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> | |
237fc6e7 | 46 | #include <linux/debugobjects.h> |
c0a31329 TG |
47 | |
48 | #include <asm/uaccess.h> | |
49 | ||
50 | /** | |
51 | * ktime_get - get the monotonic time in ktime_t format | |
52 | * | |
53 | * returns the time in ktime_t format | |
54 | */ | |
d316c57f | 55 | ktime_t ktime_get(void) |
c0a31329 TG |
56 | { |
57 | struct timespec now; | |
58 | ||
59 | ktime_get_ts(&now); | |
60 | ||
61 | return timespec_to_ktime(now); | |
62 | } | |
641b9e0e | 63 | EXPORT_SYMBOL_GPL(ktime_get); |
c0a31329 TG |
64 | |
65 | /** | |
66 | * ktime_get_real - get the real (wall-) time in ktime_t format | |
67 | * | |
68 | * returns the time in ktime_t format | |
69 | */ | |
d316c57f | 70 | ktime_t ktime_get_real(void) |
c0a31329 TG |
71 | { |
72 | struct timespec now; | |
73 | ||
74 | getnstimeofday(&now); | |
75 | ||
76 | return timespec_to_ktime(now); | |
77 | } | |
78 | ||
79 | EXPORT_SYMBOL_GPL(ktime_get_real); | |
80 | ||
81 | /* | |
82 | * The timer bases: | |
7978672c GA |
83 | * |
84 | * Note: If we want to add new timer bases, we have to skip the two | |
85 | * clock ids captured by the cpu-timers. We do this by holding empty | |
86 | * entries rather than doing math adjustment of the clock ids. | |
87 | * This ensures that we capture erroneous accesses to these clock ids | |
88 | * rather than moving them into the range of valid clock id's. | |
c0a31329 | 89 | */ |
54cdfdb4 | 90 | DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) = |
c0a31329 | 91 | { |
3c8aa39d TG |
92 | |
93 | .clock_base = | |
c0a31329 | 94 | { |
3c8aa39d TG |
95 | { |
96 | .index = CLOCK_REALTIME, | |
97 | .get_time = &ktime_get_real, | |
54cdfdb4 | 98 | .resolution = KTIME_LOW_RES, |
3c8aa39d TG |
99 | }, |
100 | { | |
101 | .index = CLOCK_MONOTONIC, | |
102 | .get_time = &ktime_get, | |
54cdfdb4 | 103 | .resolution = KTIME_LOW_RES, |
3c8aa39d TG |
104 | }, |
105 | } | |
c0a31329 TG |
106 | }; |
107 | ||
108 | /** | |
109 | * ktime_get_ts - get the monotonic clock in timespec format | |
c0a31329 TG |
110 | * @ts: pointer to timespec variable |
111 | * | |
112 | * The function calculates the monotonic clock from the realtime | |
113 | * clock and the wall_to_monotonic offset and stores the result | |
72fd4a35 | 114 | * in normalized timespec format in the variable pointed to by @ts. |
c0a31329 TG |
115 | */ |
116 | void ktime_get_ts(struct timespec *ts) | |
117 | { | |
118 | struct timespec tomono; | |
119 | unsigned long seq; | |
120 | ||
121 | do { | |
122 | seq = read_seqbegin(&xtime_lock); | |
123 | getnstimeofday(ts); | |
124 | tomono = wall_to_monotonic; | |
125 | ||
126 | } while (read_seqretry(&xtime_lock, seq)); | |
127 | ||
128 | set_normalized_timespec(ts, ts->tv_sec + tomono.tv_sec, | |
129 | ts->tv_nsec + tomono.tv_nsec); | |
130 | } | |
69778e32 | 131 | EXPORT_SYMBOL_GPL(ktime_get_ts); |
c0a31329 | 132 | |
92127c7a TG |
133 | /* |
134 | * Get the coarse grained time at the softirq based on xtime and | |
135 | * wall_to_monotonic. | |
136 | */ | |
3c8aa39d | 137 | static void hrtimer_get_softirq_time(struct hrtimer_cpu_base *base) |
92127c7a TG |
138 | { |
139 | ktime_t xtim, tomono; | |
ad28d94a | 140 | struct timespec xts, tom; |
92127c7a TG |
141 | unsigned long seq; |
142 | ||
143 | do { | |
144 | seq = read_seqbegin(&xtime_lock); | |
2c6b47de | 145 | xts = current_kernel_time(); |
ad28d94a | 146 | tom = wall_to_monotonic; |
92127c7a TG |
147 | } while (read_seqretry(&xtime_lock, seq)); |
148 | ||
f4304ab2 | 149 | xtim = timespec_to_ktime(xts); |
ad28d94a | 150 | tomono = timespec_to_ktime(tom); |
3c8aa39d TG |
151 | base->clock_base[CLOCK_REALTIME].softirq_time = xtim; |
152 | base->clock_base[CLOCK_MONOTONIC].softirq_time = | |
153 | ktime_add(xtim, tomono); | |
92127c7a TG |
154 | } |
155 | ||
c0a31329 TG |
156 | /* |
157 | * Functions and macros which are different for UP/SMP systems are kept in a | |
158 | * single place | |
159 | */ | |
160 | #ifdef CONFIG_SMP | |
161 | ||
c0a31329 TG |
162 | /* |
163 | * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock | |
164 | * means that all timers which are tied to this base via timer->base are | |
165 | * locked, and the base itself is locked too. | |
166 | * | |
167 | * So __run_timers/migrate_timers can safely modify all timers which could | |
168 | * be found on the lists/queues. | |
169 | * | |
170 | * When the timer's base is locked, and the timer removed from list, it is | |
171 | * possible to set timer->base = NULL and drop the lock: the timer remains | |
172 | * locked. | |
173 | */ | |
3c8aa39d TG |
174 | static |
175 | struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer, | |
176 | unsigned long *flags) | |
c0a31329 | 177 | { |
3c8aa39d | 178 | struct hrtimer_clock_base *base; |
c0a31329 TG |
179 | |
180 | for (;;) { | |
181 | base = timer->base; | |
182 | if (likely(base != NULL)) { | |
3c8aa39d | 183 | spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
184 | if (likely(base == timer->base)) |
185 | return base; | |
186 | /* The timer has migrated to another CPU: */ | |
3c8aa39d | 187 | spin_unlock_irqrestore(&base->cpu_base->lock, *flags); |
c0a31329 TG |
188 | } |
189 | cpu_relax(); | |
190 | } | |
191 | } | |
192 | ||
193 | /* | |
194 | * Switch the timer base to the current CPU when possible. | |
195 | */ | |
3c8aa39d TG |
196 | static inline struct hrtimer_clock_base * |
197 | switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base) | |
c0a31329 | 198 | { |
3c8aa39d TG |
199 | struct hrtimer_clock_base *new_base; |
200 | struct hrtimer_cpu_base *new_cpu_base; | |
c0a31329 | 201 | |
3c8aa39d TG |
202 | new_cpu_base = &__get_cpu_var(hrtimer_bases); |
203 | new_base = &new_cpu_base->clock_base[base->index]; | |
c0a31329 TG |
204 | |
205 | if (base != new_base) { | |
206 | /* | |
207 | * We are trying to schedule the timer on the local CPU. | |
208 | * However we can't change timer's base while it is running, | |
209 | * so we keep it on the same CPU. No hassle vs. reprogramming | |
210 | * the event source in the high resolution case. The softirq | |
211 | * code will take care of this when the timer function has | |
212 | * completed. There is no conflict as we hold the lock until | |
213 | * the timer is enqueued. | |
214 | */ | |
54cdfdb4 | 215 | if (unlikely(hrtimer_callback_running(timer))) |
c0a31329 TG |
216 | return base; |
217 | ||
218 | /* See the comment in lock_timer_base() */ | |
219 | timer->base = NULL; | |
3c8aa39d TG |
220 | spin_unlock(&base->cpu_base->lock); |
221 | spin_lock(&new_base->cpu_base->lock); | |
c0a31329 TG |
222 | timer->base = new_base; |
223 | } | |
224 | return new_base; | |
225 | } | |
226 | ||
227 | #else /* CONFIG_SMP */ | |
228 | ||
3c8aa39d | 229 | static inline struct hrtimer_clock_base * |
c0a31329 TG |
230 | lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) |
231 | { | |
3c8aa39d | 232 | struct hrtimer_clock_base *base = timer->base; |
c0a31329 | 233 | |
3c8aa39d | 234 | spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
235 | |
236 | return base; | |
237 | } | |
238 | ||
54cdfdb4 | 239 | # define switch_hrtimer_base(t, b) (b) |
c0a31329 TG |
240 | |
241 | #endif /* !CONFIG_SMP */ | |
242 | ||
243 | /* | |
244 | * Functions for the union type storage format of ktime_t which are | |
