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> | |
9984de1a | 35 | #include <linux/export.h> |
c0a31329 TG |
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 | 46 | #include <linux/sched.h> |
cf4aebc2 | 47 | #include <linux/sched/sysctl.h> |
eea08f32 | 48 | #include <linux/timer.h> |
c0a31329 TG |
49 | |
50 | #include <asm/uaccess.h> | |
51 | ||
c6a2a177 XG |
52 | #include <trace/events/timer.h> |
53 | ||
c0a31329 TG |
54 | /* |
55 | * The timer bases: | |
7978672c | 56 | * |
e06383db JS |
57 | * There are more clockids then hrtimer bases. Thus, we index |
58 | * into the timer bases by the hrtimer_base_type enum. When trying | |
59 | * to reach a base using a clockid, hrtimer_clockid_to_base() | |
60 | * is used to convert from clockid to the proper hrtimer_base_type. | |
c0a31329 | 61 | */ |
54cdfdb4 | 62 | DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) = |
c0a31329 | 63 | { |
3c8aa39d TG |
64 | |
65 | .clock_base = | |
c0a31329 | 66 | { |
3c8aa39d | 67 | { |
ab8177bc TG |
68 | .index = HRTIMER_BASE_MONOTONIC, |
69 | .clockid = CLOCK_MONOTONIC, | |
3c8aa39d | 70 | .get_time = &ktime_get, |
54cdfdb4 | 71 | .resolution = KTIME_LOW_RES, |
3c8aa39d | 72 | }, |
68fa61c0 TG |
73 | { |
74 | .index = HRTIMER_BASE_REALTIME, | |
75 | .clockid = CLOCK_REALTIME, | |
76 | .get_time = &ktime_get_real, | |
77 | .resolution = KTIME_LOW_RES, | |
78 | }, | |
70a08cca | 79 | { |
ab8177bc TG |
80 | .index = HRTIMER_BASE_BOOTTIME, |
81 | .clockid = CLOCK_BOOTTIME, | |
70a08cca JS |
82 | .get_time = &ktime_get_boottime, |
83 | .resolution = KTIME_LOW_RES, | |
84 | }, | |
3c8aa39d | 85 | } |
c0a31329 TG |
86 | }; |
87 | ||
942c3c5c | 88 | static const int hrtimer_clock_to_base_table[MAX_CLOCKS] = { |
ce31332d TG |
89 | [CLOCK_REALTIME] = HRTIMER_BASE_REALTIME, |
90 | [CLOCK_MONOTONIC] = HRTIMER_BASE_MONOTONIC, | |
91 | [CLOCK_BOOTTIME] = HRTIMER_BASE_BOOTTIME, | |
92 | }; | |
e06383db JS |
93 | |
94 | static inline int hrtimer_clockid_to_base(clockid_t clock_id) | |
95 | { | |
96 | return hrtimer_clock_to_base_table[clock_id]; | |
97 | } | |
98 | ||
99 | ||
92127c7a TG |
100 | /* |
101 | * Get the coarse grained time at the softirq based on xtime and | |
102 | * wall_to_monotonic. | |
103 | */ | |
3c8aa39d | 104 | static void hrtimer_get_softirq_time(struct hrtimer_cpu_base *base) |
92127c7a | 105 | { |
70a08cca | 106 | ktime_t xtim, mono, boot; |
314ac371 | 107 | struct timespec xts, tom, slp; |
92127c7a | 108 | |
314ac371 | 109 | get_xtime_and_monotonic_and_sleep_offset(&xts, &tom, &slp); |
92127c7a | 110 | |
f4304ab2 | 111 | xtim = timespec_to_ktime(xts); |
70a08cca JS |
112 | mono = ktime_add(xtim, timespec_to_ktime(tom)); |
113 | boot = ktime_add(mono, timespec_to_ktime(slp)); | |
e06383db | 114 | base->clock_base[HRTIMER_BASE_REALTIME].softirq_time = xtim; |
70a08cca JS |
115 | base->clock_base[HRTIMER_BASE_MONOTONIC].softirq_time = mono; |
116 | base->clock_base[HRTIMER_BASE_BOOTTIME].softirq_time = boot; | |
92127c7a TG |
117 | } |
118 | ||
c0a31329 TG |
119 | /* |
120 | * Functions and macros which are different for UP/SMP systems are kept in a | |
121 | * single place | |
122 | */ | |
123 | #ifdef CONFIG_SMP | |
124 | ||
c0a31329 TG |
125 | /* |
126 | * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock | |
127 | * means that all timers which are tied to this base via timer->base are | |
128 | * locked, and the base itself is locked too. | |
129 | * | |
130 | * So __run_timers/migrate_timers can safely modify all timers which could | |
131 | * be found on the lists/queues. | |
132 | * | |
133 | * When the timer's base is locked, and the timer removed from list, it is | |
134 | * possible to set timer->base = NULL and drop the lock: the timer remains | |
135 | * locked. | |
136 | */ | |
3c8aa39d TG |
137 | static |
138 | struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer, | |
139 | unsigned long *flags) | |
c0a31329 | 140 | { |
3c8aa39d | 141 | struct hrtimer_clock_base *base; |
c0a31329 TG |
142 | |
143 | for (;;) { | |
144 | base = timer->base; | |
145 | if (likely(base != NULL)) { | |
ecb49d1a | 146 | raw_spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
147 | if (likely(base == timer->base)) |
148 | return base; | |
149 | /* The timer has migrated to another CPU: */ | |
ecb49d1a | 150 | raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags); |
c0a31329 TG |
151 | } |
152 | cpu_relax(); | |
153 | } | |
154 | } | |
155 | ||
6ff7041d TG |
156 | |
157 | /* | |
158 | * Get the preferred target CPU for NOHZ | |
159 | */ | |
160 | static int hrtimer_get_target(int this_cpu, int pinned) | |
161 | { | |
162 | #ifdef CONFIG_NO_HZ | |
83cd4fe2 VP |
163 | if (!pinned && get_sysctl_timer_migration() && idle_cpu(this_cpu)) |
164 | return get_nohz_timer_target(); | |
6ff7041d TG |
165 | #endif |
166 | return this_cpu; | |
167 | } | |
168 | ||
169 | /* | |
170 | * With HIGHRES=y we do not migrate the timer when it is expiring | |
171 | * before the next event on the target cpu because we cannot reprogram | |
172 | * the target cpu hardware and we would cause it to fire late. | |
173 | * | |
174 | * Called with cpu_base->lock of target cpu held. | |
175 | */ | |
176 | static int | |
177 | hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base) | |
178 | { | |
179 | #ifdef CONFIG_HIGH_RES_TIMERS | |
180 | ktime_t expires; | |
181 | ||
182 | if (!new_base->cpu_base->hres_active) | |
183 | return 0; | |
184 | ||
185 | expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset); | |
186 | return expires.tv64 <= new_base->cpu_base->expires_next.tv64; | |
187 | #else | |
188 | return 0; | |
189 | #endif | |
190 | } | |
191 | ||
c0a31329 TG |
192 | /* |
193 | * Switch the timer base to the current CPU when possible. | |
194 | */ | |
3c8aa39d | 195 | static inline struct hrtimer_clock_base * |
597d0275 AB |
196 | switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base, |
197 | int pinned) | |
c0a31329 | 198 | { |
3c8aa39d TG |
199 | struct hrtimer_clock_base *new_base; |
200 | struct hrtimer_cpu_base *new_cpu_base; | |
6ff7041d TG |
201 | int this_cpu = smp_processor_id(); |
202 | int cpu = hrtimer_get_target(this_cpu, pinned); | |
ab8177bc | 203 | int basenum = base->index; |
c0a31329 | 204 | |
eea08f32 AB |
205 | again: |
206 | new_cpu_base = &per_cpu(hrtimer_bases, cpu); | |
e06383db | 207 | new_base = &new_cpu_base->clock_base[basenum]; |
c0a31329 TG |
208 | |
209 | if (base != new_base) { | |
210 | /* | |
6ff7041d | 211 | * We are trying to move timer to new_base. |
c0a31329 TG |
212 | * However we can't change timer's base while it is running, |
213 | * so we keep it on the same CPU. No hassle vs. reprogramming | |
214 | * the event source in the high resolution case. The softirq | |
215 | * code will take care of this when the timer function has | |
216 | * completed. There is no conflict as we hold the lock until | |
217 | * the timer is enqueued. | |
218 | */ | |
54cdfdb4 | 219 | if (unlikely(hrtimer_callback_running(timer))) |
c0a31329 TG |
220 | return base; |
221 | ||
222 | /* See the comment in lock_timer_base() */ | |
223 | timer->base = NULL; | |
ecb49d1a TG |
224 | raw_spin_unlock(&base->cpu_base->lock); |
225 | raw_spin_lock(&new_base->cpu_base->lock); | |
eea08f32 | 226 | |
6ff7041d TG |
227 | if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) { |
228 | cpu = this_cpu; | |
ecb49d1a TG |
229 | raw_spin_unlock(&new_base->cpu_base->lock); |
230 | raw_spin_lock(&base->cpu_base->lock); | |
6ff7041d TG |
231 | timer->base = base; |
232 | goto again; | |
eea08f32 | 233 | } |
c0a31329 TG |
234 | timer->base = new_base; |
235 | } | |
236 | return new_base; | |
237 | } | |
238 | ||
239 | #else /* CONFIG_SMP */ | |
240 | ||
3c8aa39d | 241 | static inline struct hrtimer_clock_base * |
c0a31329 TG |
242 | lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) |
243 | { | |
3c8aa39d | 244 | struct hrtimer_clock_base *base = timer->base; |
c0a31329 | 245 | |
ecb49d1a | 246 | raw_spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
247 | |
248 | return base; | |
249 | } | |
250 | ||
eea08f32 | 251 | # define switch_hrtimer_base(t, b, p) (b) |
c0a31329 TG |
252 | |
253 | #endif /* !CONFIG_SMP */ | |
254 | ||
255 | /* | |
256 | * Functions for the union type storage format of ktime_t which are | |
257 | * too large for inlining: | |
258 | */ | |
259 | #if BITS_PER_LONG < 64 | |
260 | # ifndef CONFIG_KTIME_SCALAR | |
261 | /** | |
262 | * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable | |
c0a31329 TG |
263 | * @kt: addend |
264 | * @nsec: the scalar nsec value to add | |
265 | * | |
266 | * Returns the sum of kt and nsec in ktime_t format | |
267 | */ | |
268 | ktime_t ktime_add_ns(const ktime_t kt, u64 nsec) | |
269 | { | |
270 | ktime_t tmp; | |
