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/sysctl.h>
41#include <linux/sched/nohz.h>
42#include <linux/sched/debug.h>
43#include <linux/slab.h>
44#include <linux/compat.h>
45#include <linux/random.h>
46#include <linux/sysctl.h>
47
48#include <linux/uaccess.h>
49#include <asm/unistd.h>
50#include <asm/div64.h>
51#include <asm/timex.h>
52#include <asm/io.h>
53
54#include "tick-internal.h"
55#include "timer_migration.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 therefore each
67 * level has a different granularity.
68 *
69 * The level granularity is: LVL_CLK_DIV ^ level
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 * therefore 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/*
191 * If multiple bases need to be locked, use the base ordering for lock
192 * nesting, i.e. lowest number first.
193 */
194# define NR_BASES 3
195# define BASE_LOCAL 0
196# define BASE_GLOBAL 1
197# define BASE_DEF 2
198#else
199# define NR_BASES 1
200# define BASE_LOCAL 0
201# define BASE_GLOBAL 0
202# define BASE_DEF 0
203#endif
204
205/**
206 * struct timer_base - Per CPU timer base (number of base depends on config)
207 * @lock: Lock protecting the timer_base
208 * @running_timer: When expiring timers, the lock is dropped. To make
209 * sure not to race against deleting/modifying a
210 * currently running timer, the pointer is set to the
211 * timer, which expires at the moment. If no timer is
212 * running, the pointer is NULL.
213 * @expiry_lock: PREEMPT_RT only: Lock is taken in softirq around
214 * timer expiry callback execution and when trying to
215 * delete a running timer and it wasn't successful in
216 * the first glance. It prevents priority inversion
217 * when callback was preempted on a remote CPU and a
218 * caller tries to delete the running timer. It also
219 * prevents a life lock, when the task which tries to
220 * delete a timer preempted the softirq thread which
221 * is running the timer callback function.
222 * @timer_waiters: PREEMPT_RT only: Tells, if there is a waiter
223 * waiting for the end of the timer callback function
224 * execution.
225 * @clk: clock of the timer base; is updated before enqueue
226 * of a timer; during expiry, it is 1 offset ahead of
227 * jiffies to avoid endless requeuing to current
228 * jiffies
229 * @next_expiry: expiry value of the first timer; it is updated when
230 * finding the next timer and during enqueue; the
231 * value is not valid, when next_expiry_recalc is set
232 * @cpu: Number of CPU the timer base belongs to
233 * @next_expiry_recalc: States, whether a recalculation of next_expiry is
234 * required. Value is set true, when a timer was
235 * deleted.
236 * @is_idle: Is set, when timer_base is idle. It is triggered by NOHZ
237 * code. This state is only used in standard
238 * base. Deferrable timers, which are enqueued remotely
239 * never wake up an idle CPU. So no matter of supporting it
240 * for this base.
241 * @timers_pending: Is set, when a timer is pending in the base. It is only
242 * reliable when next_expiry_recalc is not set.
243 * @pending_map: bitmap of the timer wheel; each bit reflects a
244 * bucket of the wheel. When a bit is set, at least a
245 * single timer is enqueued in the related bucket.
246 * @vectors: Array of lists; Each array member reflects a bucket
247 * of the timer wheel. The list contains all timers
248 * which are enqueued into a specific bucket.
249 */
250struct timer_base {
251 raw_spinlock_t lock;
252 struct timer_list *running_timer;
253#ifdef CONFIG_PREEMPT_RT
254 spinlock_t expiry_lock;
255 atomic_t timer_waiters;
256#endif
257 unsigned long clk;
258 unsigned long next_expiry;
259 unsigned int cpu;
260 bool next_expiry_recalc;
261 bool is_idle;
262 bool timers_pending;
263 DECLARE_BITMAP(pending_map, WHEEL_SIZE);
264 struct hlist_head vectors[WHEEL_SIZE];
265} ____cacheline_aligned;
266
267static DEFINE_PER_CPU(struct timer_base, timer_bases[NR_BASES]);
268
269#ifdef CONFIG_NO_HZ_COMMON
270
271static DEFINE_STATIC_KEY_FALSE(timers_nohz_active);
272static DEFINE_MUTEX(timer_keys_mutex);
273
274static void timer_update_keys(struct work_struct *work);
275static DECLARE_WORK(timer_update_work, timer_update_keys);
276
277#ifdef CONFIG_SMP
278static unsigned int sysctl_timer_migration = 1;
279
280DEFINE_STATIC_KEY_FALSE(timers_migration_enabled);
281
282static void timers_update_migration(void)
283{
284 if (sysctl_timer_migration && tick_nohz_active)
285 static_branch_enable(&timers_migration_enabled);
286 else
287 static_branch_disable(&timers_migration_enabled);
288}
289
290#ifdef CONFIG_SYSCTL
291static int timer_migration_handler(const struct ctl_table *table, int write,
292 void *buffer, size_t *lenp, loff_t *ppos)
293{
294 int ret;
295
296 mutex_lock(lock: &timer_keys_mutex);
297 ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
298 if (!ret && write)
299 timers_update_migration();
300 mutex_unlock(lock: &timer_keys_mutex);
301 return ret;
302}
303
304static const struct ctl_table timer_sysctl[] = {
305 {
306 .procname = "timer_migration",
307 .data = &sysctl_timer_migration,
308 .maxlen = sizeof(unsigned int),
309 .mode = 0644,
310 .proc_handler = timer_migration_handler,
311 .extra1 = SYSCTL_ZERO,
312 .extra2 = SYSCTL_ONE,
313 },
314};
315
316static int __init timer_sysctl_init(void)
317{
318 register_sysctl("kernel", timer_sysctl);
319 return 0;
320}
321device_initcall(timer_sysctl_init);
322#endif /* CONFIG_SYSCTL */
323#else /* CONFIG_SMP */
324static inline void timers_update_migration(void) { }
325#endif /* !CONFIG_SMP */
326
327static void timer_update_keys(struct work_struct *work)
328{
329 mutex_lock(lock: &timer_keys_mutex);
330 timers_update_migration();
331 static_branch_enable(&timers_nohz_active);
332 mutex_unlock(lock: &timer_keys_mutex);
333}
334
335void timers_update_nohz(void)
336{
337 schedule_work(work: &timer_update_work);
338}
339
340static inline bool is_timers_nohz_active(void)
341{
342 return static_branch_unlikely(&timers_nohz_active);
343}
344#else
345static inline bool is_timers_nohz_active(void) { return false; }
346#endif /* NO_HZ_COMMON */
347
348static unsigned long round_jiffies_common(unsigned long j, int cpu,
349 bool force_up)
350{
351 int rem;
352 unsigned long original = j;
353
354 /*
355 * We don't want all cpus firing their timers at once hitting the
356 * same lock or cachelines, so we skew each extra cpu with an extra
357 * 3 jiffies. This 3 jiffies came originally from the mm/ code which
358 * already did this.
359 * The skew is done by adding 3*cpunr, then round, then subtract this
360 * extra offset again.
361 */
362 j += cpu * 3;
363
364 rem = j % HZ;
365
366 /*
367 * If the target jiffy is just after a whole second (which can happen
368 * due to delays of the timer irq, long irq off times etc etc) then
369 * we should round down to the whole second, not up. Use 1/4th second
370 * as cutoff for this rounding as an extreme upper bound for this.
371 * But never round down if @force_up is set.
372 */
373 if (rem < HZ/4 && !force_up) /* round down */
374 j = j - rem;
375 else /* round up */
376 j = j - rem + HZ;
377
378 /* now that we have rounded, subtract the extra skew again */
379 j -= cpu * 3;
380
381 /*
382 * Make sure j is still in the future. Otherwise return the
383 * unmodified value.
384 */
385 return time_is_after_jiffies(j) ? j : original;
386}
387
388/**
389 * __round_jiffies_relative - function to round jiffies to a full second
390 * @j: the time in (relative) jiffies that should be rounded
391 * @cpu: the processor number on which the timeout will happen
392 *
393 * __round_jiffies_relative() rounds a time delta in the future (in jiffies)
394 * up or down to (approximately) full seconds. This is useful for timers
395 * for which the exact time they fire does not matter too much, as long as
396 * they fire approximately every X seconds.
397 *
398 * By rounding these timers to whole seconds, all such timers will fire
399 * at the same time, rather than at various times spread out. The goal
400 * of this is to have the CPU wake up less, which saves power.
401 *
402 * The exact rounding is skewed for each processor to avoid all
403 * processors firing at the exact same time, which could lead
404 * to lock contention or spurious cache line bouncing.
405 *
406 * The return value is the rounded version of the @j parameter.
407 */
408unsigned long __round_jiffies_relative(unsigned long j, int cpu)
409{
410 unsigned long j0 = jiffies;
411
412 /* Use j0 because jiffies might change while we run */
413 return round_jiffies_common(j: j + j0, cpu, force_up: false) - j0;
414}
415EXPORT_SYMBOL_GPL(__round_jiffies_relative);
416
417/**
418 * round_jiffies - function to round jiffies to a full second
419 * @j: the time in (absolute) jiffies that should be rounded
420 *
421 * round_jiffies() rounds an absolute time in the future (in jiffies)
422 * up or down to (approximately) full seconds. This is useful for timers
423 * for which the exact time they fire does not matter too much, as long as
424 * they fire approximately every X seconds.
425 *
426 * By rounding these timers to whole seconds, all such timers will fire
427 * at the same time, rather than at various times spread out. The goal
428 * of this is to have the CPU wake up less, which saves power.
429 *
430 * The return value is the rounded version of the @j parameter.
431 */
432unsigned long round_jiffies(unsigned long j)
433{
434 return round_jiffies_common(j, raw_smp_processor_id(), force_up: false);
435}
436EXPORT_SYMBOL_GPL(round_jiffies);
437
438/**
439 * round_jiffies_relative - function to round jiffies to a full second
440 * @j: the time in (relative) jiffies that should be rounded
441 *
442 * round_jiffies_relative() rounds a time delta in the future (in jiffies)
443 * up or down to (approximately) full seconds. This is useful for timers
444 * for which the exact time they fire does not matter too much, as long as
445 * they fire approximately every X seconds.
446 *
447 * By rounding these timers to whole seconds, all such timers will fire
448 * at the same time, rather than at various times spread out. The goal
449 * of this is to have the CPU wake up less, which saves power.
450 *
451 * The return value is the rounded version of the @j parameter.
452 */
453unsigned long round_jiffies_relative(unsigned long j)
454{
455 return __round_jiffies_relative(j, raw_smp_processor_id());
456}
457EXPORT_SYMBOL_GPL(round_jiffies_relative);
458
459/**
460 * __round_jiffies_up_relative - function to round jiffies up to a full second
461 * @j: the time in (relative) jiffies that should be rounded
462 * @cpu: the processor number on which the timeout will happen
463 *
464 * This is the same as __round_jiffies_relative() except that it will never
465 * round down. This is useful for timeouts for which the exact time
466 * of firing does not matter too much, as long as they don't fire too
467 * early.
