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 | ||
c0a31329 TG |
57 | /* |
58 | * The timer bases: | |
7978672c | 59 | * |
e06383db JS |
60 | * There are more clockids then hrtimer bases. Thus, we index |
61 | * into the timer bases by the hrtimer_base_type enum. When trying | |
62 | * to reach a base using a clockid, hrtimer_clockid_to_base() | |
63 | * is used to convert from clockid to the proper hrtimer_base_type. | |
c0a31329 | 64 | */ |
54cdfdb4 | 65 | DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) = |
c0a31329 | 66 | { |
3c8aa39d | 67 | |
84cc8fd2 | 68 | .lock = __RAW_SPIN_LOCK_UNLOCKED(hrtimer_bases.lock), |
3c8aa39d | 69 | .clock_base = |
c0a31329 | 70 | { |
3c8aa39d | 71 | { |
ab8177bc TG |
72 | .index = HRTIMER_BASE_MONOTONIC, |
73 | .clockid = CLOCK_MONOTONIC, | |
3c8aa39d | 74 | .get_time = &ktime_get, |
54cdfdb4 | 75 | .resolution = KTIME_LOW_RES, |
3c8aa39d | 76 | }, |
68fa61c0 TG |
77 | { |
78 | .index = HRTIMER_BASE_REALTIME, | |
79 | .clockid = CLOCK_REALTIME, | |
80 | .get_time = &ktime_get_real, | |
81 | .resolution = KTIME_LOW_RES, | |
82 | }, | |
70a08cca | 83 | { |
ab8177bc TG |
84 | .index = HRTIMER_BASE_BOOTTIME, |
85 | .clockid = CLOCK_BOOTTIME, | |
70a08cca JS |
86 | .get_time = &ktime_get_boottime, |
87 | .resolution = KTIME_LOW_RES, | |
88 | }, | |
90adda98 JS |
89 | { |
90 | .index = HRTIMER_BASE_TAI, | |
91 | .clockid = CLOCK_TAI, | |
92 | .get_time = &ktime_get_clocktai, | |
93 | .resolution = KTIME_LOW_RES, | |
94 | }, | |
3c8aa39d | 95 | } |
c0a31329 TG |
96 | }; |
97 | ||
942c3c5c | 98 | static const int hrtimer_clock_to_base_table[MAX_CLOCKS] = { |
ce31332d TG |
99 | [CLOCK_REALTIME] = HRTIMER_BASE_REALTIME, |
100 | [CLOCK_MONOTONIC] = HRTIMER_BASE_MONOTONIC, | |
101 | [CLOCK_BOOTTIME] = HRTIMER_BASE_BOOTTIME, | |
90adda98 | 102 | [CLOCK_TAI] = HRTIMER_BASE_TAI, |
ce31332d | 103 | }; |
e06383db JS |
104 | |
105 | static inline int hrtimer_clockid_to_base(clockid_t clock_id) | |
106 | { | |
107 | return hrtimer_clock_to_base_table[clock_id]; | |
108 | } | |
109 | ||
110 | ||
92127c7a TG |
111 | /* |
112 | * Get the coarse grained time at the softirq based on xtime and | |
113 | * wall_to_monotonic. | |
114 | */ | |
3c8aa39d | 115 | static void hrtimer_get_softirq_time(struct hrtimer_cpu_base *base) |
92127c7a | 116 | { |
76f41088 JS |
117 | ktime_t xtim, mono, boot, tai; |
118 | ktime_t off_real, off_boot, off_tai; | |
92127c7a | 119 | |
76f41088 JS |
120 | mono = ktime_get_update_offsets_tick(&off_real, &off_boot, &off_tai); |
121 | boot = ktime_add(mono, off_boot); | |
122 | xtim = ktime_add(mono, off_real); | |
123 | tai = ktime_add(xtim, off_tai); | |
92127c7a | 124 | |
e06383db | 125 | base->clock_base[HRTIMER_BASE_REALTIME].softirq_time = xtim; |
70a08cca JS |
126 | base->clock_base[HRTIMER_BASE_MONOTONIC].softirq_time = mono; |
127 | base->clock_base[HRTIMER_BASE_BOOTTIME].softirq_time = boot; | |
76f41088 | 128 | base->clock_base[HRTIMER_BASE_TAI].softirq_time = tai; |
92127c7a TG |
129 | } |
130 | ||
c0a31329 TG |
131 | /* |
132 | * Functions and macros which are different for UP/SMP systems are kept in a | |
133 | * single place | |
134 | */ | |
135 | #ifdef CONFIG_SMP | |
136 | ||
c0a31329 TG |
137 | /* |
138 | * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock | |
139 | * means that all timers which are tied to this base via timer->base are | |
140 | * locked, and the base itself is locked too. | |
141 | * | |
142 | * So __run_timers/migrate_timers can safely modify all timers which could | |
143 | * be found on the lists/queues. | |
144 | * | |
145 | * When the timer's base is locked, and the timer removed from list, it is | |
146 | * possible to set timer->base = NULL and drop the lock: the timer remains | |
147 | * locked. | |
148 | */ | |
3c8aa39d TG |
149 | static |
150 | struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer, | |
151 | unsigned long *flags) | |
c0a31329 | 152 | { |
3c8aa39d | 153 | struct hrtimer_clock_base *base; |
c0a31329 TG |
154 | |
155 | for (;;) { | |
156 | base = timer->base; | |
157 | if (likely(base != NULL)) { | |
ecb49d1a | 158 | raw_spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
159 | if (likely(base == timer->base)) |
160 | return base; | |
161 | /* The timer has migrated to another CPU: */ | |
ecb49d1a | 162 | raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags); |
c0a31329 TG |
163 | } |
164 | cpu_relax(); | |
165 | } | |
166 | } | |
167 | ||
6ff7041d TG |
168 | /* |
169 | * With HIGHRES=y we do not migrate the timer when it is expiring | |
170 | * before the next event on the target cpu because we cannot reprogram | |
171 | * the target cpu hardware and we would cause it to fire late. | |
172 | * | |
173 | * Called with cpu_base->lock of target cpu held. | |
174 | */ | |
175 | static int | |
176 | hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base) | |
177 | { | |
178 | #ifdef CONFIG_HIGH_RES_TIMERS | |
179 | ktime_t expires; | |
180 | ||
181 | if (!new_base->cpu_base->hres_active) | |
182 | return 0; | |
183 | ||
184 | expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset); | |
185 | return expires.tv64 <= new_base->cpu_base->expires_next.tv64; | |
186 | #else | |
187 | return 0; | |
188 | #endif | |
189 | } | |
190 | ||
c0a31329 TG |
191 | /* |
192 | * Switch the timer base to the current CPU when possible. | |
193 | */ | |
3c8aa39d | 194 | static inline struct hrtimer_clock_base * |
597d0275 AB |
195 | switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base, |
196 | int pinned) | |
c0a31329 | 197 | { |
3c8aa39d TG |
198 | struct hrtimer_clock_base *new_base; |
199 | struct hrtimer_cpu_base *new_cpu_base; | |
6ff7041d | 200 | int this_cpu = smp_processor_id(); |
6201b4d6 | 201 | int cpu = get_nohz_timer_target(pinned); |
ab8177bc | 202 | int basenum = base->index; |
c0a31329 | 203 | |
eea08f32 AB |
204 | again: |
205 | new_cpu_base = &per_cpu(hrtimer_bases, cpu); | |
e06383db | 206 | new_base = &new_cpu_base->clock_base[basenum]; |
c0a31329 TG |
207 | |
208 | if (base != new_base) { | |
209 | /* | |
6ff7041d | 210 | * We are trying to move timer to new_base. |
c0a31329 TG |
211 | * However we can't change timer's base while it is running, |
212 | * so we keep it on the same CPU. No hassle vs. reprogramming | |
213 | * the event source in the high resolution case. The softirq | |
214 | * code will take care of this when the timer function has | |
215 | * completed. There is no conflict as we hold the lock until | |
216 | * the timer is enqueued. | |
217 | */ | |
54cdfdb4 | 218 | if (unlikely(hrtimer_callback_running(timer))) |
c0a31329 TG |
219 | return base; |
220 | ||
221 | /* See the comment in lock_timer_base() */ | |
222 | timer->base = NULL; | |
ecb49d1a TG |
223 | raw_spin_unlock(&base->cpu_base->lock); |
224 | raw_spin_lock(&new_base->cpu_base->lock); | |
eea08f32 | 225 | |
6ff7041d TG |
226 | if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) { |
227 | cpu = this_cpu; | |
ecb49d1a TG |
228 | raw_spin_unlock(&new_base->cpu_base->lock); |
229 | raw_spin_lock(&base->cpu_base->lock); | |
6ff7041d TG |
230 | timer->base = base; |
231 | goto again; | |
eea08f32 | 232 | } |
c0a31329 | 233 | timer->base = new_base; |
012a45e3 LM |
234 | } else { |
235 | if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) { | |
236 | cpu = this_cpu; | |
237 | goto again; | |
238 | } | |
c0a31329 TG |
239 | } |
240 | return new_base; | |
241 | } | |
242 | ||
243 | #else /* CONFIG_SMP */ | |
244 | ||
3c8aa39d | 245 | static inline struct hrtimer_clock_base * |
c0a31329 TG |
246 | lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) |
247 | { | |
3c8aa39d | 248 | struct hrtimer_clock_base *base = timer->base; |
c0a31329 | 249 | |
ecb49d1a | 250 | raw_spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
251 | |
252 | return base; | |
253 | } | |
254 | ||
eea08f32 | 255 | # define switch_hrtimer_base(t, b, p) (b) |
c0a31329 TG |
256 | |
257 | #endif /* !CONFIG_SMP */ | |
258 | ||
259 | /* | |
260 | * Functions for the union type storage format of ktime_t which are | |
261 | * too large for inlining: | |
262 | */ | |
263 | #if BITS_PER_LONG < 64 | |
c0a31329 TG |
264 | /* |
265 | * Divide a ktime value by a nanosecond value | |
266 | */ | |
4d672e7a | 267 | u64 ktime_divns(const ktime_t kt, s64 div) |
c0a31329 | 268 | { |
900cfa46 | 269 | u64 dclc; |
c0a31329 TG |
270 | int sft = 0; |
271 | ||
900cfa46 | 272 | dclc = ktime_to_ns(kt); |
c0a31329 TG |
273 | /* Make sure the divisor is less than 2^32: */ |
