[PATCH] sys_alarm() unsigned signed conversion fixup
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / kernel / timer.c
1 /*
2 * linux/kernel/timer.c
3 *
4 * Kernel internal timers, kernel timekeeping, basic process system calls
5 *
6 * Copyright (C) 1991, 1992 Linus Torvalds
7 *
8 * 1997-01-28 Modified by Finn Arne Gangstad to make timers scale better.
9 *
10 * 1997-09-10 Updated NTP code according to technical memorandum Jan '96
11 * "A Kernel Model for Precision Timekeeping" by Dave Mills
12 * 1998-12-24 Fixed a xtime SMP race (we need the xtime_lock rw spinlock to
13 * serialize accesses to xtime/lost_ticks).
14 * Copyright (C) 1998 Andrea Arcangeli
15 * 1999-03-10 Improved NTP compatibility by Ulrich Windl
16 * 2002-05-31 Move sys_sysinfo here and make its locking sane, Robert Love
17 * 2000-10-05 Implemented scalable SMP per-CPU timer handling.
18 * Copyright (C) 2000, 2001, 2002 Ingo Molnar
19 * Designed by David S. Miller, Alexey Kuznetsov and Ingo Molnar
20 */
21
22 #include <linux/kernel_stat.h>
23 #include <linux/module.h>
24 #include <linux/interrupt.h>
25 #include <linux/percpu.h>
26 #include <linux/init.h>
27 #include <linux/mm.h>
28 #include <linux/swap.h>
29 #include <linux/notifier.h>
30 #include <linux/thread_info.h>
31 #include <linux/time.h>
32 #include <linux/jiffies.h>
33 #include <linux/posix-timers.h>
34 #include <linux/cpu.h>
35 #include <linux/syscalls.h>
36 #include <linux/delay.h>
37
38 #include <asm/uaccess.h>
39 #include <asm/unistd.h>
40 #include <asm/div64.h>
41 #include <asm/timex.h>
42 #include <asm/io.h>
43
44 #ifdef CONFIG_TIME_INTERPOLATION
45 static void time_interpolator_update(long delta_nsec);
46 #else
47 #define time_interpolator_update(x)
48 #endif
49
50 u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
51
52 EXPORT_SYMBOL(jiffies_64);
53
54 /*
55 * per-CPU timer vector definitions:
56 */
57
58 #define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6)
59 #define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8)
60 #define TVN_SIZE (1 << TVN_BITS)
61 #define TVR_SIZE (1 << TVR_BITS)
62 #define TVN_MASK (TVN_SIZE - 1)
63 #define TVR_MASK (TVR_SIZE - 1)
64
65 struct timer_base_s {
66 spinlock_t lock;
67 struct timer_list *running_timer;
68 };
69
70 typedef struct tvec_s {
71 struct list_head vec[TVN_SIZE];
72 } tvec_t;
73
74 typedef struct tvec_root_s {
75 struct list_head vec[TVR_SIZE];
76 } tvec_root_t;
77
78 struct tvec_t_base_s {
79 struct timer_base_s t_base;
80 unsigned long timer_jiffies;
81 tvec_root_t tv1;
82 tvec_t tv2;
83 tvec_t tv3;
84 tvec_t tv4;
85 tvec_t tv5;
86 } ____cacheline_aligned_in_smp;
87
88 typedef struct tvec_t_base_s tvec_base_t;
89 static DEFINE_PER_CPU(tvec_base_t *, tvec_bases);
90 static tvec_base_t boot_tvec_bases;
91
92 static inline void set_running_timer(tvec_base_t *base,
93 struct timer_list *timer)
94 {
95 #ifdef CONFIG_SMP
96 base->t_base.running_timer = timer;
97 #endif
98 }
99
100 static void internal_add_timer(tvec_base_t *base, struct timer_list *timer)
101 {
102 unsigned long expires = timer->expires;
103 unsigned long idx = expires - base->timer_jiffies;
104 struct list_head *vec;
105
106 if (idx < TVR_SIZE) {
107 int i = expires & TVR_MASK;
108 vec = base->tv1.vec + i;
109 } else if (idx < 1 << (TVR_BITS + TVN_BITS)) {
110 int i = (expires >> TVR_BITS) & TVN_MASK;
111 vec = base->tv2.vec + i;
112 } else if (idx < 1 << (TVR_BITS + 2 * TVN_BITS)) {
113 int i = (expires >> (TVR_BITS + TVN_BITS)) & TVN_MASK;
114 vec = base->tv3.vec + i;
115 } else if (idx < 1 << (TVR_BITS + 3 * TVN_BITS)) {
116 int i = (expires >> (TVR_BITS + 2 * TVN_BITS)) & TVN_MASK;
117 vec = base->tv4.vec + i;
118 } else if ((signed long) idx < 0) {
119 /*
120 * Can happen if you add a timer with expires == jiffies,
121 * or you set a timer to go off in the past
122 */
123 vec = base->tv1.vec + (base->timer_jiffies & TVR_MASK);
124 } else {
125 int i;
126 /* If the timeout is larger than 0xffffffff on 64-bit
127 * architectures then we use the maximum timeout:
128 */
129 if (idx > 0xffffffffUL) {
130 idx = 0xffffffffUL;
131 expires = idx + base->timer_jiffies;
132 }
133 i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK;
134 vec = base->tv5.vec + i;
135 }
136 /*
137 * Timers are FIFO:
138 */
139 list_add_tail(&timer->entry, vec);
140 }
141
142 typedef struct timer_base_s timer_base_t;
143 /*
144 * Used by TIMER_INITIALIZER, we can't use per_cpu(tvec_bases)
145 * at compile time, and we need timer->base to lock the timer.
146 */
147 timer_base_t __init_timer_base
148 ____cacheline_aligned_in_smp = { .lock = SPIN_LOCK_UNLOCKED };
149 EXPORT_SYMBOL(__init_timer_base);
150
151 /***
152 * init_timer - initialize a timer.
153 * @timer: the timer to be initialized
154 *
155 * init_timer() must be done to a timer prior calling *any* of the
156 * other timer functions.
157 */
158 void fastcall init_timer(struct timer_list *timer)
159 {
160 timer->entry.next = NULL;
161 timer->base = &per_cpu(tvec_bases, raw_smp_processor_id())->t_base;
162 }
163 EXPORT_SYMBOL(init_timer);
164
165 static inline void detach_timer(struct timer_list *timer,
166 int clear_pending)
167 {
168 struct list_head *entry = &timer->entry;
169
170 __list_del(entry->prev, entry->next);
171 if (clear_pending)
172 entry->next = NULL;
173 entry->prev = LIST_POISON2;
174 }
175
176 /*
177 * We are using hashed locking: holding per_cpu(tvec_bases).t_base.lock
178 * means that all timers which are tied to this base via timer->base are
179 * locked, and the base itself is locked too.
180 *
181 * So __run_timers/migrate_timers can safely modify all timers which could
182 * be found on ->tvX lists.
