]>
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> | |
35 | #include <linux/module.h> | |
36 | #include <linux/percpu.h> | |
37 | #include <linux/hrtimer.h> | |
38 | #include <linux/notifier.h> | |
39 | #include <linux/syscalls.h> | |
54cdfdb4 | 40 | #include <linux/kallsyms.h> |
c0a31329 | 41 | #include <linux/interrupt.h> |
79bf2bb3 | 42 | #include <linux/tick.h> |
54cdfdb4 TG |
43 | #include <linux/seq_file.h> |
44 | #include <linux/err.h> | |
237fc6e7 | 45 | #include <linux/debugobjects.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 | ||
c0a31329 TG |
155 | /* |
156 | * Functions and macros which are different for UP/SMP systems are kept in a | |
157 | * single place | |
158 | */ | |
159 | #ifdef CONFIG_SMP | |
160 | ||
c0a31329 TG |
161 | /* |
162 | * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock | |
163 | * means that all timers which are tied to this base via timer->base are | |
164 | * locked, and the base itself is locked too. | |
165 | * | |
166 | * So __run_timers/migrate_timers can safely modify all timers which could | |
167 | * be found on the lists/queues. | |
168 | * | |
169 | * When the timer's base is locked, and the timer removed from list, it is | |
170 | * possible to set timer->base = NULL and drop the lock: the timer remains | |
171 | * locked. | |
172 | */ | |
3c8aa39d TG |
173 | static |
174 | struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer, | |
175 | unsigned long *flags) | |
c0a31329 | 176 | { |
3c8aa39d | 177 | struct hrtimer_clock_base *base; |
c0a31329 TG |
178 | |
179 | for (;;) { | |
180 | base = timer->base; | |
181 | if (likely(base != NULL)) { | |
3c8aa39d | 182 | spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
183 | if (likely(base == timer->base)) |
184 | return base; | |
185 | /* The timer has migrated to another CPU: */ | |
3c8aa39d | 186 | spin_unlock_irqrestore(&base->cpu_base->lock, *flags); |
c0a31329 TG |
187 | } |
188 | cpu_relax(); | |
189 | } | |
190 | } | |
191 | ||
192 | /* | |
193 | * Switch the timer base to the current CPU when possible. | |
194 | */ | |
3c8aa39d TG |
195 | static inline struct hrtimer_clock_base * |
196 | switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base) | |
c0a31329 | 197 | { |
3c8aa39d TG |
198 | struct hrtimer_clock_base *new_base; |
199 | struct hrtimer_cpu_base *new_cpu_base; | |
c0a31329 | 200 | |
3c8aa39d TG |
201 | new_cpu_base = &__get_cpu_var(hrtimer_bases); |
202 | new_base = &new_cpu_base->clock_base[base->index]; | |
c0a31329 TG |
203 | |
204 | if (base != new_base) { | |
205 | /* | |
206 | * We are trying to schedule the timer on the local CPU. | |
207 | * However we can't change timer's base while it is running, | |
208 | * so we keep it on the same CPU. No hassle vs. reprogramming | |
209 | * the event source in the high resolution case. The softirq | |
210 | * code will take care of this when the timer function has | |
211 | * completed. There is no conflict as we hold the lock until | |
212 | * the timer is enqueued. | |
213 | */ | |
54cdfdb4 | 214 | if (unlikely(hrtimer_callback_running(timer))) |
c0a31329 TG |
215 | return base; |
216 | ||
217 | /* See the comment in lock_timer_base() */ | |
218 | timer->base = NULL; | |
3c8aa39d TG |
219 | spin_unlock(&base->cpu_base->lock); |
220 | spin_lock(&new_base->cpu_base->lock); | |
c0a31329 TG |
221 | timer->base = new_base; |
222 | } | |
223 | return new_base; | |
224 | } | |
225 | ||
226 | #else /* CONFIG_SMP */ | |
227 | ||
3c8aa39d | 228 | static inline struct hrtimer_clock_base * |
c0a31329 TG |
229 | lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) |
230 | { | |
3c8aa39d | 231 | struct hrtimer_clock_base *base = timer->base; |
c0a31329 | 232 | |
3c8aa39d | 233 | spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
234 | |
235 | return base; | |
236 | } | |
237 | ||
54cdfdb4 | 238 | # define switch_hrtimer_base(t, b) (b) |
c0a31329 TG |
239 | |
240 | #endif /* !CONFIG_SMP */ | |
241 | ||
242 | /* | |
243 | * Functions for the union type storage format of ktime_t which are | |
244 | * too large for inlining: | |
245 | */ | |
246 | #if BITS_PER_LONG < 64 | |
247 | # ifndef CONFIG_KTIME_SCALAR | |
248 | /** | |
249 | * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable | |
c0a31329 TG |
250 | * @kt: addend |
251 | * @nsec: the scalar nsec value to add | |
252 | * | |
253 | * Returns the sum of kt and nsec in ktime_t format | |
254 | */ | |
255 | ktime_t ktime_add_ns(const ktime_t kt, u64 nsec) | |
256 | { | |
257 | ktime_t tmp; | |
258 | ||
259 | if (likely(nsec < NSEC_PER_SEC)) { | |
260 | tmp.tv64 = nsec; | |
261 | } else { | |
262 | unsigned long rem = do_div(nsec, NSEC_PER_SEC); | |
263 | ||
264 | tmp = ktime_set((long)nsec, rem); | |
265 | } | |
266 | ||
267 | return ktime_add(kt, tmp); | |
268 | } | |
b8b8fd2d DH |
269 | |
270 | EXPORT_SYMBOL_GPL(ktime_add_ns); | |
a272378d ACM |
271 | |
272 | /** | |
273 | * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable | |
274 | * @kt: minuend | |
275 | * @nsec: the scalar nsec value to subtract | |
276 | * | |
277 | * Returns the subtraction of @nsec from @kt in ktime_t format | |
278 | */ | |
279 | ktime_t ktime_sub_ns(const ktime_t kt, u64 nsec) | |
280 | { | |
281 | ktime_t tmp; | |
282 | ||
283 | if (likely(nsec < NSEC_PER_SEC)) { | |
284 | tmp.tv64 = nsec; | |
285 | } else { | |
286 | unsigned long rem = do_div(nsec, NSEC_PER_SEC); | |
287 | ||
288 | tmp = ktime_set((long)nsec, rem); | |
289 | } | |
290 | ||
291 | return ktime_sub(kt, tmp); | |
292 | } | |
293 | ||
294 | EXPORT_SYMBOL_GPL(ktime_sub_ns); | |
c0a31329 TG |
295 | # endif /* !CONFIG_KTIME_SCALAR */ |
296 | ||
297 | /* | |
298 | * Divide a ktime value by a nanosecond value | |
299 | */ | |
4d672e7a | 300 | u64 ktime_divns(const ktime_t kt, s64 div) |
c0a31329 | 301 | { |
900cfa46 | 302 | u64 dclc; |
c0a31329 TG |
303 | int sft = 0; |
304 | ||
900cfa46 | 305 | dclc = ktime_to_ns(kt); |
c0a31329 TG |
306 | /* Make sure the divisor is less than 2^32: */ |
307 | while (div >> 32) { | |
308 | sft++; | |
309 | div >>= 1; | |
310 | } | |
311 | dclc >>= sft; | |
312 | do_div(dclc, (unsigned long) div); | |
313 | ||
4d672e7a | 314 | return dclc; |
c0a31329 | 315 | } |
c0a31329 TG |
316 | #endif /* BITS_PER_LONG >= 64 */ |
317 | ||
5a7780e7 TG |
318 | /* |
319 | * Add two ktime values and do a safety check for overflow: | |
320 | */ | |
321 | ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs) | |
322 | { | |
323 | ktime_t res = ktime_add(lhs, rhs); | |
324 | ||
325 | /* | |
326 | * We use KTIME_SEC_MAX here, the maximum timeout which we can | |
327 | * return to user space in a timespec: | |
328 | */ | |
329 | if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64) | |
330 | res = ktime_set(KTIME_SEC_MAX, 0); | |
331 | ||
332 | return res; | |
333 | } | |
334 | ||
237fc6e7 TG |
335 | #ifdef CONFIG_DEBUG_OBJECTS_TIMERS |
336 | ||
337 | static struct debug_obj_descr hrtimer_debug_descr; | |
338 | ||
339 | /* | |
340 | * fixup_init is called when: | |
341 | * - an active object is initialized | |
342 | */ | |
343 | static int hrtimer_fixup_init(void *addr, enum debug_obj_state state) | |
344 | { | |
345 | struct hrtimer *timer = addr; | |
346 | ||
347 | switch (state) { | |
