blob: f33afb0407bcdc5c6d6a21e06cd41dde40d46d84 [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>
35#include <linux/module.h>
36#include <linux/percpu.h>
37#include <linux/hrtimer.h>
38#include <linux/notifier.h>
39#include <linux/syscalls.h>
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -080040#include <linux/kallsyms.h>
Thomas Gleixnerc0a31322006-01-09 20:52:32 -080041#include <linux/interrupt.h>
Thomas Gleixner79bf2bb2007-02-16 01:28:03 -080042#include <linux/tick.h>
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -080043#include <linux/seq_file.h>
44#include <linux/err.h>
Thomas Gleixner237fc6e2008-04-30 00:55:04 -070045#include <linux/debugobjects.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 Anzinger7978672c2006-02-01 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 stultz2c6b47d2007-07-24 17:47:43 -0700144 xts = current_kernel_time();
Thomas Gleixnerad28d942007-03-16 13:38:21 -0800145 tom = wall_to_monotonic;
Thomas Gleixner92127c72006-03-26 01:38:05 -0800146 } while (read_seqretry(&xtime_lock, seq));
147
john stultzf4304ab2007-02-16 01:27:26 -0800148 xtim = timespec_to_ktime(xts);
Thomas Gleixnerad28d942007-03-16 13:38:21 -0800149 tomono = timespec_to_ktime(tom);
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800150 base->clock_base[CLOCK_REALTIME].softirq_time = xtim;
151 base->clock_base[CLOCK_MONOTONIC].softirq_time =
152 ktime_add(xtim, tomono);
Thomas Gleixner92127c72006-03-26 01:38:05 -0800153}
154
155/*
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800156 * Functions and macros which are different for UP/SMP systems are kept in a
157 * single place
158 */
159#ifdef CONFIG_SMP
160
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800161/*
162 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
163 * means that all timers which are tied to this base via timer->base are
164 * locked, and the base itself is locked too.
165 *
166 * So __run_timers/migrate_timers can safely modify all timers which could
167 * be found on the lists/queues.
168 *
169 * When the timer's base is locked, and the timer removed from list, it is
170 * possible to set timer->base = NULL and drop the lock: the timer remains
171 * locked.
172 */
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800173static
174struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
175 unsigned long *flags)
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800176{
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800177 struct hrtimer_clock_base *base;
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800178
179 for (;;) {
180 base = timer->base;
181 if (likely(base != NULL)) {
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800182 spin_lock_irqsave(&base->cpu_base->lock, *flags);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800183 if (likely(base == timer->base))
184 return base;
185 /* The timer has migrated to another CPU: */
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800186 spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800187 }
188 cpu_relax();
189 }
190}
191
192/*
193 * Switch the timer base to the current CPU when possible.
194 */
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800195static inline struct hrtimer_clock_base *
196switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base)
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800197{
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800198 struct hrtimer_clock_base *new_base;
199 struct hrtimer_cpu_base *new_cpu_base;
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800200
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800201 new_cpu_base = &__get_cpu_var(hrtimer_bases);
202 new_base = &new_cpu_base->clock_base[base->index];
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800203
204 if (base != new_base) {
205 /*
206 * We are trying to schedule the timer on the local CPU.
207 * However we can't change timer's base while it is running,
208 * so we keep it on the same CPU. No hassle vs. reprogramming
209 * the event source in the high resolution case. The softirq
210 * code will take care of this when the timer function has
211 * completed. There is no conflict as we hold the lock until
212 * the timer is enqueued.
213 */
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800214 if (unlikely(hrtimer_callback_running(timer)))
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800215 return base;
216
217 /* See the comment in lock_timer_base() */
218 timer->base = NULL;
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800219 spin_unlock(&base->cpu_base->lock);
220 spin_lock(&new_base->cpu_base->lock);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800221 timer->base = new_base;
222 }
223 return new_base;
224}
225
226#else /* CONFIG_SMP */
227
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800228static inline struct hrtimer_clock_base *
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800229lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
230{
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800231 struct hrtimer_clock_base *base = timer->base;
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800232
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800233 spin_lock_irqsave(&base->cpu_base->lock, *flags);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800234
235 return base;
236}
237
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800238# define switch_hrtimer_base(t, b) (b)
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800239
240#endif /* !CONFIG_SMP */
241
242/*
243 * Functions for the union type storage format of ktime_t which are
244 * too large for inlining:
245 */
246#if BITS_PER_LONG < 64
247# ifndef CONFIG_KTIME_SCALAR
248/**
249 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800250 * @kt: addend
251 * @nsec: the scalar nsec value to add
252 *
253 * Returns the sum of kt and nsec in ktime_t format
254 */
255ktime_t ktime_add_ns(const ktime_t kt, u64 nsec)
256{
257 ktime_t tmp;
258
259 if (likely(nsec < NSEC_PER_SEC)) {
260 tmp.tv64 = nsec;
261 } else {
262 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
263
264 tmp = ktime_set((long)nsec, rem);
265 }
266
267 return ktime_add(kt, tmp);
268}
David Howellsb8b8fd22007-04-27 15:31:24 -0700269
270EXPORT_SYMBOL_GPL(ktime_add_ns);
Arnaldo Carvalho de Meloa2723782007-08-19 17:16:05 -0700271
272/**
273 * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable
274 * @kt: minuend
275 * @nsec: the scalar nsec value to subtract
276 *
277 * Returns the subtraction of @nsec from @kt in ktime_t format
278 */
279ktime_t ktime_sub_ns(const ktime_t kt, u64 nsec)
280{
281 ktime_t tmp;
282
283 if (likely(nsec < NSEC_PER_SEC)) {
284 tmp.tv64 = nsec;
285 } else {
286 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
287
288 tmp = ktime_set((long)nsec, rem);
289 }
290
291 return ktime_sub(kt, tmp);
292}
293
294EXPORT_SYMBOL_GPL(ktime_sub_ns);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800295# endif /* !CONFIG_KTIME_SCALAR */
296
297/*
298 * Divide a ktime value by a nanosecond value
299 */
Davide Libenzi4d672e72008-02-04 22:27:26 -0800300u64 ktime_divns(const ktime_t kt, s64 div)
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800301{
Carlos R. Mafra900cfa42008-05-22 19:25:11 -0300302 u64 dclc;
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800303 int sft = 0;
304
Carlos R. Mafra900cfa42008-05-22 19:25:11 -0300305 dclc = ktime_to_ns(kt);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800306 /* Make sure the divisor is less than 2^32: */
307 while (div >> 32) {
308 sft++;
309 div >>= 1;
310 }
311 dclc >>= sft;
312 do_div(dclc, (unsigned long) div);
313
Davide Libenzi4d672e72008-02-04 22:27:26 -0800314 return dclc;
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800315}
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800316#endif /* BITS_PER_LONG >= 64 */
317
Peter Zijlstrad3d74452008-01-25 21:08:31 +0100318/*
Thomas Gleixner5a7780e2008-02-13 09:20:43 +0100319 * Add two ktime values and do a safety check for overflow:
320 */
321ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
322{
323 ktime_t res = ktime_add(lhs, rhs);
324
325 /*
326 * We use KTIME_SEC_MAX here, the maximum timeout which we can
327 * return to user space in a timespec:
328 */
329 if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64)
330 res = ktime_set(KTIME_SEC_MAX, 0);
331
332 return res;
333}
334
Thomas Gleixner237fc6e2008-04-30 00:55:04 -0700335#ifdef CONFIG_DEBUG_OBJECTS_TIMERS
336
337static struct debug_obj_descr hrtimer_debug_descr;
338
339/*
340 * fixup_init is called when:
341 * - an active object is initialized
342 */
343static int hrtimer_fixup_init(void *addr, enum debug_obj_state state)
344{
345 struct hrtimer *timer = addr;
346
347 switch (state) {
348 case ODEBUG_STATE_ACTIVE:
349 hrtimer_cancel(timer);
350 debug_object_init(timer, &hrtimer_debug_descr);
351 return 1;
352 default:
353 return 0;
354 }
355}
356
357/*
358 * fixup_activate is called when:
359 * - an active object is activated
360 * - an unknown object is activated (might be a statically initialized object)
361 */
362static int hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
363{
364 switch (state) {
365
366 case ODEBUG_STATE_NOTAVAILABLE:
367 WARN_ON_ONCE(1);
368 return 0;
369
370 case ODEBUG_STATE_ACTIVE:
371 WARN_ON(1);
372
373 default:
374 return 0;
375 }
376}
377
378/*
379 * fixup_free is called when:
380 * - an active object is freed
381 */
382static int hrtimer_fixup_free(void *addr, enum debug_obj_state state)
383{
384 struct hrtimer *timer = addr;
385
386 switch (state) {
387 case ODEBUG_STATE_ACTIVE:
388 hrtimer_cancel(timer);
389 debug_object_free(timer, &hrtimer_debug_descr);
390 return 1;
391 default:
392 return 0;
393 }
394}
395
396static struct debug_obj_descr hrtimer_debug_descr = {
397 .name = "hrtimer",
398 .fixup_init = hrtimer_fixup_init,
399 .fixup_activate = hrtimer_fixup_activate,
400 .fixup_free = hrtimer_fixup_free,
401};
402
403static inline void debug_hrtimer_init(struct hrtimer *timer)
404{
405 debug_object_init(timer, &hrtimer_debug_descr);
406}
407
408static inline void debug_hrtimer_activate(struct hrtimer *timer)
409{
410 debug_object_activate(timer, &hrtimer_debug_descr);
411}
412
413static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
414{
415 debug_object_deactivate(timer, &hrtimer_debug_descr);
416}
417
418static inline void debug_hrtimer_free(struct hrtimer *timer)
419{
420 debug_object_free(timer, &hrtimer_debug_descr);
421}
422
423static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
424 enum hrtimer_mode mode);
425
426void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
427 enum hrtimer_mode mode)
428{
429 debug_object_init_on_stack(timer, &hrtimer_debug_descr);
430 __hrtimer_init(timer, clock_id, mode);
431}
432
433void destroy_hrtimer_on_stack(struct hrtimer *timer)
434{
435 debug_object_free(timer, &hrtimer_debug_descr);
436}
437
438#else
439static inline void debug_hrtimer_init(struct hrtimer *timer) { }
440static inline void debug_hrtimer_activate(struct hrtimer *timer) { }
441static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
442#endif
443
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800444/* High resolution timer related functions */
445#ifdef CONFIG_HIGH_RES_TIMERS
446
447/*
448 * High resolution timer enabled ?
