ia64/linux-2.6.18-xen.hg

view kernel/timer.c @ 912:dd42cdb0ab89

[IA64] Build blktap2 driver by default in x86 builds.

add CONFIG_XEN_BLKDEV_TAP2=y to buildconfigs/linux-defconfig_xen_ia64.

Signed-off-by: Isaku Yamahata <yamahata@valinux.co.jp>
author Isaku Yamahata <yamahata@valinux.co.jp>
date Mon Jun 29 12:09:16 2009 +0900 (2009-06-29)
parents 61ed8662b69c
children
line source
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 */
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>
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>
44 #ifdef CONFIG_TIME_INTERPOLATION
45 static void time_interpolator_update(long delta_nsec);
46 #else
47 #define time_interpolator_update(x)
48 #endif
50 u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
52 EXPORT_SYMBOL(jiffies_64);
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)
64 typedef struct tvec_s {
65 struct list_head vec[TVN_SIZE];
66 } tvec_t;
68 typedef struct tvec_root_s {
69 struct list_head vec[TVR_SIZE];
70 } tvec_root_t;
72 struct 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;
83 typedef struct tvec_t_base_s tvec_base_t;
85 tvec_base_t boot_tvec_bases;
86 EXPORT_SYMBOL(boot_tvec_bases);
87 static DEFINE_PER_CPU(tvec_base_t *, tvec_bases) = &boot_tvec_bases;
89 static 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 }
97 static 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;
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 }
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 */
146 void fastcall init_timer(struct timer_list *timer)
147 {
148 timer->entry.next = NULL;
149 timer->base = __raw_get_cpu_var(tvec_bases);
150 }
151 EXPORT_SYMBOL(init_timer);
153 static inline void detach_timer(struct timer_list *timer,
154 int clear_pending)
155 {
156 struct list_head *entry = &timer->entry;
158 __list_del(entry->prev, entry->next);
159 if (clear_pending)
160 entry->next = NULL;
161 entry->prev = LIST_POISON2;
162 }
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 */
176 static tvec_base_t *lock_timer_base(struct timer_list *timer,
177 unsigned long *flags)
178 {
179 tvec_base_t *base;
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 }
194 int __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;
200 BUG_ON(!timer->function);
202 base = lock_timer_base(timer, &flags);
204 if (timer_pending(timer)) {
205 detach_timer(timer, 0);
206 ret = 1;
207 }
209 new_base = __get_cpu_var(tvec_bases);
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 }
229 timer->expires = expires;
230 internal_add_timer(base, timer);
231 spin_unlock_irqrestore(&base->lock, flags);
233 return ret;
234 }
236 EXPORT_SYMBOL(__mod_timer);
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 */
245 void add_timer_on(struct timer_list *timer, int cpu)
246 {
247 tvec_base_t *base = per_cpu(tvec_bases, cpu);
248 unsigned long flags;
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 }
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 */
277 int mod_timer(struct timer_list *timer, unsigned long expires)
278 {
279 BUG_ON(!timer->function);
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;
289 return __mod_timer(timer, expires);
290 }
292 EXPORT_SYMBOL(mod_timer);
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 */
305 int del_timer(struct timer_list *timer)
306 {
307 tvec_base_t *base;
308 unsigned long flags;
309 int ret = 0;
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 }
320 return ret;
321 }
323 EXPORT_SYMBOL(del_timer);
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 */
332 int try_to_del_timer_sync(struct timer_list *timer)
333 {
334 tvec_base_t *base;
335 unsigned long flags;
336 int ret = -1;
338 base = lock_timer_base(timer, &flags);
340 if (base->running_timer == timer)
341 goto out;
343 ret = 0;
344 if (timer_pending(timer)) {
345 detach_timer(timer, 1);
346 ret = 1;
347 }
348 out:
349 spin_unlock_irqrestore(&base->lock, flags);
351 return ret;
352 }
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 */
371 int 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 }
381 EXPORT_SYMBOL(del_timer_sync);
382 #endif
384 static 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;
390 list_replace_init(tv->vec + index, &tv_list);
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 }
401 return index;
402 }
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)
413 static inline void __run_timers(tvec_base_t *base)
414 {
415 struct timer_list *timer;
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;
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;
437 timer = list_entry(head->next,struct timer_list,entry);
438 fn = timer->function;
439 data = timer->data;
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 }
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 */
469 unsigned 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;
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 = min_t(unsigned long,
489 softlockup_get_next_event(),
490 hr_expires) + jiffies;
492 base = __get_cpu_var(tvec_bases);
493 spin_lock(&base->lock);
494 expires = base->timer_jiffies + (LONG_MAX >> 1);
495 list = NULL;
497 /* Look for timer events in tv1. */
498 j = base->timer_jiffies & TVR_MASK;
499 do {
500 list_for_each_entry(nte, base->tv1.vec + j, entry) {
501 expires = nte->expires;
502 if (j < (base->timer_jiffies & TVR_MASK))
503 list = base->tv2.vec + (INDEX(0));
504 goto found;
505 }
506 j = (j + 1) & TVR_MASK;
507 } while (j != (base->timer_jiffies & TVR_MASK));
509 /* Check tv2-tv5. */
510 varray[0] = &base->tv2;
511 varray[1] = &base->tv3;
512 varray[2] = &base->tv4;
513 varray[3] = &base->tv5;
514 for (i = 0; i < 4; i++) {
515 j = INDEX(i);
516 do {
517 if (list_empty(varray[i]->vec + j)) {
518 j = (j + 1) & TVN_MASK;
519 continue;
520 }
521 list_for_each_entry(nte, varray[i]->vec + j, entry)
522 if (time_before(nte->expires, expires))
523 expires = nte->expires;
524 if (j < (INDEX(i)) && i < 3)
525 list = varray[i + 1]->vec + (INDEX(i + 1));
526 goto found;
527 } while (j != (INDEX(i)));
528 }
529 found:
530 if (list) {
531 /*
532 * The search wrapped. We need to look at the next list
533 * from next tv element that would cascade into tv element
534 * where we found the timer element.
