Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0
2 : /*
3 : * Kernel internal timers
4 : *
5 : * Copyright (C) 1991, 1992 Linus Torvalds
6 : *
7 : * 1997-01-28 Modified by Finn Arne Gangstad to make timers scale better.
8 : *
9 : * 1997-09-10 Updated NTP code according to technical memorandum Jan '96
10 : * "A Kernel Model for Precision Timekeeping" by Dave Mills
11 : * 1998-12-24 Fixed a xtime SMP race (we need the xtime_lock rw spinlock to
12 : * serialize accesses to xtime/lost_ticks).
13 : * Copyright (C) 1998 Andrea Arcangeli
14 : * 1999-03-10 Improved NTP compatibility by Ulrich Windl
15 : * 2002-05-31 Move sys_sysinfo here and make its locking sane, Robert Love
16 : * 2000-10-05 Implemented scalable SMP per-CPU timer handling.
17 : * Copyright (C) 2000, 2001, 2002 Ingo Molnar
18 : * Designed by David S. Miller, Alexey Kuznetsov and Ingo Molnar
19 : */
20 :
21 : #include <linux/kernel_stat.h>
22 : #include <linux/export.h>
23 : #include <linux/interrupt.h>
24 : #include <linux/percpu.h>
25 : #include <linux/init.h>
26 : #include <linux/mm.h>
27 : #include <linux/swap.h>
28 : #include <linux/pid_namespace.h>
29 : #include <linux/notifier.h>
30 : #include <linux/thread_info.h>
31 : #include <linux/time.h>
32 : #include <linux/jiffies.h>
33 : #include <linux/posix-timers.h>
34 : #include <linux/cpu.h>
35 : #include <linux/syscalls.h>
36 : #include <linux/delay.h>
37 : #include <linux/tick.h>
38 : #include <linux/kallsyms.h>
39 : #include <linux/irq_work.h>
40 : #include <linux/sched/signal.h>
41 : #include <linux/sched/sysctl.h>
42 : #include <linux/sched/nohz.h>
43 : #include <linux/sched/debug.h>
44 : #include <linux/slab.h>
45 : #include <linux/compat.h>
46 : #include <linux/random.h>
47 :
48 : #include <linux/uaccess.h>
49 : #include <asm/unistd.h>
50 : #include <asm/div64.h>
51 : #include <asm/timex.h>
52 : #include <asm/io.h>
53 :
54 : #include "tick-internal.h"
55 :
56 : #define CREATE_TRACE_POINTS
57 : #include <trace/events/timer.h>
58 :
59 : __visible u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
60 :
61 : EXPORT_SYMBOL(jiffies_64);
62 :
63 : /*
64 : * The timer wheel has LVL_DEPTH array levels. Each level provides an array of
65 : * LVL_SIZE buckets. Each level is driven by its own clock and therefor each
66 : * level has a different granularity.
67 : *
68 : * The level granularity is: LVL_CLK_DIV ^ lvl
69 : * The level clock frequency is: HZ / (LVL_CLK_DIV ^ level)
70 : *
71 : * The array level of a newly armed timer depends on the relative expiry
72 : * time. The farther the expiry time is away the higher the array level and
73 : * therefor the granularity becomes.
74 : *
75 : * Contrary to the original timer wheel implementation, which aims for 'exact'
76 : * expiry of the timers, this implementation removes the need for recascading
77 : * the timers into the lower array levels. The previous 'classic' timer wheel
78 : * implementation of the kernel already violated the 'exact' expiry by adding
79 : * slack to the expiry time to provide batched expiration. The granularity
80 : * levels provide implicit batching.
81 : *
82 : * This is an optimization of the original timer wheel implementation for the
83 : * majority of the timer wheel use cases: timeouts. The vast majority of
84 : * timeout timers (networking, disk I/O ...) are canceled before expiry. If
85 : * the timeout expires it indicates that normal operation is disturbed, so it
86 : * does not matter much whether the timeout comes with a slight delay.
87 : *
88 : * The only exception to this are networking timers with a small expiry
89 : * time. They rely on the granularity. Those fit into the first wheel level,
90 : * which has HZ granularity.
91 : *
92 : * We don't have cascading anymore. timers with a expiry time above the
93 : * capacity of the last wheel level are force expired at the maximum timeout
94 : * value of the last wheel level. From data sampling we know that the maximum
95 : * value observed is 5 days (network connection tracking), so this should not
96 : * be an issue.
97 : *
98 : * The currently chosen array constants values are a good compromise between
99 : * array size and granularity.
100 : *
101 : * This results in the following granularity and range levels:
102 : *
103 : * HZ 1000 steps
104 : * Level Offset Granularity Range
105 : * 0 0 1 ms 0 ms - 63 ms
106 : * 1 64 8 ms 64 ms - 511 ms
107 : * 2 128 64 ms 512 ms - 4095 ms (512ms - ~4s)
108 : * 3 192 512 ms 4096 ms - 32767 ms (~4s - ~32s)
109 : * 4 256 4096 ms (~4s) 32768 ms - 262143 ms (~32s - ~4m)
110 : * 5 320 32768 ms (~32s) 262144 ms - 2097151 ms (~4m - ~34m)
111 : * 6 384 262144 ms (~4m) 2097152 ms - 16777215 ms (~34m - ~4h)
112 : * 7 448 2097152 ms (~34m) 16777216 ms - 134217727 ms (~4h - ~1d)
113 : * 8 512 16777216 ms (~4h) 134217728 ms - 1073741822 ms (~1d - ~12d)
114 : *
115 : * HZ 300
116 : * Level Offset Granularity Range
117 : * 0 0 3 ms 0 ms - 210 ms
118 : * 1 64 26 ms 213 ms - 1703 ms (213ms - ~1s)
119 : * 2 128 213 ms 1706 ms - 13650 ms (~1s - ~13s)
120 : * 3 192 1706 ms (~1s) 13653 ms - 109223 ms (~13s - ~1m)
121 : * 4 256 13653 ms (~13s) 109226 ms - 873810 ms (~1m - ~14m)
122 : * 5 320 109226 ms (~1m) 873813 ms - 6990503 ms (~14m - ~1h)
123 : * 6 384 873813 ms (~14m) 6990506 ms - 55924050 ms (~1h - ~15h)
124 : * 7 448 6990506 ms (~1h) 55924053 ms - 447392423 ms (~15h - ~5d)
125 : * 8 512 55924053 ms (~15h) 447392426 ms - 3579139406 ms (~5d - ~41d)
126 : *
127 : * HZ 250
128 : * Level Offset Granularity Range
129 : * 0 0 4 ms 0 ms - 255 ms
130 : * 1 64 32 ms 256 ms - 2047 ms (256ms - ~2s)
131 : * 2 128 256 ms 2048 ms - 16383 ms (~2s - ~16s)
132 : * 3 192 2048 ms (~2s) 16384 ms - 131071 ms (~16s - ~2m)
133 : * 4 256 16384 ms (~16s) 131072 ms - 1048575 ms (~2m - ~17m)
134 : * 5 320 131072 ms (~2m) 1048576 ms - 8388607 ms (~17m - ~2h)
135 : * 6 384 1048576 ms (~17m) 8388608 ms - 67108863 ms (~2h - ~18h)
136 : * 7 448 8388608 ms (~2h) 67108864 ms - 536870911 ms (~18h - ~6d)
137 : * 8 512 67108864 ms (~18h) 536870912 ms - 4294967288 ms (~6d - ~49d)
138 : *
139 : * HZ 100
140 : * Level Offset Granularity Range
141 : * 0 0 10 ms 0 ms - 630 ms
142 : * 1 64 80 ms 640 ms - 5110 ms (640ms - ~5s)
143 : * 2 128 640 ms 5120 ms - 40950 ms (~5s - ~40s)
144 : * 3 192 5120 ms (~5s) 40960 ms - 327670 ms (~40s - ~5m)
145 : * 4 256 40960 ms (~40s) 327680 ms - 2621430 ms (~5m - ~43m)
146 : * 5 320 327680 ms (~5m) 2621440 ms - 20971510 ms (~43m - ~5h)
147 : * 6 384 2621440 ms (~43m) 20971520 ms - 167772150 ms (~5h - ~1d)
148 : * 7 448 20971520 ms (~5h) 167772160 ms - 1342177270 ms (~1d - ~15d)
149 : */
150 :
151 : /* Clock divisor for the next level */
152 : #define LVL_CLK_SHIFT 3
153 : #define LVL_CLK_DIV (1UL << LVL_CLK_SHIFT)
154 : #define LVL_CLK_MASK (LVL_CLK_DIV - 1)
155 : #define LVL_SHIFT(n) ((n) * LVL_CLK_SHIFT)
156 : #define LVL_GRAN(n) (1UL << LVL_SHIFT(n))
157 :
158 : /*
159 : * The time start value for each level to select the bucket at enqueue
160 : * time. We start from the last possible delta of the previous level
161 : * so that we can later add an extra LVL_GRAN(n) to n (see calc_index()).
162 : */
163 : #define LVL_START(n) ((LVL_SIZE - 1) << (((n) - 1) * LVL_CLK_SHIFT))
164 :
165 : /* Size of each clock level */
166 : #define LVL_BITS 6
167 : #define LVL_SIZE (1UL << LVL_BITS)
168 : #define LVL_MASK (LVL_SIZE - 1)
169 : #define LVL_OFFS(n) ((n) * LVL_SIZE)
170 :
171 : /* Level depth */
172 : #if HZ > 100
173 : # define LVL_DEPTH 9
174 : # else
175 : # define LVL_DEPTH 8
176 : #endif
177 :
178 : /* The cutoff (max. capacity of the wheel) */
179 : #define WHEEL_TIMEOUT_CUTOFF (LVL_START(LVL_DEPTH))
180 : #define WHEEL_TIMEOUT_MAX (WHEEL_TIMEOUT_CUTOFF - LVL_GRAN(LVL_DEPTH - 1))
181 :
182 : /*
183 : * The resulting wheel size. If NOHZ is configured we allocate two
184 : * wheels so we have a separate storage for the deferrable timers.
