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 | { |
70a08cca | 117 | ktime_t xtim, mono, boot; |
314ac371 | 118 | struct timespec xts, tom, slp; |
90adda98 | 119 | s32 tai_offset; |
92127c7a | 120 | |
314ac371 | 121 | get_xtime_and_monotonic_and_sleep_offset(&xts, &tom, &slp); |
90adda98 | 122 | tai_offset = timekeeping_get_tai_offset(); |
92127c7a | 123 | |
f4304ab2 | 124 | xtim = timespec_to_ktime(xts); |
70a08cca JS |
125 | mono = ktime_add(xtim, timespec_to_ktime(tom)); |
126 | boot = ktime_add(mono, timespec_to_ktime(slp)); | |
e06383db | 127 | base->clock_base[HRTIMER_BASE_REALTIME].softirq_time = xtim; |
70a08cca JS |
128 | base->clock_base[HRTIMER_BASE_MONOTONIC].softirq_time = mono; |
129 | base->clock_base[HRTIMER_BASE_BOOTTIME].softirq_time = boot; | |
90adda98 JS |
130 | base->clock_base[HRTIMER_BASE_TAI].softirq_time = |
131 | ktime_add(xtim, ktime_set(tai_offset, 0)); | |
92127c7a TG |
132 | } |
133 | ||
c0a31329 TG |
134 | /* |
135 | * Functions and macros which are different for UP/SMP systems are kept in a | |
136 | * single place | |
137 | */ | |
138 | #ifdef CONFIG_SMP | |
139 | ||
c0a31329 TG |
140 | /* |
141 | * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock | |
142 | * means that all timers which are tied to this base via timer->base are | |
143 | * locked, and the base itself is locked too. | |
144 | * | |
145 | * So __run_timers/migrate_timers can safely modify all timers which could | |
146 | * be found on the lists/queues. | |
147 | * | |
148 | * When the timer's base is locked, and the timer removed from list, it is | |
149 | * possible to set timer->base = NULL and drop the lock: the timer remains | |
150 | * locked. | |
151 | */ | |
3c8aa39d TG |
152 | static |
153 | struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer, | |
154 | unsigned long *flags) | |
c0a31329 | 155 | { |
3c8aa39d | 156 | struct hrtimer_clock_base *base; |
c0a31329 TG |
157 | |
158 | for (;;) { | |
159 | base = timer->base; | |
160 | if (likely(base != NULL)) { | |
ecb49d1a | 161 | raw_spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
162 | if (likely(base == timer->base)) |
163 | return base; | |
164 | /* The timer has migrated to another CPU: */ | |
ecb49d1a | 165 | raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags); |
c0a31329 TG |
166 | } |
167 | cpu_relax(); | |
168 | } | |
169 | } | |
170 | ||
6ff7041d TG |
171 | |
172 | /* | |
173 | * Get the preferred target CPU for NOHZ | |
174 | */ | |
175 | static int hrtimer_get_target(int this_cpu, int pinned) | |
176 | { | |
3451d024 | 177 | #ifdef CONFIG_NO_HZ_COMMON |
83cd4fe2 VP |
178 | if (!pinned && get_sysctl_timer_migration() && idle_cpu(this_cpu)) |
179 | return get_nohz_timer_target(); | |
6ff7041d TG |
180 | #endif |
181 | return this_cpu; | |
182 | } | |
183 | ||
184 | /* | |
185 | * With HIGHRES=y we do not migrate the timer when it is expiring | |
186 | * before the next event on the target cpu because we cannot reprogram | |
187 | * the target cpu hardware and we would cause it to fire late. | |
188 | * | |
189 | * Called with cpu_base->lock of target cpu held. | |
190 | */ | |
191 | static int | |
192 | hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base) | |
193 | { | |
194 | #ifdef CONFIG_HIGH_RES_TIMERS | |
195 | ktime_t expires; | |
196 | ||
197 | if (!new_base->cpu_base->hres_active) | |
198 | return 0; | |
199 | ||
200 | expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset); | |
201 | return expires.tv64 <= new_base->cpu_base->expires_next.tv64; | |
202 | #else | |
203 | return 0; | |
204 | #endif | |
205 | } | |
206 | ||
c0a31329 TG |
207 | /* |
208 | * Switch the timer base to the current CPU when possible. | |
209 | */ | |
3c8aa39d | 210 | static inline struct hrtimer_clock_base * |
597d0275 AB |
211 | switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base, |
212 | int pinned) | |
c0a31329 | 213 | { |
3c8aa39d TG |
214 | struct hrtimer_clock_base *new_base; |
215 | struct hrtimer_cpu_base *new_cpu_base; | |
6ff7041d TG |
216 | int this_cpu = smp_processor_id(); |
217 | int cpu = hrtimer_get_target(this_cpu, pinned); | |
ab8177bc | 218 | int basenum = base->index; |
c0a31329 | 219 | |
eea08f32 AB |
220 | again: |
221 | new_cpu_base = &per_cpu(hrtimer_bases, cpu); | |
e06383db | 222 | new_base = &new_cpu_base->clock_base[basenum]; |
c0a31329 TG |
223 | |
224 | if (base != new_base) { | |
225 | /* | |
6ff7041d | 226 | * We are trying to move timer to new_base. |
c0a31329 TG |
227 | * However we can't change timer's base while it is running, |
228 | * so we keep it on the same CPU. No hassle vs. reprogramming | |
229 | * the event source in the high resolution case. The softirq | |
230 | * code will take care of this when the timer function has | |
231 | * completed. There is no conflict as we hold the lock until | |
232 | * the timer is enqueued. | |
233 | */ | |
54cdfdb4 | 234 | if (unlikely(hrtimer_callback_running(timer))) |
c0a31329 TG |
235 | return base; |
236 | ||
237 | /* See the comment in lock_timer_base() */ | |
238 | timer->base = NULL; | |
ecb49d1a TG |
239 | raw_spin_unlock(&base->cpu_base->lock); |
240 | raw_spin_lock(&new_base->cpu_base->lock); | |
eea08f32 | 241 | |
6ff7041d TG |
242 | if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) { |
243 | cpu = this_cpu; | |
ecb49d1a TG |
244 | raw_spin_unlock(&new_base->cpu_base->lock); |
245 | raw_spin_lock(&base->cpu_base->lock); | |
6ff7041d TG |
246 | timer->base = base; |
247 | goto again; | |
eea08f32 | 248 | } |
c0a31329 TG |
249 | timer->base = new_base; |
250 | } | |
251 | return new_base; | |
252 | } | |
253 | ||
254 | #else /* CONFIG_SMP */ | |
255 | ||
3c8aa39d | 256 | static inline struct hrtimer_clock_base * |
c0a31329 TG |
257 | lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) |
258 | { | |
3c8aa39d | 259 | struct hrtimer_clock_base *base = timer->base; |
c0a31329 | 260 | |
ecb49d1a | 261 | raw_spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
262 | |
263 | return base; | |
264 | } | |
265 | ||
eea08f32 | 266 | # define switch_hrtimer_base(t, b, p) (b) |
c0a31329 TG |
267 | |
268 | #endif /* !CONFIG_SMP */ | |
269 | ||
270 | /* | |
271 | * Functions for the union type storage format of ktime_t which are | |
272 | * too large for inlining: | |
273 | */ | |
274 | #if BITS_PER_LONG < 64 | |
275 | # ifndef CONFIG_KTIME_SCALAR | |
276 | /** | |
277 | * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable | |
c0a31329 TG |
278 | * @kt: addend |
279 | * @nsec: the scalar nsec value to add | |
280 | * | |
281 | * Returns the sum of kt and nsec in ktime_t format | |
282 | */ | |
283 | ktime_t ktime_add_ns(const ktime_t kt, u64 nsec) | |
284 | { | |
285 | ktime_t tmp; | |
286 | ||
287 | if (likely(nsec < NSEC_PER_SEC)) { | |
288 | tmp.tv64 = nsec; | |
289 | } else { | |
290 | unsigned long rem = do_div(nsec, NSEC_PER_SEC); | |
291 | ||
51fd36f3 DE |
292 | /* Make sure nsec fits into long */ |
293 | if (unlikely(nsec > KTIME_SEC_MAX)) | |
294 | return (ktime_t){ .tv64 = KTIME_MAX }; | |
295 | ||
c0a31329 TG |
296 | tmp = ktime_set((long)nsec, rem); |
297 | } | |
298 | ||
299 | return ktime_add(kt, tmp); | |
300 | } | |
b8b8fd2d DH |
301 | |
302 | EXPORT_SYMBOL_GPL(ktime_add_ns); | |
a272378d ACM |
303 | |
304 | /** | |
305 | * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable | |
306 | * @kt: minuend | |
307 | * @nsec: the scalar nsec value to subtract | |
308 | * | |
309 | * Returns the subtraction of @nsec from @kt in ktime_t format | |
310 | */ | |
311 | ktime_t ktime_sub_ns(const ktime_t kt, u64 nsec) | |
312 | { | |
313 | ktime_t tmp; | |
314 | ||
315 | if (likely(nsec < NSEC_PER_SEC)) { | |
316 | tmp.tv64 = nsec; | |
317 | } else { | |
318 | unsigned long rem = do_div(nsec, NSEC_PER_SEC); | |
319 | ||
320 | tmp = ktime_set((long)nsec, rem); | |
321 | } | |
322 | ||
323 | return ktime_sub(kt, tmp); | |
324 | } | |
325 | ||
326 | EXPORT_SYMBOL_GPL(ktime_sub_ns); | |
c0a31329 TG |
327 | # endif /* !CONFIG_KTIME_SCALAR */ |
328 | ||
329 | /* | |
330 | * Divide a ktime value by a nanosecond value | |
331 | */ | |
4d672e7a | 332 | u64 ktime_divns(const ktime_t kt, s64 div) |
c0a31329 | 333 | { |
900cfa46 | 334 | u64 dclc; |
c0a31329 TG |
335 | int sft = 0; |
336 | ||
900cfa46 | 337 | dclc = ktime_to_ns(kt); |
c0a31329 TG |
338 | /* Make sure the divisor is less than 2^32: */ |
339 | while (div >> 32) { | |
340 | sft++; | |
341 | div >>= 1; | |
342 | } | |
343 | dclc >>= sft; | |
344 | do_div(dclc, (unsigned long) div); | |
345 | ||
4d672e7a | 346 | return dclc; |
c0a31329 | 347 | } |
