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> |
174cd4b1 | 41 | #include <linux/sched/signal.h> |
cf4aebc2 | 42 | #include <linux/sched/sysctl.h> |
370c9135 | 43 | #include <linux/sched/nohz.h> |
b17b0153 | 44 | #include <linux/sched/debug.h> |
5a0e3ad6 | 45 | #include <linux/slab.h> |
1a0df594 | 46 | #include <linux/compat.h> |
1da177e4 | 47 | |
7c0f6ba6 | 48 | #include <linux/uaccess.h> |
1da177e4 LT |
49 | #include <asm/unistd.h> |
50 | #include <asm/div64.h> | |
51 | #include <asm/timex.h> | |
52 | #include <asm/io.h> | |
53 | ||
c1ad348b TG |
54 | #include "tick-internal.h" |
55 | ||
2b022e3d XG |
56 | #define CREATE_TRACE_POINTS |
57 | #include <trace/events/timer.h> | |
58 | ||
40747ffa | 59 | __visible u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES; |
ecea8d19 TG |
60 | |
61 | EXPORT_SYMBOL(jiffies_64); | |
62 | ||
1da177e4 | 63 | /* |
500462a9 TG |
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) | |
1da177e4 | 149 | */ |
1da177e4 | 150 | |
500462a9 TG |
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)) | |
1da177e4 | 157 | |
500462a9 TG |
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 | |
1da177e4 | 196 | |
494af3ed | 197 | struct timer_base { |
2287d866 | 198 | raw_spinlock_t lock; |
500462a9 TG |
199 | struct timer_list *running_timer; |
200 | unsigned long clk; | |
a683f390 | 201 | unsigned long next_expiry; |
500462a9 | 202 | unsigned int cpu; |
a683f390 | 203 | bool is_idle; |
2fe59f50 | 204 | bool must_forward_clk; |
500462a9 TG |
205 | DECLARE_BITMAP(pending_map, WHEEL_SIZE); |
206 | struct hlist_head vectors[WHEEL_SIZE]; | |
6e453a67 | 207 | } ____cacheline_aligned; |
e52b1db3 | 208 | |
500462a9 | 209 | static DEFINE_PER_CPU(struct timer_base, timer_bases[NR_BASES]); |
6e453a67 | 210 | |
ae67bada TG |
211 | #ifdef CONFIG_NO_HZ_COMMON |
212 | ||
14c80341 | 213 | static DEFINE_STATIC_KEY_FALSE(timers_nohz_active); |
ae67bada TG |
214 | static DEFINE_MUTEX(timer_keys_mutex); |
215 | ||
216 | static void timer_update_keys(struct work_struct *work); | |
217 | static DECLARE_WORK(timer_update_work, timer_update_keys); | |
218 | ||
219 | #ifdef CONFIG_SMP | |
bc7a34b8 TG |
220 | unsigned int sysctl_timer_migration = 1; |
221 | ||
ae67bada TG |
222 | DEFINE_STATIC_KEY_FALSE(timers_migration_enabled); |
223 | ||
224 | static void timers_update_migration(void) | |
bc7a34b8 | 225 | { |
ae67bada TG |
226 | if (sysctl_timer_migration && tick_nohz_active) |
227 | static_branch_enable(&timers_migration_enabled); | |
228 | else | |
229 | static_branch_disable(&timers_migration_enabled); | |
230 | } | |
231 | #else | |
232 | static inline void timers_update_migration(void) { } | |
233 | #endif /* !CONFIG_SMP */ | |
bc7a34b8 | 234 | |
ae67bada TG |
235 | static void timer_update_keys(struct work_struct *work) |
236 | { | |
237 | mutex_lock(&timer_keys_mutex); | |
238 | timers_update_migration(); | |
239 | static_branch_enable(&timers_nohz_active); | |
240 | mutex_unlock(&timer_keys_mutex); | |
241 | } | |
bc7a34b8 | 242 | |
ae67bada TG |
243 | void timers_update_nohz(void) |
244 | { | |
245 | schedule_work(&timer_update_work); | |
bc7a34b8 TG |
246 | } |
247 | ||
248 | int timer_migration_handler(struct ctl_table *table, int write, | |
249 | void __user *buffer, size_t *lenp, | |
250 | loff_t *ppos) | |
251 | { | |
bc7a34b8 TG |
252 | int ret; |
253 | ||
ae67bada | 254 | mutex_lock(&timer_keys_mutex); |
b94bf594 | 255 | ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos); |
bc7a34b8 | 256 | if (!ret && write) |
ae67bada TG |
257 | timers_update_migration(); |
258 | mutex_unlock(&timer_keys_mutex); | |
bc7a34b8 TG |
259 | return ret; |
260 | } | |
14c80341 AMG |
261 | |
262 | static inline bool is_timers_nohz_active(void) | |
263 | { | |
264 | return static_branch_unlikely(&timers_nohz_active); | |
265 | } | |
266 | #else | |
267 | static inline bool is_timers_nohz_active(void) { return false; } | |
ae67bada | 268 | #endif /* NO_HZ_COMMON */ |
bc7a34b8 | 269 | |
9c133c46 AS |
270 | static unsigned long round_jiffies_common(unsigned long j, int cpu, |
271 | bool force_up) | |
4c36a5de AV |
272 | { |
273 | int rem; | |
274 | unsigned long original = j; | |
275 | ||
276 | /* | |
277 | * We don't want all cpus firing their timers at once hitting the | |
278 | * same lock or cachelines, so we skew each extra cpu with an extra | |
279 | * 3 jiffies. This 3 jiffies came originally from the mm/ code which | |
280 | * already did this. | |
281 | * The skew is done by adding 3*cpunr, then round, then subtract this | |
282 | * extra offset again. | |
283 | */ | |
284 | j += cpu * 3; | |
285 | ||
286 | rem = j % HZ; | |
287 | ||
288 | /* | |
289 | * If the target jiffie is just after a whole second (which can happen | |
290 | * due to delays of the timer irq, long irq off times etc etc) then | |
291 | * we should round down to the whole second, not up. Use 1/4th second | |
292 | * as cutoff for this rounding as an extreme upper bound for this. | |
9c133c46 | 293 | * But never round down if @force_up is set. |
4c36a5de | 294 | */ |
9c133c46 | 295 | if (rem < HZ/4 && !force_up) /* round down */ |
4c36a5de AV |
296 | j = j - rem; |
297 | else /* round up */ | |
298 | j = j - rem + HZ; | |
299 | ||
300 | /* now that we have rounded, subtract the extra skew again */ | |
301 | j -= cpu * 3; | |
302 | ||
9e04d380 BVA |
303 | /* |
304 | * Make sure j is still in the future. Otherwise return the | |
305 | * unmodified value. | |
306 | */ | |
307 | return time_is_after_jiffies(j) ? j : original; | |
4c36a5de | 308 | } |
9c133c46 AS |
309 | |
310 | /** | |
311 | * __round_jiffies - function to round jiffies to a full second | |
312 | * @j: the time in (absolute) jiffies that should be rounded | |
313 | * @cpu: the processor number on which the timeout will happen | |
314 | * | |
315 | * __round_jiffies() rounds an absolute time in the future (in jiffies) | |
316 | * up or down to (approximately) full seconds. This is useful for timers | |
317 | * for which the exact time they fire does not matter too much, as long as | |
318 | * they fire approximately every X seconds. | |
319 | * | |
320 | * By rounding these timers to whole seconds, all such timers will fire | |
321 | * at the same time, rather than at various times spread out. The goal | |
322 | * of this is to have the CPU wake up less, which saves power. | |
323 | * | |
324 | * The exact rounding is skewed for each processor to avoid all | |
325 | * processors firing at the exact same time, which could lead | |
326 | * to lock contention or spurious cache line bouncing. | |
327 | * | |
328 | * The return value is the rounded version of the @j parameter. | |
329 | */ | |
330 | unsigned long __round_jiffies(unsigned long j, int cpu) | |
331 | { | |
332 | return round_jiffies_common(j, cpu, false); | |
333 | } | |
4c36a5de AV |
334 | EXPORT_SYMBOL_GPL(__round_jiffies); |
335 | ||
336 | /** | |
337 | * __round_jiffies_relative - function to round jiffies to a full second | |
338 | * @j: the time in (relative) jiffies that should be rounded | |
339 | * @cpu: the processor number on which the timeout will happen | |
340 | * | |
72fd4a35 | 341 | * __round_jiffies_relative() rounds a time delta in the future (in jiffies) |
4c36a5de AV |
342 | * up or down to (approximately) full seconds. This is useful for timers |
343 | * for which the exact time they fire does not matter too much, as long as | |
344 | * they fire approximately every X seconds. | |
345 | * | |
346 | * By rounding these timers to whole seconds, all such timers will fire | |
347 | * at the same time, rather than at various times spread out. The goal | |
348 | * of this is to have the CPU wake up less, which saves power. | |
349 | * | |
350 | * The exact rounding is skewed for each processor to avoid all | |
351 | * processors firing at the exact same time, which could lead | |
352 | * to lock contention or spurious cache line bouncing. | |
353 | * | |
72fd4a35 | 354 | * The return value is the rounded version of the @j parameter. |
4c36a5de AV |
355 | */ |
356 | unsigned long __round_jiffies_relative(unsigned long j, int cpu) | |
357 | { | |
9c133c46 AS |
358 | unsigned long j0 = jiffies; |
359 | ||
360 | /* Use j0 because jiffies might change while we run */ | |
361 | return round_jiffies_common(j + j0, cpu, false) - j0; | |
4c36a5de AV |
362 | } |
363 | EXPORT_SYMBOL_GPL(__round_jiffies_relative); | |
364 | ||
365 | /** | |
366 | * round_jiffies - function to round jiffies to a full second | |
367 | * @j: the time in (absolute) jiffies that should be rounded | |
368 | * | |
72fd4a35 | 369 | * round_jiffies() rounds an absolute time in the future (in jiffies) |
4c36a5de AV |
370 | * up or down to (approximately) full seconds. This is useful for timers |
371 | * for which the exact time they fire does not matter too much, as long as | |
372 | * they fire approximately every X seconds. | |
373 | * | |
374 | * By rounding these timers to whole seconds, all such timers will fire | |
375 | * at the same time, rather than at various times spread out. The goal | |
376 | * of this is to have the CPU wake up less, which saves power. | |
377 | * | |
72fd4a35 | 378 | * The return value is the rounded version of the @j parameter. |
4c36a5de AV |
379 | */ |
380 | unsigned long round_jiffies(unsigned long j) | |
381 | { | |
