]>
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), |
887d9dc9 | 70 | .seq = SEQCNT_ZERO(hrtimer_bases.seq), |
3c8aa39d | 71 | .clock_base = |
c0a31329 | 72 | { |
3c8aa39d | 73 | { |
ab8177bc TG |
74 | .index = HRTIMER_BASE_MONOTONIC, |
75 | .clockid = CLOCK_MONOTONIC, | |
3c8aa39d | 76 | .get_time = &ktime_get, |
3c8aa39d | 77 | }, |
68fa61c0 TG |
78 | { |
79 | .index = HRTIMER_BASE_REALTIME, | |
80 | .clockid = CLOCK_REALTIME, | |
81 | .get_time = &ktime_get_real, | |
68fa61c0 | 82 | }, |
70a08cca | 83 | { |
ab8177bc TG |
84 | .index = HRTIMER_BASE_BOOTTIME, |
85 | .clockid = CLOCK_BOOTTIME, | |
70a08cca | 86 | .get_time = &ktime_get_boottime, |
70a08cca | 87 | }, |
90adda98 JS |
88 | { |
89 | .index = HRTIMER_BASE_TAI, | |
90 | .clockid = CLOCK_TAI, | |
91 | .get_time = &ktime_get_clocktai, | |
90adda98 | 92 | }, |
3c8aa39d | 93 | } |
c0a31329 TG |
94 | }; |
95 | ||
942c3c5c | 96 | static const int hrtimer_clock_to_base_table[MAX_CLOCKS] = { |
ce31332d TG |
97 | [CLOCK_REALTIME] = HRTIMER_BASE_REALTIME, |
98 | [CLOCK_MONOTONIC] = HRTIMER_BASE_MONOTONIC, | |
99 | [CLOCK_BOOTTIME] = HRTIMER_BASE_BOOTTIME, | |
90adda98 | 100 | [CLOCK_TAI] = HRTIMER_BASE_TAI, |
ce31332d | 101 | }; |
e06383db JS |
102 | |
103 | static inline int hrtimer_clockid_to_base(clockid_t clock_id) | |
104 | { | |
105 | return hrtimer_clock_to_base_table[clock_id]; | |
106 | } | |
107 | ||
c0a31329 TG |
108 | /* |
109 | * Functions and macros which are different for UP/SMP systems are kept in a | |
110 | * single place | |
111 | */ | |
112 | #ifdef CONFIG_SMP | |
113 | ||
887d9dc9 PZ |
114 | /* |
115 | * We require the migration_base for lock_hrtimer_base()/switch_hrtimer_base() | |
116 | * such that hrtimer_callback_running() can unconditionally dereference | |
117 | * timer->base->cpu_base | |
118 | */ | |
119 | static struct hrtimer_cpu_base migration_cpu_base = { | |
120 | .seq = SEQCNT_ZERO(migration_cpu_base), | |
121 | .clock_base = { { .cpu_base = &migration_cpu_base, }, }, | |
122 | }; | |
123 | ||
124 | #define migration_base migration_cpu_base.clock_base[0] | |
125 | ||
c0a31329 TG |
126 | /* |
127 | * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock | |
128 | * means that all timers which are tied to this base via timer->base are | |
129 | * locked, and the base itself is locked too. | |
130 | * | |
131 | * So __run_timers/migrate_timers can safely modify all timers which could | |
132 | * be found on the lists/queues. | |
133 | * | |
134 | * When the timer's base is locked, and the timer removed from list, it is | |
887d9dc9 PZ |
135 | * possible to set timer->base = &migration_base and drop the lock: the timer |
136 | * remains locked. | |
c0a31329 | 137 | */ |
3c8aa39d TG |
138 | static |
139 | struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer, | |
140 | unsigned long *flags) | |
c0a31329 | 141 | { |
3c8aa39d | 142 | struct hrtimer_clock_base *base; |
c0a31329 TG |
143 | |
144 | for (;;) { | |
145 | base = timer->base; | |
887d9dc9 | 146 | if (likely(base != &migration_base)) { |
ecb49d1a | 147 | raw_spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
148 | if (likely(base == timer->base)) |
149 | return base; | |
150 | /* The timer has migrated to another CPU: */ | |
ecb49d1a | 151 | raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags); |
c0a31329 TG |
152 | } |
153 | cpu_relax(); | |
154 | } | |
155 | } | |
156 | ||
6ff7041d TG |
157 | /* |
158 | * With HIGHRES=y we do not migrate the timer when it is expiring | |
159 | * before the next event on the target cpu because we cannot reprogram | |
160 | * the target cpu hardware and we would cause it to fire late. | |
161 | * | |
162 | * Called with cpu_base->lock of target cpu held. | |
163 | */ | |
164 | static int | |
165 | hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base) | |
166 | { | |
167 | #ifdef CONFIG_HIGH_RES_TIMERS | |
168 | ktime_t expires; | |
169 | ||
170 | if (!new_base->cpu_base->hres_active) | |
171 | return 0; | |
172 | ||
173 | expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset); | |
174 | return expires.tv64 <= new_base->cpu_base->expires_next.tv64; | |
175 | #else | |
176 | return 0; | |
177 | #endif | |
178 | } | |
179 | ||
bc7a34b8 TG |
180 | #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON) |
181 | static inline | |
182 | struct hrtimer_cpu_base *get_target_base(struct hrtimer_cpu_base *base, | |
183 | int pinned) | |
184 | { | |
185 | if (pinned || !base->migration_enabled) | |
186 | return this_cpu_ptr(&hrtimer_bases); | |
187 | return &per_cpu(hrtimer_bases, get_nohz_timer_target()); | |
188 | } | |
189 | #else | |
190 | static inline | |
191 | struct hrtimer_cpu_base *get_target_base(struct hrtimer_cpu_base *base, | |
192 | int pinned) | |
193 | { | |
194 | return this_cpu_ptr(&hrtimer_bases); | |
195 | } | |
196 | #endif | |
197 | ||
c0a31329 TG |
198 | /* |
199 | * Switch the timer base to the current CPU when possible. | |
200 | */ | |
3c8aa39d | 201 | static inline struct hrtimer_clock_base * |
597d0275 AB |
202 | switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base, |
203 | int pinned) | |
c0a31329 | 204 | { |
bc7a34b8 | 205 | struct hrtimer_cpu_base *new_cpu_base, *this_base; |
3c8aa39d | 206 | struct hrtimer_clock_base *new_base; |
ab8177bc | 207 | int basenum = base->index; |
c0a31329 | 208 | |
bc7a34b8 TG |
209 | this_base = this_cpu_ptr(&hrtimer_bases); |
210 | new_cpu_base = get_target_base(this_base, pinned); | |
eea08f32 | 211 | again: |
e06383db | 212 | new_base = &new_cpu_base->clock_base[basenum]; |
c0a31329 TG |
213 | |
214 | if (base != new_base) { | |
215 | /* | |
6ff7041d | 216 | * We are trying to move timer to new_base. |
c0a31329 TG |
217 | * However we can't change timer's base while it is running, |
218 | * so we keep it on the same CPU. No hassle vs. reprogramming | |
219 | * the event source in the high resolution case. The softirq | |
220 | * code will take care of this when the timer function has | |
221 | * completed. There is no conflict as we hold the lock until | |
222 | * the timer is enqueued. | |
223 | */ | |
54cdfdb4 | 224 | if (unlikely(hrtimer_callback_running(timer))) |
c0a31329 TG |
225 | return base; |
226 | ||
887d9dc9 PZ |
227 | /* See the comment in lock_hrtimer_base() */ |
228 | timer->base = &migration_base; | |
ecb49d1a TG |
229 | raw_spin_unlock(&base->cpu_base->lock); |
230 | raw_spin_lock(&new_base->cpu_base->lock); | |
eea08f32 | 231 | |
bc7a34b8 TG |
232 | if (new_cpu_base != this_base && |
233 | hrtimer_check_target(timer, new_base)) { | |
ecb49d1a TG |
234 | raw_spin_unlock(&new_base->cpu_base->lock); |
235 | raw_spin_lock(&base->cpu_base->lock); | |
bc7a34b8 | 236 | new_cpu_base = this_base; |
6ff7041d TG |
237 | timer->base = base; |
238 | goto again; | |
eea08f32 | 239 | } |
c0a31329 | 240 | timer->base = new_base; |
012a45e3 | 241 | } else { |
bc7a34b8 TG |
242 | if (new_cpu_base != this_base && |
243 | hrtimer_check_target(timer, new_base)) { | |
244 | new_cpu_base = this_base; | |
012a45e3 LM |
245 | goto again; |
246 | } | |
c0a31329 TG |
247 | } |
248 | return new_base; | |
249 | } | |
250 | ||
251 | #else /* CONFIG_SMP */ | |
252 | ||
3c8aa39d | 253 | static inline struct hrtimer_clock_base * |
c0a31329 TG |
254 | lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) |
255 | { | |
3c8aa39d | 256 | struct hrtimer_clock_base *base = timer->base; |
c0a31329 | 257 | |
ecb49d1a | 258 | raw_spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
259 | |
260 | return base; | |
261 | } | |
262 | ||
eea08f32 | 263 | # define switch_hrtimer_base(t, b, p) (b) |
c0a31329 TG |
264 | |
265 | #endif /* !CONFIG_SMP */ | |
266 | ||
267 | /* | |
268 | * Functions for the union type storage format of ktime_t which are | |
269 | * too large for inlining: | |
270 | */ | |
271 | #if BITS_PER_LONG < 64 | |
c0a31329 TG |
272 | /* |
273 | * Divide a ktime value by a nanosecond value | |
274 | */ | |
8b618628 | 275 | u64 __ktime_divns(const ktime_t kt, s64 div) |
c0a31329 | 276 | { |
900cfa46 | 277 | u64 dclc; |
c0a31329 TG |
278 | int sft = 0; |
279 | ||
900cfa46 | 280 | dclc = ktime_to_ns(kt); |
c0a31329 TG |
