Merge remote-tracking branch 'kspp/for-next/kspp'
[deliverable/linux.git] / kernel / time / timer.c
1 /*
2 * linux/kernel/timer.c
3 *
4 * Kernel internal timers
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/export.h>
24 #include <linux/interrupt.h>
25 #include <linux/percpu.h>
26 #include <linux/init.h>
27 #include <linux/mm.h>
28 #include <linux/swap.h>
29 #include <linux/pid_namespace.h>
30 #include <linux/notifier.h>
31 #include <linux/thread_info.h>
32 #include <linux/time.h>
33 #include <linux/jiffies.h>
34 #include <linux/posix-timers.h>
35 #include <linux/cpu.h>
36 #include <linux/syscalls.h>
37 #include <linux/delay.h>
38 #include <linux/tick.h>
39 #include <linux/kallsyms.h>
40 #include <linux/irq_work.h>
41 #include <linux/sched.h>
42 #include <linux/sched/sysctl.h>
43 #include <linux/slab.h>
44 #include <linux/compat.h>
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
52 #include "tick-internal.h"
53
54 #define CREATE_TRACE_POINTS
55 #include <trace/events/timer.h>
56
57 __visible u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
58
59 EXPORT_SYMBOL(jiffies_64);
60
61 /*
62 * The timer wheel has LVL_DEPTH array levels. Each level provides an array of
63 * LVL_SIZE buckets. Each level is driven by its own clock and therefor each
64 * level has a different granularity.
65 *
66 * The level granularity is: LVL_CLK_DIV ^ lvl
67 * The level clock frequency is: HZ / (LVL_CLK_DIV ^ level)
68 *
69 * The array level of a newly armed timer depends on the relative expiry
70 * time. The farther the expiry time is away the higher the array level and
71 * therefor the granularity becomes.
72 *
73 * Contrary to the original timer wheel implementation, which aims for 'exact'
74 * expiry of the timers, this implementation removes the need for recascading
75 * the timers into the lower array levels. The previous 'classic' timer wheel
76 * implementation of the kernel already violated the 'exact' expiry by adding
77 * slack to the expiry time to provide batched expiration. The granularity
78 * levels provide implicit batching.
79 *
80 * This is an optimization of the original timer wheel implementation for the
81 * majority of the timer wheel use cases: timeouts. The vast majority of
82 * timeout timers (networking, disk I/O ...) are canceled before expiry. If
83 * the timeout expires it indicates that normal operation is disturbed, so it
84 * does not matter much whether the timeout comes with a slight delay.
85 *
86 * The only exception to this are networking timers with a small expiry
87 * time. They rely on the granularity. Those fit into the first wheel level,
88 * which has HZ granularity.
89 *
90 * We don't have cascading anymore. timers with a expiry time above the
91 * capacity of the last wheel level are force expired at the maximum timeout
92 * value of the last wheel level. From data sampling we know that the maximum
93 * value observed is 5 days (network connection tracking), so this should not
94 * be an issue.
95 *
96 * The currently chosen array constants values are a good compromise between
97 * array size and granularity.
98 *
99 * This results in the following granularity and range levels:
100 *
101 * HZ 1000 steps
102 * Level Offset Granularity Range
103 * 0 0 1 ms 0 ms - 63 ms
104 * 1 64 8 ms 64 ms - 511 ms
105 * 2 128 64 ms 512 ms - 4095 ms (512ms - ~4s)
106 * 3 192 512 ms 4096 ms - 32767 ms (~4s - ~32s)
107 * 4 256 4096 ms (~4s) 32768 ms - 262143 ms (~32s - ~4m)
108 * 5 320 32768 ms (~32s) 262144 ms - 2097151 ms (~4m - ~34m)
109 * 6 384 262144 ms (~4m) 2097152 ms - 16777215 ms (~34m - ~4h)
110 * 7 448 2097152 ms (~34m) 16777216 ms - 134217727 ms (~4h - ~1d)
111 * 8 512 16777216 ms (~4h) 134217728 ms - 1073741822 ms (~1d - ~12d)
112 *
113 * HZ 300
114 * Level Offset Granularity Range
115 * 0 0 3 ms 0 ms - 210 ms
116 * 1 64 26 ms 213 ms - 1703 ms (213ms - ~1s)
117 * 2 128 213 ms 1706 ms - 13650 ms (~1s - ~13s)
118 * 3 192 1706 ms (~1s) 13653 ms - 109223 ms (~13s - ~1m)
119 * 4 256 13653 ms (~13s) 109226 ms - 873810 ms (~1m - ~14m)
120 * 5 320 109226 ms (~1m) 873813 ms - 6990503 ms (~14m - ~1h)
121 * 6 384 873813 ms (~14m) 6990506 ms - 55924050 ms (~1h - ~15h)
122 * 7 448 6990506 ms (~1h) 55924053 ms - 447392423 ms (~15h - ~5d)
123 * 8 512 55924053 ms (~15h) 447392426 ms - 3579139406 ms (~5d - ~41d)
124 *
125 * HZ 250
126 * Level Offset Granularity Range
127 * 0 0 4 ms 0 ms - 255 ms
128 * 1 64 32 ms 256 ms - 2047 ms (256ms - ~2s)
129 * 2 128 256 ms 2048 ms - 16383 ms (~2s - ~16s)
130 * 3 192 2048 ms (~2s) 16384 ms - 131071 ms (~16s - ~2m)
131 * 4 256 16384 ms (~16s) 131072 ms - 1048575 ms (~2m - ~17m)
132 * 5 320 131072 ms (~2m) 1048576 ms - 8388607 ms (~17m - ~2h)
133 * 6 384 1048576 ms (~17m) 8388608 ms - 67108863 ms (~2h - ~18h)
134 * 7 448 8388608 ms (~2h) 67108864 ms - 536870911 ms (~18h - ~6d)
135 * 8 512 67108864 ms (~18h) 536870912 ms - 4294967288 ms (~6d - ~49d)
136 *
137 * HZ 100
138 * Level Offset Granularity Range
139 * 0 0 10 ms 0 ms - 630 ms
140 * 1 64 80 ms 640 ms - 5110 ms (640ms - ~5s)
141 * 2 128 640 ms 5120 ms - 40950 ms (~5s - ~40s)
142 * 3 192 5120 ms (~5s) 40960 ms - 327670 ms (~40s - ~5m)
143 * 4 256 40960 ms (~40s) 327680 ms - 2621430 ms (~5m - ~43m)
144 * 5 320 327680 ms (~5m) 2621440 ms - 20971510 ms (~43m - ~5h)
145 * 6 384 2621440 ms (~43m) 20971520 ms - 167772150 ms (~5h - ~1d)
146 * 7 448 20971520 ms (~5h) 167772160 ms - 1342177270 ms (~1d - ~15d)
147 */
148
149 /* Clock divisor for the next level */
150 #define LVL_CLK_SHIFT 3
151 #define LVL_CLK_DIV (1UL << LVL_CLK_SHIFT)
152 #define LVL_CLK_MASK (LVL_CLK_DIV - 1)
153 #define LVL_SHIFT(n) ((n) * LVL_CLK_SHIFT)
154 #define LVL_GRAN(n) (1UL << LVL_SHIFT(n))
155
156 /*
157 * The time start value for each level to select the bucket at enqueue
158 * time.
159 */
160 #define LVL_START(n) ((LVL_SIZE - 1) << (((n) - 1) * LVL_CLK_SHIFT))
161
162 /* Size of each clock level */
163 #define LVL_BITS 6
164 #define LVL_SIZE (1UL << LVL_BITS)
165 #define LVL_MASK (LVL_SIZE - 1)
166 #define LVL_OFFS(n) ((n) * LVL_SIZE)
167
168 /* Level depth */
169 #if HZ > 100
170 # define LVL_DEPTH 9
171 # else
172 # define LVL_DEPTH 8
173 #endif
174
175 /* The cutoff (max. capacity of the wheel) */
176 #define WHEEL_TIMEOUT_CUTOFF (LVL_START(LVL_DEPTH))
177 #define WHEEL_TIMEOUT_MAX (WHEEL_TIMEOUT_CUTOFF - LVL_GRAN(LVL_DEPTH - 1))
178
179 /*
180 * The resulting wheel size. If NOHZ is configured we allocate two
181 * wheels so we have a separate storage for the deferrable timers.