271 | ||
272 | if (likely(nsec < NSEC_PER_SEC)) { | |
273 | tmp.tv64 = nsec; | |
274 | } else { | |
275 | unsigned long rem = do_div(nsec, NSEC_PER_SEC); | |
276 | ||
277 | tmp = ktime_set((long)nsec, rem); | |
278 | } | |
279 | ||
280 | return ktime_add(kt, tmp); | |
281 | } | |
b8b8fd2d DH |
282 | |
283 | EXPORT_SYMBOL_GPL(ktime_add_ns); | |
a272378d ACM |
284 | |
285 | /** | |
286 | * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable | |
287 | * @kt: minuend | |
288 | * @nsec: the scalar nsec value to subtract | |
289 | * | |
290 | * Returns the subtraction of @nsec from @kt in ktime_t format | |
291 | */ | |
292 | ktime_t ktime_sub_ns(const ktime_t kt, u64 nsec) | |
293 | { | |
294 | ktime_t tmp; | |
295 | ||
296 | if (likely(nsec < NSEC_PER_SEC)) { | |
297 | tmp.tv64 = nsec; | |
298 | } else { | |
299 | unsigned long rem = do_div(nsec, NSEC_PER_SEC); | |
300 | ||
301 | tmp = ktime_set((long)nsec, rem); | |
302 | } | |
303 | ||
304 | return ktime_sub(kt, tmp); | |
305 | } | |
306 | ||
307 | EXPORT_SYMBOL_GPL(ktime_sub_ns); | |
c0a31329 TG |
308 | # endif /* !CONFIG_KTIME_SCALAR */ |
309 | ||
310 | /* | |
311 | * Divide a ktime value by a nanosecond value | |
312 | */ | |
4d672e7a | 313 | u64 ktime_divns(const ktime_t kt, s64 div) |
c0a31329 | 314 | { |
900cfa46 | 315 | u64 dclc; |
c0a31329 TG |
316 | int sft = 0; |
317 | ||
900cfa46 | 318 | dclc = ktime_to_ns(kt); |
c0a31329 TG |
319 | /* Make sure the divisor is less than 2^32: */ |
320 | while (div >> 32) { | |
321 | sft++; | |
322 | div >>= 1; | |
323 | } | |
324 | dclc >>= sft; | |
325 | do_div(dclc, (unsigned long) div); | |
326 | ||
4d672e7a | 327 | return dclc; |
c0a31329 | 328 | } |
c0a31329 TG |
329 | #endif /* BITS_PER_LONG >= 64 */ |
330 | ||
5a7780e7 TG |
331 | /* |
332 | * Add two ktime values and do a safety check for overflow: | |
333 | */ | |
334 | ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs) | |
335 | { | |
336 | ktime_t res = ktime_add(lhs, rhs); | |
337 | ||
338 | /* | |
339 | * We use KTIME_SEC_MAX here, the maximum timeout which we can | |
340 | * return to user space in a timespec: | |
341 | */ | |
342 | if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64) | |
343 | res = ktime_set(KTIME_SEC_MAX, 0); | |
344 | ||
345 | return res; | |
346 | } | |
347 | ||
8daa21e6 AB |
348 | EXPORT_SYMBOL_GPL(ktime_add_safe); |
349 | ||
237fc6e7 TG |
350 | #ifdef CONFIG_DEBUG_OBJECTS_TIMERS |
351 | ||
352 | static struct debug_obj_descr hrtimer_debug_descr; | |
353 | ||
99777288 SG |
354 | static void *hrtimer_debug_hint(void *addr) |
355 | { | |
356 | return ((struct hrtimer *) addr)->function; | |
357 | } | |
358 | ||
237fc6e7 TG |
359 | /* |
360 | * fixup_init is called when: | |
361 | * - an active object is initialized | |
362 | */ | |
363 | static int hrtimer_fixup_init(void *addr, enum debug_obj_state state) | |
364 | { | |
365 | struct hrtimer *timer = addr; | |
366 | ||
367 | switch (state) { | |
368 | case ODEBUG_STATE_ACTIVE: | |
369 | hrtimer_cancel(timer); | |
370 | debug_object_init(timer, &hrtimer_debug_descr); | |
371 | return 1; | |
372 | default: | |
373 | return 0; | |
374 | } | |
375 | } | |
376 | ||
377 | /* | |
378 | * fixup_activate is called when: | |
379 | * - an active object is activated | |
380 | * - an unknown object is activated (might be a statically initialized object) | |
381 | */ | |
382 | static int hrtimer_fixup_activate(void *addr, enum debug_obj_state state) | |
383 | { | |
384 | switch (state) { | |
385 | ||
386 | case ODEBUG_STATE_NOTAVAILABLE: | |
387 | WARN_ON_ONCE(1); | |
388 | return 0; | |
389 | ||
390 | case ODEBUG_STATE_ACTIVE: | |
391 | WARN_ON(1); | |
392 | ||
393 | default: | |
394 | return 0; | |
395 | } | |
396 | } | |
397 | ||
398 | /* | |
399 | * fixup_free is called when: | |
400 | * - an active object is freed | |
401 | */ | |
402 | static int hrtimer_fixup_free(void *addr, enum debug_obj_state state) | |
403 | { | |
404 | struct hrtimer *timer = addr; | |
405 | ||
406 | switch (state) { | |
407 | case ODEBUG_STATE_ACTIVE: | |
408 | hrtimer_cancel(timer); | |
409 | debug_object_free(timer, &hrtimer_debug_descr); | |
410 | return 1; | |
411 | default: | |
412 | return 0; | |
413 | } | |
414 | } | |
415 | ||
416 | static struct debug_obj_descr hrtimer_debug_descr = { | |
417 | .name = "hrtimer", | |
99777288 | 418 | .debug_hint = hrtimer_debug_hint, |
237fc6e7 TG |
419 | .fixup_init = hrtimer_fixup_init, |
420 | .fixup_activate = hrtimer_fixup_activate, | |
421 | .fixup_free = hrtimer_fixup_free, | |
422 | }; | |
423 | ||
424 | static inline void debug_hrtimer_init(struct hrtimer *timer) | |
425 | { | |
426 | debug_object_init(timer, &hrtimer_debug_descr); | |
427 | } | |
428 | ||
429 | static inline void debug_hrtimer_activate(struct hrtimer *timer) | |
430 | { | |
431 | debug_object_activate(timer, &hrtimer_debug_descr); | |
432 | } | |
433 | ||
434 | static inline void debug_hrtimer_deactivate(struct hrtimer *timer) | |
435 | { | |
436 | debug_object_deactivate(timer, &hrtimer_debug_descr); | |
437 | } | |
438 | ||
439 | static inline void debug_hrtimer_free(struct hrtimer *timer) | |
440 | { | |
441 | debug_object_free(timer, &hrtimer_debug_descr); | |
442 | } | |
443 | ||
444 | static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id, | |
445 | enum hrtimer_mode mode); | |
446 | ||
447 | void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id, | |
448 | enum hrtimer_mode mode) | |
449 | { | |
450 | debug_object_init_on_stack(timer, &hrtimer_debug_descr); | |
451 | __hrtimer_init(timer, clock_id, mode); | |
452 | } | |
2bc481cf | 453 | EXPORT_SYMBOL_GPL(hrtimer_init_on_stack); |
237fc6e7 TG |
454 | |
455 | void destroy_hrtimer_on_stack(struct hrtimer *timer) | |
456 | { | |
457 | debug_object_free(timer, &hrtimer_debug_descr); | |
458 | } | |
459 | ||
460 | #else | |
461 | static inline void debug_hrtimer_init(struct hrtimer *timer) { } | |
462 | static inline void debug_hrtimer_activate(struct hrtimer *timer) { } | |
463 | static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { } | |
464 | #endif | |
465 | ||
c6a2a177 XG |
466 | static inline void |
467 | debug_init(struct hrtimer *timer, clockid_t clockid, | |
468 | enum hrtimer_mode mode) | |
469 | { | |
470 | debug_hrtimer_init(timer); | |
471 | trace_hrtimer_init(timer, clockid, mode); | |
472 | } | |
473 | ||
474 | static inline void debug_activate(struct hrtimer *timer) | |
475 | { | |
476 | debug_hrtimer_activate(timer); | |
477 | trace_hrtimer_start(timer); | |
478 | } | |
479 | ||
480 | static inline void debug_deactivate(struct hrtimer *timer) | |
481 | { | |
482 | debug_hrtimer_deactivate(timer); | |
483 | trace_hrtimer_cancel(timer); | |
484 | } | |
485 | ||
54cdfdb4 TG |
486 | /* High resolution timer related functions */ |
487 | #ifdef CONFIG_HIGH_RES_TIMERS | |
488 | ||
489 | /* | |
490 | * High resolution timer enabled ? | |
491 | */ | |
492 | static int hrtimer_hres_enabled __read_mostly = 1; | |
493 | ||
494 | /* | |
495 | * Enable / Disable high resolution mode | |
496 | */ | |
497 | static int __init setup_hrtimer_hres(char *str) | |
498 | { | |
499 | if (!strcmp(str, "off")) | |
500 | hrtimer_hres_enabled = 0; | |
501 | else if (!strcmp(str, "on")) | |
502 | hrtimer_hres_enabled = 1; | |
503 | else | |
504 | return 0; | |
505 | return 1; | |
506 | } | |
507 | ||
508 | __setup("highres=", setup_hrtimer_hres); | |
509 | ||
510 | /* | |
511 | * hrtimer_high_res_enabled - query, if the highres mode is enabled | |
512 | */ | |
513 | static inline int hrtimer_is_hres_enabled(void) | |
514 | { | |
515 | return hrtimer_hres_enabled; | |
516 | } | |
517 | ||
518 | /* | |
519 | * Is the high resolution mode active ? | |
520 | */ | |
521 | static inline int hrtimer_hres_active(void) | |
522 | { | |
909ea964 | 523 | return __this_cpu_read(hrtimer_bases.hres_active); |
54cdfdb4 TG |
524 | } |
525 | ||
526 | /* | |
527 | * Reprogram the event source with checking both queues for the | |
528 | * next event | |
529 | * Called with interrupts disabled and base->lock held | |
530 | */ | |
7403f41f AC |
531 | static void |
532 | hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal) | |
54cdfdb4 TG |
533 | { |
534 | int i; | |
535 | struct hrtimer_clock_base *base = cpu_base->clock_base; | |
7403f41f | 536 | ktime_t expires, expires_next; |
54cdfdb4 | 537 | |
7403f41f | 538 | expires_next.tv64 = KTIME_MAX; |
54cdfdb4 TG |
539 | |
540 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) { | |
541 | struct hrtimer *timer; | |
998adc3d | 542 | struct timerqueue_node *next; |
54cdfdb4 | 543 | |
998adc3d JS |
544 | next = timerqueue_getnext(&base->active); |
545 | if (!next) | |
54cdfdb4 | 546 | continue; |
998adc3d JS |
547 | timer = container_of(next, struct hrtimer, node); |
