blob: 23c03f43e1962d6f0dfa4cb8a716080cf671a48c [file] [log] [blame]
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001/*
2 * linux/kernel/hrtimer.c
3 *
Thomas Gleixner3c8aa392007-02-16 01:27:50 -08004 * Copyright(C) 2005-2006, Thomas Gleixner <tglx@linutronix.de>
Thomas Gleixner79bf2bb2007-02-16 01:28:03 -08005 * Copyright(C) 2005-2007, Red Hat, Inc., Ingo Molnar
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -08006 * Copyright(C) 2006-2007 Timesys Corp., Thomas Gleixner
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08007 *
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 *
Thomas Gleixner66188fa2006-02-01 03:05:13 -080025 * Help, testing, suggestions, bugfixes, improvements were
26 * provided by:
27 *
28 * George Anzinger, Andrew Morton, Steven Rostedt, Roman Zippel
29 * et. al.
30 *
Thomas Gleixnerc0a31322006-01-09 20:52:32 -080031 * For licencing details see kernel-base/COPYING
32 */
33
34#include <linux/cpu.h>
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -080035#include <linux/irq.h>
Thomas Gleixnerc0a31322006-01-09 20:52:32 -080036#include <linux/module.h>
37#include <linux/percpu.h>
38#include <linux/hrtimer.h>
39#include <linux/notifier.h>
40#include <linux/syscalls.h>
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -080041#include <linux/kallsyms.h>
Thomas Gleixnerc0a31322006-01-09 20:52:32 -080042#include <linux/interrupt.h>
Thomas Gleixner79bf2bb2007-02-16 01:28:03 -080043#include <linux/tick.h>
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -080044#include <linux/seq_file.h>
45#include <linux/err.h>
Thomas Gleixnerc0a31322006-01-09 20:52:32 -080046
47#include <asm/uaccess.h>
48
49/**
50 * ktime_get - get the monotonic time in ktime_t format
51 *
52 * returns the time in ktime_t format
53 */
Thomas Gleixnerd316c572007-02-16 01:28:00 -080054ktime_t ktime_get(void)
Thomas Gleixnerc0a31322006-01-09 20:52:32 -080055{
56 struct timespec now;
57
58 ktime_get_ts(&now);
59
60 return timespec_to_ktime(now);
61}
Patrick McHardy641b9e02007-03-16 01:18:42 -070062EXPORT_SYMBOL_GPL(ktime_get);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -080063
64/**
65 * ktime_get_real - get the real (wall-) time in ktime_t format
66 *
67 * returns the time in ktime_t format
68 */
Thomas Gleixnerd316c572007-02-16 01:28:00 -080069ktime_t ktime_get_real(void)
Thomas Gleixnerc0a31322006-01-09 20:52:32 -080070{
71 struct timespec now;
72
73 getnstimeofday(&now);
74
75 return timespec_to_ktime(now);
76}
77
78EXPORT_SYMBOL_GPL(ktime_get_real);
79
80/*
81 * The timer bases:
George Anzinger79786722006-02-01 03:05:11 -080082 *
83 * Note: If we want to add new timer bases, we have to skip the two
84 * clock ids captured by the cpu-timers. We do this by holding empty
85 * entries rather than doing math adjustment of the clock ids.
86 * This ensures that we capture erroneous accesses to these clock ids
87 * rather than moving them into the range of valid clock id's.
Thomas Gleixnerc0a31322006-01-09 20:52:32 -080088 */
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -080089DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
Thomas Gleixnerc0a31322006-01-09 20:52:32 -080090{
Thomas Gleixner3c8aa392007-02-16 01:27:50 -080091
92 .clock_base =
Thomas Gleixnerc0a31322006-01-09 20:52:32 -080093 {
Thomas Gleixner3c8aa392007-02-16 01:27:50 -080094 {
95 .index = CLOCK_REALTIME,
96 .get_time = &ktime_get_real,
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -080097 .resolution = KTIME_LOW_RES,
Thomas Gleixner3c8aa392007-02-16 01:27:50 -080098 },
99 {
100 .index = CLOCK_MONOTONIC,
101 .get_time = &ktime_get,
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800102 .resolution = KTIME_LOW_RES,
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800103 },
104 }
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800105};
106
107/**
108 * ktime_get_ts - get the monotonic clock in timespec format
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800109 * @ts: pointer to timespec variable
110 *
111 * The function calculates the monotonic clock from the realtime
112 * clock and the wall_to_monotonic offset and stores the result
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800113 * in normalized timespec format in the variable pointed to by @ts.
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800114 */
115void ktime_get_ts(struct timespec *ts)
116{
117 struct timespec tomono;
118 unsigned long seq;
119
120 do {
121 seq = read_seqbegin(&xtime_lock);
122 getnstimeofday(ts);
123 tomono = wall_to_monotonic;
124
125 } while (read_seqretry(&xtime_lock, seq));
126
127 set_normalized_timespec(ts, ts->tv_sec + tomono.tv_sec,
128 ts->tv_nsec + tomono.tv_nsec);
129}
Matt Helsley69778e32006-01-09 20:52:39 -0800130EXPORT_SYMBOL_GPL(ktime_get_ts);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800131
132/*
Thomas Gleixner92127c72006-03-26 01:38:05 -0800133 * Get the coarse grained time at the softirq based on xtime and
134 * wall_to_monotonic.
135 */
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800136static void hrtimer_get_softirq_time(struct hrtimer_cpu_base *base)
Thomas Gleixner92127c72006-03-26 01:38:05 -0800137{
138 ktime_t xtim, tomono;
Thomas Gleixnerad28d942007-03-16 13:38:21 -0800139 struct timespec xts, tom;
Thomas Gleixner92127c72006-03-26 01:38:05 -0800140 unsigned long seq;
141
142 do {
143 seq = read_seqbegin(&xtime_lock);
john stultzf4304ab2007-02-16 01:27:26 -0800144#ifdef CONFIG_NO_HZ
145 getnstimeofday(&xts);
146#else
147 xts = xtime;
148#endif
Thomas Gleixnerad28d942007-03-16 13:38:21 -0800149 tom = wall_to_monotonic;
Thomas Gleixner92127c72006-03-26 01:38:05 -0800150 } while (read_seqretry(&xtime_lock, seq));
151
john stultzf4304ab2007-02-16 01:27:26 -0800152 xtim = timespec_to_ktime(xts);
Thomas Gleixnerad28d942007-03-16 13:38:21 -0800153 tomono = timespec_to_ktime(tom);
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800154 base->clock_base[CLOCK_REALTIME].softirq_time = xtim;
155 base->clock_base[CLOCK_MONOTONIC].softirq_time =
156 ktime_add(xtim, tomono);
Thomas Gleixner92127c72006-03-26 01:38:05 -0800157}
158
159/*
Thomas Gleixner303e9672007-02-16 01:27:51 -0800160 * Helper function to check, whether the timer is running the callback
161 * function
162 */
163static inline int hrtimer_callback_running(struct hrtimer *timer)
164{
165 return timer->state & HRTIMER_STATE_CALLBACK;
166}
167
168/*
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800169 * Functions and macros which are different for UP/SMP systems are kept in a
170 * single place
171 */
172#ifdef CONFIG_SMP
173
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800174/*
175 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
176 * means that all timers which are tied to this base via timer->base are
177 * locked, and the base itself is locked too.
