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> |
1da177e4 LT |
40 | |
41 | #include <asm/uaccess.h> | |
42 | #include <asm/unistd.h> | |
43 | #include <asm/div64.h> | |
44 | #include <asm/timex.h> | |
45 | #include <asm/io.h> | |
46 | ||
ecea8d19 TG |
47 | u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES; |
48 | ||
49 | EXPORT_SYMBOL(jiffies_64); | |
50 | ||
1da177e4 LT |
51 | /* |
52 | * per-CPU timer vector definitions: | |
53 | */ | |
1da177e4 LT |
54 | #define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6) |
55 | #define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8) | |
56 | #define TVN_SIZE (1 << TVN_BITS) | |
57 | #define TVR_SIZE (1 << TVR_BITS) | |
58 | #define TVN_MASK (TVN_SIZE - 1) | |
59 | #define TVR_MASK (TVR_SIZE - 1) | |
60 | ||
a6fa8e5a | 61 | struct tvec { |
1da177e4 | 62 | struct list_head vec[TVN_SIZE]; |
a6fa8e5a | 63 | }; |
1da177e4 | 64 | |
a6fa8e5a | 65 | struct tvec_root { |
1da177e4 | 66 | struct list_head vec[TVR_SIZE]; |
a6fa8e5a | 67 | }; |
1da177e4 | 68 | |
a6fa8e5a | 69 | struct tvec_base { |
3691c519 ON |
70 | spinlock_t lock; |
71 | struct timer_list *running_timer; | |
1da177e4 | 72 | unsigned long timer_jiffies; |
a6fa8e5a PM |
73 | struct tvec_root tv1; |
74 | struct tvec tv2; | |
75 | struct tvec tv3; | |
76 | struct tvec tv4; | |
77 | struct tvec tv5; | |
6e453a67 | 78 | } ____cacheline_aligned; |
1da177e4 | 79 | |
a6fa8e5a | 80 | struct tvec_base boot_tvec_bases; |
3691c519 | 81 | EXPORT_SYMBOL(boot_tvec_bases); |
a6fa8e5a | 82 | static DEFINE_PER_CPU(struct tvec_base *, tvec_bases) = &boot_tvec_bases; |
1da177e4 | 83 | |
6e453a67 | 84 | /* |
a6fa8e5a | 85 | * Note that all tvec_bases are 2 byte aligned and lower bit of |
6e453a67 VP |
86 | * base in timer_list is guaranteed to be zero. Use the LSB for |
87 | * the new flag to indicate whether the timer is deferrable | |
88 | */ | |
89 | #define TBASE_DEFERRABLE_FLAG (0x1) | |
90 | ||
91 | /* Functions below help us manage 'deferrable' flag */ | |
a6fa8e5a | 92 | static inline unsigned int tbase_get_deferrable(struct tvec_base *base) |
6e453a67 | 93 | { |
e9910846 | 94 | return ((unsigned int)(unsigned long)base & TBASE_DEFERRABLE_FLAG); |
6e453a67 VP |
95 | } |
96 | ||
a6fa8e5a | 97 | static inline struct tvec_base *tbase_get_base(struct tvec_base *base) |
6e453a67 | 98 | { |
a6fa8e5a | 99 | return ((struct tvec_base *)((unsigned long)base & ~TBASE_DEFERRABLE_FLAG)); |
6e453a67 VP |
100 | } |
101 | ||
102 | static inline void timer_set_deferrable(struct timer_list *timer) | |
103 | { | |
a6fa8e5a | 104 | timer->base = ((struct tvec_base *)((unsigned long)(timer->base) | |
6819457d | 105 | TBASE_DEFERRABLE_FLAG)); |
6e453a67 VP |
106 | } |
107 | ||
108 | static inline void | |
a6fa8e5a | 109 | timer_set_base(struct timer_list *timer, struct tvec_base *new_base) |
6e453a67 | 110 | { |
a6fa8e5a | 111 | timer->base = (struct tvec_base *)((unsigned long)(new_base) | |
6819457d | 112 | tbase_get_deferrable(timer->base)); |
6e453a67 VP |
113 | } |
114 | ||
9c133c46 AS |
115 | static unsigned long round_jiffies_common(unsigned long j, int cpu, |
116 | bool force_up) | |
4c36a5de AV |
117 | { |
118 | int rem; | |
119 | unsigned long original = j; | |
120 | ||
121 | /* | |
122 | * We don't want all cpus firing their timers at once hitting the | |
123 | * same lock or cachelines, so we skew each extra cpu with an extra | |
124 | * 3 jiffies. This 3 jiffies came originally from the mm/ code which | |
125 | * already did this. | |
126 | * The skew is done by adding 3*cpunr, then round, then subtract this | |
127 | * extra offset again. | |
128 | */ | |
129 | j += cpu * 3; | |
130 | ||
131 | rem = j % HZ; | |
132 | ||
133 | /* | |
134 | * If the target jiffie is just after a whole second (which can happen | |
135 | * due to delays of the timer irq, long irq off times etc etc) then | |
136 | * we should round down to the whole second, not up. Use 1/4th second | |
137 | * as cutoff for this rounding as an extreme upper bound for this. | |
9c133c46 | 138 | * But never round down if @force_up is set. |
4c36a5de | 139 | */ |
9c133c46 | 140 | if (rem < HZ/4 && !force_up) /* round down */ |
4c36a5de AV |
141 | j = j - rem; |
142 | else /* round up */ | |
143 | j = j - rem + HZ; | |
144 | ||
145 | /* now that we have rounded, subtract the extra skew again */ | |
146 | j -= cpu * 3; | |
147 | ||
148 | if (j <= jiffies) /* rounding ate our timeout entirely; */ | |
149 | return original; | |
150 | return j; | |
151 | } | |
9c133c46 AS |
152 | |
153 | /** | |
154 | * __round_jiffies - function to round jiffies to a full second | |
155 | * @j: the time in (absolute) jiffies that should be rounded | |
156 | * @cpu: the processor number on which the timeout will happen | |
157 | * | |
158 | * __round_jiffies() rounds an absolute time in the future (in jiffies) | |
159 | * up or down to (approximately) full seconds. This is useful for timers | |
160 | * for which the exact time they fire does not matter too much, as long as | |
161 | * they fire approximately every X seconds. | |
162 | * | |
163 | * By rounding these timers to whole seconds, all such timers will fire | |
164 | * at the same time, rather than at various times spread out. The goal | |
165 | * of this is to have the CPU wake up less, which saves power. | |
166 | * | |
167 | * The exact rounding is skewed for each processor to avoid all | |
168 | * processors firing at the exact same time, which could lead | |
169 | * to lock contention or spurious cache line bouncing. | |
170 | * | |
171 | * The return value is the rounded version of the @j parameter. | |
172 | */ | |
173 | unsigned long __round_jiffies(unsigned long j, int cpu) | |
174 | { | |
175 | return round_jiffies_common(j, cpu, false); | |
176 | } | |
4c36a5de AV |
177 | EXPORT_SYMBOL_GPL(__round_jiffies); |
178 | ||
179 | /** | |
180 | * __round_jiffies_relative - function to round jiffies to a full second | |
181 | * @j: the time in (relative) jiffies that should be rounded | |
182 | * @cpu: the processor number on which the timeout will happen | |
183 | * | |
72fd4a35 | 184 | * __round_jiffies_relative() rounds a time delta in the future (in jiffies) |
4c36a5de AV |
185 | * up or down to (approximately) full seconds. This is useful for timers |
186 | * for which the exact time they fire does not matter too much, as long as | |
187 | * they fire approximately every X seconds. | |
188 | * | |
189 | * By rounding these timers to whole seconds, all such timers will fire | |
190 | * at the same time, rather than at various times spread out. The goal | |
191 | * of this is to have the CPU wake up less, which saves power. | |
192 | * | |
193 | * The exact rounding is skewed for each processor to avoid all | |
194 | * processors firing at the exact same time, which could lead | |
195 | * to lock contention or spurious cache line bouncing. | |
196 | * | |
72fd4a35 | 197 | * The return value is the rounded version of the @j parameter. |
4c36a5de AV |
198 | */ |
199 | unsigned long __round_jiffies_relative(unsigned long j, int cpu) | |
200 | { | |
9c133c46 AS |
201 | unsigned long j0 = jiffies; |
202 | ||
203 | /* Use j0 because jiffies might change while we run */ | |
204 | return round_jiffies_common(j + j0, cpu, false) - j0; | |
4c36a5de AV |
205 | } |
206 | EXPORT_SYMBOL_GPL(__round_jiffies_relative); | |
207 | ||
208 | /** | |
209 | * round_jiffies - function to round jiffies to a full second | |
210 | * @j: the time in (absolute) jiffies that should be rounded | |
211 | * | |
72fd4a35 | 212 | * round_jiffies() rounds an absolute time in the future (in jiffies) |
4c36a5de AV |
213 | * up or down to (approximately) full seconds. This is useful for timers |
214 | * for which the exact time they fire does not matter too much, as long as | |
215 | * they fire approximately every X seconds. | |
216 | * | |
217 | * By rounding these timers to whole seconds, all such timers will fire | |
218 | * at the same time, rather than at various times spread out. The goal | |