245 | * too large for inlining: | |
246 | */ | |
247 | #if BITS_PER_LONG < 64 | |
248 | # ifndef CONFIG_KTIME_SCALAR | |
249 | /** | |
250 | * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable | |
c0a31329 TG |
251 | * @kt: addend |
252 | * @nsec: the scalar nsec value to add | |
253 | * | |
254 | * Returns the sum of kt and nsec in ktime_t format | |
255 | */ | |
256 | ktime_t ktime_add_ns(const ktime_t kt, u64 nsec) | |
257 | { | |
258 | ktime_t tmp; | |
259 | ||
260 | if (likely(nsec < NSEC_PER_SEC)) { | |
261 | tmp.tv64 = nsec; | |
262 | } else { | |
263 | unsigned long rem = do_div(nsec, NSEC_PER_SEC); | |
264 | ||
265 | tmp = ktime_set((long)nsec, rem); | |
266 | } | |
267 | ||
268 | return ktime_add(kt, tmp); | |
269 | } | |
b8b8fd2d DH |
270 | |
271 | EXPORT_SYMBOL_GPL(ktime_add_ns); | |
a272378d ACM |
272 | |
273 | /** | |
274 | * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable | |
275 | * @kt: minuend | |
276 | * @nsec: the scalar nsec value to subtract | |
277 | * | |
278 | * Returns the subtraction of @nsec from @kt in ktime_t format | |
279 | */ | |
280 | ktime_t ktime_sub_ns(const ktime_t kt, u64 nsec) | |
281 | { | |
282 | ktime_t tmp; | |
283 | ||
284 | if (likely(nsec < NSEC_PER_SEC)) { | |
285 | tmp.tv64 = nsec; | |
286 | } else { | |
287 | unsigned long rem = do_div(nsec, NSEC_PER_SEC); | |
288 | ||
289 | tmp = ktime_set((long)nsec, rem); | |
290 | } | |
291 | ||
292 | return ktime_sub(kt, tmp); | |
293 | } | |
294 | ||
295 | EXPORT_SYMBOL_GPL(ktime_sub_ns); | |
c0a31329 TG |
296 | # endif /* !CONFIG_KTIME_SCALAR */ |
297 | ||
298 | /* | |
299 | * Divide a ktime value by a nanosecond value | |
300 | */ | |
4d672e7a | 301 | u64 ktime_divns(const ktime_t kt, s64 div) |
c0a31329 TG |
302 | { |
303 | u64 dclc, inc, dns; | |
304 | int sft = 0; | |
305 | ||
306 | dclc = dns = ktime_to_ns(kt); | |
307 | inc = div; | |
308 | /* Make sure the divisor is less than 2^32: */ | |
309 | while (div >> 32) { | |
310 | sft++; | |
311 | div >>= 1; | |
312 | } | |
313 | dclc >>= sft; | |
314 | do_div(dclc, (unsigned long) div); | |
315 | ||
4d672e7a | 316 | return dclc; |
c0a31329 | 317 | } |
c0a31329 TG |
318 | #endif /* BITS_PER_LONG >= 64 */ |
319 | ||
5a7780e7 TG |
320 | /* |
321 | * Add two ktime values and do a safety check for overflow: | |
322 | */ | |
323 | ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs) | |
324 | { | |
325 | ktime_t res = ktime_add(lhs, rhs); | |
326 | ||
327 | /* | |
328 | * We use KTIME_SEC_MAX here, the maximum timeout which we can | |
329 | * return to user space in a timespec: | |
330 | */ | |
331 | if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64) | |
332 | res = ktime_set(KTIME_SEC_MAX, 0); | |
333 | ||
334 | return res; | |
335 | } | |
336 | ||
237fc6e7 TG |
337 | #ifdef CONFIG_DEBUG_OBJECTS_TIMERS |
338 | ||
339 | static struct debug_obj_descr hrtimer_debug_descr; | |
340 | ||
341 | /* | |
342 | * fixup_init is called when: | |
343 | * - an active object is initialized | |
344 | */ | |
345 | static int hrtimer_fixup_init(void *addr, enum debug_obj_state state) | |
346 | { | |
347 | struct hrtimer *timer = addr; | |
348 | ||
349 | switch (state) { | |
350 | case ODEBUG_STATE_ACTIVE: | |
351 | hrtimer_cancel(timer); | |
352 | debug_object_init(timer, &hrtimer_debug_descr); | |
353 | return 1; | |
354 | default: | |
355 | return 0; | |
356 | } | |
357 | } | |
358 | ||
359 | /* | |
360 | * fixup_activate is called when: | |
361 | * - an active object is activated | |
362 | * - an unknown object is activated (might be a statically initialized object) | |
363 | */ | |
364 | static int hrtimer_fixup_activate(void *addr, enum debug_obj_state state) | |
365 | { | |
366 | switch (state) { | |
367 | ||
368 | case ODEBUG_STATE_NOTAVAILABLE: | |
369 | WARN_ON_ONCE(1); | |
370 | return 0; | |
371 | ||
372 | case ODEBUG_STATE_ACTIVE: | |
373 | WARN_ON(1); | |
374 | ||
375 | default: | |
376 | return 0; | |
377 | } | |
378 | } | |
379 | ||
380 | /* | |
381 | * fixup_free is called when: | |
382 | * - an active object is freed | |
383 | */ | |
384 | static int hrtimer_fixup_free(void *addr, enum debug_obj_state state) | |
385 | { | |
386 | struct hrtimer *timer = addr; | |
387 | ||
388 | switch (state) { | |
389 | case ODEBUG_STATE_ACTIVE: | |
390 | hrtimer_cancel(timer); | |
391 | debug_object_free(timer, &hrtimer_debug_descr); | |
392 | return 1; | |
393 | default: | |
394 | return 0; | |
395 | } | |
396 | } | |
397 | ||
398 | static struct debug_obj_descr hrtimer_debug_descr = { | |
399 | .name = "hrtimer", | |
400 | .fixup_init = hrtimer_fixup_init, | |
401 | .fixup_activate = hrtimer_fixup_activate, | |
402 | .fixup_free = hrtimer_fixup_free, | |
403 | }; | |
404 | ||
405 | static inline void debug_hrtimer_init(struct hrtimer *timer) | |
406 | { | |
407 | debug_object_init(timer, &hrtimer_debug_descr); | |
408 | } | |
409 | ||
410 | static inline void debug_hrtimer_activate(struct hrtimer *timer) | |
411 | { | |
412 | debug_object_activate(timer, &hrtimer_debug_descr); | |
413 | } | |
414 | ||
415 | static inline void debug_hrtimer_deactivate(struct hrtimer *timer) | |
416 | { | |
417 | debug_object_deactivate(timer, &hrtimer_debug_descr); | |
418 | } | |
419 | ||
420 | static inline void debug_hrtimer_free(struct hrtimer *timer) | |
421 | { | |
422 | debug_object_free(timer, &hrtimer_debug_descr); | |
423 | } | |
424 | ||
425 | static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id, | |
426 | enum hrtimer_mode mode); | |
427 | ||
428 | void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id, | |
429 | enum hrtimer_mode mode) | |
430 | { | |
431 | debug_object_init_on_stack(timer, &hrtimer_debug_descr); | |
432 | __hrtimer_init(timer, clock_id, mode); | |
433 | } | |
434 | ||
435 | void destroy_hrtimer_on_stack(struct hrtimer *timer) | |
436 | { | |
437 | debug_object_free(timer, &hrtimer_debug_descr); | |
438 | } | |
439 | ||
440 | #else | |
441 | static inline void debug_hrtimer_init(struct hrtimer *timer) { } | |
442 | static inline void debug_hrtimer_activate(struct hrtimer *timer) { } | |
443 | static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { } | |
444 | #endif | |
445 | ||
d3d74453 PZ |
446 | /* |
447 | * Check, whether the timer is on the callback pending list | |
448 | */ | |
449 | static inline int hrtimer_cb_pending(const struct hrtimer *timer) | |
450 | { | |
451 | return timer->state & HRTIMER_STATE_PENDING; | |
452 | } | |
453 | ||
454 | /* | |
455 | * Remove a timer from the callback pending list | |
456 | */ | |
457 | static inline void hrtimer_remove_cb_pending(struct hrtimer *timer) | |
458 | { | |
459 | list_del_init(&timer->cb_entry); | |
460 | } | |
461 | ||
54cdfdb4 TG |
462 | /* High resolution timer related functions */ |
463 | #ifdef CONFIG_HIGH_RES_TIMERS | |
464 | ||
465 | /* | |
466 | * High resolution timer enabled ? | |
467 | */ | |
468 | static int hrtimer_hres_enabled __read_mostly = 1; | |
469 | ||
470 | /* | |
471 | * Enable / Disable high resolution mode | |
472 | */ | |
473 | static int __init setup_hrtimer_hres(char *str) | |
474 | { | |
475 | if (!strcmp(str, "off")) | |
476 | hrtimer_hres_enabled = 0; | |
477 | else if (!strcmp(str, "on")) | |
478 | hrtimer_hres_enabled = 1; | |
479 | else | |
480 | return 0; | |
481 | return 1; | |
482 | } | |
483 | ||
