex.h> | |
47 | #include <asm/io.h> | |
48 | ||
ecea8d19 TG |
49 | u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES; |
50 | ||
51 | EXPORT_SYMBOL(jiffies_64); | |
52 | ||
1da177e4 LT |
53 | /* |
54 | * per-CPU timer vector definitions: | |
55 | */ | |
1da177e4 LT |
56 | #define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6) |
57 | #define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8) | |
58 | #define TVN_SIZE (1 << TVN_BITS) | |
59 | #define TVR_SIZE (1 << TVR_BITS) | |
60 | #define TVN_MASK (TVN_SIZE - 1) | |
61 | #define TVR_MASK (TVR_SIZE - 1) | |
62 | ||
a6fa8e5a | 63 | struct tvec { |
1da177e4 | 64 | struct list_head vec[TVN_SIZE]; |
a6fa8e5a | 65 | }; |
1da177e4 | 66 | |
a6fa8e5a | 67 | struct tvec_root { |
1da177e4 | 68 | struct list_head vec[TVR_SIZE]; |
a6fa8e5a | 69 | }; |
1da177e4 | 70 | |
a6fa8e5a | 71 | struct tvec_base { |
3691c519 ON |
72 | spinlock_t lock; |
73 | struct timer_list *running_timer; | |
1da177e4 | 74 | unsigned long timer_jiffies; |
a6fa8e5a PM |
75 | struct tvec_root tv1; |
76 | struct tvec tv2; | |
77 | struct tvec tv3; | |
78 | struct tvec tv4; | |
79 | struct tvec tv5; | |
6e453a67 | 80 | } ____cacheline_aligned; |
1da177e4 | 81 | |
a6fa8e5a | 82 | struct tvec_base boot_tvec_bases; |
3691c519 | 83 | EXPORT_SYMBOL(boot_tvec_bases); |
a6fa8e5a | 84 | static DEFINE_PER_CPU(struct tvec_base *, tvec_bases) = &boot_tvec_bases; |
1da177e4 | 85 | |
6e453a67 | 86 | /* |
a6fa8e5a | 87 | * Note that all tvec_bases are 2 byte aligned and lower bit of |
6e453a67 VP |
88 | * base in timer_list is guaranteed to be zero. Use the LSB for |
89 | * the new flag to indicate whether the timer is deferrable | |
90 | */ | |
91 | #define TBASE_DEFERRABLE_FLAG (0x1) | |
92 | ||
93 | /* Functions below help us manage 'deferrable' flag */ | |
a6fa8e5a | 94 | static inline unsigned int tbase_get_deferrable(struct tvec_base *base) |
6e453a67 | 95 | { |
e9910846 | 96 | return ((unsigned int)(unsigned long)base & TBASE_DEFERRABLE_FLAG); |
6e453a67 VP |
97 | } |
98 | ||
a6fa8e5a | 99 | static inline struct tvec_base *tbase_get_base(struct tvec_base *base) |
6e453a67 | 100 | { |
a6fa8e5a | 101 | return ((struct tvec_base *)((unsigned long)base & ~TBASE_DEFERRABLE_FLAG)); |
6e453a67 VP |
102 | } |
103 | ||
104 | static inline void timer_set_deferrable(struct timer_list *timer) | |
105 | { | |
a6fa8e5a | 106 | timer->base = ((struct tvec_base *)((unsigned long)(timer->base) | |
6819457d | 107 | TBASE_DEFERRABLE_FLAG)); |
6e453a67 VP |
108 | } |
109 | ||
110 | static inline void | |
a6fa8e5a | 111 | timer_set_base(struct timer_list *timer, struct tvec_base *new_base) |
6e453a67 | 112 | { |
a6fa8e5a | 113 | timer->base = (struct tvec_base *)((unsigned long)(new_base) | |
6819457d | 114 | tbase_get_deferrable(timer->base)); |
6e453a67 VP |
115 | } |
116 | ||
9c133c46 AS |
117 | static unsigned long round_jiffies_common(unsigned long j, int cpu, |
118 | bool force_up) | |
4c36a5de AV |
119 | { |
120 | int rem; | |
121 | unsigned long original = j; | |
122 | ||
123 | /* | |
124 | * We don't want all cpus firing their timers at once hitting the | |
125 | * same lock or cachelines, so we skew each extra cpu with an extra | |
126 | * 3 jiffies. This 3 jiffies came originally from the mm/ code which | |
127 | * already did this. | |
128 | * The skew is done by adding 3*cpunr, then round, then subtract this | |
129 | * extra offset again. | |
130 | */ | |
131 | j += cpu * 3; | |
132 | ||
133 | rem = j % HZ; | |
134 | ||
135 | /* | |
136 | * If the target jiffie is just after a whole second (which can happen | |
137 | * due to delays of the timer irq, long irq off times etc etc) then | |
138 | * we should round down to the whole second, not up. Use 1/4th second | |
139 | * as cutoff for this rounding as an extreme upper bound for this. | |
9c133c46 | 140 | * But never round down if @force_up is set. |
4c36a5de | 141 | */ |
9c133c46 | 142 | if (rem < HZ/4 && !force_up) /* round down */ |
4c36a5de AV |
143 | j = j - rem; |
144 | else /* round up */ | |
145 | j = j - rem + HZ; | |
146 | ||
147 | /* now that we have rounded, subtract the extra skew again */ | |
148 | j -= cpu * 3; | |
149 | ||
150 | if (j <= jiffies) /* rounding ate our timeout entirely; */ | |
151 | return original; | |
152 | return j; | |
153 | } | |
9c133c46 AS |
154 | |
155 | /** | |
156 | * __round_jiffies - function to round jiffies to a full second | |
157 | * @j: the time in (absolute) jiffies that should be rounded | |
158 | * @cpu: the processor number on which the timeout will happen | |
159 | * | |
160 | * __round_jiffies() rounds an absolute time in the future (in jiffies) | |
161 | * up or down to (approximately) full seconds. This is useful for timers | |
162 | * for which the exact time they fire does not matter too much, as long as | |
163 | * they fire approximately every X seconds. | |
164 | * | |
165 | * By rounding these timers to whole seconds, all such timers will fire | |
166 | * at the same time, rather than at various times spread out. The goal | |
167 | * of this is to have the CPU wake up less, which saves power. | |
168 | * | |
169 | * The exact rounding is skewed for each processor to avoid all | |
170 | * processors firing at the exact same time, which could lead | |
171 | * to lock contention or spurious cache line bouncing. | |
172 | * | |
173 | * The return value is the rounded version of the @j parameter. | |
174 | */ | |
175 | unsigned long __round_jiffies(unsigned long j, int cpu) | |
176 | { | |
177 | return round_jiffies_common(j, cpu, false); | |
178 | } | |
4c36a5de AV |
179 | EXPORT_SYMBOL_GPL(__round_jiffies); |
180 | ||
181 | /** | |
182 | * __round_jiffies_relative - function to round jiffies to a full second | |
183 | * @j: the time in (relative) jiffies that should be rounded | |
184 | * @cpu: the processor number on which the timeout will happen | |
185 | * | |
72fd4a35 | 186 | * __round_jiffies_relative() rounds a time delta in the future (in jiffies) |
4c36a5de AV |
187 | * up or down to (approximately) full seconds. This is useful for timers |
188 | * for which the exact time they fire does not matter too much, as long as | |
189 | * they fire approximately every X seconds. | |
190 | * | |
191 | * By rounding these timers to whole seconds, all such timers will fire | |
192 | * at the same time, rather than at various times spread out. The goal | |
193 | * of this is to have the CPU wake up less, which saves power. | |
194 | * | |
195 | * The exact rounding is skewed for each processor to avoid all | |
196 | * processors firing at the exact same time, which could lead | |
197 | * to lock contention or spurious cache line bouncing. | |
198 | * | |
72fd4a35 | 199 | * The return value is the rounded version of the @j parameter. |
4c36a5de AV |
200 | */ |
201 | unsigned long __round_jiffies_relative(unsigned long j, int cpu) | |
202 | { | |
9c133c46 AS |
203 | unsigned long j0 = jiffies; |
204 | ||
205 | /* Use j0 because jiffies might change while we run */ | |
206 | return round_jiffies_common(j + j0, cpu, false) - j0; | |
4c36a5de AV |
207 | } |
208 | EXPORT_SYMBOL_GPL(__round_jiffies_relative); | |
209 | ||
210 | /** | |
211 | * round_jiffies - function to round jiffies to a full second | |
212 | * @j: the time in (absolute) jiffies that should be rounded | |
213 | * | |
72fd4a35 | 214 | * round_jiffies() rounds an absolute time in the future (in jiffies) |
4c36a5de AV |
215 | * up or down to (approximately) full seconds. This is useful for timers |
216 | * for which the exact time they fire does not matter too much, as long as | |
217 | * they fire approximately every X seconds. | |
218 | * | |
219 | * By rounding these timers to whole seconds, all such timers will fire | |
