projects/linux·blob v5.1 · refs

1998 lines · 57.0 KiB

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
47#include <linux/uaccess.h>
48#include <asm/unistd.h>
49#include <asm/div64.h>
50#include <asm/timex.h>
51#include <asm/io.h>
52
53#include "tick-internal.h"
54
55#define CREATE_TRACE_POINTS
56#include <trace/events/timer.h>
57
58__visible u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
59
60EXPORT_SYMBOL(jiffies_64);
61
62/*
63 * The timer wheel has LVL_DEPTH array levels. Each level provides an array of
64 * LVL_SIZE buckets. Each level is driven by its own clock and therefor each
65 * level has a different granularity.
66 *
67 * The level granularity is: LVL_CLK_DIV ^ lvl
68 * The level clock frequency is: HZ / (LVL_CLK_DIV ^ level)
69 *
70 * The array level of a newly armed timer depends on the relative expiry
71 * time. The farther the expiry time is away the higher the array level and
72 * therefor the granularity becomes.
73 *
74 * Contrary to the original timer wheel implementation, which aims for 'exact'
75 * expiry of the timers, this implementation removes the need for recascading
76 * the timers into the lower array levels. The previous 'classic' timer wheel
77 * implementation of the kernel already violated the 'exact' expiry by adding
78 * slack to the expiry time to provide batched expiration. The granularity
79 * levels provide implicit batching.
80 *
81 * This is an optimization of the original timer wheel implementation for the
82 * majority of the timer wheel use cases: timeouts. The vast majority of
83 * timeout timers (networking, disk I/O ...) are canceled before expiry. If
84 * the timeout expires it indicates that normal operation is disturbed, so it
85 * does not matter much whether the timeout comes with a slight delay.
86 *
87 * The only exception to this are networking timers with a small expiry
88 * time. They rely on the granularity. Those fit into the first wheel level,
89 * which has HZ granularity.
90 *
91 * We don't have cascading anymore. timers with a expiry time above the
92 * capacity of the last wheel level are force expired at the maximum timeout
93 * value of the last wheel level. From data sampling we know that the maximum
94 * value observed is 5 days (network connection tracking), so this should not
95 * be an issue.
96 *
97 * The currently chosen array constants values are a good compromise between
98 * array size and granularity.
99 *
100 * This results in the following granularity and range levels:
101 *
102 * HZ 1000 steps
103 * Level Offset Granularity Range
104 * 0 0 1 ms 0 ms - 63 ms
105 * 1 64 8 ms 64 ms - 511 ms
106 * 2 128 64 ms 512 ms - 4095 ms (512ms - ~4s)
107 * 3 192 512 ms 4096 ms - 32767 ms (~4s - ~32s)
108 * 4 256 4096 ms (~4s) 32768 ms - 262143 ms (~32s - ~4m)
109 * 5 320 32768 ms (~32s) 262144 ms - 2097151 ms (~4m - ~34m)
110 * 6 384 262144 ms (~4m) 2097152 ms - 16777215 ms (~34m - ~4h)
111 * 7 448 2097152 ms (~34m) 16777216 ms - 134217727 ms (~4h - ~1d)
112 * 8 512 16777216 ms (~4h) 134217728 ms - 1073741822 ms (~1d - ~12d)
113 *
114 * HZ 300
115 * Level Offset Granularity Range
116 * 0 0 3 ms 0 ms - 210 ms
117 * 1 64 26 ms 213 ms - 1703 ms (213ms - ~1s)
118 * 2 128 213 ms 1706 ms - 13650 ms (~1s - ~13s)
119 * 3 192 1706 ms (~1s) 13653 ms - 109223 ms (~13s - ~1m)
120 * 4 256 13653 ms (~13s) 109226 ms - 873810 ms (~1m - ~14m)
121 * 5 320 109226 ms (~1m) 873813 ms - 6990503 ms (~14m - ~1h)
122 * 6 384 873813 ms (~14m) 6990506 ms - 55924050 ms (~1h - ~15h)
123 * 7 448 6990506 ms (~1h) 55924053 ms - 447392423 ms (~15h - ~5d)
124 * 8 512 55924053 ms (~15h) 447392426 ms - 3579139406 ms (~5d - ~41d)
125 *
126 * HZ 250
127 * Level Offset Granularity Range
128 * 0 0 4 ms 0 ms - 255 ms
129 * 1 64 32 ms 256 ms - 2047 ms (256ms - ~2s)
130 * 2 128 256 ms 2048 ms - 16383 ms (~2s - ~16s)
131 * 3 192 2048 ms (~2s) 16384 ms - 131071 ms (~16s - ~2m)
132 * 4 256 16384 ms (~16s) 131072 ms - 1048575 ms (~2m - ~17m)
133 * 5 320 131072 ms (~2m) 1048576 ms - 8388607 ms (~17m - ~2h)
134 * 6 384 1048576 ms (~17m) 8388608 ms - 67108863 ms (~2h - ~18h)
135 * 7 448 8388608 ms (~2h) 67108864 ms - 536870911 ms (~18h - ~6d)
136 * 8 512 67108864 ms (~18h) 536870912 ms - 4294967288 ms (~6d - ~49d)
137 *
138 * HZ 100
139 * Level Offset Granularity Range
140 * 0 0 10 ms 0 ms - 630 ms
141 * 1 64 80 ms 640 ms - 5110 ms (640ms - ~5s)
142 * 2 128 640 ms 5120 ms - 40950 ms (~5s - ~40s)
143 * 3 192 5120 ms (~5s) 40960 ms - 327670 ms (~40s - ~5m)
144 * 4 256 40960 ms (~40s) 327680 ms - 2621430 ms (~5m - ~43m)
145 * 5 320 327680 ms (~5m) 2621440 ms - 20971510 ms (~43m - ~5h)
146 * 6 384 2621440 ms (~43m) 20971520 ms - 167772150 ms (~5h - ~1d)
147 * 7 448 20971520 ms (~5h) 167772160 ms - 1342177270 ms (~1d - ~15d)
148 */
149
150/* Clock divisor for the next level */
151#define LVL_CLK_SHIFT 3
152#define LVL_CLK_DIV (1UL << LVL_CLK_SHIFT)
153#define LVL_CLK_MASK (LVL_CLK_DIV - 1)
154#define LVL_SHIFT(n) ((n) * LVL_CLK_SHIFT)
155#define LVL_GRAN(n) (1UL << LVL_SHIFT(n))
156
157/*
158 * The time start value for each level to select the bucket at enqueue
159 * time.
160 */
161#define LVL_START(n) ((LVL_SIZE - 1) << (((n) - 1) * LVL_CLK_SHIFT))
162
163/* Size of each clock level */
164#define LVL_BITS 6
165#define LVL_SIZE (1UL << LVL_BITS)
166#define LVL_MASK (LVL_SIZE - 1)
167#define LVL_OFFS(n) ((n) * LVL_SIZE)
168
169/* Level depth */
170#if HZ > 100
171# define LVL_DEPTH 9
172# else
173# define LVL_DEPTH 8
174#endif
175
176/* The cutoff (max. capacity of the wheel) */
177#define WHEEL_TIMEOUT_CUTOFF (LVL_START(LVL_DEPTH))
178#define WHEEL_TIMEOUT_MAX (WHEEL_TIMEOUT_CUTOFF - LVL_GRAN(LVL_DEPTH - 1))
179
180/*
181 * The resulting wheel size. If NOHZ is configured we allocate two
182 * wheels so we have a separate storage for the deferrable timers.
