]>
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> |
8bd75c77 | 48 | #include <linux/sched/rt.h> |
aab03e05 | 49 | #include <linux/sched/deadline.h> |
eea08f32 | 50 | #include <linux/timer.h> |
b0f8c44f | 51 | #include <linux/freezer.h> |
c0a31329 TG |
52 | |
53 | #include <asm/uaccess.h> | |
54 | ||
c6a2a177 XG |
55 | #include <trace/events/timer.h> |
56 | ||
c1797baf | 57 | #include "tick-internal.h" |
8b094cd0 | 58 | |
c0a31329 TG |
59 | /* |
60 | * The timer bases: | |
7978672c | 61 | * |
e06383db JS |
62 | * There are more clockids then hrtimer bases. Thus, we index |
63 | * into the timer bases by the hrtimer_base_type enum. When trying | |
64 | * to reach a base using a clockid, hrtimer_clockid_to_base() | |
65 | * is used to convert from clockid to the proper hrtimer_base_type. | |
c0a31329 | 66 | */ |
54cdfdb4 | 67 | DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) = |
c0a31329 | 68 | { |
84cc8fd2 | 69 | .lock = __RAW_SPIN_LOCK_UNLOCKED(hrtimer_bases.lock), |
3c8aa39d | 70 | .clock_base = |
c0a31329 | 71 | { |
3c8aa39d | 72 | { |
ab8177bc TG |
73 | .index = HRTIMER_BASE_MONOTONIC, |
74 | .clockid = CLOCK_MONOTONIC, | |
3c8aa39d | 75 | .get_time = &ktime_get, |
3c8aa39d | 76 | }, |
68fa61c0 TG |
77 | { |
78 | .index = HRTIMER_BASE_REALTIME, | |
79 | .clockid = CLOCK_REALTIME, | |
80 | .get_time = &ktime_get_real, | |
68fa61c0 | 81 | }, |
70a08cca | 82 | { |
ab8177bc TG |
83 | .index = HRTIMER_BASE_BOOTTIME, |
84 | .clockid = CLOCK_BOOTTIME, | |
70a08cca | 85 | .get_time = &ktime_get_boottime, |
70a08cca | 86 | }, |
90adda98 JS |
87 | { |
88 | .index = HRTIMER_BASE_TAI, | |
89 | .clockid = CLOCK_TAI, | |
90 | .get_time = &ktime_get_clocktai, | |
90adda98 | 91 | }, |
3c8aa39d | 92 | } |
c0a31329 TG |
93 | }; |
94 | ||
942c3c5c | 95 | static const int hrtimer_clock_to_base_table[MAX_CLOCKS] = { |
ce31332d TG |
96 | [CLOCK_REALTIME] = HRTIMER_BASE_REALTIME, |
97 | [CLOCK_MONOTONIC] = HRTIMER_BASE_MONOTONIC, | |
98 | [CLOCK_BOOTTIME] = HRTIMER_BASE_BOOTTIME, | |
90adda98 | 99 | [CLOCK_TAI] = HRTIMER_BASE_TAI, |
ce31332d | 100 | }; |
e06383db JS |
101 | |
102 | static inline int hrtimer_clockid_to_base(clockid_t clock_id) | |
103 | { | |
104 | return hrtimer_clock_to_base_table[clock_id]; | |
105 | } | |
106 | ||
c0a31329 TG |
107 | /* |
108 | * Functions and macros which are different for UP/SMP systems are kept in a | |
109 | * single place | |
110 | */ | |
111 | #ifdef CONFIG_SMP | |
112 | ||
c0a31329 TG |
113 | /* |
114 | * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock | |
115 | * means that all timers which are tied to this base via timer->base are | |
116 | * locked, and the base itself is locked too. | |
117 | * | |
118 | * So __run_timers/migrate_timers can safely modify all timers which could | |
119 | * be found on the lists/queues. | |
120 | * | |
121 | * When the timer's base is locked, and the timer removed from list, it is | |
122 | * possible to set timer->base = NULL and drop the lock: the timer remains | |
123 | * locked. | |
124 | */ | |
3c8aa39d TG |
125 | static |
126 | struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer, | |
127 | unsigned long *flags) | |
c0a31329 | 128 | { |
3c8aa39d | 129 | struct hrtimer_clock_base *base; |
c0a31329 TG |
130 | |
131 | for (;;) { | |
132 | base = timer->base; | |
133 | if (likely(base != NULL)) { | |
ecb49d1a | 134 | raw_spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
135 | if (likely(base == timer->base)) |
136 | return base; | |
137 | /* The timer has migrated to another CPU: */ | |
ecb49d1a | 138 | raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags); |
c0a31329 TG |
139 | } |
140 | cpu_relax(); | |
141 | } | |
142 | } | |
143 | ||
6ff7041d TG |
144 | /* |
145 | * With HIGHRES=y we do not migrate the timer when it is expiring | |
146 | * before the next event on the target cpu because we cannot reprogram | |
147 | * the target cpu hardware and we would cause it to fire late. | |
148 | * | |
149 | * Called with cpu_base->lock of target cpu held. | |
150 | */ | |
151 | static int | |
152 | hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base) | |
153 | { | |
154 | #ifdef CONFIG_HIGH_RES_TIMERS | |
155 | ktime_t expires; | |
156 | ||
157 | if (!new_base->cpu_base->hres_active) | |
158 | return 0; | |
159 | ||
160 | expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset); | |
161 | return expires.tv64 <= new_base->cpu_base->expires_next.tv64; | |
162 | #else | |
163 | return 0; | |
164 | #endif | |
165 | } | |
166 | ||
c0a31329 TG |
167 | /* |
168 | * Switch the timer base to the current CPU when possible. | |
169 | */ | |
3c8aa39d | 170 | static inline struct hrtimer_clock_base * |
597d0275 AB |
171 | switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base, |
172 | int pinned) | |
c0a31329 | 173 | { |
3c8aa39d TG |
174 | struct hrtimer_clock_base *new_base; |
175 | struct hrtimer_cpu_base *new_cpu_base; | |
6ff7041d | 176 | int this_cpu = smp_processor_id(); |
6201b4d6 | 177 | int cpu = get_nohz_timer_target(pinned); |
ab8177bc | 178 | int basenum = base->index; |
c0a31329 | 179 | |
eea08f32 AB |
180 | again: |
181 | new_cpu_base = &per_cpu(hrtimer_bases, cpu); | |
e06383db | 182 | new_base = &new_cpu_base->clock_base[basenum]; |
c0a31329 TG |
183 | |
184 | if (base != new_base) { | |
185 | /* | |
6ff7041d | 186 | * We are trying to move timer to new_base. |
c0a31329 TG |
187 | * However we can't change timer's base while it is running, |
188 | * so we keep it on the same CPU. No hassle vs. reprogramming | |
189 | * the event source in the high resolution case. The softirq | |
190 | * code will take care of this when the timer function has | |
191 | * completed. There is no conflict as we hold the lock until | |
192 | * the timer is enqueued. | |
193 | */ | |
54cdfdb4 | 194 | if (unlikely(hrtimer_callback_running(timer))) |
c0a31329 TG |
195 | return base; |
196 | ||
197 | /* See the comment in lock_timer_base() */ | |
198 | timer->base = NULL; | |
ecb49d1a TG |
199 | raw_spin_unlock(&base->cpu_base->lock); |
200 | raw_spin_lock(&new_base->cpu_base->lock); | |
eea08f32 | 201 | |
6ff7041d TG |
202 | if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) { |
203 | cpu = this_cpu; | |
ecb49d1a TG |
204 | raw_spin_unlock(&new_base->cpu_base->lock); |
205 | raw_spin_lock(&base->cpu_base->lock); | |
6ff7041d TG |
206 | timer->base = base; |
207 | goto again; | |
eea08f32 | 208 | } |
c0a31329 | 209 | timer->base = new_base; |
012a45e3 LM |
210 | } else { |
211 | if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) { | |
212 | cpu = this_cpu; | |
213 | goto again; | |
214 | } | |
c0a31329 TG |
215 | } |
216 | return new_base; | |
217 | } | |
218 | ||
219 | #else /* CONFIG_SMP */ | |
220 | ||
3c8aa39d | 221 | static inline struct hrtimer_clock_base * |
c0a31329 TG |
222 | lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) |
223 | { | |
3c8aa39d | 224 | struct hrtimer_clock_base *base = timer->base; |
c0a31329 | 225 | |
ecb49d1a | 226 | raw_spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
227 | |
228 | return base; | |
229 | } | |
230 | ||
eea08f32 | 231 | # define switch_hrtimer_base(t, b, p) (b) |
c0a31329 TG |
232 | |
233 | #endif /* !CONFIG_SMP */ | |
234 | ||
235 | /* | |
236 | * Functions for the union type storage format of ktime_t which are | |
237 | * too large for inlining: | |
238 | */ | |
239 | #if BITS_PER_LONG < 64 | |
c0a31329 TG |
240 | /* |
241 | * Divide a ktime value by a nanosecond value | |
242 | */ | |
8b618628 | 243 | u64 __ktime_divns(const ktime_t kt, s64 div) |
c0a31329 | 244 | { |
900cfa46 | 245 | u64 dclc; |
c0a31329 TG |
246 | int sft = 0; |
247 | ||
900cfa46 | 248 | dclc = ktime_to_ns(kt); |
c0a31329 TG |
249 | /* Make sure the divisor is less than 2^32: */ |
250 | while (div >> 32) { | |
251 | sft++; | |
252 | div >>= 1; | |
253 | } | |
254 | dclc >>= sft; | |
255 | do_div(dclc, (unsigned long) div); | |
256 | ||
4d672e7a | 257 | return dclc; |
c0a31329 | 258 | } |
8b618628 | 259 | EXPORT_SYMBOL_GPL(__ktime_divns); |
c0a31329 TG |
260 | #endif /* BITS_PER_LONG >= 64 */ |
261 | ||
5a7780e7 TG |
262 | /* |
263 | * Add two ktime values and do a safety check for overflow: | |
264 | */ | |
265 | ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs) | |
266 | { | |
267 | ktime_t res = ktime_add(lhs, rhs); | |
268 | ||
269 | /* | |
270 | * We use KTIME_SEC_MAX here, the maximum timeout which we can | |
271 | * return to user space in a timespec: | |
