* Remove a procedural layer in the scheduler clock code by having
[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 $
dcc99b62 73 * $DragonFly: src/sys/kern/kern_clock.c,v 1.44 2005/06/27 18:37:57 dillon 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) &&
984263bc
MD
506 timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) &&
507 itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0)
508 psignal(p, SIGVTALRM);
509 if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value) &&
510 itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0)
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 */
649 p->p_usched->schedulerclock(p, info->periodic, info->time);
650 }
88c4d2f6 651 if ((p = curproc) != NULL) {
dcc99b62
MD
652 /*
653 * Update resource usage integrals and maximums.
654 */
88c4d2f6
MD
655 if ((pstats = p->p_stats) != NULL &&
656 (ru = &pstats->p_ru) != NULL &&
657 (vm = p->p_vmspace) != NULL) {
658 ru->ru_ixrss += pgtok(vm->vm_tsize);
659 ru->ru_idrss += pgtok(vm->vm_dsize);
660 ru->ru_isrss += pgtok(vm->vm_ssize);
661 rss = pgtok(vmspace_resident_count(vm));
662 if (ru->ru_maxrss < rss)
663 ru->ru_maxrss = rss;
664 }
b68b7282 665 }
984263bc
MD
666}
667
668/*
a94976ad
MD
669 * Compute number of ticks for the specified amount of time. The
670 * return value is intended to be used in a clock interrupt timed
671 * operation and guarenteed to meet or exceed the requested time.
672 * If the representation overflows, return INT_MAX. The minimum return
673 * value is 1 ticks and the function will average the calculation up.
674 * If any value greater then 0 microseconds is supplied, a value
675 * of at least 2 will be returned to ensure that a near-term clock
676 * interrupt does not cause the timeout to occur (degenerately) early.
677 *
678 * Note that limit checks must take into account microseconds, which is
679 * done simply by using the smaller signed long maximum instead of
680 * the unsigned long maximum.
681 *
682 * If ints have 32 bits, then the maximum value for any timeout in
683 * 10ms ticks is 248 days.
984263bc
MD
684 */
685int
a94976ad 686tvtohz_high(struct timeval *tv)
984263bc 687{
a94976ad 688 int ticks;
1fd87d54 689 long sec, usec;
984263bc 690
984263bc
MD
691 sec = tv->tv_sec;
692 usec = tv->tv_usec;
693 if (usec < 0) {
694 sec--;
695 usec += 1000000;
696 }
697 if (sec < 0) {
698#ifdef DIAGNOSTIC
699 if (usec > 0) {
700 sec++;
701 usec -= 1000000;
702 }
703 printf("tvotohz: negative time difference %ld sec %ld usec\n",
704 sec, usec);
705#endif
706 ticks = 1;
a94976ad
MD
707 } else if (sec <= INT_MAX / hz) {
708 ticks = (int)(sec * hz +
709 ((u_long)usec + (tick - 1)) / tick) + 1;
710 } else {
711 ticks = INT_MAX;
712 }
713 return (ticks);
714}
715
716/*
717 * Compute number of ticks for the specified amount of time, erroring on
718 * the side of it being too low to ensure that sleeping the returned number
719 * of ticks will not result in a late return.
720 *
721 * The supplied timeval may not be negative and should be normalized. A
722 * return value of 0 is possible if the timeval converts to less then
723 * 1 tick.
724 *
725 * If ints have 32 bits, then the maximum value for any timeout in
726 * 10ms ticks is 248 days.
727 */
728int
729tvtohz_low(struct timeval *tv)
730{
731 int ticks;
732 long sec;
733
734 sec = tv->tv_sec;
735 if (sec <= INT_MAX / hz)
736 ticks = (int)(sec * hz + (u_long)tv->tv_usec / tick);
984263bc 737 else
984263bc 738 ticks = INT_MAX;
a94976ad 739 return (ticks);
984263bc
MD
740}
741
a94976ad 742
984263bc
MD
743/*
744 * Start profiling on a process.
745 *
746 * Kernel profiling passes proc0 which never exits and hence
747 * keeps the profile clock running constantly.
