]>
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> |
eea08f32 | 46 | #include <linux/sched.h> |
cf4aebc2 | 47 | #include <linux/sched/sysctl.h> |
8bd75c77 | 48 | #include <linux/sched/rt.h> |
aab03e05 | 49 | #include <linux/sched/deadline.h> |
eea08f32 | 50 | #include <linux/timer.h> |
b0f8c44f | 51 | #include <linux/freezer.h> |
c0a31329 TG |
52 | |
53 | #include <asm/uaccess.h> | |
54 | ||
c6a2a177 XG |
55 | #include <trace/events/timer.h> |
56 | ||
c1797baf | 57 | #include "tick-internal.h" |
8b094cd0 | 58 | |
c0a31329 TG |
59 | /* |
60 | * The timer bases: | |
7978672c | 61 | * |
e06383db JS |
62 | * There are more clockids then hrtimer bases. Thus, we index |
63 | * into the timer bases by the hrtimer_base_type enum. When trying | |
64 | * to reach a base using a clockid, hrtimer_clockid_to_base() | |
65 | * is used to convert from clockid to the proper hrtimer_base_type. | |
c0a31329 | 66 | */ |
54cdfdb4 | 67 | DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) = |
c0a31329 | 68 | { |
84cc8fd2 | 69 | .lock = __RAW_SPIN_LOCK_UNLOCKED(hrtimer_bases.lock), |
3c8aa39d | 70 | .clock_base = |
c0a31329 | 71 | { |
3c8aa39d | 72 | { |
ab8177bc TG |
73 | .index = HRTIMER_BASE_MONOTONIC, |
74 | .clockid = CLOCK_MONOTONIC, | |
3c8aa39d | 75 | .get_time = &ktime_get, |
3c8aa39d | 76 | }, |
68fa61c0 TG |
77 | { |
78 | .index = HRTIMER_BASE_REALTIME, | |
79 | .clockid = CLOCK_REALTIME, | |
80 | .get_time = &ktime_get_real, | |
68fa61c0 | 81 | }, |
70a08cca | 82 | { |
ab8177bc TG |
83 | .index = HRTIMER_BASE_BOOTTIME, |
84 | .clockid = CLOCK_BOOTTIME, | |
70a08cca | 85 | .get_time = &ktime_get_boottime, |
70a08cca | 86 | }, |
90adda98 JS |
87 | { |
88 | .index = HRTIMER_BASE_TAI, | |
89 | .clockid = CLOCK_TAI, | |
90 | .get_time = &ktime_get_clocktai, | |
90adda98 | 91 | }, |
3c8aa39d | 92 | } |
c0a31329 TG |
93 | }; |
94 | ||
942c3c5c | 95 | static const int hrtimer_clock_to_base_table[MAX_CLOCKS] = { |
ce31332d TG |
96 | [CLOCK_REALTIME] = HRTIMER_BASE_REALTIME, |
97 | [CLOCK_MONOTONIC] = HRTIMER_BASE_MONOTONIC, | |
98 | [CLOCK_BOOTTIME] = HRTIMER_BASE_BOOTTIME, | |
90adda98 | 99 | [CLOCK_TAI] = HRTIMER_BASE_TAI, |
ce31332d | 100 | }; |
e06383db JS |
101 | |
102 | static inline int hrtimer_clockid_to_base(clockid_t clock_id) | |
103 | { | |
104 | return hrtimer_clock_to_base_table[clock_id]; | |
105 | } | |
106 | ||
c0a31329 TG |
107 | /* |
108 | * Functions and macros which are different for UP/SMP systems are kept in a | |
109 | * single place | |
110 | */ | |
111 | #ifdef CONFIG_SMP | |
112 | ||
c0a31329 TG |
113 | /* |
114 | * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock | |
115 | * means that all timers which are tied to this base via timer->base are | |
116 | * locked, and the base itself is locked too. | |
117 | * | |
118 | * So __run_timers/migrate_timers can safely modify all timers which could | |
119 | * be found on the lists/queues. | |
120 | * | |
121 | * When the timer's base is locked, and the timer removed from list, it is | |
122 | * possible to set timer->base = NULL and drop the lock: the timer remains | |
123 | * locked. | |
124 | */ | |
3c8aa39d TG |
125 | static |
126 | struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer, | |
127 | unsigned long *flags) | |
c0a31329 | 128 | { |
3c8aa39d | 129 | struct hrtimer_clock_base *base; |
c0a31329 TG |
130 | |
131 | for (;;) { | |
132 | base = timer->base; | |
133 | if (likely(base != NULL)) { | |
ecb49d1a | 134 | raw_spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
135 | if (likely(base == timer->base)) |
136 | return base; | |
137 | /* The timer has migrated to another CPU: */ | |
ecb49d1a | 138 | raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags); |
c0a31329 TG |
139 | } |
140 | cpu_relax(); | |
141 | } | |
142 | } | |
143 | ||
6ff7041d TG |
144 | /* |
145 | * With HIGHRES=y we do not migrate the timer when it is expiring | |
146 | * before the next event on the target cpu because we cannot reprogram | |
147 | * the target cpu hardware and we would cause it to fire late. | |
148 | * | |
149 | * Called with cpu_base->lock of target cpu held. | |
150 | */ | |
151 | static int | |
152 | hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base) | |
153 | { | |
154 | #ifdef CONFIG_HIGH_RES_TIMERS | |
155 | ktime_t expires; | |
156 | ||
157 | if (!new_base->cpu_base->hres_active) | |
158 | return 0; | |
159 | ||
160 | expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset); | |
161 | return expires.tv64 <= new_base->cpu_base->expires_next.tv64; | |
162 | #else | |
163 | return 0; | |
164 | #endif | |
165 | } | |
166 | ||
c0a31329 TG |
167 | /* |
168 | * Switch the timer base to the current CPU when possible. | |
169 | */ | |
3c8aa39d | 170 | static inline struct hrtimer_clock_base * |
597d0275 AB |
171 | switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base, |
172 | int pinned) | |
c0a31329 | 173 | { |
3c8aa39d TG |
174 | struct hrtimer_clock_base *new_base; |
175 | struct hrtimer_cpu_base *new_cpu_base; | |
6ff7041d | 176 | int this_cpu = smp_processor_id(); |
6201b4d6 | 177 | int cpu = get_nohz_timer_target(pinned); |
ab8177bc | 178 | int basenum = base->index; |
c0a31329 | 179 | |
eea08f32 AB |
180 | again: |
181 | new_cpu_base = &per_cpu(hrtimer_bases, cpu); | |
e06383db | 182 | new_base = &new_cpu_base->clock_base[basenum]; |
c0a31329 TG |
183 | |
184 | if (base != new_base) { | |
185 | /* | |
6ff7041d | 186 | * We are trying to move timer to new_base. |
c0a31329 TG |
187 | * However we can't change timer's base while it is running, |
188 | * so we keep it on the same CPU. No hassle vs. reprogramming | |
189 | * the event source in the high resolution case. The softirq | |
190 | * code will take care of this when the timer function has | |
191 | * completed. There is no conflict as we hold the lock until | |
192 | * the timer is enqueued. | |
193 | */ | |
54cdfdb4 | 194 | if (unlikely(hrtimer_callback_running(timer))) |
c0a31329 TG |
195 | return base; |
196 | ||
197 | /* See the comment in lock_timer_base() */ | |
198 | timer->base = NULL; | |
ecb49d1a TG |
199 | raw_spin_unlock(&base->cpu_base->lock); |
200 | raw_spin_lock(&new_base->cpu_base->lock); | |
eea08f32 | 201 | |
6ff7041d TG |
202 | if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) { |
203 | cpu = this_cpu; | |
ecb49d1a TG |
204 | raw_spin_unlock(&new_base->cpu_base->lock); |
205 | raw_spin_lock(&base->cpu_base->lock); | |
6ff7041d TG |
206 | timer->base = base; |
207 | goto again; | |
eea08f32 | 208 | } |
c0a31329 | 209 | timer->base = new_base; |
012a45e3 LM |
210 | } else { |
211 | if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) { | |
212 | cpu = this_cpu; | |
213 | goto again; | |
214 | } | |
c0a31329 TG |
215 | } |
216 | return new_base; | |
217 | } | |
218 | ||
219 | #else /* CONFIG_SMP */ | |
220 | ||
3c8aa39d | 221 | static inline struct hrtimer_clock_base * |
c0a31329 TG |
222 | lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) |
223 | { | |
3c8aa39d | 224 | struct hrtimer_clock_base *base = timer->base; |
c0a31329 | 225 | |
ecb49d1a | 226 | raw_spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
227 | |
228 | return base; | |
229 | } | |
230 | ||
eea08f32 | 231 | # define switch_hrtimer_base(t, b, p) (b) |
c0a31329 TG |
232 | |
233 | #endif /* !CONFIG_SMP */ | |
234 | ||
235 | /* | |
236 | * Functions for the union type storage format of ktime_t which are | |
237 | * too large for inlining: | |
238 | */ | |
239 | #if BITS_PER_LONG < 64 | |
