Commit | Line | Data |
---|---|---|
35728b82 | 1 | // SPDX-License-Identifier: GPL-2.0 |
c0a31329 | 2 | /* |
3c8aa39d | 3 | * Copyright(C) 2005-2006, Thomas Gleixner <tglx@linutronix.de> |
79bf2bb3 | 4 | * Copyright(C) 2005-2007, Red Hat, Inc., Ingo Molnar |
54cdfdb4 | 5 | * Copyright(C) 2006-2007 Timesys Corp., Thomas Gleixner |
c0a31329 TG |
6 | * |
7 | * High-resolution kernel timers | |
8 | * | |
58c5fc2b TG |
9 | * In contrast to the low-resolution timeout API, aka timer wheel, |
10 | * hrtimers provide finer resolution and accuracy depending on system | |
11 | * configuration and capabilities. | |
c0a31329 TG |
12 | * |
13 | * Started by: Thomas Gleixner and Ingo Molnar | |
14 | * | |
15 | * Credits: | |
58c5fc2b | 16 | * Based on the original timer wheel code |
c0a31329 | 17 | * |
66188fae TG |
18 | * Help, testing, suggestions, bugfixes, improvements were |
19 | * provided by: | |
20 | * | |
21 | * George Anzinger, Andrew Morton, Steven Rostedt, Roman Zippel | |
22 | * et. al. | |
c0a31329 TG |
23 | */ |
24 | ||
25 | #include <linux/cpu.h> | |
9984de1a | 26 | #include <linux/export.h> |
c0a31329 TG |
27 | #include <linux/percpu.h> |
28 | #include <linux/hrtimer.h> | |
29 | #include <linux/notifier.h> | |
30 | #include <linux/syscalls.h> | |
31 | #include <linux/interrupt.h> | |
79bf2bb3 | 32 | #include <linux/tick.h> |
54cdfdb4 | 33 | #include <linux/err.h> |
237fc6e7 | 34 | #include <linux/debugobjects.h> |
174cd4b1 | 35 | #include <linux/sched/signal.h> |
cf4aebc2 | 36 | #include <linux/sched/sysctl.h> |
8bd75c77 | 37 | #include <linux/sched/rt.h> |
aab03e05 | 38 | #include <linux/sched/deadline.h> |
370c9135 | 39 | #include <linux/sched/nohz.h> |
b17b0153 | 40 | #include <linux/sched/debug.h> |
eea08f32 | 41 | #include <linux/timer.h> |
b0f8c44f | 42 | #include <linux/freezer.h> |
edbeda46 | 43 | #include <linux/compat.h> |
c0a31329 | 44 | |
7c0f6ba6 | 45 | #include <linux/uaccess.h> |
c0a31329 | 46 | |
c6a2a177 XG |
47 | #include <trace/events/timer.h> |
48 | ||
c1797baf | 49 | #include "tick-internal.h" |
8b094cd0 | 50 | |
c458b1d1 AMG |
51 | /* |
52 | * Masks for selecting the soft and hard context timers from | |
53 | * cpu_base->active | |
54 | */ | |
55 | #define MASK_SHIFT (HRTIMER_BASE_MONOTONIC_SOFT) | |
56 | #define HRTIMER_ACTIVE_HARD ((1U << MASK_SHIFT) - 1) | |
57 | #define HRTIMER_ACTIVE_SOFT (HRTIMER_ACTIVE_HARD << MASK_SHIFT) | |
58 | #define HRTIMER_ACTIVE_ALL (HRTIMER_ACTIVE_SOFT | HRTIMER_ACTIVE_HARD) | |
59 | ||
c0a31329 TG |
60 | /* |
61 | * The timer bases: | |
7978672c | 62 | * |
571af55a | 63 | * There are more clockids than hrtimer bases. Thus, we index |
e06383db JS |
64 | * into the timer bases by the hrtimer_base_type enum. When trying |
65 | * to reach a base using a clockid, hrtimer_clockid_to_base() | |
66 | * is used to convert from clockid to the proper hrtimer_base_type. | |
c0a31329 | 67 | */ |
54cdfdb4 | 68 | DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) = |
c0a31329 | 69 | { |
84cc8fd2 | 70 | .lock = __RAW_SPIN_LOCK_UNLOCKED(hrtimer_bases.lock), |
3c8aa39d | 71 | .clock_base = |
c0a31329 | 72 | { |
3c8aa39d | 73 | { |
ab8177bc TG |
74 | .index = HRTIMER_BASE_MONOTONIC, |
75 | .clockid = CLOCK_MONOTONIC, | |
3c8aa39d | 76 | .get_time = &ktime_get, |
3c8aa39d | 77 | }, |
68fa61c0 TG |
78 | { |
79 | .index = HRTIMER_BASE_REALTIME, | |
80 | .clockid = CLOCK_REALTIME, | |
81 | .get_time = &ktime_get_real, | |
68fa61c0 | 82 | }, |
a3ed0e43 TG |
83 | { |
84 | .index = HRTIMER_BASE_BOOTTIME, | |
85 | .clockid = CLOCK_BOOTTIME, | |
86 | .get_time = &ktime_get_boottime, | |
87 | }, | |
90adda98 JS |
88 | { |
89 | .index = HRTIMER_BASE_TAI, | |
90 | .clockid = CLOCK_TAI, | |
91 | .get_time = &ktime_get_clocktai, | |
90adda98 | 92 | }, |
98ecadd4 AMG |
93 | { |
94 | .index = HRTIMER_BASE_MONOTONIC_SOFT, | |
95 | .clockid = CLOCK_MONOTONIC, | |
96 | .get_time = &ktime_get, | |
97 | }, | |
98 | { | |
99 | .index = HRTIMER_BASE_REALTIME_SOFT, | |
100 | .clockid = CLOCK_REALTIME, | |
101 | .get_time = &ktime_get_real, | |
102 | }, | |
a3ed0e43 TG |
103 | { |
104 | .index = HRTIMER_BASE_BOOTTIME_SOFT, | |
105 | .clockid = CLOCK_BOOTTIME, | |
106 | .get_time = &ktime_get_boottime, | |
107 | }, | |
98ecadd4 AMG |
108 | { |
109 | .index = HRTIMER_BASE_TAI_SOFT, | |
110 | .clockid = CLOCK_TAI, | |
111 | .get_time = &ktime_get_clocktai, | |
112 | }, | |
3c8aa39d | 113 | } |
c0a31329 TG |
114 | }; |
115 | ||
942c3c5c | 116 | static const int hrtimer_clock_to_base_table[MAX_CLOCKS] = { |
336a9cde MZ |
117 | /* Make sure we catch unsupported clockids */ |
118 | [0 ... MAX_CLOCKS - 1] = HRTIMER_MAX_CLOCK_BASES, | |
119 | ||
ce31332d TG |
120 | [CLOCK_REALTIME] = HRTIMER_BASE_REALTIME, |
121 | [CLOCK_MONOTONIC] = HRTIMER_BASE_MONOTONIC, | |
a3ed0e43 | 122 | [CLOCK_BOOTTIME] = HRTIMER_BASE_BOOTTIME, |
90adda98 | 123 | [CLOCK_TAI] = HRTIMER_BASE_TAI, |
ce31332d | 124 | }; |
e06383db | 125 | |
c0a31329 TG |
126 | /* |
127 | * Functions and macros which are different for UP/SMP systems are kept in a | |
128 | * single place | |
129 | */ | |
130 | #ifdef CONFIG_SMP | |
131 | ||
887d9dc9 PZ |
132 | /* |
133 | * We require the migration_base for lock_hrtimer_base()/switch_hrtimer_base() | |
134 | * such that hrtimer_callback_running() can unconditionally dereference | |
135 | * timer->base->cpu_base | |
136 | */ | |
137 | static struct hrtimer_cpu_base migration_cpu_base = { | |
af5a06b5 AD |
138 | .clock_base = { { |
139 | .cpu_base = &migration_cpu_base, | |
140 | .seq = SEQCNT_RAW_SPINLOCK_ZERO(migration_cpu_base.seq, | |
141 | &migration_cpu_base.lock), | |
142 | }, }, | |
887d9dc9 PZ |
143 | }; |
144 | ||
145 | #define migration_base migration_cpu_base.clock_base[0] | |
146 | ||
5d2295f3 SAS |
147 | static inline bool is_migration_base(struct hrtimer_clock_base *base) |
148 | { | |
149 | return base == &migration_base; | |
150 | } | |
151 | ||
c0a31329 TG |
152 | /* |
153 | * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock | |
154 | * means that all timers which are tied to this base via timer->base are | |
155 | * locked, and the base itself is locked too. | |
156 | * | |
157 | * So __run_timers/migrate_timers can safely modify all timers which could | |
158 | * be found on the lists/queues. | |
159 | * | |
160 | * When the timer's base is locked, and the timer removed from list, it is | |
887d9dc9 PZ |
161 | * possible to set timer->base = &migration_base and drop the lock: the timer |
162 | * remains locked. | |
c0a31329 | 163 | */ |
3c8aa39d TG |
164 | static |
165 | struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer, | |
166 | unsigned long *flags) | |
ccaa4926 | 167 | __acquires(&timer->base->lock) |
c0a31329 | 168 | { |
3c8aa39d | 169 | struct hrtimer_clock_base *base; |
c0a31329 TG |
170 | |
171 | for (;;) { | |
ff229eee | 172 | base = READ_ONCE(timer->base); |
887d9dc9 | 173 | if (likely(base != &migration_base)) { |
ecb49d1a | 174 | raw_spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
175 | if (likely(base == timer->base)) |
176 | return base; | |
177 | /* The timer has migrated to another CPU: */ | |
ecb49d1a | 178 | raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags); |
c0a31329 TG |
179 | } |
180 | cpu_relax(); | |
181 | } | |
182 | } | |
183 | ||
6ff7041d | 184 | /* |
07a9a7ea AMG |
185 | * We do not migrate the timer when it is expiring before the next |
186 | * event on the target cpu. When high resolution is enabled, we cannot | |
187 | * reprogram the target cpu hardware and we would cause it to fire | |
188 | * late. To keep it simple, we handle the high resolution enabled and | |
189 | * disabled case similar. | |
6ff7041d TG |
190 | * |
191 | * Called with cpu_base->lock of target cpu held. | |
192 | */ | |
193 | static int | |
194 | hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base) | |
195 | { | |
6ff7041d TG |
196 | ktime_t expires; |
197 | ||
6ff7041d | 198 | expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset); |
2ac2dccc | 199 | return expires < new_base->cpu_base->expires_next; |
6ff7041d TG |
200 | } |
201 | ||
bc7a34b8 TG |
202 | static inline |
203 | struct hrtimer_cpu_base *get_target_base(struct hrtimer_cpu_base *base, | |
204 | int pinned) | |
205 | { | |
ae67bada TG |
206 | #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON) |
207 | if (static_branch_likely(&timers_migration_enabled) && !pinned) | |
208 | return &per_cpu(hrtimer_bases, get_nohz_timer_target()); | |
209 | #endif | |
662b3e19 | 210 | return base; |
bc7a34b8 | 211 | } |
bc7a34b8 | 212 | |
c0a31329 | 213 | /* |
b48362d8 FW |
214 | * We switch the timer base to a power-optimized selected CPU target, |
215 | * if: | |
216 | * - NO_HZ_COMMON is enabled | |
217 | * - timer migration is enabled | |
218 | * - the timer callback is not running | |
219 | * - the timer is not the first expiring timer on the new target | |
220 | * | |
221 | * If one of the above requirements is not fulfilled we move the timer | |
222 | * to the current CPU or leave it on the previously assigned CPU if | |
223 | * the timer callback is currently running. | |
c0a31329 | 224 | */ |
3c8aa39d | 225 | static inline struct hrtimer_clock_base * |
597d0275 AB |
226 | switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base, |
227 | int pinned) | |
c0a31329 | 228 | { |
b48362d8 | 229 | struct hrtimer_cpu_base *new_cpu_base, *this_cpu_base; |
3c8aa39d | 230 | struct hrtimer_clock_base *new_base; |
ab8177bc | 231 | int basenum = base->index; |
c0a31329 | 232 | |
b48362d8 FW |
233 | this_cpu_base = this_cpu_ptr(&hrtimer_bases); |
234 | new_cpu_base = get_target_base(this_cpu_base, pinned); | |
eea08f32 | 235 | again: |
e06383db | 236 | new_base = &new_cpu_base->clock_base[basenum]; |
c0a31329 TG |
237 | |
238 | if (base != new_base) { | |
239 | /* | |
6ff7041d | 240 | * We are trying to move timer to new_base. |
c0a31329 TG |
241 | * However we can't change timer's base while it is running, |
242 | * so we keep it on the same CPU. No hassle vs. reprogramming | |
243 | * the event source in the high resolution case. The softirq | |
244 | * code will take care of this when the timer function has | |
245 | * completed. There is no conflict as we hold the lock until | |
246 | * the timer is enqueued. | |
247 | */ | |
54cdfdb4 | 248 | if (unlikely(hrtimer_callback_running(timer))) |
c0a31329 TG |
249 | return base; |
250 | ||
887d9dc9 | 251 | /* See the comment in lock_hrtimer_base() */ |
ff229eee | 252 | WRITE_ONCE(timer->base, &migration_base); |
ecb49d1a TG |
253 | raw_spin_unlock(&base->cpu_base->lock); |
254 | raw_spin_lock(&new_base->cpu_base->lock); | |
eea08f32 | 255 | |
b48362d8 | 256 | if (new_cpu_base != this_cpu_base && |
bc7a34b8 | 257 | hrtimer_check_target(timer, new_base)) { |
ecb49d1a TG |
258 | raw_spin_unlock(&new_base->cpu_base->lock); |
259 | raw_spin_lock(&base->cpu_base->lock); | |
b48362d8 | 260 | new_cpu_base = this_cpu_base; |
ff229eee | 261 | WRITE_ONCE(timer->base, base); |
6ff7041d | 262 | goto again; |
eea08f32 | 263 | } |
ff229eee | 264 | WRITE_ONCE(timer->base, new_base); |
012a45e3 | 265 | } else { |
b48362d8 | 266 | if (new_cpu_base != this_cpu_base && |
bc7a34b8 | 267 | hrtimer_check_target(timer, new_base)) { |
b48362d8 | 268 | new_cpu_base = this_cpu_base; |
012a45e3 LM |
269 | goto again; |
270 | } | |
c0a31329 TG |
271 | } |
272 | return new_base; | |
273 | } | |
274 | ||
275 | #else /* CONFIG_SMP */ | |
276 | ||
5d2295f3 SAS |
277 | static inline bool is_migration_base(struct hrtimer_clock_base *base) |
278 | { | |
279 | return false; | |
280 | } | |
281 | ||
3c8aa39d | 282 | static inline struct hrtimer_clock_base * |
c0a31329 | 283 | lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) |
ccaa4926 | 284 | __acquires(&timer->base->cpu_base->lock) |
c0a31329 | 285 | { |
3c8aa39d | 286 | struct hrtimer_clock_base *base = timer->base; |
c0a31329 | 287 | |
ecb49d1a | 288 | raw_spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
289 | |
290 | return base; | |
291 | } | |
292 | ||
eea08f32 | 293 | # define switch_hrtimer_base(t, b, p) (b) |
c0a31329 TG |
294 | |
295 | #endif /* !CONFIG_SMP */ | |
296 | ||
297 | /* | |
298 | * Functions for the union type storage format of ktime_t which are | |
299 | * too large for inlining: | |
300 | */ | |
301 | #if BITS_PER_LONG < 64 | |
c0a31329 TG |
302 | /* |
303 | * Divide a ktime value by a nanosecond value | |
304 | */ | |
f7bcb70e | 305 | s64 __ktime_divns(const ktime_t kt, s64 div) |
