projects/linux·blob v2.6.14-rc1 · refs

1632 lines · 42.6 KiB

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
37#include <asm/uaccess.h>
38#include <asm/unistd.h>
39#include <asm/div64.h>
40#include <asm/timex.h>
41#include <asm/io.h>
42
43#ifdef CONFIG_TIME_INTERPOLATION
44static void time_interpolator_update(long delta_nsec);
45#else
46#define time_interpolator_update(x)
47#endif
48
49/*
50 * per-CPU timer vector definitions:
51 */
52
53#define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6)
54#define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8)
55#define TVN_SIZE (1 << TVN_BITS)
56#define TVR_SIZE (1 << TVR_BITS)
57#define TVN_MASK (TVN_SIZE - 1)
58#define TVR_MASK (TVR_SIZE - 1)
59
60struct timer_base_s {
61 spinlock_t lock;
62 struct timer_list *running_timer;
63};
64
65typedef struct tvec_s {
66 struct list_head vec[TVN_SIZE];
67} tvec_t;
68
69typedef struct tvec_root_s {
70 struct list_head vec[TVR_SIZE];
71} tvec_root_t;
72
73struct tvec_t_base_s {
74 struct timer_base_s t_base;
75 unsigned long timer_jiffies;
76 tvec_root_t tv1;
77 tvec_t tv2;
78 tvec_t tv3;
79 tvec_t tv4;
80 tvec_t tv5;
81} ____cacheline_aligned_in_smp;
82
83typedef struct tvec_t_base_s tvec_base_t;
84static DEFINE_PER_CPU(tvec_base_t, tvec_bases);
85
86static inline void set_running_timer(tvec_base_t *base,
87 struct timer_list *timer)
88{
89#ifdef CONFIG_SMP
90 base->t_base.running_timer = timer;
91#endif
92}
93
94static void check_timer_failed(struct timer_list *timer)
95{
96 static int whine_count;
97 if (whine_count < 16) {
98 whine_count++;
99 printk("Uninitialised timer!\n");
100 printk("This is just a warning. Your computer is OK\n");
101 printk("function=0x%p, data=0x%lx\n",
102 timer->function, timer->data);
103 dump_stack();
104 }
105 /*
106 * Now fix it up
107 */
108 timer->magic = TIMER_MAGIC;
109}
110
111static inline void check_timer(struct timer_list *timer)
112{
113 if (timer->magic != TIMER_MAGIC)
114 check_timer_failed(timer);
115}
116
117
118static void internal_add_timer(tvec_base_t *base, struct timer_list *timer)
119{
120 unsigned long expires = timer->expires;
121 unsigned long idx = expires - base->timer_jiffies;
122 struct list_head *vec;
123
124 if (idx < TVR_SIZE) {
125 int i = expires & TVR_MASK;
126 vec = base->tv1.vec + i;
127 } else if (idx < 1 << (TVR_BITS + TVN_BITS)) {
128 int i = (expires >> TVR_BITS) & TVN_MASK;
129 vec = base->tv2.vec + i;
130 } else if (idx < 1 << (TVR_BITS + 2 * TVN_BITS)) {
131 int i = (expires >> (TVR_BITS + TVN_BITS)) & TVN_MASK;
132 vec = base->tv3.vec + i;
133 } else if (idx < 1 << (TVR_BITS + 3 * TVN_BITS)) {
134 int i = (expires >> (TVR_BITS + 2 * TVN_BITS)) & TVN_MASK;
135 vec = base->tv4.vec + i;
136 } else if ((signed long) idx < 0) {
137 /*
138 * Can happen if you add a timer with expires == jiffies,
139 * or you set a timer to go off in the past
140 */
141 vec = base->tv1.vec + (base->timer_jiffies & TVR_MASK);
142 } else {
143 int i;
144 /* If the timeout is larger than 0xffffffff on 64-bit
145 * architectures then we use the maximum timeout:
146 */
147 if (idx > 0xffffffffUL) {
148 idx = 0xffffffffUL;
149 expires = idx + base->timer_jiffies;
150 }
151 i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK;
152 vec = base->tv5.vec + i;
153 }
154 /*
155 * Timers are FIFO:
156 */
157 list_add_tail(&timer->entry, vec);
158}
159
160typedef struct timer_base_s timer_base_t;
161/*
162 * Used by TIMER_INITIALIZER, we can't use per_cpu(tvec_bases)
163 * at compile time, and we need timer->base to lock the timer.
164 */
165timer_base_t __init_timer_base
166 ____cacheline_aligned_in_smp = { .lock = SPIN_LOCK_UNLOCKED };
167EXPORT_SYMBOL(__init_timer_base);
168
169/***
170 * init_timer - initialize a timer.
171 * @timer: the timer to be initialized
172 *
173 * init_timer() must be done to a timer prior calling *any* of the
174 * other timer functions.
175 */
176void fastcall init_timer(struct timer_list *timer)
177{
178 timer->entry.next = NULL;
179 timer->base = &per_cpu(tvec_bases, raw_smp_processor_id()).t_base;
180 timer->magic = TIMER_MAGIC;
181}
182EXPORT_SYMBOL(init_timer);
183
184static inline void detach_timer(struct timer_list *timer,
185 int clear_pending)
186{
187 struct list_head *entry = &timer->entry;
188
189 __list_del(entry->prev, entry->next);
190 if (clear_pending)
191 entry->next = NULL;
192 entry->prev = LIST_POISON2;
193}
194
195/*
196 * We are using hashed locking: holding per_cpu(tvec_bases).t_base.lock
197 * means that all timers which are tied to this base via timer->base are
198 * locked, and the base itself is locked too.
199 *
200 * So __run_timers/migrate_timers can safely modify all timers which could
201 * be found on ->tvX lists.
202 *
203 * When the timer's base is locked, and the timer removed from list, it is
204 * possible to set timer->base = NULL and drop the lock: the timer remains
205 * locked.
206 */
207static timer_base_t *lock_timer_base(struct timer_list *timer,
208 unsigned long *flags)
209{
210 timer_base_t *base;
211
212 for (;;) {
213 base = timer->base;
214 if (likely(base != NULL)) {
215 spin_lock_irqsave(&base->lock, *flags);
216 if (likely(base == timer->base))
217 return base;
218 /* The timer has migrated to another CPU */
219 spin_unlock_irqrestore(&base->lock, *flags);
220 }
221 cpu_relax();
222 }
223}
224
225int __mod_timer(struct timer_list *timer, unsigned long expires)
226{
227 timer_base_t *base;
228 tvec_base_t *new_base;
229 unsigned long flags;
230 int ret = 0;
231
232 BUG_ON(!timer->function);
233 check_timer(timer);
234
235 base = lock_timer_base(timer, &flags);
236
237 if (timer_pending(timer)) {
238 detach_timer(timer, 0);
239 ret = 1;
240 }
241
242 new_base = &__get_cpu_var(tvec_bases);
243
244 if (base != &new_base->t_base) {
245 /*
246 * We are trying to schedule the timer on the local CPU.
