projects/linux·blob v5.3 · refs

2002 lines · 57.1 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 trace_timer_start(timer, timer->expires, timer->flags);
541}
542
543static void
544__internal_add_timer(struct timer_base *base, struct timer_list *timer)
545{
546 unsigned int idx;
547
548 idx = calc_wheel_index(timer->expires, base->clk);
549 enqueue_timer(base, timer, idx);
550}
551
552static void
553trigger_dyntick_cpu(struct timer_base *base, struct timer_list *timer)
554{
555 if (!is_timers_nohz_active())
556 return;
557
558 /*
559 * TODO: This wants some optimizing similar to the code below, but we
560 * will do that when we switch from push to pull for deferrable timers.
561 */
562 if (timer->flags & TIMER_DEFERRABLE) {
563 if (tick_nohz_full_cpu(base->cpu))
564 wake_up_nohz_cpu(base->cpu);
565 return;
566 }
567
568 /*
569 * We might have to IPI the remote CPU if the base is idle and the
570 * timer is not deferrable. If the other CPU is on the way to idle
571 * then it can't set base->is_idle as we hold the base lock:
572 */
573 if (!base->is_idle)
574 return;
575
576 /* Check whether this is the new first expiring timer: */
577 if (time_after_eq(timer->expires, base->next_expiry))
578 return;
579
580 /*
581 * Set the next expiry time and kick the CPU so it can reevaluate the
582 * wheel:
583 */
584 base->next_expiry = timer->expires;
585 wake_up_nohz_cpu(base->cpu);
586}
587
588static void
589internal_add_timer(struct timer_base *base, struct timer_list *timer)
590{
591 __internal_add_timer(base, timer);
592 trigger_dyntick_cpu(base, timer);
593}
594
595#ifdef CONFIG_DEBUG_OBJECTS_TIMERS
596
597static struct debug_obj_descr timer_debug_descr;
598
599static void *timer_debug_hint(void *addr)
600{
601 return ((struct timer_list *) addr)->function;
602}
603
604static bool timer_is_static_object(void *addr)
605{
606 struct timer_list *timer = addr;
607
608 return (timer->entry.pprev == NULL &&
609 timer->entry.next == TIMER_ENTRY_STATIC);
610}
611
612/*
613 * fixup_init is called when:
614 * - an active object is initialized
615 */
616static bool timer_fixup_init(void *addr, enum debug_obj_state state)
617{
618 struct timer_list *timer = addr;
619
620 switch (state) {
621 case ODEBUG_STATE_ACTIVE:
622 del_timer_sync(timer);
623 debug_object_init(timer, &timer_debug_descr);
624 return true;
625 default:
626 return false;
627 }
628}
629
630/* Stub timer callback for improperly used timers. */
631static void stub_timer(struct timer_list *unused)
632{
633 WARN_ON(1);
634}
635
636/*
637 * fixup_activate is called when:
638 * - an active object is activated
639 * - an unknown non-static object is activated
640 */
641static bool timer_fixup_activate(void *addr, enum debug_obj_state state)
642{
643 struct timer_list *timer = addr;
644
645 switch (state) {
646 case ODEBUG_STATE_NOTAVAILABLE:
647 timer_setup(timer, stub_timer, 0);
648 return true;
649
650 case ODEBUG_STATE_ACTIVE:
651 WARN_ON(1);
652 /* fall through */
653 default:
654 return false;
655 }
656}
657
658/*
659 * fixup_free is called when:
660 * - an active object is freed
661 */
662static bool timer_fixup_free(void *addr, enum debug_obj_state state)
663{
664 struct timer_list *timer = addr;
665
666 switch (state) {
667 case ODEBUG_STATE_ACTIVE:
668 del_timer_sync(timer);
669 debug_object_free(timer, &timer_debug_descr);
670 return true;
671 default:
672 return false;
673 }
674}
675
676/*
677 * fixup_assert_init is called when:
678 * - an untracked/uninit-ed object is found
679 */
680static bool timer_fixup_assert_init(void *addr, enum debug_obj_state state)
681{
682 struct timer_list *timer = addr;
683
684 switch (state) {
685 case ODEBUG_STATE_NOTAVAILABLE:
686 timer_setup(timer, stub_timer, 0);
687 return true;
688 default:
689 return false;
690 }
691}
692
693static struct debug_obj_descr timer_debug_descr = {
694 .name = "timer_list",
695 .debug_hint = timer_debug_hint,
696 .is_static_object = timer_is_static_object,
697 .fixup_init = timer_fixup_init,
698 .fixup_activate = timer_fixup_activate,
699 .fixup_free = timer_fixup_free,
700 .fixup_assert_init = timer_fixup_assert_init,
701};
702
703static inline void debug_timer_init(struct timer_list *timer)
704{
705 debug_object_init(timer, &timer_debug_descr);
706}
707
708static inline void debug_timer_activate(struct timer_list *timer)
709{
710 debug_object_activate(timer, &timer_debug_descr);
711}
712
713static inline void debug_timer_deactivate(struct timer_list *timer)
714{
715 debug_object_deactivate(timer, &timer_debug_descr);
716}
717
718static inline void debug_timer_free(struct timer_list *timer)
719{
720 debug_object_free(timer, &timer_debug_descr);
721}
722
723static inline void debug_timer_assert_init(struct timer_list *timer)
724{
725 debug_object_assert_init(timer, &timer_debug_descr);
726}
727
728static void do_init_timer(struct timer_list *timer,
729 void (*func)(struct timer_list *),
730 unsigned int flags,
731 const char *name, struct lock_class_key *key);
732
733void init_timer_on_stack_key(struct timer_list *timer,
734 void (*func)(struct timer_list *),
735 unsigned int flags,
736 const char *name, struct lock_class_key *key)
737{
738 debug_object_init_on_stack(timer, &timer_debug_descr);
739 do_init_timer(timer, func, flags, name, key);
740}
741EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
742
743void destroy_timer_on_stack(struct timer_list *timer)
744{
745 debug_object_free(timer, &timer_debug_descr);
746}
747EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
748
749#else
750static inline void debug_timer_init(struct timer_list *timer) { }
751static inline void debug_timer_activate(struct timer_list *timer) { }
752static inline void debug_timer_deactivate(struct timer_list *timer) { }
753static inline void debug_timer_assert_init(struct timer_list *timer) { }
754#endif
755
756static inline void debug_init(struct timer_list *timer)
757{
758 debug_timer_init(timer);
759 trace_timer_init(timer);
760}
761
762static inline void debug_deactivate(struct timer_list *timer)
763{
764 debug_timer_deactivate(timer);
765 trace_timer_cancel(timer);
766}
767
768static inline void debug_assert_init(struct timer_list *timer)
769{
770 debug_timer_assert_init(timer);
771}
772
773static void do_init_timer(struct timer_list *timer,
774 void (*func)(struct timer_list *),
775 unsigned int flags,
776 const char *name, struct lock_class_key *key)
777{
778 timer->entry.pprev = NULL;
779 timer->function = func;
780 timer->flags = flags | raw_smp_processor_id();
781 lockdep_init_map(&timer->lockdep_map, name, key, 0);
782}
783
784/**
785 * init_timer_key - initialize a timer
786 * @timer: the timer to be initialized
787 * @func: timer callback function
788 * @flags: timer flags
789 * @name: name of the timer
790 * @key: lockdep class key of the fake lock used for tracking timer
791 * sync lock dependencies
792 *
793 * init_timer_key() must be done to a timer prior calling *any* of the
794 * other timer functions.
