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 | * With HIGHRES=y we do not migrate the timer when it is expiring | |
173 | * before the next event on the target cpu because we cannot reprogram | |
174 | * the target cpu hardware and we would cause it to fire late. | |
175 | * | |
176 | * Called with cpu_base->lock of target cpu held. | |
177 | */ | |
178 | static int | |
179 | hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base) | |
180 | { | |
181 | #ifdef CONFIG_HIGH_RES_TIMERS | |
182 | ktime_t expires; | |
183 | ||
184 | if (!new_base->cpu_base->hres_active) | |
185 | return 0; | |
186 | ||
187 | expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset); | |
188 | return expires.tv64 <= new_base->cpu_base->expires_next.tv64; | |
189 | #else | |
190 | return 0; | |
191 | #endif | |
192 | } | |
193 | ||
c0a31329 TG |
194 | /* |
195 | * Switch the timer base to the current CPU when possible. | |
196 | */ | |
3c8aa39d | 197 | static inline struct hrtimer_clock_base * |
597d0275 AB |
198 | switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base, |
199 | int pinned) | |
c0a31329 | 200 | { |
3c8aa39d TG |
201 | struct hrtimer_clock_base *new_base; |
202 | struct hrtimer_cpu_base *new_cpu_base; | |
6ff7041d | 203 | int this_cpu = smp_processor_id(); |
6201b4d6 | 204 | int cpu = get_nohz_timer_target(pinned); |
ab8177bc | 205 | int basenum = base->index; |
c0a31329 | 206 | |
eea08f32 AB |
207 | again: |
208 | new_cpu_base = &per_cpu(hrtimer_bases, cpu); | |
e06383db | 209 | new_base = &new_cpu_base->clock_base[basenum]; |
c0a31329 TG |
210 | |
211 | if (base != new_base) { | |
212 | /* | |
6ff7041d | 213 | * We are trying to move timer to new_base. |
c0a31329 TG |
214 | * However we can't change timer's base while it is running, |
215 | * so we keep it on the same CPU. No hassle vs. reprogramming | |
216 | * the event source in the high resolution case. The softirq | |
217 | * code will take care of this when the timer function has | |
218 | * completed. There is no conflict as we hold the lock until | |
219 | * the timer is enqueued. | |
220 | */ | |
54cdfdb4 | 221 | if (unlikely(hrtimer_callback_running(timer))) |
c0a31329 TG |
222 | return base; |
223 | ||
224 | /* See the comment in lock_timer_base() */ | |
225 | timer->base = NULL; | |
ecb49d1a TG |
226 | raw_spin_unlock(&base->cpu_base->lock); |
227 | raw_spin_lock(&new_base->cpu_base->lock); | |
eea08f32 | 228 | |
6ff7041d TG |
229 | if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) { |
230 | cpu = this_cpu; | |
ecb49d1a TG |
231 | raw_spin_unlock(&new_base->cpu_base->lock); |
232 | raw_spin_lock(&base->cpu_base->lock); | |
6ff7041d TG |
233 | timer->base = base; |
234 | goto again; | |
eea08f32 | 235 | } |
c0a31329 | 236 | timer->base = new_base; |
012a45e3 LM |
237 | } else { |
238 | if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) { | |
239 | cpu = this_cpu; | |
240 | goto again; | |
241 | } | |
c0a31329 TG |
242 | } |
243 | return new_base; | |
244 | } | |
245 | ||
246 | #else /* CONFIG_SMP */ | |
247 | ||
3c8aa39d | 248 | static inline struct hrtimer_clock_base * |
c0a31329 TG |
249 | lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) |
250 | { | |
3c8aa39d | 251 | struct hrtimer_clock_base *base = timer->base; |
c0a31329 | 252 | |
ecb49d1a | 253 | raw_spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
254 | |
255 | return base; | |
256 | } | |
257 | ||
eea08f32 | 258 | # define switch_hrtimer_base(t, b, p) (b) |
c0a31329 TG |
259 | |
260 | #endif /* !CONFIG_SMP */ | |
261 | ||
262 | /* | |
263 | * Functions for the union type storage format of ktime_t which are | |
264 | * too large for inlining: | |
265 | */ | |
266 | #if BITS_PER_LONG < 64 | |
267 | # ifndef CONFIG_KTIME_SCALAR | |
268 | /** | |
269 | * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable | |
c0a31329 TG |
270 | * @kt: addend |
271 | * @nsec: the scalar nsec value to add | |
272 | * | |
273 | * Returns the sum of kt and nsec in ktime_t format | |
274 | */ | |
275 | ktime_t ktime_add_ns(const ktime_t kt, u64 nsec) | |
276 | { | |
277 | ktime_t tmp; | |
278 | ||
279 | if (likely(nsec < NSEC_PER_SEC)) { | |
280 | tmp.tv64 = nsec; | |
281 | } else { | |
282 | unsigned long rem = do_div(nsec, NSEC_PER_SEC); | |
283 | ||
51fd36f3 DE |
284 | /* Make sure nsec fits into long */ |
285 | if (unlikely(nsec > KTIME_SEC_MAX)) | |
286 | return (ktime_t){ .tv64 = KTIME_MAX }; | |
287 | ||
c0a31329 TG |
288 | tmp = ktime_set((long)nsec, rem); |
289 | } | |
290 | ||
291 | return ktime_add(kt, tmp); | |
292 | } | |
b8b8fd2d DH |
293 | |
294 | EXPORT_SYMBOL_GPL(ktime_add_ns); | |
a272378d ACM |
295 | |
296 | /** | |
297 | * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable | |
298 | * @kt: minuend | |
299 | * @nsec: the scalar nsec value to subtract | |
300 | * | |
301 | * Returns the subtraction of @nsec from @kt in ktime_t format | |
302 | */ | |
303 | ktime_t ktime_sub_ns(const ktime_t kt, u64 nsec) | |
304 | { | |
305 | ktime_t tmp; | |
306 | ||
307 | if (likely(nsec < NSEC_PER_SEC)) { | |
308 | tmp.tv64 = nsec; | |
309 | } else { | |
310 | unsigned long rem = do_div(nsec, NSEC_PER_SEC); | |
311 | ||
312 | tmp = ktime_set((long)nsec, rem); | |
313 | } | |
314 | ||
315 | return ktime_sub(kt, tmp); | |
316 | } | |
317 | ||
318 | EXPORT_SYMBOL_GPL(ktime_sub_ns); | |
c0a31329 TG |
319 | # endif /* !CONFIG_KTIME_SCALAR */ |
320 | ||
321 | /* | |
322 | * Divide a ktime value by a nanosecond value | |
323 | */ | |
4d672e7a | 324 | u64 ktime_divns(const ktime_t kt, s64 div) |
c0a31329 | 325 | { |
900cfa46 | 326 | u64 dclc; |
c0a31329 TG |
327 | int sft = 0; |
328 | ||
900cfa46 | 329 | dclc = ktime_to_ns(kt); |
c0a31329 TG |
330 | /* Make sure the divisor is less than 2^32: */ |
331 | while (div >> 32) { | |
332 | sft++; | |
333 | div >>= 1; | |
334 | } | |
335 | dclc >>= sft; | |
336 | do_div(dclc, (unsigned long) div); | |
337 | ||
4d672e7a | 338 | return dclc; |
c0a31329 | 339 | } |
c0a31329 TG |
340 | #endif /* BITS_PER_LONG >= 64 */ |
341 | ||
5a7780e7 TG |
342 | /* |
343 | * Add two ktime values and do a safety check for overflow: | |
344 | */ | |
345 | ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs) | |
346 | { | |
347 | ktime_t res = ktime_add(lhs, rhs); | |
348 | ||
349 | /* | |
350 | * We use KTIME_SEC_MAX here, the maximum timeout which we can | |
351 | * return to user space in a timespec: | |
352 | */ | |
353 | if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64) | |
354 | res = ktime_set(KTIME_SEC_MAX, 0); | |
355 | ||
356 | return res; | |
357 | } | |
358 | ||
8daa21e6 AB |
359 | EXPORT_SYMBOL_GPL(ktime_add_safe); |
360 | ||
237fc6e7 TG |
361 | #ifdef CONFIG_DEBUG_OBJECTS_TIMERS |
362 | ||
363 | static struct debug_obj_descr hrtimer_debug_descr; | |
364 | ||
99777288 SG |
365 | static void *hrtimer_debug_hint(void *addr) |
366 | { | |
367 | return ((struct hrtimer *) addr)->function; | |
368 | } | |
369 | ||
237fc6e7 TG |
370 | /* |
371 | * fixup_init is called when: | |
372 | * - an active object is initialized | |
373 | */ | |
374 | static int hrtimer_fixup_init(void *addr, enum debug_obj_state state) | |
375 | { | |
376 | struct hrtimer *timer = addr; | |
377 | ||
378 | switch (state) { | |
379 | case ODEBUG_STATE_ACTIVE: | |
380 | hrtimer_cancel(timer); | |
381 | debug_object_init(timer, &hrtimer_debug_descr); | |
382 | return 1; | |
383 | default: | |
384 | return 0; | |
385 | } | |
386 | } | |
387 | ||
388 | /* | |
389 | * fixup_activate is called when: | |
390 | * - an active object is activated | |
391 | * - an unknown object is activated (might be a statically initialized object) | |
392 | */ | |
393 | static int hrtimer_fixup_activate(void *addr, enum debug_obj_state state) | |
394 | { | |
395 | switch (state) { | |
396 | ||
397 | case ODEBUG_STATE_NOTAVAILABLE: | |
398 | WARN_ON_ONCE(1); | |
399 | return 0; | |
400 | ||
401 | case ODEBUG_STATE_ACTIVE: | |
402 | WARN_ON(1); | |
403 | ||
404 | default: | |
405 | return 0; | |
406 | } | |
407 | } | |
408 | ||
409 | /* | |
410 | * fixup_free is called when: | |
411 | * - an active object is freed | |
412 | */ | |
413 | static int hrtimer_fixup_free(void *addr, enum debug_obj_state state) | |
414 | { | |
415 | struct hrtimer *timer = addr; | |
416 | ||
417 | switch (state) { | |
418 | case ODEBUG_STATE_ACTIVE: | |
