]>
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> |
c0a31329 | 54 | |
7c0f6ba6 | 55 | #include <linux/uaccess.h> |
c0a31329 | 56 | |
c6a2a177 XG |
57 | #include <trace/events/timer.h> |
58 | ||
c1797baf | 59 | #include "tick-internal.h" |
8b094cd0 | 60 | |
c0a31329 TG |
61 | /* |
62 | * The timer bases: | |
7978672c | 63 | * |
571af55a | 64 | * There are more clockids than hrtimer bases. Thus, we index |
e06383db JS |
65 | * into the timer bases by the hrtimer_base_type enum. When trying |
66 | * to reach a base using a clockid, hrtimer_clockid_to_base() | |
67 | * is used to convert from clockid to the proper hrtimer_base_type. | |
c0a31329 | 68 | */ |
54cdfdb4 | 69 | DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) = |
c0a31329 | 70 | { |
84cc8fd2 | 71 | .lock = __RAW_SPIN_LOCK_UNLOCKED(hrtimer_bases.lock), |
887d9dc9 | 72 | .seq = SEQCNT_ZERO(hrtimer_bases.seq), |
3c8aa39d | 73 | .clock_base = |
c0a31329 | 74 | { |
3c8aa39d | 75 | { |
ab8177bc TG |
76 | .index = HRTIMER_BASE_MONOTONIC, |
77 | .clockid = CLOCK_MONOTONIC, | |
3c8aa39d | 78 | .get_time = &ktime_get, |
3c8aa39d | 79 | }, |
68fa61c0 TG |
80 | { |
81 | .index = HRTIMER_BASE_REALTIME, | |
82 | .clockid = CLOCK_REALTIME, | |
83 | .get_time = &ktime_get_real, | |
68fa61c0 | 84 | }, |
70a08cca | 85 | { |
ab8177bc TG |
86 | .index = HRTIMER_BASE_BOOTTIME, |
87 | .clockid = CLOCK_BOOTTIME, | |
70a08cca | 88 | .get_time = &ktime_get_boottime, |
70a08cca | 89 | }, |
90adda98 JS |
90 | { |
91 | .index = HRTIMER_BASE_TAI, | |
92 | .clockid = CLOCK_TAI, | |
93 | .get_time = &ktime_get_clocktai, | |
90adda98 | 94 | }, |
3c8aa39d | 95 | } |
c0a31329 TG |
96 | }; |
97 | ||
942c3c5c | 98 | static const int hrtimer_clock_to_base_table[MAX_CLOCKS] = { |
336a9cde MZ |
99 | /* Make sure we catch unsupported clockids */ |
100 | [0 ... MAX_CLOCKS - 1] = HRTIMER_MAX_CLOCK_BASES, | |
101 | ||
ce31332d TG |
102 | [CLOCK_REALTIME] = HRTIMER_BASE_REALTIME, |
103 | [CLOCK_MONOTONIC] = HRTIMER_BASE_MONOTONIC, | |
104 | [CLOCK_BOOTTIME] = HRTIMER_BASE_BOOTTIME, | |
90adda98 | 105 | [CLOCK_TAI] = HRTIMER_BASE_TAI, |
ce31332d | 106 | }; |
e06383db | 107 | |
c0a31329 TG |
108 | /* |
109 | * Functions and macros which are different for UP/SMP systems are kept in a | |
110 | * single place | |
111 | */ | |
112 | #ifdef CONFIG_SMP | |
113 | ||
887d9dc9 PZ |
114 | /* |
115 | * We require the migration_base for lock_hrtimer_base()/switch_hrtimer_base() | |
116 | * such that hrtimer_callback_running() can unconditionally dereference | |
117 | * timer->base->cpu_base | |
118 | */ | |
119 | static struct hrtimer_cpu_base migration_cpu_base = { | |
120 | .seq = SEQCNT_ZERO(migration_cpu_base), | |
121 | .clock_base = { { .cpu_base = &migration_cpu_base, }, }, | |
122 | }; | |
123 | ||
124 | #define migration_base migration_cpu_base.clock_base[0] | |
125 | ||
c0a31329 TG |
126 | /* |
127 | * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock | |
128 | * means that all timers which are tied to this base via timer->base are | |
129 | * locked, and the base itself is locked too. | |
130 | * | |
131 | * So __run_timers/migrate_timers can safely modify all timers which could | |
132 | * be found on the lists/queues. | |
133 | * | |
134 | * When the timer's base is locked, and the timer removed from list, it is | |
887d9dc9 PZ |
135 | * possible to set timer->base = &migration_base and drop the lock: the timer |
136 | * remains locked. | |
c0a31329 | 137 | */ |
3c8aa39d TG |
138 | static |
139 | struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer, | |
140 | unsigned long *flags) | |
c0a31329 | 141 | { |
3c8aa39d | 142 | struct hrtimer_clock_base *base; |
c0a31329 TG |
143 | |
144 | for (;;) { | |
145 | base = timer->base; | |
887d9dc9 | 146 | if (likely(base != &migration_base)) { |
ecb49d1a | 147 | raw_spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
148 | if (likely(base == timer->base)) |
149 | return base; | |
150 | /* The timer has migrated to another CPU: */ | |
ecb49d1a | 151 | raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags); |
c0a31329 TG |
152 | } |
153 | cpu_relax(); | |
154 | } | |
155 | } | |
156 | ||
6ff7041d TG |
157 | /* |
158 | * With HIGHRES=y we do not migrate the timer when it is expiring | |
159 | * before the next event on the target cpu because we cannot reprogram | |
160 | * the target cpu hardware and we would cause it to fire late. | |
161 | * | |
162 | * Called with cpu_base->lock of target cpu held. | |
163 | */ | |
164 | static int | |
165 | hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base) | |
166 | { | |
167 | #ifdef CONFIG_HIGH_RES_TIMERS | |
168 | ktime_t expires; | |
169 | ||
170 | if (!new_base->cpu_base->hres_active) | |
171 | return 0; | |
172 | ||
173 | expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset); | |
2456e855 | 174 | return expires <= new_base->cpu_base->expires_next; |
6ff7041d TG |
175 | #else |
176 | return 0; | |
177 | #endif | |
178 | } | |
179 | ||
86721ab6 | 180 | #ifdef CONFIG_NO_HZ_COMMON |
bc7a34b8 TG |
181 | static inline |
182 | struct hrtimer_cpu_base *get_target_base(struct hrtimer_cpu_base *base, | |
183 | int pinned) | |
184 | { | |
185 | if (pinned || !base->migration_enabled) | |
662b3e19 | 186 | return base; |
bc7a34b8 TG |
187 | return &per_cpu(hrtimer_bases, get_nohz_timer_target()); |
188 | } | |
189 | #else | |
190 | static inline | |
191 | struct hrtimer_cpu_base *get_target_base(struct hrtimer_cpu_base *base, | |
192 | int pinned) | |
193 | { | |
662b3e19 | 194 | return base; |
bc7a34b8 TG |
195 | } |
196 | #endif | |
197 | ||
c0a31329 | 198 | /* |
b48362d8 FW |
199 | * We switch the timer base to a power-optimized selected CPU target, |
200 | * if: | |
201 | * - NO_HZ_COMMON is enabled | |
202 | * - timer migration is enabled | |
203 | * - the timer callback is not running | |
204 | * - the timer is not the first expiring timer on the new target | |
205 | * | |
206 | * If one of the above requirements is not fulfilled we move the timer | |
207 | * to the current CPU or leave it on the previously assigned CPU if | |
208 | * the timer callback is currently running. | |
c0a31329 | 209 | */ |
3c8aa39d | 210 | static inline struct hrtimer_clock_base * |
597d0275 AB |
211 | switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base, |
212 | int pinned) | |
c0a31329 | 213 | { |
b48362d8 | 214 | struct hrtimer_cpu_base *new_cpu_base, *this_cpu_base; |
3c8aa39d | 215 | struct hrtimer_clock_base *new_base; |
ab8177bc | 216 | int basenum = base->index; |
c0a31329 | 217 | |
b48362d8 FW |
218 | this_cpu_base = this_cpu_ptr(&hrtimer_bases); |
219 | new_cpu_base = get_target_base(this_cpu_base, pinned); | |
eea08f32 | 220 | again: |
e06383db | 221 | new_base = &new_cpu_base->clock_base[basenum]; |
c0a31329 TG |
222 | |
223 | if (base != new_base) { | |
224 | /* | |
6ff7041d | 225 | * We are trying to move timer to new_base. |
c0a31329 TG |
226 | * However we can't change timer's base while it is running, |
227 | * so we keep it on the same CPU. No hassle vs. reprogramming | |
228 | * the event source in the high resolution case. The softirq | |
229 | * code will take care of this when the timer function has | |
230 | * completed. There is no conflict as we hold the lock until | |
231 | * the timer is enqueued. | |
232 | */ | |
54cdfdb4 | 233 | if (unlikely(hrtimer_callback_running(timer))) |
c0a31329 TG |
234 | return base; |
235 | ||
887d9dc9 PZ |
236 | /* See the comment in lock_hrtimer_base() */ |
237 | timer->base = &migration_base; | |
ecb49d1a TG |
238 | raw_spin_unlock(&base->cpu_base->lock); |
239 | raw_spin_lock(&new_base->cpu_base->lock); | |
eea08f32 | 240 | |
b48362d8 | 241 | if (new_cpu_base != this_cpu_base && |
bc7a34b8 | 242 | hrtimer_check_target(timer, new_base)) { |
ecb49d1a TG |
243 | raw_spin_unlock(&new_base->cpu_base->lock); |
244 | raw_spin_lock(&base->cpu_base->lock); | |
b48362d8 | 245 | new_cpu_base = this_cpu_base; |
6ff7041d TG |
246 | timer->base = base; |
247 | goto again; | |
eea08f32 | 248 | } |
c0a31329 | 249 | timer->base = new_base; |
012a45e3 | 250 | } else { |
b48362d8 | 251 | if (new_cpu_base != this_cpu_base && |
bc7a34b8 | 252 | hrtimer_check_target(timer, new_base)) { |
b48362d8 | 253 | new_cpu_base = this_cpu_base; |
012a45e3 LM |
254 | goto again; |
255 | } | |
c0a31329 TG |
256 | } |
257 | return new_base; | |
258 | } | |
259 | ||
260 | #else /* CONFIG_SMP */ | |
261 | ||
3c8aa39d | 262 | static inline struct hrtimer_clock_base * |