449 */
450static int hrtimer_hres_enabled __read_mostly = 1;
451
452/*
453 * Enable / Disable high resolution mode
454 */
455static int __init setup_hrtimer_hres(char *str)
456{
457 if (!strcmp(str, "off"))
458 hrtimer_hres_enabled = 0;
459 else if (!strcmp(str, "on"))
460 hrtimer_hres_enabled = 1;
461 else
462 return 0;
463 return 1;
464}
465
466__setup("highres=", setup_hrtimer_hres);
467
468/*
469 * hrtimer_high_res_enabled - query, if the highres mode is enabled
470 */
471static inline int hrtimer_is_hres_enabled(void)
472{
473 return hrtimer_hres_enabled;
474}
475
476/*
477 * Is the high resolution mode active ?
478 */
479static inline int hrtimer_hres_active(void)
480{
481 return __get_cpu_var(hrtimer_bases).hres_active;
482}
483
484/*
485 * Reprogram the event source with checking both queues for the
486 * next event
487 * Called with interrupts disabled and base->lock held
488 */
489static void hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base)
490{
491 int i;
492 struct hrtimer_clock_base *base = cpu_base->clock_base;
493 ktime_t expires;
494
495 cpu_base->expires_next.tv64 = KTIME_MAX;
496
497 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
498 struct hrtimer *timer;
499
500 if (!base->first)
501 continue;
502 timer = rb_entry(base->first, struct hrtimer, node);
Arjan van de Vencc584b22008-09-01 15:02:30 -0700503 expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800504 if (expires.tv64 < cpu_base->expires_next.tv64)
505 cpu_base->expires_next = expires;
506 }
507
508 if (cpu_base->expires_next.tv64 != KTIME_MAX)
509 tick_program_event(cpu_base->expires_next, 1);
510}
511
512/*
513 * Shared reprogramming for clock_realtime and clock_monotonic
514 *
515 * When a timer is enqueued and expires earlier than the already enqueued
516 * timers, we have to check, whether it expires earlier than the timer for
517 * which the clock event device was armed.
518 *
519 * Called with interrupts disabled and base->cpu_base.lock held
520 */
521static int hrtimer_reprogram(struct hrtimer *timer,
522 struct hrtimer_clock_base *base)
523{
524 ktime_t *expires_next = &__get_cpu_var(hrtimer_bases).expires_next;
Arjan van de Vencc584b22008-09-01 15:02:30 -0700525 ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800526 int res;
527
Arjan van de Vencc584b22008-09-01 15:02:30 -0700528 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
Thomas Gleixner63070a72008-02-14 00:58:36 +0100529
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800530 /*
531 * When the callback is running, we do not reprogram the clock event
532 * device. The timer callback is either running on a different CPU or
Robert P. J. Day3a4fa0a2007-10-19 23:10:43 +0200533 * the callback is executed in the hrtimer_interrupt context. The
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800534 * reprogramming is handled either by the softirq, which called the
535 * callback or at the end of the hrtimer_interrupt.
536 */
537 if (hrtimer_callback_running(timer))
538 return 0;
539
Thomas Gleixner63070a72008-02-14 00:58:36 +0100540 /*
541 * CLOCK_REALTIME timer might be requested with an absolute
542 * expiry time which is less than base->offset. Nothing wrong
543 * about that, just avoid to call into the tick code, which
544 * has now objections against negative expiry values.
545 */
546 if (expires.tv64 < 0)
547 return -ETIME;
548
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800549 if (expires.tv64 >= expires_next->tv64)
550 return 0;
551
552 /*
553 * Clockevents returns -ETIME, when the event was in the past.
554 */
555 res = tick_program_event(expires, 0);
556 if (!IS_ERR_VALUE(res))
557 *expires_next = expires;
558 return res;
559}
560
561
562/*
563 * Retrigger next event is called after clock was set
564 *
565 * Called with interrupts disabled via on_each_cpu()
566 */
567static void retrigger_next_event(void *arg)
568{
569 struct hrtimer_cpu_base *base;
570 struct timespec realtime_offset;
571 unsigned long seq;
572
573 if (!hrtimer_hres_active())
574 return;
575
576 do {
577 seq = read_seqbegin(&xtime_lock);
578 set_normalized_timespec(&realtime_offset,
579 -wall_to_monotonic.tv_sec,
580 -wall_to_monotonic.tv_nsec);
581 } while (read_seqretry(&xtime_lock, seq));
582
583 base = &__get_cpu_var(hrtimer_bases);
584
585 /* Adjust CLOCK_REALTIME offset */
586 spin_lock(&base->lock);
587 base->clock_base[CLOCK_REALTIME].offset =
588 timespec_to_ktime(realtime_offset);
589
590 hrtimer_force_reprogram(base);
591 spin_unlock(&base->lock);
592}
593
594/*
595 * Clock realtime was set
596 *
597 * Change the offset of the realtime clock vs. the monotonic
598 * clock.
599 *
600 * We might have to reprogram the high resolution timer interrupt. On
601 * SMP we call the architecture specific code to retrigger _all_ high
602 * resolution timer interrupts. On UP we just disable interrupts and
603 * call the high resolution interrupt code.
604 */
605void clock_was_set(void)
606{
607 /* Retrigger the CPU local events everywhere */
Jens Axboe15c8b6c2008-05-09 09:39:44 +0200608 on_each_cpu(retrigger_next_event, NULL, 1);
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800609}
610
611/*
Ingo Molnar995f0542007-04-07 12:05:00 +0200612 * During resume we might have to reprogram the high resolution timer
613 * interrupt (on the local CPU):
614 */
615void hres_timers_resume(void)
616{
Peter Zijlstra1d4a7f12009-01-18 16:39:29 +0100617 WARN_ONCE(!irqs_disabled(),
618 KERN_INFO "hres_timers_resume() called with IRQs enabled!");
619
Ingo Molnar995f0542007-04-07 12:05:00 +0200620 retrigger_next_event(NULL);
621}
622
623/*
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800624 * Initialize the high resolution related parts of cpu_base
625 */
626static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
627{
628 base->expires_next.tv64 = KTIME_MAX;
629 base->hres_active = 0;
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800630}
631
632/*
633 * Initialize the high resolution related parts of a hrtimer
634 */
635static inline void hrtimer_init_timer_hres(struct hrtimer *timer)
636{
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800637}
638
Peter Zijlstraca109492008-11-25 12:43:51 +0100639
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800640/*
641 * When High resolution timers are active, try to reprogram. Note, that in case
642 * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry
643 * check happens. The timer gets enqueued into the rbtree. The reprogramming
644 * and expiry check is done in the hrtimer_interrupt or in the softirq.
