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> | |
54cdfdb4 | 35 | #include <linux/irq.h> |
c0a31329 TG |
36 | #include <linux/module.h> |
37 | #include <linux/percpu.h> | |
38 | #include <linux/hrtimer.h> | |
39 | #include <linux/notifier.h> | |
40 | #include <linux/syscalls.h> | |
54cdfdb4 | 41 | #include <linux/kallsyms.h> |
c0a31329 | 42 | #include <linux/interrupt.h> |
79bf2bb3 | 43 | #include <linux/tick.h> |
54cdfdb4 TG |
44 | #include <linux/seq_file.h> |
45 | #include <linux/err.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); | |
f4304ab2 | 144 | #ifdef CONFIG_NO_HZ |
145 | getnstimeofday(&xts); | |
146 | #else | |
147 | xts = xtime; | |
148 | #endif | |
ad28d94a | 149 | tom = wall_to_monotonic; |
92127c7a TG |
150 | } while (read_seqretry(&xtime_lock, seq)); |
151 | ||
f4304ab2 | 152 | xtim = timespec_to_ktime(xts); |
ad28d94a | 153 | tomono = timespec_to_ktime(tom); |
3c8aa39d TG |
154 | base->clock_base[CLOCK_REALTIME].softirq_time = xtim; |
155 | base->clock_base[CLOCK_MONOTONIC].softirq_time = | |
156 | ktime_add(xtim, tomono); | |
92127c7a TG |
157 | } |
158 | ||
303e967f TG |
159 | /* |
160 | * Helper function to check, whether the timer is running the callback | |
161 | * function | |
162 | */ | |
163 | static inline int hrtimer_callback_running(struct hrtimer *timer) | |
164 | { | |
165 | return timer->state & HRTIMER_STATE_CALLBACK; | |
166 | } | |
167 | ||
c0a31329 TG |
168 | /* |
169 | * Functions and macros which are different for UP/SMP systems are kept in a | |
170 | * single place | |
171 | */ | |
172 | #ifdef CONFIG_SMP | |
173 | ||
c0a31329 TG |
174 | /* |
175 | * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock | |
176 | * means that all timers which are tied to this base via timer->base are | |
177 | * locked, and the base itself is locked too. | |
178 | * | |
179 | * So __run_timers/migrate_timers can safely modify all timers which could | |
180 | * be found on the lists/queues. | |
181 | * | |
182 | * When the timer's base is locked, and the timer removed from list, it is | |
183 | * possible to set timer->base = NULL and drop the lock: the timer remains | |
184 | * locked. | |
185 | */ | |
3c8aa39d TG |
186 | static |
187 | struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer, | |
188 | unsigned long *flags) | |
c0a31329 | 189 | { |
3c8aa39d | 190 | struct hrtimer_clock_base *base; |
c0a31329 TG |
191 | |
192 | for (;;) { | |
193 | base = timer->base; | |
194 | if (likely(base != NULL)) { | |
3c8aa39d | 195 | spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
196 | if (likely(base == timer->base)) |
197 | return base; | |
198 | /* The timer has migrated to another CPU: */ | |
3c8aa39d | 199 | spin_unlock_irqrestore(&base->cpu_base->lock, *flags); |
c0a31329 TG |
200 | } |
201 | cpu_relax(); | |
202 | } | |
203 | } | |
204 | ||
205 | /* | |
206 | * Switch the timer base to the current CPU when possible. | |
207 | */ | |
3c8aa39d TG |
208 | static inline struct hrtimer_clock_base * |
209 | switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base) | |
c0a31329 | 210 | { |
3c8aa39d TG |
211 | struct hrtimer_clock_base *new_base; |
212 | struct hrtimer_cpu_base *new_cpu_base; | |
c0a31329 | 213 | |
3c8aa39d TG |
214 | new_cpu_base = &__get_cpu_var(hrtimer_bases); |
215 | new_base = &new_cpu_base->clock_base[base->index]; | |
c0a31329 TG |
216 | |
217 | if (base != new_base) { | |
218 | /* | |
219 | * We are trying to schedule the timer on the local CPU. | |
220 | * However we can't change timer's base while it is running, | |
221 | * so we keep it on the same CPU. No hassle vs. reprogramming | |
222 | * the event source in the high resolution case. The softirq | |
223 | * code will take care of this when the timer function has | |
224 | * completed. There is no conflict as we hold the lock until | |
225 | * the timer is enqueued. | |
226 | */ | |
54cdfdb4 | 227 | if (unlikely(hrtimer_callback_running(timer))) |
c0a31329 TG |
228 | return base; |
229 | ||
230 | /* See the comment in lock_timer_base() */ | |
231 | timer->base = NULL; | |
3c8aa39d TG |
232 | spin_unlock(&base->cpu_base->lock); |
233 | spin_lock(&new_base->cpu_base->lock); | |
c0a31329 TG |
234 | timer->base = new_base; |
235 | } | |
236 | return new_base; | |
237 | } | |
238 | ||
239 | #else /* CONFIG_SMP */ | |
240 | ||
3c8aa39d | 241 | static inline struct hrtimer_clock_base * |
c0a31329 TG |
242 | lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) |
243 | { | |
3c8aa39d | 244 | struct hrtimer_clock_base *base = timer->base; |
c0a31329 | 245 | |
3c8aa39d | 246 | spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
247 | |
248 | return base; | |
249 | } | |
250 | ||
54cdfdb4 | 251 | # define switch_hrtimer_base(t, b) (b) |
c0a31329 TG |
252 | |
253 | #endif /* !CONFIG_SMP */ | |
254 | ||
255 | /* | |
256 | * Functions for the union type storage format of ktime_t which are | |
257 | * too large for inlining: | |
258 | */ | |
259 | #if BITS_PER_LONG < 64 | |
260 | # ifndef CONFIG_KTIME_SCALAR | |
261 | /** | |
262 | * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable | |
c0a31329 TG |
263 | * @kt: addend |
264 | * @nsec: the scalar nsec value to add | |
265 | * | |
266 | * Returns the sum of kt and nsec in ktime_t format | |
267 | */ | |
268 | ktime_t ktime_add_ns(const ktime_t kt, u64 nsec) | |
269 | { | |
270 | ktime_t tmp; | |
271 | ||
272 | if (likely(nsec < NSEC_PER_SEC)) { | |
273 | tmp.tv64 = nsec; | |
274 | } else { | |
275 | unsigned long rem = do_div(nsec, NSEC_PER_SEC); | |
276 | ||
277 | tmp = ktime_set((long)nsec, rem); | |
278 | } | |
279 | ||
280 | return ktime_add(kt, tmp); | |
281 | } | |
b8b8fd2d DH |
282 | |
283 | EXPORT_SYMBOL_GPL(ktime_add_ns); | |
c0a31329 TG |
284 | # endif /* !CONFIG_KTIME_SCALAR */ |
285 | ||
286 | /* | |
287 | * Divide a ktime value by a nanosecond value | |
288 | */ | |
79bf2bb3 | 289 | unsigned long ktime_divns(const ktime_t kt, s64 div) |
c0a31329 TG |
290 | { |
291 | u64 dclc, inc, dns; | |
292 | int sft = 0; | |
293 | ||
