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); | |
2c6b47de | 92 | xts = current_kernel_time(); |
ad28d94a | 93 | tom = 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)) { | |
3c8aa39d | 130 | 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: */ | |
3c8aa39d | 134 | 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 | |
147 | if (!pinned && get_sysctl_timer_migration() && idle_cpu(this_cpu)) { | |
148 | int preferred_cpu = get_nohz_load_balancer(); | |
149 | ||
150 | if (preferred_cpu >= 0) | |
151 | return preferred_cpu; | |
152 | } | |
153 | #endif | |
154 | return this_cpu; | |
155 | } | |
156 | ||
157 | /* | |
158 | * With HIGHRES=y we do not migrate the timer when it is expiring | |
159 | * before the next event on the target cpu because we cannot reprogram | |
160 | * the target cpu hardware and we would cause it to fire late. | |
161 | * | |
162 | * Called with cpu_base->lock of target cpu held. | |
163 | */ | |
164 | static int | |
165 | hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base) | |
166 | { | |
167 | #ifdef CONFIG_HIGH_RES_TIMERS | |
168 | ktime_t expires; | |
169 | ||
170 | if (!new_base->cpu_base->hres_active) | |
171 | return 0; | |
172 | ||
173 | expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset); | |
174 | return expires.tv64 <= new_base->cpu_base->expires_next.tv64; | |
175 | #else | |
176 | return 0; | |
177 | #endif | |
178 | } | |
179 | ||
c0a31329 TG |
180 | /* |
181 | * Switch the timer base to the current CPU when possible. | |
182 | */ | |
3c8aa39d | 183 | static inline struct hrtimer_clock_base * |
597d0275 AB |
184 | switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base, |
185 | int pinned) | |
c0a31329 | 186 | { |
3c8aa39d TG |
187 | struct hrtimer_clock_base *new_base; |
188 | struct hrtimer_cpu_base *new_cpu_base; | |
6ff7041d TG |
189 | int this_cpu = smp_processor_id(); |
190 | int cpu = hrtimer_get_target(this_cpu, pinned); | |
c0a31329 | 191 | |
eea08f32 AB |
192 | again: |
193 | new_cpu_base = &per_cpu(hrtimer_bases, cpu); | |
3c8aa39d | 194 | new_base = &new_cpu_base->clock_base[base->index]; |
c0a31329 TG |
195 | |
196 | if (base != new_base) { | |
197 | /* | |
6ff7041d | 198 | * We are trying to move timer to new_base. |
c0a31329 TG |
199 | * However we can't change timer's base while it is running, |
200 | * so we keep it on the same CPU. No hassle vs. reprogramming | |
201 | * the event source in the high resolution case. The softirq | |
202 | * code will take care of this when the timer function has | |
203 | * completed. There is no conflict as we hold the lock until | |
204 | * the timer is enqueued. | |
205 | */ | |
54cdfdb4 | 206 | if (unlikely(hrtimer_callback_running(timer))) |
c0a31329 TG |
207 | return base; |
208 | ||
209 | /* See the comment in lock_timer_base() */ | |
210 | timer->base = NULL; | |
3c8aa39d TG |
211 | spin_unlock(&base->cpu_base->lock); |
212 | spin_lock(&new_base->cpu_base->lock); | |
eea08f32 | 213 | |
6ff7041d TG |
214 | if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) { |
215 | cpu = this_cpu; | |
216 | spin_unlock(&new_base->cpu_base->lock); | |
217 | spin_lock(&base->cpu_base->lock); | |
218 | timer->base = base; | |
219 | goto again; | |
eea08f32 | 220 | } |
c0a31329 TG |
221 | timer->base = new_base; |
222 | } | |
223 | return new_base; | |
224 | } | |
225 | ||
226 | #else /* CONFIG_SMP */ | |
227 | ||
3c8aa39d | 228 | static inline struct hrtimer_clock_base * |
c0a31329 TG |
229 | lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) |
230 | { | |
3c8aa39d | 231 | struct hrtimer_clock_base *base = timer->base; |
c0a31329 | 232 | |
3c8aa39d | 233 | spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
234 | |
235 | return base; | |
236 | } | |
237 | ||
eea08f32 | 238 | # define switch_hrtimer_base(t, b, p) (b) |
c0a31329 TG |
239 | |
240 | #endif /* !CONFIG_SMP */ | |
241 | ||
242 | /* | |
243 | * Functions for the union type storage format of ktime_t which are | |
244 | * too large for inlining: | |
245 | */ | |
246 | #if BITS_PER_LONG < 64 | |
247 | # ifndef CONFIG_KTIME_SCALAR | |
248 | /** | |
249 | * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable | |
c0a31329 TG |
250 | * @kt: addend |
251 | * @nsec: the scalar nsec value to add | |
252 | * | |
253 | * Returns the sum of kt and nsec in ktime_t format | |
254 | */ | |
255 | ktime_t ktime_add_ns(const ktime_t kt, u64 nsec) | |
256 | { | |
257 | ktime_t tmp; | |
258 | ||
259 | if (likely(nsec < NSEC_PER_SEC)) { | |
260 | tmp.tv64 = nsec; | |
261 | } else { | |
262 | unsigned long rem = do_div(nsec, NSEC_PER_SEC); | |
263 | ||
264 | tmp = ktime_set((long)nsec, rem); | |
265 | } | |
266 | ||
267 | return ktime_add(kt, tmp); | |
268 | } | |
b8b8fd2d DH |
269 | |
270 | EXPORT_SYMBOL_GPL(ktime_add_ns); | |
a272378d ACM |
271 | |
272 | /** | |
273 | * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable | |
274 | * @kt: minuend | |
275 | * @nsec: the scalar nsec value to subtract | |
276 | * | |
277 | * Returns the subtraction of @nsec from @kt in ktime_t format | |
278 | */ | |
279 | ktime_t ktime_sub_ns(const ktime_t kt, u64 nsec) | |
280 | { | |
281 | ktime_t tmp; | |
282 | ||
283 | if (likely(nsec < NSEC_PER_SEC)) { | |
284 | tmp.tv64 = nsec; | |
285 | } else { | |
286 | unsigned long rem = do_div(nsec, NSEC_PER_SEC); | |
287 | ||
288 | tmp = ktime_set((long)nsec, rem); | |
289 | } | |
290 | ||
291 | return ktime_sub(kt, tmp); | |
292 | } | |
293 | ||
294 | EXPORT_SYMBOL_GPL(ktime_sub_ns); | |
c0a31329 TG |
295 | # endif /* !CONFIG_KTIME_SCALAR */ |
296 | ||
297 | /* | |
298 | * Divide a ktime value by a nanosecond value | |
299 | */ | |
4d672e7a | 300 | u64 ktime_divns(const ktime_t kt, s64 div) |
c0a31329 | 301 | { |
900cfa46 | 302 | u64 dclc; |
c0a31329 TG |
303 | int sft = 0; |
304 | ||
900cfa46 | 305 | dclc = ktime_to_ns(kt); |
c0a31329 TG |
306 | /* Make sure the divisor is less than 2^32: */ |
307 | while (div >> 32) { | |
308 | sft++; | |
309 | div >>= 1; | |
310 | } | |
311 | dclc >>= sft; | |
312 | do_div(dclc, (unsigned long) div); | |
313 | ||
4d672e7a | 314 | return dclc; |
c0a31329 | 315 | } |
c0a31329 TG |
316 | #endif /* BITS_PER_LONG >= 64 */ |
317 | ||
5a7780e7 TG |
318 | /* |
319 | * Add two ktime values and do a safety check for overflow: | |
320 | */ | |
321 | ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs) | |
322 | { | |
323 | ktime_t res = ktime_add(lhs, rhs); | |
324 | ||
325 | /* | |
326 | * We use KTIME_SEC_MAX here, the maximum timeout which we can | |
327 | * return to user space in a timespec: | |
328 | */ | |
329 | if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64) | |
330 | res = ktime_set(KTIME_SEC_MAX, 0); | |
331 | ||
332 | return res; | |
333 | } | |
334 | ||
8daa21e6 AB |
335 | EXPORT_SYMBOL_GPL(ktime_add_safe); |
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 | } | |
2bc481cf | 434 | EXPORT_SYMBOL_GPL(hrtimer_init_on_stack); |
237fc6e7 TG |
435 | |
436 | void destroy_hrtimer_on_stack(struct hrtimer *timer) | |
437 | { | |
438 | debug_object_free(timer, &hrtimer_debug_descr); | |
439 | } | |
440 | ||
441 | #else | |
442 | static inline void debug_hrtimer_init(struct hrtimer *timer) { } | |
443 | static inline void debug_hrtimer_activate(struct hrtimer *timer) { } | |
444 | static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { } | |
445 | #endif | |
446 | ||
c6a2a177 XG |
447 | static inline void |
448 | debug_init(struct hrtimer *timer, clockid_t clockid, | |
