]>
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> |
eea08f32 AB |
46 | #include <linux/sched.h> |
47 | #include <linux/timer.h> | |
c0a31329 TG |
48 | |
49 | #include <asm/uaccess.h> | |
50 | ||
c6a2a177 XG |
51 | #include <trace/events/timer.h> |
52 | ||
c0a31329 TG |
53 | /* |
54 | * The timer bases: | |
7978672c GA |
55 | * |
56 | * Note: If we want to add new timer bases, we have to skip the two | |
57 | * clock ids captured by the cpu-timers. We do this by holding empty | |
58 | * entries rather than doing math adjustment of the clock ids. | |
59 | * This ensures that we capture erroneous accesses to these clock ids | |
60 | * rather than moving them into the range of valid clock id's. | |
c0a31329 | 61 | */ |
54cdfdb4 | 62 | DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) = |
c0a31329 | 63 | { |
3c8aa39d TG |
64 | |
65 | .clock_base = | |
c0a31329 | 66 | { |
3c8aa39d TG |
67 | { |
68 | .index = CLOCK_REALTIME, | |
69 | .get_time = &ktime_get_real, | |
54cdfdb4 | 70 | .resolution = KTIME_LOW_RES, |
3c8aa39d TG |
71 | }, |
72 | { | |
73 | .index = CLOCK_MONOTONIC, | |
74 | .get_time = &ktime_get, | |
54cdfdb4 | 75 | .resolution = KTIME_LOW_RES, |
3c8aa39d TG |
76 | }, |
77 | } | |
c0a31329 TG |
78 | }; |
79 | ||
92127c7a TG |
80 | /* |
81 | * Get the coarse grained time at the softirq based on xtime and | |
82 | * wall_to_monotonic. | |
83 | */ | |
3c8aa39d | 84 | static void hrtimer_get_softirq_time(struct hrtimer_cpu_base *base) |
92127c7a TG |
85 | { |
86 | ktime_t xtim, tomono; | |
ad28d94a | 87 | struct timespec xts, tom; |
92127c7a TG |
88 | unsigned long seq; |
89 | ||
90 | do { | |
91 | seq = read_seqbegin(&xtime_lock); | |
174bd199 | 92 | xts = __current_kernel_time(); |
8ab4351a | 93 | tom = __get_wall_to_monotonic(); |
92127c7a TG |
94 | } while (read_seqretry(&xtime_lock, seq)); |
95 | ||
f4304ab2 | 96 | xtim = timespec_to_ktime(xts); |
ad28d94a | 97 | tomono = timespec_to_ktime(tom); |
3c8aa39d TG |
98 | base->clock_base[CLOCK_REALTIME].softirq_time = xtim; |
99 | base->clock_base[CLOCK_MONOTONIC].softirq_time = | |
100 | ktime_add(xtim, tomono); | |
92127c7a TG |
101 | } |
102 | ||
c0a31329 TG |
103 | /* |
104 | * Functions and macros which are different for UP/SMP systems are kept in a | |
105 | * single place | |
106 | */ | |
107 | #ifdef CONFIG_SMP | |
108 | ||
c0a31329 TG |
109 | /* |
110 | * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock | |
111 | * means that all timers which are tied to this base via timer->base are | |
112 | * locked, and the base itself is locked too. | |
113 | * | |
114 | * So __run_timers/migrate_timers can safely modify all timers which could | |
115 | * be found on the lists/queues. | |
116 | * | |
117 | * When the timer's base is locked, and the timer removed from list, it is | |
118 | * possible to set timer->base = NULL and drop the lock: the timer remains | |
119 | * locked. | |
120 | */ | |
3c8aa39d TG |
121 | static |
122 | struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer, | |
123 | unsigned long *flags) | |
c0a31329 | 124 | { |
3c8aa39d | 125 | struct hrtimer_clock_base *base; |
c0a31329 TG |
126 | |
127 | for (;;) { | |
128 | base = timer->base; | |
129 | if (likely(base != NULL)) { | |
ecb49d1a | 130 | raw_spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
131 | if (likely(base == timer->base)) |
132 | return base; | |
133 | /* The timer has migrated to another CPU: */ | |
ecb49d1a | 134 | raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags); |
c0a31329 TG |
135 | } |
136 | cpu_relax(); | |
137 | } | |
138 | } | |
139 | ||
6ff7041d TG |
140 | |
141 | /* | |
142 | * Get the preferred target CPU for NOHZ | |
143 | */ | |
144 | static int hrtimer_get_target(int this_cpu, int pinned) | |
145 | { | |
146 | #ifdef CONFIG_NO_HZ | |
83cd4fe2 VP |
147 | if (!pinned && get_sysctl_timer_migration() && idle_cpu(this_cpu)) |
148 | return get_nohz_timer_target(); | |
6ff7041d TG |
149 | #endif |
150 | return this_cpu; | |
151 | } | |
152 | ||
153 | /* | |
154 | * With HIGHRES=y we do not migrate the timer when it is expiring | |
155 | * before the next event on the target cpu because we cannot reprogram | |
156 | * the target cpu hardware and we would cause it to fire late. | |
157 | * | |
158 | * Called with cpu_base->lock of target cpu held. | |
159 | */ | |
160 | static int | |
161 | hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base) | |
162 | { | |
163 | #ifdef CONFIG_HIGH_RES_TIMERS | |
164 | ktime_t expires; | |
165 | ||
166 | if (!new_base->cpu_base->hres_active) | |
167 | return 0; | |
168 | ||
169 | expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset); | |
170 | return expires.tv64 <= new_base->cpu_base->expires_next.tv64; | |
171 | #else | |
172 | return 0; | |
173 | #endif | |
174 | } | |
175 | ||
c0a31329 TG |
176 | /* |
177 | * Switch the timer base to the current CPU when possible. | |
178 | */ | |
3c8aa39d | 179 | static inline struct hrtimer_clock_base * |
597d0275 AB |
180 | switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base, |
181 | int pinned) | |
c0a31329 | 182 | { |
3c8aa39d TG |
183 | struct hrtimer_clock_base *new_base; |
184 | struct hrtimer_cpu_base *new_cpu_base; | |
6ff7041d TG |
185 | int this_cpu = smp_processor_id(); |
186 | int cpu = hrtimer_get_target(this_cpu, pinned); | |
c0a31329 | 187 | |
eea08f32 AB |
188 | again: |
189 | new_cpu_base = &per_cpu(hrtimer_bases, cpu); | |
3c8aa39d | 190 | new_base = &new_cpu_base->clock_base[base->index]; |
c0a31329 TG |
191 | |
192 | if (base != new_base) { | |
193 | /* | |
6ff7041d | 194 | * We are trying to move timer to new_base. |
c0a31329 TG |
195 | * However we can't change timer's base while it is running, |
196 | * so we keep it on the same CPU. No hassle vs. reprogramming | |
197 | * the event source in the high resolution case. The softirq | |
198 | * code will take care of this when the timer function has | |
199 | * completed. There is no conflict as we hold the lock until | |
200 | * the timer is enqueued. | |
201 | */ | |
54cdfdb4 | 202 | if (unlikely(hrtimer_callback_running(timer))) |
c0a31329 TG |
203 | return base; |
204 | ||
205 | /* See the comment in lock_timer_base() */ | |
206 | timer->base = NULL; | |
ecb49d1a TG |
207 | raw_spin_unlock(&base->cpu_base->lock); |
208 | raw_spin_lock(&new_base->cpu_base->lock); | |
eea08f32 | 209 | |
6ff7041d TG |
210 | if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) { |
211 | cpu = this_cpu; | |
ecb49d1a TG |
212 | raw_spin_unlock(&new_base->cpu_base->lock); |
213 | raw_spin_lock(&base->cpu_base->lock); | |
6ff7041d TG |
214 | timer->base = base; |
215 | goto again; | |
eea08f32 | 216 | } |
c0a31329 TG |
217 | timer->base = new_base; |
218 | } | |
219 | return new_base; | |
220 | } | |
221 | ||
222 | #else /* CONFIG_SMP */ | |
223 | ||
3c8aa39d | 224 | static inline struct hrtimer_clock_base * |
c0a31329 TG |
225 | lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) |
226 | { | |
3c8aa39d | 227 | struct hrtimer_clock_base *base = timer->base; |
c0a31329 | 228 | |
ecb49d1a | 229 | raw_spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
230 | |
231 | return base; | |
232 | } | |
233 | ||
eea08f32 | 234 | # define switch_hrtimer_base(t, b, p) (b) |
c0a31329 TG |
235 | |
236 | #endif /* !CONFIG_SMP */ | |
237 | ||
238 | /* | |
239 | * Functions for the union type storage format of ktime_t which are | |
240 | * too large for inlining: | |
241 | */ | |
242 | #if BITS_PER_LONG < 64 | |
243 | # ifndef CONFIG_KTIME_SCALAR | |
244 | /** | |
245 | * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable | |
c0a31329 TG |
246 | * @kt: addend |
247 | * @nsec: the scalar nsec value to add | |
248 | * | |
249 | * Returns the sum of kt and nsec in ktime_t format | |
250 | */ | |
251 | ktime_t ktime_add_ns(const ktime_t kt, u64 nsec) | |
252 | { | |
253 | ktime_t tmp; | |
254 | ||
255 | if (likely(nsec < NSEC_PER_SEC)) { | |
256 | tmp.tv64 = nsec; | |
257 | } else { | |
258 | unsigned long rem = do_div(nsec, NSEC_PER_SEC); | |
259 | ||
260 | tmp = ktime_set((long)nsec, rem); | |
261 | } | |
262 | ||
263 | return ktime_add(kt, tmp); | |
264 | } | |
b8b8fd2d DH |
265 | |
266 | EXPORT_SYMBOL_GPL(ktime_add_ns); | |
a272378d ACM |
267 | |
268 | /** | |
269 | * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable | |
270 | * @kt: minuend | |
271 | * @nsec: the scalar nsec value to subtract | |
272 | * | |
