1 /*
2 * linux/kernel/hrtimer.c
3 *
4 * Copyright(C) 2005-2006, Thomas Gleixner <tglx@linutronix.de>
5 * Copyright(C) 2005-2007, Red Hat, Inc., Ingo Molnar
6 * Copyright(C) 2006-2007 Timesys Corp., Thomas Gleixner
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 *
25 * Help, testing, suggestions, bugfixes, improvements were
26 * provided by:
27 *
28 * George Anzinger, Andrew Morton, Steven Rostedt, Roman Zippel
29 * et. al.
30 *
31 * For licencing details see kernel-base/COPYING
32 */
33
34 #include <linux/cpu.h>
35 #include <linux/export.h>
36 #include <linux/percpu.h>
37 #include <linux/hrtimer.h>
38 #include <linux/notifier.h>
39 #include <linux/syscalls.h>
40 #include <linux/kallsyms.h>
41 #include <linux/interrupt.h>
42 #include <linux/tick.h>
43 #include <linux/seq_file.h>
44 #include <linux/err.h>
45 #include <linux/debugobjects.h>
46 #include <linux/sched.h>
47 #include <linux/timer.h>
48
49 #include <asm/uaccess.h>
50
51 #include <trace/events/timer.h>
52
53 /*
54 * The timer bases:
55 *
56 * There are more clockids then hrtimer bases. Thus, we index
57 * into the timer bases by the hrtimer_base_type enum. When trying
58 * to reach a base using a clockid, hrtimer_clockid_to_base()
59 * is used to convert from clockid to the proper hrtimer_base_type.
60 */
61 DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
62 {
63
64 .lock = __RAW_SPIN_LOCK_UNLOCKED(hrtimer_bases.lock),
65 .clock_base =
66 {
67 {
68 .index = HRTIMER_BASE_MONOTONIC,
69 .clockid = CLOCK_MONOTONIC,
70 .get_time = &ktime_get,
71 .resolution = KTIME_LOW_RES,
72 },
73 {
74 .index = HRTIMER_BASE_REALTIME,
75 .clockid = CLOCK_REALTIME,
76 .get_time = &ktime_get_real,
77 .resolution = KTIME_LOW_RES,
78 },
79 {
80 .index = HRTIMER_BASE_BOOTTIME,
81 .clockid = CLOCK_BOOTTIME,
82 .get_time = &ktime_get_boottime,
83 .resolution = KTIME_LOW_RES,
84 },
85 }
86 };
87
88 static const int hrtimer_clock_to_base_table[MAX_CLOCKS] = {
89 [CLOCK_REALTIME] = HRTIMER_BASE_REALTIME,
90 [CLOCK_MONOTONIC] = HRTIMER_BASE_MONOTONIC,
91 [CLOCK_BOOTTIME] = HRTIMER_BASE_BOOTTIME,
92 };
93
hrtimer_clockid_to_base(clockid_t clock_id)94 static inline int hrtimer_clockid_to_base(clockid_t clock_id)
95 {
96 return hrtimer_clock_to_base_table[clock_id];
97 }
98
99
100 /*
101 * Get the coarse grained time at the softirq based on xtime and
102 * wall_to_monotonic.
103 */
hrtimer_get_softirq_time(struct hrtimer_cpu_base * base)104 static void hrtimer_get_softirq_time(struct hrtimer_cpu_base *base)
105 {
106 ktime_t xtim, mono, boot;
107 struct timespec xts, tom, slp;
108
109 get_xtime_and_monotonic_and_sleep_offset(&xts, &tom, &slp);
110
111 xtim = timespec_to_ktime(xts);
112 mono = ktime_add(xtim, timespec_to_ktime(tom));
113 boot = ktime_add(mono, timespec_to_ktime(slp));
114 base->clock_base[HRTIMER_BASE_REALTIME].softirq_time = xtim;
115 base->clock_base[HRTIMER_BASE_MONOTONIC].softirq_time = mono;
116 base->clock_base[HRTIMER_BASE_BOOTTIME].softirq_time = boot;
117 }
118
119 /*
120 * Functions and macros which are different for UP/SMP systems are kept in a
121 * single place
122 */
123 #ifdef CONFIG_SMP
124
125 /*
126 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
127 * means that all timers which are tied to this base via timer->base are
128 * locked, and the base itself is locked too.
129 *
130 * So __run_timers/migrate_timers can safely modify all timers which could
131 * be found on the lists/queues.
132 *
133 * When the timer's base is locked, and the timer removed from list, it is
134 * possible to set timer->base = NULL and drop the lock: the timer remains
135 * locked.
136 */
137 static
lock_hrtimer_base(const struct hrtimer * timer,unsigned long * flags)138 struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
139 unsigned long *flags)
140 {
141 struct hrtimer_clock_base *base;
142
143 for (;;) {
144 base = timer->base;
145 if (likely(base != NULL)) {
146 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
147 if (likely(base == timer->base))
148 return base;
149 /* The timer has migrated to another CPU: */
150 raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
151 }
152 cpu_relax();
153 }
154 }
155
156
157 /*
158 * Get the preferred target CPU for NOHZ
159 */
hrtimer_get_target(int this_cpu,int pinned)160 static int hrtimer_get_target(int this_cpu, int pinned)
161 {
162 #ifdef CONFIG_NO_HZ
163 if (!pinned && get_sysctl_timer_migration() && idle_cpu(this_cpu))
164 return get_nohz_timer_target();
165 #endif
166 return this_cpu;
167 }
168
169 /*
170 * With HIGHRES=y we do not migrate the timer when it is expiring
171 * before the next event on the target cpu because we cannot reprogram
172 * the target cpu hardware and we would cause it to fire late.
173 *
174 * Called with cpu_base->lock of target cpu held.
175 */
176 static int
hrtimer_check_target(struct hrtimer * timer,struct hrtimer_clock_base * new_base)177 hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base)
178 {
179 #ifdef CONFIG_HIGH_RES_TIMERS
180 ktime_t expires;
181
182 if (!new_base->cpu_base->hres_active)
183 return 0;
184
185 expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset);
186 return expires.tv64 <= new_base->cpu_base->expires_next.tv64;
187 #else
188 return 0;
189 #endif
190 }
191
192 /*
193 * Switch the timer base to the current CPU when possible.
194 */
195 static inline struct hrtimer_clock_base *
switch_hrtimer_base(struct hrtimer * timer,struct hrtimer_clock_base * base,int pinned)196 switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
197 int pinned)
198 {
199 struct hrtimer_clock_base *new_base;
200 struct hrtimer_cpu_base *new_cpu_base;
201 int this_cpu = smp_processor_id();
202 int cpu = hrtimer_get_target(this_cpu, pinned);
203 int basenum = base->index;
204
205 again:
206 new_cpu_base = &per_cpu(hrtimer_bases, cpu);
207 new_base = &new_cpu_base->clock_base[basenum];
208
209 if (base != new_base) {
210 /*
211 * We are trying to move timer to new_base.
212 * However we can't change timer's base while it is running,
213 * so we keep it on the same CPU. No hassle vs. reprogramming
214 * the event source in the high resolution case. The softirq
215 * code will take care of this when the timer function has
216 * completed. There is no conflict as we hold the lock until
217 * the timer is enqueued.
