]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - kernel/time/timer.c
UBUNTU: Ubuntu-4.13.0-45.50
[mirror_ubuntu-artful-kernel.git] / kernel / time / timer.c
1 /*
2 * linux/kernel/timer.c
3 *
4 * Kernel internal timers
5 *
6 * Copyright (C) 1991, 1992 Linus Torvalds
7 *
8 * 1997-01-28 Modified by Finn Arne Gangstad to make timers scale better.
9 *
10 * 1997-09-10 Updated NTP code according to technical memorandum Jan '96
11 * "A Kernel Model for Precision Timekeeping" by Dave Mills
12 * 1998-12-24 Fixed a xtime SMP race (we need the xtime_lock rw spinlock to
13 * serialize accesses to xtime/lost_ticks).
14 * Copyright (C) 1998 Andrea Arcangeli
15 * 1999-03-10 Improved NTP compatibility by Ulrich Windl
16 * 2002-05-31 Move sys_sysinfo here and make its locking sane, Robert Love
17 * 2000-10-05 Implemented scalable SMP per-CPU timer handling.
18 * Copyright (C) 2000, 2001, 2002 Ingo Molnar
19 * Designed by David S. Miller, Alexey Kuznetsov and Ingo Molnar
20 */
21
22 #include <linux/kernel_stat.h>
23 #include <linux/export.h>
24 #include <linux/interrupt.h>
25 #include <linux/percpu.h>
26 #include <linux/init.h>
27 #include <linux/mm.h>
28 #include <linux/swap.h>
29 #include <linux/pid_namespace.h>
30 #include <linux/notifier.h>
31 #include <linux/thread_info.h>
32 #include <linux/time.h>
33 #include <linux/jiffies.h>
34 #include <linux/posix-timers.h>
35 #include <linux/cpu.h>
36 #include <linux/syscalls.h>
37 #include <linux/delay.h>
38 #include <linux/tick.h>
39 #include <linux/kallsyms.h>
40 #include <linux/irq_work.h>
41 #include <linux/sched/signal.h>
42 #include <linux/sched/sysctl.h>
43 #include <linux/sched/nohz.h>
44 #include <linux/sched/debug.h>
45 #include <linux/slab.h>
46 #include <linux/compat.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.
161 */
162 #define LVL_START(n) ((LVL_SIZE - 1) << (((n) - 1) * LVL_CLK_SHIFT))
163
164 /* Size of each clock level */
165 #define LVL_BITS 6
166 #define LVL_SIZE (1UL << LVL_BITS)
167 #define LVL_MASK (LVL_SIZE - 1)
168 #define LVL_OFFS(n) ((n) * LVL_SIZE)
169
170 /* Level depth */
171 #if HZ > 100
172 # define LVL_DEPTH 9
173 # else
174 # define LVL_DEPTH 8
175 #endif
176
177 /* The cutoff (max. capacity of the wheel) */
178 #define WHEEL_TIMEOUT_CUTOFF (LVL_START(LVL_DEPTH))
179 #define WHEEL_TIMEOUT_MAX (WHEEL_TIMEOUT_CUTOFF - LVL_GRAN(LVL_DEPTH - 1))
180
181 /*
182 * The resulting wheel size. If NOHZ is configured we allocate two
183 * wheels so we have a separate storage for the deferrable timers.
184 */
185 #define WHEEL_SIZE (LVL_SIZE * LVL_DEPTH)
186
187 #ifdef CONFIG_NO_HZ_COMMON
188 # define NR_BASES 2
189 # define BASE_STD 0
190 # define BASE_DEF 1
191 #else
192 # define NR_BASES 1
193 # define BASE_STD 0
194 # define BASE_DEF 0
195 #endif
196
197 struct timer_base {
198 raw_spinlock_t lock;
199 struct timer_list *running_timer;
200 unsigned long clk;
201 unsigned long next_expiry;
202 unsigned int cpu;
203 bool migration_enabled;
204 bool nohz_active;
205 bool is_idle;
206 bool must_forward_clk;
207 DECLARE_BITMAP(pending_map, WHEEL_SIZE);
208 struct hlist_head vectors[WHEEL_SIZE];
209 } ____cacheline_aligned;
210
211 static DEFINE_PER_CPU(struct timer_base, timer_bases[NR_BASES]);
212
213 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
214 unsigned int sysctl_timer_migration = 1;
215
216 void timers_update_migration(bool update_nohz)
217 {
218 bool on = sysctl_timer_migration && tick_nohz_active;
219 unsigned int cpu;
220
221 /* Avoid the loop, if nothing to update */
222 if (this_cpu_read(timer_bases[BASE_STD].migration_enabled) == on)
223 return;
224
225 for_each_possible_cpu(cpu) {
226 per_cpu(timer_bases[BASE_STD].migration_enabled, cpu) = on;
227 per_cpu(timer_bases[BASE_DEF].migration_enabled, cpu) = on;
228 per_cpu(hrtimer_bases.migration_enabled, cpu) = on;
229 if (!update_nohz)
230 continue;
231 per_cpu(timer_bases[BASE_STD].nohz_active, cpu) = true;
232 per_cpu(timer_bases[BASE_DEF].nohz_active, cpu) = true;
233 per_cpu(hrtimer_bases.nohz_active, cpu) = true;
234 }
235 }
236
237 int timer_migration_handler(struct ctl_table *table, int write,
238 void __user *buffer, size_t *lenp,
239 loff_t *ppos)
240 {
241 static DEFINE_MUTEX(mutex);
242 int ret;
243
244 mutex_lock(&mutex);
245 ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
246 if (!ret && write)
247 timers_update_migration(false);
248 mutex_unlock(&mutex);
249 return ret;
250 }
251 #endif
252
253 static unsigned long round_jiffies_common(unsigned long j, int cpu,
254 bool force_up)
255 {
256 int rem;
257 unsigned long original = j;
258
259 /*
260 * We don't want all cpus firing their timers at once hitting the
261 * same lock or cachelines, so we skew each extra cpu with an extra
262 * 3 jiffies. This 3 jiffies came originally from the mm/ code which
263 * already did this.
264 * The skew is done by adding 3*cpunr, then round, then subtract this
265 * extra offset again.
266 */
267 j += cpu * 3;
268
269 rem = j % HZ;
270
271 /*
272 * If the target jiffie is just after a whole second (which can happen
273 * due to delays of the timer irq, long irq off times etc etc) then
274 * we should round down to the whole second, not up. Use 1/4th second
275 * as cutoff for this rounding as an extreme upper bound for this.
276 * But never round down if @force_up is set.
277 */
278 if (rem < HZ/4 && !force_up) /* round down */
279 j = j - rem;
280 else /* round up */
281 j = j - rem + HZ;
282
283 /* now that we have rounded, subtract the extra skew again */
284 j -= cpu * 3;
285
286 /*
287 * Make sure j is still in the future. Otherwise return the
288 * unmodified value.
289 */
290 return time_is_after_jiffies(j) ? j : original;
291 }
292
293 /**
294 * __round_jiffies - function to round jiffies to a full second
295 * @j: the time in (absolute) jiffies that should be rounded
296 * @cpu: the processor number on which the timeout will happen
297 *
298 * __round_jiffies() rounds an absolute time in the future (in jiffies)
299 * up or down to (approximately) full seconds. This is useful for timers
300 * for which the exact time they fire does not matter too much, as long as
301 * they fire approximately every X seconds.
302 *
303 * By rounding these timers to whole seconds, all such timers will fire
304 * at the same time, rather than at various times spread out. The goal
305 * of this is to have the CPU wake up less, which saves power.
306 *
307 * The exact rounding is skewed for each processor to avoid all
308 * processors firing at the exact same time, which could lead
309 * to lock contention or spurious cache line bouncing.
310 *
311 * The return value is the rounded version of the @j parameter.
