Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/kernel/timer.c | |
3 | * | |
4a22f166 | 4 | * Kernel internal timers |
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> | |
9984de1a | 23 | #include <linux/export.h> |
1da177e4 LT |
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> |
e360adbe | 40 | #include <linux/irq_work.h> |
eea08f32 | 41 | #include <linux/sched.h> |
cf4aebc2 | 42 | #include <linux/sched/sysctl.h> |
5a0e3ad6 | 43 | #include <linux/slab.h> |
1a0df594 | 44 | #include <linux/compat.h> |
1da177e4 LT |
45 | |
46 | #include <asm/uaccess.h> | |
47 | #include <asm/unistd.h> | |
48 | #include <asm/div64.h> | |
49 | #include <asm/timex.h> | |
50 | #include <asm/io.h> | |
51 | ||
c1ad348b TG |
52 | #include "tick-internal.h" |
53 | ||
2b022e3d XG |
54 | #define CREATE_TRACE_POINTS |
55 | #include <trace/events/timer.h> | |
56 | ||
40747ffa | 57 | __visible u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES; |
ecea8d19 TG |
58 | |
59 | EXPORT_SYMBOL(jiffies_64); | |
60 | ||
1da177e4 LT |
61 | /* |
62 | * per-CPU timer vector definitions: | |
63 | */ | |
1da177e4 LT |
64 | #define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6) |
65 | #define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8) | |
66 | #define TVN_SIZE (1 << TVN_BITS) | |
67 | #define TVR_SIZE (1 << TVR_BITS) | |
68 | #define TVN_MASK (TVN_SIZE - 1) | |
69 | #define TVR_MASK (TVR_SIZE - 1) | |
26cff4e2 | 70 | #define MAX_TVAL ((unsigned long)((1ULL << (TVR_BITS + 4*TVN_BITS)) - 1)) |
1da177e4 | 71 | |
a6fa8e5a | 72 | struct tvec { |
1dabbcec | 73 | struct hlist_head vec[TVN_SIZE]; |
a6fa8e5a | 74 | }; |
1da177e4 | 75 | |
a6fa8e5a | 76 | struct tvec_root { |
1dabbcec | 77 | struct hlist_head vec[TVR_SIZE]; |
a6fa8e5a | 78 | }; |
1da177e4 | 79 | |
a6fa8e5a | 80 | struct tvec_base { |
3691c519 ON |
81 | spinlock_t lock; |
82 | struct timer_list *running_timer; | |
1da177e4 | 83 | unsigned long timer_jiffies; |
97fd9ed4 | 84 | unsigned long next_timer; |
99d5f3aa | 85 | unsigned long active_timers; |
fff42158 | 86 | unsigned long all_timers; |
d6f93829 | 87 | int cpu; |
bc7a34b8 | 88 | bool migration_enabled; |
683be13a | 89 | bool nohz_active; |
a6fa8e5a PM |
90 | struct tvec_root tv1; |
91 | struct tvec tv2; | |
92 | struct tvec tv3; | |
93 | struct tvec tv4; | |
94 | struct tvec tv5; | |
6e453a67 | 95 | } ____cacheline_aligned; |
1da177e4 | 96 | |
e52b1db3 | 97 | |
0eeda71b | 98 | static DEFINE_PER_CPU(struct tvec_base, tvec_bases); |
6e453a67 | 99 | |
bc7a34b8 TG |
100 | #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON) |
101 | unsigned int sysctl_timer_migration = 1; | |
102 | ||
683be13a | 103 | void timers_update_migration(bool update_nohz) |
bc7a34b8 TG |
104 | { |
105 | bool on = sysctl_timer_migration && tick_nohz_active; | |
106 | unsigned int cpu; | |
107 | ||
108 | /* Avoid the loop, if nothing to update */ | |
109 | if (this_cpu_read(tvec_bases.migration_enabled) == on) | |
110 | return; | |
111 | ||
112 | for_each_possible_cpu(cpu) { | |
113 | per_cpu(tvec_bases.migration_enabled, cpu) = on; | |
114 | per_cpu(hrtimer_bases.migration_enabled, cpu) = on; | |
683be13a TG |
115 | if (!update_nohz) |
116 | continue; | |
117 | per_cpu(tvec_bases.nohz_active, cpu) = true; | |
118 | per_cpu(hrtimer_bases.nohz_active, cpu) = true; | |
bc7a34b8 TG |
119 | } |
120 | } | |
121 | ||
122 | int timer_migration_handler(struct ctl_table *table, int write, | |
123 | void __user *buffer, size_t *lenp, | |
124 | loff_t *ppos) | |
125 | { | |
126 | static DEFINE_MUTEX(mutex); | |
127 | int ret; | |
128 | ||
129 | mutex_lock(&mutex); | |
130 | ret = proc_dointvec(table, write, buffer, lenp, ppos); | |
131 | if (!ret && write) | |
683be13a | 132 | timers_update_migration(false); |
bc7a34b8 TG |
133 | mutex_unlock(&mutex); |
134 | return ret; | |
135 | } | |
136 | ||
137 | static inline struct tvec_base *get_target_base(struct tvec_base *base, | |
138 | int pinned) | |
139 | { | |
140 | if (pinned || !base->migration_enabled) | |
141 | return this_cpu_ptr(&tvec_bases); | |
142 | return per_cpu_ptr(&tvec_bases, get_nohz_timer_target()); | |
143 | } | |
144 | #else | |
145 | static inline struct tvec_base *get_target_base(struct tvec_base *base, | |
146 | int pinned) | |
147 | { | |
148 | return this_cpu_ptr(&tvec_bases); | |
149 | } | |
150 | #endif | |
151 | ||
9c133c46 AS |
152 | static unsigned long round_jiffies_common(unsigned long j, int cpu, |
153 | bool force_up) | |
4c36a5de AV |
154 | { |
155 | int rem; | |
156 | unsigned long original = j; | |
157 | ||
158 | /* | |
159 | * We don't want all cpus firing their timers at once hitting the | |
160 | * same lock or cachelines, so we skew each extra cpu with an extra | |
161 | * 3 jiffies. This 3 jiffies came originally from the mm/ code which | |
162 | * already did this. | |
163 | * The skew is done by adding 3*cpunr, then round, then subtract this | |
164 | * extra offset again. | |
165 | */ | |
166 | j += cpu * 3; | |
167 | ||
168 | rem = j % HZ; | |
169 | ||
170 | /* | |
171 | * If the target jiffie is just after a whole second (which can happen | |
172 | * due to delays of the timer irq, long irq off times etc etc) then | |
173 | * we should round down to the whole second, not up. Use 1/4th second | |
174 | * as cutoff for this rounding as an extreme upper bound for this. | |
9c133c46 | 175 | * But never round down if @force_up is set. |
4c36a5de | 176 | */ |
9c133c46 | 177 | if (rem < HZ/4 && !force_up) /* round down */ |
4c36a5de AV |
178 | j = j - rem; |
179 | else /* round up */ | |
180 | j = j - rem + HZ; | |
181 | ||
182 | /* now that we have rounded, subtract the extra skew again */ | |
183 | j -= cpu * 3; | |
184 | ||
9e04d380 BVA |
185 | /* |
186 | * Make sure j is still in the future. Otherwise return the | |
187 | * unmodified value. | |
188 | */ | |
189 | return time_is_after_jiffies(j) ? j : original; | |
4c36a5de | 190 | } |
9c133c46 AS |
191 | |
192 | /** | |
193 | * __round_jiffies - function to round jiffies to a full second | |
194 | * @j: the time in (absolute) jiffies that should be rounded | |
195 | * @cpu: the processor number on which the timeout will happen | |
196 | * | |
197 | * __round_jiffies() rounds an absolute time in the future (in jiffies) | |
198 | * up or down to (approximately) full seconds. This is useful for timers | |
199 | * for which the exact time they fire does not matter too much, as long as | |
200 | * they fire approximately every X seconds. | |
201 | * | |
202 | * By rounding these timers to whole seconds, all such timers will fire | |
203 | * at the same time, rather than at various times spread out. The goal | |
204 | * of this is to have the CPU wake up less, which saves power. | |
205 | * | |
206 | * The exact rounding is skewed for each processor to avoid all | |
207 | * processors firing at the exact same time, which could lead | |
208 | * to lock contention or spurious cache line bouncing. | |
209 | * | |
210 | * The return value is the rounded version of the @j parameter. | |
211 | */ | |
212 | unsigned long __round_jiffies(unsigned long j, int cpu) | |
213 | { | |
214 | return round_jiffies_common(j, cpu, false); | |
215 | } | |
4c36a5de AV |
216 | EXPORT_SYMBOL_GPL(__round_jiffies); |
217 | ||
218 | /** | |
219 | * __round_jiffies_relative - function to round jiffies to a full second | |
220 | * @j: the time in (relative) jiffies that should be rounded | |
221 | * @cpu: the processor number on which the timeout will happen | |
222 | * | |
72fd4a35 | 223 | * __round_jiffies_relative() rounds a time delta in the future (in jiffies) |
4c36a5de AV |
224 | * up or down to (approximately) full seconds. This is useful for timers |
225 | * for which the exact time they fire does not matter too much, as long as | |
226 | * they fire approximately every X seconds. | |
227 | * | |
228 | * By rounding these timers to whole seconds, all such timers will fire | |
229 | * at the same time, rather than at various times spread out. The goal | |
230 | * of this is to have the CPU wake up less, which saves power. | |
231 | * | |
232 | * The exact rounding is skewed for each processor to avoid all | |
233 | * processors firing at the exact same time, which could lead | |
234 | * to lock contention or spurious cache line bouncing. | |
235 | * | |
72fd4a35 | 236 | * The return value is the rounded version of the @j parameter. |
4c36a5de AV |
237 | */ |
238 | unsigned long __round_jiffies_relative(unsigned long j, int cpu) | |