218 */
219 if (unlikely(hrtimer_callback_running(timer)))
220 return base;
221
222 /* See the comment in lock_timer_base() */
223 timer->base = NULL;
224 raw_spin_unlock(&base->cpu_base->lock);
225 raw_spin_lock(&new_base->cpu_base->lock);
226
227 if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) {
228 cpu = this_cpu;
229 raw_spin_unlock(&new_base->cpu_base->lock);
230 raw_spin_lock(&base->cpu_base->lock);
231 timer->base = base;
232 goto again;
233 }
234 timer->base = new_base;
235 } else {
236 if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) {
237 cpu = this_cpu;
238 goto again;
239 }
240 }
241 return new_base;
242 }
243
244 #else /* CONFIG_SMP */
245
246 static inline struct hrtimer_clock_base *
lock_hrtimer_base(const struct hrtimer * timer,unsigned long * flags)247 lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
248 {
249 struct hrtimer_clock_base *base = timer->base;
250
251 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
252
253 return base;
254 }
255
256 # define switch_hrtimer_base(t, b, p) (b)
257
258 #endif /* !CONFIG_SMP */
259
260 /*
261 * Functions for the union type storage format of ktime_t which are
262 * too large for inlining:
263 */
264 #if BITS_PER_LONG < 64
265 # ifndef CONFIG_KTIME_SCALAR
266 /**
267 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
268 * @kt: addend
269 * @nsec: the scalar nsec value to add
270 *
271 * Returns the sum of kt and nsec in ktime_t format
272 */
ktime_add_ns(const ktime_t kt,u64 nsec)273 ktime_t ktime_add_ns(const ktime_t kt, u64 nsec)
274 {
275 ktime_t tmp;
276
277 if (likely(nsec < NSEC_PER_SEC)) {
278 tmp.tv64 = nsec;
279 } else {
280 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
281
282 tmp = ktime_set((long)nsec, rem);
283 }
284
285 return ktime_add(kt, tmp);
286 }
287
288 EXPORT_SYMBOL_GPL(ktime_add_ns);
289
290 /**
291 * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable
292 * @kt: minuend
293 * @nsec: the scalar nsec value to subtract
294 *
295 * Returns the subtraction of @nsec from @kt in ktime_t format
296 */
ktime_sub_ns(const ktime_t kt,u64 nsec)297 ktime_t ktime_sub_ns(const ktime_t kt, u64 nsec)
298 {
299 ktime_t tmp;
300
301 if (likely(nsec < NSEC_PER_SEC)) {
302 tmp.tv64 = nsec;
303 } else {
304 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
305
306 /* Make sure nsec fits into long */
307 if (unlikely(nsec > KTIME_SEC_MAX))
308 return (ktime_t){ .tv64 = KTIME_MAX };
309
310 tmp = ktime_set((long)nsec, rem);
311 }
312
313 return ktime_sub(kt, tmp);
314 }
315
316 EXPORT_SYMBOL_GPL(ktime_sub_ns);
317 # endif /* !CONFIG_KTIME_SCALAR */
318
319 /*
320 * Divide a ktime value by a nanosecond value
321 */
ktime_divns(const ktime_t kt,s64 div)322 u64 ktime_divns(const ktime_t kt, s64 div)
323 {
324 u64 dclc;
325 int sft = 0;
326
327 dclc = ktime_to_ns(kt);
328 /* Make sure the divisor is less than 2^32: */
329 while (div >> 32) {
330 sft++;
331 div >>= 1;
332 }
333 dclc >>= sft;
334 do_div(dclc, (unsigned long) div);
335
336 return dclc;
337 }
338 #endif /* BITS_PER_LONG >= 64 */
339
340 /*
341 * Add two ktime values and do a safety check for overflow:
342 */
ktime_add_safe(const ktime_t lhs,const ktime_t rhs)343 ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
344 {
345 ktime_t res = ktime_add(lhs, rhs);
346
347 /*
348 * We use KTIME_SEC_MAX here, the maximum timeout which we can
349 * return to user space in a timespec:
350 */
351 if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64)
352 res = ktime_set(KTIME_SEC_MAX, 0);
353
354 return res;
355 }
356
357 EXPORT_SYMBOL_GPL(ktime_add_safe);
358
359 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
360
361 static struct debug_obj_descr hrtimer_debug_descr;
362
hrtimer_debug_hint(void * addr)363 static void *hrtimer_debug_hint(void *addr)
364 {
365 return ((struct hrtimer *) addr)->function;
366 }
367
368 /*
369 * fixup_init is called when:
370 * - an active object is initialized
371 */
hrtimer_fixup_init(void * addr,enum debug_obj_state state)372 static int hrtimer_fixup_init(void *addr, enum debug_obj_state state)
373 {
374 struct hrtimer *timer = addr;
375
376 switch (state) {
377 case ODEBUG_STATE_ACTIVE:
378 hrtimer_cancel(timer);
379 debug_object_init(timer, &hrtimer_debug_descr);
380 return 1;
381 default:
382 return 0;
383 }
384 }
385
386 /*
387 * fixup_activate is called when:
388 * - an active object is activated
389 * - an unknown object is activated (might be a statically initialized object)
390 */
hrtimer_fixup_activate(void * addr,enum debug_obj_state state)391 static int hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
392 {
393 switch (state) {
394
395 case ODEBUG_STATE_NOTAVAILABLE:
396 WARN_ON_ONCE(1);
397 return 0;
398
399 case ODEBUG_STATE_ACTIVE:
400 WARN_ON(1);
401
402 default:
403 return 0;
404 }
405 }
406
407 /*
408 * fixup_free is called when:
409 * - an active object is freed
410 */
hrtimer_fixup_free(void * addr,enum debug_obj_state state)411 static int hrtimer_fixup_free(void *addr, enum debug_obj_state state)
412 {
413 struct hrtimer *timer = addr;
414
415 switch (state) {
416 case ODEBUG_STATE_ACTIVE:
417 hrtimer_cancel(timer);
418 debug_object_free(timer, &hrtimer_debug_descr);
419 return 1;
420 default:
421 return 0;
422 }
423 }
424
425 static struct debug_obj_descr hrtimer_debug_descr = {
426 .name = "hrtimer",
427 .debug_hint = hrtimer_debug_hint,
428 .fixup_init = hrtimer_fixup_init,
429 .fixup_activate = hrtimer_fixup_activate,
430 .fixup_free = hrtimer_fixup_free,
431 };
432
debug_hrtimer_init(struct hrtimer * timer)433 static inline void debug_hrtimer_init(struct hrtimer *timer)
434 {
435 debug_object_init(timer, &hrtimer_debug_descr);
436 }
437
debug_hrtimer_activate(struct hrtimer * timer)438 static inline void debug_hrtimer_activate(struct hrtimer *timer)
439 {
440 debug_object_activate(timer, &hrtimer_debug_descr);
441 }
442
debug_hrtimer_deactivate(struct hrtimer * timer)443 static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
444 {
445 debug_object_deactivate(timer, &hrtimer_debug_descr);
446 }
447
debug_hrtimer_free(struct hrtimer * timer)448 static inline void debug_hrtimer_free(struct hrtimer *timer)
449 {
450 debug_object_free(timer, &hrtimer_debug_descr);
451 }
452
453 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
454 enum hrtimer_mode mode);
455
hrtimer_init_on_stack(struct hrtimer * timer,clockid_t clock_id,enum hrtimer_mode mode)456 void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
457 enum hrtimer_mode mode)
458 {
459 debug_object_init_on_stack(timer, &hrtimer_debug_descr);
460 __hrtimer_init(timer, clock_id, mode);
461 }
462 EXPORT_SYMBOL_GPL(hrtimer_init_on_stack);
463
destroy_hrtimer_on_stack(struct hrtimer * timer)464 void destroy_hrtimer_on_stack(struct hrtimer *timer)
465 {
466 debug_object_free(timer, &hrtimer_debug_descr);
467 }
468
469 #else
debug_hrtimer_init(struct hrtimer * timer)470 static inline void debug_hrtimer_init(struct hrtimer *timer) { }
debug_hrtimer_activate(struct hrtimer * timer)471 static inline void debug_hrtimer_activate(struct hrtimer *timer) { }
debug_hrtimer_deactivate(struct hrtimer * timer)472 static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
473 #endif
474
475 static inline void
debug_init(struct hrtimer * timer,clockid_t clockid,enum hrtimer_mode mode)476 debug_init(struct hrtimer *timer, clockid_t clockid,
477 enum hrtimer_mode mode)
478 {
479 debug_hrtimer_init(timer);
480 trace_hrtimer_init(timer, clockid, mode);
481 }
482
debug_activate(struct hrtimer * timer)483 static inline void debug_activate(struct hrtimer *timer)
484 {
485 debug_hrtimer_activate(timer);
486 trace_hrtimer_start(timer);
487 }
488
debug_deactivate(struct hrtimer * timer)489 static inline void debug_deactivate(struct hrtimer *timer)
490 {
491 debug_hrtimer_deactivate(timer);
492 trace_hrtimer_cancel(timer);
493 }
494
495 /* High resolution timer related functions */
496 #ifdef CONFIG_HIGH_RES_TIMERS
497
498 /*
499 * High resolution timer enabled ?