9c133c46 | 382 | return round_jiffies_common(j, raw_smp_processor_id(), false); |
4c36a5de AV |
383 | } |
384 | EXPORT_SYMBOL_GPL(round_jiffies); | |
385 | ||
386 | /** | |
387 | * round_jiffies_relative - function to round jiffies to a full second | |
388 | * @j: the time in (relative) jiffies that should be rounded | |
389 | * | |
72fd4a35 | 390 | * round_jiffies_relative() rounds a time delta in the future (in jiffies) |
4c36a5de AV |
391 | * up or down to (approximately) full seconds. This is useful for timers |
392 | * for which the exact time they fire does not matter too much, as long as | |
393 | * they fire approximately every X seconds. | |
394 | * | |
395 | * By rounding these timers to whole seconds, all such timers will fire | |
396 | * at the same time, rather than at various times spread out. The goal | |
397 | * of this is to have the CPU wake up less, which saves power. | |
398 | * | |
72fd4a35 | 399 | * The return value is the rounded version of the @j parameter. |
4c36a5de AV |
400 | */ |
401 | unsigned long round_jiffies_relative(unsigned long j) | |
402 | { | |
403 | return __round_jiffies_relative(j, raw_smp_processor_id()); | |
404 | } | |
405 | EXPORT_SYMBOL_GPL(round_jiffies_relative); | |
406 | ||
9c133c46 AS |
407 | /** |
408 | * __round_jiffies_up - function to round jiffies up to a full second | |
409 | * @j: the time in (absolute) jiffies that should be rounded | |
410 | * @cpu: the processor number on which the timeout will happen | |
411 | * | |
412 | * This is the same as __round_jiffies() except that it will never | |
413 | * round down. This is useful for timeouts for which the exact time | |
414 | * of firing does not matter too much, as long as they don't fire too | |
415 | * early. | |
416 | */ | |
417 | unsigned long __round_jiffies_up(unsigned long j, int cpu) | |
418 | { | |
419 | return round_jiffies_common(j, cpu, true); | |
420 | } | |
421 | EXPORT_SYMBOL_GPL(__round_jiffies_up); | |
422 | ||
423 | /** | |
424 | * __round_jiffies_up_relative - function to round jiffies up to a full second | |
425 | * @j: the time in (relative) jiffies that should be rounded | |
426 | * @cpu: the processor number on which the timeout will happen | |
427 | * | |
428 | * This is the same as __round_jiffies_relative() except that it will never | |
429 | * round down. This is useful for timeouts for which the exact time | |
430 | * of firing does not matter too much, as long as they don't fire too | |
431 | * early. | |
432 | */ | |
433 | unsigned long __round_jiffies_up_relative(unsigned long j, int cpu) | |
434 | { | |
435 | unsigned long j0 = jiffies; | |
436 | ||
437 | /* Use j0 because jiffies might change while we run */ | |
438 | return round_jiffies_common(j + j0, cpu, true) - j0; | |
439 | } | |
440 | EXPORT_SYMBOL_GPL(__round_jiffies_up_relative); | |
441 | ||
442 | /** | |
443 | * round_jiffies_up - function to round jiffies up to a full second | |
444 | * @j: the time in (absolute) jiffies that should be rounded | |
445 | * | |
446 | * This is the same as round_jiffies() except that it will never | |
447 | * round down. This is useful for timeouts for which the exact time | |
448 | * of firing does not matter too much, as long as they don't fire too | |
449 | * early. | |
450 | */ | |
451 | unsigned long round_jiffies_up(unsigned long j) | |
452 | { | |
453 | return round_jiffies_common(j, raw_smp_processor_id(), true); | |
454 | } | |
455 | EXPORT_SYMBOL_GPL(round_jiffies_up); | |
456 | ||
457 | /** | |
458 | * round_jiffies_up_relative - function to round jiffies up to a full second | |
459 | * @j: the time in (relative) jiffies that should be rounded | |
460 | * | |
461 | * This is the same as round_jiffies_relative() except that it will never | |
462 | * round down. This is useful for timeouts for which the exact time | |
463 | * of firing does not matter too much, as long as they don't fire too | |
464 | * early. | |
465 | */ | |
466 | unsigned long round_jiffies_up_relative(unsigned long j) | |
467 | { | |
468 | return __round_jiffies_up_relative(j, raw_smp_processor_id()); | |
469 | } | |
470 | EXPORT_SYMBOL_GPL(round_jiffies_up_relative); | |
471 | ||
3bbb9ec9 | 472 | |
500462a9 | 473 | static inline unsigned int timer_get_idx(struct timer_list *timer) |
3bbb9ec9 | 474 | { |
500462a9 | 475 | return (timer->flags & TIMER_ARRAYMASK) >> TIMER_ARRAYSHIFT; |
3bbb9ec9 | 476 | } |
3bbb9ec9 | 477 | |
500462a9 | 478 | static inline void timer_set_idx(struct timer_list *timer, unsigned int idx) |
1da177e4 | 479 | { |
500462a9 TG |
480 | timer->flags = (timer->flags & ~TIMER_ARRAYMASK) | |
481 | idx << TIMER_ARRAYSHIFT; | |
482 | } | |
1da177e4 | 483 | |
500462a9 TG |
484 | /* |
485 | * Helper function to calculate the array index for a given expiry | |
486 | * time. | |
487 | */ | |
488 | static inline unsigned calc_index(unsigned expires, unsigned lvl) | |
489 | { | |
490 | expires = (expires + LVL_GRAN(lvl)) >> LVL_SHIFT(lvl); | |
491 | return LVL_OFFS(lvl) + (expires & LVL_MASK); | |
492 | } | |
493 | ||
ffdf0477 | 494 | static int calc_wheel_index(unsigned long expires, unsigned long clk) |
1da177e4 | 495 | { |
ffdf0477 | 496 | unsigned long delta = expires - clk; |
500462a9 TG |
497 | unsigned int idx; |
498 | ||
499 | if (delta < LVL_START(1)) { | |
500 | idx = calc_index(expires, 0); | |
501 | } else if (delta < LVL_START(2)) { | |
502 | idx = calc_index(expires, 1); | |
503 | } else if (delta < LVL_START(3)) { | |
504 | idx = calc_index(expires, 2); | |
505 | } else if (delta < LVL_START(4)) { | |
506 | idx = calc_index(expires, 3); | |
507 | } else if (delta < LVL_START(5)) { | |
508 | idx = calc_index(expires, 4); | |
509 | } else if (delta < LVL_START(6)) { | |
510 | idx = calc_index(expires, 5); | |
511 | } else if (delta < LVL_START(7)) { | |
512 | idx = calc_index(expires, 6); | |
513 | } else if (LVL_DEPTH > 8 && delta < LVL_START(8)) { | |
514 | idx = calc_index(expires, 7); | |
515 | } else if ((long) delta < 0) { | |
ffdf0477 | 516 | idx = clk & LVL_MASK; |
1da177e4 | 517 | } else { |
500462a9 TG |
518 | /* |
519 | * Force expire obscene large timeouts to expire at the | |
520 | * capacity limit of the wheel. | |
1da177e4 | 521 | */ |
500462a9 TG |
522 | if (expires >= WHEEL_TIMEOUT_CUTOFF) |
523 | expires = WHEEL_TIMEOUT_MAX; | |
1bd04bf6 | 524 | |
500462a9 | 525 | idx = calc_index(expires, LVL_DEPTH - 1); |
1da177e4 | 526 | } |
ffdf0477 AMG |
527 | return idx; |
528 | } | |
1bd04bf6 | 529 | |
ffdf0477 AMG |
530 | /* |
531 | * Enqueue the timer into the hash bucket, mark it pending in | |
532 | * the bitmap and store the index in the timer flags. | |
533 | */ | |
534 | static void enqueue_timer(struct timer_base *base, struct timer_list *timer, | |
535 | unsigned int idx) | |
536 | { | |
537 | hlist_add_head(&timer->entry, base->vectors + idx); | |
500462a9 TG |
538 | __set_bit(idx, base->pending_map); |
539 | timer_set_idx(timer, idx); | |
1da177e4 LT |
540 | } |
541 | ||
ffdf0477 AMG |
542 | static void |
543 | __internal_add_timer(struct timer_base *base, struct timer_list *timer) | |
facbb4a7 | 544 | { |
ffdf0477 AMG |
545 | unsigned int idx; |
546 | ||
547 | idx = calc_wheel_index(timer->expires, base->clk); | |
548 | enqueue_timer(base, timer, idx); | |
549 | } | |
9f6d9baa | 550 | |
ffdf0477 AMG |
551 | static void |
552 | trigger_dyntick_cpu(struct timer_base *base, struct timer_list *timer) | |
553 | { | |
ae67bada | 554 | if (!is_timers_nohz_active()) |
a683f390 | 555 | return; |
3bb475a3 | 556 | |
facbb4a7 | 557 | /* |
a683f390 TG |
558 | * TODO: This wants some optimizing similar to the code below, but we |
559 | * will do that when we switch from push to pull for deferrable timers. | |
facbb4a7 | 560 | */ |
a683f390 TG |
561 | if (timer->flags & TIMER_DEFERRABLE) { |
562 | if (tick_nohz_full_cpu(base->cpu)) | |
683be13a | 563 | wake_up_nohz_cpu(base->cpu); |
a683f390 | 564 | return; |
99d5f3aa | 565 | } |
9f6d9baa VK |
566 | |
567 | /* | |
a683f390 TG |
568 | * We might have to IPI the remote CPU if the base is idle and the |
569 | * timer is not deferrable. If the other CPU is on the way to idle | |
570 | * then it can't set base->is_idle as we hold the base lock: | |
9f6d9baa | 571 | */ |
a683f390 TG |
572 | if (!base->is_idle) |
573 | return; | |
574 | ||
575 | /* Check whether this is the new first expiring timer: */ | |
576 | if (time_after_eq(timer->expires, base->next_expiry)) | |
577 | return; | |
578 | ||
579 | /* | |
580 | * Set the next expiry time and kick the CPU so it can reevaluate the | |
581 | * wheel: | |
582 | */ | |
583 | base->next_expiry = timer->expires; | |
30587589 | 584 | wake_up_nohz_cpu(base->cpu); |
ffdf0477 AMG |
585 | } |
586 | ||
587 | static void | |
588 | internal_add_timer(struct timer_base *base, struct timer_list *timer) | |
589 | { | |
590 | __internal_add_timer(base, timer); | |
591 | trigger_dyntick_cpu(base, timer); | |
facbb4a7 TG |
592 | } |
593 | ||
c6f3a97f TG |
594 | #ifdef CONFIG_DEBUG_OBJECTS_TIMERS |
595 | ||
596 | static struct debug_obj_descr timer_debug_descr; | |
597 | ||
99777288 SG |
598 | static void *timer_debug_hint(void *addr) |
599 | { | |
600 | return ((struct timer_list *) addr)->function; | |
601 | } | |
602 | ||
b9fdac7f DC |
603 | static bool timer_is_static_object(void *addr) |
604 | { | |
605 | struct timer_list *timer = addr; | |
606 | ||
607 | return (timer->entry.pprev == NULL && | |
