xref: /freebsd/sys/kern/kern_clock.c (revision b2db760808f74bb53c232900091c9da801ebbfcc)
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 
206 				/*
207 				 * Once a thread is found in "interesting"
208 				 * state a possible ticks wrap-up needs to be
209 				 * checked.
210 				 */
211 				thread_lock(td);
212 				if (TD_ON_LOCK(td) && ticks < td->td_blktick) {
213 
214 					/*
215 					 * The thread should be blocked on a
216 					 * turnstile, simply check if the
217 					 * turnstile channel is in good state.
218 					 */
219 					MPASS(td->td_blocked != NULL);
220 
221 					tticks = ticks - td->td_blktick;
222 					thread_unlock(td);
223 					if (tticks > blkticks) {
224 
225 						/*
226 						 * Accordingly with provided
227 						 * thresholds, this thread is
228 						 * stuck for too long on a
229 						 * turnstile.
230 						 */
231 						PROC_UNLOCK(p);
232 						sx_sunlock(&allproc_lock);
233 	panic("%s: possible deadlock detected for %p, blocked for %d ticks\n",
234 						    __func__, td, tticks);
235 					}
236 				} else if (TD_IS_SLEEPING(td) &&
237 				    TD_ON_SLEEPQ(td) &&
238 				    ticks < td->td_blktick) {
239 
240 					/*
241 					 * Check if the thread is sleeping on a
242 					 * lock, otherwise skip the check.
243 					 * Drop the thread lock in order to
244 					 * avoid a LOR with the sleepqueue
245 					 * spinlock.
246 					 */
247 					wchan = td->td_wchan;
248 					tticks = ticks - td->td_slptick;
249 					thread_unlock(td);
250 					slptype = sleepq_type(wchan);
251 					if ((slptype == SLEEPQ_SX ||
252 					    slptype == SLEEPQ_LK) &&
253 					    tticks > slpticks) {
254 
255 						/*
256 						 * Accordingly with provided
257 						 * thresholds, this thread is
258 						 * stuck for too long on a
259 						 * sleepqueue.
260 						 * However, being on a
261 						 * sleepqueue, we might still
262 						 * check for the blessed
263 						 * list.
264 						 */
265 						tryl = 0;
266 						for (i = 0; blessed[i] != NULL;
267 						    i++) {
268 							if (!strcmp(blessed[i],
269 							    td->td_wmesg)) {
270 								tryl = 1;
271 								break;
272 							}
273 						}
274 						if (tryl != 0) {
275 							tryl = 0;
276 							continue;
277 						}
278 						PROC_UNLOCK(p);
279 						sx_sunlock(&allproc_lock);
280 	panic("%s: possible deadlock detected for %p, blocked for %d ticks\n",
281 						    __func__, td, tticks);
282 					}
283 				} else
284 					thread_unlock(td);
285 			}
286 			PROC_UNLOCK(p);
287 		}
288 		sx_sunlock(&allproc_lock);
289 
290 		/* Sleep for sleepfreq seconds. */
291 		pause("deadlkres", sleepfreq * hz);
292 	}
293 }
294 
295 static struct kthread_desc deadlkres_kd = {
296 	"deadlkres",
297 	deadlkres,
298 	(struct thread **)NULL
299 };
300 
301 SYSINIT(deadlkres, SI_SUB_CLOCKS, SI_ORDER_ANY, kthread_start, &deadlkres_kd);
302 
303 SYSCTL_NODE(_debug, OID_AUTO, deadlkres, CTLFLAG_RW, 0, "Deadlock resolver");
304 SYSCTL_INT(_debug_deadlkres, OID_AUTO, slptime_threshold, CTLFLAG_RW,
305     &slptime_threshold, 0,
306     "Number of seconds within is valid to sleep on a sleepqueue");
307 SYSCTL_INT(_debug_deadlkres, OID_AUTO, blktime_threshold, CTLFLAG_RW,
308     &blktime_threshold, 0,
309     "Number of seconds within is valid to block on a turnstile");
310 SYSCTL_INT(_debug_deadlkres, OID_AUTO, sleepfreq, CTLFLAG_RW, &sleepfreq, 0,
311     "Number of seconds between any deadlock resolver thread run");
312 #endif	/* DEADLKRES */
313 
314 void
315 read_cpu_time(long *cp_time)
316 {
317 	struct pcpu *pc;
318 	int i, j;
319 
320 	/* Sum up global cp_time[]. */
321 	bzero(cp_time, sizeof(long) * CPUSTATES);
322 	CPU_FOREACH(i) {
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 int	timer1hz;
377 int	timer2hz;
378 static DPCPU_DEFINE(u_int, hard_cnt);
379 static DPCPU_DEFINE(u_int, stat_cnt);
380 static DPCPU_DEFINE(u_int, prof_cnt);
381 
382 /*
383  * Initialize clock frequencies and start both clocks running.
