]>
Commit | Line | Data |
---|---|---|
c0a31329 TG |
1 | /* |
2 | * linux/kernel/hrtimer.c | |
3 | * | |
3c8aa39d | 4 | * Copyright(C) 2005-2006, Thomas Gleixner <tglx@linutronix.de> |
79bf2bb3 | 5 | * Copyright(C) 2005-2007, Red Hat, Inc., Ingo Molnar |
54cdfdb4 | 6 | * Copyright(C) 2006-2007 Timesys Corp., Thomas Gleixner |
c0a31329 TG |
7 | * |
8 | * High-resolution kernel timers | |
9 | * | |
10 | * In contrast to the low-resolution timeout API implemented in | |
11 | * kernel/timer.c, hrtimers provide finer resolution and accuracy | |
12 | * depending on system configuration and capabilities. | |
13 | * | |
14 | * These timers are currently used for: | |
15 | * - itimers | |
16 | * - POSIX timers | |
17 | * - nanosleep | |
18 | * - precise in-kernel timing | |
19 | * | |
20 | * Started by: Thomas Gleixner and Ingo Molnar | |
21 | * | |
22 | * Credits: | |
23 | * based on kernel/timer.c | |
24 | * | |
66188fae TG |
25 | * Help, testing, suggestions, bugfixes, improvements were |
26 | * provided by: | |
27 | * | |
28 | * George Anzinger, Andrew Morton, Steven Rostedt, Roman Zippel | |
29 | * et. al. | |
30 | * | |
c0a31329 TG |
31 | * For licencing details see kernel-base/COPYING |
32 | */ | |
33 | ||
34 | #include <linux/cpu.h> | |
9984de1a | 35 | #include <linux/export.h> |
c0a31329 TG |
36 | #include <linux/percpu.h> |
37 | #include <linux/hrtimer.h> | |
38 | #include <linux/notifier.h> | |
39 | #include <linux/syscalls.h> | |
54cdfdb4 | 40 | #include <linux/kallsyms.h> |
c0a31329 | 41 | #include <linux/interrupt.h> |
79bf2bb3 | 42 | #include <linux/tick.h> |
54cdfdb4 TG |
43 | #include <linux/seq_file.h> |
44 | #include <linux/err.h> | |
237fc6e7 | 45 | #include <linux/debugobjects.h> |
174cd4b1 | 46 | #include <linux/sched/signal.h> |
cf4aebc2 | 47 | #include <linux/sched/sysctl.h> |
8bd75c77 | 48 | #include <linux/sched/rt.h> |
aab03e05 | 49 | #include <linux/sched/deadline.h> |
370c9135 | 50 | #include <linux/sched/nohz.h> |
b17b0153 | 51 | #include <linux/sched/debug.h> |
eea08f32 | 52 | #include <linux/timer.h> |
b0f8c44f | 53 | #include <linux/freezer.h> |
edbeda46 | 54 | #include <linux/compat.h> |
c0a31329 | 55 | |
7c0f6ba6 | 56 | #include <linux/uaccess.h> |
c0a31329 | 57 | |
c6a2a177 XG |
58 | #include <trace/events/timer.h> |
59 | ||
c1797baf | 60 | #include "tick-internal.h" |
8b094cd0 | 61 | |
c458b1d1 AMG |
62 | /* |
63 | * Masks for selecting the soft and hard context timers from | |
64 | * cpu_base->active | |
65 | */ | |
66 | #define MASK_SHIFT (HRTIMER_BASE_MONOTONIC_SOFT) | |
67 | #define HRTIMER_ACTIVE_HARD ((1U << MASK_SHIFT) - 1) | |
68 | #define HRTIMER_ACTIVE_SOFT (HRTIMER_ACTIVE_HARD << MASK_SHIFT) | |
69 | #define HRTIMER_ACTIVE_ALL (HRTIMER_ACTIVE_SOFT | HRTIMER_ACTIVE_HARD) | |
70 | ||
c0a31329 TG |
71 | /* |
72 | * The timer bases: | |
7978672c | 73 | * |
571af55a | 74 | * There are more clockids than hrtimer bases. Thus, we index |
e06383db JS |
75 | * into the timer bases by the hrtimer_base_type enum. When trying |
76 | * to reach a base using a clockid, hrtimer_clockid_to_base() | |
77 | * is used to convert from clockid to the proper hrtimer_base_type. | |
c0a31329 | 78 | */ |
54cdfdb4 | 79 | DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) = |
c0a31329 | 80 | { |
84cc8fd2 | 81 | .lock = __RAW_SPIN_LOCK_UNLOCKED(hrtimer_bases.lock), |
3c8aa39d | 82 | .clock_base = |
c0a31329 | 83 | { |
3c8aa39d | 84 | { |
ab8177bc TG |
85 | .index = HRTIMER_BASE_MONOTONIC, |
86 | .clockid = CLOCK_MONOTONIC, | |
3c8aa39d | 87 | .get_time = &ktime_get, |
3c8aa39d | 88 | }, |
68fa61c0 TG |
89 | { |
90 | .index = HRTIMER_BASE_REALTIME, | |
91 | .clockid = CLOCK_REALTIME, | |
92 | .get_time = &ktime_get_real, | |
68fa61c0 | 93 | }, |
70a08cca | 94 | { |
ab8177bc TG |
95 | .index = HRTIMER_BASE_BOOTTIME, |
96 | .clockid = CLOCK_BOOTTIME, | |
70a08cca | 97 | .get_time = &ktime_get_boottime, |
70a08cca | 98 | }, |
90adda98 JS |
99 | { |
100 | .index = HRTIMER_BASE_TAI, | |
101 | .clockid = CLOCK_TAI, | |
102 | .get_time = &ktime_get_clocktai, | |
90adda98 | 103 | }, |
98ecadd4 AMG |
104 | { |
105 | .index = HRTIMER_BASE_MONOTONIC_SOFT, | |
106 | .clockid = CLOCK_MONOTONIC, | |
107 | .get_time = &ktime_get, | |
108 | }, | |
109 | { | |
110 | .index = HRTIMER_BASE_REALTIME_SOFT, | |
111 | .clockid = CLOCK_REALTIME, | |
112 | .get_time = &ktime_get_real, | |
113 | }, | |
114 | { | |
115 | .index = HRTIMER_BASE_BOOTTIME_SOFT, | |
116 | .clockid = CLOCK_BOOTTIME, | |
117 | .get_time = &ktime_get_boottime, | |
118 | }, | |
119 | { | |
120 | .index = HRTIMER_BASE_TAI_SOFT, | |
121 | .clockid = CLOCK_TAI, | |
122 | .get_time = &ktime_get_clocktai, | |
123 | }, | |
3c8aa39d | 124 | } |
c0a31329 TG |
125 | }; |
126 | ||
942c3c5c | 127 | static const int hrtimer_clock_to_base_table[MAX_CLOCKS] = { |
336a9cde MZ |
128 | /* Make sure we catch unsupported clockids */ |
129 | [0 ... MAX_CLOCKS - 1] = HRTIMER_MAX_CLOCK_BASES, | |
130 | ||
ce31332d TG |
131 | [CLOCK_REALTIME] = HRTIMER_BASE_REALTIME, |
132 | [CLOCK_MONOTONIC] = HRTIMER_BASE_MONOTONIC, | |
133 | [CLOCK_BOOTTIME] = HRTIMER_BASE_BOOTTIME, | |
90adda98 | 134 | [CLOCK_TAI] = HRTIMER_BASE_TAI, |
ce31332d | 135 | }; |
e06383db | 136 | |
c0a31329 TG |
137 | /* |
138 | * Functions and macros which are different for UP/SMP systems are kept in a | |
139 | * single place | |
140 | */ | |
141 | #ifdef CONFIG_SMP | |
142 | ||
887d9dc9 PZ |
143 | /* |
144 | * We require the migration_base for lock_hrtimer_base()/switch_hrtimer_base() | |
145 | * such that hrtimer_callback_running() can unconditionally dereference | |
146 | * timer->base->cpu_base | |
147 | */ | |
148 | static struct hrtimer_cpu_base migration_cpu_base = { | |
887d9dc9 PZ |
149 | .clock_base = { { .cpu_base = &migration_cpu_base, }, }, |
150 | }; | |
151 | ||
152 | #define migration_base migration_cpu_base.clock_base[0] | |
153 | ||
c0a31329 TG |
154 | /* |
155 | * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock | |
156 | * means that all timers which are tied to this base via timer->base are | |
157 | * locked, and the base itself is locked too. | |
158 | * | |
159 | * So __run_timers/migrate_timers can safely modify all timers which could | |
160 | * be found on the lists/queues. | |
161 | * | |
162 | * When the timer's base is locked, and the timer removed from list, it is | |
887d9dc9 PZ |
163 | * possible to set timer->base = &migration_base and drop the lock: the timer |
164 | * remains locked. | |
c0a31329 | 165 | */ |
3c8aa39d TG |
166 | static |
167 | struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer, | |
168 | unsigned long *flags) | |
c0a31329 | 169 | { |
3c8aa39d | 170 | struct hrtimer_clock_base *base; |
c0a31329 TG |
171 | |
172 | for (;;) { | |
173 | base = timer->base; | |
887d9dc9 | 174 | if (likely(base != &migration_base)) { |
ecb49d1a | 175 | raw_spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
176 | if (likely(base == timer->base)) |
177 | return base; | |
178 | /* The timer has migrated to another CPU: */ | |
ecb49d1a | 179 | raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags); |
c0a31329 TG |
180 | } |
181 | cpu_relax(); | |
182 | } | |
183 | } | |
184 | ||
6ff7041d | 185 | /* |
07a9a7ea AMG |
186 | * We do not migrate the timer when it is expiring before the next |
187 | * event on the target cpu. When high resolution is enabled, we cannot | |
188 | * reprogram the target cpu hardware and we would cause it to fire | |
189 | * late. To keep it simple, we handle the high resolution enabled and | |
190 | * disabled case similar. | |
6ff7041d TG |
191 | * |
192 | * Called with cpu_base->lock of target cpu held. | |
193 | */ | |
194 | static int | |
195 | hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base) | |
196 | { | |
6ff7041d TG |
197 | ktime_t expires; |
198 | ||
6ff7041d | 199 | expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset); |
2ac2dccc | 200 | return expires < new_base->cpu_base->expires_next; |
6ff7041d TG |
201 | } |
202 | ||
bc7a34b8 TG |
203 | static inline |
204 | struct hrtimer_cpu_base *get_target_base(struct hrtimer_cpu_base *base, | |
205 | int pinned) | |
206 | { | |
ae67bada TG |
207 | #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON) |
208 | if (static_branch_likely(&timers_migration_enabled) && !pinned) | |
209 | return &per_cpu(hrtimer_bases, get_nohz_timer_target()); | |
210 | #endif | |
662b3e19 | 211 | return base; |
bc7a34b8 | 212 | } |
bc7a34b8 | 213 | |
c0a31329 | 214 | /* |
b48362d8 FW |
215 | * We switch the timer base to a power-optimized selected CPU target, |
216 | * if: | |
217 | * - NO_HZ_COMMON is enabled | |
218 | * - timer migration is enabled | |
219 | * - the timer callback is not running | |
220 | * - the timer is not the first expiring timer on the new target | |
221 | * | |
222 | * If one of the above requirements is not fulfilled we move the timer | |
223 | * to the current CPU or leave it on the previously assigned CPU if | |
224 | * the timer callback is currently running. | |
c0a31329 | 225 | */ |
3c8aa39d | 226 | static inline struct hrtimer_clock_base * |
597d0275 AB |
227 | switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base, |
228 | int pinned) | |
c0a31329 | 229 | { |
b48362d8 | 230 | struct hrtimer_cpu_base *new_cpu_base, *this_cpu_base; |
3c8aa39d | 231 | struct hrtimer_clock_base *new_base; |
ab8177bc | 232 | int basenum = base->index; |
c0a31329 | 233 | |
b48362d8 FW |
234 | this_cpu_base = this_cpu_ptr(&hrtimer_bases); |
235 | new_cpu_base = get_target_base(this_cpu_base, pinned); | |
eea08f32 | 236 | again: |
e06383db | 237 | new_base = &new_cpu_base->clock_base[basenum]; |
c0a31329 TG |
238 | |
239 | if (base != new_base) { | |
240 | /* | |
6ff7041d | 241 | * We are trying to move timer to new_base. |
c0a31329 TG |
242 | * However we can't change timer's base while it is running, |
243 | * so we keep it on the same CPU. No hassle vs. reprogramming | |
