]>
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 | 75 | }, |
70a08cca JS |
76 | { |
77 | .index = CLOCK_BOOTTIME, | |
78 | .get_time = &ktime_get_boottime, | |
79 | .resolution = KTIME_LOW_RES, | |
80 | }, | |
3c8aa39d | 81 | } |
c0a31329 TG |
82 | }; |
83 | ||
942c3c5c | 84 | static const int hrtimer_clock_to_base_table[MAX_CLOCKS] = { |
ce31332d TG |
85 | [CLOCK_REALTIME] = HRTIMER_BASE_REALTIME, |
86 | [CLOCK_MONOTONIC] = HRTIMER_BASE_MONOTONIC, | |
87 | [CLOCK_BOOTTIME] = HRTIMER_BASE_BOOTTIME, | |
88 | }; | |
e06383db JS |
89 | |
90 | static inline int hrtimer_clockid_to_base(clockid_t clock_id) | |
91 | { | |
92 | return hrtimer_clock_to_base_table[clock_id]; | |
93 | } | |
94 | ||
95 | ||
92127c7a TG |
96 | /* |
97 | * Get the coarse grained time at the softirq based on xtime and | |
98 | * wall_to_monotonic. | |
99 | */ | |
3c8aa39d | 100 | static void hrtimer_get_softirq_time(struct hrtimer_cpu_base *base) |
92127c7a | 101 | { |
70a08cca | 102 | ktime_t xtim, mono, boot; |
314ac371 | 103 | struct timespec xts, tom, slp; |
92127c7a | 104 | |
314ac371 | 105 | get_xtime_and_monotonic_and_sleep_offset(&xts, &tom, &slp); |
92127c7a | 106 | |
f4304ab2 | 107 | xtim = timespec_to_ktime(xts); |
70a08cca JS |
108 | mono = ktime_add(xtim, timespec_to_ktime(tom)); |
109 | boot = ktime_add(mono, timespec_to_ktime(slp)); | |
e06383db | 110 | base->clock_base[HRTIMER_BASE_REALTIME].softirq_time = xtim; |
70a08cca JS |
111 | base->clock_base[HRTIMER_BASE_MONOTONIC].softirq_time = mono; |
112 | base->clock_base[HRTIMER_BASE_BOOTTIME].softirq_time = boot; | |
92127c7a TG |
113 | } |
114 | ||
c0a31329 TG |
115 | /* |
116 | * Functions and macros which are different for UP/SMP systems are kept in a | |
117 | * single place | |
118 | */ | |
119 | #ifdef CONFIG_SMP | |
120 | ||
c0a31329 TG |
121 | /* |
122 | * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock | |
123 | * means that all timers which are tied to this base via timer->base are | |
124 | * locked, and the base itself is locked too. | |
125 | * | |
126 | * So __run_timers/migrate_timers can safely modify all timers which could | |
127 | * be found on the lists/queues. | |
128 | * | |
129 | * When the timer's base is locked, and the timer removed from list, it is | |
130 | * possible to set timer->base = NULL and drop the lock: the timer remains | |
131 | * locked. | |
132 | */ | |
3c8aa39d TG |
133 | static |
134 | struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer, | |
135 | unsigned long *flags) | |
c0a31329 | 136 | { |
3c8aa39d | 137 | struct hrtimer_clock_base *base; |
c0a31329 TG |
138 | |
139 | for (;;) { | |
140 | base = timer->base; | |
141 | if (likely(base != NULL)) { | |
ecb49d1a | 142 | raw_spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
143 | if (likely(base == timer->base)) |
144 | return base; | |
145 | /* The timer has migrated to another CPU: */ | |
ecb49d1a | 146 | raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags); |
c0a31329 TG |
147 | } |
148 | cpu_relax(); | |
149 | } | |
150 | } | |
151 | ||
6ff7041d TG |
152 | |
153 | /* | |
154 | * Get the preferred target CPU for NOHZ | |
155 | */ | |
156 | static int hrtimer_get_target(int this_cpu, int pinned) | |
157 | { | |
158 | #ifdef CONFIG_NO_HZ | |
83cd4fe2 VP |
159 | if (!pinned && get_sysctl_timer_migration() && idle_cpu(this_cpu)) |
160 | return get_nohz_timer_target(); | |
6ff7041d TG |
161 | #endif |
162 | return this_cpu; | |
163 | } | |
164 | ||
165 | /* | |
166 | * With HIGHRES=y we do not migrate the timer when it is expiring | |
167 | * before the next event on the target cpu because we cannot reprogram | |
168 | * the target cpu hardware and we would cause it to fire late. | |
169 | * | |
170 | * Called with cpu_base->lock of target cpu held. | |
171 | */ | |
172 | static int | |
173 | hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base) | |
174 | { | |
175 | #ifdef CONFIG_HIGH_RES_TIMERS | |
176 | ktime_t expires; | |
177 | ||
178 | if (!new_base->cpu_base->hres_active) | |
179 | return 0; | |
180 | ||
181 | expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset); | |
182 | return expires.tv64 <= new_base->cpu_base->expires_next.tv64; | |
183 | #else | |
184 | return 0; | |
185 | #endif | |
186 | } | |
187 | ||
c0a31329 TG |
188 | /* |
189 | * Switch the timer base to the current CPU when possible. | |
190 | */ | |
3c8aa39d | 191 | static inline struct hrtimer_clock_base * |
597d0275 AB |
192 | switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base, |
193 | int pinned) | |
c0a31329 | 194 | { |
3c8aa39d TG |
195 | struct hrtimer_clock_base *new_base; |
196 | struct hrtimer_cpu_base *new_cpu_base; | |
6ff7041d TG |
197 | int this_cpu = smp_processor_id(); |
198 | int cpu = hrtimer_get_target(this_cpu, pinned); | |
e06383db | 199 | int basenum = hrtimer_clockid_to_base(base->index); |
c0a31329 | 200 | |
eea08f32 AB |
201 | again: |
202 | new_cpu_base = &per_cpu(hrtimer_bases, cpu); | |
e06383db | 203 | new_base = &new_cpu_base->clock_base[basenum]; |
c0a31329 TG |
204 | |
205 | if (base != new_base) { | |
206 | /* | |
6ff7041d | 207 | * We are trying to move timer to new_base. |
c0a31329 TG |
208 | * However we can't change timer's base while it is running, |
209 | * so we keep it on the same CPU. No hassle vs. reprogramming | |
210 | * the event source in the high resolution case. The softirq | |
211 | * code will take care of this when the timer function has | |
212 | * completed. There is no conflict as we hold the lock until | |
213 | * the timer is enqueued. | |
214 | */ | |
54cdfdb4 | 215 | if (unlikely(hrtimer_callback_running(timer))) |
c0a31329 TG |
216 | return base; |
217 | ||
218 | /* See the comment in lock_timer_base() */ | |
219 | timer->base = NULL; | |
ecb49d1a TG |
220 | raw_spin_unlock(&base->cpu_base->lock); |
221 | raw_spin_lock(&new_base->cpu_base->lock); | |
eea08f32 | 222 | |
6ff7041d TG |
223 | if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) { |
224 | cpu = this_cpu; | |
ecb49d1a TG |
225 | raw_spin_unlock(&new_base->cpu_base->lock); |
226 | raw_spin_lock(&base->cpu_base->lock); | |
6ff7041d TG |
227 | timer->base = base; |
228 | goto again; | |
eea08f32 | 229 | } |
c0a31329 TG |
230 | timer->base = new_base; |
231 | } | |
232 | return new_base; | |
233 | } | |
234 | ||
235 | #else /* CONFIG_SMP */ | |
236 | ||
3c8aa39d | 237 | static inline struct hrtimer_clock_base * |
c0a31329 TG |
238 | lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) |
239 | { | |
3c8aa39d | 240 | struct hrtimer_clock_base *base = timer->base; |
c0a31329 | 241 | |
ecb49d1a | 242 | raw_spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
243 | |
244 | return base; | |
245 | } | |
246 | ||
eea08f32 | 247 | # define switch_hrtimer_base(t, b, p) (b) |
c0a31329 TG |
248 | |
249 | #endif /* !CONFIG_SMP */ | |
250 | ||
251 | /* | |
252 | * Functions for the union type storage format of ktime_t which are | |
253 | * too large for inlining: | |
254 | */ | |
255 | #if BITS_PER_LONG < 64 | |
256 | # ifndef CONFIG_KTIME_SCALAR | |
257 | /** | |
258 | * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable | |
c0a31329 TG |
259 | * @kt: addend |
260 | * @nsec: the scalar nsec value to add | |
261 | * | |
262 | * Returns the sum of kt and nsec in ktime_t format | |
263 | */ | |
264 | ktime_t ktime_add_ns(const ktime_t kt, u64 nsec) | |
265 | { | |
266 | ktime_t tmp; | |
267 | ||
268 | if (likely(nsec < NSEC_PER_SEC)) { | |
269 | tmp.tv64 = nsec; | |
270 | } else { | |
271 | unsigned long rem = do_div(nsec, NSEC_PER_SEC); | |
272 | ||
273 | tmp = ktime_set((long)nsec, rem); | |
274 | } | |
275 | ||
276 | return ktime_add(kt, tmp); | |
277 | } | |
b8b8fd2d DH |
278 | |
279 | EXPORT_SYMBOL_GPL(ktime_add_ns); | |
a272378d ACM |
280 | |
281 | /** | |
282 | * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable | |
283 | * @kt: minuend | |
284 | * @nsec: the scalar nsec value to subtract | |
285 | * | |
286 | * Returns the subtraction of @nsec from @kt in ktime_t format | |
287 | */ | |
288 | ktime_t ktime_sub_ns(const ktime_t kt, u64 nsec) | |
289 | { | |
290 | ktime_t tmp; | |
291 | ||
292 | if (likely(nsec < NSEC_PER_SEC)) { | |
293 | tmp.tv64 = nsec; | |
294 | } else { | |
295 | unsigned long rem = do_div(nsec, NSEC_PER_SEC); | |
296 | ||
297 | tmp = ktime_set((long)nsec, rem); | |
298 | } | |
299 | ||
300 | return ktime_sub(kt, tmp); | |
301 | } | |
302 | ||
