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 #include <linux/sysctl.h>
48 
49 #include <linux/uaccess.h>
50 #include <asm/unistd.h>
51 #include <asm/div64.h>
52 #include <asm/timex.h>
53 #include <asm/io.h>
54 
55 #include "tick-internal.h"
56 
57 #define CREATE_TRACE_POINTS
58 #include <trace/events/timer.h>
59 
60 __visible u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
61 
62 EXPORT_SYMBOL(jiffies_64);
63 
64 /*
65  * The timer wheel has LVL_DEPTH array levels. Each level provides an array of
66  * LVL_SIZE buckets. Each level is driven by its own clock and therefor each
67  * level has a different granularity.
68  *
69  * The level granularity is:		LVL_CLK_DIV ^ lvl
70  * The level clock frequency is:	HZ / (LVL_CLK_DIV ^ level)
71  *
72  * The array level of a newly armed timer depends on the relative expiry
73  * time. The farther the expiry time is away the higher the array level and
74  * therefor the granularity becomes.
75  *
76  * Contrary to the original timer wheel implementation, which aims for 'exact'
77  * expiry of the timers, this implementation removes the need for recascading
78  * the timers into the lower array levels. The previous 'classic' timer wheel
79  * implementation of the kernel already violated the 'exact' expiry by adding
80  * slack to the expiry time to provide batched expiration. The granularity
81  * levels provide implicit batching.
82  *
83  * This is an optimization of the original timer wheel implementation for the
84  * majority of the timer wheel use cases: timeouts. The vast majority of
85  * timeout timers (networking, disk I/O ...) are canceled before expiry. If
86  * the timeout expires it indicates that normal operation is disturbed, so it
87  * does not matter much whether the timeout comes with a slight delay.
88  *
89  * The only exception to this are networking timers with a small expiry
90  * time. They rely on the granularity. Those fit into the first wheel level,
91  * which has HZ granularity.
92  *
93  * We don't have cascading anymore. timers with a expiry time above the
94  * capacity of the last wheel level are force expired at the maximum timeout
95  * value of the last wheel level. From data sampling we know that the maximum
96  * value observed is 5 days (network connection tracking), so this should not
97  * be an issue.
98  *
99  * The currently chosen array constants values are a good compromise between
100  * array size and granularity.
101  *
102  * This results in the following granularity and range levels:
103  *
104  * HZ 1000 steps
105  * Level Offset  Granularity            Range
106  *  0      0         1 ms                0 ms -         63 ms
107  *  1     64         8 ms               64 ms -        511 ms
108  *  2    128        64 ms              512 ms -       4095 ms (512ms - ~4s)
109  *  3    192       512 ms             4096 ms -      32767 ms (~4s - ~32s)
110  *  4    256      4096 ms (~4s)      32768 ms -     262143 ms (~32s - ~4m)
111  *  5    320     32768 ms (~32s)    262144 ms -    2097151 ms (~4m - ~34m)
112  *  6    384    262144 ms (~4m)    2097152 ms -   16777215 ms (~34m - ~4h)
113  *  7    448   2097152 ms (~34m)  16777216 ms -  134217727 ms (~4h - ~1d)
114  *  8    512  16777216 ms (~4h)  134217728 ms - 1073741822 ms (~1d - ~12d)
115  *
116  * HZ  300
117  * Level Offset  Granularity            Range
118  *  0	   0         3 ms                0 ms -        210 ms
119  *  1	  64        26 ms              213 ms -       1703 ms (213ms - ~1s)
120  *  2	 128       213 ms             1706 ms -      13650 ms (~1s - ~13s)
121  *  3	 192      1706 ms (~1s)      13653 ms -     109223 ms (~13s - ~1m)
122  *  4	 256     13653 ms (~13s)    109226 ms -     873810 ms (~1m - ~14m)
123  *  5	 320    109226 ms (~1m)     873813 ms -    6990503 ms (~14m - ~1h)
124  *  6	 384    873813 ms (~14m)   6990506 ms -   55924050 ms (~1h - ~15h)
125  *  7	 448   6990506 ms (~1h)   55924053 ms -  447392423 ms (~15h - ~5d)
126  *  8    512  55924053 ms (~15h) 447392426 ms - 3579139406 ms (~5d - ~41d)
127  *
128  * HZ  250
129  * Level Offset  Granularity            Range
130  *  0	   0         4 ms                0 ms -        255 ms
131  *  1	  64        32 ms              256 ms -       2047 ms (256ms - ~2s)
132  *  2	 128       256 ms             2048 ms -      16383 ms (~2s - ~16s)
133  *  3	 192      2048 ms (~2s)      16384 ms -     131071 ms (~16s - ~2m)
134  *  4	 256     16384 ms (~16s)    131072 ms -    1048575 ms (~2m - ~17m)
135  *  5	 320    131072 ms (~2m)    1048576 ms -    8388607 ms (~17m - ~2h)
136  *  6	 384   1048576 ms (~17m)   8388608 ms -   67108863 ms (~2h - ~18h)
137  *  7	 448   8388608 ms (~2h)   67108864 ms -  536870911 ms (~18h - ~6d)
138  *  8    512  67108864 ms (~18h) 536870912 ms - 4294967288 ms (~6d - ~49d)
139  *
140  * HZ  100
141  * Level Offset  Granularity            Range
142  *  0	   0         10 ms               0 ms -        630 ms
143  *  1	  64         80 ms             640 ms -       5110 ms (640ms - ~5s)
144  *  2	 128        640 ms            5120 ms -      40950 ms (~5s - ~40s)
145  *  3	 192       5120 ms (~5s)     40960 ms -     327670 ms (~40s - ~5m)
146  *  4	 256      40960 ms (~40s)   327680 ms -    2621430 ms (~5m - ~43m)
147  *  5	 320     327680 ms (~5m)   2621440 ms -   20971510 ms (~43m - ~5h)
148  *  6	 384    2621440 ms (~43m) 20971520 ms -  167772150 ms (~5h - ~1d)
149  *  7	 448   20971520 ms (~5h) 167772160 ms - 1342177270 ms (~1d - ~15d)
150  */
151 
152 /* Clock divisor for the next level */
153 #define LVL_CLK_SHIFT	3
154 #define LVL_CLK_DIV	(1UL << LVL_CLK_SHIFT)
155 #define LVL_CLK_MASK	(LVL_CLK_DIV - 1)
156 #define LVL_SHIFT(n)	((n) * LVL_CLK_SHIFT)
157 #define LVL_GRAN(n)	(1UL << LVL_SHIFT(n))
158 
159 /*
160  * The time start value for each level to select the bucket at enqueue
161  * time. We start from the last possible delta of the previous level
162  * so that we can later add an extra LVL_GRAN(n) to n (see calc_index()).
163  */
164 #define LVL_START(n)	((LVL_SIZE - 1) << (((n) - 1) * LVL_CLK_SHIFT))
165 
166 /* Size of each clock level */
167 #define LVL_BITS	6
168 #define LVL_SIZE	(1UL << LVL_BITS)
169 #define LVL_MASK	(LVL_SIZE - 1)
170 #define LVL_OFFS(n)	((n) * LVL_SIZE)
171 
172 /* Level depth */
173 #if HZ > 100
174 # define LVL_DEPTH	9
175 # else
176 # define LVL_DEPTH	8
177 #endif
178 
179 /* The cutoff (max. capacity of the wheel) */
180 #define WHEEL_TIMEOUT_CUTOFF	(LVL_START(LVL_DEPTH))
181 #define WHEEL_TIMEOUT_MAX	(WHEEL_TIMEOUT_CUTOFF - LVL_GRAN(LVL_DEPTH - 1))
182 
183 /*
184  * The resulting wheel size. If NOHZ is configured we allocate two
185  * wheels so we have a separate storage for the deferrable timers.
186  */
187 #define WHEEL_SIZE	(LVL_SIZE * LVL_DEPTH)
188 
189 #ifdef CONFIG_NO_HZ_COMMON
190 # define NR_BASES	2
191 # define BASE_STD	0
192 # define BASE_DEF	1
193 #else
194 # define NR_BASES	1
195 # define BASE_STD	0
196 # define BASE_DEF	0
197 #endif
198 
199 struct timer_base {
200 	raw_spinlock_t		lock;
201 	struct timer_list	*running_timer;
202 #ifdef CONFIG_PREEMPT_RT
203 	spinlock_t		expiry_lock;
204 	atomic_t		timer_waiters;
205 #endif
206 	unsigned long		clk;
207 	unsigned long		next_expiry;
208 	unsigned int		cpu;
209 	bool			next_expiry_recalc;
210 	bool			is_idle;
211 	bool			timers_pending;
212 	DECLARE_BITMAP(pending_map, WHEEL_SIZE);
213 	struct hlist_head	vectors[WHEEL_SIZE];
214 } ____cacheline_aligned;
215 
216 static DEFINE_PER_CPU(struct timer_base, timer_bases[NR_BASES]);
217 
218 #ifdef CONFIG_NO_HZ_COMMON
219 
220 static DEFINE_STATIC_KEY_FALSE(timers_nohz_active);
221 static DEFINE_MUTEX(timer_keys_mutex);
222 
223 static void timer_update_keys(struct work_struct *work);
224 static DECLARE_WORK(timer_update_work, timer_update_keys);
225 
226 #ifdef CONFIG_SMP
227 static unsigned int sysctl_timer_migration = 1;
228 
229 DEFINE_STATIC_KEY_FALSE(timers_migration_enabled);
230 
timers_update_migration(void)231 static void timers_update_migration(void)
232 {
233 	if (sysctl_timer_migration && tick_nohz_active)
234 		static_branch_enable(&timers_migration_enabled);
235 	else
236 		static_branch_disable(&timers_migration_enabled);
237 }
238 
239 #ifdef CONFIG_SYSCTL
timer_migration_handler(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)240 static int timer_migration_handler(struct ctl_table *table, int write,
241 			    void *buffer, size_t *lenp, loff_t *ppos)
242 {
243 	int ret;
244 
245 	mutex_lock(&timer_keys_mutex);
246 	ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
247 	if (!ret && write)
248 		timers_update_migration();
249 	mutex_unlock(&timer_keys_mutex);
250 	return ret;
251 }
252 
253 static struct ctl_table timer_sysctl[] = {
254 	{
255 		.procname	= "timer_migration",
256 		.data		= &sysctl_timer_migration,
257 		.maxlen		= sizeof(unsigned int),
258 		.mode		= 0644,
259 		.proc_handler	= timer_migration_handler,
260 		.extra1		= SYSCTL_ZERO,
261 		.extra2		= SYSCTL_ONE,
262 	},
263 	{}
264 };
265 
timer_sysctl_init(void)266 static int __init timer_sysctl_init(void)
267 {
268 	register_sysctl("kernel", timer_sysctl);
269 	return 0;
270 }
271 device_initcall(timer_sysctl_init);
272 #endif /* CONFIG_SYSCTL */
273 #else /* CONFIG_SMP */
timers_update_migration(void)274 static inline void timers_update_migration(void) { }
275 #endif /* !CONFIG_SMP */
276 
timer_update_keys(struct work_struct * work)277 static void timer_update_keys(struct work_struct *work)
278 {
279 	mutex_lock(&timer_keys_mutex);
280 	timers_update_migration();
281 	static_branch_enable(&timers_nohz_active);
282 	mutex_unlock(&timer_keys_mutex);
283 }
284 
timers_update_nohz(void)285 void timers_update_nohz(void)
286 {
287 	schedule_work(&timer_update_work);
288 }
289 
is_timers_nohz_active(void)290 static inline bool is_timers_nohz_active(void)
291 {
292 	return static_branch_unlikely(&timers_nohz_active);
293 }
294 #else
is_timers_nohz_active(void)295 static inline bool is_timers_nohz_active(void) { return false; }
296 #endif /* NO_HZ_COMMON */
297 
round_jiffies_common(unsigned long j,int cpu,bool force_up)298 static unsigned long round_jiffies_common(unsigned long j, int cpu,
299 		bool force_up)
300 {
301 	int rem;
302 	unsigned long original = j;
303 
304 	/*
305 	 * We don't want all cpus firing their timers at once hitting the
306 	 * same lock or cachelines, so we skew each extra cpu with an extra
307 	 * 3 jiffies. This 3 jiffies came originally from the mm/ code which
308 	 * already did this.
