]>
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 | 55 | * |
e06383db JS |
56 | * There are more clockids then hrtimer bases. Thus, we index |
57 | * into the timer bases by the hrtimer_base_type enum. When trying | |
58 | * to reach a base using a clockid, hrtimer_clockid_to_base() | |
59 | * is used to convert from clockid to the proper hrtimer_base_type. | |
c0a31329 | 60 | */ |
54cdfdb4 | 61 | DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) = |
c0a31329 | 62 | { |
3c8aa39d TG |
63 | |
64 | .clock_base = | |
c0a31329 | 65 | { |
3c8aa39d TG |
66 | { |
67 | .index = CLOCK_REALTIME, | |
68 | .get_time = &ktime_get_real, | |
54cdfdb4 | 69 | .resolution = KTIME_LOW_RES, |
3c8aa39d TG |
70 | }, |
71 | { | |
72 | .index = CLOCK_MONOTONIC, | |
73 | .get_time = &ktime_get, | |
54cdfdb4 | 74 | .resolution = KTIME_LOW_RES, |
3c8aa39d TG |
75 | }, |
76 | } | |
c0a31329 TG |
77 | }; |
78 | ||
e06383db JS |
79 | static int hrtimer_clock_to_base_table[MAX_CLOCKS]; |
80 | ||
81 | static inline int hrtimer_clockid_to_base(clockid_t clock_id) | |
82 | { | |
83 | return hrtimer_clock_to_base_table[clock_id]; | |
84 | } | |
85 | ||
86 | ||
92127c7a TG |
87 | /* |
88 | * Get the coarse grained time at the softirq based on xtime and | |
89 | * wall_to_monotonic. | |
90 | */ | |
3c8aa39d | 91 | static void hrtimer_get_softirq_time(struct hrtimer_cpu_base *base) |
92127c7a TG |
92 | { |
93 | ktime_t xtim, tomono; | |
ad28d94a | 94 | struct timespec xts, tom; |
92127c7a | 95 | |
48cf76f7 | 96 | get_xtime_and_monotonic_offset(&xts, &tom); |
92127c7a | 97 | |
f4304ab2 | 98 | xtim = timespec_to_ktime(xts); |
ad28d94a | 99 | tomono = timespec_to_ktime(tom); |
e06383db JS |
100 | base->clock_base[HRTIMER_BASE_REALTIME].softirq_time = xtim; |
101 | base->clock_base[HRTIMER_BASE_MONOTONIC].softirq_time = | |
3c8aa39d | 102 | ktime_add(xtim, tomono); |
92127c7a TG |
103 | } |
104 | ||
c0a31329 TG |
105 | /* |
106 | * Functions and macros which are different for UP/SMP systems are kept in a | |
107 | * single place | |
108 | */ | |
109 | #ifdef CONFIG_SMP | |
110 | ||
c0a31329 TG |
111 | /* |
112 | * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock | |
113 | * means that all timers which are tied to this base via timer->base are | |
114 | * locked, and the base itself is locked too. | |
115 | * | |
116 | * So __run_timers/migrate_timers can safely modify all timers which could | |
117 | * be found on the lists/queues. | |
118 | * | |
119 | * When the timer's base is locked, and the timer removed from list, it is | |
120 | * possible to set timer->base = NULL and drop the lock: the timer remains | |
121 | * locked. | |
122 | */ | |
3c8aa39d TG |
123 | static |
124 | struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer, | |
125 | unsigned long *flags) | |
c0a31329 | 126 | { |
3c8aa39d | 127 | struct hrtimer_clock_base *base; |
c0a31329 TG |
128 | |
129 | for (;;) { | |
130 | base = timer->base; | |
131 | if (likely(base != NULL)) { | |
ecb49d1a | 132 | raw_spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
133 | if (likely(base == timer->base)) |
134 | return base; | |
135 | /* The timer has migrated to another CPU: */ | |
ecb49d1a | 136 | raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags); |
c0a31329 TG |
137 | } |
138 | cpu_relax(); | |
139 | } | |
140 | } | |
141 | ||
6ff7041d TG |
142 | |
143 | /* | |
144 | * Get the preferred target CPU for NOHZ | |
145 | */ | |
146 | static int hrtimer_get_target(int this_cpu, int pinned) | |
147 | { | |
148 | #ifdef CONFIG_NO_HZ | |
83cd4fe2 VP |
149 | if (!pinned && get_sysctl_timer_migration() && idle_cpu(this_cpu)) |
150 | return get_nohz_timer_target(); | |
6ff7041d TG |
151 | #endif |
152 | return this_cpu; | |
153 | } | |
154 | ||
155 | /* | |
156 | * With HIGHRES=y we do not migrate the timer when it is expiring | |
157 | * before the next event on the target cpu because we cannot reprogram | |
158 | * the target cpu hardware and we would cause it to fire late. | |
159 | * | |
160 | * Called with cpu_base->lock of target cpu held. | |
161 | */ | |
162 | static int | |
163 | hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base) | |
164 | { | |
165 | #ifdef CONFIG_HIGH_RES_TIMERS | |
166 | ktime_t expires; | |
167 | ||
168 | if (!new_base->cpu_base->hres_active) | |
169 | return 0; | |
170 | ||
171 | expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset); | |
172 | return expires.tv64 <= new_base->cpu_base->expires_next.tv64; | |
173 | #else | |
174 | return 0; | |
175 | #endif | |
176 | } | |
177 | ||
c0a31329 TG |
178 | /* |
179 | * Switch the timer base to the current CPU when possible. | |
180 | */ | |
3c8aa39d | 181 | static inline struct hrtimer_clock_base * |
597d0275 AB |
182 | switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base, |
183 | int pinned) | |
c0a31329 | 184 | { |
3c8aa39d TG |
185 | struct hrtimer_clock_base *new_base; |
186 | struct hrtimer_cpu_base *new_cpu_base; | |
6ff7041d TG |
187 | int this_cpu = smp_processor_id(); |
188 | int cpu = hrtimer_get_target(this_cpu, pinned); | |
e06383db | 189 | int basenum = hrtimer_clockid_to_base(base->index); |
c0a31329 | 190 | |
eea08f32 AB |
191 | again: |
192 | new_cpu_base = &per_cpu(hrtimer_bases, cpu); | |
e06383db | 193 | new_base = &new_cpu_base->clock_base[basenum]; |
c0a31329 TG |
194 | |
195 | if (base != new_base) { | |
196 | /* | |
6ff7041d | 197 | * We are trying to move timer to new_base. |
c0a31329 TG |
198 | * However we can't change timer's base while it is running, |
199 | * so we keep it on the same CPU. No hassle vs. reprogramming | |
200 | * the event source in the high resolution case. The softirq | |
201 | * code will take care of this when the timer function has | |
202 | * completed. There is no conflict as we hold the lock until | |
203 | * the timer is enqueued. | |
204 | */ | |
54cdfdb4 | 205 | if (unlikely(hrtimer_callback_running(timer))) |
c0a31329 TG |
206 | return base; |
207 | ||
208 | /* See the comment in lock_timer_base() */ | |
209 | timer->base = NULL; | |
ecb49d1a TG |
210 | raw_spin_unlock(&base->cpu_base->lock); |
211 | raw_spin_lock(&new_base->cpu_base->lock); | |
eea08f32 | 212 | |
6ff7041d TG |
213 | if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) { |
214 | cpu = this_cpu; | |
ecb49d1a TG |
215 | raw_spin_unlock(&new_base->cpu_base->lock); |
216 | raw_spin_lock(&base->cpu_base->lock); | |
6ff7041d TG |
217 | timer->base = base; |
218 | goto again; | |
eea08f32 | 219 | } |
c0a31329 TG |
220 | timer->base = new_base; |
221 | } | |
222 | return new_base; | |
223 | } | |
224 | ||
225 | #else /* CONFIG_SMP */ | |
226 | ||
3c8aa39d | 227 | static inline struct hrtimer_clock_base * |
c0a31329 TG |
228 | lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) |
229 | { | |
3c8aa39d | 230 | struct hrtimer_clock_base *base = timer->base; |
c0a31329 | 231 | |
ecb49d1a | 232 | raw_spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
233 | |
234 | return base; | |
235 | } | |
236 | ||
eea08f32 | 237 | # define switch_hrtimer_base(t, b, p) (b) |
c0a31329 TG |
238 | |
239 | #endif /* !CONFIG_SMP */ | |
240 | ||
241 | /* | |
242 | * Functions for the union type storage format of ktime_t which are | |
243 | * too large for inlining: | |
244 | */ | |
245 | #if BITS_PER_LONG < 64 | |
246 | # ifndef CONFIG_KTIME_SCALAR | |
247 | /** | |
248 | * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable | |
c0a31329 TG |
249 | * @kt: addend |
250 | * @nsec: the scalar nsec value to add | |
251 | * | |
252 | * Returns the sum of kt and nsec in ktime_t format | |
253 | */ | |
254 | ktime_t ktime_add_ns(const ktime_t kt, u64 nsec) | |
255 | { | |
256 | ktime_t tmp; | |
257 | ||
258 | if (likely(nsec < NSEC_PER_SEC)) { | |
259 | tmp.tv64 = nsec; | |
260 | } else { | |
261 | unsigned long rem = do_div(nsec, NSEC_PER_SEC); | |
262 | ||
263 | tmp = ktime_set((long)nsec, rem); | |
264 | } | |
265 | ||
266 | return ktime_add(kt, tmp); | |
267 | } | |
b8b8fd2d DH |
268 | |
269 | EXPORT_SYMBOL_GPL(ktime_add_ns); | |
a272378d ACM |
270 | |
271 | /** | |
272 | * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable | |
273 | * @kt: minuend | |
274 | * @nsec: the scalar nsec value to subtract | |
275 | * | |
276 | * Returns the subtraction of @nsec from @kt in ktime_t format | |
277 | */ | |
278 | ktime_t ktime_sub_ns(const ktime_t kt, u64 nsec) | |
279 | { | |
280 | ktime_t tmp; | |
