Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/kernel/timer.c | |
3 | * | |
4a22f166 | 4 | * Kernel internal timers |
1da177e4 LT |
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> | |
9984de1a | 23 | #include <linux/export.h> |
1da177e4 LT |
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> | |
b488893a | 29 | #include <linux/pid_namespace.h> |
1da177e4 LT |
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> | |
97a41e26 | 37 | #include <linux/delay.h> |
79bf2bb3 | 38 | #include <linux/tick.h> |
82f67cd9 | 39 | #include <linux/kallsyms.h> |
e360adbe | 40 | #include <linux/irq_work.h> |
eea08f32 | 41 | #include <linux/sched.h> |
cf4aebc2 | 42 | #include <linux/sched/sysctl.h> |
5a0e3ad6 | 43 | #include <linux/slab.h> |
1a0df594 | 44 | #include <linux/compat.h> |
1da177e4 LT |
45 | |
46 | #include <asm/uaccess.h> | |
47 | #include <asm/unistd.h> | |
48 | #include <asm/div64.h> | |
49 | #include <asm/timex.h> | |
50 | #include <asm/io.h> | |
51 | ||
c1ad348b TG |
52 | #include "tick-internal.h" |
53 | ||
2b022e3d XG |
54 | #define CREATE_TRACE_POINTS |
55 | #include <trace/events/timer.h> | |
56 | ||
40747ffa | 57 | __visible u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES; |
ecea8d19 TG |
58 | |
59 | EXPORT_SYMBOL(jiffies_64); | |
60 | ||
1da177e4 | 61 | /* |
500462a9 TG |
62 | * The timer wheel has LVL_DEPTH array levels. Each level provides an array of |
63 | * LVL_SIZE buckets. Each level is driven by its own clock and therefor each | |
64 | * level has a different granularity. | |
65 | * | |
66 | * The level granularity is: LVL_CLK_DIV ^ lvl | |
67 | * The level clock frequency is: HZ / (LVL_CLK_DIV ^ level) | |
68 | * | |
69 | * The array level of a newly armed timer depends on the relative expiry | |
70 | * time. The farther the expiry time is away the higher the array level and | |
71 | * therefor the granularity becomes. | |
72 | * | |
73 | * Contrary to the original timer wheel implementation, which aims for 'exact' | |
74 | * expiry of the timers, this implementation removes the need for recascading | |
75 | * the timers into the lower array levels. The previous 'classic' timer wheel | |
76 | * implementation of the kernel already violated the 'exact' expiry by adding | |
77 | * slack to the expiry time to provide batched expiration. The granularity | |
78 | * levels provide implicit batching. | |
79 | * | |
80 | * This is an optimization of the original timer wheel implementation for the | |
81 | * majority of the timer wheel use cases: timeouts. The vast majority of | |
82 | * timeout timers (networking, disk I/O ...) are canceled before expiry. If | |
83 | * the timeout expires it indicates that normal operation is disturbed, so it | |
84 | * does not matter much whether the timeout comes with a slight delay. | |
85 | * | |
86 | * The only exception to this are networking timers with a small expiry | |
87 | * time. They rely on the granularity. Those fit into the first wheel level, | |
88 | * which has HZ granularity. | |
89 | * | |
90 | * We don't have cascading anymore. timers with a expiry time above the | |
91 | * capacity of the last wheel level are force expired at the maximum timeout | |
92 | * value of the last wheel level. From data sampling we know that the maximum | |
93 | * value observed is 5 days (network connection tracking), so this should not | |
94 | * be an issue. | |
95 | * | |
96 | * The currently chosen array constants values are a good compromise between | |
97 | * array size and granularity. | |
98 | * | |
99 | * This results in the following granularity and range levels: | |
100 | * | |
101 | * HZ 1000 steps | |
102 | * Level Offset Granularity Range | |
103 | * 0 0 1 ms 0 ms - 63 ms | |
104 | * 1 64 8 ms 64 ms - 511 ms | |
105 | * 2 128 64 ms 512 ms - 4095 ms (512ms - ~4s) | |
106 | * 3 192 512 ms 4096 ms - 32767 ms (~4s - ~32s) | |
107 | * 4 256 4096 ms (~4s) 32768 ms - 262143 ms (~32s - ~4m) | |
108 | * 5 320 32768 ms (~32s) 262144 ms - 2097151 ms (~4m - ~34m) | |
109 | * 6 384 262144 ms (~4m) 2097152 ms - 16777215 ms (~34m - ~4h) | |
110 | * 7 448 2097152 ms (~34m) 16777216 ms - 134217727 ms (~4h - ~1d) | |
111 | * 8 512 16777216 ms (~4h) 134217728 ms - 1073741822 ms (~1d - ~12d) | |
112 | * | |
113 | * HZ 300 | |
114 | * Level Offset Granularity Range | |
115 | * 0 0 3 ms 0 ms - 210 ms | |
116 | * 1 64 26 ms 213 ms - 1703 ms (213ms - ~1s) | |
117 | * 2 128 213 ms 1706 ms - 13650 ms (~1s - ~13s) | |
118 | * 3 192 1706 ms (~1s) 13653 ms - 109223 ms (~13s - ~1m) | |
119 | * 4 256 13653 ms (~13s) 109226 ms - 873810 ms (~1m - ~14m) | |
120 | * 5 320 109226 ms (~1m) 873813 ms - 6990503 ms (~14m - ~1h) | |
121 | * 6 384 873813 ms (~14m) 6990506 ms - 55924050 ms (~1h - ~15h) | |
122 | * 7 448 6990506 ms (~1h) 55924053 ms - 447392423 ms (~15h - ~5d) | |
123 | * 8 512 55924053 ms (~15h) 447392426 ms - 3579139406 ms (~5d - ~41d) | |
124 | * | |
125 | * HZ 250 | |
126 | * Level Offset Granularity Range | |
127 | * 0 0 4 ms 0 ms - 255 ms | |
128 | * 1 64 32 ms 256 ms - 2047 ms (256ms - ~2s) | |
129 | * 2 128 256 ms 2048 ms - 16383 ms (~2s - ~16s) | |
130 | * 3 192 2048 ms (~2s) 16384 ms - 131071 ms (~16s - ~2m) | |
131 | * 4 256 16384 ms (~16s) 131072 ms - 1048575 ms (~2m - ~17m) | |
132 | * 5 320 131072 ms (~2m) 1048576 ms - 8388607 ms (~17m - ~2h) | |
133 | * 6 384 1048576 ms (~17m) 8388608 ms - 67108863 ms (~2h - ~18h) | |
134 | * 7 448 8388608 ms (~2h) 67108864 ms - 536870911 ms (~18h - ~6d) | |
135 | * 8 512 67108864 ms (~18h) 536870912 ms - 4294967288 ms (~6d - ~49d) | |
136 | * | |
137 | * HZ 100 | |
138 | * Level Offset Granularity Range | |
139 | * 0 0 10 ms 0 ms - 630 ms | |
140 | * 1 64 80 ms 640 ms - 5110 ms (640ms - ~5s) | |
141 | * 2 128 640 ms 5120 ms - 40950 ms (~5s - ~40s) | |
142 | * 3 192 5120 ms (~5s) 40960 ms - 327670 ms (~40s - ~5m) | |
143 | * 4 256 40960 ms (~40s) 327680 ms - 2621430 ms (~5m - ~43m) | |
144 | * 5 320 327680 ms (~5m) 2621440 ms - 20971510 ms (~43m - ~5h) | |
145 | * 6 384 2621440 ms (~43m) 20971520 ms - 167772150 ms (~5h - ~1d) | |
146 | * 7 448 20971520 ms (~5h) 167772160 ms - 1342177270 ms (~1d - ~15d) | |
1da177e4 | 147 | */ |
1da177e4 | 148 | |
500462a9 TG |
149 | /* Clock divisor for the next level */ |
150 | #define LVL_CLK_SHIFT 3 | |
151 | #define LVL_CLK_DIV (1UL << LVL_CLK_SHIFT) | |
152 | #define LVL_CLK_MASK (LVL_CLK_DIV - 1) | |
153 | #define LVL_SHIFT(n) ((n) * LVL_CLK_SHIFT) | |
154 | #define LVL_GRAN(n) (1UL << LVL_SHIFT(n)) | |
155 | ||
156 | /* | |
157 | * The time start value for each level to select the bucket at enqueue | |
158 | * time. | |
159 | */ | |
160 | #define LVL_START(n) ((LVL_SIZE - 1) << (((n) - 1) * LVL_CLK_SHIFT)) | |
161 | ||
162 | /* Size of each clock level */ | |
163 | #define LVL_BITS 6 | |
164 | #define LVL_SIZE (1UL << LVL_BITS) | |
165 | #define LVL_MASK (LVL_SIZE - 1) | |
166 | #define LVL_OFFS(n) ((n) * LVL_SIZE) | |
167 | ||
168 | /* Level depth */ | |
169 | #if HZ > 100 | |
170 | # define LVL_DEPTH 9 | |
171 | # else | |
172 | # define LVL_DEPTH 8 | |
173 | #endif | |
174 | ||
175 | /* The cutoff (max. capacity of the wheel) */ | |
176 | #define WHEEL_TIMEOUT_CUTOFF (LVL_START(LVL_DEPTH)) | |
177 | #define WHEEL_TIMEOUT_MAX (WHEEL_TIMEOUT_CUTOFF - LVL_GRAN(LVL_DEPTH - 1)) | |
178 | ||
179 | /* | |
180 | * The resulting wheel size. If NOHZ is configured we allocate two | |
181 | * wheels so we have a separate storage for the deferrable timers. | |
182 | */ | |
183 | #define WHEEL_SIZE (LVL_SIZE * LVL_DEPTH) | |
184 | ||
185 | #ifdef CONFIG_NO_HZ_COMMON | |
186 | # define NR_BASES 2 | |
187 | # define BASE_STD 0 | |
188 | # define BASE_DEF 1 | |
189 | #else | |
190 | # define NR_BASES 1 | |
191 | # define BASE_STD 0 | |
192 | # define BASE_DEF 0 | |
193 | #endif | |
1da177e4 | 194 | |
494af3ed | 195 | struct timer_base { |
500462a9 TG |
196 | spinlock_t lock; |
197 | struct timer_list *running_timer; | |
198 | unsigned long clk; | |
199 | unsigned int cpu; | |
200 | bool migration_enabled; | |
201 | bool nohz_active; | |
202 | DECLARE_BITMAP(pending_map, WHEEL_SIZE); | |
203 | struct hlist_head vectors[WHEEL_SIZE]; | |
6e453a67 | 204 | } ____cacheline_aligned; |
1da177e4 | 205 | |
500462a9 | 206 | static DEFINE_PER_CPU(struct timer_base, timer_bases[NR_BASES]); |
6e453a67 | 207 | |
bc7a34b8 TG |
208 | #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON) |
209 | unsigned int sysctl_timer_migration = 1; | |
210 | ||
683be13a | 211 | void timers_update_migration(bool update_nohz) |
bc7a34b8 TG |
212 | { |
213 | bool on = sysctl_timer_migration && tick_nohz_active; | |
214 | unsigned int cpu; | |
215 | ||
216 | /* Avoid the loop, if nothing to update */ | |
500462a9 | 217 | if (this_cpu_read(timer_bases[BASE_STD].migration_enabled) == on) |
bc7a34b8 TG |
218 | return; |
219 | ||