294 | dclc = dns = ktime_to_ns(kt); | |
295 | inc = div; | |
296 | /* Make sure the divisor is less than 2^32: */ | |
297 | while (div >> 32) { | |
298 | sft++; | |
299 | div >>= 1; | |
300 | } | |
301 | dclc >>= sft; | |
302 | do_div(dclc, (unsigned long) div); | |
303 | ||
304 | return (unsigned long) dclc; | |
305 | } | |
c0a31329 TG |
306 | #endif /* BITS_PER_LONG >= 64 */ |
307 | ||
54cdfdb4 TG |
308 | /* High resolution timer related functions */ |
309 | #ifdef CONFIG_HIGH_RES_TIMERS | |
310 | ||
311 | /* | |
312 | * High resolution timer enabled ? | |
313 | */ | |
314 | static int hrtimer_hres_enabled __read_mostly = 1; | |
315 | ||
316 | /* | |
317 | * Enable / Disable high resolution mode | |
318 | */ | |
319 | static int __init setup_hrtimer_hres(char *str) | |
320 | { | |
321 | if (!strcmp(str, "off")) | |
322 | hrtimer_hres_enabled = 0; | |
323 | else if (!strcmp(str, "on")) | |
324 | hrtimer_hres_enabled = 1; | |
325 | else | |
326 | return 0; | |
327 | return 1; | |
328 | } | |
329 | ||
330 | __setup("highres=", setup_hrtimer_hres); | |
331 | ||
332 | /* | |
333 | * hrtimer_high_res_enabled - query, if the highres mode is enabled | |
334 | */ | |
335 | static inline int hrtimer_is_hres_enabled(void) | |
336 | { | |
337 | return hrtimer_hres_enabled; | |
338 | } | |
339 | ||
340 | /* | |
341 | * Is the high resolution mode active ? | |
342 | */ | |
343 | static inline int hrtimer_hres_active(void) | |
344 | { | |
345 | return __get_cpu_var(hrtimer_bases).hres_active; | |
346 | } | |
347 | ||
348 | /* | |
349 | * Reprogram the event source with checking both queues for the | |
350 | * next event | |
351 | * Called with interrupts disabled and base->lock held | |
352 | */ | |
353 | static void hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base) | |
354 | { | |
355 | int i; | |
356 | struct hrtimer_clock_base *base = cpu_base->clock_base; | |
357 | ktime_t expires; | |
358 | ||
359 | cpu_base->expires_next.tv64 = KTIME_MAX; | |
360 | ||
361 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) { | |
362 | struct hrtimer *timer; | |
363 | ||
364 | if (!base->first) | |
365 | continue; | |
366 | timer = rb_entry(base->first, struct hrtimer, node); | |
367 | expires = ktime_sub(timer->expires, base->offset); | |
368 | if (expires.tv64 < cpu_base->expires_next.tv64) | |
369 | cpu_base->expires_next = expires; | |
370 | } | |
371 | ||
372 | if (cpu_base->expires_next.tv64 != KTIME_MAX) | |
373 | tick_program_event(cpu_base->expires_next, 1); | |
374 | } | |
375 | ||
376 | /* | |
377 | * Shared reprogramming for clock_realtime and clock_monotonic | |
378 | * | |
379 | * When a timer is enqueued and expires earlier than the already enqueued | |
380 | * timers, we have to check, whether it expires earlier than the timer for | |
381 | * which the clock event device was armed. | |
382 | * | |
383 | * Called with interrupts disabled and base->cpu_base.lock held | |
384 | */ | |
385 | static int hrtimer_reprogram(struct hrtimer *timer, | |
386 | struct hrtimer_clock_base *base) | |
387 | { | |
388 | ktime_t *expires_next = &__get_cpu_var(hrtimer_bases).expires_next; | |
389 | ktime_t expires = ktime_sub(timer->expires, base->offset); | |
390 | int res; | |
391 | ||
392 | /* | |
393 | * When the callback is running, we do not reprogram the clock event | |
394 | * device. The timer callback is either running on a different CPU or | |
395 | * the callback is executed in the hrtimer_interupt context. The | |
396 | * reprogramming is handled either by the softirq, which called the | |
397 | * callback or at the end of the hrtimer_interrupt. | |
398 | */ | |
399 | if (hrtimer_callback_running(timer)) | |
400 | return 0; | |
401 | ||
402 | if (expires.tv64 >= expires_next->tv64) | |
403 | return 0; | |
404 | ||
405 | /* | |
406 | * Clockevents returns -ETIME, when the event was in the past. | |
407 | */ | |
408 | res = tick_program_event(expires, 0); | |
409 | if (!IS_ERR_VALUE(res)) | |
410 | *expires_next = expires; | |
411 | return res; | |
412 | } | |
413 | ||
414 | ||
415 | /* | |
416 | * Retrigger next event is called after clock was set | |
417 | * | |
418 | * Called with interrupts disabled via on_each_cpu() | |
419 | */ | |
420 | static void retrigger_next_event(void *arg) | |
421 | { | |
422 | struct hrtimer_cpu_base *base; | |
423 | struct timespec realtime_offset; | |
424 | unsigned long seq; | |
425 | ||
426 | if (!hrtimer_hres_active()) | |
427 | return; | |
428 | ||
429 | do { | |
430 | seq = read_seqbegin(&xtime_lock); | |
431 | set_normalized_timespec(&realtime_offset, | |
432 | -wall_to_monotonic.tv_sec, | |
433 | -wall_to_monotonic.tv_nsec); | |
434 | } while (read_seqretry(&xtime_lock, seq)); | |
435 | ||
436 | base = &__get_cpu_var(hrtimer_bases); | |
437 | ||
438 | /* Adjust CLOCK_REALTIME offset */ | |
439 | spin_lock(&base->lock); | |
440 | base->clock_base[CLOCK_REALTIME].offset = | |
441 | timespec_to_ktime(realtime_offset); | |
442 | ||
443 | hrtimer_force_reprogram(base); | |
444 | spin_unlock(&base->lock); | |
445 | } | |
446 | ||
447 | /* | |
448 | * Clock realtime was set | |
449 | * | |
450 | * Change the offset of the realtime clock vs. the monotonic | |
451 | * clock. | |
452 | * | |
453 | * We might have to reprogram the high resolution timer interrupt. On | |
454 | * SMP we call the architecture specific code to retrigger _all_ high | |
455 | * resolution timer interrupts. On UP we just disable interrupts and | |
456 | * call the high resolution interrupt code. | |
457 | */ | |
458 | void clock_was_set(void) | |
459 | { | |
460 | /* Retrigger the CPU local events everywhere */ | |
461 | on_each_cpu(retrigger_next_event, NULL, 0, 1); | |
462 | } | |
463 | ||
995f054f IM |
464 | /* |
465 | * During resume we might have to reprogram the high resolution timer | |
466 | * interrupt (on the local CPU): | |
467 | */ | |
468 | void hres_timers_resume(void) | |
469 | { | |
470 | WARN_ON_ONCE(num_online_cpus() > 1); | |
471 | ||
472 | /* Retrigger the CPU local events: */ | |