281 | ||
282 | if (likely(nsec < NSEC_PER_SEC)) { | |
283 | tmp.tv64 = nsec; | |
284 | } else { | |
285 | unsigned long rem = do_div(nsec, NSEC_PER_SEC); | |
286 | ||
287 | tmp = ktime_set((long)nsec, rem); | |
288 | } | |
289 | ||
290 | return ktime_sub(kt, tmp); | |
291 | } | |
292 | ||
293 | EXPORT_SYMBOL_GPL(ktime_sub_ns); | |
c0a31329 TG |
294 | # endif /* !CONFIG_KTIME_SCALAR */ |
295 | ||
296 | /* | |
297 | * Divide a ktime value by a nanosecond value | |
298 | */ | |
4d672e7a | 299 | u64 ktime_divns(const ktime_t kt, s64 div) |
c0a31329 | 300 | { |
900cfa46 | 301 | u64 dclc; |
c0a31329 TG |
302 | int sft = 0; |
303 | ||
900cfa46 | 304 | dclc = ktime_to_ns(kt); |
c0a31329 TG |
305 | /* Make sure the divisor is less than 2^32: */ |
306 | while (div >> 32) { | |
307 | sft++; | |
308 | div >>= 1; | |
309 | } | |
310 | dclc >>= sft; | |
311 | do_div(dclc, (unsigned long) div); | |
312 | ||
4d672e7a | 313 | return dclc; |
c0a31329 | 314 | } |
c0a31329 TG |
315 | #endif /* BITS_PER_LONG >= 64 */ |
316 | ||
5a7780e7 TG |
317 | /* |
318 | * Add two ktime values and do a safety check for overflow: | |
319 | */ | |
320 | ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs) | |
321 | { | |
322 | ktime_t res = ktime_add(lhs, rhs); | |
323 | ||
324 | /* | |
325 | * We use KTIME_SEC_MAX here, the maximum timeout which we can | |
326 | * return to user space in a timespec: | |
327 | */ | |
328 | if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64) | |
329 | res = ktime_set(KTIME_SEC_MAX, 0); | |
330 | ||
331 | return res; | |
332 | } | |
333 | ||
8daa21e6 AB |
334 | EXPORT_SYMBOL_GPL(ktime_add_safe); |
335 | ||
237fc6e7 TG |
336 | #ifdef CONFIG_DEBUG_OBJECTS_TIMERS |
337 | ||
338 | static struct debug_obj_descr hrtimer_debug_descr; | |
339 | ||
340 | /* | |
341 | * fixup_init is called when: | |
342 | * - an active object is initialized | |
343 | */ | |
344 | static int hrtimer_fixup_init(void *addr, enum debug_obj_state state) | |
345 | { | |
346 | struct hrtimer *timer = addr; | |
347 | ||
348 | switch (state) { | |
349 | case ODEBUG_STATE_ACTIVE: | |
350 | hrtimer_cancel(timer); | |
351 | debug_object_init(timer, &hrtimer_debug_descr); | |
352 | return 1; | |
353 | default: | |
354 | return 0; | |
355 | } | |
356 | } | |
357 | ||
358 | /* | |
359 | * fixup_activate is called when: | |
360 | * - an active object is activated | |
361 | * - an unknown object is activated (might be a statically initialized object) | |
362 | */ | |
363 | static int hrtimer_fixup_activate(void *addr, enum debug_obj_state state) | |
364 | { | |
365 | switch (state) { | |
366 | ||
367 | case ODEBUG_STATE_NOTAVAILABLE: | |
368 | WARN_ON_ONCE(1); | |
369 | return 0; | |
370 | ||
371 | case ODEBUG_STATE_ACTIVE: | |
372 | WARN_ON(1); | |
373 | ||
374 | default: | |
375 | return 0; | |
376 | } | |
377 | } | |
378 | ||
379 | /* | |
380 | * fixup_free is called when: | |
381 | * - an active object is freed | |
382 | */ | |
383 | static int hrtimer_fixup_free(void *addr, enum debug_obj_state state) | |
384 | { | |
385 | struct hrtimer *timer = addr; | |
386 | ||
387 | switch (state) { | |
388 | case ODEBUG_STATE_ACTIVE: | |
389 | hrtimer_cancel(timer); | |
390 | debug_object_free(timer, &hrtimer_debug_descr); | |
391 | return 1; | |
392 | default: | |
393 | return 0; | |
394 | } | |
395 | } | |
396 | ||
397 | static struct debug_obj_descr hrtimer_debug_descr = { | |
398 | .name = "hrtimer", | |
399 | .fixup_init = hrtimer_fixup_init, | |
400 | .fixup_activate = hrtimer_fixup_activate, | |
401 | .fixup_free = hrtimer_fixup_free, | |
402 | }; | |
403 | ||
404 | static inline void debug_hrtimer_init(struct hrtimer *timer) | |
405 | { | |
406 | debug_object_init(timer, &hrtimer_debug_descr); | |
407 | } | |
408 | ||
409 | static inline void debug_hrtimer_activate(struct hrtimer *timer) | |
410 | { | |
411 | debug_object_activate(timer, &hrtimer_debug_descr); | |
412 | } | |
413 | ||
414 | static inline void debug_hrtimer_deactivate(struct hrtimer *timer) | |
415 | { | |
416 | debug_object_deactivate(timer, &hrtimer_debug_descr); | |
417 | } | |
418 | ||
419 | static inline void debug_hrtimer_free(struct hrtimer *timer) | |
420 | { | |
421 | debug_object_free(timer, &hrtimer_debug_descr); | |
422 | } | |
423 | ||
424 | static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id, | |
425 | enum hrtimer_mode mode); | |
426 | ||
427 | void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id, | |
428 | enum hrtimer_mode mode) | |
429 | { | |
430 | debug_object_init_on_stack(timer, &hrtimer_debug_descr); | |
431 | __hrtimer_init(timer, clock_id, mode); | |
432 | } | |
2bc481cf | 433 | EXPORT_SYMBOL_GPL(hrtimer_init_on_stack); |
237fc6e7 TG |
434 | |
435 | void destroy_hrtimer_on_stack(struct hrtimer *timer) | |
436 | { | |
437 | debug_object_free(timer, &hrtimer_debug_descr); | |
438 | } | |
439 | ||
440 | #else | |
441 | static inline void debug_hrtimer_init(struct hrtimer *timer) { } | |
442 | static inline void debug_hrtimer_activate(struct hrtimer *timer) { } | |
443 | static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { } | |
444 | #endif | |
445 | ||
c6a2a177 XG |
446 | static inline void |
447 | debug_init(struct hrtimer *timer, clockid_t clockid, | |
448 | enum hrtimer_mode mode) | |
449 | { | |
450 | debug_hrtimer_init(timer); | |
451 | trace_hrtimer_init(timer, clockid, mode); | |
452 | } | |
453 | ||
454 | static inline void debug_activate(struct hrtimer *timer) | |
455 | { | |
456 | debug_hrtimer_activate(timer); | |
457 | trace_hrtimer_start(timer); | |
458 | } | |
459 | ||
460 | static inline void debug_deactivate(struct hrtimer *timer) | |
461 | { | |
462 | debug_hrtimer_deactivate(timer); | |
463 | trace_hrtimer_cancel(timer); | |
464 | } | |
465 | ||
54cdfdb4 TG |
466 | /* High resolution timer related functions */ |
467 | #ifdef CONFIG_HIGH_RES_TIMERS | |
468 | ||
469 | /* | |
470 | * High resolution timer enabled ? | |
471 | */ | |
472 | static int hrtimer_hres_enabled __read_mostly = 1; | |
473 | ||
474 | /* | |
475 | * Enable / Disable high resolution mode | |
476 | */ | |
477 | static int __init setup_hrtimer_hres(char *str) | |
478 | { | |
479 | if (!strcmp(str, "off")) | |
480 | hrtimer_hres_enabled = 0; | |
481 | else if (!strcmp(str, "on")) | |
482 | hrtimer_hres_enabled = 1; | |
483 | else | |
484 | return 0; | |
485 | return 1; | |
486 | } | |
487 | ||
488 | __setup("highres=", setup_hrtimer_hres); | |
489 | ||
490 | /* | |
491 | * hrtimer_high_res_enabled - query, if the highres mode is enabled | |
492 | */ | |
493 | static inline int hrtimer_is_hres_enabled(void) | |
494 | { | |
495 | return hrtimer_hres_enabled; | |
496 | } | |
497 | ||
498 | /* | |
499 | * Is the high resolution mode active ? | |
500 | */ | |
501 | static inline int hrtimer_hres_active(void) | |
502 | { | |
909ea964 | 503 | return __this_cpu_read(hrtimer_bases.hres_active); |
54cdfdb4 TG |
504 | } |
505 | ||
506 | /* | |
507 | * Reprogram the event source with checking both queues for the | |
508 | * next event | |
509 | * Called with interrupts disabled and base->lock held | |
510 | */ | |
7403f41f AC |
511 | static void |
512 | hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal) | |
54cdfdb4 TG |
513 | { |
514 | int i; | |
515 | struct hrtimer_clock_base *base = cpu_base->clock_base; | |
7403f41f | 516 | ktime_t expires, expires_next; |
54cdfdb4 | 517 | |
7403f41f | 518 | expires_next.tv64 = KTIME_MAX; |
54cdfdb4 TG |
519 | |
520 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) { | |
521 | struct hrtimer *timer; | |
998adc3d | 522 | struct timerqueue_node *next; |
54cdfdb4 | 523 | |
998adc3d JS |
524 | next = timerqueue_getnext(&base->active); |
525 | if (!next) | |
54cdfdb4 | 526 | continue; |
998adc3d JS |
527 | timer = container_of(next, struct hrtimer, node); |
528 | ||
cc584b21 | 529 | expires = ktime_sub(hrtimer_get_expires(timer), base->offset); |
b0a9b511 TG |
530 | /* |
531 | * clock_was_set() has changed base->offset so the | |
532 | * result might be negative. Fix it up to prevent a | |
533 | * false positive in clockevents_program_event() | |
534 | */ | |
535 | if (expires.tv64 < 0) | |
536 | expires.tv64 = 0; | |
7403f41f AC |
537 | if (expires.tv64 < expires_next.tv64) |
538 | expires_next = expires; | |
54cdfdb4 TG |
539 | } |
540 | ||
7403f41f AC |
541 | if (skip_equal && expires_next.tv64 == cpu_base->expires_next.tv64) |
