Introduce pthread_timedjoin_np.
[dragonfly.git] / sys / kern / kern_clock.c
CommitLineData
8c10bfcf
MD
1/*
2 * Copyright (c) 2003,2004 The DragonFly Project. All rights reserved.
3 *
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 *
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
16 * distribution.
17 * 3. Neither the name of The DragonFly Project nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific, prior written permission.
20 *
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
984263bc
MD
34 * Copyright (c) 1997, 1998 Poul-Henning Kamp <phk@FreeBSD.org>
35 * Copyright (c) 1982, 1986, 1991, 1993
36 * The Regents of the University of California. All rights reserved.
37 * (c) UNIX System Laboratories, Inc.
38 * All or some portions of this file are derived from material licensed
39 * to the University of California by American Telephone and Telegraph
40 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
41 * the permission of UNIX System Laboratories, Inc.
42 *
43 * Redistribution and use in source and binary forms, with or without
44 * modification, are permitted provided that the following conditions
45 * are met:
46 * 1. Redistributions of source code must retain the above copyright
47 * notice, this list of conditions and the following disclaimer.
48 * 2. Redistributions in binary form must reproduce the above copyright
49 * notice, this list of conditions and the following disclaimer in the
50 * documentation and/or other materials provided with the distribution.
51 * 3. All advertising materials mentioning features or use of this software
52 * must display the following acknowledgement:
53 * This product includes software developed by the University of
54 * California, Berkeley and its contributors.
55 * 4. Neither the name of the University nor the names of its contributors
56 * may be used to endorse or promote products derived from this software
57 * without specific prior written permission.
58 *
59 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
60 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
61 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
62 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
63 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
64 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
65 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
66 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
67 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
68 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
69 * SUCH DAMAGE.
70 *
71 * @(#)kern_clock.c 8.5 (Berkeley) 1/21/94
72 * $FreeBSD: src/sys/kern/kern_clock.c,v 1.105.2.10 2002/10/17 13:19:40 maxim Exp $
93328593 73 * $DragonFly: src/sys/kern/kern_clock.c,v 1.46 2005/10/08 12:24:26 corecode Exp $
984263bc
MD
74 */
75
76#include "opt_ntp.h"
77
78#include <sys/param.h>
79#include <sys/systm.h>
984263bc
MD
80#include <sys/callout.h>
81#include <sys/kernel.h>
f5d21610 82#include <sys/kinfo.h>
984263bc
MD
83#include <sys/proc.h>
84#include <sys/malloc.h>
85#include <sys/resourcevar.h>
86#include <sys/signalvar.h>
87#include <sys/timex.h>
88#include <sys/timepps.h>
89#include <vm/vm.h>
90#include <sys/lock.h>
91#include <vm/pmap.h>
92#include <vm/vm_map.h>
93#include <sys/sysctl.h>
2689779e 94#include <sys/thread2.h>
984263bc
MD
95
96#include <machine/cpu.h>
97#include <machine/limits.h>
98#include <machine/smp.h>
99
100#ifdef GPROF
101#include <sys/gmon.h>
102#endif
103
104#ifdef DEVICE_POLLING
105extern void init_device_poll(void);
106extern void hardclock_device_poll(void);
107#endif /* DEVICE_POLLING */
108
402ed7e1 109static void initclocks (void *dummy);
984263bc
MD
110SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL)
111
6ad39cae
MD
112/*
113 * Some of these don't belong here, but it's easiest to concentrate them.
9eea7f0c 114 * Note that cpu_time counts in microseconds, but most userland programs
6ad39cae
MD
115 * just compare relative times against the total by delta.
116 */
9eea7f0c
HP
117struct kinfo_cputime cputime_percpu[MAXCPU];
118#ifdef SMP
119static int
120sysctl_cputime(SYSCTL_HANDLER_ARGS)
121{
122 int cpu, error = 0;
123 size_t size = sizeof(struct kinfo_cputime);
124
125 for (cpu = 0; cpu < ncpus; ++cpu) {
126 if ((error = SYSCTL_OUT(req, &cputime_percpu[cpu], size)))
127 break;
128 }
984263bc 129
9eea7f0c
HP
130 return (error);
131}
132SYSCTL_PROC(_kern, OID_AUTO, cputime, (CTLTYPE_OPAQUE|CTLFLAG_RD), 0, 0,
133 sysctl_cputime, "S,kinfo_cputime", "CPU time statistics");
134#else
135SYSCTL_STRUCT(_kern, OID_AUTO, cputime, CTLFLAG_RD, &cpu_time, kinfo_cputime,
136 "CPU time statistics");
137#endif
984263bc 138
88c4d2f6
MD
139/*
140 * boottime is used to calculate the 'real' uptime. Do not confuse this with
141 * microuptime(). microtime() is not drift compensated. The real uptime
60b2809b
MD
142 * with compensation is nanotime() - bootime. boottime is recalculated
143 * whenever the real time is set based on the compensated elapsed time
144 * in seconds (gd->gd_time_seconds).
88c4d2f6 145 *
88c4d2f6
MD
146 * The gd_time_seconds and gd_cpuclock_base fields remain fairly monotonic.
147 * Slight adjustments to gd_cpuclock_base are made to phase-lock it to
148 * the real time.
149 */
150struct timespec boottime; /* boot time (realtime) for reference only */
88c4d2f6 151time_t time_second; /* read-only 'passive' uptime in seconds */
984263bc 152
5eb5a6bc
MD
153/*
154 * basetime is used to calculate the compensated real time of day. The
155 * basetime can be modified on a per-tick basis by the adjtime(),
156 * ntp_adjtime(), and sysctl-based time correction APIs.
157 *
158 * Note that frequency corrections can also be made by adjusting
159 * gd_cpuclock_base.
160 *
161 * basetime is a tail-chasing FIFO, updated only by cpu #0. The FIFO is
162 * used on both SMP and UP systems to avoid MP races between cpu's and
163 * interrupt races on UP systems.
164 */
165#define BASETIME_ARYSIZE 16
166#define BASETIME_ARYMASK (BASETIME_ARYSIZE - 1)
167static struct timespec basetime[BASETIME_ARYSIZE];
168static volatile int basetime_index;
169
170static int
171sysctl_get_basetime(SYSCTL_HANDLER_ARGS)
172{
173 struct timespec *bt;
174 int error;
35238fa5 175 int index;
5eb5a6bc 176
35238fa5
MD
177 /*
178 * Because basetime data and index may be updated by another cpu,
179 * a load fence is required to ensure that the data we read has
180 * not been speculatively read relative to a possibly updated index.
