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 | ||
54cdfdb4 TG |
553 | if (cpu_base->expires_next.tv64 != KTIME_MAX) |
554 | tick_program_event(cpu_base->expires_next, 1); | |
555 | } | |
556 | ||
557 | /* | |
54cdfdb4 TG |
558 | * When a timer is enqueued and expires earlier than the already enqueued |
559 | * timers, we have to check, whether it expires earlier than the timer for | |
560 | * which the clock event device was armed. | |
561 | * | |
562 | * Called with interrupts disabled and base->cpu_base.lock held | |
563 | */ | |
c6eb3f70 TG |
564 | static void hrtimer_reprogram(struct hrtimer *timer, |
565 | struct hrtimer_clock_base *base) | |
54cdfdb4 | 566 | { |
dc5df73b | 567 | struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases); |
cc584b21 | 568 | ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset); |
54cdfdb4 | 569 | |
cc584b21 | 570 | WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0); |
63070a79 | 571 | |
54cdfdb4 | 572 | /* |
c6eb3f70 TG |
573 | * If the timer is not on the current cpu, we cannot reprogram |
574 | * the other cpus clock event device. | |
54cdfdb4 | 575 | */ |
c6eb3f70 TG |
576 | if (base->cpu_base != cpu_base) |
577 | return; | |
578 | ||
579 | /* | |
580 | * If the hrtimer interrupt is running, then it will | |
581 | * reevaluate the clock bases and reprogram the clock event | |
582 | * device. The callbacks are always executed in hard interrupt | |
583 | * context so we don't need an extra check for a running | |
584 | * callback. | |
585 | */ | |
586 | if (cpu_base->in_hrtirq) | |
587 | return; | |
54cdfdb4 | 588 | |
63070a79 TG |
589 | /* |
590 | * CLOCK_REALTIME timer might be requested with an absolute | |
c6eb3f70 | 591 | * expiry time which is less than base->offset. Set it to 0. |
63070a79 TG |
592 | */ |
593 | if (expires.tv64 < 0) | |
c6eb3f70 | 594 | expires.tv64 = 0; |
63070a79 | 595 | |
41d2e494 | 596 | if (expires.tv64 >= cpu_base->expires_next.tv64) |
c6eb3f70 | 597 | return; |
9bc74919 | 598 | |
c6eb3f70 | 599 | /* Update the pointer to the next expiring timer */ |
895bdfa7 TG |
600 | cpu_base->next_timer = timer; |
601 | ||
41d2e494 TG |
602 | /* |
603 | * If a hang was detected in the last timer interrupt then we | |
604 | * do not schedule a timer which is earlier than the expiry | |
605 | * which we enforced in the hang detection. We want the system | |
606 | * to make progress. | |
607 | */ | |
608 | if (cpu_base->hang_detected) | |
c6eb3f70 | 609 | return; |
54cdfdb4 TG |
610 | |
611 | /* | |
c6eb3f70 TG |
612 | * Program the timer hardware. We enforce the expiry for |
613 | * events which are already in the past. | |
54cdfdb4 | 614 | */ |
c6eb3f70 TG |
615 | cpu_base->expires_next = expires; |
616 | tick_program_event(expires, 1); | |
54cdfdb4 TG |
617 | } |
618 | ||
54cdfdb4 TG |
619 | /* |
620 | * Initialize the high resolution related parts of cpu_base | |
621 | */ | |
622 | static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) | |
623 | { | |
624 | base->expires_next.tv64 = KTIME_MAX; | |
625 | base->hres_active = 0; | |
54cdfdb4 TG |
626 | } |
627 | ||
9ec26907 TG |
628 | /* |
629 | * Retrigger next event is called after clock was set | |
630 | * | |
631 | * Called with interrupts disabled via on_each_cpu() | |
632 | */ | |
633 | static void retrigger_next_event(void *arg) | |
634 | { | |
dc5df73b | 635 | struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases); |
9ec26907 | 636 | |
e19ffe8b | 637 | if (!base->hres_active) |
9ec26907 TG |
638 | return; |
639 | ||
9ec26907 | 640 | raw_spin_lock(&base->lock); |
5baefd6d | 641 | hrtimer_update_base(base); |
9ec26907 TG |
642 | hrtimer_force_reprogram(base, 0); |
643 | raw_spin_unlock(&base->lock); | |
644 | } | |
b12a03ce | 645 | |
54cdfdb4 TG |
646 | /* |
647 | * Switch to high resolution mode | |
648 | */ | |
f8953856 | 649 | static int hrtimer_switch_to_hres(void) |
54cdfdb4 | 650 | { |
c6eb3f70 | 651 | struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases); |
54cdfdb4 TG |
652 | |
653 | if (tick_init_highres()) { | |
820de5c3 | 654 | printk(KERN_WARNING "Could not switch to high resolution " |
c6eb3f70 | 655 | "mode on CPU %d\n", base->cpu); |
f8953856 | 656 | return 0; |
54cdfdb4 TG |
657 | } |
658 | base->hres_active = 1; | |
398ca17f | 659 | hrtimer_resolution = HIGH_RES_NSEC; |
54cdfdb4 TG |
660 | |
661 | tick_setup_sched_timer(); | |
54cdfdb4 TG |
662 | /* "Retrigger" the interrupt to get things going */ |
663 | retrigger_next_event(NULL); | |
f8953856 | 664 | return 1; |
54cdfdb4 TG |
665 | } |
666 | ||
5ec2481b TG |
667 | static void clock_was_set_work(struct work_struct *work) |
668 | { | |
669 | clock_was_set(); | |
670 | } | |
671 | ||
672 | static DECLARE_WORK(hrtimer_work, clock_was_set_work); | |
673 | ||
f55a6faa | 674 | /* |
5ec2481b TG |
675 | * Called from timekeeping and resume code to reprogramm the hrtimer |
676 | * interrupt device on all cpus. | |
f55a6faa JS |
677 | */ |
678 | void clock_was_set_delayed(void) | |
679 | { | |
5ec2481b | 680 | schedule_work(&hrtimer_work); |
f55a6faa JS |
681 | } |
682 | ||
54cdfdb4 TG |
683 | #else |
684 | ||
e19ffe8b | 685 | static inline int __hrtimer_hres_active(struct hrtimer_cpu_base *b) { return 0; } |
54cdfdb4 TG |
686 | static inline int hrtimer_hres_active(void) { return 0; } |
687 | static inline int hrtimer_is_hres_enabled(void) { return 0; } | |
f8953856 | 688 | static inline int hrtimer_switch_to_hres(void) { return 0; } |
7403f41f AC |
689 | static inline void |
690 | hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { } | |
9e1e01dd VK |
691 | static inline int hrtimer_reprogram(struct hrtimer *timer, |
692 | struct hrtimer_clock_base *base) | |
54cdfdb4 TG |
693 | { |
694 | return 0; | |
695 | } | |
54cdfdb4 | 696 | static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { } |
9ec26907 | 697 | static inline void retrigger_next_event(void *arg) { } |
54cdfdb4 TG |
698 | |
699 | #endif /* CONFIG_HIGH_RES_TIMERS */ | |
700 | ||
b12a03ce TG |
701 | /* |
702 | * Clock realtime was set | |
703 | * | |
704 | * Change the offset of the realtime clock vs. the monotonic | |
705 | * clock. | |
706 | * | |
707 | * We might have to reprogram the high resolution timer interrupt. On | |
708 | * SMP we call the architecture specific code to retrigger _all_ high | |
709 | * resolution timer interrupts. On UP we just disable interrupts and | |
710 | * call the high resolution interrupt code. | |
711 | */ | |
712 | void clock_was_set(void) | |
713 | { | |
90ff1f30 | 714 | #ifdef CONFIG_HIGH_RES_TIMERS |
b12a03ce TG |
715 | /* Retrigger the CPU local events everywhere */ |
716 | on_each_cpu(retrigger_next_event, NULL, 1); | |
9ec26907 TG |
717 | #endif |
718 | timerfd_clock_was_set(); | |
b12a03ce TG |
719 | } |
720 | ||
721 | /* | |
722 | * During resume we might have to reprogram the high resolution timer | |
7c4c3a0f DV |
723 | * interrupt on all online CPUs. However, all other CPUs will be |
724 | * stopped with IRQs interrupts disabled so the clock_was_set() call | |
5ec2481b | 725 | * must be deferred. |
b12a03ce TG |
726 | */ |
727 | void hrtimers_resume(void) | |
728 | { | |
729 | WARN_ONCE(!irqs_disabled(), | |
730 | KERN_INFO "hrtimers_resume() called with IRQs enabled!"); | |
731 | ||
5ec2481b | 732 | /* Retrigger on the local CPU */ |
b12a03ce | 733 | retrigger_next_event(NULL); |
5ec2481b TG |
734 | /* And schedule a retrigger for all others */ |
735 | clock_was_set_delayed(); | |
b12a03ce TG |
736 | } |
737 | ||
5f201907 | 738 | static inline void timer_stats_hrtimer_set_start_info(struct hrtimer *timer) |
82f67cd9 | 739 | { |
5f201907 | 740 | #ifdef CONFIG_TIMER_STATS |
82f67cd9 IM |
741 | if (timer->start_site) |
742 | return; | |
5f201907 | 743 | timer->start_site = __builtin_return_address(0); |
82f67cd9 IM |
744 | memcpy(timer->start_comm, current->comm, TASK_COMM_LEN); |
745 | timer->start_pid = current->pid; | |
5f201907 HC |
746 | #endif |
747 | } | |
748 | ||
749 | static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer *timer) | |
750 | { | |
751 | #ifdef CONFIG_TIMER_STATS | |
752 | timer->start_site = NULL; | |
753 | #endif | |
82f67cd9 | 754 | } |
5f201907 HC |
755 | |
756 | static inline void timer_stats_account_hrtimer(struct hrtimer *timer) | |
757 | { | |
758 | #ifdef CONFIG_TIMER_STATS | |
759 | if (likely(!timer_stats_active)) | |
760 | return; | |
761 | timer_stats_update_stats(timer, timer->start_pid, timer->start_site, | |
762 | timer->function, timer->start_comm, 0); | |
82f67cd9 | 763 | #endif |
5f201907 | 764 | } |
82f67cd9 | 765 | |
c0a31329 | 766 | /* |
6506f2aa | 767 | * Counterpart to lock_hrtimer_base above: |
c0a31329 TG |
768 | */ |
769 | static inline | |
770 | void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) | |
771 | { | |
ecb49d1a | 772 | raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags); |
c0a31329 TG |
773 | } |
774 | ||
775 | /** | |
776 | * hrtimer_forward - forward the timer expiry | |
c0a31329 | 777 | * @timer: hrtimer to forward |
44f21475 | 778 | * @now: forward past this time |
c0a31329 TG |
779 | * @interval: the interval to forward |
780 | * | |
781 | * Forward the timer expiry so it will expire in the future. | |
8dca6f33 | 782 | * Returns the number of overruns. |
91e5a217 TG |
783 | * |
784 | * Can be safely called from the callback function of @timer. If | |
785 | * called from other contexts @timer must neither be enqueued nor | |
786 | * running the callback and the caller needs to take care of | |
787 | * serialization. | |
788 | * | |
789 | * Note: This only updates the timer expiry value and does not requeue | |
790 | * the timer. | |
c0a31329 | 791 | */ |
4d672e7a | 792 | u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval) |
c0a31329 | 793 | { |
4d672e7a | 794 | u64 orun = 1; |
44f21475 | 795 | ktime_t delta; |
c0a31329 | 796 | |
cc584b21 | 797 | delta = ktime_sub(now, hrtimer_get_expires(timer)); |
c0a31329 TG |
798 | |
799 | if (delta.tv64 < 0) | |
800 | return 0; | |
801 | ||
5de2755c PZ |
802 | if (WARN_ON(timer->state & HRTIMER_STATE_ENQUEUED)) |
803 | return 0; | |
804 | ||
398ca17f TG |
805 | if (interval.tv64 < hrtimer_resolution) |
806 | interval.tv64 = hrtimer_resolution; | |
c9db4fa1 | 807 | |
c0a31329 | 808 | if (unlikely(delta.tv64 >= interval.tv64)) { |
df869b63 | 809 | s64 incr = ktime_to_ns(interval); |
c0a31329 TG |
810 | |
811 | orun = ktime_divns(delta, incr); | |
cc584b21 AV |
812 | hrtimer_add_expires_ns(timer, incr * orun); |
813 | if (hrtimer_get_expires_tv64(timer) > now.tv64) | |
c0a31329 TG |
814 | return orun; |
815 | /* | |
816 | * This (and the ktime_add() below) is the | |
817 | * correction for exact: | |
818 | */ | |
819 | orun++; | |
820 | } | |
cc584b21 | 821 | hrtimer_add_expires(timer, interval); |
c0a31329 TG |
822 | |
823 | return orun; | |
824 | } | |
6bdb6b62 | 825 | EXPORT_SYMBOL_GPL(hrtimer_forward); |
c0a31329 TG |
826 | |
827 | /* | |
828 | * enqueue_hrtimer - internal function to (re)start a timer | |
829 | * | |
830 | * The timer is inserted in expiry order. Insertion into the | |
831 | * red black tree is O(log(n)). Must hold the base lock. | |
a6037b61 PZ |
832 | * |
833 | * Returns 1 when the new timer is the leftmost timer in the tree. | |
c0a31329 | 834 | */ |
a6037b61 PZ |
835 | static int enqueue_hrtimer(struct hrtimer *timer, |
836 | struct hrtimer_clock_base *base) | |
c0a31329 | 837 | { |
c6a2a177 | 838 | debug_activate(timer); |
237fc6e7 | 839 | |
ab8177bc | 840 | base->cpu_base->active_bases |= 1 << base->index; |
54cdfdb4 | 841 | |
303e967f TG |
842 | /* |
