]>
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 | 124 | |
92127c7a TG |
125 | /* |
126 | * Get the coarse grained time at the softirq based on xtime and | |
127 | * wall_to_monotonic. | |
128 | */ | |
129 | static void hrtimer_get_softirq_time(struct hrtimer_base *base) | |
130 | { | |
131 | ktime_t xtim, tomono; | |
132 | unsigned long seq; | |
133 | ||
134 | do { | |
135 | seq = read_seqbegin(&xtime_lock); | |
136 | xtim = timespec_to_ktime(xtime); | |
137 | tomono = timespec_to_ktime(wall_to_monotonic); | |
138 | ||
139 | } while (read_seqretry(&xtime_lock, seq)); | |
140 | ||
141 | base[CLOCK_REALTIME].softirq_time = xtim; | |
142 | base[CLOCK_MONOTONIC].softirq_time = ktime_add(xtim, tomono); | |
143 | } | |
144 | ||
c0a31329 TG |
145 | /* |
146 | * Functions and macros which are different for UP/SMP systems are kept in a | |
147 | * single place | |
148 | */ | |
149 | #ifdef CONFIG_SMP | |
150 | ||
151 | #define set_curr_timer(b, t) do { (b)->curr_timer = (t); } while (0) | |
152 | ||
153 | /* | |
154 | * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock | |
155 | * means that all timers which are tied to this base via timer->base are | |
156 | * locked, and the base itself is locked too. | |
157 | * | |
158 | * So __run_timers/migrate_timers can safely modify all timers which could | |
159 | * be found on the lists/queues. | |
160 | * | |
161 | * When the timer's base is locked, and the timer removed from list, it is | |
162 | * possible to set timer->base = NULL and drop the lock: the timer remains | |
163 | * locked. | |
164 | */ | |
165 | static struct hrtimer_base *lock_hrtimer_base(const struct hrtimer *timer, | |
166 | unsigned long *flags) | |
167 | { | |
168 | struct hrtimer_base *base; | |
169 | ||
170 | for (;;) { | |
171 | base = timer->base; | |
172 | if (likely(base != NULL)) { | |
173 | spin_lock_irqsave(&base->lock, *flags); | |
174 | if (likely(base == timer->base)) | |
175 | return base; | |
176 | /* The timer has migrated to another CPU: */ | |
177 | spin_unlock_irqrestore(&base->lock, *flags); | |
178 | } | |
179 | cpu_relax(); | |
180 | } | |
181 | } | |
182 | ||
183 | /* | |
184 | * Switch the timer base to the current CPU when possible. | |
185 | */ | |
186 | static inline struct hrtimer_base * | |
187 | switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_base *base) | |
188 | { | |
189 | struct hrtimer_base *new_base; | |
190 | ||
191 | new_base = &__get_cpu_var(hrtimer_bases[base->index]); | |
192 | ||
193 | if (base != new_base) { | |
194 | /* | |
195 | * We are trying to schedule the timer on the local CPU. | |
196 | * However we can't change timer's base while it is running, | |
197 | * so we keep it on the same CPU. No hassle vs. reprogramming | |
198 | * the event source in the high resolution case. The softirq | |
199 | * code will take care of this when the timer function has | |
200 | * completed. There is no conflict as we hold the lock until | |
201 | * the timer is enqueued. | |
202 | */ | |
203 | if (unlikely(base->curr_timer == timer)) | |
204 | return base; | |
205 | ||
206 | /* See the comment in lock_timer_base() */ | |
207 | timer->base = NULL; | |