181 */
182 index = basetime_index;
183 cpu_lfence();
184 bt = &basetime[index];
08f95c49 185 error = SYSCTL_OUT(req, bt, sizeof(*bt));
5eb5a6bc
MD
186 return (error);
187}
188
984263bc 189SYSCTL_STRUCT(_kern, KERN_BOOTTIME, boottime, CTLFLAG_RD,
08f95c49 190 &boottime, timespec, "System boottime");
5eb5a6bc 191SYSCTL_PROC(_kern, OID_AUTO, basetime, CTLTYPE_STRUCT|CTLFLAG_RD, 0, 0,
08f95c49 192 sysctl_get_basetime, "S,timespec", "System basetime");
984263bc 193
88c4d2f6
MD
194static void hardclock(systimer_t info, struct intrframe *frame);
195static void statclock(systimer_t info, struct intrframe *frame);
196static void schedclock(systimer_t info, struct intrframe *frame);
5eb5a6bc 197static void getnanotime_nbt(struct timespec *nbt, struct timespec *tsp);
88c4d2f6
MD
198
199int ticks; /* system master ticks at hz */
da3639ef 200int clocks_running; /* tsleep/timeout clocks operational */
88c4d2f6
MD
201int64_t nsec_adj; /* ntpd per-tick adjustment in nsec << 32 */
202int64_t nsec_acc; /* accumulator */
984263bc 203
4026c000
JS
204/* NTPD time correction fields */
205int64_t ntp_tick_permanent; /* per-tick adjustment in nsec << 32 */
206int64_t ntp_tick_acc; /* accumulator for per-tick adjustment */
207int64_t ntp_delta; /* one-time correction in nsec */
208int64_t ntp_big_delta = 1000000000;
209int32_t ntp_tick_delta; /* current adjustment rate */
210int32_t ntp_default_tick_delta; /* adjustment rate for ntp_delta */
48590578
JS
211time_t ntp_leap_second; /* time of next leap second */
212int ntp_leap_insert; /* whether to insert or remove a second */
4026c000 213
984263bc 214/*
88c4d2f6 215 * Finish initializing clock frequencies and start all clocks running.
984263bc 216 */
88c4d2f6
MD
217/* ARGSUSED*/
218static void
219initclocks(void *dummy)
984263bc 220{
88c4d2f6
MD
221 cpu_initclocks();
222#ifdef DEVICE_POLLING
223 init_device_poll();
224#endif
225 /*psratio = profhz / stathz;*/
226 initclocks_pcpu();
da3639ef 227 clocks_running = 1;
984263bc
MD
228}
229
88c4d2f6
MD
230/*
231 * Called on a per-cpu basis
232 */
233void
234initclocks_pcpu(void)
235{
236 struct globaldata *gd = mycpu;
984263bc 237
88c4d2f6
MD
238 crit_enter();
239 if (gd->gd_cpuid == 0) {
240 gd->gd_time_seconds = 1;
044ee7c4 241 gd->gd_cpuclock_base = sys_cputimer->count();
88c4d2f6
MD
242 } else {
243 /* XXX */
244 gd->gd_time_seconds = globaldata_find(0)->gd_time_seconds;
245 gd->gd_cpuclock_base = globaldata_find(0)->gd_cpuclock_base;
246 }
0d1dffdf
MD
247
248 /*
249 * Use a non-queued periodic systimer to prevent multiple ticks from
250 * building up if the sysclock jumps forward (8254 gets reset). The
251 * sysclock will never jump backwards. Our time sync is based on
252 * the actual sysclock, not the ticks count.
253 */
254 systimer_init_periodic_nq(&gd->gd_hardclock, hardclock, NULL, hz);
255 systimer_init_periodic_nq(&gd->gd_statclock, statclock, NULL, stathz);
88c4d2f6 256 /* XXX correct the frequency for scheduler / estcpu tests */
0d1dffdf 257 systimer_init_periodic_nq(&gd->gd_schedclock, schedclock,
8478264a 258 NULL, ESTCPUFREQ);
88c4d2f6
MD
259 crit_exit();
260}
984263bc
MD
261
262/*
88c4d2f6
MD
263 * This sets the current real time of day. Timespecs are in seconds and
264 * nanoseconds. We do not mess with gd_time_seconds and gd_cpuclock_base,
265 * instead we adjust basetime so basetime + gd_* results in the current
266 * time of day. This way the gd_* fields are guarenteed to represent
267 * a monotonically increasing 'uptime' value.
5eb5a6bc
MD
268 *
269 * When set_timeofday() is called from userland, the system call forces it
270 * onto cpu #0 since only cpu #0 can update basetime_index.
984263bc 271 */
88c4d2f6
MD
272void
273set_timeofday(struct timespec *ts)
274{
5eb5a6bc
MD
275 struct timespec *nbt;
276 int ni;
984263bc 277
88c4d2f6
MD
278 /*
279 * XXX SMP / non-atomic basetime updates
280 */
281 crit_enter();
5eb5a6bc
MD
282 ni = (basetime_index + 1) & BASETIME_ARYMASK;
283 nbt = &basetime[ni];
284 nanouptime(nbt);
285 nbt->tv_sec = ts->tv_sec - nbt->tv_sec;
286 nbt->tv_nsec = ts->tv_nsec - nbt->tv_nsec;
287 if (nbt->tv_nsec < 0) {
288 nbt->tv_nsec += 1000000000;
289 --nbt->tv_sec;
88c4d2f6 290 }
a81931cc
MD
291
292 /*
293 * Note that basetime diverges from boottime as the clock drift is
294 * compensated for, so we cannot do away with boottime. When setting
295 * the absolute time of day the drift is 0 (for an instant) and we
296 * can simply assign boottime to basetime.
297 *
298 * Note that nanouptime() is based on gd_time_seconds which is drift
299 * compensated up to a point (it is guarenteed to remain monotonically
300 * increasing). gd_time_seconds is thus our best uptime guess and
301 * suitable for use in the boottime calculation. It is already taken
302 * into account in the basetime calculation above.
303 */
5eb5a6bc 304 boottime.tv_sec = nbt->tv_sec;
4026c000 305 ntp_delta = 0;
5eb5a6bc
MD
306
307 /*
35238fa5
MD
308 * We now have a new basetime, make sure all other cpus have it,
309 * then update the index.
