xref: /dflybsd-src/sys/kern/kern_clock.c (revision bc49aa1be5400e3bdd801519c6936e8947d5d432)
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  *
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  * $DragonFly: src/sys/kern/kern_clock.c,v 1.21 2004/07/16 05:51:09 dillon Exp $
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>
94 #include <sys/thread2.h>
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
105 extern void init_device_poll(void);
106 extern void hardclock_device_poll(void);
107 #endif /* DEVICE_POLLING */
108 
109 static void initclocks (void *dummy);
110 SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL)
111 
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  */
117 long cp_time[CPUSTATES];
118 
119 SYSCTL_OPAQUE(_kern, OID_AUTO, cp_time, CTLFLAG_RD, &cp_time, sizeof(cp_time),
120     "LU", "CPU time statistics");
121 
122 long tk_cancc;
123 long tk_nin;
124 long tk_nout;
125 long tk_rawcc;
126 
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
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).
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  */
141 struct timespec boottime;	/* boot time (realtime) for reference only */
142 struct timespec basetime;	/* base time adjusts uptime -> realtime */
143 time_t time_second;		/* read-only 'passive' uptime in seconds */
144 
145 SYSCTL_STRUCT(_kern, KERN_BOOTTIME, boottime, CTLFLAG_RD,
146     &boottime, timeval, "System boottime");
147 SYSCTL_STRUCT(_kern, OID_AUTO, basetime, CTLFLAG_RD,
148     &basetime, timeval, "System basetime");
149 
150 static void hardclock(systimer_t info, struct intrframe *frame);
151 static void statclock(systimer_t info, struct intrframe *frame);
152 static void schedclock(systimer_t info, struct intrframe *frame);
153 
154 int	ticks;			/* system master ticks at hz */
155 int64_t	nsec_adj;		/* ntpd per-tick adjustment in nsec << 32 */
156 int64_t	nsec_acc;		/* accumulator */
157 
158 /*
159  * Finish initializing clock frequencies and start all clocks running.
160  */
161 /* ARGSUSED*/
162 static void
163 initclocks(void *dummy)
164 {
165 	cpu_initclocks();
166 #ifdef DEVICE_POLLING
167 	init_device_poll();
168 #endif
169 	/*psratio = profhz / stathz;*/
170 	initclocks_pcpu();
171 }
172 
173 /*
174  * Called on a per-cpu basis
175  */
176 void
177 initclocks_pcpu(void)
178 {
179 	struct globaldata *gd = mycpu;
180 
181 	crit_enter();
182 	if (gd->gd_cpuid == 0) {
183 	    gd->gd_time_seconds = 1;
184 	    gd->gd_cpuclock_base = cputimer_count();
185 	} else {
186 	    /* XXX */
187 	    gd->gd_time_seconds = globaldata_find(0)->gd_time_seconds;
188 	    gd->gd_cpuclock_base = globaldata_find(0)->gd_cpuclock_base;
189 	}
190 	systimer_init_periodic(&gd->gd_hardclock, hardclock, NULL, hz);
191 	systimer_init_periodic(&gd->gd_statclock, statclock, NULL, stathz);
192 	/* XXX correct the frequency for scheduler / estcpu tests */
193 	systimer_init_periodic(&gd->gd_schedclock, schedclock,
194 				NULL, ESTCPUFREQ);
195 	crit_exit();
196 }
197 
198 /*
199  * This sets the current real time of day.  Timespecs are in seconds and
200  * nanoseconds.  We do not mess with gd_time_seconds and gd_cpuclock_base,
201  * instead we adjust basetime so basetime + gd_* results in the current
202  * time of day.  This way the gd_* fields are guarenteed to represent
203  * a monotonically increasing 'uptime' value.