309 	 * The skew is done by adding 3*cpunr, then round, then subtract this
310 	 * extra offset again.
311 	 */
312 	j += cpu * 3;
313 
314 	rem = j % HZ;
315 
316 	/*
317 	 * If the target jiffie is just after a whole second (which can happen
318 	 * due to delays of the timer irq, long irq off times etc etc) then
319 	 * we should round down to the whole second, not up. Use 1/4th second
320 	 * as cutoff for this rounding as an extreme upper bound for this.
321 	 * But never round down if @force_up is set.
322 	 */
323 	if (rem < HZ/4 && !force_up) /* round down */
324 		j = j - rem;
325 	else /* round up */
326 		j = j - rem + HZ;
327 
328 	/* now that we have rounded, subtract the extra skew again */
329 	j -= cpu * 3;
330 
331 	/*
332 	 * Make sure j is still in the future. Otherwise return the
333 	 * unmodified value.
334 	 */
335 	return time_is_after_jiffies(j) ? j : original;
336 }
337 
338 /**
339  * __round_jiffies - function to round jiffies to a full second
340  * @j: the time in (absolute) jiffies that should be rounded
341  * @cpu: the processor number on which the timeout will happen
342  *
343  * __round_jiffies() rounds an absolute time in the future (in jiffies)
344  * up or down to (approximately) full seconds. This is useful for timers
345  * for which the exact time they fire does not matter too much, as long as
346  * they fire approximately every X seconds.
347  *
348  * By rounding these timers to whole seconds, all such timers will fire
349  * at the same time, rather than at various times spread out. The goal
350  * of this is to have the CPU wake up less, which saves power.
351  *
352  * The exact rounding is skewed for each processor to avoid all
353  * processors firing at the exact same time, which could lead
354  * to lock contention or spurious cache line bouncing.
355  *
356  * The return value is the rounded version of the @j parameter.
357  */
__round_jiffies(unsigned long j,int cpu)358 unsigned long __round_jiffies(unsigned long j, int cpu)
359 {
360 	return round_jiffies_common(j, cpu, false);
361 }
362 EXPORT_SYMBOL_GPL(__round_jiffies);
363 
364 /**
365  * __round_jiffies_relative - function to round jiffies to a full second
366  * @j: the time in (relative) jiffies that should be rounded
367  * @cpu: the processor number on which the timeout will happen
368  *
369  * __round_jiffies_relative() rounds a time delta  in the future (in jiffies)
370  * up or down to (approximately) full seconds. This is useful for timers
371  * for which the exact time they fire does not matter too much, as long as
372  * they fire approximately every X seconds.
373  *
374  * By rounding these timers to whole seconds, all such timers will fire
375  * at the same time, rather than at various times spread out. The goal
376  * of this is to have the CPU wake up less, which saves power.
377  *
378  * The exact rounding is skewed for each processor to avoid all
379  * processors firing at the exact same time, which could lead
380  * to lock contention or spurious cache line bouncing.
381  *
382  * The return value is the rounded version of the @j parameter.
383  */
__round_jiffies_relative(unsigned long j,int cpu)384 unsigned long __round_jiffies_relative(unsigned long j, int cpu)
385 {
386 	unsigned long j0 = jiffies;
387 
388 	/* Use j0 because jiffies might change while we run */
389 	return round_jiffies_common(j + j0, cpu, false) - j0;
390 }
391 EXPORT_SYMBOL_GPL(__round_jiffies_relative);
392 
393 /**
394  * round_jiffies - function to round jiffies to a full second
395  * @j: the time in (absolute) jiffies that should be rounded
396  *
397  * round_jiffies() rounds an absolute time in the future (in jiffies)
398  * up or down to (approximately) full seconds. This is useful for timers
399  * for which the exact time they fire does not matter too much, as long as
400  * they fire approximately every X seconds.
401  *
402  * By rounding these timers to whole seconds, all such timers will fire
403  * at the same time, rather than at various times spread out. The goal
404  * of this is to have the CPU wake up less, which saves power.
405  *
406  * The return value is the rounded version of the @j parameter.
407  */
round_jiffies(unsigned long j)408 unsigned long round_jiffies(unsigned long j)
409 {
410 	return round_jiffies_common(j, raw_smp_processor_id(), false);
411 }
412 EXPORT_SYMBOL_GPL(round_jiffies);
413 
414 /**
415  * round_jiffies_relative - function to round jiffies to a full second
416  * @j: the time in (relative) jiffies that should be rounded
417  *
418  * round_jiffies_relative() rounds a time delta  in the future (in jiffies)
419  * up or down to (approximately) full seconds. This is useful for timers
420  * for which the exact time they fire does not matter too much, as long as
421  * they fire approximately every X seconds.
422  *
423  * By rounding these timers to whole seconds, all such timers will fire
424  * at the same time, rather than at various times spread out. The goal
425  * of this is to have the CPU wake up less, which saves power.
426  *
427  * The return value is the rounded version of the @j parameter.
428  */
round_jiffies_relative(unsigned long j)429 unsigned long round_jiffies_relative(unsigned long j)
430 {
431 	return __round_jiffies_relative(j, raw_smp_processor_id());
432 }
433 EXPORT_SYMBOL_GPL(round_jiffies_relative);
434 
435 /**
436  * __round_jiffies_up - function to round jiffies up to a full second
437  * @j: the time in (absolute) jiffies that should be rounded
438  * @cpu: the processor number on which the timeout will happen
439  *
440  * This is the same as __round_jiffies() except that it will never
441  * round down.  This is useful for timeouts for which the exact time
442  * of firing does not matter too much, as long as they don't fire too
443  * early.
444  */
__round_jiffies_up(unsigned long j,int cpu)445 unsigned long __round_jiffies_up(unsigned long j, int cpu)
446 {
447 	return round_jiffies_common(j, cpu, true);
448 }
449 EXPORT_SYMBOL_GPL(__round_jiffies_up);
450 
451 /**
452  * __round_jiffies_up_relative - function to round jiffies up to a full second
453  * @j: the time in (relative) jiffies that should be rounded
454  * @cpu: the processor number on which the timeout will happen
455  *
456  * This is the same as __round_jiffies_relative() except that it will never
457  * round down.  This is useful for timeouts for which the exact time
458  * of firing does not matter too much, as long as they don't fire too
459  * early.
460  */
__round_jiffies_up_relative(unsigned long j,int cpu)461 unsigned long __round_jiffies_up_relative(unsigned long j, int cpu)
462 {
463 	unsigned long j0 = jiffies;
464 
465 	/* Use j0 because jiffies might change while we run */
466 	return round_jiffies_common(j + j0, cpu, true) - j0;
467 }
468 EXPORT_SYMBOL_GPL(__round_jiffies_up_relative);
469 
470 /**
471  * round_jiffies_up - function to round jiffies up to a full second
472  * @j: the time in (absolute) jiffies that should be rounded
473  *
474  * This is the same as round_jiffies() except that it will never
475  * round down.  This is useful for timeouts for which the exact time
476  * of firing does not matter too much, as long as they don't fire too
477  * early.
478  */
round_jiffies_up(unsigned long j)479 unsigned long round_jiffies_up(unsigned long j)
480 {
481 	return round_jiffies_common(j, raw_smp_processor_id(), true);
482 }
483 EXPORT_SYMBOL_GPL(round_jiffies_up);
484 
485 /**
486  * round_jiffies_up_relative - function to round jiffies up to a full second
487  * @j: the time in (relative) jiffies that should be rounded
488  *
489  * This is the same as round_jiffies_relative() except that it will never
490  * round down.  This is useful for timeouts for which the exact time
491  * of firing does not matter too much, as long as they don't fire too
492  * early.
493  */
round_jiffies_up_relative(unsigned long j)494 unsigned long round_jiffies_up_relative(unsigned long j)
495 {
496 	return __round_jiffies_up_relative(j, raw_smp_processor_id());
497 }
498 EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
499 
500 
timer_get_idx(struct timer_list * timer)501 static inline unsigned int timer_get_idx(struct timer_list *timer)
502 {
503 	return (timer->flags & TIMER_ARRAYMASK) >> TIMER_ARRAYSHIFT;
504 }
505 
timer_set_idx(struct timer_list * timer,unsigned int idx)506 static inline void timer_set_idx(struct timer_list *timer, unsigned int idx)
507 {
508 	timer->flags = (timer->flags & ~TIMER_ARRAYMASK) |
509 			idx << TIMER_ARRAYSHIFT;
510 }
511 
512 /*
513  * Helper function to calculate the array index for a given expiry
514  * time.
515  */
calc_index(unsigned long expires,unsigned lvl,unsigned long * bucket_expiry)516 static inline unsigned calc_index(unsigned long expires, unsigned lvl,
517 				  unsigned long *bucket_expiry)
518 {
519 
520 	/*
521 	 * The timer wheel has to guarantee that a timer does not fire
522 	 * early. Early expiry can happen due to:
523 	 * - Timer is armed at the edge of a tick
524 	 * - Truncation of the expiry time in the outer wheel levels
525 	 *
526 	 * Round up with level granularity to prevent this.
527 	 */
528 	expires = (expires >> LVL_SHIFT(lvl)) + 1;
529 	*bucket_expiry = expires << LVL_SHIFT(lvl);
530 	return LVL_OFFS(lvl) + (expires & LVL_MASK);
531 }
532 
calc_wheel_index(unsigned long expires,unsigned long clk,unsigned long * bucket_expiry)533 static int calc_wheel_index(unsigned long expires, unsigned long clk,
534 			    unsigned long *bucket_expiry)
535 {
536 	unsigned long delta = expires - clk;
537 	unsigned int idx;
538 
539 	if (delta < LVL_START(1)) {
540 		idx = calc_index(expires, 0, bucket_expiry);
541 	} else if (delta < LVL_START(2)) {
542 		idx = calc_index(expires, 1, bucket_expiry);
543 	} else if (delta < LVL_START(3)) {
544 		idx = calc_index(expires, 2, bucket_expiry);
545 	} else if (delta < LVL_START(4)) {
546 		idx = calc_index(expires, 3, bucket_expiry);
547 	} else if (delta < LVL_START(5)) {
548 		idx = calc_index(expires, 4, bucket_expiry);
549 	} else if (delta < LVL_START(6)) {
550 		idx = calc_index(expires, 5, bucket_expiry);
551 	} else if (delta < LVL_START(7)) {
552 		idx = calc_index(expires, 6, bucket_expiry);
553 	} else if (LVL_DEPTH > 8 && delta < LVL_START(8)) {
554 		idx = calc_index(expires, 7, bucket_expiry);
555 	} else if ((long) delta < 0) {
556 		idx = clk & LVL_MASK;
557 		*bucket_expiry = clk;
558 	} else {
559 		/*
560 		 * Force expire obscene large timeouts to expire at the
561 		 * capacity limit of the wheel.
562 		 */
563 		if (delta >= WHEEL_TIMEOUT_CUTOFF)
564 			expires = clk + WHEEL_TIMEOUT_MAX;
565 
566 		idx = calc_index(expires, LVL_DEPTH - 1, bucket_expiry);
567 	}
568 	return idx;
569 }
570 
571 static void
trigger_dyntick_cpu(struct timer_base * base,struct timer_list * timer)572 trigger_dyntick_cpu(struct timer_base *base, struct timer_list *timer)
573 {
574 	if (!is_timers_nohz_active())
575 		return;
576 
577 	/*
578 	 * TODO: This wants some optimizing similar to the code below, but we
579 	 * will do that when we switch from push to pull for deferrable timers.