608 | timer->entry.next == TIMER_ENTRY_STATIC); | |
609 | } | |
610 | ||
c6f3a97f TG |
611 | /* |
612 | * fixup_init is called when: | |
613 | * - an active object is initialized | |
55c888d6 | 614 | */ |
e3252464 | 615 | static bool timer_fixup_init(void *addr, enum debug_obj_state state) |
c6f3a97f TG |
616 | { |
617 | struct timer_list *timer = addr; | |
618 | ||
619 | switch (state) { | |
620 | case ODEBUG_STATE_ACTIVE: | |
621 | del_timer_sync(timer); | |
622 | debug_object_init(timer, &timer_debug_descr); | |
e3252464 | 623 | return true; |
c6f3a97f | 624 | default: |
e3252464 | 625 | return false; |
c6f3a97f TG |
626 | } |
627 | } | |
628 | ||
fb16b8cf | 629 | /* Stub timer callback for improperly used timers. */ |
ba16490e | 630 | static void stub_timer(struct timer_list *unused) |
fb16b8cf SB |
631 | { |
632 | WARN_ON(1); | |
633 | } | |
634 | ||
c6f3a97f TG |
635 | /* |
636 | * fixup_activate is called when: | |
637 | * - an active object is activated | |
b9fdac7f | 638 | * - an unknown non-static object is activated |
c6f3a97f | 639 | */ |
e3252464 | 640 | static bool timer_fixup_activate(void *addr, enum debug_obj_state state) |
c6f3a97f TG |
641 | { |
642 | struct timer_list *timer = addr; | |
643 | ||
644 | switch (state) { | |
c6f3a97f | 645 | case ODEBUG_STATE_NOTAVAILABLE: |
ba16490e | 646 | timer_setup(timer, stub_timer, 0); |
b9fdac7f | 647 | return true; |
c6f3a97f TG |
648 | |
649 | case ODEBUG_STATE_ACTIVE: | |
650 | WARN_ON(1); | |
651 | ||
652 | default: | |
e3252464 | 653 | return false; |
c6f3a97f TG |
654 | } |
655 | } | |
656 | ||
657 | /* | |
658 | * fixup_free is called when: | |
659 | * - an active object is freed | |
660 | */ | |
e3252464 | 661 | static bool timer_fixup_free(void *addr, enum debug_obj_state state) |
c6f3a97f TG |
662 | { |
663 | struct timer_list *timer = addr; | |
664 | ||
665 | switch (state) { | |
666 | case ODEBUG_STATE_ACTIVE: | |
667 | del_timer_sync(timer); | |
668 | debug_object_free(timer, &timer_debug_descr); | |
e3252464 | 669 | return true; |
c6f3a97f | 670 | default: |
e3252464 | 671 | return false; |
c6f3a97f TG |
672 | } |
673 | } | |
674 | ||
dc4218bd CC |
675 | /* |
676 | * fixup_assert_init is called when: | |
677 | * - an untracked/uninit-ed object is found | |
678 | */ | |
e3252464 | 679 | static bool timer_fixup_assert_init(void *addr, enum debug_obj_state state) |
dc4218bd CC |
680 | { |
681 | struct timer_list *timer = addr; | |
682 | ||
683 | switch (state) { | |
684 | case ODEBUG_STATE_NOTAVAILABLE: | |
ba16490e | 685 | timer_setup(timer, stub_timer, 0); |
b9fdac7f | 686 | return true; |
dc4218bd | 687 | default: |
e3252464 | 688 | return false; |
dc4218bd CC |
689 | } |
690 | } | |
691 | ||
c6f3a97f | 692 | static struct debug_obj_descr timer_debug_descr = { |
dc4218bd CC |
693 | .name = "timer_list", |
694 | .debug_hint = timer_debug_hint, | |
b9fdac7f | 695 | .is_static_object = timer_is_static_object, |
dc4218bd CC |
696 | .fixup_init = timer_fixup_init, |
697 | .fixup_activate = timer_fixup_activate, | |
698 | .fixup_free = timer_fixup_free, | |
699 | .fixup_assert_init = timer_fixup_assert_init, | |
c6f3a97f TG |
700 | }; |
701 | ||
702 | static inline void debug_timer_init(struct timer_list *timer) | |
703 | { | |
704 | debug_object_init(timer, &timer_debug_descr); | |
705 | } | |
706 | ||
707 | static inline void debug_timer_activate(struct timer_list *timer) | |
708 | { | |
709 | debug_object_activate(timer, &timer_debug_descr); | |
710 | } | |
711 | ||
712 | static inline void debug_timer_deactivate(struct timer_list *timer) | |
713 | { | |
714 | debug_object_deactivate(timer, &timer_debug_descr); | |
715 | } | |
716 | ||
717 | static inline void debug_timer_free(struct timer_list *timer) | |
718 | { | |
719 | debug_object_free(timer, &timer_debug_descr); | |
720 | } | |
721 | ||
dc4218bd CC |
722 | static inline void debug_timer_assert_init(struct timer_list *timer) |
723 | { | |
724 | debug_object_assert_init(timer, &timer_debug_descr); | |
725 | } | |
726 | ||
188665b2 KC |
727 | static void do_init_timer(struct timer_list *timer, |
728 | void (*func)(struct timer_list *), | |
729 | unsigned int flags, | |
fc683995 | 730 | const char *name, struct lock_class_key *key); |
c6f3a97f | 731 | |
188665b2 KC |
732 | void init_timer_on_stack_key(struct timer_list *timer, |
733 | void (*func)(struct timer_list *), | |
734 | unsigned int flags, | |
fc683995 | 735 | const char *name, struct lock_class_key *key) |
c6f3a97f TG |
736 | { |
737 | debug_object_init_on_stack(timer, &timer_debug_descr); | |
188665b2 | 738 | do_init_timer(timer, func, flags, name, key); |
c6f3a97f | 739 | } |
6f2b9b9a | 740 | EXPORT_SYMBOL_GPL(init_timer_on_stack_key); |
c6f3a97f TG |
741 | |
742 | void destroy_timer_on_stack(struct timer_list *timer) | |
743 | { | |
744 | debug_object_free(timer, &timer_debug_descr); | |
745 | } | |
746 | EXPORT_SYMBOL_GPL(destroy_timer_on_stack); | |
747 | ||
748 | #else | |
749 | static inline void debug_timer_init(struct timer_list *timer) { } | |
750 | static inline void debug_timer_activate(struct timer_list *timer) { } | |
751 | static inline void debug_timer_deactivate(struct timer_list *timer) { } | |
dc4218bd | 752 | static inline void debug_timer_assert_init(struct timer_list *timer) { } |
c6f3a97f TG |
753 | #endif |
754 | ||
2b022e3d XG |
755 | static inline void debug_init(struct timer_list *timer) |
756 | { | |
757 | debug_timer_init(timer); | |
758 | trace_timer_init(timer); | |
759 | } | |
760 | ||
761 | static inline void | |
762 | debug_activate(struct timer_list *timer, unsigned long expires) | |
763 | { | |
764 | debug_timer_activate(timer); | |
0eeda71b | 765 | trace_timer_start(timer, expires, timer->flags); |
2b022e3d XG |
766 | } |
767 | ||
768 | static inline void debug_deactivate(struct timer_list *timer) | |
769 | { | |
770 | debug_timer_deactivate(timer); | |
771 | trace_timer_cancel(timer); | |
772 | } | |
773 | ||
dc4218bd CC |
774 | static inline void debug_assert_init(struct timer_list *timer) |
775 | { | |
776 | debug_timer_assert_init(timer); | |
777 | } | |
778 | ||
188665b2 KC |
779 | static void do_init_timer(struct timer_list *timer, |
780 | void (*func)(struct timer_list *), | |
781 | unsigned int flags, | |
fc683995 | 782 | const char *name, struct lock_class_key *key) |
55c888d6 | 783 | { |
1dabbcec | 784 | timer->entry.pprev = NULL; |
188665b2 | 785 | timer->function = func; |
0eeda71b | 786 | timer->flags = flags | raw_smp_processor_id(); |
6f2b9b9a | 787 | lockdep_init_map(&timer->lockdep_map, name, key, 0); |
55c888d6 | 788 | } |
c6f3a97f TG |
789 | |
790 | /** | |
633fe795 | 791 | * init_timer_key - initialize a timer |
c6f3a97f | 792 | * @timer: the timer to be initialized |
188665b2 | 793 | * @func: timer callback function |
fc683995 | 794 | * @flags: timer flags |
633fe795 RD |
795 | * @name: name of the timer |
796 | * @key: lockdep class key of the fake lock used for tracking timer | |
797 | * sync lock dependencies | |
c6f3a97f | 798 | * |
633fe795 | 799 | * init_timer_key() must be done to a timer prior calling *any* of the |
c6f3a97f TG |
800 | * other timer functions. |
801 | */ | |
188665b2 KC |
802 | void init_timer_key(struct timer_list *timer, |
803 | void (*func)(struct timer_list *), unsigned int flags, | |
fc683995 | 804 | const char *name, struct lock_class_key *key) |
c6f3a97f | 805 | { |
2b022e3d | 806 | debug_init(timer); |
188665b2 | 807 | do_init_timer(timer, func, flags, name, key); |
c6f3a97f | 808 | } |
6f2b9b9a | 809 | EXPORT_SYMBOL(init_timer_key); |
55c888d6 | 810 | |
ec44bc7a | 811 | static inline void detach_timer(struct timer_list *timer, bool clear_pending) |
55c888d6 | 812 | { |
1dabbcec | 813 | struct hlist_node *entry = &timer->entry; |
55c888d6 | 814 | |
2b022e3d | 815 | debug_deactivate(timer); |
c6f3a97f | 816 | |
1dabbcec | 817 | __hlist_del(entry); |
55c888d6 | 818 | if (clear_pending) |
1dabbcec TG |
819 | entry->pprev = NULL; |
820 | entry->next = LIST_POISON2; | |
55c888d6 ON |
821 | } |
822 | ||
494af3ed | 823 | static int detach_if_pending(struct timer_list *timer, struct timer_base *base, |
ec44bc7a TG |
824 | bool clear_pending) |
825 | { | |
500462a9 TG |
826 | unsigned idx = timer_get_idx(timer); |
827 | ||
ec44bc7a TG |
828 | if (!timer_pending(timer)) |
829 | return 0; | |
830 | ||
500462a9 TG |
831 | if (hlist_is_singular_node(&timer->entry, base->vectors + idx)) |
832 | __clear_bit(idx, base->pending_map); | |
833 | ||
ec44bc7a | 834 | detach_timer(timer, clear_pending); |
ec44bc7a TG |
835 | return 1; |
836 | } | |
837 | ||
500462a9 TG |
838 | static inline struct timer_base *get_timer_cpu_base(u32 tflags, u32 cpu) |
839 | { | |
840 | struct timer_base *base = per_cpu_ptr(&timer_bases[BASE_STD], cpu); | |
841 | ||
842 | /* | |
ced6d5c1 AMG |
843 | * If the timer is deferrable and NO_HZ_COMMON is set then we need |
844 | * to use the deferrable base. | |
500462a9 | 845 | */ |
ced6d5c1 | 846 | if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE)) |
500462a9 TG |
847 | base = per_cpu_ptr(&timer_bases[BASE_DEF], cpu); |
848 | return base; | |
849 | } | |
850 | ||
851 | static inline struct timer_base *get_timer_this_cpu_base(u32 tflags) | |
852 | { | |
853 | struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]); | |
854 | ||
855 | /* | |
ced6d5c1 AMG |
856 | * If the timer is deferrable and NO_HZ_COMMON is set then we need |
