xref: /openbmc/linux/kernel/time/timer.c (revision ed84ef1c)
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 	bool			timers_pending;
211 	DECLARE_BITMAP(pending_map, WHEEL_SIZE);
212 	struct hlist_head	vectors[WHEEL_SIZE];
213 } ____cacheline_aligned;
214 
215 static DEFINE_PER_CPU(struct timer_base, timer_bases[NR_BASES]);
216 
217 #ifdef CONFIG_NO_HZ_COMMON
218 
219 static DEFINE_STATIC_KEY_FALSE(timers_nohz_active);
220 static DEFINE_MUTEX(timer_keys_mutex);
221 
222 static void timer_update_keys(struct work_struct *work);
223 static DECLARE_WORK(timer_update_work, timer_update_keys);
224 
225 #ifdef CONFIG_SMP
226 unsigned int sysctl_timer_migration = 1;
227 
228 DEFINE_STATIC_KEY_FALSE(timers_migration_enabled);
229 
230 static void timers_update_migration(void)
231 {
232 	if (sysctl_timer_migration && tick_nohz_active)
233 		static_branch_enable(&timers_migration_enabled);
234 	else
235 		static_branch_disable(&timers_migration_enabled);
236 }
237 #else
238 static inline void timers_update_migration(void) { }
239 #endif /* !CONFIG_SMP */
240 
241 static void timer_update_keys(struct work_struct *work)
242 {
243 	mutex_lock(&timer_keys_mutex);
244 	timers_update_migration();
245 	static_branch_enable(&timers_nohz_active);
246 	mutex_unlock(&timer_keys_mutex);
247 }
248 
249 void timers_update_nohz(void)
250 {
251 	schedule_work(&timer_update_work);
252 }
253 
254 int timer_migration_handler(struct ctl_table *table, int write,
255 			    void *buffer, size_t *lenp, loff_t *ppos)
256 {
257 	int ret;
258 
259 	mutex_lock(&timer_keys_mutex);
260 	ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
261 	if (!ret && write)
262 		timers_update_migration();
263 	mutex_unlock(&timer_keys_mutex);
264 	return ret;
265 }
266 
267 static inline bool is_timers_nohz_active(void)
268 {
269 	return static_branch_unlikely(&timers_nohz_active);
270 }
271 #else
272 static inline bool is_timers_nohz_active(void) { return false; }
273 #endif /* NO_HZ_COMMON */
274 
275 static unsigned long round_jiffies_common(unsigned long j, int cpu,
276 		bool force_up)
277 {
278 	int rem;
279 	unsigned long original = j;
280 
281 	/*
282 	 * We don't want all cpus firing their timers at once hitting the
283 	 * same lock or cachelines, so we skew each extra cpu with an extra
284 	 * 3 jiffies. This 3 jiffies came originally from the mm/ code which
285 	 * already did this.
286 	 * The skew is done by adding 3*cpunr, then round, then subtract this
287 	 * extra offset again.
288 	 */
289 	j += cpu * 3;
290 
291 	rem = j % HZ;
292 
293 	/*
294 	 * If the target jiffie is just after a whole second (which can happen
295 	 * due to delays of the timer irq, long irq off times etc etc) then
296 	 * we should round down to the whole second, not up. Use 1/4th second
297 	 * as cutoff for this rounding as an extreme upper bound for this.
298 	 * But never round down if @force_up is set.
299 	 */
300 	if (rem < HZ/4 && !force_up) /* round down */
301 		j = j - rem;
302 	else /* round up */
303 		j = j - rem + HZ;
304 
305 	/* now that we have rounded, subtract the extra skew again */
306 	j -= cpu * 3;
307 
308 	/*
309 	 * Make sure j is still in the future. Otherwise return the
310 	 * unmodified value.
311 	 */
312 	return time_is_after_jiffies(j) ? j : original;
313 }
314 
315 /**
316  * __round_jiffies - function to round jiffies to a full second
317  * @j: the time in (absolute) jiffies that should be rounded
318  * @cpu: the processor number on which the timeout will happen
319  *
320  * __round_jiffies() rounds an absolute time in the future (in jiffies)
321  * up or down to (approximately) full seconds. This is useful for timers
322  * for which the exact time they fire does not matter too much, as long as
323  * they fire approximately every X seconds.
324  *
325  * By rounding these timers to whole seconds, all such timers will fire
326  * at the same time, rather than at various times spread out. The goal
327  * of this is to have the CPU wake up less, which saves power.
328  *
329  * The exact rounding is skewed for each processor to avoid all
330  * processors firing at the exact same time, which could lead
331  * to lock contention or spurious cache line bouncing.
332  *
333  * The return value is the rounded version of the @j parameter.
334  */
335 unsigned long __round_jiffies(unsigned long j, int cpu)
336 {
337 	return round_jiffies_common(j, cpu, false);
338 }
339 EXPORT_SYMBOL_GPL(__round_jiffies);
340 
341 /**
342  * __round_jiffies_relative - function to round jiffies to a full second
343  * @j: the time in (relative) jiffies that should be rounded
344  * @cpu: the processor number on which the timeout will happen
345  *
346  * __round_jiffies_relative() rounds a time delta  in the future (in jiffies)
347  * up or down to (approximately) full seconds. This is useful for timers
348  * for which the exact time they fire does not matter too much, as long as
349  * they fire approximately every X seconds.
350  *
351  * By rounding these timers to whole seconds, all such timers will fire
352  * at the same time, rather than at various times spread out. The goal
353  * of this is to have the CPU wake up less, which saves power.
354  *
355  * The exact rounding is skewed for each processor to avoid all
356  * processors firing at the exact same time, which could lead
357  * to lock contention or spurious cache line bouncing.
358  *
359  * The return value is the rounded version of the @j parameter.
360  */
361 unsigned long __round_jiffies_relative(unsigned long j, int cpu)
362 {
363 	unsigned long j0 = jiffies;
364 
365 	/* Use j0 because jiffies might change while we run */
366 	return round_jiffies_common(j + j0, cpu, false) - j0;
367 }
368 EXPORT_SYMBOL_GPL(__round_jiffies_relative);
369 
370 /**
371  * round_jiffies - function to round jiffies to a full second
372  * @j: the time in (absolute) jiffies that should be rounded
373  *
374  * round_jiffies() rounds an absolute time in the future (in jiffies)
375  * up or down to (approximately) full seconds. This is useful for timers
376  * for which the exact time they fire does not matter too much, as long as
377  * they fire approximately every X seconds.
378  *
379  * By rounding these timers to whole seconds, all such timers will fire
380  * at the same time, rather than at various times spread out. The goal
381  * of this is to have the CPU wake up less, which saves power.
382  *
383  * The return value is the rounded version of the @j parameter.
384  */
385 unsigned long round_jiffies(unsigned long j)
386 {
387 	return round_jiffies_common(j, raw_smp_processor_id(), false);
388 }
389 EXPORT_SYMBOL_GPL(round_jiffies);
390 
391 /**
392  * round_jiffies_relative - function to round jiffies to a full second
393  * @j: the time in (relative) jiffies that should be rounded
394  *
395  * round_jiffies_relative() rounds a time delta  in the future (in jiffies)
396  * up or down to (approximately) full seconds. This is useful for timers
397  * for which the exact time they fire does not matter too much, as long as
398  * they fire approximately every X seconds.
399  *
400  * By rounding these timers to whole seconds, all such timers will fire
401  * at the same time, rather than at various times spread out. The goal
402  * of this is to have the CPU wake up less, which saves power.
403  *
404  * The return value is the rounded version of the @j parameter.
405  */
406 unsigned long round_jiffies_relative(unsigned long j)
407 {
408 	return __round_jiffies_relative(j, raw_smp_processor_id());
409 }
410 EXPORT_SYMBOL_GPL(round_jiffies_relative);
411 
412 /**
413  * __round_jiffies_up - function to round jiffies up to a full second
414  * @j: the time in (absolute) jiffies that should be rounded
415  * @cpu: the processor number on which the timeout will happen
416  *
417  * This is the same as __round_jiffies() except that it will never
418  * round down.  This is useful for timeouts for which the exact time
419  * of firing does not matter too much, as long as they don't fire too
420  * early.
421  */
422 unsigned long __round_jiffies_up(unsigned long j, int cpu)
423 {
424 	return round_jiffies_common(j, cpu, true);
425 }
426 EXPORT_SYMBOL_GPL(__round_jiffies_up);
427 
428 /**
429  * __round_jiffies_up_relative - function to round jiffies up to a full second
430  * @j: the time in (relative) jiffies that should be rounded
431  * @cpu: the processor number on which the timeout will happen
432  *
433  * This is the same as __round_jiffies_relative() except that it will never
434  * round down.  This is useful for timeouts for which the exact time
435  * of firing does not matter too much, as long as they don't fire too
436  * early.
