xref: /linux/kernel/time/timer.c (revision 8137a49e1567726eb10fcf55ad141ac19804ca6b)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  Kernel internal timers
4  *
5  *  Copyright (C) 1991, 1992  Linus Torvalds
6  *
7  *  1997-01-28  Modified by Finn Arne Gangstad to make timers scale better.
8  *
9  *  1997-09-10  Updated NTP code according to technical memorandum Jan '96
10  *              "A Kernel Model for Precision Timekeeping" by Dave Mills
11  *  1998-12-24  Fixed a xtime SMP race (we need the xtime_lock rw spinlock to
12  *              serialize accesses to xtime/lost_ticks).
13  *                              Copyright (C) 1998  Andrea Arcangeli
14  *  1999-03-10  Improved NTP compatibility by Ulrich Windl
15  *  2002-05-31	Move sys_sysinfo here and make its locking sane, Robert Love
16  *  2000-10-05  Implemented scalable SMP per-CPU timer handling.
17  *                              Copyright (C) 2000, 2001, 2002  Ingo Molnar
18  *              Designed by David S. Miller, Alexey Kuznetsov and Ingo Molnar
19  */
20 
21 #include <linux/kernel_stat.h>
22 #include <linux/export.h>
23 #include <linux/interrupt.h>
24 #include <linux/percpu.h>
25 #include <linux/init.h>
26 #include <linux/mm.h>
27 #include <linux/swap.h>
28 #include <linux/pid_namespace.h>
29 #include <linux/notifier.h>
30 #include <linux/thread_info.h>
31 #include <linux/time.h>
32 #include <linux/jiffies.h>
33 #include <linux/posix-timers.h>
34 #include <linux/cpu.h>
35 #include <linux/syscalls.h>
36 #include <linux/delay.h>
37 #include <linux/tick.h>
38 #include <linux/kallsyms.h>
39 #include <linux/irq_work.h>
40 #include <linux/sched/signal.h>
41 #include <linux/sched/sysctl.h>
42 #include <linux/sched/nohz.h>
43 #include <linux/sched/debug.h>
44 #include <linux/slab.h>
45 #include <linux/compat.h>
46 #include <linux/random.h>
47 
48 #include <linux/uaccess.h>
49 #include <asm/unistd.h>
50 #include <asm/div64.h>
51 #include <asm/timex.h>
52 #include <asm/io.h>
53 
54 #include "tick-internal.h"
55 
56 #define CREATE_TRACE_POINTS
57 #include <trace/events/timer.h>
58 
59 __visible u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
60 
61 EXPORT_SYMBOL(jiffies_64);
62 
63 /*
64  * The timer wheel has LVL_DEPTH array levels. Each level provides an array of
65  * LVL_SIZE buckets. Each level is driven by its own clock and therefor each
66  * level has a different granularity.
67  *
68  * The level granularity is:		LVL_CLK_DIV ^ lvl
69  * The level clock frequency is:	HZ / (LVL_CLK_DIV ^ level)
70  *
71  * The array level of a newly armed timer depends on the relative expiry
72  * time. The farther the expiry time is away the higher the array level and
73  * therefor the granularity becomes.
74  *
75  * Contrary to the original timer wheel implementation, which aims for 'exact'
76  * expiry of the timers, this implementation removes the need for recascading
77  * the timers into the lower array levels. The previous 'classic' timer wheel
78  * implementation of the kernel already violated the 'exact' expiry by adding
79  * slack to the expiry time to provide batched expiration. The granularity
80  * levels provide implicit batching.
81  *
82  * This is an optimization of the original timer wheel implementation for the
83  * majority of the timer wheel use cases: timeouts. The vast majority of
84  * timeout timers (networking, disk I/O ...) are canceled before expiry. If
85  * the timeout expires it indicates that normal operation is disturbed, so it
86  * does not matter much whether the timeout comes with a slight delay.
87  *
88  * The only exception to this are networking timers with a small expiry
89  * time. They rely on the granularity. Those fit into the first wheel level,
90  * which has HZ granularity.
91  *
92  * We don't have cascading anymore. timers with a expiry time above the
93  * capacity of the last wheel level are force expired at the maximum timeout
94  * value of the last wheel level. From data sampling we know that the maximum
95  * value observed is 5 days (network connection tracking), so this should not
96  * be an issue.
97  *
98  * The currently chosen array constants values are a good compromise between
99  * array size and granularity.
100  *
101  * This results in the following granularity and range levels:
102  *
103  * HZ 1000 steps
104  * Level Offset  Granularity            Range
105  *  0      0         1 ms                0 ms -         63 ms
106  *  1     64         8 ms               64 ms -        511 ms
107  *  2    128        64 ms              512 ms -       4095 ms (512ms - ~4s)
108  *  3    192       512 ms             4096 ms -      32767 ms (~4s - ~32s)
109  *  4    256      4096 ms (~4s)      32768 ms -     262143 ms (~32s - ~4m)
110  *  5    320     32768 ms (~32s)    262144 ms -    2097151 ms (~4m - ~34m)
111  *  6    384    262144 ms (~4m)    2097152 ms -   16777215 ms (~34m - ~4h)
112  *  7    448   2097152 ms (~34m)  16777216 ms -  134217727 ms (~4h - ~1d)
113  *  8    512  16777216 ms (~4h)  134217728 ms - 1073741822 ms (~1d - ~12d)
114  *
115  * HZ  300
116  * Level Offset  Granularity            Range
117  *  0	   0         3 ms                0 ms -        210 ms
118  *  1	  64        26 ms              213 ms -       1703 ms (213ms - ~1s)
119  *  2	 128       213 ms             1706 ms -      13650 ms (~1s - ~13s)
120  *  3	 192      1706 ms (~1s)      13653 ms -     109223 ms (~13s - ~1m)
121  *  4	 256     13653 ms (~13s)    109226 ms -     873810 ms (~1m - ~14m)
122  *  5	 320    109226 ms (~1m)     873813 ms -    6990503 ms (~14m - ~1h)
123  *  6	 384    873813 ms (~14m)   6990506 ms -   55924050 ms (~1h - ~15h)
124  *  7	 448   6990506 ms (~1h)   55924053 ms -  447392423 ms (~15h - ~5d)
125  *  8    512  55924053 ms (~15h) 447392426 ms - 3579139406 ms (~5d - ~41d)
126  *
127  * HZ  250
128  * Level Offset  Granularity            Range
129  *  0	   0         4 ms                0 ms -        255 ms
130  *  1	  64        32 ms              256 ms -       2047 ms (256ms - ~2s)
131  *  2	 128       256 ms             2048 ms -      16383 ms (~2s - ~16s)
132  *  3	 192      2048 ms (~2s)      16384 ms -     131071 ms (~16s - ~2m)
133  *  4	 256     16384 ms (~16s)    131072 ms -    1048575 ms (~2m - ~17m)
134  *  5	 320    131072 ms (~2m)    1048576 ms -    8388607 ms (~17m - ~2h)
135  *  6	 384   1048576 ms (~17m)   8388608 ms -   67108863 ms (~2h - ~18h)
136  *  7	 448   8388608 ms (~2h)   67108864 ms -  536870911 ms (~18h - ~6d)
137  *  8    512  67108864 ms (~18h) 536870912 ms - 4294967288 ms (~6d - ~49d)
138  *
139  * HZ  100
140  * Level Offset  Granularity            Range
141  *  0	   0         10 ms               0 ms -        630 ms
142  *  1	  64         80 ms             640 ms -       5110 ms (640ms - ~5s)
143  *  2	 128        640 ms            5120 ms -      40950 ms (~5s - ~40s)
144  *  3	 192       5120 ms (~5s)     40960 ms -     327670 ms (~40s - ~5m)
145  *  4	 256      40960 ms (~40s)   327680 ms -    2621430 ms (~5m - ~43m)
146  *  5	 320     327680 ms (~5m)   2621440 ms -   20971510 ms (~43m - ~5h)
147  *  6	 384    2621440 ms (~43m) 20971520 ms -  167772150 ms (~5h - ~1d)
148  *  7	 448   20971520 ms (~5h) 167772160 ms - 1342177270 ms (~1d - ~15d)
149  */
150 
151 /* Clock divisor for the next level */
152 #define LVL_CLK_SHIFT	3
153 #define LVL_CLK_DIV	(1UL << LVL_CLK_SHIFT)
154 #define LVL_CLK_MASK	(LVL_CLK_DIV - 1)
155 #define LVL_SHIFT(n)	((n) * LVL_CLK_SHIFT)
156 #define LVL_GRAN(n)	(1UL << LVL_SHIFT(n))
157 
158 /*
159  * The time start value for each level to select the bucket at enqueue
160  * time. We start from the last possible delta of the previous level
161  * so that we can later add an extra LVL_GRAN(n) to n (see calc_index()).
162  */
163 #define LVL_START(n)	((LVL_SIZE - 1) << (((n) - 1) * LVL_CLK_SHIFT))
164 
165 /* Size of each clock level */
166 #define LVL_BITS	6
167 #define LVL_SIZE	(1UL << LVL_BITS)
168 #define LVL_MASK	(LVL_SIZE - 1)
169 #define LVL_OFFS(n)	((n) * LVL_SIZE)
170 
171 /* Level depth */
172 #if HZ > 100
173 # define LVL_DEPTH	9
174 # else
175 # define LVL_DEPTH	8
176 #endif
177 
178 /* The cutoff (max. capacity of the wheel) */
179 #define WHEEL_TIMEOUT_CUTOFF	(LVL_START(LVL_DEPTH))
180 #define WHEEL_TIMEOUT_MAX	(WHEEL_TIMEOUT_CUTOFF - LVL_GRAN(LVL_DEPTH - 1))
181 
182 /*
183  * The resulting wheel size. If NOHZ is configured we allocate two
184  * wheels so we have a separate storage for the deferrable timers.
