]>
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); | |
2c6b47de | 144 | xts = current_kernel_time(); |
ad28d94a | 145 | tom = wall_to_monotonic; |
92127c7a TG |
146 | } while (read_seqretry(&xtime_lock, seq)); |
147 | ||
f4304ab2 | 148 | xtim = timespec_to_ktime(xts); |
ad28d94a | 149 | tomono = timespec_to_ktime(tom); |
3c8aa39d TG |
150 | base->clock_base[CLOCK_REALTIME].softirq_time = xtim; |
151 | base->clock_base[CLOCK_MONOTONIC].softirq_time = | |
152 | ktime_add(xtim, tomono); | |
92127c7a TG |
153 | } |
154 | ||
303e967f TG |
155 | /* |
156 | * Helper function to check, whether the timer is running the callback | |
157 | * function | |
158 | */ | |
159 | static inline int hrtimer_callback_running(struct hrtimer *timer) | |
160 | { | |
161 | return timer->state & HRTIMER_STATE_CALLBACK; | |
162 | } | |
163 | ||
c0a31329 TG |
164 | /* |
165 | * Functions and macros which are different for UP/SMP systems are kept in a | |
166 | * single place | |
167 | */ | |
168 | #ifdef CONFIG_SMP | |
169 | ||
c0a31329 TG |
170 | /* |
171 | * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock | |
172 | * means that all timers which are tied to this base via timer->base are | |
173 | * locked, and the base itself is locked too. | |
174 | * | |
175 | * So __run_timers/migrate_timers can safely modify all timers which could | |
176 | * be found on the lists/queues. | |
177 | * | |
178 | * When the timer's base is locked, and the timer removed from list, it is | |
179 | * possible to set timer->base = NULL and drop the lock: the timer remains | |
180 | * locked. | |
181 | */ | |
3c8aa39d TG |
182 | static |
183 | struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer, | |
184 | unsigned long *flags) | |
c0a31329 | 185 | { |
3c8aa39d | 186 | struct hrtimer_clock_base *base; |
c0a31329 TG |
187 | |
188 | for (;;) { | |
189 | base = timer->base; | |
190 | if (likely(base != NULL)) { | |
3c8aa39d | 191 | spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
192 | if (likely(base == timer->base)) |
193 | return base; | |
194 | /* The timer has migrated to another CPU: */ | |
3c8aa39d | 195 | spin_unlock_irqrestore(&base->cpu_base->lock, *flags); |
c0a31329 TG |
196 | } |
197 | cpu_relax(); | |
198 | } | |
199 | } | |
200 | ||
201 | /* | |
202 | * Switch the timer base to the current CPU when possible. | |
203 | */ | |
3c8aa39d TG |
204 | static inline struct hrtimer_clock_base * |
205 | switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base) | |
c0a31329 | 206 | { |
3c8aa39d TG |
207 | struct hrtimer_clock_base *new_base; |
208 | struct hrtimer_cpu_base *new_cpu_base; | |
c0a31329 | 209 | |
3c8aa39d TG |
210 | new_cpu_base = &__get_cpu_var(hrtimer_bases); |
211 | new_base = &new_cpu_base->clock_base[base->index]; | |
c0a31329 TG |
212 | |
213 | if (base != new_base) { | |
214 | /* | |
215 | * We are trying to schedule the timer on the local CPU. | |
216 | * However we can't change timer's base while it is running, | |
217 | * so we keep it on the same CPU. No hassle vs. reprogramming | |
218 | * the event source in the high resolution case. The softirq | |
219 | * code will take care of this when the timer function has | |
220 | * completed. There is no conflict as we hold the lock until | |
221 | * the timer is enqueued. | |
222 | */ | |
54cdfdb4 | 223 | if (unlikely(hrtimer_callback_running(timer))) |
c0a31329 TG |
224 | return base; |
225 | ||
226 | /* See the comment in lock_timer_base() */ | |
227 | timer->base = NULL; | |
3c8aa39d TG |
228 | spin_unlock(&base->cpu_base->lock); |
229 | spin_lock(&new_base->cpu_base->lock); | |
c0a31329 TG |
230 | timer->base = new_base; |
231 | } | |
232 | return new_base; | |
233 | } | |
234 | ||
235 | #else /* CONFIG_SMP */ | |
236 | ||
3c8aa39d | 237 | static inline struct hrtimer_clock_base * |
c0a31329 TG |
238 | lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) |
239 | { | |
3c8aa39d | 240 | struct hrtimer_clock_base *base = timer->base; |
c0a31329 | 241 | |
3c8aa39d | 242 | spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
243 | |
244 | return base; | |
245 | } | |
246 | ||
54cdfdb4 | 247 | # define switch_hrtimer_base(t, b) (b) |
c0a31329 TG |
248 | |
249 | #endif /* !CONFIG_SMP */ | |
250 | ||
251 | /* | |
252 | * Functions for the union type storage format of ktime_t which are | |
253 | * too large for inlining: | |
254 | */ | |
255 | #if BITS_PER_LONG < 64 | |
256 | # ifndef CONFIG_KTIME_SCALAR | |
257 | /** | |
258 | * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable | |
c0a31329 TG |
259 | * @kt: addend |
260 | * @nsec: the scalar nsec value to add | |
261 | * | |
262 | * Returns the sum of kt and nsec in ktime_t format | |
263 | */ | |
264 | ktime_t ktime_add_ns(const ktime_t kt, u64 nsec) | |
265 | { | |
266 | ktime_t tmp; | |
267 | ||
268 | if (likely(nsec < NSEC_PER_SEC)) { | |
269 | tmp.tv64 = nsec; | |
270 | } else { | |
271 | unsigned long rem = do_div(nsec, NSEC_PER_SEC); | |
272 | ||
273 | tmp = ktime_set((long)nsec, rem); | |
274 | } | |
275 | ||
276 | return ktime_add(kt, tmp); | |
277 | } | |
b8b8fd2d DH |
278 | |
279 | EXPORT_SYMBOL_GPL(ktime_add_ns); | |
a272378d ACM |
280 | |
281 | /** | |
282 | * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable | |
283 | * @kt: minuend | |
284 | * @nsec: the scalar nsec value to subtract | |
285 | * | |
286 | * Returns the subtraction of @nsec from @kt in ktime_t format | |
287 | */ | |
288 | ktime_t ktime_sub_ns(const ktime_t kt, u64 nsec) | |
289 | { | |
290 | ktime_t tmp; | |
291 | ||
292 | if (likely(nsec < NSEC_PER_SEC)) { | |
293 | tmp.tv64 = nsec; | |
294 | } else { | |
295 | unsigned long rem = do_div(nsec, NSEC_PER_SEC); | |
296 | ||
297 | tmp = ktime_set((long)nsec, rem); | |
298 | } | |
299 | ||
300 | return ktime_sub(kt, tmp); | |
301 | } | |
302 | ||
303 | EXPORT_SYMBOL_GPL(ktime_sub_ns); | |
c0a31329 TG |
304 | # endif /* !CONFIG_KTIME_SCALAR */ |
305 | ||
306 | /* | |
307 | * Divide a ktime value by a nanosecond value | |
308 | */ | |
4d672e7a | 309 | u64 ktime_divns(const ktime_t kt, s64 div) |
c0a31329 TG |
310 | { |
311 | u64 dclc, inc, dns; | |
312 | int sft = 0; | |
313 | ||
314 | dclc = dns = ktime_to_ns(kt); | |
315 | inc = div; | |