204  */
205 void
206 set_timeofday(struct timespec *ts)
207 {
208 	struct timespec ts2;
209 
210 	/*
211 	 * XXX SMP / non-atomic basetime updates
212 	 */
213 	crit_enter();
214 	nanouptime(&ts2);
215 	basetime.tv_sec = ts->tv_sec - ts2.tv_sec;
216 	basetime.tv_nsec = ts->tv_nsec - ts2.tv_nsec;
217 	if (basetime.tv_nsec < 0) {
218 	    basetime.tv_nsec += 1000000000;
219 	    --basetime.tv_sec;
220 	}
221 	boottime.tv_sec = basetime.tv_sec - mycpu->gd_time_seconds;
222 	timedelta = 0;
223 	crit_exit();
224 }
225 
226 /*
227  * Each cpu has its own hardclock, but we only increments ticks and softticks
228  * on cpu #0.
229  *
230  * NOTE! systimer! the MP lock might not be held here.  We can only safely
231  * manipulate objects owned by the current cpu.
232  */
233 static void
234 hardclock(systimer_t info, struct intrframe *frame)
235 {
236 	sysclock_t cputicks;
237 	struct proc *p;
238 	struct pstats *pstats;
239 	struct globaldata *gd = mycpu;
240 
241 	/*
242 	 * Realtime updates are per-cpu.  Note that timer corrections as
243 	 * returned by microtime() and friends make an additional adjustment
244 	 * using a system-wise 'basetime', but the running time is always
245 	 * taken from the per-cpu globaldata area.  Since the same clock
246 	 * is distributing (XXX SMP) to all cpus, the per-cpu timebases
247 	 * stay in synch.
248 	 *
249 	 * Note that we never allow info->time (aka gd->gd_hardclock.time)
250 	 * to reverse index gd_cpuclock_base.
251 	 */
252 	cputicks = info->time - gd->gd_cpuclock_base;
253 	if (cputicks > cputimer_freq) {
254 		++gd->gd_time_seconds;
255 		gd->gd_cpuclock_base += cputimer_freq;
256 	}
257 
258 	/*
259 	 * The system-wide ticks and softticks are only updated by cpu #0.
260 	 * Callwheel actions are also (at the moment) only handled by cpu #0.
261 	 * Finally, we also do NTP related timedelta/tickdelta adjustments
262 	 * by adjusting basetime.
263 	 */
264 	if (gd->gd_cpuid == 0) {
265 	    struct timespec nts;
266 	    int leap;
267 
268 	    ++ticks;
269 
270 #ifdef DEVICE_POLLING
271 	    hardclock_device_poll();	/* mpsafe, short and quick */
272 #endif /* DEVICE_POLLING */
273 
274 	    if (TAILQ_FIRST(&callwheel[ticks & callwheelmask]) != NULL) {
275 		setsoftclock();
276 	    } else if (softticks + 1 == ticks) {
277 		++softticks;
278 	    }
279 
280 #if 0
281 	    if (tco->tc_poll_pps)
282 		tco->tc_poll_pps(tco);
283 #endif
284 	    /*
285 	     * Apply adjtime corrections.  At the moment only do this if
286 	     * we can get the MP lock to interlock with adjtime's modification
287 	     * of these variables.  Note that basetime adjustments are not
288 	     * MP safe either XXX.
289 	     */
290 	    if (timedelta != 0 && try_mplock()) {
291 		basetime.tv_nsec += tickdelta * 1000;
292 		if (basetime.tv_nsec >= 1000000000) {
293 		    basetime.tv_nsec -= 1000000000;
294 		    ++basetime.tv_sec;
295 		} else if (basetime.tv_nsec < 0) {
296 		    basetime.tv_nsec += 1000000000;
297 		    --basetime.tv_sec;
298 		}
299 		timedelta -= tickdelta;
300 		rel_mplock();
301 	    }
302 
303 	    /*
304 	     * Apply per-tick compensation.  ticks_adj adjusts for both
305 	     * offset and frequency, and could be negative.