272 | */ | |
273 | if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64) | |
274 | res = ktime_set(KTIME_SEC_MAX, 0); | |
275 | ||
276 | return res; | |
277 | } | |
278 | ||
8daa21e6 AB |
279 | EXPORT_SYMBOL_GPL(ktime_add_safe); |
280 | ||
237fc6e7 TG |
281 | #ifdef CONFIG_DEBUG_OBJECTS_TIMERS |
282 | ||
283 | static struct debug_obj_descr hrtimer_debug_descr; | |
284 | ||
99777288 SG |
285 | static void *hrtimer_debug_hint(void *addr) |
286 | { | |
287 | return ((struct hrtimer *) addr)->function; | |
288 | } | |
289 | ||
237fc6e7 TG |
290 | /* |
291 | * fixup_init is called when: | |
292 | * - an active object is initialized | |
293 | */ | |
294 | static int hrtimer_fixup_init(void *addr, enum debug_obj_state state) | |
295 | { | |
296 | struct hrtimer *timer = addr; | |
297 | ||
298 | switch (state) { | |
299 | case ODEBUG_STATE_ACTIVE: | |
300 | hrtimer_cancel(timer); | |
301 | debug_object_init(timer, &hrtimer_debug_descr); | |
302 | return 1; | |
303 | default: | |
304 | return 0; | |
305 | } | |
306 | } | |
307 | ||
308 | /* | |
309 | * fixup_activate is called when: | |
310 | * - an active object is activated | |
311 | * - an unknown object is activated (might be a statically initialized object) | |
312 | */ | |
313 | static int hrtimer_fixup_activate(void *addr, enum debug_obj_state state) | |
314 | { | |
315 | switch (state) { | |
316 | ||
317 | case ODEBUG_STATE_NOTAVAILABLE: | |
318 | WARN_ON_ONCE(1); | |
319 | return 0; | |
320 | ||
321 | case ODEBUG_STATE_ACTIVE: | |
322 | WARN_ON(1); | |
323 | ||
324 | default: | |
325 | return 0; | |
326 | } | |
327 | } | |
328 | ||
329 | /* | |
330 | * fixup_free is called when: | |
331 | * - an active object is freed | |
332 | */ | |
333 | static int hrtimer_fixup_free(void *addr, enum debug_obj_state state) | |
334 | { | |
335 | struct hrtimer *timer = addr; | |
336 | ||
337 | switch (state) { | |
338 | case ODEBUG_STATE_ACTIVE: | |
339 | hrtimer_cancel(timer); | |
340 | debug_object_free(timer, &hrtimer_debug_descr); | |
341 | return 1; | |
342 | default: | |
343 | return 0; | |
344 | } | |
345 | } | |
346 | ||
347 | static struct debug_obj_descr hrtimer_debug_descr = { | |
348 | .name = "hrtimer", | |
99777288 | 349 | .debug_hint = hrtimer_debug_hint, |
237fc6e7 TG |
350 | .fixup_init = hrtimer_fixup_init, |
351 | .fixup_activate = hrtimer_fixup_activate, | |
352 | .fixup_free = hrtimer_fixup_free, | |
353 | }; | |
354 | ||
355 | static inline void debug_hrtimer_init(struct hrtimer *timer) | |
356 | { | |
357 | debug_object_init(timer, &hrtimer_debug_descr); | |
358 | } | |
359 | ||
360 | static inline void debug_hrtimer_activate(struct hrtimer *timer) | |
361 | { | |
362 | debug_object_activate(timer, &hrtimer_debug_descr); | |
363 | } | |
364 | ||
365 | static inline void debug_hrtimer_deactivate(struct hrtimer *timer) | |
366 | { | |
367 | debug_object_deactivate(timer, &hrtimer_debug_descr); | |
368 | } | |
369 | ||
370 | static inline void debug_hrtimer_free(struct hrtimer *timer) | |
371 | { | |
372 | debug_object_free(timer, &hrtimer_debug_descr); | |
373 | } | |
374 | ||
375 | static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id, | |
376 | enum hrtimer_mode mode); | |
377 | ||
378 | void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id, | |
379 | enum hrtimer_mode mode) | |
380 | { | |
381 | debug_object_init_on_stack(timer, &hrtimer_debug_descr); | |
382 | __hrtimer_init(timer, clock_id, mode); | |
383 | } | |
2bc481cf | 384 | EXPORT_SYMBOL_GPL(hrtimer_init_on_stack); |
237fc6e7 TG |
385 | |
386 | void destroy_hrtimer_on_stack(struct hrtimer *timer) | |
387 | { | |
388 | debug_object_free(timer, &hrtimer_debug_descr); | |
389 | } | |
390 | ||
391 | #else | |
392 | static inline void debug_hrtimer_init(struct hrtimer *timer) { } | |
393 | static inline void debug_hrtimer_activate(struct hrtimer *timer) { } | |
394 | static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { } | |
395 | #endif | |
396 | ||
c6a2a177 XG |
397 | static inline void |
398 | debug_init(struct hrtimer *timer, clockid_t clockid, | |
399 | enum hrtimer_mode mode) | |
400 | { | |
401 | debug_hrtimer_init(timer); | |
402 | trace_hrtimer_init(timer, clockid, mode); | |
403 | } | |
404 | ||
405 | static inline void debug_activate(struct hrtimer *timer) | |
406 | { | |
407 | debug_hrtimer_activate(timer); | |
408 | trace_hrtimer_start(timer); | |
409 | } | |
410 | ||
411 | static inline void debug_deactivate(struct hrtimer *timer) | |
412 | { | |
413 | debug_hrtimer_deactivate(timer); | |
414 | trace_hrtimer_cancel(timer); | |
415 | } | |
416 | ||
9bc74919 | 417 | #if defined(CONFIG_NO_HZ_COMMON) || defined(CONFIG_HIGH_RES_TIMERS) |
4ebbda52 | 418 | static ktime_t __hrtimer_get_next_event(struct hrtimer_cpu_base *cpu_base) |
9bc74919 TG |
419 | { |
420 | struct hrtimer_clock_base *base = cpu_base->clock_base; | |
421 | ktime_t expires, expires_next = { .tv64 = KTIME_MAX }; | |
422 | int i; | |
423 | ||
424 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) { | |
425 | struct timerqueue_node *next; | |
426 | struct hrtimer *timer; | |
427 | ||
428 | next = timerqueue_getnext(&base->active); | |
429 | if (!next) | |
430 | continue; | |
431 | ||
432 | timer = container_of(next, struct hrtimer, node); | |
433 | expires = ktime_sub(hrtimer_get_expires(timer), base->offset); | |
434 | if (expires.tv64 < expires_next.tv64) | |
435 | expires_next = expires; | |
436 | } | |
437 | /* | |
438 | * clock_was_set() might have changed base->offset of any of | |
439 | * the clock bases so the result might be negative. Fix it up | |
440 | * to prevent a false positive in clockevents_program_event(). | |
441 | */ | |
442 | if (expires_next.tv64 < 0) | |
443 | expires_next.tv64 = 0; | |
444 | return expires_next; | |
445 | } | |
446 | #endif | |
447 | ||
21d6d52a TG |
448 | static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base) |
449 | { | |
450 | ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset; | |
451 | ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset; | |
452 | ktime_t *offs_tai = &base->clock_base[HRTIMER_BASE_TAI].offset; | |
453 | ||
868a3e91 TG |
454 | return ktime_get_update_offsets_now(&base->clock_was_set_seq, |
455 | offs_real, offs_boot, offs_tai); | |
21d6d52a TG |
456 | } |
457 | ||
54cdfdb4 TG |
458 | /* High resolution timer related functions */ |
459 | #ifdef CONFIG_HIGH_RES_TIMERS | |
460 | ||
461 | /* | |
462 | * High resolution timer enabled ? | |
463 | */ | |
464 | static int hrtimer_hres_enabled __read_mostly = 1; | |
398ca17f TG |
465 | unsigned int hrtimer_resolution __read_mostly = LOW_RES_NSEC; |
466 | EXPORT_SYMBOL_GPL(hrtimer_resolution); | |
54cdfdb4 TG |
467 | |
468 | /* | |
469 | * Enable / Disable high resolution mode | |
470 | */ | |
471 | static int __init setup_hrtimer_hres(char *str) | |
472 | { | |
473 | if (!strcmp(str, "off")) | |
474 | hrtimer_hres_enabled = 0; | |
475 | else if (!strcmp(str, "on")) | |
476 | hrtimer_hres_enabled = 1; | |
477 | else | |
478 | return 0; | |
479 | return 1; | |
480 | } | |
481 | ||
482 | __setup("highres=", setup_hrtimer_hres); | |
483 | ||
484 | /* | |
485 | * hrtimer_high_res_enabled - query, if the highres mode is enabled | |
486 | */ | |
487 | static inline int hrtimer_is_hres_enabled(void) | |
488 | { | |
489 | return hrtimer_hres_enabled; | |
490 | } | |
491 | ||
492 | /* | |
493 | * Is the high resolution mode active ? | |
494 | */ | |
e19ffe8b TG |
495 | static inline int __hrtimer_hres_active(struct hrtimer_cpu_base *cpu_base) |
496 | { | |
497 | return cpu_base->hres_active; | |
498 | } | |
499 | ||
54cdfdb4 TG |
500 | static inline int hrtimer_hres_active(void) |
501 | { | |
e19ffe8b | 502 | return __hrtimer_hres_active(this_cpu_ptr(&hrtimer_bases)); |
54cdfdb4 TG |
503 | } |
504 | ||
505 | /* | |
506 | * Reprogram the event source with checking both queues for the | |
507 | * next event | |
508 | * Called with interrupts disabled and base->lock held | |
509 | */ | |
7403f41f AC |
510 | static void |
511 | hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal) | |
54cdfdb4 | 512 | { |
21d6d52a TG |
513 | ktime_t expires_next; |
514 | ||
515 | if (!cpu_base->hres_active) | |
516 | return; | |
517 | ||
518 | expires_next = __hrtimer_get_next_event(cpu_base); | |
54cdfdb4 | 519 | |
7403f41f AC |
520 | if (skip_equal && expires_next.tv64 == cpu_base->expires_next.tv64) |
521 | return; | |
522 | ||
523 | cpu_base->expires_next.tv64 = expires_next.tv64; | |