795 */
796void init_timer_key(struct timer_list *timer,
797 void (*func)(struct timer_list *), unsigned int flags,
798 const char *name, struct lock_class_key *key)
799{
800 debug_init(timer);
801 do_init_timer(timer, func, flags, name, key);
802}
803EXPORT_SYMBOL(init_timer_key);
804
805static inline void detach_timer(struct timer_list *timer, bool clear_pending)
806{
807 struct hlist_node *entry = &timer->entry;
808
809 debug_deactivate(timer);
810
811 __hlist_del(entry);
812 if (clear_pending)
813 entry->pprev = NULL;
814 entry->next = LIST_POISON2;
815}
816
817static int detach_if_pending(struct timer_list *timer, struct timer_base *base,
818 bool clear_pending)
819{
820 unsigned idx = timer_get_idx(timer);
821
822 if (!timer_pending(timer))
823 return 0;
824
825 if (hlist_is_singular_node(&timer->entry, base->vectors + idx))
826 __clear_bit(idx, base->pending_map);
827
828 detach_timer(timer, clear_pending);
829 return 1;
830}
831
832static inline struct timer_base *get_timer_cpu_base(u32 tflags, u32 cpu)
833{
834 struct timer_base *base = per_cpu_ptr(&timer_bases[BASE_STD], cpu);
835
836 /*
837 * If the timer is deferrable and NO_HZ_COMMON is set then we need
838 * to use the deferrable base.
839 */
840 if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE))
841 base = per_cpu_ptr(&timer_bases[BASE_DEF], cpu);
842 return base;
843}
844
845static inline struct timer_base *get_timer_this_cpu_base(u32 tflags)
846{
847 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
848
849 /*
850 * If the timer is deferrable and NO_HZ_COMMON is set then we need
851 * to use the deferrable base.
852 */
853 if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE))
854 base = this_cpu_ptr(&timer_bases[BASE_DEF]);
855 return base;
856}
857
858static inline struct timer_base *get_timer_base(u32 tflags)
859{
860 return get_timer_cpu_base(tflags, tflags & TIMER_CPUMASK);
861}
862
863static inline struct timer_base *
864get_target_base(struct timer_base *base, unsigned tflags)
865{
866#if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
867 if (static_branch_likely(&timers_migration_enabled) &&
868 !(tflags & TIMER_PINNED))
869 return get_timer_cpu_base(tflags, get_nohz_timer_target());
870#endif
871 return get_timer_this_cpu_base(tflags);
872}
873
874static inline void forward_timer_base(struct timer_base *base)
875{
876#ifdef CONFIG_NO_HZ_COMMON
877 unsigned long jnow;
878
879 /*
880 * We only forward the base when we are idle or have just come out of
881 * idle (must_forward_clk logic), and have a delta between base clock
882 * and jiffies. In the common case, run_timers will take care of it.
883 */
884 if (likely(!base->must_forward_clk))
885 return;
886
887 jnow = READ_ONCE(jiffies);
888 base->must_forward_clk = base->is_idle;
889 if ((long)(jnow - base->clk) < 2)
890 return;
891
892 /*
893 * If the next expiry value is > jiffies, then we fast forward to
894 * jiffies otherwise we forward to the next expiry value.
895 */
896 if (time_after(base->next_expiry, jnow))
897 base->clk = jnow;
898 else
899 base->clk = base->next_expiry;
900#endif
901}
902
903
904/*
905 * We are using hashed locking: Holding per_cpu(timer_bases[x]).lock means
906 * that all timers which are tied to this base are locked, and the base itself
907 * is locked too.
908 *
909 * So __run_timers/migrate_timers can safely modify all timers which could
910 * be found in the base->vectors array.
911 *
912 * When a timer is migrating then the TIMER_MIGRATING flag is set and we need
913 * to wait until the migration is done.
914 */
915static struct timer_base *lock_timer_base(struct timer_list *timer,
916 unsigned long *flags)
917 __acquires(timer->base->lock)
918{
919 for (;;) {
920 struct timer_base *base;
921 u32 tf;
922
923 /*
924 * We need to use READ_ONCE() here, otherwise the compiler
925 * might re-read @tf between the check for TIMER_MIGRATING
926 * and spin_lock().