220 | * at the same time, rather than at various times spread out. The goal | |
221 | * of this is to have the CPU wake up less, which saves power. | |
222 | * | |
72fd4a35 | 223 | * The return value is the rounded version of the @j parameter. |
4c36a5de AV |
224 | */ |
225 | unsigned long round_jiffies(unsigned long j) | |
226 | { | |
9c133c46 | 227 | return round_jiffies_common(j, raw_smp_processor_id(), false); |
4c36a5de AV |
228 | } |
229 | EXPORT_SYMBOL_GPL(round_jiffies); | |
230 | ||
231 | /** | |
232 | * round_jiffies_relative - function to round jiffies to a full second | |
233 | * @j: the time in (relative) jiffies that should be rounded | |
234 | * | |
72fd4a35 | 235 | * round_jiffies_relative() rounds a time delta in the future (in jiffies) |
4c36a5de AV |
236 | * up or down to (approximately) full seconds. This is useful for timers |
237 | * for which the exact time they fire does not matter too much, as long as | |
238 | * they fire approximately every X seconds. | |
239 | * | |
240 | * By rounding these timers to whole seconds, all such timers will fire | |
241 | * at the same time, rather than at various times spread out. The goal | |
242 | * of this is to have the CPU wake up less, which saves power. | |
243 | * | |
72fd4a35 | 244 | * The return value is the rounded version of the @j parameter. |
4c36a5de AV |
245 | */ |
246 | unsigned long round_jiffies_relative(unsigned long j) | |
247 | { | |
248 | return __round_jiffies_relative(j, raw_smp_processor_id()); | |
249 | } | |
250 | EXPORT_SYMBOL_GPL(round_jiffies_relative); | |
251 | ||
9c133c46 AS |
252 | /** |
253 | * __round_jiffies_up - function to round jiffies up to a full second | |
254 | * @j: the time in (absolute) jiffies that should be rounded | |
255 | * @cpu: the processor number on which the timeout will happen | |
256 | * | |
257 | * This is the same as __round_jiffies() except that it will never | |
258 | * round down. This is useful for timeouts for which the exact time | |
259 | * of firing does not matter too much, as long as they don't fire too | |
260 | * early. | |
261 | */ | |
262 | unsigned long __round_jiffies_up(unsigned long j, int cpu) | |
263 | { | |
264 | return round_jiffies_common(j, cpu, true); | |
265 | } | |
266 | EXPORT_SYMBOL_GPL(__round_jiffies_up); | |
267 | ||
268 | /** | |
269 | * __round_jiffies_up_relative - function to round jiffies up to a full second | |
270 | * @j: the time in (relative) jiffies that should be rounded | |
271 | * @cpu: the processor number on which the timeout will happen | |
272 | * | |
273 | * This is the same as __round_jiffies_relative() except that it will never | |
274 | * round down. This is useful for timeouts for which the exact time | |
275 | * of firing does not matter too much, as long as they don't fire too | |
276 | * early. | |
277 | */ | |
278 | unsigned long __round_jiffies_up_relative(unsigned long j, int cpu) | |
279 | { | |
280 | unsigned long j0 = jiffies; | |
281 | ||
282 | /* Use j0 because jiffies might change while we run */ | |
283 | return round_jiffies_common(j + j0, cpu, true) - j0; | |
284 | } | |
285 | EXPORT_SYMBOL_GPL(__round_jiffies_up_relative); | |
286 | ||
287 | /** | |
288 | * round_jiffies_up - function to round jiffies up to a full second | |
289 | * @j: the time in (absolute) jiffies that should be rounded | |
290 | * | |
291 | * This is the same as round_jiffies() except that it will never | |
292 | * round down. This is useful for timeouts for which the exact time | |
293 | * of firing does not matter too much, as long as they don't fire too | |
294 | * early. | |
295 | */ | |
296 | unsigned long round_jiffies_up(unsigned long j) | |
297 | { | |
298 | return round_jiffies_common(j, raw_smp_processor_id(), true); | |
299 | } | |
300 | EXPORT_SYMBOL_GPL(round_jiffies_up); | |
301 | ||
302 | /** | |
303 | * round_jiffies_up_relative - function to round jiffies up to a full second | |
304 | * @j: the time in (relative) jiffies that should be rounded | |
305 | * | |
306 | * This is the same as round_jiffies_relative() except that it will never | |
307 | * round down. This is useful for timeouts for which the exact time | |
308 | * of firing does not matter too much, as long as they don't fire too | |
309 | * early. | |
310 | */ | |
311 | unsigned long round_jiffies_up_relative(unsigned long j) | |
312 | { | |
313 | return __round_jiffies_up_relative(j, raw_smp_processor_id()); | |
314 | } | |
315 | EXPORT_SYMBOL_GPL(round_jiffies_up_relative); | |
316 | ||
4c36a5de | 317 | |
a6fa8e5a | 318 | static inline void set_running_timer(struct tvec_base *base, |
1da177e4 LT |
319 | struct timer_list *timer) |
320 | { | |
321 | #ifdef CONFIG_SMP | |
3691c519 | 322 | base->running_timer = timer; |
1da177e4 LT |
323 | #endif |
324 | } | |
325 | ||
a6fa8e5a | 326 | static void internal_add_timer(struct tvec_base *base, struct timer_list *timer) |
1da177e4 LT |
327 | { |
328 | unsigned long expires = timer->expires; | |
329 | unsigned long idx = expires - base->timer_jiffies; | |
330 | struct list_head *vec; | |
331 | ||
332 | if (idx < TVR_SIZE) { | |
333 | int i = expires & TVR_MASK; | |
334 | vec = base->tv1.vec + i; | |
335 | } else if (idx < 1 << (TVR_BITS + TVN_BITS)) { | |
336 | int i = (expires >> TVR_BITS) & TVN_MASK; | |
337 | vec = base->tv2.vec + i; | |
338 | } else if (idx < 1 << (TVR_BITS + 2 * TVN_BITS)) { | |
339 | int i = (expires >> (TVR_BITS + TVN_BITS)) & TVN_MASK; | |
340 | vec = base->tv3.vec + i; | |
341 | } else if (idx < 1 << (TVR_BITS + 3 * TVN_BITS)) { | |
342 | int i = (expires >> (TVR_BITS + 2 * TVN_BITS)) & TVN_MASK; | |
343 | vec = base->tv4.vec + i; | |
344 | } else if ((signed long) idx < 0) { | |
345 | /* | |
346 | * Can happen if you add a timer with expires == jiffies, | |
347 | * or you set a timer to go off in the past | |
348 | */ | |
349 | vec = base->tv1.vec + (base->timer_jiffies & TVR_MASK); | |
350 | } else { | |
351 | int i; | |
352 | /* If the timeout is larger than 0xffffffff on 64-bit | |
353 | * architectures then we use the maximum timeout: | |
354 | */ | |
355 | if (idx > 0xffffffffUL) { | |
356 | idx = 0xffffffffUL; | |
357 | expires = idx + base->timer_jiffies; | |
358 | } | |
359 | i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK; | |
360 | vec = base->tv5.vec + i; | |
361 | } | |
362 | /* | |
363 | * Timers are FIFO: | |
364 | */ | |
365 | list_add_tail(&timer->entry, vec); | |
366 | } | |
367 | ||
82f67cd9 IM |
368 | #ifdef CONFIG_TIMER_STATS |
369 | void __timer_stats_timer_set_start_info(struct timer_list *timer, void *addr) | |
370 | { | |
371 | if (timer->start_site) | |
372 | return; | |
373 | ||
374 | timer->start_site = addr; | |
375 | memcpy(timer->start_comm, current->comm, TASK_COMM_LEN); | |
376 | timer->start_pid = current->pid; | |
377 | } | |
c5c061b8 VP |
378 | |
379 | static void timer_stats_account_timer(struct timer_list *timer) | |
380 | { | |
381 | unsigned int flag = 0; | |
382 | ||
507e1231 HC |
383 | if (likely(!timer->start_site)) |
384 | return; | |
c5c061b8 VP |
385 | if (unlikely(tbase_get_deferrable(timer->base))) |
386 | flag |= TIMER_STATS_FLAG_DEFERRABLE; | |
387 | ||
388 | timer_stats_update_stats(timer, timer->start_pid, timer->start_site, | |
389 | timer->function, timer->start_comm, flag); | |
390 | } | |
391 | ||
392 | #else | |
393 | static void timer_stats_account_timer(struct timer_list *timer) {} | |
82f67cd9 IM |
394 | #endif |
395 | ||
c6f3a97f TG |
396 | #ifdef CONFIG_DEBUG_OBJECTS_TIMERS |
397 | ||
398 | static struct debug_obj_descr timer_debug_descr; | |
399 | ||
400 | /* | |
401 | * fixup_init is called when: | |
402 | * - an active object is initialized | |
55c888d6 | 403 | */ |
c6f3a97f TG |
404 | static int timer_fixup_init(void *addr, enum debug_obj_state state) |
405 | { | |
