Commit | Line | Data |
---|---|---|
c0a31329 TG |
1 | /* |
2 | * linux/kernel/hrtimer.c | |
3 | * | |
4 | * Copyright(C) 2005, Thomas Gleixner <tglx@linutronix.de> | |
5 | * Copyright(C) 2005, Red Hat, Inc., Ingo Molnar | |
6 | * | |
7 | * High-resolution kernel timers | |
8 | * | |
9 | * In contrast to the low-resolution timeout API implemented in | |
10 | * kernel/timer.c, hrtimers provide finer resolution and accuracy | |
11 | * depending on system configuration and capabilities. | |
12 | * | |
13 | * These timers are currently used for: | |
14 | * - itimers | |
15 | * - POSIX timers | |
16 | * - nanosleep | |
17 | * - precise in-kernel timing | |
18 | * | |
19 | * Started by: Thomas Gleixner and Ingo Molnar | |
20 | * | |
21 | * Credits: | |
22 | * based on kernel/timer.c | |
23 | * | |
66188fae TG |
24 | * Help, testing, suggestions, bugfixes, improvements were |
25 | * provided by: | |
26 | * | |
27 | * George Anzinger, Andrew Morton, Steven Rostedt, Roman Zippel | |
28 | * et. al. | |
29 | * | |
c0a31329 TG |
30 | * For licencing details see kernel-base/COPYING |
31 | */ | |
32 | ||
33 | #include <linux/cpu.h> | |
34 | #include <linux/module.h> | |
35 | #include <linux/percpu.h> | |
36 | #include <linux/hrtimer.h> | |
37 | #include <linux/notifier.h> | |
38 | #include <linux/syscalls.h> | |
39 | #include <linux/interrupt.h> | |
40 | ||
41 | #include <asm/uaccess.h> | |
42 | ||
43 | /** | |
44 | * ktime_get - get the monotonic time in ktime_t format | |
45 | * | |
46 | * returns the time in ktime_t format | |
47 | */ | |
48 | static ktime_t ktime_get(void) | |
49 | { | |
50 | struct timespec now; | |
51 | ||
52 | ktime_get_ts(&now); | |
53 | ||
54 | return timespec_to_ktime(now); | |
55 | } | |
56 | ||
57 | /** | |
58 | * ktime_get_real - get the real (wall-) time in ktime_t format | |
59 | * | |
60 | * returns the time in ktime_t format | |
61 | */ | |
62 | static ktime_t ktime_get_real(void) | |
63 | { | |
64 | struct timespec now; | |
65 | ||
66 | getnstimeofday(&now); | |
67 | ||
68 | return timespec_to_ktime(now); | |
69 | } | |
70 | ||
71 | EXPORT_SYMBOL_GPL(ktime_get_real); | |
72 | ||
73 | /* | |
74 | * The timer bases: | |
7978672c GA |
75 | * |
76 | * Note: If we want to add new timer bases, we have to skip the two | |
77 | * clock ids captured by the cpu-timers. We do this by holding empty | |
78 | * entries rather than doing math adjustment of the clock ids. | |
79 | * This ensures that we capture erroneous accesses to these clock ids | |
80 | * rather than moving them into the range of valid clock id's. | |
c0a31329 TG |
81 | */ |
82 | ||
83 | #define MAX_HRTIMER_BASES 2 | |
84 | ||
85 | static DEFINE_PER_CPU(struct hrtimer_base, hrtimer_bases[MAX_HRTIMER_BASES]) = | |
86 | { | |
87 | { | |
88 | .index = CLOCK_REALTIME, | |
89 | .get_time = &ktime_get_real, | |
90 | .resolution = KTIME_REALTIME_RES, | |
91 | }, | |
92 | { | |
93 | .index = CLOCK_MONOTONIC, | |
94 | .get_time = &ktime_get, | |
95 | .resolution = KTIME_MONOTONIC_RES, | |
96 | }, | |
97 | }; | |
98 | ||
99 | /** | |
100 | * ktime_get_ts - get the monotonic clock in timespec format | |
101 | * | |
102 | * @ts: pointer to timespec variable | |
103 | * | |
104 | * The function calculates the monotonic clock from the realtime | |
105 | * clock and the wall_to_monotonic offset and stores the result | |