449 | enum hrtimer_mode mode) | |
450 | { | |
451 | debug_hrtimer_init(timer); | |
452 | trace_hrtimer_init(timer, clockid, mode); | |
453 | } | |
454 | ||
455 | static inline void debug_activate(struct hrtimer *timer) | |
456 | { | |
457 | debug_hrtimer_activate(timer); | |
458 | trace_hrtimer_start(timer); | |
459 | } | |
460 | ||
461 | static inline void debug_deactivate(struct hrtimer *timer) | |
462 | { | |
463 | debug_hrtimer_deactivate(timer); | |
464 | trace_hrtimer_cancel(timer); | |
465 | } | |
466 | ||
54cdfdb4 TG |
467 | /* High resolution timer related functions */ |
468 | #ifdef CONFIG_HIGH_RES_TIMERS | |
469 | ||
470 | /* | |
471 | * High resolution timer enabled ? | |
472 | */ | |
473 | static int hrtimer_hres_enabled __read_mostly = 1; | |
474 | ||
475 | /* | |
476 | * Enable / Disable high resolution mode | |
477 | */ | |
478 | static int __init setup_hrtimer_hres(char *str) | |
479 | { | |
480 | if (!strcmp(str, "off")) | |
481 | hrtimer_hres_enabled = 0; | |
482 | else if (!strcmp(str, "on")) | |
483 | hrtimer_hres_enabled = 1; | |
484 | else | |
485 | return 0; | |
486 | return 1; | |
487 | } | |
488 | ||
489 | __setup("highres=", setup_hrtimer_hres); | |
490 | ||
491 | /* | |
492 | * hrtimer_high_res_enabled - query, if the highres mode is enabled | |
493 | */ | |
494 | static inline int hrtimer_is_hres_enabled(void) | |
495 | { | |
496 | return hrtimer_hres_enabled; | |
497 | } | |
498 | ||
499 | /* | |
500 | * Is the high resolution mode active ? | |
501 | */ | |
502 | static inline int hrtimer_hres_active(void) | |
503 | { | |
504 | return __get_cpu_var(hrtimer_bases).hres_active; | |
505 | } | |
506 | ||
507 | /* | |
508 | * Reprogram the event source with checking both queues for the | |
509 | * next event | |
510 | * Called with interrupts disabled and base->lock held | |
511 | */ | |
7403f41f AC |
512 | static void |
513 | hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal) | |
54cdfdb4 TG |
514 | { |
515 | int i; | |
516 | struct hrtimer_clock_base *base = cpu_base->clock_base; | |
7403f41f | 517 | ktime_t expires, expires_next; |
54cdfdb4 | 518 | |
7403f41f | 519 | expires_next.tv64 = KTIME_MAX; |
54cdfdb4 TG |
520 | |
521 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) { | |
522 | struct hrtimer *timer; | |
523 | ||
524 | if (!base->first) | |
525 | continue; | |
526 | timer = rb_entry(base->first, struct hrtimer, node); | |
cc584b21 | 527 | expires = ktime_sub(hrtimer_get_expires(timer), base->offset); |
b0a9b511 TG |
528 | /* |
529 | * clock_was_set() has changed base->offset so the | |
530 | * result might be negative. Fix it up to prevent a | |
531 | * false positive in clockevents_program_event() | |
532 | */ | |
533 | if (expires.tv64 < 0) | |
534 | expires.tv64 = 0; | |
7403f41f AC |
535 | if (expires.tv64 < expires_next.tv64) |
536 | expires_next = expires; | |
54cdfdb4 TG |
537 | } |
538 | ||
7403f41f AC |
539 | if (skip_equal && expires_next.tv64 == cpu_base->expires_next.tv64) |
540 | return; | |
541 | ||
542 | cpu_base->expires_next.tv64 = expires_next.tv64; | |
543 | ||
54cdfdb4 TG |
544 | if (cpu_base->expires_next.tv64 != KTIME_MAX) |
545 | tick_program_event(cpu_base->expires_next, 1); | |
546 | } | |
547 | ||
548 | /* | |
549 | * Shared reprogramming for clock_realtime and clock_monotonic | |
550 | * | |
551 | * When a timer is enqueued and expires earlier than the already enqueued | |
552 | * timers, we have to check, whether it expires earlier than the timer for | |
553 | * which the clock event device was armed. | |
554 | * | |
555 | * Called with interrupts disabled and base->cpu_base.lock held | |
556 | */ | |
557 | static int hrtimer_reprogram(struct hrtimer *timer, | |
558 | struct hrtimer_clock_base *base) | |
559 | { | |
560 | ktime_t *expires_next = &__get_cpu_var(hrtimer_bases).expires_next; | |
cc584b21 | 561 | ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset); |
54cdfdb4 TG |
562 | int res; |
563 | ||
cc584b21 | 564 | WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0); |
63070a79 | 565 | |
54cdfdb4 TG |
566 | /* |
567 | * When the callback is running, we do not reprogram the clock event | |
568 | * device. The timer callback is either running on a different CPU or | |
3a4fa0a2 | 569 | * the callback is executed in the hrtimer_interrupt context. The |
54cdfdb4 TG |
570 | * reprogramming is handled either by the softirq, which called the |
571 | * callback or at the end of the hrtimer_interrupt. | |
572 | */ | |
573 | if (hrtimer_callback_running(timer)) | |
574 | return 0; | |
575 | ||
63070a79 TG |
576 | /* |
577 | * CLOCK_REALTIME timer might be requested with an absolute | |
578 | * expiry time which is less than base->offset. Nothing wrong | |
579 | * about that, just avoid to call into the tick code, which | |
580 | * has now objections against negative expiry values. | |
581 | */ | |
582 | if (expires.tv64 < 0) | |
583 | return -ETIME; | |
584 | ||
54cdfdb4 TG |
585 | if (expires.tv64 >= expires_next->tv64) |
586 | return 0; | |
587 | ||
588 | /* | |
589 | * Clockevents returns -ETIME, when the event was in the past. | |
590 | */ | |
591 | res = tick_program_event(expires, 0); | |
592 | if (!IS_ERR_VALUE(res)) | |
593 | *expires_next = expires; | |
594 | return res; | |
595 | } | |
596 | ||
597 | ||
598 | /* | |
599 | * Retrigger next event is called after clock was set | |
600 | * | |
601 | * Called with interrupts disabled via on_each_cpu() | |
602 | */ | |
603 | static void retrigger_next_event(void *arg) | |
604 | { | |
605 | struct hrtimer_cpu_base *base; | |
606 | struct timespec realtime_offset; | |
607 | unsigned long seq; | |
608 | ||
609 | if (!hrtimer_hres_active()) | |
610 | return; | |
611 | ||
612 | do { | |
613 | seq = read_seqbegin(&xtime_lock); | |
614 | set_normalized_timespec(&realtime_offset, | |
615 | -wall_to_monotonic.tv_sec, | |
616 | -wall_to_monotonic.tv_nsec); | |
617 | } while (read_seqretry(&xtime_lock, seq)); | |
618 | ||
619 | base = &__get_cpu_var(hrtimer_bases); | |
620 | ||
621 | /* Adjust CLOCK_REALTIME offset */ | |
622 | spin_lock(&base->lock); | |
623 | base->clock_base[CLOCK_REALTIME].offset = | |
624 | timespec_to_ktime(realtime_offset); | |
625 | ||
7403f41f | 626 | hrtimer_force_reprogram(base, 0); |
54cdfdb4 TG |
627 | spin_unlock(&base->lock); |
628 | } | |
629 | ||
630 | /* | |
631 | * Clock realtime was set | |
632 | * | |
633 | * Change the offset of the realtime clock vs. the monotonic | |
634 | * clock. | |
635 | * | |
636 | * We might have to reprogram the high resolution timer interrupt. On | |
637 | * SMP we call the architecture specific code to retrigger _all_ high | |
638 | * resolution timer interrupts. On UP we just disable interrupts and | |
639 | * call the high resolution interrupt code. | |
640 | */ | |
641 | void clock_was_set(void) | |
642 | { | |
643 | /* Retrigger the CPU local events everywhere */ | |
15c8b6c1 | 644 | on_each_cpu(retrigger_next_event, NULL, 1); |
54cdfdb4 TG |
645 | } |
646 | ||
995f054f IM |
647 | /* |
648 | * During resume we might have to reprogram the high resolution timer | |
649 | * interrupt (on the local CPU): | |
650 | */ | |
651 | void hres_timers_resume(void) | |
652 | { | |
1d4a7f1c PZ |
653 | WARN_ONCE(!irqs_disabled(), |
654 | KERN_INFO "hres_timers_resume() called with IRQs enabled!"); | |
655 | ||
995f054f IM |
656 | retrigger_next_event(NULL); |
657 | } | |
658 | ||
54cdfdb4 TG |
659 | /* |