468 */
469unsigned long __round_jiffies_up_relative(unsigned long j, int cpu)
470{
471 unsigned long j0 = jiffies;
472
473 /* Use j0 because jiffies might change while we run */
474 return round_jiffies_common(j: j + j0, cpu, force_up: true) - j0;
475}
476EXPORT_SYMBOL_GPL(__round_jiffies_up_relative);
477
478/**
479 * round_jiffies_up - function to round jiffies up to a full second
480 * @j: the time in (absolute) jiffies that should be rounded
481 *
482 * This is the same as round_jiffies() except that it will never
483 * round down. This is useful for timeouts for which the exact time
484 * of firing does not matter too much, as long as they don't fire too
485 * early.
486 */
487unsigned long round_jiffies_up(unsigned long j)
488{
489 return round_jiffies_common(j, raw_smp_processor_id(), force_up: true);
490}
491EXPORT_SYMBOL_GPL(round_jiffies_up);
492
493/**
494 * round_jiffies_up_relative - function to round jiffies up to a full second
495 * @j: the time in (relative) jiffies that should be rounded
496 *
497 * This is the same as round_jiffies_relative() except that it will never
498 * round down. This is useful for timeouts for which the exact time
499 * of firing does not matter too much, as long as they don't fire too
500 * early.
501 */
502unsigned long round_jiffies_up_relative(unsigned long j)
503{
504 return __round_jiffies_up_relative(j, raw_smp_processor_id());
505}
506EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
507
508
509static inline unsigned int timer_get_idx(struct timer_list *timer)
510{
511 return (timer->flags & TIMER_ARRAYMASK) >> TIMER_ARRAYSHIFT;
512}
513
514static inline void timer_set_idx(struct timer_list *timer, unsigned int idx)
515{
516 timer->flags = (timer->flags & ~TIMER_ARRAYMASK) |
517 idx << TIMER_ARRAYSHIFT;
518}
519
520/*
521 * Helper function to calculate the array index for a given expiry
522 * time.
523 */
524static inline unsigned calc_index(unsigned long expires, unsigned lvl,
525 unsigned long *bucket_expiry)
526{
527
528 /*
529 * The timer wheel has to guarantee that a timer does not fire
530 * early. Early expiry can happen due to:
531 * - Timer is armed at the edge of a tick
532 * - Truncation of the expiry time in the outer wheel levels
533 *
534 * Round up with level granularity to prevent this.
535 */
536 expires = (expires >> LVL_SHIFT(lvl)) + 1;
537 *bucket_expiry = expires << LVL_SHIFT(lvl);
538 return LVL_OFFS(lvl) + (expires & LVL_MASK);
539}
540
541static int calc_wheel_index(unsigned long expires, unsigned long clk,
542 unsigned long *bucket_expiry)
543{
544 unsigned long delta = expires - clk;
545 unsigned int idx;
546
547 if (delta < LVL_START(1)) {
548 idx = calc_index(expires, lvl: 0, bucket_expiry);
549 } else if (delta < LVL_START(2)) {
550 idx = calc_index(expires, lvl: 1, bucket_expiry);
551 } else if (delta < LVL_START(3)) {
552 idx = calc_index(expires, lvl: 2, bucket_expiry);
553 } else if (delta < LVL_START(4)) {
554 idx = calc_index(expires, lvl: 3, bucket_expiry);
555 } else if (delta < LVL_START(5)) {
556 idx = calc_index(expires, lvl: 4, bucket_expiry);
557 } else if (delta < LVL_START(6)) {
558 idx = calc_index(expires, lvl: 5, bucket_expiry);
559 } else if (delta < LVL_START(7)) {
560 idx = calc_index(expires, lvl: 6, bucket_expiry);
561 } else if (LVL_DEPTH > 8 && delta < LVL_START(8)) {
562 idx = calc_index(expires, lvl: 7, bucket_expiry);
563 } else if ((long) delta < 0) {
564 idx = clk & LVL_MASK;
565 *bucket_expiry = clk;
566 } else {
567 /*
568 * Force expire obscene large timeouts to expire at the
569 * capacity limit of the wheel.
570 */
571 if (delta >= WHEEL_TIMEOUT_CUTOFF)
572 expires = clk + WHEEL_TIMEOUT_MAX;
573
574 idx = calc_index(expires, LVL_DEPTH - 1, bucket_expiry);
575 }
576 return idx;
577}
578
579static void
580trigger_dyntick_cpu(struct timer_base *base, struct timer_list *timer)
581{
582 /*
583 * Deferrable timers do not prevent the CPU from entering dynticks and
584 * are not taken into account on the idle/nohz_full path. An IPI when a
585 * new deferrable timer is enqueued will wake up the remote CPU but
586 * nothing will be done with the deferrable timer base. Therefore skip
587 * the remote IPI for deferrable timers completely.
588 */
589 if (!is_timers_nohz_active() || timer->flags & TIMER_DEFERRABLE)
590 return;
591
592 /*
593 * We might have to IPI the remote CPU if the base is idle and the
594 * timer is pinned. If it is a non pinned timer, it is only queued
595 * on the remote CPU, when timer was running during queueing. Then
596 * everything is handled by remote CPU anyway. If the other CPU is
597 * on the way to idle then it can't set base->is_idle as we hold
598 * the base lock:
599 */
600 if (base->is_idle) {
601 WARN_ON_ONCE(!(timer->flags & TIMER_PINNED ||
602 tick_nohz_full_cpu(base->cpu)));
603 wake_up_nohz_cpu(cpu: base->cpu);
604 }
605}
606
607/*
608 * Enqueue the timer into the hash bucket, mark it pending in
609 * the bitmap, store the index in the timer flags then wake up
610 * the target CPU if needed.
611 */
612static void enqueue_timer(struct timer_base *base, struct timer_list *timer,
613 unsigned int idx, unsigned long bucket_expiry)
614{
615
616 hlist_add_head(n: &timer->entry, h: base->vectors + idx);
617 __set_bit(idx, base->pending_map);
618 timer_set_idx(timer, idx);
619
620 trace_timer_start(timer, bucket_expiry);
621
622 /*
623 * Check whether this is the new first expiring timer. The
624 * effective expiry time of the timer is required here
625 * (bucket_expiry) instead of timer->expires.
626 */
627 if (time_before(bucket_expiry, base->next_expiry)) {
628 /*
629 * Set the next expiry time and kick the CPU so it
630 * can reevaluate the wheel:
631 */
632 WRITE_ONCE(base->next_expiry, bucket_expiry);
633 base->timers_pending = true;
634 base->next_expiry_recalc = false;
635 trigger_dyntick_cpu(base, timer);
636 }
637}
638
639static void internal_add_timer(struct timer_base *base, struct timer_list *timer)
640{
641 unsigned long bucket_expiry;
642 unsigned int idx;
643
644 idx = calc_wheel_index(expires: timer->expires, clk: base->clk, bucket_expiry: &bucket_expiry);
645 enqueue_timer(base, timer, idx, bucket_expiry);
646}
647
648#ifdef CONFIG_DEBUG_OBJECTS_TIMERS
649
650static const struct debug_obj_descr timer_debug_descr;
651
652struct timer_hint {
653 void (*function)(struct timer_list *t);
654 long offset;
655};
656
657#define TIMER_HINT(fn, container, timr, hintfn) \
658 { \
659 .function = fn, \
660 .offset = offsetof(container, hintfn) - \
661 offsetof(container, timr) \
662 }
663
664static const struct timer_hint timer_hints[] = {
665 TIMER_HINT(delayed_work_timer_fn,
666 struct delayed_work, timer, work.func),
667 TIMER_HINT(kthread_delayed_work_timer_fn,
668 struct kthread_delayed_work, timer, work.func),
669};
670
671static void *timer_debug_hint(void *addr)
672{
673 struct timer_list *timer = addr;
674 int i;
675
676 for (i = 0; i < ARRAY_SIZE(timer_hints); i++) {
677 if (timer_hints[i].function == timer->function) {
678 void (**fn)(void) = addr + timer_hints[i].offset;
679
680 return *fn;
681 }
682 }
683
684 return timer->function;
685}
686
687static bool timer_is_static_object(void *addr)
688{
689 struct timer_list *timer = addr;
690
691 return (timer->entry.pprev == NULL &&
692 timer->entry.next == TIMER_ENTRY_STATIC);
693}
694
695/*
696 * timer_fixup_init is called when:
697 * - an active object is initialized
698 */
699static bool timer_fixup_init(void *addr, enum debug_obj_state state)
700{
701 struct timer_list *timer = addr;
702
703 switch (state) {
704 case ODEBUG_STATE_ACTIVE:
705 timer_delete_sync(timer);
706 debug_object_init(timer, &timer_debug_descr);
707 return true;
708 default:
709 return false;
710 }
711}
712
713/* Stub timer callback for improperly used timers. */
714static void stub_timer(struct timer_list *unused)
715{
716 WARN_ON(1);
717}
718
719/*
720 * timer_fixup_activate is called when:
721 * - an active object is activated
722 * - an unknown non-static object is activated
723 */
724static bool timer_fixup_activate(void *addr, enum debug_obj_state state)
725{
726 struct timer_list *timer = addr;
727
728 switch (state) {
729 case ODEBUG_STATE_NOTAVAILABLE:
730 timer_setup(timer, stub_timer, 0);
731 return true;
732
733 case ODEBUG_STATE_ACTIVE:
734 WARN_ON(1);
735 fallthrough;
736 default:
737 return false;
738 }
739}
740
741/*
742 * timer_fixup_free is called when:
743 * - an active object is freed
744 */
745static bool timer_fixup_free(void *addr, enum debug_obj_state state)
746{
747 struct timer_list *timer = addr;
748
749 switch (state) {
750 case ODEBUG_STATE_ACTIVE:
751 timer_delete_sync(timer);
752 debug_object_free(timer, &timer_debug_descr);
753 return true;
754 default:
755 return false;
756 }
757}
758
759/*
760 * timer_fixup_assert_init is called when:
761 * - an untracked/uninit-ed object is found
762 */
763static bool timer_fixup_assert_init(void *addr, enum debug_obj_state state)
764{
765 struct timer_list *timer = addr;
766
767 switch (state) {
768 case ODEBUG_STATE_NOTAVAILABLE:
769 timer_setup(timer, stub_timer, 0);
770 return true;
771 default:
772 return false;
773 }
774}
775
776static const struct debug_obj_descr timer_debug_descr = {
777 .name = "timer_list",
778 .debug_hint = timer_debug_hint,
779 .is_static_object = timer_is_static_object,
780 .fixup_init = timer_fixup_init,
781 .fixup_activate = timer_fixup_activate,
782 .fixup_free = timer_fixup_free,
783 .fixup_assert_init = timer_fixup_assert_init,
784};
785
786static inline void debug_timer_init(struct timer_list *timer)
787{
788 debug_object_init(timer, &timer_debug_descr);
789}
790
791static inline void debug_timer_activate(struct timer_list *timer)
792{
793 debug_object_activate(timer, &timer_debug_descr);
794}
795
796static inline void debug_timer_deactivate(struct timer_list *timer)
797{
798 debug_object_deactivate(timer, &timer_debug_descr);
799}
800
801static inline void debug_timer_assert_init(struct timer_list *timer)
802{
803 debug_object_assert_init(timer, &timer_debug_descr);
804}
805
806static void do_init_timer(struct timer_list *timer,
807 void (*func)(struct timer_list *),
808 unsigned int flags,
809 const char *name, struct lock_class_key *key);
810
811void timer_init_key_on_stack(struct timer_list *timer,
812 void (*func)(struct timer_list *),
813 unsigned int flags,
814 const char *name, struct lock_class_key *key)
815{
816 debug_object_init_on_stack(timer, &timer_debug_descr);
817 do_init_timer(timer, func, flags, name, key);
818}
819EXPORT_SYMBOL_GPL(timer_init_key_on_stack);
820
821void timer_destroy_on_stack(struct timer_list *timer)
822{
823 debug_object_free(timer, &timer_debug_descr);
824}
825EXPORT_SYMBOL_GPL(timer_destroy_on_stack);
826
827#else
828static inline void debug_timer_init(struct timer_list *timer) { }
829static inline void debug_timer_activate(struct timer_list *timer) { }
830static inline void debug_timer_deactivate(struct timer_list *timer) { }
831static inline void debug_timer_assert_init(struct timer_list *timer) { }
832#endif
833
834static inline void debug_init(struct timer_list *timer)
835{
836 debug_timer_init(timer);
837 trace_timer_init(timer);
838}
839
840static inline void debug_deactivate(struct timer_list *timer)
841{
842 debug_timer_deactivate(timer);
843 trace_timer_cancel(timer);
844}
845
846static inline void debug_assert_init(struct timer_list *timer)
847{
848 debug_timer_assert_init(timer);
849}
850
851static void do_init_timer(struct timer_list *timer,
852 void (*func)(struct timer_list *),
853 unsigned int flags,
854 const char *name, struct lock_class_key *key)
855{
856 timer->entry.pprev = NULL;
857 timer->function = func;
858 if (WARN_ON_ONCE(flags & ~TIMER_INIT_FLAGS))
859 flags &= TIMER_INIT_FLAGS;
860 timer->flags = flags | raw_smp_processor_id();
861 lockdep_init_map(&timer->lockdep_map, name, key, 0);
862}
863
864/**
865 * timer_init_key - initialize a timer
866 * @timer: the timer to be initialized
867 * @func: timer callback function
868 * @flags: timer flags
869 * @name: name of the timer
870 * @key: lockdep class key of the fake lock used for tracking timer
871 * sync lock dependencies
872 *
873 * timer_init_key() must be done to a timer prior to calling *any* of the
874 * other timer functions.