182 */
183 #define WHEEL_SIZE (LVL_SIZE * LVL_DEPTH)
184
185 #ifdef CONFIG_NO_HZ_COMMON
186 # define NR_BASES 2
187 # define BASE_STD 0
188 # define BASE_DEF 1
189 #else
190 # define NR_BASES 1
191 # define BASE_STD 0
192 # define BASE_DEF 0
193 #endif
194
195 struct timer_base {
196 spinlock_t lock;
197 struct timer_list *running_timer;
198 unsigned long clk;
199 unsigned long next_expiry;
200 unsigned int cpu;
201 bool migration_enabled;
202 bool nohz_active;
203 bool is_idle;
204 DECLARE_BITMAP(pending_map, WHEEL_SIZE);
205 struct hlist_head vectors[WHEEL_SIZE];
206 } ____cacheline_aligned;
207
208 static DEFINE_PER_CPU(struct timer_base, timer_bases[NR_BASES]);
209
210 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
211 unsigned int sysctl_timer_migration = 1;
212
213 void timers_update_migration(bool update_nohz)
214 {
215 bool on = sysctl_timer_migration && tick_nohz_active;
216 unsigned int cpu;
217
218 /* Avoid the loop, if nothing to update */
219 if (this_cpu_read(timer_bases[BASE_STD].migration_enabled) == on)
220 return;
221
222 for_each_possible_cpu(cpu) {
223 per_cpu(timer_bases[BASE_STD].migration_enabled, cpu) = on;
224 per_cpu(timer_bases[BASE_DEF].migration_enabled, cpu) = on;
225 per_cpu(hrtimer_bases.migration_enabled, cpu) = on;
226 if (!update_nohz)
227 continue;
228 per_cpu(timer_bases[BASE_STD].nohz_active, cpu) = true;
229 per_cpu(timer_bases[BASE_DEF].nohz_active, cpu) = true;
230 per_cpu(hrtimer_bases.nohz_active, cpu) = true;
231 }
232 }
233
234 int timer_migration_handler(struct ctl_table *table, int write,
235 void __user *buffer, size_t *lenp,
236 loff_t *ppos)
237 {
238 static DEFINE_MUTEX(mutex);
239 int ret;
240
241 mutex_lock(&mutex);
242 ret = proc_dointvec(table, write, buffer, lenp, ppos);
243 if (!ret && write)
244 timers_update_migration(false);
245 mutex_unlock(&mutex);
246 return ret;
247 }
248 #endif
249
250 static unsigned long round_jiffies_common(unsigned long j, int cpu,
251 bool force_up)
252 {
253 int rem;
254 unsigned long original = j;
255
256 /*
257 * We don't want all cpus firing their timers at once hitting the
258 * same lock or cachelines, so we skew each extra cpu with an extra
259 * 3 jiffies. This 3 jiffies came originally from the mm/ code which
260 * already did this.
261 * The skew is done by adding 3*cpunr, then round, then subtract this
262 * extra offset again.
263 */
264 j += cpu * 3;
265
266 rem = j % HZ;
267
268 /*
269 * If the target jiffie is just after a whole second (which can happen
270 * due to delays of the timer irq, long irq off times etc etc) then
271 * we should round down to the whole second, not up. Use 1/4th second
272 * as cutoff for this rounding as an extreme upper bound for this.
273 * But never round down if @force_up is set.
274 */
275 if (rem < HZ/4 && !force_up) /* round down */
276 j = j - rem;
277 else /* round up */
278 j = j - rem + HZ;
279
280 /* now that we have rounded, subtract the extra skew again */
281 j -= cpu * 3;
282
283 /*
284 * Make sure j is still in the future. Otherwise return the
285 * unmodified value.
286 */
287 return time_is_after_jiffies(j) ? j : original;
288 }
289
290 /**
291 * __round_jiffies - function to round jiffies to a full second
292 * @j: the time in (absolute) jiffies that should be rounded
293 * @cpu: the processor number on which the timeout will happen
294 *
295 * __round_jiffies() rounds an absolute time in the future (in jiffies)
296 * up or down to (approximately) full seconds. This is useful for timers
297 * for which the exact time they fire does not matter too much, as long as
298 * they fire approximately every X seconds.
299 *
300 * By rounding these timers to whole seconds, all such timers will fire
301 * at the same time, rather than at various times spread out. The goal
302 * of this is to have the CPU wake up less, which saves power.
303 *
304 * The exact rounding is skewed for each processor to avoid all
305 * processors firing at the exact same time, which could lead
306 * to lock contention or spurious cache line bouncing.
307 *
308 * The return value is the rounded version of the @j parameter.
309 */
310 unsigned long __round_jiffies(unsigned long j, int cpu)
311 {
312 return round_jiffies_common(j, cpu, false);
313 }
314 EXPORT_SYMBOL_GPL(__round_jiffies);
315
316 /**
317 * __round_jiffies_relative - function to round jiffies to a full second
318 * @j: the time in (relative) jiffies that should be rounded
319 * @cpu: the processor number on which the timeout will happen
320 *
321 * __round_jiffies_relative() rounds a time delta in the future (in jiffies)
322 * up or down to (approximately) full seconds. This is useful for timers
323 * for which the exact time they fire does not matter too much, as long as
324 * they fire approximately every X seconds.
325 *
326 * By rounding these timers to whole seconds, all such timers will fire
327 * at the same time, rather than at various times spread out. The goal
328 * of this is to have the CPU wake up less, which saves power.
329 *
330 * The exact rounding is skewed for each processor to avoid all
331 * processors firing at the exact same time, which could lead
332 * to lock contention or spurious cache line bouncing.
333 *
334 * The return value is the rounded version of the @j parameter.
335 */
336 unsigned long __round_jiffies_relative(unsigned long j, int cpu)
337 {
338 unsigned long j0 = jiffies;
339
340 /* Use j0 because jiffies might change while we run */
341 return round_jiffies_common(j + j0, cpu, false) - j0;
342 }
343 EXPORT_SYMBOL_GPL(__round_jiffies_relative);
344
345 /**
346 * round_jiffies - function to round jiffies to a full second
347 * @j: the time in (absolute) jiffies that should be rounded
348 *
349 * round_jiffies() rounds an absolute time in the future (in jiffies)
350 * up or down to (approximately) full seconds. This is useful for timers
351 * for which the exact time they fire does not matter too much, as long as
352 * they fire approximately every X seconds.
353 *
354 * By rounding these timers to whole seconds, all such timers will fire
355 * at the same time, rather than at various times spread out. The goal
356 * of this is to have the CPU wake up less, which saves power.
357 *
358 * The return value is the rounded version of the @j parameter.
359 */
360 unsigned long round_jiffies(unsigned long j)
361 {
362 return round_jiffies_common(j, raw_smp_processor_id(), false);
363 }
364 EXPORT_SYMBOL_GPL(round_jiffies);
365
366 /**
367 * round_jiffies_relative - function to round jiffies to a full second
368 * @j: the time in (relative) jiffies that should be rounded
369 *
370 * round_jiffies_relative() rounds a time delta in the future (in jiffies)
371 * up or down to (approximately) full seconds. This is useful for timers
372 * for which the exact time they fire does not matter too much, as long as
373 * they fire approximately every X seconds.
374 *
375 * By rounding these timers to whole seconds, all such timers will fire
376 * at the same time, rather than at various times spread out. The goal
377 * of this is to have the CPU wake up less, which saves power.
378 *
379 * The return value is the rounded version of the @j parameter.
380 */
381 unsigned long round_jiffies_relative(unsigned long j)
382 {
383 return __round_jiffies_relative(j, raw_smp_processor_id());
384 }
385 EXPORT_SYMBOL_GPL(round_jiffies_relative);
386
387 /**
388 * __round_jiffies_up - function to round jiffies up to a full second
389 * @j: the time in (absolute) jiffies that should be rounded
390 * @cpu: the processor number on which the timeout will happen
391 *
392 * This is the same as __round_jiffies() except that it will never
393 * round down. This is useful for timeouts for which the exact time
394 * of firing does not matter too much, as long as they don't fire too
395 * early.
396 */
397 unsigned long __round_jiffies_up(unsigned long j, int cpu)
398 {
399 return round_jiffies_common(j, cpu, true);
400 }
401 EXPORT_SYMBOL_GPL(__round_jiffies_up);
402
403 /**
404 * __round_jiffies_up_relative - function to round jiffies up to a full second
405 * @j: the time in (relative) jiffies that should be rounded
406 * @cpu: the processor number on which the timeout will happen
407 *
408 * This is the same as __round_jiffies_relative() except that it will never
409 * round down. This is useful for timeouts for which the exact time
410 * of firing does not matter too much, as long as they don't fire too
411 * early.
