projects/linux·blob v2.6.18-rc6 · refs

1912 lines · 49.4 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#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
45static void time_interpolator_update(long delta_nsec);
46#else
47#define time_interpolator_update(x)
48#endif
49
50u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
51
52EXPORT_SYMBOL(jiffies_64);
53
54/*
55 * per-CPU timer vector definitions:
56 */
57#define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6)
58#define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8)
59#define TVN_SIZE (1 << TVN_BITS)
60#define TVR_SIZE (1 << TVR_BITS)
61#define TVN_MASK (TVN_SIZE - 1)
62#define TVR_MASK (TVR_SIZE - 1)
63
64typedef struct tvec_s {
65 struct list_head vec[TVN_SIZE];
66} tvec_t;
67
68typedef struct tvec_root_s {
69 struct list_head vec[TVR_SIZE];
70} tvec_root_t;
71
72struct tvec_t_base_s {
73 spinlock_t lock;
74 struct timer_list *running_timer;
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;
84
85tvec_base_t boot_tvec_bases;
86EXPORT_SYMBOL(boot_tvec_bases);
87static DEFINE_PER_CPU(tvec_base_t *, tvec_bases) = &boot_tvec_bases;
88
89static inline void set_running_timer(tvec_base_t *base,
90 struct timer_list *timer)
91{
92#ifdef CONFIG_SMP
93 base->running_timer = timer;
94#endif
95}
96
97static void internal_add_timer(tvec_base_t *base, struct timer_list *timer)
98{
99 unsigned long expires = timer->expires;
100 unsigned long idx = expires - base->timer_jiffies;
101 struct list_head *vec;
102
103 if (idx < TVR_SIZE) {
104 int i = expires & TVR_MASK;
105 vec = base->tv1.vec + i;
106 } else if (idx < 1 << (TVR_BITS + TVN_BITS)) {
107 int i = (expires >> TVR_BITS) & TVN_MASK;
108 vec = base->tv2.vec + i;
109 } else if (idx < 1 << (TVR_BITS + 2 * TVN_BITS)) {
110 int i = (expires >> (TVR_BITS + TVN_BITS)) & TVN_MASK;
111 vec = base->tv3.vec + i;
112 } else if (idx < 1 << (TVR_BITS + 3 * TVN_BITS)) {
113 int i = (expires >> (TVR_BITS + 2 * TVN_BITS)) & TVN_MASK;
114 vec = base->tv4.vec + i;
115 } else if ((signed long) idx < 0) {
116 /*
117 * Can happen if you add a timer with expires == jiffies,
118 * or you set a timer to go off in the past
119 */
120 vec = base->tv1.vec + (base->timer_jiffies & TVR_MASK);
121 } else {
122 int i;
123 /* If the timeout is larger than 0xffffffff on 64-bit
124 * architectures then we use the maximum timeout:
125 */
126 if (idx > 0xffffffffUL) {
127 idx = 0xffffffffUL;
128 expires = idx + base->timer_jiffies;
129 }
130 i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK;
131 vec = base->tv5.vec + i;
132 }
133 /*
134 * Timers are FIFO:
135 */
136 list_add_tail(&timer->entry, vec);
137}
138
139/***
140 * init_timer - initialize a timer.
141 * @timer: the timer to be initialized
142 *
143 * init_timer() must be done to a timer prior calling *any* of the
144 * other timer functions.
145 */
146void fastcall init_timer(struct timer_list *timer)
147{
148 timer->entry.next = NULL;
149 timer->base = __raw_get_cpu_var(tvec_bases);
150}
151EXPORT_SYMBOL(init_timer);
152
153static inline void detach_timer(struct timer_list *timer,
154 int clear_pending)
155{
156 struct list_head *entry = &timer->entry;
157
158 __list_del(entry->prev, entry->next);
159 if (clear_pending)
160 entry->next = NULL;
161 entry->prev = LIST_POISON2;
162}
163
164/*
165 * We are using hashed locking: holding per_cpu(tvec_bases).lock
166 * means that all timers which are tied to this base via timer->base are
167 * locked, and the base itself is locked too.
168 *
169 * So __run_timers/migrate_timers can safely modify all timers which could
170 * be found on ->tvX lists.
171 *
172 * When the timer's base is locked, and the timer removed from list, it is
173 * possible to set timer->base = NULL and drop the lock: the timer remains
174 * locked.
175 */
176static tvec_base_t *lock_timer_base(struct timer_list *timer,
177 unsigned long *flags)
178{
179 tvec_base_t *base;
180
181 for (;;) {
182 base = timer->base;
183 if (likely(base != NULL)) {
184 spin_lock_irqsave(&base->lock, *flags);
185 if (likely(base == timer->base))
186 return base;
187 /* The timer has migrated to another CPU */
188 spin_unlock_irqrestore(&base->lock, *flags);
189 }
190 cpu_relax();
191 }
192}
193
194int __mod_timer(struct timer_list *timer, unsigned long expires)
195{
196 tvec_base_t *base, *new_base;
197 unsigned long flags;
198 int ret = 0;
199
200 BUG_ON(!timer->function);
201
202 base = lock_timer_base(timer, &flags);
203
204 if (timer_pending(timer)) {
205 detach_timer(timer, 0);
206 ret = 1;
207 }
208
209 new_base = __get_cpu_var(tvec_bases);
210
211 if (base != new_base) {
212 /*
213 * We are trying to schedule the timer on the local CPU.
214 * However we can't change timer's base while it is running,
215 * otherwise del_timer_sync() can't detect that the timer's
216 * handler yet has not finished. This also guarantees that
217 * the timer is serialized wrt itself.
218 */
219 if (likely(base->running_timer != timer)) {
220 /* See the comment in lock_timer_base() */
221 timer->base = NULL;
222 spin_unlock(&base->lock);
223 base = new_base;
224 spin_lock(&base->lock);
225 timer->base = base;
226 }
227 }
228
229 timer->expires = expires;
230 internal_add_timer(base, timer);
231 spin_unlock_irqrestore(&base->lock, flags);
232
233 return ret;
234}
235
236EXPORT_SYMBOL(__mod_timer);
237
238/***
239 * add_timer_on - start a timer on a particular CPU
240 * @timer: the timer to be added
241 * @cpu: the CPU to start it on
242 *
243 * This is not very scalable on SMP. Double adds are not possible.
244 */
245void add_timer_on(struct timer_list *timer, int cpu)
246{
247 tvec_base_t *base = per_cpu(tvec_bases, cpu);
248 unsigned long flags;
249
250 BUG_ON(timer_pending(timer) || !timer->function);
251 spin_lock_irqsave(&base->lock, flags);
252 timer->base = base;
253 internal_add_timer(base, timer);
254 spin_unlock_irqrestore(&base->lock, flags);
255}
256
257
258/***
259 * mod_timer - modify a timer's timeout
260 * @timer: the timer to be modified
261 *
262 * mod_timer is a more efficient way to update the expire field of an
263 * active timer (if the timer is inactive it will be activated)
264 *
265 * mod_timer(timer, expires) is equivalent to:
266 *
267 * del_timer(timer); timer->expires = expires; add_timer(timer);
268 *
269 * Note that if there are multiple unserialized concurrent users of the
270 * same timer, then mod_timer() is the only safe way to modify the timeout,
271 * since add_timer() cannot modify an already running timer.
272 *
273 * The function returns whether it has modified a pending timer or not.
274 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
275 * active timer returns 1.)
276 */
277int mod_timer(struct timer_list *timer, unsigned long expires)
278{
279 BUG_ON(!timer->function);
280
281 /*
282 * This is a common optimization triggered by the
283 * networking code - if the timer is re-modified
284 * to be the same thing then just return:
285 */
286 if (timer->expires == expires && timer_pending(timer))
287 return 1;
288
289 return __mod_timer(timer, expires);
290}
291
292EXPORT_SYMBOL(mod_timer);
293
294/***
295 * del_timer - deactive a timer.
