Initial import from FreeBSD RELENG_4:
[dragonfly.git] / sys / kern / kern_clock.c
CommitLineData
984263bc
MD
1/*-
2 * Copyright (c) 1997, 1998 Poul-Henning Kamp <phk@FreeBSD.org>
3 * Copyright (c) 1982, 1986, 1991, 1993
4 * The Regents of the University of California. All rights reserved.
5 * (c) UNIX System Laboratories, Inc.
6 * All or some portions of this file are derived from material licensed
7 * to the University of California by American Telephone and Telegraph
8 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
9 * the permission of UNIX System Laboratories, Inc.
10 *
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
13 * are met:
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. All advertising materials mentioning features or use of this software
20 * must display the following acknowledgement:
21 * This product includes software developed by the University of
22 * California, Berkeley and its contributors.
23 * 4. Neither the name of the University nor the names of its contributors
24 * may be used to endorse or promote products derived from this software
25 * without specific prior written permission.
26 *
27 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
28 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
31 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
32 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
33 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
35 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
36 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37 * SUCH DAMAGE.
38 *
39 * @(#)kern_clock.c 8.5 (Berkeley) 1/21/94
40 * $FreeBSD: src/sys/kern/kern_clock.c,v 1.105.2.10 2002/10/17 13:19:40 maxim Exp $
41 */
42
43#include "opt_ntp.h"
44
45#include <sys/param.h>
46#include <sys/systm.h>
47#include <sys/dkstat.h>
48#include <sys/callout.h>
49#include <sys/kernel.h>
50#include <sys/proc.h>
51#include <sys/malloc.h>
52#include <sys/resourcevar.h>
53#include <sys/signalvar.h>
54#include <sys/timex.h>
55#include <sys/timepps.h>
56#include <vm/vm.h>
57#include <sys/lock.h>
58#include <vm/pmap.h>
59#include <vm/vm_map.h>
60#include <sys/sysctl.h>
61
62#include <machine/cpu.h>
63#include <machine/limits.h>
64#include <machine/smp.h>
65
66#ifdef GPROF
67#include <sys/gmon.h>
68#endif
69
70#ifdef DEVICE_POLLING
71extern void init_device_poll(void);
72extern void hardclock_device_poll(void);
73#endif /* DEVICE_POLLING */
74
75/*
76 * Number of timecounters used to implement stable storage
77 */
78#ifndef NTIMECOUNTER
79#define NTIMECOUNTER 5
80#endif
81
82static MALLOC_DEFINE(M_TIMECOUNTER, "timecounter",
83 "Timecounter stable storage");
84
85static void initclocks __P((void *dummy));
86SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL)
87
88static void tco_forward __P((int force));
89static void tco_setscales __P((struct timecounter *tc));
90static __inline unsigned tco_delta __P((struct timecounter *tc));
91
92/* Some of these don't belong here, but it's easiest to concentrate them. */
93long cp_time[CPUSTATES];
94
95SYSCTL_OPAQUE(_kern, OID_AUTO, cp_time, CTLFLAG_RD, &cp_time, sizeof(cp_time),
96 "LU", "CPU time statistics");
97
98long tk_cancc;
99long tk_nin;
100long tk_nout;
101long tk_rawcc;
102
103time_t time_second;
104
105struct timeval boottime;
106SYSCTL_STRUCT(_kern, KERN_BOOTTIME, boottime, CTLFLAG_RD,
107 &boottime, timeval, "System boottime");
108
109/*
110 * Which update policy to use.
111 * 0 - every tick, bad hardware may fail with "calcru negative..."
112 * 1 - more resistent to the above hardware, but less efficient.
113 */
114static int tco_method;
115
116/*
117 * Implement a dummy timecounter which we can use until we get a real one
118 * in the air. This allows the console and other early stuff to use
119 * timeservices.
120 */
121
122static unsigned
123dummy_get_timecount(struct timecounter *tc)
124{
125 static unsigned now;
126 return (++now);
127}
128
129static struct timecounter dummy_timecounter = {
130 dummy_get_timecount,
131 0,
132 ~0u,
133 1000000,
134 "dummy"
135};
136
137struct timecounter *timecounter = &dummy_timecounter;
138
139/*
140 * Clock handling routines.
141 *
142 * This code is written to operate with two timers that run independently of
143 * each other.
144 *
145 * The main timer, running hz times per second, is used to trigger interval
146 * timers, timeouts and rescheduling as needed.
