projects/linux·blob v2.6.21-rc2 · refs

1926 lines · 49.8 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#include <linux/tick.h>
38#include <linux/kallsyms.h>
39
40#include <asm/uaccess.h>
41#include <asm/unistd.h>
42#include <asm/div64.h>
43#include <asm/timex.h>
44#include <asm/io.h>
45
46u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
47
48EXPORT_SYMBOL(jiffies_64);
49
50/*
51 * per-CPU timer vector definitions:
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
60typedef struct tvec_s {
61 struct list_head vec[TVN_SIZE];
62} tvec_t;
63
64typedef struct tvec_root_s {
65 struct list_head vec[TVR_SIZE];
66} tvec_root_t;
67
68struct tvec_t_base_s {
69 spinlock_t lock;
70 struct timer_list *running_timer;
71 unsigned long timer_jiffies;
72 tvec_root_t tv1;
73 tvec_t tv2;
74 tvec_t tv3;
75 tvec_t tv4;
76 tvec_t tv5;
77} ____cacheline_aligned_in_smp;
78
79typedef struct tvec_t_base_s tvec_base_t;
80
81tvec_base_t boot_tvec_bases;
82EXPORT_SYMBOL(boot_tvec_bases);
83static DEFINE_PER_CPU(tvec_base_t *, tvec_bases) = &boot_tvec_bases;
84
85/**
86 * __round_jiffies - function to round jiffies to a full second
87 * @j: the time in (absolute) jiffies that should be rounded
88 * @cpu: the processor number on which the timeout will happen
89 *
90 * __round_jiffies() rounds an absolute time in the future (in jiffies)
91 * up or down to (approximately) full seconds. This is useful for timers
92 * for which the exact time they fire does not matter too much, as long as
93 * they fire approximately every X seconds.
94 *
95 * By rounding these timers to whole seconds, all such timers will fire
96 * at the same time, rather than at various times spread out. The goal
97 * of this is to have the CPU wake up less, which saves power.
98 *
99 * The exact rounding is skewed for each processor to avoid all
100 * processors firing at the exact same time, which could lead
101 * to lock contention or spurious cache line bouncing.
102 *
103 * The return value is the rounded version of the @j parameter.
104 */
105unsigned long __round_jiffies(unsigned long j, int cpu)
106{
107 int rem;
108 unsigned long original = j;
109
110 /*
111 * We don't want all cpus firing their timers at once hitting the
112 * same lock or cachelines, so we skew each extra cpu with an extra
113 * 3 jiffies. This 3 jiffies came originally from the mm/ code which
114 * already did this.
115 * The skew is done by adding 3*cpunr, then round, then subtract this
116 * extra offset again.
117 */
118 j += cpu * 3;
119
120 rem = j % HZ;
121
122 /*
123 * If the target jiffie is just after a whole second (which can happen
124 * due to delays of the timer irq, long irq off times etc etc) then
125 * we should round down to the whole second, not up. Use 1/4th second
126 * as cutoff for this rounding as an extreme upper bound for this.
127 */
128 if (rem < HZ/4) /* round down */
129 j = j - rem;
130 else /* round up */
131 j = j - rem + HZ;
132
133 /* now that we have rounded, subtract the extra skew again */
134 j -= cpu * 3;
135
136 if (j <= jiffies) /* rounding ate our timeout entirely; */
137 return original;
138 return j;
139}
140EXPORT_SYMBOL_GPL(__round_jiffies);
141
142/**
143 * __round_jiffies_relative - function to round jiffies to a full second
144 * @j: the time in (relative) jiffies that should be rounded
145 * @cpu: the processor number on which the timeout will happen
146 *
147 * __round_jiffies_relative() rounds a time delta in the future (in jiffies)
148 * up or down to (approximately) full seconds. This is useful for timers
149 * for which the exact time they fire does not matter too much, as long as
150 * they fire approximately every X seconds.
151 *
152 * By rounding these timers to whole seconds, all such timers will fire
153 * at the same time, rather than at various times spread out. The goal
154 * of this is to have the CPU wake up less, which saves power.
155 *
156 * The exact rounding is skewed for each processor to avoid all
157 * processors firing at the exact same time, which could lead
158 * to lock contention or spurious cache line bouncing.
159 *
160 * The return value is the rounded version of the @j parameter.
161 */
162unsigned long __round_jiffies_relative(unsigned long j, int cpu)
163{
164 /*
165 * In theory the following code can skip a jiffy in case jiffies
166 * increments right between the addition and the later subtraction.
167 * However since the entire point of this function is to use approximate
168 * timeouts, it's entirely ok to not handle that.
169 */
170 return __round_jiffies(j + jiffies, cpu) - jiffies;
171}
172EXPORT_SYMBOL_GPL(__round_jiffies_relative);
173
174/**
175 * round_jiffies - function to round jiffies to a full second
176 * @j: the time in (absolute) jiffies that should be rounded
177 *
178 * round_jiffies() rounds an absolute time in the future (in jiffies)
179 * up or down to (approximately) full seconds. This is useful for timers
180 * for which the exact time they fire does not matter too much, as long as
181 * they fire approximately every X seconds.
182 *
183 * By rounding these timers to whole seconds, all such timers will fire
184 * at the same time, rather than at various times spread out. The goal
185 * of this is to have the CPU wake up less, which saves power.
186 *
187 * The return value is the rounded version of the @j parameter.
188 */
189unsigned long round_jiffies(unsigned long j)
190{
191 return __round_jiffies(j, raw_smp_processor_id());
192}
193EXPORT_SYMBOL_GPL(round_jiffies);
194
195/**
196 * round_jiffies_relative - function to round jiffies to a full second
197 * @j: the time in (relative) jiffies that should be rounded
198 *
199 * round_jiffies_relative() rounds a time delta in the future (in jiffies)
200 * up or down to (approximately) full seconds. This is useful for timers
201 * for which the exact time they fire does not matter too much, as long as
202 * they fire approximately every X seconds.
203 *
204 * By rounding these timers to whole seconds, all such timers will fire
205 * at the same time, rather than at various times spread out. The goal
206 * of this is to have the CPU wake up less, which saves power.
207 *
208 * The return value is the rounded version of the @j parameter.
209 */
210unsigned long round_jiffies_relative(unsigned long j)
211{
212 return __round_jiffies_relative(j, raw_smp_processor_id());
213}
214EXPORT_SYMBOL_GPL(round_jiffies_relative);
215
216
217static inline void set_running_timer(tvec_base_t *base,
218 struct timer_list *timer)
219{
220#ifdef CONFIG_SMP
221 base->running_timer = timer;
222#endif
223}
224
225static void internal_add_timer(tvec_base_t *base, struct timer_list *timer)
226{
227 unsigned long expires = timer->expires;
228 unsigned long idx = expires - base->timer_jiffies;
229 struct list_head *vec;
230
231 if (idx < TVR_SIZE) {
232 int i = expires & TVR_MASK;
233 vec = base->tv1.vec + i;
234 } else if (idx < 1 << (TVR_BITS + TVN_BITS)) {
235 int i = (expires >> TVR_BITS) & TVN_MASK;
236 vec = base->tv2.vec + i;
237 } else if (idx < 1 << (TVR_BITS + 2 * TVN_BITS)) {
238 int i = (expires >> (TVR_BITS + TVN_BITS)) & TVN_MASK;
239 vec = base->tv3.vec + i;
240 } else if (idx < 1 << (TVR_BITS + 3 * TVN_BITS)) {
241 int i = (expires >> (TVR_BITS + 2 * TVN_BITS)) & TVN_MASK;
242 vec = base->tv4.vec + i;
243 } else if ((signed long) idx < 0) {
244 /*
245 * Can happen if you add a timer with expires == jiffies,
246 * or you set a timer to go off in the past
247 */
248 vec = base->tv1.vec + (base->timer_jiffies & TVR_MASK);
249 } else {
250 int i;
251 /* If the timeout is larger than 0xffffffff on 64-bit
252 * architectures then we use the maximum timeout:
253 */
254 if (idx > 0xffffffffUL) {
255 idx = 0xffffffffUL;
256 expires = idx + base->timer_jiffies;
257 }
258 i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK;
259 vec = base->tv5.vec + i;
260 }
261 /*
262 * Timers are FIFO:
263 */
264 list_add_tail(&timer->entry, vec);
265}
266
267#ifdef CONFIG_TIMER_STATS
268void __timer_stats_timer_set_start_info(struct timer_list *timer, void *addr)
269{
270 if (timer->start_site)
271 return;
272
273 timer->start_site = addr;
274 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
275 timer->start_pid = current->pid;
276}
277#endif
278
279/**
280 * init_timer - initialize a timer.
