xref: /linux/kernel/time/timer.c (revision 8c749ce93ee69e789e46b3be98de9e0cbfcf8ed8)
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 
59 EXPORT_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 
72 struct tvec {
73 	struct hlist_head vec[TVN_SIZE];
74 };
75 
76 struct tvec_root {
77 	struct hlist_head vec[TVR_SIZE];
78 };
79 
80 struct 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 
98 static DEFINE_PER_CPU(struct tvec_base, tvec_bases);
99 
100 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
101 unsigned int sysctl_timer_migration = 1;
102 
103 void 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 
122 int 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 
137 static 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
145 static 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 
152 static 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  */
212 unsigned long __round_jiffies(unsigned long j, int cpu)
213 {
214 	return round_jiffies_common(j, cpu, false);
215 }
216 EXPORT_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  */
238 unsigned 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 }
245 EXPORT_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  */
262 unsigned long round_jiffies(unsigned long j)
263 {
264 	return round_jiffies_common(j, raw_smp_processor_id(), false);
265 }
266 EXPORT_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  */
283 unsigned long round_jiffies_relative(unsigned long j)
284 {
285 	return __round_jiffies_relative(j, raw_smp_processor_id());
286 }
287 EXPORT_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  */
299 unsigned long __round_jiffies_up(unsigned long j, int cpu)
300 {
301 	return round_jiffies_common(j, cpu, true);
302 }
303 EXPORT_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  */
315 unsigned 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 }
322 EXPORT_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  */
333 unsigned long round_jiffies_up(unsigned long j)
334 {
335 	return round_jiffies_common(j, raw_smp_processor_id(), true);
336 }
337 EXPORT_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  */
348 unsigned long round_jiffies_up_relative(unsigned long j)
349 {
350 	return __round_jiffies_up_relative(j, raw_smp_processor_id());
351 }
352 EXPORT_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  */
367 void set_timer_slack(struct timer_list *timer, int slack_hz)
368 {
369 	timer->slack = slack_hz;
370 }
371 EXPORT_SYMBOL_GPL(set_timer_slack);
372 
373 static 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 
415 static 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
452 void __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 
462 static 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
480 static void timer_stats_account_timer(struct timer_list *timer) {}
481 #endif
482 
483 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
484 
485 static struct debug_obj_descr timer_debug_descr;
486 
487 static 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  */
496 static 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. */
511 static 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  */
521 static 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  */
556 static 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  */
574 static 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 
597 static 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 
606 static inline void debug_timer_init(struct timer_list *timer)
607 {
608 	debug_object_init(timer, &timer_debug_descr);
609 }
610 
611 static inline void debug_timer_activate(struct timer_list *timer)
612 {
613 	debug_object_activate(timer, &timer_debug_descr);
614 }
615 
616 static inline void debug_timer_deactivate(struct timer_list *timer)
617 {
618 	debug_object_deactivate(timer, &timer_debug_descr);
619 }
620 
621 static inline void debug_timer_free(struct timer_list *timer)
622 {
623 	debug_object_free(timer, &timer_debug_descr);
624 }
625 
626 static inline void debug_timer_assert_init(struct timer_list *timer)
627 {
628 	debug_object_assert_init(timer, &timer_debug_descr);
629 }
630 
631 static void do_init_timer(struct timer_list *timer, unsigned int flags,
632 			  const char *name, struct lock_class_key *key);
633 
634 void 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 }
640 EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
641 
642 void destroy_timer_on_stack(struct timer_list *timer)
643 {
644 	debug_object_free(timer, &timer_debug_descr);
645 }
646 EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
647 
648 #else
649 static inline void debug_timer_init(struct timer_list *timer) { }
650 static inline void debug_timer_activate(struct timer_list *timer) { }
651 static inline void debug_timer_deactivate(struct timer_list *timer) { }
652 static inline void debug_timer_assert_init(struct timer_list *timer) { }
653 #endif
654 
655 static inline void debug_init(struct timer_list *timer)
656 {
657 	debug_timer_init(timer);
658 	trace_timer_init(timer);
659 }
660 