185  */
186 #define WHEEL_SIZE	(LVL_SIZE * LVL_DEPTH)
187 
188 #ifdef CONFIG_NO_HZ_COMMON
189 # define NR_BASES	2
190 # define BASE_STD	0
191 # define BASE_DEF	1
192 #else
193 # define NR_BASES	1
194 # define BASE_STD	0
195 # define BASE_DEF	0
196 #endif
197 
198 struct timer_base {
199 	raw_spinlock_t		lock;
200 	struct timer_list	*running_timer;
201 #ifdef CONFIG_PREEMPT_RT
202 	spinlock_t		expiry_lock;
203 	atomic_t		timer_waiters;
204 #endif
205 	unsigned long		clk;
206 	unsigned long		next_expiry;
207 	unsigned int		cpu;
208 	bool			next_expiry_recalc;
209 	bool			is_idle;
210 	DECLARE_BITMAP(pending_map, WHEEL_SIZE);
211 	struct hlist_head	vectors[WHEEL_SIZE];
212 } ____cacheline_aligned;
213 
214 static DEFINE_PER_CPU(struct timer_base, timer_bases[NR_BASES]);
215 
216 #ifdef CONFIG_NO_HZ_COMMON
217 
218 static DEFINE_STATIC_KEY_FALSE(timers_nohz_active);
219 static DEFINE_MUTEX(timer_keys_mutex);
220 
221 static void timer_update_keys(struct work_struct *work);
222 static DECLARE_WORK(timer_update_work, timer_update_keys);
223 
224 #ifdef CONFIG_SMP
225 unsigned int sysctl_timer_migration = 1;
226 
227 DEFINE_STATIC_KEY_FALSE(timers_migration_enabled);
228 
229 static void timers_update_migration(void)
230 {
231 	if (sysctl_timer_migration && tick_nohz_active)
232 		static_branch_enable(&timers_migration_enabled);
233 	else
234 		static_branch_disable(&timers_migration_enabled);
235 }
236 #else
237 static inline void timers_update_migration(void) { }
238 #endif /* !CONFIG_SMP */
239 
240 static void timer_update_keys(struct work_struct *work)
241 {
242 	mutex_lock(&timer_keys_mutex);
243 	timers_update_migration();
244 	static_branch_enable(&timers_nohz_active);
245 	mutex_unlock(&timer_keys_mutex);
246 }
247 
248 void timers_update_nohz(void)
249 {
250 	schedule_work(&timer_update_work);
251 }
252 
253 int timer_migration_handler(struct ctl_table *table, int write,
254 			    void *buffer, size_t *lenp, loff_t *ppos)
255 {
256 	int ret;
257 
258 	mutex_lock(&timer_keys_mutex);
259 	ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
260 	if (!ret && write)
261 		timers_update_migration();
262 	mutex_unlock(&timer_keys_mutex);
263 	return ret;
264 }
265 
266 static inline bool is_timers_nohz_active(void)
267 {
268 	return static_branch_unlikely(&timers_nohz_active);
269 }
270 #else
271 static inline bool is_timers_nohz_active(void) { return false; }
272 #endif /* NO_HZ_COMMON */
273 
274 static unsigned long round_jiffies_common(unsigned long j, int cpu,
275 		bool force_up)
276 {
277 	int rem;
278 	unsigned long original = j;
279 
280 	/*
281 	 * We don't want all cpus firing their timers at once hitting the
282 	 * same lock or cachelines, so we skew each extra cpu with an extra
283 	 * 3 jiffies. This 3 jiffies came originally from the mm/ code which
284 	 * already did this.
285 	 * The skew is done by adding 3*cpunr, then round, then subtract this
286 	 * extra offset again.
287 	 */
288 	j += cpu * 3;
289 
290 	rem = j % HZ;
291 
292 	/*
293 	 * If the target jiffie is just after a whole second (which can happen
294 	 * due to delays of the timer irq, long irq off times etc etc) then
295 	 * we should round down to the whole second, not up. Use 1/4th second
296 	 * as cutoff for this rounding as an extreme upper bound for this.
297 	 * But never round down if @force_up is set.
298 	 */
299 	if (rem < HZ/4 && !force_up) /* round down */
300 		j = j - rem;
301 	else /* round up */
302 		j = j - rem + HZ;
303 
304 	/* now that we have rounded, subtract the extra skew again */
305 	j -= cpu * 3;
306 
307 	/*
308 	 * Make sure j is still in the future. Otherwise return the
309 	 * unmodified value.
310 	 */
311 	return time_is_after_jiffies(j) ? j : original;
312 }
313 
314 /**
315  * __round_jiffies - function to round jiffies to a full second
316  * @j: the time in (absolute) jiffies that should be rounded
317  * @cpu: the processor number on which the timeout will happen
318  *
319  * __round_jiffies() rounds an absolute time in the future (in jiffies)
320  * up or down to (approximately) full seconds. This is useful for timers
321  * for which the exact time they fire does not matter too much, as long as
322  * they fire approximately every X seconds.
323  *
324  * By rounding these timers to whole seconds, all such timers will fire
325  * at the same time, rather than at various times spread out. The goal
326  * of this is to have the CPU wake up less, which saves power.
327  *
328  * The exact rounding is skewed for each processor to avoid all
329  * processors firing at the exact same time, which could lead
330  * to lock contention or spurious cache line bouncing.
331  *
332  * The return value is the rounded version of the @j parameter.
333  */
334 unsigned long __round_jiffies(unsigned long j, int cpu)
335 {
336 	return round_jiffies_common(j, cpu, false);
337 }
338 EXPORT_SYMBOL_GPL(__round_jiffies);
339 
340 /**
341  * __round_jiffies_relative - function to round jiffies to a full second
342  * @j: the time in (relative) jiffies that should be rounded
343  * @cpu: the processor number on which the timeout will happen
344  *
345  * __round_jiffies_relative() rounds a time delta  in the future (in jiffies)
346  * up or down to (approximately) full seconds. This is useful for timers
347  * for which the exact time they fire does not matter too much, as long as
348  * they fire approximately every X seconds.
349  *
350  * By rounding these timers to whole seconds, all such timers will fire
351  * at the same time, rather than at various times spread out. The goal
352  * of this is to have the CPU wake up less, which saves power.
353  *
354  * The exact rounding is skewed for each processor to avoid all
355  * processors firing at the exact same time, which could lead
356  * to lock contention or spurious cache line bouncing.
357  *
358  * The return value is the rounded version of the @j parameter.
359  */
360 unsigned long __round_jiffies_relative(unsigned long j, int cpu)
361 {
362 	unsigned long j0 = jiffies;
363 
364 	/* Use j0 because jiffies might change while we run */
365 	return round_jiffies_common(j + j0, cpu, false) - j0;
366 }
367 EXPORT_SYMBOL_GPL(__round_jiffies_relative);
368 
369 /**
370  * round_jiffies - function to round jiffies to a full second
371  * @j: the time in (absolute) jiffies that should be rounded
372  *
373  * round_jiffies() rounds an absolute time in the future (in jiffies)
374  * up or down to (approximately) full seconds. This is useful for timers
375  * for which the exact time they fire does not matter too much, as long as
376  * they fire approximately every X seconds.
377  *
378  * By rounding these timers to whole seconds, all such timers will fire
379  * at the same time, rather than at various times spread out. The goal
380  * of this is to have the CPU wake up less, which saves power.
381  *
382  * The return value is the rounded version of the @j parameter.
383  */
384 unsigned long round_jiffies(unsigned long j)
385 {
386 	return round_jiffies_common(j, raw_smp_processor_id(), false);
387 }
388 EXPORT_SYMBOL_GPL(round_jiffies);
389 
390 /**
391  * round_jiffies_relative - function to round jiffies to a full second
392  * @j: the time in (relative) jiffies that should be rounded
393  *
394  * round_jiffies_relative() rounds a time delta  in the future (in jiffies)
395  * up or down to (approximately) full seconds. This is useful for timers
396  * for which the exact time they fire does not matter too much, as long as
397  * they fire approximately every X seconds.
398  *
399  * By rounding these timers to whole seconds, all such timers will fire
400  * at the same time, rather than at various times spread out. The goal
401  * of this is to have the CPU wake up less, which saves power.
402  *
403  * The return value is the rounded version of the @j parameter.
404  */
405 unsigned long round_jiffies_relative(unsigned long j)
406 {
407 	return __round_jiffies_relative(j, raw_smp_processor_id());
408 }
409 EXPORT_SYMBOL_GPL(round_jiffies_relative);
410 
411 /**
412  * __round_jiffies_up - function to round jiffies up to a full second
413  * @j: the time in (absolute) jiffies that should be rounded
414  * @cpu: the processor number on which the timeout will happen
415  *
416  * This is the same as __round_jiffies() except that it will never
417  * round down.  This is useful for timeouts for which the exact time
418  * of firing does not matter too much, as long as they don't fire too
419  * early.
420  */
421 unsigned long __round_jiffies_up(unsigned long j, int cpu)
422 {
423 	return round_jiffies_common(j, cpu, true);
424 }
425 EXPORT_SYMBOL_GPL(__round_jiffies_up);
426 
427 /**
428  * __round_jiffies_up_relative - function to round jiffies up to a full second
429  * @j: the time in (relative) jiffies that should be rounded
430  * @cpu: the processor number on which the timeout will happen
431  *
432  * This is the same as __round_jiffies_relative() except that it will never
433  * round down.  This is useful for timeouts for which the exact time
434  * of firing does not matter too much, as long as they don't fire too
435  * early.
436  */
437 unsigned long __round_jiffies_up_relative(unsigned long j, int cpu)
438 {
439 	unsigned long j0 = jiffies;
440 
441 	/* Use j0 because jiffies might change while we run */
442 	return round_jiffies_common(j + j0, cpu, true) - j0;
443 }
444 EXPORT_SYMBOL_GPL(__round_jiffies_up_relative);
445 
446 /**
447  * round_jiffies_up - function to round jiffies up to a full second
448  * @j: the time in (absolute) jiffies that should be rounded
449  *
450  * This is the same as round_jiffies() except that it will never
451  * round down.  This is useful for timeouts for which the exact time
452  * of firing does not matter too much, as long as they don't fire too
453  * early.