500 */
501 static int hrtimer_hres_enabled __read_mostly = 1;
502
503 /*
504 * Enable / Disable high resolution mode
505 */
setup_hrtimer_hres(char * str)506 static int __init setup_hrtimer_hres(char *str)
507 {
508 if (!strcmp(str, "off"))
509 hrtimer_hres_enabled = 0;
510 else if (!strcmp(str, "on"))
511 hrtimer_hres_enabled = 1;
512 else
513 return 0;
514 return 1;
515 }
516
517 __setup("highres=", setup_hrtimer_hres);
518
519 /*
520 * hrtimer_high_res_enabled - query, if the highres mode is enabled
521 */
hrtimer_is_hres_enabled(void)522 static inline int hrtimer_is_hres_enabled(void)
523 {
524 return hrtimer_hres_enabled;
525 }
526
527 /*
528 * Is the high resolution mode active ?
529 */
hrtimer_hres_active(void)530 static inline int hrtimer_hres_active(void)
531 {
532 return __this_cpu_read(hrtimer_bases.hres_active);
533 }
534
535 /*
536 * Reprogram the event source with checking both queues for the
537 * next event
538 * Called with interrupts disabled and base->lock held
539 */
540 static void
hrtimer_force_reprogram(struct hrtimer_cpu_base * cpu_base,int skip_equal)541 hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
542 {
543 int i;
544 struct hrtimer_clock_base *base = cpu_base->clock_base;
545 ktime_t expires, expires_next;
546
547 expires_next.tv64 = KTIME_MAX;
548
549 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
550 struct hrtimer *timer;
551 struct timerqueue_node *next;
552
553 next = timerqueue_getnext(&base->active);
554 if (!next)
555 continue;
556 timer = container_of(next, struct hrtimer, node);
557
558 expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
559 /*
560 * clock_was_set() has changed base->offset so the
561 * result might be negative. Fix it up to prevent a
562 * false positive in clockevents_program_event()
563 */
564 if (expires.tv64 < 0)
565 expires.tv64 = 0;
566 if (expires.tv64 < expires_next.tv64)
567 expires_next = expires;
568 }
569
570 if (skip_equal && expires_next.tv64 == cpu_base->expires_next.tv64)
571 return;
572
573 cpu_base->expires_next.tv64 = expires_next.tv64;
574
575 /*
576 * If a hang was detected in the last timer interrupt then we
577 * leave the hang delay active in the hardware. We want the
578 * system to make progress. That also prevents the following
579 * scenario:
580 * T1 expires 50ms from now
581 * T2 expires 5s from now
582 *
583 * T1 is removed, so this code is called and would reprogram
584 * the hardware to 5s from now. Any hrtimer_start after that
585 * will not reprogram the hardware due to hang_detected being
586 * set. So we'd effectivly block all timers until the T2 event
587 * fires.
588 */
589 if (cpu_base->hang_detected)
590 return;
591
592 if (cpu_base->expires_next.tv64 != KTIME_MAX)
593 tick_program_event(cpu_base->expires_next, 1);
594 }
595
596 /*
597 * Shared reprogramming for clock_realtime and clock_monotonic
598 *
599 * When a timer is enqueued and expires earlier than the already enqueued
600 * timers, we have to check, whether it expires earlier than the timer for
601 * which the clock event device was armed.
602 *
603 * Called with interrupts disabled and base->cpu_base.lock held
604 */
hrtimer_reprogram(struct hrtimer * timer,struct hrtimer_clock_base * base)605 static int hrtimer_reprogram(struct hrtimer *timer,
606 struct hrtimer_clock_base *base)
607 {
608 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
609 ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
610 int res;
611
612 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
613
614 /*
615 * When the callback is running, we do not reprogram the clock event
616 * device. The timer callback is either running on a different CPU or
617 * the callback is executed in the hrtimer_interrupt context. The
618 * reprogramming is handled either by the softirq, which called the
619 * callback or at the end of the hrtimer_interrupt.
620 */
621 if (hrtimer_callback_running(timer))
622 return 0;
623
624 /*
625 * CLOCK_REALTIME timer might be requested with an absolute
626 * expiry time which is less than base->offset. Nothing wrong
627 * about that, just avoid to call into the tick code, which
628 * has now objections against negative expiry values.
629 */
630 if (expires.tv64 < 0)
631 return -ETIME;
632
633 if (expires.tv64 >= cpu_base->expires_next.tv64)
634 return 0;
635
636 /*
637 * If a hang was detected in the last timer interrupt then we
638 * do not schedule a timer which is earlier than the expiry
639 * which we enforced in the hang detection. We want the system
640 * to make progress.
641 */
642 if (cpu_base->hang_detected)
643 return 0;
644
645 /*
646 * Clockevents returns -ETIME, when the event was in the past.
647 */
648 res = tick_program_event(expires, 0);
649 if (!IS_ERR_VALUE(res))
650 cpu_base->expires_next = expires;
651 return res;
652 }
653
654 /*
655 * Initialize the high resolution related parts of cpu_base
656 */
hrtimer_init_hres(struct hrtimer_cpu_base * base)657 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
658 {
659 base->expires_next.tv64 = KTIME_MAX;
660 base->hres_active = 0;
661 }
662
663 /*
664 * When High resolution timers are active, try to reprogram. Note, that in case
665 * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry
666 * check happens. The timer gets enqueued into the rbtree. The reprogramming
667 * and expiry check is done in the hrtimer_interrupt or in the softirq.
668 */
hrtimer_enqueue_reprogram(struct hrtimer * timer,struct hrtimer_clock_base * base)669 static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
670 struct hrtimer_clock_base *base)
671 {
672 return base->cpu_base->hres_active && hrtimer_reprogram(timer, base);
673 }
674
hrtimer_update_base(struct hrtimer_cpu_base * base)675 static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base)
676 {
677 ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset;
678 ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset;
679
680 return ktime_get_update_offsets(offs_real, offs_boot);
681 }
682
683 /*
684 * Retrigger next event is called after clock was set
685 *
686 * Called with interrupts disabled via on_each_cpu()
687 */
retrigger_next_event(void * arg)688 static void retrigger_next_event(void *arg)
689 {
690 struct hrtimer_cpu_base *base = &__get_cpu_var(hrtimer_bases);
691
692 if (!hrtimer_hres_active())
693 return;
694
695 raw_spin_lock(&base->lock);
696 hrtimer_update_base(base);
697 hrtimer_force_reprogram(base, 0);
698 raw_spin_unlock(&base->lock);
699 }
700
701 /*
702 * Switch to high resolution mode
703 */
hrtimer_switch_to_hres(void)704 static int hrtimer_switch_to_hres(void)
705 {
706 int i, cpu = smp_processor_id();
707 struct hrtimer_cpu_base *base = &per_cpu(hrtimer_bases, cpu);
708 unsigned long flags;
709
710 if (base->hres_active)
711 return 1;
712
713 local_irq_save(flags);
714
715 if (tick_init_highres()) {
716 local_irq_restore(flags);
717 printk(KERN_WARNING "Could not switch to high resolution "
718 "mode on CPU %d\n", cpu);
719 return 0;
720 }
721 base->hres_active = 1;
722 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
723 base->clock_base[i].resolution = KTIME_HIGH_RES;
724
725 tick_setup_sched_timer();
726 /* "Retrigger" the interrupt to get things going */
727 retrigger_next_event(NULL);
728 local_irq_restore(flags);
729 return 1;
730 }
731
clock_was_set_work(struct work_struct * work)732 static void clock_was_set_work(struct work_struct *work)
733 {
734 clock_was_set();
735 }
736
737 static DECLARE_WORK(hrtimer_work, clock_was_set_work);
738
739 /*
740 * Called from timekeeping and resume code to reprogramm the hrtimer
741 * interrupt device on all cpus.