220 | for_each_possible_cpu(cpu) { | |
500462a9 TG |
221 | per_cpu(timer_bases[BASE_STD].migration_enabled, cpu) = on; |
222 | per_cpu(timer_bases[BASE_DEF].migration_enabled, cpu) = on; | |
bc7a34b8 | 223 | per_cpu(hrtimer_bases.migration_enabled, cpu) = on; |
683be13a TG |
224 | if (!update_nohz) |
225 | continue; | |
500462a9 TG |
226 | per_cpu(timer_bases[BASE_STD].nohz_active, cpu) = true; |
227 | per_cpu(timer_bases[BASE_DEF].nohz_active, cpu) = true; | |
683be13a | 228 | per_cpu(hrtimer_bases.nohz_active, cpu) = true; |
bc7a34b8 TG |
229 | } |
230 | } | |
231 | ||
232 | int timer_migration_handler(struct ctl_table *table, int write, | |
233 | void __user *buffer, size_t *lenp, | |
234 | loff_t *ppos) | |
235 | { | |
236 | static DEFINE_MUTEX(mutex); | |
237 | int ret; | |
238 | ||
239 | mutex_lock(&mutex); | |
240 | ret = proc_dointvec(table, write, buffer, lenp, ppos); | |
241 | if (!ret && write) | |
683be13a | 242 | timers_update_migration(false); |
bc7a34b8 TG |
243 | mutex_unlock(&mutex); |
244 | return ret; | |
245 | } | |
bc7a34b8 TG |
246 | #endif |
247 | ||
9c133c46 AS |
248 | static unsigned long round_jiffies_common(unsigned long j, int cpu, |
249 | bool force_up) | |
4c36a5de AV |
250 | { |
251 | int rem; | |
252 | unsigned long original = j; | |
253 | ||
254 | /* | |
255 | * We don't want all cpus firing their timers at once hitting the | |
256 | * same lock or cachelines, so we skew each extra cpu with an extra | |
257 | * 3 jiffies. This 3 jiffies came originally from the mm/ code which | |
258 | * already did this. | |
259 | * The skew is done by adding 3*cpunr, then round, then subtract this | |
260 | * extra offset again. | |
261 | */ | |
262 | j += cpu * 3; | |
263 | ||
264 | rem = j % HZ; | |
265 | ||
266 | /* | |
267 | * If the target jiffie is just after a whole second (which can happen | |
268 | * due to delays of the timer irq, long irq off times etc etc) then | |
269 | * we should round down to the whole second, not up. Use 1/4th second | |
270 | * as cutoff for this rounding as an extreme upper bound for this. | |
9c133c46 | 271 | * But never round down if @force_up is set. |
4c36a5de | 272 | */ |
9c133c46 | 273 | if (rem < HZ/4 && !force_up) /* round down */ |
4c36a5de AV |
274 | j = j - rem; |
275 | else /* round up */ | |
276 | j = j - rem + HZ; | |
277 | ||
278 | /* now that we have rounded, subtract the extra skew again */ | |
279 | j -= cpu * 3; | |
280 | ||
9e04d380 BVA |
281 | /* |
282 | * Make sure j is still in the future. Otherwise return the | |
283 | * unmodified value. | |
284 | */ | |
285 | return time_is_after_jiffies(j) ? j : original; | |
4c36a5de | 286 | } |
9c133c46 AS |
287 | |
288 | /** | |
289 | * __round_jiffies - function to round jiffies to a full second | |
290 | * @j: the time in (absolute) jiffies that should be rounded | |
291 | * @cpu: the processor number on which the timeout will happen | |
292 | * | |
293 | * __round_jiffies() rounds an absolute time in the future (in jiffies) | |
294 | * up or down to (approximately) full seconds. This is useful for timers | |
295 | * for which the exact time they fire does not matter too much, as long as | |
296 | * they fire approximately every X seconds. | |
297 | * | |
298 | * By rounding these timers to whole seconds, all such timers will fire | |
299 | * at the same time, rather than at various times spread out. The goal | |
300 | * of this is to have the CPU wake up less, which saves power. | |
301 | * | |
302 | * The exact rounding is skewed for each processor to avoid all | |
303 | * processors firing at the exact same time, which could lead | |
304 | * to lock contention or spurious cache line bouncing. | |
305 | * | |
306 | * The return value is the rounded version of the @j parameter. | |
307 | */ | |
308 | unsigned long __round_jiffies(unsigned long j, int cpu) | |
309 | { | |
310 | return round_jiffies_common(j, cpu, false); | |
311 | } | |
4c36a5de AV |
312 | EXPORT_SYMBOL_GPL(__round_jiffies); |
313 | ||
314 | /** | |
315 | * __round_jiffies_relative - function to round jiffies to a full second | |
316 | * @j: the time in (relative) jiffies that should be rounded | |
317 | * @cpu: the processor number on which the timeout will happen | |
318 | * | |
72fd4a35 | 319 | * __round_jiffies_relative() rounds a time delta in the future (in jiffies) |
4c36a5de AV |
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 | * | |
72fd4a35 | 332 | * The return value is the rounded version of the @j parameter. |
4c36a5de AV |
333 | */ |
334 | unsigned long __round_jiffies_relative(unsigned long j, int cpu) | |
335 | { | |
9c133c46 AS |
336 | unsigned long j0 = jiffies; |
337 | ||
338 | /* Use j0 because jiffies might change while we run */ | |
339 | return round_jiffies_common(j + j0, cpu, false) - j0; | |
4c36a5de AV |
340 | } |
341 | EXPORT_SYMBOL_GPL(__round_jiffies_relative); | |
342 | ||
343 | /** | |
344 | * round_jiffies - function to round jiffies to a full second | |
345 | * @j: the time in (absolute) jiffies that should be rounded | |
346 | * | |
72fd4a35 | 347 | * round_jiffies() rounds an absolute time in the future (in jiffies) |
4c36a5de AV |
348 | * up or down to (approximately) full seconds. This is useful for timers |
349 | * for which the exact time they fire does not matter too much, as long as | |
350 | * they fire approximately every X seconds. | |
351 | * | |
352 | * By rounding these timers to whole seconds, all such timers will fire | |
353 | * at the same time, rather than at various times spread out. The goal | |
354 | * of this is to have the CPU wake up less, which saves power. | |
355 | * | |
72fd4a35 | 356 | * The return value is the rounded version of the @j parameter. |
4c36a5de AV |
357 | */ |
358 | unsigned long round_jiffies(unsigned long j) | |
359 | { | |
9c133c46 | 360 | return round_jiffies_common(j, raw_smp_processor_id(), false); |
4c36a5de AV |
361 | } |
362 | EXPORT_SYMBOL_GPL(round_jiffies); | |
363 | ||
364 | /** | |
365 | * round_jiffies_relative - function to round jiffies to a full second | |
366 | * @j: the time in (relative) jiffies that should be rounded | |
367 | * | |
72fd4a35 | 368 | * round_jiffies_relative() rounds a time delta in the future (in jiffies) |
4c36a5de AV |
369 | * up or down to (approximately) full seconds. This is useful for timers |
370 | * for which the exact time they fire does not matter too much, as long as | |
371 | * they fire approximately every X seconds. | |
372 | * | |
373 | * By rounding these timers to whole seconds, all such timers will fire | |
374 | * at the same time, rather than at various times spread out. The goal | |
375 | * of this is to have the CPU wake up less, which saves power. | |
376 | * | |
72fd4a35 | 377 | * The return value is the rounded version of the @j parameter. |
4c36a5de AV |
378 | */ |
379 | unsigned long round_jiffies_relative(unsigned long j) | |
380 | { | |
381 | return __round_jiffies_relative(j, raw_smp_processor_id()); | |
382 | } | |
383 | EXPORT_SYMBOL_GPL(round_jiffies_relative); | |
384 | ||
9c133c46 AS |
385 | /** |
386 | * __round_jiffies_up - function to round jiffies up to a full second | |
387 | * @j: the time in (absolute) jiffies that should be rounded | |
388 | * @cpu: the processor number on which the timeout will happen | |
389 | * | |
390 | * This is the same as __round_jiffies() except that it will never | |
391 | * round down. This is useful for timeouts for which the exact time | |
392 | * of firing does not matter too much, as long as they don't fire too | |
393 | * early. | |
394 | */ | |
395 | unsigned long __round_jiffies_up(unsigned long j, int cpu) | |
396 | { | |
397 | return round_jiffies_common(j, cpu, true); | |
398 | } | |
399 | EXPORT_SYMBOL_GPL(__round_jiffies_up); | |
400 | ||
401 | /** | |
402 | * __round_jiffies_up_relative - function to round jiffies up to a full second | |
403 | * @j: the time in (relative) jiffies that should be rounded | |
404 | * @cpu: the processor number on which the timeout will happen | |
405 | * | |
406 | * This is the same as __round_jiffies_relative() except that it will never | |
407 | * round down. This is useful for timeouts for which the exact time | |
408 | * of firing does not matter too much, as long as they don't fire too | |
409 | * early. | |
410 | */ | |
411 | unsigned long __round_jiffies_up_relative(unsigned long j, int cpu) | |
412 | { | |
413 | unsigned long j0 = jiffies; | |
414 | ||
415 | /* Use j0 because jiffies might change while we run */ | |
416 | return round_jiffies_common(j + j0, cpu, true) - j0; | |
417 | } | |
418 | EXPORT_SYMBOL_GPL(__round_jiffies_up_relative); | |
419 | ||
420 | /** | |
421 | * round_jiffies_up - function to round jiffies up to a full second | |
422 | * @j: the time in (absolute) jiffies that should be rounded | |
423 | * | |
