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#define CREATE_TRACE_POINTS
53#include <trace/events/timer.h>
54
55__visible u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
56
57EXPORT_SYMBOL(jiffies_64);
58
59/*
60 * per-CPU timer vector definitions:
61 */
62#define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6)
63#define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8)
64#define TVN_SIZE (1 << TVN_BITS)
65#define TVR_SIZE (1 << TVR_BITS)
66#define TVN_MASK (TVN_SIZE - 1)
67#define TVR_MASK (TVR_SIZE - 1)
68#define MAX_TVAL ((unsigned long)((1ULL << (TVR_BITS + 4*TVN_BITS)) - 1))
69
70struct tvec {
71	struct list_head vec[TVN_SIZE];
72};
73
74struct tvec_root {
75	struct list_head vec[TVR_SIZE];
76};
77
78struct tvec_base {
79	spinlock_t lock;
80	struct timer_list *running_timer;
81	unsigned long timer_jiffies;
82	unsigned long next_timer;
83	unsigned long active_timers;
84	unsigned long all_timers;
85	int cpu;
86	struct tvec_root tv1;
87	struct tvec tv2;
88	struct tvec tv3;
89	struct tvec tv4;
90	struct tvec tv5;
91} ____cacheline_aligned;
92
93struct tvec_base boot_tvec_bases;
94EXPORT_SYMBOL(boot_tvec_bases);
95static DEFINE_PER_CPU(struct tvec_base *, tvec_bases) = &boot_tvec_bases;
96
97/* Functions below help us manage 'deferrable' flag */
98static inline unsigned int tbase_get_deferrable(struct tvec_base *base)
99{
100	return ((unsigned int)(unsigned long)base & TIMER_DEFERRABLE);
101}
102
103static inline unsigned int tbase_get_irqsafe(struct tvec_base *base)
104{
105	return ((unsigned int)(unsigned long)base & TIMER_IRQSAFE);
106}
107
108static inline struct tvec_base *tbase_get_base(struct tvec_base *base)
109{
110	return ((struct tvec_base *)((unsigned long)base & ~TIMER_FLAG_MASK));
111}
112
113static inline void
114timer_set_base(struct timer_list *timer, struct tvec_base *new_base)
115{
116	unsigned long flags = (unsigned long)timer->base & TIMER_FLAG_MASK;
117
118	timer->base = (struct tvec_base *)((unsigned long)(new_base) | flags);
119}
120
121static unsigned long round_jiffies_common(unsigned long j, int cpu,
122		bool force_up)
123{
124	int rem;
125	unsigned long original = j;
126
127	/*
128	 * We don't want all cpus firing their timers at once hitting the
129	 * same lock or cachelines, so we skew each extra cpu with an extra
130	 * 3 jiffies. This 3 jiffies came originally from the mm/ code which
131	 * already did this.
132	 * The skew is done by adding 3*cpunr, then round, then subtract this
133	 * extra offset again.
134	 */
135	j += cpu * 3;
136
137	rem = j % HZ;
138
139	/*
140	 * If the target jiffie is just after a whole second (which can happen
141	 * due to delays of the timer irq, long irq off times etc etc) then
142	 * we should round down to the whole second, not up. Use 1/4th second
143	 * as cutoff for this rounding as an extreme upper bound for this.
144	 * But never round down if @force_up is set.
145	 */
146	if (rem < HZ/4 && !force_up) /* round down */
147		j = j - rem;
148	else /* round up */
149		j = j - rem + HZ;
150
151	/* now that we have rounded, subtract the extra skew again */
152	j -= cpu * 3;
153
154	/*
155	 * Make sure j is still in the future. Otherwise return the
156	 * unmodified value.
157	 */
158	return time_is_after_jiffies(j) ? j : original;
159}
160
161/**
162 * __round_jiffies - function to round jiffies to a full second
163 * @j: the time in (absolute) jiffies that should be rounded
164 * @cpu: the processor number on which the timeout will happen
165 *
166 * __round_jiffies() rounds an absolute time in the future (in jiffies)
167 * up or down to (approximately) full seconds. This is useful for timers
168 * for which the exact time they fire does not matter too much, as long as
169 * they fire approximately every X seconds.
170 *
171 * By rounding these timers to whole seconds, all such timers will fire
172 * at the same time, rather than at various times spread out. The goal
173 * of this is to have the CPU wake up less, which saves power.
174 *
175 * The exact rounding is skewed for each processor to avoid all
176 * processors firing at the exact same time, which could lead
177 * to lock contention or spurious cache line bouncing.
178 *
179 * The return value is the rounded version of the @j parameter.
180 */
181unsigned long __round_jiffies(unsigned long j, int cpu)
182{
183	return round_jiffies_common(j, cpu, false);
184}
185EXPORT_SYMBOL_GPL(__round_jiffies);
186
187/**
188 * __round_jiffies_relative - function to round jiffies to a full second
189 * @j: the time in (relative) jiffies that should be rounded
190 * @cpu: the processor number on which the timeout will happen
191 *
192 * __round_jiffies_relative() rounds a time delta  in the future (in jiffies)
193 * up or down to (approximately) full seconds. This is useful for timers
194 * for which the exact time they fire does not matter too much, as long as
195 * they fire approximately every X seconds.
196 *
197 * By rounding these timers to whole seconds, all such timers will fire
198 * at the same time, rather than at various times spread out. The goal
199 * of this is to have the CPU wake up less, which saves power.
200 *
201 * The exact rounding is skewed for each processor to avoid all
202 * processors firing at the exact same time, which could lead
203 * to lock contention or spurious cache line bouncing.
204 *
205 * The return value is the rounded version of the @j parameter.
206 */
207unsigned long __round_jiffies_relative(unsigned long j, int cpu)
208{
209	unsigned long j0 = jiffies;
210
211	/* Use j0 because jiffies might change while we run */
212	return round_jiffies_common(j + j0, cpu, false) - j0;
213}
214EXPORT_SYMBOL_GPL(__round_jiffies_relative);
215
216/**
217 * round_jiffies - function to round jiffies to a full second
218 * @j: the time in (absolute) jiffies that should be rounded
219 *
220 * round_jiffies() rounds an absolute time in the future (in jiffies)
221 * up or down to (approximately) full seconds. This is useful for timers
222 * for which the exact time they fire does not matter too much, as long as
223 * they fire approximately every X seconds.
224 *
225 * By rounding these timers to whole seconds, all such timers will fire
226 * at the same time, rather than at various times spread out. The goal
227 * of this is to have the CPU wake up less, which saves power.
228 *
229 * The return value is the rounded version of the @j parameter.
230 */
231unsigned long round_jiffies(unsigned long j)
232{
233	return round_jiffies_common(j, raw_smp_processor_id(), false);
234}
235EXPORT_SYMBOL_GPL(round_jiffies);
236
237/**
238 * round_jiffies_relative - function to round jiffies to a full second
239 * @j: the time in (relative) jiffies that should be rounded
240 *
241 * round_jiffies_relative() rounds a time delta  in the future (in jiffies)
242 * up or down to (approximately) full seconds. This is useful for timers
243 * for which the exact time they fire does not matter too much, as long as
244 * they fire approximately every X seconds.