306 	     */
307 	    if (nsec_adj != 0 && try_mplock()) {
308 		nsec_acc += nsec_adj;
309 		if (nsec_acc >= 0x100000000LL) {
310 		    basetime.tv_nsec += nsec_acc >> 32;
311 		    nsec_acc = (nsec_acc & 0xFFFFFFFFLL);
312 		} else if (nsec_acc <= -0x100000000LL) {
313 		    basetime.tv_nsec -= -nsec_acc >> 32;
314 		    nsec_acc = -(-nsec_acc & 0xFFFFFFFFLL);
315 		}
316 		if (basetime.tv_nsec >= 1000000000) {
317 		    basetime.tv_nsec -= 1000000000;
318 		    ++basetime.tv_sec;
319 		} else if (basetime.tv_nsec < 0) {
320 		    basetime.tv_nsec += 1000000000;
321 		    --basetime.tv_sec;
322 		}
323 		rel_mplock();
324 	    }
325 
326 	    /*
327 	     * If the realtime-adjusted seconds hand rolls over then tell
328 	     * ntp_update_second() what we did in the last second so it can
329 	     * calculate what to do in the next second.  It may also add
330 	     * or subtract a leap second.
331 	     */
332 	    getnanotime(&nts);
333 	    if (time_second != nts.tv_sec) {
334 		leap = ntp_update_second(time_second, &nsec_adj);
335 		basetime.tv_sec += leap;
336 		time_second = nts.tv_sec + leap;
337 		nsec_adj /= hz;
338 	    }
339 	}
340 
341 	/*
342 	 * ITimer handling is per-tick, per-cpu.  I don't think psignal()
343 	 * is mpsafe on curproc, so XXX get the mplock.
344 	 */
345 	if ((p = curproc) != NULL && try_mplock()) {
346 		pstats = p->p_stats;
347 		if (frame && CLKF_USERMODE(frame) &&
348 		    timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) &&
349 		    itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0)
350 			psignal(p, SIGVTALRM);
351 		if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value) &&
352 		    itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0)
353 			psignal(p, SIGPROF);
354 		rel_mplock();
355 	}
356 	setdelayed();
357 }
358 
359 /*
360  * The statistics clock typically runs at a 125Hz rate, and is intended
361  * to be frequency offset from the hardclock (typ 100Hz).  It is per-cpu.
362  *
363  * NOTE! systimer! the MP lock might not be held here.  We can only safely
364  * manipulate objects owned by the current cpu.
365  *
366  * The stats clock is responsible for grabbing a profiling sample.
367  * Most of the statistics are only used by user-level statistics programs.
368  * The main exceptions are p->p_uticks, p->p_sticks, p->p_iticks, and
369  * p->p_estcpu.
370  *
371  * Like the other clocks, the stat clock is called from what is effectively
372  * a fast interrupt, so the context should be the thread/process that got
373  * interrupted.
374  */
375 static void
376 statclock(systimer_t info, struct intrframe *frame)
377 {
378 #ifdef GPROF
379 	struct gmonparam *g;
380 	int i;
381 #endif
382 	thread_t td;
383 	struct proc *p;
384 	int bump;
385 	struct timeval tv;
386 	struct timeval *stv;
387 
388 	/*
389 	 * How big was our timeslice relative to the last time?
390 	 */
391 	microuptime(&tv);	/* mpsafe */
392 	stv = &mycpu->gd_stattv;
393 	if (stv->tv_sec == 0) {
394 	    bump = 1;
395 	} else {
396 	    bump = tv.tv_usec - stv->tv_usec +
397 		(tv.tv_sec - stv->tv_sec) * 1000000;
398 	    if (bump < 0)
399 		bump = 0;
400 	    if (bump > 1000000)
401 		bump = 1000000;
402 	}
403 	*stv = tv;
404 
405 	td = curthread;
406 	p = td->td_proc;
407 
408 	if (frame && CLKF_USERMODE(frame)) {
409 		/*
410 		 * Came from userland, handle user time and deal with
411 		 * possible process.