548 | ||
cc584b21 | 549 | expires = ktime_sub(hrtimer_get_expires(timer), base->offset); |
b0a9b511 TG |
550 | /* |
551 | * clock_was_set() has changed base->offset so the | |
552 | * result might be negative. Fix it up to prevent a | |
553 | * false positive in clockevents_program_event() | |
554 | */ | |
555 | if (expires.tv64 < 0) | |
556 | expires.tv64 = 0; | |
7403f41f AC |
557 | if (expires.tv64 < expires_next.tv64) |
558 | expires_next = expires; | |
54cdfdb4 TG |
559 | } |
560 | ||
7403f41f AC |
561 | if (skip_equal && expires_next.tv64 == cpu_base->expires_next.tv64) |
562 | return; | |
563 | ||
564 | cpu_base->expires_next.tv64 = expires_next.tv64; | |
565 | ||
54cdfdb4 TG |
566 | if (cpu_base->expires_next.tv64 != KTIME_MAX) |
567 | tick_program_event(cpu_base->expires_next, 1); | |
568 | } | |
569 | ||
570 | /* | |
571 | * Shared reprogramming for clock_realtime and clock_monotonic | |
572 | * | |
573 | * When a timer is enqueued and expires earlier than the already enqueued | |
574 | * timers, we have to check, whether it expires earlier than the timer for | |
575 | * which the clock event device was armed. | |
576 | * | |
577 | * Called with interrupts disabled and base->cpu_base.lock held | |
578 | */ | |
579 | static int hrtimer_reprogram(struct hrtimer *timer, | |
580 | struct hrtimer_clock_base *base) | |
581 | { | |
41d2e494 | 582 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
cc584b21 | 583 | ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset); |
54cdfdb4 TG |
584 | int res; |
585 | ||
cc584b21 | 586 | WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0); |
63070a79 | 587 | |
54cdfdb4 TG |
588 | /* |
589 | * When the callback is running, we do not reprogram the clock event | |
590 | * device. The timer callback is either running on a different CPU or | |
3a4fa0a2 | 591 | * the callback is executed in the hrtimer_interrupt context. The |
54cdfdb4 TG |
592 | * reprogramming is handled either by the softirq, which called the |
593 | * callback or at the end of the hrtimer_interrupt. | |
594 | */ | |
595 | if (hrtimer_callback_running(timer)) | |
596 | return 0; | |
597 | ||
63070a79 TG |
598 | /* |
599 | * CLOCK_REALTIME timer might be requested with an absolute | |
600 | * expiry time which is less than base->offset. Nothing wrong | |
601 | * about that, just avoid to call into the tick code, which | |
602 | * has now objections against negative expiry values. | |
603 | */ | |
604 | if (expires.tv64 < 0) | |
605 | return -ETIME; | |
606 | ||
41d2e494 TG |
607 | if (expires.tv64 >= cpu_base->expires_next.tv64) |
608 | return 0; | |
609 | ||
610 | /* | |
611 | * If a hang was detected in the last timer interrupt then we | |
612 | * do not schedule a timer which is earlier than the expiry | |
613 | * which we enforced in the hang detection. We want the system | |
614 | * to make progress. | |
615 | */ | |
616 | if (cpu_base->hang_detected) | |
54cdfdb4 TG |
617 | return 0; |
618 | ||
619 | /* | |
620 | * Clockevents returns -ETIME, when the event was in the past. | |
621 | */ | |
622 | res = tick_program_event(expires, 0); | |
623 | if (!IS_ERR_VALUE(res)) | |
41d2e494 | 624 | cpu_base->expires_next = expires; |
54cdfdb4 TG |
625 | return res; |
626 | } | |
627 | ||
54cdfdb4 TG |
628 | /* |
629 | * Initialize the high resolution related parts of cpu_base | |
630 | */ | |
631 | static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) | |
632 | { | |
633 | base->expires_next.tv64 = KTIME_MAX; | |
634 | base->hres_active = 0; | |
54cdfdb4 TG |
635 | } |
636 | ||
54cdfdb4 TG |
637 | /* |
638 | * When High resolution timers are active, try to reprogram. Note, that in case | |
639 | * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry | |
640 | * check happens. The timer gets enqueued into the rbtree. The reprogramming | |
641 | * and expiry check is done in the hrtimer_interrupt or in the softirq. | |
642 | */ | |
643 | static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer, | |
7f1e2ca9 PZ |
644 | struct hrtimer_clock_base *base, |
645 | int wakeup) | |
54cdfdb4 TG |
646 | { |
647 | if (base->cpu_base->hres_active && hrtimer_reprogram(timer, base)) { | |
7f1e2ca9 | 648 | if (wakeup) { |
ecb49d1a | 649 | raw_spin_unlock(&base->cpu_base->lock); |
7f1e2ca9 | 650 | raise_softirq_irqoff(HRTIMER_SOFTIRQ); |
ecb49d1a | 651 | raw_spin_lock(&base->cpu_base->lock); |
7f1e2ca9 PZ |
652 | } else |
653 | __raise_softirq_irqoff(HRTIMER_SOFTIRQ); | |
654 | ||
ca109491 | 655 | return 1; |
54cdfdb4 | 656 | } |
7f1e2ca9 | 657 | |
54cdfdb4 TG |
658 | return 0; |
659 | } | |
660 | ||
5baefd6d JS |
661 | static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base) |
662 | { | |
663 | ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset; | |
664 | ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset; | |
665 | ||
666 | return ktime_get_update_offsets(offs_real, offs_boot); | |
667 | } | |
668 | ||
9ec26907 TG |
669 | /* |
670 | * Retrigger next event is called after clock was set | |
671 | * | |
672 | * Called with interrupts disabled via on_each_cpu() | |
673 | */ | |
674 | static void retrigger_next_event(void *arg) | |
675 | { | |
676 | struct hrtimer_cpu_base *base = &__get_cpu_var(hrtimer_bases); | |
9ec26907 TG |
677 | |
678 | if (!hrtimer_hres_active()) | |
679 | return; | |
680 | ||
9ec26907 | 681 | raw_spin_lock(&base->lock); |
5baefd6d | 682 | hrtimer_update_base(base); |
9ec26907 TG |
683 | hrtimer_force_reprogram(base, 0); |
684 | raw_spin_unlock(&base->lock); | |
685 | } | |
b12a03ce | 686 | |
54cdfdb4 TG |
687 | /* |
688 | * Switch to high resolution mode | |
689 | */ | |
f8953856 | 690 | static int hrtimer_switch_to_hres(void) |
54cdfdb4 | 691 | { |
b12a03ce | 692 | int i, cpu = smp_processor_id(); |
820de5c3 | 693 | struct hrtimer_cpu_base *base = &per_cpu(hrtimer_bases, cpu); |
54cdfdb4 TG |
694 | unsigned long flags; |
695 | ||
696 | if (base->hres_active) | |
f8953856 | 697 | return 1; |
54cdfdb4 TG |
698 | |
699 | local_irq_save(flags); | |
700 | ||
701 | if (tick_init_highres()) { | |
702 | local_irq_restore(flags); | |
820de5c3 IM |
703 | printk(KERN_WARNING "Could not switch to high resolution " |
704 | "mode on CPU %d\n", cpu); | |
f8953856 | 705 | return 0; |
54cdfdb4 TG |
706 | } |
707 | base->hres_active = 1; | |
b12a03ce TG |
708 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) |
709 | base->clock_base[i].resolution = KTIME_HIGH_RES; | |
54cdfdb4 TG |
710 | |
711 | tick_setup_sched_timer(); | |
54cdfdb4 TG |
712 | /* "Retrigger" the interrupt to get things going */ |
713 | retrigger_next_event(NULL); | |
714 | local_irq_restore(flags); | |
f8953856 | 715 | return 1; |
54cdfdb4 TG |
716 | } |
717 | ||
f55a6faa JS |
718 | /* |
719 | * Called from timekeeping code to reprogramm the hrtimer interrupt | |
720 | * device. If called from the timer interrupt context we defer it to | |
721 | * softirq context. | |
722 | */ | |
723 | void clock_was_set_delayed(void) | |
724 | { | |
725 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); | |
726 | ||
727 | cpu_base->clock_was_set = 1; | |
728 | __raise_softirq_irqoff(HRTIMER_SOFTIRQ); | |
729 | } | |
730 | ||
54cdfdb4 TG |
731 | #else |
732 | ||
733 | static inline int hrtimer_hres_active(void) { return 0; } | |
734 | static inline int hrtimer_is_hres_enabled(void) { return 0; } | |
f8953856 | 735 | static inline int hrtimer_switch_to_hres(void) { return 0; } |
7403f41f AC |
736 | static inline void |
737 | hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { } | |
54cdfdb4 | 738 | static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer, |
7f1e2ca9 PZ |
739 | struct hrtimer_clock_base *base, |
740 | int wakeup) | |
54cdfdb4 TG |
741 | { |
742 | return 0; | |
743 | } | |
54cdfdb4 | 744 | static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { } |
9ec26907 | 745 | static inline void retrigger_next_event(void *arg) { } |
54cdfdb4 TG |
746 | |
747 | #endif /* CONFIG_HIGH_RES_TIMERS */ | |
748 | ||
b12a03ce TG |
749 | /* |
750 | * Clock realtime was set | |
751 | * | |
752 | * Change the offset of the realtime clock vs. the monotonic | |
753 | * clock. | |
754 | * | |
755 | * We might have to reprogram the high resolution timer interrupt. On | |
756 | * SMP we call the architecture specific code to retrigger _all_ high | |
757 | * resolution timer interrupts. On UP we just disable interrupts and | |
758 | * call the high resolution interrupt code. | |
759 | */ | |
760 | void clock_was_set(void) | |
761 | { | |
90ff1f30 | 762 | #ifdef CONFIG_HIGH_RES_TIMERS |
b12a03ce TG |
763 | /* Retrigger the CPU local events everywhere */ |
764 | on_each_cpu(retrigger_next_event, NULL, 1); | |
9ec26907 TG |
765 | #endif |
766 | timerfd_clock_was_set(); | |
