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 | } | |
62 | ||
63 | /** | |
64 | * ktime_get_real - get the real (wall-) time in ktime_t format | |
65 | * | |
66 | * returns the time in ktime_t format | |
67 | */ | |
d316c57f | 68 | ktime_t ktime_get_real(void) |
c0a31329 TG |
69 | { |
70 | struct timespec now; | |
71 | ||
72 | getnstimeofday(&now); | |
73 | ||
74 | return timespec_to_ktime(now); | |
75 | } | |
76 | ||
77 | EXPORT_SYMBOL_GPL(ktime_get_real); | |
78 | ||
79 | /* | |
80 | * The timer bases: | |
7978672c GA |
81 | * |
82 | * Note: If we want to add new timer bases, we have to skip the two | |
83 | * clock ids captured by the cpu-timers. We do this by holding empty | |
84 | * entries rather than doing math adjustment of the clock ids. | |
85 | * This ensures that we capture erroneous accesses to these clock ids | |
86 | * rather than moving them into the range of valid clock id's. | |
c0a31329 | 87 | */ |
54cdfdb4 | 88 | DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) = |
c0a31329 | 89 | { |
3c8aa39d TG |
90 | |
91 | .clock_base = | |
c0a31329 | 92 | { |
3c8aa39d TG |
93 | { |
94 | .index = CLOCK_REALTIME, | |
95 | .get_time = &ktime_get_real, | |
54cdfdb4 | 96 | .resolution = KTIME_LOW_RES, |
3c8aa39d TG |
97 | }, |
98 | { | |
99 | .index = CLOCK_MONOTONIC, | |
100 | .get_time = &ktime_get, | |
54cdfdb4 | 101 | .resolution = KTIME_LOW_RES, |
3c8aa39d TG |
102 | }, |
103 | } | |
c0a31329 TG |
104 | }; |
105 | ||
106 | /** | |
107 | * ktime_get_ts - get the monotonic clock in timespec format | |
c0a31329 TG |
108 | * @ts: pointer to timespec variable |
109 | * | |
110 | * The function calculates the monotonic clock from the realtime | |
111 | * clock and the wall_to_monotonic offset and stores the result | |
72fd4a35 | 112 | * in normalized timespec format in the variable pointed to by @ts. |
c0a31329 TG |
113 | */ |
114 | void ktime_get_ts(struct timespec *ts) | |
115 | { | |
116 | struct timespec tomono; | |
117 | unsigned long seq; | |
118 | ||
119 | do { | |
120 | seq = read_seqbegin(&xtime_lock); | |
121 | getnstimeofday(ts); | |
122 | tomono = wall_to_monotonic; | |
123 | ||
124 | } while (read_seqretry(&xtime_lock, seq)); | |
125 | ||
126 | set_normalized_timespec(ts, ts->tv_sec + tomono.tv_sec, | |
127 | ts->tv_nsec + tomono.tv_nsec); | |
128 | } | |
69778e32 | 129 | EXPORT_SYMBOL_GPL(ktime_get_ts); |
c0a31329 | 130 | |
92127c7a TG |
131 | /* |
132 | * Get the coarse grained time at the softirq based on xtime and | |
133 | * wall_to_monotonic. | |
134 | */ | |
3c8aa39d | 135 | static void hrtimer_get_softirq_time(struct hrtimer_cpu_base *base) |
92127c7a TG |
136 | { |
137 | ktime_t xtim, tomono; | |
ad28d94a | 138 | struct timespec xts, tom; |
92127c7a TG |
139 | unsigned long seq; |
140 | ||
141 | do { | |
142 | seq = read_seqbegin(&xtime_lock); | |
f4304ab2 | 143 | #ifdef CONFIG_NO_HZ |
144 | getnstimeofday(&xts); | |
145 | #else | |
146 | xts = xtime; | |
147 | #endif | |
ad28d94a | 148 | tom = wall_to_monotonic; |
92127c7a TG |
149 | } while (read_seqretry(&xtime_lock, seq)); |
150 | ||
f4304ab2 | 151 | xtim = timespec_to_ktime(xts); |
ad28d94a | 152 | tomono = timespec_to_ktime(tom); |
3c8aa39d TG |
153 | base->clock_base[CLOCK_REALTIME].softirq_time = xtim; |
154 | base->clock_base[CLOCK_MONOTONIC].softirq_time = | |
155 | ktime_add(xtim, tomono); | |
92127c7a TG |
156 | } |
157 | ||
303e967f TG |
158 | /* |
159 | * Helper function to check, whether the timer is running the callback | |
160 | * function | |
161 | */ | |
162 | static inline int hrtimer_callback_running(struct hrtimer *timer) | |
163 | { | |
164 | return timer->state & HRTIMER_STATE_CALLBACK; | |
165 | } | |
166 | ||
c0a31329 TG |
167 | /* |
168 | * Functions and macros which are different for UP/SMP systems are kept in a | |
169 | * single place | |
170 | */ | |
171 | #ifdef CONFIG_SMP | |
172 | ||
c0a31329 TG |
173 | /* |
174 | * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock | |
175 | * means that all timers which are tied to this base via timer->base are | |
176 | * locked, and the base itself is locked too. | |
177 | * | |
178 | * So __run_timers/migrate_timers can safely modify all timers which could | |
179 | * be found on the lists/queues. | |
180 | * | |
181 | * When the timer's base is locked, and the timer removed from list, it is | |
182 | * possible to set timer->base = NULL and drop the lock: the timer remains | |
183 | * locked. | |
184 | */ | |
3c8aa39d TG |
185 | static |
186 | struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer, | |
187 | unsigned long *flags) | |
c0a31329 | 188 | { |
3c8aa39d | 189 | struct hrtimer_clock_base *base; |
c0a31329 TG |
190 | |
191 | for (;;) { | |
192 | base = timer->base; | |
193 | if (likely(base != NULL)) { | |
3c8aa39d | 194 | spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
195 | if (likely(base == timer->base)) |
196 | return base; | |
197 | /* The timer has migrated to another CPU: */ | |
3c8aa39d | 198 | spin_unlock_irqrestore(&base->cpu_base->lock, *flags); |
c0a31329 TG |
199 | } |
200 | cpu_relax(); | |
201 | } | |
202 | } | |
203 | ||
204 | /* | |
205 | * Switch the timer base to the current CPU when possible. | |
206 | */ | |
3c8aa39d TG |
207 | static inline struct hrtimer_clock_base * |
208 | switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base) | |
c0a31329 | 209 | { |
3c8aa39d TG |
210 | struct hrtimer_clock_base *new_base; |
211 | struct hrtimer_cpu_base *new_cpu_base; | |
c0a31329 | 212 | |
3c8aa39d TG |
213 | new_cpu_base = &__get_cpu_var(hrtimer_bases); |
214 | new_base = &new_cpu_base->clock_base[base->index]; | |
c0a31329 TG |
215 | |
216 | if (base != new_base) { | |
217 | /* | |
218 | * We are trying to schedule the timer on the local CPU. | |
219 | * However we can't change timer's base while it is running, | |
220 | * so we keep it on the same CPU. No hassle vs. reprogramming | |
221 | * the event source in the high resolution case. The softirq | |
222 | * code will take care of this when the timer function has | |
223 | * completed. There is no conflict as we hold the lock until | |
224 | * the timer is enqueued. | |
225 | */ | |
54cdfdb4 | 226 | if (unlikely(hrtimer_callback_running(timer))) |
c0a31329 TG |