412 		 */
413 		if (p && (p->p_flag & P_PROFIL))
414 			addupc_intr(p, CLKF_PC(frame), 1);
415 		td->td_uticks += bump;
416 
417 		/*
418 		 * Charge the time as appropriate
419 		 */
420 		if (p && p->p_nice > NZERO)
421 			cp_time[CP_NICE] += bump;
422 		else
423 			cp_time[CP_USER] += bump;
424 	} else {
425 #ifdef GPROF
426 		/*
427 		 * Kernel statistics are just like addupc_intr, only easier.
428 		 */
429 		g = &_gmonparam;
430 		if (g->state == GMON_PROF_ON && frame) {
431 			i = CLKF_PC(frame) - g->lowpc;
432 			if (i < g->textsize) {
433 				i /= HISTFRACTION * sizeof(*g->kcount);
434 				g->kcount[i]++;
435 			}
436 		}
437 #endif
438 		/*
439 		 * Came from kernel mode, so we were:
440 		 * - handling an interrupt,
441 		 * - doing syscall or trap work on behalf of the current
442 		 *   user process, or
443 		 * - spinning in the idle loop.
444 		 * Whichever it is, charge the time as appropriate.
445 		 * Note that we charge interrupts to the current process,
446 		 * regardless of whether they are ``for'' that process,
447 		 * so that we know how much of its real time was spent
448 		 * in ``non-process'' (i.e., interrupt) work.
449 		 *
450 		 * XXX assume system if frame is NULL.  A NULL frame
451 		 * can occur if ipi processing is done from an splx().
452 		 */
453 		if (frame && CLKF_INTR(frame))
454 			td->td_iticks += bump;
455 		else
456 			td->td_sticks += bump;
457 
458 		if (frame && CLKF_INTR(frame)) {
459 			cp_time[CP_INTR] += bump;
460 		} else {
461 			if (td == &mycpu->gd_idlethread)
462 				cp_time[CP_IDLE] += bump;
463 			else
464 				cp_time[CP_SYS] += bump;
465 		}
466 	}
467 }
468 
469 /*
470  * The scheduler clock typically runs at a 20Hz rate.  NOTE! systimer,
471  * the MP lock might not be held.  We can safely manipulate parts of curproc
472  * but that's about it.
473  */
474 static void
475 schedclock(systimer_t info, struct intrframe *frame)
476 {
477 	struct proc *p;
478 	struct pstats *pstats;
479 	struct rusage *ru;
480 	struct vmspace *vm;
481 	long rss;
482 
483 	schedulerclock(NULL);	/* mpsafe */
484 	if ((p = curproc) != NULL) {
485 		/* Update resource usage integrals and maximums. */
486 		if ((pstats = p->p_stats) != NULL &&
487 		    (ru = &pstats->p_ru) != NULL &&
488 		    (vm = p->p_vmspace) != NULL) {
489 			ru->ru_ixrss += pgtok(vm->vm_tsize);
490 			ru->ru_idrss += pgtok(vm->vm_dsize);
491 			ru->ru_isrss += pgtok(vm->vm_ssize);
492 			rss = pgtok(vmspace_resident_count(vm));
493 			if (ru->ru_maxrss < rss)
494 				ru->ru_maxrss = rss;
495 		}
496 	}
497 }
498 
499 /*
500  * Compute number of ticks for the specified amount of time.  The
501  * return value is intended to be used in a clock interrupt timed
502  * operation and guarenteed to meet or exceed the requested time.
503  * If the representation overflows, return INT_MAX.  The minimum return
504  * value is 1 ticks and the function will average the calculation up.
505  * If any value greater then 0 microseconds is supplied, a value
506  * of at least 2 will be returned to ensure that a near-term clock
507  * interrupt does not cause the timeout to occur (degenerately) early.
508  *
509  * Note that limit checks must take into account microseconds, which is
510  * done simply by using the smaller signed long maximum instead of
511  * the unsigned long maximum.
512  *
513  * If ints have 32 bits, then the maximum value for any timeout in
514  * 10ms ticks is 248 days.