424 | * This is the same as round_jiffies() except that it will never | |
425 | * round down. This is useful for timeouts for which the exact time | |
426 | * of firing does not matter too much, as long as they don't fire too | |
427 | * early. | |
428 | */ | |
429 | unsigned long round_jiffies_up(unsigned long j) | |
430 | { | |
431 | return round_jiffies_common(j, raw_smp_processor_id(), true); | |
432 | } | |
433 | EXPORT_SYMBOL_GPL(round_jiffies_up); | |
434 | ||
435 | /** | |
436 | * round_jiffies_up_relative - function to round jiffies up to a full second | |
437 | * @j: the time in (relative) jiffies that should be rounded | |
438 | * | |
439 | * This is the same as round_jiffies_relative() except that it will never | |
440 | * round down. This is useful for timeouts for which the exact time | |
441 | * of firing does not matter too much, as long as they don't fire too | |
442 | * early. | |
443 | */ | |
444 | unsigned long round_jiffies_up_relative(unsigned long j) | |
445 | { | |
446 | return __round_jiffies_up_relative(j, raw_smp_processor_id()); | |
447 | } | |
448 | EXPORT_SYMBOL_GPL(round_jiffies_up_relative); | |
449 | ||
3bbb9ec9 | 450 | |
500462a9 TG |
451 | static inline unsigned int timer_get_idx(struct timer_list *timer) |
452 | { | |
453 | return (timer->flags & TIMER_ARRAYMASK) >> TIMER_ARRAYSHIFT; | |
454 | } | |
455 | ||
456 | static inline void timer_set_idx(struct timer_list *timer, unsigned int idx) | |
457 | { | |
458 | timer->flags = (timer->flags & ~TIMER_ARRAYMASK) | | |
459 | idx << TIMER_ARRAYSHIFT; | |
460 | } | |
461 | ||
462 | /* | |
463 | * Helper function to calculate the array index for a given expiry | |
464 | * time. | |
465 | */ | |
466 | static inline unsigned calc_index(unsigned expires, unsigned lvl) | |
467 | { | |
468 | expires = (expires + LVL_GRAN(lvl)) >> LVL_SHIFT(lvl); | |
469 | return LVL_OFFS(lvl) + (expires & LVL_MASK); | |
470 | } | |
471 | ||
facbb4a7 | 472 | static void |
494af3ed | 473 | __internal_add_timer(struct timer_base *base, struct timer_list *timer) |
1da177e4 LT |
474 | { |
475 | unsigned long expires = timer->expires; | |
500462a9 | 476 | unsigned long delta = expires - base->clk; |
1dabbcec | 477 | struct hlist_head *vec; |
500462a9 TG |
478 | unsigned int idx; |
479 | ||
480 | if (delta < LVL_START(1)) { | |
481 | idx = calc_index(expires, 0); | |
482 | } else if (delta < LVL_START(2)) { | |
483 | idx = calc_index(expires, 1); | |
484 | } else if (delta < LVL_START(3)) { | |
485 | idx = calc_index(expires, 2); | |
486 | } else if (delta < LVL_START(4)) { | |
487 | idx = calc_index(expires, 3); | |
488 | } else if (delta < LVL_START(5)) { | |
489 | idx = calc_index(expires, 4); | |
490 | } else if (delta < LVL_START(6)) { | |
491 | idx = calc_index(expires, 5); | |
492 | } else if (delta < LVL_START(7)) { | |
493 | idx = calc_index(expires, 6); | |
494 | } else if (LVL_DEPTH > 8 && delta < LVL_START(8)) { | |
495 | idx = calc_index(expires, 7); | |
496 | } else if ((long) delta < 0) { | |
497 | idx = base->clk & LVL_MASK; | |
1da177e4 | 498 | } else { |
500462a9 TG |
499 | /* |
500 | * Force expire obscene large timeouts to expire at the | |
501 | * capacity limit of the wheel. | |
1da177e4 | 502 | */ |
500462a9 TG |
503 | if (expires >= WHEEL_TIMEOUT_CUTOFF) |
504 | expires = WHEEL_TIMEOUT_MAX; | |
1bd04bf6 | 505 | |
500462a9 TG |
506 | idx = calc_index(expires, LVL_DEPTH - 1); |
507 | } | |
508 | /* | |
509 | * Enqueue the timer into the array bucket, mark it pending in | |
510 | * the bitmap and store the index in the timer flags. | |
511 | */ | |
512 | vec = base->vectors + idx; | |
1dabbcec | 513 | hlist_add_head(&timer->entry, vec); |
500462a9 TG |
514 | __set_bit(idx, base->pending_map); |
515 | timer_set_idx(timer, idx); | |
1da177e4 LT |
516 | } |
517 | ||
494af3ed | 518 | static void internal_add_timer(struct timer_base *base, struct timer_list *timer) |
facbb4a7 TG |
519 | { |
520 | __internal_add_timer(base, timer); | |
9f6d9baa VK |
521 | |
522 | /* | |
523 | * Check whether the other CPU is in dynticks mode and needs | |
500462a9 TG |
524 | * to be triggered to reevaluate the timer wheel. We are |
525 | * protected against the other CPU fiddling with the timer by | |
526 | * holding the timer base lock. This also makes sure that a | |
527 | * CPU on the way to stop its tick can not evaluate the timer | |
528 | * wheel. | |
9f6d9baa VK |
529 | * |
530 | * Spare the IPI for deferrable timers on idle targets though. | |
531 | * The next busy ticks will take care of it. Except full dynticks | |
532 | * require special care against races with idle_cpu(), lets deal | |
533 | * with that later. | |
534 | */ | |
500462a9 | 535 | if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && base->nohz_active) { |
683be13a TG |
536 | if (!(timer->flags & TIMER_DEFERRABLE) || |
537 | tick_nohz_full_cpu(base->cpu)) | |
538 | wake_up_nohz_cpu(base->cpu); | |
539 | } | |
facbb4a7 TG |
540 | } |
541 | ||
82f67cd9 IM |
542 | #ifdef CONFIG_TIMER_STATS |
543 | void __timer_stats_timer_set_start_info(struct timer_list *timer, void *addr) | |
544 | { | |
545 | if (timer->start_site) | |
546 | return; | |
547 | ||
548 | timer->start_site = addr; | |
549 | memcpy(timer->start_comm, current->comm, TASK_COMM_LEN); | |
550 | timer->start_pid = current->pid; | |
551 | } | |
c5c061b8 VP |
552 | |
553 | static void timer_stats_account_timer(struct timer_list *timer) | |
554 | { | |
3ed769bd DV |
555 | void *site; |
556 | ||
557 | /* | |
558 | * start_site can be concurrently reset by | |
559 | * timer_stats_timer_clear_start_info() | |
560 | */ | |
561 | site = READ_ONCE(timer->start_site); | |
562 | if (likely(!site)) | |
507e1231 | 563 | return; |
c5c061b8 | 564 | |
3ed769bd | 565 | timer_stats_update_stats(timer, timer->start_pid, site, |
c74441a1 TG |
566 | timer->function, timer->start_comm, |
567 | timer->flags); | |
c5c061b8 VP |
568 | } |
569 | ||
570 | #else | |
571 | static void timer_stats_account_timer(struct timer_list *timer) {} | |
82f67cd9 IM |
572 | #endif |
573 | ||
c6f3a97f TG |
574 | #ifdef CONFIG_DEBUG_OBJECTS_TIMERS |
575 | ||
576 | static struct debug_obj_descr timer_debug_descr; | |
577 | ||
99777288 SG |
578 | static void *timer_debug_hint(void *addr) |
579 | { | |
580 | return ((struct timer_list *) addr)->function; | |
581 | } | |
582 | ||
b9fdac7f DC |
583 | static bool timer_is_static_object(void *addr) |
584 | { | |
585 | struct timer_list *timer = addr; | |
586 | ||
587 | return (timer->entry.pprev == NULL && | |
588 | timer->entry.next == TIMER_ENTRY_STATIC); | |
589 | } | |
590 | ||
c6f3a97f TG |
591 | /* |
592 | * fixup_init is called when: | |
593 | * - an active object is initialized | |
55c888d6 | 594 | */ |
e3252464 | 595 | static bool timer_fixup_init(void *addr, enum debug_obj_state state) |
c6f3a97f TG |
596 | { |
597 | struct timer_list *timer = addr; | |
598 | ||
599 | switch (state) { | |
600 | case ODEBUG_STATE_ACTIVE: | |
601 | del_timer_sync(timer); | |
602 | debug_object_init(timer, &timer_debug_descr); | |
e3252464 | 603 | return true; |
c6f3a97f | 604 | default: |
e3252464 | 605 | return false; |
c6f3a97f TG |
606 | } |
607 | } | |
608 | ||
fb16b8cf SB |
609 | /* Stub timer callback for improperly used timers. */ |
610 | static void stub_timer(unsigned long data) | |
611 | { | |
612 | WARN_ON(1); | |
613 | } | |
614 | ||
c6f3a97f TG |
615 | /* |
616 | * fixup_activate is called when: | |
617 | * - an active object is activated | |
b9fdac7f | 618 | * - an unknown non-static object is activated |
c6f3a97f | 619 | */ |
e3252464 | 620 | static bool timer_fixup_activate(void *addr, enum debug_obj_state state) |
c6f3a97f TG |
621 | { |
622 | struct timer_list *timer = addr; | |
623 | ||
624 | switch (state) { | |
c6f3a97f | 625 | case ODEBUG_STATE_NOTAVAILABLE: |
b9fdac7f DC |
626 | setup_timer(timer, stub_timer, 0); |
627 | return true; | |
c6f3a97f TG |
628 | |
629 | case ODEBUG_STATE_ACTIVE: | |
630 | WARN_ON(1); | |
631 | ||
632 | default: | |
e3252464 | 633 | return false; |
c6f3a97f TG |
634 | } |
635 | } | |
636 | ||
637 | /* | |
638 | * fixup_free is called when: | |
639 | * - an active object is freed | |
640 | */ | |
e3252464 | 641 | static bool timer_fixup_free(void *addr, enum debug_obj_state state) |
c6f3a97f TG |
642 | { |
643 | struct timer_list *timer = addr; | |
644 | ||
645 | switch (state) { | |
646 | case ODEBUG_STATE_ACTIVE: | |
647 | del_timer_sync(timer); | |
648 | debug_object_free(timer, &timer_debug_descr); | |
e3252464 | 649 | return true; |
c6f3a97f | 650 | default: |
e3252464 | 651 | return false; |
c6f3a97f TG |
652 | } |
653 | } | |
654 | ||
dc4218bd CC |
655 | /* |
656 | * fixup_assert_init is called when: | |
657 | * - an untracked/uninit-ed object is found | |
658 | */ | |