5eb5a6bc 310 */
35238fa5 311 cpu_sfence();
5eb5a6bc
MD
312 basetime_index = ni;
313
88c4d2f6
MD
314 crit_exit();
315}
316
984263bc 317/*
88c4d2f6
MD
318 * Each cpu has its own hardclock, but we only increments ticks and softticks
319 * on cpu #0.
320 *
321 * NOTE! systimer! the MP lock might not be held here. We can only safely
322 * manipulate objects owned by the current cpu.
984263bc 323 */
984263bc 324static void
88c4d2f6 325hardclock(systimer_t info, struct intrframe *frame)
984263bc 326{
88c4d2f6
MD
327 sysclock_t cputicks;
328 struct proc *p;
329 struct pstats *pstats;
330 struct globaldata *gd = mycpu;
984263bc
MD
331
332 /*
88c4d2f6
MD
333 * Realtime updates are per-cpu. Note that timer corrections as
334 * returned by microtime() and friends make an additional adjustment
335 * using a system-wise 'basetime', but the running time is always
336 * taken from the per-cpu globaldata area. Since the same clock
337 * is distributing (XXX SMP) to all cpus, the per-cpu timebases
338 * stay in synch.
339 *
340 * Note that we never allow info->time (aka gd->gd_hardclock.time)
fad57d0e
MD
341 * to reverse index gd_cpuclock_base, but that it is possible for
342 * it to temporarily get behind in the seconds if something in the
343 * system locks interrupts for a long period of time. Since periodic
344 * timers count events, though everything should resynch again
345 * immediately.
984263bc 346 */
88c4d2f6 347 cputicks = info->time - gd->gd_cpuclock_base;
044ee7c4 348 if (cputicks >= sys_cputimer->freq) {
88c4d2f6 349 ++gd->gd_time_seconds;
044ee7c4 350 gd->gd_cpuclock_base += sys_cputimer->freq;
88c4d2f6 351 }
984263bc
MD
352
353 /*
92b561b7
MD
354 * The system-wide ticks counter and NTP related timedelta/tickdelta
355 * adjustments only occur on cpu #0. NTP adjustments are accomplished
356 * by updating basetime.
984263bc 357 */
88c4d2f6 358 if (gd->gd_cpuid == 0) {
5eb5a6bc 359 struct timespec *nbt;
88c4d2f6
MD
360 struct timespec nts;
361 int leap;
5eb5a6bc 362 int ni;
984263bc 363
88c4d2f6 364 ++ticks;
984263bc 365
88c4d2f6
MD
366#ifdef DEVICE_POLLING
367 hardclock_device_poll(); /* mpsafe, short and quick */
368#endif /* DEVICE_POLLING */
984263bc 369
88c4d2f6
MD
370#if 0
371 if (tco->tc_poll_pps)
372 tco->tc_poll_pps(tco);
373#endif
5eb5a6bc 374
88c4d2f6 375 /*
5eb5a6bc
MD
376 * Calculate the new basetime index. We are in a critical section
377 * on cpu #0 and can safely play with basetime_index. Start
378 * with the current basetime and then make adjustments.
379 */
380 ni = (basetime_index + 1) & BASETIME_ARYMASK;
381 nbt = &basetime[ni];
382 *nbt = basetime[basetime_index];
383
384 /*
385 * Apply adjtime corrections. (adjtime() API)
386 *
387 * adjtime() only runs on cpu #0 so our critical section is
388 * sufficient to access these variables.
88c4d2f6 389 */
4026c000 390 if (ntp_delta != 0) {
5eb5a6bc 391 nbt->tv_nsec += ntp_tick_delta;
4026c000
JS
392 ntp_delta -= ntp_tick_delta;
393 if ((ntp_delta > 0 && ntp_delta < ntp_tick_delta) ||
394 (ntp_delta < 0 && ntp_delta > ntp_tick_delta)) {
5eb5a6bc 395 ntp_tick_delta = ntp_delta;
4026c000
JS
396 }
397 }
398
5eb5a6bc
MD
399 /*
400 * Apply permanent frequency corrections. (sysctl API)
401 */
4026c000
JS
402 if (ntp_tick_permanent != 0) {
403 ntp_tick_acc += ntp_tick_permanent;
404 if (ntp_tick_acc >= (1LL << 32)) {
5eb5a6bc 405 nbt->tv_nsec += ntp_tick_acc >> 32;
331bc6f8 406 ntp_tick_acc -= (ntp_tick_acc >> 32) << 32;
4026c000 407 } else if (ntp_tick_acc <= -(1LL << 32)) {
331bc6f8 408 /* Negate ntp_tick_acc to avoid shifting the sign bit. */
5eb5a6bc 409 nbt->tv_nsec -= (-ntp_tick_acc) >> 32;
331bc6f8 410 ntp_tick_acc += ((-ntp_tick_acc) >> 32) << 32;
4026c000
JS
411 }
412 }
413
5eb5a6bc
MD
414 if (nbt->tv_nsec >= 1000000000) {
415 nbt->tv_sec++;
416 nbt->tv_nsec -= 1000000000;
417 } else if (nbt->tv_nsec < 0) {
418 nbt->tv_sec--;
419 nbt->tv_nsec += 1000000000;
88c4d2f6
MD
420 }
421
422 /*
5eb5a6bc 423 * Another per-tick compensation. (for ntp_adjtime() API)
88c4d2f6 424 */
5eb5a6bc 425 if (nsec_adj != 0) {
88c4d2f6
MD
426 nsec_acc += nsec_adj;
427 if (nsec_acc >= 0x100000000LL) {
5eb5a6bc 428 nbt->tv_nsec += nsec_acc >> 32;
88c4d2f6
MD
429 nsec_acc = (nsec_acc & 0xFFFFFFFFLL);
430 } else if (nsec_acc <= -0x100000000LL) {
5eb5a6bc 431 nbt->tv_nsec -= -nsec_acc >> 32;
88c4d2f6
MD
432 nsec_acc = -(-nsec_acc & 0xFFFFFFFFLL);
433 }
5eb5a6bc
MD
434 if (nbt->tv_nsec >= 1000000000) {
435 nbt->tv_nsec -= 1000000000;
436 ++nbt->tv_sec;
437 } else if (nbt->tv_nsec < 0) {
438 nbt->tv_nsec += 1000000000;
439 --nbt->tv_sec;
440 }
441 }
442
443 /************************************************************
444 * LEAP SECOND CORRECTION *
445 ************************************************************
446 *
447 * Taking into account all the corrections made above, figure
448 * out the new real time. If the seconds field has changed
449 * then apply any pending leap-second corrections.
