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> | |
35 | #include <linux/module.h> | |
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> |
c0a31329 TG |
46 | |
47 | #include <asm/uaccess.h> | |
48 | ||
49 | /** | |
50 | * ktime_get - get the monotonic time in ktime_t format | |
51 | * | |
52 | * returns the time in ktime_t format | |
53 | */ | |
d316c57f | 54 | ktime_t ktime_get(void) |
c0a31329 TG |
55 | { |
56 | struct timespec now; | |
57 | ||
58 | ktime_get_ts(&now); | |
59 | ||
60 | return timespec_to_ktime(now); | |
61 | } | |
641b9e0e | 62 | EXPORT_SYMBOL_GPL(ktime_get); |
c0a31329 TG |
63 | |
64 | /** | |
65 | * ktime_get_real - get the real (wall-) time in ktime_t format | |
66 | * | |
67 | * returns the time in ktime_t format | |
68 | */ | |
d316c57f | 69 | ktime_t ktime_get_real(void) |
c0a31329 TG |
70 | { |
71 | struct timespec now; | |
72 | ||
73 | getnstimeofday(&now); | |
74 | ||
75 | return timespec_to_ktime(now); | |
76 | } | |
77 | ||
78 | EXPORT_SYMBOL_GPL(ktime_get_real); | |
79 | ||
80 | /* | |
81 | * The timer bases: | |
7978672c GA |
82 | * |
83 | * Note: If we want to add new timer bases, we have to skip the two | |
84 | * clock ids captured by the cpu-timers. We do this by holding empty | |
85 | * entries rather than doing math adjustment of the clock ids. | |
86 | * This ensures that we capture erroneous accesses to these clock ids | |
87 | * rather than moving them into the range of valid clock id's. | |
c0a31329 | 88 | */ |
54cdfdb4 | 89 | DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) = |
c0a31329 | 90 | { |
3c8aa39d TG |
91 | |
92 | .clock_base = | |
c0a31329 | 93 | { |
3c8aa39d TG |
94 | { |
95 | .index = CLOCK_REALTIME, | |
96 | .get_time = &ktime_get_real, | |
54cdfdb4 | 97 | .resolution = KTIME_LOW_RES, |
3c8aa39d TG |
98 | }, |
99 | { | |
100 | .index = CLOCK_MONOTONIC, | |
101 | .get_time = &ktime_get, | |
54cdfdb4 | 102 | .resolution = KTIME_LOW_RES, |
3c8aa39d TG |
103 | }, |
104 | } | |
c0a31329 TG |
105 | }; |
106 | ||
107 | /** | |
108 | * ktime_get_ts - get the monotonic clock in timespec format | |
c0a31329 TG |
109 | * @ts: pointer to timespec variable |
110 | * | |
111 | * The function calculates the monotonic clock from the realtime | |
112 | * clock and the wall_to_monotonic offset and stores the result | |
72fd4a35 | 113 | * in normalized timespec format in the variable pointed to by @ts. |
c0a31329 TG |
114 | */ |
115 | void ktime_get_ts(struct timespec *ts) | |
116 | { | |
117 | struct timespec tomono; | |
118 | unsigned long seq; | |
119 | ||
120 | do { | |
121 | seq = read_seqbegin(&xtime_lock); | |
122 | getnstimeofday(ts); | |
123 | tomono = wall_to_monotonic; | |
124 | ||
125 | } while (read_seqretry(&xtime_lock, seq)); | |
126 | ||
127 | set_normalized_timespec(ts, ts->tv_sec + tomono.tv_sec, | |
128 | ts->tv_nsec + tomono.tv_nsec); | |
129 | } | |
69778e32 | 130 | EXPORT_SYMBOL_GPL(ktime_get_ts); |
c0a31329 | 131 | |
92127c7a TG |
132 | /* |
133 | * Get the coarse grained time at the softirq based on xtime and | |
134 | * wall_to_monotonic. | |
135 | */ | |
3c8aa39d | 136 | static void hrtimer_get_softirq_time(struct hrtimer_cpu_base *base) |
92127c7a TG |
137 | { |
138 | ktime_t xtim, tomono; | |
ad28d94a | 139 | struct timespec xts, tom; |
92127c7a TG |
140 | unsigned long seq; |
141 | ||
142 | do { | |
143 | seq = read_seqbegin(&xtime_lock); | |
2c6b47de | 144 | xts = current_kernel_time(); |
ad28d94a | 145 | tom = wall_to_monotonic; |
92127c7a TG |
146 | } while (read_seqretry(&xtime_lock, seq)); |
147 | ||
f4304ab2 | 148 | xtim = timespec_to_ktime(xts); |
ad28d94a | 149 | tomono = timespec_to_ktime(tom); |
3c8aa39d TG |
150 | base->clock_base[CLOCK_REALTIME].softirq_time = xtim; |
151 | base->clock_base[CLOCK_MONOTONIC].softirq_time = | |
152 | ktime_add(xtim, tomono); | |
92127c7a TG |
153 | } |
154 | ||
c0a31329 TG |
155 | /* |
156 | * Functions and macros which are different for UP/SMP systems are kept in a | |
157 | * single place | |
158 | */ | |
159 | #ifdef CONFIG_SMP | |
160 | ||
c0a31329 TG |
161 | /* |
162 | * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock | |
163 | * means that all timers which are tied to this base via timer->base are | |
164 | * locked, and the base itself is locked too. | |
165 | * | |
166 | * So __run_timers/migrate_timers can safely modify all timers which could | |
167 | * be found on the lists/queues. | |
168 | * | |
169 | * When the timer's base is locked, and the timer removed from list, it is | |
170 | * possible to set timer->base = NULL and drop the lock: the timer remains | |
171 | * locked. | |
172 | */ | |
3c8aa39d TG |
173 | static |
174 | struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer, | |
175 | unsigned long *flags) | |
c0a31329 | 176 | { |
3c8aa39d | 177 | struct hrtimer_clock_base *base; |
c0a31329 TG |
178 | |
179 | for (;;) { | |
180 | base = timer->base; | |
181 | if (likely(base != NULL)) { | |
3c8aa39d | 182 | spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
183 | if (likely(base == timer->base)) |
184 | return base; | |
185 | /* The timer has migrated to another CPU: */ | |
3c8aa39d | 186 | spin_unlock_irqrestore(&base->cpu_base->lock, *flags); |
c0a31329 TG |
187 | } |
188 | cpu_relax(); | |
189 | } | |
190 | } | |
191 | ||
192 | /* | |
193 | * Switch the timer base to the current CPU when possible. | |
194 | */ | |
3c8aa39d TG |
195 | static inline struct hrtimer_clock_base * |
196 | switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base) | |
c0a31329 | 197 | { |
3c8aa39d TG |
198 | struct hrtimer_clock_base *new_base; |
199 | struct hrtimer_cpu_base *new_cpu_base; | |
c0a31329 | 200 | |
3c8aa39d TG |
201 | new_cpu_base = &__get_cpu_var(hrtimer_bases); |
202 | new_base = &new_cpu_base->clock_base[base->index]; | |
c0a31329 TG |
203 | |
204 | if (base != new_base) { | |
205 | /* | |
206 | * We are trying to schedule the timer on the local CPU. | |
207 | * However we can't change timer's base while it is running, | |
208 | * so we keep it on the same CPU. No hassle vs. reprogramming | |
209 | * the event source in the high resolution case. The softirq | |
210 | * code will take care of this when the timer function has | |
211 | * completed. There is no conflict as we hold the lock until | |
212 | * the timer is enqueued. | |
213 | */ | |
54cdfdb4 | 214 | if (unlikely(hrtimer_callback_running(timer))) |
c0a31329 TG |
215 | return base; |
216 | ||
217 | /* See the comment in lock_timer_base() */ | |
218 | timer->base = NULL; | |
3c8aa39d TG |
219 | spin_unlock(&base->cpu_base->lock); |
220 | spin_lock(&new_base->cpu_base->lock); | |
c0a31329 TG |
221 | timer->base = new_base; |
222 | } | |
223 | return new_base; | |
224 | } | |
225 | ||
226 | #else /* CONFIG_SMP */ | |
227 | ||
3c8aa39d | 228 | static inline struct hrtimer_clock_base * |
c0a31329 TG |
229 | lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) |
230 | { | |
3c8aa39d | 231 | struct hrtimer_clock_base *base = timer->base; |
c0a31329 | 232 | |
3c8aa39d | 233 | spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
234 | |
235 | return base; | |
236 | } | |
237 | ||
54cdfdb4 | 238 | # define switch_hrtimer_base(t, b) (b) |
c0a31329 TG |
239 | |
240 | #endif /* !CONFIG_SMP */ | |
241 | ||
242 | /* | |
243 | * Functions for the union type storage format of ktime_t which are | |
244 | * too large for inlining: | |
245 | */ | |
246 | #if BITS_PER_LONG < 64 | |
247 | # ifndef CONFIG_KTIME_SCALAR | |
248 | /** | |
249 | * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable | |
c0a31329 TG |
250 | * @kt: addend |
251 | * @nsec: the scalar nsec value to add | |