274 | while (div >> 32) { | |
275 | sft++; | |
276 | div >>= 1; | |
277 | } | |
278 | dclc >>= sft; | |
279 | do_div(dclc, (unsigned long) div); | |
280 | ||
4d672e7a | 281 | return dclc; |
c0a31329 | 282 | } |
166afb64 | 283 | EXPORT_SYMBOL_GPL(ktime_divns); |
c0a31329 TG |
284 | #endif /* BITS_PER_LONG >= 64 */ |
285 | ||
5a7780e7 TG |
286 | /* |
287 | * Add two ktime values and do a safety check for overflow: | |
288 | */ | |
289 | ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs) | |
290 | { | |
291 | ktime_t res = ktime_add(lhs, rhs); | |
292 | ||
293 | /* | |
294 | * We use KTIME_SEC_MAX here, the maximum timeout which we can | |
295 | * return to user space in a timespec: | |
296 | */ | |
297 | if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64) | |
298 | res = ktime_set(KTIME_SEC_MAX, 0); | |
299 | ||
300 | return res; | |
301 | } | |
302 | ||
8daa21e6 AB |
303 | EXPORT_SYMBOL_GPL(ktime_add_safe); |
304 | ||
237fc6e7 TG |
305 | #ifdef CONFIG_DEBUG_OBJECTS_TIMERS |
306 | ||
307 | static struct debug_obj_descr hrtimer_debug_descr; | |
308 | ||
99777288 SG |
309 | static void *hrtimer_debug_hint(void *addr) |
310 | { | |
311 | return ((struct hrtimer *) addr)->function; | |
312 | } | |
313 | ||
237fc6e7 TG |
314 | /* |
315 | * fixup_init is called when: | |
316 | * - an active object is initialized | |
317 | */ | |
318 | static int hrtimer_fixup_init(void *addr, enum debug_obj_state state) | |
319 | { | |
320 | struct hrtimer *timer = addr; | |
321 | ||
322 | switch (state) { | |
323 | case ODEBUG_STATE_ACTIVE: | |
324 | hrtimer_cancel(timer); | |
325 | debug_object_init(timer, &hrtimer_debug_descr); | |
326 | return 1; | |
327 | default: | |
328 | return 0; | |
329 | } | |
330 | } | |
331 | ||
332 | /* | |
333 | * fixup_activate is called when: | |
334 | * - an active object is activated | |
335 | * - an unknown object is activated (might be a statically initialized object) | |
336 | */ | |
337 | static int hrtimer_fixup_activate(void *addr, enum debug_obj_state state) | |
338 | { | |
339 | switch (state) { | |
340 | ||
341 | case ODEBUG_STATE_NOTAVAILABLE: | |
342 | WARN_ON_ONCE(1); | |
343 | return 0; | |
344 | ||
345 | case ODEBUG_STATE_ACTIVE: | |
346 | WARN_ON(1); | |
347 | ||
348 | default: | |
349 | return 0; | |
350 | } | |
351 | } | |
352 | ||
353 | /* | |
354 | * fixup_free is called when: | |
355 | * - an active object is freed | |
356 | */ | |
357 | static int hrtimer_fixup_free(void *addr, enum debug_obj_state state) | |
358 | { | |
359 | struct hrtimer *timer = addr; | |
360 | ||
361 | switch (state) { | |
362 | case ODEBUG_STATE_ACTIVE: | |
363 | hrtimer_cancel(timer); | |
364 | debug_object_free(timer, &hrtimer_debug_descr); | |
365 | return 1; | |
366 | default: | |
367 | return 0; | |
368 | } | |
369 | } | |
370 | ||
371 | static struct debug_obj_descr hrtimer_debug_descr = { | |
372 | .name = "hrtimer", | |
99777288 | 373 | .debug_hint = hrtimer_debug_hint, |
237fc6e7 TG |
374 | .fixup_init = hrtimer_fixup_init, |
375 | .fixup_activate = hrtimer_fixup_activate, | |
376 | .fixup_free = hrtimer_fixup_free, | |
377 | }; | |
378 | ||
379 | static inline void debug_hrtimer_init(struct hrtimer *timer) | |
380 | { | |
381 | debug_object_init(timer, &hrtimer_debug_descr); | |
382 | } | |
383 | ||
384 | static inline void debug_hrtimer_activate(struct hrtimer *timer) | |
385 | { | |
386 | debug_object_activate(timer, &hrtimer_debug_descr); | |
387 | } | |
388 | ||
389 | static inline void debug_hrtimer_deactivate(struct hrtimer *timer) | |
390 | { | |
391 | debug_object_deactivate(timer, &hrtimer_debug_descr); | |
392 | } | |
393 | ||
394 | static inline void debug_hrtimer_free(struct hrtimer *timer) | |
395 | { | |
396 | debug_object_free(timer, &hrtimer_debug_descr); | |
397 | } | |
398 | ||
399 | static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id, | |
400 | enum hrtimer_mode mode); | |
401 | ||
402 | void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id, | |
403 | enum hrtimer_mode mode) | |
404 | { | |
405 | debug_object_init_on_stack(timer, &hrtimer_debug_descr); | |
406 | __hrtimer_init(timer, clock_id, mode); | |
407 | } | |
2bc481cf | 408 | EXPORT_SYMBOL_GPL(hrtimer_init_on_stack); |
237fc6e7 TG |
409 | |
410 | void destroy_hrtimer_on_stack(struct hrtimer *timer) | |
411 | { | |
412 | debug_object_free(timer, &hrtimer_debug_descr); | |
413 | } | |
414 | ||
415 | #else | |
416 | static inline void debug_hrtimer_init(struct hrtimer *timer) { } | |
417 | static inline void debug_hrtimer_activate(struct hrtimer *timer) { } | |
418 | static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { } | |
419 | #endif | |
420 | ||
c6a2a177 XG |
421 | static inline void |
422 | debug_init(struct hrtimer *timer, clockid_t clockid, | |
423 | enum hrtimer_mode mode) | |
424 | { | |
425 | debug_hrtimer_init(timer); | |
426 | trace_hrtimer_init(timer, clockid, mode); | |
427 | } | |
428 | ||
429 | static inline void debug_activate(struct hrtimer *timer) | |
430 | { | |
431 | debug_hrtimer_activate(timer); | |
432 | trace_hrtimer_start(timer); | |
433 | } | |
434 | ||
435 | static inline void debug_deactivate(struct hrtimer *timer) | |
436 | { | |
437 | debug_hrtimer_deactivate(timer); | |
438 | trace_hrtimer_cancel(timer); | |
439 | } | |
440 | ||
54cdfdb4 TG |
441 | /* High resolution timer related functions */ |
442 | #ifdef CONFIG_HIGH_RES_TIMERS | |
443 | ||
444 | /* | |
445 | * High resolution timer enabled ? | |
446 | */ | |
447 | static int hrtimer_hres_enabled __read_mostly = 1; | |
448 | ||
449 | /* | |
450 | * Enable / Disable high resolution mode | |
451 | */ | |
452 | static int __init setup_hrtimer_hres(char *str) | |
453 | { | |
454 | if (!strcmp(str, "off")) | |
455 | hrtimer_hres_enabled = 0; | |
456 | else if (!strcmp(str, "on")) | |
457 | hrtimer_hres_enabled = 1; | |
458 | else | |
459 | return 0; | |
460 | return 1; | |
461 | } | |
462 | ||
463 | __setup("highres=", setup_hrtimer_hres); | |
464 | ||
465 | /* | |
466 | * hrtimer_high_res_enabled - query, if the highres mode is enabled | |
467 | */ | |
468 | static inline int hrtimer_is_hres_enabled(void) | |
469 | { | |
470 | return hrtimer_hres_enabled; | |
471 | } | |
472 | ||
473 | /* | |
474 | * Is the high resolution mode active ? | |
475 | */ | |
476 | static inline int hrtimer_hres_active(void) | |
477 | { | |
909ea964 | 478 | return __this_cpu_read(hrtimer_bases.hres_active); |
54cdfdb4 TG |
479 | } |
480 | ||
481 | /* | |
482 | * Reprogram the event source with checking both queues for the | |
483 | * next event | |
484 | * Called with interrupts disabled and base->lock held | |
485 | */ | |
7403f41f AC |
486 | static void |
487 | hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal) | |
54cdfdb4 TG |
488 | { |
489 | int i; | |
490 | struct hrtimer_clock_base *base = cpu_base->clock_base; | |
7403f41f | 491 | ktime_t expires, expires_next; |
54cdfdb4 | 492 | |
7403f41f | 493 | expires_next.tv64 = KTIME_MAX; |
54cdfdb4 TG |
494 | |
495 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) { | |
496 | struct hrtimer *timer; | |
998adc3d | 497 | struct timerqueue_node *next; |
54cdfdb4 | 498 | |
998adc3d JS |
499 | next = timerqueue_getnext(&base->active); |
500 | if (!next) | |
54cdfdb4 | 501 | continue; |
998adc3d JS |
502 | timer = container_of(next, struct hrtimer, node); |
503 | ||
cc584b21 | 504 | expires = ktime_sub(hrtimer_get_expires(timer), base->offset); |
b0a9b511 TG |
505 | /* |
506 | * clock_was_set() has changed base->offset so the | |
507 | * result might be negative. Fix it up to prevent a | |
508 | * false positive in clockevents_program_event() | |
509 | */ | |
510 | if (expires.tv64 < 0) | |
511 | expires.tv64 = 0; | |
7403f41f AC |
512 | if (expires.tv64 < expires_next.tv64) |
513 | expires_next = expires; | |
54cdfdb4 TG |
514 | } |
515 | ||
7403f41f AC |
516 | if (skip_equal && expires_next.tv64 == cpu_base->expires_next.tv64) |
517 | return; | |
518 | ||
519 | cpu_base->expires_next.tv64 = expires_next.tv64; | |
520 | ||
6c6c0d5a SH |
521 | /* |
522 | * If a hang was detected in the last timer interrupt then we | |
523 | * leave the hang delay active in the hardware. We want the | |
524 | * system to make progress. That also prevents the following | |
525 | * scenario: | |
526 | * T1 expires 50ms from now | |
527 | * T2 expires 5s from now | |
528 | * | |
529 | * T1 is removed, so this code is called and would reprogram | |
530 | * the hardware to 5s from now. Any hrtimer_start after that | |