450 */
451 getnanotime_nbt(nbt, &nts);
452
32040d57
MD
453 if (time_second != nts.tv_sec) {
454 /*
455 * Apply leap second (sysctl API). Adjust nts for changes
456 * so we do not have to call getnanotime_nbt again.
457 */
458 if (ntp_leap_second) {
459 if (ntp_leap_second == nts.tv_sec) {
460 if (ntp_leap_insert) {
461 nbt->tv_sec++;
462 nts.tv_sec++;
463 } else {
464 nbt->tv_sec--;
465 nts.tv_sec--;
466 }
5eb5a6bc 467 ntp_leap_second--;
32040d57 468 }
88c4d2f6 469 }
88c4d2f6 470
32040d57
MD
471 /*
472 * Apply leap second (ntp_adjtime() API), calculate a new
473 * nsec_adj field. ntp_update_second() returns nsec_adj
474 * as a per-second value but we need it as a per-tick value.
475 */
88c4d2f6 476 leap = ntp_update_second(time_second, &nsec_adj);
88c4d2f6 477 nsec_adj /= hz;
32040d57
MD
478 nbt->tv_sec += leap;
479 nts.tv_sec += leap;
480
481 /*
482 * Update the time_second 'approximate time' global.
483 */
484 time_second = nts.tv_sec;
88c4d2f6 485 }
5eb5a6bc
MD
486
487 /*
488 * Finally, our new basetime is ready to go live!
489 */
35238fa5 490 cpu_sfence();
5eb5a6bc 491 basetime_index = ni;
88c4d2f6
MD
492 }
493
92b561b7
MD
494 /*
495 * softticks are handled for all cpus
496 */
497 hardclock_softtick(gd);
498
88c4d2f6
MD
499 /*
500 * ITimer handling is per-tick, per-cpu. I don't think psignal()
501 * is mpsafe on curproc, so XXX get the mplock.
502 */
503 if ((p = curproc) != NULL && try_mplock()) {
984263bc 504 pstats = p->p_stats;
88c4d2f6 505 if (frame && CLKF_USERMODE(frame) &&
93328593
SS
506 timevalisset(&p->p_timer[ITIMER_VIRTUAL].it_value) &&
507 itimerdecr(&p->p_timer[ITIMER_VIRTUAL], tick) == 0)
984263bc 508 psignal(p, SIGVTALRM);
93328593
SS
509 if (timevalisset(&p->p_timer[ITIMER_PROF].it_value) &&
510 itimerdecr(&p->p_timer[ITIMER_PROF], tick) == 0)
984263bc 511 psignal(p, SIGPROF);
88c4d2f6 512 rel_mplock();
984263bc 513 }
604e1e09 514 setdelayed();
88c4d2f6 515}
984263bc 516
88c4d2f6
MD
517/*
518 * The statistics clock typically runs at a 125Hz rate, and is intended
519 * to be frequency offset from the hardclock (typ 100Hz). It is per-cpu.
520 *
521 * NOTE! systimer! the MP lock might not be held here. We can only safely
522 * manipulate objects owned by the current cpu.
523 *
524 * The stats clock is responsible for grabbing a profiling sample.
525 * Most of the statistics are only used by user-level statistics programs.
526 * The main exceptions are p->p_uticks, p->p_sticks, p->p_iticks, and
527 * p->p_estcpu.
528 *
529 * Like the other clocks, the stat clock is called from what is effectively
530 * a fast interrupt, so the context should be the thread/process that got
531 * interrupted.
532 */
533static void
534statclock(systimer_t info, struct intrframe *frame)
535{
536#ifdef GPROF
537 struct gmonparam *g;
538 int i;
984263bc 539#endif
88c4d2f6
MD
540 thread_t td;
541 struct proc *p;
542 int bump;
543 struct timeval tv;
544 struct timeval *stv;
984263bc
MD
545
546 /*
88c4d2f6 547 * How big was our timeslice relative to the last time?
984263bc 548 */
88c4d2f6
MD
549 microuptime(&tv); /* mpsafe */
550 stv = &mycpu->gd_stattv;
551 if (stv->tv_sec == 0) {
552 bump = 1;
553 } else {
554 bump = tv.tv_usec - stv->tv_usec +
555 (tv.tv_sec - stv->tv_sec) * 1000000;
556 if (bump < 0)
557 bump = 0;
558 if (bump > 1000000)
559 bump = 1000000;
560 }
561 *stv = tv;
984263bc 562
88c4d2f6
MD
563 td = curthread;
564 p = td->td_proc;
984263bc 565
88c4d2f6
MD
566 if (frame && CLKF_USERMODE(frame)) {
567 /*
568 * Came from userland, handle user time and deal with
569 * possible process.
570 */
571 if (p && (p->p_flag & P_PROFIL))
572 addupc_intr(p, CLKF_PC(frame), 1);
573 td->td_uticks += bump;
984263bc 574
88c4d2f6
MD
575 /*
576 * Charge the time as appropriate
577 */
578 if (p && p->p_nice > NZERO)
9eea7f0c 579 cpu_time.cp_nice += bump;
88c4d2f6 580 else
9eea7f0c 581 cpu_time.cp_user += bump;
88c4d2f6
MD
582 } else {
583#ifdef GPROF
584 /*
585 * Kernel statistics are just like addupc_intr, only easier.
586 */
587 g = &_gmonparam;
588 if (g->state == GMON_PROF_ON && frame) {
589 i = CLKF_PC(frame) - g->lowpc;
590 if (i < g->textsize) {
591 i /= HISTFRACTION * sizeof(*g->kcount);
592 g->kcount[i]++;
593 }
594 }
595#endif
596 /*
597 * Came from kernel mode, so we were:
598 * - handling an interrupt,
599 * - doing syscall or trap work on behalf of the current
600 * user process, or
601 * - spinning in the idle loop.
602 * Whichever it is, charge the time as appropriate.
603 * Note that we charge interrupts to the current process,
604 * regardless of whether they are ``for'' that process,
605 * so that we know how much of its real time was spent
606 * in ``non-process'' (i.e., interrupt) work.