843 | * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the | |
844 | * state of a possibly running callback. | |
845 | */ | |
846 | timer->state |= HRTIMER_STATE_ENQUEUED; | |
a6037b61 | 847 | |
b97f44c9 | 848 | return timerqueue_add(&base->active, &timer->node); |
288867ec | 849 | } |
c0a31329 TG |
850 | |
851 | /* | |
852 | * __remove_hrtimer - internal function to remove a timer | |
853 | * | |
854 | * Caller must hold the base lock. | |
54cdfdb4 TG |
855 | * |
856 | * High resolution timer mode reprograms the clock event device when the | |
857 | * timer is the one which expires next. The caller can disable this by setting | |
858 | * reprogram to zero. This is useful, when the context does a reprogramming | |
859 | * anyway (e.g. timer interrupt) | |
c0a31329 | 860 | */ |
3c8aa39d | 861 | static void __remove_hrtimer(struct hrtimer *timer, |
303e967f | 862 | struct hrtimer_clock_base *base, |
54cdfdb4 | 863 | unsigned long newstate, int reprogram) |
c0a31329 | 864 | { |
e19ffe8b | 865 | struct hrtimer_cpu_base *cpu_base = base->cpu_base; |
895bdfa7 | 866 | unsigned int state = timer->state; |
e19ffe8b | 867 | |
895bdfa7 TG |
868 | timer->state = newstate; |
869 | if (!(state & HRTIMER_STATE_ENQUEUED)) | |
870 | return; | |
7403f41f | 871 | |
b97f44c9 | 872 | if (!timerqueue_del(&base->active, &timer->node)) |
e19ffe8b | 873 | cpu_base->active_bases &= ~(1 << base->index); |
d9f0acde | 874 | |
7403f41f | 875 | #ifdef CONFIG_HIGH_RES_TIMERS |
895bdfa7 TG |
876 | /* |
877 | * Note: If reprogram is false we do not update | |
878 | * cpu_base->next_timer. This happens when we remove the first | |
879 | * timer on a remote cpu. No harm as we never dereference | |
880 | * cpu_base->next_timer. So the worst thing what can happen is | |
881 | * an superflous call to hrtimer_force_reprogram() on the | |
882 | * remote cpu later on if the same timer gets enqueued again. | |
883 | */ | |
884 | if (reprogram && timer == cpu_base->next_timer) | |
885 | hrtimer_force_reprogram(cpu_base, 1); | |
7403f41f | 886 | #endif |
c0a31329 TG |
887 | } |
888 | ||
889 | /* | |
890 | * remove hrtimer, called with base lock held | |
891 | */ | |
892 | static inline int | |
3c8aa39d | 893 | remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base) |
c0a31329 | 894 | { |
303e967f | 895 | if (hrtimer_is_queued(timer)) { |
f13d4f97 | 896 | unsigned long state; |
54cdfdb4 TG |
897 | int reprogram; |
898 | ||
899 | /* | |
900 | * Remove the timer and force reprogramming when high | |
901 | * resolution mode is active and the timer is on the current | |
902 | * CPU. If we remove a timer on another CPU, reprogramming is | |
903 | * skipped. The interrupt event on this CPU is fired and | |
904 | * reprogramming happens in the interrupt handler. This is a | |
905 | * rare case and less expensive than a smp call. | |
906 | */ | |
c6a2a177 | 907 | debug_deactivate(timer); |
82f67cd9 | 908 | timer_stats_hrtimer_clear_start_info(timer); |
dc5df73b | 909 | reprogram = base->cpu_base == this_cpu_ptr(&hrtimer_bases); |
f13d4f97 SQ |
910 | /* |
911 | * We must preserve the CALLBACK state flag here, | |
912 | * otherwise we could move the timer base in | |
913 | * switch_hrtimer_base. | |
914 | */ | |
915 | state = timer->state & HRTIMER_STATE_CALLBACK; | |
916 | __remove_hrtimer(timer, base, state, reprogram); | |
c0a31329 TG |
917 | return 1; |
918 | } | |
919 | return 0; | |
920 | } | |
921 | ||
58f1f803 TG |
922 | /** |
923 | * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU | |
924 | * @timer: the timer to be added | |
925 | * @tim: expiry time | |
926 | * @delta_ns: "slack" range for the timer | |
927 | * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or | |
928 | * relative (HRTIMER_MODE_REL) | |
58f1f803 | 929 | */ |
61699e13 TG |
930 | void hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, |
931 | unsigned long delta_ns, const enum hrtimer_mode mode) | |
c0a31329 | 932 | { |
3c8aa39d | 933 | struct hrtimer_clock_base *base, *new_base; |
c0a31329 | 934 | unsigned long flags; |
61699e13 | 935 | int leftmost; |
c0a31329 TG |
936 | |
937 | base = lock_hrtimer_base(timer, &flags); | |
938 | ||
939 | /* Remove an active timer from the queue: */ | |
61699e13 | 940 | remove_hrtimer(timer, base); |
c0a31329 | 941 | |
597d0275 | 942 | if (mode & HRTIMER_MODE_REL) { |
84ea7fe3 | 943 | tim = ktime_add_safe(tim, base->get_time()); |
06027bdd IM |
944 | /* |
945 | * CONFIG_TIME_LOW_RES is a temporary way for architectures | |
946 | * to signal that they simply return xtime in | |
947 | * do_gettimeoffset(). In this case we want to round up by | |
948 | * resolution when starting a relative timer, to avoid short | |
949 | * timeouts. This will go away with the GTOD framework. | |
950 | */ | |
951 | #ifdef CONFIG_TIME_LOW_RES | |
398ca17f | 952 | tim = ktime_add_safe(tim, ktime_set(0, hrtimer_resolution)); |
06027bdd IM |
953 | #endif |
954 | } | |
237fc6e7 | 955 | |
da8f2e17 | 956 | hrtimer_set_expires_range_ns(timer, tim, delta_ns); |
c0a31329 | 957 | |
84ea7fe3 VK |
958 | /* Switch the timer base, if necessary: */ |
959 | new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED); | |
960 | ||
82f67cd9 IM |
961 | timer_stats_hrtimer_set_start_info(timer); |
962 | ||
a6037b61 | 963 | leftmost = enqueue_hrtimer(timer, new_base); |
61699e13 TG |
964 | if (!leftmost) |
965 | goto unlock; | |
49a2a075 VK |
966 | |
967 | if (!hrtimer_is_hres_active(timer)) { | |
968 | /* | |
969 | * Kick to reschedule the next tick to handle the new timer | |
970 | * on dynticks target. | |
971 | */ | |
972 | wake_up_nohz_cpu(new_base->cpu_base->cpu); | |
c6eb3f70 TG |
973 | } else { |
974 | hrtimer_reprogram(timer, new_base); | |
b22affe0 | 975 | } |
61699e13 | 976 | unlock: |
c0a31329 | 977 | unlock_hrtimer_base(timer, &flags); |
c0a31329 | 978 | } |