531 | * will not reprogram the hardware due to hang_detected being | |
532 | * set. So we'd effectivly block all timers until the T2 event | |
533 | * fires. | |
534 | */ | |
535 | if (cpu_base->hang_detected) | |
536 | return; | |
537 | ||
54cdfdb4 TG |
538 | if (cpu_base->expires_next.tv64 != KTIME_MAX) |
539 | tick_program_event(cpu_base->expires_next, 1); | |
540 | } | |
541 | ||
542 | /* | |
543 | * Shared reprogramming for clock_realtime and clock_monotonic | |
544 | * | |
545 | * When a timer is enqueued and expires earlier than the already enqueued | |
546 | * timers, we have to check, whether it expires earlier than the timer for | |
547 | * which the clock event device was armed. | |
548 | * | |
9e1e01dd VK |
549 | * Note, that in case the state has HRTIMER_STATE_CALLBACK set, no reprogramming |
550 | * and no expiry check happens. The timer gets enqueued into the rbtree. The | |
551 | * reprogramming and expiry check is done in the hrtimer_interrupt or in the | |
552 | * softirq. | |
553 | * | |
54cdfdb4 TG |
554 | * Called with interrupts disabled and base->cpu_base.lock held |
555 | */ | |
556 | static int hrtimer_reprogram(struct hrtimer *timer, | |
557 | struct hrtimer_clock_base *base) | |
558 | { | |
41d2e494 | 559 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
cc584b21 | 560 | ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset); |
54cdfdb4 TG |
561 | int res; |
562 | ||
cc584b21 | 563 | WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0); |
63070a79 | 564 | |
54cdfdb4 TG |
565 | /* |
566 | * When the callback is running, we do not reprogram the clock event | |
567 | * device. The timer callback is either running on a different CPU or | |
3a4fa0a2 | 568 | * the callback is executed in the hrtimer_interrupt context. The |
54cdfdb4 TG |
569 | * reprogramming is handled either by the softirq, which called the |
570 | * callback or at the end of the hrtimer_interrupt. | |
571 | */ | |
572 | if (hrtimer_callback_running(timer)) | |
573 | return 0; | |
574 | ||
63070a79 TG |
575 | /* |
576 | * CLOCK_REALTIME timer might be requested with an absolute | |
577 | * expiry time which is less than base->offset. Nothing wrong | |
578 | * about that, just avoid to call into the tick code, which | |
579 | * has now objections against negative expiry values. | |
580 | */ | |
581 | if (expires.tv64 < 0) | |
582 | return -ETIME; | |
583 | ||
41d2e494 TG |
584 | if (expires.tv64 >= cpu_base->expires_next.tv64) |
585 | return 0; | |
586 | ||
587 | /* | |
588 | * If a hang was detected in the last timer interrupt then we | |
589 | * do not schedule a timer which is earlier than the expiry | |
590 | * which we enforced in the hang detection. We want the system | |
591 | * to make progress. | |
592 | */ | |
593 | if (cpu_base->hang_detected) | |
54cdfdb4 TG |
594 | return 0; |
595 | ||
596 | /* | |
597 | * Clockevents returns -ETIME, when the event was in the past. | |
598 | */ | |
599 | res = tick_program_event(expires, 0); | |
600 | if (!IS_ERR_VALUE(res)) | |
41d2e494 | 601 | cpu_base->expires_next = expires; |
54cdfdb4 TG |
602 | return res; |
603 | } | |
604 | ||
54cdfdb4 TG |
605 | /* |
606 | * Initialize the high resolution related parts of cpu_base | |
607 | */ | |
608 | static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) | |
609 | { | |
610 | base->expires_next.tv64 = KTIME_MAX; | |
611 | base->hres_active = 0; | |
54cdfdb4 TG |
612 | } |
613 | ||
5baefd6d JS |
614 | static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base) |
615 | { | |
616 | ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset; | |
617 | ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset; | |
90adda98 | 618 | ktime_t *offs_tai = &base->clock_base[HRTIMER_BASE_TAI].offset; |
5baefd6d | 619 | |
76f41088 | 620 | return ktime_get_update_offsets_now(offs_real, offs_boot, offs_tai); |
5baefd6d JS |
621 | } |
622 | ||
9ec26907 TG |
623 | /* |
624 | * Retrigger next event is called after clock was set | |
625 | * | |
626 | * Called with interrupts disabled via on_each_cpu() | |
627 | */ | |
628 | static void retrigger_next_event(void *arg) | |
629 | { | |
630 | struct hrtimer_cpu_base *base = &__get_cpu_var(hrtimer_bases); | |
9ec26907 TG |
631 | |
632 | if (!hrtimer_hres_active()) | |
633 | return; | |
634 | ||
9ec26907 | 635 | raw_spin_lock(&base->lock); |
5baefd6d | 636 | hrtimer_update_base(base); |
9ec26907 TG |
637 | hrtimer_force_reprogram(base, 0); |
638 | raw_spin_unlock(&base->lock); | |
639 | } | |
b12a03ce | 640 | |
54cdfdb4 TG |
641 | /* |
642 | * Switch to high resolution mode | |
643 | */ | |
f8953856 | 644 | static int hrtimer_switch_to_hres(void) |
54cdfdb4 | 645 | { |
b12a03ce | 646 | int i, cpu = smp_processor_id(); |
820de5c3 | 647 | struct hrtimer_cpu_base *base = &per_cpu(hrtimer_bases, cpu); |
54cdfdb4 TG |
648 | unsigned long flags; |
649 | ||
650 | if (base->hres_active) | |
f8953856 | 651 | return 1; |
54cdfdb4 TG |
652 | |
653 | local_irq_save(flags); | |
654 | ||
655 | if (tick_init_highres()) { | |
656 | local_irq_restore(flags); | |
820de5c3 IM |
657 | printk(KERN_WARNING "Could not switch to high resolution " |
658 | "mode on CPU %d\n", cpu); | |
f8953856 | 659 | return 0; |
54cdfdb4 TG |
660 | } |
661 | base->hres_active = 1; | |
b12a03ce TG |
662 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) |
663 | base->clock_base[i].resolution = KTIME_HIGH_RES; | |
54cdfdb4 TG |
664 | |
665 | tick_setup_sched_timer(); | |
54cdfdb4 TG |
666 | /* "Retrigger" the interrupt to get things going */ |
667 | retrigger_next_event(NULL); | |
668 | local_irq_restore(flags); | |
f8953856 | 669 | return 1; |
54cdfdb4 TG |
670 | } |
671 | ||
5ec2481b TG |
672 | static void clock_was_set_work(struct work_struct *work) |
673 | { | |
674 | clock_was_set(); | |
675 | } | |
676 | ||
677 | static DECLARE_WORK(hrtimer_work, clock_was_set_work); | |
678 | ||
f55a6faa | 679 | /* |
5ec2481b TG |
680 | * Called from timekeeping and resume code to reprogramm the hrtimer |
681 | * interrupt device on all cpus. | |
f55a6faa JS |
682 | */ |
683 | void clock_was_set_delayed(void) | |
684 | { | |
5ec2481b | 685 | schedule_work(&hrtimer_work); |
f55a6faa JS |
686 | } |
687 | ||
54cdfdb4 TG |
688 | #else |
689 | ||
690 | static inline int hrtimer_hres_active(void) { return 0; } | |
691 | static inline int hrtimer_is_hres_enabled(void) { return 0; } | |
f8953856 | 692 | static inline int hrtimer_switch_to_hres(void) { return 0; } |
7403f41f AC |
693 | static inline void |
694 | hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { } | |
9e1e01dd VK |
695 | static inline int hrtimer_reprogram(struct hrtimer *timer, |
696 | struct hrtimer_clock_base *base) | |
54cdfdb4 TG |
697 | { |
698 | return 0; | |
699 | } | |
54cdfdb4 | 700 | static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { } |
9ec26907 | 701 | static inline void retrigger_next_event(void *arg) { } |
54cdfdb4 TG |
702 | |
703 | #endif /* CONFIG_HIGH_RES_TIMERS */ | |
704 | ||
b12a03ce TG |
705 | /* |
706 | * Clock realtime was set | |
707 | * | |
708 | * Change the offset of the realtime clock vs. the monotonic | |
709 | * clock. | |
710 | * | |
711 | * We might have to reprogram the high resolution timer interrupt. On | |
712 | * SMP we call the architecture specific code to retrigger _all_ high | |
713 | * resolution timer interrupts. On UP we just disable interrupts and | |
714 | * call the high resolution interrupt code. | |
715 | */ | |
716 | void clock_was_set(void) | |
717 | { | |
90ff1f30 | 718 | #ifdef CONFIG_HIGH_RES_TIMERS |
b12a03ce TG |
719 | /* Retrigger the CPU local events everywhere */ |
720 | on_each_cpu(retrigger_next_event, NULL, 1); | |
9ec26907 TG |
721 | #endif |
722 | timerfd_clock_was_set(); | |
b12a03ce TG |
723 | } |
724 | ||
725 | /* | |
726 | * During resume we might have to reprogram the high resolution timer | |
7c4c3a0f DV |
727 | * interrupt on all online CPUs. However, all other CPUs will be |
728 | * stopped with IRQs interrupts disabled so the clock_was_set() call | |
5ec2481b | 729 | * must be deferred. |
b12a03ce TG |
730 | */ |
731 | void hrtimers_resume(void) | |
732 | { | |
733 | WARN_ONCE(!irqs_disabled(), | |
734 | KERN_INFO "hrtimers_resume() called with IRQs enabled!"); | |
735 | ||
5ec2481b | 736 | /* Retrigger on the local CPU */ |
b12a03ce | 737 | retrigger_next_event(NULL); |
5ec2481b TG |
738 | /* And schedule a retrigger for all others */ |
739 | clock_was_set_delayed(); | |
b12a03ce TG |