312 */
313 unsigned long __round_jiffies(unsigned long j, int cpu)
314 {
315 return round_jiffies_common(j, cpu, false);
316 }
317 EXPORT_SYMBOL_GPL(__round_jiffies);
318
319 /**
320 * __round_jiffies_relative - function to round jiffies to a full second
321 * @j: the time in (relative) jiffies that should be rounded
322 * @cpu: the processor number on which the timeout will happen
323 *
324 * __round_jiffies_relative() rounds a time delta in the future (in jiffies)
325 * up or down to (approximately) full seconds. This is useful for timers
326 * for which the exact time they fire does not matter too much, as long as
327 * they fire approximately every X seconds.
328 *
329 * By rounding these timers to whole seconds, all such timers will fire
330 * at the same time, rather than at various times spread out. The goal
331 * of this is to have the CPU wake up less, which saves power.
332 *
333 * The exact rounding is skewed for each processor to avoid all
334 * processors firing at the exact same time, which could lead
335 * to lock contention or spurious cache line bouncing.
336 *
337 * The return value is the rounded version of the @j parameter.
338 */
339 unsigned long __round_jiffies_relative(unsigned long j, int cpu)
340 {
341 unsigned long j0 = jiffies;
342
343 /* Use j0 because jiffies might change while we run */
344 return round_jiffies_common(j + j0, cpu, false) - j0;
345 }
346 EXPORT_SYMBOL_GPL(__round_jiffies_relative);
347
348 /**
349 * round_jiffies - function to round jiffies to a full second
350 * @j: the time in (absolute) jiffies that should be rounded
351 *
352 * round_jiffies() rounds an absolute time in the future (in jiffies)
353 * up or down to (approximately) full seconds. This is useful for timers
354 * for which the exact time they fire does not matter too much, as long as
355 * they fire approximately every X seconds.
356 *
357 * By rounding these timers to whole seconds, all such timers will fire
358 * at the same time, rather than at various times spread out. The goal
359 * of this is to have the CPU wake up less, which saves power.
360 *
361 * The return value is the rounded version of the @j parameter.
362 */
363 unsigned long round_jiffies(unsigned long j)
364 {
365 return round_jiffies_common(j, raw_smp_processor_id(), false);
366 }
367 EXPORT_SYMBOL_GPL(round_jiffies);
368
369 /**
370 * round_jiffies_relative - function to round jiffies to a full second
371 * @j: the time in (relative) jiffies that should be rounded
372 *
373 * round_jiffies_relative() rounds a time delta in the future (in jiffies)
374 * up or down to (approximately) full seconds. This is useful for timers
375 * for which the exact time they fire does not matter too much, as long as
376 * they fire approximately every X seconds.
377 *
378 * By rounding these timers to whole seconds, all such timers will fire
379 * at the same time, rather than at various times spread out. The goal
380 * of this is to have the CPU wake up less, which saves power.
381 *
382 * The return value is the rounded version of the @j parameter.
383 */
384 unsigned long round_jiffies_relative(unsigned long j)
385 {
386 return __round_jiffies_relative(j, raw_smp_processor_id());
387 }
388 EXPORT_SYMBOL_GPL(round_jiffies_relative);
389
390 /**
391 * __round_jiffies_up - function to round jiffies up to a full second
392 * @j: the time in (absolute) jiffies that should be rounded
393 * @cpu: the processor number on which the timeout will happen
394 *
395 * This is the same as __round_jiffies() except that it will never
396 * round down. This is useful for timeouts for which the exact time
397 * of firing does not matter too much, as long as they don't fire too
398 * early.
399 */
400 unsigned long __round_jiffies_up(unsigned long j, int cpu)
401 {
402 return round_jiffies_common(j, cpu, true);
403 }
404 EXPORT_SYMBOL_GPL(__round_jiffies_up);
405
406 /**
407 * __round_jiffies_up_relative - function to round jiffies up to a full second
408 * @j: the time in (relative) jiffies that should be rounded
409 * @cpu: the processor number on which the timeout will happen
410 *
411 * This is the same as __round_jiffies_relative() except that it will never
412 * round down. This is useful for timeouts for which the exact time
413 * of firing does not matter too much, as long as they don't fire too
414 * early.
415 */
416 unsigned long __round_jiffies_up_relative(unsigned long j, int cpu)
417 {
418 unsigned long j0 = jiffies;
419
420 /* Use j0 because jiffies might change while we run */
421 return round_jiffies_common(j + j0, cpu, true) - j0;
422 }
423 EXPORT_SYMBOL_GPL(__round_jiffies_up_relative);
424
425 /**
426 * round_jiffies_up - function to round jiffies up to a full second
427 * @j: the time in (absolute) jiffies that should be rounded
428 *
429 * This is the same as round_jiffies() except that it will never
430 * round down. This is useful for timeouts for which the exact time
431 * of firing does not matter too much, as long as they don't fire too
432 * early.
433 */
434 unsigned long round_jiffies_up(unsigned long j)
435 {
436 return round_jiffies_common(j, raw_smp_processor_id(), true);
437 }
438 EXPORT_SYMBOL_GPL(round_jiffies_up);
439
440 /**
441 * round_jiffies_up_relative - function to round jiffies up to a full second
442 * @j: the time in (relative) jiffies that should be rounded
443 *
444 * This is the same as round_jiffies_relative() except that it will never
445 * round down. This is useful for timeouts for which the exact time
446 * of firing does not matter too much, as long as they don't fire too
447 * early.
448 */
449 unsigned long round_jiffies_up_relative(unsigned long j)
450 {
451 return __round_jiffies_up_relative(j, raw_smp_processor_id());
452 }
453 EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
454
455
456 static inline unsigned int timer_get_idx(struct timer_list *timer)
457 {
458 return (timer->flags & TIMER_ARRAYMASK) >> TIMER_ARRAYSHIFT;
459 }
460
461 static inline void timer_set_idx(struct timer_list *timer, unsigned int idx)
462 {
463 timer->flags = (timer->flags & ~TIMER_ARRAYMASK) |
464 idx << TIMER_ARRAYSHIFT;
465 }
466
467 /*
468 * Helper function to calculate the array index for a given expiry
469 * time.
470 */
471 static inline unsigned calc_index(unsigned expires, unsigned lvl)
472 {
473 expires = (expires + LVL_GRAN(lvl)) >> LVL_SHIFT(lvl);
474 return LVL_OFFS(lvl) + (expires & LVL_MASK);
475 }
476
477 static int calc_wheel_index(unsigned long expires, unsigned long clk)
478 {
479 unsigned long delta = expires - clk;
480 unsigned int idx;
481
482 if (delta < LVL_START(1)) {
483 idx = calc_index(expires, 0);
484 } else if (delta < LVL_START(2)) {
485 idx = calc_index(expires, 1);
486 } else if (delta < LVL_START(3)) {
487 idx = calc_index(expires, 2);
488 } else if (delta < LVL_START(4)) {
489 idx = calc_index(expires, 3);
490 } else if (delta < LVL_START(5)) {
491 idx = calc_index(expires, 4);
492 } else if (delta < LVL_START(6)) {
493 idx = calc_index(expires, 5);
494 } else if (delta < LVL_START(7)) {
495 idx = calc_index(expires, 6);
496 } else if (LVL_DEPTH > 8 && delta < LVL_START(8)) {
497 idx = calc_index(expires, 7);
498 } else if ((long) delta < 0) {
499 idx = clk & LVL_MASK;
500 } else {
501 /*
502 * Force expire obscene large timeouts to expire at the
503 * capacity limit of the wheel.
504 */
505 if (expires >= WHEEL_TIMEOUT_CUTOFF)
506 expires = WHEEL_TIMEOUT_MAX;
507
508 idx = calc_index(expires, LVL_DEPTH - 1);
509 }
510 return idx;
511 }
512
513 /*
514 * Enqueue the timer into the hash bucket, mark it pending in
515 * the bitmap and store the index in the timer flags.
