xref: /freebsd/sys/kern/kern_clock.c (revision aa12cea2ccc6e686d6d31cf67d6bc69cbc1ba744)
1 /*-
2  * Copyright (c) 1982, 1986, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)kern_clock.c	8.5 (Berkeley) 1/21/94
35  */
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 #include "opt_kdb.h"
41 #include "opt_device_polling.h"
42 #include "opt_hwpmc_hooks.h"
43 #include "opt_ntp.h"
44 #include "opt_watchdog.h"
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/callout.h>
49 #include <sys/kdb.h>
50 #include <sys/kernel.h>
51 #include <sys/kthread.h>
52 #include <sys/ktr.h>
53 #include <sys/lock.h>
54 #include <sys/mutex.h>
55 #include <sys/proc.h>
56 #include <sys/resource.h>
57 #include <sys/resourcevar.h>
58 #include <sys/sched.h>
59 #include <sys/signalvar.h>
60 #include <sys/sleepqueue.h>
61 #include <sys/smp.h>
62 #include <vm/vm.h>
63 #include <vm/pmap.h>
64 #include <vm/vm_map.h>
65 #include <sys/sysctl.h>
66 #include <sys/bus.h>
67 #include <sys/interrupt.h>
68 #include <sys/limits.h>
69 #include <sys/timetc.h>
70 
71 #ifdef GPROF
72 #include <sys/gmon.h>
73 #endif
74 
75 #ifdef HWPMC_HOOKS
76 #include <sys/pmckern.h>
77 #endif
78 
79 #ifdef DEVICE_POLLING
80 extern void hardclock_device_poll(void);
81 #endif /* DEVICE_POLLING */
82 
83 static void initclocks(void *dummy);
84 SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL);
85 
86 /* Spin-lock protecting profiling statistics. */
87 static struct mtx time_lock;
88 
89 static int
90 sysctl_kern_cp_time(SYSCTL_HANDLER_ARGS)
91 {
92 	int error;
93 	long cp_time[CPUSTATES];
94 #ifdef SCTL_MASK32
95 	int i;
96 	unsigned int cp_time32[CPUSTATES];
97 #endif
98 
99 	read_cpu_time(cp_time);
100 #ifdef SCTL_MASK32
101 	if (req->flags & SCTL_MASK32) {
102 		if (!req->oldptr)
103 			return SYSCTL_OUT(req, 0, sizeof(cp_time32));
104 		for (i = 0; i < CPUSTATES; i++)
105 			cp_time32[i] = (unsigned int)cp_time[i];
106 		error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32));
107 	} else
108 #endif
109 	{
110 		if (!req->oldptr)
111 			return SYSCTL_OUT(req, 0, sizeof(cp_time));
112 		error = SYSCTL_OUT(req, cp_time, sizeof(cp_time));
113 	}
114 	return error;
115 }
116 
117 SYSCTL_PROC(_kern, OID_AUTO, cp_time, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE,
118     0,0, sysctl_kern_cp_time, "LU", "CPU time statistics");
119 
120 static long empty[CPUSTATES];
121 
122 static int
123 sysctl_kern_cp_times(SYSCTL_HANDLER_ARGS)
124 {
125 	struct pcpu *pcpu;
126 	int error;
127 	int c;
128 	long *cp_time;
129 #ifdef SCTL_MASK32
130 	unsigned int cp_time32[CPUSTATES];
131 	int i;
132 #endif
133 
134 	if (!req->oldptr) {
135 #ifdef SCTL_MASK32
136 		if (req->flags & SCTL_MASK32)
137 			return SYSCTL_OUT(req, 0, sizeof(cp_time32) * (mp_maxid + 1));
138 		else
139 #endif
140 			return SYSCTL_OUT(req, 0, sizeof(long) * CPUSTATES * (mp_maxid + 1));
141 	}
142 	for (error = 0, c = 0; error == 0 && c <= mp_maxid; c++) {
143 		if (!CPU_ABSENT(c)) {
144 			pcpu = pcpu_find(c);
145 			cp_time = pcpu->pc_cp_time;
146 		} else {
147 			cp_time = empty;
148 		}
149 #ifdef SCTL_MASK32
150 		if (req->flags & SCTL_MASK32) {
151 			for (i = 0; i < CPUSTATES; i++)
152 				cp_time32[i] = (unsigned int)cp_time[i];
153 			error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32));
154 		} else
155 #endif
156 			error = SYSCTL_OUT(req, cp_time, sizeof(long) * CPUSTATES);
157 	}
158 	return error;
159 }
160 
161 SYSCTL_PROC(_kern, OID_AUTO, cp_times, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE,
162     0,0, sysctl_kern_cp_times, "LU", "per-CPU time statistics");
163 
164 #ifdef DEADLKRES
165 static const char *blessed[] = {
166 	"so_snd_sx",
167 	"so_rcv_sx",
168 	NULL
169 };
170 static int slptime_threshold = 1800;
171 static int blktime_threshold = 900;
172 static int sleepfreq = 3;
173 
174 static void
175 deadlkres(void)
176 {
177 	struct proc *p;
178 	struct thread *td;
179 	void *wchan;
180 	int blkticks, i, slpticks, slptype, tryl, tticks;
181 
182 	tryl = 0;
183 	for (;;) {
184 		blkticks = blktime_threshold * hz;
185 		slpticks = slptime_threshold * hz;
186 
187 		/*
188 		 * Avoid to sleep on the sx_lock in order to avoid a possible
189 		 * priority inversion problem leading to starvation.