147 *
148 * The second timer handles kernel and user profiling,
149 * and does resource use estimation. If the second timer is programmable,
150 * it is randomized to avoid aliasing between the two clocks. For example,
151 * the randomization prevents an adversary from always giving up the cpu
152 * just before its quantum expires. Otherwise, it would never accumulate
153 * cpu ticks. The mean frequency of the second timer is stathz.
154 *
155 * If no second timer exists, stathz will be zero; in this case we drive
156 * profiling and statistics off the main clock. This WILL NOT be accurate;
157 * do not do it unless absolutely necessary.
158 *
159 * The statistics clock may (or may not) be run at a higher rate while
160 * profiling. This profile clock runs at profhz. We require that profhz
161 * be an integral multiple of stathz.
162 *
163 * If the statistics clock is running fast, it must be divided by the ratio
164 * profhz/stathz for statistics. (For profiling, every tick counts.)
165 *
166 * Time-of-day is maintained using a "timecounter", which may or may
167 * not be related to the hardware generating the above mentioned
168 * interrupts.
169 */
170
171int stathz;
172int profhz;
173static int profprocs;
174int ticks;
175static int psdiv, pscnt; /* prof => stat divider */
176int psratio; /* ratio: prof / stat */
177
178/*
179 * Initialize clock frequencies and start both clocks running.
180 */
181/* ARGSUSED*/
182static void
183initclocks(dummy)
184 void *dummy;
185{
186 register int i;
187
188 /*
189 * Set divisors to 1 (normal case) and let the machine-specific
190 * code do its bit.
191 */
192 psdiv = pscnt = 1;
193 cpu_initclocks();
194
195#ifdef DEVICE_POLLING
196 init_device_poll();
197#endif
198
199 /*
200 * Compute profhz/stathz, and fix profhz if needed.
201 */
202 i = stathz ? stathz : hz;
203 if (profhz == 0)
204 profhz = i;
205 psratio = profhz / i;
206}
207
208/*
209 * The real-time timer, interrupting hz times per second.
210 */
211void
212hardclock(frame)
213 register struct clockframe *frame;
214{
215 register struct proc *p;
216
217 p = curproc;
218 if (p) {
219 register struct pstats *pstats;
220
221 /*
222 * Run current process's virtual and profile time, as needed.
223 */
224 pstats = p->p_stats;
225 if (CLKF_USERMODE(frame) &&
226 timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) &&
227 itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0)
228 psignal(p, SIGVTALRM);
229 if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value) &&
230 itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0)
231 psignal(p, SIGPROF);
232 }
233
234#if defined(SMP) && defined(BETTER_CLOCK)
235 forward_hardclock(pscnt);
236#endif
237
238 /*
239 * If no separate statistics clock is available, run it from here.
240 */
241 if (stathz == 0)
242 statclock(frame);
243
244 tco_forward(0);
245 ticks++;
246
247#ifdef DEVICE_POLLING
248 hardclock_device_poll(); /* this is very short and quick */
249#endif /* DEVICE_POLLING */
250
251 /*
252 * Process callouts at a very low cpu priority, so we don't keep the
253 * relatively high clock interrupt priority any longer than necessary.
254 */
255 if (TAILQ_FIRST(&callwheel[ticks & callwheelmask]) != NULL) {
256 if (CLKF_BASEPRI(frame)) {
257 /*
258 * Save the overhead of a software interrupt;
259 * it will happen as soon as we return, so do it now.
260 */
261 (void)splsoftclock();
262 softclock();
263 } else
264 setsoftclock();
265 } else if (softticks + 1 == ticks)
266 ++softticks;
267}
268
269/*
270 * Compute number of ticks in the specified amount of time.
271 */
272int
273tvtohz(tv)
274 struct timeval *tv;
275{
276 register unsigned long ticks;
277 register long sec, usec;
278
279 /*
280 * If the number of usecs in the whole seconds part of the time
281 * difference fits in a long, then the total number of usecs will
282 * fit in an unsigned long. Compute the total and convert it to
283 * ticks, rounding up and adding 1 to allow for the current tick
284 * to expire. Rounding also depends on unsigned long arithmetic
285 * to avoid overflow.