742 */
clock_was_set_delayed(void)743 void clock_was_set_delayed(void)
744 {
745 schedule_work(&hrtimer_work);
746 }
747
748 #else
749
hrtimer_hres_active(void)750 static inline int hrtimer_hres_active(void) { return 0; }
hrtimer_is_hres_enabled(void)751 static inline int hrtimer_is_hres_enabled(void) { return 0; }
hrtimer_switch_to_hres(void)752 static inline int hrtimer_switch_to_hres(void) { return 0; }
753 static inline void
hrtimer_force_reprogram(struct hrtimer_cpu_base * base,int skip_equal)754 hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { }
hrtimer_enqueue_reprogram(struct hrtimer * timer,struct hrtimer_clock_base * base)755 static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
756 struct hrtimer_clock_base *base)
757 {
758 return 0;
759 }
hrtimer_init_hres(struct hrtimer_cpu_base * base)760 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { }
retrigger_next_event(void * arg)761 static inline void retrigger_next_event(void *arg) { }
762
763 #endif /* CONFIG_HIGH_RES_TIMERS */
764
765 /*
766 * Clock realtime was set
767 *
768 * Change the offset of the realtime clock vs. the monotonic
769 * clock.
770 *
771 * We might have to reprogram the high resolution timer interrupt. On
772 * SMP we call the architecture specific code to retrigger _all_ high
773 * resolution timer interrupts. On UP we just disable interrupts and
774 * call the high resolution interrupt code.
775 */
clock_was_set(void)776 void clock_was_set(void)
777 {
778 #ifdef CONFIG_HIGH_RES_TIMERS
779 /* Retrigger the CPU local events everywhere */
780 on_each_cpu(retrigger_next_event, NULL, 1);
781 #endif
782 timerfd_clock_was_set();
783 }
784
785 /*
786 * During resume we might have to reprogram the high resolution timer
787 * interrupt (on the local CPU):
788 */
hrtimers_resume(void)789 void hrtimers_resume(void)
790 {
791 WARN_ONCE(!irqs_disabled(),
792 KERN_INFO "hrtimers_resume() called with IRQs enabled!");
793
794 /* Retrigger on the local CPU */
795 retrigger_next_event(NULL);
796 /* And schedule a retrigger for all others */
797 clock_was_set_delayed();
798 }
799
timer_stats_hrtimer_set_start_info(struct hrtimer * timer)800 static inline void timer_stats_hrtimer_set_start_info(struct hrtimer *timer)
801 {
802 #ifdef CONFIG_TIMER_STATS
803 if (timer->start_site)
804 return;
805 timer->start_site = __builtin_return_address(0);
806 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
807 timer->start_pid = current->pid;
808 #endif
809 }
810
timer_stats_hrtimer_clear_start_info(struct hrtimer * timer)811 static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer *timer)
812 {
813 #ifdef CONFIG_TIMER_STATS
814 timer->start_site = NULL;
815 #endif
816 }
817
timer_stats_account_hrtimer(struct hrtimer * timer)818 static inline void timer_stats_account_hrtimer(struct hrtimer *timer)
819 {
820 #ifdef CONFIG_TIMER_STATS
821 if (likely(!timer_stats_active))
822 return;
823 timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
824 timer->function, timer->start_comm, 0);
825 #endif
826 }
827
828 /*
829 * Counterpart to lock_hrtimer_base above:
830 */
831 static inline
unlock_hrtimer_base(const struct hrtimer * timer,unsigned long * flags)832 void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
833 {
834 raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
835 }
836
837 /**
838 * hrtimer_forward - forward the timer expiry
839 * @timer: hrtimer to forward
840 * @now: forward past this time
841 * @interval: the interval to forward
842 *
843 * Forward the timer expiry so it will expire in the future.
844 * Returns the number of overruns.
845 */
hrtimer_forward(struct hrtimer * timer,ktime_t now,ktime_t interval)846 u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
847 {
848 u64 orun = 1;
849 ktime_t delta;
850
851 delta = ktime_sub(now, hrtimer_get_expires(timer));
852
853 if (delta.tv64 < 0)
854 return 0;
855
856 if (interval.tv64 < timer->base->resolution.tv64)
857 interval.tv64 = timer->base->resolution.tv64;
858
859 if (unlikely(delta.tv64 >= interval.tv64)) {
860 s64 incr = ktime_to_ns(interval);
861
862 orun = ktime_divns(delta, incr);
863 hrtimer_add_expires_ns(timer, incr * orun);
864 if (hrtimer_get_expires_tv64(timer) > now.tv64)
865 return orun;
866 /*
867 * This (and the ktime_add() below) is the
868 * correction for exact:
869 */
870 orun++;
871 }
872 hrtimer_add_expires(timer, interval);
873
874 return orun;
875 }
876 EXPORT_SYMBOL_GPL(hrtimer_forward);
877
878 /*
879 * enqueue_hrtimer - internal function to (re)start a timer
880 *
881 * The timer is inserted in expiry order. Insertion into the
882 * red black tree is O(log(n)). Must hold the base lock.
883 *
884 * Returns 1 when the new timer is the leftmost timer in the tree.
885 */
enqueue_hrtimer(struct hrtimer * timer,struct hrtimer_clock_base * base)886 static int enqueue_hrtimer(struct hrtimer *timer,
887 struct hrtimer_clock_base *base)
888 {
889 debug_activate(timer);
890
891 timerqueue_add(&base->active, &timer->node);
892 base->cpu_base->active_bases |= 1 << base->index;
893
894 /*
895 * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
896 * state of a possibly running callback.
897 */
898 timer->state |= HRTIMER_STATE_ENQUEUED;
899
900 return (&timer->node == base->active.next);
901 }
902
903 /*
904 * __remove_hrtimer - internal function to remove a timer
905 *
906 * Caller must hold the base lock.
907 *
908 * High resolution timer mode reprograms the clock event device when the
909 * timer is the one which expires next. The caller can disable this by setting
910 * reprogram to zero. This is useful, when the context does a reprogramming
911 * anyway (e.g. timer interrupt)
912 */
__remove_hrtimer(struct hrtimer * timer,struct hrtimer_clock_base * base,unsigned long newstate,int reprogram)913 static void __remove_hrtimer(struct hrtimer *timer,
914 struct hrtimer_clock_base *base,
915 unsigned long newstate, int reprogram)
916 {
917 struct timerqueue_node *next_timer;
918 if (!(timer->state & HRTIMER_STATE_ENQUEUED))
919 goto out;
920
921 next_timer = timerqueue_getnext(&base->active);
922 timerqueue_del(&base->active, &timer->node);
923 if (&timer->node == next_timer) {
924 #ifdef CONFIG_HIGH_RES_TIMERS
925 /* Reprogram the clock event device. if enabled */
926 if (reprogram && hrtimer_hres_active()) {
927 ktime_t expires;
928
929 expires = ktime_sub(hrtimer_get_expires(timer),
930 base->offset);
931 if (base->cpu_base->expires_next.tv64 == expires.tv64)
932 hrtimer_force_reprogram(base->cpu_base, 1);
933 }
934 #endif
935 }
936 if (!timerqueue_getnext(&base->active))
937 base->cpu_base->active_bases &= ~(1 << base->index);
938 out:
939 timer->state = newstate;
940 }
941
942 /*
943 * remove hrtimer, called with base lock held
944 */
945 static inline int
remove_hrtimer(struct hrtimer * timer,struct hrtimer_clock_base * base)946 remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base)
947 {
948 if (hrtimer_is_queued(timer)) {
949 unsigned long state;
950 int reprogram;
951
952 /*
953 * Remove the timer and force reprogramming when high
954 * resolution mode is active and the timer is on the current
955 * CPU. If we remove a timer on another CPU, reprogramming is
956 * skipped. The interrupt event on this CPU is fired and
957 * reprogramming happens in the interrupt handler. This is a
958 * rare case and less expensive than a smp call.
959 */
960 debug_deactivate(timer);
961 timer_stats_hrtimer_clear_start_info(timer);
962 reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases);
963 /*
964 * We must preserve the CALLBACK state flag here,
965 * otherwise we could move the timer base in
966 * switch_hrtimer_base.
