xref: /openbmc/linux/kernel/time/timer.c (revision 8365a898)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  Kernel internal timers
4  *
5  *  Copyright (C) 1991, 1992  Linus Torvalds
6  *
7  *  1997-01-28  Modified by Finn Arne Gangstad to make timers scale better.
8  *
9  *  1997-09-10  Updated NTP code according to technical memorandum Jan '96
10  *              "A Kernel Model for Precision Timekeeping" by Dave Mills
11  *  1998-12-24  Fixed a xtime SMP race (we need the xtime_lock rw spinlock to
12  *              serialize accesses to xtime/lost_ticks).
13  *                              Copyright (C) 1998  Andrea Arcangeli
14  *  1999-03-10  Improved NTP compatibility by Ulrich Windl
15  *  2002-05-31	Move sys_sysinfo here and make its locking sane, Robert Love
16  *  2000-10-05  Implemented scalable SMP per-CPU timer handling.
17  *                              Copyright (C) 2000, 2001, 2002  Ingo Molnar
18  *              Designed by David S. Miller, Alexey Kuznetsov and Ingo Molnar
19  */
20 
21 #include <linux/kernel_stat.h>
22 #include <linux/export.h>
23 #include <linux/interrupt.h>
24 #include <linux/percpu.h>
25 #include <linux/init.h>
26 #include <linux/mm.h>
27 #include <linux/swap.h>
28 #include <linux/pid_namespace.h>
29 #include <linux/notifier.h>
30 #include <linux/thread_info.h>
31 #include <linux/time.h>
32 #include <linux/jiffies.h>
33 #include <linux/posix-timers.h>
34 #include <linux/cpu.h>
35 #include <linux/syscalls.h>
36 #include <linux/delay.h>
37 #include <linux/tick.h>
38 #include <linux/kallsyms.h>
39 #include <linux/irq_work.h>
40 #include <linux/sched/signal.h>
41 #include <linux/sched/sysctl.h>
42 #include <linux/sched/nohz.h>
43 #include <linux/sched/debug.h>
44 #include <linux/slab.h>
45 #include <linux/compat.h>
46 
47 #include <linux/uaccess.h>
48 #include <asm/unistd.h>
49 #include <asm/div64.h>
50 #include <asm/timex.h>
51 #include <asm/io.h>
52 
53 #include "tick-internal.h"
54 
55 #define CREATE_TRACE_POINTS
56 #include <trace/events/timer.h>
57 
58 __visible u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
59 
60 EXPORT_SYMBOL(jiffies_64);
61 
62 /*
63  * The timer wheel has LVL_DEPTH array levels. Each level provides an array of
64  * LVL_SIZE buckets. Each level is driven by its own clock and therefor each
65  * level has a different granularity.
66  *
67  * The level granularity is:		LVL_CLK_DIV ^ lvl
68  * The level clock frequency is:	HZ / (LVL_CLK_DIV ^ level)
69  *
70  * The array level of a newly armed timer depends on the relative expiry
71  * time. The farther the expiry time is away the higher the array level and
72  * therefor the granularity becomes.
73  *
74  * Contrary to the original timer wheel implementation, which aims for 'exact'
75  * expiry of the timers, this implementation removes the need for recascading
76  * the timers into the lower array levels. The previous 'classic' timer wheel
77  * implementation of the kernel already violated the 'exact' expiry by adding
78  * slack to the expiry time to provide batched expiration. The granularity
79  * levels provide implicit batching.
80  *
81  * This is an optimization of the original timer wheel implementation for the
82  * majority of the timer wheel use cases: timeouts. The vast majority of
83  * timeout timers (networking, disk I/O ...) are canceled before expiry. If
84  * the timeout expires it indicates that normal operation is disturbed, so it
85  * does not matter much whether the timeout comes with a slight delay.
86  *
87  * The only exception to this are networking timers with a small expiry
88  * time. They rely on the granularity. Those fit into the first wheel level,
89  * which has HZ granularity.
90  *
91  * We don't have cascading anymore. timers with a expiry time above the
92  * capacity of the last wheel level are force expired at the maximum timeout
93  * value of the last wheel level. From data sampling we know that the maximum
94  * value observed is 5 days (network connection tracking), so this should not
95  * be an issue.
96  *
97  * The currently chosen array constants values are a good compromise between
98  * array size and granularity.
99  *
100  * This results in the following granularity and range levels:
101  *
102  * HZ 1000 steps
103  * Level Offset  Granularity            Range
104  *  0      0         1 ms                0 ms -         63 ms
105  *  1     64         8 ms               64 ms -        511 ms
106  *  2    128        64 ms              512 ms -       4095 ms (512ms - ~4s)
107  *  3    192       512 ms             4096 ms -      32767 ms (~4s - ~32s)
108  *  4    256      4096 ms (~4s)      32768 ms -     262143 ms (~32s - ~4m)
109  *  5    320     32768 ms (~32s)    262144 ms -    2097151 ms (~4m - ~34m)
110  *  6    384    262144 ms (~4m)    2097152 ms -   16777215 ms (~34m - ~4h)
111  *  7    448   2097152 ms (~34m)  16777216 ms -  134217727 ms (~4h - ~1d)
112  *  8    512  16777216 ms (~4h)  134217728 ms - 1073741822 ms (~1d - ~12d)
113  *
114  * HZ  300
115  * Level Offset  Granularity            Range
116  *  0	   0         3 ms                0 ms -        210 ms
117  *  1	  64        26 ms              213 ms -       1703 ms (213ms - ~1s)
118  *  2	 128       213 ms             1706 ms -      13650 ms (~1s - ~13s)
119  *  3	 192      1706 ms (~1s)      13653 ms -     109223 ms (~13s - ~1m)
120  *  4	 256     13653 ms (~13s)    109226 ms -     873810 ms (~1m - ~14m)
121  *  5	 320    109226 ms (~1m)     873813 ms -    6990503 ms (~14m - ~1h)
122  *  6	 384    873813 ms (~14m)   6990506 ms -   55924050 ms (~1h - ~15h)
123  *  7	 448   6990506 ms (~1h)   55924053 ms -  447392423 ms (~15h - ~5d)
124  *  8    512  55924053 ms (~15h) 447392426 ms - 3579139406 ms (~5d - ~41d)
125  *
126  * HZ  250
127  * Level Offset  Granularity            Range
128  *  0	   0         4 ms                0 ms -        255 ms
129  *  1	  64        32 ms              256 ms -       2047 ms (256ms - ~2s)
130  *  2	 128       256 ms             2048 ms -      16383 ms (~2s - ~16s)
131  *  3	 192      2048 ms (~2s)      16384 ms -     131071 ms (~16s - ~2m)
132  *  4	 256     16384 ms (~16s)    131072 ms -    1048575 ms (~2m - ~17m)
133  *  5	 320    131072 ms (~2m)    1048576 ms -    8388607 ms (~17m - ~2h)
134  *  6	 384   1048576 ms (~17m)   8388608 ms -   67108863 ms (~2h - ~18h)
135  *  7	 448   8388608 ms (~2h)   67108864 ms -  536870911 ms (~18h - ~6d)
136  *  8    512  67108864 ms (~18h) 536870912 ms - 4294967288 ms (~6d - ~49d)
137  *
138  * HZ  100
139  * Level Offset  Granularity            Range
140  *  0	   0         10 ms               0 ms -        630 ms
141  *  1	  64         80 ms             640 ms -       5110 ms (640ms - ~5s)
142  *  2	 128        640 ms            5120 ms -      40950 ms (~5s - ~40s)
143  *  3	 192       5120 ms (~5s)     40960 ms -     327670 ms (~40s - ~5m)
144  *  4	 256      40960 ms (~40s)   327680 ms -    2621430 ms (~5m - ~43m)
145  *  5	 320     327680 ms (~5m)   2621440 ms -   20971510 ms (~43m - ~5h)
146  *  6	 384    2621440 ms (~43m) 20971520 ms -  167772150 ms (~5h - ~1d)
147  *  7	 448   20971520 ms (~5h) 167772160 ms - 1342177270 ms (~1d - ~15d)
148  */
149 
150 /* Clock divisor for the next level */
151 #define LVL_CLK_SHIFT	3
152 #define LVL_CLK_DIV	(1UL << LVL_CLK_SHIFT)
153 #define LVL_CLK_MASK	(LVL_CLK_DIV - 1)
154 #define LVL_SHIFT(n)	((n) * LVL_CLK_SHIFT)
155 #define LVL_GRAN(n)	(1UL << LVL_SHIFT(n))
156 
157 /*
158  * The time start value for each level to select the bucket at enqueue
159  * time.
160  */
161 #define LVL_START(n)	((LVL_SIZE - 1) << (((n) - 1) * LVL_CLK_SHIFT))
162 
163 /* Size of each clock level */
164 #define LVL_BITS	6
165 #define LVL_SIZE	(1UL << LVL_BITS)
166 #define LVL_MASK	(LVL_SIZE - 1)
167 #define LVL_OFFS(n)	((n) * LVL_SIZE)
168 
169 /* Level depth */
170 #if HZ > 100
171 # define LVL_DEPTH	9
172 # else
173 # define LVL_DEPTH	8
174 #endif
175 
176 /* The cutoff (max. capacity of the wheel) */
177 #define WHEEL_TIMEOUT_CUTOFF	(LVL_START(LVL_DEPTH))
178 #define WHEEL_TIMEOUT_MAX	(WHEEL_TIMEOUT_CUTOFF - LVL_GRAN(LVL_DEPTH - 1))
179 
180 /*
181  * The resulting wheel size. If NOHZ is configured we allocate two
182  * wheels so we have a separate storage for the deferrable timers.