927 */
928 tf = READ_ONCE(timer->flags);
929
930 if (!(tf & TIMER_MIGRATING)) {
931 base = get_timer_base(tf);
932 raw_spin_lock_irqsave(&base->lock, *flags);
933 if (timer->flags == tf)
934 return base;
935 raw_spin_unlock_irqrestore(&base->lock, *flags);
936 }
937 cpu_relax();
938 }
939}
940
941#define MOD_TIMER_PENDING_ONLY 0x01
942#define MOD_TIMER_REDUCE 0x02
943
944static inline int
945__mod_timer(struct timer_list *timer, unsigned long expires, unsigned int options)
946{
947 struct timer_base *base, *new_base;
948 unsigned int idx = UINT_MAX;
949 unsigned long clk = 0, flags;
950 int ret = 0;
951
952 BUG_ON(!timer->function);
953
954 /*
955 * This is a common optimization triggered by the networking code - if
956 * the timer is re-modified to have the same timeout or ends up in the
957 * same array bucket then just return:
958 */
959 if (timer_pending(timer)) {
960 /*
961 * The downside of this optimization is that it can result in
962 * larger granularity than you would get from adding a new
963 * timer with this expiry.
964 */
965 long diff = timer->expires - expires;
966
967 if (!diff)
968 return 1;
969 if (options & MOD_TIMER_REDUCE && diff <= 0)
970 return 1;
971
972 /*
973 * We lock timer base and calculate the bucket index right
974 * here. If the timer ends up in the same bucket, then we
975 * just update the expiry time and avoid the whole
976 * dequeue/enqueue dance.
977 */
978 base = lock_timer_base(timer, &flags);
979 forward_timer_base(base);
980
981 if (timer_pending(timer) && (options & MOD_TIMER_REDUCE) &&
982 time_before_eq(timer->expires, expires)) {
983 ret = 1;
984 goto out_unlock;
985 }
986
987 clk = base->clk;
988 idx = calc_wheel_index(expires, clk);
989
990 /*
991 * Retrieve and compare the array index of the pending
992 * timer. If it matches set the expiry to the new value so a
993 * subsequent call will exit in the expires check above.
994 */
995 if (idx == timer_get_idx(timer)) {
996 if (!(options & MOD_TIMER_REDUCE))
997 timer->expires = expires;
998 else if (time_after(timer->expires, expires))
999 timer->expires = expires;
1000 ret = 1;
1001 goto out_unlock;
1002 }
1003 } else {
1004 base = lock_timer_base(timer, &flags);
1005 forward_timer_base(base);
1006 }
1007
1008 ret = detach_if_pending(timer, base, false);
1009 if (!ret && (options & MOD_TIMER_PENDING_ONLY))
1010 goto out_unlock;
1011
1012 new_base = get_target_base(base, timer->flags);
1013
1014 if (base != new_base) {
1015 /*
1016 * We are trying to schedule the timer on the new base.
1017 * However we can't change timer's base while it is running,
1018 * otherwise del_timer_sync() can't detect that the timer's
1019 * handler yet has not finished. This also guarantees that the
1020 * timer is serialized wrt itself.
1021 */
1022 if (likely(base->running_timer != timer)) {
1023 /* See the comment in lock_timer_base() */
1024 timer->flags |= TIMER_MIGRATING;
1025
1026 raw_spin_unlock(&base->lock);
1027 base = new_base;
1028 raw_spin_lock(&base->lock);
1029 WRITE_ONCE(timer->flags,
1030 (timer->flags & ~TIMER_BASEMASK) | base->cpu);
1031 forward_timer_base(base);
1032 }
1033 }
1034
1035 debug_timer_activate(timer);
1036
1037 timer->expires = expires;
1038 /*
1039 * If 'idx' was calculated above and the base time did not advance
1040 * between calculating 'idx' and possibly switching the base, only
1041 * enqueue_timer() and trigger_dyntick_cpu() is required. Otherwise
1042 * we need to (re)calculate the wheel index via
1043 * internal_add_timer().
1044 */
1045 if (idx != UINT_MAX && clk == base->clk) {
1046 enqueue_timer(base, timer, idx);
1047 trigger_dyntick_cpu(base, timer);
1048 } else {
1049 internal_add_timer(base, timer);
1050 }
1051
1052out_unlock:
1053 raw_spin_unlock_irqrestore(&base->lock, flags);
1054
1055 return ret;
1056}
1057
1058/**
1059 * mod_timer_pending - modify a pending timer's timeout
1060 * @timer: the pending timer to be modified
1061 * @expires: new timeout in jiffies
1062 *
1063 * mod_timer_pending() is the same for pending timers as mod_timer(),
1064 * but will not re-activate and modify already deleted timers.
1065 *
1066 * It is useful for unserialized use of timers.
1067 */
1068int mod_timer_pending(struct timer_list *timer, unsigned long expires)
1069{
1070 return __mod_timer(timer, expires, MOD_TIMER_PENDING_ONLY);
1071}
1072EXPORT_SYMBOL(mod_timer_pending);
1073
1074/**
1075 * mod_timer - modify a timer's timeout
1076 * @timer: the timer to be modified
1077 * @expires: new timeout in jiffies
1078 *
1079 * mod_timer() is a more efficient way to update the expire field of an
1080 * active timer (if the timer is inactive it will be activated)
1081 *
1082 * mod_timer(timer, expires) is equivalent to:
1083 *
1084 * del_timer(timer); timer->expires = expires; add_timer(timer);
1085 *
1086 * Note that if there are multiple unserialized concurrent users of the
1087 * same timer, then mod_timer() is the only safe way to modify the timeout,
1088 * since add_timer() cannot modify an already running timer.
1089 *
1090 * The function returns whether it has modified a pending timer or not.
1091 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
1092 * active timer returns 1.)
1093 */
1094int mod_timer(struct timer_list *timer, unsigned long expires)
1095{
1096 return __mod_timer(timer, expires, 0);
1097}
1098EXPORT_SYMBOL(mod_timer);
1099
1100/**
1101 * timer_reduce - Modify a timer's timeout if it would reduce the timeout
1102 * @timer: The timer to be modified
1103 * @expires: New timeout in jiffies
1104 *
1105 * timer_reduce() is very similar to mod_timer(), except that it will only
1106 * modify a running timer if that would reduce the expiration time (it will
1107 * start a timer that isn't running).
1108 */
1109int timer_reduce(struct timer_list *timer, unsigned long expires)
1110{
1111 return __mod_timer(timer, expires, MOD_TIMER_REDUCE);
1112}
1113EXPORT_SYMBOL(timer_reduce);
1114
1115/**
1116 * add_timer - start a timer
1117 * @timer: the timer to be added
1118 *
1119 * The kernel will do a ->function(@timer) callback from the
1120 * timer interrupt at the ->expires point in the future. The
1121 * current time is 'jiffies'.