c0a31329 TG |
263 | lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) |
264 | { | |
3c8aa39d | 265 | struct hrtimer_clock_base *base = timer->base; |
c0a31329 | 266 | |
ecb49d1a | 267 | raw_spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
268 | |
269 | return base; | |
270 | } | |
271 | ||
eea08f32 | 272 | # define switch_hrtimer_base(t, b, p) (b) |
c0a31329 TG |
273 | |
274 | #endif /* !CONFIG_SMP */ | |
275 | ||
276 | /* | |
277 | * Functions for the union type storage format of ktime_t which are | |
278 | * too large for inlining: | |
279 | */ | |
280 | #if BITS_PER_LONG < 64 | |
c0a31329 TG |
281 | /* |
282 | * Divide a ktime value by a nanosecond value | |
283 | */ | |
f7bcb70e | 284 | s64 __ktime_divns(const ktime_t kt, s64 div) |
c0a31329 | 285 | { |
c0a31329 | 286 | int sft = 0; |
f7bcb70e JS |
287 | s64 dclc; |
288 | u64 tmp; | |
c0a31329 | 289 | |
900cfa46 | 290 | dclc = ktime_to_ns(kt); |
f7bcb70e JS |
291 | tmp = dclc < 0 ? -dclc : dclc; |
292 | ||
c0a31329 TG |
293 | /* Make sure the divisor is less than 2^32: */ |
294 | while (div >> 32) { | |
295 | sft++; | |
296 | div >>= 1; | |
297 | } | |
f7bcb70e JS |
298 | tmp >>= sft; |
299 | do_div(tmp, (unsigned long) div); | |
300 | return dclc < 0 ? -tmp : tmp; | |
c0a31329 | 301 | } |
8b618628 | 302 | EXPORT_SYMBOL_GPL(__ktime_divns); |
c0a31329 TG |
303 | #endif /* BITS_PER_LONG >= 64 */ |
304 | ||
5a7780e7 TG |
305 | /* |
306 | * Add two ktime values and do a safety check for overflow: | |
307 | */ | |
308 | ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs) | |
309 | { | |
979515c5 | 310 | ktime_t res = ktime_add_unsafe(lhs, rhs); |
5a7780e7 TG |
311 | |
312 | /* | |
313 | * We use KTIME_SEC_MAX here, the maximum timeout which we can | |
314 | * return to user space in a timespec: | |
315 | */ | |
2456e855 | 316 | if (res < 0 || res < lhs || res < rhs) |
5a7780e7 TG |
317 | res = ktime_set(KTIME_SEC_MAX, 0); |
318 | ||
319 | return res; | |
320 | } | |
321 | ||
8daa21e6 AB |
322 | EXPORT_SYMBOL_GPL(ktime_add_safe); |
323 | ||
237fc6e7 TG |
324 | #ifdef CONFIG_DEBUG_OBJECTS_TIMERS |
325 | ||
326 | static struct debug_obj_descr hrtimer_debug_descr; | |
327 | ||
99777288 SG |
328 | static void *hrtimer_debug_hint(void *addr) |
329 | { | |
330 | return ((struct hrtimer *) addr)->function; | |
331 | } | |
332 | ||
237fc6e7 TG |
333 | /* |
334 | * fixup_init is called when: | |
335 | * - an active object is initialized | |
336 | */ | |
e3252464 | 337 | static bool hrtimer_fixup_init(void *addr, enum debug_obj_state state) |
237fc6e7 TG |
338 | { |
339 | struct hrtimer *timer = addr; | |
340 | ||
341 | switch (state) { | |
342 | case ODEBUG_STATE_ACTIVE: | |
343 | hrtimer_cancel(timer); | |
344 | debug_object_init(timer, &hrtimer_debug_descr); | |
e3252464 | 345 | return true; |
237fc6e7 | 346 | default: |
e3252464 | 347 | return false; |
237fc6e7 TG |
348 | } |
349 | } | |
350 | ||
351 | /* | |
352 | * fixup_activate is called when: | |
353 | * - an active object is activated | |
b9fdac7f | 354 | * - an unknown non-static object is activated |
237fc6e7 | 355 | */ |
e3252464 | 356 | static bool hrtimer_fixup_activate(void *addr, enum debug_obj_state state) |
237fc6e7 TG |
357 | { |
358 | switch (state) { | |
237fc6e7 TG |
359 | case ODEBUG_STATE_ACTIVE: |
360 | WARN_ON(1); | |
361 | ||
362 | default: | |
e3252464 | 363 | return false; |
237fc6e7 TG |
364 | } |
365 | } | |
366 | ||
367 | /* | |
368 | * fixup_free is called when: | |
369 | * - an active object is freed | |
370 | */ | |
e3252464 | 371 | static bool hrtimer_fixup_free(void *addr, enum debug_obj_state state) |
237fc6e7 TG |
372 | { |
373 | struct hrtimer *timer = addr; | |
374 | ||
375 | switch (state) { | |
376 | case ODEBUG_STATE_ACTIVE: | |
377 | hrtimer_cancel(timer); | |
378 | debug_object_free(timer, &hrtimer_debug_descr); | |
e3252464 | 379 | return true; |
237fc6e7 | 380 | default: |
e3252464 | 381 | return false; |
237fc6e7 TG |
382 | } |
383 | } | |
384 | ||
385 | static struct debug_obj_descr hrtimer_debug_descr = { | |
386 | .name = "hrtimer", | |
99777288 | 387 | .debug_hint = hrtimer_debug_hint, |
237fc6e7 TG |
388 | .fixup_init = hrtimer_fixup_init, |
389 | .fixup_activate = hrtimer_fixup_activate, | |
390 | .fixup_free = hrtimer_fixup_free, | |
391 | }; | |
392 | ||
393 | static inline void debug_hrtimer_init(struct hrtimer *timer) | |
394 | { | |
395 | debug_object_init(timer, &hrtimer_debug_descr); | |
396 | } | |
397 | ||
398 | static inline void debug_hrtimer_activate(struct hrtimer *timer) | |
399 | { | |
400 | debug_object_activate(timer, &hrtimer_debug_descr); | |
401 | } | |
402 | ||
403 | static inline void debug_hrtimer_deactivate(struct hrtimer *timer) | |
404 | { | |
405 | debug_object_deactivate(timer, &hrtimer_debug_descr); | |
406 | } | |
407 | ||
408 | static inline void debug_hrtimer_free(struct hrtimer *timer) | |
409 | { | |
410 | debug_object_free(timer, &hrtimer_debug_descr); | |
411 | } | |
412 | ||
413 | static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id, | |
414 | enum hrtimer_mode mode); | |
415 | ||
416 | void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id, | |
417 | enum hrtimer_mode mode) | |
418 | { | |
419 | debug_object_init_on_stack(timer, &hrtimer_debug_descr); | |
420 | __hrtimer_init(timer, clock_id, mode); | |
421 | } | |
2bc481cf | 422 | EXPORT_SYMBOL_GPL(hrtimer_init_on_stack); |
237fc6e7 TG |
423 | |
424 | void destroy_hrtimer_on_stack(struct hrtimer *timer) | |
425 | { | |
426 | debug_object_free(timer, &hrtimer_debug_descr); | |
427 | } | |
c08376ac | 428 | EXPORT_SYMBOL_GPL(destroy_hrtimer_on_stack); |
237fc6e7 TG |
429 | |
430 | #else | |
431 | static inline void debug_hrtimer_init(struct hrtimer *timer) { } | |
432 | static inline void debug_hrtimer_activate(struct hrtimer *timer) { } | |
433 | static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { } | |
434 | #endif | |
435 | ||
c6a2a177 XG |
436 | static inline void |
437 | debug_init(struct hrtimer *timer, clockid_t clockid, | |
438 | enum hrtimer_mode mode) | |
439 | { | |
440 | debug_hrtimer_init(timer); | |
441 | trace_hrtimer_init(timer, clockid, mode); | |
442 | } | |
443 | ||
444 | static inline void debug_activate(struct hrtimer *timer) | |
445 | { | |
446 | debug_hrtimer_activate(timer); | |
447 | trace_hrtimer_start(timer); | |
448 | } | |
449 | ||
450 | static inline void debug_deactivate(struct hrtimer *timer) | |
451 | { | |
452 | debug_hrtimer_deactivate(timer); | |
453 | trace_hrtimer_cancel(timer); | |
454 | } | |
455 | ||
9bc74919 | 456 | #if defined(CONFIG_NO_HZ_COMMON) || defined(CONFIG_HIGH_RES_TIMERS) |
895bdfa7 TG |
457 | static inline void hrtimer_update_next_timer(struct hrtimer_cpu_base *cpu_base, |
458 | struct hrtimer *timer) | |
459 | { | |
460 | #ifdef CONFIG_HIGH_RES_TIMERS | |
461 | cpu_base->next_timer = timer; | |
462 | #endif | |
463 | } | |
464 | ||
4ebbda52 | 465 | static ktime_t __hrtimer_get_next_event(struct hrtimer_cpu_base *cpu_base) |
9bc74919 TG |
466 | { |
467 | struct hrtimer_clock_base *base = cpu_base->clock_base; | |
34aee88a | 468 | unsigned int active = cpu_base->active_bases; |
2456e855 | 469 | ktime_t expires, expires_next = KTIME_MAX; |
9bc74919 | 470 | |
895bdfa7 | 471 | hrtimer_update_next_timer(cpu_base, NULL); |
34aee88a | 472 | for (; active; base++, active >>= 1) { |
9bc74919 TG |
473 | struct timerqueue_node *next; |
474 | struct hrtimer *timer; | |
475 | ||
34aee88a | 476 | if (!(active & 0x01)) |
9bc74919 TG |
477 | continue; |
478 | ||
34aee88a | 479 | next = timerqueue_getnext(&base->active); |
9bc74919 TG |
480 | timer = container_of(next, struct hrtimer, node); |
481 | expires = ktime_sub(hrtimer_get_expires(timer), base->offset); | |
2456e855 | 482 | if (expires < expires_next) { |
9bc74919 | 483 | expires_next = expires; |
895bdfa7 TG |
484 | hrtimer_update_next_timer(cpu_base, timer); |
485 | } | |
9bc74919 TG |
486 | } |
487 | /* | |
488 | * clock_was_set() might have changed base->offset of any of | |
489 | * the clock bases so the result might be negative. Fix it up | |
490 | * to prevent a false positive in clockevents_program_event(). | |
491 | */ | |
2456e855 TG |
492 | if (expires_next < 0) |
493 | expires_next = 0; | |
9bc74919 TG |
494 | return expires_next; |
495 | } | |
496 | #endif | |
497 | ||
21d6d52a TG |
498 | static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base) |
499 | { | |
500 | ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset; | |