244 | * the event source in the high resolution case. The softirq | |
245 | * code will take care of this when the timer function has | |
246 | * completed. There is no conflict as we hold the lock until | |
247 | * the timer is enqueued. | |
248 | */ | |
54cdfdb4 | 249 | if (unlikely(hrtimer_callback_running(timer))) |
c0a31329 TG |
250 | return base; |
251 | ||
887d9dc9 PZ |
252 | /* See the comment in lock_hrtimer_base() */ |
253 | timer->base = &migration_base; | |
ecb49d1a TG |
254 | raw_spin_unlock(&base->cpu_base->lock); |
255 | raw_spin_lock(&new_base->cpu_base->lock); | |
eea08f32 | 256 | |
b48362d8 | 257 | if (new_cpu_base != this_cpu_base && |
bc7a34b8 | 258 | hrtimer_check_target(timer, new_base)) { |
ecb49d1a TG |
259 | raw_spin_unlock(&new_base->cpu_base->lock); |
260 | raw_spin_lock(&base->cpu_base->lock); | |
b48362d8 | 261 | new_cpu_base = this_cpu_base; |
6ff7041d TG |
262 | timer->base = base; |
263 | goto again; | |
eea08f32 | 264 | } |
c0a31329 | 265 | timer->base = new_base; |
012a45e3 | 266 | } else { |
b48362d8 | 267 | if (new_cpu_base != this_cpu_base && |
bc7a34b8 | 268 | hrtimer_check_target(timer, new_base)) { |
b48362d8 | 269 | new_cpu_base = this_cpu_base; |
012a45e3 LM |
270 | goto again; |
271 | } | |
c0a31329 TG |
272 | } |
273 | return new_base; | |
274 | } | |
275 | ||
276 | #else /* CONFIG_SMP */ | |
277 | ||
3c8aa39d | 278 | static inline struct hrtimer_clock_base * |
c0a31329 TG |
279 | lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) |
280 | { | |
3c8aa39d | 281 | struct hrtimer_clock_base *base = timer->base; |
c0a31329 | 282 | |
ecb49d1a | 283 | raw_spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
284 | |
285 | return base; | |
286 | } | |
287 | ||
eea08f32 | 288 | # define switch_hrtimer_base(t, b, p) (b) |
c0a31329 TG |
289 | |
290 | #endif /* !CONFIG_SMP */ | |
291 | ||
292 | /* | |
293 | * Functions for the union type storage format of ktime_t which are | |
294 | * too large for inlining: | |
295 | */ | |
296 | #if BITS_PER_LONG < 64 | |
c0a31329 TG |
297 | /* |
298 | * Divide a ktime value by a nanosecond value | |
299 | */ | |
f7bcb70e | 300 | s64 __ktime_divns(const ktime_t kt, s64 div) |
c0a31329 | 301 | { |
c0a31329 | 302 | int sft = 0; |
f7bcb70e JS |
303 | s64 dclc; |
304 | u64 tmp; | |
c0a31329 | 305 | |
900cfa46 | 306 | dclc = ktime_to_ns(kt); |
f7bcb70e JS |
307 | tmp = dclc < 0 ? -dclc : dclc; |
308 | ||
c0a31329 TG |
309 | /* Make sure the divisor is less than 2^32: */ |
310 | while (div >> 32) { | |
311 | sft++; | |
312 | div >>= 1; | |
313 | } | |
f7bcb70e JS |
314 | tmp >>= sft; |
315 | do_div(tmp, (unsigned long) div); | |
316 | return dclc < 0 ? -tmp : tmp; | |
c0a31329 | 317 | } |
8b618628 | 318 | EXPORT_SYMBOL_GPL(__ktime_divns); |
c0a31329 TG |
319 | #endif /* BITS_PER_LONG >= 64 */ |
320 | ||
5a7780e7 TG |
321 | /* |
322 | * Add two ktime values and do a safety check for overflow: | |
323 | */ | |
324 | ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs) | |
325 | { | |
979515c5 | 326 | ktime_t res = ktime_add_unsafe(lhs, rhs); |
5a7780e7 TG |
327 | |
328 | /* | |
329 | * We use KTIME_SEC_MAX here, the maximum timeout which we can | |
330 | * return to user space in a timespec: | |
331 | */ | |
2456e855 | 332 | if (res < 0 || res < lhs || res < rhs) |
5a7780e7 TG |
333 | res = ktime_set(KTIME_SEC_MAX, 0); |
334 | ||
335 | return res; | |
336 | } | |
337 | ||
8daa21e6 AB |
338 | EXPORT_SYMBOL_GPL(ktime_add_safe); |
339 | ||
237fc6e7 TG |
340 | #ifdef CONFIG_DEBUG_OBJECTS_TIMERS |
341 | ||
342 | static struct debug_obj_descr hrtimer_debug_descr; | |
343 | ||
99777288 SG |
344 | static void *hrtimer_debug_hint(void *addr) |
345 | { | |
346 | return ((struct hrtimer *) addr)->function; | |
347 | } | |
348 | ||
237fc6e7 TG |
349 | /* |
350 | * fixup_init is called when: | |
351 | * - an active object is initialized | |
352 | */ | |
e3252464 | 353 | static bool hrtimer_fixup_init(void *addr, enum debug_obj_state state) |
237fc6e7 TG |
354 | { |
355 | struct hrtimer *timer = addr; | |
356 | ||
357 | switch (state) { | |
358 | case ODEBUG_STATE_ACTIVE: | |
359 | hrtimer_cancel(timer); | |
360 | debug_object_init(timer, &hrtimer_debug_descr); | |
e3252464 | 361 | return true; |
237fc6e7 | 362 | default: |
e3252464 | 363 | return false; |
237fc6e7 TG |
364 | } |
365 | } | |
366 | ||
367 | /* | |
368 | * fixup_activate is called when: | |
369 | * - an active object is activated | |
b9fdac7f | 370 | * - an unknown non-static object is activated |
237fc6e7 | 371 | */ |
e3252464 | 372 | static bool hrtimer_fixup_activate(void *addr, enum debug_obj_state state) |
237fc6e7 TG |
373 | { |
374 | switch (state) { | |
237fc6e7 TG |
375 | case ODEBUG_STATE_ACTIVE: |
376 | WARN_ON(1); | |
377 | ||
378 | default: | |
e3252464 | 379 | return false; |
237fc6e7 TG |
380 | } |
381 | } | |
382 | ||
383 | /* | |
384 | * fixup_free is called when: | |
385 | * - an active object is freed | |
386 | */ | |
e3252464 | 387 | static bool hrtimer_fixup_free(void *addr, enum debug_obj_state state) |
237fc6e7 TG |
388 | { |
389 | struct hrtimer *timer = addr; | |
390 | ||
391 | switch (state) { | |
392 | case ODEBUG_STATE_ACTIVE: | |
393 | hrtimer_cancel(timer); | |
394 | debug_object_free(timer, &hrtimer_debug_descr); | |
e3252464 | 395 | return true; |
237fc6e7 | 396 | default: |
e3252464 | 397 | return false; |
237fc6e7 TG |
398 | } |
399 | } | |
400 | ||
401 | static struct debug_obj_descr hrtimer_debug_descr = { | |
402 | .name = "hrtimer", | |
99777288 | 403 | .debug_hint = hrtimer_debug_hint, |
237fc6e7 TG |
404 | .fixup_init = hrtimer_fixup_init, |
405 | .fixup_activate = hrtimer_fixup_activate, | |
406 | .fixup_free = hrtimer_fixup_free, | |
407 | }; | |
408 | ||
409 | static inline void debug_hrtimer_init(struct hrtimer *timer) | |
410 | { | |
411 | debug_object_init(timer, &hrtimer_debug_descr); | |
412 | } | |
413 | ||
5da70160 AMG |
414 | static inline void debug_hrtimer_activate(struct hrtimer *timer, |
415 | enum hrtimer_mode mode) | |
237fc6e7 TG |
416 | { |
417 | debug_object_activate(timer, &hrtimer_debug_descr); | |
418 | } | |
419 | ||
420 | static inline void debug_hrtimer_deactivate(struct hrtimer *timer) | |
421 | { | |
422 | debug_object_deactivate(timer, &hrtimer_debug_descr); | |
423 | } | |
424 | ||
425 | static inline void debug_hrtimer_free(struct hrtimer *timer) | |
426 | { | |
427 | debug_object_free(timer, &hrtimer_debug_descr); | |
428 | } | |
429 | ||
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 TG |
440 | |
441 | void destroy_hrtimer_on_stack(struct hrtimer *timer) | |
442 | { | |
443 | debug_object_free(timer, &hrtimer_debug_descr); | |
444 | } | |
c08376ac | 445 | EXPORT_SYMBOL_GPL(destroy_hrtimer_on_stack); |
237fc6e7 TG |
446 | |
447 | #else | |
5da70160 | 448 | |
237fc6e7 | 449 | static inline void debug_hrtimer_init(struct hrtimer *timer) { } |
5da70160 AMG |
450 | static inline void debug_hrtimer_activate(struct hrtimer *timer, |
451 | enum hrtimer_mode mode) { } | |
237fc6e7 TG |
452 | static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { } |
453 | #endif | |
454 | ||
c6a2a177 XG |
455 | static inline void |
456 | debug_init(struct hrtimer *timer, clockid_t clockid, | |
457 | enum hrtimer_mode mode) | |
458 | { | |
459 | debug_hrtimer_init(timer); | |
460 | trace_hrtimer_init(timer, clockid, mode); | |
461 | } | |
462 | ||
63e2ed36 AMG |
463 | static inline void debug_activate(struct hrtimer *timer, |
464 | enum hrtimer_mode mode) | |
c6a2a177 | 465 | { |
5da70160 | 466 | debug_hrtimer_activate(timer, mode); |
63e2ed36 | 467 | trace_hrtimer_start(timer, mode); |
c6a2a177 XG |
468 | } |
469 | ||
470 | static inline void debug_deactivate(struct hrtimer *timer) | |
471 | { | |
472 | debug_hrtimer_deactivate(timer); | |
473 | trace_hrtimer_cancel(timer); | |
474 | } | |
475 | ||
c272ca58 AMG |
476 | static struct hrtimer_clock_base * |
477 | __next_base(struct hrtimer_cpu_base *cpu_base, unsigned int *active) | |
478 | { | |
479 | unsigned int idx; | |
480 | ||
481 | if (!*active) | |
482 | return NULL; | |
483 | ||
484 | idx = __ffs(*active); | |
485 | *active &= ~(1U << idx); | |
486 | ||
487 | return &cpu_base->clock_base[idx]; | |
488 | } | |
489 | ||
490 | #define for_each_active_base(base, cpu_base, active) \ | |
491 | while ((base = __next_base((cpu_base), &(active)))) | |
492 | ||
ad38f596 AMG |
493 | static ktime_t __hrtimer_next_event_base(struct hrtimer_cpu_base *cpu_base, |
494 | unsigned int active, | |
495 | ktime_t expires_next) | |
9bc74919 | 496 | { |
c272ca58 | 497 | struct hrtimer_clock_base *base; |
ad38f596 | 498 | ktime_t expires; |
9bc74919 | 499 | |
c272ca58 | 500 | for_each_active_base(base, cpu_base, active) { |
9bc74919 TG |
501 | struct timerqueue_node *next; |
502 | struct hrtimer *timer; | |
503 | ||
34aee88a | 504 | next = timerqueue_getnext(&base->active); |
9bc74919 TG |
505 | timer = container_of(next, struct hrtimer, node); |
506 | expires = ktime_sub(hrtimer_get_expires(timer), base->offset); | |
2456e855 | 507 | if (expires < expires_next) { |
9bc74919 | 508 | expires_next = expires; |
5da70160 AMG |
509 | if (timer->is_soft) |
510 | cpu_base->softirq_next_timer = timer; | |
511 | else | |
512 | cpu_base->next_timer = timer; | |
895bdfa7 | 513 | } |
9bc74919 TG |
514 | } |
515 | /* | |
516 | * clock_was_set() might have changed base->offset of any of | |
517 | * the clock bases so the result might be negative. Fix it up | |
518 | * to prevent a false positive in clockevents_program_event(). | |
519 | */ | |
2456e855 TG |
520 | if (expires_next < 0) |
521 | expires_next = 0; | |
9bc74919 TG |
522 | return expires_next; |
523 | } | |
9bc74919 | 524 | |
c458b1d1 AMG |
525 | /* |
526 | * Recomputes cpu_base::*next_timer and returns the earliest expires_next but | |
527 | * does not set cpu_base::*expires_next, that is done by hrtimer_reprogram. | |