303 | EXPORT_SYMBOL_GPL(ktime_sub_ns); | |
c0a31329 TG |
304 | # endif /* !CONFIG_KTIME_SCALAR */ |
305 | ||
306 | /* | |
307 | * Divide a ktime value by a nanosecond value | |
308 | */ | |
4d672e7a | 309 | u64 ktime_divns(const ktime_t kt, s64 div) |
c0a31329 | 310 | { |
900cfa46 | 311 | u64 dclc; |
c0a31329 TG |
312 | int sft = 0; |
313 | ||
900cfa46 | 314 | dclc = ktime_to_ns(kt); |
c0a31329 TG |
315 | /* Make sure the divisor is less than 2^32: */ |
316 | while (div >> 32) { | |
317 | sft++; | |
318 | div >>= 1; | |
319 | } | |
320 | dclc >>= sft; | |
321 | do_div(dclc, (unsigned long) div); | |
322 | ||
4d672e7a | 323 | return dclc; |
c0a31329 | 324 | } |
c0a31329 TG |
325 | #endif /* BITS_PER_LONG >= 64 */ |
326 | ||
5a7780e7 TG |
327 | /* |
328 | * Add two ktime values and do a safety check for overflow: | |
329 | */ | |
330 | ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs) | |
331 | { | |
332 | ktime_t res = ktime_add(lhs, rhs); | |
333 | ||
334 | /* | |
335 | * We use KTIME_SEC_MAX here, the maximum timeout which we can | |
336 | * return to user space in a timespec: | |
337 | */ | |
338 | if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64) | |
339 | res = ktime_set(KTIME_SEC_MAX, 0); | |
340 | ||
341 | return res; | |
342 | } | |
343 | ||
8daa21e6 AB |
344 | EXPORT_SYMBOL_GPL(ktime_add_safe); |
345 | ||
237fc6e7 TG |
346 | #ifdef CONFIG_DEBUG_OBJECTS_TIMERS |
347 | ||
348 | static struct debug_obj_descr hrtimer_debug_descr; | |
349 | ||
99777288 SG |
350 | static void *hrtimer_debug_hint(void *addr) |
351 | { | |
352 | return ((struct hrtimer *) addr)->function; | |
353 | } | |
354 | ||
237fc6e7 TG |
355 | /* |
356 | * fixup_init is called when: | |
357 | * - an active object is initialized | |
358 | */ | |
359 | static int hrtimer_fixup_init(void *addr, enum debug_obj_state state) | |
360 | { | |
361 | struct hrtimer *timer = addr; | |
362 | ||
363 | switch (state) { | |
364 | case ODEBUG_STATE_ACTIVE: | |
365 | hrtimer_cancel(timer); | |
366 | debug_object_init(timer, &hrtimer_debug_descr); | |
367 | return 1; | |
368 | default: | |
369 | return 0; | |
370 | } | |
371 | } | |
372 | ||
373 | /* | |
374 | * fixup_activate is called when: | |
375 | * - an active object is activated | |
376 | * - an unknown object is activated (might be a statically initialized object) | |
377 | */ | |
378 | static int hrtimer_fixup_activate(void *addr, enum debug_obj_state state) | |
379 | { | |
380 | switch (state) { | |
381 | ||
382 | case ODEBUG_STATE_NOTAVAILABLE: | |
383 | WARN_ON_ONCE(1); | |
384 | return 0; | |
385 | ||
386 | case ODEBUG_STATE_ACTIVE: | |
387 | WARN_ON(1); | |
388 | ||
389 | default: | |
390 | return 0; | |
391 | } | |
392 | } | |
393 | ||
394 | /* | |
395 | * fixup_free is called when: | |
396 | * - an active object is freed | |
397 | */ | |
398 | static int hrtimer_fixup_free(void *addr, enum debug_obj_state state) | |
399 | { | |
400 | struct hrtimer *timer = addr; | |
401 | ||
402 | switch (state) { | |
403 | case ODEBUG_STATE_ACTIVE: | |
404 | hrtimer_cancel(timer); | |
405 | debug_object_free(timer, &hrtimer_debug_descr); | |
406 | return 1; | |
407 | default: | |
408 | return 0; | |
409 | } | |
410 | } | |
411 | ||
412 | static struct debug_obj_descr hrtimer_debug_descr = { | |
413 | .name = "hrtimer", | |
99777288 | 414 | .debug_hint = hrtimer_debug_hint, |
237fc6e7 TG |
415 | .fixup_init = hrtimer_fixup_init, |
416 | .fixup_activate = hrtimer_fixup_activate, | |
417 | .fixup_free = hrtimer_fixup_free, | |
418 | }; | |
419 | ||
420 | static inline void debug_hrtimer_init(struct hrtimer *timer) | |
421 | { | |
422 | debug_object_init(timer, &hrtimer_debug_descr); | |
423 | } | |
424 | ||
425 | static inline void debug_hrtimer_activate(struct hrtimer *timer) | |
426 | { | |
427 | debug_object_activate(timer, &hrtimer_debug_descr); | |
428 | } | |
429 | ||
430 | static inline void debug_hrtimer_deactivate(struct hrtimer *timer) | |
431 | { | |
432 | debug_object_deactivate(timer, &hrtimer_debug_descr); | |
433 | } | |
434 | ||
435 | static inline void debug_hrtimer_free(struct hrtimer *timer) | |
436 | { | |
437 | debug_object_free(timer, &hrtimer_debug_descr); | |
438 | } | |
439 | ||
440 | static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id, | |
441 | enum hrtimer_mode mode); | |
442 | ||
443 | void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id, | |
444 | enum hrtimer_mode mode) | |
445 | { | |
446 | debug_object_init_on_stack(timer, &hrtimer_debug_descr); | |
447 | __hrtimer_init(timer, clock_id, mode); | |
448 | } | |
2bc481cf | 449 | EXPORT_SYMBOL_GPL(hrtimer_init_on_stack); |
237fc6e7 TG |
450 | |
451 | void destroy_hrtimer_on_stack(struct hrtimer *timer) | |
452 | { | |
453 | debug_object_free(timer, &hrtimer_debug_descr); | |
454 | } | |
455 | ||
456 | #else | |
457 | static inline void debug_hrtimer_init(struct hrtimer *timer) { } | |
458 | static inline void debug_hrtimer_activate(struct hrtimer *timer) { } | |
459 | static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { } | |
460 | #endif | |
461 | ||
c6a2a177 XG |
462 | static inline void |
463 | debug_init(struct hrtimer *timer, clockid_t clockid, | |
464 | enum hrtimer_mode mode) | |
465 | { | |
466 | debug_hrtimer_init(timer); | |
467 | trace_hrtimer_init(timer, clockid, mode); | |
468 | } | |
469 | ||
470 | static inline void debug_activate(struct hrtimer *timer) | |
471 | { | |
472 | debug_hrtimer_activate(timer); | |
473 | trace_hrtimer_start(timer); | |
474 | } | |
475 | ||
476 | static inline void debug_deactivate(struct hrtimer *timer) | |
477 | { | |
478 | debug_hrtimer_deactivate(timer); | |
479 | trace_hrtimer_cancel(timer); | |
480 | } | |
481 | ||
54cdfdb4 TG |
482 | /* High resolution timer related functions */ |
483 | #ifdef CONFIG_HIGH_RES_TIMERS | |
484 | ||
485 | /* | |
486 | * High resolution timer enabled ? | |
487 | */ | |
488 | static int hrtimer_hres_enabled __read_mostly = 1; | |
489 | ||
490 | /* | |
491 | * Enable / Disable high resolution mode | |
492 | */ | |
493 | static int __init setup_hrtimer_hres(char *str) | |
494 | { | |
495 | if (!strcmp(str, "off")) | |
496 | hrtimer_hres_enabled = 0; | |
497 | else if (!strcmp(str, "on")) | |
498 | hrtimer_hres_enabled = 1; | |
499 | else | |
500 | return 0; | |
501 | return 1; | |
502 | } | |
503 | ||
504 | __setup("highres=", setup_hrtimer_hres); | |
505 | ||
506 | /* | |
507 | * hrtimer_high_res_enabled - query, if the highres mode is enabled | |
508 | */ | |
509 | static inline int hrtimer_is_hres_enabled(void) | |
510 | { | |
511 | return hrtimer_hres_enabled; | |
512 | } | |
513 | ||
514 | /* | |
515 | * Is the high resolution mode active ? | |
516 | */ | |
517 | static inline int hrtimer_hres_active(void) | |
518 | { | |
909ea964 | 519 | return __this_cpu_read(hrtimer_bases.hres_active); |
54cdfdb4 TG |
520 | } |
521 | ||
522 | /* | |
523 | * Reprogram the event source with checking both queues for the | |
524 | * next event | |
525 | * Called with interrupts disabled and base->lock held | |
526 | */ | |
7403f41f AC |
527 | static void |
528 | hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal) | |
54cdfdb4 TG |
529 | { |
530 | int i; | |
531 | struct hrtimer_clock_base *base = cpu_base->clock_base; | |
7403f41f | 532 | ktime_t expires, expires_next; |
54cdfdb4 | 533 | |
7403f41f | 534 | expires_next.tv64 = KTIME_MAX; |
54cdfdb4 TG |
535 | |
536 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) { | |
537 | struct hrtimer *timer; | |
998adc3d | 538 | struct timerqueue_node *next; |
54cdfdb4 | 539 | |
998adc3d JS |
540 | next = timerqueue_getnext(&base->active); |
541 | if (!next) | |
54cdfdb4 | 542 | continue; |
998adc3d JS |
543 | timer = container_of(next, struct hrtimer, node); |
544 | ||
cc584b21 | 545 | expires = ktime_sub(hrtimer_get_expires(timer), base->offset); |
b0a9b511 TG |
546 | /* |
547 | * clock_was_set() has changed base->offset so the | |
548 | * result might be negative. Fix it up to prevent a | |
549 | * false positive in clockevents_program_event() | |
550 | */ | |
551 | if (expires.tv64 < 0) | |
552 | expires.tv64 = 0; | |
7403f41f AC |
553 | if (expires.tv64 < expires_next.tv64) |
554 | expires_next = expires; | |
54cdfdb4 TG |
555 | } |
556 | ||
7403f41f AC |
557 | if (skip_equal && expires_next.tv64 == cpu_base->expires_next.tv64) |
558 | return; | |
559 | ||
560 | cpu_base->expires_next.tv64 = expires_next.tv64; | |
561 | ||