419 | hrtimer_cancel(timer); | |
420 | debug_object_free(timer, &hrtimer_debug_descr); | |
421 | return 1; | |
422 | default: | |
423 | return 0; | |
424 | } | |
425 | } | |
426 | ||
427 | static struct debug_obj_descr hrtimer_debug_descr = { | |
428 | .name = "hrtimer", | |
99777288 | 429 | .debug_hint = hrtimer_debug_hint, |
237fc6e7 TG |
430 | .fixup_init = hrtimer_fixup_init, |
431 | .fixup_activate = hrtimer_fixup_activate, | |
432 | .fixup_free = hrtimer_fixup_free, | |
433 | }; | |
434 | ||
435 | static inline void debug_hrtimer_init(struct hrtimer *timer) | |
436 | { | |
437 | debug_object_init(timer, &hrtimer_debug_descr); | |
438 | } | |
439 | ||
440 | static inline void debug_hrtimer_activate(struct hrtimer *timer) | |
441 | { | |
442 | debug_object_activate(timer, &hrtimer_debug_descr); | |
443 | } | |
444 | ||
445 | static inline void debug_hrtimer_deactivate(struct hrtimer *timer) | |
446 | { | |
447 | debug_object_deactivate(timer, &hrtimer_debug_descr); | |
448 | } | |
449 | ||
450 | static inline void debug_hrtimer_free(struct hrtimer *timer) | |
451 | { | |
452 | debug_object_free(timer, &hrtimer_debug_descr); | |
453 | } | |
454 | ||
455 | static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id, | |
456 | enum hrtimer_mode mode); | |
457 | ||
458 | void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id, | |
459 | enum hrtimer_mode mode) | |
460 | { | |
461 | debug_object_init_on_stack(timer, &hrtimer_debug_descr); | |
462 | __hrtimer_init(timer, clock_id, mode); | |
463 | } | |
2bc481cf | 464 | EXPORT_SYMBOL_GPL(hrtimer_init_on_stack); |
237fc6e7 TG |
465 | |
466 | void destroy_hrtimer_on_stack(struct hrtimer *timer) | |
467 | { | |
468 | debug_object_free(timer, &hrtimer_debug_descr); | |
469 | } | |
470 | ||
471 | #else | |
472 | static inline void debug_hrtimer_init(struct hrtimer *timer) { } | |
473 | static inline void debug_hrtimer_activate(struct hrtimer *timer) { } | |
474 | static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { } | |
475 | #endif | |
476 | ||
c6a2a177 XG |
477 | static inline void |
478 | debug_init(struct hrtimer *timer, clockid_t clockid, | |
479 | enum hrtimer_mode mode) | |
480 | { | |
481 | debug_hrtimer_init(timer); | |
482 | trace_hrtimer_init(timer, clockid, mode); | |
483 | } | |
484 | ||
485 | static inline void debug_activate(struct hrtimer *timer) | |
486 | { | |
487 | debug_hrtimer_activate(timer); | |
488 | trace_hrtimer_start(timer); | |
489 | } | |
490 | ||
491 | static inline void debug_deactivate(struct hrtimer *timer) | |
492 | { | |
493 | debug_hrtimer_deactivate(timer); | |
494 | trace_hrtimer_cancel(timer); | |
495 | } | |
496 | ||
54cdfdb4 TG |
497 | /* High resolution timer related functions */ |
498 | #ifdef CONFIG_HIGH_RES_TIMERS | |
499 | ||
500 | /* | |
501 | * High resolution timer enabled ? | |
502 | */ | |
503 | static int hrtimer_hres_enabled __read_mostly = 1; | |
504 | ||
505 | /* | |
506 | * Enable / Disable high resolution mode | |
507 | */ | |
508 | static int __init setup_hrtimer_hres(char *str) | |
509 | { | |
510 | if (!strcmp(str, "off")) | |
511 | hrtimer_hres_enabled = 0; | |
512 | else if (!strcmp(str, "on")) | |
513 | hrtimer_hres_enabled = 1; | |
514 | else | |
515 | return 0; | |
516 | return 1; | |
517 | } | |
518 | ||
519 | __setup("highres=", setup_hrtimer_hres); | |
520 | ||
521 | /* | |
522 | * hrtimer_high_res_enabled - query, if the highres mode is enabled | |
523 | */ | |
524 | static inline int hrtimer_is_hres_enabled(void) | |
525 | { | |
526 | return hrtimer_hres_enabled; | |
527 | } | |
528 | ||
529 | /* | |
530 | * Is the high resolution mode active ? | |
531 | */ | |
532 | static inline int hrtimer_hres_active(void) | |
533 | { | |
909ea964 | 534 | return __this_cpu_read(hrtimer_bases.hres_active); |
54cdfdb4 TG |
535 | } |
536 | ||
537 | /* | |
538 | * Reprogram the event source with checking both queues for the | |
539 | * next event | |
540 | * Called with interrupts disabled and base->lock held | |
541 | */ | |
7403f41f AC |
542 | static void |
543 | hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal) | |
54cdfdb4 TG |
544 | { |
545 | int i; | |
546 | struct hrtimer_clock_base *base = cpu_base->clock_base; | |
7403f41f | 547 | ktime_t expires, expires_next; |
54cdfdb4 | 548 | |
7403f41f | 549 | expires_next.tv64 = KTIME_MAX; |
54cdfdb4 TG |
550 | |
551 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) { | |
552 | struct hrtimer *timer; | |
998adc3d | 553 | struct timerqueue_node *next; |
54cdfdb4 | 554 | |
998adc3d JS |
555 | next = timerqueue_getnext(&base->active); |
556 | if (!next) | |
54cdfdb4 | 557 | continue; |
998adc3d JS |
558 | timer = container_of(next, struct hrtimer, node); |
559 | ||
cc584b21 | 560 | expires = ktime_sub(hrtimer_get_expires(timer), base->offset); |
b0a9b511 TG |
561 | /* |
562 | * clock_was_set() has changed base->offset so the | |
563 | * result might be negative. Fix it up to prevent a | |
564 | * false positive in clockevents_program_event() | |
565 | */ | |
566 | if (expires.tv64 < 0) | |
567 | expires.tv64 = 0; | |
7403f41f AC |
568 | if (expires.tv64 < expires_next.tv64) |
569 | expires_next = expires; | |
54cdfdb4 TG |
570 | } |
571 | ||
7403f41f AC |
572 | if (skip_equal && expires_next.tv64 == cpu_base->expires_next.tv64) |
573 | return; | |
574 | ||
575 | cpu_base->expires_next.tv64 = expires_next.tv64; | |
576 | ||
6c6c0d5a SH |
577 | /* |
578 | * If a hang was detected in the last timer interrupt then we | |
579 | * leave the hang delay active in the hardware. We want the | |
580 | * system to make progress. That also prevents the following | |
581 | * scenario: | |
582 | * T1 expires 50ms from now | |
583 | * T2 expires 5s from now | |
584 | * | |
585 | * T1 is removed, so this code is called and would reprogram | |
586 | * the hardware to 5s from now. Any hrtimer_start after that | |
587 | * will not reprogram the hardware due to hang_detected being | |
588 | * set. So we'd effectivly block all timers until the T2 event | |
589 | * fires. | |
590 | */ | |
591 | if (cpu_base->hang_detected) | |
592 | return; | |
593 | ||
54cdfdb4 TG |
594 | if (cpu_base->expires_next.tv64 != KTIME_MAX) |
595 | tick_program_event(cpu_base->expires_next, 1); | |
596 | } | |
597 | ||
598 | /* | |
599 | * Shared reprogramming for clock_realtime and clock_monotonic | |
600 | * | |
601 | * When a timer is enqueued and expires earlier than the already enqueued | |
602 | * timers, we have to check, whether it expires earlier than the timer for | |
603 | * which the clock event device was armed. | |
604 | * | |
605 | * Called with interrupts disabled and base->cpu_base.lock held | |
606 | */ | |
607 | static int hrtimer_reprogram(struct hrtimer *timer, | |
608 | struct hrtimer_clock_base *base) | |
609 | { | |
41d2e494 | 610 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
cc584b21 | 611 | ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset); |
54cdfdb4 TG |
612 | int res; |
613 | ||
cc584b21 | 614 | WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0); |
63070a79 | 615 | |
54cdfdb4 TG |
616 | /* |
617 | * When the callback is running, we do not reprogram the clock event | |
618 | * device. The timer callback is either running on a different CPU or | |
3a4fa0a2 | 619 | * the callback is executed in the hrtimer_interrupt context. The |
54cdfdb4 TG |
620 | * reprogramming is handled either by the softirq, which called the |
621 | * callback or at the end of the hrtimer_interrupt. | |
622 | */ | |
623 | if (hrtimer_callback_running(timer)) | |
624 | return 0; | |
625 | ||
63070a79 TG |
626 | /* |
627 | * CLOCK_REALTIME timer might be requested with an absolute | |
628 | * expiry time which is less than base->offset. Nothing wrong | |
629 | * about that, just avoid to call into the tick code, which | |
630 | * has now objections against negative expiry values. | |
631 | */ | |
632 | if (expires.tv64 < 0) | |
633 | return -ETIME; | |
634 | ||
41d2e494 TG |
635 | if (expires.tv64 >= cpu_base->expires_next.tv64) |
636 | return 0; | |
637 | ||
638 | /* | |
639 | * If a hang was detected in the last timer interrupt then we | |
640 | * do not schedule a timer which is earlier than the expiry | |
641 | * which we enforced in the hang detection. We want the system | |
642 | * to make progress. | |
643 | */ | |
644 | if (cpu_base->hang_detected) | |