c0a31329 TG |
240 | /* |
241 | * Divide a ktime value by a nanosecond value | |
242 | */ | |
8b618628 | 243 | u64 __ktime_divns(const ktime_t kt, s64 div) |
c0a31329 | 244 | { |
900cfa46 | 245 | u64 dclc; |
c0a31329 TG |
246 | int sft = 0; |
247 | ||
900cfa46 | 248 | dclc = ktime_to_ns(kt); |
c0a31329 TG |
249 | /* Make sure the divisor is less than 2^32: */ |
250 | while (div >> 32) { | |
251 | sft++; | |
252 | div >>= 1; | |
253 | } | |
254 | dclc >>= sft; | |
255 | do_div(dclc, (unsigned long) div); | |
256 | ||
4d672e7a | 257 | return dclc; |
c0a31329 | 258 | } |
8b618628 | 259 | EXPORT_SYMBOL_GPL(__ktime_divns); |
c0a31329 TG |
260 | #endif /* BITS_PER_LONG >= 64 */ |
261 | ||
5a7780e7 TG |
262 | /* |
263 | * Add two ktime values and do a safety check for overflow: | |
264 | */ | |
265 | ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs) | |
266 | { | |
267 | ktime_t res = ktime_add(lhs, rhs); | |
268 | ||
269 | /* | |
270 | * We use KTIME_SEC_MAX here, the maximum timeout which we can | |
271 | * return to user space in a timespec: | |
272 | */ | |
273 | if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64) | |
274 | res = ktime_set(KTIME_SEC_MAX, 0); | |
275 | ||
276 | return res; | |
277 | } | |
278 | ||
8daa21e6 AB |
279 | EXPORT_SYMBOL_GPL(ktime_add_safe); |
280 | ||
237fc6e7 TG |
281 | #ifdef CONFIG_DEBUG_OBJECTS_TIMERS |
282 | ||
283 | static struct debug_obj_descr hrtimer_debug_descr; | |
284 | ||
99777288 SG |
285 | static void *hrtimer_debug_hint(void *addr) |
286 | { | |
287 | return ((struct hrtimer *) addr)->function; | |
288 | } | |
289 | ||
237fc6e7 TG |
290 | /* |
291 | * fixup_init is called when: | |
292 | * - an active object is initialized | |
293 | */ | |
294 | static int hrtimer_fixup_init(void *addr, enum debug_obj_state state) | |
295 | { | |
296 | struct hrtimer *timer = addr; | |
297 | ||
298 | switch (state) { | |
299 | case ODEBUG_STATE_ACTIVE: | |
300 | hrtimer_cancel(timer); | |
301 | debug_object_init(timer, &hrtimer_debug_descr); | |
302 | return 1; | |
303 | default: | |
304 | return 0; | |
305 | } | |
306 | } | |
307 | ||
308 | /* | |
309 | * fixup_activate is called when: | |
310 | * - an active object is activated | |
311 | * - an unknown object is activated (might be a statically initialized object) | |
312 | */ | |
313 | static int hrtimer_fixup_activate(void *addr, enum debug_obj_state state) | |
314 | { | |
315 | switch (state) { | |
316 | ||
317 | case ODEBUG_STATE_NOTAVAILABLE: | |
318 | WARN_ON_ONCE(1); | |
319 | return 0; | |
320 | ||
321 | case ODEBUG_STATE_ACTIVE: | |
322 | WARN_ON(1); | |
323 | ||
324 | default: | |
325 | return 0; | |
326 | } | |
327 | } | |
328 | ||
329 | /* | |
330 | * fixup_free is called when: | |
331 | * - an active object is freed | |
332 | */ | |
333 | static int hrtimer_fixup_free(void *addr, enum debug_obj_state state) | |
334 | { | |
335 | struct hrtimer *timer = addr; | |
336 | ||
337 | switch (state) { | |
338 | case ODEBUG_STATE_ACTIVE: | |
339 | hrtimer_cancel(timer); | |
340 | debug_object_free(timer, &hrtimer_debug_descr); | |
341 | return 1; | |
342 | default: | |
343 | return 0; | |
344 | } | |
345 | } | |
346 | ||
347 | static struct debug_obj_descr hrtimer_debug_descr = { | |
348 | .name = "hrtimer", | |
99777288 | 349 | .debug_hint = hrtimer_debug_hint, |
237fc6e7 TG |
350 | .fixup_init = hrtimer_fixup_init, |
351 | .fixup_activate = hrtimer_fixup_activate, | |
352 | .fixup_free = hrtimer_fixup_free, | |
353 | }; | |
354 | ||
355 | static inline void debug_hrtimer_init(struct hrtimer *timer) | |
356 | { | |
357 | debug_object_init(timer, &hrtimer_debug_descr); | |
358 | } | |
359 | ||
360 | static inline void debug_hrtimer_activate(struct hrtimer *timer) | |
361 | { | |
362 | debug_object_activate(timer, &hrtimer_debug_descr); | |
363 | } | |
364 | ||
365 | static inline void debug_hrtimer_deactivate(struct hrtimer *timer) | |
366 | { | |
367 | debug_object_deactivate(timer, &hrtimer_debug_descr); | |
368 | } | |
369 | ||
370 | static inline void debug_hrtimer_free(struct hrtimer *timer) | |
371 | { | |
372 | debug_object_free(timer, &hrtimer_debug_descr); | |
373 | } | |
374 | ||
375 | static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id, | |
376 | enum hrtimer_mode mode); | |
377 | ||
378 | void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id, | |
379 | enum hrtimer_mode mode) | |
380 | { | |
381 | debug_object_init_on_stack(timer, &hrtimer_debug_descr); | |
382 | __hrtimer_init(timer, clock_id, mode); | |
383 | } | |
2bc481cf | 384 | EXPORT_SYMBOL_GPL(hrtimer_init_on_stack); |
237fc6e7 TG |
385 | |
386 | void destroy_hrtimer_on_stack(struct hrtimer *timer) | |
387 | { | |
388 | debug_object_free(timer, &hrtimer_debug_descr); | |
389 | } | |
390 | ||
391 | #else | |
392 | static inline void debug_hrtimer_init(struct hrtimer *timer) { } | |
393 | static inline void debug_hrtimer_activate(struct hrtimer *timer) { } | |
394 | static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { } | |
395 | #endif | |
396 | ||
c6a2a177 XG |
397 | static inline void |
398 | debug_init(struct hrtimer *timer, clockid_t clockid, | |
399 | enum hrtimer_mode mode) | |
400 | { | |
401 | debug_hrtimer_init(timer); | |
402 | trace_hrtimer_init(timer, clockid, mode); | |
403 | } | |
404 | ||
405 | static inline void debug_activate(struct hrtimer *timer) | |
406 | { | |
407 | debug_hrtimer_activate(timer); | |
408 | trace_hrtimer_start(timer); | |
409 | } | |
410 | ||
411 | static inline void debug_deactivate(struct hrtimer *timer) | |
412 | { | |
413 | debug_hrtimer_deactivate(timer); | |
414 | trace_hrtimer_cancel(timer); | |
415 | } | |
416 | ||
9bc74919 | 417 | #if defined(CONFIG_NO_HZ_COMMON) || defined(CONFIG_HIGH_RES_TIMERS) |
895bdfa7 TG |
418 | static inline void hrtimer_update_next_timer(struct hrtimer_cpu_base *cpu_base, |
419 | struct hrtimer *timer) | |
420 | { | |
421 | #ifdef CONFIG_HIGH_RES_TIMERS | |
422 | cpu_base->next_timer = timer; | |
423 | #endif | |
424 | } | |
425 | ||
4ebbda52 | 426 | static ktime_t __hrtimer_get_next_event(struct hrtimer_cpu_base *cpu_base) |
9bc74919 TG |
427 | { |
428 | struct hrtimer_clock_base *base = cpu_base->clock_base; | |
429 | ktime_t expires, expires_next = { .tv64 = KTIME_MAX }; | |
34aee88a | 430 | unsigned int active = cpu_base->active_bases; |
9bc74919 | 431 | |
895bdfa7 | 432 | hrtimer_update_next_timer(cpu_base, NULL); |
34aee88a | 433 | for (; active; base++, active >>= 1) { |
9bc74919 TG |
434 | struct timerqueue_node *next; |
435 | struct hrtimer *timer; | |
436 | ||
34aee88a | 437 | if (!(active & 0x01)) |
9bc74919 TG |
438 | continue; |
439 | ||
34aee88a | 440 | next = timerqueue_getnext(&base->active); |
9bc74919 TG |
441 | timer = container_of(next, struct hrtimer, node); |
442 | expires = ktime_sub(hrtimer_get_expires(timer), base->offset); | |
895bdfa7 | 443 | if (expires.tv64 < expires_next.tv64) { |
9bc74919 | 444 | expires_next = expires; |
895bdfa7 TG |
445 | hrtimer_update_next_timer(cpu_base, timer); |
446 | } | |
9bc74919 TG |
447 | } |
448 | /* | |
449 | * clock_was_set() might have changed base->offset of any of | |
450 | * the clock bases so the result might be negative. Fix it up | |
451 | * to prevent a false positive in clockevents_program_event(). | |
452 | */ | |
453 | if (expires_next.tv64 < 0) | |
454 | expires_next.tv64 = 0; | |
455 | return expires_next; | |
456 | } | |
457 | #endif | |
458 | ||
21d6d52a TG |
459 | static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base) |