281 * @timer: the timer to be initialized
282 *
283 * init_timer() must be done to a timer prior calling *any* of the
284 * other timer functions.
285 */
286void fastcall init_timer(struct timer_list *timer)
287{
288 timer->entry.next = NULL;
289 timer->base = __raw_get_cpu_var(tvec_bases);
290#ifdef CONFIG_TIMER_STATS
291 timer->start_site = NULL;
292 timer->start_pid = -1;
293 memset(timer->start_comm, 0, TASK_COMM_LEN);
294#endif
295}
296EXPORT_SYMBOL(init_timer);
297
298static inline void detach_timer(struct timer_list *timer,
299 int clear_pending)
300{
301 struct list_head *entry = &timer->entry;
302
303 __list_del(entry->prev, entry->next);
304 if (clear_pending)
305 entry->next = NULL;
306 entry->prev = LIST_POISON2;
307}
308
309/*
310 * We are using hashed locking: holding per_cpu(tvec_bases).lock
311 * means that all timers which are tied to this base via timer->base are
312 * locked, and the base itself is locked too.
313 *
314 * So __run_timers/migrate_timers can safely modify all timers which could
315 * be found on ->tvX lists.
316 *
317 * When the timer's base is locked, and the timer removed from list, it is
318 * possible to set timer->base = NULL and drop the lock: the timer remains
319 * locked.
320 */
321static tvec_base_t *lock_timer_base(struct timer_list *timer,
322 unsigned long *flags)
323 __acquires(timer->base->lock)
324{
325 tvec_base_t *base;
326
327 for (;;) {
328 base = timer->base;
329 if (likely(base != NULL)) {
330 spin_lock_irqsave(&base->lock, *flags);
331 if (likely(base == timer->base))
332 return base;
333 /* The timer has migrated to another CPU */
334 spin_unlock_irqrestore(&base->lock, *flags);
335 }
336 cpu_relax();
337 }
338}
339
340int __mod_timer(struct timer_list *timer, unsigned long expires)
341{
342 tvec_base_t *base, *new_base;
343 unsigned long flags;
344 int ret = 0;
345
346 timer_stats_timer_set_start_info(timer);
347 BUG_ON(!timer->function);
348
349 base = lock_timer_base(timer, &flags);
350
351 if (timer_pending(timer)) {
352 detach_timer(timer, 0);
353 ret = 1;
354 }
355
356 new_base = __get_cpu_var(tvec_bases);
357
358 if (base != new_base) {
359 /*
360 * We are trying to schedule the timer on the local CPU.
361 * However we can't change timer's base while it is running,
362 * otherwise del_timer_sync() can't detect that the timer's
363 * handler yet has not finished. This also guarantees that
364 * the timer is serialized wrt itself.
365 */
366 if (likely(base->running_timer != timer)) {
367 /* See the comment in lock_timer_base() */
368 timer->base = NULL;
369 spin_unlock(&base->lock);
370 base = new_base;
371 spin_lock(&base->lock);
372 timer->base = base;
373 }
374 }
375
376 timer->expires = expires;
377 internal_add_timer(base, timer);
378 spin_unlock_irqrestore(&base->lock, flags);
379
380 return ret;
381}
382
383EXPORT_SYMBOL(__mod_timer);
384
385/**
386 * add_timer_on - start a timer on a particular CPU
387 * @timer: the timer to be added
388 * @cpu: the CPU to start it on
389 *
390 * This is not very scalable on SMP. Double adds are not possible.
391 */
392void add_timer_on(struct timer_list *timer, int cpu)
393{
394 tvec_base_t *base = per_cpu(tvec_bases, cpu);
395 unsigned long flags;
396
397 timer_stats_timer_set_start_info(timer);
398 BUG_ON(timer_pending(timer) || !timer->function);
399 spin_lock_irqsave(&base->lock, flags);
400 timer->base = base;
401 internal_add_timer(base, timer);
402 spin_unlock_irqrestore(&base->lock, flags);
403}
404
405
406/**
407 * mod_timer - modify a timer's timeout
408 * @timer: the timer to be modified
409 * @expires: new timeout in jiffies
410 *
411 * mod_timer() is a more efficient way to update the expire field of an
412 * active timer (if the timer is inactive it will be activated)
413 *
414 * mod_timer(timer, expires) is equivalent to:
415 *
416 * del_timer(timer); timer->expires = expires; add_timer(timer);
417 *
418 * Note that if there are multiple unserialized concurrent users of the
419 * same timer, then mod_timer() is the only safe way to modify the timeout,
420 * since add_timer() cannot modify an already running timer.
421 *
422 * The function returns whether it has modified a pending timer or not.
423 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
424 * active timer returns 1.)
425 */
426int mod_timer(struct timer_list *timer, unsigned long expires)
427{
428 BUG_ON(!timer->function);
429
430 timer_stats_timer_set_start_info(timer);
431 /*
432 * This is a common optimization triggered by the
433 * networking code - if the timer is re-modified
434 * to be the same thing then just return:
435 */
436 if (timer->expires == expires && timer_pending(timer))
437 return 1;
438
439 return __mod_timer(timer, expires);
440}
441
442EXPORT_SYMBOL(mod_timer);
443
444/**
445 * del_timer - deactive a timer.
446 * @timer: the timer to be deactivated
447 *
448 * del_timer() deactivates a timer - this works on both active and inactive
449 * timers.
450 *
451 * The function returns whether it has deactivated a pending timer or not.
452 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
453 * active timer returns 1.)
454 */
455int del_timer(struct timer_list *timer)
456{
457 tvec_base_t *base;
458 unsigned long flags;
459 int ret = 0;
460
461 timer_stats_timer_clear_start_info(timer);
462 if (timer_pending(timer)) {
463 base = lock_timer_base(timer, &flags);
464 if (timer_pending(timer)) {
465 detach_timer(timer, 1);
466 ret = 1;
467 }
468 spin_unlock_irqrestore(&base->lock, flags);
469 }
470
471 return ret;
472}
473
474EXPORT_SYMBOL(del_timer);
475
476#ifdef CONFIG_SMP
477/**
478 * try_to_del_timer_sync - Try to deactivate a timer
479 * @timer: timer do del
480 *
481 * This function tries to deactivate a timer. Upon successful (ret >= 0)
482 * exit the timer is not queued and the handler is not running on any CPU.
483 *
484 * It must not be called from interrupt contexts.
485 */
486int try_to_del_timer_sync(struct timer_list *timer)
487{
488 tvec_base_t *base;
489 unsigned long flags;
490 int ret = -1;
491
492 base = lock_timer_base(timer, &flags);
493
494 if (base->running_timer == timer)
495 goto out;
496
497 ret = 0;
498 if (timer_pending(timer)) {
499 detach_timer(timer, 1);
500 ret = 1;
501 }
502out:
503 spin_unlock_irqrestore(&base->lock, flags);
504
505 return ret;
506}
507
508/**
509 * del_timer_sync - deactivate a timer and wait for the handler to finish.
510 * @timer: the timer to be deactivated
511 *
512 * This function only differs from del_timer() on SMP: besides deactivating
513 * the timer it also makes sure the handler has finished executing on other
514 * CPUs.
