Loading...
1/*
2 * linux/kernel/timer.c
3 *
4 * Kernel internal timers
5 *
6 * Copyright (C) 1991, 1992 Linus Torvalds
7 *
8 * 1997-01-28 Modified by Finn Arne Gangstad to make timers scale better.
9 *
10 * 1997-09-10 Updated NTP code according to technical memorandum Jan '96
11 * "A Kernel Model for Precision Timekeeping" by Dave Mills
12 * 1998-12-24 Fixed a xtime SMP race (we need the xtime_lock rw spinlock to
13 * serialize accesses to xtime/lost_ticks).
14 * Copyright (C) 1998 Andrea Arcangeli
15 * 1999-03-10 Improved NTP compatibility by Ulrich Windl
16 * 2002-05-31 Move sys_sysinfo here and make its locking sane, Robert Love
17 * 2000-10-05 Implemented scalable SMP per-CPU timer handling.
18 * Copyright (C) 2000, 2001, 2002 Ingo Molnar
19 * Designed by David S. Miller, Alexey Kuznetsov and Ingo Molnar
20 */
21
22#include <linux/kernel_stat.h>
23#include <linux/export.h>
24#include <linux/interrupt.h>
25#include <linux/percpu.h>
26#include <linux/init.h>
27#include <linux/mm.h>
28#include <linux/swap.h>
29#include <linux/pid_namespace.h>
30#include <linux/notifier.h>
31#include <linux/thread_info.h>
32#include <linux/time.h>
33#include <linux/jiffies.h>
34#include <linux/posix-timers.h>
35#include <linux/cpu.h>
36#include <linux/syscalls.h>
37#include <linux/delay.h>
38#include <linux/tick.h>
39#include <linux/kallsyms.h>
40#include <linux/irq_work.h>
41#include <linux/sched.h>
42#include <linux/sched/sysctl.h>
43#include <linux/slab.h>
44#include <linux/compat.h>
45
46#include <asm/uaccess.h>
47#include <asm/unistd.h>
48#include <asm/div64.h>
49#include <asm/timex.h>
50#include <asm/io.h>
51
52#include "tick-internal.h"
53
54#define CREATE_TRACE_POINTS
55#include <trace/events/timer.h>
56
57__visible u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
58
59EXPORT_SYMBOL(jiffies_64);
60
61/*
62 * per-CPU timer vector definitions:
63 */
64#define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6)
65#define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8)
66#define TVN_SIZE (1 << TVN_BITS)
67#define TVR_SIZE (1 << TVR_BITS)
68#define TVN_MASK (TVN_SIZE - 1)
69#define TVR_MASK (TVR_SIZE - 1)
70#define MAX_TVAL ((unsigned long)((1ULL << (TVR_BITS + 4*TVN_BITS)) - 1))
71
72struct tvec {
73 struct hlist_head vec[TVN_SIZE];
74};
75
76struct tvec_root {
77 struct hlist_head vec[TVR_SIZE];
78};
79
80struct tvec_base {
81 spinlock_t lock;
82 struct timer_list *running_timer;
83 unsigned long timer_jiffies;
84 unsigned long next_timer;
85 unsigned long active_timers;
86 unsigned long all_timers;
87 int cpu;
88 bool migration_enabled;
89 bool nohz_active;
90 struct tvec_root tv1;
91 struct tvec tv2;
92 struct tvec tv3;
93 struct tvec tv4;
94 struct tvec tv5;
95} ____cacheline_aligned;
96
97
98static DEFINE_PER_CPU(struct tvec_base, tvec_bases);
99
100#if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
101unsigned int sysctl_timer_migration = 1;
102
103void timers_update_migration(bool update_nohz)
104{
105 bool on = sysctl_timer_migration && tick_nohz_active;
106 unsigned int cpu;
107
108 /* Avoid the loop, if nothing to update */
109 if (this_cpu_read(tvec_bases.migration_enabled) == on)
110 return;
111
112 for_each_possible_cpu(cpu) {
113 per_cpu(tvec_bases.migration_enabled, cpu) = on;
114 per_cpu(hrtimer_bases.migration_enabled, cpu) = on;
115 if (!update_nohz)
116 continue;
117 per_cpu(tvec_bases.nohz_active, cpu) = true;
118 per_cpu(hrtimer_bases.nohz_active, cpu) = true;
119 }
120}
121
122int timer_migration_handler(struct ctl_table *table, int write,
123 void __user *buffer, size_t *lenp,
124 loff_t *ppos)
125{
126 static DEFINE_MUTEX(mutex);
127 int ret;
128
129 mutex_lock(&mutex);
130 ret = proc_dointvec(table, write, buffer, lenp, ppos);
131 if (!ret && write)
132 timers_update_migration(false);
133 mutex_unlock(&mutex);
134 return ret;
135}
136
137static inline struct tvec_base *get_target_base(struct tvec_base *base,
138 int pinned)
139{
140 if (pinned || !base->migration_enabled)
141 return this_cpu_ptr(&tvec_bases);
142 return per_cpu_ptr(&tvec_bases, get_nohz_timer_target());
143}
144#else
145static inline struct tvec_base *get_target_base(struct tvec_base *base,
146 int pinned)
147{
148 return this_cpu_ptr(&tvec_bases);
149}
150#endif
151
152static unsigned long round_jiffies_common(unsigned long j, int cpu,
153 bool force_up)
154{
155 int rem;
156 unsigned long original = j;
157
158 /*
159 * We don't want all cpus firing their timers at once hitting the
160 * same lock or cachelines, so we skew each extra cpu with an extra
161 * 3 jiffies. This 3 jiffies came originally from the mm/ code which
162 * already did this.
163 * The skew is done by adding 3*cpunr, then round, then subtract this
164 * extra offset again.
165 */
166 j += cpu * 3;
167
168 rem = j % HZ;
169
170 /*
171 * If the target jiffie is just after a whole second (which can happen
172 * due to delays of the timer irq, long irq off times etc etc) then
173 * we should round down to the whole second, not up. Use 1/4th second
174 * as cutoff for this rounding as an extreme upper bound for this.
175 * But never round down if @force_up is set.
176 */
177 if (rem < HZ/4 && !force_up) /* round down */
178 j = j - rem;
179 else /* round up */
180 j = j - rem + HZ;
181
182 /* now that we have rounded, subtract the extra skew again */
183 j -= cpu * 3;
184
185 /*
186 * Make sure j is still in the future. Otherwise return the
187 * unmodified value.
188 */
189 return time_is_after_jiffies(j) ? j : original;
190}
191
192/**
193 * __round_jiffies - function to round jiffies to a full second
194 * @j: the time in (absolute) jiffies that should be rounded
195 * @cpu: the processor number on which the timeout will happen
196 *
197 * __round_jiffies() rounds an absolute time in the future (in jiffies)
198 * up or down to (approximately) full seconds. This is useful for timers
199 * for which the exact time they fire does not matter too much, as long as
200 * they fire approximately every X seconds.
201 *
202 * By rounding these timers to whole seconds, all such timers will fire
203 * at the same time, rather than at various times spread out. The goal
204 * of this is to have the CPU wake up less, which saves power.
205 *
206 * The exact rounding is skewed for each processor to avoid all
207 * processors firing at the exact same time, which could lead
208 * to lock contention or spurious cache line bouncing.
209 *
210 * The return value is the rounded version of the @j parameter.
211 */
212unsigned long __round_jiffies(unsigned long j, int cpu)
213{
214 return round_jiffies_common(j, cpu, false);
215}
216EXPORT_SYMBOL_GPL(__round_jiffies);
217
218/**
219 * __round_jiffies_relative - function to round jiffies to a full second
220 * @j: the time in (relative) jiffies that should be rounded
221 * @cpu: the processor number on which the timeout will happen
222 *
223 * __round_jiffies_relative() rounds a time delta in the future (in jiffies)
224 * up or down to (approximately) full seconds. This is useful for timers
225 * for which the exact time they fire does not matter too much, as long as
226 * they fire approximately every X seconds.
227 *
228 * By rounding these timers to whole seconds, all such timers will fire
229 * at the same time, rather than at various times spread out. The goal
230 * of this is to have the CPU wake up less, which saves power.
231 *
232 * The exact rounding is skewed for each processor to avoid all
233 * processors firing at the exact same time, which could lead
234 * to lock contention or spurious cache line bouncing.
235 *
236 * The return value is the rounded version of the @j parameter.
237 */
238unsigned long __round_jiffies_relative(unsigned long j, int cpu)
239{
240 unsigned long j0 = jiffies;
241
242 /* Use j0 because jiffies might change while we run */
243 return round_jiffies_common(j + j0, cpu, false) - j0;
244}
245EXPORT_SYMBOL_GPL(__round_jiffies_relative);
246
247/**
248 * round_jiffies - function to round jiffies to a full second
249 * @j: the time in (absolute) jiffies that should be rounded
250 *
251 * round_jiffies() rounds an absolute time in the future (in jiffies)
252 * up or down to (approximately) full seconds. This is useful for timers
253 * for which the exact time they fire does not matter too much, as long as
254 * they fire approximately every X seconds.
255 *
256 * By rounding these timers to whole seconds, all such timers will fire
257 * at the same time, rather than at various times spread out. The goal
258 * of this is to have the CPU wake up less, which saves power.
259 *
260 * The return value is the rounded version of the @j parameter.
261 */
262unsigned long round_jiffies(unsigned long j)
263{
264 return round_jiffies_common(j, raw_smp_processor_id(), false);
265}
266EXPORT_SYMBOL_GPL(round_jiffies);
267
268/**
269 * round_jiffies_relative - function to round jiffies to a full second
270 * @j: the time in (relative) jiffies that should be rounded
271 *
272 * round_jiffies_relative() rounds a time delta in the future (in jiffies)
273 * up or down to (approximately) full seconds. This is useful for timers
274 * for which the exact time they fire does not matter too much, as long as
275 * they fire approximately every X seconds.
276 *
277 * By rounding these timers to whole seconds, all such timers will fire
278 * at the same time, rather than at various times spread out. The goal
279 * of this is to have the CPU wake up less, which saves power.
280 *
281 * The return value is the rounded version of the @j parameter.
282 */
283unsigned long round_jiffies_relative(unsigned long j)
284{
285 return __round_jiffies_relative(j, raw_smp_processor_id());
286}
287EXPORT_SYMBOL_GPL(round_jiffies_relative);
288
289/**
290 * __round_jiffies_up - function to round jiffies up to a full second
291 * @j: the time in (absolute) jiffies that should be rounded
292 * @cpu: the processor number on which the timeout will happen
293 *
294 * This is the same as __round_jiffies() except that it will never
295 * round down. This is useful for timeouts for which the exact time
296 * of firing does not matter too much, as long as they don't fire too
297 * early.
298 */
299unsigned long __round_jiffies_up(unsigned long j, int cpu)
300{
301 return round_jiffies_common(j, cpu, true);
302}
303EXPORT_SYMBOL_GPL(__round_jiffies_up);
304
305/**
306 * __round_jiffies_up_relative - function to round jiffies up to a full second
307 * @j: the time in (relative) jiffies that should be rounded
308 * @cpu: the processor number on which the timeout will happen
309 *
310 * This is the same as __round_jiffies_relative() except that it will never
311 * round down. This is useful for timeouts for which the exact time
312 * of firing does not matter too much, as long as they don't fire too
313 * early.
314 */
315unsigned long __round_jiffies_up_relative(unsigned long j, int cpu)
316{
317 unsigned long j0 = jiffies;
318
319 /* Use j0 because jiffies might change while we run */
320 return round_jiffies_common(j + j0, cpu, true) - j0;
321}
322EXPORT_SYMBOL_GPL(__round_jiffies_up_relative);
323
324/**
325 * round_jiffies_up - function to round jiffies up to a full second
326 * @j: the time in (absolute) jiffies that should be rounded
327 *
328 * This is the same as round_jiffies() except that it will never
329 * round down. This is useful for timeouts for which the exact time
330 * of firing does not matter too much, as long as they don't fire too
331 * early.
332 */
333unsigned long round_jiffies_up(unsigned long j)
334{
335 return round_jiffies_common(j, raw_smp_processor_id(), true);
336}
337EXPORT_SYMBOL_GPL(round_jiffies_up);
338
339/**
340 * round_jiffies_up_relative - function to round jiffies up to a full second
341 * @j: the time in (relative) jiffies that should be rounded
342 *
343 * This is the same as round_jiffies_relative() except that it will never
344 * round down. This is useful for timeouts for which the exact time
345 * of firing does not matter too much, as long as they don't fire too
346 * early.
347 */
348unsigned long round_jiffies_up_relative(unsigned long j)
349{
350 return __round_jiffies_up_relative(j, raw_smp_processor_id());
351}
352EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
353
354/**
355 * set_timer_slack - set the allowed slack for a timer
356 * @timer: the timer to be modified
357 * @slack_hz: the amount of time (in jiffies) allowed for rounding
358 *
359 * Set the amount of time, in jiffies, that a certain timer has
360 * in terms of slack. By setting this value, the timer subsystem
361 * will schedule the actual timer somewhere between
362 * the time mod_timer() asks for, and that time plus the slack.
363 *
364 * By setting the slack to -1, a percentage of the delay is used
365 * instead.
366 */
367void set_timer_slack(struct timer_list *timer, int slack_hz)
368{
369 timer->slack = slack_hz;
370}
371EXPORT_SYMBOL_GPL(set_timer_slack);
372
373static void
374__internal_add_timer(struct tvec_base *base, struct timer_list *timer)
375{
376 unsigned long expires = timer->expires;
377 unsigned long idx = expires - base->timer_jiffies;
378 struct hlist_head *vec;
379
380 if (idx < TVR_SIZE) {
381 int i = expires & TVR_MASK;
382 vec = base->tv1.vec + i;
383 } else if (idx < 1 << (TVR_BITS + TVN_BITS)) {
384 int i = (expires >> TVR_BITS) & TVN_MASK;
385 vec = base->tv2.vec + i;
386 } else if (idx < 1 << (TVR_BITS + 2 * TVN_BITS)) {
387 int i = (expires >> (TVR_BITS + TVN_BITS)) & TVN_MASK;
388 vec = base->tv3.vec + i;
389 } else if (idx < 1 << (TVR_BITS + 3 * TVN_BITS)) {
390 int i = (expires >> (TVR_BITS + 2 * TVN_BITS)) & TVN_MASK;
391 vec = base->tv4.vec + i;
392 } else if ((signed long) idx < 0) {
393 /*
394 * Can happen if you add a timer with expires == jiffies,
395 * or you set a timer to go off in the past
396 */
397 vec = base->tv1.vec + (base->timer_jiffies & TVR_MASK);
398 } else {
399 int i;
400 /* If the timeout is larger than MAX_TVAL (on 64-bit
401 * architectures or with CONFIG_BASE_SMALL=1) then we
402 * use the maximum timeout.
403 */
404 if (idx > MAX_TVAL) {
405 idx = MAX_TVAL;
406 expires = idx + base->timer_jiffies;
407 }
408 i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK;
409 vec = base->tv5.vec + i;
410 }
411
412 hlist_add_head(&timer->entry, vec);
413}
414
415static void internal_add_timer(struct tvec_base *base, struct timer_list *timer)
416{
417 /* Advance base->jiffies, if the base is empty */
418 if (!base->all_timers++)
419 base->timer_jiffies = jiffies;
420
421 __internal_add_timer(base, timer);
422 /*
423 * Update base->active_timers and base->next_timer
424 */
425 if (!(timer->flags & TIMER_DEFERRABLE)) {
426 if (!base->active_timers++ ||
427 time_before(timer->expires, base->next_timer))
428 base->next_timer = timer->expires;
429 }
430
431 /*
432 * Check whether the other CPU is in dynticks mode and needs
433 * to be triggered to reevaluate the timer wheel.
434 * We are protected against the other CPU fiddling
435 * with the timer by holding the timer base lock. This also
436 * makes sure that a CPU on the way to stop its tick can not
437 * evaluate the timer wheel.
438 *
439 * Spare the IPI for deferrable timers on idle targets though.
440 * The next busy ticks will take care of it. Except full dynticks
441 * require special care against races with idle_cpu(), lets deal
442 * with that later.