516 */
517 static void enqueue_timer(struct timer_base *base, struct timer_list *timer,
518 unsigned int idx)
519 {
520 hlist_add_head(&timer->entry, base->vectors + idx);
521 __set_bit(idx, base->pending_map);
522 timer_set_idx(timer, idx);
523 }
524
525 static void
526 __internal_add_timer(struct timer_base *base, struct timer_list *timer)
527 {
528 unsigned int idx;
529
530 idx = calc_wheel_index(timer->expires, base->clk);
531 enqueue_timer(base, timer, idx);
532 }
533
534 static void
535 trigger_dyntick_cpu(struct timer_base *base, struct timer_list *timer)
536 {
537 if (!IS_ENABLED(CONFIG_NO_HZ_COMMON) || !base->nohz_active)
538 return;
539
540 /*
541 * TODO: This wants some optimizing similar to the code below, but we
542 * will do that when we switch from push to pull for deferrable timers.
543 */
544 if (timer->flags & TIMER_DEFERRABLE) {
545 if (tick_nohz_full_cpu(base->cpu))
546 wake_up_nohz_cpu(base->cpu);
547 return;
548 }
549
550 /*
551 * We might have to IPI the remote CPU if the base is idle and the
552 * timer is not deferrable. If the other CPU is on the way to idle
553 * then it can't set base->is_idle as we hold the base lock:
554 */
555 if (!base->is_idle)
556 return;
557
558 /* Check whether this is the new first expiring timer: */
559 if (time_after_eq(timer->expires, base->next_expiry))
560 return;
561
562 /*
563 * Set the next expiry time and kick the CPU so it can reevaluate the
564 * wheel:
565 */
566 base->next_expiry = timer->expires;
567 wake_up_nohz_cpu(base->cpu);
568 }
569
570 static void
571 internal_add_timer(struct timer_base *base, struct timer_list *timer)
572 {
573 __internal_add_timer(base, timer);
574 trigger_dyntick_cpu(base, timer);
575 }
576
577 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
578
579 static struct debug_obj_descr timer_debug_descr;
580
581 static void *timer_debug_hint(void *addr)
582 {
583 return ((struct timer_list *) addr)->function;
584 }
585
586 static bool timer_is_static_object(void *addr)
587 {
588 struct timer_list *timer = addr;
589
590 return (timer->entry.pprev == NULL &&
591 timer->entry.next == TIMER_ENTRY_STATIC);
592 }
593
594 /*
595 * fixup_init is called when:
596 * - an active object is initialized
597 */
598 static bool timer_fixup_init(void *addr, enum debug_obj_state state)
599 {
600 struct timer_list *timer = addr;
601
602 switch (state) {
603 case ODEBUG_STATE_ACTIVE:
604 del_timer_sync(timer);
605 debug_object_init(timer, &timer_debug_descr);
606 return true;
607 default:
608 return false;
609 }
610 }
611
612 /* Stub timer callback for improperly used timers. */
613 static void stub_timer(unsigned long data)
614 {
615 WARN_ON(1);
616 }
617
618 /*
619 * fixup_activate is called when:
620 * - an active object is activated
621 * - an unknown non-static object is activated
622 */
623 static bool timer_fixup_activate(void *addr, enum debug_obj_state state)
624 {
625 struct timer_list *timer = addr;
626
627 switch (state) {
628 case ODEBUG_STATE_NOTAVAILABLE:
629 setup_timer(timer, stub_timer, 0);
630 return true;
631
632 case ODEBUG_STATE_ACTIVE:
633 WARN_ON(1);
634
635 default:
636 return false;
637 }
638 }
639
640 /*
641 * fixup_free is called when:
642 * - an active object is freed
643 */
644 static bool timer_fixup_free(void *addr, enum debug_obj_state state)
645 {
646 struct timer_list *timer = addr;
647
648 switch (state) {
649 case ODEBUG_STATE_ACTIVE:
650 del_timer_sync(timer);
651 debug_object_free(timer, &timer_debug_descr);
652 return true;
653 default:
654 return false;
655 }
656 }
657
658 /*
659 * fixup_assert_init is called when:
660 * - an untracked/uninit-ed object is found
661 */
662 static bool timer_fixup_assert_init(void *addr, enum debug_obj_state state)
663 {
664 struct timer_list *timer = addr;
665
666 switch (state) {
667 case ODEBUG_STATE_NOTAVAILABLE:
668 setup_timer(timer, stub_timer, 0);
669 return true;
670 default:
671 return false;
672 }
673 }
674
675 static struct debug_obj_descr timer_debug_descr = {
676 .name = "timer_list",
677 .debug_hint = timer_debug_hint,
678 .is_static_object = timer_is_static_object,
679 .fixup_init = timer_fixup_init,
680 .fixup_activate = timer_fixup_activate,
681 .fixup_free = timer_fixup_free,
682 .fixup_assert_init = timer_fixup_assert_init,
683 };
684
685 static inline void debug_timer_init(struct timer_list *timer)
686 {
687 debug_object_init(timer, &timer_debug_descr);
688 }
689
690 static inline void debug_timer_activate(struct timer_list *timer)
691 {
692 debug_object_activate(timer, &timer_debug_descr);
693 }
694
695 static inline void debug_timer_deactivate(struct timer_list *timer)
696 {
697 debug_object_deactivate(timer, &timer_debug_descr);
698 }
699
700 static inline void debug_timer_free(struct timer_list *timer)
701 {
702 debug_object_free(timer, &timer_debug_descr);
703 }
704
705 static inline void debug_timer_assert_init(struct timer_list *timer)
706 {
707 debug_object_assert_init(timer, &timer_debug_descr);
708 }
709
710 static void do_init_timer(struct timer_list *timer, unsigned int flags,
711 const char *name, struct lock_class_key *key);
712
713 void init_timer_on_stack_key(struct timer_list *timer, unsigned int flags,
714 const char *name, struct lock_class_key *key)
715 {
716 debug_object_init_on_stack(timer, &timer_debug_descr);
717 do_init_timer(timer, flags, name, key);
718 }
719 EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
720
721 void destroy_timer_on_stack(struct timer_list *timer)
722 {
723 debug_object_free(timer, &timer_debug_descr);
724 }
725 EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
726
727 #else
728 static inline void debug_timer_init(struct timer_list *timer) { }
729 static inline void debug_timer_activate(struct timer_list *timer) { }
730 static inline void debug_timer_deactivate(struct timer_list *timer) { }
731 static inline void debug_timer_assert_init(struct timer_list *timer) { }
732 #endif
733
734 static inline void debug_init(struct timer_list *timer)
735 {
736 debug_timer_init(timer);
737 trace_timer_init(timer);
738 }
739
740 static inline void
741 debug_activate(struct timer_list *timer, unsigned long expires)
742 {
743 debug_timer_activate(timer);
744 trace_timer_start(timer, expires, timer->flags);
745 }
746
747 static inline void debug_deactivate(struct timer_list *timer)
748 {
749 debug_timer_deactivate(timer);
750 trace_timer_cancel(timer);
751 }
752
753 static inline void debug_assert_init(struct timer_list *timer)
754 {
755 debug_timer_assert_init(timer);
756 }
757
758 static void do_init_timer(struct timer_list *timer, unsigned int flags,
759 const char *name, struct lock_class_key *key)
760 {
761 timer->entry.pprev = NULL;
762 timer->flags = flags | raw_smp_processor_id();
763 lockdep_init_map(&timer->lockdep_map, name, key, 0);
764 }
765
766 /**
767 * init_timer_key - initialize a timer
768 * @timer: the timer to be initialized
769 * @flags: timer flags
770 * @name: name of the timer
771 * @key: lockdep class key of the fake lock used for tracking timer
772 * sync lock dependencies
773 *
774 * init_timer_key() must be done to a timer prior calling *any* of the
775 * other timer functions.