580 	 */
581 	if (timer->flags & TIMER_DEFERRABLE) {
582 		if (tick_nohz_full_cpu(base->cpu))
583 			wake_up_nohz_cpu(base->cpu);
584 		return;
585 	}
586 
587 	/*
588 	 * We might have to IPI the remote CPU if the base is idle and the
589 	 * timer is not deferrable. If the other CPU is on the way to idle
590 	 * then it can't set base->is_idle as we hold the base lock:
591 	 */
592 	if (base->is_idle)
593 		wake_up_nohz_cpu(base->cpu);
594 }
595 
596 /*
597  * Enqueue the timer into the hash bucket, mark it pending in
598  * the bitmap, store the index in the timer flags then wake up
599  * the target CPU if needed.
600  */
enqueue_timer(struct timer_base * base,struct timer_list * timer,unsigned int idx,unsigned long bucket_expiry)601 static void enqueue_timer(struct timer_base *base, struct timer_list *timer,
602 			  unsigned int idx, unsigned long bucket_expiry)
603 {
604 
605 	hlist_add_head(&timer->entry, base->vectors + idx);
606 	__set_bit(idx, base->pending_map);
607 	timer_set_idx(timer, idx);
608 
609 	trace_timer_start(timer, timer->expires, timer->flags);
610 
611 	/*
612 	 * Check whether this is the new first expiring timer. The
613 	 * effective expiry time of the timer is required here
614 	 * (bucket_expiry) instead of timer->expires.
615 	 */
616 	if (time_before(bucket_expiry, base->next_expiry)) {
617 		/*
618 		 * Set the next expiry time and kick the CPU so it
619 		 * can reevaluate the wheel:
620 		 */
621 		base->next_expiry = bucket_expiry;
622 		base->timers_pending = true;
623 		base->next_expiry_recalc = false;
624 		trigger_dyntick_cpu(base, timer);
625 	}
626 }
627 
internal_add_timer(struct timer_base * base,struct timer_list * timer)628 static void internal_add_timer(struct timer_base *base, struct timer_list *timer)
629 {
630 	unsigned long bucket_expiry;
631 	unsigned int idx;
632 
633 	idx = calc_wheel_index(timer->expires, base->clk, &bucket_expiry);
634 	enqueue_timer(base, timer, idx, bucket_expiry);
635 }
636 
637 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
638 
639 static const struct debug_obj_descr timer_debug_descr;
640 
641 struct timer_hint {
642 	void	(*function)(struct timer_list *t);
643 	long	offset;
644 };
645 
646 #define TIMER_HINT(fn, container, timr, hintfn)			\
647 	{							\
648 		.function = fn,					\
649 		.offset	  = offsetof(container, hintfn) -	\
650 			    offsetof(container, timr)		\
651 	}
652 
653 static const struct timer_hint timer_hints[] = {
654 	TIMER_HINT(delayed_work_timer_fn,
655 		   struct delayed_work, timer, work.func),
656 	TIMER_HINT(kthread_delayed_work_timer_fn,
657 		   struct kthread_delayed_work, timer, work.func),
658 };
659 
timer_debug_hint(void * addr)660 static void *timer_debug_hint(void *addr)
661 {
662 	struct timer_list *timer = addr;
663 	int i;
664 
665 	for (i = 0; i < ARRAY_SIZE(timer_hints); i++) {
666 		if (timer_hints[i].function == timer->function) {
667 			void (**fn)(void) = addr + timer_hints[i].offset;
668 
669 			return *fn;
670 		}
671 	}
672 
673 	return timer->function;
674 }
675 
timer_is_static_object(void * addr)676 static bool timer_is_static_object(void *addr)
677 {
678 	struct timer_list *timer = addr;
679 
680 	return (timer->entry.pprev == NULL &&
681 		timer->entry.next == TIMER_ENTRY_STATIC);
682 }
683 
684 /*
685  * fixup_init is called when:
686  * - an active object is initialized
687  */
timer_fixup_init(void * addr,enum debug_obj_state state)688 static bool timer_fixup_init(void *addr, enum debug_obj_state state)
689 {
690 	struct timer_list *timer = addr;
691 
692 	switch (state) {
693 	case ODEBUG_STATE_ACTIVE:
694 		del_timer_sync(timer);
695 		debug_object_init(timer, &timer_debug_descr);
696 		return true;
697 	default:
698 		return false;
699 	}
700 }
701 
702 /* Stub timer callback for improperly used timers. */
stub_timer(struct timer_list * unused)703 static void stub_timer(struct timer_list *unused)
704 {
705 	WARN_ON(1);
706 }
707 
708 /*
709  * fixup_activate is called when:
710  * - an active object is activated
711  * - an unknown non-static object is activated
712  */
timer_fixup_activate(void * addr,enum debug_obj_state state)713 static bool timer_fixup_activate(void *addr, enum debug_obj_state state)
714 {
715 	struct timer_list *timer = addr;
716 
717 	switch (state) {
718 	case ODEBUG_STATE_NOTAVAILABLE:
719 		timer_setup(timer, stub_timer, 0);
720 		return true;
721 
722 	case ODEBUG_STATE_ACTIVE:
723 		WARN_ON(1);
724 		fallthrough;
725 	default:
726 		return false;
727 	}
728 }
729 
730 /*
731  * fixup_free is called when:
732  * - an active object is freed
733  */
timer_fixup_free(void * addr,enum debug_obj_state state)734 static bool timer_fixup_free(void *addr, enum debug_obj_state state)
735 {
736 	struct timer_list *timer = addr;
737 
738 	switch (state) {
739 	case ODEBUG_STATE_ACTIVE:
740 		del_timer_sync(timer);
741 		debug_object_free(timer, &timer_debug_descr);
742 		return true;
743 	default:
744 		return false;
745 	}
746 }
747 
748 /*
749  * fixup_assert_init is called when:
750  * - an untracked/uninit-ed object is found
751  */
timer_fixup_assert_init(void * addr,enum debug_obj_state state)752 static bool timer_fixup_assert_init(void *addr, enum debug_obj_state state)
753 {
754 	struct timer_list *timer = addr;
755 
756 	switch (state) {
757 	case ODEBUG_STATE_NOTAVAILABLE:
758 		timer_setup(timer, stub_timer, 0);
759 		return true;
760 	default:
761 		return false;
762 	}
763 }
764 
765 static const struct debug_obj_descr timer_debug_descr = {
766 	.name			= "timer_list",
767 	.debug_hint		= timer_debug_hint,
768 	.is_static_object	= timer_is_static_object,
769 	.fixup_init		= timer_fixup_init,
770 	.fixup_activate		= timer_fixup_activate,
771 	.fixup_free		= timer_fixup_free,
772 	.fixup_assert_init	= timer_fixup_assert_init,
773 };
774 
debug_timer_init(struct timer_list * timer)775 static inline void debug_timer_init(struct timer_list *timer)
776 {
777 	debug_object_init(timer, &timer_debug_descr);
778 }
779 
debug_timer_activate(struct timer_list * timer)780 static inline void debug_timer_activate(struct timer_list *timer)
781 {
782 	debug_object_activate(timer, &timer_debug_descr);
783 }
784 
debug_timer_deactivate(struct timer_list * timer)785 static inline void debug_timer_deactivate(struct timer_list *timer)
786 {
787 	debug_object_deactivate(timer, &timer_debug_descr);
788 }
789 
debug_timer_assert_init(struct timer_list * timer)790 static inline void debug_timer_assert_init(struct timer_list *timer)
791 {
792 	debug_object_assert_init(timer, &timer_debug_descr);
793 }
794 
795 static void do_init_timer(struct timer_list *timer,
796 			  void (*func)(struct timer_list *),
797 			  unsigned int flags,
798 			  const char *name, struct lock_class_key *key);
799 
init_timer_on_stack_key(struct timer_list * timer,void (* func)(struct timer_list *),unsigned int flags,const char * name,struct lock_class_key * key)800 void init_timer_on_stack_key(struct timer_list *timer,
801 			     void (*func)(struct timer_list *),
802 			     unsigned int flags,
803 			     const char *name, struct lock_class_key *key)
804 {
805 	debug_object_init_on_stack(timer, &timer_debug_descr);
806 	do_init_timer(timer, func, flags, name, key);
807 }
808 EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
809 
destroy_timer_on_stack(struct timer_list * timer)810 void destroy_timer_on_stack(struct timer_list *timer)
811 {
812 	debug_object_free(timer, &timer_debug_descr);
813 }
814 EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
815 
816 #else
debug_timer_init(struct timer_list * timer)817 static inline void debug_timer_init(struct timer_list *timer) { }
debug_timer_activate(struct timer_list * timer)818 static inline void debug_timer_activate(struct timer_list *timer) { }
debug_timer_deactivate(struct timer_list * timer)819 static inline void debug_timer_deactivate(struct timer_list *timer) { }
debug_timer_assert_init(struct timer_list * timer)820 static inline void debug_timer_assert_init(struct timer_list *timer) { }
821 #endif
822 
debug_init(struct timer_list * timer)823 static inline void debug_init(struct timer_list *timer)
824 {
825 	debug_timer_init(timer);
826 	trace_timer_init(timer);
827 }
828 
debug_deactivate(struct timer_list * timer)829 static inline void debug_deactivate(struct timer_list *timer)
830 {
831 	debug_timer_deactivate(timer);
832 	trace_timer_cancel(timer);
833 }
834 
debug_assert_init(struct timer_list * timer)835 static inline void debug_assert_init(struct timer_list *timer)
836 {
837 	debug_timer_assert_init(timer);
838 }
839 
do_init_timer(struct timer_list * timer,void (* func)(struct timer_list *),unsigned int flags,const char * name,struct lock_class_key * key)840 static void do_init_timer(struct timer_list *timer,
841 			  void (*func)(struct timer_list *),
842 			  unsigned int flags,
843 			  const char *name, struct lock_class_key *key)
844 {
845 	timer->entry.pprev = NULL;
846 	timer->function = func;
847 	if (WARN_ON_ONCE(flags & ~TIMER_INIT_FLAGS))
848 		flags &= TIMER_INIT_FLAGS;
849 	timer->flags = flags | raw_smp_processor_id();
850 	lockdep_init_map(&timer->lockdep_map, name, key, 0);
851 }
852 
853 /**
854  * init_timer_key - initialize a timer
855  * @timer: the timer to be initialized
856  * @func: timer callback function
857  * @flags: timer flags
858  * @name: name of the timer
859  * @key: lockdep class key of the fake lock used for tracking timer
860  *       sync lock dependencies
861  *
862  * init_timer_key() must be done to a timer prior calling *any* of the
863  * other timer functions.
864  */
init_timer_key(struct timer_list * timer,void (* func)(struct timer_list *),unsigned int flags,const char * name,struct lock_class_key * key)865 void init_timer_key(struct timer_list *timer,
866 		    void (*func)(struct timer_list *), unsigned int flags,
867 		    const char *name, struct lock_class_key *key)
868 {
869 	debug_init(timer);
870 	do_init_timer(timer, func, flags, name, key);
871 }
872 EXPORT_SYMBOL(init_timer_key);
873 
detach_timer(struct timer_list * timer,bool clear_pending)874 static inline void detach_timer(struct timer_list *timer, bool clear_pending)
875 {
876 	struct hlist_node *entry = &timer->entry;
877 
878 	debug_deactivate(timer);
879 
880 	__hlist_del(entry);
881 	if (clear_pending)
882 		entry->pprev = NULL;
883 	entry->next = LIST_POISON2;
884 }
885 
detach_if_pending(struct timer_list * timer,struct timer_base * base,bool clear_pending)886 static int detach_if_pending(struct timer_list *timer, struct timer_base *base,
887 			     bool clear_pending)
888 {
889 	unsigned idx = timer_get_idx(timer);
890 
891 	if (!timer_pending(timer))
892 		return 0;
893 
894 	if (hlist_is_singular_node(&timer->entry, base->vectors + idx)) {
895 		__clear_bit(idx, base->pending_map);
896 		base->next_expiry_recalc = true;
897 	}
898 
899 	detach_timer(timer, clear_pending);
900 	return 1;
901 }
902 
get_timer_cpu_base(u32 tflags,u32 cpu)903 static inline struct timer_base *get_timer_cpu_base(u32 tflags, u32 cpu)
904 {
905 	struct timer_base *base = per_cpu_ptr(&timer_bases[BASE_STD], cpu);
906 
907 	/*
908 	 * If the timer is deferrable and NO_HZ_COMMON is set then we need
909 	 * to use the deferrable base.
