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> |
cdd6c482 | 40 | #include <linux/perf_event.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; |
97fd9ed4 | 75 | unsigned long next_timer; |
a6fa8e5a PM |
76 | struct tvec_root tv1; |
77 | struct tvec tv2; | |
78 | struct tvec tv3; | |
79 | struct tvec tv4; | |
80 | struct tvec tv5; | |
6e453a67 | 81 | } ____cacheline_aligned; |
1da177e4 | 82 | |
a6fa8e5a | 83 | struct tvec_base boot_tvec_bases; |
3691c519 | 84 | EXPORT_SYMBOL(boot_tvec_bases); |
a6fa8e5a | 85 | static DEFINE_PER_CPU(struct tvec_base *, tvec_bases) = &boot_tvec_bases; |
1da177e4 | 86 | |
6e453a67 | 87 | /* |
a6fa8e5a | 88 | * Note that all tvec_bases are 2 byte aligned and lower bit of |
6e453a67 VP |
89 | * base in timer_list is guaranteed to be zero. Use the LSB for |
90 | * the new flag to indicate whether the timer is deferrable | |
91 | */ | |
92 | #define TBASE_DEFERRABLE_FLAG (0x1) | |
93 | ||
94 | /* Functions below help us manage 'deferrable' flag */ | |
a6fa8e5a | 95 | static inline unsigned int tbase_get_deferrable(struct tvec_base *base) |
6e453a67 | 96 | { |
e9910846 | 97 | return ((unsigned int)(unsigned long)base & TBASE_DEFERRABLE_FLAG); |
6e453a67 VP |
98 | } |
99 | ||
a6fa8e5a | 100 | static inline struct tvec_base *tbase_get_base(struct tvec_base *base) |
6e453a67 | 101 | { |
a6fa8e5a | 102 | return ((struct tvec_base *)((unsigned long)base & ~TBASE_DEFERRABLE_FLAG)); |
6e453a67 VP |
103 | } |
104 | ||
105 | static inline void timer_set_deferrable(struct timer_list *timer) | |
106 | { | |
a6fa8e5a | 107 | timer->base = ((struct tvec_base *)((unsigned long)(timer->base) | |
6819457d | 108 | TBASE_DEFERRABLE_FLAG)); |
6e453a67 VP |
109 | } |
110 | ||
111 | static inline void | |
a6fa8e5a | 112 | timer_set_base(struct timer_list *timer, struct tvec_base *new_base) |
6e453a67 | 113 | { |
a6fa8e5a | 114 | timer->base = (struct tvec_base *)((unsigned long)(new_base) | |
6819457d | 115 | tbase_get_deferrable(timer->base)); |
6e453a67 VP |
116 | } |
117 | ||
9c133c46 AS |
118 | static unsigned long round_jiffies_common(unsigned long j, int cpu, |
119 | bool force_up) | |
4c36a5de AV |
120 | { |
121 | int rem; | |
122 | unsigned long original = j; | |
123 | ||
124 | /* | |
125 | * We don't want all cpus firing their timers at once hitting the | |
126 | * same lock or cachelines, so we skew each extra cpu with an extra | |
127 | * 3 jiffies. This 3 jiffies came originally from the mm/ code which | |
128 | * already did this. | |
129 | * The skew is done by adding 3*cpunr, then round, then subtract this | |
130 | * extra offset again. | |
131 | */ | |
132 | j += cpu * 3; | |
133 | ||
134 | rem = j % HZ; | |
135 | ||
136 | /* | |
137 | * If the target jiffie is just after a whole second (which can happen | |
138 | * due to delays of the timer irq, long irq off times etc etc) then | |
139 | * we should round down to the whole second, not up. Use 1/4th second | |
140 | * as cutoff for this rounding as an extreme upper bound for this. | |
9c133c46 | 141 | * But never round down if @force_up is set. |
4c36a5de | 142 | */ |
9c133c46 | 143 | if (rem < HZ/4 && !force_up) /* round down */ |
4c36a5de AV |
144 | j = j - rem; |
145 | else /* round up */ | |
146 | j = j - rem + HZ; | |
147 | ||
148 | /* now that we have rounded, subtract the extra skew again */ | |
149 | j -= cpu * 3; | |
150 | ||
151 | if (j <= jiffies) /* rounding ate our timeout entirely; */ | |
152 | return original; | |
153 | return j; | |
154 | } | |
9c133c46 AS |
155 | |
156 | /** | |
157 | * __round_jiffies - function to round jiffies to a full second | |
158 | * @j: the time in (absolute) jiffies that should be rounded | |
159 | * @cpu: the processor number on which the timeout will happen | |
160 | * | |
161 | * __round_jiffies() rounds an absolute time in the future (in jiffies) | |
162 | * up or down to (approximately) full seconds. This is useful for timers | |
163 | * for which the exact time they fire does not matter too much, as long as | |
164 | * they fire approximately every X seconds. | |
165 | * | |
166 | * By rounding these timers to whole seconds, all such timers will fire | |
167 | * at the same time, rather than at various times spread out. The goal | |
168 | * of this is to have the CPU wake up less, which saves power. | |
169 | * | |
170 | * The exact rounding is skewed for each processor to avoid all | |
171 | * processors firing at the exact same time, which could lead | |
172 | * to lock contention or spurious cache line bouncing. | |
173 | * | |
174 | * The return value is the rounded version of the @j parameter. | |
175 | */ | |
176 | unsigned long __round_jiffies(unsigned long j, int cpu) | |
177 | { | |
178 | return round_jiffies_common(j, cpu, false); | |
179 | } | |
4c36a5de AV |
180 | EXPORT_SYMBOL_GPL(__round_jiffies); |
181 | ||
182 | /** | |
183 | * __round_jiffies_relative - function to round jiffies to a full second | |
184 | * @j: the time in (relative) jiffies that should be rounded | |
185 | * @cpu: the processor number on which the timeout will happen | |
186 | * | |
72fd4a35 | 187 | * __round_jiffies_relative() rounds a time delta in the future (in jiffies) |
4c36a5de AV |
188 | * up or down to (approximately) full seconds. This is useful for timers |
189 | * for which the exact time they fire does not matter too much, as long as | |
190 | * they fire approximately every X seconds. | |
191 | * | |
192 | * By rounding these timers to whole seconds, all such timers will fire | |
193 | * at the same time, rather than at various times spread out. The goal | |
194 | * of this is to have the CPU wake up less, which saves power. | |
195 | * | |
196 | * The exact rounding is skewed for each processor to avoid all | |
197 | * processors firing at the exact same time, which could lead | |
198 | * to lock contention or spurious cache line bouncing. | |
199 | * | |
72fd4a35 | 200 | * The return value is the rounded version of the @j parameter. |
4c36a5de AV |
201 | */ |
202 | unsigned long __round_jiffies_relative(unsigned long j, int cpu) | |
203 | { | |
9c133c46 AS |
204 | unsigned long j0 = jiffies; |
205 | ||
206 | /* Use j0 because jiffies might change while we run */ | |
207 | return round_jiffies_common(j + j0, cpu, false) - j0; | |
4c36a5de AV |
208 | } |
209 | EXPORT_SYMBOL_GPL(__round_jiffies_relative); | |
210 | ||
211 | /** | |
212 | * round_jiffies - function to round jiffies to a full second | |
213 | * @j: the time in (absolute) jiffies that should be rounded | |
214 | * | |
72fd4a35 | 215 | * round_jiffies() rounds an absolute time in the future (in jiffies) |
4c36a5de AV |
216 | * up or down to (approximately) full seconds. This is useful for timers |
217 | * for which the exact time they fire does not matter too much, as long as | |
218 | * they fire approximately every X seconds. | |
219 | * | |
220 | * By rounding these timers to whole seconds, all such timers will fire | |
221 | * at the same time, rather than at various times spread out. The goal | |
222 | * of this is to have the CPU wake up less, which saves power. | |
223 | * | |
72fd4a35 | 224 | * The return value is the rounded version of the @j parameter. |
4c36a5de AV |
225 | */ |
226 | unsigned long round_jiffies(unsigned long j) | |
227 | { | |
9c133c46 | 228 | return round_jiffies_common(j, raw_smp_processor_id(), false); |
4c36a5de AV |
229 | } |
230 | EXPORT_SYMBOL_GPL(round_jiffies); | |
231 | ||
232 | /** | |
233 | * round_jiffies_relative - function to round jiffies to a full second | |
234 | * @j: the time in (relative) jiffies that should be rounded | |
235 | * | |
72fd4a35 | 236 | * round_jiffies_relative() rounds a time delta in the future (in jiffies) |
4c36a5de AV |
237 | * up or down to (approximately) full seconds. This is useful for timers |