542 | return; | |
543 | ||
544 | cpu_base->expires_next.tv64 = expires_next.tv64; | |
545 | ||
54cdfdb4 TG |
546 | if (cpu_base->expires_next.tv64 != KTIME_MAX) |
547 | tick_program_event(cpu_base->expires_next, 1); | |
548 | } | |
549 | ||
550 | /* | |
551 | * Shared reprogramming for clock_realtime and clock_monotonic | |
552 | * | |
553 | * When a timer is enqueued and expires earlier than the already enqueued | |
554 | * timers, we have to check, whether it expires earlier than the timer for | |
555 | * which the clock event device was armed. | |
556 | * | |
557 | * Called with interrupts disabled and base->cpu_base.lock held | |
558 | */ | |
559 | static int hrtimer_reprogram(struct hrtimer *timer, | |
560 | struct hrtimer_clock_base *base) | |
561 | { | |
41d2e494 | 562 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
cc584b21 | 563 | ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset); |
54cdfdb4 TG |
564 | int res; |
565 | ||
cc584b21 | 566 | WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0); |
63070a79 | 567 | |
54cdfdb4 TG |
568 | /* |
569 | * When the callback is running, we do not reprogram the clock event | |
570 | * device. The timer callback is either running on a different CPU or | |
3a4fa0a2 | 571 | * the callback is executed in the hrtimer_interrupt context. The |
54cdfdb4 TG |
572 | * reprogramming is handled either by the softirq, which called the |
573 | * callback or at the end of the hrtimer_interrupt. | |
574 | */ | |
575 | if (hrtimer_callback_running(timer)) | |
576 | return 0; | |
577 | ||
63070a79 TG |
578 | /* |
579 | * CLOCK_REALTIME timer might be requested with an absolute | |
580 | * expiry time which is less than base->offset. Nothing wrong | |
581 | * about that, just avoid to call into the tick code, which | |
582 | * has now objections against negative expiry values. | |
583 | */ | |
584 | if (expires.tv64 < 0) | |
585 | return -ETIME; | |
586 | ||
41d2e494 TG |
587 | if (expires.tv64 >= cpu_base->expires_next.tv64) |
588 | return 0; | |
589 | ||
590 | /* | |
591 | * If a hang was detected in the last timer interrupt then we | |
592 | * do not schedule a timer which is earlier than the expiry | |
593 | * which we enforced in the hang detection. We want the system | |
594 | * to make progress. | |
595 | */ | |
596 | if (cpu_base->hang_detected) | |
54cdfdb4 TG |
597 | return 0; |
598 | ||
599 | /* | |
600 | * Clockevents returns -ETIME, when the event was in the past. | |
601 | */ | |
602 | res = tick_program_event(expires, 0); | |
603 | if (!IS_ERR_VALUE(res)) | |
41d2e494 | 604 | cpu_base->expires_next = expires; |
54cdfdb4 TG |
605 | return res; |
606 | } | |
607 | ||
608 | ||
609 | /* | |
610 | * Retrigger next event is called after clock was set | |
611 | * | |
612 | * Called with interrupts disabled via on_each_cpu() | |
613 | */ | |
614 | static void retrigger_next_event(void *arg) | |
615 | { | |
616 | struct hrtimer_cpu_base *base; | |
8ab4351a | 617 | struct timespec realtime_offset, wtm; |
54cdfdb4 TG |
618 | |
619 | if (!hrtimer_hres_active()) | |
620 | return; | |
621 | ||
48cf76f7 | 622 | get_xtime_and_monotonic_offset(&realtime_offset, &wtm); |
8ab4351a | 623 | set_normalized_timespec(&realtime_offset, -wtm.tv_sec, -wtm.tv_nsec); |
54cdfdb4 TG |
624 | |
625 | base = &__get_cpu_var(hrtimer_bases); | |
626 | ||
627 | /* Adjust CLOCK_REALTIME offset */ | |
ecb49d1a | 628 | raw_spin_lock(&base->lock); |
e06383db | 629 | base->clock_base[HRTIMER_BASE_REALTIME].offset = |
54cdfdb4 TG |
630 | timespec_to_ktime(realtime_offset); |
631 | ||
7403f41f | 632 | hrtimer_force_reprogram(base, 0); |
ecb49d1a | 633 | raw_spin_unlock(&base->lock); |
54cdfdb4 TG |
634 | } |
635 | ||
636 | /* | |
637 | * Clock realtime was set | |
638 | * | |
639 | * Change the offset of the realtime clock vs. the monotonic | |
640 | * clock. | |
641 | * | |
642 | * We might have to reprogram the high resolution timer interrupt. On | |
643 | * SMP we call the architecture specific code to retrigger _all_ high | |
644 | * resolution timer interrupts. On UP we just disable interrupts and | |
645 | * call the high resolution interrupt code. | |
646 | */ | |
647 | void clock_was_set(void) | |
648 | { | |
649 | /* Retrigger the CPU local events everywhere */ | |
15c8b6c1 | 650 | on_each_cpu(retrigger_next_event, NULL, 1); |
54cdfdb4 TG |
651 | } |
652 | ||
995f054f IM |
653 | /* |
654 | * During resume we might have to reprogram the high resolution timer | |
655 | * interrupt (on the local CPU): | |
656 | */ | |
657 | void hres_timers_resume(void) | |
658 | { | |
1d4a7f1c PZ |
659 | WARN_ONCE(!irqs_disabled(), |
660 | KERN_INFO "hres_timers_resume() called with IRQs enabled!"); | |
661 | ||
995f054f IM |
662 | retrigger_next_event(NULL); |
663 | } | |
664 | ||
54cdfdb4 TG |
665 | /* |
666 | * Initialize the high resolution related parts of cpu_base | |
667 | */ | |
668 | static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) | |
669 | { | |
670 | base->expires_next.tv64 = KTIME_MAX; | |
671 | base->hres_active = 0; | |
54cdfdb4 TG |
672 | } |
673 | ||
674 | /* | |
675 | * Initialize the high resolution related parts of a hrtimer | |
676 | */ | |
677 | static inline void hrtimer_init_timer_hres(struct hrtimer *timer) | |
678 | { | |
54cdfdb4 TG |
679 | } |
680 | ||
ca109491 | 681 | |
54cdfdb4 TG |
682 | /* |
683 | * When High resolution timers are active, try to reprogram. Note, that in case | |
684 | * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry | |
685 | * check happens. The timer gets enqueued into the rbtree. The reprogramming | |
686 | * and expiry check is done in the hrtimer_interrupt or in the softirq. | |
687 | */ | |
688 | static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer, | |
7f1e2ca9 PZ |
689 | struct hrtimer_clock_base *base, |
690 | int wakeup) | |
54cdfdb4 TG |
691 | { |
692 | if (base->cpu_base->hres_active && hrtimer_reprogram(timer, base)) { | |
7f1e2ca9 | 693 | if (wakeup) { |
ecb49d1a | 694 | raw_spin_unlock(&base->cpu_base->lock); |
7f1e2ca9 | 695 | raise_softirq_irqoff(HRTIMER_SOFTIRQ); |
ecb49d1a | 696 | raw_spin_lock(&base->cpu_base->lock); |
7f1e2ca9 PZ |
697 | } else |
698 | __raise_softirq_irqoff(HRTIMER_SOFTIRQ); | |
699 | ||
ca109491 | 700 | return 1; |
54cdfdb4 | 701 | } |
7f1e2ca9 | 702 | |
54cdfdb4 TG |
703 | return 0; |
704 | } | |
705 | ||
706 | /* | |
707 | * Switch to high resolution mode | |
708 | */ | |
f8953856 | 709 | static int hrtimer_switch_to_hres(void) |
54cdfdb4 | 710 | { |
820de5c3 IM |
711 | int cpu = smp_processor_id(); |
712 | struct hrtimer_cpu_base *base = &per_cpu(hrtimer_bases, cpu); | |
54cdfdb4 TG |
713 | unsigned long flags; |
714 | ||
715 | if (base->hres_active) | |
f8953856 | 716 | return 1; |
54cdfdb4 TG |
717 | |
718 | local_irq_save(flags); | |
719 | ||
720 | if (tick_init_highres()) { | |
721 | local_irq_restore(flags); | |
820de5c3 IM |
722 | printk(KERN_WARNING "Could not switch to high resolution " |
723 | "mode on CPU %d\n", cpu); | |
f8953856 | 724 | return 0; |
54cdfdb4 TG |
725 | } |
726 | base->hres_active = 1; | |
e06383db JS |
727 | base->clock_base[HRTIMER_BASE_REALTIME].resolution = KTIME_HIGH_RES; |
728 | base->clock_base[HRTIMER_BASE_MONOTONIC].resolution = KTIME_HIGH_RES; | |
54cdfdb4 TG |
729 | |
730 | tick_setup_sched_timer(); | |
731 | ||
732 | /* "Retrigger" the interrupt to get things going */ | |
733 | retrigger_next_event(NULL); | |
734 | local_irq_restore(flags); | |
f8953856 | 735 | return 1; |
54cdfdb4 TG |
736 | } |
737 | ||
738 | #else | |
739 | ||
740 | static inline int hrtimer_hres_active(void) { return 0; } | |
741 | static inline int hrtimer_is_hres_enabled(void) { return 0; } | |
f8953856 | 742 | static inline int hrtimer_switch_to_hres(void) { return 0; } |
7403f41f AC |
743 | static inline void |
744 | hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { } | |
54cdfdb4 | 745 | static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer, |
7f1e2ca9 PZ |
746 | struct hrtimer_clock_base *base, |
747 | int wakeup) | |
54cdfdb4 TG |
748 | { |
749 | return 0; | |
750 | } | |
54cdfdb4 TG |
751 | static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { } |
752 | static inline void hrtimer_init_timer_hres(struct hrtimer *timer) { } | |
753 | ||
754 | #endif /* CONFIG_HIGH_RES_TIMERS */ | |
755 | ||
5f201907 | 756 | static inline void timer_stats_hrtimer_set_start_info(struct hrtimer *timer) |