c0a31329 TG |
348 | #endif /* BITS_PER_LONG >= 64 */ |
349 | ||
5a7780e7 TG |
350 | /* |
351 | * Add two ktime values and do a safety check for overflow: | |
352 | */ | |
353 | ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs) | |
354 | { | |
355 | ktime_t res = ktime_add(lhs, rhs); | |
356 | ||
357 | /* | |
358 | * We use KTIME_SEC_MAX here, the maximum timeout which we can | |
359 | * return to user space in a timespec: | |
360 | */ | |
361 | if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64) | |
362 | res = ktime_set(KTIME_SEC_MAX, 0); | |
363 | ||
364 | return res; | |
365 | } | |
366 | ||
8daa21e6 AB |
367 | EXPORT_SYMBOL_GPL(ktime_add_safe); |
368 | ||
237fc6e7 TG |
369 | #ifdef CONFIG_DEBUG_OBJECTS_TIMERS |
370 | ||
371 | static struct debug_obj_descr hrtimer_debug_descr; | |
372 | ||
99777288 SG |
373 | static void *hrtimer_debug_hint(void *addr) |
374 | { | |
375 | return ((struct hrtimer *) addr)->function; | |
376 | } | |
377 | ||
237fc6e7 TG |
378 | /* |
379 | * fixup_init is called when: | |
380 | * - an active object is initialized | |
381 | */ | |
382 | static int hrtimer_fixup_init(void *addr, enum debug_obj_state state) | |
383 | { | |
384 | struct hrtimer *timer = addr; | |
385 | ||
386 | switch (state) { | |
387 | case ODEBUG_STATE_ACTIVE: | |
388 | hrtimer_cancel(timer); | |
389 | debug_object_init(timer, &hrtimer_debug_descr); | |
390 | return 1; | |
391 | default: | |
392 | return 0; | |
393 | } | |
394 | } | |
395 | ||
396 | /* | |
397 | * fixup_activate is called when: | |
398 | * - an active object is activated | |
399 | * - an unknown object is activated (might be a statically initialized object) | |
400 | */ | |
401 | static int hrtimer_fixup_activate(void *addr, enum debug_obj_state state) | |
402 | { | |
403 | switch (state) { | |
404 | ||
405 | case ODEBUG_STATE_NOTAVAILABLE: | |
406 | WARN_ON_ONCE(1); | |
407 | return 0; | |
408 | ||
409 | case ODEBUG_STATE_ACTIVE: | |
410 | WARN_ON(1); | |
411 | ||
412 | default: | |
413 | return 0; | |
414 | } | |
415 | } | |
416 | ||
417 | /* | |
418 | * fixup_free is called when: | |
419 | * - an active object is freed | |
420 | */ | |
421 | static int hrtimer_fixup_free(void *addr, enum debug_obj_state state) | |
422 | { | |
423 | struct hrtimer *timer = addr; | |
424 | ||
425 | switch (state) { | |
426 | case ODEBUG_STATE_ACTIVE: | |
427 | hrtimer_cancel(timer); | |
428 | debug_object_free(timer, &hrtimer_debug_descr); | |
429 | return 1; | |
430 | default: | |
431 | return 0; | |
432 | } | |
433 | } | |
434 | ||
435 | static struct debug_obj_descr hrtimer_debug_descr = { | |
436 | .name = "hrtimer", | |
99777288 | 437 | .debug_hint = hrtimer_debug_hint, |
237fc6e7 TG |
438 | .fixup_init = hrtimer_fixup_init, |
439 | .fixup_activate = hrtimer_fixup_activate, | |
440 | .fixup_free = hrtimer_fixup_free, | |
441 | }; | |
442 | ||
443 | static inline void debug_hrtimer_init(struct hrtimer *timer) | |
444 | { | |
445 | debug_object_init(timer, &hrtimer_debug_descr); | |
446 | } | |
447 | ||
448 | static inline void debug_hrtimer_activate(struct hrtimer *timer) | |
449 | { | |
450 | debug_object_activate(timer, &hrtimer_debug_descr); | |
451 | } | |
452 | ||
453 | static inline void debug_hrtimer_deactivate(struct hrtimer *timer) | |
454 | { | |
455 | debug_object_deactivate(timer, &hrtimer_debug_descr); | |
456 | } | |
457 | ||
458 | static inline void debug_hrtimer_free(struct hrtimer *timer) | |
459 | { | |
460 | debug_object_free(timer, &hrtimer_debug_descr); | |
461 | } | |
462 | ||
463 | static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id, | |
464 | enum hrtimer_mode mode); | |
465 | ||
466 | void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id, | |
467 | enum hrtimer_mode mode) | |
468 | { | |
469 | debug_object_init_on_stack(timer, &hrtimer_debug_descr); | |
470 | __hrtimer_init(timer, clock_id, mode); | |
471 | } | |
2bc481cf | 472 | EXPORT_SYMBOL_GPL(hrtimer_init_on_stack); |
237fc6e7 TG |
473 | |
474 | void destroy_hrtimer_on_stack(struct hrtimer *timer) | |
475 | { | |
476 | debug_object_free(timer, &hrtimer_debug_descr); | |
477 | } | |
478 | ||
479 | #else | |
480 | static inline void debug_hrtimer_init(struct hrtimer *timer) { } | |
481 | static inline void debug_hrtimer_activate(struct hrtimer *timer) { } | |
482 | static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { } | |
483 | #endif | |
484 | ||
c6a2a177 XG |
485 | static inline void |
486 | debug_init(struct hrtimer *timer, clockid_t clockid, | |
487 | enum hrtimer_mode mode) | |
488 | { | |
489 | debug_hrtimer_init(timer); | |
490 | trace_hrtimer_init(timer, clockid, mode); | |
491 | } | |
492 | ||
493 | static inline void debug_activate(struct hrtimer *timer) | |
494 | { | |
495 | debug_hrtimer_activate(timer); | |
496 | trace_hrtimer_start(timer); | |
497 | } | |
498 | ||
499 | static inline void debug_deactivate(struct hrtimer *timer) | |
500 | { | |
501 | debug_hrtimer_deactivate(timer); | |
502 | trace_hrtimer_cancel(timer); | |
503 | } | |
504 | ||
54cdfdb4 TG |
505 | /* High resolution timer related functions */ |
506 | #ifdef CONFIG_HIGH_RES_TIMERS | |
507 | ||
508 | /* | |
509 | * High resolution timer enabled ? | |
510 | */ | |
511 | static int hrtimer_hres_enabled __read_mostly = 1; | |
512 | ||
513 | /* | |
514 | * Enable / Disable high resolution mode | |
515 | */ | |
516 | static int __init setup_hrtimer_hres(char *str) | |
517 | { | |
518 | if (!strcmp(str, "off")) | |
519 | hrtimer_hres_enabled = 0; | |
520 | else if (!strcmp(str, "on")) | |
521 | hrtimer_hres_enabled = 1; | |
522 | else | |
523 | return 0; | |
524 | return 1; | |
525 | } | |
526 | ||
527 | __setup("highres=", setup_hrtimer_hres); | |
528 | ||
529 | /* | |
530 | * hrtimer_high_res_enabled - query, if the highres mode is enabled | |
531 | */ | |
532 | static inline int hrtimer_is_hres_enabled(void) | |
533 | { | |
534 | return hrtimer_hres_enabled; | |
535 | } | |
536 | ||
537 | /* | |
538 | * Is the high resolution mode active ? | |
539 | */ | |
540 | static inline int hrtimer_hres_active(void) | |
541 | { | |
909ea964 | 542 | return __this_cpu_read(hrtimer_bases.hres_active); |
54cdfdb4 TG |
543 | } |
544 | ||
545 | /* | |
546 | * Reprogram the event source with checking both queues for the | |
547 | * next event | |
548 | * Called with interrupts disabled and base->lock held | |
549 | */ | |
7403f41f AC |
550 | static void |
551 | hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal) | |
54cdfdb4 TG |
552 | { |
553 | int i; | |
554 | struct hrtimer_clock_base *base = cpu_base->clock_base; | |
7403f41f | 555 | ktime_t expires, expires_next; |
54cdfdb4 | 556 | |
7403f41f | 557 | expires_next.tv64 = KTIME_MAX; |
54cdfdb4 TG |
558 | |
559 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) { | |
560 | struct hrtimer *timer; | |
998adc3d | 561 | struct timerqueue_node *next; |
54cdfdb4 | 562 | |
998adc3d JS |
563 | next = timerqueue_getnext(&base->active); |
564 | if (!next) | |
54cdfdb4 | 565 | continue; |
998adc3d JS |
566 | timer = container_of(next, struct hrtimer, node); |
567 | ||
cc584b21 | 568 | expires = ktime_sub(hrtimer_get_expires(timer), base->offset); |
b0a9b511 TG |
569 | /* |
570 | * clock_was_set() has changed base->offset so the | |
571 | * result might be negative. Fix it up to prevent a | |
572 | * false positive in clockevents_program_event() | |
573 | */ | |
574 | if (expires.tv64 < 0) | |
575 | expires.tv64 = 0; | |
7403f41f AC |
576 | if (expires.tv64 < expires_next.tv64) |
577 | expires_next = expires; | |
54cdfdb4 TG |
578 | } |
579 | ||
7403f41f AC |
580 | if (skip_equal && expires_next.tv64 == cpu_base->expires_next.tv64) |
581 | return; | |
582 | ||
583 | cpu_base->expires_next.tv64 = expires_next.tv64; | |
584 | ||
54cdfdb4 TG |
585 | if (cpu_base->expires_next.tv64 != KTIME_MAX) |
586 | tick_program_event(cpu_base->expires_next, 1); | |
587 | } | |
588 | ||
589 | /* | |
590 | * Shared reprogramming for clock_realtime and clock_monotonic | |
591 | * | |
592 | * When a timer is enqueued and expires earlier than the already enqueued | |
593 | * timers, we have to check, whether it expires earlier than the timer for | |
594 | * which the clock event device was armed. | |
595 | * | |
596 | * Called with interrupts disabled and base->cpu_base.lock held | |
597 | */ | |
598 | static int hrtimer_reprogram(struct hrtimer *timer, | |
599 | struct hrtimer_clock_base *base) | |
600 | { | |
41d2e494 | 601 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
cc584b21 | 602 | ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset); |
