kernel: Make SMP support default (and non-optional).
[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 $
73 */
74
75#include "opt_ntp.h"
2b71c8f1 76#include "opt_polling.h"
b3a7093f 77#include "opt_ifpoll.h"
07522099 78#include "opt_pctrack.h"
984263bc
MD
79
80#include <sys/param.h>
81#include <sys/systm.h>
984263bc
MD
82#include <sys/callout.h>
83#include <sys/kernel.h>
f5d21610 84#include <sys/kinfo.h>
984263bc
MD
85#include <sys/proc.h>
86#include <sys/malloc.h>
d70eef28 87#include <sys/resource.h>
984263bc
MD
88#include <sys/resourcevar.h>
89#include <sys/signalvar.h>
90#include <sys/timex.h>
91#include <sys/timepps.h>
92#include <vm/vm.h>
93#include <sys/lock.h>
94#include <vm/pmap.h>
95#include <vm/vm_map.h>
5ffd1608 96#include <vm/vm_extern.h>
984263bc 97#include <sys/sysctl.h>
684a93c4 98
2689779e 99#include <sys/thread2.h>
984263bc
MD
100
101#include <machine/cpu.h>
102#include <machine/limits.h>
103#include <machine/smp.h>
d2412a2e
MD
104#include <machine/cpufunc.h>
105#include <machine/specialreg.h>
106#include <machine/clock.h>
984263bc
MD
107
108#ifdef GPROF
109#include <sys/gmon.h>
110#endif
111
112#ifdef DEVICE_POLLING
94ebffcd 113extern void init_device_poll_pcpu(int);
3e61f60e 114#endif
984263bc 115
b3a7093f
SZ
116#ifdef IFPOLL_ENABLE
117extern void ifpoll_init_pcpu(int);
118#endif
119
07522099
MD
120#ifdef DEBUG_PCTRACK
121static void do_pctrack(struct intrframe *frame, int which);
122#endif
123
402ed7e1 124static void initclocks (void *dummy);
ba39e2e0 125SYSINIT(clocks, SI_BOOT2_CLOCKS, SI_ORDER_FIRST, initclocks, NULL)
984263bc 126
6ad39cae
MD
127/*
128 * Some of these don't belong here, but it's easiest to concentrate them.
9eea7f0c 129 * Note that cpu_time counts in microseconds, but most userland programs
6ad39cae
MD
130 * just compare relative times against the total by delta.
131 */
9eea7f0c 132struct kinfo_cputime cputime_percpu[MAXCPU];
07522099
MD
133#ifdef DEBUG_PCTRACK
134struct kinfo_pcheader cputime_pcheader = { PCTRACK_SIZE, PCTRACK_ARYSIZE };
135struct kinfo_pctrack cputime_pctrack[MAXCPU][PCTRACK_SIZE];
136#endif
137
9eea7f0c
HP
138static int
139sysctl_cputime(SYSCTL_HANDLER_ARGS)
140{
141 int cpu, error = 0;
142 size_t size = sizeof(struct kinfo_cputime);
143
144 for (cpu = 0; cpu < ncpus; ++cpu) {
145 if ((error = SYSCTL_OUT(req, &cputime_percpu[cpu], size)))
146 break;
147 }
984263bc 148
9eea7f0c
HP
149 return (error);
150}
151SYSCTL_PROC(_kern, OID_AUTO, cputime, (CTLTYPE_OPAQUE|CTLFLAG_RD), 0, 0,
152 sysctl_cputime, "S,kinfo_cputime", "CPU time statistics");
984263bc 153
06636a8e
AHJ
154static int
155sysctl_cp_time(SYSCTL_HANDLER_ARGS)
156{
157 long cpu_states[5] = {0};
158 int cpu, error = 0;
159 size_t size = sizeof(cpu_states);
160
161 for (cpu = 0; cpu < ncpus; ++cpu) {
d70eef28
SG
162 cpu_states[CP_USER] += cputime_percpu[cpu].cp_user;
163 cpu_states[CP_NICE] += cputime_percpu[cpu].cp_nice;
164 cpu_states[CP_SYS] += cputime_percpu[cpu].cp_sys;
165 cpu_states[CP_INTR] += cputime_percpu[cpu].cp_intr;
166 cpu_states[CP_IDLE] += cputime_percpu[cpu].cp_idle;
06636a8e
AHJ
167 }
168
169 error = SYSCTL_OUT(req, cpu_states, size);
170
171 return (error);
172}
173
174SYSCTL_PROC(_kern, OID_AUTO, cp_time, (CTLTYPE_LONG|CTLFLAG_RD), 0, 0,
175 sysctl_cp_time, "LU", "CPU time statistics");
176
88c4d2f6
MD
177/*
178 * boottime is used to calculate the 'real' uptime. Do not confuse this with
179 * microuptime(). microtime() is not drift compensated. The real uptime
60b2809b
MD
180 * with compensation is nanotime() - bootime. boottime is recalculated
181 * whenever the real time is set based on the compensated elapsed time
182 * in seconds (gd->gd_time_seconds).
88c4d2f6 183 *
88c4d2f6
MD
184 * The gd_time_seconds and gd_cpuclock_base fields remain fairly monotonic.
185 * Slight adjustments to gd_cpuclock_base are made to phase-lock it to
186 * the real time.
187 */
188struct timespec boottime; /* boot time (realtime) for reference only */
88c4d2f6 189time_t time_second; /* read-only 'passive' uptime in seconds */
984263bc 190
5eb5a6bc
MD
191/*
192 * basetime is used to calculate the compensated real time of day. The
193 * basetime can be modified on a per-tick basis by the adjtime(),
194 * ntp_adjtime(), and sysctl-based time correction APIs.
195 *
196 * Note that frequency corrections can also be made by adjusting
197 * gd_cpuclock_base.
198 *
199 * basetime is a tail-chasing FIFO, updated only by cpu #0. The FIFO is
200 * used on both SMP and UP systems to avoid MP races between cpu's and
201 * interrupt races on UP systems.
202 */
203#define BASETIME_ARYSIZE 16
204#define BASETIME_ARYMASK (BASETIME_ARYSIZE - 1)
205static struct timespec basetime[BASETIME_ARYSIZE];
206static volatile int basetime_index;
207
208static int
209sysctl_get_basetime(SYSCTL_HANDLER_ARGS)
210{
211 struct timespec *bt;
212 int error;
35238fa5 213 int index;
5eb5a6bc 214
35238fa5
MD
215 /*
216 * Because basetime data and index may be updated by another cpu,
217 * a load fence is required to ensure that the data we read has
218 * not been speculatively read relative to a possibly updated index.
