Change the FreeBSD 5 jail sysctls to the correct DragonFly locations.
[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 $
0284027e 73 * $DragonFly: src/sys/kern/kern_clock.c,v 1.23 2004/08/02 23:20:30 dillon Exp $
984263bc
MD
74 */
75
76#include "opt_ntp.h"
77
78#include <sys/param.h>
79#include <sys/systm.h>
80#include <sys/dkstat.h>
81#include <sys/callout.h>
82#include <sys/kernel.h>
83#include <sys/proc.h>
84#include <sys/malloc.h>
85#include <sys/resourcevar.h>
86#include <sys/signalvar.h>
87#include <sys/timex.h>
88#include <sys/timepps.h>
89#include <vm/vm.h>
90#include <sys/lock.h>
91#include <vm/pmap.h>
92#include <vm/vm_map.h>
93#include <sys/sysctl.h>
2689779e 94#include <sys/thread2.h>
984263bc
MD
95
96#include <machine/cpu.h>
97#include <machine/limits.h>
98#include <machine/smp.h>
99
100#ifdef GPROF
101#include <sys/gmon.h>
102#endif
103
104#ifdef DEVICE_POLLING
105extern void init_device_poll(void);
106extern void hardclock_device_poll(void);
107#endif /* DEVICE_POLLING */
108
402ed7e1 109static void initclocks (void *dummy);
984263bc
MD
110SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL)
111
6ad39cae
MD
112/*
113 * Some of these don't belong here, but it's easiest to concentrate them.
114 * Note that cp_time[] counts in microseconds, but most userland programs
115 * just compare relative times against the total by delta.
116 */
984263bc
MD
117long cp_time[CPUSTATES];
118
119SYSCTL_OPAQUE(_kern, OID_AUTO, cp_time, CTLFLAG_RD, &cp_time, sizeof(cp_time),
120 "LU", "CPU time statistics");
121
122long tk_cancc;
123long tk_nin;
124long tk_nout;
125long tk_rawcc;
126
88c4d2f6
MD
127/*
128 * boottime is used to calculate the 'real' uptime. Do not confuse this with
129 * microuptime(). microtime() is not drift compensated. The real uptime
60b2809b
MD
130 * with compensation is nanotime() - bootime. boottime is recalculated
131 * whenever the real time is set based on the compensated elapsed time
132 * in seconds (gd->gd_time_seconds).
88c4d2f6
MD
133 *
134 * basetime is used to calculate the compensated real time of day. Chunky
135 * changes to the time, aka settimeofday(), are made by modifying basetime.
136 *
137 * The gd_time_seconds and gd_cpuclock_base fields remain fairly monotonic.
138 * Slight adjustments to gd_cpuclock_base are made to phase-lock it to
139 * the real time.
140 */
141struct timespec boottime; /* boot time (realtime) for reference only */
142struct timespec basetime; /* base time adjusts uptime -> realtime */
143time_t time_second; /* read-only 'passive' uptime in seconds */
984263bc 144
984263bc
MD
145SYSCTL_STRUCT(_kern, KERN_BOOTTIME, boottime, CTLFLAG_RD,
146 &boottime, timeval, "System boottime");
88c4d2f6
MD
147SYSCTL_STRUCT(_kern, OID_AUTO, basetime, CTLFLAG_RD,
148 &basetime, timeval, "System basetime");
984263bc 149
88c4d2f6
MD
150static void hardclock(systimer_t info, struct intrframe *frame);
151static void statclock(systimer_t info, struct intrframe *frame);
152static void schedclock(systimer_t info, struct intrframe *frame);
153
154int ticks; /* system master ticks at hz */
da3639ef 155int clocks_running; /* tsleep/timeout clocks operational */
88c4d2f6
MD
156int64_t nsec_adj; /* ntpd per-tick adjustment in nsec << 32 */
157int64_t nsec_acc; /* accumulator */
984263bc
MD
158
159/*
88c4d2f6 160 * Finish initializing clock frequencies and start all clocks running.