348 | case ODEBUG_STATE_ACTIVE: | |
349 | hrtimer_cancel(timer); | |
350 | debug_object_init(timer, &hrtimer_debug_descr); | |
351 | return 1; | |
352 | default: | |
353 | return 0; | |
354 | } | |
355 | } | |
356 | ||
357 | /* | |
358 | * fixup_activate is called when: | |
359 | * - an active object is activated | |
360 | * - an unknown object is activated (might be a statically initialized object) | |
361 | */ | |
362 | static int hrtimer_fixup_activate(void *addr, enum debug_obj_state state) | |
363 | { | |
364 | switch (state) { | |
365 | ||
366 | case ODEBUG_STATE_NOTAVAILABLE: | |
367 | WARN_ON_ONCE(1); | |
368 | return 0; | |
369 | ||
370 | case ODEBUG_STATE_ACTIVE: | |
371 | WARN_ON(1); | |
372 | ||
373 | default: | |
374 | return 0; | |
375 | } | |
376 | } | |
377 | ||
378 | /* | |
379 | * fixup_free is called when: | |
380 | * - an active object is freed | |
381 | */ | |
382 | static int hrtimer_fixup_free(void *addr, enum debug_obj_state state) | |
383 | { | |
384 | struct hrtimer *timer = addr; | |
385 | ||
386 | switch (state) { | |
387 | case ODEBUG_STATE_ACTIVE: | |
388 | hrtimer_cancel(timer); | |
389 | debug_object_free(timer, &hrtimer_debug_descr); | |
390 | return 1; | |
391 | default: | |
392 | return 0; | |
393 | } | |
394 | } | |
395 | ||
396 | static struct debug_obj_descr hrtimer_debug_descr = { | |
397 | .name = "hrtimer", | |
398 | .fixup_init = hrtimer_fixup_init, | |
399 | .fixup_activate = hrtimer_fixup_activate, | |
400 | .fixup_free = hrtimer_fixup_free, | |
401 | }; | |
402 | ||
403 | static inline void debug_hrtimer_init(struct hrtimer *timer) | |
404 | { | |
405 | debug_object_init(timer, &hrtimer_debug_descr); | |
406 | } | |
407 | ||
408 | static inline void debug_hrtimer_activate(struct hrtimer *timer) | |
409 | { | |
410 | debug_object_activate(timer, &hrtimer_debug_descr); | |
411 | } | |
412 | ||
413 | static inline void debug_hrtimer_deactivate(struct hrtimer *timer) | |
414 | { | |
415 | debug_object_deactivate(timer, &hrtimer_debug_descr); | |
416 | } | |
417 | ||
418 | static inline void debug_hrtimer_free(struct hrtimer *timer) | |
419 | { | |
420 | debug_object_free(timer, &hrtimer_debug_descr); | |
421 | } | |
422 | ||
423 | static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id, | |
424 | enum hrtimer_mode mode); | |
425 | ||
426 | void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id, | |
427 | enum hrtimer_mode mode) | |
428 | { | |
429 | debug_object_init_on_stack(timer, &hrtimer_debug_descr); | |
430 | __hrtimer_init(timer, clock_id, mode); | |
431 | } | |
432 | ||
433 | void destroy_hrtimer_on_stack(struct hrtimer *timer) | |
434 | { | |
435 | debug_object_free(timer, &hrtimer_debug_descr); | |
436 | } | |
437 | ||
438 | #else | |
439 | static inline void debug_hrtimer_init(struct hrtimer *timer) { } | |
440 | static inline void debug_hrtimer_activate(struct hrtimer *timer) { } | |
441 | static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { } | |
442 | #endif | |
443 | ||
54cdfdb4 TG |
444 | /* High resolution timer related functions */ |
445 | #ifdef CONFIG_HIGH_RES_TIMERS | |
446 | ||
447 | /* | |
448 | * High resolution timer enabled ? | |
449 | */ | |
450 | static int hrtimer_hres_enabled __read_mostly = 1; | |
451 | ||
452 | /* | |
453 | * Enable / Disable high resolution mode | |
454 | */ | |
455 | static int __init setup_hrtimer_hres(char *str) | |
456 | { | |
457 | if (!strcmp(str, "off")) | |
458 | hrtimer_hres_enabled = 0; | |
459 | else if (!strcmp(str, "on")) | |
460 | hrtimer_hres_enabled = 1; | |
461 | else | |
462 | return 0; | |
463 | return 1; | |
464 | } | |
465 | ||
466 | __setup("highres=", setup_hrtimer_hres); | |
467 | ||
468 | /* | |
469 | * hrtimer_high_res_enabled - query, if the highres mode is enabled | |
470 | */ | |
471 | static inline int hrtimer_is_hres_enabled(void) | |
472 | { | |
473 | return hrtimer_hres_enabled; | |
474 | } | |
475 | ||
476 | /* | |
477 | * Is the high resolution mode active ? | |
478 | */ | |
479 | static inline int hrtimer_hres_active(void) | |
480 | { | |
481 | return __get_cpu_var(hrtimer_bases).hres_active; | |
482 | } | |
483 | ||
484 | /* | |
485 | * Reprogram the event source with checking both queues for the | |
486 | * next event | |
487 | * Called with interrupts disabled and base->lock held | |
488 | */ | |
489 | static void hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base) | |
490 | { | |
491 | int i; | |
492 | struct hrtimer_clock_base *base = cpu_base->clock_base; | |
493 | ktime_t expires; | |
494 | ||
495 | cpu_base->expires_next.tv64 = KTIME_MAX; | |
496 | ||
497 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) { | |
498 | struct hrtimer *timer; | |
499 | ||
500 | if (!base->first) | |
501 | continue; | |
502 | timer = rb_entry(base->first, struct hrtimer, node); | |
cc584b21 | 503 | expires = ktime_sub(hrtimer_get_expires(timer), base->offset); |
54cdfdb4 TG |
504 | if (expires.tv64 < cpu_base->expires_next.tv64) |
505 | cpu_base->expires_next = expires; | |
506 | } | |
507 | ||
508 | if (cpu_base->expires_next.tv64 != KTIME_MAX) | |
509 | tick_program_event(cpu_base->expires_next, 1); | |
510 | } | |
511 | ||
512 | /* | |
513 | * Shared reprogramming for clock_realtime and clock_monotonic | |
514 | * | |
515 | * When a timer is enqueued and expires earlier than the already enqueued | |
516 | * timers, we have to check, whether it expires earlier than the timer for | |
517 | * which the clock event device was armed. | |
518 | * | |
519 | * Called with interrupts disabled and base->cpu_base.lock held | |
520 | */ | |
521 | static int hrtimer_reprogram(struct hrtimer *timer, | |
522 | struct hrtimer_clock_base *base) | |
523 | { | |
524 | ktime_t *expires_next = &__get_cpu_var(hrtimer_bases).expires_next; | |
cc584b21 | 525 | ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset); |
54cdfdb4 TG |
526 | int res; |
527 | ||
cc584b21 | 528 | WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0); |
63070a79 | 529 | |
54cdfdb4 TG |
530 | /* |
531 | * When the callback is running, we do not reprogram the clock event | |
532 | * device. The timer callback is either running on a different CPU or | |
3a4fa0a2 | 533 | * the callback is executed in the hrtimer_interrupt context. The |
54cdfdb4 TG |
534 | * reprogramming is handled either by the softirq, which called the |
535 | * callback or at the end of the hrtimer_interrupt. | |
536 | */ | |
537 | if (hrtimer_callback_running(timer)) | |
538 | return 0; | |
539 | ||
63070a79 TG |
540 | /* |
541 | * CLOCK_REALTIME timer might be requested with an absolute | |
542 | * expiry time which is less than base->offset. Nothing wrong | |
543 | * about that, just avoid to call into the tick code, which | |
544 | * has now objections against negative expiry values. | |
545 | */ | |
546 | if (expires.tv64 < 0) | |
547 | return -ETIME; | |
548 | ||
54cdfdb4 TG |
549 | if (expires.tv64 >= expires_next->tv64) |
550 | return 0; | |
551 | ||
552 | /* | |
553 | * Clockevents returns -ETIME, when the event was in the past. | |
554 | */ | |
555 | res = tick_program_event(expires, 0); | |
556 | if (!IS_ERR_VALUE(res)) | |
557 | *expires_next = expires; | |
558 | return res; | |
559 | } | |
560 | ||