443 */
444 if (base->nohz_active) {
445 if (!(timer->flags & TIMER_DEFERRABLE) ||
446 tick_nohz_full_cpu(base->cpu))
447 wake_up_nohz_cpu(base->cpu);
448 }
449}
450
451#ifdef CONFIG_TIMER_STATS
452void __timer_stats_timer_set_start_info(struct timer_list *timer, void *addr)
453{
454 if (timer->start_site)
455 return;
456
457 timer->start_site = addr;
458 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
459 timer->start_pid = current->pid;
460}
461
462static void timer_stats_account_timer(struct timer_list *timer)
463{
464 void *site;
465
466 /*
467 * start_site can be concurrently reset by
468 * timer_stats_timer_clear_start_info()
469 */
470 site = READ_ONCE(timer->start_site);
471 if (likely(!site))
472 return;
473
474 timer_stats_update_stats(timer, timer->start_pid, site,
475 timer->function, timer->start_comm,
476 timer->flags);
477}
478
479#else
480static void timer_stats_account_timer(struct timer_list *timer) {}
481#endif
482
483#ifdef CONFIG_DEBUG_OBJECTS_TIMERS
484
485static struct debug_obj_descr timer_debug_descr;
486
487static void *timer_debug_hint(void *addr)
488{
489 return ((struct timer_list *) addr)->function;
490}
491
492/*
493 * fixup_init is called when:
494 * - an active object is initialized
495 */
496static int timer_fixup_init(void *addr, enum debug_obj_state state)
497{
498 struct timer_list *timer = addr;
499
500 switch (state) {
501 case ODEBUG_STATE_ACTIVE:
502 del_timer_sync(timer);
503 debug_object_init(timer, &timer_debug_descr);
504 return 1;
505 default:
506 return 0;
507 }
508}
509
510/* Stub timer callback for improperly used timers. */
511static void stub_timer(unsigned long data)
512{
513 WARN_ON(1);
514}
515
516/*
517 * fixup_activate is called when:
518 * - an active object is activated
519 * - an unknown object is activated (might be a statically initialized object)
520 */
521static int timer_fixup_activate(void *addr, enum debug_obj_state state)
522{
523 struct timer_list *timer = addr;
524
525 switch (state) {
526
527 case ODEBUG_STATE_NOTAVAILABLE:
528 /*
529 * This is not really a fixup. The timer was
530 * statically initialized. We just make sure that it
531 * is tracked in the object tracker.
532 */
533 if (timer->entry.pprev == NULL &&
534 timer->entry.next == TIMER_ENTRY_STATIC) {
535 debug_object_init(timer, &timer_debug_descr);
536 debug_object_activate(timer, &timer_debug_descr);
537 return 0;
538 } else {
539 setup_timer(timer, stub_timer, 0);
540 return 1;
541 }
542 return 0;
543
544 case ODEBUG_STATE_ACTIVE:
545 WARN_ON(1);
546
547 default:
548 return 0;
549 }
550}
551
552/*
553 * fixup_free is called when:
554 * - an active object is freed
555 */
556static int timer_fixup_free(void *addr, enum debug_obj_state state)
557{
558 struct timer_list *timer = addr;
559
560 switch (state) {
561 case ODEBUG_STATE_ACTIVE:
562 del_timer_sync(timer);
563 debug_object_free(timer, &timer_debug_descr);
564 return 1;
565 default:
566 return 0;
567 }
568}
569
570/*
571 * fixup_assert_init is called when:
572 * - an untracked/uninit-ed object is found
573 */
574static int timer_fixup_assert_init(void *addr, enum debug_obj_state state)
575{
576 struct timer_list *timer = addr;
577
578 switch (state) {
579 case ODEBUG_STATE_NOTAVAILABLE:
580 if (timer->entry.next == TIMER_ENTRY_STATIC) {
581 /*
582 * This is not really a fixup. The timer was
583 * statically initialized. We just make sure that it
584 * is tracked in the object tracker.
585 */
586 debug_object_init(timer, &timer_debug_descr);
587 return 0;
588 } else {
589 setup_timer(timer, stub_timer, 0);
590 return 1;
591 }
592 default:
593 return 0;
594 }
595}
596
597static struct debug_obj_descr timer_debug_descr = {
598 .name = "timer_list",
599 .debug_hint = timer_debug_hint,
600 .fixup_init = timer_fixup_init,
601 .fixup_activate = timer_fixup_activate,
602 .fixup_free = timer_fixup_free,
603 .fixup_assert_init = timer_fixup_assert_init,
604};
605
606static inline void debug_timer_init(struct timer_list *timer)
607{
608 debug_object_init(timer, &timer_debug_descr);
609}
610
611static inline void debug_timer_activate(struct timer_list *timer)
612{
613 debug_object_activate(timer, &timer_debug_descr);
614}
615
616static inline void debug_timer_deactivate(struct timer_list *timer)
617{
618 debug_object_deactivate(timer, &timer_debug_descr);
619}
620
621static inline void debug_timer_free(struct timer_list *timer)
622{
623 debug_object_free(timer, &timer_debug_descr);
624}
625
626static inline void debug_timer_assert_init(struct timer_list *timer)
627{
628 debug_object_assert_init(timer, &timer_debug_descr);
629}
630
631static void do_init_timer(struct timer_list *timer, unsigned int flags,
632 const char *name, struct lock_class_key *key);
633
634void init_timer_on_stack_key(struct timer_list *timer, unsigned int flags,
635 const char *name, struct lock_class_key *key)
636{
637 debug_object_init_on_stack(timer, &timer_debug_descr);
638 do_init_timer(timer, flags, name, key);
639}
640EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
641
642void destroy_timer_on_stack(struct timer_list *timer)
643{
644 debug_object_free(timer, &timer_debug_descr);
645}
646EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
647
648#else
649static inline void debug_timer_init(struct timer_list *timer) { }
650static inline void debug_timer_activate(struct timer_list *timer) { }
651static inline void debug_timer_deactivate(struct timer_list *timer) { }
652static inline void debug_timer_assert_init(struct timer_list *timer) { }
653#endif
654
655static inline void debug_init(struct timer_list *timer)
656{
657 debug_timer_init(timer);
658 trace_timer_init(timer);
659}
660
661static inline void
662debug_activate(struct timer_list *timer, unsigned long expires)
663{
664 debug_timer_activate(timer);
665 trace_timer_start(timer, expires, timer->flags);
666}
667
668static inline void debug_deactivate(struct timer_list *timer)
669{
670 debug_timer_deactivate(timer);
671 trace_timer_cancel(timer);
672}
673
674static inline void debug_assert_init(struct timer_list *timer)
675{
676 debug_timer_assert_init(timer);
677}
678
679static void do_init_timer(struct timer_list *timer, unsigned int flags,
680 const char *name, struct lock_class_key *key)
681{
682 timer->entry.pprev = NULL;
683 timer->flags = flags | raw_smp_processor_id();
684 timer->slack = -1;
685#ifdef CONFIG_TIMER_STATS
686 timer->start_site = NULL;
687 timer->start_pid = -1;
688 memset(timer->start_comm, 0, TASK_COMM_LEN);
689#endif
690 lockdep_init_map(&timer->lockdep_map, name, key, 0);
691}
692
693/**
694 * init_timer_key - initialize a timer
695 * @timer: the timer to be initialized
696 * @flags: timer flags
697 * @name: name of the timer
698 * @key: lockdep class key of the fake lock used for tracking timer
699 * sync lock dependencies
700 *
701 * init_timer_key() must be done to a timer prior calling *any* of the
702 * other timer functions.
703 */
704void init_timer_key(struct timer_list *timer, unsigned int flags,
705 const char *name, struct lock_class_key *key)
706{
707 debug_init(timer);
708 do_init_timer(timer, flags, name, key);
709}
710EXPORT_SYMBOL(init_timer_key);
711
712static inline void detach_timer(struct timer_list *timer, bool clear_pending)
713{
714 struct hlist_node *entry = &timer->entry;
715
716 debug_deactivate(timer);
717
718 __hlist_del(entry);
719 if (clear_pending)
720 entry->pprev = NULL;
721 entry->next = LIST_POISON2;
722}
723
724static inline void
725detach_expired_timer(struct timer_list *timer, struct tvec_base *base)
726{
727 detach_timer(timer, true);
728 if (!(timer->flags & TIMER_DEFERRABLE))
729 base->active_timers--;
730 base->all_timers--;
731}
732
733static int detach_if_pending(struct timer_list *timer, struct tvec_base *base,
734 bool clear_pending)
735{
736 if (!timer_pending(timer))
737 return 0;
738
739 detach_timer(timer, clear_pending);
740 if (!(timer->flags & TIMER_DEFERRABLE)) {
741 base->active_timers--;
742 if (timer->expires == base->next_timer)
743 base->next_timer = base->timer_jiffies;
744 }
745 /* If this was the last timer, advance base->jiffies */
746 if (!--base->all_timers)
747 base->timer_jiffies = jiffies;
748 return 1;
749}
750
751/*
752 * We are using hashed locking: holding per_cpu(tvec_bases).lock
753 * means that all timers which are tied to this base via timer->base are
754 * locked, and the base itself is locked too.
755 *
756 * So __run_timers/migrate_timers can safely modify all timers which could
757 * be found on ->tvX lists.
758 *
759 * When the timer's base is locked and removed from the list, the
760 * TIMER_MIGRATING flag is set, FIXME
761 */
762static struct tvec_base *lock_timer_base(struct timer_list *timer,
763 unsigned long *flags)
764 __acquires(timer->base->lock)
765{
766 for (;;) {
767 u32 tf = timer->flags;
768 struct tvec_base *base;
769
770 if (!(tf & TIMER_MIGRATING)) {
771 base = per_cpu_ptr(&tvec_bases, tf & TIMER_CPUMASK);
772 spin_lock_irqsave(&base->lock, *flags);
773 if (timer->flags == tf)
774 return base;
775 spin_unlock_irqrestore(&base->lock, *flags);
776 }
777 cpu_relax();
778 }
779}
780
781static inline int
782__mod_timer(struct timer_list *timer, unsigned long expires,
783 bool pending_only, int pinned)
784{
785 struct tvec_base *base, *new_base;
786 unsigned long flags;
787 int ret = 0;
788
789 timer_stats_timer_set_start_info(timer);
790 BUG_ON(!timer->function);
791
792 base = lock_timer_base(timer, &flags);
793
794 ret = detach_if_pending(timer, base, false);
795 if (!ret && pending_only)
796 goto out_unlock;
797
798 debug_activate(timer, expires);
799
800 new_base = get_target_base(base, pinned);
801
802 if (base != new_base) {
803 /*
804 * We are trying to schedule the timer on the local CPU.
805 * However we can't change timer's base while it is running,
806 * otherwise del_timer_sync() can't detect that the timer's
807 * handler yet has not finished. This also guarantees that
808 * the timer is serialized wrt itself.
809 */
810 if (likely(base->running_timer != timer)) {
811 /* See the comment in lock_timer_base() */
812 timer->flags |= TIMER_MIGRATING;
813
814 spin_unlock(&base->lock);
815 base = new_base;
816 spin_lock(&base->lock);
817 WRITE_ONCE(timer->flags,
818 (timer->flags & ~TIMER_BASEMASK) | base->cpu);
819 }
820 }
821
822 timer->expires = expires;
823 internal_add_timer(base, timer);
824
825out_unlock:
826 spin_unlock_irqrestore(&base->lock, flags);
827
828 return ret;
829}
830
831/**
832 * mod_timer_pending - modify a pending timer's timeout
833 * @timer: the pending timer to be modified
834 * @expires: new timeout in jiffies
835 *
836 * mod_timer_pending() is the same for pending timers as mod_timer(),
837 * but will not re-activate and modify already deleted timers.
838 *
839 * It is useful for unserialized use of timers.
840 */
841int mod_timer_pending(struct timer_list *timer, unsigned long expires)
842{
843 return __mod_timer(timer, expires, true, TIMER_NOT_PINNED);
844}
845EXPORT_SYMBOL(mod_timer_pending);
846
847/*
848 * Decide where to put the timer while taking the slack into account
849 *
850 * Algorithm:
851 * 1) calculate the maximum (absolute) time
852 * 2) calculate the highest bit where the expires and new max are different
853 * 3) use this bit to make a mask
854 * 4) use the bitmask to round down the maximum time, so that all last
855 * bits are zeros
856 */
857static inline
858unsigned long apply_slack(struct timer_list *timer, unsigned long expires)
859{
860 unsigned long expires_limit, mask;
861 int bit;
862
863 if (timer->slack >= 0) {
864 expires_limit = expires + timer->slack;
865 } else {
866 long delta = expires - jiffies;
867
868 if (delta < 256)
869 return expires;
870
871 expires_limit = expires + delta / 256;
872 }
873 mask = expires ^ expires_limit;
874 if (mask == 0)
875 return expires;
876
877 bit = __fls(mask);
878
879 mask = (1UL << bit) - 1;
880
881 expires_limit = expires_limit & ~(mask);
882
883 return expires_limit;
884}
885
886/**
887 * mod_timer - modify a timer's timeout
888 * @timer: the timer to be modified
889 * @expires: new timeout in jiffies
890 *
891 * mod_timer() is a more efficient way to update the expire field of an
892 * active timer (if the timer is inactive it will be activated)
893 *
894 * mod_timer(timer, expires) is equivalent to:
895 *
896 * del_timer(timer); timer->expires = expires; add_timer(timer);
897 *
898 * Note that if there are multiple unserialized concurrent users of the
899 * same timer, then mod_timer() is the only safe way to modify the timeout,
900 * since add_timer() cannot modify an already running timer.
901 *
902 * The function returns whether it has modified a pending timer or not.
903 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
904 * active timer returns 1.)
905 */
906int mod_timer(struct timer_list *timer, unsigned long expires)
907{
908 expires = apply_slack(timer, expires);
909
910 /*
911 * This is a common optimization triggered by the
912 * networking code - if the timer is re-modified
913 * to be the same thing then just return:
914 */
915 if (timer_pending(timer) && timer->expires == expires)
916 return 1;
917
918 return __mod_timer(timer, expires, false, TIMER_NOT_PINNED);
919}
920EXPORT_SYMBOL(mod_timer);
921
922/**
923 * mod_timer_pinned - modify a timer's timeout
924 * @timer: the timer to be modified
925 * @expires: new timeout in jiffies
926 *
927 * mod_timer_pinned() is a way to update the expire field of an
928 * active timer (if the timer is inactive it will be activated)
929 * and to ensure that the timer is scheduled on the current CPU.
930 *
931 * Note that this does not prevent the timer from being migrated
932 * when the current CPU goes offline. If this is a problem for
933 * you, use CPU-hotplug notifiers to handle it correctly, for
934 * example, cancelling the timer when the corresponding CPU goes
935 * offline.
936 *
937 * mod_timer_pinned(timer, expires) is equivalent to:
938 *
939 * del_timer(timer); timer->expires = expires; add_timer(timer);
940 */
941int mod_timer_pinned(struct timer_list *timer, unsigned long expires)
942{
943 if (timer->expires == expires && timer_pending(timer))
944 return 1;
945
946 return __mod_timer(timer, expires, false, TIMER_PINNED);
947}
948EXPORT_SYMBOL(mod_timer_pinned);
949
950/**
951 * add_timer - start a timer
952 * @timer: the timer to be added
953 *
954 * The kernel will do a ->function(->data) callback from the
955 * timer interrupt at the ->expires point in the future. The
956 * current time is 'jiffies'.
957 *
958 * The timer's ->expires, ->function (and if the handler uses it, ->data)
959 * fields must be set prior calling this function.
960 *
961 * Timers with an ->expires field in the past will be executed in the next
962 * timer tick.
963 */
964void add_timer(struct timer_list *timer)
965{
966 BUG_ON(timer_pending(timer));
967 mod_timer(timer, timer->expires);
968}
969EXPORT_SYMBOL(add_timer);
970
971/**
972 * add_timer_on - start a timer on a particular CPU
973 * @timer: the timer to be added
974 * @cpu: the CPU to start it on
975 *
976 * This is not very scalable on SMP. Double adds are not possible.
977 */
978void add_timer_on(struct timer_list *timer, int cpu)
979{
980 struct tvec_base *new_base = per_cpu_ptr(&tvec_bases, cpu);
981 struct tvec_base *base;
982 unsigned long flags;
983
984 timer_stats_timer_set_start_info(timer);
985 BUG_ON(timer_pending(timer) || !timer->function);
986
987 /*
988 * If @timer was on a different CPU, it should be migrated with the
989 * old base locked to prevent other operations proceeding with the
990 * wrong base locked. See lock_timer_base().