183  */
184 #define WHEEL_SIZE	(LVL_SIZE * LVL_DEPTH)
185 
186 #ifdef CONFIG_NO_HZ_COMMON
187 # define NR_BASES	2
188 # define BASE_STD	0
189 # define BASE_DEF	1
190 #else
191 # define NR_BASES	1
192 # define BASE_STD	0
193 # define BASE_DEF	0
194 #endif
195 
196 struct timer_base {
197 	raw_spinlock_t		lock;
198 	struct timer_list	*running_timer;
199 #ifdef CONFIG_PREEMPT_RT
200 	spinlock_t		expiry_lock;
201 	atomic_t		timer_waiters;
202 #endif
203 	unsigned long		clk;
204 	unsigned long		next_expiry;
205 	unsigned int		cpu;
206 	bool			is_idle;
207 	bool			must_forward_clk;
208 	DECLARE_BITMAP(pending_map, WHEEL_SIZE);
209 	struct hlist_head	vectors[WHEEL_SIZE];
210 } ____cacheline_aligned;
211 
212 static DEFINE_PER_CPU(struct timer_base, timer_bases[NR_BASES]);
213 
214 #ifdef CONFIG_NO_HZ_COMMON
215 
216 static DEFINE_STATIC_KEY_FALSE(timers_nohz_active);
217 static DEFINE_MUTEX(timer_keys_mutex);
218 
219 static void timer_update_keys(struct work_struct *work);
220 static DECLARE_WORK(timer_update_work, timer_update_keys);
221 
222 #ifdef CONFIG_SMP
223 unsigned int sysctl_timer_migration = 1;
224 
225 DEFINE_STATIC_KEY_FALSE(timers_migration_enabled);
226 
227 static void timers_update_migration(void)
228 {
229 	if (sysctl_timer_migration && tick_nohz_active)
230 		static_branch_enable(&timers_migration_enabled);
231 	else
232 		static_branch_disable(&timers_migration_enabled);
233 }
234 #else
235 static inline void timers_update_migration(void) { }
236 #endif /* !CONFIG_SMP */
237 
238 static void timer_update_keys(struct work_struct *work)
239 {
240 	mutex_lock(&timer_keys_mutex);
241 	timers_update_migration();
242 	static_branch_enable(&timers_nohz_active);
243 	mutex_unlock(&timer_keys_mutex);
244 }
245 
246 void timers_update_nohz(void)
247 {
248 	schedule_work(&timer_update_work);
249 }
250 
251 int timer_migration_handler(struct ctl_table *table, int write,
252 			    void *buffer, size_t *lenp, loff_t *ppos)
253 {
254 	int ret;
255 
256 	mutex_lock(&timer_keys_mutex);
257 	ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
258 	if (!ret && write)
259 		timers_update_migration();
260 	mutex_unlock(&timer_keys_mutex);
261 	return ret;
262 }
263 
264 static inline bool is_timers_nohz_active(void)
265 {
266 	return static_branch_unlikely(&timers_nohz_active);
267 }
268 #else
269 static inline bool is_timers_nohz_active(void) { return false; }
270 #endif /* NO_HZ_COMMON */
271 
272 static unsigned long round_jiffies_common(unsigned long j, int cpu,
273 		bool force_up)
274 {
275 	int rem;
276 	unsigned long original = j;
277 
278 	/*
279 	 * We don't want all cpus firing their timers at once hitting the
280 	 * same lock or cachelines, so we skew each extra cpu with an extra
281 	 * 3 jiffies. This 3 jiffies came originally from the mm/ code which
282 	 * already did this.
283 	 * The skew is done by adding 3*cpunr, then round, then subtract this
284 	 * extra offset again.
285 	 */
286 	j += cpu * 3;
287 
288 	rem = j % HZ;
289 
290 	/*
291 	 * If the target jiffie is just after a whole second (which can happen
292 	 * due to delays of the timer irq, long irq off times etc etc) then
293 	 * we should round down to the whole second, not up. Use 1/4th second
294 	 * as cutoff for this rounding as an extreme upper bound for this.
295 	 * But never round down if @force_up is set.
296 	 */
297 	if (rem < HZ/4 && !force_up) /* round down */
298 		j = j - rem;
299 	else /* round up */
300 		j = j - rem + HZ;
301 
302 	/* now that we have rounded, subtract the extra skew again */
303 	j -= cpu * 3;
304 
305 	/*
306 	 * Make sure j is still in the future. Otherwise return the
307 	 * unmodified value.
308 	 */
309 	return time_is_after_jiffies(j) ? j : original;
310 }
311 
312 /**
313  * __round_jiffies - function to round jiffies to a full second
314  * @j: the time in (absolute) jiffies that should be rounded
315  * @cpu: the processor number on which the timeout will happen
316  *
317  * __round_jiffies() rounds an absolute time in the future (in jiffies)
318  * up or down to (approximately) full seconds. This is useful for timers
319  * for which the exact time they fire does not matter too much, as long as
320  * they fire approximately every X seconds.
321  *
322  * By rounding these timers to whole seconds, all such timers will fire
323  * at the same time, rather than at various times spread out. The goal
324  * of this is to have the CPU wake up less, which saves power.
325  *
326  * The exact rounding is skewed for each processor to avoid all
327  * processors firing at the exact same time, which could lead
328  * to lock contention or spurious cache line bouncing.
329  *
330  * The return value is the rounded version of the @j parameter.
331  */
332 unsigned long __round_jiffies(unsigned long j, int cpu)
333 {
334 	return round_jiffies_common(j, cpu, false);
335 }
336 EXPORT_SYMBOL_GPL(__round_jiffies);
337 
338 /**
339  * __round_jiffies_relative - function to round jiffies to a full second
340  * @j: the time in (relative) jiffies that should be rounded
341  * @cpu: the processor number on which the timeout will happen
342  *
343  * __round_jiffies_relative() rounds a time delta  in the future (in jiffies)
344  * up or down to (approximately) full seconds. This is useful for timers
345  * for which the exact time they fire does not matter too much, as long as
346  * they fire approximately every X seconds.
347  *
348  * By rounding these timers to whole seconds, all such timers will fire
349  * at the same time, rather than at various times spread out. The goal
350  * of this is to have the CPU wake up less, which saves power.
351  *
352  * The exact rounding is skewed for each processor to avoid all
353  * processors firing at the exact same time, which could lead
354  * to lock contention or spurious cache line bouncing.
355  *
356  * The return value is the rounded version of the @j parameter.
357  */
358 unsigned long __round_jiffies_relative(unsigned long j, int cpu)
359 {
360 	unsigned long j0 = jiffies;
361 
362 	/* Use j0 because jiffies might change while we run */
363 	return round_jiffies_common(j + j0, cpu, false) - j0;
364 }
365 EXPORT_SYMBOL_GPL(__round_jiffies_relative);
366 
367 /**
368  * round_jiffies - function to round jiffies to a full second
369  * @j: the time in (absolute) jiffies that should be rounded
370  *
371  * round_jiffies() rounds an absolute time in the future (in jiffies)
372  * up or down to (approximately) full seconds. This is useful for timers
373  * for which the exact time they fire does not matter too much, as long as
374  * they fire approximately every X seconds.
375  *
376  * By rounding these timers to whole seconds, all such timers will fire
377  * at the same time, rather than at various times spread out. The goal
378  * of this is to have the CPU wake up less, which saves power.
379  *
380  * The return value is the rounded version of the @j parameter.
381  */
382 unsigned long round_jiffies(unsigned long j)
383 {
384 	return round_jiffies_common(j, raw_smp_processor_id(), false);
385 }
386 EXPORT_SYMBOL_GPL(round_jiffies);
387 
388 /**
389  * round_jiffies_relative - function to round jiffies to a full second
390  * @j: the time in (relative) jiffies that should be rounded
391  *
392  * round_jiffies_relative() rounds a time delta  in the future (in jiffies)
393  * up or down to (approximately) full seconds. This is useful for timers
394  * for which the exact time they fire does not matter too much, as long as
395  * they fire approximately every X seconds.
396  *
397  * By rounding these timers to whole seconds, all such timers will fire
398  * at the same time, rather than at various times spread out. The goal
399  * of this is to have the CPU wake up less, which saves power.
400  *
401  * The return value is the rounded version of the @j parameter.
402  */
403 unsigned long round_jiffies_relative(unsigned long j)
404 {
405 	return __round_jiffies_relative(j, raw_smp_processor_id());
406 }
407 EXPORT_SYMBOL_GPL(round_jiffies_relative);
408 
409 /**
410  * __round_jiffies_up - function to round jiffies up to a full second
411  * @j: the time in (absolute) jiffies that should be rounded
412  * @cpu: the processor number on which the timeout will happen
413  *
414  * This is the same as __round_jiffies() except that it will never
415  * round down.  This is useful for timeouts for which the exact time
416  * of firing does not matter too much, as long as they don't fire too
417  * early.
418  */
419 unsigned long __round_jiffies_up(unsigned long j, int cpu)
420 {
421 	return round_jiffies_common(j, cpu, true);
422 }
423 EXPORT_SYMBOL_GPL(__round_jiffies_up);
424 
425 /**
426  * __round_jiffies_up_relative - function to round jiffies up to a full second
427  * @j: the time in (relative) jiffies that should be rounded
428  * @cpu: the processor number on which the timeout will happen
429  *
430  * This is the same as __round_jiffies_relative() except that it will never
431  * round down.  This is useful for timeouts for which the exact time
432  * of firing does not matter too much, as long as they don't fire too
433  * early.
434  */
435 unsigned long __round_jiffies_up_relative(unsigned long j, int cpu)
436 {
437 	unsigned long j0 = jiffies;
438 
439 	/* Use j0 because jiffies might change while we run */
440 	return round_jiffies_common(j + j0, cpu, true) - j0;
441 }
442 EXPORT_SYMBOL_GPL(__round_jiffies_up_relative);
443 
444 /**
445  * round_jiffies_up - function to round jiffies up to a full second
446  * @j: the time in (absolute) jiffies that should be rounded
447  *
448  * This is the same as round_jiffies() except that it will never
449  * round down.  This is useful for timeouts for which the exact time
450  * of firing does not matter too much, as long as they don't fire too
451  * early.