984263bc 161 */
88c4d2f6
MD
162/* ARGSUSED*/
163static void
164initclocks(void *dummy)
984263bc 165{
88c4d2f6
MD
166 cpu_initclocks();
167#ifdef DEVICE_POLLING
168 init_device_poll();
169#endif
170 /*psratio = profhz / stathz;*/
171 initclocks_pcpu();
da3639ef 172 clocks_running = 1;
984263bc
MD
173}
174
88c4d2f6
MD
175/*
176 * Called on a per-cpu basis
177 */
178void
179initclocks_pcpu(void)
180{
181 struct globaldata *gd = mycpu;
984263bc 182
88c4d2f6
MD
183 crit_enter();
184 if (gd->gd_cpuid == 0) {
185 gd->gd_time_seconds = 1;
186 gd->gd_cpuclock_base = cputimer_count();
187 } else {
188 /* XXX */
189 gd->gd_time_seconds = globaldata_find(0)->gd_time_seconds;
190 gd->gd_cpuclock_base = globaldata_find(0)->gd_cpuclock_base;
191 }
192 systimer_init_periodic(&gd->gd_hardclock, hardclock, NULL, hz);
193 systimer_init_periodic(&gd->gd_statclock, statclock, NULL, stathz);
194 /* XXX correct the frequency for scheduler / estcpu tests */
8478264a
MD
195 systimer_init_periodic(&gd->gd_schedclock, schedclock,
196 NULL, ESTCPUFREQ);
88c4d2f6
MD
197 crit_exit();
198}
984263bc 199
0284027e
MD
200/*
201 * Resynchronize gd_cpuclock_base after the system has been woken up from
202 * a sleep. It is absolutely essential that all the cpus be properly
203 * synchronized. Resynching is required because nanouptime() and friends
204 * will overflow intermediate multiplications if more then 2 seconds
205 * worth of cputimer_cont() delta has built up.
206 */
207#ifdef SMP
208
209static
210void
211restoreclocks_remote(lwkt_cpusync_t poll)
212{
213 mycpu->gd_cpuclock_base = *(sysclock_t *)poll->cs_data;
214 mycpu->gd_time_seconds = globaldata_find(0)->gd_time_seconds;
215}
216
217#endif
218
219void
220restoreclocks(void)
221{
222 sysclock_t base = cputimer_count();
223#ifdef SMP
224 lwkt_cpusync_simple(-1, restoreclocks_remote, &base);
225#else
226 mycpu->gd_cpuclock_base = base;
227#endif
228}
229
984263bc 230/*
88c4d2f6
MD
231 * This sets the current real time of day. Timespecs are in seconds and
232 * nanoseconds. We do not mess with gd_time_seconds and gd_cpuclock_base,
233 * instead we adjust basetime so basetime + gd_* results in the current
234 * time of day. This way the gd_* fields are guarenteed to represent
235 * a monotonically increasing 'uptime' value.
984263bc 236 */
88c4d2f6
MD
237void
238set_timeofday(struct timespec *ts)
239{
240 struct timespec ts2;
984263bc 241
88c4d2f6
MD
242 /*
243 * XXX SMP / non-atomic basetime updates
244 */
245 crit_enter();
246 nanouptime(&ts2);
247 basetime.tv_sec = ts->tv_sec - ts2.tv_sec;
248 basetime.tv_nsec = ts->tv_nsec - ts2.tv_nsec;
249 if (basetime.tv_nsec < 0) {
250 basetime.tv_nsec += 1000000000;
251 --basetime.tv_sec;
252 }
60b2809b 253 boottime.tv_sec = basetime.tv_sec - mycpu->gd_time_seconds;
88c4d2f6
MD
254 timedelta = 0;
255 crit_exit();
256}
257
984263bc 258/*
88c4d2f6
MD
259 * Each cpu has its own hardclock, but we only increments ticks and softticks
260 * on cpu #0.
261 *
262 * NOTE! systimer! the MP lock might not be held here. We can only safely
263 * manipulate objects owned by the current cpu.
984263bc 264 */
984263bc 265static void
88c4d2f6 266hardclock(systimer_t info, struct intrframe *frame)
984263bc 267{
88c4d2f6
MD
268 sysclock_t cputicks;
269 struct proc *p;
270 struct pstats *pstats;
271 struct globaldata *gd = mycpu;
984263bc
MD
272
273 /*
88c4d2f6
MD
274 * Realtime updates are per-cpu. Note that timer corrections as
275 * returned by microtime() and friends make an additional adjustment
276 * using a system-wise 'basetime', but the running time is always
277 * taken from the per-cpu globaldata area. Since the same clock
278 * is distributing (XXX SMP) to all cpus, the per-cpu timebases
279 * stay in synch.
280 *
281 * Note that we never allow info->time (aka gd->gd_hardclock.time)
282 * to reverse index gd_cpuclock_base.