991 */
992 base = lock_timer_base(timer, &flags);
993 if (base != new_base) {
994 timer->flags |= TIMER_MIGRATING;
995
996 spin_unlock(&base->lock);
997 base = new_base;
998 spin_lock(&base->lock);
999 WRITE_ONCE(timer->flags,
1000 (timer->flags & ~TIMER_BASEMASK) | cpu);
1001 }
1002
1003 debug_activate(timer, timer->expires);
1004 internal_add_timer(base, timer);
1005 spin_unlock_irqrestore(&base->lock, flags);
1006}
1007EXPORT_SYMBOL_GPL(add_timer_on);
1008
1009/**
1010 * del_timer - deactive a timer.
1011 * @timer: the timer to be deactivated
1012 *
1013 * del_timer() deactivates a timer - this works on both active and inactive
1014 * timers.
1015 *
1016 * The function returns whether it has deactivated a pending timer or not.
1017 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
1018 * active timer returns 1.)
1019 */
1020int del_timer(struct timer_list *timer)
1021{
1022 struct tvec_base *base;
1023 unsigned long flags;
1024 int ret = 0;
1025
1026 debug_assert_init(timer);
1027
1028 timer_stats_timer_clear_start_info(timer);
1029 if (timer_pending(timer)) {
1030 base = lock_timer_base(timer, &flags);
1031 ret = detach_if_pending(timer, base, true);
1032 spin_unlock_irqrestore(&base->lock, flags);
1033 }
1034
1035 return ret;
1036}
1037EXPORT_SYMBOL(del_timer);
1038
1039/**
1040 * try_to_del_timer_sync - Try to deactivate a timer
1041 * @timer: timer do del
1042 *
1043 * This function tries to deactivate a timer. Upon successful (ret >= 0)
1044 * exit the timer is not queued and the handler is not running on any CPU.
1045 */
1046int try_to_del_timer_sync(struct timer_list *timer)
1047{
1048 struct tvec_base *base;
1049 unsigned long flags;
1050 int ret = -1;
1051
1052 debug_assert_init(timer);
1053
1054 base = lock_timer_base(timer, &flags);
1055
1056 if (base->running_timer != timer) {
1057 timer_stats_timer_clear_start_info(timer);
1058 ret = detach_if_pending(timer, base, true);
1059 }
1060 spin_unlock_irqrestore(&base->lock, flags);
1061
1062 return ret;
1063}
1064EXPORT_SYMBOL(try_to_del_timer_sync);
1065
1066#ifdef CONFIG_SMP
1067/**
1068 * del_timer_sync - deactivate a timer and wait for the handler to finish.
1069 * @timer: the timer to be deactivated
1070 *
1071 * This function only differs from del_timer() on SMP: besides deactivating
1072 * the timer it also makes sure the handler has finished executing on other
1073 * CPUs.
1074 *
1075 * Synchronization rules: Callers must prevent restarting of the timer,
1076 * otherwise this function is meaningless. It must not be called from
1077 * interrupt contexts unless the timer is an irqsafe one. The caller must
1078 * not hold locks which would prevent completion of the timer's
1079 * handler. The timer's handler must not call add_timer_on(). Upon exit the
1080 * timer is not queued and the handler is not running on any CPU.
1081 *
1082 * Note: For !irqsafe timers, you must not hold locks that are held in
1083 * interrupt context while calling this function. Even if the lock has
1084 * nothing to do with the timer in question. Here's why:
1085 *
1086 * CPU0 CPU1
1087 * ---- ----
1088 * <SOFTIRQ>
1089 * call_timer_fn();
1090 * base->running_timer = mytimer;
1091 * spin_lock_irq(somelock);
1092 * <IRQ>
1093 * spin_lock(somelock);
1094 * del_timer_sync(mytimer);
1095 * while (base->running_timer == mytimer);
1096 *
1097 * Now del_timer_sync() will never return and never release somelock.
1098 * The interrupt on the other CPU is waiting to grab somelock but
1099 * it has interrupted the softirq that CPU0 is waiting to finish.
1100 *
1101 * The function returns whether it has deactivated a pending timer or not.
1102 */
1103int del_timer_sync(struct timer_list *timer)
1104{
1105#ifdef CONFIG_LOCKDEP
1106 unsigned long flags;
1107
1108 /*
1109 * If lockdep gives a backtrace here, please reference
1110 * the synchronization rules above.
1111 */
1112 local_irq_save(flags);
1113 lock_map_acquire(&timer->lockdep_map);
1114 lock_map_release(&timer->lockdep_map);
1115 local_irq_restore(flags);
1116#endif
1117 /*
1118 * don't use it in hardirq context, because it
1119 * could lead to deadlock.
1120 */
1121 WARN_ON(in_irq() && !(timer->flags & TIMER_IRQSAFE));
1122 for (;;) {
1123 int ret = try_to_del_timer_sync(timer);
1124 if (ret >= 0)
1125 return ret;
1126 cpu_relax();
1127 }
1128}
1129EXPORT_SYMBOL(del_timer_sync);
1130#endif
1131
1132static int cascade(struct tvec_base *base, struct tvec *tv, int index)
1133{
1134 /* cascade all the timers from tv up one level */
1135 struct timer_list *timer;
1136 struct hlist_node *tmp;
1137 struct hlist_head tv_list;
1138
1139 hlist_move_list(tv->vec + index, &tv_list);
1140
1141 /*
1142 * We are removing _all_ timers from the list, so we
1143 * don't have to detach them individually.
1144 */
1145 hlist_for_each_entry_safe(timer, tmp, &tv_list, entry) {
1146 /* No accounting, while moving them */
1147 __internal_add_timer(base, timer);
1148 }
1149
1150 return index;
1151}
1152
1153static void call_timer_fn(struct timer_list *timer, void (*fn)(unsigned long),
1154 unsigned long data)
1155{
1156 int count = preempt_count();
1157
1158#ifdef CONFIG_LOCKDEP
1159 /*
1160 * It is permissible to free the timer from inside the
1161 * function that is called from it, this we need to take into
1162 * account for lockdep too. To avoid bogus "held lock freed"
1163 * warnings as well as problems when looking into
1164 * timer->lockdep_map, make a copy and use that here.
1165 */
1166 struct lockdep_map lockdep_map;
1167
1168 lockdep_copy_map(&lockdep_map, &timer->lockdep_map);
1169#endif
1170 /*
1171 * Couple the lock chain with the lock chain at
1172 * del_timer_sync() by acquiring the lock_map around the fn()
1173 * call here and in del_timer_sync().
1174 */
1175 lock_map_acquire(&lockdep_map);
1176
1177 trace_timer_expire_entry(timer);
1178 fn(data);
1179 trace_timer_expire_exit(timer);
1180
1181 lock_map_release(&lockdep_map);
1182
1183 if (count != preempt_count()) {
1184 WARN_ONCE(1, "timer: %pF preempt leak: %08x -> %08x\n",
1185 fn, count, preempt_count());
1186 /*
1187 * Restore the preempt count. That gives us a decent
1188 * chance to survive and extract information. If the
1189 * callback kept a lock held, bad luck, but not worse
1190 * than the BUG() we had.
1191 */
1192 preempt_count_set(count);
1193 }
1194}
1195
1196#define INDEX(N) ((base->timer_jiffies >> (TVR_BITS + (N) * TVN_BITS)) & TVN_MASK)
1197
1198/**
1199 * __run_timers - run all expired timers (if any) on this CPU.
1200 * @base: the timer vector to be processed.
1201 *
1202 * This function cascades all vectors and executes all expired timer
1203 * vectors.
1204 */
1205static inline void __run_timers(struct tvec_base *base)
1206{
1207 struct timer_list *timer;
1208
1209 spin_lock_irq(&base->lock);
1210
1211 while (time_after_eq(jiffies, base->timer_jiffies)) {
1212 struct hlist_head work_list;
1213 struct hlist_head *head = &work_list;
1214 int index;
1215
1216 if (!base->all_timers) {
1217 base->timer_jiffies = jiffies;
1218 break;
1219 }
1220
1221 index = base->timer_jiffies & TVR_MASK;
1222
1223 /*
1224 * Cascade timers:
1225 */
1226 if (!index &&
1227 (!cascade(base, &base->tv2, INDEX(0))) &&
1228 (!cascade(base, &base->tv3, INDEX(1))) &&
1229 !cascade(base, &base->tv4, INDEX(2)))
1230 cascade(base, &base->tv5, INDEX(3));
1231 ++base->timer_jiffies;
1232 hlist_move_list(base->tv1.vec + index, head);
1233 while (!hlist_empty(head)) {
1234 void (*fn)(unsigned long);
1235 unsigned long data;
1236 bool irqsafe;
1237
1238 timer = hlist_entry(head->first, struct timer_list, entry);
1239 fn = timer->function;
1240 data = timer->data;
1241 irqsafe = timer->flags & TIMER_IRQSAFE;
1242
1243 timer_stats_account_timer(timer);
1244
1245 base->running_timer = timer;
1246 detach_expired_timer(timer, base);
1247
1248 if (irqsafe) {
1249 spin_unlock(&base->lock);
1250 call_timer_fn(timer, fn, data);
1251 spin_lock(&base->lock);
1252 } else {
1253 spin_unlock_irq(&base->lock);
1254 call_timer_fn(timer, fn, data);
1255 spin_lock_irq(&base->lock);
1256 }
1257 }
1258 }
1259 base->running_timer = NULL;
1260 spin_unlock_irq(&base->lock);
1261}
1262
1263#ifdef CONFIG_NO_HZ_COMMON
1264/*
1265 * Find out when the next timer event is due to happen. This
1266 * is used on S/390 to stop all activity when a CPU is idle.
1267 * This function needs to be called with interrupts disabled.
1268 */
1269static unsigned long __next_timer_interrupt(struct tvec_base *base)
1270{
1271 unsigned long timer_jiffies = base->timer_jiffies;
1272 unsigned long expires = timer_jiffies + NEXT_TIMER_MAX_DELTA;
1273 int index, slot, array, found = 0;
1274 struct timer_list *nte;
1275 struct tvec *varray[4];
1276
1277 /* Look for timer events in tv1. */
1278 index = slot = timer_jiffies & TVR_MASK;
1279 do {
1280 hlist_for_each_entry(nte, base->tv1.vec + slot, entry) {
1281 if (nte->flags & TIMER_DEFERRABLE)
1282 continue;
1283
1284 found = 1;
1285 expires = nte->expires;
1286 /* Look at the cascade bucket(s)? */
1287 if (!index || slot < index)
1288 goto cascade;
1289 return expires;
1290 }
1291 slot = (slot + 1) & TVR_MASK;
1292 } while (slot != index);
1293
1294cascade:
1295 /* Calculate the next cascade event */
1296 if (index)
1297 timer_jiffies += TVR_SIZE - index;
1298 timer_jiffies >>= TVR_BITS;
1299
1300 /* Check tv2-tv5. */
1301 varray[0] = &base->tv2;
1302 varray[1] = &base->tv3;
1303 varray[2] = &base->tv4;
1304 varray[3] = &base->tv5;
1305
1306 for (array = 0; array < 4; array++) {
1307 struct tvec *varp = varray[array];
1308
1309 index = slot = timer_jiffies & TVN_MASK;
1310 do {
1311 hlist_for_each_entry(nte, varp->vec + slot, entry) {
1312 if (nte->flags & TIMER_DEFERRABLE)
1313 continue;
1314
1315 found = 1;
1316 if (time_before(nte->expires, expires))
1317 expires = nte->expires;
1318 }
1319 /*
1320 * Do we still search for the first timer or are
1321 * we looking up the cascade buckets ?
1322 */
1323 if (found) {
1324 /* Look at the cascade bucket(s)? */
1325 if (!index || slot < index)
1326 break;
1327 return expires;
1328 }
1329 slot = (slot + 1) & TVN_MASK;
1330 } while (slot != index);
1331
1332 if (index)
1333 timer_jiffies += TVN_SIZE - index;
1334 timer_jiffies >>= TVN_BITS;
1335 }
1336 return expires;
1337}
1338
1339/*
1340 * Check, if the next hrtimer event is before the next timer wheel
1341 * event:
1342 */
1343static u64 cmp_next_hrtimer_event(u64 basem, u64 expires)
1344{
1345 u64 nextevt = hrtimer_get_next_event();
1346
1347 /*
1348 * If high resolution timers are enabled
1349 * hrtimer_get_next_event() returns KTIME_MAX.
1350 */
1351 if (expires <= nextevt)
1352 return expires;
1353
1354 /*
1355 * If the next timer is already expired, return the tick base
1356 * time so the tick is fired immediately.
1357 */
1358 if (nextevt <= basem)
1359 return basem;
1360
1361 /*
1362 * Round up to the next jiffie. High resolution timers are
1363 * off, so the hrtimers are expired in the tick and we need to
1364 * make sure that this tick really expires the timer to avoid
1365 * a ping pong of the nohz stop code.
1366 *
1367 * Use DIV_ROUND_UP_ULL to prevent gcc calling __divdi3
1368 */
1369 return DIV_ROUND_UP_ULL(nextevt, TICK_NSEC) * TICK_NSEC;
1370}
1371
1372/**
1373 * get_next_timer_interrupt - return the time (clock mono) of the next timer
1374 * @basej: base time jiffies
1375 * @basem: base time clock monotonic
1376 *
1377 * Returns the tick aligned clock monotonic time of the next pending
1378 * timer or KTIME_MAX if no timer is pending.
1379 */
1380u64 get_next_timer_interrupt(unsigned long basej, u64 basem)
1381{
1382 struct tvec_base *base = this_cpu_ptr(&tvec_bases);
1383 u64 expires = KTIME_MAX;
1384 unsigned long nextevt;
1385
1386 /*
1387 * Pretend that there is no timer pending if the cpu is offline.
1388 * Possible pending timers will be migrated later to an active cpu.
1389 */
1390 if (cpu_is_offline(smp_processor_id()))
1391 return expires;
1392
1393 spin_lock(&base->lock);
1394 if (base->active_timers) {
1395 if (time_before_eq(base->next_timer, base->timer_jiffies))
1396 base->next_timer = __next_timer_interrupt(base);
1397 nextevt = base->next_timer;
1398 if (time_before_eq(nextevt, basej))
1399 expires = basem;
1400 else
1401 expires = basem + (nextevt - basej) * TICK_NSEC;
1402 }
1403 spin_unlock(&base->lock);
1404
1405 return cmp_next_hrtimer_event(basem, expires);
1406}
1407#endif
1408
1409/*
1410 * Called from the timer interrupt handler to charge one tick to the current
1411 * process. user_tick is 1 if the tick is user time, 0 for system.
1412 */
1413void update_process_times(int user_tick)
1414{
1415 struct task_struct *p = current;
1416
1417 /* Note: this timer irq context must be accounted for as well. */
1418 account_process_tick(p, user_tick);
1419 run_local_timers();
1420 rcu_check_callbacks(user_tick);
1421#ifdef CONFIG_IRQ_WORK
1422 if (in_irq())
1423 irq_work_tick();
1424#endif
1425 scheduler_tick();
1426 run_posix_cpu_timers(p);
1427}
1428
1429/*
1430 * This function runs timers and the timer-tq in bottom half context.
1431 */
1432static void run_timer_softirq(struct softirq_action *h)
1433{
1434 struct tvec_base *base = this_cpu_ptr(&tvec_bases);
1435
1436 if (time_after_eq(jiffies, base->timer_jiffies))
1437 __run_timers(base);
1438}
1439
1440/*
1441 * Called by the local, per-CPU timer interrupt on SMP.
1442 */
1443void run_local_timers(void)
1444{
1445 hrtimer_run_queues();
1446 raise_softirq(TIMER_SOFTIRQ);
1447}
1448
1449#ifdef __ARCH_WANT_SYS_ALARM
1450
1451/*
1452 * For backwards compatibility? This can be done in libc so Alpha
1453 * and all newer ports shouldn't need it.
1454 */
1455SYSCALL_DEFINE1(alarm, unsigned int, seconds)
1456{
1457 return alarm_setitimer(seconds);
1458}
1459
1460#endif
1461
1462static void process_timeout(unsigned long __data)
1463{
1464 wake_up_process((struct task_struct *)__data);
1465}
1466
1467/**
1468 * schedule_timeout - sleep until timeout
1469 * @timeout: timeout value in jiffies
1470 *
1471 * Make the current task sleep until @timeout jiffies have
1472 * elapsed. The routine will return immediately unless
1473 * the current task state has been set (see set_current_state()).