281 | /* Make sure the divisor is less than 2^32: */ |
282 | while (div >> 32) { | |
283 | sft++; | |
284 | div >>= 1; | |
285 | } | |
286 | dclc >>= sft; | |
287 | do_div(dclc, (unsigned long) div); | |
288 | ||
4d672e7a | 289 | return dclc; |
c0a31329 | 290 | } |
8b618628 | 291 | EXPORT_SYMBOL_GPL(__ktime_divns); |
c0a31329 TG |
292 | #endif /* BITS_PER_LONG >= 64 */ |
293 | ||
5a7780e7 TG |
294 | /* |
295 | * Add two ktime values and do a safety check for overflow: | |
296 | */ | |
297 | ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs) | |
298 | { | |
299 | ktime_t res = ktime_add(lhs, rhs); | |
300 | ||
301 | /* | |
302 | * We use KTIME_SEC_MAX here, the maximum timeout which we can | |
303 | * return to user space in a timespec: | |
304 | */ | |
305 | if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64) | |
306 | res = ktime_set(KTIME_SEC_MAX, 0); | |
307 | ||
308 | return res; | |
309 | } | |
310 | ||
8daa21e6 AB |
311 | EXPORT_SYMBOL_GPL(ktime_add_safe); |
312 | ||
237fc6e7 TG |
313 | #ifdef CONFIG_DEBUG_OBJECTS_TIMERS |
314 | ||
315 | static struct debug_obj_descr hrtimer_debug_descr; | |
316 | ||
99777288 SG |
317 | static void *hrtimer_debug_hint(void *addr) |
318 | { | |
319 | return ((struct hrtimer *) addr)->function; | |
320 | } | |
321 | ||
237fc6e7 TG |
322 | /* |
323 | * fixup_init is called when: | |
324 | * - an active object is initialized | |
325 | */ | |
326 | static int hrtimer_fixup_init(void *addr, enum debug_obj_state state) | |
327 | { | |
328 | struct hrtimer *timer = addr; | |
329 | ||
330 | switch (state) { | |
331 | case ODEBUG_STATE_ACTIVE: | |
332 | hrtimer_cancel(timer); | |
333 | debug_object_init(timer, &hrtimer_debug_descr); | |
334 | return 1; | |
335 | default: | |
336 | return 0; | |
337 | } | |
338 | } | |
339 | ||
340 | /* | |
341 | * fixup_activate is called when: | |
342 | * - an active object is activated | |
343 | * - an unknown object is activated (might be a statically initialized object) | |
344 | */ | |
345 | static int hrtimer_fixup_activate(void *addr, enum debug_obj_state state) | |
346 | { | |
347 | switch (state) { | |
348 | ||
349 | case ODEBUG_STATE_NOTAVAILABLE: | |
350 | WARN_ON_ONCE(1); | |
351 | return 0; | |
352 | ||
353 | case ODEBUG_STATE_ACTIVE: | |
354 | WARN_ON(1); | |
355 | ||
356 | default: | |
357 | return 0; | |
358 | } | |
359 | } | |
360 | ||
361 | /* | |
362 | * fixup_free is called when: | |
363 | * - an active object is freed | |
364 | */ | |
365 | static int hrtimer_fixup_free(void *addr, enum debug_obj_state state) | |
366 | { | |
367 | struct hrtimer *timer = addr; | |
368 | ||
369 | switch (state) { | |
370 | case ODEBUG_STATE_ACTIVE: | |
371 | hrtimer_cancel(timer); | |
372 | debug_object_free(timer, &hrtimer_debug_descr); | |
373 | return 1; | |
374 | default: | |
375 | return 0; | |
376 | } | |
377 | } | |
378 | ||
379 | static struct debug_obj_descr hrtimer_debug_descr = { | |
380 | .name = "hrtimer", | |
99777288 | 381 | .debug_hint = hrtimer_debug_hint, |
237fc6e7 TG |
382 | .fixup_init = hrtimer_fixup_init, |
383 | .fixup_activate = hrtimer_fixup_activate, | |
384 | .fixup_free = hrtimer_fixup_free, | |
385 | }; | |
386 | ||
387 | static inline void debug_hrtimer_init(struct hrtimer *timer) | |
388 | { | |
389 | debug_object_init(timer, &hrtimer_debug_descr); | |
390 | } | |
391 | ||
392 | static inline void debug_hrtimer_activate(struct hrtimer *timer) | |
393 | { | |
394 | debug_object_activate(timer, &hrtimer_debug_descr); | |
395 | } | |
396 | ||
397 | static inline void debug_hrtimer_deactivate(struct hrtimer *timer) | |
398 | { | |
399 | debug_object_deactivate(timer, &hrtimer_debug_descr); | |
400 | } | |
401 | ||
402 | static inline void debug_hrtimer_free(struct hrtimer *timer) | |
403 | { | |
404 | debug_object_free(timer, &hrtimer_debug_descr); | |
405 | } | |
406 | ||
407 | static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id, | |
408 | enum hrtimer_mode mode); | |
409 | ||
410 | void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id, | |
411 | enum hrtimer_mode mode) | |
412 | { | |
413 | debug_object_init_on_stack(timer, &hrtimer_debug_descr); | |
414 | __hrtimer_init(timer, clock_id, mode); | |
415 | } | |
2bc481cf | 416 | EXPORT_SYMBOL_GPL(hrtimer_init_on_stack); |
237fc6e7 TG |
417 | |
418 | void destroy_hrtimer_on_stack(struct hrtimer *timer) | |
419 | { | |
420 | debug_object_free(timer, &hrtimer_debug_descr); | |
421 | } | |
422 | ||
423 | #else | |
424 | static inline void debug_hrtimer_init(struct hrtimer *timer) { } | |
425 | static inline void debug_hrtimer_activate(struct hrtimer *timer) { } | |
426 | static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { } | |
427 | #endif | |
428 | ||
c6a2a177 XG |
429 | static inline void |
430 | debug_init(struct hrtimer *timer, clockid_t clockid, | |
431 | enum hrtimer_mode mode) | |
432 | { | |
433 | debug_hrtimer_init(timer); | |
434 | trace_hrtimer_init(timer, clockid, mode); | |
435 | } | |
436 | ||
437 | static inline void debug_activate(struct hrtimer *timer) | |
438 | { | |
439 | debug_hrtimer_activate(timer); | |
440 | trace_hrtimer_start(timer); | |
441 | } | |
442 | ||
443 | static inline void debug_deactivate(struct hrtimer *timer) | |
444 | { | |
445 | debug_hrtimer_deactivate(timer); | |
446 | trace_hrtimer_cancel(timer); | |
447 | } | |
448 | ||
9bc74919 | 449 | #if defined(CONFIG_NO_HZ_COMMON) || defined(CONFIG_HIGH_RES_TIMERS) |
895bdfa7 TG |
450 | static inline void hrtimer_update_next_timer(struct hrtimer_cpu_base *cpu_base, |
451 | struct hrtimer *timer) | |
452 | { | |
453 | #ifdef CONFIG_HIGH_RES_TIMERS | |
454 | cpu_base->next_timer = timer; | |
455 | #endif | |
456 | } | |
457 | ||
4ebbda52 | 458 | static ktime_t __hrtimer_get_next_event(struct hrtimer_cpu_base *cpu_base) |
9bc74919 TG |
459 | { |
460 | struct hrtimer_clock_base *base = cpu_base->clock_base; | |
461 | ktime_t expires, expires_next = { .tv64 = KTIME_MAX }; | |
34aee88a | 462 | unsigned int active = cpu_base->active_bases; |
9bc74919 | 463 | |
895bdfa7 | 464 | hrtimer_update_next_timer(cpu_base, NULL); |
34aee88a | 465 | for (; active; base++, active >>= 1) { |
9bc74919 TG |
466 | struct timerqueue_node *next; |
467 | struct hrtimer *timer; | |
468 | ||
34aee88a | 469 | if (!(active & 0x01)) |
9bc74919 TG |
470 | continue; |
471 | ||
34aee88a | 472 | next = timerqueue_getnext(&base->active); |
9bc74919 TG |
473 | timer = container_of(next, struct hrtimer, node); |
474 | expires = ktime_sub(hrtimer_get_expires(timer), base->offset); | |
895bdfa7 | 475 | if (expires.tv64 < expires_next.tv64) { |
9bc74919 | 476 | expires_next = expires; |
895bdfa7 TG |
477 | hrtimer_update_next_timer(cpu_base, timer); |
478 | } | |
9bc74919 TG |
479 | } |
480 | /* | |
481 | * clock_was_set() might have changed base->offset of any of | |
482 | * the clock bases so the result might be negative. Fix it up | |
483 | * to prevent a false positive in clockevents_program_event(). | |
484 | */ | |
485 | if (expires_next.tv64 < 0) | |
486 | expires_next.tv64 = 0; | |
487 | return expires_next; | |
488 | } | |
489 | #endif | |
490 | ||
21d6d52a TG |
491 | static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base) |
492 | { | |
493 | ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset; | |
494 | ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset; | |
495 | ktime_t *offs_tai = &base->clock_base[HRTIMER_BASE_TAI].offset; | |
496 | ||
868a3e91 TG |
497 | return ktime_get_update_offsets_now(&base->clock_was_set_seq, |
498 | offs_real, offs_boot, offs_tai); | |
21d6d52a TG |
499 | } |
500 | ||
54cdfdb4 TG |
501 | /* High resolution timer related functions */ |
502 | #ifdef CONFIG_HIGH_RES_TIMERS | |
503 | ||
504 | /* | |
505 | * High resolution timer enabled ? | |
506 | */ | |
507 | static int hrtimer_hres_enabled __read_mostly = 1; | |
398ca17f TG |
508 | unsigned int hrtimer_resolution __read_mostly = LOW_RES_NSEC; |
509 | EXPORT_SYMBOL_GPL(hrtimer_resolution); | |
54cdfdb4 TG |
510 | |
511 | /* | |
512 | * Enable / Disable high resolution mode | |
513 | */ | |
514 | static int __init setup_hrtimer_hres(char *str) | |
515 | { | |
516 | if (!strcmp(str, "off")) | |