452  */
453 unsigned long round_jiffies_up(unsigned long j)
454 {
455 	return round_jiffies_common(j, raw_smp_processor_id(), true);
456 }
457 EXPORT_SYMBOL_GPL(round_jiffies_up);
458 
459 /**
460  * round_jiffies_up_relative - function to round jiffies up to a full second
461  * @j: the time in (relative) jiffies that should be rounded
462  *
463  * This is the same as round_jiffies_relative() except that it will never
464  * round down.  This is useful for timeouts for which the exact time
465  * of firing does not matter too much, as long as they don't fire too
466  * early.
467  */
468 unsigned long round_jiffies_up_relative(unsigned long j)
469 {
470 	return __round_jiffies_up_relative(j, raw_smp_processor_id());
471 }
472 EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
473 
474 
475 static inline unsigned int timer_get_idx(struct timer_list *timer)
476 {
477 	return (timer->flags & TIMER_ARRAYMASK) >> TIMER_ARRAYSHIFT;
478 }
479 
480 static inline void timer_set_idx(struct timer_list *timer, unsigned int idx)
481 {
482 	timer->flags = (timer->flags & ~TIMER_ARRAYMASK) |
483 			idx << TIMER_ARRAYSHIFT;
484 }
485 
486 /*
487  * Helper function to calculate the array index for a given expiry
488  * time.
489  */
490 static inline unsigned calc_index(unsigned expires, unsigned lvl)
491 {
492 	expires = (expires + LVL_GRAN(lvl)) >> LVL_SHIFT(lvl);
493 	return LVL_OFFS(lvl) + (expires & LVL_MASK);
494 }
495 
496 static int calc_wheel_index(unsigned long expires, unsigned long clk)
497 {
498 	unsigned long delta = expires - clk;
499 	unsigned int idx;
500 
501 	if (delta < LVL_START(1)) {
502 		idx = calc_index(expires, 0);
503 	} else if (delta < LVL_START(2)) {
504 		idx = calc_index(expires, 1);
505 	} else if (delta < LVL_START(3)) {
506 		idx = calc_index(expires, 2);
507 	} else if (delta < LVL_START(4)) {
508 		idx = calc_index(expires, 3);
509 	} else if (delta < LVL_START(5)) {
510 		idx = calc_index(expires, 4);
511 	} else if (delta < LVL_START(6)) {
512 		idx = calc_index(expires, 5);
513 	} else if (delta < LVL_START(7)) {
514 		idx = calc_index(expires, 6);
515 	} else if (LVL_DEPTH > 8 && delta < LVL_START(8)) {
516 		idx = calc_index(expires, 7);
517 	} else if ((long) delta < 0) {
518 		idx = clk & LVL_MASK;
519 	} else {
520 		/*
521 		 * Force expire obscene large timeouts to expire at the
522 		 * capacity limit of the wheel.
523 		 */
524 		if (expires >= WHEEL_TIMEOUT_CUTOFF)
525 			expires = WHEEL_TIMEOUT_MAX;
526 
527 		idx = calc_index(expires, LVL_DEPTH - 1);
528 	}
529 	return idx;
530 }
531 
532 /*
533  * Enqueue the timer into the hash bucket, mark it pending in
534  * the bitmap and store the index in the timer flags.
535  */
536 static void enqueue_timer(struct timer_base *base, struct timer_list *timer,
537 			  unsigned int idx)
538 {
539 	hlist_add_head(&timer->entry, base->vectors + idx);
540 	__set_bit(idx, base->pending_map);
541 	timer_set_idx(timer, idx);
542 
543 	trace_timer_start(timer, timer->expires, timer->flags);
544 }
545 
546 static void
547 __internal_add_timer(struct timer_base *base, struct timer_list *timer)
548 {
549 	unsigned int idx;
550 
551 	idx = calc_wheel_index(timer->expires, base->clk);
552 	enqueue_timer(base, timer, idx);
553 }
554 
555 static void
556 trigger_dyntick_cpu(struct timer_base *base, struct timer_list *timer)
557 {
558 	if (!is_timers_nohz_active())
559 		return;
560 
561 	/*
562 	 * TODO: This wants some optimizing similar to the code below, but we
563 	 * will do that when we switch from push to pull for deferrable timers.
564 	 */
565 	if (timer->flags & TIMER_DEFERRABLE) {
566 		if (tick_nohz_full_cpu(base->cpu))
567 			wake_up_nohz_cpu(base->cpu);
568 		return;
569 	}
570 
571 	/*
572 	 * We might have to IPI the remote CPU if the base is idle and the
573 	 * timer is not deferrable. If the other CPU is on the way to idle
574 	 * then it can't set base->is_idle as we hold the base lock:
575 	 */
576 	if (!base->is_idle)
577 		return;
578 
579 	/* Check whether this is the new first expiring timer: */
580 	if (time_after_eq(timer->expires, base->next_expiry))
581 		return;
582 
583 	/*
584 	 * Set the next expiry time and kick the CPU so it can reevaluate the
585 	 * wheel:
586 	 */
587 	base->next_expiry = timer->expires;
588 	wake_up_nohz_cpu(base->cpu);
589 }
590 
591 static void
592 internal_add_timer(struct timer_base *base, struct timer_list *timer)
593 {
594 	__internal_add_timer(base, timer);
595 	trigger_dyntick_cpu(base, timer);
596 }
597 
598 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
599 
600 static struct debug_obj_descr timer_debug_descr;
601 
602 static void *timer_debug_hint(void *addr)
603 {
604 	return ((struct timer_list *) addr)->function;
605 }
606 
607 static bool timer_is_static_object(void *addr)
608 {
609 	struct timer_list *timer = addr;
610 
611 	return (timer->entry.pprev == NULL &&
612 		timer->entry.next == TIMER_ENTRY_STATIC);
613 }
614 
615 /*
616  * fixup_init is called when:
617  * - an active object is initialized
618  */
619 static bool timer_fixup_init(void *addr, enum debug_obj_state state)
620 {
621 	struct timer_list *timer = addr;
622 
623 	switch (state) {
624 	case ODEBUG_STATE_ACTIVE:
625 		del_timer_sync(timer);
626 		debug_object_init(timer, &timer_debug_descr);
627 		return true;
628 	default:
629 		return false;
630 	}
631 }
632 
633 /* Stub timer callback for improperly used timers. */
634 static void stub_timer(struct timer_list *unused)
635 {
636 	WARN_ON(1);
637 }
638 
639 /*
640  * fixup_activate is called when:
641  * - an active object is activated
642  * - an unknown non-static object is activated
643  */
644 static bool timer_fixup_activate(void *addr, enum debug_obj_state state)
645 {
646 	struct timer_list *timer = addr;
647 
648 	switch (state) {
649 	case ODEBUG_STATE_NOTAVAILABLE:
650 		timer_setup(timer, stub_timer, 0);
651 		return true;
652 
653 	case ODEBUG_STATE_ACTIVE:
654 		WARN_ON(1);
655 		/* fall through */
656 	default:
657 		return false;
658 	}
659 }
660 
661 /*
662  * fixup_free is called when:
663  * - an active object is freed
664  */
665 static bool timer_fixup_free(void *addr, enum debug_obj_state state)
666 {
667 	struct timer_list *timer = addr;
668 
669 	switch (state) {
670 	case ODEBUG_STATE_ACTIVE:
671 		del_timer_sync(timer);
672 		debug_object_free(timer, &timer_debug_descr);
673 		return true;
674 	default:
675 		return false;
676 	}
677 }
678 
679 /*
680  * fixup_assert_init is called when:
681  * - an untracked/uninit-ed object is found
682  */
683 static bool timer_fixup_assert_init(void *addr, enum debug_obj_state state)
684 {
685 	struct timer_list *timer = addr;
686 
687 	switch (state) {
688 	case ODEBUG_STATE_NOTAVAILABLE:
689 		timer_setup(timer, stub_timer, 0);
690 		return true;
691 	default:
692 		return false;
693 	}
694 }
695 
696 static struct debug_obj_descr timer_debug_descr = {
697 	.name			= "timer_list",
698 	.debug_hint		= timer_debug_hint,
699 	.is_static_object	= timer_is_static_object,
700 	.fixup_init		= timer_fixup_init,
701 	.fixup_activate		= timer_fixup_activate,
702 	.fixup_free		= timer_fixup_free,
703 	.fixup_assert_init	= timer_fixup_assert_init,
704 };
705 
706 static inline void debug_timer_init(struct timer_list *timer)
707 {
708 	debug_object_init(timer, &timer_debug_descr);
709 }
710 
711 static inline void debug_timer_activate(struct timer_list *timer)
712 {
713 	debug_object_activate(timer, &timer_debug_descr);
714 }
715 
716 static inline void debug_timer_deactivate(struct timer_list *timer)
717 {
718 	debug_object_deactivate(timer, &timer_debug_descr);
719 }
720 
721 static inline void debug_timer_free(struct timer_list *timer)
722 {
723 	debug_object_free(timer, &timer_debug_descr);
724 }
725 
726 static inline void debug_timer_assert_init(struct timer_list *timer)
727 {
728 	debug_object_assert_init(timer, &timer_debug_descr);
729 }
730 
731 static void do_init_timer(struct timer_list *timer,
732 			  void (*func)(struct timer_list *),
733 			  unsigned int flags,
734 			  const char *name, struct lock_class_key *key);
735 
736 void init_timer_on_stack_key(struct timer_list *timer,
737 			     void (*func)(struct timer_list *),
738 			     unsigned int flags,
739 			     const char *name, struct lock_class_key *key)
740 {
741 	debug_object_init_on_stack(timer, &timer_debug_descr);
742 	do_init_timer(timer, func, flags, name, key);
743 }
744 EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
745 
746 void destroy_timer_on_stack(struct timer_list *timer)
747 {
748 	debug_object_free(timer, &timer_debug_descr);
749 }
750 EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
751 
752 #else
753 static inline void debug_timer_init(struct timer_list *timer) { }
754 static inline void debug_timer_activate(struct timer_list *timer) { }
755 static inline void debug_timer_deactivate(struct timer_list *timer) { }
756 static inline void debug_timer_assert_init(struct timer_list *timer) { }
757 #endif
758 
759 static inline void debug_init(struct timer_list *timer)
760 {
761 	debug_timer_init(timer);
762 	trace_timer_init(timer);
763 }
764 
765 static inline void debug_deactivate(struct timer_list *timer)
766 {
767 	debug_timer_deactivate(timer);
768 	trace_timer_cancel(timer);
769 }
770 
771 static inline void debug_assert_init(struct timer_list *timer)
772 {
773 	debug_timer_assert_init(timer);
774 }
775 
776 static void do_init_timer(struct timer_list *timer,
777 			  void (*func)(struct timer_list *),
778 			  unsigned int flags,
779 			  const char *name, struct lock_class_key *key)
780 {
781 	timer->entry.pprev = NULL;
782 	timer->function = func;
783 	timer->flags = flags | raw_smp_processor_id();
784 	lockdep_init_map(&timer->lockdep_map, name, key, 0);
785 }
786 
787 /**
788  * init_timer_key - initialize a timer
789  * @timer: the timer to be initialized
790  * @func: timer callback function
791  * @flags: timer flags
792  * @name: name of the timer
793  * @key: lockdep class key of the fake lock used for tracking timer
794  *       sync lock dependencies
795  *
796  * init_timer_key() must be done to a timer prior calling *any* of the
797  * other timer functions.