183 */
184#define WHEEL_SIZE (LVL_SIZE * LVL_DEPTH)
185
186#ifdef CONFIG_NO_HZ_COMMON
187# define NR_BASES 2
188# define BASE_STD 0
189# define BASE_DEF 1
190#else
191# define NR_BASES 1
192# define BASE_STD 0
193# define BASE_DEF 0
194#endif
195
196struct timer_base {
197 raw_spinlock_t lock;
198 struct timer_list *running_timer;
199 unsigned long clk;
200 unsigned long next_expiry;
201 unsigned int cpu;
202 bool is_idle;
203 bool must_forward_clk;
204 DECLARE_BITMAP(pending_map, WHEEL_SIZE);
205 struct hlist_head vectors[WHEEL_SIZE];
206} ____cacheline_aligned;
207
208static DEFINE_PER_CPU(struct timer_base, timer_bases[NR_BASES]);
209
210#ifdef CONFIG_NO_HZ_COMMON
211
212static DEFINE_STATIC_KEY_FALSE(timers_nohz_active);
213static DEFINE_MUTEX(timer_keys_mutex);
214
215static void timer_update_keys(struct work_struct *work);
216static DECLARE_WORK(timer_update_work, timer_update_keys);
217
218#ifdef CONFIG_SMP
219unsigned int sysctl_timer_migration = 1;
220
221DEFINE_STATIC_KEY_FALSE(timers_migration_enabled);
222
223static void timers_update_migration(void)
224{
225 if (sysctl_timer_migration && tick_nohz_active)
226 static_branch_enable(&timers_migration_enabled);
227 else
228 static_branch_disable(&timers_migration_enabled);
229}
230#else
231static inline void timers_update_migration(void) { }
232#endif /* !CONFIG_SMP */
233
234static void timer_update_keys(struct work_struct *work)
235{
236 mutex_lock(&timer_keys_mutex);
237 timers_update_migration();
238 static_branch_enable(&timers_nohz_active);
239 mutex_unlock(&timer_keys_mutex);
240}
241
242void timers_update_nohz(void)
243{
244 schedule_work(&timer_update_work);
245}
246
247int timer_migration_handler(struct ctl_table *table, int write,
248 void __user *buffer, size_t *lenp,
249 loff_t *ppos)
250{
251 int ret;
252
253 mutex_lock(&timer_keys_mutex);
254 ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
255 if (!ret && write)
256 timers_update_migration();
257 mutex_unlock(&timer_keys_mutex);
258 return ret;
259}
260
261static inline bool is_timers_nohz_active(void)
262{
263 return static_branch_unlikely(&timers_nohz_active);
264}
265#else
266static inline bool is_timers_nohz_active(void) { return false; }
267#endif /* NO_HZ_COMMON */
268
269static unsigned long round_jiffies_common(unsigned long j, int cpu,
270 bool force_up)
271{
272 int rem;
273 unsigned long original = j;
274
275 /*
276 * We don't want all cpus firing their timers at once hitting the
277 * same lock or cachelines, so we skew each extra cpu with an extra
278 * 3 jiffies. This 3 jiffies came originally from the mm/ code which
279 * already did this.
280 * The skew is done by adding 3*cpunr, then round, then subtract this
281 * extra offset again.
282 */
283 j += cpu * 3;
284
285 rem = j % HZ;
286
287 /*
288 * If the target jiffie is just after a whole second (which can happen
289 * due to delays of the timer irq, long irq off times etc etc) then
290 * we should round down to the whole second, not up. Use 1/4th second
291 * as cutoff for this rounding as an extreme upper bound for this.
292 * But never round down if @force_up is set.
293 */
294 if (rem < HZ/4 && !force_up) /* round down */
295 j = j - rem;
296 else /* round up */
297 j = j - rem + HZ;
298
299 /* now that we have rounded, subtract the extra skew again */
300 j -= cpu * 3;
301
302 /*
303 * Make sure j is still in the future. Otherwise return the
304 * unmodified value.
305 */
306 return time_is_after_jiffies(j) ? j : original;
307}
308
309/**
310 * __round_jiffies - function to round jiffies to a full second
311 * @j: the time in (absolute) jiffies that should be rounded
312 * @cpu: the processor number on which the timeout will happen
313 *
314 * __round_jiffies() rounds an absolute time in the future (in jiffies)
315 * up or down to (approximately) full seconds. This is useful for timers
316 * for which the exact time they fire does not matter too much, as long as
317 * they fire approximately every X seconds.
318 *
319 * By rounding these timers to whole seconds, all such timers will fire
320 * at the same time, rather than at various times spread out. The goal
321 * of this is to have the CPU wake up less, which saves power.
322 *
323 * The exact rounding is skewed for each processor to avoid all
324 * processors firing at the exact same time, which could lead
325 * to lock contention or spurious cache line bouncing.
326 *
327 * The return value is the rounded version of the @j parameter.
328 */
329unsigned long __round_jiffies(unsigned long j, int cpu)
330{
331 return round_jiffies_common(j, cpu, false);
332}
333EXPORT_SYMBOL_GPL(__round_jiffies);
334
335/**
336 * __round_jiffies_relative - function to round jiffies to a full second
337 * @j: the time in (relative) jiffies that should be rounded
338 * @cpu: the processor number on which the timeout will happen
339 *
340 * __round_jiffies_relative() rounds a time delta in the future (in jiffies)
341 * up or down to (approximately) full seconds. This is useful for timers
342 * for which the exact time they fire does not matter too much, as long as
343 * they fire approximately every X seconds.
344 *
345 * By rounding these timers to whole seconds, all such timers will fire
346 * at the same time, rather than at various times spread out. The goal
347 * of this is to have the CPU wake up less, which saves power.
348 *
349 * The exact rounding is skewed for each processor to avoid all
350 * processors firing at the exact same time, which could lead
351 * to lock contention or spurious cache line bouncing.
352 *
353 * The return value is the rounded version of the @j parameter.
354 */
355unsigned long __round_jiffies_relative(unsigned long j, int cpu)
356{
357 unsigned long j0 = jiffies;
358
359 /* Use j0 because jiffies might change while we run */
360 return round_jiffies_common(j + j0, cpu, false) - j0;
361}
362EXPORT_SYMBOL_GPL(__round_jiffies_relative);
363
364/**
365 * round_jiffies - function to round jiffies to a full second
366 * @j: the time in (absolute) jiffies that should be rounded
367 *
368 * round_jiffies() rounds an absolute time in the future (in jiffies)
369 * up or down to (approximately) full seconds. This is useful for timers
370 * for which the exact time they fire does not matter too much, as long as
371 * they fire approximately every X seconds.
372 *
373 * By rounding these timers to whole seconds, all such timers will fire
374 * at the same time, rather than at various times spread out. The goal
375 * of this is to have the CPU wake up less, which saves power.
376 *
377 * The return value is the rounded version of the @j parameter.
378 */
379unsigned long round_jiffies(unsigned long j)
380{
381 return round_jiffies_common(j, raw_smp_processor_id(), false);
382}
383EXPORT_SYMBOL_GPL(round_jiffies);
384
385/**
386 * round_jiffies_relative - function to round jiffies to a full second
387 * @j: the time in (relative) jiffies that should be rounded
388 *
389 * round_jiffies_relative() rounds a time delta in the future (in jiffies)
390 * up or down to (approximately) full seconds. This is useful for timers
391 * for which the exact time they fire does not matter too much, as long as
392 * they fire approximately every X seconds.
393 *
394 * By rounding these timers to whole seconds, all such timers will fire
395 * at the same time, rather than at various times spread out. The goal
396 * of this is to have the CPU wake up less, which saves power.
397 *
398 * The return value is the rounded version of the @j parameter.
399 */
400unsigned long round_jiffies_relative(unsigned long j)
401{
402 return __round_jiffies_relative(j, raw_smp_processor_id());
403}
404EXPORT_SYMBOL_GPL(round_jiffies_relative);
405
406/**
407 * __round_jiffies_up - function to round jiffies up to a full second
408 * @j: the time in (absolute) jiffies that should be rounded
409 * @cpu: the processor number on which the timeout will happen
410 *
411 * This is the same as __round_jiffies() except that it will never
412 * round down. This is useful for timeouts for which the exact time
413 * of firing does not matter too much, as long as they don't fire too
414 * early.
415 */
416unsigned long __round_jiffies_up(unsigned long j, int cpu)
417{
418 return round_jiffies_common(j, cpu, true);
419}
420EXPORT_SYMBOL_GPL(__round_jiffies_up);
421
422/**
423 * __round_jiffies_up_relative - function to round jiffies up to a full second
424 * @j: the time in (relative) jiffies that should be rounded
425 * @cpu: the processor number on which the timeout will happen
426 *
427 * This is the same as __round_jiffies_relative() except that it will never
428 * round down. This is useful for timeouts for which the exact time
429 * of firing does not matter too much, as long as they don't fire too
430 * early.
431 */
432unsigned long __round_jiffies_up_relative(unsigned long j, int cpu)
433{
434 unsigned long j0 = jiffies;
435
436 /* Use j0 because jiffies might change while we run */
437 return round_jiffies_common(j + j0, cpu, true) - j0;
438}
439EXPORT_SYMBOL_GPL(__round_jiffies_up_relative);
440
441/**
442 * round_jiffies_up - function to round jiffies up to a full second
443 * @j: the time in (absolute) jiffies that should be rounded
444 *
445 * This is the same as round_jiffies() except that it will never
446 * round down. This is useful for timeouts for which the exact time
447 * of firing does not matter too much, as long as they don't fire too
448 * early.