412 */
413 unsigned long __round_jiffies_up_relative(unsigned long j, int cpu)
414 {
415 unsigned long j0 = jiffies;
416
417 /* Use j0 because jiffies might change while we run */
418 return round_jiffies_common(j + j0, cpu, true) - j0;
419 }
420 EXPORT_SYMBOL_GPL(__round_jiffies_up_relative);
421
422 /**
423 * round_jiffies_up - function to round jiffies up to a full second
424 * @j: the time in (absolute) jiffies that should be rounded
425 *
426 * This is the same as round_jiffies() except that it will never
427 * round down. This is useful for timeouts for which the exact time
428 * of firing does not matter too much, as long as they don't fire too
429 * early.
430 */
431 unsigned long round_jiffies_up(unsigned long j)
432 {
433 return round_jiffies_common(j, raw_smp_processor_id(), true);
434 }
435 EXPORT_SYMBOL_GPL(round_jiffies_up);
436
437 /**
438 * round_jiffies_up_relative - function to round jiffies up to a full second
439 * @j: the time in (relative) jiffies that should be rounded
440 *
441 * This is the same as round_jiffies_relative() except that it will never
442 * round down. This is useful for timeouts for which the exact time
443 * of firing does not matter too much, as long as they don't fire too
444 * early.
445 */
446 unsigned long round_jiffies_up_relative(unsigned long j)
447 {
448 return __round_jiffies_up_relative(j, raw_smp_processor_id());
449 }
450 EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
451
452
453 static inline unsigned int timer_get_idx(struct timer_list *timer)
454 {
455 return (timer->flags & TIMER_ARRAYMASK) >> TIMER_ARRAYSHIFT;
456 }
457
458 static inline void timer_set_idx(struct timer_list *timer, unsigned int idx)
459 {
460 timer->flags = (timer->flags & ~TIMER_ARRAYMASK) |
461 idx << TIMER_ARRAYSHIFT;
462 }
463
464 /*
465 * Helper function to calculate the array index for a given expiry
466 * time.
467 */
468 static inline unsigned calc_index(unsigned expires, unsigned lvl)
469 {
470 expires = (expires + LVL_GRAN(lvl)) >> LVL_SHIFT(lvl);
471 return LVL_OFFS(lvl) + (expires & LVL_MASK);
472 }
473
474 static int calc_wheel_index(unsigned long expires, unsigned long clk)
475 {
476 unsigned long delta = expires - clk;
477 unsigned int idx;
478
479 if (delta < LVL_START(1)) {
480 idx = calc_index(expires, 0);
481 } else if (delta < LVL_START(2)) {
482 idx = calc_index(expires, 1);
483 } else if (delta < LVL_START(3)) {
484 idx = calc_index(expires, 2);
485 } else if (delta < LVL_START(4)) {
486 idx = calc_index(expires, 3);
487 } else if (delta < LVL_START(5)) {
488 idx = calc_index(expires, 4);
489 } else if (delta < LVL_START(6)) {
490 idx = calc_index(expires, 5);
491 } else if (delta < LVL_START(7)) {
492 idx = calc_index(expires, 6);
493 } else if (LVL_DEPTH > 8 && delta < LVL_START(8)) {
494 idx = calc_index(expires, 7);
495 } else if ((long) delta < 0) {
496 idx = clk & LVL_MASK;
497 } else {
498 /*
499 * Force expire obscene large timeouts to expire at the
500 * capacity limit of the wheel.
501 */
502 if (expires >= WHEEL_TIMEOUT_CUTOFF)
503 expires = WHEEL_TIMEOUT_MAX;
504
505 idx = calc_index(expires, LVL_DEPTH - 1);
506 }
507 return idx;
508 }
509
510 /*
511 * Enqueue the timer into the hash bucket, mark it pending in
512 * the bitmap and store the index in the timer flags.
513 */
514 static void enqueue_timer(struct timer_base *base, struct timer_list *timer,
515 unsigned int idx)
516 {
517 hlist_add_head(&timer->entry, base->vectors + idx);
518 __set_bit(idx, base->pending_map);
519 timer_set_idx(timer, idx);
520 }
521
522 static void
523 __internal_add_timer(struct timer_base *base, struct timer_list *timer)
524 {
525 unsigned int idx;
526
527 idx = calc_wheel_index(timer->expires, base->clk);
528 enqueue_timer(base, timer, idx);
529 }
530
531 static void
532 trigger_dyntick_cpu(struct timer_base *base, struct timer_list *timer)
533 {
534 if (!IS_ENABLED(CONFIG_NO_HZ_COMMON) || !base->nohz_active)
535 return;
536
537 /*
538 * TODO: This wants some optimizing similar to the code below, but we
539 * will do that when we switch from push to pull for deferrable timers.
540 */
541 if (timer->flags & TIMER_DEFERRABLE) {
542 if (tick_nohz_full_cpu(base->cpu))
543 wake_up_nohz_cpu(base->cpu);
544 return;
545 }
546
547 /*
548 * We might have to IPI the remote CPU if the base is idle and the
549 * timer is not deferrable. If the other CPU is on the way to idle
550 * then it can't set base->is_idle as we hold the base lock:
551 */
552 if (!base->is_idle)
553 return;
554
555 /* Check whether this is the new first expiring timer: */
556 if (time_after_eq(timer->expires, base->next_expiry))
557 return;
558
559 /*
560 * Set the next expiry time and kick the CPU so it can reevaluate the
561 * wheel:
562 */
563 base->next_expiry = timer->expires;
564 wake_up_nohz_cpu(base->cpu);
565 }
566
567 static void
568 internal_add_timer(struct timer_base *base, struct timer_list *timer)
569 {
570 __internal_add_timer(base, timer);
571 trigger_dyntick_cpu(base, timer);
572 }
573
574 #ifdef CONFIG_TIMER_STATS
575 void __timer_stats_timer_set_start_info(struct timer_list *timer, void *addr)
576 {
577 if (timer->start_site)
578 return;
579
580 timer->start_site = addr;
581 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
582 timer->start_pid = current->pid;
583 }
584
585 static void timer_stats_account_timer(struct timer_list *timer)
586 {
587 void *site;
588
589 /*
590 * start_site can be concurrently reset by
591 * timer_stats_timer_clear_start_info()
592 */
593 site = READ_ONCE(timer->start_site);
594 if (likely(!site))
595 return;
596
597 timer_stats_update_stats(timer, timer->start_pid, site,
598 timer->function, timer->start_comm,
599 timer->flags);
600 }
601
602 #else
603 static void timer_stats_account_timer(struct timer_list *timer) {}
604 #endif
605
606 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
607
608 static struct debug_obj_descr timer_debug_descr;
609
610 static void *timer_debug_hint(void *addr)
611 {
612 return ((struct timer_list *) addr)->function;
613 }
614
615 static bool timer_is_static_object(void *addr)
616 {
617 struct timer_list *timer = addr;
618
619 return (timer->entry.pprev == NULL &&
620 timer->entry.next == TIMER_ENTRY_STATIC);
621 }
622
623 /*
624 * fixup_init is called when:
625 * - an active object is initialized
626 */
627 static bool timer_fixup_init(void *addr, enum debug_obj_state state)
628 {
629 struct timer_list *timer = addr;
630
631 switch (state) {
632 case ODEBUG_STATE_ACTIVE:
633 del_timer_sync(timer);
634 debug_object_init(timer, &timer_debug_descr);
635 return true;
636 default:
637 return false;
638 }
639 }
640
641 /* Stub timer callback for improperly used timers. */
642 static void stub_timer(unsigned long data)
643 {
644 WARN_ON(1);
645 }
646
647 /*
648 * fixup_activate is called when:
649 * - an active object is activated
650 * - an unknown non-static object is activated
651 */
652 static bool timer_fixup_activate(void *addr, enum debug_obj_state state)
653 {
654 struct timer_list *timer = addr;
655
656 switch (state) {
657 case ODEBUG_STATE_NOTAVAILABLE:
658 setup_timer(timer, stub_timer, 0);
659 return true;
660
661 case ODEBUG_STATE_ACTIVE:
662 WARN_ON(1);
663
664 default:
665 return false;
666 }
667 }
668
669 /*
670 * fixup_free is called when:
671 * - an active object is freed
672 */
673 static bool timer_fixup_free(void *addr, enum debug_obj_state state)
674 {
675 struct timer_list *timer = addr;
676
677 switch (state) {
678 case ODEBUG_STATE_ACTIVE:
679 del_timer_sync(timer);
680 debug_object_free(timer, &timer_debug_descr);
681 return true;
682 default:
683 return false;