190 		 * If the lock can't be held after 100 tries, panic.
191 		 */
192 		if (!sx_try_slock(&allproc_lock)) {
193 			if (tryl > 100)
194 		panic("%s: possible deadlock detected on allproc_lock\n",
195 				    __func__);
196 			tryl++;
197 			pause("allproc_lock deadlkres", sleepfreq * hz);
198 			continue;
199 		}
200 		tryl = 0;
201 		FOREACH_PROC_IN_SYSTEM(p) {
202 			PROC_LOCK(p);
203 			FOREACH_THREAD_IN_PROC(p, td) {
204 				thread_lock(td);
205 				if (TD_ON_LOCK(td)) {
206 
207 					/*
208 					 * The thread should be blocked on a
209 					 * turnstile, simply check if the
210 					 * turnstile channel is in good state.
211 					 */
212 					MPASS(td->td_blocked != NULL);
213 
214 					/* Handle ticks wrap-up. */
215 					if (ticks < td->td_blktick)
216 						continue;
217 					tticks = ticks - td->td_blktick;
218 					thread_unlock(td);
219 					if (tticks > blkticks) {
220 
221 						/*
222 						 * Accordingly with provided
223 						 * thresholds, this thread is
224 						 * stuck for too long on a
225 						 * turnstile.
226 						 */
227 						PROC_UNLOCK(p);
228 						sx_sunlock(&allproc_lock);
229 	panic("%s: possible deadlock detected for %p, blocked for %d ticks\n",
230 						    __func__, td, tticks);
231 					}
232 				} else if (TD_IS_SLEEPING(td)) {
233 
234 					/* Handle ticks wrap-up. */
235 					if (ticks < td->td_blktick)
236 						continue;
237 
238 					/*
239 					 * Check if the thread is sleeping on a
240 					 * lock, otherwise skip the check.
241 					 * Drop the thread lock in order to
242 					 * avoid a LOR with the sleepqueue
243 					 * spinlock.
244 					 */
245 					wchan = td->td_wchan;
246 					tticks = ticks - td->td_slptick;
247 					thread_unlock(td);
248 					slptype = sleepq_type(wchan);
249 					if ((slptype == SLEEPQ_SX ||
250 					    slptype == SLEEPQ_LK) &&
251 					    tticks > slpticks) {
252 
253 						/*
254 						 * Accordingly with provided
255 						 * thresholds, this thread is
256 						 * stuck for too long on a
257 						 * sleepqueue.
258 						 * However, being on a
259 						 * sleepqueue, we might still
260 						 * check for the blessed
261 						 * list.
