Remove the asynchronous system call interface sendsys/waitsys. It was an
[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 $
5681a38a 73 * $DragonFly: src/sys/kern/kern_clock.c,v 1.52 2006/06/01 16:49:59 dillon Exp $
984263bc
MD
74 */
75
76#include "opt_ntp.h"
2b71c8f1 77#include "opt_polling.h"
984263bc
MD
78
79#include <sys/param.h>
80#include <sys/systm.h>
984263bc
MD
81#include <sys/callout.h>
82#include <sys/kernel.h>
f5d21610 83#include <sys/kinfo.h>
984263bc
MD
84#include <sys/proc.h>
85#include <sys/malloc.h>
86#include <sys/resourcevar.h>
87#include <sys/signalvar.h>
88#include <sys/timex.h>
89#include <sys/timepps.h>
90#include <vm/vm.h>
91#include <sys/lock.h>
92#include <vm/pmap.h>
93#include <vm/vm_map.h>
5ffd1608 94#include <vm/vm_extern.h>
984263bc 95#include <sys/sysctl.h>
2689779e 96#include <sys/thread2.h>
984263bc
MD
97
98#include <machine/cpu.h>
99#include <machine/limits.h>
100#include <machine/smp.h>
101
102#ifdef GPROF
103#include <sys/gmon.h>
104#endif
105
106#ifdef DEVICE_POLLING
107extern void init_device_poll(void);
3e61f60e 108#endif
984263bc 109
402ed7e1 110static void initclocks (void *dummy);
984263bc
MD
111SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL)
112
6ad39cae
MD
113/*
114 * Some of these don't belong here, but it's easiest to concentrate them.
9eea7f0c 115 * Note that cpu_time counts in microseconds, but most userland programs
6ad39cae
MD
116 * just compare relative times against the total by delta.
117 */
9eea7f0c
HP
118struct kinfo_cputime cputime_percpu[MAXCPU];
119#ifdef SMP
120static int
121sysctl_cputime(SYSCTL_HANDLER_ARGS)
122{
123 int cpu, error = 0;
124 size_t size = sizeof(struct kinfo_cputime);
125
126 for (cpu = 0; cpu < ncpus; ++cpu) {
127 if ((error = SYSCTL_OUT(req, &cputime_percpu[cpu], size)))
128 break;
129 }
984263bc 130
9eea7f0c
HP
131 return (error);
132}
133SYSCTL_PROC(_kern, OID_AUTO, cputime, (CTLTYPE_OPAQUE|CTLFLAG_RD), 0, 0,
134 sysctl_cputime, "S,kinfo_cputime", "CPU time statistics");
135#else
136SYSCTL_STRUCT(_kern, OID_AUTO, cputime, CTLFLAG_RD, &cpu_time, kinfo_cputime,
137 "CPU time statistics");
138#endif
984263bc 139
88c4d2f6
MD
140/*
141 * boottime is used to calculate the 'real' uptime. Do not confuse this with
142 * microuptime(). microtime() is not drift compensated. The real uptime
60b2809b
MD
143 * with compensation is nanotime() - bootime. boottime is recalculated
144 * whenever the real time is set based on the compensated elapsed time
145 * in seconds (gd->gd_time_seconds).
88c4d2f6 146 *
88c4d2f6
MD
147 * The gd_time_seconds and gd_cpuclock_base fields remain fairly monotonic.
148 * Slight adjustments to gd_cpuclock_base are made to phase-lock it to
149 * the real time.
150 */
151struct timespec boottime; /* boot time (realtime) for reference only */
88c4d2f6 152time_t time_second; /* read-only 'passive' uptime in seconds */
984263bc 153
5eb5a6bc
MD
154/*
155 * basetime is used to calculate the compensated real time of day. The
156 * basetime can be modified on a per-tick basis by the adjtime(),
157 * ntp_adjtime(), and sysctl-based time correction APIs.
158 *
159 * Note that frequency corrections can also be made by adjusting
160 * gd_cpuclock_base.
161 *
162 * basetime is a tail-chasing FIFO, updated only by cpu #0. The FIFO is
163 * used on both SMP and UP systems to avoid MP races between cpu's and
164 * interrupt races on UP systems.
165 */
166#define BASETIME_ARYSIZE 16
167#define BASETIME_ARYMASK (BASETIME_ARYSIZE - 1)
168static struct timespec basetime[BASETIME_ARYSIZE];
169static volatile int basetime_index;
170
171static int
172sysctl_get_basetime(SYSCTL_HANDLER_ARGS)
173{
174 struct timespec *bt;
175 int error;
35238fa5 176 int index;
5eb5a6bc 177
35238fa5
MD
178 /*
179 * Because basetime data and index may be updated by another cpu,
180 * a load fence is required to ensure that the data we read has
181 * not been speculatively read relative to a possibly updated index.
182 */
183 index = basetime_index;
184 cpu_lfence();
185 bt = &basetime[index];
08f95c49 186 error = SYSCTL_OUT(req, bt, sizeof(*bt));
5eb5a6bc
MD
187 return (error);
188}
189
984263bc 190SYSCTL_STRUCT(_kern, KERN_BOOTTIME, boottime, CTLFLAG_RD,
08f95c49 191 &boottime, timespec, "System boottime");
5eb5a6bc 192SYSCTL_PROC(_kern, OID_AUTO, basetime, CTLTYPE_STRUCT|CTLFLAG_RD, 0, 0,
08f95c49 193 sysctl_get_basetime, "S,timespec", "System basetime");
984263bc 194
88c4d2f6
MD
195static void hardclock(systimer_t info, struct intrframe *frame);
196static void statclock(systimer_t info, struct intrframe *frame);
197static void schedclock(systimer_t info, struct intrframe *frame);
5eb5a6bc 198static void getnanotime_nbt(struct timespec *nbt, struct timespec *tsp);
88c4d2f6
MD
199
200int ticks; /* system master ticks at hz */
da3639ef 201int clocks_running; /* tsleep/timeout clocks operational */
88c4d2f6
MD
202int64_t nsec_adj; /* ntpd per-tick adjustment in nsec << 32 */
203int64_t nsec_acc; /* accumulator */
984263bc 204
4026c000
JS
205/* NTPD time correction fields */
206int64_t ntp_tick_permanent; /* per-tick adjustment in nsec << 32 */
207int64_t ntp_tick_acc; /* accumulator for per-tick adjustment */
208int64_t ntp_delta; /* one-time correction in nsec */
209int64_t ntp_big_delta = 1000000000;
210int32_t ntp_tick_delta; /* current adjustment rate */
211int32_t ntp_default_tick_delta; /* adjustment rate for ntp_delta */
48590578
JS
212time_t ntp_leap_second; /* time of next leap second */
213int ntp_leap_insert; /* whether to insert or remove a second */
4026c000 214
984263bc 215/*
88c4d2f6 216 * Finish initializing clock frequencies and start all clocks running.