857 | * to use the deferrable base. | |
500462a9 | 858 | */ |
ced6d5c1 | 859 | if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE)) |
500462a9 TG |
860 | base = this_cpu_ptr(&timer_bases[BASE_DEF]); |
861 | return base; | |
862 | } | |
863 | ||
864 | static inline struct timer_base *get_timer_base(u32 tflags) | |
865 | { | |
866 | return get_timer_cpu_base(tflags, tflags & TIMER_CPUMASK); | |
867 | } | |
868 | ||
a683f390 | 869 | static inline struct timer_base * |
6bad6bcc | 870 | get_target_base(struct timer_base *base, unsigned tflags) |
500462a9 | 871 | { |
ae67bada TG |
872 | #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON) |
873 | if (static_branch_likely(&timers_migration_enabled) && | |
874 | !(tflags & TIMER_PINNED)) | |
875 | return get_timer_cpu_base(tflags, get_nohz_timer_target()); | |
500462a9 | 876 | #endif |
ae67bada | 877 | return get_timer_this_cpu_base(tflags); |
500462a9 TG |
878 | } |
879 | ||
a683f390 TG |
880 | static inline void forward_timer_base(struct timer_base *base) |
881 | { | |
ae67bada | 882 | #ifdef CONFIG_NO_HZ_COMMON |
2fe59f50 | 883 | unsigned long jnow; |
6bad6bcc | 884 | |
a683f390 | 885 | /* |
2fe59f50 NP |
886 | * We only forward the base when we are idle or have just come out of |
887 | * idle (must_forward_clk logic), and have a delta between base clock | |
888 | * and jiffies. In the common case, run_timers will take care of it. | |
a683f390 | 889 | */ |
2fe59f50 NP |
890 | if (likely(!base->must_forward_clk)) |
891 | return; | |
892 | ||
893 | jnow = READ_ONCE(jiffies); | |
894 | base->must_forward_clk = base->is_idle; | |
895 | if ((long)(jnow - base->clk) < 2) | |
a683f390 TG |
896 | return; |
897 | ||
898 | /* | |
899 | * If the next expiry value is > jiffies, then we fast forward to | |
900 | * jiffies otherwise we forward to the next expiry value. | |
901 | */ | |
6bad6bcc TG |
902 | if (time_after(base->next_expiry, jnow)) |
903 | base->clk = jnow; | |
a683f390 TG |
904 | else |
905 | base->clk = base->next_expiry; | |
a683f390 | 906 | #endif |
ae67bada | 907 | } |
a683f390 | 908 | |
a683f390 | 909 | |
55c888d6 | 910 | /* |
500462a9 TG |
911 | * We are using hashed locking: Holding per_cpu(timer_bases[x]).lock means |
912 | * that all timers which are tied to this base are locked, and the base itself | |
913 | * is locked too. | |
55c888d6 ON |
914 | * |
915 | * So __run_timers/migrate_timers can safely modify all timers which could | |
500462a9 | 916 | * be found in the base->vectors array. |
55c888d6 | 917 | * |
500462a9 TG |
918 | * When a timer is migrating then the TIMER_MIGRATING flag is set and we need |
919 | * to wait until the migration is done. | |
55c888d6 | 920 | */ |
494af3ed | 921 | static struct timer_base *lock_timer_base(struct timer_list *timer, |
500462a9 | 922 | unsigned long *flags) |
89e7e374 | 923 | __acquires(timer->base->lock) |
55c888d6 | 924 | { |
55c888d6 | 925 | for (;;) { |
494af3ed | 926 | struct timer_base *base; |
b831275a TG |
927 | u32 tf; |
928 | ||
929 | /* | |
930 | * We need to use READ_ONCE() here, otherwise the compiler | |
931 | * might re-read @tf between the check for TIMER_MIGRATING | |
932 | * and spin_lock(). | |
933 | */ | |
934 | tf = READ_ONCE(timer->flags); | |
0eeda71b TG |
935 | |
936 | if (!(tf & TIMER_MIGRATING)) { | |
500462a9 | 937 | base = get_timer_base(tf); |
2287d866 | 938 | raw_spin_lock_irqsave(&base->lock, *flags); |
0eeda71b | 939 | if (timer->flags == tf) |
55c888d6 | 940 | return base; |
2287d866 | 941 | raw_spin_unlock_irqrestore(&base->lock, *flags); |
55c888d6 ON |
942 | } |
943 | cpu_relax(); | |
944 | } | |
945 | } | |
946 | ||
b24591e2 DH |
947 | #define MOD_TIMER_PENDING_ONLY 0x01 |
948 | #define MOD_TIMER_REDUCE 0x02 | |
949 | ||
74019224 | 950 | static inline int |
b24591e2 | 951 | __mod_timer(struct timer_list *timer, unsigned long expires, unsigned int options) |
1da177e4 | 952 | { |
494af3ed | 953 | struct timer_base *base, *new_base; |
f00c0afd AMG |
954 | unsigned int idx = UINT_MAX; |
955 | unsigned long clk = 0, flags; | |
bc7a34b8 | 956 | int ret = 0; |
1da177e4 | 957 | |
4da9152a TG |
958 | BUG_ON(!timer->function); |
959 | ||
500462a9 | 960 | /* |
f00c0afd AMG |
961 | * This is a common optimization triggered by the networking code - if |
962 | * the timer is re-modified to have the same timeout or ends up in the | |
963 | * same array bucket then just return: | |
500462a9 TG |
964 | */ |
965 | if (timer_pending(timer)) { | |
2fe59f50 NP |
966 | /* |
967 | * The downside of this optimization is that it can result in | |
968 | * larger granularity than you would get from adding a new | |
969 | * timer with this expiry. | |
970 | */ | |
b24591e2 DH |
971 | long diff = timer->expires - expires; |
972 | ||
973 | if (!diff) | |
974 | return 1; | |
975 | if (options & MOD_TIMER_REDUCE && diff <= 0) | |
500462a9 | 976 | return 1; |
4da9152a | 977 | |
f00c0afd | 978 | /* |
4da9152a TG |
979 | * We lock timer base and calculate the bucket index right |
980 | * here. If the timer ends up in the same bucket, then we | |
981 | * just update the expiry time and avoid the whole | |
982 | * dequeue/enqueue dance. | |
f00c0afd | 983 | */ |
4da9152a | 984 | base = lock_timer_base(timer, &flags); |
2fe59f50 | 985 | forward_timer_base(base); |
f00c0afd | 986 | |
b24591e2 DH |
987 | if (timer_pending(timer) && (options & MOD_TIMER_REDUCE) && |
988 | time_before_eq(timer->expires, expires)) { | |
989 | ret = 1; | |
990 | goto out_unlock; | |
991 | } | |
992 | ||
4da9152a | 993 | clk = base->clk; |
f00c0afd AMG |
994 | idx = calc_wheel_index(expires, clk); |
995 | ||
996 | /* | |
997 | * Retrieve and compare the array index of the pending | |
998 | * timer. If it matches set the expiry to the new value so a | |
999 | * subsequent call will exit in the expires check above. | |
1000 | */ | |
1001 | if (idx == timer_get_idx(timer)) { | |
b24591e2 DH |
1002 | if (!(options & MOD_TIMER_REDUCE)) |
1003 | timer->expires = expires; | |
1004 | else if (time_after(timer->expires, expires)) | |
1005 | timer->expires = expires; | |
4da9152a TG |
1006 | ret = 1; |
1007 | goto out_unlock; | |
f00c0afd | 1008 | } |
4da9152a TG |
1009 | } else { |
1010 | base = lock_timer_base(timer, &flags); | |
2fe59f50 | 1011 | forward_timer_base(base); |
500462a9 TG |
1012 | } |
1013 | ||
ec44bc7a | 1014 | ret = detach_if_pending(timer, base, false); |
b24591e2 | 1015 | if (!ret && (options & MOD_TIMER_PENDING_ONLY)) |
ec44bc7a | 1016 | goto out_unlock; |
55c888d6 | 1017 | |
500462a9 | 1018 | new_base = get_target_base(base, timer->flags); |
eea08f32 | 1019 | |
3691c519 | 1020 | if (base != new_base) { |
1da177e4 | 1021 | /* |
500462a9 | 1022 | * We are trying to schedule the timer on the new base. |
55c888d6 ON |
1023 | * However we can't change timer's base while it is running, |
1024 | * otherwise del_timer_sync() can't detect that the timer's | |
500462a9 TG |
1025 | * handler yet has not finished. This also guarantees that the |
1026 | * timer is serialized wrt itself. | |
1da177e4 | 1027 | */ |
a2c348fe | 1028 | if (likely(base->running_timer != timer)) { |
55c888d6 | 1029 | /* See the comment in lock_timer_base() */ |
0eeda71b TG |
1030 | timer->flags |= TIMER_MIGRATING; |
1031 | ||
2287d866 | 1032 | raw_spin_unlock(&base->lock); |
a2c348fe | 1033 | base = new_base; |
2287d866 | 1034 | raw_spin_lock(&base->lock); |
d0023a14 ED |
1035 | WRITE_ONCE(timer->flags, |
1036 | (timer->flags & ~TIMER_BASEMASK) | base->cpu); | |
2fe59f50 | 1037 | forward_timer_base(base); |
1da177e4 LT |
1038 | } |
1039 | } | |
1040 | ||
fd45bb77 TG |
1041 | debug_activate(timer, expires); |
1042 | ||
1da177e4 | 1043 | timer->expires = expires; |
f00c0afd AMG |
1044 | /* |
1045 | * If 'idx' was calculated above and the base time did not advance | |
4da9152a TG |
1046 | * between calculating 'idx' and possibly switching the base, only |
1047 | * enqueue_timer() and trigger_dyntick_cpu() is required. Otherwise | |
1048 | * we need to (re)calculate the wheel index via | |
1049 | * internal_add_timer(). | |
f00c0afd AMG |
1050 | */ |
1051 | if (idx != UINT_MAX && clk == base->clk) { | |
1052 | enqueue_timer(base, timer, idx); | |
1053 | trigger_dyntick_cpu(base, timer); | |
1054 | } else { | |
1055 | internal_add_timer(base, timer); | |
1056 | } | |
74019224 IM |
1057 | |
1058 | out_unlock: | |
2287d866 | 1059 | raw_spin_unlock_irqrestore(&base->lock, flags); |
1da177e4 LT |
1060 | |
1061 | return ret; | |
1062 | } | |
1063 | ||
2aae4a10 | 1064 | /** |
74019224 IM |
1065 | * mod_timer_pending - modify a pending timer's timeout |
1066 | * @timer: the pending timer to be modified | |
1067 | * @expires: new timeout in jiffies | |
1da177e4 | 1068 | * |
74019224 IM |
1069 | * mod_timer_pending() is the same for pending timers as mod_timer(), |
1070 | * but will not re-activate and modify already deleted timers. | |
1071 | * | |
1072 | * It is useful for unserialized use of timers. | |
1da177e4 | 1073 | */ |
74019224 | 1074 | int mod_timer_pending(struct timer_list *timer, unsigned long expires) |
1da177e4 | 1075 | { |
b24591e2 | 1076 | return __mod_timer(timer, expires, MOD_TIMER_PENDING_ONLY); |
1da177e4 | 1077 | } |
74019224 | 1078 | EXPORT_SYMBOL(mod_timer_pending); |
1da177e4 | 1079 | |
2aae4a10 | 1080 | /** |
1da177e4 LT |
1081 | * mod_timer - modify a timer's timeout |