967 */
968 state = timer->state & HRTIMER_STATE_CALLBACK;
969 __remove_hrtimer(timer, base, state, reprogram);
970 return 1;
971 }
972 return 0;
973 }
974
__hrtimer_start_range_ns(struct hrtimer * timer,ktime_t tim,unsigned long delta_ns,const enum hrtimer_mode mode,int wakeup)975 int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
976 unsigned long delta_ns, const enum hrtimer_mode mode,
977 int wakeup)
978 {
979 struct hrtimer_clock_base *base, *new_base;
980 unsigned long flags;
981 int ret, leftmost;
982
983 base = lock_hrtimer_base(timer, &flags);
984
985 /* Remove an active timer from the queue: */
986 ret = remove_hrtimer(timer, base);
987
988 if (mode & HRTIMER_MODE_REL) {
989 tim = ktime_add_safe(tim, base->get_time());
990 /*
991 * CONFIG_TIME_LOW_RES is a temporary way for architectures
992 * to signal that they simply return xtime in
993 * do_gettimeoffset(). In this case we want to round up by
994 * resolution when starting a relative timer, to avoid short
995 * timeouts. This will go away with the GTOD framework.
996 */
997 #ifdef CONFIG_TIME_LOW_RES
998 tim = ktime_add_safe(tim, base->resolution);
999 #endif
1000 }
1001
1002 hrtimer_set_expires_range_ns(timer, tim, delta_ns);
1003
1004 /* Switch the timer base, if necessary: */
1005 new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED);
1006
1007 timer_stats_hrtimer_set_start_info(timer);
1008
1009 leftmost = enqueue_hrtimer(timer, new_base);
1010
1011 /*
1012 * Only allow reprogramming if the new base is on this CPU.
1013 * (it might still be on another CPU if the timer was pending)
1014 *
1015 * XXX send_remote_softirq() ?
1016 */
1017 if (leftmost && new_base->cpu_base == &__get_cpu_var(hrtimer_bases)
1018 && hrtimer_enqueue_reprogram(timer, new_base)) {
1019 if (wakeup) {
1020 /*
1021 * We need to drop cpu_base->lock to avoid a
1022 * lock ordering issue vs. rq->lock.
1023 */
1024 raw_spin_unlock(&new_base->cpu_base->lock);
1025 raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1026 local_irq_restore(flags);
1027 return ret;
1028 } else {
1029 __raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1030 }
1031 }
1032
1033 unlock_hrtimer_base(timer, &flags);
1034
1035 return ret;
1036 }
1037
1038 /**
1039 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
1040 * @timer: the timer to be added
1041 * @tim: expiry time
1042 * @delta_ns: "slack" range for the timer
1043 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
1044 *
1045 * Returns:
1046 * 0 on success
1047 * 1 when the timer was active
1048 */
hrtimer_start_range_ns(struct hrtimer * timer,ktime_t tim,unsigned long delta_ns,const enum hrtimer_mode mode)1049 int hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1050 unsigned long delta_ns, const enum hrtimer_mode mode)
1051 {
1052 return __hrtimer_start_range_ns(timer, tim, delta_ns, mode, 1);
1053 }
1054 EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
1055
1056 /**
1057 * hrtimer_start - (re)start an hrtimer on the current CPU
1058 * @timer: the timer to be added
1059 * @tim: expiry time
1060 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
1061 *
1062 * Returns:
1063 * 0 on success
1064 * 1 when the timer was active
1065 */
1066 int
hrtimer_start(struct hrtimer * timer,ktime_t tim,const enum hrtimer_mode mode)1067 hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode)
1068 {
1069 return __hrtimer_start_range_ns(timer, tim, 0, mode, 1);
1070 }
1071 EXPORT_SYMBOL_GPL(hrtimer_start);
1072
1073
1074 /**
1075 * hrtimer_try_to_cancel - try to deactivate a timer
1076 * @timer: hrtimer to stop
1077 *
1078 * Returns:
1079 * 0 when the timer was not active
1080 * 1 when the timer was active
1081 * -1 when the timer is currently excuting the callback function and
1082 * cannot be stopped
1083 */
hrtimer_try_to_cancel(struct hrtimer * timer)1084 int hrtimer_try_to_cancel(struct hrtimer *timer)
1085 {
1086 struct hrtimer_clock_base *base;
1087 unsigned long flags;
1088 int ret = -1;
1089
1090 base = lock_hrtimer_base(timer, &flags);
1091
1092 if (!hrtimer_callback_running(timer))
1093 ret = remove_hrtimer(timer, base);
1094
1095 unlock_hrtimer_base(timer, &flags);
1096
1097 return ret;
1098
1099 }
1100 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1101
1102 /**
1103 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1104 * @timer: the timer to be cancelled
1105 *
1106 * Returns:
1107 * 0 when the timer was not active
1108 * 1 when the timer was active
1109 */
hrtimer_cancel(struct hrtimer * timer)1110 int hrtimer_cancel(struct hrtimer *timer)
1111 {
1112 for (;;) {
1113 int ret = hrtimer_try_to_cancel(timer);
1114
1115 if (ret >= 0)
1116 return ret;
1117 cpu_relax();
1118 }
1119 }
1120 EXPORT_SYMBOL_GPL(hrtimer_cancel);
1121
1122 /**
1123 * hrtimer_get_remaining - get remaining time for the timer
1124 * @timer: the timer to read
1125 */
hrtimer_get_remaining(const struct hrtimer * timer)1126 ktime_t hrtimer_get_remaining(const struct hrtimer *timer)
1127 {
1128 unsigned long flags;
1129 ktime_t rem;
1130
1131 lock_hrtimer_base(timer, &flags);
1132 rem = hrtimer_expires_remaining(timer);
1133 unlock_hrtimer_base(timer, &flags);
1134
1135 return rem;
1136 }
1137 EXPORT_SYMBOL_GPL(hrtimer_get_remaining);
1138
1139 #ifdef CONFIG_NO_HZ
1140 /**
1141 * hrtimer_get_next_event - get the time until next expiry event
1142 *
1143 * Returns the delta to the next expiry event or KTIME_MAX if no timer
1144 * is pending.
1145 */
hrtimer_get_next_event(void)1146 ktime_t hrtimer_get_next_event(void)
1147 {
1148 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1149 struct hrtimer_clock_base *base = cpu_base->clock_base;
1150 ktime_t delta, mindelta = { .tv64 = KTIME_MAX };
1151 unsigned long flags;
1152 int i;
1153
1154 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1155
1156 if (!hrtimer_hres_active()) {
1157 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
1158 struct hrtimer *timer;
1159 struct timerqueue_node *next;
1160
1161 next = timerqueue_getnext(&base->active);
1162 if (!next)
1163 continue;
1164
1165 timer = container_of(next, struct hrtimer, node);
1166 delta.tv64 = hrtimer_get_expires_tv64(timer);
1167 delta = ktime_sub(delta, base->get_time());
1168 if (delta.tv64 < mindelta.tv64)
1169 mindelta.tv64 = delta.tv64;
1170 }
1171 }
1172
1173 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1174
1175 if (mindelta.tv64 < 0)
1176 mindelta.tv64 = 0;
1177 return mindelta;
1178 }
1179 #endif
1180
__hrtimer_init(struct hrtimer * timer,clockid_t clock_id,enum hrtimer_mode mode)1181 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1182 enum hrtimer_mode mode)
1183 {
1184 struct hrtimer_cpu_base *cpu_base;
1185 int base;
1186
1187 memset(timer, 0, sizeof(struct hrtimer));
1188
1189 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1190
1191 if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
1192 clock_id = CLOCK_MONOTONIC;
1193
1194 base = hrtimer_clockid_to_base(clock_id);
1195 timer->base = &cpu_base->clock_base[base];
1196 timerqueue_init(&timer->node);
1197
1198 #ifdef CONFIG_TIMER_STATS
1199 timer->start_site = NULL;
1200 timer->start_pid = -1;
1201 memset(timer->start_comm, 0, TASK_COMM_LEN);
1202 #endif
1203 }
1204
1205 /**
1206 * hrtimer_init - initialize a timer to the given clock
1207 * @timer: the timer to be initialized
1208 * @clock_id: the clock to be used
1209 * @mode: timer mode abs/rel
1210 */
hrtimer_init(struct hrtimer * timer,clockid_t clock_id,enum hrtimer_mode mode)1211 void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1212 enum hrtimer_mode mode)
1213 {
1214 debug_init(timer, clock_id, mode);
1215 __hrtimer_init(timer, clock_id, mode);
1216 }
1217 EXPORT_SYMBOL_GPL(hrtimer_init);
1218
1219 /**
1220 * hrtimer_get_res - get the timer resolution for a clock
1221 * @which_clock: which clock to query
1222 * @tp: pointer to timespec variable to store the resolution
1223 *
1224 * Store the resolution of the clock selected by @which_clock in the
1225 * variable pointed to by @tp.