227 | return base; |
228 | ||
229 | /* See the comment in lock_timer_base() */ | |
230 | timer->base = NULL; | |
3c8aa39d TG |
231 | spin_unlock(&base->cpu_base->lock); |
232 | spin_lock(&new_base->cpu_base->lock); | |
c0a31329 TG |
233 | timer->base = new_base; |
234 | } | |
235 | return new_base; | |
236 | } | |
237 | ||
238 | #else /* CONFIG_SMP */ | |
239 | ||
3c8aa39d | 240 | static inline struct hrtimer_clock_base * |
c0a31329 TG |
241 | lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) |
242 | { | |
3c8aa39d | 243 | struct hrtimer_clock_base *base = timer->base; |
c0a31329 | 244 | |
3c8aa39d | 245 | spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
246 | |
247 | return base; | |
248 | } | |
249 | ||
54cdfdb4 | 250 | # define switch_hrtimer_base(t, b) (b) |
c0a31329 TG |
251 | |
252 | #endif /* !CONFIG_SMP */ | |
253 | ||
254 | /* | |
255 | * Functions for the union type storage format of ktime_t which are | |
256 | * too large for inlining: | |
257 | */ | |
258 | #if BITS_PER_LONG < 64 | |
259 | # ifndef CONFIG_KTIME_SCALAR | |
260 | /** | |
261 | * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable | |
c0a31329 TG |
262 | * @kt: addend |
263 | * @nsec: the scalar nsec value to add | |
264 | * | |
265 | * Returns the sum of kt and nsec in ktime_t format | |
266 | */ | |
267 | ktime_t ktime_add_ns(const ktime_t kt, u64 nsec) | |
268 | { | |
269 | ktime_t tmp; | |
270 | ||
271 | if (likely(nsec < NSEC_PER_SEC)) { | |
272 | tmp.tv64 = nsec; | |
273 | } else { | |
274 | unsigned long rem = do_div(nsec, NSEC_PER_SEC); | |
275 | ||
276 | tmp = ktime_set((long)nsec, rem); | |
277 | } | |
278 | ||
279 | return ktime_add(kt, tmp); | |
280 | } | |
c0a31329 TG |
281 | # endif /* !CONFIG_KTIME_SCALAR */ |
282 | ||
283 | /* | |
284 | * Divide a ktime value by a nanosecond value | |
285 | */ | |
79bf2bb3 | 286 | unsigned long ktime_divns(const ktime_t kt, s64 div) |
c0a31329 TG |
287 | { |
288 | u64 dclc, inc, dns; | |
289 | int sft = 0; | |
290 | ||
291 | dclc = dns = ktime_to_ns(kt); | |
292 | inc = div; | |
293 | /* Make sure the divisor is less than 2^32: */ | |
294 | while (div >> 32) { | |
295 | sft++; | |
296 | div >>= 1; | |
297 | } | |
298 | dclc >>= sft; | |
299 | do_div(dclc, (unsigned long) div); | |
300 | ||
301 | return (unsigned long) dclc; | |
302 | } | |
c0a31329 TG |
303 | #endif /* BITS_PER_LONG >= 64 */ |
304 | ||
54cdfdb4 TG |
305 | /* High resolution timer related functions */ |
306 | #ifdef CONFIG_HIGH_RES_TIMERS | |
307 | ||
308 | /* | |
309 | * High resolution timer enabled ? | |
310 | */ | |
311 | static int hrtimer_hres_enabled __read_mostly = 1; | |
312 | ||
313 | /* | |
314 | * Enable / Disable high resolution mode | |
315 | */ | |
316 | static int __init setup_hrtimer_hres(char *str) | |
317 | { | |
318 | if (!strcmp(str, "off")) | |
319 | hrtimer_hres_enabled = 0; | |
320 | else if (!strcmp(str, "on")) | |
321 | hrtimer_hres_enabled = 1; | |
322 | else | |
323 | return 0; | |
324 | return 1; | |
325 | } | |
326 | ||
327 | __setup("highres=", setup_hrtimer_hres); | |
328 | ||
329 | /* | |
330 | * hrtimer_high_res_enabled - query, if the highres mode is enabled | |
331 | */ | |
332 | static inline int hrtimer_is_hres_enabled(void) | |
333 | { | |
334 | return hrtimer_hres_enabled; | |
335 | } | |
336 | ||
337 | /* | |
338 | * Is the high resolution mode active ? | |
339 | */ | |
340 | static inline int hrtimer_hres_active(void) | |
341 | { | |
342 | return __get_cpu_var(hrtimer_bases).hres_active; | |
343 | } | |
344 | ||
345 | /* | |
346 | * Reprogram the event source with checking both queues for the | |
347 | * next event | |
348 | * Called with interrupts disabled and base->lock held | |
349 | */ | |
350 | static void hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base) | |
351 | { | |
352 | int i; | |
353 | struct hrtimer_clock_base *base = cpu_base->clock_base; | |
354 | ktime_t expires; | |
355 | ||
356 | cpu_base->expires_next.tv64 = KTIME_MAX; | |
357 | ||
358 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) { | |
359 | struct hrtimer *timer; | |
360 | ||
361 | if (!base->first) | |
362 | continue; | |
363 | timer = rb_entry(base->first, struct hrtimer, node); | |
364 | expires = ktime_sub(timer->expires, base->offset); | |
365 | if (expires.tv64 < cpu_base->expires_next.tv64) | |
366 | cpu_base->expires_next = expires; | |
367 | } | |
368 | ||
369 | if (cpu_base->expires_next.tv64 != KTIME_MAX) | |
370 | tick_program_event(cpu_base->expires_next, 1); | |
371 | } | |
372 | ||
373 | /* | |
374 | * Shared reprogramming for clock_realtime and clock_monotonic | |
375 | * | |
376 | * When a timer is enqueued and expires earlier than the already enqueued | |
377 | * timers, we have to check, whether it expires earlier than the timer for | |
378 | * which the clock event device was armed. | |
379 | * | |
380 | * Called with interrupts disabled and base->cpu_base.lock held | |
381 | */ | |
382 | static int hrtimer_reprogram(struct hrtimer *timer, | |
383 | struct hrtimer_clock_base *base) | |
384 | { | |
385 | ktime_t *expires_next = &__get_cpu_var(hrtimer_bases).expires_next; | |
386 | ktime_t expires = ktime_sub(timer->expires, base->offset); | |
387 | int res; | |
388 | ||
389 | /* | |
390 | * When the callback is running, we do not reprogram the clock event | |
391 | * device. The timer callback is either running on a different CPU or | |
392 | * the callback is executed in the hrtimer_interupt context. The | |
393 | * reprogramming is handled either by the softirq, which called the | |
394 | * callback or at the end of the hrtimer_interrupt. | |
395 | */ | |
396 | if (hrtimer_callback_running(timer)) | |
397 | return 0; | |
398 | ||
399 | if (expires.tv64 >= expires_next->tv64) | |
400 | return 0; | |
401 | ||
402 | /* | |
403 | * Clockevents returns -ETIME, when the event was in the past. | |
404 | */ | |
405 | res = tick_program_event(expires, 0); | |
406 | if (!IS_ERR_VALUE(res)) | |
407 | *expires_next = expires; | |
408 | return res; | |
409 | } | |
410 | ||
411 | ||
412 | /* | |
413 | * Retrigger next event is called after clock was set | |
414 | * | |
415 | * Called with interrupts disabled via on_each_cpu() | |
416 | */ | |
417 | static void retrigger_next_event(void *arg) | |