e3252464 | 659 | static bool timer_fixup_assert_init(void *addr, enum debug_obj_state state) |
dc4218bd CC |
660 | { |
661 | struct timer_list *timer = addr; | |
662 | ||
663 | switch (state) { | |
664 | case ODEBUG_STATE_NOTAVAILABLE: | |
b9fdac7f DC |
665 | setup_timer(timer, stub_timer, 0); |
666 | return true; | |
dc4218bd | 667 | default: |
e3252464 | 668 | return false; |
dc4218bd CC |
669 | } |
670 | } | |
671 | ||
c6f3a97f | 672 | static struct debug_obj_descr timer_debug_descr = { |
dc4218bd CC |
673 | .name = "timer_list", |
674 | .debug_hint = timer_debug_hint, | |
b9fdac7f | 675 | .is_static_object = timer_is_static_object, |
dc4218bd CC |
676 | .fixup_init = timer_fixup_init, |
677 | .fixup_activate = timer_fixup_activate, | |
678 | .fixup_free = timer_fixup_free, | |
679 | .fixup_assert_init = timer_fixup_assert_init, | |
c6f3a97f TG |
680 | }; |
681 | ||
682 | static inline void debug_timer_init(struct timer_list *timer) | |
683 | { | |
684 | debug_object_init(timer, &timer_debug_descr); | |
685 | } | |
686 | ||
687 | static inline void debug_timer_activate(struct timer_list *timer) | |
688 | { | |
689 | debug_object_activate(timer, &timer_debug_descr); | |
690 | } | |
691 | ||
692 | static inline void debug_timer_deactivate(struct timer_list *timer) | |
693 | { | |
694 | debug_object_deactivate(timer, &timer_debug_descr); | |
695 | } | |
696 | ||
697 | static inline void debug_timer_free(struct timer_list *timer) | |
698 | { | |
699 | debug_object_free(timer, &timer_debug_descr); | |
700 | } | |
701 | ||
dc4218bd CC |
702 | static inline void debug_timer_assert_init(struct timer_list *timer) |
703 | { | |
704 | debug_object_assert_init(timer, &timer_debug_descr); | |
705 | } | |
706 | ||
fc683995 TH |
707 | static void do_init_timer(struct timer_list *timer, unsigned int flags, |
708 | const char *name, struct lock_class_key *key); | |
c6f3a97f | 709 | |
fc683995 TH |
710 | void init_timer_on_stack_key(struct timer_list *timer, unsigned int flags, |
711 | const char *name, struct lock_class_key *key) | |
c6f3a97f TG |
712 | { |
713 | debug_object_init_on_stack(timer, &timer_debug_descr); | |
fc683995 | 714 | do_init_timer(timer, flags, name, key); |
c6f3a97f | 715 | } |
6f2b9b9a | 716 | EXPORT_SYMBOL_GPL(init_timer_on_stack_key); |
c6f3a97f TG |
717 | |
718 | void destroy_timer_on_stack(struct timer_list *timer) | |
719 | { | |
720 | debug_object_free(timer, &timer_debug_descr); | |
721 | } | |
722 | EXPORT_SYMBOL_GPL(destroy_timer_on_stack); | |
723 | ||
724 | #else | |
725 | static inline void debug_timer_init(struct timer_list *timer) { } | |
726 | static inline void debug_timer_activate(struct timer_list *timer) { } | |
727 | static inline void debug_timer_deactivate(struct timer_list *timer) { } | |
dc4218bd | 728 | static inline void debug_timer_assert_init(struct timer_list *timer) { } |
c6f3a97f TG |
729 | #endif |
730 | ||
2b022e3d XG |
731 | static inline void debug_init(struct timer_list *timer) |
732 | { | |
733 | debug_timer_init(timer); | |
734 | trace_timer_init(timer); | |
735 | } | |
736 | ||
737 | static inline void | |
738 | debug_activate(struct timer_list *timer, unsigned long expires) | |
739 | { | |
740 | debug_timer_activate(timer); | |
0eeda71b | 741 | trace_timer_start(timer, expires, timer->flags); |
2b022e3d XG |
742 | } |
743 | ||
744 | static inline void debug_deactivate(struct timer_list *timer) | |
745 | { | |
746 | debug_timer_deactivate(timer); | |
747 | trace_timer_cancel(timer); | |
748 | } | |
749 | ||
dc4218bd CC |
750 | static inline void debug_assert_init(struct timer_list *timer) |
751 | { | |
752 | debug_timer_assert_init(timer); | |
753 | } | |
754 | ||
fc683995 TH |
755 | static void do_init_timer(struct timer_list *timer, unsigned int flags, |
756 | const char *name, struct lock_class_key *key) | |
55c888d6 | 757 | { |
1dabbcec | 758 | timer->entry.pprev = NULL; |
0eeda71b | 759 | timer->flags = flags | raw_smp_processor_id(); |
82f67cd9 IM |
760 | #ifdef CONFIG_TIMER_STATS |
761 | timer->start_site = NULL; | |
762 | timer->start_pid = -1; | |
763 | memset(timer->start_comm, 0, TASK_COMM_LEN); | |
764 | #endif | |
6f2b9b9a | 765 | lockdep_init_map(&timer->lockdep_map, name, key, 0); |
55c888d6 | 766 | } |
c6f3a97f TG |
767 | |
768 | /** | |
633fe795 | 769 | * init_timer_key - initialize a timer |
c6f3a97f | 770 | * @timer: the timer to be initialized |
fc683995 | 771 | * @flags: timer flags |
633fe795 RD |
772 | * @name: name of the timer |
773 | * @key: lockdep class key of the fake lock used for tracking timer | |
774 | * sync lock dependencies | |
c6f3a97f | 775 | * |
633fe795 | 776 | * init_timer_key() must be done to a timer prior calling *any* of the |
c6f3a97f TG |
777 | * other timer functions. |
778 | */ | |
fc683995 TH |
779 | void init_timer_key(struct timer_list *timer, unsigned int flags, |
780 | const char *name, struct lock_class_key *key) | |
c6f3a97f | 781 | { |
2b022e3d | 782 | debug_init(timer); |
fc683995 | 783 | do_init_timer(timer, flags, name, key); |
c6f3a97f | 784 | } |
6f2b9b9a | 785 | EXPORT_SYMBOL(init_timer_key); |
55c888d6 | 786 | |
ec44bc7a | 787 | static inline void detach_timer(struct timer_list *timer, bool clear_pending) |
55c888d6 | 788 | { |
1dabbcec | 789 | struct hlist_node *entry = &timer->entry; |
55c888d6 | 790 | |
2b022e3d | 791 | debug_deactivate(timer); |
c6f3a97f | 792 | |
1dabbcec | 793 | __hlist_del(entry); |
55c888d6 | 794 | if (clear_pending) |
1dabbcec TG |
795 | entry->pprev = NULL; |
796 | entry->next = LIST_POISON2; | |
55c888d6 ON |
797 | } |
798 | ||
494af3ed | 799 | static int detach_if_pending(struct timer_list *timer, struct timer_base *base, |
ec44bc7a TG |
800 | bool clear_pending) |
801 | { | |
500462a9 TG |
802 | unsigned idx = timer_get_idx(timer); |
803 | ||
ec44bc7a TG |
804 | if (!timer_pending(timer)) |
805 | return 0; | |
806 | ||
500462a9 TG |
807 | if (hlist_is_singular_node(&timer->entry, base->vectors + idx)) |
808 | __clear_bit(idx, base->pending_map); | |
809 | ||
ec44bc7a | 810 | detach_timer(timer, clear_pending); |
ec44bc7a TG |
811 | return 1; |
812 | } | |
813 | ||
500462a9 TG |
814 | static inline struct timer_base *get_timer_cpu_base(u32 tflags, u32 cpu) |
815 | { | |
816 | struct timer_base *base = per_cpu_ptr(&timer_bases[BASE_STD], cpu); | |
817 | ||
818 | /* | |
819 | * If the timer is deferrable and nohz is active then we need to use | |
820 | * the deferrable base. | |
821 | */ | |
822 | if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && base->nohz_active && | |
823 | (tflags & TIMER_DEFERRABLE)) | |
824 | base = per_cpu_ptr(&timer_bases[BASE_DEF], cpu); | |
825 | return base; | |
826 | } | |
827 | ||
828 | static inline struct timer_base *get_timer_this_cpu_base(u32 tflags) | |
829 | { | |
830 | struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]); | |
831 | ||
832 | /* | |
833 | * If the timer is deferrable and nohz is active then we need to use | |
834 | * the deferrable base. | |
835 | */ | |
836 | if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && base->nohz_active && | |
837 | (tflags & TIMER_DEFERRABLE)) | |
838 | base = this_cpu_ptr(&timer_bases[BASE_DEF]); | |
839 | return base; | |
840 | } | |
841 | ||
842 | static inline struct timer_base *get_timer_base(u32 tflags) | |
843 | { | |
844 | return get_timer_cpu_base(tflags, tflags & TIMER_CPUMASK); | |
845 | } | |
846 | ||
847 | static inline struct timer_base *get_target_base(struct timer_base *base, | |
848 | unsigned tflags) | |
849 | { | |
850 | #if defined(CONFIG_NO_HZ_COMMON) && defined(CONFIG_SMP) | |
851 | if ((tflags & TIMER_PINNED) || !base->migration_enabled) | |
852 | return get_timer_this_cpu_base(tflags); | |
853 | return get_timer_cpu_base(tflags, get_nohz_timer_target()); | |
854 | #else | |
855 | return get_timer_this_cpu_base(tflags); | |
856 | #endif | |
857 | } | |
858 | ||
55c888d6 | 859 | /* |
500462a9 TG |
860 | * We are using hashed locking: Holding per_cpu(timer_bases[x]).lock means |
861 | * that all timers which are tied to this base are locked, and the base itself | |
862 | * is locked too. | |
55c888d6 ON |
863 | * |
864 | * So __run_timers/migrate_timers can safely modify all timers which could | |
500462a9 | 865 | * be found in the base->vectors array. |
55c888d6 | 866 | * |
500462a9 TG |
867 | * When a timer is migrating then the TIMER_MIGRATING flag is set and we need |
868 | * to wait until the migration is done. | |
55c888d6 | 869 | */ |
494af3ed | 870 | static struct timer_base *lock_timer_base(struct timer_list *timer, |
500462a9 | 871 | unsigned long *flags) |
89e7e374 | 872 | __acquires(timer->base->lock) |
55c888d6 | 873 | { |
55c888d6 | 874 | for (;;) { |
494af3ed | 875 | struct timer_base *base; |
500462a9 | 876 | u32 tf = timer->flags; |
0eeda71b TG |
877 | |
878 | if (!(tf & TIMER_MIGRATING)) { | |