460 | { | |
461 | ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset; | |
462 | ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset; | |
463 | ktime_t *offs_tai = &base->clock_base[HRTIMER_BASE_TAI].offset; | |
464 | ||
868a3e91 TG |
465 | return ktime_get_update_offsets_now(&base->clock_was_set_seq, |
466 | offs_real, offs_boot, offs_tai); | |
21d6d52a TG |
467 | } |
468 | ||
54cdfdb4 TG |
469 | /* High resolution timer related functions */ |
470 | #ifdef CONFIG_HIGH_RES_TIMERS | |
471 | ||
472 | /* | |
473 | * High resolution timer enabled ? | |
474 | */ | |
475 | static int hrtimer_hres_enabled __read_mostly = 1; | |
398ca17f TG |
476 | unsigned int hrtimer_resolution __read_mostly = LOW_RES_NSEC; |
477 | EXPORT_SYMBOL_GPL(hrtimer_resolution); | |
54cdfdb4 TG |
478 | |
479 | /* | |
480 | * Enable / Disable high resolution mode | |
481 | */ | |
482 | static int __init setup_hrtimer_hres(char *str) | |
483 | { | |
484 | if (!strcmp(str, "off")) | |
485 | hrtimer_hres_enabled = 0; | |
486 | else if (!strcmp(str, "on")) | |
487 | hrtimer_hres_enabled = 1; | |
488 | else | |
489 | return 0; | |
490 | return 1; | |
491 | } | |
492 | ||
493 | __setup("highres=", setup_hrtimer_hres); | |
494 | ||
495 | /* | |
496 | * hrtimer_high_res_enabled - query, if the highres mode is enabled | |
497 | */ | |
498 | static inline int hrtimer_is_hres_enabled(void) | |
499 | { | |
500 | return hrtimer_hres_enabled; | |
501 | } | |
502 | ||
503 | /* | |
504 | * Is the high resolution mode active ? | |
505 | */ | |
e19ffe8b TG |
506 | static inline int __hrtimer_hres_active(struct hrtimer_cpu_base *cpu_base) |
507 | { | |
508 | return cpu_base->hres_active; | |
509 | } | |
510 | ||
54cdfdb4 TG |
511 | static inline int hrtimer_hres_active(void) |
512 | { | |
e19ffe8b | 513 | return __hrtimer_hres_active(this_cpu_ptr(&hrtimer_bases)); |
54cdfdb4 TG |
514 | } |
515 | ||
516 | /* | |
517 | * Reprogram the event source with checking both queues for the | |
518 | * next event | |
519 | * Called with interrupts disabled and base->lock held | |
520 | */ | |
7403f41f AC |
521 | static void |
522 | hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal) | |
54cdfdb4 | 523 | { |
21d6d52a TG |
524 | ktime_t expires_next; |
525 | ||
526 | if (!cpu_base->hres_active) | |
527 | return; | |
528 | ||
529 | expires_next = __hrtimer_get_next_event(cpu_base); | |
54cdfdb4 | 530 | |
7403f41f AC |
531 | if (skip_equal && expires_next.tv64 == cpu_base->expires_next.tv64) |
532 | return; | |
533 | ||
534 | cpu_base->expires_next.tv64 = expires_next.tv64; | |
535 | ||
6c6c0d5a SH |
536 | /* |
537 | * If a hang was detected in the last timer interrupt then we | |
538 | * leave the hang delay active in the hardware. We want the | |
539 | * system to make progress. That also prevents the following | |
540 | * scenario: | |
541 | * T1 expires 50ms from now | |
542 | * T2 expires 5s from now | |
543 | * | |
544 | * T1 is removed, so this code is called and would reprogram | |
545 | * the hardware to 5s from now. Any hrtimer_start after that | |
546 | * will not reprogram the hardware due to hang_detected being | |
547 | * set. So we'd effectivly block all timers until the T2 event | |
548 | * fires. | |
549 | */ | |
550 | if (cpu_base->hang_detected) | |
551 | return; | |
552 | ||
d2540875 | 553 | tick_program_event(cpu_base->expires_next, 1); |
54cdfdb4 TG |
554 | } |
555 | ||
556 | /* | |
54cdfdb4 TG |
557 | * When a timer is enqueued and expires earlier than the already enqueued |
558 | * timers, we have to check, whether it expires earlier than the timer for | |
559 | * which the clock event device was armed. | |
560 | * | |
561 | * Called with interrupts disabled and base->cpu_base.lock held | |
562 | */ | |
c6eb3f70 TG |
563 | static void hrtimer_reprogram(struct hrtimer *timer, |
564 | struct hrtimer_clock_base *base) | |
54cdfdb4 | 565 | { |
dc5df73b | 566 | struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases); |
cc584b21 | 567 | ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset); |
54cdfdb4 | 568 | |
cc584b21 | 569 | WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0); |
63070a79 | 570 | |
54cdfdb4 | 571 | /* |
c6eb3f70 TG |
572 | * If the timer is not on the current cpu, we cannot reprogram |
573 | * the other cpus clock event device. | |
54cdfdb4 | 574 | */ |
c6eb3f70 TG |
575 | if (base->cpu_base != cpu_base) |
576 | return; | |
577 | ||
578 | /* | |
579 | * If the hrtimer interrupt is running, then it will | |
580 | * reevaluate the clock bases and reprogram the clock event | |
581 | * device. The callbacks are always executed in hard interrupt | |
582 | * context so we don't need an extra check for a running | |
583 | * callback. | |
584 | */ | |
585 | if (cpu_base->in_hrtirq) | |
586 | return; | |
54cdfdb4 | 587 | |
63070a79 TG |
588 | /* |
589 | * CLOCK_REALTIME timer might be requested with an absolute | |
c6eb3f70 | 590 | * expiry time which is less than base->offset. Set it to 0. |
63070a79 TG |
591 | */ |
592 | if (expires.tv64 < 0) | |
c6eb3f70 | 593 | expires.tv64 = 0; |
63070a79 | 594 | |
41d2e494 | 595 | if (expires.tv64 >= cpu_base->expires_next.tv64) |
c6eb3f70 | 596 | return; |
9bc74919 | 597 | |
c6eb3f70 | 598 | /* Update the pointer to the next expiring timer */ |
895bdfa7 TG |
599 | cpu_base->next_timer = timer; |
600 | ||
41d2e494 TG |
601 | /* |
602 | * If a hang was detected in the last timer interrupt then we | |
603 | * do not schedule a timer which is earlier than the expiry | |
604 | * which we enforced in the hang detection. We want the system | |
605 | * to make progress. | |
606 | */ | |
607 | if (cpu_base->hang_detected) | |
c6eb3f70 | 608 | return; |
54cdfdb4 TG |
609 | |
610 | /* | |
c6eb3f70 TG |
611 | * Program the timer hardware. We enforce the expiry for |
612 | * events which are already in the past. | |
54cdfdb4 | 613 | */ |
c6eb3f70 TG |
614 | cpu_base->expires_next = expires; |
615 | tick_program_event(expires, 1); | |
54cdfdb4 TG |
616 | } |
617 | ||
54cdfdb4 TG |
618 | /* |
619 | * Initialize the high resolution related parts of cpu_base | |
620 | */ | |
621 | static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) | |
622 | { | |
623 | base->expires_next.tv64 = KTIME_MAX; | |
624 | base->hres_active = 0; | |
54cdfdb4 TG |
625 | } |
626 | ||
9ec26907 TG |
627 | /* |
628 | * Retrigger next event is called after clock was set | |
629 | * | |
630 | * Called with interrupts disabled via on_each_cpu() | |
631 | */ | |
632 | static void retrigger_next_event(void *arg) | |
633 | { | |
dc5df73b | 634 | struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases); |
9ec26907 | 635 | |
e19ffe8b | 636 | if (!base->hres_active) |
9ec26907 TG |
637 | return; |
638 | ||
9ec26907 | 639 | raw_spin_lock(&base->lock); |
5baefd6d | 640 | hrtimer_update_base(base); |
9ec26907 TG |
641 | hrtimer_force_reprogram(base, 0); |
642 | raw_spin_unlock(&base->lock); | |
643 | } | |
b12a03ce | 644 | |
54cdfdb4 TG |
645 | /* |
646 | * Switch to high resolution mode | |
647 | */ | |
f8953856 | 648 | static int hrtimer_switch_to_hres(void) |
54cdfdb4 | 649 | { |
c6eb3f70 | 650 | struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases); |
54cdfdb4 TG |
651 | |
652 | if (tick_init_highres()) { | |
820de5c3 | 653 | printk(KERN_WARNING "Could not switch to high resolution " |
c6eb3f70 | 654 | "mode on CPU %d\n", base->cpu); |
f8953856 | 655 | return 0; |
54cdfdb4 TG |
656 | } |
657 | base->hres_active = 1; | |