418 | { | |
419 | struct hrtimer_cpu_base *base; | |
420 | struct timespec realtime_offset; | |
421 | unsigned long seq; | |
422 | ||
423 | if (!hrtimer_hres_active()) | |
424 | return; | |
425 | ||
426 | do { | |
427 | seq = read_seqbegin(&xtime_lock); | |
428 | set_normalized_timespec(&realtime_offset, | |
429 | -wall_to_monotonic.tv_sec, | |
430 | -wall_to_monotonic.tv_nsec); | |
431 | } while (read_seqretry(&xtime_lock, seq)); | |
432 | ||
433 | base = &__get_cpu_var(hrtimer_bases); | |
434 | ||
435 | /* Adjust CLOCK_REALTIME offset */ | |
436 | spin_lock(&base->lock); | |
437 | base->clock_base[CLOCK_REALTIME].offset = | |
438 | timespec_to_ktime(realtime_offset); | |
439 | ||
440 | hrtimer_force_reprogram(base); | |
441 | spin_unlock(&base->lock); | |
442 | } | |
443 | ||
444 | /* | |
445 | * Clock realtime was set | |
446 | * | |
447 | * Change the offset of the realtime clock vs. the monotonic | |
448 | * clock. | |
449 | * | |
450 | * We might have to reprogram the high resolution timer interrupt. On | |
451 | * SMP we call the architecture specific code to retrigger _all_ high | |
452 | * resolution timer interrupts. On UP we just disable interrupts and | |
453 | * call the high resolution interrupt code. | |
454 | */ | |
455 | void clock_was_set(void) | |
456 | { | |
457 | /* Retrigger the CPU local events everywhere */ | |
458 | on_each_cpu(retrigger_next_event, NULL, 0, 1); | |
459 | } | |
460 | ||
995f054f IM |
461 | /* |
462 | * During resume we might have to reprogram the high resolution timer | |
463 | * interrupt (on the local CPU): | |
464 | */ | |
465 | void hres_timers_resume(void) | |
466 | { | |
467 | WARN_ON_ONCE(num_online_cpus() > 1); | |
468 | ||
469 | /* Retrigger the CPU local events: */ | |
470 | retrigger_next_event(NULL); | |
471 | } | |
472 | ||
54cdfdb4 TG |
473 | /* |
474 | * Check, whether the timer is on the callback pending list | |
475 | */ | |
476 | static inline int hrtimer_cb_pending(const struct hrtimer *timer) | |
477 | { | |
478 | return timer->state & HRTIMER_STATE_PENDING; | |
479 | } | |
480 | ||
481 | /* | |
482 | * Remove a timer from the callback pending list | |
483 | */ | |
484 | static inline void hrtimer_remove_cb_pending(struct hrtimer *timer) | |
485 | { | |
486 | list_del_init(&timer->cb_entry); | |
487 | } | |
488 | ||
489 | /* | |
490 | * Initialize the high resolution related parts of cpu_base | |
491 | */ | |
492 | static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) | |
493 | { | |
494 | base->expires_next.tv64 = KTIME_MAX; | |
495 | base->hres_active = 0; | |
496 | INIT_LIST_HEAD(&base->cb_pending); | |
497 | } | |
498 | ||
499 | /* | |
500 | * Initialize the high resolution related parts of a hrtimer | |
501 | */ | |
502 | static inline void hrtimer_init_timer_hres(struct hrtimer *timer) | |
503 | { | |
504 | INIT_LIST_HEAD(&timer->cb_entry); | |
505 | } | |
506 | ||
507 | /* | |
508 | * When High resolution timers are active, try to reprogram. Note, that in case | |
509 | * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry | |
510 | * check happens. The timer gets enqueued into the rbtree. The reprogramming | |
511 | * and expiry check is done in the hrtimer_interrupt or in the softirq. | |
512 | */ | |
513 | static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer, | |
514 | struct hrtimer_clock_base *base) | |
515 | { | |
516 | if (base->cpu_base->hres_active && hrtimer_reprogram(timer, base)) { | |
517 | ||
518 | /* Timer is expired, act upon the callback mode */ | |
519 | switch(timer->cb_mode) { | |
520 | case HRTIMER_CB_IRQSAFE_NO_RESTART: | |
521 | /* | |
522 | * We can call the callback from here. No restart | |
523 | * happens, so no danger of recursion | |
524 | */ | |
525 | BUG_ON(timer->function(timer) != HRTIMER_NORESTART); | |
526 | return 1; | |
527 | case HRTIMER_CB_IRQSAFE_NO_SOFTIRQ: | |
528 | /* | |
529 | * This is solely for the sched tick emulation with | |
530 | * dynamic tick support to ensure that we do not | |
531 | * restart the tick right on the edge and end up with | |
532 | * the tick timer in the softirq ! The calling site | |
533 | * takes care of this. | |
534 | */ | |
535 | return 1; | |
536 | case HRTIMER_CB_IRQSAFE: | |
537 | case HRTIMER_CB_SOFTIRQ: | |
538 | /* | |
539 | * Move everything else into the softirq pending list ! | |
540 | */ | |
541 | list_add_tail(&timer->cb_entry, | |
542 | &base->cpu_base->cb_pending); | |
543 | timer->state = HRTIMER_STATE_PENDING; | |
544 | raise_softirq(HRTIMER_SOFTIRQ); | |
545 | return 1; | |
546 | default: | |
547 | BUG(); | |
548 | } | |
549 | } | |
550 | return 0; | |
551 | } | |
552 | ||
553 | /* | |
554 | * Switch to high resolution mode | |
555 | */ | |
f8953856 | 556 | static int hrtimer_switch_to_hres(void) |
54cdfdb4 TG |
557 | { |
558 | struct hrtimer_cpu_base *base = &__get_cpu_var(hrtimer_bases); | |
559 | unsigned long flags; | |
560 | ||
561 | if (base->hres_active) | |
f8953856 | 562 | return 1; |
54cdfdb4 TG |
563 | |
564 | local_irq_save(flags); | |
565 | ||
566 | if (tick_init_highres()) { | |
567 | local_irq_restore(flags); | |
f8953856 | 568 | return 0; |
54cdfdb4 TG |
569 | } |
570 | base->hres_active = 1; | |
571 | base->clock_base[CLOCK_REALTIME].resolution = KTIME_HIGH_RES; | |
572 | base->clock_base[CLOCK_MONOTONIC].resolution = KTIME_HIGH_RES; | |
573 | ||
574 | tick_setup_sched_timer(); | |
575 | ||
576 | /* "Retrigger" the interrupt to get things going */ | |
577 | retrigger_next_event(NULL); | |
578 | local_irq_restore(flags); | |
579 | printk(KERN_INFO "Switched to high resolution mode on CPU %d\n", | |
580 | smp_processor_id()); | |
f8953856 | 581 | return 1; |
54cdfdb4 TG |
582 | } |
583 | ||
584 | #else | |
585 | ||
586 | static inline int hrtimer_hres_active(void) { return 0; } | |
587 | static inline int hrtimer_is_hres_enabled(void) { return 0; } | |
f8953856 | 588 | static inline int hrtimer_switch_to_hres(void) { return 0; } |
54cdfdb4 TG |
589 | static inline void hrtimer_force_reprogram(struct hrtimer_cpu_base *base) { } |
590 | static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer, | |
591 | struct hrtimer_clock_base *base) | |
592 | { | |
593 | return 0; | |
594 | } | |
595 | static inline int hrtimer_cb_pending(struct hrtimer *timer) { return 0; } | |