484 | __setup("highres=", setup_hrtimer_hres); | |
485 | ||
486 | /* | |
487 | * hrtimer_high_res_enabled - query, if the highres mode is enabled | |
488 | */ | |
489 | static inline int hrtimer_is_hres_enabled(void) | |
490 | { | |
491 | return hrtimer_hres_enabled; | |
492 | } | |
493 | ||
494 | /* | |
495 | * Is the high resolution mode active ? | |
496 | */ | |
497 | static inline int hrtimer_hres_active(void) | |
498 | { | |
499 | return __get_cpu_var(hrtimer_bases).hres_active; | |
500 | } | |
501 | ||
502 | /* | |
503 | * Reprogram the event source with checking both queues for the | |
504 | * next event | |
505 | * Called with interrupts disabled and base->lock held | |
506 | */ | |
507 | static void hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base) | |
508 | { | |
509 | int i; | |
510 | struct hrtimer_clock_base *base = cpu_base->clock_base; | |
511 | ktime_t expires; | |
512 | ||
513 | cpu_base->expires_next.tv64 = KTIME_MAX; | |
514 | ||
515 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) { | |
516 | struct hrtimer *timer; | |
517 | ||
518 | if (!base->first) | |
519 | continue; | |
520 | timer = rb_entry(base->first, struct hrtimer, node); | |
521 | expires = ktime_sub(timer->expires, base->offset); | |
522 | if (expires.tv64 < cpu_base->expires_next.tv64) | |
523 | cpu_base->expires_next = expires; | |
524 | } | |
525 | ||
526 | if (cpu_base->expires_next.tv64 != KTIME_MAX) | |
527 | tick_program_event(cpu_base->expires_next, 1); | |
528 | } | |
529 | ||
530 | /* | |
531 | * Shared reprogramming for clock_realtime and clock_monotonic | |
532 | * | |
533 | * When a timer is enqueued and expires earlier than the already enqueued | |
534 | * timers, we have to check, whether it expires earlier than the timer for | |
535 | * which the clock event device was armed. | |
536 | * | |
537 | * Called with interrupts disabled and base->cpu_base.lock held | |
538 | */ | |
539 | static int hrtimer_reprogram(struct hrtimer *timer, | |
540 | struct hrtimer_clock_base *base) | |
541 | { | |
542 | ktime_t *expires_next = &__get_cpu_var(hrtimer_bases).expires_next; | |
543 | ktime_t expires = ktime_sub(timer->expires, base->offset); | |
544 | int res; | |
545 | ||
63070a79 TG |
546 | WARN_ON_ONCE(timer->expires.tv64 < 0); |
547 | ||
54cdfdb4 TG |
548 | /* |
549 | * When the callback is running, we do not reprogram the clock event | |
550 | * device. The timer callback is either running on a different CPU or | |
3a4fa0a2 | 551 | * the callback is executed in the hrtimer_interrupt context. The |
54cdfdb4 TG |
552 | * reprogramming is handled either by the softirq, which called the |
553 | * callback or at the end of the hrtimer_interrupt. | |
554 | */ | |
555 | if (hrtimer_callback_running(timer)) | |
556 | return 0; | |
557 | ||
63070a79 TG |
558 | /* |
559 | * CLOCK_REALTIME timer might be requested with an absolute | |
560 | * expiry time which is less than base->offset. Nothing wrong | |
561 | * about that, just avoid to call into the tick code, which | |
562 | * has now objections against negative expiry values. | |
563 | */ | |
564 | if (expires.tv64 < 0) | |
565 | return -ETIME; | |
566 | ||
54cdfdb4 TG |
567 | if (expires.tv64 >= expires_next->tv64) |
568 | return 0; | |
569 | ||
570 | /* | |
571 | * Clockevents returns -ETIME, when the event was in the past. | |
572 | */ | |
573 | res = tick_program_event(expires, 0); | |
574 | if (!IS_ERR_VALUE(res)) | |
575 | *expires_next = expires; | |
576 | return res; | |
577 | } | |
578 | ||
579 | ||
580 | /* | |
581 | * Retrigger next event is called after clock was set | |
582 | * | |
583 | * Called with interrupts disabled via on_each_cpu() | |
584 | */ | |
585 | static void retrigger_next_event(void *arg) | |
586 | { | |
587 | struct hrtimer_cpu_base *base; | |
588 | struct timespec realtime_offset; | |
589 | unsigned long seq; | |
590 | ||
591 | if (!hrtimer_hres_active()) | |
592 | return; | |
593 | ||
594 | do { | |
595 | seq = read_seqbegin(&xtime_lock); | |
596 | set_normalized_timespec(&realtime_offset, | |
597 | -wall_to_monotonic.tv_sec, | |
598 | -wall_to_monotonic.tv_nsec); | |
599 | } while (read_seqretry(&xtime_lock, seq)); | |
600 | ||
601 | base = &__get_cpu_var(hrtimer_bases); | |
602 | ||
603 | /* Adjust CLOCK_REALTIME offset */ | |
604 | spin_lock(&base->lock); | |
605 | base->clock_base[CLOCK_REALTIME].offset = | |
606 | timespec_to_ktime(realtime_offset); | |
607 | ||
608 | hrtimer_force_reprogram(base); | |
609 | spin_unlock(&base->lock); | |
610 | } | |
611 | ||
612 | /* | |
613 | * Clock realtime was set | |
614 | * | |
615 | * Change the offset of the realtime clock vs. the monotonic | |
616 | * clock. | |
617 | * | |
618 | * We might have to reprogram the high resolution timer interrupt. On | |
619 | * SMP we call the architecture specific code to retrigger _all_ high | |
620 | * resolution timer interrupts. On UP we just disable interrupts and | |
621 | * call the high resolution interrupt code. | |
622 | */ | |
623 | void clock_was_set(void) | |
624 | { | |
625 | /* Retrigger the CPU local events everywhere */ | |
626 | on_each_cpu(retrigger_next_event, NULL, 0, 1); | |
627 | } | |
628 | ||
995f054f IM |
629 | /* |
630 | * During resume we might have to reprogram the high resolution timer | |
631 | * interrupt (on the local CPU): | |
632 | */ | |
633 | void hres_timers_resume(void) | |
634 | { | |
635 | WARN_ON_ONCE(num_online_cpus() > 1); | |
636 | ||
637 | /* Retrigger the CPU local events: */ | |
638 | retrigger_next_event(NULL); | |
639 | } | |
640 | ||
54cdfdb4 TG |
641 | /* |
642 | * Initialize the high resolution related parts of cpu_base | |
643 | */ | |
644 | static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) | |
645 | { | |
646 | base->expires_next.tv64 = KTIME_MAX; | |
647 | base->hres_active = 0; | |
54cdfdb4 TG |
648 | } |
649 | ||
650 | /* | |
651 | * Initialize the high resolution related parts of a hrtimer | |
652 | */ | |
653 | static inline void hrtimer_init_timer_hres(struct hrtimer *timer) | |
654 | { | |
54cdfdb4 TG |
655 | } |
656 | ||
657 | /* | |
658 | * When High resolution timers are active, try to reprogram. Note, that in case | |
659 | * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry | |
660 | * check happens. The timer gets enqueued into the rbtree. The reprogramming | |
661 | * and expiry check is done in the hrtimer_interrupt or in the softirq. | |
662 | */ | |
663 | static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer, | |
664 | struct hrtimer_clock_base *base) | |
665 | { | |
666 | if (base->cpu_base->hres_active && hrtimer_reprogram(timer, base)) { | |
667 | ||
668 | /* Timer is expired, act upon the callback mode */ | |
669 | switch(timer->cb_mode) { | |
670 | case HRTIMER_CB_IRQSAFE_NO_RESTART: | |
237fc6e7 | 671 | debug_hrtimer_deactivate(timer); |
54cdfdb4 TG |
672 | /* |
673 | * We can call the callback from here. No restart | |
674 | * happens, so no danger of recursion | |
675 | */ | |
676 | BUG_ON(timer->function(timer) != HRTIMER_NORESTART); | |
677 | return 1; | |
678 | case HRTIMER_CB_IRQSAFE_NO_SOFTIRQ: | |
679 | /* | |
680 | * This is solely for the sched tick emulation with | |
681 | * dynamic tick support to ensure that we do not | |
682 | * restart the tick right on the edge and end up with | |