245 *
246 * By rounding these timers to whole seconds, all such timers will fire
247 * at the same time, rather than at various times spread out. The goal
248 * of this is to have the CPU wake up less, which saves power.
249 *
250 * The return value is the rounded version of the @j parameter.
251 */
252unsigned long round_jiffies_relative(unsigned long j)
253{
254	return __round_jiffies_relative(j, raw_smp_processor_id());
255}
256EXPORT_SYMBOL_GPL(round_jiffies_relative);
257
258/**
259 * __round_jiffies_up - function to round jiffies up to a full second
260 * @j: the time in (absolute) jiffies that should be rounded
261 * @cpu: the processor number on which the timeout will happen
262 *
263 * This is the same as __round_jiffies() except that it will never
264 * round down.  This is useful for timeouts for which the exact time
265 * of firing does not matter too much, as long as they don't fire too
266 * early.
267 */
268unsigned long __round_jiffies_up(unsigned long j, int cpu)
269{
270	return round_jiffies_common(j, cpu, true);
271}
272EXPORT_SYMBOL_GPL(__round_jiffies_up);
273
274/**
275 * __round_jiffies_up_relative - function to round jiffies up to a full second
276 * @j: the time in (relative) jiffies that should be rounded
277 * @cpu: the processor number on which the timeout will happen
278 *
279 * This is the same as __round_jiffies_relative() except that it will never
280 * round down.  This is useful for timeouts for which the exact time
281 * of firing does not matter too much, as long as they don't fire too
282 * early.
283 */
284unsigned long __round_jiffies_up_relative(unsigned long j, int cpu)
285{
286	unsigned long j0 = jiffies;
287
288	/* Use j0 because jiffies might change while we run */
289	return round_jiffies_common(j + j0, cpu, true) - j0;
290}
291EXPORT_SYMBOL_GPL(__round_jiffies_up_relative);
292
293/**
294 * round_jiffies_up - function to round jiffies up to a full second
295 * @j: the time in (absolute) jiffies that should be rounded
296 *
297 * This is the same as round_jiffies() except that it will never
298 * round down.  This is useful for timeouts for which the exact time
299 * of firing does not matter too much, as long as they don't fire too
300 * early.
301 */
302unsigned long round_jiffies_up(unsigned long j)
303{
304	return round_jiffies_common(j, raw_smp_processor_id(), true);
305}
306EXPORT_SYMBOL_GPL(round_jiffies_up);
307
308/**
309 * round_jiffies_up_relative - function to round jiffies up to a full second
310 * @j: the time in (relative) jiffies that should be rounded
311 *
312 * This is the same as round_jiffies_relative() except that it will never
313 * round down.  This is useful for timeouts for which the exact time
314 * of firing does not matter too much, as long as they don't fire too
315 * early.
316 */
317unsigned long round_jiffies_up_relative(unsigned long j)
318{
319	return __round_jiffies_up_relative(j, raw_smp_processor_id());
320}
321EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
322
323/**
324 * set_timer_slack - set the allowed slack for a timer
325 * @timer: the timer to be modified
326 * @slack_hz: the amount of time (in jiffies) allowed for rounding
327 *
328 * Set the amount of time, in jiffies, that a certain timer has
329 * in terms of slack. By setting this value, the timer subsystem
330 * will schedule the actual timer somewhere between
331 * the time mod_timer() asks for, and that time plus the slack.
332 *
333 * By setting the slack to -1, a percentage of the delay is used
334 * instead.
335 */
336void set_timer_slack(struct timer_list *timer, int slack_hz)
337{
338	timer->slack = slack_hz;
339}
340EXPORT_SYMBOL_GPL(set_timer_slack);
341
342/*
343 * If the list is empty, catch up ->timer_jiffies to the current time.
344 * The caller must hold the tvec_base lock.  Returns true if the list
345 * was empty and therefore ->timer_jiffies was updated.
346 */
347static bool catchup_timer_jiffies(struct tvec_base *base)
348{
349	if (!base->all_timers) {
350		base->timer_jiffies = jiffies;
351		return true;
352	}
353	return false;
354}
355
356static void
357__internal_add_timer(struct tvec_base *base, struct timer_list *timer)
358{
359	unsigned long expires = timer->expires;
360	unsigned long idx = expires - base->timer_jiffies;
361	struct list_head *vec;
362
363	if (idx < TVR_SIZE) {
364		int i = expires & TVR_MASK;
365		vec = base->tv1.vec + i;
366	} else if (idx < 1 << (TVR_BITS + TVN_BITS)) {
367		int i = (expires >> TVR_BITS) & TVN_MASK;
368		vec = base->tv2.vec + i;
369	} else if (idx < 1 << (TVR_BITS + 2 * TVN_BITS)) {
370		int i = (expires >> (TVR_BITS + TVN_BITS)) & TVN_MASK;
371		vec = base->tv3.vec + i;
372	} else if (idx < 1 << (TVR_BITS + 3 * TVN_BITS)) {
373		int i = (expires >> (TVR_BITS + 2 * TVN_BITS)) & TVN_MASK;
374		vec = base->tv4.vec + i;
375	} else if ((signed long) idx < 0) {
376		/*
377		 * Can happen if you add a timer with expires == jiffies,
378		 * or you set a timer to go off in the past
379		 */
380		vec = base->tv1.vec + (base->timer_jiffies & TVR_MASK);
381	} else {
382		int i;
383		/* If the timeout is larger than MAX_TVAL (on 64-bit
384		 * architectures or with CONFIG_BASE_SMALL=1) then we
385		 * use the maximum timeout.
386		 */
387		if (idx > MAX_TVAL) {
388			idx = MAX_TVAL;
389			expires = idx + base->timer_jiffies;
390		}
391		i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK;
392		vec = base->tv5.vec + i;
393	}
394	/*
395	 * Timers are FIFO:
396	 */
397	list_add_tail(&timer->entry, vec);
398}
399
400static void internal_add_timer(struct tvec_base *base, struct timer_list *timer)
401{
402	(void)catchup_timer_jiffies(base);
403	__internal_add_timer(base, timer);
404	/*
405	 * Update base->active_timers and base->next_timer
406	 */
407	if (!tbase_get_deferrable(timer->base)) {
408		if (!base->active_timers++ ||
409		    time_before(timer->expires, base->next_timer))
410			base->next_timer = timer->expires;
411	}
412	base->all_timers++;
413
414	/*
415	 * Check whether the other CPU is in dynticks mode and needs
416	 * to be triggered to reevaluate the timer wheel.
417	 * We are protected against the other CPU fiddling
418	 * with the timer by holding the timer base lock. This also
419	 * makes sure that a CPU on the way to stop its tick can not
420	 * evaluate the timer wheel.
421	 *
422	 * Spare the IPI for deferrable timers on idle targets though.
423	 * The next busy ticks will take care of it. Except full dynticks
424	 * require special care against races with idle_cpu(), lets deal
425	 * with that later.