984263bc 283 */
88c4d2f6
MD
284 cputicks = info->time - gd->gd_cpuclock_base;
285 if (cputicks > cputimer_freq) {
286 ++gd->gd_time_seconds;
287 gd->gd_cpuclock_base += cputimer_freq;
288 }
984263bc
MD
289
290 /*
88c4d2f6
MD
291 * The system-wide ticks and softticks are only updated by cpu #0.
292 * Callwheel actions are also (at the moment) only handled by cpu #0.
293 * Finally, we also do NTP related timedelta/tickdelta adjustments
294 * by adjusting basetime.
984263bc 295 */
88c4d2f6
MD
296 if (gd->gd_cpuid == 0) {
297 struct timespec nts;
298 int leap;
984263bc 299
88c4d2f6 300 ++ticks;
984263bc 301
88c4d2f6
MD
302#ifdef DEVICE_POLLING
303 hardclock_device_poll(); /* mpsafe, short and quick */
304#endif /* DEVICE_POLLING */
984263bc 305
88c4d2f6
MD
306 if (TAILQ_FIRST(&callwheel[ticks & callwheelmask]) != NULL) {
307 setsoftclock();
308 } else if (softticks + 1 == ticks) {
309 ++softticks;
310 }
311
312#if 0
313 if (tco->tc_poll_pps)
314 tco->tc_poll_pps(tco);
315#endif
316 /*
317 * Apply adjtime corrections. At the moment only do this if
318 * we can get the MP lock to interlock with adjtime's modification
319 * of these variables. Note that basetime adjustments are not
320 * MP safe either XXX.
321 */
322 if (timedelta != 0 && try_mplock()) {
323 basetime.tv_nsec += tickdelta * 1000;
324 if (basetime.tv_nsec >= 1000000000) {
325 basetime.tv_nsec -= 1000000000;
326 ++basetime.tv_sec;
327 } else if (basetime.tv_nsec < 0) {
328 basetime.tv_nsec += 1000000000;
329 --basetime.tv_sec;
330 }
331 timedelta -= tickdelta;
332 rel_mplock();
333 }
334
335 /*
336 * Apply per-tick compensation. ticks_adj adjusts for both
337 * offset and frequency, and could be negative.
338 */
339 if (nsec_adj != 0 && try_mplock()) {
340 nsec_acc += nsec_adj;
341 if (nsec_acc >= 0x100000000LL) {
342 basetime.tv_nsec += nsec_acc >> 32;
343 nsec_acc = (nsec_acc & 0xFFFFFFFFLL);
344 } else if (nsec_acc <= -0x100000000LL) {
345 basetime.tv_nsec -= -nsec_acc >> 32;
346 nsec_acc = -(-nsec_acc & 0xFFFFFFFFLL);
347 }
348 if (basetime.tv_nsec >= 1000000000) {
349 basetime.tv_nsec -= 1000000000;
350 ++basetime.tv_sec;
351 } else if (basetime.tv_nsec < 0) {
352 basetime.tv_nsec += 1000000000;
353 --basetime.tv_sec;
354 }
355 rel_mplock();
356 }
357
358 /*
359 * If the realtime-adjusted seconds hand rolls over then tell
360 * ntp_update_second() what we did in the last second so it can
361 * calculate what to do in the next second. It may also add
362 * or subtract a leap second.
363 */
364 getnanotime(&nts);
365 if (time_second != nts.tv_sec) {
366 leap = ntp_update_second(time_second, &nsec_adj);
367 basetime.tv_sec += leap;
368 time_second = nts.tv_sec + leap;
369 nsec_adj /= hz;
370 }
371 }
372
373 /*
374 * ITimer handling is per-tick, per-cpu. I don't think psignal()
375 * is mpsafe on curproc, so XXX get the mplock.
376 */
377 if ((p = curproc) != NULL && try_mplock()) {
984263bc 378 pstats = p->p_stats;
88c4d2f6 379 if (frame && CLKF_USERMODE(frame) &&
984263bc
MD
380 timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) &&
381 itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0)
382 psignal(p, SIGVTALRM);
383 if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value) &&
384 itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0)
385 psignal(p, SIGPROF);
88c4d2f6 386 rel_mplock();
984263bc 387 }
604e1e09 388 setdelayed();
88c4d2f6 389}
984263bc 390
88c4d2f6
MD
391/*
392 * The statistics clock typically runs at a 125Hz rate, and is intended
393 * to be frequency offset from the hardclock (typ 100Hz). It is per-cpu.