239 | { | |
9c133c46 AS |
240 | unsigned long j0 = jiffies; |
241 | ||
242 | /* Use j0 because jiffies might change while we run */ | |
243 | return round_jiffies_common(j + j0, cpu, false) - j0; | |
4c36a5de AV |
244 | } |
245 | EXPORT_SYMBOL_GPL(__round_jiffies_relative); | |
246 | ||
247 | /** | |
248 | * round_jiffies - function to round jiffies to a full second | |
249 | * @j: the time in (absolute) jiffies that should be rounded | |
250 | * | |
72fd4a35 | 251 | * round_jiffies() rounds an absolute time in the future (in jiffies) |
4c36a5de AV |
252 | * up or down to (approximately) full seconds. This is useful for timers |
253 | * for which the exact time they fire does not matter too much, as long as | |
254 | * they fire approximately every X seconds. | |
255 | * | |
256 | * By rounding these timers to whole seconds, all such timers will fire | |
257 | * at the same time, rather than at various times spread out. The goal | |
258 | * of this is to have the CPU wake up less, which saves power. | |
259 | * | |
72fd4a35 | 260 | * The return value is the rounded version of the @j parameter. |
4c36a5de AV |
261 | */ |
262 | unsigned long round_jiffies(unsigned long j) | |
263 | { | |
9c133c46 | 264 | return round_jiffies_common(j, raw_smp_processor_id(), false); |
4c36a5de AV |
265 | } |
266 | EXPORT_SYMBOL_GPL(round_jiffies); | |
267 | ||
268 | /** | |
269 | * round_jiffies_relative - function to round jiffies to a full second | |
270 | * @j: the time in (relative) jiffies that should be rounded | |
271 | * | |
72fd4a35 | 272 | * round_jiffies_relative() rounds a time delta in the future (in jiffies) |
4c36a5de AV |
273 | * up or down to (approximately) full seconds. This is useful for timers |
274 | * for which the exact time they fire does not matter too much, as long as | |
275 | * they fire approximately every X seconds. | |
276 | * | |
277 | * By rounding these timers to whole seconds, all such timers will fire | |
278 | * at the same time, rather than at various times spread out. The goal | |
279 | * of this is to have the CPU wake up less, which saves power. | |
280 | * | |
72fd4a35 | 281 | * The return value is the rounded version of the @j parameter. |
4c36a5de AV |
282 | */ |
283 | unsigned long round_jiffies_relative(unsigned long j) | |
284 | { | |
285 | return __round_jiffies_relative(j, raw_smp_processor_id()); | |
286 | } | |
287 | EXPORT_SYMBOL_GPL(round_jiffies_relative); | |
288 | ||
9c133c46 AS |
289 | /** |
290 | * __round_jiffies_up - function to round jiffies up to a full second | |
291 | * @j: the time in (absolute) jiffies that should be rounded | |
292 | * @cpu: the processor number on which the timeout will happen | |
293 | * | |
294 | * This is the same as __round_jiffies() except that it will never | |
295 | * round down. This is useful for timeouts for which the exact time | |
296 | * of firing does not matter too much, as long as they don't fire too | |
297 | * early. | |
298 | */ | |
299 | unsigned long __round_jiffies_up(unsigned long j, int cpu) | |
300 | { | |
301 | return round_jiffies_common(j, cpu, true); | |
302 | } | |
303 | EXPORT_SYMBOL_GPL(__round_jiffies_up); | |
304 | ||
305 | /** | |
306 | * __round_jiffies_up_relative - function to round jiffies up to a full second | |
307 | * @j: the time in (relative) jiffies that should be rounded | |
308 | * @cpu: the processor number on which the timeout will happen | |
309 | * | |
310 | * This is the same as __round_jiffies_relative() except that it will never | |
311 | * round down. This is useful for timeouts for which the exact time | |
312 | * of firing does not matter too much, as long as they don't fire too | |
313 | * early. | |
314 | */ | |
315 | unsigned long __round_jiffies_up_relative(unsigned long j, int cpu) | |
316 | { | |
317 | unsigned long j0 = jiffies; | |
318 | ||
319 | /* Use j0 because jiffies might change while we run */ | |
320 | return round_jiffies_common(j + j0, cpu, true) - j0; | |
321 | } | |
322 | EXPORT_SYMBOL_GPL(__round_jiffies_up_relative); | |
323 | ||
324 | /** | |
325 | * round_jiffies_up - function to round jiffies up to a full second | |
326 | * @j: the time in (absolute) jiffies that should be rounded | |
327 | * | |
328 | * This is the same as round_jiffies() except that it will never | |
329 | * round down. This is useful for timeouts for which the exact time | |
330 | * of firing does not matter too much, as long as they don't fire too | |
331 | * early. | |
332 | */ | |
333 | unsigned long round_jiffies_up(unsigned long j) | |
334 | { | |
335 | return round_jiffies_common(j, raw_smp_processor_id(), true); | |
336 | } | |
337 | EXPORT_SYMBOL_GPL(round_jiffies_up); | |
338 | ||
339 | /** | |
340 | * round_jiffies_up_relative - function to round jiffies up to a full second | |
341 | * @j: the time in (relative) jiffies that should be rounded | |
342 | * | |
343 | * This is the same as round_jiffies_relative() except that it will never | |
344 | * round down. This is useful for timeouts for which the exact time | |
345 | * of firing does not matter too much, as long as they don't fire too | |
346 | * early. | |
347 | */ | |
348 | unsigned long round_jiffies_up_relative(unsigned long j) | |
349 | { | |
350 | return __round_jiffies_up_relative(j, raw_smp_processor_id()); | |
351 | } | |
352 | EXPORT_SYMBOL_GPL(round_jiffies_up_relative); | |
353 | ||
3bbb9ec9 AV |
354 | /** |
355 | * set_timer_slack - set the allowed slack for a timer | |
0caa6210 | 356 | * @timer: the timer to be modified |
3bbb9ec9 AV |
357 | * @slack_hz: the amount of time (in jiffies) allowed for rounding |
358 | * | |
359 | * Set the amount of time, in jiffies, that a certain timer has | |
360 | * in terms of slack. By setting this value, the timer subsystem | |
361 | * will schedule the actual timer somewhere between | |
362 | * the time mod_timer() asks for, and that time plus the slack. | |
363 | * | |
364 | * By setting the slack to -1, a percentage of the delay is used | |
365 | * instead. | |
366 | */ | |
367 | void set_timer_slack(struct timer_list *timer, int slack_hz) | |
368 | { | |
369 | timer->slack = slack_hz; | |
370 | } | |
371 | EXPORT_SYMBOL_GPL(set_timer_slack); | |
372 | ||
facbb4a7 TG |
373 | static void |
374 | __internal_add_timer(struct tvec_base *base, struct timer_list *timer) | |
1da177e4 LT |
375 | { |
376 | unsigned long expires = timer->expires; | |
377 | unsigned long idx = expires - base->timer_jiffies; | |
1dabbcec | 378 | struct hlist_head *vec; |
1da177e4 LT |
379 | |
380 | if (idx < TVR_SIZE) { | |
381 | int i = expires & TVR_MASK; | |
382 | vec = base->tv1.vec + i; | |
383 | } else if (idx < 1 << (TVR_BITS + TVN_BITS)) { | |
384 | int i = (expires >> TVR_BITS) & TVN_MASK; | |
385 | vec = base->tv2.vec + i; | |
386 | } else if (idx < 1 << (TVR_BITS + 2 * TVN_BITS)) { | |
387 | int i = (expires >> (TVR_BITS + TVN_BITS)) & TVN_MASK; | |
388 | vec = base->tv3.vec + i; | |
389 | } else if (idx < 1 << (TVR_BITS + 3 * TVN_BITS)) { | |
390 | int i = (expires >> (TVR_BITS + 2 * TVN_BITS)) & TVN_MASK; | |
391 | vec = base->tv4.vec + i; | |
392 | } else if ((signed long) idx < 0) { | |
393 | /* | |
394 | * Can happen if you add a timer with expires == jiffies, | |
395 | * or you set a timer to go off in the past | |
396 | */ | |
397 | vec = base->tv1.vec + (base->timer_jiffies & TVR_MASK); | |
398 | } else { | |
399 | int i; | |
26cff4e2 HC |
400 | /* If the timeout is larger than MAX_TVAL (on 64-bit |
401 | * architectures or with CONFIG_BASE_SMALL=1) then we | |
402 | * use the maximum timeout. | |
1da177e4 | 403 | */ |
26cff4e2 HC |
404 | if (idx > MAX_TVAL) { |
405 | idx = MAX_TVAL; | |
1da177e4 LT |
406 | expires = idx + base->timer_jiffies; |
407 | } | |
408 | i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK; | |
409 | vec = base->tv5.vec + i; | |
410 | } | |
1bd04bf6 | 411 | |
1dabbcec | 412 | hlist_add_head(&timer->entry, vec); |
1da177e4 LT |
413 | } |
414 | ||
facbb4a7 TG |
415 | static void internal_add_timer(struct tvec_base *base, struct timer_list *timer) |
416 | { | |
3bb475a3 TG |
417 | /* Advance base->jiffies, if the base is empty */ |
418 | if (!base->all_timers++) | |
419 | base->timer_jiffies = jiffies; | |
420 | ||
facbb4a7 TG |
421 | __internal_add_timer(base, timer); |
422 | /* | |
99d5f3aa | 423 | * Update base->active_timers and base->next_timer |
facbb4a7 | 424 | */ |
0eeda71b | 425 | if (!(timer->flags & TIMER_DEFERRABLE)) { |