54cdfdb4 TG |
645 | return 0; |
646 | ||
647 | /* | |
648 | * Clockevents returns -ETIME, when the event was in the past. | |
649 | */ | |
650 | res = tick_program_event(expires, 0); | |
651 | if (!IS_ERR_VALUE(res)) | |
41d2e494 | 652 | cpu_base->expires_next = expires; |
54cdfdb4 TG |
653 | return res; |
654 | } | |
655 | ||
54cdfdb4 TG |
656 | /* |
657 | * Initialize the high resolution related parts of cpu_base | |
658 | */ | |
659 | static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) | |
660 | { | |
661 | base->expires_next.tv64 = KTIME_MAX; | |
662 | base->hres_active = 0; | |
54cdfdb4 TG |
663 | } |
664 | ||
54cdfdb4 TG |
665 | /* |
666 | * When High resolution timers are active, try to reprogram. Note, that in case | |
667 | * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry | |
668 | * check happens. The timer gets enqueued into the rbtree. The reprogramming | |
669 | * and expiry check is done in the hrtimer_interrupt or in the softirq. | |
670 | */ | |
671 | static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer, | |
b22affe0 | 672 | struct hrtimer_clock_base *base) |
54cdfdb4 | 673 | { |
b22affe0 | 674 | return base->cpu_base->hres_active && hrtimer_reprogram(timer, base); |
54cdfdb4 TG |
675 | } |
676 | ||
5baefd6d JS |
677 | static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base) |
678 | { | |
679 | ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset; | |
680 | ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset; | |
90adda98 | 681 | ktime_t *offs_tai = &base->clock_base[HRTIMER_BASE_TAI].offset; |
5baefd6d | 682 | |
90adda98 | 683 | return ktime_get_update_offsets(offs_real, offs_boot, offs_tai); |
5baefd6d JS |
684 | } |
685 | ||
9ec26907 TG |
686 | /* |
687 | * Retrigger next event is called after clock was set | |
688 | * | |
689 | * Called with interrupts disabled via on_each_cpu() | |
690 | */ | |
691 | static void retrigger_next_event(void *arg) | |
692 | { | |
693 | struct hrtimer_cpu_base *base = &__get_cpu_var(hrtimer_bases); | |
9ec26907 TG |
694 | |
695 | if (!hrtimer_hres_active()) | |
696 | return; | |
697 | ||
9ec26907 | 698 | raw_spin_lock(&base->lock); |
5baefd6d | 699 | hrtimer_update_base(base); |
9ec26907 TG |
700 | hrtimer_force_reprogram(base, 0); |
701 | raw_spin_unlock(&base->lock); | |
702 | } | |
b12a03ce | 703 | |
54cdfdb4 TG |
704 | /* |
705 | * Switch to high resolution mode | |
706 | */ | |
f8953856 | 707 | static int hrtimer_switch_to_hres(void) |
54cdfdb4 | 708 | { |
b12a03ce | 709 | int i, cpu = smp_processor_id(); |
820de5c3 | 710 | struct hrtimer_cpu_base *base = &per_cpu(hrtimer_bases, cpu); |
54cdfdb4 TG |
711 | unsigned long flags; |
712 | ||
713 | if (base->hres_active) | |
f8953856 | 714 | return 1; |
54cdfdb4 TG |
715 | |
716 | local_irq_save(flags); | |
717 | ||
718 | if (tick_init_highres()) { | |
719 | local_irq_restore(flags); | |
820de5c3 IM |
720 | printk(KERN_WARNING "Could not switch to high resolution " |
721 | "mode on CPU %d\n", cpu); | |
f8953856 | 722 | return 0; |
54cdfdb4 TG |
723 | } |
724 | base->hres_active = 1; | |
b12a03ce TG |
725 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) |
726 | base->clock_base[i].resolution = KTIME_HIGH_RES; | |
54cdfdb4 TG |
727 | |
728 | tick_setup_sched_timer(); | |
54cdfdb4 TG |
729 | /* "Retrigger" the interrupt to get things going */ |
730 | retrigger_next_event(NULL); | |
731 | local_irq_restore(flags); | |
f8953856 | 732 | return 1; |
54cdfdb4 TG |
733 | } |
734 | ||
5ec2481b TG |
735 | static void clock_was_set_work(struct work_struct *work) |
736 | { | |
737 | clock_was_set(); | |
738 | } | |
739 | ||
740 | static DECLARE_WORK(hrtimer_work, clock_was_set_work); | |
741 | ||
f55a6faa | 742 | /* |
5ec2481b TG |
743 | * Called from timekeeping and resume code to reprogramm the hrtimer |
744 | * interrupt device on all cpus. | |
f55a6faa JS |
745 | */ |
746 | void clock_was_set_delayed(void) | |
747 | { | |
5ec2481b | 748 | schedule_work(&hrtimer_work); |
f55a6faa JS |
749 | } |
750 | ||
54cdfdb4 TG |
751 | #else |
752 | ||
753 | static inline int hrtimer_hres_active(void) { return 0; } | |
754 | static inline int hrtimer_is_hres_enabled(void) { return 0; } | |
f8953856 | 755 | static inline int hrtimer_switch_to_hres(void) { return 0; } |
7403f41f AC |
756 | static inline void |
757 | hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { } | |
54cdfdb4 | 758 | static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer, |
b22affe0 | 759 | struct hrtimer_clock_base *base) |
54cdfdb4 TG |
760 | { |
761 | return 0; | |
762 | } | |
54cdfdb4 | 763 | static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { } |
9ec26907 | 764 | static inline void retrigger_next_event(void *arg) { } |
54cdfdb4 TG |
765 | |
766 | #endif /* CONFIG_HIGH_RES_TIMERS */ | |
767 | ||
b12a03ce TG |
768 | /* |
769 | * Clock realtime was set | |
770 | * | |
771 | * Change the offset of the realtime clock vs. the monotonic | |
772 | * clock. | |
773 | * | |
774 | * We might have to reprogram the high resolution timer interrupt. On | |
775 | * SMP we call the architecture specific code to retrigger _all_ high | |
776 | * resolution timer interrupts. On UP we just disable interrupts and | |
777 | * call the high resolution interrupt code. | |
778 | */ | |
779 | void clock_was_set(void) | |
780 | { | |
90ff1f30 | 781 | #ifdef CONFIG_HIGH_RES_TIMERS |
b12a03ce TG |
782 | /* Retrigger the CPU local events everywhere */ |
783 | on_each_cpu(retrigger_next_event, NULL, 1); | |
9ec26907 TG |
784 | #endif |
785 | timerfd_clock_was_set(); | |
b12a03ce TG |
786 | } |
787 | ||
788 | /* | |
789 | * During resume we might have to reprogram the high resolution timer | |
7c4c3a0f DV |
790 | * interrupt on all online CPUs. However, all other CPUs will be |
791 | * stopped with IRQs interrupts disabled so the clock_was_set() call | |
5ec2481b | 792 | * must be deferred. |
b12a03ce TG |
793 | */ |
794 | void hrtimers_resume(void) | |
795 | { | |
796 | WARN_ONCE(!irqs_disabled(), | |
797 | KERN_INFO "hrtimers_resume() called with IRQs enabled!"); | |
798 | ||
5ec2481b | 799 | /* Retrigger on the local CPU */ |
b12a03ce | 800 | retrigger_next_event(NULL); |
5ec2481b TG |
801 | /* And schedule a retrigger for all others */ |
802 | clock_was_set_delayed(); | |
b12a03ce TG |
803 | } |
804 | ||
5f201907 | 805 | static inline void timer_stats_hrtimer_set_start_info(struct hrtimer *timer) |
82f67cd9 | 806 | { |
5f201907 | 807 | #ifdef CONFIG_TIMER_STATS |
82f67cd9 IM |
808 | if (timer->start_site) |
809 | return; | |
5f201907 | 810 | timer->start_site = __builtin_return_address(0); |
82f67cd9 IM |
811 | memcpy(timer->start_comm, current->comm, TASK_COMM_LEN); |
812 | timer->start_pid = current->pid; | |
5f201907 HC |
813 | #endif |
814 | } | |
815 | ||
816 | static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer *timer) | |
817 | { | |
818 | #ifdef CONFIG_TIMER_STATS | |
819 | timer->start_site = NULL; | |
820 | #endif | |
82f67cd9 | 821 | } |
5f201907 HC |
822 | |
823 | static inline void timer_stats_account_hrtimer(struct hrtimer *timer) | |
824 | { | |
825 | #ifdef CONFIG_TIMER_STATS | |
826 | if (likely(!timer_stats_active)) | |
827 | return; | |
828 | timer_stats_update_stats(timer, timer->start_pid, timer->start_site, | |
829 | timer->function, timer->start_comm, 0); | |
82f67cd9 | 830 | #endif |
5f201907 | 831 | } |
82f67cd9 | 832 | |
c0a31329 | 833 | /* |
6506f2aa | 834 | * Counterpart to lock_hrtimer_base above: |
c0a31329 TG |
835 | */ |
836 | static inline | |
837 | void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) | |
838 | { | |
ecb49d1a | 839 | raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags); |
c0a31329 TG |
840 | } |
841 | ||
842 | /** | |
843 | * hrtimer_forward - forward the timer expiry | |
c0a31329 | 844 | * @timer: hrtimer to forward |
44f21475 | 845 | * @now: forward past this time |
c0a31329 TG |
846 | * @interval: the interval to forward |
847 | * | |
848 | * Forward the timer expiry so it will expire in the future. | |
8dca6f33 | 849 | * Returns the number of overruns. |
c0a31329 | 850 | */ |
4d672e7a | 851 | u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval) |
c0a31329 | 852 | { |
4d672e7a | 853 | u64 orun = 1; |
44f21475 | 854 | ktime_t delta; |
c0a31329 | 855 | |