238 | * for which the exact time they fire does not matter too much, as long as | |
239 | * they fire approximately every X seconds. | |
240 | * | |
241 | * By rounding these timers to whole seconds, all such timers will fire | |
242 | * at the same time, rather than at various times spread out. The goal | |
243 | * of this is to have the CPU wake up less, which saves power. | |
244 | * | |
72fd4a35 | 245 | * The return value is the rounded version of the @j parameter. |
4c36a5de AV |
246 | */ |
247 | unsigned long round_jiffies_relative(unsigned long j) | |
248 | { | |
249 | return __round_jiffies_relative(j, raw_smp_processor_id()); | |
250 | } | |
251 | EXPORT_SYMBOL_GPL(round_jiffies_relative); | |
252 | ||
9c133c46 AS |
253 | /** |
254 | * __round_jiffies_up - function to round jiffies up to a full second | |
255 | * @j: the time in (absolute) jiffies that should be rounded | |
256 | * @cpu: the processor number on which the timeout will happen | |
257 | * | |
258 | * This is the same as __round_jiffies() except that it will never | |
259 | * round down. This is useful for timeouts for which the exact time | |
260 | * of firing does not matter too much, as long as they don't fire too | |
261 | * early. | |
262 | */ | |
263 | unsigned long __round_jiffies_up(unsigned long j, int cpu) | |
264 | { | |
265 | return round_jiffies_common(j, cpu, true); | |
266 | } | |
267 | EXPORT_SYMBOL_GPL(__round_jiffies_up); | |
268 | ||
269 | /** | |
270 | * __round_jiffies_up_relative - function to round jiffies up to a full second | |
271 | * @j: the time in (relative) jiffies that should be rounded | |
272 | * @cpu: the processor number on which the timeout will happen | |
273 | * | |
274 | * This is the same as __round_jiffies_relative() except that it will never | |
275 | * round down. This is useful for timeouts for which the exact time | |
276 | * of firing does not matter too much, as long as they don't fire too | |
277 | * early. | |
278 | */ | |
279 | unsigned long __round_jiffies_up_relative(unsigned long j, int cpu) | |
280 | { | |
281 | unsigned long j0 = jiffies; | |
282 | ||
283 | /* Use j0 because jiffies might change while we run */ | |
284 | return round_jiffies_common(j + j0, cpu, true) - j0; | |
285 | } | |
286 | EXPORT_SYMBOL_GPL(__round_jiffies_up_relative); | |
287 | ||
288 | /** | |
289 | * round_jiffies_up - function to round jiffies up to a full second | |
290 | * @j: the time in (absolute) jiffies that should be rounded | |
291 | * | |
292 | * This is the same as round_jiffies() except that it will never | |
293 | * round down. This is useful for timeouts for which the exact time | |
294 | * of firing does not matter too much, as long as they don't fire too | |
295 | * early. | |
296 | */ | |
297 | unsigned long round_jiffies_up(unsigned long j) | |
298 | { | |
299 | return round_jiffies_common(j, raw_smp_processor_id(), true); | |
300 | } | |
301 | EXPORT_SYMBOL_GPL(round_jiffies_up); | |
302 | ||
303 | /** | |
304 | * round_jiffies_up_relative - function to round jiffies up to a full second | |
305 | * @j: the time in (relative) jiffies that should be rounded | |
306 | * | |
307 | * This is the same as round_jiffies_relative() except that it will never | |
308 | * round down. This is useful for timeouts for which the exact time | |
309 | * of firing does not matter too much, as long as they don't fire too | |
310 | * early. | |
311 | */ | |
312 | unsigned long round_jiffies_up_relative(unsigned long j) | |
313 | { | |
314 | return __round_jiffies_up_relative(j, raw_smp_processor_id()); | |
315 | } | |
316 | EXPORT_SYMBOL_GPL(round_jiffies_up_relative); | |
317 | ||
4c36a5de | 318 | |
a6fa8e5a | 319 | static inline void set_running_timer(struct tvec_base *base, |
1da177e4 LT |
320 | struct timer_list *timer) |
321 | { | |
322 | #ifdef CONFIG_SMP | |
3691c519 | 323 | base->running_timer = timer; |
1da177e4 LT |
324 | #endif |
325 | } | |
326 | ||
a6fa8e5a | 327 | static void internal_add_timer(struct tvec_base *base, struct timer_list *timer) |
1da177e4 LT |
328 | { |
329 | unsigned long expires = timer->expires; | |
330 | unsigned long idx = expires - base->timer_jiffies; | |
331 | struct list_head *vec; | |
332 | ||
333 | if (idx < TVR_SIZE) { | |
334 | int i = expires & TVR_MASK; | |
335 | vec = base->tv1.vec + i; | |
336 | } else if (idx < 1 << (TVR_BITS + TVN_BITS)) { | |
337 | int i = (expires >> TVR_BITS) & TVN_MASK; | |
338 | vec = base->tv2.vec + i; | |
339 | } else if (idx < 1 << (TVR_BITS + 2 * TVN_BITS)) { | |
340 | int i = (expires >> (TVR_BITS + TVN_BITS)) & TVN_MASK; | |
341 | vec = base->tv3.vec + i; | |
342 | } else if (idx < 1 << (TVR_BITS + 3 * TVN_BITS)) { | |
343 | int i = (expires >> (TVR_BITS + 2 * TVN_BITS)) & TVN_MASK; | |
344 | vec = base->tv4.vec + i; | |
345 | } else if ((signed long) idx < 0) { | |
346 | /* | |
347 | * Can happen if you add a timer with expires == jiffies, | |
348 | * or you set a timer to go off in the past | |
349 | */ | |
350 | vec = base->tv1.vec + (base->timer_jiffies & TVR_MASK); | |
351 | } else { | |
352 | int i; | |
353 | /* If the timeout is larger than 0xffffffff on 64-bit | |
354 | * architectures then we use the maximum timeout: | |
355 | */ | |
356 | if (idx > 0xffffffffUL) { | |
357 | idx = 0xffffffffUL; | |
358 | expires = idx + base->timer_jiffies; | |
359 | } | |
360 | i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK; | |
361 | vec = base->tv5.vec + i; | |
362 | } | |
363 | /* | |
364 | * Timers are FIFO: | |
365 | */ | |
366 | list_add_tail(&timer->entry, vec); | |
367 | } | |
368 | ||
82f67cd9 IM |
369 | #ifdef CONFIG_TIMER_STATS |
370 | void __timer_stats_timer_set_start_info(struct timer_list *timer, void *addr) | |
371 | { | |
372 | if (timer->start_site) | |
373 | return; | |
374 | ||
375 | timer->start_site = addr; | |
376 | memcpy(timer->start_comm, current->comm, TASK_COMM_LEN); | |
377 | timer->start_pid = current->pid; | |
378 | } | |
c5c061b8 VP |
379 | |
380 | static void timer_stats_account_timer(struct timer_list *timer) | |
381 | { | |
382 | unsigned int flag = 0; | |
383 | ||
507e1231 HC |
384 | if (likely(!timer->start_site)) |
385 | return; | |
c5c061b8 VP |
386 | if (unlikely(tbase_get_deferrable(timer->base))) |
387 | flag |= TIMER_STATS_FLAG_DEFERRABLE; | |
388 | ||
389 | timer_stats_update_stats(timer, timer->start_pid, timer->start_site, | |
390 | timer->function, timer->start_comm, flag); | |
391 | } | |
392 | ||
393 | #else | |
394 | static void timer_stats_account_timer(struct timer_list *timer) {} | |
82f67cd9 IM |
395 | #endif |
396 | ||
c6f3a97f TG |
397 | #ifdef CONFIG_DEBUG_OBJECTS_TIMERS |
398 | ||
399 | static struct debug_obj_descr timer_debug_descr; | |
400 | ||
401 | /* | |
402 | * fixup_init is called when: | |
403 | * - an active object is initialized | |
55c888d6 | 404 | */ |
c6f3a97f TG |
405 | static int timer_fixup_init(void *addr, enum debug_obj_state state) |
406 | { | |
407 | struct timer_list *timer = addr; | |
408 | ||
409 | switch (state) { | |
410 | case ODEBUG_STATE_ACTIVE: | |
411 | del_timer_sync(timer); | |
412 | debug_object_init(timer, &timer_debug_descr); | |
413 | return 1; | |
414 | default: | |
415 | return 0; | |
416 | } | |
417 | } | |
418 | ||
419 | /* | |
420 | * fixup_activate is called when: | |
421 | * - an active object is activated | |
422 | * - an unknown object is activated (might be a statically initialized object) | |
423 | */ | |
424 | static int timer_fixup_activate(void *addr, enum debug_obj_state state) | |
425 | { | |
426 | struct timer_list *timer = addr; | |
427 | ||
428 | switch (state) { | |
429 | ||
430 | case ODEBUG_STATE_NOTAVAILABLE: | |
431 | /* | |
432 | * This is not really a fixup. The timer was | |
433 | * statically initialized. We just make sure that it | |
434 | * is tracked in the object tracker. | |
435 | */ | |