aea369b9 ON |
426 | if (!base->active_timers++ || |
427 | time_before(timer->expires, base->next_timer)) | |
99d5f3aa | 428 | base->next_timer = timer->expires; |
99d5f3aa | 429 | } |
9f6d9baa VK |
430 | |
431 | /* | |
432 | * Check whether the other CPU is in dynticks mode and needs | |
433 | * to be triggered to reevaluate the timer wheel. | |
434 | * We are protected against the other CPU fiddling | |
435 | * with the timer by holding the timer base lock. This also | |
436 | * makes sure that a CPU on the way to stop its tick can not | |
437 | * evaluate the timer wheel. | |
438 | * | |
439 | * Spare the IPI for deferrable timers on idle targets though. | |
440 | * The next busy ticks will take care of it. Except full dynticks | |
441 | * require special care against races with idle_cpu(), lets deal | |
442 | * with that later. | |
443 | */ | |
683be13a TG |
444 | if (base->nohz_active) { |
445 | if (!(timer->flags & TIMER_DEFERRABLE) || | |
446 | tick_nohz_full_cpu(base->cpu)) | |
447 | wake_up_nohz_cpu(base->cpu); | |
448 | } | |
facbb4a7 TG |
449 | } |
450 | ||
82f67cd9 IM |
451 | #ifdef CONFIG_TIMER_STATS |
452 | void __timer_stats_timer_set_start_info(struct timer_list *timer, void *addr) | |
453 | { | |
454 | if (timer->start_site) | |
455 | return; | |
456 | ||
457 | timer->start_site = addr; | |
458 | memcpy(timer->start_comm, current->comm, TASK_COMM_LEN); | |
459 | timer->start_pid = current->pid; | |
460 | } | |
c5c061b8 VP |
461 | |
462 | static void timer_stats_account_timer(struct timer_list *timer) | |
463 | { | |
3ed769bd DV |
464 | void *site; |
465 | ||
466 | /* | |
467 | * start_site can be concurrently reset by | |
468 | * timer_stats_timer_clear_start_info() | |
469 | */ | |
470 | site = READ_ONCE(timer->start_site); | |
471 | if (likely(!site)) | |
507e1231 | 472 | return; |
c5c061b8 | 473 | |
3ed769bd | 474 | timer_stats_update_stats(timer, timer->start_pid, site, |
c74441a1 TG |
475 | timer->function, timer->start_comm, |
476 | timer->flags); | |
c5c061b8 VP |
477 | } |
478 | ||
479 | #else | |
480 | static void timer_stats_account_timer(struct timer_list *timer) {} | |
82f67cd9 IM |
481 | #endif |
482 | ||
c6f3a97f TG |
483 | #ifdef CONFIG_DEBUG_OBJECTS_TIMERS |
484 | ||
485 | static struct debug_obj_descr timer_debug_descr; | |
486 | ||
99777288 SG |
487 | static void *timer_debug_hint(void *addr) |
488 | { | |
489 | return ((struct timer_list *) addr)->function; | |
490 | } | |
491 | ||
b9fdac7f DC |
492 | static bool timer_is_static_object(void *addr) |
493 | { | |
494 | struct timer_list *timer = addr; | |
495 | ||
496 | return (timer->entry.pprev == NULL && | |
497 | timer->entry.next == TIMER_ENTRY_STATIC); | |
498 | } | |
499 | ||
c6f3a97f TG |
500 | /* |
501 | * fixup_init is called when: | |
502 | * - an active object is initialized | |
55c888d6 | 503 | */ |
e3252464 | 504 | static bool timer_fixup_init(void *addr, enum debug_obj_state state) |
c6f3a97f TG |
505 | { |
506 | struct timer_list *timer = addr; | |
507 | ||
508 | switch (state) { | |
509 | case ODEBUG_STATE_ACTIVE: | |
510 | del_timer_sync(timer); | |
511 | debug_object_init(timer, &timer_debug_descr); | |
e3252464 | 512 | return true; |
c6f3a97f | 513 | default: |
e3252464 | 514 | return false; |
c6f3a97f TG |
515 | } |
516 | } | |
517 | ||
fb16b8cf SB |
518 | /* Stub timer callback for improperly used timers. */ |
519 | static void stub_timer(unsigned long data) | |
520 | { | |
521 | WARN_ON(1); | |
522 | } | |
523 | ||
c6f3a97f TG |
524 | /* |
525 | * fixup_activate is called when: | |
526 | * - an active object is activated | |
b9fdac7f | 527 | * - an unknown non-static object is activated |
c6f3a97f | 528 | */ |
e3252464 | 529 | static bool timer_fixup_activate(void *addr, enum debug_obj_state state) |
c6f3a97f TG |
530 | { |
531 | struct timer_list *timer = addr; | |
532 | ||
533 | switch (state) { | |
c6f3a97f | 534 | case ODEBUG_STATE_NOTAVAILABLE: |
b9fdac7f DC |
535 | setup_timer(timer, stub_timer, 0); |
536 | return true; | |
c6f3a97f TG |
537 | |
538 | case ODEBUG_STATE_ACTIVE: | |
539 | WARN_ON(1); | |
540 | ||
541 | default: | |
e3252464 | 542 | return false; |
c6f3a97f TG |
543 | } |
544 | } | |
545 | ||
546 | /* | |
547 | * fixup_free is called when: | |
548 | * - an active object is freed | |
549 | */ | |
e3252464 | 550 | static bool timer_fixup_free(void *addr, enum debug_obj_state state) |
c6f3a97f TG |
551 | { |
552 | struct timer_list *timer = addr; | |
553 | ||
554 | switch (state) { | |
555 | case ODEBUG_STATE_ACTIVE: | |
556 | del_timer_sync(timer); | |
557 | debug_object_free(timer, &timer_debug_descr); | |
e3252464 | 558 | return true; |
c6f3a97f | 559 | default: |
e3252464 | 560 | return false; |
c6f3a97f TG |
561 | } |
562 | } | |
563 | ||
dc4218bd CC |
564 | /* |
565 | * fixup_assert_init is called when: | |
566 | * - an untracked/uninit-ed object is found | |
567 | */ | |
e3252464 | 568 | static bool timer_fixup_assert_init(void *addr, enum debug_obj_state state) |
dc4218bd CC |
569 | { |
570 | struct timer_list *timer = addr; | |
571 | ||
572 | switch (state) { | |
573 | case ODEBUG_STATE_NOTAVAILABLE: | |
b9fdac7f DC |
574 | setup_timer(timer, stub_timer, 0); |
575 | return true; | |
dc4218bd | 576 | default: |
e3252464 | 577 | return false; |
dc4218bd CC |
578 | } |
579 | } | |
580 | ||
c6f3a97f | 581 | static struct debug_obj_descr timer_debug_descr = { |
dc4218bd CC |
582 | .name = "timer_list", |
583 | .debug_hint = timer_debug_hint, | |
b9fdac7f | 584 | .is_static_object = timer_is_static_object, |
dc4218bd CC |
585 | .fixup_init = timer_fixup_init, |
586 | .fixup_activate = timer_fixup_activate, | |
587 | .fixup_free = timer_fixup_free, | |
588 | .fixup_assert_init = timer_fixup_assert_init, | |
c6f3a97f TG |
589 | }; |
590 | ||
591 | static inline void debug_timer_init(struct timer_list *timer) | |
592 | { | |
593 | debug_object_init(timer, &timer_debug_descr); | |
594 | } | |
595 | ||
596 | static inline void debug_timer_activate(struct timer_list *timer) | |
597 | { | |
598 | debug_object_activate(timer, &timer_debug_descr); | |
599 | } | |
600 | ||
601 | static inline void debug_timer_deactivate(struct timer_list *timer) | |
602 | { | |
603 | debug_object_deactivate(timer, &timer_debug_descr); | |
604 | } | |
605 | ||
606 | static inline void debug_timer_free(struct timer_list *timer) | |
607 | { | |
608 | debug_object_free(timer, &timer_debug_descr); | |
609 | } | |
610 | ||
dc4218bd CC |
611 | static inline void debug_timer_assert_init(struct timer_list *timer) |
612 | { | |
613 | debug_object_assert_init(timer, &timer_debug_descr); | |
614 | } | |
615 | ||
fc683995 TH |
616 | static void do_init_timer(struct timer_list *timer, unsigned int flags, |
617 | const char *name, struct lock_class_key *key); | |
c6f3a97f | 618 | |
fc683995 TH |
619 | void init_timer_on_stack_key(struct timer_list *timer, unsigned int flags, |
620 | const char *name, struct lock_class_key *key) | |
c6f3a97f TG |
621 | { |
622 | debug_object_init_on_stack(timer, &timer_debug_descr); | |
fc683995 | 623 | do_init_timer(timer, flags, name, key); |
c6f3a97f | 624 | } |
6f2b9b9a | 625 | EXPORT_SYMBOL_GPL(init_timer_on_stack_key); |
c6f3a97f TG |
626 | |
627 | void destroy_timer_on_stack(struct timer_list *timer) | |
628 | { | |
629 | debug_object_free(timer, &timer_debug_descr); | |
630 | } | |
631 | EXPORT_SYMBOL_GPL(destroy_timer_on_stack); | |
632 | ||
633 | #else | |
634 | static inline void debug_timer_init(struct timer_list *timer) { } | |
635 | static inline void debug_timer_activate(struct timer_list *timer) { } | |
636 | static inline void debug_timer_deactivate(struct timer_list *timer) { } | |
dc4218bd | 637 | static inline void debug_timer_assert_init(struct timer_list *timer) { } |
c6f3a97f TG |
638 | #endif |
639 | ||
2b022e3d XG |
640 | static inline void debug_init(struct timer_list *timer) |
641 | { | |
642 | debug_timer_init(timer); | |
643 | trace_timer_init(timer); | |
644 | } | |
645 | ||
646 | static inline void | |
647 | debug_activate(struct timer_list *timer, unsigned long expires) | |
648 | { | |
649 | debug_timer_activate(timer); | |
0eeda71b | 650 | trace_timer_start(timer, expires, timer->flags); |
2b022e3d XG |
651 | } |
652 | ||
653 | static inline void debug_deactivate(struct timer_list *timer) | |
654 | { | |
655 | debug_timer_deactivate(timer); | |
656 | trace_timer_cancel(timer); | |
657 | } | |
658 | ||
dc4218bd CC |
659 | static inline void debug_assert_init(struct timer_list *timer) |