535 */
536 list_for_each_entry(nte, list, entry) {
537 if (time_before(nte->expires, expires))
538 expires = nte->expires;
539 }
540 }
541 spin_unlock(&base->lock);
543 /*
544 * It can happen that other CPUs service timer IRQs and increment
545 * jiffies, but we have not yet got a local timer tick to process
546 * the timer wheels. In that case, the expiry time can be before
547 * jiffies, but since the high-resolution timer here is relative to
548 * jiffies, the default expression when high-resolution timers are
549 * not active,
550 *
551 * time_before(MAX_JIFFY_OFFSET + jiffies, expires)
552 *
553 * would falsely evaluate to true. If that is the case, just
554 * return jiffies so that we can immediately fire the local timer
555 */
556 if (time_before(expires, jiffies))
557 return jiffies;
559 if (time_before(hr_expires, expires))
560 return hr_expires;
562 return expires;
563 }
564 #endif
566 /******************************************************************/
568 /*
569 * Timekeeping variables
570 */
571 unsigned long tick_usec = TICK_USEC; /* USER_HZ period (usec) */
572 unsigned long tick_nsec = TICK_NSEC; /* ACTHZ period (nsec) */
574 /*
575 * The current time
576 * wall_to_monotonic is what we need to add to xtime (or xtime corrected
577 * for sub jiffie times) to get to monotonic time. Monotonic is pegged
578 * at zero at system boot time, so wall_to_monotonic will be negative,
579 * however, we will ALWAYS keep the tv_nsec part positive so we can use
580 * the usual normalization.
581 */
582 struct timespec xtime __attribute__ ((aligned (16)));
583 struct timespec wall_to_monotonic __attribute__ ((aligned (16)));
585 EXPORT_SYMBOL(xtime);
587 /* Don't completely fail for HZ > 500. */
588 int tickadj = 500/HZ ? : 1; /* microsecs */
591 /*
592 * phase-lock loop variables
593 */
594 /* TIME_ERROR prevents overwriting the CMOS clock */
595 int time_state = TIME_OK; /* clock synchronization status */
596 int time_status = STA_UNSYNC; /* clock status bits */
597 long time_offset; /* time adjustment (us) */
598 long time_constant = 2; /* pll time constant */
599 long time_tolerance = MAXFREQ; /* frequency tolerance (ppm) */
600 long time_precision = 1; /* clock precision (us) */
601 long time_maxerror = NTP_PHASE_LIMIT; /* maximum error (us) */
602 long time_esterror = NTP_PHASE_LIMIT; /* estimated error (us) */
603 long time_freq = (((NSEC_PER_SEC + HZ/2) % HZ - HZ/2) << SHIFT_USEC) / NSEC_PER_USEC;
604 /* frequency offset (scaled ppm)*/
605 static long time_adj; /* tick adjust (scaled 1 / HZ) */
606 long time_reftime; /* time at last adjustment (s) */
607 long time_adjust;
608 long time_next_adjust;
610 /*
611 * this routine handles the overflow of the microsecond field
612 *
613 * The tricky bits of code to handle the accurate clock support
614 * were provided by Dave Mills (Mills@UDEL.EDU) of NTP fame.
615 * They were originally developed for SUN and DEC kernels.
616 * All the kudos should go to Dave for this stuff.
617 *
618 */
619 static void second_overflow(void)
620 {
621 long ltemp;
623 /* Bump the maxerror field */
624 time_maxerror += time_tolerance >> SHIFT_USEC;
625 if (time_maxerror > NTP_PHASE_LIMIT) {
626 time_maxerror = NTP_PHASE_LIMIT;
627 time_status |= STA_UNSYNC;
628 }
630 /*
631 * Leap second processing. If in leap-insert state at the end of the
632 * day, the system clock is set back one second; if in leap-delete
633 * state, the system clock is set ahead one second. The microtime()
634 * routine or external clock driver will insure that reported time is
635 * always monotonic. The ugly divides should be replaced.
636 */
637 switch (time_state) {
638 case TIME_OK:
639 if (time_status & STA_INS)
640 time_state = TIME_INS;
641 else if (time_status & STA_DEL)
642 time_state = TIME_DEL;
643 break;
644 case TIME_INS:
645 if (xtime.tv_sec % 86400 == 0) {
646 xtime.tv_sec--;
647 wall_to_monotonic.tv_sec++;
648 /*
649 * The timer interpolator will make time change
650 * gradually instead of an immediate jump by one second
651 */
652 time_interpolator_update(-NSEC_PER_SEC);
653 time_state = TIME_OOP;
654 clock_was_set();
655 printk(KERN_NOTICE "Clock: inserting leap second "
656 "23:59:60 UTC\n");
657 }
658 break;
659 case TIME_DEL:
660 if ((xtime.tv_sec + 1) % 86400 == 0) {
661 xtime.tv_sec++;
662 wall_to_monotonic.tv_sec--;
663 /*
664 * Use of time interpolator for a gradual change of
665 * time
666 */
667 time_interpolator_update(NSEC_PER_SEC);
668 time_state = TIME_WAIT;
669 clock_was_set();
670 printk(KERN_NOTICE "Clock: deleting leap second "
671 "23:59:59 UTC\n");
672 }
673 break;
674 case TIME_OOP:
675 time_state = TIME_WAIT;
676 break;
677 case TIME_WAIT:
678 if (!(time_status & (STA_INS | STA_DEL)))
679 time_state = TIME_OK;
680 }
682 /*
683 * Compute the phase adjustment for the next second. In PLL mode, the
684 * offset is reduced by a fixed factor times the time constant. In FLL
685 * mode the offset is used directly. In either mode, the maximum phase
686 * adjustment for each second is clamped so as to spread the adjustment
687 * over not more than the number of seconds between updates.
688 */
689 ltemp = time_offset;
690 if (!(time_status & STA_FLL))
691 ltemp = shift_right(ltemp, SHIFT_KG + time_constant);
692 ltemp = min(ltemp, (MAXPHASE / MINSEC) << SHIFT_UPDATE);
693 ltemp = max(ltemp, -(MAXPHASE / MINSEC) << SHIFT_UPDATE);
694 time_offset -= ltemp;
695 time_adj = ltemp << (SHIFT_SCALE - SHIFT_HZ - SHIFT_UPDATE);
697 /*
698 * Compute the frequency estimate and additional phase adjustment due
699 * to frequency error for the next second.