219 | * of this is to have the CPU wake up less, which saves power. | |
220 | * | |
72fd4a35 | 221 | * The return value is the rounded version of the @j parameter. |
4c36a5de AV |
222 | */ |
223 | unsigned long round_jiffies(unsigned long j) | |
224 | { | |
9c133c46 | 225 | return round_jiffies_common(j, raw_smp_processor_id(), false); |
4c36a5de AV |
226 | } |
227 | EXPORT_SYMBOL_GPL(round_jiffies); | |
228 | ||
229 | /** | |
230 | * round_jiffies_relative - function to round jiffies to a full second | |
231 | * @j: the time in (relative) jiffies that should be rounded | |
232 | * | |
72fd4a35 | 233 | * round_jiffies_relative() rounds a time delta in the future (in jiffies) |
4c36a5de AV |
234 | * up or down to (approximately) full seconds. This is useful for timers |
235 | * for which the exact time they fire does not matter too much, as long as | |
236 | * they fire approximately every X seconds. | |
237 | * | |
238 | * By rounding these timers to whole seconds, all such timers will fire | |
239 | * at the same time, rather than at various times spread out. The goal | |
240 | * of this is to have the CPU wake up less, which saves power. | |
241 | * | |
72fd4a35 | 242 | * The return value is the rounded version of the @j parameter. |
4c36a5de AV |
243 | */ |
244 | unsigned long round_jiffies_relative(unsigned long j) | |
245 | { | |
246 | return __round_jiffies_relative(j, raw_smp_processor_id()); | |
247 | } | |
248 | EXPORT_SYMBOL_GPL(round_jiffies_relative); | |
249 | ||
9c133c46 AS |
250 | /** |
251 | * __round_jiffies_up - function to round jiffies up to a full second | |
252 | * @j: the time in (absolute) jiffies that should be rounded | |
253 | * @cpu: the processor number on which the timeout will happen | |
254 | * | |
255 | * This is the same as __round_jiffies() except that it will never | |
256 | * round down. This is useful for timeouts for which the exact time | |
257 | * of firing does not matter too much, as long as they don't fire too | |
258 | * early. | |
259 | */ | |
260 | unsigned long __round_jiffies_up(unsigned long j, int cpu) | |
261 | { | |
262 | return round_jiffies_common(j, cpu, true); | |
263 | } | |
264 | EXPORT_SYMBOL_GPL(__round_jiffies_up); | |
265 | ||
266 | /** | |
267 | * __round_jiffies_up_relative - function to round jiffies up to a full second | |
268 | * @j: the time in (relative) jiffies that should be rounded | |
269 | * @cpu: the processor number on which the timeout will happen | |
270 | * | |
271 | * This is the same as __round_jiffies_relative() except that it will never | |
272 | * round down. This is useful for timeouts for which the exact time | |
273 | * of firing does not matter too much, as long as they don't fire too | |
274 | * early. | |
275 | */ | |
276 | unsigned long __round_jiffies_up_relative(unsigned long j, int cpu) | |
277 | { | |
278 | unsigned long j0 = jiffies; | |
279 | ||
280 | /* Use j0 because jiffies might change while we run */ | |
281 | return round_jiffies_common(j + j0, cpu, true) - j0; | |
282 | } | |
283 | EXPORT_SYMBOL_GPL(__round_jiffies_up_relative); | |
284 | ||
285 | /** | |
286 | * round_jiffies_up - function to round jiffies up to a full second | |
287 | * @j: the time in (absolute) jiffies that should be rounded | |
288 | * | |
289 | * This is the same as round_jiffies() except that it will never | |
290 | * round down. This is useful for timeouts for which the exact time | |
291 | * of firing does not matter too much, as long as they don't fire too | |
292 | * early. | |
293 | */ | |
294 | unsigned long round_jiffies_up(unsigned long j) | |
295 | { | |
296 | return round_jiffies_common(j, raw_smp_processor_id(), true); | |
297 | } | |
298 | EXPORT_SYMBOL_GPL(round_jiffies_up); | |
299 | ||
300 | /** | |
301 | * round_jiffies_up_relative - function to round jiffies up to a full second | |
302 | * @j: the time in (relative) jiffies that should be rounded | |
303 | * | |
304 | * This is the same as round_jiffies_relative() except that it will never | |
305 | * round down. This is useful for timeouts for which the exact time | |
306 | * of firing does not matter too much, as long as they don't fire too | |
307 | * early. | |
308 | */ | |
309 | unsigned long round_jiffies_up_relative(unsigned long j) | |
310 | { | |
311 | return __round_jiffies_up_relative(j, raw_smp_processor_id()); | |
312 | } | |
313 | EXPORT_SYMBOL_GPL(round_jiffies_up_relative); | |
314 | ||
4c36a5de | 315 | |
a6fa8e5a | 316 | static inline void set_running_timer(struct tvec_base *base, |
1da177e4 LT |
317 | struct timer_list *timer) |
318 | { | |
319 | #ifdef CONFIG_SMP | |
3691c519 | 320 | base->running_timer = timer; |
1da177e4 LT |
321 | #endif |
322 | } | |
323 | ||
a6fa8e5a | 324 | static void internal_add_timer(struct tvec_base *base, struct timer_list *timer) |
1da177e4 LT |
325 | { |
326 | unsigned long expires = timer->expires; | |
327 | unsigned long idx = expires - base->timer_jiffies; | |
328 | struct list_head *vec; | |
329 | ||
330 | if (idx < TVR_SIZE) { | |
331 | int i = expires & TVR_MASK; | |
332 | vec = base->tv1.vec + i; | |
333 | } else if (idx < 1 << (TVR_BITS + TVN_BITS)) { | |
334 | int i = (expires >> TVR_BITS) & TVN_MASK; | |
335 | vec = base->tv2.vec + i; | |
336 | } else if (idx < 1 << (TVR_BITS + 2 * TVN_BITS)) { | |
337 | int i = (expires >> (TVR_BITS + TVN_BITS)) & TVN_MASK; | |
338 | vec = base->tv3.vec + i; | |
339 | } else if (idx < 1 << (TVR_BITS + 3 * TVN_BITS)) { | |
340 | int i = (expires >> (TVR_BITS + 2 * TVN_BITS)) & TVN_MASK; | |
341 | vec = base->tv4.vec + i; | |
342 | } else if ((signed long) idx < 0) { | |
343 | /* | |
344 | * Can happen if you add a timer with expires == jiffies, | |
345 | * or you set a timer to go off in the past | |
346 | */ | |
347 | vec = base->tv1.vec + (base->timer_jiffies & TVR_MASK); | |
348 | } else { | |
349 | int i; | |
350 | /* If the timeout is larger than 0xffffffff on 64-bit | |
351 | * architectures then we use the maximum timeout: | |
352 | */ | |
353 | if (idx > 0xffffffffUL) { | |
354 | idx = 0xffffffffUL; | |
355 | expires = idx + base->timer_jiffies; | |
356 | } | |
357 | i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK; | |
358 | vec = base->tv5.vec + i; | |
359 | } | |
360 | /* | |
361 | * Timers are FIFO: | |
362 | */ | |
363 | list_add_tail(&timer->entry, vec); | |
364 | } | |
365 | ||
82f67cd9 IM |
366 | #ifdef CONFIG_TIMER_STATS |
367 | void __timer_stats_timer_set_start_info(struct timer_list *timer, void *addr) | |
368 | { | |
369 | if (timer->start_site) | |
370 | return; | |
371 | ||
372 | timer->start_site = addr; | |
373 | memcpy(timer->start_comm, current->comm, TASK_COMM_LEN); | |
374 | timer->start_pid = current->pid; | |
375 | } | |
c5c061b8 VP |
376 | |
377 | static void timer_stats_account_timer(struct timer_list *timer) | |
378 | { | |
379 | unsigned int flag = 0; | |
380 | ||
381 | if (unlikely(tbase_get_deferrable(timer->base))) | |
382 | flag |= TIMER_STATS_FLAG_DEFERRABLE; | |
383 | ||
384 | timer_stats_update_stats(timer, timer->start_pid, timer->start_site, | |
385 | timer->function, timer->start_comm, flag); | |
386 | } | |
387 | ||
388 | #else | |
389 | static void timer_stats_account_timer(struct timer_list *timer) {} | |
82f67cd9 IM |
390 | #endif |
391 | ||
c6f3a97f TG |
392 | #ifdef CONFIG_DEBUG_OBJECTS_TIMERS |
393 | ||
394 | static struct debug_obj_descr timer_debug_descr; | |
395 | ||
396 | /* | |
397 | * fixup_init is called when: | |
398 | * - an active object is initialized | |
55c888d6 | 399 | */ |
c6f3a97f TG |
400 | static int timer_fixup_init(void *addr, enum debug_obj_state state) |
401 | { | |
402 | struct timer_list *timer = addr; | |
403 | ||
404 | switch (state) { | |
405 | case ODEBUG_STATE_ACTIVE: | |
406 | del_timer_sync(timer); | |
407 | debug_object_init(timer, &timer_debug_descr); | |
408 | return 1; | |
409 | default: | |
410 | return 0; | |
411 | } | |
412 | } | |
413 | ||
414 | /* | |
415 | * fixup_activate is called when: | |
416 | * - an active object is activated | |