528 | * | |
5da70160 AMG |
529 | * When a softirq is pending, we can ignore the HRTIMER_ACTIVE_SOFT bases, |
530 | * those timers will get run whenever the softirq gets handled, at the end of | |
531 | * hrtimer_run_softirq(), hrtimer_update_softirq_timer() will re-add these bases. | |
532 | * | |
533 | * Therefore softirq values are those from the HRTIMER_ACTIVE_SOFT clock bases. | |
534 | * The !softirq values are the minima across HRTIMER_ACTIVE_ALL, unless an actual | |
535 | * softirq is pending, in which case they're the minima of HRTIMER_ACTIVE_HARD. | |
536 | * | |
c458b1d1 | 537 | * @active_mask must be one of: |
5da70160 | 538 | * - HRTIMER_ACTIVE_ALL, |
c458b1d1 AMG |
539 | * - HRTIMER_ACTIVE_SOFT, or |
540 | * - HRTIMER_ACTIVE_HARD. | |
541 | */ | |
5da70160 AMG |
542 | static ktime_t |
543 | __hrtimer_get_next_event(struct hrtimer_cpu_base *cpu_base, unsigned int active_mask) | |
ad38f596 | 544 | { |
c458b1d1 | 545 | unsigned int active; |
5da70160 | 546 | struct hrtimer *next_timer = NULL; |
ad38f596 AMG |
547 | ktime_t expires_next = KTIME_MAX; |
548 | ||
5da70160 AMG |
549 | if (!cpu_base->softirq_activated && (active_mask & HRTIMER_ACTIVE_SOFT)) { |
550 | active = cpu_base->active_bases & HRTIMER_ACTIVE_SOFT; | |
551 | cpu_base->softirq_next_timer = NULL; | |
552 | expires_next = __hrtimer_next_event_base(cpu_base, active, KTIME_MAX); | |
553 | ||
554 | next_timer = cpu_base->softirq_next_timer; | |
555 | } | |
ad38f596 | 556 | |
5da70160 AMG |
557 | if (active_mask & HRTIMER_ACTIVE_HARD) { |
558 | active = cpu_base->active_bases & HRTIMER_ACTIVE_HARD; | |
559 | cpu_base->next_timer = next_timer; | |
560 | expires_next = __hrtimer_next_event_base(cpu_base, active, expires_next); | |
561 | } | |
ad38f596 AMG |
562 | |
563 | return expires_next; | |
564 | } | |
565 | ||
21d6d52a TG |
566 | static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base) |
567 | { | |
568 | ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset; | |
569 | ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset; | |
570 | ktime_t *offs_tai = &base->clock_base[HRTIMER_BASE_TAI].offset; | |
571 | ||
5da70160 | 572 | ktime_t now = ktime_get_update_offsets_now(&base->clock_was_set_seq, |
868a3e91 | 573 | offs_real, offs_boot, offs_tai); |
5da70160 AMG |
574 | |
575 | base->clock_base[HRTIMER_BASE_REALTIME_SOFT].offset = *offs_real; | |
576 | base->clock_base[HRTIMER_BASE_BOOTTIME_SOFT].offset = *offs_boot; | |
577 | base->clock_base[HRTIMER_BASE_TAI_SOFT].offset = *offs_tai; | |
578 | ||
579 | return now; | |
21d6d52a TG |
580 | } |
581 | ||
28bfd18b AMG |
582 | /* |
583 | * Is the high resolution mode active ? | |
584 | */ | |
585 | static inline int __hrtimer_hres_active(struct hrtimer_cpu_base *cpu_base) | |
586 | { | |
587 | return IS_ENABLED(CONFIG_HIGH_RES_TIMERS) ? | |
588 | cpu_base->hres_active : 0; | |
589 | } | |
590 | ||
591 | static inline int hrtimer_hres_active(void) | |
592 | { | |
593 | return __hrtimer_hres_active(this_cpu_ptr(&hrtimer_bases)); | |
594 | } | |
595 | ||
54cdfdb4 TG |
596 | /* |
597 | * Reprogram the event source with checking both queues for the | |
598 | * next event | |
599 | * Called with interrupts disabled and base->lock held | |
600 | */ | |
7403f41f AC |
601 | static void |
602 | hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal) | |
54cdfdb4 | 603 | { |
21d6d52a TG |
604 | ktime_t expires_next; |
605 | ||
5da70160 AMG |
606 | /* |
607 | * Find the current next expiration time. | |
608 | */ | |
609 | expires_next = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_ALL); | |
610 | ||
611 | if (cpu_base->next_timer && cpu_base->next_timer->is_soft) { | |
612 | /* | |
613 | * When the softirq is activated, hrtimer has to be | |
614 | * programmed with the first hard hrtimer because soft | |
615 | * timer interrupt could occur too late. | |
616 | */ | |
617 | if (cpu_base->softirq_activated) | |
618 | expires_next = __hrtimer_get_next_event(cpu_base, | |
619 | HRTIMER_ACTIVE_HARD); | |
620 | else | |
621 | cpu_base->softirq_expires_next = expires_next; | |
622 | } | |
54cdfdb4 | 623 | |
2456e855 | 624 | if (skip_equal && expires_next == cpu_base->expires_next) |
7403f41f AC |
625 | return; |
626 | ||
2456e855 | 627 | cpu_base->expires_next = expires_next; |
7403f41f | 628 | |
6c6c0d5a | 629 | /* |
61bb4bcb AMG |
630 | * If hres is not active, hardware does not have to be |
631 | * reprogrammed yet. | |
632 | * | |
6c6c0d5a SH |
633 | * If a hang was detected in the last timer interrupt then we |
634 | * leave the hang delay active in the hardware. We want the | |
635 | * system to make progress. That also prevents the following | |
636 | * scenario: | |
637 | * T1 expires 50ms from now | |
638 | * T2 expires 5s from now | |
639 | * | |
640 | * T1 is removed, so this code is called and would reprogram | |
641 | * the hardware to 5s from now. Any hrtimer_start after that | |
642 | * will not reprogram the hardware due to hang_detected being | |
643 | * set. So we'd effectivly block all timers until the T2 event | |
644 | * fires. | |
645 | */ | |
61bb4bcb | 646 | if (!__hrtimer_hres_active(cpu_base) || cpu_base->hang_detected) |
6c6c0d5a SH |
647 | return; |
648 | ||
d2540875 | 649 | tick_program_event(cpu_base->expires_next, 1); |
54cdfdb4 TG |
650 | } |
651 | ||
ebba2c72 AMG |
652 | /* High resolution timer related functions */ |
653 | #ifdef CONFIG_HIGH_RES_TIMERS | |
654 | ||
655 | /* | |
656 | * High resolution timer enabled ? | |
657 | */ | |
658 | static bool hrtimer_hres_enabled __read_mostly = true; | |
659 | unsigned int hrtimer_resolution __read_mostly = LOW_RES_NSEC; | |
660 | EXPORT_SYMBOL_GPL(hrtimer_resolution); | |
661 | ||
662 | /* | |
663 | * Enable / Disable high resolution mode | |
664 | */ | |
665 | static int __init setup_hrtimer_hres(char *str) | |
666 | { | |
667 | return (kstrtobool(str, &hrtimer_hres_enabled) == 0); | |
668 | } | |
669 | ||
670 | __setup("highres=", setup_hrtimer_hres); | |
671 | ||
672 | /* | |
673 | * hrtimer_high_res_enabled - query, if the highres mode is enabled | |
674 | */ | |
675 | static inline int hrtimer_is_hres_enabled(void) | |
676 | { | |
677 | return hrtimer_hres_enabled; | |
678 | } | |
679 | ||
9ec26907 TG |
680 | /* |
681 | * Retrigger next event is called after clock was set | |
682 | * | |
683 | * Called with interrupts disabled via on_each_cpu() | |
684 | */ | |
685 | static void retrigger_next_event(void *arg) | |
686 | { | |
dc5df73b | 687 | struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases); |
9ec26907 | 688 | |
851cff8c | 689 | if (!__hrtimer_hres_active(base)) |
9ec26907 TG |
690 | return; |
691 | ||
9ec26907 | 692 | raw_spin_lock(&base->lock); |
5baefd6d | 693 | hrtimer_update_base(base); |
9ec26907 TG |
694 | hrtimer_force_reprogram(base, 0); |
695 | raw_spin_unlock(&base->lock); | |
696 | } | |
b12a03ce | 697 | |
54cdfdb4 TG |
698 | /* |
699 | * Switch to high resolution mode | |
700 | */ | |
75e3b37d | 701 | static void hrtimer_switch_to_hres(void) |
54cdfdb4 | 702 | { |
c6eb3f70 | 703 | struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases); |
54cdfdb4 TG |
704 | |
705 | if (tick_init_highres()) { | |
820de5c3 | 706 | printk(KERN_WARNING "Could not switch to high resolution " |
c6eb3f70 | 707 | "mode on CPU %d\n", base->cpu); |
85e1cd6e | 708 | return; |
54cdfdb4 TG |
709 | } |
710 | base->hres_active = 1; | |
398ca17f | 711 | hrtimer_resolution = HIGH_RES_NSEC; |
54cdfdb4 TG |
712 | |
713 | tick_setup_sched_timer(); | |
54cdfdb4 TG |
714 | /* "Retrigger" the interrupt to get things going */ |
715 | retrigger_next_event(NULL); | |
54cdfdb4 TG |
716 | } |
717 | ||
5ec2481b TG |
718 | static void clock_was_set_work(struct work_struct *work) |
719 | { | |
720 | clock_was_set(); | |
721 | } | |
722 | ||
723 | static DECLARE_WORK(hrtimer_work, clock_was_set_work); | |
724 | ||
f55a6faa | 725 | /* |
b4d90e9f | 726 | * Called from timekeeping and resume code to reprogram the hrtimer |
5ec2481b | 727 | * interrupt device on all cpus. |
f55a6faa JS |
728 | */ |
729 | void clock_was_set_delayed(void) | |
730 | { | |
5ec2481b | 731 | schedule_work(&hrtimer_work); |
f55a6faa JS |
732 | } |
733 | ||
54cdfdb4 TG |
734 | #else |
735 | ||
54cdfdb4 | 736 | static inline int hrtimer_is_hres_enabled(void) { return 0; } |
75e3b37d | 737 | static inline void hrtimer_switch_to_hres(void) { } |
9ec26907 | 738 | static inline void retrigger_next_event(void *arg) { } |
54cdfdb4 TG |
739 | |
740 | #endif /* CONFIG_HIGH_RES_TIMERS */ | |
741 | ||
11a9fe06 AMG |
742 | /* |
743 | * When a timer is enqueued and expires earlier than the already enqueued | |
744 | * timers, we have to check, whether it expires earlier than the timer for | |
745 | * which the clock event device was armed. | |
746 | * | |
747 | * Called with interrupts disabled and base->cpu_base.lock held | |
748 | */ | |
5da70160 | 749 | static void hrtimer_reprogram(struct hrtimer *timer, bool reprogram) |
11a9fe06 AMG |
750 | { |
751 | struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases); | |
3ec7a3ee | 752 | struct hrtimer_clock_base *base = timer->base; |
11a9fe06 AMG |
753 | ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset); |
754 | ||
755 | WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0); | |
756 | ||
5da70160 AMG |
757 | /* |
758 | * CLOCK_REALTIME timer might be requested with an absolute | |
759 | * expiry time which is less than base->offset. Set it to 0. | |
760 | */ | |
761 | if (expires < 0) | |
762 | expires = 0; | |
763 | ||
764 | if (timer->is_soft) { | |
765 | /* | |
766 | * soft hrtimer could be started on a remote CPU. In this | |
767 | * case softirq_expires_next needs to be updated on the | |
768 | * remote CPU. The soft hrtimer will not expire before the | |
769 | * first hard hrtimer on the remote CPU - | |
770 | * hrtimer_check_target() prevents this case. | |
771 | */ | |