607 *
608 * XXX assume system if frame is NULL. A NULL frame
e43a034f 609 * can occur if ipi processing is done from a crit_exit().
88c4d2f6
MD
610 */
611 if (frame && CLKF_INTR(frame))
612 td->td_iticks += bump;
613 else
614 td->td_sticks += bump;
615
616 if (frame && CLKF_INTR(frame)) {
9eea7f0c 617 cpu_time.cp_intr += bump;
88c4d2f6
MD
618 } else {
619 if (td == &mycpu->gd_idlethread)
9eea7f0c 620 cpu_time.cp_idle += bump;
88c4d2f6 621 else
9eea7f0c 622 cpu_time.cp_sys += bump;
88c4d2f6
MD
623 }
624 }
625}
626
627/*
dcc99b62 628 * The scheduler clock typically runs at a 50Hz rate. NOTE! systimer,
88c4d2f6
MD
629 * the MP lock might not be held. We can safely manipulate parts of curproc
630 * but that's about it.
dcc99b62
MD
631 *
632 * Each cpu has its own scheduler clock.
88c4d2f6
MD
633 */
634static void
635schedclock(systimer_t info, struct intrframe *frame)
636{
637 struct proc *p;
638 struct pstats *pstats;
639 struct rusage *ru;
640 struct vmspace *vm;
641 long rss;
642
dcc99b62
MD
643 if ((p = lwkt_preempted_proc()) != NULL) {
644 /*
645 * Account for cpu time used and hit the scheduler. Note
646 * that this call MUST BE MP SAFE, and the BGL IS NOT HELD
647 * HERE.
648 */
27df5e1e 649 ++p->p_cpticks;
dcc99b62
MD
650 p->p_usched->schedulerclock(p, info->periodic, info->time);
651 }
88c4d2f6 652 if ((p = curproc) != NULL) {
dcc99b62
MD
653 /*
654 * Update resource usage integrals and maximums.
655 */
88c4d2f6
MD
656 if ((pstats = p->p_stats) != NULL &&
657 (ru = &pstats->p_ru) != NULL &&
658 (vm = p->p_vmspace) != NULL) {
659 ru->ru_ixrss += pgtok(vm->vm_tsize);
660 ru->ru_idrss += pgtok(vm->vm_dsize);
661 ru->ru_isrss += pgtok(vm->vm_ssize);
662 rss = pgtok(vmspace_resident_count(vm));
663 if (ru->ru_maxrss < rss)
664 ru->ru_maxrss = rss;
665 }
b68b7282 666 }
984263bc
MD
667}
668
669/*
a94976ad
MD
670 * Compute number of ticks for the specified amount of time. The
671 * return value is intended to be used in a clock interrupt timed
672 * operation and guarenteed to meet or exceed the requested time.
673 * If the representation overflows, return INT_MAX. The minimum return
674 * value is 1 ticks and the function will average the calculation up.
675 * If any value greater then 0 microseconds is supplied, a value
676 * of at least 2 will be returned to ensure that a near-term clock
677 * interrupt does not cause the timeout to occur (degenerately) early.
678 *
679 * Note that limit checks must take into account microseconds, which is
680 * done simply by using the smaller signed long maximum instead of
681 * the unsigned long maximum.
682 *
683 * If ints have 32 bits, then the maximum value for any timeout in
684 * 10ms ticks is 248 days.
984263bc
MD
685 */
686int
a94976ad 687tvtohz_high(struct timeval *tv)
984263bc 688{
a94976ad 689 int ticks;
1fd87d54 690 long sec, usec;
984263bc 691
984263bc
MD
692 sec = tv->tv_sec;
693 usec = tv->tv_usec;
694 if (usec < 0) {
695 sec--;
696 usec += 1000000;
697 }
698 if (sec < 0) {
699#ifdef DIAGNOSTIC
700 if (usec > 0) {
701 sec++;
702 usec -= 1000000;
703 }
704 printf("tvotohz: negative time difference %ld sec %ld usec\n",
705 sec, usec);
706#endif
707 ticks = 1;
a94976ad
MD
708 } else if (sec <= INT_MAX / hz) {
709 ticks = (int)(sec * hz +
710 ((u_long)usec + (tick - 1)) / tick) + 1;
711 } else {
712 ticks = INT_MAX;
713 }
714 return (ticks);
715}
716
717/*
718 * Compute number of ticks for the specified amount of time, erroring on
719 * the side of it being too low to ensure that sleeping the returned number
720 * of ticks will not result in a late return.
721 *
722 * The supplied timeval may not be negative and should be normalized. A
723 * return value of 0 is possible if the timeval converts to less then
724 * 1 tick.
725 *
726 * If ints have 32 bits, then the maximum value for any timeout in
727 * 10ms ticks is 248 days.
728 */
729int
730tvtohz_low(struct timeval *tv)
731{
732 int ticks;
733 long sec;
734
735 sec = tv->tv_sec;
736 if (sec <= INT_MAX / hz)
737 ticks = (int)(sec * hz + (u_long)tv->tv_usec / tick);
984263bc 738 else
984263bc 739 ticks = INT_MAX;
a94976ad 740 return (ticks);
984263bc
MD
741}
742
a94976ad 743
984263bc
MD
744/*
745 * Start profiling on a process.
746 *
747 * Kernel profiling passes proc0 which never exits and hence
748 * keeps the profile clock running constantly.
749 */
750void
88c4d2f6 751startprofclock(struct proc *p)
984263bc 752{
984263bc
MD
753 if ((p->p_flag & P_PROFIL) == 0) {
754 p->p_flag |= P_PROFIL;
88c4d2f6 755#if 0 /* XXX */
984263bc 756 if (++profprocs == 1 && stathz != 0) {
e43a034f 757 crit_enter();
6ad39cae 758 psdiv = psratio;
984263bc 759 setstatclockrate(profhz);
e43a034f 760 crit_exit();
984263bc 761 }
88c4d2f6 762#endif
984263bc
MD
763 }
764}
765
766/*
767 * Stop profiling on a process.
768 */
769void
88c4d2f6 770stopprofclock(struct proc *p)
984263bc 771{
984263bc
MD
772 if (p->p_flag & P_PROFIL) {
773 p->p_flag &= ~P_PROFIL;
88c4d2f6 774#if 0 /* XXX */
984263bc 775 if (--profprocs == 0 && stathz != 0) {
e43a034f 776 crit_enter();
6ad39cae 777 psdiv = 1;
984263bc 778 setstatclockrate(stathz);
e43a034f 779 crit_exit();
984263bc 780 }
984263bc 781#endif
984263bc
MD
782 }
783}
784
785/*
786 * Return information about system clocks.