984263bc 217 */
88c4d2f6
MD
218/* ARGSUSED*/
219static void
220initclocks(void *dummy)
984263bc 221{
88c4d2f6
MD
222 cpu_initclocks();
223#ifdef DEVICE_POLLING
224 init_device_poll();
225#endif
226 /*psratio = profhz / stathz;*/
227 initclocks_pcpu();
da3639ef 228 clocks_running = 1;
984263bc
MD
229}
230
88c4d2f6
MD
231/*
232 * Called on a per-cpu basis
233 */
234void
235initclocks_pcpu(void)
236{
237 struct globaldata *gd = mycpu;
984263bc 238
88c4d2f6
MD
239 crit_enter();
240 if (gd->gd_cpuid == 0) {
241 gd->gd_time_seconds = 1;
044ee7c4 242 gd->gd_cpuclock_base = sys_cputimer->count();
88c4d2f6
MD
243 } else {
244 /* XXX */
245 gd->gd_time_seconds = globaldata_find(0)->gd_time_seconds;
246 gd->gd_cpuclock_base = globaldata_find(0)->gd_cpuclock_base;
247 }
0d1dffdf
MD
248
249 /*
250 * Use a non-queued periodic systimer to prevent multiple ticks from
251 * building up if the sysclock jumps forward (8254 gets reset). The
252 * sysclock will never jump backwards. Our time sync is based on
253 * the actual sysclock, not the ticks count.
254 */
255 systimer_init_periodic_nq(&gd->gd_hardclock, hardclock, NULL, hz);
256 systimer_init_periodic_nq(&gd->gd_statclock, statclock, NULL, stathz);
88c4d2f6 257 /* XXX correct the frequency for scheduler / estcpu tests */
0d1dffdf 258 systimer_init_periodic_nq(&gd->gd_schedclock, schedclock,
8478264a 259 NULL, ESTCPUFREQ);
88c4d2f6
MD
260 crit_exit();
261}
984263bc
MD
262
263/*
88c4d2f6
MD
264 * This sets the current real time of day. Timespecs are in seconds and
265 * nanoseconds. We do not mess with gd_time_seconds and gd_cpuclock_base,
266 * instead we adjust basetime so basetime + gd_* results in the current
267 * time of day. This way the gd_* fields are guarenteed to represent
268 * a monotonically increasing 'uptime' value.
5eb5a6bc
MD
269 *
270 * When set_timeofday() is called from userland, the system call forces it
271 * onto cpu #0 since only cpu #0 can update basetime_index.
984263bc 272 */
88c4d2f6
MD
273void
274set_timeofday(struct timespec *ts)
275{
5eb5a6bc
MD
276 struct timespec *nbt;
277 int ni;
984263bc 278
88c4d2f6
MD
279 /*
280 * XXX SMP / non-atomic basetime updates
281 */
282 crit_enter();
5eb5a6bc
MD
283 ni = (basetime_index + 1) & BASETIME_ARYMASK;
284 nbt = &basetime[ni];
285 nanouptime(nbt);
286 nbt->tv_sec = ts->tv_sec - nbt->tv_sec;
287 nbt->tv_nsec = ts->tv_nsec - nbt->tv_nsec;
288 if (nbt->tv_nsec < 0) {
289 nbt->tv_nsec += 1000000000;
290 --nbt->tv_sec;
88c4d2f6 291 }
a81931cc
MD
292
293 /*
294 * Note that basetime diverges from boottime as the clock drift is
295 * compensated for, so we cannot do away with boottime. When setting
296 * the absolute time of day the drift is 0 (for an instant) and we
297 * can simply assign boottime to basetime.
298 *
299 * Note that nanouptime() is based on gd_time_seconds which is drift
300 * compensated up to a point (it is guarenteed to remain monotonically
301 * increasing). gd_time_seconds is thus our best uptime guess and
302 * suitable for use in the boottime calculation. It is already taken
303 * into account in the basetime calculation above.
304 */
5eb5a6bc 305 boottime.tv_sec = nbt->tv_sec;
4026c000 306 ntp_delta = 0;
5eb5a6bc
MD
307
308 /*
35238fa5
MD
309 * We now have a new basetime, make sure all other cpus have it,
310 * then update the index.
5eb5a6bc 311 */
35238fa5 312 cpu_sfence();
5eb5a6bc
MD
313 basetime_index = ni;
314
88c4d2f6
MD
315 crit_exit();
316}
317
984263bc 318/*
88c4d2f6
MD
319 * Each cpu has its own hardclock, but we only increments ticks and softticks
320 * on cpu #0.
321 *
322 * NOTE! systimer! the MP lock might not be held here. We can only safely
323 * manipulate objects owned by the current cpu.
984263bc 324 */
984263bc 325static void
88c4d2f6 326hardclock(systimer_t info, struct intrframe *frame)
984263bc 327{
88c4d2f6
MD
328 sysclock_t cputicks;
329 struct proc *p;
330 struct pstats *pstats;
331 struct globaldata *gd = mycpu;
984263bc
MD
332
333 /*
88c4d2f6
MD
334 * Realtime updates are per-cpu. Note that timer corrections as
335 * returned by microtime() and friends make an additional adjustment
336 * using a system-wise 'basetime', but the running time is always
337 * taken from the per-cpu globaldata area. Since the same clock
338 * is distributing (XXX SMP) to all cpus, the per-cpu timebases
339 * stay in synch.