82f67cd9 | 757 | { |
5f201907 | 758 | #ifdef CONFIG_TIMER_STATS |
82f67cd9 IM |
759 | if (timer->start_site) |
760 | return; | |
5f201907 | 761 | timer->start_site = __builtin_return_address(0); |
82f67cd9 IM |
762 | memcpy(timer->start_comm, current->comm, TASK_COMM_LEN); |
763 | timer->start_pid = current->pid; | |
5f201907 HC |
764 | #endif |
765 | } | |
766 | ||
767 | static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer *timer) | |
768 | { | |
769 | #ifdef CONFIG_TIMER_STATS | |
770 | timer->start_site = NULL; | |
771 | #endif | |
82f67cd9 | 772 | } |
5f201907 HC |
773 | |
774 | static inline void timer_stats_account_hrtimer(struct hrtimer *timer) | |
775 | { | |
776 | #ifdef CONFIG_TIMER_STATS | |
777 | if (likely(!timer_stats_active)) | |
778 | return; | |
779 | timer_stats_update_stats(timer, timer->start_pid, timer->start_site, | |
780 | timer->function, timer->start_comm, 0); | |
82f67cd9 | 781 | #endif |
5f201907 | 782 | } |
82f67cd9 | 783 | |
c0a31329 | 784 | /* |
6506f2aa | 785 | * Counterpart to lock_hrtimer_base above: |
c0a31329 TG |
786 | */ |
787 | static inline | |
788 | void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) | |
789 | { | |
ecb49d1a | 790 | raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags); |
c0a31329 TG |
791 | } |
792 | ||
793 | /** | |
794 | * hrtimer_forward - forward the timer expiry | |
c0a31329 | 795 | * @timer: hrtimer to forward |
44f21475 | 796 | * @now: forward past this time |
c0a31329 TG |
797 | * @interval: the interval to forward |
798 | * | |
799 | * Forward the timer expiry so it will expire in the future. | |
8dca6f33 | 800 | * Returns the number of overruns. |
c0a31329 | 801 | */ |
4d672e7a | 802 | u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval) |
c0a31329 | 803 | { |
4d672e7a | 804 | u64 orun = 1; |
44f21475 | 805 | ktime_t delta; |
c0a31329 | 806 | |
cc584b21 | 807 | delta = ktime_sub(now, hrtimer_get_expires(timer)); |
c0a31329 TG |
808 | |
809 | if (delta.tv64 < 0) | |
810 | return 0; | |
811 | ||
c9db4fa1 TG |
812 | if (interval.tv64 < timer->base->resolution.tv64) |
813 | interval.tv64 = timer->base->resolution.tv64; | |
814 | ||
c0a31329 | 815 | if (unlikely(delta.tv64 >= interval.tv64)) { |
df869b63 | 816 | s64 incr = ktime_to_ns(interval); |
c0a31329 TG |
817 | |
818 | orun = ktime_divns(delta, incr); | |
cc584b21 AV |
819 | hrtimer_add_expires_ns(timer, incr * orun); |
820 | if (hrtimer_get_expires_tv64(timer) > now.tv64) | |
c0a31329 TG |
821 | return orun; |
822 | /* | |
823 | * This (and the ktime_add() below) is the | |
824 | * correction for exact: | |
825 | */ | |
826 | orun++; | |
827 | } | |
cc584b21 | 828 | hrtimer_add_expires(timer, interval); |
c0a31329 TG |
829 | |
830 | return orun; | |
831 | } | |
6bdb6b62 | 832 | EXPORT_SYMBOL_GPL(hrtimer_forward); |
c0a31329 TG |
833 | |
834 | /* | |
835 | * enqueue_hrtimer - internal function to (re)start a timer | |
836 | * | |
837 | * The timer is inserted in expiry order. Insertion into the | |
838 | * red black tree is O(log(n)). Must hold the base lock. | |
a6037b61 PZ |
839 | * |
840 | * Returns 1 when the new timer is the leftmost timer in the tree. | |
c0a31329 | 841 | */ |
a6037b61 PZ |
842 | static int enqueue_hrtimer(struct hrtimer *timer, |
843 | struct hrtimer_clock_base *base) | |
c0a31329 | 844 | { |
c6a2a177 | 845 | debug_activate(timer); |
237fc6e7 | 846 | |
998adc3d | 847 | timerqueue_add(&base->active, &timer->node); |
54cdfdb4 | 848 | |
303e967f TG |
849 | /* |
850 | * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the | |
851 | * state of a possibly running callback. | |
852 | */ | |
853 | timer->state |= HRTIMER_STATE_ENQUEUED; | |
a6037b61 | 854 | |
998adc3d | 855 | return (&timer->node == base->active.next); |
288867ec | 856 | } |
c0a31329 TG |
857 | |
858 | /* | |
859 | * __remove_hrtimer - internal function to remove a timer | |
860 | * | |
861 | * Caller must hold the base lock. | |
54cdfdb4 TG |
862 | * |
863 | * High resolution timer mode reprograms the clock event device when the | |
864 | * timer is the one which expires next. The caller can disable this by setting | |
865 | * reprogram to zero. This is useful, when the context does a reprogramming | |
866 | * anyway (e.g. timer interrupt) | |
c0a31329 | 867 | */ |
3c8aa39d | 868 | static void __remove_hrtimer(struct hrtimer *timer, |
303e967f | 869 | struct hrtimer_clock_base *base, |
54cdfdb4 | 870 | unsigned long newstate, int reprogram) |
c0a31329 | 871 | { |
7403f41f AC |
872 | if (!(timer->state & HRTIMER_STATE_ENQUEUED)) |
873 | goto out; | |
874 | ||
998adc3d | 875 | if (&timer->node == timerqueue_getnext(&base->active)) { |
7403f41f AC |
876 | #ifdef CONFIG_HIGH_RES_TIMERS |
877 | /* Reprogram the clock event device. if enabled */ | |
878 | if (reprogram && hrtimer_hres_active()) { | |
879 | ktime_t expires; | |
880 | ||
881 | expires = ktime_sub(hrtimer_get_expires(timer), | |
882 | base->offset); | |
883 | if (base->cpu_base->expires_next.tv64 == expires.tv64) | |
884 | hrtimer_force_reprogram(base->cpu_base, 1); | |
54cdfdb4 | 885 | } |
7403f41f | 886 | #endif |
54cdfdb4 | 887 | } |
998adc3d | 888 | timerqueue_del(&base->active, &timer->node); |
7403f41f | 889 | out: |
303e967f | 890 | timer->state = newstate; |
c0a31329 TG |
891 | } |
892 | ||
893 | /* | |
894 | * remove hrtimer, called with base lock held | |
895 | */ | |
896 | static inline int | |
3c8aa39d | 897 | remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base) |
c0a31329 | 898 | { |
303e967f | 899 | if (hrtimer_is_queued(timer)) { |
f13d4f97 | 900 | unsigned long state; |
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 | 913 | reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases); |
f13d4f97 SQ |
914 | /* |
915 | * We must preserve the CALLBACK state flag here, | |
916 | * otherwise we could move the timer base in | |
917 | * switch_hrtimer_base. | |
918 | */ | |
919 | state = timer->state & HRTIMER_STATE_CALLBACK; | |
920 | __remove_hrtimer(timer, base, state, reprogram); | |
c0a31329 TG |
921 | return 1; |
922 | } | |
923 | return 0; | |
924 | } | |
925 | ||
7f1e2ca9 PZ |
926 | int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, |
927 | unsigned long delta_ns, const enum hrtimer_mode mode, | |
928 | int wakeup) | |
c0a31329 | 929 | { |
3c8aa39d | 930 | struct hrtimer_clock_base *base, *new_base; |
c0a31329 | 931 | unsigned long flags; |
a6037b61 | 932 | int ret, leftmost; |
c0a31329 TG |
933 | |
934 | base = lock_hrtimer_base(timer, &flags); | |
935 | ||
936 | /* Remove an active timer from the queue: */ | |
937 | ret = remove_hrtimer(timer, base); | |
938 | ||
939 | /* Switch the timer base, if necessary: */ | |
597d0275 | 940 | new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED); |
c0a31329 | 941 | |
597d0275 | 942 | if (mode & HRTIMER_MODE_REL) { |
5a7780e7 | 943 | tim = ktime_add_safe(tim, new_base->get_time()); |
06027bdd IM |
944 | /* |
945 | * CONFIG_TIME_LOW_RES is a temporary way for architectures | |
946 | * to signal that they simply return xtime in | |
947 | * do_gettimeoffset(). In this case we want to round up by | |
948 | * resolution when starting a relative timer, to avoid short | |
949 | * timeouts. This will go away with the GTOD framework. | |
950 | */ | |
951 | #ifdef CONFIG_TIME_LOW_RES | |
5a7780e7 | 952 | tim = ktime_add_safe(tim, base->resolution); |
06027bdd IM |
953 | #endif |
954 | } | |
237fc6e7 | 955 | |
da8f2e17 | 956 | hrtimer_set_expires_range_ns(timer, tim, delta_ns); |
c0a31329 | 957 | |
82f67cd9 IM |
958 | timer_stats_hrtimer_set_start_info(timer); |
959 | ||
a6037b61 PZ |
960 | leftmost = enqueue_hrtimer(timer, new_base); |
961 | ||
935c631d IM |
962 | /* |
963 | * Only allow reprogramming if the new base is on this CPU. | |
964 | * (it might still be on another CPU if the timer was pending) | |
a6037b61 PZ |
965 | * |
966 | * XXX send_remote_softirq() ? | |
935c631d | 967 | */ |
a6037b61 | 968 | if (leftmost && new_base->cpu_base == &__get_cpu_var(hrtimer_bases)) |
7f1e2ca9 | 969 | hrtimer_enqueue_reprogram(timer, new_base, wakeup); |
c0a31329 TG |
970 | |
971 | unlock_hrtimer_base(timer, &flags); | |
972 | ||
973 | return ret; | |
974 | } | |
7f1e2ca9 PZ |
975 | |
976 | /** | |
977 | * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU | |
978 | * @timer: the timer to be added | |
979 | * @tim: expiry time | |
980 | * @delta_ns: "slack" range for the timer | |