517 | hrtimer_hres_enabled = 0; | |
518 | else if (!strcmp(str, "on")) | |
519 | hrtimer_hres_enabled = 1; | |
520 | else | |
521 | return 0; | |
522 | return 1; | |
523 | } | |
524 | ||
525 | __setup("highres=", setup_hrtimer_hres); | |
526 | ||
527 | /* | |
528 | * hrtimer_high_res_enabled - query, if the highres mode is enabled | |
529 | */ | |
530 | static inline int hrtimer_is_hres_enabled(void) | |
531 | { | |
532 | return hrtimer_hres_enabled; | |
533 | } | |
534 | ||
535 | /* | |
536 | * Is the high resolution mode active ? | |
537 | */ | |
e19ffe8b TG |
538 | static inline int __hrtimer_hres_active(struct hrtimer_cpu_base *cpu_base) |
539 | { | |
540 | return cpu_base->hres_active; | |
541 | } | |
542 | ||
54cdfdb4 TG |
543 | static inline int hrtimer_hres_active(void) |
544 | { | |
e19ffe8b | 545 | return __hrtimer_hres_active(this_cpu_ptr(&hrtimer_bases)); |
54cdfdb4 TG |
546 | } |
547 | ||
548 | /* | |
549 | * Reprogram the event source with checking both queues for the | |
550 | * next event | |
551 | * Called with interrupts disabled and base->lock held | |
552 | */ | |
7403f41f AC |
553 | static void |
554 | hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal) | |
54cdfdb4 | 555 | { |
21d6d52a TG |
556 | ktime_t expires_next; |
557 | ||
558 | if (!cpu_base->hres_active) | |
559 | return; | |
560 | ||
561 | expires_next = __hrtimer_get_next_event(cpu_base); | |
54cdfdb4 | 562 | |
7403f41f AC |
563 | if (skip_equal && expires_next.tv64 == cpu_base->expires_next.tv64) |
564 | return; | |
565 | ||
566 | cpu_base->expires_next.tv64 = expires_next.tv64; | |
567 | ||
6c6c0d5a SH |
568 | /* |
569 | * If a hang was detected in the last timer interrupt then we | |
570 | * leave the hang delay active in the hardware. We want the | |
571 | * system to make progress. That also prevents the following | |
572 | * scenario: | |
573 | * T1 expires 50ms from now | |
574 | * T2 expires 5s from now | |
575 | * | |
576 | * T1 is removed, so this code is called and would reprogram | |
577 | * the hardware to 5s from now. Any hrtimer_start after that | |
578 | * will not reprogram the hardware due to hang_detected being | |
579 | * set. So we'd effectivly block all timers until the T2 event | |
580 | * fires. | |
581 | */ | |
582 | if (cpu_base->hang_detected) | |
583 | return; | |
584 | ||
d2540875 | 585 | tick_program_event(cpu_base->expires_next, 1); |
54cdfdb4 TG |
586 | } |
587 | ||
588 | /* | |
54cdfdb4 TG |
589 | * When a timer is enqueued and expires earlier than the already enqueued |
590 | * timers, we have to check, whether it expires earlier than the timer for | |
591 | * which the clock event device was armed. | |
592 | * | |
593 | * Called with interrupts disabled and base->cpu_base.lock held | |
594 | */ | |
c6eb3f70 TG |
595 | static void hrtimer_reprogram(struct hrtimer *timer, |
596 | struct hrtimer_clock_base *base) | |
54cdfdb4 | 597 | { |
dc5df73b | 598 | struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases); |
cc584b21 | 599 | ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset); |
54cdfdb4 | 600 | |
cc584b21 | 601 | WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0); |
63070a79 | 602 | |
54cdfdb4 | 603 | /* |
c6eb3f70 TG |
604 | * If the timer is not on the current cpu, we cannot reprogram |
605 | * the other cpus clock event device. | |
54cdfdb4 | 606 | */ |
c6eb3f70 TG |
607 | if (base->cpu_base != cpu_base) |
608 | return; | |
609 | ||
610 | /* | |
611 | * If the hrtimer interrupt is running, then it will | |
612 | * reevaluate the clock bases and reprogram the clock event | |
613 | * device. The callbacks are always executed in hard interrupt | |
614 | * context so we don't need an extra check for a running | |
615 | * callback. | |
616 | */ | |
617 | if (cpu_base->in_hrtirq) | |
618 | return; | |
54cdfdb4 | 619 | |
63070a79 TG |
620 | /* |
621 | * CLOCK_REALTIME timer might be requested with an absolute | |
c6eb3f70 | 622 | * expiry time which is less than base->offset. Set it to 0. |
63070a79 TG |
623 | */ |
624 | if (expires.tv64 < 0) | |
c6eb3f70 | 625 | expires.tv64 = 0; |
63070a79 | 626 | |
41d2e494 | 627 | if (expires.tv64 >= cpu_base->expires_next.tv64) |
c6eb3f70 | 628 | return; |
9bc74919 | 629 | |
c6eb3f70 | 630 | /* Update the pointer to the next expiring timer */ |
895bdfa7 TG |
631 | cpu_base->next_timer = timer; |
632 | ||
41d2e494 TG |
633 | /* |
634 | * If a hang was detected in the last timer interrupt then we | |
635 | * do not schedule a timer which is earlier than the expiry | |
636 | * which we enforced in the hang detection. We want the system | |
637 | * to make progress. | |
638 | */ | |
639 | if (cpu_base->hang_detected) | |
c6eb3f70 | 640 | return; |
54cdfdb4 TG |
641 | |
642 | /* | |
c6eb3f70 TG |
643 | * Program the timer hardware. We enforce the expiry for |
644 | * events which are already in the past. | |
54cdfdb4 | 645 | */ |
c6eb3f70 TG |
646 | cpu_base->expires_next = expires; |
647 | tick_program_event(expires, 1); | |
54cdfdb4 TG |
648 | } |
649 | ||
54cdfdb4 TG |
650 | /* |
651 | * Initialize the high resolution related parts of cpu_base | |
652 | */ | |
653 | static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) | |
654 | { | |
655 | base->expires_next.tv64 = KTIME_MAX; | |
656 | base->hres_active = 0; | |
54cdfdb4 TG |
657 | } |
658 | ||
9ec26907 TG |
659 | /* |
660 | * Retrigger next event is called after clock was set | |
661 | * | |
662 | * Called with interrupts disabled via on_each_cpu() | |
663 | */ | |
664 | static void retrigger_next_event(void *arg) | |
665 | { | |
dc5df73b | 666 | struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases); |
9ec26907 | 667 | |
e19ffe8b | 668 | if (!base->hres_active) |
9ec26907 TG |
669 | return; |
670 | ||
9ec26907 | 671 | raw_spin_lock(&base->lock); |
5baefd6d | 672 | hrtimer_update_base(base); |
9ec26907 TG |
673 | hrtimer_force_reprogram(base, 0); |
674 | raw_spin_unlock(&base->lock); | |
675 | } | |
b12a03ce | 676 | |
54cdfdb4 TG |
677 | /* |
678 | * Switch to high resolution mode | |
679 | */ | |
f8953856 | 680 | static int hrtimer_switch_to_hres(void) |
54cdfdb4 | 681 | { |
c6eb3f70 | 682 | struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases); |
54cdfdb4 TG |
683 | |
684 | if (tick_init_highres()) { | |
820de5c3 | 685 | printk(KERN_WARNING "Could not switch to high resolution " |
c6eb3f70 | 686 | "mode on CPU %d\n", base->cpu); |
f8953856 | 687 | return 0; |
54cdfdb4 TG |
688 | } |
689 | base->hres_active = 1; | |
398ca17f | 690 | hrtimer_resolution = HIGH_RES_NSEC; |
54cdfdb4 TG |
691 | |
692 | tick_setup_sched_timer(); | |
54cdfdb4 TG |
693 | /* "Retrigger" the interrupt to get things going */ |
694 | retrigger_next_event(NULL); | |
f8953856 | 695 | return 1; |
54cdfdb4 TG |
696 | } |
697 | ||
5ec2481b TG |
698 | static void clock_was_set_work(struct work_struct *work) |
699 | { | |
700 | clock_was_set(); | |
701 | } | |
702 | ||
703 | static DECLARE_WORK(hrtimer_work, clock_was_set_work); | |
704 | ||
f55a6faa | 705 | /* |
5ec2481b TG |
706 | * Called from timekeeping and resume code to reprogramm the hrtimer |
707 | * interrupt device on all cpus. | |
f55a6faa JS |
708 | */ |
709 | void clock_was_set_delayed(void) | |
710 | { | |
5ec2481b | 711 | schedule_work(&hrtimer_work); |
f55a6faa JS |
712 | } |
713 | ||
54cdfdb4 TG |
714 | #else |
715 | ||
e19ffe8b | 716 | static inline int __hrtimer_hres_active(struct hrtimer_cpu_base *b) { return 0; } |
54cdfdb4 TG |
717 | static inline int hrtimer_hres_active(void) { return 0; } |
718 | static inline int hrtimer_is_hres_enabled(void) { return 0; } | |
f8953856 | 719 | static inline int hrtimer_switch_to_hres(void) { return 0; } |
7403f41f AC |
720 | static inline void |
721 | hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { } | |
9e1e01dd VK |
722 | static inline int hrtimer_reprogram(struct hrtimer *timer, |
723 | struct hrtimer_clock_base *base) | |
54cdfdb4 TG |
724 | { |
725 | return 0; | |
726 | } | |
54cdfdb4 | 727 | static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { } |
9ec26907 | 728 | static inline void retrigger_next_event(void *arg) { } |
54cdfdb4 TG |
729 | |