185 : */
186 : #define WHEEL_SIZE (LVL_SIZE * LVL_DEPTH)
187 :
188 : #ifdef CONFIG_NO_HZ_COMMON
189 : # define NR_BASES 2
190 : # define BASE_STD 0
191 : # define BASE_DEF 1
192 : #else
193 : # define NR_BASES 1
194 : # define BASE_STD 0
195 : # define BASE_DEF 0
196 : #endif
197 :
198 : struct timer_base {
199 : raw_spinlock_t lock;
200 : struct timer_list *running_timer;
201 : #ifdef CONFIG_PREEMPT_RT
202 : spinlock_t expiry_lock;
203 : atomic_t timer_waiters;
204 : #endif
205 : unsigned long clk;
206 : unsigned long next_expiry;
207 : unsigned int cpu;
208 : bool next_expiry_recalc;
209 : bool is_idle;
210 : DECLARE_BITMAP(pending_map, WHEEL_SIZE);
211 : struct hlist_head vectors[WHEEL_SIZE];
212 : } ____cacheline_aligned;
213 :
214 : static DEFINE_PER_CPU(struct timer_base, timer_bases[NR_BASES]);
215 :
216 : #ifdef CONFIG_NO_HZ_COMMON
217 :
218 : static DEFINE_STATIC_KEY_FALSE(timers_nohz_active);
219 : static DEFINE_MUTEX(timer_keys_mutex);
220 :
221 : static void timer_update_keys(struct work_struct *work);
222 : static DECLARE_WORK(timer_update_work, timer_update_keys);
223 :
224 : #ifdef CONFIG_SMP
225 : unsigned int sysctl_timer_migration = 1;
226 :
227 : DEFINE_STATIC_KEY_FALSE(timers_migration_enabled);
228 :
229 1 : static void timers_update_migration(void)
230 : {
231 1 : if (sysctl_timer_migration && tick_nohz_active)
232 1 : static_branch_enable(&timers_migration_enabled);
233 : else
234 0 : static_branch_disable(&timers_migration_enabled);
235 1 : }
236 : #else
237 : static inline void timers_update_migration(void) { }
238 : #endif /* !CONFIG_SMP */
239 :
240 1 : static void timer_update_keys(struct work_struct *work)
241 : {
242 1 : mutex_lock(&timer_keys_mutex);
243 1 : timers_update_migration();
244 1 : static_branch_enable(&timers_nohz_active);
245 1 : mutex_unlock(&timer_keys_mutex);
246 1 : }
247 :
248 1 : void timers_update_nohz(void)
249 : {
250 1 : schedule_work(&timer_update_work);
251 1 : }
252 :
253 0 : int timer_migration_handler(struct ctl_table *table, int write,
254 : void *buffer, size_t *lenp, loff_t *ppos)
255 : {
256 0 : int ret;
257 :
258 0 : mutex_lock(&timer_keys_mutex);
259 0 : ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
260 0 : if (!ret && write)
261 0 : timers_update_migration();
262 0 : mutex_unlock(&timer_keys_mutex);
263 0 : return ret;
264 : }
265 :
266 3661 : static inline bool is_timers_nohz_active(void)
267 : {
268 3661 : return static_branch_unlikely(&timers_nohz_active);
269 : }
270 : #else
271 : static inline bool is_timers_nohz_active(void) { return false; }
272 : #endif /* NO_HZ_COMMON */
273 :
274 220 : static unsigned long round_jiffies_common(unsigned long j, int cpu,
275 : bool force_up)
276 : {
277 220 : int rem;
278 220 : unsigned long original = j;
279 :
280 : /*
281 : * We don't want all cpus firing their timers at once hitting the
282 : * same lock or cachelines, so we skew each extra cpu with an extra
283 : * 3 jiffies. This 3 jiffies came originally from the mm/ code which
284 : * already did this.
285 : * The skew is done by adding 3*cpunr, then round, then subtract this
286 : * extra offset again.
287 : */
288 220 : j += cpu * 3;
289 :
290 220 : rem = j % HZ;
291 :
292 : /*
293 : * If the target jiffie is just after a whole second (which can happen
294 : * due to delays of the timer irq, long irq off times etc etc) then
295 : * we should round down to the whole second, not up. Use 1/4th second
296 : * as cutoff for this rounding as an extreme upper bound for this.
297 : * But never round down if @force_up is set.
298 : */
299 220 : if (rem < HZ/4 && !force_up) /* round down */
300 157 : j = j - rem;
301 : else /* round up */
302 63 : j = j - rem + HZ;
303 :
304 : /* now that we have rounded, subtract the extra skew again */
305 220 : j -= cpu * 3;
306 :
307 : /*
308 : * Make sure j is still in the future. Otherwise return the
309 : * unmodified value.
310 : */
311 220 : return time_is_after_jiffies(j) ? j : original;
312 : }
313 :
314 : /**
315 : * __round_jiffies - function to round jiffies to a full second
316 : * @j: the time in (absolute) jiffies that should be rounded
317 : * @cpu: the processor number on which the timeout will happen
318 : *
319 : * __round_jiffies() rounds an absolute time in the future (in jiffies)
320 : * up or down to (approximately) full seconds. This is useful for timers
321 : * for which the exact time they fire does not matter too much, as long as
322 : * they fire approximately every X seconds.
323 : *
324 : * By rounding these timers to whole seconds, all such timers will fire
325 : * at the same time, rather than at various times spread out. The goal
326 : * of this is to have the CPU wake up less, which saves power.
327 : *
328 : * The exact rounding is skewed for each processor to avoid all
329 : * processors firing at the exact same time, which could lead
330 : * to lock contention or spurious cache line bouncing.
331 : *
332 : * The return value is the rounded version of the @j parameter.
333 : */
334 0 : unsigned long __round_jiffies(unsigned long j, int cpu)
335 : {
336 0 : return round_jiffies_common(j, cpu, false);
337 : }
338 : EXPORT_SYMBOL_GPL(__round_jiffies);
339 :
340 : /**
341 : * __round_jiffies_relative - function to round jiffies to a full second
342 : * @j: the time in (relative) jiffies that should be rounded
343 : * @cpu: the processor number on which the timeout will happen
344 : *
345 : * __round_jiffies_relative() rounds a time delta in the future (in jiffies)
346 : * up or down to (approximately) full seconds. This is useful for timers
347 : * for which the exact time they fire does not matter too much, as long as
348 : * they fire approximately every X seconds.
349 : *
350 : * By rounding these timers to whole seconds, all such timers will fire
351 : * at the same time, rather than at various times spread out. The goal
352 : * of this is to have the CPU wake up less, which saves power.
353 : *
354 : * The exact rounding is skewed for each processor to avoid all
355 : * processors firing at the exact same time, which could lead
356 : * to lock contention or spurious cache line bouncing.
357 : *
358 : * The return value is the rounded version of the @j parameter.
359 : */
360 158 : unsigned long __round_jiffies_relative(unsigned long j, int cpu)
361 : {
362 158 : unsigned long j0 = jiffies;
363 :
364 : /* Use j0 because jiffies might change while we run */
365 158 : return round_jiffies_common(j + j0, cpu, false) - j0;
366 : }
367 : EXPORT_SYMBOL_GPL(__round_jiffies_relative);
368 :
369 : /**
370 : * round_jiffies - function to round jiffies to a full second
371 : * @j: the time in (absolute) jiffies that should be rounded
372 : *
373 : * round_jiffies() rounds an absolute time in the future (in jiffies)
374 : * up or down to (approximately) full seconds. This is useful for timers
375 : * for which the exact time they fire does not matter too much, as long as
376 : * they fire approximately every X seconds.
377 : *
378 : * By rounding these timers to whole seconds, all such timers will fire
379 : * at the same time, rather than at various times spread out. The goal
380 : * of this is to have the CPU wake up less, which saves power.
381 : *
382 : * The return value is the rounded version of the @j parameter.
383 : */
384 1 : unsigned long round_jiffies(unsigned long j)
385 : {
386 1 : return round_jiffies_common(j, raw_smp_processor_id(), false);
387 : }
388 : EXPORT_SYMBOL_GPL(round_jiffies);
389 :
390 : /**
391 : * round_jiffies_relative - function to round jiffies to a full second
392 : * @j: the time in (relative) jiffies that should be rounded
393 : *
394 : * round_jiffies_relative() rounds a time delta in the future (in jiffies)
395 : * up or down to (approximately) full seconds. This is useful for timers
396 : * for which the exact time they fire does not matter too much, as long as
397 : * they fire approximately every X seconds.
398 : *
399 : * By rounding these timers to whole seconds, all such timers will fire
400 : * at the same time, rather than at various times spread out. The goal
401 : * of this is to have the CPU wake up less, which saves power.
402 : *
403 : * The return value is the rounded version of the @j parameter.
404 : */
405 158 : unsigned long round_jiffies_relative(unsigned long j)
406 : {
407 158 : return __round_jiffies_relative(j, raw_smp_processor_id());
408 : }
409 : EXPORT_SYMBOL_GPL(round_jiffies_relative);
410 :
411 : /**
412 : * __round_jiffies_up - function to round jiffies up to a full second
413 : * @j: the time in (absolute) jiffies that should be rounded
414 : * @cpu: the processor number on which the timeout will happen
415 : *
416 : * This is the same as __round_jiffies() except that it will never
417 : * round down. This is useful for timeouts for which the exact time
418 : * of firing does not matter too much, as long as they don't fire too
419 : * early.
420 : */
421 0 : unsigned long __round_jiffies_up(unsigned long j, int cpu)
422 : {
423 0 : return round_jiffies_common(j, cpu, true);
424 : }
425 : EXPORT_SYMBOL_GPL(__round_jiffies_up);
426 :
427 : /**
428 : * __round_jiffies_up_relative - function to round jiffies up to a full second
429 : * @j: the time in (relative) jiffies that should be rounded
430 : * @cpu: the processor number on which the timeout will happen
431 : *
432 : * This is the same as __round_jiffies_relative() except that it will never
433 : * round down. This is useful for timeouts for which the exact time
434 : * of firing does not matter too much, as long as they don't fire too
435 : * early.
436 : */
437 0 : unsigned long __round_jiffies_up_relative(unsigned long j, int cpu)
438 : {
439 0 : unsigned long j0 = jiffies;
440 :
441 : /* Use j0 because jiffies might change while we run */
442 0 : return round_jiffies_common(j + j0, cpu, true) - j0;
443 : }
444 : EXPORT_SYMBOL_GPL(__round_jiffies_up_relative);
445 :
446 : /**
447 : * round_jiffies_up - function to round jiffies up to a full second
448 : * @j: the time in (absolute) jiffies that should be rounded
449 : *
450 : * This is the same as round_jiffies() except that it will never
451 : * round down. This is useful for timeouts for which the exact time
452 : * of firing does not matter too much, as long as they don't fire too
453 : * early.
454 : */
455 61 : unsigned long round_jiffies_up(unsigned long j)
456 : {
457 61 : return round_jiffies_common(j, raw_smp_processor_id(), true);
458 : }
459 : EXPORT_SYMBOL_GPL(round_jiffies_up);
460 :
461 : /**
462 : * round_jiffies_up_relative - function to round jiffies up to a full second
463 : * @j: the time in (relative) jiffies that should be rounded
464 : *
465 : * This is the same as round_jiffies_relative() except that it will never
466 : * round down. This is useful for timeouts for which the exact time
467 : * of firing does not matter too much, as long as they don't fire too
468 : * early.