426	 */
427	if (!tbase_get_deferrable(base) || tick_nohz_full_cpu(base->cpu))
428		wake_up_nohz_cpu(base->cpu);
429}
430
431#ifdef CONFIG_TIMER_STATS
432void __timer_stats_timer_set_start_info(struct timer_list *timer, void *addr)
433{
434	if (timer->start_site)
435		return;
436
437	timer->start_site = addr;
438	memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
439	timer->start_pid = current->pid;
440}
441
442static void timer_stats_account_timer(struct timer_list *timer)
443{
444	unsigned int flag = 0;
445
446	if (likely(!timer->start_site))
447		return;
448	if (unlikely(tbase_get_deferrable(timer->base)))
449		flag |= TIMER_STATS_FLAG_DEFERRABLE;
450
451	timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
452				 timer->function, timer->start_comm, flag);
453}
454
455#else
456static void timer_stats_account_timer(struct timer_list *timer) {}
457#endif
458
459#ifdef CONFIG_DEBUG_OBJECTS_TIMERS
460
461static struct debug_obj_descr timer_debug_descr;
462
463static void *timer_debug_hint(void *addr)
464{
465	return ((struct timer_list *) addr)->function;
466}
467
468/*
469 * fixup_init is called when:
470 * - an active object is initialized
471 */
472static int timer_fixup_init(void *addr, enum debug_obj_state state)
473{
474	struct timer_list *timer = addr;
475
476	switch (state) {
477	case ODEBUG_STATE_ACTIVE:
478		del_timer_sync(timer);
479		debug_object_init(timer, &timer_debug_descr);
480		return 1;
481	default:
482		return 0;
483	}
484}
485
486/* Stub timer callback for improperly used timers. */
487static void stub_timer(unsigned long data)
488{
489	WARN_ON(1);
490}
491
492/*
493 * fixup_activate is called when:
494 * - an active object is activated
495 * - an unknown object is activated (might be a statically initialized object)
496 */
497static int timer_fixup_activate(void *addr, enum debug_obj_state state)
498{
499	struct timer_list *timer = addr;
500
501	switch (state) {
502
503	case ODEBUG_STATE_NOTAVAILABLE:
504		/*
505		 * This is not really a fixup. The timer was
506		 * statically initialized. We just make sure that it
507		 * is tracked in the object tracker.
508		 */
509		if (timer->entry.next == NULL &&
510		    timer->entry.prev == TIMER_ENTRY_STATIC) {
511			debug_object_init(timer, &timer_debug_descr);
512			debug_object_activate(timer, &timer_debug_descr);
513			return 0;
514		} else {
515			setup_timer(timer, stub_timer, 0);
516			return 1;
517		}
518		return 0;
519
520	case ODEBUG_STATE_ACTIVE:
521		WARN_ON(1);
522
523	default:
524		return 0;
525	}
526}
527
528/*
529 * fixup_free is called when:
530 * - an active object is freed
531 */
532static int timer_fixup_free(void *addr, enum debug_obj_state state)
533{
534	struct timer_list *timer = addr;
535
536	switch (state) {
537	case ODEBUG_STATE_ACTIVE:
538		del_timer_sync(timer);
539		debug_object_free(timer, &timer_debug_descr);
540		return 1;
541	default:
542		return 0;
543	}
544}
545
546/*
547 * fixup_assert_init is called when:
548 * - an untracked/uninit-ed object is found
549 */
550static int timer_fixup_assert_init(void *addr, enum debug_obj_state state)
551{
552	struct timer_list *timer = addr;
553
554	switch (state) {
555	case ODEBUG_STATE_NOTAVAILABLE:
556		if (timer->entry.prev == TIMER_ENTRY_STATIC) {
557			/*
558			 * This is not really a fixup. The timer was
559			 * statically initialized. We just make sure that it
560			 * is tracked in the object tracker.
561			 */
562			debug_object_init(timer, &timer_debug_descr);
563			return 0;
564		} else {
565			setup_timer(timer, stub_timer, 0);
566			return 1;
567		}
568	default:
569		return 0;
570	}
571}
572
573static struct debug_obj_descr timer_debug_descr = {
574	.name			= "timer_list",
575	.debug_hint		= timer_debug_hint,
576	.fixup_init		= timer_fixup_init,
577	.fixup_activate		= timer_fixup_activate,
578	.fixup_free		= timer_fixup_free,
579	.fixup_assert_init	= timer_fixup_assert_init,
580};
581
582static inline void debug_timer_init(struct timer_list *timer)
583{
584	debug_object_init(timer, &timer_debug_descr);
585}
586
587static inline void debug_timer_activate(struct timer_list *timer)
588{
589	debug_object_activate(timer, &timer_debug_descr);
590}
591
592static inline void debug_timer_deactivate(struct timer_list *timer)
593{
594	debug_object_deactivate(timer, &timer_debug_descr);
595}
596
597static inline void debug_timer_free(struct timer_list *timer)
598{
599	debug_object_free(timer, &timer_debug_descr);
600}
601
602static inline void debug_timer_assert_init(struct timer_list *timer)
603{
604	debug_object_assert_init(timer, &timer_debug_descr);
605}
606
607static void do_init_timer(struct timer_list *timer, unsigned int flags,
608			  const char *name, struct lock_class_key *key);
609
610void init_timer_on_stack_key(struct timer_list *timer, unsigned int flags,
611			     const char *name, struct lock_class_key *key)
612{
613	debug_object_init_on_stack(timer, &timer_debug_descr);
614	do_init_timer(timer, flags, name, key);
615}
616EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
617
618void destroy_timer_on_stack(struct timer_list *timer)
619{
620	debug_object_free(timer, &timer_debug_descr);
621}
622EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
623
624#else
625static inline void debug_timer_init(struct timer_list *timer) { }
626static inline void debug_timer_activate(struct timer_list *timer) { }
627static inline void debug_timer_deactivate(struct timer_list *timer) { }
628static inline void debug_timer_assert_init(struct timer_list *timer) { }
629#endif
630
631static inline void debug_init(struct timer_list *timer)
632{
633	debug_timer_init(timer);
634	trace_timer_init(timer);
635}
636
637static inline void
638debug_activate(struct timer_list *timer, unsigned long expires)
639{
640	debug_timer_activate(timer);
641	trace_timer_start(timer, expires);
642}
643
644static inline void debug_deactivate(struct timer_list *timer)
645{
646	debug_timer_deactivate(timer);
647	trace_timer_cancel(timer);
648}
649
650static inline void debug_assert_init(struct timer_list *timer)
651{
652	debug_timer_assert_init(timer);
653}
654
655static void do_init_timer(struct timer_list *timer, unsigned int flags,
656			  const char *name, struct lock_class_key *key)
657{
658	struct tvec_base *base = raw_cpu_read(tvec_bases);
659
660	timer->entry.next = NULL;
661	timer->base = (void *)((unsigned long)base | flags);
662	timer->slack = -1;
663#ifdef CONFIG_TIMER_STATS
664	timer->start_site = NULL;
665	timer->start_pid = -1;
666	memset(timer->start_comm, 0, TASK_COMM_LEN);
667#endif
668	lockdep_init_map(&timer->lockdep_map, name, key, 0);
669}
670
671/**
672 * init_timer_key - initialize a timer
673 * @timer: the timer to be initialized
674 * @flags: timer flags
675 * @name: name of the timer
676 * @key: lockdep class key of the fake lock used for tracking timer
677 *       sync lock dependencies
678 *
679 * init_timer_key() must be done to a timer prior calling *any* of the
680 * other timer functions.
