Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Kernel internal timers
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 *
7 * 1997-01-28 Modified by Finn Arne Gangstad to make timers scale better.
8 *
9 * 1997-09-10 Updated NTP code according to technical memorandum Jan '96
10 * "A Kernel Model for Precision Timekeeping" by Dave Mills
11 * 1998-12-24 Fixed a xtime SMP race (we need the xtime_lock rw spinlock to
12 * serialize accesses to xtime/lost_ticks).
13 * Copyright (C) 1998 Andrea Arcangeli
14 * 1999-03-10 Improved NTP compatibility by Ulrich Windl
15 * 2002-05-31 Move sys_sysinfo here and make its locking sane, Robert Love
16 * 2000-10-05 Implemented scalable SMP per-CPU timer handling.
17 * Copyright (C) 2000, 2001, 2002 Ingo Molnar
18 * Designed by David S. Miller, Alexey Kuznetsov and Ingo Molnar
19 */
20
21#include <linux/kernel_stat.h>
22#include <linux/export.h>
23#include <linux/interrupt.h>
24#include <linux/percpu.h>
25#include <linux/init.h>
26#include <linux/mm.h>
27#include <linux/swap.h>
28#include <linux/pid_namespace.h>
29#include <linux/notifier.h>
30#include <linux/thread_info.h>
31#include <linux/time.h>
32#include <linux/jiffies.h>
33#include <linux/posix-timers.h>
34#include <linux/cpu.h>
35#include <linux/syscalls.h>
36#include <linux/delay.h>
37#include <linux/tick.h>
38#include <linux/kallsyms.h>
39#include <linux/irq_work.h>
40#include <linux/sched/signal.h>
41#include <linux/sched/sysctl.h>
42#include <linux/sched/nohz.h>
43#include <linux/sched/debug.h>
44#include <linux/slab.h>
45#include <linux/compat.h>
46#include <linux/random.h>
47#include <linux/sysctl.h>
48
49#include <linux/uaccess.h>
50#include <asm/unistd.h>
51#include <asm/div64.h>
52#include <asm/timex.h>
53#include <asm/io.h>
54
55#include "tick-internal.h"
56
57#define CREATE_TRACE_POINTS
58#include <trace/events/timer.h>
59
60__visible u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
61
62EXPORT_SYMBOL(jiffies_64);
63
64/*
65 * The timer wheel has LVL_DEPTH array levels. Each level provides an array of
66 * LVL_SIZE buckets. Each level is driven by its own clock and therefor each
67 * level has a different granularity.
68 *
69 * The level granularity is: LVL_CLK_DIV ^ lvl
70 * The level clock frequency is: HZ / (LVL_CLK_DIV ^ level)
71 *
72 * The array level of a newly armed timer depends on the relative expiry
73 * time. The farther the expiry time is away the higher the array level and
74 * therefor the granularity becomes.
75 *
76 * Contrary to the original timer wheel implementation, which aims for 'exact'
77 * expiry of the timers, this implementation removes the need for recascading
78 * the timers into the lower array levels. The previous 'classic' timer wheel
79 * implementation of the kernel already violated the 'exact' expiry by adding
80 * slack to the expiry time to provide batched expiration. The granularity
81 * levels provide implicit batching.
82 *
83 * This is an optimization of the original timer wheel implementation for the
84 * majority of the timer wheel use cases: timeouts. The vast majority of
85 * timeout timers (networking, disk I/O ...) are canceled before expiry. If
86 * the timeout expires it indicates that normal operation is disturbed, so it
87 * does not matter much whether the timeout comes with a slight delay.
88 *
89 * The only exception to this are networking timers with a small expiry
90 * time. They rely on the granularity. Those fit into the first wheel level,
91 * which has HZ granularity.
92 *
93 * We don't have cascading anymore. timers with a expiry time above the
94 * capacity of the last wheel level are force expired at the maximum timeout
95 * value of the last wheel level. From data sampling we know that the maximum
96 * value observed is 5 days (network connection tracking), so this should not
97 * be an issue.
98 *
99 * The currently chosen array constants values are a good compromise between
100 * array size and granularity.
101 *
102 * This results in the following granularity and range levels:
103 *
104 * HZ 1000 steps
105 * Level Offset Granularity Range
106 * 0 0 1 ms 0 ms - 63 ms
107 * 1 64 8 ms 64 ms - 511 ms
108 * 2 128 64 ms 512 ms - 4095 ms (512ms - ~4s)
109 * 3 192 512 ms 4096 ms - 32767 ms (~4s - ~32s)
110 * 4 256 4096 ms (~4s) 32768 ms - 262143 ms (~32s - ~4m)
111 * 5 320 32768 ms (~32s) 262144 ms - 2097151 ms (~4m - ~34m)
112 * 6 384 262144 ms (~4m) 2097152 ms - 16777215 ms (~34m - ~4h)
113 * 7 448 2097152 ms (~34m) 16777216 ms - 134217727 ms (~4h - ~1d)
114 * 8 512 16777216 ms (~4h) 134217728 ms - 1073741822 ms (~1d - ~12d)
115 *
116 * HZ 300
117 * Level Offset Granularity Range
118 * 0 0 3 ms 0 ms - 210 ms
119 * 1 64 26 ms 213 ms - 1703 ms (213ms - ~1s)
120 * 2 128 213 ms 1706 ms - 13650 ms (~1s - ~13s)
121 * 3 192 1706 ms (~1s) 13653 ms - 109223 ms (~13s - ~1m)
122 * 4 256 13653 ms (~13s) 109226 ms - 873810 ms (~1m - ~14m)
123 * 5 320 109226 ms (~1m) 873813 ms - 6990503 ms (~14m - ~1h)
124 * 6 384 873813 ms (~14m) 6990506 ms - 55924050 ms (~1h - ~15h)
125 * 7 448 6990506 ms (~1h) 55924053 ms - 447392423 ms (~15h - ~5d)
126 * 8 512 55924053 ms (~15h) 447392426 ms - 3579139406 ms (~5d - ~41d)
127 *
128 * HZ 250
129 * Level Offset Granularity Range
130 * 0 0 4 ms 0 ms - 255 ms
131 * 1 64 32 ms 256 ms - 2047 ms (256ms - ~2s)
132 * 2 128 256 ms 2048 ms - 16383 ms (~2s - ~16s)
133 * 3 192 2048 ms (~2s) 16384 ms - 131071 ms (~16s - ~2m)
134 * 4 256 16384 ms (~16s) 131072 ms - 1048575 ms (~2m - ~17m)
135 * 5 320 131072 ms (~2m) 1048576 ms - 8388607 ms (~17m - ~2h)
136 * 6 384 1048576 ms (~17m) 8388608 ms - 67108863 ms (~2h - ~18h)
137 * 7 448 8388608 ms (~2h) 67108864 ms - 536870911 ms (~18h - ~6d)
138 * 8 512 67108864 ms (~18h) 536870912 ms - 4294967288 ms (~6d - ~49d)
139 *
140 * HZ 100
141 * Level Offset Granularity Range
142 * 0 0 10 ms 0 ms - 630 ms
143 * 1 64 80 ms 640 ms - 5110 ms (640ms - ~5s)
144 * 2 128 640 ms 5120 ms - 40950 ms (~5s - ~40s)
145 * 3 192 5120 ms (~5s) 40960 ms - 327670 ms (~40s - ~5m)
146 * 4 256 40960 ms (~40s) 327680 ms - 2621430 ms (~5m - ~43m)
147 * 5 320 327680 ms (~5m) 2621440 ms - 20971510 ms (~43m - ~5h)
148 * 6 384 2621440 ms (~43m) 20971520 ms - 167772150 ms (~5h - ~1d)
149 * 7 448 20971520 ms (~5h) 167772160 ms - 1342177270 ms (~1d - ~15d)
150 */
151
152/* Clock divisor for the next level */
153#define LVL_CLK_SHIFT 3
154#define LVL_CLK_DIV (1UL << LVL_CLK_SHIFT)
155#define LVL_CLK_MASK (LVL_CLK_DIV - 1)
156#define LVL_SHIFT(n) ((n) * LVL_CLK_SHIFT)
157#define LVL_GRAN(n) (1UL << LVL_SHIFT(n))
158
159/*
160 * The time start value for each level to select the bucket at enqueue
161 * time. We start from the last possible delta of the previous level
162 * so that we can later add an extra LVL_GRAN(n) to n (see calc_index()).
163 */
164#define LVL_START(n) ((LVL_SIZE - 1) << (((n) - 1) * LVL_CLK_SHIFT))
165
166/* Size of each clock level */
167#define LVL_BITS 6
168#define LVL_SIZE (1UL << LVL_BITS)
169#define LVL_MASK (LVL_SIZE - 1)
170#define LVL_OFFS(n) ((n) * LVL_SIZE)
171
172/* Level depth */
173#if HZ > 100
174# define LVL_DEPTH 9
175# else
176# define LVL_DEPTH 8
177#endif
178
179/* The cutoff (max. capacity of the wheel) */
180#define WHEEL_TIMEOUT_CUTOFF (LVL_START(LVL_DEPTH))
181#define WHEEL_TIMEOUT_MAX (WHEEL_TIMEOUT_CUTOFF - LVL_GRAN(LVL_DEPTH - 1))
182
183/*
184 * The resulting wheel size. If NOHZ is configured we allocate two
185 * wheels so we have a separate storage for the deferrable timers.
186 */
187#define WHEEL_SIZE (LVL_SIZE * LVL_DEPTH)
188
189#ifdef CONFIG_NO_HZ_COMMON
190# define NR_BASES 2
191# define BASE_STD 0
192# define BASE_DEF 1
193#else
194# define NR_BASES 1
195# define BASE_STD 0
196# define BASE_DEF 0
197#endif
198
199struct timer_base {
200 raw_spinlock_t lock;
201 struct timer_list *running_timer;
202#ifdef CONFIG_PREEMPT_RT
203 spinlock_t expiry_lock;
204 atomic_t timer_waiters;
205#endif
206 unsigned long clk;
207 unsigned long next_expiry;
208 unsigned int cpu;
209 bool next_expiry_recalc;
210 bool is_idle;
211 bool timers_pending;
212 DECLARE_BITMAP(pending_map, WHEEL_SIZE);
213 struct hlist_head vectors[WHEEL_SIZE];
214} ____cacheline_aligned;
215
216static DEFINE_PER_CPU(struct timer_base, timer_bases[NR_BASES]);
217
218#ifdef CONFIG_NO_HZ_COMMON
219
220static DEFINE_STATIC_KEY_FALSE(timers_nohz_active);
221static DEFINE_MUTEX(timer_keys_mutex);
222
223static void timer_update_keys(struct work_struct *work);
224static DECLARE_WORK(timer_update_work, timer_update_keys);
225
226#ifdef CONFIG_SMP
227static unsigned int sysctl_timer_migration = 1;
228
229DEFINE_STATIC_KEY_FALSE(timers_migration_enabled);
230
231static void timers_update_migration(void)
232{
233 if (sysctl_timer_migration && tick_nohz_active)
234 static_branch_enable(&timers_migration_enabled);
235 else
236 static_branch_disable(&timers_migration_enabled);
237}
238
239#ifdef CONFIG_SYSCTL
240static int timer_migration_handler(struct ctl_table *table, int write,
241 void *buffer, size_t *lenp, loff_t *ppos)
242{
243 int ret;
244
245 mutex_lock(&timer_keys_mutex);
246 ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
247 if (!ret && write)
248 timers_update_migration();
249 mutex_unlock(&timer_keys_mutex);
250 return ret;
251}
252
253static struct ctl_table timer_sysctl[] = {
254 {
255 .procname = "timer_migration",
256 .data = &sysctl_timer_migration,
257 .maxlen = sizeof(unsigned int),
258 .mode = 0644,
259 .proc_handler = timer_migration_handler,
260 .extra1 = SYSCTL_ZERO,
261 .extra2 = SYSCTL_ONE,
262 },
263 {}
264};
265
266static int __init timer_sysctl_init(void)
267{
268 register_sysctl("kernel", timer_sysctl);
269 return 0;
270}
271device_initcall(timer_sysctl_init);
272#endif /* CONFIG_SYSCTL */
273#else /* CONFIG_SMP */
274static inline void timers_update_migration(void) { }
275#endif /* !CONFIG_SMP */
276
277static void timer_update_keys(struct work_struct *work)
278{
279 mutex_lock(&timer_keys_mutex);
280 timers_update_migration();
281 static_branch_enable(&timers_nohz_active);
282 mutex_unlock(&timer_keys_mutex);
283}
284
285void timers_update_nohz(void)
286{
287 schedule_work(&timer_update_work);
288}
289
290static inline bool is_timers_nohz_active(void)
291{
292 return static_branch_unlikely(&timers_nohz_active);
293}
294#else
295static inline bool is_timers_nohz_active(void) { return false; }
296#endif /* NO_HZ_COMMON */
297
298static unsigned long round_jiffies_common(unsigned long j, int cpu,
299 bool force_up)
300{
301 int rem;
302 unsigned long original = j;
303
304 /*
305 * We don't want all cpus firing their timers at once hitting the
306 * same lock or cachelines, so we skew each extra cpu with an extra
307 * 3 jiffies. This 3 jiffies came originally from the mm/ code which
308 * already did this.
309 * The skew is done by adding 3*cpunr, then round, then subtract this
310 * extra offset again.
311 */
312 j += cpu * 3;
313
314 rem = j % HZ;
315
316 /*
317 * If the target jiffie is just after a whole second (which can happen
318 * due to delays of the timer irq, long irq off times etc etc) then
319 * we should round down to the whole second, not up. Use 1/4th second
320 * as cutoff for this rounding as an extreme upper bound for this.
321 * But never round down if @force_up is set.
322 */
323 if (rem < HZ/4 && !force_up) /* round down */
324 j = j - rem;
325 else /* round up */
326 j = j - rem + HZ;
327
328 /* now that we have rounded, subtract the extra skew again */
329 j -= cpu * 3;
330
331 /*
332 * Make sure j is still in the future. Otherwise return the
333 * unmodified value.
334 */
335 return time_is_after_jiffies(j) ? j : original;
336}
337
338/**
339 * __round_jiffies - function to round jiffies to a full second
340 * @j: the time in (absolute) jiffies that should be rounded
341 * @cpu: the processor number on which the timeout will happen
342 *
343 * __round_jiffies() rounds an absolute time in the future (in jiffies)
344 * up or down to (approximately) full seconds. This is useful for timers
345 * for which the exact time they fire does not matter too much, as long as
346 * they fire approximately every X seconds.
347 *
348 * By rounding these timers to whole seconds, all such timers will fire
349 * at the same time, rather than at various times spread out. The goal
350 * of this is to have the CPU wake up less, which saves power.
351 *
352 * The exact rounding is skewed for each processor to avoid all
353 * processors firing at the exact same time, which could lead
354 * to lock contention or spurious cache line bouncing.
355 *
356 * The return value is the rounded version of the @j parameter.
357 */
358unsigned long __round_jiffies(unsigned long j, int cpu)
359{
360 return round_jiffies_common(j, cpu, false);
361}
362EXPORT_SYMBOL_GPL(__round_jiffies);
363
364/**
365 * __round_jiffies_relative - function to round jiffies to a full second
366 * @j: the time in (relative) jiffies that should be rounded
367 * @cpu: the processor number on which the timeout will happen
368 *
369 * __round_jiffies_relative() rounds a time delta in the future (in jiffies)
370 * up or down to (approximately) full seconds. This is useful for timers
371 * for which the exact time they fire does not matter too much, as long as
372 * they fire approximately every X seconds.
373 *
374 * By rounding these timers to whole seconds, all such timers will fire
375 * at the same time, rather than at various times spread out. The goal
376 * of this is to have the CPU wake up less, which saves power.
377 *
378 * The exact rounding is skewed for each processor to avoid all
379 * processors firing at the exact same time, which could lead
380 * to lock contention or spurious cache line bouncing.
381 *
382 * The return value is the rounded version of the @j parameter.
383 */
384unsigned long __round_jiffies_relative(unsigned long j, int cpu)
385{
386 unsigned long j0 = jiffies;
387
388 /* Use j0 because jiffies might change while we run */
389 return round_jiffies_common(j + j0, cpu, false) - j0;
390}
391EXPORT_SYMBOL_GPL(__round_jiffies_relative);
392
393/**
394 * round_jiffies - function to round jiffies to a full second
395 * @j: the time in (absolute) jiffies that should be rounded
396 *
397 * round_jiffies() rounds an absolute time in the future (in jiffies)
398 * up or down to (approximately) full seconds. This is useful for timers
399 * for which the exact time they fire does not matter too much, as long as
400 * they fire approximately every X seconds.