798  */
799 void init_timer_key(struct timer_list *timer,
800 		    void (*func)(struct timer_list *), unsigned int flags,
801 		    const char *name, struct lock_class_key *key)
802 {
803 	debug_init(timer);
804 	do_init_timer(timer, func, flags, name, key);
805 }
806 EXPORT_SYMBOL(init_timer_key);
807 
808 static inline void detach_timer(struct timer_list *timer, bool clear_pending)
809 {
810 	struct hlist_node *entry = &timer->entry;
811 
812 	debug_deactivate(timer);
813 
814 	__hlist_del(entry);
815 	if (clear_pending)
816 		entry->pprev = NULL;
817 	entry->next = LIST_POISON2;
818 }
819 
820 static int detach_if_pending(struct timer_list *timer, struct timer_base *base,
821 			     bool clear_pending)
822 {
823 	unsigned idx = timer_get_idx(timer);
824 
825 	if (!timer_pending(timer))
826 		return 0;
827 
828 	if (hlist_is_singular_node(&timer->entry, base->vectors + idx))
829 		__clear_bit(idx, base->pending_map);
830 
831 	detach_timer(timer, clear_pending);
832 	return 1;
833 }
834 
835 static inline struct timer_base *get_timer_cpu_base(u32 tflags, u32 cpu)
836 {
837 	struct timer_base *base = per_cpu_ptr(&timer_bases[BASE_STD], cpu);
838 
839 	/*
840 	 * If the timer is deferrable and NO_HZ_COMMON is set then we need
841 	 * to use the deferrable base.
842 	 */
843 	if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE))
844 		base = per_cpu_ptr(&timer_bases[BASE_DEF], cpu);
845 	return base;
846 }
847 
848 static inline struct timer_base *get_timer_this_cpu_base(u32 tflags)
849 {
850 	struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
851 
852 	/*
853 	 * If the timer is deferrable and NO_HZ_COMMON is set then we need
854 	 * to use the deferrable base.
855 	 */
856 	if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE))
857 		base = this_cpu_ptr(&timer_bases[BASE_DEF]);
858 	return base;
859 }
860 
861 static inline struct timer_base *get_timer_base(u32 tflags)
862 {
863 	return get_timer_cpu_base(tflags, tflags & TIMER_CPUMASK);
864 }
865 
866 static inline struct timer_base *
867 get_target_base(struct timer_base *base, unsigned tflags)
868 {
869 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
870 	if (static_branch_likely(&timers_migration_enabled) &&
871 	    !(tflags & TIMER_PINNED))
872 		return get_timer_cpu_base(tflags, get_nohz_timer_target());
873 #endif
874 	return get_timer_this_cpu_base(tflags);
875 }
876 
877 static inline void forward_timer_base(struct timer_base *base)
878 {
879 #ifdef CONFIG_NO_HZ_COMMON
880 	unsigned long jnow;
881 
882 	/*
883 	 * We only forward the base when we are idle or have just come out of
884 	 * idle (must_forward_clk logic), and have a delta between base clock
885 	 * and jiffies. In the common case, run_timers will take care of it.
886 	 */
887 	if (likely(!base->must_forward_clk))
888 		return;
889 
890 	jnow = READ_ONCE(jiffies);
891 	base->must_forward_clk = base->is_idle;
892 	if ((long)(jnow - base->clk) < 2)
893 		return;
894 
895 	/*
896 	 * If the next expiry value is > jiffies, then we fast forward to
897 	 * jiffies otherwise we forward to the next expiry value.
898 	 */
899 	if (time_after(base->next_expiry, jnow))
900 		base->clk = jnow;
901 	else
902 		base->clk = base->next_expiry;
903 #endif
904 }
905 
906 
907 /*
908  * We are using hashed locking: Holding per_cpu(timer_bases[x]).lock means
909  * that all timers which are tied to this base are locked, and the base itself
910  * is locked too.
911  *
912  * So __run_timers/migrate_timers can safely modify all timers which could
913  * be found in the base->vectors array.
914  *
915  * When a timer is migrating then the TIMER_MIGRATING flag is set and we need
916  * to wait until the migration is done.
917  */
918 static struct timer_base *lock_timer_base(struct timer_list *timer,
919 					  unsigned long *flags)
920 	__acquires(timer->base->lock)
921 {
922 	for (;;) {
923 		struct timer_base *base;
924 		u32 tf;
925 
926 		/*
927 		 * We need to use READ_ONCE() here, otherwise the compiler
928 		 * might re-read @tf between the check for TIMER_MIGRATING
929 		 * and spin_lock().
930 		 */
931 		tf = READ_ONCE(timer->flags);
932 
933 		if (!(tf & TIMER_MIGRATING)) {
934 			base = get_timer_base(tf);
935 			raw_spin_lock_irqsave(&base->lock, *flags);
936 			if (timer->flags == tf)
937 				return base;
938 			raw_spin_unlock_irqrestore(&base->lock, *flags);
939 		}
940 		cpu_relax();
941 	}
942 }
943 
944 #define MOD_TIMER_PENDING_ONLY		0x01
945 #define MOD_TIMER_REDUCE		0x02
946 #define MOD_TIMER_NOTPENDING		0x04
947 
948 static inline int
949 __mod_timer(struct timer_list *timer, unsigned long expires, unsigned int options)
950 {
951 	struct timer_base *base, *new_base;
952 	unsigned int idx = UINT_MAX;
953 	unsigned long clk = 0, flags;
954 	int ret = 0;
955 
956 	BUG_ON(!timer->function);
957 
958 	/*
959 	 * This is a common optimization triggered by the networking code - if
960 	 * the timer is re-modified to have the same timeout or ends up in the
961 	 * same array bucket then just return:
962 	 */
963 	if (!(options & MOD_TIMER_NOTPENDING) && timer_pending(timer)) {
964 		/*
965 		 * The downside of this optimization is that it can result in
966 		 * larger granularity than you would get from adding a new
967 		 * timer with this expiry.
968 		 */
969 		long diff = timer->expires - expires;
970 
971 		if (!diff)
972 			return 1;
973 		if (options & MOD_TIMER_REDUCE && diff <= 0)
974 			return 1;
975 
976 		/*
977 		 * We lock timer base and calculate the bucket index right
978 		 * here. If the timer ends up in the same bucket, then we
979 		 * just update the expiry time and avoid the whole
980 		 * dequeue/enqueue dance.
981 		 */
982 		base = lock_timer_base(timer, &flags);
983 		forward_timer_base(base);
984 
985 		if (timer_pending(timer) && (options & MOD_TIMER_REDUCE) &&
986 		    time_before_eq(timer->expires, expires)) {
987 			ret = 1;
988 			goto out_unlock;
989 		}
990 
991 		clk = base->clk;
992 		idx = calc_wheel_index(expires, clk);
993 
994 		/*
995 		 * Retrieve and compare the array index of the pending
996 		 * timer. If it matches set the expiry to the new value so a
997 		 * subsequent call will exit in the expires check above.
998 		 */
999 		if (idx == timer_get_idx(timer)) {
1000 			if (!(options & MOD_TIMER_REDUCE))
1001 				timer->expires = expires;
1002 			else if (time_after(timer->expires, expires))
1003 				timer->expires = expires;
1004 			ret = 1;
1005 			goto out_unlock;
1006 		}
1007 	} else {
1008 		base = lock_timer_base(timer, &flags);
1009 		forward_timer_base(base);
1010 	}
1011 
1012 	ret = detach_if_pending(timer, base, false);
1013 	if (!ret && (options & MOD_TIMER_PENDING_ONLY))
1014 		goto out_unlock;
1015 
1016 	new_base = get_target_base(base, timer->flags);
1017 
1018 	if (base != new_base) {
1019 		/*
1020 		 * We are trying to schedule the timer on the new base.
1021 		 * However we can't change timer's base while it is running,
1022 		 * otherwise del_timer_sync() can't detect that the timer's
1023 		 * handler yet has not finished. This also guarantees that the
1024 		 * timer is serialized wrt itself.