252 | * | |
253 | * Returns the sum of kt and nsec in ktime_t format | |
254 | */ | |
255 | ktime_t ktime_add_ns(const ktime_t kt, u64 nsec) | |
256 | { | |
257 | ktime_t tmp; | |
258 | ||
259 | if (likely(nsec < NSEC_PER_SEC)) { | |
260 | tmp.tv64 = nsec; | |
261 | } else { | |
262 | unsigned long rem = do_div(nsec, NSEC_PER_SEC); | |
263 | ||
264 | tmp = ktime_set((long)nsec, rem); | |
265 | } | |
266 | ||
267 | return ktime_add(kt, tmp); | |
268 | } | |
b8b8fd2d DH |
269 | |
270 | EXPORT_SYMBOL_GPL(ktime_add_ns); | |
a272378d ACM |
271 | |
272 | /** | |
273 | * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable | |
274 | * @kt: minuend | |
275 | * @nsec: the scalar nsec value to subtract | |
276 | * | |
277 | * Returns the subtraction of @nsec from @kt in ktime_t format | |
278 | */ | |
279 | ktime_t ktime_sub_ns(const ktime_t kt, u64 nsec) | |
280 | { | |
281 | ktime_t tmp; | |
282 | ||
283 | if (likely(nsec < NSEC_PER_SEC)) { | |
284 | tmp.tv64 = nsec; | |
285 | } else { | |
286 | unsigned long rem = do_div(nsec, NSEC_PER_SEC); | |
287 | ||
288 | tmp = ktime_set((long)nsec, rem); | |
289 | } | |
290 | ||
291 | return ktime_sub(kt, tmp); | |
292 | } | |
293 | ||
294 | EXPORT_SYMBOL_GPL(ktime_sub_ns); | |
c0a31329 TG |
295 | # endif /* !CONFIG_KTIME_SCALAR */ |
296 | ||
297 | /* | |
298 | * Divide a ktime value by a nanosecond value | |
299 | */ | |
4d672e7a | 300 | u64 ktime_divns(const ktime_t kt, s64 div) |
c0a31329 | 301 | { |
900cfa46 | 302 | u64 dclc; |
c0a31329 TG |
303 | int sft = 0; |
304 | ||
900cfa46 | 305 | dclc = ktime_to_ns(kt); |
c0a31329 TG |
306 | /* Make sure the divisor is less than 2^32: */ |
307 | while (div >> 32) { | |
308 | sft++; | |
309 | div >>= 1; | |
310 | } | |
311 | dclc >>= sft; | |
312 | do_div(dclc, (unsigned long) div); | |
313 | ||
4d672e7a | 314 | return dclc; |
c0a31329 | 315 | } |
c0a31329 TG |
316 | #endif /* BITS_PER_LONG >= 64 */ |
317 | ||
5a7780e7 TG |
318 | /* |
319 | * Add two ktime values and do a safety check for overflow: | |
320 | */ | |
321 | ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs) | |
322 | { | |
323 | ktime_t res = ktime_add(lhs, rhs); | |
324 | ||
325 | /* | |
326 | * We use KTIME_SEC_MAX here, the maximum timeout which we can | |
327 | * return to user space in a timespec: | |
328 | */ | |
329 | if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64) | |
330 | res = ktime_set(KTIME_SEC_MAX, 0); | |
331 | ||
332 | return res; | |
333 | } | |
334 | ||
237fc6e7 TG |
335 | #ifdef CONFIG_DEBUG_OBJECTS_TIMERS |
336 | ||
337 | static struct debug_obj_descr hrtimer_debug_descr; | |
338 | ||
339 | /* | |
340 | * fixup_init is called when: | |
341 | * - an active object is initialized | |
342 | */ | |
343 | static int hrtimer_fixup_init(void *addr, enum debug_obj_state state) | |
344 | { | |
345 | struct hrtimer *timer = addr; | |
346 | ||
347 | switch (state) { | |
348 | case ODEBUG_STATE_ACTIVE: | |
349 | hrtimer_cancel(timer); | |
350 | debug_object_init(timer, &hrtimer_debug_descr); | |
351 | return 1; | |
352 | default: | |
353 | return 0; | |
354 | } | |
355 | } | |
356 | ||
357 | /* | |
358 | * fixup_activate is called when: | |
359 | * - an active object is activated | |
360 | * - an unknown object is activated (might be a statically initialized object) | |
361 | */ | |
362 | static int hrtimer_fixup_activate(void *addr, enum debug_obj_state state) | |
363 | { | |
364 | switch (state) { | |
365 | ||
366 | case ODEBUG_STATE_NOTAVAILABLE: | |
367 | WARN_ON_ONCE(1); | |
368 | return 0; | |
369 | ||
370 | case ODEBUG_STATE_ACTIVE: | |
371 | WARN_ON(1); | |
372 | ||
373 | default: | |
374 | return 0; | |
375 | } | |
376 | } | |
377 | ||
378 | /* | |
379 | * fixup_free is called when: | |
380 | * - an active object is freed | |
381 | */ | |
382 | static int hrtimer_fixup_free(void *addr, enum debug_obj_state state) | |
383 | { | |
384 | struct hrtimer *timer = addr; | |
385 | ||
386 | switch (state) { | |
387 | case ODEBUG_STATE_ACTIVE: | |
388 | hrtimer_cancel(timer); | |
389 | debug_object_free(timer, &hrtimer_debug_descr); | |
390 | return 1; | |
391 | default: | |
392 | return 0; | |
393 | } | |
394 | } | |
395 | ||
396 | static struct debug_obj_descr hrtimer_debug_descr = { | |
397 | .name = "hrtimer", | |
398 | .fixup_init = hrtimer_fixup_init, | |
399 | .fixup_activate = hrtimer_fixup_activate, | |
400 | .fixup_free = hrtimer_fixup_free, | |
401 | }; | |
402 | ||
403 | static inline void debug_hrtimer_init(struct hrtimer *timer) | |
404 | { | |
405 | debug_object_init(timer, &hrtimer_debug_descr); | |
406 | } | |
407 | ||
408 | static inline void debug_hrtimer_activate(struct hrtimer *timer) | |
409 | { | |
410 | debug_object_activate(timer, &hrtimer_debug_descr); | |
411 | } | |
412 | ||
413 | static inline void debug_hrtimer_deactivate(struct hrtimer *timer) | |
414 | { | |
415 | debug_object_deactivate(timer, &hrtimer_debug_descr); | |
416 | } | |
417 | ||
418 | static inline void debug_hrtimer_free(struct hrtimer *timer) | |
419 | { | |
420 | debug_object_free(timer, &hrtimer_debug_descr); | |
421 | } | |
422 | ||
423 | static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id, | |
424 | enum hrtimer_mode mode); | |
425 | ||
426 | void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id, | |
427 | enum hrtimer_mode mode) | |
428 | { | |
429 | debug_object_init_on_stack(timer, &hrtimer_debug_descr); | |
430 | __hrtimer_init(timer, clock_id, mode); | |
431 | } | |
432 | ||
433 | void destroy_hrtimer_on_stack(struct hrtimer *timer) | |
434 | { | |
435 | debug_object_free(timer, &hrtimer_debug_descr); | |
436 | } | |
437 | ||
438 | #else | |
439 | static inline void debug_hrtimer_init(struct hrtimer *timer) { } | |
440 | static inline void debug_hrtimer_activate(struct hrtimer *timer) { } | |
441 | static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { } | |
442 | #endif | |
443 | ||
54cdfdb4 TG |
444 | /* High resolution timer related functions */ |
445 | #ifdef CONFIG_HIGH_RES_TIMERS | |
446 | ||
447 | /* | |
448 | * High resolution timer enabled ? | |
449 | */ | |
450 | static int hrtimer_hres_enabled __read_mostly = 1; | |
451 | ||
452 | /* | |
453 | * Enable / Disable high resolution mode | |
454 | */ | |
455 | static int __init setup_hrtimer_hres(char *str) | |
456 | { | |
457 | if (!strcmp(str, "off")) | |
458 | hrtimer_hres_enabled = 0; | |
459 | else if (!strcmp(str, "on")) | |
460 | hrtimer_hres_enabled = 1; | |
461 | else | |
462 | return 0; | |
463 | return 1; | |
464 | } | |
465 | ||
466 | __setup("highres=", setup_hrtimer_hres); | |
467 | ||
468 | /* | |
469 | * hrtimer_high_res_enabled - query, if the highres mode is enabled | |
470 | */ | |
471 | static inline int hrtimer_is_hres_enabled(void) | |
472 | { | |
473 | return hrtimer_hres_enabled; | |
474 | } | |
475 | ||
476 | /* | |
477 | * Is the high resolution mode active ? | |
478 | */ | |
479 | static inline int hrtimer_hres_active(void) | |
480 | { | |
481 | return __get_cpu_var(hrtimer_bases).hres_active; | |
482 | } | |
483 | ||
484 | /* | |
485 | * Reprogram the event source with checking both queues for the | |
486 | * next event | |
487 | * Called with interrupts disabled and base->lock held | |
488 | */ | |
489 | static void hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base) | |
490 | { | |
491 | int i; | |
492 | struct hrtimer_clock_base *base = cpu_base->clock_base; | |
493 | ktime_t expires; | |
494 | ||
495 | cpu_base->expires_next.tv64 = KTIME_MAX; | |
496 | ||
497 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) { | |
498 | struct hrtimer *timer; | |
499 | ||
500 | if (!base->first) | |
501 | continue; | |