524 | ||
6c6c0d5a SH |
525 | /* |
526 | * If a hang was detected in the last timer interrupt then we | |
527 | * leave the hang delay active in the hardware. We want the | |
528 | * system to make progress. That also prevents the following | |
529 | * scenario: | |
530 | * T1 expires 50ms from now | |
531 | * T2 expires 5s from now | |
532 | * | |
533 | * T1 is removed, so this code is called and would reprogram | |
534 | * the hardware to 5s from now. Any hrtimer_start after that | |
535 | * will not reprogram the hardware due to hang_detected being | |
536 | * set. So we'd effectivly block all timers until the T2 event | |
537 | * fires. | |
538 | */ | |
539 | if (cpu_base->hang_detected) | |
540 | return; | |
541 | ||
54cdfdb4 TG |
542 | if (cpu_base->expires_next.tv64 != KTIME_MAX) |
543 | tick_program_event(cpu_base->expires_next, 1); | |
544 | } | |
545 | ||
546 | /* | |
547 | * Shared reprogramming for clock_realtime and clock_monotonic | |
548 | * | |
549 | * When a timer is enqueued and expires earlier than the already enqueued | |
550 | * timers, we have to check, whether it expires earlier than the timer for | |
551 | * which the clock event device was armed. | |
552 | * | |
9e1e01dd VK |
553 | * Note, that in case the state has HRTIMER_STATE_CALLBACK set, no reprogramming |
554 | * and no expiry check happens. The timer gets enqueued into the rbtree. The | |
555 | * reprogramming and expiry check is done in the hrtimer_interrupt or in the | |
556 | * softirq. | |
557 | * | |
54cdfdb4 TG |
558 | * Called with interrupts disabled and base->cpu_base.lock held |
559 | */ | |
560 | static int hrtimer_reprogram(struct hrtimer *timer, | |
561 | struct hrtimer_clock_base *base) | |
562 | { | |
dc5df73b | 563 | struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases); |
cc584b21 | 564 | ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset); |
54cdfdb4 TG |
565 | int res; |
566 | ||
cc584b21 | 567 | WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0); |
63070a79 | 568 | |
54cdfdb4 TG |
569 | /* |
570 | * When the callback is running, we do not reprogram the clock event | |
571 | * device. The timer callback is either running on a different CPU or | |
3a4fa0a2 | 572 | * the callback is executed in the hrtimer_interrupt context. The |
54cdfdb4 TG |
573 | * reprogramming is handled either by the softirq, which called the |
574 | * callback or at the end of the hrtimer_interrupt. | |
575 | */ | |
576 | if (hrtimer_callback_running(timer)) | |
577 | return 0; | |
578 | ||
63070a79 TG |
579 | /* |
580 | * CLOCK_REALTIME timer might be requested with an absolute | |
581 | * expiry time which is less than base->offset. Nothing wrong | |
582 | * about that, just avoid to call into the tick code, which | |
583 | * has now objections against negative expiry values. | |
584 | */ | |
585 | if (expires.tv64 < 0) | |
586 | return -ETIME; | |
587 | ||
41d2e494 TG |
588 | if (expires.tv64 >= cpu_base->expires_next.tv64) |
589 | return 0; | |
590 | ||
9bc74919 TG |
591 | /* |
592 | * When the target cpu of the timer is currently executing | |
593 | * hrtimer_interrupt(), then we do not touch the clock event | |
594 | * device. hrtimer_interrupt() will reevaluate all clock bases | |
595 | * before reprogramming the device. | |
596 | */ | |
597 | if (cpu_base->in_hrtirq) | |
598 | return 0; | |
599 | ||
41d2e494 TG |
600 | /* |
601 | * If a hang was detected in the last timer interrupt then we | |
602 | * do not schedule a timer which is earlier than the expiry | |
603 | * which we enforced in the hang detection. We want the system | |
604 | * to make progress. | |
605 | */ | |
606 | if (cpu_base->hang_detected) | |
54cdfdb4 TG |
607 | return 0; |
608 | ||
609 | /* | |
610 | * Clockevents returns -ETIME, when the event was in the past. | |
611 | */ | |
612 | res = tick_program_event(expires, 0); | |
613 | if (!IS_ERR_VALUE(res)) | |
41d2e494 | 614 | cpu_base->expires_next = expires; |
54cdfdb4 TG |
615 | return res; |
616 | } | |
617 | ||
54cdfdb4 TG |
618 | /* |
619 | * Initialize the high resolution related parts of cpu_base | |
620 | */ | |
621 | static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) | |
622 | { | |
623 | base->expires_next.tv64 = KTIME_MAX; | |
624 | base->hres_active = 0; | |
54cdfdb4 TG |
625 | } |
626 | ||
9ec26907 TG |
627 | /* |
628 | * Retrigger next event is called after clock was set | |
629 | * | |
630 | * Called with interrupts disabled via on_each_cpu() | |
631 | */ | |
632 | static void retrigger_next_event(void *arg) | |
633 | { | |
dc5df73b | 634 | struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases); |
9ec26907 | 635 | |
e19ffe8b | 636 | if (!base->hres_active) |
9ec26907 TG |
637 | return; |
638 | ||
9ec26907 | 639 | raw_spin_lock(&base->lock); |
5baefd6d | 640 | hrtimer_update_base(base); |
9ec26907 TG |
641 | hrtimer_force_reprogram(base, 0); |
642 | raw_spin_unlock(&base->lock); | |
643 | } | |
b12a03ce | 644 | |
54cdfdb4 TG |
645 | /* |
646 | * Switch to high resolution mode | |
647 | */ | |
f8953856 | 648 | static int hrtimer_switch_to_hres(void) |
54cdfdb4 | 649 | { |
398ca17f | 650 | int cpu = smp_processor_id(); |
820de5c3 | 651 | struct hrtimer_cpu_base *base = &per_cpu(hrtimer_bases, cpu); |
54cdfdb4 TG |
652 | unsigned long flags; |
653 | ||
654 | if (base->hres_active) | |
f8953856 | 655 | return 1; |
54cdfdb4 TG |
656 | |
657 | local_irq_save(flags); | |
658 | ||
659 | if (tick_init_highres()) { | |
660 | local_irq_restore(flags); | |
820de5c3 IM |
661 | printk(KERN_WARNING "Could not switch to high resolution " |
662 | "mode on CPU %d\n", cpu); | |
f8953856 | 663 | return 0; |
54cdfdb4 TG |
664 | } |
665 | base->hres_active = 1; | |
398ca17f | 666 | hrtimer_resolution = HIGH_RES_NSEC; |
54cdfdb4 TG |
667 | |
668 | tick_setup_sched_timer(); | |
54cdfdb4 TG |
669 | /* "Retrigger" the interrupt to get things going */ |
670 | retrigger_next_event(NULL); | |
671 | local_irq_restore(flags); | |
f8953856 | 672 | return 1; |
54cdfdb4 TG |
673 | } |
674 | ||
5ec2481b TG |
675 | static void clock_was_set_work(struct work_struct *work) |
676 | { | |
677 | clock_was_set(); | |
678 | } | |
679 | ||
680 | static DECLARE_WORK(hrtimer_work, clock_was_set_work); | |
681 | ||
f55a6faa | 682 | /* |
5ec2481b TG |
683 | * Called from timekeeping and resume code to reprogramm the hrtimer |
684 | * interrupt device on all cpus. | |
f55a6faa JS |
685 | */ |
686 | void clock_was_set_delayed(void) | |
687 | { | |
5ec2481b | 688 | schedule_work(&hrtimer_work); |
f55a6faa JS |
689 | } |
690 | ||
54cdfdb4 TG |
691 | #else |
692 | ||
e19ffe8b | 693 | static inline int __hrtimer_hres_active(struct hrtimer_cpu_base *b) { return 0; } |
54cdfdb4 TG |
694 | static inline int hrtimer_hres_active(void) { return 0; } |
695 | static inline int hrtimer_is_hres_enabled(void) { return 0; } | |
f8953856 | 696 | static inline int hrtimer_switch_to_hres(void) { return 0; } |
7403f41f AC |
697 | static inline void |
698 | hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { } | |
9e1e01dd VK |
699 | static inline int hrtimer_reprogram(struct hrtimer *timer, |
700 | struct hrtimer_clock_base *base) | |
54cdfdb4 TG |
701 | { |
702 | return 0; | |
703 | } | |
54cdfdb4 | 704 | static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { } |
9ec26907 | 705 | static inline void retrigger_next_event(void *arg) { } |
54cdfdb4 TG |
706 | |
707 | #endif /* CONFIG_HIGH_RES_TIMERS */ | |
708 | ||
b12a03ce TG |
709 | /* |
710 | * Clock realtime was set | |
711 | * | |
712 | * Change the offset of the realtime clock vs. the monotonic | |
713 | * clock. | |
714 | * | |
715 | * We might have to reprogram the high resolution timer interrupt. On | |
716 | * SMP we call the architecture specific code to retrigger _all_ high | |
717 | * resolution timer interrupts. On UP we just disable interrupts and | |
718 | * call the high resolution interrupt code. | |
719 | */ | |
720 | void clock_was_set(void) | |
721 | { | |
90ff1f30 | 722 | #ifdef CONFIG_HIGH_RES_TIMERS |
b12a03ce TG |
723 | /* Retrigger the CPU local events everywhere */ |
724 | on_each_cpu(retrigger_next_event, NULL, 1); | |
9ec26907 TG |
725 | #endif |
726 | timerfd_clock_was_set(); | |
b12a03ce TG |
727 | } |
728 | ||
729 | /* | |
730 | * During resume we might have to reprogram the high resolution timer | |
7c4c3a0f DV |
731 | * interrupt on all online CPUs. However, all other CPUs will be |
732 | * stopped with IRQs interrupts disabled so the clock_was_set() call | |