730 | #endif /* CONFIG_HIGH_RES_TIMERS */ | |
731 | ||
b12a03ce TG |
732 | /* |
733 | * Clock realtime was set | |
734 | * | |
735 | * Change the offset of the realtime clock vs. the monotonic | |
736 | * clock. | |
737 | * | |
738 | * We might have to reprogram the high resolution timer interrupt. On | |
739 | * SMP we call the architecture specific code to retrigger _all_ high | |
740 | * resolution timer interrupts. On UP we just disable interrupts and | |
741 | * call the high resolution interrupt code. | |
742 | */ | |
743 | void clock_was_set(void) | |
744 | { | |
90ff1f30 | 745 | #ifdef CONFIG_HIGH_RES_TIMERS |
b12a03ce TG |
746 | /* Retrigger the CPU local events everywhere */ |
747 | on_each_cpu(retrigger_next_event, NULL, 1); | |
9ec26907 TG |
748 | #endif |
749 | timerfd_clock_was_set(); | |
b12a03ce TG |
750 | } |
751 | ||
752 | /* | |
753 | * During resume we might have to reprogram the high resolution timer | |
7c4c3a0f DV |
754 | * interrupt on all online CPUs. However, all other CPUs will be |
755 | * stopped with IRQs interrupts disabled so the clock_was_set() call | |
5ec2481b | 756 | * must be deferred. |
b12a03ce TG |
757 | */ |
758 | void hrtimers_resume(void) | |
759 | { | |
760 | WARN_ONCE(!irqs_disabled(), | |
761 | KERN_INFO "hrtimers_resume() called with IRQs enabled!"); | |
762 | ||
5ec2481b | 763 | /* Retrigger on the local CPU */ |
b12a03ce | 764 | retrigger_next_event(NULL); |
5ec2481b TG |
765 | /* And schedule a retrigger for all others */ |
766 | clock_was_set_delayed(); | |
b12a03ce TG |
767 | } |
768 | ||
5f201907 | 769 | static inline void timer_stats_hrtimer_set_start_info(struct hrtimer *timer) |
82f67cd9 | 770 | { |
5f201907 | 771 | #ifdef CONFIG_TIMER_STATS |
82f67cd9 IM |
772 | if (timer->start_site) |
773 | return; | |
5f201907 | 774 | timer->start_site = __builtin_return_address(0); |
82f67cd9 IM |
775 | memcpy(timer->start_comm, current->comm, TASK_COMM_LEN); |
776 | timer->start_pid = current->pid; | |
5f201907 HC |
777 | #endif |
778 | } | |
779 | ||
780 | static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer *timer) | |
781 | { | |
782 | #ifdef CONFIG_TIMER_STATS | |
783 | timer->start_site = NULL; | |
784 | #endif | |
82f67cd9 | 785 | } |
5f201907 HC |
786 | |
787 | static inline void timer_stats_account_hrtimer(struct hrtimer *timer) | |
788 | { | |
789 | #ifdef CONFIG_TIMER_STATS | |
790 | if (likely(!timer_stats_active)) | |
791 | return; | |
792 | timer_stats_update_stats(timer, timer->start_pid, timer->start_site, | |
793 | timer->function, timer->start_comm, 0); | |
82f67cd9 | 794 | #endif |
5f201907 | 795 | } |
82f67cd9 | 796 | |
c0a31329 | 797 | /* |
6506f2aa | 798 | * Counterpart to lock_hrtimer_base above: |
c0a31329 TG |
799 | */ |
800 | static inline | |
801 | void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) | |
802 | { | |
ecb49d1a | 803 | raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags); |
c0a31329 TG |
804 | } |
805 | ||
806 | /** | |
807 | * hrtimer_forward - forward the timer expiry | |
c0a31329 | 808 | * @timer: hrtimer to forward |
44f21475 | 809 | * @now: forward past this time |
c0a31329 TG |
810 | * @interval: the interval to forward |
811 | * | |
812 | * Forward the timer expiry so it will expire in the future. | |
8dca6f33 | 813 | * Returns the number of overruns. |
91e5a217 TG |
814 | * |
815 | * Can be safely called from the callback function of @timer. If | |
816 | * called from other contexts @timer must neither be enqueued nor | |
817 | * running the callback and the caller needs to take care of | |
818 | * serialization. | |
819 | * | |
820 | * Note: This only updates the timer expiry value and does not requeue | |
821 | * the timer. | |
c0a31329 | 822 | */ |
4d672e7a | 823 | u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval) |
c0a31329 | 824 | { |
4d672e7a | 825 | u64 orun = 1; |
44f21475 | 826 | ktime_t delta; |
c0a31329 | 827 | |
cc584b21 | 828 | delta = ktime_sub(now, hrtimer_get_expires(timer)); |
c0a31329 TG |
829 | |
830 | if (delta.tv64 < 0) | |
831 | return 0; | |
832 | ||
5de2755c PZ |
833 | if (WARN_ON(timer->state & HRTIMER_STATE_ENQUEUED)) |
834 | return 0; | |
835 | ||
398ca17f TG |
836 | if (interval.tv64 < hrtimer_resolution) |
837 | interval.tv64 = hrtimer_resolution; | |
c9db4fa1 | 838 | |
c0a31329 | 839 | if (unlikely(delta.tv64 >= interval.tv64)) { |
df869b63 | 840 | s64 incr = ktime_to_ns(interval); |
c0a31329 TG |
841 | |
842 | orun = ktime_divns(delta, incr); | |
cc584b21 AV |
843 | hrtimer_add_expires_ns(timer, incr * orun); |
844 | if (hrtimer_get_expires_tv64(timer) > now.tv64) | |
c0a31329 TG |
845 | return orun; |
846 | /* | |
847 | * This (and the ktime_add() below) is the | |
848 | * correction for exact: | |
849 | */ | |
850 | orun++; | |
851 | } | |
cc584b21 | 852 | hrtimer_add_expires(timer, interval); |
c0a31329 TG |
853 | |
854 | return orun; | |
855 | } | |
6bdb6b62 | 856 | EXPORT_SYMBOL_GPL(hrtimer_forward); |
c0a31329 TG |
857 | |
858 | /* | |
859 | * enqueue_hrtimer - internal function to (re)start a timer | |
860 | * | |
861 | * The timer is inserted in expiry order. Insertion into the | |
862 | * red black tree is O(log(n)). Must hold the base lock. | |
a6037b61 PZ |
863 | * |
864 | * Returns 1 when the new timer is the leftmost timer in the tree. | |
c0a31329 | 865 | */ |
a6037b61 PZ |
866 | static int enqueue_hrtimer(struct hrtimer *timer, |
867 | struct hrtimer_clock_base *base) | |
c0a31329 | 868 | { |
c6a2a177 | 869 | debug_activate(timer); |
237fc6e7 | 870 | |
ab8177bc | 871 | base->cpu_base->active_bases |= 1 << base->index; |
54cdfdb4 | 872 | |
887d9dc9 | 873 | timer->state = HRTIMER_STATE_ENQUEUED; |
a6037b61 | 874 | |
b97f44c9 | 875 | return timerqueue_add(&base->active, &timer->node); |
288867ec | 876 | } |
c0a31329 TG |
877 | |
878 | /* | |
879 | * __remove_hrtimer - internal function to remove a timer | |
880 | * | |
881 | * Caller must hold the base lock. | |
54cdfdb4 TG |
882 | * |
883 | * High resolution timer mode reprograms the clock event device when the | |
884 | * timer is the one which expires next. The caller can disable this by setting | |
885 | * reprogram to zero. This is useful, when the context does a reprogramming | |
886 | * anyway (e.g. timer interrupt) | |
c0a31329 | 887 | */ |
3c8aa39d | 888 | static void __remove_hrtimer(struct hrtimer *timer, |
303e967f | 889 | struct hrtimer_clock_base *base, |
54cdfdb4 | 890 | unsigned long newstate, int reprogram) |
c0a31329 | 891 | { |
e19ffe8b | 892 | struct hrtimer_cpu_base *cpu_base = base->cpu_base; |
895bdfa7 | 893 | unsigned int state = timer->state; |
e19ffe8b | 894 | |
895bdfa7 TG |
895 | timer->state = newstate; |
896 | if (!(state & HRTIMER_STATE_ENQUEUED)) | |
897 | return; | |
7403f41f | 898 | |
b97f44c9 | 899 | if (!timerqueue_del(&base->active, &timer->node)) |
e19ffe8b | 900 | cpu_base->active_bases &= ~(1 << base->index); |
d9f0acde | 901 | |
7403f41f | 902 | #ifdef CONFIG_HIGH_RES_TIMERS |
895bdfa7 TG |
903 | /* |
904 | * Note: If reprogram is false we do not update | |
905 | * cpu_base->next_timer. This happens when we remove the first | |
906 | * timer on a remote cpu. No harm as we never dereference | |
907 | * cpu_base->next_timer. So the worst thing what can happen is | |
908 | * an superflous call to hrtimer_force_reprogram() on the | |
909 | * remote cpu later on if the same timer gets enqueued again. | |
910 | */ | |
911 | if (reprogram && timer == cpu_base->next_timer) | |
912 | hrtimer_force_reprogram(cpu_base, 1); | |
7403f41f | 913 | #endif |
c0a31329 TG |
914 | } |
915 | ||
916 | /* | |
917 | * remove hrtimer, called with base lock held | |
918 | */ | |
919 | static inline int | |
8edfb036 | 920 | remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base, bool restart) |
c0a31329 | 921 | { |
303e967f | 922 | if (hrtimer_is_queued(timer)) { |
8edfb036 | 923 | unsigned long state = timer->state; |
54cdfdb4 TG |
924 | int reprogram; |
925 | ||
926 | /* | |
927 | * Remove the timer and force reprogramming when high | |
928 | * resolution mode is active and the timer is on the current | |
929 | * CPU. If we remove a timer on another CPU, reprogramming is | |