449 */
450unsigned long round_jiffies_up(unsigned long j)
451{
452 return round_jiffies_common(j, raw_smp_processor_id(), true);
453}
454EXPORT_SYMBOL_GPL(round_jiffies_up);
455
456/**
457 * round_jiffies_up_relative - function to round jiffies up to a full second
458 * @j: the time in (relative) jiffies that should be rounded
459 *
460 * This is the same as round_jiffies_relative() except that it will never
461 * round down. This is useful for timeouts for which the exact time
462 * of firing does not matter too much, as long as they don't fire too
463 * early.
464 */
465unsigned long round_jiffies_up_relative(unsigned long j)
466{
467 return __round_jiffies_up_relative(j, raw_smp_processor_id());
468}
469EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
470
471
472static inline unsigned int timer_get_idx(struct timer_list *timer)
473{
474 return (timer->flags & TIMER_ARRAYMASK) >> TIMER_ARRAYSHIFT;
475}
476
477static inline void timer_set_idx(struct timer_list *timer, unsigned int idx)
478{
479 timer->flags = (timer->flags & ~TIMER_ARRAYMASK) |
480 idx << TIMER_ARRAYSHIFT;
481}
482
483/*
484 * Helper function to calculate the array index for a given expiry
485 * time.
486 */
487static inline unsigned calc_index(unsigned expires, unsigned lvl)
488{
489 expires = (expires + LVL_GRAN(lvl)) >> LVL_SHIFT(lvl);
490 return LVL_OFFS(lvl) + (expires & LVL_MASK);
491}
492
493static int calc_wheel_index(unsigned long expires, unsigned long clk)
494{
495 unsigned long delta = expires - clk;
496 unsigned int idx;
497
498 if (delta < LVL_START(1)) {
499 idx = calc_index(expires, 0);
500 } else if (delta < LVL_START(2)) {
501 idx = calc_index(expires, 1);
502 } else if (delta < LVL_START(3)) {
503 idx = calc_index(expires, 2);
504 } else if (delta < LVL_START(4)) {
505 idx = calc_index(expires, 3);
506 } else if (delta < LVL_START(5)) {
507 idx = calc_index(expires, 4);
508 } else if (delta < LVL_START(6)) {
509 idx = calc_index(expires, 5);
510 } else if (delta < LVL_START(7)) {
511 idx = calc_index(expires, 6);
512 } else if (LVL_DEPTH > 8 && delta < LVL_START(8)) {
513 idx = calc_index(expires, 7);
514 } else if ((long) delta < 0) {
515 idx = clk & LVL_MASK;
516 } else {
517 /*
518 * Force expire obscene large timeouts to expire at the
519 * capacity limit of the wheel.
520 */
521 if (expires >= WHEEL_TIMEOUT_CUTOFF)
522 expires = WHEEL_TIMEOUT_MAX;
523
524 idx = calc_index(expires, LVL_DEPTH - 1);
525 }
526 return idx;
527}
528
529/*
530 * Enqueue the timer into the hash bucket, mark it pending in
531 * the bitmap and store the index in the timer flags.
532 */
533static void enqueue_timer(struct timer_base *base, struct timer_list *timer,
534 unsigned int idx)
535{
536 hlist_add_head(&timer->entry, base->vectors + idx);
537 __set_bit(idx, base->pending_map);
538 timer_set_idx(timer, idx);
539}
540
541static void
542__internal_add_timer(struct timer_base *base, struct timer_list *timer)
543{
544 unsigned int idx;
545
546 idx = calc_wheel_index(timer->expires, base->clk);
547 enqueue_timer(base, timer, idx);
548}
549
550static void
551trigger_dyntick_cpu(struct timer_base *base, struct timer_list *timer)
552{
553 if (!is_timers_nohz_active())
554 return;
555
556 /*
557 * TODO: This wants some optimizing similar to the code below, but we
558 * will do that when we switch from push to pull for deferrable timers.
559 */
560 if (timer->flags & TIMER_DEFERRABLE) {
561 if (tick_nohz_full_cpu(base->cpu))
562 wake_up_nohz_cpu(base->cpu);
563 return;
564 }
565
566 /*
567 * We might have to IPI the remote CPU if the base is idle and the
568 * timer is not deferrable. If the other CPU is on the way to idle
569 * then it can't set base->is_idle as we hold the base lock:
570 */
571 if (!base->is_idle)
572 return;
573
574 /* Check whether this is the new first expiring timer: */
575 if (time_after_eq(timer->expires, base->next_expiry))
576 return;
577
578 /*
579 * Set the next expiry time and kick the CPU so it can reevaluate the
580 * wheel:
581 */
582 base->next_expiry = timer->expires;
583 wake_up_nohz_cpu(base->cpu);
584}
585
586static void
587internal_add_timer(struct timer_base *base, struct timer_list *timer)
588{
589 __internal_add_timer(base, timer);
590 trigger_dyntick_cpu(base, timer);
591}
592
593#ifdef CONFIG_DEBUG_OBJECTS_TIMERS
594
595static struct debug_obj_descr timer_debug_descr;
596
597static void *timer_debug_hint(void *addr)
598{
599 return ((struct timer_list *) addr)->function;
600}
601
602static bool timer_is_static_object(void *addr)
603{
604 struct timer_list *timer = addr;
605
606 return (timer->entry.pprev == NULL &&
607 timer->entry.next == TIMER_ENTRY_STATIC);
608}
609
610/*
611 * fixup_init is called when:
612 * - an active object is initialized
613 */
614static bool timer_fixup_init(void *addr, enum debug_obj_state state)
615{
616 struct timer_list *timer = addr;
617
618 switch (state) {
619 case ODEBUG_STATE_ACTIVE:
620 del_timer_sync(timer);
621 debug_object_init(timer, &timer_debug_descr);
622 return true;
623 default:
624 return false;
625 }
626}
627
628/* Stub timer callback for improperly used timers. */
629static void stub_timer(struct timer_list *unused)
630{
631 WARN_ON(1);
632}
633
634/*
635 * fixup_activate is called when:
636 * - an active object is activated
637 * - an unknown non-static object is activated
638 */
639static bool timer_fixup_activate(void *addr, enum debug_obj_state state)
640{
641 struct timer_list *timer = addr;
642
643 switch (state) {
644 case ODEBUG_STATE_NOTAVAILABLE:
645 timer_setup(timer, stub_timer, 0);
646 return true;
647
648 case ODEBUG_STATE_ACTIVE:
649 WARN_ON(1);
650 /* fall through */
651 default:
652 return false;
653 }
654}
655
656/*
657 * fixup_free is called when:
658 * - an active object is freed
659 */
660static bool timer_fixup_free(void *addr, enum debug_obj_state state)
661{
662 struct timer_list *timer = addr;
663
664 switch (state) {
665 case ODEBUG_STATE_ACTIVE:
666 del_timer_sync(timer);
667 debug_object_free(timer, &timer_debug_descr);
668 return true;
669 default:
670 return false;
671 }
672}
673
674/*
675 * fixup_assert_init is called when:
676 * - an untracked/uninit-ed object is found
677 */
678static bool timer_fixup_assert_init(void *addr, enum debug_obj_state state)
679{
680 struct timer_list *timer = addr;
681
682 switch (state) {
683 case ODEBUG_STATE_NOTAVAILABLE:
684 timer_setup(timer, stub_timer, 0);
685 return true;
686 default:
687 return false;
688 }
689}
690
691static struct debug_obj_descr timer_debug_descr = {
692 .name = "timer_list",
693 .debug_hint = timer_debug_hint,
694 .is_static_object = timer_is_static_object,
695 .fixup_init = timer_fixup_init,
696 .fixup_activate = timer_fixup_activate,
697 .fixup_free = timer_fixup_free,
698 .fixup_assert_init = timer_fixup_assert_init,
699};
700
701static inline void debug_timer_init(struct timer_list *timer)
702{
703 debug_object_init(timer, &timer_debug_descr);
704}
705
706static inline void debug_timer_activate(struct timer_list *timer)
707{
708 debug_object_activate(timer, &timer_debug_descr);
709}
710
711static inline void debug_timer_deactivate(struct timer_list *timer)
712{
713 debug_object_deactivate(timer, &timer_debug_descr);
714}
715
716static inline void debug_timer_free(struct timer_list *timer)
717{
718 debug_object_free(timer, &timer_debug_descr);
719}
720
721static inline void debug_timer_assert_init(struct timer_list *timer)
722{
723 debug_object_assert_init(timer, &timer_debug_descr);
724}
725
726static void do_init_timer(struct timer_list *timer,
727 void (*func)(struct timer_list *),
728 unsigned int flags,
729 const char *name, struct lock_class_key *key);
730
731void init_timer_on_stack_key(struct timer_list *timer,
732 void (*func)(struct timer_list *),
733 unsigned int flags,
734 const char *name, struct lock_class_key *key)
735{
736 debug_object_init_on_stack(timer, &timer_debug_descr);
737 do_init_timer(timer, func, flags, name, key);
738}
739EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
740
741void destroy_timer_on_stack(struct timer_list *timer)
742{
743 debug_object_free(timer, &timer_debug_descr);
744}
745EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
746
747#else
748static inline void debug_timer_init(struct timer_list *timer) { }
749static inline void debug_timer_activate(struct timer_list *timer) { }
750static inline void debug_timer_deactivate(struct timer_list *timer) { }
751static inline void debug_timer_assert_init(struct timer_list *timer) { }
752#endif
753
754static inline void debug_init(struct timer_list *timer)
755{
756 debug_timer_init(timer);
757 trace_timer_init(timer);
758}
759
760static inline void
761debug_activate(struct timer_list *timer, unsigned long expires)
762{
763 debug_timer_activate(timer);
764 trace_timer_start(timer, expires, timer->flags);
765}
766
767static inline void debug_deactivate(struct timer_list *timer)
768{
769 debug_timer_deactivate(timer);
770 trace_timer_cancel(timer);
771}
772
773static inline void debug_assert_init(struct timer_list *timer)
774{
775 debug_timer_assert_init(timer);
776}
777
778static void do_init_timer(struct timer_list *timer,
779 void (*func)(struct timer_list *),
780 unsigned int flags,
781 const char *name, struct lock_class_key *key)
782{
783 timer->entry.pprev = NULL;
784 timer->function = func;
785 timer->flags = flags | raw_smp_processor_id();
786 lockdep_init_map(&timer->lockdep_map, name, key, 0);
787}
788
789/**
790 * init_timer_key - initialize a timer
791 * @timer: the timer to be initialized
792 * @func: timer callback function
793 * @flags: timer flags
794 * @name: name of the timer
795 * @key: lockdep class key of the fake lock used for tracking timer
796 * sync lock dependencies
797 *
798 * init_timer_key() must be done to a timer prior calling *any* of the
799 * other timer functions.