273 | * Returns the subtraction of @nsec from @kt in ktime_t format | |
274 | */ | |
275 | ktime_t ktime_sub_ns(const ktime_t kt, u64 nsec) | |
276 | { | |
277 | ktime_t tmp; | |
278 | ||
279 | if (likely(nsec < NSEC_PER_SEC)) { | |
280 | tmp.tv64 = nsec; | |
281 | } else { | |
282 | unsigned long rem = do_div(nsec, NSEC_PER_SEC); | |
283 | ||
284 | tmp = ktime_set((long)nsec, rem); | |
285 | } | |
286 | ||
287 | return ktime_sub(kt, tmp); | |
288 | } | |
289 | ||
290 | EXPORT_SYMBOL_GPL(ktime_sub_ns); | |
c0a31329 TG |
291 | # endif /* !CONFIG_KTIME_SCALAR */ |
292 | ||
293 | /* | |
294 | * Divide a ktime value by a nanosecond value | |
295 | */ | |
4d672e7a | 296 | u64 ktime_divns(const ktime_t kt, s64 div) |
c0a31329 | 297 | { |
900cfa46 | 298 | u64 dclc; |
c0a31329 TG |
299 | int sft = 0; |
300 | ||
900cfa46 | 301 | dclc = ktime_to_ns(kt); |
c0a31329 TG |
302 | /* Make sure the divisor is less than 2^32: */ |
303 | while (div >> 32) { | |
304 | sft++; | |
305 | div >>= 1; | |
306 | } | |
307 | dclc >>= sft; | |
308 | do_div(dclc, (unsigned long) div); | |
309 | ||
4d672e7a | 310 | return dclc; |
c0a31329 | 311 | } |
c0a31329 TG |
312 | #endif /* BITS_PER_LONG >= 64 */ |
313 | ||
5a7780e7 TG |
314 | /* |
315 | * Add two ktime values and do a safety check for overflow: | |
316 | */ | |
317 | ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs) | |
318 | { | |
319 | ktime_t res = ktime_add(lhs, rhs); | |
320 | ||
321 | /* | |
322 | * We use KTIME_SEC_MAX here, the maximum timeout which we can | |
323 | * return to user space in a timespec: | |
324 | */ | |
325 | if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64) | |
326 | res = ktime_set(KTIME_SEC_MAX, 0); | |
327 | ||
328 | return res; | |
329 | } | |
330 | ||
8daa21e6 AB |
331 | EXPORT_SYMBOL_GPL(ktime_add_safe); |
332 | ||
237fc6e7 TG |
333 | #ifdef CONFIG_DEBUG_OBJECTS_TIMERS |
334 | ||
335 | static struct debug_obj_descr hrtimer_debug_descr; | |
336 | ||
337 | /* | |
338 | * fixup_init is called when: | |
339 | * - an active object is initialized | |
340 | */ | |
341 | static int hrtimer_fixup_init(void *addr, enum debug_obj_state state) | |
342 | { | |
343 | struct hrtimer *timer = addr; | |
344 | ||
345 | switch (state) { | |
346 | case ODEBUG_STATE_ACTIVE: | |
347 | hrtimer_cancel(timer); | |
348 | debug_object_init(timer, &hrtimer_debug_descr); | |
349 | return 1; | |
350 | default: | |
351 | return 0; | |
352 | } | |
353 | } | |
354 | ||
355 | /* | |
356 | * fixup_activate is called when: | |
357 | * - an active object is activated | |
358 | * - an unknown object is activated (might be a statically initialized object) | |
359 | */ | |
360 | static int hrtimer_fixup_activate(void *addr, enum debug_obj_state state) | |
361 | { | |
362 | switch (state) { | |
363 | ||
364 | case ODEBUG_STATE_NOTAVAILABLE: | |
365 | WARN_ON_ONCE(1); | |
366 | return 0; | |
367 | ||
368 | case ODEBUG_STATE_ACTIVE: | |
369 | WARN_ON(1); | |
370 | ||
371 | default: | |
372 | return 0; | |
373 | } | |
374 | } | |
375 | ||
376 | /* | |
377 | * fixup_free is called when: | |
378 | * - an active object is freed | |
379 | */ | |
380 | static int hrtimer_fixup_free(void *addr, enum debug_obj_state state) | |
381 | { | |
382 | struct hrtimer *timer = addr; | |
383 | ||
384 | switch (state) { | |
385 | case ODEBUG_STATE_ACTIVE: | |
386 | hrtimer_cancel(timer); | |
387 | debug_object_free(timer, &hrtimer_debug_descr); | |
388 | return 1; | |
389 | default: | |
390 | return 0; | |
391 | } | |
392 | } | |
393 | ||
394 | static struct debug_obj_descr hrtimer_debug_descr = { | |
395 | .name = "hrtimer", | |
396 | .fixup_init = hrtimer_fixup_init, | |
397 | .fixup_activate = hrtimer_fixup_activate, | |
398 | .fixup_free = hrtimer_fixup_free, | |
399 | }; | |
400 | ||
401 | static inline void debug_hrtimer_init(struct hrtimer *timer) | |
402 | { | |
403 | debug_object_init(timer, &hrtimer_debug_descr); | |
404 | } | |
405 | ||
406 | static inline void debug_hrtimer_activate(struct hrtimer *timer) | |
407 | { | |
408 | debug_object_activate(timer, &hrtimer_debug_descr); | |
409 | } | |
410 | ||
411 | static inline void debug_hrtimer_deactivate(struct hrtimer *timer) | |
412 | { | |
413 | debug_object_deactivate(timer, &hrtimer_debug_descr); | |
414 | } | |
415 | ||
416 | static inline void debug_hrtimer_free(struct hrtimer *timer) | |
417 | { | |
418 | debug_object_free(timer, &hrtimer_debug_descr); | |
419 | } | |
420 | ||
421 | static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id, | |
422 | enum hrtimer_mode mode); | |
423 | ||
424 | void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id, | |
425 | enum hrtimer_mode mode) | |
426 | { | |
427 | debug_object_init_on_stack(timer, &hrtimer_debug_descr); | |
428 | __hrtimer_init(timer, clock_id, mode); | |
429 | } | |
2bc481cf | 430 | EXPORT_SYMBOL_GPL(hrtimer_init_on_stack); |
237fc6e7 TG |
431 | |
432 | void destroy_hrtimer_on_stack(struct hrtimer *timer) | |
433 | { | |
434 | debug_object_free(timer, &hrtimer_debug_descr); | |
435 | } | |
436 | ||
437 | #else | |
438 | static inline void debug_hrtimer_init(struct hrtimer *timer) { } | |
439 | static inline void debug_hrtimer_activate(struct hrtimer *timer) { } | |
440 | static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { } | |
441 | #endif | |
442 | ||
c6a2a177 XG |
443 | static inline void |
444 | debug_init(struct hrtimer *timer, clockid_t clockid, | |
445 | enum hrtimer_mode mode) | |
446 | { | |
447 | debug_hrtimer_init(timer); | |
448 | trace_hrtimer_init(timer, clockid, mode); | |
449 | } | |
450 | ||
451 | static inline void debug_activate(struct hrtimer *timer) | |
452 | { | |
453 | debug_hrtimer_activate(timer); | |
454 | trace_hrtimer_start(timer); | |
455 | } | |
456 | ||
457 | static inline void debug_deactivate(struct hrtimer *timer) | |
458 | { | |
459 | debug_hrtimer_deactivate(timer); | |
460 | trace_hrtimer_cancel(timer); | |
461 | } | |
462 | ||
54cdfdb4 TG |
463 | /* High resolution timer related functions */ |
464 | #ifdef CONFIG_HIGH_RES_TIMERS | |
465 | ||
466 | /* | |
467 | * High resolution timer enabled ? | |
468 | */ | |
469 | static int hrtimer_hres_enabled __read_mostly = 1; | |
470 | ||
471 | /* | |
472 | * Enable / Disable high resolution mode | |
473 | */ | |
474 | static int __init setup_hrtimer_hres(char *str) | |
475 | { | |
476 | if (!strcmp(str, "off")) | |
477 | hrtimer_hres_enabled = 0; | |
478 | else if (!strcmp(str, "on")) | |
479 | hrtimer_hres_enabled = 1; | |
480 | else | |
481 | return 0; | |
482 | return 1; | |
483 | } | |
484 | ||
485 | __setup("highres=", setup_hrtimer_hres); | |
486 | ||
487 | /* | |
488 | * hrtimer_high_res_enabled - query, if the highres mode is enabled | |
489 | */ | |
490 | static inline int hrtimer_is_hres_enabled(void) | |
491 | { | |
492 | return hrtimer_hres_enabled; | |
493 | } | |
494 | ||
495 | /* | |
496 | * Is the high resolution mode active ? | |
497 | */ | |
498 | static inline int hrtimer_hres_active(void) | |
499 | { | |
500 | return __get_cpu_var(hrtimer_bases).hres_active; | |
501 | } | |
502 | ||
503 | /* | |
504 | * Reprogram the event source with checking both queues for the | |
505 | * next event | |
506 | * Called with interrupts disabled and base->lock held | |
507 | */ | |
7403f41f AC |
508 | static void |
509 | hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal) | |
54cdfdb4 TG |
510 | { |
511 | int i; | |
512 | struct hrtimer_clock_base *base = cpu_base->clock_base; | |
7403f41f | 513 | ktime_t expires, expires_next; |
54cdfdb4 | 514 | |
7403f41f | 515 | expires_next.tv64 = KTIME_MAX; |
54cdfdb4 TG |
516 | |
517 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) { | |
518 | struct hrtimer *timer; | |
998adc3d | 519 | struct timerqueue_node *next; |
54cdfdb4 | 520 | |
998adc3d JS |
521 | next = timerqueue_getnext(&base->active); |
522 | if (!next) | |
54cdfdb4 | 523 | continue; |
998adc3d JS |
524 | timer = container_of(next, struct hrtimer, node); |
525 | ||
cc584b21 | 526 | expires = ktime_sub(hrtimer_get_expires(timer), base->offset); |
b0a9b511 TG |
527 | /* |
528 | * clock_was_set() has changed base->offset so the | |
529 | * result might be negative. Fix it up to prevent a | |
530 | * false positive in clockevents_program_event() | |
531 | */ | |
532 | if (expires.tv64 < 0) | |
533 | expires.tv64 = 0; | |
7403f41f AC |
534 | if (expires.tv64 < expires_next.tv64) |
535 | expires_next = expires; | |