247 * However we can't change timer's base while it is running,
248 * otherwise del_timer_sync() can't detect that the timer's
249 * handler yet has not finished. This also guarantees that
250 * the timer is serialized wrt itself.
251 */
252 if (unlikely(base->running_timer == timer)) {
253 /* The timer remains on a former base */
254 new_base = container_of(base, tvec_base_t, t_base);
255 } else {
256 /* See the comment in lock_timer_base() */
257 timer->base = NULL;
258 spin_unlock(&base->lock);
259 spin_lock(&new_base->t_base.lock);
260 timer->base = &new_base->t_base;
261 }
262 }
263
264 timer->expires = expires;
265 internal_add_timer(new_base, timer);
266 spin_unlock_irqrestore(&new_base->t_base.lock, flags);
267
268 return ret;
269}
270
271EXPORT_SYMBOL(__mod_timer);
272
273/***
274 * add_timer_on - start a timer on a particular CPU
275 * @timer: the timer to be added
276 * @cpu: the CPU to start it on
277 *
278 * This is not very scalable on SMP. Double adds are not possible.
279 */
280void add_timer_on(struct timer_list *timer, int cpu)
281{
282 tvec_base_t *base = &per_cpu(tvec_bases, cpu);
283 unsigned long flags;
284
285 BUG_ON(timer_pending(timer) || !timer->function);
286
287 check_timer(timer);
288
289 spin_lock_irqsave(&base->t_base.lock, flags);
290 timer->base = &base->t_base;
291 internal_add_timer(base, timer);
292 spin_unlock_irqrestore(&base->t_base.lock, flags);
293}
294
295
296/***
297 * mod_timer - modify a timer's timeout
298 * @timer: the timer to be modified
299 *
300 * mod_timer is a more efficient way to update the expire field of an
301 * active timer (if the timer is inactive it will be activated)
302 *
303 * mod_timer(timer, expires) is equivalent to:
304 *
305 * del_timer(timer); timer->expires = expires; add_timer(timer);
306 *
307 * Note that if there are multiple unserialized concurrent users of the
308 * same timer, then mod_timer() is the only safe way to modify the timeout,
309 * since add_timer() cannot modify an already running timer.
310 *
311 * The function returns whether it has modified a pending timer or not.
312 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
313 * active timer returns 1.)
314 */
315int mod_timer(struct timer_list *timer, unsigned long expires)
316{
317 BUG_ON(!timer->function);
318
319 check_timer(timer);
320
321 /*
322 * This is a common optimization triggered by the
323 * networking code - if the timer is re-modified
324 * to be the same thing then just return:
325 */
326 if (timer->expires == expires && timer_pending(timer))
327 return 1;
328
329 return __mod_timer(timer, expires);
330}
331
332EXPORT_SYMBOL(mod_timer);
333
334/***
335 * del_timer - deactive a timer.
336 * @timer: the timer to be deactivated
337 *
338 * del_timer() deactivates a timer - this works on both active and inactive
339 * timers.
340 *
341 * The function returns whether it has deactivated a pending timer or not.
342 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
343 * active timer returns 1.)
344 */
345int del_timer(struct timer_list *timer)
346{
347 timer_base_t *base;
348 unsigned long flags;
349 int ret = 0;
350
351 check_timer(timer);
352
353 if (timer_pending(timer)) {
354 base = lock_timer_base(timer, &flags);
355 if (timer_pending(timer)) {
356 detach_timer(timer, 1);
357 ret = 1;
358 }
359 spin_unlock_irqrestore(&base->lock, flags);
360 }
361
362 return ret;
363}
364
365EXPORT_SYMBOL(del_timer);
366
367#ifdef CONFIG_SMP
368/*
369 * This function tries to deactivate a timer. Upon successful (ret >= 0)
370 * exit the timer is not queued and the handler is not running on any CPU.
371 *
372 * It must not be called from interrupt contexts.
373 */
374int try_to_del_timer_sync(struct timer_list *timer)
375{
376 timer_base_t *base;
377 unsigned long flags;
378 int ret = -1;
379
380 base = lock_timer_base(timer, &flags);
381
382 if (base->running_timer == timer)
383 goto out;
384
385 ret = 0;
386 if (timer_pending(timer)) {
387 detach_timer(timer, 1);
388 ret = 1;
389 }
390out:
391 spin_unlock_irqrestore(&base->lock, flags);
392
393 return ret;
394}
395
396/***
397 * del_timer_sync - deactivate a timer and wait for the handler to finish.
398 * @timer: the timer to be deactivated
399 *
400 * This function only differs from del_timer() on SMP: besides deactivating
401 * the timer it also makes sure the handler has finished executing on other
402 * CPUs.
403 *
404 * Synchronization rules: callers must prevent restarting of the timer,
405 * otherwise this function is meaningless. It must not be called from
406 * interrupt contexts. The caller must not hold locks which would prevent
407 * completion of the timer's handler. The timer's handler must not call
408 * add_timer_on(). Upon exit the timer is not queued and the handler is
409 * not running on any CPU.
410 *
411 * The function returns whether it has deactivated a pending timer or not.
412 */
413int del_timer_sync(struct timer_list *timer)
414{
415 check_timer(timer);
416
417 for (;;) {
418 int ret = try_to_del_timer_sync(timer);
419 if (ret >= 0)
420 return ret;
421 }
422}
423
424EXPORT_SYMBOL(del_timer_sync);
425#endif
426
427static int cascade(tvec_base_t *base, tvec_t *tv, int index)
428{
429 /* cascade all the timers from tv up one level */
430 struct list_head *head, *curr;
431
432 head = tv->vec + index;
433 curr = head->next;
434 /*
435 * We are removing _all_ timers from the list, so we don't have to
436 * detach them individually, just clear the list afterwards.
437 */
438 while (curr != head) {
439 struct timer_list *tmp;
440
441 tmp = list_entry(curr, struct timer_list, entry);
442 BUG_ON(tmp->base != &base->t_base);
443 curr = curr->next;
444 internal_add_timer(base, tmp);
445 }
446 INIT_LIST_HEAD(head);
447
448 return index;
449}
450
451/***
452 * __run_timers - run all expired timers (if any) on this CPU.
453 * @base: the timer vector to be processed.
454 *
455 * This function cascades all vectors and executes all expired timer
456 * vectors.