469 : */
470 0 : unsigned long round_jiffies_up_relative(unsigned long j)
471 : {
472 0 : return __round_jiffies_up_relative(j, raw_smp_processor_id());
473 : }
474 : EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
475 :
476 :
477 10220 : static inline unsigned int timer_get_idx(struct timer_list *timer)
478 : {
479 10220 : return (timer->flags & TIMER_ARRAYMASK) >> TIMER_ARRAYSHIFT;
480 : }
481 :
482 5544 : static inline void timer_set_idx(struct timer_list *timer, unsigned int idx)
483 : {
484 5544 : timer->flags = (timer->flags & ~TIMER_ARRAYMASK) |
485 5544 : idx << TIMER_ARRAYSHIFT;
486 : }
487 :
488 : /*
489 : * Helper function to calculate the array index for a given expiry
490 : * time.
491 : */
492 5547 : static inline unsigned calc_index(unsigned long expires, unsigned lvl,
493 : unsigned long *bucket_expiry)
494 : {
495 :
496 : /*
497 : * The timer wheel has to guarantee that a timer does not fire
498 : * early. Early expiry can happen due to:
499 : * - Timer is armed at the edge of a tick
500 : * - Truncation of the expiry time in the outer wheel levels
501 : *
502 : * Round up with level granularity to prevent this.
503 : */
504 5547 : expires = (expires + LVL_GRAN(lvl)) >> LVL_SHIFT(lvl);
505 5547 : *bucket_expiry = expires << LVL_SHIFT(lvl);
506 5547 : return LVL_OFFS(lvl) + (expires & LVL_MASK);
507 : }
508 :
509 5547 : static int calc_wheel_index(unsigned long expires, unsigned long clk,
510 : unsigned long *bucket_expiry)
511 : {
512 5547 : unsigned long delta = expires - clk;
513 5547 : unsigned int idx;
514 :
515 5547 : if (delta < LVL_START(1)) {
516 4833 : idx = calc_index(expires, 0, bucket_expiry);
517 714 : } else if (delta < LVL_START(2)) {
518 303 : idx = calc_index(expires, 1, bucket_expiry);
519 411 : } else if (delta < LVL_START(3)) {
520 83 : idx = calc_index(expires, 2, bucket_expiry);
521 328 : } else if (delta < LVL_START(4)) {
522 320 : idx = calc_index(expires, 3, bucket_expiry);
523 8 : } else if (delta < LVL_START(5)) {
524 3 : idx = calc_index(expires, 4, bucket_expiry);
525 5 : } else if (delta < LVL_START(6)) {
526 4 : idx = calc_index(expires, 5, bucket_expiry);
527 1 : } else if (delta < LVL_START(7)) {
528 1 : idx = calc_index(expires, 6, bucket_expiry);
529 0 : } else if (LVL_DEPTH > 8 && delta < LVL_START(8)) {
530 0 : idx = calc_index(expires, 7, bucket_expiry);
531 0 : } else if ((long) delta < 0) {
532 0 : idx = clk & LVL_MASK;
533 0 : *bucket_expiry = clk;
534 : } else {
535 : /*
536 : * Force expire obscene large timeouts to expire at the
537 : * capacity limit of the wheel.
538 : */
539 0 : if (delta >= WHEEL_TIMEOUT_CUTOFF)
540 0 : expires = clk + WHEEL_TIMEOUT_MAX;
541 :
542 0 : idx = calc_index(expires, LVL_DEPTH - 1, bucket_expiry);
543 : }
544 5547 : return idx;
545 : }
546 :
547 : static void
548 3660 : trigger_dyntick_cpu(struct timer_base *base, struct timer_list *timer)
549 : {
550 3660 : if (!is_timers_nohz_active())
551 : return;
552 :
553 : /*
554 : * TODO: This wants some optimizing similar to the code below, but we
555 : * will do that when we switch from push to pull for deferrable timers.
556 : */
557 3647 : if (timer->flags & TIMER_DEFERRABLE) {
558 3660 : if (tick_nohz_full_cpu(base->cpu))
559 : wake_up_nohz_cpu(base->cpu);
560 : return;
561 : }
562 :
563 : /*
564 : * We might have to IPI the remote CPU if the base is idle and the
565 : * timer is not deferrable. If the other CPU is on the way to idle
566 : * then it can't set base->is_idle as we hold the base lock:
567 : */
568 3529 : if (base->is_idle)
569 10 : wake_up_nohz_cpu(base->cpu);
570 : }
571 :
572 : /*
573 : * Enqueue the timer into the hash bucket, mark it pending in
574 : * the bitmap, store the index in the timer flags then wake up
575 : * the target CPU if needed.
576 : */
577 5544 : static void enqueue_timer(struct timer_base *base, struct timer_list *timer,
578 : unsigned int idx, unsigned long bucket_expiry)
579 : {
580 :
581 5544 : hlist_add_head(&timer->entry, base->vectors + idx);
582 5544 : __set_bit(idx, base->pending_map);
583 5544 : timer_set_idx(timer, idx);
584 :
585 5544 : trace_timer_start(timer, timer->expires, timer->flags);
586 :
587 : /*
588 : * Check whether this is the new first expiring timer. The
589 : * effective expiry time of the timer is required here
590 : * (bucket_expiry) instead of timer->expires.
591 : */
592 5544 : if (time_before(bucket_expiry, base->next_expiry)) {
593 : /*
594 : * Set the next expiry time and kick the CPU so it
595 : * can reevaluate the wheel:
596 : */
597 3660 : base->next_expiry = bucket_expiry;
598 3660 : base->next_expiry_recalc = false;
599 3660 : trigger_dyntick_cpu(base, timer);
600 : }
601 5544 : }
602 :
603 5382 : static void internal_add_timer(struct timer_base *base, struct timer_list *timer)
604 : {
605 5382 : unsigned long bucket_expiry;
606 5382 : unsigned int idx;
607 :
608 5382 : idx = calc_wheel_index(timer->expires, base->clk, &bucket_expiry);
609 5382 : enqueue_timer(base, timer, idx, bucket_expiry);
610 5382 : }
611 :
612 : #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
613 :
614 : static const struct debug_obj_descr timer_debug_descr;
615 :
616 : static void *timer_debug_hint(void *addr)
617 : {
618 : return ((struct timer_list *) addr)->function;
619 : }
620 :
621 : static bool timer_is_static_object(void *addr)
622 : {
623 : struct timer_list *timer = addr;
624 :
625 : return (timer->entry.pprev == NULL &&
626 : timer->entry.next == TIMER_ENTRY_STATIC);
627 : }
628 :
629 : /*
630 : * fixup_init is called when:
631 : * - an active object is initialized
632 : */
633 : static bool timer_fixup_init(void *addr, enum debug_obj_state state)
634 : {
635 : struct timer_list *timer = addr;
636 :
637 : switch (state) {
638 : case ODEBUG_STATE_ACTIVE:
639 : del_timer_sync(timer);
640 : debug_object_init(timer, &timer_debug_descr);
641 : return true;
642 : default:
643 : return false;
644 : }
645 : }
646 :
647 : /* Stub timer callback for improperly used timers. */
648 : static void stub_timer(struct timer_list *unused)
649 : {
650 : WARN_ON(1);
651 : }
652 :
653 : /*
654 : * fixup_activate is called when:
655 : * - an active object is activated
656 : * - an unknown non-static object is activated
657 : */
658 : static bool timer_fixup_activate(void *addr, enum debug_obj_state state)
659 : {
660 : struct timer_list *timer = addr;
661 :
662 : switch (state) {
663 : case ODEBUG_STATE_NOTAVAILABLE:
664 : timer_setup(timer, stub_timer, 0);
665 : return true;
666 :
667 : case ODEBUG_STATE_ACTIVE:
668 : WARN_ON(1);
669 : fallthrough;
670 : default:
671 : return false;
672 : }
673 : }
674 :
675 : /*
676 : * fixup_free is called when:
677 : * - an active object is freed
678 : */
679 : static bool timer_fixup_free(void *addr, enum debug_obj_state state)
680 : {
681 : struct timer_list *timer = addr;
682 :
683 : switch (state) {
684 : case ODEBUG_STATE_ACTIVE:
685 : del_timer_sync(timer);
686 : debug_object_free(timer, &timer_debug_descr);
687 : return true;
688 : default:
689 : return false;
690 : }
691 : }
692 :
693 : /*
694 : * fixup_assert_init is called when:
695 : * - an untracked/uninit-ed object is found
696 : */
697 : static bool timer_fixup_assert_init(void *addr, enum debug_obj_state state)
698 : {
699 : struct timer_list *timer = addr;
700 :
701 : switch (state) {
702 : case ODEBUG_STATE_NOTAVAILABLE:
703 : timer_setup(timer, stub_timer, 0);
704 : return true;
705 : default:
706 : return false;
707 : }
708 : }
709 :
710 : static const struct debug_obj_descr timer_debug_descr = {
711 : .name = "timer_list",
712 : .debug_hint = timer_debug_hint,
713 : .is_static_object = timer_is_static_object,
714 : .fixup_init = timer_fixup_init,
715 : .fixup_activate = timer_fixup_activate,
716 : .fixup_free = timer_fixup_free,
717 : .fixup_assert_init = timer_fixup_assert_init,
718 : };
719 :
720 : static inline void debug_timer_init(struct timer_list *timer)
721 : {
722 : debug_object_init(timer, &timer_debug_descr);
723 : }
724 :
725 : static inline void debug_timer_activate(struct timer_list *timer)
726 : {
727 : debug_object_activate(timer, &timer_debug_descr);
728 : }
729 :
730 : static inline void debug_timer_deactivate(struct timer_list *timer)
731 : {
732 : debug_object_deactivate(timer, &timer_debug_descr);
733 : }
734 :
735 : static inline void debug_timer_assert_init(struct timer_list *timer)
736 : {
737 : debug_object_assert_init(timer, &timer_debug_descr);
738 : }
739 :
740 : static void do_init_timer(struct timer_list *timer,
741 : void (*func)(struct timer_list *),
742 : unsigned int flags,
743 : const char *name, struct lock_class_key *key);
744 :
745 : void init_timer_on_stack_key(struct timer_list *timer,
746 : void (*func)(struct timer_list *),
747 : unsigned int flags,
748 : const char *name, struct lock_class_key *key)
749 : {
750 : debug_object_init_on_stack(timer, &timer_debug_descr);
751 : do_init_timer(timer, func, flags, name, key);
752 : }
753 : EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
754 :
755 : void destroy_timer_on_stack(struct timer_list *timer)
756 : {
757 : debug_object_free(timer, &timer_debug_descr);
758 : }
759 : EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
760 :
761 : #else
762 5606 : static inline void debug_timer_init(struct timer_list *timer) { }
763 5544 : static inline void debug_timer_activate(struct timer_list *timer) { }
764 5511 : static inline void debug_timer_deactivate(struct timer_list *timer) { }
765 4964 : static inline void debug_timer_assert_init(struct timer_list *timer) { }
766 : #endif
767 :
768 5606 : static inline void debug_init(struct timer_list *timer)
769 : {
770 5606 : debug_timer_init(timer);
771 5606 : trace_timer_init(timer);
772 : }
773 :
774 5511 : static inline void debug_deactivate(struct timer_list *timer)
775 : {
776 5511 : debug_timer_deactivate(timer);
777 5511 : trace_timer_cancel(timer);
778 : }
779 :
780 4964 : static inline void debug_assert_init(struct timer_list *timer)
781 : {
782 4964 : debug_timer_assert_init(timer);
783 : }
784 :
785 5606 : static void do_init_timer(struct timer_list *timer,
786 : void (*func)(struct timer_list *),
787 : unsigned int flags,
788 : const char *name, struct lock_class_key *key)
789 : {
790 5606 : timer->entry.pprev = NULL;
791 5606 : timer->function = func;
792 5606 : if (WARN_ON_ONCE(flags & ~TIMER_INIT_FLAGS))
793 0 : flags &= TIMER_INIT_FLAGS;
794 5606 : timer->flags = flags | raw_smp_processor_id();
795 5606 : lockdep_init_map(&timer->lockdep_map, name, key, 0);
796 5606 : }
797 :
798 : /**
799 : * init_timer_key - initialize a timer
800 : * @timer: the timer to be initialized
801 : * @func: timer callback function
802 : * @flags: timer flags
803 : * @name: name of the timer
804 : * @key: lockdep class key of the fake lock used for tracking timer
805 : * sync lock dependencies
806 : *
807 : * init_timer_key() must be done to a timer prior calling *any* of the
808 : * other timer functions.