262 						 */
263 						tryl = 0;
264 						for (i = 0; blessed[i] != NULL;
265 						    i++) {
266 							if (!strcmp(blessed[i],
267 							    td->td_wmesg)) {
268 								tryl = 1;
269 								break;
270 							}
271 						}
272 						if (tryl != 0) {
273 							tryl = 0;
274 							continue;
275 						}
276 						PROC_UNLOCK(p);
277 						sx_sunlock(&allproc_lock);
278 	panic("%s: possible deadlock detected for %p, blocked for %d ticks\n",
279 						    __func__, td, tticks);
280 					}
281 				} else
282 					thread_unlock(td);
283 			}
284 			PROC_UNLOCK(p);
285 		}
286 		sx_sunlock(&allproc_lock);
287 
288 		/* Sleep for sleepfreq seconds. */
289 		pause("deadlkres", sleepfreq * hz);
290 	}
291 }
292 
293 static struct kthread_desc deadlkres_kd = {
294 	"deadlkres",
295 	deadlkres,
296 	(struct thread **)NULL
297 };
298 
299 SYSINIT(deadlkres, SI_SUB_CLOCKS, SI_ORDER_ANY, kthread_start, &deadlkres_kd);
300 
301 SYSCTL_NODE(_debug, OID_AUTO, deadlkres, CTLFLAG_RW, 0, "Deadlock resolver");
302 SYSCTL_INT(_debug_deadlkres, OID_AUTO, slptime_threshold, CTLFLAG_RW,
303     &slptime_threshold, 0,
304     "Number of seconds within is valid to sleep on a sleepqueue");
305 SYSCTL_INT(_debug_deadlkres, OID_AUTO, blktime_threshold, CTLFLAG_RW,
306     &blktime_threshold, 0,
307     "Number of seconds within is valid to block on a turnstile");
308 SYSCTL_INT(_debug_deadlkres, OID_AUTO, sleepfreq, CTLFLAG_RW, &sleepfreq, 0,
309     "Number of seconds between any deadlock resolver thread run");
310 #endif	/* DEADLKRES */
311 
312 void
313 read_cpu_time(long *cp_time)
314 {
315 	struct pcpu *pc;
316 	int i, j;
317 
318 	/* Sum up global cp_time[]. */
319 	bzero(cp_time, sizeof(long) * CPUSTATES);
320 	for (i = 0; i <= mp_maxid; i++) {
321 		if (CPU_ABSENT(i))
322 			continue;
323 		pc = pcpu_find(i);
324 		for (j = 0; j < CPUSTATES; j++)
325 			cp_time[j] += pc->pc_cp_time[j];
326 	}
327 }
328 
329 #ifdef SW_WATCHDOG
330 #include <sys/watchdog.h>
331 
332 static int watchdog_ticks;
333 static int watchdog_enabled;
334 static void watchdog_fire(void);
335 static void watchdog_config(void *, u_int, int *);
336 #endif /* SW_WATCHDOG */
337 
338 /*
339  * Clock handling routines.
340  *
341  * This code is written to operate with two timers that run independently of
342  * each other.
343  *
344  * The main timer, running hz times per second, is used to trigger interval
345  * timers, timeouts and rescheduling as needed.
346  *
347  * The second timer handles kernel and user profiling,
348  * and does resource use estimation.  If the second timer is programmable,
349  * it is randomized to avoid aliasing between the two clocks.  For example,
350  * the randomization prevents an adversary from always giving up the cpu
351  * just before its quantum expires.  Otherwise, it would never accumulate
352  * cpu ticks.  The mean frequency of the second timer is stathz.
353  *
354  * If no second timer exists, stathz will be zero; in this case we drive
355  * profiling and statistics off the main clock.  This WILL NOT be accurate;
356  * do not do it unless absolutely necessary.
357  *
358  * The statistics clock may (or may not) be run at a higher rate while
359  * profiling.  This profile clock runs at profhz.  We require that profhz
360  * be an integral multiple of stathz.
361  *
362  * If the statistics clock is running fast, it must be divided by the ratio
363  * profhz/stathz for statistics.  (For profiling, every tick counts.)
364  *
365  * Time-of-day is maintained using a "timecounter", which may or may
366  * not be related to the hardware generating the above mentioned
367  * interrupts.
368  */
369 
370 int	stathz;
371 int	profhz;
372 int	profprocs;
373 int	ticks;
374 int	psratio;
375 
376 /*
377  * Initialize clock frequencies and start both clocks running.
378  */
379 /* ARGSUSED*/
380 static void
381 initclocks(dummy)
382 	void *dummy;
383 {
384 	register int i;
385 
386 	/*
387 	 * Set divisors to 1 (normal case) and let the machine-specific
388 	 * code do its bit.