178 *
179 * So __run_timers/migrate_timers can safely modify all timers which could
180 * be found on the lists/queues.
181 *
182 * When the timer's base is locked, and the timer removed from list, it is
183 * possible to set timer->base = NULL and drop the lock: the timer remains
184 * locked.
185 */
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800186static
187struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
188 unsigned long *flags)
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800189{
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800190 struct hrtimer_clock_base *base;
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800191
192 for (;;) {
193 base = timer->base;
194 if (likely(base != NULL)) {
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800195 spin_lock_irqsave(&base->cpu_base->lock, *flags);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800196 if (likely(base == timer->base))
197 return base;
198 /* The timer has migrated to another CPU: */
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800199 spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800200 }
201 cpu_relax();
202 }
203}
204
205/*
206 * Switch the timer base to the current CPU when possible.
207 */
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800208static inline struct hrtimer_clock_base *
209switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base)
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800210{
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800211 struct hrtimer_clock_base *new_base;
212 struct hrtimer_cpu_base *new_cpu_base;
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800213
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800214 new_cpu_base = &__get_cpu_var(hrtimer_bases);
215 new_base = &new_cpu_base->clock_base[base->index];
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800216
217 if (base != new_base) {
218 /*
219 * We are trying to schedule the timer on the local CPU.
220 * However we can't change timer's base while it is running,
221 * so we keep it on the same CPU. No hassle vs. reprogramming
222 * the event source in the high resolution case. The softirq
223 * code will take care of this when the timer function has
224 * completed. There is no conflict as we hold the lock until
225 * the timer is enqueued.
226 */
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800227 if (unlikely(hrtimer_callback_running(timer)))
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800228 return base;
229
230 /* See the comment in lock_timer_base() */
231 timer->base = NULL;
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800232 spin_unlock(&base->cpu_base->lock);
233 spin_lock(&new_base->cpu_base->lock);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800234 timer->base = new_base;
235 }
236 return new_base;
237}
238
239#else /* CONFIG_SMP */
240
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800241static inline struct hrtimer_clock_base *
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800242lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
243{
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800244 struct hrtimer_clock_base *base = timer->base;
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800245
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800246 spin_lock_irqsave(&base->cpu_base->lock, *flags);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800247
248 return base;
249}
250
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800251# define switch_hrtimer_base(t, b) (b)
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800252
253#endif /* !CONFIG_SMP */
254
255/*
256 * Functions for the union type storage format of ktime_t which are
257 * too large for inlining:
258 */
259#if BITS_PER_LONG < 64
260# ifndef CONFIG_KTIME_SCALAR
261/**
262 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800263 * @kt: addend
264 * @nsec: the scalar nsec value to add
265 *
266 * Returns the sum of kt and nsec in ktime_t format
267 */
268ktime_t ktime_add_ns(const ktime_t kt, u64 nsec)
269{
270 ktime_t tmp;
271
272 if (likely(nsec < NSEC_PER_SEC)) {
273 tmp.tv64 = nsec;
274 } else {
275 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
276
277 tmp = ktime_set((long)nsec, rem);
278 }
279
280 return ktime_add(kt, tmp);
281}
David Howellsb8b8fd22007-04-27 15:31:24 -0700282
283EXPORT_SYMBOL_GPL(ktime_add_ns);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800284# endif /* !CONFIG_KTIME_SCALAR */
285
286/*
287 * Divide a ktime value by a nanosecond value
288 */
Thomas Gleixner79bf2bb2007-02-16 01:28:03 -0800289unsigned long ktime_divns(const ktime_t kt, s64 div)
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800290{
291 u64 dclc, inc, dns;
292 int sft = 0;
293
294 dclc = dns = ktime_to_ns(kt);
295 inc = div;
296 /* Make sure the divisor is less than 2^32: */
297 while (div >> 32) {
298 sft++;
299 div >>= 1;
300 }
301 dclc >>= sft;
302 do_div(dclc, (unsigned long) div);
303
304 return (unsigned long) dclc;
305}
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800306#endif /* BITS_PER_LONG >= 64 */
307
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800308/* High resolution timer related functions */
309#ifdef CONFIG_HIGH_RES_TIMERS
310
311/*
312 * High resolution timer enabled ?
313 */
314static int hrtimer_hres_enabled __read_mostly = 1;
315
316/*
317 * Enable / Disable high resolution mode
318 */
319static int __init setup_hrtimer_hres(char *str)
320{
321 if (!strcmp(str, "off"))
322 hrtimer_hres_enabled = 0;
323 else if (!strcmp(str, "on"))
324 hrtimer_hres_enabled = 1;
325 else
326 return 0;
327 return 1;
328}
329
330__setup("highres=", setup_hrtimer_hres);
331
332/*
333 * hrtimer_high_res_enabled - query, if the highres mode is enabled
334 */
335static inline int hrtimer_is_hres_enabled(void)
336{
337 return hrtimer_hres_enabled;
338}
339
340/*
341 * Is the high resolution mode active ?
342 */
343static inline int hrtimer_hres_active(void)
344{
345 return __get_cpu_var(hrtimer_bases).hres_active;
346}
347
348/*
349 * Reprogram the event source with checking both queues for the
350 * next event
351 * Called with interrupts disabled and base->lock held
352 */
353static void hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base)
354{
355 int i;
356 struct hrtimer_clock_base *base = cpu_base->clock_base;
357 ktime_t expires;
358
359 cpu_base->expires_next.tv64 = KTIME_MAX;
360
361 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
362 struct hrtimer *timer;
363
364 if (!base->first)
365 continue;
366 timer = rb_entry(base->first, struct hrtimer, node);
367 expires = ktime_sub(timer->expires, base->offset);
368 if (expires.tv64 < cpu_base->expires_next.tv64)
369 cpu_base->expires_next = expires;
370 }
371
372 if (cpu_base->expires_next.tv64 != KTIME_MAX)
373 tick_program_event(cpu_base->expires_next, 1);
374}
375
376/*
377 * Shared reprogramming for clock_realtime and clock_monotonic
378 *
379 * When a timer is enqueued and expires earlier than the already enqueued
380 * timers, we have to check, whether it expires earlier than the timer for
381 * which the clock event device was armed.
382 *
383 * Called with interrupts disabled and base->cpu_base.lock held
384 */
385static int hrtimer_reprogram(struct hrtimer *timer,
386 struct hrtimer_clock_base *base)
387{
388 ktime_t *expires_next = &__get_cpu_var(hrtimer_bases).expires_next;
389 ktime_t expires = ktime_sub(timer->expires, base->offset);
390 int res;
391
392 /*
393 * When the callback is running, we do not reprogram the clock event
394 * device. The timer callback is either running on a different CPU or
395 * the callback is executed in the hrtimer_interupt context. The
396 * reprogramming is handled either by the softirq, which called the
397 * callback or at the end of the hrtimer_interrupt.
398 */
399 if (hrtimer_callback_running(timer))
400 return 0;
401
402 if (expires.tv64 >= expires_next->tv64)
403 return 0;
404
405 /*
406 * Clockevents returns -ETIME, when the event was in the past.