286 *
287 * Otherwise, if the number of ticks in the whole seconds part of
288 * the time difference fits in a long, then convert the parts to
289 * ticks separately and add, using similar rounding methods and
290 * overflow avoidance. This method would work in the previous
291 * case but it is slightly slower and assumes that hz is integral.
292 *
293 * Otherwise, round the time difference down to the maximum
294 * representable value.
295 *
296 * If ints have 32 bits, then the maximum value for any timeout in
297 * 10ms ticks is 248 days.
298 */
299 sec = tv->tv_sec;
300 usec = tv->tv_usec;
301 if (usec < 0) {
302 sec--;
303 usec += 1000000;
304 }
305 if (sec < 0) {
306#ifdef DIAGNOSTIC
307 if (usec > 0) {
308 sec++;
309 usec -= 1000000;
310 }
311 printf("tvotohz: negative time difference %ld sec %ld usec\n",
312 sec, usec);
313#endif
314 ticks = 1;
315 } else if (sec <= LONG_MAX / 1000000)
316 ticks = (sec * 1000000 + (unsigned long)usec + (tick - 1))
317 / tick + 1;
318 else if (sec <= LONG_MAX / hz)
319 ticks = sec * hz
320 + ((unsigned long)usec + (tick - 1)) / tick + 1;
321 else
322 ticks = LONG_MAX;
323 if (ticks > INT_MAX)
324 ticks = INT_MAX;
325 return ((int)ticks);
326}
327
328/*
329 * Start profiling on a process.
330 *
331 * Kernel profiling passes proc0 which never exits and hence
332 * keeps the profile clock running constantly.
333 */
334void
335startprofclock(p)
336 register struct proc *p;
337{
338 int s;
339
340 if ((p->p_flag & P_PROFIL) == 0) {
341 p->p_flag |= P_PROFIL;
342 if (++profprocs == 1 && stathz != 0) {
343 s = splstatclock();
344 psdiv = pscnt = psratio;
345 setstatclockrate(profhz);
346 splx(s);
347 }
348 }
349}
350
351/*
352 * Stop profiling on a process.
353 */
354void
355stopprofclock(p)
356 register struct proc *p;
357{
358 int s;
359
360 if (p->p_flag & P_PROFIL) {
361 p->p_flag &= ~P_PROFIL;
362 if (--profprocs == 0 && stathz != 0) {
363 s = splstatclock();
364 psdiv = pscnt = 1;
365 setstatclockrate(stathz);
366 splx(s);
367 }
368 }
369}
370
371/*
372 * Statistics clock. Grab profile sample, and if divider reaches 0,
373 * do process and kernel statistics. Most of the statistics are only
374 * used by user-level statistics programs. The main exceptions are
375 * p->p_uticks, p->p_sticks, p->p_iticks, and p->p_estcpu.
376 */
377void
378statclock(frame)
379 register struct clockframe *frame;
380{
381#ifdef GPROF
382 register struct gmonparam *g;
383 int i;
384#endif
385 register struct proc *p;
386 struct pstats *pstats;
387 long rss;
388 struct rusage *ru;
389 struct vmspace *vm;
390
391 if (curproc != NULL && CLKF_USERMODE(frame)) {
392 /*
393 * Came from user mode; CPU was in user state.
394 * If this process is being profiled, record the tick.
395 */
396 p = curproc;
397 if (p->p_flag & P_PROFIL)
398 addupc_intr(p, CLKF_PC(frame), 1);
399#if defined(SMP) && defined(BETTER_CLOCK)
400 if (stathz != 0)
401 forward_statclock(pscnt);
402#endif
403 if (--pscnt > 0)
404 return;
405 /*
406 * Charge the time as appropriate.
407 */
408 p->p_uticks++;
409 if (p->p_nice > NZERO)
410 cp_time[CP_NICE]++;
411 else
412 cp_time[CP_USER]++;
413 } else {
414#ifdef GPROF
415 /*
416 * Kernel statistics are just like addupc_intr, only easier.
417 */
418 g = &_gmonparam;
419 if (g->state == GMON_PROF_ON) {
420 i = CLKF_PC(frame) - g->lowpc;
421 if (i < g->textsize) {
422 i /= HISTFRACTION * sizeof(*g->kcount);
423 g->kcount[i]++;
424 }
425 }
426#endif
427#if defined(SMP) && defined(BETTER_CLOCK)
428 if (stathz != 0)
429 forward_statclock(pscnt);
430#endif
431 if (--pscnt > 0)
432 return;
433 /*
434 * Came from kernel mode, so we were:
435 * - handling an interrupt,
436 * - doing syscall or trap work on behalf of the current
437 * user process, or
438 * - spinning in the idle loop.