454  */
455 unsigned long round_jiffies_up(unsigned long j)
456 {
457 	return round_jiffies_common(j, raw_smp_processor_id(), true);
458 }
459 EXPORT_SYMBOL_GPL(round_jiffies_up);
460 
461 /**
462  * round_jiffies_up_relative - function to round jiffies up to a full second
463  * @j: the time in (relative) jiffies that should be rounded
464  *
465  * This is the same as round_jiffies_relative() except that it will never
466  * round down.  This is useful for timeouts for which the exact time
467  * of firing does not matter too much, as long as they don't fire too
468  * early.
469  */
470 unsigned long round_jiffies_up_relative(unsigned long j)
471 {
472 	return __round_jiffies_up_relative(j, raw_smp_processor_id());
473 }
474 EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
475 
476 
477 static inline unsigned int timer_get_idx(struct timer_list *timer)
478 {
479 	return (timer->flags & TIMER_ARRAYMASK) >> TIMER_ARRAYSHIFT;
480 }
481 
482 static inline void timer_set_idx(struct timer_list *timer, unsigned int idx)
483 {
484 	timer->flags = (timer->flags & ~TIMER_ARRAYMASK) |
485 			idx << TIMER_ARRAYSHIFT;
486 }
487 
488 /*
489  * Helper function to calculate the array index for a given expiry
490  * time.
491  */
492 static inline unsigned calc_index(unsigned long expires, unsigned lvl,
493 				  unsigned long *bucket_expiry)
494 {
495 
496 	/*
497 	 * The timer wheel has to guarantee that a timer does not fire
498 	 * early. Early expiry can happen due to:
499 	 * - Timer is armed at the edge of a tick
500 	 * - Truncation of the expiry time in the outer wheel levels
501 	 *
502 	 * Round up with level granularity to prevent this.
503 	 */
504 	expires = (expires + LVL_GRAN(lvl)) >> LVL_SHIFT(lvl);
505 	*bucket_expiry = expires << LVL_SHIFT(lvl);
506 	return LVL_OFFS(lvl) + (expires & LVL_MASK);
507 }
508 
509 static int calc_wheel_index(unsigned long expires, unsigned long clk,
510 			    unsigned long *bucket_expiry)
511 {
512 	unsigned long delta = expires - clk;
513 	unsigned int idx;
514 
515 	if (delta < LVL_START(1)) {
516 		idx = calc_index(expires, 0, bucket_expiry);
517 	} else if (delta < LVL_START(2)) {
518 		idx = calc_index(expires, 1, bucket_expiry);
519 	} else if (delta < LVL_START(3)) {
520 		idx = calc_index(expires, 2, bucket_expiry);
521 	} else if (delta < LVL_START(4)) {
522 		idx = calc_index(expires, 3, bucket_expiry);
523 	} else if (delta < LVL_START(5)) {
524 		idx = calc_index(expires, 4, bucket_expiry);
525 	} else if (delta < LVL_START(6)) {
526 		idx = calc_index(expires, 5, bucket_expiry);
527 	} else if (delta < LVL_START(7)) {
528 		idx = calc_index(expires, 6, bucket_expiry);
529 	} else if (LVL_DEPTH > 8 && delta < LVL_START(8)) {
530 		idx = calc_index(expires, 7, bucket_expiry);
531 	} else if ((long) delta < 0) {
532 		idx = clk & LVL_MASK;
533 		*bucket_expiry = clk;
534 	} else {
535 		/*
536 		 * Force expire obscene large timeouts to expire at the
537 		 * capacity limit of the wheel.
538 		 */
539 		if (delta >= WHEEL_TIMEOUT_CUTOFF)
540 			expires = clk + WHEEL_TIMEOUT_MAX;
541 
542 		idx = calc_index(expires, LVL_DEPTH - 1, bucket_expiry);
543 	}
544 	return idx;
545 }
546 
547 static void
548 trigger_dyntick_cpu(struct timer_base *base, struct timer_list *timer)
549 {
550 	if (!is_timers_nohz_active())
551 		return;
552 
553 	/*
554 	 * TODO: This wants some optimizing similar to the code below, but we
555 	 * will do that when we switch from push to pull for deferrable timers.
556 	 */
557 	if (timer->flags & TIMER_DEFERRABLE) {
558 		if (tick_nohz_full_cpu(base->cpu))
559 			wake_up_nohz_cpu(base->cpu);
560 		return;
561 	}
562 
563 	/*
564 	 * We might have to IPI the remote CPU if the base is idle and the
565 	 * timer is not deferrable. If the other CPU is on the way to idle
566 	 * then it can't set base->is_idle as we hold the base lock:
567 	 */
568 	if (base->is_idle)
569 		wake_up_nohz_cpu(base->cpu);
570 }
571 
572 /*
573  * Enqueue the timer into the hash bucket, mark it pending in
574  * the bitmap, store the index in the timer flags then wake up
575  * the target CPU if needed.
576  */
577 static void enqueue_timer(struct timer_base *base, struct timer_list *timer,
578 			  unsigned int idx, unsigned long bucket_expiry)
579 {
580 
581 	hlist_add_head(&timer->entry, base->vectors + idx);
582 	__set_bit(idx, base->pending_map);
583 	timer_set_idx(timer, idx);
584 
585 	trace_timer_start(timer, timer->expires, timer->flags);
586 
587 	/*
588 	 * Check whether this is the new first expiring timer. The
589 	 * effective expiry time of the timer is required here
590 	 * (bucket_expiry) instead of timer->expires.
591 	 */
592 	if (time_before(bucket_expiry, base->next_expiry)) {
593 		/*
594 		 * Set the next expiry time and kick the CPU so it
595 		 * can reevaluate the wheel:
596 		 */
597 		base->next_expiry = bucket_expiry;
598 		base->next_expiry_recalc = false;
599 		trigger_dyntick_cpu(base, timer);
600 	}
601 }
602 
603 static void internal_add_timer(struct timer_base *base, struct timer_list *timer)
604 {
605 	unsigned long bucket_expiry;
606 	unsigned int idx;
607 
608 	idx = calc_wheel_index(timer->expires, base->clk, &bucket_expiry);
609 	enqueue_timer(base, timer, idx, bucket_expiry);
610 }
611 
612 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
613 
614 static const struct debug_obj_descr timer_debug_descr;
615 
616 static void *timer_debug_hint(void *addr)
617 {
618 	return ((struct timer_list *) addr)->function;
619 }
620 
621 static bool timer_is_static_object(void *addr)
622 {
623 	struct timer_list *timer = addr;
624 
625 	return (timer->entry.pprev == NULL &&
626 		timer->entry.next == TIMER_ENTRY_STATIC);
627 }
628 
629 /*
630  * fixup_init is called when:
631  * - an active object is initialized
632  */
633 static bool timer_fixup_init(void *addr, enum debug_obj_state state)
634 {
635 	struct timer_list *timer = addr;
636 
637 	switch (state) {
638 	case ODEBUG_STATE_ACTIVE:
639 		del_timer_sync(timer);
640 		debug_object_init(timer, &timer_debug_descr);
641 		return true;
642 	default:
643 		return false;
644 	}
645 }
646 
647 /* Stub timer callback for improperly used timers. */
648 static void stub_timer(struct timer_list *unused)
649 {
650 	WARN_ON(1);
651 }
652 
653 /*
654  * fixup_activate is called when:
655  * - an active object is activated
656  * - an unknown non-static object is activated
657  */
658 static bool timer_fixup_activate(void *addr, enum debug_obj_state state)
659 {
660 	struct timer_list *timer = addr;
661 
662 	switch (state) {
663 	case ODEBUG_STATE_NOTAVAILABLE:
664 		timer_setup(timer, stub_timer, 0);
665 		return true;
666 
667 	case ODEBUG_STATE_ACTIVE:
668 		WARN_ON(1);
669 		fallthrough;
670 	default:
671 		return false;
672 	}
673 }
674 
675 /*
676  * fixup_free is called when:
677  * - an active object is freed
678  */
679 static bool timer_fixup_free(void *addr, enum debug_obj_state state)
680 {
681 	struct timer_list *timer = addr;
682 
683 	switch (state) {
684 	case ODEBUG_STATE_ACTIVE:
685 		del_timer_sync(timer);
686 		debug_object_free(timer, &timer_debug_descr);
687 		return true;
688 	default:
689 		return false;
690 	}
691 }
692 
693 /*
694  * fixup_assert_init is called when:
695  * - an untracked/uninit-ed object is found
696  */
697 static bool timer_fixup_assert_init(void *addr, enum debug_obj_state state)
698 {
699 	struct timer_list *timer = addr;
700 
701 	switch (state) {
702 	case ODEBUG_STATE_NOTAVAILABLE:
703 		timer_setup(timer, stub_timer, 0);
704 		return true;
705 	default:
706 		return false;
707 	}
708 }
709 
710 static const struct debug_obj_descr timer_debug_descr = {
711 	.name			= "timer_list",
712 	.debug_hint		= timer_debug_hint,
713 	.is_static_object	= timer_is_static_object,
714 	.fixup_init		= timer_fixup_init,
715 	.fixup_activate		= timer_fixup_activate,
716 	.fixup_free		= timer_fixup_free,
717 	.fixup_assert_init	= timer_fixup_assert_init,
718 };
719 
720 static inline void debug_timer_init(struct timer_list *timer)
721 {
722 	debug_object_init(timer, &timer_debug_descr);
723 }
724 
725 static inline void debug_timer_activate(struct timer_list *timer)
726 {
727 	debug_object_activate(timer, &timer_debug_descr);
728 }
729 
730 static inline void debug_timer_deactivate(struct timer_list *timer)
731 {
732 	debug_object_deactivate(timer, &timer_debug_descr);
733 }
734 
735 static inline void debug_timer_assert_init(struct timer_list *timer)
736 {
737 	debug_object_assert_init(timer, &timer_debug_descr);
738 }
739 
740 static void do_init_timer(struct timer_list *timer,
741 			  void (*func)(struct timer_list *),
742 			  unsigned int flags,
743 			  const char *name, struct lock_class_key *key);
744 
745 void init_timer_on_stack_key(struct timer_list *timer,
746 			     void (*func)(struct timer_list *),
747 			     unsigned int flags,
748 			     const char *name, struct lock_class_key *key)
749 {
750 	debug_object_init_on_stack(timer, &timer_debug_descr);
751 	do_init_timer(timer, func, flags, name, key);
752 }
753 EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
754 
755 void destroy_timer_on_stack(struct timer_list *timer)
756 {
757 	debug_object_free(timer, &timer_debug_descr);
758 }
759 EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
760 
761 #else
762 static inline void debug_timer_init(struct timer_list *timer) { }
763 static inline void debug_timer_activate(struct timer_list *timer) { }
764 static inline void debug_timer_deactivate(struct timer_list *timer) { }
765 static inline void debug_timer_assert_init(struct timer_list *timer) { }
766 #endif
767 
768 static inline void debug_init(struct timer_list *timer)
769 {
770 	debug_timer_init(timer);
771 	trace_timer_init(timer);
772 }
773 
774 static inline void debug_deactivate(struct timer_list *timer)
775 {
776 	debug_timer_deactivate(timer);
777 	trace_timer_cancel(timer);
778 }
779 
780 static inline void debug_assert_init(struct timer_list *timer)
781 {
782 	debug_timer_assert_init(timer);
783 }
784 
785 static void do_init_timer(struct timer_list *timer,
786 			  void (*func)(struct timer_list *),
787 			  unsigned int flags,
788 			  const char *name, struct lock_class_key *key)
789 {
790 	timer->entry.pprev = NULL;
791 	timer->function = func;
792 	if (WARN_ON_ONCE(flags & ~TIMER_INIT_FLAGS))
793 		flags &= TIMER_INIT_FLAGS;
794 	timer->flags = flags | raw_smp_processor_id();
795 	lockdep_init_map(&timer->lockdep_map, name, key, 0);
796 }
797 
798 /**
799  * init_timer_key - initialize a timer
800  * @timer: the timer to be initialized
801  * @func: timer callback function
802  * @flags: timer flags
803  * @name: name of the timer
804  * @key: lockdep class key of the fake lock used for tracking timer
805  *       sync lock dependencies
806  *
807  * init_timer_key() must be done to a timer prior calling *any* of the
808  * other timer functions.