875 */
876void timer_init_key(struct timer_list *timer,
877 void (*func)(struct timer_list *), unsigned int flags,
878 const char *name, struct lock_class_key *key)
879{
880 debug_init(timer);
881 do_init_timer(timer, func, flags, name, key);
882}
883EXPORT_SYMBOL(timer_init_key);
884
885static inline void detach_timer(struct timer_list *timer, bool clear_pending)
886{
887 struct hlist_node *entry = &timer->entry;
888
889 debug_deactivate(timer);
890
891 __hlist_del(n: entry);
892 if (clear_pending)
893 entry->pprev = NULL;
894 entry->next = LIST_POISON2;
895}
896
897static int detach_if_pending(struct timer_list *timer, struct timer_base *base,
898 bool clear_pending)
899{
900 unsigned idx = timer_get_idx(timer);
901
902 if (!timer_pending(timer))
903 return 0;
904
905 if (hlist_is_singular_node(n: &timer->entry, h: base->vectors + idx)) {
906 __clear_bit(idx, base->pending_map);
907 base->next_expiry_recalc = true;
908 }
909
910 detach_timer(timer, clear_pending);
911 return 1;
912}
913
914static inline struct timer_base *get_timer_cpu_base(u32 tflags, u32 cpu)
915{
916 int index = tflags & TIMER_PINNED ? BASE_LOCAL : BASE_GLOBAL;
917
918 /*
919 * If the timer is deferrable and NO_HZ_COMMON is set then we need
920 * to use the deferrable base.
921 */
922 if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE))
923 index = BASE_DEF;
924
925 return per_cpu_ptr(&timer_bases[index], cpu);
926}
927
928static inline struct timer_base *get_timer_this_cpu_base(u32 tflags)
929{
930 int index = tflags & TIMER_PINNED ? BASE_LOCAL : BASE_GLOBAL;
931
932 /*
933 * If the timer is deferrable and NO_HZ_COMMON is set then we need
934 * to use the deferrable base.
935 */
936 if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE))
937 index = BASE_DEF;
938
939 return this_cpu_ptr(&timer_bases[index]);
940}
941
942static inline struct timer_base *get_timer_base(u32 tflags)
943{
944 return get_timer_cpu_base(tflags, cpu: tflags & TIMER_CPUMASK);
945}
946
947static inline void __forward_timer_base(struct timer_base *base,
948 unsigned long basej)
949{
950 /*
951 * Check whether we can forward the base. We can only do that when
952 * @basej is past base->clk otherwise we might rewind base->clk.
953 */
954 if (time_before_eq(basej, base->clk))
955 return;
956
957 /*
958 * If the next expiry value is > jiffies, then we fast forward to
959 * jiffies otherwise we forward to the next expiry value.
960 */
961 if (time_after(base->next_expiry, basej)) {
962 base->clk = basej;
963 } else {
964 if (WARN_ON_ONCE(time_before(base->next_expiry, base->clk)))
965 return;
966 base->clk = base->next_expiry;
967 }
968
969}
970
971static inline void forward_timer_base(struct timer_base *base)
972{
973 __forward_timer_base(base, READ_ONCE(jiffies));
974}
975
976/*
977 * We are using hashed locking: Holding per_cpu(timer_bases[x]).lock means
978 * that all timers which are tied to this base are locked, and the base itself
979 * is locked too.
980 *
981 * So __run_timers/migrate_timers can safely modify all timers which could
982 * be found in the base->vectors array.
983 *
984 * When a timer is migrating then the TIMER_MIGRATING flag is set and we need
985 * to wait until the migration is done.
986 */
987static struct timer_base *lock_timer_base(struct timer_list *timer,
988 unsigned long *flags)
989 __acquires(timer->base->lock)
990{
991 for (;;) {
992 struct timer_base *base;
993 u32 tf;
994
995 /*
996 * We need to use READ_ONCE() here, otherwise the compiler
997 * might re-read @tf between the check for TIMER_MIGRATING
998 * and spin_lock().
999 */
1000 tf = READ_ONCE(timer->flags);
1001
1002 if (!(tf & TIMER_MIGRATING)) {
1003 base = get_timer_base(tflags: tf);
1004 raw_spin_lock_irqsave(&base->lock, *flags);
1005 if (timer->flags == tf)
1006 return base;
1007 raw_spin_unlock_irqrestore(&base->lock, *flags);
1008 }
1009 cpu_relax();
1010 }
1011}
1012
1013#define MOD_TIMER_PENDING_ONLY 0x01
1014#define MOD_TIMER_REDUCE 0x02
1015#define MOD_TIMER_NOTPENDING 0x04
1016
1017static inline int
1018__mod_timer(struct timer_list *timer, unsigned long expires, unsigned int options)
1019{
1020 unsigned long clk = 0, flags, bucket_expiry;
1021 struct timer_base *base, *new_base;
1022 unsigned int idx = UINT_MAX;
1023 int ret = 0;
1024
1025 debug_assert_init(timer);
1026
1027 /*
1028 * This is a common optimization triggered by the networking code - if
1029 * the timer is re-modified to have the same timeout or ends up in the
1030 * same array bucket then just return:
1031 */
1032 if (!(options & MOD_TIMER_NOTPENDING) && timer_pending(timer)) {
1033 /*
1034 * The downside of this optimization is that it can result in
1035 * larger granularity than you would get from adding a new
1036 * timer with this expiry.
1037 */
1038 long diff = timer->expires - expires;
1039
1040 if (!diff)
1041 return 1;
1042 if (options & MOD_TIMER_REDUCE && diff <= 0)
1043 return 1;
1044
1045 /*
1046 * We lock timer base and calculate the bucket index right
1047 * here. If the timer ends up in the same bucket, then we
1048 * just update the expiry time and avoid the whole
1049 * dequeue/enqueue dance.
1050 */
1051 base = lock_timer_base(timer, flags: &flags);
1052 /*
1053 * Has @timer been shutdown? This needs to be evaluated
1054 * while holding base lock to prevent a race against the
1055 * shutdown code.
1056 */
1057 if (!timer->function)
1058 goto out_unlock;
1059
1060 forward_timer_base(base);
1061
1062 if (timer_pending(timer) && (options & MOD_TIMER_REDUCE) &&
1063 time_before_eq(timer->expires, expires)) {
1064 ret = 1;
1065 goto out_unlock;
1066 }
1067
1068 clk = base->clk;
1069 idx = calc_wheel_index(expires, clk, bucket_expiry: &bucket_expiry);
1070
1071 /*
1072 * Retrieve and compare the array index of the pending
1073 * timer. If it matches set the expiry to the new value so a
1074 * subsequent call will exit in the expires check above.
1075 */
1076 if (idx == timer_get_idx(timer)) {
1077 if (!(options & MOD_TIMER_REDUCE))
1078 timer->expires = expires;
1079 else if (time_after(timer->expires, expires))
1080 timer->expires = expires;
1081 ret = 1;
1082 goto out_unlock;
1083 }
1084 } else {
1085 base = lock_timer_base(timer, flags: &flags);
1086 /*
1087 * Has @timer been shutdown? This needs to be evaluated
1088 * while holding base lock to prevent a race against the
1089 * shutdown code.
1090 */
1091 if (!timer->function)
1092 goto out_unlock;
1093
1094 forward_timer_base(base);
1095 }
1096
1097 ret = detach_if_pending(timer, base, clear_pending: false);
1098 if (!ret && (options & MOD_TIMER_PENDING_ONLY))
1099 goto out_unlock;
1100
1101 new_base = get_timer_this_cpu_base(tflags: timer->flags);
1102
1103 if (base != new_base) {
1104 /*
1105 * We are trying to schedule the timer on the new base.
1106 * However we can't change timer's base while it is running,
1107 * otherwise timer_delete_sync() can't detect that the timer's
1108 * handler yet has not finished. This also guarantees that the
1109 * timer is serialized wrt itself.
1110 */
1111 if (likely(base->running_timer != timer)) {
1112 /* See the comment in lock_timer_base() */
1113 timer->flags |= TIMER_MIGRATING;
1114
1115 raw_spin_unlock(&base->lock);
1116 base = new_base;
1117 raw_spin_lock(&base->lock);
1118 WRITE_ONCE(timer->flags,
1119 (timer->flags & ~TIMER_BASEMASK) | base->cpu);
1120 forward_timer_base(base);
1121 }
1122 }
1123
1124 debug_timer_activate(timer);
1125
1126 timer->expires = expires;
1127 /*
1128 * If 'idx' was calculated above and the base time did not advance
1129 * between calculating 'idx' and possibly switching the base, only
1130 * enqueue_timer() is required. Otherwise we need to (re)calculate
1131 * the wheel index via internal_add_timer().