389 	 */
390 	mtx_init(&time_lock, "time lock", NULL, MTX_SPIN);
391 	cpu_initclocks();
392 
393 	/*
394 	 * Compute profhz/stathz, and fix profhz if needed.
395 	 */
396 	i = stathz ? stathz : hz;
397 	if (profhz == 0)
398 		profhz = i;
399 	psratio = profhz / i;
400 #ifdef SW_WATCHDOG
401 	EVENTHANDLER_REGISTER(watchdog_list, watchdog_config, NULL, 0);
402 #endif
403 }
404 
405 /*
406  * Each time the real-time timer fires, this function is called on all CPUs.
407  * Note that hardclock() calls hardclock_cpu() for the boot CPU, so only
408  * the other CPUs in the system need to call this function.
409  */
410 void
411 hardclock_cpu(int usermode)
412 {
413 	struct pstats *pstats;
414 	struct thread *td = curthread;
415 	struct proc *p = td->td_proc;
416 	int flags;
417 
418 	/*
419 	 * Run current process's virtual and profile time, as needed.
420 	 */
421 	pstats = p->p_stats;
422 	flags = 0;
423 	if (usermode &&
424 	    timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) {
425 		PROC_SLOCK(p);
426 		if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0)
427 			flags |= TDF_ALRMPEND | TDF_ASTPENDING;
428 		PROC_SUNLOCK(p);
429 	}
430 	if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) {
431 		PROC_SLOCK(p);
432 		if (itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0)
433 			flags |= TDF_PROFPEND | TDF_ASTPENDING;
434 		PROC_SUNLOCK(p);
435 	}
436 	thread_lock(td);
437 	sched_tick();
438 	td->td_flags |= flags;
439 	thread_unlock(td);
440 
441 #ifdef	HWPMC_HOOKS
442 	if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid)))
443 		PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL);
444 #endif
445 	callout_tick();
446 }
447 
448 /*
449  * The real-time timer, interrupting hz times per second.
450  */
451 void
452 hardclock(int usermode, uintfptr_t pc)
453 {
454 
455 	atomic_add_int((volatile int *)&ticks, 1);
456 	hardclock_cpu(usermode);
457 	tc_ticktock();
458 	/*
459 	 * If no separate statistics clock is available, run it from here.
460 	 *
461 	 * XXX: this only works for UP
462 	 */
463 	if (stathz == 0) {
464 		profclock(usermode, pc);
465 		statclock(usermode);
466 	}
467 #ifdef DEVICE_POLLING
468 	hardclock_device_poll();	/* this is very short and quick */
469 #endif /* DEVICE_POLLING */
470 #ifdef SW_WATCHDOG
471 	if (watchdog_enabled > 0 && --watchdog_ticks <= 0)
472 		watchdog_fire();
473 #endif /* SW_WATCHDOG */
474 }
475 
476 /*
477  * Compute number of ticks in the specified amount of time.
478  */
479 int
480 tvtohz(tv)
481 	struct timeval *tv;
482 {
483 	register unsigned long ticks;
484 	register long sec, usec;
485 
486 	/*
487 	 * If the number of usecs in the whole seconds part of the time
488 	 * difference fits in a long, then the total number of usecs will
489 	 * fit in an unsigned long.  Compute the total and convert it to
490 	 * ticks, rounding up and adding 1 to allow for the current tick
491 	 * to expire.  Rounding also depends on unsigned long arithmetic
492 	 * to avoid overflow.
493 	 *
494 	 * Otherwise, if the number of ticks in the whole seconds part of
495 	 * the time difference fits in a long, then convert the parts to
496 	 * ticks separately and add, using similar rounding methods and
497 	 * overflow avoidance.  This method would work in the previous
498 	 * case but it is slightly slower and assumes that hz is integral.
499 	 *
500 	 * Otherwise, round the time difference down to the maximum
501 	 * representable value.
502 	 *
503 	 * If ints have 32 bits, then the maximum value for any timeout in
504 	 * 10ms ticks is 248 days.
