usched_bsd4 - Topology-aware scheduling
[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 $
c730be20 73 * $DragonFly: src/sys/kern/kern_clock.c,v 1.62 2008/09/09 04:06:13 dillon Exp $
984263bc
MD
74 */
75
76#include "opt_ntp.h"
2b71c8f1 77#include "opt_polling.h"
b3a7093f 78#include "opt_ifpoll.h"
07522099 79#include "opt_pctrack.h"
984263bc
MD
80
81#include <sys/param.h>
82#include <sys/systm.h>
984263bc
MD
83#include <sys/callout.h>
84#include <sys/kernel.h>
f5d21610 85#include <sys/kinfo.h>
984263bc
MD
86#include <sys/proc.h>
87#include <sys/malloc.h>
d70eef28 88#include <sys/resource.h>
984263bc
MD
89#include <sys/resourcevar.h>
90#include <sys/signalvar.h>
91#include <sys/timex.h>
92#include <sys/timepps.h>
93#include <vm/vm.h>
94#include <sys/lock.h>
95#include <vm/pmap.h>
96#include <vm/vm_map.h>
5ffd1608 97#include <vm/vm_extern.h>
984263bc 98#include <sys/sysctl.h>
684a93c4 99
2689779e 100#include <sys/thread2.h>
984263bc
MD
101
102#include <machine/cpu.h>
103#include <machine/limits.h>
104#include <machine/smp.h>
d2412a2e
MD
105#include <machine/cpufunc.h>
106#include <machine/specialreg.h>
107#include <machine/clock.h>
984263bc
MD
108
109#ifdef GPROF
110#include <sys/gmon.h>
111#endif
112
113#ifdef DEVICE_POLLING
94ebffcd 114extern void init_device_poll_pcpu(int);
3e61f60e 115#endif
984263bc 116
b3a7093f
SZ
117#ifdef IFPOLL_ENABLE
118extern void ifpoll_init_pcpu(int);
119#endif
120
07522099
MD
121#ifdef DEBUG_PCTRACK
122static void do_pctrack(struct intrframe *frame, int which);
123#endif
124
402ed7e1 125static void initclocks (void *dummy);
ba39e2e0 126SYSINIT(clocks, SI_BOOT2_CLOCKS, SI_ORDER_FIRST, initclocks, NULL)
984263bc 127
6ad39cae
MD
128/*
129 * Some of these don't belong here, but it's easiest to concentrate them.
9eea7f0c 130 * Note that cpu_time counts in microseconds, but most userland programs
6ad39cae
MD
131 * just compare relative times against the total by delta.
132 */
9eea7f0c 133struct kinfo_cputime cputime_percpu[MAXCPU];
07522099
MD
134#ifdef DEBUG_PCTRACK
135struct kinfo_pcheader cputime_pcheader = { PCTRACK_SIZE, PCTRACK_ARYSIZE };
136struct kinfo_pctrack cputime_pctrack[MAXCPU][PCTRACK_SIZE];
137#endif
138
9eea7f0c
HP
139#ifdef SMP
140static int
141sysctl_cputime(SYSCTL_HANDLER_ARGS)
142{
143 int cpu, error = 0;
144 size_t size = sizeof(struct kinfo_cputime);
145
146 for (cpu = 0; cpu < ncpus; ++cpu) {
147 if ((error = SYSCTL_OUT(req, &cputime_percpu[cpu], size)))
148 break;
149 }
984263bc 150
9eea7f0c
HP
151 return (error);
152}
153SYSCTL_PROC(_kern, OID_AUTO, cputime, (CTLTYPE_OPAQUE|CTLFLAG_RD), 0, 0,
154 sysctl_cputime, "S,kinfo_cputime", "CPU time statistics");
155#else
156SYSCTL_STRUCT(_kern, OID_AUTO, cputime, CTLFLAG_RD, &cpu_time, kinfo_cputime,
157 "CPU time statistics");
158#endif
984263bc 159
06636a8e
AHJ
160static int
161sysctl_cp_time(SYSCTL_HANDLER_ARGS)
162{
163 long cpu_states[5] = {0};
164 int cpu, error = 0;
165 size_t size = sizeof(cpu_states);
166
167 for (cpu = 0; cpu < ncpus; ++cpu) {
d70eef28
SG
168 cpu_states[CP_USER] += cputime_percpu[cpu].cp_user;
169 cpu_states[CP_NICE] += cputime_percpu[cpu].cp_nice;
170 cpu_states[CP_SYS] += cputime_percpu[cpu].cp_sys;
171 cpu_states[CP_INTR] += cputime_percpu[cpu].cp_intr;
172 cpu_states[CP_IDLE] += cputime_percpu[cpu].cp_idle;
06636a8e
AHJ
173 }
174
175 error = SYSCTL_OUT(req, cpu_states, size);
176
177 return (error);
178}
179
180SYSCTL_PROC(_kern, OID_AUTO, cp_time, (CTLTYPE_LONG|CTLFLAG_RD), 0, 0,
181 sysctl_cp_time, "LU", "CPU time statistics");
182
88c4d2f6
MD
183/*
184 * boottime is used to calculate the 'real' uptime. Do not confuse this with
185 * microuptime(). microtime() is not drift compensated. The real uptime
60b2809b
MD
186 * with compensation is nanotime() - bootime. boottime is recalculated
187 * whenever the real time is set based on the compensated elapsed time
188 * in seconds (gd->gd_time_seconds).
88c4d2f6 189 *
88c4d2f6
MD
190 * The gd_time_seconds and gd_cpuclock_base fields remain fairly monotonic.
191 * Slight adjustments to gd_cpuclock_base are made to phase-lock it to
192 * the real time.
193 */
194struct timespec boottime; /* boot time (realtime) for reference only */
88c4d2f6 195time_t time_second; /* read-only 'passive' uptime in seconds */
984263bc 196
5eb5a6bc
MD
197/*
198 * basetime is used to calculate the compensated real time of day. The
199 * basetime can be modified on a per-tick basis by the adjtime(),
200 * ntp_adjtime(), and sysctl-based time correction APIs.
201 *
202 * Note that frequency corrections can also be made by adjusting
203 * gd_cpuclock_base.
204 *
205 * basetime is a tail-chasing FIFO, updated only by cpu #0. The FIFO is
206 * used on both SMP and UP systems to avoid MP races between cpu's and
207 * interrupt races on UP systems.
208 */
209#define BASETIME_ARYSIZE 16
210#define BASETIME_ARYMASK (BASETIME_ARYSIZE - 1)
211static struct timespec basetime[BASETIME_ARYSIZE];
212static volatile int basetime_index;
213
214static int
215sysctl_get_basetime(SYSCTL_HANDLER_ARGS)
216{
217 struct timespec *bt;
218 int error;
35238fa5 219 int index;
5eb5a6bc 220
35238fa5
MD
221 /*
222 * Because basetime data and index may be updated by another cpu,
223 * a load fence is required to ensure that the data we read has
224 * not been speculatively read relative to a possibly updated index.