501 | ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset; | |
502 | ktime_t *offs_tai = &base->clock_base[HRTIMER_BASE_TAI].offset; | |
503 | ||
868a3e91 TG |
504 | return ktime_get_update_offsets_now(&base->clock_was_set_seq, |
505 | offs_real, offs_boot, offs_tai); | |
21d6d52a TG |
506 | } |
507 | ||
54cdfdb4 TG |
508 | /* High resolution timer related functions */ |
509 | #ifdef CONFIG_HIGH_RES_TIMERS | |
510 | ||
511 | /* | |
512 | * High resolution timer enabled ? | |
513 | */ | |
4cc7ecb7 | 514 | static bool hrtimer_hres_enabled __read_mostly = true; |
398ca17f TG |
515 | unsigned int hrtimer_resolution __read_mostly = LOW_RES_NSEC; |
516 | EXPORT_SYMBOL_GPL(hrtimer_resolution); | |
54cdfdb4 TG |
517 | |
518 | /* | |
519 | * Enable / Disable high resolution mode | |
520 | */ | |
521 | static int __init setup_hrtimer_hres(char *str) | |
522 | { | |
4cc7ecb7 | 523 | return (kstrtobool(str, &hrtimer_hres_enabled) == 0); |
54cdfdb4 TG |
524 | } |
525 | ||
526 | __setup("highres=", setup_hrtimer_hres); | |
527 | ||
528 | /* | |
529 | * hrtimer_high_res_enabled - query, if the highres mode is enabled | |
530 | */ | |
531 | static inline int hrtimer_is_hres_enabled(void) | |
532 | { | |
533 | return hrtimer_hres_enabled; | |
534 | } | |
535 | ||
536 | /* | |
537 | * Is the high resolution mode active ? | |
538 | */ | |
e19ffe8b TG |
539 | static inline int __hrtimer_hres_active(struct hrtimer_cpu_base *cpu_base) |
540 | { | |
541 | return cpu_base->hres_active; | |
542 | } | |
543 | ||
54cdfdb4 TG |
544 | static inline int hrtimer_hres_active(void) |
545 | { | |
e19ffe8b | 546 | return __hrtimer_hres_active(this_cpu_ptr(&hrtimer_bases)); |
54cdfdb4 TG |
547 | } |
548 | ||
549 | /* | |
550 | * Reprogram the event source with checking both queues for the | |
551 | * next event | |
552 | * Called with interrupts disabled and base->lock held | |
553 | */ | |
7403f41f AC |
554 | static void |
555 | hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal) | |
54cdfdb4 | 556 | { |
21d6d52a TG |
557 | ktime_t expires_next; |
558 | ||
559 | if (!cpu_base->hres_active) | |
560 | return; | |
561 | ||
562 | expires_next = __hrtimer_get_next_event(cpu_base); | |
54cdfdb4 | 563 | |
2456e855 | 564 | if (skip_equal && expires_next == cpu_base->expires_next) |
7403f41f AC |
565 | return; |
566 | ||
2456e855 | 567 | cpu_base->expires_next = expires_next; |
7403f41f | 568 | |
6c6c0d5a SH |
569 | /* |
570 | * If a hang was detected in the last timer interrupt then we | |
571 | * leave the hang delay active in the hardware. We want the | |
572 | * system to make progress. That also prevents the following | |
573 | * scenario: | |
574 | * T1 expires 50ms from now | |
575 | * T2 expires 5s from now | |
576 | * | |
577 | * T1 is removed, so this code is called and would reprogram | |
578 | * the hardware to 5s from now. Any hrtimer_start after that | |
579 | * will not reprogram the hardware due to hang_detected being | |
580 | * set. So we'd effectivly block all timers until the T2 event | |
581 | * fires. | |
582 | */ | |
583 | if (cpu_base->hang_detected) | |
584 | return; | |
585 | ||
d2540875 | 586 | tick_program_event(cpu_base->expires_next, 1); |
54cdfdb4 TG |
587 | } |
588 | ||
589 | /* | |
54cdfdb4 TG |
590 | * When a timer is enqueued and expires earlier than the already enqueued |
591 | * timers, we have to check, whether it expires earlier than the timer for | |
592 | * which the clock event device was armed. | |
593 | * | |
594 | * Called with interrupts disabled and base->cpu_base.lock held | |
595 | */ | |
c6eb3f70 TG |
596 | static void hrtimer_reprogram(struct hrtimer *timer, |
597 | struct hrtimer_clock_base *base) | |
54cdfdb4 | 598 | { |
dc5df73b | 599 | struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases); |
cc584b21 | 600 | ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset); |
54cdfdb4 | 601 | |
cc584b21 | 602 | WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0); |
63070a79 | 603 | |
54cdfdb4 | 604 | /* |
c6eb3f70 TG |
605 | * If the timer is not on the current cpu, we cannot reprogram |
606 | * the other cpus clock event device. | |
54cdfdb4 | 607 | */ |
c6eb3f70 TG |
608 | if (base->cpu_base != cpu_base) |
609 | return; | |
610 | ||
611 | /* | |
612 | * If the hrtimer interrupt is running, then it will | |
613 | * reevaluate the clock bases and reprogram the clock event | |
614 | * device. The callbacks are always executed in hard interrupt | |
615 | * context so we don't need an extra check for a running | |
616 | * callback. | |
617 | */ | |
618 | if (cpu_base->in_hrtirq) | |
619 | return; | |
54cdfdb4 | 620 | |
63070a79 TG |
621 | /* |
622 | * CLOCK_REALTIME timer might be requested with an absolute | |
c6eb3f70 | 623 | * expiry time which is less than base->offset. Set it to 0. |
63070a79 | 624 | */ |
2456e855 TG |
625 | if (expires < 0) |
626 | expires = 0; | |
63070a79 | 627 | |
2456e855 | 628 | if (expires >= cpu_base->expires_next) |
c6eb3f70 | 629 | return; |
41d2e494 | 630 | |
c6eb3f70 | 631 | /* Update the pointer to the next expiring timer */ |
895bdfa7 | 632 | cpu_base->next_timer = timer; |
9bc74919 | 633 | |
41d2e494 TG |
634 | /* |
635 | * If a hang was detected in the last timer interrupt then we | |
636 | * do not schedule a timer which is earlier than the expiry | |
637 | * which we enforced in the hang detection. We want the system | |
638 | * to make progress. | |
639 | */ | |
640 | if (cpu_base->hang_detected) | |
c6eb3f70 | 641 | return; |
54cdfdb4 TG |
642 | |
643 | /* | |
c6eb3f70 TG |
644 | * Program the timer hardware. We enforce the expiry for |
645 | * events which are already in the past. | |
54cdfdb4 | 646 | */ |
c6eb3f70 TG |
647 | cpu_base->expires_next = expires; |
648 | tick_program_event(expires, 1); | |
54cdfdb4 TG |
649 | } |
650 | ||
54cdfdb4 TG |
651 | /* |
652 | * Initialize the high resolution related parts of cpu_base | |
653 | */ | |
654 | static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) | |
655 | { | |
2456e855 | 656 | base->expires_next = KTIME_MAX; |
54cdfdb4 | 657 | base->hres_active = 0; |
54cdfdb4 TG |
658 | } |
659 | ||
9ec26907 TG |
660 | /* |
661 | * Retrigger next event is called after clock was set | |
662 | * | |
663 | * Called with interrupts disabled via on_each_cpu() | |
664 | */ | |
665 | static void retrigger_next_event(void *arg) | |
666 | { | |
dc5df73b | 667 | struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases); |
9ec26907 | 668 | |
e19ffe8b | 669 | if (!base->hres_active) |
9ec26907 TG |
670 | return; |
671 | ||
9ec26907 | 672 | raw_spin_lock(&base->lock); |
5baefd6d | 673 | hrtimer_update_base(base); |
9ec26907 TG |
674 | hrtimer_force_reprogram(base, 0); |
675 | raw_spin_unlock(&base->lock); | |
676 | } | |
b12a03ce | 677 | |
54cdfdb4 TG |
678 | /* |
679 | * Switch to high resolution mode | |
680 | */ | |
75e3b37d | 681 | static void hrtimer_switch_to_hres(void) |
54cdfdb4 | 682 | { |
c6eb3f70 | 683 | struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases); |
54cdfdb4 TG |
684 | |
685 | if (tick_init_highres()) { | |
820de5c3 | 686 | printk(KERN_WARNING "Could not switch to high resolution " |
c6eb3f70 | 687 | "mode on CPU %d\n", base->cpu); |
85e1cd6e | 688 | return; |
54cdfdb4 TG |
689 | } |
690 | base->hres_active = 1; | |
398ca17f | 691 | hrtimer_resolution = HIGH_RES_NSEC; |
54cdfdb4 TG |
692 | |
693 | tick_setup_sched_timer(); | |
54cdfdb4 TG |
694 | /* "Retrigger" the interrupt to get things going */ |
695 | retrigger_next_event(NULL); | |
54cdfdb4 TG |
696 | } |
697 | ||
5ec2481b TG |
698 | static void clock_was_set_work(struct work_struct *work) |
699 | { | |
700 | clock_was_set(); | |
701 | } | |
702 | ||
703 | static DECLARE_WORK(hrtimer_work, clock_was_set_work); | |
704 | ||
f55a6faa | 705 | /* |
b4d90e9f | 706 | * Called from timekeeping and resume code to reprogram the hrtimer |
5ec2481b | 707 | * interrupt device on all cpus. |
f55a6faa JS |
708 | */ |
709 | void clock_was_set_delayed(void) | |
710 | { | |
5ec2481b | 711 | schedule_work(&hrtimer_work); |
f55a6faa JS |
712 | } |
713 | ||
54cdfdb4 TG |
714 | #else |
715 | ||
e19ffe8b | 716 | static inline int __hrtimer_hres_active(struct hrtimer_cpu_base *b) { return 0; } |
54cdfdb4 TG |
717 | static inline int hrtimer_hres_active(void) { return 0; } |
718 | static inline int hrtimer_is_hres_enabled(void) { return 0; } | |
75e3b37d | 719 | static inline void hrtimer_switch_to_hres(void) { } |
7403f41f AC |
720 | static inline void |