1226 */
hrtimer_get_res(const clockid_t which_clock,struct timespec * tp)1227 int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp)
1228 {
1229 struct hrtimer_cpu_base *cpu_base;
1230 int base = hrtimer_clockid_to_base(which_clock);
1231
1232 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1233 *tp = ktime_to_timespec(cpu_base->clock_base[base].resolution);
1234
1235 return 0;
1236 }
1237 EXPORT_SYMBOL_GPL(hrtimer_get_res);
1238
__run_hrtimer(struct hrtimer * timer,ktime_t * now)1239 static void __run_hrtimer(struct hrtimer *timer, ktime_t *now)
1240 {
1241 struct hrtimer_clock_base *base = timer->base;
1242 struct hrtimer_cpu_base *cpu_base = base->cpu_base;
1243 enum hrtimer_restart (*fn)(struct hrtimer *);
1244 int restart;
1245
1246 WARN_ON(!irqs_disabled());
1247
1248 debug_deactivate(timer);
1249 __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0);
1250 timer_stats_account_hrtimer(timer);
1251 fn = timer->function;
1252
1253 /*
1254 * Because we run timers from hardirq context, there is no chance
1255 * they get migrated to another cpu, therefore its safe to unlock
1256 * the timer base.
1257 */
1258 raw_spin_unlock(&cpu_base->lock);
1259 trace_hrtimer_expire_entry(timer, now);
1260 restart = fn(timer);
1261 trace_hrtimer_expire_exit(timer);
1262 raw_spin_lock(&cpu_base->lock);
1263
1264 /*
1265 * Note: We clear the CALLBACK bit after enqueue_hrtimer and
1266 * we do not reprogramm the event hardware. Happens either in
1267 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1268 */
1269 if (restart != HRTIMER_NORESTART) {
1270 BUG_ON(timer->state != HRTIMER_STATE_CALLBACK);
1271 enqueue_hrtimer(timer, base);
1272 }
1273
1274 WARN_ON_ONCE(!(timer->state & HRTIMER_STATE_CALLBACK));
1275
1276 timer->state &= ~HRTIMER_STATE_CALLBACK;
1277 }
1278
1279 #ifdef CONFIG_HIGH_RES_TIMERS
1280
1281 /*
1282 * High resolution timer interrupt
1283 * Called with interrupts disabled
1284 */
hrtimer_interrupt(struct clock_event_device * dev)1285 void hrtimer_interrupt(struct clock_event_device *dev)
1286 {
1287 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1288 ktime_t expires_next, now, entry_time, delta;
1289 int i, retries = 0;
1290
1291 BUG_ON(!cpu_base->hres_active);
1292 cpu_base->nr_events++;
1293 dev->next_event.tv64 = KTIME_MAX;
1294
1295 raw_spin_lock(&cpu_base->lock);
1296 entry_time = now = hrtimer_update_base(cpu_base);
1297 retry:
1298 expires_next.tv64 = KTIME_MAX;
1299 /*
1300 * We set expires_next to KTIME_MAX here with cpu_base->lock
1301 * held to prevent that a timer is enqueued in our queue via
1302 * the migration code. This does not affect enqueueing of
1303 * timers which run their callback and need to be requeued on
1304 * this CPU.
1305 */
1306 cpu_base->expires_next.tv64 = KTIME_MAX;
1307
1308 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1309 struct hrtimer_clock_base *base;
1310 struct timerqueue_node *node;
1311 ktime_t basenow;
1312
1313 if (!(cpu_base->active_bases & (1 << i)))
1314 continue;
1315
1316 base = cpu_base->clock_base + i;
1317 basenow = ktime_add(now, base->offset);
1318
1319 while ((node = timerqueue_getnext(&base->active))) {
1320 struct hrtimer *timer;
1321
1322 timer = container_of(node, struct hrtimer, node);
1323
1324 /*
1325 * The immediate goal for using the softexpires is
1326 * minimizing wakeups, not running timers at the
1327 * earliest interrupt after their soft expiration.
1328 * This allows us to avoid using a Priority Search
1329 * Tree, which can answer a stabbing querry for
1330 * overlapping intervals and instead use the simple
1331 * BST we already have.
1332 * We don't add extra wakeups by delaying timers that
1333 * are right-of a not yet expired timer, because that
1334 * timer will have to trigger a wakeup anyway.
1335 */
1336
1337 if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer)) {
1338 ktime_t expires;
1339
1340 expires = ktime_sub(hrtimer_get_expires(timer),
1341 base->offset);
1342 if (expires.tv64 < 0)
1343 expires.tv64 = KTIME_MAX;
1344 if (expires.tv64 < expires_next.tv64)
1345 expires_next = expires;
1346 break;
1347 }
1348
1349 __run_hrtimer(timer, &basenow);
1350 }
1351 }
1352
1353 /*
1354 * Store the new expiry value so the migration code can verify
1355 * against it.
1356 */
1357 cpu_base->expires_next = expires_next;
1358 raw_spin_unlock(&cpu_base->lock);
1359
1360 /* Reprogramming necessary ? */
1361 if (expires_next.tv64 == KTIME_MAX ||
1362 !tick_program_event(expires_next, 0)) {
1363 cpu_base->hang_detected = 0;
1364 return;
1365 }
1366
1367 /*
1368 * The next timer was already expired due to:
1369 * - tracing
1370 * - long lasting callbacks
1371 * - being scheduled away when running in a VM
1372 *
1373 * We need to prevent that we loop forever in the hrtimer
1374 * interrupt routine. We give it 3 attempts to avoid
1375 * overreacting on some spurious event.
1376 *
1377 * Acquire base lock for updating the offsets and retrieving
1378 * the current time.
1379 */
1380 raw_spin_lock(&cpu_base->lock);
1381 now = hrtimer_update_base(cpu_base);
1382 cpu_base->nr_retries++;
1383 if (++retries < 3)
1384 goto retry;
1385 /*
1386 * Give the system a chance to do something else than looping
1387 * here. We stored the entry time, so we know exactly how long
1388 * we spent here. We schedule the next event this amount of
1389 * time away.
1390 */
1391 cpu_base->nr_hangs++;
1392 cpu_base->hang_detected = 1;
1393 raw_spin_unlock(&cpu_base->lock);
1394 delta = ktime_sub(now, entry_time);
1395 if (delta.tv64 > cpu_base->max_hang_time.tv64)
1396 cpu_base->max_hang_time = delta;
1397 /*
1398 * Limit it to a sensible value as we enforce a longer
1399 * delay. Give the CPU at least 100ms to catch up.
1400 */
1401 if (delta.tv64 > 100 * NSEC_PER_MSEC)
1402 expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1403 else
1404 expires_next = ktime_add(now, delta);
1405 tick_program_event(expires_next, 1);
1406 printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n",
1407 ktime_to_ns(delta));
1408 }
1409
1410 /*
1411 * local version of hrtimer_peek_ahead_timers() called with interrupts
1412 * disabled.
1413 */
__hrtimer_peek_ahead_timers(void)1414 static void __hrtimer_peek_ahead_timers(void)
1415 {
1416 struct tick_device *td;
1417
1418 if (!hrtimer_hres_active())
1419 return;
1420
1421 td = &__get_cpu_var(tick_cpu_device);
1422 if (td && td->evtdev)
1423 hrtimer_interrupt(td->evtdev);
1424 }
1425
1426 /**
1427 * hrtimer_peek_ahead_timers -- run soft-expired timers now
1428 *
1429 * hrtimer_peek_ahead_timers will peek at the timer queue of
1430 * the current cpu and check if there are any timers for which
1431 * the soft expires time has passed. If any such timers exist,
1432 * they are run immediately and then removed from the timer queue.