910 	 */
911 	if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE))
912 		base = per_cpu_ptr(&timer_bases[BASE_DEF], cpu);
913 	return base;
914 }
915 
get_timer_this_cpu_base(u32 tflags)916 static inline struct timer_base *get_timer_this_cpu_base(u32 tflags)
917 {
918 	struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
919 
920 	/*
921 	 * If the timer is deferrable and NO_HZ_COMMON is set then we need
922 	 * to use the deferrable base.
923 	 */
924 	if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE))
925 		base = this_cpu_ptr(&timer_bases[BASE_DEF]);
926 	return base;
927 }
928 
get_timer_base(u32 tflags)929 static inline struct timer_base *get_timer_base(u32 tflags)
930 {
931 	return get_timer_cpu_base(tflags, tflags & TIMER_CPUMASK);
932 }
933 
934 static inline struct timer_base *
get_target_base(struct timer_base * base,unsigned tflags)935 get_target_base(struct timer_base *base, unsigned tflags)
936 {
937 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
938 	if (static_branch_likely(&timers_migration_enabled) &&
939 	    !(tflags & TIMER_PINNED))
940 		return get_timer_cpu_base(tflags, get_nohz_timer_target());
941 #endif
942 	return get_timer_this_cpu_base(tflags);
943 }
944 
forward_timer_base(struct timer_base * base)945 static inline void forward_timer_base(struct timer_base *base)
946 {
947 	unsigned long jnow = READ_ONCE(jiffies);
948 
949 	/*
950 	 * No need to forward if we are close enough below jiffies.
951 	 * Also while executing timers, base->clk is 1 offset ahead
952 	 * of jiffies to avoid endless requeuing to current jiffies.
953 	 */
954 	if ((long)(jnow - base->clk) < 1)
955 		return;
956 
957 	/*
958 	 * If the next expiry value is > jiffies, then we fast forward to
959 	 * jiffies otherwise we forward to the next expiry value.
960 	 */
961 	if (time_after(base->next_expiry, jnow)) {
962 		base->clk = jnow;
963 	} else {
964 		if (WARN_ON_ONCE(time_before(base->next_expiry, base->clk)))
965 			return;
966 		base->clk = base->next_expiry;
967 	}
968 }
969 
970 
971 /*
972  * We are using hashed locking: Holding per_cpu(timer_bases[x]).lock means
973  * that all timers which are tied to this base are locked, and the base itself
974  * is locked too.
975  *
976  * So __run_timers/migrate_timers can safely modify all timers which could
977  * be found in the base->vectors array.
978  *
979  * When a timer is migrating then the TIMER_MIGRATING flag is set and we need
980  * to wait until the migration is done.
981  */
lock_timer_base(struct timer_list * timer,unsigned long * flags)982 static struct timer_base *lock_timer_base(struct timer_list *timer,
983 					  unsigned long *flags)
984 	__acquires(timer->base->lock)
985 {
986 	for (;;) {
987 		struct timer_base *base;
988 		u32 tf;
989 
990 		/*
991 		 * We need to use READ_ONCE() here, otherwise the compiler
992 		 * might re-read @tf between the check for TIMER_MIGRATING
993 		 * and spin_lock().
994 		 */
995 		tf = READ_ONCE(timer->flags);
996 
997 		if (!(tf & TIMER_MIGRATING)) {
998 			base = get_timer_base(tf);
999 			raw_spin_lock_irqsave(&base->lock, *flags);
1000 			if (timer->flags == tf)
1001 				return base;
1002 			raw_spin_unlock_irqrestore(&base->lock, *flags);
1003 		}
1004 		cpu_relax();
1005 	}
1006 }
1007 
1008 #define MOD_TIMER_PENDING_ONLY		0x01
1009 #define MOD_TIMER_REDUCE		0x02
1010 #define MOD_TIMER_NOTPENDING		0x04
1011 
1012 static inline int
__mod_timer(struct timer_list * timer,unsigned long expires,unsigned int options)1013 __mod_timer(struct timer_list *timer, unsigned long expires, unsigned int options)
1014 {
1015 	unsigned long clk = 0, flags, bucket_expiry;
1016 	struct timer_base *base, *new_base;
1017 	unsigned int idx = UINT_MAX;
1018 	int ret = 0;
1019 
1020 	debug_assert_init(timer);
1021 
1022 	/*
1023 	 * This is a common optimization triggered by the networking code - if
1024 	 * the timer is re-modified to have the same timeout or ends up in the
1025 	 * same array bucket then just return:
1026 	 */
1027 	if (!(options & MOD_TIMER_NOTPENDING) && timer_pending(timer)) {
1028 		/*
1029 		 * The downside of this optimization is that it can result in
1030 		 * larger granularity than you would get from adding a new
1031 		 * timer with this expiry.
1032 		 */
1033 		long diff = timer->expires - expires;
1034 
1035 		if (!diff)
1036 			return 1;
1037 		if (options & MOD_TIMER_REDUCE && diff <= 0)
1038 			return 1;
1039 
1040 		/*
1041 		 * We lock timer base and calculate the bucket index right
1042 		 * here. If the timer ends up in the same bucket, then we
1043 		 * just update the expiry time and avoid the whole
1044 		 * dequeue/enqueue dance.
1045 		 */
1046 		base = lock_timer_base(timer, &flags);
1047 		/*
1048 		 * Has @timer been shutdown? This needs to be evaluated
1049 		 * while holding base lock to prevent a race against the
1050 		 * shutdown code.
1051 		 */
1052 		if (!timer->function)
1053 			goto out_unlock;
1054 
1055 		forward_timer_base(base);
1056 
1057 		if (timer_pending(timer) && (options & MOD_TIMER_REDUCE) &&
1058 		    time_before_eq(timer->expires, expires)) {
1059 			ret = 1;
1060 			goto out_unlock;
1061 		}
1062 
1063 		clk = base->clk;
1064 		idx = calc_wheel_index(expires, clk, &bucket_expiry);
1065 
1066 		/*
1067 		 * Retrieve and compare the array index of the pending
1068 		 * timer. If it matches set the expiry to the new value so a
1069 		 * subsequent call will exit in the expires check above.
1070 		 */
1071 		if (idx == timer_get_idx(timer)) {
1072 			if (!(options & MOD_TIMER_REDUCE))
1073 				timer->expires = expires;
1074 			else if (time_after(timer->expires, expires))
1075 				timer->expires = expires;
1076 			ret = 1;
1077 			goto out_unlock;
1078 		}
1079 	} else {
1080 		base = lock_timer_base(timer, &flags);
1081 		/*
1082 		 * Has @timer been shutdown? This needs to be evaluated
1083 		 * while holding base lock to prevent a race against the
1084 		 * shutdown code.
1085 		 */
1086 		if (!timer->function)
1087 			goto out_unlock;
1088 
1089 		forward_timer_base(base);
1090 	}
1091 
1092 	ret = detach_if_pending(timer, base, false);
1093 	if (!ret && (options & MOD_TIMER_PENDING_ONLY))
1094 		goto out_unlock;
1095 
1096 	new_base = get_target_base(base, timer->flags);
1097 
1098 	if (base != new_base) {
1099 		/*
1100 		 * We are trying to schedule the timer on the new base.
1101 		 * However we can't change timer's base while it is running,
1102 		 * otherwise timer_delete_sync() can't detect that the timer's
1103 		 * handler yet has not finished. This also guarantees that the
1104 		 * timer is serialized wrt itself.
1105 		 */
1106 		if (likely(base->running_timer != timer)) {
1107 			/* See the comment in lock_timer_base() */
1108 			timer->flags |= TIMER_MIGRATING;
1109 
1110 			raw_spin_unlock(&base->lock);
1111 			base = new_base;
1112 			raw_spin_lock(&base->lock);
1113 			WRITE_ONCE(timer->flags,
1114 				   (timer->flags & ~TIMER_BASEMASK) | base->cpu);
1115 			forward_timer_base(base);
1116 		}
1117 	}
1118 
1119 	debug_timer_activate(timer);
1120 
1121 	timer->expires = expires;
1122 	/*
1123 	 * If 'idx' was calculated above and the base time did not advance
1124 	 * between calculating 'idx' and possibly switching the base, only
1125 	 * enqueue_timer() is required. Otherwise we need to (re)calculate
1126 	 * the wheel index via internal_add_timer().
1127 	 */
1128 	if (idx != UINT_MAX && clk == base->clk)
1129 		enqueue_timer(base, timer, idx, bucket_expiry);
1130 	else
1131 		internal_add_timer(base, timer);
1132 
1133 out_unlock:
1134 	raw_spin_unlock_irqrestore(&base->lock, flags);
1135 
1136 	return ret;
1137 }
1138 
1139 /**
1140  * mod_timer_pending - Modify a pending timer's timeout
1141  * @timer:	The pending timer to be modified
1142  * @expires:	New absolute timeout in jiffies
1143  *
1144  * mod_timer_pending() is the same for pending timers as mod_timer(), but
1145  * will not activate inactive timers.
1146  *
1147  * If @timer->function == NULL then the start operation is silently
1148  * discarded.
1149  *
1150  * Return:
1151  * * %0 - The timer was inactive and not modified or was in
1152  *	  shutdown state and the operation was discarded
1153  * * %1 - The timer was active and requeued to expire at @expires
1154  */
mod_timer_pending(struct timer_list * timer,unsigned long expires)1155 int mod_timer_pending(struct timer_list *timer, unsigned long expires)
1156 {
1157 	return __mod_timer(timer, expires, MOD_TIMER_PENDING_ONLY);
1158 }
1159 EXPORT_SYMBOL(mod_timer_pending);
1160 
1161 /**
1162  * mod_timer - Modify a timer's timeout
1163  * @timer:	The timer to be modified
1164  * @expires:	New absolute timeout in jiffies
1165  *
1166  * mod_timer(timer, expires) is equivalent to:
1167  *
1168  *     del_timer(timer); timer->expires = expires; add_timer(timer);
1169  *
1170  * mod_timer() is more efficient than the above open coded sequence. In
1171  * case that the timer is inactive, the del_timer() part is a NOP. The
1172  * timer is in any case activated with the new expiry time @expires.
1173  *
1174  * Note that if there are multiple unserialized concurrent users of the
1175  * same timer, then mod_timer() is the only safe way to modify the timeout,
1176  * since add_timer() cannot modify an already running timer.
1177  *
1178  * If @timer->function == NULL then the start operation is silently
1179  * discarded. In this case the return value is 0 and meaningless.
1180  *
1181  * Return:
1182  * * %0 - The timer was inactive and started or was in shutdown
1183  *	  state and the operation was discarded
1184  * * %1 - The timer was active and requeued to expire at @expires or
1185  *	  the timer was active and not modified because @expires did
1186  *	  not change the effective expiry time
1187  */
mod_timer(struct timer_list * timer,unsigned long expires)1188 int mod_timer(struct timer_list *timer, unsigned long expires)
1189 {
1190 	return __mod_timer(timer, expires, 0);
1191 }
1192 EXPORT_SYMBOL(mod_timer);
1193 
1194 /**
1195  * timer_reduce - Modify a timer's timeout if it would reduce the timeout
1196  * @timer:	The timer to be modified
1197  * @expires:	New absolute timeout in jiffies
1198  *
1199  * timer_reduce() is very similar to mod_timer(), except that it will only
1200  * modify an enqueued timer if that would reduce the expiration time. If
1201  * @timer is not enqueued it starts the timer.