106 | * in normalized timespec format in the variable pointed to by ts. | |
107 | */ | |
108 | void ktime_get_ts(struct timespec *ts) | |
109 | { | |
110 | struct timespec tomono; | |
111 | unsigned long seq; | |
112 | ||
113 | do { | |
114 | seq = read_seqbegin(&xtime_lock); | |
115 | getnstimeofday(ts); | |
116 | tomono = wall_to_monotonic; | |
117 | ||
118 | } while (read_seqretry(&xtime_lock, seq)); | |
119 | ||
120 | set_normalized_timespec(ts, ts->tv_sec + tomono.tv_sec, | |
121 | ts->tv_nsec + tomono.tv_nsec); | |
122 | } | |
69778e32 | 123 | EXPORT_SYMBOL_GPL(ktime_get_ts); |
c0a31329 TG |
124 | |
125 | /* | |
126 | * Functions and macros which are different for UP/SMP systems are kept in a | |
127 | * single place | |
128 | */ | |
129 | #ifdef CONFIG_SMP | |
130 | ||
131 | #define set_curr_timer(b, t) do { (b)->curr_timer = (t); } while (0) | |
132 | ||
133 | /* | |
134 | * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock | |
135 | * means that all timers which are tied to this base via timer->base are | |
136 | * locked, and the base itself is locked too. | |
137 | * | |
138 | * So __run_timers/migrate_timers can safely modify all timers which could | |
139 | * be found on the lists/queues. | |
140 | * | |
141 | * When the timer's base is locked, and the timer removed from list, it is | |
142 | * possible to set timer->base = NULL and drop the lock: the timer remains | |
143 | * locked. | |
144 | */ | |
145 | static struct hrtimer_base *lock_hrtimer_base(const struct hrtimer *timer, | |
146 | unsigned long *flags) | |
147 | { | |
148 | struct hrtimer_base *base; | |
149 | ||
150 | for (;;) { | |
151 | base = timer->base; | |
152 | if (likely(base != NULL)) { | |
153 | spin_lock_irqsave(&base->lock, *flags); | |
154 | if (likely(base == timer->base)) | |
155 | return base; | |
156 | /* The timer has migrated to another CPU: */ | |
157 | spin_unlock_irqrestore(&base->lock, *flags); | |
158 | } | |
159 | cpu_relax(); | |
160 | } | |
161 | } | |
162 | ||
163 | /* | |
164 | * Switch the timer base to the current CPU when possible. | |
165 | */ | |
166 | static inline struct hrtimer_base * | |
167 | switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_base *base) | |
168 | { | |
169 | struct hrtimer_base *new_base; | |
170 | ||
171 | new_base = &__get_cpu_var(hrtimer_bases[base->index]); | |
172 | ||
173 | if (base != new_base) { | |
174 | /* | |
175 | * We are trying to schedule the timer on the local CPU. | |
176 | * However we can't change timer's base while it is running, | |
177 | * so we keep it on the same CPU. No hassle vs. reprogramming | |
178 | * the event source in the high resolution case. The softirq | |
179 | * code will take care of this when the timer function has | |
180 | * completed. There is no conflict as we hold the lock until | |
181 | * the timer is enqueued. | |
182 | */ | |
183 | if (unlikely(base->curr_timer == timer)) | |
184 | return base; | |
185 | ||
186 | /* See the comment in lock_timer_base() */ | |
187 | timer->base = NULL; | |
188 | spin_unlock(&base->lock); | |
189 | spin_lock(&new_base->lock); | |
190 | timer->base = new_base; | |
191 | } | |
192 | return new_base; | |
193 | } | |
194 | ||
195 | #else /* CONFIG_SMP */ | |
196 | ||
197 | #define set_curr_timer(b, t) do { } while (0) | |
198 | ||