394 *
395 * NOTE! systimer! the MP lock might not be held here. We can only safely
396 * manipulate objects owned by the current cpu.
397 *
398 * The stats clock is responsible for grabbing a profiling sample.
399 * Most of the statistics are only used by user-level statistics programs.
400 * The main exceptions are p->p_uticks, p->p_sticks, p->p_iticks, and
401 * p->p_estcpu.
402 *
403 * Like the other clocks, the stat clock is called from what is effectively
404 * a fast interrupt, so the context should be the thread/process that got
405 * interrupted.
406 */
407static void
408statclock(systimer_t info, struct intrframe *frame)
409{
410#ifdef GPROF
411 struct gmonparam *g;
412 int i;
984263bc 413#endif
88c4d2f6
MD
414 thread_t td;
415 struct proc *p;
416 int bump;
417 struct timeval tv;
418 struct timeval *stv;
984263bc
MD
419
420 /*
88c4d2f6 421 * How big was our timeslice relative to the last time?
984263bc 422 */
88c4d2f6
MD
423 microuptime(&tv); /* mpsafe */
424 stv = &mycpu->gd_stattv;
425 if (stv->tv_sec == 0) {
426 bump = 1;
427 } else {
428 bump = tv.tv_usec - stv->tv_usec +
429 (tv.tv_sec - stv->tv_sec) * 1000000;
430 if (bump < 0)
431 bump = 0;
432 if (bump > 1000000)
433 bump = 1000000;
434 }
435 *stv = tv;
984263bc 436
88c4d2f6
MD
437 td = curthread;
438 p = td->td_proc;
984263bc 439
88c4d2f6
MD
440 if (frame && CLKF_USERMODE(frame)) {
441 /*
442 * Came from userland, handle user time and deal with
443 * possible process.
444 */
445 if (p && (p->p_flag & P_PROFIL))
446 addupc_intr(p, CLKF_PC(frame), 1);
447 td->td_uticks += bump;
984263bc 448
88c4d2f6
MD
449 /*
450 * Charge the time as appropriate
451 */
452 if (p && p->p_nice > NZERO)
453 cp_time[CP_NICE] += bump;
454 else
455 cp_time[CP_USER] += bump;
456 } else {
457#ifdef GPROF
458 /*
459 * Kernel statistics are just like addupc_intr, only easier.
460 */
461 g = &_gmonparam;
462 if (g->state == GMON_PROF_ON && frame) {
463 i = CLKF_PC(frame) - g->lowpc;
464 if (i < g->textsize) {
465 i /= HISTFRACTION * sizeof(*g->kcount);
466 g->kcount[i]++;
467 }
468 }
469#endif
470 /*
471 * Came from kernel mode, so we were:
472 * - handling an interrupt,
473 * - doing syscall or trap work on behalf of the current
474 * user process, or
475 * - spinning in the idle loop.
476 * Whichever it is, charge the time as appropriate.
477 * Note that we charge interrupts to the current process,
478 * regardless of whether they are ``for'' that process,
479 * so that we know how much of its real time was spent
480 * in ``non-process'' (i.e., interrupt) work.
481 *
482 * XXX assume system if frame is NULL. A NULL frame
483 * can occur if ipi processing is done from an splx().
484 */
485 if (frame && CLKF_INTR(frame))
486 td->td_iticks += bump;
487 else
488 td->td_sticks += bump;
489
490 if (frame && CLKF_INTR(frame)) {
491 cp_time[CP_INTR] += bump;
492 } else {
493 if (td == &mycpu->gd_idlethread)
494 cp_time[CP_IDLE] += bump;
495 else
496 cp_time[CP_SYS] += bump;
497 }
498 }
499}
500
501/*
0a3f9b47 502 * The scheduler clock typically runs at a 20Hz rate. NOTE! systimer,
88c4d2f6
MD
503 * the MP lock might not be held. We can safely manipulate parts of curproc
504 * but that's about it.
505 */
506static void
507schedclock(systimer_t info, struct intrframe *frame)
508{
509 struct proc *p;
510 struct pstats *pstats;
511 struct rusage *ru;
512 struct vmspace *vm;
513 long rss;
514
515 schedulerclock(NULL); /* mpsafe */
516 if ((p = curproc) != NULL) {
517 /* Update resource usage integrals and maximums. */
518 if ((pstats = p->p_stats) != NULL &&
519 (ru = &pstats->p_ru) != NULL &&
520 (vm = p->p_vmspace) != NULL) {
521 ru->ru_ixrss += pgtok(vm->vm_tsize);
522 ru->ru_idrss += pgtok(vm->vm_dsize);
523 ru->ru_isrss += pgtok(vm->vm_ssize);
524 rss = pgtok(vmspace_resident_count(vm));
525 if (ru->ru_maxrss < rss)
526 ru->ru_maxrss = rss;
527 }
b68b7282 528 }
984263bc
MD
529}
530
531/*
a94976ad
MD
532 * Compute number of ticks for the specified amount of time. The
533 * return value is intended to be used in a clock interrupt timed
534 * operation and guarenteed to meet or exceed the requested time.