208 | spin_unlock(&base->lock); | |
209 | spin_lock(&new_base->lock); | |
210 | timer->base = new_base; | |
211 | } | |
212 | return new_base; | |
213 | } | |
214 | ||
215 | #else /* CONFIG_SMP */ | |
216 | ||
217 | #define set_curr_timer(b, t) do { } while (0) | |
218 | ||
219 | static inline struct hrtimer_base * | |
220 | lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) | |
221 | { | |
222 | struct hrtimer_base *base = timer->base; | |
223 | ||
224 | spin_lock_irqsave(&base->lock, *flags); | |
225 | ||
226 | return base; | |
227 | } | |
228 | ||
229 | #define switch_hrtimer_base(t, b) (b) | |
230 | ||
231 | #endif /* !CONFIG_SMP */ | |
232 | ||
233 | /* | |
234 | * Functions for the union type storage format of ktime_t which are | |
235 | * too large for inlining: | |
236 | */ | |
237 | #if BITS_PER_LONG < 64 | |
238 | # ifndef CONFIG_KTIME_SCALAR | |
239 | /** | |
240 | * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable | |
241 | * | |
242 | * @kt: addend | |
243 | * @nsec: the scalar nsec value to add | |
244 | * | |
245 | * Returns the sum of kt and nsec in ktime_t format | |
246 | */ | |
247 | ktime_t ktime_add_ns(const ktime_t kt, u64 nsec) | |
248 | { | |
249 | ktime_t tmp; | |
250 | ||
251 | if (likely(nsec < NSEC_PER_SEC)) { | |
252 | tmp.tv64 = nsec; | |
253 | } else { | |
254 | unsigned long rem = do_div(nsec, NSEC_PER_SEC); | |
255 | ||
256 | tmp = ktime_set((long)nsec, rem); | |
257 | } | |
258 | ||
259 | return ktime_add(kt, tmp); | |
260 | } | |
261 | ||
262 | #else /* CONFIG_KTIME_SCALAR */ | |
263 | ||
264 | # endif /* !CONFIG_KTIME_SCALAR */ | |
265 | ||
266 | /* | |
267 | * Divide a ktime value by a nanosecond value | |
268 | */ | |
df869b63 | 269 | static unsigned long ktime_divns(const ktime_t kt, s64 div) |
c0a31329 TG |
270 | { |
271 | u64 dclc, inc, dns; | |
272 | int sft = 0; | |
273 | ||
274 | dclc = dns = ktime_to_ns(kt); | |
275 | inc = div; | |
276 | /* Make sure the divisor is less than 2^32: */ | |
277 | while (div >> 32) { | |
278 | sft++; | |
279 | div >>= 1; | |
280 | } | |
281 | dclc >>= sft; | |
282 | do_div(dclc, (unsigned long) div); | |
283 | ||
284 | return (unsigned long) dclc; | |
285 | } | |
286 | ||
287 | #else /* BITS_PER_LONG < 64 */ | |
288 | # define ktime_divns(kt, div) (unsigned long)((kt).tv64 / (div)) | |
289 | #endif /* BITS_PER_LONG >= 64 */ | |
290 | ||
291 | /* | |
292 | * Counterpart to lock_timer_base above: | |
293 | */ | |
294 | static inline | |
295 | void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) | |
296 | { | |
297 | spin_unlock_irqrestore(&timer->base->lock, *flags); | |
298 | } | |
299 | ||
300 | /** | |
301 | * hrtimer_forward - forward the timer expiry | |
302 | * | |
303 | * @timer: hrtimer to forward | |
44f21475 | 304 | * @now: forward past this time |
c0a31329 TG |
305 | * @interval: the interval to forward |
306 | * | |
307 | * Forward the timer expiry so it will expire in the future. | |
8dca6f33 | 308 | * Returns the number of overruns. |
c0a31329 TG |
309 | */ |
310 | unsigned long | |
44f21475 | 311 | hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval) |
c0a31329 TG |
312 | { |
313 | unsigned long orun = 1; | |