5ec2481b | 733 | * must be deferred. |
b12a03ce TG |
734 | */ |
735 | void hrtimers_resume(void) | |
736 | { | |
737 | WARN_ONCE(!irqs_disabled(), | |
738 | KERN_INFO "hrtimers_resume() called with IRQs enabled!"); | |
739 | ||
5ec2481b | 740 | /* Retrigger on the local CPU */ |
b12a03ce | 741 | retrigger_next_event(NULL); |
5ec2481b TG |
742 | /* And schedule a retrigger for all others */ |
743 | clock_was_set_delayed(); | |
b12a03ce TG |
744 | } |
745 | ||
5f201907 | 746 | static inline void timer_stats_hrtimer_set_start_info(struct hrtimer *timer) |
82f67cd9 | 747 | { |
5f201907 | 748 | #ifdef CONFIG_TIMER_STATS |
82f67cd9 IM |
749 | if (timer->start_site) |
750 | return; | |
5f201907 | 751 | timer->start_site = __builtin_return_address(0); |
82f67cd9 IM |
752 | memcpy(timer->start_comm, current->comm, TASK_COMM_LEN); |
753 | timer->start_pid = current->pid; | |
5f201907 HC |
754 | #endif |
755 | } | |
756 | ||
757 | static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer *timer) | |
758 | { | |
759 | #ifdef CONFIG_TIMER_STATS | |
760 | timer->start_site = NULL; | |
761 | #endif | |
82f67cd9 | 762 | } |
5f201907 HC |
763 | |
764 | static inline void timer_stats_account_hrtimer(struct hrtimer *timer) | |
765 | { | |
766 | #ifdef CONFIG_TIMER_STATS | |
767 | if (likely(!timer_stats_active)) | |
768 | return; | |
769 | timer_stats_update_stats(timer, timer->start_pid, timer->start_site, | |
770 | timer->function, timer->start_comm, 0); | |
82f67cd9 | 771 | #endif |
5f201907 | 772 | } |
82f67cd9 | 773 | |
c0a31329 | 774 | /* |
6506f2aa | 775 | * Counterpart to lock_hrtimer_base above: |
c0a31329 TG |
776 | */ |
777 | static inline | |
778 | void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) | |
779 | { | |
ecb49d1a | 780 | raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags); |
c0a31329 TG |
781 | } |
782 | ||
783 | /** | |
784 | * hrtimer_forward - forward the timer expiry | |
c0a31329 | 785 | * @timer: hrtimer to forward |
44f21475 | 786 | * @now: forward past this time |
c0a31329 TG |
787 | * @interval: the interval to forward |
788 | * | |
789 | * Forward the timer expiry so it will expire in the future. | |
8dca6f33 | 790 | * Returns the number of overruns. |
91e5a217 TG |
791 | * |
792 | * Can be safely called from the callback function of @timer. If | |
793 | * called from other contexts @timer must neither be enqueued nor | |
794 | * running the callback and the caller needs to take care of | |
795 | * serialization. | |
796 | * | |
797 | * Note: This only updates the timer expiry value and does not requeue | |
798 | * the timer. | |
c0a31329 | 799 | */ |
4d672e7a | 800 | u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval) |
c0a31329 | 801 | { |
4d672e7a | 802 | u64 orun = 1; |
44f21475 | 803 | ktime_t delta; |
c0a31329 | 804 | |
cc584b21 | 805 | delta = ktime_sub(now, hrtimer_get_expires(timer)); |
c0a31329 TG |
806 | |
807 | if (delta.tv64 < 0) | |
808 | return 0; | |
809 | ||
398ca17f TG |
810 | if (interval.tv64 < hrtimer_resolution) |
811 | interval.tv64 = hrtimer_resolution; | |
c9db4fa1 | 812 | |
c0a31329 | 813 | if (unlikely(delta.tv64 >= interval.tv64)) { |
df869b63 | 814 | s64 incr = ktime_to_ns(interval); |
c0a31329 TG |
815 | |
816 | orun = ktime_divns(delta, incr); | |
cc584b21 AV |
817 | hrtimer_add_expires_ns(timer, incr * orun); |
818 | if (hrtimer_get_expires_tv64(timer) > now.tv64) | |
c0a31329 TG |
819 | return orun; |
820 | /* | |
821 | * This (and the ktime_add() below) is the | |
822 | * correction for exact: | |
823 | */ | |
824 | orun++; | |
825 | } | |
cc584b21 | 826 | hrtimer_add_expires(timer, interval); |
c0a31329 TG |
827 | |
828 | return orun; | |
829 | } | |
6bdb6b62 | 830 | EXPORT_SYMBOL_GPL(hrtimer_forward); |
c0a31329 TG |
831 | |
832 | /* | |
833 | * enqueue_hrtimer - internal function to (re)start a timer | |
834 | * | |
835 | * The timer is inserted in expiry order. Insertion into the | |
836 | * red black tree is O(log(n)). Must hold the base lock. | |
a6037b61 PZ |
837 | * |
838 | * Returns 1 when the new timer is the leftmost timer in the tree. | |
c0a31329 | 839 | */ |
a6037b61 PZ |
840 | static int enqueue_hrtimer(struct hrtimer *timer, |
841 | struct hrtimer_clock_base *base) | |
c0a31329 | 842 | { |
c6a2a177 | 843 | debug_activate(timer); |
237fc6e7 | 844 | |
998adc3d | 845 | timerqueue_add(&base->active, &timer->node); |
ab8177bc | 846 | base->cpu_base->active_bases |= 1 << base->index; |
54cdfdb4 | 847 | |
303e967f TG |
848 | /* |
849 | * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the | |
850 | * state of a possibly running callback. | |
851 | */ | |
852 | timer->state |= HRTIMER_STATE_ENQUEUED; | |
a6037b61 | 853 | |
998adc3d | 854 | return (&timer->node == base->active.next); |
288867ec | 855 | } |
c0a31329 TG |
856 | |
857 | /* | |
858 | * __remove_hrtimer - internal function to remove a timer | |
859 | * | |
860 | * Caller must hold the base lock. | |
54cdfdb4 TG |
861 | * |
862 | * High resolution timer mode reprograms the clock event device when the | |
863 | * timer is the one which expires next. The caller can disable this by setting | |
864 | * reprogram to zero. This is useful, when the context does a reprogramming | |
865 | * anyway (e.g. timer interrupt) | |
c0a31329 | 866 | */ |
3c8aa39d | 867 | static void __remove_hrtimer(struct hrtimer *timer, |
303e967f | 868 | struct hrtimer_clock_base *base, |
54cdfdb4 | 869 | unsigned long newstate, int reprogram) |
c0a31329 | 870 | { |
e19ffe8b | 871 | struct hrtimer_cpu_base *cpu_base = base->cpu_base; |
27c9cd7e | 872 | struct timerqueue_node *next_timer; |
e19ffe8b | 873 | |
7403f41f AC |
874 | if (!(timer->state & HRTIMER_STATE_ENQUEUED)) |
875 | goto out; | |
876 | ||
27c9cd7e JO |
877 | next_timer = timerqueue_getnext(&base->active); |
878 | timerqueue_del(&base->active, &timer->node); | |
d9f0acde | 879 | if (!timerqueue_getnext(&base->active)) |
e19ffe8b | 880 | cpu_base->active_bases &= ~(1 << base->index); |
d9f0acde | 881 | |
27c9cd7e | 882 | if (&timer->node == next_timer) { |
7403f41f AC |
883 | #ifdef CONFIG_HIGH_RES_TIMERS |
884 | /* Reprogram the clock event device. if enabled */ | |
e19ffe8b | 885 | if (reprogram && cpu_base->hres_active) { |
7403f41f AC |
886 | ktime_t expires; |
887 | ||
888 | expires = ktime_sub(hrtimer_get_expires(timer), | |
889 | base->offset); | |
e19ffe8b TG |
890 | if (cpu_base->expires_next.tv64 == expires.tv64) |
891 | hrtimer_force_reprogram(cpu_base, 1); | |
54cdfdb4 | 892 | } |
7403f41f | 893 | #endif |
54cdfdb4 | 894 | } |
7403f41f | 895 | out: |
303e967f | 896 | timer->state = newstate; |
c0a31329 TG |
897 | } |
898 | ||
899 | /* | |
900 | * remove hrtimer, called with base lock held | |
901 | */ | |
902 | static inline int | |
3c8aa39d | 903 | remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base) |
c0a31329 | 904 | { |
303e967f | 905 | if (hrtimer_is_queued(timer)) { |
f13d4f97 | 906 | unsigned long state; |
54cdfdb4 TG |
907 | int reprogram; |
908 | ||
909 | /* | |
910 | * Remove the timer and force reprogramming when high | |
911 | * resolution mode is active and the timer is on the current | |
912 | * CPU. If we remove a timer on another CPU, reprogramming is | |
913 | * skipped. The interrupt event on this CPU is fired and | |
914 | * reprogramming happens in the interrupt handler. This is a | |
915 | * rare case and less expensive than a smp call. | |
916 | */ | |
c6a2a177 | 917 | debug_deactivate(timer); |
82f67cd9 | 918 | timer_stats_hrtimer_clear_start_info(timer); |
dc5df73b | 919 | reprogram = base->cpu_base == this_cpu_ptr(&hrtimer_bases); |
f13d4f97 SQ |
920 | /* |
921 | * We must preserve the CALLBACK state flag here, | |
922 | * otherwise we could move the timer base in | |
923 | * switch_hrtimer_base. | |
924 | */ | |
925 | state = timer->state & HRTIMER_STATE_CALLBACK; | |
926 | __remove_hrtimer(timer, base, state, reprogram); | |
c0a31329 TG |
927 | return 1; |
928 | } | |
929 | return 0; | |
930 | } | |
931 | ||
7f1e2ca9 PZ |
932 | int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, |
933 | unsigned long delta_ns, const enum hrtimer_mode mode, | |
934 | int wakeup) | |
c0a31329 | 935 | { |
3c8aa39d | 936 | struct hrtimer_clock_base *base, *new_base; |
c0a31329 | 937 | unsigned long flags; |
a6037b61 | 938 | int ret, leftmost; |
c0a31329 TG |
939 | |
940 | base = lock_hrtimer_base(timer, &flags); | |
941 | ||
942 | /* Remove an active timer from the queue: */ | |
