xref: /freebsd/sys/kern/kern_clock.c (revision 42821a2fc9aa8656e89d0353ece77c4799e940bb)
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", sleepfreq * hz);
199 			continue;
200 		}
201 		tryl = 0;
202 		FOREACH_PROC_IN_SYSTEM(p) {
203 			PROC_LOCK(p);
204 			if (p->p_state == PRS_NEW) {
205 				PROC_UNLOCK(p);
206 				continue;
207 			}
208 			FOREACH_THREAD_IN_PROC(p, td) {
209 
210 				/*
211 				 * Once a thread is found in "interesting"
212 				 * state a possible ticks wrap-up needs to be
213 				 * checked.
214 				 */
215 				thread_lock(td);
216 				if (TD_ON_LOCK(td) && ticks < td->td_blktick) {
217 
218 					/*
219 					 * The thread should be blocked on a
220 					 * turnstile, simply check if the
221 					 * turnstile channel is in good state.
222 					 */
223 					MPASS(td->td_blocked != NULL);
224 
225 					tticks = ticks - td->td_blktick;
226 					thread_unlock(td);
227 					if (tticks > blkticks) {
228 
229 						/*
230 						 * Accordingly with provided
231 						 * thresholds, this thread is
232 						 * stuck for too long on a
233 						 * turnstile.
234 						 */
235 						PROC_UNLOCK(p);
236 						sx_sunlock(&allproc_lock);
237 	panic("%s: possible deadlock detected for %p, blocked for %d ticks\n",
238 						    __func__, td, tticks);
239 					}
240 				} else if (TD_IS_SLEEPING(td) &&
241 				    TD_ON_SLEEPQ(td) &&
242 				    ticks < td->td_blktick) {
243 
244 					/*
245 					 * Check if the thread is sleeping on a
246 					 * lock, otherwise skip the check.
247 					 * Drop the thread lock in order to
248 					 * avoid a LOR with the sleepqueue
249 					 * spinlock.
250 					 */
251 					wchan = td->td_wchan;
252 					tticks = ticks - td->td_slptick;
253 					thread_unlock(td);
254 					slptype = sleepq_type(wchan);
255 					if ((slptype == SLEEPQ_SX ||
256 					    slptype == SLEEPQ_LK) &&
257 					    tticks > slpticks) {
258 
259 						/*
260 						 * Accordingly with provided
261 						 * thresholds, this thread is
262 						 * stuck for too long on a
263 						 * sleepqueue.
264 						 * However, being on a
265 						 * sleepqueue, we might still
266 						 * check for the blessed
267 						 * list.
268 						 */
269 						tryl = 0;
270 						for (i = 0; blessed[i] != NULL;
271 						    i++) {
272 							if (!strcmp(blessed[i],
273 							    td->td_wmesg)) {
274 								tryl = 1;
275 								break;
276 							}
277 						}
278 						if (tryl != 0) {
279 							tryl = 0;
280 							continue;
281 						}
282 						PROC_UNLOCK(p);
283 						sx_sunlock(&allproc_lock);
284 	panic("%s: possible deadlock detected for %p, blocked for %d ticks\n",
285 						    __func__, td, tticks);
286 					}
287 				} else
288 					thread_unlock(td);
289 			}
290 			PROC_UNLOCK(p);
291 		}
292 		sx_sunlock(&allproc_lock);
293 
294 		/* Sleep for sleepfreq seconds. */
295 		pause("-", sleepfreq * hz);
296 	}
297 }
298 
299 static struct kthread_desc deadlkres_kd = {
300 	"deadlkres",
301 	deadlkres,
302 	(struct thread **)NULL
303 };
304 
305 SYSINIT(deadlkres, SI_SUB_CLOCKS, SI_ORDER_ANY, kthread_start, &deadlkres_kd);
306 
307 static SYSCTL_NODE(_debug, OID_AUTO, deadlkres, CTLFLAG_RW, 0,
308     "Deadlock resolver");
