]>
Commit | Line | Data |
---|---|---|
c0a31329 TG |
1 | /* |
2 | * linux/kernel/hrtimer.c | |
3 | * | |
3c8aa39d TG |
4 | * Copyright(C) 2005-2006, Thomas Gleixner <tglx@linutronix.de> |
5 | * Copyright(C) 2005-2006, Red Hat, Inc., Ingo Molnar | |
6 | * Copyright(C) 2006 Timesys Corp., Thomas Gleixner <tglx@timesys.com> | |
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> | |
40 | #include <linux/interrupt.h> | |
41 | ||
42 | #include <asm/uaccess.h> | |
43 | ||
44 | /** | |
45 | * ktime_get - get the monotonic time in ktime_t format | |
46 | * | |
47 | * returns the time in ktime_t format | |
48 | */ | |
d316c57f | 49 | ktime_t ktime_get(void) |
c0a31329 TG |
50 | { |
51 | struct timespec now; | |
52 | ||
53 | ktime_get_ts(&now); | |
54 | ||
55 | return timespec_to_ktime(now); | |
56 | } | |
57 | ||
58 | /** | |
59 | * ktime_get_real - get the real (wall-) time in ktime_t format | |
60 | * | |
61 | * returns the time in ktime_t format | |
62 | */ | |
d316c57f | 63 | ktime_t ktime_get_real(void) |
c0a31329 TG |
64 | { |
65 | struct timespec now; | |
66 | ||
67 | getnstimeofday(&now); | |
68 | ||
69 | return timespec_to_ktime(now); | |
70 | } | |
71 | ||
72 | EXPORT_SYMBOL_GPL(ktime_get_real); | |
73 | ||
74 | /* | |
75 | * The timer bases: | |
7978672c GA |
76 | * |
77 | * Note: If we want to add new timer bases, we have to skip the two | |
78 | * clock ids captured by the cpu-timers. We do this by holding empty | |
79 | * entries rather than doing math adjustment of the clock ids. | |
80 | * This ensures that we capture erroneous accesses to these clock ids | |
81 | * rather than moving them into the range of valid clock id's. | |
c0a31329 | 82 | */ |
3c8aa39d | 83 | static DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) = |
c0a31329 | 84 | { |
3c8aa39d TG |
85 | |
86 | .clock_base = | |
c0a31329 | 87 | { |
3c8aa39d TG |
88 | { |
89 | .index = CLOCK_REALTIME, | |
90 | .get_time = &ktime_get_real, | |
91 | .resolution = KTIME_REALTIME_RES, | |
92 | }, | |
93 | { | |
94 | .index = CLOCK_MONOTONIC, | |
95 | .get_time = &ktime_get, | |
96 | .resolution = KTIME_MONOTONIC_RES, | |
97 | }, | |
98 | } | |
c0a31329 TG |
99 | }; |
100 | ||
101 | /** | |
102 | * ktime_get_ts - get the monotonic clock in timespec format | |
c0a31329 TG |
103 | * @ts: pointer to timespec variable |
104 | * | |
105 | * The function calculates the monotonic clock from the realtime | |
106 | * clock and the wall_to_monotonic offset and stores the result | |
72fd4a35 | 107 | * in normalized timespec format in the variable pointed to by @ts. |
c0a31329 TG |
108 | */ |
109 | void ktime_get_ts(struct timespec *ts) | |
110 | { | |
111 | struct timespec tomono; | |
112 | unsigned long seq; | |
113 | ||
114 | do { | |
115 | seq = read_seqbegin(&xtime_lock); | |
116 | getnstimeofday(ts); | |
117 | tomono = wall_to_monotonic; | |
118 | ||
119 | } while (read_seqretry(&xtime_lock, seq)); | |
120 | ||
121 | set_normalized_timespec(ts, ts->tv_sec + tomono.tv_sec, | |
122 | ts->tv_nsec + tomono.tv_nsec); | |
123 | } | |
69778e32 | 124 | EXPORT_SYMBOL_GPL(ktime_get_ts); |
c0a31329 | 125 | |
92127c7a TG |
126 | /* |
127 | * Get the coarse grained time at the softirq based on xtime and | |
128 | * wall_to_monotonic. | |
129 | */ | |
3c8aa39d | 130 | static void hrtimer_get_softirq_time(struct hrtimer_cpu_base *base) |
92127c7a TG |
131 | { |
132 | ktime_t xtim, tomono; | |
f4304ab2 | 133 | struct timespec xts; |
92127c7a TG |
134 | unsigned long seq; |
135 | ||
136 | do { | |
137 | seq = read_seqbegin(&xtime_lock); | |
f4304ab2 JS |
138 | #ifdef CONFIG_NO_HZ |
139 | getnstimeofday(&xts); | |
140 | #else | |
141 | xts = xtime; | |
142 | #endif | |
92127c7a TG |
143 | } while (read_seqretry(&xtime_lock, seq)); |
144 | ||
f4304ab2 JS |
145 | xtim = timespec_to_ktime(xts); |
146 | tomono = timespec_to_ktime(wall_to_monotonic); | |
3c8aa39d TG |