406 | struct timer_list *timer = addr; | |
407 | ||
408 | switch (state) { | |
409 | case ODEBUG_STATE_ACTIVE: | |
410 | del_timer_sync(timer); | |
411 | debug_object_init(timer, &timer_debug_descr); | |
412 | return 1; | |
413 | default: | |
414 | return 0; | |
415 | } | |
416 | } | |
417 | ||
418 | /* | |
419 | * fixup_activate is called when: | |
420 | * - an active object is activated | |
421 | * - an unknown object is activated (might be a statically initialized object) | |
422 | */ | |
423 | static int timer_fixup_activate(void *addr, enum debug_obj_state state) | |
424 | { | |
425 | struct timer_list *timer = addr; | |
426 | ||
427 | switch (state) { | |
428 | ||
429 | case ODEBUG_STATE_NOTAVAILABLE: | |
430 | /* | |
431 | * This is not really a fixup. The timer was | |
432 | * statically initialized. We just make sure that it | |
433 | * is tracked in the object tracker. | |
434 | */ | |
435 | if (timer->entry.next == NULL && | |
436 | timer->entry.prev == TIMER_ENTRY_STATIC) { | |
437 | debug_object_init(timer, &timer_debug_descr); | |
438 | debug_object_activate(timer, &timer_debug_descr); | |
439 | return 0; | |
440 | } else { | |
441 | WARN_ON_ONCE(1); | |
442 | } | |
443 | return 0; | |
444 | ||
445 | case ODEBUG_STATE_ACTIVE: | |
446 | WARN_ON(1); | |
447 | ||
448 | default: | |
449 | return 0; | |
450 | } | |
451 | } | |
452 | ||
453 | /* | |
454 | * fixup_free is called when: | |
455 | * - an active object is freed | |
456 | */ | |
457 | static int timer_fixup_free(void *addr, enum debug_obj_state state) | |
458 | { | |
459 | struct timer_list *timer = addr; | |
460 | ||
461 | switch (state) { | |
462 | case ODEBUG_STATE_ACTIVE: | |
463 | del_timer_sync(timer); | |
464 | debug_object_free(timer, &timer_debug_descr); | |
465 | return 1; | |
466 | default: | |
467 | return 0; | |
468 | } | |
469 | } | |
470 | ||
471 | static struct debug_obj_descr timer_debug_descr = { | |
472 | .name = "timer_list", | |
473 | .fixup_init = timer_fixup_init, | |
474 | .fixup_activate = timer_fixup_activate, | |
475 | .fixup_free = timer_fixup_free, | |
476 | }; | |
477 | ||
478 | static inline void debug_timer_init(struct timer_list *timer) | |
479 | { | |
480 | debug_object_init(timer, &timer_debug_descr); | |
481 | } | |
482 | ||
483 | static inline void debug_timer_activate(struct timer_list *timer) | |
484 | { | |
485 | debug_object_activate(timer, &timer_debug_descr); | |
486 | } | |
487 | ||
488 | static inline void debug_timer_deactivate(struct timer_list *timer) | |
489 | { | |
490 | debug_object_deactivate(timer, &timer_debug_descr); | |
491 | } | |
492 | ||
493 | static inline void debug_timer_free(struct timer_list *timer) | |
494 | { | |
495 | debug_object_free(timer, &timer_debug_descr); | |
496 | } | |
497 | ||
6f2b9b9a JB |
498 | static void __init_timer(struct timer_list *timer, |
499 | const char *name, | |
500 | struct lock_class_key *key); | |
c6f3a97f | 501 | |
6f2b9b9a JB |
502 | void init_timer_on_stack_key(struct timer_list *timer, |
503 | const char *name, | |
504 | struct lock_class_key *key) | |
c6f3a97f TG |
505 | { |
506 | debug_object_init_on_stack(timer, &timer_debug_descr); | |
6f2b9b9a | 507 | __init_timer(timer, name, key); |
c6f3a97f | 508 | } |
6f2b9b9a | 509 | EXPORT_SYMBOL_GPL(init_timer_on_stack_key); |
c6f3a97f TG |
510 | |
511 | void destroy_timer_on_stack(struct timer_list *timer) | |
512 | { | |
513 | debug_object_free(timer, &timer_debug_descr); | |
514 | } | |
515 | EXPORT_SYMBOL_GPL(destroy_timer_on_stack); | |
516 | ||
517 | #else | |
518 | static inline void debug_timer_init(struct timer_list *timer) { } | |
519 | static inline void debug_timer_activate(struct timer_list *timer) { } | |
520 | static inline void debug_timer_deactivate(struct timer_list *timer) { } | |
521 | #endif | |
522 | ||
6f2b9b9a JB |
523 | static void __init_timer(struct timer_list *timer, |
524 | const char *name, | |
525 | struct lock_class_key *key) | |
55c888d6 ON |
526 | { |
527 | timer->entry.next = NULL; | |
bfe5d834 | 528 | timer->base = __raw_get_cpu_var(tvec_bases); |
82f67cd9 IM |
529 | #ifdef CONFIG_TIMER_STATS |
530 | timer->start_site = NULL; | |
531 | timer->start_pid = -1; | |
532 | memset(timer->start_comm, 0, TASK_COMM_LEN); | |
533 | #endif | |
6f2b9b9a | 534 | lockdep_init_map(&timer->lockdep_map, name, key, 0); |
55c888d6 | 535 | } |
c6f3a97f TG |
536 | |
537 | /** | |
633fe795 | 538 | * init_timer_key - initialize a timer |
c6f3a97f | 539 | * @timer: the timer to be initialized |
633fe795 RD |
540 | * @name: name of the timer |
541 | * @key: lockdep class key of the fake lock used for tracking timer | |
542 | * sync lock dependencies | |
c6f3a97f | 543 | * |
633fe795 | 544 | * init_timer_key() must be done to a timer prior calling *any* of the |
c6f3a97f TG |
545 | * other timer functions. |
546 | */ | |
6f2b9b9a JB |
547 | void init_timer_key(struct timer_list *timer, |
548 | const char *name, | |
549 | struct lock_class_key *key) | |
c6f3a97f TG |
550 | { |
551 | debug_timer_init(timer); | |
6f2b9b9a | 552 | __init_timer(timer, name, key); |
c6f3a97f | 553 | } |
6f2b9b9a | 554 | EXPORT_SYMBOL(init_timer_key); |
55c888d6 | 555 | |
6f2b9b9a JB |
556 | void init_timer_deferrable_key(struct timer_list *timer, |
557 | const char *name, | |
558 | struct lock_class_key *key) | |
6e453a67 | 559 | { |
6f2b9b9a | 560 | init_timer_key(timer, name, key); |
6e453a67 VP |
561 | timer_set_deferrable(timer); |
562 | } | |
6f2b9b9a | 563 | EXPORT_SYMBOL(init_timer_deferrable_key); |
6e453a67 | 564 | |
55c888d6 | 565 | static inline void detach_timer(struct timer_list *timer, |
82f67cd9 | 566 | int clear_pending) |
55c888d6 ON |
567 | { |
568 | struct list_head *entry = &timer->entry; | |
569 | ||
c6f3a97f TG |
570 | debug_timer_deactivate(timer); |
571 | ||
55c888d6 ON |
572 | __list_del(entry->prev, entry->next); |
573 | if (clear_pending) | |
574 | entry->next = NULL; | |
575 | entry->prev = LIST_POISON2; | |
576 | } | |
577 | ||
578 | /* | |
3691c519 | 579 | * We are using hashed locking: holding per_cpu(tvec_bases).lock |
55c888d6 ON |
580 | * means that all timers which are tied to this base via timer->base are |
581 | * locked, and the base itself is locked too. | |
582 | * | |
583 | * So __run_timers/migrate_timers can safely modify all timers which could | |
584 | * be found on ->tvX lists. | |
585 | * | |
586 | * When the timer's base is locked, and the timer removed from list, it is | |
587 | * possible to set timer->base = NULL and drop the lock: the timer remains | |
588 | * locked. | |
589 | */ | |
a6fa8e5a | 590 | static struct tvec_base *lock_timer_base(struct timer_list *timer, |
55c888d6 | 591 | unsigned long *flags) |
89e7e374 | 592 | __acquires(timer->base->lock) |
55c888d6 | 593 | { |
a6fa8e5a | 594 | struct tvec_base *base; |
55c888d6 ON |
595 | |
596 | for (;;) { | |
a6fa8e5a | 597 | struct tvec_base *prelock_base = timer->base; |
6e453a67 | 598 | base = tbase_get_base(prelock_base); |
55c888d6 ON |
599 | if (likely(base != NULL)) { |
600 | spin_lock_irqsave(&base->lock, *flags); | |
6e453a67 | 601 | if (likely(prelock_base == timer->base)) |
55c888d6 ON |
602 | return base; |
603 | /* The timer has migrated to another CPU */ | |
604 | spin_unlock_irqrestore(&base->lock, *flags); | |
605 | } | |
606 | cpu_relax(); | |
607 | } | |
608 | } | |
609 | ||
74019224 | 610 | static inline int |
597d0275 AB |
611 | __mod_timer(struct timer_list *timer, unsigned long expires, |
612 | bool pending_only, int pinned) | |
1da177e4 | 613 | { |
a6fa8e5a | 614 | struct tvec_base *base, *new_base; |
1da177e4 | 615 | unsigned long flags; |
eea08f32 | 616 | int ret = 0 , cpu; |
1da177e4 | 617 | |
82f67cd9 | 618 | timer_stats_timer_set_start_info(timer); |
1da177e4 | 619 | BUG_ON(!timer->function); |