561 | ||
562 | /* | |
563 | * Retrigger next event is called after clock was set | |
564 | * | |
565 | * Called with interrupts disabled via on_each_cpu() | |
566 | */ | |
567 | static void retrigger_next_event(void *arg) | |
568 | { | |
569 | struct hrtimer_cpu_base *base; | |
570 | struct timespec realtime_offset; | |
571 | unsigned long seq; | |
572 | ||
573 | if (!hrtimer_hres_active()) | |
574 | return; | |
575 | ||
576 | do { | |
577 | seq = read_seqbegin(&xtime_lock); | |
578 | set_normalized_timespec(&realtime_offset, | |
579 | -wall_to_monotonic.tv_sec, | |
580 | -wall_to_monotonic.tv_nsec); | |
581 | } while (read_seqretry(&xtime_lock, seq)); | |
582 | ||
583 | base = &__get_cpu_var(hrtimer_bases); | |
584 | ||
585 | /* Adjust CLOCK_REALTIME offset */ | |
586 | spin_lock(&base->lock); | |
587 | base->clock_base[CLOCK_REALTIME].offset = | |
588 | timespec_to_ktime(realtime_offset); | |
589 | ||
590 | hrtimer_force_reprogram(base); | |
591 | spin_unlock(&base->lock); | |
592 | } | |
593 | ||
594 | /* | |
595 | * Clock realtime was set | |
596 | * | |
597 | * Change the offset of the realtime clock vs. the monotonic | |
598 | * clock. | |
599 | * | |
600 | * We might have to reprogram the high resolution timer interrupt. On | |
601 | * SMP we call the architecture specific code to retrigger _all_ high | |
602 | * resolution timer interrupts. On UP we just disable interrupts and | |
603 | * call the high resolution interrupt code. | |
604 | */ | |
605 | void clock_was_set(void) | |
606 | { | |
607 | /* Retrigger the CPU local events everywhere */ | |
15c8b6c1 | 608 | on_each_cpu(retrigger_next_event, NULL, 1); |
54cdfdb4 TG |
609 | } |
610 | ||
995f054f IM |
611 | /* |
612 | * During resume we might have to reprogram the high resolution timer | |
613 | * interrupt (on the local CPU): | |
614 | */ | |
615 | void hres_timers_resume(void) | |
616 | { | |
995f054f IM |
617 | /* Retrigger the CPU local events: */ |
618 | retrigger_next_event(NULL); | |
619 | } | |
620 | ||
54cdfdb4 TG |
621 | /* |
622 | * Initialize the high resolution related parts of cpu_base | |
623 | */ | |
624 | static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) | |
625 | { | |
626 | base->expires_next.tv64 = KTIME_MAX; | |
627 | base->hres_active = 0; | |
54cdfdb4 TG |
628 | } |
629 | ||
630 | /* | |
631 | * Initialize the high resolution related parts of a hrtimer | |
632 | */ | |
633 | static inline void hrtimer_init_timer_hres(struct hrtimer *timer) | |
634 | { | |
54cdfdb4 TG |
635 | } |
636 | ||
ca109491 PZ |
637 | static void __run_hrtimer(struct hrtimer *timer); |
638 | ||
54cdfdb4 TG |
639 | /* |
640 | * When High resolution timers are active, try to reprogram. Note, that in case | |
641 | * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry | |
642 | * check happens. The timer gets enqueued into the rbtree. The reprogramming | |
643 | * and expiry check is done in the hrtimer_interrupt or in the softirq. | |
644 | */ | |
645 | static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer, | |
646 | struct hrtimer_clock_base *base) | |
647 | { | |
648 | if (base->cpu_base->hres_active && hrtimer_reprogram(timer, base)) { | |
ca109491 PZ |
649 | /* |
650 | * XXX: recursion check? | |
651 | * hrtimer_forward() should round up with timer granularity | |
652 | * so that we never get into inf recursion here, | |
653 | * it doesn't do that though | |
654 | */ | |
655 | __run_hrtimer(timer); | |
656 | return 1; | |
54cdfdb4 TG |
657 | } |
658 | return 0; | |
659 | } | |
660 | ||
661 | /* | |
662 | * Switch to high resolution mode | |
663 | */ | |
f8953856 | 664 | static int hrtimer_switch_to_hres(void) |
54cdfdb4 | 665 | { |
820de5c3 IM |
666 | int cpu = smp_processor_id(); |
667 | struct hrtimer_cpu_base *base = &per_cpu(hrtimer_bases, cpu); | |
54cdfdb4 TG |
668 | unsigned long flags; |
669 | ||
670 | if (base->hres_active) | |
f8953856 | 671 | return 1; |
54cdfdb4 TG |
672 | |
673 | local_irq_save(flags); | |
674 | ||
675 | if (tick_init_highres()) { | |
676 | local_irq_restore(flags); | |
820de5c3 IM |
677 | printk(KERN_WARNING "Could not switch to high resolution " |
678 | "mode on CPU %d\n", cpu); | |
f8953856 | 679 | return 0; |
54cdfdb4 TG |
680 | } |
681 | base->hres_active = 1; | |
682 | base->clock_base[CLOCK_REALTIME].resolution = KTIME_HIGH_RES; | |
683 | base->clock_base[CLOCK_MONOTONIC].resolution = KTIME_HIGH_RES; | |
684 | ||
685 | tick_setup_sched_timer(); | |
686 | ||
687 | /* "Retrigger" the interrupt to get things going */ | |
688 | retrigger_next_event(NULL); | |
689 | local_irq_restore(flags); | |
edfed66e | 690 | printk(KERN_DEBUG "Switched to high resolution mode on CPU %d\n", |
54cdfdb4 | 691 | smp_processor_id()); |
f8953856 | 692 | return 1; |
54cdfdb4 TG |
693 | } |
694 | ||
695 | #else | |
696 | ||
697 | static inline int hrtimer_hres_active(void) { return 0; } | |
698 | static inline int hrtimer_is_hres_enabled(void) { return 0; } | |
f8953856 | 699 | static inline int hrtimer_switch_to_hres(void) { return 0; } |
54cdfdb4 TG |
700 | static inline void hrtimer_force_reprogram(struct hrtimer_cpu_base *base) { } |
701 | static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer, | |
702 | struct hrtimer_clock_base *base) | |
703 | { | |
704 | return 0; | |
705 | } | |
54cdfdb4 TG |
706 | static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { } |
707 | static inline void hrtimer_init_timer_hres(struct hrtimer *timer) { } | |
d3d74453 PZ |
708 | static inline int hrtimer_reprogram(struct hrtimer *timer, |
709 | struct hrtimer_clock_base *base) | |
710 | { | |
711 | return 0; | |
712 | } | |
54cdfdb4 TG |
713 | |
714 | #endif /* CONFIG_HIGH_RES_TIMERS */ | |
715 | ||
82f67cd9 IM |
716 | #ifdef CONFIG_TIMER_STATS |
717 | void __timer_stats_hrtimer_set_start_info(struct hrtimer *timer, void *addr) | |
718 | { | |
719 | if (timer->start_site) | |
720 | return; | |
721 | ||
722 | timer->start_site = addr; | |
723 | memcpy(timer->start_comm, current->comm, TASK_COMM_LEN); | |
724 | timer->start_pid = current->pid; | |
725 | } | |
726 | #endif | |
727 | ||
c0a31329 | 728 | /* |
6506f2aa | 729 | * Counterpart to lock_hrtimer_base above: |
c0a31329 TG |
730 | */ |
731 | static inline | |
732 | void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) | |
733 | { | |
3c8aa39d | 734 | spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags); |
c0a31329 TG |
735 | } |
736 | ||
737 | /** | |
738 | * hrtimer_forward - forward the timer expiry | |
c0a31329 | 739 | * @timer: hrtimer to forward |
44f21475 | 740 | * @now: forward past this time |
c0a31329 TG |
741 | * @interval: the interval to forward |
742 | * | |
743 | * Forward the timer expiry so it will expire in the future. | |
8dca6f33 | 744 | * Returns the number of overruns. |
c0a31329 | 745 | */ |
4d672e7a | 746 | u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval) |
c0a31329 | 747 | { |
4d672e7a | 748 | u64 orun = 1; |
44f21475 | 749 | ktime_t delta; |
c0a31329 | 750 | |
cc584b21 | 751 | delta = ktime_sub(now, hrtimer_get_expires(timer)); |
c0a31329 TG |
752 | |
753 | if (delta.tv64 < 0) | |
754 | return 0; | |
755 | ||
c9db4fa1 TG |
756 | if (interval.tv64 < timer->base->resolution.tv64) |