596 | static inline void hrtimer_remove_cb_pending(struct hrtimer *timer) { } | |
597 | static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { } | |
598 | static inline void hrtimer_init_timer_hres(struct hrtimer *timer) { } | |
599 | ||
600 | #endif /* CONFIG_HIGH_RES_TIMERS */ | |
601 | ||
82f67cd9 IM |
602 | #ifdef CONFIG_TIMER_STATS |
603 | void __timer_stats_hrtimer_set_start_info(struct hrtimer *timer, void *addr) | |
604 | { | |
605 | if (timer->start_site) | |
606 | return; | |
607 | ||
608 | timer->start_site = addr; | |
609 | memcpy(timer->start_comm, current->comm, TASK_COMM_LEN); | |
610 | timer->start_pid = current->pid; | |
611 | } | |
612 | #endif | |
613 | ||
c0a31329 TG |
614 | /* |
615 | * Counterpart to lock_timer_base above: | |
616 | */ | |
617 | static inline | |
618 | void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) | |
619 | { | |
3c8aa39d | 620 | spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags); |
c0a31329 TG |
621 | } |
622 | ||
623 | /** | |
624 | * hrtimer_forward - forward the timer expiry | |
c0a31329 | 625 | * @timer: hrtimer to forward |
44f21475 | 626 | * @now: forward past this time |
c0a31329 TG |
627 | * @interval: the interval to forward |
628 | * | |
629 | * Forward the timer expiry so it will expire in the future. | |
8dca6f33 | 630 | * Returns the number of overruns. |
c0a31329 TG |
631 | */ |
632 | unsigned long | |
44f21475 | 633 | hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval) |
c0a31329 TG |
634 | { |
635 | unsigned long orun = 1; | |
44f21475 | 636 | ktime_t delta; |
c0a31329 TG |
637 | |
638 | delta = ktime_sub(now, timer->expires); | |
639 | ||
640 | if (delta.tv64 < 0) | |
641 | return 0; | |
642 | ||
c9db4fa1 TG |
643 | if (interval.tv64 < timer->base->resolution.tv64) |
644 | interval.tv64 = timer->base->resolution.tv64; | |
645 | ||
c0a31329 | 646 | if (unlikely(delta.tv64 >= interval.tv64)) { |
df869b63 | 647 | s64 incr = ktime_to_ns(interval); |
c0a31329 TG |
648 | |
649 | orun = ktime_divns(delta, incr); | |
650 | timer->expires = ktime_add_ns(timer->expires, incr * orun); | |
651 | if (timer->expires.tv64 > now.tv64) | |
652 | return orun; | |
653 | /* | |
654 | * This (and the ktime_add() below) is the | |
655 | * correction for exact: | |
656 | */ | |
657 | orun++; | |
658 | } | |
659 | timer->expires = ktime_add(timer->expires, interval); | |
13788ccc TG |
660 | /* |
661 | * Make sure, that the result did not wrap with a very large | |
662 | * interval. | |
663 | */ | |
664 | if (timer->expires.tv64 < 0) | |
665 | timer->expires = ktime_set(KTIME_SEC_MAX, 0); | |
c0a31329 TG |
666 | |
667 | return orun; | |
668 | } | |
669 | ||
670 | /* | |
671 | * enqueue_hrtimer - internal function to (re)start a timer | |
672 | * | |
673 | * The timer is inserted in expiry order. Insertion into the | |
674 | * red black tree is O(log(n)). Must hold the base lock. | |
675 | */ | |
3c8aa39d | 676 | static void enqueue_hrtimer(struct hrtimer *timer, |
54cdfdb4 | 677 | struct hrtimer_clock_base *base, int reprogram) |
c0a31329 TG |
678 | { |
679 | struct rb_node **link = &base->active.rb_node; | |
c0a31329 TG |
680 | struct rb_node *parent = NULL; |
681 | struct hrtimer *entry; | |
682 | ||
683 | /* | |
684 | * Find the right place in the rbtree: | |
685 | */ | |
686 | while (*link) { | |
687 | parent = *link; | |
688 | entry = rb_entry(parent, struct hrtimer, node); | |
689 | /* | |
690 | * We dont care about collisions. Nodes with | |
691 | * the same expiry time stay together. | |
692 | */ | |
693 | if (timer->expires.tv64 < entry->expires.tv64) | |
694 | link = &(*link)->rb_left; | |
288867ec | 695 | else |
c0a31329 | 696 | link = &(*link)->rb_right; |
c0a31329 TG |
697 | } |
698 | ||
699 | /* | |
288867ec TG |
700 | * Insert the timer to the rbtree and check whether it |
701 | * replaces the first pending timer | |
c0a31329 | 702 | */ |
54cdfdb4 TG |
703 | if (!base->first || timer->expires.tv64 < |
704 | rb_entry(base->first, struct hrtimer, node)->expires.tv64) { | |
705 | /* | |
706 | * Reprogram the clock event device. When the timer is already | |
707 | * expired hrtimer_enqueue_reprogram has either called the | |
708 | * callback or added it to the pending list and raised the | |
709 | * softirq. | |
710 | * | |
711 | * This is a NOP for !HIGHRES | |
712 | */ | |
713 | if (reprogram && hrtimer_enqueue_reprogram(timer, base)) | |
714 | return; | |
715 | ||
716 | base->first = &timer->node; | |
717 | } | |
718 | ||
c0a31329 TG |
719 | rb_link_node(&timer->node, parent, link); |
720 | rb_insert_color(&timer->node, &base->active); | |
303e967f TG |
721 | /* |
722 | * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the | |
723 | * state of a possibly running callback. | |
724 | */ | |
725 | timer->state |= HRTIMER_STATE_ENQUEUED; | |
288867ec | 726 | } |
c0a31329 TG |
727 | |
728 | /* | |
729 | * __remove_hrtimer - internal function to remove a timer | |
730 | * | |
731 | * Caller must hold the base lock. | |
54cdfdb4 TG |
732 | * |
733 | * High resolution timer mode reprograms the clock event device when the | |
734 | * timer is the one which expires next. The caller can disable this by setting | |
735 | * reprogram to zero. This is useful, when the context does a reprogramming | |
736 | * anyway (e.g. timer interrupt) | |
c0a31329 | 737 | */ |
3c8aa39d | 738 | static void __remove_hrtimer(struct hrtimer *timer, |
303e967f | 739 | struct hrtimer_clock_base *base, |
54cdfdb4 | 740 | unsigned long newstate, int reprogram) |
c0a31329 | 741 | { |
54cdfdb4 TG |
742 | /* High res. callback list. NOP for !HIGHRES */ |
743 | if (hrtimer_cb_pending(timer)) | |
744 | hrtimer_remove_cb_pending(timer); | |
745 | else { | |
746 | /* | |
747 | * Remove the timer from the rbtree and replace the | |
748 | * first entry pointer if necessary. | |
749 | */ | |
750 | if (base->first == &timer->node) { | |
751 | base->first = rb_next(&timer->node); | |
752 | /* Reprogram the clock event device. if enabled */ | |
753 | if (reprogram && hrtimer_hres_active()) | |
754 | hrtimer_force_reprogram(base->cpu_base); | |
755 | } | |
756 | rb_erase(&timer->node, &base->active); | |
757 | } | |
303e967f | 758 | timer->state = newstate; |