316 | /* Make sure the divisor is less than 2^32: */ | |
317 | while (div >> 32) { | |
318 | sft++; | |
319 | div >>= 1; | |
320 | } | |
321 | dclc >>= sft; | |
322 | do_div(dclc, (unsigned long) div); | |
323 | ||
4d672e7a | 324 | return dclc; |
c0a31329 | 325 | } |
c0a31329 TG |
326 | #endif /* BITS_PER_LONG >= 64 */ |
327 | ||
d3d74453 PZ |
328 | /* |
329 | * Check, whether the timer is on the callback pending list | |
330 | */ | |
331 | static inline int hrtimer_cb_pending(const struct hrtimer *timer) | |
332 | { | |
333 | return timer->state & HRTIMER_STATE_PENDING; | |
334 | } | |
335 | ||
336 | /* | |
337 | * Remove a timer from the callback pending list | |
338 | */ | |
339 | static inline void hrtimer_remove_cb_pending(struct hrtimer *timer) | |
340 | { | |
341 | list_del_init(&timer->cb_entry); | |
342 | } | |
343 | ||
54cdfdb4 TG |
344 | /* High resolution timer related functions */ |
345 | #ifdef CONFIG_HIGH_RES_TIMERS | |
346 | ||
347 | /* | |
348 | * High resolution timer enabled ? | |
349 | */ | |
350 | static int hrtimer_hres_enabled __read_mostly = 1; | |
351 | ||
352 | /* | |
353 | * Enable / Disable high resolution mode | |
354 | */ | |
355 | static int __init setup_hrtimer_hres(char *str) | |
356 | { | |
357 | if (!strcmp(str, "off")) | |
358 | hrtimer_hres_enabled = 0; | |
359 | else if (!strcmp(str, "on")) | |
360 | hrtimer_hres_enabled = 1; | |
361 | else | |
362 | return 0; | |
363 | return 1; | |
364 | } | |
365 | ||
366 | __setup("highres=", setup_hrtimer_hres); | |
367 | ||
368 | /* | |
369 | * hrtimer_high_res_enabled - query, if the highres mode is enabled | |
370 | */ | |
371 | static inline int hrtimer_is_hres_enabled(void) | |
372 | { | |
373 | return hrtimer_hres_enabled; | |
374 | } | |
375 | ||
376 | /* | |
377 | * Is the high resolution mode active ? | |
378 | */ | |
379 | static inline int hrtimer_hres_active(void) | |
380 | { | |
381 | return __get_cpu_var(hrtimer_bases).hres_active; | |
382 | } | |
383 | ||
384 | /* | |
385 | * Reprogram the event source with checking both queues for the | |
386 | * next event | |
387 | * Called with interrupts disabled and base->lock held | |
388 | */ | |
389 | static void hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base) | |
390 | { | |
391 | int i; | |
392 | struct hrtimer_clock_base *base = cpu_base->clock_base; | |
393 | ktime_t expires; | |
394 | ||
395 | cpu_base->expires_next.tv64 = KTIME_MAX; | |
396 | ||
397 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) { | |
398 | struct hrtimer *timer; | |
399 | ||
400 | if (!base->first) | |
401 | continue; | |
402 | timer = rb_entry(base->first, struct hrtimer, node); | |
403 | expires = ktime_sub(timer->expires, base->offset); | |
404 | if (expires.tv64 < cpu_base->expires_next.tv64) | |
405 | cpu_base->expires_next = expires; | |
406 | } | |
407 | ||
408 | if (cpu_base->expires_next.tv64 != KTIME_MAX) | |
409 | tick_program_event(cpu_base->expires_next, 1); | |
410 | } | |
411 | ||
412 | /* | |
413 | * Shared reprogramming for clock_realtime and clock_monotonic | |
414 | * | |
415 | * When a timer is enqueued and expires earlier than the already enqueued | |
416 | * timers, we have to check, whether it expires earlier than the timer for | |
417 | * which the clock event device was armed. | |
418 | * | |
419 | * Called with interrupts disabled and base->cpu_base.lock held | |
420 | */ | |
421 | static int hrtimer_reprogram(struct hrtimer *timer, | |
422 | struct hrtimer_clock_base *base) | |
423 | { | |
424 | ktime_t *expires_next = &__get_cpu_var(hrtimer_bases).expires_next; | |
425 | ktime_t expires = ktime_sub(timer->expires, base->offset); | |
426 | int res; | |
427 | ||
428 | /* | |
429 | * When the callback is running, we do not reprogram the clock event | |
430 | * device. The timer callback is either running on a different CPU or | |
3a4fa0a2 | 431 | * the callback is executed in the hrtimer_interrupt context. The |
54cdfdb4 TG |
432 | * reprogramming is handled either by the softirq, which called the |
433 | * callback or at the end of the hrtimer_interrupt. | |
434 | */ | |
435 | if (hrtimer_callback_running(timer)) | |
436 | return 0; | |
437 | ||
438 | if (expires.tv64 >= expires_next->tv64) | |
439 | return 0; | |
440 | ||
441 | /* | |
442 | * Clockevents returns -ETIME, when the event was in the past. | |
443 | */ | |
444 | res = tick_program_event(expires, 0); | |
445 | if (!IS_ERR_VALUE(res)) | |
446 | *expires_next = expires; | |
447 | return res; | |
448 | } | |
449 | ||
450 | ||
451 | /* | |
452 | * Retrigger next event is called after clock was set | |
453 | * | |
454 | * Called with interrupts disabled via on_each_cpu() | |
455 | */ | |
456 | static void retrigger_next_event(void *arg) | |
457 | { | |
458 | struct hrtimer_cpu_base *base; | |
459 | struct timespec realtime_offset; | |
460 | unsigned long seq; | |
461 | ||
462 | if (!hrtimer_hres_active()) | |
463 | return; | |
464 | ||
465 | do { | |
466 | seq = read_seqbegin(&xtime_lock); | |
467 | set_normalized_timespec(&realtime_offset, | |
468 | -wall_to_monotonic.tv_sec, | |
469 | -wall_to_monotonic.tv_nsec); | |
470 | } while (read_seqretry(&xtime_lock, seq)); | |
471 | ||
472 | base = &__get_cpu_var(hrtimer_bases); | |
473 | ||
474 | /* Adjust CLOCK_REALTIME offset */ | |
475 | spin_lock(&base->lock); | |
476 | base->clock_base[CLOCK_REALTIME].offset = | |
477 | timespec_to_ktime(realtime_offset); | |
478 | ||
479 | hrtimer_force_reprogram(base); | |
480 | spin_unlock(&base->lock); | |
481 | } | |
482 | ||
483 | /* | |
484 | * Clock realtime was set | |
485 | * | |
486 | * Change the offset of the realtime clock vs. the monotonic | |
487 | * clock. | |
488 | * | |
489 | * We might have to reprogram the high resolution timer interrupt. On | |
490 | * SMP we call the architecture specific code to retrigger _all_ high | |
491 | * resolution timer interrupts. On UP we just disable interrupts and | |
492 | * call the high resolution interrupt code. | |
493 | */ | |
494 | void clock_was_set(void) | |
495 | { | |
496 | /* Retrigger the CPU local events everywhere */ | |
497 | on_each_cpu(retrigger_next_event, NULL, 0, 1); | |
498 | } | |
499 | ||
995f054f IM |
500 | /* |