740 | } |
741 | ||
5f201907 | 742 | static inline void timer_stats_hrtimer_set_start_info(struct hrtimer *timer) |
82f67cd9 | 743 | { |
5f201907 | 744 | #ifdef CONFIG_TIMER_STATS |
82f67cd9 IM |
745 | if (timer->start_site) |
746 | return; | |
5f201907 | 747 | timer->start_site = __builtin_return_address(0); |
82f67cd9 IM |
748 | memcpy(timer->start_comm, current->comm, TASK_COMM_LEN); |
749 | timer->start_pid = current->pid; | |
5f201907 HC |
750 | #endif |
751 | } | |
752 | ||
753 | static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer *timer) | |
754 | { | |
755 | #ifdef CONFIG_TIMER_STATS | |
756 | timer->start_site = NULL; | |
757 | #endif | |
82f67cd9 | 758 | } |
5f201907 HC |
759 | |
760 | static inline void timer_stats_account_hrtimer(struct hrtimer *timer) | |
761 | { | |
762 | #ifdef CONFIG_TIMER_STATS | |
763 | if (likely(!timer_stats_active)) | |
764 | return; | |
765 | timer_stats_update_stats(timer, timer->start_pid, timer->start_site, | |
766 | timer->function, timer->start_comm, 0); | |
82f67cd9 | 767 | #endif |
5f201907 | 768 | } |
82f67cd9 | 769 | |
c0a31329 | 770 | /* |
6506f2aa | 771 | * Counterpart to lock_hrtimer_base above: |
c0a31329 TG |
772 | */ |
773 | static inline | |
774 | void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) | |
775 | { | |
ecb49d1a | 776 | raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags); |
c0a31329 TG |
777 | } |
778 | ||
779 | /** | |
780 | * hrtimer_forward - forward the timer expiry | |
c0a31329 | 781 | * @timer: hrtimer to forward |
44f21475 | 782 | * @now: forward past this time |
c0a31329 TG |
783 | * @interval: the interval to forward |
784 | * | |
785 | * Forward the timer expiry so it will expire in the future. | |
8dca6f33 | 786 | * Returns the number of overruns. |
c0a31329 | 787 | */ |
4d672e7a | 788 | u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval) |
c0a31329 | 789 | { |
4d672e7a | 790 | u64 orun = 1; |
44f21475 | 791 | ktime_t delta; |
c0a31329 | 792 | |
cc584b21 | 793 | delta = ktime_sub(now, hrtimer_get_expires(timer)); |
c0a31329 TG |
794 | |
795 | if (delta.tv64 < 0) | |
796 | return 0; | |
797 | ||
c9db4fa1 TG |
798 | if (interval.tv64 < timer->base->resolution.tv64) |
799 | interval.tv64 = timer->base->resolution.tv64; | |
800 | ||
c0a31329 | 801 | if (unlikely(delta.tv64 >= interval.tv64)) { |
df869b63 | 802 | s64 incr = ktime_to_ns(interval); |
c0a31329 TG |
803 | |
804 | orun = ktime_divns(delta, incr); | |
cc584b21 AV |
805 | hrtimer_add_expires_ns(timer, incr * orun); |
806 | if (hrtimer_get_expires_tv64(timer) > now.tv64) | |
c0a31329 TG |
807 | return orun; |
808 | /* | |
809 | * This (and the ktime_add() below) is the | |
810 | * correction for exact: | |
811 | */ | |
812 | orun++; | |
813 | } | |
cc584b21 | 814 | hrtimer_add_expires(timer, interval); |
c0a31329 TG |
815 | |
816 | return orun; | |
817 | } | |
6bdb6b62 | 818 | EXPORT_SYMBOL_GPL(hrtimer_forward); |
c0a31329 TG |
819 | |
820 | /* | |
821 | * enqueue_hrtimer - internal function to (re)start a timer | |
822 | * | |
823 | * The timer is inserted in expiry order. Insertion into the | |
824 | * red black tree is O(log(n)). Must hold the base lock. | |
a6037b61 PZ |
825 | * |
826 | * Returns 1 when the new timer is the leftmost timer in the tree. | |
c0a31329 | 827 | */ |
a6037b61 PZ |
828 | static int enqueue_hrtimer(struct hrtimer *timer, |
829 | struct hrtimer_clock_base *base) | |
c0a31329 | 830 | { |
c6a2a177 | 831 | debug_activate(timer); |
237fc6e7 | 832 | |
998adc3d | 833 | timerqueue_add(&base->active, &timer->node); |
ab8177bc | 834 | base->cpu_base->active_bases |= 1 << base->index; |
54cdfdb4 | 835 | |
303e967f TG |
836 | /* |
837 | * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the | |
838 | * state of a possibly running callback. | |
839 | */ | |
840 | timer->state |= HRTIMER_STATE_ENQUEUED; | |
a6037b61 | 841 | |
998adc3d | 842 | return (&timer->node == base->active.next); |
288867ec | 843 | } |
c0a31329 TG |
844 | |
845 | /* | |
846 | * __remove_hrtimer - internal function to remove a timer | |
847 | * | |
848 | * Caller must hold the base lock. | |
54cdfdb4 TG |
849 | * |
850 | * High resolution timer mode reprograms the clock event device when the | |
851 | * timer is the one which expires next. The caller can disable this by setting | |
852 | * reprogram to zero. This is useful, when the context does a reprogramming | |
853 | * anyway (e.g. timer interrupt) | |
c0a31329 | 854 | */ |
3c8aa39d | 855 | static void __remove_hrtimer(struct hrtimer *timer, |
303e967f | 856 | struct hrtimer_clock_base *base, |
54cdfdb4 | 857 | unsigned long newstate, int reprogram) |
c0a31329 | 858 | { |
27c9cd7e | 859 | struct timerqueue_node *next_timer; |
7403f41f AC |
860 | if (!(timer->state & HRTIMER_STATE_ENQUEUED)) |
861 | goto out; | |
862 | ||
27c9cd7e JO |
863 | next_timer = timerqueue_getnext(&base->active); |
864 | timerqueue_del(&base->active, &timer->node); | |
865 | if (&timer->node == next_timer) { | |
7403f41f AC |
866 | #ifdef CONFIG_HIGH_RES_TIMERS |
867 | /* Reprogram the clock event device. if enabled */ | |
868 | if (reprogram && hrtimer_hres_active()) { | |
869 | ktime_t expires; | |
870 | ||
871 | expires = ktime_sub(hrtimer_get_expires(timer), | |
872 | base->offset); | |
873 | if (base->cpu_base->expires_next.tv64 == expires.tv64) | |
874 | hrtimer_force_reprogram(base->cpu_base, 1); | |
54cdfdb4 | 875 | } |
7403f41f | 876 | #endif |
54cdfdb4 | 877 | } |
ab8177bc TG |
878 | if (!timerqueue_getnext(&base->active)) |
879 | base->cpu_base->active_bases &= ~(1 << base->index); | |
7403f41f | 880 | out: |
303e967f | 881 | timer->state = newstate; |
c0a31329 TG |
882 | } |
883 | ||
884 | /* | |
885 | * remove hrtimer, called with base lock held | |
886 | */ | |
887 | static inline int | |
3c8aa39d | 888 | remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base) |
c0a31329 | 889 | { |
303e967f | 890 | if (hrtimer_is_queued(timer)) { |
f13d4f97 | 891 | unsigned long state; |
54cdfdb4 TG |
892 | int reprogram; |
893 | ||
894 | /* | |
895 | * Remove the timer and force reprogramming when high | |
896 | * resolution mode is active and the timer is on the current | |
897 | * CPU. If we remove a timer on another CPU, reprogramming is | |
898 | * skipped. The interrupt event on this CPU is fired and | |
899 | * reprogramming happens in the interrupt handler. This is a | |
900 | * rare case and less expensive than a smp call. | |
901 | */ | |
c6a2a177 | 902 | debug_deactivate(timer); |
82f67cd9 | 903 | timer_stats_hrtimer_clear_start_info(timer); |
54cdfdb4 | 904 | reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases); |
f13d4f97 SQ |
905 | /* |
906 | * We must preserve the CALLBACK state flag here, | |
907 | * otherwise we could move the timer base in | |
908 | * switch_hrtimer_base. | |
909 | */ | |
910 | state = timer->state & HRTIMER_STATE_CALLBACK; | |
911 | __remove_hrtimer(timer, base, state, reprogram); | |
c0a31329 TG |
912 | return 1; |
913 | } | |
914 | return 0; | |
915 | } | |
916 | ||
7f1e2ca9 PZ |
917 | int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, |
918 | unsigned long delta_ns, const enum hrtimer_mode mode, | |
919 | int wakeup) | |
c0a31329 | 920 | { |
3c8aa39d | 921 | struct hrtimer_clock_base *base, *new_base; |
c0a31329 | 922 | unsigned long flags; |
a6037b61 | 923 | int ret, leftmost; |
c0a31329 TG |
924 | |
925 | base = lock_hrtimer_base(timer, &flags); | |
926 | ||
927 | /* Remove an active timer from the queue: */ | |
928 | ret = remove_hrtimer(timer, base); | |
929 | ||
597d0275 | 930 | if (mode & HRTIMER_MODE_REL) { |
84ea7fe3 | 931 | tim = ktime_add_safe(tim, base->get_time()); |
06027bdd IM |
932 | /* |
933 | * CONFIG_TIME_LOW_RES is a temporary way for architectures | |
934 | * to signal that they simply return xtime in | |
935 | * do_gettimeoffset(). In this case we want to round up by | |
936 | * resolution when starting a relative timer, to avoid short | |
937 | * timeouts. This will go away with the GTOD framework. | |
938 | */ | |
939 | #ifdef CONFIG_TIME_LOW_RES | |
5a7780e7 | 940 | tim = ktime_add_safe(tim, base->resolution); |
06027bdd IM |
941 | #endif |
942 | } | |
237fc6e7 | 943 | |
da8f2e17 | 944 | hrtimer_set_expires_range_ns(timer, tim, delta_ns); |
c0a31329 | 945 | |
84ea7fe3 VK |
946 | /* Switch the timer base, if necessary: */ |
947 | new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED); | |
948 | ||
82f67cd9 IM |
949 | timer_stats_hrtimer_set_start_info(timer); |