721 | hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { } | |
9e1e01dd VK |
722 | static inline int hrtimer_reprogram(struct hrtimer *timer, |
723 | struct hrtimer_clock_base *base) | |
54cdfdb4 TG |
724 | { |
725 | return 0; | |
726 | } | |
54cdfdb4 | 727 | static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { } |
9ec26907 | 728 | static inline void retrigger_next_event(void *arg) { } |
54cdfdb4 TG |
729 | |
730 | #endif /* CONFIG_HIGH_RES_TIMERS */ | |
731 | ||
b12a03ce TG |
732 | /* |
733 | * Clock realtime was set | |
734 | * | |
735 | * Change the offset of the realtime clock vs. the monotonic | |
736 | * clock. | |
737 | * | |
738 | * We might have to reprogram the high resolution timer interrupt. On | |
739 | * SMP we call the architecture specific code to retrigger _all_ high | |
740 | * resolution timer interrupts. On UP we just disable interrupts and | |
741 | * call the high resolution interrupt code. | |
742 | */ | |
743 | void clock_was_set(void) | |
744 | { | |
90ff1f30 | 745 | #ifdef CONFIG_HIGH_RES_TIMERS |
b12a03ce TG |
746 | /* Retrigger the CPU local events everywhere */ |
747 | on_each_cpu(retrigger_next_event, NULL, 1); | |
9ec26907 TG |
748 | #endif |
749 | timerfd_clock_was_set(); | |
b12a03ce TG |
750 | } |
751 | ||
752 | /* | |
753 | * During resume we might have to reprogram the high resolution timer | |
7c4c3a0f DV |
754 | * interrupt on all online CPUs. However, all other CPUs will be |
755 | * stopped with IRQs interrupts disabled so the clock_was_set() call | |
5ec2481b | 756 | * must be deferred. |
b12a03ce TG |
757 | */ |
758 | void hrtimers_resume(void) | |
759 | { | |
760 | WARN_ONCE(!irqs_disabled(), | |
761 | KERN_INFO "hrtimers_resume() called with IRQs enabled!"); | |
762 | ||
5ec2481b | 763 | /* Retrigger on the local CPU */ |
b12a03ce | 764 | retrigger_next_event(NULL); |
5ec2481b TG |
765 | /* And schedule a retrigger for all others */ |
766 | clock_was_set_delayed(); | |
b12a03ce TG |
767 | } |
768 | ||
c0a31329 | 769 | /* |
6506f2aa | 770 | * Counterpart to lock_hrtimer_base above: |
c0a31329 TG |
771 | */ |
772 | static inline | |
773 | void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) | |
774 | { | |
ecb49d1a | 775 | raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags); |
c0a31329 TG |
776 | } |
777 | ||
778 | /** | |
779 | * hrtimer_forward - forward the timer expiry | |
c0a31329 | 780 | * @timer: hrtimer to forward |
44f21475 | 781 | * @now: forward past this time |
c0a31329 TG |
782 | * @interval: the interval to forward |
783 | * | |
784 | * Forward the timer expiry so it will expire in the future. | |
8dca6f33 | 785 | * Returns the number of overruns. |
91e5a217 TG |
786 | * |
787 | * Can be safely called from the callback function of @timer. If | |
788 | * called from other contexts @timer must neither be enqueued nor | |
789 | * running the callback and the caller needs to take care of | |
790 | * serialization. | |
791 | * | |
792 | * Note: This only updates the timer expiry value and does not requeue | |
793 | * the timer. | |
c0a31329 | 794 | */ |
4d672e7a | 795 | u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval) |
c0a31329 | 796 | { |
4d672e7a | 797 | u64 orun = 1; |
44f21475 | 798 | ktime_t delta; |
c0a31329 | 799 | |
cc584b21 | 800 | delta = ktime_sub(now, hrtimer_get_expires(timer)); |
c0a31329 | 801 | |
2456e855 | 802 | if (delta < 0) |
c0a31329 TG |
803 | return 0; |
804 | ||
5de2755c PZ |
805 | if (WARN_ON(timer->state & HRTIMER_STATE_ENQUEUED)) |
806 | return 0; | |
807 | ||
2456e855 TG |
808 | if (interval < hrtimer_resolution) |
809 | interval = hrtimer_resolution; | |
c9db4fa1 | 810 | |
2456e855 | 811 | if (unlikely(delta >= interval)) { |
df869b63 | 812 | s64 incr = ktime_to_ns(interval); |
c0a31329 TG |
813 | |
814 | orun = ktime_divns(delta, incr); | |
cc584b21 | 815 | hrtimer_add_expires_ns(timer, incr * orun); |
2456e855 | 816 | if (hrtimer_get_expires_tv64(timer) > now) |
c0a31329 TG |
817 | return orun; |
818 | /* | |
819 | * This (and the ktime_add() below) is the | |
820 | * correction for exact: | |
821 | */ | |
822 | orun++; | |
823 | } | |
cc584b21 | 824 | hrtimer_add_expires(timer, interval); |
c0a31329 TG |
825 | |
826 | return orun; | |
827 | } | |
6bdb6b62 | 828 | EXPORT_SYMBOL_GPL(hrtimer_forward); |
c0a31329 TG |
829 | |
830 | /* | |
831 | * enqueue_hrtimer - internal function to (re)start a timer | |
832 | * | |
833 | * The timer is inserted in expiry order. Insertion into the | |
834 | * red black tree is O(log(n)). Must hold the base lock. | |
a6037b61 PZ |
835 | * |
836 | * Returns 1 when the new timer is the leftmost timer in the tree. | |
c0a31329 | 837 | */ |
a6037b61 PZ |
838 | static int enqueue_hrtimer(struct hrtimer *timer, |
839 | struct hrtimer_clock_base *base) | |
c0a31329 | 840 | { |
c6a2a177 | 841 | debug_activate(timer); |
237fc6e7 | 842 | |
ab8177bc | 843 | base->cpu_base->active_bases |= 1 << base->index; |
54cdfdb4 | 844 | |
887d9dc9 | 845 | timer->state = HRTIMER_STATE_ENQUEUED; |
a6037b61 | 846 | |
b97f44c9 | 847 | return timerqueue_add(&base->active, &timer->node); |
288867ec | 848 | } |
c0a31329 TG |
849 | |
850 | /* | |
851 | * __remove_hrtimer - internal function to remove a timer | |
852 | * | |
853 | * Caller must hold the base lock. | |
54cdfdb4 TG |
854 | * |
855 | * High resolution timer mode reprograms the clock event device when the | |
856 | * timer is the one which expires next. The caller can disable this by setting | |
857 | * reprogram to zero. This is useful, when the context does a reprogramming | |
858 | * anyway (e.g. timer interrupt) | |
c0a31329 | 859 | */ |
3c8aa39d | 860 | static void __remove_hrtimer(struct hrtimer *timer, |
303e967f | 861 | struct hrtimer_clock_base *base, |
203cbf77 | 862 | u8 newstate, int reprogram) |
c0a31329 | 863 | { |
e19ffe8b | 864 | struct hrtimer_cpu_base *cpu_base = base->cpu_base; |
203cbf77 | 865 | u8 state = timer->state; |
e19ffe8b | 866 | |
895bdfa7 TG |
867 | timer->state = newstate; |
868 | if (!(state & HRTIMER_STATE_ENQUEUED)) | |
869 | return; | |
7403f41f | 870 | |
b97f44c9 | 871 | if (!timerqueue_del(&base->active, &timer->node)) |
e19ffe8b | 872 | cpu_base->active_bases &= ~(1 << base->index); |
7403f41f | 873 | |
7403f41f | 874 | #ifdef CONFIG_HIGH_RES_TIMERS |
895bdfa7 TG |
875 | /* |
876 | * Note: If reprogram is false we do not update | |
877 | * cpu_base->next_timer. This happens when we remove the first | |
878 | * timer on a remote cpu. No harm as we never dereference | |
879 | * cpu_base->next_timer. So the worst thing what can happen is | |
880 | * an superflous call to hrtimer_force_reprogram() on the | |
881 | * remote cpu later on if the same timer gets enqueued again. | |
882 | */ | |
883 | if (reprogram && timer == cpu_base->next_timer) | |
884 | hrtimer_force_reprogram(cpu_base, 1); | |
7403f41f | 885 | #endif |
c0a31329 TG |
886 | } |
887 | ||
888 | /* | |
889 | * remove hrtimer, called with base lock held | |
890 | */ | |
891 | static inline int | |
8edfb036 | 892 | remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base, bool restart) |
c0a31329 | 893 | { |
303e967f | 894 | if (hrtimer_is_queued(timer)) { |
203cbf77 | 895 | u8 state = timer->state; |
54cdfdb4 TG |
896 | int reprogram; |
897 | ||
898 | /* | |
899 | * Remove the timer and force reprogramming when high | |
900 | * resolution mode is active and the timer is on the current | |
901 | * CPU. If we remove a timer on another CPU, reprogramming is | |
902 | * skipped. The interrupt event on this CPU is fired and | |
903 | * reprogramming happens in the interrupt handler. This is a | |
904 | * rare case and less expensive than a smp call. | |
905 | */ | |
c6a2a177 | 906 | debug_deactivate(timer); |
dc5df73b | 907 | reprogram = base->cpu_base == this_cpu_ptr(&hrtimer_bases); |
8edfb036 | 908 | |
887d9dc9 PZ |
909 | if (!restart) |
910 | state = HRTIMER_STATE_INACTIVE; | |
911 | ||
f13d4f97 | 912 | __remove_hrtimer(timer, base, state, reprogram); |
c0a31329 TG |
913 | return 1; |
914 | } | |
915 | return 0; | |
916 | } | |
917 | ||
203cbf77 TG |
918 | static inline ktime_t hrtimer_update_lowres(struct hrtimer *timer, ktime_t tim, |
919 | const enum hrtimer_mode mode) | |
920 | { | |
921 | #ifdef CONFIG_TIME_LOW_RES | |
922 | /* | |
923 | * CONFIG_TIME_LOW_RES indicates that the system has no way to return | |
924 | * granular time values. For relative timers we add hrtimer_resolution | |
925 | * (i.e. one jiffie) to prevent short timeouts. | |
926 | */ | |