cc584b21 | 856 | delta = ktime_sub(now, hrtimer_get_expires(timer)); |
c0a31329 TG |
857 | |
858 | if (delta.tv64 < 0) | |
859 | return 0; | |
860 | ||
c9db4fa1 TG |
861 | if (interval.tv64 < timer->base->resolution.tv64) |
862 | interval.tv64 = timer->base->resolution.tv64; | |
863 | ||
c0a31329 | 864 | if (unlikely(delta.tv64 >= interval.tv64)) { |
df869b63 | 865 | s64 incr = ktime_to_ns(interval); |
c0a31329 TG |
866 | |
867 | orun = ktime_divns(delta, incr); | |
cc584b21 AV |
868 | hrtimer_add_expires_ns(timer, incr * orun); |
869 | if (hrtimer_get_expires_tv64(timer) > now.tv64) | |
c0a31329 TG |
870 | return orun; |
871 | /* | |
872 | * This (and the ktime_add() below) is the | |
873 | * correction for exact: | |
874 | */ | |
875 | orun++; | |
876 | } | |
cc584b21 | 877 | hrtimer_add_expires(timer, interval); |
c0a31329 TG |
878 | |
879 | return orun; | |
880 | } | |
6bdb6b62 | 881 | EXPORT_SYMBOL_GPL(hrtimer_forward); |
c0a31329 TG |
882 | |
883 | /* | |
884 | * enqueue_hrtimer - internal function to (re)start a timer | |
885 | * | |
886 | * The timer is inserted in expiry order. Insertion into the | |
887 | * red black tree is O(log(n)). Must hold the base lock. | |
a6037b61 PZ |
888 | * |
889 | * Returns 1 when the new timer is the leftmost timer in the tree. | |
c0a31329 | 890 | */ |
a6037b61 PZ |
891 | static int enqueue_hrtimer(struct hrtimer *timer, |
892 | struct hrtimer_clock_base *base) | |
c0a31329 | 893 | { |
c6a2a177 | 894 | debug_activate(timer); |
237fc6e7 | 895 | |
998adc3d | 896 | timerqueue_add(&base->active, &timer->node); |
ab8177bc | 897 | base->cpu_base->active_bases |= 1 << base->index; |
54cdfdb4 | 898 | |
303e967f TG |
899 | /* |
900 | * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the | |
901 | * state of a possibly running callback. | |
902 | */ | |
903 | timer->state |= HRTIMER_STATE_ENQUEUED; | |
a6037b61 | 904 | |
998adc3d | 905 | return (&timer->node == base->active.next); |
288867ec | 906 | } |
c0a31329 TG |
907 | |
908 | /* | |
909 | * __remove_hrtimer - internal function to remove a timer | |
910 | * | |
911 | * Caller must hold the base lock. | |
54cdfdb4 TG |
912 | * |
913 | * High resolution timer mode reprograms the clock event device when the | |
914 | * timer is the one which expires next. The caller can disable this by setting | |
915 | * reprogram to zero. This is useful, when the context does a reprogramming | |
916 | * anyway (e.g. timer interrupt) | |
c0a31329 | 917 | */ |
3c8aa39d | 918 | static void __remove_hrtimer(struct hrtimer *timer, |
303e967f | 919 | struct hrtimer_clock_base *base, |
54cdfdb4 | 920 | unsigned long newstate, int reprogram) |
c0a31329 | 921 | { |
27c9cd7e | 922 | struct timerqueue_node *next_timer; |
7403f41f AC |
923 | if (!(timer->state & HRTIMER_STATE_ENQUEUED)) |
924 | goto out; | |
925 | ||
27c9cd7e JO |
926 | next_timer = timerqueue_getnext(&base->active); |
927 | timerqueue_del(&base->active, &timer->node); | |
928 | if (&timer->node == next_timer) { | |
7403f41f AC |
929 | #ifdef CONFIG_HIGH_RES_TIMERS |
930 | /* Reprogram the clock event device. if enabled */ | |
931 | if (reprogram && hrtimer_hres_active()) { | |
932 | ktime_t expires; | |
933 | ||
934 | expires = ktime_sub(hrtimer_get_expires(timer), | |
935 | base->offset); | |
936 | if (base->cpu_base->expires_next.tv64 == expires.tv64) | |
937 | hrtimer_force_reprogram(base->cpu_base, 1); | |
54cdfdb4 | 938 | } |
7403f41f | 939 | #endif |
54cdfdb4 | 940 | } |
ab8177bc TG |
941 | if (!timerqueue_getnext(&base->active)) |
942 | base->cpu_base->active_bases &= ~(1 << base->index); | |
7403f41f | 943 | out: |
303e967f | 944 | timer->state = newstate; |
c0a31329 TG |
945 | } |
946 | ||
947 | /* | |
948 | * remove hrtimer, called with base lock held | |
949 | */ | |
950 | static inline int | |
3c8aa39d | 951 | remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base) |
c0a31329 | 952 | { |
303e967f | 953 | if (hrtimer_is_queued(timer)) { |
f13d4f97 | 954 | unsigned long state; |
54cdfdb4 TG |
955 | int reprogram; |
956 | ||
957 | /* | |
958 | * Remove the timer and force reprogramming when high | |
959 | * resolution mode is active and the timer is on the current | |
960 | * CPU. If we remove a timer on another CPU, reprogramming is | |
961 | * skipped. The interrupt event on this CPU is fired and | |
962 | * reprogramming happens in the interrupt handler. This is a | |
963 | * rare case and less expensive than a smp call. | |
964 | */ | |
c6a2a177 | 965 | debug_deactivate(timer); |
82f67cd9 | 966 | timer_stats_hrtimer_clear_start_info(timer); |
54cdfdb4 | 967 | reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases); |
f13d4f97 SQ |
968 | /* |
969 | * We must preserve the CALLBACK state flag here, | |
970 | * otherwise we could move the timer base in | |
971 | * switch_hrtimer_base. | |
972 | */ | |
973 | state = timer->state & HRTIMER_STATE_CALLBACK; | |
974 | __remove_hrtimer(timer, base, state, reprogram); | |
c0a31329 TG |
975 | return 1; |
976 | } | |
977 | return 0; | |
978 | } | |
979 | ||
7f1e2ca9 PZ |
980 | int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, |
981 | unsigned long delta_ns, const enum hrtimer_mode mode, | |
982 | int wakeup) | |
c0a31329 | 983 | { |
3c8aa39d | 984 | struct hrtimer_clock_base *base, *new_base; |
c0a31329 | 985 | unsigned long flags; |
a6037b61 | 986 | int ret, leftmost; |
c0a31329 TG |
987 | |
988 | base = lock_hrtimer_base(timer, &flags); | |
989 | ||
990 | /* Remove an active timer from the queue: */ | |
991 | ret = remove_hrtimer(timer, base); | |
992 | ||
993 | /* Switch the timer base, if necessary: */ | |
597d0275 | 994 | new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED); |
c0a31329 | 995 | |
597d0275 | 996 | if (mode & HRTIMER_MODE_REL) { |
5a7780e7 | 997 | tim = ktime_add_safe(tim, new_base->get_time()); |
06027bdd IM |
998 | /* |
999 | * CONFIG_TIME_LOW_RES is a temporary way for architectures | |
1000 | * to signal that they simply return xtime in | |
1001 | * do_gettimeoffset(). In this case we want to round up by | |
1002 | * resolution when starting a relative timer, to avoid short | |
1003 | * timeouts. This will go away with the GTOD framework. | |
1004 | */ | |
1005 | #ifdef CONFIG_TIME_LOW_RES | |
5a7780e7 | 1006 | tim = ktime_add_safe(tim, base->resolution); |
06027bdd IM |
1007 | #endif |
1008 | } | |
237fc6e7 | 1009 | |
da8f2e17 | 1010 | hrtimer_set_expires_range_ns(timer, tim, delta_ns); |
c0a31329 | 1011 | |
82f67cd9 IM |
1012 | timer_stats_hrtimer_set_start_info(timer); |
1013 | ||
a6037b61 PZ |
1014 | leftmost = enqueue_hrtimer(timer, new_base); |
1015 | ||
935c631d IM |
1016 | /* |
1017 | * Only allow reprogramming if the new base is on this CPU. | |
1018 | * (it might still be on another CPU if the timer was pending) | |
a6037b61 PZ |
1019 | * |
1020 | * XXX send_remote_softirq() ? | |
935c631d | 1021 | */ |
b22affe0 LS |
1022 | if (leftmost && new_base->cpu_base == &__get_cpu_var(hrtimer_bases) |
1023 | && hrtimer_enqueue_reprogram(timer, new_base)) { | |
1024 | if (wakeup) { | |
1025 | /* | |
1026 | * We need to drop cpu_base->lock to avoid a | |
1027 | * lock ordering issue vs. rq->lock. | |
1028 | */ | |
1029 | raw_spin_unlock(&new_base->cpu_base->lock); | |
1030 | raise_softirq_irqoff(HRTIMER_SOFTIRQ); | |
1031 | local_irq_restore(flags); | |
1032 | return ret; | |
1033 | } else { | |
1034 | __raise_softirq_irqoff(HRTIMER_SOFTIRQ); | |
1035 | } | |
1036 | } | |
c0a31329 TG |
1037 | |
1038 | unlock_hrtimer_base(timer, &flags); | |
1039 | ||
1040 | return ret; | |
1041 | } | |
7f1e2ca9 PZ |
1042 | |
1043 | /** | |
1044 | * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU | |
1045 | * @timer: the timer to be added | |
1046 | * @tim: expiry time | |
1047 | * @delta_ns: "slack" range for the timer | |
8ffbc7d9 DD |
1048 | * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or |
1049 | * relative (HRTIMER_MODE_REL) | |
7f1e2ca9 PZ |
1050 | * |
1051 | * Returns: | |
1052 | * 0 on success | |
1053 | * 1 when the timer was active | |
1054 | */ | |
1055 | int hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, | |
1056 | unsigned long delta_ns, const enum hrtimer_mode mode) | |
1057 | { | |
1058 | return __hrtimer_start_range_ns(timer, tim, delta_ns, mode, 1); | |
1059 | } | |
da8f2e17 AV |
1060 | EXPORT_SYMBOL_GPL(hrtimer_start_range_ns); |
1061 | ||
1062 | /** | |