501 | * During resume we might have to reprogram the high resolution timer | |
502 | * interrupt (on the local CPU): | |
503 | */ | |
504 | void hres_timers_resume(void) | |
505 | { | |
506 | WARN_ON_ONCE(num_online_cpus() > 1); | |
507 | ||
508 | /* Retrigger the CPU local events: */ | |
509 | retrigger_next_event(NULL); | |
510 | } | |
511 | ||
54cdfdb4 TG |
512 | /* |
513 | * Initialize the high resolution related parts of cpu_base | |
514 | */ | |
515 | static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) | |
516 | { | |
517 | base->expires_next.tv64 = KTIME_MAX; | |
518 | base->hres_active = 0; | |
54cdfdb4 TG |
519 | } |
520 | ||
521 | /* | |
522 | * Initialize the high resolution related parts of a hrtimer | |
523 | */ | |
524 | static inline void hrtimer_init_timer_hres(struct hrtimer *timer) | |
525 | { | |
54cdfdb4 TG |
526 | } |
527 | ||
528 | /* | |
529 | * When High resolution timers are active, try to reprogram. Note, that in case | |
530 | * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry | |
531 | * check happens. The timer gets enqueued into the rbtree. The reprogramming | |
532 | * and expiry check is done in the hrtimer_interrupt or in the softirq. | |
533 | */ | |
534 | static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer, | |
535 | struct hrtimer_clock_base *base) | |
536 | { | |
537 | if (base->cpu_base->hres_active && hrtimer_reprogram(timer, base)) { | |
538 | ||
539 | /* Timer is expired, act upon the callback mode */ | |
540 | switch(timer->cb_mode) { | |
541 | case HRTIMER_CB_IRQSAFE_NO_RESTART: | |
542 | /* | |
543 | * We can call the callback from here. No restart | |
544 | * happens, so no danger of recursion | |
545 | */ | |
546 | BUG_ON(timer->function(timer) != HRTIMER_NORESTART); | |
547 | return 1; | |
548 | case HRTIMER_CB_IRQSAFE_NO_SOFTIRQ: | |
549 | /* | |
550 | * This is solely for the sched tick emulation with | |
551 | * dynamic tick support to ensure that we do not | |
552 | * restart the tick right on the edge and end up with | |
553 | * the tick timer in the softirq ! The calling site | |
554 | * takes care of this. | |
555 | */ | |
556 | return 1; | |
557 | case HRTIMER_CB_IRQSAFE: | |
558 | case HRTIMER_CB_SOFTIRQ: | |
559 | /* | |
560 | * Move everything else into the softirq pending list ! | |
561 | */ | |
562 | list_add_tail(&timer->cb_entry, | |
563 | &base->cpu_base->cb_pending); | |
564 | timer->state = HRTIMER_STATE_PENDING; | |
565 | raise_softirq(HRTIMER_SOFTIRQ); | |
566 | return 1; | |
567 | default: | |
568 | BUG(); | |
569 | } | |
570 | } | |
571 | return 0; | |
572 | } | |
573 | ||
574 | /* | |
575 | * Switch to high resolution mode | |
576 | */ | |
f8953856 | 577 | static int hrtimer_switch_to_hres(void) |
54cdfdb4 | 578 | { |
820de5c3 IM |
579 | int cpu = smp_processor_id(); |
580 | struct hrtimer_cpu_base *base = &per_cpu(hrtimer_bases, cpu); | |
54cdfdb4 TG |
581 | unsigned long flags; |
582 | ||
583 | if (base->hres_active) | |
f8953856 | 584 | return 1; |
54cdfdb4 TG |
585 | |
586 | local_irq_save(flags); | |
587 | ||
588 | if (tick_init_highres()) { | |
589 | local_irq_restore(flags); | |
820de5c3 IM |
590 | printk(KERN_WARNING "Could not switch to high resolution " |
591 | "mode on CPU %d\n", cpu); | |
f8953856 | 592 | return 0; |
54cdfdb4 TG |
593 | } |
594 | base->hres_active = 1; | |
595 | base->clock_base[CLOCK_REALTIME].resolution = KTIME_HIGH_RES; | |
596 | base->clock_base[CLOCK_MONOTONIC].resolution = KTIME_HIGH_RES; | |
597 | ||
598 | tick_setup_sched_timer(); | |
599 | ||
600 | /* "Retrigger" the interrupt to get things going */ | |
601 | retrigger_next_event(NULL); | |
602 | local_irq_restore(flags); | |
edfed66e | 603 | printk(KERN_DEBUG "Switched to high resolution mode on CPU %d\n", |
54cdfdb4 | 604 | smp_processor_id()); |
f8953856 | 605 | return 1; |
54cdfdb4 TG |
606 | } |
607 | ||
608 | #else | |
609 | ||
610 | static inline int hrtimer_hres_active(void) { return 0; } | |
611 | static inline int hrtimer_is_hres_enabled(void) { return 0; } | |
f8953856 | 612 | static inline int hrtimer_switch_to_hres(void) { return 0; } |
54cdfdb4 TG |
613 | static inline void hrtimer_force_reprogram(struct hrtimer_cpu_base *base) { } |
614 | static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer, | |
615 | struct hrtimer_clock_base *base) | |
616 | { | |
617 | return 0; | |
618 | } | |
54cdfdb4 TG |
619 | static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { } |
620 | static inline void hrtimer_init_timer_hres(struct hrtimer *timer) { } | |
d3d74453 PZ |
621 | static inline int hrtimer_reprogram(struct hrtimer *timer, |
622 | struct hrtimer_clock_base *base) | |
623 | { | |
624 | return 0; | |
625 | } | |
54cdfdb4 TG |
626 | |
627 | #endif /* CONFIG_HIGH_RES_TIMERS */ | |
628 | ||
82f67cd9 IM |
629 | #ifdef CONFIG_TIMER_STATS |
630 | void __timer_stats_hrtimer_set_start_info(struct hrtimer *timer, void *addr) | |
631 | { | |
632 | if (timer->start_site) | |
633 | return; | |
634 | ||
635 | timer->start_site = addr; | |
636 | memcpy(timer->start_comm, current->comm, TASK_COMM_LEN); | |
637 | timer->start_pid = current->pid; | |
638 | } | |
639 | #endif | |
640 | ||
c0a31329 | 641 | /* |
6506f2aa | 642 | * Counterpart to lock_hrtimer_base above: |
c0a31329 TG |
643 | */ |
644 | static inline | |
645 | void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) | |
646 | { | |
3c8aa39d | 647 | spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags); |
c0a31329 TG |
648 | } |
649 | ||
650 | /** | |
651 | * hrtimer_forward - forward the timer expiry | |
c0a31329 | 652 | * @timer: hrtimer to forward |
44f21475 | 653 | * @now: forward past this time |
c0a31329 TG |
654 | * @interval: the interval to forward |
655 | * | |
656 | * Forward the timer expiry so it will expire in the future. | |
8dca6f33 | 657 | * Returns the number of overruns. |
c0a31329 | 658 | */ |
4d672e7a | 659 | u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval) |
c0a31329 | 660 | { |
4d672e7a | 661 | u64 orun = 1; |
44f21475 | 662 | ktime_t delta; |
c0a31329 TG |
663 | |
664 | delta = ktime_sub(now, timer->expires); | |
665 | ||
666 | if (delta.tv64 < 0) | |
667 | return 0; | |
668 | ||
c9db4fa1 TG |
669 | if (interval.tv64 < timer->base->resolution.tv64) |