1da177e4 | 620 | |
55c888d6 ON |
621 | base = lock_timer_base(timer, &flags); |
622 | ||
623 | if (timer_pending(timer)) { | |
624 | detach_timer(timer, 0); | |
625 | ret = 1; | |
74019224 IM |
626 | } else { |
627 | if (pending_only) | |
628 | goto out_unlock; | |
55c888d6 ON |
629 | } |
630 | ||
c6f3a97f TG |
631 | debug_timer_activate(timer); |
632 | ||
a4a6198b | 633 | new_base = __get_cpu_var(tvec_bases); |
1da177e4 | 634 | |
eea08f32 AB |
635 | cpu = smp_processor_id(); |
636 | ||
637 | #if defined(CONFIG_NO_HZ) && defined(CONFIG_SMP) | |
638 | if (!pinned && get_sysctl_timer_migration() && idle_cpu(cpu)) { | |
639 | int preferred_cpu = get_nohz_load_balancer(); | |
640 | ||
641 | if (preferred_cpu >= 0) | |
642 | cpu = preferred_cpu; | |
643 | } | |
644 | #endif | |
645 | new_base = per_cpu(tvec_bases, cpu); | |
646 | ||
3691c519 | 647 | if (base != new_base) { |
1da177e4 | 648 | /* |
55c888d6 ON |
649 | * We are trying to schedule the timer on the local CPU. |
650 | * However we can't change timer's base while it is running, | |
651 | * otherwise del_timer_sync() can't detect that the timer's | |
652 | * handler yet has not finished. This also guarantees that | |
653 | * the timer is serialized wrt itself. | |
1da177e4 | 654 | */ |
a2c348fe | 655 | if (likely(base->running_timer != timer)) { |
55c888d6 | 656 | /* See the comment in lock_timer_base() */ |
6e453a67 | 657 | timer_set_base(timer, NULL); |
55c888d6 | 658 | spin_unlock(&base->lock); |
a2c348fe ON |
659 | base = new_base; |
660 | spin_lock(&base->lock); | |
6e453a67 | 661 | timer_set_base(timer, base); |
1da177e4 LT |
662 | } |
663 | } | |
664 | ||
1da177e4 | 665 | timer->expires = expires; |
a2c348fe | 666 | internal_add_timer(base, timer); |
74019224 IM |
667 | |
668 | out_unlock: | |
a2c348fe | 669 | spin_unlock_irqrestore(&base->lock, flags); |
1da177e4 LT |
670 | |
671 | return ret; | |
672 | } | |
673 | ||
2aae4a10 | 674 | /** |
74019224 IM |
675 | * mod_timer_pending - modify a pending timer's timeout |
676 | * @timer: the pending timer to be modified | |
677 | * @expires: new timeout in jiffies | |
1da177e4 | 678 | * |
74019224 IM |
679 | * mod_timer_pending() is the same for pending timers as mod_timer(), |
680 | * but will not re-activate and modify already deleted timers. | |
681 | * | |
682 | * It is useful for unserialized use of timers. | |
1da177e4 | 683 | */ |
74019224 | 684 | int mod_timer_pending(struct timer_list *timer, unsigned long expires) |
1da177e4 | 685 | { |
597d0275 | 686 | return __mod_timer(timer, expires, true, TIMER_NOT_PINNED); |
1da177e4 | 687 | } |
74019224 | 688 | EXPORT_SYMBOL(mod_timer_pending); |
1da177e4 | 689 | |
2aae4a10 | 690 | /** |
1da177e4 LT |
691 | * mod_timer - modify a timer's timeout |
692 | * @timer: the timer to be modified | |
2aae4a10 | 693 | * @expires: new timeout in jiffies |
1da177e4 | 694 | * |
72fd4a35 | 695 | * mod_timer() is a more efficient way to update the expire field of an |
1da177e4 LT |
696 | * active timer (if the timer is inactive it will be activated) |
697 | * | |
698 | * mod_timer(timer, expires) is equivalent to: | |
699 | * | |
700 | * del_timer(timer); timer->expires = expires; add_timer(timer); | |
701 | * | |
702 | * Note that if there are multiple unserialized concurrent users of the | |
703 | * same timer, then mod_timer() is the only safe way to modify the timeout, | |
704 | * since add_timer() cannot modify an already running timer. | |
705 | * | |
706 | * The function returns whether it has modified a pending timer or not. | |
707 | * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an | |
708 | * active timer returns 1.) | |
709 | */ | |
710 | int mod_timer(struct timer_list *timer, unsigned long expires) | |
711 | { | |
1da177e4 LT |
712 | /* |
713 | * This is a common optimization triggered by the | |
714 | * networking code - if the timer is re-modified | |
715 | * to be the same thing then just return: | |
716 | */ | |
4841158b | 717 | if (timer_pending(timer) && timer->expires == expires) |
1da177e4 LT |
718 | return 1; |
719 | ||
597d0275 | 720 | return __mod_timer(timer, expires, false, TIMER_NOT_PINNED); |
1da177e4 | 721 | } |
1da177e4 LT |
722 | EXPORT_SYMBOL(mod_timer); |
723 | ||
597d0275 AB |
724 | /** |
725 | * mod_timer_pinned - modify a timer's timeout | |
726 | * @timer: the timer to be modified | |
727 | * @expires: new timeout in jiffies | |
728 | * | |
729 | * mod_timer_pinned() is a way to update the expire field of an | |
730 | * active timer (if the timer is inactive it will be activated) | |
731 | * and not allow the timer to be migrated to a different CPU. | |
732 | * | |
733 | * mod_timer_pinned(timer, expires) is equivalent to: | |
734 | * | |
735 | * del_timer(timer); timer->expires = expires; add_timer(timer); | |
736 | */ | |
737 | int mod_timer_pinned(struct timer_list *timer, unsigned long expires) | |
738 | { | |
739 | if (timer->expires == expires && timer_pending(timer)) | |
740 | return 1; | |
741 | ||
742 | return __mod_timer(timer, expires, false, TIMER_PINNED); | |
743 | } | |
744 | EXPORT_SYMBOL(mod_timer_pinned); | |
745 | ||
74019224 IM |
746 | /** |
747 | * add_timer - start a timer | |
748 | * @timer: the timer to be added | |
749 | * | |
750 | * The kernel will do a ->function(->data) callback from the | |
751 | * timer interrupt at the ->expires point in the future. The | |
752 | * current time is 'jiffies'. | |
753 | * | |
754 | * The timer's ->expires, ->function (and if the handler uses it, ->data) | |
755 | * fields must be set prior calling this function. | |
756 | * | |
757 | * Timers with an ->expires field in the past will be executed in the next | |
758 | * timer tick. | |
759 | */ | |
760 | void add_timer(struct timer_list *timer) | |
761 | { | |
762 | BUG_ON(timer_pending(timer)); | |
763 | mod_timer(timer, timer->expires); | |
764 | } | |
765 | EXPORT_SYMBOL(add_timer); | |
766 | ||
767 | /** | |
768 | * add_timer_on - start a timer on a particular CPU | |
769 | * @timer: the timer to be added | |
770 | * @cpu: the CPU to start it on | |
771 | * | |
772 | * This is not very scalable on SMP. Double adds are not possible. | |
773 | */ | |
774 | void add_timer_on(struct timer_list *timer, int cpu) | |
775 | { | |
776 | struct tvec_base *base = per_cpu(tvec_bases, cpu); | |
777 | unsigned long flags; | |
778 | ||
779 | timer_stats_timer_set_start_info(timer); | |
780 | BUG_ON(timer_pending(timer) || !timer->function); | |
781 | spin_lock_irqsave(&base->lock, flags); | |
782 | timer_set_base(timer, base); | |
783 | debug_timer_activate(timer); | |
784 | internal_add_timer(base, timer); | |
785 | /* | |
786 | * Check whether the other CPU is idle and needs to be | |
787 | * triggered to reevaluate the timer wheel when nohz is | |
788 | * active. We are protected against the other CPU fiddling | |
789 | * with the timer by holding the timer base lock. This also | |
790 | * makes sure that a CPU on the way to idle can not evaluate | |
791 | * the timer wheel. | |
792 | */ | |
793 | wake_up_idle_cpu(cpu); | |
794 | spin_unlock_irqrestore(&base->lock, flags); | |
795 | } | |
a9862e05 | 796 | EXPORT_SYMBOL_GPL(add_timer_on); |
74019224 | 797 | |
2aae4a10 | 798 | /** |
1da177e4 LT |
799 | * del_timer - deactive a timer. |
800 | * @timer: the timer to be deactivated | |
801 | * | |
802 | * del_timer() deactivates a timer - this works on both active and inactive | |
803 | * timers. | |
804 | * | |
805 | * The function returns whether it has deactivated a pending timer or not. | |
806 | * (ie. del_timer() of an inactive timer returns 0, del_timer() of an | |
807 | * active timer returns 1.) | |
808 | */ | |
809 | int del_timer(struct timer_list *timer) | |
810 | { | |
a6fa8e5a | 811 | struct tvec_base *base; |