44f21475 | 314 | ktime_t delta; |
c0a31329 TG |
315 | |
316 | delta = ktime_sub(now, timer->expires); | |
317 | ||
318 | if (delta.tv64 < 0) | |
319 | return 0; | |
320 | ||
c9db4fa1 TG |
321 | if (interval.tv64 < timer->base->resolution.tv64) |
322 | interval.tv64 = timer->base->resolution.tv64; | |
323 | ||
c0a31329 | 324 | if (unlikely(delta.tv64 >= interval.tv64)) { |
df869b63 | 325 | s64 incr = ktime_to_ns(interval); |
c0a31329 TG |
326 | |
327 | orun = ktime_divns(delta, incr); | |
328 | timer->expires = ktime_add_ns(timer->expires, incr * orun); | |
329 | if (timer->expires.tv64 > now.tv64) | |
330 | return orun; | |
331 | /* | |
332 | * This (and the ktime_add() below) is the | |
333 | * correction for exact: | |
334 | */ | |
335 | orun++; | |
336 | } | |
337 | timer->expires = ktime_add(timer->expires, interval); | |
338 | ||
339 | return orun; | |
340 | } | |
341 | ||
342 | /* | |
343 | * enqueue_hrtimer - internal function to (re)start a timer | |
344 | * | |
345 | * The timer is inserted in expiry order. Insertion into the | |
346 | * red black tree is O(log(n)). Must hold the base lock. | |
347 | */ | |
348 | static void enqueue_hrtimer(struct hrtimer *timer, struct hrtimer_base *base) | |
349 | { | |
350 | struct rb_node **link = &base->active.rb_node; | |
c0a31329 TG |
351 | struct rb_node *parent = NULL; |
352 | struct hrtimer *entry; | |
353 | ||
354 | /* | |
355 | * Find the right place in the rbtree: | |
356 | */ | |
357 | while (*link) { | |
358 | parent = *link; | |
359 | entry = rb_entry(parent, struct hrtimer, node); | |
360 | /* | |
361 | * We dont care about collisions. Nodes with | |
362 | * the same expiry time stay together. | |
363 | */ | |
364 | if (timer->expires.tv64 < entry->expires.tv64) | |
365 | link = &(*link)->rb_left; | |
288867ec | 366 | else |
c0a31329 | 367 | link = &(*link)->rb_right; |
c0a31329 TG |
368 | } |
369 | ||
370 | /* | |
288867ec TG |
371 | * Insert the timer to the rbtree and check whether it |
372 | * replaces the first pending timer | |
c0a31329 TG |
373 | */ |
374 | rb_link_node(&timer->node, parent, link); | |
375 | rb_insert_color(&timer->node, &base->active); | |
c0a31329 | 376 | |
288867ec TG |
377 | if (!base->first || timer->expires.tv64 < |
378 | rb_entry(base->first, struct hrtimer, node)->expires.tv64) | |
379 | base->first = &timer->node; | |
380 | } | |
c0a31329 TG |
381 | |
382 | /* | |
383 | * __remove_hrtimer - internal function to remove a timer | |
384 | * | |
385 | * Caller must hold the base lock. | |
386 | */ | |
387 | static void __remove_hrtimer(struct hrtimer *timer, struct hrtimer_base *base) | |
388 | { | |
389 | /* | |
288867ec TG |
390 | * Remove the timer from the rbtree and replace the |
391 | * first entry pointer if necessary. | |
c0a31329 | 392 | */ |
288867ec TG |
393 | if (base->first == &timer->node) |
394 | base->first = rb_next(&timer->node); | |
c0a31329 | 395 | rb_erase(&timer->node, &base->active); |
b75f7a51 | 396 | timer->node.rb_parent = HRTIMER_INACTIVE; |
c0a31329 TG |
397 | } |
398 | ||
399 | /* | |
400 | * remove hrtimer, called with base lock held | |
401 | */ | |
402 | static inline int | |