800 */
801void init_timer_key(struct timer_list *timer,
802 void (*func)(struct timer_list *), unsigned int flags,
803 const char *name, struct lock_class_key *key)
804{
805 debug_init(timer);
806 do_init_timer(timer, func, flags, name, key);
807}
808EXPORT_SYMBOL(init_timer_key);
809
810static inline void detach_timer(struct timer_list *timer, bool clear_pending)
811{
812 struct hlist_node *entry = &timer->entry;
813
814 debug_deactivate(timer);
815
816 __hlist_del(entry);
817 if (clear_pending)
818 entry->pprev = NULL;
819 entry->next = LIST_POISON2;
820}
821
822static int detach_if_pending(struct timer_list *timer, struct timer_base *base,
823 bool clear_pending)
824{
825 unsigned idx = timer_get_idx(timer);
826
827 if (!timer_pending(timer))
828 return 0;
829
830 if (hlist_is_singular_node(&timer->entry, base->vectors + idx))
831 __clear_bit(idx, base->pending_map);
832
833 detach_timer(timer, clear_pending);
834 return 1;
835}
836
837static inline struct timer_base *get_timer_cpu_base(u32 tflags, u32 cpu)
838{
839 struct timer_base *base = per_cpu_ptr(&timer_bases[BASE_STD], cpu);
840
841 /*
842 * If the timer is deferrable and NO_HZ_COMMON is set then we need
843 * to use the deferrable base.
844 */
845 if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE))
846 base = per_cpu_ptr(&timer_bases[BASE_DEF], cpu);
847 return base;
848}
849
850static inline struct timer_base *get_timer_this_cpu_base(u32 tflags)
851{
852 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
853
854 /*
855 * If the timer is deferrable and NO_HZ_COMMON is set then we need
856 * to use the deferrable base.
857 */
858 if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE))
859 base = this_cpu_ptr(&timer_bases[BASE_DEF]);
860 return base;
861}
862
863static inline struct timer_base *get_timer_base(u32 tflags)
864{
865 return get_timer_cpu_base(tflags, tflags & TIMER_CPUMASK);
866}
867
868static inline struct timer_base *
869get_target_base(struct timer_base *base, unsigned tflags)
870{
871#if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
872 if (static_branch_likely(&timers_migration_enabled) &&
873 !(tflags & TIMER_PINNED))
874 return get_timer_cpu_base(tflags, get_nohz_timer_target());
875#endif
876 return get_timer_this_cpu_base(tflags);
877}
878
879static inline void forward_timer_base(struct timer_base *base)
880{
881#ifdef CONFIG_NO_HZ_COMMON
882 unsigned long jnow;
883
884 /*
885 * We only forward the base when we are idle or have just come out of
886 * idle (must_forward_clk logic), and have a delta between base clock
887 * and jiffies. In the common case, run_timers will take care of it.
888 */
889 if (likely(!base->must_forward_clk))
890 return;
891
892 jnow = READ_ONCE(jiffies);
893 base->must_forward_clk = base->is_idle;
894 if ((long)(jnow - base->clk) < 2)
895 return;
896
897 /*
898 * If the next expiry value is > jiffies, then we fast forward to
899 * jiffies otherwise we forward to the next expiry value.
900 */
901 if (time_after(base->next_expiry, jnow))
902 base->clk = jnow;
903 else
904 base->clk = base->next_expiry;
905#endif
906}
907
908
909/*
910 * We are using hashed locking: Holding per_cpu(timer_bases[x]).lock means
911 * that all timers which are tied to this base are locked, and the base itself
912 * is locked too.
913 *
914 * So __run_timers/migrate_timers can safely modify all timers which could
915 * be found in the base->vectors array.
916 *
917 * When a timer is migrating then the TIMER_MIGRATING flag is set and we need
918 * to wait until the migration is done.
919 */
920static struct timer_base *lock_timer_base(struct timer_list *timer,
921 unsigned long *flags)
922 __acquires(timer->base->lock)
923{
924 for (;;) {
925 struct timer_base *base;
926 u32 tf;
927
928 /*
929 * We need to use READ_ONCE() here, otherwise the compiler
930 * might re-read @tf between the check for TIMER_MIGRATING
931 * and spin_lock().
932 */
933 tf = READ_ONCE(timer->flags);
934
935 if (!(tf & TIMER_MIGRATING)) {
936 base = get_timer_base(tf);
937 raw_spin_lock_irqsave(&base->lock, *flags);
938 if (timer->flags == tf)
939 return base;
940 raw_spin_unlock_irqrestore(&base->lock, *flags);
941 }
942 cpu_relax();
943 }
944}
945
946#define MOD_TIMER_PENDING_ONLY 0x01
947#define MOD_TIMER_REDUCE 0x02
948
949static inline int
950__mod_timer(struct timer_list *timer, unsigned long expires, unsigned int options)
951{
952 struct timer_base *base, *new_base;
953 unsigned int idx = UINT_MAX;
954 unsigned long clk = 0, flags;
955 int ret = 0;
956
957 BUG_ON(!timer->function);
958
959 /*
960 * This is a common optimization triggered by the networking code - if
961 * the timer is re-modified to have the same timeout or ends up in the
962 * same array bucket then just return:
963 */
964 if (timer_pending(timer)) {
965 /*
966 * The downside of this optimization is that it can result in
967 * larger granularity than you would get from adding a new
968 * timer with this expiry.
969 */
970 long diff = timer->expires - expires;
971
972 if (!diff)
973 return 1;
974 if (options & MOD_TIMER_REDUCE && diff <= 0)
975 return 1;
976
977 /*
978 * We lock timer base and calculate the bucket index right
979 * here. If the timer ends up in the same bucket, then we
980 * just update the expiry time and avoid the whole
981 * dequeue/enqueue dance.
982 */
983 base = lock_timer_base(timer, &flags);
984 forward_timer_base(base);
985
986 if (timer_pending(timer) && (options & MOD_TIMER_REDUCE) &&
987 time_before_eq(timer->expires, expires)) {
988 ret = 1;
989 goto out_unlock;
990 }
991
992 clk = base->clk;
993 idx = calc_wheel_index(expires, clk);
994
995 /*
996 * Retrieve and compare the array index of the pending
997 * timer. If it matches set the expiry to the new value so a
998 * subsequent call will exit in the expires check above.