776 */
777 void init_timer_key(struct timer_list *timer, unsigned int flags,
778 const char *name, struct lock_class_key *key)
779 {
780 debug_init(timer);
781 do_init_timer(timer, flags, name, key);
782 }
783 EXPORT_SYMBOL(init_timer_key);
784
785 static inline void detach_timer(struct timer_list *timer, bool clear_pending)
786 {
787 struct hlist_node *entry = &timer->entry;
788
789 debug_deactivate(timer);
790
791 __hlist_del(entry);
792 if (clear_pending)
793 entry->pprev = NULL;
794 entry->next = LIST_POISON2;
795 }
796
797 static int detach_if_pending(struct timer_list *timer, struct timer_base *base,
798 bool clear_pending)
799 {
800 unsigned idx = timer_get_idx(timer);
801
802 if (!timer_pending(timer))
803 return 0;
804
805 if (hlist_is_singular_node(&timer->entry, base->vectors + idx))
806 __clear_bit(idx, base->pending_map);
807
808 detach_timer(timer, clear_pending);
809 return 1;
810 }
811
812 static inline struct timer_base *get_timer_cpu_base(u32 tflags, u32 cpu)
813 {
814 struct timer_base *base = per_cpu_ptr(&timer_bases[BASE_STD], cpu);
815
816 /*
817 * If the timer is deferrable and nohz is active then we need to use
818 * the deferrable base.
819 */
820 if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && base->nohz_active &&
821 (tflags & TIMER_DEFERRABLE))
822 base = per_cpu_ptr(&timer_bases[BASE_DEF], cpu);
823 return base;
824 }
825
826 static inline struct timer_base *get_timer_this_cpu_base(u32 tflags)
827 {
828 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
829
830 /*
831 * If the timer is deferrable and nohz is active then we need to use
832 * the deferrable base.
833 */
834 if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && base->nohz_active &&
835 (tflags & TIMER_DEFERRABLE))
836 base = this_cpu_ptr(&timer_bases[BASE_DEF]);
837 return base;
838 }
839
840 static inline struct timer_base *get_timer_base(u32 tflags)
841 {
842 return get_timer_cpu_base(tflags, tflags & TIMER_CPUMASK);
843 }
844
845 #ifdef CONFIG_NO_HZ_COMMON
846 static inline struct timer_base *
847 get_target_base(struct timer_base *base, unsigned tflags)
848 {
849 #ifdef CONFIG_SMP
850 if ((tflags & TIMER_PINNED) || !base->migration_enabled)
851 return get_timer_this_cpu_base(tflags);
852 return get_timer_cpu_base(tflags, get_nohz_timer_target());
853 #else
854 return get_timer_this_cpu_base(tflags);
855 #endif
856 }
857
858 static inline void forward_timer_base(struct timer_base *base)
859 {
860 unsigned long jnow;
861
862 /*
863 * We only forward the base when we are idle or have just come out of
864 * idle (must_forward_clk logic), and have a delta between base clock
865 * and jiffies. In the common case, run_timers will take care of it.
866 */
867 if (likely(!base->must_forward_clk))
868 return;
869
870 jnow = READ_ONCE(jiffies);
871 base->must_forward_clk = base->is_idle;
872 if ((long)(jnow - base->clk) < 2)
873 return;
874
875 /*
876 * If the next expiry value is > jiffies, then we fast forward to
877 * jiffies otherwise we forward to the next expiry value.
878 */
879 if (time_after(base->next_expiry, jnow))
880 base->clk = jnow;
881 else
882 base->clk = base->next_expiry;
883 }
884 #else
885 static inline struct timer_base *
886 get_target_base(struct timer_base *base, unsigned tflags)
887 {
888 return get_timer_this_cpu_base(tflags);
889 }
890
891 static inline void forward_timer_base(struct timer_base *base) { }
892 #endif
893
894
895 /*
896 * We are using hashed locking: Holding per_cpu(timer_bases[x]).lock means
897 * that all timers which are tied to this base are locked, and the base itself
898 * is locked too.
899 *
900 * So __run_timers/migrate_timers can safely modify all timers which could
901 * be found in the base->vectors array.
902 *
903 * When a timer is migrating then the TIMER_MIGRATING flag is set and we need
904 * to wait until the migration is done.
905 */
906 static struct timer_base *lock_timer_base(struct timer_list *timer,
907 unsigned long *flags)
908 __acquires(timer->base->lock)
909 {
910 for (;;) {
911 struct timer_base *base;
912 u32 tf;
913
914 /*
915 * We need to use READ_ONCE() here, otherwise the compiler
916 * might re-read @tf between the check for TIMER_MIGRATING
917 * and spin_lock().
918 */
919 tf = READ_ONCE(timer->flags);
920
921 if (!(tf & TIMER_MIGRATING)) {
922 base = get_timer_base(tf);
923 raw_spin_lock_irqsave(&base->lock, *flags);
924 if (timer->flags == tf)
925 return base;
926 raw_spin_unlock_irqrestore(&base->lock, *flags);
927 }
928 cpu_relax();
929 }
930 }
931
932 static inline int
933 __mod_timer(struct timer_list *timer, unsigned long expires, bool pending_only)
934 {
935 struct timer_base *base, *new_base;
936 unsigned int idx = UINT_MAX;
937 unsigned long clk = 0, flags;
938 int ret = 0;
939
940 BUG_ON(!timer->function);
941
942 /*
943 * This is a common optimization triggered by the networking code - if
944 * the timer is re-modified to have the same timeout or ends up in the
945 * same array bucket then just return:
946 */
947 if (timer_pending(timer)) {
948 /*
949 * The downside of this optimization is that it can result in
950 * larger granularity than you would get from adding a new
951 * timer with this expiry.
952 */
953 if (timer->expires == expires)
954 return 1;
955
956 /*
957 * We lock timer base and calculate the bucket index right
958 * here. If the timer ends up in the same bucket, then we
959 * just update the expiry time and avoid the whole
960 * dequeue/enqueue dance.
961 */
962 base = lock_timer_base(timer, &flags);
963 forward_timer_base(base);
964
965 clk = base->clk;
966 idx = calc_wheel_index(expires, clk);
967
968 /*
969 * Retrieve and compare the array index of the pending
970 * timer. If it matches set the expiry to the new value so a
971 * subsequent call will exit in the expires check above.
972 */
973 if (idx == timer_get_idx(timer)) {
974 timer->expires = expires;
975 ret = 1;
976 goto out_unlock;
977 }
978 } else {
979 base = lock_timer_base(timer, &flags);
980 forward_timer_base(base);
981 }
982
983 ret = detach_if_pending(timer, base, false);
984 if (!ret && pending_only)
985 goto out_unlock;
986
987 debug_activate(timer, expires);
988
989 new_base = get_target_base(base, timer->flags);
990
991 if (base != new_base) {
992 /*
993 * We are trying to schedule the timer on the new base.
994 * However we can't change timer's base while it is running,
995 * otherwise del_timer_sync() can't detect that the timer's
996 * handler yet has not finished. This also guarantees that the
997 * timer is serialized wrt itself.
998 */
999 if (likely(base->running_timer != timer)) {
1000 /* See the comment in lock_timer_base() */
1001 timer->flags |= TIMER_MIGRATING;
1002
1003 raw_spin_unlock(&base->lock);
1004 base = new_base;
1005 raw_spin_lock(&base->lock);
1006 WRITE_ONCE(timer->flags,
1007 (timer->flags & ~TIMER_BASEMASK) | base->cpu);
1008 forward_timer_base(base);
1009 }
1010 }
1011
1012 timer->expires = expires;
1013 /*
1014 * If 'idx' was calculated above and the base time did not advance
1015 * between calculating 'idx' and possibly switching the base, only
1016 * enqueue_timer() and trigger_dyntick_cpu() is required. Otherwise
1017 * we need to (re)calculate the wheel index via
1018 * internal_add_timer().