661 static inline void
662 debug_activate(struct timer_list *timer, unsigned long expires)
663 {
664 	debug_timer_activate(timer);
665 	trace_timer_start(timer, expires, timer->flags);
666 }
667 
668 static inline void debug_deactivate(struct timer_list *timer)
669 {
670 	debug_timer_deactivate(timer);
671 	trace_timer_cancel(timer);
672 }
673 
674 static inline void debug_assert_init(struct timer_list *timer)
675 {
676 	debug_timer_assert_init(timer);
677 }
678 
679 static 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  */
704 void 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 }
710 EXPORT_SYMBOL(init_timer_key);
711 
712 static 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 
724 static inline void
725 detach_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 
733 static 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  */
762 static 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 
781 static 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 
825 out_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  */
841 int mod_timer_pending(struct timer_list *timer, unsigned long expires)
842 {
843 	return __mod_timer(timer, expires, true, TIMER_NOT_PINNED);
844 }
845 EXPORT_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  */
857 static inline
858 unsigned 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  */
906 int 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 }
920 EXPORT_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  */
941 int 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 }
948 EXPORT_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  */
964 void add_timer(struct timer_list *timer)
965 {
966 	BUG_ON(timer_pending(timer));
967 	mod_timer(timer, timer->expires);
968 }
969 EXPORT_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  */
978 void 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 }
1007 EXPORT_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  */
1020 int 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 }
1037 EXPORT_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  */
1046 int 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 }
1064 EXPORT_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  */
1103 int 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 }
1129 EXPORT_SYMBOL(del_timer_sync);
1130 #endif
1131 
1132 static 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 
1153 static 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  */
1205 static 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  */
1269 static 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 
1294 cascade:
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  */
1343 static 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  */
1380 u64 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  */
1413 void 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  */
1432 static 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  */
1443 void 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  */
1455 SYSCALL_DEFINE1(alarm, unsigned int, seconds)
1456 {
1457 	return alarm_setitimer(seconds);
1458 }
1459 
1460 #endif
1461 
1462 static 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  */
1493 signed 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 }
1542 EXPORT_SYMBOL(schedule_timeout);
1543 
1544 /*
1545  * We can use __set_current_state() here because schedule_timeout() calls
1546  * schedule() unconditionally.
1547  */
1548 signed long __sched schedule_timeout_interruptible(signed long timeout)
1549 {
1550 	__set_current_state(TASK_INTERRUPTIBLE);
1551 	return schedule_timeout(timeout);
1552 }
1553 EXPORT_SYMBOL(schedule_timeout_interruptible);
1554 
1555 signed long __sched schedule_timeout_killable(signed long timeout)
1556 {
1557 	__set_current_state(TASK_KILLABLE);
1558 	return schedule_timeout(timeout);
1559 }
1560 EXPORT_SYMBOL(schedule_timeout_killable);
1561 
1562 signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1563 {
1564 	__set_current_state(TASK_UNINTERRUPTIBLE);
1565 	return schedule_timeout(timeout);
1566 }
1567 EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1568 
1569 #ifdef CONFIG_HOTPLUG_CPU
1570 static void migrate_timer_list(struct tvec_base *new_base, struct hlist_head *head)
1571 {
1572 	struct timer_list *timer;
1573 	int cpu = new_base->cpu;
1574 
1575 	while (!hlist_empty(head)) {
1576 		timer = hlist_entry(head->first, struct timer_list, entry);
1577 		/* We ignore the accounting on the dying cpu */
1578 		detach_timer(timer, false);
1579 		timer->flags = (timer->flags & ~TIMER_BASEMASK) | cpu;
1580 		internal_add_timer(new_base, timer);
1581 	}
1582 }
1583 
1584 static void migrate_timers(int cpu)
1585 {
1586 	struct tvec_base *old_base;
1587 	struct tvec_base *new_base;
1588 	int i;
1589 
1590 	BUG_ON(cpu_online(cpu));
1591 	old_base = per_cpu_ptr(&tvec_bases, cpu);
1592 	new_base = get_cpu_ptr(&tvec_bases);
1593 	/*
1594 	 * The caller is globally serialized and nobody else
1595 	 * takes two locks at once, deadlock is not possible.