1433 *
1434 */
hrtimer_peek_ahead_timers(void)1435 void hrtimer_peek_ahead_timers(void)
1436 {
1437 unsigned long flags;
1438
1439 local_irq_save(flags);
1440 __hrtimer_peek_ahead_timers();
1441 local_irq_restore(flags);
1442 }
1443
run_hrtimer_softirq(struct softirq_action * h)1444 static void run_hrtimer_softirq(struct softirq_action *h)
1445 {
1446 hrtimer_peek_ahead_timers();
1447 }
1448
1449 #else /* CONFIG_HIGH_RES_TIMERS */
1450
__hrtimer_peek_ahead_timers(void)1451 static inline void __hrtimer_peek_ahead_timers(void) { }
1452
1453 #endif /* !CONFIG_HIGH_RES_TIMERS */
1454
1455 /*
1456 * Called from timer softirq every jiffy, expire hrtimers:
1457 *
1458 * For HRT its the fall back code to run the softirq in the timer
1459 * softirq context in case the hrtimer initialization failed or has
1460 * not been done yet.
1461 */
hrtimer_run_pending(void)1462 void hrtimer_run_pending(void)
1463 {
1464 if (hrtimer_hres_active())
1465 return;
1466
1467 /*
1468 * This _is_ ugly: We have to check in the softirq context,
1469 * whether we can switch to highres and / or nohz mode. The
1470 * clocksource switch happens in the timer interrupt with
1471 * xtime_lock held. Notification from there only sets the
1472 * check bit in the tick_oneshot code, otherwise we might
1473 * deadlock vs. xtime_lock.
1474 */
1475 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1476 hrtimer_switch_to_hres();
1477 }
1478
1479 /*
1480 * Called from hardirq context every jiffy
1481 */
hrtimer_run_queues(void)1482 void hrtimer_run_queues(void)
1483 {
1484 struct timerqueue_node *node;
1485 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1486 struct hrtimer_clock_base *base;
1487 int index, gettime = 1;
1488
1489 if (hrtimer_hres_active())
1490 return;
1491
1492 for (index = 0; index < HRTIMER_MAX_CLOCK_BASES; index++) {
1493 base = &cpu_base->clock_base[index];
1494 if (!timerqueue_getnext(&base->active))
1495 continue;
1496
1497 if (gettime) {
1498 hrtimer_get_softirq_time(cpu_base);
1499 gettime = 0;
1500 }
1501
1502 raw_spin_lock(&cpu_base->lock);
1503
1504 while ((node = timerqueue_getnext(&base->active))) {
1505 struct hrtimer *timer;
1506
1507 timer = container_of(node, struct hrtimer, node);
1508 if (base->softirq_time.tv64 <=
1509 hrtimer_get_expires_tv64(timer))
1510 break;
1511
1512 __run_hrtimer(timer, &base->softirq_time);
1513 }
1514 raw_spin_unlock(&cpu_base->lock);
1515 }
1516 }
1517
1518 /*
1519 * Sleep related functions:
1520 */
hrtimer_wakeup(struct hrtimer * timer)1521 static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1522 {
1523 struct hrtimer_sleeper *t =
1524 container_of(timer, struct hrtimer_sleeper, timer);
1525 struct task_struct *task = t->task;
1526
1527 t->task = NULL;
1528 if (task)
1529 wake_up_process(task);
1530
1531 return HRTIMER_NORESTART;
1532 }
1533
hrtimer_init_sleeper(struct hrtimer_sleeper * sl,struct task_struct * task)1534 void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1535 {
1536 sl->timer.function = hrtimer_wakeup;
1537 sl->task = task;
1538 }
1539 EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
1540
do_nanosleep(struct hrtimer_sleeper * t,enum hrtimer_mode mode)1541 static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1542 {
1543 hrtimer_init_sleeper(t, current);
1544
1545 do {
1546 set_current_state(TASK_INTERRUPTIBLE);
1547 hrtimer_start_expires(&t->timer, mode);
1548 if (!hrtimer_active(&t->timer))
1549 t->task = NULL;
1550
1551 if (likely(t->task))
1552 schedule();
1553
1554 hrtimer_cancel(&t->timer);
1555 mode = HRTIMER_MODE_ABS;
1556
1557 } while (t->task && !signal_pending(current));
1558
1559 __set_current_state(TASK_RUNNING);
1560
1561 return t->task == NULL;
1562 }
1563
update_rmtp(struct hrtimer * timer,struct timespec __user * rmtp)1564 static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp)
1565 {
1566 struct timespec rmt;
1567 ktime_t rem;
1568
1569 rem = hrtimer_expires_remaining(timer);
1570 if (rem.tv64 <= 0)
1571 return 0;
1572 rmt = ktime_to_timespec(rem);
1573
1574 if (copy_to_user(rmtp, &rmt, sizeof(*rmtp)))
1575 return -EFAULT;
1576
1577 return 1;
1578 }
1579
hrtimer_nanosleep_restart(struct restart_block * restart)1580 long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1581 {
1582 struct hrtimer_sleeper t;
1583 struct timespec __user *rmtp;
1584 int ret = 0;
1585
1586 hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid,
1587 HRTIMER_MODE_ABS);
1588 hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
1589
1590 if (do_nanosleep(&t, HRTIMER_MODE_ABS))
1591 goto out;
1592
1593 rmtp = restart->nanosleep.rmtp;
1594 if (rmtp) {
1595 ret = update_rmtp(&t.timer, rmtp);
1596 if (ret <= 0)
1597 goto out;
1598 }
1599
1600 /* The other values in restart are already filled in */
1601 ret = -ERESTART_RESTARTBLOCK;
1602 out:
1603 destroy_hrtimer_on_stack(&t.timer);
1604 return ret;
1605 }
1606
hrtimer_nanosleep(struct timespec * rqtp,struct timespec __user * rmtp,const enum hrtimer_mode mode,const clockid_t clockid)1607 long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
1608 const enum hrtimer_mode mode, const clockid_t clockid)
1609 {
1610 struct restart_block *restart;
1611 struct hrtimer_sleeper t;
1612 int ret = 0;
1613 unsigned long slack;
1614
1615 slack = current->timer_slack_ns;
1616 if (rt_task(current))
1617 slack = 0;
1618
1619 hrtimer_init_on_stack(&t.timer, clockid, mode);
1620 hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack);
1621 if (do_nanosleep(&t, mode))
1622 goto out;
1623
1624 /* Absolute timers do not update the rmtp value and restart: */
1625 if (mode == HRTIMER_MODE_ABS) {
1626 ret = -ERESTARTNOHAND;
1627 goto out;
1628 }
1629
1630 if (rmtp) {
1631 ret = update_rmtp(&t.timer, rmtp);
1632 if (ret <= 0)
1633 goto out;
1634 }
1635
1636 restart = ¤t_thread_info()->restart_block;
1637 restart->fn = hrtimer_nanosleep_restart;
1638 restart->nanosleep.clockid = t.timer.base->clockid;
1639 restart->nanosleep.rmtp = rmtp;
1640 restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
1641
1642 ret = -ERESTART_RESTARTBLOCK;
1643 out:
1644 destroy_hrtimer_on_stack(&t.timer);
1645 return ret;
1646 }
1647
SYSCALL_DEFINE2(nanosleep,struct timespec __user *,rqtp,struct timespec __user *,rmtp)1648 SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp,
1649 struct timespec __user *, rmtp)
1650 {
1651 struct timespec tu;
1652
1653 if (copy_from_user(&tu, rqtp, sizeof(tu)))
1654 return -EFAULT;
1655
1656 if (!timespec_valid(&tu))
1657 return -EINVAL;
1658
1659 return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1660 }
1661
1662 /*
1663 * Functions related to boot-time initialization:
1664 */
init_hrtimers_cpu(int cpu)1665 static void __cpuinit init_hrtimers_cpu(int cpu)
1666 {
1667 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1668 int i;
1669
1670 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1671 cpu_base->clock_base[i].cpu_base = cpu_base;
1672 timerqueue_init_head(&cpu_base->clock_base[i].active);
1673 }
1674
1675 hrtimer_init_hres(cpu_base);
1676 }
1677
1678 #ifdef CONFIG_HOTPLUG_CPU
1679
migrate_hrtimer_list(struct hrtimer_clock_base * old_base,struct hrtimer_clock_base * new_base)1680 static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1681 struct hrtimer_clock_base *new_base)
1682 {
1683 struct hrtimer *timer;
1684 struct timerqueue_node *node;
1685
1686 while ((node = timerqueue_getnext(&old_base->active))) {
1687 timer = container_of(node, struct hrtimer, node);
1688 BUG_ON(hrtimer_callback_running(timer));
1689 debug_deactivate(timer);
1690
1691 /*
1692 * Mark it as STATE_MIGRATE not INACTIVE otherwise the
1693 * timer could be seen as !active and just vanish away
1694 * under us on another CPU
1695 */
1696 __remove_hrtimer(timer, old_base, HRTIMER_STATE_MIGRATE, 0);
1697 timer->base = new_base;
1698 /*
1699 * Enqueue the timers on the new cpu. This does not
1700 * reprogram the event device in case the timer
1701 * expires before the earliest on this CPU, but we run
1702 * hrtimer_interrupt after we migrated everything to
1703 * sort out already expired timers and reprogram the
1704 * event device.