950 | ||
a6037b61 PZ |
951 | leftmost = enqueue_hrtimer(timer, new_base); |
952 | ||
49a2a075 VK |
953 | if (!leftmost) { |
954 | unlock_hrtimer_base(timer, &flags); | |
955 | return ret; | |
956 | } | |
957 | ||
958 | if (!hrtimer_is_hres_active(timer)) { | |
959 | /* | |
960 | * Kick to reschedule the next tick to handle the new timer | |
961 | * on dynticks target. | |
962 | */ | |
963 | wake_up_nohz_cpu(new_base->cpu_base->cpu); | |
964 | } else if (new_base->cpu_base == &__get_cpu_var(hrtimer_bases) && | |
9e1e01dd | 965 | hrtimer_reprogram(timer, new_base)) { |
49a2a075 VK |
966 | /* |
967 | * Only allow reprogramming if the new base is on this CPU. | |
968 | * (it might still be on another CPU if the timer was pending) | |
969 | * | |
970 | * XXX send_remote_softirq() ? | |
971 | */ | |
b22affe0 LS |
972 | if (wakeup) { |
973 | /* | |
974 | * We need to drop cpu_base->lock to avoid a | |
975 | * lock ordering issue vs. rq->lock. | |
976 | */ | |
977 | raw_spin_unlock(&new_base->cpu_base->lock); | |
978 | raise_softirq_irqoff(HRTIMER_SOFTIRQ); | |
979 | local_irq_restore(flags); | |
980 | return ret; | |
981 | } else { | |
982 | __raise_softirq_irqoff(HRTIMER_SOFTIRQ); | |
983 | } | |
984 | } | |
c0a31329 TG |
985 | |
986 | unlock_hrtimer_base(timer, &flags); | |
987 | ||
988 | return ret; | |
989 | } | |
8588a2bb | 990 | EXPORT_SYMBOL_GPL(__hrtimer_start_range_ns); |
7f1e2ca9 PZ |
991 | |
992 | /** | |
993 | * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU | |
994 | * @timer: the timer to be added | |
995 | * @tim: expiry time | |
996 | * @delta_ns: "slack" range for the timer | |
8ffbc7d9 DD |
997 | * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or |
998 | * relative (HRTIMER_MODE_REL) | |
7f1e2ca9 PZ |
999 | * |
1000 | * Returns: | |
1001 | * 0 on success | |
1002 | * 1 when the timer was active | |
1003 | */ | |
1004 | int hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, | |
1005 | unsigned long delta_ns, const enum hrtimer_mode mode) | |
1006 | { | |
1007 | return __hrtimer_start_range_ns(timer, tim, delta_ns, mode, 1); | |
1008 | } | |
da8f2e17 AV |
1009 | EXPORT_SYMBOL_GPL(hrtimer_start_range_ns); |
1010 | ||
1011 | /** | |
e1dd7bc5 | 1012 | * hrtimer_start - (re)start an hrtimer on the current CPU |
da8f2e17 AV |
1013 | * @timer: the timer to be added |
1014 | * @tim: expiry time | |
8ffbc7d9 DD |
1015 | * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or |
1016 | * relative (HRTIMER_MODE_REL) | |
da8f2e17 AV |
1017 | * |
1018 | * Returns: | |
1019 | * 0 on success | |
1020 | * 1 when the timer was active | |
1021 | */ | |
1022 | int | |
1023 | hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode) | |
1024 | { | |
7f1e2ca9 | 1025 | return __hrtimer_start_range_ns(timer, tim, 0, mode, 1); |
da8f2e17 | 1026 | } |
8d16b764 | 1027 | EXPORT_SYMBOL_GPL(hrtimer_start); |
c0a31329 | 1028 | |
da8f2e17 | 1029 | |
c0a31329 TG |
1030 | /** |
1031 | * hrtimer_try_to_cancel - try to deactivate a timer | |
c0a31329 TG |
1032 | * @timer: hrtimer to stop |
1033 | * | |
1034 | * Returns: | |
1035 | * 0 when the timer was not active | |
1036 | * 1 when the timer was active | |
1037 | * -1 when the timer is currently excuting the callback function and | |
fa9799e3 | 1038 | * cannot be stopped |
c0a31329 TG |
1039 | */ |
1040 | int hrtimer_try_to_cancel(struct hrtimer *timer) | |
1041 | { | |
3c8aa39d | 1042 | struct hrtimer_clock_base *base; |
c0a31329 TG |
1043 | unsigned long flags; |
1044 | int ret = -1; | |
1045 | ||
1046 | base = lock_hrtimer_base(timer, &flags); | |
1047 | ||
303e967f | 1048 | if (!hrtimer_callback_running(timer)) |
c0a31329 TG |
1049 | ret = remove_hrtimer(timer, base); |
1050 | ||
1051 | unlock_hrtimer_base(timer, &flags); | |
1052 | ||
1053 | return ret; | |
1054 | ||
1055 | } | |
8d16b764 | 1056 | EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel); |
c0a31329 TG |
1057 | |
1058 | /** | |
1059 | * hrtimer_cancel - cancel a timer and wait for the handler to finish. | |
c0a31329 TG |
1060 | * @timer: the timer to be cancelled |
1061 | * | |
1062 | * Returns: | |
1063 | * 0 when the timer was not active | |
1064 | * 1 when the timer was active | |
1065 | */ | |
1066 | int hrtimer_cancel(struct hrtimer *timer) | |
1067 | { | |
1068 | for (;;) { | |
1069 | int ret = hrtimer_try_to_cancel(timer); | |
1070 | ||
1071 | if (ret >= 0) | |
1072 | return ret; | |
5ef37b19 | 1073 | cpu_relax(); |
c0a31329 TG |
1074 | } |
1075 | } | |
8d16b764 | 1076 | EXPORT_SYMBOL_GPL(hrtimer_cancel); |
c0a31329 TG |
1077 | |
1078 | /** | |
1079 | * hrtimer_get_remaining - get remaining time for the timer | |
c0a31329 TG |
1080 | * @timer: the timer to read |
1081 | */ | |
1082 | ktime_t hrtimer_get_remaining(const struct hrtimer *timer) | |
1083 | { | |
c0a31329 TG |
1084 | unsigned long flags; |
1085 | ktime_t rem; | |
1086 | ||
b3bd3de6 | 1087 | lock_hrtimer_base(timer, &flags); |
cc584b21 | 1088 | rem = hrtimer_expires_remaining(timer); |
c0a31329 TG |
1089 | unlock_hrtimer_base(timer, &flags); |
1090 | ||
1091 | return rem; | |
1092 | } | |
8d16b764 | 1093 | EXPORT_SYMBOL_GPL(hrtimer_get_remaining); |
c0a31329 | 1094 | |
3451d024 | 1095 | #ifdef CONFIG_NO_HZ_COMMON |
69239749 TL |
1096 | /** |
1097 | * hrtimer_get_next_event - get the time until next expiry event | |
1098 | * | |
1099 | * Returns the delta to the next expiry event or KTIME_MAX if no timer | |
1100 | * is pending. | |
1101 | */ | |
1102 | ktime_t hrtimer_get_next_event(void) | |
1103 | { | |
3c8aa39d TG |
1104 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
1105 | struct hrtimer_clock_base *base = cpu_base->clock_base; | |
69239749 TL |
1106 | ktime_t delta, mindelta = { .tv64 = KTIME_MAX }; |
1107 | unsigned long flags; | |
1108 | int i; | |
1109 | ||
ecb49d1a | 1110 | raw_spin_lock_irqsave(&cpu_base->lock, flags); |
3c8aa39d | 1111 | |
54cdfdb4 TG |
1112 | if (!hrtimer_hres_active()) { |
1113 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) { | |
1114 | struct hrtimer *timer; | |
998adc3d | 1115 | struct timerqueue_node *next; |
69239749 | 1116 | |
998adc3d JS |
1117 | next = timerqueue_getnext(&base->active); |
1118 | if (!next) | |
54cdfdb4 | 1119 | continue; |
3c8aa39d | 1120 | |
998adc3d | 1121 | timer = container_of(next, struct hrtimer, node); |
cc584b21 | 1122 | delta.tv64 = hrtimer_get_expires_tv64(timer); |
54cdfdb4 TG |
1123 | delta = ktime_sub(delta, base->get_time()); |
1124 | if (delta.tv64 < mindelta.tv64) | |
1125 | mindelta.tv64 = delta.tv64; | |
1126 | } | |
69239749 | 1127 | } |
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 | ||
3c8aa39d | 1145 | cpu_base = &__raw_get_cpu_var(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 TG |
1174 | |
1175 | /** | |
1176 | * hrtimer_get_res - get the timer resolution for a clock | |
c0a31329 TG |
1177 | * @which_clock: which clock to query |
1178 | * @tp: pointer to timespec variable to store the resolution | |
1179 | * | |
72fd4a35 RD |
1180 | * Store the resolution of the clock selected by @which_clock in the |
1181 | * variable pointed to by @tp. | |
c0a31329 TG |
1182 | */ |
1183 | int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp) | |
1184 | { | |
3c8aa39d | 1185 | struct hrtimer_cpu_base *cpu_base; |
e06383db | 1186 | int base = hrtimer_clockid_to_base(which_clock); |
c0a31329 | 1187 | |
3c8aa39d | 1188 | cpu_base = &__raw_get_cpu_var(hrtimer_bases); |
e06383db | 1189 | *tp = ktime_to_timespec(cpu_base->clock_base[base].resolution); |
c0a31329 TG |
1190 | |
1191 | return 0; | |
1192 | } | |
8d16b764 | 1193 | EXPORT_SYMBOL_GPL(hrtimer_get_res); |
c0a31329 | 1194 | |
c6a2a177 | 1195 | static void __run_hrtimer(struct hrtimer *timer, ktime_t *now) |
d3d74453 PZ |
1196 | { |
1197 | struct hrtimer_clock_base *base = timer->base; | |
1198 | struct hrtimer_cpu_base *cpu_base = base->cpu_base; | |
1199 | enum hrtimer_restart (*fn)(struct hrtimer *); | |
1200 | int restart; | |
1201 | ||
ca109491 PZ |
1202 | WARN_ON(!irqs_disabled()); |
1203 | ||
c6a2a177 | 1204 | debug_deactivate(timer); |
d3d74453 PZ |
1205 | __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0); |
1206 | timer_stats_account_hrtimer(timer); | |
d3d74453 | 1207 | fn = timer->function; |
ca109491 PZ |
1208 | |
1209 | /* | |
1210 | * Because we run timers from hardirq context, there is no chance | |
1211 | * they get migrated to another cpu, therefore its safe to unlock | |
1212 | * the timer base. | |
1213 | */ | |
ecb49d1a | 1214 | raw_spin_unlock(&cpu_base->lock); |