c0a31329 | 306 | { |
c0a31329 | 307 | int sft = 0; |
f7bcb70e JS |
308 | s64 dclc; |
309 | u64 tmp; | |
c0a31329 | 310 | |
900cfa46 | 311 | dclc = ktime_to_ns(kt); |
f7bcb70e JS |
312 | tmp = dclc < 0 ? -dclc : dclc; |
313 | ||
c0a31329 TG |
314 | /* Make sure the divisor is less than 2^32: */ |
315 | while (div >> 32) { | |
316 | sft++; | |
317 | div >>= 1; | |
318 | } | |
f7bcb70e | 319 | tmp >>= sft; |
38f7b0b1 | 320 | do_div(tmp, (u32) div); |
f7bcb70e | 321 | return dclc < 0 ? -tmp : tmp; |
c0a31329 | 322 | } |
8b618628 | 323 | EXPORT_SYMBOL_GPL(__ktime_divns); |
c0a31329 TG |
324 | #endif /* BITS_PER_LONG >= 64 */ |
325 | ||
5a7780e7 TG |
326 | /* |
327 | * Add two ktime values and do a safety check for overflow: | |
328 | */ | |
329 | ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs) | |
330 | { | |
979515c5 | 331 | ktime_t res = ktime_add_unsafe(lhs, rhs); |
5a7780e7 TG |
332 | |
333 | /* | |
334 | * We use KTIME_SEC_MAX here, the maximum timeout which we can | |
335 | * return to user space in a timespec: | |
336 | */ | |
2456e855 | 337 | if (res < 0 || res < lhs || res < rhs) |
5a7780e7 TG |
338 | res = ktime_set(KTIME_SEC_MAX, 0); |
339 | ||
340 | return res; | |
341 | } | |
342 | ||
8daa21e6 AB |
343 | EXPORT_SYMBOL_GPL(ktime_add_safe); |
344 | ||
237fc6e7 TG |
345 | #ifdef CONFIG_DEBUG_OBJECTS_TIMERS |
346 | ||
f9e62f31 | 347 | static const struct debug_obj_descr hrtimer_debug_descr; |
237fc6e7 | 348 | |
99777288 SG |
349 | static void *hrtimer_debug_hint(void *addr) |
350 | { | |
351 | return ((struct hrtimer *) addr)->function; | |
352 | } | |
353 | ||
237fc6e7 TG |
354 | /* |
355 | * fixup_init is called when: | |
356 | * - an active object is initialized | |
357 | */ | |
e3252464 | 358 | static bool hrtimer_fixup_init(void *addr, enum debug_obj_state state) |
237fc6e7 TG |
359 | { |
360 | struct hrtimer *timer = addr; | |
361 | ||
362 | switch (state) { | |
363 | case ODEBUG_STATE_ACTIVE: | |
364 | hrtimer_cancel(timer); | |
365 | debug_object_init(timer, &hrtimer_debug_descr); | |
e3252464 | 366 | return true; |
237fc6e7 | 367 | default: |
e3252464 | 368 | return false; |
237fc6e7 TG |
369 | } |
370 | } | |
371 | ||
372 | /* | |
373 | * fixup_activate is called when: | |
374 | * - an active object is activated | |
b9fdac7f | 375 | * - an unknown non-static object is activated |
237fc6e7 | 376 | */ |
e3252464 | 377 | static bool hrtimer_fixup_activate(void *addr, enum debug_obj_state state) |
237fc6e7 TG |
378 | { |
379 | switch (state) { | |
237fc6e7 TG |
380 | case ODEBUG_STATE_ACTIVE: |
381 | WARN_ON(1); | |
df561f66 | 382 | fallthrough; |
237fc6e7 | 383 | default: |
e3252464 | 384 | return false; |
237fc6e7 TG |
385 | } |
386 | } | |
387 | ||
388 | /* | |
389 | * fixup_free is called when: | |
390 | * - an active object is freed | |
391 | */ | |
e3252464 | 392 | static bool hrtimer_fixup_free(void *addr, enum debug_obj_state state) |
237fc6e7 TG |
393 | { |
394 | struct hrtimer *timer = addr; | |
395 | ||
396 | switch (state) { | |
397 | case ODEBUG_STATE_ACTIVE: | |
398 | hrtimer_cancel(timer); | |
399 | debug_object_free(timer, &hrtimer_debug_descr); | |
e3252464 | 400 | return true; |
237fc6e7 | 401 | default: |
e3252464 | 402 | return false; |
237fc6e7 TG |
403 | } |
404 | } | |
405 | ||
f9e62f31 | 406 | static const struct debug_obj_descr hrtimer_debug_descr = { |
237fc6e7 | 407 | .name = "hrtimer", |
99777288 | 408 | .debug_hint = hrtimer_debug_hint, |
237fc6e7 TG |
409 | .fixup_init = hrtimer_fixup_init, |
410 | .fixup_activate = hrtimer_fixup_activate, | |
411 | .fixup_free = hrtimer_fixup_free, | |
412 | }; | |
413 | ||
414 | static inline void debug_hrtimer_init(struct hrtimer *timer) | |
415 | { | |
416 | debug_object_init(timer, &hrtimer_debug_descr); | |
417 | } | |
418 | ||
5da70160 AMG |
419 | static inline void debug_hrtimer_activate(struct hrtimer *timer, |
420 | enum hrtimer_mode mode) | |
237fc6e7 TG |
421 | { |
422 | debug_object_activate(timer, &hrtimer_debug_descr); | |
423 | } | |
424 | ||
425 | static inline void debug_hrtimer_deactivate(struct hrtimer *timer) | |
426 | { | |
427 | debug_object_deactivate(timer, &hrtimer_debug_descr); | |
428 | } | |
429 | ||
237fc6e7 TG |
430 | static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id, |
431 | enum hrtimer_mode mode); | |
432 | ||
433 | void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id, | |
434 | enum hrtimer_mode mode) | |
435 | { | |
436 | debug_object_init_on_stack(timer, &hrtimer_debug_descr); | |
437 | __hrtimer_init(timer, clock_id, mode); | |
438 | } | |
2bc481cf | 439 | EXPORT_SYMBOL_GPL(hrtimer_init_on_stack); |
237fc6e7 | 440 | |
dbc1625f SAS |
441 | static void __hrtimer_init_sleeper(struct hrtimer_sleeper *sl, |
442 | clockid_t clock_id, enum hrtimer_mode mode); | |
443 | ||
444 | void hrtimer_init_sleeper_on_stack(struct hrtimer_sleeper *sl, | |
445 | clockid_t clock_id, enum hrtimer_mode mode) | |
446 | { | |
447 | debug_object_init_on_stack(&sl->timer, &hrtimer_debug_descr); | |
448 | __hrtimer_init_sleeper(sl, clock_id, mode); | |
449 | } | |
450 | EXPORT_SYMBOL_GPL(hrtimer_init_sleeper_on_stack); | |
451 | ||
237fc6e7 TG |
452 | void destroy_hrtimer_on_stack(struct hrtimer *timer) |
453 | { | |
454 | debug_object_free(timer, &hrtimer_debug_descr); | |
455 | } | |
c08376ac | 456 | EXPORT_SYMBOL_GPL(destroy_hrtimer_on_stack); |
237fc6e7 TG |
457 | |
458 | #else | |
5da70160 | 459 | |
237fc6e7 | 460 | static inline void debug_hrtimer_init(struct hrtimer *timer) { } |
5da70160 AMG |
461 | static inline void debug_hrtimer_activate(struct hrtimer *timer, |
462 | enum hrtimer_mode mode) { } | |
237fc6e7 TG |
463 | static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { } |
464 | #endif | |
465 | ||
c6a2a177 XG |
466 | static inline void |
467 | debug_init(struct hrtimer *timer, clockid_t clockid, | |
468 | enum hrtimer_mode mode) | |
469 | { | |
470 | debug_hrtimer_init(timer); | |
471 | trace_hrtimer_init(timer, clockid, mode); | |
472 | } | |
473 | ||
63e2ed36 AMG |
474 | static inline void debug_activate(struct hrtimer *timer, |
475 | enum hrtimer_mode mode) | |
c6a2a177 | 476 | { |
5da70160 | 477 | debug_hrtimer_activate(timer, mode); |
63e2ed36 | 478 | trace_hrtimer_start(timer, mode); |
c6a2a177 XG |
479 | } |
480 | ||
481 | static inline void debug_deactivate(struct hrtimer *timer) | |
482 | { | |
483 | debug_hrtimer_deactivate(timer); | |
484 | trace_hrtimer_cancel(timer); | |
485 | } | |
486 | ||
c272ca58 AMG |
487 | static struct hrtimer_clock_base * |
488 | __next_base(struct hrtimer_cpu_base *cpu_base, unsigned int *active) | |
489 | { | |
490 | unsigned int idx; | |
491 | ||
492 | if (!*active) | |
493 | return NULL; | |
494 | ||
495 | idx = __ffs(*active); | |
496 | *active &= ~(1U << idx); | |
497 | ||
498 | return &cpu_base->clock_base[idx]; | |
499 | } | |
500 | ||
501 | #define for_each_active_base(base, cpu_base, active) \ | |
502 | while ((base = __next_base((cpu_base), &(active)))) | |
503 | ||
ad38f596 | 504 | static ktime_t __hrtimer_next_event_base(struct hrtimer_cpu_base *cpu_base, |
a59855cd | 505 | const struct hrtimer *exclude, |
ad38f596 AMG |
506 | unsigned int active, |
507 | ktime_t expires_next) | |
9bc74919 | 508 | { |
c272ca58 | 509 | struct hrtimer_clock_base *base; |
ad38f596 | 510 | ktime_t expires; |
9bc74919 | 511 | |
c272ca58 | 512 | for_each_active_base(base, cpu_base, active) { |
9bc74919 TG |
513 | struct timerqueue_node *next; |
514 | struct hrtimer *timer; | |
515 | ||
34aee88a | 516 | next = timerqueue_getnext(&base->active); |
9bc74919 | 517 | timer = container_of(next, struct hrtimer, node); |
a59855cd RW |
518 | if (timer == exclude) { |
519 | /* Get to the next timer in the queue. */ | |
7d2f6abb | 520 | next = timerqueue_iterate_next(next); |
a59855cd RW |
521 | if (!next) |
522 | continue; | |
523 | ||
524 | timer = container_of(next, struct hrtimer, node); | |
525 | } | |
9bc74919 | 526 | expires = ktime_sub(hrtimer_get_expires(timer), base->offset); |
2456e855 | 527 | if (expires < expires_next) { |
9bc74919 | 528 | expires_next = expires; |
a59855cd RW |
529 | |
530 | /* Skip cpu_base update if a timer is being excluded. */ | |
531 | if (exclude) | |
532 | continue; | |
533 | ||
5da70160 AMG |
534 | if (timer->is_soft) |
535 | cpu_base->softirq_next_timer = timer; | |
536 | else | |
537 | cpu_base->next_timer = timer; | |
895bdfa7 | 538 | } |
9bc74919 TG |
539 | } |
540 | /* | |
541 | * clock_was_set() might have changed base->offset of any of | |
542 | * the clock bases so the result might be negative. Fix it up | |
543 | * to prevent a false positive in clockevents_program_event(). | |
544 | */ | |
2456e855 TG |
545 | if (expires_next < 0) |
546 | expires_next = 0; | |
9bc74919 TG |
547 | return expires_next; |
548 | } | |
9bc74919 | 549 | |
c458b1d1 | 550 | /* |
46eb1701 AMB |
551 | * Recomputes cpu_base::*next_timer and returns the earliest expires_next |
552 | * but does not set cpu_base::*expires_next, that is done by | |
553 | * hrtimer[_force]_reprogram and hrtimer_interrupt only. When updating | |
554 | * cpu_base::*expires_next right away, reprogramming logic would no longer | |
555 | * work. | |
c458b1d1 | 556 | * |
5da70160 AMG |
557 | * When a softirq is pending, we can ignore the HRTIMER_ACTIVE_SOFT bases, |
558 | * those timers will get run whenever the softirq gets handled, at the end of | |
559 | * hrtimer_run_softirq(), hrtimer_update_softirq_timer() will re-add these bases. | |
560 | * | |
561 | * Therefore softirq values are those from the HRTIMER_ACTIVE_SOFT clock bases. | |
562 | * The !softirq values are the minima across HRTIMER_ACTIVE_ALL, unless an actual | |
563 | * softirq is pending, in which case they're the minima of HRTIMER_ACTIVE_HARD. | |
564 | * | |
c458b1d1 | 565 | * @active_mask must be one of: |
5da70160 | 566 | * - HRTIMER_ACTIVE_ALL, |
c458b1d1 AMG |
567 | * - HRTIMER_ACTIVE_SOFT, or |
568 | * - HRTIMER_ACTIVE_HARD. | |
569 | */ | |
5da70160 AMG |
570 | static ktime_t |
571 | __hrtimer_get_next_event(struct hrtimer_cpu_base *cpu_base, unsigned int active_mask) | |
ad38f596 | 572 | { |
c458b1d1 | 573 | unsigned int active; |
5da70160 | 574 | struct hrtimer *next_timer = NULL; |
ad38f596 AMG |
575 | ktime_t expires_next = KTIME_MAX; |
576 | ||
5da70160 AMG |
577 | if (!cpu_base->softirq_activated && (active_mask & HRTIMER_ACTIVE_SOFT)) { |
578 | active = cpu_base->active_bases & HRTIMER_ACTIVE_SOFT; | |
579 | cpu_base->softirq_next_timer = NULL; | |
a59855cd RW |
580 | expires_next = __hrtimer_next_event_base(cpu_base, NULL, |
581 | active, KTIME_MAX); | |
5da70160 AMG |
582 | |
583 | next_timer = cpu_base->softirq_next_timer; | |
584 | } | |
ad38f596 | 585 | |
5da70160 AMG |
586 | if (active_mask & HRTIMER_ACTIVE_HARD) { |
587 | active = cpu_base->active_bases & HRTIMER_ACTIVE_HARD; | |
588 | cpu_base->next_timer = next_timer; | |
a59855cd RW |
589 | expires_next = __hrtimer_next_event_base(cpu_base, NULL, active, |
590 | expires_next); | |
5da70160 | 591 | } |
ad38f596 AMG |
592 | |
593 | return expires_next; | |
594 | } | |
595 | ||
46eb1701 AMB |
596 | static ktime_t hrtimer_update_next_event(struct hrtimer_cpu_base *cpu_base) |
597 | { | |
598 | ktime_t expires_next, soft = KTIME_MAX; | |
599 | ||
600 | /* | |
601 | * If the soft interrupt has already been activated, ignore the | |
602 | * soft bases. They will be handled in the already raised soft | |
603 | * interrupt. | |
604 | */ | |
605 | if (!cpu_base->softirq_activated) { | |
606 | soft = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_SOFT); | |
607 | /* | |
608 | * Update the soft expiry time. clock_settime() might have | |
609 | * affected it. | |
610 | */ | |
611 | cpu_base->softirq_expires_next = soft; | |
612 | } | |
613 | ||
614 | expires_next = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_HARD); | |
615 | /* | |
616 | * If a softirq timer is expiring first, update cpu_base->next_timer | |
617 | * and program the hardware with the soft expiry time. | |
618 | */ | |
619 | if (expires_next > soft) { | |
620 | cpu_base->next_timer = cpu_base->softirq_next_timer; | |
621 | expires_next = soft; | |
622 | } | |
623 | ||
624 | return expires_next; | |
625 | } | |
626 | ||
21d6d52a TG |
627 | static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base) |
628 | { | |
629 | ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset; | |
a3ed0e43 | 630 | ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset; |