219 */
220 index = basetime_index;
221 cpu_lfence();
222 bt = &basetime[index];
08f95c49 223 error = SYSCTL_OUT(req, bt, sizeof(*bt));
5eb5a6bc
MD
224 return (error);
225}
226
984263bc 227SYSCTL_STRUCT(_kern, KERN_BOOTTIME, boottime, CTLFLAG_RD,
08f95c49 228 &boottime, timespec, "System boottime");
5eb5a6bc 229SYSCTL_PROC(_kern, OID_AUTO, basetime, CTLTYPE_STRUCT|CTLFLAG_RD, 0, 0,
08f95c49 230 sysctl_get_basetime, "S,timespec", "System basetime");
984263bc 231
96d52ac8
SZ
232static void hardclock(systimer_t info, int, struct intrframe *frame);
233static void statclock(systimer_t info, int, struct intrframe *frame);
234static void schedclock(systimer_t info, int, struct intrframe *frame);
5eb5a6bc 235static void getnanotime_nbt(struct timespec *nbt, struct timespec *tsp);
88c4d2f6
MD
236
237int ticks; /* system master ticks at hz */
da3639ef 238int clocks_running; /* tsleep/timeout clocks operational */
88c4d2f6
MD
239int64_t nsec_adj; /* ntpd per-tick adjustment in nsec << 32 */
240int64_t nsec_acc; /* accumulator */
d6d39bc7 241int sched_ticks; /* global schedule clock ticks */
984263bc 242
4026c000
JS
243/* NTPD time correction fields */
244int64_t ntp_tick_permanent; /* per-tick adjustment in nsec << 32 */
245int64_t ntp_tick_acc; /* accumulator for per-tick adjustment */
246int64_t ntp_delta; /* one-time correction in nsec */
247int64_t ntp_big_delta = 1000000000;
248int32_t ntp_tick_delta; /* current adjustment rate */
249int32_t ntp_default_tick_delta; /* adjustment rate for ntp_delta */
48590578
JS
250time_t ntp_leap_second; /* time of next leap second */
251int ntp_leap_insert; /* whether to insert or remove a second */
4026c000 252
984263bc 253/*
88c4d2f6 254 * Finish initializing clock frequencies and start all clocks running.
984263bc 255 */
88c4d2f6
MD
256/* ARGSUSED*/
257static void
258initclocks(void *dummy)
984263bc 259{
88c4d2f6
MD
260 /*psratio = profhz / stathz;*/
261 initclocks_pcpu();
da3639ef 262 clocks_running = 1;
984263bc
MD
263}
264
88c4d2f6
MD
265/*
266 * Called on a per-cpu basis
267 */
268void
269initclocks_pcpu(void)
270{
271 struct globaldata *gd = mycpu;
984263bc 272
88c4d2f6
MD
273 crit_enter();
274 if (gd->gd_cpuid == 0) {
275 gd->gd_time_seconds = 1;
044ee7c4 276 gd->gd_cpuclock_base = sys_cputimer->count();
88c4d2f6
MD
277 } else {
278 /* XXX */
279 gd->gd_time_seconds = globaldata_find(0)->gd_time_seconds;
280 gd->gd_cpuclock_base = globaldata_find(0)->gd_cpuclock_base;
281 }
0d1dffdf 282
43adde98
SZ
283 systimer_intr_enable();
284
94ebffcd
SZ
285#ifdef DEVICE_POLLING
286 init_device_poll_pcpu(gd->gd_cpuid);
287#endif
288
b3a7093f
SZ
289#ifdef IFPOLL_ENABLE
290 ifpoll_init_pcpu(gd->gd_cpuid);
291#endif
292
0d1dffdf
MD
293 /*
294 * Use a non-queued periodic systimer to prevent multiple ticks from
295 * building up if the sysclock jumps forward (8254 gets reset). The
296 * sysclock will never jump backwards. Our time sync is based on
297 * the actual sysclock, not the ticks count.
298 */
299 systimer_init_periodic_nq(&gd->gd_hardclock, hardclock, NULL, hz);
300 systimer_init_periodic_nq(&gd->gd_statclock, statclock, NULL, stathz);
88c4d2f6 301 /* XXX correct the frequency for scheduler / estcpu tests */
0d1dffdf 302 systimer_init_periodic_nq(&gd->gd_schedclock, schedclock,
8478264a 303 NULL, ESTCPUFREQ);
88c4d2f6
MD
304 crit_exit();
305}
984263bc
MD
306
307/*
88c4d2f6
MD
308 * This sets the current real time of day. Timespecs are in seconds and
309 * nanoseconds. We do not mess with gd_time_seconds and gd_cpuclock_base,
310 * instead we adjust basetime so basetime + gd_* results in the current
311 * time of day. This way the gd_* fields are guarenteed to represent
312 * a monotonically increasing 'uptime' value.
5eb5a6bc
MD
313 *
314 * When set_timeofday() is called from userland, the system call forces it
315 * onto cpu #0 since only cpu #0 can update basetime_index.
984263bc 316 */
88c4d2f6
MD
317void
318set_timeofday(struct timespec *ts)
319{
5eb5a6bc
MD
320 struct timespec *nbt;
321 int ni;
984263bc 322
88c4d2f6
MD
323 /*
324 * XXX SMP / non-atomic basetime updates
325 */
326 crit_enter();
5eb5a6bc
MD
327 ni = (basetime_index + 1) & BASETIME_ARYMASK;
328 nbt = &basetime[ni];
329 nanouptime(nbt);
330 nbt->tv_sec = ts->tv_sec - nbt->tv_sec;
331 nbt->tv_nsec = ts->tv_nsec - nbt->tv_nsec;
332 if (nbt->tv_nsec < 0) {
333 nbt->tv_nsec += 1000000000;
334 --nbt->tv_sec;
88c4d2f6 335 }
a81931cc
MD
336
337 /*
338 * Note that basetime diverges from boottime as the clock drift is
339 * compensated for, so we cannot do away with boottime. When setting
340 * the absolute time of day the drift is 0 (for an instant) and we
341 * can simply assign boottime to basetime.
342 *
343 * Note that nanouptime() is based on gd_time_seconds which is drift
344 * compensated up to a point (it is guarenteed to remain monotonically
345 * increasing). gd_time_seconds is thus our best uptime guess and
346 * suitable for use in the boottime calculation. It is already taken
347 * into account in the basetime calculation above.
348 */
5eb5a6bc 349 boottime.tv_sec = nbt->tv_sec;
4026c000 350 ntp_delta = 0;
5eb5a6bc
MD
351
352 /*
35238fa5
MD
353 * We now have a new basetime, make sure all other cpus have it,
354 * then update the index.
5eb5a6bc 355 */
35238fa5 356 cpu_sfence();
5eb5a6bc
MD
357 basetime_index = ni;
358
88c4d2f6
MD
359 crit_exit();
360}
361
984263bc 362/*
88c4d2f6
MD
363 * Each cpu has its own hardclock, but we only increments ticks and softticks
364 * on cpu #0.
365 *
366 * NOTE! systimer! the MP lock might not be held here. We can only safely
367 * manipulate objects owned by the current cpu.
984263bc 368 */
984263bc 369static void
96d52ac8 370hardclock(systimer_t info, int in_ipi __unused, struct intrframe *frame)
984263bc 371{
88c4d2f6
MD
372 sysclock_t cputicks;
373 struct proc *p;
88c4d2f6 374 struct globaldata *gd = mycpu;
984263bc
MD
375
376 /*
88c4d2f6
MD
377 * Realtime updates are per-cpu. Note that timer corrections as
378 * returned by microtime() and friends make an additional adjustment
379 * using a system-wise 'basetime', but the running time is always
380 * taken from the per-cpu globaldata area. Since the same clock
381 * is distributing (XXX SMP) to all cpus, the per-cpu timebases
382 * stay in synch.
383 *
384 * Note that we never allow info->time (aka gd->gd_hardclock.time)
fad57d0e
MD
385 * to reverse index gd_cpuclock_base, but that it is possible for
386 * it to temporarily get behind in the seconds if something in the
387 * system locks interrupts for a long period of time. Since periodic
388 * timers count events, though everything should resynch again
389 * immediately.