809  */
810 void init_timer_key(struct timer_list *timer,
811 		    void (*func)(struct timer_list *), unsigned int flags,
812 		    const char *name, struct lock_class_key *key)
813 {
814 	debug_init(timer);
815 	do_init_timer(timer, func, flags, name, key);
816 }
817 EXPORT_SYMBOL(init_timer_key);
818 
819 static inline void detach_timer(struct timer_list *timer, bool clear_pending)
820 {
821 	struct hlist_node *entry = &timer->entry;
822 
823 	debug_deactivate(timer);
824 
825 	__hlist_del(entry);
826 	if (clear_pending)
827 		entry->pprev = NULL;
828 	entry->next = LIST_POISON2;
829 }
830 
831 static int detach_if_pending(struct timer_list *timer, struct timer_base *base,
832 			     bool clear_pending)
833 {
834 	unsigned idx = timer_get_idx(timer);
835 
836 	if (!timer_pending(timer))
837 		return 0;
838 
839 	if (hlist_is_singular_node(&timer->entry, base->vectors + idx)) {
840 		__clear_bit(idx, base->pending_map);
841 		base->next_expiry_recalc = true;
842 	}
843 
844 	detach_timer(timer, clear_pending);
845 	return 1;
846 }
847 
848 static inline struct timer_base *get_timer_cpu_base(u32 tflags, u32 cpu)
849 {
850 	struct timer_base *base = per_cpu_ptr(&timer_bases[BASE_STD], cpu);
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 = per_cpu_ptr(&timer_bases[BASE_DEF], cpu);
858 	return base;
859 }
860 
861 static inline struct timer_base *get_timer_this_cpu_base(u32 tflags)
862 {
863 	struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
864 
865 	/*
866 	 * If the timer is deferrable and NO_HZ_COMMON is set then we need
867 	 * to use the deferrable base.
868 	 */
869 	if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE))
870 		base = this_cpu_ptr(&timer_bases[BASE_DEF]);
871 	return base;
872 }
873 
874 static inline struct timer_base *get_timer_base(u32 tflags)
875 {
876 	return get_timer_cpu_base(tflags, tflags & TIMER_CPUMASK);
877 }
878 
879 static inline struct timer_base *
880 get_target_base(struct timer_base *base, unsigned tflags)
881 {
882 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
883 	if (static_branch_likely(&timers_migration_enabled) &&
884 	    !(tflags & TIMER_PINNED))
885 		return get_timer_cpu_base(tflags, get_nohz_timer_target());
886 #endif
887 	return get_timer_this_cpu_base(tflags);
888 }
889 
890 static inline void forward_timer_base(struct timer_base *base)
891 {
892 	unsigned long jnow = READ_ONCE(jiffies);
893 
894 	/*
895 	 * No need to forward if we are close enough below jiffies.
896 	 * Also while executing timers, base->clk is 1 offset ahead
897 	 * of jiffies to avoid endless requeuing to current jiffies.
898 	 */
899 	if ((long)(jnow - base->clk) < 1)
900 		return;
901 
902 	/*
903 	 * If the next expiry value is > jiffies, then we fast forward to
904 	 * jiffies otherwise we forward to the next expiry value.
905 	 */
906 	if (time_after(base->next_expiry, jnow)) {
907 		base->clk = jnow;
908 	} else {
909 		if (WARN_ON_ONCE(time_before(base->next_expiry, base->clk)))
910 			return;
911 		base->clk = base->next_expiry;
912 	}
913 }
914 
915 
916 /*
917  * We are using hashed locking: Holding per_cpu(timer_bases[x]).lock means
918  * that all timers which are tied to this base are locked, and the base itself
919  * is locked too.
920  *
921  * So __run_timers/migrate_timers can safely modify all timers which could
922  * be found in the base->vectors array.
923  *
924  * When a timer is migrating then the TIMER_MIGRATING flag is set and we need
925  * to wait until the migration is done.
926  */
927 static struct timer_base *lock_timer_base(struct timer_list *timer,
928 					  unsigned long *flags)
929 	__acquires(timer->base->lock)
930 {
931 	for (;;) {
932 		struct timer_base *base;
933 		u32 tf;
934 
935 		/*
936 		 * We need to use READ_ONCE() here, otherwise the compiler
937 		 * might re-read @tf between the check for TIMER_MIGRATING
938 		 * and spin_lock().
939 		 */
940 		tf = READ_ONCE(timer->flags);
941 
942 		if (!(tf & TIMER_MIGRATING)) {
943 			base = get_timer_base(tf);
944 			raw_spin_lock_irqsave(&base->lock, *flags);
945 			if (timer->flags == tf)
946 				return base;
947 			raw_spin_unlock_irqrestore(&base->lock, *flags);
948 		}
949 		cpu_relax();
950 	}
951 }
952 
953 #define MOD_TIMER_PENDING_ONLY		0x01
954 #define MOD_TIMER_REDUCE		0x02
955 #define MOD_TIMER_NOTPENDING		0x04
956 
957 static inline int
958 __mod_timer(struct timer_list *timer, unsigned long expires, unsigned int options)
959 {
960 	unsigned long clk = 0, flags, bucket_expiry;
961 	struct timer_base *base, *new_base;
962 	unsigned int idx = UINT_MAX;
963 	int ret = 0;
964 
965 	BUG_ON(!timer->function);
966 
967 	/*
968 	 * This is a common optimization triggered by the networking code - if
969 	 * the timer is re-modified to have the same timeout or ends up in the
970 	 * same array bucket then just return:
971 	 */
972 	if (!(options & MOD_TIMER_NOTPENDING) && timer_pending(timer)) {
973 		/*
974 		 * The downside of this optimization is that it can result in
975 		 * larger granularity than you would get from adding a new
976 		 * timer with this expiry.
977 		 */
978 		long diff = timer->expires - expires;
979 
980 		if (!diff)
981 			return 1;
982 		if (options & MOD_TIMER_REDUCE && diff <= 0)
983 			return 1;
984 
985 		/*
986 		 * We lock timer base and calculate the bucket index right
987 		 * here. If the timer ends up in the same bucket, then we
988 		 * just update the expiry time and avoid the whole
989 		 * dequeue/enqueue dance.
990 		 */
991 		base = lock_timer_base(timer, &flags);
992 		forward_timer_base(base);
993 
994 		if (timer_pending(timer) && (options & MOD_TIMER_REDUCE) &&
995 		    time_before_eq(timer->expires, expires)) {
996 			ret = 1;
997 			goto out_unlock;
998 		}
999 
1000 		clk = base->clk;
1001 		idx = calc_wheel_index(expires, clk, &bucket_expiry);
1002 
1003 		/*
1004 		 * Retrieve and compare the array index of the pending
1005 		 * timer. If it matches set the expiry to the new value so a
1006 		 * subsequent call will exit in the expires check above.
1007 		 */
1008 		if (idx == timer_get_idx(timer)) {
1009 			if (!(options & MOD_TIMER_REDUCE))
1010 				timer->expires = expires;
1011 			else if (time_after(timer->expires, expires))
1012 				timer->expires = expires;
1013 			ret = 1;
1014 			goto out_unlock;
1015 		}
1016 	} else {
1017 		base = lock_timer_base(timer, &flags);
1018 		forward_timer_base(base);
1019 	}
1020 
1021 	ret = detach_if_pending(timer, base, false);
1022 	if (!ret && (options & MOD_TIMER_PENDING_ONLY))
1023 		goto out_unlock;
1024 
1025 	new_base = get_target_base(base, timer->flags);
1026 
1027 	if (base != new_base) {
1028 		/*
1029 		 * We are trying to schedule the timer on the new base.