772 | struct hrtimer_cpu_base *timer_cpu_base = base->cpu_base; | |
773 | ||
774 | if (timer_cpu_base->softirq_activated) | |
775 | return; | |
776 | ||
777 | if (!ktime_before(expires, timer_cpu_base->softirq_expires_next)) | |
778 | return; | |
779 | ||
780 | timer_cpu_base->softirq_next_timer = timer; | |
781 | timer_cpu_base->softirq_expires_next = expires; | |
782 | ||
783 | if (!ktime_before(expires, timer_cpu_base->expires_next) || | |
784 | !reprogram) | |
785 | return; | |
786 | } | |
787 | ||
11a9fe06 AMG |
788 | /* |
789 | * If the timer is not on the current cpu, we cannot reprogram | |
790 | * the other cpus clock event device. | |
791 | */ | |
792 | if (base->cpu_base != cpu_base) | |
793 | return; | |
794 | ||
795 | /* | |
796 | * If the hrtimer interrupt is running, then it will | |
797 | * reevaluate the clock bases and reprogram the clock event | |
798 | * device. The callbacks are always executed in hard interrupt | |
799 | * context so we don't need an extra check for a running | |
800 | * callback. | |
801 | */ | |
802 | if (cpu_base->in_hrtirq) | |
803 | return; | |
804 | ||
11a9fe06 AMG |
805 | if (expires >= cpu_base->expires_next) |
806 | return; | |
807 | ||
808 | /* Update the pointer to the next expiring timer */ | |
809 | cpu_base->next_timer = timer; | |
14c80341 | 810 | cpu_base->expires_next = expires; |
11a9fe06 AMG |
811 | |
812 | /* | |
14c80341 AMG |
813 | * If hres is not active, hardware does not have to be |
814 | * programmed yet. | |
815 | * | |
11a9fe06 AMG |
816 | * If a hang was detected in the last timer interrupt then we |
817 | * do not schedule a timer which is earlier than the expiry | |
818 | * which we enforced in the hang detection. We want the system | |
819 | * to make progress. | |
820 | */ | |
14c80341 | 821 | if (!__hrtimer_hres_active(cpu_base) || cpu_base->hang_detected) |
11a9fe06 AMG |
822 | return; |
823 | ||
824 | /* | |
825 | * Program the timer hardware. We enforce the expiry for | |
826 | * events which are already in the past. | |
827 | */ | |
11a9fe06 AMG |
828 | tick_program_event(expires, 1); |
829 | } | |
830 | ||
b12a03ce TG |
831 | /* |
832 | * Clock realtime was set | |
833 | * | |
834 | * Change the offset of the realtime clock vs. the monotonic | |
835 | * clock. | |
836 | * | |
837 | * We might have to reprogram the high resolution timer interrupt. On | |
838 | * SMP we call the architecture specific code to retrigger _all_ high | |
839 | * resolution timer interrupts. On UP we just disable interrupts and | |
840 | * call the high resolution interrupt code. | |
841 | */ | |
842 | void clock_was_set(void) | |
843 | { | |
90ff1f30 | 844 | #ifdef CONFIG_HIGH_RES_TIMERS |
b12a03ce TG |
845 | /* Retrigger the CPU local events everywhere */ |
846 | on_each_cpu(retrigger_next_event, NULL, 1); | |
9ec26907 TG |
847 | #endif |
848 | timerfd_clock_was_set(); | |
b12a03ce TG |
849 | } |
850 | ||
851 | /* | |
852 | * During resume we might have to reprogram the high resolution timer | |
7c4c3a0f DV |
853 | * interrupt on all online CPUs. However, all other CPUs will be |
854 | * stopped with IRQs interrupts disabled so the clock_was_set() call | |
5ec2481b | 855 | * must be deferred. |
b12a03ce TG |
856 | */ |
857 | void hrtimers_resume(void) | |
858 | { | |
53bef3fd | 859 | lockdep_assert_irqs_disabled(); |
5ec2481b | 860 | /* Retrigger on the local CPU */ |
b12a03ce | 861 | retrigger_next_event(NULL); |
5ec2481b TG |
862 | /* And schedule a retrigger for all others */ |
863 | clock_was_set_delayed(); | |
b12a03ce TG |
864 | } |
865 | ||
c0a31329 | 866 | /* |
6506f2aa | 867 | * Counterpart to lock_hrtimer_base above: |
c0a31329 TG |
868 | */ |
869 | static inline | |
870 | void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) | |
871 | { | |
ecb49d1a | 872 | raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags); |
c0a31329 TG |
873 | } |
874 | ||
875 | /** | |
876 | * hrtimer_forward - forward the timer expiry | |
c0a31329 | 877 | * @timer: hrtimer to forward |
44f21475 | 878 | * @now: forward past this time |
c0a31329 TG |
879 | * @interval: the interval to forward |
880 | * | |
881 | * Forward the timer expiry so it will expire in the future. | |
8dca6f33 | 882 | * Returns the number of overruns. |
91e5a217 TG |
883 | * |
884 | * Can be safely called from the callback function of @timer. If | |
885 | * called from other contexts @timer must neither be enqueued nor | |
886 | * running the callback and the caller needs to take care of | |
887 | * serialization. | |
888 | * | |
889 | * Note: This only updates the timer expiry value and does not requeue | |
890 | * the timer. | |
c0a31329 | 891 | */ |
4d672e7a | 892 | u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval) |
c0a31329 | 893 | { |
4d672e7a | 894 | u64 orun = 1; |
44f21475 | 895 | ktime_t delta; |
c0a31329 | 896 | |
cc584b21 | 897 | delta = ktime_sub(now, hrtimer_get_expires(timer)); |
c0a31329 | 898 | |
2456e855 | 899 | if (delta < 0) |
c0a31329 TG |
900 | return 0; |
901 | ||
5de2755c PZ |
902 | if (WARN_ON(timer->state & HRTIMER_STATE_ENQUEUED)) |
903 | return 0; | |
904 | ||
2456e855 TG |
905 | if (interval < hrtimer_resolution) |
906 | interval = hrtimer_resolution; | |
c9db4fa1 | 907 | |
2456e855 | 908 | if (unlikely(delta >= interval)) { |
df869b63 | 909 | s64 incr = ktime_to_ns(interval); |
c0a31329 TG |
910 | |
911 | orun = ktime_divns(delta, incr); | |
cc584b21 | 912 | hrtimer_add_expires_ns(timer, incr * orun); |
2456e855 | 913 | if (hrtimer_get_expires_tv64(timer) > now) |
c0a31329 TG |
914 | return orun; |
915 | /* | |
916 | * This (and the ktime_add() below) is the | |
917 | * correction for exact: | |
918 | */ | |
919 | orun++; | |
920 | } | |
cc584b21 | 921 | hrtimer_add_expires(timer, interval); |
c0a31329 TG |
922 | |
923 | return orun; | |
924 | } | |
6bdb6b62 | 925 | EXPORT_SYMBOL_GPL(hrtimer_forward); |
c0a31329 TG |
926 | |
927 | /* | |
928 | * enqueue_hrtimer - internal function to (re)start a timer | |
929 | * | |
930 | * The timer is inserted in expiry order. Insertion into the | |
931 | * red black tree is O(log(n)). Must hold the base lock. | |
a6037b61 PZ |
932 | * |
933 | * Returns 1 when the new timer is the leftmost timer in the tree. | |
c0a31329 | 934 | */ |
a6037b61 | 935 | static int enqueue_hrtimer(struct hrtimer *timer, |
63e2ed36 AMG |
936 | struct hrtimer_clock_base *base, |
937 | enum hrtimer_mode mode) | |
c0a31329 | 938 | { |
63e2ed36 | 939 | debug_activate(timer, mode); |
237fc6e7 | 940 | |
ab8177bc | 941 | base->cpu_base->active_bases |= 1 << base->index; |
54cdfdb4 | 942 | |
887d9dc9 | 943 | timer->state = HRTIMER_STATE_ENQUEUED; |
a6037b61 | 944 | |
b97f44c9 | 945 | return timerqueue_add(&base->active, &timer->node); |
288867ec | 946 | } |
c0a31329 TG |
947 | |
948 | /* | |
949 | * __remove_hrtimer - internal function to remove a timer | |
950 | * | |
951 | * Caller must hold the base lock. | |
54cdfdb4 TG |
952 | * |
953 | * High resolution timer mode reprograms the clock event device when the | |
954 | * timer is the one which expires next. The caller can disable this by setting | |
955 | * reprogram to zero. This is useful, when the context does a reprogramming | |
956 | * anyway (e.g. timer interrupt) | |
c0a31329 | 957 | */ |
3c8aa39d | 958 | static void __remove_hrtimer(struct hrtimer *timer, |
303e967f | 959 | struct hrtimer_clock_base *base, |
203cbf77 | 960 | u8 newstate, int reprogram) |
c0a31329 | 961 | { |
e19ffe8b | 962 | struct hrtimer_cpu_base *cpu_base = base->cpu_base; |
203cbf77 | 963 | u8 state = timer->state; |
e19ffe8b | 964 | |
895bdfa7 TG |
965 | timer->state = newstate; |
966 | if (!(state & HRTIMER_STATE_ENQUEUED)) | |
967 | return; | |
7403f41f | 968 | |
b97f44c9 | 969 | if (!timerqueue_del(&base->active, &timer->node)) |
e19ffe8b | 970 | cpu_base->active_bases &= ~(1 << base->index); |
7403f41f | 971 | |
895bdfa7 TG |
972 | /* |
973 | * Note: If reprogram is false we do not update | |
974 | * cpu_base->next_timer. This happens when we remove the first | |
975 | * timer on a remote cpu. No harm as we never dereference | |
976 | * cpu_base->next_timer. So the worst thing what can happen is | |
977 | * an superflous call to hrtimer_force_reprogram() on the | |
978 | * remote cpu later on if the same timer gets enqueued again. | |
979 | */ | |
980 | if (reprogram && timer == cpu_base->next_timer) | |
981 | hrtimer_force_reprogram(cpu_base, 1); | |
c0a31329 TG |
982 | } |
983 | ||
984 | /* | |
985 | * remove hrtimer, called with base lock held | |
986 | */ | |
987 | static inline int | |
8edfb036 | 988 | remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base, bool restart) |
c0a31329 | 989 | { |
303e967f | 990 | if (hrtimer_is_queued(timer)) { |
203cbf77 | 991 | u8 state = timer->state; |
54cdfdb4 TG |
992 | int reprogram; |
993 | ||
994 | /* | |
995 | * Remove the timer and force reprogramming when high | |
996 | * resolution mode is active and the timer is on the current | |
997 | * CPU. If we remove a timer on another CPU, reprogramming is | |
998 | * skipped. The interrupt event on this CPU is fired and | |
999 | * reprogramming happens in the interrupt handler. This is a | |
1000 | * rare case and less expensive than a smp call. | |
1001 | */ | |
c6a2a177 | 1002 | debug_deactivate(timer); |
dc5df73b | 1003 | reprogram = base->cpu_base == this_cpu_ptr(&hrtimer_bases); |
8edfb036 | 1004 | |
887d9dc9 PZ |
1005 | if (!restart) |
1006 | state = HRTIMER_STATE_INACTIVE; | |
1007 | ||
f13d4f97 | 1008 | __remove_hrtimer(timer, base, state, reprogram); |
c0a31329 TG |
1009 | return 1; |
1010 | } | |
1011 | return 0; | |
1012 | } | |
1013 | ||
203cbf77 TG |
1014 | static inline ktime_t hrtimer_update_lowres(struct hrtimer *timer, ktime_t tim, |
1015 | const enum hrtimer_mode mode) | |
1016 | { | |
1017 | #ifdef CONFIG_TIME_LOW_RES | |
1018 | /* | |
1019 | * CONFIG_TIME_LOW_RES indicates that the system has no way to return | |