401 *
402 * By rounding these timers to whole seconds, all such timers will fire
403 * at the same time, rather than at various times spread out. The goal
404 * of this is to have the CPU wake up less, which saves power.
405 *
406 * The return value is the rounded version of the @j parameter.
407 */
408unsigned long round_jiffies(unsigned long j)
409{
410 return round_jiffies_common(j, raw_smp_processor_id(), false);
411}
412EXPORT_SYMBOL_GPL(round_jiffies);
413
414/**
415 * round_jiffies_relative - function to round jiffies to a full second
416 * @j: the time in (relative) jiffies that should be rounded
417 *
418 * round_jiffies_relative() rounds a time delta in the future (in jiffies)
419 * up or down to (approximately) full seconds. This is useful for timers
420 * for which the exact time they fire does not matter too much, as long as
421 * they fire approximately every X seconds.
422 *
423 * By rounding these timers to whole seconds, all such timers will fire
424 * at the same time, rather than at various times spread out. The goal
425 * of this is to have the CPU wake up less, which saves power.
426 *
427 * The return value is the rounded version of the @j parameter.
428 */
429unsigned long round_jiffies_relative(unsigned long j)
430{
431 return __round_jiffies_relative(j, raw_smp_processor_id());
432}
433EXPORT_SYMBOL_GPL(round_jiffies_relative);
434
435/**
436 * __round_jiffies_up - function to round jiffies up to a full second
437 * @j: the time in (absolute) jiffies that should be rounded
438 * @cpu: the processor number on which the timeout will happen
439 *
440 * This is the same as __round_jiffies() except that it will never
441 * round down. This is useful for timeouts for which the exact time
442 * of firing does not matter too much, as long as they don't fire too
443 * early.
444 */
445unsigned long __round_jiffies_up(unsigned long j, int cpu)
446{
447 return round_jiffies_common(j, cpu, true);
448}
449EXPORT_SYMBOL_GPL(__round_jiffies_up);
450
451/**
452 * __round_jiffies_up_relative - function to round jiffies up to a full second
453 * @j: the time in (relative) jiffies that should be rounded
454 * @cpu: the processor number on which the timeout will happen
455 *
456 * This is the same as __round_jiffies_relative() except that it will never
457 * round down. This is useful for timeouts for which the exact time
458 * of firing does not matter too much, as long as they don't fire too
459 * early.
460 */
461unsigned long __round_jiffies_up_relative(unsigned long j, int cpu)
462{
463 unsigned long j0 = jiffies;
464
465 /* Use j0 because jiffies might change while we run */
466 return round_jiffies_common(j + j0, cpu, true) - j0;
467}
468EXPORT_SYMBOL_GPL(__round_jiffies_up_relative);
469
470/**
471 * round_jiffies_up - function to round jiffies up to a full second
472 * @j: the time in (absolute) jiffies that should be rounded
473 *
474 * This is the same as round_jiffies() except that it will never
475 * round down. This is useful for timeouts for which the exact time
476 * of firing does not matter too much, as long as they don't fire too
477 * early.
478 */
479unsigned long round_jiffies_up(unsigned long j)
480{
481 return round_jiffies_common(j, raw_smp_processor_id(), true);
482}
483EXPORT_SYMBOL_GPL(round_jiffies_up);
484
485/**
486 * round_jiffies_up_relative - function to round jiffies up to a full second
487 * @j: the time in (relative) jiffies that should be rounded
488 *
489 * This is the same as round_jiffies_relative() except that it will never
490 * round down. This is useful for timeouts for which the exact time
491 * of firing does not matter too much, as long as they don't fire too
492 * early.
493 */
494unsigned long round_jiffies_up_relative(unsigned long j)
495{
496 return __round_jiffies_up_relative(j, raw_smp_processor_id());
497}
498EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
499
500
501static inline unsigned int timer_get_idx(struct timer_list *timer)
502{
503 return (timer->flags & TIMER_ARRAYMASK) >> TIMER_ARRAYSHIFT;
504}
505
506static inline void timer_set_idx(struct timer_list *timer, unsigned int idx)
507{
508 timer->flags = (timer->flags & ~TIMER_ARRAYMASK) |
509 idx << TIMER_ARRAYSHIFT;
510}
511
512/*
513 * Helper function to calculate the array index for a given expiry
514 * time.
515 */
516static inline unsigned calc_index(unsigned long expires, unsigned lvl,
517 unsigned long *bucket_expiry)
518{
519
520 /*
521 * The timer wheel has to guarantee that a timer does not fire
522 * early. Early expiry can happen due to:
523 * - Timer is armed at the edge of a tick
524 * - Truncation of the expiry time in the outer wheel levels
525 *
526 * Round up with level granularity to prevent this.
527 */
528 expires = (expires >> LVL_SHIFT(lvl)) + 1;
529 *bucket_expiry = expires << LVL_SHIFT(lvl);
530 return LVL_OFFS(lvl) + (expires & LVL_MASK);
531}
532
533static int calc_wheel_index(unsigned long expires, unsigned long clk,
534 unsigned long *bucket_expiry)
535{
536 unsigned long delta = expires - clk;
537 unsigned int idx;
538
539 if (delta < LVL_START(1)) {
540 idx = calc_index(expires, 0, bucket_expiry);
541 } else if (delta < LVL_START(2)) {
542 idx = calc_index(expires, 1, bucket_expiry);
543 } else if (delta < LVL_START(3)) {
544 idx = calc_index(expires, 2, bucket_expiry);
545 } else if (delta < LVL_START(4)) {
546 idx = calc_index(expires, 3, bucket_expiry);
547 } else if (delta < LVL_START(5)) {
548 idx = calc_index(expires, 4, bucket_expiry);
549 } else if (delta < LVL_START(6)) {
550 idx = calc_index(expires, 5, bucket_expiry);
551 } else if (delta < LVL_START(7)) {
552 idx = calc_index(expires, 6, bucket_expiry);
553 } else if (LVL_DEPTH > 8 && delta < LVL_START(8)) {
554 idx = calc_index(expires, 7, bucket_expiry);
555 } else if ((long) delta < 0) {
556 idx = clk & LVL_MASK;
557 *bucket_expiry = clk;
558 } else {
559 /*
560 * Force expire obscene large timeouts to expire at the
561 * capacity limit of the wheel.
562 */
563 if (delta >= WHEEL_TIMEOUT_CUTOFF)
564 expires = clk + WHEEL_TIMEOUT_MAX;
565
566 idx = calc_index(expires, LVL_DEPTH - 1, bucket_expiry);
567 }
568 return idx;
569}
570
571static void
572trigger_dyntick_cpu(struct timer_base *base, struct timer_list *timer)
573{
574 /*
575 * Deferrable timers do not prevent the CPU from entering dynticks and
576 * are not taken into account on the idle/nohz_full path. An IPI when a
577 * new deferrable timer is enqueued will wake up the remote CPU but
578 * nothing will be done with the deferrable timer base. Therefore skip
579 * the remote IPI for deferrable timers completely.
580 */
581 if (!is_timers_nohz_active() || timer->flags & TIMER_DEFERRABLE)
582 return;
583
584 /*
585 * We might have to IPI the remote CPU if the base is idle and the
586 * timer is not deferrable. If the other CPU is on the way to idle
587 * then it can't set base->is_idle as we hold the base lock:
588 */
589 if (base->is_idle)
590 wake_up_nohz_cpu(base->cpu);
591}
592
593/*
594 * Enqueue the timer into the hash bucket, mark it pending in
595 * the bitmap, store the index in the timer flags then wake up
596 * the target CPU if needed.
597 */
598static void enqueue_timer(struct timer_base *base, struct timer_list *timer,
599 unsigned int idx, unsigned long bucket_expiry)
600{
601
602 hlist_add_head(&timer->entry, base->vectors + idx);
603 __set_bit(idx, base->pending_map);
604 timer_set_idx(timer, idx);
605
606 trace_timer_start(timer, bucket_expiry);
607
608 /*
609 * Check whether this is the new first expiring timer. The
610 * effective expiry time of the timer is required here
611 * (bucket_expiry) instead of timer->expires.
612 */
613 if (time_before(bucket_expiry, base->next_expiry)) {
614 /*
615 * Set the next expiry time and kick the CPU so it
616 * can reevaluate the wheel:
617 */
618 base->next_expiry = bucket_expiry;
619 base->timers_pending = true;
620 base->next_expiry_recalc = false;
621 trigger_dyntick_cpu(base, timer);
622 }
623}
624
625static void internal_add_timer(struct timer_base *base, struct timer_list *timer)
626{
627 unsigned long bucket_expiry;
628 unsigned int idx;
629
630 idx = calc_wheel_index(timer->expires, base->clk, &bucket_expiry);
631 enqueue_timer(base, timer, idx, bucket_expiry);
632}
633
634#ifdef CONFIG_DEBUG_OBJECTS_TIMERS
635
636static const struct debug_obj_descr timer_debug_descr;
637
638struct timer_hint {
639 void (*function)(struct timer_list *t);
640 long offset;
641};
642
643#define TIMER_HINT(fn, container, timr, hintfn) \
644 { \
645 .function = fn, \
646 .offset = offsetof(container, hintfn) - \
647 offsetof(container, timr) \
648 }
649
650static const struct timer_hint timer_hints[] = {
651 TIMER_HINT(delayed_work_timer_fn,
652 struct delayed_work, timer, work.func),
653 TIMER_HINT(kthread_delayed_work_timer_fn,
654 struct kthread_delayed_work, timer, work.func),
655};
656
657static void *timer_debug_hint(void *addr)
658{
659 struct timer_list *timer = addr;
660 int i;
661
662 for (i = 0; i < ARRAY_SIZE(timer_hints); i++) {
663 if (timer_hints[i].function == timer->function) {
664 void (**fn)(void) = addr + timer_hints[i].offset;
665
666 return *fn;
667 }
668 }
669
670 return timer->function;
671}
672
673static bool timer_is_static_object(void *addr)
674{
675 struct timer_list *timer = addr;
676
677 return (timer->entry.pprev == NULL &&
678 timer->entry.next == TIMER_ENTRY_STATIC);
679}
680
681/*
682 * fixup_init is called when:
683 * - an active object is initialized
684 */
685static bool timer_fixup_init(void *addr, enum debug_obj_state state)
686{
687 struct timer_list *timer = addr;
688
689 switch (state) {
690 case ODEBUG_STATE_ACTIVE:
691 del_timer_sync(timer);
692 debug_object_init(timer, &timer_debug_descr);
693 return true;
694 default:
695 return false;
696 }
697}
698
699/* Stub timer callback for improperly used timers. */
700static void stub_timer(struct timer_list *unused)
701{
702 WARN_ON(1);
703}
704
705/*
706 * fixup_activate is called when:
707 * - an active object is activated
708 * - an unknown non-static object is activated
709 */
710static bool timer_fixup_activate(void *addr, enum debug_obj_state state)
711{
712 struct timer_list *timer = addr;
713
714 switch (state) {
715 case ODEBUG_STATE_NOTAVAILABLE:
716 timer_setup(timer, stub_timer, 0);
717 return true;
718
719 case ODEBUG_STATE_ACTIVE:
720 WARN_ON(1);
721 fallthrough;
722 default:
723 return false;
724 }
725}
726
727/*
728 * fixup_free is called when:
729 * - an active object is freed
730 */
731static bool timer_fixup_free(void *addr, enum debug_obj_state state)
732{
733 struct timer_list *timer = addr;
734
735 switch (state) {
736 case ODEBUG_STATE_ACTIVE:
737 del_timer_sync(timer);
738 debug_object_free(timer, &timer_debug_descr);
739 return true;
740 default:
741 return false;
742 }
743}
744
745/*
746 * fixup_assert_init is called when:
747 * - an untracked/uninit-ed object is found
748 */
749static bool timer_fixup_assert_init(void *addr, enum debug_obj_state state)
750{
751 struct timer_list *timer = addr;
752
753 switch (state) {
754 case ODEBUG_STATE_NOTAVAILABLE:
755 timer_setup(timer, stub_timer, 0);
756 return true;
757 default:
758 return false;
759 }
760}
761
762static const struct debug_obj_descr timer_debug_descr = {
763 .name = "timer_list",
764 .debug_hint = timer_debug_hint,
765 .is_static_object = timer_is_static_object,
766 .fixup_init = timer_fixup_init,
767 .fixup_activate = timer_fixup_activate,
768 .fixup_free = timer_fixup_free,
769 .fixup_assert_init = timer_fixup_assert_init,
770};
771
772static inline void debug_timer_init(struct timer_list *timer)
773{
774 debug_object_init(timer, &timer_debug_descr);
775}
776
777static inline void debug_timer_activate(struct timer_list *timer)
778{
779 debug_object_activate(timer, &timer_debug_descr);
780}
781
782static inline void debug_timer_deactivate(struct timer_list *timer)
783{
784 debug_object_deactivate(timer, &timer_debug_descr);
785}
786
787static inline void debug_timer_assert_init(struct timer_list *timer)
788{
789 debug_object_assert_init(timer, &timer_debug_descr);
790}
791
792static void do_init_timer(struct timer_list *timer,
793 void (*func)(struct timer_list *),
794 unsigned int flags,
795 const char *name, struct lock_class_key *key);
796
797void init_timer_on_stack_key(struct timer_list *timer,
798 void (*func)(struct timer_list *),
799 unsigned int flags,
800 const char *name, struct lock_class_key *key)
801{
802 debug_object_init_on_stack(timer, &timer_debug_descr);
803 do_init_timer(timer, func, flags, name, key);
804}
805EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
806
807void destroy_timer_on_stack(struct timer_list *timer)
808{
809 debug_object_free(timer, &timer_debug_descr);
810}
811EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
812
813#else
814static inline void debug_timer_init(struct timer_list *timer) { }
815static inline void debug_timer_activate(struct timer_list *timer) { }
816static inline void debug_timer_deactivate(struct timer_list *timer) { }
817static inline void debug_timer_assert_init(struct timer_list *timer) { }
818#endif
819
820static inline void debug_init(struct timer_list *timer)
821{
822 debug_timer_init(timer);
823 trace_timer_init(timer);
824}
825
826static inline void debug_deactivate(struct timer_list *timer)
827{
828 debug_timer_deactivate(timer);
829 trace_timer_cancel(timer);
830}
831
832static inline void debug_assert_init(struct timer_list *timer)
833{
834 debug_timer_assert_init(timer);
835}
836
837static void do_init_timer(struct timer_list *timer,
838 void (*func)(struct timer_list *),
839 unsigned int flags,
840 const char *name, struct lock_class_key *key)
841{
842 timer->entry.pprev = NULL;
843 timer->function = func;
844 if (WARN_ON_ONCE(flags & ~TIMER_INIT_FLAGS))
845 flags &= TIMER_INIT_FLAGS;
846 timer->flags = flags | raw_smp_processor_id();
847 lockdep_init_map(&timer->lockdep_map, name, key, 0);
848}
849
850/**
851 * init_timer_key - initialize a timer
852 * @timer: the timer to be initialized
853 * @func: timer callback function
854 * @flags: timer flags
855 * @name: name of the timer
856 * @key: lockdep class key of the fake lock used for tracking timer
857 * sync lock dependencies
858 *
859 * init_timer_key() must be done to a timer prior calling *any* of the
860 * other timer functions.
861 */
862void init_timer_key(struct timer_list *timer,
863 void (*func)(struct timer_list *), unsigned int flags,
864 const char *name, struct lock_class_key *key)
865{
866 debug_init(timer);
867 do_init_timer(timer, func, flags, name, key);
868}
869EXPORT_SYMBOL(init_timer_key);
870
871static inline void detach_timer(struct timer_list *timer, bool clear_pending)
872{
873 struct hlist_node *entry = &timer->entry;
874
875 debug_deactivate(timer);
876
877 __hlist_del(entry);
878 if (clear_pending)
879 entry->pprev = NULL;
880 entry->next = LIST_POISON2;
881}
882
883static int detach_if_pending(struct timer_list *timer, struct timer_base *base,
884 bool clear_pending)
885{
886 unsigned idx = timer_get_idx(timer);
887
888 if (!timer_pending(timer))
889 return 0;
890
891 if (hlist_is_singular_node(&timer->entry, base->vectors + idx)) {
892 __clear_bit(idx, base->pending_map);
893 base->next_expiry_recalc = true;
894 }
895
896 detach_timer(timer, clear_pending);
897 return 1;
898}
899
900static inline struct timer_base *get_timer_cpu_base(u32 tflags, u32 cpu)
901{
902 struct timer_base *base = per_cpu_ptr(&timer_bases[BASE_STD], cpu);
903
904 /*
905 * If the timer is deferrable and NO_HZ_COMMON is set then we need
906 * to use the deferrable base.