473 | retrigger_next_event(NULL); | |
474 | } | |
475 | ||
54cdfdb4 TG |
476 | /* |
477 | * Check, whether the timer is on the callback pending list | |
478 | */ | |
479 | static inline int hrtimer_cb_pending(const struct hrtimer *timer) | |
480 | { | |
481 | return timer->state & HRTIMER_STATE_PENDING; | |
482 | } | |
483 | ||
484 | /* | |
485 | * Remove a timer from the callback pending list | |
486 | */ | |
487 | static inline void hrtimer_remove_cb_pending(struct hrtimer *timer) | |
488 | { | |
489 | list_del_init(&timer->cb_entry); | |
490 | } | |
491 | ||
492 | /* | |
493 | * Initialize the high resolution related parts of cpu_base | |
494 | */ | |
495 | static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) | |
496 | { | |
497 | base->expires_next.tv64 = KTIME_MAX; | |
498 | base->hres_active = 0; | |
499 | INIT_LIST_HEAD(&base->cb_pending); | |
500 | } | |
501 | ||
502 | /* | |
503 | * Initialize the high resolution related parts of a hrtimer | |
504 | */ | |
505 | static inline void hrtimer_init_timer_hres(struct hrtimer *timer) | |
506 | { | |
507 | INIT_LIST_HEAD(&timer->cb_entry); | |
508 | } | |
509 | ||
510 | /* | |
511 | * When High resolution timers are active, try to reprogram. Note, that in case | |
512 | * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry | |
513 | * check happens. The timer gets enqueued into the rbtree. The reprogramming | |
514 | * and expiry check is done in the hrtimer_interrupt or in the softirq. | |
515 | */ | |
516 | static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer, | |
517 | struct hrtimer_clock_base *base) | |
518 | { | |
519 | if (base->cpu_base->hres_active && hrtimer_reprogram(timer, base)) { | |
520 | ||
521 | /* Timer is expired, act upon the callback mode */ | |
522 | switch(timer->cb_mode) { | |
523 | case HRTIMER_CB_IRQSAFE_NO_RESTART: | |
524 | /* | |
525 | * We can call the callback from here. No restart | |
526 | * happens, so no danger of recursion | |
527 | */ | |
528 | BUG_ON(timer->function(timer) != HRTIMER_NORESTART); | |
529 | return 1; | |
530 | case HRTIMER_CB_IRQSAFE_NO_SOFTIRQ: | |
531 | /* | |
532 | * This is solely for the sched tick emulation with | |
533 | * dynamic tick support to ensure that we do not | |
534 | * restart the tick right on the edge and end up with | |
535 | * the tick timer in the softirq ! The calling site | |
536 | * takes care of this. | |
537 | */ | |
538 | return 1; | |
539 | case HRTIMER_CB_IRQSAFE: | |
540 | case HRTIMER_CB_SOFTIRQ: | |
541 | /* | |
542 | * Move everything else into the softirq pending list ! | |
543 | */ | |
544 | list_add_tail(&timer->cb_entry, | |
545 | &base->cpu_base->cb_pending); | |
546 | timer->state = HRTIMER_STATE_PENDING; | |
547 | raise_softirq(HRTIMER_SOFTIRQ); | |
548 | return 1; | |
549 | default: | |
550 | BUG(); | |
551 | } | |
552 | } | |
553 | return 0; | |
554 | } | |
555 | ||
556 | /* | |
557 | * Switch to high resolution mode | |
558 | */ | |
f8953856 | 559 | static int hrtimer_switch_to_hres(void) |
54cdfdb4 | 560 | { |
820de5c3 IM |
561 | int cpu = smp_processor_id(); |
562 | struct hrtimer_cpu_base *base = &per_cpu(hrtimer_bases, cpu); | |
54cdfdb4 TG |
563 | unsigned long flags; |
564 | ||
565 | if (base->hres_active) | |
f8953856 | 566 | return 1; |
54cdfdb4 TG |
567 | |
568 | local_irq_save(flags); | |
569 | ||
570 | if (tick_init_highres()) { | |
571 | local_irq_restore(flags); | |
820de5c3 IM |
572 | printk(KERN_WARNING "Could not switch to high resolution " |
573 | "mode on CPU %d\n", cpu); | |
f8953856 | 574 | return 0; |
54cdfdb4 TG |
575 | } |
576 | base->hres_active = 1; | |
577 | base->clock_base[CLOCK_REALTIME].resolution = KTIME_HIGH_RES; | |
578 | base->clock_base[CLOCK_MONOTONIC].resolution = KTIME_HIGH_RES; | |
579 | ||
580 | tick_setup_sched_timer(); | |
581 | ||
582 | /* "Retrigger" the interrupt to get things going */ | |
583 | retrigger_next_event(NULL); | |
584 | local_irq_restore(flags); | |
585 | printk(KERN_INFO "Switched to high resolution mode on CPU %d\n", | |
586 | smp_processor_id()); | |
f8953856 | 587 | return 1; |
54cdfdb4 TG |
588 | } |
589 | ||
590 | #else | |
591 | ||
592 | static inline int hrtimer_hres_active(void) { return 0; } | |
593 | static inline int hrtimer_is_hres_enabled(void) { return 0; } | |
f8953856 | 594 | static inline int hrtimer_switch_to_hres(void) { return 0; } |
54cdfdb4 TG |
595 | static inline void hrtimer_force_reprogram(struct hrtimer_cpu_base *base) { } |
596 | static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer, | |
597 | struct hrtimer_clock_base *base) | |
598 | { | |
599 | return 0; | |
600 | } | |
601 | static inline int hrtimer_cb_pending(struct hrtimer *timer) { return 0; } | |
602 | static inline void hrtimer_remove_cb_pending(struct hrtimer *timer) { } | |
603 | static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { } | |
604 | static inline void hrtimer_init_timer_hres(struct hrtimer *timer) { } | |
605 | ||
606 | #endif /* CONFIG_HIGH_RES_TIMERS */ | |
607 | ||
82f67cd9 IM |
608 | #ifdef CONFIG_TIMER_STATS |
609 | void __timer_stats_hrtimer_set_start_info(struct hrtimer *timer, void *addr) | |
610 | { | |
611 | if (timer->start_site) | |
612 | return; | |
613 | ||
614 | timer->start_site = addr; | |
615 | memcpy(timer->start_comm, current->comm, TASK_COMM_LEN); | |
616 | timer->start_pid = current->pid; | |
617 | } | |
618 | #endif | |
619 | ||
c0a31329 TG |
620 | /* |
621 | * Counterpart to lock_timer_base above: | |
622 | */ | |
623 | static inline | |
624 | void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) | |
625 | { | |
3c8aa39d | 626 | spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags); |
c0a31329 TG |
627 | } |
628 | ||
629 | /** | |
630 | * hrtimer_forward - forward the timer expiry | |
c0a31329 | 631 | * @timer: hrtimer to forward |
44f21475 | 632 | * @now: forward past this time |
c0a31329 TG |
633 | * @interval: the interval to forward |
634 | * | |
635 | * Forward the timer expiry so it will expire in the future. | |