54cdfdb4 TG |
536 | } |
537 | ||
7403f41f AC |
538 | if (skip_equal && expires_next.tv64 == cpu_base->expires_next.tv64) |
539 | return; | |
540 | ||
541 | cpu_base->expires_next.tv64 = expires_next.tv64; | |
542 | ||
54cdfdb4 TG |
543 | if (cpu_base->expires_next.tv64 != KTIME_MAX) |
544 | tick_program_event(cpu_base->expires_next, 1); | |
545 | } | |
546 | ||
547 | /* | |
548 | * Shared reprogramming for clock_realtime and clock_monotonic | |
549 | * | |
550 | * When a timer is enqueued and expires earlier than the already enqueued | |
551 | * timers, we have to check, whether it expires earlier than the timer for | |
552 | * which the clock event device was armed. | |
553 | * | |
554 | * Called with interrupts disabled and base->cpu_base.lock held | |
555 | */ | |
556 | static int hrtimer_reprogram(struct hrtimer *timer, | |
557 | struct hrtimer_clock_base *base) | |
558 | { | |
41d2e494 | 559 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
cc584b21 | 560 | ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset); |
54cdfdb4 TG |
561 | int res; |
562 | ||
cc584b21 | 563 | WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0); |
63070a79 | 564 | |
54cdfdb4 TG |
565 | /* |
566 | * When the callback is running, we do not reprogram the clock event | |
567 | * device. The timer callback is either running on a different CPU or | |
3a4fa0a2 | 568 | * the callback is executed in the hrtimer_interrupt context. The |
54cdfdb4 TG |
569 | * reprogramming is handled either by the softirq, which called the |
570 | * callback or at the end of the hrtimer_interrupt. | |
571 | */ | |
572 | if (hrtimer_callback_running(timer)) | |
573 | return 0; | |
574 | ||
63070a79 TG |
575 | /* |
576 | * CLOCK_REALTIME timer might be requested with an absolute | |
577 | * expiry time which is less than base->offset. Nothing wrong | |
578 | * about that, just avoid to call into the tick code, which | |
579 | * has now objections against negative expiry values. | |
580 | */ | |
581 | if (expires.tv64 < 0) | |
582 | return -ETIME; | |
583 | ||
41d2e494 TG |
584 | if (expires.tv64 >= cpu_base->expires_next.tv64) |
585 | return 0; | |
586 | ||
587 | /* | |
588 | * If a hang was detected in the last timer interrupt then we | |
589 | * do not schedule a timer which is earlier than the expiry | |
590 | * which we enforced in the hang detection. We want the system | |
591 | * to make progress. | |
592 | */ | |
593 | if (cpu_base->hang_detected) | |
54cdfdb4 TG |
594 | return 0; |
595 | ||
596 | /* | |
597 | * Clockevents returns -ETIME, when the event was in the past. | |
598 | */ | |
599 | res = tick_program_event(expires, 0); | |
600 | if (!IS_ERR_VALUE(res)) | |
41d2e494 | 601 | cpu_base->expires_next = expires; |
54cdfdb4 TG |
602 | return res; |
603 | } | |
604 | ||
605 | ||
606 | /* | |
607 | * Retrigger next event is called after clock was set | |
608 | * | |
609 | * Called with interrupts disabled via on_each_cpu() | |
610 | */ | |
611 | static void retrigger_next_event(void *arg) | |
612 | { | |
613 | struct hrtimer_cpu_base *base; | |
8ab4351a | 614 | struct timespec realtime_offset, wtm; |
54cdfdb4 TG |
615 | unsigned long seq; |
616 | ||
617 | if (!hrtimer_hres_active()) | |
618 | return; | |
619 | ||
620 | do { | |
621 | seq = read_seqbegin(&xtime_lock); | |
8ab4351a | 622 | wtm = __get_wall_to_monotonic(); |
54cdfdb4 | 623 | } while (read_seqretry(&xtime_lock, seq)); |
8ab4351a | 624 | set_normalized_timespec(&realtime_offset, -wtm.tv_sec, -wtm.tv_nsec); |
54cdfdb4 TG |
625 | |
626 | base = &__get_cpu_var(hrtimer_bases); | |
627 | ||
628 | /* Adjust CLOCK_REALTIME offset */ | |
ecb49d1a | 629 | raw_spin_lock(&base->lock); |
54cdfdb4 TG |
630 | base->clock_base[CLOCK_REALTIME].offset = |
631 | timespec_to_ktime(realtime_offset); | |
632 | ||
7403f41f | 633 | hrtimer_force_reprogram(base, 0); |
ecb49d1a | 634 | raw_spin_unlock(&base->lock); |
54cdfdb4 TG |
635 | } |
636 | ||
637 | /* | |
638 | * Clock realtime was set | |
639 | * | |
640 | * Change the offset of the realtime clock vs. the monotonic | |
641 | * clock. | |
642 | * | |
643 | * We might have to reprogram the high resolution timer interrupt. On | |
644 | * SMP we call the architecture specific code to retrigger _all_ high | |
645 | * resolution timer interrupts. On UP we just disable interrupts and | |
646 | * call the high resolution interrupt code. | |
647 | */ | |
648 | void clock_was_set(void) | |
649 | { | |
650 | /* Retrigger the CPU local events everywhere */ | |
15c8b6c1 | 651 | on_each_cpu(retrigger_next_event, NULL, 1); |
54cdfdb4 TG |
652 | } |
653 | ||
995f054f IM |
654 | /* |
655 | * During resume we might have to reprogram the high resolution timer | |
656 | * interrupt (on the local CPU): | |
657 | */ | |
658 | void hres_timers_resume(void) | |
659 | { | |
1d4a7f1c PZ |
660 | WARN_ONCE(!irqs_disabled(), |
661 | KERN_INFO "hres_timers_resume() called with IRQs enabled!"); | |
662 | ||
995f054f IM |
663 | retrigger_next_event(NULL); |
664 | } | |
665 | ||
54cdfdb4 TG |
666 | /* |
667 | * Initialize the high resolution related parts of cpu_base | |
668 | */ | |
669 | static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) | |
670 | { | |
671 | base->expires_next.tv64 = KTIME_MAX; | |
672 | base->hres_active = 0; | |
54cdfdb4 TG |
673 | } |
674 | ||
675 | /* | |
676 | * Initialize the high resolution related parts of a hrtimer | |
677 | */ | |
678 | static inline void hrtimer_init_timer_hres(struct hrtimer *timer) | |
679 | { | |
54cdfdb4 TG |
680 | } |
681 | ||
ca109491 | 682 | |
54cdfdb4 TG |
683 | /* |
684 | * When High resolution timers are active, try to reprogram. Note, that in case | |
685 | * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry | |
686 | * check happens. The timer gets enqueued into the rbtree. The reprogramming | |
687 | * and expiry check is done in the hrtimer_interrupt or in the softirq. | |
688 | */ | |
689 | static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer, | |
7f1e2ca9 PZ |
690 | struct hrtimer_clock_base *base, |
691 | int wakeup) | |
54cdfdb4 TG |
692 | { |
693 | if (base->cpu_base->hres_active && hrtimer_reprogram(timer, base)) { | |
7f1e2ca9 | 694 | if (wakeup) { |
ecb49d1a | 695 | raw_spin_unlock(&base->cpu_base->lock); |
7f1e2ca9 | 696 | raise_softirq_irqoff(HRTIMER_SOFTIRQ); |
ecb49d1a | 697 | raw_spin_lock(&base->cpu_base->lock); |
7f1e2ca9 PZ |
698 | } else |
699 | __raise_softirq_irqoff(HRTIMER_SOFTIRQ); | |
700 | ||
ca109491 | 701 | return 1; |
54cdfdb4 | 702 | } |
7f1e2ca9 | 703 | |
54cdfdb4 TG |
704 | return 0; |
705 | } | |
706 | ||
707 | /* | |
708 | * Switch to high resolution mode | |
709 | */ | |
f8953856 | 710 | static int hrtimer_switch_to_hres(void) |
54cdfdb4 | 711 | { |
820de5c3 IM |
712 | int cpu = smp_processor_id(); |
713 | struct hrtimer_cpu_base *base = &per_cpu(hrtimer_bases, cpu); | |
54cdfdb4 TG |
714 | unsigned long flags; |
715 | ||
716 | if (base->hres_active) | |
f8953856 | 717 | return 1; |
54cdfdb4 TG |
718 | |
719 | local_irq_save(flags); | |
720 | ||
721 | if (tick_init_highres()) { | |
722 | local_irq_restore(flags); | |
820de5c3 IM |
723 | printk(KERN_WARNING "Could not switch to high resolution " |
724 | "mode on CPU %d\n", cpu); | |
f8953856 | 725 | return 0; |
54cdfdb4 TG |
726 | } |
727 | base->hres_active = 1; | |
728 | base->clock_base[CLOCK_REALTIME].resolution = KTIME_HIGH_RES; | |
729 | base->clock_base[CLOCK_MONOTONIC].resolution = KTIME_HIGH_RES; | |
730 | ||
731 | tick_setup_sched_timer(); | |
732 | ||
733 | /* "Retrigger" the interrupt to get things going */ | |
734 | retrigger_next_event(NULL); | |
735 | local_irq_restore(flags); | |
f8953856 | 736 | return 1; |
54cdfdb4 TG |
737 | } |
738 | ||
739 | #else | |
740 | ||
741 | static inline int hrtimer_hres_active(void) { return 0; } | |
742 | static inline int hrtimer_is_hres_enabled(void) { return 0; } | |
f8953856 | 743 | static inline int hrtimer_switch_to_hres(void) { return 0; } |
7403f41f AC |
744 | static inline void |
745 | hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { } | |
54cdfdb4 | 746 | static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer, |
7f1e2ca9 PZ |
747 | struct hrtimer_clock_base *base, |
748 | int wakeup) | |
54cdfdb4 TG |
749 | { |
750 | return 0; | |
751 | } | |
54cdfdb4 TG |
752 | static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { } |
753 | static inline void hrtimer_init_timer_hres(struct hrtimer *timer) { } | |