309 SYSCTL_INT(_debug_deadlkres, OID_AUTO, slptime_threshold, CTLFLAG_RW,
310     &slptime_threshold, 0,
311     "Number of seconds within is valid to sleep on a sleepqueue");
312 SYSCTL_INT(_debug_deadlkres, OID_AUTO, blktime_threshold, CTLFLAG_RW,
313     &blktime_threshold, 0,
314     "Number of seconds within is valid to block on a turnstile");
315 SYSCTL_INT(_debug_deadlkres, OID_AUTO, sleepfreq, CTLFLAG_RW, &sleepfreq, 0,
316     "Number of seconds between any deadlock resolver thread run");
317 #endif	/* DEADLKRES */
318 
319 void
320 read_cpu_time(long *cp_time)
321 {
322 	struct pcpu *pc;
323 	int i, j;
324 
325 	/* Sum up global cp_time[]. */
326 	bzero(cp_time, sizeof(long) * CPUSTATES);
327 	CPU_FOREACH(i) {
328 		pc = pcpu_find(i);
329 		for (j = 0; j < CPUSTATES; j++)
330 			cp_time[j] += pc->pc_cp_time[j];
331 	}
332 }
333 
334 #ifdef SW_WATCHDOG
335 #include <sys/watchdog.h>
336 
337 static int watchdog_ticks;
338 static int watchdog_enabled;
339 static void watchdog_fire(void);
340 static void watchdog_config(void *, u_int, int *);
341 #endif /* SW_WATCHDOG */
342 
343 /*
344  * Clock handling routines.
345  *
346  * This code is written to operate with two timers that run independently of
347  * each other.
348  *
349  * The main timer, running hz times per second, is used to trigger interval
350  * timers, timeouts and rescheduling as needed.
351  *
352  * The second timer handles kernel and user profiling,
353  * and does resource use estimation.  If the second timer is programmable,
354  * it is randomized to avoid aliasing between the two clocks.  For example,
355  * the randomization prevents an adversary from always giving up the cpu
356  * just before its quantum expires.  Otherwise, it would never accumulate
357  * cpu ticks.  The mean frequency of the second timer is stathz.
358  *
359  * If no second timer exists, stathz will be zero; in this case we drive
360  * profiling and statistics off the main clock.  This WILL NOT be accurate;
361  * do not do it unless absolutely necessary.
362  *
363  * The statistics clock may (or may not) be run at a higher rate while
364  * profiling.  This profile clock runs at profhz.  We require that profhz
365  * be an integral multiple of stathz.
366  *
367  * If the statistics clock is running fast, it must be divided by the ratio
368  * profhz/stathz for statistics.  (For profiling, every tick counts.)
369  *
370  * Time-of-day is maintained using a "timecounter", which may or may
371  * not be related to the hardware generating the above mentioned
372  * interrupts.
373  */
374 
375 int	stathz;
376 int	profhz;
377 int	profprocs;
378 int	ticks;
379 int	psratio;
380 
381 static DPCPU_DEFINE(int, pcputicks);	/* Per-CPU version of ticks. */
382 static int global_hardclock_run = 0;
383 
384 /*
385  * Initialize clock frequencies and start both clocks running.
386  */
387 /* ARGSUSED*/
388 static void
389 initclocks(dummy)
390 	void *dummy;
391 {
392 	register int i;
393 
394 	/*
395 	 * Set divisors to 1 (normal case) and let the machine-specific
396 	 * code do its bit.
397 	 */
398 	mtx_init(&time_lock, "time lock", NULL, MTX_DEF);
399 	cpu_initclocks();
400 
401 	/*
402 	 * Compute profhz/stathz, and fix profhz if needed.