417 | * - an unknown object is activated (might be a statically initialized object) | |
418 | */ | |
419 | static int timer_fixup_activate(void *addr, enum debug_obj_state state) | |
420 | { | |
421 | struct timer_list *timer = addr; | |
422 | ||
423 | switch (state) { | |
424 | ||
425 | case ODEBUG_STATE_NOTAVAILABLE: | |
426 | /* | |
427 | * This is not really a fixup. The timer was | |
428 | * statically initialized. We just make sure that it | |
429 | * is tracked in the object tracker. | |
430 | */ | |
431 | if (timer->entry.next == NULL && | |
432 | timer->entry.prev == TIMER_ENTRY_STATIC) { | |
433 | debug_object_init(timer, &timer_debug_descr); | |
434 | debug_object_activate(timer, &timer_debug_descr); | |
435 | return 0; | |
436 | } else { | |
437 | WARN_ON_ONCE(1); | |
438 | } | |
439 | return 0; | |
440 | ||
441 | case ODEBUG_STATE_ACTIVE: | |
442 | WARN_ON(1); | |
443 | ||
444 | default: | |
445 | return 0; | |
446 | } | |
447 | } | |
448 | ||
449 | /* | |
450 | * fixup_free is called when: | |
451 | * - an active object is freed | |
452 | */ | |
453 | static int timer_fixup_free(void *addr, enum debug_obj_state state) | |
454 | { | |
455 | struct timer_list *timer = addr; | |
456 | ||
457 | switch (state) { | |
458 | case ODEBUG_STATE_ACTIVE: | |
459 | del_timer_sync(timer); | |
460 | debug_object_free(timer, &timer_debug_descr); | |
461 | return 1; | |
462 | default: | |
463 | return 0; | |
464 | } | |
465 | } | |
466 | ||
467 | static struct debug_obj_descr timer_debug_descr = { | |
468 | .name = "timer_list", | |
469 | .fixup_init = timer_fixup_init, | |
470 | .fixup_activate = timer_fixup_activate, | |
471 | .fixup_free = timer_fixup_free, | |
472 | }; | |
473 | ||
474 | static inline void debug_timer_init(struct timer_list *timer) | |
475 | { | |
476 | debug_object_init(timer, &timer_debug_descr); | |
477 | } | |
478 | ||
479 | static inline void debug_timer_activate(struct timer_list *timer) | |
480 | { | |
481 | debug_object_activate(timer, &timer_debug_descr); | |
482 | } | |
483 | ||
484 | static inline void debug_timer_deactivate(struct timer_list *timer) | |
485 | { | |
486 | debug_object_deactivate(timer, &timer_debug_descr); | |
487 | } | |
488 | ||
489 | static inline void debug_timer_free(struct timer_list *timer) | |
490 | { | |
491 | debug_object_free(timer, &timer_debug_descr); | |
492 | } | |
493 | ||
6f2b9b9a JB |
494 | static void __init_timer(struct timer_list *timer, |
495 | const char *name, | |
496 | struct lock_class_key *key); | |
c6f3a97f | 497 | |
6f2b9b9a JB |
498 | void init_timer_on_stack_key(struct timer_list *timer, |
499 | const char *name, | |
500 | struct lock_class_key *key) | |
c6f3a97f TG |
501 | { |
502 | debug_object_init_on_stack(timer, &timer_debug_descr); | |
6f2b9b9a | 503 | __init_timer(timer, name, key); |
c6f3a97f | 504 | } |
6f2b9b9a | 505 | EXPORT_SYMBOL_GPL(init_timer_on_stack_key); |
c6f3a97f TG |
506 | |
507 | void destroy_timer_on_stack(struct timer_list *timer) | |
508 | { | |
509 | debug_object_free(timer, &timer_debug_descr); | |
510 | } | |
511 | EXPORT_SYMBOL_GPL(destroy_timer_on_stack); | |
512 | ||
513 | #else | |
514 | static inline void debug_timer_init(struct timer_list *timer) { } | |
515 | static inline void debug_timer_activate(struct timer_list *timer) { } | |
516 | static inline void debug_timer_deactivate(struct timer_list *timer) { } | |
517 | #endif | |
518 | ||
6f2b9b9a JB |
519 | static void __init_timer(struct timer_list *timer, |
520 | const char *name, | |
521 | struct lock_class_key *key) | |
55c888d6 ON |
522 | { |
523 | timer->entry.next = NULL; | |
bfe5d834 | 524 | timer->base = __raw_get_cpu_var(tvec_bases); |
82f67cd9 IM |
525 | #ifdef CONFIG_TIMER_STATS |
526 | timer->start_site = NULL; | |
527 | timer->start_pid = -1; | |
528 | memset(timer->start_comm, 0, TASK_COMM_LEN); | |
529 | #endif | |
6f2b9b9a | 530 | lockdep_init_map(&timer->lockdep_map, name, key, 0); |
55c888d6 | 531 | } |
c6f3a97f TG |
532 | |
533 | /** | |
633fe795 | 534 | * init_timer_key - initialize a timer |
c6f3a97f | 535 | * @timer: the timer to be initialized |
633fe795 RD |
536 | * @name: name of the timer |
537 | * @key: lockdep class key of the fake lock used for tracking timer | |
538 | * sync lock dependencies | |
c6f3a97f | 539 | * |
633fe795 | 540 | * init_timer_key() must be done to a timer prior calling *any* of the |
c6f3a97f TG |
541 | * other timer functions. |
542 | */ | |
6f2b9b9a JB |
543 | void init_timer_key(struct timer_list *timer, |
544 | const char *name, | |
545 | struct lock_class_key *key) | |
c6f3a97f TG |
546 | { |
547 | debug_timer_init(timer); | |
6f2b9b9a | 548 | __init_timer(timer, name, key); |
c6f3a97f | 549 | } |
6f2b9b9a | 550 | EXPORT_SYMBOL(init_timer_key); |
55c888d6 | 551 | |
6f2b9b9a JB |
552 | void init_timer_deferrable_key(struct timer_list *timer, |
553 | const char *name, | |
554 | struct lock_class_key *key) | |
6e453a67 | 555 | { |
6f2b9b9a | 556 | init_timer_key(timer, name, key); |
6e453a67 VP |
557 | timer_set_deferrable(timer); |
558 | } | |
6f2b9b9a | 559 | EXPORT_SYMBOL(init_timer_deferrable_key); |
6e453a67 | 560 | |
55c888d6 | 561 | static inline void detach_timer(struct timer_list *timer, |
82f67cd9 | 562 | int clear_pending) |
55c888d6 ON |
563 | { |
564 | struct list_head *entry = &timer->entry; | |
565 | ||
c6f3a97f TG |
566 | debug_timer_deactivate(timer); |
567 | ||
55c888d6 ON |
568 | __list_del(entry->prev, entry->next); |
569 | if (clear_pending) | |
570 | entry->next = NULL; | |
571 | entry->prev = LIST_POISON2; | |
572 | } | |
573 | ||
574 | /* | |
3691c519 | 575 | * We are using hashed locking: holding per_cpu(tvec_bases).lock |
55c888d6 ON |
576 | * means that all timers which are tied to this base via timer->base are |
577 | * locked, and the base itself is locked too. | |
578 | * | |
579 | * So __run_timers/migrate_timers can safely modify all timers which could | |
580 | * be found on ->tvX lists. | |
581 | * | |
582 | * When the timer's base is locked, and the timer removed from list, it is | |
583 | * possible to set timer->base = NULL and drop the lock: the timer remains | |
584 | * locked. | |
585 | */ | |
a6fa8e5a | 586 | static struct tvec_base *lock_timer_base(struct timer_list *timer, |
55c888d6 | 587 | unsigned long *flags) |
89e7e374 | 588 | __acquires(timer->base->lock) |
55c888d6 | 589 | { |
a6fa8e5a | 590 | struct tvec_base *base; |
55c888d6 ON |
591 | |
592 | for (;;) { | |
a6fa8e5a | 593 | struct tvec_base *prelock_base = timer->base; |
6e453a67 | 594 | base = tbase_get_base(prelock_base); |
55c888d6 ON |
595 | if (likely(base != NULL)) { |
596 | spin_lock_irqsave(&base->lock, *flags); | |
6e453a67 | 597 | if (likely(prelock_base == timer->base)) |
55c888d6 ON |
598 | return base; |
599 | /* The timer has migrated to another CPU */ | |
600 | spin_unlock_irqrestore(&base->lock, *flags); | |
601 | } | |
602 | cpu_relax(); | |
603 | } | |
604 | } | |
605 | ||
74019224 | 606 | static inline int |
597d0275 AB |
607 | __mod_timer(struct timer_list *timer, unsigned long expires, |
608 | bool pending_only, int pinned) | |
1da177e4 | 609 | { |
a6fa8e5a | 610 | struct tvec_base *base, *new_base; |
1da177e4 | 611 | unsigned long flags; |
74019224 IM |
612 | int ret; |
613 | ||
614 | ret = 0; | |
1da177e4 | 615 | |
82f67cd9 | 616 | timer_stats_timer_set_start_info(timer); |
1da177e4 | 617 | BUG_ON(!timer->function); |
1da177e4 | 618 | |
55c888d6 ON |
619 | base = lock_timer_base(timer, &flags); |
620 | ||
621 | if (timer_pending(timer)) { | |
622 | detach_timer(timer, 0); | |
623 | ret = 1; | |
74019224 IM |
624 | } else { |
625 | if (pending_only) | |
626 | goto out_unlock; | |
55c888d6 ON |
627 | } |
628 | ||
c6f3a97f TG |
629 | debug_timer_activate(timer); |
630 | ||
a4a6198b | 631 | new_base = __get_cpu_var(tvec_bases); |
1da177e4 | 632 | |
3691c519 | 633 | if (base != new_base) { |
1da177e4 | 634 | /* |
55c888d6 ON |
635 | * We are trying to schedule the timer on the local CPU. |
636 | * However we can't change timer's base while it is running, | |