981 | * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL) | |
982 | * | |
983 | * Returns: | |
984 | * 0 on success | |
985 | * 1 when the timer was active | |
986 | */ | |
987 | int hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, | |
988 | unsigned long delta_ns, const enum hrtimer_mode mode) | |
989 | { | |
990 | return __hrtimer_start_range_ns(timer, tim, delta_ns, mode, 1); | |
991 | } | |
da8f2e17 AV |
992 | EXPORT_SYMBOL_GPL(hrtimer_start_range_ns); |
993 | ||
994 | /** | |
e1dd7bc5 | 995 | * hrtimer_start - (re)start an hrtimer on the current CPU |
da8f2e17 AV |
996 | * @timer: the timer to be added |
997 | * @tim: expiry time | |
998 | * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL) | |
999 | * | |
1000 | * Returns: | |
1001 | * 0 on success | |
1002 | * 1 when the timer was active | |
1003 | */ | |
1004 | int | |
1005 | hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode) | |
1006 | { | |
7f1e2ca9 | 1007 | return __hrtimer_start_range_ns(timer, tim, 0, mode, 1); |
da8f2e17 | 1008 | } |
8d16b764 | 1009 | EXPORT_SYMBOL_GPL(hrtimer_start); |
c0a31329 | 1010 | |
da8f2e17 | 1011 | |
c0a31329 TG |
1012 | /** |
1013 | * hrtimer_try_to_cancel - try to deactivate a timer | |
c0a31329 TG |
1014 | * @timer: hrtimer to stop |
1015 | * | |
1016 | * Returns: | |
1017 | * 0 when the timer was not active | |
1018 | * 1 when the timer was active | |
1019 | * -1 when the timer is currently excuting the callback function and | |
fa9799e3 | 1020 | * cannot be stopped |
c0a31329 TG |
1021 | */ |
1022 | int hrtimer_try_to_cancel(struct hrtimer *timer) | |
1023 | { | |
3c8aa39d | 1024 | struct hrtimer_clock_base *base; |
c0a31329 TG |
1025 | unsigned long flags; |
1026 | int ret = -1; | |
1027 | ||
1028 | base = lock_hrtimer_base(timer, &flags); | |
1029 | ||
303e967f | 1030 | if (!hrtimer_callback_running(timer)) |
c0a31329 TG |
1031 | ret = remove_hrtimer(timer, base); |
1032 | ||
1033 | unlock_hrtimer_base(timer, &flags); | |
1034 | ||
1035 | return ret; | |
1036 | ||
1037 | } | |
8d16b764 | 1038 | EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel); |
c0a31329 TG |
1039 | |
1040 | /** | |
1041 | * hrtimer_cancel - cancel a timer and wait for the handler to finish. | |
c0a31329 TG |
1042 | * @timer: the timer to be cancelled |
1043 | * | |
1044 | * Returns: | |
1045 | * 0 when the timer was not active | |
1046 | * 1 when the timer was active | |
1047 | */ | |
1048 | int hrtimer_cancel(struct hrtimer *timer) | |
1049 | { | |
1050 | for (;;) { | |
1051 | int ret = hrtimer_try_to_cancel(timer); | |
1052 | ||
1053 | if (ret >= 0) | |
1054 | return ret; | |
5ef37b19 | 1055 | cpu_relax(); |
c0a31329 TG |
1056 | } |
1057 | } | |
8d16b764 | 1058 | EXPORT_SYMBOL_GPL(hrtimer_cancel); |
c0a31329 TG |
1059 | |
1060 | /** | |
1061 | * hrtimer_get_remaining - get remaining time for the timer | |
c0a31329 TG |
1062 | * @timer: the timer to read |
1063 | */ | |
1064 | ktime_t hrtimer_get_remaining(const struct hrtimer *timer) | |
1065 | { | |
c0a31329 TG |
1066 | unsigned long flags; |
1067 | ktime_t rem; | |
1068 | ||
b3bd3de6 | 1069 | lock_hrtimer_base(timer, &flags); |
cc584b21 | 1070 | rem = hrtimer_expires_remaining(timer); |
c0a31329 TG |
1071 | unlock_hrtimer_base(timer, &flags); |
1072 | ||
1073 | return rem; | |
1074 | } | |
8d16b764 | 1075 | EXPORT_SYMBOL_GPL(hrtimer_get_remaining); |
c0a31329 | 1076 | |
ee9c5785 | 1077 | #ifdef CONFIG_NO_HZ |
69239749 TL |
1078 | /** |
1079 | * hrtimer_get_next_event - get the time until next expiry event | |
1080 | * | |
1081 | * Returns the delta to the next expiry event or KTIME_MAX if no timer | |
1082 | * is pending. | |
1083 | */ | |
1084 | ktime_t hrtimer_get_next_event(void) | |
1085 | { | |
3c8aa39d TG |
1086 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
1087 | struct hrtimer_clock_base *base = cpu_base->clock_base; | |
69239749 TL |
1088 | ktime_t delta, mindelta = { .tv64 = KTIME_MAX }; |
1089 | unsigned long flags; | |
1090 | int i; | |
1091 | ||
ecb49d1a | 1092 | raw_spin_lock_irqsave(&cpu_base->lock, flags); |
3c8aa39d | 1093 | |
54cdfdb4 TG |
1094 | if (!hrtimer_hres_active()) { |
1095 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) { | |
1096 | struct hrtimer *timer; | |
998adc3d | 1097 | struct timerqueue_node *next; |
69239749 | 1098 | |
998adc3d JS |
1099 | next = timerqueue_getnext(&base->active); |
1100 | if (!next) | |
54cdfdb4 | 1101 | continue; |
3c8aa39d | 1102 | |
998adc3d | 1103 | timer = container_of(next, struct hrtimer, node); |
cc584b21 | 1104 | delta.tv64 = hrtimer_get_expires_tv64(timer); |
54cdfdb4 TG |
1105 | delta = ktime_sub(delta, base->get_time()); |
1106 | if (delta.tv64 < mindelta.tv64) | |
1107 | mindelta.tv64 = delta.tv64; | |
1108 | } | |
69239749 | 1109 | } |
3c8aa39d | 1110 | |
ecb49d1a | 1111 | raw_spin_unlock_irqrestore(&cpu_base->lock, flags); |
3c8aa39d | 1112 | |
69239749 TL |
1113 | if (mindelta.tv64 < 0) |
1114 | mindelta.tv64 = 0; | |
1115 | return mindelta; | |
1116 | } | |
1117 | #endif | |
1118 | ||
237fc6e7 TG |
1119 | static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id, |
1120 | enum hrtimer_mode mode) | |
c0a31329 | 1121 | { |
3c8aa39d | 1122 | struct hrtimer_cpu_base *cpu_base; |
e06383db | 1123 | int base; |
c0a31329 | 1124 | |
7978672c GA |
1125 | memset(timer, 0, sizeof(struct hrtimer)); |
1126 | ||
3c8aa39d | 1127 | cpu_base = &__raw_get_cpu_var(hrtimer_bases); |
c0a31329 | 1128 | |
c9cb2e3d | 1129 | if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS) |
7978672c GA |
1130 | clock_id = CLOCK_MONOTONIC; |
1131 | ||
e06383db JS |
1132 | base = hrtimer_clockid_to_base(clock_id); |
1133 | timer->base = &cpu_base->clock_base[base]; | |
54cdfdb4 | 1134 | hrtimer_init_timer_hres(timer); |
998adc3d | 1135 | timerqueue_init(&timer->node); |
82f67cd9 IM |
1136 | |
1137 | #ifdef CONFIG_TIMER_STATS | |
1138 | timer->start_site = NULL; | |
1139 | timer->start_pid = -1; | |
1140 | memset(timer->start_comm, 0, TASK_COMM_LEN); | |
1141 | #endif | |
c0a31329 | 1142 | } |
237fc6e7 TG |
1143 | |
1144 | /** | |
1145 | * hrtimer_init - initialize a timer to the given clock | |
1146 | * @timer: the timer to be initialized | |
1147 | * @clock_id: the clock to be used | |
1148 | * @mode: timer mode abs/rel | |
1149 | */ | |
1150 | void hrtimer_init(struct hrtimer *timer, clockid_t clock_id, | |
1151 | enum hrtimer_mode mode) | |
1152 | { | |
c6a2a177 | 1153 | debug_init(timer, clock_id, mode); |
237fc6e7 TG |
1154 | __hrtimer_init(timer, clock_id, mode); |
1155 | } | |
8d16b764 | 1156 | EXPORT_SYMBOL_GPL(hrtimer_init); |
c0a31329 TG |
1157 | |
1158 | /** | |
1159 | * hrtimer_get_res - get the timer resolution for a clock | |
c0a31329 TG |
1160 | * @which_clock: which clock to query |
1161 | * @tp: pointer to timespec variable to store the resolution | |
1162 | * | |
72fd4a35 RD |
1163 | * Store the resolution of the clock selected by @which_clock in the |
1164 | * variable pointed to by @tp. | |
c0a31329 TG |
1165 | */ |
1166 | int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp) | |
1167 | { | |
3c8aa39d | 1168 | struct hrtimer_cpu_base *cpu_base; |
e06383db | 1169 | int base = hrtimer_clockid_to_base(which_clock); |
c0a31329 | 1170 | |
3c8aa39d | 1171 | cpu_base = &__raw_get_cpu_var(hrtimer_bases); |
e06383db | 1172 | *tp = ktime_to_timespec(cpu_base->clock_base[base].resolution); |
c0a31329 TG |
1173 | |
1174 | return 0; | |
1175 | } | |
8d16b764 | 1176 | EXPORT_SYMBOL_GPL(hrtimer_get_res); |
c0a31329 | 1177 | |
c6a2a177 | 1178 | static void __run_hrtimer(struct hrtimer *timer, ktime_t *now) |
d3d74453 PZ |
1179 | { |
1180 | struct hrtimer_clock_base *base = timer->base; | |
1181 | struct hrtimer_cpu_base *cpu_base = base->cpu_base; | |
1182 | enum hrtimer_restart (*fn)(struct hrtimer *); | |
1183 | int restart; | |
1184 | ||
ca109491 PZ |
1185 | WARN_ON(!irqs_disabled()); |
1186 | ||
c6a2a177 | 1187 | debug_deactivate(timer); |
d3d74453 PZ |
1188 | __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0); |
1189 | timer_stats_account_hrtimer(timer); | |
d3d74453 | 1190 | fn = timer->function; |
ca109491 PZ |
1191 | |
1192 | /* | |
1193 | * Because we run timers from hardirq context, there is no chance | |
1194 | * they get migrated to another cpu, therefore its safe to unlock | |
1195 | * the timer base. | |
1196 | */ | |
ecb49d1a | 1197 | raw_spin_unlock(&cpu_base->lock); |