502 | timer = rb_entry(base->first, struct hrtimer, node); | |
cc584b21 | 503 | expires = ktime_sub(hrtimer_get_expires(timer), base->offset); |
54cdfdb4 TG |
504 | if (expires.tv64 < cpu_base->expires_next.tv64) |
505 | cpu_base->expires_next = expires; | |
506 | } | |
507 | ||
508 | if (cpu_base->expires_next.tv64 != KTIME_MAX) | |
509 | tick_program_event(cpu_base->expires_next, 1); | |
510 | } | |
511 | ||
512 | /* | |
513 | * Shared reprogramming for clock_realtime and clock_monotonic | |
514 | * | |
515 | * When a timer is enqueued and expires earlier than the already enqueued | |
516 | * timers, we have to check, whether it expires earlier than the timer for | |
517 | * which the clock event device was armed. | |
518 | * | |
519 | * Called with interrupts disabled and base->cpu_base.lock held | |
520 | */ | |
521 | static int hrtimer_reprogram(struct hrtimer *timer, | |
522 | struct hrtimer_clock_base *base) | |
523 | { | |
524 | ktime_t *expires_next = &__get_cpu_var(hrtimer_bases).expires_next; | |
cc584b21 | 525 | ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset); |
54cdfdb4 TG |
526 | int res; |
527 | ||
cc584b21 | 528 | WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0); |
63070a79 | 529 | |
54cdfdb4 TG |
530 | /* |
531 | * When the callback is running, we do not reprogram the clock event | |
532 | * device. The timer callback is either running on a different CPU or | |
3a4fa0a2 | 533 | * the callback is executed in the hrtimer_interrupt context. The |
54cdfdb4 TG |
534 | * reprogramming is handled either by the softirq, which called the |
535 | * callback or at the end of the hrtimer_interrupt. | |
536 | */ | |
537 | if (hrtimer_callback_running(timer)) | |
538 | return 0; | |
539 | ||
63070a79 TG |
540 | /* |
541 | * CLOCK_REALTIME timer might be requested with an absolute | |
542 | * expiry time which is less than base->offset. Nothing wrong | |
543 | * about that, just avoid to call into the tick code, which | |
544 | * has now objections against negative expiry values. | |
545 | */ | |
546 | if (expires.tv64 < 0) | |
547 | return -ETIME; | |
548 | ||
54cdfdb4 TG |
549 | if (expires.tv64 >= expires_next->tv64) |
550 | return 0; | |
551 | ||
552 | /* | |
553 | * Clockevents returns -ETIME, when the event was in the past. | |
554 | */ | |
555 | res = tick_program_event(expires, 0); | |
556 | if (!IS_ERR_VALUE(res)) | |
557 | *expires_next = expires; | |
558 | return res; | |
559 | } | |
560 | ||
561 | ||
562 | /* | |
563 | * Retrigger next event is called after clock was set | |
564 | * | |
565 | * Called with interrupts disabled via on_each_cpu() | |
566 | */ | |
567 | static void retrigger_next_event(void *arg) | |
568 | { | |
569 | struct hrtimer_cpu_base *base; | |
570 | struct timespec realtime_offset; | |
571 | unsigned long seq; | |
572 | ||
573 | if (!hrtimer_hres_active()) | |
574 | return; | |
575 | ||
576 | do { | |
577 | seq = read_seqbegin(&xtime_lock); | |
578 | set_normalized_timespec(&realtime_offset, | |
579 | -wall_to_monotonic.tv_sec, | |
580 | -wall_to_monotonic.tv_nsec); | |
581 | } while (read_seqretry(&xtime_lock, seq)); | |
582 | ||
583 | base = &__get_cpu_var(hrtimer_bases); | |
584 | ||
585 | /* Adjust CLOCK_REALTIME offset */ | |
586 | spin_lock(&base->lock); | |
587 | base->clock_base[CLOCK_REALTIME].offset = | |
588 | timespec_to_ktime(realtime_offset); | |
589 | ||
590 | hrtimer_force_reprogram(base); | |
591 | spin_unlock(&base->lock); | |
592 | } | |
593 | ||
594 | /* | |
595 | * Clock realtime was set | |
596 | * | |
597 | * Change the offset of the realtime clock vs. the monotonic | |
598 | * clock. | |
599 | * | |
600 | * We might have to reprogram the high resolution timer interrupt. On | |
601 | * SMP we call the architecture specific code to retrigger _all_ high | |
602 | * resolution timer interrupts. On UP we just disable interrupts and | |
603 | * call the high resolution interrupt code. | |
604 | */ | |
605 | void clock_was_set(void) | |
606 | { | |
607 | /* Retrigger the CPU local events everywhere */ | |
15c8b6c1 | 608 | on_each_cpu(retrigger_next_event, NULL, 1); |
54cdfdb4 TG |
609 | } |
610 | ||
995f054f IM |
611 | /* |
612 | * During resume we might have to reprogram the high resolution timer | |
613 | * interrupt (on the local CPU): | |
614 | */ | |
615 | void hres_timers_resume(void) | |
616 | { | |
1d4a7f1c PZ |
617 | WARN_ONCE(!irqs_disabled(), |
618 | KERN_INFO "hres_timers_resume() called with IRQs enabled!"); | |
619 | ||
995f054f IM |
620 | retrigger_next_event(NULL); |
621 | } | |
622 | ||
54cdfdb4 TG |
623 | /* |
624 | * Initialize the high resolution related parts of cpu_base | |
625 | */ | |
626 | static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) | |
627 | { | |
628 | base->expires_next.tv64 = KTIME_MAX; | |
629 | base->hres_active = 0; | |
54cdfdb4 TG |
630 | } |
631 | ||
632 | /* | |
633 | * Initialize the high resolution related parts of a hrtimer | |
634 | */ | |
635 | static inline void hrtimer_init_timer_hres(struct hrtimer *timer) | |
636 | { | |
54cdfdb4 TG |
637 | } |
638 | ||
ca109491 | 639 | |
54cdfdb4 TG |
640 | /* |
641 | * When High resolution timers are active, try to reprogram. Note, that in case | |
642 | * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry | |
643 | * check happens. The timer gets enqueued into the rbtree. The reprogramming | |
644 | * and expiry check is done in the hrtimer_interrupt or in the softirq. | |
645 | */ | |
646 | static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer, | |
647 | struct hrtimer_clock_base *base) | |
648 | { | |
649 | if (base->cpu_base->hres_active && hrtimer_reprogram(timer, base)) { | |
a6037b61 PZ |
650 | spin_unlock(&base->cpu_base->lock); |
651 | raise_softirq_irqoff(HRTIMER_SOFTIRQ); | |
652 | spin_lock(&base->cpu_base->lock); | |
ca109491 | 653 | return 1; |
54cdfdb4 TG |
654 | } |
655 | return 0; | |
656 | } | |
657 | ||
658 | /* | |
659 | * Switch to high resolution mode | |
660 | */ | |
f8953856 | 661 | static int hrtimer_switch_to_hres(void) |
54cdfdb4 | 662 | { |
820de5c3 IM |
663 | int cpu = smp_processor_id(); |
664 | struct hrtimer_cpu_base *base = &per_cpu(hrtimer_bases, cpu); | |
54cdfdb4 TG |
665 | unsigned long flags; |
666 | ||
667 | if (base->hres_active) | |
f8953856 | 668 | return 1; |
54cdfdb4 TG |
669 | |
670 | local_irq_save(flags); | |
671 | ||
672 | if (tick_init_highres()) { | |
673 | local_irq_restore(flags); | |
820de5c3 IM |
674 | printk(KERN_WARNING "Could not switch to high resolution " |
675 | "mode on CPU %d\n", cpu); | |
f8953856 | 676 | return 0; |
54cdfdb4 TG |
677 | } |
678 | base->hres_active = 1; | |
679 | base->clock_base[CLOCK_REALTIME].resolution = KTIME_HIGH_RES; | |
680 | base->clock_base[CLOCK_MONOTONIC].resolution = KTIME_HIGH_RES; | |
681 | ||
682 | tick_setup_sched_timer(); | |
683 | ||
684 | /* "Retrigger" the interrupt to get things going */ | |
685 | retrigger_next_event(NULL); | |
686 | local_irq_restore(flags); | |
edfed66e | 687 | printk(KERN_DEBUG "Switched to high resolution mode on CPU %d\n", |
54cdfdb4 | 688 | smp_processor_id()); |
f8953856 | 689 | return 1; |
54cdfdb4 TG |
690 | } |
691 | ||
692 | #else | |
693 | ||
694 | static inline int hrtimer_hres_active(void) { return 0; } | |
695 | static inline int hrtimer_is_hres_enabled(void) { return 0; } | |
f8953856 | 696 | static inline int hrtimer_switch_to_hres(void) { return 0; } |
54cdfdb4 TG |
697 | static inline void hrtimer_force_reprogram(struct hrtimer_cpu_base *base) { } |
698 | static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer, | |
699 | struct hrtimer_clock_base *base) | |
700 | { | |
701 | return 0; | |