1202  *
1203  * If @timer->function == NULL then the start operation is silently
1204  * discarded.
1205  *
1206  * Return:
1207  * * %0 - The timer was inactive and started or was in shutdown
1208  *	  state and the operation was discarded
1209  * * %1 - The timer was active and requeued to expire at @expires or
1210  *	  the timer was active and not modified because @expires
1211  *	  did not change the effective expiry time such that the
1212  *	  timer would expire earlier than already scheduled
1213  */
timer_reduce(struct timer_list * timer,unsigned long expires)1214 int timer_reduce(struct timer_list *timer, unsigned long expires)
1215 {
1216 	return __mod_timer(timer, expires, MOD_TIMER_REDUCE);
1217 }
1218 EXPORT_SYMBOL(timer_reduce);
1219 
1220 /**
1221  * add_timer - Start a timer
1222  * @timer:	The timer to be started
1223  *
1224  * Start @timer to expire at @timer->expires in the future. @timer->expires
1225  * is the absolute expiry time measured in 'jiffies'. When the timer expires
1226  * timer->function(timer) will be invoked from soft interrupt context.
1227  *
1228  * The @timer->expires and @timer->function fields must be set prior
1229  * to calling this function.
1230  *
1231  * If @timer->function == NULL then the start operation is silently
1232  * discarded.
1233  *
1234  * If @timer->expires is already in the past @timer will be queued to
1235  * expire at the next timer tick.
1236  *
1237  * This can only operate on an inactive timer. Attempts to invoke this on
1238  * an active timer are rejected with a warning.
1239  */
add_timer(struct timer_list * timer)1240 void add_timer(struct timer_list *timer)
1241 {
1242 	if (WARN_ON_ONCE(timer_pending(timer)))
1243 		return;
1244 	__mod_timer(timer, timer->expires, MOD_TIMER_NOTPENDING);
1245 }
1246 EXPORT_SYMBOL(add_timer);
1247 
1248 /**
1249  * add_timer_on - Start a timer on a particular CPU
1250  * @timer:	The timer to be started
1251  * @cpu:	The CPU to start it on
1252  *
1253  * Same as add_timer() except that it starts the timer on the given CPU.
1254  *
1255  * See add_timer() for further details.
1256  */
add_timer_on(struct timer_list * timer,int cpu)1257 void add_timer_on(struct timer_list *timer, int cpu)
1258 {
1259 	struct timer_base *new_base, *base;
1260 	unsigned long flags;
1261 
1262 	debug_assert_init(timer);
1263 
1264 	if (WARN_ON_ONCE(timer_pending(timer)))
1265 		return;
1266 
1267 	new_base = get_timer_cpu_base(timer->flags, cpu);
1268 
1269 	/*
1270 	 * If @timer was on a different CPU, it should be migrated with the
1271 	 * old base locked to prevent other operations proceeding with the
1272 	 * wrong base locked.  See lock_timer_base().
1273 	 */
1274 	base = lock_timer_base(timer, &flags);
1275 	/*
1276 	 * Has @timer been shutdown? This needs to be evaluated while
1277 	 * holding base lock to prevent a race against the shutdown code.
1278 	 */
1279 	if (!timer->function)
1280 		goto out_unlock;
1281 
1282 	if (base != new_base) {
1283 		timer->flags |= TIMER_MIGRATING;
1284 
1285 		raw_spin_unlock(&base->lock);
1286 		base = new_base;
1287 		raw_spin_lock(&base->lock);
1288 		WRITE_ONCE(timer->flags,
1289 			   (timer->flags & ~TIMER_BASEMASK) | cpu);
1290 	}
1291 	forward_timer_base(base);
1292 
1293 	debug_timer_activate(timer);
1294 	internal_add_timer(base, timer);
1295 out_unlock:
1296 	raw_spin_unlock_irqrestore(&base->lock, flags);
1297 }
1298 EXPORT_SYMBOL_GPL(add_timer_on);
1299 
1300 /**
1301  * __timer_delete - Internal function: Deactivate a timer
1302  * @timer:	The timer to be deactivated
1303  * @shutdown:	If true, this indicates that the timer is about to be
1304  *		shutdown permanently.
1305  *
1306  * If @shutdown is true then @timer->function is set to NULL under the
1307  * timer base lock which prevents further rearming of the time. In that
1308  * case any attempt to rearm @timer after this function returns will be
1309  * silently ignored.
1310  *
1311  * Return:
1312  * * %0 - The timer was not pending
1313  * * %1 - The timer was pending and deactivated
1314  */
__timer_delete(struct timer_list * timer,bool shutdown)1315 static int __timer_delete(struct timer_list *timer, bool shutdown)
1316 {
1317 	struct timer_base *base;
1318 	unsigned long flags;
1319 	int ret = 0;
1320 
1321 	debug_assert_init(timer);
1322 
1323 	/*
1324 	 * If @shutdown is set then the lock has to be taken whether the
1325 	 * timer is pending or not to protect against a concurrent rearm
1326 	 * which might hit between the lockless pending check and the lock
1327 	 * aquisition. By taking the lock it is ensured that such a newly
1328 	 * enqueued timer is dequeued and cannot end up with
1329 	 * timer->function == NULL in the expiry code.
1330 	 *
1331 	 * If timer->function is currently executed, then this makes sure
1332 	 * that the callback cannot requeue the timer.
1333 	 */
1334 	if (timer_pending(timer) || shutdown) {
1335 		base = lock_timer_base(timer, &flags);
1336 		ret = detach_if_pending(timer, base, true);
1337 		if (shutdown)
1338 			timer->function = NULL;
1339 		raw_spin_unlock_irqrestore(&base->lock, flags);
1340 	}
1341 
1342 	return ret;
1343 }
1344 
1345 /**
1346  * timer_delete - Deactivate a timer
1347  * @timer:	The timer to be deactivated
1348  *
1349  * The function only deactivates a pending timer, but contrary to
1350  * timer_delete_sync() it does not take into account whether the timer's
1351  * callback function is concurrently executed on a different CPU or not.
1352  * It neither prevents rearming of the timer.  If @timer can be rearmed
1353  * concurrently then the return value of this function is meaningless.
1354  *
1355  * Return:
1356  * * %0 - The timer was not pending
1357  * * %1 - The timer was pending and deactivated
1358  */
timer_delete(struct timer_list * timer)1359 int timer_delete(struct timer_list *timer)
1360 {
1361 	return __timer_delete(timer, false);
1362 }
1363 EXPORT_SYMBOL(timer_delete);
1364 
1365 /**
1366  * timer_shutdown - Deactivate a timer and prevent rearming
1367  * @timer:	The timer to be deactivated
1368  *
1369  * The function does not wait for an eventually running timer callback on a
1370  * different CPU but it prevents rearming of the timer. Any attempt to arm
1371  * @timer after this function returns will be silently ignored.
1372  *
1373  * This function is useful for teardown code and should only be used when
1374  * timer_shutdown_sync() cannot be invoked due to locking or context constraints.
1375  *
1376  * Return:
1377  * * %0 - The timer was not pending
1378  * * %1 - The timer was pending
1379  */
timer_shutdown(struct timer_list * timer)1380 int timer_shutdown(struct timer_list *timer)
1381 {
1382 	return __timer_delete(timer, true);
1383 }
1384 EXPORT_SYMBOL_GPL(timer_shutdown);
1385 
1386 /**
1387  * __try_to_del_timer_sync - Internal function: Try to deactivate a timer
1388  * @timer:	Timer to deactivate
1389  * @shutdown:	If true, this indicates that the timer is about to be
1390  *		shutdown permanently.
1391  *
1392  * If @shutdown is true then @timer->function is set to NULL under the
1393  * timer base lock which prevents further rearming of the timer. Any
1394  * attempt to rearm @timer after this function returns will be silently
1395  * ignored.
1396  *
1397  * This function cannot guarantee that the timer cannot be rearmed
1398  * right after dropping the base lock if @shutdown is false. That
1399  * needs to be prevented by the calling code if necessary.
1400  *
1401  * Return:
1402  * * %0  - The timer was not pending
1403  * * %1  - The timer was pending and deactivated
1404  * * %-1 - The timer callback function is running on a different CPU
1405  */
__try_to_del_timer_sync(struct timer_list * timer,bool shutdown)1406 static int __try_to_del_timer_sync(struct timer_list *timer, bool shutdown)
1407 {
1408 	struct timer_base *base;
1409 	unsigned long flags;
1410 	int ret = -1;
1411 
1412 	debug_assert_init(timer);
1413 
1414 	base = lock_timer_base(timer, &flags);
1415 
1416 	if (base->running_timer != timer)
1417 		ret = detach_if_pending(timer, base, true);
1418 	if (shutdown)
1419 		timer->function = NULL;
1420 
1421 	raw_spin_unlock_irqrestore(&base->lock, flags);
1422 
1423 	return ret;
1424 }
1425 
1426 /**
1427  * try_to_del_timer_sync - Try to deactivate a timer
1428  * @timer:	Timer to deactivate
1429  *
1430  * This function tries to deactivate a timer. On success the timer is not
1431  * queued and the timer callback function is not running on any CPU.
1432  *
1433  * This function does not guarantee that the timer cannot be rearmed right
1434  * after dropping the base lock. That needs to be prevented by the calling
1435  * code if necessary.
1436  *
1437  * Return:
1438  * * %0  - The timer was not pending
1439  * * %1  - The timer was pending and deactivated
1440  * * %-1 - The timer callback function is running on a different CPU
1441  */
try_to_del_timer_sync(struct timer_list * timer)1442 int try_to_del_timer_sync(struct timer_list *timer)
1443 {
1444 	return __try_to_del_timer_sync(timer, false);
1445 }
1446 EXPORT_SYMBOL(try_to_del_timer_sync);
1447 
1448 #ifdef CONFIG_PREEMPT_RT
timer_base_init_expiry_lock(struct timer_base * base)1449 static __init void timer_base_init_expiry_lock(struct timer_base *base)
1450 {
1451 	spin_lock_init(&base->expiry_lock);
1452 }
1453 
timer_base_lock_expiry(struct timer_base * base)1454 static inline void timer_base_lock_expiry(struct timer_base *base)
1455 {
1456 	spin_lock(&base->expiry_lock);
1457 }
1458 
timer_base_unlock_expiry(struct timer_base * base)1459 static inline void timer_base_unlock_expiry(struct timer_base *base)
1460 {
1461 	spin_unlock(&base->expiry_lock);
1462 }
1463 
1464 /*
1465  * The counterpart to del_timer_wait_running().
1466  *
1467  * If there is a waiter for base->expiry_lock, then it was waiting for the
1468  * timer callback to finish. Drop expiry_lock and reacquire it. That allows
1469  * the waiter to acquire the lock and make progress.
1470  */
timer_sync_wait_running(struct timer_base * base)1471 static void timer_sync_wait_running(struct timer_base *base)
1472 {
1473 	if (atomic_read(&base->timer_waiters)) {
1474 		raw_spin_unlock_irq(&base->lock);
1475 		spin_unlock(&base->expiry_lock);
1476 		spin_lock(&base->expiry_lock);
1477 		raw_spin_lock_irq(&base->lock);
1478 	}
1479 }
1480 
1481 /*
1482  * This function is called on PREEMPT_RT kernels when the fast path
1483  * deletion of a timer failed because the timer callback function was
1484  * running.
1485  *
1486  * This prevents priority inversion, if the softirq thread on a remote CPU
1487  * got preempted, and it prevents a life lock when the task which tries to
1488  * delete a timer preempted the softirq thread running the timer callback
1489  * function.