54cdfdb4 TG |
603 | int res; |
604 | ||
cc584b21 | 605 | WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0); |
63070a79 | 606 | |
54cdfdb4 TG |
607 | /* |
608 | * When the callback is running, we do not reprogram the clock event | |
609 | * device. The timer callback is either running on a different CPU or | |
3a4fa0a2 | 610 | * the callback is executed in the hrtimer_interrupt context. The |
54cdfdb4 TG |
611 | * reprogramming is handled either by the softirq, which called the |
612 | * callback or at the end of the hrtimer_interrupt. | |
613 | */ | |
614 | if (hrtimer_callback_running(timer)) | |
615 | return 0; | |
616 | ||
63070a79 TG |
617 | /* |
618 | * CLOCK_REALTIME timer might be requested with an absolute | |
619 | * expiry time which is less than base->offset. Nothing wrong | |
620 | * about that, just avoid to call into the tick code, which | |
621 | * has now objections against negative expiry values. | |
622 | */ | |
623 | if (expires.tv64 < 0) | |
624 | return -ETIME; | |
625 | ||
41d2e494 TG |
626 | if (expires.tv64 >= cpu_base->expires_next.tv64) |
627 | return 0; | |
628 | ||
629 | /* | |
630 | * If a hang was detected in the last timer interrupt then we | |
631 | * do not schedule a timer which is earlier than the expiry | |
632 | * which we enforced in the hang detection. We want the system | |
633 | * to make progress. | |
634 | */ | |
635 | if (cpu_base->hang_detected) | |
54cdfdb4 TG |
636 | return 0; |
637 | ||
638 | /* | |
639 | * Clockevents returns -ETIME, when the event was in the past. | |
640 | */ | |
641 | res = tick_program_event(expires, 0); | |
642 | if (!IS_ERR_VALUE(res)) | |
41d2e494 | 643 | cpu_base->expires_next = expires; |
54cdfdb4 TG |
644 | return res; |
645 | } | |
646 | ||
54cdfdb4 TG |
647 | /* |
648 | * Initialize the high resolution related parts of cpu_base | |
649 | */ | |
650 | static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) | |
651 | { | |
652 | base->expires_next.tv64 = KTIME_MAX; | |
653 | base->hres_active = 0; | |
54cdfdb4 TG |
654 | } |
655 | ||
54cdfdb4 TG |
656 | /* |
657 | * When High resolution timers are active, try to reprogram. Note, that in case | |
658 | * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry | |
659 | * check happens. The timer gets enqueued into the rbtree. The reprogramming | |
660 | * and expiry check is done in the hrtimer_interrupt or in the softirq. | |
661 | */ | |
662 | static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer, | |
b22affe0 | 663 | struct hrtimer_clock_base *base) |
54cdfdb4 | 664 | { |
b22affe0 | 665 | return base->cpu_base->hres_active && hrtimer_reprogram(timer, base); |
54cdfdb4 TG |
666 | } |
667 | ||
5baefd6d JS |
668 | static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base) |
669 | { | |
670 | ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset; | |
671 | ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset; | |
90adda98 | 672 | ktime_t *offs_tai = &base->clock_base[HRTIMER_BASE_TAI].offset; |
5baefd6d | 673 | |
90adda98 | 674 | return ktime_get_update_offsets(offs_real, offs_boot, offs_tai); |
5baefd6d JS |
675 | } |
676 | ||
9ec26907 TG |
677 | /* |
678 | * Retrigger next event is called after clock was set | |
679 | * | |
680 | * Called with interrupts disabled via on_each_cpu() | |
681 | */ | |
682 | static void retrigger_next_event(void *arg) | |
683 | { | |
684 | struct hrtimer_cpu_base *base = &__get_cpu_var(hrtimer_bases); | |
9ec26907 TG |
685 | |
686 | if (!hrtimer_hres_active()) | |
687 | return; | |
688 | ||
9ec26907 | 689 | raw_spin_lock(&base->lock); |
5baefd6d | 690 | hrtimer_update_base(base); |
9ec26907 TG |
691 | hrtimer_force_reprogram(base, 0); |
692 | raw_spin_unlock(&base->lock); | |
693 | } | |
b12a03ce | 694 | |
54cdfdb4 TG |
695 | /* |
696 | * Switch to high resolution mode | |
697 | */ | |
f8953856 | 698 | static int hrtimer_switch_to_hres(void) |
54cdfdb4 | 699 | { |
b12a03ce | 700 | int i, cpu = smp_processor_id(); |
820de5c3 | 701 | struct hrtimer_cpu_base *base = &per_cpu(hrtimer_bases, cpu); |
54cdfdb4 TG |
702 | unsigned long flags; |
703 | ||
704 | if (base->hres_active) | |
f8953856 | 705 | return 1; |
54cdfdb4 TG |
706 | |
707 | local_irq_save(flags); | |
708 | ||
709 | if (tick_init_highres()) { | |
710 | local_irq_restore(flags); | |
820de5c3 IM |
711 | printk(KERN_WARNING "Could not switch to high resolution " |
712 | "mode on CPU %d\n", cpu); | |
f8953856 | 713 | return 0; |
54cdfdb4 TG |
714 | } |
715 | base->hres_active = 1; | |
b12a03ce TG |
716 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) |
717 | base->clock_base[i].resolution = KTIME_HIGH_RES; | |
54cdfdb4 TG |
718 | |
719 | tick_setup_sched_timer(); | |
54cdfdb4 TG |
720 | /* "Retrigger" the interrupt to get things going */ |
721 | retrigger_next_event(NULL); | |
722 | local_irq_restore(flags); | |
f8953856 | 723 | return 1; |
54cdfdb4 TG |
724 | } |
725 | ||
5ec2481b TG |
726 | static void clock_was_set_work(struct work_struct *work) |
727 | { | |
728 | clock_was_set(); | |
729 | } | |
730 | ||
731 | static DECLARE_WORK(hrtimer_work, clock_was_set_work); | |
732 | ||
f55a6faa | 733 | /* |
5ec2481b TG |
734 | * Called from timekeeping and resume code to reprogramm the hrtimer |
735 | * interrupt device on all cpus. | |
f55a6faa JS |
736 | */ |
737 | void clock_was_set_delayed(void) | |
738 | { | |
5ec2481b | 739 | schedule_work(&hrtimer_work); |
f55a6faa JS |
740 | } |
741 | ||
54cdfdb4 TG |
742 | #else |
743 | ||
744 | static inline int hrtimer_hres_active(void) { return 0; } | |
745 | static inline int hrtimer_is_hres_enabled(void) { return 0; } | |
f8953856 | 746 | static inline int hrtimer_switch_to_hres(void) { return 0; } |
7403f41f AC |
747 | static inline void |
748 | hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { } | |
54cdfdb4 | 749 | static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer, |
b22affe0 | 750 | struct hrtimer_clock_base *base) |
54cdfdb4 TG |
751 | { |
752 | return 0; | |
753 | } | |
54cdfdb4 | 754 | static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { } |
9ec26907 | 755 | static inline void retrigger_next_event(void *arg) { } |
54cdfdb4 TG |
756 | |
757 | #endif /* CONFIG_HIGH_RES_TIMERS */ | |
758 | ||
b12a03ce TG |
759 | /* |
760 | * Clock realtime was set | |
761 | * | |
762 | * Change the offset of the realtime clock vs. the monotonic | |
763 | * clock. | |
764 | * | |
765 | * We might have to reprogram the high resolution timer interrupt. On | |
766 | * SMP we call the architecture specific code to retrigger _all_ high | |
767 | * resolution timer interrupts. On UP we just disable interrupts and | |
768 | * call the high resolution interrupt code. | |
769 | */ | |
770 | void clock_was_set(void) | |
771 | { | |
90ff1f30 | 772 | #ifdef CONFIG_HIGH_RES_TIMERS |
b12a03ce TG |
773 | /* Retrigger the CPU local events everywhere */ |
774 | on_each_cpu(retrigger_next_event, NULL, 1); | |
9ec26907 TG |
775 | #endif |
776 | timerfd_clock_was_set(); | |
b12a03ce TG |
777 | } |
778 | ||
779 | /* | |
780 | * During resume we might have to reprogram the high resolution timer | |
7c4c3a0f DV |
781 | * interrupt on all online CPUs. However, all other CPUs will be |
782 | * stopped with IRQs interrupts disabled so the clock_was_set() call | |
5ec2481b | 783 | * must be deferred. |
b12a03ce TG |
784 | */ |
785 | void hrtimers_resume(void) | |
786 | { | |
787 | WARN_ONCE(!irqs_disabled(), | |
788 | KERN_INFO "hrtimers_resume() called with IRQs enabled!"); | |
789 | ||
5ec2481b | 790 | /* Retrigger on the local CPU */ |
b12a03ce | 791 | retrigger_next_event(NULL); |
5ec2481b TG |
792 | /* And schedule a retrigger for all others */ |
793 | clock_was_set_delayed(); | |
b12a03ce TG |
794 | } |
795 | ||
5f201907 | 796 | static inline void timer_stats_hrtimer_set_start_info(struct hrtimer *timer) |
82f67cd9 | 797 | { |
5f201907 | 798 | #ifdef CONFIG_TIMER_STATS |
82f67cd9 IM |
799 | if (timer->start_site) |
800 | return; | |
5f201907 | 801 | timer->start_site = __builtin_return_address(0); |
82f67cd9 IM |
802 | memcpy(timer->start_comm, current->comm, TASK_COMM_LEN); |
803 | timer->start_pid = current->pid; | |
5f201907 HC |
804 | #endif |
805 | } | |
806 | ||
807 | static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer *timer) | |
808 | { | |
809 | #ifdef CONFIG_TIMER_STATS | |
810 | timer->start_site = NULL; | |
811 | #endif | |
82f67cd9 | 812 | } |
5f201907 HC |
813 | |
814 | static inline void timer_stats_account_hrtimer(struct hrtimer *timer) | |
815 | { | |
816 | #ifdef CONFIG_TIMER_STATS | |