984263bc 390 */
88c4d2f6 391 cputicks = info->time - gd->gd_cpuclock_base;
044ee7c4 392 if (cputicks >= sys_cputimer->freq) {
88c4d2f6 393 ++gd->gd_time_seconds;
044ee7c4 394 gd->gd_cpuclock_base += sys_cputimer->freq;
88c4d2f6 395 }
984263bc
MD
396
397 /*
92b561b7
MD
398 * The system-wide ticks counter and NTP related timedelta/tickdelta
399 * adjustments only occur on cpu #0. NTP adjustments are accomplished
400 * by updating basetime.
984263bc 401 */
88c4d2f6 402 if (gd->gd_cpuid == 0) {
5eb5a6bc 403 struct timespec *nbt;
88c4d2f6
MD
404 struct timespec nts;
405 int leap;
5eb5a6bc 406 int ni;
984263bc 407
88c4d2f6 408 ++ticks;
984263bc 409
88c4d2f6
MD
410#if 0
411 if (tco->tc_poll_pps)
412 tco->tc_poll_pps(tco);
413#endif
5eb5a6bc 414
88c4d2f6 415 /*
5eb5a6bc
MD
416 * Calculate the new basetime index. We are in a critical section
417 * on cpu #0 and can safely play with basetime_index. Start
418 * with the current basetime and then make adjustments.
419 */
420 ni = (basetime_index + 1) & BASETIME_ARYMASK;
421 nbt = &basetime[ni];
422 *nbt = basetime[basetime_index];
423
424 /*
425 * Apply adjtime corrections. (adjtime() API)
426 *
427 * adjtime() only runs on cpu #0 so our critical section is
428 * sufficient to access these variables.
88c4d2f6 429 */
4026c000 430 if (ntp_delta != 0) {
5eb5a6bc 431 nbt->tv_nsec += ntp_tick_delta;
4026c000
JS
432 ntp_delta -= ntp_tick_delta;
433 if ((ntp_delta > 0 && ntp_delta < ntp_tick_delta) ||
434 (ntp_delta < 0 && ntp_delta > ntp_tick_delta)) {
5eb5a6bc 435 ntp_tick_delta = ntp_delta;
4026c000
JS
436 }
437 }
438
5eb5a6bc
MD
439 /*
440 * Apply permanent frequency corrections. (sysctl API)
441 */
4026c000
JS
442 if (ntp_tick_permanent != 0) {
443 ntp_tick_acc += ntp_tick_permanent;
444 if (ntp_tick_acc >= (1LL << 32)) {
5eb5a6bc 445 nbt->tv_nsec += ntp_tick_acc >> 32;
331bc6f8 446 ntp_tick_acc -= (ntp_tick_acc >> 32) << 32;
4026c000 447 } else if (ntp_tick_acc <= -(1LL << 32)) {
331bc6f8 448 /* Negate ntp_tick_acc to avoid shifting the sign bit. */
5eb5a6bc 449 nbt->tv_nsec -= (-ntp_tick_acc) >> 32;
331bc6f8 450 ntp_tick_acc += ((-ntp_tick_acc) >> 32) << 32;
4026c000
JS
451 }
452 }
453
5eb5a6bc
MD
454 if (nbt->tv_nsec >= 1000000000) {
455 nbt->tv_sec++;
456 nbt->tv_nsec -= 1000000000;
457 } else if (nbt->tv_nsec < 0) {
458 nbt->tv_sec--;
459 nbt->tv_nsec += 1000000000;
88c4d2f6
MD
460 }
461
462 /*
5eb5a6bc 463 * Another per-tick compensation. (for ntp_adjtime() API)
88c4d2f6 464 */
5eb5a6bc 465 if (nsec_adj != 0) {
88c4d2f6
MD
466 nsec_acc += nsec_adj;
467 if (nsec_acc >= 0x100000000LL) {
5eb5a6bc 468 nbt->tv_nsec += nsec_acc >> 32;
88c4d2f6
MD
469 nsec_acc = (nsec_acc & 0xFFFFFFFFLL);
470 } else if (nsec_acc <= -0x100000000LL) {
5eb5a6bc 471 nbt->tv_nsec -= -nsec_acc >> 32;
88c4d2f6
MD
472 nsec_acc = -(-nsec_acc & 0xFFFFFFFFLL);
473 }
5eb5a6bc
MD
474 if (nbt->tv_nsec >= 1000000000) {
475 nbt->tv_nsec -= 1000000000;
476 ++nbt->tv_sec;
477 } else if (nbt->tv_nsec < 0) {
478 nbt->tv_nsec += 1000000000;
479 --nbt->tv_sec;
480 }
481 }
482
483 /************************************************************
484 * LEAP SECOND CORRECTION *
485 ************************************************************
486 *
487 * Taking into account all the corrections made above, figure
488 * out the new real time. If the seconds field has changed
489 * then apply any pending leap-second corrections.
490 */
491 getnanotime_nbt(nbt, &nts);
492
32040d57
MD
493 if (time_second != nts.tv_sec) {
494 /*
495 * Apply leap second (sysctl API). Adjust nts for changes
496 * so we do not have to call getnanotime_nbt again.
497 */
498 if (ntp_leap_second) {
499 if (ntp_leap_second == nts.tv_sec) {
500 if (ntp_leap_insert) {
501 nbt->tv_sec++;
502 nts.tv_sec++;
503 } else {
504 nbt->tv_sec--;
505 nts.tv_sec--;
506 }
5eb5a6bc 507 ntp_leap_second--;
32040d57 508 }
88c4d2f6 509 }
88c4d2f6 510
32040d57
MD
511 /*
512 * Apply leap second (ntp_adjtime() API), calculate a new
513 * nsec_adj field. ntp_update_second() returns nsec_adj
514 * as a per-second value but we need it as a per-tick value.
515 */
88c4d2f6 516 leap = ntp_update_second(time_second, &nsec_adj);
88c4d2f6 517 nsec_adj /= hz;
32040d57
MD
518 nbt->tv_sec += leap;
519 nts.tv_sec += leap;
520
521 /*
522 * Update the time_second 'approximate time' global.
523 */
524 time_second = nts.tv_sec;
88c4d2f6 525 }
5eb5a6bc
MD
526
527 /*
528 * Finally, our new basetime is ready to go live!
529 */
35238fa5 530 cpu_sfence();
5eb5a6bc 531 basetime_index = ni;
88c4d2f6
MD
532 }
533
f9235b6d
MD
534 /*
535 * lwkt thread scheduler fair queueing
536 */
85946b6c 537 lwkt_schedulerclock(curthread);
f9235b6d 538
92b561b7
MD
539 /*
540 * softticks are handled for all cpus
541 */
542 hardclock_softtick(gd);
543
88c4d2f6 544 /*
8582ec21
MD
545 * ITimer handling is per-tick, per-cpu.
546 *
547 * We must acquire the per-process token in order for ksignal()
898e34b3
MD
548 * to be non-blocking. For the moment this requires an AST fault,
549 * the ksignal() cannot be safely issued from this hard interrupt.
550 *
551 * XXX Even the trytoken here isn't right, and itimer operation in
552 * a multi threaded environment is going to be weird at the
553 * very least.