439 * Whichever it is, charge the time as appropriate.
440 * Note that we charge interrupts to the current process,
441 * regardless of whether they are ``for'' that process,
442 * so that we know how much of its real time was spent
443 * in ``non-process'' (i.e., interrupt) work.
444 */
445 p = curproc;
446 if (CLKF_INTR(frame)) {
447 if (p != NULL)
448 p->p_iticks++;
449 cp_time[CP_INTR]++;
450 } else if (p != NULL) {
451 p->p_sticks++;
452 cp_time[CP_SYS]++;
453 } else
454 cp_time[CP_IDLE]++;
455 }
456 pscnt = psdiv;
457
458 if (p != NULL) {
459 schedclock(p);
460
461 /* Update resource usage integrals and maximums. */
462 if ((pstats = p->p_stats) != NULL &&
463 (ru = &pstats->p_ru) != NULL &&
464 (vm = p->p_vmspace) != NULL) {
465 ru->ru_ixrss += pgtok(vm->vm_tsize);
466 ru->ru_idrss += pgtok(vm->vm_dsize);
467 ru->ru_isrss += pgtok(vm->vm_ssize);
468 rss = pgtok(vmspace_resident_count(vm));
469 if (ru->ru_maxrss < rss)
470 ru->ru_maxrss = rss;
471 }
472 }
473}
474
475/*
476 * Return information about system clocks.
477 */
478static int
479sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS)
480{
481 struct clockinfo clkinfo;
482 /*
483 * Construct clockinfo structure.
484 */
485 clkinfo.hz = hz;
486 clkinfo.tick = tick;
487 clkinfo.tickadj = tickadj;
488 clkinfo.profhz = profhz;
489 clkinfo.stathz = stathz ? stathz : hz;
490 return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req));
491}
492
493SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate, CTLTYPE_STRUCT|CTLFLAG_RD,
494 0, 0, sysctl_kern_clockrate, "S,clockinfo","");
495
496static __inline unsigned
497tco_delta(struct timecounter *tc)
498{
499
500 return ((tc->tc_get_timecount(tc) - tc->tc_offset_count) &
501 tc->tc_counter_mask);
502}
503
504/*
505 * We have eight functions for looking at the clock, four for
506 * microseconds and four for nanoseconds. For each there is fast
507 * but less precise version "get{nano|micro}[up]time" which will
508 * return a time which is up to 1/HZ previous to the call, whereas
509 * the raw version "{nano|micro}[up]time" will return a timestamp
510 * which is as precise as possible. The "up" variants return the
511 * time relative to system boot, these are well suited for time
512 * interval measurements.
513 */
514
515void
516getmicrotime(struct timeval *tvp)
517{
518 struct timecounter *tc;
519
520 if (!tco_method) {
521 tc = timecounter;
522 *tvp = tc->tc_microtime;
523 } else {
524 microtime(tvp);
525 }
526}
527
528void
529getnanotime(struct timespec *tsp)
530{
531 struct timecounter *tc;
532
533 if (!tco_method) {
534 tc = timecounter;
535 *tsp = tc->tc_nanotime;
536 } else {
537 nanotime(tsp);
538 }
539}
540
541void
542microtime(struct timeval *tv)
543{
544 struct timecounter *tc;
545
546 tc = timecounter;
547 tv->tv_sec = tc->tc_offset_sec;
548 tv->tv_usec = tc->tc_offset_micro;
549 tv->tv_usec += ((u_int64_t)tco_delta(tc) * tc->tc_scale_micro) >> 32;
550 tv->tv_usec += boottime.tv_usec;
551 tv->tv_sec += boottime.tv_sec;
552 while (tv->tv_usec < 0) {
553 tv->tv_usec += 1000000;
554 if (tv->tv_sec > 0)
555 tv->tv_sec--;
556 }