999 */
1000 if (idx == timer_get_idx(timer)) {
1001 if (!(options & MOD_TIMER_REDUCE))
1002 timer->expires = expires;
1003 else if (time_after(timer->expires, expires))
1004 timer->expires = expires;
1005 ret = 1;
1006 goto out_unlock;
1007 }
1008 } else {
1009 base = lock_timer_base(timer, &flags);
1010 forward_timer_base(base);
1011 }
1012
1013 ret = detach_if_pending(timer, base, false);
1014 if (!ret && (options & MOD_TIMER_PENDING_ONLY))
1015 goto out_unlock;
1016
1017 new_base = get_target_base(base, timer->flags);
1018
1019 if (base != new_base) {
1020 /*
1021 * We are trying to schedule the timer on the new base.
1022 * However we can't change timer's base while it is running,
1023 * otherwise del_timer_sync() can't detect that the timer's
1024 * handler yet has not finished. This also guarantees that the
1025 * timer is serialized wrt itself.
1026 */
1027 if (likely(base->running_timer != timer)) {
1028 /* See the comment in lock_timer_base() */
1029 timer->flags |= TIMER_MIGRATING;
1030
1031 raw_spin_unlock(&base->lock);
1032 base = new_base;
1033 raw_spin_lock(&base->lock);
1034 WRITE_ONCE(timer->flags,
1035 (timer->flags & ~TIMER_BASEMASK) | base->cpu);
1036 forward_timer_base(base);
1037 }
1038 }
1039
1040 debug_activate(timer, expires);
1041
1042 timer->expires = expires;
1043 /*
1044 * If 'idx' was calculated above and the base time did not advance
1045 * between calculating 'idx' and possibly switching the base, only
1046 * enqueue_timer() and trigger_dyntick_cpu() is required. Otherwise
1047 * we need to (re)calculate the wheel index via
1048 * internal_add_timer().
1049 */
1050 if (idx != UINT_MAX && clk == base->clk) {
1051 enqueue_timer(base, timer, idx);
1052 trigger_dyntick_cpu(base, timer);
1053 } else {
1054 internal_add_timer(base, timer);
1055 }
1056
1057out_unlock:
1058 raw_spin_unlock_irqrestore(&base->lock, flags);
1059
1060 return ret;
1061}
1062
1063/**
1064 * mod_timer_pending - modify a pending timer's timeout
1065 * @timer: the pending timer to be modified
1066 * @expires: new timeout in jiffies
1067 *
1068 * mod_timer_pending() is the same for pending timers as mod_timer(),
1069 * but will not re-activate and modify already deleted timers.
1070 *
1071 * It is useful for unserialized use of timers.
1072 */
1073int mod_timer_pending(struct timer_list *timer, unsigned long expires)
1074{
1075 return __mod_timer(timer, expires, MOD_TIMER_PENDING_ONLY);
1076}
1077EXPORT_SYMBOL(mod_timer_pending);
1078
1079/**
1080 * mod_timer - modify a timer's timeout
1081 * @timer: the timer to be modified
1082 * @expires: new timeout in jiffies
1083 *
1084 * mod_timer() is a more efficient way to update the expire field of an
1085 * active timer (if the timer is inactive it will be activated)
1086 *
1087 * mod_timer(timer, expires) is equivalent to:
1088 *
1089 * del_timer(timer); timer->expires = expires; add_timer(timer);
1090 *
1091 * Note that if there are multiple unserialized concurrent users of the
1092 * same timer, then mod_timer() is the only safe way to modify the timeout,
1093 * since add_timer() cannot modify an already running timer.
1094 *
1095 * The function returns whether it has modified a pending timer or not.
1096 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
1097 * active timer returns 1.)
1098 */
1099int mod_timer(struct timer_list *timer, unsigned long expires)
1100{
1101 return __mod_timer(timer, expires, 0);
1102}
1103EXPORT_SYMBOL(mod_timer);
1104
1105/**
1106 * timer_reduce - Modify a timer's timeout if it would reduce the timeout
1107 * @timer: The timer to be modified
1108 * @expires: New timeout in jiffies
1109 *
1110 * timer_reduce() is very similar to mod_timer(), except that it will only
1111 * modify a running timer if that would reduce the expiration time (it will
1112 * start a timer that isn't running).
1113 */
1114int timer_reduce(struct timer_list *timer, unsigned long expires)
1115{
1116 return __mod_timer(timer, expires, MOD_TIMER_REDUCE);
1117}
1118EXPORT_SYMBOL(timer_reduce);
1119
1120/**
1121 * add_timer - start a timer
1122 * @timer: the timer to be added
1123 *
1124 * The kernel will do a ->function(@timer) callback from the
1125 * timer interrupt at the ->expires point in the future. The
1126 * current time is 'jiffies'.
1127 *
1128 * The timer's ->expires, ->function fields must be set prior calling this
1129 * function.
1130 *
1131 * Timers with an ->expires field in the past will be executed in the next
1132 * timer tick.
1133 */
1134void add_timer(struct timer_list *timer)
1135{
1136 BUG_ON(timer_pending(timer));
1137 mod_timer(timer, timer->expires);
1138}
1139EXPORT_SYMBOL(add_timer);
1140
1141/**
1142 * add_timer_on - start a timer on a particular CPU
1143 * @timer: the timer to be added
1144 * @cpu: the CPU to start it on
1145 *
1146 * This is not very scalable on SMP. Double adds are not possible.
1147 */
1148void add_timer_on(struct timer_list *timer, int cpu)
1149{
1150 struct timer_base *new_base, *base;
1151 unsigned long flags;
1152
1153 BUG_ON(timer_pending(timer) || !timer->function);
1154
1155 new_base = get_timer_cpu_base(timer->flags, cpu);
1156
1157 /*
1158 * If @timer was on a different CPU, it should be migrated with the
1159 * old base locked to prevent other operations proceeding with the
1160 * wrong base locked. See lock_timer_base().
1161 */
1162 base = lock_timer_base(timer, &flags);
1163 if (base != new_base) {
1164 timer->flags |= TIMER_MIGRATING;
1165
1166 raw_spin_unlock(&base->lock);
1167 base = new_base;
1168 raw_spin_lock(&base->lock);
1169 WRITE_ONCE(timer->flags,
1170 (timer->flags & ~TIMER_BASEMASK) | cpu);
1171 }
1172 forward_timer_base(base);
1173
1174 debug_activate(timer, timer->expires);
1175 internal_add_timer(base, timer);
1176 raw_spin_unlock_irqrestore(&base->lock, flags);
1177}
1178EXPORT_SYMBOL_GPL(add_timer_on);
1179
1180/**
1181 * del_timer - deactivate a timer.
1182 * @timer: the timer to be deactivated
1183 *
1184 * del_timer() deactivates a timer - this works on both active and inactive
1185 * timers.
1186 *
1187 * The function returns whether it has deactivated a pending timer or not.
1188 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
1189 * active timer returns 1.)
1190 */
1191int del_timer(struct timer_list *timer)
1192{
1193 struct timer_base *base;
1194 unsigned long flags;
1195 int ret = 0;
1196
1197 debug_assert_init(timer);
1198
1199 if (timer_pending(timer)) {
1200 base = lock_timer_base(timer, &flags);
1201 ret = detach_if_pending(timer, base, true);
1202 raw_spin_unlock_irqrestore(&base->lock, flags);
1203 }
1204
1205 return ret;
1206}
1207EXPORT_SYMBOL(del_timer);
1208
1209/**
1210 * try_to_del_timer_sync - Try to deactivate a timer
1211 * @timer: timer to delete
1212 *
1213 * This function tries to deactivate a timer. Upon successful (ret >= 0)
1214 * exit the timer is not queued and the handler is not running on any CPU.
1215 */
1216int try_to_del_timer_sync(struct timer_list *timer)
1217{
1218 struct timer_base *base;
1219 unsigned long flags;
1220 int ret = -1;
1221
1222 debug_assert_init(timer);
1223
1224 base = lock_timer_base(timer, &flags);
1225
1226 if (base->running_timer != timer)
1227 ret = detach_if_pending(timer, base, true);
1228
1229 raw_spin_unlock_irqrestore(&base->lock, flags);
1230
1231 return ret;
1232}
1233EXPORT_SYMBOL(try_to_del_timer_sync);
1234
1235#ifdef CONFIG_SMP
1236/**
1237 * del_timer_sync - deactivate a timer and wait for the handler to finish.
1238 * @timer: the timer to be deactivated
1239 *
1240 * This function only differs from del_timer() on SMP: besides deactivating
1241 * the timer it also makes sure the handler has finished executing on other
1242 * CPUs.
1243 *
1244 * Synchronization rules: Callers must prevent restarting of the timer,
1245 * otherwise this function is meaningless. It must not be called from
1246 * interrupt contexts unless the timer is an irqsafe one. The caller must
1247 * not hold locks which would prevent completion of the timer's
1248 * handler. The timer's handler must not call add_timer_on(). Upon exit the
1249 * timer is not queued and the handler is not running on any CPU.