403 | remove_hrtimer(struct hrtimer *timer, struct hrtimer_base *base) | |
404 | { | |
405 | if (hrtimer_active(timer)) { | |
406 | __remove_hrtimer(timer, base); | |
c0a31329 TG |
407 | return 1; |
408 | } | |
409 | return 0; | |
410 | } | |
411 | ||
412 | /** | |
413 | * hrtimer_start - (re)start an relative timer on the current CPU | |
414 | * | |
415 | * @timer: the timer to be added | |
416 | * @tim: expiry time | |
417 | * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL) | |
418 | * | |
419 | * Returns: | |
420 | * 0 on success | |
421 | * 1 when the timer was active | |
422 | */ | |
423 | int | |
424 | hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode) | |
425 | { | |
426 | struct hrtimer_base *base, *new_base; | |
427 | unsigned long flags; | |
428 | int ret; | |
429 | ||
430 | base = lock_hrtimer_base(timer, &flags); | |
431 | ||
432 | /* Remove an active timer from the queue: */ | |
433 | ret = remove_hrtimer(timer, base); | |
434 | ||
435 | /* Switch the timer base, if necessary: */ | |
436 | new_base = switch_hrtimer_base(timer, base); | |
437 | ||
06027bdd | 438 | if (mode == HRTIMER_REL) { |
c0a31329 | 439 | tim = ktime_add(tim, new_base->get_time()); |
06027bdd IM |
440 | /* |
441 | * CONFIG_TIME_LOW_RES is a temporary way for architectures | |
442 | * to signal that they simply return xtime in | |
443 | * do_gettimeoffset(). In this case we want to round up by | |
444 | * resolution when starting a relative timer, to avoid short | |
445 | * timeouts. This will go away with the GTOD framework. | |
446 | */ | |
447 | #ifdef CONFIG_TIME_LOW_RES | |
448 | tim = ktime_add(tim, base->resolution); | |
449 | #endif | |
450 | } | |
c0a31329 TG |
451 | timer->expires = tim; |
452 | ||
453 | enqueue_hrtimer(timer, new_base); | |
454 | ||
455 | unlock_hrtimer_base(timer, &flags); | |
456 | ||
457 | return ret; | |
458 | } | |
459 | ||
460 | /** | |
461 | * hrtimer_try_to_cancel - try to deactivate a timer | |
462 | * | |
463 | * @timer: hrtimer to stop | |
464 | * | |
465 | * Returns: | |
466 | * 0 when the timer was not active | |
467 | * 1 when the timer was active | |
468 | * -1 when the timer is currently excuting the callback function and | |
469 | * can not be stopped | |
470 | */ | |
471 | int hrtimer_try_to_cancel(struct hrtimer *timer) | |
472 | { | |
473 | struct hrtimer_base *base; | |
474 | unsigned long flags; | |
475 | int ret = -1; | |
476 | ||
477 | base = lock_hrtimer_base(timer, &flags); | |
478 | ||
479 | if (base->curr_timer != timer) | |
480 | ret = remove_hrtimer(timer, base); | |
481 | ||
482 | unlock_hrtimer_base(timer, &flags); | |
483 | ||
484 | return ret; | |
485 | ||
486 | } | |
487 | ||
488 | /** | |
489 | * hrtimer_cancel - cancel a timer and wait for the handler to finish. | |
490 | * | |
491 | * @timer: the timer to be cancelled | |
492 | * | |
493 | * Returns: | |
494 | * 0 when the timer was not active | |
495 | * 1 when the timer was active | |
496 | */ | |
497 | int hrtimer_cancel(struct hrtimer *timer) | |
498 | { | |
499 | for (;;) { | |
500 | int ret = hrtimer_try_to_cancel(timer); | |
501 | ||
502 | if (ret >= 0) | |
503 | return ret; | |
504 | } | |
505 | } | |
506 | ||
507 | /** | |