54cdfdb4 TG |
562 | if (cpu_base->expires_next.tv64 != KTIME_MAX) |
563 | tick_program_event(cpu_base->expires_next, 1); | |
564 | } | |
565 | ||
566 | /* | |
567 | * Shared reprogramming for clock_realtime and clock_monotonic | |
568 | * | |
569 | * When a timer is enqueued and expires earlier than the already enqueued | |
570 | * timers, we have to check, whether it expires earlier than the timer for | |
571 | * which the clock event device was armed. | |
572 | * | |
573 | * Called with interrupts disabled and base->cpu_base.lock held | |
574 | */ | |
575 | static int hrtimer_reprogram(struct hrtimer *timer, | |
576 | struct hrtimer_clock_base *base) | |
577 | { | |
41d2e494 | 578 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
cc584b21 | 579 | ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset); |
54cdfdb4 TG |
580 | int res; |
581 | ||
cc584b21 | 582 | WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0); |
63070a79 | 583 | |
54cdfdb4 TG |
584 | /* |
585 | * When the callback is running, we do not reprogram the clock event | |
586 | * device. The timer callback is either running on a different CPU or | |
3a4fa0a2 | 587 | * the callback is executed in the hrtimer_interrupt context. The |
54cdfdb4 TG |
588 | * reprogramming is handled either by the softirq, which called the |
589 | * callback or at the end of the hrtimer_interrupt. | |
590 | */ | |
591 | if (hrtimer_callback_running(timer)) | |
592 | return 0; | |
593 | ||
63070a79 TG |
594 | /* |
595 | * CLOCK_REALTIME timer might be requested with an absolute | |
596 | * expiry time which is less than base->offset. Nothing wrong | |
597 | * about that, just avoid to call into the tick code, which | |
598 | * has now objections against negative expiry values. | |
599 | */ | |
600 | if (expires.tv64 < 0) | |
601 | return -ETIME; | |
602 | ||
41d2e494 TG |
603 | if (expires.tv64 >= cpu_base->expires_next.tv64) |
604 | return 0; | |
605 | ||
606 | /* | |
607 | * If a hang was detected in the last timer interrupt then we | |
608 | * do not schedule a timer which is earlier than the expiry | |
609 | * which we enforced in the hang detection. We want the system | |
610 | * to make progress. | |
611 | */ | |
612 | if (cpu_base->hang_detected) | |
54cdfdb4 TG |
613 | return 0; |
614 | ||
615 | /* | |
616 | * Clockevents returns -ETIME, when the event was in the past. | |
617 | */ | |
618 | res = tick_program_event(expires, 0); | |
619 | if (!IS_ERR_VALUE(res)) | |
41d2e494 | 620 | cpu_base->expires_next = expires; |
54cdfdb4 TG |
621 | return res; |
622 | } | |
623 | ||
54cdfdb4 TG |
624 | /* |
625 | * Initialize the high resolution related parts of cpu_base | |
626 | */ | |
627 | static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) | |
628 | { | |
629 | base->expires_next.tv64 = KTIME_MAX; | |
630 | base->hres_active = 0; | |
54cdfdb4 TG |
631 | } |
632 | ||
54cdfdb4 TG |
633 | /* |
634 | * When High resolution timers are active, try to reprogram. Note, that in case | |
635 | * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry | |
636 | * check happens. The timer gets enqueued into the rbtree. The reprogramming | |
637 | * and expiry check is done in the hrtimer_interrupt or in the softirq. | |
638 | */ | |
639 | static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer, | |
7f1e2ca9 PZ |
640 | struct hrtimer_clock_base *base, |
641 | int wakeup) | |
54cdfdb4 TG |
642 | { |
643 | if (base->cpu_base->hres_active && hrtimer_reprogram(timer, base)) { | |
7f1e2ca9 | 644 | if (wakeup) { |
ecb49d1a | 645 | raw_spin_unlock(&base->cpu_base->lock); |
7f1e2ca9 | 646 | raise_softirq_irqoff(HRTIMER_SOFTIRQ); |
ecb49d1a | 647 | raw_spin_lock(&base->cpu_base->lock); |
7f1e2ca9 PZ |
648 | } else |
649 | __raise_softirq_irqoff(HRTIMER_SOFTIRQ); | |
650 | ||
ca109491 | 651 | return 1; |
54cdfdb4 | 652 | } |
7f1e2ca9 | 653 | |
54cdfdb4 TG |
654 | return 0; |
655 | } | |
656 | ||
9ec26907 TG |
657 | /* |
658 | * Retrigger next event is called after clock was set | |
659 | * | |
660 | * Called with interrupts disabled via on_each_cpu() | |
661 | */ | |
662 | static void retrigger_next_event(void *arg) | |
663 | { | |
664 | struct hrtimer_cpu_base *base = &__get_cpu_var(hrtimer_bases); | |
665 | struct timespec realtime_offset, xtim, wtm, sleep; | |
666 | ||
667 | if (!hrtimer_hres_active()) | |
668 | return; | |
669 | ||
670 | /* Optimized out for !HIGH_RES */ | |
671 | get_xtime_and_monotonic_and_sleep_offset(&xtim, &wtm, &sleep); | |
672 | set_normalized_timespec(&realtime_offset, -wtm.tv_sec, -wtm.tv_nsec); | |
673 | ||
674 | /* Adjust CLOCK_REALTIME offset */ | |
675 | raw_spin_lock(&base->lock); | |
676 | base->clock_base[HRTIMER_BASE_REALTIME].offset = | |
677 | timespec_to_ktime(realtime_offset); | |
678 | base->clock_base[HRTIMER_BASE_BOOTTIME].offset = | |
679 | timespec_to_ktime(sleep); | |
680 | ||
681 | hrtimer_force_reprogram(base, 0); | |
682 | raw_spin_unlock(&base->lock); | |
683 | } | |
b12a03ce | 684 | |
54cdfdb4 TG |
685 | /* |
686 | * Switch to high resolution mode | |
687 | */ | |
f8953856 | 688 | static int hrtimer_switch_to_hres(void) |
54cdfdb4 | 689 | { |
b12a03ce | 690 | int i, cpu = smp_processor_id(); |
820de5c3 | 691 | struct hrtimer_cpu_base *base = &per_cpu(hrtimer_bases, cpu); |
54cdfdb4 TG |
692 | unsigned long flags; |
693 | ||
694 | if (base->hres_active) | |
f8953856 | 695 | return 1; |
54cdfdb4 TG |
696 | |
697 | local_irq_save(flags); | |
698 | ||
699 | if (tick_init_highres()) { | |
700 | local_irq_restore(flags); | |
820de5c3 IM |
701 | printk(KERN_WARNING "Could not switch to high resolution " |
702 | "mode on CPU %d\n", cpu); | |
f8953856 | 703 | return 0; |
54cdfdb4 TG |
704 | } |
705 | base->hres_active = 1; | |
b12a03ce TG |
706 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) |
707 | base->clock_base[i].resolution = KTIME_HIGH_RES; | |
54cdfdb4 TG |
708 | |
709 | tick_setup_sched_timer(); | |
710 | ||
711 | /* "Retrigger" the interrupt to get things going */ | |
712 | retrigger_next_event(NULL); | |
713 | local_irq_restore(flags); | |
f8953856 | 714 | return 1; |
54cdfdb4 TG |
715 | } |
716 | ||
717 | #else | |
718 | ||
719 | static inline int hrtimer_hres_active(void) { return 0; } | |
720 | static inline int hrtimer_is_hres_enabled(void) { return 0; } | |
f8953856 | 721 | static inline int hrtimer_switch_to_hres(void) { return 0; } |
7403f41f AC |
722 | static inline void |
723 | hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { } | |
54cdfdb4 | 724 | static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer, |
7f1e2ca9 PZ |
725 | struct hrtimer_clock_base *base, |
726 | int wakeup) | |
54cdfdb4 TG |
727 | { |
728 | return 0; | |
729 | } | |
54cdfdb4 | 730 | static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { } |
9ec26907 | 731 | static inline void retrigger_next_event(void *arg) { } |
54cdfdb4 TG |
732 | |
733 | #endif /* CONFIG_HIGH_RES_TIMERS */ | |
734 | ||
b12a03ce TG |
735 | /* |
736 | * Clock realtime was set | |
737 | * | |
738 | * Change the offset of the realtime clock vs. the monotonic | |
739 | * clock. | |
740 | * | |
741 | * We might have to reprogram the high resolution timer interrupt. On | |
742 | * SMP we call the architecture specific code to retrigger _all_ high | |
743 | * resolution timer interrupts. On UP we just disable interrupts and | |
744 | * call the high resolution interrupt code. | |
745 | */ | |
746 | void clock_was_set(void) | |
747 | { | |
9ec26907 | 748 | #ifdef CONFIG_HIGHRES_TIMERS |
b12a03ce TG |
749 | /* Retrigger the CPU local events everywhere */ |
750 | on_each_cpu(retrigger_next_event, NULL, 1); | |
9ec26907 TG |
751 | #endif |
752 | timerfd_clock_was_set(); | |
b12a03ce TG |
753 | } |
754 | ||
755 | /* | |
756 | * During resume we might have to reprogram the high resolution timer | |
757 | * interrupt (on the local CPU): | |
758 | */ | |
759 | void hrtimers_resume(void) | |
760 | { | |
761 | WARN_ONCE(!irqs_disabled(), | |
762 | KERN_INFO "hrtimers_resume() called with IRQs enabled!"); | |
763 | ||
764 | retrigger_next_event(NULL); | |
9ec26907 | 765 | timerfd_clock_was_set(); |
b12a03ce TG |
766 | } |
767 | ||
5f201907 | 768 | static inline void timer_stats_hrtimer_set_start_info(struct hrtimer *timer) |
82f67cd9 | 769 | { |
5f201907 | 770 | #ifdef CONFIG_TIMER_STATS |
82f67cd9 IM |
771 | if (timer->start_site) |
772 | return; | |
5f201907 | 773 | timer->start_site = __builtin_return_address(0); |