645 */
646static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
647 struct hrtimer_clock_base *base)
648{
649 if (base->cpu_base->hres_active && hrtimer_reprogram(timer, base)) {
Peter Zijlstraa6037b62009-01-05 11:28:22 +0100650 spin_unlock(&base->cpu_base->lock);
651 raise_softirq_irqoff(HRTIMER_SOFTIRQ);
652 spin_lock(&base->cpu_base->lock);
Peter Zijlstraca109492008-11-25 12:43:51 +0100653 return 1;
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800654 }
655 return 0;
656}
657
658/*
659 * Switch to high resolution mode
660 */
Thomas Gleixnerf8953852007-03-06 01:42:08 -0800661static int hrtimer_switch_to_hres(void)
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800662{
Ingo Molnar820de5c2007-07-21 04:37:36 -0700663 int cpu = smp_processor_id();
664 struct hrtimer_cpu_base *base = &per_cpu(hrtimer_bases, cpu);
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800665 unsigned long flags;
666
667 if (base->hres_active)
Thomas Gleixnerf8953852007-03-06 01:42:08 -0800668 return 1;
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800669
670 local_irq_save(flags);
671
672 if (tick_init_highres()) {
673 local_irq_restore(flags);
Ingo Molnar820de5c2007-07-21 04:37:36 -0700674 printk(KERN_WARNING "Could not switch to high resolution "
675 "mode on CPU %d\n", cpu);
Thomas Gleixnerf8953852007-03-06 01:42:08 -0800676 return 0;
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800677 }
678 base->hres_active = 1;
679 base->clock_base[CLOCK_REALTIME].resolution = KTIME_HIGH_RES;
680 base->clock_base[CLOCK_MONOTONIC].resolution = KTIME_HIGH_RES;
681
682 tick_setup_sched_timer();
683
684 /* "Retrigger" the interrupt to get things going */
685 retrigger_next_event(NULL);
686 local_irq_restore(flags);
Michael Ellermanedfed662007-10-29 16:35:29 +1100687 printk(KERN_DEBUG "Switched to high resolution mode on CPU %d\n",
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800688 smp_processor_id());
Thomas Gleixnerf8953852007-03-06 01:42:08 -0800689 return 1;
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800690}
691
692#else
693
694static inline int hrtimer_hres_active(void) { return 0; }
695static inline int hrtimer_is_hres_enabled(void) { return 0; }
Thomas Gleixnerf8953852007-03-06 01:42:08 -0800696static inline int hrtimer_switch_to_hres(void) { return 0; }
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800697static inline void hrtimer_force_reprogram(struct hrtimer_cpu_base *base) { }
698static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
699 struct hrtimer_clock_base *base)
700{
701 return 0;
702}
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800703static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { }
704static inline void hrtimer_init_timer_hres(struct hrtimer *timer) { }
705
706#endif /* CONFIG_HIGH_RES_TIMERS */
707
Ingo Molnar82f67cd2007-02-16 01:28:13 -0800708#ifdef CONFIG_TIMER_STATS
709void __timer_stats_hrtimer_set_start_info(struct hrtimer *timer, void *addr)
710{
711 if (timer->start_site)
712 return;
713
714 timer->start_site = addr;
715 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
716 timer->start_pid = current->pid;
717}
718#endif
719
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800720/*
Uwe Kleine-König6506f2a2007-10-20 01:56:53 +0200721 * Counterpart to lock_hrtimer_base above:
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800722 */
723static inline
724void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
725{
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800726 spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800727}
728
729/**
730 * hrtimer_forward - forward the timer expiry
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800731 * @timer: hrtimer to forward
Roman Zippel44f21472006-03-26 01:38:06 -0800732 * @now: forward past this time
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800733 * @interval: the interval to forward
734 *
735 * Forward the timer expiry so it will expire in the future.
Jonathan Corbet8dca6f32006-01-16 15:58:55 -0700736 * Returns the number of overruns.
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800737 */
Davide Libenzi4d672e72008-02-04 22:27:26 -0800738u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800739{
Davide Libenzi4d672e72008-02-04 22:27:26 -0800740 u64 orun = 1;
Roman Zippel44f21472006-03-26 01:38:06 -0800741 ktime_t delta;
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800742
Arjan van de Vencc584b22008-09-01 15:02:30 -0700743 delta = ktime_sub(now, hrtimer_get_expires(timer));
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800744
745 if (delta.tv64 < 0)
746 return 0;
747
Thomas Gleixnerc9db4fa2006-01-12 11:47:34 +0100748 if (interval.tv64 < timer->base->resolution.tv64)
749 interval.tv64 = timer->base->resolution.tv64;
750
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800751 if (unlikely(delta.tv64 >= interval.tv64)) {
Roman Zippeldf869b62006-03-26 01:38:11 -0800752 s64 incr = ktime_to_ns(interval);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800753
754 orun = ktime_divns(delta, incr);
Arjan van de Vencc584b22008-09-01 15:02:30 -0700755 hrtimer_add_expires_ns(timer, incr * orun);
756 if (hrtimer_get_expires_tv64(timer) > now.tv64)
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800757 return orun;
758 /*
759 * This (and the ktime_add() below) is the
760 * correction for exact:
761 */
762 orun++;
763 }
Arjan van de Vencc584b22008-09-01 15:02:30 -0700764 hrtimer_add_expires(timer, interval);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800765
766 return orun;
767}
Stas Sergeev6bdb6b62007-05-08 00:31:58 -0700768EXPORT_SYMBOL_GPL(hrtimer_forward);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800769
770/*
771 * enqueue_hrtimer - internal function to (re)start a timer
772 *
773 * The timer is inserted in expiry order. Insertion into the
774 * red black tree is O(log(n)). Must hold the base lock.
Peter Zijlstraa6037b62009-01-05 11:28:22 +0100775 *
776 * Returns 1 when the new timer is the leftmost timer in the tree.
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800777 */
Peter Zijlstraa6037b62009-01-05 11:28:22 +0100778static int enqueue_hrtimer(struct hrtimer *timer,
779 struct hrtimer_clock_base *base)
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800780{
781 struct rb_node **link = &base->active.rb_node;
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800782 struct rb_node *parent = NULL;
783 struct hrtimer *entry;
Ingo Molnar99bc2fc2007-07-21 04:37:36 -0700784 int leftmost = 1;
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800785
Thomas Gleixner237fc6e2008-04-30 00:55:04 -0700786 debug_hrtimer_activate(timer);
787
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800788 /*
789 * Find the right place in the rbtree:
790 */
791 while (*link) {
792 parent = *link;
793 entry = rb_entry(parent, struct hrtimer, node);
794 /*
795 * We dont care about collisions. Nodes with
796 * the same expiry time stay together.
797 */
Arjan van de Vencc584b22008-09-01 15:02:30 -0700798 if (hrtimer_get_expires_tv64(timer) <
799 hrtimer_get_expires_tv64(entry)) {
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800800 link = &(*link)->rb_left;
Ingo Molnar99bc2fc2007-07-21 04:37:36 -0700801 } else {
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800802 link = &(*link)->rb_right;
Ingo Molnar99bc2fc2007-07-21 04:37:36 -0700803 leftmost = 0;
804 }
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800805 }
806
807 /*
Thomas Gleixner288867e2006-01-12 11:25:54 +0100808 * Insert the timer to the rbtree and check whether it
809 * replaces the first pending timer
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800810 */
Peter Zijlstraa6037b62009-01-05 11:28:22 +0100811 if (leftmost)
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800812 base->first = &timer->node;
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800813
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800814 rb_link_node(&timer->node, parent, link);
815 rb_insert_color(&timer->node, &base->active);
Thomas Gleixner303e9672007-02-16 01:27:51 -0800816 /*
817 * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
818 * state of a possibly running callback.
819 */
820 timer->state |= HRTIMER_STATE_ENQUEUED;
Peter Zijlstraa6037b62009-01-05 11:28:22 +0100821
822 return leftmost;
Thomas Gleixner288867e2006-01-12 11:25:54 +0100823}
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800824
825/*
826 * __remove_hrtimer - internal function to remove a timer
827 *
828 * Caller must hold the base lock.
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800829 *
830 * High resolution timer mode reprograms the clock event device when the
831 * timer is the one which expires next. The caller can disable this by setting
832 * reprogram to zero. This is useful, when the context does a reprogramming
833 * anyway (e.g. timer interrupt)
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800834 */
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800835static void __remove_hrtimer(struct hrtimer *timer,
Thomas Gleixner303e9672007-02-16 01:27:51 -0800836 struct hrtimer_clock_base *base,
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800837 unsigned long newstate, int reprogram)
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800838{
Peter Zijlstraca109492008-11-25 12:43:51 +0100839 if (timer->state & HRTIMER_STATE_ENQUEUED) {
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800840 /*
841 * Remove the timer from the rbtree and replace the
842 * first entry pointer if necessary.