681 */
682void init_timer_key(struct timer_list *timer, unsigned int flags,
683		    const char *name, struct lock_class_key *key)
684{
685	debug_init(timer);
686	do_init_timer(timer, flags, name, key);
687}
688EXPORT_SYMBOL(init_timer_key);
689
690static inline void detach_timer(struct timer_list *timer, bool clear_pending)
691{
692	struct list_head *entry = &timer->entry;
693
694	debug_deactivate(timer);
695
696	__list_del(entry->prev, entry->next);
697	if (clear_pending)
698		entry->next = NULL;
699	entry->prev = LIST_POISON2;
700}
701
702static inline void
703detach_expired_timer(struct timer_list *timer, struct tvec_base *base)
704{
705	detach_timer(timer, true);
706	if (!tbase_get_deferrable(timer->base))
707		base->active_timers--;
708	base->all_timers--;
709	(void)catchup_timer_jiffies(base);
710}
711
712static int detach_if_pending(struct timer_list *timer, struct tvec_base *base,
713			     bool clear_pending)
714{
715	if (!timer_pending(timer))
716		return 0;
717
718	detach_timer(timer, clear_pending);
719	if (!tbase_get_deferrable(timer->base)) {
720		base->active_timers--;
721		if (timer->expires == base->next_timer)
722			base->next_timer = base->timer_jiffies;
723	}
724	base->all_timers--;
725	(void)catchup_timer_jiffies(base);
726	return 1;
727}
728
729/*
730 * We are using hashed locking: holding per_cpu(tvec_bases).lock
731 * means that all timers which are tied to this base via timer->base are
732 * locked, and the base itself is locked too.
733 *
734 * So __run_timers/migrate_timers can safely modify all timers which could
735 * be found on ->tvX lists.
736 *
737 * When the timer's base is locked, and the timer removed from list, it is
738 * possible to set timer->base = NULL and drop the lock: the timer remains
739 * locked.
740 */
741static struct tvec_base *lock_timer_base(struct timer_list *timer,
742					unsigned long *flags)
743	__acquires(timer->base->lock)
744{
745	struct tvec_base *base;
746
747	for (;;) {
748		struct tvec_base *prelock_base = timer->base;
749		base = tbase_get_base(prelock_base);
750		if (likely(base != NULL)) {
751			spin_lock_irqsave(&base->lock, *flags);
752			if (likely(prelock_base == timer->base))
753				return base;
754			/* The timer has migrated to another CPU */
755			spin_unlock_irqrestore(&base->lock, *flags);
756		}
757		cpu_relax();
758	}
759}
760
761static inline int
762__mod_timer(struct timer_list *timer, unsigned long expires,
763						bool pending_only, int pinned)
764{
765	struct tvec_base *base, *new_base;
766	unsigned long flags;
767	int ret = 0 , cpu;
768
769	timer_stats_timer_set_start_info(timer);
770	BUG_ON(!timer->function);
771
772	base = lock_timer_base(timer, &flags);
773
774	ret = detach_if_pending(timer, base, false);
775	if (!ret && pending_only)
776		goto out_unlock;
777
778	debug_activate(timer, expires);
779
780	cpu = get_nohz_timer_target(pinned);
781	new_base = per_cpu(tvec_bases, cpu);
782
783	if (base != new_base) {
784		/*
785		 * We are trying to schedule the timer on the local CPU.
786		 * However we can't change timer's base while it is running,
787		 * otherwise del_timer_sync() can't detect that the timer's
788		 * handler yet has not finished. This also guarantees that
789		 * the timer is serialized wrt itself.
790		 */
791		if (likely(base->running_timer != timer)) {
792			/* See the comment in lock_timer_base() */
793			timer_set_base(timer, NULL);
794			spin_unlock(&base->lock);
795			base = new_base;
796			spin_lock(&base->lock);
797			timer_set_base(timer, base);
798		}
799	}
800
801	timer->expires = expires;
802	internal_add_timer(base, timer);
803
804out_unlock:
805	spin_unlock_irqrestore(&base->lock, flags);
806
807	return ret;
808}
809
810/**
811 * mod_timer_pending - modify a pending timer's timeout
812 * @timer: the pending timer to be modified
813 * @expires: new timeout in jiffies
814 *
815 * mod_timer_pending() is the same for pending timers as mod_timer(),
816 * but will not re-activate and modify already deleted timers.
817 *
818 * It is useful for unserialized use of timers.
819 */
820int mod_timer_pending(struct timer_list *timer, unsigned long expires)
821{
822	return __mod_timer(timer, expires, true, TIMER_NOT_PINNED);
823}
824EXPORT_SYMBOL(mod_timer_pending);
825
826/*
827 * Decide where to put the timer while taking the slack into account
828 *
829 * Algorithm:
830 *   1) calculate the maximum (absolute) time
831 *   2) calculate the highest bit where the expires and new max are different
832 *   3) use this bit to make a mask
833 *   4) use the bitmask to round down the maximum time, so that all last
834 *      bits are zeros
835 */
836static inline
837unsigned long apply_slack(struct timer_list *timer, unsigned long expires)
838{
839	unsigned long expires_limit, mask;
840	int bit;
841
842	if (timer->slack >= 0) {
843		expires_limit = expires + timer->slack;
844	} else {
845		long delta = expires - jiffies;
846
847		if (delta < 256)
848			return expires;
849
850		expires_limit = expires + delta / 256;
851	}
852	mask = expires ^ expires_limit;
853	if (mask == 0)
854		return expires;
855
856	bit = find_last_bit(&mask, BITS_PER_LONG);
857
858	mask = (1UL << bit) - 1;
859
860	expires_limit = expires_limit & ~(mask);
861
862	return expires_limit;
863}
864
865/**
866 * mod_timer - modify a timer's timeout
867 * @timer: the timer to be modified
868 * @expires: new timeout in jiffies
869 *
870 * mod_timer() is a more efficient way to update the expire field of an
871 * active timer (if the timer is inactive it will be activated)
872 *
873 * mod_timer(timer, expires) is equivalent to:
874 *
875 *     del_timer(timer); timer->expires = expires; add_timer(timer);
876 *
877 * Note that if there are multiple unserialized concurrent users of the
878 * same timer, then mod_timer() is the only safe way to modify the timeout,
879 * since add_timer() cannot modify an already running timer.
880 *
881 * The function returns whether it has modified a pending timer or not.
882 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
883 * active timer returns 1.)
884 */
885int mod_timer(struct timer_list *timer, unsigned long expires)
886{
887	expires = apply_slack(timer, expires);
888
889	/*
890	 * This is a common optimization triggered by the
891	 * networking code - if the timer is re-modified
892	 * to be the same thing then just return:
893	 */
894	if (timer_pending(timer) && timer->expires == expires)
895		return 1;
896
897	return __mod_timer(timer, expires, false, TIMER_NOT_PINNED);
898}
899EXPORT_SYMBOL(mod_timer);
900
901/**
902 * mod_timer_pinned - modify a timer's timeout
903 * @timer: the timer to be modified
904 * @expires: new timeout in jiffies
905 *
906 * mod_timer_pinned() is a way to update the expire field of an
907 * active timer (if the timer is inactive it will be activated)
908 * and to ensure that the timer is scheduled on the current CPU.