535 * If the representation overflows, return INT_MAX. The minimum return
536 * value is 1 ticks and the function will average the calculation up.
537 * If any value greater then 0 microseconds is supplied, a value
538 * of at least 2 will be returned to ensure that a near-term clock
539 * interrupt does not cause the timeout to occur (degenerately) early.
540 *
541 * Note that limit checks must take into account microseconds, which is
542 * done simply by using the smaller signed long maximum instead of
543 * the unsigned long maximum.
544 *
545 * If ints have 32 bits, then the maximum value for any timeout in
546 * 10ms ticks is 248 days.
984263bc
MD
547 */
548int
a94976ad 549tvtohz_high(struct timeval *tv)
984263bc 550{
a94976ad 551 int ticks;
1fd87d54 552 long sec, usec;
984263bc 553
984263bc
MD
554 sec = tv->tv_sec;
555 usec = tv->tv_usec;
556 if (usec < 0) {
557 sec--;
558 usec += 1000000;
559 }
560 if (sec < 0) {
561#ifdef DIAGNOSTIC
562 if (usec > 0) {
563 sec++;
564 usec -= 1000000;
565 }
566 printf("tvotohz: negative time difference %ld sec %ld usec\n",
567 sec, usec);
568#endif
569 ticks = 1;
a94976ad
MD
570 } else if (sec <= INT_MAX / hz) {
571 ticks = (int)(sec * hz +
572 ((u_long)usec + (tick - 1)) / tick) + 1;
573 } else {
574 ticks = INT_MAX;
575 }
576 return (ticks);
577}
578
579/*
580 * Compute number of ticks for the specified amount of time, erroring on
581 * the side of it being too low to ensure that sleeping the returned number
582 * of ticks will not result in a late return.
583 *
584 * The supplied timeval may not be negative and should be normalized. A
585 * return value of 0 is possible if the timeval converts to less then
586 * 1 tick.
587 *
588 * If ints have 32 bits, then the maximum value for any timeout in
589 * 10ms ticks is 248 days.
590 */
591int
592tvtohz_low(struct timeval *tv)
593{
594 int ticks;
595 long sec;
596
597 sec = tv->tv_sec;
598 if (sec <= INT_MAX / hz)
599 ticks = (int)(sec * hz + (u_long)tv->tv_usec / tick);
984263bc 600 else
984263bc 601 ticks = INT_MAX;
a94976ad 602 return (ticks);
984263bc
MD
603}
604
a94976ad 605
984263bc
MD
606/*
607 * Start profiling on a process.
608 *
609 * Kernel profiling passes proc0 which never exits and hence
610 * keeps the profile clock running constantly.
611 */
612void
88c4d2f6 613startprofclock(struct proc *p)
984263bc 614{
984263bc
MD
615 if ((p->p_flag & P_PROFIL) == 0) {
616 p->p_flag |= P_PROFIL;
88c4d2f6 617#if 0 /* XXX */
984263bc
MD
618 if (++profprocs == 1 && stathz != 0) {
619 s = splstatclock();
6ad39cae 620 psdiv = psratio;
984263bc
MD
621 setstatclockrate(profhz);
622 splx(s);
623 }
88c4d2f6 624#endif
984263bc
MD
625 }
626}
627
628/*
629 * Stop profiling on a process.
630 */
631void
88c4d2f6 632stopprofclock(struct proc *p)
984263bc 633{
984263bc
MD
634 if (p->p_flag & P_PROFIL) {
635 p->p_flag &= ~P_PROFIL;
88c4d2f6 636#if 0 /* XXX */
984263bc
MD
637 if (--profprocs == 0 && stathz != 0) {
638 s = splstatclock();
6ad39cae 639 psdiv = 1;
984263bc
MD
640 setstatclockrate(stathz);
641 splx(s);
642 }
984263bc 643#endif
984263bc
MD
644 }
645}
646
647/*
648 * Return information about system clocks.