340 *
341 * Note that we never allow info->time (aka gd->gd_hardclock.time)
fad57d0e
MD
342 * to reverse index gd_cpuclock_base, but that it is possible for
343 * it to temporarily get behind in the seconds if something in the
344 * system locks interrupts for a long period of time. Since periodic
345 * timers count events, though everything should resynch again
346 * immediately.
984263bc 347 */
88c4d2f6 348 cputicks = info->time - gd->gd_cpuclock_base;
044ee7c4 349 if (cputicks >= sys_cputimer->freq) {
88c4d2f6 350 ++gd->gd_time_seconds;
044ee7c4 351 gd->gd_cpuclock_base += sys_cputimer->freq;
88c4d2f6 352 }
984263bc
MD
353
354 /*
92b561b7
MD
355 * The system-wide ticks counter and NTP related timedelta/tickdelta
356 * adjustments only occur on cpu #0. NTP adjustments are accomplished
357 * by updating basetime.
984263bc 358 */
88c4d2f6 359 if (gd->gd_cpuid == 0) {
5eb5a6bc 360 struct timespec *nbt;
88c4d2f6
MD
361 struct timespec nts;
362 int leap;
5eb5a6bc 363 int ni;
984263bc 364
88c4d2f6 365 ++ticks;
984263bc 366
88c4d2f6
MD
367#if 0
368 if (tco->tc_poll_pps)
369 tco->tc_poll_pps(tco);
370#endif
5eb5a6bc 371
88c4d2f6 372 /*
5eb5a6bc
MD
373 * Calculate the new basetime index. We are in a critical section
374 * on cpu #0 and can safely play with basetime_index. Start
375 * with the current basetime and then make adjustments.
376 */
377 ni = (basetime_index + 1) & BASETIME_ARYMASK;
378 nbt = &basetime[ni];
379 *nbt = basetime[basetime_index];
380
381 /*
382 * Apply adjtime corrections. (adjtime() API)
383 *
384 * adjtime() only runs on cpu #0 so our critical section is
385 * sufficient to access these variables.
88c4d2f6 386 */
4026c000 387 if (ntp_delta != 0) {
5eb5a6bc 388 nbt->tv_nsec += ntp_tick_delta;
4026c000
JS
389 ntp_delta -= ntp_tick_delta;
390 if ((ntp_delta > 0 && ntp_delta < ntp_tick_delta) ||
391 (ntp_delta < 0 && ntp_delta > ntp_tick_delta)) {
5eb5a6bc 392 ntp_tick_delta = ntp_delta;
4026c000
JS
393 }
394 }
395
5eb5a6bc
MD
396 /*
397 * Apply permanent frequency corrections. (sysctl API)
398 */
4026c000
JS
399 if (ntp_tick_permanent != 0) {
400 ntp_tick_acc += ntp_tick_permanent;
401 if (ntp_tick_acc >= (1LL << 32)) {
5eb5a6bc 402 nbt->tv_nsec += ntp_tick_acc >> 32;
331bc6f8 403 ntp_tick_acc -= (ntp_tick_acc >> 32) << 32;
4026c000 404 } else if (ntp_tick_acc <= -(1LL << 32)) {
331bc6f8 405 /* Negate ntp_tick_acc to avoid shifting the sign bit. */
5eb5a6bc 406 nbt->tv_nsec -= (-ntp_tick_acc) >> 32;
331bc6f8 407 ntp_tick_acc += ((-ntp_tick_acc) >> 32) << 32;
4026c000
JS
408 }
409 }
410
5eb5a6bc
MD
411 if (nbt->tv_nsec >= 1000000000) {
412 nbt->tv_sec++;
413 nbt->tv_nsec -= 1000000000;
414 } else if (nbt->tv_nsec < 0) {
415 nbt->tv_sec--;
416 nbt->tv_nsec += 1000000000;
88c4d2f6
MD
417 }
418
419 /*
5eb5a6bc 420 * Another per-tick compensation. (for ntp_adjtime() API)
88c4d2f6 421 */
5eb5a6bc 422 if (nsec_adj != 0) {
88c4d2f6
MD
423 nsec_acc += nsec_adj;
424 if (nsec_acc >= 0x100000000LL) {
5eb5a6bc 425 nbt->tv_nsec += nsec_acc >> 32;
88c4d2f6
MD
426 nsec_acc = (nsec_acc & 0xFFFFFFFFLL);
427 } else if (nsec_acc <= -0x100000000LL) {
5eb5a6bc 428 nbt->tv_nsec -= -nsec_acc >> 32;
88c4d2f6
MD
429 nsec_acc = -(-nsec_acc & 0xFFFFFFFFLL);
430 }
5eb5a6bc
MD
431 if (nbt->tv_nsec >= 1000000000) {
432 nbt->tv_nsec -= 1000000000;
433 ++nbt->tv_sec;
434 } else if (nbt->tv_nsec < 0) {
435 nbt->tv_nsec += 1000000000;
436 --nbt->tv_sec;
437 }
438 }
439
440 /************************************************************
441 * LEAP SECOND CORRECTION *
442 ************************************************************
443 *
444 * Taking into account all the corrections made above, figure
445 * out the new real time. If the seconds field has changed
446 * then apply any pending leap-second corrections.
447 */
448 getnanotime_nbt(nbt, &nts);
449
32040d57
MD
450 if (time_second != nts.tv_sec) {
451 /*
452 * Apply leap second (sysctl API). Adjust nts for changes
453 * so we do not have to call getnanotime_nbt again.
454 */
455 if (ntp_leap_second) {
456 if (ntp_leap_second == nts.tv_sec) {
457 if (ntp_leap_insert) {
458 nbt->tv_sec++;
459 nts.tv_sec++;
460 } else {
461 nbt->tv_sec--;
462 nts.tv_sec--;
463 }
5eb5a6bc 464 ntp_leap_second--;
32040d57 465 }
88c4d2f6 466 }
88c4d2f6 467
32040d57
MD
468 /*
469 * Apply leap second (ntp_adjtime() API), calculate a new
470 * nsec_adj field. ntp_update_second() returns nsec_adj
471 * as a per-second value but we need it as a per-tick value.