437  */
438 unsigned long __round_jiffies_up_relative(unsigned long j, int cpu)
439 {
440 	unsigned long j0 = jiffies;
441 
442 	/* Use j0 because jiffies might change while we run */
443 	return round_jiffies_common(j + j0, cpu, true) - j0;
444 }
445 EXPORT_SYMBOL_GPL(__round_jiffies_up_relative);
446 
447 /**
448  * round_jiffies_up - function to round jiffies up to a full second
449  * @j: the time in (absolute) jiffies that should be rounded
450  *
451  * This is the same as round_jiffies() except that it will never
452  * round down.  This is useful for timeouts for which the exact time
453  * of firing does not matter too much, as long as they don't fire too
454  * early.
455  */
456 unsigned long round_jiffies_up(unsigned long j)
457 {
458 	return round_jiffies_common(j, raw_smp_processor_id(), true);
459 }
460 EXPORT_SYMBOL_GPL(round_jiffies_up);
461 
462 /**
463  * round_jiffies_up_relative - function to round jiffies up to a full second
464  * @j: the time in (relative) jiffies that should be rounded
465  *
466  * This is the same as round_jiffies_relative() except that it will never
467  * round down.  This is useful for timeouts for which the exact time
468  * of firing does not matter too much, as long as they don't fire too
469  * early.
470  */
471 unsigned long round_jiffies_up_relative(unsigned long j)
472 {
473 	return __round_jiffies_up_relative(j, raw_smp_processor_id());
474 }
475 EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
476 
477 
478 static inline unsigned int timer_get_idx(struct timer_list *timer)
479 {
480 	return (timer->flags & TIMER_ARRAYMASK) >> TIMER_ARRAYSHIFT;
481 }
482 
483 static inline void timer_set_idx(struct timer_list *timer, unsigned int idx)
484 {
485 	timer->flags = (timer->flags & ~TIMER_ARRAYMASK) |
486 			idx << TIMER_ARRAYSHIFT;
487 }
488 
489 /*
490  * Helper function to calculate the array index for a given expiry
491  * time.
492  */
493 static inline unsigned calc_index(unsigned long expires, unsigned lvl,
494 				  unsigned long *bucket_expiry)
495 {
496 
497 	/*
498 	 * The timer wheel has to guarantee that a timer does not fire
499 	 * early. Early expiry can happen due to:
500 	 * - Timer is armed at the edge of a tick
501 	 * - Truncation of the expiry time in the outer wheel levels
502 	 *
503 	 * Round up with level granularity to prevent this.
504 	 */
505 	expires = (expires + LVL_GRAN(lvl)) >> LVL_SHIFT(lvl);
506 	*bucket_expiry = expires << LVL_SHIFT(lvl);
507 	return LVL_OFFS(lvl) + (expires & LVL_MASK);
508 }
509 
510 static int calc_wheel_index(unsigned long expires, unsigned long clk,
511 			    unsigned long *bucket_expiry)
512 {
513 	unsigned long delta = expires - clk;
514 	unsigned int idx;
515 
516 	if (delta < LVL_START(1)) {
517 		idx = calc_index(expires, 0, bucket_expiry);
518 	} else if (delta < LVL_START(2)) {
519 		idx = calc_index(expires, 1, bucket_expiry);
520 	} else if (delta < LVL_START(3)) {
521 		idx = calc_index(expires, 2, bucket_expiry);
522 	} else if (delta < LVL_START(4)) {
523 		idx = calc_index(expires, 3, bucket_expiry);
524 	} else if (delta < LVL_START(5)) {
525 		idx = calc_index(expires, 4, bucket_expiry);
526 	} else if (delta < LVL_START(6)) {
527 		idx = calc_index(expires, 5, bucket_expiry);
528 	} else if (delta < LVL_START(7)) {
529 		idx = calc_index(expires, 6, bucket_expiry);
530 	} else if (LVL_DEPTH > 8 && delta < LVL_START(8)) {
531 		idx = calc_index(expires, 7, bucket_expiry);
532 	} else if ((long) delta < 0) {
533 		idx = clk & LVL_MASK;
534 		*bucket_expiry = clk;
535 	} else {
536 		/*
537 		 * Force expire obscene large timeouts to expire at the
538 		 * capacity limit of the wheel.
539 		 */
540 		if (delta >= WHEEL_TIMEOUT_CUTOFF)
541 			expires = clk + WHEEL_TIMEOUT_MAX;
542 
543 		idx = calc_index(expires, LVL_DEPTH - 1, bucket_expiry);
544 	}
545 	return idx;
546 }
547 
548 static void
549 trigger_dyntick_cpu(struct timer_base *base, struct timer_list *timer)
550 {
551 	if (!is_timers_nohz_active())
552 		return;
553 
554 	/*
555 	 * TODO: This wants some optimizing similar to the code below, but we
556 	 * will do that when we switch from push to pull for deferrable timers.
557 	 */
558 	if (timer->flags & TIMER_DEFERRABLE) {
559 		if (tick_nohz_full_cpu(base->cpu))
560 			wake_up_nohz_cpu(base->cpu);
561 		return;
562 	}
563 
564 	/*
565 	 * We might have to IPI the remote CPU if the base is idle and the
566 	 * timer is not deferrable. If the other CPU is on the way to idle
567 	 * then it can't set base->is_idle as we hold the base lock:
568 	 */
569 	if (base->is_idle)
570 		wake_up_nohz_cpu(base->cpu);
571 }
572 
573 /*
574  * Enqueue the timer into the hash bucket, mark it pending in
575  * the bitmap, store the index in the timer flags then wake up
576  * the target CPU if needed.
577  */
578 static void enqueue_timer(struct timer_base *base, struct timer_list *timer,
579 			  unsigned int idx, unsigned long bucket_expiry)
580 {
581 
582 	hlist_add_head(&timer->entry, base->vectors + idx);
583 	__set_bit(idx, base->pending_map);
584 	timer_set_idx(timer, idx);
585 
586 	trace_timer_start(timer, timer->expires, timer->flags);
587 
588 	/*
589 	 * Check whether this is the new first expiring timer. The
590 	 * effective expiry time of the timer is required here
591 	 * (bucket_expiry) instead of timer->expires.