817 | if (likely(!timer_stats_active)) | |
818 | return; | |
819 | timer_stats_update_stats(timer, timer->start_pid, timer->start_site, | |
820 | timer->function, timer->start_comm, 0); | |
82f67cd9 | 821 | #endif |
5f201907 | 822 | } |
82f67cd9 | 823 | |
c0a31329 | 824 | /* |
6506f2aa | 825 | * Counterpart to lock_hrtimer_base above: |
c0a31329 TG |
826 | */ |
827 | static inline | |
828 | void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) | |
829 | { | |
ecb49d1a | 830 | raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags); |
c0a31329 TG |
831 | } |
832 | ||
833 | /** | |
834 | * hrtimer_forward - forward the timer expiry | |
c0a31329 | 835 | * @timer: hrtimer to forward |
44f21475 | 836 | * @now: forward past this time |
c0a31329 TG |
837 | * @interval: the interval to forward |
838 | * | |
839 | * Forward the timer expiry so it will expire in the future. | |
8dca6f33 | 840 | * Returns the number of overruns. |
c0a31329 | 841 | */ |
4d672e7a | 842 | u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval) |
c0a31329 | 843 | { |
4d672e7a | 844 | u64 orun = 1; |
44f21475 | 845 | ktime_t delta; |
c0a31329 | 846 | |
cc584b21 | 847 | delta = ktime_sub(now, hrtimer_get_expires(timer)); |
c0a31329 TG |
848 | |
849 | if (delta.tv64 < 0) | |
850 | return 0; | |
851 | ||
c9db4fa1 TG |
852 | if (interval.tv64 < timer->base->resolution.tv64) |
853 | interval.tv64 = timer->base->resolution.tv64; | |
854 | ||
c0a31329 | 855 | if (unlikely(delta.tv64 >= interval.tv64)) { |
df869b63 | 856 | s64 incr = ktime_to_ns(interval); |
c0a31329 TG |
857 | |
858 | orun = ktime_divns(delta, incr); | |
cc584b21 AV |
859 | hrtimer_add_expires_ns(timer, incr * orun); |
860 | if (hrtimer_get_expires_tv64(timer) > now.tv64) | |
c0a31329 TG |
861 | return orun; |
862 | /* | |
863 | * This (and the ktime_add() below) is the | |
864 | * correction for exact: | |
865 | */ | |
866 | orun++; | |
867 | } | |
cc584b21 | 868 | hrtimer_add_expires(timer, interval); |
c0a31329 TG |
869 | |
870 | return orun; | |
871 | } | |
6bdb6b62 | 872 | EXPORT_SYMBOL_GPL(hrtimer_forward); |
c0a31329 TG |
873 | |
874 | /* | |
875 | * enqueue_hrtimer - internal function to (re)start a timer | |
876 | * | |
877 | * The timer is inserted in expiry order. Insertion into the | |
878 | * red black tree is O(log(n)). Must hold the base lock. | |
a6037b61 PZ |
879 | * |
880 | * Returns 1 when the new timer is the leftmost timer in the tree. | |
c0a31329 | 881 | */ |
a6037b61 PZ |
882 | static int enqueue_hrtimer(struct hrtimer *timer, |
883 | struct hrtimer_clock_base *base) | |
c0a31329 | 884 | { |
c6a2a177 | 885 | debug_activate(timer); |
237fc6e7 | 886 | |
998adc3d | 887 | timerqueue_add(&base->active, &timer->node); |
ab8177bc | 888 | base->cpu_base->active_bases |= 1 << base->index; |
54cdfdb4 | 889 | |
303e967f TG |
890 | /* |
891 | * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the | |
892 | * state of a possibly running callback. | |
893 | */ | |
894 | timer->state |= HRTIMER_STATE_ENQUEUED; | |
a6037b61 | 895 | |
998adc3d | 896 | return (&timer->node == base->active.next); |
288867ec | 897 | } |
c0a31329 TG |
898 | |
899 | /* | |
900 | * __remove_hrtimer - internal function to remove a timer | |
901 | * | |
902 | * Caller must hold the base lock. | |
54cdfdb4 TG |
903 | * |
904 | * High resolution timer mode reprograms the clock event device when the | |
905 | * timer is the one which expires next. The caller can disable this by setting | |
906 | * reprogram to zero. This is useful, when the context does a reprogramming | |
907 | * anyway (e.g. timer interrupt) | |
c0a31329 | 908 | */ |
3c8aa39d | 909 | static void __remove_hrtimer(struct hrtimer *timer, |
303e967f | 910 | struct hrtimer_clock_base *base, |
54cdfdb4 | 911 | unsigned long newstate, int reprogram) |
c0a31329 | 912 | { |
27c9cd7e | 913 | struct timerqueue_node *next_timer; |
7403f41f AC |
914 | if (!(timer->state & HRTIMER_STATE_ENQUEUED)) |
915 | goto out; | |
916 | ||
27c9cd7e JO |
917 | next_timer = timerqueue_getnext(&base->active); |
918 | timerqueue_del(&base->active, &timer->node); | |
919 | if (&timer->node == next_timer) { | |
7403f41f AC |
920 | #ifdef CONFIG_HIGH_RES_TIMERS |
921 | /* Reprogram the clock event device. if enabled */ | |
922 | if (reprogram && hrtimer_hres_active()) { | |
923 | ktime_t expires; | |
924 | ||
925 | expires = ktime_sub(hrtimer_get_expires(timer), | |
926 | base->offset); | |
927 | if (base->cpu_base->expires_next.tv64 == expires.tv64) | |
928 | hrtimer_force_reprogram(base->cpu_base, 1); | |
54cdfdb4 | 929 | } |
7403f41f | 930 | #endif |
54cdfdb4 | 931 | } |
ab8177bc TG |
932 | if (!timerqueue_getnext(&base->active)) |
933 | base->cpu_base->active_bases &= ~(1 << base->index); | |
7403f41f | 934 | out: |
303e967f | 935 | timer->state = newstate; |
c0a31329 TG |
936 | } |
937 | ||
938 | /* | |
939 | * remove hrtimer, called with base lock held | |
940 | */ | |
941 | static inline int | |
3c8aa39d | 942 | remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base) |
c0a31329 | 943 | { |
303e967f | 944 | if (hrtimer_is_queued(timer)) { |
f13d4f97 | 945 | unsigned long state; |
54cdfdb4 TG |
946 | int reprogram; |
947 | ||
948 | /* | |
949 | * Remove the timer and force reprogramming when high | |
950 | * resolution mode is active and the timer is on the current | |
951 | * CPU. If we remove a timer on another CPU, reprogramming is | |
952 | * skipped. The interrupt event on this CPU is fired and | |
953 | * reprogramming happens in the interrupt handler. This is a | |
954 | * rare case and less expensive than a smp call. | |
955 | */ | |
c6a2a177 | 956 | debug_deactivate(timer); |
82f67cd9 | 957 | timer_stats_hrtimer_clear_start_info(timer); |
54cdfdb4 | 958 | reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases); |
f13d4f97 SQ |
959 | /* |
960 | * We must preserve the CALLBACK state flag here, | |
961 | * otherwise we could move the timer base in | |
962 | * switch_hrtimer_base. | |
963 | */ | |
964 | state = timer->state & HRTIMER_STATE_CALLBACK; | |
965 | __remove_hrtimer(timer, base, state, reprogram); | |
c0a31329 TG |
966 | return 1; |
967 | } | |
968 | return 0; | |
969 | } | |
970 | ||
7f1e2ca9 PZ |
971 | int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, |
972 | unsigned long delta_ns, const enum hrtimer_mode mode, | |
973 | int wakeup) | |
c0a31329 | 974 | { |
3c8aa39d | 975 | struct hrtimer_clock_base *base, *new_base; |
c0a31329 | 976 | unsigned long flags; |
a6037b61 | 977 | int ret, leftmost; |
c0a31329 TG |
978 | |
979 | base = lock_hrtimer_base(timer, &flags); | |
980 | ||
981 | /* Remove an active timer from the queue: */ | |
982 | ret = remove_hrtimer(timer, base); | |
983 | ||
984 | /* Switch the timer base, if necessary: */ | |
597d0275 | 985 | new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED); |
c0a31329 | 986 | |
597d0275 | 987 | if (mode & HRTIMER_MODE_REL) { |
5a7780e7 | 988 | tim = ktime_add_safe(tim, new_base->get_time()); |
06027bdd IM |
989 | /* |
990 | * CONFIG_TIME_LOW_RES is a temporary way for architectures | |
991 | * to signal that they simply return xtime in | |
992 | * do_gettimeoffset(). In this case we want to round up by | |
993 | * resolution when starting a relative timer, to avoid short | |
994 | * timeouts. This will go away with the GTOD framework. | |
995 | */ | |
996 | #ifdef CONFIG_TIME_LOW_RES | |
5a7780e7 | 997 | tim = ktime_add_safe(tim, base->resolution); |
06027bdd IM |
998 | #endif |
999 | } | |
237fc6e7 | 1000 | |
da8f2e17 | 1001 | hrtimer_set_expires_range_ns(timer, tim, delta_ns); |
c0a31329 | 1002 | |
82f67cd9 IM |
1003 | timer_stats_hrtimer_set_start_info(timer); |
1004 | ||
a6037b61 PZ |
1005 | leftmost = enqueue_hrtimer(timer, new_base); |
1006 | ||
935c631d IM |
1007 | /* |
1008 | * Only allow reprogramming if the new base is on this CPU. | |
1009 | * (it might still be on another CPU if the timer was pending) | |
a6037b61 PZ |
1010 | * |
1011 | * XXX send_remote_softirq() ? | |
935c631d | 1012 | */ |
b22affe0 LS |
1013 | if (leftmost && new_base->cpu_base == &__get_cpu_var(hrtimer_bases) |
1014 | && hrtimer_enqueue_reprogram(timer, new_base)) { | |
1015 | if (wakeup) { | |
1016 | /* | |
1017 | * We need to drop cpu_base->lock to avoid a | |
1018 | * lock ordering issue vs. rq->lock. | |
1019 | */ | |
1020 | raw_spin_unlock(&new_base->cpu_base->lock); | |
1021 | raise_softirq_irqoff(HRTIMER_SOFTIRQ); | |
1022 | local_irq_restore(flags); | |
1023 | return ret; | |
1024 | } else { | |
1025 | __raise_softirq_irqoff(HRTIMER_SOFTIRQ); | |
1026 | } | |