637 | * otherwise del_timer_sync() can't detect that the timer's | |
638 | * handler yet has not finished. This also guarantees that | |
639 | * the timer is serialized wrt itself. | |
1da177e4 | 640 | */ |
a2c348fe | 641 | if (likely(base->running_timer != timer)) { |
55c888d6 | 642 | /* See the comment in lock_timer_base() */ |
6e453a67 | 643 | timer_set_base(timer, NULL); |
55c888d6 | 644 | spin_unlock(&base->lock); |
a2c348fe ON |
645 | base = new_base; |
646 | spin_lock(&base->lock); | |
6e453a67 | 647 | timer_set_base(timer, base); |
1da177e4 LT |
648 | } |
649 | } | |
650 | ||
1da177e4 | 651 | timer->expires = expires; |
a2c348fe | 652 | internal_add_timer(base, timer); |
74019224 IM |
653 | |
654 | out_unlock: | |
a2c348fe | 655 | spin_unlock_irqrestore(&base->lock, flags); |
1da177e4 LT |
656 | |
657 | return ret; | |
658 | } | |
659 | ||
2aae4a10 | 660 | /** |
74019224 IM |
661 | * mod_timer_pending - modify a pending timer's timeout |
662 | * @timer: the pending timer to be modified | |
663 | * @expires: new timeout in jiffies | |
1da177e4 | 664 | * |
74019224 IM |
665 | * mod_timer_pending() is the same for pending timers as mod_timer(), |
666 | * but will not re-activate and modify already deleted timers. | |
667 | * | |
668 | * It is useful for unserialized use of timers. | |
1da177e4 | 669 | */ |
74019224 | 670 | int mod_timer_pending(struct timer_list *timer, unsigned long expires) |
1da177e4 | 671 | { |
597d0275 | 672 | return __mod_timer(timer, expires, true, TIMER_NOT_PINNED); |
1da177e4 | 673 | } |
74019224 | 674 | EXPORT_SYMBOL(mod_timer_pending); |
1da177e4 | 675 | |
2aae4a10 | 676 | /** |
1da177e4 LT |
677 | * mod_timer - modify a timer's timeout |
678 | * @timer: the timer to be modified | |
2aae4a10 | 679 | * @expires: new timeout in jiffies |
1da177e4 | 680 | * |
72fd4a35 | 681 | * mod_timer() is a more efficient way to update the expire field of an |
1da177e4 LT |
682 | * active timer (if the timer is inactive it will be activated) |
683 | * | |
684 | * mod_timer(timer, expires) is equivalent to: | |
685 | * | |
686 | * del_timer(timer); timer->expires = expires; add_timer(timer); | |
687 | * | |
688 | * Note that if there are multiple unserialized concurrent users of the | |
689 | * same timer, then mod_timer() is the only safe way to modify the timeout, | |
690 | * since add_timer() cannot modify an already running timer. | |
691 | * | |
692 | * The function returns whether it has modified a pending timer or not. | |
693 | * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an | |
694 | * active timer returns 1.) | |
695 | */ | |
696 | int mod_timer(struct timer_list *timer, unsigned long expires) | |
697 | { | |
1da177e4 LT |
698 | /* |
699 | * This is a common optimization triggered by the | |
700 | * networking code - if the timer is re-modified | |
701 | * to be the same thing then just return: | |
702 | */ | |
703 | if (timer->expires == expires && timer_pending(timer)) | |
704 | return 1; | |
705 | ||
597d0275 | 706 | return __mod_timer(timer, expires, false, TIMER_NOT_PINNED); |
1da177e4 | 707 | } |
1da177e4 LT |
708 | EXPORT_SYMBOL(mod_timer); |
709 | ||
597d0275 AB |
710 | /** |
711 | * mod_timer_pinned - modify a timer's timeout | |
712 | * @timer: the timer to be modified | |
713 | * @expires: new timeout in jiffies | |
714 | * | |
715 | * mod_timer_pinned() is a way to update the expire field of an | |
716 | * active timer (if the timer is inactive it will be activated) | |
717 | * and not allow the timer to be migrated to a different CPU. | |
718 | * | |
719 | * mod_timer_pinned(timer, expires) is equivalent to: | |
720 | * | |
721 | * del_timer(timer); timer->expires = expires; add_timer(timer); | |
722 | */ | |
723 | int mod_timer_pinned(struct timer_list *timer, unsigned long expires) | |
724 | { | |
725 | if (timer->expires == expires && timer_pending(timer)) | |
726 | return 1; | |
727 | ||
728 | return __mod_timer(timer, expires, false, TIMER_PINNED); | |
729 | } | |
730 | EXPORT_SYMBOL(mod_timer_pinned); | |
731 | ||
74019224 IM |
732 | /** |
733 | * add_timer - start a timer | |
734 | * @timer: the timer to be added | |
735 | * | |
736 | * The kernel will do a ->function(->data) callback from the | |
737 | * timer interrupt at the ->expires point in the future. The | |
738 | * current time is 'jiffies'. | |
739 | * | |
740 | * The timer's ->expires, ->function (and if the handler uses it, ->data) | |
741 | * fields must be set prior calling this function. | |
742 | * | |
743 | * Timers with an ->expires field in the past will be executed in the next | |
744 | * timer tick. | |
745 | */ | |
746 | void add_timer(struct timer_list *timer) | |
747 | { | |
748 | BUG_ON(timer_pending(timer)); | |
749 | mod_timer(timer, timer->expires); | |
750 | } | |
751 | EXPORT_SYMBOL(add_timer); | |
752 | ||
753 | /** | |
754 | * add_timer_on - start a timer on a particular CPU | |
755 | * @timer: the timer to be added | |
756 | * @cpu: the CPU to start it on | |
757 | * | |
758 | * This is not very scalable on SMP. Double adds are not possible. | |
759 | */ | |
760 | void add_timer_on(struct timer_list *timer, int cpu) | |
761 | { | |
762 | struct tvec_base *base = per_cpu(tvec_bases, cpu); | |
763 | unsigned long flags; | |
764 | ||
765 | timer_stats_timer_set_start_info(timer); | |
766 | BUG_ON(timer_pending(timer) || !timer->function); | |
767 | spin_lock_irqsave(&base->lock, flags); | |
768 | timer_set_base(timer, base); | |
769 | debug_timer_activate(timer); | |
770 | internal_add_timer(base, timer); | |
771 | /* | |
772 | * Check whether the other CPU is idle and needs to be | |
773 | * triggered to reevaluate the timer wheel when nohz is | |
774 | * active. We are protected against the other CPU fiddling | |
775 | * with the timer by holding the timer base lock. This also | |
776 | * makes sure that a CPU on the way to idle can not evaluate | |
777 | * the timer wheel. | |
778 | */ | |
779 | wake_up_idle_cpu(cpu); | |
780 | spin_unlock_irqrestore(&base->lock, flags); | |
781 | } | |
782 | ||
2aae4a10 | 783 | /** |
1da177e4 LT |
784 | * del_timer - deactive a timer. |
785 | * @timer: the timer to be deactivated | |
786 | * | |
787 | * del_timer() deactivates a timer - this works on both active and inactive | |
788 | * timers. | |
789 | * | |
790 | * The function returns whether it has deactivated a pending timer or not. | |
791 | * (ie. del_timer() of an inactive timer returns 0, del_timer() of an | |
792 | * active timer returns 1.) | |
793 | */ | |
794 | int del_timer(struct timer_list *timer) | |
795 | { | |
a6fa8e5a | 796 | struct tvec_base *base; |
1da177e4 | 797 | unsigned long flags; |
55c888d6 | 798 | int ret = 0; |
1da177e4 | 799 | |
82f67cd9 | 800 | timer_stats_timer_clear_start_info(timer); |
55c888d6 ON |
801 | if (timer_pending(timer)) { |
802 | base = lock_timer_base(timer, &flags); | |
803 | if (timer_pending(timer)) { | |
804 | detach_timer(timer, 1); | |
805 | ret = 1; | |
806 | } | |
1da177e4 | 807 | spin_unlock_irqrestore(&base->lock, flags); |
1da177e4 | 808 | } |
1da177e4 | 809 | |
55c888d6 | 810 | return ret; |
1da177e4 | 811 | } |
1da177e4 LT |
812 | EXPORT_SYMBOL(del_timer); |
813 | ||
814 | #ifdef CONFIG_SMP | |
2aae4a10 REB |
815 | /** |
816 | * try_to_del_timer_sync - Try to deactivate a timer | |
817 | * @timer: timer do del | |
818 | * | |
fd450b73 ON |
819 | * This function tries to deactivate a timer. Upon successful (ret >= 0) |
820 | * exit the timer is not queued and the handler is not running on any CPU. | |
821 | * | |
822 | * It must not be called from interrupt contexts. | |
823 | */ | |
824 | int try_to_del_timer_sync(struct timer_list *timer) | |
825 | { | |
a6fa8e5a | 826 | struct tvec_base *base; |
fd450b73 ON |
827 | unsigned long flags; |
828 | int ret = -1; | |
829 | ||
830 | base = lock_timer_base(timer, &flags); | |
831 | ||
832 | if (base->running_timer == timer) | |
833 | goto out; | |
834 | ||
835 | ret = 0; | |
836 | if (timer_pending(timer)) { | |
837 | detach_timer(timer, 1); | |
838 | ret = 1; | |
839 | } | |