403 	 */
404 	i = stathz ? stathz : hz;
405 	if (profhz == 0)
406 		profhz = i;
407 	psratio = profhz / i;
408 #ifdef SW_WATCHDOG
409 	EVENTHANDLER_REGISTER(watchdog_list, watchdog_config, NULL, 0);
410 #endif
411 }
412 
413 /*
414  * Each time the real-time timer fires, this function is called on all CPUs.
415  * Note that hardclock() calls hardclock_cpu() for the boot CPU, so only
416  * the other CPUs in the system need to call this function.
417  */
418 void
419 hardclock_cpu(int usermode)
420 {
421 	struct pstats *pstats;
422 	struct thread *td = curthread;
423 	struct proc *p = td->td_proc;
424 	int flags;
425 
426 	/*
427 	 * Run current process's virtual and profile time, as needed.
428 	 */
429 	pstats = p->p_stats;
430 	flags = 0;
431 	if (usermode &&
432 	    timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) {
433 		PROC_SLOCK(p);
434 		if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0)
435 			flags |= TDF_ALRMPEND | TDF_ASTPENDING;
436 		PROC_SUNLOCK(p);
437 	}
438 	if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) {
439 		PROC_SLOCK(p);
440 		if (itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0)
441 			flags |= TDF_PROFPEND | TDF_ASTPENDING;
442 		PROC_SUNLOCK(p);
443 	}
444 	thread_lock(td);
445 	sched_tick(1);
446 	td->td_flags |= flags;
447 	thread_unlock(td);
448 
449 #ifdef	HWPMC_HOOKS
450 	if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid)))
451 		PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL);
452 #endif
453 	callout_tick();
454 }
455 
456 /*
457  * The real-time timer, interrupting hz times per second.
458  */
459 void
460 hardclock(int usermode, uintfptr_t pc)
461 {
462 
463 	atomic_add_int((volatile int *)&ticks, 1);
464 	hardclock_cpu(usermode);
465 	tc_ticktock(1);
466 	cpu_tick_calibration();
467 	/*
468 	 * If no separate statistics clock is available, run it from here.
469 	 *
470 	 * XXX: this only works for UP
471 	 */
472 	if (stathz == 0) {
473 		profclock(usermode, pc);
474 		statclock(usermode);
475 	}
476 #ifdef DEVICE_POLLING
477 	hardclock_device_poll();	/* this is very short and quick */
478 #endif /* DEVICE_POLLING */
479 #ifdef SW_WATCHDOG
480 	if (watchdog_enabled > 0 && --watchdog_ticks <= 0)
481 		watchdog_fire();
482 #endif /* SW_WATCHDOG */
483 }
484 
485 void
486 hardclock_anycpu(int cnt, int usermode)
487 {
488 	struct pstats *pstats;
489 	struct thread *td = curthread;
490 	struct proc *p = td->td_proc;
491 	int *t = DPCPU_PTR(pcputicks);
492 	int flags, global, newticks;
493 #ifdef SW_WATCHDOG
494 	int i;
495 #endif /* SW_WATCHDOG */
496 
497 	/*
498 	 * Update per-CPU and possibly global ticks values.
499 	 */
500 	*t += cnt;
501 	do {
502 		global = ticks;
503 		newticks = *t - global;
504 		if (newticks <= 0) {
505 			if (newticks < -1)
506 				*t = global - 1;
507 			newticks = 0;
508 			break;
509 		}
510 	} while (!atomic_cmpset_int(&ticks, global, *t));
511 
512 	/*
513 	 * Run current process's virtual and profile time, as needed.