384  */
385 /* ARGSUSED*/
386 static void
387 initclocks(dummy)
388 	void *dummy;
389 {
390 	register int i;
391 
392 	/*
393 	 * Set divisors to 1 (normal case) and let the machine-specific
394 	 * code do its bit.
395 	 */
396 	mtx_init(&time_lock, "time lock", NULL, MTX_DEF);
397 	cpu_initclocks();
398 
399 	/*
400 	 * Compute profhz/stathz, and fix profhz if needed.
401 	 */
402 	i = stathz ? stathz : hz;
403 	if (profhz == 0)
404 		profhz = i;
405 	psratio = profhz / i;
406 #ifdef SW_WATCHDOG
407 	EVENTHANDLER_REGISTER(watchdog_list, watchdog_config, NULL, 0);
408 #endif
409 }
410 
411 void
412 timer1clock(int usermode, uintfptr_t pc)
413 {
414 	u_int *cnt;
415 
416 	cnt = DPCPU_PTR(hard_cnt);
417 	*cnt += hz;
418 	if (*cnt >= timer1hz) {
419 		*cnt -= timer1hz;
420 		if (*cnt >= timer1hz)
421 			*cnt = 0;
422 		if (PCPU_GET(cpuid) == 0)
423 			hardclock(usermode, pc);
424 		else
425 			hardclock_cpu(usermode);
426 	}
427 	if (timer2hz == 0)
428 		timer2clock(usermode, pc);
429 }
430 
431 void
432 timer2clock(int usermode, uintfptr_t pc)
433 {
434 	u_int *cnt;
435 	int t2hz = timer2hz ? timer2hz : timer1hz;
436 
437 	cnt = DPCPU_PTR(stat_cnt);
438 	*cnt += stathz;
439 	if (*cnt >= t2hz) {
440 		*cnt -= t2hz;
441 		if (*cnt >= t2hz)
442 			*cnt = 0;
443 		statclock(usermode);
444 	}
445 	if (profprocs == 0)
446 		return;
447 	cnt = DPCPU_PTR(prof_cnt);
448 	*cnt += profhz;
449 	if (*cnt >= t2hz) {
450 		*cnt -= t2hz;
451 		if (*cnt >= t2hz)
452 			*cnt = 0;
453 		profclock(usermode, pc);
454 	}
455 }
456 
457 /*
458  * Each time the real-time timer fires, this function is called on all CPUs.
459  * Note that hardclock() calls hardclock_cpu() for the boot CPU, so only
460  * the other CPUs in the system need to call this function.
461  */
462 void
463 hardclock_cpu(int usermode)
464 {
465 	struct pstats *pstats;
466 	struct thread *td = curthread;
467 	struct proc *p = td->td_proc;
468 	int flags;
469 
470 	/*
471 	 * Run current process's virtual and profile time, as needed.