907 */
908 if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE))
909 base = per_cpu_ptr(&timer_bases[BASE_DEF], cpu);
910 return base;
911}
912
913static inline struct timer_base *get_timer_this_cpu_base(u32 tflags)
914{
915 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
916
917 /*
918 * If the timer is deferrable and NO_HZ_COMMON is set then we need
919 * to use the deferrable base.
920 */
921 if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE))
922 base = this_cpu_ptr(&timer_bases[BASE_DEF]);
923 return base;
924}
925
926static inline struct timer_base *get_timer_base(u32 tflags)
927{
928 return get_timer_cpu_base(tflags, tflags & TIMER_CPUMASK);
929}
930
931static inline struct timer_base *
932get_target_base(struct timer_base *base, unsigned tflags)
933{
934#if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
935 if (static_branch_likely(&timers_migration_enabled) &&
936 !(tflags & TIMER_PINNED))
937 return get_timer_cpu_base(tflags, get_nohz_timer_target());
938#endif
939 return get_timer_this_cpu_base(tflags);
940}
941
942static inline void __forward_timer_base(struct timer_base *base,
943 unsigned long basej)
944{
945 /*
946 * Check whether we can forward the base. We can only do that when
947 * @basej is past base->clk otherwise we might rewind base->clk.
948 */
949 if (time_before_eq(basej, base->clk))
950 return;
951
952 /*
953 * If the next expiry value is > jiffies, then we fast forward to
954 * jiffies otherwise we forward to the next expiry value.
955 */
956 if (time_after(base->next_expiry, basej)) {
957 base->clk = basej;
958 } else {
959 if (WARN_ON_ONCE(time_before(base->next_expiry, base->clk)))
960 return;
961 base->clk = base->next_expiry;
962 }
963
964}
965
966static inline void forward_timer_base(struct timer_base *base)
967{
968 __forward_timer_base(base, READ_ONCE(jiffies));
969}
970
971/*
972 * We are using hashed locking: Holding per_cpu(timer_bases[x]).lock means
973 * that all timers which are tied to this base are locked, and the base itself
974 * is locked too.
975 *
976 * So __run_timers/migrate_timers can safely modify all timers which could
977 * be found in the base->vectors array.
978 *
979 * When a timer is migrating then the TIMER_MIGRATING flag is set and we need
980 * to wait until the migration is done.
981 */
982static struct timer_base *lock_timer_base(struct timer_list *timer,
983 unsigned long *flags)
984 __acquires(timer->base->lock)
985{
986 for (;;) {
987 struct timer_base *base;
988 u32 tf;
989
990 /*
991 * We need to use READ_ONCE() here, otherwise the compiler
992 * might re-read @tf between the check for TIMER_MIGRATING
993 * and spin_lock().
994 */
995 tf = READ_ONCE(timer->flags);
996
997 if (!(tf & TIMER_MIGRATING)) {
998 base = get_timer_base(tf);
999 raw_spin_lock_irqsave(&base->lock, *flags);
1000 if (timer->flags == tf)
1001 return base;
1002 raw_spin_unlock_irqrestore(&base->lock, *flags);
1003 }
1004 cpu_relax();
1005 }
1006}
1007
1008#define MOD_TIMER_PENDING_ONLY 0x01
1009#define MOD_TIMER_REDUCE 0x02
1010#define MOD_TIMER_NOTPENDING 0x04
1011
1012static inline int
1013__mod_timer(struct timer_list *timer, unsigned long expires, unsigned int options)
1014{
1015 unsigned long clk = 0, flags, bucket_expiry;
1016 struct timer_base *base, *new_base;
1017 unsigned int idx = UINT_MAX;
1018 int ret = 0;
1019
1020 debug_assert_init(timer);
1021
1022 /*
1023 * This is a common optimization triggered by the networking code - if
1024 * the timer is re-modified to have the same timeout or ends up in the
1025 * same array bucket then just return:
1026 */
1027 if (!(options & MOD_TIMER_NOTPENDING) && timer_pending(timer)) {
1028 /*
1029 * The downside of this optimization is that it can result in
1030 * larger granularity than you would get from adding a new
1031 * timer with this expiry.
1032 */
1033 long diff = timer->expires - expires;
1034
1035 if (!diff)
1036 return 1;
1037 if (options & MOD_TIMER_REDUCE && diff <= 0)
1038 return 1;
1039
1040 /*
1041 * We lock timer base and calculate the bucket index right
1042 * here. If the timer ends up in the same bucket, then we
1043 * just update the expiry time and avoid the whole
1044 * dequeue/enqueue dance.
1045 */
1046 base = lock_timer_base(timer, &flags);
1047 /*
1048 * Has @timer been shutdown? This needs to be evaluated
1049 * while holding base lock to prevent a race against the
1050 * shutdown code.
1051 */
1052 if (!timer->function)
1053 goto out_unlock;
1054
1055 forward_timer_base(base);
1056
1057 if (timer_pending(timer) && (options & MOD_TIMER_REDUCE) &&
1058 time_before_eq(timer->expires, expires)) {
1059 ret = 1;
1060 goto out_unlock;
1061 }
1062
1063 clk = base->clk;
1064 idx = calc_wheel_index(expires, clk, &bucket_expiry);
1065
1066 /*
1067 * Retrieve and compare the array index of the pending
1068 * timer. If it matches set the expiry to the new value so a
1069 * subsequent call will exit in the expires check above.
1070 */
1071 if (idx == timer_get_idx(timer)) {
1072 if (!(options & MOD_TIMER_REDUCE))
1073 timer->expires = expires;
1074 else if (time_after(timer->expires, expires))
1075 timer->expires = expires;
1076 ret = 1;
1077 goto out_unlock;
1078 }
1079 } else {
1080 base = lock_timer_base(timer, &flags);
1081 /*
1082 * Has @timer been shutdown? This needs to be evaluated
1083 * while holding base lock to prevent a race against the
1084 * shutdown code.
1085 */
1086 if (!timer->function)
1087 goto out_unlock;
1088
1089 forward_timer_base(base);
1090 }
1091
1092 ret = detach_if_pending(timer, base, false);
1093 if (!ret && (options & MOD_TIMER_PENDING_ONLY))
1094 goto out_unlock;
1095
1096 new_base = get_target_base(base, timer->flags);
1097
1098 if (base != new_base) {
1099 /*
1100 * We are trying to schedule the timer on the new base.
1101 * However we can't change timer's base while it is running,
1102 * otherwise timer_delete_sync() can't detect that the timer's
1103 * handler yet has not finished. This also guarantees that the
1104 * timer is serialized wrt itself.
1105 */
1106 if (likely(base->running_timer != timer)) {
1107 /* See the comment in lock_timer_base() */
1108 timer->flags |= TIMER_MIGRATING;
1109
1110 raw_spin_unlock(&base->lock);
1111 base = new_base;
1112 raw_spin_lock(&base->lock);
1113 WRITE_ONCE(timer->flags,
1114 (timer->flags & ~TIMER_BASEMASK) | base->cpu);
1115 forward_timer_base(base);
1116 }
1117 }
1118
1119 debug_timer_activate(timer);
1120
1121 timer->expires = expires;
1122 /*
1123 * If 'idx' was calculated above and the base time did not advance
1124 * between calculating 'idx' and possibly switching the base, only
1125 * enqueue_timer() is required. Otherwise we need to (re)calculate
1126 * the wheel index via internal_add_timer().
1127 */
1128 if (idx != UINT_MAX && clk == base->clk)
1129 enqueue_timer(base, timer, idx, bucket_expiry);
1130 else
1131 internal_add_timer(base, timer);
1132
1133out_unlock:
1134 raw_spin_unlock_irqrestore(&base->lock, flags);
1135
1136 return ret;
1137}
1138
1139/**
1140 * mod_timer_pending - Modify a pending timer's timeout
1141 * @timer: The pending timer to be modified
1142 * @expires: New absolute timeout in jiffies
1143 *
1144 * mod_timer_pending() is the same for pending timers as mod_timer(), but
1145 * will not activate inactive timers.
1146 *
1147 * If @timer->function == NULL then the start operation is silently
1148 * discarded.
1149 *
1150 * Return:
1151 * * %0 - The timer was inactive and not modified or was in
1152 * shutdown state and the operation was discarded
1153 * * %1 - The timer was active and requeued to expire at @expires
1154 */
1155int mod_timer_pending(struct timer_list *timer, unsigned long expires)
1156{
1157 return __mod_timer(timer, expires, MOD_TIMER_PENDING_ONLY);
1158}
1159EXPORT_SYMBOL(mod_timer_pending);
1160
1161/**
1162 * mod_timer - Modify a timer's timeout
1163 * @timer: The timer to be modified
1164 * @expires: New absolute timeout in jiffies
1165 *
1166 * mod_timer(timer, expires) is equivalent to:
1167 *
1168 * del_timer(timer); timer->expires = expires; add_timer(timer);
1169 *
1170 * mod_timer() is more efficient than the above open coded sequence. In
1171 * case that the timer is inactive, the del_timer() part is a NOP. The
1172 * timer is in any case activated with the new expiry time @expires.
1173 *
1174 * Note that if there are multiple unserialized concurrent users of the
1175 * same timer, then mod_timer() is the only safe way to modify the timeout,
1176 * since add_timer() cannot modify an already running timer.
1177 *
1178 * If @timer->function == NULL then the start operation is silently
1179 * discarded. In this case the return value is 0 and meaningless.
1180 *
1181 * Return:
1182 * * %0 - The timer was inactive and started or was in shutdown
1183 * state and the operation was discarded
1184 * * %1 - The timer was active and requeued to expire at @expires or
1185 * the timer was active and not modified because @expires did
1186 * not change the effective expiry time
1187 */
1188int mod_timer(struct timer_list *timer, unsigned long expires)
1189{
1190 return __mod_timer(timer, expires, 0);
1191}
1192EXPORT_SYMBOL(mod_timer);
1193
1194/**
1195 * timer_reduce - Modify a timer's timeout if it would reduce the timeout
1196 * @timer: The timer to be modified
1197 * @expires: New absolute timeout in jiffies
1198 *
1199 * timer_reduce() is very similar to mod_timer(), except that it will only
1200 * modify an enqueued timer if that would reduce the expiration time. If
1201 * @timer is not enqueued it starts the timer.
1202 *
1203 * If @timer->function == NULL then the start operation is silently
1204 * discarded.
1205 *
1206 * Return:
1207 * * %0 - The timer was inactive and started or was in shutdown
1208 * state and the operation was discarded
1209 * * %1 - The timer was active and requeued to expire at @expires or
1210 * the timer was active and not modified because @expires
1211 * did not change the effective expiry time such that the
1212 * timer would expire earlier than already scheduled
1213 */
1214int timer_reduce(struct timer_list *timer, unsigned long expires)
1215{
1216 return __mod_timer(timer, expires, MOD_TIMER_REDUCE);
1217}
1218EXPORT_SYMBOL(timer_reduce);
1219
1220/**
1221 * add_timer - Start a timer
1222 * @timer: The timer to be started
1223 *
1224 * Start @timer to expire at @timer->expires in the future. @timer->expires
1225 * is the absolute expiry time measured in 'jiffies'. When the timer expires
1226 * timer->function(timer) will be invoked from soft interrupt context.
1227 *
1228 * The @timer->expires and @timer->function fields must be set prior
1229 * to calling this function.
1230 *
1231 * If @timer->function == NULL then the start operation is silently
1232 * discarded.
1233 *
1234 * If @timer->expires is already in the past @timer will be queued to
1235 * expire at the next timer tick.
1236 *
1237 * This can only operate on an inactive timer. Attempts to invoke this on
1238 * an active timer are rejected with a warning.
1239 */
1240void add_timer(struct timer_list *timer)
1241{
1242 if (WARN_ON_ONCE(timer_pending(timer)))
1243 return;
1244 __mod_timer(timer, timer->expires, MOD_TIMER_NOTPENDING);
1245}
1246EXPORT_SYMBOL(add_timer);
1247
1248/**
1249 * add_timer_on - Start a timer on a particular CPU
1250 * @timer: The timer to be started
1251 * @cpu: The CPU to start it on
1252 *
1253 * Same as add_timer() except that it starts the timer on the given CPU.
1254 *
1255 * See add_timer() for further details.
1256 */
1257void add_timer_on(struct timer_list *timer, int cpu)
1258{
1259 struct timer_base *new_base, *base;
1260 unsigned long flags;
1261
1262 debug_assert_init(timer);
1263
1264 if (WARN_ON_ONCE(timer_pending(timer)))
1265 return;
1266
1267 new_base = get_timer_cpu_base(timer->flags, cpu);
1268
1269 /*
1270 * If @timer was on a different CPU, it should be migrated with the
1271 * old base locked to prevent other operations proceeding with the
1272 * wrong base locked. See lock_timer_base().
1273 */
1274 base = lock_timer_base(timer, &flags);
1275 /*
1276 * Has @timer been shutdown? This needs to be evaluated while
1277 * holding base lock to prevent a race against the shutdown code.
1278 */
1279 if (!timer->function)
1280 goto out_unlock;
1281
1282 if (base != new_base) {
1283 timer->flags |= TIMER_MIGRATING;
1284
1285 raw_spin_unlock(&base->lock);
1286 base = new_base;
1287 raw_spin_lock(&base->lock);
1288 WRITE_ONCE(timer->flags,
1289 (timer->flags & ~TIMER_BASEMASK) | cpu);
1290 }
1291 forward_timer_base(base);
1292
1293 debug_timer_activate(timer);
1294 internal_add_timer(base, timer);
1295out_unlock:
1296 raw_spin_unlock_irqrestore(&base->lock, flags);
1297}
1298EXPORT_SYMBOL_GPL(add_timer_on);
1299
1300/**
1301 * __timer_delete - Internal function: Deactivate a timer
1302 * @timer: The timer to be deactivated
1303 * @shutdown: If true, this indicates that the timer is about to be
1304 * shutdown permanently.
1305 *
1306 * If @shutdown is true then @timer->function is set to NULL under the
1307 * timer base lock which prevents further rearming of the time. In that
1308 * case any attempt to rearm @timer after this function returns will be
1309 * silently ignored.
1310 *
1311 * Return:
1312 * * %0 - The timer was not pending
1313 * * %1 - The timer was pending and deactivated
1314 */
1315static int __timer_delete(struct timer_list *timer, bool shutdown)
1316{
1317 struct timer_base *base;
1318 unsigned long flags;
1319 int ret = 0;
1320
1321 debug_assert_init(timer);
1322
1323 /*
1324 * If @shutdown is set then the lock has to be taken whether the
1325 * timer is pending or not to protect against a concurrent rearm
1326 * which might hit between the lockless pending check and the lock
1327 * aquisition. By taking the lock it is ensured that such a newly
1328 * enqueued timer is dequeued and cannot end up with
1329 * timer->function == NULL in the expiry code.
1330 *
1331 * If timer->function is currently executed, then this makes sure
1332 * that the callback cannot requeue the timer.
1333 */
1334 if (timer_pending(timer) || shutdown) {
1335 base = lock_timer_base(timer, &flags);
1336 ret = detach_if_pending(timer, base, true);
1337 if (shutdown)
1338 timer->function = NULL;
1339 raw_spin_unlock_irqrestore(&base->lock, flags);
1340 }
1341
1342 return ret;
1343}
1344
1345/**
1346 * timer_delete - Deactivate a timer
1347 * @timer: The timer to be deactivated
1348 *
1349 * The function only deactivates a pending timer, but contrary to
1350 * timer_delete_sync() it does not take into account whether the timer's
1351 * callback function is concurrently executed on a different CPU or not.
1352 * It neither prevents rearming of the timer. If @timer can be rearmed
1353 * concurrently then the return value of this function is meaningless.