8dca6f33 | 636 | * Returns the number of overruns. |
c0a31329 TG |
637 | */ |
638 | unsigned long | |
44f21475 | 639 | hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval) |
c0a31329 TG |
640 | { |
641 | unsigned long orun = 1; | |
44f21475 | 642 | ktime_t delta; |
c0a31329 TG |
643 | |
644 | delta = ktime_sub(now, timer->expires); | |
645 | ||
646 | if (delta.tv64 < 0) | |
647 | return 0; | |
648 | ||
c9db4fa1 TG |
649 | if (interval.tv64 < timer->base->resolution.tv64) |
650 | interval.tv64 = timer->base->resolution.tv64; | |
651 | ||
c0a31329 | 652 | if (unlikely(delta.tv64 >= interval.tv64)) { |
df869b63 | 653 | s64 incr = ktime_to_ns(interval); |
c0a31329 TG |
654 | |
655 | orun = ktime_divns(delta, incr); | |
656 | timer->expires = ktime_add_ns(timer->expires, incr * orun); | |
657 | if (timer->expires.tv64 > now.tv64) | |
658 | return orun; | |
659 | /* | |
660 | * This (and the ktime_add() below) is the | |
661 | * correction for exact: | |
662 | */ | |
663 | orun++; | |
664 | } | |
665 | timer->expires = ktime_add(timer->expires, interval); | |
13788ccc TG |
666 | /* |
667 | * Make sure, that the result did not wrap with a very large | |
668 | * interval. | |
669 | */ | |
670 | if (timer->expires.tv64 < 0) | |
671 | timer->expires = ktime_set(KTIME_SEC_MAX, 0); | |
c0a31329 TG |
672 | |
673 | return orun; | |
674 | } | |
6bdb6b62 | 675 | EXPORT_SYMBOL_GPL(hrtimer_forward); |
c0a31329 TG |
676 | |
677 | /* | |
678 | * enqueue_hrtimer - internal function to (re)start a timer | |
679 | * | |
680 | * The timer is inserted in expiry order. Insertion into the | |
681 | * red black tree is O(log(n)). Must hold the base lock. | |
682 | */ | |
3c8aa39d | 683 | static void enqueue_hrtimer(struct hrtimer *timer, |
54cdfdb4 | 684 | struct hrtimer_clock_base *base, int reprogram) |
c0a31329 TG |
685 | { |
686 | struct rb_node **link = &base->active.rb_node; | |
c0a31329 TG |
687 | struct rb_node *parent = NULL; |
688 | struct hrtimer *entry; | |
99bc2fcb | 689 | int leftmost = 1; |
c0a31329 TG |
690 | |
691 | /* | |
692 | * Find the right place in the rbtree: | |
693 | */ | |
694 | while (*link) { | |
695 | parent = *link; | |
696 | entry = rb_entry(parent, struct hrtimer, node); | |
697 | /* | |
698 | * We dont care about collisions. Nodes with | |
699 | * the same expiry time stay together. | |
700 | */ | |
99bc2fcb | 701 | if (timer->expires.tv64 < entry->expires.tv64) { |
c0a31329 | 702 | link = &(*link)->rb_left; |
99bc2fcb | 703 | } else { |
c0a31329 | 704 | link = &(*link)->rb_right; |
99bc2fcb IM |
705 | leftmost = 0; |
706 | } | |
c0a31329 TG |
707 | } |
708 | ||
709 | /* | |
288867ec TG |
710 | * Insert the timer to the rbtree and check whether it |
711 | * replaces the first pending timer | |
c0a31329 | 712 | */ |
99bc2fcb | 713 | if (leftmost) { |
54cdfdb4 TG |
714 | /* |
715 | * Reprogram the clock event device. When the timer is already | |
716 | * expired hrtimer_enqueue_reprogram has either called the | |
717 | * callback or added it to the pending list and raised the | |
718 | * softirq. | |
719 | * | |
720 | * This is a NOP for !HIGHRES | |
721 | */ | |
722 | if (reprogram && hrtimer_enqueue_reprogram(timer, base)) | |
723 | return; | |
724 | ||
725 | base->first = &timer->node; | |
726 | } | |
727 | ||
c0a31329 TG |
728 | rb_link_node(&timer->node, parent, link); |
729 | rb_insert_color(&timer->node, &base->active); | |
303e967f TG |
730 | /* |
731 | * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the | |
732 | * state of a possibly running callback. | |
733 | */ | |
734 | timer->state |= HRTIMER_STATE_ENQUEUED; | |
288867ec | 735 | } |
c0a31329 TG |
736 | |
737 | /* | |
738 | * __remove_hrtimer - internal function to remove a timer | |
739 | * | |
740 | * Caller must hold the base lock. | |
54cdfdb4 TG |
741 | * |
742 | * High resolution timer mode reprograms the clock event device when the | |
743 | * timer is the one which expires next. The caller can disable this by setting | |
744 | * reprogram to zero. This is useful, when the context does a reprogramming | |
745 | * anyway (e.g. timer interrupt) | |
c0a31329 | 746 | */ |
3c8aa39d | 747 | static void __remove_hrtimer(struct hrtimer *timer, |
303e967f | 748 | struct hrtimer_clock_base *base, |
54cdfdb4 | 749 | unsigned long newstate, int reprogram) |
c0a31329 | 750 | { |
54cdfdb4 TG |
751 | /* High res. callback list. NOP for !HIGHRES */ |
752 | if (hrtimer_cb_pending(timer)) | |
753 | hrtimer_remove_cb_pending(timer); | |
754 | else { | |
755 | /* | |
756 | * Remove the timer from the rbtree and replace the | |
757 | * first entry pointer if necessary. | |
758 | */ | |
759 | if (base->first == &timer->node) { | |
760 | base->first = rb_next(&timer->node); | |
761 | /* Reprogram the clock event device. if enabled */ | |
762 | if (reprogram && hrtimer_hres_active()) | |
763 | hrtimer_force_reprogram(base->cpu_base); | |
764 | } | |
765 | rb_erase(&timer->node, &base->active); | |
766 | } | |
303e967f | 767 | timer->state = newstate; |
c0a31329 TG |
768 | } |
769 | ||
770 | /* | |
771 | * remove hrtimer, called with base lock held | |
772 | */ | |
773 | static inline int | |
3c8aa39d | 774 | remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base) |
c0a31329 | 775 | { |
303e967f | 776 | if (hrtimer_is_queued(timer)) { |
54cdfdb4 TG |
777 | int reprogram; |
778 | ||
779 | /* | |
780 | * Remove the timer and force reprogramming when high | |
781 | * resolution mode is active and the timer is on the current | |
782 | * CPU. If we remove a timer on another CPU, reprogramming is | |
783 | * skipped. The interrupt event on this CPU is fired and | |
784 | * reprogramming happens in the interrupt handler. This is a | |
785 | * rare case and less expensive than a smp call. | |
786 | */ | |
82f67cd9 | 787 | timer_stats_hrtimer_clear_start_info(timer); |
54cdfdb4 TG |
788 | reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases); |
789 | __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE, | |
790 | reprogram); | |
c0a31329 TG |
791 | return 1; |