1025 		 */
1026 		if (likely(base->running_timer != timer)) {
1027 			/* See the comment in lock_timer_base() */
1028 			timer->flags |= TIMER_MIGRATING;
1029 
1030 			raw_spin_unlock(&base->lock);
1031 			base = new_base;
1032 			raw_spin_lock(&base->lock);
1033 			WRITE_ONCE(timer->flags,
1034 				   (timer->flags & ~TIMER_BASEMASK) | base->cpu);
1035 			forward_timer_base(base);
1036 		}
1037 	}
1038 
1039 	debug_timer_activate(timer);
1040 
1041 	timer->expires = expires;
1042 	/*
1043 	 * If 'idx' was calculated above and the base time did not advance
1044 	 * between calculating 'idx' and possibly switching the base, only
1045 	 * enqueue_timer() and trigger_dyntick_cpu() is required. Otherwise
1046 	 * we need to (re)calculate the wheel index via
1047 	 * internal_add_timer().
1048 	 */
1049 	if (idx != UINT_MAX && clk == base->clk) {
1050 		enqueue_timer(base, timer, idx);
1051 		trigger_dyntick_cpu(base, timer);
1052 	} else {
1053 		internal_add_timer(base, timer);
1054 	}
1055 
1056 out_unlock:
1057 	raw_spin_unlock_irqrestore(&base->lock, flags);
1058 
1059 	return ret;
1060 }
1061 
1062 /**
1063  * mod_timer_pending - modify a pending timer's timeout
1064  * @timer: the pending timer to be modified
1065  * @expires: new timeout in jiffies
1066  *
1067  * mod_timer_pending() is the same for pending timers as mod_timer(),
1068  * but will not re-activate and modify already deleted timers.
1069  *
1070  * It is useful for unserialized use of timers.
1071  */
1072 int mod_timer_pending(struct timer_list *timer, unsigned long expires)
1073 {
1074 	return __mod_timer(timer, expires, MOD_TIMER_PENDING_ONLY);
1075 }
1076 EXPORT_SYMBOL(mod_timer_pending);
1077 
1078 /**
1079  * mod_timer - modify a timer's timeout
1080  * @timer: the timer to be modified
1081  * @expires: new timeout in jiffies
1082  *
1083  * mod_timer() is a more efficient way to update the expire field of an
1084  * active timer (if the timer is inactive it will be activated)
1085  *
1086  * mod_timer(timer, expires) is equivalent to:
1087  *
1088  *     del_timer(timer); timer->expires = expires; add_timer(timer);
1089  *
1090  * Note that if there are multiple unserialized concurrent users of the
1091  * same timer, then mod_timer() is the only safe way to modify the timeout,
1092  * since add_timer() cannot modify an already running timer.
1093  *
1094  * The function returns whether it has modified a pending timer or not.
1095  * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
1096  * active timer returns 1.)
1097  */
1098 int mod_timer(struct timer_list *timer, unsigned long expires)
1099 {
1100 	return __mod_timer(timer, expires, 0);
1101 }
1102 EXPORT_SYMBOL(mod_timer);
1103 
1104 /**
1105  * timer_reduce - Modify a timer's timeout if it would reduce the timeout
1106  * @timer:	The timer to be modified
1107  * @expires:	New timeout in jiffies
1108  *
1109  * timer_reduce() is very similar to mod_timer(), except that it will only
1110  * modify a running timer if that would reduce the expiration time (it will
1111  * start a timer that isn't running).
1112  */
1113 int timer_reduce(struct timer_list *timer, unsigned long expires)
1114 {
1115 	return __mod_timer(timer, expires, MOD_TIMER_REDUCE);
1116 }
1117 EXPORT_SYMBOL(timer_reduce);
1118 
1119 /**
1120  * add_timer - start a timer
1121  * @timer: the timer to be added
1122  *
1123  * The kernel will do a ->function(@timer) callback from the
1124  * timer interrupt at the ->expires point in the future. The
1125  * current time is 'jiffies'.
1126  *
1127  * The timer's ->expires, ->function fields must be set prior calling this
1128  * function.
1129  *
1130  * Timers with an ->expires field in the past will be executed in the next
1131  * timer tick.
1132  */
1133 void add_timer(struct timer_list *timer)
1134 {
1135 	BUG_ON(timer_pending(timer));
1136 	__mod_timer(timer, timer->expires, MOD_TIMER_NOTPENDING);
1137 }
1138 EXPORT_SYMBOL(add_timer);
1139 
1140 /**
1141  * add_timer_on - start a timer on a particular CPU
1142  * @timer: the timer to be added
1143  * @cpu: the CPU to start it on
1144  *
1145  * This is not very scalable on SMP. Double adds are not possible.
1146  */
1147 void add_timer_on(struct timer_list *timer, int cpu)
1148 {
1149 	struct timer_base *new_base, *base;
1150 	unsigned long flags;
1151 
1152 	BUG_ON(timer_pending(timer) || !timer->function);
1153 
1154 	new_base = get_timer_cpu_base(timer->flags, cpu);
1155 
1156 	/*
1157 	 * If @timer was on a different CPU, it should be migrated with the
1158 	 * old base locked to prevent other operations proceeding with the
1159 	 * wrong base locked.  See lock_timer_base().
1160 	 */
1161 	base = lock_timer_base(timer, &flags);
1162 	if (base != new_base) {
1163 		timer->flags |= TIMER_MIGRATING;
1164 
1165 		raw_spin_unlock(&base->lock);
1166 		base = new_base;
1167 		raw_spin_lock(&base->lock);
1168 		WRITE_ONCE(timer->flags,
1169 			   (timer->flags & ~TIMER_BASEMASK) | cpu);
1170 	}
1171 	forward_timer_base(base);
1172 
1173 	debug_timer_activate(timer);
1174 	internal_add_timer(base, timer);
1175 	raw_spin_unlock_irqrestore(&base->lock, flags);
1176 }
1177 EXPORT_SYMBOL_GPL(add_timer_on);
1178 
1179 /**
1180  * del_timer - deactivate a timer.
1181  * @timer: the timer to be deactivated
1182  *
1183  * del_timer() deactivates a timer - this works on both active and inactive
1184  * timers.
1185  *
1186  * The function returns whether it has deactivated a pending timer or not.
1187  * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
1188  * active timer returns 1.)
1189  */
1190 int del_timer(struct timer_list *timer)
1191 {
1192 	struct timer_base *base;
1193 	unsigned long flags;
1194 	int ret = 0;
1195 
1196 	debug_assert_init(timer);
1197 
1198 	if (timer_pending(timer)) {
1199 		base = lock_timer_base(timer, &flags);
1200 		ret = detach_if_pending(timer, base, true);
1201 		raw_spin_unlock_irqrestore(&base->lock, flags);
1202 	}
1203 
1204 	return ret;
1205 }
1206 EXPORT_SYMBOL(del_timer);
1207 
1208 /**
1209  * try_to_del_timer_sync - Try to deactivate a timer
1210  * @timer: timer to delete
1211  *
1212  * This function tries to deactivate a timer. Upon successful (ret >= 0)
1213  * exit the timer is not queued and the handler is not running on any CPU.
1214  */
1215 int try_to_del_timer_sync(struct timer_list *timer)
1216 {
1217 	struct timer_base *base;
1218 	unsigned long flags;
1219 	int ret = -1;
1220 
1221 	debug_assert_init(timer);
1222 
1223 	base = lock_timer_base(timer, &flags);
1224 
1225 	if (base->running_timer != timer)
1226 		ret = detach_if_pending(timer, base, true);
1227 
1228 	raw_spin_unlock_irqrestore(&base->lock, flags);
1229 
1230 	return ret;
1231 }
1232 EXPORT_SYMBOL(try_to_del_timer_sync);
1233 
1234 #ifdef CONFIG_PREEMPT_RT
1235 static __init void timer_base_init_expiry_lock(struct timer_base *base)
1236 {
1237 	spin_lock_init(&base->expiry_lock);
1238 }
1239 
1240 static inline void timer_base_lock_expiry(struct timer_base *base)
1241 {
1242 	spin_lock(&base->expiry_lock);
1243 }
1244 
1245 static inline void timer_base_unlock_expiry(struct timer_base *base)
1246 {
1247 	spin_unlock(&base->expiry_lock);
1248 }
1249 
1250 /*
1251  * The counterpart to del_timer_wait_running().
1252  *
1253  * If there is a waiter for base->expiry_lock, then it was waiting for the
1254  * timer callback to finish. Drop expiry_lock and reaquire it. That allows
1255  * the waiter to acquire the lock and make progress.
1256  */
1257 static void timer_sync_wait_running(struct timer_base *base)
1258 {
1259 	if (atomic_read(&base->timer_waiters)) {
1260 		spin_unlock(&base->expiry_lock);
1261 		spin_lock(&base->expiry_lock);
1262 	}
1263 }
1264 
1265 /*
1266  * This function is called on PREEMPT_RT kernels when the fast path
1267  * deletion of a timer failed because the timer callback function was
1268  * running.
1269  *
1270  * This prevents priority inversion, if the softirq thread on a remote CPU
1271  * got preempted, and it prevents a life lock when the task which tries to
1272  * delete a timer preempted the softirq thread running the timer callback
1273  * function.
1274  */
1275 static void del_timer_wait_running(struct timer_list *timer)
1276 {
1277 	u32 tf;
1278 
1279 	tf = READ_ONCE(timer->flags);
1280 	if (!(tf & TIMER_MIGRATING)) {
1281 		struct timer_base *base = get_timer_base(tf);
1282 
1283 		/*
1284 		 * Mark the base as contended and grab the expiry lock,
1285 		 * which is held by the softirq across the timer
1286 		 * callback. Drop the lock immediately so the softirq can
1287 		 * expire the next timer. In theory the timer could already
1288 		 * be running again, but that's more than unlikely and just
1289 		 * causes another wait loop.
1290 		 */
1291 		atomic_inc(&base->timer_waiters);
1292 		spin_lock_bh(&base->expiry_lock);
1293 		atomic_dec(&base->timer_waiters);
1294 		spin_unlock_bh(&base->expiry_lock);
1295 	}
1296 }
1297 #else
1298 static inline void timer_base_init_expiry_lock(struct timer_base *base) { }
1299 static inline void timer_base_lock_expiry(struct timer_base *base) { }
1300 static inline void timer_base_unlock_expiry(struct timer_base *base) { }
1301 static inline void timer_sync_wait_running(struct timer_base *base) { }
1302 static inline void del_timer_wait_running(struct timer_list *timer) { }
1303 #endif
1304 
1305 #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT_RT)
1306 /**
1307  * del_timer_sync - deactivate a timer and wait for the handler to finish.
