xref: /freebsd/sys/kern/kern_clock.c (revision aa64588d28258aef88cc33b8043112e8856948d0)
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 	"getblk",
167 	"so_snd_sx",
168 	"so_rcv_sx",
169 	NULL
170 };
171 static int slptime_threshold = 1800;
172 static int blktime_threshold = 900;
173 static int sleepfreq = 3;
174 
175 static void
176 deadlkres(void)
177 {
178 	struct proc *p;
179 	struct thread *td;
180 	void *wchan;
181 	int blkticks, i, slpticks, slptype, tryl, tticks;
182 
183 	tryl = 0;
184 	for (;;) {
185 		blkticks = blktime_threshold * hz;
186 		slpticks = slptime_threshold * hz;
187 
188 		/*
189 		 * Avoid to sleep on the sx_lock in order to avoid a possible
190 		 * priority inversion problem leading to starvation.
191 		 * If the lock can't be held after 100 tries, panic.
192 		 */
193 		if (!sx_try_slock(&allproc_lock)) {
194 			if (tryl > 100)
195 		panic("%s: possible deadlock detected on allproc_lock\n",
196 				    __func__);
197 			tryl++;
198 			pause("allproc_lock deadlkres", sleepfreq * hz);
199 			continue;
200 		}
201 		tryl = 0;
202 		FOREACH_PROC_IN_SYSTEM(p) {
203 			PROC_LOCK(p);
204 			FOREACH_THREAD_IN_PROC(p, td) {
205 				thread_lock(td);
206 				if (TD_ON_LOCK(td)) {
207 
208 					/*
209 					 * The thread should be blocked on a
210 					 * turnstile, simply check if the
211 					 * turnstile channel is in good state.
212 					 */
213 					MPASS(td->td_blocked != NULL);
214 
215 					/* Handle ticks wrap-up. */
216 					if (ticks < td->td_blktick)
217 						continue;
218 					tticks = ticks - td->td_blktick;
219 					thread_unlock(td);
220 					if (tticks > blkticks) {
221 
222 						/*
223 						 * Accordingly with provided
224 						 * thresholds, this thread is
225 						 * stuck for too long on a
226 						 * turnstile.
227 						 */
228 						PROC_UNLOCK(p);
229 						sx_sunlock(&allproc_lock);
230 	panic("%s: possible deadlock detected for %p, blocked for %d ticks\n",
231 						    __func__, td, tticks);
232 					}
233 				} else if (TD_IS_SLEEPING(td)) {
234 
235 					/* Handle ticks wrap-up. */
236 					if (ticks < td->td_blktick)
237 						continue;
238 
239 					/*
240 					 * Check if the thread is sleeping on a
241 					 * lock, otherwise skip the check.
242 					 * Drop the thread lock in order to
243 					 * avoid a LOR with the sleepqueue
244 					 * spinlock.
245 					 */
246 					wchan = td->td_wchan;
247 					tticks = ticks - td->td_slptick;
248 					thread_unlock(td);
249 					slptype = sleepq_type(wchan);
250 					if ((slptype == SLEEPQ_SX ||
251 					    slptype == SLEEPQ_LK) &&
252 					    tticks > slpticks) {
253 
254 						/*
255 						 * Accordingly with provided
256 						 * thresholds, this thread is
257 						 * stuck for too long on a
258 						 * sleepqueue.
259 						 * However, being on a
260 						 * sleepqueue, we might still
261 						 * check for the blessed
262 						 * list.