508 | * hrtimer_get_remaining - get remaining time for the timer | |
509 | * | |
510 | * @timer: the timer to read | |
511 | */ | |
512 | ktime_t hrtimer_get_remaining(const struct hrtimer *timer) | |
513 | { | |
514 | struct hrtimer_base *base; | |
515 | unsigned long flags; | |
516 | ktime_t rem; | |
517 | ||
518 | base = lock_hrtimer_base(timer, &flags); | |
519 | rem = ktime_sub(timer->expires, timer->base->get_time()); | |
520 | unlock_hrtimer_base(timer, &flags); | |
521 | ||
522 | return rem; | |
523 | } | |
524 | ||
69239749 TL |
525 | #ifdef CONFIG_NO_IDLE_HZ |
526 | /** | |
527 | * hrtimer_get_next_event - get the time until next expiry event | |
528 | * | |
529 | * Returns the delta to the next expiry event or KTIME_MAX if no timer | |
530 | * is pending. | |
531 | */ | |
532 | ktime_t hrtimer_get_next_event(void) | |
533 | { | |
534 | struct hrtimer_base *base = __get_cpu_var(hrtimer_bases); | |
535 | ktime_t delta, mindelta = { .tv64 = KTIME_MAX }; | |
536 | unsigned long flags; | |
537 | int i; | |
538 | ||
539 | for (i = 0; i < MAX_HRTIMER_BASES; i++, base++) { | |
540 | struct hrtimer *timer; | |
541 | ||
542 | spin_lock_irqsave(&base->lock, flags); | |
543 | if (!base->first) { | |
544 | spin_unlock_irqrestore(&base->lock, flags); | |
545 | continue; | |
546 | } | |
547 | timer = rb_entry(base->first, struct hrtimer, node); | |
548 | delta.tv64 = timer->expires.tv64; | |
549 | spin_unlock_irqrestore(&base->lock, flags); | |
550 | delta = ktime_sub(delta, base->get_time()); | |
551 | if (delta.tv64 < mindelta.tv64) | |
552 | mindelta.tv64 = delta.tv64; | |
553 | } | |
554 | if (mindelta.tv64 < 0) | |
555 | mindelta.tv64 = 0; | |
556 | return mindelta; | |
557 | } | |
558 | #endif | |
559 | ||
c0a31329 | 560 | /** |
7978672c | 561 | * hrtimer_init - initialize a timer to the given clock |
c0a31329 | 562 | * |
7978672c | 563 | * @timer: the timer to be initialized |
c0a31329 | 564 | * @clock_id: the clock to be used |
7978672c | 565 | * @mode: timer mode abs/rel |
c0a31329 | 566 | */ |
7978672c GA |
567 | void hrtimer_init(struct hrtimer *timer, clockid_t clock_id, |
568 | enum hrtimer_mode mode) | |
c0a31329 TG |
569 | { |
570 | struct hrtimer_base *bases; | |
571 | ||
7978672c GA |
572 | memset(timer, 0, sizeof(struct hrtimer)); |
573 | ||
c0a31329 | 574 | bases = per_cpu(hrtimer_bases, raw_smp_processor_id()); |
c0a31329 | 575 | |
7978672c GA |
576 | if (clock_id == CLOCK_REALTIME && mode != HRTIMER_ABS) |
577 | clock_id = CLOCK_MONOTONIC; | |
578 | ||
579 | timer->base = &bases[clock_id]; | |
b75f7a51 | 580 | timer->node.rb_parent = HRTIMER_INACTIVE; |
c0a31329 TG |
581 | } |
582 | ||
583 | /** | |
584 | * hrtimer_get_res - get the timer resolution for a clock | |
585 | * | |
586 | * @which_clock: which clock to query | |
587 | * @tp: pointer to timespec variable to store the resolution | |
588 | * | |
589 | * Store the resolution of the clock selected by which_clock in the | |
590 | * variable pointed to by tp. | |
591 | */ | |
592 | int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp) | |
593 | { | |
594 | struct hrtimer_base *bases; | |
595 | ||
c0a31329 | 596 | bases = per_cpu(hrtimer_bases, raw_smp_processor_id()); |