1250 *
1251 * Note: For !irqsafe timers, you must not hold locks that are held in
1252 * interrupt context while calling this function. Even if the lock has
1253 * nothing to do with the timer in question. Here's why::
1254 *
1255 * CPU0 CPU1
1256 * ---- ----
1257 * <SOFTIRQ>
1258 * call_timer_fn();
1259 * base->running_timer = mytimer;
1260 * spin_lock_irq(somelock);
1261 * <IRQ>
1262 * spin_lock(somelock);
1263 * del_timer_sync(mytimer);
1264 * while (base->running_timer == mytimer);
1265 *
1266 * Now del_timer_sync() will never return and never release somelock.
1267 * The interrupt on the other CPU is waiting to grab somelock but
1268 * it has interrupted the softirq that CPU0 is waiting to finish.
1269 *
1270 * The function returns whether it has deactivated a pending timer or not.
1271 */
1272int del_timer_sync(struct timer_list *timer)
1273{
1274#ifdef CONFIG_LOCKDEP
1275 unsigned long flags;
1276
1277 /*
1278 * If lockdep gives a backtrace here, please reference
1279 * the synchronization rules above.
1280 */
1281 local_irq_save(flags);
1282 lock_map_acquire(&timer->lockdep_map);
1283 lock_map_release(&timer->lockdep_map);
1284 local_irq_restore(flags);
1285#endif
1286 /*
1287 * don't use it in hardirq context, because it
1288 * could lead to deadlock.
1289 */
1290 WARN_ON(in_irq() && !(timer->flags & TIMER_IRQSAFE));
1291 for (;;) {
1292 int ret = try_to_del_timer_sync(timer);
1293 if (ret >= 0)
1294 return ret;
1295 cpu_relax();
1296 }
1297}
1298EXPORT_SYMBOL(del_timer_sync);
1299#endif
1300
1301static void call_timer_fn(struct timer_list *timer, void (*fn)(struct timer_list *))
1302{
1303 int count = preempt_count();
1304
1305#ifdef CONFIG_LOCKDEP
1306 /*
1307 * It is permissible to free the timer from inside the
1308 * function that is called from it, this we need to take into
1309 * account for lockdep too. To avoid bogus "held lock freed"
1310 * warnings as well as problems when looking into
1311 * timer->lockdep_map, make a copy and use that here.
1312 */
1313 struct lockdep_map lockdep_map;
1314
1315 lockdep_copy_map(&lockdep_map, &timer->lockdep_map);
1316#endif
1317 /*
1318 * Couple the lock chain with the lock chain at
1319 * del_timer_sync() by acquiring the lock_map around the fn()
1320 * call here and in del_timer_sync().
1321 */
1322 lock_map_acquire(&lockdep_map);
1323
1324 trace_timer_expire_entry(timer);
1325 fn(timer);
1326 trace_timer_expire_exit(timer);
1327
1328 lock_map_release(&lockdep_map);
1329
1330 if (count != preempt_count()) {
1331 WARN_ONCE(1, "timer: %pF preempt leak: %08x -> %08x\n",
1332 fn, count, preempt_count());
1333 /*
1334 * Restore the preempt count. That gives us a decent
1335 * chance to survive and extract information. If the
1336 * callback kept a lock held, bad luck, but not worse
1337 * than the BUG() we had.
1338 */
1339 preempt_count_set(count);
1340 }
1341}
1342
1343static void expire_timers(struct timer_base *base, struct hlist_head *head)
1344{
1345 while (!hlist_empty(head)) {
1346 struct timer_list *timer;
1347 void (*fn)(struct timer_list *);
1348
1349 timer = hlist_entry(head->first, struct timer_list, entry);
1350
1351 base->running_timer = timer;
1352 detach_timer(timer, true);
1353
1354 fn = timer->function;
1355
1356 if (timer->flags & TIMER_IRQSAFE) {
1357 raw_spin_unlock(&base->lock);
1358 call_timer_fn(timer, fn);
1359 raw_spin_lock(&base->lock);
1360 } else {
1361 raw_spin_unlock_irq(&base->lock);
1362 call_timer_fn(timer, fn);
1363 raw_spin_lock_irq(&base->lock);
1364 }
1365 }
1366}
1367
1368static int __collect_expired_timers(struct timer_base *base,
1369 struct hlist_head *heads)
1370{
1371 unsigned long clk = base->clk;
1372 struct hlist_head *vec;
1373 int i, levels = 0;
1374 unsigned int idx;
1375
1376 for (i = 0; i < LVL_DEPTH; i++) {
1377 idx = (clk & LVL_MASK) + i * LVL_SIZE;
1378
1379 if (__test_and_clear_bit(idx, base->pending_map)) {
1380 vec = base->vectors + idx;
1381 hlist_move_list(vec, heads++);
1382 levels++;
1383 }
1384 /* Is it time to look at the next level? */
1385 if (clk & LVL_CLK_MASK)
1386 break;
1387 /* Shift clock for the next level granularity */
1388 clk >>= LVL_CLK_SHIFT;
1389 }
1390 return levels;
1391}
1392
1393#ifdef CONFIG_NO_HZ_COMMON
1394/*
1395 * Find the next pending bucket of a level. Search from level start (@offset)
1396 * + @clk upwards and if nothing there, search from start of the level
1397 * (@offset) up to @offset + clk.
1398 */
1399static int next_pending_bucket(struct timer_base *base, unsigned offset,
1400 unsigned clk)
1401{
1402 unsigned pos, start = offset + clk;
1403 unsigned end = offset + LVL_SIZE;
1404
1405 pos = find_next_bit(base->pending_map, end, start);
1406 if (pos < end)
1407 return pos - start;
1408
1409 pos = find_next_bit(base->pending_map, start, offset);
1410 return pos < start ? pos + LVL_SIZE - start : -1;
1411}
1412
1413/*
1414 * Search the first expiring timer in the various clock levels. Caller must
1415 * hold base->lock.
1416 */
1417static unsigned long __next_timer_interrupt(struct timer_base *base)
1418{
1419 unsigned long clk, next, adj;
1420 unsigned lvl, offset = 0;
1421
1422 next = base->clk + NEXT_TIMER_MAX_DELTA;
1423 clk = base->clk;
1424 for (lvl = 0; lvl < LVL_DEPTH; lvl++, offset += LVL_SIZE) {
1425 int pos = next_pending_bucket(base, offset, clk & LVL_MASK);
1426
1427 if (pos >= 0) {
1428 unsigned long tmp = clk + (unsigned long) pos;
1429
1430 tmp <<= LVL_SHIFT(lvl);
1431 if (time_before(tmp, next))
1432 next = tmp;
1433 }
1434 /*
1435 * Clock for the next level. If the current level clock lower
1436 * bits are zero, we look at the next level as is. If not we
1437 * need to advance it by one because that's going to be the
1438 * next expiring bucket in that level. base->clk is the next
1439 * expiring jiffie. So in case of:
1440 *
1441 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1442 * 0 0 0 0 0 0
1443 *
1444 * we have to look at all levels @index 0. With
1445 *
1446 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1447 * 0 0 0 0 0 2
1448 *
1449 * LVL0 has the next expiring bucket @index 2. The upper
1450 * levels have the next expiring bucket @index 1.
1451 *
1452 * In case that the propagation wraps the next level the same
1453 * rules apply:
1454 *
1455 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1456 * 0 0 0 0 F 2
1457 *
1458 * So after looking at LVL0 we get:
1459 *
1460 * LVL5 LVL4 LVL3 LVL2 LVL1
1461 * 0 0 0 1 0
1462 *
1463 * So no propagation from LVL1 to LVL2 because that happened
1464 * with the add already, but then we need to propagate further
1465 * from LVL2 to LVL3.
1466 *
1467 * So the simple check whether the lower bits of the current
1468 * level are 0 or not is sufficient for all cases.
1469 */
1470 adj = clk & LVL_CLK_MASK ? 1 : 0;
1471 clk >>= LVL_CLK_SHIFT;
1472 clk += adj;
1473 }
1474 return next;
1475}
1476
1477/*
1478 * Check, if the next hrtimer event is before the next timer wheel
1479 * event:
1480 */
1481static u64 cmp_next_hrtimer_event(u64 basem, u64 expires)
1482{
1483 u64 nextevt = hrtimer_get_next_event();
1484
1485 /*
1486 * If high resolution timers are enabled
1487 * hrtimer_get_next_event() returns KTIME_MAX.