436 | if (timer->entry.next == NULL && | |
437 | timer->entry.prev == TIMER_ENTRY_STATIC) { | |
438 | debug_object_init(timer, &timer_debug_descr); | |
439 | debug_object_activate(timer, &timer_debug_descr); | |
440 | return 0; | |
441 | } else { | |
442 | WARN_ON_ONCE(1); | |
443 | } | |
444 | return 0; | |
445 | ||
446 | case ODEBUG_STATE_ACTIVE: | |
447 | WARN_ON(1); | |
448 | ||
449 | default: | |
450 | return 0; | |
451 | } | |
452 | } | |
453 | ||
454 | /* | |
455 | * fixup_free is called when: | |
456 | * - an active object is freed | |
457 | */ | |
458 | static int timer_fixup_free(void *addr, enum debug_obj_state state) | |
459 | { | |
460 | struct timer_list *timer = addr; | |
461 | ||
462 | switch (state) { | |
463 | case ODEBUG_STATE_ACTIVE: | |
464 | del_timer_sync(timer); | |
465 | debug_object_free(timer, &timer_debug_descr); | |
466 | return 1; | |
467 | default: | |
468 | return 0; | |
469 | } | |
470 | } | |
471 | ||
472 | static struct debug_obj_descr timer_debug_descr = { | |
473 | .name = "timer_list", | |
474 | .fixup_init = timer_fixup_init, | |
475 | .fixup_activate = timer_fixup_activate, | |
476 | .fixup_free = timer_fixup_free, | |
477 | }; | |
478 | ||
479 | static inline void debug_timer_init(struct timer_list *timer) | |
480 | { | |
481 | debug_object_init(timer, &timer_debug_descr); | |
482 | } | |
483 | ||
484 | static inline void debug_timer_activate(struct timer_list *timer) | |
485 | { | |
486 | debug_object_activate(timer, &timer_debug_descr); | |
487 | } | |
488 | ||
489 | static inline void debug_timer_deactivate(struct timer_list *timer) | |
490 | { | |
491 | debug_object_deactivate(timer, &timer_debug_descr); | |
492 | } | |
493 | ||
494 | static inline void debug_timer_free(struct timer_list *timer) | |
495 | { | |
496 | debug_object_free(timer, &timer_debug_descr); | |
497 | } | |
498 | ||
6f2b9b9a JB |
499 | static void __init_timer(struct timer_list *timer, |
500 | const char *name, | |
501 | struct lock_class_key *key); | |
c6f3a97f | 502 | |
6f2b9b9a JB |
503 | void init_timer_on_stack_key(struct timer_list *timer, |
504 | const char *name, | |
505 | struct lock_class_key *key) | |
c6f3a97f TG |
506 | { |
507 | debug_object_init_on_stack(timer, &timer_debug_descr); | |
6f2b9b9a | 508 | __init_timer(timer, name, key); |
c6f3a97f | 509 | } |
6f2b9b9a | 510 | EXPORT_SYMBOL_GPL(init_timer_on_stack_key); |
c6f3a97f TG |
511 | |
512 | void destroy_timer_on_stack(struct timer_list *timer) | |
513 | { | |
514 | debug_object_free(timer, &timer_debug_descr); | |
515 | } | |
516 | EXPORT_SYMBOL_GPL(destroy_timer_on_stack); | |
517 | ||
518 | #else | |
519 | static inline void debug_timer_init(struct timer_list *timer) { } | |
520 | static inline void debug_timer_activate(struct timer_list *timer) { } | |
521 | static inline void debug_timer_deactivate(struct timer_list *timer) { } | |
522 | #endif | |
523 | ||
6f2b9b9a JB |
524 | static void __init_timer(struct timer_list *timer, |
525 | const char *name, | |
526 | struct lock_class_key *key) | |
55c888d6 ON |
527 | { |
528 | timer->entry.next = NULL; | |
bfe5d834 | 529 | timer->base = __raw_get_cpu_var(tvec_bases); |
82f67cd9 IM |
530 | #ifdef CONFIG_TIMER_STATS |
531 | timer->start_site = NULL; | |
532 | timer->start_pid = -1; | |
533 | memset(timer->start_comm, 0, TASK_COMM_LEN); | |
534 | #endif | |
6f2b9b9a | 535 | lockdep_init_map(&timer->lockdep_map, name, key, 0); |
55c888d6 | 536 | } |
c6f3a97f TG |
537 | |
538 | /** | |
633fe795 | 539 | * init_timer_key - initialize a timer |
c6f3a97f | 540 | * @timer: the timer to be initialized |
633fe795 RD |
541 | * @name: name of the timer |
542 | * @key: lockdep class key of the fake lock used for tracking timer | |
543 | * sync lock dependencies | |
c6f3a97f | 544 | * |
633fe795 | 545 | * init_timer_key() must be done to a timer prior calling *any* of the |
c6f3a97f TG |
546 | * other timer functions. |
547 | */ | |
6f2b9b9a JB |
548 | void init_timer_key(struct timer_list *timer, |
549 | const char *name, | |
550 | struct lock_class_key *key) | |
c6f3a97f TG |
551 | { |
552 | debug_timer_init(timer); | |
6f2b9b9a | 553 | __init_timer(timer, name, key); |
c6f3a97f | 554 | } |
6f2b9b9a | 555 | EXPORT_SYMBOL(init_timer_key); |
55c888d6 | 556 | |
6f2b9b9a JB |
557 | void init_timer_deferrable_key(struct timer_list *timer, |
558 | const char *name, | |
559 | struct lock_class_key *key) | |
6e453a67 | 560 | { |
6f2b9b9a | 561 | init_timer_key(timer, name, key); |
6e453a67 VP |
562 | timer_set_deferrable(timer); |
563 | } | |
6f2b9b9a | 564 | EXPORT_SYMBOL(init_timer_deferrable_key); |
6e453a67 | 565 | |
55c888d6 | 566 | static inline void detach_timer(struct timer_list *timer, |
82f67cd9 | 567 | int clear_pending) |
55c888d6 ON |
568 | { |
569 | struct list_head *entry = &timer->entry; | |
570 | ||
c6f3a97f TG |
571 | debug_timer_deactivate(timer); |
572 | ||
55c888d6 ON |
573 | __list_del(entry->prev, entry->next); |
574 | if (clear_pending) | |
575 | entry->next = NULL; | |
576 | entry->prev = LIST_POISON2; | |
577 | } | |
578 | ||
579 | /* | |
3691c519 | 580 | * We are using hashed locking: holding per_cpu(tvec_bases).lock |
55c888d6 ON |
581 | * means that all timers which are tied to this base via timer->base are |
582 | * locked, and the base itself is locked too. | |
583 | * | |
584 | * So __run_timers/migrate_timers can safely modify all timers which could | |
585 | * be found on ->tvX lists. | |
586 | * | |
587 | * When the timer's base is locked, and the timer removed from list, it is | |
588 | * possible to set timer->base = NULL and drop the lock: the timer remains | |
589 | * locked. | |
590 | */ | |
a6fa8e5a | 591 | static struct tvec_base *lock_timer_base(struct timer_list *timer, |
55c888d6 | 592 | unsigned long *flags) |
89e7e374 | 593 | __acquires(timer->base->lock) |
55c888d6 | 594 | { |
a6fa8e5a | 595 | struct tvec_base *base; |
55c888d6 ON |
596 | |
597 | for (;;) { | |
a6fa8e5a | 598 | struct tvec_base *prelock_base = timer->base; |
6e453a67 | 599 | base = tbase_get_base(prelock_base); |
55c888d6 ON |
600 | if (likely(base != NULL)) { |
601 | spin_lock_irqsave(&base->lock, *flags); | |
6e453a67 | 602 | if (likely(prelock_base == timer->base)) |
55c888d6 ON |
603 | return base; |
604 | /* The timer has migrated to another CPU */ | |
605 | spin_unlock_irqrestore(&base->lock, *flags); | |
606 | } | |
607 | cpu_relax(); | |
608 | } | |
609 | } | |
610 | ||
74019224 | 611 | static inline int |
597d0275 AB |
612 | __mod_timer(struct timer_list *timer, unsigned long expires, |
613 | bool pending_only, int pinned) | |
1da177e4 | 614 | { |
a6fa8e5a | 615 | struct tvec_base *base, *new_base; |
1da177e4 | 616 | unsigned long flags; |
eea08f32 | 617 | int ret = 0 , cpu; |
1da177e4 | 618 | |
82f67cd9 | 619 | timer_stats_timer_set_start_info(timer); |
1da177e4 | 620 | BUG_ON(!timer->function); |
1da177e4 | 621 | |
55c888d6 ON |
622 | base = lock_timer_base(timer, &flags); |
623 | ||
624 | if (timer_pending(timer)) { | |
625 | detach_timer(timer, 0); | |
97fd9ed4 MS |
626 | if (timer->expires == base->next_timer && |
627 | !tbase_get_deferrable(timer->base)) | |
628 | base->next_timer = base->timer_jiffies; | |
55c888d6 | 629 | ret = 1; |
74019224 IM |
630 | } else { |
631 | if (pending_only) | |
632 | goto out_unlock; | |
55c888d6 ON |
633 | } |
634 | ||
c6f3a97f TG |
635 | debug_timer_activate(timer); |
636 | ||
a4a6198b | 637 | new_base = __get_cpu_var(tvec_bases); |
1da177e4 | 638 | |
eea08f32 AB |
639 | cpu = smp_processor_id(); |
640 | ||
641 | #if defined(CONFIG_NO_HZ) && defined(CONFIG_SMP) | |
642 | if (!pinned && get_sysctl_timer_migration() && idle_cpu(cpu)) { | |
643 | int preferred_cpu = get_nohz_load_balancer(); | |
644 | ||
645 | if (preferred_cpu >= 0) | |
646 | cpu = preferred_cpu; | |
647 | } | |
648 | #endif | |