515 *
516 * Synchronization rules: Callers must prevent restarting of the timer,
517 * otherwise this function is meaningless. It must not be called from
518 * interrupt contexts. The caller must not hold locks which would prevent
519 * completion of the timer's handler. The timer's handler must not call
520 * add_timer_on(). Upon exit the timer is not queued and the handler is
521 * not running on any CPU.
522 *
523 * The function returns whether it has deactivated a pending timer or not.
524 */
525int del_timer_sync(struct timer_list *timer)
526{
527 for (;;) {
528 int ret = try_to_del_timer_sync(timer);
529 if (ret >= 0)
530 return ret;
531 cpu_relax();
532 }
533}
534
535EXPORT_SYMBOL(del_timer_sync);
536#endif
537
538static int cascade(tvec_base_t *base, tvec_t *tv, int index)
539{
540 /* cascade all the timers from tv up one level */
541 struct timer_list *timer, *tmp;
542 struct list_head tv_list;
543
544 list_replace_init(tv->vec + index, &tv_list);
545
546 /*
547 * We are removing _all_ timers from the list, so we
548 * don't have to detach them individually.
549 */
550 list_for_each_entry_safe(timer, tmp, &tv_list, entry) {
551 BUG_ON(timer->base != base);
552 internal_add_timer(base, timer);
553 }
554
555 return index;
556}
557
558#define INDEX(N) ((base->timer_jiffies >> (TVR_BITS + (N) * TVN_BITS)) & TVN_MASK)
559
560/**
561 * __run_timers - run all expired timers (if any) on this CPU.
562 * @base: the timer vector to be processed.
563 *
564 * This function cascades all vectors and executes all expired timer
565 * vectors.
566 */
567static inline void __run_timers(tvec_base_t *base)
568{
569 struct timer_list *timer;
570
571 spin_lock_irq(&base->lock);
572 while (time_after_eq(jiffies, base->timer_jiffies)) {
573 struct list_head work_list;
574 struct list_head *head = &work_list;
575 int index = base->timer_jiffies & TVR_MASK;
576
577 /*
578 * Cascade timers:
579 */
580 if (!index &&
581 (!cascade(base, &base->tv2, INDEX(0))) &&
582 (!cascade(base, &base->tv3, INDEX(1))) &&
583 !cascade(base, &base->tv4, INDEX(2)))
584 cascade(base, &base->tv5, INDEX(3));
585 ++base->timer_jiffies;
586 list_replace_init(base->tv1.vec + index, &work_list);
587 while (!list_empty(head)) {
588 void (*fn)(unsigned long);
589 unsigned long data;
590
591 timer = list_entry(head->next,struct timer_list,entry);
592 fn = timer->function;
593 data = timer->data;
594
595 timer_stats_account_timer(timer);
596
597 set_running_timer(base, timer);
598 detach_timer(timer, 1);
599 spin_unlock_irq(&base->lock);
600 {
601 int preempt_count = preempt_count();
602 fn(data);
603 if (preempt_count != preempt_count()) {
604 printk(KERN_WARNING "huh, entered %p "
605 "with preempt_count %08x, exited"
606 " with %08x?\n",
607 fn, preempt_count,
608 preempt_count());
609 BUG();
610 }
611 }
612 spin_lock_irq(&base->lock);
613 }
614 }
615 set_running_timer(base, NULL);
616 spin_unlock_irq(&base->lock);
617}
618
619#if defined(CONFIG_NO_IDLE_HZ) || defined(CONFIG_NO_HZ)
620/*
621 * Find out when the next timer event is due to happen. This
622 * is used on S/390 to stop all activity when a cpus is idle.
623 * This functions needs to be called disabled.
624 */
625static unsigned long __next_timer_interrupt(tvec_base_t *base)
626{
627 unsigned long timer_jiffies = base->timer_jiffies;
628 unsigned long expires = timer_jiffies + (LONG_MAX >> 1);
629 int index, slot, array, found = 0;
630 struct timer_list *nte;
631 tvec_t *varray[4];
632
633 /* Look for timer events in tv1. */
634 index = slot = timer_jiffies & TVR_MASK;
635 do {
636 list_for_each_entry(nte, base->tv1.vec + slot, entry) {
637 found = 1;
638 expires = nte->expires;
639 /* Look at the cascade bucket(s)? */
640 if (!index || slot < index)
641 goto cascade;
642 return expires;
643 }
644 slot = (slot + 1) & TVR_MASK;
645 } while (slot != index);
646
647cascade:
648 /* Calculate the next cascade event */
649 if (index)
650 timer_jiffies += TVR_SIZE - index;
651 timer_jiffies >>= TVR_BITS;
652
653 /* Check tv2-tv5. */
654 varray[0] = &base->tv2;
655 varray[1] = &base->tv3;
656 varray[2] = &base->tv4;
657 varray[3] = &base->tv5;
658
659 for (array = 0; array < 4; array++) {
660 tvec_t *varp = varray[array];
661
662 index = slot = timer_jiffies & TVN_MASK;
663 do {
664 list_for_each_entry(nte, varp->vec + slot, entry) {
665 found = 1;
666 if (time_before(nte->expires, expires))
667 expires = nte->expires;
668 }
669 /*
670 * Do we still search for the first timer or are
671 * we looking up the cascade buckets ?
672 */
673 if (found) {
674 /* Look at the cascade bucket(s)? */
675 if (!index || slot < index)
676 break;
677 return expires;
678 }
679 slot = (slot + 1) & TVN_MASK;
680 } while (slot != index);
681
682 if (index)
683 timer_jiffies += TVN_SIZE - index;
684 timer_jiffies >>= TVN_BITS;
685 }
686 return expires;
687}
688
689/*
690 * Check, if the next hrtimer event is before the next timer wheel
691 * event:
692 */
693static unsigned long cmp_next_hrtimer_event(unsigned long now,
694 unsigned long expires)
695{
696 ktime_t hr_delta = hrtimer_get_next_event();
697 struct timespec tsdelta;
698
699 if (hr_delta.tv64 == KTIME_MAX)
700 return expires;
701
702 if (hr_delta.tv64 <= TICK_NSEC)
703 return now;
704
705 tsdelta = ktime_to_timespec(hr_delta);
706 now += timespec_to_jiffies(&tsdelta);
707 if (time_before(now, expires))
708 return now;
709 return expires;
710}
711
712/**
713 * next_timer_interrupt - return the jiffy of the next pending timer
714 */
715unsigned long get_next_timer_interrupt(unsigned long now)
716{
717 tvec_base_t *base = __get_cpu_var(tvec_bases);
718 unsigned long expires;
719
720 spin_lock(&base->lock);
721 expires = __next_timer_interrupt(base);
722 spin_unlock(&base->lock);
723
724 if (time_before_eq(expires, now))
725 return now;
726
727 return cmp_next_hrtimer_event(now, expires);
728}
729
730#ifdef CONFIG_NO_IDLE_HZ
731unsigned long next_timer_interrupt(void)
732{
733 return get_next_timer_interrupt(jiffies);
734}
735#endif
736
737#endif
738
739/******************************************************************/
740
741/*
742 * The current time
743 * wall_to_monotonic is what we need to add to xtime (or xtime corrected
744 * for sub jiffie times) to get to monotonic time. Monotonic is pegged
745 * at zero at system boot time, so wall_to_monotonic will be negative,
746 * however, we will ALWAYS keep the tv_nsec part positive so we can use
747 * the usual normalization.