660 | { | |
661 | debug_timer_assert_init(timer); | |
662 | } | |
663 | ||
fc683995 TH |
664 | static void do_init_timer(struct timer_list *timer, unsigned int flags, |
665 | const char *name, struct lock_class_key *key) | |
55c888d6 | 666 | { |
1dabbcec | 667 | timer->entry.pprev = NULL; |
0eeda71b | 668 | timer->flags = flags | raw_smp_processor_id(); |
3bbb9ec9 | 669 | timer->slack = -1; |
82f67cd9 IM |
670 | #ifdef CONFIG_TIMER_STATS |
671 | timer->start_site = NULL; | |
672 | timer->start_pid = -1; | |
673 | memset(timer->start_comm, 0, TASK_COMM_LEN); | |
674 | #endif | |
6f2b9b9a | 675 | lockdep_init_map(&timer->lockdep_map, name, key, 0); |
55c888d6 | 676 | } |
c6f3a97f TG |
677 | |
678 | /** | |
633fe795 | 679 | * init_timer_key - initialize a timer |
c6f3a97f | 680 | * @timer: the timer to be initialized |
fc683995 | 681 | * @flags: timer flags |
633fe795 RD |
682 | * @name: name of the timer |
683 | * @key: lockdep class key of the fake lock used for tracking timer | |
684 | * sync lock dependencies | |
c6f3a97f | 685 | * |
633fe795 | 686 | * init_timer_key() must be done to a timer prior calling *any* of the |
c6f3a97f TG |
687 | * other timer functions. |
688 | */ | |
fc683995 TH |
689 | void init_timer_key(struct timer_list *timer, unsigned int flags, |
690 | const char *name, struct lock_class_key *key) | |
c6f3a97f | 691 | { |
2b022e3d | 692 | debug_init(timer); |
fc683995 | 693 | do_init_timer(timer, flags, name, key); |
c6f3a97f | 694 | } |
6f2b9b9a | 695 | EXPORT_SYMBOL(init_timer_key); |
55c888d6 | 696 | |
ec44bc7a | 697 | static inline void detach_timer(struct timer_list *timer, bool clear_pending) |
55c888d6 | 698 | { |
1dabbcec | 699 | struct hlist_node *entry = &timer->entry; |
55c888d6 | 700 | |
2b022e3d | 701 | debug_deactivate(timer); |
c6f3a97f | 702 | |
1dabbcec | 703 | __hlist_del(entry); |
55c888d6 | 704 | if (clear_pending) |
1dabbcec TG |
705 | entry->pprev = NULL; |
706 | entry->next = LIST_POISON2; | |
55c888d6 ON |
707 | } |
708 | ||
99d5f3aa TG |
709 | static inline void |
710 | detach_expired_timer(struct timer_list *timer, struct tvec_base *base) | |
711 | { | |
712 | detach_timer(timer, true); | |
0eeda71b | 713 | if (!(timer->flags & TIMER_DEFERRABLE)) |
e52b1db3 | 714 | base->active_timers--; |
fff42158 | 715 | base->all_timers--; |
99d5f3aa TG |
716 | } |
717 | ||
ec44bc7a TG |
718 | static int detach_if_pending(struct timer_list *timer, struct tvec_base *base, |
719 | bool clear_pending) | |
720 | { | |
721 | if (!timer_pending(timer)) | |
722 | return 0; | |
723 | ||
724 | detach_timer(timer, clear_pending); | |
0eeda71b | 725 | if (!(timer->flags & TIMER_DEFERRABLE)) { |
e52b1db3 | 726 | base->active_timers--; |
99d5f3aa TG |
727 | if (timer->expires == base->next_timer) |
728 | base->next_timer = base->timer_jiffies; | |
729 | } | |
3bb475a3 TG |
730 | /* If this was the last timer, advance base->jiffies */ |
731 | if (!--base->all_timers) | |
732 | base->timer_jiffies = jiffies; | |
ec44bc7a TG |
733 | return 1; |
734 | } | |
735 | ||
55c888d6 | 736 | /* |
3691c519 | 737 | * We are using hashed locking: holding per_cpu(tvec_bases).lock |
55c888d6 ON |
738 | * means that all timers which are tied to this base via timer->base are |
739 | * locked, and the base itself is locked too. | |
740 | * | |
741 | * So __run_timers/migrate_timers can safely modify all timers which could | |
742 | * be found on ->tvX lists. | |
743 | * | |
0eeda71b TG |
744 | * When the timer's base is locked and removed from the list, the |
745 | * TIMER_MIGRATING flag is set, FIXME | |
55c888d6 | 746 | */ |
a6fa8e5a | 747 | static struct tvec_base *lock_timer_base(struct timer_list *timer, |
55c888d6 | 748 | unsigned long *flags) |
89e7e374 | 749 | __acquires(timer->base->lock) |
55c888d6 | 750 | { |
55c888d6 | 751 | for (;;) { |
0eeda71b TG |
752 | u32 tf = timer->flags; |
753 | struct tvec_base *base; | |
754 | ||
755 | if (!(tf & TIMER_MIGRATING)) { | |
756 | base = per_cpu_ptr(&tvec_bases, tf & TIMER_CPUMASK); | |
55c888d6 | 757 | spin_lock_irqsave(&base->lock, *flags); |
0eeda71b | 758 | if (timer->flags == tf) |
55c888d6 | 759 | return base; |
55c888d6 ON |
760 | spin_unlock_irqrestore(&base->lock, *flags); |
761 | } | |
762 | cpu_relax(); | |
763 | } | |
764 | } | |
765 | ||
74019224 | 766 | static inline int |
597d0275 | 767 | __mod_timer(struct timer_list *timer, unsigned long expires, |
bc7a34b8 | 768 | bool pending_only, int pinned) |
1da177e4 | 769 | { |
a6fa8e5a | 770 | struct tvec_base *base, *new_base; |
1da177e4 | 771 | unsigned long flags; |
bc7a34b8 | 772 | int ret = 0; |
1da177e4 | 773 | |
82f67cd9 | 774 | timer_stats_timer_set_start_info(timer); |
1da177e4 | 775 | BUG_ON(!timer->function); |
1da177e4 | 776 | |
55c888d6 ON |
777 | base = lock_timer_base(timer, &flags); |
778 | ||
ec44bc7a TG |
779 | ret = detach_if_pending(timer, base, false); |
780 | if (!ret && pending_only) | |
781 | goto out_unlock; | |
55c888d6 | 782 | |
2b022e3d | 783 | debug_activate(timer, expires); |
c6f3a97f | 784 | |
bc7a34b8 | 785 | new_base = get_target_base(base, pinned); |
eea08f32 | 786 | |
3691c519 | 787 | if (base != new_base) { |
1da177e4 | 788 | /* |
55c888d6 ON |
789 | * We are trying to schedule the timer on the local CPU. |
790 | * However we can't change timer's base while it is running, | |
791 | * otherwise del_timer_sync() can't detect that the timer's | |
792 | * handler yet has not finished. This also guarantees that | |
793 | * the timer is serialized wrt itself. | |
1da177e4 | 794 | */ |
a2c348fe | 795 | if (likely(base->running_timer != timer)) { |
55c888d6 | 796 | /* See the comment in lock_timer_base() */ |
0eeda71b TG |
797 | timer->flags |= TIMER_MIGRATING; |
798 | ||
55c888d6 | 799 | spin_unlock(&base->lock); |
a2c348fe ON |
800 | base = new_base; |
801 | spin_lock(&base->lock); | |
d0023a14 ED |
802 | WRITE_ONCE(timer->flags, |
803 | (timer->flags & ~TIMER_BASEMASK) | base->cpu); | |
1da177e4 LT |
804 | } |
805 | } | |
806 | ||
1da177e4 | 807 | timer->expires = expires; |
a2c348fe | 808 | internal_add_timer(base, timer); |
74019224 IM |
809 | |
810 | out_unlock: | |
a2c348fe | 811 | spin_unlock_irqrestore(&base->lock, flags); |
1da177e4 LT |
812 | |
813 | return ret; | |
814 | } | |
815 | ||
2aae4a10 | 816 | /** |
74019224 IM |
817 | * mod_timer_pending - modify a pending timer's timeout |
818 | * @timer: the pending timer to be modified | |
819 | * @expires: new timeout in jiffies | |
1da177e4 | 820 | * |
74019224 IM |
821 | * mod_timer_pending() is the same for pending timers as mod_timer(), |
822 | * but will not re-activate and modify already deleted timers. | |
823 | * | |
824 | * It is useful for unserialized use of timers. | |
1da177e4 | 825 | */ |
74019224 | 826 | int mod_timer_pending(struct timer_list *timer, unsigned long expires) |
1da177e4 | 827 | { |
597d0275 | 828 | return __mod_timer(timer, expires, true, TIMER_NOT_PINNED); |
1da177e4 | 829 | } |
74019224 | 830 | EXPORT_SYMBOL(mod_timer_pending); |
1da177e4 | 831 | |
3bbb9ec9 AV |
832 | /* |
833 | * Decide where to put the timer while taking the slack into account | |
834 | * | |
835 | * Algorithm: | |
836 | * 1) calculate the maximum (absolute) time | |
837 | * 2) calculate the highest bit where the expires and new max are different | |
838 | * 3) use this bit to make a mask | |
839 | * 4) use the bitmask to round down the maximum time, so that all last | |
840 | * bits are zeros | |
841 | */ | |
842 | static inline | |
843 | unsigned long apply_slack(struct timer_list *timer, unsigned long expires) | |
844 | { | |
845 | unsigned long expires_limit, mask; | |
846 | int bit; | |
847 | ||
8e63d779 | 848 | if (timer->slack >= 0) { |
f00e047e | 849 | expires_limit = expires + timer->slack; |
8e63d779 | 850 | } else { |
1c3cc116 SAS |
851 | long delta = expires - jiffies; |
852 | ||
853 | if (delta < 256) | |
854 | return expires; | |
3bbb9ec9 | 855 | |
1c3cc116 | 856 | expires_limit = expires + delta / 256; |
8e63d779 | 857 | } |
3bbb9ec9 | 858 | mask = expires ^ expires_limit; |
3bbb9ec9 AV |
859 | if (mask == 0) |
860 | return expires; | |
861 | ||
9fc4468d | 862 | bit = __fls(mask); |
3bbb9ec9 | 863 | |
98a01e77 | 864 | mask = (1UL << bit) - 1; |
3bbb9ec9 AV |
865 | |
866 | expires_limit = expires_limit & ~(mask); | |
867 | ||
868 | return expires_limit; | |
869 | } | |
870 | ||
2aae4a10 | 871 | /** |
1da177e4 LT |
872 | * mod_timer - modify a timer's timeout |