660 | * Initialize the high resolution related parts of cpu_base | |
661 | */ | |
662 | static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) | |
663 | { | |
664 | base->expires_next.tv64 = KTIME_MAX; | |
665 | base->hres_active = 0; | |
54cdfdb4 TG |
666 | } |
667 | ||
668 | /* | |
669 | * Initialize the high resolution related parts of a hrtimer | |
670 | */ | |
671 | static inline void hrtimer_init_timer_hres(struct hrtimer *timer) | |
672 | { | |
54cdfdb4 TG |
673 | } |
674 | ||
ca109491 | 675 | |
54cdfdb4 TG |
676 | /* |
677 | * When High resolution timers are active, try to reprogram. Note, that in case | |
678 | * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry | |
679 | * check happens. The timer gets enqueued into the rbtree. The reprogramming | |
680 | * and expiry check is done in the hrtimer_interrupt or in the softirq. | |
681 | */ | |
682 | static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer, | |
7f1e2ca9 PZ |
683 | struct hrtimer_clock_base *base, |
684 | int wakeup) | |
54cdfdb4 TG |
685 | { |
686 | if (base->cpu_base->hres_active && hrtimer_reprogram(timer, base)) { | |
7f1e2ca9 PZ |
687 | if (wakeup) { |
688 | spin_unlock(&base->cpu_base->lock); | |
689 | raise_softirq_irqoff(HRTIMER_SOFTIRQ); | |
690 | spin_lock(&base->cpu_base->lock); | |
691 | } else | |
692 | __raise_softirq_irqoff(HRTIMER_SOFTIRQ); | |
693 | ||
ca109491 | 694 | return 1; |
54cdfdb4 | 695 | } |
7f1e2ca9 | 696 | |
54cdfdb4 TG |
697 | return 0; |
698 | } | |
699 | ||
700 | /* | |
701 | * Switch to high resolution mode | |
702 | */ | |
f8953856 | 703 | static int hrtimer_switch_to_hres(void) |
54cdfdb4 | 704 | { |
820de5c3 IM |
705 | int cpu = smp_processor_id(); |
706 | struct hrtimer_cpu_base *base = &per_cpu(hrtimer_bases, cpu); | |
54cdfdb4 TG |
707 | unsigned long flags; |
708 | ||
709 | if (base->hres_active) | |
f8953856 | 710 | return 1; |
54cdfdb4 TG |
711 | |
712 | local_irq_save(flags); | |
713 | ||
714 | if (tick_init_highres()) { | |
715 | local_irq_restore(flags); | |
820de5c3 IM |
716 | printk(KERN_WARNING "Could not switch to high resolution " |
717 | "mode on CPU %d\n", cpu); | |
f8953856 | 718 | return 0; |
54cdfdb4 TG |
719 | } |
720 | base->hres_active = 1; | |
721 | base->clock_base[CLOCK_REALTIME].resolution = KTIME_HIGH_RES; | |
722 | base->clock_base[CLOCK_MONOTONIC].resolution = KTIME_HIGH_RES; | |
723 | ||
724 | tick_setup_sched_timer(); | |
725 | ||
726 | /* "Retrigger" the interrupt to get things going */ | |
727 | retrigger_next_event(NULL); | |
728 | local_irq_restore(flags); | |
edfed66e | 729 | printk(KERN_DEBUG "Switched to high resolution mode on CPU %d\n", |
54cdfdb4 | 730 | smp_processor_id()); |
f8953856 | 731 | return 1; |
54cdfdb4 TG |
732 | } |
733 | ||
734 | #else | |
735 | ||
736 | static inline int hrtimer_hres_active(void) { return 0; } | |
737 | static inline int hrtimer_is_hres_enabled(void) { return 0; } | |
f8953856 | 738 | static inline int hrtimer_switch_to_hres(void) { return 0; } |
7403f41f AC |
739 | static inline void |
740 | hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { } | |
54cdfdb4 | 741 | static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer, |
7f1e2ca9 PZ |
742 | struct hrtimer_clock_base *base, |
743 | int wakeup) | |
54cdfdb4 TG |
744 | { |
745 | return 0; | |
746 | } | |
54cdfdb4 TG |
747 | static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { } |
748 | static inline void hrtimer_init_timer_hres(struct hrtimer *timer) { } | |
749 | ||
750 | #endif /* CONFIG_HIGH_RES_TIMERS */ | |
751 | ||
82f67cd9 IM |
752 | #ifdef CONFIG_TIMER_STATS |
753 | void __timer_stats_hrtimer_set_start_info(struct hrtimer *timer, void *addr) | |
754 | { | |
755 | if (timer->start_site) | |
756 | return; | |
757 | ||
758 | timer->start_site = addr; | |
759 | memcpy(timer->start_comm, current->comm, TASK_COMM_LEN); | |
760 | timer->start_pid = current->pid; | |
761 | } | |
762 | #endif | |
763 | ||
c0a31329 | 764 | /* |
6506f2aa | 765 | * Counterpart to lock_hrtimer_base above: |
c0a31329 TG |
766 | */ |
767 | static inline | |
768 | void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) | |
769 | { | |
3c8aa39d | 770 | spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags); |
c0a31329 TG |
771 | } |
772 | ||
773 | /** | |
774 | * hrtimer_forward - forward the timer expiry | |
c0a31329 | 775 | * @timer: hrtimer to forward |
44f21475 | 776 | * @now: forward past this time |
c0a31329 TG |
777 | * @interval: the interval to forward |
778 | * | |
779 | * Forward the timer expiry so it will expire in the future. | |
8dca6f33 | 780 | * Returns the number of overruns. |
c0a31329 | 781 | */ |
4d672e7a | 782 | u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval) |
c0a31329 | 783 | { |
4d672e7a | 784 | u64 orun = 1; |
44f21475 | 785 | ktime_t delta; |
c0a31329 | 786 | |
cc584b21 | 787 | delta = ktime_sub(now, hrtimer_get_expires(timer)); |
c0a31329 TG |
788 | |
789 | if (delta.tv64 < 0) | |
790 | return 0; | |
791 | ||
c9db4fa1 TG |
792 | if (interval.tv64 < timer->base->resolution.tv64) |
793 | interval.tv64 = timer->base->resolution.tv64; | |
794 | ||
c0a31329 | 795 | if (unlikely(delta.tv64 >= interval.tv64)) { |
df869b63 | 796 | s64 incr = ktime_to_ns(interval); |
c0a31329 TG |
797 | |
798 | orun = ktime_divns(delta, incr); | |
cc584b21 AV |
799 | hrtimer_add_expires_ns(timer, incr * orun); |
800 | if (hrtimer_get_expires_tv64(timer) > now.tv64) | |
c0a31329 TG |
801 | return orun; |
802 | /* | |
803 | * This (and the ktime_add() below) is the | |
804 | * correction for exact: | |
805 | */ | |
806 | orun++; | |
807 | } | |
cc584b21 | 808 | hrtimer_add_expires(timer, interval); |
c0a31329 TG |
809 | |
810 | return orun; | |
811 | } | |
6bdb6b62 | 812 | EXPORT_SYMBOL_GPL(hrtimer_forward); |
c0a31329 TG |
813 | |
814 | /* | |
815 | * enqueue_hrtimer - internal function to (re)start a timer | |
816 | * | |
817 | * The timer is inserted in expiry order. Insertion into the | |
818 | * red black tree is O(log(n)). Must hold the base lock. | |
a6037b61 PZ |
819 | * |
820 | * Returns 1 when the new timer is the leftmost timer in the tree. | |
c0a31329 | 821 | */ |
a6037b61 PZ |
822 | static int enqueue_hrtimer(struct hrtimer *timer, |
823 | struct hrtimer_clock_base *base) | |
c0a31329 TG |
824 | { |
825 | struct rb_node **link = &base->active.rb_node; | |
c0a31329 TG |
826 | struct rb_node *parent = NULL; |
827 | struct hrtimer *entry; | |
99bc2fcb | 828 | int leftmost = 1; |
c0a31329 | 829 | |
c6a2a177 | 830 | debug_activate(timer); |
237fc6e7 | 831 | |
c0a31329 TG |
832 | /* |
833 | * Find the right place in the rbtree: | |
834 | */ | |
835 | while (*link) { | |
836 | parent = *link; | |
837 | entry = rb_entry(parent, struct hrtimer, node); | |
838 | /* | |
839 | * We dont care about collisions. Nodes with | |
840 | * the same expiry time stay together. | |
841 | */ | |
cc584b21 AV |
842 | if (hrtimer_get_expires_tv64(timer) < |
843 | hrtimer_get_expires_tv64(entry)) { | |
c0a31329 | 844 | link = &(*link)->rb_left; |
99bc2fcb | 845 | } else { |
c0a31329 | 846 | link = &(*link)->rb_right; |
99bc2fcb IM |
847 | leftmost = 0; |
848 | } | |
c0a31329 TG |
849 | } |
850 | ||
851 | /* | |
288867ec TG |
852 | * Insert the timer to the rbtree and check whether it |
853 | * replaces the first pending timer | |
c0a31329 | 854 | */ |
a6037b61 | 855 | if (leftmost) |
54cdfdb4 | 856 | base->first = &timer->node; |
54cdfdb4 | 857 | |
c0a31329 TG |