505 	 */
506 	sec = tv->tv_sec;
507 	usec = tv->tv_usec;
508 	if (usec < 0) {
509 		sec--;
510 		usec += 1000000;
511 	}
512 	if (sec < 0) {
513 #ifdef DIAGNOSTIC
514 		if (usec > 0) {
515 			sec++;
516 			usec -= 1000000;
517 		}
518 		printf("tvotohz: negative time difference %ld sec %ld usec\n",
519 		       sec, usec);
520 #endif
521 		ticks = 1;
522 	} else if (sec <= LONG_MAX / 1000000)
523 		ticks = (sec * 1000000 + (unsigned long)usec + (tick - 1))
524 			/ tick + 1;
525 	else if (sec <= LONG_MAX / hz)
526 		ticks = sec * hz
527 			+ ((unsigned long)usec + (tick - 1)) / tick + 1;
528 	else
529 		ticks = LONG_MAX;
530 	if (ticks > INT_MAX)
531 		ticks = INT_MAX;
532 	return ((int)ticks);
533 }
534 
535 /*
536  * Start profiling on a process.
537  *
538  * Kernel profiling passes proc0 which never exits and hence
539  * keeps the profile clock running constantly.
540  */
541 void
542 startprofclock(p)
543 	register struct proc *p;
544 {
545 
546 	PROC_LOCK_ASSERT(p, MA_OWNED);
547 	if (p->p_flag & P_STOPPROF)
548 		return;
549 	if ((p->p_flag & P_PROFIL) == 0) {
550 		p->p_flag |= P_PROFIL;
551 		mtx_lock_spin(&time_lock);
552 		if (++profprocs == 1)
553 			cpu_startprofclock();
554 		mtx_unlock_spin(&time_lock);
555 	}
556 }
557 
558 /*
559  * Stop profiling on a process.
560  */
561 void
562 stopprofclock(p)
563 	register struct proc *p;
564 {
565 
566 	PROC_LOCK_ASSERT(p, MA_OWNED);
567 	if (p->p_flag & P_PROFIL) {
568 		if (p->p_profthreads != 0) {
569 			p->p_flag |= P_STOPPROF;
570 			while (p->p_profthreads != 0)
571 				msleep(&p->p_profthreads, &p->p_mtx, PPAUSE,
572 				    "stopprof", 0);
573 			p->p_flag &= ~P_STOPPROF;
574 		}
575 		if ((p->p_flag & P_PROFIL) == 0)
576 			return;
577 		p->p_flag &= ~P_PROFIL;
578 		mtx_lock_spin(&time_lock);
579 		if (--profprocs == 0)
580 			cpu_stopprofclock();
581 		mtx_unlock_spin(&time_lock);
582 	}
583 }
584 
585 /*
586  * Statistics clock.  Updates rusage information and calls the scheduler
587  * to adjust priorities of the active thread.
588  *
589  * This should be called by all active processors.
590  */
591 void
592 statclock(int usermode)
593 {
594 	struct rusage *ru;
595 	struct vmspace *vm;
596 	struct thread *td;
597 	struct proc *p;
598 	long rss;
599 	long *cp_time;
600 
601 	td = curthread;
602 	p = td->td_proc;
603 
604 	cp_time = (long *)PCPU_PTR(cp_time);
605 	if (usermode) {
606 		/*
607 		 * Charge the time as appropriate.
608 		 */
609 		td->td_uticks++;
610 		if (p->p_nice > NZERO)
611 			cp_time[CP_NICE]++;
612 		else
613 			cp_time[CP_USER]++;
614 	} else {
615 		/*
616 		 * Came from kernel mode, so we were:
617 		 * - handling an interrupt,
618 		 * - doing syscall or trap work on behalf of the current
619 		 *   user process, or
620 		 * - spinning in the idle loop.
621 		 * Whichever it is, charge the time as appropriate.
622 		 * Note that we charge interrupts to the current process,
623 		 * regardless of whether they are ``for'' that process,
624 		 * so that we know how much of its real time was spent
625 		 * in ``non-process'' (i.e., interrupt) work.