199 | static inline struct hrtimer_base * | |
200 | lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) | |
201 | { | |
202 | struct hrtimer_base *base = timer->base; | |
203 | ||
204 | spin_lock_irqsave(&base->lock, *flags); | |
205 | ||
206 | return base; | |
207 | } | |
208 | ||
209 | #define switch_hrtimer_base(t, b) (b) | |
210 | ||
211 | #endif /* !CONFIG_SMP */ | |
212 | ||
213 | /* | |
214 | * Functions for the union type storage format of ktime_t which are | |
215 | * too large for inlining: | |
216 | */ | |
217 | #if BITS_PER_LONG < 64 | |
218 | # ifndef CONFIG_KTIME_SCALAR | |
219 | /** | |
220 | * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable | |
221 | * | |
222 | * @kt: addend | |
223 | * @nsec: the scalar nsec value to add | |
224 | * | |
225 | * Returns the sum of kt and nsec in ktime_t format | |
226 | */ | |
227 | ktime_t ktime_add_ns(const ktime_t kt, u64 nsec) | |
228 | { | |
229 | ktime_t tmp; | |
230 | ||
231 | if (likely(nsec < NSEC_PER_SEC)) { | |
232 | tmp.tv64 = nsec; | |
233 | } else { | |
234 | unsigned long rem = do_div(nsec, NSEC_PER_SEC); | |
235 | ||
236 | tmp = ktime_set((long)nsec, rem); | |
237 | } | |
238 | ||
239 | return ktime_add(kt, tmp); | |
240 | } | |
241 | ||
242 | #else /* CONFIG_KTIME_SCALAR */ | |
243 | ||
244 | # endif /* !CONFIG_KTIME_SCALAR */ | |
245 | ||
246 | /* | |
247 | * Divide a ktime value by a nanosecond value | |
248 | */ | |
249 | static unsigned long ktime_divns(const ktime_t kt, nsec_t div) | |
250 | { | |
251 | u64 dclc, inc, dns; | |
252 | int sft = 0; | |
253 | ||
254 | dclc = dns = ktime_to_ns(kt); | |
255 | inc = div; | |
256 | /* Make sure the divisor is less than 2^32: */ | |
257 | while (div >> 32) { | |
258 | sft++; | |
259 | div >>= 1; | |
260 | } | |
261 | dclc >>= sft; | |
262 | do_div(dclc, (unsigned long) div); | |
263 | ||
264 | return (unsigned long) dclc; | |
265 | } | |
266 | ||
267 | #else /* BITS_PER_LONG < 64 */ | |
268 | # define ktime_divns(kt, div) (unsigned long)((kt).tv64 / (div)) | |
269 | #endif /* BITS_PER_LONG >= 64 */ | |
270 | ||
271 | /* | |
272 | * Counterpart to lock_timer_base above: | |
273 | */ | |
274 | static inline | |
275 | void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) | |
276 | { | |
277 | spin_unlock_irqrestore(&timer->base->lock, *flags); | |
278 | } | |
279 | ||
280 | /** | |
281 | * hrtimer_forward - forward the timer expiry | |
282 | * | |
283 | * @timer: hrtimer to forward | |
284 | * @interval: the interval to forward | |
285 | * | |
286 | * Forward the timer expiry so it will expire in the future. | |
8dca6f33 | 287 | * Returns the number of overruns. |
c0a31329 TG |
288 | */ |
289 | unsigned long | |
c9db4fa1 | 290 | hrtimer_forward(struct hrtimer *timer, ktime_t interval) |
c0a31329 TG |
291 | { |
292 | unsigned long orun = 1; | |
293 | ktime_t delta, now; | |
294 | ||
295 | now = timer->base->get_time(); | |
296 | ||
297 | delta = ktime_sub(now, timer->expires); | |
298 | ||
299 | if (delta.tv64 < 0) | |
300 | return 0; | |
301 | ||
c9db4fa1 TG |
302 | if (interval.tv64 < timer->base->resolution.tv64) |
303 | interval.tv64 = timer->base->resolution.tv64; | |
304 | ||
c0a31329 TG |
305 | if (unlikely(delta.tv64 >= interval.tv64)) { |
306 | nsec_t incr = ktime_to_ns(interval); | |
307 | ||
308 | orun = ktime_divns(delta, incr); | |