700 */
701 ltemp = time_freq;
702 time_adj += shift_right(ltemp,(SHIFT_USEC + SHIFT_HZ - SHIFT_SCALE));
704 #if HZ == 100
705 /*
706 * Compensate for (HZ==100) != (1 << SHIFT_HZ). Add 25% and 3.125% to
707 * get 128.125; => only 0.125% error (p. 14)
708 */
709 time_adj += shift_right(time_adj, 2) + shift_right(time_adj, 5);
710 #endif
711 #if HZ == 250
712 /*
713 * Compensate for (HZ==250) != (1 << SHIFT_HZ). Add 1.5625% and
714 * 0.78125% to get 255.85938; => only 0.05% error (p. 14)
715 */
716 time_adj += shift_right(time_adj, 6) + shift_right(time_adj, 7);
717 #endif
718 #if HZ == 1000
719 /*
720 * Compensate for (HZ==1000) != (1 << SHIFT_HZ). Add 1.5625% and
721 * 0.78125% to get 1023.4375; => only 0.05% error (p. 14)
722 */
723 time_adj += shift_right(time_adj, 6) + shift_right(time_adj, 7);
724 #endif
725 }
727 /*
728 * Returns how many microseconds we need to add to xtime this tick
729 * in doing an adjustment requested with adjtime.
730 */
731 static long adjtime_adjustment(void)
732 {
733 long time_adjust_step;
735 time_adjust_step = time_adjust;
736 if (time_adjust_step) {
737 /*
738 * We are doing an adjtime thing. Prepare time_adjust_step to
739 * be within bounds. Note that a positive time_adjust means we
740 * want the clock to run faster.
741 *
742 * Limit the amount of the step to be in the range
743 * -tickadj .. +tickadj
744 */
745 time_adjust_step = min(time_adjust_step, (long)tickadj);
746 time_adjust_step = max(time_adjust_step, (long)-tickadj);
747 }
748 return time_adjust_step;
749 }
751 /* in the NTP reference this is called "hardclock()" */
752 static void update_ntp_one_tick(void)
753 {
754 long time_adjust_step;
756 time_adjust_step = adjtime_adjustment();
757 if (time_adjust_step)
758 /* Reduce by this step the amount of time left */
759 time_adjust -= time_adjust_step;
761 /* Changes by adjtime() do not take effect till next tick. */
762 if (time_next_adjust != 0) {
763 time_adjust = time_next_adjust;
764 time_next_adjust = 0;
765 }
766 }
768 /*
769 * Return how long ticks are at the moment, that is, how much time
770 * update_wall_time_one_tick will add to xtime next time we call it
771 * (assuming no calls to do_adjtimex in the meantime).
772 * The return value is in fixed-point nanoseconds shifted by the
773 * specified number of bits to the right of the binary point.
774 * This function has no side-effects.
775 */
776 u64 current_tick_length(void)
777 {
778 long delta_nsec;
779 u64 ret;
781 /* calculate the finest interval NTP will allow.
782 * ie: nanosecond value shifted by (SHIFT_SCALE - 10)
783 */
784 delta_nsec = tick_nsec + adjtime_adjustment() * 1000;
785 ret = (u64)delta_nsec << TICK_LENGTH_SHIFT;
786 ret += (s64)time_adj << (TICK_LENGTH_SHIFT - (SHIFT_SCALE - 10));
788 return ret;
789 }
791 /* XXX - all of this timekeeping code should be later moved to time.c */
792 #include <linux/clocksource.h>
793 static struct clocksource *clock; /* pointer to current clocksource */
795 #ifdef CONFIG_GENERIC_TIME
796 /**
797 * __get_nsec_offset - Returns nanoseconds since last call to periodic_hook
798 *
799 * private function, must hold xtime_lock lock when being
800 * called. Returns the number of nanoseconds since the
801 * last call to update_wall_time() (adjusted by NTP scaling)
802 */
803 static inline s64 __get_nsec_offset(void)
804 {
805 cycle_t cycle_now, cycle_delta;
806 s64 ns_offset;
808 /* read clocksource: */
809 cycle_now = clocksource_read(clock);
811 /* calculate the delta since the last update_wall_time: */
812 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
814 /* convert to nanoseconds: */
815 ns_offset = cyc2ns(clock, cycle_delta);
817 return ns_offset;
818 }
820 /**
821 * __get_realtime_clock_ts - Returns the time of day in a timespec
822 * @ts: pointer to the timespec to be set
823 *
824 * Returns the time of day in a timespec. Used by
825 * do_gettimeofday() and get_realtime_clock_ts().
826 */
827 static inline void __get_realtime_clock_ts(struct timespec *ts)
828 {
829 unsigned long seq;
830 s64 nsecs;
832 do {
833 seq = read_seqbegin(&xtime_lock);
835 *ts = xtime;
836 nsecs = __get_nsec_offset();
838 } while (read_seqretry(&xtime_lock, seq));
840 timespec_add_ns(ts, nsecs);
841 }
843 /**
844 * getnstimeofday - Returns the time of day in a timespec
845 * @ts: pointer to the timespec to be set
846 *
847 * Returns the time of day in a timespec.