702 | } | |
54cdfdb4 TG |
703 | static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { } |
704 | static inline void hrtimer_init_timer_hres(struct hrtimer *timer) { } | |
705 | ||
706 | #endif /* CONFIG_HIGH_RES_TIMERS */ | |
707 | ||
82f67cd9 IM |
708 | #ifdef CONFIG_TIMER_STATS |
709 | void __timer_stats_hrtimer_set_start_info(struct hrtimer *timer, void *addr) | |
710 | { | |
711 | if (timer->start_site) | |
712 | return; | |
713 | ||
714 | timer->start_site = addr; | |
715 | memcpy(timer->start_comm, current->comm, TASK_COMM_LEN); | |
716 | timer->start_pid = current->pid; | |
717 | } | |
718 | #endif | |
719 | ||
c0a31329 | 720 | /* |
6506f2aa | 721 | * Counterpart to lock_hrtimer_base above: |
c0a31329 TG |
722 | */ |
723 | static inline | |
724 | void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) | |
725 | { | |
3c8aa39d | 726 | spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags); |
c0a31329 TG |
727 | } |
728 | ||
729 | /** | |
730 | * hrtimer_forward - forward the timer expiry | |
c0a31329 | 731 | * @timer: hrtimer to forward |
44f21475 | 732 | * @now: forward past this time |
c0a31329 TG |
733 | * @interval: the interval to forward |
734 | * | |
735 | * Forward the timer expiry so it will expire in the future. | |
8dca6f33 | 736 | * Returns the number of overruns. |
c0a31329 | 737 | */ |
4d672e7a | 738 | u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval) |
c0a31329 | 739 | { |
4d672e7a | 740 | u64 orun = 1; |
44f21475 | 741 | ktime_t delta; |
c0a31329 | 742 | |
cc584b21 | 743 | delta = ktime_sub(now, hrtimer_get_expires(timer)); |
c0a31329 TG |
744 | |
745 | if (delta.tv64 < 0) | |
746 | return 0; | |
747 | ||
c9db4fa1 TG |
748 | if (interval.tv64 < timer->base->resolution.tv64) |
749 | interval.tv64 = timer->base->resolution.tv64; | |
750 | ||
c0a31329 | 751 | if (unlikely(delta.tv64 >= interval.tv64)) { |
df869b63 | 752 | s64 incr = ktime_to_ns(interval); |
c0a31329 TG |
753 | |
754 | orun = ktime_divns(delta, incr); | |
cc584b21 AV |
755 | hrtimer_add_expires_ns(timer, incr * orun); |
756 | if (hrtimer_get_expires_tv64(timer) > now.tv64) | |
c0a31329 TG |
757 | return orun; |
758 | /* | |
759 | * This (and the ktime_add() below) is the | |
760 | * correction for exact: | |
761 | */ | |
762 | orun++; | |
763 | } | |
cc584b21 | 764 | hrtimer_add_expires(timer, interval); |
c0a31329 TG |
765 | |
766 | return orun; | |
767 | } | |
6bdb6b62 | 768 | EXPORT_SYMBOL_GPL(hrtimer_forward); |
c0a31329 TG |
769 | |
770 | /* | |
771 | * enqueue_hrtimer - internal function to (re)start a timer | |
772 | * | |
773 | * The timer is inserted in expiry order. Insertion into the | |
774 | * red black tree is O(log(n)). Must hold the base lock. | |
a6037b61 PZ |
775 | * |
776 | * Returns 1 when the new timer is the leftmost timer in the tree. | |
c0a31329 | 777 | */ |
a6037b61 PZ |
778 | static int enqueue_hrtimer(struct hrtimer *timer, |
779 | struct hrtimer_clock_base *base) | |
c0a31329 TG |
780 | { |
781 | struct rb_node **link = &base->active.rb_node; | |
c0a31329 TG |
782 | struct rb_node *parent = NULL; |
783 | struct hrtimer *entry; | |
99bc2fcb | 784 | int leftmost = 1; |
c0a31329 | 785 | |
237fc6e7 TG |
786 | debug_hrtimer_activate(timer); |
787 | ||
c0a31329 TG |
788 | /* |
789 | * Find the right place in the rbtree: | |
790 | */ | |
791 | while (*link) { | |
792 | parent = *link; | |
793 | entry = rb_entry(parent, struct hrtimer, node); | |
794 | /* | |
795 | * We dont care about collisions. Nodes with | |
796 | * the same expiry time stay together. | |
797 | */ | |
cc584b21 AV |
798 | if (hrtimer_get_expires_tv64(timer) < |
799 | hrtimer_get_expires_tv64(entry)) { | |
c0a31329 | 800 | link = &(*link)->rb_left; |
99bc2fcb | 801 | } else { |
c0a31329 | 802 | link = &(*link)->rb_right; |
99bc2fcb IM |
803 | leftmost = 0; |
804 | } | |
c0a31329 TG |
805 | } |
806 | ||
807 | /* | |
288867ec TG |
808 | * Insert the timer to the rbtree and check whether it |
809 | * replaces the first pending timer | |
c0a31329 | 810 | */ |
a6037b61 | 811 | if (leftmost) |
54cdfdb4 | 812 | base->first = &timer->node; |
54cdfdb4 | 813 | |
c0a31329 TG |
814 | rb_link_node(&timer->node, parent, link); |
815 | rb_insert_color(&timer->node, &base->active); | |
303e967f TG |
816 | /* |
817 | * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the | |
818 | * state of a possibly running callback. | |
819 | */ | |
820 | timer->state |= HRTIMER_STATE_ENQUEUED; | |
a6037b61 PZ |
821 | |
822 | return leftmost; | |
288867ec | 823 | } |
c0a31329 TG |
824 | |
825 | /* | |
826 | * __remove_hrtimer - internal function to remove a timer | |
827 | * | |
828 | * Caller must hold the base lock. | |
54cdfdb4 TG |
829 | * |
830 | * High resolution timer mode reprograms the clock event device when the | |
831 | * timer is the one which expires next. The caller can disable this by setting | |
832 | * reprogram to zero. This is useful, when the context does a reprogramming | |
833 | * anyway (e.g. timer interrupt) | |
c0a31329 | 834 | */ |
3c8aa39d | 835 | static void __remove_hrtimer(struct hrtimer *timer, |
303e967f | 836 | struct hrtimer_clock_base *base, |
54cdfdb4 | 837 | unsigned long newstate, int reprogram) |
c0a31329 | 838 | { |
ca109491 | 839 | if (timer->state & HRTIMER_STATE_ENQUEUED) { |
54cdfdb4 TG |
840 | /* |
841 | * Remove the timer from the rbtree and replace the | |
842 | * first entry pointer if necessary. | |
843 | */ | |
844 | if (base->first == &timer->node) { | |
845 | base->first = rb_next(&timer->node); | |
846 | /* Reprogram the clock event device. if enabled */ | |
847 | if (reprogram && hrtimer_hres_active()) | |
848 | hrtimer_force_reprogram(base->cpu_base); | |
849 | } | |
850 | rb_erase(&timer->node, &base->active); | |
851 | } | |
303e967f | 852 | timer->state = newstate; |
c0a31329 TG |
853 | } |
854 | ||
855 | /* | |
856 | * remove hrtimer, called with base lock held | |
857 | */ | |
858 | static inline int | |
3c8aa39d | 859 | remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base) |
c0a31329 | 860 | { |
303e967f | 861 | if (hrtimer_is_queued(timer)) { |
54cdfdb4 TG |
862 | int reprogram; |
863 | ||
864 | /* | |
865 | * Remove the timer and force reprogramming when high | |
866 | * resolution mode is active and the timer is on the current | |
867 | * CPU. If we remove a timer on another CPU, reprogramming is | |
868 | * skipped. The interrupt event on this CPU is fired and | |
869 | * reprogramming happens in the interrupt handler. This is a | |
870 | * rare case and less expensive than a smp call. | |
871 | */ | |
237fc6e7 | 872 | debug_hrtimer_deactivate(timer); |
82f67cd9 | 873 | timer_stats_hrtimer_clear_start_info(timer); |
54cdfdb4 TG |
874 | reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases); |
875 | __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE, | |
876 | reprogram); | |
c0a31329 TG |
877 | return 1; |
878 | } | |
879 | return 0; | |
880 | } | |
881 | ||
882 | /** | |
e1dd7bc5 | 883 | * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU |
c0a31329 TG |
884 | * @timer: the timer to be added |
885 | * @tim: expiry time | |
da8f2e17 | 886 | * @delta_ns: "slack" range for the timer |
c0a31329 TG |
887 | * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL) |
888 | * | |
889 | * Returns: | |
890 | * 0 on success | |
891 | * 1 when the timer was active | |
892 | */ | |
893 | int | |
da8f2e17 AV |
894 | hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, unsigned long delta_ns, |
895 | const enum hrtimer_mode mode) | |
c0a31329 | 896 | { |
3c8aa39d | 897 | struct hrtimer_clock_base *base, *new_base; |
c0a31329 | 898 | unsigned long flags; |