1490  */
del_timer_wait_running(struct timer_list * timer)1491 static void del_timer_wait_running(struct timer_list *timer)
1492 {
1493 	u32 tf;
1494 
1495 	tf = READ_ONCE(timer->flags);
1496 	if (!(tf & (TIMER_MIGRATING | TIMER_IRQSAFE))) {
1497 		struct timer_base *base = get_timer_base(tf);
1498 
1499 		/*
1500 		 * Mark the base as contended and grab the expiry lock,
1501 		 * which is held by the softirq across the timer
1502 		 * callback. Drop the lock immediately so the softirq can
1503 		 * expire the next timer. In theory the timer could already
1504 		 * be running again, but that's more than unlikely and just
1505 		 * causes another wait loop.
1506 		 */
1507 		atomic_inc(&base->timer_waiters);
1508 		spin_lock_bh(&base->expiry_lock);
1509 		atomic_dec(&base->timer_waiters);
1510 		spin_unlock_bh(&base->expiry_lock);
1511 	}
1512 }
1513 #else
timer_base_init_expiry_lock(struct timer_base * base)1514 static inline void timer_base_init_expiry_lock(struct timer_base *base) { }
timer_base_lock_expiry(struct timer_base * base)1515 static inline void timer_base_lock_expiry(struct timer_base *base) { }
timer_base_unlock_expiry(struct timer_base * base)1516 static inline void timer_base_unlock_expiry(struct timer_base *base) { }
timer_sync_wait_running(struct timer_base * base)1517 static inline void timer_sync_wait_running(struct timer_base *base) { }
del_timer_wait_running(struct timer_list * timer)1518 static inline void del_timer_wait_running(struct timer_list *timer) { }
1519 #endif
1520 
1521 /**
1522  * __timer_delete_sync - Internal function: Deactivate a timer and wait
1523  *			 for the handler to finish.
1524  * @timer:	The timer to be deactivated
1525  * @shutdown:	If true, @timer->function will be set to NULL under the
1526  *		timer base lock which prevents rearming of @timer
1527  *
1528  * If @shutdown is not set the timer can be rearmed later. If the timer can
1529  * be rearmed concurrently, i.e. after dropping the base lock then the
1530  * return value is meaningless.
1531  *
1532  * If @shutdown is set then @timer->function is set to NULL under timer
1533  * base lock which prevents rearming of the timer. Any attempt to rearm
1534  * a shutdown timer is silently ignored.
1535  *
1536  * If the timer should be reused after shutdown it has to be initialized
1537  * again.
1538  *
1539  * Return:
1540  * * %0	- The timer was not pending
1541  * * %1	- The timer was pending and deactivated
1542  */
__timer_delete_sync(struct timer_list * timer,bool shutdown)1543 static int __timer_delete_sync(struct timer_list *timer, bool shutdown)
1544 {
1545 	int ret;
1546 
1547 #ifdef CONFIG_LOCKDEP
1548 	unsigned long flags;
1549 
1550 	/*
1551 	 * If lockdep gives a backtrace here, please reference
1552 	 * the synchronization rules above.
1553 	 */
1554 	local_irq_save(flags);
1555 	lock_map_acquire(&timer->lockdep_map);
1556 	lock_map_release(&timer->lockdep_map);
1557 	local_irq_restore(flags);
1558 #endif
1559 	/*
1560 	 * don't use it in hardirq context, because it
1561 	 * could lead to deadlock.
1562 	 */
1563 	WARN_ON(in_hardirq() && !(timer->flags & TIMER_IRQSAFE));
1564 
1565 	/*
1566 	 * Must be able to sleep on PREEMPT_RT because of the slowpath in
1567 	 * del_timer_wait_running().
1568 	 */
1569 	if (IS_ENABLED(CONFIG_PREEMPT_RT) && !(timer->flags & TIMER_IRQSAFE))
1570 		lockdep_assert_preemption_enabled();
1571 
1572 	do {
1573 		ret = __try_to_del_timer_sync(timer, shutdown);
1574 
1575 		if (unlikely(ret < 0)) {
1576 			del_timer_wait_running(timer);
1577 			cpu_relax();
1578 		}
1579 	} while (ret < 0);
1580 
1581 	return ret;
1582 }
1583 
1584 /**
1585  * timer_delete_sync - Deactivate a timer and wait for the handler to finish.
1586  * @timer:	The timer to be deactivated
1587  *
1588  * Synchronization rules: Callers must prevent restarting of the timer,
1589  * otherwise this function is meaningless. It must not be called from
1590  * interrupt contexts unless the timer is an irqsafe one. The caller must
1591  * not hold locks which would prevent completion of the timer's callback
1592  * function. The timer's handler must not call add_timer_on(). Upon exit
1593  * the timer is not queued and the handler is not running on any CPU.
1594  *
1595  * For !irqsafe timers, the caller must not hold locks that are held in
1596  * interrupt context. Even if the lock has nothing to do with the timer in
1597  * question.  Here's why::
1598  *
1599  *    CPU0                             CPU1
1600  *    ----                             ----
1601  *                                     <SOFTIRQ>
1602  *                                       call_timer_fn();
1603  *                                       base->running_timer = mytimer;
1604  *    spin_lock_irq(somelock);
1605  *                                     <IRQ>
1606  *                                        spin_lock(somelock);
1607  *    timer_delete_sync(mytimer);
1608  *    while (base->running_timer == mytimer);
1609  *
1610  * Now timer_delete_sync() will never return and never release somelock.
1611  * The interrupt on the other CPU is waiting to grab somelock but it has
1612  * interrupted the softirq that CPU0 is waiting to finish.
1613  *
1614  * This function cannot guarantee that the timer is not rearmed again by
1615  * some concurrent or preempting code, right after it dropped the base
1616  * lock. If there is the possibility of a concurrent rearm then the return
1617  * value of the function is meaningless.
1618  *
1619  * If such a guarantee is needed, e.g. for teardown situations then use
1620  * timer_shutdown_sync() instead.
1621  *
1622  * Return:
1623  * * %0	- The timer was not pending
1624  * * %1	- The timer was pending and deactivated
1625  */
timer_delete_sync(struct timer_list * timer)1626 int timer_delete_sync(struct timer_list *timer)
1627 {
1628 	return __timer_delete_sync(timer, false);
1629 }
1630 EXPORT_SYMBOL(timer_delete_sync);
1631 
1632 /**
1633  * timer_shutdown_sync - Shutdown a timer and prevent rearming
1634  * @timer: The timer to be shutdown
1635  *
1636  * When the function returns it is guaranteed that:
1637  *   - @timer is not queued
1638  *   - The callback function of @timer is not running
1639  *   - @timer cannot be enqueued again. Any attempt to rearm
1640  *     @timer is silently ignored.
1641  *
1642  * See timer_delete_sync() for synchronization rules.
1643  *
1644  * This function is useful for final teardown of an infrastructure where
1645  * the timer is subject to a circular dependency problem.
1646  *
1647  * A common pattern for this is a timer and a workqueue where the timer can
1648  * schedule work and work can arm the timer. On shutdown the workqueue must
1649  * be destroyed and the timer must be prevented from rearming. Unless the
1650  * code has conditionals like 'if (mything->in_shutdown)' to prevent that
1651  * there is no way to get this correct with timer_delete_sync().
1652  *
1653  * timer_shutdown_sync() is solving the problem. The correct ordering of
1654  * calls in this case is:
1655  *
1656  *	timer_shutdown_sync(&mything->timer);
1657  *	workqueue_destroy(&mything->workqueue);
1658  *
1659  * After this 'mything' can be safely freed.
1660  *
1661  * This obviously implies that the timer is not required to be functional
1662  * for the rest of the shutdown operation.
1663  *
1664  * Return:
1665  * * %0 - The timer was not pending
1666  * * %1 - The timer was pending
1667  */
timer_shutdown_sync(struct timer_list * timer)1668 int timer_shutdown_sync(struct timer_list *timer)
1669 {
1670 	return __timer_delete_sync(timer, true);
1671 }
1672 EXPORT_SYMBOL_GPL(timer_shutdown_sync);
1673 
call_timer_fn(struct timer_list * timer,void (* fn)(struct timer_list *),unsigned long baseclk)1674 static void call_timer_fn(struct timer_list *timer,
1675 			  void (*fn)(struct timer_list *),
1676 			  unsigned long baseclk)
1677 {
1678 	int count = preempt_count();
1679 
1680 #ifdef CONFIG_LOCKDEP
1681 	/*
1682 	 * It is permissible to free the timer from inside the
1683 	 * function that is called from it, this we need to take into
1684 	 * account for lockdep too. To avoid bogus "held lock freed"
1685 	 * warnings as well as problems when looking into
1686 	 * timer->lockdep_map, make a copy and use that here.
1687 	 */
1688 	struct lockdep_map lockdep_map;
1689 
1690 	lockdep_copy_map(&lockdep_map, &timer->lockdep_map);
1691 #endif
1692 	/*
1693 	 * Couple the lock chain with the lock chain at
1694 	 * timer_delete_sync() by acquiring the lock_map around the fn()
1695 	 * call here and in timer_delete_sync().
1696 	 */
1697 	lock_map_acquire(&lockdep_map);
1698 
1699 	trace_timer_expire_entry(timer, baseclk);
1700 	fn(timer);
1701 	trace_timer_expire_exit(timer);
1702 
1703 	lock_map_release(&lockdep_map);
1704 
1705 	if (count != preempt_count()) {
1706 		WARN_ONCE(1, "timer: %pS preempt leak: %08x -> %08x\n",
1707 			  fn, count, preempt_count());
1708 		/*
1709 		 * Restore the preempt count. That gives us a decent
1710 		 * chance to survive and extract information. If the
1711 		 * callback kept a lock held, bad luck, but not worse
1712 		 * than the BUG() we had.
1713 		 */
1714 		preempt_count_set(count);
1715 	}
1716 }
1717 
expire_timers(struct timer_base * base,struct hlist_head * head)1718 static void expire_timers(struct timer_base *base, struct hlist_head *head)
1719 {
1720 	/*
1721 	 * This value is required only for tracing. base->clk was
1722 	 * incremented directly before expire_timers was called. But expiry
1723 	 * is related to the old base->clk value.
1724 	 */
1725 	unsigned long baseclk = base->clk - 1;
1726 
1727 	while (!hlist_empty(head)) {
1728 		struct timer_list *timer;
1729 		void (*fn)(struct timer_list *);
1730 
1731 		timer = hlist_entry(head->first, struct timer_list, entry);
1732 
1733 		base->running_timer = timer;
1734 		detach_timer(timer, true);
1735 
1736 		fn = timer->function;
1737 
1738 		if (WARN_ON_ONCE(!fn)) {
1739 			/* Should never happen. Emphasis on should! */
1740 			base->running_timer = NULL;
1741 			continue;
1742 		}
1743 
1744 		if (timer->flags & TIMER_IRQSAFE) {
1745 			raw_spin_unlock(&base->lock);
1746 			call_timer_fn(timer, fn, baseclk);
1747 			raw_spin_lock(&base->lock);
1748 			base->running_timer = NULL;
1749 		} else {
1750 			raw_spin_unlock_irq(&base->lock);
1751 			call_timer_fn(timer, fn, baseclk);
1752 			raw_spin_lock_irq(&base->lock);
1753 			base->running_timer = NULL;
1754 			timer_sync_wait_running(base);
1755 		}
1756 	}
1757 }
1758 
collect_expired_timers(struct timer_base * base,struct hlist_head * heads)1759 static int collect_expired_timers(struct timer_base *base,
1760 				  struct hlist_head *heads)
1761 {
1762 	unsigned long clk = base->clk = base->next_expiry;
1763 	struct hlist_head *vec;
1764 	int i, levels = 0;
1765 	unsigned int idx;
1766 
1767 	for (i = 0; i < LVL_DEPTH; i++) {
1768 		idx = (clk & LVL_MASK) + i * LVL_SIZE;
1769 
1770 		if (__test_and_clear_bit(idx, base->pending_map)) {
1771 			vec = base->vectors + idx;
1772 			hlist_move_list(vec, heads++);
1773 			levels++;
1774 		}
1775 		/* Is it time to look at the next level? */
1776 		if (clk & LVL_CLK_MASK)
1777 			break;
1778 		/* Shift clock for the next level granularity */
1779 		clk >>= LVL_CLK_SHIFT;
1780 	}
1781 	return levels;
1782 }
1783 
1784 /*
1785  * Find the next pending bucket of a level. Search from level start (@offset)
1786  * + @clk upwards and if nothing there, search from start of the level
1787  * (@offset) up to @offset + clk.