843 */
844 if (base->first == &timer->node) {
845 base->first = rb_next(&timer->node);
846 /* Reprogram the clock event device. if enabled */
847 if (reprogram && hrtimer_hres_active())
848 hrtimer_force_reprogram(base->cpu_base);
849 }
850 rb_erase(&timer->node, &base->active);
851 }
Thomas Gleixner303e9672007-02-16 01:27:51 -0800852 timer->state = newstate;
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800853}
854
855/*
856 * remove hrtimer, called with base lock held
857 */
858static inline int
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800859remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base)
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800860{
Thomas Gleixner303e9672007-02-16 01:27:51 -0800861 if (hrtimer_is_queued(timer)) {
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800862 int reprogram;
863
864 /*
865 * Remove the timer and force reprogramming when high
866 * resolution mode is active and the timer is on the current
867 * CPU. If we remove a timer on another CPU, reprogramming is
868 * skipped. The interrupt event on this CPU is fired and
869 * reprogramming happens in the interrupt handler. This is a
870 * rare case and less expensive than a smp call.
871 */
Thomas Gleixner237fc6e2008-04-30 00:55:04 -0700872 debug_hrtimer_deactivate(timer);
Ingo Molnar82f67cd2007-02-16 01:28:13 -0800873 timer_stats_hrtimer_clear_start_info(timer);
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -0800874 reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases);
875 __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE,
876 reprogram);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800877 return 1;
878 }
879 return 0;
880}
881
882/**
Thomas Gleixnere1dd7bc2008-10-20 13:33:36 +0200883 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800884 * @timer: the timer to be added
885 * @tim: expiry time
Arjan van de Venda8f2e12008-09-07 10:47:46 -0700886 * @delta_ns: "slack" range for the timer
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800887 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
888 *
889 * Returns:
890 * 0 on success
891 * 1 when the timer was active
892 */
893int
Arjan van de Venda8f2e12008-09-07 10:47:46 -0700894hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, unsigned long delta_ns,
895 const enum hrtimer_mode mode)
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800896{
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800897 struct hrtimer_clock_base *base, *new_base;
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800898 unsigned long flags;
Peter Zijlstraa6037b62009-01-05 11:28:22 +0100899 int ret, leftmost;
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800900
901 base = lock_hrtimer_base(timer, &flags);
902
903 /* Remove an active timer from the queue: */
904 ret = remove_hrtimer(timer, base);
905
906 /* Switch the timer base, if necessary: */
907 new_base = switch_hrtimer_base(timer, base);
908
Thomas Gleixnerc9cb2e32007-02-16 01:27:49 -0800909 if (mode == HRTIMER_MODE_REL) {
Thomas Gleixner5a7780e2008-02-13 09:20:43 +0100910 tim = ktime_add_safe(tim, new_base->get_time());
Ingo Molnar06027bd2006-02-14 13:53:15 -0800911 /*
912 * CONFIG_TIME_LOW_RES is a temporary way for architectures
913 * to signal that they simply return xtime in
914 * do_gettimeoffset(). In this case we want to round up by
915 * resolution when starting a relative timer, to avoid short
916 * timeouts. This will go away with the GTOD framework.
917 */
918#ifdef CONFIG_TIME_LOW_RES
Thomas Gleixner5a7780e2008-02-13 09:20:43 +0100919 tim = ktime_add_safe(tim, base->resolution);
Ingo Molnar06027bd2006-02-14 13:53:15 -0800920#endif
921 }
Thomas Gleixner237fc6e2008-04-30 00:55:04 -0700922
Arjan van de Venda8f2e12008-09-07 10:47:46 -0700923 hrtimer_set_expires_range_ns(timer, tim, delta_ns);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800924
Ingo Molnar82f67cd2007-02-16 01:28:13 -0800925 timer_stats_hrtimer_set_start_info(timer);
926
Peter Zijlstraa6037b62009-01-05 11:28:22 +0100927 leftmost = enqueue_hrtimer(timer, new_base);
928
Ingo Molnar935c6312007-03-28 13:17:18 +0200929 /*
930 * Only allow reprogramming if the new base is on this CPU.
931 * (it might still be on another CPU if the timer was pending)
Peter Zijlstraa6037b62009-01-05 11:28:22 +0100932 *
933 * XXX send_remote_softirq() ?
Ingo Molnar935c6312007-03-28 13:17:18 +0200934 */
Peter Zijlstraa6037b62009-01-05 11:28:22 +0100935 if (leftmost && new_base->cpu_base == &__get_cpu_var(hrtimer_bases))
936 hrtimer_enqueue_reprogram(timer, new_base);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800937
938 unlock_hrtimer_base(timer, &flags);
939
940 return ret;
941}
Arjan van de Venda8f2e12008-09-07 10:47:46 -0700942EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
943
944/**
Thomas Gleixnere1dd7bc2008-10-20 13:33:36 +0200945 * hrtimer_start - (re)start an hrtimer on the current CPU
Arjan van de Venda8f2e12008-09-07 10:47:46 -0700946 * @timer: the timer to be added
947 * @tim: expiry time
948 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
949 *
950 * Returns:
951 * 0 on success
952 * 1 when the timer was active
953 */
954int
955hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode)
956{
957 return hrtimer_start_range_ns(timer, tim, 0, mode);
958}
Stephen Hemminger8d16b762006-05-30 21:26:09 -0700959EXPORT_SYMBOL_GPL(hrtimer_start);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800960
Arjan van de Venda8f2e12008-09-07 10:47:46 -0700961
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800962/**
963 * hrtimer_try_to_cancel - try to deactivate a timer
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800964 * @timer: hrtimer to stop
965 *
966 * Returns:
967 * 0 when the timer was not active
968 * 1 when the timer was active
969 * -1 when the timer is currently excuting the callback function and
Randy Dunlapfa9799e2006-06-25 05:49:15 -0700970 * cannot be stopped
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800971 */
972int hrtimer_try_to_cancel(struct hrtimer *timer)
973{
Thomas Gleixner3c8aa392007-02-16 01:27:50 -0800974 struct hrtimer_clock_base *base;
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800975 unsigned long flags;
976 int ret = -1;
977
978 base = lock_hrtimer_base(timer, &flags);
979
Thomas Gleixner303e9672007-02-16 01:27:51 -0800980 if (!hrtimer_callback_running(timer))
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800981 ret = remove_hrtimer(timer, base);
982
983 unlock_hrtimer_base(timer, &flags);
984
985 return ret;
986
987}
Stephen Hemminger8d16b762006-05-30 21:26:09 -0700988EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800989
990/**
991 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800992 * @timer: the timer to be cancelled
993 *
994 * Returns:
995 * 0 when the timer was not active
996 * 1 when the timer was active
997 */
998int hrtimer_cancel(struct hrtimer *timer)
999{
1000 for (;;) {
1001 int ret = hrtimer_try_to_cancel(timer);
1002
1003 if (ret >= 0)
1004 return ret;
Joe Korty5ef37b12006-04-10 22:54:13 -07001005 cpu_relax();
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001006 }
1007}
Stephen Hemminger8d16b762006-05-30 21:26:09 -07001008EXPORT_SYMBOL_GPL(hrtimer_cancel);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001009
1010/**
1011 * hrtimer_get_remaining - get remaining time for the timer
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001012 * @timer: the timer to read
1013 */
1014ktime_t hrtimer_get_remaining(const struct hrtimer *timer)
1015{
Thomas Gleixner3c8aa392007-02-16 01:27:50 -08001016 struct hrtimer_clock_base *base;
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001017 unsigned long flags;
1018 ktime_t rem;
1019
1020 base = lock_hrtimer_base(timer, &flags);
Arjan van de Vencc584b22008-09-01 15:02:30 -07001021 rem = hrtimer_expires_remaining(timer);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001022 unlock_hrtimer_base(timer, &flags);
1023
1024 return rem;
1025}
Stephen Hemminger8d16b762006-05-30 21:26:09 -07001026EXPORT_SYMBOL_GPL(hrtimer_get_remaining);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001027
Russell Kingee9c5782008-04-20 13:59:33 +01001028#ifdef CONFIG_NO_HZ
Tony Lindgren69239742006-03-06 15:42:45 -08001029/**
1030 * hrtimer_get_next_event - get the time until next expiry event
1031 *
1032 * Returns the delta to the next expiry event or KTIME_MAX if no timer
1033 * is pending.