457 */
458#define INDEX(N) (base->timer_jiffies >> (TVR_BITS + N * TVN_BITS)) & TVN_MASK
459
460static inline void __run_timers(tvec_base_t *base)
461{
462 struct timer_list *timer;
463
464 spin_lock_irq(&base->t_base.lock);
465 while (time_after_eq(jiffies, base->timer_jiffies)) {
466 struct list_head work_list = LIST_HEAD_INIT(work_list);
467 struct list_head *head = &work_list;
468 int index = base->timer_jiffies & TVR_MASK;
469
470 /*
471 * Cascade timers:
472 */
473 if (!index &&
474 (!cascade(base, &base->tv2, INDEX(0))) &&
475 (!cascade(base, &base->tv3, INDEX(1))) &&
476 !cascade(base, &base->tv4, INDEX(2)))
477 cascade(base, &base->tv5, INDEX(3));
478 ++base->timer_jiffies;
479 list_splice_init(base->tv1.vec + index, &work_list);
480 while (!list_empty(head)) {
481 void (*fn)(unsigned long);
482 unsigned long data;
483
484 timer = list_entry(head->next,struct timer_list,entry);
485 fn = timer->function;
486 data = timer->data;
487
488 set_running_timer(base, timer);
489 detach_timer(timer, 1);
490 spin_unlock_irq(&base->t_base.lock);
491 {
492 int preempt_count = preempt_count();
493 fn(data);
494 if (preempt_count != preempt_count()) {
495 printk(KERN_WARNING "huh, entered %p "
496 "with preempt_count %08x, exited"
497 " with %08x?\n",
498 fn, preempt_count,
499 preempt_count());
500 BUG();
501 }
502 }
503 spin_lock_irq(&base->t_base.lock);
504 }
505 }
506 set_running_timer(base, NULL);
507 spin_unlock_irq(&base->t_base.lock);
508}
509
510#ifdef CONFIG_NO_IDLE_HZ
511/*
512 * Find out when the next timer event is due to happen. This
513 * is used on S/390 to stop all activity when a cpus is idle.
514 * This functions needs to be called disabled.
515 */
516unsigned long next_timer_interrupt(void)
517{
518 tvec_base_t *base;
519 struct list_head *list;
520 struct timer_list *nte;
521 unsigned long expires;
522 tvec_t *varray[4];
523 int i, j;
524
525 base = &__get_cpu_var(tvec_bases);
526 spin_lock(&base->t_base.lock);
527 expires = base->timer_jiffies + (LONG_MAX >> 1);
528 list = 0;
529
530 /* Look for timer events in tv1. */
531 j = base->timer_jiffies & TVR_MASK;
532 do {
533 list_for_each_entry(nte, base->tv1.vec + j, entry) {
534 expires = nte->expires;
535 if (j < (base->timer_jiffies & TVR_MASK))
536 list = base->tv2.vec + (INDEX(0));
537 goto found;
538 }
539 j = (j + 1) & TVR_MASK;
540 } while (j != (base->timer_jiffies & TVR_MASK));
541
542 /* Check tv2-tv5. */
543 varray[0] = &base->tv2;
544 varray[1] = &base->tv3;
545 varray[2] = &base->tv4;
546 varray[3] = &base->tv5;
547 for (i = 0; i < 4; i++) {
548 j = INDEX(i);
549 do {
550 if (list_empty(varray[i]->vec + j)) {
551 j = (j + 1) & TVN_MASK;
552 continue;
553 }
554 list_for_each_entry(nte, varray[i]->vec + j, entry)
555 if (time_before(nte->expires, expires))
556 expires = nte->expires;
557 if (j < (INDEX(i)) && i < 3)
558 list = varray[i + 1]->vec + (INDEX(i + 1));
559 goto found;
560 } while (j != (INDEX(i)));
561 }
562found:
563 if (list) {
564 /*
565 * The search wrapped. We need to look at the next list
566 * from next tv element that would cascade into tv element
567 * where we found the timer element.
568 */
569 list_for_each_entry(nte, list, entry) {
570 if (time_before(nte->expires, expires))
571 expires = nte->expires;
572 }
573 }
574 spin_unlock(&base->t_base.lock);
575 return expires;
576}
577#endif
578
579/******************************************************************/
580
581/*
582 * Timekeeping variables
583 */
584unsigned long tick_usec = TICK_USEC; /* USER_HZ period (usec) */
585unsigned long tick_nsec = TICK_NSEC; /* ACTHZ period (nsec) */
586
587/*
588 * The current time
589 * wall_to_monotonic is what we need to add to xtime (or xtime corrected
590 * for sub jiffie times) to get to monotonic time. Monotonic is pegged
591 * at zero at system boot time, so wall_to_monotonic will be negative,
592 * however, we will ALWAYS keep the tv_nsec part positive so we can use
593 * the usual normalization.
594 */
595struct timespec xtime __attribute__ ((aligned (16)));
596struct timespec wall_to_monotonic __attribute__ ((aligned (16)));
597
598EXPORT_SYMBOL(xtime);
599
600/* Don't completely fail for HZ > 500. */
601int tickadj = 500/HZ ? : 1; /* microsecs */
602
603
604/*
605 * phase-lock loop variables
606 */
607/* TIME_ERROR prevents overwriting the CMOS clock */
608int time_state = TIME_OK; /* clock synchronization status */
609int time_status = STA_UNSYNC; /* clock status bits */
610long time_offset; /* time adjustment (us) */
611long time_constant = 2; /* pll time constant */
612long time_tolerance = MAXFREQ; /* frequency tolerance (ppm) */
613long time_precision = 1; /* clock precision (us) */
614long time_maxerror = NTP_PHASE_LIMIT; /* maximum error (us) */
615long time_esterror = NTP_PHASE_LIMIT; /* estimated error (us) */
616static long time_phase; /* phase offset (scaled us) */
617long time_freq = (((NSEC_PER_SEC + HZ/2) % HZ - HZ/2) << SHIFT_USEC) / NSEC_PER_USEC;
618 /* frequency offset (scaled ppm)*/
619static long time_adj; /* tick adjust (scaled 1 / HZ) */
620long time_reftime; /* time at last adjustment (s) */
621long time_adjust;
622long time_next_adjust;
623
624/*
625 * this routine handles the overflow of the microsecond field
626 *
627 * The tricky bits of code to handle the accurate clock support
628 * were provided by Dave Mills ([email protected]) of NTP fame.
629 * They were originally developed for SUN and DEC kernels.
630 * All the kudos should go to Dave for this stuff.
631 *
632 */
633static void second_overflow(void)
634{
635 long ltemp;
636
637 /* Bump the maxerror field */
638 time_maxerror += time_tolerance >> SHIFT_USEC;
639 if ( time_maxerror > NTP_PHASE_LIMIT ) {
640 time_maxerror = NTP_PHASE_LIMIT;
641 time_status |= STA_UNSYNC;
642 }
643
644 /*
645 * Leap second processing. If in leap-insert state at
646 * the end of the day, the system clock is set back one
647 * second; if in leap-delete state, the system clock is
648 * set ahead one second. The microtime() routine or
649 * external clock driver will insure that reported time
650 * is always monotonic. The ugly divides should be
651 * replaced.