1132 */
1133 if (idx != UINT_MAX && clk == base->clk)
1134 enqueue_timer(base, timer, idx, bucket_expiry);
1135 else
1136 internal_add_timer(base, timer);
1137
1138out_unlock:
1139 raw_spin_unlock_irqrestore(&base->lock, flags);
1140
1141 return ret;
1142}
1143
1144/**
1145 * mod_timer_pending - Modify a pending timer's timeout
1146 * @timer: The pending timer to be modified
1147 * @expires: New absolute timeout in jiffies
1148 *
1149 * mod_timer_pending() is the same for pending timers as mod_timer(), but
1150 * will not activate inactive timers.
1151 *
1152 * If @timer->function == NULL then the start operation is silently
1153 * discarded.
1154 *
1155 * Return:
1156 * * %0 - The timer was inactive and not modified or was in
1157 * shutdown state and the operation was discarded
1158 * * %1 - The timer was active and requeued to expire at @expires
1159 */
1160int mod_timer_pending(struct timer_list *timer, unsigned long expires)
1161{
1162 return __mod_timer(timer, expires, MOD_TIMER_PENDING_ONLY);
1163}
1164EXPORT_SYMBOL(mod_timer_pending);
1165
1166/**
1167 * mod_timer - Modify a timer's timeout
1168 * @timer: The timer to be modified
1169 * @expires: New absolute timeout in jiffies
1170 *
1171 * mod_timer(timer, expires) is equivalent to:
1172 *
1173 * timer_delete(timer); timer->expires = expires; add_timer(timer);
1174 *
1175 * mod_timer() is more efficient than the above open coded sequence. In
1176 * case that the timer is inactive, the timer_delete() part is a NOP. The
1177 * timer is in any case activated with the new expiry time @expires.
1178 *
1179 * Note that if there are multiple unserialized concurrent users of the
1180 * same timer, then mod_timer() is the only safe way to modify the timeout,
1181 * since add_timer() cannot modify an already running timer.
1182 *
1183 * If @timer->function == NULL then the start operation is silently
1184 * discarded. In this case the return value is 0 and meaningless.
1185 *
1186 * Return:
1187 * * %0 - The timer was inactive and started or was in shutdown
1188 * state and the operation was discarded
1189 * * %1 - The timer was active and requeued to expire at @expires or
1190 * the timer was active and not modified because @expires did
1191 * not change the effective expiry time
1192 */
1193int mod_timer(struct timer_list *timer, unsigned long expires)
1194{
1195 return __mod_timer(timer, expires, options: 0);
1196}
1197EXPORT_SYMBOL(mod_timer);
1198
1199/**
1200 * timer_reduce - Modify a timer's timeout if it would reduce the timeout
1201 * @timer: The timer to be modified
1202 * @expires: New absolute timeout in jiffies
1203 *
1204 * timer_reduce() is very similar to mod_timer(), except that it will only
1205 * modify an enqueued timer if that would reduce the expiration time. If
1206 * @timer is not enqueued it starts the timer.
1207 *
1208 * If @timer->function == NULL then the start operation is silently
1209 * discarded.
1210 *
1211 * Return:
1212 * * %0 - The timer was inactive and started or was in shutdown
1213 * state and the operation was discarded
1214 * * %1 - The timer was active and requeued to expire at @expires or
1215 * the timer was active and not modified because @expires
1216 * did not change the effective expiry time such that the
1217 * timer would expire earlier than already scheduled
1218 */
1219int timer_reduce(struct timer_list *timer, unsigned long expires)
1220{
1221 return __mod_timer(timer, expires, MOD_TIMER_REDUCE);
1222}
1223EXPORT_SYMBOL(timer_reduce);
1224
1225/**
1226 * add_timer - Start a timer
1227 * @timer: The timer to be started
1228 *
1229 * Start @timer to expire at @timer->expires in the future. @timer->expires
1230 * is the absolute expiry time measured in 'jiffies'. When the timer expires
1231 * timer->function(timer) will be invoked from soft interrupt context.
1232 *
1233 * The @timer->expires and @timer->function fields must be set prior
1234 * to calling this function.
1235 *
1236 * If @timer->function == NULL then the start operation is silently
1237 * discarded.
1238 *
1239 * If @timer->expires is already in the past @timer will be queued to
1240 * expire at the next timer tick.
1241 *
1242 * This can only operate on an inactive timer. Attempts to invoke this on
1243 * an active timer are rejected with a warning.
1244 */
1245void add_timer(struct timer_list *timer)
1246{
1247 if (WARN_ON_ONCE(timer_pending(timer)))
1248 return;
1249 __mod_timer(timer, expires: timer->expires, MOD_TIMER_NOTPENDING);
1250}
1251EXPORT_SYMBOL(add_timer);
1252
1253/**
1254 * add_timer_local() - Start a timer on the local CPU
1255 * @timer: The timer to be started
1256 *
1257 * Same as add_timer() except that the timer flag TIMER_PINNED is set.
1258 *
1259 * See add_timer() for further details.
1260 */
1261void add_timer_local(struct timer_list *timer)
1262{
1263 if (WARN_ON_ONCE(timer_pending(timer)))
1264 return;
1265 timer->flags |= TIMER_PINNED;
1266 __mod_timer(timer, expires: timer->expires, MOD_TIMER_NOTPENDING);
1267}
1268EXPORT_SYMBOL(add_timer_local);
1269
1270/**
1271 * add_timer_global() - Start a timer without TIMER_PINNED flag set
1272 * @timer: The timer to be started
1273 *
1274 * Same as add_timer() except that the timer flag TIMER_PINNED is unset.
1275 *
1276 * See add_timer() for further details.
1277 */
1278void add_timer_global(struct timer_list *timer)
1279{
1280 if (WARN_ON_ONCE(timer_pending(timer)))
1281 return;
1282 timer->flags &= ~TIMER_PINNED;
1283 __mod_timer(timer, expires: timer->expires, MOD_TIMER_NOTPENDING);
1284}
1285EXPORT_SYMBOL(add_timer_global);
1286
1287/**
1288 * add_timer_on - Start a timer on a particular CPU
1289 * @timer: The timer to be started
1290 * @cpu: The CPU to start it on
1291 *
1292 * Same as add_timer() except that it starts the timer on the given CPU and
1293 * the TIMER_PINNED flag is set. When timer shouldn't be a pinned timer in
1294 * the next round, add_timer_global() should be used instead as it unsets
1295 * the TIMER_PINNED flag.
1296 *
1297 * See add_timer() for further details.
1298 */
1299void add_timer_on(struct timer_list *timer, int cpu)
1300{
1301 struct timer_base *new_base, *base;
1302 unsigned long flags;
1303
1304 debug_assert_init(timer);
1305
1306 if (WARN_ON_ONCE(timer_pending(timer)))
1307 return;
1308
1309 /* Make sure timer flags have TIMER_PINNED flag set */
1310 timer->flags |= TIMER_PINNED;
1311
1312 new_base = get_timer_cpu_base(tflags: timer->flags, cpu);
1313
1314 /*
1315 * If @timer was on a different CPU, it should be migrated with the
1316 * old base locked to prevent other operations proceeding with the
1317 * wrong base locked. See lock_timer_base().
1318 */
1319 base = lock_timer_base(timer, flags: &flags);
1320 /*
1321 * Has @timer been shutdown? This needs to be evaluated while
1322 * holding base lock to prevent a race against the shutdown code.
1323 */
1324 if (!timer->function)
1325 goto out_unlock;
1326
1327 if (base != new_base) {
1328 timer->flags |= TIMER_MIGRATING;
1329
1330 raw_spin_unlock(&base->lock);
1331 base = new_base;
1332 raw_spin_lock(&base->lock);
1333 WRITE_ONCE(timer->flags,
1334 (timer->flags & ~TIMER_BASEMASK) | cpu);
1335 }
1336 forward_timer_base(base);
1337
1338 debug_timer_activate(timer);
1339 internal_add_timer(base, timer);
1340out_unlock:
1341 raw_spin_unlock_irqrestore(&base->lock, flags);
1342}
1343EXPORT_SYMBOL_GPL(add_timer_on);
1344
1345/**
1346 * __timer_delete - Internal function: Deactivate a timer
1347 * @timer: The timer to be deactivated
1348 * @shutdown: If true, this indicates that the timer is about to be
1349 * shutdown permanently.
1350 *
1351 * If @shutdown is true then @timer->function is set to NULL under the
1352 * timer base lock which prevents further rearming of the time. In that
1353 * case any attempt to rearm @timer after this function returns will be
1354 * silently ignored.
1355 *
1356 * Return:
1357 * * %0 - The timer was not pending
1358 * * %1 - The timer was pending and deactivated
1359 */
1360static int __timer_delete(struct timer_list *timer, bool shutdown)
1361{
1362 struct timer_base *base;
1363 unsigned long flags;
1364 int ret = 0;
1365
1366 debug_assert_init(timer);
1367
1368 /*
1369 * If @shutdown is set then the lock has to be taken whether the
1370 * timer is pending or not to protect against a concurrent rearm
1371 * which might hit between the lockless pending check and the lock
1372 * acquisition. By taking the lock it is ensured that such a newly
1373 * enqueued timer is dequeued and cannot end up with
1374 * timer->function == NULL in the expiry code.
1375 *
1376 * If timer->function is currently executed, then this makes sure
1377 * that the callback cannot requeue the timer.
1378 */
1379 if (timer_pending(timer) || shutdown) {
1380 base = lock_timer_base(timer, flags: &flags);
1381 ret = detach_if_pending(timer, base, clear_pending: true);
1382 if (shutdown)
1383 timer->function = NULL;
1384 raw_spin_unlock_irqrestore(&base->lock, flags);
1385 }
1386
1387 return ret;
1388}
1389
1390/**
1391 * timer_delete - Deactivate a timer
1392 * @timer: The timer to be deactivated
1393 *
1394 * The function only deactivates a pending timer, but contrary to
1395 * timer_delete_sync() it does not take into account whether the timer's
1396 * callback function is concurrently executed on a different CPU or not.
1397 * It neither prevents rearming of the timer. If @timer can be rearmed
1398 * concurrently then the return value of this function is meaningless.
1399 *
1400 * Return:
1401 * * %0 - The timer was not pending
1402 * * %1 - The timer was pending and deactivated
1403 */
1404int timer_delete(struct timer_list *timer)
1405{
1406 return __timer_delete(timer, shutdown: false);
1407}
1408EXPORT_SYMBOL(timer_delete);
1409
1410/**
1411 * timer_shutdown - Deactivate a timer and prevent rearming
1412 * @timer: The timer to be deactivated
1413 *
1414 * The function does not wait for an eventually running timer callback on a
1415 * different CPU but it prevents rearming of the timer. Any attempt to arm
1416 * @timer after this function returns will be silently ignored.