1030 		 * However we can't change timer's base while it is running,
1031 		 * otherwise del_timer_sync() can't detect that the timer's
1032 		 * handler yet has not finished. This also guarantees that the
1033 		 * timer is serialized wrt itself.
1034 		 */
1035 		if (likely(base->running_timer != timer)) {
1036 			/* See the comment in lock_timer_base() */
1037 			timer->flags |= TIMER_MIGRATING;
1038 
1039 			raw_spin_unlock(&base->lock);
1040 			base = new_base;
1041 			raw_spin_lock(&base->lock);
1042 			WRITE_ONCE(timer->flags,
1043 				   (timer->flags & ~TIMER_BASEMASK) | base->cpu);
1044 			forward_timer_base(base);
1045 		}
1046 	}
1047 
1048 	debug_timer_activate(timer);
1049 
1050 	timer->expires = expires;
1051 	/*
1052 	 * If 'idx' was calculated above and the base time did not advance
1053 	 * between calculating 'idx' and possibly switching the base, only
1054 	 * enqueue_timer() is required. Otherwise we need to (re)calculate
1055 	 * the wheel index via internal_add_timer().
1056 	 */
1057 	if (idx != UINT_MAX && clk == base->clk)
1058 		enqueue_timer(base, timer, idx, bucket_expiry);
1059 	else
1060 		internal_add_timer(base, timer);
1061 
1062 out_unlock:
1063 	raw_spin_unlock_irqrestore(&base->lock, flags);
1064 
1065 	return ret;
1066 }
1067 
1068 /**
1069  * mod_timer_pending - modify a pending timer's timeout
1070  * @timer: the pending timer to be modified
1071  * @expires: new timeout in jiffies
1072  *
1073  * mod_timer_pending() is the same for pending timers as mod_timer(),
1074  * but will not re-activate and modify already deleted timers.
1075  *
1076  * It is useful for unserialized use of timers.
1077  */
1078 int mod_timer_pending(struct timer_list *timer, unsigned long expires)
1079 {
1080 	return __mod_timer(timer, expires, MOD_TIMER_PENDING_ONLY);
1081 }
1082 EXPORT_SYMBOL(mod_timer_pending);
1083 
1084 /**
1085  * mod_timer - modify a timer's timeout
1086  * @timer: the timer to be modified
1087  * @expires: new timeout in jiffies
1088  *
1089  * mod_timer() is a more efficient way to update the expire field of an
1090  * active timer (if the timer is inactive it will be activated)
1091  *
1092  * mod_timer(timer, expires) is equivalent to:
1093  *
1094  *     del_timer(timer); timer->expires = expires; add_timer(timer);
1095  *
1096  * Note that if there are multiple unserialized concurrent users of the
1097  * same timer, then mod_timer() is the only safe way to modify the timeout,
1098  * since add_timer() cannot modify an already running timer.
1099  *
1100  * The function returns whether it has modified a pending timer or not.
1101  * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
1102  * active timer returns 1.)
1103  */
1104 int mod_timer(struct timer_list *timer, unsigned long expires)
1105 {
1106 	return __mod_timer(timer, expires, 0);
1107 }
1108 EXPORT_SYMBOL(mod_timer);
1109 
1110 /**
1111  * timer_reduce - Modify a timer's timeout if it would reduce the timeout
1112  * @timer:	The timer to be modified
1113  * @expires:	New timeout in jiffies
1114  *
1115  * timer_reduce() is very similar to mod_timer(), except that it will only
1116  * modify a running timer if that would reduce the expiration time (it will
1117  * start a timer that isn't running).
1118  */
1119 int timer_reduce(struct timer_list *timer, unsigned long expires)
1120 {
1121 	return __mod_timer(timer, expires, MOD_TIMER_REDUCE);
1122 }
1123 EXPORT_SYMBOL(timer_reduce);
1124 
1125 /**
1126  * add_timer - start a timer
1127  * @timer: the timer to be added
1128  *
1129  * The kernel will do a ->function(@timer) callback from the
1130  * timer interrupt at the ->expires point in the future. The
1131  * current time is 'jiffies'.
1132  *
1133  * The timer's ->expires, ->function fields must be set prior calling this
1134  * function.
1135  *
1136  * Timers with an ->expires field in the past will be executed in the next
1137  * timer tick.
1138  */
1139 void add_timer(struct timer_list *timer)
1140 {
1141 	BUG_ON(timer_pending(timer));
1142 	__mod_timer(timer, timer->expires, MOD_TIMER_NOTPENDING);
1143 }
1144 EXPORT_SYMBOL(add_timer);
1145 
1146 /**
1147  * add_timer_on - start a timer on a particular CPU
1148  * @timer: the timer to be added
1149  * @cpu: the CPU to start it on
1150  *
1151  * This is not very scalable on SMP. Double adds are not possible.
1152  */
1153 void add_timer_on(struct timer_list *timer, int cpu)
1154 {
1155 	struct timer_base *new_base, *base;
1156 	unsigned long flags;
1157 
1158 	BUG_ON(timer_pending(timer) || !timer->function);
1159 
1160 	new_base = get_timer_cpu_base(timer->flags, cpu);
1161 
1162 	/*
1163 	 * If @timer was on a different CPU, it should be migrated with the
1164 	 * old base locked to prevent other operations proceeding with the
1165 	 * wrong base locked.  See lock_timer_base().
1166 	 */
1167 	base = lock_timer_base(timer, &flags);
1168 	if (base != new_base) {
1169 		timer->flags |= TIMER_MIGRATING;
1170 
1171 		raw_spin_unlock(&base->lock);
1172 		base = new_base;
1173 		raw_spin_lock(&base->lock);
1174 		WRITE_ONCE(timer->flags,
1175 			   (timer->flags & ~TIMER_BASEMASK) | cpu);
1176 	}
1177 	forward_timer_base(base);
1178 
1179 	debug_timer_activate(timer);
1180 	internal_add_timer(base, timer);
1181 	raw_spin_unlock_irqrestore(&base->lock, flags);
1182 }
1183 EXPORT_SYMBOL_GPL(add_timer_on);
1184 
1185 /**
1186  * del_timer - deactivate a timer.
1187  * @timer: the timer to be deactivated
1188  *
1189  * del_timer() deactivates a timer - this works on both active and inactive
1190  * timers.
1191  *
1192  * The function returns whether it has deactivated a pending timer or not.
1193  * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
1194  * active timer returns 1.)
1195  */
1196 int del_timer(struct timer_list *timer)
1197 {
1198 	struct timer_base *base;
1199 	unsigned long flags;
1200 	int ret = 0;
1201 
1202 	debug_assert_init(timer);
1203 
1204 	if (timer_pending(timer)) {
1205 		base = lock_timer_base(timer, &flags);
1206 		ret = detach_if_pending(timer, base, true);
1207 		raw_spin_unlock_irqrestore(&base->lock, flags);
1208 	}
1209 
1210 	return ret;
1211 }
1212 EXPORT_SYMBOL(del_timer);
1213 
1214 /**
1215  * try_to_del_timer_sync - Try to deactivate a timer
1216  * @timer: timer to delete
1217  *
1218  * This function tries to deactivate a timer. Upon successful (ret >= 0)
1219  * exit the timer is not queued and the handler is not running on any CPU.
1220  */
1221 int try_to_del_timer_sync(struct timer_list *timer)
1222 {
1223 	struct timer_base *base;
1224 	unsigned long flags;
1225 	int ret = -1;
1226 
1227 	debug_assert_init(timer);
1228 
1229 	base = lock_timer_base(timer, &flags);
1230 
1231 	if (base->running_timer != timer)
1232 		ret = detach_if_pending(timer, base, true);
1233 
1234 	raw_spin_unlock_irqrestore(&base->lock, flags);
1235 
1236 	return ret;
1237 }
1238 EXPORT_SYMBOL(try_to_del_timer_sync);
1239 
1240 #ifdef CONFIG_PREEMPT_RT
1241 static __init void timer_base_init_expiry_lock(struct timer_base *base)
1242 {
1243 	spin_lock_init(&base->expiry_lock);
1244 }
1245 
1246 static inline void timer_base_lock_expiry(struct timer_base *base)
1247 {
1248 	spin_lock(&base->expiry_lock);
1249 }
1250 
1251 static inline void timer_base_unlock_expiry(struct timer_base *base)
1252 {
1253 	spin_unlock(&base->expiry_lock);
1254 }
1255 
1256 /*
1257  * The counterpart to del_timer_wait_running().
1258  *
1259  * If there is a waiter for base->expiry_lock, then it was waiting for the
1260  * timer callback to finish. Drop expiry_lock and reacquire it. That allows
1261  * the waiter to acquire the lock and make progress.
1262  */
1263 static void timer_sync_wait_running(struct timer_base *base)
1264 {
1265 	if (atomic_read(&base->timer_waiters)) {
1266 		spin_unlock(&base->expiry_lock);
1267 		spin_lock(&base->expiry_lock);
1268 	}
1269 }
1270 
1271 /*
1272  * This function is called on PREEMPT_RT kernels when the fast path
1273  * deletion of a timer failed because the timer callback function was
1274  * running.
1275  *
1276  * This prevents priority inversion, if the softirq thread on a remote CPU
1277  * got preempted, and it prevents a life lock when the task which tries to
1278  * delete a timer preempted the softirq thread running the timer callback
1279  * function.
1280  */
1281 static void del_timer_wait_running(struct timer_list *timer)
1282 {
1283 	u32 tf;
1284 
1285 	tf = READ_ONCE(timer->flags);
1286 	if (!(tf & (TIMER_MIGRATING | TIMER_IRQSAFE))) {
1287 		struct timer_base *base = get_timer_base(tf);
1288 
1289 		/*
1290 		 * Mark the base as contended and grab the expiry lock,
1291 		 * which is held by the softirq across the timer
1292 		 * callback. Drop the lock immediately so the softirq can
1293 		 * expire the next timer. In theory the timer could already
1294 		 * be running again, but that's more than unlikely and just
1295 		 * causes another wait loop.