792 | } | |
793 | return 0; | |
794 | } | |
795 | ||
796 | /** | |
797 | * hrtimer_start - (re)start an relative timer on the current CPU | |
c0a31329 TG |
798 | * @timer: the timer to be added |
799 | * @tim: expiry time | |
800 | * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL) | |
801 | * | |
802 | * Returns: | |
803 | * 0 on success | |
804 | * 1 when the timer was active | |
805 | */ | |
806 | int | |
807 | hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode) | |
808 | { | |
3c8aa39d | 809 | struct hrtimer_clock_base *base, *new_base; |
c0a31329 TG |
810 | unsigned long flags; |
811 | int ret; | |
812 | ||
813 | base = lock_hrtimer_base(timer, &flags); | |
814 | ||
815 | /* Remove an active timer from the queue: */ | |
816 | ret = remove_hrtimer(timer, base); | |
817 | ||
818 | /* Switch the timer base, if necessary: */ | |
819 | new_base = switch_hrtimer_base(timer, base); | |
820 | ||
c9cb2e3d | 821 | if (mode == HRTIMER_MODE_REL) { |
c0a31329 | 822 | tim = ktime_add(tim, new_base->get_time()); |
06027bdd IM |
823 | /* |
824 | * CONFIG_TIME_LOW_RES is a temporary way for architectures | |
825 | * to signal that they simply return xtime in | |
826 | * do_gettimeoffset(). In this case we want to round up by | |
827 | * resolution when starting a relative timer, to avoid short | |
828 | * timeouts. This will go away with the GTOD framework. | |
829 | */ | |
830 | #ifdef CONFIG_TIME_LOW_RES | |
831 | tim = ktime_add(tim, base->resolution); | |
832 | #endif | |
833 | } | |
c0a31329 TG |
834 | timer->expires = tim; |
835 | ||
82f67cd9 IM |
836 | timer_stats_hrtimer_set_start_info(timer); |
837 | ||
935c631d IM |
838 | /* |
839 | * Only allow reprogramming if the new base is on this CPU. | |
840 | * (it might still be on another CPU if the timer was pending) | |
841 | */ | |
842 | enqueue_hrtimer(timer, new_base, | |
843 | new_base->cpu_base == &__get_cpu_var(hrtimer_bases)); | |
c0a31329 TG |
844 | |
845 | unlock_hrtimer_base(timer, &flags); | |
846 | ||
847 | return ret; | |
848 | } | |
8d16b764 | 849 | EXPORT_SYMBOL_GPL(hrtimer_start); |
c0a31329 TG |
850 | |
851 | /** | |
852 | * hrtimer_try_to_cancel - try to deactivate a timer | |
c0a31329 TG |
853 | * @timer: hrtimer to stop |
854 | * | |
855 | * Returns: | |
856 | * 0 when the timer was not active | |
857 | * 1 when the timer was active | |
858 | * -1 when the timer is currently excuting the callback function and | |
fa9799e3 | 859 | * cannot be stopped |
c0a31329 TG |
860 | */ |
861 | int hrtimer_try_to_cancel(struct hrtimer *timer) | |
862 | { | |
3c8aa39d | 863 | struct hrtimer_clock_base *base; |
c0a31329 TG |
864 | unsigned long flags; |
865 | int ret = -1; | |
866 | ||
867 | base = lock_hrtimer_base(timer, &flags); | |
868 | ||
303e967f | 869 | if (!hrtimer_callback_running(timer)) |
c0a31329 TG |
870 | ret = remove_hrtimer(timer, base); |
871 | ||
872 | unlock_hrtimer_base(timer, &flags); | |
873 | ||
874 | return ret; | |
875 | ||
876 | } | |
8d16b764 | 877 | EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel); |
c0a31329 TG |
878 | |
879 | /** | |
880 | * hrtimer_cancel - cancel a timer and wait for the handler to finish. | |
c0a31329 TG |
881 | * @timer: the timer to be cancelled |
882 | * | |
883 | * Returns: | |
884 | * 0 when the timer was not active | |
885 | * 1 when the timer was active | |
886 | */ | |
887 | int hrtimer_cancel(struct hrtimer *timer) | |
888 | { | |
889 | for (;;) { | |
890 | int ret = hrtimer_try_to_cancel(timer); | |
891 | ||
892 | if (ret >= 0) | |
893 | return ret; | |
5ef37b19 | 894 | cpu_relax(); |
c0a31329 TG |
895 | } |
896 | } | |
8d16b764 | 897 | EXPORT_SYMBOL_GPL(hrtimer_cancel); |
c0a31329 TG |
898 | |
899 | /** | |
900 | * hrtimer_get_remaining - get remaining time for the timer | |
c0a31329 TG |
901 | * @timer: the timer to read |
902 | */ | |
903 | ktime_t hrtimer_get_remaining(const struct hrtimer *timer) | |
904 | { | |
3c8aa39d | 905 | struct hrtimer_clock_base *base; |
c0a31329 TG |
906 | unsigned long flags; |
907 | ktime_t rem; | |
908 | ||
909 | base = lock_hrtimer_base(timer, &flags); | |
3c8aa39d | 910 | rem = ktime_sub(timer->expires, base->get_time()); |
c0a31329 TG |
911 | unlock_hrtimer_base(timer, &flags); |
912 | ||
913 | return rem; | |
914 | } | |
8d16b764 | 915 | EXPORT_SYMBOL_GPL(hrtimer_get_remaining); |
c0a31329 | 916 | |
fd064b9b | 917 | #if defined(CONFIG_NO_IDLE_HZ) || defined(CONFIG_NO_HZ) |
69239749 TL |
918 | /** |
919 | * hrtimer_get_next_event - get the time until next expiry event | |
920 | * | |
921 | * Returns the delta to the next expiry event or KTIME_MAX if no timer | |
922 | * is pending. | |
923 | */ | |
924 | ktime_t hrtimer_get_next_event(void) | |
925 | { | |
3c8aa39d TG |
926 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
927 | struct hrtimer_clock_base *base = cpu_base->clock_base; | |
69239749 TL |
928 | ktime_t delta, mindelta = { .tv64 = KTIME_MAX }; |
929 | unsigned long flags; | |
930 | int i; | |
931 | ||
3c8aa39d TG |
932 | spin_lock_irqsave(&cpu_base->lock, flags); |
933 | ||
54cdfdb4 TG |
934 | if (!hrtimer_hres_active()) { |
935 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) { | |
936 | struct hrtimer *timer; | |
69239749 | 937 | |
54cdfdb4 TG |
938 | if (!base->first) |
939 | continue; | |
3c8aa39d | 940 | |
54cdfdb4 TG |
941 | timer = rb_entry(base->first, struct hrtimer, node); |
942 | delta.tv64 = timer->expires.tv64; | |
943 | delta = ktime_sub(delta, base->get_time()); | |
944 | if (delta.tv64 < mindelta.tv64) | |
945 | mindelta.tv64 = delta.tv64; | |
946 | } | |
69239749 | 947 | } |
3c8aa39d TG |
948 | |
949 | spin_unlock_irqrestore(&cpu_base->lock, flags); | |
950 | ||
69239749 TL |
951 | if (mindelta.tv64 < 0) |
952 | mindelta.tv64 = 0; | |
953 | return mindelta; | |
954 | } | |
955 | #endif | |
956 | ||
c0a31329 | 957 | /** |
7978672c | 958 | * hrtimer_init - initialize a timer to the given clock |
7978672c | 959 | * @timer: the timer to be initialized |
c0a31329 | 960 | * @clock_id: the clock to be used |