1027 | } | |
c0a31329 TG |
1028 | |
1029 | unlock_hrtimer_base(timer, &flags); | |
1030 | ||
1031 | return ret; | |
1032 | } | |
7f1e2ca9 PZ |
1033 | |
1034 | /** | |
1035 | * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU | |
1036 | * @timer: the timer to be added | |
1037 | * @tim: expiry time | |
1038 | * @delta_ns: "slack" range for the timer | |
8ffbc7d9 DD |
1039 | * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or |
1040 | * relative (HRTIMER_MODE_REL) | |
7f1e2ca9 PZ |
1041 | * |
1042 | * Returns: | |
1043 | * 0 on success | |
1044 | * 1 when the timer was active | |
1045 | */ | |
1046 | int hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, | |
1047 | unsigned long delta_ns, const enum hrtimer_mode mode) | |
1048 | { | |
1049 | return __hrtimer_start_range_ns(timer, tim, delta_ns, mode, 1); | |
1050 | } | |
da8f2e17 AV |
1051 | EXPORT_SYMBOL_GPL(hrtimer_start_range_ns); |
1052 | ||
1053 | /** | |
e1dd7bc5 | 1054 | * hrtimer_start - (re)start an hrtimer on the current CPU |
da8f2e17 AV |
1055 | * @timer: the timer to be added |
1056 | * @tim: expiry time | |
8ffbc7d9 DD |
1057 | * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or |
1058 | * relative (HRTIMER_MODE_REL) | |
da8f2e17 AV |
1059 | * |
1060 | * Returns: | |
1061 | * 0 on success | |
1062 | * 1 when the timer was active | |
1063 | */ | |
1064 | int | |
1065 | hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode) | |
1066 | { | |
7f1e2ca9 | 1067 | return __hrtimer_start_range_ns(timer, tim, 0, mode, 1); |
da8f2e17 | 1068 | } |
8d16b764 | 1069 | EXPORT_SYMBOL_GPL(hrtimer_start); |
c0a31329 | 1070 | |
da8f2e17 | 1071 | |
c0a31329 TG |
1072 | /** |
1073 | * hrtimer_try_to_cancel - try to deactivate a timer | |
c0a31329 TG |
1074 | * @timer: hrtimer to stop |
1075 | * | |
1076 | * Returns: | |
1077 | * 0 when the timer was not active | |
1078 | * 1 when the timer was active | |
1079 | * -1 when the timer is currently excuting the callback function and | |
fa9799e3 | 1080 | * cannot be stopped |
c0a31329 TG |
1081 | */ |
1082 | int hrtimer_try_to_cancel(struct hrtimer *timer) | |
1083 | { | |
3c8aa39d | 1084 | struct hrtimer_clock_base *base; |
c0a31329 TG |
1085 | unsigned long flags; |
1086 | int ret = -1; | |
1087 | ||
1088 | base = lock_hrtimer_base(timer, &flags); | |
1089 | ||
303e967f | 1090 | if (!hrtimer_callback_running(timer)) |
c0a31329 TG |
1091 | ret = remove_hrtimer(timer, base); |
1092 | ||
1093 | unlock_hrtimer_base(timer, &flags); | |
1094 | ||
1095 | return ret; | |
1096 | ||
1097 | } | |
8d16b764 | 1098 | EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel); |
c0a31329 TG |
1099 | |
1100 | /** | |
1101 | * hrtimer_cancel - cancel a timer and wait for the handler to finish. | |
c0a31329 TG |
1102 | * @timer: the timer to be cancelled |
1103 | * | |
1104 | * Returns: | |
1105 | * 0 when the timer was not active | |
1106 | * 1 when the timer was active | |
1107 | */ | |
1108 | int hrtimer_cancel(struct hrtimer *timer) | |
1109 | { | |
1110 | for (;;) { | |
1111 | int ret = hrtimer_try_to_cancel(timer); | |
1112 | ||
1113 | if (ret >= 0) | |
1114 | return ret; | |
5ef37b19 | 1115 | cpu_relax(); |
c0a31329 TG |
1116 | } |
1117 | } | |
8d16b764 | 1118 | EXPORT_SYMBOL_GPL(hrtimer_cancel); |
c0a31329 TG |
1119 | |
1120 | /** | |
1121 | * hrtimer_get_remaining - get remaining time for the timer | |
c0a31329 TG |
1122 | * @timer: the timer to read |
1123 | */ | |
1124 | ktime_t hrtimer_get_remaining(const struct hrtimer *timer) | |
1125 | { | |
c0a31329 TG |
1126 | unsigned long flags; |
1127 | ktime_t rem; | |
1128 | ||
b3bd3de6 | 1129 | lock_hrtimer_base(timer, &flags); |
cc584b21 | 1130 | rem = hrtimer_expires_remaining(timer); |
c0a31329 TG |
1131 | unlock_hrtimer_base(timer, &flags); |
1132 | ||
1133 | return rem; | |
1134 | } | |
8d16b764 | 1135 | EXPORT_SYMBOL_GPL(hrtimer_get_remaining); |
c0a31329 | 1136 | |
3451d024 | 1137 | #ifdef CONFIG_NO_HZ_COMMON |
69239749 TL |
1138 | /** |
1139 | * hrtimer_get_next_event - get the time until next expiry event | |
1140 | * | |
1141 | * Returns the delta to the next expiry event or KTIME_MAX if no timer | |
1142 | * is pending. | |
1143 | */ | |
1144 | ktime_t hrtimer_get_next_event(void) | |
1145 | { | |
3c8aa39d TG |
1146 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
1147 | struct hrtimer_clock_base *base = cpu_base->clock_base; | |
69239749 TL |
1148 | ktime_t delta, mindelta = { .tv64 = KTIME_MAX }; |
1149 | unsigned long flags; | |
1150 | int i; | |
1151 | ||
ecb49d1a | 1152 | raw_spin_lock_irqsave(&cpu_base->lock, flags); |
3c8aa39d | 1153 | |
54cdfdb4 TG |
1154 | if (!hrtimer_hres_active()) { |
1155 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) { | |
1156 | struct hrtimer *timer; | |
998adc3d | 1157 | struct timerqueue_node *next; |
69239749 | 1158 | |
998adc3d JS |
1159 | next = timerqueue_getnext(&base->active); |
1160 | if (!next) | |
54cdfdb4 | 1161 | continue; |
3c8aa39d | 1162 | |
998adc3d | 1163 | timer = container_of(next, struct hrtimer, node); |
cc584b21 | 1164 | delta.tv64 = hrtimer_get_expires_tv64(timer); |
54cdfdb4 TG |
1165 | delta = ktime_sub(delta, base->get_time()); |
1166 | if (delta.tv64 < mindelta.tv64) | |
1167 | mindelta.tv64 = delta.tv64; | |
1168 | } | |
69239749 | 1169 | } |
3c8aa39d | 1170 | |
ecb49d1a | 1171 | raw_spin_unlock_irqrestore(&cpu_base->lock, flags); |
3c8aa39d | 1172 | |
69239749 TL |
1173 | if (mindelta.tv64 < 0) |
1174 | mindelta.tv64 = 0; | |
1175 | return mindelta; | |
1176 | } | |
1177 | #endif | |
1178 | ||
237fc6e7 TG |
1179 | static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id, |
1180 | enum hrtimer_mode mode) | |
c0a31329 | 1181 | { |
3c8aa39d | 1182 | struct hrtimer_cpu_base *cpu_base; |
e06383db | 1183 | int base; |
c0a31329 | 1184 | |
7978672c GA |
1185 | memset(timer, 0, sizeof(struct hrtimer)); |
1186 | ||
3c8aa39d | 1187 | cpu_base = &__raw_get_cpu_var(hrtimer_bases); |
c0a31329 | 1188 | |
c9cb2e3d | 1189 | if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS) |
7978672c GA |
1190 | clock_id = CLOCK_MONOTONIC; |
1191 | ||
e06383db JS |
1192 | base = hrtimer_clockid_to_base(clock_id); |
1193 | timer->base = &cpu_base->clock_base[base]; | |
998adc3d | 1194 | timerqueue_init(&timer->node); |
82f67cd9 IM |
1195 | |
1196 | #ifdef CONFIG_TIMER_STATS | |
1197 | timer->start_site = NULL; | |
1198 | timer->start_pid = -1; | |
1199 | memset(timer->start_comm, 0, TASK_COMM_LEN); | |
1200 | #endif | |
c0a31329 | 1201 | } |
237fc6e7 TG |
1202 | |
1203 | /** | |
1204 | * hrtimer_init - initialize a timer to the given clock | |
1205 | * @timer: the timer to be initialized | |
1206 | * @clock_id: the clock to be used | |
1207 | * @mode: timer mode abs/rel | |
1208 | */ | |
1209 | void hrtimer_init(struct hrtimer *timer, clockid_t clock_id, | |
1210 | enum hrtimer_mode mode) | |
1211 | { | |
c6a2a177 | 1212 | debug_init(timer, clock_id, mode); |
237fc6e7 TG |
1213 | __hrtimer_init(timer, clock_id, mode); |
1214 | } | |
8d16b764 | 1215 | EXPORT_SYMBOL_GPL(hrtimer_init); |
c0a31329 TG |
1216 | |
1217 | /** | |
1218 | * hrtimer_get_res - get the timer resolution for a clock | |
c0a31329 TG |
1219 | * @which_clock: which clock to query |
1220 | * @tp: pointer to timespec variable to store the resolution | |
1221 | * | |
72fd4a35 RD |
1222 | * Store the resolution of the clock selected by @which_clock in the |
1223 | * variable pointed to by @tp. | |
c0a31329 TG |
1224 | */ |
1225 | int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp) | |
1226 | { | |
3c8aa39d | 1227 | struct hrtimer_cpu_base *cpu_base; |
e06383db | 1228 | int base = hrtimer_clockid_to_base(which_clock); |
c0a31329 | 1229 | |
3c8aa39d | 1230 | cpu_base = &__raw_get_cpu_var(hrtimer_bases); |
e06383db | 1231 | *tp = ktime_to_timespec(cpu_base->clock_base[base].resolution); |
c0a31329 TG |
1232 | |
1233 | return 0; | |
1234 | } | |
8d16b764 | 1235 | EXPORT_SYMBOL_GPL(hrtimer_get_res); |
c0a31329 | 1236 | |
c6a2a177 | 1237 | static void __run_hrtimer(struct hrtimer *timer, ktime_t *now) |
d3d74453 PZ |
1238 | { |
1239 | struct hrtimer_clock_base *base = timer->base; | |
1240 | struct hrtimer_cpu_base *cpu_base = base->cpu_base; | |
1241 | enum hrtimer_restart (*fn)(struct hrtimer *); | |
1242 | int restart; | |
1243 | ||
ca109491 PZ |
1244 | WARN_ON(!irqs_disabled()); |
1245 | ||
c6a2a177 | 1246 | debug_deactivate(timer); |
d3d74453 PZ |
1247 | __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0); |
1248 | timer_stats_account_hrtimer(timer); | |
d3d74453 | 1249 | fn = timer->function; |
ca109491 PZ |
1250 | |