684 }
685 }
686
687 /*
688 * fixup_assert_init is called when:
689 * - an untracked/uninit-ed object is found
690 */
691 static bool timer_fixup_assert_init(void *addr, enum debug_obj_state state)
692 {
693 struct timer_list *timer = addr;
694
695 switch (state) {
696 case ODEBUG_STATE_NOTAVAILABLE:
697 setup_timer(timer, stub_timer, 0);
698 return true;
699 default:
700 return false;
701 }
702 }
703
704 static struct debug_obj_descr timer_debug_descr = {
705 .name = "timer_list",
706 .debug_hint = timer_debug_hint,
707 .is_static_object = timer_is_static_object,
708 .fixup_init = timer_fixup_init,
709 .fixup_activate = timer_fixup_activate,
710 .fixup_free = timer_fixup_free,
711 .fixup_assert_init = timer_fixup_assert_init,
712 };
713
714 static inline void debug_timer_init(struct timer_list *timer)
715 {
716 debug_object_init(timer, &timer_debug_descr);
717 }
718
719 static inline void debug_timer_activate(struct timer_list *timer)
720 {
721 debug_object_activate(timer, &timer_debug_descr);
722 }
723
724 static inline void debug_timer_deactivate(struct timer_list *timer)
725 {
726 debug_object_deactivate(timer, &timer_debug_descr);
727 }
728
729 static inline void debug_timer_free(struct timer_list *timer)
730 {
731 debug_object_free(timer, &timer_debug_descr);
732 }
733
734 static inline void debug_timer_assert_init(struct timer_list *timer)
735 {
736 debug_object_assert_init(timer, &timer_debug_descr);
737 }
738
739 static void do_init_timer(struct timer_list *timer, unsigned int flags,
740 const char *name, struct lock_class_key *key);
741
742 void init_timer_on_stack_key(struct timer_list *timer, unsigned int flags,
743 const char *name, struct lock_class_key *key)
744 {
745 debug_object_init_on_stack(timer, &timer_debug_descr);
746 do_init_timer(timer, flags, name, key);
747 }
748 EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
749
750 void destroy_timer_on_stack(struct timer_list *timer)
751 {
752 debug_object_free(timer, &timer_debug_descr);
753 }
754 EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
755
756 #else
757 static inline void debug_timer_init(struct timer_list *timer) { }
758 static inline void debug_timer_activate(struct timer_list *timer) { }
759 static inline void debug_timer_deactivate(struct timer_list *timer) { }
760 static inline void debug_timer_assert_init(struct timer_list *timer) { }
761 #endif
762
763 static inline void debug_init(struct timer_list *timer)
764 {
765 debug_timer_init(timer);
766 trace_timer_init(timer);
767 }
768
769 static inline void
770 debug_activate(struct timer_list *timer, unsigned long expires)
771 {
772 debug_timer_activate(timer);
773 trace_timer_start(timer, expires, timer->flags);
774 }
775
776 static inline void debug_deactivate(struct timer_list *timer)
777 {
778 debug_timer_deactivate(timer);
779 trace_timer_cancel(timer);
780 }
781
782 static inline void debug_assert_init(struct timer_list *timer)
783 {
784 debug_timer_assert_init(timer);
785 }
786
787 static void do_init_timer(struct timer_list *timer, unsigned int flags,
788 const char *name, struct lock_class_key *key)
789 {
790 timer->entry.pprev = NULL;
791 timer->flags = flags | raw_smp_processor_id();
792 #ifdef CONFIG_TIMER_STATS
793 timer->start_site = NULL;
794 timer->start_pid = -1;
795 memset(timer->start_comm, 0, TASK_COMM_LEN);
796 #endif
797 lockdep_init_map(&timer->lockdep_map, name, key, 0);
798 }
799
800 /**
801 * init_timer_key - initialize a timer
802 * @timer: the timer to be initialized
803 * @flags: timer flags
804 * @name: name of the timer
805 * @key: lockdep class key of the fake lock used for tracking timer
806 * sync lock dependencies
807 *
808 * init_timer_key() must be done to a timer prior calling *any* of the
809 * other timer functions.
810 */
811 void init_timer_key(struct timer_list *timer, unsigned int flags,
812 const char *name, struct lock_class_key *key)
813 {
814 debug_init(timer);
815 do_init_timer(timer, flags, name, key);
816 }
817 EXPORT_SYMBOL(init_timer_key);
818
819 static inline void detach_timer(struct timer_list *timer, bool clear_pending)
820 {
821 struct hlist_node *entry = &timer->entry;
822
823 debug_deactivate(timer);
824
825 __hlist_del(entry);
826 if (clear_pending)
827 entry->pprev = NULL;
828 entry->next = LIST_POISON2;
829 }
830
831 static int detach_if_pending(struct timer_list *timer, struct timer_base *base,
832 bool clear_pending)
833 {
834 unsigned idx = timer_get_idx(timer);
835
836 if (!timer_pending(timer))
837 return 0;
838
839 if (hlist_is_singular_node(&timer->entry, base->vectors + idx))
840 __clear_bit(idx, base->pending_map);
841
842 detach_timer(timer, clear_pending);
843 return 1;
844 }
845
846 static inline struct timer_base *get_timer_cpu_base(u32 tflags, u32 cpu)
847 {
848 struct timer_base *base = per_cpu_ptr(&timer_bases[BASE_STD], cpu);
849
850 /*
851 * If the timer is deferrable and nohz is active then we need to use
852 * the deferrable base.
853 */
854 if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && base->nohz_active &&
855 (tflags & TIMER_DEFERRABLE))
856 base = per_cpu_ptr(&timer_bases[BASE_DEF], cpu);
857 return base;
858 }
859
860 static inline struct timer_base *get_timer_this_cpu_base(u32 tflags)
861 {
862 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
863
864 /*
865 * If the timer is deferrable and nohz is active then we need to use
866 * the deferrable base.
867 */
868 if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && base->nohz_active &&
869 (tflags & TIMER_DEFERRABLE))
870 base = this_cpu_ptr(&timer_bases[BASE_DEF]);
871 return base;
872 }
873
874 static inline struct timer_base *get_timer_base(u32 tflags)
875 {
876 return get_timer_cpu_base(tflags, tflags & TIMER_CPUMASK);
877 }
878
879 #ifdef CONFIG_NO_HZ_COMMON
880 static inline struct timer_base *
881 __get_target_base(struct timer_base *base, unsigned tflags)
882 {
883 #ifdef CONFIG_SMP
884 if ((tflags & TIMER_PINNED) || !base->migration_enabled)
885 return get_timer_this_cpu_base(tflags);
886 return get_timer_cpu_base(tflags, get_nohz_timer_target());
887 #else
888 return get_timer_this_cpu_base(tflags);
889 #endif
890 }
891
892 static inline void forward_timer_base(struct timer_base *base)
893 {
894 /*
895 * We only forward the base when it's idle and we have a delta between
896 * base clock and jiffies.
897 */
898 if (!base->is_idle || (long) (jiffies - base->clk) < 2)
899 return;
900
901 /*
902 * If the next expiry value is > jiffies, then we fast forward to
903 * jiffies otherwise we forward to the next expiry value.
904 */
905 if (time_after(base->next_expiry, jiffies))
906 base->clk = jiffies;
907 else
908 base->clk = base->next_expiry;
909 }
910 #else
911 static inline struct timer_base *
912 __get_target_base(struct timer_base *base, unsigned tflags)
913 {
914 return get_timer_this_cpu_base(tflags);
915 }
916
917 static inline void forward_timer_base(struct timer_base *base) { }
918 #endif
919
920 static inline struct timer_base *
921 get_target_base(struct timer_base *base, unsigned tflags)
922 {
923 struct timer_base *target = __get_target_base(base, tflags);
924
925 forward_timer_base(target);
926 return target;
927 }
928
929 /*
930 * We are using hashed locking: Holding per_cpu(timer_bases[x]).lock means
931 * that all timers which are tied to this base are locked, and the base itself
932 * is locked too.
933 *
934 * So __run_timers/migrate_timers can safely modify all timers which could
935 * be found in the base->vectors array.
936 *
937 * When a timer is migrating then the TIMER_MIGRATING flag is set and we need
938 * to wait until the migration is done.