21d6d52a TG |
631 | ktime_t *offs_tai = &base->clock_base[HRTIMER_BASE_TAI].offset; |
632 | ||
5da70160 | 633 | ktime_t now = ktime_get_update_offsets_now(&base->clock_was_set_seq, |
a3ed0e43 | 634 | offs_real, offs_boot, offs_tai); |
5da70160 AMG |
635 | |
636 | base->clock_base[HRTIMER_BASE_REALTIME_SOFT].offset = *offs_real; | |
a3ed0e43 | 637 | base->clock_base[HRTIMER_BASE_BOOTTIME_SOFT].offset = *offs_boot; |
5da70160 AMG |
638 | base->clock_base[HRTIMER_BASE_TAI_SOFT].offset = *offs_tai; |
639 | ||
640 | return now; | |
21d6d52a TG |
641 | } |
642 | ||
28bfd18b AMG |
643 | /* |
644 | * Is the high resolution mode active ? | |
645 | */ | |
646 | static inline int __hrtimer_hres_active(struct hrtimer_cpu_base *cpu_base) | |
647 | { | |
648 | return IS_ENABLED(CONFIG_HIGH_RES_TIMERS) ? | |
649 | cpu_base->hres_active : 0; | |
650 | } | |
651 | ||
652 | static inline int hrtimer_hres_active(void) | |
653 | { | |
654 | return __hrtimer_hres_active(this_cpu_ptr(&hrtimer_bases)); | |
655 | } | |
656 | ||
f80e2148 TG |
657 | static void __hrtimer_reprogram(struct hrtimer_cpu_base *cpu_base, |
658 | struct hrtimer *next_timer, | |
659 | ktime_t expires_next) | |
54cdfdb4 | 660 | { |
2456e855 | 661 | cpu_base->expires_next = expires_next; |
7403f41f | 662 | |
6c6c0d5a | 663 | /* |
61bb4bcb AMG |
664 | * If hres is not active, hardware does not have to be |
665 | * reprogrammed yet. | |
666 | * | |
6c6c0d5a SH |
667 | * If a hang was detected in the last timer interrupt then we |
668 | * leave the hang delay active in the hardware. We want the | |
669 | * system to make progress. That also prevents the following | |
670 | * scenario: | |
671 | * T1 expires 50ms from now | |
672 | * T2 expires 5s from now | |
673 | * | |
674 | * T1 is removed, so this code is called and would reprogram | |
675 | * the hardware to 5s from now. Any hrtimer_start after that | |
676 | * will not reprogram the hardware due to hang_detected being | |
4bf07f65 | 677 | * set. So we'd effectively block all timers until the T2 event |
6c6c0d5a SH |
678 | * fires. |
679 | */ | |
61bb4bcb | 680 | if (!__hrtimer_hres_active(cpu_base) || cpu_base->hang_detected) |
6c6c0d5a SH |
681 | return; |
682 | ||
b14bca97 PZ |
683 | tick_program_event(expires_next, 1); |
684 | } | |
685 | ||
686 | /* | |
687 | * Reprogram the event source with checking both queues for the | |
688 | * next event | |
689 | * Called with interrupts disabled and base->lock held | |
690 | */ | |
691 | static void | |
692 | hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal) | |
693 | { | |
694 | ktime_t expires_next; | |
695 | ||
696 | expires_next = hrtimer_update_next_event(cpu_base); | |
697 | ||
f80e2148 TG |
698 | if (skip_equal && expires_next == cpu_base->expires_next) |
699 | return; | |
700 | ||
701 | __hrtimer_reprogram(cpu_base, cpu_base->next_timer, expires_next); | |
54cdfdb4 TG |
702 | } |
703 | ||
ebba2c72 AMG |
704 | /* High resolution timer related functions */ |
705 | #ifdef CONFIG_HIGH_RES_TIMERS | |
706 | ||
707 | /* | |
708 | * High resolution timer enabled ? | |
709 | */ | |
710 | static bool hrtimer_hres_enabled __read_mostly = true; | |
711 | unsigned int hrtimer_resolution __read_mostly = LOW_RES_NSEC; | |
712 | EXPORT_SYMBOL_GPL(hrtimer_resolution); | |
713 | ||
714 | /* | |
715 | * Enable / Disable high resolution mode | |
716 | */ | |
717 | static int __init setup_hrtimer_hres(char *str) | |
718 | { | |
719 | return (kstrtobool(str, &hrtimer_hres_enabled) == 0); | |
720 | } | |
721 | ||
722 | __setup("highres=", setup_hrtimer_hres); | |
723 | ||
724 | /* | |
725 | * hrtimer_high_res_enabled - query, if the highres mode is enabled | |
726 | */ | |
727 | static inline int hrtimer_is_hres_enabled(void) | |
728 | { | |
729 | return hrtimer_hres_enabled; | |
730 | } | |
731 | ||
e71a4153 | 732 | static void retrigger_next_event(void *arg); |
b12a03ce | 733 | |
54cdfdb4 TG |
734 | /* |
735 | * Switch to high resolution mode | |
736 | */ | |
75e3b37d | 737 | static void hrtimer_switch_to_hres(void) |
54cdfdb4 | 738 | { |
c6eb3f70 | 739 | struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases); |
54cdfdb4 TG |
740 | |
741 | if (tick_init_highres()) { | |
7a6e5537 GU |
742 | pr_warn("Could not switch to high resolution mode on CPU %u\n", |
743 | base->cpu); | |
85e1cd6e | 744 | return; |
54cdfdb4 TG |
745 | } |
746 | base->hres_active = 1; | |
398ca17f | 747 | hrtimer_resolution = HIGH_RES_NSEC; |
54cdfdb4 TG |
748 | |
749 | tick_setup_sched_timer(); | |
54cdfdb4 TG |
750 | /* "Retrigger" the interrupt to get things going */ |
751 | retrigger_next_event(NULL); | |
54cdfdb4 TG |
752 | } |
753 | ||
754 | #else | |
755 | ||
54cdfdb4 | 756 | static inline int hrtimer_is_hres_enabled(void) { return 0; } |
75e3b37d | 757 | static inline void hrtimer_switch_to_hres(void) { } |
54cdfdb4 TG |
758 | |
759 | #endif /* CONFIG_HIGH_RES_TIMERS */ | |
e71a4153 TG |
760 | /* |
761 | * Retrigger next event is called after clock was set with interrupts | |
762 | * disabled through an SMP function call or directly from low level | |
763 | * resume code. | |
764 | * | |
765 | * This is only invoked when: | |
766 | * - CONFIG_HIGH_RES_TIMERS is enabled. | |
767 | * - CONFIG_NOHZ_COMMON is enabled | |
768 | * | |
769 | * For the other cases this function is empty and because the call sites | |
770 | * are optimized out it vanishes as well, i.e. no need for lots of | |
771 | * #ifdeffery. | |
772 | */ | |
773 | static void retrigger_next_event(void *arg) | |
774 | { | |
775 | struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases); | |
776 | ||
777 | /* | |
778 | * When high resolution mode or nohz is active, then the offsets of | |
779 | * CLOCK_REALTIME/TAI/BOOTTIME have to be updated. Otherwise the | |
780 | * next tick will take care of that. | |
781 | * | |
782 | * If high resolution mode is active then the next expiring timer | |
783 | * must be reevaluated and the clock event device reprogrammed if | |
784 | * necessary. | |
785 | * | |
786 | * In the NOHZ case the update of the offset and the reevaluation | |
787 | * of the next expiring timer is enough. The return from the SMP | |
788 | * function call will take care of the reprogramming in case the | |
789 | * CPU was in a NOHZ idle sleep. | |
790 | */ | |
791 | if (!__hrtimer_hres_active(base) && !tick_nohz_active) | |
792 | return; | |
793 | ||
794 | raw_spin_lock(&base->lock); | |
795 | hrtimer_update_base(base); | |
796 | if (__hrtimer_hres_active(base)) | |
797 | hrtimer_force_reprogram(base, 0); | |
798 | else | |
799 | hrtimer_update_next_event(base); | |
800 | raw_spin_unlock(&base->lock); | |
801 | } | |
54cdfdb4 | 802 | |
11a9fe06 AMG |
803 | /* |
804 | * When a timer is enqueued and expires earlier than the already enqueued | |
805 | * timers, we have to check, whether it expires earlier than the timer for | |
806 | * which the clock event device was armed. | |
807 | * | |
808 | * Called with interrupts disabled and base->cpu_base.lock held | |
809 | */ | |
5da70160 | 810 | static void hrtimer_reprogram(struct hrtimer *timer, bool reprogram) |
11a9fe06 AMG |
811 | { |
812 | struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases); | |
3ec7a3ee | 813 | struct hrtimer_clock_base *base = timer->base; |
11a9fe06 AMG |
814 | ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset); |
815 | ||
816 | WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0); | |
817 | ||
5da70160 AMG |
818 | /* |
819 | * CLOCK_REALTIME timer might be requested with an absolute | |
820 | * expiry time which is less than base->offset. Set it to 0. | |
821 | */ | |
822 | if (expires < 0) | |
823 | expires = 0; | |
824 | ||
825 | if (timer->is_soft) { | |
826 | /* | |
827 | * soft hrtimer could be started on a remote CPU. In this | |
828 | * case softirq_expires_next needs to be updated on the | |
829 | * remote CPU. The soft hrtimer will not expire before the | |
830 | * first hard hrtimer on the remote CPU - | |
831 | * hrtimer_check_target() prevents this case. | |
832 | */ | |
833 | struct hrtimer_cpu_base *timer_cpu_base = base->cpu_base; | |
834 | ||
835 | if (timer_cpu_base->softirq_activated) | |
836 | return; | |
837 | ||
838 | if (!ktime_before(expires, timer_cpu_base->softirq_expires_next)) | |
839 | return; | |
840 | ||
841 | timer_cpu_base->softirq_next_timer = timer; | |
842 | timer_cpu_base->softirq_expires_next = expires; | |
843 | ||
844 | if (!ktime_before(expires, timer_cpu_base->expires_next) || | |
845 | !reprogram) | |
846 | return; | |
847 | } | |
848 | ||
11a9fe06 AMG |
849 | /* |
850 | * If the timer is not on the current cpu, we cannot reprogram | |
851 | * the other cpus clock event device. | |
852 | */ | |
853 | if (base->cpu_base != cpu_base) | |
854 | return; | |
855 | ||
f80e2148 TG |
856 | if (expires >= cpu_base->expires_next) |
857 | return; | |
858 | ||
859 | /* | |
860 | * If the hrtimer interrupt is running, then it will reevaluate the | |
861 | * clock bases and reprogram the clock event device. | |
862 | */ | |
863 | if (cpu_base->in_hrtirq) | |
864 | return; | |
865 | ||
866 | cpu_base->next_timer = timer; | |
867 | ||
868 | __hrtimer_reprogram(cpu_base, timer, expires); | |
11a9fe06 AMG |
869 | } |
870 | ||
1e7f7fbc TG |
871 | static bool update_needs_ipi(struct hrtimer_cpu_base *cpu_base, |
872 | unsigned int active) | |
873 | { | |
874 | struct hrtimer_clock_base *base; | |
875 | unsigned int seq; | |
876 | ktime_t expires; | |
877 | ||
878 | /* | |
879 | * Update the base offsets unconditionally so the following | |
880 | * checks whether the SMP function call is required works. | |
881 | * | |
882 | * The update is safe even when the remote CPU is in the hrtimer | |
883 | * interrupt or the hrtimer soft interrupt and expiring affected | |
884 | * bases. Either it will see the update before handling a base or | |
885 | * it will see it when it finishes the processing and reevaluates | |
886 | * the next expiring timer. | |
887 | */ | |
888 | seq = cpu_base->clock_was_set_seq; | |
889 | hrtimer_update_base(cpu_base); | |
890 | ||
891 | /* | |
892 | * If the sequence did not change over the update then the | |
893 | * remote CPU already handled it. | |
894 | */ | |
895 | if (seq == cpu_base->clock_was_set_seq) | |
896 | return false; | |
897 | ||
898 | /* | |
899 | * If the remote CPU is currently handling an hrtimer interrupt, it | |
900 | * will reevaluate the first expiring timer of all clock bases | |
901 | * before reprogramming. Nothing to do here. | |
902 | */ | |
903 | if (cpu_base->in_hrtirq) | |
904 | return false; | |
905 | ||
906 | /* | |
907 | * Walk the affected clock bases and check whether the first expiring | |
908 | * timer in a clock base is moving ahead of the first expiring timer of | |
909 | * @cpu_base. If so, the IPI must be invoked because per CPU clock | |
910 | * event devices cannot be remotely reprogrammed. | |
911 | */ | |
912 | active &= cpu_base->active_bases; | |
913 | ||
914 | for_each_active_base(base, cpu_base, active) { | |
915 | struct timerqueue_node *next; | |
916 | ||
917 | next = timerqueue_getnext(&base->active); | |
918 | expires = ktime_sub(next->expires, base->offset); | |
919 | if (expires < cpu_base->expires_next) | |
920 | return true; | |
921 | ||
922 | /* Extra check for softirq clock bases */ | |
923 | if (base->clockid < HRTIMER_BASE_MONOTONIC_SOFT) | |
924 | continue; | |
925 | if (cpu_base->softirq_activated) | |
926 | continue; | |
927 | if (expires < cpu_base->softirq_expires_next) | |
928 | return true; | |
929 | } | |
930 | return false; | |
931 | } | |
932 | ||
b12a03ce | 933 | /* |
e71a4153 TG |
934 | * Clock was set. This might affect CLOCK_REALTIME, CLOCK_TAI and |
935 | * CLOCK_BOOTTIME (for late sleep time injection). | |
b12a03ce | 936 | * |
e71a4153 TG |
937 | * This requires to update the offsets for these clocks |
938 | * vs. CLOCK_MONOTONIC. When high resolution timers are enabled, then this | |
939 | * also requires to eventually reprogram the per CPU clock event devices | |
940 | * when the change moves an affected timer ahead of the first expiring | |
941 | * timer on that CPU. Obviously remote per CPU clock event devices cannot | |
942 | * be reprogrammed. The other reason why an IPI has to be sent is when the | |
943 | * system is in !HIGH_RES and NOHZ mode. The NOHZ mode updates the offsets | |
944 | * in the tick, which obviously might be stopped, so this has to bring out | |