c6a2a177 | 1198 | trace_hrtimer_expire_entry(timer, now); |
ca109491 | 1199 | restart = fn(timer); |
c6a2a177 | 1200 | trace_hrtimer_expire_exit(timer); |
ecb49d1a | 1201 | raw_spin_lock(&cpu_base->lock); |
d3d74453 PZ |
1202 | |
1203 | /* | |
e3f1d883 TG |
1204 | * Note: We clear the CALLBACK bit after enqueue_hrtimer and |
1205 | * we do not reprogramm the event hardware. Happens either in | |
1206 | * hrtimer_start_range_ns() or in hrtimer_interrupt() | |
d3d74453 PZ |
1207 | */ |
1208 | if (restart != HRTIMER_NORESTART) { | |
1209 | BUG_ON(timer->state != HRTIMER_STATE_CALLBACK); | |
a6037b61 | 1210 | enqueue_hrtimer(timer, base); |
d3d74453 | 1211 | } |
f13d4f97 SQ |
1212 | |
1213 | WARN_ON_ONCE(!(timer->state & HRTIMER_STATE_CALLBACK)); | |
1214 | ||
d3d74453 PZ |
1215 | timer->state &= ~HRTIMER_STATE_CALLBACK; |
1216 | } | |
1217 | ||
54cdfdb4 TG |
1218 | #ifdef CONFIG_HIGH_RES_TIMERS |
1219 | ||
1220 | /* | |
1221 | * High resolution timer interrupt | |
1222 | * Called with interrupts disabled | |
1223 | */ | |
1224 | void hrtimer_interrupt(struct clock_event_device *dev) | |
1225 | { | |
1226 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); | |
1227 | struct hrtimer_clock_base *base; | |
41d2e494 TG |
1228 | ktime_t expires_next, now, entry_time, delta; |
1229 | int i, retries = 0; | |
54cdfdb4 TG |
1230 | |
1231 | BUG_ON(!cpu_base->hres_active); | |
1232 | cpu_base->nr_events++; | |
1233 | dev->next_event.tv64 = KTIME_MAX; | |
1234 | ||
41d2e494 TG |
1235 | entry_time = now = ktime_get(); |
1236 | retry: | |
54cdfdb4 TG |
1237 | expires_next.tv64 = KTIME_MAX; |
1238 | ||
ecb49d1a | 1239 | raw_spin_lock(&cpu_base->lock); |
6ff7041d TG |
1240 | /* |
1241 | * We set expires_next to KTIME_MAX here with cpu_base->lock | |
1242 | * held to prevent that a timer is enqueued in our queue via | |
1243 | * the migration code. This does not affect enqueueing of | |
1244 | * timers which run their callback and need to be requeued on | |
1245 | * this CPU. | |
1246 | */ | |
1247 | cpu_base->expires_next.tv64 = KTIME_MAX; | |
1248 | ||
54cdfdb4 TG |
1249 | base = cpu_base->clock_base; |
1250 | ||
1251 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { | |
1252 | ktime_t basenow; | |
998adc3d | 1253 | struct timerqueue_node *node; |
54cdfdb4 | 1254 | |
54cdfdb4 TG |
1255 | basenow = ktime_add(now, base->offset); |
1256 | ||
998adc3d | 1257 | while ((node = timerqueue_getnext(&base->active))) { |
54cdfdb4 TG |
1258 | struct hrtimer *timer; |
1259 | ||
998adc3d | 1260 | timer = container_of(node, struct hrtimer, node); |
54cdfdb4 | 1261 | |
654c8e0b AV |
1262 | /* |
1263 | * The immediate goal for using the softexpires is | |
1264 | * minimizing wakeups, not running timers at the | |
1265 | * earliest interrupt after their soft expiration. | |
1266 | * This allows us to avoid using a Priority Search | |
1267 | * Tree, which can answer a stabbing querry for | |
1268 | * overlapping intervals and instead use the simple | |
1269 | * BST we already have. | |
1270 | * We don't add extra wakeups by delaying timers that | |
1271 | * are right-of a not yet expired timer, because that | |
1272 | * timer will have to trigger a wakeup anyway. | |
1273 | */ | |
1274 | ||
1275 | if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer)) { | |
54cdfdb4 TG |
1276 | ktime_t expires; |
1277 | ||
cc584b21 | 1278 | expires = ktime_sub(hrtimer_get_expires(timer), |
54cdfdb4 TG |
1279 | base->offset); |
1280 | if (expires.tv64 < expires_next.tv64) | |
1281 | expires_next = expires; | |
1282 | break; | |
1283 | } | |
1284 | ||
c6a2a177 | 1285 | __run_hrtimer(timer, &basenow); |
54cdfdb4 | 1286 | } |
54cdfdb4 TG |
1287 | base++; |
1288 | } | |
1289 | ||
6ff7041d TG |
1290 | /* |
1291 | * Store the new expiry value so the migration code can verify | |
1292 | * against it. | |
1293 | */ | |
54cdfdb4 | 1294 | cpu_base->expires_next = expires_next; |
ecb49d1a | 1295 | raw_spin_unlock(&cpu_base->lock); |
54cdfdb4 TG |
1296 | |
1297 | /* Reprogramming necessary ? */ | |
41d2e494 TG |
1298 | if (expires_next.tv64 == KTIME_MAX || |
1299 | !tick_program_event(expires_next, 0)) { | |
1300 | cpu_base->hang_detected = 0; | |
1301 | return; | |
54cdfdb4 | 1302 | } |
41d2e494 TG |
1303 | |
1304 | /* | |
1305 | * The next timer was already expired due to: | |
1306 | * - tracing | |
1307 | * - long lasting callbacks | |
1308 | * - being scheduled away when running in a VM | |
1309 | * | |
1310 | * We need to prevent that we loop forever in the hrtimer | |
1311 | * interrupt routine. We give it 3 attempts to avoid | |
1312 | * overreacting on some spurious event. | |
1313 | */ | |
1314 | now = ktime_get(); | |
1315 | cpu_base->nr_retries++; | |
1316 | if (++retries < 3) | |
1317 | goto retry; | |
1318 | /* | |
1319 | * Give the system a chance to do something else than looping | |
1320 | * here. We stored the entry time, so we know exactly how long | |
1321 | * we spent here. We schedule the next event this amount of | |
1322 | * time away. | |
1323 | */ | |
1324 | cpu_base->nr_hangs++; | |
1325 | cpu_base->hang_detected = 1; | |
1326 | delta = ktime_sub(now, entry_time); | |
1327 | if (delta.tv64 > cpu_base->max_hang_time.tv64) | |
1328 | cpu_base->max_hang_time = delta; | |
1329 | /* | |
1330 | * Limit it to a sensible value as we enforce a longer | |
1331 | * delay. Give the CPU at least 100ms to catch up. | |
1332 | */ | |
1333 | if (delta.tv64 > 100 * NSEC_PER_MSEC) | |
1334 | expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC); | |
1335 | else | |
1336 | expires_next = ktime_add(now, delta); | |
1337 | tick_program_event(expires_next, 1); | |
1338 | printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n", | |
1339 | ktime_to_ns(delta)); | |
54cdfdb4 TG |
1340 | } |
1341 | ||
8bdec955 TG |
1342 | /* |
1343 | * local version of hrtimer_peek_ahead_timers() called with interrupts | |
1344 | * disabled. | |
1345 | */ | |
1346 | static void __hrtimer_peek_ahead_timers(void) | |
1347 | { | |
1348 | struct tick_device *td; | |
1349 | ||
1350 | if (!hrtimer_hres_active()) | |
1351 | return; | |
1352 | ||
1353 | td = &__get_cpu_var(tick_cpu_device); | |
1354 | if (td && td->evtdev) | |
1355 | hrtimer_interrupt(td->evtdev); | |
1356 | } | |
1357 | ||
2e94d1f7 AV |
1358 | /** |
1359 | * hrtimer_peek_ahead_timers -- run soft-expired timers now | |
1360 | * | |
1361 | * hrtimer_peek_ahead_timers will peek at the timer queue of | |
1362 | * the current cpu and check if there are any timers for which | |
1363 | * the soft expires time has passed. If any such timers exist, | |
1364 | * they are run immediately and then removed from the timer queue. | |
1365 | * | |
1366 | */ | |
1367 | void hrtimer_peek_ahead_timers(void) | |
1368 | { | |
643bdf68 | 1369 | unsigned long flags; |
dc4304f7 | 1370 | |
2e94d1f7 | 1371 | local_irq_save(flags); |
8bdec955 | 1372 | __hrtimer_peek_ahead_timers(); |
2e94d1f7 AV |
1373 | local_irq_restore(flags); |
1374 | } | |
1375 | ||
a6037b61 PZ |
1376 | static void run_hrtimer_softirq(struct softirq_action *h) |
1377 | { | |
1378 | hrtimer_peek_ahead_timers(); | |
1379 | } | |
1380 | ||
82c5b7b5 IM |
1381 | #else /* CONFIG_HIGH_RES_TIMERS */ |
1382 | ||
1383 | static inline void __hrtimer_peek_ahead_timers(void) { } | |
1384 | ||
1385 | #endif /* !CONFIG_HIGH_RES_TIMERS */ | |
82f67cd9 | 1386 | |
d3d74453 PZ |
1387 | /* |
1388 | * Called from timer softirq every jiffy, expire hrtimers: | |
1389 | * | |
1390 | * For HRT its the fall back code to run the softirq in the timer | |
1391 | * softirq context in case the hrtimer initialization failed or has | |
1392 | * not been done yet. | |
1393 | */ | |
1394 | void hrtimer_run_pending(void) | |
1395 | { | |
d3d74453 PZ |
1396 | if (hrtimer_hres_active()) |
1397 | return; | |
54cdfdb4 | 1398 | |
d3d74453 PZ |
1399 | /* |
1400 | * This _is_ ugly: We have to check in the softirq context, | |
1401 | * whether we can switch to highres and / or nohz mode. The | |
1402 | * clocksource switch happens in the timer interrupt with | |
1403 | * xtime_lock held. Notification from there only sets the | |
1404 | * check bit in the tick_oneshot code, otherwise we might | |
1405 | * deadlock vs. xtime_lock. | |
1406 | */ | |
1407 | if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) | |
1408 | hrtimer_switch_to_hres(); | |
54cdfdb4 TG |
1409 | } |
1410 | ||
c0a31329 | 1411 | /* |
d3d74453 | 1412 | * Called from hardirq context every jiffy |