147 | base->clock_base[CLOCK_REALTIME].softirq_time = xtim; |
148 | base->clock_base[CLOCK_MONOTONIC].softirq_time = | |
149 | ktime_add(xtim, tomono); | |
92127c7a TG |
150 | } |
151 | ||
303e967f TG |
152 | /* |
153 | * Helper function to check, whether the timer is on one of the queues | |
154 | */ | |
155 | static inline int hrtimer_is_queued(struct hrtimer *timer) | |
156 | { | |
157 | return timer->state & HRTIMER_STATE_ENQUEUED; | |
158 | } | |
159 | ||
160 | /* | |
161 | * Helper function to check, whether the timer is running the callback | |
162 | * function | |
163 | */ | |
164 | static inline int hrtimer_callback_running(struct hrtimer *timer) | |
165 | { | |
166 | return timer->state & HRTIMER_STATE_CALLBACK; | |
167 | } | |
168 | ||
c0a31329 TG |
169 | /* |
170 | * Functions and macros which are different for UP/SMP systems are kept in a | |
171 | * single place | |
172 | */ | |
173 | #ifdef CONFIG_SMP | |
174 | ||
c0a31329 TG |
175 | /* |
176 | * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock | |
177 | * means that all timers which are tied to this base via timer->base are | |
178 | * locked, and the base itself is locked too. | |
179 | * | |
180 | * So __run_timers/migrate_timers can safely modify all timers which could | |
181 | * be found on the lists/queues. | |
182 | * | |
183 | * When the timer's base is locked, and the timer removed from list, it is | |
184 | * possible to set timer->base = NULL and drop the lock: the timer remains | |
185 | * locked. | |
186 | */ | |
3c8aa39d TG |
187 | static |
188 | struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer, | |
189 | unsigned long *flags) | |
c0a31329 | 190 | { |
3c8aa39d | 191 | struct hrtimer_clock_base *base; |
c0a31329 TG |
192 | |
193 | for (;;) { | |
194 | base = timer->base; | |
195 | if (likely(base != NULL)) { | |
3c8aa39d | 196 | spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
197 | if (likely(base == timer->base)) |
198 | return base; | |
199 | /* The timer has migrated to another CPU: */ | |
3c8aa39d | 200 | spin_unlock_irqrestore(&base->cpu_base->lock, *flags); |
c0a31329 TG |
201 | } |
202 | cpu_relax(); | |
203 | } | |
204 | } | |
205 | ||
206 | /* | |
207 | * Switch the timer base to the current CPU when possible. | |
208 | */ | |
3c8aa39d TG |
209 | static inline struct hrtimer_clock_base * |
210 | switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base) | |
c0a31329 | 211 | { |
3c8aa39d TG |
212 | struct hrtimer_clock_base *new_base; |
213 | struct hrtimer_cpu_base *new_cpu_base; | |
c0a31329 | 214 | |
3c8aa39d TG |
215 | new_cpu_base = &__get_cpu_var(hrtimer_bases); |
216 | new_base = &new_cpu_base->clock_base[base->index]; | |
c0a31329 TG |
217 | |
218 | if (base != new_base) { | |
219 | /* | |
220 | * We are trying to schedule the timer on the local CPU. | |
221 | * However we can't change timer's base while it is running, | |
222 | * so we keep it on the same CPU. No hassle vs. reprogramming | |
223 | * the event source in the high resolution case. The softirq | |
224 | * code will take care of this when the timer function has | |
225 | * completed. There is no conflict as we hold the lock until | |
226 | * the timer is enqueued. | |
227 | */ | |
5cfb6de7 | 228 | if (unlikely(timer->state & HRTIMER_STATE_CALLBACK)) |
c0a31329 TG |
229 | return base; |
230 | ||
231 | /* See the comment in lock_timer_base() */ | |
232 | timer->base = NULL; | |
3c8aa39d TG |
233 | spin_unlock(&base->cpu_base->lock); |
234 | spin_lock(&new_base->cpu_base->lock); | |
c0a31329 TG |
235 | timer->base = new_base; |
236 | } | |
237 | return new_base; | |
238 | } | |
239 | ||
240 | #else /* CONFIG_SMP */ | |
241 | ||
3c8aa39d | 242 | static inline struct hrtimer_clock_base * |
c0a31329 TG |
243 | lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) |
244 | { | |
3c8aa39d | 245 | struct hrtimer_clock_base *base = timer->base; |
c0a31329 | 246 | |