407 */
408 res = tick_program_event(expires, 0);
409 if (!IS_ERR_VALUE(res))
410 *expires_next = expires;
411 return res;
412}
413
414
415/*
416 * Retrigger next event is called after clock was set
417 *
418 * Called with interrupts disabled via on_each_cpu()
419 */
420static void retrigger_next_event(void *arg)
421{
422 struct hrtimer_cpu_base *base;
423 struct timespec realtime_offset;
424 unsigned long seq;
425
426 if (!hrtimer_hres_active())
427 return;
428
429 do {
430 seq = read_seqbegin(&xtime_lock);
431 set_normalized_timespec(&realtime_offset,
432 -wall_to_monotonic.tv_sec,
433 -wall_to_monotonic.tv_nsec);
434 } while (read_seqretry(&xtime_lock, seq));
435
436 base = &__get_cpu_var(hrtimer_bases);
437
438 /* Adjust CLOCK_REALTIME offset */
439 spin_lock(&base->lock);
440 base->clock_base[CLOCK_REALTIME].offset =
441 timespec_to_ktime(realtime_offset);
442
443 hrtimer_force_reprogram(base);
444 spin_unlock(&base->lock);
445}
446
447/*
448 * Clock realtime was set
449 *
450 * Change the offset of the realtime clock vs. the monotonic
451 * clock.
452 *
453 * We might have to reprogram the high resolution timer interrupt. On
454 * SMP we call the architecture specific code to retrigger _all_ high
455 * resolution timer interrupts. On UP we just disable interrupts and
456 * call the high resolution interrupt code.
457 */
458void clock_was_set(void)
459{
460 /* Retrigger the CPU local events everywhere */
461 on_each_cpu(retrigger_next_event, NULL, 0, 1);
462}
463
464/*
Ingo Molnar995f0542007-04-07 12:05:00 +0200465 * During resume we might have to reprogram the high resolution timer
466 * interrupt (on the local CPU):
467 */
468void hres_timers_resume(void)
469{
470 WARN_ON_ONCE(num_online_cpus() > 1);
471
472 /* Retrigger the CPU local events: */
473 retrigger_next_event(NULL);
474}
475
476/*
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800477 * Check, whether the timer is on the callback pending list
478 */
479static inline int hrtimer_cb_pending(const struct hrtimer *timer)
480{
481 return timer->state & HRTIMER_STATE_PENDING;
482}
483
484/*
485 * Remove a timer from the callback pending list
486 */
487static inline void hrtimer_remove_cb_pending(struct hrtimer *timer)
488{
489 list_del_init(&timer->cb_entry);
490}
491
492/*
493 * Initialize the high resolution related parts of cpu_base
494 */
495static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
496{
497 base->expires_next.tv64 = KTIME_MAX;
498 base->hres_active = 0;
499 INIT_LIST_HEAD(&base->cb_pending);
500}
501
502/*
503 * Initialize the high resolution related parts of a hrtimer
504 */
505static inline void hrtimer_init_timer_hres(struct hrtimer *timer)
506{
507 INIT_LIST_HEAD(&timer->cb_entry);
508}
509
510/*
511 * When High resolution timers are active, try to reprogram. Note, that in case
512 * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry
513 * check happens. The timer gets enqueued into the rbtree. The reprogramming
514 * and expiry check is done in the hrtimer_interrupt or in the softirq.
515 */
516static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
517 struct hrtimer_clock_base *base)
518{
519 if (base->cpu_base->hres_active && hrtimer_reprogram(timer, base)) {
520
521 /* Timer is expired, act upon the callback mode */
522 switch(timer->cb_mode) {
523 case HRTIMER_CB_IRQSAFE_NO_RESTART:
524 /*
525 * We can call the callback from here. No restart
526 * happens, so no danger of recursion
527 */
528 BUG_ON(timer->function(timer) != HRTIMER_NORESTART);
529 return 1;
530 case HRTIMER_CB_IRQSAFE_NO_SOFTIRQ:
531 /*
532 * This is solely for the sched tick emulation with
533 * dynamic tick support to ensure that we do not
534 * restart the tick right on the edge and end up with
535 * the tick timer in the softirq ! The calling site
536 * takes care of this.
537 */
538 return 1;
539 case HRTIMER_CB_IRQSAFE:
540 case HRTIMER_CB_SOFTIRQ:
541 /*
542 * Move everything else into the softirq pending list !
543 */
544 list_add_tail(&timer->cb_entry,
545 &base->cpu_base->cb_pending);
546 timer->state = HRTIMER_STATE_PENDING;
547 raise_softirq(HRTIMER_SOFTIRQ);
548 return 1;
549 default:
550 BUG();
551 }
552 }
553 return 0;
554}
555
556/*
557 * Switch to high resolution mode
558 */
Thomas Gleixnerf8953852007-03-06 01:42:08 -0800559static int hrtimer_switch_to_hres(void)
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800560{
561 struct hrtimer_cpu_base *base = &__get_cpu_var(hrtimer_bases);
562 unsigned long flags;
563
564 if (base->hres_active)
Thomas Gleixnerf8953852007-03-06 01:42:08 -0800565 return 1;
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800566
567 local_irq_save(flags);
568
569 if (tick_init_highres()) {
570 local_irq_restore(flags);
Thomas Gleixnerf8953852007-03-06 01:42:08 -0800571 return 0;
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800572 }
573 base->hres_active = 1;
574 base->clock_base[CLOCK_REALTIME].resolution = KTIME_HIGH_RES;
575 base->clock_base[CLOCK_MONOTONIC].resolution = KTIME_HIGH_RES;
576
577 tick_setup_sched_timer();
578
579 /* "Retrigger" the interrupt to get things going */
580 retrigger_next_event(NULL);
581 local_irq_restore(flags);
582 printk(KERN_INFO "Switched to high resolution mode on CPU %d\n",
583 smp_processor_id());
Thomas Gleixnerf8953852007-03-06 01:42:08 -0800584 return 1;
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800585}
586
587#else
588
589static inline int hrtimer_hres_active(void) { return 0; }
590static inline int hrtimer_is_hres_enabled(void) { return 0; }
Thomas Gleixnerf8953852007-03-06 01:42:08 -0800591static inline int hrtimer_switch_to_hres(void) { return 0; }
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800592static inline void hrtimer_force_reprogram(struct hrtimer_cpu_base *base) { }
593static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
594 struct hrtimer_clock_base *base)
595{
596 return 0;
597}
598static inline int hrtimer_cb_pending(struct hrtimer *timer) { return 0; }
599static inline void hrtimer_remove_cb_pending(struct hrtimer *timer) { }
600static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { }
601static inline void hrtimer_init_timer_hres(struct hrtimer *timer) { }
602
603#endif /* CONFIG_HIGH_RES_TIMERS */
604
Ingo Molnar82f67cd2007-02-16 01:28:13 -0800605#ifdef CONFIG_TIMER_STATS
606void __timer_stats_hrtimer_set_start_info(struct hrtimer *timer, void *addr)
607{
608 if (timer->start_site)
609 return;
610
611 timer->start_site = addr;
612 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
613 timer->start_pid = current->pid;
614}
615#endif
616
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800617/*
618 * Counterpart to lock_timer_base above:
619 */
620static inline
621void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
622{
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800623 spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800624}
625
626/**
627 * hrtimer_forward - forward the timer expiry
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800628 * @timer: hrtimer to forward
Roman Zippel44f21472006-03-26 01:38:06 -0800629 * @now: forward past this time
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800630 * @interval: the interval to forward
631 *
632 * Forward the timer expiry so it will expire in the future.