e1dd7bc5 | 1063 | * hrtimer_start - (re)start an hrtimer on the current CPU |
da8f2e17 AV |
1064 | * @timer: the timer to be added |
1065 | * @tim: expiry time | |
8ffbc7d9 DD |
1066 | * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or |
1067 | * relative (HRTIMER_MODE_REL) | |
da8f2e17 AV |
1068 | * |
1069 | * Returns: | |
1070 | * 0 on success | |
1071 | * 1 when the timer was active | |
1072 | */ | |
1073 | int | |
1074 | hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode) | |
1075 | { | |
7f1e2ca9 | 1076 | return __hrtimer_start_range_ns(timer, tim, 0, mode, 1); |
da8f2e17 | 1077 | } |
8d16b764 | 1078 | EXPORT_SYMBOL_GPL(hrtimer_start); |
c0a31329 | 1079 | |
da8f2e17 | 1080 | |
c0a31329 TG |
1081 | /** |
1082 | * hrtimer_try_to_cancel - try to deactivate a timer | |
c0a31329 TG |
1083 | * @timer: hrtimer to stop |
1084 | * | |
1085 | * Returns: | |
1086 | * 0 when the timer was not active | |
1087 | * 1 when the timer was active | |
1088 | * -1 when the timer is currently excuting the callback function and | |
fa9799e3 | 1089 | * cannot be stopped |
c0a31329 TG |
1090 | */ |
1091 | int hrtimer_try_to_cancel(struct hrtimer *timer) | |
1092 | { | |
3c8aa39d | 1093 | struct hrtimer_clock_base *base; |
c0a31329 TG |
1094 | unsigned long flags; |
1095 | int ret = -1; | |
1096 | ||
1097 | base = lock_hrtimer_base(timer, &flags); | |
1098 | ||
303e967f | 1099 | if (!hrtimer_callback_running(timer)) |
c0a31329 TG |
1100 | ret = remove_hrtimer(timer, base); |
1101 | ||
1102 | unlock_hrtimer_base(timer, &flags); | |
1103 | ||
1104 | return ret; | |
1105 | ||
1106 | } | |
8d16b764 | 1107 | EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel); |
c0a31329 TG |
1108 | |
1109 | /** | |
1110 | * hrtimer_cancel - cancel a timer and wait for the handler to finish. | |
c0a31329 TG |
1111 | * @timer: the timer to be cancelled |
1112 | * | |
1113 | * Returns: | |
1114 | * 0 when the timer was not active | |
1115 | * 1 when the timer was active | |
1116 | */ | |
1117 | int hrtimer_cancel(struct hrtimer *timer) | |
1118 | { | |
1119 | for (;;) { | |
1120 | int ret = hrtimer_try_to_cancel(timer); | |
1121 | ||
1122 | if (ret >= 0) | |
1123 | return ret; | |
5ef37b19 | 1124 | cpu_relax(); |
c0a31329 TG |
1125 | } |
1126 | } | |
8d16b764 | 1127 | EXPORT_SYMBOL_GPL(hrtimer_cancel); |
c0a31329 TG |
1128 | |
1129 | /** | |
1130 | * hrtimer_get_remaining - get remaining time for the timer | |
c0a31329 TG |
1131 | * @timer: the timer to read |
1132 | */ | |
1133 | ktime_t hrtimer_get_remaining(const struct hrtimer *timer) | |
1134 | { | |
c0a31329 TG |
1135 | unsigned long flags; |
1136 | ktime_t rem; | |
1137 | ||
b3bd3de6 | 1138 | lock_hrtimer_base(timer, &flags); |
cc584b21 | 1139 | rem = hrtimer_expires_remaining(timer); |
c0a31329 TG |
1140 | unlock_hrtimer_base(timer, &flags); |
1141 | ||
1142 | return rem; | |
1143 | } | |
8d16b764 | 1144 | EXPORT_SYMBOL_GPL(hrtimer_get_remaining); |
c0a31329 | 1145 | |
3451d024 | 1146 | #ifdef CONFIG_NO_HZ_COMMON |
69239749 TL |
1147 | /** |
1148 | * hrtimer_get_next_event - get the time until next expiry event | |
1149 | * | |
1150 | * Returns the delta to the next expiry event or KTIME_MAX if no timer | |
1151 | * is pending. | |
1152 | */ | |
1153 | ktime_t hrtimer_get_next_event(void) | |
1154 | { | |
3c8aa39d TG |
1155 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
1156 | struct hrtimer_clock_base *base = cpu_base->clock_base; | |
69239749 TL |
1157 | ktime_t delta, mindelta = { .tv64 = KTIME_MAX }; |
1158 | unsigned long flags; | |
1159 | int i; | |
1160 | ||
ecb49d1a | 1161 | raw_spin_lock_irqsave(&cpu_base->lock, flags); |
3c8aa39d | 1162 | |
54cdfdb4 TG |
1163 | if (!hrtimer_hres_active()) { |
1164 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) { | |
1165 | struct hrtimer *timer; | |
998adc3d | 1166 | struct timerqueue_node *next; |
69239749 | 1167 | |
998adc3d JS |
1168 | next = timerqueue_getnext(&base->active); |
1169 | if (!next) | |
54cdfdb4 | 1170 | continue; |
3c8aa39d | 1171 | |
998adc3d | 1172 | timer = container_of(next, struct hrtimer, node); |
cc584b21 | 1173 | delta.tv64 = hrtimer_get_expires_tv64(timer); |
54cdfdb4 TG |
1174 | delta = ktime_sub(delta, base->get_time()); |
1175 | if (delta.tv64 < mindelta.tv64) | |
1176 | mindelta.tv64 = delta.tv64; | |
1177 | } | |
69239749 | 1178 | } |
3c8aa39d | 1179 | |
ecb49d1a | 1180 | raw_spin_unlock_irqrestore(&cpu_base->lock, flags); |
3c8aa39d | 1181 | |
69239749 TL |
1182 | if (mindelta.tv64 < 0) |
1183 | mindelta.tv64 = 0; | |
1184 | return mindelta; | |
1185 | } | |
1186 | #endif | |
1187 | ||
237fc6e7 TG |
1188 | static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id, |
1189 | enum hrtimer_mode mode) | |
c0a31329 | 1190 | { |
3c8aa39d | 1191 | struct hrtimer_cpu_base *cpu_base; |
e06383db | 1192 | int base; |
c0a31329 | 1193 | |
7978672c GA |
1194 | memset(timer, 0, sizeof(struct hrtimer)); |
1195 | ||
3c8aa39d | 1196 | cpu_base = &__raw_get_cpu_var(hrtimer_bases); |
c0a31329 | 1197 | |
c9cb2e3d | 1198 | if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS) |
7978672c GA |
1199 | clock_id = CLOCK_MONOTONIC; |
1200 | ||
e06383db JS |
1201 | base = hrtimer_clockid_to_base(clock_id); |
1202 | timer->base = &cpu_base->clock_base[base]; | |
998adc3d | 1203 | timerqueue_init(&timer->node); |
82f67cd9 IM |
1204 | |
1205 | #ifdef CONFIG_TIMER_STATS | |
1206 | timer->start_site = NULL; | |
1207 | timer->start_pid = -1; | |
1208 | memset(timer->start_comm, 0, TASK_COMM_LEN); | |
1209 | #endif | |
c0a31329 | 1210 | } |
237fc6e7 TG |
1211 | |
1212 | /** | |
1213 | * hrtimer_init - initialize a timer to the given clock | |
1214 | * @timer: the timer to be initialized | |
1215 | * @clock_id: the clock to be used | |
1216 | * @mode: timer mode abs/rel | |
1217 | */ | |
1218 | void hrtimer_init(struct hrtimer *timer, clockid_t clock_id, | |
1219 | enum hrtimer_mode mode) | |
1220 | { | |
c6a2a177 | 1221 | debug_init(timer, clock_id, mode); |
237fc6e7 TG |
1222 | __hrtimer_init(timer, clock_id, mode); |
1223 | } | |
8d16b764 | 1224 | EXPORT_SYMBOL_GPL(hrtimer_init); |
c0a31329 TG |
1225 | |
1226 | /** | |
1227 | * hrtimer_get_res - get the timer resolution for a clock | |
c0a31329 TG |
1228 | * @which_clock: which clock to query |
1229 | * @tp: pointer to timespec variable to store the resolution | |
1230 | * | |
72fd4a35 RD |
1231 | * Store the resolution of the clock selected by @which_clock in the |
1232 | * variable pointed to by @tp. | |
c0a31329 TG |
1233 | */ |
1234 | int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp) | |
1235 | { | |
3c8aa39d | 1236 | struct hrtimer_cpu_base *cpu_base; |
e06383db | 1237 | int base = hrtimer_clockid_to_base(which_clock); |
c0a31329 | 1238 | |
3c8aa39d | 1239 | cpu_base = &__raw_get_cpu_var(hrtimer_bases); |
e06383db | 1240 | *tp = ktime_to_timespec(cpu_base->clock_base[base].resolution); |
c0a31329 TG |
1241 | |
1242 | return 0; | |
1243 | } | |
8d16b764 | 1244 | EXPORT_SYMBOL_GPL(hrtimer_get_res); |
c0a31329 | 1245 | |
c6a2a177 | 1246 | static void __run_hrtimer(struct hrtimer *timer, ktime_t *now) |
d3d74453 PZ |
1247 | { |
1248 | struct hrtimer_clock_base *base = timer->base; | |
1249 | struct hrtimer_cpu_base *cpu_base = base->cpu_base; | |
1250 | enum hrtimer_restart (*fn)(struct hrtimer *); | |
1251 | int restart; | |
1252 | ||
ca109491 PZ |
1253 | WARN_ON(!irqs_disabled()); |
1254 | ||
c6a2a177 | 1255 | debug_deactivate(timer); |
d3d74453 PZ |
1256 | __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0); |
1257 | timer_stats_account_hrtimer(timer); | |
d3d74453 | 1258 | fn = timer->function; |
ca109491 PZ |
1259 | |
1260 | /* | |
1261 | * Because we run timers from hardirq context, there is no chance | |
1262 | * they get migrated to another cpu, therefore its safe to unlock | |
1263 | * the timer base. | |
1264 | */ | |
ecb49d1a | 1265 | raw_spin_unlock(&cpu_base->lock); |
c6a2a177 | 1266 | trace_hrtimer_expire_entry(timer, now); |
ca109491 | 1267 | restart = fn(timer); |
c6a2a177 | 1268 | trace_hrtimer_expire_exit(timer); |
ecb49d1a | 1269 | raw_spin_lock(&cpu_base->lock); |
d3d74453 PZ |
1270 | |
1271 | /* | |
e3f1d883 TG |
1272 | * Note: We clear the CALLBACK bit after enqueue_hrtimer and |
1273 | * we do not reprogramm the event hardware. Happens either in | |
1274 | * hrtimer_start_range_ns() or in hrtimer_interrupt() | |