500462a9 | 879 | base = get_timer_base(tf); |
55c888d6 | 880 | spin_lock_irqsave(&base->lock, *flags); |
0eeda71b | 881 | if (timer->flags == tf) |
55c888d6 | 882 | return base; |
55c888d6 ON |
883 | spin_unlock_irqrestore(&base->lock, *flags); |
884 | } | |
885 | cpu_relax(); | |
886 | } | |
887 | } | |
888 | ||
74019224 | 889 | static inline int |
177ec0a0 | 890 | __mod_timer(struct timer_list *timer, unsigned long expires, bool pending_only) |
1da177e4 | 891 | { |
494af3ed | 892 | struct timer_base *base, *new_base; |
1da177e4 | 893 | unsigned long flags; |
bc7a34b8 | 894 | int ret = 0; |
1da177e4 | 895 | |
500462a9 TG |
896 | /* |
897 | * TODO: Calculate the array bucket of the timer right here w/o | |
898 | * holding the base lock. This allows to check not only | |
899 | * timer->expires == expires below, but also whether the timer | |
900 | * ends up in the same bucket. If we really need to requeue | |
901 | * the timer then we check whether base->clk have | |
902 | * advanced between here and locking the timer base. If | |
903 | * jiffies advanced we have to recalc the array bucket with the | |
904 | * lock held. | |
905 | */ | |
906 | ||
907 | /* | |
908 | * This is a common optimization triggered by the | |
909 | * networking code - if the timer is re-modified | |
910 | * to be the same thing then just return: | |
911 | */ | |
912 | if (timer_pending(timer)) { | |
913 | if (timer->expires == expires) | |
914 | return 1; | |
915 | } | |
916 | ||
82f67cd9 | 917 | timer_stats_timer_set_start_info(timer); |
1da177e4 | 918 | BUG_ON(!timer->function); |
1da177e4 | 919 | |
55c888d6 ON |
920 | base = lock_timer_base(timer, &flags); |
921 | ||
ec44bc7a TG |
922 | ret = detach_if_pending(timer, base, false); |
923 | if (!ret && pending_only) | |
924 | goto out_unlock; | |
55c888d6 | 925 | |
2b022e3d | 926 | debug_activate(timer, expires); |
c6f3a97f | 927 | |
500462a9 | 928 | new_base = get_target_base(base, timer->flags); |
eea08f32 | 929 | |
3691c519 | 930 | if (base != new_base) { |
1da177e4 | 931 | /* |
500462a9 | 932 | * We are trying to schedule the timer on the new base. |
55c888d6 ON |
933 | * However we can't change timer's base while it is running, |
934 | * otherwise del_timer_sync() can't detect that the timer's | |
500462a9 TG |
935 | * handler yet has not finished. This also guarantees that the |
936 | * timer is serialized wrt itself. | |
1da177e4 | 937 | */ |
a2c348fe | 938 | if (likely(base->running_timer != timer)) { |
55c888d6 | 939 | /* See the comment in lock_timer_base() */ |
0eeda71b TG |
940 | timer->flags |= TIMER_MIGRATING; |
941 | ||
55c888d6 | 942 | spin_unlock(&base->lock); |
a2c348fe ON |
943 | base = new_base; |
944 | spin_lock(&base->lock); | |
d0023a14 ED |
945 | WRITE_ONCE(timer->flags, |
946 | (timer->flags & ~TIMER_BASEMASK) | base->cpu); | |
1da177e4 LT |
947 | } |
948 | } | |
949 | ||
1da177e4 | 950 | timer->expires = expires; |
a2c348fe | 951 | internal_add_timer(base, timer); |
74019224 IM |
952 | |
953 | out_unlock: | |
a2c348fe | 954 | spin_unlock_irqrestore(&base->lock, flags); |
1da177e4 LT |
955 | |
956 | return ret; | |
957 | } | |
958 | ||
2aae4a10 | 959 | /** |
74019224 IM |
960 | * mod_timer_pending - modify a pending timer's timeout |
961 | * @timer: the pending timer to be modified | |
962 | * @expires: new timeout in jiffies | |
1da177e4 | 963 | * |
74019224 IM |
964 | * mod_timer_pending() is the same for pending timers as mod_timer(), |
965 | * but will not re-activate and modify already deleted timers. | |
966 | * | |
967 | * It is useful for unserialized use of timers. | |
1da177e4 | 968 | */ |
74019224 | 969 | int mod_timer_pending(struct timer_list *timer, unsigned long expires) |
1da177e4 | 970 | { |
177ec0a0 | 971 | return __mod_timer(timer, expires, true); |
1da177e4 | 972 | } |
74019224 | 973 | EXPORT_SYMBOL(mod_timer_pending); |
1da177e4 | 974 | |
2aae4a10 | 975 | /** |
1da177e4 LT |
976 | * mod_timer - modify a timer's timeout |
977 | * @timer: the timer to be modified | |
2aae4a10 | 978 | * @expires: new timeout in jiffies |
1da177e4 | 979 | * |
72fd4a35 | 980 | * mod_timer() is a more efficient way to update the expire field of an |
1da177e4 LT |
981 | * active timer (if the timer is inactive it will be activated) |
982 | * | |
983 | * mod_timer(timer, expires) is equivalent to: | |
984 | * | |
985 | * del_timer(timer); timer->expires = expires; add_timer(timer); | |
986 | * | |
987 | * Note that if there are multiple unserialized concurrent users of the | |
988 | * same timer, then mod_timer() is the only safe way to modify the timeout, | |
989 | * since add_timer() cannot modify an already running timer. | |
990 | * | |
991 | * The function returns whether it has modified a pending timer or not. | |
992 | * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an | |
993 | * active timer returns 1.) | |
994 | */ | |
995 | int mod_timer(struct timer_list *timer, unsigned long expires) | |
996 | { | |
177ec0a0 | 997 | return __mod_timer(timer, expires, false); |
1da177e4 | 998 | } |
1da177e4 LT |
999 | EXPORT_SYMBOL(mod_timer); |
1000 | ||
74019224 IM |
1001 | /** |
1002 | * add_timer - start a timer | |
1003 | * @timer: the timer to be added | |
1004 | * | |
1005 | * The kernel will do a ->function(->data) callback from the | |
1006 | * timer interrupt at the ->expires point in the future. The | |
1007 | * current time is 'jiffies'. | |
1008 | * | |
1009 | * The timer's ->expires, ->function (and if the handler uses it, ->data) | |
1010 | * fields must be set prior calling this function. | |
1011 | * | |
1012 | * Timers with an ->expires field in the past will be executed in the next | |
1013 | * timer tick. | |
1014 | */ | |
1015 | void add_timer(struct timer_list *timer) | |
1016 | { | |
1017 | BUG_ON(timer_pending(timer)); | |
1018 | mod_timer(timer, timer->expires); | |
1019 | } | |
1020 | EXPORT_SYMBOL(add_timer); | |
1021 | ||
1022 | /** | |
1023 | * add_timer_on - start a timer on a particular CPU | |
1024 | * @timer: the timer to be added | |
1025 | * @cpu: the CPU to start it on | |
1026 | * | |
1027 | * This is not very scalable on SMP. Double adds are not possible. | |
1028 | */ | |
1029 | void add_timer_on(struct timer_list *timer, int cpu) | |
1030 | { | |
500462a9 | 1031 | struct timer_base *new_base, *base; |
74019224 IM |
1032 | unsigned long flags; |
1033 | ||
1034 | timer_stats_timer_set_start_info(timer); | |
1035 | BUG_ON(timer_pending(timer) || !timer->function); | |
22b886dd | 1036 | |
500462a9 TG |
1037 | new_base = get_timer_cpu_base(timer->flags, cpu); |
1038 | ||
22b886dd TH |
1039 | /* |
1040 | * If @timer was on a different CPU, it should be migrated with the | |
1041 | * old base locked to prevent other operations proceeding with the | |
1042 | * wrong base locked. See lock_timer_base(). | |
1043 | */ | |
1044 | base = lock_timer_base(timer, &flags); | |
1045 | if (base != new_base) { | |
1046 | timer->flags |= TIMER_MIGRATING; | |
1047 | ||
1048 | spin_unlock(&base->lock); | |
1049 | base = new_base; | |
1050 | spin_lock(&base->lock); | |
1051 | WRITE_ONCE(timer->flags, | |
1052 | (timer->flags & ~TIMER_BASEMASK) | cpu); | |
1053 | } | |
1054 | ||
2b022e3d | 1055 | debug_activate(timer, timer->expires); |
74019224 | 1056 | internal_add_timer(base, timer); |
74019224 IM |
1057 | spin_unlock_irqrestore(&base->lock, flags); |
1058 | } | |
a9862e05 | 1059 | EXPORT_SYMBOL_GPL(add_timer_on); |
74019224 | 1060 | |
2aae4a10 | 1061 | /** |
1da177e4 LT |
1062 | * del_timer - deactive a timer. |
1063 | * @timer: the timer to be deactivated | |
1064 | * | |
1065 | * del_timer() deactivates a timer - this works on both active and inactive | |
1066 | * timers. | |
1067 | * | |
1068 | * The function returns whether it has deactivated a pending timer or not. | |
1069 | * (ie. del_timer() of an inactive timer returns 0, del_timer() of an | |
1070 | * active timer returns 1.) | |
1071 | */ | |
1072 | int del_timer(struct timer_list *timer) | |
1073 | { | |
494af3ed | 1074 | struct timer_base *base; |
1da177e4 | 1075 | unsigned long flags; |
55c888d6 | 1076 | int ret = 0; |
1da177e4 | 1077 | |
dc4218bd CC |
1078 | debug_assert_init(timer); |
1079 | ||
82f67cd9 | 1080 | timer_stats_timer_clear_start_info(timer); |
55c888d6 ON |
1081 | if (timer_pending(timer)) { |
1082 | base = lock_timer_base(timer, &flags); | |
ec44bc7a | 1083 | ret = detach_if_pending(timer, base, true); |
1da177e4 | 1084 | spin_unlock_irqrestore(&base->lock, flags); |
1da177e4 | 1085 | } |
1da177e4 | 1086 | |
55c888d6 | 1087 | return ret; |
1da177e4 | 1088 | } |
1da177e4 LT |
1089 | EXPORT_SYMBOL(del_timer); |
1090 | ||
2aae4a10 REB |
1091 | /** |
1092 | * try_to_del_timer_sync - Try to deactivate a timer | |
1093 | * @timer: timer do del | |
1094 | * | |
fd450b73 ON |