a6037b61 | 899 | int ret, leftmost; |
c0a31329 TG |
900 | |
901 | base = lock_hrtimer_base(timer, &flags); | |
902 | ||
903 | /* Remove an active timer from the queue: */ | |
904 | ret = remove_hrtimer(timer, base); | |
905 | ||
906 | /* Switch the timer base, if necessary: */ | |
907 | new_base = switch_hrtimer_base(timer, base); | |
908 | ||
c9cb2e3d | 909 | if (mode == HRTIMER_MODE_REL) { |
5a7780e7 | 910 | tim = ktime_add_safe(tim, new_base->get_time()); |
06027bdd IM |
911 | /* |
912 | * CONFIG_TIME_LOW_RES is a temporary way for architectures | |
913 | * to signal that they simply return xtime in | |
914 | * do_gettimeoffset(). In this case we want to round up by | |
915 | * resolution when starting a relative timer, to avoid short | |
916 | * timeouts. This will go away with the GTOD framework. | |
917 | */ | |
918 | #ifdef CONFIG_TIME_LOW_RES | |
5a7780e7 | 919 | tim = ktime_add_safe(tim, base->resolution); |
06027bdd IM |
920 | #endif |
921 | } | |
237fc6e7 | 922 | |
da8f2e17 | 923 | hrtimer_set_expires_range_ns(timer, tim, delta_ns); |
c0a31329 | 924 | |
82f67cd9 IM |
925 | timer_stats_hrtimer_set_start_info(timer); |
926 | ||
a6037b61 PZ |
927 | leftmost = enqueue_hrtimer(timer, new_base); |
928 | ||
935c631d IM |
929 | /* |
930 | * Only allow reprogramming if the new base is on this CPU. | |
931 | * (it might still be on another CPU if the timer was pending) | |
a6037b61 PZ |
932 | * |
933 | * XXX send_remote_softirq() ? | |
935c631d | 934 | */ |
a6037b61 PZ |
935 | if (leftmost && new_base->cpu_base == &__get_cpu_var(hrtimer_bases)) |
936 | hrtimer_enqueue_reprogram(timer, new_base); | |
c0a31329 TG |
937 | |
938 | unlock_hrtimer_base(timer, &flags); | |
939 | ||
940 | return ret; | |
941 | } | |
da8f2e17 AV |
942 | EXPORT_SYMBOL_GPL(hrtimer_start_range_ns); |
943 | ||
944 | /** | |
e1dd7bc5 | 945 | * hrtimer_start - (re)start an hrtimer on the current CPU |
da8f2e17 AV |
946 | * @timer: the timer to be added |
947 | * @tim: expiry time | |
948 | * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL) | |
949 | * | |
950 | * Returns: | |
951 | * 0 on success | |
952 | * 1 when the timer was active | |
953 | */ | |
954 | int | |
955 | hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode) | |
956 | { | |
957 | return hrtimer_start_range_ns(timer, tim, 0, mode); | |
958 | } | |
8d16b764 | 959 | EXPORT_SYMBOL_GPL(hrtimer_start); |
c0a31329 | 960 | |
da8f2e17 | 961 | |
c0a31329 TG |
962 | /** |
963 | * hrtimer_try_to_cancel - try to deactivate a timer | |
c0a31329 TG |
964 | * @timer: hrtimer to stop |
965 | * | |
966 | * Returns: | |
967 | * 0 when the timer was not active | |
968 | * 1 when the timer was active | |
969 | * -1 when the timer is currently excuting the callback function and | |
fa9799e3 | 970 | * cannot be stopped |
c0a31329 TG |
971 | */ |
972 | int hrtimer_try_to_cancel(struct hrtimer *timer) | |
973 | { | |
3c8aa39d | 974 | struct hrtimer_clock_base *base; |
c0a31329 TG |
975 | unsigned long flags; |
976 | int ret = -1; | |
977 | ||
978 | base = lock_hrtimer_base(timer, &flags); | |
979 | ||
303e967f | 980 | if (!hrtimer_callback_running(timer)) |
c0a31329 TG |
981 | ret = remove_hrtimer(timer, base); |
982 | ||
983 | unlock_hrtimer_base(timer, &flags); | |
984 | ||
985 | return ret; | |
986 | ||
987 | } | |
8d16b764 | 988 | EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel); |
c0a31329 TG |
989 | |
990 | /** | |
991 | * hrtimer_cancel - cancel a timer and wait for the handler to finish. | |
c0a31329 TG |
992 | * @timer: the timer to be cancelled |
993 | * | |
994 | * Returns: | |
995 | * 0 when the timer was not active | |
996 | * 1 when the timer was active | |
997 | */ | |
998 | int hrtimer_cancel(struct hrtimer *timer) | |
999 | { | |
1000 | for (;;) { | |
1001 | int ret = hrtimer_try_to_cancel(timer); | |
1002 | ||
1003 | if (ret >= 0) | |
1004 | return ret; | |
5ef37b19 | 1005 | cpu_relax(); |
c0a31329 TG |
1006 | } |
1007 | } | |
8d16b764 | 1008 | EXPORT_SYMBOL_GPL(hrtimer_cancel); |
c0a31329 TG |
1009 | |
1010 | /** | |
1011 | * hrtimer_get_remaining - get remaining time for the timer | |
c0a31329 TG |
1012 | * @timer: the timer to read |
1013 | */ | |
1014 | ktime_t hrtimer_get_remaining(const struct hrtimer *timer) | |
1015 | { | |
3c8aa39d | 1016 | struct hrtimer_clock_base *base; |
c0a31329 TG |
1017 | unsigned long flags; |
1018 | ktime_t rem; | |
1019 | ||
1020 | base = lock_hrtimer_base(timer, &flags); | |
cc584b21 | 1021 | rem = hrtimer_expires_remaining(timer); |
c0a31329 TG |
1022 | unlock_hrtimer_base(timer, &flags); |
1023 | ||
1024 | return rem; | |
1025 | } | |
8d16b764 | 1026 | EXPORT_SYMBOL_GPL(hrtimer_get_remaining); |
c0a31329 | 1027 | |
ee9c5785 | 1028 | #ifdef CONFIG_NO_HZ |
69239749 TL |
1029 | /** |
1030 | * hrtimer_get_next_event - get the time until next expiry event | |
1031 | * | |
1032 | * Returns the delta to the next expiry event or KTIME_MAX if no timer | |
1033 | * is pending. | |
1034 | */ | |
1035 | ktime_t hrtimer_get_next_event(void) | |
1036 | { | |
3c8aa39d TG |
1037 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
1038 | struct hrtimer_clock_base *base = cpu_base->clock_base; | |
69239749 TL |
1039 | ktime_t delta, mindelta = { .tv64 = KTIME_MAX }; |
1040 | unsigned long flags; | |
1041 | int i; | |
1042 | ||
3c8aa39d TG |
1043 | spin_lock_irqsave(&cpu_base->lock, flags); |
1044 | ||
54cdfdb4 TG |
1045 | if (!hrtimer_hres_active()) { |
1046 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) { | |
1047 | struct hrtimer *timer; | |
69239749 | 1048 | |
54cdfdb4 TG |
1049 | if (!base->first) |
1050 | continue; | |
3c8aa39d | 1051 | |
54cdfdb4 | 1052 | timer = rb_entry(base->first, struct hrtimer, node); |
cc584b21 | 1053 | delta.tv64 = hrtimer_get_expires_tv64(timer); |
54cdfdb4 TG |
1054 | delta = ktime_sub(delta, base->get_time()); |
1055 | if (delta.tv64 < mindelta.tv64) | |
1056 | mindelta.tv64 = delta.tv64; | |
1057 | } | |
69239749 | 1058 | } |
3c8aa39d TG |
1059 | |
1060 | spin_unlock_irqrestore(&cpu_base->lock, flags); | |
1061 | ||
69239749 TL |
1062 | if (mindelta.tv64 < 0) |
1063 | mindelta.tv64 = 0; | |
1064 | return mindelta; | |
1065 | } | |
1066 | #endif | |
1067 | ||
237fc6e7 TG |
1068 | static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id, |
1069 | enum hrtimer_mode mode) | |
c0a31329 | 1070 | { |
3c8aa39d | 1071 | struct hrtimer_cpu_base *cpu_base; |
c0a31329 | 1072 | |
7978672c GA |
1073 | memset(timer, 0, sizeof(struct hrtimer)); |
1074 | ||
3c8aa39d | 1075 | cpu_base = &__raw_get_cpu_var(hrtimer_bases); |
c0a31329 | 1076 | |
c9cb2e3d | 1077 | if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS) |
7978672c GA |
1078 | clock_id = CLOCK_MONOTONIC; |
1079 | ||
3c8aa39d | 1080 | timer->base = &cpu_base->clock_base[clock_id]; |
d3d74453 | 1081 | INIT_LIST_HEAD(&timer->cb_entry); |
54cdfdb4 | 1082 | hrtimer_init_timer_hres(timer); |
82f67cd9 IM |
1083 | |
1084 | #ifdef CONFIG_TIMER_STATS | |
1085 | timer->start_site = NULL; | |
1086 | timer->start_pid = -1; | |
1087 | memset(timer->start_comm, 0, TASK_COMM_LEN); | |
1088 | #endif | |
c0a31329 | 1089 | } |
237fc6e7 TG |
1090 | |
1091 | /** | |
1092 | * hrtimer_init - initialize a timer to the given clock | |
1093 | * @timer: the timer to be initialized | |
1094 | * @clock_id: the clock to be used | |
1095 | * @mode: timer mode abs/rel | |
1096 | */ | |
1097 | void hrtimer_init(struct hrtimer *timer, clockid_t clock_id, | |
1098 | enum hrtimer_mode mode) | |
1099 | { | |
1100 | debug_hrtimer_init(timer); | |
1101 | __hrtimer_init(timer, clock_id, mode); | |
1102 | } | |
8d16b764 | 1103 | EXPORT_SYMBOL_GPL(hrtimer_init); |
c0a31329 TG |