309 | timer->expires = ktime_add_ns(timer->expires, incr * orun); | |
310 | if (timer->expires.tv64 > now.tv64) | |
311 | return orun; | |
312 | /* | |
313 | * This (and the ktime_add() below) is the | |
314 | * correction for exact: | |
315 | */ | |
316 | orun++; | |
317 | } | |
318 | timer->expires = ktime_add(timer->expires, interval); | |
319 | ||
320 | return orun; | |
321 | } | |
322 | ||
323 | /* | |
324 | * enqueue_hrtimer - internal function to (re)start a timer | |
325 | * | |
326 | * The timer is inserted in expiry order. Insertion into the | |
327 | * red black tree is O(log(n)). Must hold the base lock. | |
328 | */ | |
329 | static void enqueue_hrtimer(struct hrtimer *timer, struct hrtimer_base *base) | |
330 | { | |
331 | struct rb_node **link = &base->active.rb_node; | |
c0a31329 TG |
332 | struct rb_node *parent = NULL; |
333 | struct hrtimer *entry; | |
334 | ||
335 | /* | |
336 | * Find the right place in the rbtree: | |
337 | */ | |
338 | while (*link) { | |
339 | parent = *link; | |
340 | entry = rb_entry(parent, struct hrtimer, node); | |
341 | /* | |
342 | * We dont care about collisions. Nodes with | |
343 | * the same expiry time stay together. | |
344 | */ | |
345 | if (timer->expires.tv64 < entry->expires.tv64) | |
346 | link = &(*link)->rb_left; | |
288867ec | 347 | else |
c0a31329 | 348 | link = &(*link)->rb_right; |
c0a31329 TG |
349 | } |
350 | ||
351 | /* | |
288867ec TG |
352 | * Insert the timer to the rbtree and check whether it |
353 | * replaces the first pending timer | |
c0a31329 TG |
354 | */ |
355 | rb_link_node(&timer->node, parent, link); | |
356 | rb_insert_color(&timer->node, &base->active); | |
c0a31329 TG |
357 | |
358 | timer->state = HRTIMER_PENDING; | |
c0a31329 | 359 | |
288867ec TG |
360 | if (!base->first || timer->expires.tv64 < |
361 | rb_entry(base->first, struct hrtimer, node)->expires.tv64) | |
362 | base->first = &timer->node; | |
363 | } | |
c0a31329 TG |
364 | |
365 | /* | |
366 | * __remove_hrtimer - internal function to remove a timer | |
367 | * | |
368 | * Caller must hold the base lock. | |
369 | */ | |
370 | static void __remove_hrtimer(struct hrtimer *timer, struct hrtimer_base *base) | |
371 | { | |
372 | /* | |
288867ec TG |
373 | * Remove the timer from the rbtree and replace the |
374 | * first entry pointer if necessary. | |
c0a31329 | 375 | */ |
288867ec TG |
376 | if (base->first == &timer->node) |
377 | base->first = rb_next(&timer->node); | |
c0a31329 TG |
378 | rb_erase(&timer->node, &base->active); |
379 | } | |
380 | ||
381 | /* | |
382 | * remove hrtimer, called with base lock held | |
383 | */ | |
384 | static inline int | |
385 | remove_hrtimer(struct hrtimer *timer, struct hrtimer_base *base) | |
386 | { | |
387 | if (hrtimer_active(timer)) { | |
388 | __remove_hrtimer(timer, base); | |
389 | timer->state = HRTIMER_INACTIVE; | |
390 | return 1; | |
391 | } | |
392 | return 0; | |
393 | } | |
394 | ||
395 | /** | |
396 | * hrtimer_start - (re)start an relative timer on the current CPU | |
397 | * | |
398 | * @timer: the timer to be added | |
399 | * @tim: expiry time | |
400 | * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL) | |
401 | * | |
402 | * Returns: | |
403 | * 0 on success | |
404 | * 1 when the timer was active | |
405 | */ | |
406 | int | |
407 | hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode) | |