c6a2a177 | 1215 | trace_hrtimer_expire_entry(timer, now); |
ca109491 | 1216 | restart = fn(timer); |
c6a2a177 | 1217 | trace_hrtimer_expire_exit(timer); |
ecb49d1a | 1218 | raw_spin_lock(&cpu_base->lock); |
d3d74453 PZ |
1219 | |
1220 | /* | |
e3f1d883 TG |
1221 | * Note: We clear the CALLBACK bit after enqueue_hrtimer and |
1222 | * we do not reprogramm the event hardware. Happens either in | |
1223 | * hrtimer_start_range_ns() or in hrtimer_interrupt() | |
d3d74453 PZ |
1224 | */ |
1225 | if (restart != HRTIMER_NORESTART) { | |
1226 | BUG_ON(timer->state != HRTIMER_STATE_CALLBACK); | |
a6037b61 | 1227 | enqueue_hrtimer(timer, base); |
d3d74453 | 1228 | } |
f13d4f97 SQ |
1229 | |
1230 | WARN_ON_ONCE(!(timer->state & HRTIMER_STATE_CALLBACK)); | |
1231 | ||
d3d74453 PZ |
1232 | timer->state &= ~HRTIMER_STATE_CALLBACK; |
1233 | } | |
1234 | ||
54cdfdb4 TG |
1235 | #ifdef CONFIG_HIGH_RES_TIMERS |
1236 | ||
1237 | /* | |
1238 | * High resolution timer interrupt | |
1239 | * Called with interrupts disabled | |
1240 | */ | |
1241 | void hrtimer_interrupt(struct clock_event_device *dev) | |
1242 | { | |
1243 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); | |
41d2e494 TG |
1244 | ktime_t expires_next, now, entry_time, delta; |
1245 | int i, retries = 0; | |
54cdfdb4 TG |
1246 | |
1247 | BUG_ON(!cpu_base->hres_active); | |
1248 | cpu_base->nr_events++; | |
1249 | dev->next_event.tv64 = KTIME_MAX; | |
1250 | ||
196951e9 | 1251 | raw_spin_lock(&cpu_base->lock); |
5baefd6d | 1252 | entry_time = now = hrtimer_update_base(cpu_base); |
41d2e494 | 1253 | retry: |
54cdfdb4 | 1254 | expires_next.tv64 = KTIME_MAX; |
6ff7041d TG |
1255 | /* |
1256 | * We set expires_next to KTIME_MAX here with cpu_base->lock | |
1257 | * held to prevent that a timer is enqueued in our queue via | |
1258 | * the migration code. This does not affect enqueueing of | |
1259 | * timers which run their callback and need to be requeued on | |
1260 | * this CPU. | |
1261 | */ | |
1262 | cpu_base->expires_next.tv64 = KTIME_MAX; | |
1263 | ||
54cdfdb4 | 1264 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
ab8177bc | 1265 | struct hrtimer_clock_base *base; |
998adc3d | 1266 | struct timerqueue_node *node; |
ab8177bc TG |
1267 | ktime_t basenow; |
1268 | ||
1269 | if (!(cpu_base->active_bases & (1 << i))) | |
1270 | continue; | |
54cdfdb4 | 1271 | |
ab8177bc | 1272 | base = cpu_base->clock_base + i; |
54cdfdb4 TG |
1273 | basenow = ktime_add(now, base->offset); |
1274 | ||
998adc3d | 1275 | while ((node = timerqueue_getnext(&base->active))) { |
54cdfdb4 TG |
1276 | struct hrtimer *timer; |
1277 | ||
998adc3d | 1278 | timer = container_of(node, struct hrtimer, node); |
54cdfdb4 | 1279 | |
654c8e0b AV |
1280 | /* |
1281 | * The immediate goal for using the softexpires is | |
1282 | * minimizing wakeups, not running timers at the | |
1283 | * earliest interrupt after their soft expiration. | |
1284 | * This allows us to avoid using a Priority Search | |
1285 | * Tree, which can answer a stabbing querry for | |
1286 | * overlapping intervals and instead use the simple | |
1287 | * BST we already have. | |
1288 | * We don't add extra wakeups by delaying timers that | |
1289 | * are right-of a not yet expired timer, because that | |
1290 | * timer will have to trigger a wakeup anyway. | |
1291 | */ | |
1292 | ||
1293 | if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer)) { | |
54cdfdb4 TG |
1294 | ktime_t expires; |
1295 | ||
cc584b21 | 1296 | expires = ktime_sub(hrtimer_get_expires(timer), |
54cdfdb4 | 1297 | base->offset); |
8f294b5a PB |
1298 | if (expires.tv64 < 0) |
1299 | expires.tv64 = KTIME_MAX; | |
54cdfdb4 TG |
1300 | if (expires.tv64 < expires_next.tv64) |
1301 | expires_next = expires; | |
1302 | break; | |
1303 | } | |
1304 | ||
c6a2a177 | 1305 | __run_hrtimer(timer, &basenow); |
54cdfdb4 | 1306 | } |
54cdfdb4 TG |
1307 | } |
1308 | ||
6ff7041d TG |
1309 | /* |
1310 | * Store the new expiry value so the migration code can verify | |
1311 | * against it. | |
1312 | */ | |
54cdfdb4 | 1313 | cpu_base->expires_next = expires_next; |
ecb49d1a | 1314 | raw_spin_unlock(&cpu_base->lock); |
54cdfdb4 TG |
1315 | |
1316 | /* Reprogramming necessary ? */ | |
41d2e494 TG |
1317 | if (expires_next.tv64 == KTIME_MAX || |
1318 | !tick_program_event(expires_next, 0)) { | |
1319 | cpu_base->hang_detected = 0; | |
1320 | return; | |
54cdfdb4 | 1321 | } |
41d2e494 TG |
1322 | |
1323 | /* | |
1324 | * The next timer was already expired due to: | |
1325 | * - tracing | |
1326 | * - long lasting callbacks | |
1327 | * - being scheduled away when running in a VM | |
1328 | * | |
1329 | * We need to prevent that we loop forever in the hrtimer | |
1330 | * interrupt routine. We give it 3 attempts to avoid | |
1331 | * overreacting on some spurious event. | |
5baefd6d JS |
1332 | * |
1333 | * Acquire base lock for updating the offsets and retrieving | |
1334 | * the current time. | |
41d2e494 | 1335 | */ |
196951e9 | 1336 | raw_spin_lock(&cpu_base->lock); |
5baefd6d | 1337 | now = hrtimer_update_base(cpu_base); |
41d2e494 TG |
1338 | cpu_base->nr_retries++; |
1339 | if (++retries < 3) | |
1340 | goto retry; | |
1341 | /* | |
1342 | * Give the system a chance to do something else than looping | |
1343 | * here. We stored the entry time, so we know exactly how long | |
1344 | * we spent here. We schedule the next event this amount of | |
1345 | * time away. | |
1346 | */ | |
1347 | cpu_base->nr_hangs++; | |
1348 | cpu_base->hang_detected = 1; | |
196951e9 | 1349 | raw_spin_unlock(&cpu_base->lock); |
41d2e494 TG |
1350 | delta = ktime_sub(now, entry_time); |
1351 | if (delta.tv64 > cpu_base->max_hang_time.tv64) | |
1352 | cpu_base->max_hang_time = delta; | |
1353 | /* | |
1354 | * Limit it to a sensible value as we enforce a longer | |
1355 | * delay. Give the CPU at least 100ms to catch up. | |
1356 | */ | |
1357 | if (delta.tv64 > 100 * NSEC_PER_MSEC) | |
1358 | expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC); | |
1359 | else | |
1360 | expires_next = ktime_add(now, delta); | |
1361 | tick_program_event(expires_next, 1); | |
1362 | printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n", | |
1363 | ktime_to_ns(delta)); | |
54cdfdb4 TG |
1364 | } |
1365 | ||
8bdec955 TG |
1366 | /* |
1367 | * local version of hrtimer_peek_ahead_timers() called with interrupts | |
1368 | * disabled. | |
1369 | */ | |
1370 | static void __hrtimer_peek_ahead_timers(void) | |
1371 | { | |
1372 | struct tick_device *td; | |
1373 | ||
1374 | if (!hrtimer_hres_active()) | |
1375 | return; | |
1376 | ||
1377 | td = &__get_cpu_var(tick_cpu_device); | |
1378 | if (td && td->evtdev) | |
1379 | hrtimer_interrupt(td->evtdev); | |
1380 | } | |
1381 | ||
2e94d1f7 AV |
1382 | /** |
1383 | * hrtimer_peek_ahead_timers -- run soft-expired timers now | |
1384 | * | |
1385 | * hrtimer_peek_ahead_timers will peek at the timer queue of | |
1386 | * the current cpu and check if there are any timers for which | |
1387 | * the soft expires time has passed. If any such timers exist, | |
1388 | * they are run immediately and then removed from the timer queue. | |
1389 | * | |
1390 | */ | |
1391 | void hrtimer_peek_ahead_timers(void) | |
1392 | { | |
643bdf68 | 1393 | unsigned long flags; |
dc4304f7 | 1394 | |
2e94d1f7 | 1395 | local_irq_save(flags); |
8bdec955 | 1396 | __hrtimer_peek_ahead_timers(); |
2e94d1f7 AV |
1397 | local_irq_restore(flags); |
1398 | } | |
1399 | ||
a6037b61 PZ |
1400 | static void run_hrtimer_softirq(struct softirq_action *h) |
1401 | { | |
1402 | hrtimer_peek_ahead_timers(); | |
1403 | } | |
1404 | ||
82c5b7b5 IM |
1405 | #else /* CONFIG_HIGH_RES_TIMERS */ |
1406 | ||
1407 | static inline void __hrtimer_peek_ahead_timers(void) { } | |
1408 | ||
1409 | #endif /* !CONFIG_HIGH_RES_TIMERS */ | |
82f67cd9 | 1410 | |
d3d74453 PZ |
1411 | /* |
1412 | * Called from timer softirq every jiffy, expire hrtimers: | |
1413 | * | |
1414 | * For HRT its the fall back code to run the softirq in the timer | |
1415 | * softirq context in case the hrtimer initialization failed or has | |
1416 | * not been done yet. | |
1417 | */ | |
1418 | void hrtimer_run_pending(void) | |
1419 | { | |
d3d74453 PZ |
1420 | if (hrtimer_hres_active()) |
1421 | return; | |
54cdfdb4 | 1422 | |
d3d74453 PZ |
1423 | /* |
1424 | * This _is_ ugly: We have to check in the softirq context, | |
1425 | * whether we can switch to highres and / or nohz mode. The | |
1426 | * clocksource switch happens in the timer interrupt with | |