1308  * @timer: the timer to be deactivated
1309  *
1310  * This function only differs from del_timer() on SMP: besides deactivating
1311  * the timer it also makes sure the handler has finished executing on other
1312  * CPUs.
1313  *
1314  * Synchronization rules: Callers must prevent restarting of the timer,
1315  * otherwise this function is meaningless. It must not be called from
1316  * interrupt contexts unless the timer is an irqsafe one. The caller must
1317  * not hold locks which would prevent completion of the timer's
1318  * handler. The timer's handler must not call add_timer_on(). Upon exit the
1319  * timer is not queued and the handler is not running on any CPU.
1320  *
1321  * Note: For !irqsafe timers, you must not hold locks that are held in
1322  *   interrupt context while calling this function. Even if the lock has
1323  *   nothing to do with the timer in question.  Here's why::
1324  *
1325  *    CPU0                             CPU1
1326  *    ----                             ----
1327  *                                     <SOFTIRQ>
1328  *                                       call_timer_fn();
1329  *                                       base->running_timer = mytimer;
1330  *    spin_lock_irq(somelock);
1331  *                                     <IRQ>
1332  *                                        spin_lock(somelock);
1333  *    del_timer_sync(mytimer);
1334  *    while (base->running_timer == mytimer);
1335  *
1336  * Now del_timer_sync() will never return and never release somelock.
1337  * The interrupt on the other CPU is waiting to grab somelock but
1338  * it has interrupted the softirq that CPU0 is waiting to finish.
1339  *
1340  * The function returns whether it has deactivated a pending timer or not.
1341  */
1342 int del_timer_sync(struct timer_list *timer)
1343 {
1344 	int ret;
1345 
1346 #ifdef CONFIG_LOCKDEP
1347 	unsigned long flags;
1348 
1349 	/*
1350 	 * If lockdep gives a backtrace here, please reference
1351 	 * the synchronization rules above.
1352 	 */
1353 	local_irq_save(flags);
1354 	lock_map_acquire(&timer->lockdep_map);
1355 	lock_map_release(&timer->lockdep_map);
1356 	local_irq_restore(flags);
1357 #endif
1358 	/*
1359 	 * don't use it in hardirq context, because it
1360 	 * could lead to deadlock.
1361 	 */
1362 	WARN_ON(in_irq() && !(timer->flags & TIMER_IRQSAFE));
1363 
1364 	do {
1365 		ret = try_to_del_timer_sync(timer);
1366 
1367 		if (unlikely(ret < 0)) {
1368 			del_timer_wait_running(timer);
1369 			cpu_relax();
1370 		}
1371 	} while (ret < 0);
1372 
1373 	return ret;
1374 }
1375 EXPORT_SYMBOL(del_timer_sync);
1376 #endif
1377 
1378 static void call_timer_fn(struct timer_list *timer,
1379 			  void (*fn)(struct timer_list *),
1380 			  unsigned long baseclk)
1381 {
1382 	int count = preempt_count();
1383 
1384 #ifdef CONFIG_LOCKDEP
1385 	/*
1386 	 * It is permissible to free the timer from inside the
1387 	 * function that is called from it, this we need to take into
1388 	 * account for lockdep too. To avoid bogus "held lock freed"
1389 	 * warnings as well as problems when looking into
1390 	 * timer->lockdep_map, make a copy and use that here.
1391 	 */
1392 	struct lockdep_map lockdep_map;
1393 
1394 	lockdep_copy_map(&lockdep_map, &timer->lockdep_map);
1395 #endif
1396 	/*
1397 	 * Couple the lock chain with the lock chain at
1398 	 * del_timer_sync() by acquiring the lock_map around the fn()
1399 	 * call here and in del_timer_sync().
1400 	 */
1401 	lock_map_acquire(&lockdep_map);
1402 
1403 	trace_timer_expire_entry(timer, baseclk);
1404 	fn(timer);
1405 	trace_timer_expire_exit(timer);
1406 
1407 	lock_map_release(&lockdep_map);
1408 
1409 	if (count != preempt_count()) {
1410 		WARN_ONCE(1, "timer: %pS preempt leak: %08x -> %08x\n",
1411 			  fn, count, preempt_count());
1412 		/*
1413 		 * Restore the preempt count. That gives us a decent
1414 		 * chance to survive and extract information. If the
1415 		 * callback kept a lock held, bad luck, but not worse
1416 		 * than the BUG() we had.
1417 		 */
1418 		preempt_count_set(count);
1419 	}
1420 }
1421 
1422 static void expire_timers(struct timer_base *base, struct hlist_head *head)
1423 {
1424 	/*
1425 	 * This value is required only for tracing. base->clk was
1426 	 * incremented directly before expire_timers was called. But expiry
1427 	 * is related to the old base->clk value.
1428 	 */
1429 	unsigned long baseclk = base->clk - 1;
1430 
1431 	while (!hlist_empty(head)) {
1432 		struct timer_list *timer;
1433 		void (*fn)(struct timer_list *);
1434 
1435 		timer = hlist_entry(head->first, struct timer_list, entry);
1436 
1437 		base->running_timer = timer;
1438 		detach_timer(timer, true);
1439 
1440 		fn = timer->function;
1441 
1442 		if (timer->flags & TIMER_IRQSAFE) {
1443 			raw_spin_unlock(&base->lock);
1444 			call_timer_fn(timer, fn, baseclk);
1445 			base->running_timer = NULL;
1446 			raw_spin_lock(&base->lock);
1447 		} else {
1448 			raw_spin_unlock_irq(&base->lock);
1449 			call_timer_fn(timer, fn, baseclk);
1450 			base->running_timer = NULL;
1451 			timer_sync_wait_running(base);
1452 			raw_spin_lock_irq(&base->lock);
1453 		}
1454 	}
1455 }
1456 
1457 static int __collect_expired_timers(struct timer_base *base,
1458 				    struct hlist_head *heads)
1459 {
1460 	unsigned long clk = base->clk;
1461 	struct hlist_head *vec;
1462 	int i, levels = 0;
1463 	unsigned int idx;
1464 
1465 	for (i = 0; i < LVL_DEPTH; i++) {
1466 		idx = (clk & LVL_MASK) + i * LVL_SIZE;
1467 
1468 		if (__test_and_clear_bit(idx, base->pending_map)) {
1469 			vec = base->vectors + idx;
1470 			hlist_move_list(vec, heads++);
1471 			levels++;
1472 		}
1473 		/* Is it time to look at the next level? */
1474 		if (clk & LVL_CLK_MASK)
1475 			break;
1476 		/* Shift clock for the next level granularity */
1477 		clk >>= LVL_CLK_SHIFT;
1478 	}
1479 	return levels;
1480 }
1481 
1482 #ifdef CONFIG_NO_HZ_COMMON
1483 /*
1484  * Find the next pending bucket of a level. Search from level start (@offset)
1485  * + @clk upwards and if nothing there, search from start of the level
1486  * (@offset) up to @offset + clk.
1487  */
1488 static int next_pending_bucket(struct timer_base *base, unsigned offset,
1489 			       unsigned clk)
1490 {
1491 	unsigned pos, start = offset + clk;
1492 	unsigned end = offset + LVL_SIZE;
1493 
1494 	pos = find_next_bit(base->pending_map, end, start);
1495 	if (pos < end)
1496 		return pos - start;
1497 
1498 	pos = find_next_bit(base->pending_map, start, offset);
1499 	return pos < start ? pos + LVL_SIZE - start : -1;
1500 }
1501 
1502 /*
1503  * Search the first expiring timer in the various clock levels. Caller must
1504  * hold base->lock.
1505  */
1506 static unsigned long __next_timer_interrupt(struct timer_base *base)
1507 {
1508 	unsigned long clk, next, adj;
1509 	unsigned lvl, offset = 0;
1510 
1511 	next = base->clk + NEXT_TIMER_MAX_DELTA;
1512 	clk = base->clk;
1513 	for (lvl = 0; lvl < LVL_DEPTH; lvl++, offset += LVL_SIZE) {
1514 		int pos = next_pending_bucket(base, offset, clk & LVL_MASK);
1515 
1516 		if (pos >= 0) {
1517 			unsigned long tmp = clk + (unsigned long) pos;
1518 
1519 			tmp <<= LVL_SHIFT(lvl);
1520 			if (time_before(tmp, next))
1521 				next = tmp;
1522 		}
1523 		/*
1524 		 * Clock for the next level. If the current level clock lower
1525 		 * bits are zero, we look at the next level as is. If not we
1526 		 * need to advance it by one because that's going to be the
1527 		 * next expiring bucket in that level. base->clk is the next
1528 		 * expiring jiffie. So in case of:
1529 		 *
1530 		 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1531 		 *  0    0    0    0    0    0
1532 		 *
1533 		 * we have to look at all levels @index 0. With
1534 		 *
1535 		 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1536 		 *  0    0    0    0    0    2
1537 		 *
1538 		 * LVL0 has the next expiring bucket @index 2. The upper
1539 		 * levels have the next expiring bucket @index 1.
1540 		 *
1541 		 * In case that the propagation wraps the next level the same
1542 		 * rules apply:
1543 		 *
1544 		 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1545 		 *  0    0    0    0    F    2
1546 		 *
1547 		 * So after looking at LVL0 we get:
1548 		 *
1549 		 * LVL5 LVL4 LVL3 LVL2 LVL1
1550 		 *  0    0    0    1    0
1551 		 *
1552 		 * So no propagation from LVL1 to LVL2 because that happened
1553 		 * with the add already, but then we need to propagate further
1554 		 * from LVL2 to LVL3.