225 */
226 index = basetime_index;
227 cpu_lfence();
228 bt = &basetime[index];
08f95c49 229 error = SYSCTL_OUT(req, bt, sizeof(*bt));
5eb5a6bc
MD
230 return (error);
231}
232
984263bc 233SYSCTL_STRUCT(_kern, KERN_BOOTTIME, boottime, CTLFLAG_RD,
08f95c49 234 &boottime, timespec, "System boottime");
5eb5a6bc 235SYSCTL_PROC(_kern, OID_AUTO, basetime, CTLTYPE_STRUCT|CTLFLAG_RD, 0, 0,
08f95c49 236 sysctl_get_basetime, "S,timespec", "System basetime");
984263bc 237
96d52ac8
SZ
238static void hardclock(systimer_t info, int, struct intrframe *frame);
239static void statclock(systimer_t info, int, struct intrframe *frame);
240static void schedclock(systimer_t info, int, struct intrframe *frame);
5eb5a6bc 241static void getnanotime_nbt(struct timespec *nbt, struct timespec *tsp);
88c4d2f6
MD
242
243int ticks; /* system master ticks at hz */
da3639ef 244int clocks_running; /* tsleep/timeout clocks operational */
88c4d2f6
MD
245int64_t nsec_adj; /* ntpd per-tick adjustment in nsec << 32 */
246int64_t nsec_acc; /* accumulator */
d6d39bc7 247int sched_ticks; /* global schedule clock ticks */
984263bc 248
4026c000
JS
249/* NTPD time correction fields */
250int64_t ntp_tick_permanent; /* per-tick adjustment in nsec << 32 */
251int64_t ntp_tick_acc; /* accumulator for per-tick adjustment */
252int64_t ntp_delta; /* one-time correction in nsec */
253int64_t ntp_big_delta = 1000000000;
254int32_t ntp_tick_delta; /* current adjustment rate */
255int32_t ntp_default_tick_delta; /* adjustment rate for ntp_delta */
48590578
JS
256time_t ntp_leap_second; /* time of next leap second */
257int ntp_leap_insert; /* whether to insert or remove a second */
4026c000 258
984263bc 259/*
88c4d2f6 260 * Finish initializing clock frequencies and start all clocks running.
984263bc 261 */
88c4d2f6
MD
262/* ARGSUSED*/
263static void
264initclocks(void *dummy)
984263bc 265{
88c4d2f6
MD
266 /*psratio = profhz / stathz;*/
267 initclocks_pcpu();
da3639ef 268 clocks_running = 1;
984263bc
MD
269}
270
88c4d2f6
MD
271/*
272 * Called on a per-cpu basis
273 */
274void
275initclocks_pcpu(void)
276{
277 struct globaldata *gd = mycpu;
984263bc 278
88c4d2f6
MD
279 crit_enter();
280 if (gd->gd_cpuid == 0) {
281 gd->gd_time_seconds = 1;
044ee7c4 282 gd->gd_cpuclock_base = sys_cputimer->count();
88c4d2f6
MD
283 } else {
284 /* XXX */
285 gd->gd_time_seconds = globaldata_find(0)->gd_time_seconds;
286 gd->gd_cpuclock_base = globaldata_find(0)->gd_cpuclock_base;
287 }
0d1dffdf 288
43adde98
SZ
289 systimer_intr_enable();
290
94ebffcd
SZ
291#ifdef DEVICE_POLLING
292 init_device_poll_pcpu(gd->gd_cpuid);
293#endif
294
b3a7093f
SZ
295#ifdef IFPOLL_ENABLE
296 ifpoll_init_pcpu(gd->gd_cpuid);
297#endif
298
0d1dffdf
MD
299 /*
300 * Use a non-queued periodic systimer to prevent multiple ticks from
301 * building up if the sysclock jumps forward (8254 gets reset). The
302 * sysclock will never jump backwards. Our time sync is based on
303 * the actual sysclock, not the ticks count.
304 */
305 systimer_init_periodic_nq(&gd->gd_hardclock, hardclock, NULL, hz);
306 systimer_init_periodic_nq(&gd->gd_statclock, statclock, NULL, stathz);
88c4d2f6 307 /* XXX correct the frequency for scheduler / estcpu tests */
0d1dffdf 308 systimer_init_periodic_nq(&gd->gd_schedclock, schedclock,
8478264a 309 NULL, ESTCPUFREQ);
88c4d2f6
MD
310 crit_exit();
311}
984263bc
MD
312
313/*
88c4d2f6
MD
314 * This sets the current real time of day. Timespecs are in seconds and
315 * nanoseconds. We do not mess with gd_time_seconds and gd_cpuclock_base,
316 * instead we adjust basetime so basetime + gd_* results in the current
317 * time of day. This way the gd_* fields are guarenteed to represent
318 * a monotonically increasing 'uptime' value.
5eb5a6bc
MD
319 *
320 * When set_timeofday() is called from userland, the system call forces it
321 * onto cpu #0 since only cpu #0 can update basetime_index.
984263bc 322 */
88c4d2f6
MD
323void
324set_timeofday(struct timespec *ts)
325{
5eb5a6bc
MD
326 struct timespec *nbt;
327 int ni;
984263bc 328
88c4d2f6
MD
329 /*
330 * XXX SMP / non-atomic basetime updates
331 */
332 crit_enter();
5eb5a6bc
MD
333 ni = (basetime_index + 1) & BASETIME_ARYMASK;
334 nbt = &basetime[ni];
335 nanouptime(nbt);
336 nbt->tv_sec = ts->tv_sec - nbt->tv_sec;
337 nbt->tv_nsec = ts->tv_nsec - nbt->tv_nsec;
338 if (nbt->tv_nsec < 0) {
339 nbt->tv_nsec += 1000000000;
340 --nbt->tv_sec;
88c4d2f6 341 }
a81931cc
MD
342
343 /*
344 * Note that basetime diverges from boottime as the clock drift is
345 * compensated for, so we cannot do away with boottime. When setting
346 * the absolute time of day the drift is 0 (for an instant) and we
347 * can simply assign boottime to basetime.
348 *
349 * Note that nanouptime() is based on gd_time_seconds which is drift
350 * compensated up to a point (it is guarenteed to remain monotonically
351 * increasing). gd_time_seconds is thus our best uptime guess and
352 * suitable for use in the boottime calculation. It is already taken
353 * into account in the basetime calculation above.
354 */
5eb5a6bc 355 boottime.tv_sec = nbt->tv_sec;
4026c000 356 ntp_delta = 0;
5eb5a6bc
MD
357
358 /*
35238fa5
MD
359 * We now have a new basetime, make sure all other cpus have it,
360 * then update the index.
5eb5a6bc 361 */
35238fa5 362 cpu_sfence();
5eb5a6bc
MD
363 basetime_index = ni;
364
88c4d2f6
MD
365 crit_exit();
366}
367
984263bc 368/*
88c4d2f6
MD
369 * Each cpu has its own hardclock, but we only increments ticks and softticks
370 * on cpu #0.
371 *
372 * NOTE! systimer! the MP lock might not be held here. We can only safely
373 * manipulate objects owned by the current cpu.
984263bc 374 */
984263bc 375static void
96d52ac8 376hardclock(systimer_t info, int in_ipi __unused, struct intrframe *frame)
984263bc 377{
88c4d2f6
MD
378 sysclock_t cputicks;
379 struct proc *p;
88c4d2f6 380 struct globaldata *gd = mycpu;
984263bc
MD
381
382 /*
88c4d2f6
MD
383 * Realtime updates are per-cpu. Note that timer corrections as
384 * returned by microtime() and friends make an additional adjustment
385 * using a system-wise 'basetime', but the running time is always
386 * taken from the per-cpu globaldata area. Since the same clock
387 * is distributing (XXX SMP) to all cpus, the per-cpu timebases
388 * stay in synch.