1296 		 */
1297 		atomic_inc(&base->timer_waiters);
1298 		spin_lock_bh(&base->expiry_lock);
1299 		atomic_dec(&base->timer_waiters);
1300 		spin_unlock_bh(&base->expiry_lock);
1301 	}
1302 }
1303 #else
1304 static inline void timer_base_init_expiry_lock(struct timer_base *base) { }
1305 static inline void timer_base_lock_expiry(struct timer_base *base) { }
1306 static inline void timer_base_unlock_expiry(struct timer_base *base) { }
1307 static inline void timer_sync_wait_running(struct timer_base *base) { }
1308 static inline void del_timer_wait_running(struct timer_list *timer) { }
1309 #endif
1310 
1311 #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT_RT)
1312 /**
1313  * del_timer_sync - deactivate a timer and wait for the handler to finish.
1314  * @timer: the timer to be deactivated
1315  *
1316  * This function only differs from del_timer() on SMP: besides deactivating
1317  * the timer it also makes sure the handler has finished executing on other
1318  * CPUs.
1319  *
1320  * Synchronization rules: Callers must prevent restarting of the timer,
1321  * otherwise this function is meaningless. It must not be called from
1322  * interrupt contexts unless the timer is an irqsafe one. The caller must
1323  * not hold locks which would prevent completion of the timer's
1324  * handler. The timer's handler must not call add_timer_on(). Upon exit the
1325  * timer is not queued and the handler is not running on any CPU.
1326  *
1327  * Note: For !irqsafe timers, you must not hold locks that are held in
1328  *   interrupt context while calling this function. Even if the lock has
1329  *   nothing to do with the timer in question.  Here's why::
1330  *
1331  *    CPU0                             CPU1
1332  *    ----                             ----
1333  *                                     <SOFTIRQ>
1334  *                                       call_timer_fn();
1335  *                                       base->running_timer = mytimer;
1336  *    spin_lock_irq(somelock);
1337  *                                     <IRQ>
1338  *                                        spin_lock(somelock);
1339  *    del_timer_sync(mytimer);
1340  *    while (base->running_timer == mytimer);
1341  *
1342  * Now del_timer_sync() will never return and never release somelock.
1343  * The interrupt on the other CPU is waiting to grab somelock but
1344  * it has interrupted the softirq that CPU0 is waiting to finish.
1345  *
1346  * The function returns whether it has deactivated a pending timer or not.
1347  */
1348 int del_timer_sync(struct timer_list *timer)
1349 {
1350 	int ret;
1351 
1352 #ifdef CONFIG_LOCKDEP
1353 	unsigned long flags;
1354 
1355 	/*
1356 	 * If lockdep gives a backtrace here, please reference
1357 	 * the synchronization rules above.
1358 	 */
1359 	local_irq_save(flags);
1360 	lock_map_acquire(&timer->lockdep_map);
1361 	lock_map_release(&timer->lockdep_map);
1362 	local_irq_restore(flags);
1363 #endif
1364 	/*
1365 	 * don't use it in hardirq context, because it
1366 	 * could lead to deadlock.
1367 	 */
1368 	WARN_ON(in_irq() && !(timer->flags & TIMER_IRQSAFE));
1369 
1370 	/*
1371 	 * Must be able to sleep on PREEMPT_RT because of the slowpath in
1372 	 * del_timer_wait_running().
1373 	 */
1374 	if (IS_ENABLED(CONFIG_PREEMPT_RT) && !(timer->flags & TIMER_IRQSAFE))
1375 		lockdep_assert_preemption_enabled();
1376 
1377 	do {
1378 		ret = try_to_del_timer_sync(timer);
1379 
1380 		if (unlikely(ret < 0)) {
1381 			del_timer_wait_running(timer);
1382 			cpu_relax();
1383 		}
1384 	} while (ret < 0);
1385 
1386 	return ret;
1387 }
1388 EXPORT_SYMBOL(del_timer_sync);
1389 #endif
1390 
1391 static void call_timer_fn(struct timer_list *timer,
1392 			  void (*fn)(struct timer_list *),
1393 			  unsigned long baseclk)
1394 {
1395 	int count = preempt_count();
1396 
1397 #ifdef CONFIG_LOCKDEP
1398 	/*
1399 	 * It is permissible to free the timer from inside the
1400 	 * function that is called from it, this we need to take into
1401 	 * account for lockdep too. To avoid bogus "held lock freed"
1402 	 * warnings as well as problems when looking into
1403 	 * timer->lockdep_map, make a copy and use that here.
1404 	 */
1405 	struct lockdep_map lockdep_map;
1406 
1407 	lockdep_copy_map(&lockdep_map, &timer->lockdep_map);
1408 #endif
1409 	/*
1410 	 * Couple the lock chain with the lock chain at
1411 	 * del_timer_sync() by acquiring the lock_map around the fn()
1412 	 * call here and in del_timer_sync().
1413 	 */
1414 	lock_map_acquire(&lockdep_map);
1415 
1416 	trace_timer_expire_entry(timer, baseclk);
1417 	fn(timer);
1418 	trace_timer_expire_exit(timer);
1419 
1420 	lock_map_release(&lockdep_map);
1421 
1422 	if (count != preempt_count()) {
1423 		WARN_ONCE(1, "timer: %pS preempt leak: %08x -> %08x\n",
1424 			  fn, count, preempt_count());
1425 		/*
1426 		 * Restore the preempt count. That gives us a decent
1427 		 * chance to survive and extract information. If the
1428 		 * callback kept a lock held, bad luck, but not worse
1429 		 * than the BUG() we had.
1430 		 */
1431 		preempt_count_set(count);
1432 	}
1433 }
1434 
1435 static void expire_timers(struct timer_base *base, struct hlist_head *head)
1436 {
1437 	/*
1438 	 * This value is required only for tracing. base->clk was
1439 	 * incremented directly before expire_timers was called. But expiry
1440 	 * is related to the old base->clk value.
1441 	 */
1442 	unsigned long baseclk = base->clk - 1;
1443 
1444 	while (!hlist_empty(head)) {
1445 		struct timer_list *timer;
1446 		void (*fn)(struct timer_list *);
1447 
1448 		timer = hlist_entry(head->first, struct timer_list, entry);
1449 
1450 		base->running_timer = timer;
1451 		detach_timer(timer, true);
1452 
1453 		fn = timer->function;
1454 
1455 		if (timer->flags & TIMER_IRQSAFE) {
1456 			raw_spin_unlock(&base->lock);
1457 			call_timer_fn(timer, fn, baseclk);
1458 			base->running_timer = NULL;
1459 			raw_spin_lock(&base->lock);
1460 		} else {
1461 			raw_spin_unlock_irq(&base->lock);
1462 			call_timer_fn(timer, fn, baseclk);
1463 			base->running_timer = NULL;
1464 			timer_sync_wait_running(base);
1465 			raw_spin_lock_irq(&base->lock);
1466 		}
1467 	}
1468 }
1469 
1470 static int collect_expired_timers(struct timer_base *base,
1471 				  struct hlist_head *heads)
1472 {
1473 	unsigned long clk = base->clk = base->next_expiry;
1474 	struct hlist_head *vec;
1475 	int i, levels = 0;
1476 	unsigned int idx;
1477 
1478 	for (i = 0; i < LVL_DEPTH; i++) {
1479 		idx = (clk & LVL_MASK) + i * LVL_SIZE;
1480 
1481 		if (__test_and_clear_bit(idx, base->pending_map)) {
1482 			vec = base->vectors + idx;
1483 			hlist_move_list(vec, heads++);
1484 			levels++;
1485 		}
1486 		/* Is it time to look at the next level? */
1487 		if (clk & LVL_CLK_MASK)
1488 			break;
1489 		/* Shift clock for the next level granularity */
1490 		clk >>= LVL_CLK_SHIFT;
1491 	}
1492 	return levels;
1493 }
1494 
1495 /*
1496  * Find the next pending bucket of a level. Search from level start (@offset)
1497  * + @clk upwards and if nothing there, search from start of the level
1498  * (@offset) up to @offset + clk.
1499  */
1500 static int next_pending_bucket(struct timer_base *base, unsigned offset,
1501 			       unsigned clk)
1502 {
1503 	unsigned pos, start = offset + clk;
1504 	unsigned end = offset + LVL_SIZE;
1505 
1506 	pos = find_next_bit(base->pending_map, end, start);
1507 	if (pos < end)
1508 		return pos - start;
1509 
1510 	pos = find_next_bit(base->pending_map, start, offset);
1511 	return pos < start ? pos + LVL_SIZE - start : -1;
1512 }
1513 
1514 /*
1515  * Search the first expiring timer in the various clock levels. Caller must
1516  * hold base->lock.
1517  */
1518 static unsigned long __next_timer_interrupt(struct timer_base *base)
1519 {
1520 	unsigned long clk, next, adj;
1521 	unsigned lvl, offset = 0;
1522 
1523 	next = base->clk + NEXT_TIMER_MAX_DELTA;
1524 	clk = base->clk;
1525 	for (lvl = 0; lvl < LVL_DEPTH; lvl++, offset += LVL_SIZE) {
1526 		int pos = next_pending_bucket(base, offset, clk & LVL_MASK);
1527 		unsigned long lvl_clk = clk & LVL_CLK_MASK;
1528 
1529 		if (pos >= 0) {
1530 			unsigned long tmp = clk + (unsigned long) pos;
1531 
1532 			tmp <<= LVL_SHIFT(lvl);
1533 			if (time_before(tmp, next))
1534 				next = tmp;
1535 
1536 			/*
1537 			 * If the next expiration happens before we reach
1538 			 * the next level, no need to check further.