Jonathan Corbet8dca6f32006-01-16 15:58:55 -0700633 * Returns the number of overruns.
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800634 */
635unsigned long
Roman Zippel44f21472006-03-26 01:38:06 -0800636hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800637{
638 unsigned long orun = 1;
Roman Zippel44f21472006-03-26 01:38:06 -0800639 ktime_t delta;
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800640
641 delta = ktime_sub(now, timer->expires);
642
643 if (delta.tv64 < 0)
644 return 0;
645
Thomas Gleixnerc9db4fa2006-01-12 11:47:34 +0100646 if (interval.tv64 < timer->base->resolution.tv64)
647 interval.tv64 = timer->base->resolution.tv64;
648
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800649 if (unlikely(delta.tv64 >= interval.tv64)) {
Roman Zippeldf869b62006-03-26 01:38:11 -0800650 s64 incr = ktime_to_ns(interval);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800651
652 orun = ktime_divns(delta, incr);
653 timer->expires = ktime_add_ns(timer->expires, incr * orun);
654 if (timer->expires.tv64 > now.tv64)
655 return orun;
656 /*
657 * This (and the ktime_add() below) is the
658 * correction for exact:
659 */
660 orun++;
661 }
662 timer->expires = ktime_add(timer->expires, interval);
Thomas Gleixner13788cc2007-03-16 13:38:20 -0800663 /*
664 * Make sure, that the result did not wrap with a very large
665 * interval.
666 */
667 if (timer->expires.tv64 < 0)
668 timer->expires = ktime_set(KTIME_SEC_MAX, 0);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800669
670 return orun;
671}
Stas Sergeev6bdb6b62007-05-08 00:31:58 -0700672EXPORT_SYMBOL_GPL(hrtimer_forward);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800673
674/*
675 * enqueue_hrtimer - internal function to (re)start a timer
676 *
677 * The timer is inserted in expiry order. Insertion into the
678 * red black tree is O(log(n)). Must hold the base lock.
679 */
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800680static void enqueue_hrtimer(struct hrtimer *timer,
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800681 struct hrtimer_clock_base *base, int reprogram)
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800682{
683 struct rb_node **link = &base->active.rb_node;
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800684 struct rb_node *parent = NULL;
685 struct hrtimer *entry;
686
687 /*
688 * Find the right place in the rbtree:
689 */
690 while (*link) {
691 parent = *link;
692 entry = rb_entry(parent, struct hrtimer, node);
693 /*
694 * We dont care about collisions. Nodes with
695 * the same expiry time stay together.
696 */
697 if (timer->expires.tv64 < entry->expires.tv64)
698 link = &(*link)->rb_left;
Thomas Gleixner288867e2006-01-12 11:25:54 +0100699 else
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800700 link = &(*link)->rb_right;
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800701 }
702
703 /*
Thomas Gleixner288867e2006-01-12 11:25:54 +0100704 * Insert the timer to the rbtree and check whether it
705 * replaces the first pending timer
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800706 */
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800707 if (!base->first || timer->expires.tv64 <
708 rb_entry(base->first, struct hrtimer, node)->expires.tv64) {
709 /*
710 * Reprogram the clock event device. When the timer is already
711 * expired hrtimer_enqueue_reprogram has either called the
712 * callback or added it to the pending list and raised the
713 * softirq.
714 *
715 * This is a NOP for !HIGHRES
716 */
717 if (reprogram && hrtimer_enqueue_reprogram(timer, base))
718 return;
719
720 base->first = &timer->node;
721 }
722
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800723 rb_link_node(&timer->node, parent, link);
724 rb_insert_color(&timer->node, &base->active);
Thomas Gleixner303e9672007-02-16 01:27:51 -0800725 /*
726 * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
727 * state of a possibly running callback.
728 */
729 timer->state |= HRTIMER_STATE_ENQUEUED;
Thomas Gleixner288867e2006-01-12 11:25:54 +0100730}
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800731
732/*
733 * __remove_hrtimer - internal function to remove a timer
734 *
735 * Caller must hold the base lock.
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800736 *
737 * High resolution timer mode reprograms the clock event device when the
738 * timer is the one which expires next. The caller can disable this by setting
739 * reprogram to zero. This is useful, when the context does a reprogramming
740 * anyway (e.g. timer interrupt)
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800741 */
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800742static void __remove_hrtimer(struct hrtimer *timer,
Thomas Gleixner303e9672007-02-16 01:27:51 -0800743 struct hrtimer_clock_base *base,
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800744 unsigned long newstate, int reprogram)
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800745{
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800746 /* High res. callback list. NOP for !HIGHRES */
747 if (hrtimer_cb_pending(timer))
748 hrtimer_remove_cb_pending(timer);
749 else {
750 /*
751 * Remove the timer from the rbtree and replace the
752 * first entry pointer if necessary.
753 */
754 if (base->first == &timer->node) {
755 base->first = rb_next(&timer->node);
756 /* Reprogram the clock event device. if enabled */
757 if (reprogram && hrtimer_hres_active())
758 hrtimer_force_reprogram(base->cpu_base);
759 }
760 rb_erase(&timer->node, &base->active);
761 }
Thomas Gleixner303e9672007-02-16 01:27:51 -0800762 timer->state = newstate;
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800763}
764
765/*
766 * remove hrtimer, called with base lock held
767 */
768static inline int
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800769remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base)
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800770{
Thomas Gleixner303e9672007-02-16 01:27:51 -0800771 if (hrtimer_is_queued(timer)) {
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800772 int reprogram;
773
774 /*
775 * Remove the timer and force reprogramming when high
776 * resolution mode is active and the timer is on the current
777 * CPU. If we remove a timer on another CPU, reprogramming is
778 * skipped. The interrupt event on this CPU is fired and
779 * reprogramming happens in the interrupt handler. This is a
780 * rare case and less expensive than a smp call.
781 */
Ingo Molnar82f67cd2007-02-16 01:28:13 -0800782 timer_stats_hrtimer_clear_start_info(timer);
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800783 reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases);
784 __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE,
785 reprogram);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800786 return 1;
787 }
788 return 0;
789}
790
791/**
792 * hrtimer_start - (re)start an relative timer on the current CPU
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800793 * @timer: the timer to be added
794 * @tim: expiry time
795 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
796 *
797 * Returns:
798 * 0 on success
799 * 1 when the timer was active
800 */
801int
802hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode)
803{
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800804 struct hrtimer_clock_base *base, *new_base;
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800805 unsigned long flags;
806 int ret;
807
808 base = lock_hrtimer_base(timer, &flags);
809
810 /* Remove an active timer from the queue: */
811 ret = remove_hrtimer(timer, base);
812
813 /* Switch the timer base, if necessary: */
814 new_base = switch_hrtimer_base(timer, base);
815
Thomas Gleixnerc9cb2e32007-02-16 01:27:49 -0800816 if (mode == HRTIMER_MODE_REL) {
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800817 tim = ktime_add(tim, new_base->get_time());
Ingo Molnar06027bd2006-02-14 13:53:15 -0800818 /*
819 * CONFIG_TIME_LOW_RES is a temporary way for architectures
820 * to signal that they simply return xtime in
821 * do_gettimeoffset(). In this case we want to round up by
822 * resolution when starting a relative timer, to avoid short
823 * timeouts. This will go away with the GTOD framework.