389 *
390 * Note that we never allow info->time (aka gd->gd_hardclock.time)
fad57d0e
MD
391 * to reverse index gd_cpuclock_base, but that it is possible for
392 * it to temporarily get behind in the seconds if something in the
393 * system locks interrupts for a long period of time. Since periodic
394 * timers count events, though everything should resynch again
395 * immediately.
984263bc 396 */
88c4d2f6 397 cputicks = info->time - gd->gd_cpuclock_base;
044ee7c4 398 if (cputicks >= sys_cputimer->freq) {
88c4d2f6 399 ++gd->gd_time_seconds;
044ee7c4 400 gd->gd_cpuclock_base += sys_cputimer->freq;
88c4d2f6 401 }
984263bc
MD
402
403 /*
92b561b7
MD
404 * The system-wide ticks counter and NTP related timedelta/tickdelta
405 * adjustments only occur on cpu #0. NTP adjustments are accomplished
406 * by updating basetime.
984263bc 407 */
88c4d2f6 408 if (gd->gd_cpuid == 0) {
5eb5a6bc 409 struct timespec *nbt;
88c4d2f6
MD
410 struct timespec nts;
411 int leap;
5eb5a6bc 412 int ni;
984263bc 413
88c4d2f6 414 ++ticks;
984263bc 415
88c4d2f6
MD
416#if 0
417 if (tco->tc_poll_pps)
418 tco->tc_poll_pps(tco);
419#endif
5eb5a6bc 420
88c4d2f6 421 /*
5eb5a6bc
MD
422 * Calculate the new basetime index. We are in a critical section
423 * on cpu #0 and can safely play with basetime_index. Start
424 * with the current basetime and then make adjustments.
425 */
426 ni = (basetime_index + 1) & BASETIME_ARYMASK;
427 nbt = &basetime[ni];
428 *nbt = basetime[basetime_index];
429
430 /*
431 * Apply adjtime corrections. (adjtime() API)
432 *
433 * adjtime() only runs on cpu #0 so our critical section is
434 * sufficient to access these variables.
88c4d2f6 435 */
4026c000 436 if (ntp_delta != 0) {
5eb5a6bc 437 nbt->tv_nsec += ntp_tick_delta;
4026c000
JS
438 ntp_delta -= ntp_tick_delta;
439 if ((ntp_delta > 0 && ntp_delta < ntp_tick_delta) ||
440 (ntp_delta < 0 && ntp_delta > ntp_tick_delta)) {
5eb5a6bc 441 ntp_tick_delta = ntp_delta;
4026c000
JS
442 }
443 }
444
5eb5a6bc
MD
445 /*
446 * Apply permanent frequency corrections. (sysctl API)
447 */
4026c000
JS
448 if (ntp_tick_permanent != 0) {
449 ntp_tick_acc += ntp_tick_permanent;
450 if (ntp_tick_acc >= (1LL << 32)) {
5eb5a6bc 451 nbt->tv_nsec += ntp_tick_acc >> 32;
331bc6f8 452 ntp_tick_acc -= (ntp_tick_acc >> 32) << 32;
4026c000 453 } else if (ntp_tick_acc <= -(1LL << 32)) {
331bc6f8 454 /* Negate ntp_tick_acc to avoid shifting the sign bit. */
5eb5a6bc 455 nbt->tv_nsec -= (-ntp_tick_acc) >> 32;
331bc6f8 456 ntp_tick_acc += ((-ntp_tick_acc) >> 32) << 32;
4026c000
JS
457 }
458 }
459
5eb5a6bc
MD
460 if (nbt->tv_nsec >= 1000000000) {
461 nbt->tv_sec++;
462 nbt->tv_nsec -= 1000000000;
463 } else if (nbt->tv_nsec < 0) {
464 nbt->tv_sec--;
465 nbt->tv_nsec += 1000000000;
88c4d2f6
MD
466 }
467
468 /*
5eb5a6bc 469 * Another per-tick compensation. (for ntp_adjtime() API)
88c4d2f6 470 */
5eb5a6bc 471 if (nsec_adj != 0) {
88c4d2f6
MD
472 nsec_acc += nsec_adj;
473 if (nsec_acc >= 0x100000000LL) {
5eb5a6bc 474 nbt->tv_nsec += nsec_acc >> 32;
88c4d2f6
MD
475 nsec_acc = (nsec_acc & 0xFFFFFFFFLL);
476 } else if (nsec_acc <= -0x100000000LL) {
5eb5a6bc 477 nbt->tv_nsec -= -nsec_acc >> 32;
88c4d2f6
MD
478 nsec_acc = -(-nsec_acc & 0xFFFFFFFFLL);
479 }
5eb5a6bc
MD
480 if (nbt->tv_nsec >= 1000000000) {
481 nbt->tv_nsec -= 1000000000;
482 ++nbt->tv_sec;
483 } else if (nbt->tv_nsec < 0) {
484 nbt->tv_nsec += 1000000000;
485 --nbt->tv_sec;
486 }
487 }
488
489 /************************************************************
490 * LEAP SECOND CORRECTION *
491 ************************************************************
492 *
493 * Taking into account all the corrections made above, figure
494 * out the new real time. If the seconds field has changed
495 * then apply any pending leap-second corrections.
496 */
497 getnanotime_nbt(nbt, &nts);
498
32040d57
MD
499 if (time_second != nts.tv_sec) {
500 /*
501 * Apply leap second (sysctl API). Adjust nts for changes
502 * so we do not have to call getnanotime_nbt again.
503 */
504 if (ntp_leap_second) {
505 if (ntp_leap_second == nts.tv_sec) {
506 if (ntp_leap_insert) {
507 nbt->tv_sec++;
508 nts.tv_sec++;
509 } else {
510 nbt->tv_sec--;
511 nts.tv_sec--;
512 }
5eb5a6bc 513 ntp_leap_second--;
32040d57 514 }
88c4d2f6 515 }
88c4d2f6 516
32040d57
MD
517 /*
518 * Apply leap second (ntp_adjtime() API), calculate a new
519 * nsec_adj field. ntp_update_second() returns nsec_adj
520 * as a per-second value but we need it as a per-tick value.
521 */
88c4d2f6 522 leap = ntp_update_second(time_second, &nsec_adj);
88c4d2f6 523 nsec_adj /= hz;
32040d57
MD
524 nbt->tv_sec += leap;
525 nts.tv_sec += leap;
526
527 /*
528 * Update the time_second 'approximate time' global.
529 */
530 time_second = nts.tv_sec;
88c4d2f6 531 }
5eb5a6bc
MD
532
533 /*
534 * Finally, our new basetime is ready to go live!
535 */
35238fa5 536 cpu_sfence();
5eb5a6bc 537 basetime_index = ni;
88c4d2f6
MD
538 }
539
540 /*
f9235b6d
MD
541 * lwkt thread scheduler fair queueing
542 */
85946b6c 543 lwkt_schedulerclock(curthread);
f9235b6d
MD
544
545 /*
92b561b7
MD
546 * softticks are handled for all cpus
547 */
548 hardclock_softtick(gd);
549
550 /*
8582ec21
MD
551 * ITimer handling is per-tick, per-cpu.
552 *
553 * We must acquire the per-process token in order for ksignal()
898e34b3
MD
554 * to be non-blocking. For the moment this requires an AST fault,
555 * the ksignal() cannot be safely issued from this hard interrupt.
556 *
557 * XXX Even the trytoken here isn't right, and itimer operation in
558 * a multi threaded environment is going to be weird at the
559 * very least.