909 *
910 * Note that this does not prevent the timer from being migrated
911 * when the current CPU goes offline.  If this is a problem for
912 * you, use CPU-hotplug notifiers to handle it correctly, for
913 * example, cancelling the timer when the corresponding CPU goes
914 * offline.
915 *
916 * mod_timer_pinned(timer, expires) is equivalent to:
917 *
918 *     del_timer(timer); timer->expires = expires; add_timer(timer);
919 */
920int mod_timer_pinned(struct timer_list *timer, unsigned long expires)
921{
922	if (timer->expires == expires && timer_pending(timer))
923		return 1;
924
925	return __mod_timer(timer, expires, false, TIMER_PINNED);
926}
927EXPORT_SYMBOL(mod_timer_pinned);
928
929/**
930 * add_timer - start a timer
931 * @timer: the timer to be added
932 *
933 * The kernel will do a ->function(->data) callback from the
934 * timer interrupt at the ->expires point in the future. The
935 * current time is 'jiffies'.
936 *
937 * The timer's ->expires, ->function (and if the handler uses it, ->data)
938 * fields must be set prior calling this function.
939 *
940 * Timers with an ->expires field in the past will be executed in the next
941 * timer tick.
942 */
943void add_timer(struct timer_list *timer)
944{
945	BUG_ON(timer_pending(timer));
946	mod_timer(timer, timer->expires);
947}
948EXPORT_SYMBOL(add_timer);
949
950/**
951 * add_timer_on - start a timer on a particular CPU
952 * @timer: the timer to be added
953 * @cpu: the CPU to start it on
954 *
955 * This is not very scalable on SMP. Double adds are not possible.
956 */
957void add_timer_on(struct timer_list *timer, int cpu)
958{
959	struct tvec_base *base = per_cpu(tvec_bases, cpu);
960	unsigned long flags;
961
962	timer_stats_timer_set_start_info(timer);
963	BUG_ON(timer_pending(timer) || !timer->function);
964	spin_lock_irqsave(&base->lock, flags);
965	timer_set_base(timer, base);
966	debug_activate(timer, timer->expires);
967	internal_add_timer(base, timer);
968	spin_unlock_irqrestore(&base->lock, flags);
969}
970EXPORT_SYMBOL_GPL(add_timer_on);
971
972/**
973 * del_timer - deactive a timer.
974 * @timer: the timer to be deactivated
975 *
976 * del_timer() deactivates a timer - this works on both active and inactive
977 * timers.
978 *
979 * The function returns whether it has deactivated a pending timer or not.
980 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
981 * active timer returns 1.)
982 */
983int del_timer(struct timer_list *timer)
984{
985	struct tvec_base *base;
986	unsigned long flags;
987	int ret = 0;
988
989	debug_assert_init(timer);
990
991	timer_stats_timer_clear_start_info(timer);
992	if (timer_pending(timer)) {
993		base = lock_timer_base(timer, &flags);
994		ret = detach_if_pending(timer, base, true);
995		spin_unlock_irqrestore(&base->lock, flags);
996	}
997
998	return ret;
999}
1000EXPORT_SYMBOL(del_timer);
1001
1002/**
1003 * try_to_del_timer_sync - Try to deactivate a timer
1004 * @timer: timer do del
1005 *
1006 * This function tries to deactivate a timer. Upon successful (ret >= 0)
1007 * exit the timer is not queued and the handler is not running on any CPU.
1008 */
1009int try_to_del_timer_sync(struct timer_list *timer)
1010{
1011	struct tvec_base *base;
1012	unsigned long flags;
1013	int ret = -1;
1014
1015	debug_assert_init(timer);
1016
1017	base = lock_timer_base(timer, &flags);
1018
1019	if (base->running_timer != timer) {
1020		timer_stats_timer_clear_start_info(timer);
1021		ret = detach_if_pending(timer, base, true);
1022	}
1023	spin_unlock_irqrestore(&base->lock, flags);
1024
1025	return ret;
1026}
1027EXPORT_SYMBOL(try_to_del_timer_sync);
1028
1029#ifdef CONFIG_SMP
1030/**
1031 * del_timer_sync - deactivate a timer and wait for the handler to finish.
1032 * @timer: the timer to be deactivated
1033 *
1034 * This function only differs from del_timer() on SMP: besides deactivating
1035 * the timer it also makes sure the handler has finished executing on other
1036 * CPUs.
1037 *
1038 * Synchronization rules: Callers must prevent restarting of the timer,
1039 * otherwise this function is meaningless. It must not be called from
1040 * interrupt contexts unless the timer is an irqsafe one. The caller must
1041 * not hold locks which would prevent completion of the timer's
1042 * handler. The timer's handler must not call add_timer_on(). Upon exit the
1043 * timer is not queued and the handler is not running on any CPU.
1044 *
1045 * Note: For !irqsafe timers, you must not hold locks that are held in
1046 *   interrupt context while calling this function. Even if the lock has
1047 *   nothing to do with the timer in question.  Here's why:
1048 *
1049 *    CPU0                             CPU1
1050 *    ----                             ----
1051 *                                   <SOFTIRQ>
1052 *                                   call_timer_fn();
1053 *                                     base->running_timer = mytimer;
1054 *  spin_lock_irq(somelock);
1055 *                                     <IRQ>
1056 *                                        spin_lock(somelock);
1057 *  del_timer_sync(mytimer);
1058 *   while (base->running_timer == mytimer);
1059 *
1060 * Now del_timer_sync() will never return and never release somelock.
1061 * The interrupt on the other CPU is waiting to grab somelock but
1062 * it has interrupted the softirq that CPU0 is waiting to finish.
1063 *
1064 * The function returns whether it has deactivated a pending timer or not.
1065 */
1066int del_timer_sync(struct timer_list *timer)
1067{
1068#ifdef CONFIG_LOCKDEP
1069	unsigned long flags;
1070
1071	/*
1072	 * If lockdep gives a backtrace here, please reference
1073	 * the synchronization rules above.
1074	 */
1075	local_irq_save(flags);
1076	lock_map_acquire(&timer->lockdep_map);
1077	lock_map_release(&timer->lockdep_map);
1078	local_irq_restore(flags);
1079#endif
1080	/*
1081	 * don't use it in hardirq context, because it
1082	 * could lead to deadlock.
1083	 */
1084	WARN_ON(in_irq() && !tbase_get_irqsafe(timer->base));
1085	for (;;) {
1086		int ret = try_to_del_timer_sync(timer);
1087		if (ret >= 0)
1088			return ret;
1089		cpu_relax();
1090	}
1091}
1092EXPORT_SYMBOL(del_timer_sync);
1093#endif
1094
1095static int cascade(struct tvec_base *base, struct tvec *tv, int index)
1096{
1097	/* cascade all the timers from tv up one level */
1098	struct timer_list *timer, *tmp;
1099	struct list_head tv_list;
1100
1101	list_replace_init(tv->vec + index, &tv_list);
1102
1103	/*
1104	 * We are removing _all_ timers from the list, so we
1105	 * don't have to detach them individually.