748 */
749struct timespec xtime __attribute__ ((aligned (16)));
750struct timespec wall_to_monotonic __attribute__ ((aligned (16)));
751
752EXPORT_SYMBOL(xtime);
753
754
755/* XXX - all of this timekeeping code should be later moved to time.c */
756#include <linux/clocksource.h>
757static struct clocksource *clock; /* pointer to current clocksource */
758
759#ifdef CONFIG_GENERIC_TIME
760/**
761 * __get_nsec_offset - Returns nanoseconds since last call to periodic_hook
762 *
763 * private function, must hold xtime_lock lock when being
764 * called. Returns the number of nanoseconds since the
765 * last call to update_wall_time() (adjusted by NTP scaling)
766 */
767static inline s64 __get_nsec_offset(void)
768{
769 cycle_t cycle_now, cycle_delta;
770 s64 ns_offset;
771
772 /* read clocksource: */
773 cycle_now = clocksource_read(clock);
774
775 /* calculate the delta since the last update_wall_time: */
776 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
777
778 /* convert to nanoseconds: */
779 ns_offset = cyc2ns(clock, cycle_delta);
780
781 return ns_offset;
782}
783
784/**
785 * __get_realtime_clock_ts - Returns the time of day in a timespec
786 * @ts: pointer to the timespec to be set
787 *
788 * Returns the time of day in a timespec. Used by
789 * do_gettimeofday() and get_realtime_clock_ts().
790 */
791static inline void __get_realtime_clock_ts(struct timespec *ts)
792{
793 unsigned long seq;
794 s64 nsecs;
795
796 do {
797 seq = read_seqbegin(&xtime_lock);
798
799 *ts = xtime;
800 nsecs = __get_nsec_offset();
801
802 } while (read_seqretry(&xtime_lock, seq));
803
804 timespec_add_ns(ts, nsecs);
805}
806
807/**
808 * getnstimeofday - Returns the time of day in a timespec
809 * @ts: pointer to the timespec to be set
810 *
811 * Returns the time of day in a timespec.
812 */
813void getnstimeofday(struct timespec *ts)
814{
815 __get_realtime_clock_ts(ts);
816}
817
818EXPORT_SYMBOL(getnstimeofday);
819
820/**
821 * do_gettimeofday - Returns the time of day in a timeval
822 * @tv: pointer to the timeval to be set
823 *
824 * NOTE: Users should be converted to using get_realtime_clock_ts()
825 */
826void do_gettimeofday(struct timeval *tv)
827{
828 struct timespec now;
829
830 __get_realtime_clock_ts(&now);
831 tv->tv_sec = now.tv_sec;
832 tv->tv_usec = now.tv_nsec/1000;
833}
834
835EXPORT_SYMBOL(do_gettimeofday);
836/**
837 * do_settimeofday - Sets the time of day
838 * @tv: pointer to the timespec variable containing the new time
839 *
840 * Sets the time of day to the new time and update NTP and notify hrtimers
841 */
842int do_settimeofday(struct timespec *tv)
843{
844 unsigned long flags;
845 time_t wtm_sec, sec = tv->tv_sec;
846 long wtm_nsec, nsec = tv->tv_nsec;
847
848 if ((unsigned long)tv->tv_nsec >= NSEC_PER_SEC)
849 return -EINVAL;
850
851 write_seqlock_irqsave(&xtime_lock, flags);
852
853 nsec -= __get_nsec_offset();
854
855 wtm_sec = wall_to_monotonic.tv_sec + (xtime.tv_sec - sec);
856 wtm_nsec = wall_to_monotonic.tv_nsec + (xtime.tv_nsec - nsec);
857
858 set_normalized_timespec(&xtime, sec, nsec);
859 set_normalized_timespec(&wall_to_monotonic, wtm_sec, wtm_nsec);
860
861 clock->error = 0;
862 ntp_clear();
863
864 write_sequnlock_irqrestore(&xtime_lock, flags);
865
866 /* signal hrtimers about time change */
867 clock_was_set();
868
869 return 0;
870}
871
872EXPORT_SYMBOL(do_settimeofday);
873
874/**
875 * change_clocksource - Swaps clocksources if a new one is available
876 *
877 * Accumulates current time interval and initializes new clocksource
878 */
879static void change_clocksource(void)
880{
881 struct clocksource *new;
882 cycle_t now;
883 u64 nsec;
884
885 new = clocksource_get_next();
886
887 if (clock == new)
888 return;
889
890 now = clocksource_read(new);
891 nsec = __get_nsec_offset();
892 timespec_add_ns(&xtime, nsec);
893
894 clock = new;
895 clock->cycle_last = now;
896
897 clock->error = 0;
898 clock->xtime_nsec = 0;
899 clocksource_calculate_interval(clock, NTP_INTERVAL_LENGTH);
900
901 tick_clock_notify();
902
903 printk(KERN_INFO "Time: %s clocksource has been installed.\n",
904 clock->name);
905}
906#else
907static inline void change_clocksource(void) { }
908#endif
909
910/**
911 * timeofday_is_continuous - check to see if timekeeping is free running
912 */
913int timekeeping_is_continuous(void)
914{
915 unsigned long seq;
916 int ret;
917
918 do {
919 seq = read_seqbegin(&xtime_lock);
920
921 ret = clock->flags & CLOCK_SOURCE_VALID_FOR_HRES;
922
923 } while (read_seqretry(&xtime_lock, seq));
924
925 return ret;
926}
927
928/**
929 * read_persistent_clock - Return time in seconds from the persistent clock.
930 *
931 * Weak dummy function for arches that do not yet support it.
932 * Returns seconds from epoch using the battery backed persistent clock.
933 * Returns zero if unsupported.
934 *
935 * XXX - Do be sure to remove it once all arches implement it.
936 */
937unsigned long __attribute__((weak)) read_persistent_clock(void)
938{
939 return 0;
940}
941
942/*
943 * timekeeping_init - Initializes the clocksource and common timekeeping values
944 */
945void __init timekeeping_init(void)
946{
947 unsigned long flags;
948 unsigned long sec = read_persistent_clock();
949
950 write_seqlock_irqsave(&xtime_lock, flags);
951
952 ntp_clear();
953
954 clock = clocksource_get_next();
955 clocksource_calculate_interval(clock, NTP_INTERVAL_LENGTH);
956 clock->cycle_last = clocksource_read(clock);
957
958 xtime.tv_sec = sec;
959 xtime.tv_nsec = 0;
960 set_normalized_timespec(&wall_to_monotonic,
961 -xtime.tv_sec, -xtime.tv_nsec);
962
963 write_sequnlock_irqrestore(&xtime_lock, flags);
964}
965
966/* flag for if timekeeping is suspended */
967static int timekeeping_suspended;
968/* time in seconds when suspend began */
969static unsigned long timekeeping_suspend_time;
970
971/**
972 * timekeeping_resume - Resumes the generic timekeeping subsystem.
973 * @dev: unused
974 *
975 * This is for the generic clocksource timekeeping.
976 * xtime/wall_to_monotonic/jiffies/etc are
977 * still managed by arch specific suspend/resume code.
978 */
979static int timekeeping_resume(struct sys_device *dev)
980{
981 unsigned long flags;
982 unsigned long now = read_persistent_clock();
983
984 write_seqlock_irqsave(&xtime_lock, flags);
985
986 if (now && (now > timekeeping_suspend_time)) {
987 unsigned long sleep_length = now - timekeeping_suspend_time;
988
989 xtime.tv_sec += sleep_length;
990 wall_to_monotonic.tv_sec -= sleep_length;
991 }
992 /* re-base the last cycle value */
993 clock->cycle_last = clocksource_read(clock);
994 clock->error = 0;
995 timekeeping_suspended = 0;
996 write_sequnlock_irqrestore(&xtime_lock, flags);
997
998 touch_softlockup_watchdog();
999 /* Resume hrtimers */
1000 clock_was_set();
1001
1002 return 0;
1003}
1004
1005static int timekeeping_suspend(struct sys_device *dev, pm_message_t state)
1006{
1007 unsigned long flags;
1008
1009 write_seqlock_irqsave(&xtime_lock, flags);
1010 timekeeping_suspended = 1;
1011 timekeeping_suspend_time = read_persistent_clock();
1012 write_sequnlock_irqrestore(&xtime_lock, flags);
1013 return 0;
1014}
1015
1016/* sysfs resume/suspend bits for timekeeping */
1017static struct sysdev_class timekeeping_sysclass = {
1018 .resume = timekeeping_resume,
1019 .suspend = timekeeping_suspend,
1020 set_kset_name("timekeeping"),
1021};
1022
1023static struct sys_device device_timer = {
1024 .id = 0,
1025 .cls = &timekeeping_sysclass,
1026};
1027
1028static int __init timekeeping_init_device(void)
1029{
1030 int error = sysdev_class_register(&timekeeping_sysclass);
1031 if (!error)
1032 error = sysdev_register(&device_timer);
1033 return error;
1034}
1035
1036device_initcall(timekeeping_init_device);
1037
1038/*
1039 * If the error is already larger, we look ahead even further
1040 * to compensate for late or lost adjustments.