296 * @timer: the timer to be deactivated
297 *
298 * del_timer() deactivates a timer - this works on both active and inactive
299 * timers.
300 *
301 * The function returns whether it has deactivated a pending timer or not.
302 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
303 * active timer returns 1.)
304 */
305int del_timer(struct timer_list *timer)
306{
307 tvec_base_t *base;
308 unsigned long flags;
309 int ret = 0;
310
311 if (timer_pending(timer)) {
312 base = lock_timer_base(timer, &flags);
313 if (timer_pending(timer)) {
314 detach_timer(timer, 1);
315 ret = 1;
316 }
317 spin_unlock_irqrestore(&base->lock, flags);
318 }
319
320 return ret;
321}
322
323EXPORT_SYMBOL(del_timer);
324
325#ifdef CONFIG_SMP
326/*
327 * This function tries to deactivate a timer. Upon successful (ret >= 0)
328 * exit the timer is not queued and the handler is not running on any CPU.
329 *
330 * It must not be called from interrupt contexts.
331 */
332int try_to_del_timer_sync(struct timer_list *timer)
333{
334 tvec_base_t *base;
335 unsigned long flags;
336 int ret = -1;
337
338 base = lock_timer_base(timer, &flags);
339
340 if (base->running_timer == timer)
341 goto out;
342
343 ret = 0;
344 if (timer_pending(timer)) {
345 detach_timer(timer, 1);
346 ret = 1;
347 }
348out:
349 spin_unlock_irqrestore(&base->lock, flags);
350
351 return ret;
352}
353
354/***
355 * del_timer_sync - deactivate a timer and wait for the handler to finish.
356 * @timer: the timer to be deactivated
357 *
358 * This function only differs from del_timer() on SMP: besides deactivating
359 * the timer it also makes sure the handler has finished executing on other
360 * CPUs.
361 *
362 * Synchronization rules: callers must prevent restarting of the timer,
363 * otherwise this function is meaningless. It must not be called from
364 * interrupt contexts. The caller must not hold locks which would prevent
365 * completion of the timer's handler. The timer's handler must not call
366 * add_timer_on(). Upon exit the timer is not queued and the handler is
367 * not running on any CPU.
368 *
369 * The function returns whether it has deactivated a pending timer or not.
370 */
371int del_timer_sync(struct timer_list *timer)
372{
373 for (;;) {
374 int ret = try_to_del_timer_sync(timer);
375 if (ret >= 0)
376 return ret;
377 cpu_relax();
378 }
379}
380
381EXPORT_SYMBOL(del_timer_sync);
382#endif
383
384static int cascade(tvec_base_t *base, tvec_t *tv, int index)
385{
386 /* cascade all the timers from tv up one level */
387 struct timer_list *timer, *tmp;
388 struct list_head tv_list;
389
390 list_replace_init(tv->vec + index, &tv_list);
391
392 /*
393 * We are removing _all_ timers from the list, so we
394 * don't have to detach them individually.
395 */
396 list_for_each_entry_safe(timer, tmp, &tv_list, entry) {
397 BUG_ON(timer->base != base);
398 internal_add_timer(base, timer);
399 }
400
401 return index;
402}
403
404/***
405 * __run_timers - run all expired timers (if any) on this CPU.
406 * @base: the timer vector to be processed.
407 *
408 * This function cascades all vectors and executes all expired timer
409 * vectors.
410 */
411#define INDEX(N) ((base->timer_jiffies >> (TVR_BITS + (N) * TVN_BITS)) & TVN_MASK)
412
413static inline void __run_timers(tvec_base_t *base)
414{
415 struct timer_list *timer;
416
417 spin_lock_irq(&base->lock);
418 while (time_after_eq(jiffies, base->timer_jiffies)) {
419 struct list_head work_list;
420 struct list_head *head = &work_list;
421 int index = base->timer_jiffies & TVR_MASK;
422
423 /*
424 * Cascade timers:
425 */
426 if (!index &&
427 (!cascade(base, &base->tv2, INDEX(0))) &&
428 (!cascade(base, &base->tv3, INDEX(1))) &&
429 !cascade(base, &base->tv4, INDEX(2)))
430 cascade(base, &base->tv5, INDEX(3));
431 ++base->timer_jiffies;
432 list_replace_init(base->tv1.vec + index, &work_list);
433 while (!list_empty(head)) {
434 void (*fn)(unsigned long);
435 unsigned long data;
436
437 timer = list_entry(head->next,struct timer_list,entry);
438 fn = timer->function;
439 data = timer->data;
440
441 set_running_timer(base, timer);
442 detach_timer(timer, 1);
443 spin_unlock_irq(&base->lock);
444 {
445 int preempt_count = preempt_count();
446 fn(data);
447 if (preempt_count != preempt_count()) {
448 printk(KERN_WARNING "huh, entered %p "
449 "with preempt_count %08x, exited"
450 " with %08x?\n",
451 fn, preempt_count,
452 preempt_count());
453 BUG();
454 }
455 }
456 spin_lock_irq(&base->lock);
457 }
458 }
459 set_running_timer(base, NULL);
460 spin_unlock_irq(&base->lock);
461}
462
463#ifdef CONFIG_NO_IDLE_HZ
464/*
465 * Find out when the next timer event is due to happen. This
466 * is used on S/390 to stop all activity when a cpus is idle.
467 * This functions needs to be called disabled.
468 */
469unsigned long next_timer_interrupt(void)
470{
471 tvec_base_t *base;
472 struct list_head *list;
473 struct timer_list *nte;
474 unsigned long expires;
475 unsigned long hr_expires = MAX_JIFFY_OFFSET;
476 ktime_t hr_delta;
477 tvec_t *varray[4];
478 int i, j;
479
480 hr_delta = hrtimer_get_next_event();
481 if (hr_delta.tv64 != KTIME_MAX) {
482 struct timespec tsdelta;
483 tsdelta = ktime_to_timespec(hr_delta);
484 hr_expires = timespec_to_jiffies(&tsdelta);
485 if (hr_expires < 3)
486 return hr_expires + jiffies;
487 }
488 hr_expires += jiffies;
489
490 base = __get_cpu_var(tvec_bases);
491 spin_lock(&base->lock);
492 expires = base->timer_jiffies + (LONG_MAX >> 1);
493 list = NULL;
494
495 /* Look for timer events in tv1. */
496 j = base->timer_jiffies & TVR_MASK;
497 do {
498 list_for_each_entry(nte, base->tv1.vec + j, entry) {
499 expires = nte->expires;
500 if (j < (base->timer_jiffies & TVR_MASK))
501 list = base->tv2.vec + (INDEX(0));
502 goto found;
503 }
504 j = (j + 1) & TVR_MASK;
505 } while (j != (base->timer_jiffies & TVR_MASK));
506
507 /* Check tv2-tv5. */
508 varray[0] = &base->tv2;
509 varray[1] = &base->tv3;
510 varray[2] = &base->tv4;
511 varray[3] = &base->tv5;
512 for (i = 0; i < 4; i++) {
513 j = INDEX(i);
514 do {
515 if (list_empty(varray[i]->vec + j)) {
516 j = (j + 1) & TVN_MASK;
517 continue;
518 }
519 list_for_each_entry(nte, varray[i]->vec + j, entry)
520 if (time_before(nte->expires, expires))
521 expires = nte->expires;
522 if (j < (INDEX(i)) && i < 3)
523 list = varray[i + 1]->vec + (INDEX(i + 1));
524 goto found;
525 } while (j != (INDEX(i)));
526 }
527found:
528 if (list) {
529 /*
530 * The search wrapped. We need to look at the next list
531 * from next tv element that would cascade into tv element
532 * where we found the timer element.