748 */
749void
88c4d2f6 750startprofclock(struct proc *p)
984263bc 751{
984263bc
MD
752 if ((p->p_flag & P_PROFIL) == 0) {
753 p->p_flag |= P_PROFIL;
88c4d2f6 754#if 0 /* XXX */
984263bc 755 if (++profprocs == 1 && stathz != 0) {
e43a034f 756 crit_enter();
6ad39cae 757 psdiv = psratio;
984263bc 758 setstatclockrate(profhz);
e43a034f 759 crit_exit();
984263bc 760 }
88c4d2f6 761#endif
984263bc
MD
762 }
763}
764
765/*
766 * Stop profiling on a process.
767 */
768void
88c4d2f6 769stopprofclock(struct proc *p)
984263bc 770{
984263bc
MD
771 if (p->p_flag & P_PROFIL) {
772 p->p_flag &= ~P_PROFIL;
88c4d2f6 773#if 0 /* XXX */
984263bc 774 if (--profprocs == 0 && stathz != 0) {
e43a034f 775 crit_enter();
6ad39cae 776 psdiv = 1;
984263bc 777 setstatclockrate(stathz);
e43a034f 778 crit_exit();
984263bc 779 }
984263bc 780#endif
984263bc
MD
781 }
782}
783
784/*
785 * Return information about system clocks.
786 */
787static int
788sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS)
789{
f5d21610 790 struct kinfo_clockinfo clkinfo;
984263bc
MD
791 /*
792 * Construct clockinfo structure.
793 */
f5d21610
JS
794 clkinfo.ci_hz = hz;
795 clkinfo.ci_tick = tick;
4026c000 796 clkinfo.ci_tickadj = ntp_default_tick_delta / 1000;
f5d21610
JS
797 clkinfo.ci_profhz = profhz;
798 clkinfo.ci_stathz = stathz ? stathz : hz;
984263bc
MD
799 return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req));
800}
801
802SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate, CTLTYPE_STRUCT|CTLFLAG_RD,
803 0, 0, sysctl_kern_clockrate, "S,clockinfo","");
804
984263bc
MD
805/*
806 * We have eight functions for looking at the clock, four for
807 * microseconds and four for nanoseconds. For each there is fast
808 * but less precise version "get{nano|micro}[up]time" which will
809 * return a time which is up to 1/HZ previous to the call, whereas
810 * the raw version "{nano|micro}[up]time" will return a timestamp
811 * which is as precise as possible. The "up" variants return the
812 * time relative to system boot, these are well suited for time
813 * interval measurements.
88c4d2f6
MD
814 *
815 * Each cpu independantly maintains the current time of day, so all
816 * we need to do to protect ourselves from changes is to do a loop
817 * check on the seconds field changing out from under us.
fad57d0e
MD
818 *
819 * The system timer maintains a 32 bit count and due to various issues
820 * it is possible for the calculated delta to occassionally exceed
044ee7c4
MD
821 * sys_cputimer->freq. If this occurs the sys_cputimer->freq64_nsec
822 * multiplication can easily overflow, so we deal with the case. For
823 * uniformity we deal with the case in the usec case too.