927 | timer->is_rel = mode & HRTIMER_MODE_REL; | |
928 | if (timer->is_rel) | |
8b0e1953 | 929 | tim = ktime_add_safe(tim, hrtimer_resolution); |
203cbf77 TG |
930 | #endif |
931 | return tim; | |
932 | } | |
933 | ||
58f1f803 TG |
934 | /** |
935 | * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU | |
936 | * @timer: the timer to be added | |
937 | * @tim: expiry time | |
938 | * @delta_ns: "slack" range for the timer | |
939 | * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or | |
940 | * relative (HRTIMER_MODE_REL) | |
58f1f803 | 941 | */ |
61699e13 | 942 | void hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, |
da8b44d5 | 943 | u64 delta_ns, const enum hrtimer_mode mode) |
c0a31329 | 944 | { |
3c8aa39d | 945 | struct hrtimer_clock_base *base, *new_base; |
c0a31329 | 946 | unsigned long flags; |
61699e13 | 947 | int leftmost; |
c0a31329 TG |
948 | |
949 | base = lock_hrtimer_base(timer, &flags); | |
950 | ||
951 | /* Remove an active timer from the queue: */ | |
8edfb036 | 952 | remove_hrtimer(timer, base, true); |
c0a31329 | 953 | |
203cbf77 | 954 | if (mode & HRTIMER_MODE_REL) |
84ea7fe3 | 955 | tim = ktime_add_safe(tim, base->get_time()); |
203cbf77 TG |
956 | |
957 | tim = hrtimer_update_lowres(timer, tim, mode); | |
237fc6e7 | 958 | |
da8f2e17 | 959 | hrtimer_set_expires_range_ns(timer, tim, delta_ns); |
c0a31329 | 960 | |
84ea7fe3 VK |
961 | /* Switch the timer base, if necessary: */ |
962 | new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED); | |
963 | ||
a6037b61 | 964 | leftmost = enqueue_hrtimer(timer, new_base); |
61699e13 TG |
965 | if (!leftmost) |
966 | goto unlock; | |
49a2a075 VK |
967 | |
968 | if (!hrtimer_is_hres_active(timer)) { | |
969 | /* | |
970 | * Kick to reschedule the next tick to handle the new timer | |
971 | * on dynticks target. | |
972 | */ | |
683be13a TG |
973 | if (new_base->cpu_base->nohz_active) |
974 | wake_up_nohz_cpu(new_base->cpu_base->cpu); | |
c6eb3f70 TG |
975 | } else { |
976 | hrtimer_reprogram(timer, new_base); | |
b22affe0 | 977 | } |
61699e13 | 978 | unlock: |
c0a31329 | 979 | unlock_hrtimer_base(timer, &flags); |
7f1e2ca9 | 980 | } |
da8f2e17 AV |
981 | EXPORT_SYMBOL_GPL(hrtimer_start_range_ns); |
982 | ||
c0a31329 TG |
983 | /** |
984 | * hrtimer_try_to_cancel - try to deactivate a timer | |
c0a31329 TG |
985 | * @timer: hrtimer to stop |
986 | * | |
987 | * Returns: | |
988 | * 0 when the timer was not active | |
989 | * 1 when the timer was active | |
0ba42a59 | 990 | * -1 when the timer is currently executing the callback function and |
fa9799e3 | 991 | * cannot be stopped |
c0a31329 TG |
992 | */ |
993 | int hrtimer_try_to_cancel(struct hrtimer *timer) | |
994 | { | |
3c8aa39d | 995 | struct hrtimer_clock_base *base; |
c0a31329 TG |
996 | unsigned long flags; |
997 | int ret = -1; | |
998 | ||
19d9f422 TG |
999 | /* |
1000 | * Check lockless first. If the timer is not active (neither | |
1001 | * enqueued nor running the callback, nothing to do here. The | |
1002 | * base lock does not serialize against a concurrent enqueue, | |
1003 | * so we can avoid taking it. | |
1004 | */ | |
1005 | if (!hrtimer_active(timer)) | |
1006 | return 0; | |
1007 | ||
c0a31329 TG |
1008 | base = lock_hrtimer_base(timer, &flags); |
1009 | ||
303e967f | 1010 | if (!hrtimer_callback_running(timer)) |
8edfb036 | 1011 | ret = remove_hrtimer(timer, base, false); |
c0a31329 TG |
1012 | |
1013 | unlock_hrtimer_base(timer, &flags); | |
1014 | ||
1015 | return ret; | |
1016 | ||
1017 | } | |
8d16b764 | 1018 | EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel); |
c0a31329 TG |
1019 | |
1020 | /** | |
1021 | * hrtimer_cancel - cancel a timer and wait for the handler to finish. | |
c0a31329 TG |
1022 | * @timer: the timer to be cancelled |
1023 | * | |
1024 | * Returns: | |
1025 | * 0 when the timer was not active | |
1026 | * 1 when the timer was active | |
1027 | */ | |
1028 | int hrtimer_cancel(struct hrtimer *timer) | |
1029 | { | |
1030 | for (;;) { | |
1031 | int ret = hrtimer_try_to_cancel(timer); | |
1032 | ||
1033 | if (ret >= 0) | |
1034 | return ret; | |
5ef37b19 | 1035 | cpu_relax(); |
c0a31329 TG |
1036 | } |
1037 | } | |
8d16b764 | 1038 | EXPORT_SYMBOL_GPL(hrtimer_cancel); |
c0a31329 TG |
1039 | |
1040 | /** | |
1041 | * hrtimer_get_remaining - get remaining time for the timer | |
c0a31329 | 1042 | * @timer: the timer to read |
203cbf77 | 1043 | * @adjust: adjust relative timers when CONFIG_TIME_LOW_RES=y |
c0a31329 | 1044 | */ |
203cbf77 | 1045 | ktime_t __hrtimer_get_remaining(const struct hrtimer *timer, bool adjust) |
c0a31329 | 1046 | { |
c0a31329 TG |
1047 | unsigned long flags; |
1048 | ktime_t rem; | |
1049 | ||
b3bd3de6 | 1050 | lock_hrtimer_base(timer, &flags); |
203cbf77 TG |
1051 | if (IS_ENABLED(CONFIG_TIME_LOW_RES) && adjust) |
1052 | rem = hrtimer_expires_remaining_adjusted(timer); | |
1053 | else | |
1054 | rem = hrtimer_expires_remaining(timer); | |
c0a31329 TG |
1055 | unlock_hrtimer_base(timer, &flags); |
1056 | ||
1057 | return rem; | |
1058 | } | |
203cbf77 | 1059 | EXPORT_SYMBOL_GPL(__hrtimer_get_remaining); |
c0a31329 | 1060 | |
3451d024 | 1061 | #ifdef CONFIG_NO_HZ_COMMON |
69239749 TL |
1062 | /** |
1063 | * hrtimer_get_next_event - get the time until next expiry event | |
1064 | * | |
c1ad348b | 1065 | * Returns the next expiry time or KTIME_MAX if no timer is pending. |
69239749 | 1066 | */ |
c1ad348b | 1067 | u64 hrtimer_get_next_event(void) |
69239749 | 1068 | { |
dc5df73b | 1069 | struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases); |
c1ad348b | 1070 | u64 expires = KTIME_MAX; |
69239749 | 1071 | unsigned long flags; |
69239749 | 1072 | |
ecb49d1a | 1073 | raw_spin_lock_irqsave(&cpu_base->lock, flags); |
3c8aa39d | 1074 | |
e19ffe8b | 1075 | if (!__hrtimer_hres_active(cpu_base)) |
2456e855 | 1076 | expires = __hrtimer_get_next_event(cpu_base); |
3c8aa39d | 1077 | |
ecb49d1a | 1078 | raw_spin_unlock_irqrestore(&cpu_base->lock, flags); |
3c8aa39d | 1079 | |
c1ad348b | 1080 | return expires; |
69239749 TL |
1081 | } |
1082 | #endif | |
1083 | ||
336a9cde MZ |
1084 | static inline int hrtimer_clockid_to_base(clockid_t clock_id) |
1085 | { | |
1086 | if (likely(clock_id < MAX_CLOCKS)) { | |
1087 | int base = hrtimer_clock_to_base_table[clock_id]; | |
1088 | ||
1089 | if (likely(base != HRTIMER_MAX_CLOCK_BASES)) | |
1090 | return base; | |
1091 | } | |
1092 | WARN(1, "Invalid clockid %d. Using MONOTONIC\n", clock_id); | |
1093 | return HRTIMER_BASE_MONOTONIC; | |
1094 | } | |
1095 | ||
237fc6e7 TG |
1096 | static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id, |
1097 | enum hrtimer_mode mode) | |
c0a31329 | 1098 | { |
3c8aa39d | 1099 | struct hrtimer_cpu_base *cpu_base; |
e06383db | 1100 | int base; |
c0a31329 | 1101 | |
7978672c GA |
1102 | memset(timer, 0, sizeof(struct hrtimer)); |
1103 | ||
22127e93 | 1104 | cpu_base = raw_cpu_ptr(&hrtimer_bases); |
c0a31329 | 1105 | |
c9cb2e3d | 1106 | if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS) |
7978672c GA |
1107 | clock_id = CLOCK_MONOTONIC; |
1108 | ||
e06383db JS |
1109 | base = hrtimer_clockid_to_base(clock_id); |
1110 | timer->base = &cpu_base->clock_base[base]; | |
998adc3d | 1111 | timerqueue_init(&timer->node); |
c0a31329 | 1112 | } |
237fc6e7 TG |
1113 | |
1114 | /** | |
1115 | * hrtimer_init - initialize a timer to the given clock | |
1116 | * @timer: the timer to be initialized | |
1117 | * @clock_id: the clock to be used | |
1118 | * @mode: timer mode abs/rel | |
1119 | */ | |
1120 | void hrtimer_init(struct hrtimer *timer, clockid_t clock_id, | |
1121 | enum hrtimer_mode mode) | |
1122 | { | |
c6a2a177 | 1123 | debug_init(timer, clock_id, mode); |
237fc6e7 TG |
1124 | __hrtimer_init(timer, clock_id, mode); |
1125 | } | |
8d16b764 | 1126 | EXPORT_SYMBOL_GPL(hrtimer_init); |
c0a31329 | 1127 | |
887d9dc9 PZ |
1128 | /* |
1129 | * A timer is active, when it is enqueued into the rbtree or the | |
1130 | * callback function is running or it's in the state of being migrated | |
1131 | * to another cpu. | |
c0a31329 | 1132 | * |
887d9dc9 | 1133 | * It is important for this function to not return a false negative. |
c0a31329 | 1134 | */ |
887d9dc9 | 1135 | bool hrtimer_active(const struct hrtimer *timer) |
c0a31329 | 1136 | { |
3c8aa39d | 1137 | struct hrtimer_cpu_base *cpu_base; |
887d9dc9 | 1138 | unsigned int seq; |
c0a31329 | 1139 | |
887d9dc9 PZ |
1140 | do { |
1141 | cpu_base = READ_ONCE(timer->base->cpu_base); | |
1142 | seq = raw_read_seqcount_begin(&cpu_base->seq); | |