649 | new_base = per_cpu(tvec_bases, cpu); | |
650 | ||
3691c519 | 651 | if (base != new_base) { |
1da177e4 | 652 | /* |
55c888d6 ON |
653 | * We are trying to schedule the timer on the local CPU. |
654 | * However we can't change timer's base while it is running, | |
655 | * otherwise del_timer_sync() can't detect that the timer's | |
656 | * handler yet has not finished. This also guarantees that | |
657 | * the timer is serialized wrt itself. | |
1da177e4 | 658 | */ |
a2c348fe | 659 | if (likely(base->running_timer != timer)) { |
55c888d6 | 660 | /* See the comment in lock_timer_base() */ |
6e453a67 | 661 | timer_set_base(timer, NULL); |
55c888d6 | 662 | spin_unlock(&base->lock); |
a2c348fe ON |
663 | base = new_base; |
664 | spin_lock(&base->lock); | |
6e453a67 | 665 | timer_set_base(timer, base); |
1da177e4 LT |
666 | } |
667 | } | |
668 | ||
1da177e4 | 669 | timer->expires = expires; |
97fd9ed4 MS |
670 | if (time_before(timer->expires, base->next_timer) && |
671 | !tbase_get_deferrable(timer->base)) | |
672 | base->next_timer = timer->expires; | |
a2c348fe | 673 | internal_add_timer(base, timer); |
74019224 IM |
674 | |
675 | out_unlock: | |
a2c348fe | 676 | spin_unlock_irqrestore(&base->lock, flags); |
1da177e4 LT |
677 | |
678 | return ret; | |
679 | } | |
680 | ||
2aae4a10 | 681 | /** |
74019224 IM |
682 | * mod_timer_pending - modify a pending timer's timeout |
683 | * @timer: the pending timer to be modified | |
684 | * @expires: new timeout in jiffies | |
1da177e4 | 685 | * |
74019224 IM |
686 | * mod_timer_pending() is the same for pending timers as mod_timer(), |
687 | * but will not re-activate and modify already deleted timers. | |
688 | * | |
689 | * It is useful for unserialized use of timers. | |
1da177e4 | 690 | */ |
74019224 | 691 | int mod_timer_pending(struct timer_list *timer, unsigned long expires) |
1da177e4 | 692 | { |
597d0275 | 693 | return __mod_timer(timer, expires, true, TIMER_NOT_PINNED); |
1da177e4 | 694 | } |
74019224 | 695 | EXPORT_SYMBOL(mod_timer_pending); |
1da177e4 | 696 | |
2aae4a10 | 697 | /** |
1da177e4 LT |
698 | * mod_timer - modify a timer's timeout |
699 | * @timer: the timer to be modified | |
2aae4a10 | 700 | * @expires: new timeout in jiffies |
1da177e4 | 701 | * |
72fd4a35 | 702 | * mod_timer() is a more efficient way to update the expire field of an |
1da177e4 LT |
703 | * active timer (if the timer is inactive it will be activated) |
704 | * | |
705 | * mod_timer(timer, expires) is equivalent to: | |
706 | * | |
707 | * del_timer(timer); timer->expires = expires; add_timer(timer); | |
708 | * | |
709 | * Note that if there are multiple unserialized concurrent users of the | |
710 | * same timer, then mod_timer() is the only safe way to modify the timeout, | |
711 | * since add_timer() cannot modify an already running timer. | |
712 | * | |
713 | * The function returns whether it has modified a pending timer or not. | |
714 | * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an | |
715 | * active timer returns 1.) | |
716 | */ | |
717 | int mod_timer(struct timer_list *timer, unsigned long expires) | |
718 | { | |
1da177e4 LT |
719 | /* |
720 | * This is a common optimization triggered by the | |
721 | * networking code - if the timer is re-modified | |
722 | * to be the same thing then just return: | |
723 | */ | |
4841158b | 724 | if (timer_pending(timer) && timer->expires == expires) |
1da177e4 LT |
725 | return 1; |
726 | ||
597d0275 | 727 | return __mod_timer(timer, expires, false, TIMER_NOT_PINNED); |
1da177e4 | 728 | } |
1da177e4 LT |
729 | EXPORT_SYMBOL(mod_timer); |
730 | ||
597d0275 AB |
731 | /** |
732 | * mod_timer_pinned - modify a timer's timeout | |
733 | * @timer: the timer to be modified | |
734 | * @expires: new timeout in jiffies | |
735 | * | |
736 | * mod_timer_pinned() is a way to update the expire field of an | |
737 | * active timer (if the timer is inactive it will be activated) | |
738 | * and not allow the timer to be migrated to a different CPU. | |
739 | * | |
740 | * mod_timer_pinned(timer, expires) is equivalent to: | |
741 | * | |
742 | * del_timer(timer); timer->expires = expires; add_timer(timer); | |
743 | */ | |
744 | int mod_timer_pinned(struct timer_list *timer, unsigned long expires) | |
745 | { | |
746 | if (timer->expires == expires && timer_pending(timer)) | |
747 | return 1; | |
748 | ||
749 | return __mod_timer(timer, expires, false, TIMER_PINNED); | |
750 | } | |
751 | EXPORT_SYMBOL(mod_timer_pinned); | |
752 | ||
74019224 IM |
753 | /** |
754 | * add_timer - start a timer | |
755 | * @timer: the timer to be added | |
756 | * | |
757 | * The kernel will do a ->function(->data) callback from the | |
758 | * timer interrupt at the ->expires point in the future. The | |
759 | * current time is 'jiffies'. | |
760 | * | |
761 | * The timer's ->expires, ->function (and if the handler uses it, ->data) | |
762 | * fields must be set prior calling this function. | |
763 | * | |
764 | * Timers with an ->expires field in the past will be executed in the next | |
765 | * timer tick. | |
766 | */ | |
767 | void add_timer(struct timer_list *timer) | |
768 | { | |
769 | BUG_ON(timer_pending(timer)); | |
770 | mod_timer(timer, timer->expires); | |
771 | } | |
772 | EXPORT_SYMBOL(add_timer); | |
773 | ||
774 | /** | |
775 | * add_timer_on - start a timer on a particular CPU | |
776 | * @timer: the timer to be added | |
777 | * @cpu: the CPU to start it on | |
778 | * | |
779 | * This is not very scalable on SMP. Double adds are not possible. | |
780 | */ | |
781 | void add_timer_on(struct timer_list *timer, int cpu) | |
782 | { | |
783 | struct tvec_base *base = per_cpu(tvec_bases, cpu); | |
784 | unsigned long flags; | |
785 | ||
786 | timer_stats_timer_set_start_info(timer); | |
787 | BUG_ON(timer_pending(timer) || !timer->function); | |
788 | spin_lock_irqsave(&base->lock, flags); | |
789 | timer_set_base(timer, base); | |
790 | debug_timer_activate(timer); | |
97fd9ed4 MS |
791 | if (time_before(timer->expires, base->next_timer) && |
792 | !tbase_get_deferrable(timer->base)) | |
793 | base->next_timer = timer->expires; | |
74019224 IM |
794 | internal_add_timer(base, timer); |
795 | /* | |
796 | * Check whether the other CPU is idle and needs to be | |
797 | * triggered to reevaluate the timer wheel when nohz is | |
798 | * active. We are protected against the other CPU fiddling | |
799 | * with the timer by holding the timer base lock. This also | |
800 | * makes sure that a CPU on the way to idle can not evaluate | |
801 | * the timer wheel. | |
802 | */ | |
803 | wake_up_idle_cpu(cpu); | |
804 | spin_unlock_irqrestore(&base->lock, flags); | |
805 | } | |
a9862e05 | 806 | EXPORT_SYMBOL_GPL(add_timer_on); |
74019224 | 807 | |
2aae4a10 | 808 | /** |
1da177e4 LT |
809 | * del_timer - deactive a timer. |
810 | * @timer: the timer to be deactivated | |
811 | * | |
812 | * del_timer() deactivates a timer - this works on both active and inactive | |
813 | * timers. | |
814 | * | |
815 | * The function returns whether it has deactivated a pending timer or not. | |
816 | * (ie. del_timer() of an inactive timer returns 0, del_timer() of an | |
817 | * active timer returns 1.) | |
818 | */ | |
819 | int del_timer(struct timer_list *timer) | |
820 | { | |
a6fa8e5a | 821 | struct tvec_base *base; |
1da177e4 | 822 | unsigned long flags; |
55c888d6 | 823 | int ret = 0; |
1da177e4 | 824 | |
82f67cd9 | 825 | timer_stats_timer_clear_start_info(timer); |
55c888d6 ON |
826 | if (timer_pending(timer)) { |
827 | base = lock_timer_base(timer, &flags); | |
828 | if (timer_pending(timer)) { | |
829 | detach_timer(timer, 1); | |
97fd9ed4 MS |
830 | if (timer->expires == base->next_timer && |
831 | !tbase_get_deferrable(timer->base)) | |
832 | base->next_timer = base->timer_jiffies; | |
55c888d6 ON |
833 | ret = 1; |
834 | } | |