824 */
825#ifdef CONFIG_TIME_LOW_RES
826 tim = ktime_add(tim, base->resolution);
827#endif
828 }
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800829 timer->expires = tim;
830
Ingo Molnar82f67cd2007-02-16 01:28:13 -0800831 timer_stats_hrtimer_set_start_info(timer);
832
Ingo Molnar935c6312007-03-28 13:17:18 +0200833 /*
834 * Only allow reprogramming if the new base is on this CPU.
835 * (it might still be on another CPU if the timer was pending)
836 */
837 enqueue_hrtimer(timer, new_base,
838 new_base->cpu_base == &__get_cpu_var(hrtimer_bases));
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800839
840 unlock_hrtimer_base(timer, &flags);
841
842 return ret;
843}
Stephen Hemminger8d16b762006-05-30 21:26:09 -0700844EXPORT_SYMBOL_GPL(hrtimer_start);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800845
846/**
847 * hrtimer_try_to_cancel - try to deactivate a timer
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800848 * @timer: hrtimer to stop
849 *
850 * Returns:
851 * 0 when the timer was not active
852 * 1 when the timer was active
853 * -1 when the timer is currently excuting the callback function and
Randy Dunlapfa9799e2006-06-25 05:49:15 -0700854 * cannot be stopped
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800855 */
856int hrtimer_try_to_cancel(struct hrtimer *timer)
857{
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800858 struct hrtimer_clock_base *base;
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800859 unsigned long flags;
860 int ret = -1;
861
862 base = lock_hrtimer_base(timer, &flags);
863
Thomas Gleixner303e9672007-02-16 01:27:51 -0800864 if (!hrtimer_callback_running(timer))
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800865 ret = remove_hrtimer(timer, base);
866
867 unlock_hrtimer_base(timer, &flags);
868
869 return ret;
870
871}
Stephen Hemminger8d16b762006-05-30 21:26:09 -0700872EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800873
874/**
875 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800876 * @timer: the timer to be cancelled
877 *
878 * Returns:
879 * 0 when the timer was not active
880 * 1 when the timer was active
881 */
882int hrtimer_cancel(struct hrtimer *timer)
883{
884 for (;;) {
885 int ret = hrtimer_try_to_cancel(timer);
886
887 if (ret >= 0)
888 return ret;
Joe Korty5ef37b12006-04-10 22:54:13 -0700889 cpu_relax();
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800890 }
891}
Stephen Hemminger8d16b762006-05-30 21:26:09 -0700892EXPORT_SYMBOL_GPL(hrtimer_cancel);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800893
894/**
895 * hrtimer_get_remaining - get remaining time for the timer
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800896 * @timer: the timer to read
897 */
898ktime_t hrtimer_get_remaining(const struct hrtimer *timer)
899{
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800900 struct hrtimer_clock_base *base;
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800901 unsigned long flags;
902 ktime_t rem;
903
904 base = lock_hrtimer_base(timer, &flags);
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800905 rem = ktime_sub(timer->expires, base->get_time());
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800906 unlock_hrtimer_base(timer, &flags);
907
908 return rem;
909}
Stephen Hemminger8d16b762006-05-30 21:26:09 -0700910EXPORT_SYMBOL_GPL(hrtimer_get_remaining);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800911
Thomas Gleixnerfd064b92007-02-16 01:27:47 -0800912#if defined(CONFIG_NO_IDLE_HZ) || defined(CONFIG_NO_HZ)
Tony Lindgren69239742006-03-06 15:42:45 -0800913/**
914 * hrtimer_get_next_event - get the time until next expiry event
915 *
916 * Returns the delta to the next expiry event or KTIME_MAX if no timer
917 * is pending.
918 */
919ktime_t hrtimer_get_next_event(void)
920{
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800921 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
922 struct hrtimer_clock_base *base = cpu_base->clock_base;
Tony Lindgren69239742006-03-06 15:42:45 -0800923 ktime_t delta, mindelta = { .tv64 = KTIME_MAX };
924 unsigned long flags;
925 int i;
926
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800927 spin_lock_irqsave(&cpu_base->lock, flags);
928
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800929 if (!hrtimer_hres_active()) {
930 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
931 struct hrtimer *timer;
Tony Lindgren69239742006-03-06 15:42:45 -0800932
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800933 if (!base->first)
934 continue;
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800935
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800936 timer = rb_entry(base->first, struct hrtimer, node);
937 delta.tv64 = timer->expires.tv64;
938 delta = ktime_sub(delta, base->get_time());
939 if (delta.tv64 < mindelta.tv64)
940 mindelta.tv64 = delta.tv64;
941 }
Tony Lindgren69239742006-03-06 15:42:45 -0800942 }
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800943
944 spin_unlock_irqrestore(&cpu_base->lock, flags);
945
Tony Lindgren69239742006-03-06 15:42:45 -0800946 if (mindelta.tv64 < 0)
947 mindelta.tv64 = 0;
948 return mindelta;
949}
950#endif
951
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800952/**
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800953 * hrtimer_init - initialize a timer to the given clock
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800954 * @timer: the timer to be initialized
955 * @clock_id: the clock to be used
George Anzinger79786722006-02-01 03:05:11 -0800956 * @mode: timer mode abs/rel
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800957 */
George Anzinger79786722006-02-01 03:05:11 -0800958void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
959 enum hrtimer_mode mode)
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800960{
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800961 struct hrtimer_cpu_base *cpu_base;
George Anzinger79786722006-02-01 03:05:11 -0800962
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800963 memset(timer, 0, sizeof(struct hrtimer));
George Anzinger79786722006-02-01 03:05:11 -0800964
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800965 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
George Anzinger79786722006-02-01 03:05:11 -0800966
Thomas Gleixnerc9cb2e32007-02-16 01:27:49 -0800967 if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
George Anzinger79786722006-02-01 03:05:11 -0800968 clock_id = CLOCK_MONOTONIC;
969
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800970 timer->base = &cpu_base->clock_base[clock_id];
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800971 hrtimer_init_timer_hres(timer);
Ingo Molnar82f67cd2007-02-16 01:28:13 -0800972
973#ifdef CONFIG_TIMER_STATS
974 timer->start_site = NULL;
975 timer->start_pid = -1;
976 memset(timer->start_comm, 0, TASK_COMM_LEN);
977#endif
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800978}
Stephen Hemminger8d16b762006-05-30 21:26:09 -0700979EXPORT_SYMBOL_GPL(hrtimer_init);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800980
981/**
982 * hrtimer_get_res - get the timer resolution for a clock
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800983 * @which_clock: which clock to query
984 * @tp: pointer to timespec variable to store the resolution
985 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800986 * Store the resolution of the clock selected by @which_clock in the
987 * variable pointed to by @tp.