649 */
650static int
651sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS)
652{
653 struct clockinfo clkinfo;
654 /*
655 * Construct clockinfo structure.
656 */
657 clkinfo.hz = hz;
658 clkinfo.tick = tick;
659 clkinfo.tickadj = tickadj;
660 clkinfo.profhz = profhz;
661 clkinfo.stathz = stathz ? stathz : hz;
662 return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req));
663}
664
665SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate, CTLTYPE_STRUCT|CTLFLAG_RD,
666 0, 0, sysctl_kern_clockrate, "S,clockinfo","");
667
984263bc
MD
668/*
669 * We have eight functions for looking at the clock, four for
670 * microseconds and four for nanoseconds. For each there is fast
671 * but less precise version "get{nano|micro}[up]time" which will
672 * return a time which is up to 1/HZ previous to the call, whereas
673 * the raw version "{nano|micro}[up]time" will return a timestamp
674 * which is as precise as possible. The "up" variants return the
675 * time relative to system boot, these are well suited for time
676 * interval measurements.
88c4d2f6
MD
677 *
678 * Each cpu independantly maintains the current time of day, so all
679 * we need to do to protect ourselves from changes is to do a loop
680 * check on the seconds field changing out from under us.
984263bc 681 */
984263bc
MD
682void
683getmicrouptime(struct timeval *tvp)
684{
88c4d2f6
MD
685 struct globaldata *gd = mycpu;
686 sysclock_t delta;
687
688 do {
689 tvp->tv_sec = gd->gd_time_seconds;
690 delta = gd->gd_hardclock.time - gd->gd_cpuclock_base;
691 } while (tvp->tv_sec != gd->gd_time_seconds);
692 tvp->tv_usec = (cputimer_freq64_usec * delta) >> 32;
693 if (tvp->tv_usec >= 1000000) {
694 tvp->tv_usec -= 1000000;
695 ++tvp->tv_sec;
984263bc
MD
696 }
697}
698
699void
700getnanouptime(struct timespec *tsp)
701{
88c4d2f6
MD
702 struct globaldata *gd = mycpu;
703 sysclock_t delta;
704
705 do {
706 tsp->tv_sec = gd->gd_time_seconds;
707 delta = gd->gd_hardclock.time - gd->gd_cpuclock_base;
708 } while (tsp->tv_sec != gd->gd_time_seconds);
709 tsp->tv_nsec = (cputimer_freq64_nsec * delta) >> 32;
710 if (tsp->tv_nsec >= 1000000000) {
711 tsp->tv_nsec -= 1000000000;
712 ++tsp->tv_sec;
984263bc
MD
713 }
714}
715
716void
88c4d2f6 717microuptime(struct timeval *tvp)
984263bc 718{
88c4d2f6
MD
719 struct globaldata *gd = mycpu;
720 sysclock_t delta;
721
722 do {
723 tvp->tv_sec = gd->gd_time_seconds;
724 delta = cputimer_count() - gd->gd_cpuclock_base;
725 } while (tvp->tv_sec != gd->gd_time_seconds);
726 tvp->tv_usec = (cputimer_freq64_usec * delta) >> 32;
727 if (tvp->tv_usec >= 1000000) {
728 tvp->tv_usec -= 1000000;
729 ++tvp->tv_sec;
984263bc
MD
730 }
731}
732
733void
88c4d2f6 734nanouptime(struct timespec *tsp)
984263bc 735{
88c4d2f6
MD
736 struct globaldata *gd = mycpu;
737 sysclock_t delta;
738
739 do {
740 tsp->tv_sec = gd->gd_time_seconds;
741 delta = cputimer_count() - gd->gd_cpuclock_base;
742 } while (tsp->tv_sec != gd->gd_time_seconds);
743 tsp->tv_nsec = (cputimer_freq64_nsec * delta) >> 32;
744 if (tsp->tv_nsec >= 1000000000) {
745 tsp->tv_nsec -= 1000000000;
746 ++tsp->tv_sec;
984263bc 747 }
984263bc