943 | ret = remove_hrtimer(timer, base); | |
944 | ||
597d0275 | 945 | if (mode & HRTIMER_MODE_REL) { |
84ea7fe3 | 946 | tim = ktime_add_safe(tim, base->get_time()); |
06027bdd IM |
947 | /* |
948 | * CONFIG_TIME_LOW_RES is a temporary way for architectures | |
949 | * to signal that they simply return xtime in | |
950 | * do_gettimeoffset(). In this case we want to round up by | |
951 | * resolution when starting a relative timer, to avoid short | |
952 | * timeouts. This will go away with the GTOD framework. | |
953 | */ | |
954 | #ifdef CONFIG_TIME_LOW_RES | |
398ca17f | 955 | tim = ktime_add_safe(tim, ktime_set(0, hrtimer_resolution)); |
06027bdd IM |
956 | #endif |
957 | } | |
237fc6e7 | 958 | |
da8f2e17 | 959 | hrtimer_set_expires_range_ns(timer, tim, delta_ns); |
c0a31329 | 960 | |
84ea7fe3 VK |
961 | /* Switch the timer base, if necessary: */ |
962 | new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED); | |
963 | ||
82f67cd9 IM |
964 | timer_stats_hrtimer_set_start_info(timer); |
965 | ||
a6037b61 PZ |
966 | leftmost = enqueue_hrtimer(timer, new_base); |
967 | ||
49a2a075 VK |
968 | if (!leftmost) { |
969 | unlock_hrtimer_base(timer, &flags); | |
970 | return ret; | |
971 | } | |
972 | ||
973 | if (!hrtimer_is_hres_active(timer)) { | |
974 | /* | |
975 | * Kick to reschedule the next tick to handle the new timer | |
976 | * on dynticks target. | |
977 | */ | |
978 | wake_up_nohz_cpu(new_base->cpu_base->cpu); | |
dc5df73b | 979 | } else if (new_base->cpu_base == this_cpu_ptr(&hrtimer_bases) && |
9e1e01dd | 980 | hrtimer_reprogram(timer, new_base)) { |
49a2a075 VK |
981 | /* |
982 | * Only allow reprogramming if the new base is on this CPU. | |
983 | * (it might still be on another CPU if the timer was pending) | |
984 | * | |
985 | * XXX send_remote_softirq() ? | |
986 | */ | |
b22affe0 LS |
987 | if (wakeup) { |
988 | /* | |
989 | * We need to drop cpu_base->lock to avoid a | |
990 | * lock ordering issue vs. rq->lock. | |
991 | */ | |
992 | raw_spin_unlock(&new_base->cpu_base->lock); | |
993 | raise_softirq_irqoff(HRTIMER_SOFTIRQ); | |
994 | local_irq_restore(flags); | |
995 | return ret; | |
996 | } else { | |
997 | __raise_softirq_irqoff(HRTIMER_SOFTIRQ); | |
998 | } | |
999 | } | |
c0a31329 TG |
1000 | |
1001 | unlock_hrtimer_base(timer, &flags); | |
1002 | ||
1003 | return ret; | |
1004 | } | |
8588a2bb | 1005 | EXPORT_SYMBOL_GPL(__hrtimer_start_range_ns); |
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 | |
8ffbc7d9 DD |
1012 | * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or |
1013 | * relative (HRTIMER_MODE_REL) | |
7f1e2ca9 PZ |
1014 | * |
1015 | * Returns: | |
1016 | * 0 on success | |
1017 | * 1 when the timer was active | |
1018 | */ | |
1019 | int hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, | |
1020 | unsigned long delta_ns, const enum hrtimer_mode mode) | |
1021 | { | |
1022 | return __hrtimer_start_range_ns(timer, tim, delta_ns, mode, 1); | |
1023 | } | |
da8f2e17 AV |
1024 | EXPORT_SYMBOL_GPL(hrtimer_start_range_ns); |
1025 | ||
1026 | /** | |
e1dd7bc5 | 1027 | * hrtimer_start - (re)start an hrtimer on the current CPU |
da8f2e17 AV |
1028 | * @timer: the timer to be added |
1029 | * @tim: expiry time | |
8ffbc7d9 DD |
1030 | * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or |
1031 | * relative (HRTIMER_MODE_REL) | |
da8f2e17 AV |
1032 | * |
1033 | * Returns: | |
1034 | * 0 on success | |
1035 | * 1 when the timer was active | |
1036 | */ | |
1037 | int | |
1038 | hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode) | |
1039 | { | |
7f1e2ca9 | 1040 | return __hrtimer_start_range_ns(timer, tim, 0, mode, 1); |
da8f2e17 | 1041 | } |
8d16b764 | 1042 | EXPORT_SYMBOL_GPL(hrtimer_start); |
c0a31329 | 1043 | |
da8f2e17 | 1044 | |
c0a31329 TG |
1045 | /** |
1046 | * hrtimer_try_to_cancel - try to deactivate a timer | |
c0a31329 TG |
1047 | * @timer: hrtimer to stop |
1048 | * | |
1049 | * Returns: | |
1050 | * 0 when the timer was not active | |
1051 | * 1 when the timer was active | |
1052 | * -1 when the timer is currently excuting the callback function and | |
fa9799e3 | 1053 | * cannot be stopped |
c0a31329 TG |
1054 | */ |
1055 | int hrtimer_try_to_cancel(struct hrtimer *timer) | |
1056 | { | |
3c8aa39d | 1057 | struct hrtimer_clock_base *base; |
c0a31329 TG |
1058 | unsigned long flags; |
1059 | int ret = -1; | |
1060 | ||
1061 | base = lock_hrtimer_base(timer, &flags); | |
1062 | ||
303e967f | 1063 | if (!hrtimer_callback_running(timer)) |
c0a31329 TG |
1064 | ret = remove_hrtimer(timer, base); |
1065 | ||
1066 | unlock_hrtimer_base(timer, &flags); | |
1067 | ||
1068 | return ret; | |
1069 | ||
1070 | } | |
8d16b764 | 1071 | EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel); |
c0a31329 TG |
1072 | |
1073 | /** | |
1074 | * hrtimer_cancel - cancel a timer and wait for the handler to finish. | |
c0a31329 TG |
1075 | * @timer: the timer to be cancelled |
1076 | * | |
1077 | * Returns: | |
1078 | * 0 when the timer was not active | |
1079 | * 1 when the timer was active | |
1080 | */ | |
1081 | int hrtimer_cancel(struct hrtimer *timer) | |
1082 | { | |
1083 | for (;;) { | |
1084 | int ret = hrtimer_try_to_cancel(timer); | |
1085 | ||
1086 | if (ret >= 0) | |
1087 | return ret; | |
5ef37b19 | 1088 | cpu_relax(); |
c0a31329 TG |
1089 | } |
1090 | } | |
8d16b764 | 1091 | EXPORT_SYMBOL_GPL(hrtimer_cancel); |
c0a31329 TG |
1092 | |
1093 | /** | |
1094 | * hrtimer_get_remaining - get remaining time for the timer | |
c0a31329 TG |
1095 | * @timer: the timer to read |
1096 | */ | |
1097 | ktime_t hrtimer_get_remaining(const struct hrtimer *timer) | |
1098 | { | |
c0a31329 TG |
1099 | unsigned long flags; |
1100 | ktime_t rem; | |
1101 | ||
b3bd3de6 | 1102 | lock_hrtimer_base(timer, &flags); |
cc584b21 | 1103 | rem = hrtimer_expires_remaining(timer); |
c0a31329 TG |
1104 | unlock_hrtimer_base(timer, &flags); |
1105 | ||
1106 | return rem; | |
1107 | } | |
8d16b764 | 1108 | EXPORT_SYMBOL_GPL(hrtimer_get_remaining); |
c0a31329 | 1109 | |
3451d024 | 1110 | #ifdef CONFIG_NO_HZ_COMMON |
69239749 TL |
1111 | /** |
1112 | * hrtimer_get_next_event - get the time until next expiry event | |
1113 | * | |
1114 | * Returns the delta to the next expiry event or KTIME_MAX if no timer | |
1115 | * is pending. | |
1116 | */ | |
1117 | ktime_t hrtimer_get_next_event(void) | |
1118 | { | |
dc5df73b | 1119 | struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases); |
9bc74919 | 1120 | ktime_t mindelta = { .tv64 = KTIME_MAX }; |
69239749 | 1121 | unsigned long flags; |
69239749 | 1122 | |
ecb49d1a | 1123 | raw_spin_lock_irqsave(&cpu_base->lock, flags); |
3c8aa39d | 1124 | |
e19ffe8b | 1125 | if (!__hrtimer_hres_active(cpu_base)) |
9bc74919 TG |
1126 | mindelta = ktime_sub(__hrtimer_get_next_event(cpu_base), |
1127 | ktime_get()); | |
3c8aa39d | 1128 | |
ecb49d1a | 1129 | raw_spin_unlock_irqrestore(&cpu_base->lock, flags); |
3c8aa39d | 1130 | |
69239749 TL |
1131 | if (mindelta.tv64 < 0) |
1132 | mindelta.tv64 = 0; | |
1133 | return mindelta; | |
1134 | } | |
1135 | #endif | |
1136 | ||
237fc6e7 TG |
1137 | static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id, |
1138 | enum hrtimer_mode mode) | |
c0a31329 | 1139 | { |
3c8aa39d | 1140 | struct hrtimer_cpu_base *cpu_base; |
e06383db | 1141 | int base; |
c0a31329 | 1142 | |
7978672c GA |
1143 | memset(timer, 0, sizeof(struct hrtimer)); |
1144 | ||
22127e93 | 1145 | cpu_base = raw_cpu_ptr(&hrtimer_bases); |
c0a31329 | 1146 | |
c9cb2e3d | 1147 | if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS) |
7978672c GA |
1148 | clock_id = CLOCK_MONOTONIC; |
1149 | ||
e06383db JS |
1150 | base = hrtimer_clockid_to_base(clock_id); |
1151 | timer->base = &cpu_base->clock_base[base]; | |
998adc3d | 1152 | timerqueue_init(&timer->node); |
82f67cd9 IM |
1153 | |
1154 | #ifdef CONFIG_TIMER_STATS | |
1155 | timer->start_site = NULL; | |
1156 | timer->start_pid = -1; | |
1157 | memset(timer->start_comm, 0, TASK_COMM_LEN); | |
1158 | #endif | |
c0a31329 | 1159 | } |
237fc6e7 TG |
1160 | |
1161 | /** | |
1162 | * hrtimer_init - initialize a timer to the given clock | |
1163 | * @timer: the timer to be initialized | |
1164 | * @clock_id: the clock to be used | |
1165 | * @mode: timer mode abs/rel | |
1166 | */ | |
1167 | void hrtimer_init(struct hrtimer *timer, clockid_t clock_id, | |
1168 | enum hrtimer_mode mode) | |
1169 | { | |
c6a2a177 | 1170 | debug_init(timer, clock_id, mode); |