809 : */
810 5606 : void init_timer_key(struct timer_list *timer,
811 : void (*func)(struct timer_list *), unsigned int flags,
812 : const char *name, struct lock_class_key *key)
813 : {
814 5606 : debug_init(timer);
815 5606 : do_init_timer(timer, func, flags, name, key);
816 1037 : }
817 : EXPORT_SYMBOL(init_timer_key);
818 :
819 5511 : static inline void detach_timer(struct timer_list *timer, bool clear_pending)
820 : {
821 5511 : struct hlist_node *entry = &timer->entry;
822 :
823 5511 : debug_deactivate(timer);
824 :
825 5515 : __hlist_del(entry);
826 5515 : if (clear_pending)
827 5350 : entry->pprev = NULL;
828 5515 : entry->next = LIST_POISON2;
829 5515 : }
830 :
831 10055 : static int detach_if_pending(struct timer_list *timer, struct timer_base *base,
832 : bool clear_pending)
833 : {
834 10055 : unsigned idx = timer_get_idx(timer);
835 :
836 10055 : if (!timer_pending(timer))
837 : return 0;
838 :
839 3260 : if (hlist_is_singular_node(&timer->entry, base->vectors + idx)) {
840 1613 : __clear_bit(idx, base->pending_map);
841 1613 : base->next_expiry_recalc = true;
842 : }
843 :
844 1630 : detach_timer(timer, clear_pending);
845 1630 : return 1;
846 : }
847 :
848 15773 : static inline struct timer_base *get_timer_cpu_base(u32 tflags, u32 cpu)
849 : {
850 15773 : struct timer_base *base = per_cpu_ptr(&timer_bases[BASE_STD], cpu);
851 :
852 : /*
853 : * If the timer is deferrable and NO_HZ_COMMON is set then we need
854 : * to use the deferrable base.
855 : */
856 15773 : if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE))
857 345 : base = per_cpu_ptr(&timer_bases[BASE_DEF], cpu);
858 15773 : return base;
859 : }
860 :
861 20 : static inline struct timer_base *get_timer_this_cpu_base(u32 tflags)
862 : {
863 40 : struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
864 :
865 : /*
866 : * If the timer is deferrable and NO_HZ_COMMON is set then we need
867 : * to use the deferrable base.
868 : */
869 20 : if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE))
870 1 : base = this_cpu_ptr(&timer_bases[BASE_DEF]);
871 20 : return base;
872 : }
873 :
874 10249 : static inline struct timer_base *get_timer_base(u32 tflags)
875 : {
876 10249 : return get_timer_cpu_base(tflags, tflags & TIMER_CPUMASK);
877 : }
878 :
879 : static inline struct timer_base *
880 5350 : get_target_base(struct timer_base *base, unsigned tflags)
881 : {
882 : #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
883 5350 : if (static_branch_likely(&timers_migration_enabled) &&
884 5334 : !(tflags & TIMER_PINNED))
885 5378 : return get_timer_cpu_base(tflags, get_nohz_timer_target());
886 : #endif
887 40 : return get_timer_this_cpu_base(tflags);
888 : }
889 :
890 5860 : static inline void forward_timer_base(struct timer_base *base)
891 : {
892 5860 : unsigned long jnow = READ_ONCE(jiffies);
893 :
894 : /*
895 : * No need to forward if we are close enough below jiffies.
896 : * Also while executing timers, base->clk is 1 offset ahead
897 : * of jiffies to avoid endless requeuing to current jffies.
898 : */
899 5860 : if ((long)(jnow - base->clk) < 1)
900 : return;
901 :
902 : /*
903 : * If the next expiry value is > jiffies, then we fast forward to
904 : * jiffies otherwise we forward to the next expiry value.
905 : */
906 1305 : if (time_after(base->next_expiry, jnow)) {
907 1123 : base->clk = jnow;
908 : } else {
909 364 : if (WARN_ON_ONCE(time_before(base->next_expiry, base->clk)))
910 : return;
911 182 : base->clk = base->next_expiry;
912 : }
913 : }
914 :
915 :
916 : /*
917 : * We are using hashed locking: Holding per_cpu(timer_bases[x]).lock means
918 : * that all timers which are tied to this base are locked, and the base itself
919 : * is locked too.
920 : *
921 : * So __run_timers/migrate_timers can safely modify all timers which could
922 : * be found in the base->vectors array.
923 : *
924 : * When a timer is migrating then the TIMER_MIGRATING flag is set and we need
925 : * to wait until the migration is done.
926 : */
927 10249 : static struct timer_base *lock_timer_base(struct timer_list *timer,
928 : unsigned long *flags)
929 : __acquires(timer->base->lock)
930 : {
931 10249 : for (;;) {
932 10249 : struct timer_base *base;
933 10249 : u32 tf;
934 :
935 : /*
936 : * We need to use READ_ONCE() here, otherwise the compiler
937 : * might re-read @tf between the check for TIMER_MIGRATING
938 : * and spin_lock().
939 : */
940 10249 : tf = READ_ONCE(timer->flags);
941 :
942 10249 : if (!(tf & TIMER_MIGRATING)) {
943 10249 : base = get_timer_base(tf);
944 10249 : raw_spin_lock_irqsave(&base->lock, *flags);
945 10249 : if (timer->flags == tf)
946 10249 : return base;
947 0 : raw_spin_unlock_irqrestore(&base->lock, *flags);
948 : }
949 0 : cpu_relax();
950 : }
951 : }
952 :
953 : #define MOD_TIMER_PENDING_ONLY 0x01
954 : #define MOD_TIMER_REDUCE 0x02
955 : #define MOD_TIMER_NOTPENDING 0x04
956 :
957 : static inline int
958 5534 : __mod_timer(struct timer_list *timer, unsigned long expires, unsigned int options)
959 : {
960 5534 : unsigned long clk = 0, flags, bucket_expiry;
961 5534 : struct timer_base *base, *new_base;
962 5534 : unsigned int idx = UINT_MAX;
963 5534 : int ret = 0;
964 :
965 5534 : BUG_ON(!timer->function);
966 :
967 : /*
968 : * This is a common optimization triggered by the networking code - if
969 : * the timer is re-modified to have the same timeout or ends up in the
970 : * same array bucket then just return:
971 : */
972 5534 : if (!(options & MOD_TIMER_NOTPENDING) && timer_pending(timer)) {
973 : /*
974 : * The downside of this optimization is that it can result in
975 : * larger granularity than you would get from adding a new
976 : * timer with this expiry.
977 : */
978 349 : long diff = timer->expires - expires;
979 :
980 349 : if (!diff)
981 : return 1;
982 165 : if (options & MOD_TIMER_REDUCE && diff <= 0)
983 : return 1;
984 :
985 : /*
986 : * We lock timer base and calculate the bucket index right
987 : * here. If the timer ends up in the same bucket, then we
988 : * just update the expiry time and avoid the whole
989 : * dequeue/enqueue dance.
990 : */
991 165 : base = lock_timer_base(timer, &flags);
992 165 : forward_timer_base(base);
993 :
994 165 : if (timer_pending(timer) && (options & MOD_TIMER_REDUCE) &&
995 0 : time_before_eq(timer->expires, expires)) {
996 0 : ret = 1;
997 0 : goto out_unlock;
998 : }
999 :
1000 165 : clk = base->clk;
1001 165 : idx = calc_wheel_index(expires, clk, &bucket_expiry);
1002 :
1003 : /*
1004 : * Retrieve and compare the array index of the pending
1005 : * timer. If it matches set the expiry to the new value so a
1006 : * subsequent call will exit in the expires check above.
1007 : */
1008 165 : if (idx == timer_get_idx(timer)) {
1009 0 : if (!(options & MOD_TIMER_REDUCE))
1010 0 : timer->expires = expires;
1011 0 : else if (time_after(timer->expires, expires))
1012 0 : timer->expires = expires;
1013 0 : ret = 1;
1014 0 : goto out_unlock;
1015 : }
1016 : } else {
1017 5185 : base = lock_timer_base(timer, &flags);
1018 5185 : forward_timer_base(base);
1019 : }
1020 :
1021 5350 : ret = detach_if_pending(timer, base, false);
1022 5350 : if (!ret && (options & MOD_TIMER_PENDING_ONLY))
1023 0 : goto out_unlock;
1024 :
1025 5350 : new_base = get_target_base(base, timer->flags);
1026 :
1027 5350 : if (base != new_base) {
1028 : /*
1029 : * We are trying to schedule the timer on the new base.
1030 : * However we can't change timer's base while it is running,
1031 : * otherwise del_timer_sync() can't detect that the timer's
1032 : * handler yet has not finished. This also guarantees that the
1033 : * timer is serialized wrt itself.