e2787630 | 597 | *tp = ktime_to_timespec(bases[which_clock].resolution); |
c0a31329 TG |
598 | |
599 | return 0; | |
600 | } | |
601 | ||
602 | /* | |
603 | * Expire the per base hrtimer-queue: | |
604 | */ | |
605 | static inline void run_hrtimer_queue(struct hrtimer_base *base) | |
606 | { | |
288867ec | 607 | struct rb_node *node; |
c0a31329 | 608 | |
92127c7a TG |
609 | if (base->get_softirq_time) |
610 | base->softirq_time = base->get_softirq_time(); | |
611 | ||
c0a31329 TG |
612 | spin_lock_irq(&base->lock); |
613 | ||
288867ec | 614 | while ((node = base->first)) { |
c0a31329 | 615 | struct hrtimer *timer; |
05cfb614 | 616 | int (*fn)(struct hrtimer *); |
c0a31329 | 617 | int restart; |
c0a31329 | 618 | |
288867ec | 619 | timer = rb_entry(node, struct hrtimer, node); |
92127c7a | 620 | if (base->softirq_time.tv64 <= timer->expires.tv64) |
c0a31329 TG |
621 | break; |
622 | ||
623 | fn = timer->function; | |
c0a31329 TG |
624 | set_curr_timer(base, timer); |
625 | __remove_hrtimer(timer, base); | |
626 | spin_unlock_irq(&base->lock); | |
627 | ||
05cfb614 | 628 | restart = fn(timer); |
c0a31329 TG |
629 | |
630 | spin_lock_irq(&base->lock); | |
631 | ||
b75f7a51 RZ |
632 | if (restart != HRTIMER_NORESTART) { |
633 | BUG_ON(hrtimer_active(timer)); | |
c0a31329 | 634 | enqueue_hrtimer(timer, base); |
b75f7a51 | 635 | } |
c0a31329 TG |
636 | } |
637 | set_curr_timer(base, NULL); | |
638 | spin_unlock_irq(&base->lock); | |
639 | } | |
640 | ||
641 | /* | |
642 | * Called from timer softirq every jiffy, expire hrtimers: | |
643 | */ | |
644 | void hrtimer_run_queues(void) | |
645 | { | |
646 | struct hrtimer_base *base = __get_cpu_var(hrtimer_bases); | |
647 | int i; | |
648 | ||
92127c7a TG |
649 | hrtimer_get_softirq_time(base); |
650 | ||
c0a31329 TG |
651 | for (i = 0; i < MAX_HRTIMER_BASES; i++) |
652 | run_hrtimer_queue(&base[i]); | |
653 | } | |
654 | ||
10c94ec1 TG |
655 | /* |
656 | * Sleep related functions: | |
657 | */ | |
658 | ||
00362e33 TG |
659 | static int hrtimer_wakeup(struct hrtimer *timer) |
660 | { | |
661 | struct hrtimer_sleeper *t = | |
662 | container_of(timer, struct hrtimer_sleeper, timer); | |
663 | struct task_struct *task = t->task; | |
664 | ||
665 | t->task = NULL; | |
666 | if (task) | |
667 | wake_up_process(task); | |
668 | ||
669 | return HRTIMER_NORESTART; | |
670 | } | |
671 | ||
672 | void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, task_t *task) | |
673 | { | |
674 | sl->timer.function = hrtimer_wakeup; | |
675 | sl->task = task; | |
676 | } | |
677 | ||
432569bb RZ |
678 | struct sleep_hrtimer { |
679 | struct hrtimer timer; | |
680 | struct task_struct *task; | |
681 | int expired; | |
682 | }; | |
10c94ec1 | 683 | |
05cfb614 | 684 | static int nanosleep_wakeup(struct hrtimer *timer) |
432569bb | 685 | { |
05cfb614 RZ |
686 | struct sleep_hrtimer *t = |
687 | container_of(timer, struct sleep_hrtimer, timer); | |
10c94ec1 | 688 | |
432569bb RZ |
689 | t->expired = 1; |
690 | wake_up_process(t->task); | |
10c94ec1 | 691 | |
432569bb | 692 | return HRTIMER_NORESTART; |
10c94ec1 TG |
693 | } |
694 | ||
432569bb | 695 | static int __sched do_nanosleep(struct sleep_hrtimer *t, enum hrtimer_mode mode) |