da8f2e17 AV |
979 | EXPORT_SYMBOL_GPL(hrtimer_start_range_ns); |
980 | ||
c0a31329 TG |
981 | /** |
982 | * hrtimer_try_to_cancel - try to deactivate a timer | |
c0a31329 TG |
983 | * @timer: hrtimer to stop |
984 | * | |
985 | * Returns: | |
986 | * 0 when the timer was not active | |
987 | * 1 when the timer was active | |
988 | * -1 when the timer is currently excuting the callback function and | |
fa9799e3 | 989 | * cannot be stopped |
c0a31329 TG |
990 | */ |
991 | int hrtimer_try_to_cancel(struct hrtimer *timer) | |
992 | { | |
3c8aa39d | 993 | struct hrtimer_clock_base *base; |
c0a31329 TG |
994 | unsigned long flags; |
995 | int ret = -1; | |
996 | ||
19d9f422 TG |
997 | /* |
998 | * Check lockless first. If the timer is not active (neither | |
999 | * enqueued nor running the callback, nothing to do here. The | |
1000 | * base lock does not serialize against a concurrent enqueue, | |
1001 | * so we can avoid taking it. | |
1002 | */ | |
1003 | if (!hrtimer_active(timer)) | |
1004 | return 0; | |
1005 | ||
c0a31329 TG |
1006 | base = lock_hrtimer_base(timer, &flags); |
1007 | ||
303e967f | 1008 | if (!hrtimer_callback_running(timer)) |
c0a31329 TG |
1009 | ret = remove_hrtimer(timer, base); |
1010 | ||
1011 | unlock_hrtimer_base(timer, &flags); | |
1012 | ||
1013 | return ret; | |
1014 | ||
1015 | } | |
8d16b764 | 1016 | EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel); |
c0a31329 TG |
1017 | |
1018 | /** | |
1019 | * hrtimer_cancel - cancel a timer and wait for the handler to finish. | |
c0a31329 TG |
1020 | * @timer: the timer to be cancelled |
1021 | * | |
1022 | * Returns: | |
1023 | * 0 when the timer was not active | |
1024 | * 1 when the timer was active | |
1025 | */ | |
1026 | int hrtimer_cancel(struct hrtimer *timer) | |
1027 | { | |
1028 | for (;;) { | |
1029 | int ret = hrtimer_try_to_cancel(timer); | |
1030 | ||
1031 | if (ret >= 0) | |
1032 | return ret; | |
5ef37b19 | 1033 | cpu_relax(); |
c0a31329 TG |
1034 | } |
1035 | } | |
8d16b764 | 1036 | EXPORT_SYMBOL_GPL(hrtimer_cancel); |
c0a31329 TG |
1037 | |
1038 | /** | |
1039 | * hrtimer_get_remaining - get remaining time for the timer | |
c0a31329 TG |
1040 | * @timer: the timer to read |
1041 | */ | |
1042 | ktime_t hrtimer_get_remaining(const struct hrtimer *timer) | |
1043 | { | |
c0a31329 TG |
1044 | unsigned long flags; |
1045 | ktime_t rem; | |
1046 | ||
b3bd3de6 | 1047 | lock_hrtimer_base(timer, &flags); |
cc584b21 | 1048 | rem = hrtimer_expires_remaining(timer); |
c0a31329 TG |
1049 | unlock_hrtimer_base(timer, &flags); |
1050 | ||
1051 | return rem; | |
1052 | } | |
8d16b764 | 1053 | EXPORT_SYMBOL_GPL(hrtimer_get_remaining); |
c0a31329 | 1054 | |
3451d024 | 1055 | #ifdef CONFIG_NO_HZ_COMMON |
69239749 TL |
1056 | /** |
1057 | * hrtimer_get_next_event - get the time until next expiry event | |
1058 | * | |
c1ad348b | 1059 | * Returns the next expiry time or KTIME_MAX if no timer is pending. |
69239749 | 1060 | */ |
c1ad348b | 1061 | u64 hrtimer_get_next_event(void) |
69239749 | 1062 | { |
dc5df73b | 1063 | struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases); |
c1ad348b | 1064 | u64 expires = KTIME_MAX; |
69239749 | 1065 | unsigned long flags; |
69239749 | 1066 | |
ecb49d1a | 1067 | raw_spin_lock_irqsave(&cpu_base->lock, flags); |
3c8aa39d | 1068 | |
e19ffe8b | 1069 | if (!__hrtimer_hres_active(cpu_base)) |
c1ad348b | 1070 | expires = __hrtimer_get_next_event(cpu_base).tv64; |
3c8aa39d | 1071 | |
ecb49d1a | 1072 | raw_spin_unlock_irqrestore(&cpu_base->lock, flags); |
3c8aa39d | 1073 | |
c1ad348b | 1074 | return expires; |
69239749 TL |
1075 | } |
1076 | #endif | |
1077 | ||
237fc6e7 TG |
1078 | static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id, |
1079 | enum hrtimer_mode mode) | |
c0a31329 | 1080 | { |
3c8aa39d | 1081 | struct hrtimer_cpu_base *cpu_base; |
e06383db | 1082 | int base; |
c0a31329 | 1083 | |
7978672c GA |
1084 | memset(timer, 0, sizeof(struct hrtimer)); |
1085 | ||
22127e93 | 1086 | cpu_base = raw_cpu_ptr(&hrtimer_bases); |
c0a31329 | 1087 | |
c9cb2e3d | 1088 | if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS) |
7978672c GA |
1089 | clock_id = CLOCK_MONOTONIC; |
1090 | ||
e06383db JS |
1091 | base = hrtimer_clockid_to_base(clock_id); |
1092 | timer->base = &cpu_base->clock_base[base]; | |
998adc3d | 1093 | timerqueue_init(&timer->node); |
82f67cd9 IM |
1094 | |
1095 | #ifdef CONFIG_TIMER_STATS | |
1096 | timer->start_site = NULL; | |
1097 | timer->start_pid = -1; | |
1098 | memset(timer->start_comm, 0, TASK_COMM_LEN); | |
1099 | #endif | |
c0a31329 | 1100 | } |
237fc6e7 TG |
1101 | |
1102 | /** | |
1103 | * hrtimer_init - initialize a timer to the given clock | |
1104 | * @timer: the timer to be initialized | |
1105 | * @clock_id: the clock to be used | |
1106 | * @mode: timer mode abs/rel | |
1107 | */ | |
1108 | void hrtimer_init(struct hrtimer *timer, clockid_t clock_id, | |
1109 | enum hrtimer_mode mode) | |
1110 | { | |
c6a2a177 | 1111 | debug_init(timer, clock_id, mode); |
237fc6e7 TG |
1112 | __hrtimer_init(timer, clock_id, mode); |
1113 | } | |
8d16b764 | 1114 | EXPORT_SYMBOL_GPL(hrtimer_init); |
c0a31329 | 1115 | |
21d6d52a TG |
1116 | static void __run_hrtimer(struct hrtimer_cpu_base *cpu_base, |
1117 | struct hrtimer_clock_base *base, | |
1118 | struct hrtimer *timer, ktime_t *now) | |
d3d74453 | 1119 | { |
d3d74453 PZ |
1120 | enum hrtimer_restart (*fn)(struct hrtimer *); |
1121 | int restart; | |
1122 | ||
ca109491 PZ |
1123 | WARN_ON(!irqs_disabled()); |
1124 | ||
c6a2a177 | 1125 | debug_deactivate(timer); |
d3d74453 PZ |
1126 | __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0); |
1127 | timer_stats_account_hrtimer(timer); | |
d3d74453 | 1128 | fn = timer->function; |
ca109491 PZ |
1129 | |
1130 | /* | |