82f67cd9 IM |
774 | memcpy(timer->start_comm, current->comm, TASK_COMM_LEN); |
775 | timer->start_pid = current->pid; | |
5f201907 HC |
776 | #endif |
777 | } | |
778 | ||
779 | static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer *timer) | |
780 | { | |
781 | #ifdef CONFIG_TIMER_STATS | |
782 | timer->start_site = NULL; | |
783 | #endif | |
82f67cd9 | 784 | } |
5f201907 HC |
785 | |
786 | static inline void timer_stats_account_hrtimer(struct hrtimer *timer) | |
787 | { | |
788 | #ifdef CONFIG_TIMER_STATS | |
789 | if (likely(!timer_stats_active)) | |
790 | return; | |
791 | timer_stats_update_stats(timer, timer->start_pid, timer->start_site, | |
792 | timer->function, timer->start_comm, 0); | |
82f67cd9 | 793 | #endif |
5f201907 | 794 | } |
82f67cd9 | 795 | |
c0a31329 | 796 | /* |
6506f2aa | 797 | * Counterpart to lock_hrtimer_base above: |
c0a31329 TG |
798 | */ |
799 | static inline | |
800 | void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) | |
801 | { | |
ecb49d1a | 802 | raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags); |
c0a31329 TG |
803 | } |
804 | ||
805 | /** | |
806 | * hrtimer_forward - forward the timer expiry | |
c0a31329 | 807 | * @timer: hrtimer to forward |
44f21475 | 808 | * @now: forward past this time |
c0a31329 TG |
809 | * @interval: the interval to forward |
810 | * | |
811 | * Forward the timer expiry so it will expire in the future. | |
8dca6f33 | 812 | * Returns the number of overruns. |
c0a31329 | 813 | */ |
4d672e7a | 814 | u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval) |
c0a31329 | 815 | { |
4d672e7a | 816 | u64 orun = 1; |
44f21475 | 817 | ktime_t delta; |
c0a31329 | 818 | |
cc584b21 | 819 | delta = ktime_sub(now, hrtimer_get_expires(timer)); |
c0a31329 TG |
820 | |
821 | if (delta.tv64 < 0) | |
822 | return 0; | |
823 | ||
c9db4fa1 TG |
824 | if (interval.tv64 < timer->base->resolution.tv64) |
825 | interval.tv64 = timer->base->resolution.tv64; | |
826 | ||
c0a31329 | 827 | if (unlikely(delta.tv64 >= interval.tv64)) { |
df869b63 | 828 | s64 incr = ktime_to_ns(interval); |
c0a31329 TG |
829 | |
830 | orun = ktime_divns(delta, incr); | |
cc584b21 AV |
831 | hrtimer_add_expires_ns(timer, incr * orun); |
832 | if (hrtimer_get_expires_tv64(timer) > now.tv64) | |
c0a31329 TG |
833 | return orun; |
834 | /* | |
835 | * This (and the ktime_add() below) is the | |
836 | * correction for exact: | |
837 | */ | |
838 | orun++; | |
839 | } | |
cc584b21 | 840 | hrtimer_add_expires(timer, interval); |
c0a31329 TG |
841 | |
842 | return orun; | |
843 | } | |
6bdb6b62 | 844 | EXPORT_SYMBOL_GPL(hrtimer_forward); |
c0a31329 TG |
845 | |
846 | /* | |
847 | * enqueue_hrtimer - internal function to (re)start a timer | |
848 | * | |
849 | * The timer is inserted in expiry order. Insertion into the | |
850 | * red black tree is O(log(n)). Must hold the base lock. | |
a6037b61 PZ |
851 | * |
852 | * Returns 1 when the new timer is the leftmost timer in the tree. | |
c0a31329 | 853 | */ |
a6037b61 PZ |
854 | static int enqueue_hrtimer(struct hrtimer *timer, |
855 | struct hrtimer_clock_base *base) | |
c0a31329 | 856 | { |
c6a2a177 | 857 | debug_activate(timer); |
237fc6e7 | 858 | |
998adc3d | 859 | timerqueue_add(&base->active, &timer->node); |
54cdfdb4 | 860 | |
303e967f TG |
861 | /* |
862 | * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the | |
863 | * state of a possibly running callback. | |
864 | */ | |
865 | timer->state |= HRTIMER_STATE_ENQUEUED; | |
a6037b61 | 866 | |
998adc3d | 867 | return (&timer->node == base->active.next); |
288867ec | 868 | } |
c0a31329 TG |
869 | |
870 | /* | |
871 | * __remove_hrtimer - internal function to remove a timer | |
872 | * | |
873 | * Caller must hold the base lock. | |
54cdfdb4 TG |
874 | * |
875 | * High resolution timer mode reprograms the clock event device when the | |
876 | * timer is the one which expires next. The caller can disable this by setting | |
877 | * reprogram to zero. This is useful, when the context does a reprogramming | |
878 | * anyway (e.g. timer interrupt) | |
c0a31329 | 879 | */ |
3c8aa39d | 880 | static void __remove_hrtimer(struct hrtimer *timer, |
303e967f | 881 | struct hrtimer_clock_base *base, |
54cdfdb4 | 882 | unsigned long newstate, int reprogram) |
c0a31329 | 883 | { |
7403f41f AC |
884 | if (!(timer->state & HRTIMER_STATE_ENQUEUED)) |
885 | goto out; | |
886 | ||
998adc3d | 887 | if (&timer->node == timerqueue_getnext(&base->active)) { |
7403f41f AC |
888 | #ifdef CONFIG_HIGH_RES_TIMERS |
889 | /* Reprogram the clock event device. if enabled */ | |
890 | if (reprogram && hrtimer_hres_active()) { | |
891 | ktime_t expires; | |
892 | ||
893 | expires = ktime_sub(hrtimer_get_expires(timer), | |
894 | base->offset); | |
895 | if (base->cpu_base->expires_next.tv64 == expires.tv64) | |
896 | hrtimer_force_reprogram(base->cpu_base, 1); | |
54cdfdb4 | 897 | } |
7403f41f | 898 | #endif |
54cdfdb4 | 899 | } |
998adc3d | 900 | timerqueue_del(&base->active, &timer->node); |
7403f41f | 901 | out: |
303e967f | 902 | timer->state = newstate; |
c0a31329 TG |
903 | } |
904 | ||
905 | /* | |
906 | * remove hrtimer, called with base lock held | |
907 | */ | |
908 | static inline int | |
3c8aa39d | 909 | remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base) |
c0a31329 | 910 | { |
303e967f | 911 | if (hrtimer_is_queued(timer)) { |
f13d4f97 | 912 | unsigned long state; |
54cdfdb4 TG |
913 | int reprogram; |
914 | ||
915 | /* | |
916 | * Remove the timer and force reprogramming when high | |
917 | * resolution mode is active and the timer is on the current | |
918 | * CPU. If we remove a timer on another CPU, reprogramming is | |
919 | * skipped. The interrupt event on this CPU is fired and | |
920 | * reprogramming happens in the interrupt handler. This is a | |
921 | * rare case and less expensive than a smp call. | |
922 | */ | |
c6a2a177 | 923 | debug_deactivate(timer); |
82f67cd9 | 924 | timer_stats_hrtimer_clear_start_info(timer); |
54cdfdb4 | 925 | reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases); |
f13d4f97 SQ |
926 | /* |
927 | * We must preserve the CALLBACK state flag here, | |
928 | * otherwise we could move the timer base in | |
929 | * switch_hrtimer_base. | |
930 | */ | |
931 | state = timer->state & HRTIMER_STATE_CALLBACK; | |
932 | __remove_hrtimer(timer, base, state, reprogram); | |
c0a31329 TG |
933 | return 1; |
934 | } | |
935 | return 0; | |
936 | } | |
937 | ||
7f1e2ca9 PZ |
938 | int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, |
939 | unsigned long delta_ns, const enum hrtimer_mode mode, | |
940 | int wakeup) | |
c0a31329 | 941 | { |
3c8aa39d | 942 | struct hrtimer_clock_base *base, *new_base; |
c0a31329 | 943 | unsigned long flags; |
a6037b61 | 944 | int ret, leftmost; |
c0a31329 TG |
945 | |
946 | base = lock_hrtimer_base(timer, &flags); | |
947 | ||
948 | /* Remove an active timer from the queue: */ | |
949 | ret = remove_hrtimer(timer, base); | |
950 | ||
951 | /* Switch the timer base, if necessary: */ | |
597d0275 | 952 | new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED); |
c0a31329 | 953 | |
597d0275 | 954 | if (mode & HRTIMER_MODE_REL) { |
5a7780e7 | 955 | tim = ktime_add_safe(tim, new_base->get_time()); |
06027bdd IM |
956 | /* |
957 | * CONFIG_TIME_LOW_RES is a temporary way for architectures | |
958 | * to signal that they simply return xtime in | |
959 | * do_gettimeoffset(). In this case we want to round up by | |
960 | * resolution when starting a relative timer, to avoid short | |
961 | * timeouts. This will go away with the GTOD framework. | |
962 | */ | |
963 | #ifdef CONFIG_TIME_LOW_RES | |
5a7780e7 | 964 | tim = ktime_add_safe(tim, base->resolution); |
06027bdd IM |
965 | #endif |
966 | } | |
237fc6e7 | 967 | |
da8f2e17 | 968 | hrtimer_set_expires_range_ns(timer, tim, delta_ns); |
c0a31329 | 969 | |
82f67cd9 IM |
970 | timer_stats_hrtimer_set_start_info(timer); |
971 | ||
a6037b61 PZ |
972 | leftmost = enqueue_hrtimer(timer, new_base); |
973 | ||
935c631d IM |
974 | /* |
975 | * Only allow reprogramming if the new base is on this CPU. | |
976 | * (it might still be on another CPU if the timer was pending) | |
a6037b61 PZ |
977 | * |
978 | * XXX send_remote_softirq() ? | |
935c631d | 979 | */ |
a6037b61 | 980 | if (leftmost && new_base->cpu_base == &__get_cpu_var(hrtimer_bases)) |
7f1e2ca9 | 981 | hrtimer_enqueue_reprogram(timer, new_base, wakeup); |