398ca17f | 658 | hrtimer_resolution = HIGH_RES_NSEC; |
54cdfdb4 TG |
659 | |
660 | tick_setup_sched_timer(); | |
54cdfdb4 TG |
661 | /* "Retrigger" the interrupt to get things going */ |
662 | retrigger_next_event(NULL); | |
f8953856 | 663 | return 1; |
54cdfdb4 TG |
664 | } |
665 | ||
5ec2481b TG |
666 | static void clock_was_set_work(struct work_struct *work) |
667 | { | |
668 | clock_was_set(); | |
669 | } | |
670 | ||
671 | static DECLARE_WORK(hrtimer_work, clock_was_set_work); | |
672 | ||
f55a6faa | 673 | /* |
5ec2481b TG |
674 | * Called from timekeeping and resume code to reprogramm the hrtimer |
675 | * interrupt device on all cpus. | |
f55a6faa JS |
676 | */ |
677 | void clock_was_set_delayed(void) | |
678 | { | |
5ec2481b | 679 | schedule_work(&hrtimer_work); |
f55a6faa JS |
680 | } |
681 | ||
54cdfdb4 TG |
682 | #else |
683 | ||
e19ffe8b | 684 | static inline int __hrtimer_hres_active(struct hrtimer_cpu_base *b) { return 0; } |
54cdfdb4 TG |
685 | static inline int hrtimer_hres_active(void) { return 0; } |
686 | static inline int hrtimer_is_hres_enabled(void) { return 0; } | |
f8953856 | 687 | static inline int hrtimer_switch_to_hres(void) { return 0; } |
7403f41f AC |
688 | static inline void |
689 | hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { } | |
9e1e01dd VK |
690 | static inline int hrtimer_reprogram(struct hrtimer *timer, |
691 | struct hrtimer_clock_base *base) | |
54cdfdb4 TG |
692 | { |
693 | return 0; | |
694 | } | |
54cdfdb4 | 695 | static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { } |
9ec26907 | 696 | static inline void retrigger_next_event(void *arg) { } |
54cdfdb4 TG |
697 | |
698 | #endif /* CONFIG_HIGH_RES_TIMERS */ | |
699 | ||
b12a03ce TG |
700 | /* |
701 | * Clock realtime was set | |
702 | * | |
703 | * Change the offset of the realtime clock vs. the monotonic | |
704 | * clock. | |
705 | * | |
706 | * We might have to reprogram the high resolution timer interrupt. On | |
707 | * SMP we call the architecture specific code to retrigger _all_ high | |
708 | * resolution timer interrupts. On UP we just disable interrupts and | |
709 | * call the high resolution interrupt code. | |
710 | */ | |
711 | void clock_was_set(void) | |
712 | { | |
90ff1f30 | 713 | #ifdef CONFIG_HIGH_RES_TIMERS |
b12a03ce TG |
714 | /* Retrigger the CPU local events everywhere */ |
715 | on_each_cpu(retrigger_next_event, NULL, 1); | |
9ec26907 TG |
716 | #endif |
717 | timerfd_clock_was_set(); | |
b12a03ce TG |
718 | } |
719 | ||
720 | /* | |
721 | * During resume we might have to reprogram the high resolution timer | |
7c4c3a0f DV |
722 | * interrupt on all online CPUs. However, all other CPUs will be |
723 | * stopped with IRQs interrupts disabled so the clock_was_set() call | |
5ec2481b | 724 | * must be deferred. |
b12a03ce TG |
725 | */ |
726 | void hrtimers_resume(void) | |
727 | { | |
728 | WARN_ONCE(!irqs_disabled(), | |
729 | KERN_INFO "hrtimers_resume() called with IRQs enabled!"); | |
730 | ||
5ec2481b | 731 | /* Retrigger on the local CPU */ |
b12a03ce | 732 | retrigger_next_event(NULL); |
5ec2481b TG |
733 | /* And schedule a retrigger for all others */ |
734 | clock_was_set_delayed(); | |
b12a03ce TG |
735 | } |
736 | ||
5f201907 | 737 | static inline void timer_stats_hrtimer_set_start_info(struct hrtimer *timer) |
82f67cd9 | 738 | { |
5f201907 | 739 | #ifdef CONFIG_TIMER_STATS |
82f67cd9 IM |
740 | if (timer->start_site) |
741 | return; | |
5f201907 | 742 | timer->start_site = __builtin_return_address(0); |
82f67cd9 IM |
743 | memcpy(timer->start_comm, current->comm, TASK_COMM_LEN); |
744 | timer->start_pid = current->pid; | |
5f201907 HC |
745 | #endif |
746 | } | |
747 | ||
748 | static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer *timer) | |
749 | { | |
750 | #ifdef CONFIG_TIMER_STATS | |
751 | timer->start_site = NULL; | |
752 | #endif | |
82f67cd9 | 753 | } |
5f201907 HC |
754 | |
755 | static inline void timer_stats_account_hrtimer(struct hrtimer *timer) | |
756 | { | |
757 | #ifdef CONFIG_TIMER_STATS | |
758 | if (likely(!timer_stats_active)) | |
759 | return; | |
760 | timer_stats_update_stats(timer, timer->start_pid, timer->start_site, | |
761 | timer->function, timer->start_comm, 0); | |
82f67cd9 | 762 | #endif |
5f201907 | 763 | } |
82f67cd9 | 764 | |
c0a31329 | 765 | /* |
6506f2aa | 766 | * Counterpart to lock_hrtimer_base above: |
c0a31329 TG |
767 | */ |
768 | static inline | |
769 | void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) | |
770 | { | |
ecb49d1a | 771 | raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags); |
c0a31329 TG |
772 | } |
773 | ||
774 | /** | |
775 | * hrtimer_forward - forward the timer expiry | |
c0a31329 | 776 | * @timer: hrtimer to forward |
44f21475 | 777 | * @now: forward past this time |
c0a31329 TG |
778 | * @interval: the interval to forward |
779 | * | |
780 | * Forward the timer expiry so it will expire in the future. | |
8dca6f33 | 781 | * Returns the number of overruns. |
91e5a217 TG |
782 | * |
783 | * Can be safely called from the callback function of @timer. If | |
784 | * called from other contexts @timer must neither be enqueued nor | |
785 | * running the callback and the caller needs to take care of | |
786 | * serialization. | |
787 | * | |
788 | * Note: This only updates the timer expiry value and does not requeue | |
789 | * the timer. | |
c0a31329 | 790 | */ |
4d672e7a | 791 | u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval) |
c0a31329 | 792 | { |
4d672e7a | 793 | u64 orun = 1; |
44f21475 | 794 | ktime_t delta; |
c0a31329 | 795 | |
cc584b21 | 796 | delta = ktime_sub(now, hrtimer_get_expires(timer)); |
c0a31329 TG |
797 | |
798 | if (delta.tv64 < 0) | |
799 | return 0; | |
800 | ||
5de2755c PZ |
801 | if (WARN_ON(timer->state & HRTIMER_STATE_ENQUEUED)) |
802 | return 0; | |
803 | ||
398ca17f TG |
804 | if (interval.tv64 < hrtimer_resolution) |
805 | interval.tv64 = hrtimer_resolution; | |
c9db4fa1 | 806 | |
c0a31329 | 807 | if (unlikely(delta.tv64 >= interval.tv64)) { |
df869b63 | 808 | s64 incr = ktime_to_ns(interval); |
c0a31329 TG |
809 | |
810 | orun = ktime_divns(delta, incr); | |
cc584b21 AV |
811 | hrtimer_add_expires_ns(timer, incr * orun); |
812 | if (hrtimer_get_expires_tv64(timer) > now.tv64) | |
c0a31329 TG |
813 | return orun; |
814 | /* | |
815 | * This (and the ktime_add() below) is the | |
816 | * correction for exact: | |
817 | */ | |
818 | orun++; | |
819 | } | |
cc584b21 | 820 | hrtimer_add_expires(timer, interval); |
c0a31329 TG |
821 | |
822 | return orun; | |
823 | } | |
6bdb6b62 | 824 | EXPORT_SYMBOL_GPL(hrtimer_forward); |
c0a31329 TG |
825 | |
826 | /* | |
827 | * enqueue_hrtimer - internal function to (re)start a timer | |
828 | * | |
829 | * The timer is inserted in expiry order. Insertion into the | |
830 | * red black tree is O(log(n)). Must hold the base lock. | |
a6037b61 PZ |
831 | * |
832 | * Returns 1 when the new timer is the leftmost timer in the tree. | |
c0a31329 | 833 | */ |
a6037b61 PZ |
834 | static int enqueue_hrtimer(struct hrtimer *timer, |
835 | struct hrtimer_clock_base *base) | |
c0a31329 | 836 | { |
c6a2a177 | 837 | debug_activate(timer); |
237fc6e7 | 838 | |
ab8177bc | 839 | base->cpu_base->active_bases |= 1 << base->index; |
54cdfdb4 | 840 | |
303e967f TG |
841 | /* |
842 | * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the | |
843 | * state of a possibly running callback. | |
844 | */ | |
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, |