930 | * skipped. The interrupt event on this CPU is fired and | |
931 | * reprogramming happens in the interrupt handler. This is a | |
932 | * rare case and less expensive than a smp call. | |
933 | */ | |
c6a2a177 | 934 | debug_deactivate(timer); |
82f67cd9 | 935 | timer_stats_hrtimer_clear_start_info(timer); |
dc5df73b | 936 | reprogram = base->cpu_base == this_cpu_ptr(&hrtimer_bases); |
8edfb036 | 937 | |
887d9dc9 PZ |
938 | if (!restart) |
939 | state = HRTIMER_STATE_INACTIVE; | |
940 | ||
f13d4f97 | 941 | __remove_hrtimer(timer, base, state, reprogram); |
c0a31329 TG |
942 | return 1; |
943 | } | |
944 | return 0; | |
945 | } | |
946 | ||
58f1f803 TG |
947 | /** |
948 | * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU | |
949 | * @timer: the timer to be added | |
950 | * @tim: expiry time | |
951 | * @delta_ns: "slack" range for the timer | |
952 | * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or | |
953 | * relative (HRTIMER_MODE_REL) | |
58f1f803 | 954 | */ |
61699e13 TG |
955 | void hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, |
956 | unsigned long delta_ns, const enum hrtimer_mode mode) | |
c0a31329 | 957 | { |
3c8aa39d | 958 | struct hrtimer_clock_base *base, *new_base; |
c0a31329 | 959 | unsigned long flags; |
61699e13 | 960 | int leftmost; |
c0a31329 TG |
961 | |
962 | base = lock_hrtimer_base(timer, &flags); | |
963 | ||
964 | /* Remove an active timer from the queue: */ | |
8edfb036 | 965 | remove_hrtimer(timer, base, true); |
c0a31329 | 966 | |
597d0275 | 967 | if (mode & HRTIMER_MODE_REL) { |
84ea7fe3 | 968 | tim = ktime_add_safe(tim, base->get_time()); |
06027bdd IM |
969 | /* |
970 | * CONFIG_TIME_LOW_RES is a temporary way for architectures | |
971 | * to signal that they simply return xtime in | |
972 | * do_gettimeoffset(). In this case we want to round up by | |
973 | * resolution when starting a relative timer, to avoid short | |
974 | * timeouts. This will go away with the GTOD framework. | |
975 | */ | |
976 | #ifdef CONFIG_TIME_LOW_RES | |
398ca17f | 977 | tim = ktime_add_safe(tim, ktime_set(0, hrtimer_resolution)); |
06027bdd IM |
978 | #endif |
979 | } | |
237fc6e7 | 980 | |
da8f2e17 | 981 | hrtimer_set_expires_range_ns(timer, tim, delta_ns); |
c0a31329 | 982 | |
84ea7fe3 VK |
983 | /* Switch the timer base, if necessary: */ |
984 | new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED); | |
985 | ||
82f67cd9 IM |
986 | timer_stats_hrtimer_set_start_info(timer); |
987 | ||
a6037b61 | 988 | leftmost = enqueue_hrtimer(timer, new_base); |
61699e13 TG |
989 | if (!leftmost) |
990 | goto unlock; | |
49a2a075 VK |
991 | |
992 | if (!hrtimer_is_hres_active(timer)) { | |
993 | /* | |
994 | * Kick to reschedule the next tick to handle the new timer | |
995 | * on dynticks target. | |
996 | */ | |
997 | wake_up_nohz_cpu(new_base->cpu_base->cpu); | |
c6eb3f70 TG |
998 | } else { |
999 | hrtimer_reprogram(timer, new_base); | |
b22affe0 | 1000 | } |
61699e13 | 1001 | unlock: |
c0a31329 | 1002 | unlock_hrtimer_base(timer, &flags); |
c0a31329 | 1003 | } |
da8f2e17 AV |
1004 | EXPORT_SYMBOL_GPL(hrtimer_start_range_ns); |
1005 | ||
c0a31329 TG |
1006 | /** |
1007 | * hrtimer_try_to_cancel - try to deactivate a timer | |
c0a31329 TG |
1008 | * @timer: hrtimer to stop |
1009 | * | |
1010 | * Returns: | |
1011 | * 0 when the timer was not active | |
1012 | * 1 when the timer was active | |
1013 | * -1 when the timer is currently excuting the callback function and | |
fa9799e3 | 1014 | * cannot be stopped |
c0a31329 TG |
1015 | */ |
1016 | int hrtimer_try_to_cancel(struct hrtimer *timer) | |
1017 | { | |
3c8aa39d | 1018 | struct hrtimer_clock_base *base; |
c0a31329 TG |
1019 | unsigned long flags; |
1020 | int ret = -1; | |
1021 | ||
19d9f422 TG |
1022 | /* |
1023 | * Check lockless first. If the timer is not active (neither | |
1024 | * enqueued nor running the callback, nothing to do here. The | |
1025 | * base lock does not serialize against a concurrent enqueue, | |
1026 | * so we can avoid taking it. | |
1027 | */ | |
1028 | if (!hrtimer_active(timer)) | |
1029 | return 0; | |
1030 | ||
c0a31329 TG |
1031 | base = lock_hrtimer_base(timer, &flags); |
1032 | ||
303e967f | 1033 | if (!hrtimer_callback_running(timer)) |
8edfb036 | 1034 | ret = remove_hrtimer(timer, base, false); |
c0a31329 TG |
1035 | |
1036 | unlock_hrtimer_base(timer, &flags); | |
1037 | ||
1038 | return ret; | |
1039 | ||
1040 | } | |
8d16b764 | 1041 | EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel); |
c0a31329 TG |
1042 | |
1043 | /** | |
1044 | * hrtimer_cancel - cancel a timer and wait for the handler to finish. | |
c0a31329 TG |
1045 | * @timer: the timer to be cancelled |
1046 | * | |
1047 | * Returns: | |
1048 | * 0 when the timer was not active | |
1049 | * 1 when the timer was active | |
1050 | */ | |
1051 | int hrtimer_cancel(struct hrtimer *timer) | |
1052 | { | |
1053 | for (;;) { | |
1054 | int ret = hrtimer_try_to_cancel(timer); | |
1055 | ||
1056 | if (ret >= 0) | |
1057 | return ret; | |
5ef37b19 | 1058 | cpu_relax(); |
c0a31329 TG |
1059 | } |
1060 | } | |
8d16b764 | 1061 | EXPORT_SYMBOL_GPL(hrtimer_cancel); |
c0a31329 TG |
1062 | |
1063 | /** | |
1064 | * hrtimer_get_remaining - get remaining time for the timer | |
c0a31329 TG |
1065 | * @timer: the timer to read |
1066 | */ | |
1067 | ktime_t hrtimer_get_remaining(const struct hrtimer *timer) | |
1068 | { | |
c0a31329 TG |
1069 | unsigned long flags; |
1070 | ktime_t rem; | |
1071 | ||
b3bd3de6 | 1072 | lock_hrtimer_base(timer, &flags); |
cc584b21 | 1073 | rem = hrtimer_expires_remaining(timer); |
c0a31329 TG |
1074 | unlock_hrtimer_base(timer, &flags); |
1075 | ||
1076 | return rem; | |
1077 | } | |
8d16b764 | 1078 | EXPORT_SYMBOL_GPL(hrtimer_get_remaining); |
c0a31329 | 1079 | |
3451d024 | 1080 | #ifdef CONFIG_NO_HZ_COMMON |
69239749 TL |
1081 | /** |
1082 | * hrtimer_get_next_event - get the time until next expiry event | |
1083 | * | |
c1ad348b | 1084 | * Returns the next expiry time or KTIME_MAX if no timer is pending. |
69239749 | 1085 | */ |
c1ad348b | 1086 | u64 hrtimer_get_next_event(void) |
69239749 | 1087 | { |
dc5df73b | 1088 | struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases); |
c1ad348b | 1089 | u64 expires = KTIME_MAX; |
69239749 | 1090 | unsigned long flags; |
69239749 | 1091 | |
ecb49d1a | 1092 | raw_spin_lock_irqsave(&cpu_base->lock, flags); |
3c8aa39d | 1093 | |
e19ffe8b | 1094 | if (!__hrtimer_hres_active(cpu_base)) |
c1ad348b | 1095 | expires = __hrtimer_get_next_event(cpu_base).tv64; |
3c8aa39d | 1096 | |
ecb49d1a | 1097 | raw_spin_unlock_irqrestore(&cpu_base->lock, flags); |
3c8aa39d | 1098 | |
c1ad348b | 1099 | return expires; |
69239749 TL |
1100 | } |
1101 | #endif | |
1102 | ||
237fc6e7 TG |
1103 | static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id, |
1104 | enum hrtimer_mode mode) | |
c0a31329 | 1105 | { |
3c8aa39d | 1106 | struct hrtimer_cpu_base *cpu_base; |
e06383db | 1107 | int base; |
c0a31329 | 1108 | |
7978672c GA |
1109 | memset(timer, 0, sizeof(struct hrtimer)); |
1110 | ||
22127e93 | 1111 | cpu_base = raw_cpu_ptr(&hrtimer_bases); |
c0a31329 | 1112 | |
c9cb2e3d | 1113 | if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS) |
7978672c GA |
1114 | clock_id = CLOCK_MONOTONIC; |
1115 | ||
e06383db JS |
1116 | base = hrtimer_clockid_to_base(clock_id); |
1117 | timer->base = &cpu_base->clock_base[base]; | |
998adc3d | 1118 | timerqueue_init(&timer->node); |
82f67cd9 IM |
1119 | |
1120 | #ifdef CONFIG_TIMER_STATS | |
1121 | timer->start_site = NULL; | |
1122 | timer->start_pid = -1; | |
1123 | memset(timer->start_comm, 0, TASK_COMM_LEN); | |
1124 | #endif | |
c0a31329 | 1125 | } |
237fc6e7 TG |
1126 | |
1127 | /** | |
1128 | * hrtimer_init - initialize a timer to the given clock | |
1129 | * @timer: the timer to be initialized | |
1130 | * @clock_id: the clock to be used | |
1131 | * @mode: timer mode abs/rel | |
1132 | */ | |
1133 | void hrtimer_init(struct hrtimer *timer, clockid_t clock_id, | |
1134 | enum hrtimer_mode mode) | |
1135 | { | |
c6a2a177 | 1136 | debug_init(timer, clock_id, mode); |
237fc6e7 TG |
1137 | __hrtimer_init(timer, clock_id, mode); |