840 | out: | |
841 | spin_unlock_irqrestore(&base->lock, flags); | |
842 | ||
843 | return ret; | |
844 | } | |
e19dff1f DH |
845 | EXPORT_SYMBOL(try_to_del_timer_sync); |
846 | ||
2aae4a10 | 847 | /** |
1da177e4 LT |
848 | * del_timer_sync - deactivate a timer and wait for the handler to finish. |
849 | * @timer: the timer to be deactivated | |
850 | * | |
851 | * This function only differs from del_timer() on SMP: besides deactivating | |
852 | * the timer it also makes sure the handler has finished executing on other | |
853 | * CPUs. | |
854 | * | |
72fd4a35 | 855 | * Synchronization rules: Callers must prevent restarting of the timer, |
1da177e4 LT |
856 | * otherwise this function is meaningless. It must not be called from |
857 | * interrupt contexts. The caller must not hold locks which would prevent | |
55c888d6 ON |
858 | * completion of the timer's handler. The timer's handler must not call |
859 | * add_timer_on(). Upon exit the timer is not queued and the handler is | |
860 | * not running on any CPU. | |
1da177e4 LT |
861 | * |
862 | * The function returns whether it has deactivated a pending timer or not. | |
1da177e4 LT |
863 | */ |
864 | int del_timer_sync(struct timer_list *timer) | |
865 | { | |
6f2b9b9a JB |
866 | #ifdef CONFIG_LOCKDEP |
867 | unsigned long flags; | |
868 | ||
869 | local_irq_save(flags); | |
870 | lock_map_acquire(&timer->lockdep_map); | |
871 | lock_map_release(&timer->lockdep_map); | |
872 | local_irq_restore(flags); | |
873 | #endif | |
874 | ||
fd450b73 ON |
875 | for (;;) { |
876 | int ret = try_to_del_timer_sync(timer); | |
877 | if (ret >= 0) | |
878 | return ret; | |
a0009652 | 879 | cpu_relax(); |
fd450b73 | 880 | } |
1da177e4 | 881 | } |
55c888d6 | 882 | EXPORT_SYMBOL(del_timer_sync); |
1da177e4 LT |
883 | #endif |
884 | ||
a6fa8e5a | 885 | static int cascade(struct tvec_base *base, struct tvec *tv, int index) |
1da177e4 LT |
886 | { |
887 | /* cascade all the timers from tv up one level */ | |
3439dd86 P |
888 | struct timer_list *timer, *tmp; |
889 | struct list_head tv_list; | |
890 | ||
891 | list_replace_init(tv->vec + index, &tv_list); | |
1da177e4 | 892 | |
1da177e4 | 893 | /* |
3439dd86 P |
894 | * We are removing _all_ timers from the list, so we |
895 | * don't have to detach them individually. | |
1da177e4 | 896 | */ |
3439dd86 | 897 | list_for_each_entry_safe(timer, tmp, &tv_list, entry) { |
6e453a67 | 898 | BUG_ON(tbase_get_base(timer->base) != base); |
3439dd86 | 899 | internal_add_timer(base, timer); |
1da177e4 | 900 | } |
1da177e4 LT |
901 | |
902 | return index; | |
903 | } | |
904 | ||
2aae4a10 REB |
905 | #define INDEX(N) ((base->timer_jiffies >> (TVR_BITS + (N) * TVN_BITS)) & TVN_MASK) |
906 | ||
907 | /** | |
1da177e4 LT |
908 | * __run_timers - run all expired timers (if any) on this CPU. |
909 | * @base: the timer vector to be processed. | |
910 | * | |
911 | * This function cascades all vectors and executes all expired timer | |
912 | * vectors. | |
913 | */ | |
a6fa8e5a | 914 | static inline void __run_timers(struct tvec_base *base) |
1da177e4 LT |
915 | { |
916 | struct timer_list *timer; | |
917 | ||
3691c519 | 918 | spin_lock_irq(&base->lock); |
1da177e4 | 919 | while (time_after_eq(jiffies, base->timer_jiffies)) { |
626ab0e6 | 920 | struct list_head work_list; |
1da177e4 | 921 | struct list_head *head = &work_list; |
6819457d | 922 | int index = base->timer_jiffies & TVR_MASK; |
626ab0e6 | 923 | |
1da177e4 LT |
924 | /* |
925 | * Cascade timers: | |
926 | */ | |
927 | if (!index && | |
928 | (!cascade(base, &base->tv2, INDEX(0))) && | |
929 | (!cascade(base, &base->tv3, INDEX(1))) && | |
930 | !cascade(base, &base->tv4, INDEX(2))) | |
931 | cascade(base, &base->tv5, INDEX(3)); | |
626ab0e6 ON |
932 | ++base->timer_jiffies; |
933 | list_replace_init(base->tv1.vec + index, &work_list); | |
55c888d6 | 934 | while (!list_empty(head)) { |
1da177e4 LT |
935 | void (*fn)(unsigned long); |
936 | unsigned long data; | |
937 | ||
b5e61818 | 938 | timer = list_first_entry(head, struct timer_list,entry); |
6819457d TG |
939 | fn = timer->function; |
940 | data = timer->data; | |
1da177e4 | 941 | |
82f67cd9 IM |
942 | timer_stats_account_timer(timer); |
943 | ||
1da177e4 | 944 | set_running_timer(base, timer); |
55c888d6 | 945 | detach_timer(timer, 1); |
6f2b9b9a | 946 | |
3691c519 | 947 | spin_unlock_irq(&base->lock); |
1da177e4 | 948 | { |
be5b4fbd | 949 | int preempt_count = preempt_count(); |
6f2b9b9a JB |
950 | |
951 | #ifdef CONFIG_LOCKDEP | |
952 | /* | |
953 | * It is permissible to free the timer from | |
954 | * inside the function that is called from | |
955 | * it, this we need to take into account for | |
956 | * lockdep too. To avoid bogus "held lock | |
957 | * freed" warnings as well as problems when | |
958 | * looking into timer->lockdep_map, make a | |
959 | * copy and use that here. | |
960 | */ | |
961 | struct lockdep_map lockdep_map = | |
962 | timer->lockdep_map; | |
963 | #endif | |
964 | /* | |
965 | * Couple the lock chain with the lock chain at | |
966 | * del_timer_sync() by acquiring the lock_map | |
967 | * around the fn() call here and in | |
968 | * del_timer_sync(). | |
969 | */ | |
970 | lock_map_acquire(&lockdep_map); | |
971 | ||
1da177e4 | 972 | fn(data); |
6f2b9b9a JB |
973 | |
974 | lock_map_release(&lockdep_map); | |
975 | ||
1da177e4 | 976 | if (preempt_count != preempt_count()) { |
4c9dc641 | 977 | printk(KERN_ERR "huh, entered %p " |
be5b4fbd JJ |
978 | "with preempt_count %08x, exited" |
979 | " with %08x?\n", | |
980 | fn, preempt_count, | |
981 | preempt_count()); | |
1da177e4 LT |
982 | BUG(); |
983 | } | |
984 | } | |
3691c519 | 985 | spin_lock_irq(&base->lock); |
1da177e4 LT |
986 | } |
987 | } | |
988 | set_running_timer(base, NULL); | |
3691c519 | 989 | spin_unlock_irq(&base->lock); |
1da177e4 LT |
990 | } |
991 | ||
ee9c5785 | 992 | #ifdef CONFIG_NO_HZ |
1da177e4 LT |
993 | /* |
994 | * Find out when the next timer event is due to happen. This | |
995 | * is used on S/390 to stop all activity when a cpus is idle. | |
996 | * This functions needs to be called disabled. | |
997 | */ | |
a6fa8e5a | 998 | static unsigned long __next_timer_interrupt(struct tvec_base *base) |
1da177e4 | 999 | { |
1cfd6849 | 1000 | unsigned long timer_jiffies = base->timer_jiffies; |
eaad084b | 1001 | unsigned long expires = timer_jiffies + NEXT_TIMER_MAX_DELTA; |
1cfd6849 | 1002 | int index, slot, array, found = 0; |
1da177e4 | 1003 | struct timer_list *nte; |
a6fa8e5a | 1004 | struct tvec *varray[4]; |
1da177e4 LT |
1005 | |
1006 | /* Look for timer events in tv1. */ | |
1cfd6849 | 1007 | index = slot = timer_jiffies & TVR_MASK; |
1da177e4 | 1008 | do { |
1cfd6849 | 1009 | list_for_each_entry(nte, base->tv1.vec + slot, entry) { |
6819457d TG |
1010 | if (tbase_get_deferrable(nte->base)) |
1011 | continue; | |
6e453a67 | 1012 | |
1cfd6849 | 1013 | found = 1; |
1da177e4 | 1014 | expires = nte->expires; |
1cfd6849 TG |
1015 | /* Look at the cascade bucket(s)? */ |
1016 | if (!index || slot < index) | |
1017 | goto cascade; | |
1018 | return expires; | |
1da177e4 | 1019 | } |
1cfd6849 TG |
1020 | slot = (slot + 1) & TVR_MASK; |
1021 | } while (slot != index); | |
1022 | ||
1023 | cascade: | |
1024 | /* Calculate the next cascade event */ | |
1025 | if (index) | |
1026 | timer_jiffies += TVR_SIZE - index; | |
1027 | timer_jiffies >>= TVR_BITS; | |
1da177e4 LT |
1028 | |
1029 | /* Check tv2-tv5. */ | |
1030 | varray[0] = &base->tv2; | |
1031 | varray[1] = &base->tv3; | |
1032 | varray[2] = &base->tv4; | |
1033 | varray[3] = &base->tv5; | |
1cfd6849 TG |
1034 | |
1035 | for (array = 0; array < 4; array++) { | |
a6fa8e5a | 1036 | struct tvec *varp = varray[array]; |
1cfd6849 TG |
1037 | |
1038 | index = slot = timer_jiffies & TVN_MASK; | |
1da177e4 | 1039 | do { |
1cfd6849 | 1040 | list_for_each_entry(nte, varp->vec + slot, entry) { |
a0419888 JH |
1041 | if (tbase_get_deferrable(nte->base)) |
1042 | continue; | |
1043 | ||
1cfd6849 | 1044 | found = 1; |
1da177e4 LT |
1045 | if (time_before(nte->expires, expires)) |