592 	 */
593 	if (time_before(bucket_expiry, base->next_expiry)) {
594 		/*
595 		 * Set the next expiry time and kick the CPU so it
596 		 * can reevaluate the wheel:
597 		 */
598 		base->next_expiry = bucket_expiry;
599 		base->timers_pending = true;
600 		base->next_expiry_recalc = false;
601 		trigger_dyntick_cpu(base, timer);
602 	}
603 }
604 
605 static void internal_add_timer(struct timer_base *base, struct timer_list *timer)
606 {
607 	unsigned long bucket_expiry;
608 	unsigned int idx;
609 
610 	idx = calc_wheel_index(timer->expires, base->clk, &bucket_expiry);
611 	enqueue_timer(base, timer, idx, bucket_expiry);
612 }
613 
614 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
615 
616 static const struct debug_obj_descr timer_debug_descr;
617 
618 static void *timer_debug_hint(void *addr)
619 {
620 	return ((struct timer_list *) addr)->function;
621 }
622 
623 static bool timer_is_static_object(void *addr)
624 {
625 	struct timer_list *timer = addr;
626 
627 	return (timer->entry.pprev == NULL &&
628 		timer->entry.next == TIMER_ENTRY_STATIC);
629 }
630 
631 /*
632  * fixup_init is called when:
633  * - an active object is initialized
634  */
635 static bool timer_fixup_init(void *addr, enum debug_obj_state state)
636 {
637 	struct timer_list *timer = addr;
638 
639 	switch (state) {
640 	case ODEBUG_STATE_ACTIVE:
641 		del_timer_sync(timer);
642 		debug_object_init(timer, &timer_debug_descr);
643 		return true;
644 	default:
645 		return false;
646 	}
647 }
648 
649 /* Stub timer callback for improperly used timers. */
650 static void stub_timer(struct timer_list *unused)
651 {
652 	WARN_ON(1);
653 }
654 
655 /*
656  * fixup_activate is called when:
657  * - an active object is activated
658  * - an unknown non-static object is activated
659  */
660 static bool timer_fixup_activate(void *addr, enum debug_obj_state state)
661 {
662 	struct timer_list *timer = addr;
663 
664 	switch (state) {
665 	case ODEBUG_STATE_NOTAVAILABLE:
666 		timer_setup(timer, stub_timer, 0);
667 		return true;
668 
669 	case ODEBUG_STATE_ACTIVE:
670 		WARN_ON(1);
671 		fallthrough;
672 	default:
673 		return false;
674 	}
675 }
676 
677 /*
678  * fixup_free is called when:
679  * - an active object is freed
680  */
681 static bool timer_fixup_free(void *addr, enum debug_obj_state state)
682 {
683 	struct timer_list *timer = addr;
684 
685 	switch (state) {
686 	case ODEBUG_STATE_ACTIVE:
687 		del_timer_sync(timer);
688 		debug_object_free(timer, &timer_debug_descr);
689 		return true;
690 	default:
691 		return false;
692 	}
693 }
694 
695 /*
696  * fixup_assert_init is called when:
697  * - an untracked/uninit-ed object is found
698  */
699 static bool timer_fixup_assert_init(void *addr, enum debug_obj_state state)
700 {
701 	struct timer_list *timer = addr;
702 
703 	switch (state) {
704 	case ODEBUG_STATE_NOTAVAILABLE:
705 		timer_setup(timer, stub_timer, 0);
706 		return true;
707 	default:
708 		return false;
709 	}
710 }
711 
712 static const struct debug_obj_descr timer_debug_descr = {
713 	.name			= "timer_list",
714 	.debug_hint		= timer_debug_hint,
715 	.is_static_object	= timer_is_static_object,
716 	.fixup_init		= timer_fixup_init,
717 	.fixup_activate		= timer_fixup_activate,
718 	.fixup_free		= timer_fixup_free,
719 	.fixup_assert_init	= timer_fixup_assert_init,
720 };
721 
722 static inline void debug_timer_init(struct timer_list *timer)
723 {
724 	debug_object_init(timer, &timer_debug_descr);
725 }
726 
727 static inline void debug_timer_activate(struct timer_list *timer)
728 {
729 	debug_object_activate(timer, &timer_debug_descr);
730 }
731 
732 static inline void debug_timer_deactivate(struct timer_list *timer)
733 {
734 	debug_object_deactivate(timer, &timer_debug_descr);
735 }
736 
737 static inline void debug_timer_assert_init(struct timer_list *timer)
738 {
739 	debug_object_assert_init(timer, &timer_debug_descr);
740 }
741 
742 static void do_init_timer(struct timer_list *timer,
743 			  void (*func)(struct timer_list *),
744 			  unsigned int flags,
745 			  const char *name, struct lock_class_key *key);
746 
747 void init_timer_on_stack_key(struct timer_list *timer,
748 			     void (*func)(struct timer_list *),
749 			     unsigned int flags,
750 			     const char *name, struct lock_class_key *key)
751 {
752 	debug_object_init_on_stack(timer, &timer_debug_descr);
753 	do_init_timer(timer, func, flags, name, key);
754 }
755 EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
756 
757 void destroy_timer_on_stack(struct timer_list *timer)
758 {
759 	debug_object_free(timer, &timer_debug_descr);
760 }
761 EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
762 
763 #else
764 static inline void debug_timer_init(struct timer_list *timer) { }
765 static inline void debug_timer_activate(struct timer_list *timer) { }
766 static inline void debug_timer_deactivate(struct timer_list *timer) { }
767 static inline void debug_timer_assert_init(struct timer_list *timer) { }
768 #endif
769 
770 static inline void debug_init(struct timer_list *timer)
771 {
772 	debug_timer_init(timer);
773 	trace_timer_init(timer);
774 }
775 
776 static inline void debug_deactivate(struct timer_list *timer)
777 {
778 	debug_timer_deactivate(timer);
779 	trace_timer_cancel(timer);
780 }
781 
782 static inline void debug_assert_init(struct timer_list *timer)
783 {
784 	debug_timer_assert_init(timer);
785 }
786 
787 static void do_init_timer(struct timer_list *timer,
788 			  void (*func)(struct timer_list *),
789 			  unsigned int flags,
790 			  const char *name, struct lock_class_key *key)
791 {
792 	timer->entry.pprev = NULL;
793 	timer->function = func;
794 	if (WARN_ON_ONCE(flags & ~TIMER_INIT_FLAGS))
795 		flags &= TIMER_INIT_FLAGS;
796 	timer->flags = flags | raw_smp_processor_id();
797 	lockdep_init_map(&timer->lockdep_map, name, key, 0);
798 }
799 
800 /**
801  * init_timer_key - initialize a timer
802  * @timer: the timer to be initialized
803  * @func: timer callback function
804  * @flags: timer flags
805  * @name: name of the timer
806  * @key: lockdep class key of the fake lock used for tracking timer
807  *       sync lock dependencies
808  *
809  * init_timer_key() must be done to a timer prior calling *any* of the
810  * other timer functions.
811  */
812 void init_timer_key(struct timer_list *timer,
813 		    void (*func)(struct timer_list *), unsigned int flags,
814 		    const char *name, struct lock_class_key *key)
815 {
816 	debug_init(timer);
817 	do_init_timer(timer, func, flags, name, key);
818 }
819 EXPORT_SYMBOL(init_timer_key);
820 
821 static inline void detach_timer(struct timer_list *timer, bool clear_pending)
822 {
823 	struct hlist_node *entry = &timer->entry;
824 
825 	debug_deactivate(timer);
826 
827 	__hlist_del(entry);
828 	if (clear_pending)
829 		entry->pprev = NULL;
830 	entry->next = LIST_POISON2;
831 }
832 
833 static int detach_if_pending(struct timer_list *timer, struct timer_base *base,
834 			     bool clear_pending)
835 {
836 	unsigned idx = timer_get_idx(timer);
837 
838 	if (!timer_pending(timer))
839 		return 0;
840 
841 	if (hlist_is_singular_node(&timer->entry, base->vectors + idx)) {
842 		__clear_bit(idx, base->pending_map);
843 		base->next_expiry_recalc = true;
844 	}
845 
846 	detach_timer(timer, clear_pending);
847 	return 1;
848 }
849 
850 static inline struct timer_base *get_timer_cpu_base(u32 tflags, u32 cpu)
851 {
852 	struct timer_base *base = per_cpu_ptr(&timer_bases[BASE_STD], cpu);
853 
854 	/*
855 	 * If the timer is deferrable and NO_HZ_COMMON is set then we need
856 	 * to use the deferrable base.
857 	 */
858 	if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE))
859 		base = per_cpu_ptr(&timer_bases[BASE_DEF], cpu);
860 	return base;
861 }
862 
863 static inline struct timer_base *get_timer_this_cpu_base(u32 tflags)
864 {
865 	struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
866 
867 	/*
868 	 * If the timer is deferrable and NO_HZ_COMMON is set then we need
869 	 * to use the deferrable base.
870 	 */
871 	if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE))
872 		base = this_cpu_ptr(&timer_bases[BASE_DEF]);
873 	return base;
874 }
875 
876 static inline struct timer_base *get_timer_base(u32 tflags)
877 {
878 	return get_timer_cpu_base(tflags, tflags & TIMER_CPUMASK);
879 }
880 
881 static inline struct timer_base *
882 get_target_base(struct timer_base *base, unsigned tflags)
883 {
884 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
885 	if (static_branch_likely(&timers_migration_enabled) &&
886 	    !(tflags & TIMER_PINNED))
887 		return get_timer_cpu_base(tflags, get_nohz_timer_target());
888 #endif
889 	return get_timer_this_cpu_base(tflags);
890 }
891 
892 static inline void forward_timer_base(struct timer_base *base)
893 {
894 	unsigned long jnow = READ_ONCE(jiffies);
895 
896 	/*
897 	 * No need to forward if we are close enough below jiffies.
898 	 * Also while executing timers, base->clk is 1 offset ahead
899 	 * of jiffies to avoid endless requeuing to current jiffies.
900 	 */
901 	if ((long)(jnow - base->clk) < 1)
902 		return;
903 
904 	/*
905 	 * If the next expiry value is > jiffies, then we fast forward to
906 	 * jiffies otherwise we forward to the next expiry value.
907 	 */
908 	if (time_after(base->next_expiry, jnow)) {
909 		base->clk = jnow;
910 	} else {
911 		if (WARN_ON_ONCE(time_before(base->next_expiry, base->clk)))
912 			return;
913 		base->clk = base->next_expiry;
914 	}
915 }
916 
917 
918 /*
919  * We are using hashed locking: Holding per_cpu(timer_bases[x]).lock means
920  * that all timers which are tied to this base are locked, and the base itself
921  * is locked too.
922  *
923  * So __run_timers/migrate_timers can safely modify all timers which could
924  * be found in the base->vectors array.
925  *
926  * When a timer is migrating then the TIMER_MIGRATING flag is set and we need
927  * to wait until the migration is done.