1131 | * Because we run timers from hardirq context, there is no chance | |
1132 | * they get migrated to another cpu, therefore its safe to unlock | |
1133 | * the timer base. | |
1134 | */ | |
ecb49d1a | 1135 | raw_spin_unlock(&cpu_base->lock); |
c6a2a177 | 1136 | trace_hrtimer_expire_entry(timer, now); |
ca109491 | 1137 | restart = fn(timer); |
c6a2a177 | 1138 | trace_hrtimer_expire_exit(timer); |
ecb49d1a | 1139 | raw_spin_lock(&cpu_base->lock); |
d3d74453 PZ |
1140 | |
1141 | /* | |
e3f1d883 TG |
1142 | * Note: We clear the CALLBACK bit after enqueue_hrtimer and |
1143 | * we do not reprogramm the event hardware. Happens either in | |
1144 | * hrtimer_start_range_ns() or in hrtimer_interrupt() | |
5de2755c PZ |
1145 | * |
1146 | * Note: Because we dropped the cpu_base->lock above, | |
1147 | * hrtimer_start_range_ns() can have popped in and enqueued the timer | |
1148 | * for us already. | |
d3d74453 | 1149 | */ |
5de2755c PZ |
1150 | if (restart != HRTIMER_NORESTART && |
1151 | !(timer->state & HRTIMER_STATE_ENQUEUED)) | |
a6037b61 | 1152 | enqueue_hrtimer(timer, base); |
f13d4f97 SQ |
1153 | |
1154 | WARN_ON_ONCE(!(timer->state & HRTIMER_STATE_CALLBACK)); | |
1155 | ||
d3d74453 PZ |
1156 | timer->state &= ~HRTIMER_STATE_CALLBACK; |
1157 | } | |
1158 | ||
21d6d52a | 1159 | static void __hrtimer_run_queues(struct hrtimer_cpu_base *cpu_base, ktime_t now) |
54cdfdb4 | 1160 | { |
34aee88a TG |
1161 | struct hrtimer_clock_base *base = cpu_base->clock_base; |
1162 | unsigned int active = cpu_base->active_bases; | |
6ff7041d | 1163 | |
34aee88a | 1164 | for (; active; base++, active >>= 1) { |
998adc3d | 1165 | struct timerqueue_node *node; |
ab8177bc TG |
1166 | ktime_t basenow; |
1167 | ||
34aee88a | 1168 | if (!(active & 0x01)) |
ab8177bc | 1169 | continue; |
54cdfdb4 | 1170 | |
54cdfdb4 TG |
1171 | basenow = ktime_add(now, base->offset); |
1172 | ||
998adc3d | 1173 | while ((node = timerqueue_getnext(&base->active))) { |
54cdfdb4 TG |
1174 | struct hrtimer *timer; |
1175 | ||
998adc3d | 1176 | timer = container_of(node, struct hrtimer, node); |
54cdfdb4 | 1177 | |
654c8e0b AV |
1178 | /* |
1179 | * The immediate goal for using the softexpires is | |
1180 | * minimizing wakeups, not running timers at the | |
1181 | * earliest interrupt after their soft expiration. | |
1182 | * This allows us to avoid using a Priority Search | |
1183 | * Tree, which can answer a stabbing querry for | |
1184 | * overlapping intervals and instead use the simple | |
1185 | * BST we already have. | |
1186 | * We don't add extra wakeups by delaying timers that | |
1187 | * are right-of a not yet expired timer, because that | |
1188 | * timer will have to trigger a wakeup anyway. | |
1189 | */ | |
9bc74919 | 1190 | if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer)) |
54cdfdb4 | 1191 | break; |
54cdfdb4 | 1192 | |
21d6d52a | 1193 | __run_hrtimer(cpu_base, base, timer, &basenow); |
54cdfdb4 | 1194 | } |
54cdfdb4 | 1195 | } |
21d6d52a TG |
1196 | } |
1197 | ||
1198 | #ifdef CONFIG_HIGH_RES_TIMERS | |
1199 | ||
1200 | /* | |
1201 | * High resolution timer interrupt | |
1202 | * Called with interrupts disabled | |
1203 | */ | |
1204 | void hrtimer_interrupt(struct clock_event_device *dev) | |
1205 | { | |
1206 | struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases); | |
1207 | ktime_t expires_next, now, entry_time, delta; | |
1208 | int retries = 0; | |
1209 | ||
1210 | BUG_ON(!cpu_base->hres_active); | |
1211 | cpu_base->nr_events++; | |
1212 | dev->next_event.tv64 = KTIME_MAX; | |
1213 | ||
1214 | raw_spin_lock(&cpu_base->lock); | |
1215 | entry_time = now = hrtimer_update_base(cpu_base); | |
1216 | retry: | |
1217 | cpu_base->in_hrtirq = 1; | |
1218 | /* | |
1219 | * We set expires_next to KTIME_MAX here with cpu_base->lock | |
1220 | * held to prevent that a timer is enqueued in our queue via | |
1221 | * the migration code. This does not affect enqueueing of | |
1222 | * timers which run their callback and need to be requeued on | |
1223 | * this CPU. | |
1224 | */ | |
1225 | cpu_base->expires_next.tv64 = KTIME_MAX; | |
1226 | ||
1227 | __hrtimer_run_queues(cpu_base, now); | |
1228 | ||
9bc74919 TG |
1229 | /* Reevaluate the clock bases for the next expiry */ |
1230 | expires_next = __hrtimer_get_next_event(cpu_base); | |
6ff7041d TG |
1231 | /* |
1232 | * Store the new expiry value so the migration code can verify | |
1233 | * against it. | |
1234 | */ | |
54cdfdb4 | 1235 | cpu_base->expires_next = expires_next; |
9bc74919 | 1236 | cpu_base->in_hrtirq = 0; |
ecb49d1a | 1237 | raw_spin_unlock(&cpu_base->lock); |
54cdfdb4 TG |
1238 | |
1239 | /* Reprogramming necessary ? */ | |
41d2e494 TG |
1240 | if (expires_next.tv64 == KTIME_MAX || |
1241 | !tick_program_event(expires_next, 0)) { | |
1242 | cpu_base->hang_detected = 0; | |
1243 | return; | |
54cdfdb4 | 1244 | } |
41d2e494 TG |
1245 | |
1246 | /* | |
1247 | * The next timer was already expired due to: | |
1248 | * - tracing | |
1249 | * - long lasting callbacks | |
1250 | * - being scheduled away when running in a VM | |
1251 | * | |
1252 | * We need to prevent that we loop forever in the hrtimer | |
1253 | * interrupt routine. We give it 3 attempts to avoid | |
1254 | * overreacting on some spurious event. | |
5baefd6d JS |
1255 | * |
1256 | * Acquire base lock for updating the offsets and retrieving | |
1257 | * the current time. | |
41d2e494 | 1258 | */ |
196951e9 | 1259 | raw_spin_lock(&cpu_base->lock); |
5baefd6d | 1260 | now = hrtimer_update_base(cpu_base); |
41d2e494 TG |
1261 | cpu_base->nr_retries++; |
1262 | if (++retries < 3) | |
1263 | goto retry; | |
1264 | /* | |
1265 | * Give the system a chance to do something else than looping | |
1266 | * here. We stored the entry time, so we know exactly how long | |