1034 */
1035ktime_t hrtimer_get_next_event(void)
1036{
Thomas Gleixner3c8aa392007-02-16 01:27:50 -08001037 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1038 struct hrtimer_clock_base *base = cpu_base->clock_base;
Tony Lindgren69239742006-03-06 15:42:45 -08001039 ktime_t delta, mindelta = { .tv64 = KTIME_MAX };
1040 unsigned long flags;
1041 int i;
1042
Thomas Gleixner3c8aa392007-02-16 01:27:50 -08001043 spin_lock_irqsave(&cpu_base->lock, flags);
1044
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -08001045 if (!hrtimer_hres_active()) {
1046 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
1047 struct hrtimer *timer;
Tony Lindgren69239742006-03-06 15:42:45 -08001048
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -08001049 if (!base->first)
1050 continue;
Thomas Gleixner3c8aa392007-02-16 01:27:50 -08001051
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -08001052 timer = rb_entry(base->first, struct hrtimer, node);
Arjan van de Vencc584b22008-09-01 15:02:30 -07001053 delta.tv64 = hrtimer_get_expires_tv64(timer);
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -08001054 delta = ktime_sub(delta, base->get_time());
1055 if (delta.tv64 < mindelta.tv64)
1056 mindelta.tv64 = delta.tv64;
1057 }
Tony Lindgren69239742006-03-06 15:42:45 -08001058 }
Thomas Gleixner3c8aa392007-02-16 01:27:50 -08001059
1060 spin_unlock_irqrestore(&cpu_base->lock, flags);
1061
Tony Lindgren69239742006-03-06 15:42:45 -08001062 if (mindelta.tv64 < 0)
1063 mindelta.tv64 = 0;
1064 return mindelta;
1065}
1066#endif
1067
Thomas Gleixner237fc6e2008-04-30 00:55:04 -07001068static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1069 enum hrtimer_mode mode)
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001070{
Thomas Gleixner3c8aa392007-02-16 01:27:50 -08001071 struct hrtimer_cpu_base *cpu_base;
George Anzinger7978672c2006-02-01 03:05:11 -08001072
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001073 memset(timer, 0, sizeof(struct hrtimer));
George Anzinger7978672c2006-02-01 03:05:11 -08001074
Thomas Gleixner3c8aa392007-02-16 01:27:50 -08001075 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
George Anzinger7978672c2006-02-01 03:05:11 -08001076
Thomas Gleixnerc9cb2e32007-02-16 01:27:49 -08001077 if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
George Anzinger7978672c2006-02-01 03:05:11 -08001078 clock_id = CLOCK_MONOTONIC;
1079
Thomas Gleixner3c8aa392007-02-16 01:27:50 -08001080 timer->base = &cpu_base->clock_base[clock_id];
Peter Zijlstrad3d74452008-01-25 21:08:31 +01001081 INIT_LIST_HEAD(&timer->cb_entry);
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -08001082 hrtimer_init_timer_hres(timer);
Ingo Molnar82f67cd2007-02-16 01:28:13 -08001083
1084#ifdef CONFIG_TIMER_STATS
1085 timer->start_site = NULL;
1086 timer->start_pid = -1;
1087 memset(timer->start_comm, 0, TASK_COMM_LEN);
1088#endif
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001089}
Thomas Gleixner237fc6e2008-04-30 00:55:04 -07001090
1091/**
1092 * hrtimer_init - initialize a timer to the given clock
1093 * @timer: the timer to be initialized
1094 * @clock_id: the clock to be used
1095 * @mode: timer mode abs/rel
1096 */
1097void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1098 enum hrtimer_mode mode)
1099{
1100 debug_hrtimer_init(timer);
1101 __hrtimer_init(timer, clock_id, mode);
1102}
Stephen Hemminger8d16b762006-05-30 21:26:09 -07001103EXPORT_SYMBOL_GPL(hrtimer_init);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001104
1105/**
1106 * hrtimer_get_res - get the timer resolution for a clock
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001107 * @which_clock: which clock to query
1108 * @tp: pointer to timespec variable to store the resolution
1109 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08001110 * Store the resolution of the clock selected by @which_clock in the
1111 * variable pointed to by @tp.
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001112 */
1113int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp)
1114{
Thomas Gleixner3c8aa392007-02-16 01:27:50 -08001115 struct hrtimer_cpu_base *cpu_base;
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001116
Thomas Gleixner3c8aa392007-02-16 01:27:50 -08001117 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1118 *tp = ktime_to_timespec(cpu_base->clock_base[which_clock].resolution);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001119
1120 return 0;
1121}
Stephen Hemminger8d16b762006-05-30 21:26:09 -07001122EXPORT_SYMBOL_GPL(hrtimer_get_res);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001123
Peter Zijlstrad3d74452008-01-25 21:08:31 +01001124static void __run_hrtimer(struct hrtimer *timer)
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001125{
Peter Zijlstrad3d74452008-01-25 21:08:31 +01001126 struct hrtimer_clock_base *base = timer->base;
1127 struct hrtimer_cpu_base *cpu_base = base->cpu_base;
1128 enum hrtimer_restart (*fn)(struct hrtimer *);
1129 int restart;
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001130
Peter Zijlstraca109492008-11-25 12:43:51 +01001131 WARN_ON(!irqs_disabled());
1132
Thomas Gleixner237fc6e2008-04-30 00:55:04 -07001133 debug_hrtimer_deactivate(timer);
Peter Zijlstrad3d74452008-01-25 21:08:31 +01001134 __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0);
1135 timer_stats_account_hrtimer(timer);
Peter Zijlstrad3d74452008-01-25 21:08:31 +01001136 fn = timer->function;
Peter Zijlstraca109492008-11-25 12:43:51 +01001137
1138 /*
1139 * Because we run timers from hardirq context, there is no chance
1140 * they get migrated to another cpu, therefore its safe to unlock
1141 * the timer base.
1142 */
1143 spin_unlock(&cpu_base->lock);
1144 restart = fn(timer);
1145 spin_lock(&cpu_base->lock);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001146
Peter Zijlstrad3d74452008-01-25 21:08:31 +01001147 /*
Thomas Gleixnere3f1d882009-01-05 11:28:23 +01001148 * Note: We clear the CALLBACK bit after enqueue_hrtimer and
1149 * we do not reprogramm the event hardware. Happens either in
1150 * hrtimer_start_range_ns() or in hrtimer_interrupt()
Peter Zijlstrad3d74452008-01-25 21:08:31 +01001151 */
1152 if (restart != HRTIMER_NORESTART) {
1153 BUG_ON(timer->state != HRTIMER_STATE_CALLBACK);
Peter Zijlstraa6037b62009-01-05 11:28:22 +01001154 enqueue_hrtimer(timer, base);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001155 }
Peter Zijlstrad3d74452008-01-25 21:08:31 +01001156 timer->state &= ~HRTIMER_STATE_CALLBACK;
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001157}
1158
Peter Zijlstrad3d74452008-01-25 21:08:31 +01001159#ifdef CONFIG_HIGH_RES_TIMERS
1160
1161/*
1162 * High resolution timer interrupt
1163 * Called with interrupts disabled
1164 */
1165void hrtimer_interrupt(struct clock_event_device *dev)
1166{
1167 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1168 struct hrtimer_clock_base *base;
1169 ktime_t expires_next, now;
Peter Zijlstraca109492008-11-25 12:43:51 +01001170 int i;
Peter Zijlstrad3d74452008-01-25 21:08:31 +01001171
1172 BUG_ON(!cpu_base->hres_active);
1173 cpu_base->nr_events++;
1174 dev->next_event.tv64 = KTIME_MAX;
1175
1176 retry:
1177 now = ktime_get();
1178
1179 expires_next.tv64 = KTIME_MAX;
1180
1181 base = cpu_base->clock_base;
1182
1183 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1184 ktime_t basenow;
1185 struct rb_node *node;
1186
1187 spin_lock(&cpu_base->lock);
1188
1189 basenow = ktime_add(now, base->offset);
1190
1191 while ((node = base->first)) {
1192 struct hrtimer *timer;
1193
1194 timer = rb_entry(node, struct hrtimer, node);
1195
Arjan van de Ven654c8e02008-09-01 15:47:08 -07001196 /*
1197 * The immediate goal for using the softexpires is
1198 * minimizing wakeups, not running timers at the
1199 * earliest interrupt after their soft expiration.
1200 * This allows us to avoid using a Priority Search
1201 * Tree, which can answer a stabbing querry for
1202 * overlapping intervals and instead use the simple
1203 * BST we already have.
1204 * We don't add extra wakeups by delaying timers that
1205 * are right-of a not yet expired timer, because that
1206 * timer will have to trigger a wakeup anyway.