945 | * the remote CPU which might sleep in idle to get this sorted. | |
b12a03ce | 946 | */ |
17a1b882 | 947 | void clock_was_set(unsigned int bases) |
b12a03ce | 948 | { |
9482fd71 | 949 | struct hrtimer_cpu_base *cpu_base = raw_cpu_ptr(&hrtimer_bases); |
81d741d3 MT |
950 | cpumask_var_t mask; |
951 | int cpu; | |
952 | ||
9482fd71 | 953 | if (!__hrtimer_hres_active(cpu_base) && !tick_nohz_active) |
e71a4153 TG |
954 | goto out_timerfd; |
955 | ||
81d741d3 MT |
956 | if (!zalloc_cpumask_var(&mask, GFP_KERNEL)) { |
957 | on_each_cpu(retrigger_next_event, NULL, 1); | |
958 | goto out_timerfd; | |
959 | } | |
960 | ||
961 | /* Avoid interrupting CPUs if possible */ | |
962 | cpus_read_lock(); | |
963 | for_each_online_cpu(cpu) { | |
81d741d3 MT |
964 | unsigned long flags; |
965 | ||
9482fd71 | 966 | cpu_base = &per_cpu(hrtimer_bases, cpu); |
81d741d3 | 967 | raw_spin_lock_irqsave(&cpu_base->lock, flags); |
1e7f7fbc TG |
968 | |
969 | if (update_needs_ipi(cpu_base, bases)) | |
81d741d3 | 970 | cpumask_set_cpu(cpu, mask); |
1e7f7fbc | 971 | |
81d741d3 MT |
972 | raw_spin_unlock_irqrestore(&cpu_base->lock, flags); |
973 | } | |
974 | ||
975 | preempt_disable(); | |
976 | smp_call_function_many(mask, retrigger_next_event, NULL, 1); | |
977 | preempt_enable(); | |
978 | cpus_read_unlock(); | |
979 | free_cpumask_var(mask); | |
e71a4153 TG |
980 | |
981 | out_timerfd: | |
9ec26907 | 982 | timerfd_clock_was_set(); |
b12a03ce TG |
983 | } |
984 | ||
8c3b5e6e TG |
985 | static void clock_was_set_work(struct work_struct *work) |
986 | { | |
17a1b882 | 987 | clock_was_set(CLOCK_SET_WALL); |
8c3b5e6e TG |
988 | } |
989 | ||
990 | static DECLARE_WORK(hrtimer_work, clock_was_set_work); | |
991 | ||
992 | /* | |
a761a67f TG |
993 | * Called from timekeeping code to reprogram the hrtimer interrupt device |
994 | * on all cpus and to notify timerfd. | |
8c3b5e6e TG |
995 | */ |
996 | void clock_was_set_delayed(void) | |
997 | { | |
998 | schedule_work(&hrtimer_work); | |
999 | } | |
1000 | ||
b12a03ce | 1001 | /* |
a761a67f TG |
1002 | * Called during resume either directly from via timekeeping_resume() |
1003 | * or in the case of s2idle from tick_unfreeze() to ensure that the | |
1004 | * hrtimers are up to date. | |
b12a03ce | 1005 | */ |
a761a67f | 1006 | void hrtimers_resume_local(void) |
b12a03ce | 1007 | { |
53bef3fd | 1008 | lockdep_assert_irqs_disabled(); |
5ec2481b | 1009 | /* Retrigger on the local CPU */ |
b12a03ce TG |
1010 | retrigger_next_event(NULL); |
1011 | } | |
1012 | ||
c0a31329 | 1013 | /* |
6506f2aa | 1014 | * Counterpart to lock_hrtimer_base above: |
c0a31329 TG |
1015 | */ |
1016 | static inline | |
1017 | void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) | |
ccaa4926 | 1018 | __releases(&timer->base->cpu_base->lock) |
c0a31329 | 1019 | { |
ecb49d1a | 1020 | raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags); |
c0a31329 TG |
1021 | } |
1022 | ||
1023 | /** | |
1024 | * hrtimer_forward - forward the timer expiry | |
c0a31329 | 1025 | * @timer: hrtimer to forward |
44f21475 | 1026 | * @now: forward past this time |
c0a31329 TG |
1027 | * @interval: the interval to forward |
1028 | * | |
1029 | * Forward the timer expiry so it will expire in the future. | |
8dca6f33 | 1030 | * Returns the number of overruns. |
91e5a217 TG |
1031 | * |
1032 | * Can be safely called from the callback function of @timer. If | |
1033 | * called from other contexts @timer must neither be enqueued nor | |
1034 | * running the callback and the caller needs to take care of | |
1035 | * serialization. | |
1036 | * | |
1037 | * Note: This only updates the timer expiry value and does not requeue | |
1038 | * the timer. | |
c0a31329 | 1039 | */ |
4d672e7a | 1040 | u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval) |
c0a31329 | 1041 | { |
4d672e7a | 1042 | u64 orun = 1; |
44f21475 | 1043 | ktime_t delta; |
c0a31329 | 1044 | |
cc584b21 | 1045 | delta = ktime_sub(now, hrtimer_get_expires(timer)); |
c0a31329 | 1046 | |
2456e855 | 1047 | if (delta < 0) |
c0a31329 TG |
1048 | return 0; |
1049 | ||
5de2755c PZ |
1050 | if (WARN_ON(timer->state & HRTIMER_STATE_ENQUEUED)) |
1051 | return 0; | |
1052 | ||
2456e855 TG |
1053 | if (interval < hrtimer_resolution) |
1054 | interval = hrtimer_resolution; | |
c9db4fa1 | 1055 | |
2456e855 | 1056 | if (unlikely(delta >= interval)) { |
df869b63 | 1057 | s64 incr = ktime_to_ns(interval); |
c0a31329 TG |
1058 | |
1059 | orun = ktime_divns(delta, incr); | |
cc584b21 | 1060 | hrtimer_add_expires_ns(timer, incr * orun); |
2456e855 | 1061 | if (hrtimer_get_expires_tv64(timer) > now) |
c0a31329 TG |
1062 | return orun; |
1063 | /* | |
1064 | * This (and the ktime_add() below) is the | |
1065 | * correction for exact: | |
1066 | */ | |
1067 | orun++; | |
1068 | } | |
cc584b21 | 1069 | hrtimer_add_expires(timer, interval); |
c0a31329 TG |
1070 | |
1071 | return orun; | |
1072 | } | |
6bdb6b62 | 1073 | EXPORT_SYMBOL_GPL(hrtimer_forward); |
c0a31329 TG |
1074 | |
1075 | /* | |
1076 | * enqueue_hrtimer - internal function to (re)start a timer | |
1077 | * | |
1078 | * The timer is inserted in expiry order. Insertion into the | |
1079 | * red black tree is O(log(n)). Must hold the base lock. | |
a6037b61 PZ |
1080 | * |
1081 | * Returns 1 when the new timer is the leftmost timer in the tree. | |
c0a31329 | 1082 | */ |
a6037b61 | 1083 | static int enqueue_hrtimer(struct hrtimer *timer, |
63e2ed36 AMG |
1084 | struct hrtimer_clock_base *base, |
1085 | enum hrtimer_mode mode) | |
c0a31329 | 1086 | { |
63e2ed36 | 1087 | debug_activate(timer, mode); |
237fc6e7 | 1088 | |
ab8177bc | 1089 | base->cpu_base->active_bases |= 1 << base->index; |
54cdfdb4 | 1090 | |
56144737 ED |
1091 | /* Pairs with the lockless read in hrtimer_is_queued() */ |
1092 | WRITE_ONCE(timer->state, HRTIMER_STATE_ENQUEUED); | |
a6037b61 | 1093 | |
b97f44c9 | 1094 | return timerqueue_add(&base->active, &timer->node); |
288867ec | 1095 | } |
c0a31329 TG |
1096 | |
1097 | /* | |
1098 | * __remove_hrtimer - internal function to remove a timer | |
1099 | * | |
1100 | * Caller must hold the base lock. | |
54cdfdb4 TG |
1101 | * |
1102 | * High resolution timer mode reprograms the clock event device when the | |
1103 | * timer is the one which expires next. The caller can disable this by setting | |
1104 | * reprogram to zero. This is useful, when the context does a reprogramming | |
1105 | * anyway (e.g. timer interrupt) | |
c0a31329 | 1106 | */ |
3c8aa39d | 1107 | static void __remove_hrtimer(struct hrtimer *timer, |
303e967f | 1108 | struct hrtimer_clock_base *base, |
203cbf77 | 1109 | u8 newstate, int reprogram) |
c0a31329 | 1110 | { |
e19ffe8b | 1111 | struct hrtimer_cpu_base *cpu_base = base->cpu_base; |
203cbf77 | 1112 | u8 state = timer->state; |
e19ffe8b | 1113 | |
56144737 ED |
1114 | /* Pairs with the lockless read in hrtimer_is_queued() */ |
1115 | WRITE_ONCE(timer->state, newstate); | |
895bdfa7 TG |
1116 | if (!(state & HRTIMER_STATE_ENQUEUED)) |
1117 | return; | |
7403f41f | 1118 | |
b97f44c9 | 1119 | if (!timerqueue_del(&base->active, &timer->node)) |
e19ffe8b | 1120 | cpu_base->active_bases &= ~(1 << base->index); |
7403f41f | 1121 | |
895bdfa7 TG |
1122 | /* |
1123 | * Note: If reprogram is false we do not update | |
1124 | * cpu_base->next_timer. This happens when we remove the first | |
1125 | * timer on a remote cpu. No harm as we never dereference | |
1126 | * cpu_base->next_timer. So the worst thing what can happen is | |
4bf07f65 | 1127 | * an superfluous call to hrtimer_force_reprogram() on the |
895bdfa7 TG |
1128 | * remote cpu later on if the same timer gets enqueued again. |
1129 | */ | |
1130 | if (reprogram && timer == cpu_base->next_timer) | |
1131 | hrtimer_force_reprogram(cpu_base, 1); | |
c0a31329 TG |
1132 | } |
1133 | ||
1134 | /* | |
1135 | * remove hrtimer, called with base lock held | |
1136 | */ | |
1137 | static inline int | |
627ef5ae TG |
1138 | remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base, |
1139 | bool restart, bool keep_local) | |
c0a31329 | 1140 | { |
56144737 ED |
1141 | u8 state = timer->state; |
1142 | ||
1143 | if (state & HRTIMER_STATE_ENQUEUED) { | |
627ef5ae | 1144 | bool reprogram; |
54cdfdb4 TG |
1145 | |
1146 | /* | |
1147 | * Remove the timer and force reprogramming when high | |
1148 | * resolution mode is active and the timer is on the current | |
1149 | * CPU. If we remove a timer on another CPU, reprogramming is | |
1150 | * skipped. The interrupt event on this CPU is fired and | |
1151 | * reprogramming happens in the interrupt handler. This is a | |
1152 | * rare case and less expensive than a smp call. | |
1153 | */ | |
c6a2a177 | 1154 | debug_deactivate(timer); |
dc5df73b | 1155 | reprogram = base->cpu_base == this_cpu_ptr(&hrtimer_bases); |
8edfb036 | 1156 | |
627ef5ae TG |
1157 | /* |
1158 | * If the timer is not restarted then reprogramming is | |
1159 | * required if the timer is local. If it is local and about | |
1160 | * to be restarted, avoid programming it twice (on removal | |
1161 | * and a moment later when it's requeued). | |
1162 | */ | |
887d9dc9 PZ |
1163 | if (!restart) |
1164 | state = HRTIMER_STATE_INACTIVE; | |
627ef5ae TG |
1165 | else |
1166 | reprogram &= !keep_local; | |
887d9dc9 | 1167 | |
f13d4f97 | 1168 | __remove_hrtimer(timer, base, state, reprogram); |
c0a31329 TG |
1169 | return 1; |
1170 | } | |
1171 | return 0; | |
1172 | } | |
1173 | ||
203cbf77 TG |
1174 | static inline ktime_t hrtimer_update_lowres(struct hrtimer *timer, ktime_t tim, |
1175 | const enum hrtimer_mode mode) | |
1176 | { | |
1177 | #ifdef CONFIG_TIME_LOW_RES | |
1178 | /* | |
1179 | * CONFIG_TIME_LOW_RES indicates that the system has no way to return | |
1180 | * granular time values. For relative timers we add hrtimer_resolution | |
1181 | * (i.e. one jiffie) to prevent short timeouts. | |
1182 | */ | |
1183 | timer->is_rel = mode & HRTIMER_MODE_REL; | |
1184 | if (timer->is_rel) | |
8b0e1953 | 1185 | tim = ktime_add_safe(tim, hrtimer_resolution); |
203cbf77 TG |
1186 | #endif |
1187 | return tim; | |
1188 | } | |
1189 | ||
5da70160 AMG |
1190 | static void |
1191 | hrtimer_update_softirq_timer(struct hrtimer_cpu_base *cpu_base, bool reprogram) | |
1192 | { | |
1193 | ktime_t expires; | |
1194 | ||
1195 | /* | |
1196 | * Find the next SOFT expiration. | |
1197 | */ | |
1198 | expires = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_SOFT); | |
1199 | ||
1200 | /* | |
1201 | * reprogramming needs to be triggered, even if the next soft | |
1202 | * hrtimer expires at the same time than the next hard | |
1203 | * hrtimer. cpu_base->softirq_expires_next needs to be updated! | |
1204 | */ | |
1205 | if (expires == KTIME_MAX) | |
1206 | return; | |
1207 | ||
1208 | /* | |
1209 | * cpu_base->*next_timer is recomputed by __hrtimer_get_next_event() | |
1210 | * cpu_base->*expires_next is only set by hrtimer_reprogram() | |
1211 | */ | |
1212 | hrtimer_reprogram(cpu_base->softirq_next_timer, reprogram); | |
1213 | } | |
1214 | ||
138a6b7a AMG |
1215 | static int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, |
1216 | u64 delta_ns, const enum hrtimer_mode mode, | |
1217 | struct hrtimer_clock_base *base) | |
c0a31329 | 1218 | { |
138a6b7a | 1219 | struct hrtimer_clock_base *new_base; |
627ef5ae | 1220 | bool force_local, first; |
c0a31329 | 1221 | |
627ef5ae TG |
1222 | /* |
1223 | * If the timer is on the local cpu base and is the first expiring | |
1224 | * timer then this might end up reprogramming the hardware twice | |
1225 | * (on removal and on enqueue). To avoid that by prevent the | |
1226 | * reprogram on removal, keep the timer local to the current CPU | |
1227 | * and enforce reprogramming after it is queued no matter whether | |
1228 | * it is the new first expiring timer again or not. | |
1229 | */ | |
1230 | force_local = base->cpu_base == this_cpu_ptr(&hrtimer_bases); | |
1231 | force_local &= base->cpu_base->next_timer == timer; | |
1232 | ||
1233 | /* | |
1234 | * Remove an active timer from the queue. In case it is not queued | |
1235 | * on the current CPU, make sure that remove_hrtimer() updates the | |
1236 | * remote data correctly. | |
1237 | * | |
1238 | * If it's on the current CPU and the first expiring timer, then | |
1239 | * skip reprogramming, keep the timer local and enforce | |
1240 | * reprogramming later if it was the first expiring timer. This | |