557 while (tv->tv_usec >= 1000000) {
558 tv->tv_usec -= 1000000;
559 tv->tv_sec++;
560 }
561}
562
563void
564nanotime(struct timespec *ts)
565{
566 unsigned count;
567 u_int64_t delta;
568 struct timecounter *tc;
569
570 tc = timecounter;
571 ts->tv_sec = tc->tc_offset_sec;
572 count = tco_delta(tc);
573 delta = tc->tc_offset_nano;
574 delta += ((u_int64_t)count * tc->tc_scale_nano_f);
575 delta >>= 32;
576 delta += ((u_int64_t)count * tc->tc_scale_nano_i);
577 delta += boottime.tv_usec * 1000;
578 ts->tv_sec += boottime.tv_sec;
579 while (delta < 0) {
580 delta += 1000000000;
581 if (ts->tv_sec > 0)
582 ts->tv_sec--;
583 }
584 while (delta >= 1000000000) {
585 delta -= 1000000000;
586 ts->tv_sec++;
587 }
588 ts->tv_nsec = delta;
589}
590
591void
592getmicrouptime(struct timeval *tvp)
593{
594 struct timecounter *tc;
595
596 if (!tco_method) {
597 tc = timecounter;
598 tvp->tv_sec = tc->tc_offset_sec;
599 tvp->tv_usec = tc->tc_offset_micro;
600 } else {
601 microuptime(tvp);
602 }
603}
604
605void
606getnanouptime(struct timespec *tsp)
607{
608 struct timecounter *tc;
609
610 if (!tco_method) {
611 tc = timecounter;
612 tsp->tv_sec = tc->tc_offset_sec;
613 tsp->tv_nsec = tc->tc_offset_nano >> 32;
614 } else {
615 nanouptime(tsp);
616 }
617}
618
619void
620microuptime(struct timeval *tv)
621{
622 struct timecounter *tc;
623
624 tc = timecounter;
625 tv->tv_sec = tc->tc_offset_sec;
626 tv->tv_usec = tc->tc_offset_micro;
627 tv->tv_usec += ((u_int64_t)tco_delta(tc) * tc->tc_scale_micro) >> 32;
628 while (tv->tv_usec < 0) {
629 tv->tv_usec += 1000000;
630 if (tv->tv_sec > 0)
631 tv->tv_sec--;
632 }
633 while (tv->tv_usec >= 1000000) {
634 tv->tv_usec -= 1000000;
635 tv->tv_sec++;
636 }
637}
638
639void
640nanouptime(struct timespec *ts)
641{
642 unsigned count;
643 u_int64_t delta;
644 struct timecounter *tc;
645
646 tc = timecounter;
647 ts->tv_sec = tc->tc_offset_sec;
648 count = tco_delta(tc);
649 delta = tc->tc_offset_nano;
650 delta += ((u_int64_t)count * tc->tc_scale_nano_f);
651 delta >>= 32;
652 delta += ((u_int64_t)count * tc->tc_scale_nano_i);
653 while (delta < 0) {
654 delta += 1000000000;
655 if (ts->tv_sec > 0)
656 ts->tv_sec--;
657 }
658 while (delta >= 1000000000) {
659 delta -= 1000000000;
660 ts->tv_sec++;
661 }
662 ts->tv_nsec = delta;
663}
664
665static void
666tco_setscales(struct timecounter *tc)
667{
668 u_int64_t scale;
669
670 scale = 1000000000LL << 32;
671 scale += tc->tc_adjustment;
672 scale /= tc->tc_tweak->tc_frequency;
673 tc->tc_scale_micro = scale / 1000;
674 tc->tc_scale_nano_f = scale & 0xffffffff;
675 tc->tc_scale_nano_i = scale >> 32;
676}
677
678void
679update_timecounter(struct timecounter *tc)
680{
681 tco_setscales(tc);
682}
683
684void
685init_timecounter(struct timecounter *tc)
686{
687 struct timespec ts1;
688 struct timecounter *t1, *t2, *t3;
689 unsigned u;
690 int i;
691
692 u = tc->tc_frequency / tc->tc_counter_mask;
693 if (u > hz) {
694 printf("Timecounter \"%s\" frequency %lu Hz"
695 " -- Insufficient hz, needs at least %u\n",
696 tc->tc_name, (u_long) tc->tc_frequency, u);
697 return;
698 }
699
700 tc->tc_adjustment = 0;