757 | interval.tv64 = timer->base->resolution.tv64; | |
758 | ||
c0a31329 | 759 | if (unlikely(delta.tv64 >= interval.tv64)) { |
df869b63 | 760 | s64 incr = ktime_to_ns(interval); |
c0a31329 TG |
761 | |
762 | orun = ktime_divns(delta, incr); | |
cc584b21 AV |
763 | hrtimer_add_expires_ns(timer, incr * orun); |
764 | if (hrtimer_get_expires_tv64(timer) > now.tv64) | |
c0a31329 TG |
765 | return orun; |
766 | /* | |
767 | * This (and the ktime_add() below) is the | |
768 | * correction for exact: | |
769 | */ | |
770 | orun++; | |
771 | } | |
cc584b21 | 772 | hrtimer_add_expires(timer, interval); |
c0a31329 TG |
773 | |
774 | return orun; | |
775 | } | |
6bdb6b62 | 776 | EXPORT_SYMBOL_GPL(hrtimer_forward); |
c0a31329 TG |
777 | |
778 | /* | |
779 | * enqueue_hrtimer - internal function to (re)start a timer | |
780 | * | |
781 | * The timer is inserted in expiry order. Insertion into the | |
782 | * red black tree is O(log(n)). Must hold the base lock. | |
783 | */ | |
3c8aa39d | 784 | static void enqueue_hrtimer(struct hrtimer *timer, |
54cdfdb4 | 785 | struct hrtimer_clock_base *base, int reprogram) |
c0a31329 TG |
786 | { |
787 | struct rb_node **link = &base->active.rb_node; | |
c0a31329 TG |
788 | struct rb_node *parent = NULL; |
789 | struct hrtimer *entry; | |
99bc2fcb | 790 | int leftmost = 1; |
c0a31329 | 791 | |
237fc6e7 TG |
792 | debug_hrtimer_activate(timer); |
793 | ||
c0a31329 TG |
794 | /* |
795 | * Find the right place in the rbtree: | |
796 | */ | |
797 | while (*link) { | |
798 | parent = *link; | |
799 | entry = rb_entry(parent, struct hrtimer, node); | |
800 | /* | |
801 | * We dont care about collisions. Nodes with | |
802 | * the same expiry time stay together. | |
803 | */ | |
cc584b21 AV |
804 | if (hrtimer_get_expires_tv64(timer) < |
805 | hrtimer_get_expires_tv64(entry)) { | |
c0a31329 | 806 | link = &(*link)->rb_left; |
99bc2fcb | 807 | } else { |
c0a31329 | 808 | link = &(*link)->rb_right; |
99bc2fcb IM |
809 | leftmost = 0; |
810 | } | |
c0a31329 TG |
811 | } |
812 | ||
813 | /* | |
288867ec TG |
814 | * Insert the timer to the rbtree and check whether it |
815 | * replaces the first pending timer | |
c0a31329 | 816 | */ |
99bc2fcb | 817 | if (leftmost) { |
54cdfdb4 TG |
818 | /* |
819 | * Reprogram the clock event device. When the timer is already | |
820 | * expired hrtimer_enqueue_reprogram has either called the | |
821 | * callback or added it to the pending list and raised the | |
822 | * softirq. | |
823 | * | |
824 | * This is a NOP for !HIGHRES | |
825 | */ | |
826 | if (reprogram && hrtimer_enqueue_reprogram(timer, base)) | |
827 | return; | |
828 | ||
829 | base->first = &timer->node; | |
830 | } | |
831 | ||
c0a31329 TG |
832 | rb_link_node(&timer->node, parent, link); |
833 | rb_insert_color(&timer->node, &base->active); | |
303e967f TG |
834 | /* |
835 | * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the | |
836 | * state of a possibly running callback. | |
837 | */ | |
838 | timer->state |= HRTIMER_STATE_ENQUEUED; | |
288867ec | 839 | } |
c0a31329 TG |
840 | |
841 | /* | |
842 | * __remove_hrtimer - internal function to remove a timer | |
843 | * | |
844 | * Caller must hold the base lock. | |
54cdfdb4 TG |
845 | * |
846 | * High resolution timer mode reprograms the clock event device when the | |
847 | * timer is the one which expires next. The caller can disable this by setting | |
848 | * reprogram to zero. This is useful, when the context does a reprogramming | |
849 | * anyway (e.g. timer interrupt) | |
c0a31329 | 850 | */ |
3c8aa39d | 851 | static void __remove_hrtimer(struct hrtimer *timer, |
303e967f | 852 | struct hrtimer_clock_base *base, |
54cdfdb4 | 853 | unsigned long newstate, int reprogram) |
c0a31329 | 854 | { |
ca109491 | 855 | if (timer->state & HRTIMER_STATE_ENQUEUED) { |
54cdfdb4 TG |
856 | /* |
857 | * Remove the timer from the rbtree and replace the | |
858 | * first entry pointer if necessary. | |
859 | */ | |
860 | if (base->first == &timer->node) { | |
861 | base->first = rb_next(&timer->node); | |
862 | /* Reprogram the clock event device. if enabled */ | |
863 | if (reprogram && hrtimer_hres_active()) | |
864 | hrtimer_force_reprogram(base->cpu_base); | |
865 | } | |
866 | rb_erase(&timer->node, &base->active); | |
867 | } | |
303e967f | 868 | timer->state = newstate; |
c0a31329 TG |
869 | } |
870 | ||
871 | /* | |
872 | * remove hrtimer, called with base lock held | |
873 | */ | |
874 | static inline int | |
3c8aa39d | 875 | remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base) |
c0a31329 | 876 | { |
303e967f | 877 | if (hrtimer_is_queued(timer)) { |
54cdfdb4 TG |
878 | int reprogram; |
879 | ||
880 | /* | |
881 | * Remove the timer and force reprogramming when high | |
882 | * resolution mode is active and the timer is on the current | |
883 | * CPU. If we remove a timer on another CPU, reprogramming is | |
884 | * skipped. The interrupt event on this CPU is fired and | |
885 | * reprogramming happens in the interrupt handler. This is a | |
886 | * rare case and less expensive than a smp call. | |
887 | */ | |
237fc6e7 | 888 | debug_hrtimer_deactivate(timer); |
82f67cd9 | 889 | timer_stats_hrtimer_clear_start_info(timer); |
54cdfdb4 TG |
890 | reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases); |
891 | __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE, | |
892 | reprogram); | |
c0a31329 TG |
893 | return 1; |
894 | } | |
895 | return 0; | |
896 | } | |
897 | ||
898 | /** | |
e1dd7bc5 | 899 | * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU |
c0a31329 TG |
900 | * @timer: the timer to be added |
901 | * @tim: expiry time | |
da8f2e17 | 902 | * @delta_ns: "slack" range for the timer |
c0a31329 TG |
903 | * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL) |
904 | * | |
905 | * Returns: | |
906 | * 0 on success | |
907 | * 1 when the timer was active | |
908 | */ | |
909 | int | |
da8f2e17 AV |
910 | hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, unsigned long delta_ns, |
911 | const enum hrtimer_mode mode) | |
c0a31329 | 912 | { |
3c8aa39d | 913 | struct hrtimer_clock_base *base, *new_base; |
c0a31329 | 914 | unsigned long flags; |
ca109491 | 915 | int ret; |
c0a31329 TG |
916 | |
917 | base = lock_hrtimer_base(timer, &flags); | |
918 | ||
919 | /* Remove an active timer from the queue: */ | |
920 | ret = remove_hrtimer(timer, base); | |
921 | ||
922 | /* Switch the timer base, if necessary: */ | |
923 | new_base = switch_hrtimer_base(timer, base); | |
924 | ||
c9cb2e3d | 925 | if (mode == HRTIMER_MODE_REL) { |
5a7780e7 | 926 | tim = ktime_add_safe(tim, new_base->get_time()); |
06027bdd IM |
927 | /* |
928 | * CONFIG_TIME_LOW_RES is a temporary way for architectures | |
929 | * to signal that they simply return xtime in | |
930 | * do_gettimeoffset(). In this case we want to round up by | |
931 | * resolution when starting a relative timer, to avoid short | |
932 | * timeouts. This will go away with the GTOD framework. | |
933 | */ | |
934 | #ifdef CONFIG_TIME_LOW_RES | |
5a7780e7 | 935 | tim = ktime_add_safe(tim, base->resolution); |
06027bdd IM |
936 | #endif |
937 | } | |
237fc6e7 | 938 | |
da8f2e17 | 939 | hrtimer_set_expires_range_ns(timer, tim, delta_ns); |
c0a31329 | 940 | |