3c8aa39d | 247 | spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
248 | |
249 | return base; | |
250 | } | |
251 | ||
252 | #define switch_hrtimer_base(t, b) (b) | |
253 | ||
254 | #endif /* !CONFIG_SMP */ | |
255 | ||
256 | /* | |
257 | * Functions for the union type storage format of ktime_t which are | |
258 | * too large for inlining: | |
259 | */ | |
260 | #if BITS_PER_LONG < 64 | |
261 | # ifndef CONFIG_KTIME_SCALAR | |
262 | /** | |
263 | * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable | |
c0a31329 TG |
264 | * @kt: addend |
265 | * @nsec: the scalar nsec value to add | |
266 | * | |
267 | * Returns the sum of kt and nsec in ktime_t format | |
268 | */ | |
269 | ktime_t ktime_add_ns(const ktime_t kt, u64 nsec) | |
270 | { | |
271 | ktime_t tmp; | |
272 | ||
273 | if (likely(nsec < NSEC_PER_SEC)) { | |
274 | tmp.tv64 = nsec; | |
275 | } else { | |
276 | unsigned long rem = do_div(nsec, NSEC_PER_SEC); | |
277 | ||
278 | tmp = ktime_set((long)nsec, rem); | |
279 | } | |
280 | ||
281 | return ktime_add(kt, tmp); | |
282 | } | |
283 | ||
284 | #else /* CONFIG_KTIME_SCALAR */ | |
285 | ||
286 | # endif /* !CONFIG_KTIME_SCALAR */ | |
287 | ||
288 | /* | |
289 | * Divide a ktime value by a nanosecond value | |
290 | */ | |
df869b63 | 291 | static unsigned long ktime_divns(const ktime_t kt, s64 div) |
c0a31329 TG |
292 | { |
293 | u64 dclc, inc, dns; | |
294 | int sft = 0; | |
295 | ||
296 | dclc = dns = ktime_to_ns(kt); | |
297 | inc = div; | |
298 | /* Make sure the divisor is less than 2^32: */ | |
299 | while (div >> 32) { | |
300 | sft++; | |
301 | div >>= 1; | |
302 | } | |
303 | dclc >>= sft; | |
304 | do_div(dclc, (unsigned long) div); | |
305 | ||
306 | return (unsigned long) dclc; | |
307 | } | |
308 | ||
309 | #else /* BITS_PER_LONG < 64 */ | |
310 | # define ktime_divns(kt, div) (unsigned long)((kt).tv64 / (div)) | |
311 | #endif /* BITS_PER_LONG >= 64 */ | |
312 | ||
313 | /* | |
314 | * Counterpart to lock_timer_base above: | |
315 | */ | |
316 | static inline | |
317 | void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) | |
318 | { | |
3c8aa39d | 319 | spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags); |
c0a31329 TG |
320 | } |
321 | ||
322 | /** | |
323 | * hrtimer_forward - forward the timer expiry | |
c0a31329 | 324 | * @timer: hrtimer to forward |
44f21475 | 325 | * @now: forward past this time |
c0a31329 TG |
326 | * @interval: the interval to forward |
327 | * | |
328 | * Forward the timer expiry so it will expire in the future. | |
8dca6f33 | 329 | * Returns the number of overruns. |
c0a31329 TG |
330 | */ |
331 | unsigned long | |
44f21475 | 332 | hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval) |
c0a31329 TG |
333 | { |
334 | unsigned long orun = 1; | |
44f21475 | 335 | ktime_t delta; |
c0a31329 TG |
336 | |
337 | delta = ktime_sub(now, timer->expires); | |
338 | ||
339 | if (delta.tv64 < 0) | |
340 | return 0; | |
341 | ||
c9db4fa1 TG |
342 | if (interval.tv64 < timer->base->resolution.tv64) |
343 | interval.tv64 = timer->base->resolution.tv64; | |
344 | ||
c0a31329 | 345 | if (unlikely(delta.tv64 >= interval.tv64)) { |
df869b63 | 346 | s64 incr = ktime_to_ns(interval); |
c0a31329 TG |
347 | |
348 | orun = ktime_divns(delta, incr); | |
349 | timer->expires = ktime_add_ns(timer->expires, incr * orun); | |
350 | if (timer->expires.tv64 > now.tv64) | |
351 | return orun; | |
352 | /* | |
353 | * This (and the ktime_add() below) is the | |
354 | * correction for exact: | |
355 | */ | |
356 | orun++; | |
357 | } | |
358 | timer->expires = ktime_add(timer->expires, interval); | |
359 | ||
360 | return orun; | |
361 | } | |
362 | ||
363 | /* | |
364 | * enqueue_hrtimer - internal function to (re)start a timer | |
365 | * | |
366 | * The timer is inserted in expiry order. Insertion into the | |
367 | * red black tree is O(log(n)). Must hold the base lock. | |