858 | rb_link_node(&timer->node, parent, link); |
859 | rb_insert_color(&timer->node, &base->active); | |
303e967f TG |
860 | /* |
861 | * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the | |
862 | * state of a possibly running callback. | |
863 | */ | |
864 | timer->state |= HRTIMER_STATE_ENQUEUED; | |
a6037b61 PZ |
865 | |
866 | return leftmost; | |
288867ec | 867 | } |
c0a31329 TG |
868 | |
869 | /* | |
870 | * __remove_hrtimer - internal function to remove a timer | |
871 | * | |
872 | * Caller must hold the base lock. | |
54cdfdb4 TG |
873 | * |
874 | * High resolution timer mode reprograms the clock event device when the | |
875 | * timer is the one which expires next. The caller can disable this by setting | |
876 | * reprogram to zero. This is useful, when the context does a reprogramming | |
877 | * anyway (e.g. timer interrupt) | |
c0a31329 | 878 | */ |
3c8aa39d | 879 | static void __remove_hrtimer(struct hrtimer *timer, |
303e967f | 880 | struct hrtimer_clock_base *base, |
54cdfdb4 | 881 | unsigned long newstate, int reprogram) |
c0a31329 | 882 | { |
7403f41f AC |
883 | if (!(timer->state & HRTIMER_STATE_ENQUEUED)) |
884 | goto out; | |
885 | ||
886 | /* | |
887 | * Remove the timer from the rbtree and replace the first | |
888 | * entry pointer if necessary. | |
889 | */ | |
890 | if (base->first == &timer->node) { | |
891 | base->first = rb_next(&timer->node); | |
892 | #ifdef CONFIG_HIGH_RES_TIMERS | |
893 | /* Reprogram the clock event device. if enabled */ | |
894 | if (reprogram && hrtimer_hres_active()) { | |
895 | ktime_t expires; | |
896 | ||
897 | expires = ktime_sub(hrtimer_get_expires(timer), | |
898 | base->offset); | |
899 | if (base->cpu_base->expires_next.tv64 == expires.tv64) | |
900 | hrtimer_force_reprogram(base->cpu_base, 1); | |
54cdfdb4 | 901 | } |
7403f41f | 902 | #endif |
54cdfdb4 | 903 | } |
7403f41f AC |
904 | rb_erase(&timer->node, &base->active); |
905 | out: | |
303e967f | 906 | timer->state = newstate; |
c0a31329 TG |
907 | } |
908 | ||
909 | /* | |
910 | * remove hrtimer, called with base lock held | |
911 | */ | |
912 | static inline int | |
3c8aa39d | 913 | remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base) |
c0a31329 | 914 | { |
303e967f | 915 | if (hrtimer_is_queued(timer)) { |
54cdfdb4 TG |
916 | int reprogram; |
917 | ||
918 | /* | |
919 | * Remove the timer and force reprogramming when high | |
920 | * resolution mode is active and the timer is on the current | |
921 | * CPU. If we remove a timer on another CPU, reprogramming is | |
922 | * skipped. The interrupt event on this CPU is fired and | |
923 | * reprogramming happens in the interrupt handler. This is a | |
924 | * rare case and less expensive than a smp call. | |
925 | */ | |
c6a2a177 | 926 | debug_deactivate(timer); |
82f67cd9 | 927 | timer_stats_hrtimer_clear_start_info(timer); |
54cdfdb4 TG |
928 | reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases); |
929 | __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE, | |
930 | reprogram); | |
c0a31329 TG |
931 | return 1; |
932 | } | |
933 | return 0; | |
934 | } | |
935 | ||
7f1e2ca9 PZ |
936 | int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, |
937 | unsigned long delta_ns, const enum hrtimer_mode mode, | |
938 | int wakeup) | |
c0a31329 | 939 | { |
3c8aa39d | 940 | struct hrtimer_clock_base *base, *new_base; |
c0a31329 | 941 | unsigned long flags; |
a6037b61 | 942 | int ret, leftmost; |
c0a31329 TG |
943 | |
944 | base = lock_hrtimer_base(timer, &flags); | |
945 | ||
946 | /* Remove an active timer from the queue: */ | |
947 | ret = remove_hrtimer(timer, base); | |
948 | ||
949 | /* Switch the timer base, if necessary: */ | |
597d0275 | 950 | new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED); |
c0a31329 | 951 | |
597d0275 | 952 | if (mode & HRTIMER_MODE_REL) { |
5a7780e7 | 953 | tim = ktime_add_safe(tim, new_base->get_time()); |
06027bdd IM |
954 | /* |
955 | * CONFIG_TIME_LOW_RES is a temporary way for architectures | |
956 | * to signal that they simply return xtime in | |
957 | * do_gettimeoffset(). In this case we want to round up by | |
958 | * resolution when starting a relative timer, to avoid short | |
959 | * timeouts. This will go away with the GTOD framework. | |
960 | */ | |
961 | #ifdef CONFIG_TIME_LOW_RES | |
5a7780e7 | 962 | tim = ktime_add_safe(tim, base->resolution); |
06027bdd IM |
963 | #endif |
964 | } | |
237fc6e7 | 965 | |
da8f2e17 | 966 | hrtimer_set_expires_range_ns(timer, tim, delta_ns); |
c0a31329 | 967 | |
82f67cd9 IM |
968 | timer_stats_hrtimer_set_start_info(timer); |
969 | ||
a6037b61 PZ |
970 | leftmost = enqueue_hrtimer(timer, new_base); |
971 | ||
935c631d IM |
972 | /* |
973 | * Only allow reprogramming if the new base is on this CPU. | |
974 | * (it might still be on another CPU if the timer was pending) | |
a6037b61 PZ |
975 | * |
976 | * XXX send_remote_softirq() ? | |
935c631d | 977 | */ |
a6037b61 | 978 | if (leftmost && new_base->cpu_base == &__get_cpu_var(hrtimer_bases)) |
7f1e2ca9 | 979 | hrtimer_enqueue_reprogram(timer, new_base, wakeup); |
c0a31329 TG |
980 | |
981 | unlock_hrtimer_base(timer, &flags); | |
982 | ||
983 | return ret; | |
984 | } | |
7f1e2ca9 PZ |
985 | |
986 | /** | |
987 | * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU | |
988 | * @timer: the timer to be added | |
989 | * @tim: expiry time | |
990 | * @delta_ns: "slack" range for the timer | |
991 | * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL) | |
992 | * | |
993 | * Returns: | |
994 | * 0 on success | |
995 | * 1 when the timer was active | |
996 | */ | |
997 | int hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, | |
998 | unsigned long delta_ns, const enum hrtimer_mode mode) | |
999 | { | |
1000 | return __hrtimer_start_range_ns(timer, tim, delta_ns, mode, 1); | |
1001 | } | |
da8f2e17 AV |
1002 | EXPORT_SYMBOL_GPL(hrtimer_start_range_ns); |
1003 | ||
1004 | /** | |
e1dd7bc5 | 1005 | * hrtimer_start - (re)start an hrtimer on the current CPU |
da8f2e17 AV |
1006 | * @timer: the timer to be added |
1007 | * @tim: expiry time | |
1008 | * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL) | |
1009 | * | |
1010 | * Returns: | |
1011 | * 0 on success | |
1012 | * 1 when the timer was active | |
1013 | */ | |
1014 | int | |
1015 | hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode) | |
1016 | { | |
7f1e2ca9 | 1017 | return __hrtimer_start_range_ns(timer, tim, 0, mode, 1); |
da8f2e17 | 1018 | } |
8d16b764 | 1019 | EXPORT_SYMBOL_GPL(hrtimer_start); |
c0a31329 | 1020 | |
da8f2e17 | 1021 | |
c0a31329 TG |
1022 | /** |
1023 | * hrtimer_try_to_cancel - try to deactivate a timer | |
c0a31329 TG |
1024 | * @timer: hrtimer to stop |
1025 | * | |
1026 | * Returns: | |
1027 | * 0 when the timer was not active | |
1028 | * 1 when the timer was active | |
1029 | * -1 when the timer is currently excuting the callback function and | |
fa9799e3 | 1030 | * cannot be stopped |
c0a31329 TG |
1031 | */ |
1032 | int hrtimer_try_to_cancel(struct hrtimer *timer) | |
1033 | { | |
3c8aa39d | 1034 | struct hrtimer_clock_base *base; |
c0a31329 TG |
1035 | unsigned long flags; |
1036 | int ret = -1; | |
1037 | ||
1038 | base = lock_hrtimer_base(timer, &flags); | |
1039 | ||
303e967f | 1040 | if (!hrtimer_callback_running(timer)) |
c0a31329 TG |
1041 | ret = remove_hrtimer(timer, base); |
1042 | ||
1043 | unlock_hrtimer_base(timer, &flags); | |
1044 | ||
1045 | return ret; | |