1596 	 */
1597 	spin_lock_irq(&new_base->lock);
1598 	spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1599 
1600 	BUG_ON(old_base->running_timer);
1601 
1602 	for (i = 0; i < TVR_SIZE; i++)
1603 		migrate_timer_list(new_base, old_base->tv1.vec + i);
1604 	for (i = 0; i < TVN_SIZE; i++) {
1605 		migrate_timer_list(new_base, old_base->tv2.vec + i);
1606 		migrate_timer_list(new_base, old_base->tv3.vec + i);
1607 		migrate_timer_list(new_base, old_base->tv4.vec + i);
1608 		migrate_timer_list(new_base, old_base->tv5.vec + i);
1609 	}
1610 
1611 	old_base->active_timers = 0;
1612 	old_base->all_timers = 0;
1613 
1614 	spin_unlock(&old_base->lock);
1615 	spin_unlock_irq(&new_base->lock);
1616 	put_cpu_ptr(&tvec_bases);
1617 }
1618 
1619 static int timer_cpu_notify(struct notifier_block *self,
1620 				unsigned long action, void *hcpu)
1621 {
1622 	switch (action) {
1623 	case CPU_DEAD:
1624 	case CPU_DEAD_FROZEN:
1625 		migrate_timers((long)hcpu);
1626 		break;
1627 	default:
1628 		break;
1629 	}
1630 
1631 	return NOTIFY_OK;
1632 }
1633 
1634 static inline void timer_register_cpu_notifier(void)
1635 {
1636 	cpu_notifier(timer_cpu_notify, 0);
1637 }
1638 #else
1639 static inline void timer_register_cpu_notifier(void) { }
1640 #endif /* CONFIG_HOTPLUG_CPU */
1641 
1642 static void __init init_timer_cpu(int cpu)
1643 {
1644 	struct tvec_base *base = per_cpu_ptr(&tvec_bases, cpu);
1645 
1646 	base->cpu = cpu;
1647 	spin_lock_init(&base->lock);
1648 
1649 	base->timer_jiffies = jiffies;
1650 	base->next_timer = base->timer_jiffies;
1651 }
1652 
1653 static void __init init_timer_cpus(void)
1654 {
1655 	int cpu;
1656 
1657 	for_each_possible_cpu(cpu)
1658 		init_timer_cpu(cpu);
1659 }
1660 
1661 void __init init_timers(void)
1662 {
1663 	init_timer_cpus();
1664 	init_timer_stats();
1665 	timer_register_cpu_notifier();
1666 	open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
1667 }
1668 
1669 /**
1670  * msleep - sleep safely even with waitqueue interruptions
1671  * @msecs: Time in milliseconds to sleep for
1672  */
1673 void msleep(unsigned int msecs)
1674 {
1675 	unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1676 
1677 	while (timeout)
1678 		timeout = schedule_timeout_uninterruptible(timeout);
1679 }
1680 
1681 EXPORT_SYMBOL(msleep);
1682 
1683 /**
1684  * msleep_interruptible - sleep waiting for signals
1685  * @msecs: Time in milliseconds to sleep for
1686  */
1687 unsigned long msleep_interruptible(unsigned int msecs)
1688 {
1689 	unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1690 
1691 	while (timeout && !signal_pending(current))
1692 		timeout = schedule_timeout_interruptible(timeout);
1693 	return jiffies_to_msecs(timeout);
1694 }
1695 
1696 EXPORT_SYMBOL(msleep_interruptible);
1697 
1698 static void __sched do_usleep_range(unsigned long min, unsigned long max)
1699 {
1700 	ktime_t kmin;
1701 	unsigned long delta;
1702 
1703 	kmin = ktime_set(0, min * NSEC_PER_USEC);
1704 	delta = (max - min) * NSEC_PER_USEC;
1705 	schedule_hrtimeout_range(&kmin, delta, HRTIMER_MODE_REL);
1706 }
1707 
1708 /**
1709  * usleep_range - Drop in replacement for udelay where wakeup is flexible
1710  * @min: Minimum time in usecs to sleep
1711  * @max: Maximum time in usecs to sleep
1712  */
1713 void __sched usleep_range(unsigned long min, unsigned long max)
1714 {
1715 	__set_current_state(TASK_UNINTERRUPTIBLE);
1716 	do_usleep_range(min, max);
1717 }
1718 EXPORT_SYMBOL(usleep_range);
1719