928  */
929 static struct timer_base *lock_timer_base(struct timer_list *timer,
930 					  unsigned long *flags)
931 	__acquires(timer->base->lock)
932 {
933 	for (;;) {
934 		struct timer_base *base;
935 		u32 tf;
936 
937 		/*
938 		 * We need to use READ_ONCE() here, otherwise the compiler
939 		 * might re-read @tf between the check for TIMER_MIGRATING
940 		 * and spin_lock().
941 		 */
942 		tf = READ_ONCE(timer->flags);
943 
944 		if (!(tf & TIMER_MIGRATING)) {
945 			base = get_timer_base(tf);
946 			raw_spin_lock_irqsave(&base->lock, *flags);
947 			if (timer->flags == tf)
948 				return base;
949 			raw_spin_unlock_irqrestore(&base->lock, *flags);
950 		}
951 		cpu_relax();
952 	}
953 }
954 
955 #define MOD_TIMER_PENDING_ONLY		0x01
956 #define MOD_TIMER_REDUCE		0x02
957 #define MOD_TIMER_NOTPENDING		0x04
958 
959 static inline int
960 __mod_timer(struct timer_list *timer, unsigned long expires, unsigned int options)
961 {
962 	unsigned long clk = 0, flags, bucket_expiry;
963 	struct timer_base *base, *new_base;
964 	unsigned int idx = UINT_MAX;
965 	int ret = 0;
966 
967 	BUG_ON(!timer->function);
968 
969 	/*
970 	 * This is a common optimization triggered by the networking code - if
971 	 * the timer is re-modified to have the same timeout or ends up in the
972 	 * same array bucket then just return:
973 	 */
974 	if (!(options & MOD_TIMER_NOTPENDING) && timer_pending(timer)) {
975 		/*
976 		 * The downside of this optimization is that it can result in
977 		 * larger granularity than you would get from adding a new
978 		 * timer with this expiry.
979 		 */
980 		long diff = timer->expires - expires;
981 
982 		if (!diff)
983 			return 1;
984 		if (options & MOD_TIMER_REDUCE && diff <= 0)
985 			return 1;
986 
987 		/*
988 		 * We lock timer base and calculate the bucket index right
989 		 * here. If the timer ends up in the same bucket, then we
990 		 * just update the expiry time and avoid the whole
991 		 * dequeue/enqueue dance.
992 		 */
993 		base = lock_timer_base(timer, &flags);
994 		forward_timer_base(base);
995 
996 		if (timer_pending(timer) && (options & MOD_TIMER_REDUCE) &&
997 		    time_before_eq(timer->expires, expires)) {
998 			ret = 1;
999 			goto out_unlock;
1000 		}
1001 
1002 		clk = base->clk;
1003 		idx = calc_wheel_index(expires, clk, &bucket_expiry);
1004 
1005 		/*
1006 		 * Retrieve and compare the array index of the pending
1007 		 * timer. If it matches set the expiry to the new value so a
1008 		 * subsequent call will exit in the expires check above.
1009 		 */
1010 		if (idx == timer_get_idx(timer)) {
1011 			if (!(options & MOD_TIMER_REDUCE))
1012 				timer->expires = expires;
1013 			else if (time_after(timer->expires, expires))
1014 				timer->expires = expires;
1015 			ret = 1;
1016 			goto out_unlock;
1017 		}
1018 	} else {
1019 		base = lock_timer_base(timer, &flags);
1020 		forward_timer_base(base);
1021 	}
1022 
1023 	ret = detach_if_pending(timer, base, false);
1024 	if (!ret && (options & MOD_TIMER_PENDING_ONLY))
1025 		goto out_unlock;
1026 
1027 	new_base = get_target_base(base, timer->flags);
1028 
1029 	if (base != new_base) {
1030 		/*
1031 		 * We are trying to schedule the timer on the new base.
1032 		 * However we can't change timer's base while it is running,
1033 		 * otherwise del_timer_sync() can't detect that the timer's
1034 		 * handler yet has not finished. This also guarantees that the
1035 		 * timer is serialized wrt itself.
1036 		 */
1037 		if (likely(base->running_timer != timer)) {
1038 			/* See the comment in lock_timer_base() */
1039 			timer->flags |= TIMER_MIGRATING;
1040 
1041 			raw_spin_unlock(&base->lock);
1042 			base = new_base;
1043 			raw_spin_lock(&base->lock);
1044 			WRITE_ONCE(timer->flags,
1045 				   (timer->flags & ~TIMER_BASEMASK) | base->cpu);
1046 			forward_timer_base(base);
1047 		}
1048 	}
1049 
1050 	debug_timer_activate(timer);
1051 
1052 	timer->expires = expires;
1053 	/*
1054 	 * If 'idx' was calculated above and the base time did not advance
1055 	 * between calculating 'idx' and possibly switching the base, only
1056 	 * enqueue_timer() is required. Otherwise we need to (re)calculate
1057 	 * the wheel index via internal_add_timer().
1058 	 */
1059 	if (idx != UINT_MAX && clk == base->clk)
1060 		enqueue_timer(base, timer, idx, bucket_expiry);
1061 	else
1062 		internal_add_timer(base, timer);
1063 
1064 out_unlock:
1065 	raw_spin_unlock_irqrestore(&base->lock, flags);
1066 
1067 	return ret;
1068 }
1069 
1070 /**
1071  * mod_timer_pending - modify a pending timer's timeout
1072  * @timer: the pending timer to be modified
1073  * @expires: new timeout in jiffies
1074  *
1075  * mod_timer_pending() is the same for pending timers as mod_timer(),
1076  * but will not re-activate and modify already deleted timers.
1077  *
1078  * It is useful for unserialized use of timers.
1079  */
1080 int mod_timer_pending(struct timer_list *timer, unsigned long expires)
1081 {
1082 	return __mod_timer(timer, expires, MOD_TIMER_PENDING_ONLY);
1083 }
1084 EXPORT_SYMBOL(mod_timer_pending);
1085 
1086 /**
1087  * mod_timer - modify a timer's timeout
1088  * @timer: the timer to be modified
1089  * @expires: new timeout in jiffies
1090  *
1091  * mod_timer() is a more efficient way to update the expire field of an
1092  * active timer (if the timer is inactive it will be activated)
1093  *
1094  * mod_timer(timer, expires) is equivalent to:
1095  *
1096  *     del_timer(timer); timer->expires = expires; add_timer(timer);
1097  *
1098  * Note that if there are multiple unserialized concurrent users of the
1099  * same timer, then mod_timer() is the only safe way to modify the timeout,
1100  * since add_timer() cannot modify an already running timer.
1101  *
1102  * The function returns whether it has modified a pending timer or not.
1103  * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
1104  * active timer returns 1.)
1105  */
1106 int mod_timer(struct timer_list *timer, unsigned long expires)
1107 {
1108 	return __mod_timer(timer, expires, 0);
1109 }
1110 EXPORT_SYMBOL(mod_timer);
1111 
1112 /**
1113  * timer_reduce - Modify a timer's timeout if it would reduce the timeout
1114  * @timer:	The timer to be modified
1115  * @expires:	New timeout in jiffies
1116  *
1117  * timer_reduce() is very similar to mod_timer(), except that it will only
1118  * modify a running timer if that would reduce the expiration time (it will
1119  * start a timer that isn't running).
1120  */
1121 int timer_reduce(struct timer_list *timer, unsigned long expires)
1122 {
1123 	return __mod_timer(timer, expires, MOD_TIMER_REDUCE);
1124 }
1125 EXPORT_SYMBOL(timer_reduce);
1126 
1127 /**
1128  * add_timer - start a timer
1129  * @timer: the timer to be added
1130  *
1131  * The kernel will do a ->function(@timer) callback from the
1132  * timer interrupt at the ->expires point in the future. The
1133  * current time is 'jiffies'.
1134  *
1135  * The timer's ->expires, ->function fields must be set prior calling this
1136  * function.
1137  *
1138  * Timers with an ->expires field in the past will be executed in the next
1139  * timer tick.
1140  */
1141 void add_timer(struct timer_list *timer)
1142 {
1143 	BUG_ON(timer_pending(timer));
1144 	__mod_timer(timer, timer->expires, MOD_TIMER_NOTPENDING);
1145 }
1146 EXPORT_SYMBOL(add_timer);
1147 
1148 /**
1149  * add_timer_on - start a timer on a particular CPU
1150  * @timer: the timer to be added
1151  * @cpu: the CPU to start it on
1152  *
1153  * This is not very scalable on SMP. Double adds are not possible.