88c4d2f6 560 */
8582ec21 561 if ((p = curproc) != NULL && lwkt_trytoken(&p->p_token)) {
3dbbd6dd 562 crit_enter_hard();
88c4d2f6 563 if (frame && CLKF_USERMODE(frame) &&
93328593 564 timevalisset(&p->p_timer[ITIMER_VIRTUAL].it_value) &&
898e34b3 565 itimerdecr(&p->p_timer[ITIMER_VIRTUAL], ustick) == 0) {
4643740a 566 p->p_flags |= P_SIGVTALRM;
898e34b3
MD
567 need_user_resched();
568 }
93328593 569 if (timevalisset(&p->p_timer[ITIMER_PROF].it_value) &&
898e34b3 570 itimerdecr(&p->p_timer[ITIMER_PROF], ustick) == 0) {
4643740a 571 p->p_flags |= P_SIGPROF;
898e34b3
MD
572 need_user_resched();
573 }
3dbbd6dd 574 crit_exit_hard();
8582ec21 575 lwkt_reltoken(&p->p_token);
984263bc 576 }
604e1e09 577 setdelayed();
88c4d2f6 578}
984263bc 579
88c4d2f6
MD
580/*
581 * The statistics clock typically runs at a 125Hz rate, and is intended
582 * to be frequency offset from the hardclock (typ 100Hz). It is per-cpu.
583 *
584 * NOTE! systimer! the MP lock might not be held here. We can only safely
585 * manipulate objects owned by the current cpu.
586 *
587 * The stats clock is responsible for grabbing a profiling sample.
588 * Most of the statistics are only used by user-level statistics programs.
589 * The main exceptions are p->p_uticks, p->p_sticks, p->p_iticks, and
590 * p->p_estcpu.
591 *
592 * Like the other clocks, the stat clock is called from what is effectively
593 * a fast interrupt, so the context should be the thread/process that got
594 * interrupted.
595 */
596static void
96d52ac8 597statclock(systimer_t info, int in_ipi, struct intrframe *frame)
88c4d2f6
MD
598{
599#ifdef GPROF
600 struct gmonparam *g;
601 int i;
984263bc 602#endif
88c4d2f6
MD
603 thread_t td;
604 struct proc *p;
605 int bump;
606 struct timeval tv;
607 struct timeval *stv;
984263bc
MD
608
609 /*
88c4d2f6 610 * How big was our timeslice relative to the last time?
984263bc 611 */
88c4d2f6
MD
612 microuptime(&tv); /* mpsafe */
613 stv = &mycpu->gd_stattv;
614 if (stv->tv_sec == 0) {
615 bump = 1;
616 } else {
617 bump = tv.tv_usec - stv->tv_usec +
618 (tv.tv_sec - stv->tv_sec) * 1000000;
619 if (bump < 0)
620 bump = 0;
621 if (bump > 1000000)
622 bump = 1000000;
623 }
624 *stv = tv;
984263bc 625
88c4d2f6
MD
626 td = curthread;
627 p = td->td_proc;
984263bc 628
88c4d2f6
MD
629 if (frame && CLKF_USERMODE(frame)) {
630 /*
631 * Came from userland, handle user time and deal with
632 * possible process.
633 */
4643740a 634 if (p && (p->p_flags & P_PROFIL))
88c4d2f6
MD
635 addupc_intr(p, CLKF_PC(frame), 1);
636 td->td_uticks += bump;
984263bc 637
88c4d2f6
MD
638 /*
639 * Charge the time as appropriate
640 */
641 if (p && p->p_nice > NZERO)
9eea7f0c 642 cpu_time.cp_nice += bump;
88c4d2f6 643 else
9eea7f0c 644 cpu_time.cp_user += bump;
88c4d2f6 645 } else {
96d52ac8
SZ
646 int intr_nest = mycpu->gd_intr_nesting_level;
647
648 if (in_ipi) {
649 /*
650 * IPI processing code will bump gd_intr_nesting_level
651 * up by one, which breaks following CLKF_INTR testing,
652 * so we substract it by one here.
653 */
654 --intr_nest;
655 }
88c4d2f6
MD
656#ifdef GPROF
657 /*
658 * Kernel statistics are just like addupc_intr, only easier.
659 */
660 g = &_gmonparam;
661 if (g->state == GMON_PROF_ON && frame) {
662 i = CLKF_PC(frame) - g->lowpc;
663 if (i < g->textsize) {
664 i /= HISTFRACTION * sizeof(*g->kcount);
665 g->kcount[i]++;
666 }
667 }
668#endif
6026c54d
SZ
669
670#define IS_INTR_RUNNING ((frame && CLKF_INTR(intr_nest)) || CLKF_INTR_TD(td))
671
88c4d2f6
MD
672 /*
673 * Came from kernel mode, so we were:
674 * - handling an interrupt,
675 * - doing syscall or trap work on behalf of the current
676 * user process, or
677 * - spinning in the idle loop.
678 * Whichever it is, charge the time as appropriate.
679 * Note that we charge interrupts to the current process,
680 * regardless of whether they are ``for'' that process,
681 * so that we know how much of its real time was spent
682 * in ``non-process'' (i.e., interrupt) work.
683 *
684 * XXX assume system if frame is NULL. A NULL frame
e43a034f 685 * can occur if ipi processing is done from a crit_exit().
88c4d2f6 686 */
6026c54d 687 if (IS_INTR_RUNNING)
88c4d2f6
MD
688 td->td_iticks += bump;
689 else
690 td->td_sticks += bump;
691
6026c54d 692 if (IS_INTR_RUNNING) {
07522099 693#ifdef DEBUG_PCTRACK
6026c54d
SZ
694 if (frame)
695 do_pctrack(frame, PCTRACK_INT);
07522099 696#endif
9eea7f0c 697 cpu_time.cp_intr += bump;
88c4d2f6 698 } else {
07522099 699 if (td == &mycpu->gd_idlethread) {
9eea7f0c 700 cpu_time.cp_idle += bump;
07522099
MD
701 } else {
702#ifdef DEBUG_PCTRACK
703 if (frame)
704 do_pctrack(frame, PCTRACK_SYS);
705#endif
9eea7f0c 706 cpu_time.cp_sys += bump;
07522099 707 }
88c4d2f6 708 }
6026c54d
SZ
709
710#undef IS_INTR_RUNNING
88c4d2f6
MD
711 }
712}
713
07522099
MD
714#ifdef DEBUG_PCTRACK
715/*
716 * Sample the PC when in the kernel or in an interrupt. User code can
717 * retrieve the information and generate a histogram or other output.