1474 *
1475 * You can set the task state as follows -
1476 *
1477 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1478 * pass before the routine returns. The routine will return 0
1479 *
1480 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1481 * delivered to the current task. In this case the remaining time
1482 * in jiffies will be returned, or 0 if the timer expired in time
1483 *
1484 * The current task state is guaranteed to be TASK_RUNNING when this
1485 * routine returns.
1486 *
1487 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1488 * the CPU away without a bound on the timeout. In this case the return
1489 * value will be %MAX_SCHEDULE_TIMEOUT.
1490 *
1491 * In all cases the return value is guaranteed to be non-negative.
1492 */
1493signed long __sched schedule_timeout(signed long timeout)
1494{
1495 struct timer_list timer;
1496 unsigned long expire;
1497
1498 switch (timeout)
1499 {
1500 case MAX_SCHEDULE_TIMEOUT:
1501 /*
1502 * These two special cases are useful to be comfortable
1503 * in the caller. Nothing more. We could take
1504 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1505 * but I' d like to return a valid offset (>=0) to allow
1506 * the caller to do everything it want with the retval.
1507 */
1508 schedule();
1509 goto out;
1510 default:
1511 /*
1512 * Another bit of PARANOID. Note that the retval will be
1513 * 0 since no piece of kernel is supposed to do a check
1514 * for a negative retval of schedule_timeout() (since it
1515 * should never happens anyway). You just have the printk()
1516 * that will tell you if something is gone wrong and where.
1517 */
1518 if (timeout < 0) {
1519 printk(KERN_ERR "schedule_timeout: wrong timeout "
1520 "value %lx\n", timeout);
1521 dump_stack();
1522 current->state = TASK_RUNNING;
1523 goto out;
1524 }
1525 }
1526
1527 expire = timeout + jiffies;
1528
1529 setup_timer_on_stack(&timer, process_timeout, (unsigned long)current);
1530 __mod_timer(&timer, expire, false, TIMER_NOT_PINNED);
1531 schedule();
1532 del_singleshot_timer_sync(&timer);
1533
1534 /* Remove the timer from the object tracker */
1535 destroy_timer_on_stack(&timer);
1536
1537 timeout = expire - jiffies;
1538
1539 out:
1540 return timeout < 0 ? 0 : timeout;
1541}
1542EXPORT_SYMBOL(schedule_timeout);
1543
1544/*
1545 * We can use __set_current_state() here because schedule_timeout() calls
1546 * schedule() unconditionally.
1547 */
1548signed long __sched schedule_timeout_interruptible(signed long timeout)
1549{
1550 __set_current_state(TASK_INTERRUPTIBLE);
1551 return schedule_timeout(timeout);
1552}
1553EXPORT_SYMBOL(schedule_timeout_interruptible);
1554
1555signed long __sched schedule_timeout_killable(signed long timeout)
1556{
1557 __set_current_state(TASK_KILLABLE);
1558 return schedule_timeout(timeout);
1559}
1560EXPORT_SYMBOL(schedule_timeout_killable);
1561
1562signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1563{
1564 __set_current_state(TASK_UNINTERRUPTIBLE);
1565 return schedule_timeout(timeout);
1566}
1567EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1568
1569/*
1570 * Like schedule_timeout_uninterruptible(), except this task will not contribute
1571 * to load average.
1572 */
1573signed long __sched schedule_timeout_idle(signed long timeout)
1574{
1575 __set_current_state(TASK_IDLE);
1576 return schedule_timeout(timeout);
1577}
1578EXPORT_SYMBOL(schedule_timeout_idle);
1579
1580#ifdef CONFIG_HOTPLUG_CPU
1581static void migrate_timer_list(struct tvec_base *new_base, struct hlist_head *head)
1582{
1583 struct timer_list *timer;
1584 int cpu = new_base->cpu;
1585
1586 while (!hlist_empty(head)) {
1587 timer = hlist_entry(head->first, struct timer_list, entry);
1588 /* We ignore the accounting on the dying cpu */
1589 detach_timer(timer, false);
1590 timer->flags = (timer->flags & ~TIMER_BASEMASK) | cpu;
1591 internal_add_timer(new_base, timer);
1592 }
1593}
1594
1595static void migrate_timers(int cpu)
1596{
1597 struct tvec_base *old_base;
1598 struct tvec_base *new_base;
1599 int i;
1600
1601 BUG_ON(cpu_online(cpu));
1602 old_base = per_cpu_ptr(&tvec_bases, cpu);
1603 new_base = get_cpu_ptr(&tvec_bases);
1604 /*
1605 * The caller is globally serialized and nobody else
1606 * takes two locks at once, deadlock is not possible.
1607 */
1608 spin_lock_irq(&new_base->lock);
1609 spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1610
1611 BUG_ON(old_base->running_timer);
1612
1613 for (i = 0; i < TVR_SIZE; i++)
1614 migrate_timer_list(new_base, old_base->tv1.vec + i);
1615 for (i = 0; i < TVN_SIZE; i++) {
1616 migrate_timer_list(new_base, old_base->tv2.vec + i);
1617 migrate_timer_list(new_base, old_base->tv3.vec + i);
1618 migrate_timer_list(new_base, old_base->tv4.vec + i);
1619 migrate_timer_list(new_base, old_base->tv5.vec + i);
1620 }
1621
1622 old_base->active_timers = 0;
1623 old_base->all_timers = 0;
1624
1625 spin_unlock(&old_base->lock);
1626 spin_unlock_irq(&new_base->lock);
1627 put_cpu_ptr(&tvec_bases);
1628}
1629
1630static int timer_cpu_notify(struct notifier_block *self,
1631 unsigned long action, void *hcpu)
1632{
1633 switch (action) {
1634 case CPU_DEAD:
1635 case CPU_DEAD_FROZEN:
1636 migrate_timers((long)hcpu);
1637 break;
1638 default:
1639 break;
1640 }
1641
1642 return NOTIFY_OK;
1643}
1644
1645static inline void timer_register_cpu_notifier(void)
1646{
1647 cpu_notifier(timer_cpu_notify, 0);
1648}
1649#else
1650static inline void timer_register_cpu_notifier(void) { }
1651#endif /* CONFIG_HOTPLUG_CPU */
1652
1653static void __init init_timer_cpu(int cpu)
1654{
1655 struct tvec_base *base = per_cpu_ptr(&tvec_bases, cpu);
1656
1657 base->cpu = cpu;
1658 spin_lock_init(&base->lock);
1659
1660 base->timer_jiffies = jiffies;
1661 base->next_timer = base->timer_jiffies;
1662}
1663
1664static void __init init_timer_cpus(void)
1665{
1666 int cpu;
1667
1668 for_each_possible_cpu(cpu)
1669 init_timer_cpu(cpu);
1670}
1671
1672void __init init_timers(void)
1673{
1674 init_timer_cpus();
1675 init_timer_stats();
1676 timer_register_cpu_notifier();
1677 open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
1678}
1679
1680/**
1681 * msleep - sleep safely even with waitqueue interruptions
1682 * @msecs: Time in milliseconds to sleep for
1683 */
1684void msleep(unsigned int msecs)
1685{
1686 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1687
1688 while (timeout)
1689 timeout = schedule_timeout_uninterruptible(timeout);
1690}
1691
1692EXPORT_SYMBOL(msleep);
1693
1694/**
1695 * msleep_interruptible - sleep waiting for signals
1696 * @msecs: Time in milliseconds to sleep for
1697 */
1698unsigned long msleep_interruptible(unsigned int msecs)
1699{
1700 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1701
1702 while (timeout && !signal_pending(current))
1703 timeout = schedule_timeout_interruptible(timeout);
1704 return jiffies_to_msecs(timeout);
1705}
1706
1707EXPORT_SYMBOL(msleep_interruptible);
1708
1709static void __sched do_usleep_range(unsigned long min, unsigned long max)
1710{
1711 ktime_t kmin;
1712 u64 delta;
1713
1714 kmin = ktime_set(0, min * NSEC_PER_USEC);
1715 delta = (u64)(max - min) * NSEC_PER_USEC;
1716 schedule_hrtimeout_range(&kmin, delta, HRTIMER_MODE_REL);
1717}
1718
1719/**
1720 * usleep_range - Drop in replacement for udelay where wakeup is flexible
1721 * @min: Minimum time in usecs to sleep
1722 * @max: Maximum time in usecs to sleep
1723 */
1724void __sched usleep_range(unsigned long min, unsigned long max)
1725{
1726 __set_current_state(TASK_UNINTERRUPTIBLE);
1727 do_usleep_range(min, max);
1728}
1729EXPORT_SYMBOL(usleep_range);
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Kernel internal timers
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 *
7 * 1997-01-28 Modified by Finn Arne Gangstad to make timers scale better.
8 *
9 * 1997-09-10 Updated NTP code according to technical memorandum Jan '96
10 * "A Kernel Model for Precision Timekeeping" by Dave Mills
11 * 1998-12-24 Fixed a xtime SMP race (we need the xtime_lock rw spinlock to
12 * serialize accesses to xtime/lost_ticks).
13 * Copyright (C) 1998 Andrea Arcangeli
14 * 1999-03-10 Improved NTP compatibility by Ulrich Windl
15 * 2002-05-31 Move sys_sysinfo here and make its locking sane, Robert Love
16 * 2000-10-05 Implemented scalable SMP per-CPU timer handling.
17 * Copyright (C) 2000, 2001, 2002 Ingo Molnar
18 * Designed by David S. Miller, Alexey Kuznetsov and Ingo Molnar
19 */
20
21#include <linux/kernel_stat.h>
22#include <linux/export.h>
23#include <linux/interrupt.h>
24#include <linux/percpu.h>
25#include <linux/init.h>
26#include <linux/mm.h>
27#include <linux/swap.h>
28#include <linux/pid_namespace.h>
29#include <linux/notifier.h>
30#include <linux/thread_info.h>
31#include <linux/time.h>
32#include <linux/jiffies.h>
33#include <linux/posix-timers.h>
34#include <linux/cpu.h>
35#include <linux/syscalls.h>
36#include <linux/delay.h>
37#include <linux/tick.h>
38#include <linux/kallsyms.h>
39#include <linux/irq_work.h>
40#include <linux/sched/signal.h>
41#include <linux/sched/sysctl.h>
42#include <linux/sched/nohz.h>
43#include <linux/sched/debug.h>
44#include <linux/slab.h>
45#include <linux/compat.h>
46#include <linux/random.h>
47
48#include <linux/uaccess.h>
49#include <asm/unistd.h>
50#include <asm/div64.h>
51#include <asm/timex.h>
52#include <asm/io.h>
53
54#include "tick-internal.h"
55
56#define CREATE_TRACE_POINTS
57#include <trace/events/timer.h>
58
59__visible u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
60
61EXPORT_SYMBOL(jiffies_64);
62
63/*
64 * The timer wheel has LVL_DEPTH array levels. Each level provides an array of
65 * LVL_SIZE buckets. Each level is driven by its own clock and therefor each
66 * level has a different granularity.
67 *
68 * The level granularity is: LVL_CLK_DIV ^ lvl
69 * The level clock frequency is: HZ / (LVL_CLK_DIV ^ level)
70 *
71 * The array level of a newly armed timer depends on the relative expiry
72 * time. The farther the expiry time is away the higher the array level and
73 * therefor the granularity becomes.
74 *
75 * Contrary to the original timer wheel implementation, which aims for 'exact'
76 * expiry of the timers, this implementation removes the need for recascading
77 * the timers into the lower array levels. The previous 'classic' timer wheel
78 * implementation of the kernel already violated the 'exact' expiry by adding
79 * slack to the expiry time to provide batched expiration. The granularity
80 * levels provide implicit batching.
81 *
82 * This is an optimization of the original timer wheel implementation for the
83 * majority of the timer wheel use cases: timeouts. The vast majority of
84 * timeout timers (networking, disk I/O ...) are canceled before expiry. If
85 * the timeout expires it indicates that normal operation is disturbed, so it
86 * does not matter much whether the timeout comes with a slight delay.
87 *
88 * The only exception to this are networking timers with a small expiry
89 * time. They rely on the granularity. Those fit into the first wheel level,
90 * which has HZ granularity.
91 *
92 * We don't have cascading anymore. timers with a expiry time above the
93 * capacity of the last wheel level are force expired at the maximum timeout
94 * value of the last wheel level. From data sampling we know that the maximum
95 * value observed is 5 days (network connection tracking), so this should not
96 * be an issue.
97 *
98 * The currently chosen array constants values are a good compromise between
99 * array size and granularity.
100 *
101 * This results in the following granularity and range levels:
102 *
103 * HZ 1000 steps
104 * Level Offset Granularity Range
105 * 0 0 1 ms 0 ms - 63 ms
106 * 1 64 8 ms 64 ms - 511 ms
107 * 2 128 64 ms 512 ms - 4095 ms (512ms - ~4s)
108 * 3 192 512 ms 4096 ms - 32767 ms (~4s - ~32s)
109 * 4 256 4096 ms (~4s) 32768 ms - 262143 ms (~32s - ~4m)
110 * 5 320 32768 ms (~32s) 262144 ms - 2097151 ms (~4m - ~34m)
111 * 6 384 262144 ms (~4m) 2097152 ms - 16777215 ms (~34m - ~4h)
112 * 7 448 2097152 ms (~34m) 16777216 ms - 134217727 ms (~4h - ~1d)
113 * 8 512 16777216 ms (~4h) 134217728 ms - 1073741822 ms (~1d - ~12d)
114 *
115 * HZ 300
116 * Level Offset Granularity Range
117 * 0 0 3 ms 0 ms - 210 ms
118 * 1 64 26 ms 213 ms - 1703 ms (213ms - ~1s)
119 * 2 128 213 ms 1706 ms - 13650 ms (~1s - ~13s)
120 * 3 192 1706 ms (~1s) 13653 ms - 109223 ms (~13s - ~1m)
121 * 4 256 13653 ms (~13s) 109226 ms - 873810 ms (~1m - ~14m)
122 * 5 320 109226 ms (~1m) 873813 ms - 6990503 ms (~14m - ~1h)
123 * 6 384 873813 ms (~14m) 6990506 ms - 55924050 ms (~1h - ~15h)
124 * 7 448 6990506 ms (~1h) 55924053 ms - 447392423 ms (~15h - ~5d)
125 * 8 512 55924053 ms (~15h) 447392426 ms - 3579139406 ms (~5d - ~41d)
126 *
127 * HZ 250
128 * Level Offset Granularity Range
129 * 0 0 4 ms 0 ms - 255 ms
130 * 1 64 32 ms 256 ms - 2047 ms (256ms - ~2s)
131 * 2 128 256 ms 2048 ms - 16383 ms (~2s - ~16s)
132 * 3 192 2048 ms (~2s) 16384 ms - 131071 ms (~16s - ~2m)
133 * 4 256 16384 ms (~16s) 131072 ms - 1048575 ms (~2m - ~17m)
134 * 5 320 131072 ms (~2m) 1048576 ms - 8388607 ms (~17m - ~2h)
135 * 6 384 1048576 ms (~17m) 8388608 ms - 67108863 ms (~2h - ~18h)
136 * 7 448 8388608 ms (~2h) 67108864 ms - 536870911 ms (~18h - ~6d)
137 * 8 512 67108864 ms (~18h) 536870912 ms - 4294967288 ms (~6d - ~49d)
138 *
139 * HZ 100
140 * Level Offset Granularity Range
141 * 0 0 10 ms 0 ms - 630 ms
142 * 1 64 80 ms 640 ms - 5110 ms (640ms - ~5s)
143 * 2 128 640 ms 5120 ms - 40950 ms (~5s - ~40s)
144 * 3 192 5120 ms (~5s) 40960 ms - 327670 ms (~40s - ~5m)
145 * 4 256 40960 ms (~40s) 327680 ms - 2621430 ms (~5m - ~43m)
146 * 5 320 327680 ms (~5m) 2621440 ms - 20971510 ms (~43m - ~5h)
147 * 6 384 2621440 ms (~43m) 20971520 ms - 167772150 ms (~5h - ~1d)
148 * 7 448 20971520 ms (~5h) 167772160 ms - 1342177270 ms (~1d - ~15d)
149 */
150
151/* Clock divisor for the next level */
152#define LVL_CLK_SHIFT 3
153#define LVL_CLK_DIV (1UL << LVL_CLK_SHIFT)
154#define LVL_CLK_MASK (LVL_CLK_DIV - 1)
155#define LVL_SHIFT(n) ((n) * LVL_CLK_SHIFT)
156#define LVL_GRAN(n) (1UL << LVL_SHIFT(n))
157
158/*
159 * The time start value for each level to select the bucket at enqueue
160 * time. We start from the last possible delta of the previous level
161 * so that we can later add an extra LVL_GRAN(n) to n (see calc_index()).