1106	 */
1107	list_for_each_entry_safe(timer, tmp, &tv_list, entry) {
1108		BUG_ON(tbase_get_base(timer->base) != base);
1109		/* No accounting, while moving them */
1110		__internal_add_timer(base, timer);
1111	}
1112
1113	return index;
1114}
1115
1116static void call_timer_fn(struct timer_list *timer, void (*fn)(unsigned long),
1117			  unsigned long data)
1118{
1119	int count = preempt_count();
1120
1121#ifdef CONFIG_LOCKDEP
1122	/*
1123	 * It is permissible to free the timer from inside the
1124	 * function that is called from it, this we need to take into
1125	 * account for lockdep too. To avoid bogus "held lock freed"
1126	 * warnings as well as problems when looking into
1127	 * timer->lockdep_map, make a copy and use that here.
1128	 */
1129	struct lockdep_map lockdep_map;
1130
1131	lockdep_copy_map(&lockdep_map, &timer->lockdep_map);
1132#endif
1133	/*
1134	 * Couple the lock chain with the lock chain at
1135	 * del_timer_sync() by acquiring the lock_map around the fn()
1136	 * call here and in del_timer_sync().
1137	 */
1138	lock_map_acquire(&lockdep_map);
1139
1140	trace_timer_expire_entry(timer);
1141	fn(data);
1142	trace_timer_expire_exit(timer);
1143
1144	lock_map_release(&lockdep_map);
1145
1146	if (count != preempt_count()) {
1147		WARN_ONCE(1, "timer: %pF preempt leak: %08x -> %08x\n",
1148			  fn, count, preempt_count());
1149		/*
1150		 * Restore the preempt count. That gives us a decent
1151		 * chance to survive and extract information. If the
1152		 * callback kept a lock held, bad luck, but not worse
1153		 * than the BUG() we had.
1154		 */
1155		preempt_count_set(count);
1156	}
1157}
1158
1159#define INDEX(N) ((base->timer_jiffies >> (TVR_BITS + (N) * TVN_BITS)) & TVN_MASK)
1160
1161/**
1162 * __run_timers - run all expired timers (if any) on this CPU.
1163 * @base: the timer vector to be processed.
1164 *
1165 * This function cascades all vectors and executes all expired timer
1166 * vectors.
1167 */
1168static inline void __run_timers(struct tvec_base *base)
1169{
1170	struct timer_list *timer;
1171
1172	spin_lock_irq(&base->lock);
1173	if (catchup_timer_jiffies(base)) {
1174		spin_unlock_irq(&base->lock);
1175		return;
1176	}
1177	while (time_after_eq(jiffies, base->timer_jiffies)) {
1178		struct list_head work_list;
1179		struct list_head *head = &work_list;
1180		int index = base->timer_jiffies & TVR_MASK;
1181
1182		/*
1183		 * Cascade timers:
1184		 */
1185		if (!index &&
1186			(!cascade(base, &base->tv2, INDEX(0))) &&
1187				(!cascade(base, &base->tv3, INDEX(1))) &&
1188					!cascade(base, &base->tv4, INDEX(2)))
1189			cascade(base, &base->tv5, INDEX(3));
1190		++base->timer_jiffies;
1191		list_replace_init(base->tv1.vec + index, head);
1192		while (!list_empty(head)) {
1193			void (*fn)(unsigned long);
1194			unsigned long data;
1195			bool irqsafe;
1196
1197			timer = list_first_entry(head, struct timer_list,entry);
1198			fn = timer->function;
1199			data = timer->data;
1200			irqsafe = tbase_get_irqsafe(timer->base);
1201
1202			timer_stats_account_timer(timer);
1203
1204			base->running_timer = timer;
1205			detach_expired_timer(timer, base);
1206
1207			if (irqsafe) {
1208				spin_unlock(&base->lock);
1209				call_timer_fn(timer, fn, data);
1210				spin_lock(&base->lock);
1211			} else {
1212				spin_unlock_irq(&base->lock);
1213				call_timer_fn(timer, fn, data);
1214				spin_lock_irq(&base->lock);
1215			}
1216		}
1217	}
1218	base->running_timer = NULL;
1219	spin_unlock_irq(&base->lock);
1220}
1221
1222#ifdef CONFIG_NO_HZ_COMMON
1223/*
1224 * Find out when the next timer event is due to happen. This
1225 * is used on S/390 to stop all activity when a CPU is idle.
1226 * This function needs to be called with interrupts disabled.
1227 */
1228static unsigned long __next_timer_interrupt(struct tvec_base *base)
1229{
1230	unsigned long timer_jiffies = base->timer_jiffies;
1231	unsigned long expires = timer_jiffies + NEXT_TIMER_MAX_DELTA;
1232	int index, slot, array, found = 0;
1233	struct timer_list *nte;
1234	struct tvec *varray[4];
1235
1236	/* Look for timer events in tv1. */
1237	index = slot = timer_jiffies & TVR_MASK;
1238	do {
1239		list_for_each_entry(nte, base->tv1.vec + slot, entry) {
1240			if (tbase_get_deferrable(nte->base))
1241				continue;
1242
1243			found = 1;
1244			expires = nte->expires;
1245			/* Look at the cascade bucket(s)? */
1246			if (!index || slot < index)
1247				goto cascade;
1248			return expires;
1249		}
1250		slot = (slot + 1) & TVR_MASK;
1251	} while (slot != index);
1252
1253cascade:
1254	/* Calculate the next cascade event */
1255	if (index)
1256		timer_jiffies += TVR_SIZE - index;
1257	timer_jiffies >>= TVR_BITS;
1258
1259	/* Check tv2-tv5. */
1260	varray[0] = &base->tv2;
1261	varray[1] = &base->tv3;
1262	varray[2] = &base->tv4;
1263	varray[3] = &base->tv5;
1264
1265	for (array = 0; array < 4; array++) {
1266		struct tvec *varp = varray[array];
1267
1268		index = slot = timer_jiffies & TVN_MASK;
1269		do {
1270			list_for_each_entry(nte, varp->vec + slot, entry) {
1271				if (tbase_get_deferrable(nte->base))
1272					continue;
1273
1274				found = 1;
1275				if (time_before(nte->expires, expires))
1276					expires = nte->expires;
1277			}
1278			/*
1279			 * Do we still search for the first timer or are
1280			 * we looking up the cascade buckets ?