1138 | } | |
8d16b764 | 1139 | EXPORT_SYMBOL_GPL(hrtimer_init); |
c0a31329 | 1140 | |
887d9dc9 PZ |
1141 | /* |
1142 | * A timer is active, when it is enqueued into the rbtree or the | |
1143 | * callback function is running or it's in the state of being migrated | |
1144 | * to another cpu. | |
1145 | * | |
1146 | * It is important for this function to not return a false negative. | |
1147 | */ | |
1148 | bool hrtimer_active(const struct hrtimer *timer) | |
1149 | { | |
1150 | struct hrtimer_cpu_base *cpu_base; | |
1151 | unsigned int seq; | |
1152 | ||
1153 | do { | |
1154 | cpu_base = READ_ONCE(timer->base->cpu_base); | |
1155 | seq = raw_read_seqcount_begin(&cpu_base->seq); | |
1156 | ||
1157 | if (timer->state != HRTIMER_STATE_INACTIVE || | |
1158 | cpu_base->running == timer) | |
1159 | return true; | |
1160 | ||
1161 | } while (read_seqcount_retry(&cpu_base->seq, seq) || | |
1162 | cpu_base != READ_ONCE(timer->base->cpu_base)); | |
1163 | ||
1164 | return false; | |
1165 | } | |
1166 | EXPORT_SYMBOL_GPL(hrtimer_active); | |
1167 | ||
1168 | /* | |
1169 | * The write_seqcount_barrier()s in __run_hrtimer() split the thing into 3 | |
1170 | * distinct sections: | |
1171 | * | |
1172 | * - queued: the timer is queued | |
1173 | * - callback: the timer is being ran | |
1174 | * - post: the timer is inactive or (re)queued | |
1175 | * | |
1176 | * On the read side we ensure we observe timer->state and cpu_base->running | |
1177 | * from the same section, if anything changed while we looked at it, we retry. | |
1178 | * This includes timer->base changing because sequence numbers alone are | |
1179 | * insufficient for that. | |
1180 | * | |
1181 | * The sequence numbers are required because otherwise we could still observe | |
1182 | * a false negative if the read side got smeared over multiple consequtive | |
1183 | * __run_hrtimer() invocations. | |
1184 | */ | |
1185 | ||
21d6d52a TG |
1186 | static void __run_hrtimer(struct hrtimer_cpu_base *cpu_base, |
1187 | struct hrtimer_clock_base *base, | |
1188 | struct hrtimer *timer, ktime_t *now) | |
d3d74453 | 1189 | { |
d3d74453 PZ |
1190 | enum hrtimer_restart (*fn)(struct hrtimer *); |
1191 | int restart; | |
1192 | ||
887d9dc9 | 1193 | lockdep_assert_held(&cpu_base->lock); |
ca109491 | 1194 | |
c6a2a177 | 1195 | debug_deactivate(timer); |
887d9dc9 PZ |
1196 | cpu_base->running = timer; |
1197 | ||
1198 | /* | |
1199 | * Separate the ->running assignment from the ->state assignment. | |
1200 | * | |
1201 | * As with a regular write barrier, this ensures the read side in | |
1202 | * hrtimer_active() cannot observe cpu_base->running == NULL && | |
1203 | * timer->state == INACTIVE. | |
1204 | */ | |
1205 | raw_write_seqcount_barrier(&cpu_base->seq); | |
1206 | ||
1207 | __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE, 0); | |
d3d74453 | 1208 | timer_stats_account_hrtimer(timer); |
d3d74453 | 1209 | fn = timer->function; |
ca109491 PZ |
1210 | |
1211 | /* | |
1212 | * Because we run timers from hardirq context, there is no chance | |
1213 | * they get migrated to another cpu, therefore its safe to unlock | |
1214 | * the timer base. | |
1215 | */ | |
ecb49d1a | 1216 | raw_spin_unlock(&cpu_base->lock); |
c6a2a177 | 1217 | trace_hrtimer_expire_entry(timer, now); |
ca109491 | 1218 | restart = fn(timer); |
c6a2a177 | 1219 | trace_hrtimer_expire_exit(timer); |
ecb49d1a | 1220 | raw_spin_lock(&cpu_base->lock); |
d3d74453 PZ |
1221 | |
1222 | /* | |
887d9dc9 | 1223 | * Note: We clear the running state after enqueue_hrtimer and |
e3f1d883 TG |
1224 | * we do not reprogramm the event hardware. Happens either in |
1225 | * hrtimer_start_range_ns() or in hrtimer_interrupt() | |
5de2755c PZ |
1226 | * |
1227 | * Note: Because we dropped the cpu_base->lock above, | |
1228 | * hrtimer_start_range_ns() can have popped in and enqueued the timer | |
1229 | * for us already. | |
d3d74453 | 1230 | */ |
5de2755c PZ |
1231 | if (restart != HRTIMER_NORESTART && |
1232 | !(timer->state & HRTIMER_STATE_ENQUEUED)) | |
a6037b61 | 1233 | enqueue_hrtimer(timer, base); |
f13d4f97 | 1234 | |
887d9dc9 PZ |
1235 | /* |
1236 | * Separate the ->running assignment from the ->state assignment. | |
1237 | * | |
1238 | * As with a regular write barrier, this ensures the read side in | |
1239 | * hrtimer_active() cannot observe cpu_base->running == NULL && | |
1240 | * timer->state == INACTIVE. | |
1241 | */ | |
1242 | raw_write_seqcount_barrier(&cpu_base->seq); | |
f13d4f97 | 1243 | |
887d9dc9 PZ |
1244 | WARN_ON_ONCE(cpu_base->running != timer); |
1245 | cpu_base->running = NULL; | |
d3d74453 PZ |
1246 | } |
1247 | ||
21d6d52a | 1248 | static void __hrtimer_run_queues(struct hrtimer_cpu_base *cpu_base, ktime_t now) |
54cdfdb4 | 1249 | { |
34aee88a TG |
1250 | struct hrtimer_clock_base *base = cpu_base->clock_base; |
1251 | unsigned int active = cpu_base->active_bases; | |
6ff7041d | 1252 | |
34aee88a | 1253 | for (; active; base++, active >>= 1) { |
998adc3d | 1254 | struct timerqueue_node *node; |
ab8177bc TG |
1255 | ktime_t basenow; |
1256 | ||
34aee88a | 1257 | if (!(active & 0x01)) |
ab8177bc | 1258 | continue; |
54cdfdb4 | 1259 | |
54cdfdb4 TG |
1260 | basenow = ktime_add(now, base->offset); |
1261 | ||
998adc3d | 1262 | while ((node = timerqueue_getnext(&base->active))) { |
54cdfdb4 TG |
1263 | struct hrtimer *timer; |
1264 | ||
998adc3d | 1265 | timer = container_of(node, struct hrtimer, node); |
54cdfdb4 | 1266 | |
654c8e0b AV |
1267 | /* |
1268 | * The immediate goal for using the softexpires is | |
1269 | * minimizing wakeups, not running timers at the | |
1270 | * earliest interrupt after their soft expiration. | |
1271 | * This allows us to avoid using a Priority Search | |
1272 | * Tree, which can answer a stabbing querry for | |
1273 | * overlapping intervals and instead use the simple | |
1274 | * BST we already have. | |
1275 | * We don't add extra wakeups by delaying timers that | |
1276 | * are right-of a not yet expired timer, because that | |
1277 | * timer will have to trigger a wakeup anyway. | |
1278 | */ | |
9bc74919 | 1279 | if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer)) |
54cdfdb4 | 1280 | break; |
54cdfdb4 | 1281 | |
21d6d52a | 1282 | __run_hrtimer(cpu_base, base, timer, &basenow); |
54cdfdb4 | 1283 | } |
54cdfdb4 | 1284 | } |
21d6d52a TG |
1285 | } |
1286 | ||
1287 | #ifdef CONFIG_HIGH_RES_TIMERS | |
1288 | ||
1289 | /* | |
1290 | * High resolution timer interrupt | |
1291 | * Called with interrupts disabled | |
1292 | */ | |
1293 | void hrtimer_interrupt(struct clock_event_device *dev) | |
1294 | { | |
1295 | struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases); | |
1296 | ktime_t expires_next, now, entry_time, delta; | |
1297 | int retries = 0; | |
1298 | ||
1299 | BUG_ON(!cpu_base->hres_active); | |
1300 | cpu_base->nr_events++; | |
1301 | dev->next_event.tv64 = KTIME_MAX; | |
1302 | ||
1303 | raw_spin_lock(&cpu_base->lock); | |
1304 | entry_time = now = hrtimer_update_base(cpu_base); | |
1305 | retry: | |
1306 | cpu_base->in_hrtirq = 1; | |
1307 | /* | |
1308 | * We set expires_next to KTIME_MAX here with cpu_base->lock | |
1309 | * held to prevent that a timer is enqueued in our queue via | |
1310 | * the migration code. This does not affect enqueueing of | |
1311 | * timers which run their callback and need to be requeued on | |
1312 | * this CPU. | |
1313 | */ | |
1314 | cpu_base->expires_next.tv64 = KTIME_MAX; | |
1315 | ||
1316 | __hrtimer_run_queues(cpu_base, now); | |
1317 | ||
9bc74919 TG |
1318 | /* Reevaluate the clock bases for the next expiry */ |
1319 | expires_next = __hrtimer_get_next_event(cpu_base); | |
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; |
9bc74919 | 1325 | cpu_base->in_hrtirq = 0; |
ecb49d1a | 1326 | raw_spin_unlock(&cpu_base->lock); |
54cdfdb4 TG |
1327 | |
1328 | /* Reprogramming necessary ? */ | |
d2540875 | 1329 | if (!tick_program_event(expires_next, 0)) { |
41d2e494 TG |
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 | 1361 | delta = ktime_sub(now, entry_time); |
a6ffebce TG |
1362 | if ((unsigned int)delta.tv64 > cpu_base->max_hang_time) |