515  */
516 int
517 tvtohz_high(struct timeval *tv)
518 {
519 	int ticks;
520 	long sec, usec;
521 
522 	sec = tv->tv_sec;
523 	usec = tv->tv_usec;
524 	if (usec < 0) {
525 		sec--;
526 		usec += 1000000;
527 	}
528 	if (sec < 0) {
529 #ifdef DIAGNOSTIC
530 		if (usec > 0) {
531 			sec++;
532 			usec -= 1000000;
533 		}
534 		printf("tvotohz: negative time difference %ld sec %ld usec\n",
535 		       sec, usec);
536 #endif
537 		ticks = 1;
538 	} else if (sec <= INT_MAX / hz) {
539 		ticks = (int)(sec * hz +
540 			    ((u_long)usec + (tick - 1)) / tick) + 1;
541 	} else {
542 		ticks = INT_MAX;
543 	}
544 	return (ticks);
545 }
546 
547 /*
548  * Compute number of ticks for the specified amount of time, erroring on
549  * the side of it being too low to ensure that sleeping the returned number
550  * of ticks will not result in a late return.
551  *
552  * The supplied timeval may not be negative and should be normalized.  A
553  * return value of 0 is possible if the timeval converts to less then
554  * 1 tick.
555  *
556  * If ints have 32 bits, then the maximum value for any timeout in
557  * 10ms ticks is 248 days.
558  */
559 int
560 tvtohz_low(struct timeval *tv)
561 {
562 	int ticks;
563 	long sec;
564 
565 	sec = tv->tv_sec;
566 	if (sec <= INT_MAX / hz)
567 		ticks = (int)(sec * hz + (u_long)tv->tv_usec / tick);
568 	else
569 		ticks = INT_MAX;
570 	return (ticks);
571 }
572 
573 
574 /*
575  * Start profiling on a process.
576  *
577  * Kernel profiling passes proc0 which never exits and hence
578  * keeps the profile clock running constantly.
579  */
580 void
581 startprofclock(struct proc *p)
582 {
583 	if ((p->p_flag & P_PROFIL) == 0) {
584 		p->p_flag |= P_PROFIL;
585 #if 0	/* XXX */
586 		if (++profprocs == 1 && stathz != 0) {
587 			s = splstatclock();
588 			psdiv = psratio;
589 			setstatclockrate(profhz);
590 			splx(s);
591 		}
592 #endif
593 	}
594 }
595 
596 /*
597  * Stop profiling on a process.
598  */
599 void
600 stopprofclock(struct proc *p)
601 {
602 	if (p->p_flag & P_PROFIL) {
603 		p->p_flag &= ~P_PROFIL;
604 #if 0	/* XXX */
605 		if (--profprocs == 0 && stathz != 0) {
606 			s = splstatclock();
607 			psdiv = 1;
608 			setstatclockrate(stathz);
609 			splx(s);
610 		}
611 #endif
612 	}
613 }
614 
615 /*
616  * Return information about system clocks.
617  */
618 static int
619 sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS)
620 {
621 	struct clockinfo clkinfo;
622 	/*
623 	 * Construct clockinfo structure.
624 	 */
625 	clkinfo.hz = hz;
626 	clkinfo.tick = tick;
627 	clkinfo.tickadj = tickadj;
628 	clkinfo.profhz = profhz;
629 	clkinfo.stathz = stathz ? stathz : hz;
630 	return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req));
631 }
632 
633 SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate, CTLTYPE_STRUCT|CTLFLAG_RD,
634 	0, 0, sysctl_kern_clockrate, "S,clockinfo","");
635 
636 /*
637  * We have eight functions for looking at the clock, four for
638  * microseconds and four for nanoseconds.  For each there is fast
639  * but less precise version "get{nano|micro}[up]time" which will
640  * return a time which is up to 1/HZ previous to the call, whereas
641  * the raw version "{nano|micro}[up]time" will return a timestamp
642  * which is as precise as possible.  The "up" variants return the
643  * time relative to system boot, these are well suited for time
644  * interval measurements.