718 */
719
720static void
721do_pctrack(struct intrframe *frame, int which)
722{
723 struct kinfo_pctrack *pctrack;
724
725 pctrack = &cputime_pctrack[mycpu->gd_cpuid][which];
726 pctrack->pc_array[pctrack->pc_index & PCTRACK_ARYMASK] =
727 (void *)CLKF_PC(frame);
728 ++pctrack->pc_index;
729}
730
731static int
732sysctl_pctrack(SYSCTL_HANDLER_ARGS)
733{
734 struct kinfo_pcheader head;
735 int error;
736 int cpu;
737 int ntrack;
738
739 head.pc_ntrack = PCTRACK_SIZE;
740 head.pc_arysize = PCTRACK_ARYSIZE;
741
742 if ((error = SYSCTL_OUT(req, &head, sizeof(head))) != 0)
743 return (error);
744
745 for (cpu = 0; cpu < ncpus; ++cpu) {
746 for (ntrack = 0; ntrack < PCTRACK_SIZE; ++ntrack) {
747 error = SYSCTL_OUT(req, &cputime_pctrack[cpu][ntrack],
748 sizeof(struct kinfo_pctrack));
749 if (error)
750 break;
751 }
752 if (error)
753 break;
754 }
755 return (error);
756}
757SYSCTL_PROC(_kern, OID_AUTO, pctrack, (CTLTYPE_OPAQUE|CTLFLAG_RD), 0, 0,
758 sysctl_pctrack, "S,kinfo_pcheader", "CPU PC tracking");
759
760#endif
761
88c4d2f6 762/*
dcc99b62 763 * The scheduler clock typically runs at a 50Hz rate. NOTE! systimer,
88c4d2f6
MD
764 * the MP lock might not be held. We can safely manipulate parts of curproc
765 * but that's about it.
dcc99b62
MD
766 *
767 * Each cpu has its own scheduler clock.
88c4d2f6
MD
768 */
769static void
96d52ac8 770schedclock(systimer_t info, int in_ipi __unused, struct intrframe *frame)
88c4d2f6 771{
553ea3c8 772 struct lwp *lp;
88c4d2f6
MD
773 struct rusage *ru;
774 struct vmspace *vm;
775 long rss;
776
553ea3c8 777 if ((lp = lwkt_preempted_proc()) != NULL) {
dcc99b62
MD
778 /*
779 * Account for cpu time used and hit the scheduler. Note
780 * that this call MUST BE MP SAFE, and the BGL IS NOT HELD
781 * HERE.
782 */
553ea3c8 783 ++lp->lwp_cpticks;
5681a38a
MD
784 lp->lwp_proc->p_usched->schedulerclock(lp, info->periodic,
785 info->time);
dcc99b62 786 }
553ea3c8 787 if ((lp = curthread->td_lwp) != NULL) {
dcc99b62
MD
788 /*
789 * Update resource usage integrals and maximums.
790 */
fde7ac71 791 if ((ru = &lp->lwp_proc->p_ru) &&
553ea3c8 792 (vm = lp->lwp_proc->p_vmspace) != NULL) {
88c4d2f6
MD
793 ru->ru_ixrss += pgtok(vm->vm_tsize);
794 ru->ru_idrss += pgtok(vm->vm_dsize);
795 ru->ru_isrss += pgtok(vm->vm_ssize);
b12defdc
MD
796 if (lwkt_trytoken(&vm->vm_map.token)) {
797 rss = pgtok(vmspace_resident_count(vm));
798 if (ru->ru_maxrss < rss)
799 ru->ru_maxrss = rss;
800 lwkt_reltoken(&vm->vm_map.token);
801 }
88c4d2f6 802 }
b68b7282 803 }
d6d39bc7
MC
804 /* Increment the global sched_ticks */
805 if (mycpu->gd_cpuid == 0)
806 ++sched_ticks;
984263bc
MD
807}
808
809/*
a94976ad
MD
810 * Compute number of ticks for the specified amount of time. The
811 * return value is intended to be used in a clock interrupt timed
812 * operation and guarenteed to meet or exceed the requested time.
813 * If the representation overflows, return INT_MAX. The minimum return
814 * value is 1 ticks and the function will average the calculation up.
815 * If any value greater then 0 microseconds is supplied, a value
816 * of at least 2 will be returned to ensure that a near-term clock
817 * interrupt does not cause the timeout to occur (degenerately) early.
818 *
819 * Note that limit checks must take into account microseconds, which is
820 * done simply by using the smaller signed long maximum instead of
821 * the unsigned long maximum.
822 *
823 * If ints have 32 bits, then the maximum value for any timeout in
824 * 10ms ticks is 248 days.
984263bc
MD
825 */
826int
a94976ad 827tvtohz_high(struct timeval *tv)
984263bc 828{
a94976ad 829 int ticks;
1fd87d54 830 long sec, usec;
984263bc 831
984263bc
MD
832 sec = tv->tv_sec;
833 usec = tv->tv_usec;
834 if (usec < 0) {
835 sec--;
836 usec += 1000000;
837 }
838 if (sec < 0) {
839#ifdef DIAGNOSTIC
840 if (usec > 0) {
841 sec++;
842 usec -= 1000000;
843 }
a591f597
MD
844 kprintf("tvtohz_high: negative time difference "
845 "%ld sec %ld usec\n",
846 sec, usec);
984263bc
MD
847#endif
848 ticks = 1;
a94976ad
MD
849 } else if (sec <= INT_MAX / hz) {
850 ticks = (int)(sec * hz +
a591f597 851 ((u_long)usec + (ustick - 1)) / ustick) + 1;
a94976ad
MD
852 } else {
853 ticks = INT_MAX;
854 }
855 return (ticks);
856}
857
a591f597
MD
858int
859tstohz_high(struct timespec *ts)
860{
861 int ticks;
862 long sec, nsec;
863
864 sec = ts->tv_sec;
865 nsec = ts->tv_nsec;
866 if (nsec < 0) {
867 sec--;
868 nsec += 1000000000;
869 }
870 if (sec < 0) {
871#ifdef DIAGNOSTIC
872 if (nsec > 0) {
873 sec++;
874 nsec -= 1000000000;
875 }
876 kprintf("tstohz_high: negative time difference "
877 "%ld sec %ld nsec\n",
878 sec, nsec);
879#endif
880 ticks = 1;
881 } else if (sec <= INT_MAX / hz) {
882 ticks = (int)(sec * hz +
883 ((u_long)nsec + (nstick - 1)) / nstick) + 1;
884 } else {
885 ticks = INT_MAX;
886 }
887 return (ticks);
888}
889
890
a94976ad
MD
891/*
892 * Compute number of ticks for the specified amount of time, erroring on
893 * the side of it being too low to ensure that sleeping the returned number
894 * of ticks will not result in a late return.
895 *
896 * The supplied timeval may not be negative and should be normalized. A
897 * return value of 0 is possible if the timeval converts to less then
898 * 1 tick.
899 *
900 * If ints have 32 bits, then the maximum value for any timeout in
901 * 10ms ticks is 248 days.
902 */
903int
904tvtohz_low(struct timeval *tv)
905{
906 int ticks;
907 long sec;
908
909 sec = tv->tv_sec;
910 if (sec <= INT_MAX / hz)
a591f597 911 ticks = (int)(sec * hz + (u_long)tv->tv_usec / ustick);
984263bc 912 else
984263bc 913 ticks = INT_MAX;
a94976ad 914 return (ticks);
984263bc
MD
915}
916
a591f597
MD
917int
918tstohz_low(struct timespec *ts)
919{
920 int ticks;
921 long sec;
922
923 sec = ts->tv_sec;
924 if (sec <= INT_MAX / hz)
925 ticks = (int)(sec * hz + (u_long)ts->tv_nsec / nstick);
926 else
927 ticks = INT_MAX;
928 return (ticks);
929}
a94976ad 930
984263bc
MD
931/*
932 * Start profiling on a process.
933 *
934 * Kernel profiling passes proc0 which never exits and hence
935 * keeps the profile clock running constantly.