1122 *
1123 * The timer's ->expires, ->function fields must be set prior calling this
1124 * function.
1125 *
1126 * Timers with an ->expires field in the past will be executed in the next
1127 * timer tick.
1128 */
1129void add_timer(struct timer_list *timer)
1130{
1131 BUG_ON(timer_pending(timer));
1132 mod_timer(timer, timer->expires);
1133}
1134EXPORT_SYMBOL(add_timer);
1135
1136/**
1137 * add_timer_on - start a timer on a particular CPU
1138 * @timer: the timer to be added
1139 * @cpu: the CPU to start it on
1140 *
1141 * This is not very scalable on SMP. Double adds are not possible.
1142 */
1143void add_timer_on(struct timer_list *timer, int cpu)
1144{
1145 struct timer_base *new_base, *base;
1146 unsigned long flags;
1147
1148 BUG_ON(timer_pending(timer) || !timer->function);
1149
1150 new_base = get_timer_cpu_base(timer->flags, cpu);
1151
1152 /*
1153 * If @timer was on a different CPU, it should be migrated with the
1154 * old base locked to prevent other operations proceeding with the
1155 * wrong base locked. See lock_timer_base().
1156 */
1157 base = lock_timer_base(timer, &flags);
1158 if (base != new_base) {
1159 timer->flags |= TIMER_MIGRATING;
1160
1161 raw_spin_unlock(&base->lock);
1162 base = new_base;
1163 raw_spin_lock(&base->lock);
1164 WRITE_ONCE(timer->flags,
1165 (timer->flags & ~TIMER_BASEMASK) | cpu);
1166 }
1167 forward_timer_base(base);
1168
1169 debug_timer_activate(timer);
1170 internal_add_timer(base, timer);
1171 raw_spin_unlock_irqrestore(&base->lock, flags);
1172}
1173EXPORT_SYMBOL_GPL(add_timer_on);
1174
1175/**
1176 * del_timer - deactivate a timer.
1177 * @timer: the timer to be deactivated
1178 *
1179 * del_timer() deactivates a timer - this works on both active and inactive
1180 * timers.
1181 *
1182 * The function returns whether it has deactivated a pending timer or not.
1183 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
1184 * active timer returns 1.)
1185 */
1186int del_timer(struct timer_list *timer)
1187{
1188 struct timer_base *base;
1189 unsigned long flags;
1190 int ret = 0;
1191
1192 debug_assert_init(timer);
1193
1194 if (timer_pending(timer)) {
1195 base = lock_timer_base(timer, &flags);
1196 ret = detach_if_pending(timer, base, true);
1197 raw_spin_unlock_irqrestore(&base->lock, flags);
1198 }
1199
1200 return ret;
1201}
1202EXPORT_SYMBOL(del_timer);
1203
1204/**
1205 * try_to_del_timer_sync - Try to deactivate a timer
1206 * @timer: timer to delete
1207 *
1208 * This function tries to deactivate a timer. Upon successful (ret >= 0)
1209 * exit the timer is not queued and the handler is not running on any CPU.
1210 */
1211int try_to_del_timer_sync(struct timer_list *timer)
1212{
1213 struct timer_base *base;
1214 unsigned long flags;
1215 int ret = -1;
1216
1217 debug_assert_init(timer);
1218
1219 base = lock_timer_base(timer, &flags);
1220
1221 if (base->running_timer != timer)
1222 ret = detach_if_pending(timer, base, true);
1223
1224 raw_spin_unlock_irqrestore(&base->lock, flags);
1225
1226 return ret;
1227}
1228EXPORT_SYMBOL(try_to_del_timer_sync);
1229
1230#ifdef CONFIG_SMP
1231/**
1232 * del_timer_sync - deactivate a timer and wait for the handler to finish.
1233 * @timer: the timer to be deactivated
1234 *
1235 * This function only differs from del_timer() on SMP: besides deactivating
1236 * the timer it also makes sure the handler has finished executing on other
1237 * CPUs.
1238 *
1239 * Synchronization rules: Callers must prevent restarting of the timer,
1240 * otherwise this function is meaningless. It must not be called from
1241 * interrupt contexts unless the timer is an irqsafe one. The caller must
1242 * not hold locks which would prevent completion of the timer's
1243 * handler. The timer's handler must not call add_timer_on(). Upon exit the
1244 * timer is not queued and the handler is not running on any CPU.
1245 *
1246 * Note: For !irqsafe timers, you must not hold locks that are held in
1247 * interrupt context while calling this function. Even if the lock has
1248 * nothing to do with the timer in question. Here's why::
1249 *
1250 * CPU0 CPU1
1251 * ---- ----
1252 * <SOFTIRQ>
1253 * call_timer_fn();
1254 * base->running_timer = mytimer;
1255 * spin_lock_irq(somelock);
1256 * <IRQ>
1257 * spin_lock(somelock);
1258 * del_timer_sync(mytimer);
1259 * while (base->running_timer == mytimer);
1260 *
1261 * Now del_timer_sync() will never return and never release somelock.
1262 * The interrupt on the other CPU is waiting to grab somelock but
1263 * it has interrupted the softirq that CPU0 is waiting to finish.
1264 *
1265 * The function returns whether it has deactivated a pending timer or not.
1266 */
1267int del_timer_sync(struct timer_list *timer)
1268{
1269#ifdef CONFIG_LOCKDEP
1270 unsigned long flags;
1271
1272 /*
1273 * If lockdep gives a backtrace here, please reference
1274 * the synchronization rules above.
1275 */
1276 local_irq_save(flags);
1277 lock_map_acquire(&timer->lockdep_map);
1278 lock_map_release(&timer->lockdep_map);
1279 local_irq_restore(flags);
1280#endif
1281 /*
1282 * don't use it in hardirq context, because it
1283 * could lead to deadlock.