1da177e4 | 812 | unsigned long flags; |
55c888d6 | 813 | int ret = 0; |
1da177e4 | 814 | |
82f67cd9 | 815 | timer_stats_timer_clear_start_info(timer); |
55c888d6 ON |
816 | if (timer_pending(timer)) { |
817 | base = lock_timer_base(timer, &flags); | |
818 | if (timer_pending(timer)) { | |
819 | detach_timer(timer, 1); | |
820 | ret = 1; | |
821 | } | |
1da177e4 | 822 | spin_unlock_irqrestore(&base->lock, flags); |
1da177e4 | 823 | } |
1da177e4 | 824 | |
55c888d6 | 825 | return ret; |
1da177e4 | 826 | } |
1da177e4 LT |
827 | EXPORT_SYMBOL(del_timer); |
828 | ||
829 | #ifdef CONFIG_SMP | |
2aae4a10 REB |
830 | /** |
831 | * try_to_del_timer_sync - Try to deactivate a timer | |
832 | * @timer: timer do del | |
833 | * | |
fd450b73 ON |
834 | * This function tries to deactivate a timer. Upon successful (ret >= 0) |
835 | * exit the timer is not queued and the handler is not running on any CPU. | |
836 | * | |
837 | * It must not be called from interrupt contexts. | |
838 | */ | |
839 | int try_to_del_timer_sync(struct timer_list *timer) | |
840 | { | |
a6fa8e5a | 841 | struct tvec_base *base; |
fd450b73 ON |
842 | unsigned long flags; |
843 | int ret = -1; | |
844 | ||
845 | base = lock_timer_base(timer, &flags); | |
846 | ||
847 | if (base->running_timer == timer) | |
848 | goto out; | |
849 | ||
850 | ret = 0; | |
851 | if (timer_pending(timer)) { | |
852 | detach_timer(timer, 1); | |
853 | ret = 1; | |
854 | } | |
855 | out: | |
856 | spin_unlock_irqrestore(&base->lock, flags); | |
857 | ||
858 | return ret; | |
859 | } | |
e19dff1f DH |
860 | EXPORT_SYMBOL(try_to_del_timer_sync); |
861 | ||
2aae4a10 | 862 | /** |
1da177e4 LT |
863 | * del_timer_sync - deactivate a timer and wait for the handler to finish. |
864 | * @timer: the timer to be deactivated | |
865 | * | |
866 | * This function only differs from del_timer() on SMP: besides deactivating | |
867 | * the timer it also makes sure the handler has finished executing on other | |
868 | * CPUs. | |
869 | * | |
72fd4a35 | 870 | * Synchronization rules: Callers must prevent restarting of the timer, |
1da177e4 LT |
871 | * otherwise this function is meaningless. It must not be called from |
872 | * interrupt contexts. The caller must not hold locks which would prevent | |
55c888d6 ON |
873 | * completion of the timer's handler. The timer's handler must not call |
874 | * add_timer_on(). Upon exit the timer is not queued and the handler is | |
875 | * not running on any CPU. | |
1da177e4 LT |
876 | * |
877 | * The function returns whether it has deactivated a pending timer or not. | |
1da177e4 LT |
878 | */ |
879 | int del_timer_sync(struct timer_list *timer) | |
880 | { | |
6f2b9b9a JB |
881 | #ifdef CONFIG_LOCKDEP |
882 | unsigned long flags; | |
883 | ||
884 | local_irq_save(flags); | |
885 | lock_map_acquire(&timer->lockdep_map); | |
886 | lock_map_release(&timer->lockdep_map); | |
887 | local_irq_restore(flags); | |
888 | #endif | |
889 | ||
fd450b73 ON |
890 | for (;;) { |
891 | int ret = try_to_del_timer_sync(timer); | |
892 | if (ret >= 0) | |
893 | return ret; | |
a0009652 | 894 | cpu_relax(); |
fd450b73 | 895 | } |
1da177e4 | 896 | } |
55c888d6 | 897 | EXPORT_SYMBOL(del_timer_sync); |
1da177e4 LT |
898 | #endif |
899 | ||
a6fa8e5a | 900 | static int cascade(struct tvec_base *base, struct tvec *tv, int index) |
1da177e4 LT |
901 | { |
902 | /* cascade all the timers from tv up one level */ | |
3439dd86 P |
903 | struct timer_list *timer, *tmp; |
904 | struct list_head tv_list; | |
905 | ||
906 | list_replace_init(tv->vec + index, &tv_list); | |
1da177e4 | 907 | |
1da177e4 | 908 | /* |
3439dd86 P |
909 | * We are removing _all_ timers from the list, so we |
910 | * don't have to detach them individually. | |
1da177e4 | 911 | */ |
3439dd86 | 912 | list_for_each_entry_safe(timer, tmp, &tv_list, entry) { |
6e453a67 | 913 | BUG_ON(tbase_get_base(timer->base) != base); |
3439dd86 | 914 | internal_add_timer(base, timer); |
1da177e4 | 915 | } |
1da177e4 LT |
916 | |
917 | return index; | |
918 | } | |
919 | ||
2aae4a10 REB |
920 | #define INDEX(N) ((base->timer_jiffies >> (TVR_BITS + (N) * TVN_BITS)) & TVN_MASK) |
921 | ||
922 | /** | |
1da177e4 LT |
923 | * __run_timers - run all expired timers (if any) on this CPU. |
924 | * @base: the timer vector to be processed. | |
925 | * | |
926 | * This function cascades all vectors and executes all expired timer | |
927 | * vectors. | |
928 | */ | |
a6fa8e5a | 929 | static inline void __run_timers(struct tvec_base *base) |
1da177e4 LT |
930 | { |
931 | struct timer_list *timer; | |
932 | ||
3691c519 | 933 | spin_lock_irq(&base->lock); |
1da177e4 | 934 | while (time_after_eq(jiffies, base->timer_jiffies)) { |
626ab0e6 | 935 | struct list_head work_list; |
1da177e4 | 936 | struct list_head *head = &work_list; |
6819457d | 937 | int index = base->timer_jiffies & TVR_MASK; |
626ab0e6 | 938 | |
1da177e4 LT |
939 | /* |
940 | * Cascade timers: | |
941 | */ | |
942 | if (!index && | |
943 | (!cascade(base, &base->tv2, INDEX(0))) && | |
944 | (!cascade(base, &base->tv3, INDEX(1))) && | |
945 | !cascade(base, &base->tv4, INDEX(2))) | |
946 | cascade(base, &base->tv5, INDEX(3)); | |
626ab0e6 ON |
947 | ++base->timer_jiffies; |
948 | list_replace_init(base->tv1.vec + index, &work_list); | |
55c888d6 | 949 | while (!list_empty(head)) { |
1da177e4 LT |
950 | void (*fn)(unsigned long); |
951 | unsigned long data; | |
952 | ||
b5e61818 | 953 | timer = list_first_entry(head, struct timer_list,entry); |
6819457d TG |
954 | fn = timer->function; |
955 | data = timer->data; | |
1da177e4 | 956 | |
82f67cd9 IM |
957 | timer_stats_account_timer(timer); |
958 | ||
1da177e4 | 959 | set_running_timer(base, timer); |
55c888d6 | 960 | detach_timer(timer, 1); |
6f2b9b9a | 961 | |
3691c519 | 962 | spin_unlock_irq(&base->lock); |
1da177e4 | 963 | { |
be5b4fbd | 964 | int preempt_count = preempt_count(); |
6f2b9b9a JB |
965 | |
966 | #ifdef CONFIG_LOCKDEP | |
967 | /* | |
968 | * It is permissible to free the timer from | |
969 | * inside the function that is called from | |
970 | * it, this we need to take into account for | |
971 | * lockdep too. To avoid bogus "held lock | |
972 | * freed" warnings as well as problems when | |
973 | * looking into timer->lockdep_map, make a | |
974 | * copy and use that here. | |
975 | */ | |
976 | struct lockdep_map lockdep_map = | |
977 | timer->lockdep_map; | |
978 | #endif | |
979 | /* | |
980 | * Couple the lock chain with the lock chain at | |
981 | * del_timer_sync() by acquiring the lock_map | |
982 | * around the fn() call here and in | |
983 | * del_timer_sync(). | |
984 | */ | |
985 | lock_map_acquire(&lockdep_map); | |
986 | ||
1da177e4 | 987 | fn(data); |
6f2b9b9a JB |
988 | |
989 | lock_map_release(&lockdep_map); | |
990 | ||
1da177e4 | 991 | if (preempt_count != preempt_count()) { |
4c9dc641 | 992 | printk(KERN_ERR "huh, entered %p " |
be5b4fbd JJ |
993 | "with preempt_count %08x, exited" |
994 | " with %08x?\n", | |
995 | fn, preempt_count, | |
996 | preempt_count()); | |
1da177e4 LT |
997 | BUG(); |
998 | } | |
999 | } | |
3691c519 | 1000 | spin_lock_irq(&base->lock); |
1da177e4 LT |
1001 | } |
1002 | } | |
1003 | set_running_timer(base, NULL); | |
3691c519 | 1004 | spin_unlock_irq(&base->lock); |
1da177e4 LT |
1005 | } |
1006 | ||
ee9c5785 | 1007 | #ifdef CONFIG_NO_HZ |
1da177e4 LT |
1008 | /* |
1009 | * Find out when the next timer event is due to happen. This | |
1010 | * is used on S/390 to stop all activity when a cpus is idle. | |
1011 | * This functions needs to be called disabled. | |
1012 | */ | |
a6fa8e5a | 1013 | static unsigned long __next_timer_interrupt(struct tvec_base *base) |
1da177e4 | 1014 | { |
1cfd6849 | 1015 | unsigned long timer_jiffies = base->timer_jiffies; |
eaad084b | 1016 | unsigned long expires = timer_jiffies + NEXT_TIMER_MAX_DELTA; |