936 */
937void
88c4d2f6 938startprofclock(struct proc *p)
984263bc 939{
4643740a
MD
940 if ((p->p_flags & P_PROFIL) == 0) {
941 p->p_flags |= P_PROFIL;
88c4d2f6 942#if 0 /* XXX */
984263bc 943 if (++profprocs == 1 && stathz != 0) {
e43a034f 944 crit_enter();
6ad39cae 945 psdiv = psratio;
984263bc 946 setstatclockrate(profhz);
e43a034f 947 crit_exit();
984263bc 948 }
88c4d2f6 949#endif
984263bc
MD
950 }
951}
952
953/*
954 * Stop profiling on a process.
616516c8
MD
955 *
956 * caller must hold p->p_token
984263bc
MD
957 */
958void
88c4d2f6 959stopprofclock(struct proc *p)
984263bc 960{
4643740a
MD
961 if (p->p_flags & P_PROFIL) {
962 p->p_flags &= ~P_PROFIL;
88c4d2f6 963#if 0 /* XXX */
984263bc 964 if (--profprocs == 0 && stathz != 0) {
e43a034f 965 crit_enter();
6ad39cae 966 psdiv = 1;
984263bc 967 setstatclockrate(stathz);
e43a034f 968 crit_exit();
984263bc 969 }
984263bc 970#endif
984263bc
MD
971 }
972}
973
974/*
975 * Return information about system clocks.
976 */
977static int
978sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS)
979{
f5d21610 980 struct kinfo_clockinfo clkinfo;
984263bc
MD
981 /*
982 * Construct clockinfo structure.
983 */
f5d21610 984 clkinfo.ci_hz = hz;
a591f597 985 clkinfo.ci_tick = ustick;
4026c000 986 clkinfo.ci_tickadj = ntp_default_tick_delta / 1000;
f5d21610
JS
987 clkinfo.ci_profhz = profhz;
988 clkinfo.ci_stathz = stathz ? stathz : hz;
984263bc
MD
989 return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req));
990}
991
992SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate, CTLTYPE_STRUCT|CTLFLAG_RD,
993 0, 0, sysctl_kern_clockrate, "S,clockinfo","");
994
984263bc
MD
995/*
996 * We have eight functions for looking at the clock, four for
997 * microseconds and four for nanoseconds. For each there is fast
998 * but less precise version "get{nano|micro}[up]time" which will
999 * return a time which is up to 1/HZ previous to the call, whereas
1000 * the raw version "{nano|micro}[up]time" will return a timestamp
1001 * which is as precise as possible. The "up" variants return the
1002 * time relative to system boot, these are well suited for time
1003 * interval measurements.
88c4d2f6
MD
1004 *
1005 * Each cpu independantly maintains the current time of day, so all
1006 * we need to do to protect ourselves from changes is to do a loop
1007 * check on the seconds field changing out from under us.
fad57d0e
MD
1008 *
1009 * The system timer maintains a 32 bit count and due to various issues
1010 * it is possible for the calculated delta to occassionally exceed
044ee7c4
MD
1011 * sys_cputimer->freq. If this occurs the sys_cputimer->freq64_nsec
1012 * multiplication can easily overflow, so we deal with the case. For
1013 * uniformity we deal with the case in the usec case too.
627531fa
MD
1014 *
1015 * All the [get][micro,nano][time,uptime]() routines are MPSAFE.
984263bc 1016 */
984263bc
MD
1017void
1018getmicrouptime(struct timeval *tvp)
1019{
88c4d2f6
MD
1020 struct globaldata *gd = mycpu;
1021 sysclock_t delta;
1022
1023 do {
1024 tvp->tv_sec = gd->gd_time_seconds;
1025 delta = gd->gd_hardclock.time - gd->gd_cpuclock_base;
1026 } while (tvp->tv_sec != gd->gd_time_seconds);
fad57d0e 1027
044ee7c4
MD
1028 if (delta >= sys_cputimer->freq) {
1029 tvp->tv_sec += delta / sys_cputimer->freq;
1030 delta %= sys_cputimer->freq;
fad57d0e 1031 }
044ee7c4 1032 tvp->tv_usec = (sys_cputimer->freq64_usec * delta) >> 32;
88c4d2f6
MD
1033 if (tvp->tv_usec >= 1000000) {
1034 tvp->tv_usec -= 1000000;
1035 ++tvp->tv_sec;
984263bc
MD
1036 }
1037}
1038
1039void
1040getnanouptime(struct timespec *tsp)
1041{
88c4d2f6
MD
1042 struct globaldata *gd = mycpu;
1043 sysclock_t delta;
1044
1045 do {
1046 tsp->tv_sec = gd->gd_time_seconds;
1047 delta = gd->gd_hardclock.time - gd->gd_cpuclock_base;
1048 } while (tsp->tv_sec != gd->gd_time_seconds);
fad57d0e 1049
044ee7c4
MD
1050 if (delta >= sys_cputimer->freq) {
1051 tsp->tv_sec += delta / sys_cputimer->freq;
1052 delta %= sys_cputimer->freq;
984263bc 1053 }
044ee7c4 1054 tsp->tv_nsec = (sys_cputimer->freq64_nsec * delta) >> 32;
984263bc
MD
1055}
1056
1057void
88c4d2f6 1058microuptime(struct timeval *tvp)
984263bc 1059{
88c4d2f6
MD
1060 struct globaldata *gd = mycpu;
1061 sysclock_t delta;
1062
1063 do {
1064 tvp->tv_sec = gd->gd_time_seconds;
044ee7c4 1065 delta = sys_cputimer->count() - gd->gd_cpuclock_base;
88c4d2f6 1066 } while (tvp->tv_sec != gd->gd_time_seconds);
fad57d0e 1067
044ee7c4
MD
1068 if (delta >= sys_cputimer->freq) {
1069 tvp->tv_sec += delta / sys_cputimer->freq;
1070 delta %= sys_cputimer->freq;
984263bc 1071 }
044ee7c4 1072 tvp->tv_usec = (sys_cputimer->freq64_usec * delta) >> 32;
984263bc
MD
1073}
1074
1075void
88c4d2f6 1076nanouptime(struct timespec *tsp)
984263bc 1077{
88c4d2f6
MD
1078 struct globaldata *gd = mycpu;
1079 sysclock_t delta;
1080
1081 do {
1082 tsp->tv_sec = gd->gd_time_seconds;