263 						 */
264 						tryl = 0;
265 						for (i = 0; blessed[i] != NULL;
266 						    i++) {
267 							if (!strcmp(blessed[i],
268 							    td->td_wmesg)) {
269 								tryl = 1;
270 								break;
271 							}
272 						}
273 						if (tryl != 0) {
274 							tryl = 0;
275 							continue;
276 						}
277 						PROC_UNLOCK(p);
278 						sx_sunlock(&allproc_lock);
279 	panic("%s: possible deadlock detected for %p, blocked for %d ticks\n",
280 						    __func__, td, tticks);
281 					}
282 				} else
283 					thread_unlock(td);
284 			}
285 			PROC_UNLOCK(p);
286 		}
287 		sx_sunlock(&allproc_lock);
288 
289 		/* Sleep for sleepfreq seconds. */
290 		pause("deadlkres", sleepfreq * hz);
291 	}
292 }
293 
294 static struct kthread_desc deadlkres_kd = {
295 	"deadlkres",
296 	deadlkres,
297 	(struct thread **)NULL
298 };
299 
300 SYSINIT(deadlkres, SI_SUB_CLOCKS, SI_ORDER_ANY, kthread_start, &deadlkres_kd);
301 
302 SYSCTL_NODE(_debug, OID_AUTO, deadlkres, CTLFLAG_RW, 0, "Deadlock resolver");
303 SYSCTL_INT(_debug_deadlkres, OID_AUTO, slptime_threshold, CTLFLAG_RW,
304     &slptime_threshold, 0,
305     "Number of seconds within is valid to sleep on a sleepqueue");
306 SYSCTL_INT(_debug_deadlkres, OID_AUTO, blktime_threshold, CTLFLAG_RW,
307     &blktime_threshold, 0,
308     "Number of seconds within is valid to block on a turnstile");
309 SYSCTL_INT(_debug_deadlkres, OID_AUTO, sleepfreq, CTLFLAG_RW, &sleepfreq, 0,
310     "Number of seconds between any deadlock resolver thread run");
311 #endif	/* DEADLKRES */
312 
313 void
314 read_cpu_time(long *cp_time)
315 {
316 	struct pcpu *pc;
317 	int i, j;
318 
319 	/* Sum up global cp_time[]. */
320 	bzero(cp_time, sizeof(long) * CPUSTATES);
321 	for (i = 0; i <= mp_maxid; i++) {
322 		if (CPU_ABSENT(i))
323 			continue;
324 		pc = pcpu_find(i);
325 		for (j = 0; j < CPUSTATES; j++)
326 			cp_time[j] += pc->pc_cp_time[j];
327 	}
328 }
329 
330 #ifdef SW_WATCHDOG
331 #include <sys/watchdog.h>
332 
333 static int watchdog_ticks;
334 static int watchdog_enabled;
335 static void watchdog_fire(void);
336 static void watchdog_config(void *, u_int, int *);
337 #endif /* SW_WATCHDOG */
338 
339 /*
340  * Clock handling routines.
341  *
342  * This code is written to operate with two timers that run independently of
343  * each other.
344  *
345  * The main timer, running hz times per second, is used to trigger interval
346  * timers, timeouts and rescheduling as needed.
347  *
348  * The second timer handles kernel and user profiling,
349  * and does resource use estimation.  If the second timer is programmable,
350  * it is randomized to avoid aliasing between the two clocks.  For example,
351  * the randomization prevents an adversary from always giving up the cpu
352  * just before its quantum expires.  Otherwise, it would never accumulate
353  * cpu ticks.  The mean frequency of the second timer is stathz.
354  *
355  * If no second timer exists, stathz will be zero; in this case we drive
356  * profiling and statistics off the main clock.  This WILL NOT be accurate;
357  * do not do it unless absolutely necessary.
358  *
359  * The statistics clock may (or may not) be run at a higher rate while
360  * profiling.  This profile clock runs at profhz.  We require that profhz
361  * be an integral multiple of stathz.
362  *
363  * If the statistics clock is running fast, it must be divided by the ratio
364  * profhz/stathz for statistics.  (For profiling, every tick counts.)
365  *
366  * Time-of-day is maintained using a "timecounter", which may or may
367  * not be related to the hardware generating the above mentioned
368  * interrupts.
369  */
370 
371 int	stathz;
372 int	profhz;
373 int	profprocs;
374 int	ticks;
375 int	psratio;
376 
377 int	timer1hz;
378 int	timer2hz;
379 static DPCPU_DEFINE(u_int, hard_cnt);
380 static DPCPU_DEFINE(u_int, stat_cnt);
381 static DPCPU_DEFINE(u_int, prof_cnt);
382 
383 /*
384  * Initialize clock frequencies and start both clocks running.
385  */
386 /* ARGSUSED*/
387 static void
388 initclocks(dummy)
389 	void *dummy;
390 {
391 	register int i;
392 
393 	/*
394 	 * Set divisors to 1 (normal case) and let the machine-specific
395 	 * code do its bit.
396 	 */
397 	mtx_init(&time_lock, "time lock", NULL, MTX_SPIN);
398 	cpu_initclocks();
399 
400 	/*
401 	 * Compute profhz/stathz, and fix profhz if needed.