1267 | * we spent here. We schedule the next event this amount of | |
1268 | * time away. | |
1269 | */ | |
1270 | cpu_base->nr_hangs++; | |
1271 | cpu_base->hang_detected = 1; | |
196951e9 | 1272 | raw_spin_unlock(&cpu_base->lock); |
41d2e494 | 1273 | delta = ktime_sub(now, entry_time); |
a6ffebce TG |
1274 | if ((unsigned int)delta.tv64 > cpu_base->max_hang_time) |
1275 | cpu_base->max_hang_time = (unsigned int) delta.tv64; | |
41d2e494 TG |
1276 | /* |
1277 | * Limit it to a sensible value as we enforce a longer | |
1278 | * delay. Give the CPU at least 100ms to catch up. | |
1279 | */ | |
1280 | if (delta.tv64 > 100 * NSEC_PER_MSEC) | |
1281 | expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC); | |
1282 | else | |
1283 | expires_next = ktime_add(now, delta); | |
1284 | tick_program_event(expires_next, 1); | |
1285 | printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n", | |
1286 | ktime_to_ns(delta)); | |
54cdfdb4 TG |
1287 | } |
1288 | ||
8bdec955 TG |
1289 | /* |
1290 | * local version of hrtimer_peek_ahead_timers() called with interrupts | |
1291 | * disabled. | |
1292 | */ | |
c6eb3f70 | 1293 | static inline void __hrtimer_peek_ahead_timers(void) |
8bdec955 TG |
1294 | { |
1295 | struct tick_device *td; | |
1296 | ||
1297 | if (!hrtimer_hres_active()) | |
1298 | return; | |
1299 | ||
22127e93 | 1300 | td = this_cpu_ptr(&tick_cpu_device); |
8bdec955 TG |
1301 | if (td && td->evtdev) |
1302 | hrtimer_interrupt(td->evtdev); | |
1303 | } | |
1304 | ||
82c5b7b5 IM |
1305 | #else /* CONFIG_HIGH_RES_TIMERS */ |
1306 | ||
1307 | static inline void __hrtimer_peek_ahead_timers(void) { } | |
1308 | ||
1309 | #endif /* !CONFIG_HIGH_RES_TIMERS */ | |
82f67cd9 | 1310 | |
d3d74453 | 1311 | /* |
c6eb3f70 | 1312 | * Called from run_local_timers in hardirq context every jiffy |
c0a31329 | 1313 | */ |
833883d9 | 1314 | void hrtimer_run_queues(void) |
c0a31329 | 1315 | { |
dc5df73b | 1316 | struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases); |
21d6d52a | 1317 | ktime_t now; |
c0a31329 | 1318 | |
e19ffe8b | 1319 | if (__hrtimer_hres_active(cpu_base)) |
3055adda DS |
1320 | return; |
1321 | ||
c6eb3f70 TG |
1322 | /* |
1323 | * This _is_ ugly: We have to check periodically, whether we | |
1324 | * can switch to highres and / or nohz mode. The clocksource | |
1325 | * switch happens with xtime_lock held. Notification from | |
1326 | * there only sets the check bit in the tick_oneshot code, | |
1327 | * otherwise we might deadlock vs. xtime_lock. | |
1328 | */ | |
1329 | if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) { | |
1330 | hrtimer_switch_to_hres(); | |
1331 | return; | |
1332 | } | |
1333 | ||
21d6d52a TG |
1334 | raw_spin_lock(&cpu_base->lock); |
1335 | now = hrtimer_update_base(cpu_base); | |
1336 | __hrtimer_run_queues(cpu_base, now); | |
1337 | raw_spin_unlock(&cpu_base->lock); | |
c0a31329 TG |
1338 | } |
1339 | ||
10c94ec1 TG |
1340 | /* |
1341 | * Sleep related functions: | |
1342 | */ | |
c9cb2e3d | 1343 | static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer) |
00362e33 TG |
1344 | { |
1345 | struct hrtimer_sleeper *t = | |
1346 | container_of(timer, struct hrtimer_sleeper, timer); | |
1347 | struct task_struct *task = t->task; | |
1348 | ||
1349 | t->task = NULL; | |
1350 | if (task) | |
1351 | wake_up_process(task); | |
1352 | ||
1353 | return HRTIMER_NORESTART; | |
1354 | } | |
1355 | ||
36c8b586 | 1356 | void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task) |
00362e33 TG |
1357 | { |
1358 | sl->timer.function = hrtimer_wakeup; | |
1359 | sl->task = task; | |
1360 | } | |
2bc481cf | 1361 | EXPORT_SYMBOL_GPL(hrtimer_init_sleeper); |
00362e33 | 1362 | |
669d7868 | 1363 | static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode) |
432569bb | 1364 | { |
669d7868 | 1365 | hrtimer_init_sleeper(t, current); |
10c94ec1 | 1366 | |
432569bb RZ |
1367 | do { |
1368 | set_current_state(TASK_INTERRUPTIBLE); | |
cc584b21 | 1369 | hrtimer_start_expires(&t->timer, mode); |
432569bb | 1370 | |
54cdfdb4 | 1371 | if (likely(t->task)) |
b0f8c44f | 1372 | freezable_schedule(); |
432569bb | 1373 | |
669d7868 | 1374 | hrtimer_cancel(&t->timer); |
c9cb2e3d | 1375 | mode = HRTIMER_MODE_ABS; |
669d7868 TG |
1376 | |
1377 | } while (t->task && !signal_pending(current)); | |
432569bb | 1378 | |
3588a085 PZ |
1379 | __set_current_state(TASK_RUNNING); |
1380 | ||
669d7868 | 1381 | return t->task == NULL; |
10c94ec1 TG |
1382 | } |
1383 | ||
080344b9 ON |
1384 | static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp) |
1385 | { | |
1386 | struct timespec rmt; | |
1387 | ktime_t rem; | |
1388 | ||
cc584b21 | 1389 | rem = hrtimer_expires_remaining(timer); |
080344b9 ON |
1390 | if (rem.tv64 <= 0) |
1391 | return 0; | |
1392 | rmt = ktime_to_timespec(rem); | |
1393 | ||
1394 | if (copy_to_user(rmtp, &rmt, sizeof(*rmtp))) | |
1395 | return -EFAULT; | |
1396 | ||
1397 | return 1; | |
1398 | } | |
1399 | ||
1711ef38 | 1400 | long __sched hrtimer_nanosleep_restart(struct restart_block *restart) |
10c94ec1 | 1401 | { |
669d7868 | 1402 | struct hrtimer_sleeper t; |
080344b9 | 1403 | struct timespec __user *rmtp; |
237fc6e7 | 1404 | int ret = 0; |
10c94ec1 | 1405 | |
ab8177bc | 1406 | hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid, |
237fc6e7 | 1407 | HRTIMER_MODE_ABS); |
cc584b21 | 1408 | hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires); |
10c94ec1 | 1409 | |
c9cb2e3d | 1410 | if (do_nanosleep(&t, HRTIMER_MODE_ABS)) |
237fc6e7 | 1411 | goto out; |
10c94ec1 | 1412 | |
029a07e0 | 1413 | rmtp = restart->nanosleep.rmtp; |
432569bb | 1414 | if (rmtp) { |
237fc6e7 | 1415 | ret = update_rmtp(&t.timer, rmtp); |
080344b9 | 1416 | if (ret <= 0) |
237fc6e7 | 1417 | goto out; |
432569bb | 1418 | } |
10c94ec1 | 1419 | |
10c94ec1 | 1420 | /* The other values in restart are already filled in */ |