533 */
534 list_for_each_entry(nte, list, entry) {
535 if (time_before(nte->expires, expires))
536 expires = nte->expires;
537 }
538 }
539 spin_unlock(&base->lock);
540
541 /*
542 * It can happen that other CPUs service timer IRQs and increment
543 * jiffies, but we have not yet got a local timer tick to process
544 * the timer wheels. In that case, the expiry time can be before
545 * jiffies, but since the high-resolution timer here is relative to
546 * jiffies, the default expression when high-resolution timers are
547 * not active,
548 *
549 * time_before(MAX_JIFFY_OFFSET + jiffies, expires)
550 *
551 * would falsely evaluate to true. If that is the case, just
552 * return jiffies so that we can immediately fire the local timer
553 */
554 if (time_before(expires, jiffies))
555 return jiffies;
556
557 if (time_before(hr_expires, expires))
558 return hr_expires;
559
560 return expires;
561}
562#endif
563
564/******************************************************************/
565
566/*
567 * Timekeeping variables
568 */
569unsigned long tick_usec = TICK_USEC; /* USER_HZ period (usec) */
570unsigned long tick_nsec = TICK_NSEC; /* ACTHZ period (nsec) */
571
572/*
573 * The current time
574 * wall_to_monotonic is what we need to add to xtime (or xtime corrected
575 * for sub jiffie times) to get to monotonic time. Monotonic is pegged
576 * at zero at system boot time, so wall_to_monotonic will be negative,
577 * however, we will ALWAYS keep the tv_nsec part positive so we can use
578 * the usual normalization.
579 */
580struct timespec xtime __attribute__ ((aligned (16)));
581struct timespec wall_to_monotonic __attribute__ ((aligned (16)));
582
583EXPORT_SYMBOL(xtime);
584
585/* Don't completely fail for HZ > 500. */
586int tickadj = 500/HZ ? : 1; /* microsecs */
587
588
589/*
590 * phase-lock loop variables
591 */
592/* TIME_ERROR prevents overwriting the CMOS clock */
593int time_state = TIME_OK; /* clock synchronization status */
594int time_status = STA_UNSYNC; /* clock status bits */
595long time_offset; /* time adjustment (us) */
596long time_constant = 2; /* pll time constant */
597long time_tolerance = MAXFREQ; /* frequency tolerance (ppm) */
598long time_precision = 1; /* clock precision (us) */
599long time_maxerror = NTP_PHASE_LIMIT; /* maximum error (us) */
600long time_esterror = NTP_PHASE_LIMIT; /* estimated error (us) */
601long time_freq = (((NSEC_PER_SEC + HZ/2) % HZ - HZ/2) << SHIFT_USEC) / NSEC_PER_USEC;
602 /* frequency offset (scaled ppm)*/
603static long time_adj; /* tick adjust (scaled 1 / HZ) */
604long time_reftime; /* time at last adjustment (s) */
605long time_adjust;
606long time_next_adjust;
607
608/*
609 * this routine handles the overflow of the microsecond field
610 *
611 * The tricky bits of code to handle the accurate clock support
612 * were provided by Dave Mills ([email protected]) of NTP fame.
613 * They were originally developed for SUN and DEC kernels.
614 * All the kudos should go to Dave for this stuff.
615 *
616 */
617static void second_overflow(void)
618{
619 long ltemp;
620
621 /* Bump the maxerror field */
622 time_maxerror += time_tolerance >> SHIFT_USEC;
623 if (time_maxerror > NTP_PHASE_LIMIT) {
624 time_maxerror = NTP_PHASE_LIMIT;
625 time_status |= STA_UNSYNC;
626 }
627
628 /*
629 * Leap second processing. If in leap-insert state at the end of the
630 * day, the system clock is set back one second; if in leap-delete
631 * state, the system clock is set ahead one second. The microtime()
632 * routine or external clock driver will insure that reported time is
633 * always monotonic. The ugly divides should be replaced.
634 */
635 switch (time_state) {
636 case TIME_OK:
637 if (time_status & STA_INS)
638 time_state = TIME_INS;
639 else if (time_status & STA_DEL)
640 time_state = TIME_DEL;
641 break;
642 case TIME_INS:
643 if (xtime.tv_sec % 86400 == 0) {
644 xtime.tv_sec--;
645 wall_to_monotonic.tv_sec++;
646 /*
647 * The timer interpolator will make time change
648 * gradually instead of an immediate jump by one second
649 */
650 time_interpolator_update(-NSEC_PER_SEC);
651 time_state = TIME_OOP;
652 clock_was_set();
653 printk(KERN_NOTICE "Clock: inserting leap second "
654 "23:59:60 UTC\n");
655 }
656 break;
657 case TIME_DEL:
658 if ((xtime.tv_sec + 1) % 86400 == 0) {
659 xtime.tv_sec++;
660 wall_to_monotonic.tv_sec--;
661 /*
662 * Use of time interpolator for a gradual change of
663 * time
664 */
665 time_interpolator_update(NSEC_PER_SEC);
666 time_state = TIME_WAIT;
667 clock_was_set();
668 printk(KERN_NOTICE "Clock: deleting leap second "
669 "23:59:59 UTC\n");
670 }
671 break;
672 case TIME_OOP:
673 time_state = TIME_WAIT;
674 break;
675 case TIME_WAIT:
676 if (!(time_status & (STA_INS | STA_DEL)))
677 time_state = TIME_OK;
678 }
679
680 /*
681 * Compute the phase adjustment for the next second. In PLL mode, the
682 * offset is reduced by a fixed factor times the time constant. In FLL
683 * mode the offset is used directly. In either mode, the maximum phase
684 * adjustment for each second is clamped so as to spread the adjustment
685 * over not more than the number of seconds between updates.
686 */
687 ltemp = time_offset;
688 if (!(time_status & STA_FLL))
689 ltemp = shift_right(ltemp, SHIFT_KG + time_constant);
690 ltemp = min(ltemp, (MAXPHASE / MINSEC) << SHIFT_UPDATE);
691 ltemp = max(ltemp, -(MAXPHASE / MINSEC) << SHIFT_UPDATE);
692 time_offset -= ltemp;
693 time_adj = ltemp << (SHIFT_SCALE - SHIFT_HZ - SHIFT_UPDATE);
694
695 /*
696 * Compute the frequency estimate and additional phase adjustment due
697 * to frequency error for the next second.
698 */
699 ltemp = time_freq;
700 time_adj += shift_right(ltemp,(SHIFT_USEC + SHIFT_HZ - SHIFT_SCALE));
701
702#if HZ == 100
703 /*
704 * Compensate for (HZ==100) != (1 << SHIFT_HZ). Add 25% and 3.125% to
705 * get 128.125; => only 0.125% error (p. 14)
706 */
707 time_adj += shift_right(time_adj, 2) + shift_right(time_adj, 5);
708#endif
709#if HZ == 250
710 /*
711 * Compensate for (HZ==250) != (1 << SHIFT_HZ). Add 1.5625% and
712 * 0.78125% to get 255.85938; => only 0.05% error (p. 14)
713 */
714 time_adj += shift_right(time_adj, 6) + shift_right(time_adj, 7);
715#endif
716#if HZ == 1000
717 /*
718 * Compensate for (HZ==1000) != (1 << SHIFT_HZ). Add 1.5625% and
719 * 0.78125% to get 1023.4375; => only 0.05% error (p. 14)
720 */
721 time_adj += shift_right(time_adj, 6) + shift_right(time_adj, 7);
722#endif
723}
724
725/*
726 * Returns how many microseconds we need to add to xtime this tick
727 * in doing an adjustment requested with adjtime.
728 */
729static long adjtime_adjustment(void)
730{
731 long time_adjust_step;
732
733 time_adjust_step = time_adjust;
734 if (time_adjust_step) {
735 /*
736 * We are doing an adjtime thing. Prepare time_adjust_step to
737 * be within bounds. Note that a positive time_adjust means we
738 * want the clock to run faster.