183 *
184 * When the timer's base is locked, and the timer removed from list, it is
185 * possible to set timer->base = NULL and drop the lock: the timer remains
186 * locked.
187 */
188 static timer_base_t *lock_timer_base(struct timer_list *timer,
189 unsigned long *flags)
190 {
191 timer_base_t *base;
192
193 for (;;) {
194 base = timer->base;
195 if (likely(base != NULL)) {
196 spin_lock_irqsave(&base->lock, *flags);
197 if (likely(base == timer->base))
198 return base;
199 /* The timer has migrated to another CPU */
200 spin_unlock_irqrestore(&base->lock, *flags);
201 }
202 cpu_relax();
203 }
204 }
205
206 int __mod_timer(struct timer_list *timer, unsigned long expires)
207 {
208 timer_base_t *base;
209 tvec_base_t *new_base;
210 unsigned long flags;
211 int ret = 0;
212
213 BUG_ON(!timer->function);
214
215 base = lock_timer_base(timer, &flags);
216
217 if (timer_pending(timer)) {
218 detach_timer(timer, 0);
219 ret = 1;
220 }
221
222 new_base = __get_cpu_var(tvec_bases);
223
224 if (base != &new_base->t_base) {
225 /*
226 * We are trying to schedule the timer on the local CPU.
227 * However we can't change timer's base while it is running,
228 * otherwise del_timer_sync() can't detect that the timer's
229 * handler yet has not finished. This also guarantees that
230 * the timer is serialized wrt itself.
231 */
232 if (unlikely(base->running_timer == timer)) {
233 /* The timer remains on a former base */
234 new_base = container_of(base, tvec_base_t, t_base);
235 } else {
236 /* See the comment in lock_timer_base() */
237 timer->base = NULL;
238 spin_unlock(&base->lock);
239 spin_lock(&new_base->t_base.lock);
240 timer->base = &new_base->t_base;
241 }
242 }
243
244 timer->expires = expires;
245 internal_add_timer(new_base, timer);
246 spin_unlock_irqrestore(&new_base->t_base.lock, flags);
247
248 return ret;
249 }
250
251 EXPORT_SYMBOL(__mod_timer);
252
253 /***
254 * add_timer_on - start a timer on a particular CPU
255 * @timer: the timer to be added
256 * @cpu: the CPU to start it on
257 *
258 * This is not very scalable on SMP. Double adds are not possible.
259 */
260 void add_timer_on(struct timer_list *timer, int cpu)
261 {
262 tvec_base_t *base = per_cpu(tvec_bases, cpu);
263 unsigned long flags;
264
265 BUG_ON(timer_pending(timer) || !timer->function);
266 spin_lock_irqsave(&base->t_base.lock, flags);
267 timer->base = &base->t_base;
268 internal_add_timer(base, timer);
269 spin_unlock_irqrestore(&base->t_base.lock, flags);
270 }
271
272
273 /***
274 * mod_timer - modify a timer's timeout
275 * @timer: the timer to be modified
276 *
277 * mod_timer is a more efficient way to update the expire field of an
278 * active timer (if the timer is inactive it will be activated)
279 *
280 * mod_timer(timer, expires) is equivalent to:
281 *
282 * del_timer(timer); timer->expires = expires; add_timer(timer);
283 *
284 * Note that if there are multiple unserialized concurrent users of the
285 * same timer, then mod_timer() is the only safe way to modify the timeout,
286 * since add_timer() cannot modify an already running timer.
287 *
288 * The function returns whether it has modified a pending timer or not.
289 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
290 * active timer returns 1.)
291 */
292 int mod_timer(struct timer_list *timer, unsigned long expires)
293 {
294 BUG_ON(!timer->function);
295
296 /*
297 * This is a common optimization triggered by the
298 * networking code - if the timer is re-modified
299 * to be the same thing then just return:
300 */
301 if (timer->expires == expires && timer_pending(timer))
302 return 1;
303
304 return __mod_timer(timer, expires);
305 }
306
307 EXPORT_SYMBOL(mod_timer);
308
309 /***
310 * del_timer - deactive a timer.
311 * @timer: the timer to be deactivated
312 *
313 * del_timer() deactivates a timer - this works on both active and inactive
314 * timers.
315 *
316 * The function returns whether it has deactivated a pending timer or not.
317 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
318 * active timer returns 1.)
319 */
320 int del_timer(struct timer_list *timer)
321 {
322 timer_base_t *base;
323 unsigned long flags;
324 int ret = 0;
325
326 if (timer_pending(timer)) {
327 base = lock_timer_base(timer, &flags);
328 if (timer_pending(timer)) {
329 detach_timer(timer, 1);
330 ret = 1;
331 }
332 spin_unlock_irqrestore(&base->lock, flags);
333 }
334
335 return ret;
336 }
337
338 EXPORT_SYMBOL(del_timer);
339
340 #ifdef CONFIG_SMP
341 /*
342 * This function tries to deactivate a timer. Upon successful (ret >= 0)
343 * exit the timer is not queued and the handler is not running on any CPU.
344 *
345 * It must not be called from interrupt contexts.
346 */
347 int try_to_del_timer_sync(struct timer_list *timer)
348 {
349 timer_base_t *base;
350 unsigned long flags;
351 int ret = -1;
352
353 base = lock_timer_base(timer, &flags);
354
355 if (base->running_timer == timer)
356 goto out;
357
358 ret = 0;
359 if (timer_pending(timer)) {
360 detach_timer(timer, 1);
361 ret = 1;
362 }
363 out:
364 spin_unlock_irqrestore(&base->lock, flags);
365
366 return ret;
367 }
368
369 /***
370 * del_timer_sync - deactivate a timer and wait for the handler to finish.
371 * @timer: the timer to be deactivated
372 *
373 * This function only differs from del_timer() on SMP: besides deactivating
374 * the timer it also makes sure the handler has finished executing on other
375 * CPUs.
376 *
377 * Synchronization rules: callers must prevent restarting of the timer,
378 * otherwise this function is meaningless. It must not be called from
379 * interrupt contexts. The caller must not hold locks which would prevent
380 * completion of the timer's handler. The timer's handler must not call
381 * add_timer_on(). Upon exit the timer is not queued and the handler is
382 * not running on any CPU.
383 *
384 * The function returns whether it has deactivated a pending timer or not.
385 */
386 int del_timer_sync(struct timer_list *timer)
387 {
388 for (;;) {
389 int ret = try_to_del_timer_sync(timer);
390 if (ret >= 0)
391 return ret;
392 }
393 }
394
395 EXPORT_SYMBOL(del_timer_sync);
396 #endif
397
398 static int cascade(tvec_base_t *base, tvec_t *tv, int index)
399 {
400 /* cascade all the timers from tv up one level */
401 struct list_head *head, *curr;
402
403 head = tv->vec + index;
404 curr = head->next;
405 /*
406 * We are removing _all_ timers from the list, so we don't have to
407 * detach them individually, just clear the list afterwards.