7978672c | 961 | * @mode: timer mode abs/rel |
c0a31329 | 962 | */ |
7978672c GA |
963 | void hrtimer_init(struct hrtimer *timer, clockid_t clock_id, |
964 | enum hrtimer_mode mode) | |
c0a31329 | 965 | { |
3c8aa39d | 966 | struct hrtimer_cpu_base *cpu_base; |
c0a31329 | 967 | |
7978672c GA |
968 | memset(timer, 0, sizeof(struct hrtimer)); |
969 | ||
3c8aa39d | 970 | cpu_base = &__raw_get_cpu_var(hrtimer_bases); |
c0a31329 | 971 | |
c9cb2e3d | 972 | if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS) |
7978672c GA |
973 | clock_id = CLOCK_MONOTONIC; |
974 | ||
3c8aa39d | 975 | timer->base = &cpu_base->clock_base[clock_id]; |
54cdfdb4 | 976 | hrtimer_init_timer_hres(timer); |
82f67cd9 IM |
977 | |
978 | #ifdef CONFIG_TIMER_STATS | |
979 | timer->start_site = NULL; | |
980 | timer->start_pid = -1; | |
981 | memset(timer->start_comm, 0, TASK_COMM_LEN); | |
982 | #endif | |
c0a31329 | 983 | } |
8d16b764 | 984 | EXPORT_SYMBOL_GPL(hrtimer_init); |
c0a31329 TG |
985 | |
986 | /** | |
987 | * hrtimer_get_res - get the timer resolution for a clock | |
c0a31329 TG |
988 | * @which_clock: which clock to query |
989 | * @tp: pointer to timespec variable to store the resolution | |
990 | * | |
72fd4a35 RD |
991 | * Store the resolution of the clock selected by @which_clock in the |
992 | * variable pointed to by @tp. | |
c0a31329 TG |
993 | */ |
994 | int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp) | |
995 | { | |
3c8aa39d | 996 | struct hrtimer_cpu_base *cpu_base; |
c0a31329 | 997 | |
3c8aa39d TG |
998 | cpu_base = &__raw_get_cpu_var(hrtimer_bases); |
999 | *tp = ktime_to_timespec(cpu_base->clock_base[which_clock].resolution); | |
c0a31329 TG |
1000 | |
1001 | return 0; | |
1002 | } | |
8d16b764 | 1003 | EXPORT_SYMBOL_GPL(hrtimer_get_res); |
c0a31329 | 1004 | |
54cdfdb4 TG |
1005 | #ifdef CONFIG_HIGH_RES_TIMERS |
1006 | ||
1007 | /* | |
1008 | * High resolution timer interrupt | |
1009 | * Called with interrupts disabled | |
1010 | */ | |
1011 | void hrtimer_interrupt(struct clock_event_device *dev) | |
1012 | { | |
1013 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); | |
1014 | struct hrtimer_clock_base *base; | |
1015 | ktime_t expires_next, now; | |
1016 | int i, raise = 0; | |
1017 | ||
1018 | BUG_ON(!cpu_base->hres_active); | |
1019 | cpu_base->nr_events++; | |
1020 | dev->next_event.tv64 = KTIME_MAX; | |
1021 | ||
1022 | retry: | |
1023 | now = ktime_get(); | |
1024 | ||
1025 | expires_next.tv64 = KTIME_MAX; | |
1026 | ||
1027 | base = cpu_base->clock_base; | |
1028 | ||
1029 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { | |
1030 | ktime_t basenow; | |
1031 | struct rb_node *node; | |
1032 | ||
1033 | spin_lock(&cpu_base->lock); | |
1034 | ||
1035 | basenow = ktime_add(now, base->offset); | |
1036 | ||
1037 | while ((node = base->first)) { | |
1038 | struct hrtimer *timer; | |
1039 | ||
1040 | timer = rb_entry(node, struct hrtimer, node); | |
1041 | ||
1042 | if (basenow.tv64 < timer->expires.tv64) { | |
1043 | ktime_t expires; | |
1044 | ||
1045 | expires = ktime_sub(timer->expires, | |
1046 | base->offset); | |
1047 | if (expires.tv64 < expires_next.tv64) | |
1048 | expires_next = expires; | |
1049 | break; | |
1050 | } | |
1051 | ||
1052 | /* Move softirq callbacks to the pending list */ | |
1053 | if (timer->cb_mode == HRTIMER_CB_SOFTIRQ) { | |
1054 | __remove_hrtimer(timer, base, | |
1055 | HRTIMER_STATE_PENDING, 0); | |
1056 | list_add_tail(&timer->cb_entry, | |
1057 | &base->cpu_base->cb_pending); | |
1058 | raise = 1; | |
1059 | continue; | |
1060 | } | |
1061 | ||
1062 | __remove_hrtimer(timer, base, | |
1063 | HRTIMER_STATE_CALLBACK, 0); | |
82f67cd9 | 1064 | timer_stats_account_hrtimer(timer); |
54cdfdb4 TG |
1065 | |
1066 | /* | |
1067 | * Note: We clear the CALLBACK bit after | |
1068 | * enqueue_hrtimer to avoid reprogramming of | |
1069 | * the event hardware. This happens at the end | |
1070 | * of this function anyway. | |
1071 | */ | |
1072 | if (timer->function(timer) != HRTIMER_NORESTART) { | |
1073 | BUG_ON(timer->state != HRTIMER_STATE_CALLBACK); | |
1074 | enqueue_hrtimer(timer, base, 0); | |
1075 | } | |
1076 | timer->state &= ~HRTIMER_STATE_CALLBACK; | |
1077 | } | |
1078 | spin_unlock(&cpu_base->lock); | |
1079 | base++; | |
1080 | } | |
1081 | ||
1082 | cpu_base->expires_next = expires_next; | |
1083 | ||
1084 | /* Reprogramming necessary ? */ | |
1085 | if (expires_next.tv64 != KTIME_MAX) { | |
1086 | if (tick_program_event(expires_next, 0)) | |
1087 | goto retry; | |
1088 | } | |
1089 | ||
1090 | /* Raise softirq ? */ | |
1091 | if (raise) | |
1092 | raise_softirq(HRTIMER_SOFTIRQ); | |
1093 | } | |
1094 | ||
1095 | static void run_hrtimer_softirq(struct softirq_action *h) | |
1096 | { | |
1097 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); | |
1098 | ||
1099 | spin_lock_irq(&cpu_base->lock); | |
1100 | ||
1101 | while (!list_empty(&cpu_base->cb_pending)) { | |
1102 | enum hrtimer_restart (*fn)(struct hrtimer *); | |
1103 | struct hrtimer *timer; | |
1104 | int restart; | |
1105 | ||
1106 | timer = list_entry(cpu_base->cb_pending.next, | |
1107 | struct hrtimer, cb_entry); | |
1108 | ||
82f67cd9 IM |
1109 | timer_stats_account_hrtimer(timer); |
1110 | ||
54cdfdb4 TG |
1111 | fn = timer->function; |
1112 | __remove_hrtimer(timer, timer->base, HRTIMER_STATE_CALLBACK, 0); | |
1113 | spin_unlock_irq(&cpu_base->lock); | |
1114 | ||
1115 | restart = fn(timer); | |
1116 | ||
1117 | spin_lock_irq(&cpu_base->lock); | |
1118 | ||
1119 | timer->state &= ~HRTIMER_STATE_CALLBACK; | |
1120 | if (restart == HRTIMER_RESTART) { | |
1121 | BUG_ON(hrtimer_active(timer)); | |
1122 | /* | |
1123 | * Enqueue the timer, allow reprogramming of the event | |
1124 | * device | |
1125 | */ | |
1126 | enqueue_hrtimer(timer, timer->base, 1); | |
1127 | } else if (hrtimer_active(timer)) { | |
1128 | /* | |
1129 | * If the timer was rearmed on another CPU, reprogram | |
1130 | * the event device. | |
1131 | */ | |
1132 | if (timer->base->first == &timer->node) | |