754 | ||
755 | #endif /* CONFIG_HIGH_RES_TIMERS */ | |
756 | ||
5f201907 | 757 | static inline void timer_stats_hrtimer_set_start_info(struct hrtimer *timer) |
82f67cd9 | 758 | { |
5f201907 | 759 | #ifdef CONFIG_TIMER_STATS |
82f67cd9 IM |
760 | if (timer->start_site) |
761 | return; | |
5f201907 | 762 | timer->start_site = __builtin_return_address(0); |
82f67cd9 IM |
763 | memcpy(timer->start_comm, current->comm, TASK_COMM_LEN); |
764 | timer->start_pid = current->pid; | |
5f201907 HC |
765 | #endif |
766 | } | |
767 | ||
768 | static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer *timer) | |
769 | { | |
770 | #ifdef CONFIG_TIMER_STATS | |
771 | timer->start_site = NULL; | |
772 | #endif | |
82f67cd9 | 773 | } |
5f201907 HC |
774 | |
775 | static inline void timer_stats_account_hrtimer(struct hrtimer *timer) | |
776 | { | |
777 | #ifdef CONFIG_TIMER_STATS | |
778 | if (likely(!timer_stats_active)) | |
779 | return; | |
780 | timer_stats_update_stats(timer, timer->start_pid, timer->start_site, | |
781 | timer->function, timer->start_comm, 0); | |
82f67cd9 | 782 | #endif |
5f201907 | 783 | } |
82f67cd9 | 784 | |
c0a31329 | 785 | /* |
6506f2aa | 786 | * Counterpart to lock_hrtimer_base above: |
c0a31329 TG |
787 | */ |
788 | static inline | |
789 | void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) | |
790 | { | |
ecb49d1a | 791 | raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags); |
c0a31329 TG |
792 | } |
793 | ||
794 | /** | |
795 | * hrtimer_forward - forward the timer expiry | |
c0a31329 | 796 | * @timer: hrtimer to forward |
44f21475 | 797 | * @now: forward past this time |
c0a31329 TG |
798 | * @interval: the interval to forward |
799 | * | |
800 | * Forward the timer expiry so it will expire in the future. | |
8dca6f33 | 801 | * Returns the number of overruns. |
c0a31329 | 802 | */ |
4d672e7a | 803 | u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval) |
c0a31329 | 804 | { |
4d672e7a | 805 | u64 orun = 1; |
44f21475 | 806 | ktime_t delta; |
c0a31329 | 807 | |
cc584b21 | 808 | delta = ktime_sub(now, hrtimer_get_expires(timer)); |
c0a31329 TG |
809 | |
810 | if (delta.tv64 < 0) | |
811 | return 0; | |
812 | ||
c9db4fa1 TG |
813 | if (interval.tv64 < timer->base->resolution.tv64) |
814 | interval.tv64 = timer->base->resolution.tv64; | |
815 | ||
c0a31329 | 816 | if (unlikely(delta.tv64 >= interval.tv64)) { |
df869b63 | 817 | s64 incr = ktime_to_ns(interval); |
c0a31329 TG |
818 | |
819 | orun = ktime_divns(delta, incr); | |
cc584b21 AV |
820 | hrtimer_add_expires_ns(timer, incr * orun); |
821 | if (hrtimer_get_expires_tv64(timer) > now.tv64) | |
c0a31329 TG |
822 | return orun; |
823 | /* | |
824 | * This (and the ktime_add() below) is the | |
825 | * correction for exact: | |
826 | */ | |
827 | orun++; | |
828 | } | |
cc584b21 | 829 | hrtimer_add_expires(timer, interval); |
c0a31329 TG |
830 | |
831 | return orun; | |
832 | } | |
6bdb6b62 | 833 | EXPORT_SYMBOL_GPL(hrtimer_forward); |
c0a31329 TG |
834 | |
835 | /* | |
836 | * enqueue_hrtimer - internal function to (re)start a timer | |
837 | * | |
838 | * The timer is inserted in expiry order. Insertion into the | |
839 | * red black tree is O(log(n)). Must hold the base lock. | |
a6037b61 PZ |
840 | * |
841 | * Returns 1 when the new timer is the leftmost timer in the tree. | |
c0a31329 | 842 | */ |
a6037b61 PZ |
843 | static int enqueue_hrtimer(struct hrtimer *timer, |
844 | struct hrtimer_clock_base *base) | |
c0a31329 | 845 | { |
c6a2a177 | 846 | debug_activate(timer); |
237fc6e7 | 847 | |
998adc3d | 848 | timerqueue_add(&base->active, &timer->node); |
c0a31329 | 849 | |
303e967f TG |
850 | /* |
851 | * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the | |
852 | * state of a possibly running callback. | |
853 | */ | |
854 | timer->state |= HRTIMER_STATE_ENQUEUED; | |
a6037b61 | 855 | |
998adc3d | 856 | return (&timer->node == base->active.next); |
288867ec | 857 | } |
c0a31329 TG |
858 | |
859 | /* | |
860 | * __remove_hrtimer - internal function to remove a timer | |
861 | * | |
862 | * Caller must hold the base lock. | |
54cdfdb4 TG |
863 | * |
864 | * High resolution timer mode reprograms the clock event device when the | |
865 | * timer is the one which expires next. The caller can disable this by setting | |
866 | * reprogram to zero. This is useful, when the context does a reprogramming | |
867 | * anyway (e.g. timer interrupt) | |
c0a31329 | 868 | */ |
3c8aa39d | 869 | static void __remove_hrtimer(struct hrtimer *timer, |
303e967f | 870 | struct hrtimer_clock_base *base, |
54cdfdb4 | 871 | unsigned long newstate, int reprogram) |
c0a31329 | 872 | { |
7403f41f AC |
873 | if (!(timer->state & HRTIMER_STATE_ENQUEUED)) |
874 | goto out; | |
875 | ||
998adc3d | 876 | if (&timer->node == timerqueue_getnext(&base->active)) { |
7403f41f AC |
877 | #ifdef CONFIG_HIGH_RES_TIMERS |
878 | /* Reprogram the clock event device. if enabled */ | |
879 | if (reprogram && hrtimer_hres_active()) { | |
880 | ktime_t expires; | |
881 | ||
882 | expires = ktime_sub(hrtimer_get_expires(timer), | |
883 | base->offset); | |
884 | if (base->cpu_base->expires_next.tv64 == expires.tv64) | |
885 | hrtimer_force_reprogram(base->cpu_base, 1); | |
54cdfdb4 | 886 | } |
7403f41f | 887 | #endif |
54cdfdb4 | 888 | } |
998adc3d | 889 | timerqueue_del(&base->active, &timer->node); |
7403f41f | 890 | out: |
303e967f | 891 | timer->state = newstate; |
c0a31329 TG |
892 | } |
893 | ||
894 | /* | |
895 | * remove hrtimer, called with base lock held | |
896 | */ | |
897 | static inline int | |
3c8aa39d | 898 | remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base) |
c0a31329 | 899 | { |
303e967f | 900 | if (hrtimer_is_queued(timer)) { |
54cdfdb4 TG |
901 | int reprogram; |
902 | ||
903 | /* | |
904 | * Remove the timer and force reprogramming when high | |
905 | * resolution mode is active and the timer is on the current | |
906 | * CPU. If we remove a timer on another CPU, reprogramming is | |
907 | * skipped. The interrupt event on this CPU is fired and | |
908 | * reprogramming happens in the interrupt handler. This is a | |
909 | * rare case and less expensive than a smp call. | |
910 | */ | |
c6a2a177 | 911 | debug_deactivate(timer); |
82f67cd9 | 912 | timer_stats_hrtimer_clear_start_info(timer); |
54cdfdb4 TG |
913 | reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases); |
914 | __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE, | |
915 | reprogram); | |
c0a31329 TG |
916 | return 1; |
917 | } | |
918 | return 0; | |
919 | } | |
920 | ||
7f1e2ca9 PZ |
921 | int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, |
922 | unsigned long delta_ns, const enum hrtimer_mode mode, | |
923 | int wakeup) | |
c0a31329 | 924 | { |
3c8aa39d | 925 | struct hrtimer_clock_base *base, *new_base; |
c0a31329 | 926 | unsigned long flags; |
a6037b61 | 927 | int ret, leftmost; |
c0a31329 TG |
928 | |
929 | base = lock_hrtimer_base(timer, &flags); | |
930 | ||
931 | /* Remove an active timer from the queue: */ | |
932 | ret = remove_hrtimer(timer, base); | |
933 | ||
934 | /* Switch the timer base, if necessary: */ | |
597d0275 | 935 | new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED); |
c0a31329 | 936 | |
597d0275 | 937 | if (mode & HRTIMER_MODE_REL) { |
5a7780e7 | 938 | tim = ktime_add_safe(tim, new_base->get_time()); |
06027bdd IM |
939 | /* |
940 | * CONFIG_TIME_LOW_RES is a temporary way for architectures | |
941 | * to signal that they simply return xtime in | |
942 | * do_gettimeoffset(). In this case we want to round up by | |
943 | * resolution when starting a relative timer, to avoid short | |
944 | * timeouts. This will go away with the GTOD framework. | |
945 | */ | |
946 | #ifdef CONFIG_TIME_LOW_RES | |
5a7780e7 | 947 | tim = ktime_add_safe(tim, base->resolution); |
06027bdd IM |
948 | #endif |
949 | } | |
237fc6e7 | 950 | |
da8f2e17 | 951 | hrtimer_set_expires_range_ns(timer, tim, delta_ns); |
c0a31329 | 952 | |
82f67cd9 IM |
953 | timer_stats_hrtimer_set_start_info(timer); |
954 | ||
a6037b61 PZ |
955 | leftmost = enqueue_hrtimer(timer, new_base); |
956 | ||
935c631d IM |
957 | /* |
958 | * Only allow reprogramming if the new base is on this CPU. | |
959 | * (it might still be on another CPU if the timer was pending) | |
a6037b61 PZ |
960 | * |
961 | * XXX send_remote_softirq() ? | |
935c631d | 962 | */ |