701 tc->tc_tweak = tc;
702 tco_setscales(tc);
703 tc->tc_offset_count = tc->tc_get_timecount(tc);
704 if (timecounter == &dummy_timecounter)
705 tc->tc_avail = tc;
706 else {
707 tc->tc_avail = timecounter->tc_tweak->tc_avail;
708 timecounter->tc_tweak->tc_avail = tc;
709 }
710 MALLOC(t1, struct timecounter *, sizeof *t1, M_TIMECOUNTER, M_WAITOK);
711 tc->tc_other = t1;
712 *t1 = *tc;
713 t2 = t1;
714 for (i = 1; i < NTIMECOUNTER; i++) {
715 MALLOC(t3, struct timecounter *, sizeof *t3,
716 M_TIMECOUNTER, M_WAITOK);
717 *t3 = *tc;
718 t3->tc_other = t2;
719 t2 = t3;
720 }
721 t1->tc_other = t3;
722 tc = t1;
723
724 printf("Timecounter \"%s\" frequency %lu Hz\n",
725 tc->tc_name, (u_long)tc->tc_frequency);
726
727 /* XXX: For now always start using the counter. */
728 tc->tc_offset_count = tc->tc_get_timecount(tc);
729 nanouptime(&ts1);
730 tc->tc_offset_nano = (u_int64_t)ts1.tv_nsec << 32;
731 tc->tc_offset_micro = ts1.tv_nsec / 1000;
732 tc->tc_offset_sec = ts1.tv_sec;
733 timecounter = tc;
734}
735
736void
737set_timecounter(struct timespec *ts)
738{
739 struct timespec ts2;
740
741 nanouptime(&ts2);
742 boottime.tv_sec = ts->tv_sec - ts2.tv_sec;
743 boottime.tv_usec = (ts->tv_nsec - ts2.tv_nsec) / 1000;
744 if (boottime.tv_usec < 0) {
745 boottime.tv_usec += 1000000;
746 boottime.tv_sec--;
747 }
748 /* fiddle all the little crinkly bits around the fiords... */
749 tco_forward(1);
750}
751
752static void
753switch_timecounter(struct timecounter *newtc)
754{
755 int s;
756 struct timecounter *tc;
757 struct timespec ts;
758
759 s = splclock();
760 tc = timecounter;
761 if (newtc->tc_tweak == tc->tc_tweak) {
762 splx(s);
763 return;
764 }
765 newtc = newtc->tc_tweak->tc_other;
766 nanouptime(&ts);
767 newtc->tc_offset_sec = ts.tv_sec;
768 newtc->tc_offset_nano = (u_int64_t)ts.tv_nsec << 32;
769 newtc->tc_offset_micro = ts.tv_nsec / 1000;
770 newtc->tc_offset_count = newtc->tc_get_timecount(newtc);
771 tco_setscales(newtc);
772 timecounter = newtc;
773 splx(s);
774}
775
776static struct timecounter *
777sync_other_counter(void)
778{
779 struct timecounter *tc, *tcn, *tco;
780 unsigned delta;
781
782 tco = timecounter;
783 tc = tco->tc_other;
784 tcn = tc->tc_other;
785 *tc = *tco;
786 tc->tc_other = tcn;
787 delta = tco_delta(tc);
788 tc->tc_offset_count += delta;
789 tc->tc_offset_count &= tc->tc_counter_mask;
790 tc->tc_offset_nano += (u_int64_t)delta * tc->tc_scale_nano_f;
791 tc->tc_offset_nano += (u_int64_t)delta * tc->tc_scale_nano_i << 32;
792 return (tc);
793}
794
795static void
796tco_forward(int force)
797{
798 struct timecounter *tc, *tco;
799 struct timeval tvt;
800
801 tco = timecounter;
802 tc = sync_other_counter();
803 /*
804 * We may be inducing a tiny error here, the tc_poll_pps() may
805 * process a latched count which happens after the tco_delta()
806 * in sync_other_counter(), which would extend the previous
807 * counters parameters into the domain of this new one.
808 * Since the timewindow is very small for this, the error is
809 * going to be only a few weenieseconds (as Dave Mills would
810 * say), so lets just not talk more about it, OK ?