1417 *
1418 * This function is useful for teardown code and should only be used when
1419 * timer_shutdown_sync() cannot be invoked due to locking or context constraints.
1420 *
1421 * Return:
1422 * * %0 - The timer was not pending
1423 * * %1 - The timer was pending
1424 */
1425int timer_shutdown(struct timer_list *timer)
1426{
1427 return __timer_delete(timer, shutdown: true);
1428}
1429EXPORT_SYMBOL_GPL(timer_shutdown);
1430
1431/**
1432 * __try_to_del_timer_sync - Internal function: Try to deactivate a timer
1433 * @timer: Timer to deactivate
1434 * @shutdown: If true, this indicates that the timer is about to be
1435 * shutdown permanently.
1436 *
1437 * If @shutdown is true then @timer->function is set to NULL under the
1438 * timer base lock which prevents further rearming of the timer. Any
1439 * attempt to rearm @timer after this function returns will be silently
1440 * ignored.
1441 *
1442 * This function cannot guarantee that the timer cannot be rearmed
1443 * right after dropping the base lock if @shutdown is false. That
1444 * needs to be prevented by the calling code if necessary.
1445 *
1446 * Return:
1447 * * %0 - The timer was not pending
1448 * * %1 - The timer was pending and deactivated
1449 * * %-1 - The timer callback function is running on a different CPU
1450 */
1451static int __try_to_del_timer_sync(struct timer_list *timer, bool shutdown)
1452{
1453 struct timer_base *base;
1454 unsigned long flags;
1455 int ret = -1;
1456
1457 debug_assert_init(timer);
1458
1459 base = lock_timer_base(timer, flags: &flags);
1460
1461 if (base->running_timer != timer)
1462 ret = detach_if_pending(timer, base, clear_pending: true);
1463 if (shutdown)
1464 timer->function = NULL;
1465
1466 raw_spin_unlock_irqrestore(&base->lock, flags);
1467
1468 return ret;
1469}
1470
1471/**
1472 * timer_delete_sync_try - Try to deactivate a timer
1473 * @timer: Timer to deactivate
1474 *
1475 * This function tries to deactivate a timer. On success the timer is not
1476 * queued and the timer callback function is not running on any CPU.
1477 *
1478 * This function does not guarantee that the timer cannot be rearmed right
1479 * after dropping the base lock. That needs to be prevented by the calling
1480 * code if necessary.
1481 *
1482 * Return:
1483 * * %0 - The timer was not pending
1484 * * %1 - The timer was pending and deactivated
1485 * * %-1 - The timer callback function is running on a different CPU
1486 */
1487int timer_delete_sync_try(struct timer_list *timer)
1488{
1489 return __try_to_del_timer_sync(timer, shutdown: false);
1490}
1491EXPORT_SYMBOL(timer_delete_sync_try);
1492
1493#ifdef CONFIG_PREEMPT_RT
1494static __init void timer_base_init_expiry_lock(struct timer_base *base)
1495{
1496 spin_lock_init(&base->expiry_lock);
1497}
1498
1499static inline void timer_base_lock_expiry(struct timer_base *base)
1500{
1501 spin_lock(&base->expiry_lock);
1502}
1503
1504static inline void timer_base_unlock_expiry(struct timer_base *base)
1505{
1506 spin_unlock(&base->expiry_lock);
1507}
1508
1509/*
1510 * The counterpart to del_timer_wait_running().
1511 *
1512 * If there is a waiter for base->expiry_lock, then it was waiting for the
1513 * timer callback to finish. Drop expiry_lock and reacquire it. That allows
1514 * the waiter to acquire the lock and make progress.
1515 */
1516static void timer_sync_wait_running(struct timer_base *base)
1517 __releases(&base->lock) __releases(&base->expiry_lock)
1518 __acquires(&base->expiry_lock) __acquires(&base->lock)
1519{
1520 if (atomic_read(&base->timer_waiters)) {
1521 raw_spin_unlock_irq(&base->lock);
1522 spin_unlock(&base->expiry_lock);
1523 spin_lock(&base->expiry_lock);
1524 raw_spin_lock_irq(&base->lock);
1525 }
1526}
1527
1528/*
1529 * This function is called on PREEMPT_RT kernels when the fast path
1530 * deletion of a timer failed because the timer callback function was
1531 * running.
1532 *
1533 * This prevents priority inversion, if the softirq thread on a remote CPU
1534 * got preempted, and it prevents a life lock when the task which tries to
1535 * delete a timer preempted the softirq thread running the timer callback
1536 * function.
1537 */
1538static void del_timer_wait_running(struct timer_list *timer)
1539{
1540 u32 tf;
1541
1542 tf = READ_ONCE(timer->flags);
1543 if (!(tf & (TIMER_MIGRATING | TIMER_IRQSAFE))) {
1544 struct timer_base *base = get_timer_base(tf);
1545
1546 /*
1547 * Mark the base as contended and grab the expiry lock,
1548 * which is held by the softirq across the timer
1549 * callback. Drop the lock immediately so the softirq can
1550 * expire the next timer. In theory the timer could already
1551 * be running again, but that's more than unlikely and just
1552 * causes another wait loop.
1553 */
1554 atomic_inc(&base->timer_waiters);
1555 spin_lock_bh(&base->expiry_lock);
1556 atomic_dec(&base->timer_waiters);
1557 spin_unlock_bh(&base->expiry_lock);
1558 }
1559}
1560#else
1561static inline void timer_base_init_expiry_lock(struct timer_base *base) { }
1562static inline void timer_base_lock_expiry(struct timer_base *base) { }
1563static inline void timer_base_unlock_expiry(struct timer_base *base) { }
1564static inline void timer_sync_wait_running(struct timer_base *base) { }
1565static inline void del_timer_wait_running(struct timer_list *timer) { }
1566#endif
1567
1568/**
1569 * __timer_delete_sync - Internal function: Deactivate a timer and wait
1570 * for the handler to finish.
1571 * @timer: The timer to be deactivated
1572 * @shutdown: If true, @timer->function will be set to NULL under the
1573 * timer base lock which prevents rearming of @timer
1574 *
1575 * If @shutdown is not set the timer can be rearmed later. If the timer can
1576 * be rearmed concurrently, i.e. after dropping the base lock then the
1577 * return value is meaningless.
1578 *
1579 * If @shutdown is set then @timer->function is set to NULL under timer
1580 * base lock which prevents rearming of the timer. Any attempt to rearm
1581 * a shutdown timer is silently ignored.
1582 *
1583 * If the timer should be reused after shutdown it has to be initialized
1584 * again.
1585 *
1586 * Return:
1587 * * %0 - The timer was not pending
1588 * * %1 - The timer was pending and deactivated
1589 */
1590static int __timer_delete_sync(struct timer_list *timer, bool shutdown)
1591{
1592 int ret;
1593
1594#ifdef CONFIG_LOCKDEP
1595 unsigned long flags;
1596
1597 /*
1598 * If lockdep gives a backtrace here, please reference
1599 * the synchronization rules above.
1600 */
1601 local_irq_save(flags);
1602 lock_map_acquire(&timer->lockdep_map);
1603 lock_map_release(&timer->lockdep_map);
1604 local_irq_restore(flags);
1605#endif
1606 /*
1607 * don't use it in hardirq context, because it
1608 * could lead to deadlock.
1609 */
1610 WARN_ON(in_hardirq() && !(timer->flags & TIMER_IRQSAFE));
1611
1612 /*
1613 * Must be able to sleep on PREEMPT_RT because of the slowpath in
1614 * del_timer_wait_running().
1615 */
1616 if (IS_ENABLED(CONFIG_PREEMPT_RT) && !(timer->flags & TIMER_IRQSAFE))
1617 lockdep_assert_preemption_enabled();
1618
1619 do {
1620 ret = __try_to_del_timer_sync(timer, shutdown);
1621
1622 if (unlikely(ret < 0)) {
1623 del_timer_wait_running(timer);
1624 cpu_relax();
1625 }
1626 } while (ret < 0);
1627
1628 return ret;
1629}
1630
1631/**
1632 * timer_delete_sync - Deactivate a timer and wait for the handler to finish.
1633 * @timer: The timer to be deactivated
1634 *
1635 * Synchronization rules: Callers must prevent restarting of the timer,
1636 * otherwise this function is meaningless. It must not be called from
1637 * interrupt contexts unless the timer is an irqsafe one. The caller must
1638 * not hold locks which would prevent completion of the timer's callback
1639 * function. The timer's handler must not call add_timer_on(). Upon exit
1640 * the timer is not queued and the handler is not running on any CPU.
1641 *
1642 * For !irqsafe timers, the caller must not hold locks that are held in
1643 * interrupt context. Even if the lock has nothing to do with the timer in
1644 * question. Here's why::
1645 *
1646 * CPU0 CPU1
1647 * ---- ----
1648 * <SOFTIRQ>
1649 * call_timer_fn();
1650 * base->running_timer = mytimer;
1651 * spin_lock_irq(somelock);
1652 * <IRQ>
1653 * spin_lock(somelock);
1654 * timer_delete_sync(mytimer);
1655 * while (base->running_timer == mytimer);
1656 *
1657 * Now timer_delete_sync() will never return and never release somelock.
1658 * The interrupt on the other CPU is waiting to grab somelock but it has
1659 * interrupted the softirq that CPU0 is waiting to finish.
1660 *
1661 * This function cannot guarantee that the timer is not rearmed again by
1662 * some concurrent or preempting code, right after it dropped the base
1663 * lock. If there is the possibility of a concurrent rearm then the return
1664 * value of the function is meaningless.
1665 *
1666 * If such a guarantee is needed, e.g. for teardown situations then use
1667 * timer_shutdown_sync() instead.
1668 *
1669 * Return:
1670 * * %0 - The timer was not pending
1671 * * %1 - The timer was pending and deactivated
1672 */
1673int timer_delete_sync(struct timer_list *timer)
1674{
1675 return __timer_delete_sync(timer, shutdown: false);
1676}
1677EXPORT_SYMBOL(timer_delete_sync);
1678
1679/**
1680 * timer_shutdown_sync - Shutdown a timer and prevent rearming
1681 * @timer: The timer to be shutdown
1682 *
1683 * When the function returns it is guaranteed that:
1684 * - @timer is not queued
1685 * - The callback function of @timer is not running
1686 * - @timer cannot be enqueued again. Any attempt to rearm
1687 * @timer is silently ignored.
1688 *
1689 * See timer_delete_sync() for synchronization rules.
1690 *
1691 * This function is useful for final teardown of an infrastructure where
1692 * the timer is subject to a circular dependency problem.
1693 *
1694 * A common pattern for this is a timer and a workqueue where the timer can
1695 * schedule work and work can arm the timer. On shutdown the workqueue must
1696 * be destroyed and the timer must be prevented from rearming. Unless the
1697 * code has conditionals like 'if (mything->in_shutdown)' to prevent that
1698 * there is no way to get this correct with timer_delete_sync().