984263bc 824 */
984263bc
MD
825void
826getmicrouptime(struct timeval *tvp)
827{
88c4d2f6
MD
828 struct globaldata *gd = mycpu;
829 sysclock_t delta;
830
831 do {
832 tvp->tv_sec = gd->gd_time_seconds;
833 delta = gd->gd_hardclock.time - gd->gd_cpuclock_base;
834 } while (tvp->tv_sec != gd->gd_time_seconds);
fad57d0e 835
044ee7c4
MD
836 if (delta >= sys_cputimer->freq) {
837 tvp->tv_sec += delta / sys_cputimer->freq;
838 delta %= sys_cputimer->freq;
fad57d0e 839 }
044ee7c4 840 tvp->tv_usec = (sys_cputimer->freq64_usec * delta) >> 32;
88c4d2f6
MD
841 if (tvp->tv_usec >= 1000000) {
842 tvp->tv_usec -= 1000000;
843 ++tvp->tv_sec;
984263bc
MD
844 }
845}
846
847void
848getnanouptime(struct timespec *tsp)
849{
88c4d2f6
MD
850 struct globaldata *gd = mycpu;
851 sysclock_t delta;
852
853 do {
854 tsp->tv_sec = gd->gd_time_seconds;
855 delta = gd->gd_hardclock.time - gd->gd_cpuclock_base;
856 } while (tsp->tv_sec != gd->gd_time_seconds);
fad57d0e 857
044ee7c4
MD
858 if (delta >= sys_cputimer->freq) {
859 tsp->tv_sec += delta / sys_cputimer->freq;
860 delta %= sys_cputimer->freq;
984263bc 861 }
044ee7c4 862 tsp->tv_nsec = (sys_cputimer->freq64_nsec * delta) >> 32;
984263bc
MD
863}
864
865void
88c4d2f6 866microuptime(struct timeval *tvp)
984263bc 867{
88c4d2f6
MD
868 struct globaldata *gd = mycpu;
869 sysclock_t delta;
870
871 do {
872 tvp->tv_sec = gd->gd_time_seconds;
044ee7c4 873 delta = sys_cputimer->count() - gd->gd_cpuclock_base;
88c4d2f6 874 } while (tvp->tv_sec != gd->gd_time_seconds);
fad57d0e 875
044ee7c4
MD
876 if (delta >= sys_cputimer->freq) {
877 tvp->tv_sec += delta / sys_cputimer->freq;
878 delta %= sys_cputimer->freq;
984263bc 879 }
044ee7c4 880 tvp->tv_usec = (sys_cputimer->freq64_usec * delta) >> 32;
984263bc
MD
881}
882
883void
88c4d2f6 884nanouptime(struct timespec *tsp)
984263bc 885{
88c4d2f6
MD
886 struct globaldata *gd = mycpu;
887 sysclock_t delta;
888
889 do {
890 tsp->tv_sec = gd->gd_time_seconds;
044ee7c4 891 delta = sys_cputimer->count() - gd->gd_cpuclock_base;
88c4d2f6 892 } while (tsp->tv_sec != gd->gd_time_seconds);
fad57d0e 893
044ee7c4
MD
894 if (delta >= sys_cputimer->freq) {
895 tsp->tv_sec += delta / sys_cputimer->freq;
896 delta %= sys_cputimer->freq;
984263bc 897 }
044ee7c4 898 tsp->tv_nsec = (sys_cputimer->freq64_nsec * delta) >> 32;
984263bc
MD
899}
900
88c4d2f6
MD
901/*
902 * realtime routines
903 */
984263bc
MD
904
905void
88c4d2f6 906getmicrotime(struct timeval *tvp)
984263bc 907{
88c4d2f6 908 struct globaldata *gd = mycpu;
5eb5a6bc 909 struct timespec *bt;
88c4d2f6 910 sysclock_t delta;
984263bc 911
88c4d2f6
MD
912 do {
913 tvp->tv_sec = gd->gd_time_seconds;
914 delta = gd->gd_hardclock.time - gd->gd_cpuclock_base;
915 } while (tvp->tv_sec != gd->gd_time_seconds);
fad57d0e 916
044ee7c4
MD
917 if (delta >= sys_cputimer->freq) {
918 tvp->tv_sec += delta / sys_cputimer->freq;
919 delta %= sys_cputimer->freq;
fad57d0e 920 }
044ee7c4 921 tvp->tv_usec = (sys_cputimer->freq64_usec * delta) >> 32;
984263bc 922
5eb5a6bc
MD
923 bt = &basetime[basetime_index];
924 tvp->tv_sec += bt->tv_sec;
925 tvp->tv_usec += bt->tv_nsec / 1000;
88c4d2f6
MD
926 while (tvp->tv_usec >= 1000000) {
927 tvp->tv_usec -= 1000000;
928 ++tvp->tv_sec;
984263bc 929 }
984263bc
MD
930}
931
932void
88c4d2f6 933getnanotime(struct timespec *tsp)
984263bc 934{
88c4d2f6 935 struct globaldata *gd = mycpu;
5eb5a6bc 936 struct timespec *bt;