472 */
88c4d2f6 473 leap = ntp_update_second(time_second, &nsec_adj);
88c4d2f6 474 nsec_adj /= hz;
32040d57
MD
475 nbt->tv_sec += leap;
476 nts.tv_sec += leap;
477
478 /*
479 * Update the time_second 'approximate time' global.
480 */
481 time_second = nts.tv_sec;
88c4d2f6 482 }
5eb5a6bc
MD
483
484 /*
485 * Finally, our new basetime is ready to go live!
486 */
35238fa5 487 cpu_sfence();
5eb5a6bc 488 basetime_index = ni;
5ffd1608
MD
489
490 /*
491 * Figure out how badly the system is starved for memory
492 */
493 vm_fault_ratecheck();
88c4d2f6
MD
494 }
495
92b561b7
MD
496 /*
497 * softticks are handled for all cpus
498 */
499 hardclock_softtick(gd);
500
88c4d2f6
MD
501 /*
502 * ITimer handling is per-tick, per-cpu. I don't think psignal()
503 * is mpsafe on curproc, so XXX get the mplock.
504 */
505 if ((p = curproc) != NULL && try_mplock()) {
984263bc 506 pstats = p->p_stats;
88c4d2f6 507 if (frame && CLKF_USERMODE(frame) &&
93328593
SS
508 timevalisset(&p->p_timer[ITIMER_VIRTUAL].it_value) &&
509 itimerdecr(&p->p_timer[ITIMER_VIRTUAL], tick) == 0)
984263bc 510 psignal(p, SIGVTALRM);
93328593
SS
511 if (timevalisset(&p->p_timer[ITIMER_PROF].it_value) &&
512 itimerdecr(&p->p_timer[ITIMER_PROF], tick) == 0)
984263bc 513 psignal(p, SIGPROF);
88c4d2f6 514 rel_mplock();
984263bc 515 }
604e1e09 516 setdelayed();
88c4d2f6 517}
984263bc 518
88c4d2f6
MD
519/*
520 * The statistics clock typically runs at a 125Hz rate, and is intended
521 * to be frequency offset from the hardclock (typ 100Hz). It is per-cpu.
522 *
523 * NOTE! systimer! the MP lock might not be held here. We can only safely
524 * manipulate objects owned by the current cpu.
525 *
526 * The stats clock is responsible for grabbing a profiling sample.
527 * Most of the statistics are only used by user-level statistics programs.
528 * The main exceptions are p->p_uticks, p->p_sticks, p->p_iticks, and
529 * p->p_estcpu.
530 *
531 * Like the other clocks, the stat clock is called from what is effectively
532 * a fast interrupt, so the context should be the thread/process that got
533 * interrupted.
534 */
535static void
536statclock(systimer_t info, struct intrframe *frame)
537{
538#ifdef GPROF
539 struct gmonparam *g;
540 int i;
984263bc 541#endif
88c4d2f6
MD
542 thread_t td;
543 struct proc *p;
544 int bump;
545 struct timeval tv;
546 struct timeval *stv;
984263bc
MD
547
548 /*
88c4d2f6 549 * How big was our timeslice relative to the last time?
984263bc 550 */
88c4d2f6
MD
551 microuptime(&tv); /* mpsafe */
552 stv = &mycpu->gd_stattv;
553 if (stv->tv_sec == 0) {
554 bump = 1;
555 } else {
556 bump = tv.tv_usec - stv->tv_usec +
557 (tv.tv_sec - stv->tv_sec) * 1000000;
558 if (bump < 0)
559 bump = 0;
560 if (bump > 1000000)
561 bump = 1000000;
562 }
563 *stv = tv;
984263bc 564
88c4d2f6
MD
565 td = curthread;
566 p = td->td_proc;
984263bc 567
88c4d2f6
MD
568 if (frame && CLKF_USERMODE(frame)) {
569 /*
570 * Came from userland, handle user time and deal with
571 * possible process.
572 */
573 if (p && (p->p_flag & P_PROFIL))
574 addupc_intr(p, CLKF_PC(frame), 1);
575 td->td_uticks += bump;
984263bc 576
88c4d2f6
MD
577 /*
578 * Charge the time as appropriate
579 */
580 if (p && p->p_nice > NZERO)
9eea7f0c 581 cpu_time.cp_nice += bump;
88c4d2f6 582 else
9eea7f0c 583 cpu_time.cp_user += bump;
88c4d2f6
MD
584 } else {
585#ifdef GPROF
586 /*
587 * Kernel statistics are just like addupc_intr, only easier.
588 */
589 g = &_gmonparam;
590 if (g->state == GMON_PROF_ON && frame) {
591 i = CLKF_PC(frame) - g->lowpc;
592 if (i < g->textsize) {
593 i /= HISTFRACTION * sizeof(*g->kcount);
594 g->kcount[i]++;
595 }
596 }
597#endif
598 /*
599 * Came from kernel mode, so we were:
600 * - handling an interrupt,
601 * - doing syscall or trap work on behalf of the current
602 * user process, or
603 * - spinning in the idle loop.
604 * Whichever it is, charge the time as appropriate.
605 * Note that we charge interrupts to the current process,
606 * regardless of whether they are ``for'' that process,
607 * so that we know how much of its real time was spent
608 * in ``non-process'' (i.e., interrupt) work.
609 *
610 * XXX assume system if frame is NULL. A NULL frame
e43a034f 611 * can occur if ipi processing is done from a crit_exit().
88c4d2f6
MD
612 */
613 if (frame && CLKF_INTR(frame))
614 td->td_iticks += bump;
615 else
616 td->td_sticks += bump;
617
618 if (frame && CLKF_INTR(frame)) {
9eea7f0c 619 cpu_time.cp_intr += bump;
88c4d2f6
MD
620 } else {
621 if (td == &mycpu->gd_idlethread)
9eea7f0c 622 cpu_time.cp_idle += bump;
88c4d2f6 623 else
9eea7f0c 624 cpu_time.cp_sys += bump;
88c4d2f6
MD
625 }
626 }
627}
628
629/*
dcc99b62 630 * The scheduler clock typically runs at a 50Hz rate. NOTE! systimer,
88c4d2f6
MD
631 * the MP lock might not be held. We can safely manipulate parts of curproc
632 * but that's about it.
dcc99b62
MD
633 *
634 * Each cpu has its own scheduler clock.