c0a31329 TG |
982 | |
983 | unlock_hrtimer_base(timer, &flags); | |
984 | ||
985 | return ret; | |
986 | } | |
7f1e2ca9 PZ |
987 | |
988 | /** | |
989 | * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU | |
990 | * @timer: the timer to be added | |
991 | * @tim: expiry time | |
992 | * @delta_ns: "slack" range for the timer | |
993 | * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL) | |
994 | * | |
995 | * Returns: | |
996 | * 0 on success | |
997 | * 1 when the timer was active | |
998 | */ | |
999 | int hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, | |
1000 | unsigned long delta_ns, const enum hrtimer_mode mode) | |
1001 | { | |
1002 | return __hrtimer_start_range_ns(timer, tim, delta_ns, mode, 1); | |
1003 | } | |
da8f2e17 AV |
1004 | EXPORT_SYMBOL_GPL(hrtimer_start_range_ns); |
1005 | ||
1006 | /** | |
e1dd7bc5 | 1007 | * hrtimer_start - (re)start an hrtimer on the current CPU |
da8f2e17 AV |
1008 | * @timer: the timer to be added |
1009 | * @tim: expiry time | |
1010 | * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL) | |
1011 | * | |
1012 | * Returns: | |
1013 | * 0 on success | |
1014 | * 1 when the timer was active | |
1015 | */ | |
1016 | int | |
1017 | hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode) | |
1018 | { | |
7f1e2ca9 | 1019 | return __hrtimer_start_range_ns(timer, tim, 0, mode, 1); |
da8f2e17 | 1020 | } |
8d16b764 | 1021 | EXPORT_SYMBOL_GPL(hrtimer_start); |
c0a31329 | 1022 | |
da8f2e17 | 1023 | |
c0a31329 TG |
1024 | /** |
1025 | * hrtimer_try_to_cancel - try to deactivate a timer | |
c0a31329 TG |
1026 | * @timer: hrtimer to stop |
1027 | * | |
1028 | * Returns: | |
1029 | * 0 when the timer was not active | |
1030 | * 1 when the timer was active | |
1031 | * -1 when the timer is currently excuting the callback function and | |
fa9799e3 | 1032 | * cannot be stopped |
c0a31329 TG |
1033 | */ |
1034 | int hrtimer_try_to_cancel(struct hrtimer *timer) | |
1035 | { | |
3c8aa39d | 1036 | struct hrtimer_clock_base *base; |
c0a31329 TG |
1037 | unsigned long flags; |
1038 | int ret = -1; | |
1039 | ||
1040 | base = lock_hrtimer_base(timer, &flags); | |
1041 | ||
303e967f | 1042 | if (!hrtimer_callback_running(timer)) |
c0a31329 TG |
1043 | ret = remove_hrtimer(timer, base); |
1044 | ||
1045 | unlock_hrtimer_base(timer, &flags); | |
1046 | ||
1047 | return ret; | |
1048 | ||
1049 | } | |
8d16b764 | 1050 | EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel); |
c0a31329 TG |
1051 | |
1052 | /** | |
1053 | * hrtimer_cancel - cancel a timer and wait for the handler to finish. | |
c0a31329 TG |
1054 | * @timer: the timer to be cancelled |
1055 | * | |
1056 | * Returns: | |
1057 | * 0 when the timer was not active | |
1058 | * 1 when the timer was active | |
1059 | */ | |
1060 | int hrtimer_cancel(struct hrtimer *timer) | |
1061 | { | |
1062 | for (;;) { | |
1063 | int ret = hrtimer_try_to_cancel(timer); | |
1064 | ||
1065 | if (ret >= 0) | |
1066 | return ret; | |
5ef37b19 | 1067 | cpu_relax(); |
c0a31329 TG |
1068 | } |
1069 | } | |
8d16b764 | 1070 | EXPORT_SYMBOL_GPL(hrtimer_cancel); |
c0a31329 TG |
1071 | |
1072 | /** | |
1073 | * hrtimer_get_remaining - get remaining time for the timer | |
c0a31329 TG |
1074 | * @timer: the timer to read |
1075 | */ | |
1076 | ktime_t hrtimer_get_remaining(const struct hrtimer *timer) | |
1077 | { | |
c0a31329 TG |
1078 | unsigned long flags; |
1079 | ktime_t rem; | |
1080 | ||
b3bd3de6 | 1081 | lock_hrtimer_base(timer, &flags); |
cc584b21 | 1082 | rem = hrtimer_expires_remaining(timer); |
c0a31329 TG |
1083 | unlock_hrtimer_base(timer, &flags); |
1084 | ||
1085 | return rem; | |
1086 | } | |
8d16b764 | 1087 | EXPORT_SYMBOL_GPL(hrtimer_get_remaining); |
c0a31329 | 1088 | |
ee9c5785 | 1089 | #ifdef CONFIG_NO_HZ |
69239749 TL |
1090 | /** |
1091 | * hrtimer_get_next_event - get the time until next expiry event | |
1092 | * | |
1093 | * Returns the delta to the next expiry event or KTIME_MAX if no timer | |
1094 | * is pending. | |
1095 | */ | |
1096 | ktime_t hrtimer_get_next_event(void) | |
1097 | { | |
3c8aa39d TG |
1098 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
1099 | struct hrtimer_clock_base *base = cpu_base->clock_base; | |
69239749 TL |
1100 | ktime_t delta, mindelta = { .tv64 = KTIME_MAX }; |
1101 | unsigned long flags; | |
1102 | int i; | |
1103 | ||
ecb49d1a | 1104 | raw_spin_lock_irqsave(&cpu_base->lock, flags); |
3c8aa39d | 1105 | |
54cdfdb4 TG |
1106 | if (!hrtimer_hres_active()) { |
1107 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) { | |
1108 | struct hrtimer *timer; | |
998adc3d | 1109 | struct timerqueue_node *next; |
69239749 | 1110 | |
998adc3d JS |
1111 | next = timerqueue_getnext(&base->active); |
1112 | if (!next) | |
54cdfdb4 | 1113 | continue; |
3c8aa39d | 1114 | |
998adc3d | 1115 | timer = container_of(next, struct hrtimer, node); |
cc584b21 | 1116 | delta.tv64 = hrtimer_get_expires_tv64(timer); |
54cdfdb4 TG |
1117 | delta = ktime_sub(delta, base->get_time()); |
1118 | if (delta.tv64 < mindelta.tv64) | |
1119 | mindelta.tv64 = delta.tv64; | |
1120 | } | |
69239749 | 1121 | } |
3c8aa39d | 1122 | |
ecb49d1a | 1123 | raw_spin_unlock_irqrestore(&cpu_base->lock, flags); |
3c8aa39d | 1124 | |
69239749 TL |
1125 | if (mindelta.tv64 < 0) |
1126 | mindelta.tv64 = 0; | |
1127 | return mindelta; | |
1128 | } | |
1129 | #endif | |
1130 | ||
237fc6e7 TG |
1131 | static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id, |
1132 | enum hrtimer_mode mode) | |
c0a31329 | 1133 | { |
3c8aa39d | 1134 | struct hrtimer_cpu_base *cpu_base; |
e06383db | 1135 | int base; |
c0a31329 | 1136 | |
7978672c GA |
1137 | memset(timer, 0, sizeof(struct hrtimer)); |
1138 | ||
3c8aa39d | 1139 | cpu_base = &__raw_get_cpu_var(hrtimer_bases); |
c0a31329 | 1140 | |
c9cb2e3d | 1141 | if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS) |
7978672c GA |
1142 | clock_id = CLOCK_MONOTONIC; |
1143 | ||
e06383db JS |
1144 | base = hrtimer_clockid_to_base(clock_id); |
1145 | timer->base = &cpu_base->clock_base[base]; | |
998adc3d | 1146 | timerqueue_init(&timer->node); |
82f67cd9 IM |
1147 | |
1148 | #ifdef CONFIG_TIMER_STATS | |
1149 | timer->start_site = NULL; | |
1150 | timer->start_pid = -1; | |
1151 | memset(timer->start_comm, 0, TASK_COMM_LEN); | |
1152 | #endif | |
c0a31329 | 1153 | } |
237fc6e7 TG |
1154 | |
1155 | /** | |
1156 | * hrtimer_init - initialize a timer to the given clock | |
1157 | * @timer: the timer to be initialized | |
1158 | * @clock_id: the clock to be used | |
1159 | * @mode: timer mode abs/rel | |
1160 | */ | |
1161 | void hrtimer_init(struct hrtimer *timer, clockid_t clock_id, | |
1162 | enum hrtimer_mode mode) | |
1163 | { | |
c6a2a177 | 1164 | debug_init(timer, clock_id, mode); |
237fc6e7 TG |
1165 | __hrtimer_init(timer, clock_id, mode); |
1166 | } | |
8d16b764 | 1167 | EXPORT_SYMBOL_GPL(hrtimer_init); |
c0a31329 TG |
1168 | |
1169 | /** | |
1170 | * hrtimer_get_res - get the timer resolution for a clock | |
c0a31329 TG |
1171 | * @which_clock: which clock to query |
1172 | * @tp: pointer to timespec variable to store the resolution | |
1173 | * | |
72fd4a35 RD |
1174 | * Store the resolution of the clock selected by @which_clock in the |
1175 | * variable pointed to by @tp. | |
c0a31329 TG |
1176 | */ |
1177 | int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp) | |
1178 | { | |
3c8aa39d | 1179 | struct hrtimer_cpu_base *cpu_base; |
e06383db | 1180 | int base = hrtimer_clockid_to_base(which_clock); |
c0a31329 | 1181 | |
3c8aa39d | 1182 | cpu_base = &__raw_get_cpu_var(hrtimer_bases); |
e06383db | 1183 | *tp = ktime_to_timespec(cpu_base->clock_base[base].resolution); |
c0a31329 TG |
1184 | |
1185 | return 0; | |
1186 | } | |
8d16b764 | 1187 | EXPORT_SYMBOL_GPL(hrtimer_get_res); |
c0a31329 | 1188 | |
c6a2a177 | 1189 | static void __run_hrtimer(struct hrtimer *timer, ktime_t *now) |
d3d74453 PZ |
1190 | { |
1191 | struct hrtimer_clock_base *base = timer->base; | |
1192 | struct hrtimer_cpu_base *cpu_base = base->cpu_base; | |
1193 | enum hrtimer_restart (*fn)(struct hrtimer *); | |
1194 | int restart; | |
1195 | ||
ca109491 PZ |
1196 | WARN_ON(!irqs_disabled()); |
1197 | ||
c6a2a177 | 1198 | debug_deactivate(timer); |
d3d74453 PZ |
1199 | __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0); |
1200 | timer_stats_account_hrtimer(timer); | |
d3d74453 | 1201 | fn = timer->function; |
ca109491 PZ |
1202 | |
1203 | /* | |