683 | * the tick timer in the softirq ! The calling site | |
684 | * takes care of this. | |
685 | */ | |
237fc6e7 | 686 | debug_hrtimer_deactivate(timer); |
54cdfdb4 TG |
687 | return 1; |
688 | case HRTIMER_CB_IRQSAFE: | |
689 | case HRTIMER_CB_SOFTIRQ: | |
690 | /* | |
691 | * Move everything else into the softirq pending list ! | |
692 | */ | |
693 | list_add_tail(&timer->cb_entry, | |
694 | &base->cpu_base->cb_pending); | |
695 | timer->state = HRTIMER_STATE_PENDING; | |
54cdfdb4 TG |
696 | return 1; |
697 | default: | |
698 | BUG(); | |
699 | } | |
700 | } | |
701 | return 0; | |
702 | } | |
703 | ||
704 | /* | |
705 | * Switch to high resolution mode | |
706 | */ | |
f8953856 | 707 | static int hrtimer_switch_to_hres(void) |
54cdfdb4 | 708 | { |
820de5c3 IM |
709 | int cpu = smp_processor_id(); |
710 | struct hrtimer_cpu_base *base = &per_cpu(hrtimer_bases, cpu); | |
54cdfdb4 TG |
711 | unsigned long flags; |
712 | ||
713 | if (base->hres_active) | |
f8953856 | 714 | return 1; |
54cdfdb4 TG |
715 | |
716 | local_irq_save(flags); | |
717 | ||
718 | if (tick_init_highres()) { | |
719 | local_irq_restore(flags); | |
820de5c3 IM |
720 | printk(KERN_WARNING "Could not switch to high resolution " |
721 | "mode on CPU %d\n", cpu); | |
f8953856 | 722 | return 0; |
54cdfdb4 TG |
723 | } |
724 | base->hres_active = 1; | |
725 | base->clock_base[CLOCK_REALTIME].resolution = KTIME_HIGH_RES; | |
726 | base->clock_base[CLOCK_MONOTONIC].resolution = KTIME_HIGH_RES; | |
727 | ||
728 | tick_setup_sched_timer(); | |
729 | ||
730 | /* "Retrigger" the interrupt to get things going */ | |
731 | retrigger_next_event(NULL); | |
732 | local_irq_restore(flags); | |
edfed66e | 733 | printk(KERN_DEBUG "Switched to high resolution mode on CPU %d\n", |
54cdfdb4 | 734 | smp_processor_id()); |
f8953856 | 735 | return 1; |
54cdfdb4 TG |
736 | } |
737 | ||
0c96c597 TG |
738 | static inline void hrtimer_raise_softirq(void) |
739 | { | |
740 | raise_softirq(HRTIMER_SOFTIRQ); | |
741 | } | |
742 | ||
54cdfdb4 TG |
743 | #else |
744 | ||
745 | static inline int hrtimer_hres_active(void) { return 0; } | |
746 | static inline int hrtimer_is_hres_enabled(void) { return 0; } | |
f8953856 | 747 | static inline int hrtimer_switch_to_hres(void) { return 0; } |
54cdfdb4 TG |
748 | static inline void hrtimer_force_reprogram(struct hrtimer_cpu_base *base) { } |
749 | static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer, | |
750 | struct hrtimer_clock_base *base) | |
751 | { | |
752 | return 0; | |
753 | } | |
54cdfdb4 TG |
754 | static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { } |
755 | static inline void hrtimer_init_timer_hres(struct hrtimer *timer) { } | |
d3d74453 PZ |
756 | static inline int hrtimer_reprogram(struct hrtimer *timer, |
757 | struct hrtimer_clock_base *base) | |
758 | { | |
759 | return 0; | |
760 | } | |
0c96c597 | 761 | static inline void hrtimer_raise_softirq(void) { } |
54cdfdb4 TG |
762 | |
763 | #endif /* CONFIG_HIGH_RES_TIMERS */ | |
764 | ||
82f67cd9 IM |
765 | #ifdef CONFIG_TIMER_STATS |
766 | void __timer_stats_hrtimer_set_start_info(struct hrtimer *timer, void *addr) | |
767 | { | |
768 | if (timer->start_site) | |
769 | return; | |
770 | ||
771 | timer->start_site = addr; | |
772 | memcpy(timer->start_comm, current->comm, TASK_COMM_LEN); | |
773 | timer->start_pid = current->pid; | |
774 | } | |
775 | #endif | |
776 | ||
c0a31329 | 777 | /* |
6506f2aa | 778 | * Counterpart to lock_hrtimer_base above: |
c0a31329 TG |
779 | */ |
780 | static inline | |
781 | void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) | |
782 | { | |
3c8aa39d | 783 | spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags); |
c0a31329 TG |
784 | } |
785 | ||
786 | /** | |
787 | * hrtimer_forward - forward the timer expiry | |
c0a31329 | 788 | * @timer: hrtimer to forward |
44f21475 | 789 | * @now: forward past this time |
c0a31329 TG |
790 | * @interval: the interval to forward |
791 | * | |
792 | * Forward the timer expiry so it will expire in the future. | |
8dca6f33 | 793 | * Returns the number of overruns. |
c0a31329 | 794 | */ |
4d672e7a | 795 | u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval) |
c0a31329 | 796 | { |
4d672e7a | 797 | u64 orun = 1; |
44f21475 | 798 | ktime_t delta; |
c0a31329 TG |
799 | |
800 | delta = ktime_sub(now, timer->expires); | |
801 | ||
802 | if (delta.tv64 < 0) | |
803 | return 0; | |
804 | ||
c9db4fa1 TG |
805 | if (interval.tv64 < timer->base->resolution.tv64) |
806 | interval.tv64 = timer->base->resolution.tv64; | |
807 | ||
c0a31329 | 808 | if (unlikely(delta.tv64 >= interval.tv64)) { |
df869b63 | 809 | s64 incr = ktime_to_ns(interval); |
c0a31329 TG |
810 | |
811 | orun = ktime_divns(delta, incr); | |
812 | timer->expires = ktime_add_ns(timer->expires, incr * orun); | |
813 | if (timer->expires.tv64 > now.tv64) | |
814 | return orun; | |
815 | /* | |
816 | * This (and the ktime_add() below) is the | |
817 | * correction for exact: | |
818 | */ | |
819 | orun++; | |
820 | } | |
5a7780e7 | 821 | timer->expires = ktime_add_safe(timer->expires, interval); |
c0a31329 TG |
822 | |
823 | return orun; | |
824 | } | |
6bdb6b62 | 825 | EXPORT_SYMBOL_GPL(hrtimer_forward); |
c0a31329 TG |
826 | |
827 | /* | |
828 | * enqueue_hrtimer - internal function to (re)start a timer | |
829 | * | |
830 | * The timer is inserted in expiry order. Insertion into the | |
831 | * red black tree is O(log(n)). Must hold the base lock. | |
832 | */ | |
3c8aa39d | 833 | static void enqueue_hrtimer(struct hrtimer *timer, |
54cdfdb4 | 834 | struct hrtimer_clock_base *base, int reprogram) |
c0a31329 TG |
835 | { |
836 | struct rb_node **link = &base->active.rb_node; | |
c0a31329 TG |
837 | struct rb_node *parent = NULL; |
838 | struct hrtimer *entry; | |
99bc2fcb | 839 | int leftmost = 1; |
c0a31329 | 840 | |
237fc6e7 TG |
841 | debug_hrtimer_activate(timer); |
842 | ||
c0a31329 TG |
843 | /* |
844 | * Find the right place in the rbtree: | |
845 | */ | |
846 | while (*link) { | |
847 | parent = *link; | |
848 | entry = rb_entry(parent, struct hrtimer, node); | |
849 | /* | |
850 | * We dont care about collisions. Nodes with | |
851 | * the same expiry time stay together. | |
852 | */ | |
99bc2fcb | 853 | if (timer->expires.tv64 < entry->expires.tv64) { |
c0a31329 | 854 | link = &(*link)->rb_left; |
99bc2fcb | 855 | } else { |
c0a31329 | 856 | link = &(*link)->rb_right; |
99bc2fcb IM |
857 | leftmost = 0; |
858 | } | |
c0a31329 TG |
859 | } |
860 | ||
861 | /* | |
288867ec TG |
862 | * Insert the timer to the rbtree and check whether it |
863 | * replaces the first pending timer | |
c0a31329 | 864 | */ |
99bc2fcb | 865 | if (leftmost) { |
54cdfdb4 TG |
866 | /* |
867 | * Reprogram the clock event device. When the timer is already | |
868 | * expired hrtimer_enqueue_reprogram has either called the | |
869 | * callback or added it to the pending list and raised the | |
870 | * softirq. | |
871 | * | |
872 | * This is a NOP for !HIGHRES | |
873 | */ | |
874 | if (reprogram && hrtimer_enqueue_reprogram(timer, base)) | |
875 | return; | |
876 | ||
877 | base->first = &timer->node; | |
878 | } | |
879 | ||
c0a31329 TG |
880 | rb_link_node(&timer->node, parent, link); |
881 | rb_insert_color(&timer->node, &base->active); | |
303e967f TG |
882 | /* |
883 | * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the | |