1788  */
next_pending_bucket(struct timer_base * base,unsigned offset,unsigned clk)1789 static int next_pending_bucket(struct timer_base *base, unsigned offset,
1790 			       unsigned clk)
1791 {
1792 	unsigned pos, start = offset + clk;
1793 	unsigned end = offset + LVL_SIZE;
1794 
1795 	pos = find_next_bit(base->pending_map, end, start);
1796 	if (pos < end)
1797 		return pos - start;
1798 
1799 	pos = find_next_bit(base->pending_map, start, offset);
1800 	return pos < start ? pos + LVL_SIZE - start : -1;
1801 }
1802 
1803 /*
1804  * Search the first expiring timer in the various clock levels. Caller must
1805  * hold base->lock.
1806  */
__next_timer_interrupt(struct timer_base * base)1807 static unsigned long __next_timer_interrupt(struct timer_base *base)
1808 {
1809 	unsigned long clk, next, adj;
1810 	unsigned lvl, offset = 0;
1811 
1812 	next = base->clk + NEXT_TIMER_MAX_DELTA;
1813 	clk = base->clk;
1814 	for (lvl = 0; lvl < LVL_DEPTH; lvl++, offset += LVL_SIZE) {
1815 		int pos = next_pending_bucket(base, offset, clk & LVL_MASK);
1816 		unsigned long lvl_clk = clk & LVL_CLK_MASK;
1817 
1818 		if (pos >= 0) {
1819 			unsigned long tmp = clk + (unsigned long) pos;
1820 
1821 			tmp <<= LVL_SHIFT(lvl);
1822 			if (time_before(tmp, next))
1823 				next = tmp;
1824 
1825 			/*
1826 			 * If the next expiration happens before we reach
1827 			 * the next level, no need to check further.
1828 			 */
1829 			if (pos <= ((LVL_CLK_DIV - lvl_clk) & LVL_CLK_MASK))
1830 				break;
1831 		}
1832 		/*
1833 		 * Clock for the next level. If the current level clock lower
1834 		 * bits are zero, we look at the next level as is. If not we
1835 		 * need to advance it by one because that's going to be the
1836 		 * next expiring bucket in that level. base->clk is the next
1837 		 * expiring jiffie. So in case of:
1838 		 *
1839 		 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1840 		 *  0    0    0    0    0    0
1841 		 *
1842 		 * we have to look at all levels @index 0. With
1843 		 *
1844 		 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1845 		 *  0    0    0    0    0    2
1846 		 *
1847 		 * LVL0 has the next expiring bucket @index 2. The upper
1848 		 * levels have the next expiring bucket @index 1.
1849 		 *
1850 		 * In case that the propagation wraps the next level the same
1851 		 * rules apply:
1852 		 *
1853 		 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1854 		 *  0    0    0    0    F    2
1855 		 *
1856 		 * So after looking at LVL0 we get:
1857 		 *
1858 		 * LVL5 LVL4 LVL3 LVL2 LVL1
1859 		 *  0    0    0    1    0
1860 		 *
1861 		 * So no propagation from LVL1 to LVL2 because that happened
1862 		 * with the add already, but then we need to propagate further
1863 		 * from LVL2 to LVL3.
1864 		 *
1865 		 * So the simple check whether the lower bits of the current
1866 		 * level are 0 or not is sufficient for all cases.
1867 		 */
1868 		adj = lvl_clk ? 1 : 0;
1869 		clk >>= LVL_CLK_SHIFT;
1870 		clk += adj;
1871 	}
1872 
1873 	base->next_expiry_recalc = false;
1874 	base->timers_pending = !(next == base->clk + NEXT_TIMER_MAX_DELTA);
1875 
1876 	return next;
1877 }
1878 
1879 #ifdef CONFIG_NO_HZ_COMMON
1880 /*
1881  * Check, if the next hrtimer event is before the next timer wheel
1882  * event:
1883  */
cmp_next_hrtimer_event(u64 basem,u64 expires)1884 static u64 cmp_next_hrtimer_event(u64 basem, u64 expires)
1885 {
1886 	u64 nextevt = hrtimer_get_next_event();
1887 
1888 	/*
1889 	 * If high resolution timers are enabled
1890 	 * hrtimer_get_next_event() returns KTIME_MAX.
1891 	 */
1892 	if (expires <= nextevt)
1893 		return expires;
1894 
1895 	/*
1896 	 * If the next timer is already expired, return the tick base
1897 	 * time so the tick is fired immediately.
1898 	 */
1899 	if (nextevt <= basem)
1900 		return basem;
1901 
1902 	/*
1903 	 * Round up to the next jiffie. High resolution timers are
1904 	 * off, so the hrtimers are expired in the tick and we need to
1905 	 * make sure that this tick really expires the timer to avoid
1906 	 * a ping pong of the nohz stop code.
1907 	 *
1908 	 * Use DIV_ROUND_UP_ULL to prevent gcc calling __divdi3
1909 	 */
1910 	return DIV_ROUND_UP_ULL(nextevt, TICK_NSEC) * TICK_NSEC;
1911 }
1912 
1913 /**
1914  * get_next_timer_interrupt - return the time (clock mono) of the next timer
1915  * @basej:	base time jiffies
1916  * @basem:	base time clock monotonic
1917  *
1918  * Returns the tick aligned clock monotonic time of the next pending
1919  * timer or KTIME_MAX if no timer is pending.
1920  */
get_next_timer_interrupt(unsigned long basej,u64 basem)1921 u64 get_next_timer_interrupt(unsigned long basej, u64 basem)
1922 {
1923 	struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1924 	u64 expires = KTIME_MAX;
1925 	unsigned long nextevt;
1926 
1927 	/*
1928 	 * Pretend that there is no timer pending if the cpu is offline.
1929 	 * Possible pending timers will be migrated later to an active cpu.
1930 	 */
1931 	if (cpu_is_offline(smp_processor_id()))
1932 		return expires;
1933 
1934 	raw_spin_lock(&base->lock);
1935 	if (base->next_expiry_recalc)
1936 		base->next_expiry = __next_timer_interrupt(base);
1937 	nextevt = base->next_expiry;
1938 
1939 	/*
1940 	 * We have a fresh next event. Check whether we can forward the
1941 	 * base. We can only do that when @basej is past base->clk
1942 	 * otherwise we might rewind base->clk.
1943 	 */
1944 	if (time_after(basej, base->clk)) {
1945 		if (time_after(nextevt, basej))
1946 			base->clk = basej;
1947 		else if (time_after(nextevt, base->clk))
1948 			base->clk = nextevt;
1949 	}
1950 
1951 	if (time_before_eq(nextevt, basej)) {
1952 		expires = basem;
1953 		base->is_idle = false;
1954 	} else {
1955 		if (base->timers_pending)
1956 			expires = basem + (u64)(nextevt - basej) * TICK_NSEC;
1957 		/*
1958 		 * If we expect to sleep more than a tick, mark the base idle.
1959 		 * Also the tick is stopped so any added timer must forward
1960 		 * the base clk itself to keep granularity small. This idle
1961 		 * logic is only maintained for the BASE_STD base, deferrable
1962 		 * timers may still see large granularity skew (by design).
1963 		 */
1964 		if ((expires - basem) > TICK_NSEC)
1965 			base->is_idle = true;
1966 	}
1967 	raw_spin_unlock(&base->lock);
1968 
1969 	return cmp_next_hrtimer_event(basem, expires);
1970 }
1971 
1972 /**
1973  * timer_clear_idle - Clear the idle state of the timer base
1974  *
1975  * Called with interrupts disabled
1976  */
timer_clear_idle(void)1977 void timer_clear_idle(void)
1978 {
1979 	struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1980 
1981 	/*
1982 	 * We do this unlocked. The worst outcome is a remote enqueue sending
1983 	 * a pointless IPI, but taking the lock would just make the window for
1984 	 * sending the IPI a few instructions smaller for the cost of taking
1985 	 * the lock in the exit from idle path.
1986 	 */
1987 	base->is_idle = false;
1988 }
1989 #endif
1990 
1991 /**
1992  * __run_timers - run all expired timers (if any) on this CPU.
1993  * @base: the timer vector to be processed.
1994  */
__run_timers(struct timer_base * base)1995 static inline void __run_timers(struct timer_base *base)
1996 {
1997 	struct hlist_head heads[LVL_DEPTH];
1998 	int levels;
1999 
2000 	if (time_before(jiffies, base->next_expiry))
2001 		return;
2002 
2003 	timer_base_lock_expiry(base);
2004 	raw_spin_lock_irq(&base->lock);
2005 
2006 	while (time_after_eq(jiffies, base->clk) &&
2007 	       time_after_eq(jiffies, base->next_expiry)) {
2008 		levels = collect_expired_timers(base, heads);
2009 		/*
2010 		 * The two possible reasons for not finding any expired
2011 		 * timer at this clk are that all matching timers have been
2012 		 * dequeued or no timer has been queued since
2013 		 * base::next_expiry was set to base::clk +
2014 		 * NEXT_TIMER_MAX_DELTA.
2015 		 */
2016 		WARN_ON_ONCE(!levels && !base->next_expiry_recalc
2017 			     && base->timers_pending);
2018 		base->clk++;
2019 		base->next_expiry = __next_timer_interrupt(base);
2020 
2021 		while (levels--)
2022 			expire_timers(base, heads + levels);
2023 	}
2024 	raw_spin_unlock_irq(&base->lock);
2025 	timer_base_unlock_expiry(base);
2026 }
2027 
2028 /*
2029  * This function runs timers and the timer-tq in bottom half context.
2030  */
run_timer_softirq(struct softirq_action * h)2031 static __latent_entropy void run_timer_softirq(struct softirq_action *h)
2032 {
2033 	struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
2034 
2035 	__run_timers(base);
2036 	if (IS_ENABLED(CONFIG_NO_HZ_COMMON))
2037 		__run_timers(this_cpu_ptr(&timer_bases[BASE_DEF]));
2038 }
2039 
2040 /*
2041  * Called by the local, per-CPU timer interrupt on SMP.
2042  */
run_local_timers(void)2043 static void run_local_timers(void)
2044 {
2045 	struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
2046 
2047 	hrtimer_run_queues();
2048 	/* Raise the softirq only if required. */
2049 	if (time_before(jiffies, base->next_expiry)) {
2050 		if (!IS_ENABLED(CONFIG_NO_HZ_COMMON))
2051 			return;
2052 		/* CPU is awake, so check the deferrable base. */
2053 		base++;
2054 		if (time_before(jiffies, base->next_expiry))
2055 			return;
2056 	}
2057 	raise_softirq(TIMER_SOFTIRQ);
2058 }
2059 
2060 /*
2061  * Called from the timer interrupt handler to charge one tick to the current
2062  * process.  user_tick is 1 if the tick is user time, 0 for system.