1539 			 */
1540 			if (pos <= ((LVL_CLK_DIV - lvl_clk) & LVL_CLK_MASK))
1541 				break;
1542 		}
1543 		/*
1544 		 * Clock for the next level. If the current level clock lower
1545 		 * bits are zero, we look at the next level as is. If not we
1546 		 * need to advance it by one because that's going to be the
1547 		 * next expiring bucket in that level. base->clk is the next
1548 		 * expiring jiffie. So in case of:
1549 		 *
1550 		 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1551 		 *  0    0    0    0    0    0
1552 		 *
1553 		 * we have to look at all levels @index 0. With
1554 		 *
1555 		 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1556 		 *  0    0    0    0    0    2
1557 		 *
1558 		 * LVL0 has the next expiring bucket @index 2. The upper
1559 		 * levels have the next expiring bucket @index 1.
1560 		 *
1561 		 * In case that the propagation wraps the next level the same
1562 		 * rules apply:
1563 		 *
1564 		 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1565 		 *  0    0    0    0    F    2
1566 		 *
1567 		 * So after looking at LVL0 we get:
1568 		 *
1569 		 * LVL5 LVL4 LVL3 LVL2 LVL1
1570 		 *  0    0    0    1    0
1571 		 *
1572 		 * So no propagation from LVL1 to LVL2 because that happened
1573 		 * with the add already, but then we need to propagate further
1574 		 * from LVL2 to LVL3.
1575 		 *
1576 		 * So the simple check whether the lower bits of the current
1577 		 * level are 0 or not is sufficient for all cases.
1578 		 */
1579 		adj = lvl_clk ? 1 : 0;
1580 		clk >>= LVL_CLK_SHIFT;
1581 		clk += adj;
1582 	}
1583 
1584 	base->next_expiry_recalc = false;
1585 
1586 	return next;
1587 }
1588 
1589 #ifdef CONFIG_NO_HZ_COMMON
1590 /*
1591  * Check, if the next hrtimer event is before the next timer wheel
1592  * event:
1593  */
1594 static u64 cmp_next_hrtimer_event(u64 basem, u64 expires)
1595 {
1596 	u64 nextevt = hrtimer_get_next_event();
1597 
1598 	/*
1599 	 * If high resolution timers are enabled
1600 	 * hrtimer_get_next_event() returns KTIME_MAX.
1601 	 */
1602 	if (expires <= nextevt)
1603 		return expires;
1604 
1605 	/*
1606 	 * If the next timer is already expired, return the tick base
1607 	 * time so the tick is fired immediately.
1608 	 */
1609 	if (nextevt <= basem)
1610 		return basem;
1611 
1612 	/*
1613 	 * Round up to the next jiffie. High resolution timers are
1614 	 * off, so the hrtimers are expired in the tick and we need to
1615 	 * make sure that this tick really expires the timer to avoid
1616 	 * a ping pong of the nohz stop code.
1617 	 *
1618 	 * Use DIV_ROUND_UP_ULL to prevent gcc calling __divdi3
1619 	 */
1620 	return DIV_ROUND_UP_ULL(nextevt, TICK_NSEC) * TICK_NSEC;
1621 }
1622 
1623 /**
1624  * get_next_timer_interrupt - return the time (clock mono) of the next timer
1625  * @basej:	base time jiffies
1626  * @basem:	base time clock monotonic
1627  *
1628  * Returns the tick aligned clock monotonic time of the next pending
1629  * timer or KTIME_MAX if no timer is pending.
1630  */
1631 u64 get_next_timer_interrupt(unsigned long basej, u64 basem)
1632 {
1633 	struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1634 	u64 expires = KTIME_MAX;
1635 	unsigned long nextevt;
1636 	bool is_max_delta;
1637 
1638 	/*
1639 	 * Pretend that there is no timer pending if the cpu is offline.
1640 	 * Possible pending timers will be migrated later to an active cpu.
1641 	 */
1642 	if (cpu_is_offline(smp_processor_id()))
1643 		return expires;
1644 
1645 	raw_spin_lock(&base->lock);
1646 	if (base->next_expiry_recalc)
1647 		base->next_expiry = __next_timer_interrupt(base);
1648 	nextevt = base->next_expiry;
1649 	is_max_delta = (nextevt == base->clk + NEXT_TIMER_MAX_DELTA);
1650 
1651 	/*
1652 	 * We have a fresh next event. Check whether we can forward the
1653 	 * base. We can only do that when @basej is past base->clk
1654 	 * otherwise we might rewind base->clk.
1655 	 */
1656 	if (time_after(basej, base->clk)) {
1657 		if (time_after(nextevt, basej))
1658 			base->clk = basej;
1659 		else if (time_after(nextevt, base->clk))
1660 			base->clk = nextevt;
1661 	}
1662 
1663 	if (time_before_eq(nextevt, basej)) {
1664 		expires = basem;
1665 		base->is_idle = false;
1666 	} else {
1667 		if (!is_max_delta)
1668 			expires = basem + (u64)(nextevt - basej) * TICK_NSEC;
1669 		/*
1670 		 * If we expect to sleep more than a tick, mark the base idle.
1671 		 * Also the tick is stopped so any added timer must forward
1672 		 * the base clk itself to keep granularity small. This idle
1673 		 * logic is only maintained for the BASE_STD base, deferrable
1674 		 * timers may still see large granularity skew (by design).
1675 		 */
1676 		if ((expires - basem) > TICK_NSEC)
1677 			base->is_idle = true;
1678 	}
1679 	raw_spin_unlock(&base->lock);
1680 
1681 	return cmp_next_hrtimer_event(basem, expires);
1682 }
1683 
1684 /**
1685  * timer_clear_idle - Clear the idle state of the timer base
1686  *
1687  * Called with interrupts disabled
1688  */
1689 void timer_clear_idle(void)
1690 {
1691 	struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1692 
1693 	/*
1694 	 * We do this unlocked. The worst outcome is a remote enqueue sending
1695 	 * a pointless IPI, but taking the lock would just make the window for
1696 	 * sending the IPI a few instructions smaller for the cost of taking
1697 	 * the lock in the exit from idle path.
1698 	 */
1699 	base->is_idle = false;
1700 }
1701 #endif
1702 
1703 /**
1704  * __run_timers - run all expired timers (if any) on this CPU.
1705  * @base: the timer vector to be processed.
1706  */
1707 static inline void __run_timers(struct timer_base *base)
1708 {
1709 	struct hlist_head heads[LVL_DEPTH];
1710 	int levels;
1711 
1712 	if (time_before(jiffies, base->next_expiry))
1713 		return;
1714 
1715 	timer_base_lock_expiry(base);
1716 	raw_spin_lock_irq(&base->lock);
1717 
1718 	while (time_after_eq(jiffies, base->clk) &&
1719 	       time_after_eq(jiffies, base->next_expiry)) {
1720 		levels = collect_expired_timers(base, heads);
1721 		/*
1722 		 * The only possible reason for not finding any expired
1723 		 * timer at this clk is that all matching timers have been
1724 		 * dequeued.
1725 		 */
1726 		WARN_ON_ONCE(!levels && !base->next_expiry_recalc);
1727 		base->clk++;
1728 		base->next_expiry = __next_timer_interrupt(base);
1729 
1730 		while (levels--)
1731 			expire_timers(base, heads + levels);
1732 	}
1733 	raw_spin_unlock_irq(&base->lock);
1734 	timer_base_unlock_expiry(base);
1735 }
1736 
1737 /*
1738  * This function runs timers and the timer-tq in bottom half context.
1739  */
1740 static __latent_entropy void run_timer_softirq(struct softirq_action *h)
1741 {
1742 	struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1743 
1744 	__run_timers(base);
1745 	if (IS_ENABLED(CONFIG_NO_HZ_COMMON))
1746 		__run_timers(this_cpu_ptr(&timer_bases[BASE_DEF]));
1747 }
1748 
1749 /*
1750  * Called by the local, per-CPU timer interrupt on SMP.
1751  */
1752 static void run_local_timers(void)
1753 {
1754 	struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1755 
1756 	hrtimer_run_queues();
1757 	/* Raise the softirq only if required. */
1758 	if (time_before(jiffies, base->next_expiry)) {
1759 		if (!IS_ENABLED(CONFIG_NO_HZ_COMMON))
1760 			return;
1761 		/* CPU is awake, so check the deferrable base. */
1762 		base++;
1763 		if (time_before(jiffies, base->next_expiry))
1764 			return;
1765 	}
1766 	raise_softirq(TIMER_SOFTIRQ);
1767 }
1768 
1769 /*
1770  * Called from the timer interrupt handler to charge one tick to the current
1771  * process.  user_tick is 1 if the tick is user time, 0 for system.
1772  */
1773 void update_process_times(int user_tick)
1774 {
1775 	struct task_struct *p = current;
1776 
1777 	PRANDOM_ADD_NOISE(jiffies, user_tick, p, 0);
1778 
1779 	/* Note: this timer irq context must be accounted for as well. */
1780 	account_process_tick(p, user_tick);
1781 	run_local_timers();
1782 	rcu_sched_clock_irq(user_tick);
1783 #ifdef CONFIG_IRQ_WORK
1784 	if (in_irq())
1785 		irq_work_tick();
1786 #endif
1787 	scheduler_tick();
1788 	if (IS_ENABLED(CONFIG_POSIX_TIMERS))
1789 		run_posix_cpu_timers();
1790 }
1791 
1792 /*
1793  * Since schedule_timeout()'s timer is defined on the stack, it must store
1794  * the target task on the stack as well.
1795  */
1796 struct process_timer {
1797 	struct timer_list timer;
1798 	struct task_struct *task;
1799 };
1800 
1801 static void process_timeout(struct timer_list *t)
1802 {
1803 	struct process_timer *timeout = from_timer(timeout, t, timer);
1804 
1805 	wake_up_process(timeout->task);
1806 }
1807 
1808 /**
1809  * schedule_timeout - sleep until timeout
1810  * @timeout: timeout value in jiffies
1811  *
1812  * Make the current task sleep until @timeout jiffies have elapsed.