939 */
940 static struct timer_base *lock_timer_base(struct timer_list *timer,
941 unsigned long *flags)
942 __acquires(timer->base->lock)
943 {
944 for (;;) {
945 struct timer_base *base;
946 u32 tf = timer->flags;
947
948 if (!(tf & TIMER_MIGRATING)) {
949 base = get_timer_base(tf);
950 spin_lock_irqsave(&base->lock, *flags);
951 if (timer->flags == tf)
952 return base;
953 spin_unlock_irqrestore(&base->lock, *flags);
954 }
955 cpu_relax();
956 }
957 }
958
959 static inline int
960 __mod_timer(struct timer_list *timer, unsigned long expires, bool pending_only)
961 {
962 struct timer_base *base, *new_base;
963 unsigned int idx = UINT_MAX;
964 unsigned long clk = 0, flags;
965 int ret = 0;
966
967 /*
968 * This is a common optimization triggered by the networking code - if
969 * the timer is re-modified to have the same timeout or ends up in the
970 * same array bucket then just return:
971 */
972 if (timer_pending(timer)) {
973 if (timer->expires == expires)
974 return 1;
975 /*
976 * Take the current timer_jiffies of base, but without holding
977 * the lock!
978 */
979 base = get_timer_base(timer->flags);
980 clk = base->clk;
981
982 idx = calc_wheel_index(expires, clk);
983
984 /*
985 * Retrieve and compare the array index of the pending
986 * timer. If it matches set the expiry to the new value so a
987 * subsequent call will exit in the expires check above.
988 */
989 if (idx == timer_get_idx(timer)) {
990 timer->expires = expires;
991 return 1;
992 }
993 }
994
995 timer_stats_timer_set_start_info(timer);
996 BUG_ON(!timer->function);
997
998 base = lock_timer_base(timer, &flags);
999
1000 ret = detach_if_pending(timer, base, false);
1001 if (!ret && pending_only)
1002 goto out_unlock;
1003
1004 debug_activate(timer, expires);
1005
1006 new_base = get_target_base(base, timer->flags);
1007
1008 if (base != new_base) {
1009 /*
1010 * We are trying to schedule the timer on the new base.
1011 * However we can't change timer's base while it is running,
1012 * otherwise del_timer_sync() can't detect that the timer's
1013 * handler yet has not finished. This also guarantees that the
1014 * timer is serialized wrt itself.
1015 */
1016 if (likely(base->running_timer != timer)) {
1017 /* See the comment in lock_timer_base() */
1018 timer->flags |= TIMER_MIGRATING;
1019
1020 spin_unlock(&base->lock);
1021 base = new_base;
1022 spin_lock(&base->lock);
1023 WRITE_ONCE(timer->flags,
1024 (timer->flags & ~TIMER_BASEMASK) | base->cpu);
1025 }
1026 }
1027
1028 timer->expires = expires;
1029 /*
1030 * If 'idx' was calculated above and the base time did not advance
1031 * between calculating 'idx' and taking the lock, only enqueue_timer()
1032 * and trigger_dyntick_cpu() is required. Otherwise we need to
1033 * (re)calculate the wheel index via internal_add_timer().
1034 */
1035 if (idx != UINT_MAX && clk == base->clk) {
1036 enqueue_timer(base, timer, idx);
1037 trigger_dyntick_cpu(base, timer);
1038 } else {
1039 internal_add_timer(base, timer);
1040 }
1041
1042 out_unlock:
1043 spin_unlock_irqrestore(&base->lock, flags);
1044
1045 return ret;
1046 }
1047
1048 /**
1049 * mod_timer_pending - modify a pending timer's timeout
1050 * @timer: the pending timer to be modified
1051 * @expires: new timeout in jiffies
1052 *
1053 * mod_timer_pending() is the same for pending timers as mod_timer(),
1054 * but will not re-activate and modify already deleted timers.
1055 *
1056 * It is useful for unserialized use of timers.
1057 */
1058 int mod_timer_pending(struct timer_list *timer, unsigned long expires)
1059 {
1060 return __mod_timer(timer, expires, true);
1061 }
1062 EXPORT_SYMBOL(mod_timer_pending);
1063
1064 /**
1065 * mod_timer - modify a timer's timeout
1066 * @timer: the timer to be modified
1067 * @expires: new timeout in jiffies
1068 *
1069 * mod_timer() is a more efficient way to update the expire field of an
1070 * active timer (if the timer is inactive it will be activated)
1071 *
1072 * mod_timer(timer, expires) is equivalent to:
1073 *
1074 * del_timer(timer); timer->expires = expires; add_timer(timer);
1075 *
1076 * Note that if there are multiple unserialized concurrent users of the
1077 * same timer, then mod_timer() is the only safe way to modify the timeout,
1078 * since add_timer() cannot modify an already running timer.
1079 *
1080 * The function returns whether it has modified a pending timer or not.
1081 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
1082 * active timer returns 1.)
1083 */
1084 int mod_timer(struct timer_list *timer, unsigned long expires)
1085 {
1086 return __mod_timer(timer, expires, false);
1087 }
1088 EXPORT_SYMBOL(mod_timer);
1089
1090 /**
1091 * add_timer - start a timer
1092 * @timer: the timer to be added
1093 *
1094 * The kernel will do a ->function(->data) callback from the
1095 * timer interrupt at the ->expires point in the future. The
1096 * current time is 'jiffies'.
1097 *
1098 * The timer's ->expires, ->function (and if the handler uses it, ->data)
1099 * fields must be set prior calling this function.
1100 *
1101 * Timers with an ->expires field in the past will be executed in the next
1102 * timer tick.
1103 */
1104 void add_timer(struct timer_list *timer)
1105 {
1106 BUG_ON(timer_pending(timer));
1107 mod_timer(timer, timer->expires);
1108 }
1109 EXPORT_SYMBOL(add_timer);
1110
1111 /**
1112 * add_timer_on - start a timer on a particular CPU
1113 * @timer: the timer to be added
1114 * @cpu: the CPU to start it on
1115 *
1116 * This is not very scalable on SMP. Double adds are not possible.
1117 */
1118 void add_timer_on(struct timer_list *timer, int cpu)
1119 {
1120 struct timer_base *new_base, *base;
1121 unsigned long flags;
1122
1123 timer_stats_timer_set_start_info(timer);
1124 BUG_ON(timer_pending(timer) || !timer->function);
1125
1126 new_base = get_timer_cpu_base(timer->flags, cpu);
1127
1128 /*
1129 * If @timer was on a different CPU, it should be migrated with the
1130 * old base locked to prevent other operations proceeding with the
1131 * wrong base locked. See lock_timer_base().
1132 */
1133 base = lock_timer_base(timer, &flags);
1134 if (base != new_base) {
1135 timer->flags |= TIMER_MIGRATING;
1136
1137 spin_unlock(&base->lock);
1138 base = new_base;
1139 spin_lock(&base->lock);
1140 WRITE_ONCE(timer->flags,
1141 (timer->flags & ~TIMER_BASEMASK) | cpu);
1142 }
1143
1144 debug_activate(timer, timer->expires);
1145 internal_add_timer(base, timer);
1146 spin_unlock_irqrestore(&base->lock, flags);
1147 }
1148 EXPORT_SYMBOL_GPL(add_timer_on);
1149
1150 /**
1151 * del_timer - deactive a timer.
1152 * @timer: the timer to be deactivated
1153 *
1154 * del_timer() deactivates a timer - this works on both active and inactive
1155 * timers.
1156 *
1157 * The function returns whether it has deactivated a pending timer or not.
1158 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
1159 * active timer returns 1.)
1160 */
1161 int del_timer(struct timer_list *timer)
1162 {
1163 struct timer_base *base;
1164 unsigned long flags;
1165 int ret = 0;
1166
1167 debug_assert_init(timer);
1168
1169 timer_stats_timer_clear_start_info(timer);
1170 if (timer_pending(timer)) {
1171 base = lock_timer_base(timer, &flags);
1172 ret = detach_if_pending(timer, base, true);
1173 spin_unlock_irqrestore(&base->lock, flags);
1174 }
1175
1176 return ret;
1177 }
1178 EXPORT_SYMBOL(del_timer);
1179
1180 /**
1181 * try_to_del_timer_sync - Try to deactivate a timer
1182 * @timer: timer do del
1183 *
1184 * This function tries to deactivate a timer. Upon successful (ret >= 0)
1185 * exit the timer is not queued and the handler is not running on any CPU.