1095 | * This function tries to deactivate a timer. Upon successful (ret >= 0) |
1096 | * exit the timer is not queued and the handler is not running on any CPU. | |
fd450b73 ON |
1097 | */ |
1098 | int try_to_del_timer_sync(struct timer_list *timer) | |
1099 | { | |
494af3ed | 1100 | struct timer_base *base; |
fd450b73 ON |
1101 | unsigned long flags; |
1102 | int ret = -1; | |
1103 | ||
dc4218bd CC |
1104 | debug_assert_init(timer); |
1105 | ||
fd450b73 ON |
1106 | base = lock_timer_base(timer, &flags); |
1107 | ||
ec44bc7a TG |
1108 | if (base->running_timer != timer) { |
1109 | timer_stats_timer_clear_start_info(timer); | |
1110 | ret = detach_if_pending(timer, base, true); | |
fd450b73 | 1111 | } |
fd450b73 ON |
1112 | spin_unlock_irqrestore(&base->lock, flags); |
1113 | ||
1114 | return ret; | |
1115 | } | |
e19dff1f DH |
1116 | EXPORT_SYMBOL(try_to_del_timer_sync); |
1117 | ||
6f1bc451 | 1118 | #ifdef CONFIG_SMP |
2aae4a10 | 1119 | /** |
1da177e4 LT |
1120 | * del_timer_sync - deactivate a timer and wait for the handler to finish. |
1121 | * @timer: the timer to be deactivated | |
1122 | * | |
1123 | * This function only differs from del_timer() on SMP: besides deactivating | |
1124 | * the timer it also makes sure the handler has finished executing on other | |
1125 | * CPUs. | |
1126 | * | |
72fd4a35 | 1127 | * Synchronization rules: Callers must prevent restarting of the timer, |
1da177e4 | 1128 | * otherwise this function is meaningless. It must not be called from |
c5f66e99 TH |
1129 | * interrupt contexts unless the timer is an irqsafe one. The caller must |
1130 | * not hold locks which would prevent completion of the timer's | |
1131 | * handler. The timer's handler must not call add_timer_on(). Upon exit the | |
1132 | * timer is not queued and the handler is not running on any CPU. | |
1da177e4 | 1133 | * |
c5f66e99 TH |
1134 | * Note: For !irqsafe timers, you must not hold locks that are held in |
1135 | * interrupt context while calling this function. Even if the lock has | |
1136 | * nothing to do with the timer in question. Here's why: | |
48228f7b SR |
1137 | * |
1138 | * CPU0 CPU1 | |
1139 | * ---- ---- | |
1140 | * <SOFTIRQ> | |
1141 | * call_timer_fn(); | |
1142 | * base->running_timer = mytimer; | |
1143 | * spin_lock_irq(somelock); | |
1144 | * <IRQ> | |
1145 | * spin_lock(somelock); | |
1146 | * del_timer_sync(mytimer); | |
1147 | * while (base->running_timer == mytimer); | |
1148 | * | |
1149 | * Now del_timer_sync() will never return and never release somelock. | |
1150 | * The interrupt on the other CPU is waiting to grab somelock but | |
1151 | * it has interrupted the softirq that CPU0 is waiting to finish. | |
1152 | * | |
1da177e4 | 1153 | * The function returns whether it has deactivated a pending timer or not. |
1da177e4 LT |
1154 | */ |
1155 | int del_timer_sync(struct timer_list *timer) | |
1156 | { | |
6f2b9b9a | 1157 | #ifdef CONFIG_LOCKDEP |
f266a511 PZ |
1158 | unsigned long flags; |
1159 | ||
48228f7b SR |
1160 | /* |
1161 | * If lockdep gives a backtrace here, please reference | |
1162 | * the synchronization rules above. | |
1163 | */ | |
7ff20792 | 1164 | local_irq_save(flags); |
6f2b9b9a JB |
1165 | lock_map_acquire(&timer->lockdep_map); |
1166 | lock_map_release(&timer->lockdep_map); | |
7ff20792 | 1167 | local_irq_restore(flags); |
6f2b9b9a | 1168 | #endif |
466bd303 YZ |
1169 | /* |
1170 | * don't use it in hardirq context, because it | |
1171 | * could lead to deadlock. | |
1172 | */ | |
0eeda71b | 1173 | WARN_ON(in_irq() && !(timer->flags & TIMER_IRQSAFE)); |
fd450b73 ON |
1174 | for (;;) { |
1175 | int ret = try_to_del_timer_sync(timer); | |
1176 | if (ret >= 0) | |
1177 | return ret; | |
a0009652 | 1178 | cpu_relax(); |
fd450b73 | 1179 | } |
1da177e4 | 1180 | } |
55c888d6 | 1181 | EXPORT_SYMBOL(del_timer_sync); |
1da177e4 LT |
1182 | #endif |
1183 | ||
576da126 TG |
1184 | static void call_timer_fn(struct timer_list *timer, void (*fn)(unsigned long), |
1185 | unsigned long data) | |
1186 | { | |
4a2b4b22 | 1187 | int count = preempt_count(); |
576da126 TG |
1188 | |
1189 | #ifdef CONFIG_LOCKDEP | |
1190 | /* | |
1191 | * It is permissible to free the timer from inside the | |
1192 | * function that is called from it, this we need to take into | |
1193 | * account for lockdep too. To avoid bogus "held lock freed" | |
1194 | * warnings as well as problems when looking into | |
1195 | * timer->lockdep_map, make a copy and use that here. | |
1196 | */ | |
4d82a1de PZ |
1197 | struct lockdep_map lockdep_map; |
1198 | ||
1199 | lockdep_copy_map(&lockdep_map, &timer->lockdep_map); | |
576da126 TG |
1200 | #endif |
1201 | /* | |
1202 | * Couple the lock chain with the lock chain at | |
1203 | * del_timer_sync() by acquiring the lock_map around the fn() | |
1204 | * call here and in del_timer_sync(). | |
1205 | */ | |
1206 | lock_map_acquire(&lockdep_map); | |
1207 | ||
1208 | trace_timer_expire_entry(timer); | |
1209 | fn(data); | |
1210 | trace_timer_expire_exit(timer); | |
1211 | ||
1212 | lock_map_release(&lockdep_map); | |
1213 | ||
4a2b4b22 | 1214 | if (count != preempt_count()) { |
802702e0 | 1215 | WARN_ONCE(1, "timer: %pF preempt leak: %08x -> %08x\n", |
4a2b4b22 | 1216 | fn, count, preempt_count()); |
802702e0 TG |
1217 | /* |
1218 | * Restore the preempt count. That gives us a decent | |
1219 | * chance to survive and extract information. If the | |
1220 | * callback kept a lock held, bad luck, but not worse | |
1221 | * than the BUG() we had. | |
1222 | */ | |
4a2b4b22 | 1223 | preempt_count_set(count); |
576da126 TG |
1224 | } |
1225 | } | |
1226 | ||
500462a9 TG |
1227 | static void expire_timers(struct timer_base *base, struct hlist_head *head) |
1228 | { | |
1229 | while (!hlist_empty(head)) { | |
1230 | struct timer_list *timer; | |
1231 | void (*fn)(unsigned long); | |
1232 | unsigned long data; | |
1233 | ||
1234 | timer = hlist_entry(head->first, struct timer_list, entry); | |
1235 | timer_stats_account_timer(timer); | |
1236 | ||
1237 | base->running_timer = timer; | |
1238 | detach_timer(timer, true); | |
1239 | ||
1240 | fn = timer->function; | |
1241 | data = timer->data; | |
1242 | ||
1243 | if (timer->flags & TIMER_IRQSAFE) { | |
1244 | spin_unlock(&base->lock); | |
1245 | call_timer_fn(timer, fn, data); | |
1246 | spin_lock(&base->lock); | |
1247 | } else { | |
1248 | spin_unlock_irq(&base->lock); | |
1249 | call_timer_fn(timer, fn, data); | |
1250 | spin_lock_irq(&base->lock); | |
1251 | } | |
1252 | } | |
1253 | } | |
1254 | ||
1255 | static int collect_expired_timers(struct timer_base *base, | |
1256 | struct hlist_head *heads) | |
1257 | { | |
1258 | unsigned long clk = base->clk; | |
1259 | struct hlist_head *vec; | |
1260 | int i, levels = 0; | |
1261 | unsigned int idx; | |
1262 | ||
1263 | for (i = 0; i < LVL_DEPTH; i++) { | |
1264 | idx = (clk & LVL_MASK) + i * LVL_SIZE; | |
1265 | ||
1266 | if (__test_and_clear_bit(idx, base->pending_map)) { | |
1267 | vec = base->vectors + idx; | |
1268 | hlist_move_list(vec, heads++); | |
1269 | levels++; | |
1270 | } | |
1271 | /* Is it time to look at the next level? */ | |
1272 | if (clk & LVL_CLK_MASK) | |
1273 | break; | |
1274 | /* Shift clock for the next level granularity */ | |
1275 | clk >>= LVL_CLK_SHIFT; | |
1276 | } | |
1277 | return levels; | |
1278 | } | |
2aae4a10 | 1279 | |
3451d024 | 1280 | #ifdef CONFIG_NO_HZ_COMMON |
1da177e4 | 1281 | /* |
500462a9 TG |
1282 | * Find the next pending bucket of a level. Search from @offset + @clk upwards |
1283 | * and if nothing there, search from start of the level (@offset) up to | |
1284 | * @offset + clk. | |
1285 | */ | |
1286 | static int next_pending_bucket(struct timer_base *base, unsigned offset, | |
1287 | unsigned clk) | |
1288 | { | |
1289 | unsigned pos, start = offset + clk; | |
1290 | unsigned end = offset + LVL_SIZE; | |
1291 | ||
1292 | pos = find_next_bit(base->pending_map, end, start); | |
1293 | if (pos < end) | |
1294 | return pos - start; | |
1295 | ||
1296 | pos = find_next_bit(base->pending_map, start, offset); | |
1297 | return pos < start ? pos + LVL_SIZE - start : -1; | |
1298 | } | |
1299 | ||
1300 | /* | |
1301 | * Search the first expiring timer in the various clock levels. | |
1da177e4 | 1302 | */ |
494af3ed | 1303 | static unsigned long __next_timer_interrupt(struct timer_base *base) |
1da177e4 | 1304 | { |
500462a9 TG |
1305 | unsigned long clk, next, adj; |
1306 | unsigned lvl, offset = 0; | |
1307 | ||
1308 | spin_lock(&base->lock); | |
1309 | next = base->clk + NEXT_TIMER_MAX_DELTA; | |
1310 | clk = base->clk; | |
1311 | for (lvl = 0; lvl < LVL_DEPTH; lvl++, offset += LVL_SIZE) { | |
1312 | int pos = next_pending_bucket(base, offset, clk & LVL_MASK); | |
1313 | ||
1314 | if (pos >= 0) { | |
1315 | unsigned long tmp = clk + (unsigned long) pos; | |
1316 | ||