54cdfdb4 | 862 | unsigned long newstate, int reprogram) |
c0a31329 | 863 | { |
e19ffe8b | 864 | struct hrtimer_cpu_base *cpu_base = base->cpu_base; |
895bdfa7 | 865 | unsigned int 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); |
d9f0acde | 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)) { |
8edfb036 | 895 | unsigned long 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); |
82f67cd9 | 907 | timer_stats_hrtimer_clear_start_info(timer); |
dc5df73b | 908 | reprogram = base->cpu_base == this_cpu_ptr(&hrtimer_bases); |
8edfb036 PZ |
909 | |
910 | if (!restart) { | |
911 | /* | |
912 | * We must preserve the CALLBACK state flag here, | |
913 | * otherwise we could move the timer base in | |
914 | * switch_hrtimer_base. | |
915 | */ | |
916 | state &= HRTIMER_STATE_CALLBACK; | |
917 | } | |
f13d4f97 | 918 | __remove_hrtimer(timer, base, state, reprogram); |
c0a31329 TG |
919 | return 1; |
920 | } | |
921 | return 0; | |
922 | } | |
923 | ||
58f1f803 TG |
924 | /** |
925 | * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU | |
926 | * @timer: the timer to be added | |
927 | * @tim: expiry time | |
928 | * @delta_ns: "slack" range for the timer | |
929 | * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or | |
930 | * relative (HRTIMER_MODE_REL) | |
58f1f803 | 931 | */ |
61699e13 TG |
932 | void hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, |
933 | unsigned long delta_ns, const enum hrtimer_mode mode) | |
c0a31329 | 934 | { |
3c8aa39d | 935 | struct hrtimer_clock_base *base, *new_base; |
c0a31329 | 936 | unsigned long flags; |
61699e13 | 937 | int leftmost; |
c0a31329 TG |
938 | |
939 | base = lock_hrtimer_base(timer, &flags); | |
940 | ||
941 | /* Remove an active timer from the queue: */ | |
8edfb036 | 942 | remove_hrtimer(timer, base, true); |
c0a31329 | 943 | |
597d0275 | 944 | if (mode & HRTIMER_MODE_REL) { |
84ea7fe3 | 945 | tim = ktime_add_safe(tim, base->get_time()); |
06027bdd IM |
946 | /* |
947 | * CONFIG_TIME_LOW_RES is a temporary way for architectures | |
948 | * to signal that they simply return xtime in | |
949 | * do_gettimeoffset(). In this case we want to round up by | |
950 | * resolution when starting a relative timer, to avoid short | |
951 | * timeouts. This will go away with the GTOD framework. | |
952 | */ | |
953 | #ifdef CONFIG_TIME_LOW_RES | |
398ca17f | 954 | tim = ktime_add_safe(tim, ktime_set(0, hrtimer_resolution)); |
06027bdd IM |
955 | #endif |
956 | } | |
237fc6e7 | 957 | |
da8f2e17 | 958 | hrtimer_set_expires_range_ns(timer, tim, delta_ns); |
c0a31329 | 959 | |
84ea7fe3 VK |
960 | /* Switch the timer base, if necessary: */ |
961 | new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED); | |
962 | ||
82f67cd9 IM |
963 | timer_stats_hrtimer_set_start_info(timer); |
964 | ||
a6037b61 | 965 | leftmost = enqueue_hrtimer(timer, new_base); |
61699e13 TG |
966 | if (!leftmost) |
967 | goto unlock; | |
49a2a075 VK |
968 | |
969 | if (!hrtimer_is_hres_active(timer)) { | |
970 | /* | |
971 | * Kick to reschedule the next tick to handle the new timer | |
972 | * on dynticks target. | |
973 | */ | |
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); |
c0a31329 | 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 | |
990 | * -1 when the timer is currently excuting 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 TG |
1042 | * @timer: the timer to read |
1043 | */ | |
1044 | ktime_t hrtimer_get_remaining(const struct hrtimer *timer) | |
1045 | { | |
c0a31329 TG |
1046 | unsigned long flags; |
1047 | ktime_t rem; | |
1048 | ||
b3bd3de6 | 1049 | lock_hrtimer_base(timer, &flags); |
cc584b21 | 1050 | rem = hrtimer_expires_remaining(timer); |
c0a31329 TG |
1051 | unlock_hrtimer_base(timer, &flags); |
1052 | ||
1053 | return rem; | |
1054 | } | |
8d16b764 | 1055 | EXPORT_SYMBOL_GPL(hrtimer_get_remaining); |
c0a31329 | 1056 | |
3451d024 | 1057 | #ifdef CONFIG_NO_HZ_COMMON |
69239749 TL |
1058 | /** |
1059 | * hrtimer_get_next_event - get the time until next expiry event | |
1060 | * | |
c1ad348b | 1061 | * Returns the next expiry time or KTIME_MAX if no timer is pending. |
69239749 | 1062 | */ |
c1ad348b | 1063 | u64 hrtimer_get_next_event(void) |
69239749 | 1064 | { |
dc5df73b | 1065 | struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases); |
c1ad348b | 1066 | u64 expires = KTIME_MAX; |
69239749 | 1067 | unsigned long flags; |
69239749 | 1068 | |
ecb49d1a | 1069 | raw_spin_lock_irqsave(&cpu_base->lock, flags); |
3c8aa39d | 1070 | |
e19ffe8b | 1071 | if (!__hrtimer_hres_active(cpu_base)) |
c1ad348b | 1072 | expires = __hrtimer_get_next_event(cpu_base).tv64; |
3c8aa39d | 1073 | |
ecb49d1a | 1074 | raw_spin_unlock_irqrestore(&cpu_base->lock, flags); |
3c8aa39d | 1075 | |
c1ad348b | 1076 | return expires; |
69239749 TL |
1077 | } |
1078 | #endif | |
1079 | ||
237fc6e7 TG |
1080 | static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id, |
1081 | enum hrtimer_mode mode) | |
c0a31329 | 1082 | { |
3c8aa39d | 1083 | struct hrtimer_cpu_base *cpu_base; |
e06383db | 1084 | int base; |
c0a31329 | 1085 | |
7978672c GA |
1086 | memset(timer, 0, sizeof(struct hrtimer)); |
1087 | ||
22127e93 | 1088 | cpu_base = raw_cpu_ptr(&hrtimer_bases); |
c0a31329 | 1089 | |
c9cb2e3d | 1090 | if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS) |
7978672c GA |
1091 | clock_id = CLOCK_MONOTONIC; |
1092 | ||
e06383db JS |
1093 | base = hrtimer_clockid_to_base(clock_id); |
1094 | timer->base = &cpu_base->clock_base[base]; | |
998adc3d | 1095 | timerqueue_init(&timer->node); |
82f67cd9 IM |
1096 | |
1097 | #ifdef CONFIG_TIMER_STATS | |
1098 | timer->start_site = NULL; | |
1099 | timer->start_pid = -1; | |
1100 | memset(timer->start_comm, 0, TASK_COMM_LEN); | |
1101 | #endif | |
c0a31329 | 1102 | } |
237fc6e7 TG |
1103 | |
1104 | /** | |
1105 | * hrtimer_init - initialize a timer to the given clock | |
1106 | * @timer: the timer to be initialized | |
1107 | * @clock_id: the clock to be used | |
1108 | * @mode: timer mode abs/rel | |
1109 | */ | |
1110 | void hrtimer_init(struct hrtimer *timer, clockid_t clock_id, | |
1111 | enum hrtimer_mode mode) | |
1112 | { | |
c6a2a177 | 1113 | debug_init(timer, clock_id, mode); |
237fc6e7 TG |
1114 | __hrtimer_init(timer, clock_id, mode); |
1115 | } | |
8d16b764 | 1116 | EXPORT_SYMBOL_GPL(hrtimer_init); |
c0a31329 | 1117 | |
21d6d52a TG |
1118 | static void __run_hrtimer(struct hrtimer_cpu_base *cpu_base, |
1119 | struct hrtimer_clock_base *base, | |
1120 | struct hrtimer *timer, ktime_t *now) | |
d3d74453 | 1121 | { |
d3d74453 PZ |
1122 | enum hrtimer_restart (*fn)(struct hrtimer *); |
1123 | int restart; | |
1124 | ||
ca109491 PZ |
1125 | WARN_ON(!irqs_disabled()); |
1126 | ||
c6a2a177 | 1127 | debug_deactivate(timer); |
d3d74453 PZ |
1128 | __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0); |
1129 | timer_stats_account_hrtimer(timer); | |
d3d74453 | 1130 | fn = timer->function; |
ca109491 PZ |
1131 | |
1132 | /* | |
1133 | * Because we run timers from hardirq context, there is no chance | |
1134 | * they get migrated to another cpu, therefore its safe to unlock | |
1135 | * the timer base. | |
1136 | */ | |
ecb49d1a | 1137 | raw_spin_unlock(&cpu_base->lock); |
c6a2a177 | 1138 | trace_hrtimer_expire_entry(timer, now); |
ca109491 | 1139 | restart = fn(timer); |
c6a2a177 | 1140 | trace_hrtimer_expire_exit(timer); |
ecb49d1a | 1141 | raw_spin_lock(&cpu_base->lock); |
d3d74453 PZ |
1142 | |
1143 | /* | |
e3f1d883 TG |
1144 | * Note: We clear the CALLBACK bit after enqueue_hrtimer and |