1354 *
1355 * Return:
1356 * * %0 - The timer was not pending
1357 * * %1 - The timer was pending and deactivated
1358 */
1359int timer_delete(struct timer_list *timer)
1360{
1361 return __timer_delete(timer, false);
1362}
1363EXPORT_SYMBOL(timer_delete);
1364
1365/**
1366 * timer_shutdown - Deactivate a timer and prevent rearming
1367 * @timer: The timer to be deactivated
1368 *
1369 * The function does not wait for an eventually running timer callback on a
1370 * different CPU but it prevents rearming of the timer. Any attempt to arm
1371 * @timer after this function returns will be silently ignored.
1372 *
1373 * This function is useful for teardown code and should only be used when
1374 * timer_shutdown_sync() cannot be invoked due to locking or context constraints.
1375 *
1376 * Return:
1377 * * %0 - The timer was not pending
1378 * * %1 - The timer was pending
1379 */
1380int timer_shutdown(struct timer_list *timer)
1381{
1382 return __timer_delete(timer, true);
1383}
1384EXPORT_SYMBOL_GPL(timer_shutdown);
1385
1386/**
1387 * __try_to_del_timer_sync - Internal function: Try to deactivate a timer
1388 * @timer: Timer to deactivate
1389 * @shutdown: If true, this indicates that the timer is about to be
1390 * shutdown permanently.
1391 *
1392 * If @shutdown is true then @timer->function is set to NULL under the
1393 * timer base lock which prevents further rearming of the timer. Any
1394 * attempt to rearm @timer after this function returns will be silently
1395 * ignored.
1396 *
1397 * This function cannot guarantee that the timer cannot be rearmed
1398 * right after dropping the base lock if @shutdown is false. That
1399 * needs to be prevented by the calling code if necessary.
1400 *
1401 * Return:
1402 * * %0 - The timer was not pending
1403 * * %1 - The timer was pending and deactivated
1404 * * %-1 - The timer callback function is running on a different CPU
1405 */
1406static int __try_to_del_timer_sync(struct timer_list *timer, bool shutdown)
1407{
1408 struct timer_base *base;
1409 unsigned long flags;
1410 int ret = -1;
1411
1412 debug_assert_init(timer);
1413
1414 base = lock_timer_base(timer, &flags);
1415
1416 if (base->running_timer != timer)
1417 ret = detach_if_pending(timer, base, true);
1418 if (shutdown)
1419 timer->function = NULL;
1420
1421 raw_spin_unlock_irqrestore(&base->lock, flags);
1422
1423 return ret;
1424}
1425
1426/**
1427 * try_to_del_timer_sync - Try to deactivate a timer
1428 * @timer: Timer to deactivate
1429 *
1430 * This function tries to deactivate a timer. On success the timer is not
1431 * queued and the timer callback function is not running on any CPU.
1432 *
1433 * This function does not guarantee that the timer cannot be rearmed right
1434 * after dropping the base lock. That needs to be prevented by the calling
1435 * code if necessary.
1436 *
1437 * Return:
1438 * * %0 - The timer was not pending
1439 * * %1 - The timer was pending and deactivated
1440 * * %-1 - The timer callback function is running on a different CPU
1441 */
1442int try_to_del_timer_sync(struct timer_list *timer)
1443{
1444 return __try_to_del_timer_sync(timer, false);
1445}
1446EXPORT_SYMBOL(try_to_del_timer_sync);
1447
1448#ifdef CONFIG_PREEMPT_RT
1449static __init void timer_base_init_expiry_lock(struct timer_base *base)
1450{
1451 spin_lock_init(&base->expiry_lock);
1452}
1453
1454static inline void timer_base_lock_expiry(struct timer_base *base)
1455{
1456 spin_lock(&base->expiry_lock);
1457}
1458
1459static inline void timer_base_unlock_expiry(struct timer_base *base)
1460{
1461 spin_unlock(&base->expiry_lock);
1462}
1463
1464/*
1465 * The counterpart to del_timer_wait_running().
1466 *
1467 * If there is a waiter for base->expiry_lock, then it was waiting for the
1468 * timer callback to finish. Drop expiry_lock and reacquire it. That allows
1469 * the waiter to acquire the lock and make progress.
1470 */
1471static void timer_sync_wait_running(struct timer_base *base)
1472{
1473 if (atomic_read(&base->timer_waiters)) {
1474 raw_spin_unlock_irq(&base->lock);
1475 spin_unlock(&base->expiry_lock);
1476 spin_lock(&base->expiry_lock);
1477 raw_spin_lock_irq(&base->lock);
1478 }
1479}
1480
1481/*
1482 * This function is called on PREEMPT_RT kernels when the fast path
1483 * deletion of a timer failed because the timer callback function was
1484 * running.
1485 *
1486 * This prevents priority inversion, if the softirq thread on a remote CPU
1487 * got preempted, and it prevents a life lock when the task which tries to
1488 * delete a timer preempted the softirq thread running the timer callback
1489 * function.
1490 */
1491static void del_timer_wait_running(struct timer_list *timer)
1492{
1493 u32 tf;
1494
1495 tf = READ_ONCE(timer->flags);
1496 if (!(tf & (TIMER_MIGRATING | TIMER_IRQSAFE))) {
1497 struct timer_base *base = get_timer_base(tf);
1498
1499 /*
1500 * Mark the base as contended and grab the expiry lock,
1501 * which is held by the softirq across the timer
1502 * callback. Drop the lock immediately so the softirq can
1503 * expire the next timer. In theory the timer could already
1504 * be running again, but that's more than unlikely and just
1505 * causes another wait loop.
1506 */
1507 atomic_inc(&base->timer_waiters);
1508 spin_lock_bh(&base->expiry_lock);
1509 atomic_dec(&base->timer_waiters);
1510 spin_unlock_bh(&base->expiry_lock);
1511 }
1512}
1513#else
1514static inline void timer_base_init_expiry_lock(struct timer_base *base) { }
1515static inline void timer_base_lock_expiry(struct timer_base *base) { }
1516static inline void timer_base_unlock_expiry(struct timer_base *base) { }
1517static inline void timer_sync_wait_running(struct timer_base *base) { }
1518static inline void del_timer_wait_running(struct timer_list *timer) { }
1519#endif
1520
1521/**
1522 * __timer_delete_sync - Internal function: Deactivate a timer and wait
1523 * for the handler to finish.
1524 * @timer: The timer to be deactivated
1525 * @shutdown: If true, @timer->function will be set to NULL under the
1526 * timer base lock which prevents rearming of @timer
1527 *
1528 * If @shutdown is not set the timer can be rearmed later. If the timer can
1529 * be rearmed concurrently, i.e. after dropping the base lock then the
1530 * return value is meaningless.
1531 *
1532 * If @shutdown is set then @timer->function is set to NULL under timer
1533 * base lock which prevents rearming of the timer. Any attempt to rearm
1534 * a shutdown timer is silently ignored.
1535 *
1536 * If the timer should be reused after shutdown it has to be initialized
1537 * again.
1538 *
1539 * Return:
1540 * * %0 - The timer was not pending
1541 * * %1 - The timer was pending and deactivated
1542 */
1543static int __timer_delete_sync(struct timer_list *timer, bool shutdown)
1544{
1545 int ret;
1546
1547#ifdef CONFIG_LOCKDEP
1548 unsigned long flags;
1549
1550 /*
1551 * If lockdep gives a backtrace here, please reference
1552 * the synchronization rules above.
1553 */
1554 local_irq_save(flags);
1555 lock_map_acquire(&timer->lockdep_map);
1556 lock_map_release(&timer->lockdep_map);
1557 local_irq_restore(flags);
1558#endif
1559 /*
1560 * don't use it in hardirq context, because it
1561 * could lead to deadlock.
1562 */
1563 WARN_ON(in_hardirq() && !(timer->flags & TIMER_IRQSAFE));
1564
1565 /*
1566 * Must be able to sleep on PREEMPT_RT because of the slowpath in
1567 * del_timer_wait_running().
1568 */
1569 if (IS_ENABLED(CONFIG_PREEMPT_RT) && !(timer->flags & TIMER_IRQSAFE))
1570 lockdep_assert_preemption_enabled();
1571
1572 do {
1573 ret = __try_to_del_timer_sync(timer, shutdown);
1574
1575 if (unlikely(ret < 0)) {
1576 del_timer_wait_running(timer);
1577 cpu_relax();
1578 }
1579 } while (ret < 0);
1580
1581 return ret;
1582}
1583
1584/**
1585 * timer_delete_sync - Deactivate a timer and wait for the handler to finish.
1586 * @timer: The timer to be deactivated
1587 *
1588 * Synchronization rules: Callers must prevent restarting of the timer,
1589 * otherwise this function is meaningless. It must not be called from
1590 * interrupt contexts unless the timer is an irqsafe one. The caller must
1591 * not hold locks which would prevent completion of the timer's callback
1592 * function. The timer's handler must not call add_timer_on(). Upon exit
1593 * the timer is not queued and the handler is not running on any CPU.
1594 *
1595 * For !irqsafe timers, the caller must not hold locks that are held in
1596 * interrupt context. Even if the lock has nothing to do with the timer in
1597 * question. Here's why::
1598 *
1599 * CPU0 CPU1
1600 * ---- ----
1601 * <SOFTIRQ>
1602 * call_timer_fn();
1603 * base->running_timer = mytimer;
1604 * spin_lock_irq(somelock);
1605 * <IRQ>
1606 * spin_lock(somelock);
1607 * timer_delete_sync(mytimer);
1608 * while (base->running_timer == mytimer);
1609 *
1610 * Now timer_delete_sync() will never return and never release somelock.
1611 * The interrupt on the other CPU is waiting to grab somelock but it has
1612 * interrupted the softirq that CPU0 is waiting to finish.
1613 *
1614 * This function cannot guarantee that the timer is not rearmed again by
1615 * some concurrent or preempting code, right after it dropped the base
1616 * lock. If there is the possibility of a concurrent rearm then the return
1617 * value of the function is meaningless.
1618 *
1619 * If such a guarantee is needed, e.g. for teardown situations then use
1620 * timer_shutdown_sync() instead.
1621 *
1622 * Return:
1623 * * %0 - The timer was not pending
1624 * * %1 - The timer was pending and deactivated
1625 */
1626int timer_delete_sync(struct timer_list *timer)
1627{
1628 return __timer_delete_sync(timer, false);
1629}
1630EXPORT_SYMBOL(timer_delete_sync);
1631
1632/**
1633 * timer_shutdown_sync - Shutdown a timer and prevent rearming
1634 * @timer: The timer to be shutdown
1635 *
1636 * When the function returns it is guaranteed that:
1637 * - @timer is not queued
1638 * - The callback function of @timer is not running
1639 * - @timer cannot be enqueued again. Any attempt to rearm
1640 * @timer is silently ignored.
1641 *
1642 * See timer_delete_sync() for synchronization rules.
1643 *
1644 * This function is useful for final teardown of an infrastructure where
1645 * the timer is subject to a circular dependency problem.
1646 *
1647 * A common pattern for this is a timer and a workqueue where the timer can
1648 * schedule work and work can arm the timer. On shutdown the workqueue must
1649 * be destroyed and the timer must be prevented from rearming. Unless the
1650 * code has conditionals like 'if (mything->in_shutdown)' to prevent that
1651 * there is no way to get this correct with timer_delete_sync().
1652 *
1653 * timer_shutdown_sync() is solving the problem. The correct ordering of
1654 * calls in this case is:
1655 *
1656 * timer_shutdown_sync(&mything->timer);
1657 * workqueue_destroy(&mything->workqueue);
1658 *
1659 * After this 'mything' can be safely freed.
1660 *
1661 * This obviously implies that the timer is not required to be functional
1662 * for the rest of the shutdown operation.
1663 *
1664 * Return:
1665 * * %0 - The timer was not pending
1666 * * %1 - The timer was pending
1667 */
1668int timer_shutdown_sync(struct timer_list *timer)
1669{
1670 return __timer_delete_sync(timer, true);
1671}
1672EXPORT_SYMBOL_GPL(timer_shutdown_sync);
1673
1674static void call_timer_fn(struct timer_list *timer,
1675 void (*fn)(struct timer_list *),
1676 unsigned long baseclk)
1677{
1678 int count = preempt_count();
1679
1680#ifdef CONFIG_LOCKDEP
1681 /*
1682 * It is permissible to free the timer from inside the
1683 * function that is called from it, this we need to take into
1684 * account for lockdep too. To avoid bogus "held lock freed"
1685 * warnings as well as problems when looking into
1686 * timer->lockdep_map, make a copy and use that here.
1687 */
1688 struct lockdep_map lockdep_map;
1689
1690 lockdep_copy_map(&lockdep_map, &timer->lockdep_map);
1691#endif
1692 /*
1693 * Couple the lock chain with the lock chain at
1694 * timer_delete_sync() by acquiring the lock_map around the fn()
1695 * call here and in timer_delete_sync().
1696 */
1697 lock_map_acquire(&lockdep_map);
1698
1699 trace_timer_expire_entry(timer, baseclk);
1700 fn(timer);
1701 trace_timer_expire_exit(timer);
1702
1703 lock_map_release(&lockdep_map);
1704
1705 if (count != preempt_count()) {
1706 WARN_ONCE(1, "timer: %pS preempt leak: %08x -> %08x\n",
1707 fn, count, preempt_count());
1708 /*
1709 * Restore the preempt count. That gives us a decent
1710 * chance to survive and extract information. If the
1711 * callback kept a lock held, bad luck, but not worse
1712 * than the BUG() we had.
1713 */
1714 preempt_count_set(count);
1715 }
1716}
1717
1718static void expire_timers(struct timer_base *base, struct hlist_head *head)
1719{
1720 /*
1721 * This value is required only for tracing. base->clk was
1722 * incremented directly before expire_timers was called. But expiry
1723 * is related to the old base->clk value.
1724 */
1725 unsigned long baseclk = base->clk - 1;
1726
1727 while (!hlist_empty(head)) {
1728 struct timer_list *timer;
1729 void (*fn)(struct timer_list *);
1730
1731 timer = hlist_entry(head->first, struct timer_list, entry);
1732
1733 base->running_timer = timer;
1734 detach_timer(timer, true);
1735
1736 fn = timer->function;
1737
1738 if (WARN_ON_ONCE(!fn)) {
1739 /* Should never happen. Emphasis on should! */
1740 base->running_timer = NULL;
1741 continue;
1742 }
1743
1744 if (timer->flags & TIMER_IRQSAFE) {
1745 raw_spin_unlock(&base->lock);
1746 call_timer_fn(timer, fn, baseclk);
1747 raw_spin_lock(&base->lock);
1748 base->running_timer = NULL;
1749 } else {
1750 raw_spin_unlock_irq(&base->lock);
1751 call_timer_fn(timer, fn, baseclk);
1752 raw_spin_lock_irq(&base->lock);
1753 base->running_timer = NULL;
1754 timer_sync_wait_running(base);
1755 }
1756 }
1757}
1758
1759static int collect_expired_timers(struct timer_base *base,
1760 struct hlist_head *heads)
1761{
1762 unsigned long clk = base->clk = base->next_expiry;
1763 struct hlist_head *vec;
1764 int i, levels = 0;
1765 unsigned int idx;
1766
1767 for (i = 0; i < LVL_DEPTH; i++) {
1768 idx = (clk & LVL_MASK) + i * LVL_SIZE;
1769
1770 if (__test_and_clear_bit(idx, base->pending_map)) {
1771 vec = base->vectors + idx;
1772 hlist_move_list(vec, heads++);
1773 levels++;
1774 }
1775 /* Is it time to look at the next level? */
1776 if (clk & LVL_CLK_MASK)
1777 break;
1778 /* Shift clock for the next level granularity */
1779 clk >>= LVL_CLK_SHIFT;
1780 }
1781 return levels;
1782}
1783
1784/*
1785 * Find the next pending bucket of a level. Search from level start (@offset)
1786 * + @clk upwards and if nothing there, search from start of the level
1787 * (@offset) up to @offset + clk.
1788 */
1789static int next_pending_bucket(struct timer_base *base, unsigned offset,
1790 unsigned clk)
1791{
1792 unsigned pos, start = offset + clk;
1793 unsigned end = offset + LVL_SIZE;
1794
1795 pos = find_next_bit(base->pending_map, end, start);
1796 if (pos < end)
1797 return pos - start;
1798
1799 pos = find_next_bit(base->pending_map, start, offset);
1800 return pos < start ? pos + LVL_SIZE - start : -1;
1801}
1802
1803/*
1804 * Search the first expiring timer in the various clock levels. Caller must
1805 * hold base->lock.