848 */
849 void getnstimeofday(struct timespec *ts)
850 {
851 __get_realtime_clock_ts(ts);
852 }
854 EXPORT_SYMBOL(getnstimeofday);
856 /**
857 * do_gettimeofday - Returns the time of day in a timeval
858 * @tv: pointer to the timeval to be set
859 *
860 * NOTE: Users should be converted to using get_realtime_clock_ts()
861 */
862 void do_gettimeofday(struct timeval *tv)
863 {
864 struct timespec now;
866 __get_realtime_clock_ts(&now);
867 tv->tv_sec = now.tv_sec;
868 tv->tv_usec = now.tv_nsec/1000;
869 }
871 EXPORT_SYMBOL(do_gettimeofday);
872 /**
873 * do_settimeofday - Sets the time of day
874 * @tv: pointer to the timespec variable containing the new time
875 *
876 * Sets the time of day to the new time and update NTP and notify hrtimers
877 */
878 int do_settimeofday(struct timespec *tv)
879 {
880 unsigned long flags;
881 time_t wtm_sec, sec = tv->tv_sec;
882 long wtm_nsec, nsec = tv->tv_nsec;
884 if ((unsigned long)tv->tv_nsec >= NSEC_PER_SEC)
885 return -EINVAL;
887 write_seqlock_irqsave(&xtime_lock, flags);
889 nsec -= __get_nsec_offset();
891 wtm_sec = wall_to_monotonic.tv_sec + (xtime.tv_sec - sec);
892 wtm_nsec = wall_to_monotonic.tv_nsec + (xtime.tv_nsec - nsec);
894 set_normalized_timespec(&xtime, sec, nsec);
895 set_normalized_timespec(&wall_to_monotonic, wtm_sec, wtm_nsec);
897 clock->error = 0;
898 ntp_clear();
900 write_sequnlock_irqrestore(&xtime_lock, flags);
902 /* signal hrtimers about time change */
903 clock_was_set();
905 return 0;
906 }
908 EXPORT_SYMBOL(do_settimeofday);
910 /**
911 * change_clocksource - Swaps clocksources if a new one is available
912 *
913 * Accumulates current time interval and initializes new clocksource
914 */
915 static int change_clocksource(void)
916 {
917 struct clocksource *new;
918 cycle_t now;
919 u64 nsec;
920 new = clocksource_get_next();
921 if (clock != new) {
922 now = clocksource_read(new);
923 nsec = __get_nsec_offset();
924 timespec_add_ns(&xtime, nsec);
926 clock = new;
927 clock->cycle_last = now;
928 printk(KERN_INFO "Time: %s clocksource has been installed.\n",
929 clock->name);
930 return 1;
931 } else if (clock->update_callback) {
932 return clock->update_callback();
933 }
934 return 0;
935 }
936 #else
937 #define change_clocksource() (0)
938 #endif
940 /**
941 * timeofday_is_continuous - check to see if timekeeping is free running
942 */
943 int timekeeping_is_continuous(void)
944 {
945 unsigned long seq;
946 int ret;
948 do {
949 seq = read_seqbegin(&xtime_lock);
951 ret = clock->is_continuous;
953 } while (read_seqretry(&xtime_lock, seq));
955 return ret;
956 }
958 /*
959 * timekeeping_init - Initializes the clocksource and common timekeeping values
960 */
961 void __init timekeeping_init(void)
962 {
963 unsigned long flags;
965 write_seqlock_irqsave(&xtime_lock, flags);
966 clock = clocksource_get_next();
967 clocksource_calculate_interval(clock, tick_nsec);
968 clock->cycle_last = clocksource_read(clock);
969 ntp_clear();
970 write_sequnlock_irqrestore(&xtime_lock, flags);
971 }
974 static int timekeeping_suspended;
975 /*
976 * timekeeping_resume - Resumes the generic timekeeping subsystem.
977 * @dev: unused
978 *
979 * This is for the generic clocksource timekeeping.
980 * xtime/wall_to_monotonic/jiffies/wall_jiffies/etc are
981 * still managed by arch specific suspend/resume code.
982 */
983 static int timekeeping_resume(struct sys_device *dev)
984 {
985 unsigned long flags;
987 write_seqlock_irqsave(&xtime_lock, flags);
988 /* restart the last cycle value */
989 clock->cycle_last = clocksource_read(clock);
990 clock->error = 0;
991 timekeeping_suspended = 0;
992 write_sequnlock_irqrestore(&xtime_lock, flags);
993 return 0;
994 }
996 static int timekeeping_suspend(struct sys_device *dev, pm_message_t state)
997 {
998 unsigned long flags;
1000 write_seqlock_irqsave(&xtime_lock, flags);
1001 timekeeping_suspended = 1;
1002 write_sequnlock_irqrestore(&xtime_lock, flags);
1003 return 0;
1006 /* sysfs resume/suspend bits for timekeeping */
1007 static struct sysdev_class timekeeping_sysclass = {
1008 .resume = timekeeping_resume,
1009 .suspend = timekeeping_suspend,
1010 set_kset_name("timekeeping"),
1011 };
1013 static struct sys_device device_timer = {
1014 .id = 0,
1015 .cls = &timekeeping_sysclass,
1016 };
1018 static int __init timekeeping_init_device(void)
1020 int error = sysdev_class_register(&timekeeping_sysclass);
1021 if (!error)
1022 error = sysdev_register(&device_timer);
1023 return error;
1026 device_initcall(timekeeping_init_device);
1028 /*
1029 * If the error is already larger, we look ahead even further
1030 * to compensate for late or lost adjustments.
1031 */
1032 static __always_inline int clocksource_bigadjust(s64 error, s64 *interval, s64 *offset)
1034 s64 tick_error, i;
1035 u32 look_ahead, adj;
1036 s32 error2, mult;
1038 /*
1039 * Use the current error value to determine how much to look ahead.
1040 * The larger the error the slower we adjust for it to avoid problems
1041 * with losing too many ticks, otherwise we would overadjust and
1042 * produce an even larger error. The smaller the adjustment the
1043 * faster we try to adjust for it, as lost ticks can do less harm
1044 * here. This is tuned so that an error of about 1 msec is adusted
1045 * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks).
1046 */
1047 error2 = clock->error >> (TICK_LENGTH_SHIFT + 22 - 2 * SHIFT_HZ);
1048 error2 = abs(error2);
1049 for (look_ahead = 0; error2 > 0; look_ahead++)
1050 error2 >>= 2;
1052 /*
1053 * Now calculate the error in (1 << look_ahead) ticks, but first
1054 * remove the single look ahead already included in the error.
1055 */
1056 tick_error = current_tick_length() >> (TICK_LENGTH_SHIFT - clock->shift + 1);
1057 tick_error -= clock->xtime_interval >> 1;
1058 error = ((error - tick_error) >> look_ahead) + tick_error;
1060 /* Finally calculate the adjustment shift value. */
1061 i = *interval;
1062 mult = 1;
1063 if (error < 0) {
1064 error = -error;
1065 *interval = -*interval;
1066 *offset = -*offset;
1067 mult = -1;
1069 for (adj = 0; error > i; adj++)
1070 error >>= 1;
1072 *interval <<= adj;
1073 *offset <<= adj;
1074 return mult << adj;
1077 /*
1078 * Adjust the multiplier to reduce the error value,
1079 * this is optimized for the most common adjustments of -1,0,1,
1080 * for other values we can do a bit more work.