1104 | |
1105 | /** | |
1106 | * hrtimer_get_res - get the timer resolution for a clock | |
c0a31329 TG |
1107 | * @which_clock: which clock to query |
1108 | * @tp: pointer to timespec variable to store the resolution | |
1109 | * | |
72fd4a35 RD |
1110 | * Store the resolution of the clock selected by @which_clock in the |
1111 | * variable pointed to by @tp. | |
c0a31329 TG |
1112 | */ |
1113 | int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp) | |
1114 | { | |
3c8aa39d | 1115 | struct hrtimer_cpu_base *cpu_base; |
c0a31329 | 1116 | |
3c8aa39d TG |
1117 | cpu_base = &__raw_get_cpu_var(hrtimer_bases); |
1118 | *tp = ktime_to_timespec(cpu_base->clock_base[which_clock].resolution); | |
c0a31329 TG |
1119 | |
1120 | return 0; | |
1121 | } | |
8d16b764 | 1122 | EXPORT_SYMBOL_GPL(hrtimer_get_res); |
c0a31329 | 1123 | |
d3d74453 PZ |
1124 | static void __run_hrtimer(struct hrtimer *timer) |
1125 | { | |
1126 | struct hrtimer_clock_base *base = timer->base; | |
1127 | struct hrtimer_cpu_base *cpu_base = base->cpu_base; | |
1128 | enum hrtimer_restart (*fn)(struct hrtimer *); | |
1129 | int restart; | |
1130 | ||
ca109491 PZ |
1131 | WARN_ON(!irqs_disabled()); |
1132 | ||
237fc6e7 | 1133 | debug_hrtimer_deactivate(timer); |
d3d74453 PZ |
1134 | __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0); |
1135 | timer_stats_account_hrtimer(timer); | |
d3d74453 | 1136 | fn = timer->function; |
ca109491 PZ |
1137 | |
1138 | /* | |
1139 | * Because we run timers from hardirq context, there is no chance | |
1140 | * they get migrated to another cpu, therefore its safe to unlock | |
1141 | * the timer base. | |
1142 | */ | |
1143 | spin_unlock(&cpu_base->lock); | |
1144 | restart = fn(timer); | |
1145 | spin_lock(&cpu_base->lock); | |
d3d74453 PZ |
1146 | |
1147 | /* | |
e3f1d883 TG |
1148 | * Note: We clear the CALLBACK bit after enqueue_hrtimer and |
1149 | * we do not reprogramm the event hardware. Happens either in | |
1150 | * hrtimer_start_range_ns() or in hrtimer_interrupt() | |
d3d74453 PZ |
1151 | */ |
1152 | if (restart != HRTIMER_NORESTART) { | |
1153 | BUG_ON(timer->state != HRTIMER_STATE_CALLBACK); | |
a6037b61 | 1154 | enqueue_hrtimer(timer, base); |
d3d74453 PZ |
1155 | } |
1156 | timer->state &= ~HRTIMER_STATE_CALLBACK; | |
1157 | } | |
1158 | ||
54cdfdb4 TG |
1159 | #ifdef CONFIG_HIGH_RES_TIMERS |
1160 | ||
1161 | /* | |
1162 | * High resolution timer interrupt | |
1163 | * Called with interrupts disabled | |
1164 | */ | |
1165 | void hrtimer_interrupt(struct clock_event_device *dev) | |
1166 | { | |
1167 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); | |
1168 | struct hrtimer_clock_base *base; | |
1169 | ktime_t expires_next, now; | |
ca109491 | 1170 | int i; |
54cdfdb4 TG |
1171 | |
1172 | BUG_ON(!cpu_base->hres_active); | |
1173 | cpu_base->nr_events++; | |
1174 | dev->next_event.tv64 = KTIME_MAX; | |
1175 | ||
1176 | retry: | |
1177 | now = ktime_get(); | |
1178 | ||
1179 | expires_next.tv64 = KTIME_MAX; | |
1180 | ||
1181 | base = cpu_base->clock_base; | |
1182 | ||
1183 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { | |
1184 | ktime_t basenow; | |
1185 | struct rb_node *node; | |
1186 | ||
1187 | spin_lock(&cpu_base->lock); | |
1188 | ||
1189 | basenow = ktime_add(now, base->offset); | |
1190 | ||
1191 | while ((node = base->first)) { | |
1192 | struct hrtimer *timer; | |
1193 | ||
1194 | timer = rb_entry(node, struct hrtimer, node); | |
1195 | ||
654c8e0b AV |
1196 | /* |
1197 | * The immediate goal for using the softexpires is | |
1198 | * minimizing wakeups, not running timers at the | |
1199 | * earliest interrupt after their soft expiration. | |
1200 | * This allows us to avoid using a Priority Search | |
1201 | * Tree, which can answer a stabbing querry for | |
1202 | * overlapping intervals and instead use the simple | |
1203 | * BST we already have. | |
1204 | * We don't add extra wakeups by delaying timers that | |
1205 | * are right-of a not yet expired timer, because that | |
1206 | * timer will have to trigger a wakeup anyway. | |
1207 | */ | |
1208 | ||
1209 | if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer)) { | |
54cdfdb4 TG |
1210 | ktime_t expires; |
1211 | ||
cc584b21 | 1212 | expires = ktime_sub(hrtimer_get_expires(timer), |
54cdfdb4 TG |
1213 | base->offset); |
1214 | if (expires.tv64 < expires_next.tv64) | |
1215 | expires_next = expires; | |
1216 | break; | |
1217 | } | |
1218 | ||
d3d74453 | 1219 | __run_hrtimer(timer); |
54cdfdb4 TG |
1220 | } |
1221 | spin_unlock(&cpu_base->lock); | |
1222 | base++; | |
1223 | } | |
1224 | ||
1225 | cpu_base->expires_next = expires_next; | |
1226 | ||
1227 | /* Reprogramming necessary ? */ | |
1228 | if (expires_next.tv64 != KTIME_MAX) { | |
1229 | if (tick_program_event(expires_next, 0)) | |
1230 | goto retry; | |
1231 | } | |
54cdfdb4 TG |
1232 | } |
1233 | ||
8bdec955 TG |
1234 | /* |
1235 | * local version of hrtimer_peek_ahead_timers() called with interrupts | |
1236 | * disabled. | |
1237 | */ | |
1238 | static void __hrtimer_peek_ahead_timers(void) | |
1239 | { | |
1240 | struct tick_device *td; | |
1241 | ||
1242 | if (!hrtimer_hres_active()) | |
1243 | return; | |
1244 | ||
1245 | td = &__get_cpu_var(tick_cpu_device); | |
1246 | if (td && td->evtdev) | |
1247 | hrtimer_interrupt(td->evtdev); | |
1248 | } | |
1249 | ||
2e94d1f7 AV |
1250 | /** |
1251 | * hrtimer_peek_ahead_timers -- run soft-expired timers now | |
1252 | * | |
1253 | * hrtimer_peek_ahead_timers will peek at the timer queue of | |
1254 | * the current cpu and check if there are any timers for which | |
1255 | * the soft expires time has passed. If any such timers exist, | |
1256 | * they are run immediately and then removed from the timer queue. | |
1257 | * | |
1258 | */ | |
1259 | void hrtimer_peek_ahead_timers(void) | |
1260 | { | |
643bdf68 | 1261 | unsigned long flags; |
dc4304f7 | 1262 | |
2e94d1f7 | 1263 | local_irq_save(flags); |
8bdec955 | 1264 | __hrtimer_peek_ahead_timers(); |
2e94d1f7 AV |
1265 | local_irq_restore(flags); |
1266 | } | |
1267 | ||
a6037b61 PZ |
1268 | static void run_hrtimer_softirq(struct softirq_action *h) |
1269 | { | |
1270 | hrtimer_peek_ahead_timers(); | |
1271 | } | |
1272 | ||
82c5b7b5 IM |
1273 | #else /* CONFIG_HIGH_RES_TIMERS */ |
1274 | ||
1275 | static inline void __hrtimer_peek_ahead_timers(void) { } | |
1276 | ||
1277 | #endif /* !CONFIG_HIGH_RES_TIMERS */ | |
82f67cd9 | 1278 | |
d3d74453 PZ |
1279 | /* |
1280 | * Called from timer softirq every jiffy, expire hrtimers: | |
1281 | * | |
1282 | * For HRT its the fall back code to run the softirq in the timer | |
1283 | * softirq context in case the hrtimer initialization failed or has | |
1284 | * not been done yet. | |
1285 | */ | |
1286 | void hrtimer_run_pending(void) | |
1287 | { | |
d3d74453 PZ |
1288 | if (hrtimer_hres_active()) |
1289 | return; | |
54cdfdb4 | 1290 | |
d3d74453 PZ |
1291 | /* |
1292 | * This _is_ ugly: We have to check in the softirq context, | |
1293 | * whether we can switch to highres and / or nohz mode. The | |
1294 | * clocksource switch happens in the timer interrupt with | |
1295 | * xtime_lock held. Notification from there only sets the | |
1296 | * check bit in the tick_oneshot code, otherwise we might | |
1297 | * deadlock vs. xtime_lock. | |
1298 | */ | |
1299 | if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) | |
1300 | hrtimer_switch_to_hres(); | |
54cdfdb4 TG |
1301 | } |
1302 | ||
c0a31329 | 1303 | /* |
d3d74453 | 1304 | * Called from hardirq context every jiffy |
c0a31329 | 1305 | */ |
833883d9 | 1306 | void hrtimer_run_queues(void) |
c0a31329 | 1307 | { |
288867ec | 1308 | struct rb_node *node; |