d3d74453 PZ |
1275 | */ |
1276 | if (restart != HRTIMER_NORESTART) { | |
1277 | BUG_ON(timer->state != HRTIMER_STATE_CALLBACK); | |
a6037b61 | 1278 | enqueue_hrtimer(timer, base); |
d3d74453 | 1279 | } |
f13d4f97 SQ |
1280 | |
1281 | WARN_ON_ONCE(!(timer->state & HRTIMER_STATE_CALLBACK)); | |
1282 | ||
d3d74453 PZ |
1283 | timer->state &= ~HRTIMER_STATE_CALLBACK; |
1284 | } | |
1285 | ||
54cdfdb4 TG |
1286 | #ifdef CONFIG_HIGH_RES_TIMERS |
1287 | ||
1288 | /* | |
1289 | * High resolution timer interrupt | |
1290 | * Called with interrupts disabled | |
1291 | */ | |
1292 | void hrtimer_interrupt(struct clock_event_device *dev) | |
1293 | { | |
1294 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); | |
41d2e494 TG |
1295 | ktime_t expires_next, now, entry_time, delta; |
1296 | int i, retries = 0; | |
54cdfdb4 TG |
1297 | |
1298 | BUG_ON(!cpu_base->hres_active); | |
1299 | cpu_base->nr_events++; | |
1300 | dev->next_event.tv64 = KTIME_MAX; | |
1301 | ||
196951e9 | 1302 | raw_spin_lock(&cpu_base->lock); |
5baefd6d | 1303 | entry_time = now = hrtimer_update_base(cpu_base); |
41d2e494 | 1304 | retry: |
54cdfdb4 | 1305 | expires_next.tv64 = KTIME_MAX; |
6ff7041d TG |
1306 | /* |
1307 | * We set expires_next to KTIME_MAX here with cpu_base->lock | |
1308 | * held to prevent that a timer is enqueued in our queue via | |
1309 | * the migration code. This does not affect enqueueing of | |
1310 | * timers which run their callback and need to be requeued on | |
1311 | * this CPU. | |
1312 | */ | |
1313 | cpu_base->expires_next.tv64 = KTIME_MAX; | |
1314 | ||
54cdfdb4 | 1315 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
ab8177bc | 1316 | struct hrtimer_clock_base *base; |
998adc3d | 1317 | struct timerqueue_node *node; |
ab8177bc TG |
1318 | ktime_t basenow; |
1319 | ||
1320 | if (!(cpu_base->active_bases & (1 << i))) | |
1321 | continue; | |
54cdfdb4 | 1322 | |
ab8177bc | 1323 | base = cpu_base->clock_base + i; |
54cdfdb4 TG |
1324 | basenow = ktime_add(now, base->offset); |
1325 | ||
998adc3d | 1326 | while ((node = timerqueue_getnext(&base->active))) { |
54cdfdb4 TG |
1327 | struct hrtimer *timer; |
1328 | ||
998adc3d | 1329 | timer = container_of(node, struct hrtimer, node); |
54cdfdb4 | 1330 | |
654c8e0b AV |
1331 | /* |
1332 | * The immediate goal for using the softexpires is | |
1333 | * minimizing wakeups, not running timers at the | |
1334 | * earliest interrupt after their soft expiration. | |
1335 | * This allows us to avoid using a Priority Search | |
1336 | * Tree, which can answer a stabbing querry for | |
1337 | * overlapping intervals and instead use the simple | |
1338 | * BST we already have. | |
1339 | * We don't add extra wakeups by delaying timers that | |
1340 | * are right-of a not yet expired timer, because that | |
1341 | * timer will have to trigger a wakeup anyway. | |
1342 | */ | |
1343 | ||
1344 | if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer)) { | |
54cdfdb4 TG |
1345 | ktime_t expires; |
1346 | ||
cc584b21 | 1347 | expires = ktime_sub(hrtimer_get_expires(timer), |
54cdfdb4 | 1348 | base->offset); |
8f294b5a PB |
1349 | if (expires.tv64 < 0) |
1350 | expires.tv64 = KTIME_MAX; | |
54cdfdb4 TG |
1351 | if (expires.tv64 < expires_next.tv64) |
1352 | expires_next = expires; | |
1353 | break; | |
1354 | } | |
1355 | ||
c6a2a177 | 1356 | __run_hrtimer(timer, &basenow); |
54cdfdb4 | 1357 | } |
54cdfdb4 TG |
1358 | } |
1359 | ||
6ff7041d TG |
1360 | /* |
1361 | * Store the new expiry value so the migration code can verify | |
1362 | * against it. | |
1363 | */ | |
54cdfdb4 | 1364 | cpu_base->expires_next = expires_next; |
ecb49d1a | 1365 | raw_spin_unlock(&cpu_base->lock); |
54cdfdb4 TG |
1366 | |
1367 | /* Reprogramming necessary ? */ | |
41d2e494 TG |
1368 | if (expires_next.tv64 == KTIME_MAX || |
1369 | !tick_program_event(expires_next, 0)) { | |
1370 | cpu_base->hang_detected = 0; | |
1371 | return; | |
54cdfdb4 | 1372 | } |
41d2e494 TG |
1373 | |
1374 | /* | |
1375 | * The next timer was already expired due to: | |
1376 | * - tracing | |
1377 | * - long lasting callbacks | |
1378 | * - being scheduled away when running in a VM | |
1379 | * | |
1380 | * We need to prevent that we loop forever in the hrtimer | |
1381 | * interrupt routine. We give it 3 attempts to avoid | |
1382 | * overreacting on some spurious event. | |
5baefd6d JS |
1383 | * |
1384 | * Acquire base lock for updating the offsets and retrieving | |
1385 | * the current time. | |
41d2e494 | 1386 | */ |
196951e9 | 1387 | raw_spin_lock(&cpu_base->lock); |
5baefd6d | 1388 | now = hrtimer_update_base(cpu_base); |
41d2e494 TG |
1389 | cpu_base->nr_retries++; |
1390 | if (++retries < 3) | |
1391 | goto retry; | |
1392 | /* | |
1393 | * Give the system a chance to do something else than looping | |
1394 | * here. We stored the entry time, so we know exactly how long | |
1395 | * we spent here. We schedule the next event this amount of | |
1396 | * time away. | |
1397 | */ | |
1398 | cpu_base->nr_hangs++; | |
1399 | cpu_base->hang_detected = 1; | |
196951e9 | 1400 | raw_spin_unlock(&cpu_base->lock); |
41d2e494 TG |
1401 | delta = ktime_sub(now, entry_time); |
1402 | if (delta.tv64 > cpu_base->max_hang_time.tv64) | |
1403 | cpu_base->max_hang_time = delta; | |
1404 | /* | |
1405 | * Limit it to a sensible value as we enforce a longer | |
1406 | * delay. Give the CPU at least 100ms to catch up. | |
1407 | */ | |
1408 | if (delta.tv64 > 100 * NSEC_PER_MSEC) | |
1409 | expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC); | |
1410 | else | |
1411 | expires_next = ktime_add(now, delta); | |
1412 | tick_program_event(expires_next, 1); | |
1413 | printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n", | |
1414 | ktime_to_ns(delta)); | |
54cdfdb4 TG |
1415 | } |
1416 | ||
8bdec955 TG |
1417 | /* |
1418 | * local version of hrtimer_peek_ahead_timers() called with interrupts | |
1419 | * disabled. | |
1420 | */ | |
1421 | static void __hrtimer_peek_ahead_timers(void) | |
1422 | { | |
1423 | struct tick_device *td; | |
1424 | ||
1425 | if (!hrtimer_hres_active()) | |
1426 | return; | |
1427 | ||
1428 | td = &__get_cpu_var(tick_cpu_device); | |
1429 | if (td && td->evtdev) | |
1430 | hrtimer_interrupt(td->evtdev); | |
1431 | } | |
1432 | ||
2e94d1f7 AV |
1433 | /** |
1434 | * hrtimer_peek_ahead_timers -- run soft-expired timers now | |
1435 | * | |
1436 | * hrtimer_peek_ahead_timers will peek at the timer queue of | |
1437 | * the current cpu and check if there are any timers for which | |
1438 | * the soft expires time has passed. If any such timers exist, | |
1439 | * they are run immediately and then removed from the timer queue. | |
1440 | * | |
1441 | */ | |
1442 | void hrtimer_peek_ahead_timers(void) | |
1443 | { | |
643bdf68 | 1444 | unsigned long flags; |
dc4304f7 | 1445 | |
2e94d1f7 | 1446 | local_irq_save(flags); |
8bdec955 | 1447 | __hrtimer_peek_ahead_timers(); |
2e94d1f7 AV |
1448 | local_irq_restore(flags); |
1449 | } | |
1450 | ||
a6037b61 PZ |
1451 | static void run_hrtimer_softirq(struct softirq_action *h) |
1452 | { | |
1453 | hrtimer_peek_ahead_timers(); | |
1454 | } | |
1455 | ||
82c5b7b5 IM |
1456 | #else /* CONFIG_HIGH_RES_TIMERS */ |
1457 | ||
1458 | static inline void __hrtimer_peek_ahead_timers(void) { } | |
1459 | ||
1460 | #endif /* !CONFIG_HIGH_RES_TIMERS */ | |
82f67cd9 | 1461 | |
d3d74453 PZ |
1462 | /* |
1463 | * Called from timer softirq every jiffy, expire hrtimers: | |
1464 | * | |
1465 | * For HRT its the fall back code to run the softirq in the timer | |
1466 | * softirq context in case the hrtimer initialization failed or has | |
1467 | * not been done yet. | |
1468 | */ | |
1469 | void hrtimer_run_pending(void) | |
1470 | { | |
d3d74453 PZ |
1471 | if (hrtimer_hres_active()) |
1472 | return; | |
54cdfdb4 | 1473 | |
d3d74453 PZ |
1474 | /* |
1475 | * This _is_ ugly: We have to check in the softirq context, | |
1476 | * whether we can switch to highres and / or nohz mode. The | |
1477 | * clocksource switch happens in the timer interrupt with | |
1478 | * xtime_lock held. Notification from there only sets the | |
1479 | * check bit in the tick_oneshot code, otherwise we might | |
1480 | * deadlock vs. xtime_lock. | |
1481 | */ | |
1482 | if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) | |
1483 | hrtimer_switch_to_hres(); | |