1207 */
1208
1209 if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer)) {
Peter Zijlstrad3d74452008-01-25 21:08:31 +01001210 ktime_t expires;
1211
Arjan van de Vencc584b22008-09-01 15:02:30 -07001212 expires = ktime_sub(hrtimer_get_expires(timer),
Peter Zijlstrad3d74452008-01-25 21:08:31 +01001213 base->offset);
1214 if (expires.tv64 < expires_next.tv64)
1215 expires_next = expires;
1216 break;
1217 }
1218
Peter Zijlstrad3d74452008-01-25 21:08:31 +01001219 __run_hrtimer(timer);
1220 }
1221 spin_unlock(&cpu_base->lock);
1222 base++;
1223 }
1224
1225 cpu_base->expires_next = expires_next;
1226
1227 /* Reprogramming necessary ? */
1228 if (expires_next.tv64 != KTIME_MAX) {
1229 if (tick_program_event(expires_next, 0))
1230 goto retry;
1231 }
Peter Zijlstrad3d74452008-01-25 21:08:31 +01001232}
1233
Thomas Gleixner8bdec952009-01-05 11:28:19 +01001234/*
1235 * local version of hrtimer_peek_ahead_timers() called with interrupts
1236 * disabled.
1237 */
1238static void __hrtimer_peek_ahead_timers(void)
1239{
1240 struct tick_device *td;
1241
1242 if (!hrtimer_hres_active())
1243 return;
1244
1245 td = &__get_cpu_var(tick_cpu_device);
1246 if (td && td->evtdev)
1247 hrtimer_interrupt(td->evtdev);
1248}
1249
Arjan van de Ven2e94d1f2008-09-10 16:06:00 -07001250/**
1251 * hrtimer_peek_ahead_timers -- run soft-expired timers now
1252 *
1253 * hrtimer_peek_ahead_timers will peek at the timer queue of
1254 * the current cpu and check if there are any timers for which
1255 * the soft expires time has passed. If any such timers exist,
1256 * they are run immediately and then removed from the timer queue.
1257 *
1258 */
1259void hrtimer_peek_ahead_timers(void)
1260{
Thomas Gleixner643bdf62008-10-20 13:38:11 +02001261 unsigned long flags;
Arjan van de Vendc4304f2008-10-13 10:32:15 -04001262
Arjan van de Ven2e94d1f2008-09-10 16:06:00 -07001263 local_irq_save(flags);
Thomas Gleixner8bdec952009-01-05 11:28:19 +01001264 __hrtimer_peek_ahead_timers();
Arjan van de Ven2e94d1f2008-09-10 16:06:00 -07001265 local_irq_restore(flags);
1266}
1267
Peter Zijlstraa6037b62009-01-05 11:28:22 +01001268static void run_hrtimer_softirq(struct softirq_action *h)
1269{
1270 hrtimer_peek_ahead_timers();
1271}
1272
Ingo Molnar82c5b7b2009-01-05 14:11:10 +01001273#else /* CONFIG_HIGH_RES_TIMERS */
1274
1275static inline void __hrtimer_peek_ahead_timers(void) { }
1276
1277#endif /* !CONFIG_HIGH_RES_TIMERS */
Peter Zijlstrad3d74452008-01-25 21:08:31 +01001278
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001279/*
1280 * Called from timer softirq every jiffy, expire hrtimers:
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -08001281 *
1282 * For HRT its the fall back code to run the softirq in the timer
1283 * softirq context in case the hrtimer initialization failed or has
1284 * not been done yet.
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001285 */
Peter Zijlstrad3d74452008-01-25 21:08:31 +01001286void hrtimer_run_pending(void)
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001287{
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -08001288 if (hrtimer_hres_active())
1289 return;
1290
Thomas Gleixner79bf2bb2007-02-16 01:28:03 -08001291 /*
1292 * This _is_ ugly: We have to check in the softirq context,
1293 * whether we can switch to highres and / or nohz mode. The
1294 * clocksource switch happens in the timer interrupt with
1295 * xtime_lock held. Notification from there only sets the
1296 * check bit in the tick_oneshot code, otherwise we might
1297 * deadlock vs. xtime_lock.
1298 */
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -08001299 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
Peter Zijlstrad3d74452008-01-25 21:08:31 +01001300 hrtimer_switch_to_hres();
Peter Zijlstrad3d74452008-01-25 21:08:31 +01001301}
1302
1303/*
1304 * Called from hardirq context every jiffy
1305 */
Peter Zijlstrad3d74452008-01-25 21:08:31 +01001306void hrtimer_run_queues(void)
1307{
Dimitri Sivanich833883d2008-04-18 13:39:00 -07001308 struct rb_node *node;
Peter Zijlstrad3d74452008-01-25 21:08:31 +01001309 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
Dimitri Sivanich833883d2008-04-18 13:39:00 -07001310 struct hrtimer_clock_base *base;
1311 int index, gettime = 1;
Peter Zijlstrad3d74452008-01-25 21:08:31 +01001312
1313 if (hrtimer_hres_active())
1314 return;
Thomas Gleixner79bf2bb2007-02-16 01:28:03 -08001315
Dimitri Sivanich833883d2008-04-18 13:39:00 -07001316 for (index = 0; index < HRTIMER_MAX_CLOCK_BASES; index++) {
1317 base = &cpu_base->clock_base[index];
Thomas Gleixner92127c72006-03-26 01:38:05 -08001318
Dimitri Sivanich833883d2008-04-18 13:39:00 -07001319 if (!base->first)
1320 continue;
1321
Mark McLoughlind7cfb602008-09-19 13:13:44 +01001322 if (gettime) {
Dimitri Sivanich833883d2008-04-18 13:39:00 -07001323 hrtimer_get_softirq_time(cpu_base);
1324 gettime = 0;
1325 }
1326
Dimitri Sivanich833883d2008-04-18 13:39:00 -07001327 spin_lock(&cpu_base->lock);
1328
1329 while ((node = base->first)) {
1330 struct hrtimer *timer;
1331
1332 timer = rb_entry(node, struct hrtimer, node);
Arjan van de Vencc584b22008-09-01 15:02:30 -07001333 if (base->softirq_time.tv64 <=
1334 hrtimer_get_expires_tv64(timer))
Dimitri Sivanich833883d2008-04-18 13:39:00 -07001335 break;
1336
Dimitri Sivanich833883d2008-04-18 13:39:00 -07001337 __run_hrtimer(timer);
1338 }
1339 spin_unlock(&cpu_base->lock);
1340 }
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001341}
1342
1343/*
Thomas Gleixner10c94ec2006-01-09 20:52:35 -08001344 * Sleep related functions:
1345 */
Thomas Gleixnerc9cb2e32007-02-16 01:27:49 -08001346static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
Thomas Gleixner00362e32006-03-31 02:31:17 -08001347{
1348 struct hrtimer_sleeper *t =
1349 container_of(timer, struct hrtimer_sleeper, timer);
1350 struct task_struct *task = t->task;
1351
1352 t->task = NULL;
1353 if (task)
1354 wake_up_process(task);
1355
1356 return HRTIMER_NORESTART;
1357}
1358
Ingo Molnar36c8b582006-07-03 00:25:41 -07001359void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
Thomas Gleixner00362e32006-03-31 02:31:17 -08001360{
1361 sl->timer.function = hrtimer_wakeup;
1362 sl->task = task;
1363}
1364
Thomas Gleixner669d7862006-03-31 02:31:19 -08001365static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
Thomas Gleixner10c94ec2006-01-09 20:52:35 -08001366{
Thomas Gleixner669d7862006-03-31 02:31:19 -08001367 hrtimer_init_sleeper(t, current);
Thomas Gleixner10c94ec2006-01-09 20:52:35 -08001368
Roman Zippel432569b2006-03-26 01:38:08 -08001369 do {
1370 set_current_state(TASK_INTERRUPTIBLE);
Arjan van de Vencc584b22008-09-01 15:02:30 -07001371 hrtimer_start_expires(&t->timer, mode);
Peter Zijlstra37bb6cb2008-01-25 21:08:32 +01001372 if (!hrtimer_active(&t->timer))
1373 t->task = NULL;
Roman Zippel432569b2006-03-26 01:38:08 -08001374
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -08001375 if (likely(t->task))
1376 schedule();
Roman Zippel432569b2006-03-26 01:38:08 -08001377
Thomas Gleixner669d7862006-03-31 02:31:19 -08001378 hrtimer_cancel(&t->timer);
Thomas Gleixnerc9cb2e32007-02-16 01:27:49 -08001379 mode = HRTIMER_MODE_ABS;
Roman Zippel432569b2006-03-26 01:38:08 -08001380
Thomas Gleixner669d7862006-03-31 02:31:19 -08001381 } while (t->task && !signal_pending(current));
1382
Peter Zijlstra3588a082008-02-01 17:45:13 +01001383 __set_current_state(TASK_RUNNING);
1384
Thomas Gleixner669d7862006-03-31 02:31:19 -08001385 return t->task == NULL;