408 */
409 while (curr != head) {
410 struct timer_list *tmp;
411
412 tmp = list_entry(curr, struct timer_list, entry);
413 BUG_ON(tmp->base != &base->t_base);
414 curr = curr->next;
415 internal_add_timer(base, tmp);
416 }
417 INIT_LIST_HEAD(head);
418
419 return index;
420 }
421
422 /***
423 * __run_timers - run all expired timers (if any) on this CPU.
424 * @base: the timer vector to be processed.
425 *
426 * This function cascades all vectors and executes all expired timer
427 * vectors.
428 */
429 #define INDEX(N) (base->timer_jiffies >> (TVR_BITS + N * TVN_BITS)) & TVN_MASK
430
431 static inline void __run_timers(tvec_base_t *base)
432 {
433 struct timer_list *timer;
434
435 spin_lock_irq(&base->t_base.lock);
436 while (time_after_eq(jiffies, base->timer_jiffies)) {
437 struct list_head work_list = LIST_HEAD_INIT(work_list);
438 struct list_head *head = &work_list;
439 int index = base->timer_jiffies & TVR_MASK;
440
441 /*
442 * Cascade timers:
443 */
444 if (!index &&
445 (!cascade(base, &base->tv2, INDEX(0))) &&
446 (!cascade(base, &base->tv3, INDEX(1))) &&
447 !cascade(base, &base->tv4, INDEX(2)))
448 cascade(base, &base->tv5, INDEX(3));
449 ++base->timer_jiffies;
450 list_splice_init(base->tv1.vec + index, &work_list);
451 while (!list_empty(head)) {
452 void (*fn)(unsigned long);
453 unsigned long data;
454
455 timer = list_entry(head->next,struct timer_list,entry);
456 fn = timer->function;
457 data = timer->data;
458
459 set_running_timer(base, timer);
460 detach_timer(timer, 1);
461 spin_unlock_irq(&base->t_base.lock);
462 {
463 int preempt_count = preempt_count();
464 fn(data);
465 if (preempt_count != preempt_count()) {
466 printk(KERN_WARNING "huh, entered %p "
467 "with preempt_count %08x, exited"
468 " with %08x?\n",
469 fn, preempt_count,
470 preempt_count());
471 BUG();
472 }
473 }
474 spin_lock_irq(&base->t_base.lock);
475 }
476 }
477 set_running_timer(base, NULL);
478 spin_unlock_irq(&base->t_base.lock);
479 }
480
481 #ifdef CONFIG_NO_IDLE_HZ
482 /*
483 * Find out when the next timer event is due to happen. This
484 * is used on S/390 to stop all activity when a cpus is idle.
485 * This functions needs to be called disabled.
486 */
487 unsigned long next_timer_interrupt(void)
488 {
489 tvec_base_t *base;
490 struct list_head *list;
491 struct timer_list *nte;
492 unsigned long expires;
493 unsigned long hr_expires = MAX_JIFFY_OFFSET;
494 ktime_t hr_delta;
495 tvec_t *varray[4];
496 int i, j;
497
498 hr_delta = hrtimer_get_next_event();
499 if (hr_delta.tv64 != KTIME_MAX) {
500 struct timespec tsdelta;
501 tsdelta = ktime_to_timespec(hr_delta);
502 hr_expires = timespec_to_jiffies(&tsdelta);
503 if (hr_expires < 3)
504 return hr_expires + jiffies;
505 }
506 hr_expires += jiffies;
507
508 base = __get_cpu_var(tvec_bases);
509 spin_lock(&base->t_base.lock);
510 expires = base->timer_jiffies + (LONG_MAX >> 1);
511 list = NULL;
512
513 /* Look for timer events in tv1. */
514 j = base->timer_jiffies & TVR_MASK;
515 do {
516 list_for_each_entry(nte, base->tv1.vec + j, entry) {
517 expires = nte->expires;
518 if (j < (base->timer_jiffies & TVR_MASK))
519 list = base->tv2.vec + (INDEX(0));
520 goto found;
521 }
522 j = (j + 1) & TVR_MASK;
523 } while (j != (base->timer_jiffies & TVR_MASK));
524
525 /* Check tv2-tv5. */
526 varray[0] = &base->tv2;
527 varray[1] = &base->tv3;
528 varray[2] = &base->tv4;
529 varray[3] = &base->tv5;
530 for (i = 0; i < 4; i++) {
531 j = INDEX(i);
532 do {
533 if (list_empty(varray[i]->vec + j)) {
534 j = (j + 1) & TVN_MASK;
535 continue;
536 }
537 list_for_each_entry(nte, varray[i]->vec + j, entry)
538 if (time_before(nte->expires, expires))
539 expires = nte->expires;
540 if (j < (INDEX(i)) && i < 3)
541 list = varray[i + 1]->vec + (INDEX(i + 1));
542 goto found;
543 } while (j != (INDEX(i)));
544 }
545 found:
546 if (list) {
547 /*
548 * The search wrapped. We need to look at the next list
549 * from next tv element that would cascade into tv element
550 * where we found the timer element.
551 */
552 list_for_each_entry(nte, list, entry) {
553 if (time_before(nte->expires, expires))
554 expires = nte->expires;
555 }
556 }
557 spin_unlock(&base->t_base.lock);
558
559 if (time_before(hr_expires, expires))
560 return hr_expires;
561
562 return expires;
563 }
564 #endif
565
566 /******************************************************************/
567
568 /*
569 * Timekeeping variables
570 */
571 unsigned long tick_usec = TICK_USEC; /* USER_HZ period (usec) */
572 unsigned long tick_nsec = TICK_NSEC; /* ACTHZ period (nsec) */
573
574 /*
575 * The current time
576 * wall_to_monotonic is what we need to add to xtime (or xtime corrected
577 * for sub jiffie times) to get to monotonic time. Monotonic is pegged
578 * at zero at system boot time, so wall_to_monotonic will be negative,
579 * however, we will ALWAYS keep the tv_nsec part positive so we can use
580 * the usual normalization.
581 */
582 struct timespec xtime __attribute__ ((aligned (16)));
583 struct timespec wall_to_monotonic __attribute__ ((aligned (16)));
584
585 EXPORT_SYMBOL(xtime);
586
587 /* Don't completely fail for HZ > 500. */
588 int tickadj = 500/HZ ? : 1; /* microsecs */
589
590
591 /*
592 * phase-lock loop variables
593 */
594 /* TIME_ERROR prevents overwriting the CMOS clock */
595 int time_state = TIME_OK; /* clock synchronization status */
596 int time_status = STA_UNSYNC; /* clock status bits */
597 long time_offset; /* time adjustment (us) */
598 long time_constant = 2; /* pll time constant */
599 long time_tolerance = MAXFREQ; /* frequency tolerance (ppm) */
600 long time_precision = 1; /* clock precision (us) */
601 long time_maxerror = NTP_PHASE_LIMIT; /* maximum error (us) */
602 long time_esterror = NTP_PHASE_LIMIT; /* estimated error (us) */
603 static long time_phase; /* phase offset (scaled us) */
604 long time_freq = (((NSEC_PER_SEC + HZ/2) % HZ - HZ/2) << SHIFT_USEC) / NSEC_PER_USEC;
605 /* frequency offset (scaled ppm)*/
606 static long time_adj; /* tick adjust (scaled 1 / HZ) */
607 long time_reftime; /* time at last adjustment (s) */
608 long time_adjust;
609 long time_next_adjust;
610
611 /*
612 * this routine handles the overflow of the microsecond field
613 *
614 * The tricky bits of code to handle the accurate clock support
615 * were provided by Dave Mills (Mills@UDEL.EDU) of NTP fame.