787 */
788static int
789sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS)
790{
f5d21610 791 struct kinfo_clockinfo clkinfo;
984263bc
MD
792 /*
793 * Construct clockinfo structure.
794 */
f5d21610
JS
795 clkinfo.ci_hz = hz;
796 clkinfo.ci_tick = tick;
4026c000 797 clkinfo.ci_tickadj = ntp_default_tick_delta / 1000;
f5d21610
JS
798 clkinfo.ci_profhz = profhz;
799 clkinfo.ci_stathz = stathz ? stathz : hz;
984263bc
MD
800 return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req));
801}
802
803SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate, CTLTYPE_STRUCT|CTLFLAG_RD,
804 0, 0, sysctl_kern_clockrate, "S,clockinfo","");
805
984263bc
MD
806/*
807 * We have eight functions for looking at the clock, four for
808 * microseconds and four for nanoseconds. For each there is fast
809 * but less precise version "get{nano|micro}[up]time" which will
810 * return a time which is up to 1/HZ previous to the call, whereas
811 * the raw version "{nano|micro}[up]time" will return a timestamp
812 * which is as precise as possible. The "up" variants return the
813 * time relative to system boot, these are well suited for time
814 * interval measurements.
88c4d2f6
MD
815 *
816 * Each cpu independantly maintains the current time of day, so all
817 * we need to do to protect ourselves from changes is to do a loop
818 * check on the seconds field changing out from under us.
fad57d0e
MD
819 *
820 * The system timer maintains a 32 bit count and due to various issues
821 * it is possible for the calculated delta to occassionally exceed
044ee7c4
MD
822 * sys_cputimer->freq. If this occurs the sys_cputimer->freq64_nsec
823 * multiplication can easily overflow, so we deal with the case. For
824 * uniformity we deal with the case in the usec case too.
984263bc 825 */
984263bc
MD
826void
827getmicrouptime(struct timeval *tvp)
828{
88c4d2f6
MD
829 struct globaldata *gd = mycpu;
830 sysclock_t delta;
831
832 do {
833 tvp->tv_sec = gd->gd_time_seconds;
834 delta = gd->gd_hardclock.time - gd->gd_cpuclock_base;
835 } while (tvp->tv_sec != gd->gd_time_seconds);
fad57d0e 836
044ee7c4
MD
837 if (delta >= sys_cputimer->freq) {
838 tvp->tv_sec += delta / sys_cputimer->freq;
839 delta %= sys_cputimer->freq;
fad57d0e 840 }
044ee7c4 841 tvp->tv_usec = (sys_cputimer->freq64_usec * delta) >> 32;
88c4d2f6
MD
842 if (tvp->tv_usec >= 1000000) {
843 tvp->tv_usec -= 1000000;
844 ++tvp->tv_sec;
984263bc
MD
845 }
846}
847
848void
849getnanouptime(struct timespec *tsp)
850{
88c4d2f6
MD
851 struct globaldata *gd = mycpu;
852 sysclock_t delta;
853
854 do {
855 tsp->tv_sec = gd->gd_time_seconds;
856 delta = gd->gd_hardclock.time - gd->gd_cpuclock_base;
857 } while (tsp->tv_sec != gd->gd_time_seconds);
fad57d0e 858
044ee7c4
MD
859 if (delta >= sys_cputimer->freq) {
860 tsp->tv_sec += delta / sys_cputimer->freq;
861 delta %= sys_cputimer->freq;
984263bc 862 }
044ee7c4 863 tsp->tv_nsec = (sys_cputimer->freq64_nsec * delta) >> 32;
984263bc
MD
864}
865
866void
88c4d2f6 867microuptime(struct timeval *tvp)
984263bc 868{
88c4d2f6
MD
869 struct globaldata *gd = mycpu;
870 sysclock_t delta;
871
872 do {
873 tvp->tv_sec = gd->gd_time_seconds;
044ee7c4 874 delta = sys_cputimer->count() - gd->gd_cpuclock_base;
88c4d2f6 875 } while (tvp->tv_sec != gd->gd_time_seconds);
fad57d0e 876
044ee7c4
MD
877 if (delta >= sys_cputimer->freq) {
878 tvp->tv_sec += delta / sys_cputimer->freq;
879 delta %= sys_cputimer->freq;
984263bc 880 }
044ee7c4 881 tvp->tv_usec = (sys_cputimer->freq64_usec * delta) >> 32;
984263bc
MD
882}
883
884void
88c4d2f6 885nanouptime(struct timespec *tsp)
984263bc 886{
88c4d2f6
MD
887 struct globaldata *gd = mycpu;
888 sysclock_t delta;
889
890 do {
891 tsp->tv_sec = gd->gd_time_seconds;
044ee7c4 892 delta = sys_cputimer->count() - gd->gd_cpuclock_base;
88c4d2f6 893 } while (tsp->tv_sec != gd->gd_time_seconds);
fad57d0e 894
044ee7c4
MD
895 if (delta >= sys_cputimer->freq) {
896 tsp->tv_sec += delta / sys_cputimer->freq;
897 delta %= sys_cputimer->freq;
984263bc 898 }
044ee7c4 899 tsp->tv_nsec = (sys_cputimer->freq64_nsec * delta) >> 32;
984263bc
MD
900}
901
88c4d2f6
MD
902/*
903 * realtime routines
904 */
984263bc
MD
905
906void
88c4d2f6 907getmicrotime(struct timeval *tvp)
984263bc 908{
88c4d2f6 909 struct globaldata *gd = mycpu;