237fc6e7 TG |
1171 | __hrtimer_init(timer, clock_id, mode); |
1172 | } | |
8d16b764 | 1173 | EXPORT_SYMBOL_GPL(hrtimer_init); |
c0a31329 | 1174 | |
21d6d52a TG |
1175 | static void __run_hrtimer(struct hrtimer_cpu_base *cpu_base, |
1176 | struct hrtimer_clock_base *base, | |
1177 | struct hrtimer *timer, ktime_t *now) | |
d3d74453 | 1178 | { |
d3d74453 PZ |
1179 | enum hrtimer_restart (*fn)(struct hrtimer *); |
1180 | int restart; | |
1181 | ||
ca109491 PZ |
1182 | WARN_ON(!irqs_disabled()); |
1183 | ||
c6a2a177 | 1184 | debug_deactivate(timer); |
d3d74453 PZ |
1185 | __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0); |
1186 | timer_stats_account_hrtimer(timer); | |
d3d74453 | 1187 | fn = timer->function; |
ca109491 PZ |
1188 | |
1189 | /* | |
1190 | * Because we run timers from hardirq context, there is no chance | |
1191 | * they get migrated to another cpu, therefore its safe to unlock | |
1192 | * the timer base. | |
1193 | */ | |
ecb49d1a | 1194 | raw_spin_unlock(&cpu_base->lock); |
c6a2a177 | 1195 | trace_hrtimer_expire_entry(timer, now); |
ca109491 | 1196 | restart = fn(timer); |
c6a2a177 | 1197 | trace_hrtimer_expire_exit(timer); |
ecb49d1a | 1198 | raw_spin_lock(&cpu_base->lock); |
d3d74453 PZ |
1199 | |
1200 | /* | |
e3f1d883 TG |
1201 | * Note: We clear the CALLBACK bit after enqueue_hrtimer and |
1202 | * we do not reprogramm the event hardware. Happens either in | |
1203 | * hrtimer_start_range_ns() or in hrtimer_interrupt() | |
d3d74453 PZ |
1204 | */ |
1205 | if (restart != HRTIMER_NORESTART) { | |
1206 | BUG_ON(timer->state != HRTIMER_STATE_CALLBACK); | |
a6037b61 | 1207 | enqueue_hrtimer(timer, base); |
d3d74453 | 1208 | } |
f13d4f97 SQ |
1209 | |
1210 | WARN_ON_ONCE(!(timer->state & HRTIMER_STATE_CALLBACK)); | |
1211 | ||
d3d74453 PZ |
1212 | timer->state &= ~HRTIMER_STATE_CALLBACK; |
1213 | } | |
1214 | ||
21d6d52a | 1215 | static void __hrtimer_run_queues(struct hrtimer_cpu_base *cpu_base, ktime_t now) |
54cdfdb4 | 1216 | { |
21d6d52a | 1217 | int i; |
6ff7041d | 1218 | |
54cdfdb4 | 1219 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
ab8177bc | 1220 | struct hrtimer_clock_base *base; |
998adc3d | 1221 | struct timerqueue_node *node; |
ab8177bc TG |
1222 | ktime_t basenow; |
1223 | ||
1224 | if (!(cpu_base->active_bases & (1 << i))) | |
1225 | continue; | |
54cdfdb4 | 1226 | |
ab8177bc | 1227 | base = cpu_base->clock_base + i; |
54cdfdb4 TG |
1228 | basenow = ktime_add(now, base->offset); |
1229 | ||
998adc3d | 1230 | while ((node = timerqueue_getnext(&base->active))) { |
54cdfdb4 TG |
1231 | struct hrtimer *timer; |
1232 | ||
998adc3d | 1233 | timer = container_of(node, struct hrtimer, node); |
54cdfdb4 | 1234 | |
654c8e0b AV |
1235 | /* |
1236 | * The immediate goal for using the softexpires is | |
1237 | * minimizing wakeups, not running timers at the | |
1238 | * earliest interrupt after their soft expiration. | |
1239 | * This allows us to avoid using a Priority Search | |
1240 | * Tree, which can answer a stabbing querry for | |
1241 | * overlapping intervals and instead use the simple | |
1242 | * BST we already have. | |
1243 | * We don't add extra wakeups by delaying timers that | |
1244 | * are right-of a not yet expired timer, because that | |
1245 | * timer will have to trigger a wakeup anyway. | |
1246 | */ | |
9bc74919 | 1247 | if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer)) |
54cdfdb4 | 1248 | break; |
54cdfdb4 | 1249 | |
21d6d52a | 1250 | __run_hrtimer(cpu_base, base, timer, &basenow); |
54cdfdb4 | 1251 | } |
54cdfdb4 | 1252 | } |
21d6d52a TG |
1253 | } |
1254 | ||
1255 | #ifdef CONFIG_HIGH_RES_TIMERS | |
1256 | ||
1257 | /* | |
1258 | * High resolution timer interrupt | |
1259 | * Called with interrupts disabled | |
1260 | */ | |
1261 | void hrtimer_interrupt(struct clock_event_device *dev) | |
1262 | { | |
1263 | struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases); | |
1264 | ktime_t expires_next, now, entry_time, delta; | |
1265 | int retries = 0; | |
1266 | ||
1267 | BUG_ON(!cpu_base->hres_active); | |
1268 | cpu_base->nr_events++; | |
1269 | dev->next_event.tv64 = KTIME_MAX; | |
1270 | ||
1271 | raw_spin_lock(&cpu_base->lock); | |
1272 | entry_time = now = hrtimer_update_base(cpu_base); | |
1273 | retry: | |
1274 | cpu_base->in_hrtirq = 1; | |
1275 | /* | |
1276 | * We set expires_next to KTIME_MAX here with cpu_base->lock | |
1277 | * held to prevent that a timer is enqueued in our queue via | |
1278 | * the migration code. This does not affect enqueueing of | |
1279 | * timers which run their callback and need to be requeued on | |
1280 | * this CPU. | |
1281 | */ | |
1282 | cpu_base->expires_next.tv64 = KTIME_MAX; | |
1283 | ||
1284 | __hrtimer_run_queues(cpu_base, now); | |
1285 | ||
9bc74919 TG |
1286 | /* Reevaluate the clock bases for the next expiry */ |
1287 | expires_next = __hrtimer_get_next_event(cpu_base); | |
6ff7041d TG |
1288 | /* |
1289 | * Store the new expiry value so the migration code can verify | |
1290 | * against it. | |
1291 | */ | |
54cdfdb4 | 1292 | cpu_base->expires_next = expires_next; |
9bc74919 | 1293 | cpu_base->in_hrtirq = 0; |
ecb49d1a | 1294 | raw_spin_unlock(&cpu_base->lock); |
54cdfdb4 TG |
1295 | |
1296 | /* Reprogramming necessary ? */ | |
41d2e494 TG |
1297 | if (expires_next.tv64 == KTIME_MAX || |
1298 | !tick_program_event(expires_next, 0)) { | |
1299 | cpu_base->hang_detected = 0; | |
1300 | return; | |
54cdfdb4 | 1301 | } |
41d2e494 TG |
1302 | |
1303 | /* | |
1304 | * The next timer was already expired due to: | |
1305 | * - tracing | |
1306 | * - long lasting callbacks | |
1307 | * - being scheduled away when running in a VM | |
1308 | * | |
1309 | * We need to prevent that we loop forever in the hrtimer | |
1310 | * interrupt routine. We give it 3 attempts to avoid | |
1311 | * overreacting on some spurious event. | |
5baefd6d JS |
1312 | * |
1313 | * Acquire base lock for updating the offsets and retrieving | |
1314 | * the current time. | |
41d2e494 | 1315 | */ |
196951e9 | 1316 | raw_spin_lock(&cpu_base->lock); |
5baefd6d | 1317 | now = hrtimer_update_base(cpu_base); |
41d2e494 TG |
1318 | cpu_base->nr_retries++; |
1319 | if (++retries < 3) | |
1320 | goto retry; | |
1321 | /* | |
1322 | * Give the system a chance to do something else than looping | |
1323 | * here. We stored the entry time, so we know exactly how long | |
1324 | * we spent here. We schedule the next event this amount of | |
1325 | * time away. | |
1326 | */ | |
1327 | cpu_base->nr_hangs++; | |
1328 | cpu_base->hang_detected = 1; | |
196951e9 | 1329 | raw_spin_unlock(&cpu_base->lock); |
41d2e494 | 1330 | delta = ktime_sub(now, entry_time); |
a6ffebce TG |
1331 | if ((unsigned int)delta.tv64 > cpu_base->max_hang_time) |
1332 | cpu_base->max_hang_time = (unsigned int) delta.tv64; | |
41d2e494 TG |
1333 | /* |
1334 | * Limit it to a sensible value as we enforce a longer | |
1335 | * delay. Give the CPU at least 100ms to catch up. | |
1336 | */ | |
1337 | if (delta.tv64 > 100 * NSEC_PER_MSEC) | |
1338 | expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC); | |
1339 | else | |
1340 | expires_next = ktime_add(now, delta); | |
1341 | tick_program_event(expires_next, 1); | |
1342 | printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n", | |
1343 | ktime_to_ns(delta)); | |
54cdfdb4 TG |
1344 | } |
1345 | ||
8bdec955 TG |
1346 | /* |
1347 | * local version of hrtimer_peek_ahead_timers() called with interrupts | |
1348 | * disabled. | |
1349 | */ | |
1350 | static void __hrtimer_peek_ahead_timers(void) | |
1351 | { | |
1352 | struct tick_device *td; | |
1353 | ||
1354 | if (!hrtimer_hres_active()) | |
1355 | return; | |
1356 | ||
22127e93 | 1357 | td = this_cpu_ptr(&tick_cpu_device); |
8bdec955 TG |
1358 | if (td && td->evtdev) |
1359 | hrtimer_interrupt(td->evtdev); | |
1360 | } | |
1361 | ||
2e94d1f7 AV |
1362 | /** |
1363 | * hrtimer_peek_ahead_timers -- run soft-expired timers now | |
1364 | * | |
1365 | * hrtimer_peek_ahead_timers will peek at the timer queue of | |
1366 | * the current cpu and check if there are any timers for which | |
1367 | * the soft expires time has passed. If any such timers exist, | |
1368 | * they are run immediately and then removed from the timer queue. | |
1369 | * | |
1370 | */ | |
1371 | void hrtimer_peek_ahead_timers(void) | |
1372 | { | |
643bdf68 | 1373 | unsigned long flags; |
dc4304f7 | 1374 | |
2e94d1f7 | 1375 | local_irq_save(flags); |
8bdec955 | 1376 | __hrtimer_peek_ahead_timers(); |