408 | { | |
409 | struct hrtimer_base *base, *new_base; | |
410 | unsigned long flags; | |
411 | int ret; | |
412 | ||
413 | base = lock_hrtimer_base(timer, &flags); | |
414 | ||
415 | /* Remove an active timer from the queue: */ | |
416 | ret = remove_hrtimer(timer, base); | |
417 | ||
418 | /* Switch the timer base, if necessary: */ | |
419 | new_base = switch_hrtimer_base(timer, base); | |
420 | ||
421 | if (mode == HRTIMER_REL) | |
422 | tim = ktime_add(tim, new_base->get_time()); | |
423 | timer->expires = tim; | |
424 | ||
425 | enqueue_hrtimer(timer, new_base); | |
426 | ||
427 | unlock_hrtimer_base(timer, &flags); | |
428 | ||
429 | return ret; | |
430 | } | |
431 | ||
432 | /** | |
433 | * hrtimer_try_to_cancel - try to deactivate a timer | |
434 | * | |
435 | * @timer: hrtimer to stop | |
436 | * | |
437 | * Returns: | |
438 | * 0 when the timer was not active | |
439 | * 1 when the timer was active | |
440 | * -1 when the timer is currently excuting the callback function and | |
441 | * can not be stopped | |
442 | */ | |
443 | int hrtimer_try_to_cancel(struct hrtimer *timer) | |
444 | { | |
445 | struct hrtimer_base *base; | |
446 | unsigned long flags; | |
447 | int ret = -1; | |
448 | ||
449 | base = lock_hrtimer_base(timer, &flags); | |
450 | ||
451 | if (base->curr_timer != timer) | |
452 | ret = remove_hrtimer(timer, base); | |
453 | ||
454 | unlock_hrtimer_base(timer, &flags); | |
455 | ||
456 | return ret; | |
457 | ||
458 | } | |
459 | ||
460 | /** | |
461 | * hrtimer_cancel - cancel a timer and wait for the handler to finish. | |
462 | * | |
463 | * @timer: the timer to be cancelled | |
464 | * | |
465 | * Returns: | |
466 | * 0 when the timer was not active | |
467 | * 1 when the timer was active | |
468 | */ | |
469 | int hrtimer_cancel(struct hrtimer *timer) | |
470 | { | |
471 | for (;;) { | |
472 | int ret = hrtimer_try_to_cancel(timer); | |
473 | ||
474 | if (ret >= 0) | |
475 | return ret; | |
476 | } | |
477 | } | |
478 | ||
479 | /** | |
480 | * hrtimer_get_remaining - get remaining time for the timer | |
481 | * | |
482 | * @timer: the timer to read | |
483 | */ | |
484 | ktime_t hrtimer_get_remaining(const struct hrtimer *timer) | |
485 | { | |
486 | struct hrtimer_base *base; | |
487 | unsigned long flags; | |
488 | ktime_t rem; | |
489 | ||
490 | base = lock_hrtimer_base(timer, &flags); | |
491 | rem = ktime_sub(timer->expires, timer->base->get_time()); | |
492 | unlock_hrtimer_base(timer, &flags); | |
493 | ||
494 | return rem; | |
495 | } | |
496 | ||
497 | /** | |
7978672c | 498 | * hrtimer_init - initialize a timer to the given clock |
c0a31329 | 499 | * |
7978672c | 500 | * @timer: the timer to be initialized |
c0a31329 | 501 | * @clock_id: the clock to be used |
7978672c | 502 | * @mode: timer mode abs/rel |
c0a31329 | 503 | */ |
7978672c GA |
504 | void hrtimer_init(struct hrtimer *timer, clockid_t clock_id, |
505 | enum hrtimer_mode mode) | |
c0a31329 TG |
506 | { |
507 | struct hrtimer_base *bases; | |
508 | ||
7978672c GA |
509 | memset(timer, 0, sizeof(struct hrtimer)); |
510 | ||
c0a31329 | 511 | bases = per_cpu(hrtimer_bases, raw_smp_processor_id()); |
c0a31329 | 512 | |
7978672c GA |
513 | if (clock_id == CLOCK_REALTIME && mode != HRTIMER_ABS) |
514 | clock_id = CLOCK_MONOTONIC; | |
515 | ||