833883d9 DS |
1309 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
1310 | struct hrtimer_clock_base *base; | |
1311 | int index, gettime = 1; | |
c0a31329 | 1312 | |
833883d9 | 1313 | if (hrtimer_hres_active()) |
3055adda DS |
1314 | return; |
1315 | ||
833883d9 DS |
1316 | for (index = 0; index < HRTIMER_MAX_CLOCK_BASES; index++) { |
1317 | base = &cpu_base->clock_base[index]; | |
c0a31329 | 1318 | |
833883d9 | 1319 | if (!base->first) |
d3d74453 | 1320 | continue; |
833883d9 | 1321 | |
d7cfb60c | 1322 | if (gettime) { |
833883d9 DS |
1323 | hrtimer_get_softirq_time(cpu_base); |
1324 | gettime = 0; | |
b75f7a51 | 1325 | } |
d3d74453 | 1326 | |
833883d9 | 1327 | spin_lock(&cpu_base->lock); |
c0a31329 | 1328 | |
833883d9 DS |
1329 | while ((node = base->first)) { |
1330 | struct hrtimer *timer; | |
54cdfdb4 | 1331 | |
833883d9 | 1332 | timer = rb_entry(node, struct hrtimer, node); |
cc584b21 AV |
1333 | if (base->softirq_time.tv64 <= |
1334 | hrtimer_get_expires_tv64(timer)) | |
833883d9 DS |
1335 | break; |
1336 | ||
833883d9 DS |
1337 | __run_hrtimer(timer); |
1338 | } | |
1339 | spin_unlock(&cpu_base->lock); | |
1340 | } | |
c0a31329 TG |
1341 | } |
1342 | ||
10c94ec1 TG |
1343 | /* |
1344 | * Sleep related functions: | |
1345 | */ | |
c9cb2e3d | 1346 | static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer) |
00362e33 TG |
1347 | { |
1348 | struct hrtimer_sleeper *t = | |
1349 | container_of(timer, struct hrtimer_sleeper, timer); | |
1350 | struct task_struct *task = t->task; | |
1351 | ||
1352 | t->task = NULL; | |
1353 | if (task) | |
1354 | wake_up_process(task); | |
1355 | ||
1356 | return HRTIMER_NORESTART; | |
1357 | } | |
1358 | ||
36c8b586 | 1359 | void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task) |
00362e33 TG |
1360 | { |
1361 | sl->timer.function = hrtimer_wakeup; | |
1362 | sl->task = task; | |
1363 | } | |
1364 | ||
669d7868 | 1365 | static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode) |
432569bb | 1366 | { |
669d7868 | 1367 | hrtimer_init_sleeper(t, current); |
10c94ec1 | 1368 | |
432569bb RZ |
1369 | do { |
1370 | set_current_state(TASK_INTERRUPTIBLE); | |
cc584b21 | 1371 | hrtimer_start_expires(&t->timer, mode); |
37bb6cb4 PZ |
1372 | if (!hrtimer_active(&t->timer)) |
1373 | t->task = NULL; | |
432569bb | 1374 | |
54cdfdb4 TG |
1375 | if (likely(t->task)) |
1376 | schedule(); | |
432569bb | 1377 | |
669d7868 | 1378 | hrtimer_cancel(&t->timer); |
c9cb2e3d | 1379 | mode = HRTIMER_MODE_ABS; |
669d7868 TG |
1380 | |
1381 | } while (t->task && !signal_pending(current)); | |
432569bb | 1382 | |
3588a085 PZ |
1383 | __set_current_state(TASK_RUNNING); |
1384 | ||
669d7868 | 1385 | return t->task == NULL; |
10c94ec1 TG |
1386 | } |
1387 | ||
080344b9 ON |
1388 | static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp) |
1389 | { | |
1390 | struct timespec rmt; | |
1391 | ktime_t rem; | |
1392 | ||
cc584b21 | 1393 | rem = hrtimer_expires_remaining(timer); |
080344b9 ON |
1394 | if (rem.tv64 <= 0) |
1395 | return 0; | |
1396 | rmt = ktime_to_timespec(rem); | |
1397 | ||
1398 | if (copy_to_user(rmtp, &rmt, sizeof(*rmtp))) | |
1399 | return -EFAULT; | |
1400 | ||
1401 | return 1; | |
1402 | } | |
1403 | ||
1711ef38 | 1404 | long __sched hrtimer_nanosleep_restart(struct restart_block *restart) |
10c94ec1 | 1405 | { |
669d7868 | 1406 | struct hrtimer_sleeper t; |
080344b9 | 1407 | struct timespec __user *rmtp; |
237fc6e7 | 1408 | int ret = 0; |
10c94ec1 | 1409 | |
237fc6e7 TG |
1410 | hrtimer_init_on_stack(&t.timer, restart->nanosleep.index, |
1411 | HRTIMER_MODE_ABS); | |
cc584b21 | 1412 | hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires); |
10c94ec1 | 1413 | |
c9cb2e3d | 1414 | if (do_nanosleep(&t, HRTIMER_MODE_ABS)) |
237fc6e7 | 1415 | goto out; |
10c94ec1 | 1416 | |
029a07e0 | 1417 | rmtp = restart->nanosleep.rmtp; |
432569bb | 1418 | if (rmtp) { |
237fc6e7 | 1419 | ret = update_rmtp(&t.timer, rmtp); |
080344b9 | 1420 | if (ret <= 0) |
237fc6e7 | 1421 | goto out; |
432569bb | 1422 | } |
10c94ec1 | 1423 | |
10c94ec1 | 1424 | /* The other values in restart are already filled in */ |
237fc6e7 TG |
1425 | ret = -ERESTART_RESTARTBLOCK; |
1426 | out: | |
1427 | destroy_hrtimer_on_stack(&t.timer); | |
1428 | return ret; | |
10c94ec1 TG |
1429 | } |
1430 | ||
080344b9 | 1431 | long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp, |
10c94ec1 TG |
1432 | const enum hrtimer_mode mode, const clockid_t clockid) |
1433 | { | |
1434 | struct restart_block *restart; | |
669d7868 | 1435 | struct hrtimer_sleeper t; |
237fc6e7 | 1436 | int ret = 0; |
3bd01206 AV |
1437 | unsigned long slack; |
1438 | ||
1439 | slack = current->timer_slack_ns; | |
1440 | if (rt_task(current)) | |
1441 | slack = 0; | |
10c94ec1 | 1442 | |
237fc6e7 | 1443 | hrtimer_init_on_stack(&t.timer, clockid, mode); |
3bd01206 | 1444 | hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack); |
432569bb | 1445 | if (do_nanosleep(&t, mode)) |
237fc6e7 | 1446 | goto out; |
10c94ec1 | 1447 | |
7978672c | 1448 | /* Absolute timers do not update the rmtp value and restart: */ |
237fc6e7 TG |
1449 | if (mode == HRTIMER_MODE_ABS) { |
1450 | ret = -ERESTARTNOHAND; | |
1451 | goto out; | |
1452 | } | |
10c94ec1 | 1453 | |
432569bb | 1454 | if (rmtp) { |
237fc6e7 | 1455 | ret = update_rmtp(&t.timer, rmtp); |
080344b9 | 1456 | if (ret <= 0) |
237fc6e7 | 1457 | goto out; |
432569bb | 1458 | } |
10c94ec1 TG |
1459 | |
1460 | restart = ¤t_thread_info()->restart_block; | |
1711ef38 | 1461 | restart->fn = hrtimer_nanosleep_restart; |
029a07e0 TG |
1462 | restart->nanosleep.index = t.timer.base->index; |
1463 | restart->nanosleep.rmtp = rmtp; | |
cc584b21 | 1464 | restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer); |
10c94ec1 | 1465 | |
237fc6e7 TG |
1466 | ret = -ERESTART_RESTARTBLOCK; |
1467 | out: | |
1468 | destroy_hrtimer_on_stack(&t.timer); | |
1469 | return ret; | |
10c94ec1 TG |
1470 | } |
1471 | ||
58fd3aa2 HC |
1472 | SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp, |
1473 | struct timespec __user *, rmtp) | |
6ba1b912 | 1474 | { |
080344b9 | 1475 | struct timespec tu; |
6ba1b912 TG |
1476 | |
1477 | if (copy_from_user(&tu, rqtp, sizeof(tu))) | |
1478 | return -EFAULT; | |
1479 | ||
1480 | if (!timespec_valid(&tu)) | |
1481 | return -EINVAL; | |
1482 | ||
080344b9 | 1483 | return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC); |
6ba1b912 TG |
1484 | } |
1485 | ||
c0a31329 TG |
1486 | /* |
1487 | * Functions related to boot-time initialization: | |
1488 | */ | |
0ec160dd | 1489 | static void __cpuinit init_hrtimers_cpu(int cpu) |
c0a31329 | 1490 | { |
3c8aa39d | 1491 | struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu); |
c0a31329 TG |
1492 | int i; |
1493 | ||
3c8aa39d | 1494 | spin_lock_init(&cpu_base->lock); |
3c8aa39d TG |
1495 | |
1496 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) | |
1497 | cpu_base->clock_base[i].cpu_base = cpu_base; | |
1498 | ||
54cdfdb4 | 1499 | hrtimer_init_hres(cpu_base); |
c0a31329 TG |
1500 | } |
1501 | ||
1502 | #ifdef CONFIG_HOTPLUG_CPU | |
1503 | ||
ca109491 | 1504 | static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base, |
37810659 | 1505 | struct hrtimer_clock_base *new_base) |
c0a31329 TG |
1506 | { |
1507 | struct hrtimer *timer; | |
1508 | struct rb_node *node; | |
1509 | ||
1510 | while ((node = rb_first(&old_base->active))) { | |
1511 | timer = rb_entry(node, struct hrtimer, node); | |
54cdfdb4 | 1512 | BUG_ON(hrtimer_callback_running(timer)); |
237fc6e7 | 1513 | debug_hrtimer_deactivate(timer); |
b00c1a99 TG |
1514 | |
1515 | /* | |
1516 | * Mark it as STATE_MIGRATE not INACTIVE otherwise the | |