1806 *
1807 * Store next expiry time in base->next_expiry.
1808 */
1809static void next_expiry_recalc(struct timer_base *base)
1810{
1811 unsigned long clk, next, adj;
1812 unsigned lvl, offset = 0;
1813
1814 next = base->clk + NEXT_TIMER_MAX_DELTA;
1815 clk = base->clk;
1816 for (lvl = 0; lvl < LVL_DEPTH; lvl++, offset += LVL_SIZE) {
1817 int pos = next_pending_bucket(base, offset, clk & LVL_MASK);
1818 unsigned long lvl_clk = clk & LVL_CLK_MASK;
1819
1820 if (pos >= 0) {
1821 unsigned long tmp = clk + (unsigned long) pos;
1822
1823 tmp <<= LVL_SHIFT(lvl);
1824 if (time_before(tmp, next))
1825 next = tmp;
1826
1827 /*
1828 * If the next expiration happens before we reach
1829 * the next level, no need to check further.
1830 */
1831 if (pos <= ((LVL_CLK_DIV - lvl_clk) & LVL_CLK_MASK))
1832 break;
1833 }
1834 /*
1835 * Clock for the next level. If the current level clock lower
1836 * bits are zero, we look at the next level as is. If not we
1837 * need to advance it by one because that's going to be the
1838 * next expiring bucket in that level. base->clk is the next
1839 * expiring jiffie. So in case of:
1840 *
1841 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1842 * 0 0 0 0 0 0
1843 *
1844 * we have to look at all levels @index 0. With
1845 *
1846 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1847 * 0 0 0 0 0 2
1848 *
1849 * LVL0 has the next expiring bucket @index 2. The upper
1850 * levels have the next expiring bucket @index 1.
1851 *
1852 * In case that the propagation wraps the next level the same
1853 * rules apply:
1854 *
1855 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1856 * 0 0 0 0 F 2
1857 *
1858 * So after looking at LVL0 we get:
1859 *
1860 * LVL5 LVL4 LVL3 LVL2 LVL1
1861 * 0 0 0 1 0
1862 *
1863 * So no propagation from LVL1 to LVL2 because that happened
1864 * with the add already, but then we need to propagate further
1865 * from LVL2 to LVL3.
1866 *
1867 * So the simple check whether the lower bits of the current
1868 * level are 0 or not is sufficient for all cases.
1869 */
1870 adj = lvl_clk ? 1 : 0;
1871 clk >>= LVL_CLK_SHIFT;
1872 clk += adj;
1873 }
1874
1875 base->next_expiry = next;
1876 base->next_expiry_recalc = false;
1877 base->timers_pending = !(next == base->clk + NEXT_TIMER_MAX_DELTA);
1878}
1879
1880#ifdef CONFIG_NO_HZ_COMMON
1881/*
1882 * Check, if the next hrtimer event is before the next timer wheel
1883 * event:
1884 */
1885static u64 cmp_next_hrtimer_event(u64 basem, u64 expires)
1886{
1887 u64 nextevt = hrtimer_get_next_event();
1888
1889 /*
1890 * If high resolution timers are enabled
1891 * hrtimer_get_next_event() returns KTIME_MAX.
1892 */
1893 if (expires <= nextevt)
1894 return expires;
1895
1896 /*
1897 * If the next timer is already expired, return the tick base
1898 * time so the tick is fired immediately.
1899 */
1900 if (nextevt <= basem)
1901 return basem;
1902
1903 /*
1904 * Round up to the next jiffie. High resolution timers are
1905 * off, so the hrtimers are expired in the tick and we need to
1906 * make sure that this tick really expires the timer to avoid
1907 * a ping pong of the nohz stop code.
1908 *
1909 * Use DIV_ROUND_UP_ULL to prevent gcc calling __divdi3
1910 */
1911 return DIV_ROUND_UP_ULL(nextevt, TICK_NSEC) * TICK_NSEC;
1912}
1913
1914/**
1915 * get_next_timer_interrupt - return the time (clock mono) of the next timer
1916 * @basej: base time jiffies
1917 * @basem: base time clock monotonic
1918 *
1919 * Returns the tick aligned clock monotonic time of the next pending
1920 * timer or KTIME_MAX if no timer is pending.
1921 */
1922u64 get_next_timer_interrupt(unsigned long basej, u64 basem)
1923{
1924 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1925 unsigned long nextevt = basej + NEXT_TIMER_MAX_DELTA;
1926 u64 expires = KTIME_MAX;
1927 bool was_idle;
1928
1929 /*
1930 * Pretend that there is no timer pending if the cpu is offline.
1931 * Possible pending timers will be migrated later to an active cpu.
1932 */
1933 if (cpu_is_offline(smp_processor_id()))
1934 return expires;
1935
1936 raw_spin_lock(&base->lock);
1937 if (base->next_expiry_recalc)
1938 next_expiry_recalc(base);
1939
1940 /*
1941 * We have a fresh next event. Check whether we can forward the
1942 * base.
1943 */
1944 __forward_timer_base(base, basej);
1945
1946 if (base->timers_pending) {
1947 nextevt = base->next_expiry;
1948
1949 /* If we missed a tick already, force 0 delta */
1950 if (time_before(nextevt, basej))
1951 nextevt = basej;
1952 expires = basem + (u64)(nextevt - basej) * TICK_NSEC;
1953 } else {
1954 /*
1955 * Move next_expiry for the empty base into the future to
1956 * prevent a unnecessary raise of the timer softirq when the
1957 * next_expiry value will be reached even if there is no timer
1958 * pending.
1959 */
1960 base->next_expiry = nextevt;
1961 }
1962
1963 /*
1964 * Base is idle if the next event is more than a tick away.
1965 *
1966 * If the base is marked idle then any timer add operation must forward
1967 * the base clk itself to keep granularity small. This idle logic is
1968 * only maintained for the BASE_STD base, deferrable timers may still
1969 * see large granularity skew (by design).
1970 */
1971 was_idle = base->is_idle;
1972 base->is_idle = time_after(nextevt, basej + 1);
1973 if (was_idle != base->is_idle)
1974 trace_timer_base_idle(base->is_idle, base->cpu);
1975
1976 raw_spin_unlock(&base->lock);
1977
1978 return cmp_next_hrtimer_event(basem, expires);
1979}
1980
1981/**
1982 * timer_clear_idle - Clear the idle state of the timer base
1983 *
1984 * Called with interrupts disabled
1985 */
1986void timer_clear_idle(void)
1987{
1988 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1989
1990 /*
1991 * We do this unlocked. The worst outcome is a remote enqueue sending
1992 * a pointless IPI, but taking the lock would just make the window for
1993 * sending the IPI a few instructions smaller for the cost of taking
1994 * the lock in the exit from idle path.
1995 */
1996 if (base->is_idle) {
1997 base->is_idle = false;
1998 trace_timer_base_idle(false, smp_processor_id());
1999 }
2000}
2001#endif
2002
2003/**
2004 * __run_timers - run all expired timers (if any) on this CPU.
2005 * @base: the timer vector to be processed.
2006 */
2007static inline void __run_timers(struct timer_base *base)
2008{
2009 struct hlist_head heads[LVL_DEPTH];
2010 int levels;
2011
2012 if (time_before(jiffies, base->next_expiry))
2013 return;
2014
2015 timer_base_lock_expiry(base);
2016 raw_spin_lock_irq(&base->lock);
2017
2018 while (time_after_eq(jiffies, base->clk) &&
2019 time_after_eq(jiffies, base->next_expiry)) {
2020 levels = collect_expired_timers(base, heads);
2021 /*
2022 * The two possible reasons for not finding any expired
2023 * timer at this clk are that all matching timers have been
2024 * dequeued or no timer has been queued since
2025 * base::next_expiry was set to base::clk +
2026 * NEXT_TIMER_MAX_DELTA.
2027 */
2028 WARN_ON_ONCE(!levels && !base->next_expiry_recalc
2029 && base->timers_pending);
2030 /*
2031 * While executing timers, base->clk is set 1 offset ahead of
2032 * jiffies to avoid endless requeuing to current jiffies.
2033 */
2034 base->clk++;
2035 next_expiry_recalc(base);
2036
2037 while (levels--)
2038 expire_timers(base, heads + levels);
2039 }
2040 raw_spin_unlock_irq(&base->lock);
2041 timer_base_unlock_expiry(base);
2042}
2043
2044/*
2045 * This function runs timers and the timer-tq in bottom half context.
2046 */
2047static __latent_entropy void run_timer_softirq(struct softirq_action *h)
2048{
2049 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
2050
2051 __run_timers(base);
2052 if (IS_ENABLED(CONFIG_NO_HZ_COMMON))
2053 __run_timers(this_cpu_ptr(&timer_bases[BASE_DEF]));
2054}
2055
2056/*
2057 * Called by the local, per-CPU timer interrupt on SMP.
2058 */
2059static void run_local_timers(void)
2060{
2061 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
2062
2063 hrtimer_run_queues();
2064 /* Raise the softirq only if required. */
2065 if (time_before(jiffies, base->next_expiry)) {
2066 if (!IS_ENABLED(CONFIG_NO_HZ_COMMON))
2067 return;
2068 /* CPU is awake, so check the deferrable base. */
2069 base++;
2070 if (time_before(jiffies, base->next_expiry))
2071 return;
2072 }
2073 raise_softirq(TIMER_SOFTIRQ);
2074}
2075
2076/*
2077 * Called from the timer interrupt handler to charge one tick to the current
2078 * process. user_tick is 1 if the tick is user time, 0 for system.
2079 */
2080void update_process_times(int user_tick)
2081{
2082 struct task_struct *p = current;
2083
2084 /* Note: this timer irq context must be accounted for as well. */
2085 account_process_tick(p, user_tick);
2086 run_local_timers();
2087 rcu_sched_clock_irq(user_tick);
2088#ifdef CONFIG_IRQ_WORK
2089 if (in_irq())
2090 irq_work_tick();
2091#endif
2092 scheduler_tick();
2093 if (IS_ENABLED(CONFIG_POSIX_TIMERS))
2094 run_posix_cpu_timers();
2095}
2096
2097/*
2098 * Since schedule_timeout()'s timer is defined on the stack, it must store
2099 * the target task on the stack as well.
2100 */
2101struct process_timer {
2102 struct timer_list timer;
2103 struct task_struct *task;
2104};
2105
2106static void process_timeout(struct timer_list *t)
2107{
2108 struct process_timer *timeout = from_timer(timeout, t, timer);
2109
2110 wake_up_process(timeout->task);
2111}
2112
2113/**
2114 * schedule_timeout - sleep until timeout
2115 * @timeout: timeout value in jiffies
2116 *
2117 * Make the current task sleep until @timeout jiffies have elapsed.
2118 * The function behavior depends on the current task state
2119 * (see also set_current_state() description):
2120 *
2121 * %TASK_RUNNING - the scheduler is called, but the task does not sleep
2122 * at all. That happens because sched_submit_work() does nothing for
2123 * tasks in %TASK_RUNNING state.
2124 *
2125 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
2126 * pass before the routine returns unless the current task is explicitly
2127 * woken up, (e.g. by wake_up_process()).
2128 *
2129 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
2130 * delivered to the current task or the current task is explicitly woken
2131 * up.
2132 *
2133 * The current task state is guaranteed to be %TASK_RUNNING when this
2134 * routine returns.
2135 *
2136 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
2137 * the CPU away without a bound on the timeout. In this case the return
2138 * value will be %MAX_SCHEDULE_TIMEOUT.
2139 *
2140 * Returns 0 when the timer has expired otherwise the remaining time in
2141 * jiffies will be returned. In all cases the return value is guaranteed
2142 * to be non-negative.
2143 */
2144signed long __sched schedule_timeout(signed long timeout)
2145{
2146 struct process_timer timer;
2147 unsigned long expire;
2148
2149 switch (timeout)
2150 {
2151 case MAX_SCHEDULE_TIMEOUT:
2152 /*
2153 * These two special cases are useful to be comfortable
2154 * in the caller. Nothing more. We could take
2155 * MAX_SCHEDULE_TIMEOUT from one of the negative value
2156 * but I' d like to return a valid offset (>=0) to allow
2157 * the caller to do everything it want with the retval.
2158 */
2159 schedule();
2160 goto out;
2161 default:
2162 /*
2163 * Another bit of PARANOID. Note that the retval will be
2164 * 0 since no piece of kernel is supposed to do a check
2165 * for a negative retval of schedule_timeout() (since it
2166 * should never happens anyway). You just have the printk()
2167 * that will tell you if something is gone wrong and where.
2168 */
2169 if (timeout < 0) {
2170 printk(KERN_ERR "schedule_timeout: wrong timeout "
2171 "value %lx\n", timeout);
2172 dump_stack();
2173 __set_current_state(TASK_RUNNING);
2174 goto out;
2175 }
2176 }
2177
2178 expire = timeout + jiffies;
2179
2180 timer.task = current;
2181 timer_setup_on_stack(&timer.timer, process_timeout, 0);
2182 __mod_timer(&timer.timer, expire, MOD_TIMER_NOTPENDING);
2183 schedule();
2184 del_timer_sync(&timer.timer);
2185
2186 /* Remove the timer from the object tracker */
2187 destroy_timer_on_stack(&timer.timer);
2188
2189 timeout = expire - jiffies;
2190
2191 out:
2192 return timeout < 0 ? 0 : timeout;
2193}
2194EXPORT_SYMBOL(schedule_timeout);
2195
2196/*
2197 * We can use __set_current_state() here because schedule_timeout() calls
2198 * schedule() unconditionally.
2199 */
2200signed long __sched schedule_timeout_interruptible(signed long timeout)
2201{
2202 __set_current_state(TASK_INTERRUPTIBLE);
2203 return schedule_timeout(timeout);
2204}
2205EXPORT_SYMBOL(schedule_timeout_interruptible);
2206
2207signed long __sched schedule_timeout_killable(signed long timeout)
2208{
2209 __set_current_state(TASK_KILLABLE);
2210 return schedule_timeout(timeout);
2211}
2212EXPORT_SYMBOL(schedule_timeout_killable);
2213
2214signed long __sched schedule_timeout_uninterruptible(signed long timeout)
2215{
2216 __set_current_state(TASK_UNINTERRUPTIBLE);
2217 return schedule_timeout(timeout);
2218}
2219EXPORT_SYMBOL(schedule_timeout_uninterruptible);
2220
2221/*
2222 * Like schedule_timeout_uninterruptible(), except this task will not contribute
2223 * to load average.
2224 */
2225signed long __sched schedule_timeout_idle(signed long timeout)
2226{
2227 __set_current_state(TASK_IDLE);
2228 return schedule_timeout(timeout);
2229}
2230EXPORT_SYMBOL(schedule_timeout_idle);
2231
2232#ifdef CONFIG_HOTPLUG_CPU
2233static void migrate_timer_list(struct timer_base *new_base, struct hlist_head *head)
2234{
2235 struct timer_list *timer;
2236 int cpu = new_base->cpu;
2237
2238 while (!hlist_empty(head)) {
2239 timer = hlist_entry(head->first, struct timer_list, entry);
2240 detach_timer(timer, false);
2241 timer->flags = (timer->flags & ~TIMER_BASEMASK) | cpu;
2242 internal_add_timer(new_base, timer);
2243 }
2244}
2245
2246int timers_prepare_cpu(unsigned int cpu)
2247{
2248 struct timer_base *base;
2249 int b;
2250
2251 for (b = 0; b < NR_BASES; b++) {
2252 base = per_cpu_ptr(&timer_bases[b], cpu);
2253 base->clk = jiffies;
2254 base->next_expiry = base->clk + NEXT_TIMER_MAX_DELTA;
2255 base->next_expiry_recalc = false;
2256 base->timers_pending = false;
2257 base->is_idle = false;
2258 }
2259 return 0;
2260}
2261
2262int timers_dead_cpu(unsigned int cpu)
2263{
2264 struct timer_base *old_base;
2265 struct timer_base *new_base;
2266 int b, i;
2267
2268 for (b = 0; b < NR_BASES; b++) {
2269 old_base = per_cpu_ptr(&timer_bases[b], cpu);
2270 new_base = get_cpu_ptr(&timer_bases[b]);
2271 /*
2272 * The caller is globally serialized and nobody else
2273 * takes two locks at once, deadlock is not possible.
2274 */
2275 raw_spin_lock_irq(&new_base->lock);
2276 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
2277
2278 /*
2279 * The current CPUs base clock might be stale. Update it
2280 * before moving the timers over.