873 | * @timer: the timer to be modified | |
2aae4a10 | 874 | * @expires: new timeout in jiffies |
1da177e4 | 875 | * |
72fd4a35 | 876 | * mod_timer() is a more efficient way to update the expire field of an |
1da177e4 LT |
877 | * active timer (if the timer is inactive it will be activated) |
878 | * | |
879 | * mod_timer(timer, expires) is equivalent to: | |
880 | * | |
881 | * del_timer(timer); timer->expires = expires; add_timer(timer); | |
882 | * | |
883 | * Note that if there are multiple unserialized concurrent users of the | |
884 | * same timer, then mod_timer() is the only safe way to modify the timeout, | |
885 | * since add_timer() cannot modify an already running timer. | |
886 | * | |
887 | * The function returns whether it has modified a pending timer or not. | |
888 | * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an | |
889 | * active timer returns 1.) | |
890 | */ | |
891 | int mod_timer(struct timer_list *timer, unsigned long expires) | |
892 | { | |
1c3cc116 SAS |
893 | expires = apply_slack(timer, expires); |
894 | ||
1da177e4 LT |
895 | /* |
896 | * This is a common optimization triggered by the | |
897 | * networking code - if the timer is re-modified | |
898 | * to be the same thing then just return: | |
899 | */ | |
4841158b | 900 | if (timer_pending(timer) && timer->expires == expires) |
1da177e4 LT |
901 | return 1; |
902 | ||
597d0275 | 903 | return __mod_timer(timer, expires, false, TIMER_NOT_PINNED); |
1da177e4 | 904 | } |
1da177e4 LT |
905 | EXPORT_SYMBOL(mod_timer); |
906 | ||
597d0275 AB |
907 | /** |
908 | * mod_timer_pinned - modify a timer's timeout | |
909 | * @timer: the timer to be modified | |
910 | * @expires: new timeout in jiffies | |
911 | * | |
912 | * mod_timer_pinned() is a way to update the expire field of an | |
913 | * active timer (if the timer is inactive it will be activated) | |
048a0e8f PM |
914 | * and to ensure that the timer is scheduled on the current CPU. |
915 | * | |
916 | * Note that this does not prevent the timer from being migrated | |
917 | * when the current CPU goes offline. If this is a problem for | |
918 | * you, use CPU-hotplug notifiers to handle it correctly, for | |
919 | * example, cancelling the timer when the corresponding CPU goes | |
920 | * offline. | |
597d0275 AB |
921 | * |
922 | * mod_timer_pinned(timer, expires) is equivalent to: | |
923 | * | |
924 | * del_timer(timer); timer->expires = expires; add_timer(timer); | |
925 | */ | |
926 | int mod_timer_pinned(struct timer_list *timer, unsigned long expires) | |
927 | { | |
928 | if (timer->expires == expires && timer_pending(timer)) | |
929 | return 1; | |
930 | ||
931 | return __mod_timer(timer, expires, false, TIMER_PINNED); | |
932 | } | |
933 | EXPORT_SYMBOL(mod_timer_pinned); | |
934 | ||
74019224 IM |
935 | /** |
936 | * add_timer - start a timer | |
937 | * @timer: the timer to be added | |
938 | * | |
939 | * The kernel will do a ->function(->data) callback from the | |
940 | * timer interrupt at the ->expires point in the future. The | |
941 | * current time is 'jiffies'. | |
942 | * | |
943 | * The timer's ->expires, ->function (and if the handler uses it, ->data) | |
944 | * fields must be set prior calling this function. | |
945 | * | |
946 | * Timers with an ->expires field in the past will be executed in the next | |
947 | * timer tick. | |
948 | */ | |
949 | void add_timer(struct timer_list *timer) | |
950 | { | |
951 | BUG_ON(timer_pending(timer)); | |
952 | mod_timer(timer, timer->expires); | |
953 | } | |
954 | EXPORT_SYMBOL(add_timer); | |
955 | ||
956 | /** | |
957 | * add_timer_on - start a timer on a particular CPU | |
958 | * @timer: the timer to be added | |
959 | * @cpu: the CPU to start it on | |
960 | * | |
961 | * This is not very scalable on SMP. Double adds are not possible. | |
962 | */ | |
963 | void add_timer_on(struct timer_list *timer, int cpu) | |
964 | { | |
22b886dd TH |
965 | struct tvec_base *new_base = per_cpu_ptr(&tvec_bases, cpu); |
966 | struct tvec_base *base; | |
74019224 IM |
967 | unsigned long flags; |
968 | ||
969 | timer_stats_timer_set_start_info(timer); | |
970 | BUG_ON(timer_pending(timer) || !timer->function); | |
22b886dd TH |
971 | |
972 | /* | |
973 | * If @timer was on a different CPU, it should be migrated with the | |
974 | * old base locked to prevent other operations proceeding with the | |
975 | * wrong base locked. See lock_timer_base(). | |
976 | */ | |
977 | base = lock_timer_base(timer, &flags); | |
978 | if (base != new_base) { | |
979 | timer->flags |= TIMER_MIGRATING; | |
980 | ||
981 | spin_unlock(&base->lock); | |
982 | base = new_base; | |
983 | spin_lock(&base->lock); | |
984 | WRITE_ONCE(timer->flags, | |
985 | (timer->flags & ~TIMER_BASEMASK) | cpu); | |
986 | } | |
987 | ||
2b022e3d | 988 | debug_activate(timer, timer->expires); |
74019224 | 989 | internal_add_timer(base, timer); |
74019224 IM |
990 | spin_unlock_irqrestore(&base->lock, flags); |
991 | } | |
a9862e05 | 992 | EXPORT_SYMBOL_GPL(add_timer_on); |
74019224 | 993 | |
2aae4a10 | 994 | /** |
1da177e4 LT |
995 | * del_timer - deactive a timer. |
996 | * @timer: the timer to be deactivated | |
997 | * | |
998 | * del_timer() deactivates a timer - this works on both active and inactive | |
999 | * timers. | |
1000 | * | |
1001 | * The function returns whether it has deactivated a pending timer or not. | |
1002 | * (ie. del_timer() of an inactive timer returns 0, del_timer() of an | |
1003 | * active timer returns 1.) | |
1004 | */ | |
1005 | int del_timer(struct timer_list *timer) | |
1006 | { | |
a6fa8e5a | 1007 | struct tvec_base *base; |
1da177e4 | 1008 | unsigned long flags; |
55c888d6 | 1009 | int ret = 0; |
1da177e4 | 1010 | |
dc4218bd CC |
1011 | debug_assert_init(timer); |
1012 | ||
82f67cd9 | 1013 | timer_stats_timer_clear_start_info(timer); |
55c888d6 ON |
1014 | if (timer_pending(timer)) { |
1015 | base = lock_timer_base(timer, &flags); | |
ec44bc7a | 1016 | ret = detach_if_pending(timer, base, true); |
1da177e4 | 1017 | spin_unlock_irqrestore(&base->lock, flags); |
1da177e4 | 1018 | } |
1da177e4 | 1019 | |
55c888d6 | 1020 | return ret; |
1da177e4 | 1021 | } |
1da177e4 LT |
1022 | EXPORT_SYMBOL(del_timer); |
1023 | ||
2aae4a10 REB |
1024 | /** |
1025 | * try_to_del_timer_sync - Try to deactivate a timer | |
1026 | * @timer: timer do del | |
1027 | * | |
fd450b73 ON |
1028 | * This function tries to deactivate a timer. Upon successful (ret >= 0) |
1029 | * exit the timer is not queued and the handler is not running on any CPU. | |
fd450b73 ON |
1030 | */ |
1031 | int try_to_del_timer_sync(struct timer_list *timer) | |
1032 | { | |
a6fa8e5a | 1033 | struct tvec_base *base; |
fd450b73 ON |
1034 | unsigned long flags; |
1035 | int ret = -1; | |
1036 | ||
dc4218bd CC |
1037 | debug_assert_init(timer); |
1038 | ||
fd450b73 ON |
1039 | base = lock_timer_base(timer, &flags); |
1040 | ||
ec44bc7a TG |
1041 | if (base->running_timer != timer) { |
1042 | timer_stats_timer_clear_start_info(timer); | |
1043 | ret = detach_if_pending(timer, base, true); | |
fd450b73 | 1044 | } |
fd450b73 ON |
1045 | spin_unlock_irqrestore(&base->lock, flags); |
1046 | ||
1047 | return ret; | |
1048 | } | |
e19dff1f DH |
1049 | EXPORT_SYMBOL(try_to_del_timer_sync); |
1050 | ||
6f1bc451 | 1051 | #ifdef CONFIG_SMP |
2aae4a10 | 1052 | /** |
1da177e4 LT |
1053 | * del_timer_sync - deactivate a timer and wait for the handler to finish. |
1054 | * @timer: the timer to be deactivated | |
1055 | * | |
1056 | * This function only differs from del_timer() on SMP: besides deactivating | |
1057 | * the timer it also makes sure the handler has finished executing on other | |
1058 | * CPUs. | |
1059 | * | |
72fd4a35 | 1060 | * Synchronization rules: Callers must prevent restarting of the timer, |
1da177e4 | 1061 | * otherwise this function is meaningless. It must not be called from |
c5f66e99 TH |
1062 | * interrupt contexts unless the timer is an irqsafe one. The caller must |
1063 | * not hold locks which would prevent completion of the timer's | |
1064 | * handler. The timer's handler must not call add_timer_on(). Upon exit the | |
1065 | * timer is not queued and the handler is not running on any CPU. | |
1da177e4 | 1066 | * |
c5f66e99 TH |
1067 | * Note: For !irqsafe timers, you must not hold locks that are held in |
1068 | * interrupt context while calling this function. Even if the lock has | |
1069 | * nothing to do with the timer in question. Here's why: | |