1046 | ||
1047 | } | |
8d16b764 | 1048 | EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel); |
c0a31329 TG |
1049 | |
1050 | /** | |
1051 | * hrtimer_cancel - cancel a timer and wait for the handler to finish. | |
c0a31329 TG |
1052 | * @timer: the timer to be cancelled |
1053 | * | |
1054 | * Returns: | |
1055 | * 0 when the timer was not active | |
1056 | * 1 when the timer was active | |
1057 | */ | |
1058 | int hrtimer_cancel(struct hrtimer *timer) | |
1059 | { | |
1060 | for (;;) { | |
1061 | int ret = hrtimer_try_to_cancel(timer); | |
1062 | ||
1063 | if (ret >= 0) | |
1064 | return ret; | |
5ef37b19 | 1065 | cpu_relax(); |
c0a31329 TG |
1066 | } |
1067 | } | |
8d16b764 | 1068 | EXPORT_SYMBOL_GPL(hrtimer_cancel); |
c0a31329 TG |
1069 | |
1070 | /** | |
1071 | * hrtimer_get_remaining - get remaining time for the timer | |
c0a31329 TG |
1072 | * @timer: the timer to read |
1073 | */ | |
1074 | ktime_t hrtimer_get_remaining(const struct hrtimer *timer) | |
1075 | { | |
3c8aa39d | 1076 | struct hrtimer_clock_base *base; |
c0a31329 TG |
1077 | unsigned long flags; |
1078 | ktime_t rem; | |
1079 | ||
1080 | base = lock_hrtimer_base(timer, &flags); | |
cc584b21 | 1081 | rem = hrtimer_expires_remaining(timer); |
c0a31329 TG |
1082 | unlock_hrtimer_base(timer, &flags); |
1083 | ||
1084 | return rem; | |
1085 | } | |
8d16b764 | 1086 | EXPORT_SYMBOL_GPL(hrtimer_get_remaining); |
c0a31329 | 1087 | |
ee9c5785 | 1088 | #ifdef CONFIG_NO_HZ |
69239749 TL |
1089 | /** |
1090 | * hrtimer_get_next_event - get the time until next expiry event | |
1091 | * | |
1092 | * Returns the delta to the next expiry event or KTIME_MAX if no timer | |
1093 | * is pending. | |
1094 | */ | |
1095 | ktime_t hrtimer_get_next_event(void) | |
1096 | { | |
3c8aa39d TG |
1097 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
1098 | struct hrtimer_clock_base *base = cpu_base->clock_base; | |
69239749 TL |
1099 | ktime_t delta, mindelta = { .tv64 = KTIME_MAX }; |
1100 | unsigned long flags; | |
1101 | int i; | |
1102 | ||
3c8aa39d TG |
1103 | spin_lock_irqsave(&cpu_base->lock, flags); |
1104 | ||
54cdfdb4 TG |
1105 | if (!hrtimer_hres_active()) { |
1106 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) { | |
1107 | struct hrtimer *timer; | |
69239749 | 1108 | |
54cdfdb4 TG |
1109 | if (!base->first) |
1110 | continue; | |
3c8aa39d | 1111 | |
54cdfdb4 | 1112 | timer = rb_entry(base->first, struct hrtimer, node); |
cc584b21 | 1113 | delta.tv64 = hrtimer_get_expires_tv64(timer); |
54cdfdb4 TG |
1114 | delta = ktime_sub(delta, base->get_time()); |
1115 | if (delta.tv64 < mindelta.tv64) | |
1116 | mindelta.tv64 = delta.tv64; | |
1117 | } | |
69239749 | 1118 | } |
3c8aa39d TG |
1119 | |
1120 | spin_unlock_irqrestore(&cpu_base->lock, flags); | |
1121 | ||
69239749 TL |
1122 | if (mindelta.tv64 < 0) |
1123 | mindelta.tv64 = 0; | |
1124 | return mindelta; | |
1125 | } | |
1126 | #endif | |
1127 | ||
237fc6e7 TG |
1128 | static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id, |
1129 | enum hrtimer_mode mode) | |
c0a31329 | 1130 | { |
3c8aa39d | 1131 | struct hrtimer_cpu_base *cpu_base; |
c0a31329 | 1132 | |
7978672c GA |
1133 | memset(timer, 0, sizeof(struct hrtimer)); |
1134 | ||
3c8aa39d | 1135 | cpu_base = &__raw_get_cpu_var(hrtimer_bases); |
c0a31329 | 1136 | |
c9cb2e3d | 1137 | if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS) |
7978672c GA |
1138 | clock_id = CLOCK_MONOTONIC; |
1139 | ||
3c8aa39d | 1140 | timer->base = &cpu_base->clock_base[clock_id]; |
54cdfdb4 | 1141 | hrtimer_init_timer_hres(timer); |
82f67cd9 IM |
1142 | |
1143 | #ifdef CONFIG_TIMER_STATS | |
1144 | timer->start_site = NULL; | |
1145 | timer->start_pid = -1; | |
1146 | memset(timer->start_comm, 0, TASK_COMM_LEN); | |
1147 | #endif | |
c0a31329 | 1148 | } |
237fc6e7 TG |
1149 | |
1150 | /** | |
1151 | * hrtimer_init - initialize a timer to the given clock | |
1152 | * @timer: the timer to be initialized | |
1153 | * @clock_id: the clock to be used | |
1154 | * @mode: timer mode abs/rel | |
1155 | */ | |
1156 | void hrtimer_init(struct hrtimer *timer, clockid_t clock_id, | |
1157 | enum hrtimer_mode mode) | |
1158 | { | |
c6a2a177 | 1159 | debug_init(timer, clock_id, mode); |
237fc6e7 TG |
1160 | __hrtimer_init(timer, clock_id, mode); |
1161 | } | |
8d16b764 | 1162 | EXPORT_SYMBOL_GPL(hrtimer_init); |
c0a31329 TG |
1163 | |
1164 | /** | |
1165 | * hrtimer_get_res - get the timer resolution for a clock | |
c0a31329 TG |
1166 | * @which_clock: which clock to query |
1167 | * @tp: pointer to timespec variable to store the resolution | |
1168 | * | |
72fd4a35 RD |
1169 | * Store the resolution of the clock selected by @which_clock in the |
1170 | * variable pointed to by @tp. | |
c0a31329 TG |
1171 | */ |
1172 | int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp) | |
1173 | { | |
3c8aa39d | 1174 | struct hrtimer_cpu_base *cpu_base; |
c0a31329 | 1175 | |
3c8aa39d TG |
1176 | cpu_base = &__raw_get_cpu_var(hrtimer_bases); |
1177 | *tp = ktime_to_timespec(cpu_base->clock_base[which_clock].resolution); | |
c0a31329 TG |
1178 | |
1179 | return 0; | |
1180 | } | |
8d16b764 | 1181 | EXPORT_SYMBOL_GPL(hrtimer_get_res); |
c0a31329 | 1182 | |
c6a2a177 | 1183 | static void __run_hrtimer(struct hrtimer *timer, ktime_t *now) |
d3d74453 PZ |
1184 | { |
1185 | struct hrtimer_clock_base *base = timer->base; | |
1186 | struct hrtimer_cpu_base *cpu_base = base->cpu_base; | |
1187 | enum hrtimer_restart (*fn)(struct hrtimer *); | |
1188 | int restart; | |
1189 | ||
ca109491 PZ |
1190 | WARN_ON(!irqs_disabled()); |
1191 | ||
c6a2a177 | 1192 | debug_deactivate(timer); |
d3d74453 PZ |
1193 | __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0); |
1194 | timer_stats_account_hrtimer(timer); | |
d3d74453 | 1195 | fn = timer->function; |
ca109491 PZ |
1196 | |
1197 | /* | |
1198 | * Because we run timers from hardirq context, there is no chance | |
1199 | * they get migrated to another cpu, therefore its safe to unlock | |
1200 | * the timer base. | |
1201 | */ | |
1202 | spin_unlock(&cpu_base->lock); | |
c6a2a177 | 1203 | trace_hrtimer_expire_entry(timer, now); |
ca109491 | 1204 | restart = fn(timer); |
c6a2a177 | 1205 | trace_hrtimer_expire_exit(timer); |
ca109491 | 1206 | spin_lock(&cpu_base->lock); |
d3d74453 PZ |
1207 | |
1208 | /* | |
e3f1d883 TG |
1209 | * Note: We clear the CALLBACK bit after enqueue_hrtimer and |
1210 | * we do not reprogramm the event hardware. Happens either in | |
1211 | * hrtimer_start_range_ns() or in hrtimer_interrupt() | |
d3d74453 PZ |
1212 | */ |
1213 | if (restart != HRTIMER_NORESTART) { | |
1214 | BUG_ON(timer->state != HRTIMER_STATE_CALLBACK); | |
a6037b61 | 1215 | enqueue_hrtimer(timer, base); |
d3d74453 PZ |
1216 | } |
1217 | timer->state &= ~HRTIMER_STATE_CALLBACK; | |
1218 | } | |
1219 | ||
54cdfdb4 TG |
1220 | #ifdef CONFIG_HIGH_RES_TIMERS |
1221 | ||
7f22391c FW |
1222 | static int force_clock_reprogram; |
1223 | ||
1224 | /* | |
1225 | * After 5 iteration's attempts, we consider that hrtimer_interrupt() | |
1226 | * is hanging, which could happen with something that slows the interrupt | |
1227 | * such as the tracing. Then we force the clock reprogramming for each future | |
1228 | * hrtimer interrupts to avoid infinite loops and use the min_delta_ns | |
1229 | * threshold that we will overwrite. | |
1230 | * The next tick event will be scheduled to 3 times we currently spend on | |