82f67cd9 IM |
941 | timer_stats_hrtimer_set_start_info(timer); |
942 | ||
935c631d IM |
943 | /* |
944 | * Only allow reprogramming if the new base is on this CPU. | |
945 | * (it might still be on another CPU if the timer was pending) | |
946 | */ | |
947 | enqueue_hrtimer(timer, new_base, | |
948 | new_base->cpu_base == &__get_cpu_var(hrtimer_bases)); | |
c0a31329 TG |
949 | |
950 | unlock_hrtimer_base(timer, &flags); | |
951 | ||
952 | return ret; | |
953 | } | |
da8f2e17 AV |
954 | EXPORT_SYMBOL_GPL(hrtimer_start_range_ns); |
955 | ||
956 | /** | |
e1dd7bc5 | 957 | * hrtimer_start - (re)start an hrtimer on the current CPU |
da8f2e17 AV |
958 | * @timer: the timer to be added |
959 | * @tim: expiry time | |
960 | * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL) | |
961 | * | |
962 | * Returns: | |
963 | * 0 on success | |
964 | * 1 when the timer was active | |
965 | */ | |
966 | int | |
967 | hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode) | |
968 | { | |
969 | return hrtimer_start_range_ns(timer, tim, 0, mode); | |
970 | } | |
8d16b764 | 971 | EXPORT_SYMBOL_GPL(hrtimer_start); |
c0a31329 | 972 | |
da8f2e17 | 973 | |
c0a31329 TG |
974 | /** |
975 | * hrtimer_try_to_cancel - try to deactivate a timer | |
c0a31329 TG |
976 | * @timer: hrtimer to stop |
977 | * | |
978 | * Returns: | |
979 | * 0 when the timer was not active | |
980 | * 1 when the timer was active | |
981 | * -1 when the timer is currently excuting the callback function and | |
fa9799e3 | 982 | * cannot be stopped |
c0a31329 TG |
983 | */ |
984 | int hrtimer_try_to_cancel(struct hrtimer *timer) | |
985 | { | |
3c8aa39d | 986 | struct hrtimer_clock_base *base; |
c0a31329 TG |
987 | unsigned long flags; |
988 | int ret = -1; | |
989 | ||
990 | base = lock_hrtimer_base(timer, &flags); | |
991 | ||
303e967f | 992 | if (!hrtimer_callback_running(timer)) |
c0a31329 TG |
993 | ret = remove_hrtimer(timer, base); |
994 | ||
995 | unlock_hrtimer_base(timer, &flags); | |
996 | ||
997 | return ret; | |
998 | ||
999 | } | |
8d16b764 | 1000 | EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel); |
c0a31329 TG |
1001 | |
1002 | /** | |
1003 | * hrtimer_cancel - cancel a timer and wait for the handler to finish. | |
c0a31329 TG |
1004 | * @timer: the timer to be cancelled |
1005 | * | |
1006 | * Returns: | |
1007 | * 0 when the timer was not active | |
1008 | * 1 when the timer was active | |
1009 | */ | |
1010 | int hrtimer_cancel(struct hrtimer *timer) | |
1011 | { | |
1012 | for (;;) { | |
1013 | int ret = hrtimer_try_to_cancel(timer); | |
1014 | ||
1015 | if (ret >= 0) | |
1016 | return ret; | |
5ef37b19 | 1017 | cpu_relax(); |
c0a31329 TG |
1018 | } |
1019 | } | |
8d16b764 | 1020 | EXPORT_SYMBOL_GPL(hrtimer_cancel); |
c0a31329 TG |
1021 | |
1022 | /** | |
1023 | * hrtimer_get_remaining - get remaining time for the timer | |
c0a31329 TG |
1024 | * @timer: the timer to read |
1025 | */ | |
1026 | ktime_t hrtimer_get_remaining(const struct hrtimer *timer) | |
1027 | { | |
3c8aa39d | 1028 | struct hrtimer_clock_base *base; |
c0a31329 TG |
1029 | unsigned long flags; |
1030 | ktime_t rem; | |
1031 | ||
1032 | base = lock_hrtimer_base(timer, &flags); | |
cc584b21 | 1033 | rem = hrtimer_expires_remaining(timer); |
c0a31329 TG |
1034 | unlock_hrtimer_base(timer, &flags); |
1035 | ||
1036 | return rem; | |
1037 | } | |
8d16b764 | 1038 | EXPORT_SYMBOL_GPL(hrtimer_get_remaining); |
c0a31329 | 1039 | |
ee9c5785 | 1040 | #ifdef CONFIG_NO_HZ |
69239749 TL |
1041 | /** |
1042 | * hrtimer_get_next_event - get the time until next expiry event | |
1043 | * | |
1044 | * Returns the delta to the next expiry event or KTIME_MAX if no timer | |
1045 | * is pending. | |
1046 | */ | |
1047 | ktime_t hrtimer_get_next_event(void) | |
1048 | { | |
3c8aa39d TG |
1049 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
1050 | struct hrtimer_clock_base *base = cpu_base->clock_base; | |
69239749 TL |
1051 | ktime_t delta, mindelta = { .tv64 = KTIME_MAX }; |
1052 | unsigned long flags; | |
1053 | int i; | |
1054 | ||
3c8aa39d TG |
1055 | spin_lock_irqsave(&cpu_base->lock, flags); |
1056 | ||
54cdfdb4 TG |
1057 | if (!hrtimer_hres_active()) { |
1058 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) { | |
1059 | struct hrtimer *timer; | |
69239749 | 1060 | |
54cdfdb4 TG |
1061 | if (!base->first) |
1062 | continue; | |
3c8aa39d | 1063 | |
54cdfdb4 | 1064 | timer = rb_entry(base->first, struct hrtimer, node); |
cc584b21 | 1065 | delta.tv64 = hrtimer_get_expires_tv64(timer); |
54cdfdb4 TG |
1066 | delta = ktime_sub(delta, base->get_time()); |
1067 | if (delta.tv64 < mindelta.tv64) | |
1068 | mindelta.tv64 = delta.tv64; | |
1069 | } | |
69239749 | 1070 | } |
3c8aa39d TG |
1071 | |
1072 | spin_unlock_irqrestore(&cpu_base->lock, flags); | |
1073 | ||
69239749 TL |
1074 | if (mindelta.tv64 < 0) |
1075 | mindelta.tv64 = 0; | |
1076 | return mindelta; | |
1077 | } | |
1078 | #endif | |
1079 | ||
237fc6e7 TG |
1080 | static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id, |
1081 | enum hrtimer_mode mode) | |
c0a31329 | 1082 | { |
3c8aa39d | 1083 | struct hrtimer_cpu_base *cpu_base; |
c0a31329 | 1084 | |
7978672c GA |
1085 | memset(timer, 0, sizeof(struct hrtimer)); |
1086 | ||
3c8aa39d | 1087 | cpu_base = &__raw_get_cpu_var(hrtimer_bases); |
c0a31329 | 1088 | |
c9cb2e3d | 1089 | if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS) |
7978672c GA |
1090 | clock_id = CLOCK_MONOTONIC; |
1091 | ||
3c8aa39d | 1092 | timer->base = &cpu_base->clock_base[clock_id]; |
d3d74453 | 1093 | INIT_LIST_HEAD(&timer->cb_entry); |
54cdfdb4 | 1094 | hrtimer_init_timer_hres(timer); |
82f67cd9 IM |
1095 | |
1096 | #ifdef CONFIG_TIMER_STATS | |
1097 | timer->start_site = NULL; | |
1098 | timer->start_pid = -1; | |
1099 | memset(timer->start_comm, 0, TASK_COMM_LEN); | |
1100 | #endif | |
c0a31329 | 1101 | } |
237fc6e7 TG |
1102 | |
1103 | /** | |
1104 | * hrtimer_init - initialize a timer to the given clock | |
1105 | * @timer: the timer to be initialized | |
1106 | * @clock_id: the clock to be used | |
1107 | * @mode: timer mode abs/rel | |
1108 | */ | |
1109 | void hrtimer_init(struct hrtimer *timer, clockid_t clock_id, | |
1110 | enum hrtimer_mode mode) | |
1111 | { | |
1112 | debug_hrtimer_init(timer); | |
1113 | __hrtimer_init(timer, clock_id, mode); | |
1114 | } | |
8d16b764 | 1115 | EXPORT_SYMBOL_GPL(hrtimer_init); |
c0a31329 TG |
1116 | |
1117 | /** | |
1118 | * hrtimer_get_res - get the timer resolution for a clock | |
c0a31329 TG |
1119 | * @which_clock: which clock to query |
1120 | * @tp: pointer to timespec variable to store the resolution | |
1121 | * | |
72fd4a35 RD |
1122 | * Store the resolution of the clock selected by @which_clock in the |
1123 | * variable pointed to by @tp. | |
c0a31329 TG |
1124 | */ |
1125 | int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp) | |
1126 | { | |
3c8aa39d | 1127 | struct hrtimer_cpu_base *cpu_base; |
c0a31329 | 1128 | |
3c8aa39d TG |
1129 | cpu_base = &__raw_get_cpu_var(hrtimer_bases); |
1130 | *tp = ktime_to_timespec(cpu_base->clock_base[which_clock].resolution); | |
c0a31329 TG |
1131 | |
1132 | return 0; | |
1133 | } | |
8d16b764 | 1134 | EXPORT_SYMBOL_GPL(hrtimer_get_res); |
c0a31329 | 1135 | |