88c4d2f6 937 sysclock_t delta;
984263bc 938
88c4d2f6
MD
939 do {
940 tsp->tv_sec = gd->gd_time_seconds;
941 delta = gd->gd_hardclock.time - gd->gd_cpuclock_base;
942 } while (tsp->tv_sec != gd->gd_time_seconds);
fad57d0e 943
044ee7c4
MD
944 if (delta >= sys_cputimer->freq) {
945 tsp->tv_sec += delta / sys_cputimer->freq;
946 delta %= sys_cputimer->freq;
fad57d0e 947 }
044ee7c4 948 tsp->tv_nsec = (sys_cputimer->freq64_nsec * delta) >> 32;
984263bc 949
5eb5a6bc
MD
950 bt = &basetime[basetime_index];
951 tsp->tv_sec += bt->tv_sec;
952 tsp->tv_nsec += bt->tv_nsec;
88c4d2f6
MD
953 while (tsp->tv_nsec >= 1000000000) {
954 tsp->tv_nsec -= 1000000000;
955 ++tsp->tv_sec;
984263bc 956 }
984263bc
MD
957}
958
5eb5a6bc
MD
959static void
960getnanotime_nbt(struct timespec *nbt, struct timespec *tsp)
961{
962 struct globaldata *gd = mycpu;
963 sysclock_t delta;
964
965 do {
966 tsp->tv_sec = gd->gd_time_seconds;
967 delta = gd->gd_hardclock.time - gd->gd_cpuclock_base;
968 } while (tsp->tv_sec != gd->gd_time_seconds);
969
044ee7c4
MD
970 if (delta >= sys_cputimer->freq) {
971 tsp->tv_sec += delta / sys_cputimer->freq;
972 delta %= sys_cputimer->freq;
5eb5a6bc 973 }
044ee7c4 974 tsp->tv_nsec = (sys_cputimer->freq64_nsec * delta) >> 32;
5eb5a6bc
MD
975
976 tsp->tv_sec += nbt->tv_sec;
977 tsp->tv_nsec += nbt->tv_nsec;
978 while (tsp->tv_nsec >= 1000000000) {
979 tsp->tv_nsec -= 1000000000;
980 ++tsp->tv_sec;
981 }
982}
983
984
88c4d2f6
MD
985void
986microtime(struct timeval *tvp)
984263bc 987{
88c4d2f6 988 struct globaldata *gd = mycpu;
5eb5a6bc 989 struct timespec *bt;
88c4d2f6 990 sysclock_t delta;
984263bc 991
88c4d2f6
MD
992 do {
993 tvp->tv_sec = gd->gd_time_seconds;
044ee7c4 994 delta = sys_cputimer->count() - gd->gd_cpuclock_base;
88c4d2f6 995 } while (tvp->tv_sec != gd->gd_time_seconds);
fad57d0e 996
044ee7c4
MD
997 if (delta >= sys_cputimer->freq) {
998 tvp->tv_sec += delta / sys_cputimer->freq;
999 delta %= sys_cputimer->freq;
fad57d0e 1000 }
044ee7c4 1001 tvp->tv_usec = (sys_cputimer->freq64_usec * delta) >> 32;
984263bc 1002
5eb5a6bc
MD
1003 bt = &basetime[basetime_index];
1004 tvp->tv_sec += bt->tv_sec;
1005 tvp->tv_usec += bt->tv_nsec / 1000;
88c4d2f6
MD
1006 while (tvp->tv_usec >= 1000000) {
1007 tvp->tv_usec -= 1000000;
1008 ++tvp->tv_sec;
984263bc 1009 }
984263bc
MD
1010}
1011
88c4d2f6
MD
1012void
1013nanotime(struct timespec *tsp)
1014{
1015 struct globaldata *gd = mycpu;
5eb5a6bc 1016 struct timespec *bt;
88c4d2f6 1017 sysclock_t delta;
984263bc 1018
88c4d2f6
MD
1019 do {
1020 tsp->tv_sec = gd->gd_time_seconds;
044ee7c4 1021 delta = sys_cputimer->count() - gd->gd_cpuclock_base;
88c4d2f6 1022 } while (tsp->tv_sec != gd->gd_time_seconds);
fad57d0e 1023
044ee7c4
MD
1024 if (delta >= sys_cputimer->freq) {
1025 tsp->tv_sec += delta / sys_cputimer->freq;
1026 delta %= sys_cputimer->freq;
fad57d0e 1027 }
044ee7c4 1028 tsp->tv_nsec = (sys_cputimer->freq64_nsec * delta) >> 32;
984263bc 1029
5eb5a6bc
MD
1030 bt = &basetime[basetime_index];
1031 tsp->tv_sec += bt->tv_sec;
1032 tsp->tv_nsec += bt->tv_nsec;
88c4d2f6
MD
1033 while (tsp->tv_nsec >= 1000000000) {
1034 tsp->tv_nsec -= 1000000000;
1035 ++tsp->tv_sec;
984263bc 1036 }
984263bc
MD
1037}
1038
25b804e7
MD
1039/*
1040 * note: this is not exactly synchronized with real time. To do that we
1041 * would have to do what microtime does and check for a nanoseconds overflow.