1082 | * @timer: the timer to be modified | |
2aae4a10 | 1083 | * @expires: new timeout in jiffies |
1da177e4 | 1084 | * |
72fd4a35 | 1085 | * mod_timer() is a more efficient way to update the expire field of an |
1da177e4 LT |
1086 | * active timer (if the timer is inactive it will be activated) |
1087 | * | |
1088 | * mod_timer(timer, expires) is equivalent to: | |
1089 | * | |
1090 | * del_timer(timer); timer->expires = expires; add_timer(timer); | |
1091 | * | |
1092 | * Note that if there are multiple unserialized concurrent users of the | |
1093 | * same timer, then mod_timer() is the only safe way to modify the timeout, | |
1094 | * since add_timer() cannot modify an already running timer. | |
1095 | * | |
1096 | * The function returns whether it has modified a pending timer or not. | |
1097 | * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an | |
1098 | * active timer returns 1.) | |
1099 | */ | |
1100 | int mod_timer(struct timer_list *timer, unsigned long expires) | |
1101 | { | |
b24591e2 | 1102 | return __mod_timer(timer, expires, 0); |
1da177e4 | 1103 | } |
1da177e4 LT |
1104 | EXPORT_SYMBOL(mod_timer); |
1105 | ||
b24591e2 DH |
1106 | /** |
1107 | * timer_reduce - Modify a timer's timeout if it would reduce the timeout | |
1108 | * @timer: The timer to be modified | |
1109 | * @expires: New timeout in jiffies | |
1110 | * | |
1111 | * timer_reduce() is very similar to mod_timer(), except that it will only | |
1112 | * modify a running timer if that would reduce the expiration time (it will | |
1113 | * start a timer that isn't running). | |
1114 | */ | |
1115 | int timer_reduce(struct timer_list *timer, unsigned long expires) | |
1116 | { | |
1117 | return __mod_timer(timer, expires, MOD_TIMER_REDUCE); | |
1118 | } | |
1119 | EXPORT_SYMBOL(timer_reduce); | |
1120 | ||
74019224 IM |
1121 | /** |
1122 | * add_timer - start a timer | |
1123 | * @timer: the timer to be added | |
1124 | * | |
c1eba5bc | 1125 | * The kernel will do a ->function(@timer) callback from the |
74019224 IM |
1126 | * timer interrupt at the ->expires point in the future. The |
1127 | * current time is 'jiffies'. | |
1128 | * | |
c1eba5bc KC |
1129 | * The timer's ->expires, ->function fields must be set prior calling this |
1130 | * function. | |
74019224 IM |
1131 | * |
1132 | * Timers with an ->expires field in the past will be executed in the next | |
1133 | * timer tick. | |
1134 | */ | |
1135 | void add_timer(struct timer_list *timer) | |
1136 | { | |
1137 | BUG_ON(timer_pending(timer)); | |
1138 | mod_timer(timer, timer->expires); | |
1139 | } | |
1140 | EXPORT_SYMBOL(add_timer); | |
1141 | ||
1142 | /** | |
1143 | * add_timer_on - start a timer on a particular CPU | |
1144 | * @timer: the timer to be added | |
1145 | * @cpu: the CPU to start it on | |
1146 | * | |
1147 | * This is not very scalable on SMP. Double adds are not possible. | |
1148 | */ | |
1149 | void add_timer_on(struct timer_list *timer, int cpu) | |
1150 | { | |
500462a9 | 1151 | struct timer_base *new_base, *base; |
74019224 IM |
1152 | unsigned long flags; |
1153 | ||
74019224 | 1154 | BUG_ON(timer_pending(timer) || !timer->function); |
22b886dd | 1155 | |
500462a9 TG |
1156 | new_base = get_timer_cpu_base(timer->flags, cpu); |
1157 | ||
22b886dd TH |
1158 | /* |
1159 | * If @timer was on a different CPU, it should be migrated with the | |
1160 | * old base locked to prevent other operations proceeding with the | |
1161 | * wrong base locked. See lock_timer_base(). | |
1162 | */ | |
1163 | base = lock_timer_base(timer, &flags); | |
1164 | if (base != new_base) { | |
1165 | timer->flags |= TIMER_MIGRATING; | |
1166 | ||
2287d866 | 1167 | raw_spin_unlock(&base->lock); |
22b886dd | 1168 | base = new_base; |
2287d866 | 1169 | raw_spin_lock(&base->lock); |
22b886dd TH |
1170 | WRITE_ONCE(timer->flags, |
1171 | (timer->flags & ~TIMER_BASEMASK) | cpu); | |
1172 | } | |
2fe59f50 | 1173 | forward_timer_base(base); |
22b886dd | 1174 | |
2b022e3d | 1175 | debug_activate(timer, timer->expires); |
74019224 | 1176 | internal_add_timer(base, timer); |
2287d866 | 1177 | raw_spin_unlock_irqrestore(&base->lock, flags); |
74019224 | 1178 | } |
a9862e05 | 1179 | EXPORT_SYMBOL_GPL(add_timer_on); |
74019224 | 1180 | |
2aae4a10 | 1181 | /** |
0ba42a59 | 1182 | * del_timer - deactivate a timer. |
1da177e4 LT |
1183 | * @timer: the timer to be deactivated |
1184 | * | |
1185 | * del_timer() deactivates a timer - this works on both active and inactive | |
1186 | * timers. | |
1187 | * | |
1188 | * The function returns whether it has deactivated a pending timer or not. | |
1189 | * (ie. del_timer() of an inactive timer returns 0, del_timer() of an | |
1190 | * active timer returns 1.) | |
1191 | */ | |
1192 | int del_timer(struct timer_list *timer) | |
1193 | { | |
494af3ed | 1194 | struct timer_base *base; |
1da177e4 | 1195 | unsigned long flags; |
55c888d6 | 1196 | int ret = 0; |
1da177e4 | 1197 | |
dc4218bd CC |
1198 | debug_assert_init(timer); |
1199 | ||
55c888d6 ON |
1200 | if (timer_pending(timer)) { |
1201 | base = lock_timer_base(timer, &flags); | |
ec44bc7a | 1202 | ret = detach_if_pending(timer, base, true); |
2287d866 | 1203 | raw_spin_unlock_irqrestore(&base->lock, flags); |
1da177e4 | 1204 | } |
1da177e4 | 1205 | |
55c888d6 | 1206 | return ret; |
1da177e4 | 1207 | } |
1da177e4 LT |
1208 | EXPORT_SYMBOL(del_timer); |
1209 | ||
2aae4a10 REB |
1210 | /** |
1211 | * try_to_del_timer_sync - Try to deactivate a timer | |
d15bc69a | 1212 | * @timer: timer to delete |
2aae4a10 | 1213 | * |
fd450b73 ON |
1214 | * This function tries to deactivate a timer. Upon successful (ret >= 0) |
1215 | * exit the timer is not queued and the handler is not running on any CPU. | |
fd450b73 ON |
1216 | */ |
1217 | int try_to_del_timer_sync(struct timer_list *timer) | |
1218 | { | |
494af3ed | 1219 | struct timer_base *base; |
fd450b73 ON |
1220 | unsigned long flags; |
1221 | int ret = -1; | |
1222 | ||
dc4218bd CC |
1223 | debug_assert_init(timer); |
1224 | ||
fd450b73 ON |
1225 | base = lock_timer_base(timer, &flags); |
1226 | ||
dfb4357d | 1227 | if (base->running_timer != timer) |
ec44bc7a | 1228 | ret = detach_if_pending(timer, base, true); |
dfb4357d | 1229 | |
2287d866 | 1230 | raw_spin_unlock_irqrestore(&base->lock, flags); |
fd450b73 ON |
1231 | |
1232 | return ret; | |
1233 | } | |
e19dff1f DH |
1234 | EXPORT_SYMBOL(try_to_del_timer_sync); |
1235 | ||
6f1bc451 | 1236 | #ifdef CONFIG_SMP |
2aae4a10 | 1237 | /** |
1da177e4 LT |
1238 | * del_timer_sync - deactivate a timer and wait for the handler to finish. |
1239 | * @timer: the timer to be deactivated | |
1240 | * | |
1241 | * This function only differs from del_timer() on SMP: besides deactivating | |
1242 | * the timer it also makes sure the handler has finished executing on other | |
1243 | * CPUs. | |
1244 | * | |
72fd4a35 | 1245 | * Synchronization rules: Callers must prevent restarting of the timer, |
1da177e4 | 1246 | * otherwise this function is meaningless. It must not be called from |
c5f66e99 TH |
1247 | * interrupt contexts unless the timer is an irqsafe one. The caller must |
1248 | * not hold locks which would prevent completion of the timer's | |
1249 | * handler. The timer's handler must not call add_timer_on(). Upon exit the | |
1250 | * timer is not queued and the handler is not running on any CPU. | |
1da177e4 | 1251 | * |
c5f66e99 TH |
1252 | * Note: For !irqsafe timers, you must not hold locks that are held in |
1253 | * interrupt context while calling this function. Even if the lock has | |
bf9c96be | 1254 | * nothing to do with the timer in question. Here's why:: |
48228f7b SR |
1255 | * |
1256 | * CPU0 CPU1 | |
1257 | * ---- ---- | |
bf9c96be MCC |
1258 | * <SOFTIRQ> |
1259 | * call_timer_fn(); | |
1260 | * base->running_timer = mytimer; | |
1261 | * spin_lock_irq(somelock); | |
48228f7b SR |
1262 | * <IRQ> |
1263 | * spin_lock(somelock); | |
bf9c96be MCC |
1264 | * del_timer_sync(mytimer); |
1265 | * while (base->running_timer == mytimer); | |
48228f7b SR |
1266 | * |
1267 | * Now del_timer_sync() will never return and never release somelock. | |
1268 | * The interrupt on the other CPU is waiting to grab somelock but | |
1269 | * it has interrupted the softirq that CPU0 is waiting to finish. | |
1270 | * | |
1da177e4 | 1271 | * The function returns whether it has deactivated a pending timer or not. |
1da177e4 LT |
1272 | */ |
1273 | int del_timer_sync(struct timer_list *timer) | |
1274 | { | |
6f2b9b9a | 1275 | #ifdef CONFIG_LOCKDEP |
f266a511 PZ |
1276 | unsigned long flags; |
1277 | ||
48228f7b SR |
1278 | /* |
1279 | * If lockdep gives a backtrace here, please reference | |
1280 | * the synchronization rules above. | |
1281 | */ | |
7ff20792 | 1282 | local_irq_save(flags); |
6f2b9b9a JB |
1283 | lock_map_acquire(&timer->lockdep_map); |
1284 | lock_map_release(&timer->lockdep_map); | |
7ff20792 | 1285 | local_irq_restore(flags); |
6f2b9b9a | 1286 | #endif |
466bd303 YZ |
1287 | /* |
1288 | * don't use it in hardirq context, because it | |
1289 | * could lead to deadlock. | |
1290 | */ | |
0eeda71b | 1291 | WARN_ON(in_irq() && !(timer->flags & TIMER_IRQSAFE)); |
fd450b73 ON |
1292 | for (;;) { |
1293 | int ret = try_to_del_timer_sync(timer); | |
1294 | if (ret >= 0) | |
1295 | return ret; | |
a0009652 | 1296 | cpu_relax(); |
fd450b73 | 1297 | } |
1da177e4 | 1298 | } |