1363 | cpu_base->max_hang_time = (unsigned int) delta.tv64; | |
41d2e494 TG |
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 | */ | |
c6eb3f70 | 1381 | static inline void __hrtimer_peek_ahead_timers(void) |
8bdec955 TG |
1382 | { |
1383 | struct tick_device *td; | |
1384 | ||
1385 | if (!hrtimer_hres_active()) | |
1386 | return; | |
1387 | ||
22127e93 | 1388 | td = this_cpu_ptr(&tick_cpu_device); |
8bdec955 TG |
1389 | if (td && td->evtdev) |
1390 | hrtimer_interrupt(td->evtdev); | |
1391 | } | |
1392 | ||
82c5b7b5 IM |
1393 | #else /* CONFIG_HIGH_RES_TIMERS */ |
1394 | ||
1395 | static inline void __hrtimer_peek_ahead_timers(void) { } | |
1396 | ||
1397 | #endif /* !CONFIG_HIGH_RES_TIMERS */ | |
82f67cd9 | 1398 | |
d3d74453 | 1399 | /* |
c6eb3f70 | 1400 | * Called from run_local_timers in hardirq context every jiffy |
c0a31329 | 1401 | */ |
833883d9 | 1402 | void hrtimer_run_queues(void) |
c0a31329 | 1403 | { |
dc5df73b | 1404 | struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases); |
21d6d52a | 1405 | ktime_t now; |
c0a31329 | 1406 | |
e19ffe8b | 1407 | if (__hrtimer_hres_active(cpu_base)) |
3055adda DS |
1408 | return; |
1409 | ||
c6eb3f70 TG |
1410 | /* |
1411 | * This _is_ ugly: We have to check periodically, whether we | |
1412 | * can switch to highres and / or nohz mode. The clocksource | |
1413 | * switch happens with xtime_lock held. Notification from | |
1414 | * there only sets the check bit in the tick_oneshot code, | |
1415 | * otherwise we might deadlock vs. xtime_lock. | |
1416 | */ | |
1417 | if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) { | |
1418 | hrtimer_switch_to_hres(); | |
1419 | return; | |
1420 | } | |
1421 | ||
21d6d52a TG |
1422 | raw_spin_lock(&cpu_base->lock); |
1423 | now = hrtimer_update_base(cpu_base); | |
1424 | __hrtimer_run_queues(cpu_base, now); | |
1425 | raw_spin_unlock(&cpu_base->lock); | |
c0a31329 TG |
1426 | } |
1427 | ||
10c94ec1 TG |
1428 | /* |
1429 | * Sleep related functions: | |
1430 | */ | |
c9cb2e3d | 1431 | static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer) |
00362e33 TG |
1432 | { |
1433 | struct hrtimer_sleeper *t = | |
1434 | container_of(timer, struct hrtimer_sleeper, timer); | |
1435 | struct task_struct *task = t->task; | |
1436 | ||
1437 | t->task = NULL; | |
1438 | if (task) | |
1439 | wake_up_process(task); | |
1440 | ||
1441 | return HRTIMER_NORESTART; | |
1442 | } | |
1443 | ||
36c8b586 | 1444 | void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task) |
00362e33 TG |
1445 | { |
1446 | sl->timer.function = hrtimer_wakeup; | |
1447 | sl->task = task; | |
1448 | } | |
2bc481cf | 1449 | EXPORT_SYMBOL_GPL(hrtimer_init_sleeper); |
00362e33 | 1450 | |
669d7868 | 1451 | static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode) |
432569bb | 1452 | { |
669d7868 | 1453 | hrtimer_init_sleeper(t, current); |
10c94ec1 | 1454 | |
432569bb RZ |
1455 | do { |
1456 | set_current_state(TASK_INTERRUPTIBLE); | |
cc584b21 | 1457 | hrtimer_start_expires(&t->timer, mode); |
432569bb | 1458 | |
54cdfdb4 | 1459 | if (likely(t->task)) |
b0f8c44f | 1460 | freezable_schedule(); |
432569bb | 1461 | |
669d7868 | 1462 | hrtimer_cancel(&t->timer); |
c9cb2e3d | 1463 | mode = HRTIMER_MODE_ABS; |
669d7868 TG |
1464 | |
1465 | } while (t->task && !signal_pending(current)); | |
432569bb | 1466 | |
3588a085 PZ |
1467 | __set_current_state(TASK_RUNNING); |
1468 | ||
669d7868 | 1469 | return t->task == NULL; |
10c94ec1 TG |
1470 | } |
1471 | ||
080344b9 ON |
1472 | static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp) |
1473 | { | |
1474 | struct timespec rmt; | |
1475 | ktime_t rem; | |
1476 | ||
cc584b21 | 1477 | rem = hrtimer_expires_remaining(timer); |
080344b9 ON |
1478 | if (rem.tv64 <= 0) |
1479 | return 0; | |
1480 | rmt = ktime_to_timespec(rem); | |
1481 | ||
1482 | if (copy_to_user(rmtp, &rmt, sizeof(*rmtp))) | |
1483 | return -EFAULT; | |
1484 | ||
1485 | return 1; | |
1486 | } | |
1487 | ||
1711ef38 | 1488 | long __sched hrtimer_nanosleep_restart(struct restart_block *restart) |
10c94ec1 | 1489 | { |
669d7868 | 1490 | struct hrtimer_sleeper t; |
080344b9 | 1491 | struct timespec __user *rmtp; |
237fc6e7 | 1492 | int ret = 0; |
10c94ec1 | 1493 | |
ab8177bc | 1494 | hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid, |
237fc6e7 | 1495 | HRTIMER_MODE_ABS); |
cc584b21 | 1496 | hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires); |
10c94ec1 | 1497 | |
c9cb2e3d | 1498 | if (do_nanosleep(&t, HRTIMER_MODE_ABS)) |
237fc6e7 | 1499 | goto out; |
10c94ec1 | 1500 | |
029a07e0 | 1501 | rmtp = restart->nanosleep.rmtp; |
432569bb | 1502 | if (rmtp) { |
237fc6e7 | 1503 | ret = update_rmtp(&t.timer, rmtp); |
080344b9 | 1504 | if (ret <= 0) |
237fc6e7 | 1505 | goto out; |
432569bb | 1506 | } |
10c94ec1 | 1507 | |
10c94ec1 | 1508 | /* The other values in restart are already filled in */ |
237fc6e7 TG |
1509 | ret = -ERESTART_RESTARTBLOCK; |
1510 | out: | |
1511 | destroy_hrtimer_on_stack(&t.timer); | |
1512 | return ret; | |
10c94ec1 TG |
1513 | } |
1514 | ||
080344b9 | 1515 | long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp, |
10c94ec1 TG |
1516 | const enum hrtimer_mode mode, const clockid_t clockid) |
1517 | { | |
1518 | struct restart_block *restart; | |
669d7868 | 1519 | struct hrtimer_sleeper t; |
237fc6e7 | 1520 | int ret = 0; |
3bd01206 AV |
1521 | unsigned long slack; |
1522 | ||
1523 | slack = current->timer_slack_ns; | |
aab03e05 | 1524 | if (dl_task(current) || rt_task(current)) |
3bd01206 | 1525 | slack = 0; |
10c94ec1 | 1526 | |
237fc6e7 | 1527 | hrtimer_init_on_stack(&t.timer, clockid, mode); |
3bd01206 | 1528 | hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack); |
432569bb | 1529 | if (do_nanosleep(&t, mode)) |
237fc6e7 | 1530 | goto out; |
10c94ec1 | 1531 | |
7978672c | 1532 | /* Absolute timers do not update the rmtp value and restart: */ |
237fc6e7 TG |
1533 | if (mode == HRTIMER_MODE_ABS) { |
1534 | ret = -ERESTARTNOHAND; | |
1535 | goto out; | |
1536 | } | |
10c94ec1 | 1537 | |
432569bb | 1538 | if (rmtp) { |
237fc6e7 | 1539 | ret = update_rmtp(&t.timer, rmtp); |
080344b9 | 1540 | if (ret <= 0) |
237fc6e7 | 1541 | goto out; |
432569bb | 1542 | } |
10c94ec1 | 1543 | |
f56141e3 | 1544 | restart = ¤t->restart_block; |
1711ef38 | 1545 | restart->fn = hrtimer_nanosleep_restart; |
ab8177bc | 1546 | restart->nanosleep.clockid = t.timer.base->clockid; |
029a07e0 | 1547 | restart->nanosleep.rmtp = rmtp; |
cc584b21 | 1548 | restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer); |
10c94ec1 | 1549 | |
237fc6e7 TG |
1550 | ret = -ERESTART_RESTARTBLOCK; |
1551 | out: | |
1552 | destroy_hrtimer_on_stack(&t.timer); | |
1553 | return ret; | |
10c94ec1 TG |
1554 | } |
1555 | ||
58fd3aa2 HC |
1556 | SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp, |
1557 | struct timespec __user *, rmtp) | |
6ba1b912 | 1558 | { |
080344b9 | 1559 | struct timespec tu; |
6ba1b912 TG |
1560 | |
1561 | if (copy_from_user(&tu, rqtp, sizeof(tu))) | |
1562 | return -EFAULT; | |
1563 | ||
1564 | if (!timespec_valid(&tu)) | |
1565 | return -EINVAL; | |
1566 | ||
080344b9 | 1567 | return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC); |
6ba1b912 TG |
1568 | } |
1569 | ||
c0a31329 TG |
1570 | /* |
1571 | * Functions related to boot-time initialization: | |
1572 | */ | |
0db0628d | 1573 | static void init_hrtimers_cpu(int cpu) |
c0a31329 | 1574 | { |
3c8aa39d | 1575 | struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu); |
c0a31329 TG |
1576 | int i; |
1577 | ||
998adc3d | 1578 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
3c8aa39d | 1579 | cpu_base->clock_base[i].cpu_base = cpu_base; |
998adc3d JS |
1580 | timerqueue_init_head(&cpu_base->clock_base[i].active); |
1581 | } | |
3c8aa39d | 1582 | |
cddd0248 | 1583 | cpu_base->cpu = cpu; |
54cdfdb4 | 1584 | hrtimer_init_hres(cpu_base); |
c0a31329 TG |
1585 | } |
1586 | ||
1587 | #ifdef CONFIG_HOTPLUG_CPU | |
1588 | ||
ca109491 | 1589 | static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base, |
37810659 | 1590 | struct hrtimer_clock_base *new_base) |
c0a31329 TG |
1591 | { |
1592 | struct hrtimer *timer; | |
998adc3d | 1593 | struct timerqueue_node *node; |