1042 */
1043time_t
1044get_approximate_time_t(void)
1045{
1046 struct globaldata *gd = mycpu;
5eb5a6bc
MD
1047 struct timespec *bt;
1048
1049 bt = &basetime[basetime_index];
1050 return(gd->gd_time_seconds + bt->tv_sec);
25b804e7
MD
1051}
1052
984263bc
MD
1053int
1054pps_ioctl(u_long cmd, caddr_t data, struct pps_state *pps)
1055{
1056 pps_params_t *app;
1057 struct pps_fetch_args *fapi;
1058#ifdef PPS_SYNC
1059 struct pps_kcbind_args *kapi;
1060#endif
1061
1062 switch (cmd) {
1063 case PPS_IOC_CREATE:
1064 return (0);
1065 case PPS_IOC_DESTROY:
1066 return (0);
1067 case PPS_IOC_SETPARAMS:
1068 app = (pps_params_t *)data;
1069 if (app->mode & ~pps->ppscap)
1070 return (EINVAL);
1071 pps->ppsparam = *app;
1072 return (0);
1073 case PPS_IOC_GETPARAMS:
1074 app = (pps_params_t *)data;
1075 *app = pps->ppsparam;
1076 app->api_version = PPS_API_VERS_1;
1077 return (0);
1078 case PPS_IOC_GETCAP:
1079 *(int*)data = pps->ppscap;
1080 return (0);
1081 case PPS_IOC_FETCH:
1082 fapi = (struct pps_fetch_args *)data;
1083 if (fapi->tsformat && fapi->tsformat != PPS_TSFMT_TSPEC)
1084 return (EINVAL);
1085 if (fapi->timeout.tv_sec || fapi->timeout.tv_nsec)
1086 return (EOPNOTSUPP);
1087 pps->ppsinfo.current_mode = pps->ppsparam.mode;
1088 fapi->pps_info_buf = pps->ppsinfo;
1089 return (0);
1090 case PPS_IOC_KCBIND:
1091#ifdef PPS_SYNC
1092 kapi = (struct pps_kcbind_args *)data;
1093 /* XXX Only root should be able to do this */
1094 if (kapi->tsformat && kapi->tsformat != PPS_TSFMT_TSPEC)
1095 return (EINVAL);
1096 if (kapi->kernel_consumer != PPS_KC_HARDPPS)
1097 return (EINVAL);
1098 if (kapi->edge & ~pps->ppscap)
1099 return (EINVAL);
1100 pps->kcmode = kapi->edge;
1101 return (0);
1102#else
1103 return (EOPNOTSUPP);
1104#endif
1105 default:
1106 return (ENOTTY);
1107 }
1108}
1109
1110void
1111pps_init(struct pps_state *pps)
1112{
1113 pps->ppscap |= PPS_TSFMT_TSPEC;
1114 if (pps->ppscap & PPS_CAPTUREASSERT)
1115 pps->ppscap |= PPS_OFFSETASSERT;
1116 if (pps->ppscap & PPS_CAPTURECLEAR)
1117 pps->ppscap |= PPS_OFFSETCLEAR;
1118}
1119
1120void
88c4d2f6 1121pps_event(struct pps_state *pps, sysclock_t count, int event)
984263bc 1122{
88c4d2f6
MD
1123 struct globaldata *gd;
1124 struct timespec *tsp;
1125 struct timespec *osp;
5eb5a6bc 1126 struct timespec *bt;