884 | * state of a possibly running callback. | |
885 | */ | |
886 | timer->state |= HRTIMER_STATE_ENQUEUED; | |
288867ec | 887 | } |
c0a31329 TG |
888 | |
889 | /* | |
890 | * __remove_hrtimer - internal function to remove a timer | |
891 | * | |
892 | * Caller must hold the base lock. | |
54cdfdb4 TG |
893 | * |
894 | * High resolution timer mode reprograms the clock event device when the | |
895 | * timer is the one which expires next. The caller can disable this by setting | |
896 | * reprogram to zero. This is useful, when the context does a reprogramming | |
897 | * anyway (e.g. timer interrupt) | |
c0a31329 | 898 | */ |
3c8aa39d | 899 | static void __remove_hrtimer(struct hrtimer *timer, |
303e967f | 900 | struct hrtimer_clock_base *base, |
54cdfdb4 | 901 | unsigned long newstate, int reprogram) |
c0a31329 | 902 | { |
54cdfdb4 TG |
903 | /* High res. callback list. NOP for !HIGHRES */ |
904 | if (hrtimer_cb_pending(timer)) | |
905 | hrtimer_remove_cb_pending(timer); | |
906 | else { | |
907 | /* | |
908 | * Remove the timer from the rbtree and replace the | |
909 | * first entry pointer if necessary. | |
910 | */ | |
911 | if (base->first == &timer->node) { | |
912 | base->first = rb_next(&timer->node); | |
913 | /* Reprogram the clock event device. if enabled */ | |
914 | if (reprogram && hrtimer_hres_active()) | |
915 | hrtimer_force_reprogram(base->cpu_base); | |
916 | } | |
917 | rb_erase(&timer->node, &base->active); | |
918 | } | |
303e967f | 919 | timer->state = newstate; |
c0a31329 TG |
920 | } |
921 | ||
922 | /* | |
923 | * remove hrtimer, called with base lock held | |
924 | */ | |
925 | static inline int | |
3c8aa39d | 926 | remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base) |
c0a31329 | 927 | { |
303e967f | 928 | if (hrtimer_is_queued(timer)) { |
54cdfdb4 TG |
929 | int reprogram; |
930 | ||
931 | /* | |
932 | * Remove the timer and force reprogramming when high | |
933 | * resolution mode is active and the timer is on the current | |
934 | * CPU. If we remove a timer on another CPU, reprogramming is | |
935 | * skipped. The interrupt event on this CPU is fired and | |
936 | * reprogramming happens in the interrupt handler. This is a | |
937 | * rare case and less expensive than a smp call. | |
938 | */ | |
237fc6e7 | 939 | debug_hrtimer_deactivate(timer); |
82f67cd9 | 940 | timer_stats_hrtimer_clear_start_info(timer); |
54cdfdb4 TG |
941 | reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases); |
942 | __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE, | |
943 | reprogram); | |
c0a31329 TG |
944 | return 1; |
945 | } | |
946 | return 0; | |
947 | } | |
948 | ||
949 | /** | |
950 | * hrtimer_start - (re)start an relative timer on the current CPU | |
c0a31329 TG |
951 | * @timer: the timer to be added |
952 | * @tim: expiry time | |
953 | * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL) | |
954 | * | |
955 | * Returns: | |
956 | * 0 on success | |
957 | * 1 when the timer was active | |
958 | */ | |
959 | int | |
960 | hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode) | |
961 | { | |
3c8aa39d | 962 | struct hrtimer_clock_base *base, *new_base; |
c0a31329 | 963 | unsigned long flags; |
0c96c597 | 964 | int ret, raise; |
c0a31329 TG |
965 | |
966 | base = lock_hrtimer_base(timer, &flags); | |
967 | ||
968 | /* Remove an active timer from the queue: */ | |
969 | ret = remove_hrtimer(timer, base); | |
970 | ||
971 | /* Switch the timer base, if necessary: */ | |
972 | new_base = switch_hrtimer_base(timer, base); | |
973 | ||
c9cb2e3d | 974 | if (mode == HRTIMER_MODE_REL) { |
5a7780e7 | 975 | tim = ktime_add_safe(tim, new_base->get_time()); |
06027bdd IM |
976 | /* |
977 | * CONFIG_TIME_LOW_RES is a temporary way for architectures | |
978 | * to signal that they simply return xtime in | |
979 | * do_gettimeoffset(). In this case we want to round up by | |
980 | * resolution when starting a relative timer, to avoid short | |
981 | * timeouts. This will go away with the GTOD framework. | |
982 | */ | |
983 | #ifdef CONFIG_TIME_LOW_RES | |
5a7780e7 | 984 | tim = ktime_add_safe(tim, base->resolution); |
06027bdd IM |
985 | #endif |
986 | } | |
237fc6e7 | 987 | |
c0a31329 TG |
988 | timer->expires = tim; |
989 | ||
82f67cd9 IM |
990 | timer_stats_hrtimer_set_start_info(timer); |
991 | ||
935c631d IM |
992 | /* |
993 | * Only allow reprogramming if the new base is on this CPU. | |
994 | * (it might still be on another CPU if the timer was pending) | |
995 | */ | |
996 | enqueue_hrtimer(timer, new_base, | |
997 | new_base->cpu_base == &__get_cpu_var(hrtimer_bases)); | |
c0a31329 | 998 | |
0c96c597 TG |
999 | /* |
1000 | * The timer may be expired and moved to the cb_pending | |
1001 | * list. We can not raise the softirq with base lock held due | |
1002 | * to a possible deadlock with runqueue lock. | |
1003 | */ | |
1004 | raise = timer->state == HRTIMER_STATE_PENDING; | |
1005 | ||
c0a31329 TG |
1006 | unlock_hrtimer_base(timer, &flags); |
1007 | ||
0c96c597 TG |
1008 | if (raise) |
1009 | hrtimer_raise_softirq(); | |
1010 | ||
c0a31329 TG |
1011 | return ret; |
1012 | } | |
8d16b764 | 1013 | EXPORT_SYMBOL_GPL(hrtimer_start); |
c0a31329 TG |
1014 | |
1015 | /** | |
1016 | * hrtimer_try_to_cancel - try to deactivate a timer | |
c0a31329 TG |
1017 | * @timer: hrtimer to stop |
1018 | * | |
1019 | * Returns: | |
1020 | * 0 when the timer was not active | |
1021 | * 1 when the timer was active | |
1022 | * -1 when the timer is currently excuting the callback function and | |
fa9799e3 | 1023 | * cannot be stopped |
c0a31329 TG |
1024 | */ |
1025 | int hrtimer_try_to_cancel(struct hrtimer *timer) | |
1026 | { | |
3c8aa39d | 1027 | struct hrtimer_clock_base *base; |
c0a31329 TG |
1028 | unsigned long flags; |
1029 | int ret = -1; | |
1030 | ||
1031 | base = lock_hrtimer_base(timer, &flags); | |
1032 | ||
303e967f | 1033 | if (!hrtimer_callback_running(timer)) |
c0a31329 TG |
1034 | ret = remove_hrtimer(timer, base); |
1035 | ||
1036 | unlock_hrtimer_base(timer, &flags); | |
1037 | ||
1038 | return ret; | |
1039 | ||
1040 | } | |
8d16b764 | 1041 | EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel); |
c0a31329 TG |
1042 | |
1043 | /** | |
1044 | * hrtimer_cancel - cancel a timer and wait for the handler to finish. | |
c0a31329 TG |
1045 | * @timer: the timer to be cancelled |
1046 | * | |
1047 | * Returns: | |
1048 | * 0 when the timer was not active | |
1049 | * 1 when the timer was active | |
1050 | */ | |
1051 | int hrtimer_cancel(struct hrtimer *timer) | |
1052 | { | |
1053 | for (;;) { | |
1054 | int ret = hrtimer_try_to_cancel(timer); | |
1055 | ||
1056 | if (ret >= 0) | |
1057 | return ret; | |
5ef37b19 | 1058 | cpu_relax(); |
c0a31329 TG |
1059 | } |
1060 | } | |
8d16b764 | 1061 | EXPORT_SYMBOL_GPL(hrtimer_cancel); |
c0a31329 TG |
1062 | |
1063 | /** | |
1064 | * hrtimer_get_remaining - get remaining time for the timer | |
c0a31329 TG |
1065 | * @timer: the timer to read |
1066 | */ | |
1067 | ktime_t hrtimer_get_remaining(const struct hrtimer *timer) | |
1068 | { | |
3c8aa39d | 1069 | struct hrtimer_clock_base *base; |
c0a31329 TG |
1070 | unsigned long flags; |
1071 | ktime_t rem; | |
1072 | ||
1073 | base = lock_hrtimer_base(timer, &flags); | |
3c8aa39d | 1074 | rem = ktime_sub(timer->expires, base->get_time()); |