1186 */
1187 int try_to_del_timer_sync(struct timer_list *timer)
1188 {
1189 struct timer_base *base;
1190 unsigned long flags;
1191 int ret = -1;
1192
1193 debug_assert_init(timer);
1194
1195 base = lock_timer_base(timer, &flags);
1196
1197 if (base->running_timer != timer) {
1198 timer_stats_timer_clear_start_info(timer);
1199 ret = detach_if_pending(timer, base, true);
1200 }
1201 spin_unlock_irqrestore(&base->lock, flags);
1202
1203 return ret;
1204 }
1205 EXPORT_SYMBOL(try_to_del_timer_sync);
1206
1207 #ifdef CONFIG_SMP
1208 /**
1209 * del_timer_sync - deactivate a timer and wait for the handler to finish.
1210 * @timer: the timer to be deactivated
1211 *
1212 * This function only differs from del_timer() on SMP: besides deactivating
1213 * the timer it also makes sure the handler has finished executing on other
1214 * CPUs.
1215 *
1216 * Synchronization rules: Callers must prevent restarting of the timer,
1217 * otherwise this function is meaningless. It must not be called from
1218 * interrupt contexts unless the timer is an irqsafe one. The caller must
1219 * not hold locks which would prevent completion of the timer's
1220 * handler. The timer's handler must not call add_timer_on(). Upon exit the
1221 * timer is not queued and the handler is not running on any CPU.
1222 *
1223 * Note: For !irqsafe timers, you must not hold locks that are held in
1224 * interrupt context while calling this function. Even if the lock has
1225 * nothing to do with the timer in question. Here's why:
1226 *
1227 * CPU0 CPU1
1228 * ---- ----
1229 * <SOFTIRQ>
1230 * call_timer_fn();
1231 * base->running_timer = mytimer;
1232 * spin_lock_irq(somelock);
1233 * <IRQ>
1234 * spin_lock(somelock);
1235 * del_timer_sync(mytimer);
1236 * while (base->running_timer == mytimer);
1237 *
1238 * Now del_timer_sync() will never return and never release somelock.
1239 * The interrupt on the other CPU is waiting to grab somelock but
1240 * it has interrupted the softirq that CPU0 is waiting to finish.
1241 *
1242 * The function returns whether it has deactivated a pending timer or not.
1243 */
1244 int del_timer_sync(struct timer_list *timer)
1245 {
1246 #ifdef CONFIG_LOCKDEP
1247 unsigned long flags;
1248
1249 /*
1250 * If lockdep gives a backtrace here, please reference
1251 * the synchronization rules above.
1252 */
1253 local_irq_save(flags);
1254 lock_map_acquire(&timer->lockdep_map);
1255 lock_map_release(&timer->lockdep_map);
1256 local_irq_restore(flags);
1257 #endif
1258 /*
1259 * don't use it in hardirq context, because it
1260 * could lead to deadlock.
1261 */
1262 WARN_ON(in_irq() && !(timer->flags & TIMER_IRQSAFE));
1263 for (;;) {
1264 int ret = try_to_del_timer_sync(timer);
1265 if (ret >= 0)
1266 return ret;
1267 cpu_relax();
1268 }
1269 }
1270 EXPORT_SYMBOL(del_timer_sync);
1271 #endif
1272
1273 static void call_timer_fn(struct timer_list *timer, void (*fn)(unsigned long),
1274 unsigned long data)
1275 {
1276 int count = preempt_count();
1277
1278 #ifdef CONFIG_LOCKDEP
1279 /*
1280 * It is permissible to free the timer from inside the
1281 * function that is called from it, this we need to take into
1282 * account for lockdep too. To avoid bogus "held lock freed"
1283 * warnings as well as problems when looking into
1284 * timer->lockdep_map, make a copy and use that here.
1285 */
1286 struct lockdep_map lockdep_map;
1287
1288 lockdep_copy_map(&lockdep_map, &timer->lockdep_map);
1289 #endif
1290 /*
1291 * Couple the lock chain with the lock chain at
1292 * del_timer_sync() by acquiring the lock_map around the fn()
1293 * call here and in del_timer_sync().
1294 */
1295 lock_map_acquire(&lockdep_map);
1296
1297 trace_timer_expire_entry(timer);
1298 fn(data);
1299 trace_timer_expire_exit(timer);
1300
1301 lock_map_release(&lockdep_map);
1302
1303 if (count != preempt_count()) {
1304 WARN_ONCE(1, "timer: %pF preempt leak: %08x -> %08x\n",
1305 fn, count, preempt_count());
1306 /*
1307 * Restore the preempt count. That gives us a decent
1308 * chance to survive and extract information. If the
1309 * callback kept a lock held, bad luck, but not worse
1310 * than the BUG() we had.
1311 */
1312 preempt_count_set(count);
1313 }
1314 }
1315
1316 static void expire_timers(struct timer_base *base, struct hlist_head *head)
1317 {
1318 while (!hlist_empty(head)) {
1319 struct timer_list *timer;
1320 void (*fn)(unsigned long);
1321 unsigned long data;
1322
1323 timer = hlist_entry(head->first, struct timer_list, entry);
1324 timer_stats_account_timer(timer);
1325
1326 base->running_timer = timer;
1327 detach_timer(timer, true);
1328
1329 fn = timer->function;
1330 data = timer->data;
1331
1332 if (timer->flags & TIMER_IRQSAFE) {
1333 spin_unlock(&base->lock);
1334 call_timer_fn(timer, fn, data);
1335 spin_lock(&base->lock);
1336 } else {
1337 spin_unlock_irq(&base->lock);
1338 call_timer_fn(timer, fn, data);
1339 spin_lock_irq(&base->lock);
1340 }
1341 }
1342 }
1343
1344 static int __collect_expired_timers(struct timer_base *base,
1345 struct hlist_head *heads)
1346 {
1347 unsigned long clk = base->clk;
1348 struct hlist_head *vec;
1349 int i, levels = 0;
1350 unsigned int idx;
1351
1352 for (i = 0; i < LVL_DEPTH; i++) {
1353 idx = (clk & LVL_MASK) + i * LVL_SIZE;
1354
1355 if (__test_and_clear_bit(idx, base->pending_map)) {
1356 vec = base->vectors + idx;
1357 hlist_move_list(vec, heads++);
1358 levels++;
1359 }
1360 /* Is it time to look at the next level? */
1361 if (clk & LVL_CLK_MASK)
1362 break;
1363 /* Shift clock for the next level granularity */
1364 clk >>= LVL_CLK_SHIFT;
1365 }
1366 return levels;
1367 }
1368
1369 #ifdef CONFIG_NO_HZ_COMMON
1370 /*
1371 * Find the next pending bucket of a level. Search from level start (@offset)
1372 * + @clk upwards and if nothing there, search from start of the level
1373 * (@offset) up to @offset + clk.
1374 */
1375 static int next_pending_bucket(struct timer_base *base, unsigned offset,
1376 unsigned clk)
1377 {
1378 unsigned pos, start = offset + clk;
1379 unsigned end = offset + LVL_SIZE;
1380
1381 pos = find_next_bit(base->pending_map, end, start);
1382 if (pos < end)
1383 return pos - start;
1384
1385 pos = find_next_bit(base->pending_map, start, offset);
1386 return pos < start ? pos + LVL_SIZE - start : -1;
1387 }
1388
1389 /*
1390 * Search the first expiring timer in the various clock levels. Caller must
1391 * hold base->lock.
1392 */
1393 static unsigned long __next_timer_interrupt(struct timer_base *base)
1394 {
1395 unsigned long clk, next, adj;
1396 unsigned lvl, offset = 0;
1397
1398 next = base->clk + NEXT_TIMER_MAX_DELTA;
1399 clk = base->clk;
1400 for (lvl = 0; lvl < LVL_DEPTH; lvl++, offset += LVL_SIZE) {
1401 int pos = next_pending_bucket(base, offset, clk & LVL_MASK);
1402
1403 if (pos >= 0) {
1404 unsigned long tmp = clk + (unsigned long) pos;
1405
1406 tmp <<= LVL_SHIFT(lvl);
1407 if (time_before(tmp, next))
1408 next = tmp;
1409 }
1410 /*
1411 * Clock for the next level. If the current level clock lower
1412 * bits are zero, we look at the next level as is. If not we
1413 * need to advance it by one because that's going to be the
1414 * next expiring bucket in that level. base->clk is the next
1415 * expiring jiffie. So in case of:
1416 *
1417 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1418 * 0 0 0 0 0 0
1419 *
1420 * we have to look at all levels @index 0. With
1421 *
1422 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1423 * 0 0 0 0 0 2
1424 *
1425 * LVL0 has the next expiring bucket @index 2. The upper
1426 * levels have the next expiring bucket @index 1.