1488 */
1489 if (expires <= nextevt)
1490 return expires;
1491
1492 /*
1493 * If the next timer is already expired, return the tick base
1494 * time so the tick is fired immediately.
1495 */
1496 if (nextevt <= basem)
1497 return basem;
1498
1499 /*
1500 * Round up to the next jiffie. High resolution timers are
1501 * off, so the hrtimers are expired in the tick and we need to
1502 * make sure that this tick really expires the timer to avoid
1503 * a ping pong of the nohz stop code.
1504 *
1505 * Use DIV_ROUND_UP_ULL to prevent gcc calling __divdi3
1506 */
1507 return DIV_ROUND_UP_ULL(nextevt, TICK_NSEC) * TICK_NSEC;
1508}
1509
1510/**
1511 * get_next_timer_interrupt - return the time (clock mono) of the next timer
1512 * @basej: base time jiffies
1513 * @basem: base time clock monotonic
1514 *
1515 * Returns the tick aligned clock monotonic time of the next pending
1516 * timer or KTIME_MAX if no timer is pending.
1517 */
1518u64 get_next_timer_interrupt(unsigned long basej, u64 basem)
1519{
1520 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1521 u64 expires = KTIME_MAX;
1522 unsigned long nextevt;
1523 bool is_max_delta;
1524
1525 /*
1526 * Pretend that there is no timer pending if the cpu is offline.
1527 * Possible pending timers will be migrated later to an active cpu.
1528 */
1529 if (cpu_is_offline(smp_processor_id()))
1530 return expires;
1531
1532 raw_spin_lock(&base->lock);
1533 nextevt = __next_timer_interrupt(base);
1534 is_max_delta = (nextevt == base->clk + NEXT_TIMER_MAX_DELTA);
1535 base->next_expiry = nextevt;
1536 /*
1537 * We have a fresh next event. Check whether we can forward the
1538 * base. We can only do that when @basej is past base->clk
1539 * otherwise we might rewind base->clk.
1540 */
1541 if (time_after(basej, base->clk)) {
1542 if (time_after(nextevt, basej))
1543 base->clk = basej;
1544 else if (time_after(nextevt, base->clk))
1545 base->clk = nextevt;
1546 }
1547
1548 if (time_before_eq(nextevt, basej)) {
1549 expires = basem;
1550 base->is_idle = false;
1551 } else {
1552 if (!is_max_delta)
1553 expires = basem + (u64)(nextevt - basej) * TICK_NSEC;
1554 /*
1555 * If we expect to sleep more than a tick, mark the base idle.
1556 * Also the tick is stopped so any added timer must forward
1557 * the base clk itself to keep granularity small. This idle
1558 * logic is only maintained for the BASE_STD base, deferrable
1559 * timers may still see large granularity skew (by design).
1560 */
1561 if ((expires - basem) > TICK_NSEC) {
1562 base->must_forward_clk = true;
1563 base->is_idle = true;
1564 }
1565 }
1566 raw_spin_unlock(&base->lock);
1567
1568 return cmp_next_hrtimer_event(basem, expires);
1569}
1570
1571/**
1572 * timer_clear_idle - Clear the idle state of the timer base
1573 *
1574 * Called with interrupts disabled
1575 */
1576void timer_clear_idle(void)
1577{
1578 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1579
1580 /*
1581 * We do this unlocked. The worst outcome is a remote enqueue sending
1582 * a pointless IPI, but taking the lock would just make the window for
1583 * sending the IPI a few instructions smaller for the cost of taking
1584 * the lock in the exit from idle path.
1585 */
1586 base->is_idle = false;
1587}
1588
1589static int collect_expired_timers(struct timer_base *base,
1590 struct hlist_head *heads)
1591{
1592 /*
1593 * NOHZ optimization. After a long idle sleep we need to forward the
1594 * base to current jiffies. Avoid a loop by searching the bitfield for
1595 * the next expiring timer.
1596 */
1597 if ((long)(jiffies - base->clk) > 2) {
1598 unsigned long next = __next_timer_interrupt(base);
1599
1600 /*
1601 * If the next timer is ahead of time forward to current
1602 * jiffies, otherwise forward to the next expiry time:
1603 */
1604 if (time_after(next, jiffies)) {
1605 /*
1606 * The call site will increment base->clk and then
1607 * terminate the expiry loop immediately.
1608 */
1609 base->clk = jiffies;
1610 return 0;
1611 }
1612 base->clk = next;
1613 }
1614 return __collect_expired_timers(base, heads);
1615}
1616#else
1617static inline int collect_expired_timers(struct timer_base *base,
1618 struct hlist_head *heads)
1619{
1620 return __collect_expired_timers(base, heads);
1621}
1622#endif
1623
1624/*
1625 * Called from the timer interrupt handler to charge one tick to the current
1626 * process. user_tick is 1 if the tick is user time, 0 for system.
1627 */
1628void update_process_times(int user_tick)
1629{
1630 struct task_struct *p = current;
1631
1632 /* Note: this timer irq context must be accounted for as well. */
1633 account_process_tick(p, user_tick);
1634 run_local_timers();
1635 rcu_sched_clock_irq(user_tick);
1636#ifdef CONFIG_IRQ_WORK
1637 if (in_irq())
1638 irq_work_tick();
1639#endif
1640 scheduler_tick();
1641 if (IS_ENABLED(CONFIG_POSIX_TIMERS))
1642 run_posix_cpu_timers(p);
1643}
1644
1645/**
1646 * __run_timers - run all expired timers (if any) on this CPU.
1647 * @base: the timer vector to be processed.
1648 */
1649static inline void __run_timers(struct timer_base *base)
1650{
1651 struct hlist_head heads[LVL_DEPTH];
1652 int levels;
1653
1654 if (!time_after_eq(jiffies, base->clk))
1655 return;
1656
1657 raw_spin_lock_irq(&base->lock);
1658
1659 /*
1660 * timer_base::must_forward_clk must be cleared before running
1661 * timers so that any timer functions that call mod_timer() will
1662 * not try to forward the base. Idle tracking / clock forwarding
1663 * logic is only used with BASE_STD timers.
1664 *
1665 * The must_forward_clk flag is cleared unconditionally also for
1666 * the deferrable base. The deferrable base is not affected by idle
1667 * tracking and never forwarded, so clearing the flag is a NOOP.
1668 *
1669 * The fact that the deferrable base is never forwarded can cause
1670 * large variations in granularity for deferrable timers, but they
1671 * can be deferred for long periods due to idle anyway.
1672 */
1673 base->must_forward_clk = false;
1674
1675 while (time_after_eq(jiffies, base->clk)) {
1676
1677 levels = collect_expired_timers(base, heads);
1678 base->clk++;
1679
1680 while (levels--)
1681 expire_timers(base, heads + levels);
1682 }
1683 base->running_timer = NULL;
1684 raw_spin_unlock_irq(&base->lock);
1685}
1686
1687/*
1688 * This function runs timers and the timer-tq in bottom half context.
1689 */
1690static __latent_entropy void run_timer_softirq(struct softirq_action *h)
1691{
1692 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1693
1694 __run_timers(base);
1695 if (IS_ENABLED(CONFIG_NO_HZ_COMMON))
1696 __run_timers(this_cpu_ptr(&timer_bases[BASE_DEF]));
1697}
1698
1699/*
1700 * Called by the local, per-CPU timer interrupt on SMP.
1701 */
1702void run_local_timers(void)
1703{
1704 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1705
1706 hrtimer_run_queues();
1707 /* Raise the softirq only if required. */
1708 if (time_before(jiffies, base->clk)) {
1709 if (!IS_ENABLED(CONFIG_NO_HZ_COMMON))
1710 return;
1711 /* CPU is awake, so check the deferrable base. */
1712 base++;
1713 if (time_before(jiffies, base->clk))
1714 return;
1715 }
1716 raise_softirq(TIMER_SOFTIRQ);
1717}
1718
1719/*
1720 * Since schedule_timeout()'s timer is defined on the stack, it must store
1721 * the target task on the stack as well.
1722 */
1723struct process_timer {
1724 struct timer_list timer;
1725 struct task_struct *task;
1726};
1727
1728static void process_timeout(struct timer_list *t)
1729{
1730 struct process_timer *timeout = from_timer(timeout, t, timer);
1731
1732 wake_up_process(timeout->task);
1733}
1734
1735/**
1736 * schedule_timeout - sleep until timeout
1737 * @timeout: timeout value in jiffies
1738 *
1739 * Make the current task sleep until @timeout jiffies have
1740 * elapsed. The routine will return immediately unless
1741 * the current task state has been set (see set_current_state()).