b12a03ce TG |
767 | } |
768 | ||
769 | /* | |
770 | * During resume we might have to reprogram the high resolution timer | |
771 | * interrupt (on the local CPU): | |
772 | */ | |
773 | void hrtimers_resume(void) | |
774 | { | |
775 | WARN_ONCE(!irqs_disabled(), | |
776 | KERN_INFO "hrtimers_resume() called with IRQs enabled!"); | |
777 | ||
778 | retrigger_next_event(NULL); | |
9ec26907 | 779 | timerfd_clock_was_set(); |
b12a03ce TG |
780 | } |
781 | ||
5f201907 | 782 | static inline void timer_stats_hrtimer_set_start_info(struct hrtimer *timer) |
82f67cd9 | 783 | { |
5f201907 | 784 | #ifdef CONFIG_TIMER_STATS |
82f67cd9 IM |
785 | if (timer->start_site) |
786 | return; | |
5f201907 | 787 | timer->start_site = __builtin_return_address(0); |
82f67cd9 IM |
788 | memcpy(timer->start_comm, current->comm, TASK_COMM_LEN); |
789 | timer->start_pid = current->pid; | |
5f201907 HC |
790 | #endif |
791 | } | |
792 | ||
793 | static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer *timer) | |
794 | { | |
795 | #ifdef CONFIG_TIMER_STATS | |
796 | timer->start_site = NULL; | |
797 | #endif | |
82f67cd9 | 798 | } |
5f201907 HC |
799 | |
800 | static inline void timer_stats_account_hrtimer(struct hrtimer *timer) | |
801 | { | |
802 | #ifdef CONFIG_TIMER_STATS | |
803 | if (likely(!timer_stats_active)) | |
804 | return; | |
805 | timer_stats_update_stats(timer, timer->start_pid, timer->start_site, | |
806 | timer->function, timer->start_comm, 0); | |
82f67cd9 | 807 | #endif |
5f201907 | 808 | } |
82f67cd9 | 809 | |
c0a31329 | 810 | /* |
6506f2aa | 811 | * Counterpart to lock_hrtimer_base above: |
c0a31329 TG |
812 | */ |
813 | static inline | |
814 | void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) | |
815 | { | |
ecb49d1a | 816 | raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags); |
c0a31329 TG |
817 | } |
818 | ||
819 | /** | |
820 | * hrtimer_forward - forward the timer expiry | |
c0a31329 | 821 | * @timer: hrtimer to forward |
44f21475 | 822 | * @now: forward past this time |
c0a31329 TG |
823 | * @interval: the interval to forward |
824 | * | |
825 | * Forward the timer expiry so it will expire in the future. | |
8dca6f33 | 826 | * Returns the number of overruns. |
c0a31329 | 827 | */ |
4d672e7a | 828 | u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval) |
c0a31329 | 829 | { |
4d672e7a | 830 | u64 orun = 1; |
44f21475 | 831 | ktime_t delta; |
c0a31329 | 832 | |
cc584b21 | 833 | delta = ktime_sub(now, hrtimer_get_expires(timer)); |
c0a31329 TG |
834 | |
835 | if (delta.tv64 < 0) | |
836 | return 0; | |
837 | ||
c9db4fa1 TG |
838 | if (interval.tv64 < timer->base->resolution.tv64) |
839 | interval.tv64 = timer->base->resolution.tv64; | |
840 | ||
c0a31329 | 841 | if (unlikely(delta.tv64 >= interval.tv64)) { |
df869b63 | 842 | s64 incr = ktime_to_ns(interval); |
c0a31329 TG |
843 | |
844 | orun = ktime_divns(delta, incr); | |
cc584b21 AV |
845 | hrtimer_add_expires_ns(timer, incr * orun); |
846 | if (hrtimer_get_expires_tv64(timer) > now.tv64) | |
c0a31329 TG |
847 | return orun; |
848 | /* | |
849 | * This (and the ktime_add() below) is the | |
850 | * correction for exact: | |
851 | */ | |
852 | orun++; | |
853 | } | |
cc584b21 | 854 | hrtimer_add_expires(timer, interval); |
c0a31329 TG |
855 | |
856 | return orun; | |
857 | } | |
6bdb6b62 | 858 | EXPORT_SYMBOL_GPL(hrtimer_forward); |
c0a31329 TG |
859 | |
860 | /* | |
861 | * enqueue_hrtimer - internal function to (re)start a timer | |
862 | * | |
863 | * The timer is inserted in expiry order. Insertion into the | |
864 | * red black tree is O(log(n)). Must hold the base lock. | |
a6037b61 PZ |
865 | * |
866 | * Returns 1 when the new timer is the leftmost timer in the tree. | |
c0a31329 | 867 | */ |
a6037b61 PZ |
868 | static int enqueue_hrtimer(struct hrtimer *timer, |
869 | struct hrtimer_clock_base *base) | |
c0a31329 | 870 | { |
c6a2a177 | 871 | debug_activate(timer); |
237fc6e7 | 872 | |
998adc3d | 873 | timerqueue_add(&base->active, &timer->node); |
ab8177bc | 874 | base->cpu_base->active_bases |= 1 << base->index; |
54cdfdb4 | 875 | |
303e967f TG |
876 | /* |
877 | * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the | |
878 | * state of a possibly running callback. | |
879 | */ | |
880 | timer->state |= HRTIMER_STATE_ENQUEUED; | |
a6037b61 | 881 | |
998adc3d | 882 | return (&timer->node == base->active.next); |
288867ec | 883 | } |
c0a31329 TG |
884 | |
885 | /* | |
886 | * __remove_hrtimer - internal function to remove a timer | |
887 | * | |
888 | * Caller must hold the base lock. | |
54cdfdb4 TG |
889 | * |
890 | * High resolution timer mode reprograms the clock event device when the | |
891 | * timer is the one which expires next. The caller can disable this by setting | |
892 | * reprogram to zero. This is useful, when the context does a reprogramming | |
893 | * anyway (e.g. timer interrupt) | |
c0a31329 | 894 | */ |
3c8aa39d | 895 | static void __remove_hrtimer(struct hrtimer *timer, |
303e967f | 896 | struct hrtimer_clock_base *base, |
54cdfdb4 | 897 | unsigned long newstate, int reprogram) |
c0a31329 | 898 | { |
27c9cd7e | 899 | struct timerqueue_node *next_timer; |
7403f41f AC |
900 | if (!(timer->state & HRTIMER_STATE_ENQUEUED)) |
901 | goto out; | |
902 | ||
27c9cd7e JO |
903 | next_timer = timerqueue_getnext(&base->active); |
904 | timerqueue_del(&base->active, &timer->node); | |
905 | if (&timer->node == next_timer) { | |
7403f41f AC |
906 | #ifdef CONFIG_HIGH_RES_TIMERS |
907 | /* Reprogram the clock event device. if enabled */ | |
908 | if (reprogram && hrtimer_hres_active()) { | |
909 | ktime_t expires; | |
910 | ||
911 | expires = ktime_sub(hrtimer_get_expires(timer), | |
912 | base->offset); | |
913 | if (base->cpu_base->expires_next.tv64 == expires.tv64) | |
914 | hrtimer_force_reprogram(base->cpu_base, 1); | |
54cdfdb4 | 915 | } |
7403f41f | 916 | #endif |
54cdfdb4 | 917 | } |
ab8177bc TG |
918 | if (!timerqueue_getnext(&base->active)) |
919 | base->cpu_base->active_bases &= ~(1 << base->index); | |
7403f41f | 920 | out: |
303e967f | 921 | timer->state = newstate; |
c0a31329 TG |
922 | } |
923 | ||
924 | /* | |
925 | * remove hrtimer, called with base lock held | |
926 | */ | |
927 | static inline int | |
3c8aa39d | 928 | remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base) |
c0a31329 | 929 | { |
303e967f | 930 | if (hrtimer_is_queued(timer)) { |
f13d4f97 | 931 | unsigned long state; |
54cdfdb4 TG |
932 | int reprogram; |
933 | ||
934 | /* | |
935 | * Remove the timer and force reprogramming when high | |
936 | * resolution mode is active and the timer is on the current | |
937 | * CPU. If we remove a timer on another CPU, reprogramming is | |
938 | * skipped. The interrupt event on this CPU is fired and | |
939 | * reprogramming happens in the interrupt handler. This is a | |
940 | * rare case and less expensive than a smp call. | |
941 | */ | |
c6a2a177 | 942 | debug_deactivate(timer); |
82f67cd9 | 943 | timer_stats_hrtimer_clear_start_info(timer); |
54cdfdb4 | 944 | reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases); |
f13d4f97 SQ |
945 | /* |
946 | * We must preserve the CALLBACK state flag here, | |
947 | * otherwise we could move the timer base in | |
948 | * switch_hrtimer_base. | |
949 | */ | |
950 | state = timer->state & HRTIMER_STATE_CALLBACK; | |
951 | __remove_hrtimer(timer, base, state, reprogram); | |
c0a31329 TG |
952 | return 1; |
953 | } | |
954 | return 0; | |
955 | } | |
956 | ||
7f1e2ca9 PZ |
957 | int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, |
958 | unsigned long delta_ns, const enum hrtimer_mode mode, | |
959 | int wakeup) | |
c0a31329 | 960 | { |
3c8aa39d | 961 | struct hrtimer_clock_base *base, *new_base; |
c0a31329 | 962 | unsigned long flags; |
a6037b61 | 963 | int ret, leftmost; |
c0a31329 TG |
964 | |
965 | base = lock_hrtimer_base(timer, &flags); | |
966 | ||
967 | /* Remove an active timer from the queue: */ | |
968 | ret = remove_hrtimer(timer, base); | |
969 | ||
970 | /* Switch the timer base, if necessary: */ | |
597d0275 | 971 | new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED); |
c0a31329 | 972 | |
597d0275 | 973 | if (mode & HRTIMER_MODE_REL) { |
5a7780e7 | 974 | tim = ktime_add_safe(tim, new_base->get_time()); |
06027bdd IM |
975 | /* |
976 | * CONFIG_TIME_LOW_RES is a temporary way for architectures | |
977 | * to signal that they simply return xtime in | |
978 | * do_gettimeoffset(). In this case we want to round up by | |
979 | * resolution when starting a relative timer, to avoid short | |
980 | * timeouts. This will go away with the GTOD framework. | |
981 | */ | |
982 | #ifdef CONFIG_TIME_LOW_RES | |
5a7780e7 | 983 | tim = ktime_add_safe(tim, base->resolution); |
06027bdd IM |
984 | #endif |