162 */
163#define LVL_START(n) ((LVL_SIZE - 1) << (((n) - 1) * LVL_CLK_SHIFT))
164
165/* Size of each clock level */
166#define LVL_BITS 6
167#define LVL_SIZE (1UL << LVL_BITS)
168#define LVL_MASK (LVL_SIZE - 1)
169#define LVL_OFFS(n) ((n) * LVL_SIZE)
170
171/* Level depth */
172#if HZ > 100
173# define LVL_DEPTH 9
174# else
175# define LVL_DEPTH 8
176#endif
177
178/* The cutoff (max. capacity of the wheel) */
179#define WHEEL_TIMEOUT_CUTOFF (LVL_START(LVL_DEPTH))
180#define WHEEL_TIMEOUT_MAX (WHEEL_TIMEOUT_CUTOFF - LVL_GRAN(LVL_DEPTH - 1))
181
182/*
183 * The resulting wheel size. If NOHZ is configured we allocate two
184 * wheels so we have a separate storage for the deferrable timers.
185 */
186#define WHEEL_SIZE (LVL_SIZE * LVL_DEPTH)
187
188#ifdef CONFIG_NO_HZ_COMMON
189# define NR_BASES 2
190# define BASE_STD 0
191# define BASE_DEF 1
192#else
193# define NR_BASES 1
194# define BASE_STD 0
195# define BASE_DEF 0
196#endif
197
198struct timer_base {
199 raw_spinlock_t lock;
200 struct timer_list *running_timer;
201#ifdef CONFIG_PREEMPT_RT
202 spinlock_t expiry_lock;
203 atomic_t timer_waiters;
204#endif
205 unsigned long clk;
206 unsigned long next_expiry;
207 unsigned int cpu;
208 bool next_expiry_recalc;
209 bool is_idle;
210 DECLARE_BITMAP(pending_map, WHEEL_SIZE);
211 struct hlist_head vectors[WHEEL_SIZE];
212} ____cacheline_aligned;
213
214static DEFINE_PER_CPU(struct timer_base, timer_bases[NR_BASES]);
215
216#ifdef CONFIG_NO_HZ_COMMON
217
218static DEFINE_STATIC_KEY_FALSE(timers_nohz_active);
219static DEFINE_MUTEX(timer_keys_mutex);
220
221static void timer_update_keys(struct work_struct *work);
222static DECLARE_WORK(timer_update_work, timer_update_keys);
223
224#ifdef CONFIG_SMP
225unsigned int sysctl_timer_migration = 1;
226
227DEFINE_STATIC_KEY_FALSE(timers_migration_enabled);
228
229static void timers_update_migration(void)
230{
231 if (sysctl_timer_migration && tick_nohz_active)
232 static_branch_enable(&timers_migration_enabled);
233 else
234 static_branch_disable(&timers_migration_enabled);
235}
236#else
237static inline void timers_update_migration(void) { }
238#endif /* !CONFIG_SMP */
239
240static void timer_update_keys(struct work_struct *work)
241{
242 mutex_lock(&timer_keys_mutex);
243 timers_update_migration();
244 static_branch_enable(&timers_nohz_active);
245 mutex_unlock(&timer_keys_mutex);
246}
247
248void timers_update_nohz(void)
249{
250 schedule_work(&timer_update_work);
251}
252
253int timer_migration_handler(struct ctl_table *table, int write,
254 void *buffer, size_t *lenp, loff_t *ppos)
255{
256 int ret;
257
258 mutex_lock(&timer_keys_mutex);
259 ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
260 if (!ret && write)
261 timers_update_migration();
262 mutex_unlock(&timer_keys_mutex);
263 return ret;
264}
265
266static inline bool is_timers_nohz_active(void)
267{
268 return static_branch_unlikely(&timers_nohz_active);
269}
270#else
271static inline bool is_timers_nohz_active(void) { return false; }
272#endif /* NO_HZ_COMMON */
273
274static unsigned long round_jiffies_common(unsigned long j, int cpu,
275 bool force_up)
276{
277 int rem;
278 unsigned long original = j;
279
280 /*
281 * We don't want all cpus firing their timers at once hitting the
282 * same lock or cachelines, so we skew each extra cpu with an extra
283 * 3 jiffies. This 3 jiffies came originally from the mm/ code which
284 * already did this.
285 * The skew is done by adding 3*cpunr, then round, then subtract this
286 * extra offset again.
287 */
288 j += cpu * 3;
289
290 rem = j % HZ;
291
292 /*
293 * If the target jiffie is just after a whole second (which can happen
294 * due to delays of the timer irq, long irq off times etc etc) then
295 * we should round down to the whole second, not up. Use 1/4th second
296 * as cutoff for this rounding as an extreme upper bound for this.
297 * But never round down if @force_up is set.
298 */
299 if (rem < HZ/4 && !force_up) /* round down */
300 j = j - rem;
301 else /* round up */
302 j = j - rem + HZ;
303
304 /* now that we have rounded, subtract the extra skew again */
305 j -= cpu * 3;
306
307 /*
308 * Make sure j is still in the future. Otherwise return the
309 * unmodified value.
310 */
311 return time_is_after_jiffies(j) ? j : original;
312}
313
314/**
315 * __round_jiffies - function to round jiffies to a full second
316 * @j: the time in (absolute) jiffies that should be rounded
317 * @cpu: the processor number on which the timeout will happen
318 *
319 * __round_jiffies() rounds an absolute time in the future (in jiffies)
320 * up or down to (approximately) full seconds. This is useful for timers
321 * for which the exact time they fire does not matter too much, as long as
322 * they fire approximately every X seconds.
323 *
324 * By rounding these timers to whole seconds, all such timers will fire
325 * at the same time, rather than at various times spread out. The goal
326 * of this is to have the CPU wake up less, which saves power.
327 *
328 * The exact rounding is skewed for each processor to avoid all
329 * processors firing at the exact same time, which could lead
330 * to lock contention or spurious cache line bouncing.
331 *
332 * The return value is the rounded version of the @j parameter.
333 */
334unsigned long __round_jiffies(unsigned long j, int cpu)
335{
336 return round_jiffies_common(j, cpu, false);
337}
338EXPORT_SYMBOL_GPL(__round_jiffies);
339
340/**
341 * __round_jiffies_relative - function to round jiffies to a full second
342 * @j: the time in (relative) jiffies that should be rounded
343 * @cpu: the processor number on which the timeout will happen
344 *
345 * __round_jiffies_relative() rounds a time delta in the future (in jiffies)
346 * up or down to (approximately) full seconds. This is useful for timers
347 * for which the exact time they fire does not matter too much, as long as
348 * they fire approximately every X seconds.
349 *
350 * By rounding these timers to whole seconds, all such timers will fire
351 * at the same time, rather than at various times spread out. The goal
352 * of this is to have the CPU wake up less, which saves power.
353 *
354 * The exact rounding is skewed for each processor to avoid all
355 * processors firing at the exact same time, which could lead
356 * to lock contention or spurious cache line bouncing.
357 *
358 * The return value is the rounded version of the @j parameter.
359 */
360unsigned long __round_jiffies_relative(unsigned long j, int cpu)
361{
362 unsigned long j0 = jiffies;
363
364 /* Use j0 because jiffies might change while we run */
365 return round_jiffies_common(j + j0, cpu, false) - j0;
366}
367EXPORT_SYMBOL_GPL(__round_jiffies_relative);
368
369/**
370 * round_jiffies - function to round jiffies to a full second
371 * @j: the time in (absolute) jiffies that should be rounded
372 *
373 * round_jiffies() rounds an absolute time in the future (in jiffies)
374 * up or down to (approximately) full seconds. This is useful for timers
375 * for which the exact time they fire does not matter too much, as long as
376 * they fire approximately every X seconds.
377 *
378 * By rounding these timers to whole seconds, all such timers will fire
379 * at the same time, rather than at various times spread out. The goal
380 * of this is to have the CPU wake up less, which saves power.
381 *
382 * The return value is the rounded version of the @j parameter.
383 */
384unsigned long round_jiffies(unsigned long j)
385{
386 return round_jiffies_common(j, raw_smp_processor_id(), false);
387}
388EXPORT_SYMBOL_GPL(round_jiffies);
389
390/**
391 * round_jiffies_relative - function to round jiffies to a full second
392 * @j: the time in (relative) jiffies that should be rounded
393 *
394 * round_jiffies_relative() rounds a time delta in the future (in jiffies)
395 * up or down to (approximately) full seconds. This is useful for timers
396 * for which the exact time they fire does not matter too much, as long as
397 * they fire approximately every X seconds.
398 *
399 * By rounding these timers to whole seconds, all such timers will fire
400 * at the same time, rather than at various times spread out. The goal
401 * of this is to have the CPU wake up less, which saves power.
402 *
403 * The return value is the rounded version of the @j parameter.
404 */
405unsigned long round_jiffies_relative(unsigned long j)
406{
407 return __round_jiffies_relative(j, raw_smp_processor_id());
408}
409EXPORT_SYMBOL_GPL(round_jiffies_relative);
410
411/**
412 * __round_jiffies_up - function to round jiffies up to a full second
413 * @j: the time in (absolute) jiffies that should be rounded
414 * @cpu: the processor number on which the timeout will happen
415 *
416 * This is the same as __round_jiffies() except that it will never
417 * round down. This is useful for timeouts for which the exact time
418 * of firing does not matter too much, as long as they don't fire too
419 * early.
420 */
421unsigned long __round_jiffies_up(unsigned long j, int cpu)
422{
423 return round_jiffies_common(j, cpu, true);
424}
425EXPORT_SYMBOL_GPL(__round_jiffies_up);
426
427/**
428 * __round_jiffies_up_relative - function to round jiffies up to a full second
429 * @j: the time in (relative) jiffies that should be rounded
430 * @cpu: the processor number on which the timeout will happen
431 *
432 * This is the same as __round_jiffies_relative() except that it will never
433 * round down. This is useful for timeouts for which the exact time
434 * of firing does not matter too much, as long as they don't fire too
435 * early.
436 */
437unsigned long __round_jiffies_up_relative(unsigned long j, int cpu)
438{
439 unsigned long j0 = jiffies;
440
441 /* Use j0 because jiffies might change while we run */
442 return round_jiffies_common(j + j0, cpu, true) - j0;
443}
444EXPORT_SYMBOL_GPL(__round_jiffies_up_relative);
445
446/**
447 * round_jiffies_up - function to round jiffies up to a full second
448 * @j: the time in (absolute) jiffies that should be rounded
449 *
450 * This is the same as round_jiffies() except that it will never
451 * round down. This is useful for timeouts for which the exact time
452 * of firing does not matter too much, as long as they don't fire too
453 * early.
454 */
455unsigned long round_jiffies_up(unsigned long j)
456{
457 return round_jiffies_common(j, raw_smp_processor_id(), true);
458}
459EXPORT_SYMBOL_GPL(round_jiffies_up);
460
461/**
462 * round_jiffies_up_relative - function to round jiffies up to a full second
463 * @j: the time in (relative) jiffies that should be rounded
464 *
465 * This is the same as round_jiffies_relative() except that it will never
466 * round down. This is useful for timeouts for which the exact time
467 * of firing does not matter too much, as long as they don't fire too
468 * early.
469 */
470unsigned long round_jiffies_up_relative(unsigned long j)
471{
472 return __round_jiffies_up_relative(j, raw_smp_processor_id());
473}
474EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
475
476
477static inline unsigned int timer_get_idx(struct timer_list *timer)
478{
479 return (timer->flags & TIMER_ARRAYMASK) >> TIMER_ARRAYSHIFT;
480}
481
482static inline void timer_set_idx(struct timer_list *timer, unsigned int idx)
483{
484 timer->flags = (timer->flags & ~TIMER_ARRAYMASK) |
485 idx << TIMER_ARRAYSHIFT;
486}
487
488/*
489 * Helper function to calculate the array index for a given expiry
490 * time.
491 */
492static inline unsigned calc_index(unsigned long expires, unsigned lvl,
493 unsigned long *bucket_expiry)
494{
495
496 /*
497 * The timer wheel has to guarantee that a timer does not fire
498 * early. Early expiry can happen due to:
499 * - Timer is armed at the edge of a tick
500 * - Truncation of the expiry time in the outer wheel levels
501 *
502 * Round up with level granularity to prevent this.
503 */
504 expires = (expires + LVL_GRAN(lvl)) >> LVL_SHIFT(lvl);
505 *bucket_expiry = expires << LVL_SHIFT(lvl);
506 return LVL_OFFS(lvl) + (expires & LVL_MASK);
507}
508
509static int calc_wheel_index(unsigned long expires, unsigned long clk,
510 unsigned long *bucket_expiry)
511{
512 unsigned long delta = expires - clk;
513 unsigned int idx;
514
515 if (delta < LVL_START(1)) {
516 idx = calc_index(expires, 0, bucket_expiry);
517 } else if (delta < LVL_START(2)) {
518 idx = calc_index(expires, 1, bucket_expiry);
519 } else if (delta < LVL_START(3)) {
520 idx = calc_index(expires, 2, bucket_expiry);
521 } else if (delta < LVL_START(4)) {
522 idx = calc_index(expires, 3, bucket_expiry);
523 } else if (delta < LVL_START(5)) {
524 idx = calc_index(expires, 4, bucket_expiry);
525 } else if (delta < LVL_START(6)) {
526 idx = calc_index(expires, 5, bucket_expiry);
527 } else if (delta < LVL_START(7)) {
528 idx = calc_index(expires, 6, bucket_expiry);
529 } else if (LVL_DEPTH > 8 && delta < LVL_START(8)) {
530 idx = calc_index(expires, 7, bucket_expiry);
531 } else if ((long) delta < 0) {
532 idx = clk & LVL_MASK;
533 *bucket_expiry = clk;
534 } else {
535 /*
536 * Force expire obscene large timeouts to expire at the
537 * capacity limit of the wheel.
538 */
539 if (delta >= WHEEL_TIMEOUT_CUTOFF)
540 expires = clk + WHEEL_TIMEOUT_MAX;
541
542 idx = calc_index(expires, LVL_DEPTH - 1, bucket_expiry);
543 }
544 return idx;
545}
546
547static void
548trigger_dyntick_cpu(struct timer_base *base, struct timer_list *timer)
549{
550 if (!is_timers_nohz_active())
551 return;
552
553 /*
554 * TODO: This wants some optimizing similar to the code below, but we
555 * will do that when we switch from push to pull for deferrable timers.
556 */
557 if (timer->flags & TIMER_DEFERRABLE) {
558 if (tick_nohz_full_cpu(base->cpu))
559 wake_up_nohz_cpu(base->cpu);
560 return;
561 }
562
563 /*
564 * We might have to IPI the remote CPU if the base is idle and the
565 * timer is not deferrable. If the other CPU is on the way to idle
566 * then it can't set base->is_idle as we hold the base lock:
567 */
568 if (base->is_idle)
569 wake_up_nohz_cpu(base->cpu);
570}
571
572/*
573 * Enqueue the timer into the hash bucket, mark it pending in
574 * the bitmap, store the index in the timer flags then wake up
575 * the target CPU if needed.
576 */
577static void enqueue_timer(struct timer_base *base, struct timer_list *timer,
578 unsigned int idx, unsigned long bucket_expiry)
579{
580
581 hlist_add_head(&timer->entry, base->vectors + idx);
582 __set_bit(idx, base->pending_map);
583 timer_set_idx(timer, idx);
584
585 trace_timer_start(timer, timer->expires, timer->flags);
586
587 /*
588 * Check whether this is the new first expiring timer. The
589 * effective expiry time of the timer is required here
590 * (bucket_expiry) instead of timer->expires.