1020 | * granular time values. For relative timers we add hrtimer_resolution | |
1021 | * (i.e. one jiffie) to prevent short timeouts. | |
1022 | */ | |
1023 | timer->is_rel = mode & HRTIMER_MODE_REL; | |
1024 | if (timer->is_rel) | |
8b0e1953 | 1025 | tim = ktime_add_safe(tim, hrtimer_resolution); |
203cbf77 TG |
1026 | #endif |
1027 | return tim; | |
1028 | } | |
1029 | ||
5da70160 AMG |
1030 | static void |
1031 | hrtimer_update_softirq_timer(struct hrtimer_cpu_base *cpu_base, bool reprogram) | |
1032 | { | |
1033 | ktime_t expires; | |
1034 | ||
1035 | /* | |
1036 | * Find the next SOFT expiration. | |
1037 | */ | |
1038 | expires = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_SOFT); | |
1039 | ||
1040 | /* | |
1041 | * reprogramming needs to be triggered, even if the next soft | |
1042 | * hrtimer expires at the same time than the next hard | |
1043 | * hrtimer. cpu_base->softirq_expires_next needs to be updated! | |
1044 | */ | |
1045 | if (expires == KTIME_MAX) | |
1046 | return; | |
1047 | ||
1048 | /* | |
1049 | * cpu_base->*next_timer is recomputed by __hrtimer_get_next_event() | |
1050 | * cpu_base->*expires_next is only set by hrtimer_reprogram() | |
1051 | */ | |
1052 | hrtimer_reprogram(cpu_base->softirq_next_timer, reprogram); | |
1053 | } | |
1054 | ||
138a6b7a AMG |
1055 | static int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, |
1056 | u64 delta_ns, const enum hrtimer_mode mode, | |
1057 | struct hrtimer_clock_base *base) | |
c0a31329 | 1058 | { |
138a6b7a | 1059 | struct hrtimer_clock_base *new_base; |
c0a31329 TG |
1060 | |
1061 | /* Remove an active timer from the queue: */ | |
8edfb036 | 1062 | remove_hrtimer(timer, base, true); |
c0a31329 | 1063 | |
203cbf77 | 1064 | if (mode & HRTIMER_MODE_REL) |
84ea7fe3 | 1065 | tim = ktime_add_safe(tim, base->get_time()); |
203cbf77 TG |
1066 | |
1067 | tim = hrtimer_update_lowres(timer, tim, mode); | |
237fc6e7 | 1068 | |
da8f2e17 | 1069 | hrtimer_set_expires_range_ns(timer, tim, delta_ns); |
c0a31329 | 1070 | |
84ea7fe3 VK |
1071 | /* Switch the timer base, if necessary: */ |
1072 | new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED); | |
1073 | ||
138a6b7a AMG |
1074 | return enqueue_hrtimer(timer, new_base, mode); |
1075 | } | |
5da70160 | 1076 | |
138a6b7a AMG |
1077 | /** |
1078 | * hrtimer_start_range_ns - (re)start an hrtimer | |
1079 | * @timer: the timer to be added | |
1080 | * @tim: expiry time | |
1081 | * @delta_ns: "slack" range for the timer | |
1082 | * @mode: timer mode: absolute (HRTIMER_MODE_ABS) or | |
5da70160 AMG |
1083 | * relative (HRTIMER_MODE_REL), and pinned (HRTIMER_MODE_PINNED); |
1084 | * softirq based mode is considered for debug purpose only! | |
138a6b7a AMG |
1085 | */ |
1086 | void hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, | |
1087 | u64 delta_ns, const enum hrtimer_mode mode) | |
1088 | { | |
1089 | struct hrtimer_clock_base *base; | |
1090 | unsigned long flags; | |
1091 | ||
5da70160 AMG |
1092 | /* |
1093 | * Check whether the HRTIMER_MODE_SOFT bit and hrtimer.is_soft | |
1094 | * match. | |
1095 | */ | |
1096 | WARN_ON_ONCE(!(mode & HRTIMER_MODE_SOFT) ^ !timer->is_soft); | |
1097 | ||
138a6b7a AMG |
1098 | base = lock_hrtimer_base(timer, &flags); |
1099 | ||
1100 | if (__hrtimer_start_range_ns(timer, tim, delta_ns, mode, base)) | |
5da70160 | 1101 | hrtimer_reprogram(timer, true); |
49a2a075 | 1102 | |
c0a31329 | 1103 | unlock_hrtimer_base(timer, &flags); |
7f1e2ca9 | 1104 | } |
da8f2e17 AV |
1105 | EXPORT_SYMBOL_GPL(hrtimer_start_range_ns); |
1106 | ||
c0a31329 TG |
1107 | /** |
1108 | * hrtimer_try_to_cancel - try to deactivate a timer | |
c0a31329 TG |
1109 | * @timer: hrtimer to stop |
1110 | * | |
1111 | * Returns: | |
1112 | * 0 when the timer was not active | |
1113 | * 1 when the timer was active | |
0ba42a59 | 1114 | * -1 when the timer is currently executing the callback function and |
fa9799e3 | 1115 | * cannot be stopped |
c0a31329 TG |
1116 | */ |
1117 | int hrtimer_try_to_cancel(struct hrtimer *timer) | |
1118 | { | |
3c8aa39d | 1119 | struct hrtimer_clock_base *base; |
c0a31329 TG |
1120 | unsigned long flags; |
1121 | int ret = -1; | |
1122 | ||
19d9f422 TG |
1123 | /* |
1124 | * Check lockless first. If the timer is not active (neither | |
1125 | * enqueued nor running the callback, nothing to do here. The | |
1126 | * base lock does not serialize against a concurrent enqueue, | |
1127 | * so we can avoid taking it. | |
1128 | */ | |
1129 | if (!hrtimer_active(timer)) | |
1130 | return 0; | |
1131 | ||
c0a31329 TG |
1132 | base = lock_hrtimer_base(timer, &flags); |
1133 | ||
303e967f | 1134 | if (!hrtimer_callback_running(timer)) |
8edfb036 | 1135 | ret = remove_hrtimer(timer, base, false); |
c0a31329 TG |
1136 | |
1137 | unlock_hrtimer_base(timer, &flags); | |
1138 | ||
1139 | return ret; | |
1140 | ||
1141 | } | |
8d16b764 | 1142 | EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel); |
c0a31329 TG |
1143 | |
1144 | /** | |
1145 | * hrtimer_cancel - cancel a timer and wait for the handler to finish. | |
c0a31329 TG |
1146 | * @timer: the timer to be cancelled |
1147 | * | |
1148 | * Returns: | |
1149 | * 0 when the timer was not active | |
1150 | * 1 when the timer was active | |
1151 | */ | |
1152 | int hrtimer_cancel(struct hrtimer *timer) | |
1153 | { | |
1154 | for (;;) { | |
1155 | int ret = hrtimer_try_to_cancel(timer); | |
1156 | ||
1157 | if (ret >= 0) | |
1158 | return ret; | |
5ef37b19 | 1159 | cpu_relax(); |
c0a31329 TG |
1160 | } |
1161 | } | |
8d16b764 | 1162 | EXPORT_SYMBOL_GPL(hrtimer_cancel); |
c0a31329 TG |
1163 | |
1164 | /** | |
1165 | * hrtimer_get_remaining - get remaining time for the timer | |
c0a31329 | 1166 | * @timer: the timer to read |
203cbf77 | 1167 | * @adjust: adjust relative timers when CONFIG_TIME_LOW_RES=y |
c0a31329 | 1168 | */ |
203cbf77 | 1169 | ktime_t __hrtimer_get_remaining(const struct hrtimer *timer, bool adjust) |
c0a31329 | 1170 | { |
c0a31329 TG |
1171 | unsigned long flags; |
1172 | ktime_t rem; | |
1173 | ||
b3bd3de6 | 1174 | lock_hrtimer_base(timer, &flags); |
203cbf77 TG |
1175 | if (IS_ENABLED(CONFIG_TIME_LOW_RES) && adjust) |
1176 | rem = hrtimer_expires_remaining_adjusted(timer); | |
1177 | else | |
1178 | rem = hrtimer_expires_remaining(timer); | |
c0a31329 TG |
1179 | unlock_hrtimer_base(timer, &flags); |
1180 | ||
1181 | return rem; | |
1182 | } | |
203cbf77 | 1183 | EXPORT_SYMBOL_GPL(__hrtimer_get_remaining); |
c0a31329 | 1184 | |
3451d024 | 1185 | #ifdef CONFIG_NO_HZ_COMMON |
69239749 TL |
1186 | /** |
1187 | * hrtimer_get_next_event - get the time until next expiry event | |
1188 | * | |
c1ad348b | 1189 | * Returns the next expiry time or KTIME_MAX if no timer is pending. |
69239749 | 1190 | */ |
c1ad348b | 1191 | u64 hrtimer_get_next_event(void) |
69239749 | 1192 | { |
dc5df73b | 1193 | struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases); |
c1ad348b | 1194 | u64 expires = KTIME_MAX; |
69239749 | 1195 | unsigned long flags; |
69239749 | 1196 | |
ecb49d1a | 1197 | raw_spin_lock_irqsave(&cpu_base->lock, flags); |
3c8aa39d | 1198 | |
e19ffe8b | 1199 | if (!__hrtimer_hres_active(cpu_base)) |
5da70160 | 1200 | expires = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_ALL); |
3c8aa39d | 1201 | |
ecb49d1a | 1202 | raw_spin_unlock_irqrestore(&cpu_base->lock, flags); |
3c8aa39d | 1203 | |
c1ad348b | 1204 | return expires; |
69239749 TL |
1205 | } |
1206 | #endif | |
1207 | ||
336a9cde MZ |
1208 | static inline int hrtimer_clockid_to_base(clockid_t clock_id) |
1209 | { | |
1210 | if (likely(clock_id < MAX_CLOCKS)) { | |
1211 | int base = hrtimer_clock_to_base_table[clock_id]; | |
1212 | ||
1213 | if (likely(base != HRTIMER_MAX_CLOCK_BASES)) | |
1214 | return base; | |
1215 | } | |
1216 | WARN(1, "Invalid clockid %d. Using MONOTONIC\n", clock_id); | |
1217 | return HRTIMER_BASE_MONOTONIC; | |
1218 | } | |
1219 | ||
237fc6e7 TG |
1220 | static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id, |
1221 | enum hrtimer_mode mode) | |
c0a31329 | 1222 | { |
3c8aa39d | 1223 | struct hrtimer_cpu_base *cpu_base; |
e06383db | 1224 | int base; |
c0a31329 | 1225 | |
7978672c GA |
1226 | memset(timer, 0, sizeof(struct hrtimer)); |
1227 | ||
22127e93 | 1228 | cpu_base = raw_cpu_ptr(&hrtimer_bases); |
c0a31329 | 1229 | |
48d0c9be AMG |
1230 | /* |
1231 | * POSIX magic: Relative CLOCK_REALTIME timers are not affected by | |
1232 | * clock modifications, so they needs to become CLOCK_MONOTONIC to | |
1233 | * ensure POSIX compliance. | |
1234 | */ | |
1235 | if (clock_id == CLOCK_REALTIME && mode & HRTIMER_MODE_REL) | |
7978672c GA |
1236 | clock_id = CLOCK_MONOTONIC; |
1237 | ||
e06383db JS |
1238 | base = hrtimer_clockid_to_base(clock_id); |
1239 | timer->base = &cpu_base->clock_base[base]; | |
998adc3d | 1240 | timerqueue_init(&timer->node); |
c0a31329 | 1241 | } |
237fc6e7 TG |
1242 | |
1243 | /** | |
1244 | * hrtimer_init - initialize a timer to the given clock | |
1245 | * @timer: the timer to be initialized | |
1246 | * @clock_id: the clock to be used | |
6de6250c AMG |
1247 | * @mode: timer mode: absolute (HRTIMER_MODE_ABS) or |
1248 | * relative (HRTIMER_MODE_REL); pinned is not considered here! | |
237fc6e7 TG |
1249 | */ |
1250 | void hrtimer_init(struct hrtimer *timer, clockid_t clock_id, | |
1251 | enum hrtimer_mode mode) | |
1252 | { | |
c6a2a177 | 1253 | debug_init(timer, clock_id, mode); |
237fc6e7 TG |
1254 | __hrtimer_init(timer, clock_id, mode); |
1255 | } | |
8d16b764 | 1256 | EXPORT_SYMBOL_GPL(hrtimer_init); |
c0a31329 | 1257 | |
887d9dc9 PZ |
1258 | /* |
1259 | * A timer is active, when it is enqueued into the rbtree or the | |
1260 | * callback function is running or it's in the state of being migrated | |
1261 | * to another cpu. | |
c0a31329 | 1262 | * |
887d9dc9 | 1263 | * It is important for this function to not return a false negative. |