368 | */ | |
3c8aa39d TG |
369 | static void enqueue_hrtimer(struct hrtimer *timer, |
370 | struct hrtimer_clock_base *base) | |
c0a31329 TG |
371 | { |
372 | struct rb_node **link = &base->active.rb_node; | |
c0a31329 TG |
373 | struct rb_node *parent = NULL; |
374 | struct hrtimer *entry; | |
375 | ||
376 | /* | |
377 | * Find the right place in the rbtree: | |
378 | */ | |
379 | while (*link) { | |
380 | parent = *link; | |
381 | entry = rb_entry(parent, struct hrtimer, node); | |
382 | /* | |
383 | * We dont care about collisions. Nodes with | |
384 | * the same expiry time stay together. | |
385 | */ | |
386 | if (timer->expires.tv64 < entry->expires.tv64) | |
387 | link = &(*link)->rb_left; | |
288867ec | 388 | else |
c0a31329 | 389 | link = &(*link)->rb_right; |
c0a31329 TG |
390 | } |
391 | ||
392 | /* | |
288867ec TG |
393 | * Insert the timer to the rbtree and check whether it |
394 | * replaces the first pending timer | |
c0a31329 TG |
395 | */ |
396 | rb_link_node(&timer->node, parent, link); | |
397 | rb_insert_color(&timer->node, &base->active); | |
303e967f TG |
398 | /* |
399 | * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the | |
400 | * state of a possibly running callback. | |
401 | */ | |
402 | timer->state |= HRTIMER_STATE_ENQUEUED; | |
c0a31329 | 403 | |
288867ec TG |
404 | if (!base->first || timer->expires.tv64 < |
405 | rb_entry(base->first, struct hrtimer, node)->expires.tv64) | |
406 | base->first = &timer->node; | |
407 | } | |
c0a31329 TG |
408 | |
409 | /* | |
410 | * __remove_hrtimer - internal function to remove a timer | |
411 | * | |
412 | * Caller must hold the base lock. | |
413 | */ | |
3c8aa39d | 414 | static void __remove_hrtimer(struct hrtimer *timer, |
303e967f TG |
415 | struct hrtimer_clock_base *base, |
416 | unsigned long newstate) | |
c0a31329 TG |
417 | { |
418 | /* | |
288867ec TG |
419 | * Remove the timer from the rbtree and replace the |
420 | * first entry pointer if necessary. | |
c0a31329 | 421 | */ |
288867ec TG |
422 | if (base->first == &timer->node) |
423 | base->first = rb_next(&timer->node); | |
c0a31329 | 424 | rb_erase(&timer->node, &base->active); |
303e967f | 425 | timer->state = newstate; |
c0a31329 TG |
426 | } |
427 | ||
428 | /* | |
429 | * remove hrtimer, called with base lock held | |
430 | */ | |
431 | static inline int | |
3c8aa39d | 432 | remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base) |
c0a31329 | 433 | { |
303e967f TG |
434 | if (hrtimer_is_queued(timer)) { |
435 | __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE); | |
c0a31329 TG |
436 | return 1; |
437 | } | |
438 | return 0; | |
439 | } | |
440 | ||
441 | /** | |
442 | * hrtimer_start - (re)start an relative timer on the current CPU | |
c0a31329 TG |
443 | * @timer: the timer to be added |
444 | * @tim: expiry time | |
445 | * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL) | |
446 | * | |
447 | * Returns: | |
448 | * 0 on success | |
449 | * 1 when the timer was active | |
450 | */ | |
451 | int | |
452 | hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode) | |
453 | { | |
3c8aa39d | 454 | struct hrtimer_clock_base *base, *new_base; |
c0a31329 TG |
455 | unsigned long flags; |
456 | int ret; | |
457 | ||
458 | base = lock_hrtimer_base(timer, &flags); | |
459 | ||
460 | /* Remove an active timer from the queue: */ | |
461 | ret = remove_hrtimer(timer, base); | |
462 | ||
463 | /* Switch the timer base, if necessary: */ | |
464 | new_base = switch_hrtimer_base(timer, base); | |
465 | ||
c9cb2e3d | 466 | if (mode == HRTIMER_MODE_REL) { |
c0a31329 | 467 | tim = ktime_add(tim, new_base->get_time()); |
06027bdd IM |
468 | /* |
469 | * CONFIG_TIME_LOW_RES is a temporary way for architectures | |
470 | * to signal that they simply return xtime in | |
471 | * do_gettimeoffset(). In this case we want to round up by | |
472 | * resolution when starting a relative timer, to avoid short | |