670 | interval.tv64 = timer->base->resolution.tv64; | |
671 | ||
c0a31329 | 672 | if (unlikely(delta.tv64 >= interval.tv64)) { |
df869b63 | 673 | s64 incr = ktime_to_ns(interval); |
c0a31329 TG |
674 | |
675 | orun = ktime_divns(delta, incr); | |
676 | timer->expires = ktime_add_ns(timer->expires, incr * orun); | |
677 | if (timer->expires.tv64 > now.tv64) | |
678 | return orun; | |
679 | /* | |
680 | * This (and the ktime_add() below) is the | |
681 | * correction for exact: | |
682 | */ | |
683 | orun++; | |
684 | } | |
685 | timer->expires = ktime_add(timer->expires, interval); | |
13788ccc TG |
686 | /* |
687 | * Make sure, that the result did not wrap with a very large | |
688 | * interval. | |
689 | */ | |
690 | if (timer->expires.tv64 < 0) | |
691 | timer->expires = ktime_set(KTIME_SEC_MAX, 0); | |
c0a31329 TG |
692 | |
693 | return orun; | |
694 | } | |
6bdb6b62 | 695 | EXPORT_SYMBOL_GPL(hrtimer_forward); |
c0a31329 TG |
696 | |
697 | /* | |
698 | * enqueue_hrtimer - internal function to (re)start a timer | |
699 | * | |
700 | * The timer is inserted in expiry order. Insertion into the | |
701 | * red black tree is O(log(n)). Must hold the base lock. | |
702 | */ | |
3c8aa39d | 703 | static void enqueue_hrtimer(struct hrtimer *timer, |
54cdfdb4 | 704 | struct hrtimer_clock_base *base, int reprogram) |
c0a31329 TG |
705 | { |
706 | struct rb_node **link = &base->active.rb_node; | |
c0a31329 TG |
707 | struct rb_node *parent = NULL; |
708 | struct hrtimer *entry; | |
99bc2fcb | 709 | int leftmost = 1; |
c0a31329 TG |
710 | |
711 | /* | |
712 | * Find the right place in the rbtree: | |
713 | */ | |
714 | while (*link) { | |
715 | parent = *link; | |
716 | entry = rb_entry(parent, struct hrtimer, node); | |
717 | /* | |
718 | * We dont care about collisions. Nodes with | |
719 | * the same expiry time stay together. | |
720 | */ | |
99bc2fcb | 721 | if (timer->expires.tv64 < entry->expires.tv64) { |
c0a31329 | 722 | link = &(*link)->rb_left; |
99bc2fcb | 723 | } else { |
c0a31329 | 724 | link = &(*link)->rb_right; |
99bc2fcb IM |
725 | leftmost = 0; |
726 | } | |
c0a31329 TG |
727 | } |
728 | ||
729 | /* | |
288867ec TG |
730 | * Insert the timer to the rbtree and check whether it |
731 | * replaces the first pending timer | |
c0a31329 | 732 | */ |
99bc2fcb | 733 | if (leftmost) { |
54cdfdb4 TG |
734 | /* |
735 | * Reprogram the clock event device. When the timer is already | |
736 | * expired hrtimer_enqueue_reprogram has either called the | |
737 | * callback or added it to the pending list and raised the | |
738 | * softirq. | |
739 | * | |
740 | * This is a NOP for !HIGHRES | |
741 | */ | |
742 | if (reprogram && hrtimer_enqueue_reprogram(timer, base)) | |
743 | return; | |
744 | ||
745 | base->first = &timer->node; | |
746 | } | |
747 | ||
c0a31329 TG |
748 | rb_link_node(&timer->node, parent, link); |
749 | rb_insert_color(&timer->node, &base->active); | |
303e967f TG |
750 | /* |
751 | * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the | |
752 | * state of a possibly running callback. | |
753 | */ | |
754 | timer->state |= HRTIMER_STATE_ENQUEUED; | |
288867ec | 755 | } |
c0a31329 TG |
756 | |
757 | /* | |
758 | * __remove_hrtimer - internal function to remove a timer | |
759 | * | |
760 | * Caller must hold the base lock. | |
54cdfdb4 TG |
761 | * |
762 | * High resolution timer mode reprograms the clock event device when the | |
763 | * timer is the one which expires next. The caller can disable this by setting | |
764 | * reprogram to zero. This is useful, when the context does a reprogramming | |
765 | * anyway (e.g. timer interrupt) | |
c0a31329 | 766 | */ |
3c8aa39d | 767 | static void __remove_hrtimer(struct hrtimer *timer, |
303e967f | 768 | struct hrtimer_clock_base *base, |
54cdfdb4 | 769 | unsigned long newstate, int reprogram) |
c0a31329 | 770 | { |
54cdfdb4 TG |
771 | /* High res. callback list. NOP for !HIGHRES */ |
772 | if (hrtimer_cb_pending(timer)) | |
773 | hrtimer_remove_cb_pending(timer); | |
774 | else { | |
775 | /* | |
776 | * Remove the timer from the rbtree and replace the | |
777 | * first entry pointer if necessary. | |
778 | */ | |
779 | if (base->first == &timer->node) { | |
780 | base->first = rb_next(&timer->node); | |
781 | /* Reprogram the clock event device. if enabled */ | |
782 | if (reprogram && hrtimer_hres_active()) | |
783 | hrtimer_force_reprogram(base->cpu_base); | |
784 | } | |
785 | rb_erase(&timer->node, &base->active); | |
786 | } | |
303e967f | 787 | timer->state = newstate; |
c0a31329 TG |
788 | } |
789 | ||
790 | /* | |
791 | * remove hrtimer, called with base lock held | |
792 | */ | |
793 | static inline int | |
3c8aa39d | 794 | remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base) |
c0a31329 | 795 | { |
303e967f | 796 | if (hrtimer_is_queued(timer)) { |
54cdfdb4 TG |
797 | int reprogram; |
798 | ||
799 | /* | |
800 | * Remove the timer and force reprogramming when high | |
801 | * resolution mode is active and the timer is on the current | |
802 | * CPU. If we remove a timer on another CPU, reprogramming is | |
803 | * skipped. The interrupt event on this CPU is fired and | |
804 | * reprogramming happens in the interrupt handler. This is a | |
805 | * rare case and less expensive than a smp call. | |
806 | */ | |
82f67cd9 | 807 | timer_stats_hrtimer_clear_start_info(timer); |
54cdfdb4 TG |
808 | reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases); |
809 | __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE, | |
810 | reprogram); | |
c0a31329 TG |
811 | return 1; |
812 | } | |
813 | return 0; | |
814 | } | |
815 | ||
816 | /** | |
817 | * hrtimer_start - (re)start an relative timer on the current CPU | |
c0a31329 TG |
818 | * @timer: the timer to be added |
819 | * @tim: expiry time | |
820 | * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL) | |
821 | * | |
822 | * Returns: | |
823 | * 0 on success | |
824 | * 1 when the timer was active | |
825 | */ | |
826 | int | |
827 | hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode) | |
828 | { | |
3c8aa39d | 829 | struct hrtimer_clock_base *base, *new_base; |
c0a31329 TG |
830 | unsigned long flags; |
831 | int ret; | |
832 | ||