a6037b61 | 963 | if (leftmost && new_base->cpu_base == &__get_cpu_var(hrtimer_bases)) |
7f1e2ca9 | 964 | hrtimer_enqueue_reprogram(timer, new_base, wakeup); |
c0a31329 TG |
965 | |
966 | unlock_hrtimer_base(timer, &flags); | |
967 | ||
968 | return ret; | |
969 | } | |
7f1e2ca9 PZ |
970 | |
971 | /** | |
972 | * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU | |
973 | * @timer: the timer to be added | |
974 | * @tim: expiry time | |
975 | * @delta_ns: "slack" range for the timer | |
976 | * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL) | |
977 | * | |
978 | * Returns: | |
979 | * 0 on success | |
980 | * 1 when the timer was active | |
981 | */ | |
982 | int hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, | |
983 | unsigned long delta_ns, const enum hrtimer_mode mode) | |
984 | { | |
985 | return __hrtimer_start_range_ns(timer, tim, delta_ns, mode, 1); | |
986 | } | |
da8f2e17 AV |
987 | EXPORT_SYMBOL_GPL(hrtimer_start_range_ns); |
988 | ||
989 | /** | |
e1dd7bc5 | 990 | * hrtimer_start - (re)start an hrtimer on the current CPU |
da8f2e17 AV |
991 | * @timer: the timer to be added |
992 | * @tim: expiry time | |
993 | * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL) | |
994 | * | |
995 | * Returns: | |
996 | * 0 on success | |
997 | * 1 when the timer was active | |
998 | */ | |
999 | int | |
1000 | hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode) | |
1001 | { | |
7f1e2ca9 | 1002 | return __hrtimer_start_range_ns(timer, tim, 0, mode, 1); |
da8f2e17 | 1003 | } |
8d16b764 | 1004 | EXPORT_SYMBOL_GPL(hrtimer_start); |
c0a31329 | 1005 | |
da8f2e17 | 1006 | |
c0a31329 TG |
1007 | /** |
1008 | * hrtimer_try_to_cancel - try to deactivate a timer | |
c0a31329 TG |
1009 | * @timer: hrtimer to stop |
1010 | * | |
1011 | * Returns: | |
1012 | * 0 when the timer was not active | |
1013 | * 1 when the timer was active | |
1014 | * -1 when the timer is currently excuting the callback function and | |
fa9799e3 | 1015 | * cannot be stopped |
c0a31329 TG |
1016 | */ |
1017 | int hrtimer_try_to_cancel(struct hrtimer *timer) | |
1018 | { | |
3c8aa39d | 1019 | struct hrtimer_clock_base *base; |
c0a31329 TG |
1020 | unsigned long flags; |
1021 | int ret = -1; | |
1022 | ||
1023 | base = lock_hrtimer_base(timer, &flags); | |
1024 | ||
303e967f | 1025 | if (!hrtimer_callback_running(timer)) |
c0a31329 TG |
1026 | ret = remove_hrtimer(timer, base); |
1027 | ||
1028 | unlock_hrtimer_base(timer, &flags); | |
1029 | ||
1030 | return ret; | |
1031 | ||
1032 | } | |
8d16b764 | 1033 | EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel); |
c0a31329 TG |
1034 | |
1035 | /** | |
1036 | * hrtimer_cancel - cancel a timer and wait for the handler to finish. | |
c0a31329 TG |
1037 | * @timer: the timer to be cancelled |
1038 | * | |
1039 | * Returns: | |
1040 | * 0 when the timer was not active | |
1041 | * 1 when the timer was active | |
1042 | */ | |
1043 | int hrtimer_cancel(struct hrtimer *timer) | |
1044 | { | |
1045 | for (;;) { | |
1046 | int ret = hrtimer_try_to_cancel(timer); | |
1047 | ||
1048 | if (ret >= 0) | |
1049 | return ret; | |
5ef37b19 | 1050 | cpu_relax(); |
c0a31329 TG |
1051 | } |
1052 | } | |
8d16b764 | 1053 | EXPORT_SYMBOL_GPL(hrtimer_cancel); |
c0a31329 TG |
1054 | |
1055 | /** | |
1056 | * hrtimer_get_remaining - get remaining time for the timer | |
c0a31329 TG |
1057 | * @timer: the timer to read |
1058 | */ | |
1059 | ktime_t hrtimer_get_remaining(const struct hrtimer *timer) | |
1060 | { | |
3c8aa39d | 1061 | struct hrtimer_clock_base *base; |
c0a31329 TG |
1062 | unsigned long flags; |
1063 | ktime_t rem; | |
1064 | ||
1065 | base = lock_hrtimer_base(timer, &flags); | |
cc584b21 | 1066 | rem = hrtimer_expires_remaining(timer); |
c0a31329 TG |
1067 | unlock_hrtimer_base(timer, &flags); |
1068 | ||
1069 | return rem; | |
1070 | } | |
8d16b764 | 1071 | EXPORT_SYMBOL_GPL(hrtimer_get_remaining); |
c0a31329 | 1072 | |
ee9c5785 | 1073 | #ifdef CONFIG_NO_HZ |
69239749 TL |
1074 | /** |
1075 | * hrtimer_get_next_event - get the time until next expiry event | |
1076 | * | |
1077 | * Returns the delta to the next expiry event or KTIME_MAX if no timer | |
1078 | * is pending. | |
1079 | */ | |
1080 | ktime_t hrtimer_get_next_event(void) | |
1081 | { | |
3c8aa39d TG |
1082 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
1083 | struct hrtimer_clock_base *base = cpu_base->clock_base; | |
69239749 TL |
1084 | ktime_t delta, mindelta = { .tv64 = KTIME_MAX }; |
1085 | unsigned long flags; | |
1086 | int i; | |
1087 | ||
ecb49d1a | 1088 | raw_spin_lock_irqsave(&cpu_base->lock, flags); |
3c8aa39d | 1089 | |
54cdfdb4 TG |
1090 | if (!hrtimer_hres_active()) { |
1091 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) { | |
1092 | struct hrtimer *timer; | |
998adc3d | 1093 | struct timerqueue_node *next; |
69239749 | 1094 | |
998adc3d JS |
1095 | next = timerqueue_getnext(&base->active); |
1096 | if (!next) | |
54cdfdb4 | 1097 | continue; |
3c8aa39d | 1098 | |
998adc3d | 1099 | timer = container_of(next, struct hrtimer, node); |
cc584b21 | 1100 | delta.tv64 = hrtimer_get_expires_tv64(timer); |
54cdfdb4 TG |
1101 | delta = ktime_sub(delta, base->get_time()); |
1102 | if (delta.tv64 < mindelta.tv64) | |
1103 | mindelta.tv64 = delta.tv64; | |
1104 | } | |
69239749 | 1105 | } |
3c8aa39d | 1106 | |
ecb49d1a | 1107 | raw_spin_unlock_irqrestore(&cpu_base->lock, flags); |
3c8aa39d | 1108 | |
69239749 TL |
1109 | if (mindelta.tv64 < 0) |
1110 | mindelta.tv64 = 0; | |
1111 | return mindelta; | |
1112 | } | |
1113 | #endif | |
1114 | ||
237fc6e7 TG |
1115 | static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id, |
1116 | enum hrtimer_mode mode) | |
c0a31329 | 1117 | { |
3c8aa39d | 1118 | struct hrtimer_cpu_base *cpu_base; |
c0a31329 | 1119 | |
7978672c GA |
1120 | memset(timer, 0, sizeof(struct hrtimer)); |
1121 | ||
3c8aa39d | 1122 | cpu_base = &__raw_get_cpu_var(hrtimer_bases); |
c0a31329 | 1123 | |
c9cb2e3d | 1124 | if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS) |
7978672c GA |
1125 | clock_id = CLOCK_MONOTONIC; |
1126 | ||
3c8aa39d | 1127 | timer->base = &cpu_base->clock_base[clock_id]; |
54cdfdb4 | 1128 | hrtimer_init_timer_hres(timer); |
998adc3d | 1129 | timerqueue_init(&timer->node); |
82f67cd9 IM |
1130 | |
1131 | #ifdef CONFIG_TIMER_STATS | |
1132 | timer->start_site = NULL; | |
1133 | timer->start_pid = -1; | |
1134 | memset(timer->start_comm, 0, TASK_COMM_LEN); | |
1135 | #endif | |
c0a31329 | 1136 | } |
237fc6e7 TG |
1137 | |
1138 | /** | |
1139 | * hrtimer_init - initialize a timer to the given clock | |
1140 | * @timer: the timer to be initialized | |
1141 | * @clock_id: the clock to be used | |
1142 | * @mode: timer mode abs/rel | |
1143 | */ | |
1144 | void hrtimer_init(struct hrtimer *timer, clockid_t clock_id, | |
1145 | enum hrtimer_mode mode) | |
1146 | { | |
c6a2a177 | 1147 | debug_init(timer, clock_id, mode); |
237fc6e7 TG |
1148 | __hrtimer_init(timer, clock_id, mode); |
1149 | } | |
8d16b764 | 1150 | EXPORT_SYMBOL_GPL(hrtimer_init); |
c0a31329 TG |
1151 | |
1152 | /** | |
1153 | * hrtimer_get_res - get the timer resolution for a clock | |
c0a31329 TG |
1154 | * @which_clock: which clock to query |
1155 | * @tp: pointer to timespec variable to store the resolution | |
1156 | * | |
72fd4a35 RD |
1157 | * Store the resolution of the clock selected by @which_clock in the |
1158 | * variable pointed to by @tp. | |
c0a31329 TG |
1159 | */ |
1160 | int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp) | |
1161 | { | |
3c8aa39d | 1162 | struct hrtimer_cpu_base *cpu_base; |
c0a31329 | 1163 | |
3c8aa39d TG |
1164 | cpu_base = &__raw_get_cpu_var(hrtimer_bases); |
1165 | *tp = ktime_to_timespec(cpu_base->clock_base[which_clock].resolution); | |
c0a31329 TG |
1166 | |
1167 | return 0; | |
1168 | } | |
8d16b764 | 1169 | EXPORT_SYMBOL_GPL(hrtimer_get_res); |
c0a31329 | 1170 | |
c6a2a177 | 1171 | static void __run_hrtimer(struct hrtimer *timer, ktime_t *now) |
d3d74453 PZ |
1172 | { |
1173 | struct hrtimer_clock_base *base = timer->base; | |
1174 | struct hrtimer_cpu_base *cpu_base = base->cpu_base; | |
1175 | enum hrtimer_restart (*fn)(struct hrtimer *); | |
1176 | int restart; | |
1177 | ||
ca109491 PZ |
1178 | WARN_ON(!irqs_disabled()); |
1179 | ||
c6a2a177 | 1180 | debug_deactivate(timer); |
d3d74453 PZ |
1181 | __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0); |
1182 | timer_stats_account_hrtimer(timer); | |