2281 */
2282 forward_timer_base(new_base);
2283
2284 WARN_ON_ONCE(old_base->running_timer);
2285 old_base->running_timer = NULL;
2286
2287 for (i = 0; i < WHEEL_SIZE; i++)
2288 migrate_timer_list(new_base, old_base->vectors + i);
2289
2290 raw_spin_unlock(&old_base->lock);
2291 raw_spin_unlock_irq(&new_base->lock);
2292 put_cpu_ptr(&timer_bases);
2293 }
2294 return 0;
2295}
2296
2297#endif /* CONFIG_HOTPLUG_CPU */
2298
2299static void __init init_timer_cpu(int cpu)
2300{
2301 struct timer_base *base;
2302 int i;
2303
2304 for (i = 0; i < NR_BASES; i++) {
2305 base = per_cpu_ptr(&timer_bases[i], cpu);
2306 base->cpu = cpu;
2307 raw_spin_lock_init(&base->lock);
2308 base->clk = jiffies;
2309 base->next_expiry = base->clk + NEXT_TIMER_MAX_DELTA;
2310 timer_base_init_expiry_lock(base);
2311 }
2312}
2313
2314static void __init init_timer_cpus(void)
2315{
2316 int cpu;
2317
2318 for_each_possible_cpu(cpu)
2319 init_timer_cpu(cpu);
2320}
2321
2322void __init init_timers(void)
2323{
2324 init_timer_cpus();
2325 posix_cputimers_init_work();
2326 open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
2327}
2328
2329/**
2330 * msleep - sleep safely even with waitqueue interruptions
2331 * @msecs: Time in milliseconds to sleep for
2332 */
2333void msleep(unsigned int msecs)
2334{
2335 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
2336
2337 while (timeout)
2338 timeout = schedule_timeout_uninterruptible(timeout);
2339}
2340
2341EXPORT_SYMBOL(msleep);
2342
2343/**
2344 * msleep_interruptible - sleep waiting for signals
2345 * @msecs: Time in milliseconds to sleep for
2346 */
2347unsigned long msleep_interruptible(unsigned int msecs)
2348{
2349 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
2350
2351 while (timeout && !signal_pending(current))
2352 timeout = schedule_timeout_interruptible(timeout);
2353 return jiffies_to_msecs(timeout);
2354}
2355
2356EXPORT_SYMBOL(msleep_interruptible);
2357
2358/**
2359 * usleep_range_state - Sleep for an approximate time in a given state
2360 * @min: Minimum time in usecs to sleep
2361 * @max: Maximum time in usecs to sleep
2362 * @state: State of the current task that will be while sleeping
2363 *
2364 * In non-atomic context where the exact wakeup time is flexible, use
2365 * usleep_range_state() instead of udelay(). The sleep improves responsiveness
2366 * by avoiding the CPU-hogging busy-wait of udelay(), and the range reduces
2367 * power usage by allowing hrtimers to take advantage of an already-
2368 * scheduled interrupt instead of scheduling a new one just for this sleep.
2369 */
2370void __sched usleep_range_state(unsigned long min, unsigned long max,
2371 unsigned int state)
2372{
2373 ktime_t exp = ktime_add_us(ktime_get(), min);
2374 u64 delta = (u64)(max - min) * NSEC_PER_USEC;
2375
2376 for (;;) {
2377 __set_current_state(state);
2378 /* Do not return before the requested sleep time has elapsed */
2379 if (!schedule_hrtimeout_range(&exp, delta, HRTIMER_MODE_ABS))
2380 break;
2381 }
2382}
2383EXPORT_SYMBOL(usleep_range_state);
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
59EXPORT_SYMBOL(jiffies_64);
60
61/*
62 * per-CPU timer vector definitions:
63 */
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)
70#define MAX_TVAL ((unsigned long)((1ULL << (TVR_BITS + 4*TVN_BITS)) - 1))
71
72struct tvec {
73 struct hlist_head vec[TVN_SIZE];
74};
75
76struct tvec_root {
77 struct hlist_head vec[TVR_SIZE];
78};
79
80struct tvec_base {
81 spinlock_t lock;
82 struct timer_list *running_timer;
83 unsigned long timer_jiffies;
84 unsigned long next_timer;
85 unsigned long active_timers;
86 unsigned long all_timers;
87 int cpu;
88 bool migration_enabled;
89 bool nohz_active;
90 struct tvec_root tv1;
91 struct tvec tv2;
92 struct tvec tv3;
93 struct tvec tv4;
94 struct tvec tv5;
95} ____cacheline_aligned;
96
97
98static DEFINE_PER_CPU(struct tvec_base, tvec_bases);
99
100#if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
101unsigned int sysctl_timer_migration = 1;
102
103void timers_update_migration(bool update_nohz)
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;
115 if (!update_nohz)
116 continue;
117 per_cpu(tvec_bases.nohz_active, cpu) = true;
118 per_cpu(hrtimer_bases.nohz_active, cpu) = true;
119 }
120}
121
122int 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)
132 timers_update_migration(false);
133 mutex_unlock(&mutex);
134 return ret;
135}
136
137static 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
145static 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
152static unsigned long round_jiffies_common(unsigned long j, int cpu,
153 bool force_up)
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.
175 * But never round down if @force_up is set.
176 */
177 if (rem < HZ/4 && !force_up) /* round down */
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
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;
190}
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 */
212unsigned long __round_jiffies(unsigned long j, int cpu)
213{
214 return round_jiffies_common(j, cpu, false);
215}
216EXPORT_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 *
223 * __round_jiffies_relative() rounds a time delta in the future (in jiffies)
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 *
236 * The return value is the rounded version of the @j parameter.
237 */
238unsigned long __round_jiffies_relative(unsigned long j, int cpu)
239{
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;
244}
245EXPORT_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 *
251 * round_jiffies() rounds an absolute time in the future (in jiffies)
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 *
260 * The return value is the rounded version of the @j parameter.
261 */
262unsigned long round_jiffies(unsigned long j)
263{
264 return round_jiffies_common(j, raw_smp_processor_id(), false);
265}
266EXPORT_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 *
272 * round_jiffies_relative() rounds a time delta in the future (in jiffies)
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 *
281 * The return value is the rounded version of the @j parameter.
282 */
283unsigned long round_jiffies_relative(unsigned long j)
284{
285 return __round_jiffies_relative(j, raw_smp_processor_id());
286}
287EXPORT_SYMBOL_GPL(round_jiffies_relative);
288
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 */
299unsigned long __round_jiffies_up(unsigned long j, int cpu)
300{
301 return round_jiffies_common(j, cpu, true);
302}
303EXPORT_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 */
315unsigned 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}
322EXPORT_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 */
333unsigned long round_jiffies_up(unsigned long j)
334{
335 return round_jiffies_common(j, raw_smp_processor_id(), true);
336}
337EXPORT_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 */
348unsigned long round_jiffies_up_relative(unsigned long j)
349{
350 return __round_jiffies_up_relative(j, raw_smp_processor_id());
351}
352EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
353
354/**
355 * set_timer_slack - set the allowed slack for a timer
356 * @timer: the timer to be modified
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 */
367void set_timer_slack(struct timer_list *timer, int slack_hz)
368{
369 timer->slack = slack_hz;
370}
371EXPORT_SYMBOL_GPL(set_timer_slack);
372
373static void
374__internal_add_timer(struct tvec_base *base, struct timer_list *timer)
375{
376 unsigned long expires = timer->expires;
377 unsigned long idx = expires - base->timer_jiffies;
378 struct hlist_head *vec;
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;
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.
403 */
404 if (idx > MAX_TVAL) {
405 idx = MAX_TVAL;
406 expires = idx + base->timer_jiffies;
407 }
408 i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK;
409 vec = base->tv5.vec + i;
410 }
411
412 hlist_add_head(&timer->entry, vec);
413}
414
415static void internal_add_timer(struct tvec_base *base, struct timer_list *timer)
416{
417 /* Advance base->jiffies, if the base is empty */
418 if (!base->all_timers++)
419 base->timer_jiffies = jiffies;
420
421 __internal_add_timer(base, timer);
422 /*
423 * Update base->active_timers and base->next_timer
424 */
425 if (!(timer->flags & TIMER_DEFERRABLE)) {
426 if (!base->active_timers++ ||
427 time_before(timer->expires, base->next_timer))
428 base->next_timer = timer->expires;
429 }
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 */
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 }
449}
450
451#ifdef CONFIG_TIMER_STATS
452void __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}
461
462static void timer_stats_account_timer(struct timer_list *timer)
463{
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))
472 return;
473
474 timer_stats_update_stats(timer, timer->start_pid, site,
475 timer->function, timer->start_comm,
476 timer->flags);
477}
478
479#else
480static void timer_stats_account_timer(struct timer_list *timer) {}
481#endif
482
483#ifdef CONFIG_DEBUG_OBJECTS_TIMERS
484
485static struct debug_obj_descr timer_debug_descr;
486
487static void *timer_debug_hint(void *addr)
488{
489 return ((struct timer_list *) addr)->function;
490}
491
492/*
493 * fixup_init is called when:
494 * - an active object is initialized
495 */
496static int timer_fixup_init(void *addr, enum debug_obj_state state)
497{
498 struct timer_list *timer = addr;
499
500 switch (state) {
501 case ODEBUG_STATE_ACTIVE:
502 del_timer_sync(timer);
503 debug_object_init(timer, &timer_debug_descr);
504 return 1;
505 default:
506 return 0;
507 }
508}
509
510/* Stub timer callback for improperly used timers. */
511static void stub_timer(unsigned long data)
512{
513 WARN_ON(1);
514}
515
516/*
517 * fixup_activate is called when:
518 * - an active object is activated
519 * - an unknown object is activated (might be a statically initialized object)
520 */
521static int timer_fixup_activate(void *addr, enum debug_obj_state state)
522{
523 struct timer_list *timer = addr;
524
525 switch (state) {
526
527 case ODEBUG_STATE_NOTAVAILABLE:
528 /*
529 * This is not really a fixup. The timer was
530 * statically initialized. We just make sure that it
531 * is tracked in the object tracker.
532 */
533 if (timer->entry.pprev == NULL &&
534 timer->entry.next == TIMER_ENTRY_STATIC) {
535 debug_object_init(timer, &timer_debug_descr);
536 debug_object_activate(timer, &timer_debug_descr);
537 return 0;
538 } else {
539 setup_timer(timer, stub_timer, 0);
540 return 1;
541 }
542 return 0;
543
544 case ODEBUG_STATE_ACTIVE:
545 WARN_ON(1);
546
547 default:
548 return 0;
549 }
550}
551
552/*
553 * fixup_free is called when:
554 * - an active object is freed
555 */
556static int timer_fixup_free(void *addr, enum debug_obj_state state)
557{
558 struct timer_list *timer = addr;
559
560 switch (state) {
561 case ODEBUG_STATE_ACTIVE:
562 del_timer_sync(timer);
563 debug_object_free(timer, &timer_debug_descr);
564 return 1;
565 default:
566 return 0;
567 }
568}
569
570/*
571 * fixup_assert_init is called when:
572 * - an untracked/uninit-ed object is found
573 */
574static int timer_fixup_assert_init(void *addr, enum debug_obj_state state)
575{
576 struct timer_list *timer = addr;
577
578 switch (state) {
579 case ODEBUG_STATE_NOTAVAILABLE:
580 if (timer->entry.next == TIMER_ENTRY_STATIC) {
581 /*
582 * This is not really a fixup. The timer was
583 * statically initialized. We just make sure that it
584 * is tracked in the object tracker.
585 */
586 debug_object_init(timer, &timer_debug_descr);
587 return 0;
588 } else {
589 setup_timer(timer, stub_timer, 0);
590 return 1;
591 }
592 default:
593 return 0;
594 }
595}
596
597static struct debug_obj_descr timer_debug_descr = {
598 .name = "timer_list",
599 .debug_hint = timer_debug_hint,
600 .fixup_init = timer_fixup_init,
601 .fixup_activate = timer_fixup_activate,
602 .fixup_free = timer_fixup_free,
603 .fixup_assert_init = timer_fixup_assert_init,
604};
605
606static inline void debug_timer_init(struct timer_list *timer)
607{
608 debug_object_init(timer, &timer_debug_descr);
609}
610
611static inline void debug_timer_activate(struct timer_list *timer)
612{
613 debug_object_activate(timer, &timer_debug_descr);
614}
615
616static inline void debug_timer_deactivate(struct timer_list *timer)
617{
618 debug_object_deactivate(timer, &timer_debug_descr);
619}
620
621static inline void debug_timer_free(struct timer_list *timer)
622{
623 debug_object_free(timer, &timer_debug_descr);
624}
625
626static inline void debug_timer_assert_init(struct timer_list *timer)
627{
628 debug_object_assert_init(timer, &timer_debug_descr);
629}
630
631static void do_init_timer(struct timer_list *timer, unsigned int flags,
632 const char *name, struct lock_class_key *key);
633
634void init_timer_on_stack_key(struct timer_list *timer, unsigned int flags,
635 const char *name, struct lock_class_key *key)
636{
637 debug_object_init_on_stack(timer, &timer_debug_descr);
638 do_init_timer(timer, flags, name, key);
639}
640EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
641
642void destroy_timer_on_stack(struct timer_list *timer)
643{
644 debug_object_free(timer, &timer_debug_descr);
645}
646EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
647
648#else
649static inline void debug_timer_init(struct timer_list *timer) { }
650static inline void debug_timer_activate(struct timer_list *timer) { }
651static inline void debug_timer_deactivate(struct timer_list *timer) { }
652static inline void debug_timer_assert_init(struct timer_list *timer) { }
653#endif
654
655static inline void debug_init(struct timer_list *timer)
656{
657 debug_timer_init(timer);
658 trace_timer_init(timer);
659}
660
661static inline void
662debug_activate(struct timer_list *timer, unsigned long expires)
663{
664 debug_timer_activate(timer);
665 trace_timer_start(timer, expires, timer->flags);
666}
667
668static inline void debug_deactivate(struct timer_list *timer)
669{
670 debug_timer_deactivate(timer);
671 trace_timer_cancel(timer);
672}
673
674static inline void debug_assert_init(struct timer_list *timer)
675{
676 debug_timer_assert_init(timer);
677}
678
679static void do_init_timer(struct timer_list *timer, unsigned int flags,
680 const char *name, struct lock_class_key *key)
681{
682 timer->entry.pprev = NULL;
683 timer->flags = flags | raw_smp_processor_id();
684 timer->slack = -1;
685#ifdef CONFIG_TIMER_STATS
686 timer->start_site = NULL;
687 timer->start_pid = -1;
688 memset(timer->start_comm, 0, TASK_COMM_LEN);
689#endif
690 lockdep_init_map(&timer->lockdep_map, name, key, 0);
691}
692
693/**
694 * init_timer_key - initialize a timer
695 * @timer: the timer to be initialized
696 * @flags: timer flags
697 * @name: name of the timer
698 * @key: lockdep class key of the fake lock used for tracking timer
699 * sync lock dependencies
700 *
701 * init_timer_key() must be done to a timer prior calling *any* of the
702 * other timer functions.
703 */
704void init_timer_key(struct timer_list *timer, unsigned int flags,
705 const char *name, struct lock_class_key *key)
706{
707 debug_init(timer);
708 do_init_timer(timer, flags, name, key);
709}
710EXPORT_SYMBOL(init_timer_key);
711
712static inline void detach_timer(struct timer_list *timer, bool clear_pending)
713{
714 struct hlist_node *entry = &timer->entry;
715
716 debug_deactivate(timer);
717
718 __hlist_del(entry);
719 if (clear_pending)
720 entry->pprev = NULL;
721 entry->next = LIST_POISON2;
722}
723
724static inline void
725detach_expired_timer(struct timer_list *timer, struct tvec_base *base)
726{
727 detach_timer(timer, true);
728 if (!(timer->flags & TIMER_DEFERRABLE))
729 base->active_timers--;
730 base->all_timers--;
731}
732
733static int detach_if_pending(struct timer_list *timer, struct tvec_base *base,
734 bool clear_pending)
735{
736 if (!timer_pending(timer))
737 return 0;
738
739 detach_timer(timer, clear_pending);
740 if (!(timer->flags & TIMER_DEFERRABLE)) {
741 base->active_timers--;
742 if (timer->expires == base->next_timer)
743 base->next_timer = base->timer_jiffies;
744 }
745 /* If this was the last timer, advance base->jiffies */
746 if (!--base->all_timers)
747 base->timer_jiffies = jiffies;
748 return 1;
749}
750
751/*
752 * We are using hashed locking: holding per_cpu(tvec_bases).lock
753 * means that all timers which are tied to this base via timer->base are
754 * locked, and the base itself is locked too.