472 	 */
473 	pstats = p->p_stats;
474 	flags = 0;
475 	if (usermode &&
476 	    timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) {
477 		PROC_SLOCK(p);
478 		if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0)
479 			flags |= TDF_ALRMPEND | TDF_ASTPENDING;
480 		PROC_SUNLOCK(p);
481 	}
482 	if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) {
483 		PROC_SLOCK(p);
484 		if (itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0)
485 			flags |= TDF_PROFPEND | TDF_ASTPENDING;
486 		PROC_SUNLOCK(p);
487 	}
488 	thread_lock(td);
489 	sched_tick();
490 	td->td_flags |= flags;
491 	thread_unlock(td);
492 
493 #ifdef	HWPMC_HOOKS
494 	if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid)))
495 		PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL);
496 #endif
497 	callout_tick();
498 }
499 
500 /*
501  * The real-time timer, interrupting hz times per second.
502  */
503 void
504 hardclock(int usermode, uintfptr_t pc)
505 {
506 
507 	atomic_add_int((volatile int *)&ticks, 1);
508 	hardclock_cpu(usermode);
509 	tc_ticktock();
510 	/*
511 	 * If no separate statistics clock is available, run it from here.
512 	 *
513 	 * XXX: this only works for UP
514 	 */
515 	if (stathz == 0) {
516 		profclock(usermode, pc);
517 		statclock(usermode);
518 	}
519 #ifdef DEVICE_POLLING
520 	hardclock_device_poll();	/* this is very short and quick */
521 #endif /* DEVICE_POLLING */
522 #ifdef SW_WATCHDOG
523 	if (watchdog_enabled > 0 && --watchdog_ticks <= 0)
524 		watchdog_fire();
525 #endif /* SW_WATCHDOG */
526 }
527 
528 /*
529  * Compute number of ticks in the specified amount of time.
530  */
531 int
532 tvtohz(tv)
533 	struct timeval *tv;
534 {
535 	register unsigned long ticks;
536 	register long sec, usec;
537 
538 	/*
539 	 * If the number of usecs in the whole seconds part of the time
540 	 * difference fits in a long, then the total number of usecs will
541 	 * fit in an unsigned long.  Compute the total and convert it to
542 	 * ticks, rounding up and adding 1 to allow for the current tick
543 	 * to expire.  Rounding also depends on unsigned long arithmetic
544 	 * to avoid overflow.
545 	 *
546 	 * Otherwise, if the number of ticks in the whole seconds part of
547 	 * the time difference fits in a long, then convert the parts to
548 	 * ticks separately and add, using similar rounding methods and
549 	 * overflow avoidance.  This method would work in the previous
550 	 * case but it is slightly slower and assumes that hz is integral.
551 	 *
552 	 * Otherwise, round the time difference down to the maximum
553 	 * representable value.
554 	 *
555 	 * If ints have 32 bits, then the maximum value for any timeout in
556 	 * 10ms ticks is 248 days.
557 	 */
558 	sec = tv->tv_sec;
559 	usec = tv->tv_usec;
560 	if (usec < 0) {
561 		sec--;
562 		usec += 1000000;
563 	}
564 	if (sec < 0) {
565 #ifdef DIAGNOSTIC
566 		if (usec > 0) {
567 			sec++;
568 			usec -= 1000000;
569 		}
570 		printf("tvotohz: negative time difference %ld sec %ld usec\n",
571 		       sec, usec);
572 #endif
573 		ticks = 1;
574 	} else if (sec <= LONG_MAX / 1000000)
575 		ticks = (sec * 1000000 + (unsigned long)usec + (tick - 1))
576 			/ tick + 1;
577 	else if (sec <= LONG_MAX / hz)
578 		ticks = sec * hz
579 			+ ((unsigned long)usec + (tick - 1)) / tick + 1;
580 	else
581 		ticks = LONG_MAX;
582 	if (ticks > INT_MAX)
583 		ticks = INT_MAX;
584 	return ((int)ticks);
585 }
586 
587 /*
588  * Start profiling on a process.