1317 | tmp <<= LVL_SHIFT(lvl); | |
1318 | if (time_before(tmp, next)) | |
1319 | next = tmp; | |
1da177e4 | 1320 | } |
500462a9 TG |
1321 | /* |
1322 | * Clock for the next level. If the current level clock lower | |
1323 | * bits are zero, we look at the next level as is. If not we | |
1324 | * need to advance it by one because that's going to be the | |
1325 | * next expiring bucket in that level. base->clk is the next | |
1326 | * expiring jiffie. So in case of: | |
1327 | * | |
1328 | * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0 | |
1329 | * 0 0 0 0 0 0 | |
1330 | * | |
1331 | * we have to look at all levels @index 0. With | |
1332 | * | |
1333 | * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0 | |
1334 | * 0 0 0 0 0 2 | |
1335 | * | |
1336 | * LVL0 has the next expiring bucket @index 2. The upper | |
1337 | * levels have the next expiring bucket @index 1. | |
1338 | * | |
1339 | * In case that the propagation wraps the next level the same | |
1340 | * rules apply: | |
1341 | * | |
1342 | * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0 | |
1343 | * 0 0 0 0 F 2 | |
1344 | * | |
1345 | * So after looking at LVL0 we get: | |
1346 | * | |
1347 | * LVL5 LVL4 LVL3 LVL2 LVL1 | |
1348 | * 0 0 0 1 0 | |
1349 | * | |
1350 | * So no propagation from LVL1 to LVL2 because that happened | |
1351 | * with the add already, but then we need to propagate further | |
1352 | * from LVL2 to LVL3. | |
1353 | * | |
1354 | * So the simple check whether the lower bits of the current | |
1355 | * level are 0 or not is sufficient for all cases. | |
1356 | */ | |
1357 | adj = clk & LVL_CLK_MASK ? 1 : 0; | |
1358 | clk >>= LVL_CLK_SHIFT; | |
1359 | clk += adj; | |
1da177e4 | 1360 | } |
500462a9 TG |
1361 | spin_unlock(&base->lock); |
1362 | return next; | |
1cfd6849 | 1363 | } |
69239749 | 1364 | |
1cfd6849 TG |
1365 | /* |
1366 | * Check, if the next hrtimer event is before the next timer wheel | |
1367 | * event: | |
1368 | */ | |
c1ad348b | 1369 | static u64 cmp_next_hrtimer_event(u64 basem, u64 expires) |
1cfd6849 | 1370 | { |
c1ad348b | 1371 | u64 nextevt = hrtimer_get_next_event(); |
0662b713 | 1372 | |
9501b6cf | 1373 | /* |
c1ad348b TG |
1374 | * If high resolution timers are enabled |
1375 | * hrtimer_get_next_event() returns KTIME_MAX. | |
9501b6cf | 1376 | */ |
c1ad348b TG |
1377 | if (expires <= nextevt) |
1378 | return expires; | |
eaad084b TG |
1379 | |
1380 | /* | |
c1ad348b TG |
1381 | * If the next timer is already expired, return the tick base |
1382 | * time so the tick is fired immediately. | |
eaad084b | 1383 | */ |
c1ad348b TG |
1384 | if (nextevt <= basem) |
1385 | return basem; | |
eaad084b | 1386 | |
9501b6cf | 1387 | /* |
c1ad348b TG |
1388 | * Round up to the next jiffie. High resolution timers are |
1389 | * off, so the hrtimers are expired in the tick and we need to | |
1390 | * make sure that this tick really expires the timer to avoid | |
1391 | * a ping pong of the nohz stop code. | |
1392 | * | |
1393 | * Use DIV_ROUND_UP_ULL to prevent gcc calling __divdi3 | |
9501b6cf | 1394 | */ |
c1ad348b | 1395 | return DIV_ROUND_UP_ULL(nextevt, TICK_NSEC) * TICK_NSEC; |
1da177e4 | 1396 | } |
1cfd6849 TG |
1397 | |
1398 | /** | |
c1ad348b TG |
1399 | * get_next_timer_interrupt - return the time (clock mono) of the next timer |
1400 | * @basej: base time jiffies | |
1401 | * @basem: base time clock monotonic | |
1402 | * | |
1403 | * Returns the tick aligned clock monotonic time of the next pending | |
1404 | * timer or KTIME_MAX if no timer is pending. | |
1cfd6849 | 1405 | */ |
c1ad348b | 1406 | u64 get_next_timer_interrupt(unsigned long basej, u64 basem) |
1cfd6849 | 1407 | { |
500462a9 | 1408 | struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]); |
c1ad348b TG |
1409 | u64 expires = KTIME_MAX; |
1410 | unsigned long nextevt; | |
1cfd6849 | 1411 | |
dbd87b5a HC |
1412 | /* |
1413 | * Pretend that there is no timer pending if the cpu is offline. | |
1414 | * Possible pending timers will be migrated later to an active cpu. | |
1415 | */ | |
1416 | if (cpu_is_offline(smp_processor_id())) | |
e40468a5 TG |
1417 | return expires; |
1418 | ||
500462a9 TG |
1419 | nextevt = __next_timer_interrupt(base); |
1420 | if (time_before_eq(nextevt, basej)) | |
1421 | expires = basem; | |
1422 | else | |
1423 | expires = basem + (nextevt - basej) * TICK_NSEC; | |
1cfd6849 | 1424 | |
c1ad348b | 1425 | return cmp_next_hrtimer_event(basem, expires); |
1cfd6849 | 1426 | } |
1da177e4 LT |
1427 | #endif |
1428 | ||
1da177e4 | 1429 | /* |
5b4db0c2 | 1430 | * Called from the timer interrupt handler to charge one tick to the current |
1da177e4 LT |
1431 | * process. user_tick is 1 if the tick is user time, 0 for system. |
1432 | */ | |
1433 | void update_process_times(int user_tick) | |
1434 | { | |
1435 | struct task_struct *p = current; | |
1da177e4 LT |
1436 | |
1437 | /* Note: this timer irq context must be accounted for as well. */ | |
fa13a5a1 | 1438 | account_process_tick(p, user_tick); |
1da177e4 | 1439 | run_local_timers(); |
c3377c2d | 1440 | rcu_check_callbacks(user_tick); |
e360adbe PZ |
1441 | #ifdef CONFIG_IRQ_WORK |
1442 | if (in_irq()) | |
76a33061 | 1443 | irq_work_tick(); |
e360adbe | 1444 | #endif |
1da177e4 | 1445 | scheduler_tick(); |
6819457d | 1446 | run_posix_cpu_timers(p); |
1da177e4 LT |
1447 | } |
1448 | ||
73420fea AMG |
1449 | /** |
1450 | * __run_timers - run all expired timers (if any) on this CPU. | |
1451 | * @base: the timer vector to be processed. | |
1452 | */ | |
1453 | static inline void __run_timers(struct timer_base *base) | |
1454 | { | |
1455 | struct hlist_head heads[LVL_DEPTH]; | |
1456 | int levels; | |
1457 | ||
1458 | if (!time_after_eq(jiffies, base->clk)) | |
1459 | return; | |
1460 | ||
1461 | spin_lock_irq(&base->lock); | |
1462 | ||
1463 | while (time_after_eq(jiffies, base->clk)) { | |
1464 | ||
1465 | levels = collect_expired_timers(base, heads); | |
1466 | base->clk++; | |
1467 | ||
1468 | while (levels--) | |
1469 | expire_timers(base, heads + levels); | |
1470 | } | |
1471 | base->running_timer = NULL; | |
1472 | spin_unlock_irq(&base->lock); | |
1473 | } | |
1474 | ||
1da177e4 LT |
1475 | /* |
1476 | * This function runs timers and the timer-tq in bottom half context. | |
1477 | */ | |
1478 | static void run_timer_softirq(struct softirq_action *h) | |
1479 | { | |
500462a9 | 1480 | struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]); |
1da177e4 | 1481 | |
500462a9 TG |
1482 | __run_timers(base); |
1483 | if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && base->nohz_active) | |
1484 | __run_timers(this_cpu_ptr(&timer_bases[BASE_DEF])); | |
1da177e4 LT |
1485 | } |
1486 | ||
1487 | /* | |
1488 | * Called by the local, per-CPU timer interrupt on SMP. | |
1489 | */ | |
1490 | void run_local_timers(void) | |
1491 | { | |
d3d74453 | 1492 | hrtimer_run_queues(); |
1da177e4 LT |
1493 | raise_softirq(TIMER_SOFTIRQ); |
1494 | } | |
1495 | ||
1da177e4 LT |
1496 | #ifdef __ARCH_WANT_SYS_ALARM |
1497 | ||
1498 | /* | |
1499 | * For backwards compatibility? This can be done in libc so Alpha | |
1500 | * and all newer ports shouldn't need it. | |
1501 | */ | |
58fd3aa2 | 1502 | SYSCALL_DEFINE1(alarm, unsigned int, seconds) |
1da177e4 | 1503 | { |
c08b8a49 | 1504 | return alarm_setitimer(seconds); |
1da177e4 LT |
1505 | } |
1506 | ||
1507 | #endif | |
1508 | ||
1da177e4 LT |
1509 | static void process_timeout(unsigned long __data) |
1510 | { | |
36c8b586 | 1511 | wake_up_process((struct task_struct *)__data); |
1da177e4 LT |
1512 | } |
1513 | ||
1514 | /** | |
1515 | * schedule_timeout - sleep until timeout | |
1516 | * @timeout: timeout value in jiffies | |
1517 | * | |
1518 | * Make the current task sleep until @timeout jiffies have | |
1519 | * elapsed. The routine will return immediately unless | |
1520 | * the current task state has been set (see set_current_state()). | |
1521 | * | |
1522 | * You can set the task state as follows - | |
1523 | * | |
1524 | * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to | |
1525 | * pass before the routine returns. The routine will return 0 | |
1526 | * | |
1527 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
1528 | * delivered to the current task. In this case the remaining time | |
1529 | * in jiffies will be returned, or 0 if the timer expired in time | |
1530 | * | |
1531 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1532 | * routine returns. | |
1533 | * | |
1534 | * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule | |
1535 | * the CPU away without a bound on the timeout. In this case the return | |
1536 | * value will be %MAX_SCHEDULE_TIMEOUT. | |
1537 | * | |
1538 | * In all cases the return value is guaranteed to be non-negative. | |
1539 | */ | |
7ad5b3a5 | 1540 | signed long __sched schedule_timeout(signed long timeout) |