54cdfdb4 TG |
1484 | } |
1485 | ||
c0a31329 | 1486 | /* |
d3d74453 | 1487 | * Called from hardirq context every jiffy |
c0a31329 | 1488 | */ |
833883d9 | 1489 | void hrtimer_run_queues(void) |
c0a31329 | 1490 | { |
998adc3d | 1491 | struct timerqueue_node *node; |
833883d9 DS |
1492 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
1493 | struct hrtimer_clock_base *base; | |
1494 | int index, gettime = 1; | |
c0a31329 | 1495 | |
833883d9 | 1496 | if (hrtimer_hres_active()) |
3055adda DS |
1497 | return; |
1498 | ||
833883d9 DS |
1499 | for (index = 0; index < HRTIMER_MAX_CLOCK_BASES; index++) { |
1500 | base = &cpu_base->clock_base[index]; | |
b007c389 | 1501 | if (!timerqueue_getnext(&base->active)) |
d3d74453 | 1502 | continue; |
833883d9 | 1503 | |
d7cfb60c | 1504 | if (gettime) { |
833883d9 DS |
1505 | hrtimer_get_softirq_time(cpu_base); |
1506 | gettime = 0; | |
b75f7a51 | 1507 | } |
d3d74453 | 1508 | |
ecb49d1a | 1509 | raw_spin_lock(&cpu_base->lock); |
c0a31329 | 1510 | |
b007c389 | 1511 | while ((node = timerqueue_getnext(&base->active))) { |
833883d9 | 1512 | struct hrtimer *timer; |
54cdfdb4 | 1513 | |
998adc3d | 1514 | timer = container_of(node, struct hrtimer, node); |
cc584b21 AV |
1515 | if (base->softirq_time.tv64 <= |
1516 | hrtimer_get_expires_tv64(timer)) | |
833883d9 DS |
1517 | break; |
1518 | ||
c6a2a177 | 1519 | __run_hrtimer(timer, &base->softirq_time); |
833883d9 | 1520 | } |
ecb49d1a | 1521 | raw_spin_unlock(&cpu_base->lock); |
833883d9 | 1522 | } |
c0a31329 TG |
1523 | } |
1524 | ||
10c94ec1 TG |
1525 | /* |
1526 | * Sleep related functions: | |
1527 | */ | |
c9cb2e3d | 1528 | static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer) |
00362e33 TG |
1529 | { |
1530 | struct hrtimer_sleeper *t = | |
1531 | container_of(timer, struct hrtimer_sleeper, timer); | |
1532 | struct task_struct *task = t->task; | |
1533 | ||
1534 | t->task = NULL; | |
1535 | if (task) | |
1536 | wake_up_process(task); | |
1537 | ||
1538 | return HRTIMER_NORESTART; | |
1539 | } | |
1540 | ||
36c8b586 | 1541 | void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task) |
00362e33 TG |
1542 | { |
1543 | sl->timer.function = hrtimer_wakeup; | |
1544 | sl->task = task; | |
1545 | } | |
2bc481cf | 1546 | EXPORT_SYMBOL_GPL(hrtimer_init_sleeper); |
00362e33 | 1547 | |
669d7868 | 1548 | static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode) |
432569bb | 1549 | { |
669d7868 | 1550 | hrtimer_init_sleeper(t, current); |
10c94ec1 | 1551 | |
432569bb RZ |
1552 | do { |
1553 | set_current_state(TASK_INTERRUPTIBLE); | |
cc584b21 | 1554 | hrtimer_start_expires(&t->timer, mode); |
37bb6cb4 PZ |
1555 | if (!hrtimer_active(&t->timer)) |
1556 | t->task = NULL; | |
432569bb | 1557 | |
54cdfdb4 | 1558 | if (likely(t->task)) |
b0f8c44f | 1559 | freezable_schedule(); |
432569bb | 1560 | |
669d7868 | 1561 | hrtimer_cancel(&t->timer); |
c9cb2e3d | 1562 | mode = HRTIMER_MODE_ABS; |
669d7868 TG |
1563 | |
1564 | } while (t->task && !signal_pending(current)); | |
432569bb | 1565 | |
3588a085 PZ |
1566 | __set_current_state(TASK_RUNNING); |
1567 | ||
669d7868 | 1568 | return t->task == NULL; |
10c94ec1 TG |
1569 | } |
1570 | ||
080344b9 ON |
1571 | static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp) |
1572 | { | |
1573 | struct timespec rmt; | |
1574 | ktime_t rem; | |
1575 | ||
cc584b21 | 1576 | rem = hrtimer_expires_remaining(timer); |
080344b9 ON |
1577 | if (rem.tv64 <= 0) |
1578 | return 0; | |
1579 | rmt = ktime_to_timespec(rem); | |
1580 | ||
1581 | if (copy_to_user(rmtp, &rmt, sizeof(*rmtp))) | |
1582 | return -EFAULT; | |
1583 | ||
1584 | return 1; | |
1585 | } | |
1586 | ||
1711ef38 | 1587 | long __sched hrtimer_nanosleep_restart(struct restart_block *restart) |
10c94ec1 | 1588 | { |
669d7868 | 1589 | struct hrtimer_sleeper t; |
080344b9 | 1590 | struct timespec __user *rmtp; |
237fc6e7 | 1591 | int ret = 0; |
10c94ec1 | 1592 | |
ab8177bc | 1593 | hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid, |
237fc6e7 | 1594 | HRTIMER_MODE_ABS); |
cc584b21 | 1595 | hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires); |
10c94ec1 | 1596 | |
c9cb2e3d | 1597 | if (do_nanosleep(&t, HRTIMER_MODE_ABS)) |
237fc6e7 | 1598 | goto out; |
10c94ec1 | 1599 | |
029a07e0 | 1600 | rmtp = restart->nanosleep.rmtp; |
432569bb | 1601 | if (rmtp) { |
237fc6e7 | 1602 | ret = update_rmtp(&t.timer, rmtp); |
080344b9 | 1603 | if (ret <= 0) |
237fc6e7 | 1604 | goto out; |
432569bb | 1605 | } |
10c94ec1 | 1606 | |
10c94ec1 | 1607 | /* The other values in restart are already filled in */ |
237fc6e7 TG |
1608 | ret = -ERESTART_RESTARTBLOCK; |
1609 | out: | |
1610 | destroy_hrtimer_on_stack(&t.timer); | |
1611 | return ret; | |
10c94ec1 TG |
1612 | } |
1613 | ||
080344b9 | 1614 | long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp, |
10c94ec1 TG |
1615 | const enum hrtimer_mode mode, const clockid_t clockid) |
1616 | { | |
1617 | struct restart_block *restart; | |
669d7868 | 1618 | struct hrtimer_sleeper t; |
237fc6e7 | 1619 | int ret = 0; |
3bd01206 AV |
1620 | unsigned long slack; |
1621 | ||
1622 | slack = current->timer_slack_ns; | |
aab03e05 | 1623 | if (dl_task(current) || rt_task(current)) |
3bd01206 | 1624 | slack = 0; |
10c94ec1 | 1625 | |
237fc6e7 | 1626 | hrtimer_init_on_stack(&t.timer, clockid, mode); |
3bd01206 | 1627 | hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack); |
432569bb | 1628 | if (do_nanosleep(&t, mode)) |
237fc6e7 | 1629 | goto out; |
10c94ec1 | 1630 | |
7978672c | 1631 | /* Absolute timers do not update the rmtp value and restart: */ |
237fc6e7 TG |
1632 | if (mode == HRTIMER_MODE_ABS) { |
1633 | ret = -ERESTARTNOHAND; | |
1634 | goto out; | |
1635 | } | |
10c94ec1 | 1636 | |
432569bb | 1637 | if (rmtp) { |
237fc6e7 | 1638 | ret = update_rmtp(&t.timer, rmtp); |
080344b9 | 1639 | if (ret <= 0) |
237fc6e7 | 1640 | goto out; |
432569bb | 1641 | } |
10c94ec1 TG |
1642 | |
1643 | restart = ¤t_thread_info()->restart_block; | |
1711ef38 | 1644 | restart->fn = hrtimer_nanosleep_restart; |
ab8177bc | 1645 | restart->nanosleep.clockid = t.timer.base->clockid; |
029a07e0 | 1646 | restart->nanosleep.rmtp = rmtp; |
cc584b21 | 1647 | restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer); |
10c94ec1 | 1648 | |
237fc6e7 TG |
1649 | ret = -ERESTART_RESTARTBLOCK; |
1650 | out: | |
1651 | destroy_hrtimer_on_stack(&t.timer); | |
1652 | return ret; | |
10c94ec1 TG |
1653 | } |
1654 | ||
58fd3aa2 HC |
1655 | SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp, |
1656 | struct timespec __user *, rmtp) | |
6ba1b912 | 1657 | { |
080344b9 | 1658 | struct timespec tu; |
6ba1b912 TG |
1659 | |
1660 | if (copy_from_user(&tu, rqtp, sizeof(tu))) | |
1661 | return -EFAULT; | |
1662 | ||
1663 | if (!timespec_valid(&tu)) | |
1664 | return -EINVAL; | |
1665 | ||
080344b9 | 1666 | return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC); |
6ba1b912 TG |
1667 | } |
1668 | ||
c0a31329 TG |
1669 | /* |
1670 | * Functions related to boot-time initialization: | |
1671 | */ | |
0db0628d | 1672 | static void init_hrtimers_cpu(int cpu) |
c0a31329 | 1673 | { |
3c8aa39d | 1674 | struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu); |
c0a31329 TG |
1675 | int i; |
1676 | ||
998adc3d | 1677 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
3c8aa39d | 1678 | cpu_base->clock_base[i].cpu_base = cpu_base; |
998adc3d JS |
1679 | timerqueue_init_head(&cpu_base->clock_base[i].active); |
1680 | } | |
3c8aa39d | 1681 | |
54cdfdb4 | 1682 | hrtimer_init_hres(cpu_base); |
c0a31329 TG |
1683 | } |
1684 | ||
1685 | #ifdef CONFIG_HOTPLUG_CPU | |
1686 | ||
ca109491 | 1687 | static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base, |
37810659 | 1688 | struct hrtimer_clock_base *new_base) |
c0a31329 TG |
1689 | { |
1690 | struct hrtimer *timer; | |
998adc3d | 1691 | struct timerqueue_node *node; |
c0a31329 | 1692 | |
998adc3d JS |
1693 | while ((node = timerqueue_getnext(&old_base->active))) { |
1694 | timer = container_of(node, struct hrtimer, node); | |
54cdfdb4 | 1695 | BUG_ON(hrtimer_callback_running(timer)); |
c6a2a177 | 1696 | debug_deactivate(timer); |
b00c1a99 TG |
1697 | |
1698 | /* | |
1699 | * Mark it as STATE_MIGRATE not INACTIVE otherwise the | |
1700 | * timer could be seen as !active and just vanish away | |
1701 | * under us on another CPU | |