2e94d1f7 AV |
1377 | local_irq_restore(flags); |
1378 | } | |
1379 | ||
a6037b61 PZ |
1380 | static void run_hrtimer_softirq(struct softirq_action *h) |
1381 | { | |
1382 | hrtimer_peek_ahead_timers(); | |
1383 | } | |
1384 | ||
82c5b7b5 IM |
1385 | #else /* CONFIG_HIGH_RES_TIMERS */ |
1386 | ||
1387 | static inline void __hrtimer_peek_ahead_timers(void) { } | |
1388 | ||
1389 | #endif /* !CONFIG_HIGH_RES_TIMERS */ | |
82f67cd9 | 1390 | |
d3d74453 PZ |
1391 | /* |
1392 | * Called from timer softirq every jiffy, expire hrtimers: | |
1393 | * | |
1394 | * For HRT its the fall back code to run the softirq in the timer | |
1395 | * softirq context in case the hrtimer initialization failed or has | |
1396 | * not been done yet. | |
1397 | */ | |
1398 | void hrtimer_run_pending(void) | |
1399 | { | |
d3d74453 PZ |
1400 | if (hrtimer_hres_active()) |
1401 | return; | |
54cdfdb4 | 1402 | |
d3d74453 PZ |
1403 | /* |
1404 | * This _is_ ugly: We have to check in the softirq context, | |
1405 | * whether we can switch to highres and / or nohz mode. The | |
1406 | * clocksource switch happens in the timer interrupt with | |
1407 | * xtime_lock held. Notification from there only sets the | |
1408 | * check bit in the tick_oneshot code, otherwise we might | |
1409 | * deadlock vs. xtime_lock. | |
1410 | */ | |
1411 | if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) | |
1412 | hrtimer_switch_to_hres(); | |
54cdfdb4 TG |
1413 | } |
1414 | ||
c0a31329 | 1415 | /* |
d3d74453 | 1416 | * Called from hardirq context every jiffy |
c0a31329 | 1417 | */ |
833883d9 | 1418 | void hrtimer_run_queues(void) |
c0a31329 | 1419 | { |
dc5df73b | 1420 | struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases); |
21d6d52a | 1421 | ktime_t now; |
c0a31329 | 1422 | |
e19ffe8b | 1423 | if (__hrtimer_hres_active(cpu_base)) |
3055adda DS |
1424 | return; |
1425 | ||
21d6d52a TG |
1426 | raw_spin_lock(&cpu_base->lock); |
1427 | now = hrtimer_update_base(cpu_base); | |
1428 | __hrtimer_run_queues(cpu_base, now); | |
1429 | raw_spin_unlock(&cpu_base->lock); | |
c0a31329 TG |
1430 | } |
1431 | ||
10c94ec1 TG |
1432 | /* |
1433 | * Sleep related functions: | |
1434 | */ | |
c9cb2e3d | 1435 | static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer) |
00362e33 TG |
1436 | { |
1437 | struct hrtimer_sleeper *t = | |
1438 | container_of(timer, struct hrtimer_sleeper, timer); | |
1439 | struct task_struct *task = t->task; | |
1440 | ||
1441 | t->task = NULL; | |
1442 | if (task) | |
1443 | wake_up_process(task); | |
1444 | ||
1445 | return HRTIMER_NORESTART; | |
1446 | } | |
1447 | ||
36c8b586 | 1448 | void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task) |
00362e33 TG |
1449 | { |
1450 | sl->timer.function = hrtimer_wakeup; | |
1451 | sl->task = task; | |
1452 | } | |
2bc481cf | 1453 | EXPORT_SYMBOL_GPL(hrtimer_init_sleeper); |
00362e33 | 1454 | |
669d7868 | 1455 | static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode) |
432569bb | 1456 | { |
669d7868 | 1457 | hrtimer_init_sleeper(t, current); |
10c94ec1 | 1458 | |
432569bb RZ |
1459 | do { |
1460 | set_current_state(TASK_INTERRUPTIBLE); | |
cc584b21 | 1461 | hrtimer_start_expires(&t->timer, mode); |
37bb6cb4 PZ |
1462 | if (!hrtimer_active(&t->timer)) |
1463 | t->task = NULL; | |
432569bb | 1464 | |
54cdfdb4 | 1465 | if (likely(t->task)) |
b0f8c44f | 1466 | freezable_schedule(); |
432569bb | 1467 | |
669d7868 | 1468 | hrtimer_cancel(&t->timer); |
c9cb2e3d | 1469 | mode = HRTIMER_MODE_ABS; |
669d7868 TG |
1470 | |
1471 | } while (t->task && !signal_pending(current)); | |
432569bb | 1472 | |
3588a085 PZ |
1473 | __set_current_state(TASK_RUNNING); |
1474 | ||
669d7868 | 1475 | return t->task == NULL; |
10c94ec1 TG |
1476 | } |
1477 | ||
080344b9 ON |
1478 | static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp) |
1479 | { | |
1480 | struct timespec rmt; | |
1481 | ktime_t rem; | |
1482 | ||
cc584b21 | 1483 | rem = hrtimer_expires_remaining(timer); |
080344b9 ON |
1484 | if (rem.tv64 <= 0) |
1485 | return 0; | |
1486 | rmt = ktime_to_timespec(rem); | |
1487 | ||
1488 | if (copy_to_user(rmtp, &rmt, sizeof(*rmtp))) | |
1489 | return -EFAULT; | |
1490 | ||
1491 | return 1; | |
1492 | } | |
1493 | ||
1711ef38 | 1494 | long __sched hrtimer_nanosleep_restart(struct restart_block *restart) |
10c94ec1 | 1495 | { |
669d7868 | 1496 | struct hrtimer_sleeper t; |
080344b9 | 1497 | struct timespec __user *rmtp; |
237fc6e7 | 1498 | int ret = 0; |
10c94ec1 | 1499 | |
ab8177bc | 1500 | hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid, |
237fc6e7 | 1501 | HRTIMER_MODE_ABS); |
cc584b21 | 1502 | hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires); |
10c94ec1 | 1503 | |
c9cb2e3d | 1504 | if (do_nanosleep(&t, HRTIMER_MODE_ABS)) |
237fc6e7 | 1505 | goto out; |
10c94ec1 | 1506 | |
029a07e0 | 1507 | rmtp = restart->nanosleep.rmtp; |
432569bb | 1508 | if (rmtp) { |
237fc6e7 | 1509 | ret = update_rmtp(&t.timer, rmtp); |
080344b9 | 1510 | if (ret <= 0) |
237fc6e7 | 1511 | goto out; |
432569bb | 1512 | } |
10c94ec1 | 1513 | |
10c94ec1 | 1514 | /* The other values in restart are already filled in */ |
237fc6e7 TG |
1515 | ret = -ERESTART_RESTARTBLOCK; |
1516 | out: | |
1517 | destroy_hrtimer_on_stack(&t.timer); | |
1518 | return ret; | |
10c94ec1 TG |
1519 | } |
1520 | ||
080344b9 | 1521 | long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp, |
10c94ec1 TG |
1522 | const enum hrtimer_mode mode, const clockid_t clockid) |
1523 | { | |
1524 | struct restart_block *restart; | |
669d7868 | 1525 | struct hrtimer_sleeper t; |
237fc6e7 | 1526 | int ret = 0; |
3bd01206 AV |
1527 | unsigned long slack; |
1528 | ||
1529 | slack = current->timer_slack_ns; | |
aab03e05 | 1530 | if (dl_task(current) || rt_task(current)) |
3bd01206 | 1531 | slack = 0; |
10c94ec1 | 1532 | |
237fc6e7 | 1533 | hrtimer_init_on_stack(&t.timer, clockid, mode); |
3bd01206 | 1534 | hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack); |
432569bb | 1535 | if (do_nanosleep(&t, mode)) |
237fc6e7 | 1536 | goto out; |
10c94ec1 | 1537 | |
7978672c | 1538 | /* Absolute timers do not update the rmtp value and restart: */ |
237fc6e7 TG |
1539 | if (mode == HRTIMER_MODE_ABS) { |
1540 | ret = -ERESTARTNOHAND; | |
1541 | goto out; | |
1542 | } | |
10c94ec1 | 1543 | |
432569bb | 1544 | if (rmtp) { |
237fc6e7 | 1545 | ret = update_rmtp(&t.timer, rmtp); |
080344b9 | 1546 | if (ret <= 0) |
237fc6e7 | 1547 | goto out; |
432569bb | 1548 | } |
10c94ec1 | 1549 | |
f56141e3 | 1550 | restart = ¤t->restart_block; |
1711ef38 | 1551 | restart->fn = hrtimer_nanosleep_restart; |
ab8177bc | 1552 | restart->nanosleep.clockid = t.timer.base->clockid; |
029a07e0 | 1553 | restart->nanosleep.rmtp = rmtp; |
cc584b21 | 1554 | restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer); |
10c94ec1 | 1555 | |
237fc6e7 TG |
1556 | ret = -ERESTART_RESTARTBLOCK; |
1557 | out: | |
1558 | destroy_hrtimer_on_stack(&t.timer); | |
1559 | return ret; | |
10c94ec1 TG |
1560 | } |
1561 | ||
58fd3aa2 HC |
1562 | SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp, |
1563 | struct timespec __user *, rmtp) | |
6ba1b912 | 1564 | { |
080344b9 | 1565 | struct timespec tu; |
6ba1b912 TG |
1566 | |
1567 | if (copy_from_user(&tu, rqtp, sizeof(tu))) | |
1568 | return -EFAULT; | |
1569 | ||
1570 | if (!timespec_valid(&tu)) | |
1571 | return -EINVAL; | |
1572 | ||
080344b9 | 1573 | return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC); |
6ba1b912 TG |
1574 | } |
1575 | ||
c0a31329 TG |
1576 | /* |
1577 | * Functions related to boot-time initialization: | |
1578 | */ | |
0db0628d | 1579 | static void init_hrtimers_cpu(int cpu) |
c0a31329 | 1580 | { |
3c8aa39d | 1581 | struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu); |
c0a31329 TG |
1582 | int i; |
1583 | ||
998adc3d | 1584 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
3c8aa39d | 1585 | cpu_base->clock_base[i].cpu_base = cpu_base; |
998adc3d JS |
1586 | timerqueue_init_head(&cpu_base->clock_base[i].active); |
1587 | } | |
3c8aa39d | 1588 | |
cddd0248 | 1589 | cpu_base->cpu = cpu; |
54cdfdb4 | 1590 | hrtimer_init_hres(cpu_base); |
c0a31329 TG |
1591 | } |
1592 | ||
1593 | #ifdef CONFIG_HOTPLUG_CPU | |
1594 | ||
ca109491 | 1595 | static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base, |
37810659 | 1596 | struct hrtimer_clock_base *new_base) |
c0a31329 TG |
1597 | { |
1598 | struct hrtimer *timer; | |
998adc3d | 1599 | struct timerqueue_node *node; |