1034 : */
1035 318 : if (likely(base->running_timer != timer)) {
1036 : /* See the comment in lock_timer_base() */
1037 316 : timer->flags |= TIMER_MIGRATING;
1038 :
1039 316 : raw_spin_unlock(&base->lock);
1040 316 : base = new_base;
1041 316 : raw_spin_lock(&base->lock);
1042 316 : WRITE_ONCE(timer->flags,
1043 : (timer->flags & ~TIMER_BASEMASK) | base->cpu);
1044 316 : forward_timer_base(base);
1045 : }
1046 : }
1047 :
1048 5350 : debug_timer_activate(timer);
1049 :
1050 5350 : timer->expires = expires;
1051 : /*
1052 : * If 'idx' was calculated above and the base time did not advance
1053 : * between calculating 'idx' and possibly switching the base, only
1054 : * enqueue_timer() is required. Otherwise we need to (re)calculate
1055 : * the wheel index via internal_add_timer().
1056 : */
1057 5350 : if (idx != UINT_MAX && clk == base->clk)
1058 162 : enqueue_timer(base, timer, idx, bucket_expiry);
1059 : else
1060 5188 : internal_add_timer(base, timer);
1061 :
1062 5350 : out_unlock:
1063 5350 : raw_spin_unlock_irqrestore(&base->lock, flags);
1064 :
1065 5350 : return ret;
1066 : }
1067 :
1068 : /**
1069 : * mod_timer_pending - modify a pending timer's timeout
1070 : * @timer: the pending timer to be modified
1071 : * @expires: new timeout in jiffies
1072 : *
1073 : * mod_timer_pending() is the same for pending timers as mod_timer(),
1074 : * but will not re-activate and modify already deleted timers.
1075 : *
1076 : * It is useful for unserialized use of timers.
1077 : */
1078 0 : int mod_timer_pending(struct timer_list *timer, unsigned long expires)
1079 : {
1080 0 : return __mod_timer(timer, expires, MOD_TIMER_PENDING_ONLY);
1081 : }
1082 : EXPORT_SYMBOL(mod_timer_pending);
1083 :
1084 : /**
1085 : * mod_timer - modify a timer's timeout
1086 : * @timer: the timer to be modified
1087 : * @expires: new timeout in jiffies
1088 : *
1089 : * mod_timer() is a more efficient way to update the expire field of an
1090 : * active timer (if the timer is inactive it will be activated)
1091 : *
1092 : * mod_timer(timer, expires) is equivalent to:
1093 : *
1094 : * del_timer(timer); timer->expires = expires; add_timer(timer);
1095 : *
1096 : * Note that if there are multiple unserialized concurrent users of the
1097 : * same timer, then mod_timer() is the only safe way to modify the timeout,
1098 : * since add_timer() cannot modify an already running timer.
1099 : *
1100 : * The function returns whether it has modified a pending timer or not.
1101 : * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
1102 : * active timer returns 1.)
1103 : */
1104 411 : int mod_timer(struct timer_list *timer, unsigned long expires)
1105 : {
1106 411 : return __mod_timer(timer, expires, 0);
1107 : }
1108 : EXPORT_SYMBOL(mod_timer);
1109 :
1110 : /**
1111 : * timer_reduce - Modify a timer's timeout if it would reduce the timeout
1112 : * @timer: The timer to be modified
1113 : * @expires: New timeout in jiffies
1114 : *
1115 : * timer_reduce() is very similar to mod_timer(), except that it will only
1116 : * modify a running timer if that would reduce the expiration time (it will
1117 : * start a timer that isn't running).
1118 : */
1119 1 : int timer_reduce(struct timer_list *timer, unsigned long expires)
1120 : {
1121 1 : return __mod_timer(timer, expires, MOD_TIMER_REDUCE);
1122 : }
1123 : EXPORT_SYMBOL(timer_reduce);
1124 :
1125 : /**
1126 : * add_timer - start a timer
1127 : * @timer: the timer to be added
1128 : *
1129 : * The kernel will do a ->function(@timer) callback from the
1130 : * timer interrupt at the ->expires point in the future. The
1131 : * current time is 'jiffies'.
1132 : *
1133 : * The timer's ->expires, ->function fields must be set prior calling this
1134 : * function.
1135 : *
1136 : * Timers with an ->expires field in the past will be executed in the next
1137 : * timer tick.
1138 : */
1139 553 : void add_timer(struct timer_list *timer)
1140 : {
1141 553 : BUG_ON(timer_pending(timer));
1142 553 : __mod_timer(timer, timer->expires, MOD_TIMER_NOTPENDING);
1143 553 : }
1144 : EXPORT_SYMBOL(add_timer);
1145 :
1146 : /**
1147 : * add_timer_on - start a timer on a particular CPU
1148 : * @timer: the timer to be added
1149 : * @cpu: the CPU to start it on
1150 : *
1151 : * This is not very scalable on SMP. Double adds are not possible.
1152 : */
1153 194 : void add_timer_on(struct timer_list *timer, int cpu)
1154 : {
1155 194 : struct timer_base *new_base, *base;
1156 194 : unsigned long flags;
1157 :
1158 194 : BUG_ON(timer_pending(timer) || !timer->function);
1159 :
1160 194 : new_base = get_timer_cpu_base(timer->flags, cpu);
1161 :
1162 : /*
1163 : * If @timer was on a different CPU, it should be migrated with the
1164 : * old base locked to prevent other operations proceeding with the
1165 : * wrong base locked. See lock_timer_base().
1166 : */
1167 194 : base = lock_timer_base(timer, &flags);
1168 194 : if (base != new_base) {
1169 72 : timer->flags |= TIMER_MIGRATING;
1170 :
1171 72 : raw_spin_unlock(&base->lock);
1172 72 : base = new_base;
1173 72 : raw_spin_lock(&base->lock);
1174 72 : WRITE_ONCE(timer->flags,
1175 : (timer->flags & ~TIMER_BASEMASK) | cpu);
1176 : }
1177 194 : forward_timer_base(base);
1178 :
1179 194 : debug_timer_activate(timer);
1180 194 : internal_add_timer(base, timer);
1181 194 : raw_spin_unlock_irqrestore(&base->lock, flags);
1182 194 : }
1183 : EXPORT_SYMBOL_GPL(add_timer_on);
1184 :
1185 : /**
1186 : * del_timer - deactivate a timer.
1187 : * @timer: the timer to be deactivated
1188 : *
1189 : * del_timer() deactivates a timer - this works on both active and inactive
1190 : * timers.
1191 : *
1192 : * The function returns whether it has deactivated a pending timer or not.
1193 : * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
1194 : * active timer returns 1.)
1195 : */
1196 272 : int del_timer(struct timer_list *timer)
1197 : {
1198 272 : struct timer_base *base;
1199 272 : unsigned long flags;
1200 272 : int ret = 0;
1201 :
1202 272 : debug_assert_init(timer);
1203 :
1204 272 : if (timer_pending(timer)) {
1205 13 : base = lock_timer_base(timer, &flags);
1206 13 : ret = detach_if_pending(timer, base, true);
1207 13 : raw_spin_unlock_irqrestore(&base->lock, flags);
1208 : }
1209 :
1210 272 : return ret;
1211 : }
1212 : EXPORT_SYMBOL(del_timer);
1213 :
1214 : /**
1215 : * try_to_del_timer_sync - Try to deactivate a timer
1216 : * @timer: timer to delete
1217 : *
1218 : * This function tries to deactivate a timer. Upon successful (ret >= 0)
1219 : * exit the timer is not queued and the handler is not running on any CPU.
1220 : */
1221 4692 : int try_to_del_timer_sync(struct timer_list *timer)
1222 : {
1223 4692 : struct timer_base *base;
1224 4692 : unsigned long flags;
1225 4692 : int ret = -1;
1226 :
1227 4692 : debug_assert_init(timer);
1228 :
1229 4692 : base = lock_timer_base(timer, &flags);
1230 :
1231 4692 : if (base->running_timer != timer)
1232 4692 : ret = detach_if_pending(timer, base, true);
1233 :
1234 4692 : raw_spin_unlock_irqrestore(&base->lock, flags);
1235 :
1236 4692 : return ret;
1237 : }
1238 : EXPORT_SYMBOL(try_to_del_timer_sync);
1239 :
1240 0 : bool timer_curr_running(struct timer_list *timer)
1241 : {
1242 0 : int i;
1243 :
1244 0 : for (i = 0; i < NR_BASES; i++) {
1245 0 : struct timer_base *base = this_cpu_ptr(&timer_bases[i]);
1246 :
1247 0 : if (base->running_timer == timer)
1248 : return true;
1249 : }
1250 :
1251 : return false;
1252 : }
1253 :
1254 : #ifdef CONFIG_PREEMPT_RT
1255 : static __init void timer_base_init_expiry_lock(struct timer_base *base)
1256 : {
1257 : spin_lock_init(&base->expiry_lock);
1258 : }
1259 :
1260 : static inline void timer_base_lock_expiry(struct timer_base *base)
1261 : {
1262 : spin_lock(&base->expiry_lock);
1263 : }
1264 :
1265 : static inline void timer_base_unlock_expiry(struct timer_base *base)
1266 : {
1267 : spin_unlock(&base->expiry_lock);
1268 : }
1269 :
1270 : /*
1271 : * The counterpart to del_timer_wait_running().
1272 : *
1273 : * If there is a waiter for base->expiry_lock, then it was waiting for the
1274 : * timer callback to finish. Drop expiry_lock and reaquire it. That allows
1275 : * the waiter to acquire the lock and make progress.
1276 : */
1277 : static void timer_sync_wait_running(struct timer_base *base)
1278 : {
1279 : if (atomic_read(&base->timer_waiters)) {
1280 : spin_unlock(&base->expiry_lock);
1281 : spin_lock(&base->expiry_lock);
1282 : }
1283 : }
1284 :
1285 : /*
1286 : * This function is called on PREEMPT_RT kernels when the fast path
1287 : * deletion of a timer failed because the timer callback function was
1288 : * running.
1289 : *
1290 : * This prevents priority inversion, if the softirq thread on a remote CPU
1291 : * got preempted, and it prevents a life lock when the task which tries to
1292 : * delete a timer preempted the softirq thread running the timer callback
1293 : * function.
1294 : */
1295 : static void del_timer_wait_running(struct timer_list *timer)
1296 : {
1297 : u32 tf;
1298 :
1299 : tf = READ_ONCE(timer->flags);
1300 : if (!(tf & (TIMER_MIGRATING | TIMER_IRQSAFE))) {
1301 : struct timer_base *base = get_timer_base(tf);
1302 :
1303 : /*
1304 : * Mark the base as contended and grab the expiry lock,
1305 : * which is held by the softirq across the timer
1306 : * callback. Drop the lock immediately so the softirq can
1307 : * expire the next timer. In theory the timer could already
1308 : * be running again, but that's more than unlikely and just
1309 : * causes another wait loop.