88c4d2f6 554 */
8582ec21 555 if ((p = curproc) != NULL && lwkt_trytoken(&p->p_token)) {
3dbbd6dd 556 crit_enter_hard();
88c4d2f6 557 if (frame && CLKF_USERMODE(frame) &&
93328593 558 timevalisset(&p->p_timer[ITIMER_VIRTUAL].it_value) &&
898e34b3 559 itimerdecr(&p->p_timer[ITIMER_VIRTUAL], ustick) == 0) {
4643740a 560 p->p_flags |= P_SIGVTALRM;
898e34b3
MD
561 need_user_resched();
562 }
93328593 563 if (timevalisset(&p->p_timer[ITIMER_PROF].it_value) &&
898e34b3 564 itimerdecr(&p->p_timer[ITIMER_PROF], ustick) == 0) {
4643740a 565 p->p_flags |= P_SIGPROF;
898e34b3
MD
566 need_user_resched();
567 }
3dbbd6dd 568 crit_exit_hard();
8582ec21 569 lwkt_reltoken(&p->p_token);
984263bc 570 }
604e1e09 571 setdelayed();
88c4d2f6 572}
984263bc 573
88c4d2f6
MD
574/*
575 * The statistics clock typically runs at a 125Hz rate, and is intended
576 * to be frequency offset from the hardclock (typ 100Hz). It is per-cpu.
577 *
578 * NOTE! systimer! the MP lock might not be held here. We can only safely
579 * manipulate objects owned by the current cpu.
580 *
581 * The stats clock is responsible for grabbing a profiling sample.
582 * Most of the statistics are only used by user-level statistics programs.
583 * The main exceptions are p->p_uticks, p->p_sticks, p->p_iticks, and
584 * p->p_estcpu.
585 *
586 * Like the other clocks, the stat clock is called from what is effectively
587 * a fast interrupt, so the context should be the thread/process that got
588 * interrupted.
589 */
590static void
96d52ac8 591statclock(systimer_t info, int in_ipi, struct intrframe *frame)
88c4d2f6
MD
592{
593#ifdef GPROF
594 struct gmonparam *g;
595 int i;
984263bc 596#endif
88c4d2f6
MD
597 thread_t td;
598 struct proc *p;
599 int bump;
600 struct timeval tv;
601 struct timeval *stv;
984263bc
MD
602
603 /*
88c4d2f6 604 * How big was our timeslice relative to the last time?
984263bc 605 */
88c4d2f6
MD
606 microuptime(&tv); /* mpsafe */
607 stv = &mycpu->gd_stattv;
608 if (stv->tv_sec == 0) {
609 bump = 1;
610 } else {
611 bump = tv.tv_usec - stv->tv_usec +
612 (tv.tv_sec - stv->tv_sec) * 1000000;
613 if (bump < 0)
614 bump = 0;
615 if (bump > 1000000)
616 bump = 1000000;
617 }
618 *stv = tv;
984263bc 619
88c4d2f6
MD
620 td = curthread;
621 p = td->td_proc;
984263bc 622
88c4d2f6
MD
623 if (frame && CLKF_USERMODE(frame)) {
624 /*
625 * Came from userland, handle user time and deal with
626 * possible process.
627 */
4643740a 628 if (p && (p->p_flags & P_PROFIL))
88c4d2f6
MD
629 addupc_intr(p, CLKF_PC(frame), 1);
630 td->td_uticks += bump;
984263bc 631
88c4d2f6
MD
632 /*
633 * Charge the time as appropriate
634 */
635 if (p && p->p_nice > NZERO)
9eea7f0c 636 cpu_time.cp_nice += bump;
88c4d2f6 637 else
9eea7f0c 638 cpu_time.cp_user += bump;
88c4d2f6 639 } else {
96d52ac8
SZ
640 int intr_nest = mycpu->gd_intr_nesting_level;
641
642 if (in_ipi) {
643 /*
644 * IPI processing code will bump gd_intr_nesting_level
645 * up by one, which breaks following CLKF_INTR testing,
646 * so we substract it by one here.
647 */
648 --intr_nest;
649 }
88c4d2f6
MD
650#ifdef GPROF
651 /*
652 * Kernel statistics are just like addupc_intr, only easier.
653 */
654 g = &_gmonparam;
655 if (g->state == GMON_PROF_ON && frame) {
656 i = CLKF_PC(frame) - g->lowpc;
657 if (i < g->textsize) {
658 i /= HISTFRACTION * sizeof(*g->kcount);
659 g->kcount[i]++;
660 }
661 }
662#endif
6026c54d
SZ
663
664#define IS_INTR_RUNNING ((frame && CLKF_INTR(intr_nest)) || CLKF_INTR_TD(td))
665
88c4d2f6
MD
666 /*
667 * Came from kernel mode, so we were:
668 * - handling an interrupt,
669 * - doing syscall or trap work on behalf of the current
670 * user process, or
671 * - spinning in the idle loop.
672 * Whichever it is, charge the time as appropriate.
673 * Note that we charge interrupts to the current process,
674 * regardless of whether they are ``for'' that process,
675 * so that we know how much of its real time was spent
676 * in ``non-process'' (i.e., interrupt) work.
677 *
678 * XXX assume system if frame is NULL. A NULL frame
e43a034f 679 * can occur if ipi processing is done from a crit_exit().
88c4d2f6 680 */
6026c54d 681 if (IS_INTR_RUNNING)
88c4d2f6
MD
682 td->td_iticks += bump;
683 else
684 td->td_sticks += bump;
685
6026c54d 686 if (IS_INTR_RUNNING) {
07522099 687#ifdef DEBUG_PCTRACK
6026c54d
SZ
688 if (frame)
689 do_pctrack(frame, PCTRACK_INT);
07522099 690#endif
9eea7f0c 691 cpu_time.cp_intr += bump;
88c4d2f6 692 } else {
07522099 693 if (td == &mycpu->gd_idlethread) {
9eea7f0c 694 cpu_time.cp_idle += bump;
07522099
MD
695 } else {
696#ifdef DEBUG_PCTRACK
697 if (frame)
698 do_pctrack(frame, PCTRACK_SYS);
699#endif
9eea7f0c 700 cpu_time.cp_sys += bump;
07522099 701 }
88c4d2f6 702 }
6026c54d
SZ
703
704#undef IS_INTR_RUNNING
88c4d2f6
MD
705 }
706}
707
07522099
MD
708#ifdef DEBUG_PCTRACK
709/*
710 * Sample the PC when in the kernel or in an interrupt. User code can
711 * retrieve the information and generate a histogram or other output.
712 */
713
714static void
715do_pctrack(struct intrframe *frame, int which)
716{
717 struct kinfo_pctrack *pctrack;
718
719 pctrack = &cputime_pctrack[mycpu->gd_cpuid][which];
720 pctrack->pc_array[pctrack->pc_index & PCTRACK_ARYMASK] =
721 (void *)CLKF_PC(frame);
722 ++pctrack->pc_index;
723}
724
725static int
726sysctl_pctrack(SYSCTL_HANDLER_ARGS)
727{
728 struct kinfo_pcheader head;
729 int error;
730 int cpu;
731 int ntrack;
732
733 head.pc_ntrack = PCTRACK_SIZE;
734 head.pc_arysize = PCTRACK_ARYSIZE;
735
736 if ((error = SYSCTL_OUT(req, &head, sizeof(head))) != 0)
737 return (error);
738
739 for (cpu = 0; cpu < ncpus; ++cpu) {
740 for (ntrack = 0; ntrack < PCTRACK_SIZE; ++ntrack) {
741 error = SYSCTL_OUT(req, &cputime_pctrack[cpu][ntrack],
742 sizeof(struct kinfo_pctrack));
743 if (error)
744 break;
745 }
746 if (error)
747 break;
748 }
749 return (error);
750}
751SYSCTL_PROC(_kern, OID_AUTO, pctrack, (CTLTYPE_OPAQUE|CTLFLAG_RD), 0, 0,
752 sysctl_pctrack, "S,kinfo_pcheader", "CPU PC tracking");
753
754#endif
755
88c4d2f6 756/*
dcc99b62 757 * The scheduler clock typically runs at a 50Hz rate. NOTE! systimer,
88c4d2f6
MD
758 * the MP lock might not be held. We can safely manipulate parts of curproc
759 * but that's about it.
dcc99b62
MD
760 *
761 * Each cpu has its own scheduler clock.