1699 *
1700 * timer_shutdown_sync() is solving the problem. The correct ordering of
1701 * calls in this case is:
1702 *
1703 * timer_shutdown_sync(&mything->timer);
1704 * workqueue_destroy(&mything->workqueue);
1705 *
1706 * After this 'mything' can be safely freed.
1707 *
1708 * This obviously implies that the timer is not required to be functional
1709 * for the rest of the shutdown operation.
1710 *
1711 * Return:
1712 * * %0 - The timer was not pending
1713 * * %1 - The timer was pending
1714 */
1715int timer_shutdown_sync(struct timer_list *timer)
1716{
1717 return __timer_delete_sync(timer, shutdown: true);
1718}
1719EXPORT_SYMBOL_GPL(timer_shutdown_sync);
1720
1721static void call_timer_fn(struct timer_list *timer,
1722 void (*fn)(struct timer_list *),
1723 unsigned long baseclk)
1724{
1725 int count = preempt_count();
1726
1727#ifdef CONFIG_LOCKDEP
1728 /*
1729 * It is permissible to free the timer from inside the
1730 * function that is called from it, this we need to take into
1731 * account for lockdep too. To avoid bogus "held lock freed"
1732 * warnings as well as problems when looking into
1733 * timer->lockdep_map, make a copy and use that here.
1734 */
1735 struct lockdep_map lockdep_map;
1736
1737 lockdep_copy_map(&lockdep_map, &timer->lockdep_map);
1738#endif
1739 /*
1740 * Couple the lock chain with the lock chain at
1741 * timer_delete_sync() by acquiring the lock_map around the fn()
1742 * call here and in timer_delete_sync().
1743 */
1744 lock_map_acquire(&lockdep_map);
1745
1746 trace_timer_expire_entry(timer, baseclk);
1747 fn(timer);
1748 trace_timer_expire_exit(timer);
1749
1750 lock_map_release(&lockdep_map);
1751
1752 if (count != preempt_count()) {
1753 WARN_ONCE(1, "timer: %pS preempt leak: %08x -> %08x\n",
1754 fn, count, preempt_count());
1755 /*
1756 * Restore the preempt count. That gives us a decent
1757 * chance to survive and extract information. If the
1758 * callback kept a lock held, bad luck, but not worse
1759 * than the BUG() we had.
1760 */
1761 preempt_count_set(pc: count);
1762 }
1763}
1764
1765static void expire_timers(struct timer_base *base, struct hlist_head *head)
1766{
1767 /*
1768 * This value is required only for tracing. base->clk was
1769 * incremented directly before expire_timers was called. But expiry
1770 * is related to the old base->clk value.
1771 */
1772 unsigned long baseclk = base->clk - 1;
1773
1774 while (!hlist_empty(h: head)) {
1775 struct timer_list *timer;
1776 void (*fn)(struct timer_list *);
1777
1778 timer = hlist_entry(head->first, struct timer_list, entry);
1779
1780 base->running_timer = timer;
1781 detach_timer(timer, clear_pending: true);
1782
1783 fn = timer->function;
1784
1785 if (WARN_ON_ONCE(!fn)) {
1786 /* Should never happen. Emphasis on should! */
1787 base->running_timer = NULL;
1788 continue;
1789 }
1790
1791 if (timer->flags & TIMER_IRQSAFE) {
1792 raw_spin_unlock(&base->lock);
1793 call_timer_fn(timer, fn, baseclk);
1794 raw_spin_lock(&base->lock);
1795 base->running_timer = NULL;
1796 } else {
1797 raw_spin_unlock_irq(&base->lock);
1798 call_timer_fn(timer, fn, baseclk);
1799 raw_spin_lock_irq(&base->lock);
1800 base->running_timer = NULL;
1801 timer_sync_wait_running(base);
1802 }
1803 }
1804}
1805
1806static int collect_expired_timers(struct timer_base *base,
1807 struct hlist_head *heads)
1808{
1809 unsigned long clk = base->clk = base->next_expiry;
1810 struct hlist_head *vec;
1811 int i, levels = 0;
1812 unsigned int idx;
1813
1814 for (i = 0; i < LVL_DEPTH; i++) {
1815 idx = (clk & LVL_MASK) + i * LVL_SIZE;
1816
1817 if (__test_and_clear_bit(idx, base->pending_map)) {
1818 vec = base->vectors + idx;
1819 hlist_move_list(old: vec, new: heads++);
1820 levels++;
1821 }
1822 /* Is it time to look at the next level? */
1823 if (clk & LVL_CLK_MASK)
1824 break;
1825 /* Shift clock for the next level granularity */
1826 clk >>= LVL_CLK_SHIFT;
1827 }
1828 return levels;
1829}
1830
1831/*
1832 * Find the next pending bucket of a level. Search from level start (@offset)
1833 * + @clk upwards and if nothing there, search from start of the level
1834 * (@offset) up to @offset + clk.
1835 */
1836static int next_pending_bucket(struct timer_base *base, unsigned offset,
1837 unsigned clk)
1838{
1839 unsigned pos, start = offset + clk;
1840 unsigned end = offset + LVL_SIZE;
1841
1842 pos = find_next_bit(addr: base->pending_map, size: end, offset: start);
1843 if (pos < end)
1844 return pos - start;
1845
1846 pos = find_next_bit(addr: base->pending_map, size: start, offset);
1847 return pos < start ? pos + LVL_SIZE - start : -1;
1848}
1849
1850/*
1851 * Search the first expiring timer in the various clock levels. Caller must
1852 * hold base->lock.
1853 *
1854 * Store next expiry time in base->next_expiry.
1855 */
1856static void timer_recalc_next_expiry(struct timer_base *base)
1857{
1858 unsigned long clk, next, adj;
1859 unsigned lvl, offset = 0;
1860
1861 next = base->clk + TIMER_NEXT_MAX_DELTA;
1862 clk = base->clk;
1863 for (lvl = 0; lvl < LVL_DEPTH; lvl++, offset += LVL_SIZE) {
1864 int pos = next_pending_bucket(base, offset, clk: clk & LVL_MASK);
1865 unsigned long lvl_clk = clk & LVL_CLK_MASK;
1866
1867 if (pos >= 0) {
1868 unsigned long tmp = clk + (unsigned long) pos;
1869
1870 tmp <<= LVL_SHIFT(lvl);
1871 if (time_before(tmp, next))
1872 next = tmp;
1873
1874 /*
1875 * If the next expiration happens before we reach
1876 * the next level, no need to check further.
1877 */
1878 if (pos <= ((LVL_CLK_DIV - lvl_clk) & LVL_CLK_MASK))
1879 break;
1880 }
1881 /*
1882 * Clock for the next level. If the current level clock lower
1883 * bits are zero, we look at the next level as is. If not we
1884 * need to advance it by one because that's going to be the
1885 * next expiring bucket in that level. base->clk is the next
1886 * expiring jiffy. So in case of:
1887 *
1888 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1889 * 0 0 0 0 0 0
1890 *
1891 * we have to look at all levels @index 0. With
1892 *
1893 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1894 * 0 0 0 0 0 2
1895 *
1896 * LVL0 has the next expiring bucket @index 2. The upper
1897 * levels have the next expiring bucket @index 1.
1898 *
1899 * In case that the propagation wraps the next level the same
1900 * rules apply:
1901 *
1902 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1903 * 0 0 0 0 F 2
1904 *
1905 * So after looking at LVL0 we get:
1906 *
1907 * LVL5 LVL4 LVL3 LVL2 LVL1
1908 * 0 0 0 1 0
1909 *
1910 * So no propagation from LVL1 to LVL2 because that happened
1911 * with the add already, but then we need to propagate further
1912 * from LVL2 to LVL3.
1913 *
1914 * So the simple check whether the lower bits of the current
1915 * level are 0 or not is sufficient for all cases.
1916 */
1917 adj = lvl_clk ? 1 : 0;
1918 clk >>= LVL_CLK_SHIFT;
1919 clk += adj;
1920 }
1921
1922 WRITE_ONCE(base->next_expiry, next);
1923 base->next_expiry_recalc = false;
1924 base->timers_pending = !(next == base->clk + TIMER_NEXT_MAX_DELTA);
1925}
1926
1927#ifdef CONFIG_NO_HZ_COMMON
1928/*
1929 * Check, if the next hrtimer event is before the next timer wheel
1930 * event:
1931 */
1932static u64 cmp_next_hrtimer_event(u64 basem, u64 expires)
1933{
1934 u64 nextevt = hrtimer_get_next_event();
1935
1936 /*
1937 * If high resolution timers are enabled
1938 * hrtimer_get_next_event() returns KTIME_MAX.
1939 */
1940 if (expires <= nextevt)
1941 return expires;
1942
1943 /*
1944 * If the next timer is already expired, return the tick base
1945 * time so the tick is fired immediately.
1946 */
1947 if (nextevt <= basem)
1948 return basem;
1949
1950 /*
1951 * Round up to the next jiffy. High resolution timers are
1952 * off, so the hrtimers are expired in the tick and we need to
1953 * make sure that this tick really expires the timer to avoid
1954 * a ping pong of the nohz stop code.
1955 *
1956 * Use DIV_ROUND_UP_ULL to prevent gcc calling __divdi3
1957 */
1958 return DIV_ROUND_UP_ULL(nextevt, TICK_NSEC) * TICK_NSEC;
1959}
1960
1961static unsigned long next_timer_interrupt(struct timer_base *base,
1962 unsigned long basej)
1963{
1964 if (base->next_expiry_recalc)
1965 timer_recalc_next_expiry(base);
1966
1967 /*
1968 * Move next_expiry for the empty base into the future to prevent an
1969 * unnecessary raise of the timer softirq when the next_expiry value
1970 * will be reached even if there is no timer pending.
1971 *
1972 * This update is also required to make timer_base::next_expiry values
1973 * easy comparable to find out which base holds the first pending timer.
1974 */
1975 if (!base->timers_pending)
1976 WRITE_ONCE(base->next_expiry, basej + TIMER_NEXT_MAX_DELTA);
1977
1978 return base->next_expiry;
1979}
1980
1981static unsigned long fetch_next_timer_interrupt(unsigned long basej, u64 basem,
1982 struct timer_base *base_local,
1983 struct timer_base *base_global,
1984 struct timer_events *tevt)
1985{
1986 unsigned long nextevt, nextevt_local, nextevt_global;
1987 bool local_first;
1988
1989 nextevt_local = next_timer_interrupt(base: base_local, basej);
1990 nextevt_global = next_timer_interrupt(base: base_global, basej);
1991
1992 local_first = time_before_eq(nextevt_local, nextevt_global);
1993
1994 nextevt = local_first ? nextevt_local : nextevt_global;
1995
1996 /*
1997 * If the @nextevt is at max. one tick away, use @nextevt and store
1998 * it in the local expiry value. The next global event is irrelevant in
1999 * this case and can be left as KTIME_MAX.