1813  * The function behavior depends on the current task state
1814  * (see also set_current_state() description):
1815  *
1816  * %TASK_RUNNING - the scheduler is called, but the task does not sleep
1817  * at all. That happens because sched_submit_work() does nothing for
1818  * tasks in %TASK_RUNNING state.
1819  *
1820  * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1821  * pass before the routine returns unless the current task is explicitly
1822  * woken up, (e.g. by wake_up_process()).
1823  *
1824  * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1825  * delivered to the current task or the current task is explicitly woken
1826  * up.
1827  *
1828  * The current task state is guaranteed to be %TASK_RUNNING when this
1829  * routine returns.
1830  *
1831  * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1832  * the CPU away without a bound on the timeout. In this case the return
1833  * value will be %MAX_SCHEDULE_TIMEOUT.
1834  *
1835  * Returns 0 when the timer has expired otherwise the remaining time in
1836  * jiffies will be returned. In all cases the return value is guaranteed
1837  * to be non-negative.
1838  */
1839 signed long __sched schedule_timeout(signed long timeout)
1840 {
1841 	struct process_timer timer;
1842 	unsigned long expire;
1843 
1844 	switch (timeout)
1845 	{
1846 	case MAX_SCHEDULE_TIMEOUT:
1847 		/*
1848 		 * These two special cases are useful to be comfortable
1849 		 * in the caller. Nothing more. We could take
1850 		 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1851 		 * but I' d like to return a valid offset (>=0) to allow
1852 		 * the caller to do everything it want with the retval.
1853 		 */
1854 		schedule();
1855 		goto out;
1856 	default:
1857 		/*
1858 		 * Another bit of PARANOID. Note that the retval will be
1859 		 * 0 since no piece of kernel is supposed to do a check
1860 		 * for a negative retval of schedule_timeout() (since it
1861 		 * should never happens anyway). You just have the printk()
1862 		 * that will tell you if something is gone wrong and where.
1863 		 */
1864 		if (timeout < 0) {
1865 			printk(KERN_ERR "schedule_timeout: wrong timeout "
1866 				"value %lx\n", timeout);
1867 			dump_stack();
1868 			__set_current_state(TASK_RUNNING);
1869 			goto out;
1870 		}
1871 	}
1872 
1873 	expire = timeout + jiffies;
1874 
1875 	timer.task = current;
1876 	timer_setup_on_stack(&timer.timer, process_timeout, 0);
1877 	__mod_timer(&timer.timer, expire, MOD_TIMER_NOTPENDING);
1878 	schedule();
1879 	del_singleshot_timer_sync(&timer.timer);
1880 
1881 	/* Remove the timer from the object tracker */
1882 	destroy_timer_on_stack(&timer.timer);
1883 
1884 	timeout = expire - jiffies;
1885 
1886  out:
1887 	return timeout < 0 ? 0 : timeout;
1888 }
1889 EXPORT_SYMBOL(schedule_timeout);
1890 
1891 /*
1892  * We can use __set_current_state() here because schedule_timeout() calls
1893  * schedule() unconditionally.
1894  */
1895 signed long __sched schedule_timeout_interruptible(signed long timeout)
1896 {
1897 	__set_current_state(TASK_INTERRUPTIBLE);
1898 	return schedule_timeout(timeout);
1899 }
1900 EXPORT_SYMBOL(schedule_timeout_interruptible);
1901 
1902 signed long __sched schedule_timeout_killable(signed long timeout)
1903 {
1904 	__set_current_state(TASK_KILLABLE);
1905 	return schedule_timeout(timeout);
1906 }
1907 EXPORT_SYMBOL(schedule_timeout_killable);
1908 
1909 signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1910 {
1911 	__set_current_state(TASK_UNINTERRUPTIBLE);
1912 	return schedule_timeout(timeout);
1913 }
1914 EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1915 
1916 /*
1917  * Like schedule_timeout_uninterruptible(), except this task will not contribute
1918  * to load average.
1919  */
1920 signed long __sched schedule_timeout_idle(signed long timeout)
1921 {
1922 	__set_current_state(TASK_IDLE);
1923 	return schedule_timeout(timeout);
1924 }
1925 EXPORT_SYMBOL(schedule_timeout_idle);
1926 
1927 #ifdef CONFIG_HOTPLUG_CPU
1928 static void migrate_timer_list(struct timer_base *new_base, struct hlist_head *head)
1929 {
1930 	struct timer_list *timer;
1931 	int cpu = new_base->cpu;
1932 
1933 	while (!hlist_empty(head)) {
1934 		timer = hlist_entry(head->first, struct timer_list, entry);
1935 		detach_timer(timer, false);
1936 		timer->flags = (timer->flags & ~TIMER_BASEMASK) | cpu;
1937 		internal_add_timer(new_base, timer);
1938 	}
1939 }
1940 
1941 int timers_prepare_cpu(unsigned int cpu)
1942 {
1943 	struct timer_base *base;
1944 	int b;
1945 
1946 	for (b = 0; b < NR_BASES; b++) {
1947 		base = per_cpu_ptr(&timer_bases[b], cpu);
1948 		base->clk = jiffies;
1949 		base->next_expiry = base->clk + NEXT_TIMER_MAX_DELTA;
1950 		base->is_idle = false;
1951 	}
1952 	return 0;
1953 }
1954 
1955 int timers_dead_cpu(unsigned int cpu)
1956 {
1957 	struct timer_base *old_base;
1958 	struct timer_base *new_base;
1959 	int b, i;
1960 
1961 	BUG_ON(cpu_online(cpu));
1962 
1963 	for (b = 0; b < NR_BASES; b++) {
1964 		old_base = per_cpu_ptr(&timer_bases[b], cpu);
1965 		new_base = get_cpu_ptr(&timer_bases[b]);
1966 		/*
1967 		 * The caller is globally serialized and nobody else
1968 		 * takes two locks at once, deadlock is not possible.
1969 		 */
1970 		raw_spin_lock_irq(&new_base->lock);
1971 		raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1972 
1973 		/*
1974 		 * The current CPUs base clock might be stale. Update it
1975 		 * before moving the timers over.
1976 		 */
1977 		forward_timer_base(new_base);
1978 
1979 		BUG_ON(old_base->running_timer);
1980 
1981 		for (i = 0; i < WHEEL_SIZE; i++)
1982 			migrate_timer_list(new_base, old_base->vectors + i);
1983 
1984 		raw_spin_unlock(&old_base->lock);
1985 		raw_spin_unlock_irq(&new_base->lock);
1986 		put_cpu_ptr(&timer_bases);
1987 	}
1988 	return 0;
1989 }
1990 
1991 #endif /* CONFIG_HOTPLUG_CPU */
1992 
1993 static void __init init_timer_cpu(int cpu)
1994 {
1995 	struct timer_base *base;
1996 	int i;
1997 
1998 	for (i = 0; i < NR_BASES; i++) {
1999 		base = per_cpu_ptr(&timer_bases[i], cpu);
2000 		base->cpu = cpu;
2001 		raw_spin_lock_init(&base->lock);
2002 		base->clk = jiffies;
2003 		base->next_expiry = base->clk + NEXT_TIMER_MAX_DELTA;
2004 		timer_base_init_expiry_lock(base);
2005 	}
2006 }
2007 
2008 static void __init init_timer_cpus(void)
2009 {
2010 	int cpu;
2011 
2012 	for_each_possible_cpu(cpu)
2013 		init_timer_cpu(cpu);
2014 }
2015 
2016 void __init init_timers(void)
2017 {
2018 	init_timer_cpus();
2019 	posix_cputimers_init_work();
2020 	open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
2021 }
2022 
2023 /**
2024  * msleep - sleep safely even with waitqueue interruptions
2025  * @msecs: Time in milliseconds to sleep for
2026  */
2027 void msleep(unsigned int msecs)
2028 {
2029 	unsigned long timeout = msecs_to_jiffies(msecs) + 1;
2030 
2031 	while (timeout)
2032 		timeout = schedule_timeout_uninterruptible(timeout);
2033 }
2034 
2035 EXPORT_SYMBOL(msleep);
2036 
2037 /**
2038  * msleep_interruptible - sleep waiting for signals
2039  * @msecs: Time in milliseconds to sleep for
2040  */
2041 unsigned long msleep_interruptible(unsigned int msecs)
2042 {
2043 	unsigned long timeout = msecs_to_jiffies(msecs) + 1;
2044 
2045 	while (timeout && !signal_pending(current))
2046 		timeout = schedule_timeout_interruptible(timeout);
2047 	return jiffies_to_msecs(timeout);
2048 }
2049 
2050 EXPORT_SYMBOL(msleep_interruptible);
2051 
2052 /**
2053  * usleep_range - Sleep for an approximate time
2054  * @min: Minimum time in usecs to sleep
2055  * @max: Maximum time in usecs to sleep
2056  *
2057  * In non-atomic context where the exact wakeup time is flexible, use
2058  * usleep_range() instead of udelay().  The sleep improves responsiveness
2059  * by avoiding the CPU-hogging busy-wait of udelay(), and the range reduces
2060  * power usage by allowing hrtimers to take advantage of an already-
2061  * scheduled interrupt instead of scheduling a new one just for this sleep.
2062  */
2063 void __sched usleep_range(unsigned long min, unsigned long max)
2064 {
2065 	ktime_t exp = ktime_add_us(ktime_get(), min);
2066 	u64 delta = (u64)(max - min) * NSEC_PER_USEC;
2067 
2068 	for (;;) {
2069 		__set_current_state(TASK_UNINTERRUPTIBLE);
2070 		/* Do not return before the requested sleep time has elapsed */
2071 		if (!schedule_hrtimeout_range(&exp, delta, HRTIMER_MODE_ABS))
2072 			break;
2073 	}
2074 }
2075 EXPORT_SYMBOL(usleep_range);
2076