1133 | hrtimer_reprogram(timer, timer->base); | |
1134 | } | |
1135 | } | |
1136 | spin_unlock_irq(&cpu_base->lock); | |
1137 | } | |
1138 | ||
1139 | #endif /* CONFIG_HIGH_RES_TIMERS */ | |
1140 | ||
c0a31329 TG |
1141 | /* |
1142 | * Expire the per base hrtimer-queue: | |
1143 | */ | |
3c8aa39d TG |
1144 | static inline void run_hrtimer_queue(struct hrtimer_cpu_base *cpu_base, |
1145 | int index) | |
c0a31329 | 1146 | { |
288867ec | 1147 | struct rb_node *node; |
3c8aa39d | 1148 | struct hrtimer_clock_base *base = &cpu_base->clock_base[index]; |
c0a31329 | 1149 | |
3055adda DS |
1150 | if (!base->first) |
1151 | return; | |
1152 | ||
92127c7a TG |
1153 | if (base->get_softirq_time) |
1154 | base->softirq_time = base->get_softirq_time(); | |
1155 | ||
3c8aa39d | 1156 | spin_lock_irq(&cpu_base->lock); |
c0a31329 | 1157 | |
288867ec | 1158 | while ((node = base->first)) { |
c0a31329 | 1159 | struct hrtimer *timer; |
c9cb2e3d | 1160 | enum hrtimer_restart (*fn)(struct hrtimer *); |
c0a31329 | 1161 | int restart; |
c0a31329 | 1162 | |
288867ec | 1163 | timer = rb_entry(node, struct hrtimer, node); |
92127c7a | 1164 | if (base->softirq_time.tv64 <= timer->expires.tv64) |
c0a31329 TG |
1165 | break; |
1166 | ||
f8953856 TG |
1167 | #ifdef CONFIG_HIGH_RES_TIMERS |
1168 | WARN_ON_ONCE(timer->cb_mode == HRTIMER_CB_IRQSAFE_NO_SOFTIRQ); | |
1169 | #endif | |
82f67cd9 IM |
1170 | timer_stats_account_hrtimer(timer); |
1171 | ||
c0a31329 | 1172 | fn = timer->function; |
54cdfdb4 | 1173 | __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0); |
3c8aa39d | 1174 | spin_unlock_irq(&cpu_base->lock); |
c0a31329 | 1175 | |
05cfb614 | 1176 | restart = fn(timer); |
c0a31329 | 1177 | |
3c8aa39d | 1178 | spin_lock_irq(&cpu_base->lock); |
c0a31329 | 1179 | |
303e967f | 1180 | timer->state &= ~HRTIMER_STATE_CALLBACK; |
b75f7a51 RZ |
1181 | if (restart != HRTIMER_NORESTART) { |
1182 | BUG_ON(hrtimer_active(timer)); | |
54cdfdb4 | 1183 | enqueue_hrtimer(timer, base, 0); |
b75f7a51 | 1184 | } |
c0a31329 | 1185 | } |
3c8aa39d | 1186 | spin_unlock_irq(&cpu_base->lock); |
c0a31329 TG |
1187 | } |
1188 | ||
1189 | /* | |
1190 | * Called from timer softirq every jiffy, expire hrtimers: | |
54cdfdb4 TG |
1191 | * |
1192 | * For HRT its the fall back code to run the softirq in the timer | |
1193 | * softirq context in case the hrtimer initialization failed or has | |
1194 | * not been done yet. | |
c0a31329 TG |
1195 | */ |
1196 | void hrtimer_run_queues(void) | |
1197 | { | |
3c8aa39d | 1198 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
c0a31329 TG |
1199 | int i; |
1200 | ||
54cdfdb4 TG |
1201 | if (hrtimer_hres_active()) |
1202 | return; | |
1203 | ||
79bf2bb3 TG |
1204 | /* |
1205 | * This _is_ ugly: We have to check in the softirq context, | |
1206 | * whether we can switch to highres and / or nohz mode. The | |
1207 | * clocksource switch happens in the timer interrupt with | |
1208 | * xtime_lock held. Notification from there only sets the | |
1209 | * check bit in the tick_oneshot code, otherwise we might | |
1210 | * deadlock vs. xtime_lock. | |
1211 | */ | |
54cdfdb4 | 1212 | if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) |
f8953856 TG |
1213 | if (hrtimer_switch_to_hres()) |
1214 | return; | |
79bf2bb3 | 1215 | |
3c8aa39d | 1216 | hrtimer_get_softirq_time(cpu_base); |
92127c7a | 1217 | |
3c8aa39d TG |
1218 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) |
1219 | run_hrtimer_queue(cpu_base, i); | |
c0a31329 TG |
1220 | } |
1221 | ||
10c94ec1 TG |
1222 | /* |
1223 | * Sleep related functions: | |
1224 | */ | |
c9cb2e3d | 1225 | static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer) |
00362e33 TG |
1226 | { |
1227 | struct hrtimer_sleeper *t = | |
1228 | container_of(timer, struct hrtimer_sleeper, timer); | |
1229 | struct task_struct *task = t->task; | |
1230 | ||
1231 | t->task = NULL; | |
1232 | if (task) | |
1233 | wake_up_process(task); | |
1234 | ||
1235 | return HRTIMER_NORESTART; | |
1236 | } | |
1237 | ||
36c8b586 | 1238 | void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task) |
00362e33 TG |
1239 | { |
1240 | sl->timer.function = hrtimer_wakeup; | |
1241 | sl->task = task; | |
54cdfdb4 TG |
1242 | #ifdef CONFIG_HIGH_RES_TIMERS |
1243 | sl->timer.cb_mode = HRTIMER_CB_IRQSAFE_NO_RESTART; | |
1244 | #endif | |
00362e33 TG |
1245 | } |
1246 | ||
669d7868 | 1247 | static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode) |
432569bb | 1248 | { |
669d7868 | 1249 | hrtimer_init_sleeper(t, current); |
10c94ec1 | 1250 | |
432569bb RZ |
1251 | do { |
1252 | set_current_state(TASK_INTERRUPTIBLE); | |
1253 | hrtimer_start(&t->timer, t->timer.expires, mode); | |
1254 | ||
54cdfdb4 TG |
1255 | if (likely(t->task)) |
1256 | schedule(); | |
432569bb | 1257 | |
669d7868 | 1258 | hrtimer_cancel(&t->timer); |
c9cb2e3d | 1259 | mode = HRTIMER_MODE_ABS; |
669d7868 TG |
1260 | |
1261 | } while (t->task && !signal_pending(current)); | |
432569bb | 1262 | |
669d7868 | 1263 | return t->task == NULL; |
10c94ec1 TG |
1264 | } |
1265 | ||
1711ef38 | 1266 | long __sched hrtimer_nanosleep_restart(struct restart_block *restart) |
10c94ec1 | 1267 | { |
669d7868 | 1268 | struct hrtimer_sleeper t; |
ea13dbc8 IM |
1269 | struct timespec __user *rmtp; |
1270 | struct timespec tu; | |
432569bb | 1271 | ktime_t time; |
10c94ec1 TG |
1272 | |
1273 | restart->fn = do_no_restart_syscall; | |
1274 | ||
c9cb2e3d | 1275 | hrtimer_init(&t.timer, restart->arg0, HRTIMER_MODE_ABS); |
1711ef38 | 1276 | t.timer.expires.tv64 = ((u64)restart->arg3 << 32) | (u64) restart->arg2; |
10c94ec1 | 1277 | |
c9cb2e3d | 1278 | if (do_nanosleep(&t, HRTIMER_MODE_ABS)) |
10c94ec1 TG |
1279 | return 0; |
1280 | ||
1711ef38 | 1281 | rmtp = (struct timespec __user *) restart->arg1; |
432569bb RZ |
1282 | if (rmtp) { |
1283 | time = ktime_sub(t.timer.expires, t.timer.base->get_time()); | |
1284 | if (time.tv64 <= 0) | |
1285 | return 0; | |
1286 | tu = ktime_to_timespec(time); | |
1287 | if (copy_to_user(rmtp, &tu, sizeof(tu))) | |
1288 | return -EFAULT; | |
1289 | } | |