c0a31329 | 1413 | */ |
833883d9 | 1414 | void hrtimer_run_queues(void) |
c0a31329 | 1415 | { |
998adc3d | 1416 | struct timerqueue_node *node; |
833883d9 DS |
1417 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
1418 | struct hrtimer_clock_base *base; | |
1419 | int index, gettime = 1; | |
c0a31329 | 1420 | |
833883d9 | 1421 | if (hrtimer_hres_active()) |
3055adda DS |
1422 | return; |
1423 | ||
833883d9 DS |
1424 | for (index = 0; index < HRTIMER_MAX_CLOCK_BASES; index++) { |
1425 | base = &cpu_base->clock_base[index]; | |
b007c389 | 1426 | if (!timerqueue_getnext(&base->active)) |
d3d74453 | 1427 | continue; |
833883d9 | 1428 | |
d7cfb60c | 1429 | if (gettime) { |
833883d9 DS |
1430 | hrtimer_get_softirq_time(cpu_base); |
1431 | gettime = 0; | |
b75f7a51 | 1432 | } |
d3d74453 | 1433 | |
ecb49d1a | 1434 | raw_spin_lock(&cpu_base->lock); |
c0a31329 | 1435 | |
b007c389 | 1436 | while ((node = timerqueue_getnext(&base->active))) { |
833883d9 | 1437 | struct hrtimer *timer; |
54cdfdb4 | 1438 | |
998adc3d | 1439 | timer = container_of(node, struct hrtimer, node); |
cc584b21 AV |
1440 | if (base->softirq_time.tv64 <= |
1441 | hrtimer_get_expires_tv64(timer)) | |
833883d9 DS |
1442 | break; |
1443 | ||
c6a2a177 | 1444 | __run_hrtimer(timer, &base->softirq_time); |
833883d9 | 1445 | } |
ecb49d1a | 1446 | raw_spin_unlock(&cpu_base->lock); |
833883d9 | 1447 | } |
c0a31329 TG |
1448 | } |
1449 | ||
10c94ec1 TG |
1450 | /* |
1451 | * Sleep related functions: | |
1452 | */ | |
c9cb2e3d | 1453 | static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer) |
00362e33 TG |
1454 | { |
1455 | struct hrtimer_sleeper *t = | |
1456 | container_of(timer, struct hrtimer_sleeper, timer); | |
1457 | struct task_struct *task = t->task; | |
1458 | ||
1459 | t->task = NULL; | |
1460 | if (task) | |
1461 | wake_up_process(task); | |
1462 | ||
1463 | return HRTIMER_NORESTART; | |
1464 | } | |
1465 | ||
36c8b586 | 1466 | void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task) |
00362e33 TG |
1467 | { |
1468 | sl->timer.function = hrtimer_wakeup; | |
1469 | sl->task = task; | |
1470 | } | |
2bc481cf | 1471 | EXPORT_SYMBOL_GPL(hrtimer_init_sleeper); |
00362e33 | 1472 | |
669d7868 | 1473 | static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode) |
432569bb | 1474 | { |
669d7868 | 1475 | hrtimer_init_sleeper(t, current); |
10c94ec1 | 1476 | |
432569bb RZ |
1477 | do { |
1478 | set_current_state(TASK_INTERRUPTIBLE); | |
cc584b21 | 1479 | hrtimer_start_expires(&t->timer, mode); |
37bb6cb4 PZ |
1480 | if (!hrtimer_active(&t->timer)) |
1481 | t->task = NULL; | |
432569bb | 1482 | |
54cdfdb4 TG |
1483 | if (likely(t->task)) |
1484 | schedule(); | |
432569bb | 1485 | |
669d7868 | 1486 | hrtimer_cancel(&t->timer); |
c9cb2e3d | 1487 | mode = HRTIMER_MODE_ABS; |
669d7868 TG |
1488 | |
1489 | } while (t->task && !signal_pending(current)); | |
432569bb | 1490 | |
3588a085 PZ |
1491 | __set_current_state(TASK_RUNNING); |
1492 | ||
669d7868 | 1493 | return t->task == NULL; |
10c94ec1 TG |
1494 | } |
1495 | ||
080344b9 ON |
1496 | static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp) |
1497 | { | |
1498 | struct timespec rmt; | |
1499 | ktime_t rem; | |
1500 | ||
cc584b21 | 1501 | rem = hrtimer_expires_remaining(timer); |
080344b9 ON |
1502 | if (rem.tv64 <= 0) |
1503 | return 0; | |
1504 | rmt = ktime_to_timespec(rem); | |
1505 | ||
1506 | if (copy_to_user(rmtp, &rmt, sizeof(*rmtp))) | |
1507 | return -EFAULT; | |
1508 | ||
1509 | return 1; | |
1510 | } | |
1511 | ||
1711ef38 | 1512 | long __sched hrtimer_nanosleep_restart(struct restart_block *restart) |
10c94ec1 | 1513 | { |
669d7868 | 1514 | struct hrtimer_sleeper t; |
080344b9 | 1515 | struct timespec __user *rmtp; |
237fc6e7 | 1516 | int ret = 0; |
10c94ec1 | 1517 | |
237fc6e7 TG |
1518 | hrtimer_init_on_stack(&t.timer, restart->nanosleep.index, |
1519 | HRTIMER_MODE_ABS); | |
cc584b21 | 1520 | hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires); |
10c94ec1 | 1521 | |
c9cb2e3d | 1522 | if (do_nanosleep(&t, HRTIMER_MODE_ABS)) |
237fc6e7 | 1523 | goto out; |
10c94ec1 | 1524 | |
029a07e0 | 1525 | rmtp = restart->nanosleep.rmtp; |
432569bb | 1526 | if (rmtp) { |
237fc6e7 | 1527 | ret = update_rmtp(&t.timer, rmtp); |
080344b9 | 1528 | if (ret <= 0) |
237fc6e7 | 1529 | goto out; |
432569bb | 1530 | } |
10c94ec1 | 1531 | |
10c94ec1 | 1532 | /* The other values in restart are already filled in */ |
237fc6e7 TG |
1533 | ret = -ERESTART_RESTARTBLOCK; |
1534 | out: | |
1535 | destroy_hrtimer_on_stack(&t.timer); | |
1536 | return ret; | |
10c94ec1 TG |
1537 | } |
1538 | ||
080344b9 | 1539 | long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp, |
10c94ec1 TG |
1540 | const enum hrtimer_mode mode, const clockid_t clockid) |
1541 | { | |
1542 | struct restart_block *restart; | |
669d7868 | 1543 | struct hrtimer_sleeper t; |
237fc6e7 | 1544 | int ret = 0; |
3bd01206 AV |
1545 | unsigned long slack; |
1546 | ||
1547 | slack = current->timer_slack_ns; | |
1548 | if (rt_task(current)) | |
1549 | slack = 0; | |
10c94ec1 | 1550 | |
237fc6e7 | 1551 | hrtimer_init_on_stack(&t.timer, clockid, mode); |
3bd01206 | 1552 | hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack); |
432569bb | 1553 | if (do_nanosleep(&t, mode)) |
237fc6e7 | 1554 | goto out; |
10c94ec1 | 1555 | |
7978672c | 1556 | /* Absolute timers do not update the rmtp value and restart: */ |
237fc6e7 TG |
1557 | if (mode == HRTIMER_MODE_ABS) { |
1558 | ret = -ERESTARTNOHAND; | |
1559 | goto out; | |
1560 | } | |
10c94ec1 | 1561 | |
432569bb | 1562 | if (rmtp) { |
237fc6e7 | 1563 | ret = update_rmtp(&t.timer, rmtp); |
080344b9 | 1564 | if (ret <= 0) |
237fc6e7 | 1565 | goto out; |
432569bb | 1566 | } |
10c94ec1 TG |
1567 | |
1568 | restart = ¤t_thread_info()->restart_block; | |
1711ef38 | 1569 | restart->fn = hrtimer_nanosleep_restart; |
029a07e0 TG |
1570 | restart->nanosleep.index = t.timer.base->index; |
1571 | restart->nanosleep.rmtp = rmtp; | |
cc584b21 | 1572 | restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer); |
10c94ec1 | 1573 | |
237fc6e7 TG |
1574 | ret = -ERESTART_RESTARTBLOCK; |
1575 | out: | |
1576 | destroy_hrtimer_on_stack(&t.timer); | |
1577 | return ret; | |
10c94ec1 TG |
1578 | } |
1579 | ||
58fd3aa2 HC |
1580 | SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp, |
1581 | struct timespec __user *, rmtp) | |
6ba1b912 | 1582 | { |
080344b9 | 1583 | struct timespec tu; |
6ba1b912 TG |
1584 | |
1585 | if (copy_from_user(&tu, rqtp, sizeof(tu))) | |
1586 | return -EFAULT; | |
1587 | ||
1588 | if (!timespec_valid(&tu)) | |
1589 | return -EINVAL; | |
1590 | ||
080344b9 | 1591 | return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC); |
6ba1b912 TG |
1592 | } |
1593 | ||
c0a31329 TG |
1594 | /* |
1595 | * Functions related to boot-time initialization: | |
1596 | */ | |
0ec160dd | 1597 | static void __cpuinit init_hrtimers_cpu(int cpu) |
c0a31329 | 1598 | { |
3c8aa39d | 1599 | struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu); |
c0a31329 TG |
1600 | int i; |
1601 | ||
ecb49d1a | 1602 | raw_spin_lock_init(&cpu_base->lock); |
3c8aa39d | 1603 | |
998adc3d | 1604 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
3c8aa39d | 1605 | cpu_base->clock_base[i].cpu_base = cpu_base; |
998adc3d JS |
1606 | timerqueue_init_head(&cpu_base->clock_base[i].active); |
1607 | } | |
3c8aa39d | 1608 | |
54cdfdb4 | 1609 | hrtimer_init_hres(cpu_base); |
c0a31329 TG |
1610 | } |
1611 | ||
1612 | #ifdef CONFIG_HOTPLUG_CPU | |
1613 | ||
ca109491 | 1614 | static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base, |
37810659 | 1615 | struct hrtimer_clock_base *new_base) |
c0a31329 TG |
1616 | { |
1617 | struct hrtimer *timer; | |
998adc3d | 1618 | struct timerqueue_node *node; |
c0a31329 | 1619 | |
998adc3d JS |
1620 | while ((node = timerqueue_getnext(&old_base->active))) { |
1621 | timer = container_of(node, struct hrtimer, node); | |
54cdfdb4 | 1622 | BUG_ON(hrtimer_callback_running(timer)); |
c6a2a177 | 1623 | debug_deactivate(timer); |
b00c1a99 TG |
1624 | |
1625 | /* | |
1626 | * Mark it as STATE_MIGRATE not INACTIVE otherwise the | |
1627 | * timer could be seen as !active and just vanish away | |
1628 | * under us on another CPU | |
1629 | */ | |
1630 | __remove_hrtimer(timer, old_base, HRTIMER_STATE_MIGRATE, 0); | |
c0a31329 | 1631 | timer->base = new_base; |
54cdfdb4 | 1632 | /* |