739 *
740 * Limit the amount of the step to be in the range
741 * -tickadj .. +tickadj
742 */
743 time_adjust_step = min(time_adjust_step, (long)tickadj);
744 time_adjust_step = max(time_adjust_step, (long)-tickadj);
745 }
746 return time_adjust_step;
747}
748
749/* in the NTP reference this is called "hardclock()" */
750static void update_ntp_one_tick(void)
751{
752 long time_adjust_step;
753
754 time_adjust_step = adjtime_adjustment();
755 if (time_adjust_step)
756 /* Reduce by this step the amount of time left */
757 time_adjust -= time_adjust_step;
758
759 /* Changes by adjtime() do not take effect till next tick. */
760 if (time_next_adjust != 0) {
761 time_adjust = time_next_adjust;
762 time_next_adjust = 0;
763 }
764}
765
766/*
767 * Return how long ticks are at the moment, that is, how much time
768 * update_wall_time_one_tick will add to xtime next time we call it
769 * (assuming no calls to do_adjtimex in the meantime).
770 * The return value is in fixed-point nanoseconds shifted by the
771 * specified number of bits to the right of the binary point.
772 * This function has no side-effects.
773 */
774u64 current_tick_length(void)
775{
776 long delta_nsec;
777 u64 ret;
778
779 /* calculate the finest interval NTP will allow.
780 * ie: nanosecond value shifted by (SHIFT_SCALE - 10)
781 */
782 delta_nsec = tick_nsec + adjtime_adjustment() * 1000;
783 ret = (u64)delta_nsec << TICK_LENGTH_SHIFT;
784 ret += (s64)time_adj << (TICK_LENGTH_SHIFT - (SHIFT_SCALE - 10));
785
786 return ret;
787}
788
789/* XXX - all of this timekeeping code should be later moved to time.c */
790#include <linux/clocksource.h>
791static struct clocksource *clock; /* pointer to current clocksource */
792
793#ifdef CONFIG_GENERIC_TIME
794/**
795 * __get_nsec_offset - Returns nanoseconds since last call to periodic_hook
796 *
797 * private function, must hold xtime_lock lock when being
798 * called. Returns the number of nanoseconds since the
799 * last call to update_wall_time() (adjusted by NTP scaling)
800 */
801static inline s64 __get_nsec_offset(void)
802{
803 cycle_t cycle_now, cycle_delta;
804 s64 ns_offset;
805
806 /* read clocksource: */
807 cycle_now = clocksource_read(clock);
808
809 /* calculate the delta since the last update_wall_time: */
810 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
811
812 /* convert to nanoseconds: */
813 ns_offset = cyc2ns(clock, cycle_delta);
814
815 return ns_offset;
816}
817
818/**
819 * __get_realtime_clock_ts - Returns the time of day in a timespec
820 * @ts: pointer to the timespec to be set
821 *
822 * Returns the time of day in a timespec. Used by
823 * do_gettimeofday() and get_realtime_clock_ts().
824 */
825static inline void __get_realtime_clock_ts(struct timespec *ts)
826{
827 unsigned long seq;
828 s64 nsecs;
829
830 do {
831 seq = read_seqbegin(&xtime_lock);
832
833 *ts = xtime;
834 nsecs = __get_nsec_offset();
835
836 } while (read_seqretry(&xtime_lock, seq));
837
838 timespec_add_ns(ts, nsecs);
839}
840
841/**
842 * getnstimeofday - Returns the time of day in a timespec
843 * @ts: pointer to the timespec to be set
844 *
845 * Returns the time of day in a timespec.
846 */
847void getnstimeofday(struct timespec *ts)
848{
849 __get_realtime_clock_ts(ts);
850}
851
852EXPORT_SYMBOL(getnstimeofday);
853
854/**
855 * do_gettimeofday - Returns the time of day in a timeval
856 * @tv: pointer to the timeval to be set
857 *
858 * NOTE: Users should be converted to using get_realtime_clock_ts()
859 */
860void do_gettimeofday(struct timeval *tv)
861{
862 struct timespec now;
863
864 __get_realtime_clock_ts(&now);
865 tv->tv_sec = now.tv_sec;
866 tv->tv_usec = now.tv_nsec/1000;
867}
868
869EXPORT_SYMBOL(do_gettimeofday);
870/**
871 * do_settimeofday - Sets the time of day
872 * @tv: pointer to the timespec variable containing the new time
873 *
874 * Sets the time of day to the new time and update NTP and notify hrtimers
875 */
876int do_settimeofday(struct timespec *tv)
877{
878 unsigned long flags;
879 time_t wtm_sec, sec = tv->tv_sec;
880 long wtm_nsec, nsec = tv->tv_nsec;
881
882 if ((unsigned long)tv->tv_nsec >= NSEC_PER_SEC)
883 return -EINVAL;
884
885 write_seqlock_irqsave(&xtime_lock, flags);
886
887 nsec -= __get_nsec_offset();
888
889 wtm_sec = wall_to_monotonic.tv_sec + (xtime.tv_sec - sec);
890 wtm_nsec = wall_to_monotonic.tv_nsec + (xtime.tv_nsec - nsec);
891
892 set_normalized_timespec(&xtime, sec, nsec);
893 set_normalized_timespec(&wall_to_monotonic, wtm_sec, wtm_nsec);
894
895 clock->error = 0;
896 ntp_clear();
897
898 write_sequnlock_irqrestore(&xtime_lock, flags);
899
900 /* signal hrtimers about time change */
901 clock_was_set();
902
903 return 0;
904}
905
906EXPORT_SYMBOL(do_settimeofday);
907
908/**
909 * change_clocksource - Swaps clocksources if a new one is available
910 *
911 * Accumulates current time interval and initializes new clocksource
912 */
913static int change_clocksource(void)
914{
915 struct clocksource *new;
916 cycle_t now;
917 u64 nsec;
918 new = clocksource_get_next();
919 if (clock != new) {
920 now = clocksource_read(new);
921 nsec = __get_nsec_offset();
922 timespec_add_ns(&xtime, nsec);
923
924 clock = new;
925 clock->cycle_last = now;
926 printk(KERN_INFO "Time: %s clocksource has been installed.\n",
927 clock->name);
928 return 1;
929 } else if (clock->update_callback) {
930 return clock->update_callback();
931 }
932 return 0;
933}
934#else
935#define change_clocksource() (0)
936#endif
937
938/**
939 * timeofday_is_continuous - check to see if timekeeping is free running
940 */
941int timekeeping_is_continuous(void)
942{
943 unsigned long seq;
944 int ret;
945
946 do {
947 seq = read_seqbegin(&xtime_lock);
948
949 ret = clock->is_continuous;
950
951 } while (read_seqretry(&xtime_lock, seq));
952
953 return ret;
954}
955
956/*
957 * timekeeping_init - Initializes the clocksource and common timekeeping values
958 */
959void __init timekeeping_init(void)
960{
961 unsigned long flags;
962
963 write_seqlock_irqsave(&xtime_lock, flags);
964 clock = clocksource_get_next();
965 clocksource_calculate_interval(clock, tick_nsec);
966 clock->cycle_last = clocksource_read(clock);
967 ntp_clear();
968 write_sequnlock_irqrestore(&xtime_lock, flags);
969}
970
971
972static int timekeeping_suspended;
973/*
974 * timekeeping_resume - Resumes the generic timekeeping subsystem.
975 * @dev: unused
976 *
977 * This is for the generic clocksource timekeeping.
978 * xtime/wall_to_monotonic/jiffies/wall_jiffies/etc are
979 * still managed by arch specific suspend/resume code.