616 * They were originally developed for SUN and DEC kernels.
617 * All the kudos should go to Dave for this stuff.
618 *
619 */
620 static void second_overflow(void)
621 {
622 long ltemp;
623
624 /* Bump the maxerror field */
625 time_maxerror += time_tolerance >> SHIFT_USEC;
626 if (time_maxerror > NTP_PHASE_LIMIT) {
627 time_maxerror = NTP_PHASE_LIMIT;
628 time_status |= STA_UNSYNC;
629 }
630
631 /*
632 * Leap second processing. If in leap-insert state at the end of the
633 * day, the system clock is set back one second; if in leap-delete
634 * state, the system clock is set ahead one second. The microtime()
635 * routine or external clock driver will insure that reported time is
636 * always monotonic. The ugly divides should be replaced.
637 */
638 switch (time_state) {
639 case TIME_OK:
640 if (time_status & STA_INS)
641 time_state = TIME_INS;
642 else if (time_status & STA_DEL)
643 time_state = TIME_DEL;
644 break;
645 case TIME_INS:
646 if (xtime.tv_sec % 86400 == 0) {
647 xtime.tv_sec--;
648 wall_to_monotonic.tv_sec++;
649 /*
650 * The timer interpolator will make time change
651 * gradually instead of an immediate jump by one second
652 */
653 time_interpolator_update(-NSEC_PER_SEC);
654 time_state = TIME_OOP;
655 clock_was_set();
656 printk(KERN_NOTICE "Clock: inserting leap second "
657 "23:59:60 UTC\n");
658 }
659 break;
660 case TIME_DEL:
661 if ((xtime.tv_sec + 1) % 86400 == 0) {
662 xtime.tv_sec++;
663 wall_to_monotonic.tv_sec--;
664 /*
665 * Use of time interpolator for a gradual change of
666 * time
667 */
668 time_interpolator_update(NSEC_PER_SEC);
669 time_state = TIME_WAIT;
670 clock_was_set();
671 printk(KERN_NOTICE "Clock: deleting leap second "
672 "23:59:59 UTC\n");
673 }
674 break;
675 case TIME_OOP:
676 time_state = TIME_WAIT;
677 break;
678 case TIME_WAIT:
679 if (!(time_status & (STA_INS | STA_DEL)))
680 time_state = TIME_OK;
681 }
682
683 /*
684 * Compute the phase adjustment for the next second. In PLL mode, the
685 * offset is reduced by a fixed factor times the time constant. In FLL
686 * mode the offset is used directly. In either mode, the maximum phase
687 * adjustment for each second is clamped so as to spread the adjustment
688 * over not more than the number of seconds between updates.
689 */
690 ltemp = time_offset;
691 if (!(time_status & STA_FLL))
692 ltemp = shift_right(ltemp, SHIFT_KG + time_constant);
693 ltemp = min(ltemp, (MAXPHASE / MINSEC) << SHIFT_UPDATE);
694 ltemp = max(ltemp, -(MAXPHASE / MINSEC) << SHIFT_UPDATE);
695 time_offset -= ltemp;
696 time_adj = ltemp << (SHIFT_SCALE - SHIFT_HZ - SHIFT_UPDATE);
697
698 /*
699 * Compute the frequency estimate and additional phase adjustment due
700 * to frequency error for the next second. When the PPS signal is
701 * engaged, gnaw on the watchdog counter and update the frequency
702 * computed by the pll and the PPS signal.
703 */
704 pps_valid++;
705 if (pps_valid == PPS_VALID) { /* PPS signal lost */
706 pps_jitter = MAXTIME;
707 pps_stabil = MAXFREQ;
708 time_status &= ~(STA_PPSSIGNAL | STA_PPSJITTER |
709 STA_PPSWANDER | STA_PPSERROR);
710 }
711 ltemp = time_freq + pps_freq;
712 time_adj += shift_right(ltemp,(SHIFT_USEC + SHIFT_HZ - SHIFT_SCALE));
713
714 #if HZ == 100
715 /*
716 * Compensate for (HZ==100) != (1 << SHIFT_HZ). Add 25% and 3.125% to
717 * get 128.125; => only 0.125% error (p. 14)
718 */
719 time_adj += shift_right(time_adj, 2) + shift_right(time_adj, 5);
720 #endif
721 #if HZ == 250
722 /*
723 * Compensate for (HZ==250) != (1 << SHIFT_HZ). Add 1.5625% and
724 * 0.78125% to get 255.85938; => only 0.05% error (p. 14)
725 */
726 time_adj += shift_right(time_adj, 6) + shift_right(time_adj, 7);
727 #endif
728 #if HZ == 1000
729 /*
730 * Compensate for (HZ==1000) != (1 << SHIFT_HZ). Add 1.5625% and
731 * 0.78125% to get 1023.4375; => only 0.05% error (p. 14)
732 */
733 time_adj += shift_right(time_adj, 6) + shift_right(time_adj, 7);
734 #endif
735 }
736
737 /*
738 * Returns how many microseconds we need to add to xtime this tick
739 * in doing an adjustment requested with adjtime.
740 */
741 static long adjtime_adjustment(void)
742 {
743 long time_adjust_step;
744
745 time_adjust_step = time_adjust;
746 if (time_adjust_step) {
747 /*
748 * We are doing an adjtime thing. Prepare time_adjust_step to
749 * be within bounds. Note that a positive time_adjust means we
750 * want the clock to run faster.
751 *
752 * Limit the amount of the step to be in the range
753 * -tickadj .. +tickadj
754 */
755 time_adjust_step = min(time_adjust_step, (long)tickadj);
756 time_adjust_step = max(time_adjust_step, (long)-tickadj);
757 }
758 return time_adjust_step;
759 }
760
761 /* in the NTP reference this is called "hardclock()" */
762 static void update_wall_time_one_tick(void)
763 {
764 long time_adjust_step, delta_nsec;
765
766 time_adjust_step = adjtime_adjustment();
767 if (time_adjust_step)
768 /* Reduce by this step the amount of time left */
769 time_adjust -= time_adjust_step;
770 delta_nsec = tick_nsec + time_adjust_step * 1000;
771 /*
772 * Advance the phase, once it gets to one microsecond, then
773 * advance the tick more.