1204 | * Because we run timers from hardirq context, there is no chance | |
1205 | * they get migrated to another cpu, therefore its safe to unlock | |
1206 | * the timer base. | |
1207 | */ | |
ecb49d1a | 1208 | raw_spin_unlock(&cpu_base->lock); |
c6a2a177 | 1209 | trace_hrtimer_expire_entry(timer, now); |
ca109491 | 1210 | restart = fn(timer); |
c6a2a177 | 1211 | trace_hrtimer_expire_exit(timer); |
ecb49d1a | 1212 | raw_spin_lock(&cpu_base->lock); |
d3d74453 PZ |
1213 | |
1214 | /* | |
e3f1d883 TG |
1215 | * Note: We clear the CALLBACK bit after enqueue_hrtimer and |
1216 | * we do not reprogramm the event hardware. Happens either in | |
1217 | * hrtimer_start_range_ns() or in hrtimer_interrupt() | |
d3d74453 PZ |
1218 | */ |
1219 | if (restart != HRTIMER_NORESTART) { | |
1220 | BUG_ON(timer->state != HRTIMER_STATE_CALLBACK); | |
a6037b61 | 1221 | enqueue_hrtimer(timer, base); |
d3d74453 | 1222 | } |
f13d4f97 SQ |
1223 | |
1224 | WARN_ON_ONCE(!(timer->state & HRTIMER_STATE_CALLBACK)); | |
1225 | ||
d3d74453 PZ |
1226 | timer->state &= ~HRTIMER_STATE_CALLBACK; |
1227 | } | |
1228 | ||
54cdfdb4 TG |
1229 | #ifdef CONFIG_HIGH_RES_TIMERS |
1230 | ||
1231 | /* | |
1232 | * High resolution timer interrupt | |
1233 | * Called with interrupts disabled | |
1234 | */ | |
1235 | void hrtimer_interrupt(struct clock_event_device *dev) | |
1236 | { | |
1237 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); | |
1238 | struct hrtimer_clock_base *base; | |
41d2e494 TG |
1239 | ktime_t expires_next, now, entry_time, delta; |
1240 | int i, retries = 0; | |
54cdfdb4 TG |
1241 | |
1242 | BUG_ON(!cpu_base->hres_active); | |
1243 | cpu_base->nr_events++; | |
1244 | dev->next_event.tv64 = KTIME_MAX; | |
1245 | ||
41d2e494 TG |
1246 | entry_time = now = ktime_get(); |
1247 | retry: | |
54cdfdb4 TG |
1248 | expires_next.tv64 = KTIME_MAX; |
1249 | ||
ecb49d1a | 1250 | raw_spin_lock(&cpu_base->lock); |
6ff7041d TG |
1251 | /* |
1252 | * We set expires_next to KTIME_MAX here with cpu_base->lock | |
1253 | * held to prevent that a timer is enqueued in our queue via | |
1254 | * the migration code. This does not affect enqueueing of | |
1255 | * timers which run their callback and need to be requeued on | |
1256 | * this CPU. | |
1257 | */ | |
1258 | cpu_base->expires_next.tv64 = KTIME_MAX; | |
1259 | ||
54cdfdb4 TG |
1260 | base = cpu_base->clock_base; |
1261 | ||
1262 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { | |
1263 | ktime_t basenow; | |
998adc3d | 1264 | struct timerqueue_node *node; |
54cdfdb4 | 1265 | |
54cdfdb4 TG |
1266 | basenow = ktime_add(now, base->offset); |
1267 | ||
998adc3d | 1268 | while ((node = timerqueue_getnext(&base->active))) { |
54cdfdb4 TG |
1269 | struct hrtimer *timer; |
1270 | ||
998adc3d | 1271 | timer = container_of(node, struct hrtimer, node); |
54cdfdb4 | 1272 | |
654c8e0b AV |
1273 | /* |
1274 | * The immediate goal for using the softexpires is | |
1275 | * minimizing wakeups, not running timers at the | |
1276 | * earliest interrupt after their soft expiration. | |
1277 | * This allows us to avoid using a Priority Search | |
1278 | * Tree, which can answer a stabbing querry for | |
1279 | * overlapping intervals and instead use the simple | |
1280 | * BST we already have. | |
1281 | * We don't add extra wakeups by delaying timers that | |
1282 | * are right-of a not yet expired timer, because that | |
1283 | * timer will have to trigger a wakeup anyway. | |
1284 | */ | |
1285 | ||
1286 | if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer)) { | |
54cdfdb4 TG |
1287 | ktime_t expires; |
1288 | ||
cc584b21 | 1289 | expires = ktime_sub(hrtimer_get_expires(timer), |
54cdfdb4 TG |
1290 | base->offset); |
1291 | if (expires.tv64 < expires_next.tv64) | |
1292 | expires_next = expires; | |
1293 | break; | |
1294 | } | |
1295 | ||
c6a2a177 | 1296 | __run_hrtimer(timer, &basenow); |
54cdfdb4 | 1297 | } |
54cdfdb4 TG |
1298 | base++; |
1299 | } | |
1300 | ||
6ff7041d TG |
1301 | /* |
1302 | * Store the new expiry value so the migration code can verify | |
1303 | * against it. | |
1304 | */ | |
54cdfdb4 | 1305 | cpu_base->expires_next = expires_next; |
ecb49d1a | 1306 | raw_spin_unlock(&cpu_base->lock); |
54cdfdb4 TG |
1307 | |
1308 | /* Reprogramming necessary ? */ | |
41d2e494 TG |
1309 | if (expires_next.tv64 == KTIME_MAX || |
1310 | !tick_program_event(expires_next, 0)) { | |
1311 | cpu_base->hang_detected = 0; | |
1312 | return; | |
54cdfdb4 | 1313 | } |
41d2e494 TG |
1314 | |
1315 | /* | |
1316 | * The next timer was already expired due to: | |
1317 | * - tracing | |
1318 | * - long lasting callbacks | |
1319 | * - being scheduled away when running in a VM | |
1320 | * | |
1321 | * We need to prevent that we loop forever in the hrtimer | |
1322 | * interrupt routine. We give it 3 attempts to avoid | |
1323 | * overreacting on some spurious event. | |
1324 | */ | |
1325 | now = ktime_get(); | |
1326 | cpu_base->nr_retries++; | |
1327 | if (++retries < 3) | |
1328 | goto retry; | |
1329 | /* | |
1330 | * Give the system a chance to do something else than looping | |
1331 | * here. We stored the entry time, so we know exactly how long | |
1332 | * we spent here. We schedule the next event this amount of | |
1333 | * time away. | |
1334 | */ | |
1335 | cpu_base->nr_hangs++; | |
1336 | cpu_base->hang_detected = 1; | |
1337 | delta = ktime_sub(now, entry_time); | |
1338 | if (delta.tv64 > cpu_base->max_hang_time.tv64) | |
1339 | cpu_base->max_hang_time = delta; | |
1340 | /* | |
1341 | * Limit it to a sensible value as we enforce a longer | |
1342 | * delay. Give the CPU at least 100ms to catch up. | |
1343 | */ | |
1344 | if (delta.tv64 > 100 * NSEC_PER_MSEC) | |
1345 | expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC); | |
1346 | else | |
1347 | expires_next = ktime_add(now, delta); | |
1348 | tick_program_event(expires_next, 1); | |
1349 | printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n", | |
1350 | ktime_to_ns(delta)); | |
54cdfdb4 TG |
1351 | } |
1352 | ||
8bdec955 TG |
1353 | /* |
1354 | * local version of hrtimer_peek_ahead_timers() called with interrupts | |
1355 | * disabled. | |
1356 | */ | |
1357 | static void __hrtimer_peek_ahead_timers(void) | |
1358 | { | |
1359 | struct tick_device *td; | |
1360 | ||
1361 | if (!hrtimer_hres_active()) | |
1362 | return; | |
1363 | ||
1364 | td = &__get_cpu_var(tick_cpu_device); | |
1365 | if (td && td->evtdev) | |
1366 | hrtimer_interrupt(td->evtdev); | |
1367 | } | |
1368 | ||
2e94d1f7 AV |
1369 | /** |
1370 | * hrtimer_peek_ahead_timers -- run soft-expired timers now | |
1371 | * | |
1372 | * hrtimer_peek_ahead_timers will peek at the timer queue of | |
1373 | * the current cpu and check if there are any timers for which | |
1374 | * the soft expires time has passed. If any such timers exist, | |
1375 | * they are run immediately and then removed from the timer queue. | |
1376 | * | |
1377 | */ | |
1378 | void hrtimer_peek_ahead_timers(void) | |
1379 | { | |
643bdf68 | 1380 | unsigned long flags; |
dc4304f7 | 1381 | |
2e94d1f7 | 1382 | local_irq_save(flags); |
8bdec955 | 1383 | __hrtimer_peek_ahead_timers(); |
2e94d1f7 AV |
1384 | local_irq_restore(flags); |
1385 | } | |
1386 | ||
a6037b61 PZ |
1387 | static void run_hrtimer_softirq(struct softirq_action *h) |
1388 | { | |
1389 | hrtimer_peek_ahead_timers(); | |
1390 | } | |
1391 | ||
82c5b7b5 IM |
1392 | #else /* CONFIG_HIGH_RES_TIMERS */ |
1393 | ||
1394 | static inline void __hrtimer_peek_ahead_timers(void) { } | |
1395 | ||
1396 | #endif /* !CONFIG_HIGH_RES_TIMERS */ | |
82f67cd9 | 1397 | |
d3d74453 PZ |
1398 | /* |
1399 | * Called from timer softirq every jiffy, expire hrtimers: | |
1400 | * | |
1401 | * For HRT its the fall back code to run the softirq in the timer | |
1402 | * softirq context in case the hrtimer initialization failed or has | |
1403 | * not been done yet. | |
1404 | */ | |
1405 | void hrtimer_run_pending(void) | |
1406 | { | |
d3d74453 PZ |
1407 | if (hrtimer_hres_active()) |
1408 | return; | |
54cdfdb4 | 1409 | |
d3d74453 PZ |
1410 | /* |
1411 | * This _is_ ugly: We have to check in the softirq context, | |
1412 | * whether we can switch to highres and / or nohz mode. The | |
1413 | * clocksource switch happens in the timer interrupt with | |
1414 | * xtime_lock held. Notification from there only sets the | |
1415 | * check bit in the tick_oneshot code, otherwise we might | |