652 */
653 switch (time_state) {
654
655 case TIME_OK:
656 if (time_status & STA_INS)
657 time_state = TIME_INS;
658 else if (time_status & STA_DEL)
659 time_state = TIME_DEL;
660 break;
661
662 case TIME_INS:
663 if (xtime.tv_sec % 86400 == 0) {
664 xtime.tv_sec--;
665 wall_to_monotonic.tv_sec++;
666 /* The timer interpolator will make time change gradually instead
667 * of an immediate jump by one second.
668 */
669 time_interpolator_update(-NSEC_PER_SEC);
670 time_state = TIME_OOP;
671 clock_was_set();
672 printk(KERN_NOTICE "Clock: inserting leap second 23:59:60 UTC\n");
673 }
674 break;
675
676 case TIME_DEL:
677 if ((xtime.tv_sec + 1) % 86400 == 0) {
678 xtime.tv_sec++;
679 wall_to_monotonic.tv_sec--;
680 /* Use of time interpolator for a gradual change of time */
681 time_interpolator_update(NSEC_PER_SEC);
682 time_state = TIME_WAIT;
683 clock_was_set();
684 printk(KERN_NOTICE "Clock: deleting leap second 23:59:59 UTC\n");
685 }
686 break;
687
688 case TIME_OOP:
689 time_state = TIME_WAIT;
690 break;
691
692 case TIME_WAIT:
693 if (!(time_status & (STA_INS | STA_DEL)))
694 time_state = TIME_OK;
695 }
696
697 /*
698 * Compute the phase adjustment for the next second. In
699 * PLL mode, the offset is reduced by a fixed factor
700 * times the time constant. In FLL mode the offset is
701 * used directly. In either mode, the maximum phase
702 * adjustment for each second is clamped so as to spread
703 * the adjustment over not more than the number of
704 * seconds between updates.
705 */
706 if (time_offset < 0) {
707 ltemp = -time_offset;
708 if (!(time_status & STA_FLL))
709 ltemp >>= SHIFT_KG + time_constant;
710 if (ltemp > (MAXPHASE / MINSEC) << SHIFT_UPDATE)
711 ltemp = (MAXPHASE / MINSEC) << SHIFT_UPDATE;
712 time_offset += ltemp;
713 time_adj = -ltemp << (SHIFT_SCALE - SHIFT_HZ - SHIFT_UPDATE);
714 } else {
715 ltemp = time_offset;
716 if (!(time_status & STA_FLL))
717 ltemp >>= SHIFT_KG + time_constant;
718 if (ltemp > (MAXPHASE / MINSEC) << SHIFT_UPDATE)
719 ltemp = (MAXPHASE / MINSEC) << SHIFT_UPDATE;
720 time_offset -= ltemp;
721 time_adj = ltemp << (SHIFT_SCALE - SHIFT_HZ - SHIFT_UPDATE);
722 }
723
724 /*
725 * Compute the frequency estimate and additional phase
726 * adjustment due to frequency error for the next
727 * second. When the PPS signal is engaged, gnaw on the
728 * watchdog counter and update the frequency computed by
729 * the pll and the PPS signal.
730 */
731 pps_valid++;
732 if (pps_valid == PPS_VALID) { /* PPS signal lost */
733 pps_jitter = MAXTIME;
734 pps_stabil = MAXFREQ;
735 time_status &= ~(STA_PPSSIGNAL | STA_PPSJITTER |
736 STA_PPSWANDER | STA_PPSERROR);
737 }
738 ltemp = time_freq + pps_freq;
739 if (ltemp < 0)
740 time_adj -= -ltemp >>
741 (SHIFT_USEC + SHIFT_HZ - SHIFT_SCALE);
742 else
743 time_adj += ltemp >>
744 (SHIFT_USEC + SHIFT_HZ - SHIFT_SCALE);
745
746#if HZ == 100
747 /* Compensate for (HZ==100) != (1 << SHIFT_HZ).
748 * Add 25% and 3.125% to get 128.125; => only 0.125% error (p. 14)
749 */
750 if (time_adj < 0)
751 time_adj -= (-time_adj >> 2) + (-time_adj >> 5);
752 else
753 time_adj += (time_adj >> 2) + (time_adj >> 5);
754#endif
755#if HZ == 1000
756 /* Compensate for (HZ==1000) != (1 << SHIFT_HZ).
757 * Add 1.5625% and 0.78125% to get 1023.4375; => only 0.05% error (p. 14)
758 */
759 if (time_adj < 0)
760 time_adj -= (-time_adj >> 6) + (-time_adj >> 7);
761 else
762 time_adj += (time_adj >> 6) + (time_adj >> 7);
763#endif
764}
765
766/* in the NTP reference this is called "hardclock()" */
767static void update_wall_time_one_tick(void)
768{
769 long time_adjust_step, delta_nsec;
770
771 if ( (time_adjust_step = time_adjust) != 0 ) {
772 /* We are doing an adjtime thing.
773 *
774 * Prepare time_adjust_step to be within bounds.
775 * Note that a positive time_adjust means we want the clock
776 * to run faster.
777 *
778 * Limit the amount of the step to be in the range
779 * -tickadj .. +tickadj
780 */
781 if (time_adjust > tickadj)
782 time_adjust_step = tickadj;
783 else if (time_adjust < -tickadj)
784 time_adjust_step = -tickadj;
785
786 /* Reduce by this step the amount of time left */
787 time_adjust -= time_adjust_step;
788 }
789 delta_nsec = tick_nsec + time_adjust_step * 1000;
790 /*
791 * Advance the phase, once it gets to one microsecond, then
792 * advance the tick more.
793 */
794 time_phase += time_adj;
795 if (time_phase <= -FINENSEC) {
796 long ltemp = -time_phase >> (SHIFT_SCALE - 10);
797 time_phase += ltemp << (SHIFT_SCALE - 10);
798 delta_nsec -= ltemp;
799 }
800 else if (time_phase >= FINENSEC) {
801 long ltemp = time_phase >> (SHIFT_SCALE - 10);
802 time_phase -= ltemp << (SHIFT_SCALE - 10);
803 delta_nsec += ltemp;
804 }
805 xtime.tv_nsec += delta_nsec;
806 time_interpolator_update(delta_nsec);
807
808 /* Changes by adjtime() do not take effect till next tick. */
809 if (time_next_adjust != 0) {
810 time_adjust = time_next_adjust;
811 time_next_adjust = 0;
812 }
813}
814
815/*
816 * Using a loop looks inefficient, but "ticks" is
817 * usually just one (we shouldn't be losing ticks,
818 * we're doing this this way mainly for interrupt
819 * latency reasons, not because we think we'll
820 * have lots of lost timer ticks
821 */
822static void update_wall_time(unsigned long ticks)
823{
824 do {
825 ticks--;
826 update_wall_time_one_tick();
827 if (xtime.tv_nsec >= 1000000000) {
828 xtime.tv_nsec -= 1000000000;
829 xtime.tv_sec++;
830 second_overflow();
831 }
832 } while (ticks);
833}
834
835/*
836 * Called from the timer interrupt handler to charge one tick to the current
837 * process. user_tick is 1 if the tick is user time, 0 for system.