1241 | * avoids programming the underlying clock event twice (once at | |
1242 | * removal and once after enqueue). | |
1243 | */ | |
1244 | remove_hrtimer(timer, base, true, force_local); | |
c0a31329 | 1245 | |
203cbf77 | 1246 | if (mode & HRTIMER_MODE_REL) |
84ea7fe3 | 1247 | tim = ktime_add_safe(tim, base->get_time()); |
203cbf77 TG |
1248 | |
1249 | tim = hrtimer_update_lowres(timer, tim, mode); | |
237fc6e7 | 1250 | |
da8f2e17 | 1251 | hrtimer_set_expires_range_ns(timer, tim, delta_ns); |
c0a31329 | 1252 | |
84ea7fe3 | 1253 | /* Switch the timer base, if necessary: */ |
627ef5ae TG |
1254 | if (!force_local) { |
1255 | new_base = switch_hrtimer_base(timer, base, | |
1256 | mode & HRTIMER_MODE_PINNED); | |
1257 | } else { | |
1258 | new_base = base; | |
1259 | } | |
1260 | ||
1261 | first = enqueue_hrtimer(timer, new_base, mode); | |
1262 | if (!force_local) | |
1263 | return first; | |
84ea7fe3 | 1264 | |
627ef5ae TG |
1265 | /* |
1266 | * Timer was forced to stay on the current CPU to avoid | |
1267 | * reprogramming on removal and enqueue. Force reprogram the | |
1268 | * hardware by evaluating the new first expiring timer. | |
1269 | */ | |
1270 | hrtimer_force_reprogram(new_base->cpu_base, 1); | |
1271 | return 0; | |
138a6b7a | 1272 | } |
5da70160 | 1273 | |
138a6b7a AMG |
1274 | /** |
1275 | * hrtimer_start_range_ns - (re)start an hrtimer | |
1276 | * @timer: the timer to be added | |
1277 | * @tim: expiry time | |
1278 | * @delta_ns: "slack" range for the timer | |
1279 | * @mode: timer mode: absolute (HRTIMER_MODE_ABS) or | |
5da70160 AMG |
1280 | * relative (HRTIMER_MODE_REL), and pinned (HRTIMER_MODE_PINNED); |
1281 | * softirq based mode is considered for debug purpose only! | |
138a6b7a AMG |
1282 | */ |
1283 | void hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, | |
1284 | u64 delta_ns, const enum hrtimer_mode mode) | |
1285 | { | |
1286 | struct hrtimer_clock_base *base; | |
1287 | unsigned long flags; | |
1288 | ||
5da70160 AMG |
1289 | /* |
1290 | * Check whether the HRTIMER_MODE_SOFT bit and hrtimer.is_soft | |
0ab6a3dd TG |
1291 | * match on CONFIG_PREEMPT_RT = n. With PREEMPT_RT check the hard |
1292 | * expiry mode because unmarked timers are moved to softirq expiry. | |
5da70160 | 1293 | */ |
0ab6a3dd TG |
1294 | if (!IS_ENABLED(CONFIG_PREEMPT_RT)) |
1295 | WARN_ON_ONCE(!(mode & HRTIMER_MODE_SOFT) ^ !timer->is_soft); | |
1296 | else | |
1297 | WARN_ON_ONCE(!(mode & HRTIMER_MODE_HARD) ^ !timer->is_hard); | |
5da70160 | 1298 | |
138a6b7a AMG |
1299 | base = lock_hrtimer_base(timer, &flags); |
1300 | ||
1301 | if (__hrtimer_start_range_ns(timer, tim, delta_ns, mode, base)) | |
5da70160 | 1302 | hrtimer_reprogram(timer, true); |
49a2a075 | 1303 | |
c0a31329 | 1304 | unlock_hrtimer_base(timer, &flags); |
7f1e2ca9 | 1305 | } |
da8f2e17 AV |
1306 | EXPORT_SYMBOL_GPL(hrtimer_start_range_ns); |
1307 | ||
c0a31329 TG |
1308 | /** |
1309 | * hrtimer_try_to_cancel - try to deactivate a timer | |
c0a31329 TG |
1310 | * @timer: hrtimer to stop |
1311 | * | |
1312 | * Returns: | |
51633704 MCC |
1313 | * |
1314 | * * 0 when the timer was not active | |
1315 | * * 1 when the timer was active | |
1316 | * * -1 when the timer is currently executing the callback function and | |
fa9799e3 | 1317 | * cannot be stopped |
c0a31329 TG |
1318 | */ |
1319 | int hrtimer_try_to_cancel(struct hrtimer *timer) | |
1320 | { | |
3c8aa39d | 1321 | struct hrtimer_clock_base *base; |
c0a31329 TG |
1322 | unsigned long flags; |
1323 | int ret = -1; | |
1324 | ||
19d9f422 TG |
1325 | /* |
1326 | * Check lockless first. If the timer is not active (neither | |
1327 | * enqueued nor running the callback, nothing to do here. The | |
1328 | * base lock does not serialize against a concurrent enqueue, | |
1329 | * so we can avoid taking it. | |
1330 | */ | |
1331 | if (!hrtimer_active(timer)) | |
1332 | return 0; | |
1333 | ||
c0a31329 TG |
1334 | base = lock_hrtimer_base(timer, &flags); |
1335 | ||
303e967f | 1336 | if (!hrtimer_callback_running(timer)) |
627ef5ae | 1337 | ret = remove_hrtimer(timer, base, false, false); |
c0a31329 TG |
1338 | |
1339 | unlock_hrtimer_base(timer, &flags); | |
1340 | ||
1341 | return ret; | |
1342 | ||
1343 | } | |
8d16b764 | 1344 | EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel); |
c0a31329 | 1345 | |
f61eff83 AMG |
1346 | #ifdef CONFIG_PREEMPT_RT |
1347 | static void hrtimer_cpu_base_init_expiry_lock(struct hrtimer_cpu_base *base) | |
1348 | { | |
1349 | spin_lock_init(&base->softirq_expiry_lock); | |
1350 | } | |
1351 | ||
1352 | static void hrtimer_cpu_base_lock_expiry(struct hrtimer_cpu_base *base) | |
1353 | { | |
1354 | spin_lock(&base->softirq_expiry_lock); | |
1355 | } | |
1356 | ||
1357 | static void hrtimer_cpu_base_unlock_expiry(struct hrtimer_cpu_base *base) | |
1358 | { | |
1359 | spin_unlock(&base->softirq_expiry_lock); | |
1360 | } | |
1361 | ||
1362 | /* | |
1363 | * The counterpart to hrtimer_cancel_wait_running(). | |
1364 | * | |
1365 | * If there is a waiter for cpu_base->expiry_lock, then it was waiting for | |
4bf07f65 | 1366 | * the timer callback to finish. Drop expiry_lock and reacquire it. That |
f61eff83 AMG |
1367 | * allows the waiter to acquire the lock and make progress. |
1368 | */ | |
1369 | static void hrtimer_sync_wait_running(struct hrtimer_cpu_base *cpu_base, | |
1370 | unsigned long flags) | |
1371 | { | |
1372 | if (atomic_read(&cpu_base->timer_waiters)) { | |
1373 | raw_spin_unlock_irqrestore(&cpu_base->lock, flags); | |
1374 | spin_unlock(&cpu_base->softirq_expiry_lock); | |
1375 | spin_lock(&cpu_base->softirq_expiry_lock); | |
1376 | raw_spin_lock_irq(&cpu_base->lock); | |
1377 | } | |
1378 | } | |
1379 | ||
1380 | /* | |
1381 | * This function is called on PREEMPT_RT kernels when the fast path | |
1382 | * deletion of a timer failed because the timer callback function was | |
1383 | * running. | |
1384 | * | |
0bee3b60 FW |
1385 | * This prevents priority inversion: if the soft irq thread is preempted |
1386 | * in the middle of a timer callback, then calling del_timer_sync() can | |
1387 | * lead to two issues: | |
1388 | * | |
1389 | * - If the caller is on a remote CPU then it has to spin wait for the timer | |
1390 | * handler to complete. This can result in unbound priority inversion. | |
1391 | * | |
1392 | * - If the caller originates from the task which preempted the timer | |
1393 | * handler on the same CPU, then spin waiting for the timer handler to | |
1394 | * complete is never going to end. | |
f61eff83 AMG |
1395 | */ |
1396 | void hrtimer_cancel_wait_running(const struct hrtimer *timer) | |
1397 | { | |
dd2261ed JG |
1398 | /* Lockless read. Prevent the compiler from reloading it below */ |
1399 | struct hrtimer_clock_base *base = READ_ONCE(timer->base); | |
f61eff83 | 1400 | |
68b2c8c1 JG |
1401 | /* |
1402 | * Just relax if the timer expires in hard interrupt context or if | |
1403 | * it is currently on the migration base. | |
1404 | */ | |
5d2295f3 | 1405 | if (!timer->is_soft || is_migration_base(base)) { |
f61eff83 AMG |
1406 | cpu_relax(); |
1407 | return; | |
1408 | } | |
1409 | ||
1410 | /* | |
1411 | * Mark the base as contended and grab the expiry lock, which is | |
1412 | * held by the softirq across the timer callback. Drop the lock | |
1413 | * immediately so the softirq can expire the next timer. In theory | |
1414 | * the timer could already be running again, but that's more than | |
1415 | * unlikely and just causes another wait loop. | |
1416 | */ | |
1417 | atomic_inc(&base->cpu_base->timer_waiters); | |
1418 | spin_lock_bh(&base->cpu_base->softirq_expiry_lock); | |
1419 | atomic_dec(&base->cpu_base->timer_waiters); | |
1420 | spin_unlock_bh(&base->cpu_base->softirq_expiry_lock); | |
1421 | } | |
1422 | #else | |
1423 | static inline void | |
1424 | hrtimer_cpu_base_init_expiry_lock(struct hrtimer_cpu_base *base) { } | |
1425 | static inline void | |
1426 | hrtimer_cpu_base_lock_expiry(struct hrtimer_cpu_base *base) { } | |
1427 | static inline void | |
1428 | hrtimer_cpu_base_unlock_expiry(struct hrtimer_cpu_base *base) { } | |
1429 | static inline void hrtimer_sync_wait_running(struct hrtimer_cpu_base *base, | |
1430 | unsigned long flags) { } | |
1431 | #endif | |
1432 | ||
c0a31329 TG |
1433 | /** |
1434 | * hrtimer_cancel - cancel a timer and wait for the handler to finish. | |
c0a31329 TG |
1435 | * @timer: the timer to be cancelled |
1436 | * | |
1437 | * Returns: | |
1438 | * 0 when the timer was not active | |
1439 | * 1 when the timer was active | |
1440 | */ | |
1441 | int hrtimer_cancel(struct hrtimer *timer) | |
1442 | { | |
f61eff83 | 1443 | int ret; |
c0a31329 | 1444 | |
f61eff83 AMG |
1445 | do { |
1446 | ret = hrtimer_try_to_cancel(timer); | |
1447 | ||
1448 | if (ret < 0) | |
1449 | hrtimer_cancel_wait_running(timer); | |
1450 | } while (ret < 0); | |
1451 | return ret; | |
c0a31329 | 1452 | } |
8d16b764 | 1453 | EXPORT_SYMBOL_GPL(hrtimer_cancel); |
c0a31329 TG |
1454 | |
1455 | /** | |
66981c37 | 1456 | * __hrtimer_get_remaining - get remaining time for the timer |
c0a31329 | 1457 | * @timer: the timer to read |
203cbf77 | 1458 | * @adjust: adjust relative timers when CONFIG_TIME_LOW_RES=y |
c0a31329 | 1459 | */ |
203cbf77 | 1460 | ktime_t __hrtimer_get_remaining(const struct hrtimer *timer, bool adjust) |
c0a31329 | 1461 | { |
c0a31329 TG |
1462 | unsigned long flags; |
1463 | ktime_t rem; | |
1464 | ||
b3bd3de6 | 1465 | lock_hrtimer_base(timer, &flags); |
203cbf77 TG |
1466 | if (IS_ENABLED(CONFIG_TIME_LOW_RES) && adjust) |
1467 | rem = hrtimer_expires_remaining_adjusted(timer); | |
1468 | else | |
1469 | rem = hrtimer_expires_remaining(timer); | |
c0a31329 TG |
1470 | unlock_hrtimer_base(timer, &flags); |
1471 | ||
1472 | return rem; | |
1473 | } | |
203cbf77 | 1474 | EXPORT_SYMBOL_GPL(__hrtimer_get_remaining); |
c0a31329 | 1475 | |
3451d024 | 1476 | #ifdef CONFIG_NO_HZ_COMMON |
69239749 TL |
1477 | /** |
1478 | * hrtimer_get_next_event - get the time until next expiry event | |
1479 | * | |
c1ad348b | 1480 | * Returns the next expiry time or KTIME_MAX if no timer is pending. |
69239749 | 1481 | */ |
c1ad348b | 1482 | u64 hrtimer_get_next_event(void) |
69239749 | 1483 | { |
dc5df73b | 1484 | struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases); |
c1ad348b | 1485 | u64 expires = KTIME_MAX; |
69239749 | 1486 | unsigned long flags; |
69239749 | 1487 | |
ecb49d1a | 1488 | raw_spin_lock_irqsave(&cpu_base->lock, flags); |
3c8aa39d | 1489 | |
e19ffe8b | 1490 | if (!__hrtimer_hres_active(cpu_base)) |
5da70160 | 1491 | expires = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_ALL); |
3c8aa39d | 1492 | |
ecb49d1a | 1493 | raw_spin_unlock_irqrestore(&cpu_base->lock, flags); |
3c8aa39d | 1494 | |
c1ad348b | 1495 | return expires; |
69239749 | 1496 | } |
a59855cd RW |
1497 | |
1498 | /** | |
1499 | * hrtimer_next_event_without - time until next expiry event w/o one timer | |
1500 | * @exclude: timer to exclude | |
1501 | * | |
1502 | * Returns the next expiry time over all timers except for the @exclude one or | |
1503 | * KTIME_MAX if none of them is pending. | |
1504 | */ | |
1505 | u64 hrtimer_next_event_without(const struct hrtimer *exclude) | |
1506 | { | |
1507 | struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases); | |
1508 | u64 expires = KTIME_MAX; | |
1509 | unsigned long flags; | |
1510 | ||
1511 | raw_spin_lock_irqsave(&cpu_base->lock, flags); | |
1512 | ||
1513 | if (__hrtimer_hres_active(cpu_base)) { | |
1514 | unsigned int active; | |
1515 | ||
1516 | if (!cpu_base->softirq_activated) { | |
1517 | active = cpu_base->active_bases & HRTIMER_ACTIVE_SOFT; | |
1518 | expires = __hrtimer_next_event_base(cpu_base, exclude, | |
1519 | active, KTIME_MAX); | |
1520 | } | |
1521 | active = cpu_base->active_bases & HRTIMER_ACTIVE_HARD; | |
1522 | expires = __hrtimer_next_event_base(cpu_base, exclude, active, | |
1523 | expires); | |
1524 | } | |
1525 | ||
1526 | raw_spin_unlock_irqrestore(&cpu_base->lock, flags); | |
1527 | ||
1528 | return expires; | |
1529 | } | |
69239749 TL |
1530 | #endif |
1531 | ||
336a9cde MZ |
1532 | static inline int hrtimer_clockid_to_base(clockid_t clock_id) |
1533 | { | |
1534 | if (likely(clock_id < MAX_CLOCKS)) { | |
1535 | int base = hrtimer_clock_to_base_table[clock_id]; | |
1536 | ||
1537 | if (likely(base != HRTIMER_MAX_CLOCK_BASES)) | |
1538 | return base; | |
1539 | } | |
1540 | WARN(1, "Invalid clockid %d. Using MONOTONIC\n", clock_id); | |