88c4d2f6
MD
635 */
636static void
637schedclock(systimer_t info, struct intrframe *frame)
638{
553ea3c8 639 struct lwp *lp;
88c4d2f6
MD
640 struct pstats *pstats;
641 struct rusage *ru;
642 struct vmspace *vm;
643 long rss;
644
553ea3c8 645 if ((lp = lwkt_preempted_proc()) != NULL) {
dcc99b62
MD
646 /*
647 * Account for cpu time used and hit the scheduler. Note
648 * that this call MUST BE MP SAFE, and the BGL IS NOT HELD
649 * HERE.
650 */
553ea3c8 651 ++lp->lwp_cpticks;
5681a38a
MD
652 lp->lwp_proc->p_usched->schedulerclock(lp, info->periodic,
653 info->time);
dcc99b62 654 }
553ea3c8 655 if ((lp = curthread->td_lwp) != NULL) {
dcc99b62
MD
656 /*
657 * Update resource usage integrals and maximums.
658 */
553ea3c8 659 if ((pstats = lp->lwp_stats) != NULL &&
88c4d2f6 660 (ru = &pstats->p_ru) != NULL &&
553ea3c8 661 (vm = lp->lwp_proc->p_vmspace) != NULL) {
88c4d2f6
MD
662 ru->ru_ixrss += pgtok(vm->vm_tsize);
663 ru->ru_idrss += pgtok(vm->vm_dsize);
664 ru->ru_isrss += pgtok(vm->vm_ssize);
665 rss = pgtok(vmspace_resident_count(vm));
666 if (ru->ru_maxrss < rss)
667 ru->ru_maxrss = rss;
668 }
b68b7282 669 }
984263bc
MD
670}
671
672/*
a94976ad
MD
673 * Compute number of ticks for the specified amount of time. The
674 * return value is intended to be used in a clock interrupt timed
675 * operation and guarenteed to meet or exceed the requested time.
676 * If the representation overflows, return INT_MAX. The minimum return
677 * value is 1 ticks and the function will average the calculation up.
678 * If any value greater then 0 microseconds is supplied, a value
679 * of at least 2 will be returned to ensure that a near-term clock
680 * interrupt does not cause the timeout to occur (degenerately) early.
681 *
682 * Note that limit checks must take into account microseconds, which is
683 * done simply by using the smaller signed long maximum instead of
684 * the unsigned long maximum.
685 *
686 * If ints have 32 bits, then the maximum value for any timeout in
687 * 10ms ticks is 248 days.
984263bc
MD
688 */
689int
a94976ad 690tvtohz_high(struct timeval *tv)
984263bc 691{
a94976ad 692 int ticks;
1fd87d54 693 long sec, usec;
984263bc 694
984263bc
MD
695 sec = tv->tv_sec;
696 usec = tv->tv_usec;
697 if (usec < 0) {
698 sec--;
699 usec += 1000000;
700 }
701 if (sec < 0) {
702#ifdef DIAGNOSTIC
703 if (usec > 0) {
704 sec++;
705 usec -= 1000000;
706 }
707 printf("tvotohz: negative time difference %ld sec %ld usec\n",
708 sec, usec);
709#endif
710 ticks = 1;
a94976ad
MD
711 } else if (sec <= INT_MAX / hz) {
712 ticks = (int)(sec * hz +
713 ((u_long)usec + (tick - 1)) / tick) + 1;
714 } else {
715 ticks = INT_MAX;
716 }
717 return (ticks);
718}
719
720/*
721 * Compute number of ticks for the specified amount of time, erroring on
722 * the side of it being too low to ensure that sleeping the returned number
723 * of ticks will not result in a late return.
724 *
725 * The supplied timeval may not be negative and should be normalized. A
726 * return value of 0 is possible if the timeval converts to less then
727 * 1 tick.
728 *
729 * If ints have 32 bits, then the maximum value for any timeout in
730 * 10ms ticks is 248 days.
731 */
732int
733tvtohz_low(struct timeval *tv)
734{
735 int ticks;
736 long sec;
737
738 sec = tv->tv_sec;
739 if (sec <= INT_MAX / hz)
740 ticks = (int)(sec * hz + (u_long)tv->tv_usec / tick);
984263bc 741 else
984263bc 742 ticks = INT_MAX;
a94976ad 743 return (ticks);
984263bc
MD
744}
745
a94976ad 746
984263bc
MD
747/*
748 * Start profiling on a process.
749 *
750 * Kernel profiling passes proc0 which never exits and hence
751 * keeps the profile clock running constantly.
752 */
753void
88c4d2f6 754startprofclock(struct proc *p)
984263bc 755{
984263bc
MD
756 if ((p->p_flag & P_PROFIL) == 0) {
757 p->p_flag |= P_PROFIL;
88c4d2f6 758#if 0 /* XXX */
984263bc 759 if (++profprocs == 1 && stathz != 0) {
e43a034f 760 crit_enter();
6ad39cae 761 psdiv = psratio;
984263bc 762 setstatclockrate(profhz);
e43a034f 763 crit_exit();
984263bc 764 }
88c4d2f6 765#endif
984263bc
MD
766 }
767}
768
769/*
770 * Stop profiling on a process.
771 */
772void
88c4d2f6 773stopprofclock(struct proc *p)
984263bc 774{
984263bc
MD
775 if (p->p_flag & P_PROFIL) {
776 p->p_flag &= ~P_PROFIL;
88c4d2f6 777#if 0 /* XXX */
984263bc 778 if (--profprocs == 0 && stathz != 0) {
e43a034f 779 crit_enter();
6ad39cae 780 psdiv = 1;
984263bc 781 setstatclockrate(stathz);
e43a034f 782 crit_exit();
984263bc 783 }
984263bc 784#endif
984263bc
MD
785 }
786}
787
788/*
789 * Return information about system clocks.
790 */
791static int
792sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS)
793{
f5d21610 794 struct kinfo_clockinfo clkinfo;
984263bc
MD
795 /*
796 * Construct clockinfo structure.