237fc6e7 TG |
1421 | ret = -ERESTART_RESTARTBLOCK; |
1422 | out: | |
1423 | destroy_hrtimer_on_stack(&t.timer); | |
1424 | return ret; | |
10c94ec1 TG |
1425 | } |
1426 | ||
080344b9 | 1427 | long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp, |
10c94ec1 TG |
1428 | const enum hrtimer_mode mode, const clockid_t clockid) |
1429 | { | |
1430 | struct restart_block *restart; | |
669d7868 | 1431 | struct hrtimer_sleeper t; |
237fc6e7 | 1432 | int ret = 0; |
3bd01206 AV |
1433 | unsigned long slack; |
1434 | ||
1435 | slack = current->timer_slack_ns; | |
aab03e05 | 1436 | if (dl_task(current) || rt_task(current)) |
3bd01206 | 1437 | slack = 0; |
10c94ec1 | 1438 | |
237fc6e7 | 1439 | hrtimer_init_on_stack(&t.timer, clockid, mode); |
3bd01206 | 1440 | hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack); |
432569bb | 1441 | if (do_nanosleep(&t, mode)) |
237fc6e7 | 1442 | goto out; |
10c94ec1 | 1443 | |
7978672c | 1444 | /* Absolute timers do not update the rmtp value and restart: */ |
237fc6e7 TG |
1445 | if (mode == HRTIMER_MODE_ABS) { |
1446 | ret = -ERESTARTNOHAND; | |
1447 | goto out; | |
1448 | } | |
10c94ec1 | 1449 | |
432569bb | 1450 | if (rmtp) { |
237fc6e7 | 1451 | ret = update_rmtp(&t.timer, rmtp); |
080344b9 | 1452 | if (ret <= 0) |
237fc6e7 | 1453 | goto out; |
432569bb | 1454 | } |
10c94ec1 | 1455 | |
f56141e3 | 1456 | restart = ¤t->restart_block; |
1711ef38 | 1457 | restart->fn = hrtimer_nanosleep_restart; |
ab8177bc | 1458 | restart->nanosleep.clockid = t.timer.base->clockid; |
029a07e0 | 1459 | restart->nanosleep.rmtp = rmtp; |
cc584b21 | 1460 | restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer); |
10c94ec1 | 1461 | |
237fc6e7 TG |
1462 | ret = -ERESTART_RESTARTBLOCK; |
1463 | out: | |
1464 | destroy_hrtimer_on_stack(&t.timer); | |
1465 | return ret; | |
10c94ec1 TG |
1466 | } |
1467 | ||
58fd3aa2 HC |
1468 | SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp, |
1469 | struct timespec __user *, rmtp) | |
6ba1b912 | 1470 | { |
080344b9 | 1471 | struct timespec tu; |
6ba1b912 TG |
1472 | |
1473 | if (copy_from_user(&tu, rqtp, sizeof(tu))) | |
1474 | return -EFAULT; | |
1475 | ||
1476 | if (!timespec_valid(&tu)) | |
1477 | return -EINVAL; | |
1478 | ||
080344b9 | 1479 | return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC); |
6ba1b912 TG |
1480 | } |
1481 | ||
c0a31329 TG |
1482 | /* |
1483 | * Functions related to boot-time initialization: | |
1484 | */ | |
0db0628d | 1485 | static void init_hrtimers_cpu(int cpu) |
c0a31329 | 1486 | { |
3c8aa39d | 1487 | struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu); |
c0a31329 TG |
1488 | int i; |
1489 | ||
998adc3d | 1490 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
3c8aa39d | 1491 | cpu_base->clock_base[i].cpu_base = cpu_base; |
998adc3d JS |
1492 | timerqueue_init_head(&cpu_base->clock_base[i].active); |
1493 | } | |
3c8aa39d | 1494 | |
cddd0248 | 1495 | cpu_base->cpu = cpu; |
54cdfdb4 | 1496 | hrtimer_init_hres(cpu_base); |
c0a31329 TG |
1497 | } |
1498 | ||
1499 | #ifdef CONFIG_HOTPLUG_CPU | |
1500 | ||
ca109491 | 1501 | static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base, |
37810659 | 1502 | struct hrtimer_clock_base *new_base) |
c0a31329 TG |
1503 | { |
1504 | struct hrtimer *timer; | |
998adc3d | 1505 | struct timerqueue_node *node; |
c0a31329 | 1506 | |
998adc3d JS |
1507 | while ((node = timerqueue_getnext(&old_base->active))) { |
1508 | timer = container_of(node, struct hrtimer, node); | |
54cdfdb4 | 1509 | BUG_ON(hrtimer_callback_running(timer)); |
c6a2a177 | 1510 | debug_deactivate(timer); |
b00c1a99 TG |
1511 | |
1512 | /* | |
1513 | * Mark it as STATE_MIGRATE not INACTIVE otherwise the | |
1514 | * timer could be seen as !active and just vanish away | |
1515 | * under us on another CPU | |
1516 | */ | |
1517 | __remove_hrtimer(timer, old_base, HRTIMER_STATE_MIGRATE, 0); | |
c0a31329 | 1518 | timer->base = new_base; |
54cdfdb4 | 1519 | /* |
e3f1d883 TG |
1520 | * Enqueue the timers on the new cpu. This does not |
1521 | * reprogram the event device in case the timer | |
1522 | * expires before the earliest on this CPU, but we run | |
1523 | * hrtimer_interrupt after we migrated everything to | |
1524 | * sort out already expired timers and reprogram the | |
1525 | * event device. | |
54cdfdb4 | 1526 | */ |
a6037b61 | 1527 | enqueue_hrtimer(timer, new_base); |
41e1022e | 1528 | |
b00c1a99 TG |
1529 | /* Clear the migration state bit */ |
1530 | timer->state &= ~HRTIMER_STATE_MIGRATE; | |
c0a31329 TG |
1531 | } |
1532 | } | |
1533 | ||
d5fd43c4 | 1534 | static void migrate_hrtimers(int scpu) |
c0a31329 | 1535 | { |
3c8aa39d | 1536 | struct hrtimer_cpu_base *old_base, *new_base; |
731a55ba | 1537 | int i; |
c0a31329 | 1538 | |
37810659 | 1539 | BUG_ON(cpu_online(scpu)); |
37810659 | 1540 | tick_cancel_sched_timer(scpu); |
731a55ba TG |
1541 | |
1542 | local_irq_disable(); | |
1543 | old_base = &per_cpu(hrtimer_bases, scpu); | |
dc5df73b | 1544 | new_base = this_cpu_ptr(&hrtimer_bases); |
d82f0b0f ON |
1545 | /* |
1546 | * The caller is globally serialized and nobody else | |
1547 | * takes two locks at once, deadlock is not possible. | |
1548 | */ | |
ecb49d1a TG |
1549 | raw_spin_lock(&new_base->lock); |
1550 | raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING); | |
c0a31329 | 1551 | |
3c8aa39d | 1552 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
ca109491 | 1553 | migrate_hrtimer_list(&old_base->clock_base[i], |
37810659 | 1554 | &new_base->clock_base[i]); |
c0a31329 TG |
1555 | } |
1556 | ||
ecb49d1a TG |
1557 | raw_spin_unlock(&old_base->lock); |
1558 | raw_spin_unlock(&new_base->lock); | |
37810659 | 1559 | |
731a55ba TG |
1560 | /* Check, if we got expired work to do */ |
1561 | __hrtimer_peek_ahead_timers(); | |