1041 */
1042static __always_inline int clocksource_bigadjust(s64 error, s64 *interval,
1043 s64 *offset)
1044{
1045 s64 tick_error, i;
1046 u32 look_ahead, adj;
1047 s32 error2, mult;
1048
1049 /*
1050 * Use the current error value to determine how much to look ahead.
1051 * The larger the error the slower we adjust for it to avoid problems
1052 * with losing too many ticks, otherwise we would overadjust and
1053 * produce an even larger error. The smaller the adjustment the
1054 * faster we try to adjust for it, as lost ticks can do less harm
1055 * here. This is tuned so that an error of about 1 msec is adusted
1056 * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks).
1057 */
1058 error2 = clock->error >> (TICK_LENGTH_SHIFT + 22 - 2 * SHIFT_HZ);
1059 error2 = abs(error2);
1060 for (look_ahead = 0; error2 > 0; look_ahead++)
1061 error2 >>= 2;
1062
1063 /*
1064 * Now calculate the error in (1 << look_ahead) ticks, but first
1065 * remove the single look ahead already included in the error.
1066 */
1067 tick_error = current_tick_length() >>
1068 (TICK_LENGTH_SHIFT - clock->shift + 1);
1069 tick_error -= clock->xtime_interval >> 1;
1070 error = ((error - tick_error) >> look_ahead) + tick_error;
1071
1072 /* Finally calculate the adjustment shift value. */
1073 i = *interval;
1074 mult = 1;
1075 if (error < 0) {
1076 error = -error;
1077 *interval = -*interval;
1078 *offset = -*offset;
1079 mult = -1;
1080 }
1081 for (adj = 0; error > i; adj++)
1082 error >>= 1;
1083
1084 *interval <<= adj;
1085 *offset <<= adj;
1086 return mult << adj;
1087}
1088
1089/*
1090 * Adjust the multiplier to reduce the error value,
1091 * this is optimized for the most common adjustments of -1,0,1,
1092 * for other values we can do a bit more work.
1093 */
1094static void clocksource_adjust(struct clocksource *clock, s64 offset)
1095{
1096 s64 error, interval = clock->cycle_interval;
1097 int adj;
1098
1099 error = clock->error >> (TICK_LENGTH_SHIFT - clock->shift - 1);
1100 if (error > interval) {
1101 error >>= 2;
1102 if (likely(error <= interval))
1103 adj = 1;
1104 else
1105 adj = clocksource_bigadjust(error, &interval, &offset);
1106 } else if (error < -interval) {
1107 error >>= 2;
1108 if (likely(error >= -interval)) {
1109 adj = -1;
1110 interval = -interval;
1111 offset = -offset;
1112 } else
1113 adj = clocksource_bigadjust(error, &interval, &offset);
1114 } else
1115 return;
1116
1117 clock->mult += adj;
1118 clock->xtime_interval += interval;
1119 clock->xtime_nsec -= offset;
1120 clock->error -= (interval - offset) <<
1121 (TICK_LENGTH_SHIFT - clock->shift);
1122}
1123
1124/**
1125 * update_wall_time - Uses the current clocksource to increment the wall time
1126 *
1127 * Called from the timer interrupt, must hold a write on xtime_lock.
1128 */
1129static void update_wall_time(void)
1130{
1131 cycle_t offset;
1132
1133 /* Make sure we're fully resumed: */
1134 if (unlikely(timekeeping_suspended))
1135 return;
1136
1137#ifdef CONFIG_GENERIC_TIME
1138 offset = (clocksource_read(clock) - clock->cycle_last) & clock->mask;
1139#else
1140 offset = clock->cycle_interval;
1141#endif
1142 clock->xtime_nsec += (s64)xtime.tv_nsec << clock->shift;
1143
1144 /* normally this loop will run just once, however in the
1145 * case of lost or late ticks, it will accumulate correctly.
1146 */
1147 while (offset >= clock->cycle_interval) {
1148 /* accumulate one interval */
1149 clock->xtime_nsec += clock->xtime_interval;
1150 clock->cycle_last += clock->cycle_interval;
1151 offset -= clock->cycle_interval;
1152
1153 if (clock->xtime_nsec >= (u64)NSEC_PER_SEC << clock->shift) {
1154 clock->xtime_nsec -= (u64)NSEC_PER_SEC << clock->shift;
1155 xtime.tv_sec++;
1156 second_overflow();
1157 }
1158
1159 /* interpolator bits */
1160 time_interpolator_update(clock->xtime_interval
1161 >> clock->shift);
1162
1163 /* accumulate error between NTP and clock interval */
1164 clock->error += current_tick_length();
1165 clock->error -= clock->xtime_interval << (TICK_LENGTH_SHIFT - clock->shift);
1166 }
1167
1168 /* correct the clock when NTP error is too big */
1169 clocksource_adjust(clock, offset);
1170
1171 /* store full nanoseconds into xtime */
1172 xtime.tv_nsec = (s64)clock->xtime_nsec >> clock->shift;
1173 clock->xtime_nsec -= (s64)xtime.tv_nsec << clock->shift;
1174
1175 /* check to see if there is a new clocksource to use */
1176 change_clocksource();
1177 update_vsyscall(&xtime, clock);
1178}
1179
1180/*
1181 * Called from the timer interrupt handler to charge one tick to the current
1182 * process. user_tick is 1 if the tick is user time, 0 for system.
1183 */
1184void update_process_times(int user_tick)
1185{
1186 struct task_struct *p = current;
1187 int cpu = smp_processor_id();
1188
1189 /* Note: this timer irq context must be accounted for as well. */
1190 if (user_tick)
1191 account_user_time(p, jiffies_to_cputime(1));
1192 else
1193 account_system_time(p, HARDIRQ_OFFSET, jiffies_to_cputime(1));
1194 run_local_timers();
1195 if (rcu_pending(cpu))
1196 rcu_check_callbacks(cpu, user_tick);
1197 scheduler_tick();
1198 run_posix_cpu_timers(p);
1199}
1200
1201/*
1202 * Nr of active tasks - counted in fixed-point numbers
1203 */
1204static unsigned long count_active_tasks(void)
1205{
1206 return nr_active() * FIXED_1;
1207}
1208
1209/*
1210 * Hmm.. Changed this, as the GNU make sources (load.c) seems to
1211 * imply that avenrun[] is the standard name for this kind of thing.
1212 * Nothing else seems to be standardized: the fractional size etc
1213 * all seem to differ on different machines.
1214 *
1215 * Requires xtime_lock to access.
1216 */
1217unsigned long avenrun[3];
1218
1219EXPORT_SYMBOL(avenrun);
1220
1221/*
1222 * calc_load - given tick count, update the avenrun load estimates.
1223 * This is called while holding a write_lock on xtime_lock.