1da177e4 | 835 | spin_unlock_irqrestore(&base->lock, flags); |
1da177e4 | 836 | } |
1da177e4 | 837 | |
55c888d6 | 838 | return ret; |
1da177e4 | 839 | } |
1da177e4 LT |
840 | EXPORT_SYMBOL(del_timer); |
841 | ||
842 | #ifdef CONFIG_SMP | |
2aae4a10 REB |
843 | /** |
844 | * try_to_del_timer_sync - Try to deactivate a timer | |
845 | * @timer: timer do del | |
846 | * | |
fd450b73 ON |
847 | * This function tries to deactivate a timer. Upon successful (ret >= 0) |
848 | * exit the timer is not queued and the handler is not running on any CPU. | |
849 | * | |
850 | * It must not be called from interrupt contexts. | |
851 | */ | |
852 | int try_to_del_timer_sync(struct timer_list *timer) | |
853 | { | |
a6fa8e5a | 854 | struct tvec_base *base; |
fd450b73 ON |
855 | unsigned long flags; |
856 | int ret = -1; | |
857 | ||
858 | base = lock_timer_base(timer, &flags); | |
859 | ||
860 | if (base->running_timer == timer) | |
861 | goto out; | |
862 | ||
863 | ret = 0; | |
864 | if (timer_pending(timer)) { | |
865 | detach_timer(timer, 1); | |
97fd9ed4 MS |
866 | if (timer->expires == base->next_timer && |
867 | !tbase_get_deferrable(timer->base)) | |
868 | base->next_timer = base->timer_jiffies; | |
fd450b73 ON |
869 | ret = 1; |
870 | } | |
871 | out: | |
872 | spin_unlock_irqrestore(&base->lock, flags); | |
873 | ||
874 | return ret; | |
875 | } | |
e19dff1f DH |
876 | EXPORT_SYMBOL(try_to_del_timer_sync); |
877 | ||
2aae4a10 | 878 | /** |
1da177e4 LT |
879 | * del_timer_sync - deactivate a timer and wait for the handler to finish. |
880 | * @timer: the timer to be deactivated | |
881 | * | |
882 | * This function only differs from del_timer() on SMP: besides deactivating | |
883 | * the timer it also makes sure the handler has finished executing on other | |
884 | * CPUs. | |
885 | * | |
72fd4a35 | 886 | * Synchronization rules: Callers must prevent restarting of the timer, |
1da177e4 LT |
887 | * otherwise this function is meaningless. It must not be called from |
888 | * interrupt contexts. The caller must not hold locks which would prevent | |
55c888d6 ON |
889 | * completion of the timer's handler. The timer's handler must not call |
890 | * add_timer_on(). Upon exit the timer is not queued and the handler is | |
891 | * not running on any CPU. | |
1da177e4 LT |
892 | * |
893 | * The function returns whether it has deactivated a pending timer or not. | |
1da177e4 LT |
894 | */ |
895 | int del_timer_sync(struct timer_list *timer) | |
896 | { | |
6f2b9b9a JB |
897 | #ifdef CONFIG_LOCKDEP |
898 | unsigned long flags; | |
899 | ||
900 | local_irq_save(flags); | |
901 | lock_map_acquire(&timer->lockdep_map); | |
902 | lock_map_release(&timer->lockdep_map); | |
903 | local_irq_restore(flags); | |
904 | #endif | |
905 | ||
fd450b73 ON |
906 | for (;;) { |
907 | int ret = try_to_del_timer_sync(timer); | |
908 | if (ret >= 0) | |
909 | return ret; | |
a0009652 | 910 | cpu_relax(); |
fd450b73 | 911 | } |
1da177e4 | 912 | } |
55c888d6 | 913 | EXPORT_SYMBOL(del_timer_sync); |
1da177e4 LT |
914 | #endif |
915 | ||
a6fa8e5a | 916 | static int cascade(struct tvec_base *base, struct tvec *tv, int index) |
1da177e4 LT |
917 | { |
918 | /* cascade all the timers from tv up one level */ | |
3439dd86 P |
919 | struct timer_list *timer, *tmp; |
920 | struct list_head tv_list; | |
921 | ||
922 | list_replace_init(tv->vec + index, &tv_list); | |
1da177e4 | 923 | |
1da177e4 | 924 | /* |
3439dd86 P |
925 | * We are removing _all_ timers from the list, so we |
926 | * don't have to detach them individually. | |
1da177e4 | 927 | */ |
3439dd86 | 928 | list_for_each_entry_safe(timer, tmp, &tv_list, entry) { |
6e453a67 | 929 | BUG_ON(tbase_get_base(timer->base) != base); |
3439dd86 | 930 | internal_add_timer(base, timer); |
1da177e4 | 931 | } |
1da177e4 LT |
932 | |
933 | return index; | |
934 | } | |
935 | ||
2aae4a10 REB |
936 | #define INDEX(N) ((base->timer_jiffies >> (TVR_BITS + (N) * TVN_BITS)) & TVN_MASK) |
937 | ||
938 | /** | |
1da177e4 LT |
939 | * __run_timers - run all expired timers (if any) on this CPU. |
940 | * @base: the timer vector to be processed. | |
941 | * | |
942 | * This function cascades all vectors and executes all expired timer | |
943 | * vectors. | |
944 | */ | |
a6fa8e5a | 945 | static inline void __run_timers(struct tvec_base *base) |
1da177e4 LT |
946 | { |
947 | struct timer_list *timer; | |
948 | ||
3691c519 | 949 | spin_lock_irq(&base->lock); |
1da177e4 | 950 | while (time_after_eq(jiffies, base->timer_jiffies)) { |
626ab0e6 | 951 | struct list_head work_list; |
1da177e4 | 952 | struct list_head *head = &work_list; |
6819457d | 953 | int index = base->timer_jiffies & TVR_MASK; |
626ab0e6 | 954 | |
1da177e4 LT |
955 | /* |
956 | * Cascade timers: | |
957 | */ | |
958 | if (!index && | |
959 | (!cascade(base, &base->tv2, INDEX(0))) && | |
960 | (!cascade(base, &base->tv3, INDEX(1))) && | |
961 | !cascade(base, &base->tv4, INDEX(2))) | |
962 | cascade(base, &base->tv5, INDEX(3)); | |
626ab0e6 ON |
963 | ++base->timer_jiffies; |
964 | list_replace_init(base->tv1.vec + index, &work_list); | |
55c888d6 | 965 | while (!list_empty(head)) { |
1da177e4 LT |
966 | void (*fn)(unsigned long); |
967 | unsigned long data; | |
968 | ||
b5e61818 | 969 | timer = list_first_entry(head, struct timer_list,entry); |
6819457d TG |
970 | fn = timer->function; |
971 | data = timer->data; | |
1da177e4 | 972 | |
82f67cd9 IM |
973 | timer_stats_account_timer(timer); |
974 | ||
1da177e4 | 975 | set_running_timer(base, timer); |
55c888d6 | 976 | detach_timer(timer, 1); |
6f2b9b9a | 977 | |
3691c519 | 978 | spin_unlock_irq(&base->lock); |
1da177e4 | 979 | { |
be5b4fbd | 980 | int preempt_count = preempt_count(); |
6f2b9b9a JB |
981 | |
982 | #ifdef CONFIG_LOCKDEP | |
983 | /* | |
984 | * It is permissible to free the timer from | |
985 | * inside the function that is called from | |
986 | * it, this we need to take into account for | |
987 | * lockdep too. To avoid bogus "held lock | |
988 | * freed" warnings as well as problems when | |
989 | * looking into timer->lockdep_map, make a | |
990 | * copy and use that here. | |
991 | */ | |
992 | struct lockdep_map lockdep_map = | |
993 | timer->lockdep_map; | |
994 | #endif | |
995 | /* | |
996 | * Couple the lock chain with the lock chain at | |
997 | * del_timer_sync() by acquiring the lock_map | |
998 | * around the fn() call here and in | |
999 | * del_timer_sync(). | |
1000 | */ | |
1001 | lock_map_acquire(&lockdep_map); | |
1002 | ||
1da177e4 | 1003 | fn(data); |
6f2b9b9a JB |
1004 | |
1005 | lock_map_release(&lockdep_map); | |
1006 | ||
1da177e4 | 1007 | if (preempt_count != preempt_count()) { |
4c9dc641 | 1008 | printk(KERN_ERR "huh, entered %p " |
be5b4fbd JJ |
1009 | "with preempt_count %08x, exited" |
1010 | " with %08x?\n", | |
1011 | fn, preempt_count, | |
1012 | preempt_count()); | |
1da177e4 LT |
1013 | BUG(); |
1014 | } | |
1015 | } | |
3691c519 | 1016 | spin_lock_irq(&base->lock); |
1da177e4 LT |
1017 | } |
1018 | } | |
1019 | set_running_timer(base, NULL); | |
3691c519 | 1020 | spin_unlock_irq(&base->lock); |
1da177e4 LT |
1021 | } |
1022 | ||
ee9c5785 | 1023 | #ifdef CONFIG_NO_HZ |
1da177e4 LT |
1024 | /* |
1025 | * Find out when the next timer event is due to happen. This | |
90cba64a RD |
1026 | * is used on S/390 to stop all activity when a CPU is idle. |
1027 | * This function needs to be called with interrupts disabled. | |
1da177e4 | 1028 | */ |
a6fa8e5a | 1029 | static unsigned long __next_timer_interrupt(struct tvec_base *base) |
1da177e4 | 1030 | { |
1cfd6849 | 1031 | unsigned long timer_jiffies = base->timer_jiffies; |
eaad084b | 1032 | unsigned long expires = timer_jiffies + NEXT_TIMER_MAX_DELTA; |
1cfd6849 | 1033 | int index, slot, array, found = 0; |
1da177e4 | 1034 | struct timer_list *nte; |
a6fa8e5a | 1035 | struct tvec *varray[4]; |