48228f7b SR |
1070 | * |
1071 | * CPU0 CPU1 | |
1072 | * ---- ---- | |
1073 | * <SOFTIRQ> | |
1074 | * call_timer_fn(); | |
1075 | * base->running_timer = mytimer; | |
1076 | * spin_lock_irq(somelock); | |
1077 | * <IRQ> | |
1078 | * spin_lock(somelock); | |
1079 | * del_timer_sync(mytimer); | |
1080 | * while (base->running_timer == mytimer); | |
1081 | * | |
1082 | * Now del_timer_sync() will never return and never release somelock. | |
1083 | * The interrupt on the other CPU is waiting to grab somelock but | |
1084 | * it has interrupted the softirq that CPU0 is waiting to finish. | |
1085 | * | |
1da177e4 | 1086 | * The function returns whether it has deactivated a pending timer or not. |
1da177e4 LT |
1087 | */ |
1088 | int del_timer_sync(struct timer_list *timer) | |
1089 | { | |
6f2b9b9a | 1090 | #ifdef CONFIG_LOCKDEP |
f266a511 PZ |
1091 | unsigned long flags; |
1092 | ||
48228f7b SR |
1093 | /* |
1094 | * If lockdep gives a backtrace here, please reference | |
1095 | * the synchronization rules above. | |
1096 | */ | |
7ff20792 | 1097 | local_irq_save(flags); |
6f2b9b9a JB |
1098 | lock_map_acquire(&timer->lockdep_map); |
1099 | lock_map_release(&timer->lockdep_map); | |
7ff20792 | 1100 | local_irq_restore(flags); |
6f2b9b9a | 1101 | #endif |
466bd303 YZ |
1102 | /* |
1103 | * don't use it in hardirq context, because it | |
1104 | * could lead to deadlock. | |
1105 | */ | |
0eeda71b | 1106 | WARN_ON(in_irq() && !(timer->flags & TIMER_IRQSAFE)); |
fd450b73 ON |
1107 | for (;;) { |
1108 | int ret = try_to_del_timer_sync(timer); | |
1109 | if (ret >= 0) | |
1110 | return ret; | |
a0009652 | 1111 | cpu_relax(); |
fd450b73 | 1112 | } |
1da177e4 | 1113 | } |
55c888d6 | 1114 | EXPORT_SYMBOL(del_timer_sync); |
1da177e4 LT |
1115 | #endif |
1116 | ||
a6fa8e5a | 1117 | static int cascade(struct tvec_base *base, struct tvec *tv, int index) |
1da177e4 LT |
1118 | { |
1119 | /* cascade all the timers from tv up one level */ | |
1dabbcec TG |
1120 | struct timer_list *timer; |
1121 | struct hlist_node *tmp; | |
1122 | struct hlist_head tv_list; | |
3439dd86 | 1123 | |
1dabbcec | 1124 | hlist_move_list(tv->vec + index, &tv_list); |
1da177e4 | 1125 | |
1da177e4 | 1126 | /* |
3439dd86 P |
1127 | * We are removing _all_ timers from the list, so we |
1128 | * don't have to detach them individually. | |
1da177e4 | 1129 | */ |
1dabbcec | 1130 | hlist_for_each_entry_safe(timer, tmp, &tv_list, entry) { |
facbb4a7 TG |
1131 | /* No accounting, while moving them */ |
1132 | __internal_add_timer(base, timer); | |
1da177e4 | 1133 | } |
1da177e4 LT |
1134 | |
1135 | return index; | |
1136 | } | |
1137 | ||
576da126 TG |
1138 | static void call_timer_fn(struct timer_list *timer, void (*fn)(unsigned long), |
1139 | unsigned long data) | |
1140 | { | |
4a2b4b22 | 1141 | int count = preempt_count(); |
576da126 TG |
1142 | |
1143 | #ifdef CONFIG_LOCKDEP | |
1144 | /* | |
1145 | * It is permissible to free the timer from inside the | |
1146 | * function that is called from it, this we need to take into | |
1147 | * account for lockdep too. To avoid bogus "held lock freed" | |
1148 | * warnings as well as problems when looking into | |
1149 | * timer->lockdep_map, make a copy and use that here. | |
1150 | */ | |
4d82a1de PZ |
1151 | struct lockdep_map lockdep_map; |
1152 | ||
1153 | lockdep_copy_map(&lockdep_map, &timer->lockdep_map); | |
576da126 TG |
1154 | #endif |
1155 | /* | |
1156 | * Couple the lock chain with the lock chain at | |
1157 | * del_timer_sync() by acquiring the lock_map around the fn() | |
1158 | * call here and in del_timer_sync(). | |
1159 | */ | |
1160 | lock_map_acquire(&lockdep_map); | |
1161 | ||
1162 | trace_timer_expire_entry(timer); | |
1163 | fn(data); | |
1164 | trace_timer_expire_exit(timer); | |
1165 | ||
1166 | lock_map_release(&lockdep_map); | |
1167 | ||
4a2b4b22 | 1168 | if (count != preempt_count()) { |
802702e0 | 1169 | WARN_ONCE(1, "timer: %pF preempt leak: %08x -> %08x\n", |
4a2b4b22 | 1170 | fn, count, preempt_count()); |
802702e0 TG |
1171 | /* |
1172 | * Restore the preempt count. That gives us a decent | |
1173 | * chance to survive and extract information. If the | |
1174 | * callback kept a lock held, bad luck, but not worse | |
1175 | * than the BUG() we had. | |
1176 | */ | |
4a2b4b22 | 1177 | preempt_count_set(count); |
576da126 TG |
1178 | } |
1179 | } | |
1180 | ||
2aae4a10 REB |
1181 | #define INDEX(N) ((base->timer_jiffies >> (TVR_BITS + (N) * TVN_BITS)) & TVN_MASK) |
1182 | ||
1183 | /** | |
1da177e4 LT |
1184 | * __run_timers - run all expired timers (if any) on this CPU. |
1185 | * @base: the timer vector to be processed. | |
1186 | * | |
1187 | * This function cascades all vectors and executes all expired timer | |
1188 | * vectors. | |
1189 | */ | |
a6fa8e5a | 1190 | static inline void __run_timers(struct tvec_base *base) |
1da177e4 LT |
1191 | { |
1192 | struct timer_list *timer; | |
1193 | ||
3691c519 | 1194 | spin_lock_irq(&base->lock); |
3bb475a3 | 1195 | |
1da177e4 | 1196 | while (time_after_eq(jiffies, base->timer_jiffies)) { |
1dabbcec TG |
1197 | struct hlist_head work_list; |
1198 | struct hlist_head *head = &work_list; | |
3bb475a3 TG |
1199 | int index; |
1200 | ||
1201 | if (!base->all_timers) { | |
1202 | base->timer_jiffies = jiffies; | |
1203 | break; | |
1204 | } | |
1205 | ||
1206 | index = base->timer_jiffies & TVR_MASK; | |
626ab0e6 | 1207 | |
1da177e4 LT |
1208 | /* |
1209 | * Cascade timers: | |
1210 | */ | |
1211 | if (!index && | |
1212 | (!cascade(base, &base->tv2, INDEX(0))) && | |
1213 | (!cascade(base, &base->tv3, INDEX(1))) && | |
1214 | !cascade(base, &base->tv4, INDEX(2))) | |
1215 | cascade(base, &base->tv5, INDEX(3)); | |
626ab0e6 | 1216 | ++base->timer_jiffies; |
1dabbcec TG |
1217 | hlist_move_list(base->tv1.vec + index, head); |
1218 | while (!hlist_empty(head)) { | |
1da177e4 LT |
1219 | void (*fn)(unsigned long); |
1220 | unsigned long data; | |
c5f66e99 | 1221 | bool irqsafe; |
1da177e4 | 1222 | |
1dabbcec | 1223 | timer = hlist_entry(head->first, struct timer_list, entry); |
6819457d TG |
1224 | fn = timer->function; |
1225 | data = timer->data; | |
0eeda71b | 1226 | irqsafe = timer->flags & TIMER_IRQSAFE; |
1da177e4 | 1227 | |
82f67cd9 IM |
1228 | timer_stats_account_timer(timer); |
1229 | ||
6f1bc451 | 1230 | base->running_timer = timer; |
99d5f3aa | 1231 | detach_expired_timer(timer, base); |
6f2b9b9a | 1232 | |
c5f66e99 TH |
1233 | if (irqsafe) { |
1234 | spin_unlock(&base->lock); | |
1235 | call_timer_fn(timer, fn, data); | |
1236 | spin_lock(&base->lock); | |
1237 | } else { | |
1238 | spin_unlock_irq(&base->lock); | |
1239 | call_timer_fn(timer, fn, data); | |
1240 | spin_lock_irq(&base->lock); | |
1241 | } | |
1da177e4 LT |
1242 | } |
1243 | } | |
6f1bc451 | 1244 | base->running_timer = NULL; |
3691c519 | 1245 | spin_unlock_irq(&base->lock); |
1da177e4 LT |
1246 | } |
1247 | ||
3451d024 | 1248 | #ifdef CONFIG_NO_HZ_COMMON |
1da177e4 LT |
1249 | /* |
1250 | * Find out when the next timer event is due to happen. This | |
90cba64a RD |
1251 | * is used on S/390 to stop all activity when a CPU is idle. |
1252 | * This function needs to be called with interrupts disabled. | |
1da177e4 | 1253 | */ |
a6fa8e5a | 1254 | static unsigned long __next_timer_interrupt(struct tvec_base *base) |
1da177e4 | 1255 | { |
1cfd6849 | 1256 | unsigned long timer_jiffies = base->timer_jiffies; |
eaad084b | 1257 | unsigned long expires = timer_jiffies + NEXT_TIMER_MAX_DELTA; |
1cfd6849 | 1258 | int index, slot, array, found = 0; |
1da177e4 | 1259 | struct timer_list *nte; |
a6fa8e5a | 1260 | struct tvec *varray[4]; |
1da177e4 LT |
1261 | |
1262 | /* Look for timer events in tv1. */ | |
1cfd6849 | 1263 | index = slot = timer_jiffies & TVR_MASK; |
1da177e4 | 1264 | do { |
1dabbcec | 1265 | hlist_for_each_entry(nte, base->tv1.vec + slot, entry) { |
0eeda71b | 1266 | if (nte->flags & TIMER_DEFERRABLE) |
6819457d | 1267 | continue; |
6e453a67 | 1268 | |
1cfd6849 | 1269 | found = 1; |
1da177e4 | 1270 | expires = nte->expires; |
1cfd6849 TG |
1271 | /* Look at the cascade bucket(s)? */ |
1272 | if (!index || slot < index) | |
1273 | goto cascade; | |
1274 | return expires; | |
1da177e4 | 1275 | } |
1cfd6849 TG |
1276 | slot = (slot + 1) & TVR_MASK; |
1277 | } while (slot != index); | |
1278 | ||
1279 | cascade: | |
1280 | /* Calculate the next cascade event */ | |
1281 | if (index) | |
1282 | timer_jiffies += TVR_SIZE - index; | |
1283 | timer_jiffies >>= TVR_BITS; | |