1046 | expires = nte->expires; | |
1cfd6849 TG |
1047 | } |
1048 | /* | |
1049 | * Do we still search for the first timer or are | |
1050 | * we looking up the cascade buckets ? | |
1051 | */ | |
1052 | if (found) { | |
1053 | /* Look at the cascade bucket(s)? */ | |
1054 | if (!index || slot < index) | |
1055 | break; | |
1056 | return expires; | |
1057 | } | |
1058 | slot = (slot + 1) & TVN_MASK; | |
1059 | } while (slot != index); | |
1060 | ||
1061 | if (index) | |
1062 | timer_jiffies += TVN_SIZE - index; | |
1063 | timer_jiffies >>= TVN_BITS; | |
1da177e4 | 1064 | } |
1cfd6849 TG |
1065 | return expires; |
1066 | } | |
69239749 | 1067 | |
1cfd6849 TG |
1068 | /* |
1069 | * Check, if the next hrtimer event is before the next timer wheel | |
1070 | * event: | |
1071 | */ | |
1072 | static unsigned long cmp_next_hrtimer_event(unsigned long now, | |
1073 | unsigned long expires) | |
1074 | { | |
1075 | ktime_t hr_delta = hrtimer_get_next_event(); | |
1076 | struct timespec tsdelta; | |
9501b6cf | 1077 | unsigned long delta; |
1cfd6849 TG |
1078 | |
1079 | if (hr_delta.tv64 == KTIME_MAX) | |
1080 | return expires; | |
0662b713 | 1081 | |
9501b6cf TG |
1082 | /* |
1083 | * Expired timer available, let it expire in the next tick | |
1084 | */ | |
1085 | if (hr_delta.tv64 <= 0) | |
1086 | return now + 1; | |
69239749 | 1087 | |
1cfd6849 | 1088 | tsdelta = ktime_to_timespec(hr_delta); |
9501b6cf | 1089 | delta = timespec_to_jiffies(&tsdelta); |
eaad084b TG |
1090 | |
1091 | /* | |
1092 | * Limit the delta to the max value, which is checked in | |
1093 | * tick_nohz_stop_sched_tick(): | |
1094 | */ | |
1095 | if (delta > NEXT_TIMER_MAX_DELTA) | |
1096 | delta = NEXT_TIMER_MAX_DELTA; | |
1097 | ||
9501b6cf TG |
1098 | /* |
1099 | * Take rounding errors in to account and make sure, that it | |
1100 | * expires in the next tick. Otherwise we go into an endless | |
1101 | * ping pong due to tick_nohz_stop_sched_tick() retriggering | |
1102 | * the timer softirq | |
1103 | */ | |
1104 | if (delta < 1) | |
1105 | delta = 1; | |
1106 | now += delta; | |
1cfd6849 TG |
1107 | if (time_before(now, expires)) |
1108 | return now; | |
1da177e4 LT |
1109 | return expires; |
1110 | } | |
1cfd6849 TG |
1111 | |
1112 | /** | |
8dce39c2 | 1113 | * get_next_timer_interrupt - return the jiffy of the next pending timer |
05fb6bf0 | 1114 | * @now: current time (in jiffies) |
1cfd6849 | 1115 | */ |
fd064b9b | 1116 | unsigned long get_next_timer_interrupt(unsigned long now) |
1cfd6849 | 1117 | { |
a6fa8e5a | 1118 | struct tvec_base *base = __get_cpu_var(tvec_bases); |
fd064b9b | 1119 | unsigned long expires; |
1cfd6849 TG |
1120 | |
1121 | spin_lock(&base->lock); | |
1122 | expires = __next_timer_interrupt(base); | |
1123 | spin_unlock(&base->lock); | |
1124 | ||
1125 | if (time_before_eq(expires, now)) | |
1126 | return now; | |
1127 | ||
1128 | return cmp_next_hrtimer_event(now, expires); | |
1129 | } | |
1da177e4 LT |
1130 | #endif |
1131 | ||
1da177e4 | 1132 | /* |
5b4db0c2 | 1133 | * Called from the timer interrupt handler to charge one tick to the current |
1da177e4 LT |
1134 | * process. user_tick is 1 if the tick is user time, 0 for system. |
1135 | */ | |
1136 | void update_process_times(int user_tick) | |
1137 | { | |
1138 | struct task_struct *p = current; | |
1139 | int cpu = smp_processor_id(); | |
1140 | ||
1141 | /* Note: this timer irq context must be accounted for as well. */ | |
fa13a5a1 | 1142 | account_process_tick(p, user_tick); |
1da177e4 LT |
1143 | run_local_timers(); |
1144 | if (rcu_pending(cpu)) | |
1145 | rcu_check_callbacks(cpu, user_tick); | |
b845b517 | 1146 | printk_tick(); |
1da177e4 | 1147 | scheduler_tick(); |
6819457d | 1148 | run_posix_cpu_timers(p); |
1da177e4 LT |
1149 | } |
1150 | ||
1151 | /* | |
1152 | * Nr of active tasks - counted in fixed-point numbers | |
1153 | */ | |
1154 | static unsigned long count_active_tasks(void) | |
1155 | { | |
db1b1fef | 1156 | return nr_active() * FIXED_1; |
1da177e4 LT |
1157 | } |
1158 | ||
1159 | /* | |
1160 | * Hmm.. Changed this, as the GNU make sources (load.c) seems to | |
1161 | * imply that avenrun[] is the standard name for this kind of thing. | |
1162 | * Nothing else seems to be standardized: the fractional size etc | |
1163 | * all seem to differ on different machines. | |
1164 | * | |
1165 | * Requires xtime_lock to access. | |
1166 | */ | |
1167 | unsigned long avenrun[3]; | |
1168 | ||
1169 | EXPORT_SYMBOL(avenrun); | |
1170 | ||
1171 | /* | |
1172 | * calc_load - given tick count, update the avenrun load estimates. | |
1173 | * This is called while holding a write_lock on xtime_lock. | |
1174 | */ | |
1175 | static inline void calc_load(unsigned long ticks) | |
1176 | { | |
1177 | unsigned long active_tasks; /* fixed-point */ | |
1178 | static int count = LOAD_FREQ; | |
1179 | ||
cd7175ed ED |
1180 | count -= ticks; |
1181 | if (unlikely(count < 0)) { | |
1182 | active_tasks = count_active_tasks(); | |
1183 | do { | |
1184 | CALC_LOAD(avenrun[0], EXP_1, active_tasks); | |
1185 | CALC_LOAD(avenrun[1], EXP_5, active_tasks); | |
1186 | CALC_LOAD(avenrun[2], EXP_15, active_tasks); | |
1187 | count += LOAD_FREQ; | |
1188 | } while (count < 0); | |
1da177e4 LT |
1189 | } |
1190 | } | |
1191 | ||
1da177e4 LT |
1192 | /* |
1193 | * This function runs timers and the timer-tq in bottom half context. | |
1194 | */ | |
1195 | static void run_timer_softirq(struct softirq_action *h) | |
1196 | { | |
a6fa8e5a | 1197 | struct tvec_base *base = __get_cpu_var(tvec_bases); |
1da177e4 | 1198 | |
d3d74453 | 1199 | hrtimer_run_pending(); |
82f67cd9 | 1200 | |
1da177e4 LT |
1201 | if (time_after_eq(jiffies, base->timer_jiffies)) |
1202 | __run_timers(base); | |
1203 | } | |
1204 | ||
1205 | /* | |
1206 | * Called by the local, per-CPU timer interrupt on SMP. | |
1207 | */ | |
1208 | void run_local_timers(void) | |
1209 | { | |
d3d74453 | 1210 | hrtimer_run_queues(); |
1da177e4 | 1211 | raise_softirq(TIMER_SOFTIRQ); |
6687a97d | 1212 | softlockup_tick(); |
1da177e4 LT |
1213 | } |
1214 | ||
1215 | /* | |
1216 | * Called by the timer interrupt. xtime_lock must already be taken | |
1217 | * by the timer IRQ! | |
1218 | */ | |
3171a030 | 1219 | static inline void update_times(unsigned long ticks) |
1da177e4 | 1220 | { |
ad596171 | 1221 | update_wall_time(); |
1da177e4 LT |
1222 | calc_load(ticks); |
1223 | } | |
6819457d | 1224 | |
1da177e4 LT |
1225 | /* |
1226 | * The 64-bit jiffies value is not atomic - you MUST NOT read it | |
1227 | * without sampling the sequence number in xtime_lock. | |
1228 | * jiffies is defined in the linker script... | |
1229 | */ | |
1230 | ||
3171a030 | 1231 | void do_timer(unsigned long ticks) |
1da177e4 | 1232 | { |
3171a030 AN |
1233 | jiffies_64 += ticks; |
1234 | update_times(ticks); | |
1da177e4 LT |
1235 | } |
1236 | ||
1237 | #ifdef __ARCH_WANT_SYS_ALARM | |
1238 | ||
1239 | /* | |
1240 | * For backwards compatibility? This can be done in libc so Alpha | |
1241 | * and all newer ports shouldn't need it. | |
1242 | */ | |
58fd3aa2 | 1243 | SYSCALL_DEFINE1(alarm, unsigned int, seconds) |
1da177e4 | 1244 | { |
c08b8a49 | 1245 | return alarm_setitimer(seconds); |
1da177e4 LT |
1246 | } |
1247 | ||
1248 | #endif | |
1249 | ||
1250 | #ifndef __alpha__ | |
1251 | ||
1252 | /* | |
1253 | * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this | |
1254 | * should be moved into arch/i386 instead? | |
1255 | */ | |
1256 | ||
1257 | /** | |
1258 | * sys_getpid - return the thread group id of the current process | |
1259 | * | |
1260 | * Note, despite the name, this returns the tgid not the pid. The tgid and | |
1261 | * the pid are identical unless CLONE_THREAD was specified on clone() in | |
1262 | * which case the tgid is the same in all threads of the same group. | |
1263 | * | |
1264 | * This is SMP safe as current->tgid does not change. | |
1265 | */ | |
58fd3aa2 | 1266 | SYSCALL_DEFINE0(getpid) |
1da177e4 | 1267 | { |
b488893a | 1268 | return task_tgid_vnr(current); |
1da177e4 LT |
1269 | } |
1270 | ||