1da177e4 LT |
1036 | |
1037 | /* Look for timer events in tv1. */ | |
1cfd6849 | 1038 | index = slot = timer_jiffies & TVR_MASK; |
1da177e4 | 1039 | do { |
1cfd6849 | 1040 | list_for_each_entry(nte, base->tv1.vec + slot, entry) { |
6819457d TG |
1041 | if (tbase_get_deferrable(nte->base)) |
1042 | continue; | |
6e453a67 | 1043 | |
1cfd6849 | 1044 | found = 1; |
1da177e4 | 1045 | expires = nte->expires; |
1cfd6849 TG |
1046 | /* Look at the cascade bucket(s)? */ |
1047 | if (!index || slot < index) | |
1048 | goto cascade; | |
1049 | return expires; | |
1da177e4 | 1050 | } |
1cfd6849 TG |
1051 | slot = (slot + 1) & TVR_MASK; |
1052 | } while (slot != index); | |
1053 | ||
1054 | cascade: | |
1055 | /* Calculate the next cascade event */ | |
1056 | if (index) | |
1057 | timer_jiffies += TVR_SIZE - index; | |
1058 | timer_jiffies >>= TVR_BITS; | |
1da177e4 LT |
1059 | |
1060 | /* Check tv2-tv5. */ | |
1061 | varray[0] = &base->tv2; | |
1062 | varray[1] = &base->tv3; | |
1063 | varray[2] = &base->tv4; | |
1064 | varray[3] = &base->tv5; | |
1cfd6849 TG |
1065 | |
1066 | for (array = 0; array < 4; array++) { | |
a6fa8e5a | 1067 | struct tvec *varp = varray[array]; |
1cfd6849 TG |
1068 | |
1069 | index = slot = timer_jiffies & TVN_MASK; | |
1da177e4 | 1070 | do { |
1cfd6849 | 1071 | list_for_each_entry(nte, varp->vec + slot, entry) { |
a0419888 JH |
1072 | if (tbase_get_deferrable(nte->base)) |
1073 | continue; | |
1074 | ||
1cfd6849 | 1075 | found = 1; |
1da177e4 LT |
1076 | if (time_before(nte->expires, expires)) |
1077 | expires = nte->expires; | |
1cfd6849 TG |
1078 | } |
1079 | /* | |
1080 | * Do we still search for the first timer or are | |
1081 | * we looking up the cascade buckets ? | |
1082 | */ | |
1083 | if (found) { | |
1084 | /* Look at the cascade bucket(s)? */ | |
1085 | if (!index || slot < index) | |
1086 | break; | |
1087 | return expires; | |
1088 | } | |
1089 | slot = (slot + 1) & TVN_MASK; | |
1090 | } while (slot != index); | |
1091 | ||
1092 | if (index) | |
1093 | timer_jiffies += TVN_SIZE - index; | |
1094 | timer_jiffies >>= TVN_BITS; | |
1da177e4 | 1095 | } |
1cfd6849 TG |
1096 | return expires; |
1097 | } | |
69239749 | 1098 | |
1cfd6849 TG |
1099 | /* |
1100 | * Check, if the next hrtimer event is before the next timer wheel | |
1101 | * event: | |
1102 | */ | |
1103 | static unsigned long cmp_next_hrtimer_event(unsigned long now, | |
1104 | unsigned long expires) | |
1105 | { | |
1106 | ktime_t hr_delta = hrtimer_get_next_event(); | |
1107 | struct timespec tsdelta; | |
9501b6cf | 1108 | unsigned long delta; |
1cfd6849 TG |
1109 | |
1110 | if (hr_delta.tv64 == KTIME_MAX) | |
1111 | return expires; | |
0662b713 | 1112 | |
9501b6cf TG |
1113 | /* |
1114 | * Expired timer available, let it expire in the next tick | |
1115 | */ | |
1116 | if (hr_delta.tv64 <= 0) | |
1117 | return now + 1; | |
69239749 | 1118 | |
1cfd6849 | 1119 | tsdelta = ktime_to_timespec(hr_delta); |
9501b6cf | 1120 | delta = timespec_to_jiffies(&tsdelta); |
eaad084b TG |
1121 | |
1122 | /* | |
1123 | * Limit the delta to the max value, which is checked in | |
1124 | * tick_nohz_stop_sched_tick(): | |
1125 | */ | |
1126 | if (delta > NEXT_TIMER_MAX_DELTA) | |
1127 | delta = NEXT_TIMER_MAX_DELTA; | |
1128 | ||
9501b6cf TG |
1129 | /* |
1130 | * Take rounding errors in to account and make sure, that it | |
1131 | * expires in the next tick. Otherwise we go into an endless | |
1132 | * ping pong due to tick_nohz_stop_sched_tick() retriggering | |
1133 | * the timer softirq | |
1134 | */ | |
1135 | if (delta < 1) | |
1136 | delta = 1; | |
1137 | now += delta; | |
1cfd6849 TG |
1138 | if (time_before(now, expires)) |
1139 | return now; | |
1da177e4 LT |
1140 | return expires; |
1141 | } | |
1cfd6849 TG |
1142 | |
1143 | /** | |
8dce39c2 | 1144 | * get_next_timer_interrupt - return the jiffy of the next pending timer |
05fb6bf0 | 1145 | * @now: current time (in jiffies) |
1cfd6849 | 1146 | */ |
fd064b9b | 1147 | unsigned long get_next_timer_interrupt(unsigned long now) |
1cfd6849 | 1148 | { |
a6fa8e5a | 1149 | struct tvec_base *base = __get_cpu_var(tvec_bases); |
fd064b9b | 1150 | unsigned long expires; |
1cfd6849 TG |
1151 | |
1152 | spin_lock(&base->lock); | |
97fd9ed4 MS |
1153 | if (time_before_eq(base->next_timer, base->timer_jiffies)) |
1154 | base->next_timer = __next_timer_interrupt(base); | |
1155 | expires = base->next_timer; | |
1cfd6849 TG |
1156 | spin_unlock(&base->lock); |
1157 | ||
1158 | if (time_before_eq(expires, now)) | |
1159 | return now; | |
1160 | ||
1161 | return cmp_next_hrtimer_event(now, expires); | |
1162 | } | |
1da177e4 LT |
1163 | #endif |
1164 | ||
1da177e4 | 1165 | /* |
5b4db0c2 | 1166 | * Called from the timer interrupt handler to charge one tick to the current |
1da177e4 LT |
1167 | * process. user_tick is 1 if the tick is user time, 0 for system. |
1168 | */ | |
1169 | void update_process_times(int user_tick) | |
1170 | { | |
1171 | struct task_struct *p = current; | |
1172 | int cpu = smp_processor_id(); | |
1173 | ||
1174 | /* Note: this timer irq context must be accounted for as well. */ | |
fa13a5a1 | 1175 | account_process_tick(p, user_tick); |
1da177e4 | 1176 | run_local_timers(); |
a157229c | 1177 | rcu_check_callbacks(cpu, user_tick); |
b845b517 | 1178 | printk_tick(); |
1da177e4 | 1179 | scheduler_tick(); |
6819457d | 1180 | run_posix_cpu_timers(p); |
1da177e4 LT |
1181 | } |
1182 | ||
1da177e4 LT |
1183 | /* |
1184 | * This function runs timers and the timer-tq in bottom half context. | |
1185 | */ | |
1186 | static void run_timer_softirq(struct softirq_action *h) | |
1187 | { | |
a6fa8e5a | 1188 | struct tvec_base *base = __get_cpu_var(tvec_bases); |
1da177e4 | 1189 | |
cdd6c482 | 1190 | perf_event_do_pending(); |
925d519a | 1191 | |
d3d74453 | 1192 | hrtimer_run_pending(); |
82f67cd9 | 1193 | |
1da177e4 LT |
1194 | if (time_after_eq(jiffies, base->timer_jiffies)) |
1195 | __run_timers(base); | |
1196 | } | |
1197 | ||
1198 | /* | |
1199 | * Called by the local, per-CPU timer interrupt on SMP. | |
1200 | */ | |
1201 | void run_local_timers(void) | |
1202 | { | |
d3d74453 | 1203 | hrtimer_run_queues(); |
1da177e4 | 1204 | raise_softirq(TIMER_SOFTIRQ); |
6687a97d | 1205 | softlockup_tick(); |
1da177e4 LT |
1206 | } |
1207 | ||
1da177e4 LT |
1208 | /* |
1209 | * The 64-bit jiffies value is not atomic - you MUST NOT read it | |
1210 | * without sampling the sequence number in xtime_lock. | |
1211 | * jiffies is defined in the linker script... | |
1212 | */ | |
1213 | ||
3171a030 | 1214 | void do_timer(unsigned long ticks) |
1da177e4 | 1215 | { |
3171a030 | 1216 | jiffies_64 += ticks; |
dce48a84 TG |
1217 | update_wall_time(); |
1218 | calc_global_load(); | |
1da177e4 LT |
1219 | } |
1220 | ||
1221 | #ifdef __ARCH_WANT_SYS_ALARM | |
1222 | ||
1223 | /* | |
1224 | * For backwards compatibility? This can be done in libc so Alpha | |
1225 | * and all newer ports shouldn't need it. | |
1226 | */ | |
58fd3aa2 | 1227 | SYSCALL_DEFINE1(alarm, unsigned int, seconds) |
1da177e4 | 1228 | { |
c08b8a49 | 1229 | return alarm_setitimer(seconds); |
1da177e4 LT |
1230 | } |
1231 | ||
1232 | #endif | |
1233 | ||
1234 | #ifndef __alpha__ | |
1235 | ||
1236 | /* | |
1237 | * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this | |
1238 | * should be moved into arch/i386 instead? | |
1239 | */ | |
1240 | ||
1241 | /** | |
1242 | * sys_getpid - return the thread group id of the current process | |
1243 | * | |
1244 | * Note, despite the name, this returns the tgid not the pid. The tgid and | |
1245 | * the pid are identical unless CLONE_THREAD was specified on clone() in | |
1246 | * which case the tgid is the same in all threads of the same group. | |
1247 | * | |
1248 | * This is SMP safe as current->tgid does not change. | |
1249 | */ | |
58fd3aa2 | 1250 | SYSCALL_DEFINE0(getpid) |