838 */
839void update_process_times(int user_tick)
840{
841 struct task_struct *p = current;
842 int cpu = smp_processor_id();
843
844 /* Note: this timer irq context must be accounted for as well. */
845 if (user_tick)
846 account_user_time(p, jiffies_to_cputime(1));
847 else
848 account_system_time(p, HARDIRQ_OFFSET, jiffies_to_cputime(1));
849 run_local_timers();
850 if (rcu_pending(cpu))
851 rcu_check_callbacks(cpu, user_tick);
852 scheduler_tick();
853 run_posix_cpu_timers(p);
854}
855
856/*
857 * Nr of active tasks - counted in fixed-point numbers
858 */
859static unsigned long count_active_tasks(void)
860{
861 return (nr_running() + nr_uninterruptible()) * FIXED_1;
862}
863
864/*
865 * Hmm.. Changed this, as the GNU make sources (load.c) seems to
866 * imply that avenrun[] is the standard name for this kind of thing.
867 * Nothing else seems to be standardized: the fractional size etc
868 * all seem to differ on different machines.
869 *
870 * Requires xtime_lock to access.
871 */
872unsigned long avenrun[3];
873
874EXPORT_SYMBOL(avenrun);
875
876/*
877 * calc_load - given tick count, update the avenrun load estimates.
878 * This is called while holding a write_lock on xtime_lock.
879 */
880static inline void calc_load(unsigned long ticks)
881{
882 unsigned long active_tasks; /* fixed-point */
883 static int count = LOAD_FREQ;
884
885 count -= ticks;
886 if (count < 0) {
887 count += LOAD_FREQ;
888 active_tasks = count_active_tasks();
889 CALC_LOAD(avenrun[0], EXP_1, active_tasks);
890 CALC_LOAD(avenrun[1], EXP_5, active_tasks);
891 CALC_LOAD(avenrun[2], EXP_15, active_tasks);
892 }
893}
894
895/* jiffies at the most recent update of wall time */
896unsigned long wall_jiffies = INITIAL_JIFFIES;
897
898/*
899 * This read-write spinlock protects us from races in SMP while
900 * playing with xtime and avenrun.
901 */
902#ifndef ARCH_HAVE_XTIME_LOCK
903seqlock_t xtime_lock __cacheline_aligned_in_smp = SEQLOCK_UNLOCKED;
904
905EXPORT_SYMBOL(xtime_lock);
906#endif
907
908/*
909 * This function runs timers and the timer-tq in bottom half context.
910 */
911static void run_timer_softirq(struct softirq_action *h)
912{
913 tvec_base_t *base = &__get_cpu_var(tvec_bases);
914
915 if (time_after_eq(jiffies, base->timer_jiffies))
916 __run_timers(base);
917}
918
919/*
920 * Called by the local, per-CPU timer interrupt on SMP.
921 */
922void run_local_timers(void)
923{
924 raise_softirq(TIMER_SOFTIRQ);
925}
926
927/*
928 * Called by the timer interrupt. xtime_lock must already be taken
929 * by the timer IRQ!
930 */
931static inline void update_times(void)
932{
933 unsigned long ticks;
934
935 ticks = jiffies - wall_jiffies;
936 if (ticks) {
937 wall_jiffies += ticks;
938 update_wall_time(ticks);
939 }
940 calc_load(ticks);
941}
942
943/*
944 * The 64-bit jiffies value is not atomic - you MUST NOT read it
945 * without sampling the sequence number in xtime_lock.
946 * jiffies is defined in the linker script...
947 */
948
949void do_timer(struct pt_regs *regs)
950{
951 jiffies_64++;
952 update_times();
953 softlockup_tick(regs);
954}
955
956#ifdef __ARCH_WANT_SYS_ALARM
957
958/*
959 * For backwards compatibility? This can be done in libc so Alpha
960 * and all newer ports shouldn't need it.
961 */
962asmlinkage unsigned long sys_alarm(unsigned int seconds)
963{
964 struct itimerval it_new, it_old;
965 unsigned int oldalarm;
966
967 it_new.it_interval.tv_sec = it_new.it_interval.tv_usec = 0;
968 it_new.it_value.tv_sec = seconds;
969 it_new.it_value.tv_usec = 0;
970 do_setitimer(ITIMER_REAL, &it_new, &it_old);
971 oldalarm = it_old.it_value.tv_sec;
972 /* ehhh.. We can't return 0 if we have an alarm pending.. */
973 /* And we'd better return too much than too little anyway */
974 if ((!oldalarm && it_old.it_value.tv_usec) || it_old.it_value.tv_usec >= 500000)
975 oldalarm++;
976 return oldalarm;
977}
978
979#endif
980
981#ifndef __alpha__
982
983/*
984 * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this
985 * should be moved into arch/i386 instead?
986 */
987
988/**
989 * sys_getpid - return the thread group id of the current process
990 *
991 * Note, despite the name, this returns the tgid not the pid. The tgid and
992 * the pid are identical unless CLONE_THREAD was specified on clone() in
993 * which case the tgid is the same in all threads of the same group.
994 *
995 * This is SMP safe as current->tgid does not change.
996 */
997asmlinkage long sys_getpid(void)
998{
999 return current->tgid;
1000}
1001
1002/*
1003 * Accessing ->group_leader->real_parent is not SMP-safe, it could
1004 * change from under us. However, rather than getting any lock
1005 * we can use an optimistic algorithm: get the parent
1006 * pid, and go back and check that the parent is still
1007 * the same. If it has changed (which is extremely unlikely
1008 * indeed), we just try again..
1009 *
1010 * NOTE! This depends on the fact that even if we _do_
1011 * get an old value of "parent", we can happily dereference
1012 * the pointer (it was and remains a dereferencable kernel pointer
1013 * no matter what): we just can't necessarily trust the result
1014 * until we know that the parent pointer is valid.
1015 *
1016 * NOTE2: ->group_leader never changes from under us.