833 | base = lock_hrtimer_base(timer, &flags); | |
834 | ||
835 | /* Remove an active timer from the queue: */ | |
836 | ret = remove_hrtimer(timer, base); | |
837 | ||
838 | /* Switch the timer base, if necessary: */ | |
839 | new_base = switch_hrtimer_base(timer, base); | |
840 | ||
c9cb2e3d | 841 | if (mode == HRTIMER_MODE_REL) { |
c0a31329 | 842 | tim = ktime_add(tim, new_base->get_time()); |
06027bdd IM |
843 | /* |
844 | * CONFIG_TIME_LOW_RES is a temporary way for architectures | |
845 | * to signal that they simply return xtime in | |
846 | * do_gettimeoffset(). In this case we want to round up by | |
847 | * resolution when starting a relative timer, to avoid short | |
848 | * timeouts. This will go away with the GTOD framework. | |
849 | */ | |
850 | #ifdef CONFIG_TIME_LOW_RES | |
851 | tim = ktime_add(tim, base->resolution); | |
852 | #endif | |
62f0f61e TG |
853 | /* |
854 | * Careful here: User space might have asked for a | |
855 | * very long sleep, so the add above might result in a | |
856 | * negative number, which enqueues the timer in front | |
857 | * of the queue. | |
858 | */ | |
859 | if (tim.tv64 < 0) | |
860 | tim.tv64 = KTIME_MAX; | |
06027bdd | 861 | } |
c0a31329 TG |
862 | timer->expires = tim; |
863 | ||
82f67cd9 IM |
864 | timer_stats_hrtimer_set_start_info(timer); |
865 | ||
935c631d IM |
866 | /* |
867 | * Only allow reprogramming if the new base is on this CPU. | |
868 | * (it might still be on another CPU if the timer was pending) | |
869 | */ | |
870 | enqueue_hrtimer(timer, new_base, | |
871 | new_base->cpu_base == &__get_cpu_var(hrtimer_bases)); | |
c0a31329 TG |
872 | |
873 | unlock_hrtimer_base(timer, &flags); | |
874 | ||
875 | return ret; | |
876 | } | |
8d16b764 | 877 | EXPORT_SYMBOL_GPL(hrtimer_start); |
c0a31329 TG |
878 | |
879 | /** | |
880 | * hrtimer_try_to_cancel - try to deactivate a timer | |
c0a31329 TG |
881 | * @timer: hrtimer to stop |
882 | * | |
883 | * Returns: | |
884 | * 0 when the timer was not active | |
885 | * 1 when the timer was active | |
886 | * -1 when the timer is currently excuting the callback function and | |
fa9799e3 | 887 | * cannot be stopped |
c0a31329 TG |
888 | */ |
889 | int hrtimer_try_to_cancel(struct hrtimer *timer) | |
890 | { | |
3c8aa39d | 891 | struct hrtimer_clock_base *base; |
c0a31329 TG |
892 | unsigned long flags; |
893 | int ret = -1; | |
894 | ||
895 | base = lock_hrtimer_base(timer, &flags); | |
896 | ||
303e967f | 897 | if (!hrtimer_callback_running(timer)) |
c0a31329 TG |
898 | ret = remove_hrtimer(timer, base); |
899 | ||
900 | unlock_hrtimer_base(timer, &flags); | |
901 | ||
902 | return ret; | |
903 | ||
904 | } | |
8d16b764 | 905 | EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel); |
c0a31329 TG |
906 | |
907 | /** | |
908 | * hrtimer_cancel - cancel a timer and wait for the handler to finish. | |
c0a31329 TG |
909 | * @timer: the timer to be cancelled |
910 | * | |
911 | * Returns: | |
912 | * 0 when the timer was not active | |
913 | * 1 when the timer was active | |
914 | */ | |
915 | int hrtimer_cancel(struct hrtimer *timer) | |
916 | { | |
917 | for (;;) { | |
918 | int ret = hrtimer_try_to_cancel(timer); | |
919 | ||
920 | if (ret >= 0) | |
921 | return ret; | |
5ef37b19 | 922 | cpu_relax(); |
c0a31329 TG |
923 | } |
924 | } | |
8d16b764 | 925 | EXPORT_SYMBOL_GPL(hrtimer_cancel); |
c0a31329 TG |
926 | |
927 | /** | |
928 | * hrtimer_get_remaining - get remaining time for the timer | |
c0a31329 TG |
929 | * @timer: the timer to read |
930 | */ | |
931 | ktime_t hrtimer_get_remaining(const struct hrtimer *timer) | |
932 | { | |
3c8aa39d | 933 | struct hrtimer_clock_base *base; |
c0a31329 TG |
934 | unsigned long flags; |
935 | ktime_t rem; | |
936 | ||
937 | base = lock_hrtimer_base(timer, &flags); | |
3c8aa39d | 938 | rem = ktime_sub(timer->expires, base->get_time()); |
c0a31329 TG |
939 | unlock_hrtimer_base(timer, &flags); |
940 | ||
941 | return rem; | |
942 | } | |
8d16b764 | 943 | EXPORT_SYMBOL_GPL(hrtimer_get_remaining); |
c0a31329 | 944 | |
fd064b9b | 945 | #if defined(CONFIG_NO_IDLE_HZ) || defined(CONFIG_NO_HZ) |
69239749 TL |
946 | /** |
947 | * hrtimer_get_next_event - get the time until next expiry event | |
948 | * | |
949 | * Returns the delta to the next expiry event or KTIME_MAX if no timer | |
950 | * is pending. | |
951 | */ | |
952 | ktime_t hrtimer_get_next_event(void) | |
953 | { | |
3c8aa39d TG |
954 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
955 | struct hrtimer_clock_base *base = cpu_base->clock_base; | |
69239749 TL |
956 | ktime_t delta, mindelta = { .tv64 = KTIME_MAX }; |
957 | unsigned long flags; | |
958 | int i; | |
959 | ||
3c8aa39d TG |
960 | spin_lock_irqsave(&cpu_base->lock, flags); |
961 | ||
54cdfdb4 TG |
962 | if (!hrtimer_hres_active()) { |
963 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) { | |
964 | struct hrtimer *timer; | |
69239749 | 965 | |
54cdfdb4 TG |
966 | if (!base->first) |
967 | continue; | |
3c8aa39d | 968 | |
54cdfdb4 TG |
969 | timer = rb_entry(base->first, struct hrtimer, node); |
970 | delta.tv64 = timer->expires.tv64; | |
971 | delta = ktime_sub(delta, base->get_time()); | |
972 | if (delta.tv64 < mindelta.tv64) | |
973 | mindelta.tv64 = delta.tv64; | |
974 | } | |
69239749 | 975 | } |
3c8aa39d TG |
976 | |
977 | spin_unlock_irqrestore(&cpu_base->lock, flags); | |
978 | ||
69239749 TL |
979 | if (mindelta.tv64 < 0) |
980 | mindelta.tv64 = 0; | |
981 | return mindelta; | |
982 | } | |
983 | #endif | |
984 | ||
c0a31329 | 985 | /** |
7978672c | 986 | * hrtimer_init - initialize a timer to the given clock |
7978672c | 987 | * @timer: the timer to be initialized |
c0a31329 | 988 | * @clock_id: the clock to be used |
7978672c | 989 | * @mode: timer mode abs/rel |
c0a31329 | 990 | */ |
7978672c GA |
991 | void hrtimer_init(struct hrtimer *timer, clockid_t clock_id, |
992 | enum hrtimer_mode mode) | |
c0a31329 | 993 | { |
3c8aa39d | 994 | struct hrtimer_cpu_base *cpu_base; |
c0a31329 | 995 | |
7978672c GA |
996 | memset(timer, 0, sizeof(struct hrtimer)); |
997 | ||
3c8aa39d | 998 | cpu_base = &__raw_get_cpu_var(hrtimer_bases); |
c0a31329 | 999 | |
c9cb2e3d | 1000 | if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS) |
7978672c GA |