589  *
590  * Kernel profiling passes proc0 which never exits and hence
591  * keeps the profile clock running constantly.
592  */
593 void
594 startprofclock(p)
595 	register struct proc *p;
596 {
597 
598 	PROC_LOCK_ASSERT(p, MA_OWNED);
599 	if (p->p_flag & P_STOPPROF)
600 		return;
601 	if ((p->p_flag & P_PROFIL) == 0) {
602 		p->p_flag |= P_PROFIL;
603 		mtx_lock(&time_lock);
604 		if (++profprocs == 1)
605 			cpu_startprofclock();
606 		mtx_unlock(&time_lock);
607 	}
608 }
609 
610 /*
611  * Stop profiling on a process.
612  */
613 void
614 stopprofclock(p)
615 	register struct proc *p;
616 {
617 
618 	PROC_LOCK_ASSERT(p, MA_OWNED);
619 	if (p->p_flag & P_PROFIL) {
620 		if (p->p_profthreads != 0) {
621 			p->p_flag |= P_STOPPROF;
622 			while (p->p_profthreads != 0)
623 				msleep(&p->p_profthreads, &p->p_mtx, PPAUSE,
624 				    "stopprof", 0);
625 			p->p_flag &= ~P_STOPPROF;
626 		}
627 		if ((p->p_flag & P_PROFIL) == 0)
628 			return;
629 		p->p_flag &= ~P_PROFIL;
630 		mtx_lock(&time_lock);
631 		if (--profprocs == 0)
632 			cpu_stopprofclock();
633 		mtx_unlock(&time_lock);
634 	}
635 }
636 
637 /*
638  * Statistics clock.  Updates rusage information and calls the scheduler
639  * to adjust priorities of the active thread.
640  *
641  * This should be called by all active processors.
642  */
643 void
644 statclock(int usermode)
645 {
646 	struct rusage *ru;
647 	struct vmspace *vm;
648 	struct thread *td;
649 	struct proc *p;
650 	long rss;
651 	long *cp_time;
652 
653 	td = curthread;
654 	p = td->td_proc;
655 
656 	cp_time = (long *)PCPU_PTR(cp_time);
657 	if (usermode) {
658 		/*
659 		 * Charge the time as appropriate.
660 		 */
661 		td->td_uticks++;
662 		if (p->p_nice > NZERO)
663 			cp_time[CP_NICE]++;
664 		else
665 			cp_time[CP_USER]++;
666 	} else {
667 		/*
668 		 * Came from kernel mode, so we were:
669 		 * - handling an interrupt,
670 		 * - doing syscall or trap work on behalf of the current
671 		 *   user process, or
672 		 * - spinning in the idle loop.
673 		 * Whichever it is, charge the time as appropriate.
674 		 * Note that we charge interrupts to the current process,
675 		 * regardless of whether they are ``for'' that process,
676 		 * so that we know how much of its real time was spent
677 		 * in ``non-process'' (i.e., interrupt) work.
678 		 */
679 		if ((td->td_pflags & TDP_ITHREAD) ||
680 		    td->td_intr_nesting_level >= 2) {
681 			td->td_iticks++;
682 			cp_time[CP_INTR]++;
683 		} else {
684 			td->td_pticks++;
685 			td->td_sticks++;
686 			if (!TD_IS_IDLETHREAD(td))
687 				cp_time[CP_SYS]++;
688 			else
689 				cp_time[CP_IDLE]++;
690 		}
691 	}
692 
693 	/* Update resource usage integrals and maximums. */
694 	MPASS(p->p_vmspace != NULL);
695 	vm = p->p_vmspace;
696 	ru = &td->td_ru;
697 	ru->ru_ixrss += pgtok(vm->vm_tsize);
698 	ru->ru_idrss += pgtok(vm->vm_dsize);
699 	ru->ru_isrss += pgtok(vm->vm_ssize);
700 	rss = pgtok(vmspace_resident_count(vm));
701 	if (ru->ru_maxrss < rss)
702 		ru->ru_maxrss = rss;
703 	KTR_POINT2(KTR_SCHED, "thread", sched_tdname(td), "statclock",
704 	    "prio:%d", td->td_priority, "stathz:%d", (stathz)?stathz:hz);
705 	thread_lock_flags(td, MTX_QUIET);
706 	sched_clock(td);
707 	thread_unlock(td);
708 }
709 
710 void
711 profclock(int usermode, uintfptr_t pc)
712 {
713 	struct thread *td;
714 #ifdef GPROF
715 	struct gmonparam *g;
716 	uintfptr_t i;
717 #endif
718 
719 	td = curthread;
720 	if (usermode) {
721 		/*
722 		 * Came from user mode; CPU was in user state.