c0a31329 | 1143 | |
887d9dc9 PZ |
1144 | if (timer->state != HRTIMER_STATE_INACTIVE || |
1145 | cpu_base->running == timer) | |
1146 | return true; | |
1147 | ||
1148 | } while (read_seqcount_retry(&cpu_base->seq, seq) || | |
1149 | cpu_base != READ_ONCE(timer->base->cpu_base)); | |
1150 | ||
1151 | return false; | |
c0a31329 | 1152 | } |
887d9dc9 | 1153 | EXPORT_SYMBOL_GPL(hrtimer_active); |
c0a31329 | 1154 | |
887d9dc9 PZ |
1155 | /* |
1156 | * The write_seqcount_barrier()s in __run_hrtimer() split the thing into 3 | |
1157 | * distinct sections: | |
1158 | * | |
1159 | * - queued: the timer is queued | |
1160 | * - callback: the timer is being ran | |
1161 | * - post: the timer is inactive or (re)queued | |
1162 | * | |
1163 | * On the read side we ensure we observe timer->state and cpu_base->running | |
1164 | * from the same section, if anything changed while we looked at it, we retry. | |
1165 | * This includes timer->base changing because sequence numbers alone are | |
1166 | * insufficient for that. | |
1167 | * | |
1168 | * The sequence numbers are required because otherwise we could still observe | |
1169 | * a false negative if the read side got smeared over multiple consequtive | |
1170 | * __run_hrtimer() invocations. | |
1171 | */ | |
1172 | ||
21d6d52a TG |
1173 | static void __run_hrtimer(struct hrtimer_cpu_base *cpu_base, |
1174 | struct hrtimer_clock_base *base, | |
1175 | struct hrtimer *timer, ktime_t *now) | |
d3d74453 | 1176 | { |
d3d74453 PZ |
1177 | enum hrtimer_restart (*fn)(struct hrtimer *); |
1178 | int restart; | |
1179 | ||
887d9dc9 | 1180 | lockdep_assert_held(&cpu_base->lock); |
ca109491 | 1181 | |
c6a2a177 | 1182 | debug_deactivate(timer); |
887d9dc9 PZ |
1183 | cpu_base->running = timer; |
1184 | ||
1185 | /* | |
1186 | * Separate the ->running assignment from the ->state assignment. | |
1187 | * | |
1188 | * As with a regular write barrier, this ensures the read side in | |
1189 | * hrtimer_active() cannot observe cpu_base->running == NULL && | |
1190 | * timer->state == INACTIVE. | |
1191 | */ | |
1192 | raw_write_seqcount_barrier(&cpu_base->seq); | |
1193 | ||
1194 | __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE, 0); | |
d3d74453 | 1195 | fn = timer->function; |
ca109491 | 1196 | |
203cbf77 TG |
1197 | /* |
1198 | * Clear the 'is relative' flag for the TIME_LOW_RES case. If the | |
1199 | * timer is restarted with a period then it becomes an absolute | |
1200 | * timer. If its not restarted it does not matter. | |
1201 | */ | |
1202 | if (IS_ENABLED(CONFIG_TIME_LOW_RES)) | |
1203 | timer->is_rel = false; | |
1204 | ||
ca109491 PZ |
1205 | /* |
1206 | * Because we run timers from hardirq context, there is no chance | |
1207 | * they get migrated to another cpu, therefore its safe to unlock | |
1208 | * the timer base. | |
1209 | */ | |
ecb49d1a | 1210 | raw_spin_unlock(&cpu_base->lock); |
c6a2a177 | 1211 | trace_hrtimer_expire_entry(timer, now); |
ca109491 | 1212 | restart = fn(timer); |
c6a2a177 | 1213 | trace_hrtimer_expire_exit(timer); |
ecb49d1a | 1214 | raw_spin_lock(&cpu_base->lock); |
d3d74453 PZ |
1215 | |
1216 | /* | |
887d9dc9 | 1217 | * Note: We clear the running state after enqueue_hrtimer and |
b4d90e9f | 1218 | * we do not reprogram the event hardware. Happens either in |
e3f1d883 | 1219 | * hrtimer_start_range_ns() or in hrtimer_interrupt() |
5de2755c PZ |
1220 | * |
1221 | * Note: Because we dropped the cpu_base->lock above, | |
1222 | * hrtimer_start_range_ns() can have popped in and enqueued the timer | |
1223 | * for us already. | |
d3d74453 | 1224 | */ |
5de2755c PZ |
1225 | if (restart != HRTIMER_NORESTART && |
1226 | !(timer->state & HRTIMER_STATE_ENQUEUED)) | |
a6037b61 | 1227 | enqueue_hrtimer(timer, base); |
f13d4f97 | 1228 | |
887d9dc9 PZ |
1229 | /* |
1230 | * Separate the ->running assignment from the ->state assignment. | |
1231 | * | |
1232 | * As with a regular write barrier, this ensures the read side in | |
1233 | * hrtimer_active() cannot observe cpu_base->running == NULL && | |
1234 | * timer->state == INACTIVE. | |
1235 | */ | |
1236 | raw_write_seqcount_barrier(&cpu_base->seq); | |
f13d4f97 | 1237 | |
887d9dc9 PZ |
1238 | WARN_ON_ONCE(cpu_base->running != timer); |
1239 | cpu_base->running = NULL; | |
d3d74453 PZ |
1240 | } |
1241 | ||
21d6d52a | 1242 | static void __hrtimer_run_queues(struct hrtimer_cpu_base *cpu_base, ktime_t now) |
54cdfdb4 | 1243 | { |
34aee88a TG |
1244 | struct hrtimer_clock_base *base = cpu_base->clock_base; |
1245 | unsigned int active = cpu_base->active_bases; | |
6ff7041d | 1246 | |
34aee88a | 1247 | for (; active; base++, active >>= 1) { |
998adc3d | 1248 | struct timerqueue_node *node; |
ab8177bc TG |
1249 | ktime_t basenow; |
1250 | ||
34aee88a | 1251 | if (!(active & 0x01)) |
ab8177bc | 1252 | continue; |
54cdfdb4 | 1253 | |
54cdfdb4 TG |
1254 | basenow = ktime_add(now, base->offset); |
1255 | ||
998adc3d | 1256 | while ((node = timerqueue_getnext(&base->active))) { |
54cdfdb4 TG |
1257 | struct hrtimer *timer; |
1258 | ||
998adc3d | 1259 | timer = container_of(node, struct hrtimer, node); |
54cdfdb4 | 1260 | |
654c8e0b AV |
1261 | /* |
1262 | * The immediate goal for using the softexpires is | |
1263 | * minimizing wakeups, not running timers at the | |
1264 | * earliest interrupt after their soft expiration. | |
1265 | * This allows us to avoid using a Priority Search | |
1266 | * Tree, which can answer a stabbing querry for | |
1267 | * overlapping intervals and instead use the simple | |
1268 | * BST we already have. | |
1269 | * We don't add extra wakeups by delaying timers that | |
1270 | * are right-of a not yet expired timer, because that | |
1271 | * timer will have to trigger a wakeup anyway. | |
1272 | */ | |
2456e855 | 1273 | if (basenow < hrtimer_get_softexpires_tv64(timer)) |
54cdfdb4 | 1274 | break; |
54cdfdb4 | 1275 | |
21d6d52a | 1276 | __run_hrtimer(cpu_base, base, timer, &basenow); |
54cdfdb4 | 1277 | } |
54cdfdb4 | 1278 | } |
21d6d52a TG |
1279 | } |
1280 | ||
1281 | #ifdef CONFIG_HIGH_RES_TIMERS | |
1282 | ||
1283 | /* | |
1284 | * High resolution timer interrupt | |
1285 | * Called with interrupts disabled | |
1286 | */ | |
1287 | void hrtimer_interrupt(struct clock_event_device *dev) | |
1288 | { | |
1289 | struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases); | |
1290 | ktime_t expires_next, now, entry_time, delta; | |
1291 | int retries = 0; | |
1292 | ||
1293 | BUG_ON(!cpu_base->hres_active); | |
1294 | cpu_base->nr_events++; | |
2456e855 | 1295 | dev->next_event = KTIME_MAX; |
21d6d52a TG |
1296 | |
1297 | raw_spin_lock(&cpu_base->lock); | |
1298 | entry_time = now = hrtimer_update_base(cpu_base); | |
1299 | retry: | |
1300 | cpu_base->in_hrtirq = 1; | |
1301 | /* | |
1302 | * We set expires_next to KTIME_MAX here with cpu_base->lock | |
1303 | * held to prevent that a timer is enqueued in our queue via | |
1304 | * the migration code. This does not affect enqueueing of | |
1305 | * timers which run their callback and need to be requeued on | |
1306 | * this CPU. | |
1307 | */ | |
2456e855 | 1308 | cpu_base->expires_next = KTIME_MAX; |
21d6d52a TG |
1309 | |
1310 | __hrtimer_run_queues(cpu_base, now); | |
1311 | ||
9bc74919 TG |
1312 | /* Reevaluate the clock bases for the next expiry */ |
1313 | expires_next = __hrtimer_get_next_event(cpu_base); | |
6ff7041d TG |
1314 | /* |
1315 | * Store the new expiry value so the migration code can verify | |
1316 | * against it. | |
1317 | */ | |
54cdfdb4 | 1318 | cpu_base->expires_next = expires_next; |
9bc74919 | 1319 | cpu_base->in_hrtirq = 0; |
ecb49d1a | 1320 | raw_spin_unlock(&cpu_base->lock); |
54cdfdb4 TG |
1321 | |
1322 | /* Reprogramming necessary ? */ | |
d2540875 | 1323 | if (!tick_program_event(expires_next, 0)) { |
41d2e494 TG |
1324 | cpu_base->hang_detected = 0; |
1325 | return; | |
54cdfdb4 | 1326 | } |
41d2e494 TG |
1327 | |
1328 | /* | |
1329 | * The next timer was already expired due to: | |
1330 | * - tracing | |
1331 | * - long lasting callbacks | |
1332 | * - being scheduled away when running in a VM | |
1333 | * | |
1334 | * We need to prevent that we loop forever in the hrtimer | |
1335 | * interrupt routine. We give it 3 attempts to avoid | |
1336 | * overreacting on some spurious event. | |
5baefd6d JS |
1337 | * |
1338 | * Acquire base lock for updating the offsets and retrieving | |
1339 | * the current time. | |
41d2e494 | 1340 | */ |
196951e9 | 1341 | raw_spin_lock(&cpu_base->lock); |