811 */
812 if (tco->tc_poll_pps)
813 tco->tc_poll_pps(tco);
814 if (timedelta != 0) {
815 tvt = boottime;
816 tvt.tv_usec += tickdelta;
817 if (tvt.tv_usec >= 1000000) {
818 tvt.tv_sec++;
819 tvt.tv_usec -= 1000000;
820 } else if (tvt.tv_usec < 0) {
821 tvt.tv_sec--;
822 tvt.tv_usec += 1000000;
823 }
824 boottime = tvt;
825 timedelta -= tickdelta;
826 }
827
828 while (tc->tc_offset_nano >= 1000000000ULL << 32) {
829 tc->tc_offset_nano -= 1000000000ULL << 32;
830 tc->tc_offset_sec++;
831 ntp_update_second(tc); /* XXX only needed if xntpd runs */
832 tco_setscales(tc);
833 force++;
834 }
835
836 if (tco_method && !force)
837 return;
838
839 tc->tc_offset_micro = (tc->tc_offset_nano / 1000) >> 32;
840
841 /* Figure out the wall-clock time */
842 tc->tc_nanotime.tv_sec = tc->tc_offset_sec + boottime.tv_sec;
843 tc->tc_nanotime.tv_nsec =
844 (tc->tc_offset_nano >> 32) + boottime.tv_usec * 1000;
845 tc->tc_microtime.tv_usec = tc->tc_offset_micro + boottime.tv_usec;
846 while (tc->tc_nanotime.tv_nsec >= 1000000000) {
847 tc->tc_nanotime.tv_nsec -= 1000000000;
848 tc->tc_microtime.tv_usec -= 1000000;
849 tc->tc_nanotime.tv_sec++;
850 }
851 time_second = tc->tc_microtime.tv_sec = tc->tc_nanotime.tv_sec;
852
853 timecounter = tc;
854}
855
856SYSCTL_NODE(_kern, OID_AUTO, timecounter, CTLFLAG_RW, 0, "");
857
858SYSCTL_INT(_kern_timecounter, OID_AUTO, method, CTLFLAG_RW, &tco_method, 0,
859 "This variable determines the method used for updating timecounters. "
860 "If the default algorithm (0) fails with \"calcru negative...\" messages "
861 "try the alternate algorithm (1) which handles bad hardware better."
862
863);
864
865static int
866sysctl_kern_timecounter_hardware(SYSCTL_HANDLER_ARGS)
867{
868 char newname[32];
869 struct timecounter *newtc, *tc;
870 int error;
871
872 tc = timecounter->tc_tweak;
873 strncpy(newname, tc->tc_name, sizeof(newname));
874 error = sysctl_handle_string(oidp, &newname[0], sizeof(newname), req);
875 if (error == 0 && req->newptr != NULL &&
876 strcmp(newname, tc->tc_name) != 0) {
877 for (newtc = tc->tc_avail; newtc != tc;
878 newtc = newtc->tc_avail) {
879 if (strcmp(newname, newtc->tc_name) == 0) {
880 /* Warm up new timecounter. */
881 (void)newtc->tc_get_timecount(newtc);
882
883 switch_timecounter(newtc);
884 return (0);
885 }
886 }
887 return (EINVAL);
888 }
889 return (error);
890}
891
892SYSCTL_PROC(_kern_timecounter, OID_AUTO, hardware, CTLTYPE_STRING | CTLFLAG_RW,
893 0, 0, sysctl_kern_timecounter_hardware, "A", "");
894
895
896int
897pps_ioctl(u_long cmd, caddr_t data, struct pps_state *pps)
898{
899 pps_params_t *app;
900 struct pps_fetch_args *fapi;
901#ifdef PPS_SYNC
902 struct pps_kcbind_args *kapi;
903#endif
904
905 switch (cmd) {
906 case PPS_IOC_CREATE:
907 return (0);
908 case PPS_IOC_DESTROY:
909 return (0);
910 case PPS_IOC_SETPARAMS:
911 app = (pps_params_t *)data;
912 if (app->mode & ~pps->ppscap)
913 return (EINVAL);
914 pps->ppsparam = *app;
915 return (0);
916 case PPS_IOC_GETPARAMS:
917 app = (pps_params_t *)data;
918 *app = pps->ppsparam;
919 app->api_version = PPS_API_VERS_1;
920 return (0);
921 case PPS_IOC_GETCAP:
922 *(int*)data = pps->ppscap;
923 return (0);
924 case PPS_IOC_FETCH:
925 fapi = (struct pps_fetch_args *)data;