514 	 */
515 	pstats = p->p_stats;
516 	flags = 0;
517 	if (usermode &&
518 	    timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) {
519 		PROC_SLOCK(p);
520 		if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL],
521 		    tick * cnt) == 0)
522 			flags |= TDF_ALRMPEND | TDF_ASTPENDING;
523 		PROC_SUNLOCK(p);
524 	}
525 	if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) {
526 		PROC_SLOCK(p);
527 		if (itimerdecr(&pstats->p_timer[ITIMER_PROF],
528 		    tick * cnt) == 0)
529 			flags |= TDF_PROFPEND | TDF_ASTPENDING;
530 		PROC_SUNLOCK(p);
531 	}
532 	thread_lock(td);
533 	sched_tick(cnt);
534 	td->td_flags |= flags;
535 	thread_unlock(td);
536 
537 #ifdef	HWPMC_HOOKS
538 	if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid)))
539 		PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL);
540 #endif
541 	callout_tick();
542 	/* We are in charge to handle this tick duty. */
543 	if (newticks > 0) {
544 		/* Dangerous and no need to call these things concurrently. */
545 		if (atomic_cmpset_acq_int(&global_hardclock_run, 0, 1)) {
546 			tc_ticktock(newticks);
547 #ifdef DEVICE_POLLING
548 			/* This is very short and quick. */
549 			hardclock_device_poll();
550 #endif /* DEVICE_POLLING */
551 			atomic_store_rel_int(&global_hardclock_run, 0);
552 		}
553 #ifdef SW_WATCHDOG
554 		if (watchdog_enabled > 0) {
555 			i = atomic_fetchadd_int(&watchdog_ticks, -newticks);
556 			if (i > 0 && i <= newticks)
557 				watchdog_fire();
558 		}
559 #endif /* SW_WATCHDOG */
560 	}
561 	if (curcpu == CPU_FIRST())
562 		cpu_tick_calibration();
563 }
564 
565 void
566 hardclock_sync(int cpu)
567 {
568 	int	*t = DPCPU_ID_PTR(cpu, pcputicks);
569 
570 	*t = ticks;
571 }
572 
573 /*
574  * Compute number of ticks in the specified amount of time.
575  */
576 int
577 tvtohz(tv)
578 	struct timeval *tv;
579 {
580 	register unsigned long ticks;
581 	register long sec, usec;
582 
583 	/*
584 	 * If the number of usecs in the whole seconds part of the time
585 	 * difference fits in a long, then the total number of usecs will
586 	 * fit in an unsigned long.  Compute the total and convert it to
587 	 * ticks, rounding up and adding 1 to allow for the current tick
588 	 * to expire.  Rounding also depends on unsigned long arithmetic
589 	 * to avoid overflow.
590 	 *
591 	 * Otherwise, if the number of ticks in the whole seconds part of
592 	 * the time difference fits in a long, then convert the parts to
593 	 * ticks separately and add, using similar rounding methods and
594 	 * overflow avoidance.  This method would work in the previous
595 	 * case but it is slightly slower and assumes that hz is integral.
596 	 *
597 	 * Otherwise, round the time difference down to the maximum
598 	 * representable value.
599 	 *
600 	 * If ints have 32 bits, then the maximum value for any timeout in
601 	 * 10ms ticks is 248 days.
602 	 */
603 	sec = tv->tv_sec;
604 	usec = tv->tv_usec;
605 	if (usec < 0) {
606 		sec--;
607 		usec += 1000000;
608 	}
609 	if (sec < 0) {
610 #ifdef DIAGNOSTIC
611 		if (usec > 0) {
612 			sec++;
613 			usec -= 1000000;
614 		}
615 		printf("tvotohz: negative time difference %ld sec %ld usec\n",
616 		       sec, usec);
617 #endif
618 		ticks = 1;
619 	} else if (sec <= LONG_MAX / 1000000)
620 		ticks = (sec * 1000000 + (unsigned long)usec + (tick - 1))
621 			/ tick + 1;
622 	else if (sec <= LONG_MAX / hz)
623 		ticks = sec * hz
624 			+ ((unsigned long)usec + (tick - 1)) / tick + 1;
625 	else
626 		ticks = LONG_MAX;
627 	if (ticks > INT_MAX)
628 		ticks = INT_MAX;
629 	return ((int)ticks);
630 }
631 
632 /*
633  * Start profiling on a process.
634  *
635  * Kernel profiling passes proc0 which never exits and hence
636  * keeps the profile clock running constantly.