1081 */
1082 static void clocksource_adjust(struct clocksource *clock, s64 offset)
1084 s64 error, interval = clock->cycle_interval;
1085 int adj;
1087 error = clock->error >> (TICK_LENGTH_SHIFT - clock->shift - 1);
1088 if (error > interval) {
1089 error >>= 2;
1090 if (likely(error <= interval))
1091 adj = 1;
1092 else
1093 adj = clocksource_bigadjust(error, &interval, &offset);
1094 } else if (error < -interval) {
1095 error >>= 2;
1096 if (likely(error >= -interval)) {
1097 adj = -1;
1098 interval = -interval;
1099 offset = -offset;
1100 } else
1101 adj = clocksource_bigadjust(error, &interval, &offset);
1102 } else
1103 return;
1105 clock->mult += adj;
1106 clock->xtime_interval += interval;
1107 clock->xtime_nsec -= offset;
1108 clock->error -= (interval - offset) << (TICK_LENGTH_SHIFT - clock->shift);
1111 /*
1112 * update_wall_time - Uses the current clocksource to increment the wall time
1114 * Called from the timer interrupt, must hold a write on xtime_lock.
1115 */
1116 static void update_wall_time(void)
1118 cycle_t offset;
1120 /* Make sure we're fully resumed: */
1121 if (unlikely(timekeeping_suspended))
1122 return;
1124 #ifdef CONFIG_GENERIC_TIME
1125 offset = (clocksource_read(clock) - clock->cycle_last) & clock->mask;
1126 #else
1127 offset = clock->cycle_interval;
1128 #endif
1129 clock->xtime_nsec += (s64)xtime.tv_nsec << clock->shift;
1131 /* normally this loop will run just once, however in the
1132 * case of lost or late ticks, it will accumulate correctly.
1133 */
1134 while (offset >= clock->cycle_interval) {
1135 /* accumulate one interval */
1136 clock->xtime_nsec += clock->xtime_interval;
1137 clock->cycle_last += clock->cycle_interval;
1138 offset -= clock->cycle_interval;
1140 if (clock->xtime_nsec >= (u64)NSEC_PER_SEC << clock->shift) {
1141 clock->xtime_nsec -= (u64)NSEC_PER_SEC << clock->shift;
1142 xtime.tv_sec++;
1143 second_overflow();
1146 /* interpolator bits */
1147 time_interpolator_update(clock->xtime_interval
1148 >> clock->shift);
1149 /* increment the NTP state machine */
1150 update_ntp_one_tick();
1152 /* accumulate error between NTP and clock interval */
1153 clock->error += current_tick_length();
1154 clock->error -= clock->xtime_interval << (TICK_LENGTH_SHIFT - clock->shift);
1157 /* correct the clock when NTP error is too big */
1158 clocksource_adjust(clock, offset);
1160 /* store full nanoseconds into xtime */
1161 xtime.tv_nsec = (s64)clock->xtime_nsec >> clock->shift;
1162 clock->xtime_nsec -= (s64)xtime.tv_nsec << clock->shift;
1164 /* check to see if there is a new clocksource to use */
1165 if (change_clocksource()) {
1166 clock->error = 0;
1167 clock->xtime_nsec = 0;
1168 clocksource_calculate_interval(clock, tick_nsec);
1172 /*
1173 * Called from the timer interrupt handler to charge one tick to the current
1174 * process. user_tick is 1 if the tick is user time, 0 for system.
1175 */
1176 void update_process_times(int user_tick)
1178 struct task_struct *p = current;
1179 int cpu = smp_processor_id();
1181 /* Note: this timer irq context must be accounted for as well. */
1182 if (user_tick)
1183 account_user_time(p, jiffies_to_cputime(1));
1184 else
1185 account_system_time(p, HARDIRQ_OFFSET, jiffies_to_cputime(1));
1186 run_local_timers();
1187 if (rcu_pending(cpu))
1188 rcu_check_callbacks(cpu, user_tick);
1189 scheduler_tick();
1190 run_posix_cpu_timers(p);
1193 /*
1194 * Nr of active tasks - counted in fixed-point numbers
1195 */
1196 static unsigned long count_active_tasks(void)
1198 return nr_active() * FIXED_1;
1201 /*
1202 * Hmm.. Changed this, as the GNU make sources (load.c) seems to
1203 * imply that avenrun[] is the standard name for this kind of thing.
1204 * Nothing else seems to be standardized: the fractional size etc
1205 * all seem to differ on different machines.
1207 * Requires xtime_lock to access.
1208 */
1209 unsigned long avenrun[3];
1211 EXPORT_SYMBOL(avenrun);
1213 /*
1214 * calc_load - given tick count, update the avenrun load estimates.
1215 * This is called while holding a write_lock on xtime_lock.
1216 */
1217 static inline void calc_load(unsigned long ticks)
1219 unsigned long active_tasks; /* fixed-point */
1220 static int count = LOAD_FREQ;
1222 count -= ticks;
1223 if (count < 0) {
1224 count += LOAD_FREQ;
1225 active_tasks = count_active_tasks();
1226 CALC_LOAD(avenrun[0], EXP_1, active_tasks);
1227 CALC_LOAD(avenrun[1], EXP_5, active_tasks);
1228 CALC_LOAD(avenrun[2], EXP_15, active_tasks);
1232 /* jiffies at the most recent update of wall time */
1233 unsigned long wall_jiffies = INITIAL_JIFFIES;
1235 /*
1236 * This read-write spinlock protects us from races in SMP while
1237 * playing with xtime and avenrun.
1238 */
1239 #ifndef ARCH_HAVE_XTIME_LOCK
1240 __cacheline_aligned_in_smp DEFINE_SEQLOCK(xtime_lock);
1242 EXPORT_SYMBOL(xtime_lock);
1243 #endif
1245 /*
1246 * This function runs timers and the timer-tq in bottom half context.
1247 */
1248 static void run_timer_softirq(struct softirq_action *h)
1250 tvec_base_t *base = __get_cpu_var(tvec_bases);
1252 hrtimer_run_queues();
1253 if (time_after_eq(jiffies, base->timer_jiffies))
1254 __run_timers(base);
1257 /*
1258 * Called by the local, per-CPU timer interrupt on SMP.
1259 */
1260 void run_local_timers(void)
1262 raise_softirq(TIMER_SOFTIRQ);
1263 softlockup_tick();
1266 /*
1267 * Called by the timer interrupt. xtime_lock must already be taken
1268 * by the timer IRQ!