473 | * timeouts. This will go away with the GTOD framework. | |
474 | */ | |
475 | #ifdef CONFIG_TIME_LOW_RES | |
476 | tim = ktime_add(tim, base->resolution); | |
477 | #endif | |
478 | } | |
c0a31329 TG |
479 | timer->expires = tim; |
480 | ||
481 | enqueue_hrtimer(timer, new_base); | |
482 | ||
483 | unlock_hrtimer_base(timer, &flags); | |
484 | ||
485 | return ret; | |
486 | } | |
8d16b764 | 487 | EXPORT_SYMBOL_GPL(hrtimer_start); |
c0a31329 TG |
488 | |
489 | /** | |
490 | * hrtimer_try_to_cancel - try to deactivate a timer | |
c0a31329 TG |
491 | * @timer: hrtimer to stop |
492 | * | |
493 | * Returns: | |
494 | * 0 when the timer was not active | |
495 | * 1 when the timer was active | |
496 | * -1 when the timer is currently excuting the callback function and | |
fa9799e3 | 497 | * cannot be stopped |
c0a31329 TG |
498 | */ |
499 | int hrtimer_try_to_cancel(struct hrtimer *timer) | |
500 | { | |
3c8aa39d | 501 | struct hrtimer_clock_base *base; |
c0a31329 TG |
502 | unsigned long flags; |
503 | int ret = -1; | |
504 | ||
505 | base = lock_hrtimer_base(timer, &flags); | |
506 | ||
303e967f | 507 | if (!hrtimer_callback_running(timer)) |
c0a31329 TG |
508 | ret = remove_hrtimer(timer, base); |
509 | ||
510 | unlock_hrtimer_base(timer, &flags); | |
511 | ||
512 | return ret; | |
513 | ||
514 | } | |
8d16b764 | 515 | EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel); |
c0a31329 TG |
516 | |
517 | /** | |
518 | * hrtimer_cancel - cancel a timer and wait for the handler to finish. | |
c0a31329 TG |
519 | * @timer: the timer to be cancelled |
520 | * | |
521 | * Returns: | |
522 | * 0 when the timer was not active | |
523 | * 1 when the timer was active | |
524 | */ | |
525 | int hrtimer_cancel(struct hrtimer *timer) | |
526 | { | |
527 | for (;;) { | |
528 | int ret = hrtimer_try_to_cancel(timer); | |
529 | ||
530 | if (ret >= 0) | |
531 | return ret; | |
5ef37b19 | 532 | cpu_relax(); |
c0a31329 TG |
533 | } |
534 | } | |
8d16b764 | 535 | EXPORT_SYMBOL_GPL(hrtimer_cancel); |
c0a31329 TG |
536 | |
537 | /** | |
538 | * hrtimer_get_remaining - get remaining time for the timer | |
c0a31329 TG |
539 | * @timer: the timer to read |
540 | */ | |
541 | ktime_t hrtimer_get_remaining(const struct hrtimer *timer) | |
542 | { | |
3c8aa39d | 543 | struct hrtimer_clock_base *base; |
c0a31329 TG |
544 | unsigned long flags; |
545 | ktime_t rem; | |
546 | ||
547 | base = lock_hrtimer_base(timer, &flags); | |
3c8aa39d | 548 | rem = ktime_sub(timer->expires, base->get_time()); |
c0a31329 TG |
549 | unlock_hrtimer_base(timer, &flags); |
550 | ||
551 | return rem; | |
552 | } | |
8d16b764 | 553 | EXPORT_SYMBOL_GPL(hrtimer_get_remaining); |
c0a31329 | 554 | |
fd064b9b | 555 | #if defined(CONFIG_NO_IDLE_HZ) || defined(CONFIG_NO_HZ) |
69239749 TL |
556 | /** |
557 | * hrtimer_get_next_event - get the time until next expiry event | |
558 | * | |
559 | * Returns the delta to the next expiry event or KTIME_MAX if no timer | |
560 | * is pending. | |
561 | */ | |
562 | ktime_t hrtimer_get_next_event(void) | |
563 | { | |
3c8aa39d TG |
564 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
565 | struct hrtimer_clock_base *base = cpu_base->clock_base; | |
69239749 TL |
566 | ktime_t delta, mindelta = { .tv64 = KTIME_MAX }; |
567 | unsigned long flags; | |
568 | int i; | |
569 | ||
3c8aa39d TG |
570 | spin_lock_irqsave(&cpu_base->lock, flags); |
571 | ||
572 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) { | |
69239749 TL |
573 | struct hrtimer *timer; |
574 | ||
3c8aa39d | 575 | if (!base->first) |
69239749 | 576 | continue; |
3c8aa39d | 577 | |
69239749 TL |
578 | timer = rb_entry(base->first, struct hrtimer, node); |
579 | delta.tv64 = timer->expires.tv64; | |
69239749 TL |
580 | delta = ktime_sub(delta, base->get_time()); |
581 | if (delta.tv64 < mindelta.tv64) | |
582 | mindelta.tv64 = delta.tv64; | |