5baefd6d | 1342 | now = hrtimer_update_base(cpu_base); |
41d2e494 TG |
1343 | cpu_base->nr_retries++; |
1344 | if (++retries < 3) | |
1345 | goto retry; | |
1346 | /* | |
1347 | * Give the system a chance to do something else than looping | |
1348 | * here. We stored the entry time, so we know exactly how long | |
1349 | * we spent here. We schedule the next event this amount of | |
1350 | * time away. | |
1351 | */ | |
1352 | cpu_base->nr_hangs++; | |
1353 | cpu_base->hang_detected = 1; | |
196951e9 | 1354 | raw_spin_unlock(&cpu_base->lock); |
41d2e494 | 1355 | delta = ktime_sub(now, entry_time); |
2456e855 TG |
1356 | if ((unsigned int)delta > cpu_base->max_hang_time) |
1357 | cpu_base->max_hang_time = (unsigned int) delta; | |
41d2e494 TG |
1358 | /* |
1359 | * Limit it to a sensible value as we enforce a longer | |
1360 | * delay. Give the CPU at least 100ms to catch up. | |
1361 | */ | |
2456e855 | 1362 | if (delta > 100 * NSEC_PER_MSEC) |
41d2e494 TG |
1363 | expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC); |
1364 | else | |
1365 | expires_next = ktime_add(now, delta); | |
1366 | tick_program_event(expires_next, 1); | |
1367 | printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n", | |
1368 | ktime_to_ns(delta)); | |
54cdfdb4 TG |
1369 | } |
1370 | ||
016da201 | 1371 | /* called with interrupts disabled */ |
c6eb3f70 | 1372 | static inline void __hrtimer_peek_ahead_timers(void) |
8bdec955 TG |
1373 | { |
1374 | struct tick_device *td; | |
1375 | ||
1376 | if (!hrtimer_hres_active()) | |
1377 | return; | |
1378 | ||
22127e93 | 1379 | td = this_cpu_ptr(&tick_cpu_device); |
8bdec955 TG |
1380 | if (td && td->evtdev) |
1381 | hrtimer_interrupt(td->evtdev); | |
1382 | } | |
1383 | ||
82c5b7b5 IM |
1384 | #else /* CONFIG_HIGH_RES_TIMERS */ |
1385 | ||
1386 | static inline void __hrtimer_peek_ahead_timers(void) { } | |
1387 | ||
1388 | #endif /* !CONFIG_HIGH_RES_TIMERS */ | |
82f67cd9 | 1389 | |
d3d74453 | 1390 | /* |
c6eb3f70 | 1391 | * Called from run_local_timers in hardirq context every jiffy |
d3d74453 | 1392 | */ |
833883d9 | 1393 | void hrtimer_run_queues(void) |
d3d74453 | 1394 | { |
dc5df73b | 1395 | struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases); |
21d6d52a | 1396 | ktime_t now; |
c0a31329 | 1397 | |
e19ffe8b | 1398 | if (__hrtimer_hres_active(cpu_base)) |
d3d74453 | 1399 | return; |
54cdfdb4 | 1400 | |
d3d74453 | 1401 | /* |
c6eb3f70 TG |
1402 | * This _is_ ugly: We have to check periodically, whether we |
1403 | * can switch to highres and / or nohz mode. The clocksource | |
1404 | * switch happens with xtime_lock held. Notification from | |
1405 | * there only sets the check bit in the tick_oneshot code, | |
1406 | * otherwise we might deadlock vs. xtime_lock. | |
d3d74453 | 1407 | */ |
c6eb3f70 | 1408 | if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) { |
d3d74453 | 1409 | hrtimer_switch_to_hres(); |
3055adda | 1410 | return; |
833883d9 | 1411 | } |
c6eb3f70 | 1412 | |
21d6d52a TG |
1413 | raw_spin_lock(&cpu_base->lock); |
1414 | now = hrtimer_update_base(cpu_base); | |
1415 | __hrtimer_run_queues(cpu_base, now); | |
1416 | raw_spin_unlock(&cpu_base->lock); | |
c0a31329 TG |
1417 | } |
1418 | ||
10c94ec1 TG |
1419 | /* |
1420 | * Sleep related functions: | |
1421 | */ | |
c9cb2e3d | 1422 | static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer) |
00362e33 TG |
1423 | { |
1424 | struct hrtimer_sleeper *t = | |
1425 | container_of(timer, struct hrtimer_sleeper, timer); | |
1426 | struct task_struct *task = t->task; | |
1427 | ||
1428 | t->task = NULL; | |
1429 | if (task) | |
1430 | wake_up_process(task); | |
1431 | ||
1432 | return HRTIMER_NORESTART; | |
1433 | } | |
1434 | ||
36c8b586 | 1435 | void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task) |
00362e33 TG |
1436 | { |
1437 | sl->timer.function = hrtimer_wakeup; | |
1438 | sl->task = task; | |
1439 | } | |
2bc481cf | 1440 | EXPORT_SYMBOL_GPL(hrtimer_init_sleeper); |
00362e33 | 1441 | |
669d7868 | 1442 | static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode) |
432569bb | 1443 | { |
a7602681 | 1444 | struct timespec __user *rmtp; |
669d7868 | 1445 | hrtimer_init_sleeper(t, current); |
10c94ec1 | 1446 | |
432569bb RZ |
1447 | do { |
1448 | set_current_state(TASK_INTERRUPTIBLE); | |
cc584b21 | 1449 | hrtimer_start_expires(&t->timer, mode); |
432569bb | 1450 | |
54cdfdb4 | 1451 | if (likely(t->task)) |
b0f8c44f | 1452 | freezable_schedule(); |
432569bb | 1453 | |
669d7868 | 1454 | hrtimer_cancel(&t->timer); |
c9cb2e3d | 1455 | mode = HRTIMER_MODE_ABS; |
669d7868 TG |
1456 | |
1457 | } while (t->task && !signal_pending(current)); | |
432569bb | 1458 | |
3588a085 PZ |
1459 | __set_current_state(TASK_RUNNING); |
1460 | ||
a7602681 | 1461 | if (!t->task) |
080344b9 | 1462 | return 0; |
080344b9 | 1463 | |
a7602681 AV |
1464 | rmtp = current->restart_block.nanosleep.rmtp; |
1465 | if (rmtp) { | |
1466 | struct timespec rmt; | |
1467 | ktime_t rem = hrtimer_expires_remaining(&t->timer); | |
1468 | if (rem <= 0) | |
1469 | return 0; | |
1470 | rmt = ktime_to_timespec(rem); | |
1471 | ||
1472 | if (copy_to_user(rmtp, &rmt, sizeof(*rmtp))) | |
1473 | return -EFAULT; | |
1474 | } | |
1475 | return -ERESTART_RESTARTBLOCK; | |
080344b9 ON |
1476 | } |
1477 | ||
1711ef38 | 1478 | long __sched hrtimer_nanosleep_restart(struct restart_block *restart) |
10c94ec1 | 1479 | { |
669d7868 | 1480 | struct hrtimer_sleeper t; |
a7602681 | 1481 | int ret; |
10c94ec1 | 1482 | |
ab8177bc | 1483 | hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid, |
237fc6e7 | 1484 | HRTIMER_MODE_ABS); |
cc584b21 | 1485 | hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires); |
10c94ec1 | 1486 | |
a7602681 | 1487 | ret = do_nanosleep(&t, HRTIMER_MODE_ABS); |
237fc6e7 TG |
1488 | destroy_hrtimer_on_stack(&t.timer); |
1489 | return ret; | |
10c94ec1 TG |
1490 | } |
1491 | ||
192a82f9 | 1492 | long hrtimer_nanosleep(struct timespec64 *rqtp, |
10c94ec1 TG |
1493 | const enum hrtimer_mode mode, const clockid_t clockid) |
1494 | { | |
a7602681 | 1495 | struct restart_block *restart; |
669d7868 | 1496 | struct hrtimer_sleeper t; |
237fc6e7 | 1497 | int ret = 0; |
da8b44d5 | 1498 | u64 slack; |
3bd01206 AV |
1499 | |
1500 | slack = current->timer_slack_ns; | |
aab03e05 | 1501 | if (dl_task(current) || rt_task(current)) |
3bd01206 | 1502 | slack = 0; |
10c94ec1 | 1503 | |
237fc6e7 | 1504 | hrtimer_init_on_stack(&t.timer, clockid, mode); |
ad196384 | 1505 | hrtimer_set_expires_range_ns(&t.timer, timespec64_to_ktime(*rqtp), slack); |
a7602681 AV |
1506 | ret = do_nanosleep(&t, mode); |
1507 | if (ret != -ERESTART_RESTARTBLOCK) | |
237fc6e7 | 1508 | goto out; |
10c94ec1 | 1509 | |
7978672c | 1510 | /* Absolute timers do not update the rmtp value and restart: */ |
237fc6e7 TG |
1511 | if (mode == HRTIMER_MODE_ABS) { |
1512 | ret = -ERESTARTNOHAND; | |
1513 | goto out; | |
1514 | } | |
10c94ec1 | 1515 | |
a7602681 | 1516 | restart = ¤t->restart_block; |
1711ef38 | 1517 | restart->fn = hrtimer_nanosleep_restart; |
ab8177bc | 1518 | restart->nanosleep.clockid = t.timer.base->clockid; |
cc584b21 | 1519 | restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer); |
237fc6e7 TG |
1520 | out: |
1521 | destroy_hrtimer_on_stack(&t.timer); | |
1522 | return ret; | |
10c94ec1 TG |
1523 | } |
1524 | ||
58fd3aa2 HC |
1525 | SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp, |
1526 | struct timespec __user *, rmtp) | |
6ba1b912 | 1527 | { |
ad196384 | 1528 | struct timespec64 tu64; |
080344b9 | 1529 | struct timespec tu; |
6ba1b912 TG |
1530 | |
1531 | if (copy_from_user(&tu, rqtp, sizeof(tu))) | |
1532 | return -EFAULT; | |
1533 | ||
ad196384 DD |
1534 | tu64 = timespec_to_timespec64(tu); |
1535 | if (!timespec64_valid(&tu64)) | |
6ba1b912 TG |
1536 | return -EINVAL; |
1537 | ||
192a82f9 AV |
1538 | current->restart_block.nanosleep.rmtp = rmtp; |
1539 | return hrtimer_nanosleep(&tu64, HRTIMER_MODE_REL, CLOCK_MONOTONIC); | |
6ba1b912 TG |
1540 | } |
1541 | ||
c0a31329 TG |
1542 | /* |
1543 | * Functions related to boot-time initialization: | |
1544 | */ | |
27590dc1 | 1545 | int hrtimers_prepare_cpu(unsigned int cpu) |
c0a31329 | 1546 | { |
3c8aa39d | 1547 | struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu); |
c0a31329 TG |
1548 | int i; |
1549 | ||
998adc3d | 1550 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
3c8aa39d | 1551 | cpu_base->clock_base[i].cpu_base = cpu_base; |
998adc3d JS |
1552 | timerqueue_init_head(&cpu_base->clock_base[i].active); |