2063  */
update_process_times(int user_tick)2064 void update_process_times(int user_tick)
2065 {
2066 	struct task_struct *p = current;
2067 
2068 	/* Note: this timer irq context must be accounted for as well. */
2069 	account_process_tick(p, user_tick);
2070 	run_local_timers();
2071 	rcu_sched_clock_irq(user_tick);
2072 #ifdef CONFIG_IRQ_WORK
2073 	if (in_irq())
2074 		irq_work_tick();
2075 #endif
2076 	scheduler_tick();
2077 	if (IS_ENABLED(CONFIG_POSIX_TIMERS))
2078 		run_posix_cpu_timers();
2079 }
2080 
2081 /*
2082  * Since schedule_timeout()'s timer is defined on the stack, it must store
2083  * the target task on the stack as well.
2084  */
2085 struct process_timer {
2086 	struct timer_list timer;
2087 	struct task_struct *task;
2088 };
2089 
process_timeout(struct timer_list * t)2090 static void process_timeout(struct timer_list *t)
2091 {
2092 	struct process_timer *timeout = from_timer(timeout, t, timer);
2093 
2094 	wake_up_process(timeout->task);
2095 }
2096 
2097 /**
2098  * schedule_timeout - sleep until timeout
2099  * @timeout: timeout value in jiffies
2100  *
2101  * Make the current task sleep until @timeout jiffies have elapsed.
2102  * The function behavior depends on the current task state
2103  * (see also set_current_state() description):
2104  *
2105  * %TASK_RUNNING - the scheduler is called, but the task does not sleep
2106  * at all. That happens because sched_submit_work() does nothing for
2107  * tasks in %TASK_RUNNING state.
2108  *
2109  * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
2110  * pass before the routine returns unless the current task is explicitly
2111  * woken up, (e.g. by wake_up_process()).
2112  *
2113  * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
2114  * delivered to the current task or the current task is explicitly woken
2115  * up.
2116  *
2117  * The current task state is guaranteed to be %TASK_RUNNING when this
2118  * routine returns.
2119  *
2120  * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
2121  * the CPU away without a bound on the timeout. In this case the return
2122  * value will be %MAX_SCHEDULE_TIMEOUT.
2123  *
2124  * Returns 0 when the timer has expired otherwise the remaining time in
2125  * jiffies will be returned. In all cases the return value is guaranteed
2126  * to be non-negative.
2127  */
schedule_timeout(signed long timeout)2128 signed long __sched schedule_timeout(signed long timeout)
2129 {
2130 	struct process_timer timer;
2131 	unsigned long expire;
2132 
2133 	switch (timeout)
2134 	{
2135 	case MAX_SCHEDULE_TIMEOUT:
2136 		/*
2137 		 * These two special cases are useful to be comfortable
2138 		 * in the caller. Nothing more. We could take
2139 		 * MAX_SCHEDULE_TIMEOUT from one of the negative value
2140 		 * but I' d like to return a valid offset (>=0) to allow
2141 		 * the caller to do everything it want with the retval.
2142 		 */
2143 		schedule();
2144 		goto out;
2145 	default:
2146 		/*
2147 		 * Another bit of PARANOID. Note that the retval will be
2148 		 * 0 since no piece of kernel is supposed to do a check
2149 		 * for a negative retval of schedule_timeout() (since it
2150 		 * should never happens anyway). You just have the printk()
2151 		 * that will tell you if something is gone wrong and where.
2152 		 */
2153 		if (timeout < 0) {
2154 			printk(KERN_ERR "schedule_timeout: wrong timeout "
2155 				"value %lx\n", timeout);
2156 			dump_stack();
2157 			__set_current_state(TASK_RUNNING);
2158 			goto out;
2159 		}
2160 	}
2161 
2162 	expire = timeout + jiffies;
2163 
2164 	timer.task = current;
2165 	timer_setup_on_stack(&timer.timer, process_timeout, 0);
2166 	__mod_timer(&timer.timer, expire, MOD_TIMER_NOTPENDING);
2167 	schedule();
2168 	del_timer_sync(&timer.timer);
2169 
2170 	/* Remove the timer from the object tracker */
2171 	destroy_timer_on_stack(&timer.timer);
2172 
2173 	timeout = expire - jiffies;
2174 
2175  out:
2176 	return timeout < 0 ? 0 : timeout;
2177 }
2178 EXPORT_SYMBOL(schedule_timeout);
2179 
2180 /*
2181  * We can use __set_current_state() here because schedule_timeout() calls
2182  * schedule() unconditionally.
2183  */
schedule_timeout_interruptible(signed long timeout)2184 signed long __sched schedule_timeout_interruptible(signed long timeout)
2185 {
2186 	__set_current_state(TASK_INTERRUPTIBLE);
2187 	return schedule_timeout(timeout);
2188 }
2189 EXPORT_SYMBOL(schedule_timeout_interruptible);
2190 
schedule_timeout_killable(signed long timeout)2191 signed long __sched schedule_timeout_killable(signed long timeout)
2192 {
2193 	__set_current_state(TASK_KILLABLE);
2194 	return schedule_timeout(timeout);
2195 }
2196 EXPORT_SYMBOL(schedule_timeout_killable);
2197 
schedule_timeout_uninterruptible(signed long timeout)2198 signed long __sched schedule_timeout_uninterruptible(signed long timeout)
2199 {
2200 	__set_current_state(TASK_UNINTERRUPTIBLE);
2201 	return schedule_timeout(timeout);
2202 }
2203 EXPORT_SYMBOL(schedule_timeout_uninterruptible);
2204 
2205 /*
2206  * Like schedule_timeout_uninterruptible(), except this task will not contribute
2207  * to load average.
2208  */
schedule_timeout_idle(signed long timeout)2209 signed long __sched schedule_timeout_idle(signed long timeout)
2210 {
2211 	__set_current_state(TASK_IDLE);
2212 	return schedule_timeout(timeout);
2213 }
2214 EXPORT_SYMBOL(schedule_timeout_idle);
2215 
2216 #ifdef CONFIG_HOTPLUG_CPU
migrate_timer_list(struct timer_base * new_base,struct hlist_head * head)2217 static void migrate_timer_list(struct timer_base *new_base, struct hlist_head *head)
2218 {
2219 	struct timer_list *timer;
2220 	int cpu = new_base->cpu;
2221 
2222 	while (!hlist_empty(head)) {
2223 		timer = hlist_entry(head->first, struct timer_list, entry);
2224 		detach_timer(timer, false);
2225 		timer->flags = (timer->flags & ~TIMER_BASEMASK) | cpu;
2226 		internal_add_timer(new_base, timer);
2227 	}
2228 }
2229 
timers_prepare_cpu(unsigned int cpu)2230 int timers_prepare_cpu(unsigned int cpu)
2231 {
2232 	struct timer_base *base;
2233 	int b;
2234 
2235 	for (b = 0; b < NR_BASES; b++) {
2236 		base = per_cpu_ptr(&timer_bases[b], cpu);
2237 		base->clk = jiffies;
2238 		base->next_expiry = base->clk + NEXT_TIMER_MAX_DELTA;
2239 		base->next_expiry_recalc = false;
2240 		base->timers_pending = false;
2241 		base->is_idle = false;
2242 	}
2243 	return 0;
2244 }
2245 
timers_dead_cpu(unsigned int cpu)2246 int timers_dead_cpu(unsigned int cpu)
2247 {
2248 	struct timer_base *old_base;
2249 	struct timer_base *new_base;
2250 	int b, i;
2251 
2252 	for (b = 0; b < NR_BASES; b++) {
2253 		old_base = per_cpu_ptr(&timer_bases[b], cpu);
2254 		new_base = get_cpu_ptr(&timer_bases[b]);
2255 		/*
2256 		 * The caller is globally serialized and nobody else
2257 		 * takes two locks at once, deadlock is not possible.
2258 		 */
2259 		raw_spin_lock_irq(&new_base->lock);
2260 		raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
2261 
2262 		/*
2263 		 * The current CPUs base clock might be stale. Update it
2264 		 * before moving the timers over.
2265 		 */
2266 		forward_timer_base(new_base);
2267 
2268 		WARN_ON_ONCE(old_base->running_timer);
2269 		old_base->running_timer = NULL;
2270 
2271 		for (i = 0; i < WHEEL_SIZE; i++)
2272 			migrate_timer_list(new_base, old_base->vectors + i);
2273 
2274 		raw_spin_unlock(&old_base->lock);
2275 		raw_spin_unlock_irq(&new_base->lock);
2276 		put_cpu_ptr(&timer_bases);
2277 	}
2278 	return 0;
2279 }
2280 
2281 #endif /* CONFIG_HOTPLUG_CPU */
2282 
init_timer_cpu(int cpu)2283 static void __init init_timer_cpu(int cpu)
2284 {
2285 	struct timer_base *base;
2286 	int i;
2287 
2288 	for (i = 0; i < NR_BASES; i++) {
2289 		base = per_cpu_ptr(&timer_bases[i], cpu);
2290 		base->cpu = cpu;
2291 		raw_spin_lock_init(&base->lock);
2292 		base->clk = jiffies;
2293 		base->next_expiry = base->clk + NEXT_TIMER_MAX_DELTA;
2294 		timer_base_init_expiry_lock(base);
2295 	}
2296 }
2297 
init_timer_cpus(void)2298 static void __init init_timer_cpus(void)
2299 {
2300 	int cpu;
2301 
2302 	for_each_possible_cpu(cpu)
2303 		init_timer_cpu(cpu);
2304 }
2305 
init_timers(void)2306 void __init init_timers(void)
2307 {
2308 	init_timer_cpus();
2309 	posix_cputimers_init_work();
2310 	open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
2311 }
2312 
2313 /**
2314  * msleep - sleep safely even with waitqueue interruptions
2315  * @msecs: Time in milliseconds to sleep for
2316  */
msleep(unsigned int msecs)2317 void msleep(unsigned int msecs)
2318 {
2319 	unsigned long timeout = msecs_to_jiffies(msecs) + 1;
2320 
2321 	while (timeout)
2322 		timeout = schedule_timeout_uninterruptible(timeout);
2323 }
2324 
2325 EXPORT_SYMBOL(msleep);
2326 
2327 /**
2328  * msleep_interruptible - sleep waiting for signals
2329  * @msecs: Time in milliseconds to sleep for
2330  */
msleep_interruptible(unsigned int msecs)2331 unsigned long msleep_interruptible(unsigned int msecs)
2332 {
2333 	unsigned long timeout = msecs_to_jiffies(msecs) + 1;
2334 
2335 	while (timeout && !signal_pending(current))
2336 		timeout = schedule_timeout_interruptible(timeout);
2337 	return jiffies_to_msecs(timeout);
2338 }
2339 
2340 EXPORT_SYMBOL(msleep_interruptible);
2341 
2342 /**
2343  * usleep_range_state - Sleep for an approximate time in a given state
2344  * @min:	Minimum time in usecs to sleep
2345  * @max:	Maximum time in usecs to sleep
2346  * @state:	State of the current task that will be while sleeping
2347  *
2348  * In non-atomic context where the exact wakeup time is flexible, use
2349  * usleep_range_state() instead of udelay().  The sleep improves responsiveness
2350  * by avoiding the CPU-hogging busy-wait of udelay(), and the range reduces
2351  * power usage by allowing hrtimers to take advantage of an already-
2352  * scheduled interrupt instead of scheduling a new one just for this sleep.
2353  */
usleep_range_state(unsigned long min,unsigned long max,unsigned int state)2354 void __sched usleep_range_state(unsigned long min, unsigned long max,
2355 				unsigned int state)
2356 {
2357 	ktime_t exp = ktime_add_us(ktime_get(), min);
2358 	u64 delta = (u64)(max - min) * NSEC_PER_USEC;
2359 
2360 	for (;;) {
2361 		__set_current_state(state);
2362 		/* Do not return before the requested sleep time has elapsed */
2363 		if (!schedule_hrtimeout_range(&exp, delta, HRTIMER_MODE_ABS))
2364 			break;
2365 	}
2366 }
2367 EXPORT_SYMBOL(usleep_range_state);
2368