c0a31329 TG |
759 | } |
760 | ||
761 | /* | |
762 | * remove hrtimer, called with base lock held | |
763 | */ | |
764 | static inline int | |
3c8aa39d | 765 | remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base) |
c0a31329 | 766 | { |
303e967f | 767 | if (hrtimer_is_queued(timer)) { |
54cdfdb4 TG |
768 | int reprogram; |
769 | ||
770 | /* | |
771 | * Remove the timer and force reprogramming when high | |
772 | * resolution mode is active and the timer is on the current | |
773 | * CPU. If we remove a timer on another CPU, reprogramming is | |
774 | * skipped. The interrupt event on this CPU is fired and | |
775 | * reprogramming happens in the interrupt handler. This is a | |
776 | * rare case and less expensive than a smp call. | |
777 | */ | |
82f67cd9 | 778 | timer_stats_hrtimer_clear_start_info(timer); |
54cdfdb4 TG |
779 | reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases); |
780 | __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE, | |
781 | reprogram); | |
c0a31329 TG |
782 | return 1; |
783 | } | |
784 | return 0; | |
785 | } | |
786 | ||
787 | /** | |
788 | * hrtimer_start - (re)start an relative timer on the current CPU | |
c0a31329 TG |
789 | * @timer: the timer to be added |
790 | * @tim: expiry time | |
791 | * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL) | |
792 | * | |
793 | * Returns: | |
794 | * 0 on success | |
795 | * 1 when the timer was active | |
796 | */ | |
797 | int | |
798 | hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode) | |
799 | { | |
3c8aa39d | 800 | struct hrtimer_clock_base *base, *new_base; |
c0a31329 TG |
801 | unsigned long flags; |
802 | int ret; | |
803 | ||
804 | base = lock_hrtimer_base(timer, &flags); | |
805 | ||
806 | /* Remove an active timer from the queue: */ | |
807 | ret = remove_hrtimer(timer, base); | |
808 | ||
809 | /* Switch the timer base, if necessary: */ | |
810 | new_base = switch_hrtimer_base(timer, base); | |
811 | ||
c9cb2e3d | 812 | if (mode == HRTIMER_MODE_REL) { |
c0a31329 | 813 | tim = ktime_add(tim, new_base->get_time()); |
06027bdd IM |
814 | /* |
815 | * CONFIG_TIME_LOW_RES is a temporary way for architectures | |
816 | * to signal that they simply return xtime in | |
817 | * do_gettimeoffset(). In this case we want to round up by | |
818 | * resolution when starting a relative timer, to avoid short | |
819 | * timeouts. This will go away with the GTOD framework. | |
820 | */ | |
821 | #ifdef CONFIG_TIME_LOW_RES | |
822 | tim = ktime_add(tim, base->resolution); | |
823 | #endif | |
824 | } | |
c0a31329 TG |
825 | timer->expires = tim; |
826 | ||
82f67cd9 IM |
827 | timer_stats_hrtimer_set_start_info(timer); |
828 | ||
935c631d IM |
829 | /* |
830 | * Only allow reprogramming if the new base is on this CPU. | |
831 | * (it might still be on another CPU if the timer was pending) | |
832 | */ | |
833 | enqueue_hrtimer(timer, new_base, | |
834 | new_base->cpu_base == &__get_cpu_var(hrtimer_bases)); | |
c0a31329 TG |
835 | |
836 | unlock_hrtimer_base(timer, &flags); | |
837 | ||
838 | return ret; | |
839 | } | |
8d16b764 | 840 | EXPORT_SYMBOL_GPL(hrtimer_start); |
c0a31329 TG |
841 | |
842 | /** | |
843 | * hrtimer_try_to_cancel - try to deactivate a timer | |
c0a31329 TG |
844 | * @timer: hrtimer to stop |
845 | * | |
846 | * Returns: | |
847 | * 0 when the timer was not active | |
848 | * 1 when the timer was active | |
849 | * -1 when the timer is currently excuting the callback function and | |
fa9799e3 | 850 | * cannot be stopped |
c0a31329 TG |
851 | */ |
852 | int hrtimer_try_to_cancel(struct hrtimer *timer) | |
853 | { | |
3c8aa39d | 854 | struct hrtimer_clock_base *base; |
c0a31329 TG |
855 | unsigned long flags; |
856 | int ret = -1; | |
857 | ||
858 | base = lock_hrtimer_base(timer, &flags); | |
859 | ||
303e967f | 860 | if (!hrtimer_callback_running(timer)) |
c0a31329 TG |
861 | ret = remove_hrtimer(timer, base); |
862 | ||
863 | unlock_hrtimer_base(timer, &flags); | |
864 | ||
865 | return ret; | |
866 | ||
867 | } | |
8d16b764 | 868 | EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel); |
c0a31329 TG |
869 | |
870 | /** | |
871 | * hrtimer_cancel - cancel a timer and wait for the handler to finish. | |
c0a31329 TG |
872 | * @timer: the timer to be cancelled |
873 | * | |
874 | * Returns: | |
875 | * 0 when the timer was not active | |
876 | * 1 when the timer was active | |
877 | */ | |
878 | int hrtimer_cancel(struct hrtimer *timer) | |
879 | { | |
880 | for (;;) { | |
881 | int ret = hrtimer_try_to_cancel(timer); | |
882 | ||
883 | if (ret >= 0) | |
884 | return ret; | |
5ef37b19 | 885 | cpu_relax(); |
c0a31329 TG |
886 | } |
887 | } | |
8d16b764 | 888 | EXPORT_SYMBOL_GPL(hrtimer_cancel); |
c0a31329 TG |
889 | |
890 | /** | |
891 | * hrtimer_get_remaining - get remaining time for the timer | |
c0a31329 TG |
892 | * @timer: the timer to read |
893 | */ | |
894 | ktime_t hrtimer_get_remaining(const struct hrtimer *timer) | |
895 | { | |
3c8aa39d | 896 | struct hrtimer_clock_base *base; |
c0a31329 TG |
897 | unsigned long flags; |
898 | ktime_t rem; | |
899 | ||
900 | base = lock_hrtimer_base(timer, &flags); | |
3c8aa39d | 901 | rem = ktime_sub(timer->expires, base->get_time()); |
c0a31329 TG |
902 | unlock_hrtimer_base(timer, &flags); |
903 | ||
904 | return rem; | |
905 | } | |
8d16b764 | 906 | EXPORT_SYMBOL_GPL(hrtimer_get_remaining); |
c0a31329 | 907 | |
fd064b9b | 908 | #if defined(CONFIG_NO_IDLE_HZ) || defined(CONFIG_NO_HZ) |
69239749 TL |
909 | /** |
910 | * hrtimer_get_next_event - get the time until next expiry event | |
911 | * | |
912 | * Returns the delta to the next expiry event or KTIME_MAX if no timer | |
913 | * is pending. | |
914 | */ | |
915 | ktime_t hrtimer_get_next_event(void) | |
916 | { | |
3c8aa39d TG |
917 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
918 | struct hrtimer_clock_base *base = cpu_base->clock_base; | |
69239749 TL |
919 | ktime_t delta, mindelta = { .tv64 = KTIME_MAX }; |
920 | unsigned long flags; | |
921 | int i; | |
922 | ||
3c8aa39d TG |
923 | spin_lock_irqsave(&cpu_base->lock, flags); |
924 | ||
54cdfdb4 TG |
925 | if (!hrtimer_hres_active()) { |
926 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) { | |
927 | struct hrtimer *timer; | |
69239749 | 928 | |