583 | } | |
3c8aa39d TG |
584 | |
585 | spin_unlock_irqrestore(&cpu_base->lock, flags); | |
586 | ||
69239749 TL |
587 | if (mindelta.tv64 < 0) |
588 | mindelta.tv64 = 0; | |
589 | return mindelta; | |
590 | } | |
591 | #endif | |
592 | ||
c0a31329 | 593 | /** |
7978672c | 594 | * hrtimer_init - initialize a timer to the given clock |
7978672c | 595 | * @timer: the timer to be initialized |
c0a31329 | 596 | * @clock_id: the clock to be used |
7978672c | 597 | * @mode: timer mode abs/rel |
c0a31329 | 598 | */ |
7978672c GA |
599 | void hrtimer_init(struct hrtimer *timer, clockid_t clock_id, |
600 | enum hrtimer_mode mode) | |
c0a31329 | 601 | { |
3c8aa39d | 602 | struct hrtimer_cpu_base *cpu_base; |
c0a31329 | 603 | |
7978672c GA |
604 | memset(timer, 0, sizeof(struct hrtimer)); |
605 | ||
3c8aa39d | 606 | cpu_base = &__raw_get_cpu_var(hrtimer_bases); |
c0a31329 | 607 | |
c9cb2e3d | 608 | if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS) |
7978672c GA |
609 | clock_id = CLOCK_MONOTONIC; |
610 | ||
3c8aa39d | 611 | timer->base = &cpu_base->clock_base[clock_id]; |
c0a31329 | 612 | } |
8d16b764 | 613 | EXPORT_SYMBOL_GPL(hrtimer_init); |
c0a31329 TG |
614 | |
615 | /** | |
616 | * hrtimer_get_res - get the timer resolution for a clock | |
c0a31329 TG |
617 | * @which_clock: which clock to query |
618 | * @tp: pointer to timespec variable to store the resolution | |
619 | * | |
72fd4a35 RD |
620 | * Store the resolution of the clock selected by @which_clock in the |
621 | * variable pointed to by @tp. | |
c0a31329 TG |
622 | */ |
623 | int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp) | |
624 | { | |
3c8aa39d | 625 | struct hrtimer_cpu_base *cpu_base; |
c0a31329 | 626 | |
3c8aa39d TG |
627 | cpu_base = &__raw_get_cpu_var(hrtimer_bases); |
628 | *tp = ktime_to_timespec(cpu_base->clock_base[which_clock].resolution); | |
c0a31329 TG |
629 | |
630 | return 0; | |
631 | } | |
8d16b764 | 632 | EXPORT_SYMBOL_GPL(hrtimer_get_res); |
c0a31329 TG |
633 | |
634 | /* | |
635 | * Expire the per base hrtimer-queue: | |
636 | */ | |
3c8aa39d TG |
637 | static inline void run_hrtimer_queue(struct hrtimer_cpu_base *cpu_base, |
638 | int index) | |
c0a31329 | 639 | { |
288867ec | 640 | struct rb_node *node; |
3c8aa39d | 641 | struct hrtimer_clock_base *base = &cpu_base->clock_base[index]; |
c0a31329 | 642 | |
3055adda DS |
643 | if (!base->first) |
644 | return; | |
645 | ||
92127c7a TG |
646 | if (base->get_softirq_time) |
647 | base->softirq_time = base->get_softirq_time(); | |
648 | ||
3c8aa39d | 649 | spin_lock_irq(&cpu_base->lock); |
c0a31329 | 650 | |
288867ec | 651 | while ((node = base->first)) { |
c0a31329 | 652 | struct hrtimer *timer; |
c9cb2e3d | 653 | enum hrtimer_restart (*fn)(struct hrtimer *); |
c0a31329 | 654 | int restart; |
c0a31329 | 655 | |
288867ec | 656 | timer = rb_entry(node, struct hrtimer, node); |
92127c7a | 657 | if (base->softirq_time.tv64 <= timer->expires.tv64) |
c0a31329 TG |
658 | break; |
659 | ||
660 | fn = timer->function; | |
303e967f | 661 | __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK); |
3c8aa39d | 662 | spin_unlock_irq(&cpu_base->lock); |
c0a31329 | 663 | |
05cfb614 | 664 | restart = fn(timer); |
c0a31329 | 665 | |
3c8aa39d | 666 | spin_lock_irq(&cpu_base->lock); |
c0a31329 | 667 | |
303e967f | 668 | timer->state &= ~HRTIMER_STATE_CALLBACK; |
b75f7a51 RZ |
669 | if (restart != HRTIMER_NORESTART) { |
670 | BUG_ON(hrtimer_active(timer)); | |
c0a31329 | 671 | enqueue_hrtimer(timer, base); |
b75f7a51 | 672 | } |
c0a31329 | 673 | } |
3c8aa39d | 674 | spin_unlock_irq(&cpu_base->lock); |
c0a31329 TG |
675 | } |
676 | ||
677 | /* | |
678 | * Called from timer softirq every jiffy, expire hrtimers: | |
679 | */ | |
680 | void hrtimer_run_queues(void) | |
681 | { | |
3c8aa39d | 682 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
c0a31329 TG |
683 | int i; |
684 | ||