1281			 */
1282			if (found) {
1283				/* Look at the cascade bucket(s)? */
1284				if (!index || slot < index)
1285					break;
1286				return expires;
1287			}
1288			slot = (slot + 1) & TVN_MASK;
1289		} while (slot != index);
1290
1291		if (index)
1292			timer_jiffies += TVN_SIZE - index;
1293		timer_jiffies >>= TVN_BITS;
1294	}
1295	return expires;
1296}
1297
1298/*
1299 * Check, if the next hrtimer event is before the next timer wheel
1300 * event:
1301 */
1302static unsigned long cmp_next_hrtimer_event(unsigned long now,
1303					    unsigned long expires)
1304{
1305	ktime_t hr_delta = hrtimer_get_next_event();
1306	struct timespec tsdelta;
1307	unsigned long delta;
1308
1309	if (hr_delta.tv64 == KTIME_MAX)
1310		return expires;
1311
1312	/*
1313	 * Expired timer available, let it expire in the next tick
1314	 */
1315	if (hr_delta.tv64 <= 0)
1316		return now + 1;
1317
1318	tsdelta = ktime_to_timespec(hr_delta);
1319	delta = timespec_to_jiffies(&tsdelta);
1320
1321	/*
1322	 * Limit the delta to the max value, which is checked in
1323	 * tick_nohz_stop_sched_tick():
1324	 */
1325	if (delta > NEXT_TIMER_MAX_DELTA)
1326		delta = NEXT_TIMER_MAX_DELTA;
1327
1328	/*
1329	 * Take rounding errors in to account and make sure, that it
1330	 * expires in the next tick. Otherwise we go into an endless
1331	 * ping pong due to tick_nohz_stop_sched_tick() retriggering
1332	 * the timer softirq
1333	 */
1334	if (delta < 1)
1335		delta = 1;
1336	now += delta;
1337	if (time_before(now, expires))
1338		return now;
1339	return expires;
1340}
1341
1342/**
1343 * get_next_timer_interrupt - return the jiffy of the next pending timer
1344 * @now: current time (in jiffies)
1345 */
1346unsigned long get_next_timer_interrupt(unsigned long now)
1347{
1348	struct tvec_base *base = __this_cpu_read(tvec_bases);
1349	unsigned long expires = now + NEXT_TIMER_MAX_DELTA;
1350
1351	/*
1352	 * Pretend that there is no timer pending if the cpu is offline.
1353	 * Possible pending timers will be migrated later to an active cpu.
1354	 */
1355	if (cpu_is_offline(smp_processor_id()))
1356		return expires;
1357
1358	spin_lock(&base->lock);
1359	if (base->active_timers) {
1360		if (time_before_eq(base->next_timer, base->timer_jiffies))
1361			base->next_timer = __next_timer_interrupt(base);
1362		expires = base->next_timer;
1363	}
1364	spin_unlock(&base->lock);
1365
1366	if (time_before_eq(expires, now))
1367		return now;
1368
1369	return cmp_next_hrtimer_event(now, expires);
1370}
1371#endif
1372
1373/*
1374 * Called from the timer interrupt handler to charge one tick to the current
1375 * process.  user_tick is 1 if the tick is user time, 0 for system.
1376 */
1377void update_process_times(int user_tick)
1378{
1379	struct task_struct *p = current;
1380	int cpu = smp_processor_id();
1381
1382	/* Note: this timer irq context must be accounted for as well. */
1383	account_process_tick(p, user_tick);
1384	run_local_timers();
1385	rcu_check_callbacks(cpu, user_tick);
1386#ifdef CONFIG_IRQ_WORK
1387	if (in_irq())
1388		irq_work_tick();
1389#endif
1390	scheduler_tick();
1391	run_posix_cpu_timers(p);
1392}
1393
1394/*
1395 * This function runs timers and the timer-tq in bottom half context.
1396 */
1397static void run_timer_softirq(struct softirq_action *h)
1398{
1399	struct tvec_base *base = __this_cpu_read(tvec_bases);
1400
1401	hrtimer_run_pending();
1402
1403	if (time_after_eq(jiffies, base->timer_jiffies))
1404		__run_timers(base);
1405}
1406
1407/*
1408 * Called by the local, per-CPU timer interrupt on SMP.
1409 */
1410void run_local_timers(void)
1411{
1412	hrtimer_run_queues();
1413	raise_softirq(TIMER_SOFTIRQ);
1414}
1415
1416#ifdef __ARCH_WANT_SYS_ALARM
1417
1418/*
1419 * For backwards compatibility?  This can be done in libc so Alpha
1420 * and all newer ports shouldn't need it.
1421 */
1422SYSCALL_DEFINE1(alarm, unsigned int, seconds)
1423{
1424	return alarm_setitimer(seconds);
1425}
1426
1427#endif
1428
1429static void process_timeout(unsigned long __data)
1430{
1431	wake_up_process((struct task_struct *)__data);
1432}
1433
1434/**
1435 * schedule_timeout - sleep until timeout
1436 * @timeout: timeout value in jiffies
1437 *
1438 * Make the current task sleep until @timeout jiffies have
1439 * elapsed. The routine will return immediately unless
1440 * the current task state has been set (see set_current_state()).
1441 *
1442 * You can set the task state as follows -
1443 *
1444 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1445 * pass before the routine returns. The routine will return 0
1446 *
1447 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1448 * delivered to the current task. In this case the remaining time
1449 * in jiffies will be returned, or 0 if the timer expired in time
1450 *
1451 * The current task state is guaranteed to be TASK_RUNNING when this
1452 * routine returns.
1453 *
1454 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1455 * the CPU away without a bound on the timeout. In this case the return
1456 * value will be %MAX_SCHEDULE_TIMEOUT.
1457 *
1458 * In all cases the return value is guaranteed to be non-negative.
1459 */
1460signed long __sched schedule_timeout(signed long timeout)
1461{
1462	struct timer_list timer;
1463	unsigned long expire;
1464
1465	switch (timeout)
1466	{
1467	case MAX_SCHEDULE_TIMEOUT:
1468		/*
1469		 * These two special cases are useful to be comfortable
1470		 * in the caller. Nothing more. We could take
1471		 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1472		 * but I' d like to return a valid offset (>=0) to allow
1473		 * the caller to do everything it want with the retval.
1474		 */
1475		schedule();
1476		goto out;
1477	default:
1478		/*
1479		 * Another bit of PARANOID. Note that the retval will be
1480		 * 0 since no piece of kernel is supposed to do a check
1481		 * for a negative retval of schedule_timeout() (since it
1482		 * should never happens anyway). You just have the printk()
1483		 * that will tell you if something is gone wrong and where.
1484		 */
1485		if (timeout < 0) {
1486			printk(KERN_ERR "schedule_timeout: wrong timeout "
1487				"value %lx\n", timeout);
1488			dump_stack();
1489			current->state = TASK_RUNNING;
1490			goto out;
1491		}
1492	}
1493
1494	expire = timeout + jiffies;
1495
1496	setup_timer_on_stack(&timer, process_timeout, (unsigned long)current);
1497	__mod_timer(&timer, expire, false, TIMER_NOT_PINNED);
1498	schedule();
1499	del_singleshot_timer_sync(&timer);
1500
1501	/* Remove the timer from the object tracker */
1502	destroy_timer_on_stack(&timer);
1503
1504	timeout = expire - jiffies;
1505
1506 out:
1507	return timeout < 0 ? 0 : timeout;
1508}
1509EXPORT_SYMBOL(schedule_timeout);
1510
1511/*
1512 * We can use __set_current_state() here because schedule_timeout() calls
1513 * schedule() unconditionally.