1310 : */
1311 : atomic_inc(&base->timer_waiters);
1312 : spin_lock_bh(&base->expiry_lock);
1313 : atomic_dec(&base->timer_waiters);
1314 : spin_unlock_bh(&base->expiry_lock);
1315 : }
1316 : }
1317 : #else
1318 8 : static inline void timer_base_init_expiry_lock(struct timer_base *base) { }
1319 4508 : static inline void timer_base_lock_expiry(struct timer_base *base) { }
1320 4517 : static inline void timer_base_unlock_expiry(struct timer_base *base) { }
1321 3286 : static inline void timer_sync_wait_running(struct timer_base *base) { }
1322 0 : static inline void del_timer_wait_running(struct timer_list *timer) { }
1323 : #endif
1324 :
1325 : #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT_RT)
1326 : /**
1327 : * del_timer_sync - deactivate a timer and wait for the handler to finish.
1328 : * @timer: the timer to be deactivated
1329 : *
1330 : * This function only differs from del_timer() on SMP: besides deactivating
1331 : * the timer it also makes sure the handler has finished executing on other
1332 : * CPUs.
1333 : *
1334 : * Synchronization rules: Callers must prevent restarting of the timer,
1335 : * otherwise this function is meaningless. It must not be called from
1336 : * interrupt contexts unless the timer is an irqsafe one. The caller must
1337 : * not hold locks which would prevent completion of the timer's
1338 : * handler. The timer's handler must not call add_timer_on(). Upon exit the
1339 : * timer is not queued and the handler is not running on any CPU.
1340 : *
1341 : * Note: For !irqsafe timers, you must not hold locks that are held in
1342 : * interrupt context while calling this function. Even if the lock has
1343 : * nothing to do with the timer in question. Here's why::
1344 : *
1345 : * CPU0 CPU1
1346 : * ---- ----
1347 : * <SOFTIRQ>
1348 : * call_timer_fn();
1349 : * base->running_timer = mytimer;
1350 : * spin_lock_irq(somelock);
1351 : * <IRQ>
1352 : * spin_lock(somelock);
1353 : * del_timer_sync(mytimer);
1354 : * while (base->running_timer == mytimer);
1355 : *
1356 : * Now del_timer_sync() will never return and never release somelock.
1357 : * The interrupt on the other CPU is waiting to grab somelock but
1358 : * it has interrupted the softirq that CPU0 is waiting to finish.
1359 : *
1360 : * The function returns whether it has deactivated a pending timer or not.
1361 : */
1362 4692 : int del_timer_sync(struct timer_list *timer)
1363 : {
1364 4692 : int ret;
1365 :
1366 : #ifdef CONFIG_LOCKDEP
1367 4692 : unsigned long flags;
1368 :
1369 : /*
1370 : * If lockdep gives a backtrace here, please reference
1371 : * the synchronization rules above.
1372 : */
1373 9384 : local_irq_save(flags);
1374 4692 : lock_map_acquire(&timer->lockdep_map);
1375 4692 : lock_map_release(&timer->lockdep_map);
1376 4692 : local_irq_restore(flags);
1377 : #endif
1378 : /*
1379 : * don't use it in hardirq context, because it
1380 : * could lead to deadlock.
1381 : */
1382 9384 : WARN_ON(in_irq() && !(timer->flags & TIMER_IRQSAFE));
1383 :
1384 : /*
1385 : * Must be able to sleep on PREEMPT_RT because of the slowpath in
1386 : * del_timer_wait_running().
1387 : */
1388 : if (IS_ENABLED(CONFIG_PREEMPT_RT) && !(timer->flags & TIMER_IRQSAFE))
1389 4692 : lockdep_assert_preemption_enabled();
1390 :
1391 4692 : do {
1392 4692 : ret = try_to_del_timer_sync(timer);
1393 :
1394 4692 : if (unlikely(ret < 0)) {
1395 0 : del_timer_wait_running(timer);
1396 0 : cpu_relax();
1397 : }
1398 4692 : } while (ret < 0);
1399 :
1400 4692 : return ret;
1401 : }
1402 : EXPORT_SYMBOL(del_timer_sync);
1403 : #endif
1404 :
1405 3887 : static void call_timer_fn(struct timer_list *timer,
1406 : void (*fn)(struct timer_list *),
1407 : unsigned long baseclk)
1408 : {
1409 3887 : int count = preempt_count();
1410 :
1411 : #ifdef CONFIG_LOCKDEP
1412 : /*
1413 : * It is permissible to free the timer from inside the
1414 : * function that is called from it, this we need to take into
1415 : * account for lockdep too. To avoid bogus "held lock freed"
1416 : * warnings as well as problems when looking into
1417 : * timer->lockdep_map, make a copy and use that here.
1418 : */
1419 3887 : struct lockdep_map lockdep_map;
1420 :
1421 3887 : lockdep_copy_map(&lockdep_map, &timer->lockdep_map);
1422 : #endif
1423 : /*
1424 : * Couple the lock chain with the lock chain at
1425 : * del_timer_sync() by acquiring the lock_map around the fn()
1426 : * call here and in del_timer_sync().
1427 : */
1428 3887 : lock_map_acquire(&lockdep_map);
1429 :
1430 3887 : trace_timer_expire_entry(timer, baseclk);
1431 3887 : fn(timer);
1432 3890 : trace_timer_expire_exit(timer);
1433 :
1434 3890 : lock_map_release(&lockdep_map);
1435 :
1436 3890 : if (count != preempt_count()) {
1437 0 : WARN_ONCE(1, "timer: %pS preempt leak: %08x -> %08x\n",
1438 : fn, count, preempt_count());
1439 : /*
1440 : * Restore the preempt count. That gives us a decent
1441 : * chance to survive and extract information. If the
1442 : * callback kept a lock held, bad luck, but not worse
1443 : * than the BUG() we had.
1444 : */
1445 0 : preempt_count_set(count);
1446 : }
1447 3890 : }
1448 :
1449 3796 : static void expire_timers(struct timer_base *base, struct hlist_head *head)
1450 : {
1451 : /*
1452 : * This value is required only for tracing. base->clk was
1453 : * incremented directly before expire_timers was called. But expiry
1454 : * is related to the old base->clk value.
1455 : */
1456 3796 : unsigned long baseclk = base->clk - 1;
1457 :
1458 7686 : while (!hlist_empty(head)) {
1459 3884 : struct timer_list *timer;
1460 3884 : void (*fn)(struct timer_list *);
1461 :
1462 3884 : timer = hlist_entry(head->first, struct timer_list, entry);
1463 :
1464 3884 : base->running_timer = timer;
1465 3884 : detach_timer(timer, true);
1466 :
1467 3890 : fn = timer->function;
1468 :
1469 3890 : if (timer->flags & TIMER_IRQSAFE) {
1470 604 : raw_spin_unlock(&base->lock);
1471 604 : call_timer_fn(timer, fn, baseclk);
1472 602 : base->running_timer = NULL;
1473 602 : raw_spin_lock(&base->lock);
1474 : } else {
1475 3286 : raw_spin_unlock_irq(&base->lock);
1476 3285 : call_timer_fn(timer, fn, baseclk);
1477 3286 : base->running_timer = NULL;
1478 3286 : timer_sync_wait_running(base);
1479 3286 : raw_spin_lock_irq(&base->lock);
1480 : }
1481 : }
1482 3802 : }
1483 :
1484 4611 : static int collect_expired_timers(struct timer_base *base,
1485 : struct hlist_head *heads)
1486 : {
1487 4611 : unsigned long clk = base->clk = base->next_expiry;
1488 4611 : struct hlist_head *vec;
1489 4611 : int i, levels = 0;
1490 4611 : unsigned int idx;
1491 :
1492 5588 : for (i = 0; i < LVL_DEPTH; i++) {
1493 5588 : idx = (clk & LVL_MASK) + i * LVL_SIZE;
1494 :
1495 5588 : if (__test_and_clear_bit(idx, base->pending_map)) {
1496 3797 : vec = base->vectors + idx;
1497 3797 : hlist_move_list(vec, heads++);
1498 3797 : levels++;
1499 : }
1500 : /* Is it time to look at the next level? */
1501 5593 : if (clk & LVL_CLK_MASK)
1502 : break;
1503 : /* Shift clock for the next level granularity */
1504 977 : clk >>= LVL_CLK_SHIFT;
1505 : }
1506 4616 : return levels;
1507 : }
1508 :
1509 : /*
1510 : * Find the next pending bucket of a level. Search from level start (@offset)
1511 : * + @clk upwards and if nothing there, search from start of the level
1512 : * (@offset) up to @offset + clk.
1513 : */
1514 26633 : static int next_pending_bucket(struct timer_base *base, unsigned offset,
1515 : unsigned clk)
1516 : {
1517 26633 : unsigned pos, start = offset + clk;
1518 26633 : unsigned end = offset + LVL_SIZE;
1519 :
1520 26633 : pos = find_next_bit(base->pending_map, end, start);
1521 26646 : if (pos < end)
1522 5650 : return pos - start;
1523 :
1524 20996 : pos = find_next_bit(base->pending_map, start, offset);
1525 21004 : return pos < start ? pos + LVL_SIZE - start : -1;
1526 : }
1527 :
1528 : /*
1529 : * Search the first expiring timer in the various clock levels. Caller must
1530 : * hold base->lock.
1531 : */
1532 4718 : static unsigned long __next_timer_interrupt(struct timer_base *base)
1533 : {
1534 4718 : unsigned long clk, next, adj;
1535 4718 : unsigned lvl, offset = 0;
1536 :
1537 4718 : next = base->clk + NEXT_TIMER_MAX_DELTA;
1538 4718 : clk = base->clk;
1539 29090 : for (lvl = 0; lvl < LVL_DEPTH; lvl++, offset += LVL_SIZE) {
1540 26638 : int pos = next_pending_bucket(base, offset, clk & LVL_MASK);
1541 26656 : unsigned long lvl_clk = clk & LVL_CLK_MASK;
1542 :
1543 26656 : if (pos >= 0) {
1544 7962 : unsigned long tmp = clk + (unsigned long) pos;
1545 :
1546 7962 : tmp <<= LVL_SHIFT(lvl);
1547 7962 : if (time_before(tmp, next))
1548 4586 : next = tmp;
1549 :
1550 : /*
1551 : * If the next expiration happens before we reach
1552 : * the next level, no need to check further.