3c8aa39d | 685 | hrtimer_get_softirq_time(cpu_base); |
92127c7a | 686 | |
3c8aa39d TG |
687 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) |
688 | run_hrtimer_queue(cpu_base, i); | |
c0a31329 TG |
689 | } |
690 | ||
10c94ec1 TG |
691 | /* |
692 | * Sleep related functions: | |
693 | */ | |
c9cb2e3d | 694 | static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer) |
00362e33 TG |
695 | { |
696 | struct hrtimer_sleeper *t = | |
697 | container_of(timer, struct hrtimer_sleeper, timer); | |
698 | struct task_struct *task = t->task; | |
699 | ||
700 | t->task = NULL; | |
701 | if (task) | |
702 | wake_up_process(task); | |
703 | ||
704 | return HRTIMER_NORESTART; | |
705 | } | |
706 | ||
36c8b586 | 707 | void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task) |
00362e33 TG |
708 | { |
709 | sl->timer.function = hrtimer_wakeup; | |
710 | sl->task = task; | |
711 | } | |
712 | ||
669d7868 | 713 | static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode) |
432569bb | 714 | { |
669d7868 | 715 | hrtimer_init_sleeper(t, current); |
10c94ec1 | 716 | |
432569bb RZ |
717 | do { |
718 | set_current_state(TASK_INTERRUPTIBLE); | |
719 | hrtimer_start(&t->timer, t->timer.expires, mode); | |
720 | ||
721 | schedule(); | |
722 | ||
669d7868 | 723 | hrtimer_cancel(&t->timer); |
c9cb2e3d | 724 | mode = HRTIMER_MODE_ABS; |
669d7868 TG |
725 | |
726 | } while (t->task && !signal_pending(current)); | |
432569bb | 727 | |
669d7868 | 728 | return t->task == NULL; |
10c94ec1 TG |
729 | } |
730 | ||
1711ef38 | 731 | long __sched hrtimer_nanosleep_restart(struct restart_block *restart) |
10c94ec1 | 732 | { |
669d7868 | 733 | struct hrtimer_sleeper t; |
ea13dbc8 IM |
734 | struct timespec __user *rmtp; |
735 | struct timespec tu; | |
432569bb | 736 | ktime_t time; |
10c94ec1 TG |
737 | |
738 | restart->fn = do_no_restart_syscall; | |
739 | ||
c9cb2e3d | 740 | hrtimer_init(&t.timer, restart->arg0, HRTIMER_MODE_ABS); |
1711ef38 | 741 | t.timer.expires.tv64 = ((u64)restart->arg3 << 32) | (u64) restart->arg2; |
10c94ec1 | 742 | |
c9cb2e3d | 743 | if (do_nanosleep(&t, HRTIMER_MODE_ABS)) |
10c94ec1 TG |
744 | return 0; |
745 | ||
1711ef38 | 746 | rmtp = (struct timespec __user *) restart->arg1; |
432569bb RZ |
747 | if (rmtp) { |
748 | time = ktime_sub(t.timer.expires, t.timer.base->get_time()); | |
749 | if (time.tv64 <= 0) | |
750 | return 0; | |
751 | tu = ktime_to_timespec(time); | |
752 | if (copy_to_user(rmtp, &tu, sizeof(tu))) | |
753 | return -EFAULT; | |
754 | } | |
10c94ec1 | 755 | |
1711ef38 | 756 | restart->fn = hrtimer_nanosleep_restart; |
10c94ec1 TG |
757 | |
758 | /* The other values in restart are already filled in */ | |
759 | return -ERESTART_RESTARTBLOCK; | |
760 | } | |
761 | ||
10c94ec1 TG |
762 | long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp, |
763 | const enum hrtimer_mode mode, const clockid_t clockid) | |
764 | { | |
765 | struct restart_block *restart; | |
669d7868 | 766 | struct hrtimer_sleeper t; |
10c94ec1 TG |
767 | struct timespec tu; |
768 | ktime_t rem; | |
769 | ||
432569bb RZ |
770 | hrtimer_init(&t.timer, clockid, mode); |
771 | t.timer.expires = timespec_to_ktime(*rqtp); | |
772 | if (do_nanosleep(&t, mode)) | |
10c94ec1 TG |
773 | return 0; |
774 | ||
7978672c | 775 | /* Absolute timers do not update the rmtp value and restart: */ |
c9cb2e3d | 776 | if (mode == HRTIMER_MODE_ABS) |
10c94ec1 TG |
777 | return -ERESTARTNOHAND; |
778 | ||
432569bb RZ |
779 | if (rmtp) { |
780 | rem = ktime_sub(t.timer.expires, t.timer.base->get_time()); | |
781 | if (rem.tv64 <= 0) | |
782 | return 0; | |
783 | tu = ktime_to_timespec(rem); | |
784 | if (copy_to_user(rmtp, &tu, sizeof(tu))) | |
785 | return -EFAULT; | |
786 | } | |
10c94ec1 TG |
787 | |
788 | restart = ¤t_thread_info()->restart_block; | |
1711ef38 TA |
789 | restart->fn = hrtimer_nanosleep_restart; |
790 | restart->arg0 = (unsigned long) t.timer.base->index; | |