d3d74453 | 1183 | fn = timer->function; |
ca109491 PZ |
1184 | |
1185 | /* | |
1186 | * Because we run timers from hardirq context, there is no chance | |
1187 | * they get migrated to another cpu, therefore its safe to unlock | |
1188 | * the timer base. | |
1189 | */ | |
ecb49d1a | 1190 | raw_spin_unlock(&cpu_base->lock); |
c6a2a177 | 1191 | trace_hrtimer_expire_entry(timer, now); |
ca109491 | 1192 | restart = fn(timer); |
c6a2a177 | 1193 | trace_hrtimer_expire_exit(timer); |
ecb49d1a | 1194 | raw_spin_lock(&cpu_base->lock); |
d3d74453 PZ |
1195 | |
1196 | /* | |
e3f1d883 TG |
1197 | * Note: We clear the CALLBACK bit after enqueue_hrtimer and |
1198 | * we do not reprogramm the event hardware. Happens either in | |
1199 | * hrtimer_start_range_ns() or in hrtimer_interrupt() | |
d3d74453 PZ |
1200 | */ |
1201 | if (restart != HRTIMER_NORESTART) { | |
1202 | BUG_ON(timer->state != HRTIMER_STATE_CALLBACK); | |
a6037b61 | 1203 | enqueue_hrtimer(timer, base); |
d3d74453 PZ |
1204 | } |
1205 | timer->state &= ~HRTIMER_STATE_CALLBACK; | |
1206 | } | |
1207 | ||
54cdfdb4 TG |
1208 | #ifdef CONFIG_HIGH_RES_TIMERS |
1209 | ||
1210 | /* | |
1211 | * High resolution timer interrupt | |
1212 | * Called with interrupts disabled | |
1213 | */ | |
1214 | void hrtimer_interrupt(struct clock_event_device *dev) | |
1215 | { | |
1216 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); | |
1217 | struct hrtimer_clock_base *base; | |
41d2e494 TG |
1218 | ktime_t expires_next, now, entry_time, delta; |
1219 | int i, retries = 0; | |
54cdfdb4 TG |
1220 | |
1221 | BUG_ON(!cpu_base->hres_active); | |
1222 | cpu_base->nr_events++; | |
1223 | dev->next_event.tv64 = KTIME_MAX; | |
1224 | ||
41d2e494 TG |
1225 | entry_time = now = ktime_get(); |
1226 | retry: | |
54cdfdb4 TG |
1227 | expires_next.tv64 = KTIME_MAX; |
1228 | ||
ecb49d1a | 1229 | raw_spin_lock(&cpu_base->lock); |
6ff7041d TG |
1230 | /* |
1231 | * We set expires_next to KTIME_MAX here with cpu_base->lock | |
1232 | * held to prevent that a timer is enqueued in our queue via | |
1233 | * the migration code. This does not affect enqueueing of | |
1234 | * timers which run their callback and need to be requeued on | |
1235 | * this CPU. | |
1236 | */ | |
1237 | cpu_base->expires_next.tv64 = KTIME_MAX; | |
1238 | ||
54cdfdb4 TG |
1239 | base = cpu_base->clock_base; |
1240 | ||
1241 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { | |
1242 | ktime_t basenow; | |
998adc3d | 1243 | struct timerqueue_node *node; |
54cdfdb4 | 1244 | |
54cdfdb4 TG |
1245 | basenow = ktime_add(now, base->offset); |
1246 | ||
998adc3d | 1247 | while ((node = timerqueue_getnext(&base->active))) { |
54cdfdb4 TG |
1248 | struct hrtimer *timer; |
1249 | ||
998adc3d | 1250 | timer = container_of(node, struct hrtimer, node); |
54cdfdb4 | 1251 | |
654c8e0b AV |
1252 | /* |
1253 | * The immediate goal for using the softexpires is | |
1254 | * minimizing wakeups, not running timers at the | |
1255 | * earliest interrupt after their soft expiration. | |
1256 | * This allows us to avoid using a Priority Search | |
1257 | * Tree, which can answer a stabbing querry for | |
1258 | * overlapping intervals and instead use the simple | |
1259 | * BST we already have. | |
1260 | * We don't add extra wakeups by delaying timers that | |
1261 | * are right-of a not yet expired timer, because that | |
1262 | * timer will have to trigger a wakeup anyway. | |
1263 | */ | |
1264 | ||
1265 | if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer)) { | |
54cdfdb4 TG |
1266 | ktime_t expires; |
1267 | ||
cc584b21 | 1268 | expires = ktime_sub(hrtimer_get_expires(timer), |
54cdfdb4 TG |
1269 | base->offset); |
1270 | if (expires.tv64 < expires_next.tv64) | |
1271 | expires_next = expires; | |
1272 | break; | |
1273 | } | |
1274 | ||
c6a2a177 | 1275 | __run_hrtimer(timer, &basenow); |
54cdfdb4 | 1276 | } |
54cdfdb4 TG |
1277 | base++; |
1278 | } | |
1279 | ||
6ff7041d TG |
1280 | /* |
1281 | * Store the new expiry value so the migration code can verify | |
1282 | * against it. | |
1283 | */ | |
54cdfdb4 | 1284 | cpu_base->expires_next = expires_next; |
ecb49d1a | 1285 | raw_spin_unlock(&cpu_base->lock); |
54cdfdb4 TG |
1286 | |
1287 | /* Reprogramming necessary ? */ | |
41d2e494 TG |
1288 | if (expires_next.tv64 == KTIME_MAX || |
1289 | !tick_program_event(expires_next, 0)) { | |
1290 | cpu_base->hang_detected = 0; | |
1291 | return; | |
54cdfdb4 | 1292 | } |
41d2e494 TG |
1293 | |
1294 | /* | |
1295 | * The next timer was already expired due to: | |
1296 | * - tracing | |
1297 | * - long lasting callbacks | |
1298 | * - being scheduled away when running in a VM | |
1299 | * | |
1300 | * We need to prevent that we loop forever in the hrtimer | |
1301 | * interrupt routine. We give it 3 attempts to avoid | |
1302 | * overreacting on some spurious event. | |
1303 | */ | |
1304 | now = ktime_get(); | |
1305 | cpu_base->nr_retries++; | |
1306 | if (++retries < 3) | |
1307 | goto retry; | |
1308 | /* | |
1309 | * Give the system a chance to do something else than looping | |
1310 | * here. We stored the entry time, so we know exactly how long | |
1311 | * we spent here. We schedule the next event this amount of | |
1312 | * time away. | |
1313 | */ | |
1314 | cpu_base->nr_hangs++; | |
1315 | cpu_base->hang_detected = 1; | |
1316 | delta = ktime_sub(now, entry_time); | |
1317 | if (delta.tv64 > cpu_base->max_hang_time.tv64) | |
1318 | cpu_base->max_hang_time = delta; | |
1319 | /* | |
1320 | * Limit it to a sensible value as we enforce a longer | |
1321 | * delay. Give the CPU at least 100ms to catch up. | |
1322 | */ | |
1323 | if (delta.tv64 > 100 * NSEC_PER_MSEC) | |
1324 | expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC); | |
1325 | else | |
1326 | expires_next = ktime_add(now, delta); | |
1327 | tick_program_event(expires_next, 1); | |
1328 | printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n", | |
1329 | ktime_to_ns(delta)); | |
54cdfdb4 TG |
1330 | } |
1331 | ||
8bdec955 TG |
1332 | /* |
1333 | * local version of hrtimer_peek_ahead_timers() called with interrupts | |
1334 | * disabled. | |
1335 | */ | |
1336 | static void __hrtimer_peek_ahead_timers(void) | |
1337 | { | |
1338 | struct tick_device *td; | |
1339 | ||
1340 | if (!hrtimer_hres_active()) | |
1341 | return; | |
1342 | ||
1343 | td = &__get_cpu_var(tick_cpu_device); | |
1344 | if (td && td->evtdev) | |
1345 | hrtimer_interrupt(td->evtdev); | |
1346 | } | |
1347 | ||
2e94d1f7 AV |
1348 | /** |
1349 | * hrtimer_peek_ahead_timers -- run soft-expired timers now | |
1350 | * | |
1351 | * hrtimer_peek_ahead_timers will peek at the timer queue of | |
1352 | * the current cpu and check if there are any timers for which | |
1353 | * the soft expires time has passed. If any such timers exist, | |
1354 | * they are run immediately and then removed from the timer queue. | |
1355 | * | |
1356 | */ | |
1357 | void hrtimer_peek_ahead_timers(void) | |
1358 | { | |
643bdf68 | 1359 | unsigned long flags; |
dc4304f7 | 1360 | |
2e94d1f7 | 1361 | local_irq_save(flags); |
8bdec955 | 1362 | __hrtimer_peek_ahead_timers(); |
2e94d1f7 AV |
1363 | local_irq_restore(flags); |
1364 | } | |
1365 | ||
a6037b61 PZ |
1366 | static void run_hrtimer_softirq(struct softirq_action *h) |
1367 | { | |
1368 | hrtimer_peek_ahead_timers(); | |
1369 | } | |
1370 | ||
82c5b7b5 IM |
1371 | #else /* CONFIG_HIGH_RES_TIMERS */ |
1372 | ||
1373 | static inline void __hrtimer_peek_ahead_timers(void) { } | |
1374 | ||
1375 | #endif /* !CONFIG_HIGH_RES_TIMERS */ | |
82f67cd9 | 1376 | |
d3d74453 PZ |
1377 | /* |
1378 | * Called from timer softirq every jiffy, expire hrtimers: | |
1379 | * | |
1380 | * For HRT its the fall back code to run the softirq in the timer | |
1381 | * softirq context in case the hrtimer initialization failed or has | |
1382 | * not been done yet. | |
1383 | */ | |
1384 | void hrtimer_run_pending(void) | |
1385 | { | |
d3d74453 PZ |
1386 | if (hrtimer_hres_active()) |
1387 | return; | |
54cdfdb4 | 1388 | |
d3d74453 PZ |
1389 | /* |
1390 | * This _is_ ugly: We have to check in the softirq context, | |
1391 | * whether we can switch to highres and / or nohz mode. The | |
1392 | * clocksource switch happens in the timer interrupt with | |
1393 | * xtime_lock held. Notification from there only sets the | |
1394 | * check bit in the tick_oneshot code, otherwise we might | |
1395 | * deadlock vs. xtime_lock. | |
1396 | */ | |