985 | } | |
237fc6e7 | 986 | |
da8f2e17 | 987 | hrtimer_set_expires_range_ns(timer, tim, delta_ns); |
c0a31329 | 988 | |
82f67cd9 IM |
989 | timer_stats_hrtimer_set_start_info(timer); |
990 | ||
a6037b61 PZ |
991 | leftmost = enqueue_hrtimer(timer, new_base); |
992 | ||
935c631d IM |
993 | /* |
994 | * Only allow reprogramming if the new base is on this CPU. | |
995 | * (it might still be on another CPU if the timer was pending) | |
a6037b61 PZ |
996 | * |
997 | * XXX send_remote_softirq() ? | |
935c631d | 998 | */ |
a6037b61 | 999 | if (leftmost && new_base->cpu_base == &__get_cpu_var(hrtimer_bases)) |
7f1e2ca9 | 1000 | hrtimer_enqueue_reprogram(timer, new_base, wakeup); |
c0a31329 TG |
1001 | |
1002 | unlock_hrtimer_base(timer, &flags); | |
1003 | ||
1004 | return ret; | |
1005 | } | |
7f1e2ca9 PZ |
1006 | |
1007 | /** | |
1008 | * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU | |
1009 | * @timer: the timer to be added | |
1010 | * @tim: expiry time | |
1011 | * @delta_ns: "slack" range for the timer | |
1012 | * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL) | |
1013 | * | |
1014 | * Returns: | |
1015 | * 0 on success | |
1016 | * 1 when the timer was active | |
1017 | */ | |
1018 | int hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, | |
1019 | unsigned long delta_ns, const enum hrtimer_mode mode) | |
1020 | { | |
1021 | return __hrtimer_start_range_ns(timer, tim, delta_ns, mode, 1); | |
1022 | } | |
da8f2e17 AV |
1023 | EXPORT_SYMBOL_GPL(hrtimer_start_range_ns); |
1024 | ||
1025 | /** | |
e1dd7bc5 | 1026 | * hrtimer_start - (re)start an hrtimer on the current CPU |
da8f2e17 AV |
1027 | * @timer: the timer to be added |
1028 | * @tim: expiry time | |
1029 | * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL) | |
1030 | * | |
1031 | * Returns: | |
1032 | * 0 on success | |
1033 | * 1 when the timer was active | |
1034 | */ | |
1035 | int | |
1036 | hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode) | |
1037 | { | |
7f1e2ca9 | 1038 | return __hrtimer_start_range_ns(timer, tim, 0, mode, 1); |
da8f2e17 | 1039 | } |
8d16b764 | 1040 | EXPORT_SYMBOL_GPL(hrtimer_start); |
c0a31329 | 1041 | |
da8f2e17 | 1042 | |
c0a31329 TG |
1043 | /** |
1044 | * hrtimer_try_to_cancel - try to deactivate a timer | |
c0a31329 TG |
1045 | * @timer: hrtimer to stop |
1046 | * | |
1047 | * Returns: | |
1048 | * 0 when the timer was not active | |
1049 | * 1 when the timer was active | |
1050 | * -1 when the timer is currently excuting the callback function and | |
fa9799e3 | 1051 | * cannot be stopped |
c0a31329 TG |
1052 | */ |
1053 | int hrtimer_try_to_cancel(struct hrtimer *timer) | |
1054 | { | |
3c8aa39d | 1055 | struct hrtimer_clock_base *base; |
c0a31329 TG |
1056 | unsigned long flags; |
1057 | int ret = -1; | |
1058 | ||
1059 | base = lock_hrtimer_base(timer, &flags); | |
1060 | ||
303e967f | 1061 | if (!hrtimer_callback_running(timer)) |
c0a31329 TG |
1062 | ret = remove_hrtimer(timer, base); |
1063 | ||
1064 | unlock_hrtimer_base(timer, &flags); | |
1065 | ||
1066 | return ret; | |
1067 | ||
1068 | } | |
8d16b764 | 1069 | EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel); |
c0a31329 TG |
1070 | |
1071 | /** | |
1072 | * hrtimer_cancel - cancel a timer and wait for the handler to finish. | |
c0a31329 TG |
1073 | * @timer: the timer to be cancelled |
1074 | * | |
1075 | * Returns: | |
1076 | * 0 when the timer was not active | |
1077 | * 1 when the timer was active | |
1078 | */ | |
1079 | int hrtimer_cancel(struct hrtimer *timer) | |
1080 | { | |
1081 | for (;;) { | |
1082 | int ret = hrtimer_try_to_cancel(timer); | |
1083 | ||
1084 | if (ret >= 0) | |
1085 | return ret; | |
5ef37b19 | 1086 | cpu_relax(); |
c0a31329 TG |
1087 | } |
1088 | } | |
8d16b764 | 1089 | EXPORT_SYMBOL_GPL(hrtimer_cancel); |
c0a31329 TG |
1090 | |
1091 | /** | |
1092 | * hrtimer_get_remaining - get remaining time for the timer | |
c0a31329 TG |
1093 | * @timer: the timer to read |
1094 | */ | |
1095 | ktime_t hrtimer_get_remaining(const struct hrtimer *timer) | |
1096 | { | |
c0a31329 TG |
1097 | unsigned long flags; |
1098 | ktime_t rem; | |
1099 | ||
b3bd3de6 | 1100 | lock_hrtimer_base(timer, &flags); |
cc584b21 | 1101 | rem = hrtimer_expires_remaining(timer); |
c0a31329 TG |
1102 | unlock_hrtimer_base(timer, &flags); |
1103 | ||
1104 | return rem; | |
1105 | } | |
8d16b764 | 1106 | EXPORT_SYMBOL_GPL(hrtimer_get_remaining); |
c0a31329 | 1107 | |
ee9c5785 | 1108 | #ifdef CONFIG_NO_HZ |
69239749 TL |
1109 | /** |
1110 | * hrtimer_get_next_event - get the time until next expiry event | |
1111 | * | |
1112 | * Returns the delta to the next expiry event or KTIME_MAX if no timer | |
1113 | * is pending. | |
1114 | */ | |
1115 | ktime_t hrtimer_get_next_event(void) | |
1116 | { | |
3c8aa39d TG |
1117 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
1118 | struct hrtimer_clock_base *base = cpu_base->clock_base; | |
69239749 TL |
1119 | ktime_t delta, mindelta = { .tv64 = KTIME_MAX }; |
1120 | unsigned long flags; | |
1121 | int i; | |
1122 | ||
ecb49d1a | 1123 | raw_spin_lock_irqsave(&cpu_base->lock, flags); |
3c8aa39d | 1124 | |
54cdfdb4 TG |
1125 | if (!hrtimer_hres_active()) { |
1126 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) { | |
1127 | struct hrtimer *timer; | |
998adc3d | 1128 | struct timerqueue_node *next; |
69239749 | 1129 | |
998adc3d JS |
1130 | next = timerqueue_getnext(&base->active); |
1131 | if (!next) | |
54cdfdb4 | 1132 | continue; |
3c8aa39d | 1133 | |
998adc3d | 1134 | timer = container_of(next, struct hrtimer, node); |
cc584b21 | 1135 | delta.tv64 = hrtimer_get_expires_tv64(timer); |
54cdfdb4 TG |
1136 | delta = ktime_sub(delta, base->get_time()); |
1137 | if (delta.tv64 < mindelta.tv64) | |
1138 | mindelta.tv64 = delta.tv64; | |
1139 | } | |
69239749 | 1140 | } |
3c8aa39d | 1141 | |
ecb49d1a | 1142 | raw_spin_unlock_irqrestore(&cpu_base->lock, flags); |
3c8aa39d | 1143 | |
69239749 TL |
1144 | if (mindelta.tv64 < 0) |
1145 | mindelta.tv64 = 0; | |
1146 | return mindelta; | |
1147 | } | |
1148 | #endif | |
1149 | ||
237fc6e7 TG |
1150 | static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id, |
1151 | enum hrtimer_mode mode) | |
c0a31329 | 1152 | { |
3c8aa39d | 1153 | struct hrtimer_cpu_base *cpu_base; |
e06383db | 1154 | int base; |
c0a31329 | 1155 | |
7978672c GA |
1156 | memset(timer, 0, sizeof(struct hrtimer)); |
1157 | ||
3c8aa39d | 1158 | cpu_base = &__raw_get_cpu_var(hrtimer_bases); |
c0a31329 | 1159 | |
c9cb2e3d | 1160 | if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS) |
7978672c GA |
1161 | clock_id = CLOCK_MONOTONIC; |
1162 | ||
e06383db JS |
1163 | base = hrtimer_clockid_to_base(clock_id); |
1164 | timer->base = &cpu_base->clock_base[base]; | |
998adc3d | 1165 | timerqueue_init(&timer->node); |
82f67cd9 IM |
1166 | |
1167 | #ifdef CONFIG_TIMER_STATS | |
1168 | timer->start_site = NULL; | |
1169 | timer->start_pid = -1; | |
1170 | memset(timer->start_comm, 0, TASK_COMM_LEN); | |
1171 | #endif | |
c0a31329 | 1172 | } |
237fc6e7 TG |
1173 | |
1174 | /** | |
1175 | * hrtimer_init - initialize a timer to the given clock | |
1176 | * @timer: the timer to be initialized | |
1177 | * @clock_id: the clock to be used | |
1178 | * @mode: timer mode abs/rel | |
1179 | */ | |
1180 | void hrtimer_init(struct hrtimer *timer, clockid_t clock_id, | |
1181 | enum hrtimer_mode mode) | |
1182 | { | |
c6a2a177 | 1183 | debug_init(timer, clock_id, mode); |
237fc6e7 TG |
1184 | __hrtimer_init(timer, clock_id, mode); |
1185 | } | |
8d16b764 | 1186 | EXPORT_SYMBOL_GPL(hrtimer_init); |
c0a31329 TG |
1187 | |
1188 | /** | |
1189 | * hrtimer_get_res - get the timer resolution for a clock | |
c0a31329 TG |
1190 | * @which_clock: which clock to query |
1191 | * @tp: pointer to timespec variable to store the resolution | |
1192 | * | |
72fd4a35 RD |
1193 | * Store the resolution of the clock selected by @which_clock in the |
1194 | * variable pointed to by @tp. | |
c0a31329 TG |
1195 | */ |
1196 | int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp) | |
1197 | { | |
3c8aa39d | 1198 | struct hrtimer_cpu_base *cpu_base; |
e06383db | 1199 | int base = hrtimer_clockid_to_base(which_clock); |
c0a31329 | 1200 | |
3c8aa39d | 1201 | cpu_base = &__raw_get_cpu_var(hrtimer_bases); |
e06383db | 1202 | *tp = ktime_to_timespec(cpu_base->clock_base[base].resolution); |
c0a31329 TG |
1203 | |
1204 | return 0; | |
1205 | } | |
8d16b764 | 1206 | EXPORT_SYMBOL_GPL(hrtimer_get_res); |
c0a31329 | 1207 | |
c6a2a177 | 1208 | static void __run_hrtimer(struct hrtimer *timer, ktime_t *now) |
d3d74453 PZ |
1209 | { |
1210 | struct hrtimer_clock_base *base = timer->base; | |
1211 | struct hrtimer_cpu_base *cpu_base = base->cpu_base; | |
1212 | enum hrtimer_restart (*fn)(struct hrtimer *); | |