88c4d2f6
MD
762 */
763static void
96d52ac8 764schedclock(systimer_t info, int in_ipi __unused, struct intrframe *frame)
88c4d2f6 765{
553ea3c8 766 struct lwp *lp;
88c4d2f6
MD
767 struct rusage *ru;
768 struct vmspace *vm;
769 long rss;
770
553ea3c8 771 if ((lp = lwkt_preempted_proc()) != NULL) {
dcc99b62
MD
772 /*
773 * Account for cpu time used and hit the scheduler. Note
774 * that this call MUST BE MP SAFE, and the BGL IS NOT HELD
775 * HERE.
776 */
553ea3c8 777 ++lp->lwp_cpticks;
de4d4cb0
MD
778 usched_schedulerclock(lp, info->periodic, info->time);
779 } else {
780 usched_schedulerclock(NULL, info->periodic, info->time);
dcc99b62 781 }
553ea3c8 782 if ((lp = curthread->td_lwp) != NULL) {
dcc99b62
MD
783 /*
784 * Update resource usage integrals and maximums.
785 */
fde7ac71 786 if ((ru = &lp->lwp_proc->p_ru) &&
553ea3c8 787 (vm = lp->lwp_proc->p_vmspace) != NULL) {
88c4d2f6
MD
788 ru->ru_ixrss += pgtok(vm->vm_tsize);
789 ru->ru_idrss += pgtok(vm->vm_dsize);
790 ru->ru_isrss += pgtok(vm->vm_ssize);
b12defdc
MD
791 if (lwkt_trytoken(&vm->vm_map.token)) {
792 rss = pgtok(vmspace_resident_count(vm));
793 if (ru->ru_maxrss < rss)
794 ru->ru_maxrss = rss;
795 lwkt_reltoken(&vm->vm_map.token);
796 }
88c4d2f6 797 }
b68b7282 798 }
d6d39bc7
MC
799 /* Increment the global sched_ticks */
800 if (mycpu->gd_cpuid == 0)
801 ++sched_ticks;
984263bc
MD
802}
803
804/*
a94976ad
MD
805 * Compute number of ticks for the specified amount of time. The
806 * return value is intended to be used in a clock interrupt timed
807 * operation and guarenteed to meet or exceed the requested time.
808 * If the representation overflows, return INT_MAX. The minimum return
809 * value is 1 ticks and the function will average the calculation up.
810 * If any value greater then 0 microseconds is supplied, a value
811 * of at least 2 will be returned to ensure that a near-term clock
812 * interrupt does not cause the timeout to occur (degenerately) early.
813 *
814 * Note that limit checks must take into account microseconds, which is
815 * done simply by using the smaller signed long maximum instead of
816 * the unsigned long maximum.
817 *
818 * If ints have 32 bits, then the maximum value for any timeout in
819 * 10ms ticks is 248 days.
984263bc
MD
820 */
821int
a94976ad 822tvtohz_high(struct timeval *tv)
984263bc 823{
a94976ad 824 int ticks;
1fd87d54 825 long sec, usec;
984263bc 826
984263bc
MD
827 sec = tv->tv_sec;
828 usec = tv->tv_usec;
829 if (usec < 0) {
830 sec--;
831 usec += 1000000;
832 }
833 if (sec < 0) {
834#ifdef DIAGNOSTIC
835 if (usec > 0) {
836 sec++;
837 usec -= 1000000;
838 }
a591f597
MD
839 kprintf("tvtohz_high: negative time difference "
840 "%ld sec %ld usec\n",
841 sec, usec);
984263bc
MD
842#endif
843 ticks = 1;
a94976ad
MD
844 } else if (sec <= INT_MAX / hz) {
845 ticks = (int)(sec * hz +
a591f597 846 ((u_long)usec + (ustick - 1)) / ustick) + 1;
a94976ad
MD
847 } else {
848 ticks = INT_MAX;
849 }
850 return (ticks);
851}
852
a591f597
MD
853int
854tstohz_high(struct timespec *ts)
855{
856 int ticks;
857 long sec, nsec;
858
859 sec = ts->tv_sec;
860 nsec = ts->tv_nsec;
861 if (nsec < 0) {
862 sec--;
863 nsec += 1000000000;
864 }
865 if (sec < 0) {
866#ifdef DIAGNOSTIC
867 if (nsec > 0) {
868 sec++;
869 nsec -= 1000000000;
870 }
871 kprintf("tstohz_high: negative time difference "
872 "%ld sec %ld nsec\n",
873 sec, nsec);
874#endif
875 ticks = 1;
876 } else if (sec <= INT_MAX / hz) {
877 ticks = (int)(sec * hz +
878 ((u_long)nsec + (nstick - 1)) / nstick) + 1;
879 } else {
880 ticks = INT_MAX;
881 }
882 return (ticks);
883}
884
885
a94976ad
MD
886/*
887 * Compute number of ticks for the specified amount of time, erroring on
888 * the side of it being too low to ensure that sleeping the returned number
889 * of ticks will not result in a late return.
890 *
891 * The supplied timeval may not be negative and should be normalized. A
892 * return value of 0 is possible if the timeval converts to less then
893 * 1 tick.
894 *
895 * If ints have 32 bits, then the maximum value for any timeout in
896 * 10ms ticks is 248 days.
897 */
898int
899tvtohz_low(struct timeval *tv)
900{
901 int ticks;
902 long sec;
903
904 sec = tv->tv_sec;
905 if (sec <= INT_MAX / hz)
a591f597 906 ticks = (int)(sec * hz + (u_long)tv->tv_usec / ustick);
984263bc 907 else
984263bc 908 ticks = INT_MAX;
a94976ad 909 return (ticks);
984263bc
MD
910}
911
a591f597
MD
912int
913tstohz_low(struct timespec *ts)
914{
915 int ticks;
916 long sec;
917
918 sec = ts->tv_sec;
919 if (sec <= INT_MAX / hz)
920 ticks = (int)(sec * hz + (u_long)ts->tv_nsec / nstick);
921 else
922 ticks = INT_MAX;
923 return (ticks);
924}
a94976ad 925
984263bc
MD
926/*
927 * Start profiling on a process.
928 *
929 * Kernel profiling passes proc0 which never exits and hence
930 * keeps the profile clock running constantly.