1cfd6849 | 1017 | int index, slot, array, found = 0; |
1da177e4 | 1018 | struct timer_list *nte; |
a6fa8e5a | 1019 | struct tvec *varray[4]; |
1da177e4 LT |
1020 | |
1021 | /* Look for timer events in tv1. */ | |
1cfd6849 | 1022 | index = slot = timer_jiffies & TVR_MASK; |
1da177e4 | 1023 | do { |
1cfd6849 | 1024 | list_for_each_entry(nte, base->tv1.vec + slot, entry) { |
6819457d TG |
1025 | if (tbase_get_deferrable(nte->base)) |
1026 | continue; | |
6e453a67 | 1027 | |
1cfd6849 | 1028 | found = 1; |
1da177e4 | 1029 | expires = nte->expires; |
1cfd6849 TG |
1030 | /* Look at the cascade bucket(s)? */ |
1031 | if (!index || slot < index) | |
1032 | goto cascade; | |
1033 | return expires; | |
1da177e4 | 1034 | } |
1cfd6849 TG |
1035 | slot = (slot + 1) & TVR_MASK; |
1036 | } while (slot != index); | |
1037 | ||
1038 | cascade: | |
1039 | /* Calculate the next cascade event */ | |
1040 | if (index) | |
1041 | timer_jiffies += TVR_SIZE - index; | |
1042 | timer_jiffies >>= TVR_BITS; | |
1da177e4 LT |
1043 | |
1044 | /* Check tv2-tv5. */ | |
1045 | varray[0] = &base->tv2; | |
1046 | varray[1] = &base->tv3; | |
1047 | varray[2] = &base->tv4; | |
1048 | varray[3] = &base->tv5; | |
1cfd6849 TG |
1049 | |
1050 | for (array = 0; array < 4; array++) { | |
a6fa8e5a | 1051 | struct tvec *varp = varray[array]; |
1cfd6849 TG |
1052 | |
1053 | index = slot = timer_jiffies & TVN_MASK; | |
1da177e4 | 1054 | do { |
1cfd6849 | 1055 | list_for_each_entry(nte, varp->vec + slot, entry) { |
a0419888 JH |
1056 | if (tbase_get_deferrable(nte->base)) |
1057 | continue; | |
1058 | ||
1cfd6849 | 1059 | found = 1; |
1da177e4 LT |
1060 | if (time_before(nte->expires, expires)) |
1061 | expires = nte->expires; | |
1cfd6849 TG |
1062 | } |
1063 | /* | |
1064 | * Do we still search for the first timer or are | |
1065 | * we looking up the cascade buckets ? | |
1066 | */ | |
1067 | if (found) { | |
1068 | /* Look at the cascade bucket(s)? */ | |
1069 | if (!index || slot < index) | |
1070 | break; | |
1071 | return expires; | |
1072 | } | |
1073 | slot = (slot + 1) & TVN_MASK; | |
1074 | } while (slot != index); | |
1075 | ||
1076 | if (index) | |
1077 | timer_jiffies += TVN_SIZE - index; | |
1078 | timer_jiffies >>= TVN_BITS; | |
1da177e4 | 1079 | } |
1cfd6849 TG |
1080 | return expires; |
1081 | } | |
69239749 | 1082 | |
1cfd6849 TG |
1083 | /* |
1084 | * Check, if the next hrtimer event is before the next timer wheel | |
1085 | * event: | |
1086 | */ | |
1087 | static unsigned long cmp_next_hrtimer_event(unsigned long now, | |
1088 | unsigned long expires) | |
1089 | { | |
1090 | ktime_t hr_delta = hrtimer_get_next_event(); | |
1091 | struct timespec tsdelta; | |
9501b6cf | 1092 | unsigned long delta; |
1cfd6849 TG |
1093 | |
1094 | if (hr_delta.tv64 == KTIME_MAX) | |
1095 | return expires; | |
0662b713 | 1096 | |
9501b6cf TG |
1097 | /* |
1098 | * Expired timer available, let it expire in the next tick | |
1099 | */ | |
1100 | if (hr_delta.tv64 <= 0) | |
1101 | return now + 1; | |
69239749 | 1102 | |
1cfd6849 | 1103 | tsdelta = ktime_to_timespec(hr_delta); |
9501b6cf | 1104 | delta = timespec_to_jiffies(&tsdelta); |
eaad084b TG |
1105 | |
1106 | /* | |
1107 | * Limit the delta to the max value, which is checked in | |
1108 | * tick_nohz_stop_sched_tick(): | |
1109 | */ | |
1110 | if (delta > NEXT_TIMER_MAX_DELTA) | |
1111 | delta = NEXT_TIMER_MAX_DELTA; | |
1112 | ||
9501b6cf TG |
1113 | /* |
1114 | * Take rounding errors in to account and make sure, that it | |
1115 | * expires in the next tick. Otherwise we go into an endless | |
1116 | * ping pong due to tick_nohz_stop_sched_tick() retriggering | |
1117 | * the timer softirq | |
1118 | */ | |
1119 | if (delta < 1) | |
1120 | delta = 1; | |
1121 | now += delta; | |
1cfd6849 TG |
1122 | if (time_before(now, expires)) |
1123 | return now; | |
1da177e4 LT |
1124 | return expires; |
1125 | } | |
1cfd6849 TG |
1126 | |
1127 | /** | |
8dce39c2 | 1128 | * get_next_timer_interrupt - return the jiffy of the next pending timer |
05fb6bf0 | 1129 | * @now: current time (in jiffies) |
1cfd6849 | 1130 | */ |
fd064b9b | 1131 | unsigned long get_next_timer_interrupt(unsigned long now) |
1cfd6849 | 1132 | { |
a6fa8e5a | 1133 | struct tvec_base *base = __get_cpu_var(tvec_bases); |
fd064b9b | 1134 | unsigned long expires; |
1cfd6849 TG |
1135 | |
1136 | spin_lock(&base->lock); | |
1137 | expires = __next_timer_interrupt(base); | |
1138 | spin_unlock(&base->lock); | |
1139 | ||
1140 | if (time_before_eq(expires, now)) | |
1141 | return now; | |
1142 | ||
1143 | return cmp_next_hrtimer_event(now, expires); | |
1144 | } | |
1da177e4 LT |
1145 | #endif |
1146 | ||
1da177e4 | 1147 | /* |
5b4db0c2 | 1148 | * Called from the timer interrupt handler to charge one tick to the current |
1da177e4 LT |
1149 | * process. user_tick is 1 if the tick is user time, 0 for system. |
1150 | */ | |
1151 | void update_process_times(int user_tick) | |
1152 | { | |
1153 | struct task_struct *p = current; | |
1154 | int cpu = smp_processor_id(); | |
1155 | ||
1156 | /* Note: this timer irq context must be accounted for as well. */ | |
fa13a5a1 | 1157 | account_process_tick(p, user_tick); |
1da177e4 LT |
1158 | run_local_timers(); |
1159 | if (rcu_pending(cpu)) | |
1160 | rcu_check_callbacks(cpu, user_tick); | |
b845b517 | 1161 | printk_tick(); |
1da177e4 | 1162 | scheduler_tick(); |
6819457d | 1163 | run_posix_cpu_timers(p); |
1da177e4 LT |
1164 | } |
1165 | ||
1da177e4 LT |
1166 | /* |
1167 | * This function runs timers and the timer-tq in bottom half context. | |
1168 | */ | |
1169 | static void run_timer_softirq(struct softirq_action *h) | |
1170 | { | |
a6fa8e5a | 1171 | struct tvec_base *base = __get_cpu_var(tvec_bases); |
1da177e4 | 1172 | |
925d519a PZ |
1173 | perf_counter_do_pending(); |
1174 | ||
d3d74453 | 1175 | hrtimer_run_pending(); |
82f67cd9 | 1176 | |
1da177e4 LT |
1177 | if (time_after_eq(jiffies, base->timer_jiffies)) |
1178 | __run_timers(base); | |
1179 | } | |
1180 | ||
1181 | /* | |
1182 | * Called by the local, per-CPU timer interrupt on SMP. | |
1183 | */ | |
1184 | void run_local_timers(void) | |
1185 | { | |
d3d74453 | 1186 | hrtimer_run_queues(); |
1da177e4 | 1187 | raise_softirq(TIMER_SOFTIRQ); |
6687a97d | 1188 | softlockup_tick(); |
1da177e4 LT |
1189 | } |
1190 | ||
1da177e4 LT |
1191 | /* |
1192 | * The 64-bit jiffies value is not atomic - you MUST NOT read it | |
1193 | * without sampling the sequence number in xtime_lock. | |
1194 | * jiffies is defined in the linker script... | |
1195 | */ | |
1196 | ||
3171a030 | 1197 | void do_timer(unsigned long ticks) |
1da177e4 | 1198 | { |
3171a030 | 1199 | jiffies_64 += ticks; |
dce48a84 TG |
1200 | update_wall_time(); |
1201 | calc_global_load(); | |
1da177e4 LT |
1202 | } |
1203 | ||
1204 | #ifdef __ARCH_WANT_SYS_ALARM | |
1205 | ||
1206 | /* | |
1207 | * For backwards compatibility? This can be done in libc so Alpha | |
1208 | * and all newer ports shouldn't need it. | |
1209 | */ | |
58fd3aa2 | 1210 | SYSCALL_DEFINE1(alarm, unsigned int, seconds) |
1da177e4 | 1211 | { |
c08b8a49 | 1212 | return alarm_setitimer(seconds); |
1da177e4 LT |
1213 | } |
1214 | ||
1215 | #endif | |
1216 | ||
1217 | #ifndef __alpha__ | |
1218 | ||
1219 | /* | |
1220 | * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this | |
1221 | * should be moved into arch/i386 instead? | |
1222 | */ | |
1223 | ||
1224 | /** | |
1225 | * sys_getpid - return the thread group id of the current process | |
1226 | * | |
1227 | * Note, despite the name, this returns the tgid not the pid. The tgid and | |
1228 | * the pid are identical unless CLONE_THREAD was specified on clone() in | |
1229 | * which case the tgid is the same in all threads of the same group. | |