1da177e4 | 1251 | { |
b488893a | 1252 | return task_tgid_vnr(current); |
1da177e4 LT |
1253 | } |
1254 | ||
1255 | /* | |
6997a6fa KK |
1256 | * Accessing ->real_parent is not SMP-safe, it could |
1257 | * change from under us. However, we can use a stale | |
1258 | * value of ->real_parent under rcu_read_lock(), see | |
1259 | * release_task()->call_rcu(delayed_put_task_struct). | |
1da177e4 | 1260 | */ |
dbf040d9 | 1261 | SYSCALL_DEFINE0(getppid) |
1da177e4 LT |
1262 | { |
1263 | int pid; | |
1da177e4 | 1264 | |
6997a6fa | 1265 | rcu_read_lock(); |
6c5f3e7b | 1266 | pid = task_tgid_vnr(current->real_parent); |
6997a6fa | 1267 | rcu_read_unlock(); |
1da177e4 | 1268 | |
1da177e4 LT |
1269 | return pid; |
1270 | } | |
1271 | ||
dbf040d9 | 1272 | SYSCALL_DEFINE0(getuid) |
1da177e4 LT |
1273 | { |
1274 | /* Only we change this so SMP safe */ | |
76aac0e9 | 1275 | return current_uid(); |
1da177e4 LT |
1276 | } |
1277 | ||
dbf040d9 | 1278 | SYSCALL_DEFINE0(geteuid) |
1da177e4 LT |
1279 | { |
1280 | /* Only we change this so SMP safe */ | |
76aac0e9 | 1281 | return current_euid(); |
1da177e4 LT |
1282 | } |
1283 | ||
dbf040d9 | 1284 | SYSCALL_DEFINE0(getgid) |
1da177e4 LT |
1285 | { |
1286 | /* Only we change this so SMP safe */ | |
76aac0e9 | 1287 | return current_gid(); |
1da177e4 LT |
1288 | } |
1289 | ||
dbf040d9 | 1290 | SYSCALL_DEFINE0(getegid) |
1da177e4 LT |
1291 | { |
1292 | /* Only we change this so SMP safe */ | |
76aac0e9 | 1293 | return current_egid(); |
1da177e4 LT |
1294 | } |
1295 | ||
1296 | #endif | |
1297 | ||
1298 | static void process_timeout(unsigned long __data) | |
1299 | { | |
36c8b586 | 1300 | wake_up_process((struct task_struct *)__data); |
1da177e4 LT |
1301 | } |
1302 | ||
1303 | /** | |
1304 | * schedule_timeout - sleep until timeout | |
1305 | * @timeout: timeout value in jiffies | |
1306 | * | |
1307 | * Make the current task sleep until @timeout jiffies have | |
1308 | * elapsed. The routine will return immediately unless | |
1309 | * the current task state has been set (see set_current_state()). | |
1310 | * | |
1311 | * You can set the task state as follows - | |
1312 | * | |
1313 | * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to | |
1314 | * pass before the routine returns. The routine will return 0 | |
1315 | * | |
1316 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
1317 | * delivered to the current task. In this case the remaining time | |
1318 | * in jiffies will be returned, or 0 if the timer expired in time | |
1319 | * | |
1320 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1321 | * routine returns. | |
1322 | * | |
1323 | * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule | |
1324 | * the CPU away without a bound on the timeout. In this case the return | |
1325 | * value will be %MAX_SCHEDULE_TIMEOUT. | |
1326 | * | |
1327 | * In all cases the return value is guaranteed to be non-negative. | |
1328 | */ | |
7ad5b3a5 | 1329 | signed long __sched schedule_timeout(signed long timeout) |
1da177e4 LT |
1330 | { |
1331 | struct timer_list timer; | |
1332 | unsigned long expire; | |
1333 | ||
1334 | switch (timeout) | |
1335 | { | |
1336 | case MAX_SCHEDULE_TIMEOUT: | |
1337 | /* | |
1338 | * These two special cases are useful to be comfortable | |
1339 | * in the caller. Nothing more. We could take | |
1340 | * MAX_SCHEDULE_TIMEOUT from one of the negative value | |
1341 | * but I' d like to return a valid offset (>=0) to allow | |
1342 | * the caller to do everything it want with the retval. | |
1343 | */ | |
1344 | schedule(); | |
1345 | goto out; | |
1346 | default: | |
1347 | /* | |
1348 | * Another bit of PARANOID. Note that the retval will be | |
1349 | * 0 since no piece of kernel is supposed to do a check | |
1350 | * for a negative retval of schedule_timeout() (since it | |
1351 | * should never happens anyway). You just have the printk() | |
1352 | * that will tell you if something is gone wrong and where. | |
1353 | */ | |
5b149bcc | 1354 | if (timeout < 0) { |
1da177e4 | 1355 | printk(KERN_ERR "schedule_timeout: wrong timeout " |
5b149bcc AM |
1356 | "value %lx\n", timeout); |
1357 | dump_stack(); | |
1da177e4 LT |
1358 | current->state = TASK_RUNNING; |
1359 | goto out; | |
1360 | } | |
1361 | } | |
1362 | ||
1363 | expire = timeout + jiffies; | |
1364 | ||
c6f3a97f | 1365 | setup_timer_on_stack(&timer, process_timeout, (unsigned long)current); |
597d0275 | 1366 | __mod_timer(&timer, expire, false, TIMER_NOT_PINNED); |
1da177e4 LT |
1367 | schedule(); |
1368 | del_singleshot_timer_sync(&timer); | |
1369 | ||
c6f3a97f TG |
1370 | /* Remove the timer from the object tracker */ |
1371 | destroy_timer_on_stack(&timer); | |
1372 | ||
1da177e4 LT |
1373 | timeout = expire - jiffies; |
1374 | ||
1375 | out: | |
1376 | return timeout < 0 ? 0 : timeout; | |
1377 | } | |
1da177e4 LT |
1378 | EXPORT_SYMBOL(schedule_timeout); |
1379 | ||
8a1c1757 AM |
1380 | /* |
1381 | * We can use __set_current_state() here because schedule_timeout() calls | |
1382 | * schedule() unconditionally. | |
1383 | */ | |
64ed93a2 NA |
1384 | signed long __sched schedule_timeout_interruptible(signed long timeout) |
1385 | { | |
a5a0d52c AM |
1386 | __set_current_state(TASK_INTERRUPTIBLE); |
1387 | return schedule_timeout(timeout); | |
64ed93a2 NA |
1388 | } |
1389 | EXPORT_SYMBOL(schedule_timeout_interruptible); | |
1390 | ||
294d5cc2 MW |
1391 | signed long __sched schedule_timeout_killable(signed long timeout) |
1392 | { | |
1393 | __set_current_state(TASK_KILLABLE); | |
1394 | return schedule_timeout(timeout); | |
1395 | } | |
1396 | EXPORT_SYMBOL(schedule_timeout_killable); | |
1397 | ||
64ed93a2 NA |
1398 | signed long __sched schedule_timeout_uninterruptible(signed long timeout) |
1399 | { | |
a5a0d52c AM |
1400 | __set_current_state(TASK_UNINTERRUPTIBLE); |
1401 | return schedule_timeout(timeout); | |
64ed93a2 NA |
1402 | } |
1403 | EXPORT_SYMBOL(schedule_timeout_uninterruptible); | |
1404 | ||
1da177e4 | 1405 | /* Thread ID - the internal kernel "pid" */ |
58fd3aa2 | 1406 | SYSCALL_DEFINE0(gettid) |
1da177e4 | 1407 | { |
b488893a | 1408 | return task_pid_vnr(current); |
1da177e4 LT |
1409 | } |
1410 | ||
2aae4a10 | 1411 | /** |
d4d23add | 1412 | * do_sysinfo - fill in sysinfo struct |
2aae4a10 | 1413 | * @info: pointer to buffer to fill |
6819457d | 1414 | */ |
d4d23add | 1415 | int do_sysinfo(struct sysinfo *info) |
1da177e4 | 1416 | { |
1da177e4 LT |
1417 | unsigned long mem_total, sav_total; |
1418 | unsigned int mem_unit, bitcount; | |
2d02494f | 1419 | struct timespec tp; |
1da177e4 | 1420 | |
d4d23add | 1421 | memset(info, 0, sizeof(struct sysinfo)); |
1da177e4 | 1422 | |
2d02494f TG |
1423 | ktime_get_ts(&tp); |
1424 | monotonic_to_bootbased(&tp); | |
1425 | info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0); | |
1da177e4 | 1426 | |
2d02494f | 1427 | get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT); |
1da177e4 | 1428 | |
2d02494f | 1429 | info->procs = nr_threads; |
1da177e4 | 1430 | |
d4d23add KM |
1431 | si_meminfo(info); |
1432 | si_swapinfo(info); | |
1da177e4 LT |
1433 | |
1434 | /* | |
1435 | * If the sum of all the available memory (i.e. ram + swap) | |
1436 | * is less than can be stored in a 32 bit unsigned long then | |
1437 | * we can be binary compatible with 2.2.x kernels. If not, | |
1438 | * well, in that case 2.2.x was broken anyways... | |
1439 | * | |
1440 | * -Erik Andersen <andersee@debian.org> | |
1441 | */ | |
1442 | ||
d4d23add KM |
1443 | mem_total = info->totalram + info->totalswap; |
1444 | if (mem_total < info->totalram || mem_total < info->totalswap) | |
1da177e4 LT |
1445 | goto out; |
1446 | bitcount = 0; | |
d4d23add | 1447 | mem_unit = info->mem_unit; |
1da177e4 LT |
1448 | while (mem_unit > 1) { |
1449 | bitcount++; | |
1450 | mem_unit >>= 1; | |