54cdfdb4 TG |
929 | if (!base->first) |
930 | continue; | |
3c8aa39d | 931 | |
54cdfdb4 TG |
932 | timer = rb_entry(base->first, struct hrtimer, node); |
933 | delta.tv64 = timer->expires.tv64; | |
934 | delta = ktime_sub(delta, base->get_time()); | |
935 | if (delta.tv64 < mindelta.tv64) | |
936 | mindelta.tv64 = delta.tv64; | |
937 | } | |
69239749 | 938 | } |
3c8aa39d TG |
939 | |
940 | spin_unlock_irqrestore(&cpu_base->lock, flags); | |
941 | ||
69239749 TL |
942 | if (mindelta.tv64 < 0) |
943 | mindelta.tv64 = 0; | |
944 | return mindelta; | |
945 | } | |
946 | #endif | |
947 | ||
c0a31329 | 948 | /** |
7978672c | 949 | * hrtimer_init - initialize a timer to the given clock |
7978672c | 950 | * @timer: the timer to be initialized |
c0a31329 | 951 | * @clock_id: the clock to be used |
7978672c | 952 | * @mode: timer mode abs/rel |
c0a31329 | 953 | */ |
7978672c GA |
954 | void hrtimer_init(struct hrtimer *timer, clockid_t clock_id, |
955 | enum hrtimer_mode mode) | |
c0a31329 | 956 | { |
3c8aa39d | 957 | struct hrtimer_cpu_base *cpu_base; |
c0a31329 | 958 | |
7978672c GA |
959 | memset(timer, 0, sizeof(struct hrtimer)); |
960 | ||
3c8aa39d | 961 | cpu_base = &__raw_get_cpu_var(hrtimer_bases); |
c0a31329 | 962 | |
c9cb2e3d | 963 | if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS) |
7978672c GA |
964 | clock_id = CLOCK_MONOTONIC; |
965 | ||
3c8aa39d | 966 | timer->base = &cpu_base->clock_base[clock_id]; |
54cdfdb4 | 967 | hrtimer_init_timer_hres(timer); |
82f67cd9 IM |
968 | |
969 | #ifdef CONFIG_TIMER_STATS | |
970 | timer->start_site = NULL; | |
971 | timer->start_pid = -1; | |
972 | memset(timer->start_comm, 0, TASK_COMM_LEN); | |
973 | #endif | |
c0a31329 | 974 | } |
8d16b764 | 975 | EXPORT_SYMBOL_GPL(hrtimer_init); |
c0a31329 TG |
976 | |
977 | /** | |
978 | * hrtimer_get_res - get the timer resolution for a clock | |
c0a31329 TG |
979 | * @which_clock: which clock to query |
980 | * @tp: pointer to timespec variable to store the resolution | |
981 | * | |
72fd4a35 RD |
982 | * Store the resolution of the clock selected by @which_clock in the |
983 | * variable pointed to by @tp. | |
c0a31329 TG |
984 | */ |
985 | int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp) | |
986 | { | |
3c8aa39d | 987 | struct hrtimer_cpu_base *cpu_base; |
c0a31329 | 988 | |
3c8aa39d TG |
989 | cpu_base = &__raw_get_cpu_var(hrtimer_bases); |
990 | *tp = ktime_to_timespec(cpu_base->clock_base[which_clock].resolution); | |
c0a31329 TG |
991 | |
992 | return 0; | |
993 | } | |
8d16b764 | 994 | EXPORT_SYMBOL_GPL(hrtimer_get_res); |
c0a31329 | 995 | |
54cdfdb4 TG |
996 | #ifdef CONFIG_HIGH_RES_TIMERS |
997 | ||
998 | /* | |
999 | * High resolution timer interrupt | |
1000 | * Called with interrupts disabled | |
1001 | */ | |
1002 | void hrtimer_interrupt(struct clock_event_device *dev) | |
1003 | { | |
1004 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); | |
1005 | struct hrtimer_clock_base *base; | |
1006 | ktime_t expires_next, now; | |
1007 | int i, raise = 0; | |
1008 | ||
1009 | BUG_ON(!cpu_base->hres_active); | |
1010 | cpu_base->nr_events++; | |
1011 | dev->next_event.tv64 = KTIME_MAX; | |
1012 | ||
1013 | retry: | |
1014 | now = ktime_get(); | |
1015 | ||
1016 | expires_next.tv64 = KTIME_MAX; | |
1017 | ||
1018 | base = cpu_base->clock_base; | |
1019 | ||
1020 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { | |
1021 | ktime_t basenow; | |
1022 | struct rb_node *node; | |
1023 | ||
1024 | spin_lock(&cpu_base->lock); | |
1025 | ||
1026 | basenow = ktime_add(now, base->offset); | |
1027 | ||
1028 | while ((node = base->first)) { | |
1029 | struct hrtimer *timer; | |
1030 | ||
1031 | timer = rb_entry(node, struct hrtimer, node); | |
1032 | ||
1033 | if (basenow.tv64 < timer->expires.tv64) { | |
1034 | ktime_t expires; | |
1035 | ||
1036 | expires = ktime_sub(timer->expires, | |
1037 | base->offset); | |
1038 | if (expires.tv64 < expires_next.tv64) | |
1039 | expires_next = expires; | |
1040 | break; | |
1041 | } | |
1042 | ||
1043 | /* Move softirq callbacks to the pending list */ | |
1044 | if (timer->cb_mode == HRTIMER_CB_SOFTIRQ) { | |
1045 | __remove_hrtimer(timer, base, | |
1046 | HRTIMER_STATE_PENDING, 0); | |
1047 | list_add_tail(&timer->cb_entry, | |
1048 | &base->cpu_base->cb_pending); | |
1049 | raise = 1; | |
1050 | continue; | |
1051 | } | |
1052 | ||
1053 | __remove_hrtimer(timer, base, | |
1054 | HRTIMER_STATE_CALLBACK, 0); | |
82f67cd9 | 1055 | timer_stats_account_hrtimer(timer); |
54cdfdb4 TG |
1056 | |
1057 | /* | |
1058 | * Note: We clear the CALLBACK bit after | |
1059 | * enqueue_hrtimer to avoid reprogramming of | |
1060 | * the event hardware. This happens at the end | |
1061 | * of this function anyway. | |
1062 | */ | |
1063 | if (timer->function(timer) != HRTIMER_NORESTART) { | |
1064 | BUG_ON(timer->state != HRTIMER_STATE_CALLBACK); | |
1065 | enqueue_hrtimer(timer, base, 0); | |
1066 | } | |
1067 | timer->state &= ~HRTIMER_STATE_CALLBACK; | |
1068 | } | |
1069 | spin_unlock(&cpu_base->lock); | |
1070 | base++; | |
1071 | } | |
1072 | ||
1073 | cpu_base->expires_next = expires_next; | |
1074 | ||
1075 | /* Reprogramming necessary ? */ | |
1076 | if (expires_next.tv64 != KTIME_MAX) { | |
1077 | if (tick_program_event(expires_next, 0)) | |
1078 | goto retry; | |
1079 | } | |
1080 | ||
1081 | /* Raise softirq ? */ | |
1082 | if (raise) | |
1083 | raise_softirq(HRTIMER_SOFTIRQ); | |
1084 | } | |
1085 | ||
1086 | static void run_hrtimer_softirq(struct softirq_action *h) | |
1087 | { | |
1088 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); | |
1089 | ||
1090 | spin_lock_irq(&cpu_base->lock); | |
1091 | ||
1092 | while (!list_empty(&cpu_base->cb_pending)) { | |
1093 | enum hrtimer_restart (*fn)(struct hrtimer *); | |
1094 | struct hrtimer *timer; | |
1095 | int restart; | |
1096 | ||
1097 | timer = list_entry(cpu_base->cb_pending.next, | |
1098 | struct hrtimer, cb_entry); | |
1099 | ||
82f67cd9 IM |
1100 | timer_stats_account_hrtimer(timer); |
1101 | ||
54cdfdb4 TG |
1102 | fn = timer->function; |
1103 | __remove_hrtimer(timer, timer->base, HRTIMER_STATE_CALLBACK, 0); | |
1104 | spin_unlock_irq(&cpu_base->lock); | |
1105 | ||
1106 | restart = fn(timer); | |
1107 | ||