980 */
981static int timekeeping_resume(struct sys_device *dev)
982{
983 unsigned long flags;
984
985 write_seqlock_irqsave(&xtime_lock, flags);
986 /* restart the last cycle value */
987 clock->cycle_last = clocksource_read(clock);
988 clock->error = 0;
989 timekeeping_suspended = 0;
990 write_sequnlock_irqrestore(&xtime_lock, flags);
991 return 0;
992}
993
994static int timekeeping_suspend(struct sys_device *dev, pm_message_t state)
995{
996 unsigned long flags;
997
998 write_seqlock_irqsave(&xtime_lock, flags);
999 timekeeping_suspended = 1;
1000 write_sequnlock_irqrestore(&xtime_lock, flags);
1001 return 0;
1002}
1003
1004/* sysfs resume/suspend bits for timekeeping */
1005static struct sysdev_class timekeeping_sysclass = {
1006 .resume = timekeeping_resume,
1007 .suspend = timekeeping_suspend,
1008 set_kset_name("timekeeping"),
1009};
1010
1011static struct sys_device device_timer = {
1012 .id = 0,
1013 .cls = &timekeeping_sysclass,
1014};
1015
1016static int __init timekeeping_init_device(void)
1017{
1018 int error = sysdev_class_register(&timekeeping_sysclass);
1019 if (!error)
1020 error = sysdev_register(&device_timer);
1021 return error;
1022}
1023
1024device_initcall(timekeeping_init_device);
1025
1026/*
1027 * If the error is already larger, we look ahead even further
1028 * to compensate for late or lost adjustments.
1029 */
1030static __always_inline int clocksource_bigadjust(s64 error, s64 *interval, s64 *offset)
1031{
1032 s64 tick_error, i;
1033 u32 look_ahead, adj;
1034 s32 error2, mult;
1035
1036 /*
1037 * Use the current error value to determine how much to look ahead.
1038 * The larger the error the slower we adjust for it to avoid problems
1039 * with losing too many ticks, otherwise we would overadjust and
1040 * produce an even larger error. The smaller the adjustment the
1041 * faster we try to adjust for it, as lost ticks can do less harm
1042 * here. This is tuned so that an error of about 1 msec is adusted
1043 * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks).
1044 */
1045 error2 = clock->error >> (TICK_LENGTH_SHIFT + 22 - 2 * SHIFT_HZ);
1046 error2 = abs(error2);
1047 for (look_ahead = 0; error2 > 0; look_ahead++)
1048 error2 >>= 2;
1049
1050 /*
1051 * Now calculate the error in (1 << look_ahead) ticks, but first
1052 * remove the single look ahead already included in the error.
1053 */
1054 tick_error = current_tick_length() >> (TICK_LENGTH_SHIFT - clock->shift + 1);
1055 tick_error -= clock->xtime_interval >> 1;
1056 error = ((error - tick_error) >> look_ahead) + tick_error;
1057
1058 /* Finally calculate the adjustment shift value. */
1059 i = *interval;
1060 mult = 1;
1061 if (error < 0) {
1062 error = -error;
1063 *interval = -*interval;
1064 *offset = -*offset;
1065 mult = -1;
1066 }
1067 for (adj = 0; error > i; adj++)
1068 error >>= 1;
1069
1070 *interval <<= adj;
1071 *offset <<= adj;
1072 return mult << adj;
1073}
1074
1075/*
1076 * Adjust the multiplier to reduce the error value,
1077 * this is optimized for the most common adjustments of -1,0,1,
1078 * for other values we can do a bit more work.
1079 */
1080static void clocksource_adjust(struct clocksource *clock, s64 offset)
1081{
1082 s64 error, interval = clock->cycle_interval;
1083 int adj;
1084
1085 error = clock->error >> (TICK_LENGTH_SHIFT - clock->shift - 1);
1086 if (error > interval) {
1087 error >>= 2;
1088 if (likely(error <= interval))
1089 adj = 1;
1090 else
1091 adj = clocksource_bigadjust(error, &interval, &offset);
1092 } else if (error < -interval) {
1093 error >>= 2;
1094 if (likely(error >= -interval)) {
1095 adj = -1;
1096 interval = -interval;
1097 offset = -offset;
1098 } else
1099 adj = clocksource_bigadjust(error, &interval, &offset);
1100 } else
1101 return;
1102
1103 clock->mult += adj;
1104 clock->xtime_interval += interval;
1105 clock->xtime_nsec -= offset;
1106 clock->error -= (interval - offset) << (TICK_LENGTH_SHIFT - clock->shift);
1107}
1108
1109/*
1110 * update_wall_time - Uses the current clocksource to increment the wall time
1111 *
1112 * Called from the timer interrupt, must hold a write on xtime_lock.
1113 */
1114static void update_wall_time(void)
1115{
1116 cycle_t offset;
1117
1118 /* Make sure we're fully resumed: */
1119 if (unlikely(timekeeping_suspended))
1120 return;
1121
1122#ifdef CONFIG_GENERIC_TIME
1123 offset = (clocksource_read(clock) - clock->cycle_last) & clock->mask;
1124#else
1125 offset = clock->cycle_interval;
1126#endif
1127 clock->xtime_nsec += (s64)xtime.tv_nsec << clock->shift;
1128
1129 /* normally this loop will run just once, however in the
1130 * case of lost or late ticks, it will accumulate correctly.
1131 */
1132 while (offset >= clock->cycle_interval) {
1133 /* accumulate one interval */
1134 clock->xtime_nsec += clock->xtime_interval;
1135 clock->cycle_last += clock->cycle_interval;
1136 offset -= clock->cycle_interval;
1137
1138 if (clock->xtime_nsec >= (u64)NSEC_PER_SEC << clock->shift) {
1139 clock->xtime_nsec -= (u64)NSEC_PER_SEC << clock->shift;
1140 xtime.tv_sec++;
1141 second_overflow();
1142 }
1143
1144 /* interpolator bits */
1145 time_interpolator_update(clock->xtime_interval
1146 >> clock->shift);
1147 /* increment the NTP state machine */
1148 update_ntp_one_tick();
1149
1150 /* accumulate error between NTP and clock interval */
1151 clock->error += current_tick_length();
1152 clock->error -= clock->xtime_interval << (TICK_LENGTH_SHIFT - clock->shift);
1153 }
1154
1155 /* correct the clock when NTP error is too big */
1156 clocksource_adjust(clock, offset);
1157
1158 /* store full nanoseconds into xtime */
1159 xtime.tv_nsec = (s64)clock->xtime_nsec >> clock->shift;
1160 clock->xtime_nsec -= (s64)xtime.tv_nsec << clock->shift;
1161
1162 /* check to see if there is a new clocksource to use */
1163 if (change_clocksource()) {
1164 clock->error = 0;
1165 clock->xtime_nsec = 0;
1166 clocksource_calculate_interval(clock, tick_nsec);
1167 }
1168}
1169
1170/*
1171 * Called from the timer interrupt handler to charge one tick to the current
1172 * process. user_tick is 1 if the tick is user time, 0 for system.
1173 */
1174void update_process_times(int user_tick)
1175{
1176 struct task_struct *p = current;
1177 int cpu = smp_processor_id();
1178
1179 /* Note: this timer irq context must be accounted for as well. */
1180 if (user_tick)
1181 account_user_time(p, jiffies_to_cputime(1));
1182 else
1183 account_system_time(p, HARDIRQ_OFFSET, jiffies_to_cputime(1));
1184 run_local_timers();
1185 if (rcu_pending(cpu))
1186 rcu_check_callbacks(cpu, user_tick);
1187 scheduler_tick();
1188 run_posix_cpu_timers(p);
1189}
1190
1191/*
1192 * Nr of active tasks - counted in fixed-point numbers
1193 */
1194static unsigned long count_active_tasks(void)
1195{
1196 return nr_active() * FIXED_1;
1197}
1198
1199/*
1200 * Hmm.. Changed this, as the GNU make sources (load.c) seems to
1201 * imply that avenrun[] is the standard name for this kind of thing.
1202 * Nothing else seems to be standardized: the fractional size etc
1203 * all seem to differ on different machines.
1204 *
1205 * Requires xtime_lock to access.
1206 */
1207unsigned long avenrun[3];
1208
1209EXPORT_SYMBOL(avenrun);
1210
1211/*
1212 * calc_load - given tick count, update the avenrun load estimates.
1213 * This is called while holding a write_lock on xtime_lock.