1231 | * hrtimer_interrupt(). This gives a good compromise, the cpus will spend | |
1232 | * 1/4 of their time to process the hrtimer interrupts. This is enough to | |
1233 | * let it running without serious starvation. | |
1234 | */ | |
1235 | ||
1236 | static inline void | |
1237 | hrtimer_interrupt_hanging(struct clock_event_device *dev, | |
1238 | ktime_t try_time) | |
1239 | { | |
1240 | force_clock_reprogram = 1; | |
1241 | dev->min_delta_ns = (unsigned long)try_time.tv64 * 3; | |
1242 | printk(KERN_WARNING "hrtimer: interrupt too slow, " | |
1243 | "forcing clock min delta to %lu ns\n", dev->min_delta_ns); | |
1244 | } | |
54cdfdb4 TG |
1245 | /* |
1246 | * High resolution timer interrupt | |
1247 | * Called with interrupts disabled | |
1248 | */ | |
1249 | void hrtimer_interrupt(struct clock_event_device *dev) | |
1250 | { | |
1251 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); | |
1252 | struct hrtimer_clock_base *base; | |
1253 | ktime_t expires_next, now; | |
7f22391c | 1254 | int nr_retries = 0; |
ca109491 | 1255 | int i; |
54cdfdb4 TG |
1256 | |
1257 | BUG_ON(!cpu_base->hres_active); | |
1258 | cpu_base->nr_events++; | |
1259 | dev->next_event.tv64 = KTIME_MAX; | |
1260 | ||
1261 | retry: | |
7f22391c FW |
1262 | /* 5 retries is enough to notice a hang */ |
1263 | if (!(++nr_retries % 5)) | |
1264 | hrtimer_interrupt_hanging(dev, ktime_sub(ktime_get(), now)); | |
1265 | ||
54cdfdb4 TG |
1266 | now = ktime_get(); |
1267 | ||
1268 | expires_next.tv64 = KTIME_MAX; | |
1269 | ||
6ff7041d TG |
1270 | spin_lock(&cpu_base->lock); |
1271 | /* | |
1272 | * We set expires_next to KTIME_MAX here with cpu_base->lock | |
1273 | * held to prevent that a timer is enqueued in our queue via | |
1274 | * the migration code. This does not affect enqueueing of | |
1275 | * timers which run their callback and need to be requeued on | |
1276 | * this CPU. | |
1277 | */ | |
1278 | cpu_base->expires_next.tv64 = KTIME_MAX; | |
1279 | ||
54cdfdb4 TG |
1280 | base = cpu_base->clock_base; |
1281 | ||
1282 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { | |
1283 | ktime_t basenow; | |
1284 | struct rb_node *node; | |
1285 | ||
54cdfdb4 TG |
1286 | basenow = ktime_add(now, base->offset); |
1287 | ||
1288 | while ((node = base->first)) { | |
1289 | struct hrtimer *timer; | |
1290 | ||
1291 | timer = rb_entry(node, struct hrtimer, node); | |
1292 | ||
654c8e0b AV |
1293 | /* |
1294 | * The immediate goal for using the softexpires is | |
1295 | * minimizing wakeups, not running timers at the | |
1296 | * earliest interrupt after their soft expiration. | |
1297 | * This allows us to avoid using a Priority Search | |
1298 | * Tree, which can answer a stabbing querry for | |
1299 | * overlapping intervals and instead use the simple | |
1300 | * BST we already have. | |
1301 | * We don't add extra wakeups by delaying timers that | |
1302 | * are right-of a not yet expired timer, because that | |
1303 | * timer will have to trigger a wakeup anyway. | |
1304 | */ | |
1305 | ||
1306 | if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer)) { | |
54cdfdb4 TG |
1307 | ktime_t expires; |
1308 | ||
cc584b21 | 1309 | expires = ktime_sub(hrtimer_get_expires(timer), |
54cdfdb4 TG |
1310 | base->offset); |
1311 | if (expires.tv64 < expires_next.tv64) | |
1312 | expires_next = expires; | |
1313 | break; | |
1314 | } | |
1315 | ||
c6a2a177 | 1316 | __run_hrtimer(timer, &basenow); |
54cdfdb4 | 1317 | } |
54cdfdb4 TG |
1318 | base++; |
1319 | } | |
1320 | ||
6ff7041d TG |
1321 | /* |
1322 | * Store the new expiry value so the migration code can verify | |
1323 | * against it. | |
1324 | */ | |
54cdfdb4 | 1325 | cpu_base->expires_next = expires_next; |
6ff7041d | 1326 | spin_unlock(&cpu_base->lock); |
54cdfdb4 TG |
1327 | |
1328 | /* Reprogramming necessary ? */ | |
1329 | if (expires_next.tv64 != KTIME_MAX) { | |
7f22391c | 1330 | if (tick_program_event(expires_next, force_clock_reprogram)) |
54cdfdb4 TG |
1331 | goto retry; |
1332 | } | |
54cdfdb4 TG |
1333 | } |
1334 | ||
8bdec955 TG |
1335 | /* |
1336 | * local version of hrtimer_peek_ahead_timers() called with interrupts | |
1337 | * disabled. | |
1338 | */ | |
1339 | static void __hrtimer_peek_ahead_timers(void) | |
1340 | { | |
1341 | struct tick_device *td; | |
1342 | ||
1343 | if (!hrtimer_hres_active()) | |
1344 | return; | |
1345 | ||
1346 | td = &__get_cpu_var(tick_cpu_device); | |
1347 | if (td && td->evtdev) | |
1348 | hrtimer_interrupt(td->evtdev); | |
1349 | } | |
1350 | ||
2e94d1f7 AV |
1351 | /** |
1352 | * hrtimer_peek_ahead_timers -- run soft-expired timers now | |
1353 | * | |
1354 | * hrtimer_peek_ahead_timers will peek at the timer queue of | |
1355 | * the current cpu and check if there are any timers for which | |
1356 | * the soft expires time has passed. If any such timers exist, | |
1357 | * they are run immediately and then removed from the timer queue. | |
1358 | * | |
1359 | */ | |
1360 | void hrtimer_peek_ahead_timers(void) | |
1361 | { | |
643bdf68 | 1362 | unsigned long flags; |
dc4304f7 | 1363 | |
2e94d1f7 | 1364 | local_irq_save(flags); |
8bdec955 | 1365 | __hrtimer_peek_ahead_timers(); |
2e94d1f7 AV |
1366 | local_irq_restore(flags); |
1367 | } | |
1368 | ||
a6037b61 PZ |
1369 | static void run_hrtimer_softirq(struct softirq_action *h) |
1370 | { | |
1371 | hrtimer_peek_ahead_timers(); | |
1372 | } | |
1373 | ||
82c5b7b5 IM |
1374 | #else /* CONFIG_HIGH_RES_TIMERS */ |
1375 | ||
1376 | static inline void __hrtimer_peek_ahead_timers(void) { } | |
1377 | ||
1378 | #endif /* !CONFIG_HIGH_RES_TIMERS */ | |
82f67cd9 | 1379 | |
d3d74453 PZ |
1380 | /* |
1381 | * Called from timer softirq every jiffy, expire hrtimers: | |
1382 | * | |
1383 | * For HRT its the fall back code to run the softirq in the timer | |
1384 | * softirq context in case the hrtimer initialization failed or has | |
1385 | * not been done yet. | |
1386 | */ | |
1387 | void hrtimer_run_pending(void) | |
1388 | { | |
d3d74453 PZ |
1389 | if (hrtimer_hres_active()) |
1390 | return; | |
54cdfdb4 | 1391 | |
d3d74453 PZ |
1392 | /* |
1393 | * This _is_ ugly: We have to check in the softirq context, | |
1394 | * whether we can switch to highres and / or nohz mode. The | |
1395 | * clocksource switch happens in the timer interrupt with | |
1396 | * xtime_lock held. Notification from there only sets the | |
1397 | * check bit in the tick_oneshot code, otherwise we might | |
1398 | * deadlock vs. xtime_lock. | |
1399 | */ | |
1400 | if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) | |
1401 | hrtimer_switch_to_hres(); | |
54cdfdb4 TG |
1402 | } |
1403 | ||
c0a31329 | 1404 | /* |
d3d74453 | 1405 | * Called from hardirq context every jiffy |
c0a31329 | 1406 | */ |
833883d9 | 1407 | void hrtimer_run_queues(void) |
c0a31329 | 1408 | { |
288867ec | 1409 | struct rb_node *node; |
833883d9 DS |
1410 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
1411 | struct hrtimer_clock_base *base; | |
1412 | int index, gettime = 1; | |
c0a31329 | 1413 | |
833883d9 | 1414 | if (hrtimer_hres_active()) |
3055adda DS |
1415 | return; |
1416 | ||
833883d9 DS |
1417 | for (index = 0; index < HRTIMER_MAX_CLOCK_BASES; index++) { |
1418 | base = &cpu_base->clock_base[index]; | |
c0a31329 | 1419 | |
833883d9 | 1420 | if (!base->first) |
d3d74453 | 1421 | continue; |
833883d9 | 1422 | |
d7cfb60c | 1423 | if (gettime) { |
833883d9 DS |
1424 | hrtimer_get_softirq_time(cpu_base); |
1425 | gettime = 0; | |
b75f7a51 | 1426 | } |
d3d74453 | 1427 | |