723 		 * If this process is being profiled, record the tick.
724 		 * if there is no related user location yet, don't
725 		 * bother trying to count it.
726 		 */
727 		if (td->td_proc->p_flag & P_PROFIL)
728 			addupc_intr(td, pc, 1);
729 	}
730 #ifdef GPROF
731 	else {
732 		/*
733 		 * Kernel statistics are just like addupc_intr, only easier.
734 		 */
735 		g = &_gmonparam;
736 		if (g->state == GMON_PROF_ON && pc >= g->lowpc) {
737 			i = PC_TO_I(g, pc);
738 			if (i < g->textsize) {
739 				KCOUNT(g, i)++;
740 			}
741 		}
742 	}
743 #endif
744 }
745 
746 /*
747  * Return information about system clocks.
748  */
749 static int
750 sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS)
751 {
752 	struct clockinfo clkinfo;
753 	/*
754 	 * Construct clockinfo structure.
755 	 */
756 	bzero(&clkinfo, sizeof(clkinfo));
757 	clkinfo.hz = hz;
758 	clkinfo.tick = tick;
759 	clkinfo.profhz = profhz;
760 	clkinfo.stathz = stathz ? stathz : hz;
761 	return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req));
762 }
763 
764 SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate,
765 	CTLTYPE_STRUCT|CTLFLAG_RD|CTLFLAG_MPSAFE,
766 	0, 0, sysctl_kern_clockrate, "S,clockinfo",
767 	"Rate and period of various kernel clocks");
768 
769 #ifdef SW_WATCHDOG
770 
771 static void
772 watchdog_config(void *unused __unused, u_int cmd, int *error)
773 {
774 	u_int u;
775 
776 	u = cmd & WD_INTERVAL;
777 	if (u >= WD_TO_1SEC) {
778 		watchdog_ticks = (1 << (u - WD_TO_1SEC)) * hz;
779 		watchdog_enabled = 1;
780 		*error = 0;
781 	} else {
782 		watchdog_enabled = 0;
783 	}
784 }
785 
786 /*
787  * Handle a watchdog timeout by dumping interrupt information and
788  * then either dropping to DDB or panicking.
789  */
790 static void
791 watchdog_fire(void)
792 {
793 	int nintr;
794 	uint64_t inttotal;
795 	u_long *curintr;
796 	char *curname;
797 
798 	curintr = intrcnt;
799 	curname = intrnames;
800 	inttotal = 0;
801 	nintr = eintrcnt - intrcnt;
802 
803 	printf("interrupt                   total\n");
804 	while (--nintr >= 0) {
805 		if (*curintr)
806 			printf("%-12s %20lu\n", curname, *curintr);
807 		curname += strlen(curname) + 1;
808 		inttotal += *curintr++;
809 	}
810 	printf("Total        %20ju\n", (uintmax_t)inttotal);
811 
812 #if defined(KDB) && !defined(KDB_UNATTENDED)
813 	kdb_backtrace();
814 	kdb_enter(KDB_WHY_WATCHDOG, "watchdog timeout");
815 #else
816 	panic("watchdog timeout");
817 #endif
818 }
819 
820 #endif /* SW_WATCHDOG */
821