1416 | * deadlock vs. xtime_lock. | |
1417 | */ | |
1418 | if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) | |
1419 | hrtimer_switch_to_hres(); | |
54cdfdb4 TG |
1420 | } |
1421 | ||
c0a31329 | 1422 | /* |
d3d74453 | 1423 | * Called from hardirq context every jiffy |
c0a31329 | 1424 | */ |
833883d9 | 1425 | void hrtimer_run_queues(void) |
c0a31329 | 1426 | { |
998adc3d | 1427 | struct timerqueue_node *node; |
833883d9 DS |
1428 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
1429 | struct hrtimer_clock_base *base; | |
1430 | int index, gettime = 1; | |
c0a31329 | 1431 | |
833883d9 | 1432 | if (hrtimer_hres_active()) |
3055adda DS |
1433 | return; |
1434 | ||
833883d9 DS |
1435 | for (index = 0; index < HRTIMER_MAX_CLOCK_BASES; index++) { |
1436 | base = &cpu_base->clock_base[index]; | |
b007c389 | 1437 | if (!timerqueue_getnext(&base->active)) |
d3d74453 | 1438 | continue; |
833883d9 | 1439 | |
d7cfb60c | 1440 | if (gettime) { |
833883d9 DS |
1441 | hrtimer_get_softirq_time(cpu_base); |
1442 | gettime = 0; | |
b75f7a51 | 1443 | } |
d3d74453 | 1444 | |
ecb49d1a | 1445 | raw_spin_lock(&cpu_base->lock); |
c0a31329 | 1446 | |
b007c389 | 1447 | while ((node = timerqueue_getnext(&base->active))) { |
833883d9 | 1448 | struct hrtimer *timer; |
54cdfdb4 | 1449 | |
998adc3d | 1450 | timer = container_of(node, struct hrtimer, node); |
cc584b21 AV |
1451 | if (base->softirq_time.tv64 <= |
1452 | hrtimer_get_expires_tv64(timer)) | |
833883d9 DS |
1453 | break; |
1454 | ||
c6a2a177 | 1455 | __run_hrtimer(timer, &base->softirq_time); |
833883d9 | 1456 | } |
ecb49d1a | 1457 | raw_spin_unlock(&cpu_base->lock); |
833883d9 | 1458 | } |
c0a31329 TG |
1459 | } |
1460 | ||
10c94ec1 TG |
1461 | /* |
1462 | * Sleep related functions: | |
1463 | */ | |
c9cb2e3d | 1464 | static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer) |
00362e33 TG |
1465 | { |
1466 | struct hrtimer_sleeper *t = | |
1467 | container_of(timer, struct hrtimer_sleeper, timer); | |
1468 | struct task_struct *task = t->task; | |
1469 | ||
1470 | t->task = NULL; | |
1471 | if (task) | |
1472 | wake_up_process(task); | |
1473 | ||
1474 | return HRTIMER_NORESTART; | |
1475 | } | |
1476 | ||
36c8b586 | 1477 | void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task) |
00362e33 TG |
1478 | { |
1479 | sl->timer.function = hrtimer_wakeup; | |
1480 | sl->task = task; | |
1481 | } | |
2bc481cf | 1482 | EXPORT_SYMBOL_GPL(hrtimer_init_sleeper); |
00362e33 | 1483 | |
669d7868 | 1484 | static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode) |
432569bb | 1485 | { |
669d7868 | 1486 | hrtimer_init_sleeper(t, current); |
10c94ec1 | 1487 | |
432569bb RZ |
1488 | do { |
1489 | set_current_state(TASK_INTERRUPTIBLE); | |
cc584b21 | 1490 | hrtimer_start_expires(&t->timer, mode); |
37bb6cb4 PZ |
1491 | if (!hrtimer_active(&t->timer)) |
1492 | t->task = NULL; | |
432569bb | 1493 | |
54cdfdb4 TG |
1494 | if (likely(t->task)) |
1495 | schedule(); | |
432569bb | 1496 | |
669d7868 | 1497 | hrtimer_cancel(&t->timer); |
c9cb2e3d | 1498 | mode = HRTIMER_MODE_ABS; |
669d7868 TG |
1499 | |
1500 | } while (t->task && !signal_pending(current)); | |
432569bb | 1501 | |
3588a085 PZ |
1502 | __set_current_state(TASK_RUNNING); |
1503 | ||
669d7868 | 1504 | return t->task == NULL; |
10c94ec1 TG |
1505 | } |
1506 | ||
080344b9 ON |
1507 | static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp) |
1508 | { | |
1509 | struct timespec rmt; | |
1510 | ktime_t rem; | |
1511 | ||
cc584b21 | 1512 | rem = hrtimer_expires_remaining(timer); |
080344b9 ON |
1513 | if (rem.tv64 <= 0) |
1514 | return 0; | |
1515 | rmt = ktime_to_timespec(rem); | |
1516 | ||
1517 | if (copy_to_user(rmtp, &rmt, sizeof(*rmtp))) | |
1518 | return -EFAULT; | |
1519 | ||
1520 | return 1; | |
1521 | } | |
1522 | ||
1711ef38 | 1523 | long __sched hrtimer_nanosleep_restart(struct restart_block *restart) |
10c94ec1 | 1524 | { |
669d7868 | 1525 | struct hrtimer_sleeper t; |
080344b9 | 1526 | struct timespec __user *rmtp; |
237fc6e7 | 1527 | int ret = 0; |
10c94ec1 | 1528 | |
237fc6e7 TG |
1529 | hrtimer_init_on_stack(&t.timer, restart->nanosleep.index, |
1530 | HRTIMER_MODE_ABS); | |
cc584b21 | 1531 | hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires); |
10c94ec1 | 1532 | |
c9cb2e3d | 1533 | if (do_nanosleep(&t, HRTIMER_MODE_ABS)) |
237fc6e7 | 1534 | goto out; |
10c94ec1 | 1535 | |
029a07e0 | 1536 | rmtp = restart->nanosleep.rmtp; |
432569bb | 1537 | if (rmtp) { |
237fc6e7 | 1538 | ret = update_rmtp(&t.timer, rmtp); |
080344b9 | 1539 | if (ret <= 0) |
237fc6e7 | 1540 | goto out; |
432569bb | 1541 | } |
10c94ec1 | 1542 | |
10c94ec1 | 1543 | /* The other values in restart are already filled in */ |
237fc6e7 TG |
1544 | ret = -ERESTART_RESTARTBLOCK; |
1545 | out: | |
1546 | destroy_hrtimer_on_stack(&t.timer); | |
1547 | return ret; | |
10c94ec1 TG |
1548 | } |
1549 | ||
080344b9 | 1550 | long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp, |
10c94ec1 TG |
1551 | const enum hrtimer_mode mode, const clockid_t clockid) |
1552 | { | |
1553 | struct restart_block *restart; | |
669d7868 | 1554 | struct hrtimer_sleeper t; |
237fc6e7 | 1555 | int ret = 0; |
3bd01206 AV |
1556 | unsigned long slack; |
1557 | ||
1558 | slack = current->timer_slack_ns; | |
1559 | if (rt_task(current)) | |
1560 | slack = 0; | |
10c94ec1 | 1561 | |
237fc6e7 | 1562 | hrtimer_init_on_stack(&t.timer, clockid, mode); |
3bd01206 | 1563 | hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack); |
432569bb | 1564 | if (do_nanosleep(&t, mode)) |
237fc6e7 | 1565 | goto out; |
10c94ec1 | 1566 | |
7978672c | 1567 | /* Absolute timers do not update the rmtp value and restart: */ |
237fc6e7 TG |
1568 | if (mode == HRTIMER_MODE_ABS) { |
1569 | ret = -ERESTARTNOHAND; | |
1570 | goto out; | |
1571 | } | |
10c94ec1 | 1572 | |
432569bb | 1573 | if (rmtp) { |
237fc6e7 | 1574 | ret = update_rmtp(&t.timer, rmtp); |
080344b9 | 1575 | if (ret <= 0) |
237fc6e7 | 1576 | goto out; |
432569bb | 1577 | } |
10c94ec1 TG |
1578 | |
1579 | restart = ¤t_thread_info()->restart_block; | |
1711ef38 | 1580 | restart->fn = hrtimer_nanosleep_restart; |
029a07e0 TG |
1581 | restart->nanosleep.index = t.timer.base->index; |
1582 | restart->nanosleep.rmtp = rmtp; | |
cc584b21 | 1583 | restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer); |
10c94ec1 | 1584 | |
237fc6e7 TG |
1585 | ret = -ERESTART_RESTARTBLOCK; |
1586 | out: | |
1587 | destroy_hrtimer_on_stack(&t.timer); | |
1588 | return ret; | |
10c94ec1 TG |
1589 | } |
1590 | ||
58fd3aa2 HC |
1591 | SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp, |
1592 | struct timespec __user *, rmtp) | |
6ba1b912 | 1593 | { |
080344b9 | 1594 | struct timespec tu; |
6ba1b912 TG |
1595 | |
1596 | if (copy_from_user(&tu, rqtp, sizeof(tu))) | |
1597 | return -EFAULT; | |
1598 | ||
1599 | if (!timespec_valid(&tu)) | |
1600 | return -EINVAL; | |
1601 | ||
080344b9 | 1602 | return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC); |
6ba1b912 TG |
1603 | } |
1604 | ||
c0a31329 TG |
1605 | /* |
1606 | * Functions related to boot-time initialization: | |
1607 | */ | |
0ec160dd | 1608 | static void __cpuinit init_hrtimers_cpu(int cpu) |
c0a31329 | 1609 | { |
3c8aa39d | 1610 | struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu); |
c0a31329 TG |
1611 | int i; |
1612 | ||
ecb49d1a | 1613 | raw_spin_lock_init(&cpu_base->lock); |
3c8aa39d | 1614 | |
998adc3d | 1615 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
3c8aa39d | 1616 | cpu_base->clock_base[i].cpu_base = cpu_base; |
998adc3d JS |
1617 | timerqueue_init_head(&cpu_base->clock_base[i].active); |
1618 | } | |
3c8aa39d | 1619 | |
54cdfdb4 | 1620 | hrtimer_init_hres(cpu_base); |
c0a31329 TG |
1621 | } |
1622 | ||
1623 | #ifdef CONFIG_HOTPLUG_CPU | |
1624 | ||
ca109491 | 1625 | static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base, |
37810659 | 1626 | struct hrtimer_clock_base *new_base) |
c0a31329 TG |
1627 | { |
1628 | struct hrtimer *timer; | |
998adc3d | 1629 | struct timerqueue_node *node; |
c0a31329 | 1630 | |
998adc3d JS |
1631 | while ((node = timerqueue_getnext(&old_base->active))) { |
1632 | timer = container_of(node, struct hrtimer, node); | |
54cdfdb4 | 1633 | BUG_ON(hrtimer_callback_running(timer)); |
c6a2a177 | 1634 | debug_deactivate(timer); |
b00c1a99 TG |
1635 | |
1636 | /* | |
1637 | * Mark it as STATE_MIGRATE not INACTIVE otherwise the | |
1638 | * timer could be seen as !active and just vanish away | |