1213 | int restart; | |
1214 | ||
ca109491 PZ |
1215 | WARN_ON(!irqs_disabled()); |
1216 | ||
c6a2a177 | 1217 | debug_deactivate(timer); |
d3d74453 PZ |
1218 | __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0); |
1219 | timer_stats_account_hrtimer(timer); | |
d3d74453 | 1220 | fn = timer->function; |
ca109491 PZ |
1221 | |
1222 | /* | |
1223 | * Because we run timers from hardirq context, there is no chance | |
1224 | * they get migrated to another cpu, therefore its safe to unlock | |
1225 | * the timer base. | |
1226 | */ | |
ecb49d1a | 1227 | raw_spin_unlock(&cpu_base->lock); |
c6a2a177 | 1228 | trace_hrtimer_expire_entry(timer, now); |
ca109491 | 1229 | restart = fn(timer); |
c6a2a177 | 1230 | trace_hrtimer_expire_exit(timer); |
ecb49d1a | 1231 | raw_spin_lock(&cpu_base->lock); |
d3d74453 PZ |
1232 | |
1233 | /* | |
e3f1d883 TG |
1234 | * Note: We clear the CALLBACK bit after enqueue_hrtimer and |
1235 | * we do not reprogramm the event hardware. Happens either in | |
1236 | * hrtimer_start_range_ns() or in hrtimer_interrupt() | |
d3d74453 PZ |
1237 | */ |
1238 | if (restart != HRTIMER_NORESTART) { | |
1239 | BUG_ON(timer->state != HRTIMER_STATE_CALLBACK); | |
a6037b61 | 1240 | enqueue_hrtimer(timer, base); |
d3d74453 | 1241 | } |
f13d4f97 SQ |
1242 | |
1243 | WARN_ON_ONCE(!(timer->state & HRTIMER_STATE_CALLBACK)); | |
1244 | ||
d3d74453 PZ |
1245 | timer->state &= ~HRTIMER_STATE_CALLBACK; |
1246 | } | |
1247 | ||
54cdfdb4 TG |
1248 | #ifdef CONFIG_HIGH_RES_TIMERS |
1249 | ||
1250 | /* | |
1251 | * High resolution timer interrupt | |
1252 | * Called with interrupts disabled | |
1253 | */ | |
1254 | void hrtimer_interrupt(struct clock_event_device *dev) | |
1255 | { | |
1256 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); | |
41d2e494 TG |
1257 | ktime_t expires_next, now, entry_time, delta; |
1258 | int i, retries = 0; | |
54cdfdb4 TG |
1259 | |
1260 | BUG_ON(!cpu_base->hres_active); | |
1261 | cpu_base->nr_events++; | |
1262 | dev->next_event.tv64 = KTIME_MAX; | |
1263 | ||
196951e9 | 1264 | raw_spin_lock(&cpu_base->lock); |
5baefd6d | 1265 | entry_time = now = hrtimer_update_base(cpu_base); |
41d2e494 | 1266 | retry: |
54cdfdb4 | 1267 | expires_next.tv64 = KTIME_MAX; |
6ff7041d TG |
1268 | /* |
1269 | * We set expires_next to KTIME_MAX here with cpu_base->lock | |
1270 | * held to prevent that a timer is enqueued in our queue via | |
1271 | * the migration code. This does not affect enqueueing of | |
1272 | * timers which run their callback and need to be requeued on | |
1273 | * this CPU. | |
1274 | */ | |
1275 | cpu_base->expires_next.tv64 = KTIME_MAX; | |
1276 | ||
54cdfdb4 | 1277 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
ab8177bc | 1278 | struct hrtimer_clock_base *base; |
998adc3d | 1279 | struct timerqueue_node *node; |
ab8177bc TG |
1280 | ktime_t basenow; |
1281 | ||
1282 | if (!(cpu_base->active_bases & (1 << i))) | |
1283 | continue; | |
54cdfdb4 | 1284 | |
ab8177bc | 1285 | base = cpu_base->clock_base + i; |
54cdfdb4 TG |
1286 | basenow = ktime_add(now, base->offset); |
1287 | ||
998adc3d | 1288 | while ((node = timerqueue_getnext(&base->active))) { |
54cdfdb4 TG |
1289 | struct hrtimer *timer; |
1290 | ||
998adc3d | 1291 | timer = container_of(node, struct hrtimer, node); |
54cdfdb4 | 1292 | |
654c8e0b AV |
1293 | /* |
1294 | * The immediate goal for using the softexpires is | |
1295 | * minimizing wakeups, not running timers at the | |
1296 | * earliest interrupt after their soft expiration. | |
1297 | * This allows us to avoid using a Priority Search | |
1298 | * Tree, which can answer a stabbing querry for | |
1299 | * overlapping intervals and instead use the simple | |
1300 | * BST we already have. | |
1301 | * We don't add extra wakeups by delaying timers that | |
1302 | * are right-of a not yet expired timer, because that | |
1303 | * timer will have to trigger a wakeup anyway. | |
1304 | */ | |
1305 | ||
1306 | if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer)) { | |
54cdfdb4 TG |
1307 | ktime_t expires; |
1308 | ||
cc584b21 | 1309 | expires = ktime_sub(hrtimer_get_expires(timer), |
54cdfdb4 TG |
1310 | base->offset); |
1311 | if (expires.tv64 < expires_next.tv64) | |
1312 | expires_next = expires; | |
1313 | break; | |
1314 | } | |
1315 | ||
c6a2a177 | 1316 | __run_hrtimer(timer, &basenow); |
54cdfdb4 | 1317 | } |
54cdfdb4 TG |
1318 | } |
1319 | ||
6ff7041d TG |
1320 | /* |
1321 | * Store the new expiry value so the migration code can verify | |
1322 | * against it. | |
1323 | */ | |
54cdfdb4 | 1324 | cpu_base->expires_next = expires_next; |
ecb49d1a | 1325 | raw_spin_unlock(&cpu_base->lock); |
54cdfdb4 TG |
1326 | |
1327 | /* Reprogramming necessary ? */ | |
41d2e494 TG |
1328 | if (expires_next.tv64 == KTIME_MAX || |
1329 | !tick_program_event(expires_next, 0)) { | |
1330 | cpu_base->hang_detected = 0; | |
1331 | return; | |
54cdfdb4 | 1332 | } |
41d2e494 TG |
1333 | |
1334 | /* | |
1335 | * The next timer was already expired due to: | |
1336 | * - tracing | |
1337 | * - long lasting callbacks | |
1338 | * - being scheduled away when running in a VM | |
1339 | * | |
1340 | * We need to prevent that we loop forever in the hrtimer | |
1341 | * interrupt routine. We give it 3 attempts to avoid | |
1342 | * overreacting on some spurious event. | |
5baefd6d JS |
1343 | * |
1344 | * Acquire base lock for updating the offsets and retrieving | |
1345 | * the current time. | |
41d2e494 | 1346 | */ |
196951e9 | 1347 | raw_spin_lock(&cpu_base->lock); |
5baefd6d | 1348 | now = hrtimer_update_base(cpu_base); |
41d2e494 TG |
1349 | cpu_base->nr_retries++; |
1350 | if (++retries < 3) | |
1351 | goto retry; | |
1352 | /* | |
1353 | * Give the system a chance to do something else than looping | |
1354 | * here. We stored the entry time, so we know exactly how long | |
1355 | * we spent here. We schedule the next event this amount of | |
1356 | * time away. | |
1357 | */ | |
1358 | cpu_base->nr_hangs++; | |
1359 | cpu_base->hang_detected = 1; | |
196951e9 | 1360 | raw_spin_unlock(&cpu_base->lock); |
41d2e494 TG |
1361 | delta = ktime_sub(now, entry_time); |
1362 | if (delta.tv64 > cpu_base->max_hang_time.tv64) | |
1363 | cpu_base->max_hang_time = delta; | |
1364 | /* | |
1365 | * Limit it to a sensible value as we enforce a longer | |
1366 | * delay. Give the CPU at least 100ms to catch up. | |
1367 | */ | |
1368 | if (delta.tv64 > 100 * NSEC_PER_MSEC) | |
1369 | expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC); | |
1370 | else | |
1371 | expires_next = ktime_add(now, delta); | |
1372 | tick_program_event(expires_next, 1); | |
1373 | printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n", | |
1374 | ktime_to_ns(delta)); | |
54cdfdb4 TG |
1375 | } |
1376 | ||
8bdec955 TG |
1377 | /* |
1378 | * local version of hrtimer_peek_ahead_timers() called with interrupts | |
1379 | * disabled. | |
1380 | */ | |
1381 | static void __hrtimer_peek_ahead_timers(void) | |
1382 | { | |
1383 | struct tick_device *td; | |
1384 | ||
1385 | if (!hrtimer_hres_active()) | |
1386 | return; | |
1387 | ||
1388 | td = &__get_cpu_var(tick_cpu_device); | |
1389 | if (td && td->evtdev) | |
1390 | hrtimer_interrupt(td->evtdev); | |
1391 | } | |
1392 | ||
2e94d1f7 AV |
1393 | /** |
1394 | * hrtimer_peek_ahead_timers -- run soft-expired timers now | |
1395 | * | |
1396 | * hrtimer_peek_ahead_timers will peek at the timer queue of | |
1397 | * the current cpu and check if there are any timers for which | |
1398 | * the soft expires time has passed. If any such timers exist, | |
1399 | * they are run immediately and then removed from the timer queue. | |
1400 | * | |
1401 | */ | |
1402 | void hrtimer_peek_ahead_timers(void) | |
1403 | { | |
643bdf68 | 1404 | unsigned long flags; |
dc4304f7 | 1405 | |
2e94d1f7 | 1406 | local_irq_save(flags); |
8bdec955 | 1407 | __hrtimer_peek_ahead_timers(); |
2e94d1f7 AV |
1408 | local_irq_restore(flags); |
1409 | } | |
1410 | ||
a6037b61 PZ |
1411 | static void run_hrtimer_softirq(struct softirq_action *h) |
1412 | { | |
f55a6faa JS |
1413 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
1414 | ||
1415 | if (cpu_base->clock_was_set) { | |
1416 | cpu_base->clock_was_set = 0; | |
1417 | clock_was_set(); | |
1418 | } | |
1419 | ||
a6037b61 PZ |
1420 | hrtimer_peek_ahead_timers(); |
1421 | } | |
1422 | ||
82c5b7b5 IM |
1423 | #else /* CONFIG_HIGH_RES_TIMERS */ |
1424 | ||
1425 | static inline void __hrtimer_peek_ahead_timers(void) { } | |
1426 | ||
1427 | #endif /* !CONFIG_HIGH_RES_TIMERS */ | |
82f67cd9 | 1428 | |
d3d74453 PZ |
1429 | /* |
1430 | * Called from timer softirq every jiffy, expire hrtimers: | |
1431 | * | |
1432 | * For HRT its the fall back code to run the softirq in the timer | |