1019 */
1020 if (idx != UINT_MAX && clk == base->clk) {
1021 enqueue_timer(base, timer, idx);
1022 trigger_dyntick_cpu(base, timer);
1023 } else {
1024 internal_add_timer(base, timer);
1025 }
1026
1027 out_unlock:
1028 raw_spin_unlock_irqrestore(&base->lock, flags);
1029
1030 return ret;
1031 }
1032
1033 /**
1034 * mod_timer_pending - modify a pending timer's timeout
1035 * @timer: the pending timer to be modified
1036 * @expires: new timeout in jiffies
1037 *
1038 * mod_timer_pending() is the same for pending timers as mod_timer(),
1039 * but will not re-activate and modify already deleted timers.
1040 *
1041 * It is useful for unserialized use of timers.
1042 */
1043 int mod_timer_pending(struct timer_list *timer, unsigned long expires)
1044 {
1045 return __mod_timer(timer, expires, true);
1046 }
1047 EXPORT_SYMBOL(mod_timer_pending);
1048
1049 /**
1050 * mod_timer - modify a timer's timeout
1051 * @timer: the timer to be modified
1052 * @expires: new timeout in jiffies
1053 *
1054 * mod_timer() is a more efficient way to update the expire field of an
1055 * active timer (if the timer is inactive it will be activated)
1056 *
1057 * mod_timer(timer, expires) is equivalent to:
1058 *
1059 * del_timer(timer); timer->expires = expires; add_timer(timer);
1060 *
1061 * Note that if there are multiple unserialized concurrent users of the
1062 * same timer, then mod_timer() is the only safe way to modify the timeout,
1063 * since add_timer() cannot modify an already running timer.
1064 *
1065 * The function returns whether it has modified a pending timer or not.
1066 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
1067 * active timer returns 1.)
1068 */
1069 int mod_timer(struct timer_list *timer, unsigned long expires)
1070 {
1071 return __mod_timer(timer, expires, false);
1072 }
1073 EXPORT_SYMBOL(mod_timer);
1074
1075 /**
1076 * add_timer - start a timer
1077 * @timer: the timer to be added
1078 *
1079 * The kernel will do a ->function(->data) callback from the
1080 * timer interrupt at the ->expires point in the future. The
1081 * current time is 'jiffies'.
1082 *
1083 * The timer's ->expires, ->function (and if the handler uses it, ->data)
1084 * fields must be set prior calling this function.
1085 *
1086 * Timers with an ->expires field in the past will be executed in the next
1087 * timer tick.
1088 */
1089 void add_timer(struct timer_list *timer)
1090 {
1091 BUG_ON(timer_pending(timer));
1092 mod_timer(timer, timer->expires);
1093 }
1094 EXPORT_SYMBOL(add_timer);
1095
1096 /**
1097 * add_timer_on - start a timer on a particular CPU
1098 * @timer: the timer to be added
1099 * @cpu: the CPU to start it on
1100 *
1101 * This is not very scalable on SMP. Double adds are not possible.
1102 */
1103 void add_timer_on(struct timer_list *timer, int cpu)
1104 {
1105 struct timer_base *new_base, *base;
1106 unsigned long flags;
1107
1108 BUG_ON(timer_pending(timer) || !timer->function);
1109
1110 new_base = get_timer_cpu_base(timer->flags, cpu);
1111
1112 /*
1113 * If @timer was on a different CPU, it should be migrated with the
1114 * old base locked to prevent other operations proceeding with the
1115 * wrong base locked. See lock_timer_base().
1116 */
1117 base = lock_timer_base(timer, &flags);
1118 if (base != new_base) {
1119 timer->flags |= TIMER_MIGRATING;
1120
1121 raw_spin_unlock(&base->lock);
1122 base = new_base;
1123 raw_spin_lock(&base->lock);
1124 WRITE_ONCE(timer->flags,
1125 (timer->flags & ~TIMER_BASEMASK) | cpu);
1126 }
1127 forward_timer_base(base);
1128
1129 debug_activate(timer, timer->expires);
1130 internal_add_timer(base, timer);
1131 raw_spin_unlock_irqrestore(&base->lock, flags);
1132 }
1133 EXPORT_SYMBOL_GPL(add_timer_on);
1134
1135 /**
1136 * del_timer - deactivate a timer.
1137 * @timer: the timer to be deactivated
1138 *
1139 * del_timer() deactivates a timer - this works on both active and inactive
1140 * timers.
1141 *
1142 * The function returns whether it has deactivated a pending timer or not.
1143 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
1144 * active timer returns 1.)
1145 */
1146 int del_timer(struct timer_list *timer)
1147 {
1148 struct timer_base *base;
1149 unsigned long flags;
1150 int ret = 0;
1151
1152 debug_assert_init(timer);
1153
1154 if (timer_pending(timer)) {
1155 base = lock_timer_base(timer, &flags);
1156 ret = detach_if_pending(timer, base, true);
1157 raw_spin_unlock_irqrestore(&base->lock, flags);
1158 }
1159
1160 return ret;
1161 }
1162 EXPORT_SYMBOL(del_timer);
1163
1164 /**
1165 * try_to_del_timer_sync - Try to deactivate a timer
1166 * @timer: timer to delete
1167 *
1168 * This function tries to deactivate a timer. Upon successful (ret >= 0)
1169 * exit the timer is not queued and the handler is not running on any CPU.
1170 */
1171 int try_to_del_timer_sync(struct timer_list *timer)
1172 {
1173 struct timer_base *base;
1174 unsigned long flags;
1175 int ret = -1;
1176
1177 debug_assert_init(timer);
1178
1179 base = lock_timer_base(timer, &flags);
1180
1181 if (base->running_timer != timer)
1182 ret = detach_if_pending(timer, base, true);
1183
1184 raw_spin_unlock_irqrestore(&base->lock, flags);
1185
1186 return ret;
1187 }
1188 EXPORT_SYMBOL(try_to_del_timer_sync);
1189
1190 #ifdef CONFIG_SMP
1191 /**
1192 * del_timer_sync - deactivate a timer and wait for the handler to finish.
1193 * @timer: the timer to be deactivated
1194 *
1195 * This function only differs from del_timer() on SMP: besides deactivating
1196 * the timer it also makes sure the handler has finished executing on other
1197 * CPUs.
1198 *
1199 * Synchronization rules: Callers must prevent restarting of the timer,
1200 * otherwise this function is meaningless. It must not be called from
1201 * interrupt contexts unless the timer is an irqsafe one. The caller must
1202 * not hold locks which would prevent completion of the timer's
1203 * handler. The timer's handler must not call add_timer_on(). Upon exit the
1204 * timer is not queued and the handler is not running on any CPU.
1205 *
1206 * Note: For !irqsafe timers, you must not hold locks that are held in
1207 * interrupt context while calling this function. Even if the lock has
1208 * nothing to do with the timer in question. Here's why:
1209 *
1210 * CPU0 CPU1
1211 * ---- ----
1212 * <SOFTIRQ>
1213 * call_timer_fn();
1214 * base->running_timer = mytimer;
1215 * spin_lock_irq(somelock);
1216 * <IRQ>
1217 * spin_lock(somelock);
1218 * del_timer_sync(mytimer);
1219 * while (base->running_timer == mytimer);
1220 *
1221 * Now del_timer_sync() will never return and never release somelock.
1222 * The interrupt on the other CPU is waiting to grab somelock but
1223 * it has interrupted the softirq that CPU0 is waiting to finish.
1224 *
1225 * The function returns whether it has deactivated a pending timer or not.
1226 */
1227 int del_timer_sync(struct timer_list *timer)
1228 {
1229 #ifdef CONFIG_LOCKDEP
1230 unsigned long flags;
1231
1232 /*
1233 * If lockdep gives a backtrace here, please reference
1234 * the synchronization rules above.