1224 */
1225static inline void calc_load(unsigned long ticks)
1226{
1227 unsigned long active_tasks; /* fixed-point */
1228 static int count = LOAD_FREQ;
1229
1230 count -= ticks;
1231 if (unlikely(count < 0)) {
1232 active_tasks = count_active_tasks();
1233 do {
1234 CALC_LOAD(avenrun[0], EXP_1, active_tasks);
1235 CALC_LOAD(avenrun[1], EXP_5, active_tasks);
1236 CALC_LOAD(avenrun[2], EXP_15, active_tasks);
1237 count += LOAD_FREQ;
1238 } while (count < 0);
1239 }
1240}
1241
1242/*
1243 * This read-write spinlock protects us from races in SMP while
1244 * playing with xtime and avenrun.
1245 */
1246__attribute__((weak)) __cacheline_aligned_in_smp DEFINE_SEQLOCK(xtime_lock);
1247
1248EXPORT_SYMBOL(xtime_lock);
1249
1250/*
1251 * This function runs timers and the timer-tq in bottom half context.
1252 */
1253static void run_timer_softirq(struct softirq_action *h)
1254{
1255 tvec_base_t *base = __get_cpu_var(tvec_bases);
1256
1257 hrtimer_run_queues();
1258
1259 if (time_after_eq(jiffies, base->timer_jiffies))
1260 __run_timers(base);
1261}
1262
1263/*
1264 * Called by the local, per-CPU timer interrupt on SMP.
1265 */
1266void run_local_timers(void)
1267{
1268 raise_softirq(TIMER_SOFTIRQ);
1269 softlockup_tick();
1270}
1271
1272/*
1273 * Called by the timer interrupt. xtime_lock must already be taken
1274 * by the timer IRQ!
1275 */
1276static inline void update_times(unsigned long ticks)
1277{
1278 update_wall_time();
1279 calc_load(ticks);
1280}
1281
1282/*
1283 * The 64-bit jiffies value is not atomic - you MUST NOT read it
1284 * without sampling the sequence number in xtime_lock.
1285 * jiffies is defined in the linker script...
1286 */
1287
1288void do_timer(unsigned long ticks)
1289{
1290 jiffies_64 += ticks;
1291 update_times(ticks);
1292}
1293
1294#ifdef __ARCH_WANT_SYS_ALARM
1295
1296/*
1297 * For backwards compatibility? This can be done in libc so Alpha
1298 * and all newer ports shouldn't need it.
1299 */
1300asmlinkage unsigned long sys_alarm(unsigned int seconds)
1301{
1302 return alarm_setitimer(seconds);
1303}
1304
1305#endif
1306
1307#ifndef __alpha__
1308
1309/*
1310 * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this
1311 * should be moved into arch/i386 instead?
1312 */
1313
1314/**
1315 * sys_getpid - return the thread group id of the current process
1316 *
1317 * Note, despite the name, this returns the tgid not the pid. The tgid and
1318 * the pid are identical unless CLONE_THREAD was specified on clone() in
1319 * which case the tgid is the same in all threads of the same group.
1320 *
1321 * This is SMP safe as current->tgid does not change.
1322 */
1323asmlinkage long sys_getpid(void)
1324{
1325 return current->tgid;
1326}
1327
1328/*
1329 * Accessing ->real_parent is not SMP-safe, it could
1330 * change from under us. However, we can use a stale
1331 * value of ->real_parent under rcu_read_lock(), see
1332 * release_task()->call_rcu(delayed_put_task_struct).
1333 */
1334asmlinkage long sys_getppid(void)
1335{
1336 int pid;
1337
1338 rcu_read_lock();
1339 pid = rcu_dereference(current->real_parent)->tgid;
1340 rcu_read_unlock();
1341
1342 return pid;
1343}
1344
1345asmlinkage long sys_getuid(void)
1346{
1347 /* Only we change this so SMP safe */
1348 return current->uid;
1349}
1350
1351asmlinkage long sys_geteuid(void)
1352{
1353 /* Only we change this so SMP safe */
1354 return current->euid;
1355}
1356
1357asmlinkage long sys_getgid(void)
1358{
1359 /* Only we change this so SMP safe */
1360 return current->gid;
1361}
1362
1363asmlinkage long sys_getegid(void)
1364{
1365 /* Only we change this so SMP safe */
1366 return current->egid;
1367}
1368
1369#endif
1370
1371static void process_timeout(unsigned long __data)
1372{
1373 wake_up_process((struct task_struct *)__data);
1374}
1375
1376/**
1377 * schedule_timeout - sleep until timeout
1378 * @timeout: timeout value in jiffies
1379 *
1380 * Make the current task sleep until @timeout jiffies have
1381 * elapsed. The routine will return immediately unless
1382 * the current task state has been set (see set_current_state()).
1383 *
1384 * You can set the task state as follows -
1385 *
1386 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1387 * pass before the routine returns. The routine will return 0
1388 *
1389 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1390 * delivered to the current task. In this case the remaining time
1391 * in jiffies will be returned, or 0 if the timer expired in time
1392 *
1393 * The current task state is guaranteed to be TASK_RUNNING when this
1394 * routine returns.
1395 *
1396 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1397 * the CPU away without a bound on the timeout. In this case the return
1398 * value will be %MAX_SCHEDULE_TIMEOUT.
1399 *
1400 * In all cases the return value is guaranteed to be non-negative.
1401 */
1402fastcall signed long __sched schedule_timeout(signed long timeout)
1403{
1404 struct timer_list timer;
1405 unsigned long expire;
1406
1407 switch (timeout)
1408 {
1409 case MAX_SCHEDULE_TIMEOUT:
1410 /*
1411 * These two special cases are useful to be comfortable
1412 * in the caller. Nothing more. We could take
1413 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1414 * but I' d like to return a valid offset (>=0) to allow
1415 * the caller to do everything it want with the retval.
1416 */
1417 schedule();
1418 goto out;
1419 default:
1420 /*
1421 * Another bit of PARANOID. Note that the retval will be
1422 * 0 since no piece of kernel is supposed to do a check
1423 * for a negative retval of schedule_timeout() (since it
1424 * should never happens anyway). You just have the printk()
1425 * that will tell you if something is gone wrong and where.
1426 */
1427 if (timeout < 0) {
1428 printk(KERN_ERR "schedule_timeout: wrong timeout "
1429 "value %lx\n", timeout);
1430 dump_stack();
1431 current->state = TASK_RUNNING;
1432 goto out;
1433 }
1434 }
1435
1436 expire = timeout + jiffies;
1437
1438 setup_timer(&timer, process_timeout, (unsigned long)current);
1439 __mod_timer(&timer, expire);
1440 schedule();
1441 del_singleshot_timer_sync(&timer);
1442
1443 timeout = expire - jiffies;
1444
1445 out:
1446 return timeout < 0 ? 0 : timeout;
1447}
1448EXPORT_SYMBOL(schedule_timeout);
1449
1450/*
1451 * We can use __set_current_state() here because schedule_timeout() calls
1452 * schedule() unconditionally.
1453 */
1454signed long __sched schedule_timeout_interruptible(signed long timeout)
1455{
1456 __set_current_state(TASK_INTERRUPTIBLE);
1457 return schedule_timeout(timeout);
1458}
1459EXPORT_SYMBOL(schedule_timeout_interruptible);
1460
1461signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1462{
1463 __set_current_state(TASK_UNINTERRUPTIBLE);
1464 return schedule_timeout(timeout);
1465}
1466EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1467
1468/* Thread ID - the internal kernel "pid" */
1469asmlinkage long sys_gettid(void)
1470{
1471 return current->pid;
1472}
1473
1474/**
1475 * do_sysinfo - fill in sysinfo struct
1476 * @info: pointer to buffer to fill
1477 */
1478int do_sysinfo(struct sysinfo *info)
1479{
1480 unsigned long mem_total, sav_total;
1481 unsigned int mem_unit, bitcount;
1482 unsigned long seq;
1483
1484 memset(info, 0, sizeof(struct sysinfo));
1485
1486 do {
1487 struct timespec tp;
1488 seq = read_seqbegin(&xtime_lock);
1489
1490 /*
1491 * This is annoying. The below is the same thing
1492 * posix_get_clock_monotonic() does, but it wants to
1493 * take the lock which we want to cover the loads stuff
1494 * too.