591 */
592 if (time_before(bucket_expiry, base->next_expiry)) {
593 /*
594 * Set the next expiry time and kick the CPU so it
595 * can reevaluate the wheel:
596 */
597 base->next_expiry = bucket_expiry;
598 base->next_expiry_recalc = false;
599 trigger_dyntick_cpu(base, timer);
600 }
601}
602
603static void internal_add_timer(struct timer_base *base, struct timer_list *timer)
604{
605 unsigned long bucket_expiry;
606 unsigned int idx;
607
608 idx = calc_wheel_index(timer->expires, base->clk, &bucket_expiry);
609 enqueue_timer(base, timer, idx, bucket_expiry);
610}
611
612#ifdef CONFIG_DEBUG_OBJECTS_TIMERS
613
614static struct debug_obj_descr timer_debug_descr;
615
616static void *timer_debug_hint(void *addr)
617{
618 return ((struct timer_list *) addr)->function;
619}
620
621static bool timer_is_static_object(void *addr)
622{
623 struct timer_list *timer = addr;
624
625 return (timer->entry.pprev == NULL &&
626 timer->entry.next == TIMER_ENTRY_STATIC);
627}
628
629/*
630 * fixup_init is called when:
631 * - an active object is initialized
632 */
633static bool timer_fixup_init(void *addr, enum debug_obj_state state)
634{
635 struct timer_list *timer = addr;
636
637 switch (state) {
638 case ODEBUG_STATE_ACTIVE:
639 del_timer_sync(timer);
640 debug_object_init(timer, &timer_debug_descr);
641 return true;
642 default:
643 return false;
644 }
645}
646
647/* Stub timer callback for improperly used timers. */
648static void stub_timer(struct timer_list *unused)
649{
650 WARN_ON(1);
651}
652
653/*
654 * fixup_activate is called when:
655 * - an active object is activated
656 * - an unknown non-static object is activated
657 */
658static bool timer_fixup_activate(void *addr, enum debug_obj_state state)
659{
660 struct timer_list *timer = addr;
661
662 switch (state) {
663 case ODEBUG_STATE_NOTAVAILABLE:
664 timer_setup(timer, stub_timer, 0);
665 return true;
666
667 case ODEBUG_STATE_ACTIVE:
668 WARN_ON(1);
669 fallthrough;
670 default:
671 return false;
672 }
673}
674
675/*
676 * fixup_free is called when:
677 * - an active object is freed
678 */
679static bool timer_fixup_free(void *addr, enum debug_obj_state state)
680{
681 struct timer_list *timer = addr;
682
683 switch (state) {
684 case ODEBUG_STATE_ACTIVE:
685 del_timer_sync(timer);
686 debug_object_free(timer, &timer_debug_descr);
687 return true;
688 default:
689 return false;
690 }
691}
692
693/*
694 * fixup_assert_init is called when:
695 * - an untracked/uninit-ed object is found
696 */
697static bool timer_fixup_assert_init(void *addr, enum debug_obj_state state)
698{
699 struct timer_list *timer = addr;
700
701 switch (state) {
702 case ODEBUG_STATE_NOTAVAILABLE:
703 timer_setup(timer, stub_timer, 0);
704 return true;
705 default:
706 return false;
707 }
708}
709
710static struct debug_obj_descr timer_debug_descr = {
711 .name = "timer_list",
712 .debug_hint = timer_debug_hint,
713 .is_static_object = timer_is_static_object,
714 .fixup_init = timer_fixup_init,
715 .fixup_activate = timer_fixup_activate,
716 .fixup_free = timer_fixup_free,
717 .fixup_assert_init = timer_fixup_assert_init,
718};
719
720static inline void debug_timer_init(struct timer_list *timer)
721{
722 debug_object_init(timer, &timer_debug_descr);
723}
724
725static inline void debug_timer_activate(struct timer_list *timer)
726{
727 debug_object_activate(timer, &timer_debug_descr);
728}
729
730static inline void debug_timer_deactivate(struct timer_list *timer)
731{
732 debug_object_deactivate(timer, &timer_debug_descr);
733}
734
735static inline void debug_timer_free(struct timer_list *timer)
736{
737 debug_object_free(timer, &timer_debug_descr);
738}
739
740static inline void debug_timer_assert_init(struct timer_list *timer)
741{
742 debug_object_assert_init(timer, &timer_debug_descr);
743}
744
745static void do_init_timer(struct timer_list *timer,
746 void (*func)(struct timer_list *),
747 unsigned int flags,
748 const char *name, struct lock_class_key *key);
749
750void init_timer_on_stack_key(struct timer_list *timer,
751 void (*func)(struct timer_list *),
752 unsigned int flags,
753 const char *name, struct lock_class_key *key)
754{
755 debug_object_init_on_stack(timer, &timer_debug_descr);
756 do_init_timer(timer, func, flags, name, key);
757}
758EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
759
760void destroy_timer_on_stack(struct timer_list *timer)
761{
762 debug_object_free(timer, &timer_debug_descr);
763}
764EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
765
766#else
767static inline void debug_timer_init(struct timer_list *timer) { }
768static inline void debug_timer_activate(struct timer_list *timer) { }
769static inline void debug_timer_deactivate(struct timer_list *timer) { }
770static inline void debug_timer_assert_init(struct timer_list *timer) { }
771#endif
772
773static inline void debug_init(struct timer_list *timer)
774{
775 debug_timer_init(timer);
776 trace_timer_init(timer);
777}
778
779static inline void debug_deactivate(struct timer_list *timer)
780{
781 debug_timer_deactivate(timer);
782 trace_timer_cancel(timer);
783}
784
785static inline void debug_assert_init(struct timer_list *timer)
786{
787 debug_timer_assert_init(timer);
788}
789
790static void do_init_timer(struct timer_list *timer,
791 void (*func)(struct timer_list *),
792 unsigned int flags,
793 const char *name, struct lock_class_key *key)
794{
795 timer->entry.pprev = NULL;
796 timer->function = func;
797 timer->flags = flags | raw_smp_processor_id();
798 lockdep_init_map(&timer->lockdep_map, name, key, 0);
799}
800
801/**
802 * init_timer_key - initialize a timer
803 * @timer: the timer to be initialized
804 * @func: timer callback function
805 * @flags: timer flags
806 * @name: name of the timer
807 * @key: lockdep class key of the fake lock used for tracking timer
808 * sync lock dependencies
809 *
810 * init_timer_key() must be done to a timer prior calling *any* of the
811 * other timer functions.
812 */
813void init_timer_key(struct timer_list *timer,
814 void (*func)(struct timer_list *), unsigned int flags,
815 const char *name, struct lock_class_key *key)
816{
817 debug_init(timer);
818 do_init_timer(timer, func, flags, name, key);
819}
820EXPORT_SYMBOL(init_timer_key);
821
822static inline void detach_timer(struct timer_list *timer, bool clear_pending)
823{
824 struct hlist_node *entry = &timer->entry;
825
826 debug_deactivate(timer);
827
828 __hlist_del(entry);
829 if (clear_pending)
830 entry->pprev = NULL;
831 entry->next = LIST_POISON2;
832}
833
834static int detach_if_pending(struct timer_list *timer, struct timer_base *base,
835 bool clear_pending)
836{
837 unsigned idx = timer_get_idx(timer);
838
839 if (!timer_pending(timer))
840 return 0;
841
842 if (hlist_is_singular_node(&timer->entry, base->vectors + idx)) {
843 __clear_bit(idx, base->pending_map);
844 base->next_expiry_recalc = true;
845 }
846
847 detach_timer(timer, clear_pending);
848 return 1;
849}
850
851static inline struct timer_base *get_timer_cpu_base(u32 tflags, u32 cpu)
852{
853 struct timer_base *base = per_cpu_ptr(&timer_bases[BASE_STD], cpu);
854
855 /*
856 * If the timer is deferrable and NO_HZ_COMMON is set then we need
857 * to use the deferrable base.
858 */
859 if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE))
860 base = per_cpu_ptr(&timer_bases[BASE_DEF], cpu);
861 return base;
862}
863
864static inline struct timer_base *get_timer_this_cpu_base(u32 tflags)
865{
866 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
867
868 /*
869 * If the timer is deferrable and NO_HZ_COMMON is set then we need
870 * to use the deferrable base.
871 */
872 if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE))
873 base = this_cpu_ptr(&timer_bases[BASE_DEF]);
874 return base;
875}
876
877static inline struct timer_base *get_timer_base(u32 tflags)
878{
879 return get_timer_cpu_base(tflags, tflags & TIMER_CPUMASK);
880}
881
882static inline struct timer_base *
883get_target_base(struct timer_base *base, unsigned tflags)
884{
885#if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
886 if (static_branch_likely(&timers_migration_enabled) &&
887 !(tflags & TIMER_PINNED))
888 return get_timer_cpu_base(tflags, get_nohz_timer_target());
889#endif
890 return get_timer_this_cpu_base(tflags);
891}
892
893static inline void forward_timer_base(struct timer_base *base)
894{
895 unsigned long jnow = READ_ONCE(jiffies);
896
897 /*
898 * No need to forward if we are close enough below jiffies.
899 * Also while executing timers, base->clk is 1 offset ahead
900 * of jiffies to avoid endless requeuing to current jffies.
901 */
902 if ((long)(jnow - base->clk) < 1)
903 return;
904
905 /*
906 * If the next expiry value is > jiffies, then we fast forward to
907 * jiffies otherwise we forward to the next expiry value.
908 */
909 if (time_after(base->next_expiry, jnow)) {
910 base->clk = jnow;
911 } else {
912 if (WARN_ON_ONCE(time_before(base->next_expiry, base->clk)))
913 return;
914 base->clk = base->next_expiry;
915 }
916}
917
918
919/*
920 * We are using hashed locking: Holding per_cpu(timer_bases[x]).lock means
921 * that all timers which are tied to this base are locked, and the base itself
922 * is locked too.
923 *
924 * So __run_timers/migrate_timers can safely modify all timers which could
925 * be found in the base->vectors array.
926 *
927 * When a timer is migrating then the TIMER_MIGRATING flag is set and we need
928 * to wait until the migration is done.
929 */
930static struct timer_base *lock_timer_base(struct timer_list *timer,
931 unsigned long *flags)
932 __acquires(timer->base->lock)
933{
934 for (;;) {
935 struct timer_base *base;
936 u32 tf;
937
938 /*
939 * We need to use READ_ONCE() here, otherwise the compiler
940 * might re-read @tf between the check for TIMER_MIGRATING
941 * and spin_lock().
942 */
943 tf = READ_ONCE(timer->flags);
944
945 if (!(tf & TIMER_MIGRATING)) {
946 base = get_timer_base(tf);
947 raw_spin_lock_irqsave(&base->lock, *flags);
948 if (timer->flags == tf)
949 return base;
950 raw_spin_unlock_irqrestore(&base->lock, *flags);
951 }
952 cpu_relax();
953 }
954}
955
956#define MOD_TIMER_PENDING_ONLY 0x01
957#define MOD_TIMER_REDUCE 0x02
958#define MOD_TIMER_NOTPENDING 0x04
959
960static inline int
961__mod_timer(struct timer_list *timer, unsigned long expires, unsigned int options)
962{
963 unsigned long clk = 0, flags, bucket_expiry;
964 struct timer_base *base, *new_base;
965 unsigned int idx = UINT_MAX;
966 int ret = 0;
967
968 BUG_ON(!timer->function);
969
970 /*
971 * This is a common optimization triggered by the networking code - if
972 * the timer is re-modified to have the same timeout or ends up in the
973 * same array bucket then just return:
974 */
975 if (!(options & MOD_TIMER_NOTPENDING) && timer_pending(timer)) {
976 /*
977 * The downside of this optimization is that it can result in
978 * larger granularity than you would get from adding a new
979 * timer with this expiry.
980 */
981 long diff = timer->expires - expires;
982
983 if (!diff)
984 return 1;
985 if (options & MOD_TIMER_REDUCE && diff <= 0)
986 return 1;
987
988 /*
989 * We lock timer base and calculate the bucket index right
990 * here. If the timer ends up in the same bucket, then we
991 * just update the expiry time and avoid the whole
992 * dequeue/enqueue dance.
993 */
994 base = lock_timer_base(timer, &flags);
995 forward_timer_base(base);
996
997 if (timer_pending(timer) && (options & MOD_TIMER_REDUCE) &&
998 time_before_eq(timer->expires, expires)) {
999 ret = 1;
1000 goto out_unlock;
1001 }
1002
1003 clk = base->clk;
1004 idx = calc_wheel_index(expires, clk, &bucket_expiry);
1005
1006 /*
1007 * Retrieve and compare the array index of the pending
1008 * timer. If it matches set the expiry to the new value so a
1009 * subsequent call will exit in the expires check above.
1010 */
1011 if (idx == timer_get_idx(timer)) {
1012 if (!(options & MOD_TIMER_REDUCE))
1013 timer->expires = expires;
1014 else if (time_after(timer->expires, expires))
1015 timer->expires = expires;
1016 ret = 1;
1017 goto out_unlock;
1018 }
1019 } else {
1020 base = lock_timer_base(timer, &flags);
1021 forward_timer_base(base);
1022 }
1023
1024 ret = detach_if_pending(timer, base, false);
1025 if (!ret && (options & MOD_TIMER_PENDING_ONLY))
1026 goto out_unlock;
1027
1028 new_base = get_target_base(base, timer->flags);
1029
1030 if (base != new_base) {
1031 /*
1032 * We are trying to schedule the timer on the new base.
1033 * However we can't change timer's base while it is running,
1034 * otherwise del_timer_sync() can't detect that the timer's
1035 * handler yet has not finished. This also guarantees that the
1036 * timer is serialized wrt itself.
1037 */
1038 if (likely(base->running_timer != timer)) {
1039 /* See the comment in lock_timer_base() */
1040 timer->flags |= TIMER_MIGRATING;
1041
1042 raw_spin_unlock(&base->lock);
1043 base = new_base;
1044 raw_spin_lock(&base->lock);
1045 WRITE_ONCE(timer->flags,
1046 (timer->flags & ~TIMER_BASEMASK) | base->cpu);
1047 forward_timer_base(base);
1048 }
1049 }
1050
1051 debug_timer_activate(timer);
1052
1053 timer->expires = expires;
1054 /*
1055 * If 'idx' was calculated above and the base time did not advance
1056 * between calculating 'idx' and possibly switching the base, only
1057 * enqueue_timer() is required. Otherwise we need to (re)calculate
1058 * the wheel index via internal_add_timer().
1059 */
1060 if (idx != UINT_MAX && clk == base->clk)
1061 enqueue_timer(base, timer, idx, bucket_expiry);
1062 else
1063 internal_add_timer(base, timer);
1064
1065out_unlock:
1066 raw_spin_unlock_irqrestore(&base->lock, flags);
1067
1068 return ret;
1069}
1070
1071/**
1072 * mod_timer_pending - modify a pending timer's timeout
1073 * @timer: the pending timer to be modified
1074 * @expires: new timeout in jiffies
1075 *
1076 * mod_timer_pending() is the same for pending timers as mod_timer(),
1077 * but will not re-activate and modify already deleted timers.
1078 *
1079 * It is useful for unserialized use of timers.
1080 */
1081int mod_timer_pending(struct timer_list *timer, unsigned long expires)
1082{
1083 return __mod_timer(timer, expires, MOD_TIMER_PENDING_ONLY);
1084}
1085EXPORT_SYMBOL(mod_timer_pending);
1086
1087/**
1088 * mod_timer - modify a timer's timeout
1089 * @timer: the timer to be modified
1090 * @expires: new timeout in jiffies
1091 *
1092 * mod_timer() is a more efficient way to update the expire field of an
1093 * active timer (if the timer is inactive it will be activated)
1094 *
1095 * mod_timer(timer, expires) is equivalent to:
1096 *
1097 * del_timer(timer); timer->expires = expires; add_timer(timer);
1098 *
1099 * Note that if there are multiple unserialized concurrent users of the
1100 * same timer, then mod_timer() is the only safe way to modify the timeout,
1101 * since add_timer() cannot modify an already running timer.
1102 *
1103 * The function returns whether it has modified a pending timer or not.
1104 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
1105 * active timer returns 1.)
1106 */
1107int mod_timer(struct timer_list *timer, unsigned long expires)
1108{
1109 return __mod_timer(timer, expires, 0);
1110}
1111EXPORT_SYMBOL(mod_timer);
1112
1113/**
1114 * timer_reduce - Modify a timer's timeout if it would reduce the timeout
1115 * @timer: The timer to be modified
1116 * @expires: New timeout in jiffies
1117 *
1118 * timer_reduce() is very similar to mod_timer(), except that it will only
1119 * modify a running timer if that would reduce the expiration time (it will
1120 * start a timer that isn't running).