637  */
638 void
639 startprofclock(p)
640 	register struct proc *p;
641 {
642 
643 	PROC_LOCK_ASSERT(p, MA_OWNED);
644 	if (p->p_flag & P_STOPPROF)
645 		return;
646 	if ((p->p_flag & P_PROFIL) == 0) {
647 		p->p_flag |= P_PROFIL;
648 		mtx_lock(&time_lock);
649 		if (++profprocs == 1)
650 			cpu_startprofclock();
651 		mtx_unlock(&time_lock);
652 	}
653 }
654 
655 /*
656  * Stop profiling on a process.
657  */
658 void
659 stopprofclock(p)
660 	register struct proc *p;
661 {
662 
663 	PROC_LOCK_ASSERT(p, MA_OWNED);
664 	if (p->p_flag & P_PROFIL) {
665 		if (p->p_profthreads != 0) {
666 			p->p_flag |= P_STOPPROF;
667 			while (p->p_profthreads != 0)
668 				msleep(&p->p_profthreads, &p->p_mtx, PPAUSE,
669 				    "stopprof", 0);
670 			p->p_flag &= ~P_STOPPROF;
671 		}
672 		if ((p->p_flag & P_PROFIL) == 0)
673 			return;
674 		p->p_flag &= ~P_PROFIL;
675 		mtx_lock(&time_lock);
676 		if (--profprocs == 0)
677 			cpu_stopprofclock();
678 		mtx_unlock(&time_lock);
679 	}
680 }
681 
682 /*
683  * Statistics clock.  Updates rusage information and calls the scheduler
684  * to adjust priorities of the active thread.
685  *
686  * This should be called by all active processors.
687  */
688 void
689 statclock(int usermode)
690 {
691 	struct rusage *ru;
692 	struct vmspace *vm;
693 	struct thread *td;
694 	struct proc *p;
695 	long rss;
696 	long *cp_time;
697 
698 	td = curthread;
699 	p = td->td_proc;
700 
701 	cp_time = (long *)PCPU_PTR(cp_time);
702 	if (usermode) {
703 		/*
704 		 * Charge the time as appropriate.
705 		 */
706 		td->td_uticks++;
707 		if (p->p_nice > NZERO)
708 			cp_time[CP_NICE]++;
709 		else
710 			cp_time[CP_USER]++;
711 	} else {
712 		/*
713 		 * Came from kernel mode, so we were:
714 		 * - handling an interrupt,
715 		 * - doing syscall or trap work on behalf of the current
716 		 *   user process, or
717 		 * - spinning in the idle loop.
718 		 * Whichever it is, charge the time as appropriate.
719 		 * Note that we charge interrupts to the current process,
720 		 * regardless of whether they are ``for'' that process,
721 		 * so that we know how much of its real time was spent
722 		 * in ``non-process'' (i.e., interrupt) work.
723 		 */
724 		if ((td->td_pflags & TDP_ITHREAD) ||
725 		    td->td_intr_nesting_level >= 2) {
726 			td->td_iticks++;
727 			cp_time[CP_INTR]++;
728 		} else {
729 			td->td_pticks++;
730 			td->td_sticks++;
731 			if (!TD_IS_IDLETHREAD(td))
732 				cp_time[CP_SYS]++;
733 			else
734 				cp_time[CP_IDLE]++;
735 		}
736 	}
737 
738 	/* Update resource usage integrals and maximums. */
739 	MPASS(p->p_vmspace != NULL);
740 	vm = p->p_vmspace;
741 	ru = &td->td_ru;
742 	ru->ru_ixrss += pgtok(vm->vm_tsize);
743 	ru->ru_idrss += pgtok(vm->vm_dsize);
744 	ru->ru_isrss += pgtok(vm->vm_ssize);
745 	rss = pgtok(vmspace_resident_count(vm));
746 	if (ru->ru_maxrss < rss)
747 		ru->ru_maxrss = rss;
748 	KTR_POINT2(KTR_SCHED, "thread", sched_tdname(td), "statclock",
749 	    "prio:%d", td->td_priority, "stathz:%d", (stathz)?stathz:hz);
750 	thread_lock_flags(td, MTX_QUIET);
751 	sched_clock(td);
752 	thread_unlock(td);
753 }
754 
755 void
756 profclock(int usermode, uintfptr_t pc)
757 {
758 	struct thread *td;
759 #ifdef GPROF
760 	struct gmonparam *g;
761 	uintfptr_t i;
762 #endif
763 
764 	td = curthread;
765 	if (usermode) {
766 		/*
767 		 * Came from user mode; CPU was in user state.