Thomas Gleixner10c94ec2006-01-09 20:52:35 -08001386}
1387
Oleg Nesterov080344b2008-02-01 17:29:05 +03001388static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp)
1389{
1390 struct timespec rmt;
1391 ktime_t rem;
1392
Arjan van de Vencc584b22008-09-01 15:02:30 -07001393 rem = hrtimer_expires_remaining(timer);
Oleg Nesterov080344b2008-02-01 17:29:05 +03001394 if (rem.tv64 <= 0)
1395 return 0;
1396 rmt = ktime_to_timespec(rem);
1397
1398 if (copy_to_user(rmtp, &rmt, sizeof(*rmtp)))
1399 return -EFAULT;
1400
1401 return 1;
1402}
1403
Toyo Abe1711ef32006-09-29 02:00:28 -07001404long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
Thomas Gleixner10c94ec2006-01-09 20:52:35 -08001405{
Thomas Gleixner669d7862006-03-31 02:31:19 -08001406 struct hrtimer_sleeper t;
Oleg Nesterov080344b2008-02-01 17:29:05 +03001407 struct timespec __user *rmtp;
Thomas Gleixner237fc6e2008-04-30 00:55:04 -07001408 int ret = 0;
Thomas Gleixner10c94ec2006-01-09 20:52:35 -08001409
Thomas Gleixner237fc6e2008-04-30 00:55:04 -07001410 hrtimer_init_on_stack(&t.timer, restart->nanosleep.index,
1411 HRTIMER_MODE_ABS);
Arjan van de Vencc584b22008-09-01 15:02:30 -07001412 hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
Thomas Gleixner10c94ec2006-01-09 20:52:35 -08001413
Thomas Gleixnerc9cb2e32007-02-16 01:27:49 -08001414 if (do_nanosleep(&t, HRTIMER_MODE_ABS))
Thomas Gleixner237fc6e2008-04-30 00:55:04 -07001415 goto out;
Thomas Gleixner10c94ec2006-01-09 20:52:35 -08001416
Thomas Gleixner029a07e2008-02-10 09:17:43 +01001417 rmtp = restart->nanosleep.rmtp;
Roman Zippel432569b2006-03-26 01:38:08 -08001418 if (rmtp) {
Thomas Gleixner237fc6e2008-04-30 00:55:04 -07001419 ret = update_rmtp(&t.timer, rmtp);
Oleg Nesterov080344b2008-02-01 17:29:05 +03001420 if (ret <= 0)
Thomas Gleixner237fc6e2008-04-30 00:55:04 -07001421 goto out;
Roman Zippel432569b2006-03-26 01:38:08 -08001422 }
Thomas Gleixner10c94ec2006-01-09 20:52:35 -08001423
Thomas Gleixner10c94ec2006-01-09 20:52:35 -08001424 /* The other values in restart are already filled in */
Thomas Gleixner237fc6e2008-04-30 00:55:04 -07001425 ret = -ERESTART_RESTARTBLOCK;
1426out:
1427 destroy_hrtimer_on_stack(&t.timer);
1428 return ret;
Thomas Gleixner10c94ec2006-01-09 20:52:35 -08001429}
1430
Oleg Nesterov080344b2008-02-01 17:29:05 +03001431long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
Thomas Gleixner10c94ec2006-01-09 20:52:35 -08001432 const enum hrtimer_mode mode, const clockid_t clockid)
1433{
1434 struct restart_block *restart;
Thomas Gleixner669d7862006-03-31 02:31:19 -08001435 struct hrtimer_sleeper t;
Thomas Gleixner237fc6e2008-04-30 00:55:04 -07001436 int ret = 0;
Arjan van de Ven3bd01202008-09-08 08:58:59 -07001437 unsigned long slack;
1438
1439 slack = current->timer_slack_ns;
1440 if (rt_task(current))
1441 slack = 0;
Thomas Gleixner10c94ec2006-01-09 20:52:35 -08001442
Thomas Gleixner237fc6e2008-04-30 00:55:04 -07001443 hrtimer_init_on_stack(&t.timer, clockid, mode);
Arjan van de Ven3bd01202008-09-08 08:58:59 -07001444 hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack);
Roman Zippel432569b2006-03-26 01:38:08 -08001445 if (do_nanosleep(&t, mode))
Thomas Gleixner237fc6e2008-04-30 00:55:04 -07001446 goto out;
Thomas Gleixner10c94ec2006-01-09 20:52:35 -08001447
George Anzinger7978672c2006-02-01 03:05:11 -08001448 /* Absolute timers do not update the rmtp value and restart: */
Thomas Gleixner237fc6e2008-04-30 00:55:04 -07001449 if (mode == HRTIMER_MODE_ABS) {
1450 ret = -ERESTARTNOHAND;
1451 goto out;
1452 }
Thomas Gleixner10c94ec2006-01-09 20:52:35 -08001453
Roman Zippel432569b2006-03-26 01:38:08 -08001454 if (rmtp) {
Thomas Gleixner237fc6e2008-04-30 00:55:04 -07001455 ret = update_rmtp(&t.timer, rmtp);
Oleg Nesterov080344b2008-02-01 17:29:05 +03001456 if (ret <= 0)
Thomas Gleixner237fc6e2008-04-30 00:55:04 -07001457 goto out;
Roman Zippel432569b2006-03-26 01:38:08 -08001458 }
Thomas Gleixner10c94ec2006-01-09 20:52:35 -08001459
1460 restart = &current_thread_info()->restart_block;
Toyo Abe1711ef32006-09-29 02:00:28 -07001461 restart->fn = hrtimer_nanosleep_restart;
Thomas Gleixner029a07e2008-02-10 09:17:43 +01001462 restart->nanosleep.index = t.timer.base->index;
1463 restart->nanosleep.rmtp = rmtp;
Arjan van de Vencc584b22008-09-01 15:02:30 -07001464 restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
Thomas Gleixner10c94ec2006-01-09 20:52:35 -08001465
Thomas Gleixner237fc6e2008-04-30 00:55:04 -07001466 ret = -ERESTART_RESTARTBLOCK;
1467out:
1468 destroy_hrtimer_on_stack(&t.timer);
1469 return ret;
Thomas Gleixner10c94ec2006-01-09 20:52:35 -08001470}
1471
Heiko Carstens58fd3aa2009-01-14 14:14:03 +01001472SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp,
1473 struct timespec __user *, rmtp)
Thomas Gleixner6ba1b912006-01-09 20:52:36 -08001474{
Oleg Nesterov080344b2008-02-01 17:29:05 +03001475 struct timespec tu;
Thomas Gleixner6ba1b912006-01-09 20:52:36 -08001476
1477 if (copy_from_user(&tu, rqtp, sizeof(tu)))
1478 return -EFAULT;
1479
1480 if (!timespec_valid(&tu))
1481 return -EINVAL;
1482
Oleg Nesterov080344b2008-02-01 17:29:05 +03001483 return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
Thomas Gleixner6ba1b912006-01-09 20:52:36 -08001484}
1485
Thomas Gleixner10c94ec2006-01-09 20:52:35 -08001486/*
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001487 * Functions related to boot-time initialization:
1488 */
Randy Dunlap0ec160d2008-01-21 17:18:24 -08001489static void __cpuinit init_hrtimers_cpu(int cpu)
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001490{
Thomas Gleixner3c8aa392007-02-16 01:27:50 -08001491 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001492 int i;
1493
Thomas Gleixner3c8aa392007-02-16 01:27:50 -08001494 spin_lock_init(&cpu_base->lock);
Thomas Gleixner3c8aa392007-02-16 01:27:50 -08001495
1496 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
1497 cpu_base->clock_base[i].cpu_base = cpu_base;
1498
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -08001499 hrtimer_init_hres(cpu_base);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001500}
1501
1502#ifdef CONFIG_HOTPLUG_CPU
1503
Peter Zijlstraca109492008-11-25 12:43:51 +01001504static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
Peter Zijlstra37810652008-12-04 11:17:10 +01001505 struct hrtimer_clock_base *new_base)
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001506{
1507 struct hrtimer *timer;
1508 struct rb_node *node;
1509
1510 while ((node = rb_first(&old_base->active))) {
1511 timer = rb_entry(node, struct hrtimer, node);
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -08001512 BUG_ON(hrtimer_callback_running(timer));
Thomas Gleixner237fc6e2008-04-30 00:55:04 -07001513 debug_hrtimer_deactivate(timer);
Thomas Gleixnerb00c1a92008-09-29 15:44:46 +02001514
1515 /*
1516 * Mark it as STATE_MIGRATE not INACTIVE otherwise the
1517 * timer could be seen as !active and just vanish away
1518 * under us on another CPU
1519 */
1520 __remove_hrtimer(timer, old_base, HRTIMER_STATE_MIGRATE, 0);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001521 timer->base = new_base;
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -08001522 /*
Thomas Gleixnere3f1d882009-01-05 11:28:23 +01001523 * Enqueue the timers on the new cpu. This does not
1524 * reprogram the event device in case the timer
1525 * expires before the earliest on this CPU, but we run
1526 * hrtimer_interrupt after we migrated everything to
1527 * sort out already expired timers and reprogram the
1528 * event device.