1001 | clock_id = CLOCK_MONOTONIC; |
1002 | ||
3c8aa39d | 1003 | timer->base = &cpu_base->clock_base[clock_id]; |
d3d74453 | 1004 | INIT_LIST_HEAD(&timer->cb_entry); |
54cdfdb4 | 1005 | hrtimer_init_timer_hres(timer); |
82f67cd9 IM |
1006 | |
1007 | #ifdef CONFIG_TIMER_STATS | |
1008 | timer->start_site = NULL; | |
1009 | timer->start_pid = -1; | |
1010 | memset(timer->start_comm, 0, TASK_COMM_LEN); | |
1011 | #endif | |
c0a31329 | 1012 | } |
8d16b764 | 1013 | EXPORT_SYMBOL_GPL(hrtimer_init); |
c0a31329 TG |
1014 | |
1015 | /** | |
1016 | * hrtimer_get_res - get the timer resolution for a clock | |
c0a31329 TG |
1017 | * @which_clock: which clock to query |
1018 | * @tp: pointer to timespec variable to store the resolution | |
1019 | * | |
72fd4a35 RD |
1020 | * Store the resolution of the clock selected by @which_clock in the |
1021 | * variable pointed to by @tp. | |
c0a31329 TG |
1022 | */ |
1023 | int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp) | |
1024 | { | |
3c8aa39d | 1025 | struct hrtimer_cpu_base *cpu_base; |
c0a31329 | 1026 | |
3c8aa39d TG |
1027 | cpu_base = &__raw_get_cpu_var(hrtimer_bases); |
1028 | *tp = ktime_to_timespec(cpu_base->clock_base[which_clock].resolution); | |
c0a31329 TG |
1029 | |
1030 | return 0; | |
1031 | } | |
8d16b764 | 1032 | EXPORT_SYMBOL_GPL(hrtimer_get_res); |
c0a31329 | 1033 | |
d3d74453 PZ |
1034 | static void run_hrtimer_pending(struct hrtimer_cpu_base *cpu_base) |
1035 | { | |
1036 | spin_lock_irq(&cpu_base->lock); | |
1037 | ||
1038 | while (!list_empty(&cpu_base->cb_pending)) { | |
1039 | enum hrtimer_restart (*fn)(struct hrtimer *); | |
1040 | struct hrtimer *timer; | |
1041 | int restart; | |
1042 | ||
1043 | timer = list_entry(cpu_base->cb_pending.next, | |
1044 | struct hrtimer, cb_entry); | |
1045 | ||
1046 | timer_stats_account_hrtimer(timer); | |
1047 | ||
1048 | fn = timer->function; | |
1049 | __remove_hrtimer(timer, timer->base, HRTIMER_STATE_CALLBACK, 0); | |
1050 | spin_unlock_irq(&cpu_base->lock); | |
1051 | ||
1052 | restart = fn(timer); | |
1053 | ||
1054 | spin_lock_irq(&cpu_base->lock); | |
1055 | ||
1056 | timer->state &= ~HRTIMER_STATE_CALLBACK; | |
1057 | if (restart == HRTIMER_RESTART) { | |
1058 | BUG_ON(hrtimer_active(timer)); | |
1059 | /* | |
1060 | * Enqueue the timer, allow reprogramming of the event | |
1061 | * device | |
1062 | */ | |
1063 | enqueue_hrtimer(timer, timer->base, 1); | |
1064 | } else if (hrtimer_active(timer)) { | |
1065 | /* | |
1066 | * If the timer was rearmed on another CPU, reprogram | |
1067 | * the event device. | |
1068 | */ | |
1069 | if (timer->base->first == &timer->node) | |
1070 | hrtimer_reprogram(timer, timer->base); | |
1071 | } | |
1072 | } | |
1073 | spin_unlock_irq(&cpu_base->lock); | |
1074 | } | |
1075 | ||
1076 | static void __run_hrtimer(struct hrtimer *timer) | |
1077 | { | |
1078 | struct hrtimer_clock_base *base = timer->base; | |
1079 | struct hrtimer_cpu_base *cpu_base = base->cpu_base; | |
1080 | enum hrtimer_restart (*fn)(struct hrtimer *); | |
1081 | int restart; | |
1082 | ||
1083 | __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0); | |
1084 | timer_stats_account_hrtimer(timer); | |
1085 | ||
1086 | fn = timer->function; | |
1087 | if (timer->cb_mode == HRTIMER_CB_IRQSAFE_NO_SOFTIRQ) { | |
1088 | /* | |
1089 | * Used for scheduler timers, avoid lock inversion with | |
1090 | * rq->lock and tasklist_lock. | |
1091 | * | |
1092 | * These timers are required to deal with enqueue expiry | |
1093 | * themselves and are not allowed to migrate. | |
1094 | */ | |
1095 | spin_unlock(&cpu_base->lock); | |
1096 | restart = fn(timer); | |
1097 | spin_lock(&cpu_base->lock); | |
1098 | } else | |
1099 | restart = fn(timer); | |
1100 | ||
1101 | /* | |
1102 | * Note: We clear the CALLBACK bit after enqueue_hrtimer to avoid | |
1103 | * reprogramming of the event hardware. This happens at the end of this | |
1104 | * function anyway. | |
1105 | */ | |
1106 | if (restart != HRTIMER_NORESTART) { | |
1107 | BUG_ON(timer->state != HRTIMER_STATE_CALLBACK); | |
1108 | enqueue_hrtimer(timer, base, 0); | |
1109 | } | |
1110 | timer->state &= ~HRTIMER_STATE_CALLBACK; | |
1111 | } | |
1112 | ||
54cdfdb4 TG |
1113 | #ifdef CONFIG_HIGH_RES_TIMERS |
1114 | ||
1115 | /* | |
1116 | * High resolution timer interrupt | |
1117 | * Called with interrupts disabled | |
1118 | */ | |
1119 | void hrtimer_interrupt(struct clock_event_device *dev) | |
1120 | { | |
1121 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); | |
1122 | struct hrtimer_clock_base *base; | |
1123 | ktime_t expires_next, now; | |
1124 | int i, raise = 0; | |
1125 | ||
1126 | BUG_ON(!cpu_base->hres_active); | |
1127 | cpu_base->nr_events++; | |
1128 | dev->next_event.tv64 = KTIME_MAX; | |
1129 | ||
1130 | retry: | |
1131 | now = ktime_get(); | |
1132 | ||
1133 | expires_next.tv64 = KTIME_MAX; | |
1134 | ||
1135 | base = cpu_base->clock_base; | |
1136 | ||
1137 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { | |
1138 | ktime_t basenow; | |
1139 | struct rb_node *node; | |
1140 | ||
1141 | spin_lock(&cpu_base->lock); | |
1142 | ||
1143 | basenow = ktime_add(now, base->offset); | |
1144 | ||
1145 | while ((node = base->first)) { | |
1146 | struct hrtimer *timer; | |
1147 | ||
1148 | timer = rb_entry(node, struct hrtimer, node); | |
1149 | ||
1150 | if (basenow.tv64 < timer->expires.tv64) { | |
1151 | ktime_t expires; | |
1152 | ||
1153 | expires = ktime_sub(timer->expires, | |
1154 | base->offset); | |
1155 | if (expires.tv64 < expires_next.tv64) | |
1156 | expires_next = expires; | |
1157 | break; | |
1158 | } | |
1159 | ||
1160 | /* Move softirq callbacks to the pending list */ | |
1161 | if (timer->cb_mode == HRTIMER_CB_SOFTIRQ) { | |
1162 | __remove_hrtimer(timer, base, | |
1163 | HRTIMER_STATE_PENDING, 0); | |
1164 | list_add_tail(&timer->cb_entry, | |
1165 | &base->cpu_base->cb_pending); | |
1166 | raise = 1; | |
1167 | continue; | |
1168 | } | |
1169 | ||
d3d74453 | 1170 | __run_hrtimer(timer); |
54cdfdb4 TG |
1171 | } |
1172 | spin_unlock(&cpu_base->lock); | |
1173 | base++; | |
1174 | } | |
1175 | ||
1176 | cpu_base->expires_next = expires_next; | |
1177 | ||
1178 | /* Reprogramming necessary ? */ | |