c0a31329 | 1264 | */ |
887d9dc9 | 1265 | bool hrtimer_active(const struct hrtimer *timer) |
c0a31329 | 1266 | { |
3f0b9e8e | 1267 | struct hrtimer_clock_base *base; |
887d9dc9 | 1268 | unsigned int seq; |
c0a31329 | 1269 | |
887d9dc9 | 1270 | do { |
3f0b9e8e AMG |
1271 | base = READ_ONCE(timer->base); |
1272 | seq = raw_read_seqcount_begin(&base->seq); | |
c0a31329 | 1273 | |
887d9dc9 | 1274 | if (timer->state != HRTIMER_STATE_INACTIVE || |
3f0b9e8e | 1275 | base->running == timer) |
887d9dc9 PZ |
1276 | return true; |
1277 | ||
3f0b9e8e AMG |
1278 | } while (read_seqcount_retry(&base->seq, seq) || |
1279 | base != READ_ONCE(timer->base)); | |
887d9dc9 PZ |
1280 | |
1281 | return false; | |
c0a31329 | 1282 | } |
887d9dc9 | 1283 | EXPORT_SYMBOL_GPL(hrtimer_active); |
c0a31329 | 1284 | |
887d9dc9 PZ |
1285 | /* |
1286 | * The write_seqcount_barrier()s in __run_hrtimer() split the thing into 3 | |
1287 | * distinct sections: | |
1288 | * | |
1289 | * - queued: the timer is queued | |
1290 | * - callback: the timer is being ran | |
1291 | * - post: the timer is inactive or (re)queued | |
1292 | * | |
1293 | * On the read side we ensure we observe timer->state and cpu_base->running | |
1294 | * from the same section, if anything changed while we looked at it, we retry. | |
1295 | * This includes timer->base changing because sequence numbers alone are | |
1296 | * insufficient for that. | |
1297 | * | |
1298 | * The sequence numbers are required because otherwise we could still observe | |
1299 | * a false negative if the read side got smeared over multiple consequtive | |
1300 | * __run_hrtimer() invocations. | |
1301 | */ | |
1302 | ||
21d6d52a TG |
1303 | static void __run_hrtimer(struct hrtimer_cpu_base *cpu_base, |
1304 | struct hrtimer_clock_base *base, | |
dd934aa8 AMG |
1305 | struct hrtimer *timer, ktime_t *now, |
1306 | unsigned long flags) | |
d3d74453 | 1307 | { |
d3d74453 PZ |
1308 | enum hrtimer_restart (*fn)(struct hrtimer *); |
1309 | int restart; | |
1310 | ||
887d9dc9 | 1311 | lockdep_assert_held(&cpu_base->lock); |
ca109491 | 1312 | |
c6a2a177 | 1313 | debug_deactivate(timer); |
3f0b9e8e | 1314 | base->running = timer; |
887d9dc9 PZ |
1315 | |
1316 | /* | |
1317 | * Separate the ->running assignment from the ->state assignment. | |
1318 | * | |
1319 | * As with a regular write barrier, this ensures the read side in | |
3f0b9e8e | 1320 | * hrtimer_active() cannot observe base->running == NULL && |
887d9dc9 PZ |
1321 | * timer->state == INACTIVE. |
1322 | */ | |
3f0b9e8e | 1323 | raw_write_seqcount_barrier(&base->seq); |
887d9dc9 PZ |
1324 | |
1325 | __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE, 0); | |
d3d74453 | 1326 | fn = timer->function; |
ca109491 | 1327 | |
203cbf77 TG |
1328 | /* |
1329 | * Clear the 'is relative' flag for the TIME_LOW_RES case. If the | |
1330 | * timer is restarted with a period then it becomes an absolute | |
1331 | * timer. If its not restarted it does not matter. | |
1332 | */ | |
1333 | if (IS_ENABLED(CONFIG_TIME_LOW_RES)) | |
1334 | timer->is_rel = false; | |
1335 | ||
ca109491 | 1336 | /* |
d05ca13b TG |
1337 | * The timer is marked as running in the CPU base, so it is |
1338 | * protected against migration to a different CPU even if the lock | |
1339 | * is dropped. | |
ca109491 | 1340 | */ |
dd934aa8 | 1341 | raw_spin_unlock_irqrestore(&cpu_base->lock, flags); |
c6a2a177 | 1342 | trace_hrtimer_expire_entry(timer, now); |
ca109491 | 1343 | restart = fn(timer); |
c6a2a177 | 1344 | trace_hrtimer_expire_exit(timer); |
dd934aa8 | 1345 | raw_spin_lock_irq(&cpu_base->lock); |
d3d74453 PZ |
1346 | |
1347 | /* | |
887d9dc9 | 1348 | * Note: We clear the running state after enqueue_hrtimer and |
b4d90e9f | 1349 | * we do not reprogram the event hardware. Happens either in |
e3f1d883 | 1350 | * hrtimer_start_range_ns() or in hrtimer_interrupt() |
5de2755c PZ |
1351 | * |
1352 | * Note: Because we dropped the cpu_base->lock above, | |
1353 | * hrtimer_start_range_ns() can have popped in and enqueued the timer | |
1354 | * for us already. | |
d3d74453 | 1355 | */ |
5de2755c PZ |
1356 | if (restart != HRTIMER_NORESTART && |
1357 | !(timer->state & HRTIMER_STATE_ENQUEUED)) | |
63e2ed36 | 1358 | enqueue_hrtimer(timer, base, HRTIMER_MODE_ABS); |
f13d4f97 | 1359 | |
887d9dc9 PZ |
1360 | /* |
1361 | * Separate the ->running assignment from the ->state assignment. | |
1362 | * | |
1363 | * As with a regular write barrier, this ensures the read side in | |
3f0b9e8e | 1364 | * hrtimer_active() cannot observe base->running.timer == NULL && |
887d9dc9 PZ |
1365 | * timer->state == INACTIVE. |
1366 | */ | |
3f0b9e8e | 1367 | raw_write_seqcount_barrier(&base->seq); |
f13d4f97 | 1368 | |
3f0b9e8e AMG |
1369 | WARN_ON_ONCE(base->running != timer); |
1370 | base->running = NULL; | |
d3d74453 PZ |
1371 | } |
1372 | ||
dd934aa8 | 1373 | static void __hrtimer_run_queues(struct hrtimer_cpu_base *cpu_base, ktime_t now, |
c458b1d1 | 1374 | unsigned long flags, unsigned int active_mask) |
54cdfdb4 | 1375 | { |
c272ca58 | 1376 | struct hrtimer_clock_base *base; |
c458b1d1 | 1377 | unsigned int active = cpu_base->active_bases & active_mask; |
6ff7041d | 1378 | |
c272ca58 | 1379 | for_each_active_base(base, cpu_base, active) { |
998adc3d | 1380 | struct timerqueue_node *node; |
ab8177bc TG |
1381 | ktime_t basenow; |
1382 | ||
54cdfdb4 TG |
1383 | basenow = ktime_add(now, base->offset); |
1384 | ||
998adc3d | 1385 | while ((node = timerqueue_getnext(&base->active))) { |
54cdfdb4 TG |
1386 | struct hrtimer *timer; |
1387 | ||
998adc3d | 1388 | timer = container_of(node, struct hrtimer, node); |
54cdfdb4 | 1389 | |
654c8e0b AV |
1390 | /* |
1391 | * The immediate goal for using the softexpires is | |
1392 | * minimizing wakeups, not running timers at the | |
1393 | * earliest interrupt after their soft expiration. | |
1394 | * This allows us to avoid using a Priority Search | |
1395 | * Tree, which can answer a stabbing querry for | |
1396 | * overlapping intervals and instead use the simple | |
1397 | * BST we already have. | |
1398 | * We don't add extra wakeups by delaying timers that | |
1399 | * are right-of a not yet expired timer, because that | |
1400 | * timer will have to trigger a wakeup anyway. | |
1401 | */ | |
2456e855 | 1402 | if (basenow < hrtimer_get_softexpires_tv64(timer)) |
54cdfdb4 | 1403 | break; |
54cdfdb4 | 1404 | |
dd934aa8 | 1405 | __run_hrtimer(cpu_base, base, timer, &basenow, flags); |
54cdfdb4 | 1406 | } |
54cdfdb4 | 1407 | } |
21d6d52a TG |
1408 | } |
1409 | ||
5da70160 AMG |
1410 | static __latent_entropy void hrtimer_run_softirq(struct softirq_action *h) |
1411 | { | |
1412 | struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases); | |
1413 | unsigned long flags; | |
1414 | ktime_t now; | |
1415 | ||
1416 | raw_spin_lock_irqsave(&cpu_base->lock, flags); | |
1417 | ||
1418 | now = hrtimer_update_base(cpu_base); | |
1419 | __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_SOFT); | |
1420 | ||
1421 | cpu_base->softirq_activated = 0; | |
1422 | hrtimer_update_softirq_timer(cpu_base, true); | |
1423 | ||
1424 | raw_spin_unlock_irqrestore(&cpu_base->lock, flags); | |
1425 | } | |
1426 | ||
21d6d52a TG |
1427 | #ifdef CONFIG_HIGH_RES_TIMERS |
1428 | ||
1429 | /* | |
1430 | * High resolution timer interrupt | |
1431 | * Called with interrupts disabled | |
1432 | */ | |
1433 | void hrtimer_interrupt(struct clock_event_device *dev) | |
1434 | { | |
1435 | struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases); | |
1436 | ktime_t expires_next, now, entry_time, delta; | |
dd934aa8 | 1437 | unsigned long flags; |
21d6d52a TG |
1438 | int retries = 0; |
1439 | ||
1440 | BUG_ON(!cpu_base->hres_active); | |
1441 | cpu_base->nr_events++; | |
2456e855 | 1442 | dev->next_event = KTIME_MAX; |
21d6d52a | 1443 | |
dd934aa8 | 1444 | raw_spin_lock_irqsave(&cpu_base->lock, flags); |
21d6d52a TG |
1445 | entry_time = now = hrtimer_update_base(cpu_base); |
1446 | retry: | |
1447 | cpu_base->in_hrtirq = 1; | |
1448 | /* | |
1449 | * We set expires_next to KTIME_MAX here with cpu_base->lock | |
1450 | * held to prevent that a timer is enqueued in our queue via | |
1451 | * the migration code. This does not affect enqueueing of | |
1452 | * timers which run their callback and need to be requeued on | |
1453 | * this CPU. | |
1454 | */ | |
2456e855 | 1455 | cpu_base->expires_next = KTIME_MAX; |
21d6d52a | 1456 | |
5da70160 AMG |
1457 | if (!ktime_before(now, cpu_base->softirq_expires_next)) { |
1458 | cpu_base->softirq_expires_next = KTIME_MAX; | |
1459 | cpu_base->softirq_activated = 1; | |
1460 | raise_softirq_irqoff(HRTIMER_SOFTIRQ); | |
1461 | } | |
1462 | ||
c458b1d1 | 1463 | __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_HARD); |
21d6d52a | 1464 | |
9bc74919 | 1465 | /* Reevaluate the clock bases for the next expiry */ |
5da70160 | 1466 | expires_next = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_ALL); |
6ff7041d TG |
1467 | /* |
1468 | * Store the new expiry value so the migration code can verify | |
1469 | * against it. | |
1470 | */ | |
54cdfdb4 | 1471 | cpu_base->expires_next = expires_next; |
9bc74919 | 1472 | cpu_base->in_hrtirq = 0; |
dd934aa8 | 1473 | raw_spin_unlock_irqrestore(&cpu_base->lock, flags); |
54cdfdb4 TG |
1474 | |
1475 | /* Reprogramming necessary ? */ | |
d2540875 | 1476 | if (!tick_program_event(expires_next, 0)) { |
41d2e494 TG |
1477 | cpu_base->hang_detected = 0; |
1478 | return; | |
54cdfdb4 | 1479 | } |
41d2e494 TG |
1480 | |
1481 | /* | |
1482 | * The next timer was already expired due to: | |
1483 | * - tracing | |
1484 | * - long lasting callbacks | |
1485 | * - being scheduled away when running in a VM | |
1486 | * | |
1487 | * We need to prevent that we loop forever in the hrtimer | |
1488 | * interrupt routine. We give it 3 attempts to avoid | |
1489 | * overreacting on some spurious event. | |