1639 | * under us on another CPU | |
1640 | */ | |
1641 | __remove_hrtimer(timer, old_base, HRTIMER_STATE_MIGRATE, 0); | |
c0a31329 | 1642 | timer->base = new_base; |
54cdfdb4 | 1643 | /* |
e3f1d883 TG |
1644 | * Enqueue the timers on the new cpu. This does not |
1645 | * reprogram the event device in case the timer | |
1646 | * expires before the earliest on this CPU, but we run | |
1647 | * hrtimer_interrupt after we migrated everything to | |
1648 | * sort out already expired timers and reprogram the | |
1649 | * event device. | |
54cdfdb4 | 1650 | */ |
a6037b61 | 1651 | enqueue_hrtimer(timer, new_base); |
41e1022e | 1652 | |
b00c1a99 TG |
1653 | /* Clear the migration state bit */ |
1654 | timer->state &= ~HRTIMER_STATE_MIGRATE; | |
c0a31329 TG |
1655 | } |
1656 | } | |
1657 | ||
d5fd43c4 | 1658 | static void migrate_hrtimers(int scpu) |
c0a31329 | 1659 | { |
3c8aa39d | 1660 | struct hrtimer_cpu_base *old_base, *new_base; |
731a55ba | 1661 | int i; |
c0a31329 | 1662 | |
37810659 | 1663 | BUG_ON(cpu_online(scpu)); |
37810659 | 1664 | tick_cancel_sched_timer(scpu); |
731a55ba TG |
1665 | |
1666 | local_irq_disable(); | |
1667 | old_base = &per_cpu(hrtimer_bases, scpu); | |
1668 | new_base = &__get_cpu_var(hrtimer_bases); | |
d82f0b0f ON |
1669 | /* |
1670 | * The caller is globally serialized and nobody else | |
1671 | * takes two locks at once, deadlock is not possible. | |
1672 | */ | |
ecb49d1a TG |
1673 | raw_spin_lock(&new_base->lock); |
1674 | raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING); | |
c0a31329 | 1675 | |
3c8aa39d | 1676 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
ca109491 | 1677 | migrate_hrtimer_list(&old_base->clock_base[i], |
37810659 | 1678 | &new_base->clock_base[i]); |
c0a31329 TG |
1679 | } |
1680 | ||
ecb49d1a TG |
1681 | raw_spin_unlock(&old_base->lock); |
1682 | raw_spin_unlock(&new_base->lock); | |
37810659 | 1683 | |
731a55ba TG |
1684 | /* Check, if we got expired work to do */ |
1685 | __hrtimer_peek_ahead_timers(); | |
1686 | local_irq_enable(); | |
c0a31329 | 1687 | } |
37810659 | 1688 | |
c0a31329 TG |
1689 | #endif /* CONFIG_HOTPLUG_CPU */ |
1690 | ||
8c78f307 | 1691 | static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self, |
c0a31329 TG |
1692 | unsigned long action, void *hcpu) |
1693 | { | |
b2e3c0ad | 1694 | int scpu = (long)hcpu; |
c0a31329 TG |
1695 | |
1696 | switch (action) { | |
1697 | ||
1698 | case CPU_UP_PREPARE: | |
8bb78442 | 1699 | case CPU_UP_PREPARE_FROZEN: |
37810659 | 1700 | init_hrtimers_cpu(scpu); |
c0a31329 TG |
1701 | break; |
1702 | ||
1703 | #ifdef CONFIG_HOTPLUG_CPU | |
94df7de0 SD |
1704 | case CPU_DYING: |
1705 | case CPU_DYING_FROZEN: | |
1706 | clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DYING, &scpu); | |
1707 | break; | |
c0a31329 | 1708 | case CPU_DEAD: |
8bb78442 | 1709 | case CPU_DEAD_FROZEN: |
b2e3c0ad | 1710 | { |
37810659 | 1711 | clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &scpu); |
d5fd43c4 | 1712 | migrate_hrtimers(scpu); |
c0a31329 | 1713 | break; |
b2e3c0ad | 1714 | } |
c0a31329 TG |
1715 | #endif |
1716 | ||
1717 | default: | |
1718 | break; | |
1719 | } | |
1720 | ||
1721 | return NOTIFY_OK; | |
1722 | } | |
1723 | ||
8c78f307 | 1724 | static struct notifier_block __cpuinitdata hrtimers_nb = { |
c0a31329 TG |
1725 | .notifier_call = hrtimer_cpu_notify, |
1726 | }; | |
1727 | ||
1728 | void __init hrtimers_init(void) | |
1729 | { | |
1730 | hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE, | |
1731 | (void *)(long)smp_processor_id()); | |
1732 | register_cpu_notifier(&hrtimers_nb); | |
a6037b61 PZ |
1733 | #ifdef CONFIG_HIGH_RES_TIMERS |
1734 | open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq); | |
1735 | #endif | |
c0a31329 TG |
1736 | } |
1737 | ||
7bb67439 | 1738 | /** |
351b3f7a | 1739 | * schedule_hrtimeout_range_clock - sleep until timeout |
7bb67439 | 1740 | * @expires: timeout value (ktime_t) |
654c8e0b | 1741 | * @delta: slack in expires timeout (ktime_t) |
7bb67439 | 1742 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL |
351b3f7a | 1743 | * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME |
7bb67439 | 1744 | */ |
351b3f7a CE |
1745 | int __sched |
1746 | schedule_hrtimeout_range_clock(ktime_t *expires, unsigned long delta, | |
1747 | const enum hrtimer_mode mode, int clock) | |
7bb67439 AV |
1748 | { |
1749 | struct hrtimer_sleeper t; | |
1750 | ||
1751 | /* | |
1752 | * Optimize when a zero timeout value is given. It does not | |
1753 | * matter whether this is an absolute or a relative time. | |
1754 | */ | |
1755 | if (expires && !expires->tv64) { | |
1756 | __set_current_state(TASK_RUNNING); | |
1757 | return 0; | |
1758 | } | |
1759 | ||
1760 | /* | |
43b21013 | 1761 | * A NULL parameter means "infinite" |
7bb67439 AV |
1762 | */ |
1763 | if (!expires) { | |
1764 | schedule(); | |
1765 | __set_current_state(TASK_RUNNING); | |
1766 | return -EINTR; | |
1767 | } | |
1768 | ||
351b3f7a | 1769 | hrtimer_init_on_stack(&t.timer, clock, mode); |
654c8e0b | 1770 | hrtimer_set_expires_range_ns(&t.timer, *expires, delta); |
7bb67439 AV |
1771 | |
1772 | hrtimer_init_sleeper(&t, current); | |
1773 | ||
cc584b21 | 1774 | hrtimer_start_expires(&t.timer, mode); |
7bb67439 AV |
1775 | if (!hrtimer_active(&t.timer)) |
1776 | t.task = NULL; | |
1777 | ||
1778 | if (likely(t.task)) | |
1779 | schedule(); | |
1780 | ||
1781 | hrtimer_cancel(&t.timer); | |
1782 | destroy_hrtimer_on_stack(&t.timer); | |
1783 | ||
1784 | __set_current_state(TASK_RUNNING); | |
1785 | ||
1786 | return !t.task ? 0 : -EINTR; | |
1787 | } | |
351b3f7a CE |
1788 | |
1789 | /** | |
1790 | * schedule_hrtimeout_range - sleep until timeout | |
1791 | * @expires: timeout value (ktime_t) | |
1792 | * @delta: slack in expires timeout (ktime_t) | |
1793 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL | |
1794 | * | |
1795 | * Make the current task sleep until the given expiry time has | |
1796 | * elapsed. The routine will return immediately unless | |
1797 | * the current task state has been set (see set_current_state()). | |
1798 | * | |
1799 | * The @delta argument gives the kernel the freedom to schedule the | |
1800 | * actual wakeup to a time that is both power and performance friendly. | |
1801 | * The kernel give the normal best effort behavior for "@expires+@delta", | |
1802 | * but may decide to fire the timer earlier, but no earlier than @expires. | |
1803 | * | |
1804 | * You can set the task state as follows - | |
1805 | * | |
1806 | * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to | |
1807 | * pass before the routine returns. | |
1808 | * | |
1809 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
1810 | * delivered to the current task. | |
1811 | * | |
1812 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1813 | * routine returns. | |
1814 | * | |
1815 | * Returns 0 when the timer has expired otherwise -EINTR | |
1816 | */ | |
1817 | int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta, | |
1818 | const enum hrtimer_mode mode) | |
1819 | { | |
1820 | return schedule_hrtimeout_range_clock(expires, delta, mode, | |
1821 | CLOCK_MONOTONIC); | |
1822 | } | |
654c8e0b AV |
1823 | EXPORT_SYMBOL_GPL(schedule_hrtimeout_range); |
1824 | ||
1825 | /** | |
1826 | * schedule_hrtimeout - sleep until timeout | |
1827 | * @expires: timeout value (ktime_t) | |
1828 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL | |
1829 | * | |
1830 | * Make the current task sleep until the given expiry time has | |
1831 | * elapsed. The routine will return immediately unless | |
1832 | * the current task state has been set (see set_current_state()). | |
1833 | * | |
1834 | * You can set the task state as follows - | |
1835 | * | |
1836 | * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to | |
1837 | * pass before the routine returns. | |
1838 | * | |
1839 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
1840 | * delivered to the current task. | |
1841 | * | |
1842 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1843 | * routine returns. | |
1844 | * | |
1845 | * Returns 0 when the timer has expired otherwise -EINTR | |
1846 | */ | |
1847 | int __sched schedule_hrtimeout(ktime_t *expires, | |
1848 | const enum hrtimer_mode mode) | |
1849 | { | |
1850 | return schedule_hrtimeout_range(expires, 0, mode); | |
1851 | } | |
7bb67439 | 1852 | EXPORT_SYMBOL_GPL(schedule_hrtimeout); |