791 | restart->arg1 = (unsigned long) rmtp; | |
792 | restart->arg2 = t.timer.expires.tv64 & 0xFFFFFFFF; | |
793 | restart->arg3 = t.timer.expires.tv64 >> 32; | |
10c94ec1 TG |
794 | |
795 | return -ERESTART_RESTARTBLOCK; | |
796 | } | |
797 | ||
6ba1b912 TG |
798 | asmlinkage long |
799 | sys_nanosleep(struct timespec __user *rqtp, struct timespec __user *rmtp) | |
800 | { | |
801 | struct timespec tu; | |
802 | ||
803 | if (copy_from_user(&tu, rqtp, sizeof(tu))) | |
804 | return -EFAULT; | |
805 | ||
806 | if (!timespec_valid(&tu)) | |
807 | return -EINVAL; | |
808 | ||
c9cb2e3d | 809 | return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC); |
6ba1b912 TG |
810 | } |
811 | ||
c0a31329 TG |
812 | /* |
813 | * Functions related to boot-time initialization: | |
814 | */ | |
815 | static void __devinit init_hrtimers_cpu(int cpu) | |
816 | { | |
3c8aa39d | 817 | struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu); |
c0a31329 TG |
818 | int i; |
819 | ||
3c8aa39d TG |
820 | spin_lock_init(&cpu_base->lock); |
821 | lockdep_set_class(&cpu_base->lock, &cpu_base->lock_key); | |
822 | ||
823 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) | |
824 | cpu_base->clock_base[i].cpu_base = cpu_base; | |
825 | ||
c0a31329 TG |
826 | } |
827 | ||
828 | #ifdef CONFIG_HOTPLUG_CPU | |
829 | ||
3c8aa39d TG |
830 | static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base, |
831 | struct hrtimer_clock_base *new_base) | |
c0a31329 TG |
832 | { |
833 | struct hrtimer *timer; | |
834 | struct rb_node *node; | |
835 | ||
836 | while ((node = rb_first(&old_base->active))) { | |
837 | timer = rb_entry(node, struct hrtimer, node); | |
303e967f TG |
838 | BUG_ON(timer->state & HRTIMER_STATE_CALLBACK); |
839 | __remove_hrtimer(timer, old_base, HRTIMER_STATE_INACTIVE); | |
c0a31329 TG |
840 | timer->base = new_base; |
841 | enqueue_hrtimer(timer, new_base); | |
842 | } | |
843 | } | |
844 | ||
845 | static void migrate_hrtimers(int cpu) | |
846 | { | |
3c8aa39d | 847 | struct hrtimer_cpu_base *old_base, *new_base; |
c0a31329 TG |
848 | int i; |
849 | ||
850 | BUG_ON(cpu_online(cpu)); | |
3c8aa39d TG |
851 | old_base = &per_cpu(hrtimer_bases, cpu); |
852 | new_base = &get_cpu_var(hrtimer_bases); | |
c0a31329 TG |
853 | |
854 | local_irq_disable(); | |
855 | ||
3c8aa39d TG |
856 | spin_lock(&new_base->lock); |
857 | spin_lock(&old_base->lock); | |
c0a31329 | 858 | |
3c8aa39d | 859 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
3c8aa39d TG |
860 | migrate_hrtimer_list(&old_base->clock_base[i], |
861 | &new_base->clock_base[i]); | |
c0a31329 | 862 | } |
3c8aa39d TG |
863 | spin_unlock(&old_base->lock); |
864 | spin_unlock(&new_base->lock); | |
c0a31329 TG |
865 | |
866 | local_irq_enable(); | |
867 | put_cpu_var(hrtimer_bases); | |
868 | } | |
869 | #endif /* CONFIG_HOTPLUG_CPU */ | |
870 | ||
8c78f307 | 871 | static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self, |
c0a31329 TG |
872 | unsigned long action, void *hcpu) |
873 | { | |
874 | long cpu = (long)hcpu; | |
875 | ||
876 | switch (action) { | |
877 | ||
878 | case CPU_UP_PREPARE: | |
879 | init_hrtimers_cpu(cpu); | |
880 | break; | |
881 | ||
882 | #ifdef CONFIG_HOTPLUG_CPU | |
883 | case CPU_DEAD: | |
d316c57f | 884 | clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &cpu); |
c0a31329 TG |
885 | migrate_hrtimers(cpu); |
886 | break; | |
887 | #endif | |
888 | ||
889 | default: | |
890 | break; | |
891 | } | |
892 | ||
893 | return NOTIFY_OK; | |
894 | } | |
895 | ||
8c78f307 | 896 | static struct notifier_block __cpuinitdata hrtimers_nb = { |
c0a31329 TG |
897 | .notifier_call = hrtimer_cpu_notify, |
898 | }; | |
899 | ||
900 | void __init hrtimers_init(void) | |
901 | { | |
902 | hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE, | |
903 | (void *)(long)smp_processor_id()); | |
904 | register_cpu_notifier(&hrtimers_nb); | |
905 | } | |
906 |