e3f1d883 TG |
1633 | * Enqueue the timers on the new cpu. This does not |
1634 | * reprogram the event device in case the timer | |
1635 | * expires before the earliest on this CPU, but we run | |
1636 | * hrtimer_interrupt after we migrated everything to | |
1637 | * sort out already expired timers and reprogram the | |
1638 | * event device. | |
54cdfdb4 | 1639 | */ |
a6037b61 | 1640 | enqueue_hrtimer(timer, new_base); |
41e1022e | 1641 | |
b00c1a99 TG |
1642 | /* Clear the migration state bit */ |
1643 | timer->state &= ~HRTIMER_STATE_MIGRATE; | |
c0a31329 TG |
1644 | } |
1645 | } | |
1646 | ||
d5fd43c4 | 1647 | static void migrate_hrtimers(int scpu) |
c0a31329 | 1648 | { |
3c8aa39d | 1649 | struct hrtimer_cpu_base *old_base, *new_base; |
731a55ba | 1650 | int i; |
c0a31329 | 1651 | |
37810659 | 1652 | BUG_ON(cpu_online(scpu)); |
37810659 | 1653 | tick_cancel_sched_timer(scpu); |
731a55ba TG |
1654 | |
1655 | local_irq_disable(); | |
1656 | old_base = &per_cpu(hrtimer_bases, scpu); | |
1657 | new_base = &__get_cpu_var(hrtimer_bases); | |
d82f0b0f ON |
1658 | /* |
1659 | * The caller is globally serialized and nobody else | |
1660 | * takes two locks at once, deadlock is not possible. | |
1661 | */ | |
ecb49d1a TG |
1662 | raw_spin_lock(&new_base->lock); |
1663 | raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING); | |
c0a31329 | 1664 | |
3c8aa39d | 1665 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
ca109491 | 1666 | migrate_hrtimer_list(&old_base->clock_base[i], |
37810659 | 1667 | &new_base->clock_base[i]); |
c0a31329 TG |
1668 | } |
1669 | ||
ecb49d1a TG |
1670 | raw_spin_unlock(&old_base->lock); |
1671 | raw_spin_unlock(&new_base->lock); | |
37810659 | 1672 | |
731a55ba TG |
1673 | /* Check, if we got expired work to do */ |
1674 | __hrtimer_peek_ahead_timers(); | |
1675 | local_irq_enable(); | |
c0a31329 | 1676 | } |
37810659 | 1677 | |
c0a31329 TG |
1678 | #endif /* CONFIG_HOTPLUG_CPU */ |
1679 | ||
8c78f307 | 1680 | static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self, |
c0a31329 TG |
1681 | unsigned long action, void *hcpu) |
1682 | { | |
b2e3c0ad | 1683 | int scpu = (long)hcpu; |
c0a31329 TG |
1684 | |
1685 | switch (action) { | |
1686 | ||
1687 | case CPU_UP_PREPARE: | |
8bb78442 | 1688 | case CPU_UP_PREPARE_FROZEN: |
37810659 | 1689 | init_hrtimers_cpu(scpu); |
c0a31329 TG |
1690 | break; |
1691 | ||
1692 | #ifdef CONFIG_HOTPLUG_CPU | |
94df7de0 SD |
1693 | case CPU_DYING: |
1694 | case CPU_DYING_FROZEN: | |
1695 | clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DYING, &scpu); | |
1696 | break; | |
c0a31329 | 1697 | case CPU_DEAD: |
8bb78442 | 1698 | case CPU_DEAD_FROZEN: |
b2e3c0ad | 1699 | { |
37810659 | 1700 | clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &scpu); |
d5fd43c4 | 1701 | migrate_hrtimers(scpu); |
c0a31329 | 1702 | break; |
b2e3c0ad | 1703 | } |
c0a31329 TG |
1704 | #endif |
1705 | ||
1706 | default: | |
1707 | break; | |
1708 | } | |
1709 | ||
1710 | return NOTIFY_OK; | |
1711 | } | |
1712 | ||
8c78f307 | 1713 | static struct notifier_block __cpuinitdata hrtimers_nb = { |
c0a31329 TG |
1714 | .notifier_call = hrtimer_cpu_notify, |
1715 | }; | |
1716 | ||
1717 | void __init hrtimers_init(void) | |
1718 | { | |
e06383db JS |
1719 | hrtimer_clock_to_base_table[CLOCK_REALTIME] = HRTIMER_BASE_REALTIME; |
1720 | hrtimer_clock_to_base_table[CLOCK_MONOTONIC] = HRTIMER_BASE_MONOTONIC; | |
1721 | ||
c0a31329 TG |
1722 | hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE, |
1723 | (void *)(long)smp_processor_id()); | |
1724 | register_cpu_notifier(&hrtimers_nb); | |
a6037b61 PZ |
1725 | #ifdef CONFIG_HIGH_RES_TIMERS |
1726 | open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq); | |
1727 | #endif | |
c0a31329 TG |
1728 | } |
1729 | ||
7bb67439 | 1730 | /** |
351b3f7a | 1731 | * schedule_hrtimeout_range_clock - sleep until timeout |
7bb67439 | 1732 | * @expires: timeout value (ktime_t) |
654c8e0b | 1733 | * @delta: slack in expires timeout (ktime_t) |
7bb67439 | 1734 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL |
351b3f7a | 1735 | * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME |
7bb67439 | 1736 | */ |
351b3f7a CE |
1737 | int __sched |
1738 | schedule_hrtimeout_range_clock(ktime_t *expires, unsigned long delta, | |
1739 | const enum hrtimer_mode mode, int clock) | |
7bb67439 AV |
1740 | { |
1741 | struct hrtimer_sleeper t; | |
1742 | ||
1743 | /* | |
1744 | * Optimize when a zero timeout value is given. It does not | |
1745 | * matter whether this is an absolute or a relative time. | |
1746 | */ | |
1747 | if (expires && !expires->tv64) { | |
1748 | __set_current_state(TASK_RUNNING); | |
1749 | return 0; | |
1750 | } | |
1751 | ||
1752 | /* | |
43b21013 | 1753 | * A NULL parameter means "infinite" |
7bb67439 AV |
1754 | */ |
1755 | if (!expires) { | |
1756 | schedule(); | |
1757 | __set_current_state(TASK_RUNNING); | |
1758 | return -EINTR; | |
1759 | } | |
1760 | ||
351b3f7a | 1761 | hrtimer_init_on_stack(&t.timer, clock, mode); |
654c8e0b | 1762 | hrtimer_set_expires_range_ns(&t.timer, *expires, delta); |
7bb67439 AV |
1763 | |
1764 | hrtimer_init_sleeper(&t, current); | |
1765 | ||
cc584b21 | 1766 | hrtimer_start_expires(&t.timer, mode); |
7bb67439 AV |
1767 | if (!hrtimer_active(&t.timer)) |
1768 | t.task = NULL; | |
1769 | ||
1770 | if (likely(t.task)) | |
1771 | schedule(); | |
1772 | ||
1773 | hrtimer_cancel(&t.timer); | |
1774 | destroy_hrtimer_on_stack(&t.timer); | |
1775 | ||
1776 | __set_current_state(TASK_RUNNING); | |
1777 | ||
1778 | return !t.task ? 0 : -EINTR; | |
1779 | } | |
351b3f7a CE |
1780 | |
1781 | /** | |
1782 | * schedule_hrtimeout_range - sleep until timeout | |
1783 | * @expires: timeout value (ktime_t) | |
1784 | * @delta: slack in expires timeout (ktime_t) | |
1785 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL | |
1786 | * | |
1787 | * Make the current task sleep until the given expiry time has | |
1788 | * elapsed. The routine will return immediately unless | |
1789 | * the current task state has been set (see set_current_state()). | |
1790 | * | |
1791 | * The @delta argument gives the kernel the freedom to schedule the | |
1792 | * actual wakeup to a time that is both power and performance friendly. | |
1793 | * The kernel give the normal best effort behavior for "@expires+@delta", | |
1794 | * but may decide to fire the timer earlier, but no earlier than @expires. | |
1795 | * | |
1796 | * You can set the task state as follows - | |
1797 | * | |
1798 | * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to | |
1799 | * pass before the routine returns. | |
1800 | * | |
1801 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
1802 | * delivered to the current task. | |
1803 | * | |
1804 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1805 | * routine returns. | |
1806 | * | |
1807 | * Returns 0 when the timer has expired otherwise -EINTR | |
1808 | */ | |
1809 | int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta, | |
1810 | const enum hrtimer_mode mode) | |
1811 | { | |
1812 | return schedule_hrtimeout_range_clock(expires, delta, mode, | |
1813 | CLOCK_MONOTONIC); | |
1814 | } | |
654c8e0b AV |
1815 | EXPORT_SYMBOL_GPL(schedule_hrtimeout_range); |
1816 | ||
1817 | /** | |
1818 | * schedule_hrtimeout - sleep until timeout | |
1819 | * @expires: timeout value (ktime_t) | |
1820 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL | |
1821 | * | |
1822 | * Make the current task sleep until the given expiry time has | |
1823 | * elapsed. The routine will return immediately unless | |
1824 | * the current task state has been set (see set_current_state()). | |
1825 | * | |
1826 | * You can set the task state as follows - | |
1827 | * | |
1828 | * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to | |
1829 | * pass before the routine returns. | |
1830 | * | |
1831 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
1832 | * delivered to the current task. | |
1833 | * | |
1834 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1835 | * routine returns. | |
1836 | * | |
1837 | * Returns 0 when the timer has expired otherwise -EINTR | |
1838 | */ | |
1839 | int __sched schedule_hrtimeout(ktime_t *expires, | |
1840 | const enum hrtimer_mode mode) | |
1841 | { | |
1842 | return schedule_hrtimeout_range(expires, 0, mode); | |
1843 | } | |
7bb67439 | 1844 | EXPORT_SYMBOL_GPL(schedule_hrtimeout); |