c0a31329 TG |
1075 | unlock_hrtimer_base(timer, &flags); |
1076 | ||
1077 | return rem; | |
1078 | } | |
8d16b764 | 1079 | EXPORT_SYMBOL_GPL(hrtimer_get_remaining); |
c0a31329 | 1080 | |
fd064b9b | 1081 | #if defined(CONFIG_NO_IDLE_HZ) || defined(CONFIG_NO_HZ) |
69239749 TL |
1082 | /** |
1083 | * hrtimer_get_next_event - get the time until next expiry event | |
1084 | * | |
1085 | * Returns the delta to the next expiry event or KTIME_MAX if no timer | |
1086 | * is pending. | |
1087 | */ | |
1088 | ktime_t hrtimer_get_next_event(void) | |
1089 | { | |
3c8aa39d TG |
1090 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
1091 | struct hrtimer_clock_base *base = cpu_base->clock_base; | |
69239749 TL |
1092 | ktime_t delta, mindelta = { .tv64 = KTIME_MAX }; |
1093 | unsigned long flags; | |
1094 | int i; | |
1095 | ||
3c8aa39d TG |
1096 | spin_lock_irqsave(&cpu_base->lock, flags); |
1097 | ||
54cdfdb4 TG |
1098 | if (!hrtimer_hres_active()) { |
1099 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) { | |
1100 | struct hrtimer *timer; | |
69239749 | 1101 | |
54cdfdb4 TG |
1102 | if (!base->first) |
1103 | continue; | |
3c8aa39d | 1104 | |
54cdfdb4 TG |
1105 | timer = rb_entry(base->first, struct hrtimer, node); |
1106 | delta.tv64 = timer->expires.tv64; | |
1107 | delta = ktime_sub(delta, base->get_time()); | |
1108 | if (delta.tv64 < mindelta.tv64) | |
1109 | mindelta.tv64 = delta.tv64; | |
1110 | } | |
69239749 | 1111 | } |
3c8aa39d TG |
1112 | |
1113 | spin_unlock_irqrestore(&cpu_base->lock, flags); | |
1114 | ||
69239749 TL |
1115 | if (mindelta.tv64 < 0) |
1116 | mindelta.tv64 = 0; | |
1117 | return mindelta; | |
1118 | } | |
1119 | #endif | |
1120 | ||
237fc6e7 TG |
1121 | static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id, |
1122 | enum hrtimer_mode mode) | |
c0a31329 | 1123 | { |
3c8aa39d | 1124 | struct hrtimer_cpu_base *cpu_base; |
c0a31329 | 1125 | |
7978672c GA |
1126 | memset(timer, 0, sizeof(struct hrtimer)); |
1127 | ||
3c8aa39d | 1128 | cpu_base = &__raw_get_cpu_var(hrtimer_bases); |
c0a31329 | 1129 | |
c9cb2e3d | 1130 | if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS) |
7978672c GA |
1131 | clock_id = CLOCK_MONOTONIC; |
1132 | ||
3c8aa39d | 1133 | timer->base = &cpu_base->clock_base[clock_id]; |
d3d74453 | 1134 | INIT_LIST_HEAD(&timer->cb_entry); |
54cdfdb4 | 1135 | hrtimer_init_timer_hres(timer); |
82f67cd9 IM |
1136 | |
1137 | #ifdef CONFIG_TIMER_STATS | |
1138 | timer->start_site = NULL; | |
1139 | timer->start_pid = -1; | |
1140 | memset(timer->start_comm, 0, TASK_COMM_LEN); | |
1141 | #endif | |
c0a31329 | 1142 | } |
237fc6e7 TG |
1143 | |
1144 | /** | |
1145 | * hrtimer_init - initialize a timer to the given clock | |
1146 | * @timer: the timer to be initialized | |
1147 | * @clock_id: the clock to be used | |
1148 | * @mode: timer mode abs/rel | |
1149 | */ | |
1150 | void hrtimer_init(struct hrtimer *timer, clockid_t clock_id, | |
1151 | enum hrtimer_mode mode) | |
1152 | { | |
1153 | debug_hrtimer_init(timer); | |
1154 | __hrtimer_init(timer, clock_id, mode); | |
1155 | } | |
8d16b764 | 1156 | EXPORT_SYMBOL_GPL(hrtimer_init); |
c0a31329 TG |
1157 | |
1158 | /** | |
1159 | * hrtimer_get_res - get the timer resolution for a clock | |
c0a31329 TG |
1160 | * @which_clock: which clock to query |
1161 | * @tp: pointer to timespec variable to store the resolution | |
1162 | * | |
72fd4a35 RD |
1163 | * Store the resolution of the clock selected by @which_clock in the |
1164 | * variable pointed to by @tp. | |
c0a31329 TG |
1165 | */ |
1166 | int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp) | |
1167 | { | |
3c8aa39d | 1168 | struct hrtimer_cpu_base *cpu_base; |
c0a31329 | 1169 | |
3c8aa39d TG |
1170 | cpu_base = &__raw_get_cpu_var(hrtimer_bases); |
1171 | *tp = ktime_to_timespec(cpu_base->clock_base[which_clock].resolution); | |
c0a31329 TG |
1172 | |
1173 | return 0; | |
1174 | } | |
8d16b764 | 1175 | EXPORT_SYMBOL_GPL(hrtimer_get_res); |
c0a31329 | 1176 | |
d3d74453 PZ |
1177 | static void run_hrtimer_pending(struct hrtimer_cpu_base *cpu_base) |
1178 | { | |
1179 | spin_lock_irq(&cpu_base->lock); | |
1180 | ||
1181 | while (!list_empty(&cpu_base->cb_pending)) { | |
1182 | enum hrtimer_restart (*fn)(struct hrtimer *); | |
1183 | struct hrtimer *timer; | |
1184 | int restart; | |
1185 | ||
1186 | timer = list_entry(cpu_base->cb_pending.next, | |
1187 | struct hrtimer, cb_entry); | |
1188 | ||
237fc6e7 | 1189 | debug_hrtimer_deactivate(timer); |
d3d74453 PZ |
1190 | timer_stats_account_hrtimer(timer); |
1191 | ||
1192 | fn = timer->function; | |
1193 | __remove_hrtimer(timer, timer->base, HRTIMER_STATE_CALLBACK, 0); | |
1194 | spin_unlock_irq(&cpu_base->lock); | |
1195 | ||
1196 | restart = fn(timer); | |
1197 | ||
1198 | spin_lock_irq(&cpu_base->lock); | |
1199 | ||
1200 | timer->state &= ~HRTIMER_STATE_CALLBACK; | |
1201 | if (restart == HRTIMER_RESTART) { | |
1202 | BUG_ON(hrtimer_active(timer)); | |
1203 | /* | |
1204 | * Enqueue the timer, allow reprogramming of the event | |
1205 | * device | |
1206 | */ | |
1207 | enqueue_hrtimer(timer, timer->base, 1); | |
1208 | } else if (hrtimer_active(timer)) { | |
1209 | /* | |
1210 | * If the timer was rearmed on another CPU, reprogram | |
1211 | * the event device. | |
1212 | */ | |
d7b41a24 BS |
1213 | struct hrtimer_clock_base *base = timer->base; |
1214 | ||
1215 | if (base->first == &timer->node && | |
1216 | hrtimer_reprogram(timer, base)) { | |
1217 | /* | |
1218 | * Timer is expired. Thus move it from tree to | |
1219 | * pending list again. | |
1220 | */ | |
1221 | __remove_hrtimer(timer, base, | |
1222 | HRTIMER_STATE_PENDING, 0); | |
1223 | list_add_tail(&timer->cb_entry, | |
1224 | &base->cpu_base->cb_pending); | |
1225 | } | |
d3d74453 PZ |
1226 | } |
1227 | } | |
1228 | spin_unlock_irq(&cpu_base->lock); | |
1229 | } | |
1230 | ||
1231 | static void __run_hrtimer(struct hrtimer *timer) | |
1232 | { | |
1233 | struct hrtimer_clock_base *base = timer->base; | |
1234 | struct hrtimer_cpu_base *cpu_base = base->cpu_base; | |
1235 | enum hrtimer_restart (*fn)(struct hrtimer *); | |
1236 | int restart; | |
1237 | ||
237fc6e7 | 1238 | debug_hrtimer_deactivate(timer); |
d3d74453 PZ |
1239 | __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0); |
1240 | timer_stats_account_hrtimer(timer); | |
1241 | ||
1242 | fn = timer->function; | |
1243 | if (timer->cb_mode == HRTIMER_CB_IRQSAFE_NO_SOFTIRQ) { | |
1244 | /* | |
1245 | * Used for scheduler timers, avoid lock inversion with | |
1246 | * rq->lock and tasklist_lock. | |
1247 | * | |
1248 | * These timers are required to deal with enqueue expiry | |
1249 | * themselves and are not allowed to migrate. | |
1250 | */ | |
1251 | spin_unlock(&cpu_base->lock); | |
1252 | restart = fn(timer); | |
1253 | spin_lock(&cpu_base->lock); | |
1254 | } else | |
1255 | restart = fn(timer); | |
1256 | ||
1257 | /* | |
1258 | * Note: We clear the CALLBACK bit after enqueue_hrtimer to avoid | |
1259 | * reprogramming of the event hardware. This happens at the end of this | |
1260 | * function anyway. | |
1261 | */ | |
1262 | if (restart != HRTIMER_NORESTART) { | |
1263 | BUG_ON(timer->state != HRTIMER_STATE_CALLBACK); | |
1264 | enqueue_hrtimer(timer, base, 0); | |
1265 | } | |
1266 | timer->state &= ~HRTIMER_STATE_CALLBACK; | |