1214 */
1215static inline void calc_load(unsigned long ticks)
1216{
1217 unsigned long active_tasks; /* fixed-point */
1218 static int count = LOAD_FREQ;
1219
1220 count -= ticks;
1221 if (count < 0) {
1222 count += LOAD_FREQ;
1223 active_tasks = count_active_tasks();
1224 CALC_LOAD(avenrun[0], EXP_1, active_tasks);
1225 CALC_LOAD(avenrun[1], EXP_5, active_tasks);
1226 CALC_LOAD(avenrun[2], EXP_15, active_tasks);
1227 }
1228}
1229
1230/* jiffies at the most recent update of wall time */
1231unsigned long wall_jiffies = INITIAL_JIFFIES;
1232
1233/*
1234 * This read-write spinlock protects us from races in SMP while
1235 * playing with xtime and avenrun.
1236 */
1237#ifndef ARCH_HAVE_XTIME_LOCK
1238__cacheline_aligned_in_smp DEFINE_SEQLOCK(xtime_lock);
1239
1240EXPORT_SYMBOL(xtime_lock);
1241#endif
1242
1243/*
1244 * This function runs timers and the timer-tq in bottom half context.
1245 */
1246static void run_timer_softirq(struct softirq_action *h)
1247{
1248 tvec_base_t *base = __get_cpu_var(tvec_bases);
1249
1250 hrtimer_run_queues();
1251 if (time_after_eq(jiffies, base->timer_jiffies))
1252 __run_timers(base);
1253}
1254
1255/*
1256 * Called by the local, per-CPU timer interrupt on SMP.
1257 */
1258void run_local_timers(void)
1259{
1260 raise_softirq(TIMER_SOFTIRQ);
1261 softlockup_tick();
1262}
1263
1264/*
1265 * Called by the timer interrupt. xtime_lock must already be taken
1266 * by the timer IRQ!
1267 */
1268static inline void update_times(void)
1269{
1270 unsigned long ticks;
1271
1272 ticks = jiffies - wall_jiffies;
1273 wall_jiffies += ticks;
1274 update_wall_time();
1275 calc_load(ticks);
1276}
1277
1278/*
1279 * The 64-bit jiffies value is not atomic - you MUST NOT read it
1280 * without sampling the sequence number in xtime_lock.
1281 * jiffies is defined in the linker script...
1282 */
1283
1284void do_timer(struct pt_regs *regs)
1285{
1286 jiffies_64++;
1287 /* prevent loading jiffies before storing new jiffies_64 value. */
1288 barrier();
1289 update_times();
1290}
1291
1292#ifdef __ARCH_WANT_SYS_ALARM
1293
1294/*
1295 * For backwards compatibility? This can be done in libc so Alpha
1296 * and all newer ports shouldn't need it.
1297 */
1298asmlinkage unsigned long sys_alarm(unsigned int seconds)
1299{
1300 return alarm_setitimer(seconds);
1301}
1302
1303#endif
1304
1305#ifndef __alpha__
1306
1307/*
1308 * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this
1309 * should be moved into arch/i386 instead?
1310 */
1311
1312/**
1313 * sys_getpid - return the thread group id of the current process
1314 *
1315 * Note, despite the name, this returns the tgid not the pid. The tgid and
1316 * the pid are identical unless CLONE_THREAD was specified on clone() in
1317 * which case the tgid is the same in all threads of the same group.
1318 *
1319 * This is SMP safe as current->tgid does not change.
1320 */
1321asmlinkage long sys_getpid(void)
1322{
1323 return current->tgid;
1324}
1325
1326/*
1327 * Accessing ->real_parent is not SMP-safe, it could
1328 * change from under us. However, we can use a stale
1329 * value of ->real_parent under rcu_read_lock(), see
1330 * release_task()->call_rcu(delayed_put_task_struct).
1331 */
1332asmlinkage long sys_getppid(void)
1333{
1334 int pid;
1335
1336 rcu_read_lock();
1337 pid = rcu_dereference(current->real_parent)->tgid;
1338 rcu_read_unlock();
1339
1340 return pid;
1341}
1342
1343asmlinkage long sys_getuid(void)
1344{
1345 /* Only we change this so SMP safe */
1346 return current->uid;
1347}
1348
1349asmlinkage long sys_geteuid(void)
1350{
1351 /* Only we change this so SMP safe */
1352 return current->euid;
1353}
1354
1355asmlinkage long sys_getgid(void)
1356{
1357 /* Only we change this so SMP safe */
1358 return current->gid;
1359}
1360
1361asmlinkage long sys_getegid(void)
1362{
1363 /* Only we change this so SMP safe */
1364 return current->egid;
1365}
1366
1367#endif
1368
1369static void process_timeout(unsigned long __data)
1370{
1371 wake_up_process((struct task_struct *)__data);
1372}
1373
1374/**
1375 * schedule_timeout - sleep until timeout
1376 * @timeout: timeout value in jiffies
1377 *
1378 * Make the current task sleep until @timeout jiffies have
1379 * elapsed. The routine will return immediately unless
1380 * the current task state has been set (see set_current_state()).
1381 *
1382 * You can set the task state as follows -
1383 *
1384 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1385 * pass before the routine returns. The routine will return 0
1386 *
1387 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1388 * delivered to the current task. In this case the remaining time
1389 * in jiffies will be returned, or 0 if the timer expired in time
1390 *
1391 * The current task state is guaranteed to be TASK_RUNNING when this
1392 * routine returns.
1393 *
1394 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1395 * the CPU away without a bound on the timeout. In this case the return
1396 * value will be %MAX_SCHEDULE_TIMEOUT.
1397 *
1398 * In all cases the return value is guaranteed to be non-negative.
1399 */
1400fastcall signed long __sched schedule_timeout(signed long timeout)
1401{
1402 struct timer_list timer;
1403 unsigned long expire;
1404
1405 switch (timeout)
1406 {
1407 case MAX_SCHEDULE_TIMEOUT:
1408 /*
1409 * These two special cases are useful to be comfortable
1410 * in the caller. Nothing more. We could take
1411 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1412 * but I' d like to return a valid offset (>=0) to allow
1413 * the caller to do everything it want with the retval.
1414 */
1415 schedule();
1416 goto out;
1417 default:
1418 /*
1419 * Another bit of PARANOID. Note that the retval will be
1420 * 0 since no piece of kernel is supposed to do a check
1421 * for a negative retval of schedule_timeout() (since it
1422 * should never happens anyway). You just have the printk()
1423 * that will tell you if something is gone wrong and where.
1424 */
1425 if (timeout < 0)
1426 {
1427 printk(KERN_ERR "schedule_timeout: wrong timeout "
1428 "value %lx from %p\n", timeout,
1429 __builtin_return_address(0));
1430 current->state = TASK_RUNNING;
1431 goto out;
1432 }
1433 }
1434
1435 expire = timeout + jiffies;
1436
1437 setup_timer(&timer, process_timeout, (unsigned long)current);
1438 __mod_timer(&timer, expire);
1439 schedule();
1440 del_singleshot_timer_sync(&timer);
1441
1442 timeout = expire - jiffies;
1443
1444 out:
1445 return timeout < 0 ? 0 : timeout;
1446}
1447EXPORT_SYMBOL(schedule_timeout);
1448
1449/*
1450 * We can use __set_current_state() here because schedule_timeout() calls
1451 * schedule() unconditionally.