1517 | * timer could be seen as !active and just vanish away | |
1518 | * under us on another CPU | |
1519 | */ | |
1520 | __remove_hrtimer(timer, old_base, HRTIMER_STATE_MIGRATE, 0); | |
c0a31329 | 1521 | timer->base = new_base; |
54cdfdb4 | 1522 | /* |
e3f1d883 TG |
1523 | * Enqueue the timers on the new cpu. This does not |
1524 | * reprogram the event device in case the timer | |
1525 | * expires before the earliest on this CPU, but we run | |
1526 | * hrtimer_interrupt after we migrated everything to | |
1527 | * sort out already expired timers and reprogram the | |
1528 | * event device. | |
54cdfdb4 | 1529 | */ |
a6037b61 | 1530 | enqueue_hrtimer(timer, new_base); |
41e1022e | 1531 | |
b00c1a99 TG |
1532 | /* Clear the migration state bit */ |
1533 | timer->state &= ~HRTIMER_STATE_MIGRATE; | |
c0a31329 TG |
1534 | } |
1535 | } | |
1536 | ||
d5fd43c4 | 1537 | static void migrate_hrtimers(int scpu) |
c0a31329 | 1538 | { |
3c8aa39d | 1539 | struct hrtimer_cpu_base *old_base, *new_base; |
731a55ba | 1540 | int i; |
c0a31329 | 1541 | |
37810659 | 1542 | BUG_ON(cpu_online(scpu)); |
37810659 | 1543 | tick_cancel_sched_timer(scpu); |
731a55ba TG |
1544 | |
1545 | local_irq_disable(); | |
1546 | old_base = &per_cpu(hrtimer_bases, scpu); | |
1547 | new_base = &__get_cpu_var(hrtimer_bases); | |
d82f0b0f ON |
1548 | /* |
1549 | * The caller is globally serialized and nobody else | |
1550 | * takes two locks at once, deadlock is not possible. | |
1551 | */ | |
731a55ba | 1552 | spin_lock(&new_base->lock); |
8e60e05f | 1553 | spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING); |
c0a31329 | 1554 | |
3c8aa39d | 1555 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
ca109491 | 1556 | migrate_hrtimer_list(&old_base->clock_base[i], |
37810659 | 1557 | &new_base->clock_base[i]); |
c0a31329 TG |
1558 | } |
1559 | ||
8e60e05f | 1560 | spin_unlock(&old_base->lock); |
731a55ba | 1561 | spin_unlock(&new_base->lock); |
37810659 | 1562 | |
731a55ba TG |
1563 | /* Check, if we got expired work to do */ |
1564 | __hrtimer_peek_ahead_timers(); | |
1565 | local_irq_enable(); | |
c0a31329 | 1566 | } |
37810659 | 1567 | |
c0a31329 TG |
1568 | #endif /* CONFIG_HOTPLUG_CPU */ |
1569 | ||
8c78f307 | 1570 | static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self, |
c0a31329 TG |
1571 | unsigned long action, void *hcpu) |
1572 | { | |
b2e3c0ad | 1573 | int scpu = (long)hcpu; |
c0a31329 TG |
1574 | |
1575 | switch (action) { | |
1576 | ||
1577 | case CPU_UP_PREPARE: | |
8bb78442 | 1578 | case CPU_UP_PREPARE_FROZEN: |
37810659 | 1579 | init_hrtimers_cpu(scpu); |
c0a31329 TG |
1580 | break; |
1581 | ||
1582 | #ifdef CONFIG_HOTPLUG_CPU | |
1583 | case CPU_DEAD: | |
8bb78442 | 1584 | case CPU_DEAD_FROZEN: |
b2e3c0ad | 1585 | { |
37810659 | 1586 | clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &scpu); |
d5fd43c4 | 1587 | migrate_hrtimers(scpu); |
c0a31329 | 1588 | break; |
b2e3c0ad | 1589 | } |
c0a31329 TG |
1590 | #endif |
1591 | ||
1592 | default: | |
1593 | break; | |
1594 | } | |
1595 | ||
1596 | return NOTIFY_OK; | |
1597 | } | |
1598 | ||
8c78f307 | 1599 | static struct notifier_block __cpuinitdata hrtimers_nb = { |
c0a31329 TG |
1600 | .notifier_call = hrtimer_cpu_notify, |
1601 | }; | |
1602 | ||
1603 | void __init hrtimers_init(void) | |
1604 | { | |
1605 | hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE, | |
1606 | (void *)(long)smp_processor_id()); | |
1607 | register_cpu_notifier(&hrtimers_nb); | |
a6037b61 PZ |
1608 | #ifdef CONFIG_HIGH_RES_TIMERS |
1609 | open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq); | |
1610 | #endif | |
c0a31329 TG |
1611 | } |
1612 | ||
7bb67439 | 1613 | /** |
654c8e0b | 1614 | * schedule_hrtimeout_range - sleep until timeout |
7bb67439 | 1615 | * @expires: timeout value (ktime_t) |
654c8e0b | 1616 | * @delta: slack in expires timeout (ktime_t) |
7bb67439 AV |
1617 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL |
1618 | * | |
1619 | * Make the current task sleep until the given expiry time has | |
1620 | * elapsed. The routine will return immediately unless | |
1621 | * the current task state has been set (see set_current_state()). | |
1622 | * | |
654c8e0b AV |
1623 | * The @delta argument gives the kernel the freedom to schedule the |
1624 | * actual wakeup to a time that is both power and performance friendly. | |
1625 | * The kernel give the normal best effort behavior for "@expires+@delta", | |
1626 | * but may decide to fire the timer earlier, but no earlier than @expires. | |
1627 | * | |
7bb67439 AV |
1628 | * You can set the task state as follows - |
1629 | * | |
1630 | * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to | |
1631 | * pass before the routine returns. | |
1632 | * | |
1633 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
1634 | * delivered to the current task. | |
1635 | * | |
1636 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1637 | * routine returns. | |
1638 | * | |
1639 | * Returns 0 when the timer has expired otherwise -EINTR | |
1640 | */ | |
654c8e0b | 1641 | int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta, |
7bb67439 AV |
1642 | const enum hrtimer_mode mode) |
1643 | { | |
1644 | struct hrtimer_sleeper t; | |
1645 | ||
1646 | /* | |
1647 | * Optimize when a zero timeout value is given. It does not | |
1648 | * matter whether this is an absolute or a relative time. | |
1649 | */ | |
1650 | if (expires && !expires->tv64) { | |
1651 | __set_current_state(TASK_RUNNING); | |
1652 | return 0; | |
1653 | } | |
1654 | ||
1655 | /* | |
1656 | * A NULL parameter means "inifinte" | |
1657 | */ | |
1658 | if (!expires) { | |
1659 | schedule(); | |
1660 | __set_current_state(TASK_RUNNING); | |
1661 | return -EINTR; | |
1662 | } | |
1663 | ||
1664 | hrtimer_init_on_stack(&t.timer, CLOCK_MONOTONIC, mode); | |
654c8e0b | 1665 | hrtimer_set_expires_range_ns(&t.timer, *expires, delta); |
7bb67439 AV |
1666 | |
1667 | hrtimer_init_sleeper(&t, current); | |
1668 | ||
cc584b21 | 1669 | hrtimer_start_expires(&t.timer, mode); |
7bb67439 AV |
1670 | if (!hrtimer_active(&t.timer)) |
1671 | t.task = NULL; | |
1672 | ||
1673 | if (likely(t.task)) | |
1674 | schedule(); | |
1675 | ||
1676 | hrtimer_cancel(&t.timer); | |
1677 | destroy_hrtimer_on_stack(&t.timer); | |
1678 | ||
1679 | __set_current_state(TASK_RUNNING); | |
1680 | ||
1681 | return !t.task ? 0 : -EINTR; | |
1682 | } | |
654c8e0b AV |
1683 | EXPORT_SYMBOL_GPL(schedule_hrtimeout_range); |
1684 | ||
1685 | /** | |
1686 | * schedule_hrtimeout - sleep until timeout | |
1687 | * @expires: timeout value (ktime_t) | |
1688 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL | |
1689 | * | |
1690 | * Make the current task sleep until the given expiry time has | |
1691 | * elapsed. The routine will return immediately unless | |
1692 | * the current task state has been set (see set_current_state()). | |
1693 | * | |
1694 | * You can set the task state as follows - | |
1695 | * | |
1696 | * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to | |
1697 | * pass before the routine returns. | |
1698 | * | |
1699 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
1700 | * delivered to the current task. | |
1701 | * | |
1702 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1703 | * routine returns. | |
1704 | * | |
1705 | * Returns 0 when the timer has expired otherwise -EINTR | |
1706 | */ | |
1707 | int __sched schedule_hrtimeout(ktime_t *expires, | |
1708 | const enum hrtimer_mode mode) | |
1709 | { | |
1710 | return schedule_hrtimeout_range(expires, 0, mode); | |
1711 | } | |
7bb67439 | 1712 | EXPORT_SYMBOL_GPL(schedule_hrtimeout); |