1267 | } | |
1268 | ||
54cdfdb4 TG |
1269 | #ifdef CONFIG_HIGH_RES_TIMERS |
1270 | ||
1271 | /* | |
1272 | * High resolution timer interrupt | |
1273 | * Called with interrupts disabled | |
1274 | */ | |
1275 | void hrtimer_interrupt(struct clock_event_device *dev) | |
1276 | { | |
1277 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); | |
1278 | struct hrtimer_clock_base *base; | |
1279 | ktime_t expires_next, now; | |
1280 | int i, raise = 0; | |
1281 | ||
1282 | BUG_ON(!cpu_base->hres_active); | |
1283 | cpu_base->nr_events++; | |
1284 | dev->next_event.tv64 = KTIME_MAX; | |
1285 | ||
1286 | retry: | |
1287 | now = ktime_get(); | |
1288 | ||
1289 | expires_next.tv64 = KTIME_MAX; | |
1290 | ||
1291 | base = cpu_base->clock_base; | |
1292 | ||
1293 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { | |
1294 | ktime_t basenow; | |
1295 | struct rb_node *node; | |
1296 | ||
1297 | spin_lock(&cpu_base->lock); | |
1298 | ||
1299 | basenow = ktime_add(now, base->offset); | |
1300 | ||
1301 | while ((node = base->first)) { | |
1302 | struct hrtimer *timer; | |
1303 | ||
1304 | timer = rb_entry(node, struct hrtimer, node); | |
1305 | ||
1306 | if (basenow.tv64 < timer->expires.tv64) { | |
1307 | ktime_t expires; | |
1308 | ||
1309 | expires = ktime_sub(timer->expires, | |
1310 | base->offset); | |
1311 | if (expires.tv64 < expires_next.tv64) | |
1312 | expires_next = expires; | |
1313 | break; | |
1314 | } | |
1315 | ||
1316 | /* Move softirq callbacks to the pending list */ | |
1317 | if (timer->cb_mode == HRTIMER_CB_SOFTIRQ) { | |
1318 | __remove_hrtimer(timer, base, | |
1319 | HRTIMER_STATE_PENDING, 0); | |
1320 | list_add_tail(&timer->cb_entry, | |
1321 | &base->cpu_base->cb_pending); | |
1322 | raise = 1; | |
1323 | continue; | |
1324 | } | |
1325 | ||
d3d74453 | 1326 | __run_hrtimer(timer); |
54cdfdb4 TG |
1327 | } |
1328 | spin_unlock(&cpu_base->lock); | |
1329 | base++; | |
1330 | } | |
1331 | ||
1332 | cpu_base->expires_next = expires_next; | |
1333 | ||
1334 | /* Reprogramming necessary ? */ | |
1335 | if (expires_next.tv64 != KTIME_MAX) { | |
1336 | if (tick_program_event(expires_next, 0)) | |
1337 | goto retry; | |
1338 | } | |
1339 | ||
1340 | /* Raise softirq ? */ | |
1341 | if (raise) | |
1342 | raise_softirq(HRTIMER_SOFTIRQ); | |
1343 | } | |
1344 | ||
1345 | static void run_hrtimer_softirq(struct softirq_action *h) | |
1346 | { | |
d3d74453 PZ |
1347 | run_hrtimer_pending(&__get_cpu_var(hrtimer_bases)); |
1348 | } | |
54cdfdb4 | 1349 | |
d3d74453 | 1350 | #endif /* CONFIG_HIGH_RES_TIMERS */ |
82f67cd9 | 1351 | |
d3d74453 PZ |
1352 | /* |
1353 | * Called from timer softirq every jiffy, expire hrtimers: | |
1354 | * | |
1355 | * For HRT its the fall back code to run the softirq in the timer | |
1356 | * softirq context in case the hrtimer initialization failed or has | |
1357 | * not been done yet. | |
1358 | */ | |
1359 | void hrtimer_run_pending(void) | |
1360 | { | |
1361 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); | |
54cdfdb4 | 1362 | |
d3d74453 PZ |
1363 | if (hrtimer_hres_active()) |
1364 | return; | |
54cdfdb4 | 1365 | |
d3d74453 PZ |
1366 | /* |
1367 | * This _is_ ugly: We have to check in the softirq context, | |
1368 | * whether we can switch to highres and / or nohz mode. The | |
1369 | * clocksource switch happens in the timer interrupt with | |
1370 | * xtime_lock held. Notification from there only sets the | |
1371 | * check bit in the tick_oneshot code, otherwise we might | |
1372 | * deadlock vs. xtime_lock. | |
1373 | */ | |
1374 | if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) | |
1375 | hrtimer_switch_to_hres(); | |
54cdfdb4 | 1376 | |
d3d74453 | 1377 | run_hrtimer_pending(cpu_base); |
54cdfdb4 TG |
1378 | } |
1379 | ||
c0a31329 | 1380 | /* |
d3d74453 | 1381 | * Called from hardirq context every jiffy |
c0a31329 | 1382 | */ |
833883d9 | 1383 | void hrtimer_run_queues(void) |
c0a31329 | 1384 | { |
288867ec | 1385 | struct rb_node *node; |
833883d9 DS |
1386 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
1387 | struct hrtimer_clock_base *base; | |
1388 | int index, gettime = 1; | |
c0a31329 | 1389 | |
833883d9 | 1390 | if (hrtimer_hres_active()) |
3055adda DS |
1391 | return; |
1392 | ||
833883d9 DS |
1393 | for (index = 0; index < HRTIMER_MAX_CLOCK_BASES; index++) { |
1394 | base = &cpu_base->clock_base[index]; | |
c0a31329 | 1395 | |
833883d9 | 1396 | if (!base->first) |
d3d74453 | 1397 | continue; |
833883d9 | 1398 | |
259aae86 TG |
1399 | if (base->get_softirq_time) |
1400 | base->softirq_time = base->get_softirq_time(); | |
1401 | else if (gettime) { | |
833883d9 DS |
1402 | hrtimer_get_softirq_time(cpu_base); |
1403 | gettime = 0; | |
b75f7a51 | 1404 | } |
d3d74453 | 1405 | |
833883d9 | 1406 | spin_lock(&cpu_base->lock); |
c0a31329 | 1407 | |
833883d9 DS |
1408 | while ((node = base->first)) { |
1409 | struct hrtimer *timer; | |
54cdfdb4 | 1410 | |
833883d9 DS |
1411 | timer = rb_entry(node, struct hrtimer, node); |
1412 | if (base->softirq_time.tv64 <= timer->expires.tv64) | |
1413 | break; | |
1414 | ||
1415 | if (timer->cb_mode == HRTIMER_CB_SOFTIRQ) { | |
1416 | __remove_hrtimer(timer, base, | |
1417 | HRTIMER_STATE_PENDING, 0); | |
1418 | list_add_tail(&timer->cb_entry, | |
1419 | &base->cpu_base->cb_pending); | |
1420 | continue; | |
1421 | } | |
92127c7a | 1422 | |
833883d9 DS |
1423 | __run_hrtimer(timer); |
1424 | } | |
1425 | spin_unlock(&cpu_base->lock); | |
1426 | } | |
c0a31329 TG |
1427 | } |
1428 | ||
10c94ec1 TG |
1429 | /* |
1430 | * Sleep related functions: | |
1431 | */ | |
c9cb2e3d | 1432 | static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer) |
00362e33 TG |
1433 | { |
1434 | struct hrtimer_sleeper *t = | |
1435 | container_of(timer, struct hrtimer_sleeper, timer); | |
1436 | struct task_struct *task = t->task; | |
1437 | ||
1438 | t->task = NULL; | |
1439 | if (task) | |
1440 | wake_up_process(task); | |
1441 | ||
1442 | return HRTIMER_NORESTART; | |
1443 | } | |
1444 | ||
36c8b586 | 1445 | void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task) |
00362e33 TG |
1446 | { |
1447 | sl->timer.function = hrtimer_wakeup; | |
1448 | sl->task = task; | |
54cdfdb4 | 1449 | #ifdef CONFIG_HIGH_RES_TIMERS |
37bb6cb4 | 1450 | sl->timer.cb_mode = HRTIMER_CB_IRQSAFE_NO_SOFTIRQ; |
54cdfdb4 | 1451 | #endif |
00362e33 TG |
1452 | } |
1453 | ||
669d7868 | 1454 | static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode) |
432569bb | 1455 | { |
669d7868 | 1456 | hrtimer_init_sleeper(t, current); |
10c94ec1 | 1457 | |
432569bb RZ |
1458 | do { |
1459 | set_current_state(TASK_INTERRUPTIBLE); | |
1460 | hrtimer_start(&t->timer, t->timer.expires, mode); | |
37bb6cb4 PZ |
1461 | if (!hrtimer_active(&t->timer)) |
1462 | t->task = NULL; | |
432569bb | 1463 | |
54cdfdb4 TG |
1464 | if (likely(t->task)) |
1465 | schedule(); | |
432569bb | 1466 | |
669d7868 | 1467 | hrtimer_cancel(&t->timer); |
c9cb2e3d | 1468 | mode = HRTIMER_MODE_ABS; |
669d7868 TG |
1469 | |
1470 | } while (t->task && !signal_pending(current)); | |
432569bb | 1471 | |
3588a085 PZ |
1472 | __set_current_state(TASK_RUNNING); |
1473 | ||
669d7868 | 1474 | return t->task == NULL; |
10c94ec1 TG |
1475 | } |
1476 | ||
080344b9 ON |
1477 | static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp) |