d3d74453 PZ |
1136 | static void __run_hrtimer(struct hrtimer *timer) |
1137 | { | |
1138 | struct hrtimer_clock_base *base = timer->base; | |
1139 | struct hrtimer_cpu_base *cpu_base = base->cpu_base; | |
1140 | enum hrtimer_restart (*fn)(struct hrtimer *); | |
1141 | int restart; | |
1142 | ||
ca109491 PZ |
1143 | WARN_ON(!irqs_disabled()); |
1144 | ||
237fc6e7 | 1145 | debug_hrtimer_deactivate(timer); |
d3d74453 PZ |
1146 | __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0); |
1147 | timer_stats_account_hrtimer(timer); | |
d3d74453 | 1148 | fn = timer->function; |
ca109491 PZ |
1149 | |
1150 | /* | |
1151 | * Because we run timers from hardirq context, there is no chance | |
1152 | * they get migrated to another cpu, therefore its safe to unlock | |
1153 | * the timer base. | |
1154 | */ | |
1155 | spin_unlock(&cpu_base->lock); | |
1156 | restart = fn(timer); | |
1157 | spin_lock(&cpu_base->lock); | |
d3d74453 PZ |
1158 | |
1159 | /* | |
1160 | * Note: We clear the CALLBACK bit after enqueue_hrtimer to avoid | |
1161 | * reprogramming of the event hardware. This happens at the end of this | |
1162 | * function anyway. | |
1163 | */ | |
1164 | if (restart != HRTIMER_NORESTART) { | |
1165 | BUG_ON(timer->state != HRTIMER_STATE_CALLBACK); | |
1166 | enqueue_hrtimer(timer, base, 0); | |
1167 | } | |
1168 | timer->state &= ~HRTIMER_STATE_CALLBACK; | |
1169 | } | |
1170 | ||
54cdfdb4 TG |
1171 | #ifdef CONFIG_HIGH_RES_TIMERS |
1172 | ||
1173 | /* | |
1174 | * High resolution timer interrupt | |
1175 | * Called with interrupts disabled | |
1176 | */ | |
1177 | void hrtimer_interrupt(struct clock_event_device *dev) | |
1178 | { | |
1179 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); | |
1180 | struct hrtimer_clock_base *base; | |
1181 | ktime_t expires_next, now; | |
ca109491 | 1182 | int i; |
54cdfdb4 TG |
1183 | |
1184 | BUG_ON(!cpu_base->hres_active); | |
1185 | cpu_base->nr_events++; | |
1186 | dev->next_event.tv64 = KTIME_MAX; | |
1187 | ||
1188 | retry: | |
1189 | now = ktime_get(); | |
1190 | ||
1191 | expires_next.tv64 = KTIME_MAX; | |
1192 | ||
1193 | base = cpu_base->clock_base; | |
1194 | ||
1195 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { | |
1196 | ktime_t basenow; | |
1197 | struct rb_node *node; | |
1198 | ||
1199 | spin_lock(&cpu_base->lock); | |
1200 | ||
1201 | basenow = ktime_add(now, base->offset); | |
1202 | ||
1203 | while ((node = base->first)) { | |
1204 | struct hrtimer *timer; | |
1205 | ||
1206 | timer = rb_entry(node, struct hrtimer, node); | |
1207 | ||
654c8e0b AV |
1208 | /* |
1209 | * The immediate goal for using the softexpires is | |
1210 | * minimizing wakeups, not running timers at the | |
1211 | * earliest interrupt after their soft expiration. | |
1212 | * This allows us to avoid using a Priority Search | |
1213 | * Tree, which can answer a stabbing querry for | |
1214 | * overlapping intervals and instead use the simple | |
1215 | * BST we already have. | |
1216 | * We don't add extra wakeups by delaying timers that | |
1217 | * are right-of a not yet expired timer, because that | |
1218 | * timer will have to trigger a wakeup anyway. | |
1219 | */ | |
1220 | ||
1221 | if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer)) { | |
54cdfdb4 TG |
1222 | ktime_t expires; |
1223 | ||
cc584b21 | 1224 | expires = ktime_sub(hrtimer_get_expires(timer), |
54cdfdb4 TG |
1225 | base->offset); |
1226 | if (expires.tv64 < expires_next.tv64) | |
1227 | expires_next = expires; | |
1228 | break; | |
1229 | } | |
1230 | ||
d3d74453 | 1231 | __run_hrtimer(timer); |
54cdfdb4 TG |
1232 | } |
1233 | spin_unlock(&cpu_base->lock); | |
1234 | base++; | |
1235 | } | |
1236 | ||
1237 | cpu_base->expires_next = expires_next; | |
1238 | ||
1239 | /* Reprogramming necessary ? */ | |
1240 | if (expires_next.tv64 != KTIME_MAX) { | |
1241 | if (tick_program_event(expires_next, 0)) | |
1242 | goto retry; | |
1243 | } | |
54cdfdb4 TG |
1244 | } |
1245 | ||
8bdec955 TG |
1246 | /* |
1247 | * local version of hrtimer_peek_ahead_timers() called with interrupts | |
1248 | * disabled. | |
1249 | */ | |
1250 | static void __hrtimer_peek_ahead_timers(void) | |
1251 | { | |
1252 | struct tick_device *td; | |
1253 | ||
1254 | if (!hrtimer_hres_active()) | |
1255 | return; | |
1256 | ||
1257 | td = &__get_cpu_var(tick_cpu_device); | |
1258 | if (td && td->evtdev) | |
1259 | hrtimer_interrupt(td->evtdev); | |
1260 | } | |
1261 | ||
2e94d1f7 AV |
1262 | /** |
1263 | * hrtimer_peek_ahead_timers -- run soft-expired timers now | |
1264 | * | |
1265 | * hrtimer_peek_ahead_timers will peek at the timer queue of | |
1266 | * the current cpu and check if there are any timers for which | |
1267 | * the soft expires time has passed. If any such timers exist, | |
1268 | * they are run immediately and then removed from the timer queue. | |
1269 | * | |
1270 | */ | |
1271 | void hrtimer_peek_ahead_timers(void) | |
1272 | { | |
643bdf68 | 1273 | unsigned long flags; |
dc4304f7 | 1274 | |
2e94d1f7 | 1275 | local_irq_save(flags); |
8bdec955 | 1276 | __hrtimer_peek_ahead_timers(); |
2e94d1f7 AV |
1277 | local_irq_restore(flags); |
1278 | } | |
1279 | ||
d3d74453 | 1280 | #endif /* CONFIG_HIGH_RES_TIMERS */ |
82f67cd9 | 1281 | |
d3d74453 PZ |
1282 | /* |
1283 | * Called from timer softirq every jiffy, expire hrtimers: | |
1284 | * | |
1285 | * For HRT its the fall back code to run the softirq in the timer | |
1286 | * softirq context in case the hrtimer initialization failed or has | |
1287 | * not been done yet. | |
1288 | */ | |
1289 | void hrtimer_run_pending(void) | |
1290 | { | |
d3d74453 PZ |
1291 | if (hrtimer_hres_active()) |
1292 | return; | |
54cdfdb4 | 1293 | |
d3d74453 PZ |
1294 | /* |
1295 | * This _is_ ugly: We have to check in the softirq context, | |
1296 | * whether we can switch to highres and / or nohz mode. The | |
1297 | * clocksource switch happens in the timer interrupt with | |
1298 | * xtime_lock held. Notification from there only sets the | |
1299 | * check bit in the tick_oneshot code, otherwise we might | |
1300 | * deadlock vs. xtime_lock. | |
1301 | */ | |
1302 | if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) | |
1303 | hrtimer_switch_to_hres(); | |
54cdfdb4 TG |
1304 | } |
1305 | ||
c0a31329 | 1306 | /* |
d3d74453 | 1307 | * Called from hardirq context every jiffy |
c0a31329 | 1308 | */ |
833883d9 | 1309 | void hrtimer_run_queues(void) |
c0a31329 | 1310 | { |
288867ec | 1311 | struct rb_node *node; |
833883d9 DS |
1312 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
1313 | struct hrtimer_clock_base *base; | |
1314 | int index, gettime = 1; | |
c0a31329 | 1315 | |
833883d9 | 1316 | if (hrtimer_hres_active()) |
3055adda DS |
1317 | return; |
1318 | ||
833883d9 DS |
1319 | for (index = 0; index < HRTIMER_MAX_CLOCK_BASES; index++) { |
1320 | base = &cpu_base->clock_base[index]; | |
c0a31329 | 1321 | |
833883d9 | 1322 | if (!base->first) |
d3d74453 | 1323 | continue; |
833883d9 | 1324 | |
d7cfb60c | 1325 | if (gettime) { |
833883d9 DS |
1326 | hrtimer_get_softirq_time(cpu_base); |
1327 | gettime = 0; | |
b75f7a51 | 1328 | } |
d3d74453 | 1329 | |
833883d9 | 1330 | spin_lock(&cpu_base->lock); |
c0a31329 | 1331 | |
833883d9 DS |