931 */
932void
88c4d2f6 933startprofclock(struct proc *p)
984263bc 934{
4643740a
MD
935 if ((p->p_flags & P_PROFIL) == 0) {
936 p->p_flags |= P_PROFIL;
88c4d2f6 937#if 0 /* XXX */
984263bc 938 if (++profprocs == 1 && stathz != 0) {
e43a034f 939 crit_enter();
6ad39cae 940 psdiv = psratio;
984263bc 941 setstatclockrate(profhz);
e43a034f 942 crit_exit();
984263bc 943 }
88c4d2f6 944#endif
984263bc
MD
945 }
946}
947
948/*
949 * Stop profiling on a process.
616516c8
MD
950 *
951 * caller must hold p->p_token
984263bc
MD
952 */
953void
88c4d2f6 954stopprofclock(struct proc *p)
984263bc 955{
4643740a
MD
956 if (p->p_flags & P_PROFIL) {
957 p->p_flags &= ~P_PROFIL;
88c4d2f6 958#if 0 /* XXX */
984263bc 959 if (--profprocs == 0 && stathz != 0) {
e43a034f 960 crit_enter();
6ad39cae 961 psdiv = 1;
984263bc 962 setstatclockrate(stathz);
e43a034f 963 crit_exit();
984263bc 964 }
984263bc 965#endif
984263bc
MD
966 }
967}
968
969/*
970 * Return information about system clocks.
971 */
972static int
973sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS)
974{
f5d21610 975 struct kinfo_clockinfo clkinfo;
984263bc
MD
976 /*
977 * Construct clockinfo structure.
978 */
f5d21610 979 clkinfo.ci_hz = hz;
a591f597 980 clkinfo.ci_tick = ustick;
4026c000 981 clkinfo.ci_tickadj = ntp_default_tick_delta / 1000;
f5d21610
JS
982 clkinfo.ci_profhz = profhz;
983 clkinfo.ci_stathz = stathz ? stathz : hz;
984263bc
MD
984 return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req));
985}
986
987SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate, CTLTYPE_STRUCT|CTLFLAG_RD,
988 0, 0, sysctl_kern_clockrate, "S,clockinfo","");
989
984263bc
MD
990/*
991 * We have eight functions for looking at the clock, four for
992 * microseconds and four for nanoseconds. For each there is fast
993 * but less precise version "get{nano|micro}[up]time" which will
994 * return a time which is up to 1/HZ previous to the call, whereas
995 * the raw version "{nano|micro}[up]time" will return a timestamp
996 * which is as precise as possible. The "up" variants return the
997 * time relative to system boot, these are well suited for time
998 * interval measurements.
88c4d2f6
MD
999 *
1000 * Each cpu independantly maintains the current time of day, so all
1001 * we need to do to protect ourselves from changes is to do a loop
1002 * check on the seconds field changing out from under us.
fad57d0e
MD
1003 *
1004 * The system timer maintains a 32 bit count and due to various issues
1005 * it is possible for the calculated delta to occassionally exceed
044ee7c4
MD
1006 * sys_cputimer->freq. If this occurs the sys_cputimer->freq64_nsec
1007 * multiplication can easily overflow, so we deal with the case. For
1008 * uniformity we deal with the case in the usec case too.
627531fa
MD
1009 *
1010 * All the [get][micro,nano][time,uptime]() routines are MPSAFE.
984263bc 1011 */
984263bc
MD
1012void
1013getmicrouptime(struct timeval *tvp)
1014{
88c4d2f6
MD
1015 struct globaldata *gd = mycpu;
1016 sysclock_t delta;
1017
1018 do {
1019 tvp->tv_sec = gd->gd_time_seconds;
1020 delta = gd->gd_hardclock.time - gd->gd_cpuclock_base;
1021 } while (tvp->tv_sec != gd->gd_time_seconds);
fad57d0e 1022
044ee7c4
MD
1023 if (delta >= sys_cputimer->freq) {
1024 tvp->tv_sec += delta / sys_cputimer->freq;
1025 delta %= sys_cputimer->freq;
fad57d0e 1026 }
044ee7c4 1027 tvp->tv_usec = (sys_cputimer->freq64_usec * delta) >> 32;
88c4d2f6
MD
1028 if (tvp->tv_usec >= 1000000) {
1029 tvp->tv_usec -= 1000000;
1030 ++tvp->tv_sec;
984263bc
MD
1031 }
1032}
1033
1034void
1035getnanouptime(struct timespec *tsp)
1036{
88c4d2f6
MD
1037 struct globaldata *gd = mycpu;
1038 sysclock_t delta;
1039
1040 do {
1041 tsp->tv_sec = gd->gd_time_seconds;
1042 delta = gd->gd_hardclock.time - gd->gd_cpuclock_base;
1043 } while (tsp->tv_sec != gd->gd_time_seconds);
fad57d0e 1044
044ee7c4
MD
1045 if (delta >= sys_cputimer->freq) {
1046 tsp->tv_sec += delta / sys_cputimer->freq;
1047 delta %= sys_cputimer->freq;
984263bc 1048 }
044ee7c4 1049 tsp->tv_nsec = (sys_cputimer->freq64_nsec * delta) >> 32;
984263bc
MD
1050}
1051
1052void
88c4d2f6 1053microuptime(struct timeval *tvp)
984263bc 1054{
88c4d2f6
MD
1055 struct globaldata *gd = mycpu;
1056 sysclock_t delta;
1057
1058 do {
1059 tvp->tv_sec = gd->gd_time_seconds;
044ee7c4 1060 delta = sys_cputimer->count() - gd->gd_cpuclock_base;
88c4d2f6 1061 } while (tvp->tv_sec != gd->gd_time_seconds);
fad57d0e 1062
044ee7c4
MD
1063 if (delta >= sys_cputimer->freq) {
1064 tvp->tv_sec += delta / sys_cputimer->freq;
1065 delta %= sys_cputimer->freq;
984263bc 1066 }
044ee7c4 1067 tvp->tv_usec = (sys_cputimer->freq64_usec * delta) >> 32;
984263bc
MD
1068}
1069
1070void
88c4d2f6 1071nanouptime(struct timespec *tsp)
984263bc 1072{
88c4d2f6
MD
1073 struct globaldata *gd = mycpu;
1074 sysclock_t delta;
1075
1076 do {
1077 tsp->tv_sec = gd->gd_time_seconds;