1271 | /* | |
6997a6fa KK |
1272 | * Accessing ->real_parent is not SMP-safe, it could |
1273 | * change from under us. However, we can use a stale | |
1274 | * value of ->real_parent under rcu_read_lock(), see | |
1275 | * release_task()->call_rcu(delayed_put_task_struct). | |
1da177e4 | 1276 | */ |
dbf040d9 | 1277 | SYSCALL_DEFINE0(getppid) |
1da177e4 LT |
1278 | { |
1279 | int pid; | |
1da177e4 | 1280 | |
6997a6fa | 1281 | rcu_read_lock(); |
6c5f3e7b | 1282 | pid = task_tgid_vnr(current->real_parent); |
6997a6fa | 1283 | rcu_read_unlock(); |
1da177e4 | 1284 | |
1da177e4 LT |
1285 | return pid; |
1286 | } | |
1287 | ||
dbf040d9 | 1288 | SYSCALL_DEFINE0(getuid) |
1da177e4 LT |
1289 | { |
1290 | /* Only we change this so SMP safe */ | |
76aac0e9 | 1291 | return current_uid(); |
1da177e4 LT |
1292 | } |
1293 | ||
dbf040d9 | 1294 | SYSCALL_DEFINE0(geteuid) |
1da177e4 LT |
1295 | { |
1296 | /* Only we change this so SMP safe */ | |
76aac0e9 | 1297 | return current_euid(); |
1da177e4 LT |
1298 | } |
1299 | ||
dbf040d9 | 1300 | SYSCALL_DEFINE0(getgid) |
1da177e4 LT |
1301 | { |
1302 | /* Only we change this so SMP safe */ | |
76aac0e9 | 1303 | return current_gid(); |
1da177e4 LT |
1304 | } |
1305 | ||
dbf040d9 | 1306 | SYSCALL_DEFINE0(getegid) |
1da177e4 LT |
1307 | { |
1308 | /* Only we change this so SMP safe */ | |
76aac0e9 | 1309 | return current_egid(); |
1da177e4 LT |
1310 | } |
1311 | ||
1312 | #endif | |
1313 | ||
1314 | static void process_timeout(unsigned long __data) | |
1315 | { | |
36c8b586 | 1316 | wake_up_process((struct task_struct *)__data); |
1da177e4 LT |
1317 | } |
1318 | ||
1319 | /** | |
1320 | * schedule_timeout - sleep until timeout | |
1321 | * @timeout: timeout value in jiffies | |
1322 | * | |
1323 | * Make the current task sleep until @timeout jiffies have | |
1324 | * elapsed. The routine will return immediately unless | |
1325 | * the current task state has been set (see set_current_state()). | |
1326 | * | |
1327 | * You can set the task state as follows - | |
1328 | * | |
1329 | * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to | |
1330 | * pass before the routine returns. The routine will return 0 | |
1331 | * | |
1332 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
1333 | * delivered to the current task. In this case the remaining time | |
1334 | * in jiffies will be returned, or 0 if the timer expired in time | |
1335 | * | |
1336 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1337 | * routine returns. | |
1338 | * | |
1339 | * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule | |
1340 | * the CPU away without a bound on the timeout. In this case the return | |
1341 | * value will be %MAX_SCHEDULE_TIMEOUT. | |
1342 | * | |
1343 | * In all cases the return value is guaranteed to be non-negative. | |
1344 | */ | |
7ad5b3a5 | 1345 | signed long __sched schedule_timeout(signed long timeout) |
1da177e4 LT |
1346 | { |
1347 | struct timer_list timer; | |
1348 | unsigned long expire; | |
1349 | ||
1350 | switch (timeout) | |
1351 | { | |
1352 | case MAX_SCHEDULE_TIMEOUT: | |
1353 | /* | |
1354 | * These two special cases are useful to be comfortable | |
1355 | * in the caller. Nothing more. We could take | |
1356 | * MAX_SCHEDULE_TIMEOUT from one of the negative value | |
1357 | * but I' d like to return a valid offset (>=0) to allow | |
1358 | * the caller to do everything it want with the retval. | |
1359 | */ | |
1360 | schedule(); | |
1361 | goto out; | |
1362 | default: | |
1363 | /* | |
1364 | * Another bit of PARANOID. Note that the retval will be | |
1365 | * 0 since no piece of kernel is supposed to do a check | |
1366 | * for a negative retval of schedule_timeout() (since it | |
1367 | * should never happens anyway). You just have the printk() | |
1368 | * that will tell you if something is gone wrong and where. | |
1369 | */ | |
5b149bcc | 1370 | if (timeout < 0) { |
1da177e4 | 1371 | printk(KERN_ERR "schedule_timeout: wrong timeout " |
5b149bcc AM |
1372 | "value %lx\n", timeout); |
1373 | dump_stack(); | |
1da177e4 LT |
1374 | current->state = TASK_RUNNING; |
1375 | goto out; | |
1376 | } | |
1377 | } | |
1378 | ||
1379 | expire = timeout + jiffies; | |
1380 | ||
c6f3a97f | 1381 | setup_timer_on_stack(&timer, process_timeout, (unsigned long)current); |
597d0275 | 1382 | __mod_timer(&timer, expire, false, TIMER_NOT_PINNED); |
1da177e4 LT |
1383 | schedule(); |
1384 | del_singleshot_timer_sync(&timer); | |
1385 | ||
c6f3a97f TG |
1386 | /* Remove the timer from the object tracker */ |
1387 | destroy_timer_on_stack(&timer); | |
1388 | ||
1da177e4 LT |
1389 | timeout = expire - jiffies; |
1390 | ||
1391 | out: | |
1392 | return timeout < 0 ? 0 : timeout; | |
1393 | } | |
1da177e4 LT |
1394 | EXPORT_SYMBOL(schedule_timeout); |
1395 | ||
8a1c1757 AM |
1396 | /* |
1397 | * We can use __set_current_state() here because schedule_timeout() calls | |
1398 | * schedule() unconditionally. | |
1399 | */ | |
64ed93a2 NA |
1400 | signed long __sched schedule_timeout_interruptible(signed long timeout) |
1401 | { | |
a5a0d52c AM |
1402 | __set_current_state(TASK_INTERRUPTIBLE); |
1403 | return schedule_timeout(timeout); | |
64ed93a2 NA |
1404 | } |
1405 | EXPORT_SYMBOL(schedule_timeout_interruptible); | |
1406 | ||
294d5cc2 MW |
1407 | signed long __sched schedule_timeout_killable(signed long timeout) |
1408 | { | |
1409 | __set_current_state(TASK_KILLABLE); | |
1410 | return schedule_timeout(timeout); | |
1411 | } | |
1412 | EXPORT_SYMBOL(schedule_timeout_killable); | |
1413 | ||
64ed93a2 NA |
1414 | signed long __sched schedule_timeout_uninterruptible(signed long timeout) |
1415 | { | |
a5a0d52c AM |
1416 | __set_current_state(TASK_UNINTERRUPTIBLE); |
1417 | return schedule_timeout(timeout); | |
64ed93a2 NA |
1418 | } |
1419 | EXPORT_SYMBOL(schedule_timeout_uninterruptible); | |
1420 | ||
1da177e4 | 1421 | /* Thread ID - the internal kernel "pid" */ |
58fd3aa2 | 1422 | SYSCALL_DEFINE0(gettid) |
1da177e4 | 1423 | { |
b488893a | 1424 | return task_pid_vnr(current); |
1da177e4 LT |
1425 | } |
1426 | ||
2aae4a10 | 1427 | /** |
d4d23add | 1428 | * do_sysinfo - fill in sysinfo struct |
2aae4a10 | 1429 | * @info: pointer to buffer to fill |
6819457d | 1430 | */ |
d4d23add | 1431 | int do_sysinfo(struct sysinfo *info) |
1da177e4 | 1432 | { |
1da177e4 LT |
1433 | unsigned long mem_total, sav_total; |
1434 | unsigned int mem_unit, bitcount; | |
1435 | unsigned long seq; | |
1436 | ||
d4d23add | 1437 | memset(info, 0, sizeof(struct sysinfo)); |
1da177e4 LT |
1438 | |
1439 | do { | |
1440 | struct timespec tp; | |
1441 | seq = read_seqbegin(&xtime_lock); | |
1442 | ||
1443 | /* | |
1444 | * This is annoying. The below is the same thing | |
1445 | * posix_get_clock_monotonic() does, but it wants to | |
1446 | * take the lock which we want to cover the loads stuff | |
1447 | * too. | |
1448 | */ | |
1449 | ||
1450 | getnstimeofday(&tp); | |
1451 | tp.tv_sec += wall_to_monotonic.tv_sec; | |
1452 | tp.tv_nsec += wall_to_monotonic.tv_nsec; | |
d6214141 | 1453 | monotonic_to_bootbased(&tp); |
1da177e4 LT |
1454 | if (tp.tv_nsec - NSEC_PER_SEC >= 0) { |
1455 | tp.tv_nsec = tp.tv_nsec - NSEC_PER_SEC; | |
1456 | tp.tv_sec++; | |
1457 | } | |
d4d23add | 1458 | info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0); |
1da177e4 | 1459 | |
d4d23add KM |
1460 | info->loads[0] = avenrun[0] << (SI_LOAD_SHIFT - FSHIFT); |
1461 | info->loads[1] = avenrun[1] << (SI_LOAD_SHIFT - FSHIFT); | |
1462 | info->loads[2] = avenrun[2] << (SI_LOAD_SHIFT - FSHIFT); | |
1da177e4 | 1463 | |
d4d23add | 1464 | info->procs = nr_threads; |
1da177e4 LT |
1465 | } while (read_seqretry(&xtime_lock, seq)); |
1466 | ||
d4d23add KM |
1467 | si_meminfo(info); |
1468 | si_swapinfo(info); | |
1da177e4 LT |
1469 | |
1470 | /* | |
1471 | * If the sum of all the available memory (i.e. ram + swap) | |
1472 | * is less than can be stored in a 32 bit unsigned long then | |
1473 | * we can be binary compatible with 2.2.x kernels. If not, | |
1474 | * well, in that case 2.2.x was broken anyways... | |