768 		 * If this process is being profiled, record the tick.
769 		 * if there is no related user location yet, don't
770 		 * bother trying to count it.
771 		 */
772 		if (td->td_proc->p_flag & P_PROFIL)
773 			addupc_intr(td, pc, 1);
774 	}
775 #ifdef GPROF
776 	else {
777 		/*
778 		 * Kernel statistics are just like addupc_intr, only easier.
779 		 */
780 		g = &_gmonparam;
781 		if (g->state == GMON_PROF_ON && pc >= g->lowpc) {
782 			i = PC_TO_I(g, pc);
783 			if (i < g->textsize) {
784 				KCOUNT(g, i)++;
785 			}
786 		}
787 	}
788 #endif
789 }
790 
791 /*
792  * Return information about system clocks.
793  */
794 static int
795 sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS)
796 {
797 	struct clockinfo clkinfo;
798 	/*
799 	 * Construct clockinfo structure.
800 	 */
801 	bzero(&clkinfo, sizeof(clkinfo));
802 	clkinfo.hz = hz;
803 	clkinfo.tick = tick;
804 	clkinfo.profhz = profhz;
805 	clkinfo.stathz = stathz ? stathz : hz;
806 	return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req));
807 }
808 
809 SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate,
810 	CTLTYPE_STRUCT|CTLFLAG_RD|CTLFLAG_MPSAFE,
811 	0, 0, sysctl_kern_clockrate, "S,clockinfo",
812 	"Rate and period of various kernel clocks");
813 
814 #ifdef SW_WATCHDOG
815 
816 static void
817 watchdog_config(void *unused __unused, u_int cmd, int *error)
818 {
819 	u_int u;
820 
821 	u = cmd & WD_INTERVAL;
822 	if (u >= WD_TO_1SEC) {
823 		watchdog_ticks = (1 << (u - WD_TO_1SEC)) * hz;
824 		watchdog_enabled = 1;
825 		*error = 0;
826 	} else {
827 		watchdog_enabled = 0;
828 	}
829 }
830 
831 /*
832  * Handle a watchdog timeout by dumping interrupt information and
833  * then either dropping to DDB or panicking.
834  */
835 static void
836 watchdog_fire(void)
837 {
838 	int nintr;
839 	uint64_t inttotal;
840 	u_long *curintr;
841 	char *curname;
842 
843 	curintr = intrcnt;
844 	curname = intrnames;
845 	inttotal = 0;
846 	nintr = sintrcnt / sizeof(u_long);
847 
848 	printf("interrupt                   total\n");
849 	while (--nintr >= 0) {
850 		if (*curintr)
851 			printf("%-12s %20lu\n", curname, *curintr);
852 		curname += strlen(curname) + 1;
853 		inttotal += *curintr++;
854 	}
855 	printf("Total        %20ju\n", (uintmax_t)inttotal);
856 
857 #if defined(KDB) && !defined(KDB_UNATTENDED)
858 	kdb_backtrace();
859 	kdb_enter(KDB_WHY_WATCHDOG, "watchdog timeout");
860 #else
861 	panic("watchdog timeout");
862 #endif
863 }
864 
865 #endif /* SW_WATCHDOG */
866