1514 */
1515signed long __sched schedule_timeout_interruptible(signed long timeout)
1516{
1517	__set_current_state(TASK_INTERRUPTIBLE);
1518	return schedule_timeout(timeout);
1519}
1520EXPORT_SYMBOL(schedule_timeout_interruptible);
1521
1522signed long __sched schedule_timeout_killable(signed long timeout)
1523{
1524	__set_current_state(TASK_KILLABLE);
1525	return schedule_timeout(timeout);
1526}
1527EXPORT_SYMBOL(schedule_timeout_killable);
1528
1529signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1530{
1531	__set_current_state(TASK_UNINTERRUPTIBLE);
1532	return schedule_timeout(timeout);
1533}
1534EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1535
1536static int init_timers_cpu(int cpu)
1537{
1538	int j;
1539	struct tvec_base *base;
1540	static char tvec_base_done[NR_CPUS];
1541
1542	if (!tvec_base_done[cpu]) {
1543		static char boot_done;
1544
1545		if (boot_done) {
1546			/*
1547			 * The APs use this path later in boot
1548			 */
1549			base = kzalloc_node(sizeof(*base), GFP_KERNEL,
1550					    cpu_to_node(cpu));
1551			if (!base)
1552				return -ENOMEM;
1553
1554			/* Make sure tvec_base has TIMER_FLAG_MASK bits free */
1555			if (WARN_ON(base != tbase_get_base(base))) {
1556				kfree(base);
1557				return -ENOMEM;
1558			}
1559			per_cpu(tvec_bases, cpu) = base;
1560		} else {
1561			/*
1562			 * This is for the boot CPU - we use compile-time
1563			 * static initialisation because per-cpu memory isn't
1564			 * ready yet and because the memory allocators are not
1565			 * initialised either.
1566			 */
1567			boot_done = 1;
1568			base = &boot_tvec_bases;
1569		}
1570		spin_lock_init(&base->lock);
1571		tvec_base_done[cpu] = 1;
1572		base->cpu = cpu;
1573	} else {
1574		base = per_cpu(tvec_bases, cpu);
1575	}
1576
1577
1578	for (j = 0; j < TVN_SIZE; j++) {
1579		INIT_LIST_HEAD(base->tv5.vec + j);
1580		INIT_LIST_HEAD(base->tv4.vec + j);
1581		INIT_LIST_HEAD(base->tv3.vec + j);
1582		INIT_LIST_HEAD(base->tv2.vec + j);
1583	}
1584	for (j = 0; j < TVR_SIZE; j++)
1585		INIT_LIST_HEAD(base->tv1.vec + j);
1586
1587	base->timer_jiffies = jiffies;
1588	base->next_timer = base->timer_jiffies;
1589	base->active_timers = 0;
1590	base->all_timers = 0;
1591	return 0;
1592}
1593
1594#ifdef CONFIG_HOTPLUG_CPU
1595static void migrate_timer_list(struct tvec_base *new_base, struct list_head *head)
1596{
1597	struct timer_list *timer;
1598
1599	while (!list_empty(head)) {
1600		timer = list_first_entry(head, struct timer_list, entry);
1601		/* We ignore the accounting on the dying cpu */
1602		detach_timer(timer, false);
1603		timer_set_base(timer, new_base);
1604		internal_add_timer(new_base, timer);
1605	}
1606}
1607
1608static void migrate_timers(int cpu)
1609{
1610	struct tvec_base *old_base;
1611	struct tvec_base *new_base;
1612	int i;
1613
1614	BUG_ON(cpu_online(cpu));
1615	old_base = per_cpu(tvec_bases, cpu);
1616	new_base = get_cpu_var(tvec_bases);
1617	/*
1618	 * The caller is globally serialized and nobody else
1619	 * takes two locks at once, deadlock is not possible.
1620	 */
1621	spin_lock_irq(&new_base->lock);
1622	spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1623
1624	BUG_ON(old_base->running_timer);
1625
1626	for (i = 0; i < TVR_SIZE; i++)
1627		migrate_timer_list(new_base, old_base->tv1.vec + i);
1628	for (i = 0; i < TVN_SIZE; i++) {
1629		migrate_timer_list(new_base, old_base->tv2.vec + i);
1630		migrate_timer_list(new_base, old_base->tv3.vec + i);
1631		migrate_timer_list(new_base, old_base->tv4.vec + i);
1632		migrate_timer_list(new_base, old_base->tv5.vec + i);
1633	}
1634
1635	spin_unlock(&old_base->lock);
1636	spin_unlock_irq(&new_base->lock);
1637	put_cpu_var(tvec_bases);
1638}
1639#endif /* CONFIG_HOTPLUG_CPU */
1640
1641static int timer_cpu_notify(struct notifier_block *self,
1642				unsigned long action, void *hcpu)
1643{
1644	long cpu = (long)hcpu;
1645	int err;
1646
1647	switch(action) {
1648	case CPU_UP_PREPARE:
1649	case CPU_UP_PREPARE_FROZEN:
1650		err = init_timers_cpu(cpu);
1651		if (err < 0)
1652			return notifier_from_errno(err);
1653		break;
1654#ifdef CONFIG_HOTPLUG_CPU
1655	case CPU_DEAD:
1656	case CPU_DEAD_FROZEN:
1657		migrate_timers(cpu);
1658		break;
1659#endif
1660	default:
1661		break;
1662	}
1663	return NOTIFY_OK;
1664}
1665
1666static struct notifier_block timers_nb = {
1667	.notifier_call	= timer_cpu_notify,
1668};
1669
1670
1671void __init init_timers(void)
1672{
1673	int err;
1674
1675	/* ensure there are enough low bits for flags in timer->base pointer */
1676	BUILD_BUG_ON(__alignof__(struct tvec_base) & TIMER_FLAG_MASK);
1677
1678	err = timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE,
1679			       (void *)(long)smp_processor_id());
1680	BUG_ON(err != NOTIFY_OK);
1681
1682	init_timer_stats();
1683	register_cpu_notifier(&timers_nb);
1684	open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
1685}
1686
1687/**
1688 * msleep - sleep safely even with waitqueue interruptions
1689 * @msecs: Time in milliseconds to sleep for
1690 */
1691void msleep(unsigned int msecs)
1692{
1693	unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1694
1695	while (timeout)
1696		timeout = schedule_timeout_uninterruptible(timeout);
1697}
1698
1699EXPORT_SYMBOL(msleep);
1700
1701/**
1702 * msleep_interruptible - sleep waiting for signals
1703 * @msecs: Time in milliseconds to sleep for
1704 */
1705unsigned long msleep_interruptible(unsigned int msecs)
1706{
1707	unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1708
1709	while (timeout && !signal_pending(current))
1710		timeout = schedule_timeout_interruptible(timeout);
1711	return jiffies_to_msecs(timeout);
1712}
1713
1714EXPORT_SYMBOL(msleep_interruptible);
1715
1716static int __sched do_usleep_range(unsigned long min, unsigned long max)
1717{
1718	ktime_t kmin;
1719	unsigned long delta;
1720
1721	kmin = ktime_set(0, min * NSEC_PER_USEC);
1722	delta = (max - min) * NSEC_PER_USEC;
1723	return schedule_hrtimeout_range(&kmin, delta, HRTIMER_MODE_REL);
1724}
1725
1726/**
1727 * usleep_range - Drop in replacement for udelay where wakeup is flexible
1728 * @min: Minimum time in usecs to sleep
1729 * @max: Maximum time in usecs to sleep
1730 */
1731void usleep_range(unsigned long min, unsigned long max)
1732{
1733	__set_current_state(TASK_UNINTERRUPTIBLE);
1734	do_usleep_range(min, max);
1735}
1736EXPORT_SYMBOL(usleep_range);
1737