1108 | spin_lock_irq(&cpu_base->lock); | |
1109 | ||
1110 | timer->state &= ~HRTIMER_STATE_CALLBACK; | |
1111 | if (restart == HRTIMER_RESTART) { | |
1112 | BUG_ON(hrtimer_active(timer)); | |
1113 | /* | |
1114 | * Enqueue the timer, allow reprogramming of the event | |
1115 | * device | |
1116 | */ | |
1117 | enqueue_hrtimer(timer, timer->base, 1); | |
1118 | } else if (hrtimer_active(timer)) { | |
1119 | /* | |
1120 | * If the timer was rearmed on another CPU, reprogram | |
1121 | * the event device. | |
1122 | */ | |
1123 | if (timer->base->first == &timer->node) | |
1124 | hrtimer_reprogram(timer, timer->base); | |
1125 | } | |
1126 | } | |
1127 | spin_unlock_irq(&cpu_base->lock); | |
1128 | } | |
1129 | ||
1130 | #endif /* CONFIG_HIGH_RES_TIMERS */ | |
1131 | ||
c0a31329 TG |
1132 | /* |
1133 | * Expire the per base hrtimer-queue: | |
1134 | */ | |
3c8aa39d TG |
1135 | static inline void run_hrtimer_queue(struct hrtimer_cpu_base *cpu_base, |
1136 | int index) | |
c0a31329 | 1137 | { |
288867ec | 1138 | struct rb_node *node; |
3c8aa39d | 1139 | struct hrtimer_clock_base *base = &cpu_base->clock_base[index]; |
c0a31329 | 1140 | |
3055adda DS |
1141 | if (!base->first) |
1142 | return; | |
1143 | ||
92127c7a TG |
1144 | if (base->get_softirq_time) |
1145 | base->softirq_time = base->get_softirq_time(); | |
1146 | ||
3c8aa39d | 1147 | spin_lock_irq(&cpu_base->lock); |
c0a31329 | 1148 | |
288867ec | 1149 | while ((node = base->first)) { |
c0a31329 | 1150 | struct hrtimer *timer; |
c9cb2e3d | 1151 | enum hrtimer_restart (*fn)(struct hrtimer *); |
c0a31329 | 1152 | int restart; |
c0a31329 | 1153 | |
288867ec | 1154 | timer = rb_entry(node, struct hrtimer, node); |
92127c7a | 1155 | if (base->softirq_time.tv64 <= timer->expires.tv64) |
c0a31329 TG |
1156 | break; |
1157 | ||
f8953856 TG |
1158 | #ifdef CONFIG_HIGH_RES_TIMERS |
1159 | WARN_ON_ONCE(timer->cb_mode == HRTIMER_CB_IRQSAFE_NO_SOFTIRQ); | |
1160 | #endif | |
82f67cd9 IM |
1161 | timer_stats_account_hrtimer(timer); |
1162 | ||
c0a31329 | 1163 | fn = timer->function; |
54cdfdb4 | 1164 | __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0); |
3c8aa39d | 1165 | spin_unlock_irq(&cpu_base->lock); |
c0a31329 | 1166 | |
05cfb614 | 1167 | restart = fn(timer); |
c0a31329 | 1168 | |
3c8aa39d | 1169 | spin_lock_irq(&cpu_base->lock); |
c0a31329 | 1170 | |
303e967f | 1171 | timer->state &= ~HRTIMER_STATE_CALLBACK; |
b75f7a51 RZ |
1172 | if (restart != HRTIMER_NORESTART) { |
1173 | BUG_ON(hrtimer_active(timer)); | |
54cdfdb4 | 1174 | enqueue_hrtimer(timer, base, 0); |
b75f7a51 | 1175 | } |
c0a31329 | 1176 | } |
3c8aa39d | 1177 | spin_unlock_irq(&cpu_base->lock); |
c0a31329 TG |
1178 | } |
1179 | ||
1180 | /* | |
1181 | * Called from timer softirq every jiffy, expire hrtimers: | |
54cdfdb4 TG |
1182 | * |
1183 | * For HRT its the fall back code to run the softirq in the timer | |
1184 | * softirq context in case the hrtimer initialization failed or has | |
1185 | * not been done yet. | |
c0a31329 TG |
1186 | */ |
1187 | void hrtimer_run_queues(void) | |
1188 | { | |
3c8aa39d | 1189 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
c0a31329 TG |
1190 | int i; |
1191 | ||
54cdfdb4 TG |
1192 | if (hrtimer_hres_active()) |
1193 | return; | |
1194 | ||
79bf2bb3 TG |
1195 | /* |
1196 | * This _is_ ugly: We have to check in the softirq context, | |
1197 | * whether we can switch to highres and / or nohz mode. The | |
1198 | * clocksource switch happens in the timer interrupt with | |
1199 | * xtime_lock held. Notification from there only sets the | |
1200 | * check bit in the tick_oneshot code, otherwise we might | |
1201 | * deadlock vs. xtime_lock. | |
1202 | */ | |
54cdfdb4 | 1203 | if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) |
f8953856 TG |
1204 | if (hrtimer_switch_to_hres()) |
1205 | return; | |
79bf2bb3 | 1206 | |
3c8aa39d | 1207 | hrtimer_get_softirq_time(cpu_base); |
92127c7a | 1208 | |
3c8aa39d TG |
1209 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) |
1210 | run_hrtimer_queue(cpu_base, i); | |
c0a31329 TG |
1211 | } |
1212 | ||
10c94ec1 TG |
1213 | /* |
1214 | * Sleep related functions: | |
1215 | */ | |
c9cb2e3d | 1216 | static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer) |
00362e33 TG |
1217 | { |
1218 | struct hrtimer_sleeper *t = | |
1219 | container_of(timer, struct hrtimer_sleeper, timer); | |
1220 | struct task_struct *task = t->task; | |
1221 | ||
1222 | t->task = NULL; | |
1223 | if (task) | |
1224 | wake_up_process(task); | |
1225 | ||
1226 | return HRTIMER_NORESTART; | |
1227 | } | |
1228 | ||
36c8b586 | 1229 | void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task) |
00362e33 TG |
1230 | { |
1231 | sl->timer.function = hrtimer_wakeup; | |
1232 | sl->task = task; | |
54cdfdb4 TG |
1233 | #ifdef CONFIG_HIGH_RES_TIMERS |
1234 | sl->timer.cb_mode = HRTIMER_CB_IRQSAFE_NO_RESTART; | |
1235 | #endif | |
00362e33 TG |
1236 | } |
1237 | ||
669d7868 | 1238 | static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode) |
432569bb | 1239 | { |
669d7868 | 1240 | hrtimer_init_sleeper(t, current); |
10c94ec1 | 1241 | |
432569bb RZ |
1242 | do { |
1243 | set_current_state(TASK_INTERRUPTIBLE); | |
1244 | hrtimer_start(&t->timer, t->timer.expires, mode); | |
1245 | ||
54cdfdb4 TG |
1246 | if (likely(t->task)) |
1247 | schedule(); | |
432569bb | 1248 | |
669d7868 | 1249 | hrtimer_cancel(&t->timer); |
c9cb2e3d | 1250 | mode = HRTIMER_MODE_ABS; |
669d7868 TG |
1251 | |
1252 | } while (t->task && !signal_pending(current)); | |
432569bb | 1253 | |
669d7868 | 1254 | return t->task == NULL; |
10c94ec1 TG |
1255 | } |
1256 | ||
1711ef38 | 1257 | long __sched hrtimer_nanosleep_restart(struct restart_block *restart) |
10c94ec1 | 1258 | { |
669d7868 | 1259 | struct hrtimer_sleeper t; |
ea13dbc8 IM |
1260 | struct timespec __user *rmtp; |
1261 | struct timespec tu; | |
432569bb | 1262 | ktime_t time; |
10c94ec1 TG |
1263 | |
1264 | restart->fn = do_no_restart_syscall; | |
1265 | ||
c9cb2e3d | 1266 | hrtimer_init(&t.timer, restart->arg0, HRTIMER_MODE_ABS); |
1711ef38 | 1267 | t.timer.expires.tv64 = ((u64)restart->arg3 << 32) | (u64) restart->arg2; |
10c94ec1 | 1268 | |
c9cb2e3d | 1269 | if (do_nanosleep(&t, HRTIMER_MODE_ABS)) |
10c94ec1 TG |
1270 | return 0; |
1271 | ||