1154  */
1155 void add_timer_on(struct timer_list *timer, int cpu)
1156 {
1157 	struct timer_base *new_base, *base;
1158 	unsigned long flags;
1159 
1160 	BUG_ON(timer_pending(timer) || !timer->function);
1161 
1162 	new_base = get_timer_cpu_base(timer->flags, cpu);
1163 
1164 	/*
1165 	 * If @timer was on a different CPU, it should be migrated with the
1166 	 * old base locked to prevent other operations proceeding with the
1167 	 * wrong base locked.  See lock_timer_base().
1168 	 */
1169 	base = lock_timer_base(timer, &flags);
1170 	if (base != new_base) {
1171 		timer->flags |= TIMER_MIGRATING;
1172 
1173 		raw_spin_unlock(&base->lock);
1174 		base = new_base;
1175 		raw_spin_lock(&base->lock);
1176 		WRITE_ONCE(timer->flags,
1177 			   (timer->flags & ~TIMER_BASEMASK) | cpu);
1178 	}
1179 	forward_timer_base(base);
1180 
1181 	debug_timer_activate(timer);
1182 	internal_add_timer(base, timer);
1183 	raw_spin_unlock_irqrestore(&base->lock, flags);
1184 }
1185 EXPORT_SYMBOL_GPL(add_timer_on);
1186 
1187 /**
1188  * del_timer - deactivate a timer.
1189  * @timer: the timer to be deactivated
1190  *
1191  * del_timer() deactivates a timer - this works on both active and inactive
1192  * timers.
1193  *
1194  * The function returns whether it has deactivated a pending timer or not.
1195  * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
1196  * active timer returns 1.)
1197  */
1198 int del_timer(struct timer_list *timer)
1199 {
1200 	struct timer_base *base;
1201 	unsigned long flags;
1202 	int ret = 0;
1203 
1204 	debug_assert_init(timer);
1205 
1206 	if (timer_pending(timer)) {
1207 		base = lock_timer_base(timer, &flags);
1208 		ret = detach_if_pending(timer, base, true);
1209 		raw_spin_unlock_irqrestore(&base->lock, flags);
1210 	}
1211 
1212 	return ret;
1213 }
1214 EXPORT_SYMBOL(del_timer);
1215 
1216 /**
1217  * try_to_del_timer_sync - Try to deactivate a timer
1218  * @timer: timer to delete
1219  *
1220  * This function tries to deactivate a timer. Upon successful (ret >= 0)
1221  * exit the timer is not queued and the handler is not running on any CPU.
1222  */
1223 int try_to_del_timer_sync(struct timer_list *timer)
1224 {
1225 	struct timer_base *base;
1226 	unsigned long flags;
1227 	int ret = -1;
1228 
1229 	debug_assert_init(timer);
1230 
1231 	base = lock_timer_base(timer, &flags);
1232 
1233 	if (base->running_timer != timer)
1234 		ret = detach_if_pending(timer, base, true);
1235 
1236 	raw_spin_unlock_irqrestore(&base->lock, flags);
1237 
1238 	return ret;
1239 }
1240 EXPORT_SYMBOL(try_to_del_timer_sync);
1241 
1242 #ifdef CONFIG_PREEMPT_RT
1243 static __init void timer_base_init_expiry_lock(struct timer_base *base)
1244 {
1245 	spin_lock_init(&base->expiry_lock);
1246 }
1247 
1248 static inline void timer_base_lock_expiry(struct timer_base *base)
1249 {
1250 	spin_lock(&base->expiry_lock);
1251 }
1252 
1253 static inline void timer_base_unlock_expiry(struct timer_base *base)
1254 {
1255 	spin_unlock(&base->expiry_lock);
1256 }
1257 
1258 /*
1259  * The counterpart to del_timer_wait_running().
1260  *
1261  * If there is a waiter for base->expiry_lock, then it was waiting for the
1262  * timer callback to finish. Drop expiry_lock and reacquire it. That allows
1263  * the waiter to acquire the lock and make progress.
1264  */
1265 static void timer_sync_wait_running(struct timer_base *base)
1266 {
1267 	if (atomic_read(&base->timer_waiters)) {
1268 		raw_spin_unlock_irq(&base->lock);
1269 		spin_unlock(&base->expiry_lock);
1270 		spin_lock(&base->expiry_lock);
1271 		raw_spin_lock_irq(&base->lock);
1272 	}
1273 }
1274 
1275 /*
1276  * This function is called on PREEMPT_RT kernels when the fast path
1277  * deletion of a timer failed because the timer callback function was
1278  * running.
1279  *
1280  * This prevents priority inversion, if the softirq thread on a remote CPU
1281  * got preempted, and it prevents a life lock when the task which tries to
1282  * delete a timer preempted the softirq thread running the timer callback
1283  * function.
1284  */
1285 static void del_timer_wait_running(struct timer_list *timer)
1286 {
1287 	u32 tf;
1288 
1289 	tf = READ_ONCE(timer->flags);
1290 	if (!(tf & (TIMER_MIGRATING | TIMER_IRQSAFE))) {
1291 		struct timer_base *base = get_timer_base(tf);
1292 
1293 		/*
1294 		 * Mark the base as contended and grab the expiry lock,
1295 		 * which is held by the softirq across the timer
1296 		 * callback. Drop the lock immediately so the softirq can
1297 		 * expire the next timer. In theory the timer could already
1298 		 * be running again, but that's more than unlikely and just
1299 		 * causes another wait loop.
1300 		 */
1301 		atomic_inc(&base->timer_waiters);
1302 		spin_lock_bh(&base->expiry_lock);
1303 		atomic_dec(&base->timer_waiters);
1304 		spin_unlock_bh(&base->expiry_lock);
1305 	}
1306 }
1307 #else
1308 static inline void timer_base_init_expiry_lock(struct timer_base *base) { }
1309 static inline void timer_base_lock_expiry(struct timer_base *base) { }
1310 static inline void timer_base_unlock_expiry(struct timer_base *base) { }
1311 static inline void timer_sync_wait_running(struct timer_base *base) { }
1312 static inline void del_timer_wait_running(struct timer_list *timer) { }
1313 #endif
1314 
1315 #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT_RT)
1316 /**
1317  * del_timer_sync - deactivate a timer and wait for the handler to finish.
1318  * @timer: the timer to be deactivated
1319  *
1320  * This function only differs from del_timer() on SMP: besides deactivating
1321  * the timer it also makes sure the handler has finished executing on other
1322  * CPUs.
1323  *
1324  * Synchronization rules: Callers must prevent restarting of the timer,
1325  * otherwise this function is meaningless. It must not be called from
1326  * interrupt contexts unless the timer is an irqsafe one. The caller must
1327  * not hold locks which would prevent completion of the timer's
1328  * handler. The timer's handler must not call add_timer_on(). Upon exit the
1329  * timer is not queued and the handler is not running on any CPU.
1330  *
1331  * Note: For !irqsafe timers, you must not hold locks that are held in
1332  *   interrupt context while calling this function. Even if the lock has
1333  *   nothing to do with the timer in question.  Here's why::
1334  *
1335  *    CPU0                             CPU1
1336  *    ----                             ----
1337  *                                     <SOFTIRQ>
1338  *                                       call_timer_fn();
1339  *                                       base->running_timer = mytimer;
1340  *    spin_lock_irq(somelock);
1341  *                                     <IRQ>
1342  *                                        spin_lock(somelock);
1343  *    del_timer_sync(mytimer);
1344  *    while (base->running_timer == mytimer);
1345  *
1346  * Now del_timer_sync() will never return and never release somelock.
1347  * The interrupt on the other CPU is waiting to grab somelock but
1348  * it has interrupted the softirq that CPU0 is waiting to finish.
1349  *
1350  * The function returns whether it has deactivated a pending timer or not.
1351  */
1352 int del_timer_sync(struct timer_list *timer)
1353 {
1354 	int ret;
1355 
1356 #ifdef CONFIG_LOCKDEP
1357 	unsigned long flags;
1358 
1359 	/*
1360 	 * If lockdep gives a backtrace here, please reference
1361 	 * the synchronization rules above.
1362 	 */
1363 	local_irq_save(flags);
1364 	lock_map_acquire(&timer->lockdep_map);
1365 	lock_map_release(&timer->lockdep_map);
1366 	local_irq_restore(flags);
1367 #endif
1368 	/*
1369 	 * don't use it in hardirq context, because it
1370 	 * could lead to deadlock.
1371 	 */
1372 	WARN_ON(in_irq() && !(timer->flags & TIMER_IRQSAFE));
1373 
1374 	/*
1375 	 * Must be able to sleep on PREEMPT_RT because of the slowpath in
1376 	 * del_timer_wait_running().