MD
748}
749
88c4d2f6
MD
750/*
751 * realtime routines
752 */
984263bc
MD
753
754void
88c4d2f6 755getmicrotime(struct timeval *tvp)
984263bc 756{
88c4d2f6
MD
757 struct globaldata *gd = mycpu;
758 sysclock_t delta;
984263bc 759
88c4d2f6
MD
760 do {
761 tvp->tv_sec = gd->gd_time_seconds;
762 delta = gd->gd_hardclock.time - gd->gd_cpuclock_base;
763 } while (tvp->tv_sec != gd->gd_time_seconds);
764 tvp->tv_usec = (cputimer_freq64_usec * delta) >> 32;
984263bc 765
88c4d2f6
MD
766 tvp->tv_sec += basetime.tv_sec;
767 tvp->tv_usec += basetime.tv_nsec / 1000;
768 while (tvp->tv_usec >= 1000000) {
769 tvp->tv_usec -= 1000000;
770 ++tvp->tv_sec;
984263bc 771 }
984263bc
MD
772}
773
774void
88c4d2f6 775getnanotime(struct timespec *tsp)
984263bc 776{
88c4d2f6
MD
777 struct globaldata *gd = mycpu;
778 sysclock_t delta;
984263bc 779
88c4d2f6
MD
780 do {
781 tsp->tv_sec = gd->gd_time_seconds;
782 delta = gd->gd_hardclock.time - gd->gd_cpuclock_base;
783 } while (tsp->tv_sec != gd->gd_time_seconds);
784 tsp->tv_nsec = (cputimer_freq64_nsec * delta) >> 32;
984263bc 785
88c4d2f6
MD
786 tsp->tv_sec += basetime.tv_sec;
787 tsp->tv_nsec += basetime.tv_nsec;
788 while (tsp->tv_nsec >= 1000000000) {
789 tsp->tv_nsec -= 1000000000;
790 ++tsp->tv_sec;
984263bc 791 }
984263bc
MD
792}
793
88c4d2f6
MD
794void
795microtime(struct timeval *tvp)
984263bc 796{
88c4d2f6
MD
797 struct globaldata *gd = mycpu;
798 sysclock_t delta;
984263bc 799
88c4d2f6
MD
800 do {
801 tvp->tv_sec = gd->gd_time_seconds;
802 delta = cputimer_count() - gd->gd_cpuclock_base;
803 } while (tvp->tv_sec != gd->gd_time_seconds);
804 tvp->tv_usec = (cputimer_freq64_usec * delta) >> 32;
984263bc 805
88c4d2f6
MD
806 tvp->tv_sec += basetime.tv_sec;
807 tvp->tv_usec += basetime.tv_nsec / 1000;
808 while (tvp->tv_usec >= 1000000) {
809 tvp->tv_usec -= 1000000;
810 ++tvp->tv_sec;
984263bc 811 }
984263bc
MD
812}
813
88c4d2f6
MD
814void
815nanotime(struct timespec *tsp)
816{
817 struct globaldata *gd = mycpu;
818 sysclock_t delta;
984263bc 819
88c4d2f6
MD
820 do {
821 tsp->tv_sec = gd->gd_time_seconds;
822 delta = cputimer_count() - gd->gd_cpuclock_base;
823 } while (tsp->tv_sec != gd->gd_time_seconds);
824 tsp->tv_nsec = (cputimer_freq64_nsec * delta) >> 32;
984263bc 825
88c4d2f6
MD
826 tsp->tv_sec += basetime.tv_sec;
827 tsp->tv_nsec += basetime.tv_nsec;
828 while (tsp->tv_nsec >= 1000000000) {
829 tsp->tv_nsec -= 1000000000;
830 ++tsp->tv_sec;
984263bc 831 }
984263bc
MD
832}
833
984263bc
MD
834int
835pps_ioctl(u_long cmd, caddr_t data, struct pps_state *pps)
836{
837 pps_params_t *app;
838 struct pps_fetch_args *fapi;
839#ifdef PPS_SYNC
840 struct pps_kcbind_args *kapi;
841#endif
842
843 switch (cmd) {
844 case PPS_IOC_CREATE:
845 return (0);
846 case PPS_IOC_DESTROY:
847 return (0);
848 case PPS_IOC_SETPARAMS:
849 app = (pps_params_t *)data;
850 if (app->mode & ~pps->ppscap)
851 return (EINVAL);
852 pps->ppsparam = *app;
853 return (0);
854 case PPS_IOC_GETPARAMS:
855 app = (pps_params_t *)data;
856 *app = pps->ppsparam;
857 app->api_version = PPS_API_VERS_1;
858 return (0);
859 case PPS_IOC_GETCAP:
860 *(int*)data = pps->ppscap;