1451 | sav_total = mem_total; | |
1452 | mem_total <<= 1; | |
1453 | if (mem_total < sav_total) | |
1454 | goto out; | |
1455 | } | |
1456 | ||
1457 | /* | |
1458 | * If mem_total did not overflow, multiply all memory values by | |
d4d23add | 1459 | * info->mem_unit and set it to 1. This leaves things compatible |
1da177e4 LT |
1460 | * with 2.2.x, and also retains compatibility with earlier 2.4.x |
1461 | * kernels... | |
1462 | */ | |
1463 | ||
d4d23add KM |
1464 | info->mem_unit = 1; |
1465 | info->totalram <<= bitcount; | |
1466 | info->freeram <<= bitcount; | |
1467 | info->sharedram <<= bitcount; | |
1468 | info->bufferram <<= bitcount; | |
1469 | info->totalswap <<= bitcount; | |
1470 | info->freeswap <<= bitcount; | |
1471 | info->totalhigh <<= bitcount; | |
1472 | info->freehigh <<= bitcount; | |
1473 | ||
1474 | out: | |
1475 | return 0; | |
1476 | } | |
1477 | ||
1e7bfb21 | 1478 | SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info) |
d4d23add KM |
1479 | { |
1480 | struct sysinfo val; | |
1481 | ||
1482 | do_sysinfo(&val); | |
1da177e4 | 1483 | |
1da177e4 LT |
1484 | if (copy_to_user(info, &val, sizeof(struct sysinfo))) |
1485 | return -EFAULT; | |
1486 | ||
1487 | return 0; | |
1488 | } | |
1489 | ||
b4be6258 | 1490 | static int __cpuinit init_timers_cpu(int cpu) |
1da177e4 LT |
1491 | { |
1492 | int j; | |
a6fa8e5a | 1493 | struct tvec_base *base; |
b4be6258 | 1494 | static char __cpuinitdata tvec_base_done[NR_CPUS]; |
55c888d6 | 1495 | |
ba6edfcd | 1496 | if (!tvec_base_done[cpu]) { |
a4a6198b JB |
1497 | static char boot_done; |
1498 | ||
a4a6198b | 1499 | if (boot_done) { |
ba6edfcd AM |
1500 | /* |
1501 | * The APs use this path later in boot | |
1502 | */ | |
94f6030c CL |
1503 | base = kmalloc_node(sizeof(*base), |
1504 | GFP_KERNEL | __GFP_ZERO, | |
a4a6198b JB |
1505 | cpu_to_node(cpu)); |
1506 | if (!base) | |
1507 | return -ENOMEM; | |
6e453a67 VP |
1508 | |
1509 | /* Make sure that tvec_base is 2 byte aligned */ | |
1510 | if (tbase_get_deferrable(base)) { | |
1511 | WARN_ON(1); | |
1512 | kfree(base); | |
1513 | return -ENOMEM; | |
1514 | } | |
ba6edfcd | 1515 | per_cpu(tvec_bases, cpu) = base; |
a4a6198b | 1516 | } else { |
ba6edfcd AM |
1517 | /* |
1518 | * This is for the boot CPU - we use compile-time | |
1519 | * static initialisation because per-cpu memory isn't | |
1520 | * ready yet and because the memory allocators are not | |
1521 | * initialised either. | |
1522 | */ | |
a4a6198b | 1523 | boot_done = 1; |
ba6edfcd | 1524 | base = &boot_tvec_bases; |
a4a6198b | 1525 | } |
ba6edfcd AM |
1526 | tvec_base_done[cpu] = 1; |
1527 | } else { | |
1528 | base = per_cpu(tvec_bases, cpu); | |
a4a6198b | 1529 | } |
ba6edfcd | 1530 | |
3691c519 | 1531 | spin_lock_init(&base->lock); |
d730e882 | 1532 | |
1da177e4 LT |
1533 | for (j = 0; j < TVN_SIZE; j++) { |
1534 | INIT_LIST_HEAD(base->tv5.vec + j); | |
1535 | INIT_LIST_HEAD(base->tv4.vec + j); | |
1536 | INIT_LIST_HEAD(base->tv3.vec + j); | |
1537 | INIT_LIST_HEAD(base->tv2.vec + j); | |
1538 | } | |
1539 | for (j = 0; j < TVR_SIZE; j++) | |
1540 | INIT_LIST_HEAD(base->tv1.vec + j); | |
1541 | ||
1542 | base->timer_jiffies = jiffies; | |
97fd9ed4 | 1543 | base->next_timer = base->timer_jiffies; |
a4a6198b | 1544 | return 0; |
1da177e4 LT |
1545 | } |
1546 | ||
1547 | #ifdef CONFIG_HOTPLUG_CPU | |
a6fa8e5a | 1548 | static void migrate_timer_list(struct tvec_base *new_base, struct list_head *head) |
1da177e4 LT |
1549 | { |
1550 | struct timer_list *timer; | |
1551 | ||
1552 | while (!list_empty(head)) { | |
b5e61818 | 1553 | timer = list_first_entry(head, struct timer_list, entry); |
55c888d6 | 1554 | detach_timer(timer, 0); |
6e453a67 | 1555 | timer_set_base(timer, new_base); |
97fd9ed4 MS |
1556 | if (time_before(timer->expires, new_base->next_timer) && |
1557 | !tbase_get_deferrable(timer->base)) | |
1558 | new_base->next_timer = timer->expires; | |
1da177e4 | 1559 | internal_add_timer(new_base, timer); |
1da177e4 | 1560 | } |
1da177e4 LT |
1561 | } |
1562 | ||
48ccf3da | 1563 | static void __cpuinit migrate_timers(int cpu) |
1da177e4 | 1564 | { |
a6fa8e5a PM |
1565 | struct tvec_base *old_base; |
1566 | struct tvec_base *new_base; | |
1da177e4 LT |
1567 | int i; |
1568 | ||
1569 | BUG_ON(cpu_online(cpu)); | |
a4a6198b JB |
1570 | old_base = per_cpu(tvec_bases, cpu); |
1571 | new_base = get_cpu_var(tvec_bases); | |
d82f0b0f ON |
1572 | /* |
1573 | * The caller is globally serialized and nobody else | |
1574 | * takes two locks at once, deadlock is not possible. | |
1575 | */ | |
1576 | spin_lock_irq(&new_base->lock); | |
0d180406 | 1577 | spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING); |
3691c519 ON |
1578 | |
1579 | BUG_ON(old_base->running_timer); | |
1da177e4 | 1580 | |
1da177e4 | 1581 | for (i = 0; i < TVR_SIZE; i++) |
55c888d6 ON |
1582 | migrate_timer_list(new_base, old_base->tv1.vec + i); |
1583 | for (i = 0; i < TVN_SIZE; i++) { | |
1584 | migrate_timer_list(new_base, old_base->tv2.vec + i); | |
1585 | migrate_timer_list(new_base, old_base->tv3.vec + i); | |
1586 | migrate_timer_list(new_base, old_base->tv4.vec + i); | |
1587 | migrate_timer_list(new_base, old_base->tv5.vec + i); | |
1588 | } | |
1589 | ||
0d180406 | 1590 | spin_unlock(&old_base->lock); |
d82f0b0f | 1591 | spin_unlock_irq(&new_base->lock); |
1da177e4 | 1592 | put_cpu_var(tvec_bases); |
1da177e4 LT |
1593 | } |
1594 | #endif /* CONFIG_HOTPLUG_CPU */ | |
1595 | ||
8c78f307 | 1596 | static int __cpuinit timer_cpu_notify(struct notifier_block *self, |
1da177e4 LT |
1597 | unsigned long action, void *hcpu) |
1598 | { | |
1599 | long cpu = (long)hcpu; | |
1600 | switch(action) { | |
1601 | case CPU_UP_PREPARE: | |
8bb78442 | 1602 | case CPU_UP_PREPARE_FROZEN: |
a4a6198b JB |
1603 | if (init_timers_cpu(cpu) < 0) |
1604 | return NOTIFY_BAD; | |
1da177e4 LT |
1605 | break; |
1606 | #ifdef CONFIG_HOTPLUG_CPU | |
1607 | case CPU_DEAD: | |
8bb78442 | 1608 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
1609 | migrate_timers(cpu); |
1610 | break; | |
1611 | #endif | |
1612 | default: | |
1613 | break; | |
1614 | } | |
1615 | return NOTIFY_OK; | |
1616 | } | |
1617 | ||
8c78f307 | 1618 | static struct notifier_block __cpuinitdata timers_nb = { |
1da177e4 LT |
1619 | .notifier_call = timer_cpu_notify, |
1620 | }; | |
1621 | ||
1622 | ||
1623 | void __init init_timers(void) | |
1624 | { | |
07dccf33 | 1625 | int err = timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE, |
1da177e4 | 1626 | (void *)(long)smp_processor_id()); |
07dccf33 | 1627 | |
82f67cd9 IM |
1628 | init_timer_stats(); |
1629 | ||
07dccf33 | 1630 | BUG_ON(err == NOTIFY_BAD); |
1da177e4 | 1631 | register_cpu_notifier(&timers_nb); |
962cf36c | 1632 | open_softirq(TIMER_SOFTIRQ, run_timer_softirq); |
1da177e4 LT |
1633 | } |
1634 | ||
1da177e4 LT |
1635 | /** |
1636 | * msleep - sleep safely even with waitqueue interruptions | |
1637 | * @msecs: Time in milliseconds to sleep for | |
1638 | */ | |
1639 | void msleep(unsigned int msecs) | |
1640 | { | |
1641 | unsigned long timeout = msecs_to_jiffies(msecs) + 1; | |
1642 | ||
75bcc8c5 NA |
1643 | while (timeout) |
1644 | timeout = schedule_timeout_uninterruptible(timeout); | |
1da177e4 LT |
1645 | } |
1646 | ||
1647 | EXPORT_SYMBOL(msleep); | |
1648 | ||
1649 | /** | |
96ec3efd | 1650 | * msleep_interruptible - sleep waiting for signals |
1da177e4 LT |
1651 | * @msecs: Time in milliseconds to sleep for |
1652 | */ | |
1653 | unsigned long msleep_interruptible(unsigned int msecs) | |
1654 | { | |
1655 | unsigned long timeout = msecs_to_jiffies(msecs) + 1; | |
1656 | ||
75bcc8c5 NA |
1657 | while (timeout && !signal_pending(current)) |
1658 | timeout = schedule_timeout_interruptible(timeout); | |
1da177e4 LT |
1659 | return jiffies_to_msecs(timeout); |
1660 | } | |
1661 | ||
1662 | EXPORT_SYMBOL(msleep_interruptible); |