402 	 */
403 	i = stathz ? stathz : hz;
404 	if (profhz == 0)
405 		profhz = i;
406 	psratio = profhz / i;
407 #ifdef SW_WATCHDOG
408 	EVENTHANDLER_REGISTER(watchdog_list, watchdog_config, NULL, 0);
409 #endif
410 }
411 
412 void
413 timer1clock(int usermode, uintfptr_t pc)
414 {
415 	u_int *cnt;
416 
417 	cnt = DPCPU_PTR(hard_cnt);
418 	*cnt += hz;
419 	if (*cnt >= timer1hz) {
420 		*cnt -= timer1hz;
421 		if (*cnt >= timer1hz)
422 			*cnt = 0;
423 		if (PCPU_GET(cpuid) == 0)
424 			hardclock(usermode, pc);
425 		else
426 			hardclock_cpu(usermode);
427 	}
428 	if (timer2hz == 0)
429 		timer2clock(usermode, pc);
430 }
431 
432 void
433 timer2clock(int usermode, uintfptr_t pc)
434 {
435 	u_int *cnt;
436 	int t2hz = timer2hz ? timer2hz : timer1hz;
437 
438 	cnt = DPCPU_PTR(stat_cnt);
439 	*cnt += stathz;
440 	if (*cnt >= t2hz) {
441 		*cnt -= t2hz;
442 		if (*cnt >= t2hz)
443 			*cnt = 0;
444 		statclock(usermode);
445 	}
446 	if (profprocs == 0)
447 		return;
448 	cnt = DPCPU_PTR(prof_cnt);
449 	*cnt += profhz;
450 	if (*cnt >= t2hz) {
451 		*cnt -= t2hz;
452 		if (*cnt >= t2hz)
453 			*cnt = 0;
454 			profclock(usermode, pc);
455 	}
456 }
457 
458 /*
459  * Each time the real-time timer fires, this function is called on all CPUs.
460  * Note that hardclock() calls hardclock_cpu() for the boot CPU, so only
461  * the other CPUs in the system need to call this function.
462  */
463 void
464 hardclock_cpu(int usermode)
465 {
466 	struct pstats *pstats;
467 	struct thread *td = curthread;
468 	struct proc *p = td->td_proc;
469 	int flags;
470 
471 	/*
472 	 * Run current process's virtual and profile time, as needed.
473 	 */
474 	pstats = p->p_stats;
475 	flags = 0;
476 	if (usermode &&
477 	    timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) {
478 		PROC_SLOCK(p);
479 		if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0)
480 			flags |= TDF_ALRMPEND | TDF_ASTPENDING;
481 		PROC_SUNLOCK(p);
482 	}
483 	if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) {
484 		PROC_SLOCK(p);
485 		if (itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0)
486 			flags |= TDF_PROFPEND | TDF_ASTPENDING;
487 		PROC_SUNLOCK(p);
488 	}
489 	thread_lock(td);
490 	sched_tick();
491 	td->td_flags |= flags;
492 	thread_unlock(td);
493 
494 #ifdef	HWPMC_HOOKS
495 	if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid)))
496 		PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL);
497 #endif
498 	callout_tick();
499 }
500 
501 /*
502  * The real-time timer, interrupting hz times per second.
503  */
504 void
505 hardclock(int usermode, uintfptr_t pc)
506 {
507 
508 	atomic_add_int((volatile int *)&ticks, 1);
509 	hardclock_cpu(usermode);
510 	tc_ticktock();
511 	/*
512 	 * If no separate statistics clock is available, run it from here.
513 	 *
514 	 * XXX: this only works for UP
515 	 */
516 	if (stathz == 0) {
517 		profclock(usermode, pc);
518 		statclock(usermode);
519 	}
520 #ifdef DEVICE_POLLING
521 	hardclock_device_poll();	/* this is very short and quick */
522 #endif /* DEVICE_POLLING */
523 #ifdef SW_WATCHDOG
524 	if (watchdog_enabled > 0 && --watchdog_ticks <= 0)
525 		watchdog_fire();
526 #endif /* SW_WATCHDOG */
527 }
528 
529 /*
530  * Compute number of ticks in the specified amount of time.
531  */
532 int
533 tvtohz(tv)
534 	struct timeval *tv;
535 {
536 	register unsigned long ticks;
537 	register long sec, usec;
538 
539 	/*
540 	 * If the number of usecs in the whole seconds part of the time
541 	 * difference fits in a long, then the total number of usecs will
542 	 * fit in an unsigned long.  Compute the total and convert it to
543 	 * ticks, rounding up and adding 1 to allow for the current tick
544 	 * to expire.  Rounding also depends on unsigned long arithmetic
545 	 * to avoid overflow.
546 	 *
547 	 * Otherwise, if the number of ticks in the whole seconds part of
548 	 * the time difference fits in a long, then convert the parts to
549 	 * ticks separately and add, using similar rounding methods and
550 	 * overflow avoidance.  This method would work in the previous
551 	 * case but it is slightly slower and assumes that hz is integral.