1145 | * we do not reprogramm the event hardware. Happens either in | |
1146 | * hrtimer_start_range_ns() or in hrtimer_interrupt() | |
5de2755c PZ |
1147 | * |
1148 | * Note: Because we dropped the cpu_base->lock above, | |
1149 | * hrtimer_start_range_ns() can have popped in and enqueued the timer | |
1150 | * for us already. | |
d3d74453 | 1151 | */ |
5de2755c PZ |
1152 | if (restart != HRTIMER_NORESTART && |
1153 | !(timer->state & HRTIMER_STATE_ENQUEUED)) | |
a6037b61 | 1154 | enqueue_hrtimer(timer, base); |
f13d4f97 SQ |
1155 | |
1156 | WARN_ON_ONCE(!(timer->state & HRTIMER_STATE_CALLBACK)); | |
1157 | ||
d3d74453 PZ |
1158 | timer->state &= ~HRTIMER_STATE_CALLBACK; |
1159 | } | |
1160 | ||
21d6d52a | 1161 | static void __hrtimer_run_queues(struct hrtimer_cpu_base *cpu_base, ktime_t now) |
54cdfdb4 | 1162 | { |
34aee88a TG |
1163 | struct hrtimer_clock_base *base = cpu_base->clock_base; |
1164 | unsigned int active = cpu_base->active_bases; | |
6ff7041d | 1165 | |
34aee88a | 1166 | for (; active; base++, active >>= 1) { |
998adc3d | 1167 | struct timerqueue_node *node; |
ab8177bc TG |
1168 | ktime_t basenow; |
1169 | ||
34aee88a | 1170 | if (!(active & 0x01)) |
ab8177bc | 1171 | continue; |
54cdfdb4 | 1172 | |
54cdfdb4 TG |
1173 | basenow = ktime_add(now, base->offset); |
1174 | ||
998adc3d | 1175 | while ((node = timerqueue_getnext(&base->active))) { |
54cdfdb4 TG |
1176 | struct hrtimer *timer; |
1177 | ||
998adc3d | 1178 | timer = container_of(node, struct hrtimer, node); |
54cdfdb4 | 1179 | |
654c8e0b AV |
1180 | /* |
1181 | * The immediate goal for using the softexpires is | |
1182 | * minimizing wakeups, not running timers at the | |
1183 | * earliest interrupt after their soft expiration. | |
1184 | * This allows us to avoid using a Priority Search | |
1185 | * Tree, which can answer a stabbing querry for | |
1186 | * overlapping intervals and instead use the simple | |
1187 | * BST we already have. | |
1188 | * We don't add extra wakeups by delaying timers that | |
1189 | * are right-of a not yet expired timer, because that | |
1190 | * timer will have to trigger a wakeup anyway. | |
1191 | */ | |
9bc74919 | 1192 | if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer)) |
54cdfdb4 | 1193 | break; |
54cdfdb4 | 1194 | |
21d6d52a | 1195 | __run_hrtimer(cpu_base, base, timer, &basenow); |
54cdfdb4 | 1196 | } |
54cdfdb4 | 1197 | } |
21d6d52a TG |
1198 | } |
1199 | ||
1200 | #ifdef CONFIG_HIGH_RES_TIMERS | |
1201 | ||
1202 | /* | |
1203 | * High resolution timer interrupt | |
1204 | * Called with interrupts disabled | |
1205 | */ | |
1206 | void hrtimer_interrupt(struct clock_event_device *dev) | |
1207 | { | |
1208 | struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases); | |
1209 | ktime_t expires_next, now, entry_time, delta; | |
1210 | int retries = 0; | |
1211 | ||
1212 | BUG_ON(!cpu_base->hres_active); | |
1213 | cpu_base->nr_events++; | |
1214 | dev->next_event.tv64 = KTIME_MAX; | |
1215 | ||
1216 | raw_spin_lock(&cpu_base->lock); | |
1217 | entry_time = now = hrtimer_update_base(cpu_base); | |
1218 | retry: | |
1219 | cpu_base->in_hrtirq = 1; | |
1220 | /* | |
1221 | * We set expires_next to KTIME_MAX here with cpu_base->lock | |
1222 | * held to prevent that a timer is enqueued in our queue via | |
1223 | * the migration code. This does not affect enqueueing of | |
1224 | * timers which run their callback and need to be requeued on | |
1225 | * this CPU. | |
1226 | */ | |
1227 | cpu_base->expires_next.tv64 = KTIME_MAX; | |
1228 | ||
1229 | __hrtimer_run_queues(cpu_base, now); | |
1230 | ||
9bc74919 TG |
1231 | /* Reevaluate the clock bases for the next expiry */ |
1232 | expires_next = __hrtimer_get_next_event(cpu_base); | |
6ff7041d TG |
1233 | /* |
1234 | * Store the new expiry value so the migration code can verify | |
1235 | * against it. | |
1236 | */ | |
54cdfdb4 | 1237 | cpu_base->expires_next = expires_next; |
9bc74919 | 1238 | cpu_base->in_hrtirq = 0; |
ecb49d1a | 1239 | raw_spin_unlock(&cpu_base->lock); |
54cdfdb4 TG |
1240 | |
1241 | /* Reprogramming necessary ? */ | |
d2540875 | 1242 | if (!tick_program_event(expires_next, 0)) { |
41d2e494 TG |
1243 | cpu_base->hang_detected = 0; |
1244 | return; | |
54cdfdb4 | 1245 | } |
41d2e494 TG |
1246 | |
1247 | /* | |
1248 | * The next timer was already expired due to: | |
1249 | * - tracing | |
1250 | * - long lasting callbacks | |
1251 | * - being scheduled away when running in a VM | |
1252 | * | |
1253 | * We need to prevent that we loop forever in the hrtimer | |
1254 | * interrupt routine. We give it 3 attempts to avoid | |
1255 | * overreacting on some spurious event. | |
5baefd6d JS |
1256 | * |
1257 | * Acquire base lock for updating the offsets and retrieving | |
1258 | * the current time. | |
41d2e494 | 1259 | */ |
196951e9 | 1260 | raw_spin_lock(&cpu_base->lock); |
5baefd6d | 1261 | now = hrtimer_update_base(cpu_base); |
41d2e494 TG |
1262 | cpu_base->nr_retries++; |
1263 | if (++retries < 3) | |
1264 | goto retry; | |
1265 | /* | |
1266 | * Give the system a chance to do something else than looping | |
1267 | * here. We stored the entry time, so we know exactly how long | |
1268 | * we spent here. We schedule the next event this amount of | |
1269 | * time away. | |
1270 | */ | |
1271 | cpu_base->nr_hangs++; | |
1272 | cpu_base->hang_detected = 1; | |
196951e9 | 1273 | raw_spin_unlock(&cpu_base->lock); |
41d2e494 | 1274 | delta = ktime_sub(now, entry_time); |
a6ffebce TG |
1275 | if ((unsigned int)delta.tv64 > cpu_base->max_hang_time) |
1276 | cpu_base->max_hang_time = (unsigned int) delta.tv64; | |
41d2e494 TG |
1277 | /* |
1278 | * Limit it to a sensible value as we enforce a longer | |
1279 | * delay. Give the CPU at least 100ms to catch up. | |
1280 | */ | |
1281 | if (delta.tv64 > 100 * NSEC_PER_MSEC) | |
1282 | expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC); | |
1283 | else | |
1284 | expires_next = ktime_add(now, delta); | |
1285 | tick_program_event(expires_next, 1); | |
1286 | printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n", | |
1287 | ktime_to_ns(delta)); | |
54cdfdb4 TG |
1288 | } |
1289 | ||
8bdec955 TG |
1290 | /* |
1291 | * local version of hrtimer_peek_ahead_timers() called with interrupts | |
1292 | * disabled. | |
1293 | */ | |
c6eb3f70 | 1294 | static inline void __hrtimer_peek_ahead_timers(void) |
8bdec955 TG |
1295 | { |
1296 | struct tick_device *td; | |
1297 | ||
1298 | if (!hrtimer_hres_active()) | |
1299 | return; | |
1300 | ||
22127e93 | 1301 | td = this_cpu_ptr(&tick_cpu_device); |
8bdec955 TG |
1302 | if (td && td->evtdev) |
1303 | hrtimer_interrupt(td->evtdev); | |
1304 | } | |
1305 | ||
82c5b7b5 IM |
1306 | #else /* CONFIG_HIGH_RES_TIMERS */ |
1307 | ||
1308 | static inline void __hrtimer_peek_ahead_timers(void) { } | |
1309 | ||
1310 | #endif /* !CONFIG_HIGH_RES_TIMERS */ | |
82f67cd9 | 1311 | |
d3d74453 | 1312 | /* |
c6eb3f70 | 1313 | * Called from run_local_timers in hardirq context every jiffy |
c0a31329 | 1314 | */ |
833883d9 | 1315 | void hrtimer_run_queues(void) |
c0a31329 | 1316 | { |
dc5df73b | 1317 | struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases); |
21d6d52a | 1318 | ktime_t now; |
c0a31329 | 1319 | |
e19ffe8b | 1320 | if (__hrtimer_hres_active(cpu_base)) |
3055adda DS |
1321 | return; |
1322 | ||
c6eb3f70 TG |
1323 | /* |
1324 | * This _is_ ugly: We have to check periodically, whether we | |
1325 | * can switch to highres and / or nohz mode. The clocksource | |