1017 */
1018asmlinkage long sys_getppid(void)
1019{
1020 int pid;
1021 struct task_struct *me = current;
1022 struct task_struct *parent;
1023
1024 parent = me->group_leader->real_parent;
1025 for (;;) {
1026 pid = parent->tgid;
1027#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1028{
1029 struct task_struct *old = parent;
1030
1031 /*
1032 * Make sure we read the pid before re-reading the
1033 * parent pointer:
1034 */
1035 smp_rmb();
1036 parent = me->group_leader->real_parent;
1037 if (old != parent)
1038 continue;
1039}
1040#endif
1041 break;
1042 }
1043 return pid;
1044}
1045
1046asmlinkage long sys_getuid(void)
1047{
1048 /* Only we change this so SMP safe */
1049 return current->uid;
1050}
1051
1052asmlinkage long sys_geteuid(void)
1053{
1054 /* Only we change this so SMP safe */
1055 return current->euid;
1056}
1057
1058asmlinkage long sys_getgid(void)
1059{
1060 /* Only we change this so SMP safe */
1061 return current->gid;
1062}
1063
1064asmlinkage long sys_getegid(void)
1065{
1066 /* Only we change this so SMP safe */
1067 return current->egid;
1068}
1069
1070#endif
1071
1072static void process_timeout(unsigned long __data)
1073{
1074 wake_up_process((task_t *)__data);
1075}
1076
1077/**
1078 * schedule_timeout - sleep until timeout
1079 * @timeout: timeout value in jiffies
1080 *
1081 * Make the current task sleep until @timeout jiffies have
1082 * elapsed. The routine will return immediately unless
1083 * the current task state has been set (see set_current_state()).
1084 *
1085 * You can set the task state as follows -
1086 *
1087 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1088 * pass before the routine returns. The routine will return 0
1089 *
1090 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1091 * delivered to the current task. In this case the remaining time
1092 * in jiffies will be returned, or 0 if the timer expired in time
1093 *
1094 * The current task state is guaranteed to be TASK_RUNNING when this
1095 * routine returns.
1096 *
1097 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1098 * the CPU away without a bound on the timeout. In this case the return
1099 * value will be %MAX_SCHEDULE_TIMEOUT.
1100 *
1101 * In all cases the return value is guaranteed to be non-negative.
1102 */
1103fastcall signed long __sched schedule_timeout(signed long timeout)
1104{
1105 struct timer_list timer;
1106 unsigned long expire;
1107
1108 switch (timeout)
1109 {
1110 case MAX_SCHEDULE_TIMEOUT:
1111 /*
1112 * These two special cases are useful to be comfortable
1113 * in the caller. Nothing more. We could take
1114 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1115 * but I' d like to return a valid offset (>=0) to allow
1116 * the caller to do everything it want with the retval.
1117 */
1118 schedule();
1119 goto out;
1120 default:
1121 /*
1122 * Another bit of PARANOID. Note that the retval will be
1123 * 0 since no piece of kernel is supposed to do a check
1124 * for a negative retval of schedule_timeout() (since it
1125 * should never happens anyway). You just have the printk()
1126 * that will tell you if something is gone wrong and where.
1127 */
1128 if (timeout < 0)
1129 {
1130 printk(KERN_ERR "schedule_timeout: wrong timeout "
1131 "value %lx from %p\n", timeout,
1132 __builtin_return_address(0));
1133 current->state = TASK_RUNNING;
1134 goto out;
1135 }
1136 }
1137
1138 expire = timeout + jiffies;
1139
1140 init_timer(&timer);
1141 timer.expires = expire;
1142 timer.data = (unsigned long) current;
1143 timer.function = process_timeout;
1144
1145 add_timer(&timer);
1146 schedule();
1147 del_singleshot_timer_sync(&timer);
1148
1149 timeout = expire - jiffies;
1150
1151 out:
1152 return timeout < 0 ? 0 : timeout;
1153}
1154
1155EXPORT_SYMBOL(schedule_timeout);
1156
1157signed long __sched schedule_timeout_interruptible(signed long timeout)
1158{
1159 set_current_state(TASK_INTERRUPTIBLE);
1160 return schedule_timeout(timeout);
1161}
1162EXPORT_SYMBOL(schedule_timeout_interruptible);
1163
1164signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1165{
1166 set_current_state(TASK_UNINTERRUPTIBLE);
1167 return schedule_timeout(timeout);
1168}
1169EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1170
1171/* Thread ID - the internal kernel "pid" */
1172asmlinkage long sys_gettid(void)
1173{
1174 return current->pid;
1175}
1176
1177static long __sched nanosleep_restart(struct restart_block *restart)
1178{
1179 unsigned long expire = restart->arg0, now = jiffies;
1180 struct timespec __user *rmtp = (struct timespec __user *) restart->arg1;
1181 long ret;
1182
1183 /* Did it expire while we handled signals? */
1184 if (!time_after(expire, now))
1185 return 0;
1186
1187 expire = schedule_timeout_interruptible(expire - now);
1188
1189 ret = 0;
1190 if (expire) {
1191 struct timespec t;
1192 jiffies_to_timespec(expire, &t);
1193
1194 ret = -ERESTART_RESTARTBLOCK;
1195 if (rmtp && copy_to_user(rmtp, &t, sizeof(t)))
1196 ret = -EFAULT;
1197 /* The 'restart' block is already filled in */
1198 }
1199 return ret;
1200}
1201
1202asmlinkage long sys_nanosleep(struct timespec __user *rqtp, struct timespec __user *rmtp)
1203{
1204 struct timespec t;
1205 unsigned long expire;
1206 long ret;
1207
1208 if (copy_from_user(&t, rqtp, sizeof(t)))
1209 return -EFAULT;
1210
1211 if ((t.tv_nsec >= 1000000000L) || (t.tv_nsec < 0) || (t.tv_sec < 0))
1212 return -EINVAL;
1213
1214 expire = timespec_to_jiffies(&t) + (t.tv_sec || t.tv_nsec);
1215 expire = schedule_timeout_interruptible(expire);
1216
1217 ret = 0;
1218 if (expire) {
1219 struct restart_block *restart;
1220 jiffies_to_timespec(expire, &t);
1221 if (rmtp && copy_to_user(rmtp, &t, sizeof(t)))
1222 return -EFAULT;
1223
1224 restart = &current_thread_info()->restart_block;
1225 restart->fn = nanosleep_restart;
1226 restart->arg0 = jiffies + expire;
1227 restart->arg1 = (unsigned long) rmtp;
1228 ret = -ERESTART_RESTARTBLOCK;
1229 }
1230 return ret;
1231}
1232
1233/*
1234 * sys_sysinfo - fill in sysinfo struct
1235 */
1236asmlinkage long sys_sysinfo(struct sysinfo __user *info)
1237{
1238 struct sysinfo val;
1239 unsigned long mem_total, sav_total;
1240 unsigned int mem_unit, bitcount;
1241 unsigned long seq;
1242
1243 memset((char *)&val, 0, sizeof(struct sysinfo));
1244
1245 do {
1246 struct timespec tp;
1247 seq = read_seqbegin(&xtime_lock);
1248
1249 /*
1250 * This is annoying. The below is the same thing
1251 * posix_get_clock_monotonic() does, but it wants to
1252 * take the lock which we want to cover the loads stuff
1253 * too.