044ee7c4 1078 delta = sys_cputimer->count() - gd->gd_cpuclock_base;
88c4d2f6 1079 } while (tsp->tv_sec != gd->gd_time_seconds);
fad57d0e 1080
044ee7c4
MD
1081 if (delta >= sys_cputimer->freq) {
1082 tsp->tv_sec += delta / sys_cputimer->freq;
1083 delta %= sys_cputimer->freq;
984263bc 1084 }
044ee7c4 1085 tsp->tv_nsec = (sys_cputimer->freq64_nsec * delta) >> 32;
984263bc
MD
1086}
1087
88c4d2f6
MD
1088/*
1089 * realtime routines
1090 */
984263bc 1091void
88c4d2f6 1092getmicrotime(struct timeval *tvp)
984263bc 1093{
88c4d2f6 1094 struct globaldata *gd = mycpu;
5eb5a6bc 1095 struct timespec *bt;
88c4d2f6 1096 sysclock_t delta;
984263bc 1097
88c4d2f6
MD
1098 do {
1099 tvp->tv_sec = gd->gd_time_seconds;
1100 delta = gd->gd_hardclock.time - gd->gd_cpuclock_base;
1101 } while (tvp->tv_sec != gd->gd_time_seconds);
fad57d0e 1102
044ee7c4
MD
1103 if (delta >= sys_cputimer->freq) {
1104 tvp->tv_sec += delta / sys_cputimer->freq;
1105 delta %= sys_cputimer->freq;
fad57d0e 1106 }
044ee7c4 1107 tvp->tv_usec = (sys_cputimer->freq64_usec * delta) >> 32;
984263bc 1108
5eb5a6bc
MD
1109 bt = &basetime[basetime_index];
1110 tvp->tv_sec += bt->tv_sec;
1111 tvp->tv_usec += bt->tv_nsec / 1000;
88c4d2f6
MD
1112 while (tvp->tv_usec >= 1000000) {
1113 tvp->tv_usec -= 1000000;
1114 ++tvp->tv_sec;
984263bc 1115 }
984263bc
MD
1116}
1117
1118void
88c4d2f6 1119getnanotime(struct timespec *tsp)
984263bc 1120{
88c4d2f6 1121 struct globaldata *gd = mycpu;
5eb5a6bc 1122 struct timespec *bt;
88c4d2f6 1123 sysclock_t delta;
984263bc 1124
88c4d2f6
MD
1125 do {
1126 tsp->tv_sec = gd->gd_time_seconds;
1127 delta = gd->gd_hardclock.time - gd->gd_cpuclock_base;
1128 } while (tsp->tv_sec != gd->gd_time_seconds);
fad57d0e 1129
044ee7c4
MD
1130 if (delta >= sys_cputimer->freq) {
1131 tsp->tv_sec += delta / sys_cputimer->freq;
1132 delta %= sys_cputimer->freq;
fad57d0e 1133 }
044ee7c4 1134 tsp->tv_nsec = (sys_cputimer->freq64_nsec * delta) >> 32;
984263bc 1135
5eb5a6bc
MD
1136 bt = &basetime[basetime_index];
1137 tsp->tv_sec += bt->tv_sec;
1138 tsp->tv_nsec += bt->tv_nsec;
88c4d2f6
MD
1139 while (tsp->tv_nsec >= 1000000000) {
1140 tsp->tv_nsec -= 1000000000;
1141 ++tsp->tv_sec;
984263bc 1142 }
984263bc
MD
1143}
1144
5eb5a6bc
MD
1145static void
1146getnanotime_nbt(struct timespec *nbt, struct timespec *tsp)
1147{
1148 struct globaldata *gd = mycpu;
1149 sysclock_t delta;
1150
1151 do {
1152 tsp->tv_sec = gd->gd_time_seconds;
1153 delta = gd->gd_hardclock.time - gd->gd_cpuclock_base;
1154 } while (tsp->tv_sec != gd->gd_time_seconds);
1155
044ee7c4
MD
1156 if (delta >= sys_cputimer->freq) {
1157 tsp->tv_sec += delta / sys_cputimer->freq;
1158 delta %= sys_cputimer->freq;
5eb5a6bc 1159 }
044ee7c4 1160 tsp->tv_nsec = (sys_cputimer->freq64_nsec * delta) >> 32;
5eb5a6bc
MD
1161
1162 tsp->tv_sec += nbt->tv_sec;
1163 tsp->tv_nsec += nbt->tv_nsec;
1164 while (tsp->tv_nsec >= 1000000000) {
1165 tsp->tv_nsec -= 1000000000;
1166 ++tsp->tv_sec;
1167 }
1168}
1169
1170
88c4d2f6
MD
1171void
1172microtime(struct timeval *tvp)
984263bc 1173{
88c4d2f6 1174 struct globaldata *gd = mycpu;
5eb5a6bc 1175 struct timespec *bt;
88c4d2f6 1176 sysclock_t delta;
984263bc 1177
88c4d2f6
MD
1178 do {
1179 tvp->tv_sec = gd->gd_time_seconds;
044ee7c4 1180 delta = sys_cputimer->count() - gd->gd_cpuclock_base;
88c4d2f6 1181 } while (tvp->tv_sec != gd->gd_time_seconds);
fad57d0e 1182
044ee7c4
MD
1183 if (delta >= sys_cputimer->freq) {
1184 tvp->tv_sec += delta / sys_cputimer->freq;
1185 delta %= sys_cputimer->freq;
fad57d0e 1186 }
044ee7c4 1187 tvp->tv_usec = (sys_cputimer->freq64_usec * delta) >> 32;
984263bc 1188
5eb5a6bc
MD
1189 bt = &basetime[basetime_index];
1190 tvp->tv_sec += bt->tv_sec;
1191 tvp->tv_usec += bt->tv_nsec / 1000;
88c4d2f6
MD
1192 while (tvp->tv_usec >= 1000000) {
1193 tvp->tv_usec -= 1000000;
1194 ++tvp->tv_sec;
984263bc 1195 }
984263bc
MD
1196}
1197
88c4d2f6
MD
1198void
1199nanotime(struct timespec *tsp)
1200{
1201 struct globaldata *gd = mycpu;
5eb5a6bc 1202 struct timespec *bt;
88c4d2f6 1203 sysclock_t delta;
984263bc 1204
88c4d2f6
MD
1205 do {
1206 tsp->tv_sec = gd->gd_time_seconds;
044ee7c4 1207 delta = sys_cputimer->count() - gd->gd_cpuclock_base;
88c4d2f6 1208 } while (tsp->tv_sec != gd->gd_time_seconds);
fad57d0e 1209
044ee7c4
MD
1210 if (delta >= sys_cputimer->freq) {
1211 tsp->tv_sec += delta / sys_cputimer->freq;
1212 delta %= sys_cputimer->freq;
fad57d0e 1213 }
044ee7c4 1214 tsp->tv_nsec = (sys_cputimer->freq64_nsec * delta) >> 32;
984263bc 1215
5eb5a6bc
MD
1216 bt = &basetime[basetime_index];
1217 tsp->tv_sec += bt->tv_sec;
1218 tsp->tv_nsec += bt->tv_nsec;
88c4d2f6
MD
1219 while (tsp->tv_nsec >= 1000000000) {
1220 tsp->tv_nsec -= 1000000000;
1221 ++tsp->tv_sec;
984263bc 1222 }
984263bc
MD
1223}
1224
25b804e7
MD
1225/*
1226 * note: this is not exactly synchronized with real time. To do that we
1227 * would have to do what microtime does and check for a nanoseconds overflow.