5eb5a6bc 910 struct timespec *bt;
88c4d2f6 911 sysclock_t delta;
984263bc 912
88c4d2f6
MD
913 do {
914 tvp->tv_sec = gd->gd_time_seconds;
915 delta = gd->gd_hardclock.time - gd->gd_cpuclock_base;
916 } while (tvp->tv_sec != gd->gd_time_seconds);
fad57d0e 917
044ee7c4
MD
918 if (delta >= sys_cputimer->freq) {
919 tvp->tv_sec += delta / sys_cputimer->freq;
920 delta %= sys_cputimer->freq;
fad57d0e 921 }
044ee7c4 922 tvp->tv_usec = (sys_cputimer->freq64_usec * delta) >> 32;
984263bc 923
5eb5a6bc
MD
924 bt = &basetime[basetime_index];
925 tvp->tv_sec += bt->tv_sec;
926 tvp->tv_usec += bt->tv_nsec / 1000;
88c4d2f6
MD
927 while (tvp->tv_usec >= 1000000) {
928 tvp->tv_usec -= 1000000;
929 ++tvp->tv_sec;
984263bc 930 }
984263bc
MD
931}
932
933void
88c4d2f6 934getnanotime(struct timespec *tsp)
984263bc 935{
88c4d2f6 936 struct globaldata *gd = mycpu;
5eb5a6bc 937 struct timespec *bt;
88c4d2f6 938 sysclock_t delta;
984263bc 939
88c4d2f6
MD
940 do {
941 tsp->tv_sec = gd->gd_time_seconds;
942 delta = gd->gd_hardclock.time - gd->gd_cpuclock_base;
943 } while (tsp->tv_sec != gd->gd_time_seconds);
fad57d0e 944
044ee7c4
MD
945 if (delta >= sys_cputimer->freq) {
946 tsp->tv_sec += delta / sys_cputimer->freq;
947 delta %= sys_cputimer->freq;
fad57d0e 948 }
044ee7c4 949 tsp->tv_nsec = (sys_cputimer->freq64_nsec * delta) >> 32;
984263bc 950
5eb5a6bc
MD
951 bt = &basetime[basetime_index];
952 tsp->tv_sec += bt->tv_sec;
953 tsp->tv_nsec += bt->tv_nsec;
88c4d2f6
MD
954 while (tsp->tv_nsec >= 1000000000) {
955 tsp->tv_nsec -= 1000000000;
956 ++tsp->tv_sec;
984263bc 957 }
984263bc
MD
958}
959
5eb5a6bc
MD
960static void
961getnanotime_nbt(struct timespec *nbt, struct timespec *tsp)
962{
963 struct globaldata *gd = mycpu;
964 sysclock_t delta;
965
966 do {
967 tsp->tv_sec = gd->gd_time_seconds;
968 delta = gd->gd_hardclock.time - gd->gd_cpuclock_base;
969 } while (tsp->tv_sec != gd->gd_time_seconds);
970
044ee7c4
MD
971 if (delta >= sys_cputimer->freq) {
972 tsp->tv_sec += delta / sys_cputimer->freq;
973 delta %= sys_cputimer->freq;
5eb5a6bc 974 }
044ee7c4 975 tsp->tv_nsec = (sys_cputimer->freq64_nsec * delta) >> 32;
5eb5a6bc
MD
976
977 tsp->tv_sec += nbt->tv_sec;
978 tsp->tv_nsec += nbt->tv_nsec;
979 while (tsp->tv_nsec >= 1000000000) {
980 tsp->tv_nsec -= 1000000000;
981 ++tsp->tv_sec;
982 }
983}
984
985
88c4d2f6
MD
986void
987microtime(struct timeval *tvp)
984263bc 988{
88c4d2f6 989 struct globaldata *gd = mycpu;
5eb5a6bc 990 struct timespec *bt;
88c4d2f6 991 sysclock_t delta;
984263bc 992
88c4d2f6
MD
993 do {
994 tvp->tv_sec = gd->gd_time_seconds;
044ee7c4 995 delta = sys_cputimer->count() - gd->gd_cpuclock_base;
88c4d2f6 996 } while (tvp->tv_sec != gd->gd_time_seconds);
fad57d0e 997
044ee7c4
MD
998 if (delta >= sys_cputimer->freq) {
999 tvp->tv_sec += delta / sys_cputimer->freq;
1000 delta %= sys_cputimer->freq;
fad57d0e 1001 }
044ee7c4 1002 tvp->tv_usec = (sys_cputimer->freq64_usec * delta) >> 32;
984263bc 1003
5eb5a6bc
MD
1004 bt = &basetime[basetime_index];
1005 tvp->tv_sec += bt->tv_sec;
1006 tvp->tv_usec += bt->tv_nsec / 1000;
88c4d2f6
MD
1007 while (tvp->tv_usec >= 1000000) {
1008 tvp->tv_usec -= 1000000;
1009 ++tvp->tv_sec;
984263bc 1010 }
984263bc
MD
1011}
1012
88c4d2f6
MD
1013void
1014nanotime(struct timespec *tsp)
1015{
1016 struct globaldata *gd = mycpu;
5eb5a6bc 1017 struct timespec *bt;
88c4d2f6 1018 sysclock_t delta;
984263bc 1019
88c4d2f6
MD
1020 do {
1021 tsp->tv_sec = gd->gd_time_seconds;
044ee7c4 1022 delta = sys_cputimer->count() - gd->gd_cpuclock_base;
88c4d2f6 1023 } while (tsp->tv_sec != gd->gd_time_seconds);
fad57d0e 1024
044ee7c4
MD
1025 if (delta >= sys_cputimer->freq) {
1026 tsp->tv_sec += delta / sys_cputimer->freq;
1027 delta %= sys_cputimer->freq;
fad57d0e 1028 }
044ee7c4 1029 tsp->tv_nsec = (sys_cputimer->freq64_nsec * delta) >> 32;
984263bc 1030
5eb5a6bc
MD
1031 bt = &basetime[basetime_index];
1032 tsp->tv_sec += bt->tv_sec;
1033 tsp->tv_nsec += bt->tv_nsec;
88c4d2f6
MD
1034 while (tsp->tv_nsec >= 1000000000) {
1035 tsp->tv_nsec -= 1000000000;
1036 ++tsp->tv_sec;
984263bc 1037 }
984263bc
MD
1038}
1039
25b804e7
MD
1040/*
1041 * note: this is not exactly synchronized with real time. To do that we
1042 * would have to do what microtime does and check for a nanoseconds overflow.