645  *
646  * Each cpu independantly maintains the current time of day, so all
647  * we need to do to protect ourselves from changes is to do a loop
648  * check on the seconds field changing out from under us.
649  */
650 void
651 getmicrouptime(struct timeval *tvp)
652 {
653 	struct globaldata *gd = mycpu;
654 	sysclock_t delta;
655 
656 	do {
657 		tvp->tv_sec = gd->gd_time_seconds;
658 		delta = gd->gd_hardclock.time - gd->gd_cpuclock_base;
659 	} while (tvp->tv_sec != gd->gd_time_seconds);
660 	tvp->tv_usec = (cputimer_freq64_usec * delta) >> 32;
661 	if (tvp->tv_usec >= 1000000) {
662 		tvp->tv_usec -= 1000000;
663 		++tvp->tv_sec;
664 	}
665 }
666 
667 void
668 getnanouptime(struct timespec *tsp)
669 {
670 	struct globaldata *gd = mycpu;
671 	sysclock_t delta;
672 
673 	do {
674 		tsp->tv_sec = gd->gd_time_seconds;
675 		delta = gd->gd_hardclock.time - gd->gd_cpuclock_base;
676 	} while (tsp->tv_sec != gd->gd_time_seconds);
677 	tsp->tv_nsec = (cputimer_freq64_nsec * delta) >> 32;
678 	if (tsp->tv_nsec >= 1000000000) {
679 		tsp->tv_nsec -= 1000000000;
680 		++tsp->tv_sec;
681 	}
682 }
683 
684 void
685 microuptime(struct timeval *tvp)
686 {
687 	struct globaldata *gd = mycpu;
688 	sysclock_t delta;
689 
690 	do {
691 		tvp->tv_sec = gd->gd_time_seconds;
692 		delta = cputimer_count() - gd->gd_cpuclock_base;
693 	} while (tvp->tv_sec != gd->gd_time_seconds);
694 	tvp->tv_usec = (cputimer_freq64_usec * delta) >> 32;
695 	if (tvp->tv_usec >= 1000000) {
696 		tvp->tv_usec -= 1000000;
697 		++tvp->tv_sec;
698 	}
699 }
700 
701 void
702 nanouptime(struct timespec *tsp)
703 {
704 	struct globaldata *gd = mycpu;
705 	sysclock_t delta;
706 
707 	do {
708 		tsp->tv_sec = gd->gd_time_seconds;
709 		delta = cputimer_count() - gd->gd_cpuclock_base;
710 	} while (tsp->tv_sec != gd->gd_time_seconds);
711 	tsp->tv_nsec = (cputimer_freq64_nsec * delta) >> 32;
712 	if (tsp->tv_nsec >= 1000000000) {
713 		tsp->tv_nsec -= 1000000000;
714 		++tsp->tv_sec;
715 	}
716 }
717 
718 /*
719  * realtime routines
720  */
721 
722 void
723 getmicrotime(struct timeval *tvp)
724 {
725 	struct globaldata *gd = mycpu;
726 	sysclock_t delta;
727 
728 	do {
729 		tvp->tv_sec = gd->gd_time_seconds;
730 		delta = gd->gd_hardclock.time - gd->gd_cpuclock_base;
731 	} while (tvp->tv_sec != gd->gd_time_seconds);
732 	tvp->tv_usec = (cputimer_freq64_usec * delta) >> 32;
733 
734 	tvp->tv_sec += basetime.tv_sec;
735 	tvp->tv_usec += basetime.tv_nsec / 1000;
736 	while (tvp->tv_usec >= 1000000) {
737 		tvp->tv_usec -= 1000000;
738 		++tvp->tv_sec;
739 	}
740 }
741 
742 void
743 getnanotime(struct timespec *tsp)
744 {
745 	struct globaldata *gd = mycpu;
746 	sysclock_t delta;