1230 | * | |
1231 | * This is SMP safe as current->tgid does not change. | |
1232 | */ | |
58fd3aa2 | 1233 | SYSCALL_DEFINE0(getpid) |
1da177e4 | 1234 | { |
b488893a | 1235 | return task_tgid_vnr(current); |
1da177e4 LT |
1236 | } |
1237 | ||
1238 | /* | |
6997a6fa KK |
1239 | * Accessing ->real_parent is not SMP-safe, it could |
1240 | * change from under us. However, we can use a stale | |
1241 | * value of ->real_parent under rcu_read_lock(), see | |
1242 | * release_task()->call_rcu(delayed_put_task_struct). | |
1da177e4 | 1243 | */ |
dbf040d9 | 1244 | SYSCALL_DEFINE0(getppid) |
1da177e4 LT |
1245 | { |
1246 | int pid; | |
1da177e4 | 1247 | |
6997a6fa | 1248 | rcu_read_lock(); |
6c5f3e7b | 1249 | pid = task_tgid_vnr(current->real_parent); |
6997a6fa | 1250 | rcu_read_unlock(); |
1da177e4 | 1251 | |
1da177e4 LT |
1252 | return pid; |
1253 | } | |
1254 | ||
dbf040d9 | 1255 | SYSCALL_DEFINE0(getuid) |
1da177e4 LT |
1256 | { |
1257 | /* Only we change this so SMP safe */ | |
76aac0e9 | 1258 | return current_uid(); |
1da177e4 LT |
1259 | } |
1260 | ||
dbf040d9 | 1261 | SYSCALL_DEFINE0(geteuid) |
1da177e4 LT |
1262 | { |
1263 | /* Only we change this so SMP safe */ | |
76aac0e9 | 1264 | return current_euid(); |
1da177e4 LT |
1265 | } |
1266 | ||
dbf040d9 | 1267 | SYSCALL_DEFINE0(getgid) |
1da177e4 LT |
1268 | { |
1269 | /* Only we change this so SMP safe */ | |
76aac0e9 | 1270 | return current_gid(); |
1da177e4 LT |
1271 | } |
1272 | ||
dbf040d9 | 1273 | SYSCALL_DEFINE0(getegid) |
1da177e4 LT |
1274 | { |
1275 | /* Only we change this so SMP safe */ | |
76aac0e9 | 1276 | return current_egid(); |
1da177e4 LT |
1277 | } |
1278 | ||
1279 | #endif | |
1280 | ||
1281 | static void process_timeout(unsigned long __data) | |
1282 | { | |
36c8b586 | 1283 | wake_up_process((struct task_struct *)__data); |
1da177e4 LT |
1284 | } |
1285 | ||
1286 | /** | |
1287 | * schedule_timeout - sleep until timeout | |
1288 | * @timeout: timeout value in jiffies | |
1289 | * | |
1290 | * Make the current task sleep until @timeout jiffies have | |
1291 | * elapsed. The routine will return immediately unless | |
1292 | * the current task state has been set (see set_current_state()). | |
1293 | * | |
1294 | * You can set the task state as follows - | |
1295 | * | |
1296 | * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to | |
1297 | * pass before the routine returns. The routine will return 0 | |
1298 | * | |
1299 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
1300 | * delivered to the current task. In this case the remaining time | |
1301 | * in jiffies will be returned, or 0 if the timer expired in time | |
1302 | * | |
1303 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1304 | * routine returns. | |
1305 | * | |
1306 | * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule | |
1307 | * the CPU away without a bound on the timeout. In this case the return | |
1308 | * value will be %MAX_SCHEDULE_TIMEOUT. | |
1309 | * | |
1310 | * In all cases the return value is guaranteed to be non-negative. | |
1311 | */ | |
7ad5b3a5 | 1312 | signed long __sched schedule_timeout(signed long timeout) |
1da177e4 LT |
1313 | { |
1314 | struct timer_list timer; | |
1315 | unsigned long expire; | |
1316 | ||
1317 | switch (timeout) | |
1318 | { | |
1319 | case MAX_SCHEDULE_TIMEOUT: | |
1320 | /* | |
1321 | * These two special cases are useful to be comfortable | |
1322 | * in the caller. Nothing more. We could take | |
1323 | * MAX_SCHEDULE_TIMEOUT from one of the negative value | |
1324 | * but I' d like to return a valid offset (>=0) to allow | |
1325 | * the caller to do everything it want with the retval. | |
1326 | */ | |
1327 | schedule(); | |
1328 | goto out; | |
1329 | default: | |
1330 | /* | |
1331 | * Another bit of PARANOID. Note that the retval will be | |
1332 | * 0 since no piece of kernel is supposed to do a check | |
1333 | * for a negative retval of schedule_timeout() (since it | |
1334 | * should never happens anyway). You just have the printk() | |
1335 | * that will tell you if something is gone wrong and where. | |
1336 | */ | |
5b149bcc | 1337 | if (timeout < 0) { |
1da177e4 | 1338 | printk(KERN_ERR "schedule_timeout: wrong timeout " |
5b149bcc AM |
1339 | "value %lx\n", timeout); |
1340 | dump_stack(); | |
1da177e4 LT |
1341 | current->state = TASK_RUNNING; |
1342 | goto out; | |
1343 | } | |
1344 | } | |
1345 | ||
1346 | expire = timeout + jiffies; | |
1347 | ||
c6f3a97f | 1348 | setup_timer_on_stack(&timer, process_timeout, (unsigned long)current); |
597d0275 | 1349 | __mod_timer(&timer, expire, false, TIMER_NOT_PINNED); |
1da177e4 LT |
1350 | schedule(); |
1351 | del_singleshot_timer_sync(&timer); | |
1352 | ||
c6f3a97f TG |
1353 | /* Remove the timer from the object tracker */ |
1354 | destroy_timer_on_stack(&timer); | |
1355 | ||
1da177e4 LT |
1356 | timeout = expire - jiffies; |
1357 | ||
1358 | out: | |
1359 | return timeout < 0 ? 0 : timeout; | |
1360 | } | |
1da177e4 LT |
1361 | EXPORT_SYMBOL(schedule_timeout); |
1362 | ||
8a1c1757 AM |
1363 | /* |
1364 | * We can use __set_current_state() here because schedule_timeout() calls | |
1365 | * schedule() unconditionally. | |
1366 | */ | |
64ed93a2 NA |
1367 | signed long __sched schedule_timeout_interruptible(signed long timeout) |
1368 | { | |
a5a0d52c AM |
1369 | __set_current_state(TASK_INTERRUPTIBLE); |
1370 | return schedule_timeout(timeout); | |
64ed93a2 NA |
1371 | } |
1372 | EXPORT_SYMBOL(schedule_timeout_interruptible); | |
1373 | ||
294d5cc2 MW |
1374 | signed long __sched schedule_timeout_killable(signed long timeout) |
1375 | { | |
1376 | __set_current_state(TASK_KILLABLE); | |
1377 | return schedule_timeout(timeout); | |
1378 | } | |
1379 | EXPORT_SYMBOL(schedule_timeout_killable); | |
1380 | ||
64ed93a2 NA |
1381 | signed long __sched schedule_timeout_uninterruptible(signed long timeout) |
1382 | { | |
a5a0d52c AM |
1383 | __set_current_state(TASK_UNINTERRUPTIBLE); |
1384 | return schedule_timeout(timeout); | |
64ed93a2 NA |
1385 | } |
1386 | EXPORT_SYMBOL(schedule_timeout_uninterruptible); | |
1387 | ||
1da177e4 | 1388 | /* Thread ID - the internal kernel "pid" */ |
58fd3aa2 | 1389 | SYSCALL_DEFINE0(gettid) |
1da177e4 | 1390 | { |
b488893a | 1391 | return task_pid_vnr(current); |
1da177e4 LT |
1392 | } |
1393 | ||
2aae4a10 | 1394 | /** |
d4d23add | 1395 | * do_sysinfo - fill in sysinfo struct |
2aae4a10 | 1396 | * @info: pointer to buffer to fill |
6819457d | 1397 | */ |
d4d23add | 1398 | int do_sysinfo(struct sysinfo *info) |
1da177e4 | 1399 | { |
1da177e4 LT |
1400 | unsigned long mem_total, sav_total; |
1401 | unsigned int mem_unit, bitcount; | |
2d02494f | 1402 | struct timespec tp; |
1da177e4 | 1403 | |
d4d23add | 1404 | memset(info, 0, sizeof(struct sysinfo)); |
1da177e4 | 1405 | |
2d02494f TG |
1406 | ktime_get_ts(&tp); |
1407 | monotonic_to_bootbased(&tp); | |
1408 | info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0); | |
1da177e4 | 1409 | |
2d02494f | 1410 | get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT); |
1da177e4 | 1411 | |
2d02494f | 1412 | info->procs = nr_threads; |
1da177e4 | 1413 | |
d4d23add KM |
1414 | si_meminfo(info); |
1415 | si_swapinfo(info); | |
1da177e4 LT |
1416 | |
1417 | /* | |
1418 | * If the sum of all the available memory (i.e. ram + swap) | |
1419 | * is less than can be stored in a 32 bit unsigned long then | |
1420 | * we can be binary compatible with 2.2.x kernels. If not, | |
1421 | * well, in that case 2.2.x was broken anyways... | |
1422 | * | |
1423 | * -Erik Andersen <andersee@debian.org> | |
1424 | */ | |
1425 | ||
d4d23add KM |
1426 | mem_total = info->totalram + info->totalswap; |
1427 | if (mem_total < info->totalram || mem_total < info->totalswap) | |
1da177e4 LT |
1428 | goto out; |