55c888d6 | 1299 | EXPORT_SYMBOL(del_timer_sync); |
1da177e4 LT |
1300 | #endif |
1301 | ||
354b46b1 | 1302 | static void call_timer_fn(struct timer_list *timer, void (*fn)(struct timer_list *)) |
576da126 | 1303 | { |
4a2b4b22 | 1304 | int count = preempt_count(); |
576da126 TG |
1305 | |
1306 | #ifdef CONFIG_LOCKDEP | |
1307 | /* | |
1308 | * It is permissible to free the timer from inside the | |
1309 | * function that is called from it, this we need to take into | |
1310 | * account for lockdep too. To avoid bogus "held lock freed" | |
1311 | * warnings as well as problems when looking into | |
1312 | * timer->lockdep_map, make a copy and use that here. | |
1313 | */ | |
4d82a1de PZ |
1314 | struct lockdep_map lockdep_map; |
1315 | ||
1316 | lockdep_copy_map(&lockdep_map, &timer->lockdep_map); | |
576da126 TG |
1317 | #endif |
1318 | /* | |
1319 | * Couple the lock chain with the lock chain at | |
1320 | * del_timer_sync() by acquiring the lock_map around the fn() | |
1321 | * call here and in del_timer_sync(). | |
1322 | */ | |
1323 | lock_map_acquire(&lockdep_map); | |
1324 | ||
1325 | trace_timer_expire_entry(timer); | |
354b46b1 | 1326 | fn(timer); |
576da126 TG |
1327 | trace_timer_expire_exit(timer); |
1328 | ||
1329 | lock_map_release(&lockdep_map); | |
1330 | ||
4a2b4b22 | 1331 | if (count != preempt_count()) { |
802702e0 | 1332 | WARN_ONCE(1, "timer: %pF preempt leak: %08x -> %08x\n", |
4a2b4b22 | 1333 | fn, count, preempt_count()); |
802702e0 TG |
1334 | /* |
1335 | * Restore the preempt count. That gives us a decent | |
1336 | * chance to survive and extract information. If the | |
1337 | * callback kept a lock held, bad luck, but not worse | |
1338 | * than the BUG() we had. | |
1339 | */ | |
4a2b4b22 | 1340 | preempt_count_set(count); |
576da126 TG |
1341 | } |
1342 | } | |
1343 | ||
500462a9 | 1344 | static void expire_timers(struct timer_base *base, struct hlist_head *head) |
1da177e4 | 1345 | { |
500462a9 TG |
1346 | while (!hlist_empty(head)) { |
1347 | struct timer_list *timer; | |
354b46b1 | 1348 | void (*fn)(struct timer_list *); |
1da177e4 | 1349 | |
500462a9 | 1350 | timer = hlist_entry(head->first, struct timer_list, entry); |
3bb475a3 | 1351 | |
500462a9 TG |
1352 | base->running_timer = timer; |
1353 | detach_timer(timer, true); | |
3bb475a3 | 1354 | |
500462a9 | 1355 | fn = timer->function; |
500462a9 TG |
1356 | |
1357 | if (timer->flags & TIMER_IRQSAFE) { | |
2287d866 | 1358 | raw_spin_unlock(&base->lock); |
c1eba5bc | 1359 | call_timer_fn(timer, fn); |
2287d866 | 1360 | raw_spin_lock(&base->lock); |
500462a9 | 1361 | } else { |
2287d866 | 1362 | raw_spin_unlock_irq(&base->lock); |
c1eba5bc | 1363 | call_timer_fn(timer, fn); |
2287d866 | 1364 | raw_spin_lock_irq(&base->lock); |
3bb475a3 | 1365 | } |
500462a9 TG |
1366 | } |
1367 | } | |
3bb475a3 | 1368 | |
23696838 AMG |
1369 | static int __collect_expired_timers(struct timer_base *base, |
1370 | struct hlist_head *heads) | |
500462a9 TG |
1371 | { |
1372 | unsigned long clk = base->clk; | |
1373 | struct hlist_head *vec; | |
1374 | int i, levels = 0; | |
1375 | unsigned int idx; | |
626ab0e6 | 1376 | |
500462a9 TG |
1377 | for (i = 0; i < LVL_DEPTH; i++) { |
1378 | idx = (clk & LVL_MASK) + i * LVL_SIZE; | |
1379 | ||
1380 | if (__test_and_clear_bit(idx, base->pending_map)) { | |
1381 | vec = base->vectors + idx; | |
1382 | hlist_move_list(vec, heads++); | |
1383 | levels++; | |
1da177e4 | 1384 | } |
500462a9 TG |
1385 | /* Is it time to look at the next level? */ |
1386 | if (clk & LVL_CLK_MASK) | |
1387 | break; | |
1388 | /* Shift clock for the next level granularity */ | |
1389 | clk >>= LVL_CLK_SHIFT; | |
1da177e4 | 1390 | } |
500462a9 | 1391 | return levels; |
1da177e4 LT |
1392 | } |
1393 | ||
3451d024 | 1394 | #ifdef CONFIG_NO_HZ_COMMON |
1da177e4 | 1395 | /* |
23696838 AMG |
1396 | * Find the next pending bucket of a level. Search from level start (@offset) |
1397 | * + @clk upwards and if nothing there, search from start of the level | |
1398 | * (@offset) up to @offset + clk. | |
1da177e4 | 1399 | */ |
500462a9 TG |
1400 | static int next_pending_bucket(struct timer_base *base, unsigned offset, |
1401 | unsigned clk) | |
1402 | { | |
1403 | unsigned pos, start = offset + clk; | |
1404 | unsigned end = offset + LVL_SIZE; | |
1405 | ||
1406 | pos = find_next_bit(base->pending_map, end, start); | |
1407 | if (pos < end) | |
1408 | return pos - start; | |
1409 | ||
1410 | pos = find_next_bit(base->pending_map, start, offset); | |
1411 | return pos < start ? pos + LVL_SIZE - start : -1; | |
1412 | } | |
1413 | ||
1414 | /* | |
23696838 AMG |
1415 | * Search the first expiring timer in the various clock levels. Caller must |
1416 | * hold base->lock. | |
1da177e4 | 1417 | */ |
494af3ed | 1418 | static unsigned long __next_timer_interrupt(struct timer_base *base) |
1da177e4 | 1419 | { |
500462a9 TG |
1420 | unsigned long clk, next, adj; |
1421 | unsigned lvl, offset = 0; | |
1422 | ||
500462a9 TG |
1423 | next = base->clk + NEXT_TIMER_MAX_DELTA; |
1424 | clk = base->clk; | |
1425 | for (lvl = 0; lvl < LVL_DEPTH; lvl++, offset += LVL_SIZE) { | |
1426 | int pos = next_pending_bucket(base, offset, clk & LVL_MASK); | |
1427 | ||
1428 | if (pos >= 0) { | |
1429 | unsigned long tmp = clk + (unsigned long) pos; | |
1430 | ||
1431 | tmp <<= LVL_SHIFT(lvl); | |
1432 | if (time_before(tmp, next)) | |
1433 | next = tmp; | |
1da177e4 | 1434 | } |
500462a9 TG |
1435 | /* |
1436 | * Clock for the next level. If the current level clock lower | |
1437 | * bits are zero, we look at the next level as is. If not we | |
1438 | * need to advance it by one because that's going to be the | |
1439 | * next expiring bucket in that level. base->clk is the next | |
1440 | * expiring jiffie. So in case of: | |
1441 | * | |
1442 | * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0 | |
1443 | * 0 0 0 0 0 0 | |
1444 | * | |
1445 | * we have to look at all levels @index 0. With | |
1446 | * | |
1447 | * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0 | |
1448 | * 0 0 0 0 0 2 | |
1449 | * | |
1450 | * LVL0 has the next expiring bucket @index 2. The upper | |
1451 | * levels have the next expiring bucket @index 1. | |
1452 | * | |
1453 | * In case that the propagation wraps the next level the same | |
1454 | * rules apply: | |
1455 | * | |
1456 | * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0 | |
1457 | * 0 0 0 0 F 2 | |
1458 | * | |
1459 | * So after looking at LVL0 we get: | |
1460 | * | |
1461 | * LVL5 LVL4 LVL3 LVL2 LVL1 | |
1462 | * 0 0 0 1 0 | |
1463 | * | |
1464 | * So no propagation from LVL1 to LVL2 because that happened | |
1465 | * with the add already, but then we need to propagate further | |
1466 | * from LVL2 to LVL3. | |
1467 | * | |
1468 | * So the simple check whether the lower bits of the current | |
1469 | * level are 0 or not is sufficient for all cases. | |
1470 | */ | |
1471 | adj = clk & LVL_CLK_MASK ? 1 : 0; | |
1472 | clk >>= LVL_CLK_SHIFT; | |
1473 | clk += adj; | |
1da177e4 | 1474 | } |
500462a9 | 1475 | return next; |
1cfd6849 | 1476 | } |
69239749 | 1477 | |
1cfd6849 TG |
1478 | /* |
1479 | * Check, if the next hrtimer event is before the next timer wheel | |
1480 | * event: | |
1481 | */ | |
c1ad348b | 1482 | static u64 cmp_next_hrtimer_event(u64 basem, u64 expires) |
1cfd6849 | 1483 | { |
c1ad348b | 1484 | u64 nextevt = hrtimer_get_next_event(); |
0662b713 | 1485 | |
9501b6cf | 1486 | /* |
c1ad348b TG |
1487 | * If high resolution timers are enabled |
1488 | * hrtimer_get_next_event() returns KTIME_MAX. | |
9501b6cf | 1489 | */ |
c1ad348b TG |
1490 | if (expires <= nextevt) |
1491 | return expires; | |
eaad084b TG |
1492 | |
1493 | /* | |
c1ad348b TG |
1494 | * If the next timer is already expired, return the tick base |
1495 | * time so the tick is fired immediately. | |
eaad084b | 1496 | */ |
c1ad348b TG |
1497 | if (nextevt <= basem) |
1498 | return basem; | |
eaad084b | 1499 | |
9501b6cf | 1500 | /* |
c1ad348b TG |
1501 | * Round up to the next jiffie. High resolution timers are |
1502 | * off, so the hrtimers are expired in the tick and we need to | |
1503 | * make sure that this tick really expires the timer to avoid | |
1504 | * a ping pong of the nohz stop code. | |
1505 | * | |
1506 | * Use DIV_ROUND_UP_ULL to prevent gcc calling __divdi3 | |
9501b6cf | 1507 | */ |
c1ad348b | 1508 | return DIV_ROUND_UP_ULL(nextevt, TICK_NSEC) * TICK_NSEC; |
1da177e4 | 1509 | } |
1cfd6849 TG |
1510 | |
1511 | /** | |
c1ad348b TG |
1512 | * get_next_timer_interrupt - return the time (clock mono) of the next timer |
1513 | * @basej: base time jiffies | |
1514 | * @basem: base time clock monotonic | |
1515 | * | |
1516 | * Returns the tick aligned clock monotonic time of the next pending | |
1517 | * timer or KTIME_MAX if no timer is pending. | |
1cfd6849 | 1518 | */ |
c1ad348b | 1519 | u64 get_next_timer_interrupt(unsigned long basej, u64 basem) |
1cfd6849 | 1520 | { |
500462a9 | 1521 | struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]); |
c1ad348b TG |
1522 | u64 expires = KTIME_MAX; |
1523 | unsigned long nextevt; | |
46c8f0b0 | 1524 | bool is_max_delta; |
1cfd6849 | 1525 | |
dbd87b5a HC |
1526 | /* |
1527 | * Pretend that there is no timer pending if the cpu is offline. | |
1528 | * Possible pending timers will be migrated later to an active cpu. | |
1529 | */ | |
1530 | if (cpu_is_offline(smp_processor_id())) | |