1da177e4 LT |
1284 | |
1285 | /* Check tv2-tv5. */ | |
1286 | varray[0] = &base->tv2; | |
1287 | varray[1] = &base->tv3; | |
1288 | varray[2] = &base->tv4; | |
1289 | varray[3] = &base->tv5; | |
1cfd6849 TG |
1290 | |
1291 | for (array = 0; array < 4; array++) { | |
a6fa8e5a | 1292 | struct tvec *varp = varray[array]; |
1cfd6849 TG |
1293 | |
1294 | index = slot = timer_jiffies & TVN_MASK; | |
1da177e4 | 1295 | do { |
1dabbcec | 1296 | hlist_for_each_entry(nte, varp->vec + slot, entry) { |
0eeda71b | 1297 | if (nte->flags & TIMER_DEFERRABLE) |
a0419888 JH |
1298 | continue; |
1299 | ||
1cfd6849 | 1300 | found = 1; |
1da177e4 LT |
1301 | if (time_before(nte->expires, expires)) |
1302 | expires = nte->expires; | |
1cfd6849 TG |
1303 | } |
1304 | /* | |
1305 | * Do we still search for the first timer or are | |
1306 | * we looking up the cascade buckets ? | |
1307 | */ | |
1308 | if (found) { | |
1309 | /* Look at the cascade bucket(s)? */ | |
1310 | if (!index || slot < index) | |
1311 | break; | |
1312 | return expires; | |
1313 | } | |
1314 | slot = (slot + 1) & TVN_MASK; | |
1315 | } while (slot != index); | |
1316 | ||
1317 | if (index) | |
1318 | timer_jiffies += TVN_SIZE - index; | |
1319 | timer_jiffies >>= TVN_BITS; | |
1da177e4 | 1320 | } |
1cfd6849 TG |
1321 | return expires; |
1322 | } | |
69239749 | 1323 | |
1cfd6849 TG |
1324 | /* |
1325 | * Check, if the next hrtimer event is before the next timer wheel | |
1326 | * event: | |
1327 | */ | |
c1ad348b | 1328 | static u64 cmp_next_hrtimer_event(u64 basem, u64 expires) |
1cfd6849 | 1329 | { |
c1ad348b | 1330 | u64 nextevt = hrtimer_get_next_event(); |
0662b713 | 1331 | |
9501b6cf | 1332 | /* |
c1ad348b TG |
1333 | * If high resolution timers are enabled |
1334 | * hrtimer_get_next_event() returns KTIME_MAX. | |
9501b6cf | 1335 | */ |
c1ad348b TG |
1336 | if (expires <= nextevt) |
1337 | return expires; | |
eaad084b TG |
1338 | |
1339 | /* | |
c1ad348b TG |
1340 | * If the next timer is already expired, return the tick base |
1341 | * time so the tick is fired immediately. | |
eaad084b | 1342 | */ |
c1ad348b TG |
1343 | if (nextevt <= basem) |
1344 | return basem; | |
eaad084b | 1345 | |
9501b6cf | 1346 | /* |
c1ad348b TG |
1347 | * Round up to the next jiffie. High resolution timers are |
1348 | * off, so the hrtimers are expired in the tick and we need to | |
1349 | * make sure that this tick really expires the timer to avoid | |
1350 | * a ping pong of the nohz stop code. | |
1351 | * | |
1352 | * Use DIV_ROUND_UP_ULL to prevent gcc calling __divdi3 | |
9501b6cf | 1353 | */ |
c1ad348b | 1354 | return DIV_ROUND_UP_ULL(nextevt, TICK_NSEC) * TICK_NSEC; |
1da177e4 | 1355 | } |
1cfd6849 TG |
1356 | |
1357 | /** | |
c1ad348b TG |
1358 | * get_next_timer_interrupt - return the time (clock mono) of the next timer |
1359 | * @basej: base time jiffies | |
1360 | * @basem: base time clock monotonic | |
1361 | * | |
1362 | * Returns the tick aligned clock monotonic time of the next pending | |
1363 | * timer or KTIME_MAX if no timer is pending. | |
1cfd6849 | 1364 | */ |
c1ad348b | 1365 | u64 get_next_timer_interrupt(unsigned long basej, u64 basem) |
1cfd6849 | 1366 | { |
0eeda71b | 1367 | struct tvec_base *base = this_cpu_ptr(&tvec_bases); |
c1ad348b TG |
1368 | u64 expires = KTIME_MAX; |
1369 | unsigned long nextevt; | |
1cfd6849 | 1370 | |
dbd87b5a HC |
1371 | /* |
1372 | * Pretend that there is no timer pending if the cpu is offline. | |
1373 | * Possible pending timers will be migrated later to an active cpu. | |
1374 | */ | |
1375 | if (cpu_is_offline(smp_processor_id())) | |
e40468a5 TG |
1376 | return expires; |
1377 | ||
1cfd6849 | 1378 | spin_lock(&base->lock); |
e40468a5 TG |
1379 | if (base->active_timers) { |
1380 | if (time_before_eq(base->next_timer, base->timer_jiffies)) | |
1381 | base->next_timer = __next_timer_interrupt(base); | |
c1ad348b TG |
1382 | nextevt = base->next_timer; |
1383 | if (time_before_eq(nextevt, basej)) | |
1384 | expires = basem; | |
1385 | else | |
1386 | expires = basem + (nextevt - basej) * TICK_NSEC; | |
e40468a5 | 1387 | } |
1cfd6849 TG |
1388 | spin_unlock(&base->lock); |
1389 | ||
c1ad348b | 1390 | return cmp_next_hrtimer_event(basem, expires); |
1cfd6849 | 1391 | } |
1da177e4 LT |
1392 | #endif |
1393 | ||
1da177e4 | 1394 | /* |
5b4db0c2 | 1395 | * Called from the timer interrupt handler to charge one tick to the current |
1da177e4 LT |
1396 | * process. user_tick is 1 if the tick is user time, 0 for system. |
1397 | */ | |
1398 | void update_process_times(int user_tick) | |
1399 | { | |
1400 | struct task_struct *p = current; | |
1da177e4 LT |
1401 | |
1402 | /* Note: this timer irq context must be accounted for as well. */ | |
fa13a5a1 | 1403 | account_process_tick(p, user_tick); |
1da177e4 | 1404 | run_local_timers(); |
c3377c2d | 1405 | rcu_check_callbacks(user_tick); |
e360adbe PZ |
1406 | #ifdef CONFIG_IRQ_WORK |
1407 | if (in_irq()) | |
76a33061 | 1408 | irq_work_tick(); |
e360adbe | 1409 | #endif |
1da177e4 | 1410 | scheduler_tick(); |
6819457d | 1411 | run_posix_cpu_timers(p); |
1da177e4 LT |
1412 | } |
1413 | ||
1da177e4 LT |
1414 | /* |
1415 | * This function runs timers and the timer-tq in bottom half context. | |
1416 | */ | |
1417 | static void run_timer_softirq(struct softirq_action *h) | |
1418 | { | |
0eeda71b | 1419 | struct tvec_base *base = this_cpu_ptr(&tvec_bases); |
1da177e4 LT |
1420 | |
1421 | if (time_after_eq(jiffies, base->timer_jiffies)) | |
1422 | __run_timers(base); | |
1423 | } | |
1424 | ||
1425 | /* | |
1426 | * Called by the local, per-CPU timer interrupt on SMP. | |
1427 | */ | |
1428 | void run_local_timers(void) | |
1429 | { | |
d3d74453 | 1430 | hrtimer_run_queues(); |
1da177e4 LT |
1431 | raise_softirq(TIMER_SOFTIRQ); |
1432 | } | |
1433 | ||
1da177e4 LT |
1434 | #ifdef __ARCH_WANT_SYS_ALARM |
1435 | ||
1436 | /* | |
1437 | * For backwards compatibility? This can be done in libc so Alpha | |
1438 | * and all newer ports shouldn't need it. | |
1439 | */ | |
58fd3aa2 | 1440 | SYSCALL_DEFINE1(alarm, unsigned int, seconds) |
1da177e4 | 1441 | { |
c08b8a49 | 1442 | return alarm_setitimer(seconds); |
1da177e4 LT |
1443 | } |
1444 | ||
1445 | #endif | |
1446 | ||
1da177e4 LT |
1447 | static void process_timeout(unsigned long __data) |
1448 | { | |
36c8b586 | 1449 | wake_up_process((struct task_struct *)__data); |
1da177e4 LT |
1450 | } |
1451 | ||
1452 | /** | |
1453 | * schedule_timeout - sleep until timeout | |
1454 | * @timeout: timeout value in jiffies | |
1455 | * | |
1456 | * Make the current task sleep until @timeout jiffies have | |
1457 | * elapsed. The routine will return immediately unless | |
1458 | * the current task state has been set (see set_current_state()). | |
1459 | * | |
1460 | * You can set the task state as follows - | |
1461 | * | |
1462 | * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to | |
1463 | * pass before the routine returns. The routine will return 0 | |
1464 | * | |
1465 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
1466 | * delivered to the current task. In this case the remaining time | |
1467 | * in jiffies will be returned, or 0 if the timer expired in time | |
1468 | * | |
1469 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1470 | * routine returns. | |
1471 | * | |
1472 | * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule | |
1473 | * the CPU away without a bound on the timeout. In this case the return | |
1474 | * value will be %MAX_SCHEDULE_TIMEOUT. | |
1475 | * | |
1476 | * In all cases the return value is guaranteed to be non-negative. | |
1477 | */ | |
7ad5b3a5 | 1478 | signed long __sched schedule_timeout(signed long timeout) |
1da177e4 LT |
1479 | { |
1480 | struct timer_list timer; | |
1481 | unsigned long expire; | |
1482 | ||
1483 | switch (timeout) | |
1484 | { | |
1485 | case MAX_SCHEDULE_TIMEOUT: | |
1486 | /* | |
1487 | * These two special cases are useful to be comfortable | |
1488 | * in the caller. Nothing more. We could take | |
1489 | * MAX_SCHEDULE_TIMEOUT from one of the negative value | |
1490 | * but I' d like to return a valid offset (>=0) to allow | |
1491 | * the caller to do everything it want with the retval. | |
1492 | */ | |
1493 | schedule(); | |
1494 | goto out; | |
1495 | default: | |
1496 | /* | |
1497 | * Another bit of PARANOID. Note that the retval will be | |
1498 | * 0 since no piece of kernel is supposed to do a check | |
1499 | * for a negative retval of schedule_timeout() (since it | |