1235 */
1236 local_irq_save(flags);
1237 lock_map_acquire(&timer->lockdep_map);
1238 lock_map_release(&timer->lockdep_map);
1239 local_irq_restore(flags);
1240 #endif
1241 /*
1242 * don't use it in hardirq context, because it
1243 * could lead to deadlock.
1244 */
1245 WARN_ON(in_irq() && !(timer->flags & TIMER_IRQSAFE));
1246 for (;;) {
1247 int ret = try_to_del_timer_sync(timer);
1248 if (ret >= 0)
1249 return ret;
1250 cpu_relax();
1251 }
1252 }
1253 EXPORT_SYMBOL(del_timer_sync);
1254 #endif
1255
1256 static void call_timer_fn(struct timer_list *timer, void (*fn)(unsigned long),
1257 unsigned long data)
1258 {
1259 int count = preempt_count();
1260
1261 #ifdef CONFIG_LOCKDEP
1262 /*
1263 * It is permissible to free the timer from inside the
1264 * function that is called from it, this we need to take into
1265 * account for lockdep too. To avoid bogus "held lock freed"
1266 * warnings as well as problems when looking into
1267 * timer->lockdep_map, make a copy and use that here.
1268 */
1269 struct lockdep_map lockdep_map;
1270
1271 lockdep_copy_map(&lockdep_map, &timer->lockdep_map);
1272 #endif
1273 /*
1274 * Couple the lock chain with the lock chain at
1275 * del_timer_sync() by acquiring the lock_map around the fn()
1276 * call here and in del_timer_sync().
1277 */
1278 lock_map_acquire(&lockdep_map);
1279
1280 trace_timer_expire_entry(timer);
1281 fn(data);
1282 trace_timer_expire_exit(timer);
1283
1284 lock_map_release(&lockdep_map);
1285
1286 if (count != preempt_count()) {
1287 WARN_ONCE(1, "timer: %pF preempt leak: %08x -> %08x\n",
1288 fn, count, preempt_count());
1289 /*
1290 * Restore the preempt count. That gives us a decent
1291 * chance to survive and extract information. If the
1292 * callback kept a lock held, bad luck, but not worse
1293 * than the BUG() we had.
1294 */
1295 preempt_count_set(count);
1296 }
1297 }
1298
1299 static void expire_timers(struct timer_base *base, struct hlist_head *head)
1300 {
1301 while (!hlist_empty(head)) {
1302 struct timer_list *timer;
1303 void (*fn)(unsigned long);
1304 unsigned long data;
1305
1306 timer = hlist_entry(head->first, struct timer_list, entry);
1307
1308 base->running_timer = timer;
1309 detach_timer(timer, true);
1310
1311 fn = timer->function;
1312 data = timer->data;
1313
1314 if (timer->flags & TIMER_IRQSAFE) {
1315 raw_spin_unlock(&base->lock);
1316 call_timer_fn(timer, fn, data);
1317 raw_spin_lock(&base->lock);
1318 } else {
1319 raw_spin_unlock_irq(&base->lock);
1320 call_timer_fn(timer, fn, data);
1321 raw_spin_lock_irq(&base->lock);
1322 }
1323 }
1324 }
1325
1326 static int __collect_expired_timers(struct timer_base *base,
1327 struct hlist_head *heads)
1328 {
1329 unsigned long clk = base->clk;
1330 struct hlist_head *vec;
1331 int i, levels = 0;
1332 unsigned int idx;
1333
1334 for (i = 0; i < LVL_DEPTH; i++) {
1335 idx = (clk & LVL_MASK) + i * LVL_SIZE;
1336
1337 if (__test_and_clear_bit(idx, base->pending_map)) {
1338 vec = base->vectors + idx;
1339 hlist_move_list(vec, heads++);
1340 levels++;
1341 }
1342 /* Is it time to look at the next level? */
1343 if (clk & LVL_CLK_MASK)
1344 break;
1345 /* Shift clock for the next level granularity */
1346 clk >>= LVL_CLK_SHIFT;
1347 }
1348 return levels;
1349 }
1350
1351 #ifdef CONFIG_NO_HZ_COMMON
1352 /*
1353 * Find the next pending bucket of a level. Search from level start (@offset)
1354 * + @clk upwards and if nothing there, search from start of the level
1355 * (@offset) up to @offset + clk.
1356 */
1357 static int next_pending_bucket(struct timer_base *base, unsigned offset,
1358 unsigned clk)
1359 {
1360 unsigned pos, start = offset + clk;
1361 unsigned end = offset + LVL_SIZE;
1362
1363 pos = find_next_bit(base->pending_map, end, start);
1364 if (pos < end)
1365 return pos - start;
1366
1367 pos = find_next_bit(base->pending_map, start, offset);
1368 return pos < start ? pos + LVL_SIZE - start : -1;
1369 }
1370
1371 /*
1372 * Search the first expiring timer in the various clock levels. Caller must
1373 * hold base->lock.
1374 */
1375 static unsigned long __next_timer_interrupt(struct timer_base *base)
1376 {
1377 unsigned long clk, next, adj;
1378 unsigned lvl, offset = 0;
1379
1380 next = base->clk + NEXT_TIMER_MAX_DELTA;
1381 clk = base->clk;
1382 for (lvl = 0; lvl < LVL_DEPTH; lvl++, offset += LVL_SIZE) {
1383 int pos = next_pending_bucket(base, offset, clk & LVL_MASK);
1384
1385 if (pos >= 0) {
1386 unsigned long tmp = clk + (unsigned long) pos;
1387
1388 tmp <<= LVL_SHIFT(lvl);
1389 if (time_before(tmp, next))
1390 next = tmp;
1391 }
1392 /*
1393 * Clock for the next level. If the current level clock lower
1394 * bits are zero, we look at the next level as is. If not we
1395 * need to advance it by one because that's going to be the
1396 * next expiring bucket in that level. base->clk is the next
1397 * expiring jiffie. So in case of:
1398 *
1399 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1400 * 0 0 0 0 0 0
1401 *
1402 * we have to look at all levels @index 0. With
1403 *
1404 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1405 * 0 0 0 0 0 2
1406 *
1407 * LVL0 has the next expiring bucket @index 2. The upper
1408 * levels have the next expiring bucket @index 1.
1409 *
1410 * In case that the propagation wraps the next level the same
1411 * rules apply:
1412 *
1413 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1414 * 0 0 0 0 F 2
1415 *
1416 * So after looking at LVL0 we get:
1417 *
1418 * LVL5 LVL4 LVL3 LVL2 LVL1
1419 * 0 0 0 1 0
1420 *
1421 * So no propagation from LVL1 to LVL2 because that happened
1422 * with the add already, but then we need to propagate further
1423 * from LVL2 to LVL3.
1424 *
1425 * So the simple check whether the lower bits of the current
1426 * level are 0 or not is sufficient for all cases.
1427 */
1428 adj = clk & LVL_CLK_MASK ? 1 : 0;
1429 clk >>= LVL_CLK_SHIFT;
1430 clk += adj;
1431 }
1432 return next;
1433 }
1434
1435 /*
1436 * Check, if the next hrtimer event is before the next timer wheel
1437 * event:
1438 */
1439 static u64 cmp_next_hrtimer_event(u64 basem, u64 expires)
1440 {
1441 u64 nextevt = hrtimer_get_next_event();
1442
1443 /*
1444 * If high resolution timers are enabled
1445 * hrtimer_get_next_event() returns KTIME_MAX.
1446 */
1447 if (expires <= nextevt)
1448 return expires;
1449
1450 /*
1451 * If the next timer is already expired, return the tick base
1452 * time so the tick is fired immediately.
1453 */
1454 if (nextevt <= basem)
1455 return basem;
1456
1457 /*
1458 * Round up to the next jiffie. High resolution timers are
1459 * off, so the hrtimers are expired in the tick and we need to
1460 * make sure that this tick really expires the timer to avoid
1461 * a ping pong of the nohz stop code.