1269 */
1270 static inline void update_times(void)
1272 unsigned long ticks;
1274 ticks = jiffies - wall_jiffies;
1275 wall_jiffies += ticks;
1276 update_wall_time();
1277 calc_load(ticks);
1280 /*
1281 * The 64-bit jiffies value is not atomic - you MUST NOT read it
1282 * without sampling the sequence number in xtime_lock.
1283 * jiffies is defined in the linker script...
1284 */
1286 void do_timer(struct pt_regs *regs)
1288 jiffies_64++;
1289 /* prevent loading jiffies before storing new jiffies_64 value. */
1290 barrier();
1291 update_times();
1294 #ifdef __ARCH_WANT_SYS_ALARM
1296 /*
1297 * For backwards compatibility? This can be done in libc so Alpha
1298 * and all newer ports shouldn't need it.
1299 */
1300 asmlinkage unsigned long sys_alarm(unsigned int seconds)
1302 return alarm_setitimer(seconds);
1305 #endif
1307 #ifndef __alpha__
1309 /*
1310 * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this
1311 * should be moved into arch/i386 instead?
1312 */
1314 /**
1315 * sys_getpid - return the thread group id of the current process
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.
1321 * This is SMP safe as current->tgid does not change.
1322 */
1323 asmlinkage long sys_getpid(void)
1325 return current->tgid;
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 */
1334 asmlinkage long sys_getppid(void)
1336 int pid;
1338 rcu_read_lock();
1339 pid = rcu_dereference(current->real_parent)->tgid;
1340 rcu_read_unlock();
1342 return pid;
1345 asmlinkage long sys_getuid(void)
1347 /* Only we change this so SMP safe */
1348 return current->uid;
1351 asmlinkage long sys_geteuid(void)
1353 /* Only we change this so SMP safe */
1354 return current->euid;
1357 asmlinkage long sys_getgid(void)
1359 /* Only we change this so SMP safe */
1360 return current->gid;
1363 asmlinkage long sys_getegid(void)
1365 /* Only we change this so SMP safe */
1366 return current->egid;
1369 #endif
1371 static void process_timeout(unsigned long __data)
1373 wake_up_process((struct task_struct *)__data);
1376 /**
1377 * schedule_timeout - sleep until timeout
1378 * @timeout: timeout value in jiffies
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()).
1384 * You can set the task state as follows -
1386 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1387 * pass before the routine returns. The routine will return 0
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
1393 * The current task state is guaranteed to be TASK_RUNNING when this
1394 * routine returns.
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.
1400 * In all cases the return value is guaranteed to be non-negative.
1401 */
1402 fastcall signed long __sched schedule_timeout(signed long timeout)
1404 struct timer_list timer;
1405 unsigned long expire;
1407 switch (timeout)
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)
1429 printk(KERN_ERR "schedule_timeout: wrong timeout "
1430 "value %lx from %p\n", timeout,
1431 __builtin_return_address(0));
1432 current->state = TASK_RUNNING;
1433 goto out;
1437 expire = timeout + jiffies;
1439 setup_timer(&timer, process_timeout, (unsigned long)current);
1440 __mod_timer(&timer, expire);
1441 schedule();
1442 del_singleshot_timer_sync(&timer);
1444 timeout = expire - jiffies;
1446 out:
1447 return timeout < 0 ? 0 : timeout;
1449 EXPORT_SYMBOL(schedule_timeout);
1451 /*
1452 * We can use __set_current_state() here because schedule_timeout() calls
1453 * schedule() unconditionally.
1454 */
1455 signed long __sched schedule_timeout_interruptible(signed long timeout)
1457 __set_current_state(TASK_INTERRUPTIBLE);
1458 return schedule_timeout(timeout);
1460 EXPORT_SYMBOL(schedule_timeout_interruptible);
1462 signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1464 __set_current_state(TASK_UNINTERRUPTIBLE);
1465 return schedule_timeout(timeout);
1467 EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1469 /* Thread ID - the internal kernel "pid" */
1470 asmlinkage long sys_gettid(void)
1472 return current->pid;
1475 /*
1476 * sys_sysinfo - fill in sysinfo struct
1477 */
1478 asmlinkage long sys_sysinfo(struct sysinfo __user *info)
1480 struct sysinfo val;
1481 unsigned long mem_total, sav_total;
1482 unsigned int mem_unit, bitcount;
1483 unsigned long seq;
1485 memset((char *)&val, 0, sizeof(struct sysinfo));
1487 do {
1488 struct timespec tp;
1489 seq = read_seqbegin(&xtime_lock);
1491 /*
1492 * This is annoying. The below is the same thing
1493 * posix_get_clock_monotonic() does, but it wants to
1494 * take the lock which we want to cover the loads stuff
1495 * too.
1496 */
1498 getnstimeofday(&tp);
1499 tp.tv_sec += wall_to_monotonic.tv_sec;
1500 tp.tv_nsec += wall_to_monotonic.tv_nsec;
1501 if (tp.tv_nsec - NSEC_PER_SEC >= 0) {
1502 tp.tv_nsec = tp.tv_nsec - NSEC_PER_SEC;
1503 tp.tv_sec++;
1505 val.uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
1507 val.loads[0] = avenrun[0] << (SI_LOAD_SHIFT - FSHIFT);
1508 val.loads[1] = avenrun[1] << (SI_LOAD_SHIFT - FSHIFT);
1509 val.loads[2] = avenrun[2] << (SI_LOAD_SHIFT - FSHIFT);
1511 val.procs = nr_threads;
1512 } while (read_seqretry(&xtime_lock, seq));
1514 si_meminfo(&val);
1515 si_swapinfo(&val);
1517 /*
1518 * If the sum of all the available memory (i.e. ram + swap)
1519 * is less than can be stored in a 32 bit unsigned long then
1520 * we can be binary compatible with 2.2.x kernels. If not,
1521 * well, in that case 2.2.x was broken anyways...
1523 * -Erik Andersen <andersee@debian.org>
1524 */
1526 mem_total = val.totalram + val.totalswap;
1527 if (mem_total < val.totalram || mem_total < val.totalswap)
1528 goto out;
1529 bitcount = 0;
1530 mem_unit = val.mem_unit;
1531 while (mem_unit > 1) {
1532 bitcount++;
1533 mem_unit >>= 1;
1534 sav_total = mem_total;
1535 mem_total <<= 1;
1536 if (mem_total < sav_total)
1537 goto out;
1540 /*
1541 * If mem_total did not overflow, multiply all memory values by
1542 * val.mem_unit and set it to 1. This leaves things compatible
1543 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1544 * kernels...