1121 */
1122int timer_reduce(struct timer_list *timer, unsigned long expires)
1123{
1124 return __mod_timer(timer, expires, MOD_TIMER_REDUCE);
1125}
1126EXPORT_SYMBOL(timer_reduce);
1127
1128/**
1129 * add_timer - start a timer
1130 * @timer: the timer to be added
1131 *
1132 * The kernel will do a ->function(@timer) callback from the
1133 * timer interrupt at the ->expires point in the future. The
1134 * current time is 'jiffies'.
1135 *
1136 * The timer's ->expires, ->function fields must be set prior calling this
1137 * function.
1138 *
1139 * Timers with an ->expires field in the past will be executed in the next
1140 * timer tick.
1141 */
1142void add_timer(struct timer_list *timer)
1143{
1144 BUG_ON(timer_pending(timer));
1145 __mod_timer(timer, timer->expires, MOD_TIMER_NOTPENDING);
1146}
1147EXPORT_SYMBOL(add_timer);
1148
1149/**
1150 * add_timer_on - start a timer on a particular CPU
1151 * @timer: the timer to be added
1152 * @cpu: the CPU to start it on
1153 *
1154 * This is not very scalable on SMP. Double adds are not possible.
1155 */
1156void add_timer_on(struct timer_list *timer, int cpu)
1157{
1158 struct timer_base *new_base, *base;
1159 unsigned long flags;
1160
1161 BUG_ON(timer_pending(timer) || !timer->function);
1162
1163 new_base = get_timer_cpu_base(timer->flags, cpu);
1164
1165 /*
1166 * If @timer was on a different CPU, it should be migrated with the
1167 * old base locked to prevent other operations proceeding with the
1168 * wrong base locked. See lock_timer_base().
1169 */
1170 base = lock_timer_base(timer, &flags);
1171 if (base != new_base) {
1172 timer->flags |= TIMER_MIGRATING;
1173
1174 raw_spin_unlock(&base->lock);
1175 base = new_base;
1176 raw_spin_lock(&base->lock);
1177 WRITE_ONCE(timer->flags,
1178 (timer->flags & ~TIMER_BASEMASK) | cpu);
1179 }
1180 forward_timer_base(base);
1181
1182 debug_timer_activate(timer);
1183 internal_add_timer(base, timer);
1184 raw_spin_unlock_irqrestore(&base->lock, flags);
1185}
1186EXPORT_SYMBOL_GPL(add_timer_on);
1187
1188/**
1189 * del_timer - deactivate a timer.
1190 * @timer: the timer to be deactivated
1191 *
1192 * del_timer() deactivates a timer - this works on both active and inactive
1193 * timers.
1194 *
1195 * The function returns whether it has deactivated a pending timer or not.
1196 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
1197 * active timer returns 1.)
1198 */
1199int del_timer(struct timer_list *timer)
1200{
1201 struct timer_base *base;
1202 unsigned long flags;
1203 int ret = 0;
1204
1205 debug_assert_init(timer);
1206
1207 if (timer_pending(timer)) {
1208 base = lock_timer_base(timer, &flags);
1209 ret = detach_if_pending(timer, base, true);
1210 raw_spin_unlock_irqrestore(&base->lock, flags);
1211 }
1212
1213 return ret;
1214}
1215EXPORT_SYMBOL(del_timer);
1216
1217/**
1218 * try_to_del_timer_sync - Try to deactivate a timer
1219 * @timer: timer to delete
1220 *
1221 * This function tries to deactivate a timer. Upon successful (ret >= 0)
1222 * exit the timer is not queued and the handler is not running on any CPU.
1223 */
1224int try_to_del_timer_sync(struct timer_list *timer)
1225{
1226 struct timer_base *base;
1227 unsigned long flags;
1228 int ret = -1;
1229
1230 debug_assert_init(timer);
1231
1232 base = lock_timer_base(timer, &flags);
1233
1234 if (base->running_timer != timer)
1235 ret = detach_if_pending(timer, base, true);
1236
1237 raw_spin_unlock_irqrestore(&base->lock, flags);
1238
1239 return ret;
1240}
1241EXPORT_SYMBOL(try_to_del_timer_sync);
1242
1243#ifdef CONFIG_PREEMPT_RT
1244static __init void timer_base_init_expiry_lock(struct timer_base *base)
1245{
1246 spin_lock_init(&base->expiry_lock);
1247}
1248
1249static inline void timer_base_lock_expiry(struct timer_base *base)
1250{
1251 spin_lock(&base->expiry_lock);
1252}
1253
1254static inline void timer_base_unlock_expiry(struct timer_base *base)
1255{
1256 spin_unlock(&base->expiry_lock);
1257}
1258
1259/*
1260 * The counterpart to del_timer_wait_running().
1261 *
1262 * If there is a waiter for base->expiry_lock, then it was waiting for the
1263 * timer callback to finish. Drop expiry_lock and reaquire it. That allows
1264 * the waiter to acquire the lock and make progress.
1265 */
1266static void timer_sync_wait_running(struct timer_base *base)
1267{
1268 if (atomic_read(&base->timer_waiters)) {
1269 spin_unlock(&base->expiry_lock);
1270 spin_lock(&base->expiry_lock);
1271 }
1272}
1273
1274/*
1275 * This function is called on PREEMPT_RT kernels when the fast path
1276 * deletion of a timer failed because the timer callback function was
1277 * running.
1278 *
1279 * This prevents priority inversion, if the softirq thread on a remote CPU
1280 * got preempted, and it prevents a life lock when the task which tries to
1281 * delete a timer preempted the softirq thread running the timer callback
1282 * function.
1283 */
1284static void del_timer_wait_running(struct timer_list *timer)
1285{
1286 u32 tf;
1287
1288 tf = READ_ONCE(timer->flags);
1289 if (!(tf & TIMER_MIGRATING)) {
1290 struct timer_base *base = get_timer_base(tf);
1291
1292 /*
1293 * Mark the base as contended and grab the expiry lock,
1294 * which is held by the softirq across the timer
1295 * callback. Drop the lock immediately so the softirq can
1296 * expire the next timer. In theory the timer could already
1297 * be running again, but that's more than unlikely and just
1298 * causes another wait loop.
1299 */
1300 atomic_inc(&base->timer_waiters);
1301 spin_lock_bh(&base->expiry_lock);
1302 atomic_dec(&base->timer_waiters);
1303 spin_unlock_bh(&base->expiry_lock);
1304 }
1305}
1306#else
1307static inline void timer_base_init_expiry_lock(struct timer_base *base) { }
1308static inline void timer_base_lock_expiry(struct timer_base *base) { }
1309static inline void timer_base_unlock_expiry(struct timer_base *base) { }
1310static inline void timer_sync_wait_running(struct timer_base *base) { }
1311static inline void del_timer_wait_running(struct timer_list *timer) { }
1312#endif
1313
1314#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT_RT)
1315/**
1316 * del_timer_sync - deactivate a timer and wait for the handler to finish.
1317 * @timer: the timer to be deactivated
1318 *
1319 * This function only differs from del_timer() on SMP: besides deactivating
1320 * the timer it also makes sure the handler has finished executing on other
1321 * CPUs.
1322 *
1323 * Synchronization rules: Callers must prevent restarting of the timer,
1324 * otherwise this function is meaningless. It must not be called from
1325 * interrupt contexts unless the timer is an irqsafe one. The caller must
1326 * not hold locks which would prevent completion of the timer's
1327 * handler. The timer's handler must not call add_timer_on(). Upon exit the
1328 * timer is not queued and the handler is not running on any CPU.
1329 *
1330 * Note: For !irqsafe timers, you must not hold locks that are held in
1331 * interrupt context while calling this function. Even if the lock has
1332 * nothing to do with the timer in question. Here's why::
1333 *
1334 * CPU0 CPU1
1335 * ---- ----
1336 * <SOFTIRQ>
1337 * call_timer_fn();
1338 * base->running_timer = mytimer;
1339 * spin_lock_irq(somelock);
1340 * <IRQ>
1341 * spin_lock(somelock);
1342 * del_timer_sync(mytimer);
1343 * while (base->running_timer == mytimer);
1344 *
1345 * Now del_timer_sync() will never return and never release somelock.
1346 * The interrupt on the other CPU is waiting to grab somelock but
1347 * it has interrupted the softirq that CPU0 is waiting to finish.
1348 *
1349 * The function returns whether it has deactivated a pending timer or not.
1350 */
1351int del_timer_sync(struct timer_list *timer)
1352{
1353 int ret;
1354
1355#ifdef CONFIG_LOCKDEP
1356 unsigned long flags;
1357
1358 /*
1359 * If lockdep gives a backtrace here, please reference
1360 * the synchronization rules above.
1361 */
1362 local_irq_save(flags);
1363 lock_map_acquire(&timer->lockdep_map);
1364 lock_map_release(&timer->lockdep_map);
1365 local_irq_restore(flags);
1366#endif
1367 /*
1368 * don't use it in hardirq context, because it
1369 * could lead to deadlock.
1370 */
1371 WARN_ON(in_irq() && !(timer->flags & TIMER_IRQSAFE));
1372
1373 do {
1374 ret = try_to_del_timer_sync(timer);
1375
1376 if (unlikely(ret < 0)) {
1377 del_timer_wait_running(timer);
1378 cpu_relax();
1379 }
1380 } while (ret < 0);
1381
1382 return ret;
1383}
1384EXPORT_SYMBOL(del_timer_sync);
1385#endif
1386
1387static void call_timer_fn(struct timer_list *timer,
1388 void (*fn)(struct timer_list *),
1389 unsigned long baseclk)
1390{
1391 int count = preempt_count();
1392
1393#ifdef CONFIG_LOCKDEP
1394 /*
1395 * It is permissible to free the timer from inside the
1396 * function that is called from it, this we need to take into
1397 * account for lockdep too. To avoid bogus "held lock freed"
1398 * warnings as well as problems when looking into
1399 * timer->lockdep_map, make a copy and use that here.
1400 */
1401 struct lockdep_map lockdep_map;
1402
1403 lockdep_copy_map(&lockdep_map, &timer->lockdep_map);
1404#endif
1405 /*
1406 * Couple the lock chain with the lock chain at
1407 * del_timer_sync() by acquiring the lock_map around the fn()
1408 * call here and in del_timer_sync().
1409 */
1410 lock_map_acquire(&lockdep_map);
1411
1412 trace_timer_expire_entry(timer, baseclk);
1413 fn(timer);
1414 trace_timer_expire_exit(timer);
1415
1416 lock_map_release(&lockdep_map);
1417
1418 if (count != preempt_count()) {
1419 WARN_ONCE(1, "timer: %pS preempt leak: %08x -> %08x\n",
1420 fn, count, preempt_count());
1421 /*
1422 * Restore the preempt count. That gives us a decent
1423 * chance to survive and extract information. If the
1424 * callback kept a lock held, bad luck, but not worse
1425 * than the BUG() we had.
1426 */
1427 preempt_count_set(count);
1428 }
1429}
1430
1431static void expire_timers(struct timer_base *base, struct hlist_head *head)
1432{
1433 /*
1434 * This value is required only for tracing. base->clk was
1435 * incremented directly before expire_timers was called. But expiry
1436 * is related to the old base->clk value.
1437 */
1438 unsigned long baseclk = base->clk - 1;
1439
1440 while (!hlist_empty(head)) {
1441 struct timer_list *timer;
1442 void (*fn)(struct timer_list *);
1443
1444 timer = hlist_entry(head->first, struct timer_list, entry);
1445
1446 base->running_timer = timer;
1447 detach_timer(timer, true);
1448
1449 fn = timer->function;
1450
1451 if (timer->flags & TIMER_IRQSAFE) {
1452 raw_spin_unlock(&base->lock);
1453 call_timer_fn(timer, fn, baseclk);
1454 base->running_timer = NULL;
1455 raw_spin_lock(&base->lock);
1456 } else {
1457 raw_spin_unlock_irq(&base->lock);
1458 call_timer_fn(timer, fn, baseclk);
1459 base->running_timer = NULL;
1460 timer_sync_wait_running(base);
1461 raw_spin_lock_irq(&base->lock);
1462 }
1463 }
1464}
1465
1466static int collect_expired_timers(struct timer_base *base,
1467 struct hlist_head *heads)
1468{
1469 unsigned long clk = base->clk = base->next_expiry;
1470 struct hlist_head *vec;
1471 int i, levels = 0;
1472 unsigned int idx;
1473
1474 for (i = 0; i < LVL_DEPTH; i++) {
1475 idx = (clk & LVL_MASK) + i * LVL_SIZE;
1476
1477 if (__test_and_clear_bit(idx, base->pending_map)) {
1478 vec = base->vectors + idx;
1479 hlist_move_list(vec, heads++);
1480 levels++;
1481 }
1482 /* Is it time to look at the next level? */
1483 if (clk & LVL_CLK_MASK)
1484 break;
1485 /* Shift clock for the next level granularity */
1486 clk >>= LVL_CLK_SHIFT;
1487 }
1488 return levels;
1489}
1490
1491/*
1492 * Find the next pending bucket of a level. Search from level start (@offset)
1493 * + @clk upwards and if nothing there, search from start of the level
1494 * (@offset) up to @offset + clk.
1495 */
1496static int next_pending_bucket(struct timer_base *base, unsigned offset,
1497 unsigned clk)
1498{
1499 unsigned pos, start = offset + clk;
1500 unsigned end = offset + LVL_SIZE;
1501
1502 pos = find_next_bit(base->pending_map, end, start);
1503 if (pos < end)
1504 return pos - start;
1505
1506 pos = find_next_bit(base->pending_map, start, offset);
1507 return pos < start ? pos + LVL_SIZE - start : -1;
1508}
1509
1510/*
1511 * Search the first expiring timer in the various clock levels. Caller must
1512 * hold base->lock.
1513 */
1514static unsigned long __next_timer_interrupt(struct timer_base *base)
1515{
1516 unsigned long clk, next, adj;
1517 unsigned lvl, offset = 0;
1518
1519 next = base->clk + NEXT_TIMER_MAX_DELTA;
1520 clk = base->clk;
1521 for (lvl = 0; lvl < LVL_DEPTH; lvl++, offset += LVL_SIZE) {
1522 int pos = next_pending_bucket(base, offset, clk & LVL_MASK);
1523 unsigned long lvl_clk = clk & LVL_CLK_MASK;
1524
1525 if (pos >= 0) {
1526 unsigned long tmp = clk + (unsigned long) pos;
1527
1528 tmp <<= LVL_SHIFT(lvl);
1529 if (time_before(tmp, next))
1530 next = tmp;
1531
1532 /*
1533 * If the next expiration happens before we reach
1534 * the next level, no need to check further.
1535 */
1536 if (pos <= ((LVL_CLK_DIV - lvl_clk) & LVL_CLK_MASK))
1537 break;
1538 }
1539 /*
1540 * Clock for the next level. If the current level clock lower
1541 * bits are zero, we look at the next level as is. If not we
1542 * need to advance it by one because that's going to be the
1543 * next expiring bucket in that level. base->clk is the next
1544 * expiring jiffie. So in case of:
1545 *
1546 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1547 * 0 0 0 0 0 0
1548 *
1549 * we have to look at all levels @index 0. With
1550 *
1551 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1552 * 0 0 0 0 0 2
1553 *
1554 * LVL0 has the next expiring bucket @index 2. The upper
1555 * levels have the next expiring bucket @index 1.
1556 *
1557 * In case that the propagation wraps the next level the same
1558 * rules apply:
1559 *
1560 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1561 * 0 0 0 0 F 2
1562 *
1563 * So after looking at LVL0 we get:
1564 *
1565 * LVL5 LVL4 LVL3 LVL2 LVL1
1566 * 0 0 0 1 0
1567 *
1568 * So no propagation from LVL1 to LVL2 because that happened
1569 * with the add already, but then we need to propagate further
1570 * from LVL2 to LVL3.
1571 *
1572 * So the simple check whether the lower bits of the current
1573 * level are 0 or not is sufficient for all cases.
1574 */
1575 adj = lvl_clk ? 1 : 0;
1576 clk >>= LVL_CLK_SHIFT;
1577 clk += adj;
1578 }
1579
1580 base->next_expiry_recalc = false;
1581
1582 return next;
1583}
1584
1585#ifdef CONFIG_NO_HZ_COMMON
1586/*
1587 * Check, if the next hrtimer event is before the next timer wheel
1588 * event:
1589 */
1590static u64 cmp_next_hrtimer_event(u64 basem, u64 expires)
1591{
1592 u64 nextevt = hrtimer_get_next_event();
1593
1594 /*
1595 * If high resolution timers are enabled
1596 * hrtimer_get_next_event() returns KTIME_MAX.