1254 */
1255
1256 getnstimeofday(&tp);
1257 tp.tv_sec += wall_to_monotonic.tv_sec;
1258 tp.tv_nsec += wall_to_monotonic.tv_nsec;
1259 if (tp.tv_nsec - NSEC_PER_SEC >= 0) {
1260 tp.tv_nsec = tp.tv_nsec - NSEC_PER_SEC;
1261 tp.tv_sec++;
1262 }
1263 val.uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
1264
1265 val.loads[0] = avenrun[0] << (SI_LOAD_SHIFT - FSHIFT);
1266 val.loads[1] = avenrun[1] << (SI_LOAD_SHIFT - FSHIFT);
1267 val.loads[2] = avenrun[2] << (SI_LOAD_SHIFT - FSHIFT);
1268
1269 val.procs = nr_threads;
1270 } while (read_seqretry(&xtime_lock, seq));
1271
1272 si_meminfo(&val);
1273 si_swapinfo(&val);
1274
1275 /*
1276 * If the sum of all the available memory (i.e. ram + swap)
1277 * is less than can be stored in a 32 bit unsigned long then
1278 * we can be binary compatible with 2.2.x kernels. If not,
1279 * well, in that case 2.2.x was broken anyways...
1280 *
1281 * -Erik Andersen <[email protected]>
1282 */
1283
1284 mem_total = val.totalram + val.totalswap;
1285 if (mem_total < val.totalram || mem_total < val.totalswap)
1286 goto out;
1287 bitcount = 0;
1288 mem_unit = val.mem_unit;
1289 while (mem_unit > 1) {
1290 bitcount++;
1291 mem_unit >>= 1;
1292 sav_total = mem_total;
1293 mem_total <<= 1;
1294 if (mem_total < sav_total)
1295 goto out;
1296 }
1297
1298 /*
1299 * If mem_total did not overflow, multiply all memory values by
1300 * val.mem_unit and set it to 1. This leaves things compatible
1301 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1302 * kernels...
1303 */
1304
1305 val.mem_unit = 1;
1306 val.totalram <<= bitcount;
1307 val.freeram <<= bitcount;
1308 val.sharedram <<= bitcount;
1309 val.bufferram <<= bitcount;
1310 val.totalswap <<= bitcount;
1311 val.freeswap <<= bitcount;
1312 val.totalhigh <<= bitcount;
1313 val.freehigh <<= bitcount;
1314
1315 out:
1316 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
1317 return -EFAULT;
1318
1319 return 0;
1320}
1321
1322static void __devinit init_timers_cpu(int cpu)
1323{
1324 int j;
1325 tvec_base_t *base;
1326
1327 base = &per_cpu(tvec_bases, cpu);
1328 spin_lock_init(&base->t_base.lock);
1329 for (j = 0; j < TVN_SIZE; j++) {
1330 INIT_LIST_HEAD(base->tv5.vec + j);
1331 INIT_LIST_HEAD(base->tv4.vec + j);
1332 INIT_LIST_HEAD(base->tv3.vec + j);
1333 INIT_LIST_HEAD(base->tv2.vec + j);
1334 }
1335 for (j = 0; j < TVR_SIZE; j++)
1336 INIT_LIST_HEAD(base->tv1.vec + j);
1337
1338 base->timer_jiffies = jiffies;
1339}
1340
1341#ifdef CONFIG_HOTPLUG_CPU
1342static void migrate_timer_list(tvec_base_t *new_base, struct list_head *head)
1343{
1344 struct timer_list *timer;
1345
1346 while (!list_empty(head)) {
1347 timer = list_entry(head->next, struct timer_list, entry);
1348 detach_timer(timer, 0);
1349 timer->base = &new_base->t_base;
1350 internal_add_timer(new_base, timer);
1351 }
1352}
1353
1354static void __devinit migrate_timers(int cpu)
1355{
1356 tvec_base_t *old_base;
1357 tvec_base_t *new_base;
1358 int i;
1359
1360 BUG_ON(cpu_online(cpu));
1361 old_base = &per_cpu(tvec_bases, cpu);
1362 new_base = &get_cpu_var(tvec_bases);
1363
1364 local_irq_disable();
1365 spin_lock(&new_base->t_base.lock);
1366 spin_lock(&old_base->t_base.lock);
1367
1368 if (old_base->t_base.running_timer)
1369 BUG();
1370 for (i = 0; i < TVR_SIZE; i++)
1371 migrate_timer_list(new_base, old_base->tv1.vec + i);
1372 for (i = 0; i < TVN_SIZE; i++) {
1373 migrate_timer_list(new_base, old_base->tv2.vec + i);
1374 migrate_timer_list(new_base, old_base->tv3.vec + i);
1375 migrate_timer_list(new_base, old_base->tv4.vec + i);
1376 migrate_timer_list(new_base, old_base->tv5.vec + i);
1377 }
1378
1379 spin_unlock(&old_base->t_base.lock);
1380 spin_unlock(&new_base->t_base.lock);
1381 local_irq_enable();
1382 put_cpu_var(tvec_bases);
1383}
1384#endif /* CONFIG_HOTPLUG_CPU */
1385
1386static int __devinit timer_cpu_notify(struct notifier_block *self,
1387 unsigned long action, void *hcpu)
1388{
1389 long cpu = (long)hcpu;
1390 switch(action) {
1391 case CPU_UP_PREPARE:
1392 init_timers_cpu(cpu);
1393 break;
1394#ifdef CONFIG_HOTPLUG_CPU
1395 case CPU_DEAD:
1396 migrate_timers(cpu);
1397 break;
1398#endif
1399 default:
1400 break;
1401 }
1402 return NOTIFY_OK;
1403}
1404
1405static struct notifier_block __devinitdata timers_nb = {
1406 .notifier_call = timer_cpu_notify,
1407};
1408
1409
1410void __init init_timers(void)
1411{
1412 timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE,
1413 (void *)(long)smp_processor_id());
1414 register_cpu_notifier(&timers_nb);
1415 open_softirq(TIMER_SOFTIRQ, run_timer_softirq, NULL);
1416}
1417
1418#ifdef CONFIG_TIME_INTERPOLATION
1419
1420struct time_interpolator *time_interpolator;
1421static struct time_interpolator *time_interpolator_list;
1422static DEFINE_SPINLOCK(time_interpolator_lock);
1423
1424static inline u64 time_interpolator_get_cycles(unsigned int src)
1425{
1426 unsigned long (*x)(void);
1427
1428 switch (src)
1429 {
1430 case TIME_SOURCE_FUNCTION:
1431 x = time_interpolator->addr;
1432 return x();
1433
1434 case TIME_SOURCE_MMIO64 :
1435 return readq((void __iomem *) time_interpolator->addr);
1436
1437 case TIME_SOURCE_MMIO32 :
1438 return readl((void __iomem *) time_interpolator->addr);
1439
1440 default: return get_cycles();
1441 }
1442}
1443
1444static inline u64 time_interpolator_get_counter(int writelock)
1445{
1446 unsigned int src = time_interpolator->source;
1447
1448 if (time_interpolator->jitter)
1449 {
1450 u64 lcycle;
1451 u64 now;
1452
1453 do {
1454 lcycle = time_interpolator->last_cycle;
1455 now = time_interpolator_get_cycles(src);
1456 if (lcycle && time_after(lcycle, now))
1457 return lcycle;
1458
1459 /* When holding the xtime write lock, there's no need
1460 * to add the overhead of the cmpxchg. Readers are
1461 * force to retry until the write lock is released.