1452 */
1453signed long __sched schedule_timeout_interruptible(signed long timeout)
1454{
1455 __set_current_state(TASK_INTERRUPTIBLE);
1456 return schedule_timeout(timeout);
1457}
1458EXPORT_SYMBOL(schedule_timeout_interruptible);
1459
1460signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1461{
1462 __set_current_state(TASK_UNINTERRUPTIBLE);
1463 return schedule_timeout(timeout);
1464}
1465EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1466
1467/* Thread ID - the internal kernel "pid" */
1468asmlinkage long sys_gettid(void)
1469{
1470 return current->pid;
1471}
1472
1473/*
1474 * sys_sysinfo - fill in sysinfo struct
1475 */
1476asmlinkage long sys_sysinfo(struct sysinfo __user *info)
1477{
1478 struct sysinfo val;
1479 unsigned long mem_total, sav_total;
1480 unsigned int mem_unit, bitcount;
1481 unsigned long seq;
1482
1483 memset((char *)&val, 0, sizeof(struct sysinfo));
1484
1485 do {
1486 struct timespec tp;
1487 seq = read_seqbegin(&xtime_lock);
1488
1489 /*
1490 * This is annoying. The below is the same thing
1491 * posix_get_clock_monotonic() does, but it wants to
1492 * take the lock which we want to cover the loads stuff
1493 * too.
1494 */
1495
1496 getnstimeofday(&tp);
1497 tp.tv_sec += wall_to_monotonic.tv_sec;
1498 tp.tv_nsec += wall_to_monotonic.tv_nsec;
1499 if (tp.tv_nsec - NSEC_PER_SEC >= 0) {
1500 tp.tv_nsec = tp.tv_nsec - NSEC_PER_SEC;
1501 tp.tv_sec++;
1502 }
1503 val.uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
1504
1505 val.loads[0] = avenrun[0] << (SI_LOAD_SHIFT - FSHIFT);
1506 val.loads[1] = avenrun[1] << (SI_LOAD_SHIFT - FSHIFT);
1507 val.loads[2] = avenrun[2] << (SI_LOAD_SHIFT - FSHIFT);
1508
1509 val.procs = nr_threads;
1510 } while (read_seqretry(&xtime_lock, seq));
1511
1512 si_meminfo(&val);
1513 si_swapinfo(&val);
1514
1515 /*
1516 * If the sum of all the available memory (i.e. ram + swap)
1517 * is less than can be stored in a 32 bit unsigned long then
1518 * we can be binary compatible with 2.2.x kernels. If not,
1519 * well, in that case 2.2.x was broken anyways...
1520 *
1521 * -Erik Andersen <[email protected]>
1522 */
1523
1524 mem_total = val.totalram + val.totalswap;
1525 if (mem_total < val.totalram || mem_total < val.totalswap)
1526 goto out;
1527 bitcount = 0;
1528 mem_unit = val.mem_unit;
1529 while (mem_unit > 1) {
1530 bitcount++;
1531 mem_unit >>= 1;
1532 sav_total = mem_total;
1533 mem_total <<= 1;
1534 if (mem_total < sav_total)
1535 goto out;
1536 }
1537
1538 /*
1539 * If mem_total did not overflow, multiply all memory values by
1540 * val.mem_unit and set it to 1. This leaves things compatible
1541 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1542 * kernels...
1543 */
1544
1545 val.mem_unit = 1;
1546 val.totalram <<= bitcount;
1547 val.freeram <<= bitcount;
1548 val.sharedram <<= bitcount;
1549 val.bufferram <<= bitcount;
1550 val.totalswap <<= bitcount;
1551 val.freeswap <<= bitcount;
1552 val.totalhigh <<= bitcount;
1553 val.freehigh <<= bitcount;
1554
1555 out:
1556 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
1557 return -EFAULT;
1558
1559 return 0;
1560}
1561
1562/*
1563 * lockdep: we want to track each per-CPU base as a separate lock-class,
1564 * but timer-bases are kmalloc()-ed, so we need to attach separate
1565 * keys to them:
1566 */
1567static struct lock_class_key base_lock_keys[NR_CPUS];
1568
1569static int __devinit init_timers_cpu(int cpu)
1570{
1571 int j;
1572 tvec_base_t *base;
1573 static char __devinitdata tvec_base_done[NR_CPUS];
1574
1575 if (!tvec_base_done[cpu]) {
1576 static char boot_done;
1577
1578 if (boot_done) {
1579 /*
1580 * The APs use this path later in boot
1581 */
1582 base = kmalloc_node(sizeof(*base), GFP_KERNEL,
1583 cpu_to_node(cpu));
1584 if (!base)
1585 return -ENOMEM;
1586 memset(base, 0, sizeof(*base));
1587 per_cpu(tvec_bases, cpu) = base;
1588 } else {
1589 /*
1590 * This is for the boot CPU - we use compile-time
1591 * static initialisation because per-cpu memory isn't
1592 * ready yet and because the memory allocators are not
1593 * initialised either.
1594 */
1595 boot_done = 1;
1596 base = &boot_tvec_bases;
1597 }
1598 tvec_base_done[cpu] = 1;
1599 } else {
1600 base = per_cpu(tvec_bases, cpu);
1601 }
1602
1603 spin_lock_init(&base->lock);
1604 lockdep_set_class(&base->lock, base_lock_keys + cpu);
1605
1606 for (j = 0; j < TVN_SIZE; j++) {
1607 INIT_LIST_HEAD(base->tv5.vec + j);
1608 INIT_LIST_HEAD(base->tv4.vec + j);
1609 INIT_LIST_HEAD(base->tv3.vec + j);
1610 INIT_LIST_HEAD(base->tv2.vec + j);
1611 }
1612 for (j = 0; j < TVR_SIZE; j++)
1613 INIT_LIST_HEAD(base->tv1.vec + j);
1614
1615 base->timer_jiffies = jiffies;
1616 return 0;
1617}
1618
1619#ifdef CONFIG_HOTPLUG_CPU
1620static void migrate_timer_list(tvec_base_t *new_base, struct list_head *head)
1621{
1622 struct timer_list *timer;
1623
1624 while (!list_empty(head)) {
1625 timer = list_entry(head->next, struct timer_list, entry);
1626 detach_timer(timer, 0);
1627 timer->base = new_base;
1628 internal_add_timer(new_base, timer);
1629 }
1630}
1631
1632static void __devinit migrate_timers(int cpu)
1633{
1634 tvec_base_t *old_base;
1635 tvec_base_t *new_base;
1636 int i;
1637
1638 BUG_ON(cpu_online(cpu));
1639 old_base = per_cpu(tvec_bases, cpu);
1640 new_base = get_cpu_var(tvec_bases);
1641
1642 local_irq_disable();
1643 spin_lock(&new_base->lock);
1644 spin_lock(&old_base->lock);
1645
1646 BUG_ON(old_base->running_timer);
1647
1648 for (i = 0; i < TVR_SIZE; i++)
1649 migrate_timer_list(new_base, old_base->tv1.vec + i);
1650 for (i = 0; i < TVN_SIZE; i++) {
1651 migrate_timer_list(new_base, old_base->tv2.vec + i);
1652 migrate_timer_list(new_base, old_base->tv3.vec + i);
1653 migrate_timer_list(new_base, old_base->tv4.vec + i);
1654 migrate_timer_list(new_base, old_base->tv5.vec + i);
1655 }
1656
1657 spin_unlock(&old_base->lock);
1658 spin_unlock(&new_base->lock);
1659 local_irq_enable();
1660 put_cpu_var(tvec_bases);
1661}
1662#endif /* CONFIG_HOTPLUG_CPU */
1663
1664static int __cpuinit timer_cpu_notify(struct notifier_block *self,
1665 unsigned long action, void *hcpu)
1666{
1667 long cpu = (long)hcpu;
1668 switch(action) {
1669 case CPU_UP_PREPARE:
1670 if (init_timers_cpu(cpu) < 0)
1671 return NOTIFY_BAD;
1672 break;
1673#ifdef CONFIG_HOTPLUG_CPU
1674 case CPU_DEAD:
1675 migrate_timers(cpu);
1676 break;
1677#endif
1678 default:
1679 break;
1680 }
1681 return NOTIFY_OK;
1682}
1683
1684static struct notifier_block __cpuinitdata timers_nb = {
1685 .notifier_call = timer_cpu_notify,
1686};
1687
1688
1689void __init init_timers(void)
1690{
1691 timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE,
1692 (void *)(long)smp_processor_id());
1693 register_cpu_notifier(&timers_nb);
1694 open_softirq(TIMER_SOFTIRQ, run_timer_softirq, NULL);
1695}
1696
1697#ifdef CONFIG_TIME_INTERPOLATION
1698
1699struct time_interpolator *time_interpolator __read_mostly;
1700static struct time_interpolator *time_interpolator_list __read_mostly;
1701static DEFINE_SPINLOCK(time_interpolator_lock);
1702
1703static inline u64 time_interpolator_get_cycles(unsigned int src)
1704{
1705 unsigned long (*x)(void);
1706
1707 switch (src)
1708 {
1709 case TIME_SOURCE_FUNCTION:
1710 x = time_interpolator->addr;
1711 return x();
1712
1713 case TIME_SOURCE_MMIO64 :
1714 return readq_relaxed((void __iomem *)time_interpolator->addr);
1715
1716 case TIME_SOURCE_MMIO32 :
1717 return readl_relaxed((void __iomem *)time_interpolator->addr);
1718
1719 default: return get_cycles();
1720 }
1721}
1722
1723static inline u64 time_interpolator_get_counter(int writelock)
1724{
1725 unsigned int src = time_interpolator->source;
1726
1727 if (time_interpolator->jitter)
1728 {
1729 u64 lcycle;
1730 u64 now;
1731
1732 do {
1733 lcycle = time_interpolator->last_cycle;
1734 now = time_interpolator_get_cycles(src);
1735 if (lcycle && time_after(lcycle, now))
1736 return lcycle;
1737
1738 /* When holding the xtime write lock, there's no need
1739 * to add the overhead of the cmpxchg. Readers are
1740 * force to retry until the write lock is released.