747 
748 	do {
749 		tsp->tv_sec = gd->gd_time_seconds;
750 		delta = gd->gd_hardclock.time - gd->gd_cpuclock_base;
751 	} while (tsp->tv_sec != gd->gd_time_seconds);
752 	tsp->tv_nsec = (cputimer_freq64_nsec * delta) >> 32;
753 
754 	tsp->tv_sec += basetime.tv_sec;
755 	tsp->tv_nsec += basetime.tv_nsec;
756 	while (tsp->tv_nsec >= 1000000000) {
757 		tsp->tv_nsec -= 1000000000;
758 		++tsp->tv_sec;
759 	}
760 }
761 
762 void
763 microtime(struct timeval *tvp)
764 {
765 	struct globaldata *gd = mycpu;
766 	sysclock_t delta;
767 
768 	do {
769 		tvp->tv_sec = gd->gd_time_seconds;
770 		delta = cputimer_count() - gd->gd_cpuclock_base;
771 	} while (tvp->tv_sec != gd->gd_time_seconds);
772 	tvp->tv_usec = (cputimer_freq64_usec * delta) >> 32;
773 
774 	tvp->tv_sec += basetime.tv_sec;
775 	tvp->tv_usec += basetime.tv_nsec / 1000;
776 	while (tvp->tv_usec >= 1000000) {
777 		tvp->tv_usec -= 1000000;
778 		++tvp->tv_sec;
779 	}
780 }
781 
782 void
783 nanotime(struct timespec *tsp)
784 {
785 	struct globaldata *gd = mycpu;
786 	sysclock_t delta;
787 
788 	do {
789 		tsp->tv_sec = gd->gd_time_seconds;
790 		delta = cputimer_count() - gd->gd_cpuclock_base;
791 	} while (tsp->tv_sec != gd->gd_time_seconds);
792 	tsp->tv_nsec = (cputimer_freq64_nsec * delta) >> 32;
793 
794 	tsp->tv_sec += basetime.tv_sec;
795 	tsp->tv_nsec += basetime.tv_nsec;
796 	while (tsp->tv_nsec >= 1000000000) {
797 		tsp->tv_nsec -= 1000000000;
798 		++tsp->tv_sec;
799 	}
800 }
801 
802 int
803 pps_ioctl(u_long cmd, caddr_t data, struct pps_state *pps)
804 {
805 	pps_params_t *app;
806 	struct pps_fetch_args *fapi;
807 #ifdef PPS_SYNC
808 	struct pps_kcbind_args *kapi;
809 #endif
810 
811 	switch (cmd) {
812 	case PPS_IOC_CREATE:
813 		return (0);
814 	case PPS_IOC_DESTROY:
815 		return (0);
816 	case PPS_IOC_SETPARAMS:
817 		app = (pps_params_t *)data;
818 		if (app->mode & ~pps->ppscap)
819 			return (EINVAL);
820 		pps->ppsparam = *app;
821 		return (0);
822 	case PPS_IOC_GETPARAMS:
823 		app = (pps_params_t *)data;
824 		*app = pps->ppsparam;
825 		app->api_version = PPS_API_VERS_1;
826 		return (0);
827 	case PPS_IOC_GETCAP:
828 		*(int*)data = pps->ppscap;
829 		return (0);
830 	case PPS_IOC_FETCH:
831 		fapi = (struct pps_fetch_args *)data;
832 		if (fapi->tsformat && fapi->tsformat != PPS_TSFMT_TSPEC)
833 			return (EINVAL);
834 		if (fapi->timeout.tv_sec || fapi->timeout.tv_nsec)
835 			return (EOPNOTSUPP);
836 		pps->ppsinfo.current_mode = pps->ppsparam.mode;
837 		fapi->pps_info_buf = pps->ppsinfo;
838 		return (0);
839 	case PPS_IOC_KCBIND:
840 #ifdef PPS_SYNC
841 		kapi = (struct pps_kcbind_args *)data;
842 		/* XXX Only root should be able to do this */