1284 */
1285 WARN_ON(in_irq() && !(timer->flags & TIMER_IRQSAFE));
1286 for (;;) {
1287 int ret = try_to_del_timer_sync(timer);
1288 if (ret >= 0)
1289 return ret;
1290 cpu_relax();
1291 }
1292}
1293EXPORT_SYMBOL(del_timer_sync);
1294#endif
1295
1296static void call_timer_fn(struct timer_list *timer,
1297 void (*fn)(struct timer_list *),
1298 unsigned long baseclk)
1299{
1300 int count = preempt_count();
1301
1302#ifdef CONFIG_LOCKDEP
1303 /*
1304 * It is permissible to free the timer from inside the
1305 * function that is called from it, this we need to take into
1306 * account for lockdep too. To avoid bogus "held lock freed"
1307 * warnings as well as problems when looking into
1308 * timer->lockdep_map, make a copy and use that here.
1309 */
1310 struct lockdep_map lockdep_map;
1311
1312 lockdep_copy_map(&lockdep_map, &timer->lockdep_map);
1313#endif
1314 /*
1315 * Couple the lock chain with the lock chain at
1316 * del_timer_sync() by acquiring the lock_map around the fn()
1317 * call here and in del_timer_sync().
1318 */
1319 lock_map_acquire(&lockdep_map);
1320
1321 trace_timer_expire_entry(timer, baseclk);
1322 fn(timer);
1323 trace_timer_expire_exit(timer);
1324
1325 lock_map_release(&lockdep_map);
1326
1327 if (count != preempt_count()) {
1328 WARN_ONCE(1, "timer: %pS preempt leak: %08x -> %08x\n",
1329 fn, count, preempt_count());
1330 /*
1331 * Restore the preempt count. That gives us a decent
1332 * chance to survive and extract information. If the
1333 * callback kept a lock held, bad luck, but not worse
1334 * than the BUG() we had.
1335 */
1336 preempt_count_set(count);
1337 }
1338}
1339
1340static void expire_timers(struct timer_base *base, struct hlist_head *head)
1341{
1342 /*
1343 * This value is required only for tracing. base->clk was
1344 * incremented directly before expire_timers was called. But expiry
1345 * is related to the old base->clk value.
1346 */
1347 unsigned long baseclk = base->clk - 1;
1348
1349 while (!hlist_empty(head)) {
1350 struct timer_list *timer;
1351 void (*fn)(struct timer_list *);
1352
1353 timer = hlist_entry(head->first, struct timer_list, entry);
1354
1355 base->running_timer = timer;
1356 detach_timer(timer, true);
1357
1358 fn = timer->function;
1359
1360 if (timer->flags & TIMER_IRQSAFE) {
1361 raw_spin_unlock(&base->lock);
1362 call_timer_fn(timer, fn, baseclk);
1363 raw_spin_lock(&base->lock);
1364 } else {
1365 raw_spin_unlock_irq(&base->lock);
1366 call_timer_fn(timer, fn, baseclk);
1367 raw_spin_lock_irq(&base->lock);
1368 }
1369 }
1370}
1371
1372static int __collect_expired_timers(struct timer_base *base,
1373 struct hlist_head *heads)
1374{
1375 unsigned long clk = base->clk;
1376 struct hlist_head *vec;
1377 int i, levels = 0;
1378 unsigned int idx;
1379
1380 for (i = 0; i < LVL_DEPTH; i++) {
1381 idx = (clk & LVL_MASK) + i * LVL_SIZE;
1382
1383 if (__test_and_clear_bit(idx, base->pending_map)) {
1384 vec = base->vectors + idx;
1385 hlist_move_list(vec, heads++);
1386 levels++;
1387 }
1388 /* Is it time to look at the next level? */
1389 if (clk & LVL_CLK_MASK)
1390 break;
1391 /* Shift clock for the next level granularity */
1392 clk >>= LVL_CLK_SHIFT;
1393 }
1394 return levels;
1395}
1396
1397#ifdef CONFIG_NO_HZ_COMMON
1398/*
1399 * Find the next pending bucket of a level. Search from level start (@offset)
1400 * + @clk upwards and if nothing there, search from start of the level
1401 * (@offset) up to @offset + clk.
1402 */
1403static int next_pending_bucket(struct timer_base *base, unsigned offset,
1404 unsigned clk)
1405{
1406 unsigned pos, start = offset + clk;
1407 unsigned end = offset + LVL_SIZE;
1408
1409 pos = find_next_bit(base->pending_map, end, start);
1410 if (pos < end)
1411 return pos - start;
1412
1413 pos = find_next_bit(base->pending_map, start, offset);
1414 return pos < start ? pos + LVL_SIZE - start : -1;
1415}
1416
1417/*
1418 * Search the first expiring timer in the various clock levels. Caller must
1419 * hold base->lock.
1420 */
1421static unsigned long __next_timer_interrupt(struct timer_base *base)
1422{
1423 unsigned long clk, next, adj;
1424 unsigned lvl, offset = 0;
1425
1426 next = base->clk + NEXT_TIMER_MAX_DELTA;
1427 clk = base->clk;
1428 for (lvl = 0; lvl < LVL_DEPTH; lvl++, offset += LVL_SIZE) {
1429 int pos = next_pending_bucket(base, offset, clk & LVL_MASK);
1430
1431 if (pos >= 0) {
1432 unsigned long tmp = clk + (unsigned long) pos;
1433
1434 tmp <<= LVL_SHIFT(lvl);
1435 if (time_before(tmp, next))
1436 next = tmp;
1437 }
1438 /*
1439 * Clock for the next level. If the current level clock lower
1440 * bits are zero, we look at the next level as is. If not we
1441 * need to advance it by one because that's going to be the
1442 * next expiring bucket in that level. base->clk is the next
1443 * expiring jiffie. So in case of:
1444 *
1445 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1446 * 0 0 0 0 0 0
1447 *
1448 * we have to look at all levels @index 0. With
1449 *
1450 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1451 * 0 0 0 0 0 2
1452 *
1453 * LVL0 has the next expiring bucket @index 2. The upper
1454 * levels have the next expiring bucket @index 1.