1711ef38 | 1272 | rmtp = (struct timespec __user *) restart->arg1; |
432569bb RZ |
1273 | if (rmtp) { |
1274 | time = ktime_sub(t.timer.expires, t.timer.base->get_time()); | |
1275 | if (time.tv64 <= 0) | |
1276 | return 0; | |
1277 | tu = ktime_to_timespec(time); | |
1278 | if (copy_to_user(rmtp, &tu, sizeof(tu))) | |
1279 | return -EFAULT; | |
1280 | } | |
10c94ec1 | 1281 | |
1711ef38 | 1282 | restart->fn = hrtimer_nanosleep_restart; |
10c94ec1 TG |
1283 | |
1284 | /* The other values in restart are already filled in */ | |
1285 | return -ERESTART_RESTARTBLOCK; | |
1286 | } | |
1287 | ||
10c94ec1 TG |
1288 | long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp, |
1289 | const enum hrtimer_mode mode, const clockid_t clockid) | |
1290 | { | |
1291 | struct restart_block *restart; | |
669d7868 | 1292 | struct hrtimer_sleeper t; |
10c94ec1 TG |
1293 | struct timespec tu; |
1294 | ktime_t rem; | |
1295 | ||
432569bb RZ |
1296 | hrtimer_init(&t.timer, clockid, mode); |
1297 | t.timer.expires = timespec_to_ktime(*rqtp); | |
1298 | if (do_nanosleep(&t, mode)) | |
10c94ec1 TG |
1299 | return 0; |
1300 | ||
7978672c | 1301 | /* Absolute timers do not update the rmtp value and restart: */ |
c9cb2e3d | 1302 | if (mode == HRTIMER_MODE_ABS) |
10c94ec1 TG |
1303 | return -ERESTARTNOHAND; |
1304 | ||
432569bb RZ |
1305 | if (rmtp) { |
1306 | rem = ktime_sub(t.timer.expires, t.timer.base->get_time()); | |
1307 | if (rem.tv64 <= 0) | |
1308 | return 0; | |
1309 | tu = ktime_to_timespec(rem); | |
1310 | if (copy_to_user(rmtp, &tu, sizeof(tu))) | |
1311 | return -EFAULT; | |
1312 | } | |
10c94ec1 TG |
1313 | |
1314 | restart = ¤t_thread_info()->restart_block; | |
1711ef38 TA |
1315 | restart->fn = hrtimer_nanosleep_restart; |
1316 | restart->arg0 = (unsigned long) t.timer.base->index; | |
1317 | restart->arg1 = (unsigned long) rmtp; | |
1318 | restart->arg2 = t.timer.expires.tv64 & 0xFFFFFFFF; | |
1319 | restart->arg3 = t.timer.expires.tv64 >> 32; | |
10c94ec1 TG |
1320 | |
1321 | return -ERESTART_RESTARTBLOCK; | |
1322 | } | |
1323 | ||
6ba1b912 TG |
1324 | asmlinkage long |
1325 | sys_nanosleep(struct timespec __user *rqtp, struct timespec __user *rmtp) | |
1326 | { | |
1327 | struct timespec tu; | |
1328 | ||
1329 | if (copy_from_user(&tu, rqtp, sizeof(tu))) | |
1330 | return -EFAULT; | |
1331 | ||
1332 | if (!timespec_valid(&tu)) | |
1333 | return -EINVAL; | |
1334 | ||
c9cb2e3d | 1335 | return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC); |
6ba1b912 TG |
1336 | } |
1337 | ||
c0a31329 TG |
1338 | /* |
1339 | * Functions related to boot-time initialization: | |
1340 | */ | |
1341 | static void __devinit init_hrtimers_cpu(int cpu) | |
1342 | { | |
3c8aa39d | 1343 | struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu); |
c0a31329 TG |
1344 | int i; |
1345 | ||
3c8aa39d TG |
1346 | spin_lock_init(&cpu_base->lock); |
1347 | lockdep_set_class(&cpu_base->lock, &cpu_base->lock_key); | |
1348 | ||
1349 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) | |
1350 | cpu_base->clock_base[i].cpu_base = cpu_base; | |
1351 | ||
54cdfdb4 | 1352 | hrtimer_init_hres(cpu_base); |
c0a31329 TG |
1353 | } |
1354 | ||
1355 | #ifdef CONFIG_HOTPLUG_CPU | |
1356 | ||
3c8aa39d TG |
1357 | static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base, |
1358 | struct hrtimer_clock_base *new_base) | |
c0a31329 TG |
1359 | { |
1360 | struct hrtimer *timer; | |
1361 | struct rb_node *node; | |
1362 | ||
1363 | while ((node = rb_first(&old_base->active))) { | |
1364 | timer = rb_entry(node, struct hrtimer, node); | |
54cdfdb4 TG |
1365 | BUG_ON(hrtimer_callback_running(timer)); |
1366 | __remove_hrtimer(timer, old_base, HRTIMER_STATE_INACTIVE, 0); | |
c0a31329 | 1367 | timer->base = new_base; |
54cdfdb4 TG |
1368 | /* |
1369 | * Enqueue the timer. Allow reprogramming of the event device | |
1370 | */ | |
1371 | enqueue_hrtimer(timer, new_base, 1); | |
c0a31329 TG |
1372 | } |
1373 | } | |
1374 | ||
1375 | static void migrate_hrtimers(int cpu) | |
1376 | { | |
3c8aa39d | 1377 | struct hrtimer_cpu_base *old_base, *new_base; |
c0a31329 TG |
1378 | int i; |
1379 | ||
1380 | BUG_ON(cpu_online(cpu)); | |
3c8aa39d TG |
1381 | old_base = &per_cpu(hrtimer_bases, cpu); |
1382 | new_base = &get_cpu_var(hrtimer_bases); | |
c0a31329 | 1383 | |
54cdfdb4 TG |
1384 | tick_cancel_sched_timer(cpu); |
1385 | ||
c0a31329 | 1386 | local_irq_disable(); |
e81ce1f7 HC |
1387 | double_spin_lock(&new_base->lock, &old_base->lock, |
1388 | smp_processor_id() < cpu); | |
c0a31329 | 1389 | |
3c8aa39d | 1390 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
3c8aa39d TG |
1391 | migrate_hrtimer_list(&old_base->clock_base[i], |
1392 | &new_base->clock_base[i]); | |
c0a31329 TG |
1393 | } |
1394 | ||
e81ce1f7 HC |
1395 | double_spin_unlock(&new_base->lock, &old_base->lock, |
1396 | smp_processor_id() < cpu); | |
c0a31329 TG |
1397 | local_irq_enable(); |
1398 | put_cpu_var(hrtimer_bases); | |
1399 | } | |
1400 | #endif /* CONFIG_HOTPLUG_CPU */ | |
1401 | ||
8c78f307 | 1402 | static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self, |
c0a31329 TG |
1403 | unsigned long action, void *hcpu) |
1404 | { | |
1405 | long cpu = (long)hcpu; | |
1406 | ||
1407 | switch (action) { | |
1408 | ||
1409 | case CPU_UP_PREPARE: | |
1410 | init_hrtimers_cpu(cpu); | |
1411 | break; | |
1412 | ||
1413 | #ifdef CONFIG_HOTPLUG_CPU | |
1414 | case CPU_DEAD: | |
d316c57f | 1415 | clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &cpu); |
c0a31329 TG |
1416 | migrate_hrtimers(cpu); |
1417 | break; | |
1418 | #endif | |
1419 | ||
1420 | default: | |
1421 | break; | |
1422 | } | |
1423 | ||
1424 | return NOTIFY_OK; | |
1425 | } | |
1426 | ||
8c78f307 | 1427 | static struct notifier_block __cpuinitdata hrtimers_nb = { |
c0a31329 TG |
1428 | .notifier_call = hrtimer_cpu_notify, |
1429 | }; | |
1430 | ||
1431 | void __init hrtimers_init(void) | |
1432 | { | |
1433 | hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE, | |
1434 | (void *)(long)smp_processor_id()); | |
1435 | register_cpu_notifier(&hrtimers_nb); | |
54cdfdb4 TG |
1436 | #ifdef CONFIG_HIGH_RES_TIMERS |
1437 | open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq, NULL); | |
1438 | #endif | |
c0a31329 TG |
1439 | } |
1440 |