2000 */
2001 if (time_before_eq(nextevt, basej + 1)) {
2002 /* If we missed a tick already, force 0 delta */
2003 if (time_before(nextevt, basej))
2004 nextevt = basej;
2005 tevt->local = basem + (u64)(nextevt - basej) * TICK_NSEC;
2006
2007 /*
2008 * This is required for the remote check only but it doesn't
2009 * hurt, when it is done for both call sites:
2010 *
2011 * * The remote callers will only take care of the global timers
2012 * as local timers will be handled by CPU itself. When not
2013 * updating tevt->global with the already missed first global
2014 * timer, it is possible that it will be missed completely.
2015 *
2016 * * The local callers will ignore the tevt->global anyway, when
2017 * nextevt is max. one tick away.
2018 */
2019 if (!local_first)
2020 tevt->global = tevt->local;
2021 return nextevt;
2022 }
2023
2024 /*
2025 * Update tevt.* values:
2026 *
2027 * If the local queue expires first, then the global event can be
2028 * ignored. If the global queue is empty, nothing to do either.
2029 */
2030 if (!local_first && base_global->timers_pending)
2031 tevt->global = basem + (u64)(nextevt_global - basej) * TICK_NSEC;
2032
2033 if (base_local->timers_pending)
2034 tevt->local = basem + (u64)(nextevt_local - basej) * TICK_NSEC;
2035
2036 return nextevt;
2037}
2038
2039# ifdef CONFIG_SMP
2040/**
2041 * fetch_next_timer_interrupt_remote() - Store next timers into @tevt
2042 * @basej: base time jiffies
2043 * @basem: base time clock monotonic
2044 * @tevt: Pointer to the storage for the expiry values
2045 * @cpu: Remote CPU
2046 *
2047 * Stores the next pending local and global timer expiry values in the
2048 * struct pointed to by @tevt. If a queue is empty the corresponding
2049 * field is set to KTIME_MAX. If local event expires before global
2050 * event, global event is set to KTIME_MAX as well.
2051 *
2052 * Caller needs to make sure timer base locks are held (use
2053 * timer_lock_remote_bases() for this purpose).
2054 */
2055void fetch_next_timer_interrupt_remote(unsigned long basej, u64 basem,
2056 struct timer_events *tevt,
2057 unsigned int cpu)
2058{
2059 struct timer_base *base_local, *base_global;
2060
2061 /* Preset local / global events */
2062 tevt->local = tevt->global = KTIME_MAX;
2063
2064 base_local = per_cpu_ptr(&timer_bases[BASE_LOCAL], cpu);
2065 base_global = per_cpu_ptr(&timer_bases[BASE_GLOBAL], cpu);
2066
2067 lockdep_assert_held(&base_local->lock);
2068 lockdep_assert_held(&base_global->lock);
2069
2070 fetch_next_timer_interrupt(basej, basem, base_local, base_global, tevt);
2071}
2072
2073/**
2074 * timer_unlock_remote_bases - unlock timer bases of cpu
2075 * @cpu: Remote CPU
2076 *
2077 * Unlocks the remote timer bases.
2078 */
2079void timer_unlock_remote_bases(unsigned int cpu)
2080 __releases(timer_bases[BASE_LOCAL]->lock)
2081 __releases(timer_bases[BASE_GLOBAL]->lock)
2082{
2083 struct timer_base *base_local, *base_global;
2084
2085 base_local = per_cpu_ptr(&timer_bases[BASE_LOCAL], cpu);
2086 base_global = per_cpu_ptr(&timer_bases[BASE_GLOBAL], cpu);
2087
2088 raw_spin_unlock(&base_global->lock);
2089 raw_spin_unlock(&base_local->lock);
2090}
2091
2092/**
2093 * timer_lock_remote_bases - lock timer bases of cpu
2094 * @cpu: Remote CPU
2095 *
2096 * Locks the remote timer bases.
2097 */
2098void timer_lock_remote_bases(unsigned int cpu)
2099 __acquires(timer_bases[BASE_LOCAL]->lock)
2100 __acquires(timer_bases[BASE_GLOBAL]->lock)
2101{
2102 struct timer_base *base_local, *base_global;
2103
2104 base_local = per_cpu_ptr(&timer_bases[BASE_LOCAL], cpu);
2105 base_global = per_cpu_ptr(&timer_bases[BASE_GLOBAL], cpu);
2106
2107 lockdep_assert_irqs_disabled();
2108
2109 raw_spin_lock(&base_local->lock);
2110 raw_spin_lock_nested(&base_global->lock, SINGLE_DEPTH_NESTING);
2111}
2112
2113/**
2114 * timer_base_is_idle() - Return whether timer base is set idle
2115 *
2116 * Returns value of local timer base is_idle value.
2117 */
2118bool timer_base_is_idle(void)
2119{
2120 return __this_cpu_read(timer_bases[BASE_LOCAL].is_idle);
2121}
2122
2123static void __run_timer_base(struct timer_base *base);
2124
2125/**
2126 * timer_expire_remote() - expire global timers of cpu
2127 * @cpu: Remote CPU
2128 *
2129 * Expire timers of global base of remote CPU.
2130 */
2131void timer_expire_remote(unsigned int cpu)
2132{
2133 struct timer_base *base = per_cpu_ptr(&timer_bases[BASE_GLOBAL], cpu);
2134
2135 __run_timer_base(base);
2136}
2137
2138static void timer_use_tmigr(unsigned long basej, u64 basem,
2139 unsigned long *nextevt, bool *tick_stop_path,
2140 bool timer_base_idle, struct timer_events *tevt)
2141{
2142 u64 next_tmigr;
2143
2144 if (timer_base_idle)
2145 next_tmigr = tmigr_cpu_new_timer(nextevt: tevt->global);
2146 else if (tick_stop_path)
2147 next_tmigr = tmigr_cpu_deactivate(nextevt: tevt->global);
2148 else
2149 next_tmigr = tmigr_quick_check(nextevt: tevt->global);
2150
2151 /*
2152 * If the CPU is the last going idle in timer migration hierarchy, make
2153 * sure the CPU will wake up in time to handle remote timers.
2154 * next_tmigr == KTIME_MAX if other CPUs are still active.
2155 */
2156 if (next_tmigr < tevt->local) {
2157 u64 tmp;
2158
2159 /* If we missed a tick already, force 0 delta */
2160 if (next_tmigr < basem)
2161 next_tmigr = basem;
2162
2163 tmp = div_u64(dividend: next_tmigr - basem, TICK_NSEC);
2164
2165 *nextevt = basej + (unsigned long)tmp;
2166 tevt->local = next_tmigr;
2167 }
2168}
2169# else
2170static void timer_use_tmigr(unsigned long basej, u64 basem,
2171 unsigned long *nextevt, bool *tick_stop_path,
2172 bool timer_base_idle, struct timer_events *tevt)
2173{
2174 /*
2175 * Make sure first event is written into tevt->local to not miss a
2176 * timer on !SMP systems.
2177 */
2178 tevt->local = min_t(u64, tevt->local, tevt->global);
2179}
2180# endif /* CONFIG_SMP */
2181
2182static inline u64 __get_next_timer_interrupt(unsigned long basej, u64 basem,
2183 bool *idle)
2184{
2185 struct timer_events tevt = { .local = KTIME_MAX, .global = KTIME_MAX };
2186 struct timer_base *base_local, *base_global;
2187 unsigned long nextevt;
2188 bool idle_is_possible;
2189
2190 /*
2191 * When the CPU is offline, the tick is cancelled and nothing is supposed
2192 * to try to stop it.
2193 */
2194 if (WARN_ON_ONCE(cpu_is_offline(smp_processor_id()))) {
2195 if (idle)
2196 *idle = true;
2197 return tevt.local;
2198 }
2199
2200 base_local = this_cpu_ptr(&timer_bases[BASE_LOCAL]);
2201 base_global = this_cpu_ptr(&timer_bases[BASE_GLOBAL]);
2202
2203 raw_spin_lock(&base_local->lock);
2204 raw_spin_lock_nested(&base_global->lock, SINGLE_DEPTH_NESTING);
2205
2206 nextevt = fetch_next_timer_interrupt(basej, basem, base_local,
2207 base_global, tevt: &tevt);
2208
2209 /*
2210 * If the next event is only one jiffy ahead there is no need to call
2211 * timer migration hierarchy related functions. The value for the next
2212 * global timer in @tevt struct equals then KTIME_MAX. This is also
2213 * true, when the timer base is idle.
2214 *
2215 * The proper timer migration hierarchy function depends on the callsite
2216 * and whether timer base is idle or not. @nextevt will be updated when
2217 * this CPU needs to handle the first timer migration hierarchy
2218 * event. See timer_use_tmigr() for detailed information.
2219 */
2220 idle_is_possible = time_after(nextevt, basej + 1);
2221 if (idle_is_possible)
2222 timer_use_tmigr(basej, basem, nextevt: &nextevt, tick_stop_path: idle,
2223 timer_base_idle: base_local->is_idle, tevt: &tevt);
2224
2225 /*
2226 * We have a fresh next event. Check whether we can forward the
2227 * base.
2228 */
2229 __forward_timer_base(base: base_local, basej);
2230 __forward_timer_base(base: base_global, basej);
2231
2232 /*
2233 * Set base->is_idle only when caller is timer_base_try_to_set_idle()
2234 */
2235 if (idle) {
2236 /*
2237 * Bases are idle if the next event is more than a tick
2238 * away. Caution: @nextevt could have changed by enqueueing a
2239 * global timer into timer migration hierarchy. Therefore a new
2240 * check is required here.
2241 *
2242 * If the base is marked idle then any timer add operation must
2243 * forward the base clk itself to keep granularity small. This
2244 * idle logic is only maintained for the BASE_LOCAL and
2245 * BASE_GLOBAL base, deferrable timers may still see large
2246 * granularity skew (by design).
2247 */
2248 if (!base_local->is_idle && time_after(nextevt, basej + 1)) {
2249 base_local->is_idle = true;
2250 /*
2251 * Global timers queued locally while running in a task
2252 * in nohz_full mode need a self-IPI to kick reprogramming
2253 * in IRQ tail.
2254 */
2255 if (tick_nohz_full_cpu(cpu: base_local->cpu))
2256 base_global->is_idle = true;
2257 trace_timer_base_idle(is_idle: true, cpu: base_local->cpu);
2258 }
2259 *idle = base_local->is_idle;
2260
2261 /*
2262 * When timer base is not set idle, undo the effect of
2263 * tmigr_cpu_deactivate() to prevent inconsistent states - active
2264 * timer base but inactive timer migration hierarchy.
2265 *
2266 * When timer base was already marked idle, nothing will be
2267 * changed here.