1462 */
1463 if (writelock) {
1464 time_interpolator->last_cycle = now;
1465 return now;
1466 }
1467 /* Keep track of the last timer value returned. The use of cmpxchg here
1468 * will cause contention in an SMP environment.
1469 */
1470 } while (unlikely(cmpxchg(&time_interpolator->last_cycle, lcycle, now) != lcycle));
1471 return now;
1472 }
1473 else
1474 return time_interpolator_get_cycles(src);
1475}
1476
1477void time_interpolator_reset(void)
1478{
1479 time_interpolator->offset = 0;
1480 time_interpolator->last_counter = time_interpolator_get_counter(1);
1481}
1482
1483#define GET_TI_NSECS(count,i) (((((count) - i->last_counter) & (i)->mask) * (i)->nsec_per_cyc) >> (i)->shift)
1484
1485unsigned long time_interpolator_get_offset(void)
1486{
1487 /* If we do not have a time interpolator set up then just return zero */
1488 if (!time_interpolator)
1489 return 0;
1490
1491 return time_interpolator->offset +
1492 GET_TI_NSECS(time_interpolator_get_counter(0), time_interpolator);
1493}
1494
1495#define INTERPOLATOR_ADJUST 65536
1496#define INTERPOLATOR_MAX_SKIP 10*INTERPOLATOR_ADJUST
1497
1498static void time_interpolator_update(long delta_nsec)
1499{
1500 u64 counter;
1501 unsigned long offset;
1502
1503 /* If there is no time interpolator set up then do nothing */
1504 if (!time_interpolator)
1505 return;
1506
1507 /* The interpolator compensates for late ticks by accumulating
1508 * the late time in time_interpolator->offset. A tick earlier than
1509 * expected will lead to a reset of the offset and a corresponding
1510 * jump of the clock forward. Again this only works if the
1511 * interpolator clock is running slightly slower than the regular clock
1512 * and the tuning logic insures that.
1513 */
1514
1515 counter = time_interpolator_get_counter(1);
1516 offset = time_interpolator->offset + GET_TI_NSECS(counter, time_interpolator);
1517
1518 if (delta_nsec < 0 || (unsigned long) delta_nsec < offset)
1519 time_interpolator->offset = offset - delta_nsec;
1520 else {
1521 time_interpolator->skips++;
1522 time_interpolator->ns_skipped += delta_nsec - offset;
1523 time_interpolator->offset = 0;
1524 }
1525 time_interpolator->last_counter = counter;
1526
1527 /* Tuning logic for time interpolator invoked every minute or so.
1528 * Decrease interpolator clock speed if no skips occurred and an offset is carried.
1529 * Increase interpolator clock speed if we skip too much time.
1530 */
1531 if (jiffies % INTERPOLATOR_ADJUST == 0)
1532 {
1533 if (time_interpolator->skips == 0 && time_interpolator->offset > TICK_NSEC)
1534 time_interpolator->nsec_per_cyc--;
1535 if (time_interpolator->ns_skipped > INTERPOLATOR_MAX_SKIP && time_interpolator->offset == 0)
1536 time_interpolator->nsec_per_cyc++;
1537 time_interpolator->skips = 0;
1538 time_interpolator->ns_skipped = 0;
1539 }
1540}
1541
1542static inline int
1543is_better_time_interpolator(struct time_interpolator *new)
1544{
1545 if (!time_interpolator)
1546 return 1;
1547 return new->frequency > 2*time_interpolator->frequency ||
1548 (unsigned long)new->drift < (unsigned long)time_interpolator->drift;
1549}
1550
1551void
1552register_time_interpolator(struct time_interpolator *ti)
1553{
1554 unsigned long flags;
1555
1556 /* Sanity check */
1557 if (ti->frequency == 0 || ti->mask == 0)
1558 BUG();
1559
1560 ti->nsec_per_cyc = ((u64)NSEC_PER_SEC << ti->shift) / ti->frequency;
1561 spin_lock(&time_interpolator_lock);
1562 write_seqlock_irqsave(&xtime_lock, flags);
1563 if (is_better_time_interpolator(ti)) {
1564 time_interpolator = ti;
1565 time_interpolator_reset();
1566 }
1567 write_sequnlock_irqrestore(&xtime_lock, flags);
1568
1569 ti->next = time_interpolator_list;
1570 time_interpolator_list = ti;
1571 spin_unlock(&time_interpolator_lock);
1572}
1573
1574void
1575unregister_time_interpolator(struct time_interpolator *ti)
1576{
1577 struct time_interpolator *curr, **prev;
1578 unsigned long flags;
1579
1580 spin_lock(&time_interpolator_lock);
1581 prev = &time_interpolator_list;
1582 for (curr = *prev; curr; curr = curr->next) {
1583 if (curr == ti) {
1584 *prev = curr->next;
1585 break;
1586 }
1587 prev = &curr->next;
1588 }
1589
1590 write_seqlock_irqsave(&xtime_lock, flags);
1591 if (ti == time_interpolator) {
1592 /* we lost the best time-interpolator: */
1593 time_interpolator = NULL;
1594 /* find the next-best interpolator */
1595 for (curr = time_interpolator_list; curr; curr = curr->next)
1596 if (is_better_time_interpolator(curr))
1597 time_interpolator = curr;
1598 time_interpolator_reset();
1599 }
1600 write_sequnlock_irqrestore(&xtime_lock, flags);
1601 spin_unlock(&time_interpolator_lock);
1602}
1603#endif /* CONFIG_TIME_INTERPOLATION */
1604
1605/**
1606 * msleep - sleep safely even with waitqueue interruptions
1607 * @msecs: Time in milliseconds to sleep for
1608 */
1609void msleep(unsigned int msecs)
1610{
1611 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1612
1613 while (timeout)
1614 timeout = schedule_timeout_uninterruptible(timeout);
1615}
1616
1617EXPORT_SYMBOL(msleep);
1618
1619/**
1620 * msleep_interruptible - sleep waiting for signals
1621 * @msecs: Time in milliseconds to sleep for
1622 */
1623unsigned long msleep_interruptible(unsigned int msecs)
1624{
1625 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1626
1627 while (timeout && !signal_pending(current))
1628 timeout = schedule_timeout_interruptible(timeout);
1629 return jiffies_to_msecs(timeout);
1630}
1631
1632EXPORT_SYMBOL(msleep_interruptible);