1702 | */ | |
1703 | __remove_hrtimer(timer, old_base, HRTIMER_STATE_MIGRATE, 0); | |
c0a31329 | 1704 | timer->base = new_base; |
54cdfdb4 | 1705 | /* |
e3f1d883 TG |
1706 | * Enqueue the timers on the new cpu. This does not |
1707 | * reprogram the event device in case the timer | |
1708 | * expires before the earliest on this CPU, but we run | |
1709 | * hrtimer_interrupt after we migrated everything to | |
1710 | * sort out already expired timers and reprogram the | |
1711 | * event device. | |
54cdfdb4 | 1712 | */ |
a6037b61 | 1713 | enqueue_hrtimer(timer, new_base); |
41e1022e | 1714 | |
b00c1a99 TG |
1715 | /* Clear the migration state bit */ |
1716 | timer->state &= ~HRTIMER_STATE_MIGRATE; | |
c0a31329 TG |
1717 | } |
1718 | } | |
1719 | ||
d5fd43c4 | 1720 | static void migrate_hrtimers(int scpu) |
c0a31329 | 1721 | { |
3c8aa39d | 1722 | struct hrtimer_cpu_base *old_base, *new_base; |
731a55ba | 1723 | int i; |
c0a31329 | 1724 | |
37810659 | 1725 | BUG_ON(cpu_online(scpu)); |
37810659 | 1726 | tick_cancel_sched_timer(scpu); |
731a55ba TG |
1727 | |
1728 | local_irq_disable(); | |
1729 | old_base = &per_cpu(hrtimer_bases, scpu); | |
1730 | new_base = &__get_cpu_var(hrtimer_bases); | |
d82f0b0f ON |
1731 | /* |
1732 | * The caller is globally serialized and nobody else | |
1733 | * takes two locks at once, deadlock is not possible. | |
1734 | */ | |
ecb49d1a TG |
1735 | raw_spin_lock(&new_base->lock); |
1736 | raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING); | |
c0a31329 | 1737 | |
3c8aa39d | 1738 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
ca109491 | 1739 | migrate_hrtimer_list(&old_base->clock_base[i], |
37810659 | 1740 | &new_base->clock_base[i]); |
c0a31329 TG |
1741 | } |
1742 | ||
ecb49d1a TG |
1743 | raw_spin_unlock(&old_base->lock); |
1744 | raw_spin_unlock(&new_base->lock); | |
37810659 | 1745 | |
731a55ba TG |
1746 | /* Check, if we got expired work to do */ |
1747 | __hrtimer_peek_ahead_timers(); | |
1748 | local_irq_enable(); | |
c0a31329 | 1749 | } |
37810659 | 1750 | |
c0a31329 TG |
1751 | #endif /* CONFIG_HOTPLUG_CPU */ |
1752 | ||
0db0628d | 1753 | static int hrtimer_cpu_notify(struct notifier_block *self, |
c0a31329 TG |
1754 | unsigned long action, void *hcpu) |
1755 | { | |
b2e3c0ad | 1756 | int scpu = (long)hcpu; |
c0a31329 TG |
1757 | |
1758 | switch (action) { | |
1759 | ||
1760 | case CPU_UP_PREPARE: | |
8bb78442 | 1761 | case CPU_UP_PREPARE_FROZEN: |
37810659 | 1762 | init_hrtimers_cpu(scpu); |
c0a31329 TG |
1763 | break; |
1764 | ||
1765 | #ifdef CONFIG_HOTPLUG_CPU | |
94df7de0 SD |
1766 | case CPU_DYING: |
1767 | case CPU_DYING_FROZEN: | |
1768 | clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DYING, &scpu); | |
1769 | break; | |
c0a31329 | 1770 | case CPU_DEAD: |
8bb78442 | 1771 | case CPU_DEAD_FROZEN: |
b2e3c0ad | 1772 | { |
37810659 | 1773 | clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &scpu); |
d5fd43c4 | 1774 | migrate_hrtimers(scpu); |
c0a31329 | 1775 | break; |
b2e3c0ad | 1776 | } |
c0a31329 TG |
1777 | #endif |
1778 | ||
1779 | default: | |
1780 | break; | |
1781 | } | |
1782 | ||
1783 | return NOTIFY_OK; | |
1784 | } | |
1785 | ||
0db0628d | 1786 | static struct notifier_block hrtimers_nb = { |
c0a31329 TG |
1787 | .notifier_call = hrtimer_cpu_notify, |
1788 | }; | |
1789 | ||
1790 | void __init hrtimers_init(void) | |
1791 | { | |
1792 | hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE, | |
1793 | (void *)(long)smp_processor_id()); | |
1794 | register_cpu_notifier(&hrtimers_nb); | |
a6037b61 PZ |
1795 | #ifdef CONFIG_HIGH_RES_TIMERS |
1796 | open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq); | |
1797 | #endif | |
c0a31329 TG |
1798 | } |
1799 | ||
7bb67439 | 1800 | /** |
351b3f7a | 1801 | * schedule_hrtimeout_range_clock - sleep until timeout |
7bb67439 | 1802 | * @expires: timeout value (ktime_t) |
654c8e0b | 1803 | * @delta: slack in expires timeout (ktime_t) |
7bb67439 | 1804 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL |
351b3f7a | 1805 | * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME |
7bb67439 | 1806 | */ |
351b3f7a CE |
1807 | int __sched |
1808 | schedule_hrtimeout_range_clock(ktime_t *expires, unsigned long delta, | |
1809 | const enum hrtimer_mode mode, int clock) | |
7bb67439 AV |
1810 | { |
1811 | struct hrtimer_sleeper t; | |
1812 | ||
1813 | /* | |
1814 | * Optimize when a zero timeout value is given. It does not | |
1815 | * matter whether this is an absolute or a relative time. | |
1816 | */ | |
1817 | if (expires && !expires->tv64) { | |
1818 | __set_current_state(TASK_RUNNING); | |
1819 | return 0; | |
1820 | } | |
1821 | ||
1822 | /* | |
43b21013 | 1823 | * A NULL parameter means "infinite" |
7bb67439 AV |
1824 | */ |
1825 | if (!expires) { | |
1826 | schedule(); | |
1827 | __set_current_state(TASK_RUNNING); | |
1828 | return -EINTR; | |
1829 | } | |
1830 | ||
351b3f7a | 1831 | hrtimer_init_on_stack(&t.timer, clock, mode); |
654c8e0b | 1832 | hrtimer_set_expires_range_ns(&t.timer, *expires, delta); |
7bb67439 AV |
1833 | |
1834 | hrtimer_init_sleeper(&t, current); | |
1835 | ||
cc584b21 | 1836 | hrtimer_start_expires(&t.timer, mode); |
7bb67439 AV |
1837 | if (!hrtimer_active(&t.timer)) |
1838 | t.task = NULL; | |
1839 | ||
1840 | if (likely(t.task)) | |
1841 | schedule(); | |
1842 | ||
1843 | hrtimer_cancel(&t.timer); | |
1844 | destroy_hrtimer_on_stack(&t.timer); | |
1845 | ||
1846 | __set_current_state(TASK_RUNNING); | |
1847 | ||
1848 | return !t.task ? 0 : -EINTR; | |
1849 | } | |
351b3f7a CE |
1850 | |
1851 | /** | |
1852 | * schedule_hrtimeout_range - sleep until timeout | |
1853 | * @expires: timeout value (ktime_t) | |
1854 | * @delta: slack in expires timeout (ktime_t) | |
1855 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL | |
1856 | * | |
1857 | * Make the current task sleep until the given expiry time has | |
1858 | * elapsed. The routine will return immediately unless | |
1859 | * the current task state has been set (see set_current_state()). | |
1860 | * | |
1861 | * The @delta argument gives the kernel the freedom to schedule the | |
1862 | * actual wakeup to a time that is both power and performance friendly. | |
1863 | * The kernel give the normal best effort behavior for "@expires+@delta", | |
1864 | * but may decide to fire the timer earlier, but no earlier than @expires. | |
1865 | * | |
1866 | * You can set the task state as follows - | |
1867 | * | |
1868 | * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to | |
1869 | * pass before the routine returns. | |
1870 | * | |
1871 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
1872 | * delivered to the current task. | |
1873 | * | |
1874 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1875 | * routine returns. | |
1876 | * | |
1877 | * Returns 0 when the timer has expired otherwise -EINTR | |
1878 | */ | |
1879 | int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta, | |
1880 | const enum hrtimer_mode mode) | |
1881 | { | |
1882 | return schedule_hrtimeout_range_clock(expires, delta, mode, | |
1883 | CLOCK_MONOTONIC); | |
1884 | } | |
654c8e0b AV |
1885 | EXPORT_SYMBOL_GPL(schedule_hrtimeout_range); |
1886 | ||
1887 | /** | |
1888 | * schedule_hrtimeout - sleep until timeout | |
1889 | * @expires: timeout value (ktime_t) | |
1890 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL | |
1891 | * | |
1892 | * Make the current task sleep until the given expiry time has | |
1893 | * elapsed. The routine will return immediately unless | |
1894 | * the current task state has been set (see set_current_state()). | |
1895 | * | |
1896 | * You can set the task state as follows - | |
1897 | * | |
1898 | * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to | |
1899 | * pass before the routine returns. | |
1900 | * | |
1901 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
1902 | * delivered to the current task. | |
1903 | * | |
1904 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1905 | * routine returns. | |
1906 | * | |
1907 | * Returns 0 when the timer has expired otherwise -EINTR | |
1908 | */ | |
1909 | int __sched schedule_hrtimeout(ktime_t *expires, | |
1910 | const enum hrtimer_mode mode) | |
1911 | { | |
1912 | return schedule_hrtimeout_range(expires, 0, mode); | |
1913 | } | |
7bb67439 | 1914 | EXPORT_SYMBOL_GPL(schedule_hrtimeout); |