1553 : */
1554 7962 : if (pos <= ((LVL_CLK_DIV - lvl_clk) & LVL_CLK_MASK))
1555 : break;
1556 : }
1557 : /*
1558 : * Clock for the next level. If the current level clock lower
1559 : * bits are zero, we look at the next level as is. If not we
1560 : * need to advance it by one because that's going to be the
1561 : * next expiring bucket in that level. base->clk is the next
1562 : * expiring jiffie. So in case of:
1563 : *
1564 : * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1565 : * 0 0 0 0 0 0
1566 : *
1567 : * we have to look at all levels @index 0. With
1568 : *
1569 : * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1570 : * 0 0 0 0 0 2
1571 : *
1572 : * LVL0 has the next expiring bucket @index 2. The upper
1573 : * levels have the next expiring bucket @index 1.
1574 : *
1575 : * In case that the propagation wraps the next level the same
1576 : * rules apply:
1577 : *
1578 : * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1579 : * 0 0 0 0 F 2
1580 : *
1581 : * So after looking at LVL0 we get:
1582 : *
1583 : * LVL5 LVL4 LVL3 LVL2 LVL1
1584 : * 0 0 0 1 0
1585 : *
1586 : * So no propagation from LVL1 to LVL2 because that happened
1587 : * with the add already, but then we need to propagate further
1588 : * from LVL2 to LVL3.
1589 : *
1590 : * So the simple check whether the lower bits of the current
1591 : * level are 0 or not is sufficient for all cases.
1592 : */
1593 24372 : adj = lvl_clk ? 1 : 0;
1594 24372 : clk >>= LVL_CLK_SHIFT;
1595 24372 : clk += adj;
1596 : }
1597 :
1598 4736 : base->next_expiry_recalc = false;
1599 :
1600 4736 : return next;
1601 : }
1602 :
1603 : #ifdef CONFIG_NO_HZ_COMMON
1604 : /*
1605 : * Check, if the next hrtimer event is before the next timer wheel
1606 : * event:
1607 : */
1608 1221 : static u64 cmp_next_hrtimer_event(u64 basem, u64 expires)
1609 : {
1610 1221 : u64 nextevt = hrtimer_get_next_event();
1611 :
1612 : /*
1613 : * If high resolution timers are enabled
1614 : * hrtimer_get_next_event() returns KTIME_MAX.
1615 : */
1616 1218 : if (expires <= nextevt)
1617 : return expires;
1618 :
1619 : /*
1620 : * If the next timer is already expired, return the tick base
1621 : * time so the tick is fired immediately.
1622 : */
1623 76 : if (nextevt <= basem)
1624 : return basem;
1625 :
1626 : /*
1627 : * Round up to the next jiffie. High resolution timers are
1628 : * off, so the hrtimers are expired in the tick and we need to
1629 : * make sure that this tick really expires the timer to avoid
1630 : * a ping pong of the nohz stop code.
1631 : *
1632 : * Use DIV_ROUND_UP_ULL to prevent gcc calling __divdi3
1633 : */
1634 75 : return DIV_ROUND_UP_ULL(nextevt, TICK_NSEC) * TICK_NSEC;
1635 : }
1636 :
1637 : /**
1638 : * get_next_timer_interrupt - return the time (clock mono) of the next timer
1639 : * @basej: base time jiffies
1640 : * @basem: base time clock monotonic
1641 : *
1642 : * Returns the tick aligned clock monotonic time of the next pending
1643 : * timer or KTIME_MAX if no timer is pending.
1644 : */
1645 1218 : u64 get_next_timer_interrupt(unsigned long basej, u64 basem)
1646 : {
1647 1218 : struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1648 1220 : u64 expires = KTIME_MAX;
1649 1220 : unsigned long nextevt;
1650 1220 : bool is_max_delta;
1651 :
1652 : /*
1653 : * Pretend that there is no timer pending if the cpu is offline.
1654 : * Possible pending timers will be migrated later to an active cpu.
1655 : */
1656 1220 : if (cpu_is_offline(smp_processor_id()))
1657 : return expires;
1658 :
1659 1221 : raw_spin_lock(&base->lock);
1660 1223 : if (base->next_expiry_recalc)
1661 117 : base->next_expiry = __next_timer_interrupt(base);
1662 1223 : nextevt = base->next_expiry;
1663 1223 : is_max_delta = (nextevt == base->clk + NEXT_TIMER_MAX_DELTA);
1664 :
1665 : /*
1666 : * We have a fresh next event. Check whether we can forward the
1667 : * base. We can only do that when @basej is past base->clk
1668 : * otherwise we might rewind base->clk.
1669 : */
1670 1223 : if (time_after(basej, base->clk)) {
1671 345 : if (time_after(nextevt, basej))
1672 345 : base->clk = basej;
1673 0 : else if (time_after(nextevt, base->clk))
1674 0 : base->clk = nextevt;
1675 : }
1676 :
1677 1223 : if (time_before_eq(nextevt, basej)) {
1678 2 : expires = basem;
1679 2 : base->is_idle = false;
1680 : } else {
1681 1221 : if (!is_max_delta)
1682 1217 : expires = basem + (u64)(nextevt - basej) * TICK_NSEC;
1683 : /*
1684 : * If we expect to sleep more than a tick, mark the base idle.
1685 : * Also the tick is stopped so any added timer must forward
1686 : * the base clk itself to keep granularity small. This idle
1687 : * logic is only maintained for the BASE_STD base, deferrable
1688 : * timers may still see large granularity skew (by design).
1689 : */
1690 1221 : if ((expires - basem) > TICK_NSEC)
1691 1096 : base->is_idle = true;
1692 : }
1693 1223 : raw_spin_unlock(&base->lock);
1694 :
1695 1222 : return cmp_next_hrtimer_event(basem, expires);
1696 : }
1697 :
1698 : /**
1699 : * timer_clear_idle - Clear the idle state of the timer base
1700 : *
1701 : * Called with interrupts disabled
1702 : */
1703 17376 : void timer_clear_idle(void)
1704 : {
1705 17376 : struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1706 :
1707 : /*
1708 : * We do this unlocked. The worst outcome is a remote enqueue sending
1709 : * a pointless IPI, but taking the lock would just make the window for
1710 : * sending the IPI a few instructions smaller for the cost of taking
1711 : * the lock in the exit from idle path.
1712 : */
1713 17410 : base->is_idle = false;
1714 17410 : }
1715 : #endif
1716 :
1717 : /**
1718 : * __run_timers - run all expired timers (if any) on this CPU.
1719 : * @base: the timer vector to be processed.
1720 : */
1721 8972 : static inline void __run_timers(struct timer_base *base)
1722 : {
1723 8972 : struct hlist_head heads[LVL_DEPTH];
1724 8972 : int levels;
1725 :
1726 8972 : if (time_before(jiffies, base->next_expiry))
1727 4464 : return;
1728 :
1729 4508 : timer_base_lock_expiry(base);
1730 4508 : raw_spin_lock_irq(&base->lock);
1731 :
1732 9123 : while (time_after_eq(jiffies, base->clk) &&
1733 5173 : time_after_eq(jiffies, base->next_expiry)) {
1734 4614 : levels = collect_expired_timers(base, heads);
1735 : /*
1736 : * The only possible reason for not finding any expired
1737 : * timer at this clk is that all matching timers have been
1738 : * dequeued.
1739 : */
1740 9220 : WARN_ON_ONCE(!levels && !base->next_expiry_recalc);
1741 4610 : base->clk++;
1742 4610 : base->next_expiry = __next_timer_interrupt(base);
1743 :
1744 8412 : while (levels--)
1745 3797 : expire_timers(base, heads + levels);
1746 : }
1747 4518 : raw_spin_unlock_irq(&base->lock);
1748 4517 : timer_base_unlock_expiry(base);
1749 : }
1750 :
1751 : /*
1752 : * This function runs timers and the timer-tq in bottom half context.
1753 : */
1754 4476 : static __latent_entropy void run_timer_softirq(struct softirq_action *h)
1755 : {
1756 4476 : struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1757 :
1758 4479 : __run_timers(base);
1759 4488 : if (IS_ENABLED(CONFIG_NO_HZ_COMMON))
1760 4488 : __run_timers(this_cpu_ptr(&timer_bases[BASE_DEF]));
1761 4492 : }
1762 :
1763 : /*
1764 : * Called by the local, per-CPU timer interrupt on SMP.
1765 : */
1766 28314 : static void run_local_timers(void)
1767 : {
1768 28314 : struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1769 :
1770 28332 : hrtimer_run_queues();
1771 : /* Raise the softirq only if required. */
1772 28560 : if (time_before(jiffies, base->next_expiry)) {
1773 24039 : if (!IS_ENABLED(CONFIG_NO_HZ_COMMON))
1774 : return;
1775 : /* CPU is awake, so check the deferrable base. */
1776 24039 : base++;
1777 24039 : if (time_before(jiffies, base->next_expiry))
1778 : return;
1779 : }
1780 4624 : raise_softirq(TIMER_SOFTIRQ);
1781 : }
1782 :
1783 : /*
1784 : * Called from the timer interrupt handler to charge one tick to the current
1785 : * process. user_tick is 1 if the tick is user time, 0 for system.
1786 : */
1787 28034 : void update_process_times(int user_tick)
1788 : {
1789 28034 : struct task_struct *p = current;
1790 :
1791 28034 : PRANDOM_ADD_NOISE(jiffies, user_tick, p, 0);
1792 :
1793 : /* Note: this timer irq context must be accounted for as well. */
1794 28034 : account_process_tick(p, user_tick);
1795 28396 : run_local_timers();
1796 28468 : rcu_sched_clock_irq(user_tick);
1797 : #ifdef CONFIG_IRQ_WORK
1798 28280 : if (in_irq())
1799 28294 : irq_work_tick();
1800 : #endif
1801 28211 : scheduler_tick();
1802 28288 : if (IS_ENABLED(CONFIG_POSIX_TIMERS))
1803 28288 : run_posix_cpu_timers();
1804 28528 : }
1805 :
1806 : /*
1807 : * Since schedule_timeout()'s timer is defined on the stack, it must store
1808 : * the target task on the stack as well.
1809 : */
1810 : struct process_timer {
1811 : struct timer_list timer;
1812 : struct task_struct *task;
1813 : };
1814 :
1815 3189 : static void process_timeout(struct timer_list *t)
1816 : {
1817 3189 : struct process_timer *timeout = from_timer(timeout, t, timer);
1818 :
1819 3189 : wake_up_process(timeout->task);
1820 3188 : }
1821 :
1822 : /**
1823 : * schedule_timeout - sleep until timeout
1824 : * @timeout: timeout value in jiffies
1825 : *
1826 : * Make the current task sleep until @timeout jiffies have elapsed.