797 */
f5d21610
JS
798 clkinfo.ci_hz = hz;
799 clkinfo.ci_tick = tick;
4026c000 800 clkinfo.ci_tickadj = ntp_default_tick_delta / 1000;
f5d21610
JS
801 clkinfo.ci_profhz = profhz;
802 clkinfo.ci_stathz = stathz ? stathz : hz;
984263bc
MD
803 return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req));
804}
805
806SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate, CTLTYPE_STRUCT|CTLFLAG_RD,
807 0, 0, sysctl_kern_clockrate, "S,clockinfo","");
808
984263bc
MD
809/*
810 * We have eight functions for looking at the clock, four for
811 * microseconds and four for nanoseconds. For each there is fast
812 * but less precise version "get{nano|micro}[up]time" which will
813 * return a time which is up to 1/HZ previous to the call, whereas
814 * the raw version "{nano|micro}[up]time" will return a timestamp
815 * which is as precise as possible. The "up" variants return the
816 * time relative to system boot, these are well suited for time
817 * interval measurements.
88c4d2f6
MD
818 *
819 * Each cpu independantly maintains the current time of day, so all
820 * we need to do to protect ourselves from changes is to do a loop
821 * check on the seconds field changing out from under us.
fad57d0e
MD
822 *
823 * The system timer maintains a 32 bit count and due to various issues
824 * it is possible for the calculated delta to occassionally exceed
044ee7c4
MD
825 * sys_cputimer->freq. If this occurs the sys_cputimer->freq64_nsec
826 * multiplication can easily overflow, so we deal with the case. For
827 * uniformity we deal with the case in the usec case too.
984263bc 828 */
984263bc
MD
829void
830getmicrouptime(struct timeval *tvp)
831{
88c4d2f6
MD
832 struct globaldata *gd = mycpu;
833 sysclock_t delta;
834
835 do {
836 tvp->tv_sec = gd->gd_time_seconds;
837 delta = gd->gd_hardclock.time - gd->gd_cpuclock_base;
838 } while (tvp->tv_sec != gd->gd_time_seconds);
fad57d0e 839
044ee7c4
MD
840 if (delta >= sys_cputimer->freq) {
841 tvp->tv_sec += delta / sys_cputimer->freq;
842 delta %= sys_cputimer->freq;
fad57d0e 843 }
044ee7c4 844 tvp->tv_usec = (sys_cputimer->freq64_usec * delta) >> 32;
88c4d2f6
MD
845 if (tvp->tv_usec >= 1000000) {
846 tvp->tv_usec -= 1000000;
847 ++tvp->tv_sec;
984263bc
MD
848 }
849}
850
851void
852getnanouptime(struct timespec *tsp)
853{
88c4d2f6
MD
854 struct globaldata *gd = mycpu;
855 sysclock_t delta;
856
857 do {
858 tsp->tv_sec = gd->gd_time_seconds;
859 delta = gd->gd_hardclock.time - gd->gd_cpuclock_base;
860 } while (tsp->tv_sec != gd->gd_time_seconds);
fad57d0e 861
044ee7c4
MD
862 if (delta >= sys_cputimer->freq) {
863 tsp->tv_sec += delta / sys_cputimer->freq;
864 delta %= sys_cputimer->freq;
984263bc 865 }
044ee7c4 866 tsp->tv_nsec = (sys_cputimer->freq64_nsec * delta) >> 32;
984263bc
MD
867}
868
869void
88c4d2f6 870microuptime(struct timeval *tvp)
984263bc 871{
88c4d2f6
MD
872 struct globaldata *gd = mycpu;
873 sysclock_t delta;
874
875 do {
876 tvp->tv_sec = gd->gd_time_seconds;
044ee7c4 877 delta = sys_cputimer->count() - gd->gd_cpuclock_base;
88c4d2f6 878 } while (tvp->tv_sec != gd->gd_time_seconds);
fad57d0e 879
044ee7c4
MD
880 if (delta >= sys_cputimer->freq) {
881 tvp->tv_sec += delta / sys_cputimer->freq;
882 delta %= sys_cputimer->freq;
984263bc 883 }
044ee7c4 884 tvp->tv_usec = (sys_cputimer->freq64_usec * delta) >> 32;
984263bc
MD
885}
886
887void
88c4d2f6 888nanouptime(struct timespec *tsp)
984263bc 889{
88c4d2f6
MD
890 struct globaldata *gd = mycpu;
891 sysclock_t delta;
892
893 do {
894 tsp->tv_sec = gd->gd_time_seconds;
044ee7c4 895 delta = sys_cputimer->count() - gd->gd_cpuclock_base;
88c4d2f6 896 } while (tsp->tv_sec != gd->gd_time_seconds);
fad57d0e 897
044ee7c4
MD
898 if (delta >= sys_cputimer->freq) {
899 tsp->tv_sec += delta / sys_cputimer->freq;
900 delta %= sys_cputimer->freq;
984263bc 901 }
044ee7c4 902 tsp->tv_nsec = (sys_cputimer->freq64_nsec * delta) >> 32;
984263bc
MD
903}
904
88c4d2f6
MD
905/*
906 * realtime routines
907 */
984263bc
MD
908
909void
88c4d2f6 910getmicrotime(struct timeval *tvp)
984263bc 911{
88c4d2f6 912 struct globaldata *gd = mycpu;
5eb5a6bc 913 struct timespec *bt;
88c4d2f6 914 sysclock_t delta;
984263bc 915
88c4d2f6
MD