1455 *
1456 * In case that the propagation wraps the next level the same
1457 * rules apply:
1458 *
1459 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1460 * 0 0 0 0 F 2
1461 *
1462 * So after looking at LVL0 we get:
1463 *
1464 * LVL5 LVL4 LVL3 LVL2 LVL1
1465 * 0 0 0 1 0
1466 *
1467 * So no propagation from LVL1 to LVL2 because that happened
1468 * with the add already, but then we need to propagate further
1469 * from LVL2 to LVL3.
1470 *
1471 * So the simple check whether the lower bits of the current
1472 * level are 0 or not is sufficient for all cases.
1473 */
1474 adj = clk & LVL_CLK_MASK ? 1 : 0;
1475 clk >>= LVL_CLK_SHIFT;
1476 clk += adj;
1477 }
1478 return next;
1479}
1480
1481/*
1482 * Check, if the next hrtimer event is before the next timer wheel
1483 * event:
1484 */
1485static u64 cmp_next_hrtimer_event(u64 basem, u64 expires)
1486{
1487 u64 nextevt = hrtimer_get_next_event();
1488
1489 /*
1490 * If high resolution timers are enabled
1491 * hrtimer_get_next_event() returns KTIME_MAX.
1492 */
1493 if (expires <= nextevt)
1494 return expires;
1495
1496 /*
1497 * If the next timer is already expired, return the tick base
1498 * time so the tick is fired immediately.
1499 */
1500 if (nextevt <= basem)
1501 return basem;
1502
1503 /*
1504 * Round up to the next jiffie. High resolution timers are
1505 * off, so the hrtimers are expired in the tick and we need to
1506 * make sure that this tick really expires the timer to avoid
1507 * a ping pong of the nohz stop code.
1508 *
1509 * Use DIV_ROUND_UP_ULL to prevent gcc calling __divdi3
1510 */
1511 return DIV_ROUND_UP_ULL(nextevt, TICK_NSEC) * TICK_NSEC;
1512}
1513
1514/**
1515 * get_next_timer_interrupt - return the time (clock mono) of the next timer
1516 * @basej: base time jiffies
1517 * @basem: base time clock monotonic
1518 *
1519 * Returns the tick aligned clock monotonic time of the next pending
1520 * timer or KTIME_MAX if no timer is pending.
1521 */
1522u64 get_next_timer_interrupt(unsigned long basej, u64 basem)
1523{
1524 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1525 u64 expires = KTIME_MAX;
1526 unsigned long nextevt;
1527 bool is_max_delta;
1528
1529 /*
1530 * Pretend that there is no timer pending if the cpu is offline.
1531 * Possible pending timers will be migrated later to an active cpu.
1532 */
1533 if (cpu_is_offline(smp_processor_id()))
1534 return expires;
1535
1536 raw_spin_lock(&base->lock);
1537 nextevt = __next_timer_interrupt(base);
1538 is_max_delta = (nextevt == base->clk + NEXT_TIMER_MAX_DELTA);
1539 base->next_expiry = nextevt;
1540 /*
1541 * We have a fresh next event. Check whether we can forward the
1542 * base. We can only do that when @basej is past base->clk
1543 * otherwise we might rewind base->clk.
1544 */
1545 if (time_after(basej, base->clk)) {
1546 if (time_after(nextevt, basej))
1547 base->clk = basej;
1548 else if (time_after(nextevt, base->clk))
1549 base->clk = nextevt;
1550 }
1551
1552 if (time_before_eq(nextevt, basej)) {
1553 expires = basem;
1554 base->is_idle = false;
1555 } else {
1556 if (!is_max_delta)
1557 expires = basem + (u64)(nextevt - basej) * TICK_NSEC;
1558 /*
1559 * If we expect to sleep more than a tick, mark the base idle.
1560 * Also the tick is stopped so any added timer must forward
1561 * the base clk itself to keep granularity small. This idle
1562 * logic is only maintained for the BASE_STD base, deferrable
1563 * timers may still see large granularity skew (by design).
1564 */
1565 if ((expires - basem) > TICK_NSEC) {
1566 base->must_forward_clk = true;
1567 base->is_idle = true;
1568 }
1569 }
1570 raw_spin_unlock(&base->lock);
1571
1572 return cmp_next_hrtimer_event(basem, expires);
1573}
1574
1575/**
1576 * timer_clear_idle - Clear the idle state of the timer base
1577 *
1578 * Called with interrupts disabled
1579 */
1580void timer_clear_idle(void)
1581{
1582 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1583
1584 /*
1585 * We do this unlocked. The worst outcome is a remote enqueue sending
1586 * a pointless IPI, but taking the lock would just make the window for
1587 * sending the IPI a few instructions smaller for the cost of taking
1588 * the lock in the exit from idle path.
1589 */
1590 base->is_idle = false;
1591}
1592
1593static int collect_expired_timers(struct timer_base *base,
1594 struct hlist_head *heads)
1595{
1596 /*
1597 * NOHZ optimization. After a long idle sleep we need to forward the
1598 * base to current jiffies. Avoid a loop by searching the bitfield for
1599 * the next expiring timer.
1600 */
1601 if ((long)(jiffies - base->clk) > 2) {
1602 unsigned long next = __next_timer_interrupt(base);
1603
1604 /*
1605 * If the next timer is ahead of time forward to current
1606 * jiffies, otherwise forward to the next expiry time:
1607 */
1608 if (time_after(next, jiffies)) {
1609 /*
1610 * The call site will increment base->clk and then
1611 * terminate the expiry loop immediately.
1612 */
1613 base->clk = jiffies;
1614 return 0;
1615 }
1616 base->clk = next;
1617 }
1618 return __collect_expired_timers(base, heads);
1619}
1620#else
1621static inline int collect_expired_timers(struct timer_base *base,
1622 struct hlist_head *heads)
1623{
1624 return __collect_expired_timers(base, heads);
1625}
1626#endif
1627
1628/*
1629 * Called from the timer interrupt handler to charge one tick to the current
1630 * process. user_tick is 1 if the tick is user time, 0 for system.