1179 | if (expires_next.tv64 != KTIME_MAX) { | |
1180 | if (tick_program_event(expires_next, 0)) | |
1181 | goto retry; | |
1182 | } | |
1183 | ||
1184 | /* Raise softirq ? */ | |
1185 | if (raise) | |
1186 | raise_softirq(HRTIMER_SOFTIRQ); | |
1187 | } | |
1188 | ||
1189 | static void run_hrtimer_softirq(struct softirq_action *h) | |
1190 | { | |
d3d74453 PZ |
1191 | run_hrtimer_pending(&__get_cpu_var(hrtimer_bases)); |
1192 | } | |
54cdfdb4 | 1193 | |
d3d74453 | 1194 | #endif /* CONFIG_HIGH_RES_TIMERS */ |
82f67cd9 | 1195 | |
d3d74453 PZ |
1196 | /* |
1197 | * Called from timer softirq every jiffy, expire hrtimers: | |
1198 | * | |
1199 | * For HRT its the fall back code to run the softirq in the timer | |
1200 | * softirq context in case the hrtimer initialization failed or has | |
1201 | * not been done yet. | |
1202 | */ | |
1203 | void hrtimer_run_pending(void) | |
1204 | { | |
1205 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); | |
54cdfdb4 | 1206 | |
d3d74453 PZ |
1207 | if (hrtimer_hres_active()) |
1208 | return; | |
54cdfdb4 | 1209 | |
d3d74453 PZ |
1210 | /* |
1211 | * This _is_ ugly: We have to check in the softirq context, | |
1212 | * whether we can switch to highres and / or nohz mode. The | |
1213 | * clocksource switch happens in the timer interrupt with | |
1214 | * xtime_lock held. Notification from there only sets the | |
1215 | * check bit in the tick_oneshot code, otherwise we might | |
1216 | * deadlock vs. xtime_lock. | |
1217 | */ | |
1218 | if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) | |
1219 | hrtimer_switch_to_hres(); | |
54cdfdb4 | 1220 | |
d3d74453 | 1221 | run_hrtimer_pending(cpu_base); |
54cdfdb4 TG |
1222 | } |
1223 | ||
c0a31329 | 1224 | /* |
d3d74453 | 1225 | * Called from hardirq context every jiffy |
c0a31329 | 1226 | */ |
3c8aa39d TG |
1227 | static inline void run_hrtimer_queue(struct hrtimer_cpu_base *cpu_base, |
1228 | int index) | |
c0a31329 | 1229 | { |
288867ec | 1230 | struct rb_node *node; |
3c8aa39d | 1231 | struct hrtimer_clock_base *base = &cpu_base->clock_base[index]; |
c0a31329 | 1232 | |
3055adda DS |
1233 | if (!base->first) |
1234 | return; | |
1235 | ||
92127c7a TG |
1236 | if (base->get_softirq_time) |
1237 | base->softirq_time = base->get_softirq_time(); | |
1238 | ||
d3d74453 | 1239 | spin_lock(&cpu_base->lock); |
c0a31329 | 1240 | |
288867ec | 1241 | while ((node = base->first)) { |
c0a31329 | 1242 | struct hrtimer *timer; |
c0a31329 | 1243 | |
288867ec | 1244 | timer = rb_entry(node, struct hrtimer, node); |
92127c7a | 1245 | if (base->softirq_time.tv64 <= timer->expires.tv64) |
c0a31329 TG |
1246 | break; |
1247 | ||
d3d74453 PZ |
1248 | if (timer->cb_mode == HRTIMER_CB_SOFTIRQ) { |
1249 | __remove_hrtimer(timer, base, HRTIMER_STATE_PENDING, 0); | |
1250 | list_add_tail(&timer->cb_entry, | |
1251 | &base->cpu_base->cb_pending); | |
1252 | continue; | |
b75f7a51 | 1253 | } |
d3d74453 PZ |
1254 | |
1255 | __run_hrtimer(timer); | |
c0a31329 | 1256 | } |
d3d74453 | 1257 | spin_unlock(&cpu_base->lock); |
c0a31329 TG |
1258 | } |
1259 | ||
c0a31329 TG |
1260 | void hrtimer_run_queues(void) |
1261 | { | |
3c8aa39d | 1262 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
c0a31329 TG |
1263 | int i; |
1264 | ||
54cdfdb4 TG |
1265 | if (hrtimer_hres_active()) |
1266 | return; | |
1267 | ||
3c8aa39d | 1268 | hrtimer_get_softirq_time(cpu_base); |
92127c7a | 1269 | |
3c8aa39d TG |
1270 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) |
1271 | run_hrtimer_queue(cpu_base, i); | |
c0a31329 TG |
1272 | } |
1273 | ||
10c94ec1 TG |
1274 | /* |
1275 | * Sleep related functions: | |
1276 | */ | |
c9cb2e3d | 1277 | static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer) |
00362e33 TG |
1278 | { |
1279 | struct hrtimer_sleeper *t = | |
1280 | container_of(timer, struct hrtimer_sleeper, timer); | |
1281 | struct task_struct *task = t->task; | |
1282 | ||
1283 | t->task = NULL; | |
1284 | if (task) | |
1285 | wake_up_process(task); | |
1286 | ||
1287 | return HRTIMER_NORESTART; | |
1288 | } | |
1289 | ||
36c8b586 | 1290 | void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task) |
00362e33 TG |
1291 | { |
1292 | sl->timer.function = hrtimer_wakeup; | |
1293 | sl->task = task; | |
54cdfdb4 | 1294 | #ifdef CONFIG_HIGH_RES_TIMERS |
37bb6cb4 | 1295 | sl->timer.cb_mode = HRTIMER_CB_IRQSAFE_NO_SOFTIRQ; |
54cdfdb4 | 1296 | #endif |
00362e33 TG |
1297 | } |
1298 | ||
669d7868 | 1299 | static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode) |
432569bb | 1300 | { |
669d7868 | 1301 | hrtimer_init_sleeper(t, current); |
10c94ec1 | 1302 | |
432569bb RZ |
1303 | do { |
1304 | set_current_state(TASK_INTERRUPTIBLE); | |
1305 | hrtimer_start(&t->timer, t->timer.expires, mode); | |
37bb6cb4 PZ |
1306 | if (!hrtimer_active(&t->timer)) |
1307 | t->task = NULL; | |
432569bb | 1308 | |
54cdfdb4 TG |
1309 | if (likely(t->task)) |
1310 | schedule(); | |
432569bb | 1311 | |
669d7868 | 1312 | hrtimer_cancel(&t->timer); |
c9cb2e3d | 1313 | mode = HRTIMER_MODE_ABS; |
669d7868 TG |
1314 | |
1315 | } while (t->task && !signal_pending(current)); | |
432569bb | 1316 | |
3588a085 PZ |
1317 | __set_current_state(TASK_RUNNING); |
1318 | ||
669d7868 | 1319 | return t->task == NULL; |
10c94ec1 TG |
1320 | } |
1321 | ||
1711ef38 | 1322 | long __sched hrtimer_nanosleep_restart(struct restart_block *restart) |
10c94ec1 | 1323 | { |
669d7868 | 1324 | struct hrtimer_sleeper t; |
04c22714 | 1325 | struct timespec *rmtp; |
432569bb | 1326 | ktime_t time; |
10c94ec1 TG |
1327 | |
1328 | restart->fn = do_no_restart_syscall; | |
1329 | ||
c9cb2e3d | 1330 | hrtimer_init(&t.timer, restart->arg0, HRTIMER_MODE_ABS); |
1711ef38 | 1331 | t.timer.expires.tv64 = ((u64)restart->arg3 << 32) | (u64) restart->arg2; |
10c94ec1 | 1332 | |
c9cb2e3d | 1333 | if (do_nanosleep(&t, HRTIMER_MODE_ABS)) |
10c94ec1 TG |
1334 | return 0; |
1335 | ||
04c22714 | 1336 | rmtp = (struct timespec *)restart->arg1; |
432569bb RZ |
1337 | if (rmtp) { |
1338 | time = ktime_sub(t.timer.expires, t.timer.base->get_time()); | |
1339 | if (time.tv64 <= 0) | |
1340 | return 0; | |
04c22714 | 1341 | *rmtp = ktime_to_timespec(time); |
432569bb | 1342 | } |
10c94ec1 | 1343 | |
1711ef38 | 1344 | restart->fn = hrtimer_nanosleep_restart; |