916 do {
917 tvp->tv_sec = gd->gd_time_seconds;
918 delta = gd->gd_hardclock.time - gd->gd_cpuclock_base;
919 } while (tvp->tv_sec != gd->gd_time_seconds);
fad57d0e 920
044ee7c4
MD
921 if (delta >= sys_cputimer->freq) {
922 tvp->tv_sec += delta / sys_cputimer->freq;
923 delta %= sys_cputimer->freq;
fad57d0e 924 }
044ee7c4 925 tvp->tv_usec = (sys_cputimer->freq64_usec * delta) >> 32;
984263bc 926
5eb5a6bc
MD
927 bt = &basetime[basetime_index];
928 tvp->tv_sec += bt->tv_sec;
929 tvp->tv_usec += bt->tv_nsec / 1000;
88c4d2f6
MD
930 while (tvp->tv_usec >= 1000000) {
931 tvp->tv_usec -= 1000000;
932 ++tvp->tv_sec;
984263bc 933 }
984263bc
MD
934}
935
936void
88c4d2f6 937getnanotime(struct timespec *tsp)
984263bc 938{
88c4d2f6 939 struct globaldata *gd = mycpu;
5eb5a6bc 940 struct timespec *bt;
88c4d2f6 941 sysclock_t delta;
984263bc 942
88c4d2f6
MD
943 do {
944 tsp->tv_sec = gd->gd_time_seconds;
945 delta = gd->gd_hardclock.time - gd->gd_cpuclock_base;
946 } while (tsp->tv_sec != gd->gd_time_seconds);
fad57d0e 947
044ee7c4
MD
948 if (delta >= sys_cputimer->freq) {
949 tsp->tv_sec += delta / sys_cputimer->freq;
950 delta %= sys_cputimer->freq;
fad57d0e 951 }
044ee7c4 952 tsp->tv_nsec = (sys_cputimer->freq64_nsec * delta) >> 32;
984263bc 953
5eb5a6bc
MD
954 bt = &basetime[basetime_index];
955 tsp->tv_sec += bt->tv_sec;
956 tsp->tv_nsec += bt->tv_nsec;
88c4d2f6
MD
957 while (tsp->tv_nsec >= 1000000000) {
958 tsp->tv_nsec -= 1000000000;
959 ++tsp->tv_sec;
984263bc 960 }
984263bc
MD
961}
962
5eb5a6bc
MD
963static void
964getnanotime_nbt(struct timespec *nbt, struct timespec *tsp)
965{
966 struct globaldata *gd = mycpu;
967 sysclock_t delta;
968
969 do {
970 tsp->tv_sec = gd->gd_time_seconds;
971 delta = gd->gd_hardclock.time - gd->gd_cpuclock_base;
972 } while (tsp->tv_sec != gd->gd_time_seconds);
973
044ee7c4
MD
974 if (delta >= sys_cputimer->freq) {
975 tsp->tv_sec += delta / sys_cputimer->freq;
976 delta %= sys_cputimer->freq;
5eb5a6bc 977 }
044ee7c4 978 tsp->tv_nsec = (sys_cputimer->freq64_nsec * delta) >> 32;
5eb5a6bc
MD
979
980 tsp->tv_sec += nbt->tv_sec;
981 tsp->tv_nsec += nbt->tv_nsec;
982 while (tsp->tv_nsec >= 1000000000) {
983 tsp->tv_nsec -= 1000000000;
984 ++tsp->tv_sec;
985 }
986}
987
988
88c4d2f6
MD
989void
990microtime(struct timeval *tvp)
984263bc 991{
88c4d2f6 992 struct globaldata *gd = mycpu;
5eb5a6bc 993 struct timespec *bt;
88c4d2f6 994 sysclock_t delta;
984263bc 995
88c4d2f6
MD
996 do {
997 tvp->tv_sec = gd->gd_time_seconds;
044ee7c4 998 delta = sys_cputimer->count() - gd->gd_cpuclock_base;
88c4d2f6 999 } while (tvp->tv_sec != gd->gd_time_seconds);
fad57d0e 1000
044ee7c4
MD
1001 if (delta >= sys_cputimer->freq) {
1002 tvp->tv_sec += delta / sys_cputimer->freq;
1003 delta %= sys_cputimer->freq;
fad57d0e 1004 }
044ee7c4 1005 tvp->tv_usec = (sys_cputimer->freq64_usec * delta) >> 32;
984263bc 1006
5eb5a6bc
MD
1007 bt = &basetime[basetime_index];
1008 tvp->tv_sec += bt->tv_sec;
1009 tvp->tv_usec += bt->tv_nsec / 1000;
88c4d2f6
MD
1010 while (tvp->tv_usec >= 1000000) {
1011 tvp->tv_usec -= 1000000;
1012 ++tvp->tv_sec;
984263bc 1013 }
984263bc
MD
1014}
1015
88c4d2f6
MD
1016void
1017nanotime(struct timespec *tsp)
1018{
1019 struct globaldata *gd = mycpu;
5eb5a6bc 1020 struct timespec *bt;
88c4d2f6 1021 sysclock_t delta;
984263bc 1022
88c4d2f6
MD
1023 do {
1024 tsp->tv_sec = gd->gd_time_seconds;
044ee7c4 1025 delta = sys_cputimer->count() - gd->gd_cpuclock_base;
88c4d2f6 1026 } while (tsp->tv_sec != gd->gd_time_seconds);
fad57d0e 1027
044ee7c4
MD
1028 if (delta >= sys_cputimer->freq) {
1029 tsp->tv_sec += delta / sys_cputimer->freq;
1030 delta %= sys_cputimer->freq;
fad57d0e 1031 }
044ee7c4 1032 tsp->tv_nsec = (sys_cputimer->freq64_nsec * delta) >> 32;
984263bc 1033
5eb5a6bc
MD
1034 bt = &basetime[basetime_index];
1035 tsp->tv_sec += bt->tv_sec;
1036 tsp->tv_nsec += bt->tv_nsec;
88c4d2f6
MD
1037 while (tsp->tv_nsec >= 1000000000) {
1038 tsp->tv_nsec -= 1000000000;
1039 ++tsp->tv_sec;
984263bc 1040 }
984263bc
MD
1041}
1042
25b804e7
MD
1043/*
1044 * note: this is not exactly synchronized with real time. To do that we
1045 * would have to do what microtime does and check for a nanoseconds overflow.