5baefd6d JS |
1490 | * |
1491 | * Acquire base lock for updating the offsets and retrieving | |
1492 | * the current time. | |
41d2e494 | 1493 | */ |
dd934aa8 | 1494 | raw_spin_lock_irqsave(&cpu_base->lock, flags); |
5baefd6d | 1495 | now = hrtimer_update_base(cpu_base); |
41d2e494 TG |
1496 | cpu_base->nr_retries++; |
1497 | if (++retries < 3) | |
1498 | goto retry; | |
1499 | /* | |
1500 | * Give the system a chance to do something else than looping | |
1501 | * here. We stored the entry time, so we know exactly how long | |
1502 | * we spent here. We schedule the next event this amount of | |
1503 | * time away. | |
1504 | */ | |
1505 | cpu_base->nr_hangs++; | |
1506 | cpu_base->hang_detected = 1; | |
dd934aa8 AMG |
1507 | raw_spin_unlock_irqrestore(&cpu_base->lock, flags); |
1508 | ||
41d2e494 | 1509 | delta = ktime_sub(now, entry_time); |
2456e855 TG |
1510 | if ((unsigned int)delta > cpu_base->max_hang_time) |
1511 | cpu_base->max_hang_time = (unsigned int) delta; | |
41d2e494 TG |
1512 | /* |
1513 | * Limit it to a sensible value as we enforce a longer | |
1514 | * delay. Give the CPU at least 100ms to catch up. | |
1515 | */ | |
2456e855 | 1516 | if (delta > 100 * NSEC_PER_MSEC) |
41d2e494 TG |
1517 | expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC); |
1518 | else | |
1519 | expires_next = ktime_add(now, delta); | |
1520 | tick_program_event(expires_next, 1); | |
1521 | printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n", | |
1522 | ktime_to_ns(delta)); | |
54cdfdb4 TG |
1523 | } |
1524 | ||
016da201 | 1525 | /* called with interrupts disabled */ |
c6eb3f70 | 1526 | static inline void __hrtimer_peek_ahead_timers(void) |
8bdec955 TG |
1527 | { |
1528 | struct tick_device *td; | |
1529 | ||
1530 | if (!hrtimer_hres_active()) | |
1531 | return; | |
1532 | ||
22127e93 | 1533 | td = this_cpu_ptr(&tick_cpu_device); |
8bdec955 TG |
1534 | if (td && td->evtdev) |
1535 | hrtimer_interrupt(td->evtdev); | |
1536 | } | |
1537 | ||
82c5b7b5 IM |
1538 | #else /* CONFIG_HIGH_RES_TIMERS */ |
1539 | ||
1540 | static inline void __hrtimer_peek_ahead_timers(void) { } | |
1541 | ||
1542 | #endif /* !CONFIG_HIGH_RES_TIMERS */ | |
82f67cd9 | 1543 | |
d3d74453 | 1544 | /* |
c6eb3f70 | 1545 | * Called from run_local_timers in hardirq context every jiffy |
d3d74453 | 1546 | */ |
833883d9 | 1547 | void hrtimer_run_queues(void) |
d3d74453 | 1548 | { |
dc5df73b | 1549 | struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases); |
dd934aa8 | 1550 | unsigned long flags; |
21d6d52a | 1551 | ktime_t now; |
c0a31329 | 1552 | |
e19ffe8b | 1553 | if (__hrtimer_hres_active(cpu_base)) |
d3d74453 | 1554 | return; |
54cdfdb4 | 1555 | |
d3d74453 | 1556 | /* |
c6eb3f70 TG |
1557 | * This _is_ ugly: We have to check periodically, whether we |
1558 | * can switch to highres and / or nohz mode. The clocksource | |
1559 | * switch happens with xtime_lock held. Notification from | |
1560 | * there only sets the check bit in the tick_oneshot code, | |
1561 | * otherwise we might deadlock vs. xtime_lock. | |
d3d74453 | 1562 | */ |
c6eb3f70 | 1563 | if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) { |
d3d74453 | 1564 | hrtimer_switch_to_hres(); |
3055adda | 1565 | return; |
833883d9 | 1566 | } |
c6eb3f70 | 1567 | |
dd934aa8 | 1568 | raw_spin_lock_irqsave(&cpu_base->lock, flags); |
21d6d52a | 1569 | now = hrtimer_update_base(cpu_base); |
5da70160 AMG |
1570 | |
1571 | if (!ktime_before(now, cpu_base->softirq_expires_next)) { | |
1572 | cpu_base->softirq_expires_next = KTIME_MAX; | |
1573 | cpu_base->softirq_activated = 1; | |
1574 | raise_softirq_irqoff(HRTIMER_SOFTIRQ); | |
1575 | } | |
1576 | ||
c458b1d1 | 1577 | __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_HARD); |
dd934aa8 | 1578 | raw_spin_unlock_irqrestore(&cpu_base->lock, flags); |
c0a31329 TG |
1579 | } |
1580 | ||
10c94ec1 TG |
1581 | /* |
1582 | * Sleep related functions: | |
1583 | */ | |
c9cb2e3d | 1584 | static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer) |
00362e33 TG |
1585 | { |
1586 | struct hrtimer_sleeper *t = | |
1587 | container_of(timer, struct hrtimer_sleeper, timer); | |
1588 | struct task_struct *task = t->task; | |
1589 | ||
1590 | t->task = NULL; | |
1591 | if (task) | |
1592 | wake_up_process(task); | |
1593 | ||
1594 | return HRTIMER_NORESTART; | |
1595 | } | |
1596 | ||
36c8b586 | 1597 | void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task) |
00362e33 TG |
1598 | { |
1599 | sl->timer.function = hrtimer_wakeup; | |
1600 | sl->task = task; | |
1601 | } | |
2bc481cf | 1602 | EXPORT_SYMBOL_GPL(hrtimer_init_sleeper); |
00362e33 | 1603 | |
c0edd7c9 | 1604 | int nanosleep_copyout(struct restart_block *restart, struct timespec64 *ts) |
ce41aaf4 AV |
1605 | { |
1606 | switch(restart->nanosleep.type) { | |
1607 | #ifdef CONFIG_COMPAT | |
1608 | case TT_COMPAT: | |
c0edd7c9 | 1609 | if (compat_put_timespec64(ts, restart->nanosleep.compat_rmtp)) |
ce41aaf4 AV |
1610 | return -EFAULT; |
1611 | break; | |
1612 | #endif | |
1613 | case TT_NATIVE: | |
c0edd7c9 | 1614 | if (put_timespec64(ts, restart->nanosleep.rmtp)) |
ce41aaf4 AV |
1615 | return -EFAULT; |
1616 | break; | |
1617 | default: | |
1618 | BUG(); | |
1619 | } | |
1620 | return -ERESTART_RESTARTBLOCK; | |
1621 | } | |
1622 | ||
669d7868 | 1623 | static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode) |
432569bb | 1624 | { |
edbeda46 AV |
1625 | struct restart_block *restart; |
1626 | ||
669d7868 | 1627 | hrtimer_init_sleeper(t, current); |
10c94ec1 | 1628 | |
432569bb RZ |
1629 | do { |
1630 | set_current_state(TASK_INTERRUPTIBLE); | |
cc584b21 | 1631 | hrtimer_start_expires(&t->timer, mode); |
432569bb | 1632 | |
54cdfdb4 | 1633 | if (likely(t->task)) |
b0f8c44f | 1634 | freezable_schedule(); |
432569bb | 1635 | |
669d7868 | 1636 | hrtimer_cancel(&t->timer); |
c9cb2e3d | 1637 | mode = HRTIMER_MODE_ABS; |
669d7868 TG |
1638 | |
1639 | } while (t->task && !signal_pending(current)); | |
432569bb | 1640 | |
3588a085 PZ |
1641 | __set_current_state(TASK_RUNNING); |
1642 | ||
a7602681 | 1643 | if (!t->task) |
080344b9 | 1644 | return 0; |
080344b9 | 1645 | |
edbeda46 AV |
1646 | restart = ¤t->restart_block; |
1647 | if (restart->nanosleep.type != TT_NONE) { | |
a7602681 | 1648 | ktime_t rem = hrtimer_expires_remaining(&t->timer); |
c0edd7c9 | 1649 | struct timespec64 rmt; |
edbeda46 | 1650 | |
a7602681 AV |
1651 | if (rem <= 0) |
1652 | return 0; | |
c0edd7c9 | 1653 | rmt = ktime_to_timespec64(rem); |
a7602681 | 1654 | |
ce41aaf4 | 1655 | return nanosleep_copyout(restart, &rmt); |
a7602681 AV |
1656 | } |
1657 | return -ERESTART_RESTARTBLOCK; | |
080344b9 ON |
1658 | } |
1659 | ||
fb923c4a | 1660 | static long __sched hrtimer_nanosleep_restart(struct restart_block *restart) |
10c94ec1 | 1661 | { |
669d7868 | 1662 | struct hrtimer_sleeper t; |
a7602681 | 1663 | int ret; |
10c94ec1 | 1664 | |
ab8177bc | 1665 | hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid, |
237fc6e7 | 1666 | HRTIMER_MODE_ABS); |
cc584b21 | 1667 | hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires); |
10c94ec1 | 1668 | |
a7602681 | 1669 | ret = do_nanosleep(&t, HRTIMER_MODE_ABS); |
237fc6e7 TG |
1670 | destroy_hrtimer_on_stack(&t.timer); |
1671 | return ret; | |
10c94ec1 TG |
1672 | } |
1673 | ||
938e7cf2 | 1674 | long hrtimer_nanosleep(const struct timespec64 *rqtp, |
10c94ec1 TG |
1675 | const enum hrtimer_mode mode, const clockid_t clockid) |
1676 | { | |
a7602681 | 1677 | struct restart_block *restart; |
669d7868 | 1678 | struct hrtimer_sleeper t; |
237fc6e7 | 1679 | int ret = 0; |
da8b44d5 | 1680 | u64 slack; |
3bd01206 AV |
1681 | |
1682 | slack = current->timer_slack_ns; | |
aab03e05 | 1683 | if (dl_task(current) || rt_task(current)) |
3bd01206 | 1684 | slack = 0; |
10c94ec1 | 1685 | |
237fc6e7 | 1686 | hrtimer_init_on_stack(&t.timer, clockid, mode); |
ad196384 | 1687 | hrtimer_set_expires_range_ns(&t.timer, timespec64_to_ktime(*rqtp), slack); |
a7602681 AV |
1688 | ret = do_nanosleep(&t, mode); |
1689 | if (ret != -ERESTART_RESTARTBLOCK) | |
237fc6e7 | 1690 | goto out; |
10c94ec1 | 1691 | |
7978672c | 1692 | /* Absolute timers do not update the rmtp value and restart: */ |
237fc6e7 TG |
1693 | if (mode == HRTIMER_MODE_ABS) { |
1694 | ret = -ERESTARTNOHAND; | |
1695 | goto out; | |
1696 | } | |
10c94ec1 | 1697 | |
a7602681 | 1698 | restart = ¤t->restart_block; |
1711ef38 | 1699 | restart->fn = hrtimer_nanosleep_restart; |
ab8177bc | 1700 | restart->nanosleep.clockid = t.timer.base->clockid; |
cc584b21 | 1701 | restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer); |
237fc6e7 TG |
1702 | out: |
1703 | destroy_hrtimer_on_stack(&t.timer); | |
1704 | return ret; | |
10c94ec1 TG |
1705 | } |
1706 | ||
58fd3aa2 HC |
1707 | SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp, |
1708 | struct timespec __user *, rmtp) | |
6ba1b912 | 1709 | { |
c0edd7c9 | 1710 | struct timespec64 tu; |
6ba1b912 | 1711 | |
c0edd7c9 | 1712 | if (get_timespec64(&tu, rqtp)) |
6ba1b912 TG |
1713 | return -EFAULT; |
1714 | ||
c0edd7c9 | 1715 | if (!timespec64_valid(&tu)) |
6ba1b912 TG |
1716 | return -EINVAL; |
1717 | ||
edbeda46 | 1718 | current->restart_block.nanosleep.type = rmtp ? TT_NATIVE : TT_NONE; |
192a82f9 | 1719 | current->restart_block.nanosleep.rmtp = rmtp; |
c0edd7c9 | 1720 | return hrtimer_nanosleep(&tu, HRTIMER_MODE_REL, CLOCK_MONOTONIC); |
6ba1b912 TG |
1721 | } |
1722 | ||
edbeda46 AV |
1723 | #ifdef CONFIG_COMPAT |
1724 | ||
1725 | COMPAT_SYSCALL_DEFINE2(nanosleep, struct compat_timespec __user *, rqtp, | |
1726 | struct compat_timespec __user *, rmtp) | |
1727 | { | |
c0edd7c9 | 1728 | struct timespec64 tu; |
edbeda46 | 1729 | |
c0edd7c9 | 1730 | if (compat_get_timespec64(&tu, rqtp)) |
edbeda46 AV |
1731 | return -EFAULT; |
1732 | ||
c0edd7c9 | 1733 | if (!timespec64_valid(&tu)) |
edbeda46 AV |
1734 | return -EINVAL; |
1735 | ||