1377 	 */
1378 	if (IS_ENABLED(CONFIG_PREEMPT_RT) && !(timer->flags & TIMER_IRQSAFE))
1379 		lockdep_assert_preemption_enabled();
1380 
1381 	do {
1382 		ret = try_to_del_timer_sync(timer);
1383 
1384 		if (unlikely(ret < 0)) {
1385 			del_timer_wait_running(timer);
1386 			cpu_relax();
1387 		}
1388 	} while (ret < 0);
1389 
1390 	return ret;
1391 }
1392 EXPORT_SYMBOL(del_timer_sync);
1393 #endif
1394 
1395 static void call_timer_fn(struct timer_list *timer,
1396 			  void (*fn)(struct timer_list *),
1397 			  unsigned long baseclk)
1398 {
1399 	int count = preempt_count();
1400 
1401 #ifdef CONFIG_LOCKDEP
1402 	/*
1403 	 * It is permissible to free the timer from inside the
1404 	 * function that is called from it, this we need to take into
1405 	 * account for lockdep too. To avoid bogus "held lock freed"
1406 	 * warnings as well as problems when looking into
1407 	 * timer->lockdep_map, make a copy and use that here.
1408 	 */
1409 	struct lockdep_map lockdep_map;
1410 
1411 	lockdep_copy_map(&lockdep_map, &timer->lockdep_map);
1412 #endif
1413 	/*
1414 	 * Couple the lock chain with the lock chain at
1415 	 * del_timer_sync() by acquiring the lock_map around the fn()
1416 	 * call here and in del_timer_sync().
1417 	 */
1418 	lock_map_acquire(&lockdep_map);
1419 
1420 	trace_timer_expire_entry(timer, baseclk);
1421 	fn(timer);
1422 	trace_timer_expire_exit(timer);
1423 
1424 	lock_map_release(&lockdep_map);
1425 
1426 	if (count != preempt_count()) {
1427 		WARN_ONCE(1, "timer: %pS preempt leak: %08x -> %08x\n",
1428 			  fn, count, preempt_count());
1429 		/*
1430 		 * Restore the preempt count. That gives us a decent
1431 		 * chance to survive and extract information. If the
1432 		 * callback kept a lock held, bad luck, but not worse
1433 		 * than the BUG() we had.
1434 		 */
1435 		preempt_count_set(count);
1436 	}
1437 }
1438 
1439 static void expire_timers(struct timer_base *base, struct hlist_head *head)
1440 {
1441 	/*
1442 	 * This value is required only for tracing. base->clk was
1443 	 * incremented directly before expire_timers was called. But expiry
1444 	 * is related to the old base->clk value.
1445 	 */
1446 	unsigned long baseclk = base->clk - 1;
1447 
1448 	while (!hlist_empty(head)) {
1449 		struct timer_list *timer;
1450 		void (*fn)(struct timer_list *);
1451 
1452 		timer = hlist_entry(head->first, struct timer_list, entry);
1453 
1454 		base->running_timer = timer;
1455 		detach_timer(timer, true);
1456 
1457 		fn = timer->function;
1458 
1459 		if (timer->flags & TIMER_IRQSAFE) {
1460 			raw_spin_unlock(&base->lock);
1461 			call_timer_fn(timer, fn, baseclk);
1462 			raw_spin_lock(&base->lock);
1463 			base->running_timer = NULL;
1464 		} else {
1465 			raw_spin_unlock_irq(&base->lock);
1466 			call_timer_fn(timer, fn, baseclk);
1467 			raw_spin_lock_irq(&base->lock);
1468 			base->running_timer = NULL;
1469 			timer_sync_wait_running(base);
1470 		}
1471 	}
1472 }
1473 
1474 static int collect_expired_timers(struct timer_base *base,
1475 				  struct hlist_head *heads)
1476 {
1477 	unsigned long clk = base->clk = base->next_expiry;
1478 	struct hlist_head *vec;
1479 	int i, levels = 0;
1480 	unsigned int idx;
1481 
1482 	for (i = 0; i < LVL_DEPTH; i++) {
1483 		idx = (clk & LVL_MASK) + i * LVL_SIZE;
1484 
1485 		if (__test_and_clear_bit(idx, base->pending_map)) {
1486 			vec = base->vectors + idx;
1487 			hlist_move_list(vec, heads++);
1488 			levels++;
1489 		}
1490 		/* Is it time to look at the next level? */
1491 		if (clk & LVL_CLK_MASK)
1492 			break;
1493 		/* Shift clock for the next level granularity */
1494 		clk >>= LVL_CLK_SHIFT;
1495 	}
1496 	return levels;
1497 }
1498 
1499 /*
1500  * Find the next pending bucket of a level. Search from level start (@offset)
1501  * + @clk upwards and if nothing there, search from start of the level
1502  * (@offset) up to @offset + clk.
1503  */
1504 static int next_pending_bucket(struct timer_base *base, unsigned offset,
1505 			       unsigned clk)
1506 {
1507 	unsigned pos, start = offset + clk;
1508 	unsigned end = offset + LVL_SIZE;
1509 
1510 	pos = find_next_bit(base->pending_map, end, start);
1511 	if (pos < end)
1512 		return pos - start;
1513 
1514 	pos = find_next_bit(base->pending_map, start, offset);
1515 	return pos < start ? pos + LVL_SIZE - start : -1;
1516 }
1517 
1518 /*
1519  * Search the first expiring timer in the various clock levels. Caller must
1520  * hold base->lock.
1521  */
1522 static unsigned long __next_timer_interrupt(struct timer_base *base)
1523 {
1524 	unsigned long clk, next, adj;
1525 	unsigned lvl, offset = 0;
1526 
1527 	next = base->clk + NEXT_TIMER_MAX_DELTA;
1528 	clk = base->clk;
1529 	for (lvl = 0; lvl < LVL_DEPTH; lvl++, offset += LVL_SIZE) {
1530 		int pos = next_pending_bucket(base, offset, clk & LVL_MASK);
1531 		unsigned long lvl_clk = clk & LVL_CLK_MASK;
1532 
1533 		if (pos >= 0) {
1534 			unsigned long tmp = clk + (unsigned long) pos;
1535 
1536 			tmp <<= LVL_SHIFT(lvl);
1537 			if (time_before(tmp, next))
1538 				next = tmp;
1539 
1540 			/*
1541 			 * If the next expiration happens before we reach
1542 			 * the next level, no need to check further.
1543 			 */
1544 			if (pos <= ((LVL_CLK_DIV - lvl_clk) & LVL_CLK_MASK))
1545 				break;
1546 		}
1547 		/*
1548 		 * Clock for the next level. If the current level clock lower
1549 		 * bits are zero, we look at the next level as is. If not we
1550 		 * need to advance it by one because that's going to be the
1551 		 * next expiring bucket in that level. base->clk is the next
1552 		 * expiring jiffie. So in case of:
1553 		 *
1554 		 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1555 		 *  0    0    0    0    0    0
1556 		 *
1557 		 * we have to look at all levels @index 0. With
1558 		 *
1559 		 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1560 		 *  0    0    0    0    0    2
1561 		 *
1562 		 * LVL0 has the next expiring bucket @index 2. The upper
1563 		 * levels have the next expiring bucket @index 1.
1564 		 *
1565 		 * In case that the propagation wraps the next level the same
1566 		 * rules apply:
1567 		 *
1568 		 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1569 		 *  0    0    0    0    F    2
1570 		 *
1571 		 * So after looking at LVL0 we get:
1572 		 *
1573 		 * LVL5 LVL4 LVL3 LVL2 LVL1
1574 		 *  0    0    0    1    0
1575 		 *
1576 		 * So no propagation from LVL1 to LVL2 because that happened
1577 		 * with the add already, but then we need to propagate further
1578 		 * from LVL2 to LVL3.
1579 		 *
1580 		 * So the simple check whether the lower bits of the current
1581 		 * level are 0 or not is sufficient for all cases.
1582 		 */
1583 		adj = lvl_clk ? 1 : 0;
1584 		clk >>= LVL_CLK_SHIFT;
1585 		clk += adj;
1586 	}
1587 
1588 	base->next_expiry_recalc = false;
1589 	base->timers_pending = !(next == base->clk + NEXT_TIMER_MAX_DELTA);
1590 
1591 	return next;
1592 }
1593 
1594 #ifdef CONFIG_NO_HZ_COMMON
1595 /*
1596  * Check, if the next hrtimer event is before the next timer wheel
1597  * event:
1598  */
1599 static u64 cmp_next_hrtimer_event(u64 basem, u64 expires)
1600 {
1601 	u64 nextevt = hrtimer_get_next_event();
1602 
1603 	/*
1604 	 * If high resolution timers are enabled
1605 	 * hrtimer_get_next_event() returns KTIME_MAX.
1606 	 */
1607 	if (expires <= nextevt)
1608 		return expires;
1609 
1610 	/*
1611 	 * If the next timer is already expired, return the tick base
1612 	 * time so the tick is fired immediately.