1475 | * | |
1476 | * -Erik Andersen <andersee@debian.org> | |
1477 | */ | |
1478 | ||
d4d23add KM |
1479 | mem_total = info->totalram + info->totalswap; |
1480 | if (mem_total < info->totalram || mem_total < info->totalswap) | |
1da177e4 LT |
1481 | goto out; |
1482 | bitcount = 0; | |
d4d23add | 1483 | mem_unit = info->mem_unit; |
1da177e4 LT |
1484 | while (mem_unit > 1) { |
1485 | bitcount++; | |
1486 | mem_unit >>= 1; | |
1487 | sav_total = mem_total; | |
1488 | mem_total <<= 1; | |
1489 | if (mem_total < sav_total) | |
1490 | goto out; | |
1491 | } | |
1492 | ||
1493 | /* | |
1494 | * If mem_total did not overflow, multiply all memory values by | |
d4d23add | 1495 | * info->mem_unit and set it to 1. This leaves things compatible |
1da177e4 LT |
1496 | * with 2.2.x, and also retains compatibility with earlier 2.4.x |
1497 | * kernels... | |
1498 | */ | |
1499 | ||
d4d23add KM |
1500 | info->mem_unit = 1; |
1501 | info->totalram <<= bitcount; | |
1502 | info->freeram <<= bitcount; | |
1503 | info->sharedram <<= bitcount; | |
1504 | info->bufferram <<= bitcount; | |
1505 | info->totalswap <<= bitcount; | |
1506 | info->freeswap <<= bitcount; | |
1507 | info->totalhigh <<= bitcount; | |
1508 | info->freehigh <<= bitcount; | |
1509 | ||
1510 | out: | |
1511 | return 0; | |
1512 | } | |
1513 | ||
1e7bfb21 | 1514 | SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info) |
d4d23add KM |
1515 | { |
1516 | struct sysinfo val; | |
1517 | ||
1518 | do_sysinfo(&val); | |
1da177e4 | 1519 | |
1da177e4 LT |
1520 | if (copy_to_user(info, &val, sizeof(struct sysinfo))) |
1521 | return -EFAULT; | |
1522 | ||
1523 | return 0; | |
1524 | } | |
1525 | ||
b4be6258 | 1526 | static int __cpuinit init_timers_cpu(int cpu) |
1da177e4 LT |
1527 | { |
1528 | int j; | |
a6fa8e5a | 1529 | struct tvec_base *base; |
b4be6258 | 1530 | static char __cpuinitdata tvec_base_done[NR_CPUS]; |
55c888d6 | 1531 | |
ba6edfcd | 1532 | if (!tvec_base_done[cpu]) { |
a4a6198b JB |
1533 | static char boot_done; |
1534 | ||
a4a6198b | 1535 | if (boot_done) { |
ba6edfcd AM |
1536 | /* |
1537 | * The APs use this path later in boot | |
1538 | */ | |
94f6030c CL |
1539 | base = kmalloc_node(sizeof(*base), |
1540 | GFP_KERNEL | __GFP_ZERO, | |
a4a6198b JB |
1541 | cpu_to_node(cpu)); |
1542 | if (!base) | |
1543 | return -ENOMEM; | |
6e453a67 VP |
1544 | |
1545 | /* Make sure that tvec_base is 2 byte aligned */ | |
1546 | if (tbase_get_deferrable(base)) { | |
1547 | WARN_ON(1); | |
1548 | kfree(base); | |
1549 | return -ENOMEM; | |
1550 | } | |
ba6edfcd | 1551 | per_cpu(tvec_bases, cpu) = base; |
a4a6198b | 1552 | } else { |
ba6edfcd AM |
1553 | /* |
1554 | * This is for the boot CPU - we use compile-time | |
1555 | * static initialisation because per-cpu memory isn't | |
1556 | * ready yet and because the memory allocators are not | |
1557 | * initialised either. | |
1558 | */ | |
a4a6198b | 1559 | boot_done = 1; |
ba6edfcd | 1560 | base = &boot_tvec_bases; |
a4a6198b | 1561 | } |
ba6edfcd AM |
1562 | tvec_base_done[cpu] = 1; |
1563 | } else { | |
1564 | base = per_cpu(tvec_bases, cpu); | |
a4a6198b | 1565 | } |
ba6edfcd | 1566 | |
3691c519 | 1567 | spin_lock_init(&base->lock); |
d730e882 | 1568 | |
1da177e4 LT |
1569 | for (j = 0; j < TVN_SIZE; j++) { |
1570 | INIT_LIST_HEAD(base->tv5.vec + j); | |
1571 | INIT_LIST_HEAD(base->tv4.vec + j); | |
1572 | INIT_LIST_HEAD(base->tv3.vec + j); | |
1573 | INIT_LIST_HEAD(base->tv2.vec + j); | |
1574 | } | |
1575 | for (j = 0; j < TVR_SIZE; j++) | |
1576 | INIT_LIST_HEAD(base->tv1.vec + j); | |
1577 | ||
1578 | base->timer_jiffies = jiffies; | |
a4a6198b | 1579 | return 0; |
1da177e4 LT |
1580 | } |
1581 | ||
1582 | #ifdef CONFIG_HOTPLUG_CPU | |
a6fa8e5a | 1583 | static void migrate_timer_list(struct tvec_base *new_base, struct list_head *head) |
1da177e4 LT |
1584 | { |
1585 | struct timer_list *timer; | |
1586 | ||
1587 | while (!list_empty(head)) { | |
b5e61818 | 1588 | timer = list_first_entry(head, struct timer_list, entry); |
55c888d6 | 1589 | detach_timer(timer, 0); |
6e453a67 | 1590 | timer_set_base(timer, new_base); |
1da177e4 | 1591 | internal_add_timer(new_base, timer); |
1da177e4 | 1592 | } |
1da177e4 LT |
1593 | } |
1594 | ||
48ccf3da | 1595 | static void __cpuinit migrate_timers(int cpu) |
1da177e4 | 1596 | { |
a6fa8e5a PM |
1597 | struct tvec_base *old_base; |
1598 | struct tvec_base *new_base; | |
1da177e4 LT |
1599 | int i; |
1600 | ||
1601 | BUG_ON(cpu_online(cpu)); | |
a4a6198b JB |
1602 | old_base = per_cpu(tvec_bases, cpu); |
1603 | new_base = get_cpu_var(tvec_bases); | |
d82f0b0f ON |
1604 | /* |
1605 | * The caller is globally serialized and nobody else | |
1606 | * takes two locks at once, deadlock is not possible. | |
1607 | */ | |
1608 | spin_lock_irq(&new_base->lock); | |
0d180406 | 1609 | spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING); |
3691c519 ON |
1610 | |
1611 | BUG_ON(old_base->running_timer); | |
1da177e4 | 1612 | |
1da177e4 | 1613 | for (i = 0; i < TVR_SIZE; i++) |
55c888d6 ON |
1614 | migrate_timer_list(new_base, old_base->tv1.vec + i); |
1615 | for (i = 0; i < TVN_SIZE; i++) { | |
1616 | migrate_timer_list(new_base, old_base->tv2.vec + i); | |
1617 | migrate_timer_list(new_base, old_base->tv3.vec + i); | |
1618 | migrate_timer_list(new_base, old_base->tv4.vec + i); | |
1619 | migrate_timer_list(new_base, old_base->tv5.vec + i); | |
1620 | } | |
1621 | ||
0d180406 | 1622 | spin_unlock(&old_base->lock); |
d82f0b0f | 1623 | spin_unlock_irq(&new_base->lock); |
1da177e4 | 1624 | put_cpu_var(tvec_bases); |
1da177e4 LT |
1625 | } |
1626 | #endif /* CONFIG_HOTPLUG_CPU */ | |
1627 | ||
8c78f307 | 1628 | static int __cpuinit timer_cpu_notify(struct notifier_block *self, |
1da177e4 LT |
1629 | unsigned long action, void *hcpu) |
1630 | { | |
1631 | long cpu = (long)hcpu; | |
1632 | switch(action) { | |
1633 | case CPU_UP_PREPARE: | |
8bb78442 | 1634 | case CPU_UP_PREPARE_FROZEN: |
a4a6198b JB |
1635 | if (init_timers_cpu(cpu) < 0) |
1636 | return NOTIFY_BAD; | |
1da177e4 LT |
1637 | break; |
1638 | #ifdef CONFIG_HOTPLUG_CPU | |
1639 | case CPU_DEAD: | |
8bb78442 | 1640 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
1641 | migrate_timers(cpu); |
1642 | break; | |
1643 | #endif | |
1644 | default: | |
1645 | break; | |
1646 | } | |
1647 | return NOTIFY_OK; | |
1648 | } | |
1649 | ||
8c78f307 | 1650 | static struct notifier_block __cpuinitdata timers_nb = { |
1da177e4 LT |
1651 | .notifier_call = timer_cpu_notify, |
1652 | }; | |
1653 | ||
1654 | ||
1655 | void __init init_timers(void) | |
1656 | { | |
07dccf33 | 1657 | int err = timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE, |
1da177e4 | 1658 | (void *)(long)smp_processor_id()); |
07dccf33 | 1659 | |
82f67cd9 IM |
1660 | init_timer_stats(); |
1661 | ||
07dccf33 | 1662 | BUG_ON(err == NOTIFY_BAD); |
1da177e4 | 1663 | register_cpu_notifier(&timers_nb); |
962cf36c | 1664 | open_softirq(TIMER_SOFTIRQ, run_timer_softirq); |
1da177e4 LT |
1665 | } |
1666 | ||
1da177e4 LT |
1667 | /** |
1668 | * msleep - sleep safely even with waitqueue interruptions | |
1669 | * @msecs: Time in milliseconds to sleep for | |
1670 | */ | |
1671 | void msleep(unsigned int msecs) | |
1672 | { | |
1673 | unsigned long timeout = msecs_to_jiffies(msecs) + 1; | |
1674 | ||
75bcc8c5 NA |
1675 | while (timeout) |
1676 | timeout = schedule_timeout_uninterruptible(timeout); | |
1da177e4 LT |
1677 | } |
1678 | ||
1679 | EXPORT_SYMBOL(msleep); | |
1680 | ||
1681 | /** | |
96ec3efd | 1682 | * msleep_interruptible - sleep waiting for signals |
1da177e4 LT |
1683 | * @msecs: Time in milliseconds to sleep for |
1684 | */ | |
1685 | unsigned long msleep_interruptible(unsigned int msecs) | |
1686 | { | |
1687 | unsigned long timeout = msecs_to_jiffies(msecs) + 1; | |
1688 | ||
75bcc8c5 NA |
1689 | while (timeout && !signal_pending(current)) |
1690 | timeout = schedule_timeout_interruptible(timeout); | |
1da177e4 LT |
1691 | return jiffies_to_msecs(timeout); |
1692 | } | |
1693 | ||
1694 | EXPORT_SYMBOL(msleep_interruptible); |