552 	 *
553 	 * Otherwise, round the time difference down to the maximum
554 	 * representable value.
555 	 *
556 	 * If ints have 32 bits, then the maximum value for any timeout in
557 	 * 10ms ticks is 248 days.
558 	 */
559 	sec = tv->tv_sec;
560 	usec = tv->tv_usec;
561 	if (usec < 0) {
562 		sec--;
563 		usec += 1000000;
564 	}
565 	if (sec < 0) {
566 #ifdef DIAGNOSTIC
567 		if (usec > 0) {
568 			sec++;
569 			usec -= 1000000;
570 		}
571 		printf("tvotohz: negative time difference %ld sec %ld usec\n",
572 		       sec, usec);
573 #endif
574 		ticks = 1;
575 	} else if (sec <= LONG_MAX / 1000000)
576 		ticks = (sec * 1000000 + (unsigned long)usec + (tick - 1))
577 			/ tick + 1;
578 	else if (sec <= LONG_MAX / hz)
579 		ticks = sec * hz
580 			+ ((unsigned long)usec + (tick - 1)) / tick + 1;
581 	else
582 		ticks = LONG_MAX;
583 	if (ticks > INT_MAX)
584 		ticks = INT_MAX;
585 	return ((int)ticks);
586 }
587 
588 /*
589  * Start profiling on a process.
590  *
591  * Kernel profiling passes proc0 which never exits and hence
592  * keeps the profile clock running constantly.
593  */
594 void
595 startprofclock(p)
596 	register struct proc *p;
597 {
598 
599 	PROC_LOCK_ASSERT(p, MA_OWNED);
600 	if (p->p_flag & P_STOPPROF)
601 		return;
602 	if ((p->p_flag & P_PROFIL) == 0) {
603 		p->p_flag |= P_PROFIL;
604 		mtx_lock_spin(&time_lock);
605 		if (++profprocs == 1)
606 			cpu_startprofclock();
607 		mtx_unlock_spin(&time_lock);
608 	}
609 }
610 
611 /*
612  * Stop profiling on a process.
613  */
614 void
615 stopprofclock(p)
616 	register struct proc *p;
617 {
618 
619 	PROC_LOCK_ASSERT(p, MA_OWNED);
620 	if (p->p_flag & P_PROFIL) {
621 		if (p->p_profthreads != 0) {
622 			p->p_flag |= P_STOPPROF;
623 			while (p->p_profthreads != 0)
624 				msleep(&p->p_profthreads, &p->p_mtx, PPAUSE,
625 				    "stopprof", 0);
626 			p->p_flag &= ~P_STOPPROF;
627 		}
628 		if ((p->p_flag & P_PROFIL) == 0)
629 			return;
630 		p->p_flag &= ~P_PROFIL;
631 		mtx_lock_spin(&time_lock);
632 		if (--profprocs == 0)
633 			cpu_stopprofclock();
634 		mtx_unlock_spin(&time_lock);
635 	}
636 }
637 
638 /*
639  * Statistics clock.  Updates rusage information and calls the scheduler
640  * to adjust priorities of the active thread.
641  *
642  * This should be called by all active processors.
643  */
644 void
645 statclock(int usermode)
646 {
647 	struct rusage *ru;
648 	struct vmspace *vm;
649 	struct thread *td;
650 	struct proc *p;
651 	long rss;
652 	long *cp_time;
653 
654 	td = curthread;
655 	p = td->td_proc;
656 
657 	cp_time = (long *)PCPU_PTR(cp_time);
658 	if (usermode) {
659 		/*
660 		 * Charge the time as appropriate.
661 		 */
662 		td->td_uticks++;
663 		if (p->p_nice > NZERO)
664 			cp_time[CP_NICE]++;
665 		else
666 			cp_time[CP_USER]++;
667 	} else {
668 		/*
669 		 * Came from kernel mode, so we were:
670 		 * - handling an interrupt,
671 		 * - doing syscall or trap work on behalf of the current
672 		 *   user process, or
673 		 * - spinning in the idle loop.
674 		 * Whichever it is, charge the time as appropriate.
675 		 * Note that we charge interrupts to the current process,
676 		 * regardless of whether they are ``for'' that process,
677 		 * so that we know how much of its real time was spent
678 		 * in ``non-process'' (i.e., interrupt) work.