1251 | /* | |
1252 | * Because we run timers from hardirq context, there is no chance | |
1253 | * they get migrated to another cpu, therefore its safe to unlock | |
1254 | * the timer base. | |
1255 | */ | |
ecb49d1a | 1256 | raw_spin_unlock(&cpu_base->lock); |
c6a2a177 | 1257 | trace_hrtimer_expire_entry(timer, now); |
ca109491 | 1258 | restart = fn(timer); |
c6a2a177 | 1259 | trace_hrtimer_expire_exit(timer); |
ecb49d1a | 1260 | raw_spin_lock(&cpu_base->lock); |
d3d74453 PZ |
1261 | |
1262 | /* | |
e3f1d883 TG |
1263 | * Note: We clear the CALLBACK bit after enqueue_hrtimer and |
1264 | * we do not reprogramm the event hardware. Happens either in | |
1265 | * hrtimer_start_range_ns() or in hrtimer_interrupt() | |
d3d74453 PZ |
1266 | */ |
1267 | if (restart != HRTIMER_NORESTART) { | |
1268 | BUG_ON(timer->state != HRTIMER_STATE_CALLBACK); | |
a6037b61 | 1269 | enqueue_hrtimer(timer, base); |
d3d74453 | 1270 | } |
f13d4f97 SQ |
1271 | |
1272 | WARN_ON_ONCE(!(timer->state & HRTIMER_STATE_CALLBACK)); | |
1273 | ||
d3d74453 PZ |
1274 | timer->state &= ~HRTIMER_STATE_CALLBACK; |
1275 | } | |
1276 | ||
54cdfdb4 TG |
1277 | #ifdef CONFIG_HIGH_RES_TIMERS |
1278 | ||
1279 | /* | |
1280 | * High resolution timer interrupt | |
1281 | * Called with interrupts disabled | |
1282 | */ | |
1283 | void hrtimer_interrupt(struct clock_event_device *dev) | |
1284 | { | |
1285 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); | |
41d2e494 TG |
1286 | ktime_t expires_next, now, entry_time, delta; |
1287 | int i, retries = 0; | |
54cdfdb4 TG |
1288 | |
1289 | BUG_ON(!cpu_base->hres_active); | |
1290 | cpu_base->nr_events++; | |
1291 | dev->next_event.tv64 = KTIME_MAX; | |
1292 | ||
196951e9 | 1293 | raw_spin_lock(&cpu_base->lock); |
5baefd6d | 1294 | entry_time = now = hrtimer_update_base(cpu_base); |
41d2e494 | 1295 | retry: |
54cdfdb4 | 1296 | expires_next.tv64 = KTIME_MAX; |
6ff7041d TG |
1297 | /* |
1298 | * We set expires_next to KTIME_MAX here with cpu_base->lock | |
1299 | * held to prevent that a timer is enqueued in our queue via | |
1300 | * the migration code. This does not affect enqueueing of | |
1301 | * timers which run their callback and need to be requeued on | |
1302 | * this CPU. | |
1303 | */ | |
1304 | cpu_base->expires_next.tv64 = KTIME_MAX; | |
1305 | ||
54cdfdb4 | 1306 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
ab8177bc | 1307 | struct hrtimer_clock_base *base; |
998adc3d | 1308 | struct timerqueue_node *node; |
ab8177bc TG |
1309 | ktime_t basenow; |
1310 | ||
1311 | if (!(cpu_base->active_bases & (1 << i))) | |
1312 | continue; | |
54cdfdb4 | 1313 | |
ab8177bc | 1314 | base = cpu_base->clock_base + i; |
54cdfdb4 TG |
1315 | basenow = ktime_add(now, base->offset); |
1316 | ||
998adc3d | 1317 | while ((node = timerqueue_getnext(&base->active))) { |
54cdfdb4 TG |
1318 | struct hrtimer *timer; |
1319 | ||
998adc3d | 1320 | timer = container_of(node, struct hrtimer, node); |
54cdfdb4 | 1321 | |
654c8e0b AV |
1322 | /* |
1323 | * The immediate goal for using the softexpires is | |
1324 | * minimizing wakeups, not running timers at the | |
1325 | * earliest interrupt after their soft expiration. | |
1326 | * This allows us to avoid using a Priority Search | |
1327 | * Tree, which can answer a stabbing querry for | |
1328 | * overlapping intervals and instead use the simple | |
1329 | * BST we already have. | |
1330 | * We don't add extra wakeups by delaying timers that | |
1331 | * are right-of a not yet expired timer, because that | |
1332 | * timer will have to trigger a wakeup anyway. | |
1333 | */ | |
1334 | ||
1335 | if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer)) { | |
54cdfdb4 TG |
1336 | ktime_t expires; |
1337 | ||
cc584b21 | 1338 | expires = ktime_sub(hrtimer_get_expires(timer), |
54cdfdb4 | 1339 | base->offset); |
8f294b5a PB |
1340 | if (expires.tv64 < 0) |
1341 | expires.tv64 = KTIME_MAX; | |
54cdfdb4 TG |
1342 | if (expires.tv64 < expires_next.tv64) |
1343 | expires_next = expires; | |
1344 | break; | |
1345 | } | |
1346 | ||
c6a2a177 | 1347 | __run_hrtimer(timer, &basenow); |
54cdfdb4 | 1348 | } |
54cdfdb4 TG |
1349 | } |
1350 | ||
6ff7041d TG |
1351 | /* |
1352 | * Store the new expiry value so the migration code can verify | |
1353 | * against it. | |
1354 | */ | |
54cdfdb4 | 1355 | cpu_base->expires_next = expires_next; |
ecb49d1a | 1356 | raw_spin_unlock(&cpu_base->lock); |
54cdfdb4 TG |
1357 | |
1358 | /* Reprogramming necessary ? */ | |
41d2e494 TG |
1359 | if (expires_next.tv64 == KTIME_MAX || |
1360 | !tick_program_event(expires_next, 0)) { | |
1361 | cpu_base->hang_detected = 0; | |
1362 | return; | |
54cdfdb4 | 1363 | } |
41d2e494 TG |
1364 | |
1365 | /* | |
1366 | * The next timer was already expired due to: | |
1367 | * - tracing | |
1368 | * - long lasting callbacks | |
1369 | * - being scheduled away when running in a VM | |
1370 | * | |
1371 | * We need to prevent that we loop forever in the hrtimer | |
1372 | * interrupt routine. We give it 3 attempts to avoid | |
1373 | * overreacting on some spurious event. | |
5baefd6d JS |
1374 | * |
1375 | * Acquire base lock for updating the offsets and retrieving | |
1376 | * the current time. | |
41d2e494 | 1377 | */ |
196951e9 | 1378 | raw_spin_lock(&cpu_base->lock); |
5baefd6d | 1379 | now = hrtimer_update_base(cpu_base); |
41d2e494 TG |
1380 | cpu_base->nr_retries++; |
1381 | if (++retries < 3) | |
1382 | goto retry; | |
1383 | /* | |
1384 | * Give the system a chance to do something else than looping | |
1385 | * here. We stored the entry time, so we know exactly how long | |
1386 | * we spent here. We schedule the next event this amount of | |
1387 | * time away. | |
1388 | */ | |
1389 | cpu_base->nr_hangs++; | |
1390 | cpu_base->hang_detected = 1; | |
196951e9 | 1391 | raw_spin_unlock(&cpu_base->lock); |
41d2e494 TG |
1392 | delta = ktime_sub(now, entry_time); |
1393 | if (delta.tv64 > cpu_base->max_hang_time.tv64) | |
1394 | cpu_base->max_hang_time = delta; | |
1395 | /* | |
1396 | * Limit it to a sensible value as we enforce a longer | |
1397 | * delay. Give the CPU at least 100ms to catch up. | |
1398 | */ | |
1399 | if (delta.tv64 > 100 * NSEC_PER_MSEC) | |
1400 | expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC); | |
1401 | else | |
1402 | expires_next = ktime_add(now, delta); | |
1403 | tick_program_event(expires_next, 1); | |
1404 | printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n", | |
1405 | ktime_to_ns(delta)); | |
54cdfdb4 TG |
1406 | } |
1407 | ||
8bdec955 TG |
1408 | /* |
1409 | * local version of hrtimer_peek_ahead_timers() called with interrupts | |
1410 | * disabled. | |
1411 | */ | |
1412 | static void __hrtimer_peek_ahead_timers(void) | |
1413 | { | |
1414 | struct tick_device *td; | |
1415 | ||
1416 | if (!hrtimer_hres_active()) | |
1417 | return; | |
1418 | ||
1419 | td = &__get_cpu_var(tick_cpu_device); | |
1420 | if (td && td->evtdev) | |
1421 | hrtimer_interrupt(td->evtdev); | |
1422 | } | |
1423 | ||
2e94d1f7 AV |
1424 | /** |
1425 | * hrtimer_peek_ahead_timers -- run soft-expired timers now | |
1426 | * | |
1427 | * hrtimer_peek_ahead_timers will peek at the timer queue of | |
1428 | * the current cpu and check if there are any timers for which | |
1429 | * the soft expires time has passed. If any such timers exist, | |
1430 | * they are run immediately and then removed from the timer queue. | |
1431 | * | |
1432 | */ | |
1433 | void hrtimer_peek_ahead_timers(void) | |
1434 | { | |
643bdf68 | 1435 | unsigned long flags; |
dc4304f7 | 1436 | |
2e94d1f7 | 1437 | local_irq_save(flags); |
8bdec955 | 1438 | __hrtimer_peek_ahead_timers(); |
2e94d1f7 AV |
1439 | local_irq_restore(flags); |
1440 | } | |
1441 | ||
a6037b61 PZ |
1442 | static void run_hrtimer_softirq(struct softirq_action *h) |
1443 | { | |
1444 | hrtimer_peek_ahead_timers(); | |
1445 | } | |
1446 | ||
82c5b7b5 IM |
1447 | #else /* CONFIG_HIGH_RES_TIMERS */ |
1448 | ||
1449 | static inline void __hrtimer_peek_ahead_timers(void) { } | |
1450 | ||
1451 | #endif /* !CONFIG_HIGH_RES_TIMERS */ | |
82f67cd9 | 1452 | |
d3d74453 PZ |
1453 | /* |
1454 | * Called from timer softirq every jiffy, expire hrtimers: | |
1455 | * | |
1456 | * For HRT its the fall back code to run the softirq in the timer | |
1457 | * softirq context in case the hrtimer initialization failed or has | |
1458 | * not been done yet. | |
1459 | */ | |
1460 | void hrtimer_run_pending(void) | |
1461 | { | |
d3d74453 PZ |
1462 | if (hrtimer_hres_active()) |
1463 | return; | |
54cdfdb4 | 1464 | |
d3d74453 PZ |
1465 | /* |