1228 */
1229time_t
1230get_approximate_time_t(void)
1231{
1232 struct globaldata *gd = mycpu;
5eb5a6bc
MD
1233 struct timespec *bt;
1234
1235 bt = &basetime[basetime_index];
1236 return(gd->gd_time_seconds + bt->tv_sec);
25b804e7
MD
1237}
1238
984263bc
MD
1239int
1240pps_ioctl(u_long cmd, caddr_t data, struct pps_state *pps)
1241{
1242 pps_params_t *app;
1243 struct pps_fetch_args *fapi;
1244#ifdef PPS_SYNC
1245 struct pps_kcbind_args *kapi;
1246#endif
1247
1248 switch (cmd) {
1249 case PPS_IOC_CREATE:
1250 return (0);
1251 case PPS_IOC_DESTROY:
1252 return (0);
1253 case PPS_IOC_SETPARAMS:
1254 app = (pps_params_t *)data;
1255 if (app->mode & ~pps->ppscap)
1256 return (EINVAL);
1257 pps->ppsparam = *app;
1258 return (0);
1259 case PPS_IOC_GETPARAMS:
1260 app = (pps_params_t *)data;
1261 *app = pps->ppsparam;
1262 app->api_version = PPS_API_VERS_1;
1263 return (0);
1264 case PPS_IOC_GETCAP:
1265 *(int*)data = pps->ppscap;
1266 return (0);
1267 case PPS_IOC_FETCH:
1268 fapi = (struct pps_fetch_args *)data;
1269 if (fapi->tsformat && fapi->tsformat != PPS_TSFMT_TSPEC)
1270 return (EINVAL);
1271 if (fapi->timeout.tv_sec || fapi->timeout.tv_nsec)
1272 return (EOPNOTSUPP);
1273 pps->ppsinfo.current_mode = pps->ppsparam.mode;
1274 fapi->pps_info_buf = pps->ppsinfo;
1275 return (0);
1276 case PPS_IOC_KCBIND:
1277#ifdef PPS_SYNC
1278 kapi = (struct pps_kcbind_args *)data;
1279 /* XXX Only root should be able to do this */
1280 if (kapi->tsformat && kapi->tsformat != PPS_TSFMT_TSPEC)
1281 return (EINVAL);
1282 if (kapi->kernel_consumer != PPS_KC_HARDPPS)
1283 return (EINVAL);
1284 if (kapi->edge & ~pps->ppscap)
1285 return (EINVAL);
1286 pps->kcmode = kapi->edge;
1287 return (0);
1288#else
1289 return (EOPNOTSUPP);
1290#endif
1291 default:
1292 return (ENOTTY);
1293 }
1294}
1295
1296void
1297pps_init(struct pps_state *pps)
1298{
1299 pps->ppscap |= PPS_TSFMT_TSPEC;
1300 if (pps->ppscap & PPS_CAPTUREASSERT)
1301 pps->ppscap |= PPS_OFFSETASSERT;
1302 if (pps->ppscap & PPS_CAPTURECLEAR)
1303 pps->ppscap |= PPS_OFFSETCLEAR;
1304}
1305
1306void
88c4d2f6 1307pps_event(struct pps_state *pps, sysclock_t count, int event)
984263bc 1308{
88c4d2f6
MD
1309 struct globaldata *gd;
1310 struct timespec *tsp;
1311 struct timespec *osp;
5eb5a6bc 1312 struct timespec *bt;
88c4d2f6
MD
1313 struct timespec ts;
1314 sysclock_t *pcount;
1315#ifdef PPS_SYNC
1316 sysclock_t tcount;
1317#endif
1318 sysclock_t delta;
1319 pps_seq_t *pseq;
1320 int foff;
1321 int fhard;
1322
1323 gd = mycpu;
984263bc
MD
1324
1325 /* Things would be easier with arrays... */
1326 if (event == PPS_CAPTUREASSERT) {
1327 tsp = &pps->ppsinfo.assert_timestamp;
1328 osp = &pps->ppsparam.assert_offset;
1329 foff = pps->ppsparam.mode & PPS_OFFSETASSERT;
1330 fhard = pps->kcmode & PPS_CAPTUREASSERT;
1331 pcount = &pps->ppscount[0];
1332 pseq = &pps->ppsinfo.assert_sequence;
1333 } else {
1334 tsp = &pps->ppsinfo.clear_timestamp;
1335 osp = &pps->ppsparam.clear_offset;
1336 foff = pps->ppsparam.mode & PPS_OFFSETCLEAR;
1337 fhard = pps->kcmode & PPS_CAPTURECLEAR;
1338 pcount = &pps->ppscount[1];
1339 pseq = &pps->ppsinfo.clear_sequence;
1340 }
1341
984263bc
MD
1342 /* Nothing really happened */
1343 if (*pcount == count)
1344 return;
1345
1346 *pcount = count;
1347
88c4d2f6
MD
1348 do {
1349 ts.tv_sec = gd->gd_time_seconds;
1350 delta = count - gd->gd_cpuclock_base;
1351 } while (ts.tv_sec != gd->gd_time_seconds);
fad57d0e 1352
044ee7c4
MD
1353 if (delta >= sys_cputimer->freq) {
1354 ts.tv_sec += delta / sys_cputimer->freq;
1355 delta %= sys_cputimer->freq;
88c4d2f6 1356 }
044ee7c4 1357 ts.tv_nsec = (sys_cputimer->freq64_nsec * delta) >> 32;
5eb5a6bc
MD
1358 bt = &basetime[basetime_index];
1359 ts.tv_sec += bt->tv_sec;
1360 ts.tv_nsec += bt->tv_nsec;
88c4d2f6
MD
1361 while (ts.tv_nsec >= 1000000000) {
1362 ts.tv_nsec -= 1000000000;
1363 ++ts.tv_sec;
984263bc 1364 }
984263bc
MD
1365
1366 (*pseq)++;
1367 *tsp = ts;
1368
1369 if (foff) {
1370 timespecadd(tsp, osp);
1371 if (tsp->tv_nsec < 0) {
1372 tsp->tv_nsec += 1000000000;
1373 tsp->tv_sec -= 1;
1374 }
1375 }
1376#ifdef PPS_SYNC
1377 if (fhard) {
1378 /* magic, at its best... */
1379 tcount = count - pps->ppscount[2];
1380 pps->ppscount[2] = count;
044ee7c4
MD
1381 if (tcount >= sys_cputimer->freq) {
1382 delta = (1000000000 * (tcount / sys_cputimer->freq) +
1383 sys_cputimer->freq64_nsec *
1384 (tcount % sys_cputimer->freq)) >> 32;
fad57d0e 1385 } else {
044ee7c4 1386 delta = (sys_cputimer->freq64_nsec * tcount) >> 32;
fad57d0e 1387 }
984263bc
MD
1388 hardpps(tsp, delta);
1389 }
1390#endif
1391}
88c4d2f6 1392
d2412a2e
MD
1393/*
1394 * Return the tsc target value for a delay of (ns).
1395 *
1396 * Returns -1 if the TSC is not supported.
1397 */
1398int64_t
1399tsc_get_target(int ns)
1400{
1401#if defined(_RDTSC_SUPPORTED_)
1402 if (cpu_feature & CPUID_TSC) {
1403 return (rdtsc() + tsc_frequency * ns / (int64_t)1000000000);
1404 }
1405#endif
1406 return(-1);
1407}
1408
1409/*
1410 * Compare the tsc against the passed target
1411 *
1412 * Returns +1 if the target has been reached
1413 * Returns 0 if the target has not yet been reached
1414 * Returns -1 if the TSC is not supported.
1415 *
1416 * Typical use: while (tsc_test_target(target) == 0) { ...poll... }
1417 */
1418int
1419tsc_test_target(int64_t target)
1420{
1421#if defined(_RDTSC_SUPPORTED_)
1422 if (cpu_feature & CPUID_TSC) {
1423 if ((int64_t)(target - rdtsc()) <= 0)
1424 return(1);
1425 return(0);
1426 }
d2412a2e 1427#endif
2e537993 1428 return(-1);
d2412a2e 1429}
b12defdc
MD
1430
1431/*
1432 * Delay the specified number of nanoseconds using the tsc. This function
1433 * returns immediately if the TSC is not supported. At least one cpu_pause()
1434 * will be issued.
1435 */
1436void
1437tsc_delay(int ns)
1438{
1439 int64_t clk;
1440
1441 clk = tsc_get_target(ns);
1442 cpu_pause();
1443 while (tsc_test_target(clk) == 0)
1444 cpu_pause();
1445}