1827 : * The function behavior depends on the current task state
1828 : * (see also set_current_state() description):
1829 : *
1830 : * %TASK_RUNNING - the scheduler is called, but the task does not sleep
1831 : * at all. That happens because sched_submit_work() does nothing for
1832 : * tasks in %TASK_RUNNING state.
1833 : *
1834 : * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1835 : * pass before the routine returns unless the current task is explicitly
1836 : * woken up, (e.g. by wake_up_process()).
1837 : *
1838 : * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1839 : * delivered to the current task or the current task is explicitly woken
1840 : * up.
1841 : *
1842 : * The current task state is guaranteed to be %TASK_RUNNING when this
1843 : * routine returns.
1844 : *
1845 : * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1846 : * the CPU away without a bound on the timeout. In this case the return
1847 : * value will be %MAX_SCHEDULE_TIMEOUT.
1848 : *
1849 : * Returns 0 when the timer has expired otherwise the remaining time in
1850 : * jiffies will be returned. In all cases the return value is guaranteed
1851 : * to be non-negative.
1852 : */
1853 4999 : signed long __sched schedule_timeout(signed long timeout)
1854 : {
1855 4999 : struct process_timer timer;
1856 4999 : unsigned long expire;
1857 :
1858 4999 : switch (timeout)
1859 : {
1860 430 : case MAX_SCHEDULE_TIMEOUT:
1861 : /*
1862 : * These two special cases are useful to be comfortable
1863 : * in the caller. Nothing more. We could take
1864 : * MAX_SCHEDULE_TIMEOUT from one of the negative value
1865 : * but I' d like to return a valid offset (>=0) to allow
1866 : * the caller to do everything it want with the retval.
1867 : */
1868 430 : schedule();
1869 429 : goto out;
1870 4569 : default:
1871 : /*
1872 : * Another bit of PARANOID. Note that the retval will be
1873 : * 0 since no piece of kernel is supposed to do a check
1874 : * for a negative retval of schedule_timeout() (since it
1875 : * should never happens anyway). You just have the printk()
1876 : * that will tell you if something is gone wrong and where.
1877 : */
1878 4569 : if (timeout < 0) {
1879 0 : printk(KERN_ERR "schedule_timeout: wrong timeout "
1880 : "value %lx\n", timeout);
1881 0 : dump_stack();
1882 0 : current->state = TASK_RUNNING;
1883 0 : goto out;
1884 : }
1885 : }
1886 :
1887 4569 : expire = timeout + jiffies;
1888 :
1889 4569 : timer.task = current;
1890 4569 : timer_setup_on_stack(&timer.timer, process_timeout, 0);
1891 4569 : __mod_timer(&timer.timer, expire, MOD_TIMER_NOTPENDING);
1892 4569 : schedule();
1893 4566 : del_singleshot_timer_sync(&timer.timer);
1894 :
1895 : /* Remove the timer from the object tracker */
1896 4566 : destroy_timer_on_stack(&timer.timer);
1897 :
1898 4566 : timeout = expire - jiffies;
1899 :
1900 4995 : out:
1901 4995 : return timeout < 0 ? 0 : timeout;
1902 : }
1903 : EXPORT_SYMBOL(schedule_timeout);
1904 :
1905 : /*
1906 : * We can use __set_current_state() here because schedule_timeout() calls
1907 : * schedule() unconditionally.
1908 : */
1909 0 : signed long __sched schedule_timeout_interruptible(signed long timeout)
1910 : {
1911 0 : __set_current_state(TASK_INTERRUPTIBLE);
1912 0 : return schedule_timeout(timeout);
1913 : }
1914 : EXPORT_SYMBOL(schedule_timeout_interruptible);
1915 :
1916 0 : signed long __sched schedule_timeout_killable(signed long timeout)
1917 : {
1918 0 : __set_current_state(TASK_KILLABLE);
1919 0 : return schedule_timeout(timeout);
1920 : }
1921 : EXPORT_SYMBOL(schedule_timeout_killable);
1922 :
1923 0 : signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1924 : {
1925 0 : __set_current_state(TASK_UNINTERRUPTIBLE);
1926 0 : return schedule_timeout(timeout);
1927 : }
1928 : EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1929 :
1930 : /*
1931 : * Like schedule_timeout_uninterruptible(), except this task will not contribute
1932 : * to load average.
1933 : */
1934 0 : signed long __sched schedule_timeout_idle(signed long timeout)
1935 : {
1936 0 : __set_current_state(TASK_IDLE);
1937 0 : return schedule_timeout(timeout);
1938 : }
1939 : EXPORT_SYMBOL(schedule_timeout_idle);
1940 :
1941 : #ifdef CONFIG_HOTPLUG_CPU
1942 0 : static void migrate_timer_list(struct timer_base *new_base, struct hlist_head *head)
1943 : {
1944 0 : struct timer_list *timer;
1945 0 : int cpu = new_base->cpu;
1946 :
1947 0 : while (!hlist_empty(head)) {
1948 0 : timer = hlist_entry(head->first, struct timer_list, entry);
1949 0 : detach_timer(timer, false);
1950 0 : timer->flags = (timer->flags & ~TIMER_BASEMASK) | cpu;
1951 0 : internal_add_timer(new_base, timer);
1952 : }
1953 0 : }
1954 :
1955 3 : int timers_prepare_cpu(unsigned int cpu)
1956 : {
1957 3 : struct timer_base *base;
1958 3 : int b;
1959 :
1960 9 : for (b = 0; b < NR_BASES; b++) {
1961 6 : base = per_cpu_ptr(&timer_bases[b], cpu);
1962 6 : base->clk = jiffies;
1963 6 : base->next_expiry = base->clk + NEXT_TIMER_MAX_DELTA;
1964 6 : base->is_idle = false;
1965 : }
1966 3 : return 0;
1967 : }
1968 :
1969 0 : int timers_dead_cpu(unsigned int cpu)
1970 : {
1971 0 : struct timer_base *old_base;
1972 0 : struct timer_base *new_base;
1973 0 : int b, i;
1974 :
1975 0 : BUG_ON(cpu_online(cpu));
1976 :
1977 0 : for (b = 0; b < NR_BASES; b++) {
1978 0 : old_base = per_cpu_ptr(&timer_bases[b], cpu);
1979 0 : new_base = get_cpu_ptr(&timer_bases[b]);
1980 : /*
1981 : * The caller is globally serialized and nobody else
1982 : * takes two locks at once, deadlock is not possible.
1983 : */
1984 0 : raw_spin_lock_irq(&new_base->lock);
1985 0 : raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1986 :
1987 : /*
1988 : * The current CPUs base clock might be stale. Update it
1989 : * before moving the timers over.
1990 : */
1991 0 : forward_timer_base(new_base);
1992 :
1993 0 : BUG_ON(old_base->running_timer);
1994 :
1995 0 : for (i = 0; i < WHEEL_SIZE; i++)
1996 0 : migrate_timer_list(new_base, old_base->vectors + i);
1997 :
1998 0 : raw_spin_unlock(&old_base->lock);
1999 0 : raw_spin_unlock_irq(&new_base->lock);
2000 0 : put_cpu_ptr(&timer_bases);
2001 : }
2002 0 : return 0;
2003 : }
2004 :
2005 : #endif /* CONFIG_HOTPLUG_CPU */
2006 :
2007 4 : static void __init init_timer_cpu(int cpu)
2008 : {
2009 4 : struct timer_base *base;
2010 4 : int i;
2011 :
2012 12 : for (i = 0; i < NR_BASES; i++) {
2013 8 : base = per_cpu_ptr(&timer_bases[i], cpu);
2014 8 : base->cpu = cpu;
2015 8 : raw_spin_lock_init(&base->lock);
2016 8 : base->clk = jiffies;
2017 8 : base->next_expiry = base->clk + NEXT_TIMER_MAX_DELTA;
2018 8 : timer_base_init_expiry_lock(base);
2019 : }
2020 4 : }
2021 :
2022 1 : static void __init init_timer_cpus(void)
2023 : {
2024 1 : int cpu;
2025 :
2026 5 : for_each_possible_cpu(cpu)
2027 4 : init_timer_cpu(cpu);
2028 1 : }
2029 :
2030 1 : void __init init_timers(void)
2031 : {
2032 1 : init_timer_cpus();
2033 1 : posix_cputimers_init_work();
2034 1 : open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
2035 1 : }
2036 :
2037 : /**
2038 : * msleep - sleep safely even with waitqueue interruptions
2039 : * @msecs: Time in milliseconds to sleep for
2040 : */
2041 0 : void msleep(unsigned int msecs)
2042 : {
2043 0 : unsigned long timeout = msecs_to_jiffies(msecs) + 1;
2044 :
2045 0 : while (timeout)
2046 0 : timeout = schedule_timeout_uninterruptible(timeout);
2047 0 : }
2048 :
2049 : EXPORT_SYMBOL(msleep);
2050 :
2051 : /**
2052 : * msleep_interruptible - sleep waiting for signals
2053 : * @msecs: Time in milliseconds to sleep for
2054 : */
2055 0 : unsigned long msleep_interruptible(unsigned int msecs)
2056 : {
2057 0 : unsigned long timeout = msecs_to_jiffies(msecs) + 1;
2058 :
2059 0 : while (timeout && !signal_pending(current))
2060 0 : timeout = schedule_timeout_interruptible(timeout);
2061 0 : return jiffies_to_msecs(timeout);
2062 : }
2063 :
2064 : EXPORT_SYMBOL(msleep_interruptible);
2065 :
2066 : /**
2067 : * usleep_range - Sleep for an approximate time
2068 : * @min: Minimum time in usecs to sleep
2069 : * @max: Maximum time in usecs to sleep
2070 : *
2071 : * In non-atomic context where the exact wakeup time is flexible, use
2072 : * usleep_range() instead of udelay(). The sleep improves responsiveness
2073 : * by avoiding the CPU-hogging busy-wait of udelay(), and the range reduces
2074 : * power usage by allowing hrtimers to take advantage of an already-
2075 : * scheduled interrupt instead of scheduling a new one just for this sleep.
2076 : */
2077 0 : void __sched usleep_range(unsigned long min, unsigned long max)
2078 : {
2079 0 : ktime_t exp = ktime_add_us(ktime_get(), min);
2080 0 : u64 delta = (u64)(max - min) * NSEC_PER_USEC;
2081 :
2082 0 : for (;;) {
2083 0 : __set_current_state(TASK_UNINTERRUPTIBLE);
2084 : /* Do not return before the requested sleep time has elapsed */
2085 0 : if (!schedule_hrtimeout_range(&exp, delta, HRTIMER_MODE_ABS))
2086 : break;
2087 : }
2088 0 : }
2089 : EXPORT_SYMBOL(usleep_range);
|