1545 */
1547 val.mem_unit = 1;
1548 val.totalram <<= bitcount;
1549 val.freeram <<= bitcount;
1550 val.sharedram <<= bitcount;
1551 val.bufferram <<= bitcount;
1552 val.totalswap <<= bitcount;
1553 val.freeswap <<= bitcount;
1554 val.totalhigh <<= bitcount;
1555 val.freehigh <<= bitcount;
1557 out:
1558 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
1559 return -EFAULT;
1561 return 0;
1564 /*
1565 * lockdep: we want to track each per-CPU base as a separate lock-class,
1566 * but timer-bases are kmalloc()-ed, so we need to attach separate
1567 * keys to them:
1568 */
1569 static struct lock_class_key base_lock_keys[NR_CPUS];
1571 static int __devinit init_timers_cpu(int cpu)
1573 int j;
1574 tvec_base_t *base;
1575 static char __devinitdata tvec_base_done[NR_CPUS];
1577 if (!tvec_base_done[cpu]) {
1578 static char boot_done;
1580 if (boot_done) {
1581 /*
1582 * The APs use this path later in boot
1583 */
1584 base = kmalloc_node(sizeof(*base), GFP_KERNEL,
1585 cpu_to_node(cpu));
1586 if (!base)
1587 return -ENOMEM;
1588 memset(base, 0, sizeof(*base));
1589 per_cpu(tvec_bases, cpu) = base;
1590 } else {
1591 /*
1592 * This is for the boot CPU - we use compile-time
1593 * static initialisation because per-cpu memory isn't
1594 * ready yet and because the memory allocators are not
1595 * initialised either.
1596 */
1597 boot_done = 1;
1598 base = &boot_tvec_bases;
1600 tvec_base_done[cpu] = 1;
1601 } else {
1602 base = per_cpu(tvec_bases, cpu);
1605 spin_lock_init(&base->lock);
1606 lockdep_set_class(&base->lock, base_lock_keys + cpu);
1608 for (j = 0; j < TVN_SIZE; j++) {
1609 INIT_LIST_HEAD(base->tv5.vec + j);
1610 INIT_LIST_HEAD(base->tv4.vec + j);
1611 INIT_LIST_HEAD(base->tv3.vec + j);
1612 INIT_LIST_HEAD(base->tv2.vec + j);
1614 for (j = 0; j < TVR_SIZE; j++)
1615 INIT_LIST_HEAD(base->tv1.vec + j);
1617 base->timer_jiffies = jiffies;
1618 return 0;
1621 #ifdef CONFIG_HOTPLUG_CPU
1622 static void migrate_timer_list(tvec_base_t *new_base, struct list_head *head)
1624 struct timer_list *timer;
1626 while (!list_empty(head)) {
1627 timer = list_entry(head->next, struct timer_list, entry);
1628 detach_timer(timer, 0);
1629 timer->base = new_base;
1630 internal_add_timer(new_base, timer);
1634 static void __devinit migrate_timers(int cpu)
1636 tvec_base_t *old_base;
1637 tvec_base_t *new_base;
1638 int i;
1640 BUG_ON(cpu_online(cpu));
1641 old_base = per_cpu(tvec_bases, cpu);
1642 new_base = get_cpu_var(tvec_bases);
1644 local_irq_disable();
1645 spin_lock(&new_base->lock);
1646 spin_lock(&old_base->lock);
1648 BUG_ON(old_base->running_timer);
1650 for (i = 0; i < TVR_SIZE; i++)
1651 migrate_timer_list(new_base, old_base->tv1.vec + i);
1652 for (i = 0; i < TVN_SIZE; i++) {
1653 migrate_timer_list(new_base, old_base->tv2.vec + i);
1654 migrate_timer_list(new_base, old_base->tv3.vec + i);
1655 migrate_timer_list(new_base, old_base->tv4.vec + i);
1656 migrate_timer_list(new_base, old_base->tv5.vec + i);
1659 spin_unlock(&old_base->lock);
1660 spin_unlock(&new_base->lock);
1661 local_irq_enable();
1662 put_cpu_var(tvec_bases);
1664 #endif /* CONFIG_HOTPLUG_CPU */
1666 static int __cpuinit timer_cpu_notify(struct notifier_block *self,
1667 unsigned long action, void *hcpu)
1669 long cpu = (long)hcpu;
1670 switch(action) {
1671 case CPU_UP_PREPARE:
1672 if (init_timers_cpu(cpu) < 0)
1673 return NOTIFY_BAD;
1674 break;
1675 #ifdef CONFIG_HOTPLUG_CPU
1676 case CPU_DEAD:
1677 migrate_timers(cpu);
1678 break;
1679 #endif
1680 default:
1681 break;
1683 return NOTIFY_OK;
1686 static struct notifier_block __cpuinitdata timers_nb = {
1687 .notifier_call = timer_cpu_notify,
1688 };
1691 void __init init_timers(void)
1693 timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE,
1694 (void *)(long)smp_processor_id());
1695 register_cpu_notifier(&timers_nb);
1696 open_softirq(TIMER_SOFTIRQ, run_timer_softirq, NULL);
1699 #ifdef CONFIG_TIME_INTERPOLATION
1701 struct time_interpolator *time_interpolator __read_mostly;
1702 static struct time_interpolator *time_interpolator_list __read_mostly;
1703 static DEFINE_SPINLOCK(time_interpolator_lock);
1705 static inline u64 time_interpolator_get_cycles(unsigned int src)
1707 unsigned long (*x)(void);
1709 switch (src)
1711 case TIME_SOURCE_FUNCTION:
1712 x = time_interpolator->addr;
1713 return x();
1715 case TIME_SOURCE_MMIO64 :
1716 return readq_relaxed((void __iomem *)time_interpolator->addr);
1718 case TIME_SOURCE_MMIO32 :
1719 return readl_relaxed((void __iomem *)time_interpolator->addr);
1721 default: return get_cycles();
1725 static inline u64 time_interpolator_get_counter(int writelock)
1727 unsigned int src = time_interpolator->source;
1729 if (time_interpolator->jitter)
1731 u64 lcycle;
1732 u64 now;
1734 do {
1735 lcycle = time_interpolator->last_cycle;
1736 now = time_interpolator_get_cycles(src);
1737 if (lcycle && time_after(lcycle, now))
1738 return lcycle;
1740 /* When holding the xtime write lock, there's no need
1741 * to add the overhead of the cmpxchg. Readers are
1742 * force to retry until the write lock is released.