10c94ec1 TG |
1345 | |
1346 | /* The other values in restart are already filled in */ | |
1347 | return -ERESTART_RESTARTBLOCK; | |
1348 | } | |
1349 | ||
04c22714 | 1350 | long hrtimer_nanosleep(struct timespec *rqtp, struct timespec *rmtp, |
10c94ec1 TG |
1351 | const enum hrtimer_mode mode, const clockid_t clockid) |
1352 | { | |
1353 | struct restart_block *restart; | |
669d7868 | 1354 | struct hrtimer_sleeper t; |
10c94ec1 TG |
1355 | ktime_t rem; |
1356 | ||
432569bb RZ |
1357 | hrtimer_init(&t.timer, clockid, mode); |
1358 | t.timer.expires = timespec_to_ktime(*rqtp); | |
1359 | if (do_nanosleep(&t, mode)) | |
10c94ec1 TG |
1360 | return 0; |
1361 | ||
7978672c | 1362 | /* Absolute timers do not update the rmtp value and restart: */ |
c9cb2e3d | 1363 | if (mode == HRTIMER_MODE_ABS) |
10c94ec1 TG |
1364 | return -ERESTARTNOHAND; |
1365 | ||
432569bb RZ |
1366 | if (rmtp) { |
1367 | rem = ktime_sub(t.timer.expires, t.timer.base->get_time()); | |
1368 | if (rem.tv64 <= 0) | |
1369 | return 0; | |
04c22714 | 1370 | *rmtp = ktime_to_timespec(rem); |
432569bb | 1371 | } |
10c94ec1 TG |
1372 | |
1373 | restart = ¤t_thread_info()->restart_block; | |
1711ef38 TA |
1374 | restart->fn = hrtimer_nanosleep_restart; |
1375 | restart->arg0 = (unsigned long) t.timer.base->index; | |
1376 | restart->arg1 = (unsigned long) rmtp; | |
1377 | restart->arg2 = t.timer.expires.tv64 & 0xFFFFFFFF; | |
1378 | restart->arg3 = t.timer.expires.tv64 >> 32; | |
10c94ec1 TG |
1379 | |
1380 | return -ERESTART_RESTARTBLOCK; | |
1381 | } | |
1382 | ||
6ba1b912 TG |
1383 | asmlinkage long |
1384 | sys_nanosleep(struct timespec __user *rqtp, struct timespec __user *rmtp) | |
1385 | { | |
04c22714 AB |
1386 | struct timespec tu, rmt; |
1387 | int ret; | |
6ba1b912 TG |
1388 | |
1389 | if (copy_from_user(&tu, rqtp, sizeof(tu))) | |
1390 | return -EFAULT; | |
1391 | ||
1392 | if (!timespec_valid(&tu)) | |
1393 | return -EINVAL; | |
1394 | ||
04c22714 AB |
1395 | ret = hrtimer_nanosleep(&tu, rmtp ? &rmt : NULL, HRTIMER_MODE_REL, |
1396 | CLOCK_MONOTONIC); | |
1397 | ||
1398 | if (ret && rmtp) { | |
1399 | if (copy_to_user(rmtp, &rmt, sizeof(*rmtp))) | |
1400 | return -EFAULT; | |
1401 | } | |
1402 | ||
1403 | return ret; | |
6ba1b912 TG |
1404 | } |
1405 | ||
c0a31329 TG |
1406 | /* |
1407 | * Functions related to boot-time initialization: | |
1408 | */ | |
0ec160dd | 1409 | static void __cpuinit init_hrtimers_cpu(int cpu) |
c0a31329 | 1410 | { |
3c8aa39d | 1411 | struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu); |
c0a31329 TG |
1412 | int i; |
1413 | ||
3c8aa39d TG |
1414 | spin_lock_init(&cpu_base->lock); |
1415 | lockdep_set_class(&cpu_base->lock, &cpu_base->lock_key); | |
1416 | ||
1417 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) | |
1418 | cpu_base->clock_base[i].cpu_base = cpu_base; | |
1419 | ||
d3d74453 | 1420 | INIT_LIST_HEAD(&cpu_base->cb_pending); |
54cdfdb4 | 1421 | hrtimer_init_hres(cpu_base); |
c0a31329 TG |
1422 | } |
1423 | ||
1424 | #ifdef CONFIG_HOTPLUG_CPU | |
1425 | ||
3c8aa39d TG |
1426 | static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base, |
1427 | struct hrtimer_clock_base *new_base) | |
c0a31329 TG |
1428 | { |
1429 | struct hrtimer *timer; | |
1430 | struct rb_node *node; | |
1431 | ||
1432 | while ((node = rb_first(&old_base->active))) { | |
1433 | timer = rb_entry(node, struct hrtimer, node); | |
54cdfdb4 TG |
1434 | BUG_ON(hrtimer_callback_running(timer)); |
1435 | __remove_hrtimer(timer, old_base, HRTIMER_STATE_INACTIVE, 0); | |
c0a31329 | 1436 | timer->base = new_base; |
54cdfdb4 TG |
1437 | /* |
1438 | * Enqueue the timer. Allow reprogramming of the event device | |
1439 | */ | |
1440 | enqueue_hrtimer(timer, new_base, 1); | |
c0a31329 TG |
1441 | } |
1442 | } | |
1443 | ||
1444 | static void migrate_hrtimers(int cpu) | |
1445 | { | |
3c8aa39d | 1446 | struct hrtimer_cpu_base *old_base, *new_base; |
c0a31329 TG |
1447 | int i; |
1448 | ||
1449 | BUG_ON(cpu_online(cpu)); | |
3c8aa39d TG |
1450 | old_base = &per_cpu(hrtimer_bases, cpu); |
1451 | new_base = &get_cpu_var(hrtimer_bases); | |
c0a31329 | 1452 | |
54cdfdb4 TG |
1453 | tick_cancel_sched_timer(cpu); |
1454 | ||
c0a31329 | 1455 | local_irq_disable(); |
e81ce1f7 HC |
1456 | double_spin_lock(&new_base->lock, &old_base->lock, |
1457 | smp_processor_id() < cpu); | |
c0a31329 | 1458 | |
3c8aa39d | 1459 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
3c8aa39d TG |
1460 | migrate_hrtimer_list(&old_base->clock_base[i], |
1461 | &new_base->clock_base[i]); | |
c0a31329 TG |
1462 | } |
1463 | ||
e81ce1f7 HC |
1464 | double_spin_unlock(&new_base->lock, &old_base->lock, |
1465 | smp_processor_id() < cpu); | |
c0a31329 TG |
1466 | local_irq_enable(); |
1467 | put_cpu_var(hrtimer_bases); | |
1468 | } | |
1469 | #endif /* CONFIG_HOTPLUG_CPU */ | |
1470 | ||
8c78f307 | 1471 | static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self, |
c0a31329 TG |
1472 | unsigned long action, void *hcpu) |
1473 | { | |
7713a7d1 | 1474 | unsigned int cpu = (long)hcpu; |
c0a31329 TG |
1475 | |
1476 | switch (action) { | |
1477 | ||
1478 | case CPU_UP_PREPARE: | |
8bb78442 | 1479 | case CPU_UP_PREPARE_FROZEN: |
c0a31329 TG |
1480 | init_hrtimers_cpu(cpu); |
1481 | break; | |
1482 | ||
1483 | #ifdef CONFIG_HOTPLUG_CPU | |
1484 | case CPU_DEAD: | |
8bb78442 | 1485 | case CPU_DEAD_FROZEN: |
d316c57f | 1486 | clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &cpu); |
c0a31329 TG |
1487 | migrate_hrtimers(cpu); |
1488 | break; | |
1489 | #endif | |
1490 | ||
1491 | default: | |
1492 | break; | |
1493 | } | |
1494 | ||
1495 | return NOTIFY_OK; | |
1496 | } | |
1497 | ||
8c78f307 | 1498 | static struct notifier_block __cpuinitdata hrtimers_nb = { |
c0a31329 TG |
1499 | .notifier_call = hrtimer_cpu_notify, |
1500 | }; | |
1501 | ||
1502 | void __init hrtimers_init(void) | |
1503 | { | |
1504 | hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE, | |
1505 | (void *)(long)smp_processor_id()); | |
1506 | register_cpu_notifier(&hrtimers_nb); | |
54cdfdb4 TG |
1507 | #ifdef CONFIG_HIGH_RES_TIMERS |
1508 | open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq, NULL); | |
1509 | #endif | |
c0a31329 TG |
1510 | } |
1511 |