1046 */
1047time_t
1048get_approximate_time_t(void)
1049{
1050 struct globaldata *gd = mycpu;
5eb5a6bc
MD
1051 struct timespec *bt;
1052
1053 bt = &basetime[basetime_index];
1054 return(gd->gd_time_seconds + bt->tv_sec);
25b804e7
MD
1055}
1056
984263bc
MD
1057int
1058pps_ioctl(u_long cmd, caddr_t data, struct pps_state *pps)
1059{
1060 pps_params_t *app;
1061 struct pps_fetch_args *fapi;
1062#ifdef PPS_SYNC
1063 struct pps_kcbind_args *kapi;
1064#endif
1065
1066 switch (cmd) {
1067 case PPS_IOC_CREATE:
1068 return (0);
1069 case PPS_IOC_DESTROY:
1070 return (0);
1071 case PPS_IOC_SETPARAMS:
1072 app = (pps_params_t *)data;
1073 if (app->mode & ~pps->ppscap)
1074 return (EINVAL);
1075 pps->ppsparam = *app;
1076 return (0);
1077 case PPS_IOC_GETPARAMS:
1078 app = (pps_params_t *)data;
1079 *app = pps->ppsparam;
1080 app->api_version = PPS_API_VERS_1;
1081 return (0);
1082 case PPS_IOC_GETCAP:
1083 *(int*)data = pps->ppscap;
1084 return (0);
1085 case PPS_IOC_FETCH:
1086 fapi = (struct pps_fetch_args *)data;
1087 if (fapi->tsformat && fapi->tsformat != PPS_TSFMT_TSPEC)
1088 return (EINVAL);
1089 if (fapi->timeout.tv_sec || fapi->timeout.tv_nsec)
1090 return (EOPNOTSUPP);
1091 pps->ppsinfo.current_mode = pps->ppsparam.mode;
1092 fapi->pps_info_buf = pps->ppsinfo;
1093 return (0);
1094 case PPS_IOC_KCBIND:
1095#ifdef PPS_SYNC
1096 kapi = (struct pps_kcbind_args *)data;
1097 /* XXX Only root should be able to do this */
1098 if (kapi->tsformat && kapi->tsformat != PPS_TSFMT_TSPEC)
1099 return (EINVAL);
1100 if (kapi->kernel_consumer != PPS_KC_HARDPPS)
1101 return (EINVAL);
1102 if (kapi->edge & ~pps->ppscap)
1103 return (EINVAL);
1104 pps->kcmode = kapi->edge;
1105 return (0);
1106#else
1107 return (EOPNOTSUPP);
1108#endif
1109 default:
1110 return (ENOTTY);
1111 }
1112}
1113
1114void
1115pps_init(struct pps_state *pps)
1116{
1117 pps->ppscap |= PPS_TSFMT_TSPEC;
1118 if (pps->ppscap & PPS_CAPTUREASSERT)
1119 pps->ppscap |= PPS_OFFSETASSERT;
1120 if (pps->ppscap & PPS_CAPTURECLEAR)
1121 pps->ppscap |= PPS_OFFSETCLEAR;
1122}
1123
1124void
88c4d2f6 1125pps_event(struct pps_state *pps, sysclock_t count, int event)
984263bc 1126{
88c4d2f6
MD
1127 struct globaldata *gd;
1128 struct timespec *tsp;
1129 struct timespec *osp;
5eb5a6bc 1130 struct timespec *bt;
88c4d2f6
MD
1131 struct timespec ts;
1132 sysclock_t *pcount;
1133#ifdef PPS_SYNC
1134 sysclock_t tcount;
1135#endif
1136 sysclock_t delta;
1137 pps_seq_t *pseq;
1138 int foff;
1139 int fhard;
1140
1141 gd = mycpu;
984263bc
MD
1142
1143 /* Things would be easier with arrays... */
1144 if (event == PPS_CAPTUREASSERT) {
1145 tsp = &pps->ppsinfo.assert_timestamp;
1146 osp = &pps->ppsparam.assert_offset;
1147 foff = pps->ppsparam.mode & PPS_OFFSETASSERT;
1148 fhard = pps->kcmode & PPS_CAPTUREASSERT;
1149 pcount = &pps->ppscount[0];
1150 pseq = &pps->ppsinfo.assert_sequence;
1151 } else {
1152 tsp = &pps->ppsinfo.clear_timestamp;
1153 osp = &pps->ppsparam.clear_offset;
1154 foff = pps->ppsparam.mode & PPS_OFFSETCLEAR;
1155 fhard = pps->kcmode & PPS_CAPTURECLEAR;
1156 pcount = &pps->ppscount[1];
1157 pseq = &pps->ppsinfo.clear_sequence;
1158 }
1159
984263bc
MD
1160 /* Nothing really happened */
1161 if (*pcount == count)
1162 return;
1163
1164 *pcount = count;
1165
88c4d2f6
MD
1166 do {
1167 ts.tv_sec = gd->gd_time_seconds;
1168 delta = count - gd->gd_cpuclock_base;
1169 } while (ts.tv_sec != gd->gd_time_seconds);
fad57d0e 1170
044ee7c4
MD
1171 if (delta >= sys_cputimer->freq) {
1172 ts.tv_sec += delta / sys_cputimer->freq;
1173 delta %= sys_cputimer->freq;
88c4d2f6 1174 }
044ee7c4 1175 ts.tv_nsec = (sys_cputimer->freq64_nsec * delta) >> 32;
5eb5a6bc
MD
1176 bt = &basetime[basetime_index];
1177 ts.tv_sec += bt->tv_sec;
1178 ts.tv_nsec += bt->tv_nsec;
88c4d2f6
MD
1179 while (ts.tv_nsec >= 1000000000) {
1180 ts.tv_nsec -= 1000000000;
1181 ++ts.tv_sec;
984263bc 1182 }
984263bc
MD
1183
1184 (*pseq)++;
1185 *tsp = ts;
1186
1187 if (foff) {
1188 timespecadd(tsp, osp);
1189 if (tsp->tv_nsec < 0) {
1190 tsp->tv_nsec += 1000000000;
1191 tsp->tv_sec -= 1;
1192 }
1193 }
1194#ifdef PPS_SYNC
1195 if (fhard) {
1196 /* magic, at its best... */
1197 tcount = count - pps->ppscount[2];
1198 pps->ppscount[2] = count;
044ee7c4
MD
1199 if (tcount >= sys_cputimer->freq) {
1200 delta = (1000000000 * (tcount / sys_cputimer->freq) +
1201 sys_cputimer->freq64_nsec *
1202 (tcount % sys_cputimer->freq)) >> 32;
fad57d0e 1203 } else {
044ee7c4 1204 delta = (sys_cputimer->freq64_nsec * tcount) >> 32;
fad57d0e 1205 }
984263bc
MD
1206 hardpps(tsp, delta);
1207 }
1208#endif
1209}
88c4d2f6 1210