1743 */
1744 if (writelock) {
1745 time_interpolator->last_cycle = now;
1746 return now;
1748 /* Keep track of the last timer value returned. The use of cmpxchg here
1749 * will cause contention in an SMP environment.
1750 */
1751 } while (unlikely(cmpxchg(&time_interpolator->last_cycle, lcycle, now) != lcycle));
1752 return now;
1754 else
1755 return time_interpolator_get_cycles(src);
1758 void time_interpolator_reset(void)
1760 time_interpolator->offset = 0;
1761 time_interpolator->last_counter = time_interpolator_get_counter(1);
1764 #define GET_TI_NSECS(count,i) (((((count) - i->last_counter) & (i)->mask) * (i)->nsec_per_cyc) >> (i)->shift)
1766 unsigned long time_interpolator_get_offset(void)
1768 /* If we do not have a time interpolator set up then just return zero */
1769 if (!time_interpolator)
1770 return 0;
1772 return time_interpolator->offset +
1773 GET_TI_NSECS(time_interpolator_get_counter(0), time_interpolator);
1776 #define INTERPOLATOR_ADJUST 65536
1777 #define INTERPOLATOR_MAX_SKIP 10*INTERPOLATOR_ADJUST
1779 static void time_interpolator_update(long delta_nsec)
1781 u64 counter;
1782 unsigned long offset;
1784 /* If there is no time interpolator set up then do nothing */
1785 if (!time_interpolator)
1786 return;
1788 /*
1789 * The interpolator compensates for late ticks by accumulating the late
1790 * time in time_interpolator->offset. A tick earlier than expected will
1791 * lead to a reset of the offset and a corresponding jump of the clock
1792 * forward. Again this only works if the interpolator clock is running
1793 * slightly slower than the regular clock and the tuning logic insures
1794 * that.
1795 */
1797 counter = time_interpolator_get_counter(1);
1798 offset = time_interpolator->offset +
1799 GET_TI_NSECS(counter, time_interpolator);
1801 if (delta_nsec < 0 || (unsigned long) delta_nsec < offset)
1802 time_interpolator->offset = offset - delta_nsec;
1803 else {
1804 time_interpolator->skips++;
1805 time_interpolator->ns_skipped += delta_nsec - offset;
1806 time_interpolator->offset = 0;
1808 time_interpolator->last_counter = counter;
1810 /* Tuning logic for time interpolator invoked every minute or so.
1811 * Decrease interpolator clock speed if no skips occurred and an offset is carried.
1812 * Increase interpolator clock speed if we skip too much time.
1813 */
1814 if (jiffies % INTERPOLATOR_ADJUST == 0)
1816 if (time_interpolator->skips == 0 && time_interpolator->offset > tick_nsec)
1817 time_interpolator->nsec_per_cyc--;
1818 if (time_interpolator->ns_skipped > INTERPOLATOR_MAX_SKIP && time_interpolator->offset == 0)
1819 time_interpolator->nsec_per_cyc++;
1820 time_interpolator->skips = 0;
1821 time_interpolator->ns_skipped = 0;
1825 static inline int
1826 is_better_time_interpolator(struct time_interpolator *new)
1828 if (!time_interpolator)
1829 return 1;
1830 return new->frequency > 2*time_interpolator->frequency ||
1831 (unsigned long)new->drift < (unsigned long)time_interpolator->drift;
1834 void
1835 register_time_interpolator(struct time_interpolator *ti)
1837 unsigned long flags;
1839 /* Sanity check */
1840 BUG_ON(ti->frequency == 0 || ti->mask == 0);
1842 ti->nsec_per_cyc = ((u64)NSEC_PER_SEC << ti->shift) / ti->frequency;
1843 spin_lock(&time_interpolator_lock);
1844 write_seqlock_irqsave(&xtime_lock, flags);
1845 if (is_better_time_interpolator(ti)) {
1846 time_interpolator = ti;
1847 time_interpolator_reset();
1849 write_sequnlock_irqrestore(&xtime_lock, flags);
1851 ti->next = time_interpolator_list;
1852 time_interpolator_list = ti;
1853 spin_unlock(&time_interpolator_lock);
1856 void
1857 unregister_time_interpolator(struct time_interpolator *ti)
1859 struct time_interpolator *curr, **prev;
1860 unsigned long flags;
1862 spin_lock(&time_interpolator_lock);
1863 prev = &time_interpolator_list;
1864 for (curr = *prev; curr; curr = curr->next) {
1865 if (curr == ti) {
1866 *prev = curr->next;
1867 break;
1869 prev = &curr->next;
1872 write_seqlock_irqsave(&xtime_lock, flags);
1873 if (ti == time_interpolator) {
1874 /* we lost the best time-interpolator: */
1875 time_interpolator = NULL;
1876 /* find the next-best interpolator */
1877 for (curr = time_interpolator_list; curr; curr = curr->next)
1878 if (is_better_time_interpolator(curr))
1879 time_interpolator = curr;
1880 time_interpolator_reset();
1882 write_sequnlock_irqrestore(&xtime_lock, flags);
1883 spin_unlock(&time_interpolator_lock);
1885 #endif /* CONFIG_TIME_INTERPOLATION */
1887 /**
1888 * msleep - sleep safely even with waitqueue interruptions
1889 * @msecs: Time in milliseconds to sleep for
1890 */
1891 void msleep(unsigned int msecs)
1893 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1895 while (timeout)
1896 timeout = schedule_timeout_uninterruptible(timeout);
1899 EXPORT_SYMBOL(msleep);
1901 /**
1902 * msleep_interruptible - sleep waiting for signals
1903 * @msecs: Time in milliseconds to sleep for
1904 */
1905 unsigned long msleep_interruptible(unsigned int msecs)
1907 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1909 while (timeout && !signal_pending(current))
1910 timeout = schedule_timeout_interruptible(timeout);
1911 return jiffies_to_msecs(timeout);
1914 EXPORT_SYMBOL(msleep_interruptible);