679 		 */
680 		if ((td->td_pflags & TDP_ITHREAD) ||
681 		    td->td_intr_nesting_level >= 2) {
682 			td->td_iticks++;
683 			cp_time[CP_INTR]++;
684 		} else {
685 			td->td_pticks++;
686 			td->td_sticks++;
687 			if (!TD_IS_IDLETHREAD(td))
688 				cp_time[CP_SYS]++;
689 			else
690 				cp_time[CP_IDLE]++;
691 		}
692 	}
693 
694 	/* Update resource usage integrals and maximums. */
695 	MPASS(p->p_vmspace != NULL);
696 	vm = p->p_vmspace;
697 	ru = &td->td_ru;
698 	ru->ru_ixrss += pgtok(vm->vm_tsize);
699 	ru->ru_idrss += pgtok(vm->vm_dsize);
700 	ru->ru_isrss += pgtok(vm->vm_ssize);
701 	rss = pgtok(vmspace_resident_count(vm));
702 	if (ru->ru_maxrss < rss)
703 		ru->ru_maxrss = rss;
704 	KTR_POINT2(KTR_SCHED, "thread", sched_tdname(td), "statclock",
705 	    "prio:%d", td->td_priority, "stathz:%d", (stathz)?stathz:hz);
706 	thread_lock_flags(td, MTX_QUIET);
707 	sched_clock(td);
708 	thread_unlock(td);
709 }
710 
711 void
712 profclock(int usermode, uintfptr_t pc)
713 {
714 	struct thread *td;
715 #ifdef GPROF
716 	struct gmonparam *g;
717 	uintfptr_t i;
718 #endif
719 
720 	td = curthread;
721 	if (usermode) {
722 		/*
723 		 * Came from user mode; CPU was in user state.
724 		 * If this process is being profiled, record the tick.
725 		 * if there is no related user location yet, don't
726 		 * bother trying to count it.
727 		 */
728 		if (td->td_proc->p_flag & P_PROFIL)
729 			addupc_intr(td, pc, 1);
730 	}
731 #ifdef GPROF
732 	else {
733 		/*
734 		 * Kernel statistics are just like addupc_intr, only easier.
735 		 */
736 		g = &_gmonparam;
737 		if (g->state == GMON_PROF_ON && pc >= g->lowpc) {
738 			i = PC_TO_I(g, pc);
739 			if (i < g->textsize) {
740 				KCOUNT(g, i)++;
741 			}
742 		}
743 	}
744 #endif
745 }
746 
747 /*
748  * Return information about system clocks.
749  */
750 static int
751 sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS)
752 {
753 	struct clockinfo clkinfo;
754 	/*
755 	 * Construct clockinfo structure.
756 	 */
757 	bzero(&clkinfo, sizeof(clkinfo));
758 	clkinfo.hz = hz;
759 	clkinfo.tick = tick;
760 	clkinfo.profhz = profhz;
761 	clkinfo.stathz = stathz ? stathz : hz;
762 	return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req));
763 }
764 
765 SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate,
766 	CTLTYPE_STRUCT|CTLFLAG_RD|CTLFLAG_MPSAFE,
767 	0, 0, sysctl_kern_clockrate, "S,clockinfo",
768 	"Rate and period of various kernel clocks");
769 
770 #ifdef SW_WATCHDOG
771 
772 static void
773 watchdog_config(void *unused __unused, u_int cmd, int *error)
774 {
775 	u_int u;
776 
777 	u = cmd & WD_INTERVAL;
778 	if (u >= WD_TO_1SEC) {
779 		watchdog_ticks = (1 << (u - WD_TO_1SEC)) * hz;
780 		watchdog_enabled = 1;
781 		*error = 0;
782 	} else {
783 		watchdog_enabled = 0;
784 	}
785 }
786 
787 /*
788  * Handle a watchdog timeout by dumping interrupt information and
789  * then either dropping to DDB or panicking.
790  */
791 static void
792 watchdog_fire(void)
793 {
794 	int nintr;
795 	u_int64_t inttotal;
796 	u_long *curintr;
797 	char *curname;
798 
799 	curintr = intrcnt;
800 	curname = intrnames;
801 	inttotal = 0;
802 	nintr = eintrcnt - intrcnt;
803 
804 	printf("interrupt                   total\n");
805 	while (--nintr >= 0) {
806 		if (*curintr)
807 			printf("%-12s %20lu\n", curname, *curintr);
808 		curname += strlen(curname) + 1;
809 		inttotal += *curintr++;
810 	}
811 	printf("Total        %20ju\n", (uintmax_t)inttotal);
812 
813 #if defined(KDB) && !defined(KDB_UNATTENDED)
814 	kdb_backtrace();
815 	kdb_enter(KDB_WHY_WATCHDOG, "watchdog timeout");
816 #else
817 	panic("watchdog timeout");
818 #endif
819 }
820 
821 #endif /* SW_WATCHDOG */
822