1043 */
1044time_t
1045get_approximate_time_t(void)
1046{
1047 struct globaldata *gd = mycpu;
5eb5a6bc
MD
1048 struct timespec *bt;
1049
1050 bt = &basetime[basetime_index];
1051 return(gd->gd_time_seconds + bt->tv_sec);
25b804e7
MD
1052}
1053
984263bc
MD
1054int
1055pps_ioctl(u_long cmd, caddr_t data, struct pps_state *pps)
1056{
1057 pps_params_t *app;
1058 struct pps_fetch_args *fapi;
1059#ifdef PPS_SYNC
1060 struct pps_kcbind_args *kapi;
1061#endif
1062
1063 switch (cmd) {
1064 case PPS_IOC_CREATE:
1065 return (0);
1066 case PPS_IOC_DESTROY:
1067 return (0);
1068 case PPS_IOC_SETPARAMS:
1069 app = (pps_params_t *)data;
1070 if (app->mode & ~pps->ppscap)
1071 return (EINVAL);
1072 pps->ppsparam = *app;
1073 return (0);
1074 case PPS_IOC_GETPARAMS:
1075 app = (pps_params_t *)data;
1076 *app = pps->ppsparam;
1077 app->api_version = PPS_API_VERS_1;
1078 return (0);
1079 case PPS_IOC_GETCAP:
1080 *(int*)data = pps->ppscap;
1081 return (0);
1082 case PPS_IOC_FETCH:
1083 fapi = (struct pps_fetch_args *)data;
1084 if (fapi->tsformat && fapi->tsformat != PPS_TSFMT_TSPEC)
1085 return (EINVAL);
1086 if (fapi->timeout.tv_sec || fapi->timeout.tv_nsec)
1087 return (EOPNOTSUPP);
1088 pps->ppsinfo.current_mode = pps->ppsparam.mode;
1089 fapi->pps_info_buf = pps->ppsinfo;
1090 return (0);
1091 case PPS_IOC_KCBIND:
1092#ifdef PPS_SYNC
1093 kapi = (struct pps_kcbind_args *)data;
1094 /* XXX Only root should be able to do this */
1095 if (kapi->tsformat && kapi->tsformat != PPS_TSFMT_TSPEC)
1096 return (EINVAL);
1097 if (kapi->kernel_consumer != PPS_KC_HARDPPS)
1098 return (EINVAL);
1099 if (kapi->edge & ~pps->ppscap)
1100 return (EINVAL);
1101 pps->kcmode = kapi->edge;
1102 return (0);
1103#else
1104 return (EOPNOTSUPP);
1105#endif
1106 default:
1107 return (ENOTTY);
1108 }
1109}
1110
1111void
1112pps_init(struct pps_state *pps)
1113{
1114 pps->ppscap |= PPS_TSFMT_TSPEC;
1115 if (pps->ppscap & PPS_CAPTUREASSERT)
1116 pps->ppscap |= PPS_OFFSETASSERT;
1117 if (pps->ppscap & PPS_CAPTURECLEAR)
1118 pps->ppscap |= PPS_OFFSETCLEAR;
1119}
1120
1121void
88c4d2f6 1122pps_event(struct pps_state *pps, sysclock_t count, int event)
984263bc 1123{
88c4d2f6
MD
1124 struct globaldata *gd;
1125 struct timespec *tsp;
1126 struct timespec *osp;
5eb5a6bc 1127 struct timespec *bt;
88c4d2f6
MD
1128 struct timespec ts;
1129 sysclock_t *pcount;
1130#ifdef PPS_SYNC
1131 sysclock_t tcount;
1132#endif
1133 sysclock_t delta;
1134 pps_seq_t *pseq;
1135 int foff;
1136 int fhard;
1137
1138 gd = mycpu;
984263bc
MD
1139
1140 /* Things would be easier with arrays... */
1141 if (event == PPS_CAPTUREASSERT) {
1142 tsp = &pps->ppsinfo.assert_timestamp;
1143 osp = &pps->ppsparam.assert_offset;
1144 foff = pps->ppsparam.mode & PPS_OFFSETASSERT;
1145 fhard = pps->kcmode & PPS_CAPTUREASSERT;
1146 pcount = &pps->ppscount[0];
1147 pseq = &pps->ppsinfo.assert_sequence;
1148 } else {
1149 tsp = &pps->ppsinfo.clear_timestamp;
1150 osp = &pps->ppsparam.clear_offset;
1151 foff = pps->ppsparam.mode & PPS_OFFSETCLEAR;
1152 fhard = pps->kcmode & PPS_CAPTURECLEAR;
1153 pcount = &pps->ppscount[1];
1154 pseq = &pps->ppsinfo.clear_sequence;
1155 }
1156
984263bc
MD
1157 /* Nothing really happened */
1158 if (*pcount == count)
1159 return;
1160
1161 *pcount = count;
1162
88c4d2f6
MD
1163 do {
1164 ts.tv_sec = gd->gd_time_seconds;
1165 delta = count - gd->gd_cpuclock_base;
1166 } while (ts.tv_sec != gd->gd_time_seconds);
fad57d0e 1167
044ee7c4
MD
1168 if (delta >= sys_cputimer->freq) {
1169 ts.tv_sec += delta / sys_cputimer->freq;
1170 delta %= sys_cputimer->freq;
88c4d2f6 1171 }
044ee7c4 1172 ts.tv_nsec = (sys_cputimer->freq64_nsec * delta) >> 32;
5eb5a6bc
MD
1173 bt = &basetime[basetime_index];
1174 ts.tv_sec += bt->tv_sec;
1175 ts.tv_nsec += bt->tv_nsec;
88c4d2f6
MD
1176 while (ts.tv_nsec >= 1000000000) {
1177 ts.tv_nsec -= 1000000000;
1178 ++ts.tv_sec;
984263bc 1179 }
984263bc
MD
1180
1181 (*pseq)++;
1182 *tsp = ts;
1183
1184 if (foff) {
1185 timespecadd(tsp, osp);
1186 if (tsp->tv_nsec < 0) {
1187 tsp->tv_nsec += 1000000000;
1188 tsp->tv_sec -= 1;
1189 }
1190 }
1191#ifdef PPS_SYNC
1192 if (fhard) {
1193 /* magic, at its best... */
1194 tcount = count - pps->ppscount[2];
1195 pps->ppscount[2] = count;
044ee7c4
MD
1196 if (tcount >= sys_cputimer->freq) {
1197 delta = (1000000000 * (tcount / sys_cputimer->freq) +
1198 sys_cputimer->freq64_nsec *
1199 (tcount % sys_cputimer->freq)) >> 32;
fad57d0e 1200 } else {
044ee7c4 1201 delta = (sys_cputimer->freq64_nsec * tcount) >> 32;
fad57d0e 1202 }
984263bc
MD
1203 hardpps(tsp, delta);
1204 }
1205#endif
1206}
88c4d2f6 1207