755 *
756 * So __run_timers/migrate_timers can safely modify all timers which could
757 * be found on ->tvX lists.
758 *
759 * When the timer's base is locked and removed from the list, the
760 * TIMER_MIGRATING flag is set, FIXME
761 */
762static struct tvec_base *lock_timer_base(struct timer_list *timer,
763 unsigned long *flags)
764 __acquires(timer->base->lock)
765{
766 for (;;) {
767 u32 tf = timer->flags;
768 struct tvec_base *base;
769
770 if (!(tf & TIMER_MIGRATING)) {
771 base = per_cpu_ptr(&tvec_bases, tf & TIMER_CPUMASK);
772 spin_lock_irqsave(&base->lock, *flags);
773 if (timer->flags == tf)
774 return base;
775 spin_unlock_irqrestore(&base->lock, *flags);
776 }
777 cpu_relax();
778 }
779}
780
781static inline int
782__mod_timer(struct timer_list *timer, unsigned long expires,
783 bool pending_only, int pinned)
784{
785 struct tvec_base *base, *new_base;
786 unsigned long flags;
787 int ret = 0;
788
789 timer_stats_timer_set_start_info(timer);
790 BUG_ON(!timer->function);
791
792 base = lock_timer_base(timer, &flags);
793
794 ret = detach_if_pending(timer, base, false);
795 if (!ret && pending_only)
796 goto out_unlock;
797
798 debug_activate(timer, expires);
799
800 new_base = get_target_base(base, pinned);
801
802 if (base != new_base) {
803 /*
804 * We are trying to schedule the timer on the local CPU.
805 * However we can't change timer's base while it is running,
806 * otherwise del_timer_sync() can't detect that the timer's
807 * handler yet has not finished. This also guarantees that
808 * the timer is serialized wrt itself.
809 */
810 if (likely(base->running_timer != timer)) {
811 /* See the comment in lock_timer_base() */
812 timer->flags |= TIMER_MIGRATING;
813
814 spin_unlock(&base->lock);
815 base = new_base;
816 spin_lock(&base->lock);
817 WRITE_ONCE(timer->flags,
818 (timer->flags & ~TIMER_BASEMASK) | base->cpu);
819 }
820 }
821
822 timer->expires = expires;
823 internal_add_timer(base, timer);
824
825out_unlock:
826 spin_unlock_irqrestore(&base->lock, flags);
827
828 return ret;
829}
830
831/**
832 * mod_timer_pending - modify a pending timer's timeout
833 * @timer: the pending timer to be modified
834 * @expires: new timeout in jiffies
835 *
836 * mod_timer_pending() is the same for pending timers as mod_timer(),
837 * but will not re-activate and modify already deleted timers.
838 *
839 * It is useful for unserialized use of timers.
840 */
841int mod_timer_pending(struct timer_list *timer, unsigned long expires)
842{
843 return __mod_timer(timer, expires, true, TIMER_NOT_PINNED);
844}
845EXPORT_SYMBOL(mod_timer_pending);
846
847/*
848 * Decide where to put the timer while taking the slack into account
849 *
850 * Algorithm:
851 * 1) calculate the maximum (absolute) time
852 * 2) calculate the highest bit where the expires and new max are different
853 * 3) use this bit to make a mask
854 * 4) use the bitmask to round down the maximum time, so that all last
855 * bits are zeros
856 */
857static inline
858unsigned long apply_slack(struct timer_list *timer, unsigned long expires)
859{
860 unsigned long expires_limit, mask;
861 int bit;
862
863 if (timer->slack >= 0) {
864 expires_limit = expires + timer->slack;
865 } else {
866 long delta = expires - jiffies;
867
868 if (delta < 256)
869 return expires;
870
871 expires_limit = expires + delta / 256;
872 }
873 mask = expires ^ expires_limit;
874 if (mask == 0)
875 return expires;
876
877 bit = __fls(mask);
878
879 mask = (1UL << bit) - 1;
880
881 expires_limit = expires_limit & ~(mask);
882
883 return expires_limit;
884}
885
886/**
887 * mod_timer - modify a timer's timeout
888 * @timer: the timer to be modified
889 * @expires: new timeout in jiffies
890 *
891 * mod_timer() is a more efficient way to update the expire field of an
892 * active timer (if the timer is inactive it will be activated)
893 *
894 * mod_timer(timer, expires) is equivalent to:
895 *
896 * del_timer(timer); timer->expires = expires; add_timer(timer);
897 *
898 * Note that if there are multiple unserialized concurrent users of the
899 * same timer, then mod_timer() is the only safe way to modify the timeout,
900 * since add_timer() cannot modify an already running timer.
901 *
902 * The function returns whether it has modified a pending timer or not.
903 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
904 * active timer returns 1.)
905 */
906int mod_timer(struct timer_list *timer, unsigned long expires)
907{
908 expires = apply_slack(timer, expires);
909
910 /*
911 * This is a common optimization triggered by the
912 * networking code - if the timer is re-modified
913 * to be the same thing then just return:
914 */
915 if (timer_pending(timer) && timer->expires == expires)
916 return 1;
917
918 return __mod_timer(timer, expires, false, TIMER_NOT_PINNED);
919}
920EXPORT_SYMBOL(mod_timer);
921
922/**
923 * mod_timer_pinned - modify a timer's timeout
924 * @timer: the timer to be modified
925 * @expires: new timeout in jiffies
926 *
927 * mod_timer_pinned() is a way to update the expire field of an
928 * active timer (if the timer is inactive it will be activated)
929 * and to ensure that the timer is scheduled on the current CPU.
930 *
931 * Note that this does not prevent the timer from being migrated
932 * when the current CPU goes offline. If this is a problem for
933 * you, use CPU-hotplug notifiers to handle it correctly, for
934 * example, cancelling the timer when the corresponding CPU goes
935 * offline.
936 *
937 * mod_timer_pinned(timer, expires) is equivalent to:
938 *
939 * del_timer(timer); timer->expires = expires; add_timer(timer);
940 */
941int mod_timer_pinned(struct timer_list *timer, unsigned long expires)
942{
943 if (timer->expires == expires && timer_pending(timer))
944 return 1;
945
946 return __mod_timer(timer, expires, false, TIMER_PINNED);
947}
948EXPORT_SYMBOL(mod_timer_pinned);
949
950/**
951 * add_timer - start a timer
952 * @timer: the timer to be added
953 *
954 * The kernel will do a ->function(->data) callback from the
955 * timer interrupt at the ->expires point in the future. The
956 * current time is 'jiffies'.
957 *
958 * The timer's ->expires, ->function (and if the handler uses it, ->data)
959 * fields must be set prior calling this function.
960 *
961 * Timers with an ->expires field in the past will be executed in the next
962 * timer tick.
963 */
964void add_timer(struct timer_list *timer)
965{
966 BUG_ON(timer_pending(timer));
967 mod_timer(timer, timer->expires);
968}
969EXPORT_SYMBOL(add_timer);
970
971/**
972 * add_timer_on - start a timer on a particular CPU
973 * @timer: the timer to be added
974 * @cpu: the CPU to start it on
975 *
976 * This is not very scalable on SMP. Double adds are not possible.
977 */
978void add_timer_on(struct timer_list *timer, int cpu)
979{
980 struct tvec_base *new_base = per_cpu_ptr(&tvec_bases, cpu);
981 struct tvec_base *base;
982 unsigned long flags;
983
984 timer_stats_timer_set_start_info(timer);
985 BUG_ON(timer_pending(timer) || !timer->function);
986
987 /*
988 * If @timer was on a different CPU, it should be migrated with the
989 * old base locked to prevent other operations proceeding with the
990 * wrong base locked. See lock_timer_base().
991 */
992 base = lock_timer_base(timer, &flags);
993 if (base != new_base) {
994 timer->flags |= TIMER_MIGRATING;
995
996 spin_unlock(&base->lock);
997 base = new_base;
998 spin_lock(&base->lock);
999 WRITE_ONCE(timer->flags,
1000 (timer->flags & ~TIMER_BASEMASK) | cpu);
1001 }
1002
1003 debug_activate(timer, timer->expires);
1004 internal_add_timer(base, timer);
1005 spin_unlock_irqrestore(&base->lock, flags);
1006}
1007EXPORT_SYMBOL_GPL(add_timer_on);
1008
1009/**
1010 * del_timer - deactive a timer.
1011 * @timer: the timer to be deactivated
1012 *
1013 * del_timer() deactivates a timer - this works on both active and inactive
1014 * timers.
1015 *
1016 * The function returns whether it has deactivated a pending timer or not.
1017 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
1018 * active timer returns 1.)
1019 */
1020int del_timer(struct timer_list *timer)
1021{
1022 struct tvec_base *base;
1023 unsigned long flags;
1024 int ret = 0;
1025
1026 debug_assert_init(timer);
1027
1028 timer_stats_timer_clear_start_info(timer);
1029 if (timer_pending(timer)) {
1030 base = lock_timer_base(timer, &flags);
1031 ret = detach_if_pending(timer, base, true);
1032 spin_unlock_irqrestore(&base->lock, flags);
1033 }
1034
1035 return ret;
1036}
1037EXPORT_SYMBOL(del_timer);
1038
1039/**
1040 * try_to_del_timer_sync - Try to deactivate a timer
1041 * @timer: timer do del
1042 *
1043 * This function tries to deactivate a timer. Upon successful (ret >= 0)
1044 * exit the timer is not queued and the handler is not running on any CPU.
1045 */
1046int try_to_del_timer_sync(struct timer_list *timer)
1047{
1048 struct tvec_base *base;
1049 unsigned long flags;
1050 int ret = -1;
1051
1052 debug_assert_init(timer);
1053
1054 base = lock_timer_base(timer, &flags);
1055
1056 if (base->running_timer != timer) {
1057 timer_stats_timer_clear_start_info(timer);
1058 ret = detach_if_pending(timer, base, true);
1059 }
1060 spin_unlock_irqrestore(&base->lock, flags);
1061
1062 return ret;
1063}
1064EXPORT_SYMBOL(try_to_del_timer_sync);
1065
1066#ifdef CONFIG_SMP
1067/**
1068 * del_timer_sync - deactivate a timer and wait for the handler to finish.
1069 * @timer: the timer to be deactivated
1070 *
1071 * This function only differs from del_timer() on SMP: besides deactivating
1072 * the timer it also makes sure the handler has finished executing on other
1073 * CPUs.
1074 *
1075 * Synchronization rules: Callers must prevent restarting of the timer,
1076 * otherwise this function is meaningless. It must not be called from
1077 * interrupt contexts unless the timer is an irqsafe one. The caller must
1078 * not hold locks which would prevent completion of the timer's
1079 * handler. The timer's handler must not call add_timer_on(). Upon exit the
1080 * timer is not queued and the handler is not running on any CPU.
1081 *
1082 * Note: For !irqsafe timers, you must not hold locks that are held in
1083 * interrupt context while calling this function. Even if the lock has
1084 * nothing to do with the timer in question. Here's why:
1085 *
1086 * CPU0 CPU1
1087 * ---- ----
1088 * <SOFTIRQ>
1089 * call_timer_fn();
1090 * base->running_timer = mytimer;
1091 * spin_lock_irq(somelock);
1092 * <IRQ>
1093 * spin_lock(somelock);
1094 * del_timer_sync(mytimer);
1095 * while (base->running_timer == mytimer);
1096 *
1097 * Now del_timer_sync() will never return and never release somelock.
1098 * The interrupt on the other CPU is waiting to grab somelock but
1099 * it has interrupted the softirq that CPU0 is waiting to finish.
1100 *
1101 * The function returns whether it has deactivated a pending timer or not.
1102 */
1103int del_timer_sync(struct timer_list *timer)
1104{
1105#ifdef CONFIG_LOCKDEP
1106 unsigned long flags;
1107
1108 /*
1109 * If lockdep gives a backtrace here, please reference
1110 * the synchronization rules above.
1111 */
1112 local_irq_save(flags);
1113 lock_map_acquire(&timer->lockdep_map);
1114 lock_map_release(&timer->lockdep_map);
1115 local_irq_restore(flags);
1116#endif
1117 /*
1118 * don't use it in hardirq context, because it
1119 * could lead to deadlock.
1120 */
1121 WARN_ON(in_irq() && !(timer->flags & TIMER_IRQSAFE));
1122 for (;;) {
1123 int ret = try_to_del_timer_sync(timer);
1124 if (ret >= 0)
1125 return ret;
1126 cpu_relax();
1127 }
1128}
1129EXPORT_SYMBOL(del_timer_sync);
1130#endif
1131
1132static int cascade(struct tvec_base *base, struct tvec *tv, int index)
1133{
1134 /* cascade all the timers from tv up one level */
1135 struct timer_list *timer;
1136 struct hlist_node *tmp;
1137 struct hlist_head tv_list;
1138
1139 hlist_move_list(tv->vec + index, &tv_list);
1140
1141 /*
1142 * We are removing _all_ timers from the list, so we
1143 * don't have to detach them individually.
1144 */
1145 hlist_for_each_entry_safe(timer, tmp, &tv_list, entry) {
1146 /* No accounting, while moving them */
1147 __internal_add_timer(base, timer);
1148 }
1149
1150 return index;
1151}
1152
1153static void call_timer_fn(struct timer_list *timer, void (*fn)(unsigned long),
1154 unsigned long data)
1155{
1156 int count = preempt_count();
1157
1158#ifdef CONFIG_LOCKDEP
1159 /*
1160 * It is permissible to free the timer from inside the
1161 * function that is called from it, this we need to take into
1162 * account for lockdep too. To avoid bogus "held lock freed"
1163 * warnings as well as problems when looking into
1164 * timer->lockdep_map, make a copy and use that here.
1165 */
1166 struct lockdep_map lockdep_map;
1167
1168 lockdep_copy_map(&lockdep_map, &timer->lockdep_map);
1169#endif
1170 /*
1171 * Couple the lock chain with the lock chain at
1172 * del_timer_sync() by acquiring the lock_map around the fn()
1173 * call here and in del_timer_sync().
1174 */
1175 lock_map_acquire(&lockdep_map);
1176
1177 trace_timer_expire_entry(timer);
1178 fn(data);
1179 trace_timer_expire_exit(timer);
1180
1181 lock_map_release(&lockdep_map);
1182
1183 if (count != preempt_count()) {
1184 WARN_ONCE(1, "timer: %pF preempt leak: %08x -> %08x\n",
1185 fn, count, preempt_count());
1186 /*
1187 * Restore the preempt count. That gives us a decent
1188 * chance to survive and extract information. If the
1189 * callback kept a lock held, bad luck, but not worse
1190 * than the BUG() we had.
1191 */
1192 preempt_count_set(count);
1193 }
1194}
1195
1196#define INDEX(N) ((base->timer_jiffies >> (TVR_BITS + (N) * TVN_BITS)) & TVN_MASK)
1197
1198/**
1199 * __run_timers - run all expired timers (if any) on this CPU.
1200 * @base: the timer vector to be processed.
1201 *
1202 * This function cascades all vectors and executes all expired timer
1203 * vectors.
1204 */
1205static inline void __run_timers(struct tvec_base *base)
1206{
1207 struct timer_list *timer;
1208
1209 spin_lock_irq(&base->lock);
1210
1211 while (time_after_eq(jiffies, base->timer_jiffies)) {
1212 struct hlist_head work_list;
1213 struct hlist_head *head = &work_list;
1214 int index;
1215
1216 if (!base->all_timers) {
1217 base->timer_jiffies = jiffies;
1218 break;
1219 }
1220
1221 index = base->timer_jiffies & TVR_MASK;
1222
1223 /*
1224 * Cascade timers:
1225 */
1226 if (!index &&
1227 (!cascade(base, &base->tv2, INDEX(0))) &&
1228 (!cascade(base, &base->tv3, INDEX(1))) &&
1229 !cascade(base, &base->tv4, INDEX(2)))
1230 cascade(base, &base->tv5, INDEX(3));
1231 ++base->timer_jiffies;
1232 hlist_move_list(base->tv1.vec + index, head);
1233 while (!hlist_empty(head)) {
1234 void (*fn)(unsigned long);
1235 unsigned long data;
1236 bool irqsafe;
1237
1238 timer = hlist_entry(head->first, struct timer_list, entry);
1239 fn = timer->function;
1240 data = timer->data;
1241 irqsafe = timer->flags & TIMER_IRQSAFE;
1242
1243 timer_stats_account_timer(timer);
1244
1245 base->running_timer = timer;
1246 detach_expired_timer(timer, base);
1247
1248 if (irqsafe) {
1249 spin_unlock(&base->lock);
1250 call_timer_fn(timer, fn, data);
1251 spin_lock(&base->lock);
1252 } else {
1253 spin_unlock_irq(&base->lock);
1254 call_timer_fn(timer, fn, data);
1255 spin_lock_irq(&base->lock);
1256 }
1257 }
1258 }
1259 base->running_timer = NULL;
1260 spin_unlock_irq(&base->lock);
1261}
1262
1263#ifdef CONFIG_NO_HZ_COMMON
1264/*
1265 * Find out when the next timer event is due to happen. This
1266 * is used on S/390 to stop all activity when a CPU is idle.