774 */
775 time_phase += time_adj;
776 if ((time_phase >= FINENSEC) || (time_phase <= -FINENSEC)) {
777 long ltemp = shift_right(time_phase, (SHIFT_SCALE - 10));
778 time_phase -= ltemp << (SHIFT_SCALE - 10);
779 delta_nsec += ltemp;
780 }
781 xtime.tv_nsec += delta_nsec;
782 time_interpolator_update(delta_nsec);
783
784 /* Changes by adjtime() do not take effect till next tick. */
785 if (time_next_adjust != 0) {
786 time_adjust = time_next_adjust;
787 time_next_adjust = 0;
788 }
789 }
790
791 /*
792 * Return how long ticks are at the moment, that is, how much time
793 * update_wall_time_one_tick will add to xtime next time we call it
794 * (assuming no calls to do_adjtimex in the meantime).
795 * The return value is in fixed-point nanoseconds with SHIFT_SCALE-10
796 * bits to the right of the binary point.
797 * This function has no side-effects.
798 */
799 u64 current_tick_length(void)
800 {
801 long delta_nsec;
802
803 delta_nsec = tick_nsec + adjtime_adjustment() * 1000;
804 return ((u64) delta_nsec << (SHIFT_SCALE - 10)) + time_adj;
805 }
806
807 /*
808 * Using a loop looks inefficient, but "ticks" is
809 * usually just one (we shouldn't be losing ticks,
810 * we're doing this this way mainly for interrupt
811 * latency reasons, not because we think we'll
812 * have lots of lost timer ticks
813 */
814 static void update_wall_time(unsigned long ticks)
815 {
816 do {
817 ticks--;
818 update_wall_time_one_tick();
819 if (xtime.tv_nsec >= 1000000000) {
820 xtime.tv_nsec -= 1000000000;
821 xtime.tv_sec++;
822 second_overflow();
823 }
824 } while (ticks);
825 }
826
827 /*
828 * Called from the timer interrupt handler to charge one tick to the current
829 * process. user_tick is 1 if the tick is user time, 0 for system.
830 */
831 void update_process_times(int user_tick)
832 {
833 struct task_struct *p = current;
834 int cpu = smp_processor_id();
835
836 /* Note: this timer irq context must be accounted for as well. */
837 if (user_tick)
838 account_user_time(p, jiffies_to_cputime(1));
839 else
840 account_system_time(p, HARDIRQ_OFFSET, jiffies_to_cputime(1));
841 run_local_timers();
842 if (rcu_pending(cpu))
843 rcu_check_callbacks(cpu, user_tick);
844 scheduler_tick();
845 run_posix_cpu_timers(p);
846 }
847
848 /*
849 * Nr of active tasks - counted in fixed-point numbers
850 */
851 static unsigned long count_active_tasks(void)
852 {
853 return (nr_running() + nr_uninterruptible()) * FIXED_1;
854 }
855
856 /*
857 * Hmm.. Changed this, as the GNU make sources (load.c) seems to
858 * imply that avenrun[] is the standard name for this kind of thing.
859 * Nothing else seems to be standardized: the fractional size etc
860 * all seem to differ on different machines.
861 *
862 * Requires xtime_lock to access.
863 */
864 unsigned long avenrun[3];
865
866 EXPORT_SYMBOL(avenrun);
867
868 /*
869 * calc_load - given tick count, update the avenrun load estimates.
870 * This is called while holding a write_lock on xtime_lock.
871 */
872 static inline void calc_load(unsigned long ticks)
873 {
874 unsigned long active_tasks; /* fixed-point */
875 static int count = LOAD_FREQ;
876
877 count -= ticks;
878 if (count < 0) {
879 count += LOAD_FREQ;
880 active_tasks = count_active_tasks();
881 CALC_LOAD(avenrun[0], EXP_1, active_tasks);
882 CALC_LOAD(avenrun[1], EXP_5, active_tasks);
883 CALC_LOAD(avenrun[2], EXP_15, active_tasks);
884 }
885 }
886
887 /* jiffies at the most recent update of wall time */
888 unsigned long wall_jiffies = INITIAL_JIFFIES;
889
890 /*
891 * This read-write spinlock protects us from races in SMP while
892 * playing with xtime and avenrun.
893 */
894 #ifndef ARCH_HAVE_XTIME_LOCK
895 seqlock_t xtime_lock __cacheline_aligned_in_smp = SEQLOCK_UNLOCKED;
896
897 EXPORT_SYMBOL(xtime_lock);
898 #endif
899
900 /*
901 * This function runs timers and the timer-tq in bottom half context.
902 */
903 static void run_timer_softirq(struct softirq_action *h)
904 {
905 tvec_base_t *base = __get_cpu_var(tvec_bases);
906
907 hrtimer_run_queues();
908 if (time_after_eq(jiffies, base->timer_jiffies))
909 __run_timers(base);
910 }
911
912 /*
913 * Called by the local, per-CPU timer interrupt on SMP.
914 */
915 void run_local_timers(void)
916 {
917 raise_softirq(TIMER_SOFTIRQ);
918 softlockup_tick();
919 }
920
921 /*
922 * Called by the timer interrupt. xtime_lock must already be taken
923 * by the timer IRQ!
924 */
925 static inline void update_times(void)
926 {
927 unsigned long ticks;
928
929 ticks = jiffies - wall_jiffies;
930 if (ticks) {
931 wall_jiffies += ticks;
932 update_wall_time(ticks);
933 }
934 calc_load(ticks);
935 }
936
937 /*
938 * The 64-bit jiffies value is not atomic - you MUST NOT read it
939 * without sampling the sequence number in xtime_lock.
940 * jiffies is defined in the linker script...
941 */
942
943 void do_timer(struct pt_regs *regs)
944 {
945 jiffies_64++;
946 /* prevent loading jiffies before storing new jiffies_64 value. */
947 barrier();
948 update_times();
949 }
950
951 #ifdef __ARCH_WANT_SYS_ALARM
952
953 /*
954 * For backwards compatibility? This can be done in libc so Alpha
955 * and all newer ports shouldn't need it.
956 */
957 asmlinkage unsigned long sys_alarm(unsigned int seconds)
958 {
959 return alarm_setitimer(seconds);
960 }
961
962 #endif
963
964 #ifndef __alpha__
965
966 /*
967 * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this
968 * should be moved into arch/i386 instead?
969 */
970
971 /**
972 * sys_getpid - return the thread group id of the current process
973 *
974 * Note, despite the name, this returns the tgid not the pid. The tgid and
975 * the pid are identical unless CLONE_THREAD was specified on clone() in
976 * which case the tgid is the same in all threads of the same group.
977 *
978 * This is SMP safe as current->tgid does not change.
979 */
980 asmlinkage long sys_getpid(void)
981 {
982 return current->tgid;
983 }
984
985 /*
986 * Accessing ->group_leader->real_parent is not SMP-safe, it could
987 * change from under us. However, rather than getting any lock
988 * we can use an optimistic algorithm: get the parent
989 * pid, and go back and check that the parent is still
990 * the same. If it has changed (which is extremely unlikely
991 * indeed), we just try again..