1631 */
1632void update_process_times(int user_tick)
1633{
1634 struct task_struct *p = current;
1635
1636 /* Note: this timer irq context must be accounted for as well. */
1637 account_process_tick(p, user_tick);
1638 run_local_timers();
1639 rcu_sched_clock_irq(user_tick);
1640#ifdef CONFIG_IRQ_WORK
1641 if (in_irq())
1642 irq_work_tick();
1643#endif
1644 scheduler_tick();
1645 if (IS_ENABLED(CONFIG_POSIX_TIMERS))
1646 run_posix_cpu_timers(p);
1647}
1648
1649/**
1650 * __run_timers - run all expired timers (if any) on this CPU.
1651 * @base: the timer vector to be processed.
1652 */
1653static inline void __run_timers(struct timer_base *base)
1654{
1655 struct hlist_head heads[LVL_DEPTH];
1656 int levels;
1657
1658 if (!time_after_eq(jiffies, base->clk))
1659 return;
1660
1661 raw_spin_lock_irq(&base->lock);
1662
1663 /*
1664 * timer_base::must_forward_clk must be cleared before running
1665 * timers so that any timer functions that call mod_timer() will
1666 * not try to forward the base. Idle tracking / clock forwarding
1667 * logic is only used with BASE_STD timers.
1668 *
1669 * The must_forward_clk flag is cleared unconditionally also for
1670 * the deferrable base. The deferrable base is not affected by idle
1671 * tracking and never forwarded, so clearing the flag is a NOOP.
1672 *
1673 * The fact that the deferrable base is never forwarded can cause
1674 * large variations in granularity for deferrable timers, but they
1675 * can be deferred for long periods due to idle anyway.
1676 */
1677 base->must_forward_clk = false;
1678
1679 while (time_after_eq(jiffies, base->clk)) {
1680
1681 levels = collect_expired_timers(base, heads);
1682 base->clk++;
1683
1684 while (levels--)
1685 expire_timers(base, heads + levels);
1686 }
1687 base->running_timer = NULL;
1688 raw_spin_unlock_irq(&base->lock);
1689}
1690
1691/*
1692 * This function runs timers and the timer-tq in bottom half context.
1693 */
1694static __latent_entropy void run_timer_softirq(struct softirq_action *h)
1695{
1696 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1697
1698 __run_timers(base);
1699 if (IS_ENABLED(CONFIG_NO_HZ_COMMON))
1700 __run_timers(this_cpu_ptr(&timer_bases[BASE_DEF]));
1701}
1702
1703/*
1704 * Called by the local, per-CPU timer interrupt on SMP.
1705 */
1706void run_local_timers(void)
1707{
1708 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1709
1710 hrtimer_run_queues();
1711 /* Raise the softirq only if required. */
1712 if (time_before(jiffies, base->clk)) {
1713 if (!IS_ENABLED(CONFIG_NO_HZ_COMMON))
1714 return;
1715 /* CPU is awake, so check the deferrable base. */
1716 base++;
1717 if (time_before(jiffies, base->clk))
1718 return;
1719 }
1720 raise_softirq(TIMER_SOFTIRQ);
1721}
1722
1723/*
1724 * Since schedule_timeout()'s timer is defined on the stack, it must store
1725 * the target task on the stack as well.
1726 */
1727struct process_timer {
1728 struct timer_list timer;
1729 struct task_struct *task;
1730};
1731
1732static void process_timeout(struct timer_list *t)
1733{
1734 struct process_timer *timeout = from_timer(timeout, t, timer);
1735
1736 wake_up_process(timeout->task);
1737}
1738
1739/**
1740 * schedule_timeout - sleep until timeout
1741 * @timeout: timeout value in jiffies
1742 *
1743 * Make the current task sleep until @timeout jiffies have
1744 * elapsed. The routine will return immediately unless
1745 * the current task state has been set (see set_current_state()).
1746 *
1747 * You can set the task state as follows -
1748 *
1749 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1750 * pass before the routine returns unless the current task is explicitly
1751 * woken up, (e.g. by wake_up_process())".
1752 *
1753 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1754 * delivered to the current task or the current task is explicitly woken
1755 * up.
1756 *
1757 * The current task state is guaranteed to be TASK_RUNNING when this
1758 * routine returns.
1759 *
1760 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1761 * the CPU away without a bound on the timeout. In this case the return
1762 * value will be %MAX_SCHEDULE_TIMEOUT.
1763 *
1764 * Returns 0 when the timer has expired otherwise the remaining time in
1765 * jiffies will be returned. In all cases the return value is guaranteed
1766 * to be non-negative.
1767 */
1768signed long __sched schedule_timeout(signed long timeout)
1769{
1770 struct process_timer timer;
1771 unsigned long expire;
1772
1773 switch (timeout)
1774 {
1775 case MAX_SCHEDULE_TIMEOUT:
1776 /*
1777 * These two special cases are useful to be comfortable
1778 * in the caller. Nothing more. We could take
1779 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1780 * but I' d like to return a valid offset (>=0) to allow
1781 * the caller to do everything it want with the retval.
1782 */
1783 schedule();
1784 goto out;
1785 default:
1786 /*
1787 * Another bit of PARANOID. Note that the retval will be
1788 * 0 since no piece of kernel is supposed to do a check
1789 * for a negative retval of schedule_timeout() (since it
1790 * should never happens anyway). You just have the printk()
1791 * that will tell you if something is gone wrong and where.
1792 */
1793 if (timeout < 0) {
1794 printk(KERN_ERR "schedule_timeout: wrong timeout "
1795 "value %lx\n", timeout);
1796 dump_stack();
1797 current->state = TASK_RUNNING;
1798 goto out;
1799 }
1800 }
1801
1802 expire = timeout + jiffies;
1803
1804 timer.task = current;
1805 timer_setup_on_stack(&timer.timer, process_timeout, 0);
1806 __mod_timer(&timer.timer, expire, 0);
1807 schedule();
1808 del_singleshot_timer_sync(&timer.timer);
1809
1810 /* Remove the timer from the object tracker */
1811 destroy_timer_on_stack(&timer.timer);
1812
1813 timeout = expire - jiffies;
1814
1815 out:
1816 return timeout < 0 ? 0 : timeout;
1817}
1818EXPORT_SYMBOL(schedule_timeout);
1819
1820/*
1821 * We can use __set_current_state() here because schedule_timeout() calls
1822 * schedule() unconditionally.