1553 | } | |
3c8aa39d | 1554 | |
cddd0248 | 1555 | cpu_base->cpu = cpu; |
54cdfdb4 | 1556 | hrtimer_init_hres(cpu_base); |
27590dc1 | 1557 | return 0; |
c0a31329 TG |
1558 | } |
1559 | ||
1560 | #ifdef CONFIG_HOTPLUG_CPU | |
1561 | ||
ca109491 | 1562 | static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base, |
37810659 | 1563 | struct hrtimer_clock_base *new_base) |
c0a31329 TG |
1564 | { |
1565 | struct hrtimer *timer; | |
998adc3d | 1566 | struct timerqueue_node *node; |
c0a31329 | 1567 | |
998adc3d JS |
1568 | while ((node = timerqueue_getnext(&old_base->active))) { |
1569 | timer = container_of(node, struct hrtimer, node); | |
54cdfdb4 | 1570 | BUG_ON(hrtimer_callback_running(timer)); |
c6a2a177 | 1571 | debug_deactivate(timer); |
b00c1a99 TG |
1572 | |
1573 | /* | |
c04dca02 | 1574 | * Mark it as ENQUEUED not INACTIVE otherwise the |
b00c1a99 TG |
1575 | * timer could be seen as !active and just vanish away |
1576 | * under us on another CPU | |
1577 | */ | |
c04dca02 | 1578 | __remove_hrtimer(timer, old_base, HRTIMER_STATE_ENQUEUED, 0); |
c0a31329 | 1579 | timer->base = new_base; |
54cdfdb4 | 1580 | /* |
e3f1d883 TG |
1581 | * Enqueue the timers on the new cpu. This does not |
1582 | * reprogram the event device in case the timer | |
1583 | * expires before the earliest on this CPU, but we run | |
1584 | * hrtimer_interrupt after we migrated everything to | |
1585 | * sort out already expired timers and reprogram the | |
1586 | * event device. | |
54cdfdb4 | 1587 | */ |
a6037b61 | 1588 | enqueue_hrtimer(timer, new_base); |
c0a31329 TG |
1589 | } |
1590 | } | |
1591 | ||
27590dc1 | 1592 | int hrtimers_dead_cpu(unsigned int scpu) |
c0a31329 | 1593 | { |
3c8aa39d | 1594 | struct hrtimer_cpu_base *old_base, *new_base; |
731a55ba | 1595 | int i; |
c0a31329 | 1596 | |
37810659 | 1597 | BUG_ON(cpu_online(scpu)); |
37810659 | 1598 | tick_cancel_sched_timer(scpu); |
731a55ba TG |
1599 | |
1600 | local_irq_disable(); | |
1601 | old_base = &per_cpu(hrtimer_bases, scpu); | |
dc5df73b | 1602 | new_base = this_cpu_ptr(&hrtimer_bases); |
d82f0b0f ON |
1603 | /* |
1604 | * The caller is globally serialized and nobody else | |
1605 | * takes two locks at once, deadlock is not possible. | |
1606 | */ | |
ecb49d1a TG |
1607 | raw_spin_lock(&new_base->lock); |
1608 | raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING); | |
c0a31329 | 1609 | |
3c8aa39d | 1610 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
ca109491 | 1611 | migrate_hrtimer_list(&old_base->clock_base[i], |
37810659 | 1612 | &new_base->clock_base[i]); |
c0a31329 TG |
1613 | } |
1614 | ||
ecb49d1a TG |
1615 | raw_spin_unlock(&old_base->lock); |
1616 | raw_spin_unlock(&new_base->lock); | |
37810659 | 1617 | |
731a55ba TG |
1618 | /* Check, if we got expired work to do */ |
1619 | __hrtimer_peek_ahead_timers(); | |
1620 | local_irq_enable(); | |
27590dc1 | 1621 | return 0; |
c0a31329 | 1622 | } |
37810659 | 1623 | |
c0a31329 TG |
1624 | #endif /* CONFIG_HOTPLUG_CPU */ |
1625 | ||
c0a31329 TG |
1626 | void __init hrtimers_init(void) |
1627 | { | |
27590dc1 | 1628 | hrtimers_prepare_cpu(smp_processor_id()); |
c0a31329 TG |
1629 | } |
1630 | ||
7bb67439 | 1631 | /** |
351b3f7a | 1632 | * schedule_hrtimeout_range_clock - sleep until timeout |
7bb67439 | 1633 | * @expires: timeout value (ktime_t) |
654c8e0b | 1634 | * @delta: slack in expires timeout (ktime_t) |
7bb67439 | 1635 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL |
351b3f7a | 1636 | * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME |
7bb67439 | 1637 | */ |
351b3f7a | 1638 | int __sched |
da8b44d5 | 1639 | schedule_hrtimeout_range_clock(ktime_t *expires, u64 delta, |
351b3f7a | 1640 | const enum hrtimer_mode mode, int clock) |
7bb67439 AV |
1641 | { |
1642 | struct hrtimer_sleeper t; | |
1643 | ||
1644 | /* | |
1645 | * Optimize when a zero timeout value is given. It does not | |
1646 | * matter whether this is an absolute or a relative time. | |
1647 | */ | |
2456e855 | 1648 | if (expires && *expires == 0) { |
7bb67439 AV |
1649 | __set_current_state(TASK_RUNNING); |
1650 | return 0; | |
1651 | } | |
1652 | ||
1653 | /* | |
43b21013 | 1654 | * A NULL parameter means "infinite" |
7bb67439 AV |
1655 | */ |
1656 | if (!expires) { | |
1657 | schedule(); | |
7bb67439 AV |
1658 | return -EINTR; |
1659 | } | |
1660 | ||
351b3f7a | 1661 | hrtimer_init_on_stack(&t.timer, clock, mode); |
654c8e0b | 1662 | hrtimer_set_expires_range_ns(&t.timer, *expires, delta); |
7bb67439 AV |
1663 | |
1664 | hrtimer_init_sleeper(&t, current); | |
1665 | ||
cc584b21 | 1666 | hrtimer_start_expires(&t.timer, mode); |
7bb67439 AV |
1667 | |
1668 | if (likely(t.task)) | |
1669 | schedule(); | |
1670 | ||
1671 | hrtimer_cancel(&t.timer); | |
1672 | destroy_hrtimer_on_stack(&t.timer); | |
1673 | ||
1674 | __set_current_state(TASK_RUNNING); | |
1675 | ||
1676 | return !t.task ? 0 : -EINTR; | |
1677 | } | |
351b3f7a CE |
1678 | |
1679 | /** | |
1680 | * schedule_hrtimeout_range - sleep until timeout | |
1681 | * @expires: timeout value (ktime_t) | |
1682 | * @delta: slack in expires timeout (ktime_t) | |
1683 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL | |
1684 | * | |
1685 | * Make the current task sleep until the given expiry time has | |
1686 | * elapsed. The routine will return immediately unless | |
1687 | * the current task state has been set (see set_current_state()). | |
1688 | * | |
1689 | * The @delta argument gives the kernel the freedom to schedule the | |
1690 | * actual wakeup to a time that is both power and performance friendly. | |
1691 | * The kernel give the normal best effort behavior for "@expires+@delta", | |
1692 | * but may decide to fire the timer earlier, but no earlier than @expires. | |
1693 | * | |
1694 | * You can set the task state as follows - | |
1695 | * | |
1696 | * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to | |
4b7e9cf9 DA |
1697 | * pass before the routine returns unless the current task is explicitly |
1698 | * woken up, (e.g. by wake_up_process()). | |
351b3f7a CE |
1699 | * |
1700 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
4b7e9cf9 DA |
1701 | * delivered to the current task or the current task is explicitly woken |
1702 | * up. | |
351b3f7a CE |
1703 | * |
1704 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1705 | * routine returns. | |
1706 | * | |
4b7e9cf9 DA |
1707 | * Returns 0 when the timer has expired. If the task was woken before the |
1708 | * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or | |
1709 | * by an explicit wakeup, it returns -EINTR. | |
351b3f7a | 1710 | */ |
da8b44d5 | 1711 | int __sched schedule_hrtimeout_range(ktime_t *expires, u64 delta, |
351b3f7a CE |
1712 | const enum hrtimer_mode mode) |
1713 | { | |
1714 | return schedule_hrtimeout_range_clock(expires, delta, mode, | |
1715 | CLOCK_MONOTONIC); | |
1716 | } | |
654c8e0b AV |
1717 | EXPORT_SYMBOL_GPL(schedule_hrtimeout_range); |
1718 | ||
1719 | /** | |
1720 | * schedule_hrtimeout - sleep until timeout | |
1721 | * @expires: timeout value (ktime_t) | |
1722 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL | |
1723 | * | |
1724 | * Make the current task sleep until the given expiry time has | |
1725 | * elapsed. The routine will return immediately unless | |
1726 | * the current task state has been set (see set_current_state()). | |
1727 | * | |
1728 | * You can set the task state as follows - | |
1729 | * | |
1730 | * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to | |
4b7e9cf9 DA |
1731 | * pass before the routine returns unless the current task is explicitly |
1732 | * woken up, (e.g. by wake_up_process()). | |
654c8e0b AV |
1733 | * |
1734 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
4b7e9cf9 DA |
1735 | * delivered to the current task or the current task is explicitly woken |
1736 | * up. | |
654c8e0b AV |
1737 | * |
1738 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1739 | * routine returns. | |
1740 | * | |
4b7e9cf9 DA |
1741 | * Returns 0 when the timer has expired. If the task was woken before the |
1742 | * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or | |
1743 | * by an explicit wakeup, it returns -EINTR. | |
654c8e0b AV |
1744 | */ |
1745 | int __sched schedule_hrtimeout(ktime_t *expires, | |
1746 | const enum hrtimer_mode mode) | |
1747 | { | |
1748 | return schedule_hrtimeout_range(expires, 0, mode); | |
1749 | } | |
7bb67439 | 1750 | EXPORT_SYMBOL_GPL(schedule_hrtimeout); |