1427 | * xtime_lock held. Notification from there only sets the | |
1428 | * check bit in the tick_oneshot code, otherwise we might | |
1429 | * deadlock vs. xtime_lock. | |
1430 | */ | |
1431 | if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) | |
1432 | hrtimer_switch_to_hres(); | |
54cdfdb4 TG |
1433 | } |
1434 | ||
c0a31329 | 1435 | /* |
d3d74453 | 1436 | * Called from hardirq context every jiffy |
c0a31329 | 1437 | */ |
833883d9 | 1438 | void hrtimer_run_queues(void) |
c0a31329 | 1439 | { |
998adc3d | 1440 | struct timerqueue_node *node; |
833883d9 DS |
1441 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
1442 | struct hrtimer_clock_base *base; | |
1443 | int index, gettime = 1; | |
c0a31329 | 1444 | |
833883d9 | 1445 | if (hrtimer_hres_active()) |
3055adda DS |
1446 | return; |
1447 | ||
833883d9 DS |
1448 | for (index = 0; index < HRTIMER_MAX_CLOCK_BASES; index++) { |
1449 | base = &cpu_base->clock_base[index]; | |
b007c389 | 1450 | if (!timerqueue_getnext(&base->active)) |
d3d74453 | 1451 | continue; |
833883d9 | 1452 | |
d7cfb60c | 1453 | if (gettime) { |
833883d9 DS |
1454 | hrtimer_get_softirq_time(cpu_base); |
1455 | gettime = 0; | |
b75f7a51 | 1456 | } |
d3d74453 | 1457 | |
ecb49d1a | 1458 | raw_spin_lock(&cpu_base->lock); |
c0a31329 | 1459 | |
b007c389 | 1460 | while ((node = timerqueue_getnext(&base->active))) { |
833883d9 | 1461 | struct hrtimer *timer; |
54cdfdb4 | 1462 | |
998adc3d | 1463 | timer = container_of(node, struct hrtimer, node); |
cc584b21 AV |
1464 | if (base->softirq_time.tv64 <= |
1465 | hrtimer_get_expires_tv64(timer)) | |
833883d9 DS |
1466 | break; |
1467 | ||
c6a2a177 | 1468 | __run_hrtimer(timer, &base->softirq_time); |
833883d9 | 1469 | } |
ecb49d1a | 1470 | raw_spin_unlock(&cpu_base->lock); |
833883d9 | 1471 | } |
c0a31329 TG |
1472 | } |
1473 | ||
10c94ec1 TG |
1474 | /* |
1475 | * Sleep related functions: | |
1476 | */ | |
c9cb2e3d | 1477 | static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer) |
00362e33 TG |
1478 | { |
1479 | struct hrtimer_sleeper *t = | |
1480 | container_of(timer, struct hrtimer_sleeper, timer); | |
1481 | struct task_struct *task = t->task; | |
1482 | ||
1483 | t->task = NULL; | |
1484 | if (task) | |
1485 | wake_up_process(task); | |
1486 | ||
1487 | return HRTIMER_NORESTART; | |
1488 | } | |
1489 | ||
36c8b586 | 1490 | void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task) |
00362e33 TG |
1491 | { |
1492 | sl->timer.function = hrtimer_wakeup; | |
1493 | sl->task = task; | |
1494 | } | |
2bc481cf | 1495 | EXPORT_SYMBOL_GPL(hrtimer_init_sleeper); |
00362e33 | 1496 | |
669d7868 | 1497 | static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode) |
432569bb | 1498 | { |
669d7868 | 1499 | hrtimer_init_sleeper(t, current); |
10c94ec1 | 1500 | |
432569bb RZ |
1501 | do { |
1502 | set_current_state(TASK_INTERRUPTIBLE); | |
cc584b21 | 1503 | hrtimer_start_expires(&t->timer, mode); |
37bb6cb4 PZ |
1504 | if (!hrtimer_active(&t->timer)) |
1505 | t->task = NULL; | |
432569bb | 1506 | |
54cdfdb4 | 1507 | if (likely(t->task)) |
b0f8c44f | 1508 | freezable_schedule(); |
432569bb | 1509 | |
669d7868 | 1510 | hrtimer_cancel(&t->timer); |
c9cb2e3d | 1511 | mode = HRTIMER_MODE_ABS; |
669d7868 TG |
1512 | |
1513 | } while (t->task && !signal_pending(current)); | |
432569bb | 1514 | |
3588a085 PZ |
1515 | __set_current_state(TASK_RUNNING); |
1516 | ||
669d7868 | 1517 | return t->task == NULL; |
10c94ec1 TG |
1518 | } |
1519 | ||
080344b9 ON |
1520 | static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp) |
1521 | { | |
1522 | struct timespec rmt; | |
1523 | ktime_t rem; | |
1524 | ||
cc584b21 | 1525 | rem = hrtimer_expires_remaining(timer); |
080344b9 ON |
1526 | if (rem.tv64 <= 0) |
1527 | return 0; | |
1528 | rmt = ktime_to_timespec(rem); | |
1529 | ||
1530 | if (copy_to_user(rmtp, &rmt, sizeof(*rmtp))) | |
1531 | return -EFAULT; | |
1532 | ||
1533 | return 1; | |
1534 | } | |
1535 | ||
1711ef38 | 1536 | long __sched hrtimer_nanosleep_restart(struct restart_block *restart) |
10c94ec1 | 1537 | { |
669d7868 | 1538 | struct hrtimer_sleeper t; |
080344b9 | 1539 | struct timespec __user *rmtp; |
237fc6e7 | 1540 | int ret = 0; |
10c94ec1 | 1541 | |
ab8177bc | 1542 | hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid, |
237fc6e7 | 1543 | HRTIMER_MODE_ABS); |
cc584b21 | 1544 | hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires); |
10c94ec1 | 1545 | |
c9cb2e3d | 1546 | if (do_nanosleep(&t, HRTIMER_MODE_ABS)) |
237fc6e7 | 1547 | goto out; |
10c94ec1 | 1548 | |
029a07e0 | 1549 | rmtp = restart->nanosleep.rmtp; |
432569bb | 1550 | if (rmtp) { |
237fc6e7 | 1551 | ret = update_rmtp(&t.timer, rmtp); |
080344b9 | 1552 | if (ret <= 0) |
237fc6e7 | 1553 | goto out; |
432569bb | 1554 | } |
10c94ec1 | 1555 | |
10c94ec1 | 1556 | /* The other values in restart are already filled in */ |
237fc6e7 TG |
1557 | ret = -ERESTART_RESTARTBLOCK; |
1558 | out: | |
1559 | destroy_hrtimer_on_stack(&t.timer); | |
1560 | return ret; | |
10c94ec1 TG |
1561 | } |
1562 | ||
080344b9 | 1563 | long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp, |
10c94ec1 TG |
1564 | const enum hrtimer_mode mode, const clockid_t clockid) |
1565 | { | |
1566 | struct restart_block *restart; | |
669d7868 | 1567 | struct hrtimer_sleeper t; |
237fc6e7 | 1568 | int ret = 0; |
3bd01206 AV |
1569 | unsigned long slack; |
1570 | ||
1571 | slack = current->timer_slack_ns; | |
aab03e05 | 1572 | if (dl_task(current) || rt_task(current)) |
3bd01206 | 1573 | slack = 0; |
10c94ec1 | 1574 | |
237fc6e7 | 1575 | hrtimer_init_on_stack(&t.timer, clockid, mode); |
3bd01206 | 1576 | hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack); |
432569bb | 1577 | if (do_nanosleep(&t, mode)) |
237fc6e7 | 1578 | goto out; |
10c94ec1 | 1579 | |
7978672c | 1580 | /* Absolute timers do not update the rmtp value and restart: */ |
237fc6e7 TG |
1581 | if (mode == HRTIMER_MODE_ABS) { |
1582 | ret = -ERESTARTNOHAND; | |
1583 | goto out; | |
1584 | } | |
10c94ec1 | 1585 | |
432569bb | 1586 | if (rmtp) { |
237fc6e7 | 1587 | ret = update_rmtp(&t.timer, rmtp); |
080344b9 | 1588 | if (ret <= 0) |
237fc6e7 | 1589 | goto out; |
432569bb | 1590 | } |
10c94ec1 TG |
1591 | |
1592 | restart = ¤t_thread_info()->restart_block; | |
1711ef38 | 1593 | restart->fn = hrtimer_nanosleep_restart; |
ab8177bc | 1594 | restart->nanosleep.clockid = t.timer.base->clockid; |
029a07e0 | 1595 | restart->nanosleep.rmtp = rmtp; |
cc584b21 | 1596 | restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer); |
10c94ec1 | 1597 | |
237fc6e7 TG |
1598 | ret = -ERESTART_RESTARTBLOCK; |
1599 | out: | |
1600 | destroy_hrtimer_on_stack(&t.timer); | |
1601 | return ret; | |
10c94ec1 TG |
1602 | } |
1603 | ||
58fd3aa2 HC |
1604 | SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp, |
1605 | struct timespec __user *, rmtp) | |
6ba1b912 | 1606 | { |
080344b9 | 1607 | struct timespec tu; |
6ba1b912 TG |
1608 | |
1609 | if (copy_from_user(&tu, rqtp, sizeof(tu))) | |
1610 | return -EFAULT; | |
1611 | ||
1612 | if (!timespec_valid(&tu)) | |
1613 | return -EINVAL; | |
1614 | ||
080344b9 | 1615 | return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC); |
6ba1b912 TG |
1616 | } |
1617 | ||
c0a31329 TG |
1618 | /* |
1619 | * Functions related to boot-time initialization: | |
1620 | */ | |
0db0628d | 1621 | static void init_hrtimers_cpu(int cpu) |
c0a31329 | 1622 | { |
3c8aa39d | 1623 | struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu); |
c0a31329 TG |
1624 | int i; |
1625 | ||
998adc3d | 1626 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
3c8aa39d | 1627 | cpu_base->clock_base[i].cpu_base = cpu_base; |
998adc3d JS |
1628 | timerqueue_init_head(&cpu_base->clock_base[i].active); |
1629 | } | |
3c8aa39d | 1630 | |
cddd0248 | 1631 | cpu_base->cpu = cpu; |
54cdfdb4 | 1632 | hrtimer_init_hres(cpu_base); |
c0a31329 TG |
1633 | } |
1634 | ||
1635 | #ifdef CONFIG_HOTPLUG_CPU | |
1636 | ||
ca109491 | 1637 | static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base, |
37810659 | 1638 | struct hrtimer_clock_base *new_base) |
c0a31329 TG |
1639 | { |
1640 | struct hrtimer *timer; | |
998adc3d | 1641 | struct timerqueue_node *node; |
c0a31329 | 1642 | |
998adc3d JS |
1643 | while ((node = timerqueue_getnext(&old_base->active))) { |
1644 | timer = container_of(node, struct hrtimer, node); | |
54cdfdb4 | 1645 | BUG_ON(hrtimer_callback_running(timer)); |
c6a2a177 | 1646 | debug_deactivate(timer); |
b00c1a99 TG |
1647 | |
1648 | /* | |