1427 *
1428 * In case that the propagation wraps the next level the same
1429 * rules apply:
1430 *
1431 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1432 * 0 0 0 0 F 2
1433 *
1434 * So after looking at LVL0 we get:
1435 *
1436 * LVL5 LVL4 LVL3 LVL2 LVL1
1437 * 0 0 0 1 0
1438 *
1439 * So no propagation from LVL1 to LVL2 because that happened
1440 * with the add already, but then we need to propagate further
1441 * from LVL2 to LVL3.
1442 *
1443 * So the simple check whether the lower bits of the current
1444 * level are 0 or not is sufficient for all cases.
1445 */
1446 adj = clk & LVL_CLK_MASK ? 1 : 0;
1447 clk >>= LVL_CLK_SHIFT;
1448 clk += adj;
1449 }
1450 return next;
1451 }
1452
1453 /*
1454 * Check, if the next hrtimer event is before the next timer wheel
1455 * event:
1456 */
1457 static u64 cmp_next_hrtimer_event(u64 basem, u64 expires)
1458 {
1459 u64 nextevt = hrtimer_get_next_event();
1460
1461 /*
1462 * If high resolution timers are enabled
1463 * hrtimer_get_next_event() returns KTIME_MAX.
1464 */
1465 if (expires <= nextevt)
1466 return expires;
1467
1468 /*
1469 * If the next timer is already expired, return the tick base
1470 * time so the tick is fired immediately.
1471 */
1472 if (nextevt <= basem)
1473 return basem;
1474
1475 /*
1476 * Round up to the next jiffie. High resolution timers are
1477 * off, so the hrtimers are expired in the tick and we need to
1478 * make sure that this tick really expires the timer to avoid
1479 * a ping pong of the nohz stop code.
1480 *
1481 * Use DIV_ROUND_UP_ULL to prevent gcc calling __divdi3
1482 */
1483 return DIV_ROUND_UP_ULL(nextevt, TICK_NSEC) * TICK_NSEC;
1484 }
1485
1486 /**
1487 * get_next_timer_interrupt - return the time (clock mono) of the next timer
1488 * @basej: base time jiffies
1489 * @basem: base time clock monotonic
1490 *
1491 * Returns the tick aligned clock monotonic time of the next pending
1492 * timer or KTIME_MAX if no timer is pending.
1493 */
1494 u64 get_next_timer_interrupt(unsigned long basej, u64 basem)
1495 {
1496 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1497 u64 expires = KTIME_MAX;
1498 unsigned long nextevt;
1499 bool is_max_delta;
1500
1501 /*
1502 * Pretend that there is no timer pending if the cpu is offline.
1503 * Possible pending timers will be migrated later to an active cpu.
1504 */
1505 if (cpu_is_offline(smp_processor_id()))
1506 return expires;
1507
1508 spin_lock(&base->lock);
1509 nextevt = __next_timer_interrupt(base);
1510 is_max_delta = (nextevt == base->clk + NEXT_TIMER_MAX_DELTA);
1511 base->next_expiry = nextevt;
1512 /*
1513 * We have a fresh next event. Check whether we can forward the base:
1514 */
1515 if (time_after(nextevt, jiffies))
1516 base->clk = jiffies;
1517 else if (time_after(nextevt, base->clk))
1518 base->clk = nextevt;
1519
1520 if (time_before_eq(nextevt, basej)) {
1521 expires = basem;
1522 base->is_idle = false;
1523 } else {
1524 if (!is_max_delta)
1525 expires = basem + (nextevt - basej) * TICK_NSEC;
1526 /*
1527 * If we expect to sleep more than a tick, mark the base idle:
1528 */
1529 if ((expires - basem) > TICK_NSEC)
1530 base->is_idle = true;
1531 }
1532 spin_unlock(&base->lock);
1533
1534 return cmp_next_hrtimer_event(basem, expires);
1535 }
1536
1537 /**
1538 * timer_clear_idle - Clear the idle state of the timer base
1539 *
1540 * Called with interrupts disabled
1541 */
1542 void timer_clear_idle(void)
1543 {
1544 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1545
1546 /*
1547 * We do this unlocked. The worst outcome is a remote enqueue sending
1548 * a pointless IPI, but taking the lock would just make the window for
1549 * sending the IPI a few instructions smaller for the cost of taking
1550 * the lock in the exit from idle path.
1551 */
1552 base->is_idle = false;
1553 }
1554
1555 static int collect_expired_timers(struct timer_base *base,
1556 struct hlist_head *heads)
1557 {
1558 /*
1559 * NOHZ optimization. After a long idle sleep we need to forward the
1560 * base to current jiffies. Avoid a loop by searching the bitfield for
1561 * the next expiring timer.
1562 */
1563 if ((long)(jiffies - base->clk) > 2) {
1564 unsigned long next = __next_timer_interrupt(base);
1565
1566 /*
1567 * If the next timer is ahead of time forward to current
1568 * jiffies, otherwise forward to the next expiry time:
1569 */
1570 if (time_after(next, jiffies)) {
1571 /* The call site will increment clock! */
1572 base->clk = jiffies - 1;
1573 return 0;
1574 }
1575 base->clk = next;
1576 }
1577 return __collect_expired_timers(base, heads);
1578 }
1579 #else
1580 static inline int collect_expired_timers(struct timer_base *base,
1581 struct hlist_head *heads)
1582 {
1583 return __collect_expired_timers(base, heads);
1584 }
1585 #endif
1586
1587 /*
1588 * Called from the timer interrupt handler to charge one tick to the current
1589 * process. user_tick is 1 if the tick is user time, 0 for system.
1590 */
1591 void update_process_times(int user_tick)
1592 {
1593 struct task_struct *p = current;
1594
1595 /* Note: this timer irq context must be accounted for as well. */
1596 account_process_tick(p, user_tick);
1597 run_local_timers();
1598 rcu_check_callbacks(user_tick);
1599 #ifdef CONFIG_IRQ_WORK
1600 if (in_irq())
1601 irq_work_tick();
1602 #endif
1603 scheduler_tick();
1604 run_posix_cpu_timers(p);
1605 }
1606
1607 /**
1608 * __run_timers - run all expired timers (if any) on this CPU.
1609 * @base: the timer vector to be processed.
1610 */
1611 static inline void __run_timers(struct timer_base *base)
1612 {
1613 struct hlist_head heads[LVL_DEPTH];
1614 int levels;
1615
1616 if (!time_after_eq(jiffies, base->clk))
1617 return;
1618
1619 spin_lock_irq(&base->lock);
1620
1621 while (time_after_eq(jiffies, base->clk)) {
1622
1623 levels = collect_expired_timers(base, heads);
1624 base->clk++;
1625
1626 while (levels--)
1627 expire_timers(base, heads + levels);
1628 }
1629 base->running_timer = NULL;
1630 spin_unlock_irq(&base->lock);
1631 }
1632
1633 /*
1634 * This function runs timers and the timer-tq in bottom half context.
1635 */
1636 static __latent_entropy void run_timer_softirq(struct softirq_action *h)
1637 {
1638 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1639
1640 __run_timers(base);
1641 if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && base->nohz_active)
1642 __run_timers(this_cpu_ptr(&timer_bases[BASE_DEF]));
1643 }
1644
1645 /*
1646 * Called by the local, per-CPU timer interrupt on SMP.
1647 */
1648 void run_local_timers(void)
1649 {
1650 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1651
1652 hrtimer_run_queues();
1653 /* Raise the softirq only if required. */
1654 if (time_before(jiffies, base->clk)) {
1655 if (!IS_ENABLED(CONFIG_NO_HZ_COMMON) || !base->nohz_active)
1656 return;
1657 /* CPU is awake, so check the deferrable base. */
1658 base++;
1659 if (time_before(jiffies, base->clk))
1660 return;
1661 }
1662 raise_softirq(TIMER_SOFTIRQ);
1663 }
1664
1665 #ifdef __ARCH_WANT_SYS_ALARM
1666
1667 /*
1668 * For backwards compatibility? This can be done in libc so Alpha
1669 * and all newer ports shouldn't need it.