1562 | local_irq_enable(); | |
c0a31329 | 1563 | } |
37810659 | 1564 | |
c0a31329 TG |
1565 | #endif /* CONFIG_HOTPLUG_CPU */ |
1566 | ||
0db0628d | 1567 | static int hrtimer_cpu_notify(struct notifier_block *self, |
c0a31329 TG |
1568 | unsigned long action, void *hcpu) |
1569 | { | |
b2e3c0ad | 1570 | int scpu = (long)hcpu; |
c0a31329 TG |
1571 | |
1572 | switch (action) { | |
1573 | ||
1574 | case CPU_UP_PREPARE: | |
8bb78442 | 1575 | case CPU_UP_PREPARE_FROZEN: |
37810659 | 1576 | init_hrtimers_cpu(scpu); |
c0a31329 TG |
1577 | break; |
1578 | ||
1579 | #ifdef CONFIG_HOTPLUG_CPU | |
1580 | case CPU_DEAD: | |
8bb78442 | 1581 | case CPU_DEAD_FROZEN: |
d5fd43c4 | 1582 | migrate_hrtimers(scpu); |
c0a31329 TG |
1583 | break; |
1584 | #endif | |
1585 | ||
1586 | default: | |
1587 | break; | |
1588 | } | |
1589 | ||
1590 | return NOTIFY_OK; | |
1591 | } | |
1592 | ||
0db0628d | 1593 | static struct notifier_block hrtimers_nb = { |
c0a31329 TG |
1594 | .notifier_call = hrtimer_cpu_notify, |
1595 | }; | |
1596 | ||
1597 | void __init hrtimers_init(void) | |
1598 | { | |
1599 | hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE, | |
1600 | (void *)(long)smp_processor_id()); | |
1601 | register_cpu_notifier(&hrtimers_nb); | |
1602 | } | |
1603 | ||
7bb67439 | 1604 | /** |
351b3f7a | 1605 | * schedule_hrtimeout_range_clock - sleep until timeout |
7bb67439 | 1606 | * @expires: timeout value (ktime_t) |
654c8e0b | 1607 | * @delta: slack in expires timeout (ktime_t) |
7bb67439 | 1608 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL |
351b3f7a | 1609 | * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME |
7bb67439 | 1610 | */ |
351b3f7a CE |
1611 | int __sched |
1612 | schedule_hrtimeout_range_clock(ktime_t *expires, unsigned long delta, | |
1613 | const enum hrtimer_mode mode, int clock) | |
7bb67439 AV |
1614 | { |
1615 | struct hrtimer_sleeper t; | |
1616 | ||
1617 | /* | |
1618 | * Optimize when a zero timeout value is given. It does not | |
1619 | * matter whether this is an absolute or a relative time. | |
1620 | */ | |
1621 | if (expires && !expires->tv64) { | |
1622 | __set_current_state(TASK_RUNNING); | |
1623 | return 0; | |
1624 | } | |
1625 | ||
1626 | /* | |
43b21013 | 1627 | * A NULL parameter means "infinite" |
7bb67439 AV |
1628 | */ |
1629 | if (!expires) { | |
1630 | schedule(); | |
7bb67439 AV |
1631 | return -EINTR; |
1632 | } | |
1633 | ||
351b3f7a | 1634 | hrtimer_init_on_stack(&t.timer, clock, mode); |
654c8e0b | 1635 | hrtimer_set_expires_range_ns(&t.timer, *expires, delta); |
7bb67439 AV |
1636 | |
1637 | hrtimer_init_sleeper(&t, current); | |
1638 | ||
cc584b21 | 1639 | hrtimer_start_expires(&t.timer, mode); |
7bb67439 AV |
1640 | |
1641 | if (likely(t.task)) | |
1642 | schedule(); | |
1643 | ||
1644 | hrtimer_cancel(&t.timer); | |
1645 | destroy_hrtimer_on_stack(&t.timer); | |
1646 | ||
1647 | __set_current_state(TASK_RUNNING); | |
1648 | ||
1649 | return !t.task ? 0 : -EINTR; | |
1650 | } | |
351b3f7a CE |
1651 | |
1652 | /** | |
1653 | * schedule_hrtimeout_range - sleep until timeout | |
1654 | * @expires: timeout value (ktime_t) | |
1655 | * @delta: slack in expires timeout (ktime_t) | |
1656 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL | |
1657 | * | |
1658 | * Make the current task sleep until the given expiry time has | |
1659 | * elapsed. The routine will return immediately unless | |
1660 | * the current task state has been set (see set_current_state()). | |
1661 | * | |
1662 | * The @delta argument gives the kernel the freedom to schedule the | |
1663 | * actual wakeup to a time that is both power and performance friendly. | |
1664 | * The kernel give the normal best effort behavior for "@expires+@delta", | |
1665 | * but may decide to fire the timer earlier, but no earlier than @expires. | |
1666 | * | |
1667 | * You can set the task state as follows - | |
1668 | * | |
1669 | * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to | |
1670 | * pass before the routine returns. | |
1671 | * | |
1672 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
1673 | * delivered to the current task. | |
1674 | * | |
1675 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1676 | * routine returns. | |
1677 | * | |
1678 | * Returns 0 when the timer has expired otherwise -EINTR | |
1679 | */ | |
1680 | int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta, | |
1681 | const enum hrtimer_mode mode) | |
1682 | { | |
1683 | return schedule_hrtimeout_range_clock(expires, delta, mode, | |
1684 | CLOCK_MONOTONIC); | |
1685 | } | |
654c8e0b AV |
1686 | EXPORT_SYMBOL_GPL(schedule_hrtimeout_range); |
1687 | ||
1688 | /** | |
1689 | * schedule_hrtimeout - sleep until timeout | |
1690 | * @expires: timeout value (ktime_t) | |
1691 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL | |
1692 | * | |
1693 | * Make the current task sleep until the given expiry time has | |
1694 | * elapsed. The routine will return immediately unless | |
1695 | * the current task state has been set (see set_current_state()). | |
1696 | * | |
1697 | * You can set the task state as follows - | |
1698 | * | |
1699 | * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to | |
1700 | * pass before the routine returns. | |
1701 | * | |
1702 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
1703 | * delivered to the current task. | |
1704 | * | |
1705 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1706 | * routine returns. | |
1707 | * | |
1708 | * Returns 0 when the timer has expired otherwise -EINTR | |
1709 | */ | |
1710 | int __sched schedule_hrtimeout(ktime_t *expires, | |
1711 | const enum hrtimer_mode mode) | |
1712 | { | |
1713 | return schedule_hrtimeout_range(expires, 0, mode); | |
1714 | } | |
7bb67439 | 1715 | EXPORT_SYMBOL_GPL(schedule_hrtimeout); |