1613 	 */
1614 	if (nextevt <= basem)
1615 		return basem;
1616 
1617 	/*
1618 	 * Round up to the next jiffie. High resolution timers are
1619 	 * off, so the hrtimers are expired in the tick and we need to
1620 	 * make sure that this tick really expires the timer to avoid
1621 	 * a ping pong of the nohz stop code.
1622 	 *
1623 	 * Use DIV_ROUND_UP_ULL to prevent gcc calling __divdi3
1624 	 */
1625 	return DIV_ROUND_UP_ULL(nextevt, TICK_NSEC) * TICK_NSEC;
1626 }
1627 
1628 /**
1629  * get_next_timer_interrupt - return the time (clock mono) of the next timer
1630  * @basej:	base time jiffies
1631  * @basem:	base time clock monotonic
1632  *
1633  * Returns the tick aligned clock monotonic time of the next pending
1634  * timer or KTIME_MAX if no timer is pending.
1635  */
1636 u64 get_next_timer_interrupt(unsigned long basej, u64 basem)
1637 {
1638 	struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1639 	u64 expires = KTIME_MAX;
1640 	unsigned long nextevt;
1641 
1642 	/*
1643 	 * Pretend that there is no timer pending if the cpu is offline.
1644 	 * Possible pending timers will be migrated later to an active cpu.
1645 	 */
1646 	if (cpu_is_offline(smp_processor_id()))
1647 		return expires;
1648 
1649 	raw_spin_lock(&base->lock);
1650 	if (base->next_expiry_recalc)
1651 		base->next_expiry = __next_timer_interrupt(base);
1652 	nextevt = base->next_expiry;
1653 
1654 	/*
1655 	 * We have a fresh next event. Check whether we can forward the
1656 	 * base. We can only do that when @basej is past base->clk
1657 	 * otherwise we might rewind base->clk.
1658 	 */
1659 	if (time_after(basej, base->clk)) {
1660 		if (time_after(nextevt, basej))
1661 			base->clk = basej;
1662 		else if (time_after(nextevt, base->clk))
1663 			base->clk = nextevt;
1664 	}
1665 
1666 	if (time_before_eq(nextevt, basej)) {
1667 		expires = basem;
1668 		base->is_idle = false;
1669 	} else {
1670 		if (base->timers_pending)
1671 			expires = basem + (u64)(nextevt - basej) * TICK_NSEC;
1672 		/*
1673 		 * If we expect to sleep more than a tick, mark the base idle.
1674 		 * Also the tick is stopped so any added timer must forward
1675 		 * the base clk itself to keep granularity small. This idle
1676 		 * logic is only maintained for the BASE_STD base, deferrable
1677 		 * timers may still see large granularity skew (by design).
1678 		 */
1679 		if ((expires - basem) > TICK_NSEC)
1680 			base->is_idle = true;
1681 	}
1682 	raw_spin_unlock(&base->lock);
1683 
1684 	return cmp_next_hrtimer_event(basem, expires);
1685 }
1686 
1687 /**
1688  * timer_clear_idle - Clear the idle state of the timer base
1689  *
1690  * Called with interrupts disabled
1691  */
1692 void timer_clear_idle(void)
1693 {
1694 	struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1695 
1696 	/*
1697 	 * We do this unlocked. The worst outcome is a remote enqueue sending
1698 	 * a pointless IPI, but taking the lock would just make the window for
1699 	 * sending the IPI a few instructions smaller for the cost of taking
1700 	 * the lock in the exit from idle path.
1701 	 */
1702 	base->is_idle = false;
1703 }
1704 #endif
1705 
1706 /**
1707  * __run_timers - run all expired timers (if any) on this CPU.
1708  * @base: the timer vector to be processed.
1709  */
1710 static inline void __run_timers(struct timer_base *base)
1711 {
1712 	struct hlist_head heads[LVL_DEPTH];
1713 	int levels;
1714 
1715 	if (time_before(jiffies, base->next_expiry))
1716 		return;
1717 
1718 	timer_base_lock_expiry(base);
1719 	raw_spin_lock_irq(&base->lock);
1720 
1721 	while (time_after_eq(jiffies, base->clk) &&
1722 	       time_after_eq(jiffies, base->next_expiry)) {
1723 		levels = collect_expired_timers(base, heads);
1724 		/*
1725 		 * The only possible reason for not finding any expired
1726 		 * timer at this clk is that all matching timers have been
1727 		 * dequeued.
1728 		 */
1729 		WARN_ON_ONCE(!levels && !base->next_expiry_recalc);
1730 		base->clk++;
1731 		base->next_expiry = __next_timer_interrupt(base);
1732 
1733 		while (levels--)
1734 			expire_timers(base, heads + levels);
1735 	}
1736 	raw_spin_unlock_irq(&base->lock);
1737 	timer_base_unlock_expiry(base);
1738 }
1739 
1740 /*
1741  * This function runs timers and the timer-tq in bottom half context.
1742  */
1743 static __latent_entropy void run_timer_softirq(struct softirq_action *h)
1744 {
1745 	struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1746 
1747 	__run_timers(base);
1748 	if (IS_ENABLED(CONFIG_NO_HZ_COMMON))
1749 		__run_timers(this_cpu_ptr(&timer_bases[BASE_DEF]));
1750 }
1751 
1752 /*
1753  * Called by the local, per-CPU timer interrupt on SMP.
1754  */
1755 static void run_local_timers(void)
1756 {
1757 	struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1758 
1759 	hrtimer_run_queues();
1760 	/* Raise the softirq only if required. */
1761 	if (time_before(jiffies, base->next_expiry)) {
1762 		if (!IS_ENABLED(CONFIG_NO_HZ_COMMON))
1763 			return;
1764 		/* CPU is awake, so check the deferrable base. */
1765 		base++;
1766 		if (time_before(jiffies, base->next_expiry))
1767 			return;
1768 	}
1769 	raise_softirq(TIMER_SOFTIRQ);
1770 }
1771 
1772 /*
1773  * Called from the timer interrupt handler to charge one tick to the current
1774  * process.  user_tick is 1 if the tick is user time, 0 for system.
1775  */
1776 void update_process_times(int user_tick)
1777 {
1778 	struct task_struct *p = current;
1779 
1780 	PRANDOM_ADD_NOISE(jiffies, user_tick, p, 0);
1781 
1782 	/* Note: this timer irq context must be accounted for as well. */
1783 	account_process_tick(p, user_tick);
1784 	run_local_timers();
1785 	rcu_sched_clock_irq(user_tick);
1786 #ifdef CONFIG_IRQ_WORK
1787 	if (in_irq())
1788 		irq_work_tick();
1789 #endif
1790 	scheduler_tick();
1791 	if (IS_ENABLED(CONFIG_POSIX_TIMERS))
1792 		run_posix_cpu_timers();
1793 }
1794 
1795 /*
1796  * Since schedule_timeout()'s timer is defined on the stack, it must store
1797  * the target task on the stack as well.
1798  */
1799 struct process_timer {
1800 	struct timer_list timer;
1801 	struct task_struct *task;
1802 };
1803 
1804 static void process_timeout(struct timer_list *t)
1805 {
1806 	struct process_timer *timeout = from_timer(timeout, t, timer);
1807 
1808 	wake_up_process(timeout->task);
1809 }
1810 
1811 /**
1812  * schedule_timeout - sleep until timeout
1813  * @timeout: timeout value in jiffies
1814  *
1815  * Make the current task sleep until @timeout jiffies have elapsed.
1816  * The function behavior depends on the current task state
1817  * (see also set_current_state() description):
1818  *
1819  * %TASK_RUNNING - the scheduler is called, but the task does not sleep
1820  * at all. That happens because sched_submit_work() does nothing for
1821  * tasks in %TASK_RUNNING state.
1822  *
1823  * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1824  * pass before the routine returns unless the current task is explicitly
1825  * woken up, (e.g. by wake_up_process()).
1826  *
1827  * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1828  * delivered to the current task or the current task is explicitly woken
1829  * up.
1830  *
1831  * The current task state is guaranteed to be %TASK_RUNNING when this
1832  * routine returns.
1833  *
1834  * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1835  * the CPU away without a bound on the timeout. In this case the return
1836  * value will be %MAX_SCHEDULE_TIMEOUT.
1837  *
1838  * Returns 0 when the timer has expired otherwise the remaining time in
1839  * jiffies will be returned. In all cases the return value is guaranteed
1840  * to be non-negative.