1462 *
1463 * Use DIV_ROUND_UP_ULL to prevent gcc calling __divdi3
1464 */
1465 return DIV_ROUND_UP_ULL(nextevt, TICK_NSEC) * TICK_NSEC;
1466 }
1467
1468 /**
1469 * get_next_timer_interrupt - return the time (clock mono) of the next timer
1470 * @basej: base time jiffies
1471 * @basem: base time clock monotonic
1472 *
1473 * Returns the tick aligned clock monotonic time of the next pending
1474 * timer or KTIME_MAX if no timer is pending.
1475 */
1476 u64 get_next_timer_interrupt(unsigned long basej, u64 basem)
1477 {
1478 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1479 u64 expires = KTIME_MAX;
1480 unsigned long nextevt;
1481 bool is_max_delta;
1482
1483 /*
1484 * Pretend that there is no timer pending if the cpu is offline.
1485 * Possible pending timers will be migrated later to an active cpu.
1486 */
1487 if (cpu_is_offline(smp_processor_id()))
1488 return expires;
1489
1490 raw_spin_lock(&base->lock);
1491 nextevt = __next_timer_interrupt(base);
1492 is_max_delta = (nextevt == base->clk + NEXT_TIMER_MAX_DELTA);
1493 base->next_expiry = nextevt;
1494 /*
1495 * We have a fresh next event. Check whether we can forward the
1496 * base. We can only do that when @basej is past base->clk
1497 * otherwise we might rewind base->clk.
1498 */
1499 if (time_after(basej, base->clk)) {
1500 if (time_after(nextevt, basej))
1501 base->clk = basej;
1502 else if (time_after(nextevt, base->clk))
1503 base->clk = nextevt;
1504 }
1505
1506 if (time_before_eq(nextevt, basej)) {
1507 expires = basem;
1508 base->is_idle = false;
1509 } else {
1510 if (!is_max_delta)
1511 expires = basem + (u64)(nextevt - basej) * TICK_NSEC;
1512 /*
1513 * If we expect to sleep more than a tick, mark the base idle.
1514 * Also the tick is stopped so any added timer must forward
1515 * the base clk itself to keep granularity small. This idle
1516 * logic is only maintained for the BASE_STD base, deferrable
1517 * timers may still see large granularity skew (by design).
1518 */
1519 if ((expires - basem) > TICK_NSEC) {
1520 base->must_forward_clk = true;
1521 base->is_idle = true;
1522 }
1523 }
1524 raw_spin_unlock(&base->lock);
1525
1526 return cmp_next_hrtimer_event(basem, expires);
1527 }
1528
1529 /**
1530 * timer_clear_idle - Clear the idle state of the timer base
1531 *
1532 * Called with interrupts disabled
1533 */
1534 void timer_clear_idle(void)
1535 {
1536 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1537
1538 /*
1539 * We do this unlocked. The worst outcome is a remote enqueue sending
1540 * a pointless IPI, but taking the lock would just make the window for
1541 * sending the IPI a few instructions smaller for the cost of taking
1542 * the lock in the exit from idle path.
1543 */
1544 base->is_idle = false;
1545 }
1546
1547 static int collect_expired_timers(struct timer_base *base,
1548 struct hlist_head *heads)
1549 {
1550 /*
1551 * NOHZ optimization. After a long idle sleep we need to forward the
1552 * base to current jiffies. Avoid a loop by searching the bitfield for
1553 * the next expiring timer.
1554 */
1555 if ((long)(jiffies - base->clk) > 2) {
1556 unsigned long next = __next_timer_interrupt(base);
1557
1558 /*
1559 * If the next timer is ahead of time forward to current
1560 * jiffies, otherwise forward to the next expiry time:
1561 */
1562 if (time_after(next, jiffies)) {
1563 /* The call site will increment clock! */
1564 base->clk = jiffies - 1;
1565 return 0;
1566 }
1567 base->clk = next;
1568 }
1569 return __collect_expired_timers(base, heads);
1570 }
1571 #else
1572 static inline int collect_expired_timers(struct timer_base *base,
1573 struct hlist_head *heads)
1574 {
1575 return __collect_expired_timers(base, heads);
1576 }
1577 #endif
1578
1579 /*
1580 * Called from the timer interrupt handler to charge one tick to the current
1581 * process. user_tick is 1 if the tick is user time, 0 for system.
1582 */
1583 void update_process_times(int user_tick)
1584 {
1585 struct task_struct *p = current;
1586
1587 /* Note: this timer irq context must be accounted for as well. */
1588 account_process_tick(p, user_tick);
1589 run_local_timers();
1590 rcu_check_callbacks(user_tick);
1591 #ifdef CONFIG_IRQ_WORK
1592 if (in_irq())
1593 irq_work_tick();
1594 #endif
1595 scheduler_tick();
1596 if (IS_ENABLED(CONFIG_POSIX_TIMERS))
1597 run_posix_cpu_timers(p);
1598 }
1599
1600 /**
1601 * __run_timers - run all expired timers (if any) on this CPU.
1602 * @base: the timer vector to be processed.
1603 */
1604 static inline void __run_timers(struct timer_base *base)
1605 {
1606 struct hlist_head heads[LVL_DEPTH];
1607 int levels;
1608
1609 if (!time_after_eq(jiffies, base->clk))
1610 return;
1611
1612 raw_spin_lock_irq(&base->lock);
1613
1614 while (time_after_eq(jiffies, base->clk)) {
1615
1616 levels = collect_expired_timers(base, heads);
1617 base->clk++;
1618
1619 while (levels--)
1620 expire_timers(base, heads + levels);
1621 }
1622 base->running_timer = NULL;
1623 raw_spin_unlock_irq(&base->lock);
1624 }
1625
1626 /*
1627 * This function runs timers and the timer-tq in bottom half context.
1628 */
1629 static __latent_entropy void run_timer_softirq(struct softirq_action *h)
1630 {
1631 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1632
1633 /*
1634 * must_forward_clk must be cleared before running timers so that any
1635 * timer functions that call mod_timer will not try to forward the
1636 * base. idle trcking / clock forwarding logic is only used with
1637 * BASE_STD timers.
1638 *
1639 * The deferrable base does not do idle tracking at all, so we do
1640 * not forward it. This can result in very large variations in
1641 * granularity for deferrable timers, but they can be deferred for
1642 * long periods due to idle.
1643 */
1644 base->must_forward_clk = false;
1645
1646 __run_timers(base);
1647 if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && base->nohz_active)
1648 __run_timers(this_cpu_ptr(&timer_bases[BASE_DEF]));
1649 }
1650
1651 /*
1652 * Called by the local, per-CPU timer interrupt on SMP.
1653 */
1654 void run_local_timers(void)
1655 {
1656 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1657
1658 hrtimer_run_queues();
1659 /* Raise the softirq only if required. */
1660 if (time_before(jiffies, base->clk)) {
1661 if (!IS_ENABLED(CONFIG_NO_HZ_COMMON) || !base->nohz_active)
1662 return;
1663 /* CPU is awake, so check the deferrable base. */
1664 base++;
1665 if (time_before(jiffies, base->clk))
1666 return;
1667 }
1668 raise_softirq(TIMER_SOFTIRQ);
1669 }
1670
1671 static void process_timeout(unsigned long __data)
1672 {
1673 wake_up_process((struct task_struct *)__data);
1674 }
1675
1676 /**
1677 * schedule_timeout - sleep until timeout
1678 * @timeout: timeout value in jiffies
1679 *
1680 * Make the current task sleep until @timeout jiffies have
1681 * elapsed. The routine will return immediately unless
1682 * the current task state has been set (see set_current_state()).
1683 *
1684 * You can set the task state as follows -
1685 *
1686 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1687 * pass before the routine returns unless the current task is explicitly
1688 * woken up, (e.g. by wake_up_process())".
1689 *
1690 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1691 * delivered to the current task or the current task is explicitly woken
1692 * up.