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800988 */
989int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp)
990{
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800991 struct hrtimer_cpu_base *cpu_base;
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800992
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800993 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
994 *tp = ktime_to_timespec(cpu_base->clock_base[which_clock].resolution);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800995
996 return 0;
997}
Stephen Hemminger8d16b762006-05-30 21:26:09 -0700998EXPORT_SYMBOL_GPL(hrtimer_get_res);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800999
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -08001000#ifdef CONFIG_HIGH_RES_TIMERS
1001
1002/*
1003 * High resolution timer interrupt
1004 * Called with interrupts disabled
1005 */
1006void hrtimer_interrupt(struct clock_event_device *dev)
1007{
1008 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1009 struct hrtimer_clock_base *base;
1010 ktime_t expires_next, now;
1011 int i, raise = 0;
1012
1013 BUG_ON(!cpu_base->hres_active);
1014 cpu_base->nr_events++;
1015 dev->next_event.tv64 = KTIME_MAX;
1016
1017 retry:
1018 now = ktime_get();
1019
1020 expires_next.tv64 = KTIME_MAX;
1021
1022 base = cpu_base->clock_base;
1023
1024 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1025 ktime_t basenow;
1026 struct rb_node *node;
1027
1028 spin_lock(&cpu_base->lock);
1029
1030 basenow = ktime_add(now, base->offset);
1031
1032 while ((node = base->first)) {
1033 struct hrtimer *timer;
1034
1035 timer = rb_entry(node, struct hrtimer, node);
1036
1037 if (basenow.tv64 < timer->expires.tv64) {
1038 ktime_t expires;
1039
1040 expires = ktime_sub(timer->expires,
1041 base->offset);
1042 if (expires.tv64 < expires_next.tv64)
1043 expires_next = expires;
1044 break;
1045 }
1046
1047 /* Move softirq callbacks to the pending list */
1048 if (timer->cb_mode == HRTIMER_CB_SOFTIRQ) {
1049 __remove_hrtimer(timer, base,
1050 HRTIMER_STATE_PENDING, 0);
1051 list_add_tail(&timer->cb_entry,
1052 &base->cpu_base->cb_pending);
1053 raise = 1;
1054 continue;
1055 }
1056
1057 __remove_hrtimer(timer, base,
1058 HRTIMER_STATE_CALLBACK, 0);
Ingo Molnar82f67cd2007-02-16 01:28:13 -08001059 timer_stats_account_hrtimer(timer);
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -08001060
1061 /*
1062 * Note: We clear the CALLBACK bit after
1063 * enqueue_hrtimer to avoid reprogramming of
1064 * the event hardware. This happens at the end
1065 * of this function anyway.
1066 */
1067 if (timer->function(timer) != HRTIMER_NORESTART) {
1068 BUG_ON(timer->state != HRTIMER_STATE_CALLBACK);
1069 enqueue_hrtimer(timer, base, 0);
1070 }
1071 timer->state &= ~HRTIMER_STATE_CALLBACK;
1072 }
1073 spin_unlock(&cpu_base->lock);
1074 base++;
1075 }
1076
1077 cpu_base->expires_next = expires_next;
1078
1079 /* Reprogramming necessary ? */
1080 if (expires_next.tv64 != KTIME_MAX) {
1081 if (tick_program_event(expires_next, 0))
1082 goto retry;
1083 }
1084
1085 /* Raise softirq ? */
1086 if (raise)
1087 raise_softirq(HRTIMER_SOFTIRQ);
1088}
1089
1090static void run_hrtimer_softirq(struct softirq_action *h)
1091{
1092 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1093
1094 spin_lock_irq(&cpu_base->lock);
1095
1096 while (!list_empty(&cpu_base->cb_pending)) {
1097 enum hrtimer_restart (*fn)(struct hrtimer *);
1098 struct hrtimer *timer;
1099 int restart;
1100
1101 timer = list_entry(cpu_base->cb_pending.next,
1102 struct hrtimer, cb_entry);
1103
Ingo Molnar82f67cd2007-02-16 01:28:13 -08001104 timer_stats_account_hrtimer(timer);
1105
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -08001106 fn = timer->function;
1107 __remove_hrtimer(timer, timer->base, HRTIMER_STATE_CALLBACK, 0);
1108 spin_unlock_irq(&cpu_base->lock);
1109
1110 restart = fn(timer);
1111
1112 spin_lock_irq(&cpu_base->lock);
1113
1114 timer->state &= ~HRTIMER_STATE_CALLBACK;
1115 if (restart == HRTIMER_RESTART) {
1116 BUG_ON(hrtimer_active(timer));
1117 /*
1118 * Enqueue the timer, allow reprogramming of the event
1119 * device
1120 */
1121 enqueue_hrtimer(timer, timer->base, 1);
1122 } else if (hrtimer_active(timer)) {
1123 /*
1124 * If the timer was rearmed on another CPU, reprogram
1125 * the event device.
1126 */
1127 if (timer->base->first == &timer->node)
1128 hrtimer_reprogram(timer, timer->base);
1129 }
1130 }
1131 spin_unlock_irq(&cpu_base->lock);
1132}
1133
1134#endif /* CONFIG_HIGH_RES_TIMERS */
1135
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001136/*
1137 * Expire the per base hrtimer-queue:
1138 */
Thomas Gleixner3c8aa392007-02-16 01:27:50 -08001139static inline void run_hrtimer_queue(struct hrtimer_cpu_base *cpu_base,
1140 int index)
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001141{
Thomas Gleixner288867e2006-01-12 11:25:54 +01001142 struct rb_node *node;
Thomas Gleixner3c8aa392007-02-16 01:27:50 -08001143 struct hrtimer_clock_base *base = &cpu_base->clock_base[index];
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001144
Dimitri Sivanich3055add2006-03-31 02:31:20 -08001145 if (!base->first)
1146 return;
1147
Thomas Gleixner92127c72006-03-26 01:38:05 -08001148 if (base->get_softirq_time)
1149 base->softirq_time = base->get_softirq_time();
1150
Thomas Gleixner3c8aa392007-02-16 01:27:50 -08001151 spin_lock_irq(&cpu_base->lock);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001152
Thomas Gleixner288867e2006-01-12 11:25:54 +01001153 while ((node = base->first)) {
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001154 struct hrtimer *timer;
Thomas Gleixnerc9cb2e32007-02-16 01:27:49 -08001155 enum hrtimer_restart (*fn)(struct hrtimer *);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001156 int restart;
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001157
Thomas Gleixner288867e2006-01-12 11:25:54 +01001158 timer = rb_entry(node, struct hrtimer, node);
Thomas Gleixner92127c72006-03-26 01:38:05 -08001159 if (base->softirq_time.tv64 <= timer->expires.tv64)
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001160 break;
1161
Thomas Gleixnerf8953852007-03-06 01:42:08 -08001162#ifdef CONFIG_HIGH_RES_TIMERS
1163 WARN_ON_ONCE(timer->cb_mode == HRTIMER_CB_IRQSAFE_NO_SOFTIRQ);
1164#endif
Ingo Molnar82f67cd2007-02-16 01:28:13 -08001165 timer_stats_account_hrtimer(timer);
1166
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001167 fn = timer->function;
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -08001168 __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0);
Thomas Gleixner3c8aa392007-02-16 01:27:50 -08001169 spin_unlock_irq(&cpu_base->lock);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001170
Roman Zippel05cfb612006-03-26 01:38:12 -08001171 restart = fn(timer);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001172
Thomas Gleixner3c8aa392007-02-16 01:27:50 -08001173 spin_lock_irq(&cpu_base->lock);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001174
Thomas Gleixner303e9672007-02-16 01:27:51 -08001175 timer->state &= ~HRTIMER_STATE_CALLBACK;
Roman Zippelb75f7a52006-03-26 01:38:09 -08001176 if (restart != HRTIMER_NORESTART) {
1177 BUG_ON(hrtimer_active(timer));
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -08001178 enqueue_hrtimer(timer, base, 0);
Roman Zippelb75f7a52006-03-26 01:38:09 -08001179 }
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001180 }
Thomas Gleixner3c8aa392007-02-16 01:27:50 -08001181 spin_unlock_irq(&cpu_base->lock);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001182}
1183
1184/*
1185 * Called from timer softirq every jiffy, expire hrtimers:
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -08001186 *
1187 * For HRT its the fall back code to run the softirq in the timer
1188 * softirq context in case the hrtimer initialization failed or has
1189 * not been done yet.