Thomas Gleixner54cdfdb2007-02-16 01:28:11 -08001529 */
Peter Zijlstraa6037b62009-01-05 11:28:22 +01001530 enqueue_hrtimer(timer, new_base);
Thomas Gleixner41e10222008-09-29 14:09:39 +02001531
Thomas Gleixnerb00c1a92008-09-29 15:44:46 +02001532 /* Clear the migration state bit */
1533 timer->state &= ~HRTIMER_STATE_MIGRATE;
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001534 }
1535}
1536
Thomas Gleixnerd5fd43c2009-01-05 11:28:20 +01001537static void migrate_hrtimers(int scpu)
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001538{
Thomas Gleixner3c8aa392007-02-16 01:27:50 -08001539 struct hrtimer_cpu_base *old_base, *new_base;
Thomas Gleixner731a55b2009-01-05 11:28:21 +01001540 int i;
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001541
Peter Zijlstra37810652008-12-04 11:17:10 +01001542 BUG_ON(cpu_online(scpu));
Peter Zijlstra37810652008-12-04 11:17:10 +01001543 tick_cancel_sched_timer(scpu);
Thomas Gleixner731a55b2009-01-05 11:28:21 +01001544
1545 local_irq_disable();
1546 old_base = &per_cpu(hrtimer_bases, scpu);
1547 new_base = &__get_cpu_var(hrtimer_bases);
Oleg Nesterovd82f0b02008-08-20 16:46:04 -07001548 /*
1549 * The caller is globally serialized and nobody else
1550 * takes two locks at once, deadlock is not possible.
1551 */
Thomas Gleixner731a55b2009-01-05 11:28:21 +01001552 spin_lock(&new_base->lock);
Oleg Nesterov8e60e052008-04-04 20:54:10 +02001553 spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001554
Thomas Gleixner3c8aa392007-02-16 01:27:50 -08001555 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
Peter Zijlstraca109492008-11-25 12:43:51 +01001556 migrate_hrtimer_list(&old_base->clock_base[i],
Peter Zijlstra37810652008-12-04 11:17:10 +01001557 &new_base->clock_base[i]);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001558 }
1559
Oleg Nesterov8e60e052008-04-04 20:54:10 +02001560 spin_unlock(&old_base->lock);
Thomas Gleixner731a55b2009-01-05 11:28:21 +01001561 spin_unlock(&new_base->lock);
Peter Zijlstra37810652008-12-04 11:17:10 +01001562
Thomas Gleixner731a55b2009-01-05 11:28:21 +01001563 /* Check, if we got expired work to do */
1564 __hrtimer_peek_ahead_timers();
1565 local_irq_enable();
Peter Zijlstra37810652008-12-04 11:17:10 +01001566}
1567
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001568#endif /* CONFIG_HOTPLUG_CPU */
1569
Chandra Seetharaman8c78f302006-07-30 03:03:35 -07001570static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self,
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001571 unsigned long action, void *hcpu)
1572{
Ingo Molnarb2e3c0a2008-12-19 00:48:27 +01001573 int scpu = (long)hcpu;
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001574
1575 switch (action) {
1576
1577 case CPU_UP_PREPARE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07001578 case CPU_UP_PREPARE_FROZEN:
Peter Zijlstra37810652008-12-04 11:17:10 +01001579 init_hrtimers_cpu(scpu);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001580 break;
1581
1582#ifdef CONFIG_HOTPLUG_CPU
1583 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07001584 case CPU_DEAD_FROZEN:
Ingo Molnarb2e3c0a2008-12-19 00:48:27 +01001585 {
Peter Zijlstra37810652008-12-04 11:17:10 +01001586 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &scpu);
Thomas Gleixnerd5fd43c2009-01-05 11:28:20 +01001587 migrate_hrtimers(scpu);
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001588 break;
Ingo Molnarb2e3c0a2008-12-19 00:48:27 +01001589 }
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001590#endif
1591
1592 default:
1593 break;
1594 }
1595
1596 return NOTIFY_OK;
1597}
1598
Chandra Seetharaman8c78f302006-07-30 03:03:35 -07001599static struct notifier_block __cpuinitdata hrtimers_nb = {
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001600 .notifier_call = hrtimer_cpu_notify,
1601};
1602
1603void __init hrtimers_init(void)
1604{
1605 hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
1606 (void *)(long)smp_processor_id());
1607 register_cpu_notifier(&hrtimers_nb);
Peter Zijlstraa6037b62009-01-05 11:28:22 +01001608#ifdef CONFIG_HIGH_RES_TIMERS
1609 open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq);
1610#endif
Thomas Gleixnerc0a31322006-01-09 20:52:32 -08001611}
1612
Arjan van de Ven7bb67432008-08-31 08:05:58 -07001613/**
Arjan van de Ven654c8e02008-09-01 15:47:08 -07001614 * schedule_hrtimeout_range - sleep until timeout
1615 * @expires: timeout value (ktime_t)
1616 * @delta: slack in expires timeout (ktime_t)
1617 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1618 *
1619 * Make the current task sleep until the given expiry time has
1620 * elapsed. The routine will return immediately unless
1621 * the current task state has been set (see set_current_state()).
1622 *
1623 * The @delta argument gives the kernel the freedom to schedule the
1624 * actual wakeup to a time that is both power and performance friendly.
1625 * The kernel give the normal best effort behavior for "@expires+@delta",
1626 * but may decide to fire the timer earlier, but no earlier than @expires.
1627 *
1628 * You can set the task state as follows -
1629 *
1630 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1631 * pass before the routine returns.
1632 *
1633 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1634 * delivered to the current task.
1635 *
1636 * The current task state is guaranteed to be TASK_RUNNING when this
1637 * routine returns.
1638 *
1639 * Returns 0 when the timer has expired otherwise -EINTR
1640 */
1641int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta,
1642 const enum hrtimer_mode mode)
1643{
1644 struct hrtimer_sleeper t;
1645
1646 /*
1647 * Optimize when a zero timeout value is given. It does not
1648 * matter whether this is an absolute or a relative time.
1649 */
1650 if (expires && !expires->tv64) {
1651 __set_current_state(TASK_RUNNING);
1652 return 0;
1653 }
1654
1655 /*
1656 * A NULL parameter means "inifinte"
1657 */
1658 if (!expires) {
1659 schedule();
1660 __set_current_state(TASK_RUNNING);
1661 return -EINTR;
1662 }
1663
1664 hrtimer_init_on_stack(&t.timer, CLOCK_MONOTONIC, mode);
1665 hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
1666
1667 hrtimer_init_sleeper(&t, current);
1668
1669 hrtimer_start_expires(&t.timer, mode);
1670 if (!hrtimer_active(&t.timer))
1671 t.task = NULL;
1672
1673 if (likely(t.task))
1674 schedule();
1675
1676 hrtimer_cancel(&t.timer);
1677 destroy_hrtimer_on_stack(&t.timer);
1678
1679 __set_current_state(TASK_RUNNING);
1680
1681 return !t.task ? 0 : -EINTR;
1682}
1683EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
1684
1685/**
Arjan van de Ven7bb67432008-08-31 08:05:58 -07001686 * schedule_hrtimeout - sleep until timeout
1687 * @expires: timeout value (ktime_t)
1688 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1689 *
1690 * Make the current task sleep until the given expiry time has
1691 * elapsed. The routine will return immediately unless
1692 * the current task state has been set (see set_current_state()).
1693 *
1694 * You can set the task state as follows -
1695 *
1696 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1697 * pass before the routine returns.
1698 *
1699 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1700 * delivered to the current task.
1701 *
1702 * The current task state is guaranteed to be TASK_RUNNING when this
1703 * routine returns.
1704 *
1705 * Returns 0 when the timer has expired otherwise -EINTR
1706 */
1707int __sched schedule_hrtimeout(ktime_t *expires,
1708 const enum hrtimer_mode mode)
1709{
Arjan van de Ven654c8e02008-09-01 15:47:08 -07001710 return schedule_hrtimeout_range(expires, 0, mode);
Arjan van de Ven7bb67432008-08-31 08:05:58 -07001711}
1712EXPORT_SYMBOL_GPL(schedule_hrtimeout);