1841  */
1842 signed long __sched schedule_timeout(signed long timeout)
1843 {
1844 	struct process_timer timer;
1845 	unsigned long expire;
1846 
1847 	switch (timeout)
1848 	{
1849 	case MAX_SCHEDULE_TIMEOUT:
1850 		/*
1851 		 * These two special cases are useful to be comfortable
1852 		 * in the caller. Nothing more. We could take
1853 		 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1854 		 * but I' d like to return a valid offset (>=0) to allow
1855 		 * the caller to do everything it want with the retval.
1856 		 */
1857 		schedule();
1858 		goto out;
1859 	default:
1860 		/*
1861 		 * Another bit of PARANOID. Note that the retval will be
1862 		 * 0 since no piece of kernel is supposed to do a check
1863 		 * for a negative retval of schedule_timeout() (since it
1864 		 * should never happens anyway). You just have the printk()
1865 		 * that will tell you if something is gone wrong and where.
1866 		 */
1867 		if (timeout < 0) {
1868 			printk(KERN_ERR "schedule_timeout: wrong timeout "
1869 				"value %lx\n", timeout);
1870 			dump_stack();
1871 			__set_current_state(TASK_RUNNING);
1872 			goto out;
1873 		}
1874 	}
1875 
1876 	expire = timeout + jiffies;
1877 
1878 	timer.task = current;
1879 	timer_setup_on_stack(&timer.timer, process_timeout, 0);
1880 	__mod_timer(&timer.timer, expire, MOD_TIMER_NOTPENDING);
1881 	schedule();
1882 	del_singleshot_timer_sync(&timer.timer);
1883 
1884 	/* Remove the timer from the object tracker */
1885 	destroy_timer_on_stack(&timer.timer);
1886 
1887 	timeout = expire - jiffies;
1888 
1889  out:
1890 	return timeout < 0 ? 0 : timeout;
1891 }
1892 EXPORT_SYMBOL(schedule_timeout);
1893 
1894 /*
1895  * We can use __set_current_state() here because schedule_timeout() calls
1896  * schedule() unconditionally.
1897  */
1898 signed long __sched schedule_timeout_interruptible(signed long timeout)
1899 {
1900 	__set_current_state(TASK_INTERRUPTIBLE);
1901 	return schedule_timeout(timeout);
1902 }
1903 EXPORT_SYMBOL(schedule_timeout_interruptible);
1904 
1905 signed long __sched schedule_timeout_killable(signed long timeout)
1906 {
1907 	__set_current_state(TASK_KILLABLE);
1908 	return schedule_timeout(timeout);
1909 }
1910 EXPORT_SYMBOL(schedule_timeout_killable);
1911 
1912 signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1913 {
1914 	__set_current_state(TASK_UNINTERRUPTIBLE);
1915 	return schedule_timeout(timeout);
1916 }
1917 EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1918 
1919 /*
1920  * Like schedule_timeout_uninterruptible(), except this task will not contribute
1921  * to load average.
1922  */
1923 signed long __sched schedule_timeout_idle(signed long timeout)
1924 {
1925 	__set_current_state(TASK_IDLE);
1926 	return schedule_timeout(timeout);
1927 }
1928 EXPORT_SYMBOL(schedule_timeout_idle);
1929 
1930 #ifdef CONFIG_HOTPLUG_CPU
1931 static void migrate_timer_list(struct timer_base *new_base, struct hlist_head *head)
1932 {
1933 	struct timer_list *timer;
1934 	int cpu = new_base->cpu;
1935 
1936 	while (!hlist_empty(head)) {
1937 		timer = hlist_entry(head->first, struct timer_list, entry);
1938 		detach_timer(timer, false);
1939 		timer->flags = (timer->flags & ~TIMER_BASEMASK) | cpu;
1940 		internal_add_timer(new_base, timer);
1941 	}
1942 }
1943 
1944 int timers_prepare_cpu(unsigned int cpu)
1945 {
1946 	struct timer_base *base;
1947 	int b;
1948 
1949 	for (b = 0; b < NR_BASES; b++) {
1950 		base = per_cpu_ptr(&timer_bases[b], cpu);
1951 		base->clk = jiffies;
1952 		base->next_expiry = base->clk + NEXT_TIMER_MAX_DELTA;
1953 		base->timers_pending = false;
1954 		base->is_idle = false;
1955 	}
1956 	return 0;
1957 }
1958 
1959 int timers_dead_cpu(unsigned int cpu)
1960 {
1961 	struct timer_base *old_base;
1962 	struct timer_base *new_base;
1963 	int b, i;
1964 
1965 	BUG_ON(cpu_online(cpu));
1966 
1967 	for (b = 0; b < NR_BASES; b++) {
1968 		old_base = per_cpu_ptr(&timer_bases[b], cpu);
1969 		new_base = get_cpu_ptr(&timer_bases[b]);
1970 		/*
1971 		 * The caller is globally serialized and nobody else
1972 		 * takes two locks at once, deadlock is not possible.
1973 		 */
1974 		raw_spin_lock_irq(&new_base->lock);
1975 		raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1976 
1977 		/*
1978 		 * The current CPUs base clock might be stale. Update it
1979 		 * before moving the timers over.
1980 		 */
1981 		forward_timer_base(new_base);
1982 
1983 		BUG_ON(old_base->running_timer);
1984 
1985 		for (i = 0; i < WHEEL_SIZE; i++)
1986 			migrate_timer_list(new_base, old_base->vectors + i);
1987 
1988 		raw_spin_unlock(&old_base->lock);
1989 		raw_spin_unlock_irq(&new_base->lock);
1990 		put_cpu_ptr(&timer_bases);
1991 	}
1992 	return 0;
1993 }
1994 
1995 #endif /* CONFIG_HOTPLUG_CPU */
1996 
1997 static void __init init_timer_cpu(int cpu)
1998 {
1999 	struct timer_base *base;
2000 	int i;
2001 
2002 	for (i = 0; i < NR_BASES; i++) {
2003 		base = per_cpu_ptr(&timer_bases[i], cpu);
2004 		base->cpu = cpu;
2005 		raw_spin_lock_init(&base->lock);
2006 		base->clk = jiffies;
2007 		base->next_expiry = base->clk + NEXT_TIMER_MAX_DELTA;
2008 		timer_base_init_expiry_lock(base);
2009 	}
2010 }
2011 
2012 static void __init init_timer_cpus(void)
2013 {
2014 	int cpu;
2015 
2016 	for_each_possible_cpu(cpu)
2017 		init_timer_cpu(cpu);
2018 }
2019 
2020 void __init init_timers(void)
2021 {
2022 	init_timer_cpus();
2023 	posix_cputimers_init_work();
2024 	open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
2025 }
2026 
2027 /**
2028  * msleep - sleep safely even with waitqueue interruptions
2029  * @msecs: Time in milliseconds to sleep for
2030  */
2031 void msleep(unsigned int msecs)
2032 {
2033 	unsigned long timeout = msecs_to_jiffies(msecs) + 1;
2034 
2035 	while (timeout)
2036 		timeout = schedule_timeout_uninterruptible(timeout);
2037 }
2038 
2039 EXPORT_SYMBOL(msleep);
2040 
2041 /**
2042  * msleep_interruptible - sleep waiting for signals
2043  * @msecs: Time in milliseconds to sleep for
2044  */
2045 unsigned long msleep_interruptible(unsigned int msecs)
2046 {
2047 	unsigned long timeout = msecs_to_jiffies(msecs) + 1;
2048 
2049 	while (timeout && !signal_pending(current))
2050 		timeout = schedule_timeout_interruptible(timeout);
2051 	return jiffies_to_msecs(timeout);
2052 }
2053 
2054 EXPORT_SYMBOL(msleep_interruptible);
2055 
2056 /**
2057  * usleep_range - Sleep for an approximate time
2058  * @min: Minimum time in usecs to sleep
2059  * @max: Maximum time in usecs to sleep
2060  *
2061  * In non-atomic context where the exact wakeup time is flexible, use
2062  * usleep_range() instead of udelay().  The sleep improves responsiveness
2063  * by avoiding the CPU-hogging busy-wait of udelay(), and the range reduces
2064  * power usage by allowing hrtimers to take advantage of an already-
2065  * scheduled interrupt instead of scheduling a new one just for this sleep.
2066  */
2067 void __sched usleep_range(unsigned long min, unsigned long max)
2068 {
2069 	ktime_t exp = ktime_add_us(ktime_get(), min);
2070 	u64 delta = (u64)(max - min) * NSEC_PER_USEC;
2071 
2072 	for (;;) {
2073 		__set_current_state(TASK_UNINTERRUPTIBLE);
2074 		/* Do not return before the requested sleep time has elapsed */
2075 		if (!schedule_hrtimeout_range(&exp, delta, HRTIMER_MODE_ABS))
2076 			break;
2077 	}
2078 }
2079 EXPORT_SYMBOL(usleep_range);
2080