1541 | return HRTIMER_BASE_MONOTONIC; | |
1542 | } | |
1543 | ||
237fc6e7 TG |
1544 | static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id, |
1545 | enum hrtimer_mode mode) | |
c0a31329 | 1546 | { |
42f42da4 | 1547 | bool softtimer = !!(mode & HRTIMER_MODE_SOFT); |
3c8aa39d | 1548 | struct hrtimer_cpu_base *cpu_base; |
f5c2f021 SAS |
1549 | int base; |
1550 | ||
1551 | /* | |
4bf07f65 | 1552 | * On PREEMPT_RT enabled kernels hrtimers which are not explicitly |
f5c2f021 SAS |
1553 | * marked for hard interrupt expiry mode are moved into soft |
1554 | * interrupt context for latency reasons and because the callbacks | |
1555 | * can invoke functions which might sleep on RT, e.g. spin_lock(). | |
1556 | */ | |
1557 | if (IS_ENABLED(CONFIG_PREEMPT_RT) && !(mode & HRTIMER_MODE_HARD)) | |
1558 | softtimer = true; | |
c0a31329 | 1559 | |
7978672c GA |
1560 | memset(timer, 0, sizeof(struct hrtimer)); |
1561 | ||
22127e93 | 1562 | cpu_base = raw_cpu_ptr(&hrtimer_bases); |
c0a31329 | 1563 | |
48d0c9be AMG |
1564 | /* |
1565 | * POSIX magic: Relative CLOCK_REALTIME timers are not affected by | |
1566 | * clock modifications, so they needs to become CLOCK_MONOTONIC to | |
1567 | * ensure POSIX compliance. | |
1568 | */ | |
1569 | if (clock_id == CLOCK_REALTIME && mode & HRTIMER_MODE_REL) | |
7978672c GA |
1570 | clock_id = CLOCK_MONOTONIC; |
1571 | ||
f5c2f021 | 1572 | base = softtimer ? HRTIMER_MAX_CLOCK_BASES / 2 : 0; |
42f42da4 AMG |
1573 | base += hrtimer_clockid_to_base(clock_id); |
1574 | timer->is_soft = softtimer; | |
40db1739 | 1575 | timer->is_hard = !!(mode & HRTIMER_MODE_HARD); |
e06383db | 1576 | timer->base = &cpu_base->clock_base[base]; |
998adc3d | 1577 | timerqueue_init(&timer->node); |
c0a31329 | 1578 | } |
237fc6e7 TG |
1579 | |
1580 | /** | |
1581 | * hrtimer_init - initialize a timer to the given clock | |
1582 | * @timer: the timer to be initialized | |
1583 | * @clock_id: the clock to be used | |
4bf07f65 | 1584 | * @mode: The modes which are relevant for initialization: |
42f42da4 AMG |
1585 | * HRTIMER_MODE_ABS, HRTIMER_MODE_REL, HRTIMER_MODE_ABS_SOFT, |
1586 | * HRTIMER_MODE_REL_SOFT | |
1587 | * | |
1588 | * The PINNED variants of the above can be handed in, | |
1589 | * but the PINNED bit is ignored as pinning happens | |
1590 | * when the hrtimer is started | |
237fc6e7 TG |
1591 | */ |
1592 | void hrtimer_init(struct hrtimer *timer, clockid_t clock_id, | |
1593 | enum hrtimer_mode mode) | |
1594 | { | |
c6a2a177 | 1595 | debug_init(timer, clock_id, mode); |
237fc6e7 TG |
1596 | __hrtimer_init(timer, clock_id, mode); |
1597 | } | |
8d16b764 | 1598 | EXPORT_SYMBOL_GPL(hrtimer_init); |
c0a31329 | 1599 | |
887d9dc9 PZ |
1600 | /* |
1601 | * A timer is active, when it is enqueued into the rbtree or the | |
1602 | * callback function is running or it's in the state of being migrated | |
1603 | * to another cpu. | |
c0a31329 | 1604 | * |
887d9dc9 | 1605 | * It is important for this function to not return a false negative. |
c0a31329 | 1606 | */ |
887d9dc9 | 1607 | bool hrtimer_active(const struct hrtimer *timer) |
c0a31329 | 1608 | { |
3f0b9e8e | 1609 | struct hrtimer_clock_base *base; |
887d9dc9 | 1610 | unsigned int seq; |
c0a31329 | 1611 | |
887d9dc9 | 1612 | do { |
3f0b9e8e AMG |
1613 | base = READ_ONCE(timer->base); |
1614 | seq = raw_read_seqcount_begin(&base->seq); | |
c0a31329 | 1615 | |
887d9dc9 | 1616 | if (timer->state != HRTIMER_STATE_INACTIVE || |
3f0b9e8e | 1617 | base->running == timer) |
887d9dc9 PZ |
1618 | return true; |
1619 | ||
3f0b9e8e AMG |
1620 | } while (read_seqcount_retry(&base->seq, seq) || |
1621 | base != READ_ONCE(timer->base)); | |
887d9dc9 PZ |
1622 | |
1623 | return false; | |
c0a31329 | 1624 | } |
887d9dc9 | 1625 | EXPORT_SYMBOL_GPL(hrtimer_active); |
c0a31329 | 1626 | |
887d9dc9 PZ |
1627 | /* |
1628 | * The write_seqcount_barrier()s in __run_hrtimer() split the thing into 3 | |
1629 | * distinct sections: | |
1630 | * | |
1631 | * - queued: the timer is queued | |
1632 | * - callback: the timer is being ran | |
1633 | * - post: the timer is inactive or (re)queued | |
1634 | * | |
1635 | * On the read side we ensure we observe timer->state and cpu_base->running | |
1636 | * from the same section, if anything changed while we looked at it, we retry. | |
1637 | * This includes timer->base changing because sequence numbers alone are | |
1638 | * insufficient for that. | |
1639 | * | |
1640 | * The sequence numbers are required because otherwise we could still observe | |
4bf07f65 | 1641 | * a false negative if the read side got smeared over multiple consecutive |
887d9dc9 PZ |
1642 | * __run_hrtimer() invocations. |
1643 | */ | |
1644 | ||
21d6d52a TG |
1645 | static void __run_hrtimer(struct hrtimer_cpu_base *cpu_base, |
1646 | struct hrtimer_clock_base *base, | |
dd934aa8 | 1647 | struct hrtimer *timer, ktime_t *now, |
eb5a4d0a | 1648 | unsigned long flags) __must_hold(&cpu_base->lock) |
d3d74453 | 1649 | { |
d3d74453 | 1650 | enum hrtimer_restart (*fn)(struct hrtimer *); |
73d20564 | 1651 | bool expires_in_hardirq; |
d3d74453 PZ |
1652 | int restart; |
1653 | ||
887d9dc9 | 1654 | lockdep_assert_held(&cpu_base->lock); |
ca109491 | 1655 | |
c6a2a177 | 1656 | debug_deactivate(timer); |
3f0b9e8e | 1657 | base->running = timer; |
887d9dc9 PZ |
1658 | |
1659 | /* | |
1660 | * Separate the ->running assignment from the ->state assignment. | |
1661 | * | |
1662 | * As with a regular write barrier, this ensures the read side in | |
3f0b9e8e | 1663 | * hrtimer_active() cannot observe base->running == NULL && |
887d9dc9 PZ |
1664 | * timer->state == INACTIVE. |
1665 | */ | |
3f0b9e8e | 1666 | raw_write_seqcount_barrier(&base->seq); |
887d9dc9 PZ |
1667 | |
1668 | __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE, 0); | |
d3d74453 | 1669 | fn = timer->function; |
ca109491 | 1670 | |
203cbf77 TG |
1671 | /* |
1672 | * Clear the 'is relative' flag for the TIME_LOW_RES case. If the | |
1673 | * timer is restarted with a period then it becomes an absolute | |
1674 | * timer. If its not restarted it does not matter. | |
1675 | */ | |
1676 | if (IS_ENABLED(CONFIG_TIME_LOW_RES)) | |
1677 | timer->is_rel = false; | |
1678 | ||
ca109491 | 1679 | /* |
d05ca13b TG |
1680 | * The timer is marked as running in the CPU base, so it is |
1681 | * protected against migration to a different CPU even if the lock | |
1682 | * is dropped. | |
ca109491 | 1683 | */ |
dd934aa8 | 1684 | raw_spin_unlock_irqrestore(&cpu_base->lock, flags); |
c6a2a177 | 1685 | trace_hrtimer_expire_entry(timer, now); |
73d20564 | 1686 | expires_in_hardirq = lockdep_hrtimer_enter(timer); |
40db1739 | 1687 | |
ca109491 | 1688 | restart = fn(timer); |
40db1739 | 1689 | |
73d20564 | 1690 | lockdep_hrtimer_exit(expires_in_hardirq); |
c6a2a177 | 1691 | trace_hrtimer_expire_exit(timer); |
dd934aa8 | 1692 | raw_spin_lock_irq(&cpu_base->lock); |
d3d74453 PZ |
1693 | |
1694 | /* | |
887d9dc9 | 1695 | * Note: We clear the running state after enqueue_hrtimer and |
b4d90e9f | 1696 | * we do not reprogram the event hardware. Happens either in |
e3f1d883 | 1697 | * hrtimer_start_range_ns() or in hrtimer_interrupt() |
5de2755c PZ |
1698 | * |
1699 | * Note: Because we dropped the cpu_base->lock above, | |
1700 | * hrtimer_start_range_ns() can have popped in and enqueued the timer | |
1701 | * for us already. | |
d3d74453 | 1702 | */ |
5de2755c PZ |
1703 | if (restart != HRTIMER_NORESTART && |
1704 | !(timer->state & HRTIMER_STATE_ENQUEUED)) | |
63e2ed36 | 1705 | enqueue_hrtimer(timer, base, HRTIMER_MODE_ABS); |
f13d4f97 | 1706 | |
887d9dc9 PZ |
1707 | /* |
1708 | * Separate the ->running assignment from the ->state assignment. | |
1709 | * | |
1710 | * As with a regular write barrier, this ensures the read side in | |
3f0b9e8e | 1711 | * hrtimer_active() cannot observe base->running.timer == NULL && |
887d9dc9 PZ |
1712 | * timer->state == INACTIVE. |
1713 | */ | |
3f0b9e8e | 1714 | raw_write_seqcount_barrier(&base->seq); |
f13d4f97 | 1715 | |
3f0b9e8e AMG |
1716 | WARN_ON_ONCE(base->running != timer); |
1717 | base->running = NULL; | |
d3d74453 PZ |
1718 | } |
1719 | ||
dd934aa8 | 1720 | static void __hrtimer_run_queues(struct hrtimer_cpu_base *cpu_base, ktime_t now, |
c458b1d1 | 1721 | unsigned long flags, unsigned int active_mask) |
54cdfdb4 | 1722 | { |
c272ca58 | 1723 | struct hrtimer_clock_base *base; |
c458b1d1 | 1724 | unsigned int active = cpu_base->active_bases & active_mask; |
6ff7041d | 1725 | |
c272ca58 | 1726 | for_each_active_base(base, cpu_base, active) { |
998adc3d | 1727 | struct timerqueue_node *node; |
ab8177bc TG |
1728 | ktime_t basenow; |
1729 | ||
54cdfdb4 TG |
1730 | basenow = ktime_add(now, base->offset); |
1731 | ||
998adc3d | 1732 | while ((node = timerqueue_getnext(&base->active))) { |
54cdfdb4 TG |
1733 | struct hrtimer *timer; |
1734 | ||
998adc3d | 1735 | timer = container_of(node, struct hrtimer, node); |
54cdfdb4 | 1736 | |
654c8e0b AV |
1737 | /* |
1738 | * The immediate goal for using the softexpires is | |
1739 | * minimizing wakeups, not running timers at the | |
1740 | * earliest interrupt after their soft expiration. | |
1741 | * This allows us to avoid using a Priority Search | |
4bf07f65 | 1742 | * Tree, which can answer a stabbing query for |
654c8e0b AV |
1743 | * overlapping intervals and instead use the simple |
1744 | * BST we already have. | |
1745 | * We don't add extra wakeups by delaying timers that | |
1746 | * are right-of a not yet expired timer, because that | |
1747 | * timer will have to trigger a wakeup anyway. | |
1748 | */ | |
2456e855 | 1749 | if (basenow < hrtimer_get_softexpires_tv64(timer)) |
54cdfdb4 | 1750 | break; |
54cdfdb4 | 1751 | |
dd934aa8 | 1752 | __run_hrtimer(cpu_base, base, timer, &basenow, flags); |
f61eff83 AMG |
1753 | if (active_mask == HRTIMER_ACTIVE_SOFT) |
1754 | hrtimer_sync_wait_running(cpu_base, flags); | |
54cdfdb4 | 1755 | } |
54cdfdb4 | 1756 | } |
21d6d52a TG |
1757 | } |
1758 | ||
5da70160 AMG |
1759 | static __latent_entropy void hrtimer_run_softirq(struct softirq_action *h) |
1760 | { | |
1761 | struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases); | |
1762 | unsigned long flags; | |
1763 | ktime_t now; | |
1764 | ||
f61eff83 | 1765 | hrtimer_cpu_base_lock_expiry(cpu_base); |
5da70160 AMG |
1766 | raw_spin_lock_irqsave(&cpu_base->lock, flags); |
1767 | ||
1768 | now = hrtimer_update_base(cpu_base); | |
1769 | __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_SOFT); | |
1770 | ||
1771 | cpu_base->softirq_activated = 0; | |
1772 | hrtimer_update_softirq_timer(cpu_base, true); | |
1773 | ||
1774 | raw_spin_unlock_irqrestore(&cpu_base->lock, flags); | |
f61eff83 | 1775 | hrtimer_cpu_base_unlock_expiry(cpu_base); |
5da70160 AMG |
1776 | } |
1777 | ||
21d6d52a TG |
1778 | #ifdef CONFIG_HIGH_RES_TIMERS |
1779 | ||
1780 | /* | |
1781 | * High resolution timer interrupt | |
1782 | * Called with interrupts disabled | |
1783 | */ | |
1784 | void hrtimer_interrupt(struct clock_event_device *dev) | |
1785 | { | |
1786 | struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases); | |
1787 | ktime_t expires_next, now, entry_time, delta; | |
dd934aa8 | 1788 | unsigned long flags; |
21d6d52a TG |
1789 | int retries = 0; |
1790 | ||
1791 | BUG_ON(!cpu_base->hres_active); | |
1792 | cpu_base->nr_events++; | |
2456e855 | 1793 | dev->next_event = KTIME_MAX; |
21d6d52a | 1794 | |
dd934aa8 | 1795 | raw_spin_lock_irqsave(&cpu_base->lock, flags); |
21d6d52a TG |
1796 | entry_time = now = hrtimer_update_base(cpu_base); |
1797 | retry: | |
1798 | cpu_base->in_hrtirq = 1; | |
1799 | /* | |
1800 | * We set expires_next to KTIME_MAX here with cpu_base->lock | |
1801 | * held to prevent that a timer is enqueued in our queue via | |
1802 | * the migration code. This does not affect enqueueing of | |
1803 | * timers which run their callback and need to be requeued on | |
1804 | * this CPU. | |
1805 | */ | |
2456e855 | 1806 | cpu_base->expires_next = KTIME_MAX; |
21d6d52a | 1807 | |
5da70160 AMG |
1808 | if (!ktime_before(now, cpu_base->softirq_expires_next)) { |
1809 | cpu_base->softirq_expires_next = KTIME_MAX; | |
1810 | cpu_base->softirq_activated = 1; | |
1811 | raise_softirq_irqoff(HRTIMER_SOFTIRQ); | |
1812 | } | |
1813 | ||
c458b1d1 | 1814 | __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_HARD); |
21d6d52a | 1815 | |
46eb1701 AMB |
1816 | /* Reevaluate the clock bases for the [soft] next expiry */ |