1670 */
1671 SYSCALL_DEFINE1(alarm, unsigned int, seconds)
1672 {
1673 return alarm_setitimer(seconds);
1674 }
1675
1676 #endif
1677
1678 static void process_timeout(unsigned long __data)
1679 {
1680 wake_up_process((struct task_struct *)__data);
1681 }
1682
1683 /**
1684 * schedule_timeout - sleep until timeout
1685 * @timeout: timeout value in jiffies
1686 *
1687 * Make the current task sleep until @timeout jiffies have
1688 * elapsed. The routine will return immediately unless
1689 * the current task state has been set (see set_current_state()).
1690 *
1691 * You can set the task state as follows -
1692 *
1693 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1694 * pass before the routine returns. The routine will return 0
1695 *
1696 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1697 * delivered to the current task. In this case the remaining time
1698 * in jiffies will be returned, or 0 if the timer expired in time
1699 *
1700 * The current task state is guaranteed to be TASK_RUNNING when this
1701 * routine returns.
1702 *
1703 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1704 * the CPU away without a bound on the timeout. In this case the return
1705 * value will be %MAX_SCHEDULE_TIMEOUT.
1706 *
1707 * In all cases the return value is guaranteed to be non-negative.
1708 */
1709 signed long __sched schedule_timeout(signed long timeout)
1710 {
1711 struct timer_list timer;
1712 unsigned long expire;
1713
1714 switch (timeout)
1715 {
1716 case MAX_SCHEDULE_TIMEOUT:
1717 /*
1718 * These two special cases are useful to be comfortable
1719 * in the caller. Nothing more. We could take
1720 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1721 * but I' d like to return a valid offset (>=0) to allow
1722 * the caller to do everything it want with the retval.
1723 */
1724 schedule();
1725 goto out;
1726 default:
1727 /*
1728 * Another bit of PARANOID. Note that the retval will be
1729 * 0 since no piece of kernel is supposed to do a check
1730 * for a negative retval of schedule_timeout() (since it
1731 * should never happens anyway). You just have the printk()
1732 * that will tell you if something is gone wrong and where.
1733 */
1734 if (timeout < 0) {
1735 printk(KERN_ERR "schedule_timeout: wrong timeout "
1736 "value %lx\n", timeout);
1737 dump_stack();
1738 current->state = TASK_RUNNING;
1739 goto out;
1740 }
1741 }
1742
1743 expire = timeout + jiffies;
1744
1745 setup_timer_on_stack(&timer, process_timeout, (unsigned long)current);
1746 __mod_timer(&timer, expire, false);
1747 schedule();
1748 del_singleshot_timer_sync(&timer);
1749
1750 /* Remove the timer from the object tracker */
1751 destroy_timer_on_stack(&timer);
1752
1753 timeout = expire - jiffies;
1754
1755 out:
1756 return timeout < 0 ? 0 : timeout;
1757 }
1758 EXPORT_SYMBOL(schedule_timeout);
1759
1760 /*
1761 * We can use __set_current_state() here because schedule_timeout() calls
1762 * schedule() unconditionally.
1763 */
1764 signed long __sched schedule_timeout_interruptible(signed long timeout)
1765 {
1766 __set_current_state(TASK_INTERRUPTIBLE);
1767 return schedule_timeout(timeout);
1768 }
1769 EXPORT_SYMBOL(schedule_timeout_interruptible);
1770
1771 signed long __sched schedule_timeout_killable(signed long timeout)
1772 {
1773 __set_current_state(TASK_KILLABLE);
1774 return schedule_timeout(timeout);
1775 }
1776 EXPORT_SYMBOL(schedule_timeout_killable);
1777
1778 signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1779 {
1780 __set_current_state(TASK_UNINTERRUPTIBLE);
1781 return schedule_timeout(timeout);
1782 }
1783 EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1784
1785 /*
1786 * Like schedule_timeout_uninterruptible(), except this task will not contribute
1787 * to load average.
1788 */
1789 signed long __sched schedule_timeout_idle(signed long timeout)
1790 {
1791 __set_current_state(TASK_IDLE);
1792 return schedule_timeout(timeout);
1793 }
1794 EXPORT_SYMBOL(schedule_timeout_idle);
1795
1796 #ifdef CONFIG_HOTPLUG_CPU
1797 static void migrate_timer_list(struct timer_base *new_base, struct hlist_head *head)
1798 {
1799 struct timer_list *timer;
1800 int cpu = new_base->cpu;
1801
1802 while (!hlist_empty(head)) {
1803 timer = hlist_entry(head->first, struct timer_list, entry);
1804 detach_timer(timer, false);
1805 timer->flags = (timer->flags & ~TIMER_BASEMASK) | cpu;
1806 internal_add_timer(new_base, timer);
1807 }
1808 }
1809
1810 int timers_dead_cpu(unsigned int cpu)
1811 {
1812 struct timer_base *old_base;
1813 struct timer_base *new_base;
1814 int b, i;
1815
1816 BUG_ON(cpu_online(cpu));
1817
1818 for (b = 0; b < NR_BASES; b++) {
1819 old_base = per_cpu_ptr(&timer_bases[b], cpu);
1820 new_base = get_cpu_ptr(&timer_bases[b]);
1821 /*
1822 * The caller is globally serialized and nobody else
1823 * takes two locks at once, deadlock is not possible.
1824 */
1825 spin_lock_irq(&new_base->lock);
1826 spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1827
1828 BUG_ON(old_base->running_timer);
1829
1830 for (i = 0; i < WHEEL_SIZE; i++)
1831 migrate_timer_list(new_base, old_base->vectors + i);
1832
1833 spin_unlock(&old_base->lock);
1834 spin_unlock_irq(&new_base->lock);
1835 put_cpu_ptr(&timer_bases);
1836 }
1837 return 0;
1838 }
1839
1840 #endif /* CONFIG_HOTPLUG_CPU */
1841
1842 static void __init init_timer_cpu(int cpu)
1843 {
1844 struct timer_base *base;
1845 int i;
1846
1847 for (i = 0; i < NR_BASES; i++) {
1848 base = per_cpu_ptr(&timer_bases[i], cpu);
1849 base->cpu = cpu;
1850 spin_lock_init(&base->lock);
1851 base->clk = jiffies;
1852 }
1853 }
1854
1855 static void __init init_timer_cpus(void)
1856 {
1857 int cpu;
1858
1859 for_each_possible_cpu(cpu)
1860 init_timer_cpu(cpu);
1861 }
1862
1863 void __init init_timers(void)
1864 {
1865 init_timer_cpus();
1866 init_timer_stats();
1867 open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
1868 }
1869
1870 /**
1871 * msleep - sleep safely even with waitqueue interruptions
1872 * @msecs: Time in milliseconds to sleep for
1873 */
1874 void msleep(unsigned int msecs)
1875 {
1876 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1877
1878 while (timeout)
1879 timeout = schedule_timeout_uninterruptible(timeout);
1880 }
1881
1882 EXPORT_SYMBOL(msleep);
1883
1884 /**
1885 * msleep_interruptible - sleep waiting for signals
1886 * @msecs: Time in milliseconds to sleep for
1887 */
1888 unsigned long msleep_interruptible(unsigned int msecs)
1889 {
1890 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1891
1892 while (timeout && !signal_pending(current))
1893 timeout = schedule_timeout_interruptible(timeout);
1894 return jiffies_to_msecs(timeout);
1895 }
1896
1897 EXPORT_SYMBOL(msleep_interruptible);
1898
1899 static void __sched do_usleep_range(unsigned long min, unsigned long max)
1900 {
1901 ktime_t kmin;
1902 u64 delta;
1903
1904 kmin = ktime_set(0, min * NSEC_PER_USEC);
1905 delta = (u64)(max - min) * NSEC_PER_USEC;
1906 schedule_hrtimeout_range(&kmin, delta, HRTIMER_MODE_REL);
1907 }
1908
1909 /**
1910 * usleep_range - Sleep for an approximate time
1911 * @min: Minimum time in usecs to sleep
1912 * @max: Maximum time in usecs to sleep
1913 *
1914 * In non-atomic context where the exact wakeup time is flexible, use
1915 * usleep_range() instead of udelay(). The sleep improves responsiveness
1916 * by avoiding the CPU-hogging busy-wait of udelay(), and the range reduces
1917 * power usage by allowing hrtimers to take advantage of an already-
1918 * scheduled interrupt instead of scheduling a new one just for this sleep.
1919 */
1920 void __sched usleep_range(unsigned long min, unsigned long max)
1921 {
1922 __set_current_state(TASK_UNINTERRUPTIBLE);
1923 do_usleep_range(min, max);
1924 }
1925 EXPORT_SYMBOL(usleep_range);
This page took 0.072883 seconds and 5 git commands to generate.