1429 | bitcount = 0; | |
d4d23add | 1430 | mem_unit = info->mem_unit; |
1da177e4 LT |
1431 | while (mem_unit > 1) { |
1432 | bitcount++; | |
1433 | mem_unit >>= 1; | |
1434 | sav_total = mem_total; | |
1435 | mem_total <<= 1; | |
1436 | if (mem_total < sav_total) | |
1437 | goto out; | |
1438 | } | |
1439 | ||
1440 | /* | |
1441 | * If mem_total did not overflow, multiply all memory values by | |
d4d23add | 1442 | * info->mem_unit and set it to 1. This leaves things compatible |
1da177e4 LT |
1443 | * with 2.2.x, and also retains compatibility with earlier 2.4.x |
1444 | * kernels... | |
1445 | */ | |
1446 | ||
d4d23add KM |
1447 | info->mem_unit = 1; |
1448 | info->totalram <<= bitcount; | |
1449 | info->freeram <<= bitcount; | |
1450 | info->sharedram <<= bitcount; | |
1451 | info->bufferram <<= bitcount; | |
1452 | info->totalswap <<= bitcount; | |
1453 | info->freeswap <<= bitcount; | |
1454 | info->totalhigh <<= bitcount; | |
1455 | info->freehigh <<= bitcount; | |
1456 | ||
1457 | out: | |
1458 | return 0; | |
1459 | } | |
1460 | ||
1e7bfb21 | 1461 | SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info) |
d4d23add KM |
1462 | { |
1463 | struct sysinfo val; | |
1464 | ||
1465 | do_sysinfo(&val); | |
1da177e4 | 1466 | |
1da177e4 LT |
1467 | if (copy_to_user(info, &val, sizeof(struct sysinfo))) |
1468 | return -EFAULT; | |
1469 | ||
1470 | return 0; | |
1471 | } | |
1472 | ||
b4be6258 | 1473 | static int __cpuinit init_timers_cpu(int cpu) |
1da177e4 LT |
1474 | { |
1475 | int j; | |
a6fa8e5a | 1476 | struct tvec_base *base; |
b4be6258 | 1477 | static char __cpuinitdata tvec_base_done[NR_CPUS]; |
55c888d6 | 1478 | |
ba6edfcd | 1479 | if (!tvec_base_done[cpu]) { |
a4a6198b JB |
1480 | static char boot_done; |
1481 | ||
a4a6198b | 1482 | if (boot_done) { |
ba6edfcd AM |
1483 | /* |
1484 | * The APs use this path later in boot | |
1485 | */ | |
94f6030c CL |
1486 | base = kmalloc_node(sizeof(*base), |
1487 | GFP_KERNEL | __GFP_ZERO, | |
a4a6198b JB |
1488 | cpu_to_node(cpu)); |
1489 | if (!base) | |
1490 | return -ENOMEM; | |
6e453a67 VP |
1491 | |
1492 | /* Make sure that tvec_base is 2 byte aligned */ | |
1493 | if (tbase_get_deferrable(base)) { | |
1494 | WARN_ON(1); | |
1495 | kfree(base); | |
1496 | return -ENOMEM; | |
1497 | } | |
ba6edfcd | 1498 | per_cpu(tvec_bases, cpu) = base; |
a4a6198b | 1499 | } else { |
ba6edfcd AM |
1500 | /* |
1501 | * This is for the boot CPU - we use compile-time | |
1502 | * static initialisation because per-cpu memory isn't | |
1503 | * ready yet and because the memory allocators are not | |
1504 | * initialised either. | |
1505 | */ | |
a4a6198b | 1506 | boot_done = 1; |
ba6edfcd | 1507 | base = &boot_tvec_bases; |
a4a6198b | 1508 | } |
ba6edfcd AM |
1509 | tvec_base_done[cpu] = 1; |
1510 | } else { | |
1511 | base = per_cpu(tvec_bases, cpu); | |
a4a6198b | 1512 | } |
ba6edfcd | 1513 | |
3691c519 | 1514 | spin_lock_init(&base->lock); |
d730e882 | 1515 | |
1da177e4 LT |
1516 | for (j = 0; j < TVN_SIZE; j++) { |
1517 | INIT_LIST_HEAD(base->tv5.vec + j); | |
1518 | INIT_LIST_HEAD(base->tv4.vec + j); | |
1519 | INIT_LIST_HEAD(base->tv3.vec + j); | |
1520 | INIT_LIST_HEAD(base->tv2.vec + j); | |
1521 | } | |
1522 | for (j = 0; j < TVR_SIZE; j++) | |
1523 | INIT_LIST_HEAD(base->tv1.vec + j); | |
1524 | ||
1525 | base->timer_jiffies = jiffies; | |
a4a6198b | 1526 | return 0; |
1da177e4 LT |
1527 | } |
1528 | ||
1529 | #ifdef CONFIG_HOTPLUG_CPU | |
a6fa8e5a | 1530 | static void migrate_timer_list(struct tvec_base *new_base, struct list_head *head) |
1da177e4 LT |
1531 | { |
1532 | struct timer_list *timer; | |
1533 | ||
1534 | while (!list_empty(head)) { | |
b5e61818 | 1535 | timer = list_first_entry(head, struct timer_list, entry); |
55c888d6 | 1536 | detach_timer(timer, 0); |
6e453a67 | 1537 | timer_set_base(timer, new_base); |
1da177e4 | 1538 | internal_add_timer(new_base, timer); |
1da177e4 | 1539 | } |
1da177e4 LT |
1540 | } |
1541 | ||
48ccf3da | 1542 | static void __cpuinit migrate_timers(int cpu) |
1da177e4 | 1543 | { |
a6fa8e5a PM |
1544 | struct tvec_base *old_base; |
1545 | struct tvec_base *new_base; | |
1da177e4 LT |
1546 | int i; |
1547 | ||
1548 | BUG_ON(cpu_online(cpu)); | |
a4a6198b JB |
1549 | old_base = per_cpu(tvec_bases, cpu); |
1550 | new_base = get_cpu_var(tvec_bases); | |
d82f0b0f ON |
1551 | /* |
1552 | * The caller is globally serialized and nobody else | |
1553 | * takes two locks at once, deadlock is not possible. | |
1554 | */ | |
1555 | spin_lock_irq(&new_base->lock); | |
0d180406 | 1556 | spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING); |
3691c519 ON |
1557 | |
1558 | BUG_ON(old_base->running_timer); | |
1da177e4 | 1559 | |
1da177e4 | 1560 | for (i = 0; i < TVR_SIZE; i++) |
55c888d6 ON |
1561 | migrate_timer_list(new_base, old_base->tv1.vec + i); |
1562 | for (i = 0; i < TVN_SIZE; i++) { | |
1563 | migrate_timer_list(new_base, old_base->tv2.vec + i); | |
1564 | migrate_timer_list(new_base, old_base->tv3.vec + i); | |
1565 | migrate_timer_list(new_base, old_base->tv4.vec + i); | |
1566 | migrate_timer_list(new_base, old_base->tv5.vec + i); | |
1567 | } | |
1568 | ||
0d180406 | 1569 | spin_unlock(&old_base->lock); |
d82f0b0f | 1570 | spin_unlock_irq(&new_base->lock); |
1da177e4 | 1571 | put_cpu_var(tvec_bases); |
1da177e4 LT |
1572 | } |
1573 | #endif /* CONFIG_HOTPLUG_CPU */ | |
1574 | ||
8c78f307 | 1575 | static int __cpuinit timer_cpu_notify(struct notifier_block *self, |
1da177e4 LT |
1576 | unsigned long action, void *hcpu) |
1577 | { | |
1578 | long cpu = (long)hcpu; | |
1579 | switch(action) { | |
1580 | case CPU_UP_PREPARE: | |
8bb78442 | 1581 | case CPU_UP_PREPARE_FROZEN: |
a4a6198b JB |
1582 | if (init_timers_cpu(cpu) < 0) |
1583 | return NOTIFY_BAD; | |
1da177e4 LT |
1584 | break; |
1585 | #ifdef CONFIG_HOTPLUG_CPU | |
1586 | case CPU_DEAD: | |
8bb78442 | 1587 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
1588 | migrate_timers(cpu); |
1589 | break; | |
1590 | #endif | |
1591 | default: | |
1592 | break; | |
1593 | } | |
1594 | return NOTIFY_OK; | |
1595 | } | |
1596 | ||
8c78f307 | 1597 | static struct notifier_block __cpuinitdata timers_nb = { |
1da177e4 LT |
1598 | .notifier_call = timer_cpu_notify, |
1599 | }; | |
1600 | ||
1601 | ||
1602 | void __init init_timers(void) | |
1603 | { | |
07dccf33 | 1604 | int err = timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE, |
1da177e4 | 1605 | (void *)(long)smp_processor_id()); |
07dccf33 | 1606 | |
82f67cd9 IM |
1607 | init_timer_stats(); |
1608 | ||
07dccf33 | 1609 | BUG_ON(err == NOTIFY_BAD); |
1da177e4 | 1610 | register_cpu_notifier(&timers_nb); |
962cf36c | 1611 | open_softirq(TIMER_SOFTIRQ, run_timer_softirq); |
1da177e4 LT |
1612 | } |
1613 | ||
1da177e4 LT |
1614 | /** |
1615 | * msleep - sleep safely even with waitqueue interruptions | |
1616 | * @msecs: Time in milliseconds to sleep for | |
1617 | */ | |
1618 | void msleep(unsigned int msecs) | |
1619 | { | |
1620 | unsigned long timeout = msecs_to_jiffies(msecs) + 1; | |
1621 | ||
75bcc8c5 NA |
1622 | while (timeout) |
1623 | timeout = schedule_timeout_uninterruptible(timeout); | |
1da177e4 LT |
1624 | } |
1625 | ||
1626 | EXPORT_SYMBOL(msleep); | |
1627 | ||
1628 | /** | |
96ec3efd | 1629 | * msleep_interruptible - sleep waiting for signals |
1da177e4 LT |
1630 | * @msecs: Time in milliseconds to sleep for |
1631 | */ | |
1632 | unsigned long msleep_interruptible(unsigned int msecs) | |
1633 | { | |
1634 | unsigned long timeout = msecs_to_jiffies(msecs) + 1; | |
1635 | ||
75bcc8c5 NA |
1636 | while (timeout && !signal_pending(current)) |
1637 | timeout = schedule_timeout_interruptible(timeout); | |
1da177e4 LT |
1638 | return jiffies_to_msecs(timeout); |
1639 | } | |
1640 | ||
1641 | EXPORT_SYMBOL(msleep_interruptible); |