1823 */
1824signed long __sched schedule_timeout_interruptible(signed long timeout)
1825{
1826 __set_current_state(TASK_INTERRUPTIBLE);
1827 return schedule_timeout(timeout);
1828}
1829EXPORT_SYMBOL(schedule_timeout_interruptible);
1830
1831signed long __sched schedule_timeout_killable(signed long timeout)
1832{
1833 __set_current_state(TASK_KILLABLE);
1834 return schedule_timeout(timeout);
1835}
1836EXPORT_SYMBOL(schedule_timeout_killable);
1837
1838signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1839{
1840 __set_current_state(TASK_UNINTERRUPTIBLE);
1841 return schedule_timeout(timeout);
1842}
1843EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1844
1845/*
1846 * Like schedule_timeout_uninterruptible(), except this task will not contribute
1847 * to load average.
1848 */
1849signed long __sched schedule_timeout_idle(signed long timeout)
1850{
1851 __set_current_state(TASK_IDLE);
1852 return schedule_timeout(timeout);
1853}
1854EXPORT_SYMBOL(schedule_timeout_idle);
1855
1856#ifdef CONFIG_HOTPLUG_CPU
1857static void migrate_timer_list(struct timer_base *new_base, struct hlist_head *head)
1858{
1859 struct timer_list *timer;
1860 int cpu = new_base->cpu;
1861
1862 while (!hlist_empty(head)) {
1863 timer = hlist_entry(head->first, struct timer_list, entry);
1864 detach_timer(timer, false);
1865 timer->flags = (timer->flags & ~TIMER_BASEMASK) | cpu;
1866 internal_add_timer(new_base, timer);
1867 }
1868}
1869
1870int timers_prepare_cpu(unsigned int cpu)
1871{
1872 struct timer_base *base;
1873 int b;
1874
1875 for (b = 0; b < NR_BASES; b++) {
1876 base = per_cpu_ptr(&timer_bases[b], cpu);
1877 base->clk = jiffies;
1878 base->next_expiry = base->clk + NEXT_TIMER_MAX_DELTA;
1879 base->is_idle = false;
1880 base->must_forward_clk = true;
1881 }
1882 return 0;
1883}
1884
1885int timers_dead_cpu(unsigned int cpu)
1886{
1887 struct timer_base *old_base;
1888 struct timer_base *new_base;
1889 int b, i;
1890
1891 BUG_ON(cpu_online(cpu));
1892
1893 for (b = 0; b < NR_BASES; b++) {
1894 old_base = per_cpu_ptr(&timer_bases[b], cpu);
1895 new_base = get_cpu_ptr(&timer_bases[b]);
1896 /*
1897 * The caller is globally serialized and nobody else
1898 * takes two locks at once, deadlock is not possible.
1899 */
1900 raw_spin_lock_irq(&new_base->lock);
1901 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1902
1903 /*
1904 * The current CPUs base clock might be stale. Update it
1905 * before moving the timers over.
1906 */
1907 forward_timer_base(new_base);
1908
1909 BUG_ON(old_base->running_timer);
1910
1911 for (i = 0; i < WHEEL_SIZE; i++)
1912 migrate_timer_list(new_base, old_base->vectors + i);
1913
1914 raw_spin_unlock(&old_base->lock);
1915 raw_spin_unlock_irq(&new_base->lock);
1916 put_cpu_ptr(&timer_bases);
1917 }
1918 return 0;
1919}
1920
1921#endif /* CONFIG_HOTPLUG_CPU */
1922
1923static void __init init_timer_cpu(int cpu)
1924{
1925 struct timer_base *base;
1926 int i;
1927
1928 for (i = 0; i < NR_BASES; i++) {
1929 base = per_cpu_ptr(&timer_bases[i], cpu);
1930 base->cpu = cpu;
1931 raw_spin_lock_init(&base->lock);
1932 base->clk = jiffies;
1933 }
1934}
1935
1936static void __init init_timer_cpus(void)
1937{
1938 int cpu;
1939
1940 for_each_possible_cpu(cpu)
1941 init_timer_cpu(cpu);
1942}
1943
1944void __init init_timers(void)
1945{
1946 init_timer_cpus();
1947 open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
1948}
1949
1950/**
1951 * msleep - sleep safely even with waitqueue interruptions
1952 * @msecs: Time in milliseconds to sleep for
1953 */
1954void msleep(unsigned int msecs)
1955{
1956 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1957
1958 while (timeout)
1959 timeout = schedule_timeout_uninterruptible(timeout);
1960}
1961
1962EXPORT_SYMBOL(msleep);
1963
1964/**
1965 * msleep_interruptible - sleep waiting for signals
1966 * @msecs: Time in milliseconds to sleep for
1967 */
1968unsigned long msleep_interruptible(unsigned int msecs)
1969{
1970 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1971
1972 while (timeout && !signal_pending(current))
1973 timeout = schedule_timeout_interruptible(timeout);
1974 return jiffies_to_msecs(timeout);
1975}
1976
1977EXPORT_SYMBOL(msleep_interruptible);
1978
1979/**
1980 * usleep_range - Sleep for an approximate time
1981 * @min: Minimum time in usecs to sleep
1982 * @max: Maximum time in usecs to sleep
1983 *
1984 * In non-atomic context where the exact wakeup time is flexible, use
1985 * usleep_range() instead of udelay(). The sleep improves responsiveness
1986 * by avoiding the CPU-hogging busy-wait of udelay(), and the range reduces
1987 * power usage by allowing hrtimers to take advantage of an already-
1988 * scheduled interrupt instead of scheduling a new one just for this sleep.
1989 */
1990void __sched usleep_range(unsigned long min, unsigned long max)
1991{
1992 ktime_t exp = ktime_add_us(ktime_get(), min);
1993 u64 delta = (u64)(max - min) * NSEC_PER_USEC;
1994
1995 for (;;) {
1996 __set_current_state(TASK_UNINTERRUPTIBLE);
1997 /* Do not return before the requested sleep time has elapsed */
1998 if (!schedule_hrtimeout_range(&exp, delta, HRTIMER_MODE_ABS))
1999 break;
2000 }
2001}
2002EXPORT_SYMBOL(usleep_range);