1736 | current->restart_block.nanosleep.type = rmtp ? TT_COMPAT : TT_NONE; | |
1737 | current->restart_block.nanosleep.compat_rmtp = rmtp; | |
c0edd7c9 | 1738 | return hrtimer_nanosleep(&tu, HRTIMER_MODE_REL, CLOCK_MONOTONIC); |
edbeda46 AV |
1739 | } |
1740 | #endif | |
1741 | ||
c0a31329 TG |
1742 | /* |
1743 | * Functions related to boot-time initialization: | |
1744 | */ | |
27590dc1 | 1745 | int hrtimers_prepare_cpu(unsigned int cpu) |
c0a31329 | 1746 | { |
3c8aa39d | 1747 | struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu); |
c0a31329 TG |
1748 | int i; |
1749 | ||
998adc3d | 1750 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
3c8aa39d | 1751 | cpu_base->clock_base[i].cpu_base = cpu_base; |
998adc3d JS |
1752 | timerqueue_init_head(&cpu_base->clock_base[i].active); |
1753 | } | |
3c8aa39d | 1754 | |
cddd0248 | 1755 | cpu_base->cpu = cpu; |
28bfd18b | 1756 | cpu_base->hres_active = 0; |
07a9a7ea | 1757 | cpu_base->expires_next = KTIME_MAX; |
5da70160 | 1758 | cpu_base->softirq_expires_next = KTIME_MAX; |
27590dc1 | 1759 | return 0; |
c0a31329 TG |
1760 | } |
1761 | ||
1762 | #ifdef CONFIG_HOTPLUG_CPU | |
1763 | ||
ca109491 | 1764 | static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base, |
37810659 | 1765 | struct hrtimer_clock_base *new_base) |
c0a31329 TG |
1766 | { |
1767 | struct hrtimer *timer; | |
998adc3d | 1768 | struct timerqueue_node *node; |
c0a31329 | 1769 | |
998adc3d JS |
1770 | while ((node = timerqueue_getnext(&old_base->active))) { |
1771 | timer = container_of(node, struct hrtimer, node); | |
54cdfdb4 | 1772 | BUG_ON(hrtimer_callback_running(timer)); |
c6a2a177 | 1773 | debug_deactivate(timer); |
b00c1a99 TG |
1774 | |
1775 | /* | |
c04dca02 | 1776 | * Mark it as ENQUEUED not INACTIVE otherwise the |
b00c1a99 TG |
1777 | * timer could be seen as !active and just vanish away |
1778 | * under us on another CPU | |
1779 | */ | |
c04dca02 | 1780 | __remove_hrtimer(timer, old_base, HRTIMER_STATE_ENQUEUED, 0); |
c0a31329 | 1781 | timer->base = new_base; |
54cdfdb4 | 1782 | /* |
e3f1d883 TG |
1783 | * Enqueue the timers on the new cpu. This does not |
1784 | * reprogram the event device in case the timer | |
1785 | * expires before the earliest on this CPU, but we run | |
1786 | * hrtimer_interrupt after we migrated everything to | |
1787 | * sort out already expired timers and reprogram the | |
1788 | * event device. | |
54cdfdb4 | 1789 | */ |
63e2ed36 | 1790 | enqueue_hrtimer(timer, new_base, HRTIMER_MODE_ABS); |
c0a31329 TG |
1791 | } |
1792 | } | |
1793 | ||
27590dc1 | 1794 | int hrtimers_dead_cpu(unsigned int scpu) |
c0a31329 | 1795 | { |
3c8aa39d | 1796 | struct hrtimer_cpu_base *old_base, *new_base; |
731a55ba | 1797 | int i; |
c0a31329 | 1798 | |
37810659 | 1799 | BUG_ON(cpu_online(scpu)); |
37810659 | 1800 | tick_cancel_sched_timer(scpu); |
731a55ba | 1801 | |
5da70160 AMG |
1802 | /* |
1803 | * this BH disable ensures that raise_softirq_irqoff() does | |
1804 | * not wakeup ksoftirqd (and acquire the pi-lock) while | |
1805 | * holding the cpu_base lock | |
1806 | */ | |
1807 | local_bh_disable(); | |
731a55ba TG |
1808 | local_irq_disable(); |
1809 | old_base = &per_cpu(hrtimer_bases, scpu); | |
dc5df73b | 1810 | new_base = this_cpu_ptr(&hrtimer_bases); |
d82f0b0f ON |
1811 | /* |
1812 | * The caller is globally serialized and nobody else | |
1813 | * takes two locks at once, deadlock is not possible. | |
1814 | */ | |
ecb49d1a TG |
1815 | raw_spin_lock(&new_base->lock); |
1816 | raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING); | |
c0a31329 | 1817 | |
3c8aa39d | 1818 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
ca109491 | 1819 | migrate_hrtimer_list(&old_base->clock_base[i], |
37810659 | 1820 | &new_base->clock_base[i]); |
c0a31329 TG |
1821 | } |
1822 | ||
5da70160 AMG |
1823 | /* |
1824 | * The migration might have changed the first expiring softirq | |
1825 | * timer on this CPU. Update it. | |
1826 | */ | |
1827 | hrtimer_update_softirq_timer(new_base, false); | |
1828 | ||
ecb49d1a TG |
1829 | raw_spin_unlock(&old_base->lock); |
1830 | raw_spin_unlock(&new_base->lock); | |
37810659 | 1831 | |
731a55ba TG |
1832 | /* Check, if we got expired work to do */ |
1833 | __hrtimer_peek_ahead_timers(); | |
1834 | local_irq_enable(); | |
5da70160 | 1835 | local_bh_enable(); |
27590dc1 | 1836 | return 0; |
c0a31329 | 1837 | } |
37810659 | 1838 | |
c0a31329 TG |
1839 | #endif /* CONFIG_HOTPLUG_CPU */ |
1840 | ||
c0a31329 TG |
1841 | void __init hrtimers_init(void) |
1842 | { | |
27590dc1 | 1843 | hrtimers_prepare_cpu(smp_processor_id()); |
5da70160 | 1844 | open_softirq(HRTIMER_SOFTIRQ, hrtimer_run_softirq); |
c0a31329 TG |
1845 | } |
1846 | ||
7bb67439 | 1847 | /** |
351b3f7a | 1848 | * schedule_hrtimeout_range_clock - sleep until timeout |
7bb67439 | 1849 | * @expires: timeout value (ktime_t) |
654c8e0b | 1850 | * @delta: slack in expires timeout (ktime_t) |
90777713 AMG |
1851 | * @mode: timer mode |
1852 | * @clock_id: timer clock to be used | |
7bb67439 | 1853 | */ |
351b3f7a | 1854 | int __sched |
da8b44d5 | 1855 | schedule_hrtimeout_range_clock(ktime_t *expires, u64 delta, |
90777713 | 1856 | const enum hrtimer_mode mode, clockid_t clock_id) |
7bb67439 AV |
1857 | { |
1858 | struct hrtimer_sleeper t; | |
1859 | ||
1860 | /* | |
1861 | * Optimize when a zero timeout value is given. It does not | |
1862 | * matter whether this is an absolute or a relative time. | |
1863 | */ | |
2456e855 | 1864 | if (expires && *expires == 0) { |
7bb67439 AV |
1865 | __set_current_state(TASK_RUNNING); |
1866 | return 0; | |
1867 | } | |
1868 | ||
1869 | /* | |
43b21013 | 1870 | * A NULL parameter means "infinite" |
7bb67439 AV |
1871 | */ |
1872 | if (!expires) { | |
1873 | schedule(); | |
7bb67439 AV |
1874 | return -EINTR; |
1875 | } | |
1876 | ||
90777713 | 1877 | hrtimer_init_on_stack(&t.timer, clock_id, mode); |
654c8e0b | 1878 | hrtimer_set_expires_range_ns(&t.timer, *expires, delta); |
7bb67439 AV |
1879 | |
1880 | hrtimer_init_sleeper(&t, current); | |
1881 | ||
cc584b21 | 1882 | hrtimer_start_expires(&t.timer, mode); |
7bb67439 AV |
1883 | |
1884 | if (likely(t.task)) | |
1885 | schedule(); | |
1886 | ||
1887 | hrtimer_cancel(&t.timer); | |
1888 | destroy_hrtimer_on_stack(&t.timer); | |
1889 | ||
1890 | __set_current_state(TASK_RUNNING); | |
1891 | ||
1892 | return !t.task ? 0 : -EINTR; | |
1893 | } | |
351b3f7a CE |
1894 | |
1895 | /** | |
1896 | * schedule_hrtimeout_range - sleep until timeout | |
1897 | * @expires: timeout value (ktime_t) | |
1898 | * @delta: slack in expires timeout (ktime_t) | |
90777713 | 1899 | * @mode: timer mode |
351b3f7a CE |
1900 | * |
1901 | * Make the current task sleep until the given expiry time has | |
1902 | * elapsed. The routine will return immediately unless | |
1903 | * the current task state has been set (see set_current_state()). | |
1904 | * | |
1905 | * The @delta argument gives the kernel the freedom to schedule the | |
1906 | * actual wakeup to a time that is both power and performance friendly. | |
1907 | * The kernel give the normal best effort behavior for "@expires+@delta", | |
1908 | * but may decide to fire the timer earlier, but no earlier than @expires. | |
1909 | * | |
1910 | * You can set the task state as follows - | |
1911 | * | |
1912 | * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to | |
4b7e9cf9 DA |
1913 | * pass before the routine returns unless the current task is explicitly |
1914 | * woken up, (e.g. by wake_up_process()). | |
351b3f7a CE |
1915 | * |
1916 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
4b7e9cf9 DA |
1917 | * delivered to the current task or the current task is explicitly woken |
1918 | * up. | |
351b3f7a CE |
1919 | * |
1920 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1921 | * routine returns. | |
1922 | * | |
4b7e9cf9 DA |
1923 | * Returns 0 when the timer has expired. If the task was woken before the |
1924 | * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or | |
1925 | * by an explicit wakeup, it returns -EINTR. | |
351b3f7a | 1926 | */ |
da8b44d5 | 1927 | int __sched schedule_hrtimeout_range(ktime_t *expires, u64 delta, |
351b3f7a CE |
1928 | const enum hrtimer_mode mode) |
1929 | { | |
1930 | return schedule_hrtimeout_range_clock(expires, delta, mode, | |
1931 | CLOCK_MONOTONIC); | |
1932 | } | |
654c8e0b AV |
1933 | EXPORT_SYMBOL_GPL(schedule_hrtimeout_range); |
1934 | ||
1935 | /** | |
1936 | * schedule_hrtimeout - sleep until timeout | |
1937 | * @expires: timeout value (ktime_t) | |
90777713 | 1938 | * @mode: timer mode |
654c8e0b AV |
1939 | * |
1940 | * Make the current task sleep until the given expiry time has | |
1941 | * elapsed. The routine will return immediately unless | |
1942 | * the current task state has been set (see set_current_state()). | |
1943 | * | |
1944 | * You can set the task state as follows - | |
1945 | * | |
1946 | * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to | |
4b7e9cf9 DA |
1947 | * pass before the routine returns unless the current task is explicitly |
1948 | * woken up, (e.g. by wake_up_process()). | |
654c8e0b AV |
1949 | * |
1950 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
4b7e9cf9 DA |
1951 | * delivered to the current task or the current task is explicitly woken |
1952 | * up. | |
654c8e0b AV |
1953 | * |
1954 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1955 | * routine returns. | |
1956 | * | |
4b7e9cf9 DA |
1957 | * Returns 0 when the timer has expired. If the task was woken before the |
1958 | * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or | |
1959 | * by an explicit wakeup, it returns -EINTR. | |
654c8e0b AV |
1960 | */ |
1961 | int __sched schedule_hrtimeout(ktime_t *expires, | |
1962 | const enum hrtimer_mode mode) | |
1963 | { | |
1964 | return schedule_hrtimeout_range(expires, 0, mode); | |
1965 | } | |
7bb67439 | 1966 | EXPORT_SYMBOL_GPL(schedule_hrtimeout); |