861 return (0);
862 case PPS_IOC_FETCH:
863 fapi = (struct pps_fetch_args *)data;
864 if (fapi->tsformat && fapi->tsformat != PPS_TSFMT_TSPEC)
865 return (EINVAL);
866 if (fapi->timeout.tv_sec || fapi->timeout.tv_nsec)
867 return (EOPNOTSUPP);
868 pps->ppsinfo.current_mode = pps->ppsparam.mode;
869 fapi->pps_info_buf = pps->ppsinfo;
870 return (0);
871 case PPS_IOC_KCBIND:
872#ifdef PPS_SYNC
873 kapi = (struct pps_kcbind_args *)data;
874 /* XXX Only root should be able to do this */
875 if (kapi->tsformat && kapi->tsformat != PPS_TSFMT_TSPEC)
876 return (EINVAL);
877 if (kapi->kernel_consumer != PPS_KC_HARDPPS)
878 return (EINVAL);
879 if (kapi->edge & ~pps->ppscap)
880 return (EINVAL);
881 pps->kcmode = kapi->edge;
882 return (0);
883#else
884 return (EOPNOTSUPP);
885#endif
886 default:
887 return (ENOTTY);
888 }
889}
890
891void
892pps_init(struct pps_state *pps)
893{
894 pps->ppscap |= PPS_TSFMT_TSPEC;
895 if (pps->ppscap & PPS_CAPTUREASSERT)
896 pps->ppscap |= PPS_OFFSETASSERT;
897 if (pps->ppscap & PPS_CAPTURECLEAR)
898 pps->ppscap |= PPS_OFFSETCLEAR;
899}
900
901void
88c4d2f6 902pps_event(struct pps_state *pps, sysclock_t count, int event)
984263bc 903{
88c4d2f6
MD
904 struct globaldata *gd;
905 struct timespec *tsp;
906 struct timespec *osp;
907 struct timespec ts;
908 sysclock_t *pcount;
909#ifdef PPS_SYNC
910 sysclock_t tcount;
911#endif
912 sysclock_t delta;
913 pps_seq_t *pseq;
914 int foff;
915 int fhard;
916
917 gd = mycpu;
984263bc
MD
918
919 /* Things would be easier with arrays... */
920 if (event == PPS_CAPTUREASSERT) {
921 tsp = &pps->ppsinfo.assert_timestamp;
922 osp = &pps->ppsparam.assert_offset;
923 foff = pps->ppsparam.mode & PPS_OFFSETASSERT;
924 fhard = pps->kcmode & PPS_CAPTUREASSERT;
925 pcount = &pps->ppscount[0];
926 pseq = &pps->ppsinfo.assert_sequence;
927 } else {
928 tsp = &pps->ppsinfo.clear_timestamp;
929 osp = &pps->ppsparam.clear_offset;
930 foff = pps->ppsparam.mode & PPS_OFFSETCLEAR;
931 fhard = pps->kcmode & PPS_CAPTURECLEAR;
932 pcount = &pps->ppscount[1];
933 pseq = &pps->ppsinfo.clear_sequence;
934 }
935
984263bc
MD
936 /* Nothing really happened */
937 if (*pcount == count)
938 return;
939
940 *pcount = count;
941
88c4d2f6
MD
942 do {
943 ts.tv_sec = gd->gd_time_seconds;
944 delta = count - gd->gd_cpuclock_base;
945 } while (ts.tv_sec != gd->gd_time_seconds);
946 if (delta > cputimer_freq) {
947 ts.tv_sec += delta / cputimer_freq;
948 delta %= cputimer_freq;
949 }
950 ts.tv_nsec = (cputimer_freq64_nsec * delta) >> 32;
951 ts.tv_sec += basetime.tv_sec;
952 ts.tv_nsec += basetime.tv_nsec;
953 while (ts.tv_nsec >= 1000000000) {
954 ts.tv_nsec -= 1000000000;
955 ++ts.tv_sec;
984263bc 956 }
984263bc
MD
957
958 (*pseq)++;
959 *tsp = ts;
960
961 if (foff) {
962 timespecadd(tsp, osp);
963 if (tsp->tv_nsec < 0) {
964 tsp->tv_nsec += 1000000000;
965 tsp->tv_sec -= 1;
966 }
967 }
968#ifdef PPS_SYNC
969 if (fhard) {
970 /* magic, at its best... */
971 tcount = count - pps->ppscount[2];
972 pps->ppscount[2] = count;
88c4d2f6 973 delta = (cputimer_freq64_nsec * tcount) >> 32;
984263bc
MD
974 hardpps(tsp, delta);
975 }
976#endif
977}
88c4d2f6 978