516 | timer->base = &bases[clock_id]; | |
c0a31329 TG |
517 | } |
518 | ||
519 | /** | |
520 | * hrtimer_get_res - get the timer resolution for a clock | |
521 | * | |
522 | * @which_clock: which clock to query | |
523 | * @tp: pointer to timespec variable to store the resolution | |
524 | * | |
525 | * Store the resolution of the clock selected by which_clock in the | |
526 | * variable pointed to by tp. | |
527 | */ | |
528 | int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp) | |
529 | { | |
530 | struct hrtimer_base *bases; | |
531 | ||
c0a31329 | 532 | bases = per_cpu(hrtimer_bases, raw_smp_processor_id()); |
e2787630 | 533 | *tp = ktime_to_timespec(bases[which_clock].resolution); |
c0a31329 TG |
534 | |
535 | return 0; | |
536 | } | |
537 | ||
538 | /* | |
539 | * Expire the per base hrtimer-queue: | |
540 | */ | |
541 | static inline void run_hrtimer_queue(struct hrtimer_base *base) | |
542 | { | |
543 | ktime_t now = base->get_time(); | |
288867ec | 544 | struct rb_node *node; |
c0a31329 TG |
545 | |
546 | spin_lock_irq(&base->lock); | |
547 | ||
288867ec | 548 | while ((node = base->first)) { |
c0a31329 TG |
549 | struct hrtimer *timer; |
550 | int (*fn)(void *); | |
551 | int restart; | |
552 | void *data; | |
553 | ||
288867ec | 554 | timer = rb_entry(node, struct hrtimer, node); |
c0a31329 TG |
555 | if (now.tv64 <= timer->expires.tv64) |
556 | break; | |
557 | ||
558 | fn = timer->function; | |
559 | data = timer->data; | |
560 | set_curr_timer(base, timer); | |
ff60a5dc | 561 | timer->state = HRTIMER_RUNNING; |
c0a31329 TG |
562 | __remove_hrtimer(timer, base); |
563 | spin_unlock_irq(&base->lock); | |
564 | ||
565 | /* | |
566 | * fn == NULL is special case for the simplest timer | |
567 | * variant - wake up process and do not restart: | |
568 | */ | |
569 | if (!fn) { | |
570 | wake_up_process(data); | |
571 | restart = HRTIMER_NORESTART; | |
572 | } else | |
573 | restart = fn(data); | |
574 | ||
575 | spin_lock_irq(&base->lock); | |
576 | ||
ff60a5dc | 577 | /* Another CPU has added back the timer */ |
578 | if (timer->state != HRTIMER_RUNNING) | |
579 | continue; | |
580 | ||
c0a31329 TG |
581 | if (restart == HRTIMER_RESTART) |
582 | enqueue_hrtimer(timer, base); | |
583 | else | |
584 | timer->state = HRTIMER_EXPIRED; | |
585 | } | |
586 | set_curr_timer(base, NULL); | |
587 | spin_unlock_irq(&base->lock); | |
588 | } | |
589 | ||
590 | /* | |
591 | * Called from timer softirq every jiffy, expire hrtimers: | |
592 | */ | |
593 | void hrtimer_run_queues(void) | |
594 | { | |
595 | struct hrtimer_base *base = __get_cpu_var(hrtimer_bases); | |
596 | int i; | |
597 | ||
598 | for (i = 0; i < MAX_HRTIMER_BASES; i++) | |
599 | run_hrtimer_queue(&base[i]); | |
600 | } | |
601 | ||
10c94ec1 TG |
602 | /* |
603 | * Sleep related functions: | |
604 | */ | |
605 | ||
606 | /** | |
607 | * schedule_hrtimer - sleep until timeout | |
608 | * | |
609 | * @timer: hrtimer variable initialized with the correct clock base | |
610 | * @mode: timeout value is abs/rel | |
611 | * | |
612 | * Make the current task sleep until @timeout is | |
613 | * elapsed. | |
614 | * | |
615 | * You can set the task state as follows - | |
616 | * | |
617 | * %TASK_UNINTERRUPTIBLE - at least @timeout is guaranteed to | |
618 | * pass before the routine returns. The routine will return 0 | |