1705 */
1706 enqueue_hrtimer(timer, new_base);
1707
1708 /* Clear the migration state bit */
1709 timer->state &= ~HRTIMER_STATE_MIGRATE;
1710 }
1711 }
1712
migrate_hrtimers(int scpu)1713 static void migrate_hrtimers(int scpu)
1714 {
1715 struct hrtimer_cpu_base *old_base, *new_base;
1716 int i;
1717
1718 BUG_ON(cpu_online(scpu));
1719 tick_cancel_sched_timer(scpu);
1720
1721 local_irq_disable();
1722 old_base = &per_cpu(hrtimer_bases, scpu);
1723 new_base = &__get_cpu_var(hrtimer_bases);
1724 /*
1725 * The caller is globally serialized and nobody else
1726 * takes two locks at once, deadlock is not possible.
1727 */
1728 raw_spin_lock(&new_base->lock);
1729 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1730
1731 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1732 migrate_hrtimer_list(&old_base->clock_base[i],
1733 &new_base->clock_base[i]);
1734 }
1735
1736 raw_spin_unlock(&old_base->lock);
1737 raw_spin_unlock(&new_base->lock);
1738
1739 /* Check, if we got expired work to do */
1740 __hrtimer_peek_ahead_timers();
1741 local_irq_enable();
1742 }
1743
1744 #endif /* CONFIG_HOTPLUG_CPU */
1745
hrtimer_cpu_notify(struct notifier_block * self,unsigned long action,void * hcpu)1746 static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self,
1747 unsigned long action, void *hcpu)
1748 {
1749 int scpu = (long)hcpu;
1750
1751 switch (action) {
1752
1753 case CPU_UP_PREPARE:
1754 case CPU_UP_PREPARE_FROZEN:
1755 init_hrtimers_cpu(scpu);
1756 break;
1757
1758 #ifdef CONFIG_HOTPLUG_CPU
1759 case CPU_DYING:
1760 case CPU_DYING_FROZEN:
1761 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DYING, &scpu);
1762 break;
1763 case CPU_DEAD:
1764 case CPU_DEAD_FROZEN:
1765 {
1766 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &scpu);
1767 migrate_hrtimers(scpu);
1768 break;
1769 }
1770 #endif
1771
1772 default:
1773 break;
1774 }
1775
1776 return NOTIFY_OK;
1777 }
1778
1779 static struct notifier_block __cpuinitdata hrtimers_nb = {
1780 .notifier_call = hrtimer_cpu_notify,
1781 };
1782
hrtimers_init(void)1783 void __init hrtimers_init(void)
1784 {
1785 hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
1786 (void *)(long)smp_processor_id());
1787 register_cpu_notifier(&hrtimers_nb);
1788 #ifdef CONFIG_HIGH_RES_TIMERS
1789 open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq);
1790 #endif
1791 }
1792
1793 /**
1794 * schedule_hrtimeout_range_clock - sleep until timeout
1795 * @expires: timeout value (ktime_t)
1796 * @delta: slack in expires timeout (ktime_t)
1797 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1798 * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME
1799 */
1800 int __sched
schedule_hrtimeout_range_clock(ktime_t * expires,unsigned long delta,const enum hrtimer_mode mode,int clock)1801 schedule_hrtimeout_range_clock(ktime_t *expires, unsigned long delta,
1802 const enum hrtimer_mode mode, int clock)
1803 {
1804 struct hrtimer_sleeper t;
1805
1806 /*
1807 * Optimize when a zero timeout value is given. It does not
1808 * matter whether this is an absolute or a relative time.
1809 */
1810 if (expires && !expires->tv64) {
1811 __set_current_state(TASK_RUNNING);
1812 return 0;
1813 }
1814
1815 /*
1816 * A NULL parameter means "infinite"
1817 */
1818 if (!expires) {
1819 schedule();
1820 __set_current_state(TASK_RUNNING);
1821 return -EINTR;
1822 }
1823
1824 hrtimer_init_on_stack(&t.timer, clock, mode);
1825 hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
1826
1827 hrtimer_init_sleeper(&t, current);
1828
1829 hrtimer_start_expires(&t.timer, mode);
1830 if (!hrtimer_active(&t.timer))
1831 t.task = NULL;
1832
1833 if (likely(t.task))
1834 schedule();
1835
1836 hrtimer_cancel(&t.timer);
1837 destroy_hrtimer_on_stack(&t.timer);
1838
1839 __set_current_state(TASK_RUNNING);
1840
1841 return !t.task ? 0 : -EINTR;
1842 }
1843
1844 /**
1845 * schedule_hrtimeout_range - sleep until timeout
1846 * @expires: timeout value (ktime_t)
1847 * @delta: slack in expires timeout (ktime_t)
1848 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1849 *
1850 * Make the current task sleep until the given expiry time has
1851 * elapsed. The routine will return immediately unless
1852 * the current task state has been set (see set_current_state()).
1853 *
1854 * The @delta argument gives the kernel the freedom to schedule the
1855 * actual wakeup to a time that is both power and performance friendly.
1856 * The kernel give the normal best effort behavior for "@expires+@delta",
1857 * but may decide to fire the timer earlier, but no earlier than @expires.
1858 *
1859 * You can set the task state as follows -
1860 *
1861 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1862 * pass before the routine returns.
1863 *
1864 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1865 * delivered to the current task.
1866 *
1867 * The current task state is guaranteed to be TASK_RUNNING when this
1868 * routine returns.
1869 *
1870 * Returns 0 when the timer has expired otherwise -EINTR
1871 */
schedule_hrtimeout_range(ktime_t * expires,unsigned long delta,const enum hrtimer_mode mode)1872 int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta,
1873 const enum hrtimer_mode mode)
1874 {
1875 return schedule_hrtimeout_range_clock(expires, delta, mode,
1876 CLOCK_MONOTONIC);
1877 }
1878 EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
1879
1880 /**
1881 * schedule_hrtimeout - sleep until timeout
1882 * @expires: timeout value (ktime_t)
1883 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1884 *
1885 * Make the current task sleep until the given expiry time has
1886 * elapsed. The routine will return immediately unless
1887 * the current task state has been set (see set_current_state()).
1888 *
1889 * You can set the task state as follows -
1890 *
1891 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1892 * pass before the routine returns.
1893 *
1894 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1895 * delivered to the current task.
1896 *
1897 * The current task state is guaranteed to be TASK_RUNNING when this
1898 * routine returns.
1899 *
1900 * Returns 0 when the timer has expired otherwise -EINTR
1901 */
schedule_hrtimeout(ktime_t * expires,const enum hrtimer_mode mode)1902 int __sched schedule_hrtimeout(ktime_t *expires,
1903 const enum hrtimer_mode mode)
1904 {
1905 return schedule_hrtimeout_range(expires, 0, mode);
1906 }
1907 EXPORT_SYMBOL_GPL(schedule_hrtimeout);
1908