1597 */
1598 if (expires <= nextevt)
1599 return expires;
1600
1601 /*
1602 * If the next timer is already expired, return the tick base
1603 * time so the tick is fired immediately.
1604 */
1605 if (nextevt <= basem)
1606 return basem;
1607
1608 /*
1609 * Round up to the next jiffie. High resolution timers are
1610 * off, so the hrtimers are expired in the tick and we need to
1611 * make sure that this tick really expires the timer to avoid
1612 * a ping pong of the nohz stop code.
1613 *
1614 * Use DIV_ROUND_UP_ULL to prevent gcc calling __divdi3
1615 */
1616 return DIV_ROUND_UP_ULL(nextevt, TICK_NSEC) * TICK_NSEC;
1617}
1618
1619/**
1620 * get_next_timer_interrupt - return the time (clock mono) of the next timer
1621 * @basej: base time jiffies
1622 * @basem: base time clock monotonic
1623 *
1624 * Returns the tick aligned clock monotonic time of the next pending
1625 * timer or KTIME_MAX if no timer is pending.
1626 */
1627u64 get_next_timer_interrupt(unsigned long basej, u64 basem)
1628{
1629 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1630 u64 expires = KTIME_MAX;
1631 unsigned long nextevt;
1632 bool is_max_delta;
1633
1634 /*
1635 * Pretend that there is no timer pending if the cpu is offline.
1636 * Possible pending timers will be migrated later to an active cpu.
1637 */
1638 if (cpu_is_offline(smp_processor_id()))
1639 return expires;
1640
1641 raw_spin_lock(&base->lock);
1642 if (base->next_expiry_recalc)
1643 base->next_expiry = __next_timer_interrupt(base);
1644 nextevt = base->next_expiry;
1645 is_max_delta = (nextevt == base->clk + NEXT_TIMER_MAX_DELTA);
1646
1647 /*
1648 * We have a fresh next event. Check whether we can forward the
1649 * base. We can only do that when @basej is past base->clk
1650 * otherwise we might rewind base->clk.
1651 */
1652 if (time_after(basej, base->clk)) {
1653 if (time_after(nextevt, basej))
1654 base->clk = basej;
1655 else if (time_after(nextevt, base->clk))
1656 base->clk = nextevt;
1657 }
1658
1659 if (time_before_eq(nextevt, basej)) {
1660 expires = basem;
1661 base->is_idle = false;
1662 } else {
1663 if (!is_max_delta)
1664 expires = basem + (u64)(nextevt - basej) * TICK_NSEC;
1665 /*
1666 * If we expect to sleep more than a tick, mark the base idle.
1667 * Also the tick is stopped so any added timer must forward
1668 * the base clk itself to keep granularity small. This idle
1669 * logic is only maintained for the BASE_STD base, deferrable
1670 * timers may still see large granularity skew (by design).
1671 */
1672 if ((expires - basem) > TICK_NSEC)
1673 base->is_idle = true;
1674 }
1675 raw_spin_unlock(&base->lock);
1676
1677 return cmp_next_hrtimer_event(basem, expires);
1678}
1679
1680/**
1681 * timer_clear_idle - Clear the idle state of the timer base
1682 *
1683 * Called with interrupts disabled
1684 */
1685void timer_clear_idle(void)
1686{
1687 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1688
1689 /*
1690 * We do this unlocked. The worst outcome is a remote enqueue sending
1691 * a pointless IPI, but taking the lock would just make the window for
1692 * sending the IPI a few instructions smaller for the cost of taking
1693 * the lock in the exit from idle path.
1694 */
1695 base->is_idle = false;
1696}
1697#endif
1698
1699/*
1700 * Called from the timer interrupt handler to charge one tick to the current
1701 * process. user_tick is 1 if the tick is user time, 0 for system.
1702 */
1703void update_process_times(int user_tick)
1704{
1705 struct task_struct *p = current;
1706
1707 /* Note: this timer irq context must be accounted for as well. */
1708 account_process_tick(p, user_tick);
1709 run_local_timers();
1710 rcu_sched_clock_irq(user_tick);
1711#ifdef CONFIG_IRQ_WORK
1712 if (in_irq())
1713 irq_work_tick();
1714#endif
1715 scheduler_tick();
1716 if (IS_ENABLED(CONFIG_POSIX_TIMERS))
1717 run_posix_cpu_timers();
1718
1719 /* The current CPU might make use of net randoms without receiving IRQs
1720 * to renew them often enough. Let's update the net_rand_state from a
1721 * non-constant value that's not affine to the number of calls to make
1722 * sure it's updated when there's some activity (we don't care in idle).
1723 */
1724 this_cpu_add(net_rand_state.s1, rol32(jiffies, 24) + user_tick);
1725}
1726
1727/**
1728 * __run_timers - run all expired timers (if any) on this CPU.
1729 * @base: the timer vector to be processed.
1730 */
1731static inline void __run_timers(struct timer_base *base)
1732{
1733 struct hlist_head heads[LVL_DEPTH];
1734 int levels;
1735
1736 if (time_before(jiffies, base->next_expiry))
1737 return;
1738
1739 timer_base_lock_expiry(base);
1740 raw_spin_lock_irq(&base->lock);
1741
1742 while (time_after_eq(jiffies, base->clk) &&
1743 time_after_eq(jiffies, base->next_expiry)) {
1744 levels = collect_expired_timers(base, heads);
1745 /*
1746 * The only possible reason for not finding any expired
1747 * timer at this clk is that all matching timers have been
1748 * dequeued.
1749 */
1750 WARN_ON_ONCE(!levels && !base->next_expiry_recalc);
1751 base->clk++;
1752 base->next_expiry = __next_timer_interrupt(base);
1753
1754 while (levels--)
1755 expire_timers(base, heads + levels);
1756 }
1757 raw_spin_unlock_irq(&base->lock);
1758 timer_base_unlock_expiry(base);
1759}
1760
1761/*
1762 * This function runs timers and the timer-tq in bottom half context.
1763 */
1764static __latent_entropy void run_timer_softirq(struct softirq_action *h)
1765{
1766 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1767
1768 __run_timers(base);
1769 if (IS_ENABLED(CONFIG_NO_HZ_COMMON))
1770 __run_timers(this_cpu_ptr(&timer_bases[BASE_DEF]));
1771}
1772
1773/*
1774 * Called by the local, per-CPU timer interrupt on SMP.
1775 */
1776void run_local_timers(void)
1777{
1778 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1779
1780 hrtimer_run_queues();
1781 /* Raise the softirq only if required. */
1782 if (time_before(jiffies, base->next_expiry)) {
1783 if (!IS_ENABLED(CONFIG_NO_HZ_COMMON))
1784 return;
1785 /* CPU is awake, so check the deferrable base. */
1786 base++;
1787 if (time_before(jiffies, base->next_expiry))
1788 return;
1789 }
1790 raise_softirq(TIMER_SOFTIRQ);
1791}
1792
1793/*
1794 * Since schedule_timeout()'s timer is defined on the stack, it must store
1795 * the target task on the stack as well.
1796 */
1797struct process_timer {
1798 struct timer_list timer;
1799 struct task_struct *task;
1800};
1801
1802static void process_timeout(struct timer_list *t)
1803{
1804 struct process_timer *timeout = from_timer(timeout, t, timer);
1805
1806 wake_up_process(timeout->task);
1807}
1808
1809/**
1810 * schedule_timeout - sleep until timeout
1811 * @timeout: timeout value in jiffies
1812 *
1813 * Make the current task sleep until @timeout jiffies have elapsed.
1814 * The function behavior depends on the current task state
1815 * (see also set_current_state() description):
1816 *
1817 * %TASK_RUNNING - the scheduler is called, but the task does not sleep
1818 * at all. That happens because sched_submit_work() does nothing for
1819 * tasks in %TASK_RUNNING state.
1820 *
1821 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1822 * pass before the routine returns unless the current task is explicitly
1823 * woken up, (e.g. by wake_up_process()).
1824 *
1825 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1826 * delivered to the current task or the current task is explicitly woken
1827 * up.
1828 *
1829 * The current task state is guaranteed to be %TASK_RUNNING when this
1830 * routine returns.
1831 *
1832 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1833 * the CPU away without a bound on the timeout. In this case the return
1834 * value will be %MAX_SCHEDULE_TIMEOUT.
1835 *
1836 * Returns 0 when the timer has expired otherwise the remaining time in
1837 * jiffies will be returned. In all cases the return value is guaranteed
1838 * to be non-negative.
1839 */
1840signed long __sched schedule_timeout(signed long timeout)
1841{
1842 struct process_timer timer;
1843 unsigned long expire;
1844
1845 switch (timeout)
1846 {
1847 case MAX_SCHEDULE_TIMEOUT:
1848 /*
1849 * These two special cases are useful to be comfortable
1850 * in the caller. Nothing more. We could take
1851 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1852 * but I' d like to return a valid offset (>=0) to allow
1853 * the caller to do everything it want with the retval.
1854 */
1855 schedule();
1856 goto out;
1857 default:
1858 /*
1859 * Another bit of PARANOID. Note that the retval will be
1860 * 0 since no piece of kernel is supposed to do a check
1861 * for a negative retval of schedule_timeout() (since it
1862 * should never happens anyway). You just have the printk()
1863 * that will tell you if something is gone wrong and where.
1864 */
1865 if (timeout < 0) {
1866 printk(KERN_ERR "schedule_timeout: wrong timeout "
1867 "value %lx\n", timeout);
1868 dump_stack();
1869 current->state = TASK_RUNNING;
1870 goto out;
1871 }
1872 }
1873
1874 expire = timeout + jiffies;
1875
1876 timer.task = current;
1877 timer_setup_on_stack(&timer.timer, process_timeout, 0);
1878 __mod_timer(&timer.timer, expire, MOD_TIMER_NOTPENDING);
1879 schedule();
1880 del_singleshot_timer_sync(&timer.timer);
1881
1882 /* Remove the timer from the object tracker */
1883 destroy_timer_on_stack(&timer.timer);
1884
1885 timeout = expire - jiffies;
1886
1887 out:
1888 return timeout < 0 ? 0 : timeout;
1889}
1890EXPORT_SYMBOL(schedule_timeout);
1891
1892/*
1893 * We can use __set_current_state() here because schedule_timeout() calls
1894 * schedule() unconditionally.
1895 */
1896signed long __sched schedule_timeout_interruptible(signed long timeout)
1897{
1898 __set_current_state(TASK_INTERRUPTIBLE);
1899 return schedule_timeout(timeout);
1900}
1901EXPORT_SYMBOL(schedule_timeout_interruptible);
1902
1903signed long __sched schedule_timeout_killable(signed long timeout)
1904{
1905 __set_current_state(TASK_KILLABLE);
1906 return schedule_timeout(timeout);
1907}
1908EXPORT_SYMBOL(schedule_timeout_killable);
1909
1910signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1911{
1912 __set_current_state(TASK_UNINTERRUPTIBLE);
1913 return schedule_timeout(timeout);
1914}
1915EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1916
1917/*
1918 * Like schedule_timeout_uninterruptible(), except this task will not contribute
1919 * to load average.
1920 */
1921signed long __sched schedule_timeout_idle(signed long timeout)
1922{
1923 __set_current_state(TASK_IDLE);
1924 return schedule_timeout(timeout);
1925}
1926EXPORT_SYMBOL(schedule_timeout_idle);
1927
1928#ifdef CONFIG_HOTPLUG_CPU
1929static void migrate_timer_list(struct timer_base *new_base, struct hlist_head *head)
1930{
1931 struct timer_list *timer;
1932 int cpu = new_base->cpu;
1933
1934 while (!hlist_empty(head)) {
1935 timer = hlist_entry(head->first, struct timer_list, entry);
1936 detach_timer(timer, false);
1937 timer->flags = (timer->flags & ~TIMER_BASEMASK) | cpu;
1938 internal_add_timer(new_base, timer);
1939 }
1940}
1941
1942int timers_prepare_cpu(unsigned int cpu)
1943{
1944 struct timer_base *base;
1945 int b;
1946
1947 for (b = 0; b < NR_BASES; b++) {
1948 base = per_cpu_ptr(&timer_bases[b], cpu);
1949 base->clk = jiffies;
1950 base->next_expiry = base->clk + NEXT_TIMER_MAX_DELTA;
1951 base->is_idle = false;
1952 }
1953 return 0;
1954}
1955
1956int timers_dead_cpu(unsigned int cpu)
1957{
1958 struct timer_base *old_base;
1959 struct timer_base *new_base;
1960 int b, i;
1961
1962 BUG_ON(cpu_online(cpu));
1963
1964 for (b = 0; b < NR_BASES; b++) {
1965 old_base = per_cpu_ptr(&timer_bases[b], cpu);
1966 new_base = get_cpu_ptr(&timer_bases[b]);
1967 /*
1968 * The caller is globally serialized and nobody else
1969 * takes two locks at once, deadlock is not possible.
1970 */
1971 raw_spin_lock_irq(&new_base->lock);
1972 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1973
1974 /*
1975 * The current CPUs base clock might be stale. Update it
1976 * before moving the timers over.
1977 */
1978 forward_timer_base(new_base);
1979
1980 BUG_ON(old_base->running_timer);
1981
1982 for (i = 0; i < WHEEL_SIZE; i++)
1983 migrate_timer_list(new_base, old_base->vectors + i);
1984
1985 raw_spin_unlock(&old_base->lock);
1986 raw_spin_unlock_irq(&new_base->lock);
1987 put_cpu_ptr(&timer_bases);
1988 }
1989 return 0;
1990}
1991
1992#endif /* CONFIG_HOTPLUG_CPU */
1993
1994static void __init init_timer_cpu(int cpu)
1995{
1996 struct timer_base *base;
1997 int i;
1998
1999 for (i = 0; i < NR_BASES; i++) {
2000 base = per_cpu_ptr(&timer_bases[i], cpu);
2001 base->cpu = cpu;
2002 raw_spin_lock_init(&base->lock);
2003 base->clk = jiffies;
2004 base->next_expiry = base->clk + NEXT_TIMER_MAX_DELTA;
2005 timer_base_init_expiry_lock(base);
2006 }
2007}
2008
2009static void __init init_timer_cpus(void)
2010{
2011 int cpu;
2012
2013 for_each_possible_cpu(cpu)
2014 init_timer_cpu(cpu);
2015}
2016
2017void __init init_timers(void)
2018{
2019 init_timer_cpus();
2020 posix_cputimers_init_work();
2021 open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
2022}
2023
2024/**
2025 * msleep - sleep safely even with waitqueue interruptions
2026 * @msecs: Time in milliseconds to sleep for
2027 */
2028void msleep(unsigned int msecs)
2029{
2030 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
2031
2032 while (timeout)
2033 timeout = schedule_timeout_uninterruptible(timeout);
2034}
2035
2036EXPORT_SYMBOL(msleep);
2037
2038/**
2039 * msleep_interruptible - sleep waiting for signals
2040 * @msecs: Time in milliseconds to sleep for
2041 */
2042unsigned long msleep_interruptible(unsigned int msecs)
2043{
2044 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
2045
2046 while (timeout && !signal_pending(current))
2047 timeout = schedule_timeout_interruptible(timeout);
2048 return jiffies_to_msecs(timeout);
2049}
2050
2051EXPORT_SYMBOL(msleep_interruptible);
2052
2053/**
2054 * usleep_range - Sleep for an approximate time
2055 * @min: Minimum time in usecs to sleep
2056 * @max: Maximum time in usecs to sleep
2057 *
2058 * In non-atomic context where the exact wakeup time is flexible, use
2059 * usleep_range() instead of udelay(). The sleep improves responsiveness
2060 * by avoiding the CPU-hogging busy-wait of udelay(), and the range reduces
2061 * power usage by allowing hrtimers to take advantage of an already-
2062 * scheduled interrupt instead of scheduling a new one just for this sleep.
2063 */
2064void __sched usleep_range(unsigned long min, unsigned long max)
2065{
2066 ktime_t exp = ktime_add_us(ktime_get(), min);
2067 u64 delta = (u64)(max - min) * NSEC_PER_USEC;
2068
2069 for (;;) {
2070 __set_current_state(TASK_UNINTERRUPTIBLE);
2071 /* Do not return before the requested sleep time has elapsed */
2072 if (!schedule_hrtimeout_range(&exp, delta, HRTIMER_MODE_ABS))
2073 break;
2074 }
2075}
2076EXPORT_SYMBOL(usleep_range);