1326 | * switch happens with xtime_lock held. Notification from | |
1327 | * there only sets the check bit in the tick_oneshot code, | |
1328 | * otherwise we might deadlock vs. xtime_lock. | |
1329 | */ | |
1330 | if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) { | |
1331 | hrtimer_switch_to_hres(); | |
1332 | return; | |
1333 | } | |
1334 | ||
21d6d52a TG |
1335 | raw_spin_lock(&cpu_base->lock); |
1336 | now = hrtimer_update_base(cpu_base); | |
1337 | __hrtimer_run_queues(cpu_base, now); | |
1338 | raw_spin_unlock(&cpu_base->lock); | |
c0a31329 TG |
1339 | } |
1340 | ||
10c94ec1 TG |
1341 | /* |
1342 | * Sleep related functions: | |
1343 | */ | |
c9cb2e3d | 1344 | static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer) |
00362e33 TG |
1345 | { |
1346 | struct hrtimer_sleeper *t = | |
1347 | container_of(timer, struct hrtimer_sleeper, timer); | |
1348 | struct task_struct *task = t->task; | |
1349 | ||
1350 | t->task = NULL; | |
1351 | if (task) | |
1352 | wake_up_process(task); | |
1353 | ||
1354 | return HRTIMER_NORESTART; | |
1355 | } | |
1356 | ||
36c8b586 | 1357 | void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task) |
00362e33 TG |
1358 | { |
1359 | sl->timer.function = hrtimer_wakeup; | |
1360 | sl->task = task; | |
1361 | } | |
2bc481cf | 1362 | EXPORT_SYMBOL_GPL(hrtimer_init_sleeper); |
00362e33 | 1363 | |
669d7868 | 1364 | static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode) |
432569bb | 1365 | { |
669d7868 | 1366 | hrtimer_init_sleeper(t, current); |
10c94ec1 | 1367 | |
432569bb RZ |
1368 | do { |
1369 | set_current_state(TASK_INTERRUPTIBLE); | |
cc584b21 | 1370 | hrtimer_start_expires(&t->timer, mode); |
432569bb | 1371 | |
54cdfdb4 | 1372 | if (likely(t->task)) |
b0f8c44f | 1373 | freezable_schedule(); |
432569bb | 1374 | |
669d7868 | 1375 | hrtimer_cancel(&t->timer); |
c9cb2e3d | 1376 | mode = HRTIMER_MODE_ABS; |
669d7868 TG |
1377 | |
1378 | } while (t->task && !signal_pending(current)); | |
432569bb | 1379 | |
3588a085 PZ |
1380 | __set_current_state(TASK_RUNNING); |
1381 | ||
669d7868 | 1382 | return t->task == NULL; |
10c94ec1 TG |
1383 | } |
1384 | ||
080344b9 ON |
1385 | static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp) |
1386 | { | |
1387 | struct timespec rmt; | |
1388 | ktime_t rem; | |
1389 | ||
cc584b21 | 1390 | rem = hrtimer_expires_remaining(timer); |
080344b9 ON |
1391 | if (rem.tv64 <= 0) |
1392 | return 0; | |
1393 | rmt = ktime_to_timespec(rem); | |
1394 | ||
1395 | if (copy_to_user(rmtp, &rmt, sizeof(*rmtp))) | |
1396 | return -EFAULT; | |
1397 | ||
1398 | return 1; | |
1399 | } | |
1400 | ||
1711ef38 | 1401 | long __sched hrtimer_nanosleep_restart(struct restart_block *restart) |
10c94ec1 | 1402 | { |
669d7868 | 1403 | struct hrtimer_sleeper t; |
080344b9 | 1404 | struct timespec __user *rmtp; |
237fc6e7 | 1405 | int ret = 0; |
10c94ec1 | 1406 | |
ab8177bc | 1407 | hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid, |
237fc6e7 | 1408 | HRTIMER_MODE_ABS); |
cc584b21 | 1409 | hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires); |
10c94ec1 | 1410 | |
c9cb2e3d | 1411 | if (do_nanosleep(&t, HRTIMER_MODE_ABS)) |
237fc6e7 | 1412 | goto out; |
10c94ec1 | 1413 | |
029a07e0 | 1414 | rmtp = restart->nanosleep.rmtp; |
432569bb | 1415 | if (rmtp) { |
237fc6e7 | 1416 | ret = update_rmtp(&t.timer, rmtp); |
080344b9 | 1417 | if (ret <= 0) |
237fc6e7 | 1418 | goto out; |
432569bb | 1419 | } |
10c94ec1 | 1420 | |
10c94ec1 | 1421 | /* The other values in restart are already filled in */ |
237fc6e7 TG |
1422 | ret = -ERESTART_RESTARTBLOCK; |
1423 | out: | |
1424 | destroy_hrtimer_on_stack(&t.timer); | |
1425 | return ret; | |
10c94ec1 TG |
1426 | } |
1427 | ||
080344b9 | 1428 | long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp, |
10c94ec1 TG |
1429 | const enum hrtimer_mode mode, const clockid_t clockid) |
1430 | { | |
1431 | struct restart_block *restart; | |
669d7868 | 1432 | struct hrtimer_sleeper t; |
237fc6e7 | 1433 | int ret = 0; |
3bd01206 AV |
1434 | unsigned long slack; |
1435 | ||
1436 | slack = current->timer_slack_ns; | |
aab03e05 | 1437 | if (dl_task(current) || rt_task(current)) |
3bd01206 | 1438 | slack = 0; |
10c94ec1 | 1439 | |
237fc6e7 | 1440 | hrtimer_init_on_stack(&t.timer, clockid, mode); |
3bd01206 | 1441 | hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack); |
432569bb | 1442 | if (do_nanosleep(&t, mode)) |
237fc6e7 | 1443 | goto out; |
10c94ec1 | 1444 | |
7978672c | 1445 | /* Absolute timers do not update the rmtp value and restart: */ |
237fc6e7 TG |
1446 | if (mode == HRTIMER_MODE_ABS) { |
1447 | ret = -ERESTARTNOHAND; | |
1448 | goto out; | |
1449 | } | |
10c94ec1 | 1450 | |
432569bb | 1451 | if (rmtp) { |
237fc6e7 | 1452 | ret = update_rmtp(&t.timer, rmtp); |
080344b9 | 1453 | if (ret <= 0) |
237fc6e7 | 1454 | goto out; |
432569bb | 1455 | } |
10c94ec1 | 1456 | |
f56141e3 | 1457 | restart = ¤t->restart_block; |
1711ef38 | 1458 | restart->fn = hrtimer_nanosleep_restart; |
ab8177bc | 1459 | restart->nanosleep.clockid = t.timer.base->clockid; |
029a07e0 | 1460 | restart->nanosleep.rmtp = rmtp; |
cc584b21 | 1461 | restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer); |
10c94ec1 | 1462 | |
237fc6e7 TG |
1463 | ret = -ERESTART_RESTARTBLOCK; |
1464 | out: | |
1465 | destroy_hrtimer_on_stack(&t.timer); | |
1466 | return ret; | |
10c94ec1 TG |
1467 | } |
1468 | ||
58fd3aa2 HC |
1469 | SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp, |
1470 | struct timespec __user *, rmtp) | |
6ba1b912 | 1471 | { |
080344b9 | 1472 | struct timespec tu; |
6ba1b912 TG |
1473 | |
1474 | if (copy_from_user(&tu, rqtp, sizeof(tu))) | |
1475 | return -EFAULT; | |
1476 | ||
1477 | if (!timespec_valid(&tu)) | |
1478 | return -EINVAL; | |
1479 | ||
080344b9 | 1480 | return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC); |
6ba1b912 TG |
1481 | } |
1482 | ||
c0a31329 TG |
1483 | /* |
1484 | * Functions related to boot-time initialization: | |
1485 | */ | |
0db0628d | 1486 | static void init_hrtimers_cpu(int cpu) |
c0a31329 | 1487 | { |
3c8aa39d | 1488 | struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu); |
c0a31329 TG |
1489 | int i; |
1490 | ||
998adc3d | 1491 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
3c8aa39d | 1492 | cpu_base->clock_base[i].cpu_base = cpu_base; |
998adc3d JS |
1493 | timerqueue_init_head(&cpu_base->clock_base[i].active); |
1494 | } | |
3c8aa39d | 1495 | |
cddd0248 | 1496 | cpu_base->cpu = cpu; |
54cdfdb4 | 1497 | hrtimer_init_hres(cpu_base); |
c0a31329 TG |
1498 | } |
1499 | ||
1500 | #ifdef CONFIG_HOTPLUG_CPU | |
1501 | ||
ca109491 | 1502 | static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base, |
37810659 | 1503 | struct hrtimer_clock_base *new_base) |
c0a31329 TG |
1504 | { |
1505 | struct hrtimer *timer; | |
998adc3d | 1506 | struct timerqueue_node *node; |
c0a31329 | 1507 | |
998adc3d JS |
1508 | while ((node = timerqueue_getnext(&old_base->active))) { |
1509 | timer = container_of(node, struct hrtimer, node); | |
54cdfdb4 | 1510 | BUG_ON(hrtimer_callback_running(timer)); |
c6a2a177 | 1511 | debug_deactivate(timer); |
b00c1a99 TG |
1512 | |
1513 | /* | |
c04dca02 | 1514 | * Mark it as ENQUEUED not INACTIVE otherwise the |
b00c1a99 TG |
1515 | * timer could be seen as !active and just vanish away |
1516 | * under us on another CPU | |
1517 | */ | |
c04dca02 | 1518 | __remove_hrtimer(timer, old_base, HRTIMER_STATE_ENQUEUED, 0); |
c0a31329 | 1519 | timer->base = new_base; |
54cdfdb4 | 1520 | /* |