1332 | while ((node = base->first)) { |
1333 | struct hrtimer *timer; | |
54cdfdb4 | 1334 | |
833883d9 | 1335 | timer = rb_entry(node, struct hrtimer, node); |
cc584b21 AV |
1336 | if (base->softirq_time.tv64 <= |
1337 | hrtimer_get_expires_tv64(timer)) | |
833883d9 DS |
1338 | break; |
1339 | ||
833883d9 DS |
1340 | __run_hrtimer(timer); |
1341 | } | |
1342 | spin_unlock(&cpu_base->lock); | |
1343 | } | |
c0a31329 TG |
1344 | } |
1345 | ||
10c94ec1 TG |
1346 | /* |
1347 | * Sleep related functions: | |
1348 | */ | |
c9cb2e3d | 1349 | static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer) |
00362e33 TG |
1350 | { |
1351 | struct hrtimer_sleeper *t = | |
1352 | container_of(timer, struct hrtimer_sleeper, timer); | |
1353 | struct task_struct *task = t->task; | |
1354 | ||
1355 | t->task = NULL; | |
1356 | if (task) | |
1357 | wake_up_process(task); | |
1358 | ||
1359 | return HRTIMER_NORESTART; | |
1360 | } | |
1361 | ||
36c8b586 | 1362 | void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task) |
00362e33 TG |
1363 | { |
1364 | sl->timer.function = hrtimer_wakeup; | |
1365 | sl->task = task; | |
1366 | } | |
1367 | ||
669d7868 | 1368 | static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode) |
432569bb | 1369 | { |
669d7868 | 1370 | hrtimer_init_sleeper(t, current); |
10c94ec1 | 1371 | |
432569bb RZ |
1372 | do { |
1373 | set_current_state(TASK_INTERRUPTIBLE); | |
cc584b21 | 1374 | hrtimer_start_expires(&t->timer, mode); |
37bb6cb4 PZ |
1375 | if (!hrtimer_active(&t->timer)) |
1376 | t->task = NULL; | |
432569bb | 1377 | |
54cdfdb4 TG |
1378 | if (likely(t->task)) |
1379 | schedule(); | |
432569bb | 1380 | |
669d7868 | 1381 | hrtimer_cancel(&t->timer); |
c9cb2e3d | 1382 | mode = HRTIMER_MODE_ABS; |
669d7868 TG |
1383 | |
1384 | } while (t->task && !signal_pending(current)); | |
432569bb | 1385 | |
3588a085 PZ |
1386 | __set_current_state(TASK_RUNNING); |
1387 | ||
669d7868 | 1388 | return t->task == NULL; |
10c94ec1 TG |
1389 | } |
1390 | ||
080344b9 ON |
1391 | static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp) |
1392 | { | |
1393 | struct timespec rmt; | |
1394 | ktime_t rem; | |
1395 | ||
cc584b21 | 1396 | rem = hrtimer_expires_remaining(timer); |
080344b9 ON |
1397 | if (rem.tv64 <= 0) |
1398 | return 0; | |
1399 | rmt = ktime_to_timespec(rem); | |
1400 | ||
1401 | if (copy_to_user(rmtp, &rmt, sizeof(*rmtp))) | |
1402 | return -EFAULT; | |
1403 | ||
1404 | return 1; | |
1405 | } | |
1406 | ||
1711ef38 | 1407 | long __sched hrtimer_nanosleep_restart(struct restart_block *restart) |
10c94ec1 | 1408 | { |
669d7868 | 1409 | struct hrtimer_sleeper t; |
080344b9 | 1410 | struct timespec __user *rmtp; |
237fc6e7 | 1411 | int ret = 0; |
10c94ec1 | 1412 | |
237fc6e7 TG |
1413 | hrtimer_init_on_stack(&t.timer, restart->nanosleep.index, |
1414 | HRTIMER_MODE_ABS); | |
cc584b21 | 1415 | hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires); |
10c94ec1 | 1416 | |
c9cb2e3d | 1417 | if (do_nanosleep(&t, HRTIMER_MODE_ABS)) |
237fc6e7 | 1418 | goto out; |
10c94ec1 | 1419 | |
029a07e0 | 1420 | rmtp = restart->nanosleep.rmtp; |
432569bb | 1421 | if (rmtp) { |
237fc6e7 | 1422 | ret = update_rmtp(&t.timer, rmtp); |
080344b9 | 1423 | if (ret <= 0) |
237fc6e7 | 1424 | goto out; |
432569bb | 1425 | } |
10c94ec1 | 1426 | |
10c94ec1 | 1427 | /* The other values in restart are already filled in */ |
237fc6e7 TG |
1428 | ret = -ERESTART_RESTARTBLOCK; |
1429 | out: | |
1430 | destroy_hrtimer_on_stack(&t.timer); | |
1431 | return ret; | |
10c94ec1 TG |
1432 | } |
1433 | ||
080344b9 | 1434 | long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp, |
10c94ec1 TG |
1435 | const enum hrtimer_mode mode, const clockid_t clockid) |
1436 | { | |
1437 | struct restart_block *restart; | |
669d7868 | 1438 | struct hrtimer_sleeper t; |
237fc6e7 | 1439 | int ret = 0; |
3bd01206 AV |
1440 | unsigned long slack; |
1441 | ||
1442 | slack = current->timer_slack_ns; | |
1443 | if (rt_task(current)) | |
1444 | slack = 0; | |
10c94ec1 | 1445 | |
237fc6e7 | 1446 | hrtimer_init_on_stack(&t.timer, clockid, mode); |
3bd01206 | 1447 | hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack); |
432569bb | 1448 | if (do_nanosleep(&t, mode)) |
237fc6e7 | 1449 | goto out; |
10c94ec1 | 1450 | |
7978672c | 1451 | /* Absolute timers do not update the rmtp value and restart: */ |
237fc6e7 TG |
1452 | if (mode == HRTIMER_MODE_ABS) { |
1453 | ret = -ERESTARTNOHAND; | |
1454 | goto out; | |
1455 | } | |
10c94ec1 | 1456 | |
432569bb | 1457 | if (rmtp) { |
237fc6e7 | 1458 | ret = update_rmtp(&t.timer, rmtp); |
080344b9 | 1459 | if (ret <= 0) |
237fc6e7 | 1460 | goto out; |
432569bb | 1461 | } |
10c94ec1 TG |
1462 | |
1463 | restart = ¤t_thread_info()->restart_block; | |
1711ef38 | 1464 | restart->fn = hrtimer_nanosleep_restart; |
029a07e0 TG |
1465 | restart->nanosleep.index = t.timer.base->index; |
1466 | restart->nanosleep.rmtp = rmtp; | |
cc584b21 | 1467 | restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer); |
10c94ec1 | 1468 | |
237fc6e7 TG |
1469 | ret = -ERESTART_RESTARTBLOCK; |
1470 | out: | |
1471 | destroy_hrtimer_on_stack(&t.timer); | |
1472 | return ret; | |
10c94ec1 TG |
1473 | } |
1474 | ||
6ba1b912 TG |
1475 | asmlinkage long |
1476 | sys_nanosleep(struct timespec __user *rqtp, struct timespec __user *rmtp) | |
1477 | { | |
080344b9 | 1478 | struct timespec tu; |
6ba1b912 TG |
1479 | |
1480 | if (copy_from_user(&tu, rqtp, sizeof(tu))) | |
1481 | return -EFAULT; | |
1482 | ||
1483 | if (!timespec_valid(&tu)) | |
1484 | return -EINVAL; | |
1485 | ||
080344b9 | 1486 | return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC); |
6ba1b912 TG |
1487 | } |
1488 | ||
c0a31329 TG |
1489 | /* |
1490 | * Functions related to boot-time initialization: | |
1491 | */ | |
0ec160dd | 1492 | static void __cpuinit init_hrtimers_cpu(int cpu) |
c0a31329 | 1493 | { |
3c8aa39d | 1494 | struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu); |
c0a31329 TG |
1495 | int i; |
1496 | ||
3c8aa39d | 1497 | spin_lock_init(&cpu_base->lock); |
3c8aa39d TG |
1498 | |
1499 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) | |
1500 | cpu_base->clock_base[i].cpu_base = cpu_base; | |
1501 | ||
54cdfdb4 | 1502 | hrtimer_init_hres(cpu_base); |
c0a31329 TG |
1503 | } |
1504 | ||
1505 | #ifdef CONFIG_HOTPLUG_CPU | |
1506 | ||
ca109491 | 1507 | static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base, |
37810659 | 1508 | struct hrtimer_clock_base *new_base) |
c0a31329 TG |
1509 | { |
1510 | struct hrtimer *timer; | |
1511 | struct rb_node *node; | |
1512 | ||
1513 | while ((node = rb_first(&old_base->active))) { | |
1514 | timer = rb_entry(node, struct hrtimer, node); | |
54cdfdb4 | 1515 | BUG_ON(hrtimer_callback_running(timer)); |
237fc6e7 | 1516 | debug_hrtimer_deactivate(timer); |
b00c1a99 TG |
1517 | |
1518 | /* | |
1519 | * Mark it as STATE_MIGRATE not INACTIVE otherwise the | |
1520 | * timer could be seen as !active and just vanish away | |
1521 | * under us on another CPU | |
1522 | */ | |
1523 | __remove_hrtimer(timer, old_base, HRTIMER_STATE_MIGRATE, 0); | |
c0a31329 | 1524 | timer->base = new_base; |
54cdfdb4 | 1525 | /* |
37810659 PZ |
1526 | * Enqueue the timers on the new cpu, but do not reprogram |