1466 | * This _is_ ugly: We have to check in the softirq context, | |
1467 | * whether we can switch to highres and / or nohz mode. The | |
1468 | * clocksource switch happens in the timer interrupt with | |
1469 | * xtime_lock held. Notification from there only sets the | |
1470 | * check bit in the tick_oneshot code, otherwise we might | |
1471 | * deadlock vs. xtime_lock. | |
1472 | */ | |
1473 | if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) | |
1474 | hrtimer_switch_to_hres(); | |
54cdfdb4 TG |
1475 | } |
1476 | ||
c0a31329 | 1477 | /* |
d3d74453 | 1478 | * Called from hardirq context every jiffy |
c0a31329 | 1479 | */ |
833883d9 | 1480 | void hrtimer_run_queues(void) |
c0a31329 | 1481 | { |
998adc3d | 1482 | struct timerqueue_node *node; |
833883d9 DS |
1483 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
1484 | struct hrtimer_clock_base *base; | |
1485 | int index, gettime = 1; | |
c0a31329 | 1486 | |
833883d9 | 1487 | if (hrtimer_hres_active()) |
3055adda DS |
1488 | return; |
1489 | ||
833883d9 DS |
1490 | for (index = 0; index < HRTIMER_MAX_CLOCK_BASES; index++) { |
1491 | base = &cpu_base->clock_base[index]; | |
b007c389 | 1492 | if (!timerqueue_getnext(&base->active)) |
d3d74453 | 1493 | continue; |
833883d9 | 1494 | |
d7cfb60c | 1495 | if (gettime) { |
833883d9 DS |
1496 | hrtimer_get_softirq_time(cpu_base); |
1497 | gettime = 0; | |
b75f7a51 | 1498 | } |
d3d74453 | 1499 | |
ecb49d1a | 1500 | raw_spin_lock(&cpu_base->lock); |
c0a31329 | 1501 | |
b007c389 | 1502 | while ((node = timerqueue_getnext(&base->active))) { |
833883d9 | 1503 | struct hrtimer *timer; |
54cdfdb4 | 1504 | |
998adc3d | 1505 | timer = container_of(node, struct hrtimer, node); |
cc584b21 AV |
1506 | if (base->softirq_time.tv64 <= |
1507 | hrtimer_get_expires_tv64(timer)) | |
833883d9 DS |
1508 | break; |
1509 | ||
c6a2a177 | 1510 | __run_hrtimer(timer, &base->softirq_time); |
833883d9 | 1511 | } |
ecb49d1a | 1512 | raw_spin_unlock(&cpu_base->lock); |
833883d9 | 1513 | } |
c0a31329 TG |
1514 | } |
1515 | ||
10c94ec1 TG |
1516 | /* |
1517 | * Sleep related functions: | |
1518 | */ | |
c9cb2e3d | 1519 | static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer) |
00362e33 TG |
1520 | { |
1521 | struct hrtimer_sleeper *t = | |
1522 | container_of(timer, struct hrtimer_sleeper, timer); | |
1523 | struct task_struct *task = t->task; | |
1524 | ||
1525 | t->task = NULL; | |
1526 | if (task) | |
1527 | wake_up_process(task); | |
1528 | ||
1529 | return HRTIMER_NORESTART; | |
1530 | } | |
1531 | ||
36c8b586 | 1532 | void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task) |
00362e33 TG |
1533 | { |
1534 | sl->timer.function = hrtimer_wakeup; | |
1535 | sl->task = task; | |
1536 | } | |
2bc481cf | 1537 | EXPORT_SYMBOL_GPL(hrtimer_init_sleeper); |
00362e33 | 1538 | |
669d7868 | 1539 | static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode) |
432569bb | 1540 | { |
669d7868 | 1541 | hrtimer_init_sleeper(t, current); |
10c94ec1 | 1542 | |
432569bb RZ |
1543 | do { |
1544 | set_current_state(TASK_INTERRUPTIBLE); | |
cc584b21 | 1545 | hrtimer_start_expires(&t->timer, mode); |
37bb6cb4 PZ |
1546 | if (!hrtimer_active(&t->timer)) |
1547 | t->task = NULL; | |
432569bb | 1548 | |
54cdfdb4 | 1549 | if (likely(t->task)) |
b0f8c44f | 1550 | freezable_schedule(); |
432569bb | 1551 | |
669d7868 | 1552 | hrtimer_cancel(&t->timer); |
c9cb2e3d | 1553 | mode = HRTIMER_MODE_ABS; |
669d7868 TG |
1554 | |
1555 | } while (t->task && !signal_pending(current)); | |
432569bb | 1556 | |
3588a085 PZ |
1557 | __set_current_state(TASK_RUNNING); |
1558 | ||
669d7868 | 1559 | return t->task == NULL; |
10c94ec1 TG |
1560 | } |
1561 | ||
080344b9 ON |
1562 | static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp) |
1563 | { | |
1564 | struct timespec rmt; | |
1565 | ktime_t rem; | |
1566 | ||
cc584b21 | 1567 | rem = hrtimer_expires_remaining(timer); |
080344b9 ON |
1568 | if (rem.tv64 <= 0) |
1569 | return 0; | |
1570 | rmt = ktime_to_timespec(rem); | |
1571 | ||
1572 | if (copy_to_user(rmtp, &rmt, sizeof(*rmtp))) | |
1573 | return -EFAULT; | |
1574 | ||
1575 | return 1; | |
1576 | } | |
1577 | ||
1711ef38 | 1578 | long __sched hrtimer_nanosleep_restart(struct restart_block *restart) |
10c94ec1 | 1579 | { |
669d7868 | 1580 | struct hrtimer_sleeper t; |
080344b9 | 1581 | struct timespec __user *rmtp; |
237fc6e7 | 1582 | int ret = 0; |
10c94ec1 | 1583 | |
ab8177bc | 1584 | hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid, |
237fc6e7 | 1585 | HRTIMER_MODE_ABS); |
cc584b21 | 1586 | hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires); |
10c94ec1 | 1587 | |
c9cb2e3d | 1588 | if (do_nanosleep(&t, HRTIMER_MODE_ABS)) |
237fc6e7 | 1589 | goto out; |
10c94ec1 | 1590 | |
029a07e0 | 1591 | rmtp = restart->nanosleep.rmtp; |
432569bb | 1592 | if (rmtp) { |
237fc6e7 | 1593 | ret = update_rmtp(&t.timer, rmtp); |
080344b9 | 1594 | if (ret <= 0) |
237fc6e7 | 1595 | goto out; |
432569bb | 1596 | } |
10c94ec1 | 1597 | |
10c94ec1 | 1598 | /* The other values in restart are already filled in */ |
237fc6e7 TG |
1599 | ret = -ERESTART_RESTARTBLOCK; |
1600 | out: | |
1601 | destroy_hrtimer_on_stack(&t.timer); | |
1602 | return ret; | |
10c94ec1 TG |
1603 | } |
1604 | ||
080344b9 | 1605 | long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp, |
10c94ec1 TG |
1606 | const enum hrtimer_mode mode, const clockid_t clockid) |
1607 | { | |
1608 | struct restart_block *restart; | |
669d7868 | 1609 | struct hrtimer_sleeper t; |
237fc6e7 | 1610 | int ret = 0; |
3bd01206 AV |
1611 | unsigned long slack; |
1612 | ||
1613 | slack = current->timer_slack_ns; | |
aab03e05 | 1614 | if (dl_task(current) || rt_task(current)) |
3bd01206 | 1615 | slack = 0; |
10c94ec1 | 1616 | |
237fc6e7 | 1617 | hrtimer_init_on_stack(&t.timer, clockid, mode); |
3bd01206 | 1618 | hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack); |
432569bb | 1619 | if (do_nanosleep(&t, mode)) |
237fc6e7 | 1620 | goto out; |
10c94ec1 | 1621 | |
7978672c | 1622 | /* Absolute timers do not update the rmtp value and restart: */ |
237fc6e7 TG |
1623 | if (mode == HRTIMER_MODE_ABS) { |
1624 | ret = -ERESTARTNOHAND; | |
1625 | goto out; | |
1626 | } | |
10c94ec1 | 1627 | |
432569bb | 1628 | if (rmtp) { |
237fc6e7 | 1629 | ret = update_rmtp(&t.timer, rmtp); |
080344b9 | 1630 | if (ret <= 0) |
237fc6e7 | 1631 | goto out; |
432569bb | 1632 | } |
10c94ec1 TG |
1633 | |
1634 | restart = ¤t_thread_info()->restart_block; | |
1711ef38 | 1635 | restart->fn = hrtimer_nanosleep_restart; |
ab8177bc | 1636 | restart->nanosleep.clockid = t.timer.base->clockid; |
029a07e0 | 1637 | restart->nanosleep.rmtp = rmtp; |
cc584b21 | 1638 | restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer); |
10c94ec1 | 1639 | |
237fc6e7 TG |
1640 | ret = -ERESTART_RESTARTBLOCK; |
1641 | out: | |
1642 | destroy_hrtimer_on_stack(&t.timer); | |
1643 | return ret; | |
10c94ec1 TG |
1644 | } |
1645 | ||
58fd3aa2 HC |
1646 | SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp, |
1647 | struct timespec __user *, rmtp) | |
6ba1b912 | 1648 | { |
080344b9 | 1649 | struct timespec tu; |
6ba1b912 TG |
1650 | |
1651 | if (copy_from_user(&tu, rqtp, sizeof(tu))) | |
1652 | return -EFAULT; | |
1653 | ||
1654 | if (!timespec_valid(&tu)) | |
1655 | return -EINVAL; | |
1656 | ||
080344b9 | 1657 | return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC); |
6ba1b912 TG |
1658 | } |
1659 | ||
c0a31329 TG |
1660 | /* |
1661 | * Functions related to boot-time initialization: | |
1662 | */ | |
0db0628d | 1663 | static void init_hrtimers_cpu(int cpu) |
c0a31329 | 1664 | { |
3c8aa39d | 1665 | struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu); |
c0a31329 TG |
1666 | int i; |
1667 | ||
998adc3d | 1668 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
3c8aa39d | 1669 | cpu_base->clock_base[i].cpu_base = cpu_base; |
998adc3d JS |
1670 | timerqueue_init_head(&cpu_base->clock_base[i].active); |
1671 | } | |
3c8aa39d | 1672 | |
54cdfdb4 | 1673 | hrtimer_init_hres(cpu_base); |
c0a31329 TG |
1674 | } |
1675 | ||
1676 | #ifdef CONFIG_HOTPLUG_CPU | |
1677 | ||
ca109491 | 1678 | static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base, |
37810659 | 1679 | struct hrtimer_clock_base *new_base) |
c0a31329 TG |
1680 | { |
1681 | struct hrtimer *timer; | |
998adc3d | 1682 | struct timerqueue_node *node; |
c0a31329 | 1683 | |
998adc3d JS |
1684 | while ((node = timerqueue_getnext(&old_base->active))) { |
1685 | timer = container_of(node, struct hrtimer, node); | |
54cdfdb4 | 1686 | BUG_ON(hrtimer_callback_running(timer)); |