c0a31329 | 1594 | |
998adc3d JS |
1595 | while ((node = timerqueue_getnext(&old_base->active))) { |
1596 | timer = container_of(node, struct hrtimer, node); | |
54cdfdb4 | 1597 | BUG_ON(hrtimer_callback_running(timer)); |
c6a2a177 | 1598 | debug_deactivate(timer); |
b00c1a99 TG |
1599 | |
1600 | /* | |
c04dca02 | 1601 | * Mark it as ENQUEUED not INACTIVE otherwise the |
b00c1a99 TG |
1602 | * timer could be seen as !active and just vanish away |
1603 | * under us on another CPU | |
1604 | */ | |
c04dca02 | 1605 | __remove_hrtimer(timer, old_base, HRTIMER_STATE_ENQUEUED, 0); |
c0a31329 | 1606 | timer->base = new_base; |
54cdfdb4 | 1607 | /* |
e3f1d883 TG |
1608 | * Enqueue the timers on the new cpu. This does not |
1609 | * reprogram the event device in case the timer | |
1610 | * expires before the earliest on this CPU, but we run | |
1611 | * hrtimer_interrupt after we migrated everything to | |
1612 | * sort out already expired timers and reprogram the | |
1613 | * event device. | |
54cdfdb4 | 1614 | */ |
a6037b61 | 1615 | enqueue_hrtimer(timer, new_base); |
c0a31329 TG |
1616 | } |
1617 | } | |
1618 | ||
d5fd43c4 | 1619 | static void migrate_hrtimers(int scpu) |
c0a31329 | 1620 | { |
3c8aa39d | 1621 | struct hrtimer_cpu_base *old_base, *new_base; |
731a55ba | 1622 | int i; |
c0a31329 | 1623 | |
37810659 | 1624 | BUG_ON(cpu_online(scpu)); |
37810659 | 1625 | tick_cancel_sched_timer(scpu); |
731a55ba TG |
1626 | |
1627 | local_irq_disable(); | |
1628 | old_base = &per_cpu(hrtimer_bases, scpu); | |
dc5df73b | 1629 | new_base = this_cpu_ptr(&hrtimer_bases); |
d82f0b0f ON |
1630 | /* |
1631 | * The caller is globally serialized and nobody else | |
1632 | * takes two locks at once, deadlock is not possible. | |
1633 | */ | |
ecb49d1a TG |
1634 | raw_spin_lock(&new_base->lock); |
1635 | raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING); | |
c0a31329 | 1636 | |
3c8aa39d | 1637 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
ca109491 | 1638 | migrate_hrtimer_list(&old_base->clock_base[i], |
37810659 | 1639 | &new_base->clock_base[i]); |
c0a31329 TG |
1640 | } |
1641 | ||
ecb49d1a TG |
1642 | raw_spin_unlock(&old_base->lock); |
1643 | raw_spin_unlock(&new_base->lock); | |
37810659 | 1644 | |
731a55ba TG |
1645 | /* Check, if we got expired work to do */ |
1646 | __hrtimer_peek_ahead_timers(); | |
1647 | local_irq_enable(); | |
c0a31329 | 1648 | } |
37810659 | 1649 | |
c0a31329 TG |
1650 | #endif /* CONFIG_HOTPLUG_CPU */ |
1651 | ||
0db0628d | 1652 | static int hrtimer_cpu_notify(struct notifier_block *self, |
c0a31329 TG |
1653 | unsigned long action, void *hcpu) |
1654 | { | |
b2e3c0ad | 1655 | int scpu = (long)hcpu; |
c0a31329 TG |
1656 | |
1657 | switch (action) { | |
1658 | ||
1659 | case CPU_UP_PREPARE: | |
8bb78442 | 1660 | case CPU_UP_PREPARE_FROZEN: |
37810659 | 1661 | init_hrtimers_cpu(scpu); |
c0a31329 TG |
1662 | break; |
1663 | ||
1664 | #ifdef CONFIG_HOTPLUG_CPU | |
1665 | case CPU_DEAD: | |
8bb78442 | 1666 | case CPU_DEAD_FROZEN: |
d5fd43c4 | 1667 | migrate_hrtimers(scpu); |
c0a31329 TG |
1668 | break; |
1669 | #endif | |
1670 | ||
1671 | default: | |
1672 | break; | |
1673 | } | |
1674 | ||
1675 | return NOTIFY_OK; | |
1676 | } | |
1677 | ||
0db0628d | 1678 | static struct notifier_block hrtimers_nb = { |
c0a31329 TG |
1679 | .notifier_call = hrtimer_cpu_notify, |
1680 | }; | |
1681 | ||
1682 | void __init hrtimers_init(void) | |
1683 | { | |
1684 | hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE, | |
1685 | (void *)(long)smp_processor_id()); | |
1686 | register_cpu_notifier(&hrtimers_nb); | |
1687 | } | |
1688 | ||
7bb67439 | 1689 | /** |
351b3f7a | 1690 | * schedule_hrtimeout_range_clock - sleep until timeout |
7bb67439 | 1691 | * @expires: timeout value (ktime_t) |
654c8e0b | 1692 | * @delta: slack in expires timeout (ktime_t) |
7bb67439 | 1693 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL |
351b3f7a | 1694 | * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME |
7bb67439 | 1695 | */ |
351b3f7a CE |
1696 | int __sched |
1697 | schedule_hrtimeout_range_clock(ktime_t *expires, unsigned long delta, | |
1698 | const enum hrtimer_mode mode, int clock) | |
7bb67439 AV |
1699 | { |
1700 | struct hrtimer_sleeper t; | |
1701 | ||
1702 | /* | |
1703 | * Optimize when a zero timeout value is given. It does not | |
1704 | * matter whether this is an absolute or a relative time. | |
1705 | */ | |
1706 | if (expires && !expires->tv64) { | |
1707 | __set_current_state(TASK_RUNNING); | |
1708 | return 0; | |
1709 | } | |
1710 | ||
1711 | /* | |
43b21013 | 1712 | * A NULL parameter means "infinite" |
7bb67439 AV |
1713 | */ |
1714 | if (!expires) { | |
1715 | schedule(); | |
7bb67439 AV |
1716 | return -EINTR; |
1717 | } | |
1718 | ||
351b3f7a | 1719 | hrtimer_init_on_stack(&t.timer, clock, mode); |
654c8e0b | 1720 | hrtimer_set_expires_range_ns(&t.timer, *expires, delta); |
7bb67439 AV |
1721 | |
1722 | hrtimer_init_sleeper(&t, current); | |
1723 | ||
cc584b21 | 1724 | hrtimer_start_expires(&t.timer, mode); |
7bb67439 AV |
1725 | |
1726 | if (likely(t.task)) | |
1727 | schedule(); | |
1728 | ||
1729 | hrtimer_cancel(&t.timer); | |
1730 | destroy_hrtimer_on_stack(&t.timer); | |
1731 | ||
1732 | __set_current_state(TASK_RUNNING); | |
1733 | ||
1734 | return !t.task ? 0 : -EINTR; | |
1735 | } | |
351b3f7a CE |
1736 | |
1737 | /** | |
1738 | * schedule_hrtimeout_range - sleep until timeout | |
1739 | * @expires: timeout value (ktime_t) | |
1740 | * @delta: slack in expires timeout (ktime_t) | |
1741 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL | |
1742 | * | |
1743 | * Make the current task sleep until the given expiry time has | |
1744 | * elapsed. The routine will return immediately unless | |
1745 | * the current task state has been set (see set_current_state()). | |
1746 | * | |
1747 | * The @delta argument gives the kernel the freedom to schedule the | |
1748 | * actual wakeup to a time that is both power and performance friendly. | |
1749 | * The kernel give the normal best effort behavior for "@expires+@delta", | |
1750 | * but may decide to fire the timer earlier, but no earlier than @expires. | |
1751 | * | |
1752 | * You can set the task state as follows - | |
1753 | * | |
1754 | * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to | |
1755 | * pass before the routine returns. | |
1756 | * | |
1757 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
1758 | * delivered to the current task. | |
1759 | * | |
1760 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1761 | * routine returns. | |
1762 | * | |
1763 | * Returns 0 when the timer has expired otherwise -EINTR | |
1764 | */ | |
1765 | int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta, | |
1766 | const enum hrtimer_mode mode) | |
1767 | { | |
1768 | return schedule_hrtimeout_range_clock(expires, delta, mode, | |
1769 | CLOCK_MONOTONIC); | |
1770 | } | |
654c8e0b AV |
1771 | EXPORT_SYMBOL_GPL(schedule_hrtimeout_range); |
1772 | ||
1773 | /** | |
1774 | * schedule_hrtimeout - sleep until timeout | |
1775 | * @expires: timeout value (ktime_t) | |
1776 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL | |
1777 | * | |
1778 | * Make the current task sleep until the given expiry time has | |
1779 | * elapsed. The routine will return immediately unless | |
1780 | * the current task state has been set (see set_current_state()). | |
1781 | * | |
1782 | * You can set the task state as follows - | |
1783 | * | |
1784 | * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to | |
1785 | * pass before the routine returns. | |
1786 | * | |
1787 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
1788 | * delivered to the current task. | |
1789 | * | |
1790 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1791 | * routine returns. | |
1792 | * | |
1793 | * Returns 0 when the timer has expired otherwise -EINTR | |
1794 | */ | |
1795 | int __sched schedule_hrtimeout(ktime_t *expires, | |
1796 | const enum hrtimer_mode mode) | |
1797 | { | |
1798 | return schedule_hrtimeout_range(expires, 0, mode); | |
1799 | } | |
7bb67439 | 1800 | EXPORT_SYMBOL_GPL(schedule_hrtimeout); |