c0a31329 | 1600 | |
998adc3d JS |
1601 | while ((node = timerqueue_getnext(&old_base->active))) { |
1602 | timer = container_of(node, struct hrtimer, node); | |
54cdfdb4 | 1603 | BUG_ON(hrtimer_callback_running(timer)); |
c6a2a177 | 1604 | debug_deactivate(timer); |
b00c1a99 TG |
1605 | |
1606 | /* | |
1607 | * Mark it as STATE_MIGRATE not INACTIVE otherwise the | |
1608 | * timer could be seen as !active and just vanish away | |
1609 | * under us on another CPU | |
1610 | */ | |
1611 | __remove_hrtimer(timer, old_base, HRTIMER_STATE_MIGRATE, 0); | |
c0a31329 | 1612 | timer->base = new_base; |
54cdfdb4 | 1613 | /* |
e3f1d883 TG |
1614 | * Enqueue the timers on the new cpu. This does not |
1615 | * reprogram the event device in case the timer | |
1616 | * expires before the earliest on this CPU, but we run | |
1617 | * hrtimer_interrupt after we migrated everything to | |
1618 | * sort out already expired timers and reprogram the | |
1619 | * event device. | |
54cdfdb4 | 1620 | */ |
a6037b61 | 1621 | enqueue_hrtimer(timer, new_base); |
41e1022e | 1622 | |
b00c1a99 TG |
1623 | /* Clear the migration state bit */ |
1624 | timer->state &= ~HRTIMER_STATE_MIGRATE; | |
c0a31329 TG |
1625 | } |
1626 | } | |
1627 | ||
d5fd43c4 | 1628 | static void migrate_hrtimers(int scpu) |
c0a31329 | 1629 | { |
3c8aa39d | 1630 | struct hrtimer_cpu_base *old_base, *new_base; |
731a55ba | 1631 | int i; |
c0a31329 | 1632 | |
37810659 | 1633 | BUG_ON(cpu_online(scpu)); |
37810659 | 1634 | tick_cancel_sched_timer(scpu); |
731a55ba TG |
1635 | |
1636 | local_irq_disable(); | |
1637 | old_base = &per_cpu(hrtimer_bases, scpu); | |
dc5df73b | 1638 | new_base = this_cpu_ptr(&hrtimer_bases); |
d82f0b0f ON |
1639 | /* |
1640 | * The caller is globally serialized and nobody else | |
1641 | * takes two locks at once, deadlock is not possible. | |
1642 | */ | |
ecb49d1a TG |
1643 | raw_spin_lock(&new_base->lock); |
1644 | raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING); | |
c0a31329 | 1645 | |
3c8aa39d | 1646 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
ca109491 | 1647 | migrate_hrtimer_list(&old_base->clock_base[i], |
37810659 | 1648 | &new_base->clock_base[i]); |
c0a31329 TG |
1649 | } |
1650 | ||
ecb49d1a TG |
1651 | raw_spin_unlock(&old_base->lock); |
1652 | raw_spin_unlock(&new_base->lock); | |
37810659 | 1653 | |
731a55ba TG |
1654 | /* Check, if we got expired work to do */ |
1655 | __hrtimer_peek_ahead_timers(); | |
1656 | local_irq_enable(); | |
c0a31329 | 1657 | } |
37810659 | 1658 | |
c0a31329 TG |
1659 | #endif /* CONFIG_HOTPLUG_CPU */ |
1660 | ||
0db0628d | 1661 | static int hrtimer_cpu_notify(struct notifier_block *self, |
c0a31329 TG |
1662 | unsigned long action, void *hcpu) |
1663 | { | |
b2e3c0ad | 1664 | int scpu = (long)hcpu; |
c0a31329 TG |
1665 | |
1666 | switch (action) { | |
1667 | ||
1668 | case CPU_UP_PREPARE: | |
8bb78442 | 1669 | case CPU_UP_PREPARE_FROZEN: |
37810659 | 1670 | init_hrtimers_cpu(scpu); |
c0a31329 TG |
1671 | break; |
1672 | ||
1673 | #ifdef CONFIG_HOTPLUG_CPU | |
1674 | case CPU_DEAD: | |
8bb78442 | 1675 | case CPU_DEAD_FROZEN: |
d5fd43c4 | 1676 | migrate_hrtimers(scpu); |
c0a31329 TG |
1677 | break; |
1678 | #endif | |
1679 | ||
1680 | default: | |
1681 | break; | |
1682 | } | |
1683 | ||
1684 | return NOTIFY_OK; | |
1685 | } | |
1686 | ||
0db0628d | 1687 | static struct notifier_block hrtimers_nb = { |
c0a31329 TG |
1688 | .notifier_call = hrtimer_cpu_notify, |
1689 | }; | |
1690 | ||
1691 | void __init hrtimers_init(void) | |
1692 | { | |
1693 | hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE, | |
1694 | (void *)(long)smp_processor_id()); | |
1695 | register_cpu_notifier(&hrtimers_nb); | |
a6037b61 PZ |
1696 | #ifdef CONFIG_HIGH_RES_TIMERS |
1697 | open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq); | |
1698 | #endif | |
c0a31329 TG |
1699 | } |
1700 | ||
7bb67439 | 1701 | /** |
351b3f7a | 1702 | * schedule_hrtimeout_range_clock - sleep until timeout |
7bb67439 | 1703 | * @expires: timeout value (ktime_t) |
654c8e0b | 1704 | * @delta: slack in expires timeout (ktime_t) |
7bb67439 | 1705 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL |
351b3f7a | 1706 | * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME |
7bb67439 | 1707 | */ |
351b3f7a CE |
1708 | int __sched |
1709 | schedule_hrtimeout_range_clock(ktime_t *expires, unsigned long delta, | |
1710 | const enum hrtimer_mode mode, int clock) | |
7bb67439 AV |
1711 | { |
1712 | struct hrtimer_sleeper t; | |
1713 | ||
1714 | /* | |
1715 | * Optimize when a zero timeout value is given. It does not | |
1716 | * matter whether this is an absolute or a relative time. | |
1717 | */ | |
1718 | if (expires && !expires->tv64) { | |
1719 | __set_current_state(TASK_RUNNING); | |
1720 | return 0; | |
1721 | } | |
1722 | ||
1723 | /* | |
43b21013 | 1724 | * A NULL parameter means "infinite" |
7bb67439 AV |
1725 | */ |
1726 | if (!expires) { | |
1727 | schedule(); | |
7bb67439 AV |
1728 | return -EINTR; |
1729 | } | |
1730 | ||
351b3f7a | 1731 | hrtimer_init_on_stack(&t.timer, clock, mode); |
654c8e0b | 1732 | hrtimer_set_expires_range_ns(&t.timer, *expires, delta); |
7bb67439 AV |
1733 | |
1734 | hrtimer_init_sleeper(&t, current); | |
1735 | ||
cc584b21 | 1736 | hrtimer_start_expires(&t.timer, mode); |
7bb67439 AV |
1737 | if (!hrtimer_active(&t.timer)) |
1738 | t.task = NULL; | |
1739 | ||
1740 | if (likely(t.task)) | |
1741 | schedule(); | |
1742 | ||
1743 | hrtimer_cancel(&t.timer); | |
1744 | destroy_hrtimer_on_stack(&t.timer); | |
1745 | ||
1746 | __set_current_state(TASK_RUNNING); | |
1747 | ||
1748 | return !t.task ? 0 : -EINTR; | |
1749 | } | |
351b3f7a CE |
1750 | |
1751 | /** | |
1752 | * schedule_hrtimeout_range - sleep until timeout | |
1753 | * @expires: timeout value (ktime_t) | |
1754 | * @delta: slack in expires timeout (ktime_t) | |
1755 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL | |
1756 | * | |
1757 | * Make the current task sleep until the given expiry time has | |
1758 | * elapsed. The routine will return immediately unless | |
1759 | * the current task state has been set (see set_current_state()). | |
1760 | * | |
1761 | * The @delta argument gives the kernel the freedom to schedule the | |
1762 | * actual wakeup to a time that is both power and performance friendly. | |
1763 | * The kernel give the normal best effort behavior for "@expires+@delta", | |
1764 | * but may decide to fire the timer earlier, but no earlier than @expires. | |
1765 | * | |
1766 | * You can set the task state as follows - | |
1767 | * | |
1768 | * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to | |
1769 | * pass before the routine returns. | |
1770 | * | |
1771 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
1772 | * delivered to the current task. | |
1773 | * | |
1774 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1775 | * routine returns. | |
1776 | * | |
1777 | * Returns 0 when the timer has expired otherwise -EINTR | |
1778 | */ | |
1779 | int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta, | |
1780 | const enum hrtimer_mode mode) | |
1781 | { | |
1782 | return schedule_hrtimeout_range_clock(expires, delta, mode, | |
1783 | CLOCK_MONOTONIC); | |
1784 | } | |
654c8e0b AV |
1785 | EXPORT_SYMBOL_GPL(schedule_hrtimeout_range); |
1786 | ||
1787 | /** | |
1788 | * schedule_hrtimeout - sleep until timeout | |
1789 | * @expires: timeout value (ktime_t) | |
1790 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL | |
1791 | * | |
1792 | * Make the current task sleep until the given expiry time has | |
1793 | * elapsed. The routine will return immediately unless | |
1794 | * the current task state has been set (see set_current_state()). | |
1795 | * | |
1796 | * You can set the task state as follows - | |
1797 | * | |
1798 | * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to | |
1799 | * pass before the routine returns. | |
1800 | * | |
1801 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
1802 | * delivered to the current task. | |
1803 | * | |
1804 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1805 | * routine returns. | |
1806 | * | |
1807 | * Returns 0 when the timer has expired otherwise -EINTR | |
1808 | */ | |
1809 | int __sched schedule_hrtimeout(ktime_t *expires, | |
1810 | const enum hrtimer_mode mode) | |
1811 | { | |
1812 | return schedule_hrtimeout_range(expires, 0, mode); | |
1813 | } | |
7bb67439 | 1814 | EXPORT_SYMBOL_GPL(schedule_hrtimeout); |