044ee7c4 1083 delta = sys_cputimer->count() - gd->gd_cpuclock_base;
88c4d2f6 1084 } while (tsp->tv_sec != gd->gd_time_seconds);
fad57d0e 1085
044ee7c4
MD
1086 if (delta >= sys_cputimer->freq) {
1087 tsp->tv_sec += delta / sys_cputimer->freq;
1088 delta %= sys_cputimer->freq;
984263bc 1089 }
044ee7c4 1090 tsp->tv_nsec = (sys_cputimer->freq64_nsec * delta) >> 32;
984263bc
MD
1091}
1092
88c4d2f6
MD
1093/*
1094 * realtime routines
1095 */
984263bc 1096void
88c4d2f6 1097getmicrotime(struct timeval *tvp)
984263bc 1098{
88c4d2f6 1099 struct globaldata *gd = mycpu;
5eb5a6bc 1100 struct timespec *bt;
88c4d2f6 1101 sysclock_t delta;
984263bc 1102
88c4d2f6
MD
1103 do {
1104 tvp->tv_sec = gd->gd_time_seconds;
1105 delta = gd->gd_hardclock.time - gd->gd_cpuclock_base;
1106 } while (tvp->tv_sec != gd->gd_time_seconds);
fad57d0e 1107
044ee7c4
MD
1108 if (delta >= sys_cputimer->freq) {
1109 tvp->tv_sec += delta / sys_cputimer->freq;
1110 delta %= sys_cputimer->freq;
fad57d0e 1111 }
044ee7c4 1112 tvp->tv_usec = (sys_cputimer->freq64_usec * delta) >> 32;
984263bc 1113
5eb5a6bc
MD
1114 bt = &basetime[basetime_index];
1115 tvp->tv_sec += bt->tv_sec;
1116 tvp->tv_usec += bt->tv_nsec / 1000;
88c4d2f6
MD
1117 while (tvp->tv_usec >= 1000000) {
1118 tvp->tv_usec -= 1000000;
1119 ++tvp->tv_sec;
984263bc 1120 }
984263bc
MD
1121}
1122
1123void
88c4d2f6 1124getnanotime(struct timespec *tsp)
984263bc 1125{
88c4d2f6 1126 struct globaldata *gd = mycpu;
5eb5a6bc 1127 struct timespec *bt;
88c4d2f6 1128 sysclock_t delta;
984263bc 1129
88c4d2f6
MD
1130 do {
1131 tsp->tv_sec = gd->gd_time_seconds;
1132 delta = gd->gd_hardclock.time - gd->gd_cpuclock_base;
1133 } while (tsp->tv_sec != gd->gd_time_seconds);
fad57d0e 1134
044ee7c4
MD
1135 if (delta >= sys_cputimer->freq) {
1136 tsp->tv_sec += delta / sys_cputimer->freq;
1137 delta %= sys_cputimer->freq;
fad57d0e 1138 }
044ee7c4 1139 tsp->tv_nsec = (sys_cputimer->freq64_nsec * delta) >> 32;
984263bc 1140
5eb5a6bc
MD
1141 bt = &basetime[basetime_index];
1142 tsp->tv_sec += bt->tv_sec;
1143 tsp->tv_nsec += bt->tv_nsec;
88c4d2f6
MD
1144 while (tsp->tv_nsec >= 1000000000) {
1145 tsp->tv_nsec -= 1000000000;
1146 ++tsp->tv_sec;
984263bc 1147 }
984263bc
MD
1148}
1149
5eb5a6bc
MD
1150static void
1151getnanotime_nbt(struct timespec *nbt, struct timespec *tsp)
1152{
1153 struct globaldata *gd = mycpu;
1154 sysclock_t delta;
1155
1156 do {
1157 tsp->tv_sec = gd->gd_time_seconds;
1158 delta = gd->gd_hardclock.time - gd->gd_cpuclock_base;
1159 } while (tsp->tv_sec != gd->gd_time_seconds);
1160
044ee7c4
MD
1161 if (delta >= sys_cputimer->freq) {
1162 tsp->tv_sec += delta / sys_cputimer->freq;
1163 delta %= sys_cputimer->freq;
5eb5a6bc 1164 }
044ee7c4 1165 tsp->tv_nsec = (sys_cputimer->freq64_nsec * delta) >> 32;
5eb5a6bc
MD
1166
1167 tsp->tv_sec += nbt->tv_sec;
1168 tsp->tv_nsec += nbt->tv_nsec;
1169 while (tsp->tv_nsec >= 1000000000) {
1170 tsp->tv_nsec -= 1000000000;
1171 ++tsp->tv_sec;
1172 }
1173}
1174
1175
88c4d2f6
MD
1176void
1177microtime(struct timeval *tvp)
984263bc 1178{
88c4d2f6 1179 struct globaldata *gd = mycpu;
5eb5a6bc 1180 struct timespec *bt;
88c4d2f6 1181 sysclock_t delta;
984263bc 1182
88c4d2f6
MD
1183 do {
1184 tvp->tv_sec = gd->gd_time_seconds;
044ee7c4 1185 delta = sys_cputimer->count() - gd->gd_cpuclock_base;
88c4d2f6 1186 } while (tvp->tv_sec != gd->gd_time_seconds);
fad57d0e 1187
044ee7c4
MD
1188 if (delta >= sys_cputimer->freq) {
1189 tvp->tv_sec += delta / sys_cputimer->freq;
1190 delta %= sys_cputimer->freq;
fad57d0e 1191 }
044ee7c4 1192 tvp->tv_usec = (sys_cputimer->freq64_usec * delta) >> 32;
984263bc 1193
5eb5a6bc
MD
1194 bt = &basetime[basetime_index];
1195 tvp->tv_sec += bt->tv_sec;
1196 tvp->tv_usec += bt->tv_nsec / 1000;
88c4d2f6
MD
1197 while (tvp->tv_usec >= 1000000) {
1198 tvp->tv_usec -= 1000000;
1199 ++tvp->tv_sec;
984263bc 1200 }
984263bc
MD
1201}
1202
88c4d2f6
MD
1203void
1204nanotime(struct timespec *tsp)
1205{
1206 struct globaldata *gd = mycpu;
5eb5a6bc 1207 struct timespec *bt;
88c4d2f6 1208 sysclock_t delta;
984263bc 1209
88c4d2f6
MD
1210 do {
1211 tsp->tv_sec = gd->gd_time_seconds;
044ee7c4 1212 delta = sys_cputimer->count() - gd->gd_cpuclock_base;
88c4d2f6 1213 } while (tsp->tv_sec != gd->gd_time_seconds);
fad57d0e 1214
044ee7c4
MD
1215 if (delta >= sys_cputimer->freq) {
1216 tsp->tv_sec += delta / sys_cputimer->freq;
1217 delta %= sys_cputimer->freq;
fad57d0e 1218 }
044ee7c4 1219 tsp->tv_nsec = (sys_cputimer->freq64_nsec * delta) >> 32;
984263bc 1220
5eb5a6bc
MD
1221 bt = &basetime[basetime_index];
1222 tsp->tv_sec += bt->tv_sec;
1223 tsp->tv_nsec += bt->tv_nsec;
88c4d2f6
MD
1224 while (tsp->tv_nsec >= 1000000000) {
1225 tsp->tv_nsec -= 1000000000;
1226 ++tsp->tv_sec;
984263bc 1227 }
984263bc
MD
1228}
1229
25b804e7
MD
1230/*
1231 * note: this is not exactly synchronized with real time. To do that we
1232 * would have to do what microtime does and check for a nanoseconds overflow.