619 | * | |
620 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
621 | * delivered to the current task. In this case the remaining time | |
622 | * will be returned | |
623 | * | |
624 | * The current task state is guaranteed to be TASK_RUNNING when this | |
625 | * routine returns. | |
626 | */ | |
627 | static ktime_t __sched | |
628 | schedule_hrtimer(struct hrtimer *timer, const enum hrtimer_mode mode) | |
629 | { | |
630 | /* fn stays NULL, meaning single-shot wakeup: */ | |
631 | timer->data = current; | |
632 | ||
633 | hrtimer_start(timer, timer->expires, mode); | |
634 | ||
635 | schedule(); | |
636 | hrtimer_cancel(timer); | |
637 | ||
638 | /* Return the remaining time: */ | |
639 | if (timer->state != HRTIMER_EXPIRED) | |
640 | return ktime_sub(timer->expires, timer->base->get_time()); | |
641 | else | |
642 | return (ktime_t) {.tv64 = 0 }; | |
643 | } | |
644 | ||
645 | static inline ktime_t __sched | |
646 | schedule_hrtimer_interruptible(struct hrtimer *timer, | |
647 | const enum hrtimer_mode mode) | |
648 | { | |
649 | set_current_state(TASK_INTERRUPTIBLE); | |
650 | ||
651 | return schedule_hrtimer(timer, mode); | |
652 | } | |
653 | ||
7978672c | 654 | static long __sched nanosleep_restart(struct restart_block *restart) |
10c94ec1 | 655 | { |
ea13dbc8 IM |
656 | struct timespec __user *rmtp; |
657 | struct timespec tu; | |
10c94ec1 TG |
658 | void *rfn_save = restart->fn; |
659 | struct hrtimer timer; | |
660 | ktime_t rem; | |
661 | ||
662 | restart->fn = do_no_restart_syscall; | |
663 | ||
7978672c | 664 | hrtimer_init(&timer, (clockid_t) restart->arg3, HRTIMER_ABS); |
10c94ec1 TG |
665 | |
666 | timer.expires.tv64 = ((u64)restart->arg1 << 32) | (u64) restart->arg0; | |
667 | ||
668 | rem = schedule_hrtimer_interruptible(&timer, HRTIMER_ABS); | |
669 | ||
670 | if (rem.tv64 <= 0) | |
671 | return 0; | |
672 | ||
673 | rmtp = (struct timespec __user *) restart->arg2; | |
674 | tu = ktime_to_timespec(rem); | |
675 | if (rmtp && copy_to_user(rmtp, &tu, sizeof(tu))) | |
676 | return -EFAULT; | |
677 | ||
678 | restart->fn = rfn_save; | |
679 | ||
680 | /* The other values in restart are already filled in */ | |
681 | return -ERESTART_RESTARTBLOCK; | |
682 | } | |
683 | ||
10c94ec1 TG |
684 | long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp, |
685 | const enum hrtimer_mode mode, const clockid_t clockid) | |
686 | { | |
687 | struct restart_block *restart; | |
688 | struct hrtimer timer; | |
689 | struct timespec tu; | |
690 | ktime_t rem; | |
691 | ||
7978672c | 692 | hrtimer_init(&timer, clockid, mode); |
10c94ec1 TG |
693 | |
694 | timer.expires = timespec_to_ktime(*rqtp); | |
695 | ||
696 | rem = schedule_hrtimer_interruptible(&timer, mode); | |
697 | if (rem.tv64 <= 0) | |
698 | return 0; | |
699 | ||
7978672c | 700 | /* Absolute timers do not update the rmtp value and restart: */ |
10c94ec1 TG |
701 | if (mode == HRTIMER_ABS) |
702 | return -ERESTARTNOHAND; | |
703 | ||
704 | tu = ktime_to_timespec(rem); | |
705 | ||
706 | if (rmtp && copy_to_user(rmtp, &tu, sizeof(tu))) | |
707 | return -EFAULT; | |
708 | ||
709 | restart = ¤t_thread_info()->restart_block; | |
7978672c | 710 | restart->fn = nanosleep_restart; |
10c94ec1 TG |
711 | restart->arg0 = timer.expires.tv64 & 0xFFFFFFFF; |
712 | restart->arg1 = timer.expires.tv64 >> 32; | |