992 *
993 * NOTE! This depends on the fact that even if we _do_
994 * get an old value of "parent", we can happily dereference
995 * the pointer (it was and remains a dereferencable kernel pointer
996 * no matter what): we just can't necessarily trust the result
997 * until we know that the parent pointer is valid.
998 *
999 * NOTE2: ->group_leader never changes from under us.
1000 */
1001 asmlinkage long sys_getppid(void)
1002 {
1003 int pid;
1004 struct task_struct *me = current;
1005 struct task_struct *parent;
1006
1007 parent = me->group_leader->real_parent;
1008 for (;;) {
1009 pid = parent->tgid;
1010 #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1011 {
1012 struct task_struct *old = parent;
1013
1014 /*
1015 * Make sure we read the pid before re-reading the
1016 * parent pointer:
1017 */
1018 smp_rmb();
1019 parent = me->group_leader->real_parent;
1020 if (old != parent)
1021 continue;
1022 }
1023 #endif
1024 break;
1025 }
1026 return pid;
1027 }
1028
1029 asmlinkage long sys_getuid(void)
1030 {
1031 /* Only we change this so SMP safe */
1032 return current->uid;
1033 }
1034
1035 asmlinkage long sys_geteuid(void)
1036 {
1037 /* Only we change this so SMP safe */
1038 return current->euid;
1039 }
1040
1041 asmlinkage long sys_getgid(void)
1042 {
1043 /* Only we change this so SMP safe */
1044 return current->gid;
1045 }
1046
1047 asmlinkage long sys_getegid(void)
1048 {
1049 /* Only we change this so SMP safe */
1050 return current->egid;
1051 }
1052
1053 #endif
1054
1055 static void process_timeout(unsigned long __data)
1056 {
1057 wake_up_process((task_t *)__data);
1058 }
1059
1060 /**
1061 * schedule_timeout - sleep until timeout
1062 * @timeout: timeout value in jiffies
1063 *
1064 * Make the current task sleep until @timeout jiffies have
1065 * elapsed. The routine will return immediately unless
1066 * the current task state has been set (see set_current_state()).
1067 *
1068 * You can set the task state as follows -
1069 *
1070 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1071 * pass before the routine returns. The routine will return 0
1072 *
1073 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1074 * delivered to the current task. In this case the remaining time
1075 * in jiffies will be returned, or 0 if the timer expired in time
1076 *
1077 * The current task state is guaranteed to be TASK_RUNNING when this
1078 * routine returns.
1079 *
1080 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1081 * the CPU away without a bound on the timeout. In this case the return
1082 * value will be %MAX_SCHEDULE_TIMEOUT.
1083 *
1084 * In all cases the return value is guaranteed to be non-negative.
1085 */
1086 fastcall signed long __sched schedule_timeout(signed long timeout)
1087 {
1088 struct timer_list timer;
1089 unsigned long expire;
1090
1091 switch (timeout)
1092 {
1093 case MAX_SCHEDULE_TIMEOUT:
1094 /*
1095 * These two special cases are useful to be comfortable
1096 * in the caller. Nothing more. We could take
1097 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1098 * but I' d like to return a valid offset (>=0) to allow
1099 * the caller to do everything it want with the retval.
1100 */
1101 schedule();
1102 goto out;
1103 default:
1104 /*
1105 * Another bit of PARANOID. Note that the retval will be
1106 * 0 since no piece of kernel is supposed to do a check
1107 * for a negative retval of schedule_timeout() (since it
1108 * should never happens anyway). You just have the printk()
1109 * that will tell you if something is gone wrong and where.
1110 */
1111 if (timeout < 0)
1112 {
1113 printk(KERN_ERR "schedule_timeout: wrong timeout "
1114 "value %lx from %p\n", timeout,
1115 __builtin_return_address(0));
1116 current->state = TASK_RUNNING;
1117 goto out;
1118 }
1119 }
1120
1121 expire = timeout + jiffies;
1122
1123 setup_timer(&timer, process_timeout, (unsigned long)current);
1124 __mod_timer(&timer, expire);
1125 schedule();
1126 del_singleshot_timer_sync(&timer);
1127
1128 timeout = expire - jiffies;
1129
1130 out:
1131 return timeout < 0 ? 0 : timeout;
1132 }
1133 EXPORT_SYMBOL(schedule_timeout);
1134
1135 /*
1136 * We can use __set_current_state() here because schedule_timeout() calls
1137 * schedule() unconditionally.
1138 */
1139 signed long __sched schedule_timeout_interruptible(signed long timeout)
1140 {
1141 __set_current_state(TASK_INTERRUPTIBLE);
1142 return schedule_timeout(timeout);
1143 }
1144 EXPORT_SYMBOL(schedule_timeout_interruptible);
1145
1146 signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1147 {
1148 __set_current_state(TASK_UNINTERRUPTIBLE);
1149 return schedule_timeout(timeout);
1150 }
1151 EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1152
1153 /* Thread ID - the internal kernel "pid" */
1154 asmlinkage long sys_gettid(void)
1155 {
1156 return current->pid;
1157 }
1158
1159 /*
1160 * sys_sysinfo - fill in sysinfo struct
1161 */
1162 asmlinkage long sys_sysinfo(struct sysinfo __user *info)
1163 {
1164 struct sysinfo val;
1165 unsigned long mem_total, sav_total;
1166 unsigned int mem_unit, bitcount;
1167 unsigned long seq;
1168
1169 memset((char *)&val, 0, sizeof(struct sysinfo));
1170
1171 do {
1172 struct timespec tp;
1173 seq = read_seqbegin(&xtime_lock);
1174
1175 /*
1176 * This is annoying. The below is the same thing
1177 * posix_get_clock_monotonic() does, but it wants to
1178 * take the lock which we want to cover the loads stuff
1179 * too.
1180 */
1181
1182 getnstimeofday(&tp);
1183 tp.tv_sec += wall_to_monotonic.tv_sec;
1184 tp.tv_nsec += wall_to_monotonic.tv_nsec;
1185 if (tp.tv_nsec - NSEC_PER_SEC >= 0) {
1186 tp.tv_nsec = tp.tv_nsec - NSEC_PER_SEC;
1187 tp.tv_sec++;
1188 }
1189 val.uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
1190
1191 val.loads[0] = avenrun[0] << (SI_LOAD_SHIFT - FSHIFT);
1192 val.loads[1] = avenrun[1] << (SI_LOAD_SHIFT - FSHIFT);
1193 val.loads[2] = avenrun[2] << (SI_LOAD_SHIFT - FSHIFT);
1194
1195 val.procs = nr_threads;
1196 } while (read_seqretry(&xtime_lock, seq));
1197
1198 si_meminfo(&val);
1199 si_swapinfo(&val);
1200
1201 /*
1202 * If the sum of all the available memory (i.e. ram + swap)
1203 * is less than can be stored in a 32 bit unsigned long then
1204 * we can be binary compatible with 2.2.x kernels. If not,
1205 * well, in that case 2.2.x was broken anyways...