1433 | * softirq context in case the hrtimer initialization failed or has | |
1434 | * not been done yet. | |
1435 | */ | |
1436 | void hrtimer_run_pending(void) | |
1437 | { | |
d3d74453 PZ |
1438 | if (hrtimer_hres_active()) |
1439 | return; | |
54cdfdb4 | 1440 | |
d3d74453 PZ |
1441 | /* |
1442 | * This _is_ ugly: We have to check in the softirq context, | |
1443 | * whether we can switch to highres and / or nohz mode. The | |
1444 | * clocksource switch happens in the timer interrupt with | |
1445 | * xtime_lock held. Notification from there only sets the | |
1446 | * check bit in the tick_oneshot code, otherwise we might | |
1447 | * deadlock vs. xtime_lock. | |
1448 | */ | |
1449 | if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) | |
1450 | hrtimer_switch_to_hres(); | |
54cdfdb4 TG |
1451 | } |
1452 | ||
c0a31329 | 1453 | /* |
d3d74453 | 1454 | * Called from hardirq context every jiffy |
c0a31329 | 1455 | */ |
833883d9 | 1456 | void hrtimer_run_queues(void) |
c0a31329 | 1457 | { |
998adc3d | 1458 | struct timerqueue_node *node; |
833883d9 DS |
1459 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
1460 | struct hrtimer_clock_base *base; | |
1461 | int index, gettime = 1; | |
c0a31329 | 1462 | |
833883d9 | 1463 | if (hrtimer_hres_active()) |
3055adda DS |
1464 | return; |
1465 | ||
833883d9 DS |
1466 | for (index = 0; index < HRTIMER_MAX_CLOCK_BASES; index++) { |
1467 | base = &cpu_base->clock_base[index]; | |
b007c389 | 1468 | if (!timerqueue_getnext(&base->active)) |
d3d74453 | 1469 | continue; |
833883d9 | 1470 | |
d7cfb60c | 1471 | if (gettime) { |
833883d9 DS |
1472 | hrtimer_get_softirq_time(cpu_base); |
1473 | gettime = 0; | |
b75f7a51 | 1474 | } |
d3d74453 | 1475 | |
ecb49d1a | 1476 | raw_spin_lock(&cpu_base->lock); |
c0a31329 | 1477 | |
b007c389 | 1478 | while ((node = timerqueue_getnext(&base->active))) { |
833883d9 | 1479 | struct hrtimer *timer; |
54cdfdb4 | 1480 | |
998adc3d | 1481 | timer = container_of(node, struct hrtimer, node); |
cc584b21 AV |
1482 | if (base->softirq_time.tv64 <= |
1483 | hrtimer_get_expires_tv64(timer)) | |
833883d9 DS |
1484 | break; |
1485 | ||
c6a2a177 | 1486 | __run_hrtimer(timer, &base->softirq_time); |
833883d9 | 1487 | } |
ecb49d1a | 1488 | raw_spin_unlock(&cpu_base->lock); |
833883d9 | 1489 | } |
c0a31329 TG |
1490 | } |
1491 | ||
10c94ec1 TG |
1492 | /* |
1493 | * Sleep related functions: | |
1494 | */ | |
c9cb2e3d | 1495 | static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer) |
00362e33 TG |
1496 | { |
1497 | struct hrtimer_sleeper *t = | |
1498 | container_of(timer, struct hrtimer_sleeper, timer); | |
1499 | struct task_struct *task = t->task; | |
1500 | ||
1501 | t->task = NULL; | |
1502 | if (task) | |
1503 | wake_up_process(task); | |
1504 | ||
1505 | return HRTIMER_NORESTART; | |
1506 | } | |
1507 | ||
36c8b586 | 1508 | void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task) |
00362e33 TG |
1509 | { |
1510 | sl->timer.function = hrtimer_wakeup; | |
1511 | sl->task = task; | |
1512 | } | |
2bc481cf | 1513 | EXPORT_SYMBOL_GPL(hrtimer_init_sleeper); |
00362e33 | 1514 | |
669d7868 | 1515 | static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode) |
432569bb | 1516 | { |
669d7868 | 1517 | hrtimer_init_sleeper(t, current); |
10c94ec1 | 1518 | |
432569bb RZ |
1519 | do { |
1520 | set_current_state(TASK_INTERRUPTIBLE); | |
cc584b21 | 1521 | hrtimer_start_expires(&t->timer, mode); |
37bb6cb4 PZ |
1522 | if (!hrtimer_active(&t->timer)) |
1523 | t->task = NULL; | |
432569bb | 1524 | |
54cdfdb4 TG |
1525 | if (likely(t->task)) |
1526 | schedule(); | |
432569bb | 1527 | |
669d7868 | 1528 | hrtimer_cancel(&t->timer); |
c9cb2e3d | 1529 | mode = HRTIMER_MODE_ABS; |
669d7868 TG |
1530 | |
1531 | } while (t->task && !signal_pending(current)); | |
432569bb | 1532 | |
3588a085 PZ |
1533 | __set_current_state(TASK_RUNNING); |
1534 | ||
669d7868 | 1535 | return t->task == NULL; |
10c94ec1 TG |
1536 | } |
1537 | ||
080344b9 ON |
1538 | static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp) |
1539 | { | |
1540 | struct timespec rmt; | |
1541 | ktime_t rem; | |
1542 | ||
cc584b21 | 1543 | rem = hrtimer_expires_remaining(timer); |
080344b9 ON |
1544 | if (rem.tv64 <= 0) |
1545 | return 0; | |
1546 | rmt = ktime_to_timespec(rem); | |
1547 | ||
1548 | if (copy_to_user(rmtp, &rmt, sizeof(*rmtp))) | |
1549 | return -EFAULT; | |
1550 | ||
1551 | return 1; | |
1552 | } | |
1553 | ||
1711ef38 | 1554 | long __sched hrtimer_nanosleep_restart(struct restart_block *restart) |
10c94ec1 | 1555 | { |
669d7868 | 1556 | struct hrtimer_sleeper t; |
080344b9 | 1557 | struct timespec __user *rmtp; |
237fc6e7 | 1558 | int ret = 0; |
10c94ec1 | 1559 | |
ab8177bc | 1560 | hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid, |
237fc6e7 | 1561 | HRTIMER_MODE_ABS); |
cc584b21 | 1562 | hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires); |
10c94ec1 | 1563 | |
c9cb2e3d | 1564 | if (do_nanosleep(&t, HRTIMER_MODE_ABS)) |
237fc6e7 | 1565 | goto out; |
10c94ec1 | 1566 | |
029a07e0 | 1567 | rmtp = restart->nanosleep.rmtp; |
432569bb | 1568 | if (rmtp) { |
237fc6e7 | 1569 | ret = update_rmtp(&t.timer, rmtp); |
080344b9 | 1570 | if (ret <= 0) |
237fc6e7 | 1571 | goto out; |
432569bb | 1572 | } |
10c94ec1 | 1573 | |
10c94ec1 | 1574 | /* The other values in restart are already filled in */ |
237fc6e7 TG |
1575 | ret = -ERESTART_RESTARTBLOCK; |
1576 | out: | |
1577 | destroy_hrtimer_on_stack(&t.timer); | |
1578 | return ret; | |
10c94ec1 TG |
1579 | } |
1580 | ||
080344b9 | 1581 | long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp, |
10c94ec1 TG |
1582 | const enum hrtimer_mode mode, const clockid_t clockid) |
1583 | { | |
1584 | struct restart_block *restart; | |
669d7868 | 1585 | struct hrtimer_sleeper t; |
237fc6e7 | 1586 | int ret = 0; |
3bd01206 AV |
1587 | unsigned long slack; |
1588 | ||
1589 | slack = current->timer_slack_ns; | |
1590 | if (rt_task(current)) | |
1591 | slack = 0; | |
10c94ec1 | 1592 | |
237fc6e7 | 1593 | hrtimer_init_on_stack(&t.timer, clockid, mode); |
3bd01206 | 1594 | hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack); |
432569bb | 1595 | if (do_nanosleep(&t, mode)) |
237fc6e7 | 1596 | goto out; |
10c94ec1 | 1597 | |
7978672c | 1598 | /* Absolute timers do not update the rmtp value and restart: */ |
237fc6e7 TG |
1599 | if (mode == HRTIMER_MODE_ABS) { |
1600 | ret = -ERESTARTNOHAND; | |
1601 | goto out; | |
1602 | } | |
10c94ec1 | 1603 | |
432569bb | 1604 | if (rmtp) { |
237fc6e7 | 1605 | ret = update_rmtp(&t.timer, rmtp); |
080344b9 | 1606 | if (ret <= 0) |
237fc6e7 | 1607 | goto out; |
432569bb | 1608 | } |
10c94ec1 TG |
1609 | |
1610 | restart = ¤t_thread_info()->restart_block; | |
1711ef38 | 1611 | restart->fn = hrtimer_nanosleep_restart; |
ab8177bc | 1612 | restart->nanosleep.clockid = t.timer.base->clockid; |
029a07e0 | 1613 | restart->nanosleep.rmtp = rmtp; |
cc584b21 | 1614 | restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer); |
10c94ec1 | 1615 | |
237fc6e7 TG |
1616 | ret = -ERESTART_RESTARTBLOCK; |
1617 | out: | |
1618 | destroy_hrtimer_on_stack(&t.timer); | |
1619 | return ret; | |
10c94ec1 TG |
1620 | } |
1621 | ||
58fd3aa2 HC |
1622 | SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp, |
1623 | struct timespec __user *, rmtp) | |
6ba1b912 | 1624 | { |
080344b9 | 1625 | struct timespec tu; |
6ba1b912 TG |
1626 | |
1627 | if (copy_from_user(&tu, rqtp, sizeof(tu))) | |
1628 | return -EFAULT; | |
1629 | ||
1630 | if (!timespec_valid(&tu)) | |
1631 | return -EINVAL; | |
1632 | ||
080344b9 | 1633 | return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC); |
6ba1b912 TG |
1634 | } |
1635 | ||
c0a31329 TG |
1636 | /* |
1637 | * Functions related to boot-time initialization: | |
1638 | */ | |
0ec160dd | 1639 | static void __cpuinit init_hrtimers_cpu(int cpu) |
c0a31329 | 1640 | { |
3c8aa39d | 1641 | struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu); |
c0a31329 TG |
1642 | int i; |
1643 | ||
ecb49d1a | 1644 | raw_spin_lock_init(&cpu_base->lock); |
3c8aa39d | 1645 | |
998adc3d | 1646 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
3c8aa39d | 1647 | cpu_base->clock_base[i].cpu_base = cpu_base; |
998adc3d JS |
1648 | timerqueue_init_head(&cpu_base->clock_base[i].active); |
1649 | } | |
3c8aa39d | 1650 | |
54cdfdb4 | 1651 | hrtimer_init_hres(cpu_base); |
c0a31329 TG |
1652 | } |
1653 | ||
1654 | #ifdef CONFIG_HOTPLUG_CPU | |
1655 | ||
ca109491 | 1656 | static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base, |
37810659 | 1657 | struct hrtimer_clock_base *new_base) |
c0a31329 TG |
1658 | { |
1659 | struct hrtimer *timer; | |
998adc3d | 1660 | struct timerqueue_node *node; |
c0a31329 | 1661 | |
998adc3d JS |
1662 | while ((node = timerqueue_getnext(&old_base->active))) { |