1397 | if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) | |
1398 | hrtimer_switch_to_hres(); | |
54cdfdb4 TG |
1399 | } |
1400 | ||
c0a31329 | 1401 | /* |
d3d74453 | 1402 | * Called from hardirq context every jiffy |
c0a31329 | 1403 | */ |
833883d9 | 1404 | void hrtimer_run_queues(void) |
c0a31329 | 1405 | { |
998adc3d | 1406 | struct timerqueue_node *node; |
833883d9 DS |
1407 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
1408 | struct hrtimer_clock_base *base; | |
1409 | int index, gettime = 1; | |
c0a31329 | 1410 | |
833883d9 | 1411 | if (hrtimer_hres_active()) |
3055adda DS |
1412 | return; |
1413 | ||
833883d9 | 1414 | for (index = 0; index < HRTIMER_MAX_CLOCK_BASES; index++) { |
998adc3d | 1415 | struct timerqueue_node *next; |
c0a31329 | 1416 | |
998adc3d JS |
1417 | base = &cpu_base->clock_base[index]; |
1418 | next = timerqueue_getnext(&base->active); | |
1419 | if (!next) | |
d3d74453 | 1420 | continue; |
833883d9 | 1421 | |
d7cfb60c | 1422 | if (gettime) { |
833883d9 DS |
1423 | hrtimer_get_softirq_time(cpu_base); |
1424 | gettime = 0; | |
b75f7a51 | 1425 | } |
d3d74453 | 1426 | |
ecb49d1a | 1427 | raw_spin_lock(&cpu_base->lock); |
c0a31329 | 1428 | |
998adc3d | 1429 | while ((node = next)) { |
833883d9 | 1430 | struct hrtimer *timer; |
54cdfdb4 | 1431 | |
998adc3d | 1432 | timer = container_of(node, struct hrtimer, node); |
cc584b21 AV |
1433 | if (base->softirq_time.tv64 <= |
1434 | hrtimer_get_expires_tv64(timer)) | |
833883d9 DS |
1435 | break; |
1436 | ||
c6a2a177 | 1437 | __run_hrtimer(timer, &base->softirq_time); |
833883d9 | 1438 | } |
ecb49d1a | 1439 | raw_spin_unlock(&cpu_base->lock); |
833883d9 | 1440 | } |
c0a31329 TG |
1441 | } |
1442 | ||
10c94ec1 TG |
1443 | /* |
1444 | * Sleep related functions: | |
1445 | */ | |
c9cb2e3d | 1446 | static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer) |
00362e33 TG |
1447 | { |
1448 | struct hrtimer_sleeper *t = | |
1449 | container_of(timer, struct hrtimer_sleeper, timer); | |
1450 | struct task_struct *task = t->task; | |
1451 | ||
1452 | t->task = NULL; | |
1453 | if (task) | |
1454 | wake_up_process(task); | |
1455 | ||
1456 | return HRTIMER_NORESTART; | |
1457 | } | |
1458 | ||
36c8b586 | 1459 | void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task) |
00362e33 TG |
1460 | { |
1461 | sl->timer.function = hrtimer_wakeup; | |
1462 | sl->task = task; | |
1463 | } | |
2bc481cf | 1464 | EXPORT_SYMBOL_GPL(hrtimer_init_sleeper); |
00362e33 | 1465 | |
669d7868 | 1466 | static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode) |
432569bb | 1467 | { |
669d7868 | 1468 | hrtimer_init_sleeper(t, current); |
10c94ec1 | 1469 | |
432569bb RZ |
1470 | do { |
1471 | set_current_state(TASK_INTERRUPTIBLE); | |
cc584b21 | 1472 | hrtimer_start_expires(&t->timer, mode); |
37bb6cb4 PZ |
1473 | if (!hrtimer_active(&t->timer)) |
1474 | t->task = NULL; | |
432569bb | 1475 | |
54cdfdb4 TG |
1476 | if (likely(t->task)) |
1477 | schedule(); | |
432569bb | 1478 | |
669d7868 | 1479 | hrtimer_cancel(&t->timer); |
c9cb2e3d | 1480 | mode = HRTIMER_MODE_ABS; |
669d7868 TG |
1481 | |
1482 | } while (t->task && !signal_pending(current)); | |
432569bb | 1483 | |
3588a085 PZ |
1484 | __set_current_state(TASK_RUNNING); |
1485 | ||
669d7868 | 1486 | return t->task == NULL; |
10c94ec1 TG |
1487 | } |
1488 | ||
080344b9 ON |
1489 | static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp) |
1490 | { | |
1491 | struct timespec rmt; | |
1492 | ktime_t rem; | |
1493 | ||
cc584b21 | 1494 | rem = hrtimer_expires_remaining(timer); |
080344b9 ON |
1495 | if (rem.tv64 <= 0) |
1496 | return 0; | |
1497 | rmt = ktime_to_timespec(rem); | |
1498 | ||
1499 | if (copy_to_user(rmtp, &rmt, sizeof(*rmtp))) | |
1500 | return -EFAULT; | |
1501 | ||
1502 | return 1; | |
1503 | } | |
1504 | ||
1711ef38 | 1505 | long __sched hrtimer_nanosleep_restart(struct restart_block *restart) |
10c94ec1 | 1506 | { |
669d7868 | 1507 | struct hrtimer_sleeper t; |
080344b9 | 1508 | struct timespec __user *rmtp; |
237fc6e7 | 1509 | int ret = 0; |
10c94ec1 | 1510 | |
237fc6e7 TG |
1511 | hrtimer_init_on_stack(&t.timer, restart->nanosleep.index, |
1512 | HRTIMER_MODE_ABS); | |
cc584b21 | 1513 | hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires); |
10c94ec1 | 1514 | |
c9cb2e3d | 1515 | if (do_nanosleep(&t, HRTIMER_MODE_ABS)) |
237fc6e7 | 1516 | goto out; |
10c94ec1 | 1517 | |
029a07e0 | 1518 | rmtp = restart->nanosleep.rmtp; |
432569bb | 1519 | if (rmtp) { |
237fc6e7 | 1520 | ret = update_rmtp(&t.timer, rmtp); |
080344b9 | 1521 | if (ret <= 0) |
237fc6e7 | 1522 | goto out; |
432569bb | 1523 | } |
10c94ec1 | 1524 | |
10c94ec1 | 1525 | /* The other values in restart are already filled in */ |
237fc6e7 TG |
1526 | ret = -ERESTART_RESTARTBLOCK; |
1527 | out: | |
1528 | destroy_hrtimer_on_stack(&t.timer); | |
1529 | return ret; | |
10c94ec1 TG |
1530 | } |
1531 | ||
080344b9 | 1532 | long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp, |
10c94ec1 TG |
1533 | const enum hrtimer_mode mode, const clockid_t clockid) |
1534 | { | |
1535 | struct restart_block *restart; | |
669d7868 | 1536 | struct hrtimer_sleeper t; |
237fc6e7 | 1537 | int ret = 0; |
3bd01206 AV |
1538 | unsigned long slack; |
1539 | ||
1540 | slack = current->timer_slack_ns; | |
1541 | if (rt_task(current)) | |
1542 | slack = 0; | |
10c94ec1 | 1543 | |
237fc6e7 | 1544 | hrtimer_init_on_stack(&t.timer, clockid, mode); |
3bd01206 | 1545 | hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack); |
432569bb | 1546 | if (do_nanosleep(&t, mode)) |
237fc6e7 | 1547 | goto out; |
10c94ec1 | 1548 | |
7978672c | 1549 | /* Absolute timers do not update the rmtp value and restart: */ |
237fc6e7 TG |
1550 | if (mode == HRTIMER_MODE_ABS) { |
1551 | ret = -ERESTARTNOHAND; | |
1552 | goto out; | |
1553 | } | |
10c94ec1 | 1554 | |
432569bb | 1555 | if (rmtp) { |
237fc6e7 | 1556 | ret = update_rmtp(&t.timer, rmtp); |
080344b9 | 1557 | if (ret <= 0) |
237fc6e7 | 1558 | goto out; |
432569bb | 1559 | } |
10c94ec1 TG |
1560 | |
1561 | restart = ¤t_thread_info()->restart_block; | |
1711ef38 | 1562 | restart->fn = hrtimer_nanosleep_restart; |
029a07e0 TG |
1563 | restart->nanosleep.index = t.timer.base->index; |
1564 | restart->nanosleep.rmtp = rmtp; | |
cc584b21 | 1565 | restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer); |
10c94ec1 | 1566 | |
237fc6e7 TG |
1567 | ret = -ERESTART_RESTARTBLOCK; |
1568 | out: | |
1569 | destroy_hrtimer_on_stack(&t.timer); | |
1570 | return ret; | |
10c94ec1 TG |
1571 | } |
1572 | ||
58fd3aa2 HC |
1573 | SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp, |
1574 | struct timespec __user *, rmtp) | |
6ba1b912 | 1575 | { |
080344b9 | 1576 | struct timespec tu; |
6ba1b912 TG |
1577 | |
1578 | if (copy_from_user(&tu, rqtp, sizeof(tu))) | |
1579 | return -EFAULT; | |
1580 | ||
1581 | if (!timespec_valid(&tu)) | |
1582 | return -EINVAL; | |
1583 | ||
080344b9 | 1584 | return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC); |
6ba1b912 TG |
1585 | } |
1586 | ||
c0a31329 TG |
1587 | /* |
1588 | * Functions related to boot-time initialization: | |
1589 | */ | |
0ec160dd | 1590 | static void __cpuinit init_hrtimers_cpu(int cpu) |
c0a31329 | 1591 | { |
3c8aa39d | 1592 | struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu); |
c0a31329 TG |
1593 | int i; |
1594 | ||
ecb49d1a | 1595 | raw_spin_lock_init(&cpu_base->lock); |
3c8aa39d | 1596 | |
998adc3d | 1597 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
3c8aa39d | 1598 | cpu_base->clock_base[i].cpu_base = cpu_base; |
998adc3d JS |
1599 | timerqueue_init_head(&cpu_base->clock_base[i].active); |
1600 | } | |
3c8aa39d | 1601 | |
54cdfdb4 | 1602 | hrtimer_init_hres(cpu_base); |
c0a31329 TG |
1603 | } |
1604 | ||
1605 | #ifdef CONFIG_HOTPLUG_CPU | |
1606 | ||
ca109491 | 1607 | static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base, |
37810659 | 1608 | struct hrtimer_clock_base *new_base) |
c0a31329 TG |
1609 | { |
1610 | struct hrtimer *timer; | |
998adc3d | 1611 | struct timerqueue_node *node; |
c0a31329 | 1612 | |
998adc3d JS |
1613 | while ((node = timerqueue_getnext(&old_base->active))) { |
1614 | timer = container_of(node, struct hrtimer, node); | |
54cdfdb4 | 1615 | BUG_ON(hrtimer_callback_running(timer)); |
c6a2a177 | 1616 | debug_deactivate(timer); |
b00c1a99 TG |
1617 | |
1618 | /* | |
1619 | * Mark it as STATE_MIGRATE not INACTIVE otherwise the | |