1206 *
1207 * -Erik Andersen <andersee@debian.org>
1208 */
1209
1210 mem_total = val.totalram + val.totalswap;
1211 if (mem_total < val.totalram || mem_total < val.totalswap)
1212 goto out;
1213 bitcount = 0;
1214 mem_unit = val.mem_unit;
1215 while (mem_unit > 1) {
1216 bitcount++;
1217 mem_unit >>= 1;
1218 sav_total = mem_total;
1219 mem_total <<= 1;
1220 if (mem_total < sav_total)
1221 goto out;
1222 }
1223
1224 /*
1225 * If mem_total did not overflow, multiply all memory values by
1226 * val.mem_unit and set it to 1. This leaves things compatible
1227 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1228 * kernels...
1229 */
1230
1231 val.mem_unit = 1;
1232 val.totalram <<= bitcount;
1233 val.freeram <<= bitcount;
1234 val.sharedram <<= bitcount;
1235 val.bufferram <<= bitcount;
1236 val.totalswap <<= bitcount;
1237 val.freeswap <<= bitcount;
1238 val.totalhigh <<= bitcount;
1239 val.freehigh <<= bitcount;
1240
1241 out:
1242 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
1243 return -EFAULT;
1244
1245 return 0;
1246 }
1247
1248 static int __devinit init_timers_cpu(int cpu)
1249 {
1250 int j;
1251 tvec_base_t *base;
1252
1253 base = per_cpu(tvec_bases, cpu);
1254 if (!base) {
1255 static char boot_done;
1256
1257 /*
1258 * Cannot do allocation in init_timers as that runs before the
1259 * allocator initializes (and would waste memory if there are
1260 * more possible CPUs than will ever be installed/brought up).
1261 */
1262 if (boot_done) {
1263 base = kmalloc_node(sizeof(*base), GFP_KERNEL,
1264 cpu_to_node(cpu));
1265 if (!base)
1266 return -ENOMEM;
1267 memset(base, 0, sizeof(*base));
1268 } else {
1269 base = &boot_tvec_bases;
1270 boot_done = 1;
1271 }
1272 per_cpu(tvec_bases, cpu) = base;
1273 }
1274 spin_lock_init(&base->t_base.lock);
1275 for (j = 0; j < TVN_SIZE; j++) {
1276 INIT_LIST_HEAD(base->tv5.vec + j);
1277 INIT_LIST_HEAD(base->tv4.vec + j);
1278 INIT_LIST_HEAD(base->tv3.vec + j);
1279 INIT_LIST_HEAD(base->tv2.vec + j);
1280 }
1281 for (j = 0; j < TVR_SIZE; j++)
1282 INIT_LIST_HEAD(base->tv1.vec + j);
1283
1284 base->timer_jiffies = jiffies;
1285 return 0;
1286 }
1287
1288 #ifdef CONFIG_HOTPLUG_CPU
1289 static void migrate_timer_list(tvec_base_t *new_base, struct list_head *head)
1290 {
1291 struct timer_list *timer;
1292
1293 while (!list_empty(head)) {
1294 timer = list_entry(head->next, struct timer_list, entry);
1295 detach_timer(timer, 0);
1296 timer->base = &new_base->t_base;
1297 internal_add_timer(new_base, timer);
1298 }
1299 }
1300
1301 static void __devinit migrate_timers(int cpu)
1302 {
1303 tvec_base_t *old_base;
1304 tvec_base_t *new_base;
1305 int i;
1306
1307 BUG_ON(cpu_online(cpu));
1308 old_base = per_cpu(tvec_bases, cpu);
1309 new_base = get_cpu_var(tvec_bases);
1310
1311 local_irq_disable();
1312 spin_lock(&new_base->t_base.lock);
1313 spin_lock(&old_base->t_base.lock);
1314
1315 if (old_base->t_base.running_timer)
1316 BUG();
1317 for (i = 0; i < TVR_SIZE; i++)
1318 migrate_timer_list(new_base, old_base->tv1.vec + i);
1319 for (i = 0; i < TVN_SIZE; i++) {
1320 migrate_timer_list(new_base, old_base->tv2.vec + i);
1321 migrate_timer_list(new_base, old_base->tv3.vec + i);
1322 migrate_timer_list(new_base, old_base->tv4.vec + i);
1323 migrate_timer_list(new_base, old_base->tv5.vec + i);
1324 }
1325
1326 spin_unlock(&old_base->t_base.lock);
1327 spin_unlock(&new_base->t_base.lock);
1328 local_irq_enable();
1329 put_cpu_var(tvec_bases);
1330 }
1331 #endif /* CONFIG_HOTPLUG_CPU */
1332
1333 static int __devinit timer_cpu_notify(struct notifier_block *self,
1334 unsigned long action, void *hcpu)
1335 {
1336 long cpu = (long)hcpu;
1337 switch(action) {
1338 case CPU_UP_PREPARE:
1339 if (init_timers_cpu(cpu) < 0)
1340 return NOTIFY_BAD;
1341 break;
1342 #ifdef CONFIG_HOTPLUG_CPU
1343 case CPU_DEAD:
1344 migrate_timers(cpu);
1345 break;
1346 #endif
1347 default:
1348 break;
1349 }
1350 return NOTIFY_OK;
1351 }
1352
1353 static struct notifier_block __devinitdata timers_nb = {
1354 .notifier_call = timer_cpu_notify,
1355 };
1356
1357
1358 void __init init_timers(void)
1359 {
1360 timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE,
1361 (void *)(long)smp_processor_id());
1362 register_cpu_notifier(&timers_nb);
1363 open_softirq(TIMER_SOFTIRQ, run_timer_softirq, NULL);
1364 }
1365
1366 #ifdef CONFIG_TIME_INTERPOLATION
1367
1368 struct time_interpolator *time_interpolator __read_mostly;
1369 static struct time_interpolator *time_interpolator_list __read_mostly;
1370 static DEFINE_SPINLOCK(time_interpolator_lock);
1371
1372 static inline u64 time_interpolator_get_cycles(unsigned int src)
1373 {
1374 unsigned long (*x)(void);
1375
1376 switch (src)
1377 {
1378 case TIME_SOURCE_FUNCTION:
1379 x = time_interpolator->addr;
1380 return x();
1381
1382 case TIME_SOURCE_MMIO64 :
1383 return readq_relaxed((void __iomem *)time_interpolator->addr);
1384
1385 case TIME_SOURCE_MMIO32 :
1386 return readl_relaxed((void __iomem *)time_interpolator->addr);
1387
1388 default: return get_cycles();
1389 }
1390 }
1391
1392 static inline u64 time_interpolator_get_counter(int writelock)
1393 {
1394 unsigned int src = time_interpolator->source;
1395
1396 if (time_interpolator->jitter)
1397 {
1398 u64 lcycle;
1399 u64 now;
1400
1401 do {
1402 lcycle = time_interpolator->last_cycle;
1403 now = time_interpolator_get_cycles(src);
1404 if (lcycle && time_after(lcycle, now))
1405 return lcycle;
1406
1407 /* When holding the xtime write lock, there's no need
1408 * to add the overhead of the cmpxchg. Readers are
1409 * force to retry until the write lock is released.