1663 | timer = container_of(node, struct hrtimer, node); | |
54cdfdb4 | 1664 | BUG_ON(hrtimer_callback_running(timer)); |
c6a2a177 | 1665 | debug_deactivate(timer); |
b00c1a99 TG |
1666 | |
1667 | /* | |
1668 | * Mark it as STATE_MIGRATE not INACTIVE otherwise the | |
1669 | * timer could be seen as !active and just vanish away | |
1670 | * under us on another CPU | |
1671 | */ | |
1672 | __remove_hrtimer(timer, old_base, HRTIMER_STATE_MIGRATE, 0); | |
c0a31329 | 1673 | timer->base = new_base; |
54cdfdb4 | 1674 | /* |
e3f1d883 TG |
1675 | * Enqueue the timers on the new cpu. This does not |
1676 | * reprogram the event device in case the timer | |
1677 | * expires before the earliest on this CPU, but we run | |
1678 | * hrtimer_interrupt after we migrated everything to | |
1679 | * sort out already expired timers and reprogram the | |
1680 | * event device. | |
54cdfdb4 | 1681 | */ |
a6037b61 | 1682 | enqueue_hrtimer(timer, new_base); |
41e1022e | 1683 | |
b00c1a99 TG |
1684 | /* Clear the migration state bit */ |
1685 | timer->state &= ~HRTIMER_STATE_MIGRATE; | |
c0a31329 TG |
1686 | } |
1687 | } | |
1688 | ||
d5fd43c4 | 1689 | static void migrate_hrtimers(int scpu) |
c0a31329 | 1690 | { |
3c8aa39d | 1691 | struct hrtimer_cpu_base *old_base, *new_base; |
731a55ba | 1692 | int i; |
c0a31329 | 1693 | |
37810659 | 1694 | BUG_ON(cpu_online(scpu)); |
37810659 | 1695 | tick_cancel_sched_timer(scpu); |
731a55ba TG |
1696 | |
1697 | local_irq_disable(); | |
1698 | old_base = &per_cpu(hrtimer_bases, scpu); | |
1699 | new_base = &__get_cpu_var(hrtimer_bases); | |
d82f0b0f ON |
1700 | /* |
1701 | * The caller is globally serialized and nobody else | |
1702 | * takes two locks at once, deadlock is not possible. | |
1703 | */ | |
ecb49d1a TG |
1704 | raw_spin_lock(&new_base->lock); |
1705 | raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING); | |
c0a31329 | 1706 | |
3c8aa39d | 1707 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
ca109491 | 1708 | migrate_hrtimer_list(&old_base->clock_base[i], |
37810659 | 1709 | &new_base->clock_base[i]); |
c0a31329 TG |
1710 | } |
1711 | ||
ecb49d1a TG |
1712 | raw_spin_unlock(&old_base->lock); |
1713 | raw_spin_unlock(&new_base->lock); | |
37810659 | 1714 | |
731a55ba TG |
1715 | /* Check, if we got expired work to do */ |
1716 | __hrtimer_peek_ahead_timers(); | |
1717 | local_irq_enable(); | |
c0a31329 | 1718 | } |
37810659 | 1719 | |
c0a31329 TG |
1720 | #endif /* CONFIG_HOTPLUG_CPU */ |
1721 | ||
8c78f307 | 1722 | static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self, |
c0a31329 TG |
1723 | unsigned long action, void *hcpu) |
1724 | { | |
b2e3c0ad | 1725 | int scpu = (long)hcpu; |
c0a31329 TG |
1726 | |
1727 | switch (action) { | |
1728 | ||
1729 | case CPU_UP_PREPARE: | |
8bb78442 | 1730 | case CPU_UP_PREPARE_FROZEN: |
37810659 | 1731 | init_hrtimers_cpu(scpu); |
c0a31329 TG |
1732 | break; |
1733 | ||
1734 | #ifdef CONFIG_HOTPLUG_CPU | |
94df7de0 SD |
1735 | case CPU_DYING: |
1736 | case CPU_DYING_FROZEN: | |
1737 | clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DYING, &scpu); | |
1738 | break; | |
c0a31329 | 1739 | case CPU_DEAD: |
8bb78442 | 1740 | case CPU_DEAD_FROZEN: |
b2e3c0ad | 1741 | { |
37810659 | 1742 | clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &scpu); |
d5fd43c4 | 1743 | migrate_hrtimers(scpu); |
c0a31329 | 1744 | break; |
b2e3c0ad | 1745 | } |
c0a31329 TG |
1746 | #endif |
1747 | ||
1748 | default: | |
1749 | break; | |
1750 | } | |
1751 | ||
1752 | return NOTIFY_OK; | |
1753 | } | |
1754 | ||
8c78f307 | 1755 | static struct notifier_block __cpuinitdata hrtimers_nb = { |
c0a31329 TG |
1756 | .notifier_call = hrtimer_cpu_notify, |
1757 | }; | |
1758 | ||
1759 | void __init hrtimers_init(void) | |
1760 | { | |
1761 | hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE, | |
1762 | (void *)(long)smp_processor_id()); | |
1763 | register_cpu_notifier(&hrtimers_nb); | |
a6037b61 PZ |
1764 | #ifdef CONFIG_HIGH_RES_TIMERS |
1765 | open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq); | |
1766 | #endif | |
c0a31329 TG |
1767 | } |
1768 | ||
7bb67439 | 1769 | /** |
351b3f7a | 1770 | * schedule_hrtimeout_range_clock - sleep until timeout |
7bb67439 | 1771 | * @expires: timeout value (ktime_t) |
654c8e0b | 1772 | * @delta: slack in expires timeout (ktime_t) |
7bb67439 | 1773 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL |
351b3f7a | 1774 | * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME |
7bb67439 | 1775 | */ |
351b3f7a CE |
1776 | int __sched |
1777 | schedule_hrtimeout_range_clock(ktime_t *expires, unsigned long delta, | |
1778 | const enum hrtimer_mode mode, int clock) | |
7bb67439 AV |
1779 | { |
1780 | struct hrtimer_sleeper t; | |
1781 | ||
1782 | /* | |
1783 | * Optimize when a zero timeout value is given. It does not | |
1784 | * matter whether this is an absolute or a relative time. | |
1785 | */ | |
1786 | if (expires && !expires->tv64) { | |
1787 | __set_current_state(TASK_RUNNING); | |
1788 | return 0; | |
1789 | } | |
1790 | ||
1791 | /* | |
43b21013 | 1792 | * A NULL parameter means "infinite" |
7bb67439 AV |
1793 | */ |
1794 | if (!expires) { | |
1795 | schedule(); | |
1796 | __set_current_state(TASK_RUNNING); | |
1797 | return -EINTR; | |
1798 | } | |
1799 | ||
351b3f7a | 1800 | hrtimer_init_on_stack(&t.timer, clock, mode); |
654c8e0b | 1801 | hrtimer_set_expires_range_ns(&t.timer, *expires, delta); |
7bb67439 AV |
1802 | |
1803 | hrtimer_init_sleeper(&t, current); | |
1804 | ||
cc584b21 | 1805 | hrtimer_start_expires(&t.timer, mode); |
7bb67439 AV |
1806 | if (!hrtimer_active(&t.timer)) |
1807 | t.task = NULL; | |
1808 | ||
1809 | if (likely(t.task)) | |
1810 | schedule(); | |
1811 | ||
1812 | hrtimer_cancel(&t.timer); | |
1813 | destroy_hrtimer_on_stack(&t.timer); | |
1814 | ||
1815 | __set_current_state(TASK_RUNNING); | |
1816 | ||
1817 | return !t.task ? 0 : -EINTR; | |
1818 | } | |
351b3f7a CE |
1819 | |
1820 | /** | |
1821 | * schedule_hrtimeout_range - sleep until timeout | |
1822 | * @expires: timeout value (ktime_t) | |
1823 | * @delta: slack in expires timeout (ktime_t) | |
1824 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL | |
1825 | * | |
1826 | * Make the current task sleep until the given expiry time has | |
1827 | * elapsed. The routine will return immediately unless | |
1828 | * the current task state has been set (see set_current_state()). | |
1829 | * | |
1830 | * The @delta argument gives the kernel the freedom to schedule the | |
1831 | * actual wakeup to a time that is both power and performance friendly. | |
1832 | * The kernel give the normal best effort behavior for "@expires+@delta", | |
1833 | * but may decide to fire the timer earlier, but no earlier than @expires. | |
1834 | * | |
1835 | * You can set the task state as follows - | |
1836 | * | |
1837 | * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to | |
1838 | * pass before the routine returns. | |
1839 | * | |
1840 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
1841 | * delivered to the current task. | |
1842 | * | |
1843 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1844 | * routine returns. | |
1845 | * | |
1846 | * Returns 0 when the timer has expired otherwise -EINTR | |
1847 | */ | |
1848 | int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta, | |
1849 | const enum hrtimer_mode mode) | |
1850 | { | |
1851 | return schedule_hrtimeout_range_clock(expires, delta, mode, | |
1852 | CLOCK_MONOTONIC); | |
1853 | } | |
654c8e0b AV |
1854 | EXPORT_SYMBOL_GPL(schedule_hrtimeout_range); |
1855 | ||
1856 | /** | |
1857 | * schedule_hrtimeout - sleep until timeout | |
1858 | * @expires: timeout value (ktime_t) | |
1859 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL | |
1860 | * | |
1861 | * Make the current task sleep until the given expiry time has | |
1862 | * elapsed. The routine will return immediately unless | |
1863 | * the current task state has been set (see set_current_state()). | |
1864 | * | |
1865 | * You can set the task state as follows - | |
1866 | * | |
1867 | * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to | |
1868 | * pass before the routine returns. | |
1869 | * | |
1870 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
1871 | * delivered to the current task. | |
1872 | * | |
1873 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1874 | * routine returns. | |
1875 | * | |
1876 | * Returns 0 when the timer has expired otherwise -EINTR | |
1877 | */ | |
1878 | int __sched schedule_hrtimeout(ktime_t *expires, | |
1879 | const enum hrtimer_mode mode) | |
1880 | { | |
1881 | return schedule_hrtimeout_range(expires, 0, mode); | |
1882 | } | |
7bb67439 | 1883 | EXPORT_SYMBOL_GPL(schedule_hrtimeout); |