e3f1d883 TG |
1521 | * Enqueue the timers on the new cpu. This does not |
1522 | * reprogram the event device in case the timer | |
1523 | * expires before the earliest on this CPU, but we run | |
1524 | * hrtimer_interrupt after we migrated everything to | |
1525 | * sort out already expired timers and reprogram the | |
1526 | * event device. | |
54cdfdb4 | 1527 | */ |
a6037b61 | 1528 | enqueue_hrtimer(timer, new_base); |
c0a31329 TG |
1529 | } |
1530 | } | |
1531 | ||
d5fd43c4 | 1532 | static void migrate_hrtimers(int scpu) |
c0a31329 | 1533 | { |
3c8aa39d | 1534 | struct hrtimer_cpu_base *old_base, *new_base; |
731a55ba | 1535 | int i; |
c0a31329 | 1536 | |
37810659 | 1537 | BUG_ON(cpu_online(scpu)); |
37810659 | 1538 | tick_cancel_sched_timer(scpu); |
731a55ba TG |
1539 | |
1540 | local_irq_disable(); | |
1541 | old_base = &per_cpu(hrtimer_bases, scpu); | |
dc5df73b | 1542 | new_base = this_cpu_ptr(&hrtimer_bases); |
d82f0b0f ON |
1543 | /* |
1544 | * The caller is globally serialized and nobody else | |
1545 | * takes two locks at once, deadlock is not possible. | |
1546 | */ | |
ecb49d1a TG |
1547 | raw_spin_lock(&new_base->lock); |
1548 | raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING); | |
c0a31329 | 1549 | |
3c8aa39d | 1550 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
ca109491 | 1551 | migrate_hrtimer_list(&old_base->clock_base[i], |
37810659 | 1552 | &new_base->clock_base[i]); |
c0a31329 TG |
1553 | } |
1554 | ||
ecb49d1a TG |
1555 | raw_spin_unlock(&old_base->lock); |
1556 | raw_spin_unlock(&new_base->lock); | |
37810659 | 1557 | |
731a55ba TG |
1558 | /* Check, if we got expired work to do */ |
1559 | __hrtimer_peek_ahead_timers(); | |
1560 | local_irq_enable(); | |
c0a31329 | 1561 | } |
37810659 | 1562 | |
c0a31329 TG |
1563 | #endif /* CONFIG_HOTPLUG_CPU */ |
1564 | ||
0db0628d | 1565 | static int hrtimer_cpu_notify(struct notifier_block *self, |
c0a31329 TG |
1566 | unsigned long action, void *hcpu) |
1567 | { | |
b2e3c0ad | 1568 | int scpu = (long)hcpu; |
c0a31329 TG |
1569 | |
1570 | switch (action) { | |
1571 | ||
1572 | case CPU_UP_PREPARE: | |
8bb78442 | 1573 | case CPU_UP_PREPARE_FROZEN: |
37810659 | 1574 | init_hrtimers_cpu(scpu); |
c0a31329 TG |
1575 | break; |
1576 | ||
1577 | #ifdef CONFIG_HOTPLUG_CPU | |
1578 | case CPU_DEAD: | |
8bb78442 | 1579 | case CPU_DEAD_FROZEN: |
d5fd43c4 | 1580 | migrate_hrtimers(scpu); |
c0a31329 TG |
1581 | break; |
1582 | #endif | |
1583 | ||
1584 | default: | |
1585 | break; | |
1586 | } | |
1587 | ||
1588 | return NOTIFY_OK; | |
1589 | } | |
1590 | ||
0db0628d | 1591 | static struct notifier_block hrtimers_nb = { |
c0a31329 TG |
1592 | .notifier_call = hrtimer_cpu_notify, |
1593 | }; | |
1594 | ||
1595 | void __init hrtimers_init(void) | |
1596 | { | |
1597 | hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE, | |
1598 | (void *)(long)smp_processor_id()); | |
1599 | register_cpu_notifier(&hrtimers_nb); | |
1600 | } | |
1601 | ||
7bb67439 | 1602 | /** |
351b3f7a | 1603 | * schedule_hrtimeout_range_clock - sleep until timeout |
7bb67439 | 1604 | * @expires: timeout value (ktime_t) |
654c8e0b | 1605 | * @delta: slack in expires timeout (ktime_t) |
7bb67439 | 1606 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL |
351b3f7a | 1607 | * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME |
7bb67439 | 1608 | */ |
351b3f7a CE |
1609 | int __sched |
1610 | schedule_hrtimeout_range_clock(ktime_t *expires, unsigned long delta, | |
1611 | const enum hrtimer_mode mode, int clock) | |
7bb67439 AV |
1612 | { |
1613 | struct hrtimer_sleeper t; | |
1614 | ||
1615 | /* | |
1616 | * Optimize when a zero timeout value is given. It does not | |
1617 | * matter whether this is an absolute or a relative time. | |
1618 | */ | |
1619 | if (expires && !expires->tv64) { | |
1620 | __set_current_state(TASK_RUNNING); | |
1621 | return 0; | |
1622 | } | |
1623 | ||
1624 | /* | |
43b21013 | 1625 | * A NULL parameter means "infinite" |
7bb67439 AV |
1626 | */ |
1627 | if (!expires) { | |
1628 | schedule(); | |
7bb67439 AV |
1629 | return -EINTR; |
1630 | } | |
1631 | ||
351b3f7a | 1632 | hrtimer_init_on_stack(&t.timer, clock, mode); |
654c8e0b | 1633 | hrtimer_set_expires_range_ns(&t.timer, *expires, delta); |
7bb67439 AV |
1634 | |
1635 | hrtimer_init_sleeper(&t, current); | |
1636 | ||
cc584b21 | 1637 | hrtimer_start_expires(&t.timer, mode); |
7bb67439 AV |
1638 | |
1639 | if (likely(t.task)) | |
1640 | schedule(); | |
1641 | ||
1642 | hrtimer_cancel(&t.timer); | |
1643 | destroy_hrtimer_on_stack(&t.timer); | |
1644 | ||
1645 | __set_current_state(TASK_RUNNING); | |
1646 | ||
1647 | return !t.task ? 0 : -EINTR; | |
1648 | } | |
351b3f7a CE |
1649 | |
1650 | /** | |
1651 | * schedule_hrtimeout_range - sleep until timeout | |
1652 | * @expires: timeout value (ktime_t) | |
1653 | * @delta: slack in expires timeout (ktime_t) | |
1654 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL | |
1655 | * | |
1656 | * Make the current task sleep until the given expiry time has | |
1657 | * elapsed. The routine will return immediately unless | |
1658 | * the current task state has been set (see set_current_state()). | |
1659 | * | |
1660 | * The @delta argument gives the kernel the freedom to schedule the | |
1661 | * actual wakeup to a time that is both power and performance friendly. | |
1662 | * The kernel give the normal best effort behavior for "@expires+@delta", | |
1663 | * but may decide to fire the timer earlier, but no earlier than @expires. | |
1664 | * | |
1665 | * You can set the task state as follows - | |
1666 | * | |
1667 | * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to | |
1668 | * pass before the routine returns. | |
1669 | * | |
1670 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
1671 | * delivered to the current task. | |
1672 | * | |
1673 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1674 | * routine returns. | |
1675 | * | |
1676 | * Returns 0 when the timer has expired otherwise -EINTR | |
1677 | */ | |
1678 | int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta, | |
1679 | const enum hrtimer_mode mode) | |
1680 | { | |
1681 | return schedule_hrtimeout_range_clock(expires, delta, mode, | |
1682 | CLOCK_MONOTONIC); | |
1683 | } | |
654c8e0b AV |
1684 | EXPORT_SYMBOL_GPL(schedule_hrtimeout_range); |
1685 | ||
1686 | /** | |
1687 | * schedule_hrtimeout - sleep until timeout | |
1688 | * @expires: timeout value (ktime_t) | |
1689 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL | |
1690 | * | |
1691 | * Make the current task sleep until the given expiry time has | |
1692 | * elapsed. The routine will return immediately unless | |
1693 | * the current task state has been set (see set_current_state()). | |
1694 | * | |
1695 | * You can set the task state as follows - | |
1696 | * | |
1697 | * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to | |
1698 | * pass before the routine returns. | |
1699 | * | |
1700 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
1701 | * delivered to the current task. | |
1702 | * | |
1703 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1704 | * routine returns. | |
1705 | * | |
1706 | * Returns 0 when the timer has expired otherwise -EINTR | |
1707 | */ | |
1708 | int __sched schedule_hrtimeout(ktime_t *expires, | |
1709 | const enum hrtimer_mode mode) | |
1710 | { | |
1711 | return schedule_hrtimeout_range(expires, 0, mode); | |
1712 | } | |
7bb67439 | 1713 | EXPORT_SYMBOL_GPL(schedule_hrtimeout); |