1da177e4 LT |
1541 | { |
1542 | struct timer_list timer; | |
1543 | unsigned long expire; | |
1544 | ||
1545 | switch (timeout) | |
1546 | { | |
1547 | case MAX_SCHEDULE_TIMEOUT: | |
1548 | /* | |
1549 | * These two special cases are useful to be comfortable | |
1550 | * in the caller. Nothing more. We could take | |
1551 | * MAX_SCHEDULE_TIMEOUT from one of the negative value | |
1552 | * but I' d like to return a valid offset (>=0) to allow | |
1553 | * the caller to do everything it want with the retval. | |
1554 | */ | |
1555 | schedule(); | |
1556 | goto out; | |
1557 | default: | |
1558 | /* | |
1559 | * Another bit of PARANOID. Note that the retval will be | |
1560 | * 0 since no piece of kernel is supposed to do a check | |
1561 | * for a negative retval of schedule_timeout() (since it | |
1562 | * should never happens anyway). You just have the printk() | |
1563 | * that will tell you if something is gone wrong and where. | |
1564 | */ | |
5b149bcc | 1565 | if (timeout < 0) { |
1da177e4 | 1566 | printk(KERN_ERR "schedule_timeout: wrong timeout " |
5b149bcc AM |
1567 | "value %lx\n", timeout); |
1568 | dump_stack(); | |
1da177e4 LT |
1569 | current->state = TASK_RUNNING; |
1570 | goto out; | |
1571 | } | |
1572 | } | |
1573 | ||
1574 | expire = timeout + jiffies; | |
1575 | ||
c6f3a97f | 1576 | setup_timer_on_stack(&timer, process_timeout, (unsigned long)current); |
177ec0a0 | 1577 | __mod_timer(&timer, expire, false); |
1da177e4 LT |
1578 | schedule(); |
1579 | del_singleshot_timer_sync(&timer); | |
1580 | ||
c6f3a97f TG |
1581 | /* Remove the timer from the object tracker */ |
1582 | destroy_timer_on_stack(&timer); | |
1583 | ||
1da177e4 LT |
1584 | timeout = expire - jiffies; |
1585 | ||
1586 | out: | |
1587 | return timeout < 0 ? 0 : timeout; | |
1588 | } | |
1da177e4 LT |
1589 | EXPORT_SYMBOL(schedule_timeout); |
1590 | ||
8a1c1757 AM |
1591 | /* |
1592 | * We can use __set_current_state() here because schedule_timeout() calls | |
1593 | * schedule() unconditionally. | |
1594 | */ | |
64ed93a2 NA |
1595 | signed long __sched schedule_timeout_interruptible(signed long timeout) |
1596 | { | |
a5a0d52c AM |
1597 | __set_current_state(TASK_INTERRUPTIBLE); |
1598 | return schedule_timeout(timeout); | |
64ed93a2 NA |
1599 | } |
1600 | EXPORT_SYMBOL(schedule_timeout_interruptible); | |
1601 | ||
294d5cc2 MW |
1602 | signed long __sched schedule_timeout_killable(signed long timeout) |
1603 | { | |
1604 | __set_current_state(TASK_KILLABLE); | |
1605 | return schedule_timeout(timeout); | |
1606 | } | |
1607 | EXPORT_SYMBOL(schedule_timeout_killable); | |
1608 | ||
64ed93a2 NA |
1609 | signed long __sched schedule_timeout_uninterruptible(signed long timeout) |
1610 | { | |
a5a0d52c AM |
1611 | __set_current_state(TASK_UNINTERRUPTIBLE); |
1612 | return schedule_timeout(timeout); | |
64ed93a2 NA |
1613 | } |
1614 | EXPORT_SYMBOL(schedule_timeout_uninterruptible); | |
1615 | ||
69b27baf AM |
1616 | /* |
1617 | * Like schedule_timeout_uninterruptible(), except this task will not contribute | |
1618 | * to load average. | |
1619 | */ | |
1620 | signed long __sched schedule_timeout_idle(signed long timeout) | |
1621 | { | |
1622 | __set_current_state(TASK_IDLE); | |
1623 | return schedule_timeout(timeout); | |
1624 | } | |
1625 | EXPORT_SYMBOL(schedule_timeout_idle); | |
1626 | ||
1da177e4 | 1627 | #ifdef CONFIG_HOTPLUG_CPU |
494af3ed | 1628 | static void migrate_timer_list(struct timer_base *new_base, struct hlist_head *head) |
1da177e4 LT |
1629 | { |
1630 | struct timer_list *timer; | |
0eeda71b | 1631 | int cpu = new_base->cpu; |
1da177e4 | 1632 | |
1dabbcec TG |
1633 | while (!hlist_empty(head)) { |
1634 | timer = hlist_entry(head->first, struct timer_list, entry); | |
ec44bc7a | 1635 | detach_timer(timer, false); |
0eeda71b | 1636 | timer->flags = (timer->flags & ~TIMER_BASEMASK) | cpu; |
1da177e4 | 1637 | internal_add_timer(new_base, timer); |
1da177e4 | 1638 | } |
1da177e4 LT |
1639 | } |
1640 | ||
0db0628d | 1641 | static void migrate_timers(int cpu) |
1da177e4 | 1642 | { |
494af3ed TG |
1643 | struct timer_base *old_base; |
1644 | struct timer_base *new_base; | |
500462a9 | 1645 | int b, i; |
1da177e4 LT |
1646 | |
1647 | BUG_ON(cpu_online(cpu)); | |
55c888d6 | 1648 | |
500462a9 TG |
1649 | for (b = 0; b < NR_BASES; b++) { |
1650 | old_base = per_cpu_ptr(&timer_bases[b], cpu); | |
1651 | new_base = get_cpu_ptr(&timer_bases[b]); | |
1652 | /* | |
1653 | * The caller is globally serialized and nobody else | |
1654 | * takes two locks at once, deadlock is not possible. | |
1655 | */ | |
1656 | spin_lock_irq(&new_base->lock); | |
1657 | spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING); | |
1658 | ||
1659 | BUG_ON(old_base->running_timer); | |
1660 | ||
1661 | for (i = 0; i < WHEEL_SIZE; i++) | |
1662 | migrate_timer_list(new_base, old_base->vectors + i); | |
8def9060 | 1663 | |
500462a9 TG |
1664 | spin_unlock(&old_base->lock); |
1665 | spin_unlock_irq(&new_base->lock); | |
1666 | put_cpu_ptr(&timer_bases); | |
1667 | } | |
1da177e4 | 1668 | } |
1da177e4 | 1669 | |
0db0628d | 1670 | static int timer_cpu_notify(struct notifier_block *self, |
1da177e4 LT |
1671 | unsigned long action, void *hcpu) |
1672 | { | |
8def9060 | 1673 | switch (action) { |
1da177e4 | 1674 | case CPU_DEAD: |
8bb78442 | 1675 | case CPU_DEAD_FROZEN: |
8def9060 | 1676 | migrate_timers((long)hcpu); |
1da177e4 | 1677 | break; |
1da177e4 LT |
1678 | default: |
1679 | break; | |
1680 | } | |
3650b57f | 1681 | |
1da177e4 LT |
1682 | return NOTIFY_OK; |
1683 | } | |
1684 | ||
3650b57f PZ |
1685 | static inline void timer_register_cpu_notifier(void) |
1686 | { | |
1687 | cpu_notifier(timer_cpu_notify, 0); | |
1688 | } | |
1689 | #else | |
1690 | static inline void timer_register_cpu_notifier(void) { } | |
1691 | #endif /* CONFIG_HOTPLUG_CPU */ | |
1da177e4 | 1692 | |
0eeda71b | 1693 | static void __init init_timer_cpu(int cpu) |
8def9060 | 1694 | { |
500462a9 TG |
1695 | struct timer_base *base; |
1696 | int i; | |
8def9060 | 1697 | |
500462a9 TG |
1698 | for (i = 0; i < NR_BASES; i++) { |
1699 | base = per_cpu_ptr(&timer_bases[i], cpu); | |
1700 | base->cpu = cpu; | |
1701 | spin_lock_init(&base->lock); | |
1702 | base->clk = jiffies; | |
1703 | } | |
8def9060 VK |
1704 | } |
1705 | ||
1706 | static void __init init_timer_cpus(void) | |
1da177e4 | 1707 | { |
8def9060 VK |
1708 | int cpu; |
1709 | ||
0eeda71b TG |
1710 | for_each_possible_cpu(cpu) |
1711 | init_timer_cpu(cpu); | |
8def9060 | 1712 | } |
e52b1db3 | 1713 | |
8def9060 VK |
1714 | void __init init_timers(void) |
1715 | { | |
8def9060 | 1716 | init_timer_cpus(); |
c24a4a36 | 1717 | init_timer_stats(); |
3650b57f | 1718 | timer_register_cpu_notifier(); |
962cf36c | 1719 | open_softirq(TIMER_SOFTIRQ, run_timer_softirq); |
1da177e4 LT |
1720 | } |
1721 | ||
1da177e4 LT |
1722 | /** |
1723 | * msleep - sleep safely even with waitqueue interruptions | |
1724 | * @msecs: Time in milliseconds to sleep for | |
1725 | */ | |
1726 | void msleep(unsigned int msecs) | |
1727 | { | |
1728 | unsigned long timeout = msecs_to_jiffies(msecs) + 1; | |
1729 | ||
75bcc8c5 NA |
1730 | while (timeout) |
1731 | timeout = schedule_timeout_uninterruptible(timeout); | |
1da177e4 LT |
1732 | } |
1733 | ||
1734 | EXPORT_SYMBOL(msleep); | |
1735 | ||
1736 | /** | |
96ec3efd | 1737 | * msleep_interruptible - sleep waiting for signals |
1da177e4 LT |
1738 | * @msecs: Time in milliseconds to sleep for |
1739 | */ | |
1740 | unsigned long msleep_interruptible(unsigned int msecs) | |
1741 | { | |
1742 | unsigned long timeout = msecs_to_jiffies(msecs) + 1; | |
1743 | ||
75bcc8c5 NA |
1744 | while (timeout && !signal_pending(current)) |
1745 | timeout = schedule_timeout_interruptible(timeout); | |
1da177e4 LT |
1746 | return jiffies_to_msecs(timeout); |
1747 | } | |
1748 | ||
1749 | EXPORT_SYMBOL(msleep_interruptible); | |
5e7f5a17 | 1750 | |
6deba083 | 1751 | static void __sched do_usleep_range(unsigned long min, unsigned long max) |
5e7f5a17 PP |
1752 | { |
1753 | ktime_t kmin; | |
da8b44d5 | 1754 | u64 delta; |
5e7f5a17 PP |
1755 | |
1756 | kmin = ktime_set(0, min * NSEC_PER_USEC); | |
da8b44d5 | 1757 | delta = (u64)(max - min) * NSEC_PER_USEC; |
6deba083 | 1758 | schedule_hrtimeout_range(&kmin, delta, HRTIMER_MODE_REL); |
5e7f5a17 PP |
1759 | } |
1760 | ||
1761 | /** | |
1762 | * usleep_range - Drop in replacement for udelay where wakeup is flexible | |
1763 | * @min: Minimum time in usecs to sleep | |
1764 | * @max: Maximum time in usecs to sleep | |
1765 | */ | |
2ad5d327 | 1766 | void __sched usleep_range(unsigned long min, unsigned long max) |
5e7f5a17 PP |
1767 | { |
1768 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
1769 | do_usleep_range(min, max); | |
1770 | } | |
1771 | EXPORT_SYMBOL(usleep_range); |