88c4d2f6
MD
1127 struct timespec ts;
1128 sysclock_t *pcount;
1129#ifdef PPS_SYNC
1130 sysclock_t tcount;
1131#endif
1132 sysclock_t delta;
1133 pps_seq_t *pseq;
1134 int foff;
1135 int fhard;
1136
1137 gd = mycpu;
984263bc
MD
1138
1139 /* Things would be easier with arrays... */
1140 if (event == PPS_CAPTUREASSERT) {
1141 tsp = &pps->ppsinfo.assert_timestamp;
1142 osp = &pps->ppsparam.assert_offset;
1143 foff = pps->ppsparam.mode & PPS_OFFSETASSERT;
1144 fhard = pps->kcmode & PPS_CAPTUREASSERT;
1145 pcount = &pps->ppscount[0];
1146 pseq = &pps->ppsinfo.assert_sequence;
1147 } else {
1148 tsp = &pps->ppsinfo.clear_timestamp;
1149 osp = &pps->ppsparam.clear_offset;
1150 foff = pps->ppsparam.mode & PPS_OFFSETCLEAR;
1151 fhard = pps->kcmode & PPS_CAPTURECLEAR;
1152 pcount = &pps->ppscount[1];
1153 pseq = &pps->ppsinfo.clear_sequence;
1154 }
1155
984263bc
MD
1156 /* Nothing really happened */
1157 if (*pcount == count)
1158 return;
1159
1160 *pcount = count;
1161
88c4d2f6
MD
1162 do {
1163 ts.tv_sec = gd->gd_time_seconds;
1164 delta = count - gd->gd_cpuclock_base;
1165 } while (ts.tv_sec != gd->gd_time_seconds);
fad57d0e 1166
044ee7c4
MD
1167 if (delta >= sys_cputimer->freq) {
1168 ts.tv_sec += delta / sys_cputimer->freq;
1169 delta %= sys_cputimer->freq;
88c4d2f6 1170 }
044ee7c4 1171 ts.tv_nsec = (sys_cputimer->freq64_nsec * delta) >> 32;
5eb5a6bc
MD
1172 bt = &basetime[basetime_index];
1173 ts.tv_sec += bt->tv_sec;
1174 ts.tv_nsec += bt->tv_nsec;
88c4d2f6
MD
1175 while (ts.tv_nsec >= 1000000000) {
1176 ts.tv_nsec -= 1000000000;
1177 ++ts.tv_sec;
984263bc 1178 }
984263bc
MD
1179
1180 (*pseq)++;
1181 *tsp = ts;
1182
1183 if (foff) {
1184 timespecadd(tsp, osp);
1185 if (tsp->tv_nsec < 0) {
1186 tsp->tv_nsec += 1000000000;
1187 tsp->tv_sec -= 1;
1188 }
1189 }
1190#ifdef PPS_SYNC
1191 if (fhard) {
1192 /* magic, at its best... */
1193 tcount = count - pps->ppscount[2];
1194 pps->ppscount[2] = count;
044ee7c4
MD
1195 if (tcount >= sys_cputimer->freq) {
1196 delta = (1000000000 * (tcount / sys_cputimer->freq) +
1197 sys_cputimer->freq64_nsec *
1198 (tcount % sys_cputimer->freq)) >> 32;
fad57d0e 1199 } else {
044ee7c4 1200 delta = (sys_cputimer->freq64_nsec * tcount) >> 32;
fad57d0e 1201 }
984263bc
MD
1202 hardpps(tsp, delta);
1203 }
1204#endif
1205}
88c4d2f6 1206