1233 */
1234time_t
1235get_approximate_time_t(void)
1236{
1237 struct globaldata *gd = mycpu;
5eb5a6bc
MD
1238 struct timespec *bt;
1239
1240 bt = &basetime[basetime_index];
1241 return(gd->gd_time_seconds + bt->tv_sec);
25b804e7
MD
1242}
1243
984263bc
MD
1244int
1245pps_ioctl(u_long cmd, caddr_t data, struct pps_state *pps)
1246{
1247 pps_params_t *app;
1248 struct pps_fetch_args *fapi;
1249#ifdef PPS_SYNC
1250 struct pps_kcbind_args *kapi;
1251#endif
1252
1253 switch (cmd) {
1254 case PPS_IOC_CREATE:
1255 return (0);
1256 case PPS_IOC_DESTROY:
1257 return (0);
1258 case PPS_IOC_SETPARAMS:
1259 app = (pps_params_t *)data;
1260 if (app->mode & ~pps->ppscap)
1261 return (EINVAL);
1262 pps->ppsparam = *app;
1263 return (0);
1264 case PPS_IOC_GETPARAMS:
1265 app = (pps_params_t *)data;
1266 *app = pps->ppsparam;
1267 app->api_version = PPS_API_VERS_1;
1268 return (0);
1269 case PPS_IOC_GETCAP:
1270 *(int*)data = pps->ppscap;
1271 return (0);
1272 case PPS_IOC_FETCH:
1273 fapi = (struct pps_fetch_args *)data;
1274 if (fapi->tsformat && fapi->tsformat != PPS_TSFMT_TSPEC)
1275 return (EINVAL);
1276 if (fapi->timeout.tv_sec || fapi->timeout.tv_nsec)
1277 return (EOPNOTSUPP);
1278 pps->ppsinfo.current_mode = pps->ppsparam.mode;
1279 fapi->pps_info_buf = pps->ppsinfo;
1280 return (0);
1281 case PPS_IOC_KCBIND:
1282#ifdef PPS_SYNC
1283 kapi = (struct pps_kcbind_args *)data;
1284 /* XXX Only root should be able to do this */
1285 if (kapi->tsformat && kapi->tsformat != PPS_TSFMT_TSPEC)
1286 return (EINVAL);
1287 if (kapi->kernel_consumer != PPS_KC_HARDPPS)
1288 return (EINVAL);
1289 if (kapi->edge & ~pps->ppscap)
1290 return (EINVAL);
1291 pps->kcmode = kapi->edge;
1292 return (0);
1293#else
1294 return (EOPNOTSUPP);
1295#endif
1296 default:
1297 return (ENOTTY);
1298 }
1299}
1300
1301void
1302pps_init(struct pps_state *pps)
1303{
1304 pps->ppscap |= PPS_TSFMT_TSPEC;
1305 if (pps->ppscap & PPS_CAPTUREASSERT)
1306 pps->ppscap |= PPS_OFFSETASSERT;
1307 if (pps->ppscap & PPS_CAPTURECLEAR)
1308 pps->ppscap |= PPS_OFFSETCLEAR;
1309}
1310
1311void
88c4d2f6 1312pps_event(struct pps_state *pps, sysclock_t count, int event)
984263bc 1313{
88c4d2f6
MD
1314 struct globaldata *gd;
1315 struct timespec *tsp;
1316 struct timespec *osp;
5eb5a6bc 1317 struct timespec *bt;
88c4d2f6
MD
1318 struct timespec ts;
1319 sysclock_t *pcount;
1320#ifdef PPS_SYNC
1321 sysclock_t tcount;
1322#endif
1323 sysclock_t delta;
1324 pps_seq_t *pseq;
1325 int foff;
1326 int fhard;
1327
1328 gd = mycpu;
984263bc
MD
1329
1330 /* Things would be easier with arrays... */
1331 if (event == PPS_CAPTUREASSERT) {
1332 tsp = &pps->ppsinfo.assert_timestamp;
1333 osp = &pps->ppsparam.assert_offset;
1334 foff = pps->ppsparam.mode & PPS_OFFSETASSERT;
1335 fhard = pps->kcmode & PPS_CAPTUREASSERT;
1336 pcount = &pps->ppscount[0];
1337 pseq = &pps->ppsinfo.assert_sequence;
1338 } else {
1339 tsp = &pps->ppsinfo.clear_timestamp;
1340 osp = &pps->ppsparam.clear_offset;
1341 foff = pps->ppsparam.mode & PPS_OFFSETCLEAR;
1342 fhard = pps->kcmode & PPS_CAPTURECLEAR;
1343 pcount = &pps->ppscount[1];
1344 pseq = &pps->ppsinfo.clear_sequence;
1345 }
1346
984263bc
MD
1347 /* Nothing really happened */
1348 if (*pcount == count)
1349 return;
1350
1351 *pcount = count;
1352
88c4d2f6
MD
1353 do {
1354 ts.tv_sec = gd->gd_time_seconds;
1355 delta = count - gd->gd_cpuclock_base;
1356 } while (ts.tv_sec != gd->gd_time_seconds);
fad57d0e 1357
044ee7c4
MD
1358 if (delta >= sys_cputimer->freq) {
1359 ts.tv_sec += delta / sys_cputimer->freq;
1360 delta %= sys_cputimer->freq;
88c4d2f6 1361 }
044ee7c4 1362 ts.tv_nsec = (sys_cputimer->freq64_nsec * delta) >> 32;
5eb5a6bc
MD
1363 bt = &basetime[basetime_index];
1364 ts.tv_sec += bt->tv_sec;
1365 ts.tv_nsec += bt->tv_nsec;
88c4d2f6
MD
1366 while (ts.tv_nsec >= 1000000000) {
1367 ts.tv_nsec -= 1000000000;
1368 ++ts.tv_sec;
984263bc 1369 }
984263bc
MD
1370
1371 (*pseq)++;
1372 *tsp = ts;
1373
1374 if (foff) {
1375 timespecadd(tsp, osp);
1376 if (tsp->tv_nsec < 0) {
1377 tsp->tv_nsec += 1000000000;
1378 tsp->tv_sec -= 1;
1379 }
1380 }
1381#ifdef PPS_SYNC
1382 if (fhard) {
1383 /* magic, at its best... */
1384 tcount = count - pps->ppscount[2];
1385 pps->ppscount[2] = count;
044ee7c4
MD
1386 if (tcount >= sys_cputimer->freq) {
1387 delta = (1000000000 * (tcount / sys_cputimer->freq) +
1388 sys_cputimer->freq64_nsec *
1389 (tcount % sys_cputimer->freq)) >> 32;
fad57d0e 1390 } else {
044ee7c4 1391 delta = (sys_cputimer->freq64_nsec * tcount) >> 32;
fad57d0e 1392 }
984263bc
MD
1393 hardpps(tsp, delta);
1394 }
1395#endif
1396}
88c4d2f6 1397
d2412a2e
MD
1398/*
1399 * Return the tsc target value for a delay of (ns).
1400 *
1401 * Returns -1 if the TSC is not supported.
1402 */
1403int64_t
1404tsc_get_target(int ns)
1405{
1406#if defined(_RDTSC_SUPPORTED_)
1407 if (cpu_feature & CPUID_TSC) {
1408 return (rdtsc() + tsc_frequency * ns / (int64_t)1000000000);
1409 }
1410#endif
1411 return(-1);
1412}
1413
1414/*
1415 * Compare the tsc against the passed target
1416 *
1417 * Returns +1 if the target has been reached
1418 * Returns 0 if the target has not yet been reached
1419 * Returns -1 if the TSC is not supported.
1420 *
1421 * Typical use: while (tsc_test_target(target) == 0) { ...poll... }
1422 */
1423int
1424tsc_test_target(int64_t target)
1425{
1426#if defined(_RDTSC_SUPPORTED_)
1427 if (cpu_feature & CPUID_TSC) {
1428 if ((int64_t)(target - rdtsc()) <= 0)
1429 return(1);
1430 return(0);
1431 }
d2412a2e 1432#endif
2e537993 1433 return(-1);
d2412a2e 1434}
b12defdc
MD
1435
1436/*
1437 * Delay the specified number of nanoseconds using the tsc. This function
1438 * returns immediately if the TSC is not supported. At least one cpu_pause()
1439 * will be issued.
1440 */
1441void
1442tsc_delay(int ns)
1443{
1444 int64_t clk;
1445
1446 clk = tsc_get_target(ns);
1447 cpu_pause();
1448 while (tsc_test_target(clk) == 0)
1449 cpu_pause();
1450}