1495 */
1496
1497 getnstimeofday(&tp);
1498 tp.tv_sec += wall_to_monotonic.tv_sec;
1499 tp.tv_nsec += wall_to_monotonic.tv_nsec;
1500 if (tp.tv_nsec - NSEC_PER_SEC >= 0) {
1501 tp.tv_nsec = tp.tv_nsec - NSEC_PER_SEC;
1502 tp.tv_sec++;
1503 }
1504 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
1505
1506 info->loads[0] = avenrun[0] << (SI_LOAD_SHIFT - FSHIFT);
1507 info->loads[1] = avenrun[1] << (SI_LOAD_SHIFT - FSHIFT);
1508 info->loads[2] = avenrun[2] << (SI_LOAD_SHIFT - FSHIFT);
1509
1510 info->procs = nr_threads;
1511 } while (read_seqretry(&xtime_lock, seq));
1512
1513 si_meminfo(info);
1514 si_swapinfo(info);
1515
1516 /*
1517 * If the sum of all the available memory (i.e. ram + swap)
1518 * is less than can be stored in a 32 bit unsigned long then
1519 * we can be binary compatible with 2.2.x kernels. If not,
1520 * well, in that case 2.2.x was broken anyways...
1521 *
1522 * -Erik Andersen <[email protected]>
1523 */
1524
1525 mem_total = info->totalram + info->totalswap;
1526 if (mem_total < info->totalram || mem_total < info->totalswap)
1527 goto out;
1528 bitcount = 0;
1529 mem_unit = info->mem_unit;
1530 while (mem_unit > 1) {
1531 bitcount++;
1532 mem_unit >>= 1;
1533 sav_total = mem_total;
1534 mem_total <<= 1;
1535 if (mem_total < sav_total)
1536 goto out;
1537 }
1538
1539 /*
1540 * If mem_total did not overflow, multiply all memory values by
1541 * info->mem_unit and set it to 1. This leaves things compatible
1542 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1543 * kernels...
1544 */
1545
1546 info->mem_unit = 1;
1547 info->totalram <<= bitcount;
1548 info->freeram <<= bitcount;
1549 info->sharedram <<= bitcount;
1550 info->bufferram <<= bitcount;
1551 info->totalswap <<= bitcount;
1552 info->freeswap <<= bitcount;
1553 info->totalhigh <<= bitcount;
1554 info->freehigh <<= bitcount;
1555
1556out:
1557 return 0;
1558}
1559
1560asmlinkage long sys_sysinfo(struct sysinfo __user *info)
1561{
1562 struct sysinfo val;
1563
1564 do_sysinfo(&val);
1565
1566 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
1567 return -EFAULT;
1568
1569 return 0;
1570}
1571
1572/*
1573 * lockdep: we want to track each per-CPU base as a separate lock-class,
1574 * but timer-bases are kmalloc()-ed, so we need to attach separate
1575 * keys to them:
1576 */
1577static struct lock_class_key base_lock_keys[NR_CPUS];
1578
1579static int __devinit init_timers_cpu(int cpu)
1580{
1581 int j;
1582 tvec_base_t *base;
1583 static char __devinitdata tvec_base_done[NR_CPUS];
1584
1585 if (!tvec_base_done[cpu]) {
1586 static char boot_done;
1587
1588 if (boot_done) {
1589 /*
1590 * The APs use this path later in boot
1591 */
1592 base = kmalloc_node(sizeof(*base), GFP_KERNEL,
1593 cpu_to_node(cpu));
1594 if (!base)
1595 return -ENOMEM;
1596 memset(base, 0, sizeof(*base));
1597 per_cpu(tvec_bases, cpu) = base;
1598 } else {
1599 /*
1600 * This is for the boot CPU - we use compile-time
1601 * static initialisation because per-cpu memory isn't
1602 * ready yet and because the memory allocators are not
1603 * initialised either.
1604 */
1605 boot_done = 1;
1606 base = &boot_tvec_bases;
1607 }
1608 tvec_base_done[cpu] = 1;
1609 } else {
1610 base = per_cpu(tvec_bases, cpu);
1611 }
1612
1613 spin_lock_init(&base->lock);
1614 lockdep_set_class(&base->lock, base_lock_keys + cpu);
1615
1616 for (j = 0; j < TVN_SIZE; j++) {
1617 INIT_LIST_HEAD(base->tv5.vec + j);
1618 INIT_LIST_HEAD(base->tv4.vec + j);
1619 INIT_LIST_HEAD(base->tv3.vec + j);
1620 INIT_LIST_HEAD(base->tv2.vec + j);
1621 }
1622 for (j = 0; j < TVR_SIZE; j++)
1623 INIT_LIST_HEAD(base->tv1.vec + j);
1624
1625 base->timer_jiffies = jiffies;
1626 return 0;
1627}
1628
1629#ifdef CONFIG_HOTPLUG_CPU
1630static void migrate_timer_list(tvec_base_t *new_base, struct list_head *head)
1631{
1632 struct timer_list *timer;
1633
1634 while (!list_empty(head)) {
1635 timer = list_entry(head->next, struct timer_list, entry);
1636 detach_timer(timer, 0);
1637 timer->base = new_base;
1638 internal_add_timer(new_base, timer);
1639 }
1640}
1641
1642static void __devinit migrate_timers(int cpu)
1643{
1644 tvec_base_t *old_base;
1645 tvec_base_t *new_base;
1646 int i;
1647
1648 BUG_ON(cpu_online(cpu));
1649 old_base = per_cpu(tvec_bases, cpu);
1650 new_base = get_cpu_var(tvec_bases);
1651
1652 local_irq_disable();
1653 spin_lock(&new_base->lock);
1654 spin_lock(&old_base->lock);
1655
1656 BUG_ON(old_base->running_timer);
1657
1658 for (i = 0; i < TVR_SIZE; i++)
1659 migrate_timer_list(new_base, old_base->tv1.vec + i);
1660 for (i = 0; i < TVN_SIZE; i++) {
1661 migrate_timer_list(new_base, old_base->tv2.vec + i);
1662 migrate_timer_list(new_base, old_base->tv3.vec + i);
1663 migrate_timer_list(new_base, old_base->tv4.vec + i);
1664 migrate_timer_list(new_base, old_base->tv5.vec + i);
1665 }
1666
1667 spin_unlock(&old_base->lock);
1668 spin_unlock(&new_base->lock);
1669 local_irq_enable();
1670 put_cpu_var(tvec_bases);
1671}
1672#endif /* CONFIG_HOTPLUG_CPU */
1673
1674static int __cpuinit timer_cpu_notify(struct notifier_block *self,
1675 unsigned long action, void *hcpu)
1676{
1677 long cpu = (long)hcpu;
1678 switch(action) {
1679 case CPU_UP_PREPARE:
1680 if (init_timers_cpu(cpu) < 0)
1681 return NOTIFY_BAD;
1682 break;
1683#ifdef CONFIG_HOTPLUG_CPU
1684 case CPU_DEAD:
1685 migrate_timers(cpu);
1686 break;
1687#endif
1688 default:
1689 break;
1690 }
1691 return NOTIFY_OK;
1692}
1693
1694static struct notifier_block __cpuinitdata timers_nb = {
1695 .notifier_call = timer_cpu_notify,
1696};
1697
1698
1699void __init init_timers(void)
1700{
1701 int err = timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE,
1702 (void *)(long)smp_processor_id());
1703
1704 init_timer_stats();
1705
1706 BUG_ON(err == NOTIFY_BAD);
1707 register_cpu_notifier(&timers_nb);
1708 open_softirq(TIMER_SOFTIRQ, run_timer_softirq, NULL);
1709}
1710
1711#ifdef CONFIG_TIME_INTERPOLATION
1712
1713struct time_interpolator *time_interpolator __read_mostly;
1714static struct time_interpolator *time_interpolator_list __read_mostly;
1715static DEFINE_SPINLOCK(time_interpolator_lock);
1716
1717static inline cycles_t time_interpolator_get_cycles(unsigned int src)
1718{
1719 unsigned long (*x)(void);
1720
1721 switch (src)
1722 {
1723 case TIME_SOURCE_FUNCTION:
1724 x = time_interpolator->addr;
1725 return x();
1726
1727 case TIME_SOURCE_MMIO64 :
1728 return readq_relaxed((void __iomem *)time_interpolator->addr);
1729
1730 case TIME_SOURCE_MMIO32 :
1731 return readl_relaxed((void __iomem *)time_interpolator->addr);
1732
1733 default: return get_cycles();
1734 }
1735}
1736
1737static inline u64 time_interpolator_get_counter(int writelock)
1738{
1739 unsigned int src = time_interpolator->source;
1740
1741 if (time_interpolator->jitter)
1742 {
1743 cycles_t lcycle;
1744 cycles_t now;
1745
1746 do {
1747 lcycle = time_interpolator->last_cycle;
1748 now = time_interpolator_get_cycles(src);
1749 if (lcycle && time_after(lcycle, now))
1750 return lcycle;
1751
1752 /* When holding the xtime write lock, there's no need
1753 * to add the overhead of the cmpxchg. Readers are
1754 * force to retry until the write lock is released.