10c94ec1 | 696 | { |
432569bb | 697 | t->timer.function = nanosleep_wakeup; |
432569bb RZ |
698 | t->task = current; |
699 | t->expired = 0; | |
10c94ec1 | 700 | |
432569bb RZ |
701 | do { |
702 | set_current_state(TASK_INTERRUPTIBLE); | |
703 | hrtimer_start(&t->timer, t->timer.expires, mode); | |
704 | ||
705 | schedule(); | |
706 | ||
707 | if (unlikely(!t->expired)) { | |
708 | hrtimer_cancel(&t->timer); | |
709 | mode = HRTIMER_ABS; | |
710 | } | |
711 | } while (!t->expired && !signal_pending(current)); | |
712 | ||
713 | return t->expired; | |
10c94ec1 TG |
714 | } |
715 | ||
7978672c | 716 | static long __sched nanosleep_restart(struct restart_block *restart) |
10c94ec1 | 717 | { |
432569bb | 718 | struct sleep_hrtimer t; |
ea13dbc8 IM |
719 | struct timespec __user *rmtp; |
720 | struct timespec tu; | |
432569bb | 721 | ktime_t time; |
10c94ec1 TG |
722 | |
723 | restart->fn = do_no_restart_syscall; | |
724 | ||
432569bb RZ |
725 | hrtimer_init(&t.timer, restart->arg3, HRTIMER_ABS); |
726 | t.timer.expires.tv64 = ((u64)restart->arg1 << 32) | (u64) restart->arg0; | |
10c94ec1 | 727 | |
432569bb | 728 | if (do_nanosleep(&t, HRTIMER_ABS)) |
10c94ec1 TG |
729 | return 0; |
730 | ||
731 | rmtp = (struct timespec __user *) restart->arg2; | |
432569bb RZ |
732 | if (rmtp) { |
733 | time = ktime_sub(t.timer.expires, t.timer.base->get_time()); | |
734 | if (time.tv64 <= 0) | |
735 | return 0; | |
736 | tu = ktime_to_timespec(time); | |
737 | if (copy_to_user(rmtp, &tu, sizeof(tu))) | |
738 | return -EFAULT; | |
739 | } | |
10c94ec1 | 740 | |
432569bb | 741 | restart->fn = nanosleep_restart; |
10c94ec1 TG |
742 | |
743 | /* The other values in restart are already filled in */ | |
744 | return -ERESTART_RESTARTBLOCK; | |
745 | } | |
746 | ||
10c94ec1 TG |
747 | long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp, |
748 | const enum hrtimer_mode mode, const clockid_t clockid) | |
749 | { | |
750 | struct restart_block *restart; | |
432569bb | 751 | struct sleep_hrtimer t; |
10c94ec1 TG |
752 | struct timespec tu; |
753 | ktime_t rem; | |
754 | ||
432569bb RZ |
755 | hrtimer_init(&t.timer, clockid, mode); |
756 | t.timer.expires = timespec_to_ktime(*rqtp); | |
757 | if (do_nanosleep(&t, mode)) | |
10c94ec1 TG |
758 | return 0; |
759 | ||
7978672c | 760 | /* Absolute timers do not update the rmtp value and restart: */ |
10c94ec1 TG |
761 | if (mode == HRTIMER_ABS) |
762 | return -ERESTARTNOHAND; | |
763 | ||
432569bb RZ |
764 | if (rmtp) { |
765 | rem = ktime_sub(t.timer.expires, t.timer.base->get_time()); | |
766 | if (rem.tv64 <= 0) | |
767 | return 0; | |
768 | tu = ktime_to_timespec(rem); | |
769 | if (copy_to_user(rmtp, &tu, sizeof(tu))) | |
770 | return -EFAULT; | |
771 | } | |
10c94ec1 TG |
772 | |
773 | restart = ¤t_thread_info()->restart_block; | |
7978672c | 774 | restart->fn = nanosleep_restart; |
432569bb RZ |
775 | restart->arg0 = t.timer.expires.tv64 & 0xFFFFFFFF; |
776 | restart->arg1 = t.timer.expires.tv64 >> 32; | |
10c94ec1 | 777 | restart->arg2 = (unsigned long) rmtp; |
432569bb | 778 | restart->arg3 = (unsigned long) t.timer.base->index; |
10c94ec1 TG |
779 | |