1267 * This function needs to be called with interrupts disabled.
1268 */
1269static unsigned long __next_timer_interrupt(struct tvec_base *base)
1270{
1271 unsigned long timer_jiffies = base->timer_jiffies;
1272 unsigned long expires = timer_jiffies + NEXT_TIMER_MAX_DELTA;
1273 int index, slot, array, found = 0;
1274 struct timer_list *nte;
1275 struct tvec *varray[4];
1276
1277 /* Look for timer events in tv1. */
1278 index = slot = timer_jiffies & TVR_MASK;
1279 do {
1280 hlist_for_each_entry(nte, base->tv1.vec + slot, entry) {
1281 if (nte->flags & TIMER_DEFERRABLE)
1282 continue;
1283
1284 found = 1;
1285 expires = nte->expires;
1286 /* Look at the cascade bucket(s)? */
1287 if (!index || slot < index)
1288 goto cascade;
1289 return expires;
1290 }
1291 slot = (slot + 1) & TVR_MASK;
1292 } while (slot != index);
1293
1294cascade:
1295 /* Calculate the next cascade event */
1296 if (index)
1297 timer_jiffies += TVR_SIZE - index;
1298 timer_jiffies >>= TVR_BITS;
1299
1300 /* Check tv2-tv5. */
1301 varray[0] = &base->tv2;
1302 varray[1] = &base->tv3;
1303 varray[2] = &base->tv4;
1304 varray[3] = &base->tv5;
1305
1306 for (array = 0; array < 4; array++) {
1307 struct tvec *varp = varray[array];
1308
1309 index = slot = timer_jiffies & TVN_MASK;
1310 do {
1311 hlist_for_each_entry(nte, varp->vec + slot, entry) {
1312 if (nte->flags & TIMER_DEFERRABLE)
1313 continue;
1314
1315 found = 1;
1316 if (time_before(nte->expires, expires))
1317 expires = nte->expires;
1318 }
1319 /*
1320 * Do we still search for the first timer or are
1321 * we looking up the cascade buckets ?
1322 */
1323 if (found) {
1324 /* Look at the cascade bucket(s)? */
1325 if (!index || slot < index)
1326 break;
1327 return expires;
1328 }
1329 slot = (slot + 1) & TVN_MASK;
1330 } while (slot != index);
1331
1332 if (index)
1333 timer_jiffies += TVN_SIZE - index;
1334 timer_jiffies >>= TVN_BITS;
1335 }
1336 return expires;
1337}
1338
1339/*
1340 * Check, if the next hrtimer event is before the next timer wheel
1341 * event:
1342 */
1343static u64 cmp_next_hrtimer_event(u64 basem, u64 expires)
1344{
1345 u64 nextevt = hrtimer_get_next_event();
1346
1347 /*
1348 * If high resolution timers are enabled
1349 * hrtimer_get_next_event() returns KTIME_MAX.
1350 */
1351 if (expires <= nextevt)
1352 return expires;
1353
1354 /*
1355 * If the next timer is already expired, return the tick base
1356 * time so the tick is fired immediately.
1357 */
1358 if (nextevt <= basem)
1359 return basem;
1360
1361 /*
1362 * Round up to the next jiffie. High resolution timers are
1363 * off, so the hrtimers are expired in the tick and we need to
1364 * make sure that this tick really expires the timer to avoid
1365 * a ping pong of the nohz stop code.
1366 *
1367 * Use DIV_ROUND_UP_ULL to prevent gcc calling __divdi3
1368 */
1369 return DIV_ROUND_UP_ULL(nextevt, TICK_NSEC) * TICK_NSEC;
1370}
1371
1372/**
1373 * get_next_timer_interrupt - return the time (clock mono) of the next timer
1374 * @basej: base time jiffies
1375 * @basem: base time clock monotonic
1376 *
1377 * Returns the tick aligned clock monotonic time of the next pending
1378 * timer or KTIME_MAX if no timer is pending.
1379 */
1380u64 get_next_timer_interrupt(unsigned long basej, u64 basem)
1381{
1382 struct tvec_base *base = this_cpu_ptr(&tvec_bases);
1383 u64 expires = KTIME_MAX;
1384 unsigned long nextevt;
1385
1386 /*
1387 * Pretend that there is no timer pending if the cpu is offline.
1388 * Possible pending timers will be migrated later to an active cpu.
1389 */
1390 if (cpu_is_offline(smp_processor_id()))
1391 return expires;
1392
1393 spin_lock(&base->lock);
1394 if (base->active_timers) {
1395 if (time_before_eq(base->next_timer, base->timer_jiffies))
1396 base->next_timer = __next_timer_interrupt(base);
1397 nextevt = base->next_timer;
1398 if (time_before_eq(nextevt, basej))
1399 expires = basem;
1400 else
1401 expires = basem + (nextevt - basej) * TICK_NSEC;
1402 }
1403 spin_unlock(&base->lock);
1404
1405 return cmp_next_hrtimer_event(basem, expires);
1406}
1407#endif
1408
1409/*
1410 * Called from the timer interrupt handler to charge one tick to the current
1411 * process. user_tick is 1 if the tick is user time, 0 for system.
1412 */
1413void update_process_times(int user_tick)
1414{
1415 struct task_struct *p = current;
1416
1417 /* Note: this timer irq context must be accounted for as well. */
1418 account_process_tick(p, user_tick);
1419 run_local_timers();
1420 rcu_check_callbacks(user_tick);
1421#ifdef CONFIG_IRQ_WORK
1422 if (in_irq())
1423 irq_work_tick();
1424#endif
1425 scheduler_tick();
1426 run_posix_cpu_timers(p);
1427}
1428
1429/*
1430 * This function runs timers and the timer-tq in bottom half context.
1431 */
1432static void run_timer_softirq(struct softirq_action *h)
1433{
1434 struct tvec_base *base = this_cpu_ptr(&tvec_bases);
1435
1436 if (time_after_eq(jiffies, base->timer_jiffies))
1437 __run_timers(base);
1438}
1439
1440/*
1441 * Called by the local, per-CPU timer interrupt on SMP.
1442 */
1443void run_local_timers(void)
1444{
1445 hrtimer_run_queues();
1446 raise_softirq(TIMER_SOFTIRQ);
1447}
1448
1449#ifdef __ARCH_WANT_SYS_ALARM
1450
1451/*
1452 * For backwards compatibility? This can be done in libc so Alpha
1453 * and all newer ports shouldn't need it.
1454 */
1455SYSCALL_DEFINE1(alarm, unsigned int, seconds)
1456{
1457 return alarm_setitimer(seconds);
1458}
1459
1460#endif
1461
1462static void process_timeout(unsigned long __data)
1463{
1464 wake_up_process((struct task_struct *)__data);
1465}
1466
1467/**
1468 * schedule_timeout - sleep until timeout
1469 * @timeout: timeout value in jiffies
1470 *
1471 * Make the current task sleep until @timeout jiffies have
1472 * elapsed. The routine will return immediately unless
1473 * the current task state has been set (see set_current_state()).
1474 *
1475 * You can set the task state as follows -
1476 *
1477 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1478 * pass before the routine returns. The routine will return 0
1479 *
1480 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1481 * delivered to the current task. In this case the remaining time
1482 * in jiffies will be returned, or 0 if the timer expired in time
1483 *
1484 * The current task state is guaranteed to be TASK_RUNNING when this
1485 * routine returns.
1486 *
1487 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1488 * the CPU away without a bound on the timeout. In this case the return
1489 * value will be %MAX_SCHEDULE_TIMEOUT.
1490 *
1491 * In all cases the return value is guaranteed to be non-negative.
1492 */
1493signed long __sched schedule_timeout(signed long timeout)
1494{
1495 struct timer_list timer;
1496 unsigned long expire;
1497
1498 switch (timeout)
1499 {
1500 case MAX_SCHEDULE_TIMEOUT:
1501 /*
1502 * These two special cases are useful to be comfortable
1503 * in the caller. Nothing more. We could take
1504 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1505 * but I' d like to return a valid offset (>=0) to allow
1506 * the caller to do everything it want with the retval.
1507 */
1508 schedule();
1509 goto out;
1510 default:
1511 /*
1512 * Another bit of PARANOID. Note that the retval will be
1513 * 0 since no piece of kernel is supposed to do a check
1514 * for a negative retval of schedule_timeout() (since it
1515 * should never happens anyway). You just have the printk()
1516 * that will tell you if something is gone wrong and where.
1517 */
1518 if (timeout < 0) {
1519 printk(KERN_ERR "schedule_timeout: wrong timeout "
1520 "value %lx\n", timeout);
1521 dump_stack();
1522 current->state = TASK_RUNNING;
1523 goto out;
1524 }
1525 }
1526
1527 expire = timeout + jiffies;
1528
1529 setup_timer_on_stack(&timer, process_timeout, (unsigned long)current);
1530 __mod_timer(&timer, expire, false, TIMER_NOT_PINNED);
1531 schedule();
1532 del_singleshot_timer_sync(&timer);
1533
1534 /* Remove the timer from the object tracker */
1535 destroy_timer_on_stack(&timer);
1536
1537 timeout = expire - jiffies;
1538
1539 out:
1540 return timeout < 0 ? 0 : timeout;
1541}
1542EXPORT_SYMBOL(schedule_timeout);
1543
1544/*
1545 * We can use __set_current_state() here because schedule_timeout() calls
1546 * schedule() unconditionally.
1547 */
1548signed long __sched schedule_timeout_interruptible(signed long timeout)
1549{
1550 __set_current_state(TASK_INTERRUPTIBLE);
1551 return schedule_timeout(timeout);
1552}
1553EXPORT_SYMBOL(schedule_timeout_interruptible);
1554
1555signed long __sched schedule_timeout_killable(signed long timeout)
1556{
1557 __set_current_state(TASK_KILLABLE);
1558 return schedule_timeout(timeout);
1559}
1560EXPORT_SYMBOL(schedule_timeout_killable);
1561
1562signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1563{
1564 __set_current_state(TASK_UNINTERRUPTIBLE);
1565 return schedule_timeout(timeout);
1566}
1567EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1568
1569/*
1570 * Like schedule_timeout_uninterruptible(), except this task will not contribute
1571 * to load average.
1572 */
1573signed long __sched schedule_timeout_idle(signed long timeout)
1574{
1575 __set_current_state(TASK_IDLE);
1576 return schedule_timeout(timeout);
1577}
1578EXPORT_SYMBOL(schedule_timeout_idle);
1579
1580#ifdef CONFIG_HOTPLUG_CPU
1581static void migrate_timer_list(struct tvec_base *new_base, struct hlist_head *head)
1582{
1583 struct timer_list *timer;
1584 int cpu = new_base->cpu;
1585
1586 while (!hlist_empty(head)) {
1587 timer = hlist_entry(head->first, struct timer_list, entry);
1588 /* We ignore the accounting on the dying cpu */
1589 detach_timer(timer, false);
1590 timer->flags = (timer->flags & ~TIMER_BASEMASK) | cpu;
1591 internal_add_timer(new_base, timer);
1592 }
1593}
1594
1595static void migrate_timers(int cpu)
1596{
1597 struct tvec_base *old_base;
1598 struct tvec_base *new_base;
1599 int i;
1600
1601 BUG_ON(cpu_online(cpu));
1602 old_base = per_cpu_ptr(&tvec_bases, cpu);
1603 new_base = get_cpu_ptr(&tvec_bases);
1604 /*
1605 * The caller is globally serialized and nobody else
1606 * takes two locks at once, deadlock is not possible.
1607 */
1608 spin_lock_irq(&new_base->lock);
1609 spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1610
1611 BUG_ON(old_base->running_timer);
1612
1613 for (i = 0; i < TVR_SIZE; i++)
1614 migrate_timer_list(new_base, old_base->tv1.vec + i);
1615 for (i = 0; i < TVN_SIZE; i++) {
1616 migrate_timer_list(new_base, old_base->tv2.vec + i);
1617 migrate_timer_list(new_base, old_base->tv3.vec + i);
1618 migrate_timer_list(new_base, old_base->tv4.vec + i);
1619 migrate_timer_list(new_base, old_base->tv5.vec + i);
1620 }
1621
1622 old_base->active_timers = 0;
1623 old_base->all_timers = 0;
1624
1625 spin_unlock(&old_base->lock);
1626 spin_unlock_irq(&new_base->lock);
1627 put_cpu_ptr(&tvec_bases);
1628}
1629
1630static int timer_cpu_notify(struct notifier_block *self,
1631 unsigned long action, void *hcpu)
1632{
1633 switch (action) {
1634 case CPU_DEAD:
1635 case CPU_DEAD_FROZEN:
1636 migrate_timers((long)hcpu);
1637 break;
1638 default:
1639 break;
1640 }
1641
1642 return NOTIFY_OK;
1643}
1644
1645static inline void timer_register_cpu_notifier(void)
1646{
1647 cpu_notifier(timer_cpu_notify, 0);
1648}
1649#else
1650static inline void timer_register_cpu_notifier(void) { }
1651#endif /* CONFIG_HOTPLUG_CPU */
1652
1653static void __init init_timer_cpu(int cpu)
1654{
1655 struct tvec_base *base = per_cpu_ptr(&tvec_bases, cpu);
1656
1657 base->cpu = cpu;
1658 spin_lock_init(&base->lock);
1659
1660 base->timer_jiffies = jiffies;
1661 base->next_timer = base->timer_jiffies;
1662}
1663
1664static void __init init_timer_cpus(void)
1665{
1666 int cpu;
1667
1668 for_each_possible_cpu(cpu)
1669 init_timer_cpu(cpu);
1670}
1671
1672void __init init_timers(void)
1673{
1674 init_timer_cpus();
1675 init_timer_stats();
1676 timer_register_cpu_notifier();
1677 open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
1678}
1679
1680/**
1681 * msleep - sleep safely even with waitqueue interruptions
1682 * @msecs: Time in milliseconds to sleep for
1683 */
1684void msleep(unsigned int msecs)
1685{
1686 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1687
1688 while (timeout)
1689 timeout = schedule_timeout_uninterruptible(timeout);
1690}
1691
1692EXPORT_SYMBOL(msleep);
1693
1694/**
1695 * msleep_interruptible - sleep waiting for signals
1696 * @msecs: Time in milliseconds to sleep for
1697 */
1698unsigned long msleep_interruptible(unsigned int msecs)
1699{
1700 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1701
1702 while (timeout && !signal_pending(current))
1703 timeout = schedule_timeout_interruptible(timeout);
1704 return jiffies_to_msecs(timeout);
1705}
1706
1707EXPORT_SYMBOL(msleep_interruptible);
1708
1709static void __sched do_usleep_range(unsigned long min, unsigned long max)
1710{
1711 ktime_t kmin;
1712 u64 delta;
1713
1714 kmin = ktime_set(0, min * NSEC_PER_USEC);
1715 delta = (u64)(max - min) * NSEC_PER_USEC;
1716 schedule_hrtimeout_range(&kmin, delta, HRTIMER_MODE_REL);
1717}
1718
1719/**
1720 * usleep_range - Drop in replacement for udelay where wakeup is flexible
1721 * @min: Minimum time in usecs to sleep
1722 * @max: Maximum time in usecs to sleep
1723 */
1724void __sched usleep_range(unsigned long min, unsigned long max)
1725{
1726 __set_current_state(TASK_UNINTERRUPTIBLE);
1727 do_usleep_range(min, max);
1728}
1729EXPORT_SYMBOL(usleep_range);