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001190 */
1191void hrtimer_run_queues(void)
1192{
Thomas Gleixner3c8aa392007-02-16 01:27:50 -08001193 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001194 int i;
1195
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -08001196 if (hrtimer_hres_active())
1197 return;
1198
Thomas Gleixner79bf2bb2007-02-16 01:28:03 -08001199 /*
1200 * This _is_ ugly: We have to check in the softirq context,
1201 * whether we can switch to highres and / or nohz mode. The
1202 * clocksource switch happens in the timer interrupt with
1203 * xtime_lock held. Notification from there only sets the
1204 * check bit in the tick_oneshot code, otherwise we might
1205 * deadlock vs. xtime_lock.
1206 */
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -08001207 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
Thomas Gleixnerf8953852007-03-06 01:42:08 -08001208 if (hrtimer_switch_to_hres())
1209 return;
Thomas Gleixner79bf2bb2007-02-16 01:28:03 -08001210
Thomas Gleixner3c8aa392007-02-16 01:27:50 -08001211 hrtimer_get_softirq_time(cpu_base);
Thomas Gleixner92127c72006-03-26 01:38:05 -08001212
Thomas Gleixner3c8aa392007-02-16 01:27:50 -08001213 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
1214 run_hrtimer_queue(cpu_base, i);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001215}
1216
1217/*
Thomas Gleixner10c94ec2006-01-09 20:52:35 -08001218 * Sleep related functions:
1219 */
Thomas Gleixnerc9cb2e32007-02-16 01:27:49 -08001220static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
Thomas Gleixner00362e32006-03-31 02:31:17 -08001221{
1222 struct hrtimer_sleeper *t =
1223 container_of(timer, struct hrtimer_sleeper, timer);
1224 struct task_struct *task = t->task;
1225
1226 t->task = NULL;
1227 if (task)
1228 wake_up_process(task);
1229
1230 return HRTIMER_NORESTART;
1231}
1232
Ingo Molnar36c8b582006-07-03 00:25:41 -07001233void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
Thomas Gleixner00362e32006-03-31 02:31:17 -08001234{
1235 sl->timer.function = hrtimer_wakeup;
1236 sl->task = task;
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -08001237#ifdef CONFIG_HIGH_RES_TIMERS
1238 sl->timer.cb_mode = HRTIMER_CB_IRQSAFE_NO_RESTART;
1239#endif
Thomas Gleixner00362e32006-03-31 02:31:17 -08001240}
1241
Thomas Gleixner669d7862006-03-31 02:31:19 -08001242static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
Thomas Gleixner10c94ec2006-01-09 20:52:35 -08001243{
Thomas Gleixner669d7862006-03-31 02:31:19 -08001244 hrtimer_init_sleeper(t, current);
Thomas Gleixner10c94ec2006-01-09 20:52:35 -08001245
Roman Zippel432569b2006-03-26 01:38:08 -08001246 do {
1247 set_current_state(TASK_INTERRUPTIBLE);
1248 hrtimer_start(&t->timer, t->timer.expires, mode);
1249
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -08001250 if (likely(t->task))
1251 schedule();
Roman Zippel432569b2006-03-26 01:38:08 -08001252
Thomas Gleixner669d7862006-03-31 02:31:19 -08001253 hrtimer_cancel(&t->timer);
Thomas Gleixnerc9cb2e32007-02-16 01:27:49 -08001254 mode = HRTIMER_MODE_ABS;
Roman Zippel432569b2006-03-26 01:38:08 -08001255
Thomas Gleixner669d7862006-03-31 02:31:19 -08001256 } while (t->task && !signal_pending(current));
1257
1258 return t->task == NULL;
Thomas Gleixner10c94ec2006-01-09 20:52:35 -08001259}
1260
Toyo Abe1711ef32006-09-29 02:00:28 -07001261long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
Thomas Gleixner10c94ec2006-01-09 20:52:35 -08001262{
Thomas Gleixner669d7862006-03-31 02:31:19 -08001263 struct hrtimer_sleeper t;
Ingo Molnarea13dbc2006-01-16 10:59:41 +01001264 struct timespec __user *rmtp;
1265 struct timespec tu;
Roman Zippel432569b2006-03-26 01:38:08 -08001266 ktime_t time;
Thomas Gleixner10c94ec2006-01-09 20:52:35 -08001267
1268 restart->fn = do_no_restart_syscall;
1269
Thomas Gleixnerc9cb2e32007-02-16 01:27:49 -08001270 hrtimer_init(&t.timer, restart->arg0, HRTIMER_MODE_ABS);
Toyo Abe1711ef32006-09-29 02:00:28 -07001271 t.timer.expires.tv64 = ((u64)restart->arg3 << 32) | (u64) restart->arg2;
Thomas Gleixner10c94ec2006-01-09 20:52:35 -08001272
Thomas Gleixnerc9cb2e32007-02-16 01:27:49 -08001273 if (do_nanosleep(&t, HRTIMER_MODE_ABS))
Thomas Gleixner10c94ec2006-01-09 20:52:35 -08001274 return 0;
1275
Toyo Abe1711ef32006-09-29 02:00:28 -07001276 rmtp = (struct timespec __user *) restart->arg1;
Roman Zippel432569b2006-03-26 01:38:08 -08001277 if (rmtp) {
1278 time = ktime_sub(t.timer.expires, t.timer.base->get_time());
1279 if (time.tv64 <= 0)
1280 return 0;
1281 tu = ktime_to_timespec(time);
1282 if (copy_to_user(rmtp, &tu, sizeof(tu)))
1283 return -EFAULT;
1284 }
Thomas Gleixner10c94ec2006-01-09 20:52:35 -08001285
Toyo Abe1711ef32006-09-29 02:00:28 -07001286 restart->fn = hrtimer_nanosleep_restart;
Thomas Gleixner10c94ec2006-01-09 20:52:35 -08001287
1288 /* The other values in restart are already filled in */
1289 return -ERESTART_RESTARTBLOCK;
1290}
1291
Thomas Gleixner10c94ec2006-01-09 20:52:35 -08001292long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
1293 const enum hrtimer_mode mode, const clockid_t clockid)
1294{
1295 struct restart_block *restart;
Thomas Gleixner669d7862006-03-31 02:31:19 -08001296 struct hrtimer_sleeper t;
Thomas Gleixner10c94ec2006-01-09 20:52:35 -08001297 struct timespec tu;
1298 ktime_t rem;
1299
Roman Zippel432569b2006-03-26 01:38:08 -08001300 hrtimer_init(&t.timer, clockid, mode);
1301 t.timer.expires = timespec_to_ktime(*rqtp);
1302 if (do_nanosleep(&t, mode))
Thomas Gleixner10c94ec2006-01-09 20:52:35 -08001303 return 0;
1304
George Anzinger79786722006-02-01 03:05:11 -08001305 /* Absolute timers do not update the rmtp value and restart: */
Thomas Gleixnerc9cb2e32007-02-16 01:27:49 -08001306 if (mode == HRTIMER_MODE_ABS)
Thomas Gleixner10c94ec2006-01-09 20:52:35 -08001307 return -ERESTARTNOHAND;
1308
Roman Zippel432569b2006-03-26 01:38:08 -08001309 if (rmtp) {