1478 | { | |
1479 | struct timespec rmt; | |
1480 | ktime_t rem; | |
1481 | ||
1482 | rem = ktime_sub(timer->expires, timer->base->get_time()); | |
1483 | if (rem.tv64 <= 0) | |
1484 | return 0; | |
1485 | rmt = ktime_to_timespec(rem); | |
1486 | ||
1487 | if (copy_to_user(rmtp, &rmt, sizeof(*rmtp))) | |
1488 | return -EFAULT; | |
1489 | ||
1490 | return 1; | |
1491 | } | |
1492 | ||
1711ef38 | 1493 | long __sched hrtimer_nanosleep_restart(struct restart_block *restart) |
10c94ec1 | 1494 | { |
669d7868 | 1495 | struct hrtimer_sleeper t; |
080344b9 | 1496 | struct timespec __user *rmtp; |
237fc6e7 | 1497 | int ret = 0; |
10c94ec1 | 1498 | |
237fc6e7 TG |
1499 | hrtimer_init_on_stack(&t.timer, restart->nanosleep.index, |
1500 | HRTIMER_MODE_ABS); | |
029a07e0 | 1501 | t.timer.expires.tv64 = restart->nanosleep.expires; |
10c94ec1 | 1502 | |
c9cb2e3d | 1503 | if (do_nanosleep(&t, HRTIMER_MODE_ABS)) |
237fc6e7 | 1504 | goto out; |
10c94ec1 | 1505 | |
029a07e0 | 1506 | rmtp = restart->nanosleep.rmtp; |
432569bb | 1507 | if (rmtp) { |
237fc6e7 | 1508 | ret = update_rmtp(&t.timer, rmtp); |
080344b9 | 1509 | if (ret <= 0) |
237fc6e7 | 1510 | goto out; |
432569bb | 1511 | } |
10c94ec1 | 1512 | |
10c94ec1 | 1513 | /* The other values in restart are already filled in */ |
237fc6e7 TG |
1514 | ret = -ERESTART_RESTARTBLOCK; |
1515 | out: | |
1516 | destroy_hrtimer_on_stack(&t.timer); | |
1517 | return ret; | |
10c94ec1 TG |
1518 | } |
1519 | ||
080344b9 | 1520 | long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp, |
10c94ec1 TG |
1521 | const enum hrtimer_mode mode, const clockid_t clockid) |
1522 | { | |
1523 | struct restart_block *restart; | |
669d7868 | 1524 | struct hrtimer_sleeper t; |
237fc6e7 | 1525 | int ret = 0; |
10c94ec1 | 1526 | |
237fc6e7 | 1527 | hrtimer_init_on_stack(&t.timer, clockid, mode); |
432569bb RZ |
1528 | t.timer.expires = timespec_to_ktime(*rqtp); |
1529 | if (do_nanosleep(&t, mode)) | |
237fc6e7 | 1530 | goto out; |
10c94ec1 | 1531 | |
7978672c | 1532 | /* Absolute timers do not update the rmtp value and restart: */ |
237fc6e7 TG |
1533 | if (mode == HRTIMER_MODE_ABS) { |
1534 | ret = -ERESTARTNOHAND; | |
1535 | goto out; | |
1536 | } | |
10c94ec1 | 1537 | |
432569bb | 1538 | if (rmtp) { |
237fc6e7 | 1539 | ret = update_rmtp(&t.timer, rmtp); |
080344b9 | 1540 | if (ret <= 0) |
237fc6e7 | 1541 | goto out; |
432569bb | 1542 | } |
10c94ec1 TG |
1543 | |
1544 | restart = ¤t_thread_info()->restart_block; | |
1711ef38 | 1545 | restart->fn = hrtimer_nanosleep_restart; |
029a07e0 TG |
1546 | restart->nanosleep.index = t.timer.base->index; |
1547 | restart->nanosleep.rmtp = rmtp; | |
1548 | restart->nanosleep.expires = t.timer.expires.tv64; | |
10c94ec1 | 1549 | |
237fc6e7 TG |
1550 | ret = -ERESTART_RESTARTBLOCK; |
1551 | out: | |
1552 | destroy_hrtimer_on_stack(&t.timer); | |
1553 | return ret; | |
10c94ec1 TG |
1554 | } |
1555 | ||
6ba1b912 TG |
1556 | asmlinkage long |
1557 | sys_nanosleep(struct timespec __user *rqtp, struct timespec __user *rmtp) | |
1558 | { | |
080344b9 | 1559 | struct timespec tu; |
6ba1b912 TG |
1560 | |
1561 | if (copy_from_user(&tu, rqtp, sizeof(tu))) | |
1562 | return -EFAULT; | |
1563 | ||
1564 | if (!timespec_valid(&tu)) | |
1565 | return -EINVAL; | |
1566 | ||
080344b9 | 1567 | return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC); |
6ba1b912 TG |
1568 | } |
1569 | ||
c0a31329 TG |
1570 | /* |
1571 | * Functions related to boot-time initialization: | |
1572 | */ | |
0ec160dd | 1573 | static void __cpuinit init_hrtimers_cpu(int cpu) |
c0a31329 | 1574 | { |
3c8aa39d | 1575 | struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu); |
c0a31329 TG |
1576 | int i; |
1577 | ||
3c8aa39d | 1578 | spin_lock_init(&cpu_base->lock); |
3c8aa39d TG |
1579 | |
1580 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) | |
1581 | cpu_base->clock_base[i].cpu_base = cpu_base; | |
1582 | ||
d3d74453 | 1583 | INIT_LIST_HEAD(&cpu_base->cb_pending); |
54cdfdb4 | 1584 | hrtimer_init_hres(cpu_base); |
c0a31329 TG |
1585 | } |
1586 | ||
1587 | #ifdef CONFIG_HOTPLUG_CPU | |
1588 | ||
3c8aa39d TG |
1589 | static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base, |
1590 | struct hrtimer_clock_base *new_base) | |
c0a31329 TG |
1591 | { |
1592 | struct hrtimer *timer; | |
1593 | struct rb_node *node; | |
1594 | ||
1595 | while ((node = rb_first(&old_base->active))) { | |
1596 | timer = rb_entry(node, struct hrtimer, node); | |
54cdfdb4 | 1597 | BUG_ON(hrtimer_callback_running(timer)); |
237fc6e7 | 1598 | debug_hrtimer_deactivate(timer); |
54cdfdb4 | 1599 | __remove_hrtimer(timer, old_base, HRTIMER_STATE_INACTIVE, 0); |
c0a31329 | 1600 | timer->base = new_base; |
54cdfdb4 TG |
1601 | /* |
1602 | * Enqueue the timer. Allow reprogramming of the event device | |
1603 | */ | |
1604 | enqueue_hrtimer(timer, new_base, 1); | |
c0a31329 TG |
1605 | } |
1606 | } | |
1607 | ||
1608 | static void migrate_hrtimers(int cpu) | |
1609 | { | |
3c8aa39d | 1610 | struct hrtimer_cpu_base *old_base, *new_base; |
c0a31329 TG |
1611 | int i; |
1612 | ||
1613 | BUG_ON(cpu_online(cpu)); | |
3c8aa39d TG |
1614 | old_base = &per_cpu(hrtimer_bases, cpu); |
1615 | new_base = &get_cpu_var(hrtimer_bases); | |
c0a31329 | 1616 | |
54cdfdb4 TG |
1617 | tick_cancel_sched_timer(cpu); |
1618 | ||
c0a31329 | 1619 | local_irq_disable(); |
8e60e05f ON |
1620 | spin_lock(&new_base->lock); |
1621 | spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING); | |
c0a31329 | 1622 | |
3c8aa39d | 1623 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
3c8aa39d TG |
1624 | migrate_hrtimer_list(&old_base->clock_base[i], |
1625 | &new_base->clock_base[i]); | |
c0a31329 TG |
1626 | } |
1627 | ||
8e60e05f ON |
1628 | spin_unlock(&old_base->lock); |
1629 | spin_unlock(&new_base->lock); | |
c0a31329 TG |
1630 | local_irq_enable(); |
1631 | put_cpu_var(hrtimer_bases); | |
1632 | } | |
1633 | #endif /* CONFIG_HOTPLUG_CPU */ | |
1634 | ||
8c78f307 | 1635 | static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self, |
c0a31329 TG |
1636 | unsigned long action, void *hcpu) |
1637 | { | |
7713a7d1 | 1638 | unsigned int cpu = (long)hcpu; |
c0a31329 TG |
1639 | |
1640 | switch (action) { | |
1641 | ||
1642 | case CPU_UP_PREPARE: | |
8bb78442 | 1643 | case CPU_UP_PREPARE_FROZEN: |
c0a31329 TG |
1644 | init_hrtimers_cpu(cpu); |
1645 | break; | |
1646 | ||
1647 | #ifdef CONFIG_HOTPLUG_CPU | |
1648 | case CPU_DEAD: | |
8bb78442 | 1649 | case CPU_DEAD_FROZEN: |
d316c57f | 1650 | clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &cpu); |
c0a31329 TG |
1651 | migrate_hrtimers(cpu); |
1652 | break; | |
1653 | #endif | |
1654 | ||
1655 | default: | |
1656 | break; | |
1657 | } | |
1658 | ||
1659 | return NOTIFY_OK; | |
1660 | } | |
1661 | ||
8c78f307 | 1662 | static struct notifier_block __cpuinitdata hrtimers_nb = { |
c0a31329 TG |
1663 | .notifier_call = hrtimer_cpu_notify, |
1664 | }; | |
1665 | ||
1666 | void __init hrtimers_init(void) | |
1667 | { | |
1668 | hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE, | |
1669 | (void *)(long)smp_processor_id()); | |
1670 | register_cpu_notifier(&hrtimers_nb); | |
54cdfdb4 TG |
1671 | #ifdef CONFIG_HIGH_RES_TIMERS |
1672 | open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq, NULL); | |
1673 | #endif | |
c0a31329 TG |
1674 | } |
1675 |