1742 *
1743 * You can set the task state as follows -
1744 *
1745 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1746 * pass before the routine returns unless the current task is explicitly
1747 * woken up, (e.g. by wake_up_process())".
1748 *
1749 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1750 * delivered to the current task or the current task is explicitly woken
1751 * up.
1752 *
1753 * The current task state is guaranteed to be TASK_RUNNING when this
1754 * routine returns.
1755 *
1756 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1757 * the CPU away without a bound on the timeout. In this case the return
1758 * value will be %MAX_SCHEDULE_TIMEOUT.
1759 *
1760 * Returns 0 when the timer has expired otherwise the remaining time in
1761 * jiffies will be returned. In all cases the return value is guaranteed
1762 * to be non-negative.
1763 */
1764signed long __sched schedule_timeout(signed long timeout)
1765{
1766 struct process_timer timer;
1767 unsigned long expire;
1768
1769 switch (timeout)
1770 {
1771 case MAX_SCHEDULE_TIMEOUT:
1772 /*
1773 * These two special cases are useful to be comfortable
1774 * in the caller. Nothing more. We could take
1775 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1776 * but I' d like to return a valid offset (>=0) to allow
1777 * the caller to do everything it want with the retval.
1778 */
1779 schedule();
1780 goto out;
1781 default:
1782 /*
1783 * Another bit of PARANOID. Note that the retval will be
1784 * 0 since no piece of kernel is supposed to do a check
1785 * for a negative retval of schedule_timeout() (since it
1786 * should never happens anyway). You just have the printk()
1787 * that will tell you if something is gone wrong and where.
1788 */
1789 if (timeout < 0) {
1790 printk(KERN_ERR "schedule_timeout: wrong timeout "
1791 "value %lx\n", timeout);
1792 dump_stack();
1793 current->state = TASK_RUNNING;
1794 goto out;
1795 }
1796 }
1797
1798 expire = timeout + jiffies;
1799
1800 timer.task = current;
1801 timer_setup_on_stack(&timer.timer, process_timeout, 0);
1802 __mod_timer(&timer.timer, expire, 0);
1803 schedule();
1804 del_singleshot_timer_sync(&timer.timer);
1805
1806 /* Remove the timer from the object tracker */
1807 destroy_timer_on_stack(&timer.timer);
1808
1809 timeout = expire - jiffies;
1810
1811 out:
1812 return timeout < 0 ? 0 : timeout;
1813}
1814EXPORT_SYMBOL(schedule_timeout);
1815
1816/*
1817 * We can use __set_current_state() here because schedule_timeout() calls
1818 * schedule() unconditionally.
1819 */
1820signed long __sched schedule_timeout_interruptible(signed long timeout)
1821{
1822 __set_current_state(TASK_INTERRUPTIBLE);
1823 return schedule_timeout(timeout);
1824}
1825EXPORT_SYMBOL(schedule_timeout_interruptible);
1826
1827signed long __sched schedule_timeout_killable(signed long timeout)
1828{
1829 __set_current_state(TASK_KILLABLE);
1830 return schedule_timeout(timeout);
1831}
1832EXPORT_SYMBOL(schedule_timeout_killable);
1833
1834signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1835{
1836 __set_current_state(TASK_UNINTERRUPTIBLE);
1837 return schedule_timeout(timeout);
1838}
1839EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1840
1841/*
1842 * Like schedule_timeout_uninterruptible(), except this task will not contribute
1843 * to load average.
1844 */
1845signed long __sched schedule_timeout_idle(signed long timeout)
1846{
1847 __set_current_state(TASK_IDLE);
1848 return schedule_timeout(timeout);
1849}
1850EXPORT_SYMBOL(schedule_timeout_idle);
1851
1852#ifdef CONFIG_HOTPLUG_CPU
1853static void migrate_timer_list(struct timer_base *new_base, struct hlist_head *head)
1854{
1855 struct timer_list *timer;
1856 int cpu = new_base->cpu;
1857
1858 while (!hlist_empty(head)) {
1859 timer = hlist_entry(head->first, struct timer_list, entry);
1860 detach_timer(timer, false);
1861 timer->flags = (timer->flags & ~TIMER_BASEMASK) | cpu;
1862 internal_add_timer(new_base, timer);
1863 }
1864}
1865
1866int timers_prepare_cpu(unsigned int cpu)
1867{
1868 struct timer_base *base;
1869 int b;
1870
1871 for (b = 0; b < NR_BASES; b++) {
1872 base = per_cpu_ptr(&timer_bases[b], cpu);
1873 base->clk = jiffies;
1874 base->next_expiry = base->clk + NEXT_TIMER_MAX_DELTA;
1875 base->is_idle = false;
1876 base->must_forward_clk = true;
1877 }
1878 return 0;
1879}
1880
1881int timers_dead_cpu(unsigned int cpu)
1882{
1883 struct timer_base *old_base;
1884 struct timer_base *new_base;
1885 int b, i;
1886
1887 BUG_ON(cpu_online(cpu));
1888
1889 for (b = 0; b < NR_BASES; b++) {
1890 old_base = per_cpu_ptr(&timer_bases[b], cpu);
1891 new_base = get_cpu_ptr(&timer_bases[b]);
1892 /*
1893 * The caller is globally serialized and nobody else
1894 * takes two locks at once, deadlock is not possible.
1895 */
1896 raw_spin_lock_irq(&new_base->lock);
1897 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1898
1899 /*
1900 * The current CPUs base clock might be stale. Update it
1901 * before moving the timers over.
1902 */
1903 forward_timer_base(new_base);
1904
1905 BUG_ON(old_base->running_timer);
1906
1907 for (i = 0; i < WHEEL_SIZE; i++)
1908 migrate_timer_list(new_base, old_base->vectors + i);
1909
1910 raw_spin_unlock(&old_base->lock);
1911 raw_spin_unlock_irq(&new_base->lock);
1912 put_cpu_ptr(&timer_bases);
1913 }
1914 return 0;
1915}
1916
1917#endif /* CONFIG_HOTPLUG_CPU */
1918
1919static void __init init_timer_cpu(int cpu)
1920{
1921 struct timer_base *base;
1922 int i;
1923
1924 for (i = 0; i < NR_BASES; i++) {
1925 base = per_cpu_ptr(&timer_bases[i], cpu);
1926 base->cpu = cpu;
1927 raw_spin_lock_init(&base->lock);
1928 base->clk = jiffies;
1929 }
1930}
1931
1932static void __init init_timer_cpus(void)
1933{
1934 int cpu;
1935
1936 for_each_possible_cpu(cpu)
1937 init_timer_cpu(cpu);
1938}
1939
1940void __init init_timers(void)
1941{
1942 init_timer_cpus();
1943 open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
1944}
1945
1946/**
1947 * msleep - sleep safely even with waitqueue interruptions
1948 * @msecs: Time in milliseconds to sleep for
1949 */
1950void msleep(unsigned int msecs)
1951{
1952 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1953
1954 while (timeout)
1955 timeout = schedule_timeout_uninterruptible(timeout);
1956}
1957
1958EXPORT_SYMBOL(msleep);
1959
1960/**
1961 * msleep_interruptible - sleep waiting for signals
1962 * @msecs: Time in milliseconds to sleep for
1963 */
1964unsigned long msleep_interruptible(unsigned int msecs)
1965{
1966 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1967
1968 while (timeout && !signal_pending(current))
1969 timeout = schedule_timeout_interruptible(timeout);
1970 return jiffies_to_msecs(timeout);
1971}
1972
1973EXPORT_SYMBOL(msleep_interruptible);
1974
1975/**
1976 * usleep_range - Sleep for an approximate time
1977 * @min: Minimum time in usecs to sleep
1978 * @max: Maximum time in usecs to sleep
1979 *
1980 * In non-atomic context where the exact wakeup time is flexible, use
1981 * usleep_range() instead of udelay(). The sleep improves responsiveness
1982 * by avoiding the CPU-hogging busy-wait of udelay(), and the range reduces
1983 * power usage by allowing hrtimers to take advantage of an already-
1984 * scheduled interrupt instead of scheduling a new one just for this sleep.
1985 */
1986void __sched usleep_range(unsigned long min, unsigned long max)
1987{
1988 ktime_t exp = ktime_add_us(ktime_get(), min);
1989 u64 delta = (u64)(max - min) * NSEC_PER_USEC;
1990
1991 for (;;) {
1992 __set_current_state(TASK_UNINTERRUPTIBLE);
1993 /* Do not return before the requested sleep time has elapsed */
1994 if (!schedule_hrtimeout_range(&exp, delta, HRTIMER_MODE_ABS))
1995 break;
1996 }
1997}
1998EXPORT_SYMBOL(usleep_range);