1620 | * timer could be seen as !active and just vanish away | |
1621 | * under us on another CPU | |
1622 | */ | |
1623 | __remove_hrtimer(timer, old_base, HRTIMER_STATE_MIGRATE, 0); | |
c0a31329 | 1624 | timer->base = new_base; |
54cdfdb4 | 1625 | /* |
e3f1d883 TG |
1626 | * Enqueue the timers on the new cpu. This does not |
1627 | * reprogram the event device in case the timer | |
1628 | * expires before the earliest on this CPU, but we run | |
1629 | * hrtimer_interrupt after we migrated everything to | |
1630 | * sort out already expired timers and reprogram the | |
1631 | * event device. | |
54cdfdb4 | 1632 | */ |
a6037b61 | 1633 | enqueue_hrtimer(timer, new_base); |
41e1022e | 1634 | |
b00c1a99 TG |
1635 | /* Clear the migration state bit */ |
1636 | timer->state &= ~HRTIMER_STATE_MIGRATE; | |
c0a31329 TG |
1637 | } |
1638 | } | |
1639 | ||
d5fd43c4 | 1640 | static void migrate_hrtimers(int scpu) |
c0a31329 | 1641 | { |
3c8aa39d | 1642 | struct hrtimer_cpu_base *old_base, *new_base; |
731a55ba | 1643 | int i; |
c0a31329 | 1644 | |
37810659 | 1645 | BUG_ON(cpu_online(scpu)); |
37810659 | 1646 | tick_cancel_sched_timer(scpu); |
731a55ba TG |
1647 | |
1648 | local_irq_disable(); | |
1649 | old_base = &per_cpu(hrtimer_bases, scpu); | |
1650 | new_base = &__get_cpu_var(hrtimer_bases); | |
d82f0b0f ON |
1651 | /* |
1652 | * The caller is globally serialized and nobody else | |
1653 | * takes two locks at once, deadlock is not possible. | |
1654 | */ | |
ecb49d1a TG |
1655 | raw_spin_lock(&new_base->lock); |
1656 | raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING); | |
c0a31329 | 1657 | |
3c8aa39d | 1658 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
ca109491 | 1659 | migrate_hrtimer_list(&old_base->clock_base[i], |
37810659 | 1660 | &new_base->clock_base[i]); |
c0a31329 TG |
1661 | } |
1662 | ||
ecb49d1a TG |
1663 | raw_spin_unlock(&old_base->lock); |
1664 | raw_spin_unlock(&new_base->lock); | |
37810659 | 1665 | |
731a55ba TG |
1666 | /* Check, if we got expired work to do */ |
1667 | __hrtimer_peek_ahead_timers(); | |
1668 | local_irq_enable(); | |
c0a31329 | 1669 | } |
37810659 | 1670 | |
c0a31329 TG |
1671 | #endif /* CONFIG_HOTPLUG_CPU */ |
1672 | ||
8c78f307 | 1673 | static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self, |
c0a31329 TG |
1674 | unsigned long action, void *hcpu) |
1675 | { | |
b2e3c0ad | 1676 | int scpu = (long)hcpu; |
c0a31329 TG |
1677 | |
1678 | switch (action) { | |
1679 | ||
1680 | case CPU_UP_PREPARE: | |
8bb78442 | 1681 | case CPU_UP_PREPARE_FROZEN: |
37810659 | 1682 | init_hrtimers_cpu(scpu); |
c0a31329 TG |
1683 | break; |
1684 | ||
1685 | #ifdef CONFIG_HOTPLUG_CPU | |
94df7de0 SD |
1686 | case CPU_DYING: |
1687 | case CPU_DYING_FROZEN: | |
1688 | clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DYING, &scpu); | |
1689 | break; | |
c0a31329 | 1690 | case CPU_DEAD: |
8bb78442 | 1691 | case CPU_DEAD_FROZEN: |
b2e3c0ad | 1692 | { |
37810659 | 1693 | clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &scpu); |
d5fd43c4 | 1694 | migrate_hrtimers(scpu); |
c0a31329 | 1695 | break; |
b2e3c0ad | 1696 | } |
c0a31329 TG |
1697 | #endif |
1698 | ||
1699 | default: | |
1700 | break; | |
1701 | } | |
1702 | ||
1703 | return NOTIFY_OK; | |
1704 | } | |
1705 | ||
8c78f307 | 1706 | static struct notifier_block __cpuinitdata hrtimers_nb = { |
c0a31329 TG |
1707 | .notifier_call = hrtimer_cpu_notify, |
1708 | }; | |
1709 | ||
1710 | void __init hrtimers_init(void) | |
1711 | { | |
1712 | hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE, | |
1713 | (void *)(long)smp_processor_id()); | |
1714 | register_cpu_notifier(&hrtimers_nb); | |
a6037b61 PZ |
1715 | #ifdef CONFIG_HIGH_RES_TIMERS |
1716 | open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq); | |
1717 | #endif | |
c0a31329 TG |
1718 | } |
1719 | ||
7bb67439 | 1720 | /** |
351b3f7a | 1721 | * schedule_hrtimeout_range_clock - sleep until timeout |
7bb67439 | 1722 | * @expires: timeout value (ktime_t) |
654c8e0b | 1723 | * @delta: slack in expires timeout (ktime_t) |
7bb67439 | 1724 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL |
351b3f7a | 1725 | * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME |
7bb67439 | 1726 | */ |
351b3f7a CE |
1727 | int __sched |
1728 | schedule_hrtimeout_range_clock(ktime_t *expires, unsigned long delta, | |
1729 | const enum hrtimer_mode mode, int clock) | |
7bb67439 AV |
1730 | { |
1731 | struct hrtimer_sleeper t; | |
1732 | ||
1733 | /* | |
1734 | * Optimize when a zero timeout value is given. It does not | |
1735 | * matter whether this is an absolute or a relative time. | |
1736 | */ | |
1737 | if (expires && !expires->tv64) { | |
1738 | __set_current_state(TASK_RUNNING); | |
1739 | return 0; | |
1740 | } | |
1741 | ||
1742 | /* | |
1743 | * A NULL parameter means "inifinte" | |
1744 | */ | |
1745 | if (!expires) { | |
1746 | schedule(); | |
1747 | __set_current_state(TASK_RUNNING); | |
1748 | return -EINTR; | |
1749 | } | |
1750 | ||
351b3f7a | 1751 | hrtimer_init_on_stack(&t.timer, clock, mode); |
654c8e0b | 1752 | hrtimer_set_expires_range_ns(&t.timer, *expires, delta); |
7bb67439 AV |
1753 | |
1754 | hrtimer_init_sleeper(&t, current); | |
1755 | ||
cc584b21 | 1756 | hrtimer_start_expires(&t.timer, mode); |
7bb67439 AV |
1757 | if (!hrtimer_active(&t.timer)) |
1758 | t.task = NULL; | |
1759 | ||
1760 | if (likely(t.task)) | |
1761 | schedule(); | |
1762 | ||
1763 | hrtimer_cancel(&t.timer); | |
1764 | destroy_hrtimer_on_stack(&t.timer); | |
1765 | ||
1766 | __set_current_state(TASK_RUNNING); | |
1767 | ||
1768 | return !t.task ? 0 : -EINTR; | |
1769 | } | |
351b3f7a CE |
1770 | |
1771 | /** | |
1772 | * schedule_hrtimeout_range - sleep until timeout | |
1773 | * @expires: timeout value (ktime_t) | |
1774 | * @delta: slack in expires timeout (ktime_t) | |
1775 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL | |
1776 | * | |
1777 | * Make the current task sleep until the given expiry time has | |
1778 | * elapsed. The routine will return immediately unless | |
1779 | * the current task state has been set (see set_current_state()). | |
1780 | * | |
1781 | * The @delta argument gives the kernel the freedom to schedule the | |
1782 | * actual wakeup to a time that is both power and performance friendly. | |
1783 | * The kernel give the normal best effort behavior for "@expires+@delta", | |
1784 | * but may decide to fire the timer earlier, but no earlier than @expires. | |
1785 | * | |
1786 | * You can set the task state as follows - | |
1787 | * | |
1788 | * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to | |
1789 | * pass before the routine returns. | |
1790 | * | |
1791 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
1792 | * delivered to the current task. | |
1793 | * | |
1794 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1795 | * routine returns. | |
1796 | * | |
1797 | * Returns 0 when the timer has expired otherwise -EINTR | |
1798 | */ | |
1799 | int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta, | |
1800 | const enum hrtimer_mode mode) | |
1801 | { | |
1802 | return schedule_hrtimeout_range_clock(expires, delta, mode, | |
1803 | CLOCK_MONOTONIC); | |
1804 | } | |
654c8e0b AV |
1805 | EXPORT_SYMBOL_GPL(schedule_hrtimeout_range); |
1806 | ||
1807 | /** | |
1808 | * schedule_hrtimeout - sleep until timeout | |
1809 | * @expires: timeout value (ktime_t) | |
1810 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL | |
1811 | * | |
1812 | * Make the current task sleep until the given expiry time has | |
1813 | * elapsed. The routine will return immediately unless | |
1814 | * the current task state has been set (see set_current_state()). | |
1815 | * | |
1816 | * You can set the task state as follows - | |
1817 | * | |
1818 | * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to | |
1819 | * pass before the routine returns. | |
1820 | * | |
1821 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
1822 | * delivered to the current task. | |
1823 | * | |
1824 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1825 | * routine returns. | |
1826 | * | |
1827 | * Returns 0 when the timer has expired otherwise -EINTR | |
1828 | */ | |
1829 | int __sched schedule_hrtimeout(ktime_t *expires, | |
1830 | const enum hrtimer_mode mode) | |
1831 | { | |
1832 | return schedule_hrtimeout_range(expires, 0, mode); | |
1833 | } | |
7bb67439 | 1834 | EXPORT_SYMBOL_GPL(schedule_hrtimeout); |