1817 | expires_next = hrtimer_update_next_event(cpu_base); | |
6ff7041d TG |
1818 | /* |
1819 | * Store the new expiry value so the migration code can verify | |
1820 | * against it. | |
1821 | */ | |
54cdfdb4 | 1822 | cpu_base->expires_next = expires_next; |
9bc74919 | 1823 | cpu_base->in_hrtirq = 0; |
dd934aa8 | 1824 | raw_spin_unlock_irqrestore(&cpu_base->lock, flags); |
54cdfdb4 TG |
1825 | |
1826 | /* Reprogramming necessary ? */ | |
d2540875 | 1827 | if (!tick_program_event(expires_next, 0)) { |
41d2e494 TG |
1828 | cpu_base->hang_detected = 0; |
1829 | return; | |
54cdfdb4 | 1830 | } |
41d2e494 TG |
1831 | |
1832 | /* | |
1833 | * The next timer was already expired due to: | |
1834 | * - tracing | |
1835 | * - long lasting callbacks | |
1836 | * - being scheduled away when running in a VM | |
1837 | * | |
1838 | * We need to prevent that we loop forever in the hrtimer | |
1839 | * interrupt routine. We give it 3 attempts to avoid | |
1840 | * overreacting on some spurious event. | |
5baefd6d JS |
1841 | * |
1842 | * Acquire base lock for updating the offsets and retrieving | |
1843 | * the current time. | |
41d2e494 | 1844 | */ |
dd934aa8 | 1845 | raw_spin_lock_irqsave(&cpu_base->lock, flags); |
5baefd6d | 1846 | now = hrtimer_update_base(cpu_base); |
41d2e494 TG |
1847 | cpu_base->nr_retries++; |
1848 | if (++retries < 3) | |
1849 | goto retry; | |
1850 | /* | |
1851 | * Give the system a chance to do something else than looping | |
1852 | * here. We stored the entry time, so we know exactly how long | |
1853 | * we spent here. We schedule the next event this amount of | |
1854 | * time away. | |
1855 | */ | |
1856 | cpu_base->nr_hangs++; | |
1857 | cpu_base->hang_detected = 1; | |
dd934aa8 AMG |
1858 | raw_spin_unlock_irqrestore(&cpu_base->lock, flags); |
1859 | ||
41d2e494 | 1860 | delta = ktime_sub(now, entry_time); |
2456e855 TG |
1861 | if ((unsigned int)delta > cpu_base->max_hang_time) |
1862 | cpu_base->max_hang_time = (unsigned int) delta; | |
41d2e494 TG |
1863 | /* |
1864 | * Limit it to a sensible value as we enforce a longer | |
1865 | * delay. Give the CPU at least 100ms to catch up. | |
1866 | */ | |
2456e855 | 1867 | if (delta > 100 * NSEC_PER_MSEC) |
41d2e494 TG |
1868 | expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC); |
1869 | else | |
1870 | expires_next = ktime_add(now, delta); | |
1871 | tick_program_event(expires_next, 1); | |
7a6e5537 | 1872 | pr_warn_once("hrtimer: interrupt took %llu ns\n", ktime_to_ns(delta)); |
54cdfdb4 TG |
1873 | } |
1874 | ||
016da201 | 1875 | /* called with interrupts disabled */ |
c6eb3f70 | 1876 | static inline void __hrtimer_peek_ahead_timers(void) |
8bdec955 TG |
1877 | { |
1878 | struct tick_device *td; | |
1879 | ||
1880 | if (!hrtimer_hres_active()) | |
1881 | return; | |
1882 | ||
22127e93 | 1883 | td = this_cpu_ptr(&tick_cpu_device); |
8bdec955 TG |
1884 | if (td && td->evtdev) |
1885 | hrtimer_interrupt(td->evtdev); | |
1886 | } | |
1887 | ||
82c5b7b5 IM |
1888 | #else /* CONFIG_HIGH_RES_TIMERS */ |
1889 | ||
1890 | static inline void __hrtimer_peek_ahead_timers(void) { } | |
1891 | ||
1892 | #endif /* !CONFIG_HIGH_RES_TIMERS */ | |
82f67cd9 | 1893 | |
d3d74453 | 1894 | /* |
c6eb3f70 | 1895 | * Called from run_local_timers in hardirq context every jiffy |
d3d74453 | 1896 | */ |
833883d9 | 1897 | void hrtimer_run_queues(void) |
d3d74453 | 1898 | { |
dc5df73b | 1899 | struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases); |
dd934aa8 | 1900 | unsigned long flags; |
21d6d52a | 1901 | ktime_t now; |
c0a31329 | 1902 | |
e19ffe8b | 1903 | if (__hrtimer_hres_active(cpu_base)) |
d3d74453 | 1904 | return; |
54cdfdb4 | 1905 | |
d3d74453 | 1906 | /* |
c6eb3f70 TG |
1907 | * This _is_ ugly: We have to check periodically, whether we |
1908 | * can switch to highres and / or nohz mode. The clocksource | |
1909 | * switch happens with xtime_lock held. Notification from | |
1910 | * there only sets the check bit in the tick_oneshot code, | |
1911 | * otherwise we might deadlock vs. xtime_lock. | |
d3d74453 | 1912 | */ |
c6eb3f70 | 1913 | if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) { |
d3d74453 | 1914 | hrtimer_switch_to_hres(); |
3055adda | 1915 | return; |
833883d9 | 1916 | } |
c6eb3f70 | 1917 | |
dd934aa8 | 1918 | raw_spin_lock_irqsave(&cpu_base->lock, flags); |
21d6d52a | 1919 | now = hrtimer_update_base(cpu_base); |
5da70160 AMG |
1920 | |
1921 | if (!ktime_before(now, cpu_base->softirq_expires_next)) { | |
1922 | cpu_base->softirq_expires_next = KTIME_MAX; | |
1923 | cpu_base->softirq_activated = 1; | |
1924 | raise_softirq_irqoff(HRTIMER_SOFTIRQ); | |
1925 | } | |
1926 | ||
c458b1d1 | 1927 | __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_HARD); |
dd934aa8 | 1928 | raw_spin_unlock_irqrestore(&cpu_base->lock, flags); |
c0a31329 TG |
1929 | } |
1930 | ||
10c94ec1 TG |
1931 | /* |
1932 | * Sleep related functions: | |
1933 | */ | |
c9cb2e3d | 1934 | static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer) |
00362e33 TG |
1935 | { |
1936 | struct hrtimer_sleeper *t = | |
1937 | container_of(timer, struct hrtimer_sleeper, timer); | |
1938 | struct task_struct *task = t->task; | |
1939 | ||
1940 | t->task = NULL; | |
1941 | if (task) | |
1942 | wake_up_process(task); | |
1943 | ||
1944 | return HRTIMER_NORESTART; | |
1945 | } | |
1946 | ||
01656464 TG |
1947 | /** |
1948 | * hrtimer_sleeper_start_expires - Start a hrtimer sleeper timer | |
1949 | * @sl: sleeper to be started | |
1950 | * @mode: timer mode abs/rel | |
1951 | * | |
1952 | * Wrapper around hrtimer_start_expires() for hrtimer_sleeper based timers | |
1953 | * to allow PREEMPT_RT to tweak the delivery mode (soft/hardirq context) | |
1954 | */ | |
1955 | void hrtimer_sleeper_start_expires(struct hrtimer_sleeper *sl, | |
1956 | enum hrtimer_mode mode) | |
1957 | { | |
1842f5a4 SAS |
1958 | /* |
1959 | * Make the enqueue delivery mode check work on RT. If the sleeper | |
1960 | * was initialized for hard interrupt delivery, force the mode bit. | |
1961 | * This is a special case for hrtimer_sleepers because | |
1962 | * hrtimer_init_sleeper() determines the delivery mode on RT so the | |
1963 | * fiddling with this decision is avoided at the call sites. | |
1964 | */ | |
1965 | if (IS_ENABLED(CONFIG_PREEMPT_RT) && sl->timer.is_hard) | |
1966 | mode |= HRTIMER_MODE_HARD; | |
1967 | ||
01656464 TG |
1968 | hrtimer_start_expires(&sl->timer, mode); |
1969 | } | |
1970 | EXPORT_SYMBOL_GPL(hrtimer_sleeper_start_expires); | |
1971 | ||
dbc1625f SAS |
1972 | static void __hrtimer_init_sleeper(struct hrtimer_sleeper *sl, |
1973 | clockid_t clock_id, enum hrtimer_mode mode) | |
00362e33 | 1974 | { |
1842f5a4 | 1975 | /* |
4bf07f65 | 1976 | * On PREEMPT_RT enabled kernels hrtimers which are not explicitly |
1842f5a4 SAS |
1977 | * marked for hard interrupt expiry mode are moved into soft |
1978 | * interrupt context either for latency reasons or because the | |
1979 | * hrtimer callback takes regular spinlocks or invokes other | |
1980 | * functions which are not suitable for hard interrupt context on | |
1981 | * PREEMPT_RT. | |
1982 | * | |
1983 | * The hrtimer_sleeper callback is RT compatible in hard interrupt | |
1984 | * context, but there is a latency concern: Untrusted userspace can | |
1985 | * spawn many threads which arm timers for the same expiry time on | |
1986 | * the same CPU. That causes a latency spike due to the wakeup of | |
1987 | * a gazillion threads. | |
1988 | * | |
4bf07f65 | 1989 | * OTOH, privileged real-time user space applications rely on the |
1842f5a4 SAS |
1990 | * low latency of hard interrupt wakeups. If the current task is in |
1991 | * a real-time scheduling class, mark the mode for hard interrupt | |
1992 | * expiry. | |
1993 | */ | |
1994 | if (IS_ENABLED(CONFIG_PREEMPT_RT)) { | |
1995 | if (task_is_realtime(current) && !(mode & HRTIMER_MODE_SOFT)) | |
1996 | mode |= HRTIMER_MODE_HARD; | |
1997 | } | |
1998 | ||
dbc1625f | 1999 | __hrtimer_init(&sl->timer, clock_id, mode); |
00362e33 | 2000 | sl->timer.function = hrtimer_wakeup; |
b7449487 | 2001 | sl->task = current; |
00362e33 | 2002 | } |
dbc1625f SAS |
2003 | |
2004 | /** | |
2005 | * hrtimer_init_sleeper - initialize sleeper to the given clock | |
2006 | * @sl: sleeper to be initialized | |
2007 | * @clock_id: the clock to be used | |
2008 | * @mode: timer mode abs/rel | |
2009 | */ | |
2010 | void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, clockid_t clock_id, | |
2011 | enum hrtimer_mode mode) | |
2012 | { | |
2013 | debug_init(&sl->timer, clock_id, mode); | |
2014 | __hrtimer_init_sleeper(sl, clock_id, mode); | |
2015 | ||
2016 | } | |
2bc481cf | 2017 | EXPORT_SYMBOL_GPL(hrtimer_init_sleeper); |
00362e33 | 2018 | |
c0edd7c9 | 2019 | int nanosleep_copyout(struct restart_block *restart, struct timespec64 *ts) |
ce41aaf4 AV |
2020 | { |
2021 | switch(restart->nanosleep.type) { | |
0fe27955 | 2022 | #ifdef CONFIG_COMPAT_32BIT_TIME |
ce41aaf4 | 2023 | case TT_COMPAT: |
9afc5eee | 2024 | if (put_old_timespec32(ts, restart->nanosleep.compat_rmtp)) |
ce41aaf4 AV |
2025 | return -EFAULT; |
2026 | break; | |
2027 | #endif | |
2028 | case TT_NATIVE: | |
c0edd7c9 | 2029 | if (put_timespec64(ts, restart->nanosleep.rmtp)) |
ce41aaf4 AV |
2030 | return -EFAULT; |
2031 | break; | |
2032 | default: | |
2033 | BUG(); | |
2034 | } | |
2035 | return -ERESTART_RESTARTBLOCK; | |
2036 | } | |
2037 | ||
669d7868 | 2038 | static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode) |
432569bb | 2039 | { |
edbeda46 AV |
2040 | struct restart_block *restart; |
2041 | ||
432569bb | 2042 | do { |
f5d39b02 | 2043 | set_current_state(TASK_INTERRUPTIBLE|TASK_FREEZABLE); |
01656464 | 2044 | hrtimer_sleeper_start_expires(t, mode); |
432569bb | 2045 | |
54cdfdb4 | 2046 | if (likely(t->task)) |
f5d39b02 | 2047 | schedule(); |
432569bb | 2048 | |
669d7868 | 2049 | hrtimer_cancel(&t->timer); |
c9cb2e3d | 2050 | mode = HRTIMER_MODE_ABS; |
669d7868 TG |
2051 | |
2052 | } while (t->task && !signal_pending(current)); | |
432569bb | 2053 | |
3588a085 PZ |
2054 | __set_current_state(TASK_RUNNING); |
2055 | ||
a7602681 | 2056 | if (!t->task) |
080344b9 | 2057 | return 0; |
080344b9 | 2058 | |
edbeda46 AV |
2059 | restart = ¤t->restart_block; |
2060 | if (restart->nanosleep.type != TT_NONE) { | |
a7602681 | 2061 | ktime_t rem = hrtimer_expires_remaining(&t->timer); |
c0edd7c9 | 2062 | struct timespec64 rmt; |
edbeda46 | 2063 | |
a7602681 AV |
2064 | if (rem <= 0) |
2065 | return 0; | |
c0edd7c9 | 2066 | rmt = ktime_to_timespec64(rem); |
a7602681 | 2067 | |
ce41aaf4 | 2068 | return nanosleep_copyout(restart, &rmt); |
a7602681 AV |
2069 | } |
2070 | return -ERESTART_RESTARTBLOCK; | |
080344b9 ON |
2071 | } |
2072 | ||
fb923c4a | 2073 | static long __sched hrtimer_nanosleep_restart(struct restart_block *restart) |
10c94ec1 | 2074 | { |
669d7868 | 2075 | struct hrtimer_sleeper t; |
a7602681 | 2076 | int ret; |
10c94ec1 | 2077 | |
dbc1625f SAS |
2078 | hrtimer_init_sleeper_on_stack(&t, restart->nanosleep.clockid, |
2079 | HRTIMER_MODE_ABS); | |
cc584b21 | 2080 | hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires); |
a7602681 | 2081 | ret = do_nanosleep(&t, HRTIMER_MODE_ABS); |
237fc6e7 TG |
2082 | destroy_hrtimer_on_stack(&t.timer); |
2083 | return ret; | |
10c94ec1 TG |
2084 | } |
2085 | ||
ea2d1f7f AV |
2086 | long hrtimer_nanosleep(ktime_t rqtp, const enum hrtimer_mode mode, |
2087 | const clockid_t clockid) | |
10c94ec1 | 2088 | { |
a7602681 | 2089 | struct restart_block *restart; |
669d7868 | 2090 | struct hrtimer_sleeper t; |
237fc6e7 | 2091 | int ret = 0; |
da8b44d5 | 2092 | u64 slack; |
3bd01206 AV |
2093 | |
2094 | slack = current->timer_slack_ns; | |
c14fd3dc | 2095 | if (rt_task(current)) |
3bd01206 | 2096 | slack = 0; |
10c94ec1 | 2097 | |
dbc1625f | 2098 | hrtimer_init_sleeper_on_stack(&t, clockid, mode); |
ea2d1f7f | 2099 | hrtimer_set_expires_range_ns(&t.timer, rqtp, slack); |
a7602681 AV |
2100 | ret = do_nanosleep(&t, mode); |
2101 | if (ret != -ERESTART_RESTARTBLOCK) | |
237fc6e7 | 2102 | goto out; |
10c94ec1 | 2103 | |
7978672c | 2104 | /* Absolute timers do not update the rmtp value and restart: */ |
237fc6e7 TG |
2105 | if (mode == HRTIMER_MODE_ABS) { |
2106 | ret = -ERESTARTNOHAND; | |
2107 | goto out; | |
2108 | } | |
10c94ec1 | 2109 | |
a7602681 | 2110 | restart = ¤t->restart_block; |
ab8177bc | 2111 | restart->nanosleep.clockid = t.timer.base->clockid; |