2268 */
2269 if (!base_local->is_idle && idle_is_possible)
2270 tmigr_cpu_activate();
2271 }
2272
2273 raw_spin_unlock(&base_global->lock);
2274 raw_spin_unlock(&base_local->lock);
2275
2276 return cmp_next_hrtimer_event(basem, expires: tevt.local);
2277}
2278
2279/**
2280 * get_next_timer_interrupt() - return the time (clock mono) of the next timer
2281 * @basej: base time jiffies
2282 * @basem: base time clock monotonic
2283 *
2284 * Returns the tick aligned clock monotonic time of the next pending timer or
2285 * KTIME_MAX if no timer is pending. If timer of global base was queued into
2286 * timer migration hierarchy, first global timer is not taken into account. If
2287 * it was the last CPU of timer migration hierarchy going idle, first global
2288 * event is taken into account.
2289 */
2290u64 get_next_timer_interrupt(unsigned long basej, u64 basem)
2291{
2292 return __get_next_timer_interrupt(basej, basem, NULL);
2293}
2294
2295/**
2296 * timer_base_try_to_set_idle() - Try to set the idle state of the timer bases
2297 * @basej: base time jiffies
2298 * @basem: base time clock monotonic
2299 * @idle: pointer to store the value of timer_base->is_idle on return;
2300 * *idle contains the information whether tick was already stopped
2301 *
2302 * Returns the tick aligned clock monotonic time of the next pending timer or
2303 * KTIME_MAX if no timer is pending. When tick was already stopped KTIME_MAX is
2304 * returned as well.
2305 */
2306u64 timer_base_try_to_set_idle(unsigned long basej, u64 basem, bool *idle)
2307{
2308 if (*idle)
2309 return KTIME_MAX;
2310
2311 return __get_next_timer_interrupt(basej, basem, idle);
2312}
2313
2314/**
2315 * timer_clear_idle - Clear the idle state of the timer base
2316 *
2317 * Called with interrupts disabled
2318 */
2319void timer_clear_idle(void)
2320{
2321 /*
2322 * We do this unlocked. The worst outcome is a remote pinned timer
2323 * enqueue sending a pointless IPI, but taking the lock would just
2324 * make the window for sending the IPI a few instructions smaller
2325 * for the cost of taking the lock in the exit from idle
2326 * path. Required for BASE_LOCAL only.
2327 */
2328 __this_cpu_write(timer_bases[BASE_LOCAL].is_idle, false);
2329 if (tick_nohz_full_cpu(smp_processor_id()))
2330 __this_cpu_write(timer_bases[BASE_GLOBAL].is_idle, false);
2331 trace_timer_base_idle(is_idle: false, smp_processor_id());
2332
2333 /* Activate without holding the timer_base->lock */
2334 tmigr_cpu_activate();
2335}
2336#endif
2337
2338/**
2339 * __run_timers - run all expired timers (if any) on this CPU.
2340 * @base: the timer vector to be processed.
2341 */
2342static inline void __run_timers(struct timer_base *base)
2343{
2344 struct hlist_head heads[LVL_DEPTH];
2345 int levels;
2346
2347 lockdep_assert_held(&base->lock);
2348
2349 if (base->running_timer)
2350 return;
2351
2352 while (time_after_eq(jiffies, base->clk) &&
2353 time_after_eq(jiffies, base->next_expiry)) {
2354 levels = collect_expired_timers(base, heads);
2355 /*
2356 * The two possible reasons for not finding any expired
2357 * timer at this clk are that all matching timers have been
2358 * dequeued or no timer has been queued since
2359 * base::next_expiry was set to base::clk +
2360 * TIMER_NEXT_MAX_DELTA.
2361 */
2362 WARN_ON_ONCE(!levels && !base->next_expiry_recalc
2363 && base->timers_pending);
2364 /*
2365 * While executing timers, base->clk is set 1 offset ahead of
2366 * jiffies to avoid endless requeuing to current jiffies.
2367 */
2368 base->clk++;
2369 timer_recalc_next_expiry(base);
2370
2371 while (levels--)
2372 expire_timers(base, head: heads + levels);
2373 }
2374}
2375
2376static void __run_timer_base(struct timer_base *base)
2377{
2378 /* Can race against a remote CPU updating next_expiry under the lock */
2379 if (time_before(jiffies, READ_ONCE(base->next_expiry)))
2380 return;
2381
2382 timer_base_lock_expiry(base);
2383 raw_spin_lock_irq(&base->lock);
2384 __run_timers(base);
2385 raw_spin_unlock_irq(&base->lock);
2386 timer_base_unlock_expiry(base);
2387}
2388
2389static void run_timer_base(int index)
2390{
2391 struct timer_base *base = this_cpu_ptr(&timer_bases[index]);
2392
2393 __run_timer_base(base);
2394}
2395
2396/*
2397 * This function runs timers and the timer-tq in bottom half context.
2398 */
2399static __latent_entropy void run_timer_softirq(void)
2400{
2401 run_timer_base(BASE_LOCAL);
2402 if (IS_ENABLED(CONFIG_NO_HZ_COMMON)) {
2403 run_timer_base(BASE_GLOBAL);
2404 run_timer_base(BASE_DEF);
2405
2406 if (is_timers_nohz_active())
2407 tmigr_handle_remote();
2408 }
2409}
2410
2411/*
2412 * Called by the local, per-CPU timer interrupt on SMP.
2413 */
2414static void run_local_timers(void)
2415{
2416 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_LOCAL]);
2417
2418 hrtimer_run_queues();
2419
2420 for (int i = 0; i < NR_BASES; i++, base++) {
2421 /*
2422 * Raise the softirq only if required.
2423 *
2424 * timer_base::next_expiry can be written by a remote CPU while
2425 * holding the lock. If this write happens at the same time than
2426 * the lockless local read, sanity checker could complain about
2427 * data corruption.
2428 *
2429 * There are two possible situations where
2430 * timer_base::next_expiry is written by a remote CPU:
2431 *
2432 * 1. Remote CPU expires global timers of this CPU and updates
2433 * timer_base::next_expiry of BASE_GLOBAL afterwards in
2434 * next_timer_interrupt() or timer_recalc_next_expiry(). The
2435 * worst outcome is a superfluous raise of the timer softirq
2436 * when the not yet updated value is read.
2437 *
2438 * 2. A new first pinned timer is enqueued by a remote CPU
2439 * and therefore timer_base::next_expiry of BASE_LOCAL is
2440 * updated. When this update is missed, this isn't a
2441 * problem, as an IPI is executed nevertheless when the CPU
2442 * was idle before. When the CPU wasn't idle but the update
2443 * is missed, then the timer would expire one jiffy late -
2444 * bad luck.
2445 *
2446 * Those unlikely corner cases where the worst outcome is only a
2447 * one jiffy delay or a superfluous raise of the softirq are
2448 * not that expensive as doing the check always while holding
2449 * the lock.
2450 *
2451 * Possible remote writers are using WRITE_ONCE(). Local reader
2452 * uses therefore READ_ONCE().
2453 */
2454 if (time_after_eq(jiffies, READ_ONCE(base->next_expiry)) ||
2455 (i == BASE_DEF && tmigr_requires_handle_remote())) {
2456 raise_timer_softirq(nr: TIMER_SOFTIRQ);
2457 return;
2458 }
2459 }
2460}
2461
2462/*
2463 * Called from the timer interrupt handler to charge one tick to the current
2464 * process. user_tick is 1 if the tick is user time, 0 for system.
2465 */
2466void update_process_times(int user_tick)
2467{
2468 struct task_struct *p = current;
2469
2470 /* Note: this timer irq context must be accounted for as well. */
2471 account_process_tick(p, user: user_tick);
2472 run_local_timers();
2473 rcu_sched_clock_irq(user: user_tick);
2474#ifdef CONFIG_IRQ_WORK
2475 if (in_irq())
2476 irq_work_tick();
2477#endif
2478 sched_tick();
2479 if (IS_ENABLED(CONFIG_POSIX_TIMERS))
2480 run_posix_cpu_timers();
2481}
2482
2483#ifdef CONFIG_HOTPLUG_CPU
2484static void migrate_timer_list(struct timer_base *new_base, struct hlist_head *head)
2485{
2486 struct timer_list *timer;
2487 int cpu = new_base->cpu;
2488
2489 while (!hlist_empty(h: head)) {
2490 timer = hlist_entry(head->first, struct timer_list, entry);
2491 detach_timer(timer, clear_pending: false);
2492 timer->flags = (timer->flags & ~TIMER_BASEMASK) | cpu;
2493 internal_add_timer(base: new_base, timer);
2494 }
2495}
2496
2497int timers_prepare_cpu(unsigned int cpu)
2498{
2499 struct timer_base *base;
2500 int b;
2501
2502 for (b = 0; b < NR_BASES; b++) {
2503 base = per_cpu_ptr(&timer_bases[b], cpu);
2504 base->clk = jiffies;
2505 base->next_expiry = base->clk + TIMER_NEXT_MAX_DELTA;
2506 base->next_expiry_recalc = false;
2507 base->timers_pending = false;
2508 base->is_idle = false;
2509 }
2510 return 0;
2511}
2512
2513int timers_dead_cpu(unsigned int cpu)
2514{
2515 struct timer_base *old_base;
2516 struct timer_base *new_base;
2517 int b, i;
2518
2519 for (b = 0; b < NR_BASES; b++) {
2520 old_base = per_cpu_ptr(&timer_bases[b], cpu);
2521 new_base = get_cpu_ptr(&timer_bases[b]);
2522 /*
2523 * The caller is globally serialized and nobody else
2524 * takes two locks at once, deadlock is not possible.
2525 */
2526 raw_spin_lock_irq(&new_base->lock);
2527 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
2528
2529 /*
2530 * The current CPUs base clock might be stale. Update it
2531 * before moving the timers over.
2532 */
2533 forward_timer_base(base: new_base);
2534
2535 WARN_ON_ONCE(old_base->running_timer);
2536 old_base->running_timer = NULL;
2537
2538 for (i = 0; i < WHEEL_SIZE; i++)
2539 migrate_timer_list(new_base, head: old_base->vectors + i);
2540
2541 raw_spin_unlock(&old_base->lock);
2542 raw_spin_unlock_irq(&new_base->lock);
2543 put_cpu_ptr(&timer_bases);
2544 }
2545 return 0;
2546}
2547
2548#endif /* CONFIG_HOTPLUG_CPU */
2549
2550static void __init init_timer_cpu(int cpu)
2551{
2552 struct timer_base *base;
2553 int i;
2554
2555 for (i = 0; i < NR_BASES; i++) {
2556 base = per_cpu_ptr(&timer_bases[i], cpu);
2557 base->cpu = cpu;
2558 raw_spin_lock_init(&base->lock);
2559 base->clk = jiffies;
2560 base->next_expiry = base->clk + TIMER_NEXT_MAX_DELTA;
2561 timer_base_init_expiry_lock(base);
2562 }
2563}
2564
2565static void __init init_timer_cpus(void)
2566{
2567 int cpu;
2568
2569 for_each_possible_cpu(cpu)
2570 init_timer_cpu(cpu);
2571}
2572
2573void __init timers_init(void)
2574{
2575 init_timer_cpus();
2576 posix_cputimers_init_work();
2577 open_softirq(nr: TIMER_SOFTIRQ, action: run_timer_softirq);
2578}
2579