1555 		 *
1556 		 * So the simple check whether the lower bits of the current
1557 		 * level are 0 or not is sufficient for all cases.
1558 		 */
1559 		adj = clk & LVL_CLK_MASK ? 1 : 0;
1560 		clk >>= LVL_CLK_SHIFT;
1561 		clk += adj;
1562 	}
1563 	return next;
1564 }
1565 
1566 /*
1567  * Check, if the next hrtimer event is before the next timer wheel
1568  * event:
1569  */
1570 static u64 cmp_next_hrtimer_event(u64 basem, u64 expires)
1571 {
1572 	u64 nextevt = hrtimer_get_next_event();
1573 
1574 	/*
1575 	 * If high resolution timers are enabled
1576 	 * hrtimer_get_next_event() returns KTIME_MAX.
1577 	 */
1578 	if (expires <= nextevt)
1579 		return expires;
1580 
1581 	/*
1582 	 * If the next timer is already expired, return the tick base
1583 	 * time so the tick is fired immediately.
1584 	 */
1585 	if (nextevt <= basem)
1586 		return basem;
1587 
1588 	/*
1589 	 * Round up to the next jiffie. High resolution timers are
1590 	 * off, so the hrtimers are expired in the tick and we need to
1591 	 * make sure that this tick really expires the timer to avoid
1592 	 * a ping pong of the nohz stop code.
1593 	 *
1594 	 * Use DIV_ROUND_UP_ULL to prevent gcc calling __divdi3
1595 	 */
1596 	return DIV_ROUND_UP_ULL(nextevt, TICK_NSEC) * TICK_NSEC;
1597 }
1598 
1599 /**
1600  * get_next_timer_interrupt - return the time (clock mono) of the next timer
1601  * @basej:	base time jiffies
1602  * @basem:	base time clock monotonic
1603  *
1604  * Returns the tick aligned clock monotonic time of the next pending
1605  * timer or KTIME_MAX if no timer is pending.
1606  */
1607 u64 get_next_timer_interrupt(unsigned long basej, u64 basem)
1608 {
1609 	struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1610 	u64 expires = KTIME_MAX;
1611 	unsigned long nextevt;
1612 	bool is_max_delta;
1613 
1614 	/*
1615 	 * Pretend that there is no timer pending if the cpu is offline.
1616 	 * Possible pending timers will be migrated later to an active cpu.
1617 	 */
1618 	if (cpu_is_offline(smp_processor_id()))
1619 		return expires;
1620 
1621 	raw_spin_lock(&base->lock);
1622 	nextevt = __next_timer_interrupt(base);
1623 	is_max_delta = (nextevt == base->clk + NEXT_TIMER_MAX_DELTA);
1624 	base->next_expiry = nextevt;
1625 	/*
1626 	 * We have a fresh next event. Check whether we can forward the
1627 	 * base. We can only do that when @basej is past base->clk
1628 	 * otherwise we might rewind base->clk.
1629 	 */
1630 	if (time_after(basej, base->clk)) {
1631 		if (time_after(nextevt, basej))
1632 			base->clk = basej;
1633 		else if (time_after(nextevt, base->clk))
1634 			base->clk = nextevt;
1635 	}
1636 
1637 	if (time_before_eq(nextevt, basej)) {
1638 		expires = basem;
1639 		base->is_idle = false;
1640 	} else {
1641 		if (!is_max_delta)
1642 			expires = basem + (u64)(nextevt - basej) * TICK_NSEC;
1643 		/*
1644 		 * If we expect to sleep more than a tick, mark the base idle.
1645 		 * Also the tick is stopped so any added timer must forward
1646 		 * the base clk itself to keep granularity small. This idle
1647 		 * logic is only maintained for the BASE_STD base, deferrable
1648 		 * timers may still see large granularity skew (by design).
1649 		 */
1650 		if ((expires - basem) > TICK_NSEC) {
1651 			base->must_forward_clk = true;
1652 			base->is_idle = true;
1653 		}
1654 	}
1655 	raw_spin_unlock(&base->lock);
1656 
1657 	return cmp_next_hrtimer_event(basem, expires);
1658 }
1659 
1660 /**
1661  * timer_clear_idle - Clear the idle state of the timer base
1662  *
1663  * Called with interrupts disabled
1664  */
1665 void timer_clear_idle(void)
1666 {
1667 	struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1668 
1669 	/*
1670 	 * We do this unlocked. The worst outcome is a remote enqueue sending
1671 	 * a pointless IPI, but taking the lock would just make the window for
1672 	 * sending the IPI a few instructions smaller for the cost of taking
1673 	 * the lock in the exit from idle path.
1674 	 */
1675 	base->is_idle = false;
1676 }
1677 
1678 static int collect_expired_timers(struct timer_base *base,
1679 				  struct hlist_head *heads)
1680 {
1681 	unsigned long now = READ_ONCE(jiffies);
1682 
1683 	/*
1684 	 * NOHZ optimization. After a long idle sleep we need to forward the
1685 	 * base to current jiffies. Avoid a loop by searching the bitfield for
1686 	 * the next expiring timer.
1687 	 */
1688 	if ((long)(now - base->clk) > 2) {
1689 		unsigned long next = __next_timer_interrupt(base);
1690 
1691 		/*
1692 		 * If the next timer is ahead of time forward to current
1693 		 * jiffies, otherwise forward to the next expiry time:
1694 		 */
1695 		if (time_after(next, now)) {
1696 			/*
1697 			 * The call site will increment base->clk and then
1698 			 * terminate the expiry loop immediately.
1699 			 */
1700 			base->clk = now;
1701 			return 0;
1702 		}
1703 		base->clk = next;
1704 	}
1705 	return __collect_expired_timers(base, heads);
1706 }
1707 #else
1708 static inline int collect_expired_timers(struct timer_base *base,
1709 					 struct hlist_head *heads)
1710 {
1711 	return __collect_expired_timers(base, heads);
1712 }
1713 #endif
1714 
1715 /*
1716  * Called from the timer interrupt handler to charge one tick to the current
1717  * process.  user_tick is 1 if the tick is user time, 0 for system.
1718  */
1719 void update_process_times(int user_tick)
1720 {
1721 	struct task_struct *p = current;
1722 
1723 	/* Note: this timer irq context must be accounted for as well. */
1724 	account_process_tick(p, user_tick);
1725 	run_local_timers();
1726 	rcu_sched_clock_irq(user_tick);
1727 #ifdef CONFIG_IRQ_WORK
1728 	if (in_irq())
1729 		irq_work_tick();
1730 #endif
1731 	scheduler_tick();
1732 	if (IS_ENABLED(CONFIG_POSIX_TIMERS))
1733 		run_posix_cpu_timers();
1734 }
1735 
1736 /**
1737  * __run_timers - run all expired timers (if any) on this CPU.
1738  * @base: the timer vector to be processed.
1739  */
1740 static inline void __run_timers(struct timer_base *base)
1741 {
1742 	struct hlist_head heads[LVL_DEPTH];
1743 	int levels;
1744 
1745 	if (!time_after_eq(jiffies, base->clk))
1746 		return;
1747 
1748 	timer_base_lock_expiry(base);
1749 	raw_spin_lock_irq(&base->lock);
1750 
1751 	/*
1752 	 * timer_base::must_forward_clk must be cleared before running
1753 	 * timers so that any timer functions that call mod_timer() will
1754 	 * not try to forward the base. Idle tracking / clock forwarding
1755 	 * logic is only used with BASE_STD timers.
1756 	 *
1757 	 * The must_forward_clk flag is cleared unconditionally also for
1758 	 * the deferrable base. The deferrable base is not affected by idle
1759 	 * tracking and never forwarded, so clearing the flag is a NOOP.
1760 	 *
1761 	 * The fact that the deferrable base is never forwarded can cause
1762 	 * large variations in granularity for deferrable timers, but they
1763 	 * can be deferred for long periods due to idle anyway.
1764 	 */
1765 	base->must_forward_clk = false;
1766 
1767 	while (time_after_eq(jiffies, base->clk)) {
1768 
1769 		levels = collect_expired_timers(base, heads);
1770 		base->clk++;
1771 
1772 		while (levels--)
1773 			expire_timers(base, heads + levels);
1774 	}
1775 	raw_spin_unlock_irq(&base->lock);
1776 	timer_base_unlock_expiry(base);
1777 }
1778 
1779 /*
1780  * This function runs timers and the timer-tq in bottom half context.
1781  */
1782 static __latent_entropy void run_timer_softirq(struct softirq_action *h)
1783 {
1784 	struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1785 
1786 	__run_timers(base);
1787 	if (IS_ENABLED(CONFIG_NO_HZ_COMMON))
1788 		__run_timers(this_cpu_ptr(&timer_bases[BASE_DEF]));
1789 }
1790 
1791 /*
1792  * Called by the local, per-CPU timer interrupt on SMP.
1793  */
1794 void run_local_timers(void)
1795 {
1796 	struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1797 
1798 	hrtimer_run_queues();
1799 	/* Raise the softirq only if required. */
1800 	if (time_before(jiffies, base->clk)) {
1801 		if (!IS_ENABLED(CONFIG_NO_HZ_COMMON))
1802 			return;
1803 		/* CPU is awake, so check the deferrable base. */
1804 		base++;
1805 		if (time_before(jiffies, base->clk))
1806 			return;
1807 	}
1808 	raise_softirq(TIMER_SOFTIRQ);
1809 }
1810 
1811 /*
1812  * Since schedule_timeout()'s timer is defined on the stack, it must store
1813  * the target task on the stack as well.
1814  */
1815 struct process_timer {
1816 	struct timer_list timer;
1817 	struct task_struct *task;
1818 };
1819 
1820 static void process_timeout(struct timer_list *t)
1821 {
1822 	struct process_timer *timeout = from_timer(timeout, t, timer);
1823 
1824 	wake_up_process(timeout->task);
1825 }
1826 
1827 /**
1828  * schedule_timeout - sleep until timeout
1829  * @timeout: timeout value in jiffies
1830  *
1831  * Make the current task sleep until @timeout jiffies have elapsed.