1500 | * should never happens anyway). You just have the printk() | |
1501 | * that will tell you if something is gone wrong and where. | |
1502 | */ | |
5b149bcc | 1503 | if (timeout < 0) { |
1da177e4 | 1504 | printk(KERN_ERR "schedule_timeout: wrong timeout " |
5b149bcc AM |
1505 | "value %lx\n", timeout); |
1506 | dump_stack(); | |
1da177e4 LT |
1507 | current->state = TASK_RUNNING; |
1508 | goto out; | |
1509 | } | |
1510 | } | |
1511 | ||
1512 | expire = timeout + jiffies; | |
1513 | ||
c6f3a97f | 1514 | setup_timer_on_stack(&timer, process_timeout, (unsigned long)current); |
597d0275 | 1515 | __mod_timer(&timer, expire, false, TIMER_NOT_PINNED); |
1da177e4 LT |
1516 | schedule(); |
1517 | del_singleshot_timer_sync(&timer); | |
1518 | ||
c6f3a97f TG |
1519 | /* Remove the timer from the object tracker */ |
1520 | destroy_timer_on_stack(&timer); | |
1521 | ||
1da177e4 LT |
1522 | timeout = expire - jiffies; |
1523 | ||
1524 | out: | |
1525 | return timeout < 0 ? 0 : timeout; | |
1526 | } | |
1da177e4 LT |
1527 | EXPORT_SYMBOL(schedule_timeout); |
1528 | ||
8a1c1757 AM |
1529 | /* |
1530 | * We can use __set_current_state() here because schedule_timeout() calls | |
1531 | * schedule() unconditionally. | |
1532 | */ | |
64ed93a2 NA |
1533 | signed long __sched schedule_timeout_interruptible(signed long timeout) |
1534 | { | |
a5a0d52c AM |
1535 | __set_current_state(TASK_INTERRUPTIBLE); |
1536 | return schedule_timeout(timeout); | |
64ed93a2 NA |
1537 | } |
1538 | EXPORT_SYMBOL(schedule_timeout_interruptible); | |
1539 | ||
294d5cc2 MW |
1540 | signed long __sched schedule_timeout_killable(signed long timeout) |
1541 | { | |
1542 | __set_current_state(TASK_KILLABLE); | |
1543 | return schedule_timeout(timeout); | |
1544 | } | |
1545 | EXPORT_SYMBOL(schedule_timeout_killable); | |
1546 | ||
64ed93a2 NA |
1547 | signed long __sched schedule_timeout_uninterruptible(signed long timeout) |
1548 | { | |
a5a0d52c AM |
1549 | __set_current_state(TASK_UNINTERRUPTIBLE); |
1550 | return schedule_timeout(timeout); | |
64ed93a2 NA |
1551 | } |
1552 | EXPORT_SYMBOL(schedule_timeout_uninterruptible); | |
1553 | ||
69b27baf AM |
1554 | /* |
1555 | * Like schedule_timeout_uninterruptible(), except this task will not contribute | |
1556 | * to load average. | |
1557 | */ | |
1558 | signed long __sched schedule_timeout_idle(signed long timeout) | |
1559 | { | |
1560 | __set_current_state(TASK_IDLE); | |
1561 | return schedule_timeout(timeout); | |
1562 | } | |
1563 | EXPORT_SYMBOL(schedule_timeout_idle); | |
1564 | ||
1da177e4 | 1565 | #ifdef CONFIG_HOTPLUG_CPU |
1dabbcec | 1566 | static void migrate_timer_list(struct tvec_base *new_base, struct hlist_head *head) |
1da177e4 LT |
1567 | { |
1568 | struct timer_list *timer; | |
0eeda71b | 1569 | int cpu = new_base->cpu; |
1da177e4 | 1570 | |
1dabbcec TG |
1571 | while (!hlist_empty(head)) { |
1572 | timer = hlist_entry(head->first, struct timer_list, entry); | |
99d5f3aa | 1573 | /* We ignore the accounting on the dying cpu */ |
ec44bc7a | 1574 | detach_timer(timer, false); |
0eeda71b | 1575 | timer->flags = (timer->flags & ~TIMER_BASEMASK) | cpu; |
1da177e4 | 1576 | internal_add_timer(new_base, timer); |
1da177e4 | 1577 | } |
1da177e4 LT |
1578 | } |
1579 | ||
0db0628d | 1580 | static void migrate_timers(int cpu) |
1da177e4 | 1581 | { |
a6fa8e5a PM |
1582 | struct tvec_base *old_base; |
1583 | struct tvec_base *new_base; | |
1da177e4 LT |
1584 | int i; |
1585 | ||
1586 | BUG_ON(cpu_online(cpu)); | |
0eeda71b | 1587 | old_base = per_cpu_ptr(&tvec_bases, cpu); |
24bfcb10 | 1588 | new_base = get_cpu_ptr(&tvec_bases); |
d82f0b0f ON |
1589 | /* |
1590 | * The caller is globally serialized and nobody else | |
1591 | * takes two locks at once, deadlock is not possible. | |
1592 | */ | |
1593 | spin_lock_irq(&new_base->lock); | |
0d180406 | 1594 | spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING); |
3691c519 ON |
1595 | |
1596 | BUG_ON(old_base->running_timer); | |
1da177e4 | 1597 | |
1da177e4 | 1598 | for (i = 0; i < TVR_SIZE; i++) |
55c888d6 ON |
1599 | migrate_timer_list(new_base, old_base->tv1.vec + i); |
1600 | for (i = 0; i < TVN_SIZE; i++) { | |
1601 | migrate_timer_list(new_base, old_base->tv2.vec + i); | |
1602 | migrate_timer_list(new_base, old_base->tv3.vec + i); | |
1603 | migrate_timer_list(new_base, old_base->tv4.vec + i); | |
1604 | migrate_timer_list(new_base, old_base->tv5.vec + i); | |
1605 | } | |
1606 | ||
8def9060 VK |
1607 | old_base->active_timers = 0; |
1608 | old_base->all_timers = 0; | |
1609 | ||
0d180406 | 1610 | spin_unlock(&old_base->lock); |
d82f0b0f | 1611 | spin_unlock_irq(&new_base->lock); |
24bfcb10 | 1612 | put_cpu_ptr(&tvec_bases); |
1da177e4 | 1613 | } |
1da177e4 | 1614 | |
0db0628d | 1615 | static int timer_cpu_notify(struct notifier_block *self, |
1da177e4 LT |
1616 | unsigned long action, void *hcpu) |
1617 | { | |
8def9060 | 1618 | switch (action) { |
1da177e4 | 1619 | case CPU_DEAD: |
8bb78442 | 1620 | case CPU_DEAD_FROZEN: |
8def9060 | 1621 | migrate_timers((long)hcpu); |
1da177e4 | 1622 | break; |
1da177e4 LT |
1623 | default: |
1624 | break; | |
1625 | } | |
3650b57f | 1626 | |
1da177e4 LT |
1627 | return NOTIFY_OK; |
1628 | } | |
1629 | ||
3650b57f PZ |
1630 | static inline void timer_register_cpu_notifier(void) |
1631 | { | |
1632 | cpu_notifier(timer_cpu_notify, 0); | |
1633 | } | |
1634 | #else | |
1635 | static inline void timer_register_cpu_notifier(void) { } | |
1636 | #endif /* CONFIG_HOTPLUG_CPU */ | |
1da177e4 | 1637 | |
0eeda71b | 1638 | static void __init init_timer_cpu(int cpu) |
8def9060 | 1639 | { |
0eeda71b | 1640 | struct tvec_base *base = per_cpu_ptr(&tvec_bases, cpu); |
3650b57f | 1641 | |
8def9060 | 1642 | base->cpu = cpu; |
8def9060 VK |
1643 | spin_lock_init(&base->lock); |
1644 | ||
8def9060 VK |
1645 | base->timer_jiffies = jiffies; |
1646 | base->next_timer = base->timer_jiffies; | |
1647 | } | |
1648 | ||
1649 | static void __init init_timer_cpus(void) | |
1da177e4 | 1650 | { |
8def9060 VK |
1651 | int cpu; |
1652 | ||
0eeda71b TG |
1653 | for_each_possible_cpu(cpu) |
1654 | init_timer_cpu(cpu); | |
8def9060 | 1655 | } |
e52b1db3 | 1656 | |
8def9060 VK |
1657 | void __init init_timers(void) |
1658 | { | |
8def9060 | 1659 | init_timer_cpus(); |
c24a4a36 | 1660 | init_timer_stats(); |
3650b57f | 1661 | timer_register_cpu_notifier(); |
962cf36c | 1662 | open_softirq(TIMER_SOFTIRQ, run_timer_softirq); |
1da177e4 LT |
1663 | } |
1664 | ||
1da177e4 LT |
1665 | /** |
1666 | * msleep - sleep safely even with waitqueue interruptions | |
1667 | * @msecs: Time in milliseconds to sleep for | |
1668 | */ | |
1669 | void msleep(unsigned int msecs) | |
1670 | { | |
1671 | unsigned long timeout = msecs_to_jiffies(msecs) + 1; | |
1672 | ||
75bcc8c5 NA |
1673 | while (timeout) |
1674 | timeout = schedule_timeout_uninterruptible(timeout); | |
1da177e4 LT |
1675 | } |
1676 | ||
1677 | EXPORT_SYMBOL(msleep); | |
1678 | ||
1679 | /** | |
96ec3efd | 1680 | * msleep_interruptible - sleep waiting for signals |
1da177e4 LT |
1681 | * @msecs: Time in milliseconds to sleep for |
1682 | */ | |
1683 | unsigned long msleep_interruptible(unsigned int msecs) | |
1684 | { | |
1685 | unsigned long timeout = msecs_to_jiffies(msecs) + 1; | |
1686 | ||
75bcc8c5 NA |
1687 | while (timeout && !signal_pending(current)) |
1688 | timeout = schedule_timeout_interruptible(timeout); | |
1da177e4 LT |
1689 | return jiffies_to_msecs(timeout); |
1690 | } | |
1691 | ||
1692 | EXPORT_SYMBOL(msleep_interruptible); | |
5e7f5a17 | 1693 | |
6deba083 | 1694 | static void __sched do_usleep_range(unsigned long min, unsigned long max) |
5e7f5a17 PP |
1695 | { |
1696 | ktime_t kmin; | |
da8b44d5 | 1697 | u64 delta; |
5e7f5a17 PP |
1698 | |
1699 | kmin = ktime_set(0, min * NSEC_PER_USEC); | |
da8b44d5 | 1700 | delta = (u64)(max - min) * NSEC_PER_USEC; |
6deba083 | 1701 | schedule_hrtimeout_range(&kmin, delta, HRTIMER_MODE_REL); |
5e7f5a17 PP |
1702 | } |
1703 | ||
1704 | /** | |
1705 | * usleep_range - Drop in replacement for udelay where wakeup is flexible | |
1706 | * @min: Minimum time in usecs to sleep | |
1707 | * @max: Maximum time in usecs to sleep | |
1708 | */ | |
2ad5d327 | 1709 | void __sched usleep_range(unsigned long min, unsigned long max) |
5e7f5a17 PP |
1710 | { |
1711 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
1712 | do_usleep_range(min, max); | |
1713 | } | |
1714 | EXPORT_SYMBOL(usleep_range); |