626 		 */
627 		if ((td->td_pflags & TDP_ITHREAD) ||
628 		    td->td_intr_nesting_level >= 2) {
629 			td->td_iticks++;
630 			cp_time[CP_INTR]++;
631 		} else {
632 			td->td_pticks++;
633 			td->td_sticks++;
634 			if (!TD_IS_IDLETHREAD(td))
635 				cp_time[CP_SYS]++;
636 			else
637 				cp_time[CP_IDLE]++;
638 		}
639 	}
640 
641 	/* Update resource usage integrals and maximums. */
642 	MPASS(p->p_vmspace != NULL);
643 	vm = p->p_vmspace;
644 	ru = &td->td_ru;
645 	ru->ru_ixrss += pgtok(vm->vm_tsize);
646 	ru->ru_idrss += pgtok(vm->vm_dsize);
647 	ru->ru_isrss += pgtok(vm->vm_ssize);
648 	rss = pgtok(vmspace_resident_count(vm));
649 	if (ru->ru_maxrss < rss)
650 		ru->ru_maxrss = rss;
651 	KTR_POINT2(KTR_SCHED, "thread", sched_tdname(td), "statclock",
652 	    "prio:%d", td->td_priority, "stathz:%d", (stathz)?stathz:hz);
653 	thread_lock_flags(td, MTX_QUIET);
654 	sched_clock(td);
655 	thread_unlock(td);
656 }
657 
658 void
659 profclock(int usermode, uintfptr_t pc)
660 {
661 	struct thread *td;
662 #ifdef GPROF
663 	struct gmonparam *g;
664 	uintfptr_t i;
665 #endif
666 
667 	td = curthread;
668 	if (usermode) {
669 		/*
670 		 * Came from user mode; CPU was in user state.
671 		 * If this process is being profiled, record the tick.
672 		 * if there is no related user location yet, don't
673 		 * bother trying to count it.
674 		 */
675 		if (td->td_proc->p_flag & P_PROFIL)
676 			addupc_intr(td, pc, 1);
677 	}
678 #ifdef GPROF
679 	else {
680 		/*
681 		 * Kernel statistics are just like addupc_intr, only easier.
682 		 */
683 		g = &_gmonparam;
684 		if (g->state == GMON_PROF_ON && pc >= g->lowpc) {
685 			i = PC_TO_I(g, pc);
686 			if (i < g->textsize) {
687 				KCOUNT(g, i)++;
688 			}
689 		}
690 	}
691 #endif
692 }
693 
694 /*
695  * Return information about system clocks.
696  */
697 static int
698 sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS)
699 {
700 	struct clockinfo clkinfo;
701 	/*
702 	 * Construct clockinfo structure.
703 	 */
704 	bzero(&clkinfo, sizeof(clkinfo));
705 	clkinfo.hz = hz;
706 	clkinfo.tick = tick;
707 	clkinfo.profhz = profhz;
708 	clkinfo.stathz = stathz ? stathz : hz;
709 	return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req));
710 }
711 
712 SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate,
713 	CTLTYPE_STRUCT|CTLFLAG_RD|CTLFLAG_MPSAFE,
714 	0, 0, sysctl_kern_clockrate, "S,clockinfo",
715 	"Rate and period of various kernel clocks");
716 
717 #ifdef SW_WATCHDOG
718 
719 static void
720 watchdog_config(void *unused __unused, u_int cmd, int *error)
721 {
722 	u_int u;
723 
724 	u = cmd & WD_INTERVAL;
725 	if (u >= WD_TO_1SEC) {
726 		watchdog_ticks = (1 << (u - WD_TO_1SEC)) * hz;
727 		watchdog_enabled = 1;
728 		*error = 0;
729 	} else {
730 		watchdog_enabled = 0;
731 	}
732 }
733 
734 /*
735  * Handle a watchdog timeout by dumping interrupt information and
736  * then either dropping to DDB or panicking.
737  */
738 static void
739 watchdog_fire(void)
740 {
741 	int nintr;
742 	u_int64_t inttotal;
743 	u_long *curintr;
744 	char *curname;
745 
746 	curintr = intrcnt;
747 	curname = intrnames;
748 	inttotal = 0;
749 	nintr = eintrcnt - intrcnt;
750 
751 	printf("interrupt                   total\n");
752 	while (--nintr >= 0) {
753 		if (*curintr)
754 			printf("%-12s %20lu\n", curname, *curintr);
755 		curname += strlen(curname) + 1;
756 		inttotal += *curintr++;
757 	}
758 	printf("Total        %20ju\n", (uintmax_t)inttotal);
759 
760 #if defined(KDB) && !defined(KDB_UNATTENDED)
761 	kdb_backtrace();
762 	kdb_enter(KDB_WHY_WATCHDOG, "watchdog timeout");
763 #else
764 	panic("watchdog timeout");
765 #endif
766 }
767 
768 #endif /* SW_WATCHDOG */
769