c6a2a177 | 1687 | debug_deactivate(timer); |
b00c1a99 TG |
1688 | |
1689 | /* | |
1690 | * Mark it as STATE_MIGRATE not INACTIVE otherwise the | |
1691 | * timer could be seen as !active and just vanish away | |
1692 | * under us on another CPU | |
1693 | */ | |
1694 | __remove_hrtimer(timer, old_base, HRTIMER_STATE_MIGRATE, 0); | |
c0a31329 | 1695 | timer->base = new_base; |
54cdfdb4 | 1696 | /* |
e3f1d883 TG |
1697 | * Enqueue the timers on the new cpu. This does not |
1698 | * reprogram the event device in case the timer | |
1699 | * expires before the earliest on this CPU, but we run | |
1700 | * hrtimer_interrupt after we migrated everything to | |
1701 | * sort out already expired timers and reprogram the | |
1702 | * event device. | |
54cdfdb4 | 1703 | */ |
a6037b61 | 1704 | enqueue_hrtimer(timer, new_base); |
41e1022e | 1705 | |
b00c1a99 TG |
1706 | /* Clear the migration state bit */ |
1707 | timer->state &= ~HRTIMER_STATE_MIGRATE; | |
c0a31329 TG |
1708 | } |
1709 | } | |
1710 | ||
d5fd43c4 | 1711 | static void migrate_hrtimers(int scpu) |
c0a31329 | 1712 | { |
3c8aa39d | 1713 | struct hrtimer_cpu_base *old_base, *new_base; |
731a55ba | 1714 | int i; |
c0a31329 | 1715 | |
37810659 | 1716 | BUG_ON(cpu_online(scpu)); |
37810659 | 1717 | tick_cancel_sched_timer(scpu); |
731a55ba TG |
1718 | |
1719 | local_irq_disable(); | |
1720 | old_base = &per_cpu(hrtimer_bases, scpu); | |
1721 | new_base = &__get_cpu_var(hrtimer_bases); | |
d82f0b0f ON |
1722 | /* |
1723 | * The caller is globally serialized and nobody else | |
1724 | * takes two locks at once, deadlock is not possible. | |
1725 | */ | |
ecb49d1a TG |
1726 | raw_spin_lock(&new_base->lock); |
1727 | raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING); | |
c0a31329 | 1728 | |
3c8aa39d | 1729 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
ca109491 | 1730 | migrate_hrtimer_list(&old_base->clock_base[i], |
37810659 | 1731 | &new_base->clock_base[i]); |
c0a31329 TG |
1732 | } |
1733 | ||
ecb49d1a TG |
1734 | raw_spin_unlock(&old_base->lock); |
1735 | raw_spin_unlock(&new_base->lock); | |
37810659 | 1736 | |
731a55ba TG |
1737 | /* Check, if we got expired work to do */ |
1738 | __hrtimer_peek_ahead_timers(); | |
1739 | local_irq_enable(); | |
c0a31329 | 1740 | } |
37810659 | 1741 | |
c0a31329 TG |
1742 | #endif /* CONFIG_HOTPLUG_CPU */ |
1743 | ||
0db0628d | 1744 | static int hrtimer_cpu_notify(struct notifier_block *self, |
c0a31329 TG |
1745 | unsigned long action, void *hcpu) |
1746 | { | |
b2e3c0ad | 1747 | int scpu = (long)hcpu; |
c0a31329 TG |
1748 | |
1749 | switch (action) { | |
1750 | ||
1751 | case CPU_UP_PREPARE: | |
8bb78442 | 1752 | case CPU_UP_PREPARE_FROZEN: |
37810659 | 1753 | init_hrtimers_cpu(scpu); |
c0a31329 TG |
1754 | break; |
1755 | ||
1756 | #ifdef CONFIG_HOTPLUG_CPU | |
94df7de0 SD |
1757 | case CPU_DYING: |
1758 | case CPU_DYING_FROZEN: | |
1759 | clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DYING, &scpu); | |
1760 | break; | |
c0a31329 | 1761 | case CPU_DEAD: |
8bb78442 | 1762 | case CPU_DEAD_FROZEN: |
b2e3c0ad | 1763 | { |
37810659 | 1764 | clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &scpu); |
d5fd43c4 | 1765 | migrate_hrtimers(scpu); |
c0a31329 | 1766 | break; |
b2e3c0ad | 1767 | } |
c0a31329 TG |
1768 | #endif |
1769 | ||
1770 | default: | |
1771 | break; | |
1772 | } | |
1773 | ||
1774 | return NOTIFY_OK; | |
1775 | } | |
1776 | ||
0db0628d | 1777 | static struct notifier_block hrtimers_nb = { |
c0a31329 TG |
1778 | .notifier_call = hrtimer_cpu_notify, |
1779 | }; | |
1780 | ||
1781 | void __init hrtimers_init(void) | |
1782 | { | |
1783 | hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE, | |
1784 | (void *)(long)smp_processor_id()); | |
1785 | register_cpu_notifier(&hrtimers_nb); | |
a6037b61 PZ |
1786 | #ifdef CONFIG_HIGH_RES_TIMERS |
1787 | open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq); | |
1788 | #endif | |
c0a31329 TG |
1789 | } |
1790 | ||
7bb67439 | 1791 | /** |
351b3f7a | 1792 | * schedule_hrtimeout_range_clock - sleep until timeout |
7bb67439 | 1793 | * @expires: timeout value (ktime_t) |
654c8e0b | 1794 | * @delta: slack in expires timeout (ktime_t) |
7bb67439 | 1795 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL |
351b3f7a | 1796 | * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME |
7bb67439 | 1797 | */ |
351b3f7a CE |
1798 | int __sched |
1799 | schedule_hrtimeout_range_clock(ktime_t *expires, unsigned long delta, | |
1800 | const enum hrtimer_mode mode, int clock) | |
7bb67439 AV |
1801 | { |
1802 | struct hrtimer_sleeper t; | |
1803 | ||
1804 | /* | |
1805 | * Optimize when a zero timeout value is given. It does not | |
1806 | * matter whether this is an absolute or a relative time. | |
1807 | */ | |
1808 | if (expires && !expires->tv64) { | |
1809 | __set_current_state(TASK_RUNNING); | |
1810 | return 0; | |
1811 | } | |
1812 | ||
1813 | /* | |
43b21013 | 1814 | * A NULL parameter means "infinite" |
7bb67439 AV |
1815 | */ |
1816 | if (!expires) { | |
1817 | schedule(); | |
1818 | __set_current_state(TASK_RUNNING); | |
1819 | return -EINTR; | |
1820 | } | |
1821 | ||
351b3f7a | 1822 | hrtimer_init_on_stack(&t.timer, clock, mode); |
654c8e0b | 1823 | hrtimer_set_expires_range_ns(&t.timer, *expires, delta); |
7bb67439 AV |
1824 | |
1825 | hrtimer_init_sleeper(&t, current); | |
1826 | ||
cc584b21 | 1827 | hrtimer_start_expires(&t.timer, mode); |
7bb67439 AV |
1828 | if (!hrtimer_active(&t.timer)) |
1829 | t.task = NULL; | |
1830 | ||
1831 | if (likely(t.task)) | |
1832 | schedule(); | |
1833 | ||
1834 | hrtimer_cancel(&t.timer); | |
1835 | destroy_hrtimer_on_stack(&t.timer); | |
1836 | ||
1837 | __set_current_state(TASK_RUNNING); | |
1838 | ||
1839 | return !t.task ? 0 : -EINTR; | |
1840 | } | |
351b3f7a CE |
1841 | |
1842 | /** | |
1843 | * schedule_hrtimeout_range - sleep until timeout | |
1844 | * @expires: timeout value (ktime_t) | |
1845 | * @delta: slack in expires timeout (ktime_t) | |
1846 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL | |
1847 | * | |
1848 | * Make the current task sleep until the given expiry time has | |
1849 | * elapsed. The routine will return immediately unless | |
1850 | * the current task state has been set (see set_current_state()). | |
1851 | * | |
1852 | * The @delta argument gives the kernel the freedom to schedule the | |
1853 | * actual wakeup to a time that is both power and performance friendly. | |
1854 | * The kernel give the normal best effort behavior for "@expires+@delta", | |
1855 | * but may decide to fire the timer earlier, but no earlier than @expires. | |
1856 | * | |
1857 | * You can set the task state as follows - | |
1858 | * | |
1859 | * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to | |
1860 | * pass before the routine returns. | |
1861 | * | |
1862 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
1863 | * delivered to the current task. | |
1864 | * | |
1865 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1866 | * routine returns. | |
1867 | * | |
1868 | * Returns 0 when the timer has expired otherwise -EINTR | |
1869 | */ | |
1870 | int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta, | |
1871 | const enum hrtimer_mode mode) | |
1872 | { | |
1873 | return schedule_hrtimeout_range_clock(expires, delta, mode, | |
1874 | CLOCK_MONOTONIC); | |
1875 | } | |
654c8e0b AV |
1876 | EXPORT_SYMBOL_GPL(schedule_hrtimeout_range); |
1877 | ||
1878 | /** | |
1879 | * schedule_hrtimeout - sleep until timeout | |
1880 | * @expires: timeout value (ktime_t) | |
1881 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL | |
1882 | * | |
1883 | * Make the current task sleep until the given expiry time has | |
1884 | * elapsed. The routine will return immediately unless | |
1885 | * the current task state has been set (see set_current_state()). | |
1886 | * | |
1887 | * You can set the task state as follows - | |
1888 | * | |
1889 | * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to | |
1890 | * pass before the routine returns. | |
1891 | * | |
1892 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
1893 | * delivered to the current task. | |
1894 | * | |
1895 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1896 | * routine returns. | |
1897 | * | |
1898 | * Returns 0 when the timer has expired otherwise -EINTR | |
1899 | */ | |
1900 | int __sched schedule_hrtimeout(ktime_t *expires, | |
1901 | const enum hrtimer_mode mode) | |
1902 | { | |
1903 | return schedule_hrtimeout_range(expires, 0, mode); | |
1904 | } | |
7bb67439 | 1905 | EXPORT_SYMBOL_GPL(schedule_hrtimeout); |