1649 | * Mark it as STATE_MIGRATE not INACTIVE otherwise the | |
1650 | * timer could be seen as !active and just vanish away | |
1651 | * under us on another CPU | |
1652 | */ | |
1653 | __remove_hrtimer(timer, old_base, HRTIMER_STATE_MIGRATE, 0); | |
c0a31329 | 1654 | timer->base = new_base; |
54cdfdb4 | 1655 | /* |
e3f1d883 TG |
1656 | * Enqueue the timers on the new cpu. This does not |
1657 | * reprogram the event device in case the timer | |
1658 | * expires before the earliest on this CPU, but we run | |
1659 | * hrtimer_interrupt after we migrated everything to | |
1660 | * sort out already expired timers and reprogram the | |
1661 | * event device. | |
54cdfdb4 | 1662 | */ |
a6037b61 | 1663 | enqueue_hrtimer(timer, new_base); |
41e1022e | 1664 | |
b00c1a99 TG |
1665 | /* Clear the migration state bit */ |
1666 | timer->state &= ~HRTIMER_STATE_MIGRATE; | |
c0a31329 TG |
1667 | } |
1668 | } | |
1669 | ||
d5fd43c4 | 1670 | static void migrate_hrtimers(int scpu) |
c0a31329 | 1671 | { |
3c8aa39d | 1672 | struct hrtimer_cpu_base *old_base, *new_base; |
731a55ba | 1673 | int i; |
c0a31329 | 1674 | |
37810659 | 1675 | BUG_ON(cpu_online(scpu)); |
37810659 | 1676 | tick_cancel_sched_timer(scpu); |
731a55ba TG |
1677 | |
1678 | local_irq_disable(); | |
1679 | old_base = &per_cpu(hrtimer_bases, scpu); | |
1680 | new_base = &__get_cpu_var(hrtimer_bases); | |
d82f0b0f ON |
1681 | /* |
1682 | * The caller is globally serialized and nobody else | |
1683 | * takes two locks at once, deadlock is not possible. | |
1684 | */ | |
ecb49d1a TG |
1685 | raw_spin_lock(&new_base->lock); |
1686 | raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING); | |
c0a31329 | 1687 | |
3c8aa39d | 1688 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
ca109491 | 1689 | migrate_hrtimer_list(&old_base->clock_base[i], |
37810659 | 1690 | &new_base->clock_base[i]); |
c0a31329 TG |
1691 | } |
1692 | ||
ecb49d1a TG |
1693 | raw_spin_unlock(&old_base->lock); |
1694 | raw_spin_unlock(&new_base->lock); | |
37810659 | 1695 | |
731a55ba TG |
1696 | /* Check, if we got expired work to do */ |
1697 | __hrtimer_peek_ahead_timers(); | |
1698 | local_irq_enable(); | |
c0a31329 | 1699 | } |
37810659 | 1700 | |
c0a31329 TG |
1701 | #endif /* CONFIG_HOTPLUG_CPU */ |
1702 | ||
0db0628d | 1703 | static int hrtimer_cpu_notify(struct notifier_block *self, |
c0a31329 TG |
1704 | unsigned long action, void *hcpu) |
1705 | { | |
b2e3c0ad | 1706 | int scpu = (long)hcpu; |
c0a31329 TG |
1707 | |
1708 | switch (action) { | |
1709 | ||
1710 | case CPU_UP_PREPARE: | |
8bb78442 | 1711 | case CPU_UP_PREPARE_FROZEN: |
37810659 | 1712 | init_hrtimers_cpu(scpu); |
c0a31329 TG |
1713 | break; |
1714 | ||
1715 | #ifdef CONFIG_HOTPLUG_CPU | |
94df7de0 SD |
1716 | case CPU_DYING: |
1717 | case CPU_DYING_FROZEN: | |
1718 | clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DYING, &scpu); | |
1719 | break; | |
c0a31329 | 1720 | case CPU_DEAD: |
8bb78442 | 1721 | case CPU_DEAD_FROZEN: |
b2e3c0ad | 1722 | { |
37810659 | 1723 | clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &scpu); |
d5fd43c4 | 1724 | migrate_hrtimers(scpu); |
c0a31329 | 1725 | break; |
b2e3c0ad | 1726 | } |
c0a31329 TG |
1727 | #endif |
1728 | ||
1729 | default: | |
1730 | break; | |
1731 | } | |
1732 | ||
1733 | return NOTIFY_OK; | |
1734 | } | |
1735 | ||
0db0628d | 1736 | static struct notifier_block hrtimers_nb = { |
c0a31329 TG |
1737 | .notifier_call = hrtimer_cpu_notify, |
1738 | }; | |
1739 | ||
1740 | void __init hrtimers_init(void) | |
1741 | { | |
1742 | hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE, | |
1743 | (void *)(long)smp_processor_id()); | |
1744 | register_cpu_notifier(&hrtimers_nb); | |
a6037b61 PZ |
1745 | #ifdef CONFIG_HIGH_RES_TIMERS |
1746 | open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq); | |
1747 | #endif | |
c0a31329 TG |
1748 | } |
1749 | ||
7bb67439 | 1750 | /** |
351b3f7a | 1751 | * schedule_hrtimeout_range_clock - sleep until timeout |
7bb67439 | 1752 | * @expires: timeout value (ktime_t) |
654c8e0b | 1753 | * @delta: slack in expires timeout (ktime_t) |
7bb67439 | 1754 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL |
351b3f7a | 1755 | * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME |
7bb67439 | 1756 | */ |
351b3f7a CE |
1757 | int __sched |
1758 | schedule_hrtimeout_range_clock(ktime_t *expires, unsigned long delta, | |
1759 | const enum hrtimer_mode mode, int clock) | |
7bb67439 AV |
1760 | { |
1761 | struct hrtimer_sleeper t; | |
1762 | ||
1763 | /* | |
1764 | * Optimize when a zero timeout value is given. It does not | |
1765 | * matter whether this is an absolute or a relative time. | |
1766 | */ | |
1767 | if (expires && !expires->tv64) { | |
1768 | __set_current_state(TASK_RUNNING); | |
1769 | return 0; | |
1770 | } | |
1771 | ||
1772 | /* | |
43b21013 | 1773 | * A NULL parameter means "infinite" |
7bb67439 AV |
1774 | */ |
1775 | if (!expires) { | |
1776 | schedule(); | |
1777 | __set_current_state(TASK_RUNNING); | |
1778 | return -EINTR; | |
1779 | } | |
1780 | ||
351b3f7a | 1781 | hrtimer_init_on_stack(&t.timer, clock, mode); |
654c8e0b | 1782 | hrtimer_set_expires_range_ns(&t.timer, *expires, delta); |
7bb67439 AV |
1783 | |
1784 | hrtimer_init_sleeper(&t, current); | |
1785 | ||
cc584b21 | 1786 | hrtimer_start_expires(&t.timer, mode); |
7bb67439 AV |
1787 | if (!hrtimer_active(&t.timer)) |
1788 | t.task = NULL; | |
1789 | ||
1790 | if (likely(t.task)) | |
1791 | schedule(); | |
1792 | ||
1793 | hrtimer_cancel(&t.timer); | |
1794 | destroy_hrtimer_on_stack(&t.timer); | |
1795 | ||
1796 | __set_current_state(TASK_RUNNING); | |
1797 | ||
1798 | return !t.task ? 0 : -EINTR; | |
1799 | } | |
351b3f7a CE |
1800 | |
1801 | /** | |
1802 | * schedule_hrtimeout_range - sleep until timeout | |
1803 | * @expires: timeout value (ktime_t) | |
1804 | * @delta: slack in expires timeout (ktime_t) | |
1805 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL | |
1806 | * | |
1807 | * Make the current task sleep until the given expiry time has | |
1808 | * elapsed. The routine will return immediately unless | |
1809 | * the current task state has been set (see set_current_state()). | |
1810 | * | |
1811 | * The @delta argument gives the kernel the freedom to schedule the | |
1812 | * actual wakeup to a time that is both power and performance friendly. | |
1813 | * The kernel give the normal best effort behavior for "@expires+@delta", | |
1814 | * but may decide to fire the timer earlier, but no earlier than @expires. | |
1815 | * | |
1816 | * You can set the task state as follows - | |
1817 | * | |
1818 | * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to | |
1819 | * pass before the routine returns. | |
1820 | * | |
1821 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
1822 | * delivered to the current task. | |
1823 | * | |
1824 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1825 | * routine returns. | |
1826 | * | |
1827 | * Returns 0 when the timer has expired otherwise -EINTR | |
1828 | */ | |
1829 | int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta, | |
1830 | const enum hrtimer_mode mode) | |
1831 | { | |
1832 | return schedule_hrtimeout_range_clock(expires, delta, mode, | |
1833 | CLOCK_MONOTONIC); | |
1834 | } | |
654c8e0b AV |
1835 | EXPORT_SYMBOL_GPL(schedule_hrtimeout_range); |
1836 | ||
1837 | /** | |
1838 | * schedule_hrtimeout - sleep until timeout | |
1839 | * @expires: timeout value (ktime_t) | |
1840 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL | |
1841 | * | |
1842 | * Make the current task sleep until the given expiry time has | |
1843 | * elapsed. The routine will return immediately unless | |
1844 | * the current task state has been set (see set_current_state()). | |
1845 | * | |
1846 | * You can set the task state as follows - | |
1847 | * | |
1848 | * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to | |
1849 | * pass before the routine returns. | |
1850 | * | |
1851 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
1852 | * delivered to the current task. | |
1853 | * | |
1854 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1855 | * routine returns. | |
1856 | * | |
1857 | * Returns 0 when the timer has expired otherwise -EINTR | |
1858 | */ | |
1859 | int __sched schedule_hrtimeout(ktime_t *expires, | |
1860 | const enum hrtimer_mode mode) | |
1861 | { | |
1862 | return schedule_hrtimeout_range(expires, 0, mode); | |
1863 | } | |
7bb67439 | 1864 | EXPORT_SYMBOL_GPL(schedule_hrtimeout); |