926 if (fapi->tsformat && fapi->tsformat != PPS_TSFMT_TSPEC)
927 return (EINVAL);
928 if (fapi->timeout.tv_sec || fapi->timeout.tv_nsec)
929 return (EOPNOTSUPP);
930 pps->ppsinfo.current_mode = pps->ppsparam.mode;
931 fapi->pps_info_buf = pps->ppsinfo;
932 return (0);
933 case PPS_IOC_KCBIND:
934#ifdef PPS_SYNC
935 kapi = (struct pps_kcbind_args *)data;
936 /* XXX Only root should be able to do this */
937 if (kapi->tsformat && kapi->tsformat != PPS_TSFMT_TSPEC)
938 return (EINVAL);
939 if (kapi->kernel_consumer != PPS_KC_HARDPPS)
940 return (EINVAL);
941 if (kapi->edge & ~pps->ppscap)
942 return (EINVAL);
943 pps->kcmode = kapi->edge;
944 return (0);
945#else
946 return (EOPNOTSUPP);
947#endif
948 default:
949 return (ENOTTY);
950 }
951}
952
953void
954pps_init(struct pps_state *pps)
955{
956 pps->ppscap |= PPS_TSFMT_TSPEC;
957 if (pps->ppscap & PPS_CAPTUREASSERT)
958 pps->ppscap |= PPS_OFFSETASSERT;
959 if (pps->ppscap & PPS_CAPTURECLEAR)
960 pps->ppscap |= PPS_OFFSETCLEAR;
961}
962
963void
964pps_event(struct pps_state *pps, struct timecounter *tc, unsigned count, int event)
965{
966 struct timespec ts, *tsp, *osp;
967 u_int64_t delta;
968 unsigned tcount, *pcount;
969 int foff, fhard;
970 pps_seq_t *pseq;
971
972 /* Things would be easier with arrays... */
973 if (event == PPS_CAPTUREASSERT) {
974 tsp = &pps->ppsinfo.assert_timestamp;
975 osp = &pps->ppsparam.assert_offset;
976 foff = pps->ppsparam.mode & PPS_OFFSETASSERT;
977 fhard = pps->kcmode & PPS_CAPTUREASSERT;
978 pcount = &pps->ppscount[0];
979 pseq = &pps->ppsinfo.assert_sequence;
980 } else {
981 tsp = &pps->ppsinfo.clear_timestamp;
982 osp = &pps->ppsparam.clear_offset;
983 foff = pps->ppsparam.mode & PPS_OFFSETCLEAR;
984 fhard = pps->kcmode & PPS_CAPTURECLEAR;
985 pcount = &pps->ppscount[1];
986 pseq = &pps->ppsinfo.clear_sequence;
987 }
988
989 /* The timecounter changed: bail */
990 if (!pps->ppstc ||
991 pps->ppstc->tc_name != tc->tc_name ||
992 tc->tc_name != timecounter->tc_name) {
993 pps->ppstc = tc;
994 *pcount = count;
995 return;
996 }
997
998 /* Nothing really happened */
999 if (*pcount == count)
1000 return;
1001
1002 *pcount = count;
1003
1004 /* Convert the count to timespec */
1005 ts.tv_sec = tc->tc_offset_sec;
1006 tcount = count - tc->tc_offset_count;
1007 tcount &= tc->tc_counter_mask;
1008 delta = tc->tc_offset_nano;
1009 delta += ((u_int64_t)tcount * tc->tc_scale_nano_f);
1010 delta >>= 32;
1011 delta += ((u_int64_t)tcount * tc->tc_scale_nano_i);
1012 delta += boottime.tv_usec * 1000;
1013 ts.tv_sec += boottime.tv_sec;
1014 while (delta >= 1000000000) {
1015 delta -= 1000000000;
1016 ts.tv_sec++;
1017 }
1018 ts.tv_nsec = delta;
1019
1020 (*pseq)++;
1021 *tsp = ts;
1022
1023 if (foff) {
1024 timespecadd(tsp, osp);
1025 if (tsp->tv_nsec < 0) {
1026 tsp->tv_nsec += 1000000000;
1027 tsp->tv_sec -= 1;
1028 }
1029 }
1030#ifdef PPS_SYNC
1031 if (fhard) {
1032 /* magic, at its best... */
1033 tcount = count - pps->ppscount[2];
1034 pps->ppscount[2] = count;
1035 tcount &= tc->tc_counter_mask;
1036 delta = ((u_int64_t)tcount * tc->tc_tweak->tc_scale_nano_f);
1037 delta >>= 32;
1038 delta += ((u_int64_t)tcount * tc->tc_tweak->tc_scale_nano_i);
1039 hardpps(tsp, delta);
1040 }
1041#endif
1042}