1755 */
1756 if (writelock) {
1757 time_interpolator->last_cycle = now;
1758 return now;
1759 }
1760 /* Keep track of the last timer value returned. The use of cmpxchg here
1761 * will cause contention in an SMP environment.
1762 */
1763 } while (unlikely(cmpxchg(&time_interpolator->last_cycle, lcycle, now) != lcycle));
1764 return now;
1765 }
1766 else
1767 return time_interpolator_get_cycles(src);
1768}
1769
1770void time_interpolator_reset(void)
1771{
1772 time_interpolator->offset = 0;
1773 time_interpolator->last_counter = time_interpolator_get_counter(1);
1774}
1775
1776#define GET_TI_NSECS(count,i) (((((count) - i->last_counter) & (i)->mask) * (i)->nsec_per_cyc) >> (i)->shift)
1777
1778unsigned long time_interpolator_get_offset(void)
1779{
1780 /* If we do not have a time interpolator set up then just return zero */
1781 if (!time_interpolator)
1782 return 0;
1783
1784 return time_interpolator->offset +
1785 GET_TI_NSECS(time_interpolator_get_counter(0), time_interpolator);
1786}
1787
1788#define INTERPOLATOR_ADJUST 65536
1789#define INTERPOLATOR_MAX_SKIP 10*INTERPOLATOR_ADJUST
1790
1791void time_interpolator_update(long delta_nsec)
1792{
1793 u64 counter;
1794 unsigned long offset;
1795
1796 /* If there is no time interpolator set up then do nothing */
1797 if (!time_interpolator)
1798 return;
1799
1800 /*
1801 * The interpolator compensates for late ticks by accumulating the late
1802 * time in time_interpolator->offset. A tick earlier than expected will
1803 * lead to a reset of the offset and a corresponding jump of the clock
1804 * forward. Again this only works if the interpolator clock is running
1805 * slightly slower than the regular clock and the tuning logic insures
1806 * that.
1807 */
1808
1809 counter = time_interpolator_get_counter(1);
1810 offset = time_interpolator->offset +
1811 GET_TI_NSECS(counter, time_interpolator);
1812
1813 if (delta_nsec < 0 || (unsigned long) delta_nsec < offset)
1814 time_interpolator->offset = offset - delta_nsec;
1815 else {
1816 time_interpolator->skips++;
1817 time_interpolator->ns_skipped += delta_nsec - offset;
1818 time_interpolator->offset = 0;
1819 }
1820 time_interpolator->last_counter = counter;
1821
1822 /* Tuning logic for time interpolator invoked every minute or so.
1823 * Decrease interpolator clock speed if no skips occurred and an offset is carried.
1824 * Increase interpolator clock speed if we skip too much time.
1825 */
1826 if (jiffies % INTERPOLATOR_ADJUST == 0)
1827 {
1828 if (time_interpolator->skips == 0 && time_interpolator->offset > tick_nsec)
1829 time_interpolator->nsec_per_cyc--;
1830 if (time_interpolator->ns_skipped > INTERPOLATOR_MAX_SKIP && time_interpolator->offset == 0)
1831 time_interpolator->nsec_per_cyc++;
1832 time_interpolator->skips = 0;
1833 time_interpolator->ns_skipped = 0;
1834 }
1835}
1836
1837static inline int
1838is_better_time_interpolator(struct time_interpolator *new)
1839{
1840 if (!time_interpolator)
1841 return 1;
1842 return new->frequency > 2*time_interpolator->frequency ||
1843 (unsigned long)new->drift < (unsigned long)time_interpolator->drift;
1844}
1845
1846void
1847register_time_interpolator(struct time_interpolator *ti)
1848{
1849 unsigned long flags;
1850
1851 /* Sanity check */
1852 BUG_ON(ti->frequency == 0 || ti->mask == 0);
1853
1854 ti->nsec_per_cyc = ((u64)NSEC_PER_SEC << ti->shift) / ti->frequency;
1855 spin_lock(&time_interpolator_lock);
1856 write_seqlock_irqsave(&xtime_lock, flags);
1857 if (is_better_time_interpolator(ti)) {
1858 time_interpolator = ti;
1859 time_interpolator_reset();
1860 }
1861 write_sequnlock_irqrestore(&xtime_lock, flags);
1862
1863 ti->next = time_interpolator_list;
1864 time_interpolator_list = ti;
1865 spin_unlock(&time_interpolator_lock);
1866}
1867
1868void
1869unregister_time_interpolator(struct time_interpolator *ti)
1870{
1871 struct time_interpolator *curr, **prev;
1872 unsigned long flags;
1873
1874 spin_lock(&time_interpolator_lock);
1875 prev = &time_interpolator_list;
1876 for (curr = *prev; curr; curr = curr->next) {
1877 if (curr == ti) {
1878 *prev = curr->next;
1879 break;
1880 }
1881 prev = &curr->next;
1882 }
1883
1884 write_seqlock_irqsave(&xtime_lock, flags);
1885 if (ti == time_interpolator) {
1886 /* we lost the best time-interpolator: */
1887 time_interpolator = NULL;
1888 /* find the next-best interpolator */
1889 for (curr = time_interpolator_list; curr; curr = curr->next)
1890 if (is_better_time_interpolator(curr))
1891 time_interpolator = curr;
1892 time_interpolator_reset();
1893 }
1894 write_sequnlock_irqrestore(&xtime_lock, flags);
1895 spin_unlock(&time_interpolator_lock);
1896}
1897#endif /* CONFIG_TIME_INTERPOLATION */
1898
1899/**
1900 * msleep - sleep safely even with waitqueue interruptions
1901 * @msecs: Time in milliseconds to sleep for
1902 */
1903void msleep(unsigned int msecs)
1904{
1905 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1906
1907 while (timeout)
1908 timeout = schedule_timeout_uninterruptible(timeout);
1909}
1910
1911EXPORT_SYMBOL(msleep);
1912
1913/**
1914 * msleep_interruptible - sleep waiting for signals
1915 * @msecs: Time in milliseconds to sleep for
1916 */
1917unsigned long msleep_interruptible(unsigned int msecs)
1918{
1919 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1920
1921 while (timeout && !signal_pending(current))
1922 timeout = schedule_timeout_interruptible(timeout);
1923 return jiffies_to_msecs(timeout);
1924}
1925
1926EXPORT_SYMBOL(msleep_interruptible);