833883d9 | 1428 | spin_lock(&cpu_base->lock); |
c0a31329 | 1429 | |
833883d9 DS |
1430 | while ((node = base->first)) { |
1431 | struct hrtimer *timer; | |
54cdfdb4 | 1432 | |
833883d9 | 1433 | timer = rb_entry(node, struct hrtimer, node); |
cc584b21 AV |
1434 | if (base->softirq_time.tv64 <= |
1435 | hrtimer_get_expires_tv64(timer)) | |
833883d9 DS |
1436 | break; |
1437 | ||
c6a2a177 | 1438 | __run_hrtimer(timer, &base->softirq_time); |
833883d9 DS |
1439 | } |
1440 | spin_unlock(&cpu_base->lock); | |
1441 | } | |
c0a31329 TG |
1442 | } |
1443 | ||
10c94ec1 TG |
1444 | /* |
1445 | * Sleep related functions: | |
1446 | */ | |
c9cb2e3d | 1447 | static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer) |
00362e33 TG |
1448 | { |
1449 | struct hrtimer_sleeper *t = | |
1450 | container_of(timer, struct hrtimer_sleeper, timer); | |
1451 | struct task_struct *task = t->task; | |
1452 | ||
1453 | t->task = NULL; | |
1454 | if (task) | |
1455 | wake_up_process(task); | |
1456 | ||
1457 | return HRTIMER_NORESTART; | |
1458 | } | |
1459 | ||
36c8b586 | 1460 | void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task) |
00362e33 TG |
1461 | { |
1462 | sl->timer.function = hrtimer_wakeup; | |
1463 | sl->task = task; | |
1464 | } | |
2bc481cf | 1465 | EXPORT_SYMBOL_GPL(hrtimer_init_sleeper); |
00362e33 | 1466 | |
669d7868 | 1467 | static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode) |
432569bb | 1468 | { |
669d7868 | 1469 | hrtimer_init_sleeper(t, current); |
10c94ec1 | 1470 | |
432569bb RZ |
1471 | do { |
1472 | set_current_state(TASK_INTERRUPTIBLE); | |
cc584b21 | 1473 | hrtimer_start_expires(&t->timer, mode); |
37bb6cb4 PZ |
1474 | if (!hrtimer_active(&t->timer)) |
1475 | t->task = NULL; | |
432569bb | 1476 | |
54cdfdb4 TG |
1477 | if (likely(t->task)) |
1478 | schedule(); | |
432569bb | 1479 | |
669d7868 | 1480 | hrtimer_cancel(&t->timer); |
c9cb2e3d | 1481 | mode = HRTIMER_MODE_ABS; |
669d7868 TG |
1482 | |
1483 | } while (t->task && !signal_pending(current)); | |
432569bb | 1484 | |
3588a085 PZ |
1485 | __set_current_state(TASK_RUNNING); |
1486 | ||
669d7868 | 1487 | return t->task == NULL; |
10c94ec1 TG |
1488 | } |
1489 | ||
080344b9 ON |
1490 | static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp) |
1491 | { | |
1492 | struct timespec rmt; | |
1493 | ktime_t rem; | |
1494 | ||
cc584b21 | 1495 | rem = hrtimer_expires_remaining(timer); |
080344b9 ON |
1496 | if (rem.tv64 <= 0) |
1497 | return 0; | |
1498 | rmt = ktime_to_timespec(rem); | |
1499 | ||
1500 | if (copy_to_user(rmtp, &rmt, sizeof(*rmtp))) | |
1501 | return -EFAULT; | |
1502 | ||
1503 | return 1; | |
1504 | } | |
1505 | ||
1711ef38 | 1506 | long __sched hrtimer_nanosleep_restart(struct restart_block *restart) |
10c94ec1 | 1507 | { |
669d7868 | 1508 | struct hrtimer_sleeper t; |
080344b9 | 1509 | struct timespec __user *rmtp; |
237fc6e7 | 1510 | int ret = 0; |
10c94ec1 | 1511 | |
237fc6e7 TG |
1512 | hrtimer_init_on_stack(&t.timer, restart->nanosleep.index, |
1513 | HRTIMER_MODE_ABS); | |
cc584b21 | 1514 | hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires); |
10c94ec1 | 1515 | |
c9cb2e3d | 1516 | if (do_nanosleep(&t, HRTIMER_MODE_ABS)) |
237fc6e7 | 1517 | goto out; |
10c94ec1 | 1518 | |
029a07e0 | 1519 | rmtp = restart->nanosleep.rmtp; |
432569bb | 1520 | if (rmtp) { |
237fc6e7 | 1521 | ret = update_rmtp(&t.timer, rmtp); |
080344b9 | 1522 | if (ret <= 0) |
237fc6e7 | 1523 | goto out; |
432569bb | 1524 | } |
10c94ec1 | 1525 | |
10c94ec1 | 1526 | /* The other values in restart are already filled in */ |
237fc6e7 TG |
1527 | ret = -ERESTART_RESTARTBLOCK; |
1528 | out: | |
1529 | destroy_hrtimer_on_stack(&t.timer); | |
1530 | return ret; | |
10c94ec1 TG |
1531 | } |
1532 | ||
080344b9 | 1533 | long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp, |
10c94ec1 TG |
1534 | const enum hrtimer_mode mode, const clockid_t clockid) |
1535 | { | |
1536 | struct restart_block *restart; | |
669d7868 | 1537 | struct hrtimer_sleeper t; |
237fc6e7 | 1538 | int ret = 0; |
3bd01206 AV |
1539 | unsigned long slack; |
1540 | ||
1541 | slack = current->timer_slack_ns; | |
1542 | if (rt_task(current)) | |
1543 | slack = 0; | |
10c94ec1 | 1544 | |
237fc6e7 | 1545 | hrtimer_init_on_stack(&t.timer, clockid, mode); |
3bd01206 | 1546 | hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack); |
432569bb | 1547 | if (do_nanosleep(&t, mode)) |
237fc6e7 | 1548 | goto out; |
10c94ec1 | 1549 | |
7978672c | 1550 | /* Absolute timers do not update the rmtp value and restart: */ |
237fc6e7 TG |
1551 | if (mode == HRTIMER_MODE_ABS) { |
1552 | ret = -ERESTARTNOHAND; | |
1553 | goto out; | |
1554 | } | |
10c94ec1 | 1555 | |
432569bb | 1556 | if (rmtp) { |
237fc6e7 | 1557 | ret = update_rmtp(&t.timer, rmtp); |
080344b9 | 1558 | if (ret <= 0) |
237fc6e7 | 1559 | goto out; |
432569bb | 1560 | } |
10c94ec1 TG |
1561 | |
1562 | restart = ¤t_thread_info()->restart_block; | |
1711ef38 | 1563 | restart->fn = hrtimer_nanosleep_restart; |
029a07e0 TG |
1564 | restart->nanosleep.index = t.timer.base->index; |
1565 | restart->nanosleep.rmtp = rmtp; | |
cc584b21 | 1566 | restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer); |
10c94ec1 | 1567 | |
237fc6e7 TG |
1568 | ret = -ERESTART_RESTARTBLOCK; |
1569 | out: | |
1570 | destroy_hrtimer_on_stack(&t.timer); | |
1571 | return ret; | |
10c94ec1 TG |
1572 | } |
1573 | ||
58fd3aa2 HC |
1574 | SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp, |
1575 | struct timespec __user *, rmtp) | |
6ba1b912 | 1576 | { |
080344b9 | 1577 | struct timespec tu; |
6ba1b912 TG |
1578 | |
1579 | if (copy_from_user(&tu, rqtp, sizeof(tu))) | |
1580 | return -EFAULT; | |
1581 | ||
1582 | if (!timespec_valid(&tu)) | |
1583 | return -EINVAL; | |
1584 | ||
080344b9 | 1585 | return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC); |
6ba1b912 TG |
1586 | } |
1587 | ||
c0a31329 TG |
1588 | /* |
1589 | * Functions related to boot-time initialization: | |
1590 | */ | |
0ec160dd | 1591 | static void __cpuinit init_hrtimers_cpu(int cpu) |
c0a31329 | 1592 | { |
3c8aa39d | 1593 | struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu); |
c0a31329 TG |
1594 | int i; |
1595 | ||
3c8aa39d | 1596 | spin_lock_init(&cpu_base->lock); |
3c8aa39d TG |
1597 | |
1598 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) | |
1599 | cpu_base->clock_base[i].cpu_base = cpu_base; | |
1600 | ||
54cdfdb4 | 1601 | hrtimer_init_hres(cpu_base); |
c0a31329 TG |
1602 | } |
1603 | ||
1604 | #ifdef CONFIG_HOTPLUG_CPU | |
1605 | ||
ca109491 | 1606 | static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base, |
37810659 | 1607 | struct hrtimer_clock_base *new_base) |
c0a31329 TG |
1608 | { |
1609 | struct hrtimer *timer; | |
1610 | struct rb_node *node; | |
1611 | ||
1612 | while ((node = rb_first(&old_base->active))) { | |
1613 | timer = rb_entry(node, struct hrtimer, node); | |
54cdfdb4 | 1614 | BUG_ON(hrtimer_callback_running(timer)); |
c6a2a177 | 1615 | debug_deactivate(timer); |
b00c1a99 TG |
1616 | |
1617 | /* | |
1618 | * Mark it as STATE_MIGRATE not INACTIVE otherwise the | |
1619 | * timer could be seen as !active and just vanish away | |
1620 | * under us on another CPU | |
1621 | */ | |
1622 | __remove_hrtimer(timer, old_base, HRTIMER_STATE_MIGRATE, 0); | |
c0a31329 | 1623 | timer->base = new_base; |
54cdfdb4 | 1624 | /* |
e3f1d883 TG |
1625 | * Enqueue the timers on the new cpu. This does not |
1626 | * reprogram the event device in case the timer | |