1527 | * the timer as that would enable a deadlock between | |
1528 | * hrtimer_enqueue_reprogramm() running the timer and us still | |
1529 | * holding a nested base lock. | |
1530 | * | |
1531 | * Instead we tickle the hrtimer interrupt after the migration | |
1532 | * is done, which will run all expired timers and re-programm | |
1533 | * the timer device. | |
54cdfdb4 | 1534 | */ |
37810659 | 1535 | enqueue_hrtimer(timer, new_base, 0); |
41e1022e | 1536 | |
b00c1a99 TG |
1537 | /* Clear the migration state bit */ |
1538 | timer->state &= ~HRTIMER_STATE_MIGRATE; | |
c0a31329 TG |
1539 | } |
1540 | } | |
1541 | ||
d5fd43c4 | 1542 | static void migrate_hrtimers(int scpu) |
c0a31329 | 1543 | { |
3c8aa39d | 1544 | struct hrtimer_cpu_base *old_base, *new_base; |
731a55ba | 1545 | int i; |
c0a31329 | 1546 | |
37810659 | 1547 | BUG_ON(cpu_online(scpu)); |
37810659 | 1548 | tick_cancel_sched_timer(scpu); |
731a55ba TG |
1549 | |
1550 | local_irq_disable(); | |
1551 | old_base = &per_cpu(hrtimer_bases, scpu); | |
1552 | new_base = &__get_cpu_var(hrtimer_bases); | |
d82f0b0f ON |
1553 | /* |
1554 | * The caller is globally serialized and nobody else | |
1555 | * takes two locks at once, deadlock is not possible. | |
1556 | */ | |
731a55ba | 1557 | spin_lock(&new_base->lock); |
8e60e05f | 1558 | spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING); |
c0a31329 | 1559 | |
3c8aa39d | 1560 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
ca109491 | 1561 | migrate_hrtimer_list(&old_base->clock_base[i], |
37810659 | 1562 | &new_base->clock_base[i]); |
c0a31329 TG |
1563 | } |
1564 | ||
8e60e05f | 1565 | spin_unlock(&old_base->lock); |
731a55ba | 1566 | spin_unlock(&new_base->lock); |
37810659 | 1567 | |
731a55ba TG |
1568 | /* Check, if we got expired work to do */ |
1569 | __hrtimer_peek_ahead_timers(); | |
1570 | local_irq_enable(); | |
c0a31329 | 1571 | } |
37810659 | 1572 | |
c0a31329 TG |
1573 | #endif /* CONFIG_HOTPLUG_CPU */ |
1574 | ||
8c78f307 | 1575 | static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self, |
c0a31329 TG |
1576 | unsigned long action, void *hcpu) |
1577 | { | |
b2e3c0ad | 1578 | int scpu = (long)hcpu; |
c0a31329 TG |
1579 | |
1580 | switch (action) { | |
1581 | ||
1582 | case CPU_UP_PREPARE: | |
8bb78442 | 1583 | case CPU_UP_PREPARE_FROZEN: |
37810659 | 1584 | init_hrtimers_cpu(scpu); |
c0a31329 TG |
1585 | break; |
1586 | ||
1587 | #ifdef CONFIG_HOTPLUG_CPU | |
1588 | case CPU_DEAD: | |
8bb78442 | 1589 | case CPU_DEAD_FROZEN: |
b2e3c0ad | 1590 | { |
37810659 | 1591 | clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &scpu); |
d5fd43c4 | 1592 | migrate_hrtimers(scpu); |
c0a31329 | 1593 | break; |
b2e3c0ad | 1594 | } |
c0a31329 TG |
1595 | #endif |
1596 | ||
1597 | default: | |
1598 | break; | |
1599 | } | |
1600 | ||
1601 | return NOTIFY_OK; | |
1602 | } | |
1603 | ||
8c78f307 | 1604 | static struct notifier_block __cpuinitdata hrtimers_nb = { |
c0a31329 TG |
1605 | .notifier_call = hrtimer_cpu_notify, |
1606 | }; | |
1607 | ||
1608 | void __init hrtimers_init(void) | |
1609 | { | |
1610 | hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE, | |
1611 | (void *)(long)smp_processor_id()); | |
1612 | register_cpu_notifier(&hrtimers_nb); | |
1613 | } | |
1614 | ||
7bb67439 | 1615 | /** |
654c8e0b | 1616 | * schedule_hrtimeout_range - sleep until timeout |
7bb67439 | 1617 | * @expires: timeout value (ktime_t) |
654c8e0b | 1618 | * @delta: slack in expires timeout (ktime_t) |
7bb67439 AV |
1619 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL |
1620 | * | |
1621 | * Make the current task sleep until the given expiry time has | |
1622 | * elapsed. The routine will return immediately unless | |
1623 | * the current task state has been set (see set_current_state()). | |
1624 | * | |
654c8e0b AV |
1625 | * The @delta argument gives the kernel the freedom to schedule the |
1626 | * actual wakeup to a time that is both power and performance friendly. | |
1627 | * The kernel give the normal best effort behavior for "@expires+@delta", | |
1628 | * but may decide to fire the timer earlier, but no earlier than @expires. | |
1629 | * | |
7bb67439 AV |
1630 | * You can set the task state as follows - |
1631 | * | |
1632 | * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to | |
1633 | * pass before the routine returns. | |
1634 | * | |
1635 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
1636 | * delivered to the current task. | |
1637 | * | |
1638 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1639 | * routine returns. | |
1640 | * | |
1641 | * Returns 0 when the timer has expired otherwise -EINTR | |
1642 | */ | |
654c8e0b | 1643 | int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta, |
7bb67439 AV |
1644 | const enum hrtimer_mode mode) |
1645 | { | |
1646 | struct hrtimer_sleeper t; | |
1647 | ||
1648 | /* | |
1649 | * Optimize when a zero timeout value is given. It does not | |
1650 | * matter whether this is an absolute or a relative time. | |
1651 | */ | |
1652 | if (expires && !expires->tv64) { | |
1653 | __set_current_state(TASK_RUNNING); | |
1654 | return 0; | |
1655 | } | |
1656 | ||
1657 | /* | |
1658 | * A NULL parameter means "inifinte" | |
1659 | */ | |
1660 | if (!expires) { | |
1661 | schedule(); | |
1662 | __set_current_state(TASK_RUNNING); | |
1663 | return -EINTR; | |
1664 | } | |
1665 | ||
1666 | hrtimer_init_on_stack(&t.timer, CLOCK_MONOTONIC, mode); | |
654c8e0b | 1667 | hrtimer_set_expires_range_ns(&t.timer, *expires, delta); |
7bb67439 AV |
1668 | |
1669 | hrtimer_init_sleeper(&t, current); | |
1670 | ||
cc584b21 | 1671 | hrtimer_start_expires(&t.timer, mode); |
7bb67439 AV |
1672 | if (!hrtimer_active(&t.timer)) |
1673 | t.task = NULL; | |
1674 | ||
1675 | if (likely(t.task)) | |
1676 | schedule(); | |
1677 | ||
1678 | hrtimer_cancel(&t.timer); | |
1679 | destroy_hrtimer_on_stack(&t.timer); | |
1680 | ||
1681 | __set_current_state(TASK_RUNNING); | |
1682 | ||
1683 | return !t.task ? 0 : -EINTR; | |
1684 | } | |
654c8e0b AV |
1685 | EXPORT_SYMBOL_GPL(schedule_hrtimeout_range); |
1686 | ||
1687 | /** | |
1688 | * schedule_hrtimeout - sleep until timeout | |
1689 | * @expires: timeout value (ktime_t) | |
1690 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL | |
1691 | * | |
1692 | * Make the current task sleep until the given expiry time has | |
1693 | * elapsed. The routine will return immediately unless | |
1694 | * the current task state has been set (see set_current_state()). | |
1695 | * | |
1696 | * You can set the task state as follows - | |
1697 | * | |
1698 | * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to | |
1699 | * pass before the routine returns. | |
1700 | * | |
1701 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
1702 | * delivered to the current task. | |
1703 | * | |
1704 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1705 | * routine returns. | |
1706 | * | |
1707 | * Returns 0 when the timer has expired otherwise -EINTR | |
1708 | */ | |
1709 | int __sched schedule_hrtimeout(ktime_t *expires, | |
1710 | const enum hrtimer_mode mode) | |
1711 | { | |
1712 | return schedule_hrtimeout_range(expires, 0, mode); | |
1713 | } | |
7bb67439 | 1714 | EXPORT_SYMBOL_GPL(schedule_hrtimeout); |