1741 */
1742 if (writelock) {
1743 time_interpolator->last_cycle = now;
1744 return now;
1745 }
1746 /* Keep track of the last timer value returned. The use of cmpxchg here
1747 * will cause contention in an SMP environment.
1748 */
1749 } while (unlikely(cmpxchg(&time_interpolator->last_cycle, lcycle, now) != lcycle));
1750 return now;
1751 }
1752 else
1753 return time_interpolator_get_cycles(src);
1754}
1755
1756void time_interpolator_reset(void)
1757{
1758 time_interpolator->offset = 0;
1759 time_interpolator->last_counter = time_interpolator_get_counter(1);
1760}
1761
1762#define GET_TI_NSECS(count,i) (((((count) - i->last_counter) & (i)->mask) * (i)->nsec_per_cyc) >> (i)->shift)
1763
1764unsigned long time_interpolator_get_offset(void)
1765{
1766 /* If we do not have a time interpolator set up then just return zero */
1767 if (!time_interpolator)
1768 return 0;
1769
1770 return time_interpolator->offset +
1771 GET_TI_NSECS(time_interpolator_get_counter(0), time_interpolator);
1772}
1773
1774#define INTERPOLATOR_ADJUST 65536
1775#define INTERPOLATOR_MAX_SKIP 10*INTERPOLATOR_ADJUST
1776
1777static void time_interpolator_update(long delta_nsec)
1778{
1779 u64 counter;
1780 unsigned long offset;
1781
1782 /* If there is no time interpolator set up then do nothing */
1783 if (!time_interpolator)
1784 return;
1785
1786 /*
1787 * The interpolator compensates for late ticks by accumulating the late
1788 * time in time_interpolator->offset. A tick earlier than expected will
1789 * lead to a reset of the offset and a corresponding jump of the clock
1790 * forward. Again this only works if the interpolator clock is running
1791 * slightly slower than the regular clock and the tuning logic insures
1792 * that.
1793 */
1794
1795 counter = time_interpolator_get_counter(1);
1796 offset = time_interpolator->offset +
1797 GET_TI_NSECS(counter, time_interpolator);
1798
1799 if (delta_nsec < 0 || (unsigned long) delta_nsec < offset)
1800 time_interpolator->offset = offset - delta_nsec;
1801 else {
1802 time_interpolator->skips++;
1803 time_interpolator->ns_skipped += delta_nsec - offset;
1804 time_interpolator->offset = 0;
1805 }
1806 time_interpolator->last_counter = counter;
1807
1808 /* Tuning logic for time interpolator invoked every minute or so.
1809 * Decrease interpolator clock speed if no skips occurred and an offset is carried.
1810 * Increase interpolator clock speed if we skip too much time.
1811 */
1812 if (jiffies % INTERPOLATOR_ADJUST == 0)
1813 {
1814 if (time_interpolator->skips == 0 && time_interpolator->offset > tick_nsec)
1815 time_interpolator->nsec_per_cyc--;
1816 if (time_interpolator->ns_skipped > INTERPOLATOR_MAX_SKIP && time_interpolator->offset == 0)
1817 time_interpolator->nsec_per_cyc++;
1818 time_interpolator->skips = 0;
1819 time_interpolator->ns_skipped = 0;
1820 }
1821}
1822
1823static inline int
1824is_better_time_interpolator(struct time_interpolator *new)
1825{
1826 if (!time_interpolator)
1827 return 1;
1828 return new->frequency > 2*time_interpolator->frequency ||
1829 (unsigned long)new->drift < (unsigned long)time_interpolator->drift;
1830}
1831
1832void
1833register_time_interpolator(struct time_interpolator *ti)
1834{
1835 unsigned long flags;
1836
1837 /* Sanity check */
1838 BUG_ON(ti->frequency == 0 || ti->mask == 0);
1839
1840 ti->nsec_per_cyc = ((u64)NSEC_PER_SEC << ti->shift) / ti->frequency;
1841 spin_lock(&time_interpolator_lock);
1842 write_seqlock_irqsave(&xtime_lock, flags);
1843 if (is_better_time_interpolator(ti)) {
1844 time_interpolator = ti;
1845 time_interpolator_reset();
1846 }
1847 write_sequnlock_irqrestore(&xtime_lock, flags);
1848
1849 ti->next = time_interpolator_list;
1850 time_interpolator_list = ti;
1851 spin_unlock(&time_interpolator_lock);
1852}
1853
1854void
1855unregister_time_interpolator(struct time_interpolator *ti)
1856{
1857 struct time_interpolator *curr, **prev;
1858 unsigned long flags;
1859
1860 spin_lock(&time_interpolator_lock);
1861 prev = &time_interpolator_list;
1862 for (curr = *prev; curr; curr = curr->next) {
1863 if (curr == ti) {
1864 *prev = curr->next;
1865 break;
1866 }
1867 prev = &curr->next;
1868 }
1869
1870 write_seqlock_irqsave(&xtime_lock, flags);
1871 if (ti == time_interpolator) {
1872 /* we lost the best time-interpolator: */
1873 time_interpolator = NULL;
1874 /* find the next-best interpolator */
1875 for (curr = time_interpolator_list; curr; curr = curr->next)
1876 if (is_better_time_interpolator(curr))
1877 time_interpolator = curr;
1878 time_interpolator_reset();
1879 }
1880 write_sequnlock_irqrestore(&xtime_lock, flags);
1881 spin_unlock(&time_interpolator_lock);
1882}
1883#endif /* CONFIG_TIME_INTERPOLATION */
1884
1885/**
1886 * msleep - sleep safely even with waitqueue interruptions
1887 * @msecs: Time in milliseconds to sleep for
1888 */
1889void msleep(unsigned int msecs)
1890{
1891 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1892
1893 while (timeout)
1894 timeout = schedule_timeout_uninterruptible(timeout);
1895}
1896
1897EXPORT_SYMBOL(msleep);
1898
1899/**
1900 * msleep_interruptible - sleep waiting for signals
1901 * @msecs: Time in milliseconds to sleep for
1902 */
1903unsigned long msleep_interruptible(unsigned int msecs)
1904{
1905 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1906
1907 while (timeout && !signal_pending(current))
1908 timeout = schedule_timeout_interruptible(timeout);
1909 return jiffies_to_msecs(timeout);
1910}
1911
1912EXPORT_SYMBOL(msleep_interruptible);