1310 rem = ktime_sub(t.timer.expires, t.timer.base->get_time());
1311 if (rem.tv64 <= 0)
1312 return 0;
1313 tu = ktime_to_timespec(rem);
1314 if (copy_to_user(rmtp, &tu, sizeof(tu)))
1315 return -EFAULT;
1316 }
Thomas Gleixner10c94ec2006-01-09 20:52:35 -08001317
1318 restart = &current_thread_info()->restart_block;
Toyo Abe1711ef32006-09-29 02:00:28 -07001319 restart->fn = hrtimer_nanosleep_restart;
1320 restart->arg0 = (unsigned long) t.timer.base->index;
1321 restart->arg1 = (unsigned long) rmtp;
1322 restart->arg2 = t.timer.expires.tv64 & 0xFFFFFFFF;
1323 restart->arg3 = t.timer.expires.tv64 >> 32;
Thomas Gleixner10c94ec2006-01-09 20:52:35 -08001324
1325 return -ERESTART_RESTARTBLOCK;
1326}
1327
Thomas Gleixner6ba1b912006-01-09 20:52:36 -08001328asmlinkage long
1329sys_nanosleep(struct timespec __user *rqtp, struct timespec __user *rmtp)
1330{
1331 struct timespec tu;
1332
1333 if (copy_from_user(&tu, rqtp, sizeof(tu)))
1334 return -EFAULT;
1335
1336 if (!timespec_valid(&tu))
1337 return -EINVAL;
1338
Thomas Gleixnerc9cb2e32007-02-16 01:27:49 -08001339 return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
Thomas Gleixner6ba1b912006-01-09 20:52:36 -08001340}
1341
Thomas Gleixner10c94ec2006-01-09 20:52:35 -08001342/*
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001343 * Functions related to boot-time initialization:
1344 */
1345static void __devinit init_hrtimers_cpu(int cpu)
1346{
Thomas Gleixner3c8aa392007-02-16 01:27:50 -08001347 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001348 int i;
1349
Thomas Gleixner3c8aa392007-02-16 01:27:50 -08001350 spin_lock_init(&cpu_base->lock);
1351 lockdep_set_class(&cpu_base->lock, &cpu_base->lock_key);
1352
1353 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
1354 cpu_base->clock_base[i].cpu_base = cpu_base;
1355
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -08001356 hrtimer_init_hres(cpu_base);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001357}
1358
1359#ifdef CONFIG_HOTPLUG_CPU
1360
Thomas Gleixner3c8aa392007-02-16 01:27:50 -08001361static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1362 struct hrtimer_clock_base *new_base)
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001363{
1364 struct hrtimer *timer;
1365 struct rb_node *node;
1366
1367 while ((node = rb_first(&old_base->active))) {
1368 timer = rb_entry(node, struct hrtimer, node);
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -08001369 BUG_ON(hrtimer_callback_running(timer));
1370 __remove_hrtimer(timer, old_base, HRTIMER_STATE_INACTIVE, 0);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001371 timer->base = new_base;
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -08001372 /*
1373 * Enqueue the timer. Allow reprogramming of the event device
1374 */
1375 enqueue_hrtimer(timer, new_base, 1);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001376 }
1377}
1378
1379static void migrate_hrtimers(int cpu)
1380{
Thomas Gleixner3c8aa392007-02-16 01:27:50 -08001381 struct hrtimer_cpu_base *old_base, *new_base;
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001382 int i;
1383
1384 BUG_ON(cpu_online(cpu));
Thomas Gleixner3c8aa392007-02-16 01:27:50 -08001385 old_base = &per_cpu(hrtimer_bases, cpu);
1386 new_base = &get_cpu_var(hrtimer_bases);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001387
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -08001388 tick_cancel_sched_timer(cpu);
1389
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001390 local_irq_disable();
Heiko Carstense81ce1f2007-03-05 00:30:51 -08001391 double_spin_lock(&new_base->lock, &old_base->lock,
1392 smp_processor_id() < cpu);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001393
Thomas Gleixner3c8aa392007-02-16 01:27:50 -08001394 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
Thomas Gleixner3c8aa392007-02-16 01:27:50 -08001395 migrate_hrtimer_list(&old_base->clock_base[i],
1396 &new_base->clock_base[i]);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001397 }
1398
Heiko Carstense81ce1f2007-03-05 00:30:51 -08001399 double_spin_unlock(&new_base->lock, &old_base->lock,
1400 smp_processor_id() < cpu);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001401 local_irq_enable();
1402 put_cpu_var(hrtimer_bases);
1403}
1404#endif /* CONFIG_HOTPLUG_CPU */
1405
Chandra Seetharaman8c78f302006-07-30 03:03:35 -07001406static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self,
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001407 unsigned long action, void *hcpu)
1408{
1409 long cpu = (long)hcpu;
1410
1411 switch (action) {
1412
1413 case CPU_UP_PREPARE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07001414 case CPU_UP_PREPARE_FROZEN:
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001415 init_hrtimers_cpu(cpu);
1416 break;
1417
1418#ifdef CONFIG_HOTPLUG_CPU
1419 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07001420 case CPU_DEAD_FROZEN:
Thomas Gleixnerd316c572007-02-16 01:28:00 -08001421 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &cpu);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001422 migrate_hrtimers(cpu);
1423 break;
1424#endif
1425
1426 default:
1427 break;
1428 }
1429
1430 return NOTIFY_OK;
1431}
1432
Chandra Seetharaman8c78f302006-07-30 03:03:35 -07001433static struct notifier_block __cpuinitdata hrtimers_nb = {
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001434 .notifier_call = hrtimer_cpu_notify,
1435};
1436
1437void __init hrtimers_init(void)
1438{
1439 hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
1440 (void *)(long)smp_processor_id());
1441 register_cpu_notifier(&hrtimers_nb);
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -08001442#ifdef CONFIG_HIGH_RES_TIMERS
1443 open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq, NULL);
1444#endif
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001445}
1446