1832  * The function behavior depends on the current task state
1833  * (see also set_current_state() description):
1834  *
1835  * %TASK_RUNNING - the scheduler is called, but the task does not sleep
1836  * at all. That happens because sched_submit_work() does nothing for
1837  * tasks in %TASK_RUNNING state.
1838  *
1839  * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1840  * pass before the routine returns unless the current task is explicitly
1841  * woken up, (e.g. by wake_up_process()).
1842  *
1843  * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1844  * delivered to the current task or the current task is explicitly woken
1845  * up.
1846  *
1847  * The current task state is guaranteed to be %TASK_RUNNING when this
1848  * routine returns.
1849  *
1850  * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1851  * the CPU away without a bound on the timeout. In this case the return
1852  * value will be %MAX_SCHEDULE_TIMEOUT.
1853  *
1854  * Returns 0 when the timer has expired otherwise the remaining time in
1855  * jiffies will be returned. In all cases the return value is guaranteed
1856  * to be non-negative.
1857  */
1858 signed long __sched schedule_timeout(signed long timeout)
1859 {
1860 	struct process_timer timer;
1861 	unsigned long expire;
1862 
1863 	switch (timeout)
1864 	{
1865 	case MAX_SCHEDULE_TIMEOUT:
1866 		/*
1867 		 * These two special cases are useful to be comfortable
1868 		 * in the caller. Nothing more. We could take
1869 		 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1870 		 * but I' d like to return a valid offset (>=0) to allow
1871 		 * the caller to do everything it want with the retval.
1872 		 */
1873 		schedule();
1874 		goto out;
1875 	default:
1876 		/*
1877 		 * Another bit of PARANOID. Note that the retval will be
1878 		 * 0 since no piece of kernel is supposed to do a check
1879 		 * for a negative retval of schedule_timeout() (since it
1880 		 * should never happens anyway). You just have the printk()
1881 		 * that will tell you if something is gone wrong and where.
1882 		 */
1883 		if (timeout < 0) {
1884 			printk(KERN_ERR "schedule_timeout: wrong timeout "
1885 				"value %lx\n", timeout);
1886 			dump_stack();
1887 			current->state = TASK_RUNNING;
1888 			goto out;
1889 		}
1890 	}
1891 
1892 	expire = timeout + jiffies;
1893 
1894 	timer.task = current;
1895 	timer_setup_on_stack(&timer.timer, process_timeout, 0);
1896 	__mod_timer(&timer.timer, expire, MOD_TIMER_NOTPENDING);
1897 	schedule();
1898 	del_singleshot_timer_sync(&timer.timer);
1899 
1900 	/* Remove the timer from the object tracker */
1901 	destroy_timer_on_stack(&timer.timer);
1902 
1903 	timeout = expire - jiffies;
1904 
1905  out:
1906 	return timeout < 0 ? 0 : timeout;
1907 }
1908 EXPORT_SYMBOL(schedule_timeout);
1909 
1910 /*
1911  * We can use __set_current_state() here because schedule_timeout() calls
1912  * schedule() unconditionally.
1913  */
1914 signed long __sched schedule_timeout_interruptible(signed long timeout)
1915 {
1916 	__set_current_state(TASK_INTERRUPTIBLE);
1917 	return schedule_timeout(timeout);
1918 }
1919 EXPORT_SYMBOL(schedule_timeout_interruptible);
1920 
1921 signed long __sched schedule_timeout_killable(signed long timeout)
1922 {
1923 	__set_current_state(TASK_KILLABLE);
1924 	return schedule_timeout(timeout);
1925 }
1926 EXPORT_SYMBOL(schedule_timeout_killable);
1927 
1928 signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1929 {
1930 	__set_current_state(TASK_UNINTERRUPTIBLE);
1931 	return schedule_timeout(timeout);
1932 }
1933 EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1934 
1935 /*
1936  * Like schedule_timeout_uninterruptible(), except this task will not contribute
1937  * to load average.
1938  */
1939 signed long __sched schedule_timeout_idle(signed long timeout)
1940 {
1941 	__set_current_state(TASK_IDLE);
1942 	return schedule_timeout(timeout);
1943 }
1944 EXPORT_SYMBOL(schedule_timeout_idle);
1945 
1946 #ifdef CONFIG_HOTPLUG_CPU
1947 static void migrate_timer_list(struct timer_base *new_base, struct hlist_head *head)
1948 {
1949 	struct timer_list *timer;
1950 	int cpu = new_base->cpu;
1951 
1952 	while (!hlist_empty(head)) {
1953 		timer = hlist_entry(head->first, struct timer_list, entry);
1954 		detach_timer(timer, false);
1955 		timer->flags = (timer->flags & ~TIMER_BASEMASK) | cpu;
1956 		internal_add_timer(new_base, timer);
1957 	}
1958 }
1959 
1960 int timers_prepare_cpu(unsigned int cpu)
1961 {
1962 	struct timer_base *base;
1963 	int b;
1964 
1965 	for (b = 0; b < NR_BASES; b++) {
1966 		base = per_cpu_ptr(&timer_bases[b], cpu);
1967 		base->clk = jiffies;
1968 		base->next_expiry = base->clk + NEXT_TIMER_MAX_DELTA;
1969 		base->is_idle = false;
1970 		base->must_forward_clk = true;
1971 	}
1972 	return 0;
1973 }
1974 
1975 int timers_dead_cpu(unsigned int cpu)
1976 {
1977 	struct timer_base *old_base;
1978 	struct timer_base *new_base;
1979 	int b, i;
1980 
1981 	BUG_ON(cpu_online(cpu));
1982 
1983 	for (b = 0; b < NR_BASES; b++) {
1984 		old_base = per_cpu_ptr(&timer_bases[b], cpu);
1985 		new_base = get_cpu_ptr(&timer_bases[b]);
1986 		/*
1987 		 * The caller is globally serialized and nobody else
1988 		 * takes two locks at once, deadlock is not possible.
1989 		 */
1990 		raw_spin_lock_irq(&new_base->lock);
1991 		raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1992 
1993 		/*
1994 		 * The current CPUs base clock might be stale. Update it
1995 		 * before moving the timers over.
1996 		 */
1997 		forward_timer_base(new_base);
1998 
1999 		BUG_ON(old_base->running_timer);
2000 
2001 		for (i = 0; i < WHEEL_SIZE; i++)
2002 			migrate_timer_list(new_base, old_base->vectors + i);
2003 
2004 		raw_spin_unlock(&old_base->lock);
2005 		raw_spin_unlock_irq(&new_base->lock);
2006 		put_cpu_ptr(&timer_bases);
2007 	}
2008 	return 0;
2009 }
2010 
2011 #endif /* CONFIG_HOTPLUG_CPU */
2012 
2013 static void __init init_timer_cpu(int cpu)
2014 {
2015 	struct timer_base *base;
2016 	int i;
2017 
2018 	for (i = 0; i < NR_BASES; i++) {
2019 		base = per_cpu_ptr(&timer_bases[i], cpu);
2020 		base->cpu = cpu;
2021 		raw_spin_lock_init(&base->lock);
2022 		base->clk = jiffies;
2023 		timer_base_init_expiry_lock(base);
2024 	}
2025 }
2026 
2027 static void __init init_timer_cpus(void)
2028 {
2029 	int cpu;
2030 
2031 	for_each_possible_cpu(cpu)
2032 		init_timer_cpu(cpu);
2033 }
2034 
2035 void __init init_timers(void)
2036 {
2037 	init_timer_cpus();
2038 	open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
2039 }
2040 
2041 /**
2042  * msleep - sleep safely even with waitqueue interruptions
2043  * @msecs: Time in milliseconds to sleep for
2044  */
2045 void msleep(unsigned int msecs)
2046 {
2047 	unsigned long timeout = msecs_to_jiffies(msecs) + 1;
2048 
2049 	while (timeout)
2050 		timeout = schedule_timeout_uninterruptible(timeout);
2051 }
2052 
2053 EXPORT_SYMBOL(msleep);
2054 
2055 /**
2056  * msleep_interruptible - sleep waiting for signals
2057  * @msecs: Time in milliseconds to sleep for
2058  */
2059 unsigned long msleep_interruptible(unsigned int msecs)
2060 {
2061 	unsigned long timeout = msecs_to_jiffies(msecs) + 1;
2062 
2063 	while (timeout && !signal_pending(current))
2064 		timeout = schedule_timeout_interruptible(timeout);
2065 	return jiffies_to_msecs(timeout);
2066 }
2067 
2068 EXPORT_SYMBOL(msleep_interruptible);
2069 
2070 /**
2071  * usleep_range - Sleep for an approximate time
2072  * @min: Minimum time in usecs to sleep
2073  * @max: Maximum time in usecs to sleep
2074  *
2075  * In non-atomic context where the exact wakeup time is flexible, use
2076  * usleep_range() instead of udelay().  The sleep improves responsiveness
2077  * by avoiding the CPU-hogging busy-wait of udelay(), and the range reduces
2078  * power usage by allowing hrtimers to take advantage of an already-
2079  * scheduled interrupt instead of scheduling a new one just for this sleep.
2080  */
2081 void __sched usleep_range(unsigned long min, unsigned long max)
2082 {
2083 	ktime_t exp = ktime_add_us(ktime_get(), min);
2084 	u64 delta = (u64)(max - min) * NSEC_PER_USEC;
2085 
2086 	for (;;) {
2087 		__set_current_state(TASK_UNINTERRUPTIBLE);
2088 		/* Do not return before the requested sleep time has elapsed */
2089 		if (!schedule_hrtimeout_range(&exp, delta, HRTIMER_MODE_ABS))
2090 			break;
2091 	}
2092 }
2093 EXPORT_SYMBOL(usleep_range);
2094