1693 *
1694 * The current task state is guaranteed to be TASK_RUNNING when this
1695 * routine returns.
1696 *
1697 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1698 * the CPU away without a bound on the timeout. In this case the return
1699 * value will be %MAX_SCHEDULE_TIMEOUT.
1700 *
1701 * Returns 0 when the timer has expired otherwise the remaining time in
1702 * jiffies will be returned. In all cases the return value is guaranteed
1703 * to be non-negative.
1704 */
1705 signed long __sched schedule_timeout(signed long timeout)
1706 {
1707 struct timer_list timer;
1708 unsigned long expire;
1709
1710 switch (timeout)
1711 {
1712 case MAX_SCHEDULE_TIMEOUT:
1713 /*
1714 * These two special cases are useful to be comfortable
1715 * in the caller. Nothing more. We could take
1716 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1717 * but I' d like to return a valid offset (>=0) to allow
1718 * the caller to do everything it want with the retval.
1719 */
1720 schedule();
1721 goto out;
1722 default:
1723 /*
1724 * Another bit of PARANOID. Note that the retval will be
1725 * 0 since no piece of kernel is supposed to do a check
1726 * for a negative retval of schedule_timeout() (since it
1727 * should never happens anyway). You just have the printk()
1728 * that will tell you if something is gone wrong and where.
1729 */
1730 if (timeout < 0) {
1731 printk(KERN_ERR "schedule_timeout: wrong timeout "
1732 "value %lx\n", timeout);
1733 dump_stack();
1734 current->state = TASK_RUNNING;
1735 goto out;
1736 }
1737 }
1738
1739 expire = timeout + jiffies;
1740
1741 setup_timer_on_stack(&timer, process_timeout, (unsigned long)current);
1742 __mod_timer(&timer, expire, false);
1743 schedule();
1744 del_singleshot_timer_sync(&timer);
1745
1746 /* Remove the timer from the object tracker */
1747 destroy_timer_on_stack(&timer);
1748
1749 timeout = expire - jiffies;
1750
1751 out:
1752 return timeout < 0 ? 0 : timeout;
1753 }
1754 EXPORT_SYMBOL(schedule_timeout);
1755
1756 /*
1757 * We can use __set_current_state() here because schedule_timeout() calls
1758 * schedule() unconditionally.
1759 */
1760 signed long __sched schedule_timeout_interruptible(signed long timeout)
1761 {
1762 __set_current_state(TASK_INTERRUPTIBLE);
1763 return schedule_timeout(timeout);
1764 }
1765 EXPORT_SYMBOL(schedule_timeout_interruptible);
1766
1767 signed long __sched schedule_timeout_killable(signed long timeout)
1768 {
1769 __set_current_state(TASK_KILLABLE);
1770 return schedule_timeout(timeout);
1771 }
1772 EXPORT_SYMBOL(schedule_timeout_killable);
1773
1774 signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1775 {
1776 __set_current_state(TASK_UNINTERRUPTIBLE);
1777 return schedule_timeout(timeout);
1778 }
1779 EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1780
1781 /*
1782 * Like schedule_timeout_uninterruptible(), except this task will not contribute
1783 * to load average.
1784 */
1785 signed long __sched schedule_timeout_idle(signed long timeout)
1786 {
1787 __set_current_state(TASK_IDLE);
1788 return schedule_timeout(timeout);
1789 }
1790 EXPORT_SYMBOL(schedule_timeout_idle);
1791
1792 #ifdef CONFIG_HOTPLUG_CPU
1793 static void migrate_timer_list(struct timer_base *new_base, struct hlist_head *head)
1794 {
1795 struct timer_list *timer;
1796 int cpu = new_base->cpu;
1797
1798 while (!hlist_empty(head)) {
1799 timer = hlist_entry(head->first, struct timer_list, entry);
1800 detach_timer(timer, false);
1801 timer->flags = (timer->flags & ~TIMER_BASEMASK) | cpu;
1802 internal_add_timer(new_base, timer);
1803 }
1804 }
1805
1806 int timers_dead_cpu(unsigned int cpu)
1807 {
1808 struct timer_base *old_base;
1809 struct timer_base *new_base;
1810 int b, i;
1811
1812 BUG_ON(cpu_online(cpu));
1813
1814 for (b = 0; b < NR_BASES; b++) {
1815 old_base = per_cpu_ptr(&timer_bases[b], cpu);
1816 new_base = get_cpu_ptr(&timer_bases[b]);
1817 /*
1818 * The caller is globally serialized and nobody else
1819 * takes two locks at once, deadlock is not possible.
1820 */
1821 raw_spin_lock_irq(&new_base->lock);
1822 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1823
1824 BUG_ON(old_base->running_timer);
1825
1826 for (i = 0; i < WHEEL_SIZE; i++)
1827 migrate_timer_list(new_base, old_base->vectors + i);
1828
1829 raw_spin_unlock(&old_base->lock);
1830 raw_spin_unlock_irq(&new_base->lock);
1831 put_cpu_ptr(&timer_bases);
1832 }
1833 return 0;
1834 }
1835
1836 #endif /* CONFIG_HOTPLUG_CPU */
1837
1838 static void __init init_timer_cpu(int cpu)
1839 {
1840 struct timer_base *base;
1841 int i;
1842
1843 for (i = 0; i < NR_BASES; i++) {
1844 base = per_cpu_ptr(&timer_bases[i], cpu);
1845 base->cpu = cpu;
1846 raw_spin_lock_init(&base->lock);
1847 base->clk = jiffies;
1848 }
1849 }
1850
1851 static void __init init_timer_cpus(void)
1852 {
1853 int cpu;
1854
1855 for_each_possible_cpu(cpu)
1856 init_timer_cpu(cpu);
1857 }
1858
1859 void __init init_timers(void)
1860 {
1861 init_timer_cpus();
1862 open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
1863 }
1864
1865 /**
1866 * msleep - sleep safely even with waitqueue interruptions
1867 * @msecs: Time in milliseconds to sleep for
1868 */
1869 void msleep(unsigned int msecs)
1870 {
1871 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1872
1873 while (timeout)
1874 timeout = schedule_timeout_uninterruptible(timeout);
1875 }
1876
1877 EXPORT_SYMBOL(msleep);
1878
1879 /**
1880 * msleep_interruptible - sleep waiting for signals
1881 * @msecs: Time in milliseconds to sleep for
1882 */
1883 unsigned long msleep_interruptible(unsigned int msecs)
1884 {
1885 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1886
1887 while (timeout && !signal_pending(current))
1888 timeout = schedule_timeout_interruptible(timeout);
1889 return jiffies_to_msecs(timeout);
1890 }
1891
1892 EXPORT_SYMBOL(msleep_interruptible);
1893
1894 /**
1895 * usleep_range - Sleep for an approximate time
1896 * @min: Minimum time in usecs to sleep
1897 * @max: Maximum time in usecs to sleep
1898 *
1899 * In non-atomic context where the exact wakeup time is flexible, use
1900 * usleep_range() instead of udelay(). The sleep improves responsiveness
1901 * by avoiding the CPU-hogging busy-wait of udelay(), and the range reduces
1902 * power usage by allowing hrtimers to take advantage of an already-
1903 * scheduled interrupt instead of scheduling a new one just for this sleep.
1904 */
1905 void __sched usleep_range(unsigned long min, unsigned long max)
1906 {
1907 ktime_t exp = ktime_add_us(ktime_get(), min);
1908 u64 delta = (u64)(max - min) * NSEC_PER_USEC;
1909
1910 for (;;) {
1911 __set_current_state(TASK_UNINTERRUPTIBLE);
1912 /* Do not return before the requested sleep time has elapsed */
1913 if (!schedule_hrtimeout_range(&exp, delta, HRTIMER_MODE_ABS))
1914 break;
1915 }
1916 }
1917 EXPORT_SYMBOL(usleep_range);