e40468a5 TG |
1531 | return expires; |
1532 | ||
2287d866 | 1533 | raw_spin_lock(&base->lock); |
500462a9 | 1534 | nextevt = __next_timer_interrupt(base); |
46c8f0b0 | 1535 | is_max_delta = (nextevt == base->clk + NEXT_TIMER_MAX_DELTA); |
a683f390 TG |
1536 | base->next_expiry = nextevt; |
1537 | /* | |
041ad7bc TG |
1538 | * We have a fresh next event. Check whether we can forward the |
1539 | * base. We can only do that when @basej is past base->clk | |
1540 | * otherwise we might rewind base->clk. | |
a683f390 | 1541 | */ |
041ad7bc TG |
1542 | if (time_after(basej, base->clk)) { |
1543 | if (time_after(nextevt, basej)) | |
1544 | base->clk = basej; | |
1545 | else if (time_after(nextevt, base->clk)) | |
1546 | base->clk = nextevt; | |
1547 | } | |
23696838 | 1548 | |
a683f390 | 1549 | if (time_before_eq(nextevt, basej)) { |
500462a9 | 1550 | expires = basem; |
a683f390 TG |
1551 | base->is_idle = false; |
1552 | } else { | |
46c8f0b0 | 1553 | if (!is_max_delta) |
34f41c03 | 1554 | expires = basem + (u64)(nextevt - basej) * TICK_NSEC; |
a683f390 | 1555 | /* |
2fe59f50 NP |
1556 | * If we expect to sleep more than a tick, mark the base idle. |
1557 | * Also the tick is stopped so any added timer must forward | |
1558 | * the base clk itself to keep granularity small. This idle | |
1559 | * logic is only maintained for the BASE_STD base, deferrable | |
1560 | * timers may still see large granularity skew (by design). | |
a683f390 | 1561 | */ |
2fe59f50 NP |
1562 | if ((expires - basem) > TICK_NSEC) { |
1563 | base->must_forward_clk = true; | |
a683f390 | 1564 | base->is_idle = true; |
2fe59f50 | 1565 | } |
e40468a5 | 1566 | } |
2287d866 | 1567 | raw_spin_unlock(&base->lock); |
1cfd6849 | 1568 | |
c1ad348b | 1569 | return cmp_next_hrtimer_event(basem, expires); |
1cfd6849 | 1570 | } |
23696838 | 1571 | |
a683f390 TG |
1572 | /** |
1573 | * timer_clear_idle - Clear the idle state of the timer base | |
1574 | * | |
1575 | * Called with interrupts disabled | |
1576 | */ | |
1577 | void timer_clear_idle(void) | |
1578 | { | |
1579 | struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]); | |
1580 | ||
1581 | /* | |
1582 | * We do this unlocked. The worst outcome is a remote enqueue sending | |
1583 | * a pointless IPI, but taking the lock would just make the window for | |
1584 | * sending the IPI a few instructions smaller for the cost of taking | |
1585 | * the lock in the exit from idle path. | |
1586 | */ | |
1587 | base->is_idle = false; | |
1588 | } | |
1589 | ||
23696838 AMG |
1590 | static int collect_expired_timers(struct timer_base *base, |
1591 | struct hlist_head *heads) | |
1592 | { | |
1593 | /* | |
1594 | * NOHZ optimization. After a long idle sleep we need to forward the | |
1595 | * base to current jiffies. Avoid a loop by searching the bitfield for | |
1596 | * the next expiring timer. | |
1597 | */ | |
1598 | if ((long)(jiffies - base->clk) > 2) { | |
1599 | unsigned long next = __next_timer_interrupt(base); | |
1600 | ||
1601 | /* | |
1602 | * If the next timer is ahead of time forward to current | |
a683f390 | 1603 | * jiffies, otherwise forward to the next expiry time: |
23696838 AMG |
1604 | */ |
1605 | if (time_after(next, jiffies)) { | |
c310ce4d ZD |
1606 | /* |
1607 | * The call site will increment base->clk and then | |
1608 | * terminate the expiry loop immediately. | |
1609 | */ | |
1610 | base->clk = jiffies; | |
23696838 AMG |
1611 | return 0; |
1612 | } | |
1613 | base->clk = next; | |
1614 | } | |
1615 | return __collect_expired_timers(base, heads); | |
1616 | } | |
1617 | #else | |
1618 | static inline int collect_expired_timers(struct timer_base *base, | |
1619 | struct hlist_head *heads) | |
1620 | { | |
1621 | return __collect_expired_timers(base, heads); | |
1622 | } | |
1da177e4 LT |
1623 | #endif |
1624 | ||
1da177e4 | 1625 | /* |
5b4db0c2 | 1626 | * Called from the timer interrupt handler to charge one tick to the current |
1da177e4 LT |
1627 | * process. user_tick is 1 if the tick is user time, 0 for system. |
1628 | */ | |
1629 | void update_process_times(int user_tick) | |
1630 | { | |
1631 | struct task_struct *p = current; | |
1da177e4 LT |
1632 | |
1633 | /* Note: this timer irq context must be accounted for as well. */ | |
fa13a5a1 | 1634 | account_process_tick(p, user_tick); |
1da177e4 | 1635 | run_local_timers(); |
c3377c2d | 1636 | rcu_check_callbacks(user_tick); |
e360adbe PZ |
1637 | #ifdef CONFIG_IRQ_WORK |
1638 | if (in_irq()) | |
76a33061 | 1639 | irq_work_tick(); |
e360adbe | 1640 | #endif |
1da177e4 | 1641 | scheduler_tick(); |
baa73d9e NP |
1642 | if (IS_ENABLED(CONFIG_POSIX_TIMERS)) |
1643 | run_posix_cpu_timers(p); | |
1da177e4 LT |
1644 | } |
1645 | ||
73420fea AMG |
1646 | /** |
1647 | * __run_timers - run all expired timers (if any) on this CPU. | |
1648 | * @base: the timer vector to be processed. | |
1649 | */ | |
1650 | static inline void __run_timers(struct timer_base *base) | |
1651 | { | |
1652 | struct hlist_head heads[LVL_DEPTH]; | |
1653 | int levels; | |
1654 | ||
1655 | if (!time_after_eq(jiffies, base->clk)) | |
1656 | return; | |
1657 | ||
2287d866 | 1658 | raw_spin_lock_irq(&base->lock); |
73420fea AMG |
1659 | |
1660 | while (time_after_eq(jiffies, base->clk)) { | |
1661 | ||
1662 | levels = collect_expired_timers(base, heads); | |
1663 | base->clk++; | |
1664 | ||
1665 | while (levels--) | |
1666 | expire_timers(base, heads + levels); | |
1667 | } | |
1668 | base->running_timer = NULL; | |
2287d866 | 1669 | raw_spin_unlock_irq(&base->lock); |
73420fea AMG |
1670 | } |
1671 | ||
1da177e4 LT |
1672 | /* |
1673 | * This function runs timers and the timer-tq in bottom half context. | |
1674 | */ | |
0766f788 | 1675 | static __latent_entropy void run_timer_softirq(struct softirq_action *h) |
1da177e4 | 1676 | { |
500462a9 | 1677 | struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]); |
1da177e4 | 1678 | |
2fe59f50 NP |
1679 | /* |
1680 | * must_forward_clk must be cleared before running timers so that any | |
1681 | * timer functions that call mod_timer will not try to forward the | |
1682 | * base. idle trcking / clock forwarding logic is only used with | |
1683 | * BASE_STD timers. | |
1684 | * | |
1685 | * The deferrable base does not do idle tracking at all, so we do | |
1686 | * not forward it. This can result in very large variations in | |
1687 | * granularity for deferrable timers, but they can be deferred for | |
1688 | * long periods due to idle. | |
1689 | */ | |
1690 | base->must_forward_clk = false; | |
1691 | ||
500462a9 | 1692 | __run_timers(base); |
ced6d5c1 | 1693 | if (IS_ENABLED(CONFIG_NO_HZ_COMMON)) |
500462a9 | 1694 | __run_timers(this_cpu_ptr(&timer_bases[BASE_DEF])); |
1da177e4 LT |
1695 | } |
1696 | ||
1697 | /* | |
1698 | * Called by the local, per-CPU timer interrupt on SMP. | |
1699 | */ | |
1700 | void run_local_timers(void) | |
1701 | { | |
4e85876a TG |
1702 | struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]); |
1703 | ||
d3d74453 | 1704 | hrtimer_run_queues(); |
4e85876a TG |
1705 | /* Raise the softirq only if required. */ |
1706 | if (time_before(jiffies, base->clk)) { | |
ed4bbf79 | 1707 | if (!IS_ENABLED(CONFIG_NO_HZ_COMMON)) |
4e85876a TG |
1708 | return; |
1709 | /* CPU is awake, so check the deferrable base. */ | |
1710 | base++; | |
1711 | if (time_before(jiffies, base->clk)) | |
1712 | return; | |
1713 | } | |
1da177e4 LT |
1714 | raise_softirq(TIMER_SOFTIRQ); |
1715 | } | |
1716 | ||
58e1177b KC |
1717 | /* |
1718 | * Since schedule_timeout()'s timer is defined on the stack, it must store | |
1719 | * the target task on the stack as well. | |
1720 | */ | |
1721 | struct process_timer { | |
1722 | struct timer_list timer; | |
1723 | struct task_struct *task; | |
1724 | }; | |
1725 | ||
1726 | static void process_timeout(struct timer_list *t) | |
1da177e4 | 1727 | { |
58e1177b KC |
1728 | struct process_timer *timeout = from_timer(timeout, t, timer); |
1729 | ||
1730 | wake_up_process(timeout->task); | |
1da177e4 LT |
1731 | } |
1732 | ||
1733 | /** | |
1734 | * schedule_timeout - sleep until timeout | |
1735 | * @timeout: timeout value in jiffies | |
1736 | * | |
1737 | * Make the current task sleep until @timeout jiffies have | |
1738 | * elapsed. The routine will return immediately unless | |
1739 | * the current task state has been set (see set_current_state()). | |
1740 | * | |
1741 | * You can set the task state as follows - | |
1742 | * | |
1743 | * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to | |
4b7e9cf9 DA |
1744 | * pass before the routine returns unless the current task is explicitly |
1745 | * woken up, (e.g. by wake_up_process())". | |
1da177e4 LT |
1746 | * |
1747 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
4b7e9cf9 DA |
1748 | * delivered to the current task or the current task is explicitly woken |
1749 | * up. | |
1da177e4 LT |
1750 | * |
1751 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1752 | * routine returns. | |
1753 | * | |
1754 | * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule | |
1755 | * the CPU away without a bound on the timeout. In this case the return | |
1756 | * value will be %MAX_SCHEDULE_TIMEOUT. | |
1757 | * | |
4b7e9cf9 DA |
1758 | * Returns 0 when the timer has expired otherwise the remaining time in |
1759 | * jiffies will be returned. In all cases the return value is guaranteed | |