780 | return -ERESTART_RESTARTBLOCK; | |
781 | } | |
782 | ||
6ba1b912 TG |
783 | asmlinkage long |
784 | sys_nanosleep(struct timespec __user *rqtp, struct timespec __user *rmtp) | |
785 | { | |
786 | struct timespec tu; | |
787 | ||
788 | if (copy_from_user(&tu, rqtp, sizeof(tu))) | |
789 | return -EFAULT; | |
790 | ||
791 | if (!timespec_valid(&tu)) | |
792 | return -EINVAL; | |
793 | ||
794 | return hrtimer_nanosleep(&tu, rmtp, HRTIMER_REL, CLOCK_MONOTONIC); | |
795 | } | |
796 | ||
c0a31329 TG |
797 | /* |
798 | * Functions related to boot-time initialization: | |
799 | */ | |
800 | static void __devinit init_hrtimers_cpu(int cpu) | |
801 | { | |
802 | struct hrtimer_base *base = per_cpu(hrtimer_bases, cpu); | |
803 | int i; | |
804 | ||
7978672c | 805 | for (i = 0; i < MAX_HRTIMER_BASES; i++, base++) |
c0a31329 | 806 | spin_lock_init(&base->lock); |
c0a31329 TG |
807 | } |
808 | ||
809 | #ifdef CONFIG_HOTPLUG_CPU | |
810 | ||
811 | static void migrate_hrtimer_list(struct hrtimer_base *old_base, | |
812 | struct hrtimer_base *new_base) | |
813 | { | |
814 | struct hrtimer *timer; | |
815 | struct rb_node *node; | |
816 | ||
817 | while ((node = rb_first(&old_base->active))) { | |
818 | timer = rb_entry(node, struct hrtimer, node); | |
819 | __remove_hrtimer(timer, old_base); | |
820 | timer->base = new_base; | |
821 | enqueue_hrtimer(timer, new_base); | |
822 | } | |
823 | } | |
824 | ||
825 | static void migrate_hrtimers(int cpu) | |
826 | { | |
827 | struct hrtimer_base *old_base, *new_base; | |
828 | int i; | |
829 | ||
830 | BUG_ON(cpu_online(cpu)); | |
831 | old_base = per_cpu(hrtimer_bases, cpu); | |
832 | new_base = get_cpu_var(hrtimer_bases); | |
833 | ||
834 | local_irq_disable(); | |
835 | ||
836 | for (i = 0; i < MAX_HRTIMER_BASES; i++) { | |
837 | ||
838 | spin_lock(&new_base->lock); | |
839 | spin_lock(&old_base->lock); | |
840 | ||
841 | BUG_ON(old_base->curr_timer); | |
842 | ||
843 | migrate_hrtimer_list(old_base, new_base); | |
844 | ||
845 | spin_unlock(&old_base->lock); | |
846 | spin_unlock(&new_base->lock); | |
847 | old_base++; | |
848 | new_base++; | |
849 | } | |
850 | ||
851 | local_irq_enable(); | |
852 | put_cpu_var(hrtimer_bases); | |
853 | } | |
854 | #endif /* CONFIG_HOTPLUG_CPU */ | |
855 | ||
856 | static int __devinit hrtimer_cpu_notify(struct notifier_block *self, | |
857 | unsigned long action, void *hcpu) | |
858 | { | |
859 | long cpu = (long)hcpu; | |
860 | ||
861 | switch (action) { | |
862 | ||
863 | case CPU_UP_PREPARE: | |
864 | init_hrtimers_cpu(cpu); | |
865 | break; | |
866 | ||
867 | #ifdef CONFIG_HOTPLUG_CPU | |
868 | case CPU_DEAD: | |
869 | migrate_hrtimers(cpu); | |
870 | break; | |
871 | #endif | |
872 | ||
873 | default: | |
874 | break; | |
875 | } | |
876 | ||
877 | return NOTIFY_OK; | |
878 | } | |
879 | ||
880 | static struct notifier_block __devinitdata hrtimers_nb = { | |
881 | .notifier_call = hrtimer_cpu_notify, | |
882 | }; | |
883 | ||
884 | void __init hrtimers_init(void) | |
885 | { | |
886 | hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE, | |
887 | (void *)(long)smp_processor_id()); | |
888 | register_cpu_notifier(&hrtimers_nb); | |
889 | } | |
890 |