843 		if (kapi->tsformat && kapi->tsformat != PPS_TSFMT_TSPEC)
844 			return (EINVAL);
845 		if (kapi->kernel_consumer != PPS_KC_HARDPPS)
846 			return (EINVAL);
847 		if (kapi->edge & ~pps->ppscap)
848 			return (EINVAL);
849 		pps->kcmode = kapi->edge;
850 		return (0);
851 #else
852 		return (EOPNOTSUPP);
853 #endif
854 	default:
855 		return (ENOTTY);
856 	}
857 }
858 
859 void
860 pps_init(struct pps_state *pps)
861 {
862 	pps->ppscap |= PPS_TSFMT_TSPEC;
863 	if (pps->ppscap & PPS_CAPTUREASSERT)
864 		pps->ppscap |= PPS_OFFSETASSERT;
865 	if (pps->ppscap & PPS_CAPTURECLEAR)
866 		pps->ppscap |= PPS_OFFSETCLEAR;
867 }
868 
869 void
870 pps_event(struct pps_state *pps, sysclock_t count, int event)
871 {
872 	struct globaldata *gd;
873 	struct timespec *tsp;
874 	struct timespec *osp;
875 	struct timespec ts;
876 	sysclock_t *pcount;
877 #ifdef PPS_SYNC
878 	sysclock_t tcount;
879 #endif
880 	sysclock_t delta;
881 	pps_seq_t *pseq;
882 	int foff;
883 	int fhard;
884 
885 	gd = mycpu;
886 
887 	/* Things would be easier with arrays... */
888 	if (event == PPS_CAPTUREASSERT) {
889 		tsp = &pps->ppsinfo.assert_timestamp;
890 		osp = &pps->ppsparam.assert_offset;
891 		foff = pps->ppsparam.mode & PPS_OFFSETASSERT;
892 		fhard = pps->kcmode & PPS_CAPTUREASSERT;
893 		pcount = &pps->ppscount[0];
894 		pseq = &pps->ppsinfo.assert_sequence;
895 	} else {
896 		tsp = &pps->ppsinfo.clear_timestamp;
897 		osp = &pps->ppsparam.clear_offset;
898 		foff = pps->ppsparam.mode & PPS_OFFSETCLEAR;
899 		fhard = pps->kcmode & PPS_CAPTURECLEAR;
900 		pcount = &pps->ppscount[1];
901 		pseq = &pps->ppsinfo.clear_sequence;
902 	}
903 
904 	/* Nothing really happened */
905 	if (*pcount == count)
906 		return;
907 
908 	*pcount = count;
909 
910 	do {
911 		ts.tv_sec = gd->gd_time_seconds;
912 		delta = count - gd->gd_cpuclock_base;
913 	} while (ts.tv_sec != gd->gd_time_seconds);
914 	if (delta > cputimer_freq) {
915 		ts.tv_sec += delta / cputimer_freq;
916 		delta %= cputimer_freq;
917 	}
918 	ts.tv_nsec = (cputimer_freq64_nsec * delta) >> 32;
919 	ts.tv_sec += basetime.tv_sec;
920 	ts.tv_nsec += basetime.tv_nsec;
921 	while (ts.tv_nsec >= 1000000000) {
922 		ts.tv_nsec -= 1000000000;
923 		++ts.tv_sec;
924 	}
925 
926 	(*pseq)++;
927 	*tsp = ts;
928 
929 	if (foff) {
930 		timespecadd(tsp, osp);
931 		if (tsp->tv_nsec < 0) {
932 			tsp->tv_nsec += 1000000000;
933 			tsp->tv_sec -= 1;
934 		}
935 	}
936 #ifdef PPS_SYNC
937 	if (fhard) {
938 		/* magic, at its best... */
939 		tcount = count - pps->ppscount[2];
940 		pps->ppscount[2] = count;
941 		delta = (cputimer_freq64_nsec * tcount) >> 32;
942 		hardpps(tsp, delta);
943 	}
944 #endif
945 }
946 
947