1410 */
1411 if (writelock) {
1412 time_interpolator->last_cycle = now;
1413 return now;
1414 }
1415 /* Keep track of the last timer value returned. The use of cmpxchg here
1416 * will cause contention in an SMP environment.
1417 */
1418 } while (unlikely(cmpxchg(&time_interpolator->last_cycle, lcycle, now) != lcycle));
1419 return now;
1420 }
1421 else
1422 return time_interpolator_get_cycles(src);
1423 }
1424
1425 void time_interpolator_reset(void)
1426 {
1427 time_interpolator->offset = 0;
1428 time_interpolator->last_counter = time_interpolator_get_counter(1);
1429 }
1430
1431 #define GET_TI_NSECS(count,i) (((((count) - i->last_counter) & (i)->mask) * (i)->nsec_per_cyc) >> (i)->shift)
1432
1433 unsigned long time_interpolator_get_offset(void)
1434 {
1435 /* If we do not have a time interpolator set up then just return zero */
1436 if (!time_interpolator)
1437 return 0;
1438
1439 return time_interpolator->offset +
1440 GET_TI_NSECS(time_interpolator_get_counter(0), time_interpolator);
1441 }
1442
1443 #define INTERPOLATOR_ADJUST 65536
1444 #define INTERPOLATOR_MAX_SKIP 10*INTERPOLATOR_ADJUST
1445
1446 static void time_interpolator_update(long delta_nsec)
1447 {
1448 u64 counter;
1449 unsigned long offset;
1450
1451 /* If there is no time interpolator set up then do nothing */
1452 if (!time_interpolator)
1453 return;
1454
1455 /*
1456 * The interpolator compensates for late ticks by accumulating the late
1457 * time in time_interpolator->offset. A tick earlier than expected will
1458 * lead to a reset of the offset and a corresponding jump of the clock
1459 * forward. Again this only works if the interpolator clock is running
1460 * slightly slower than the regular clock and the tuning logic insures
1461 * that.
1462 */
1463
1464 counter = time_interpolator_get_counter(1);
1465 offset = time_interpolator->offset +
1466 GET_TI_NSECS(counter, time_interpolator);
1467
1468 if (delta_nsec < 0 || (unsigned long) delta_nsec < offset)
1469 time_interpolator->offset = offset - delta_nsec;
1470 else {
1471 time_interpolator->skips++;
1472 time_interpolator->ns_skipped += delta_nsec - offset;
1473 time_interpolator->offset = 0;
1474 }
1475 time_interpolator->last_counter = counter;
1476
1477 /* Tuning logic for time interpolator invoked every minute or so.
1478 * Decrease interpolator clock speed if no skips occurred and an offset is carried.
1479 * Increase interpolator clock speed if we skip too much time.
1480 */
1481 if (jiffies % INTERPOLATOR_ADJUST == 0)
1482 {
1483 if (time_interpolator->skips == 0 && time_interpolator->offset > TICK_NSEC)
1484 time_interpolator->nsec_per_cyc--;
1485 if (time_interpolator->ns_skipped > INTERPOLATOR_MAX_SKIP && time_interpolator->offset == 0)
1486 time_interpolator->nsec_per_cyc++;
1487 time_interpolator->skips = 0;
1488 time_interpolator->ns_skipped = 0;
1489 }
1490 }
1491
1492 static inline int
1493 is_better_time_interpolator(struct time_interpolator *new)
1494 {
1495 if (!time_interpolator)
1496 return 1;
1497 return new->frequency > 2*time_interpolator->frequency ||
1498 (unsigned long)new->drift < (unsigned long)time_interpolator->drift;
1499 }
1500
1501 void
1502 register_time_interpolator(struct time_interpolator *ti)
1503 {
1504 unsigned long flags;
1505
1506 /* Sanity check */
1507 if (ti->frequency == 0 || ti->mask == 0)
1508 BUG();
1509
1510 ti->nsec_per_cyc = ((u64)NSEC_PER_SEC << ti->shift) / ti->frequency;
1511 spin_lock(&time_interpolator_lock);
1512 write_seqlock_irqsave(&xtime_lock, flags);
1513 if (is_better_time_interpolator(ti)) {
1514 time_interpolator = ti;
1515 time_interpolator_reset();
1516 }
1517 write_sequnlock_irqrestore(&xtime_lock, flags);
1518
1519 ti->next = time_interpolator_list;
1520 time_interpolator_list = ti;
1521 spin_unlock(&time_interpolator_lock);
1522 }
1523
1524 void
1525 unregister_time_interpolator(struct time_interpolator *ti)
1526 {
1527 struct time_interpolator *curr, **prev;
1528 unsigned long flags;
1529
1530 spin_lock(&time_interpolator_lock);
1531 prev = &time_interpolator_list;
1532 for (curr = *prev; curr; curr = curr->next) {
1533 if (curr == ti) {
1534 *prev = curr->next;
1535 break;
1536 }
1537 prev = &curr->next;
1538 }
1539
1540 write_seqlock_irqsave(&xtime_lock, flags);
1541 if (ti == time_interpolator) {
1542 /* we lost the best time-interpolator: */
1543 time_interpolator = NULL;
1544 /* find the next-best interpolator */
1545 for (curr = time_interpolator_list; curr; curr = curr->next)
1546 if (is_better_time_interpolator(curr))
1547 time_interpolator = curr;
1548 time_interpolator_reset();
1549 }
1550 write_sequnlock_irqrestore(&xtime_lock, flags);
1551 spin_unlock(&time_interpolator_lock);
1552 }
1553 #endif /* CONFIG_TIME_INTERPOLATION */
1554
1555 /**
1556 * msleep - sleep safely even with waitqueue interruptions
1557 * @msecs: Time in milliseconds to sleep for
1558 */
1559 void msleep(unsigned int msecs)
1560 {
1561 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1562
1563 while (timeout)
1564 timeout = schedule_timeout_uninterruptible(timeout);
1565 }
1566
1567 EXPORT_SYMBOL(msleep);
1568
1569 /**
1570 * msleep_interruptible - sleep waiting for signals
1571 * @msecs: Time in milliseconds to sleep for
1572 */
1573 unsigned long msleep_interruptible(unsigned int msecs)
1574 {
1575 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1576
1577 while (timeout && !signal_pending(current))
1578 timeout = schedule_timeout_interruptible(timeout);
1579 return jiffies_to_msecs(timeout);
1580 }
1581
1582 EXPORT_SYMBOL(msleep_interruptible);