10c94ec1 | 1290 | |
1711ef38 | 1291 | restart->fn = hrtimer_nanosleep_restart; |
10c94ec1 TG |
1292 | |
1293 | /* The other values in restart are already filled in */ | |
1294 | return -ERESTART_RESTARTBLOCK; | |
1295 | } | |
1296 | ||
10c94ec1 TG |
1297 | long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp, |
1298 | const enum hrtimer_mode mode, const clockid_t clockid) | |
1299 | { | |
1300 | struct restart_block *restart; | |
669d7868 | 1301 | struct hrtimer_sleeper t; |
10c94ec1 TG |
1302 | struct timespec tu; |
1303 | ktime_t rem; | |
1304 | ||
432569bb RZ |
1305 | hrtimer_init(&t.timer, clockid, mode); |
1306 | t.timer.expires = timespec_to_ktime(*rqtp); | |
1307 | if (do_nanosleep(&t, mode)) | |
10c94ec1 TG |
1308 | return 0; |
1309 | ||
7978672c | 1310 | /* Absolute timers do not update the rmtp value and restart: */ |
c9cb2e3d | 1311 | if (mode == HRTIMER_MODE_ABS) |
10c94ec1 TG |
1312 | return -ERESTARTNOHAND; |
1313 | ||
432569bb RZ |
1314 | if (rmtp) { |
1315 | rem = ktime_sub(t.timer.expires, t.timer.base->get_time()); | |
1316 | if (rem.tv64 <= 0) | |
1317 | return 0; | |
1318 | tu = ktime_to_timespec(rem); | |
1319 | if (copy_to_user(rmtp, &tu, sizeof(tu))) | |
1320 | return -EFAULT; | |
1321 | } | |
10c94ec1 TG |
1322 | |
1323 | restart = ¤t_thread_info()->restart_block; | |
1711ef38 TA |
1324 | restart->fn = hrtimer_nanosleep_restart; |
1325 | restart->arg0 = (unsigned long) t.timer.base->index; | |
1326 | restart->arg1 = (unsigned long) rmtp; | |
1327 | restart->arg2 = t.timer.expires.tv64 & 0xFFFFFFFF; | |
1328 | restart->arg3 = t.timer.expires.tv64 >> 32; | |
10c94ec1 TG |
1329 | |
1330 | return -ERESTART_RESTARTBLOCK; | |
1331 | } | |
1332 | ||
6ba1b912 TG |
1333 | asmlinkage long |
1334 | sys_nanosleep(struct timespec __user *rqtp, struct timespec __user *rmtp) | |
1335 | { | |
1336 | struct timespec tu; | |
1337 | ||
1338 | if (copy_from_user(&tu, rqtp, sizeof(tu))) | |
1339 | return -EFAULT; | |
1340 | ||
1341 | if (!timespec_valid(&tu)) | |
1342 | return -EINVAL; | |
1343 | ||
c9cb2e3d | 1344 | return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC); |
6ba1b912 TG |
1345 | } |
1346 | ||
c0a31329 TG |
1347 | /* |
1348 | * Functions related to boot-time initialization: | |
1349 | */ | |
1350 | static void __devinit init_hrtimers_cpu(int cpu) | |
1351 | { | |
3c8aa39d | 1352 | struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu); |
c0a31329 TG |
1353 | int i; |
1354 | ||
3c8aa39d TG |
1355 | spin_lock_init(&cpu_base->lock); |
1356 | lockdep_set_class(&cpu_base->lock, &cpu_base->lock_key); | |
1357 | ||
1358 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) | |
1359 | cpu_base->clock_base[i].cpu_base = cpu_base; | |
1360 | ||
54cdfdb4 | 1361 | hrtimer_init_hres(cpu_base); |
c0a31329 TG |
1362 | } |
1363 | ||
1364 | #ifdef CONFIG_HOTPLUG_CPU | |
1365 | ||
3c8aa39d TG |
1366 | static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base, |
1367 | struct hrtimer_clock_base *new_base) | |
c0a31329 TG |
1368 | { |
1369 | struct hrtimer *timer; | |
1370 | struct rb_node *node; | |
1371 | ||
1372 | while ((node = rb_first(&old_base->active))) { | |
1373 | timer = rb_entry(node, struct hrtimer, node); | |
54cdfdb4 TG |
1374 | BUG_ON(hrtimer_callback_running(timer)); |
1375 | __remove_hrtimer(timer, old_base, HRTIMER_STATE_INACTIVE, 0); | |
c0a31329 | 1376 | timer->base = new_base; |
54cdfdb4 TG |
1377 | /* |
1378 | * Enqueue the timer. Allow reprogramming of the event device | |
1379 | */ | |
1380 | enqueue_hrtimer(timer, new_base, 1); | |
c0a31329 TG |
1381 | } |
1382 | } | |
1383 | ||
1384 | static void migrate_hrtimers(int cpu) | |
1385 | { | |
3c8aa39d | 1386 | struct hrtimer_cpu_base *old_base, *new_base; |
c0a31329 TG |
1387 | int i; |
1388 | ||
1389 | BUG_ON(cpu_online(cpu)); | |
3c8aa39d TG |
1390 | old_base = &per_cpu(hrtimer_bases, cpu); |
1391 | new_base = &get_cpu_var(hrtimer_bases); | |
c0a31329 | 1392 | |
54cdfdb4 TG |
1393 | tick_cancel_sched_timer(cpu); |
1394 | ||
c0a31329 | 1395 | local_irq_disable(); |
e81ce1f7 HC |
1396 | double_spin_lock(&new_base->lock, &old_base->lock, |
1397 | smp_processor_id() < cpu); | |
c0a31329 | 1398 | |
3c8aa39d | 1399 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
3c8aa39d TG |
1400 | migrate_hrtimer_list(&old_base->clock_base[i], |
1401 | &new_base->clock_base[i]); | |
c0a31329 TG |
1402 | } |
1403 | ||
e81ce1f7 HC |
1404 | double_spin_unlock(&new_base->lock, &old_base->lock, |
1405 | smp_processor_id() < cpu); | |
c0a31329 TG |
1406 | local_irq_enable(); |
1407 | put_cpu_var(hrtimer_bases); | |
1408 | } | |
1409 | #endif /* CONFIG_HOTPLUG_CPU */ | |
1410 | ||
8c78f307 | 1411 | static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self, |
c0a31329 TG |
1412 | unsigned long action, void *hcpu) |
1413 | { | |
7713a7d1 | 1414 | unsigned int cpu = (long)hcpu; |
c0a31329 TG |
1415 | |
1416 | switch (action) { | |
1417 | ||
1418 | case CPU_UP_PREPARE: | |
8bb78442 | 1419 | case CPU_UP_PREPARE_FROZEN: |
c0a31329 TG |
1420 | init_hrtimers_cpu(cpu); |
1421 | break; | |
1422 | ||
1423 | #ifdef CONFIG_HOTPLUG_CPU | |
1424 | case CPU_DEAD: | |
8bb78442 | 1425 | case CPU_DEAD_FROZEN: |
d316c57f | 1426 | clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &cpu); |
c0a31329 TG |
1427 | migrate_hrtimers(cpu); |
1428 | break; | |
1429 | #endif | |
1430 | ||
1431 | default: | |
1432 | break; | |
1433 | } | |
1434 | ||
1435 | return NOTIFY_OK; | |
1436 | } | |
1437 | ||
8c78f307 | 1438 | static struct notifier_block __cpuinitdata hrtimers_nb = { |
c0a31329 TG |
1439 | .notifier_call = hrtimer_cpu_notify, |
1440 | }; | |
1441 | ||
1442 | void __init hrtimers_init(void) | |
1443 | { | |
1444 | hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE, | |
1445 | (void *)(long)smp_processor_id()); | |
1446 | register_cpu_notifier(&hrtimers_nb); | |
54cdfdb4 TG |
1447 | #ifdef CONFIG_HIGH_RES_TIMERS |
1448 | open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq, NULL); | |
1449 | #endif | |
c0a31329 TG |
1450 | } |
1451 |