1760 | * to be non-negative. | |
1da177e4 | 1761 | */ |
7ad5b3a5 | 1762 | signed long __sched schedule_timeout(signed long timeout) |
1da177e4 | 1763 | { |
58e1177b | 1764 | struct process_timer timer; |
1da177e4 LT |
1765 | unsigned long expire; |
1766 | ||
1767 | switch (timeout) | |
1768 | { | |
1769 | case MAX_SCHEDULE_TIMEOUT: | |
1770 | /* | |
1771 | * These two special cases are useful to be comfortable | |
1772 | * in the caller. Nothing more. We could take | |
1773 | * MAX_SCHEDULE_TIMEOUT from one of the negative value | |
1774 | * but I' d like to return a valid offset (>=0) to allow | |
1775 | * the caller to do everything it want with the retval. | |
1776 | */ | |
1777 | schedule(); | |
1778 | goto out; | |
1779 | default: | |
1780 | /* | |
1781 | * Another bit of PARANOID. Note that the retval will be | |
1782 | * 0 since no piece of kernel is supposed to do a check | |
1783 | * for a negative retval of schedule_timeout() (since it | |
1784 | * should never happens anyway). You just have the printk() | |
1785 | * that will tell you if something is gone wrong and where. | |
1786 | */ | |
5b149bcc | 1787 | if (timeout < 0) { |
1da177e4 | 1788 | printk(KERN_ERR "schedule_timeout: wrong timeout " |
5b149bcc AM |
1789 | "value %lx\n", timeout); |
1790 | dump_stack(); | |
1da177e4 LT |
1791 | current->state = TASK_RUNNING; |
1792 | goto out; | |
1793 | } | |
1794 | } | |
1795 | ||
1796 | expire = timeout + jiffies; | |
1797 | ||
58e1177b KC |
1798 | timer.task = current; |
1799 | timer_setup_on_stack(&timer.timer, process_timeout, 0); | |
b24591e2 | 1800 | __mod_timer(&timer.timer, expire, 0); |
1da177e4 | 1801 | schedule(); |
58e1177b | 1802 | del_singleshot_timer_sync(&timer.timer); |
1da177e4 | 1803 | |
c6f3a97f | 1804 | /* Remove the timer from the object tracker */ |
58e1177b | 1805 | destroy_timer_on_stack(&timer.timer); |
c6f3a97f | 1806 | |
1da177e4 LT |
1807 | timeout = expire - jiffies; |
1808 | ||
1809 | out: | |
1810 | return timeout < 0 ? 0 : timeout; | |
1811 | } | |
1da177e4 LT |
1812 | EXPORT_SYMBOL(schedule_timeout); |
1813 | ||
8a1c1757 AM |
1814 | /* |
1815 | * We can use __set_current_state() here because schedule_timeout() calls | |
1816 | * schedule() unconditionally. | |
1817 | */ | |
64ed93a2 NA |
1818 | signed long __sched schedule_timeout_interruptible(signed long timeout) |
1819 | { | |
a5a0d52c AM |
1820 | __set_current_state(TASK_INTERRUPTIBLE); |
1821 | return schedule_timeout(timeout); | |
64ed93a2 NA |
1822 | } |
1823 | EXPORT_SYMBOL(schedule_timeout_interruptible); | |
1824 | ||
294d5cc2 MW |
1825 | signed long __sched schedule_timeout_killable(signed long timeout) |
1826 | { | |
1827 | __set_current_state(TASK_KILLABLE); | |
1828 | return schedule_timeout(timeout); | |
1829 | } | |
1830 | EXPORT_SYMBOL(schedule_timeout_killable); | |
1831 | ||
64ed93a2 NA |
1832 | signed long __sched schedule_timeout_uninterruptible(signed long timeout) |
1833 | { | |
a5a0d52c AM |
1834 | __set_current_state(TASK_UNINTERRUPTIBLE); |
1835 | return schedule_timeout(timeout); | |
64ed93a2 NA |
1836 | } |
1837 | EXPORT_SYMBOL(schedule_timeout_uninterruptible); | |
1838 | ||
69b27baf AM |
1839 | /* |
1840 | * Like schedule_timeout_uninterruptible(), except this task will not contribute | |
1841 | * to load average. | |
1842 | */ | |
1843 | signed long __sched schedule_timeout_idle(signed long timeout) | |
1844 | { | |
1845 | __set_current_state(TASK_IDLE); | |
1846 | return schedule_timeout(timeout); | |
1847 | } | |
1848 | EXPORT_SYMBOL(schedule_timeout_idle); | |
1849 | ||
1da177e4 | 1850 | #ifdef CONFIG_HOTPLUG_CPU |
494af3ed | 1851 | static void migrate_timer_list(struct timer_base *new_base, struct hlist_head *head) |
1da177e4 LT |
1852 | { |
1853 | struct timer_list *timer; | |
0eeda71b | 1854 | int cpu = new_base->cpu; |
1da177e4 | 1855 | |
1dabbcec TG |
1856 | while (!hlist_empty(head)) { |
1857 | timer = hlist_entry(head->first, struct timer_list, entry); | |
ec44bc7a | 1858 | detach_timer(timer, false); |
0eeda71b | 1859 | timer->flags = (timer->flags & ~TIMER_BASEMASK) | cpu; |
1da177e4 | 1860 | internal_add_timer(new_base, timer); |
1da177e4 | 1861 | } |
1da177e4 LT |
1862 | } |
1863 | ||
26456f87 TG |
1864 | int timers_prepare_cpu(unsigned int cpu) |
1865 | { | |
1866 | struct timer_base *base; | |
1867 | int b; | |
1868 | ||
1869 | for (b = 0; b < NR_BASES; b++) { | |
1870 | base = per_cpu_ptr(&timer_bases[b], cpu); | |
1871 | base->clk = jiffies; | |
1872 | base->next_expiry = base->clk + NEXT_TIMER_MAX_DELTA; | |
1873 | base->is_idle = false; | |
1874 | base->must_forward_clk = true; | |
1875 | } | |
1876 | return 0; | |
1877 | } | |
1878 | ||
24f73b99 | 1879 | int timers_dead_cpu(unsigned int cpu) |
1da177e4 | 1880 | { |
494af3ed TG |
1881 | struct timer_base *old_base; |
1882 | struct timer_base *new_base; | |
500462a9 | 1883 | int b, i; |
1da177e4 LT |
1884 | |
1885 | BUG_ON(cpu_online(cpu)); | |
55c888d6 | 1886 | |
500462a9 TG |
1887 | for (b = 0; b < NR_BASES; b++) { |
1888 | old_base = per_cpu_ptr(&timer_bases[b], cpu); | |
1889 | new_base = get_cpu_ptr(&timer_bases[b]); | |
1890 | /* | |
1891 | * The caller is globally serialized and nobody else | |
1892 | * takes two locks at once, deadlock is not possible. | |
1893 | */ | |
2287d866 SAS |
1894 | raw_spin_lock_irq(&new_base->lock); |
1895 | raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING); | |
500462a9 | 1896 | |
c52232a4 LC |
1897 | /* |
1898 | * The current CPUs base clock might be stale. Update it | |
1899 | * before moving the timers over. | |
1900 | */ | |
1901 | forward_timer_base(new_base); | |
1902 | ||
500462a9 TG |
1903 | BUG_ON(old_base->running_timer); |
1904 | ||
1905 | for (i = 0; i < WHEEL_SIZE; i++) | |
1906 | migrate_timer_list(new_base, old_base->vectors + i); | |
8def9060 | 1907 | |
2287d866 SAS |
1908 | raw_spin_unlock(&old_base->lock); |
1909 | raw_spin_unlock_irq(&new_base->lock); | |
500462a9 TG |
1910 | put_cpu_ptr(&timer_bases); |
1911 | } | |
24f73b99 | 1912 | return 0; |
1da177e4 | 1913 | } |
1da177e4 | 1914 | |
3650b57f | 1915 | #endif /* CONFIG_HOTPLUG_CPU */ |
1da177e4 | 1916 | |
0eeda71b | 1917 | static void __init init_timer_cpu(int cpu) |
8def9060 | 1918 | { |
500462a9 TG |
1919 | struct timer_base *base; |
1920 | int i; | |
8def9060 | 1921 | |
500462a9 TG |
1922 | for (i = 0; i < NR_BASES; i++) { |
1923 | base = per_cpu_ptr(&timer_bases[i], cpu); | |
1924 | base->cpu = cpu; | |
2287d866 | 1925 | raw_spin_lock_init(&base->lock); |
500462a9 TG |
1926 | base->clk = jiffies; |
1927 | } | |
8def9060 VK |
1928 | } |
1929 | ||
1930 | static void __init init_timer_cpus(void) | |
1da177e4 | 1931 | { |
8def9060 VK |
1932 | int cpu; |
1933 | ||
0eeda71b TG |
1934 | for_each_possible_cpu(cpu) |
1935 | init_timer_cpu(cpu); | |
8def9060 | 1936 | } |
e52b1db3 | 1937 | |
8def9060 VK |
1938 | void __init init_timers(void) |
1939 | { | |
8def9060 | 1940 | init_timer_cpus(); |
962cf36c | 1941 | open_softirq(TIMER_SOFTIRQ, run_timer_softirq); |
1da177e4 LT |
1942 | } |
1943 | ||
1da177e4 LT |
1944 | /** |
1945 | * msleep - sleep safely even with waitqueue interruptions | |
1946 | * @msecs: Time in milliseconds to sleep for | |
1947 | */ | |
1948 | void msleep(unsigned int msecs) | |
1949 | { | |
1950 | unsigned long timeout = msecs_to_jiffies(msecs) + 1; | |
1951 | ||
75bcc8c5 NA |
1952 | while (timeout) |
1953 | timeout = schedule_timeout_uninterruptible(timeout); | |
1da177e4 LT |
1954 | } |
1955 | ||
1956 | EXPORT_SYMBOL(msleep); | |
1957 | ||
1958 | /** | |
96ec3efd | 1959 | * msleep_interruptible - sleep waiting for signals |
1da177e4 LT |
1960 | * @msecs: Time in milliseconds to sleep for |
1961 | */ | |
1962 | unsigned long msleep_interruptible(unsigned int msecs) | |
1963 | { | |
1964 | unsigned long timeout = msecs_to_jiffies(msecs) + 1; | |
1965 | ||
75bcc8c5 NA |
1966 | while (timeout && !signal_pending(current)) |
1967 | timeout = schedule_timeout_interruptible(timeout); | |
1da177e4 LT |
1968 | return jiffies_to_msecs(timeout); |
1969 | } | |
1970 | ||
1971 | EXPORT_SYMBOL(msleep_interruptible); | |
5e7f5a17 | 1972 | |
5e7f5a17 | 1973 | /** |
b5227d03 | 1974 | * usleep_range - Sleep for an approximate time |
5e7f5a17 PP |
1975 | * @min: Minimum time in usecs to sleep |
1976 | * @max: Maximum time in usecs to sleep | |
b5227d03 BH |
1977 | * |
1978 | * In non-atomic context where the exact wakeup time is flexible, use | |
1979 | * usleep_range() instead of udelay(). The sleep improves responsiveness | |
1980 | * by avoiding the CPU-hogging busy-wait of udelay(), and the range reduces | |
1981 | * power usage by allowing hrtimers to take advantage of an already- | |
1982 | * scheduled interrupt instead of scheduling a new one just for this sleep. | |
5e7f5a17 | 1983 | */ |
2ad5d327 | 1984 | void __sched usleep_range(unsigned long min, unsigned long max) |
5e7f5a17 | 1985 | { |
6c5e9059 DA |
1986 | ktime_t exp = ktime_add_us(ktime_get(), min); |
1987 | u64 delta = (u64)(max - min) * NSEC_PER_USEC; | |
1988 | ||
1989 | for (;;) { | |
1990 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
1991 | /* Do not return before the requested sleep time has elapsed */ | |
1992 | if (!schedule_hrtimeout_range(&exp, delta, HRTIMER_MODE_ABS)) | |
1993 | break; | |
1994 | } | |
5e7f5a17 PP |
1995 | } |
1996 | EXPORT_SYMBOL(usleep_range); |