1627 | * expires before the earliest on this CPU, but we run | |
1628 | * hrtimer_interrupt after we migrated everything to | |
1629 | * sort out already expired timers and reprogram the | |
1630 | * event device. | |
54cdfdb4 | 1631 | */ |
a6037b61 | 1632 | enqueue_hrtimer(timer, new_base); |
41e1022e | 1633 | |
b00c1a99 TG |
1634 | /* Clear the migration state bit */ |
1635 | timer->state &= ~HRTIMER_STATE_MIGRATE; | |
c0a31329 TG |
1636 | } |
1637 | } | |
1638 | ||
d5fd43c4 | 1639 | static void migrate_hrtimers(int scpu) |
c0a31329 | 1640 | { |
3c8aa39d | 1641 | struct hrtimer_cpu_base *old_base, *new_base; |
731a55ba | 1642 | int i; |
c0a31329 | 1643 | |
37810659 | 1644 | BUG_ON(cpu_online(scpu)); |
37810659 | 1645 | tick_cancel_sched_timer(scpu); |
731a55ba TG |
1646 | |
1647 | local_irq_disable(); | |
1648 | old_base = &per_cpu(hrtimer_bases, scpu); | |
1649 | new_base = &__get_cpu_var(hrtimer_bases); | |
d82f0b0f ON |
1650 | /* |
1651 | * The caller is globally serialized and nobody else | |
1652 | * takes two locks at once, deadlock is not possible. | |
1653 | */ | |
731a55ba | 1654 | spin_lock(&new_base->lock); |
8e60e05f | 1655 | spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING); |
c0a31329 | 1656 | |
3c8aa39d | 1657 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
ca109491 | 1658 | migrate_hrtimer_list(&old_base->clock_base[i], |
37810659 | 1659 | &new_base->clock_base[i]); |
c0a31329 TG |
1660 | } |
1661 | ||
8e60e05f | 1662 | spin_unlock(&old_base->lock); |
731a55ba | 1663 | spin_unlock(&new_base->lock); |
37810659 | 1664 | |
731a55ba TG |
1665 | /* Check, if we got expired work to do */ |
1666 | __hrtimer_peek_ahead_timers(); | |
1667 | local_irq_enable(); | |
c0a31329 | 1668 | } |
37810659 | 1669 | |
c0a31329 TG |
1670 | #endif /* CONFIG_HOTPLUG_CPU */ |
1671 | ||
8c78f307 | 1672 | static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self, |
c0a31329 TG |
1673 | unsigned long action, void *hcpu) |
1674 | { | |
b2e3c0ad | 1675 | int scpu = (long)hcpu; |
c0a31329 TG |
1676 | |
1677 | switch (action) { | |
1678 | ||
1679 | case CPU_UP_PREPARE: | |
8bb78442 | 1680 | case CPU_UP_PREPARE_FROZEN: |
37810659 | 1681 | init_hrtimers_cpu(scpu); |
c0a31329 TG |
1682 | break; |
1683 | ||
1684 | #ifdef CONFIG_HOTPLUG_CPU | |
94df7de0 SD |
1685 | case CPU_DYING: |
1686 | case CPU_DYING_FROZEN: | |
1687 | clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DYING, &scpu); | |
1688 | break; | |
c0a31329 | 1689 | case CPU_DEAD: |
8bb78442 | 1690 | case CPU_DEAD_FROZEN: |
b2e3c0ad | 1691 | { |
37810659 | 1692 | clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &scpu); |
d5fd43c4 | 1693 | migrate_hrtimers(scpu); |
c0a31329 | 1694 | break; |
b2e3c0ad | 1695 | } |
c0a31329 TG |
1696 | #endif |
1697 | ||
1698 | default: | |
1699 | break; | |
1700 | } | |
1701 | ||
1702 | return NOTIFY_OK; | |
1703 | } | |
1704 | ||
8c78f307 | 1705 | static struct notifier_block __cpuinitdata hrtimers_nb = { |
c0a31329 TG |
1706 | .notifier_call = hrtimer_cpu_notify, |
1707 | }; | |
1708 | ||
1709 | void __init hrtimers_init(void) | |
1710 | { | |
1711 | hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE, | |
1712 | (void *)(long)smp_processor_id()); | |
1713 | register_cpu_notifier(&hrtimers_nb); | |
a6037b61 PZ |
1714 | #ifdef CONFIG_HIGH_RES_TIMERS |
1715 | open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq); | |
1716 | #endif | |
c0a31329 TG |
1717 | } |
1718 | ||
7bb67439 | 1719 | /** |
654c8e0b | 1720 | * schedule_hrtimeout_range - sleep until timeout |
7bb67439 | 1721 | * @expires: timeout value (ktime_t) |
654c8e0b | 1722 | * @delta: slack in expires timeout (ktime_t) |
7bb67439 AV |
1723 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL |
1724 | * | |
1725 | * Make the current task sleep until the given expiry time has | |
1726 | * elapsed. The routine will return immediately unless | |
1727 | * the current task state has been set (see set_current_state()). | |
1728 | * | |
654c8e0b AV |
1729 | * The @delta argument gives the kernel the freedom to schedule the |
1730 | * actual wakeup to a time that is both power and performance friendly. | |
1731 | * The kernel give the normal best effort behavior for "@expires+@delta", | |
1732 | * but may decide to fire the timer earlier, but no earlier than @expires. | |
1733 | * | |
7bb67439 AV |
1734 | * You can set the task state as follows - |
1735 | * | |
1736 | * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to | |
1737 | * pass before the routine returns. | |
1738 | * | |
1739 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
1740 | * delivered to the current task. | |
1741 | * | |
1742 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1743 | * routine returns. | |
1744 | * | |
1745 | * Returns 0 when the timer has expired otherwise -EINTR | |
1746 | */ | |
654c8e0b | 1747 | int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta, |
7bb67439 AV |
1748 | const enum hrtimer_mode mode) |
1749 | { | |
1750 | struct hrtimer_sleeper t; | |
1751 | ||
1752 | /* | |
1753 | * Optimize when a zero timeout value is given. It does not | |
1754 | * matter whether this is an absolute or a relative time. | |
1755 | */ | |
1756 | if (expires && !expires->tv64) { | |
1757 | __set_current_state(TASK_RUNNING); | |
1758 | return 0; | |
1759 | } | |
1760 | ||
1761 | /* | |
1762 | * A NULL parameter means "inifinte" | |
1763 | */ | |
1764 | if (!expires) { | |
1765 | schedule(); | |
1766 | __set_current_state(TASK_RUNNING); | |
1767 | return -EINTR; | |
1768 | } | |
1769 | ||
1770 | hrtimer_init_on_stack(&t.timer, CLOCK_MONOTONIC, mode); | |
654c8e0b | 1771 | hrtimer_set_expires_range_ns(&t.timer, *expires, delta); |
7bb67439 AV |
1772 | |
1773 | hrtimer_init_sleeper(&t, current); | |
1774 | ||
cc584b21 | 1775 | hrtimer_start_expires(&t.timer, mode); |
7bb67439 AV |
1776 | if (!hrtimer_active(&t.timer)) |
1777 | t.task = NULL; | |
1778 | ||
1779 | if (likely(t.task)) | |
1780 | schedule(); | |
1781 | ||
1782 | hrtimer_cancel(&t.timer); | |
1783 | destroy_hrtimer_on_stack(&t.timer); | |
1784 | ||
1785 | __set_current_state(TASK_RUNNING); | |
1786 | ||
1787 | return !t.task ? 0 : -EINTR; | |
1788 | } | |
654c8e0b AV |
1789 | EXPORT_SYMBOL_GPL(schedule_hrtimeout_range); |
1790 | ||
1791 | /** | |
1792 | * schedule_hrtimeout - sleep until timeout | |
1793 | * @expires: timeout value (ktime_t) | |
1794 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL | |
1795 | * | |
1796 | * Make the current task sleep until the given expiry time has | |
1797 | * elapsed. The routine will return immediately unless | |
1798 | * the current task state has been set (see set_current_state()). | |
1799 | * | |
1800 | * You can set the task state as follows - | |
1801 | * | |
1802 | * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to | |
1803 | * pass before the routine returns. | |
1804 | * | |
1805 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
1806 | * delivered to the current task. | |
1807 | * | |
1808 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1809 | * routine returns. | |
1810 | * | |
1811 | * Returns 0 when the timer has expired otherwise -EINTR | |
1812 | */ | |
1813 | int __sched schedule_hrtimeout(ktime_t *expires, | |
1814 | const enum hrtimer_mode mode) | |
1815 | { | |
1816 | return schedule_hrtimeout_range(expires, 0, mode); | |
1817 | } | |
7bb67439 | 1818 | EXPORT_SYMBOL_GPL(schedule_hrtimeout); |