713 | restart->arg2 = (unsigned long) rmtp; | |
7978672c | 714 | restart->arg3 = (unsigned long) timer.base->index; |
10c94ec1 TG |
715 | |
716 | return -ERESTART_RESTARTBLOCK; | |
717 | } | |
718 | ||
6ba1b912 TG |
719 | asmlinkage long |
720 | sys_nanosleep(struct timespec __user *rqtp, struct timespec __user *rmtp) | |
721 | { | |
722 | struct timespec tu; | |
723 | ||
724 | if (copy_from_user(&tu, rqtp, sizeof(tu))) | |
725 | return -EFAULT; | |
726 | ||
727 | if (!timespec_valid(&tu)) | |
728 | return -EINVAL; | |
729 | ||
730 | return hrtimer_nanosleep(&tu, rmtp, HRTIMER_REL, CLOCK_MONOTONIC); | |
731 | } | |
732 | ||
c0a31329 TG |
733 | /* |
734 | * Functions related to boot-time initialization: | |
735 | */ | |
736 | static void __devinit init_hrtimers_cpu(int cpu) | |
737 | { | |
738 | struct hrtimer_base *base = per_cpu(hrtimer_bases, cpu); | |
739 | int i; | |
740 | ||
7978672c | 741 | for (i = 0; i < MAX_HRTIMER_BASES; i++, base++) |
c0a31329 | 742 | spin_lock_init(&base->lock); |
c0a31329 TG |
743 | } |
744 | ||
745 | #ifdef CONFIG_HOTPLUG_CPU | |
746 | ||
747 | static void migrate_hrtimer_list(struct hrtimer_base *old_base, | |
748 | struct hrtimer_base *new_base) | |
749 | { | |
750 | struct hrtimer *timer; | |
751 | struct rb_node *node; | |
752 | ||
753 | while ((node = rb_first(&old_base->active))) { | |
754 | timer = rb_entry(node, struct hrtimer, node); | |
755 | __remove_hrtimer(timer, old_base); | |
756 | timer->base = new_base; | |
757 | enqueue_hrtimer(timer, new_base); | |
758 | } | |
759 | } | |
760 | ||
761 | static void migrate_hrtimers(int cpu) | |
762 | { | |
763 | struct hrtimer_base *old_base, *new_base; | |
764 | int i; | |
765 | ||
766 | BUG_ON(cpu_online(cpu)); | |
767 | old_base = per_cpu(hrtimer_bases, cpu); | |
768 | new_base = get_cpu_var(hrtimer_bases); | |
769 | ||
770 | local_irq_disable(); | |
771 | ||
772 | for (i = 0; i < MAX_HRTIMER_BASES; i++) { | |
773 | ||
774 | spin_lock(&new_base->lock); | |
775 | spin_lock(&old_base->lock); | |
776 | ||
777 | BUG_ON(old_base->curr_timer); | |
778 | ||
779 | migrate_hrtimer_list(old_base, new_base); | |
780 | ||
781 | spin_unlock(&old_base->lock); | |
782 | spin_unlock(&new_base->lock); | |
783 | old_base++; | |
784 | new_base++; | |
785 | } | |
786 | ||
787 | local_irq_enable(); | |
788 | put_cpu_var(hrtimer_bases); | |
789 | } | |
790 | #endif /* CONFIG_HOTPLUG_CPU */ | |
791 | ||
792 | static int __devinit hrtimer_cpu_notify(struct notifier_block *self, | |
793 | unsigned long action, void *hcpu) | |
794 | { | |
795 | long cpu = (long)hcpu; | |
796 | ||
797 | switch (action) { | |
798 | ||
799 | case CPU_UP_PREPARE: | |
800 | init_hrtimers_cpu(cpu); | |
801 | break; | |
802 | ||
803 | #ifdef CONFIG_HOTPLUG_CPU | |
804 | case CPU_DEAD: | |
805 | migrate_hrtimers(cpu); | |
806 | break; | |
807 | #endif | |
808 | ||
809 | default: | |
810 | break; | |
811 | } | |
812 | ||
813 | return NOTIFY_OK; | |
814 | } | |
815 | ||
816 | static struct notifier_block __devinitdata hrtimers_nb = { | |
817 | .notifier_call = hrtimer_cpu_notify, | |
818 | }; | |
819 | ||
820 | void __init hrtimers_init(void) | |
821 | { | |
822 | hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE, | |
823 | (void *)(long)smp_processor_id()); | |
824 | register_cpu_notifier(&hrtimers_nb); | |
825 | } | |
826 |