cc584b21 | 2112 | restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer); |
5abbe51a | 2113 | set_restart_fn(restart, hrtimer_nanosleep_restart); |
237fc6e7 TG |
2114 | out: |
2115 | destroy_hrtimer_on_stack(&t.timer); | |
2116 | return ret; | |
10c94ec1 TG |
2117 | } |
2118 | ||
3ca47e95 | 2119 | #ifdef CONFIG_64BIT |
01909974 DD |
2120 | |
2121 | SYSCALL_DEFINE2(nanosleep, struct __kernel_timespec __user *, rqtp, | |
2122 | struct __kernel_timespec __user *, rmtp) | |
6ba1b912 | 2123 | { |
c0edd7c9 | 2124 | struct timespec64 tu; |
6ba1b912 | 2125 | |
c0edd7c9 | 2126 | if (get_timespec64(&tu, rqtp)) |
6ba1b912 TG |
2127 | return -EFAULT; |
2128 | ||
c0edd7c9 | 2129 | if (!timespec64_valid(&tu)) |
6ba1b912 TG |
2130 | return -EINVAL; |
2131 | ||
9f76d591 | 2132 | current->restart_block.fn = do_no_restart_syscall; |
edbeda46 | 2133 | current->restart_block.nanosleep.type = rmtp ? TT_NATIVE : TT_NONE; |
192a82f9 | 2134 | current->restart_block.nanosleep.rmtp = rmtp; |
ea2d1f7f AV |
2135 | return hrtimer_nanosleep(timespec64_to_ktime(tu), HRTIMER_MODE_REL, |
2136 | CLOCK_MONOTONIC); | |
6ba1b912 TG |
2137 | } |
2138 | ||
01909974 DD |
2139 | #endif |
2140 | ||
b5793b0d | 2141 | #ifdef CONFIG_COMPAT_32BIT_TIME |
edbeda46 | 2142 | |
8dabe724 | 2143 | SYSCALL_DEFINE2(nanosleep_time32, struct old_timespec32 __user *, rqtp, |
9afc5eee | 2144 | struct old_timespec32 __user *, rmtp) |
edbeda46 | 2145 | { |
c0edd7c9 | 2146 | struct timespec64 tu; |
edbeda46 | 2147 | |
9afc5eee | 2148 | if (get_old_timespec32(&tu, rqtp)) |
edbeda46 AV |
2149 | return -EFAULT; |
2150 | ||
c0edd7c9 | 2151 | if (!timespec64_valid(&tu)) |
edbeda46 AV |
2152 | return -EINVAL; |
2153 | ||
9f76d591 | 2154 | current->restart_block.fn = do_no_restart_syscall; |
edbeda46 AV |
2155 | current->restart_block.nanosleep.type = rmtp ? TT_COMPAT : TT_NONE; |
2156 | current->restart_block.nanosleep.compat_rmtp = rmtp; | |
ea2d1f7f AV |
2157 | return hrtimer_nanosleep(timespec64_to_ktime(tu), HRTIMER_MODE_REL, |
2158 | CLOCK_MONOTONIC); | |
edbeda46 AV |
2159 | } |
2160 | #endif | |
2161 | ||
c0a31329 TG |
2162 | /* |
2163 | * Functions related to boot-time initialization: | |
2164 | */ | |
27590dc1 | 2165 | int hrtimers_prepare_cpu(unsigned int cpu) |
c0a31329 | 2166 | { |
3c8aa39d | 2167 | struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu); |
c0a31329 TG |
2168 | int i; |
2169 | ||
998adc3d | 2170 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
af5a06b5 AD |
2171 | struct hrtimer_clock_base *clock_b = &cpu_base->clock_base[i]; |
2172 | ||
2173 | clock_b->cpu_base = cpu_base; | |
2174 | seqcount_raw_spinlock_init(&clock_b->seq, &cpu_base->lock); | |
2175 | timerqueue_init_head(&clock_b->active); | |
998adc3d | 2176 | } |
3c8aa39d | 2177 | |
cddd0248 | 2178 | cpu_base->cpu = cpu; |
303c146d | 2179 | cpu_base->active_bases = 0; |
28bfd18b | 2180 | cpu_base->hres_active = 0; |
303c146d TG |
2181 | cpu_base->hang_detected = 0; |
2182 | cpu_base->next_timer = NULL; | |
2183 | cpu_base->softirq_next_timer = NULL; | |
07a9a7ea | 2184 | cpu_base->expires_next = KTIME_MAX; |
5da70160 | 2185 | cpu_base->softirq_expires_next = KTIME_MAX; |
f61eff83 | 2186 | hrtimer_cpu_base_init_expiry_lock(cpu_base); |
27590dc1 | 2187 | return 0; |
c0a31329 TG |
2188 | } |
2189 | ||
2190 | #ifdef CONFIG_HOTPLUG_CPU | |
2191 | ||
ca109491 | 2192 | static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base, |
37810659 | 2193 | struct hrtimer_clock_base *new_base) |
c0a31329 TG |
2194 | { |
2195 | struct hrtimer *timer; | |
998adc3d | 2196 | struct timerqueue_node *node; |
c0a31329 | 2197 | |
998adc3d JS |
2198 | while ((node = timerqueue_getnext(&old_base->active))) { |
2199 | timer = container_of(node, struct hrtimer, node); | |
54cdfdb4 | 2200 | BUG_ON(hrtimer_callback_running(timer)); |
c6a2a177 | 2201 | debug_deactivate(timer); |
b00c1a99 TG |
2202 | |
2203 | /* | |
c04dca02 | 2204 | * Mark it as ENQUEUED not INACTIVE otherwise the |
b00c1a99 TG |
2205 | * timer could be seen as !active and just vanish away |
2206 | * under us on another CPU | |
2207 | */ | |
c04dca02 | 2208 | __remove_hrtimer(timer, old_base, HRTIMER_STATE_ENQUEUED, 0); |
c0a31329 | 2209 | timer->base = new_base; |
54cdfdb4 | 2210 | /* |
e3f1d883 TG |
2211 | * Enqueue the timers on the new cpu. This does not |
2212 | * reprogram the event device in case the timer | |
2213 | * expires before the earliest on this CPU, but we run | |
2214 | * hrtimer_interrupt after we migrated everything to | |
2215 | * sort out already expired timers and reprogram the | |
2216 | * event device. | |
54cdfdb4 | 2217 | */ |
63e2ed36 | 2218 | enqueue_hrtimer(timer, new_base, HRTIMER_MODE_ABS); |
c0a31329 TG |
2219 | } |
2220 | } | |
2221 | ||
27590dc1 | 2222 | int hrtimers_dead_cpu(unsigned int scpu) |
c0a31329 | 2223 | { |
3c8aa39d | 2224 | struct hrtimer_cpu_base *old_base, *new_base; |
731a55ba | 2225 | int i; |
c0a31329 | 2226 | |
37810659 | 2227 | BUG_ON(cpu_online(scpu)); |
37810659 | 2228 | tick_cancel_sched_timer(scpu); |
731a55ba | 2229 | |
5da70160 AMG |
2230 | /* |
2231 | * this BH disable ensures that raise_softirq_irqoff() does | |
2232 | * not wakeup ksoftirqd (and acquire the pi-lock) while | |
2233 | * holding the cpu_base lock | |
2234 | */ | |
2235 | local_bh_disable(); | |
731a55ba TG |
2236 | local_irq_disable(); |
2237 | old_base = &per_cpu(hrtimer_bases, scpu); | |
dc5df73b | 2238 | new_base = this_cpu_ptr(&hrtimer_bases); |
d82f0b0f ON |
2239 | /* |
2240 | * The caller is globally serialized and nobody else | |
2241 | * takes two locks at once, deadlock is not possible. | |
2242 | */ | |
ecb49d1a TG |
2243 | raw_spin_lock(&new_base->lock); |
2244 | raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING); | |
c0a31329 | 2245 | |
3c8aa39d | 2246 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
ca109491 | 2247 | migrate_hrtimer_list(&old_base->clock_base[i], |
37810659 | 2248 | &new_base->clock_base[i]); |
c0a31329 TG |
2249 | } |
2250 | ||
5da70160 AMG |
2251 | /* |
2252 | * The migration might have changed the first expiring softirq | |
2253 | * timer on this CPU. Update it. | |
2254 | */ | |
2255 | hrtimer_update_softirq_timer(new_base, false); | |
2256 | ||
ecb49d1a TG |
2257 | raw_spin_unlock(&old_base->lock); |
2258 | raw_spin_unlock(&new_base->lock); | |
37810659 | 2259 | |
731a55ba TG |
2260 | /* Check, if we got expired work to do */ |
2261 | __hrtimer_peek_ahead_timers(); | |
2262 | local_irq_enable(); | |
5da70160 | 2263 | local_bh_enable(); |
27590dc1 | 2264 | return 0; |
c0a31329 | 2265 | } |
37810659 | 2266 | |
c0a31329 TG |
2267 | #endif /* CONFIG_HOTPLUG_CPU */ |
2268 | ||
c0a31329 TG |
2269 | void __init hrtimers_init(void) |
2270 | { | |
27590dc1 | 2271 | hrtimers_prepare_cpu(smp_processor_id()); |
5da70160 | 2272 | open_softirq(HRTIMER_SOFTIRQ, hrtimer_run_softirq); |
c0a31329 TG |
2273 | } |
2274 | ||
7bb67439 | 2275 | /** |
351b3f7a | 2276 | * schedule_hrtimeout_range_clock - sleep until timeout |
7bb67439 | 2277 | * @expires: timeout value (ktime_t) |
0c52310f | 2278 | * @delta: slack in expires timeout (ktime_t) for SCHED_OTHER tasks |
90777713 AMG |
2279 | * @mode: timer mode |
2280 | * @clock_id: timer clock to be used | |
7bb67439 | 2281 | */ |
351b3f7a | 2282 | int __sched |
da8b44d5 | 2283 | schedule_hrtimeout_range_clock(ktime_t *expires, u64 delta, |
90777713 | 2284 | const enum hrtimer_mode mode, clockid_t clock_id) |
7bb67439 AV |
2285 | { |
2286 | struct hrtimer_sleeper t; | |
2287 | ||
2288 | /* | |
2289 | * Optimize when a zero timeout value is given. It does not | |
2290 | * matter whether this is an absolute or a relative time. | |
2291 | */ | |
2456e855 | 2292 | if (expires && *expires == 0) { |
7bb67439 AV |
2293 | __set_current_state(TASK_RUNNING); |
2294 | return 0; | |
2295 | } | |
2296 | ||
2297 | /* | |
43b21013 | 2298 | * A NULL parameter means "infinite" |
7bb67439 AV |
2299 | */ |
2300 | if (!expires) { | |
2301 | schedule(); | |
7bb67439 AV |
2302 | return -EINTR; |
2303 | } | |
2304 | ||
0c52310f DB |
2305 | /* |
2306 | * Override any slack passed by the user if under | |
2307 | * rt contraints. | |
2308 | */ | |
2309 | if (rt_task(current)) | |
2310 | delta = 0; | |
2311 | ||
dbc1625f | 2312 | hrtimer_init_sleeper_on_stack(&t, clock_id, mode); |
654c8e0b | 2313 | hrtimer_set_expires_range_ns(&t.timer, *expires, delta); |
01656464 | 2314 | hrtimer_sleeper_start_expires(&t, mode); |
7bb67439 AV |
2315 | |
2316 | if (likely(t.task)) | |
2317 | schedule(); | |
2318 | ||
2319 | hrtimer_cancel(&t.timer); | |
2320 | destroy_hrtimer_on_stack(&t.timer); | |
2321 | ||
2322 | __set_current_state(TASK_RUNNING); | |
2323 | ||
2324 | return !t.task ? 0 : -EINTR; | |
2325 | } | |
151c8e49 | 2326 | EXPORT_SYMBOL_GPL(schedule_hrtimeout_range_clock); |
351b3f7a CE |
2327 | |
2328 | /** | |
2329 | * schedule_hrtimeout_range - sleep until timeout | |
2330 | * @expires: timeout value (ktime_t) | |
0c52310f | 2331 | * @delta: slack in expires timeout (ktime_t) for SCHED_OTHER tasks |
90777713 | 2332 | * @mode: timer mode |
351b3f7a CE |
2333 | * |
2334 | * Make the current task sleep until the given expiry time has | |
2335 | * elapsed. The routine will return immediately unless | |
2336 | * the current task state has been set (see set_current_state()). | |
2337 | * | |
2338 | * The @delta argument gives the kernel the freedom to schedule the | |
0c52310f DB |
2339 | * actual wakeup to a time that is both power and performance friendly |
2340 | * for regular (non RT/DL) tasks. | |
351b3f7a CE |
2341 | * The kernel give the normal best effort behavior for "@expires+@delta", |
2342 | * but may decide to fire the timer earlier, but no earlier than @expires. | |
2343 | * | |
2344 | * You can set the task state as follows - | |
2345 | * | |
2346 | * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to | |
4b7e9cf9 DA |
2347 | * pass before the routine returns unless the current task is explicitly |
2348 | * woken up, (e.g. by wake_up_process()). | |
351b3f7a CE |
2349 | * |
2350 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
4b7e9cf9 DA |
2351 | * delivered to the current task or the current task is explicitly woken |
2352 | * up. | |
351b3f7a CE |
2353 | * |
2354 | * The current task state is guaranteed to be TASK_RUNNING when this | |
2355 | * routine returns. | |
2356 | * | |
4b7e9cf9 DA |
2357 | * Returns 0 when the timer has expired. If the task was woken before the |
2358 | * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or | |
2359 | * by an explicit wakeup, it returns -EINTR. | |
351b3f7a | 2360 | */ |
da8b44d5 | 2361 | int __sched schedule_hrtimeout_range(ktime_t *expires, u64 delta, |
351b3f7a CE |
2362 | const enum hrtimer_mode mode) |
2363 | { | |
2364 | return schedule_hrtimeout_range_clock(expires, delta, mode, | |
2365 | CLOCK_MONOTONIC); | |
2366 | } | |
654c8e0b AV |
2367 | EXPORT_SYMBOL_GPL(schedule_hrtimeout_range); |
2368 | ||
2369 | /** | |
2370 | * schedule_hrtimeout - sleep until timeout | |
2371 | * @expires: timeout value (ktime_t) | |
90777713 | 2372 | * @mode: timer mode |
654c8e0b AV |
2373 | * |
2374 | * Make the current task sleep until the given expiry time has | |
2375 | * elapsed. The routine will return immediately unless | |
2376 | * the current task state has been set (see set_current_state()). | |
2377 | * | |
2378 | * You can set the task state as follows - | |
2379 | * | |
2380 | * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to | |
4b7e9cf9 DA |
2381 | * pass before the routine returns unless the current task is explicitly |
2382 | * woken up, (e.g. by wake_up_process()). | |
654c8e0b AV |
2383 | * |
2384 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
4b7e9cf9 DA |
2385 | * delivered to the current task or the current task is explicitly woken |
2386 | * up. | |
654c8e0b AV |
2387 | * |
2388 | * The current task state is guaranteed to be TASK_RUNNING when this | |
2389 | * routine returns. | |
2390 | * | |
4b7e9cf9 DA |
2391 | * Returns 0 when the timer has expired. If the task was woken before the |
2392 | * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or | |
2393 | * by an explicit wakeup, it returns -EINTR. | |
654c8e0b AV |
2394 | */ |
2395 | int __sched schedule_hrtimeout(ktime_t *expires, | |
2396 | const enum hrtimer_mode mode) | |
2397 | { | |
2398 | return schedule_hrtimeout_range(expires, 0, mode); | |
2399 | } | |
7bb67439 | 2400 | EXPORT_SYMBOL_GPL(schedule_hrtimeout); |