xref: /freebsd/sys/kern/kern_clock.c (revision eb69d1f144a6fcc765d1b9d44a5ae8082353e70b)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)kern_clock.c	8.5 (Berkeley) 1/21/94
37  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include "opt_kdb.h"
43 #include "opt_device_polling.h"
44 #include "opt_hwpmc_hooks.h"
45 #include "opt_ntp.h"
46 #include "opt_watchdog.h"
47 
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/callout.h>
51 #include <sys/kdb.h>
52 #include <sys/kernel.h>
53 #include <sys/kthread.h>
54 #include <sys/ktr.h>
55 #include <sys/lock.h>
56 #include <sys/mutex.h>
57 #include <sys/proc.h>
58 #include <sys/resource.h>
59 #include <sys/resourcevar.h>
60 #include <sys/sched.h>
61 #include <sys/sdt.h>
62 #include <sys/signalvar.h>
63 #include <sys/sleepqueue.h>
64 #include <sys/smp.h>
65 #include <vm/vm.h>
66 #include <vm/pmap.h>
67 #include <vm/vm_map.h>
68 #include <sys/sysctl.h>
69 #include <sys/bus.h>
70 #include <sys/interrupt.h>
71 #include <sys/limits.h>
72 #include <sys/timetc.h>
73 
74 #ifdef GPROF
75 #include <sys/gmon.h>
76 #endif
77 
78 #ifdef HWPMC_HOOKS
79 #include <sys/pmckern.h>
80 PMC_SOFT_DEFINE( , , clock, hard);
81 PMC_SOFT_DEFINE( , , clock, stat);
82 PMC_SOFT_DEFINE_EX( , , clock, prof, \
83     cpu_startprofclock, cpu_stopprofclock);
84 #endif
85 
86 #ifdef DEVICE_POLLING
87 extern void hardclock_device_poll(void);
88 #endif /* DEVICE_POLLING */
89 
90 static void initclocks(void *dummy);
91 SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL);
92 
93 /* Spin-lock protecting profiling statistics. */
94 static struct mtx time_lock;
95 
96 SDT_PROVIDER_DECLARE(sched);
97 SDT_PROBE_DEFINE2(sched, , , tick, "struct thread *", "struct proc *");
98 
99 static int
100 sysctl_kern_cp_time(SYSCTL_HANDLER_ARGS)
101 {
102 	int error;
103 	long cp_time[CPUSTATES];
104 #ifdef SCTL_MASK32
105 	int i;
106 	unsigned int cp_time32[CPUSTATES];
107 #endif
108 
109 	read_cpu_time(cp_time);
110 #ifdef SCTL_MASK32
111 	if (req->flags & SCTL_MASK32) {
112 		if (!req->oldptr)
113 			return SYSCTL_OUT(req, 0, sizeof(cp_time32));
114 		for (i = 0; i < CPUSTATES; i++)
115 			cp_time32[i] = (unsigned int)cp_time[i];
116 		error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32));
117 	} else
118 #endif
119 	{
120 		if (!req->oldptr)
121 			return SYSCTL_OUT(req, 0, sizeof(cp_time));
122 		error = SYSCTL_OUT(req, cp_time, sizeof(cp_time));
123 	}
124 	return error;
125 }
126 
127 SYSCTL_PROC(_kern, OID_AUTO, cp_time, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE,
128     0,0, sysctl_kern_cp_time, "LU", "CPU time statistics");
129 
130 static long empty[CPUSTATES];
131 
132 static int
133 sysctl_kern_cp_times(SYSCTL_HANDLER_ARGS)
134 {
135 	struct pcpu *pcpu;
136 	int error;
137 	int c;
138 	long *cp_time;
139 #ifdef SCTL_MASK32
140 	unsigned int cp_time32[CPUSTATES];
141 	int i;
142 #endif
143 
144 	if (!req->oldptr) {
145 #ifdef SCTL_MASK32
146 		if (req->flags & SCTL_MASK32)
147 			return SYSCTL_OUT(req, 0, sizeof(cp_time32) * (mp_maxid + 1));
148 		else
149 #endif
150 			return SYSCTL_OUT(req, 0, sizeof(long) * CPUSTATES * (mp_maxid + 1));
151 	}
152 	for (error = 0, c = 0; error == 0 && c <= mp_maxid; c++) {
153 		if (!CPU_ABSENT(c)) {
154 			pcpu = pcpu_find(c);
155 			cp_time = pcpu->pc_cp_time;
156 		} else {
157 			cp_time = empty;
158 		}
159 #ifdef SCTL_MASK32
160 		if (req->flags & SCTL_MASK32) {
161 			for (i = 0; i < CPUSTATES; i++)
162 				cp_time32[i] = (unsigned int)cp_time[i];
163 			error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32));
164 		} else
165 #endif
166 			error = SYSCTL_OUT(req, cp_time, sizeof(long) * CPUSTATES);
167 	}
168 	return error;
169 }
170 
171 SYSCTL_PROC(_kern, OID_AUTO, cp_times, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE,
172     0,0, sysctl_kern_cp_times, "LU", "per-CPU time statistics");
173 
174 #ifdef DEADLKRES
175 static const char *blessed[] = {
176 	"getblk",
177 	"so_snd_sx",
178 	"so_rcv_sx",
179 	NULL
180 };
181 static int slptime_threshold = 1800;
182 static int blktime_threshold = 900;
183 static int sleepfreq = 3;
184 
185 static void
186 deadlkres(void)
187 {
188 	struct proc *p;
189 	struct thread *td;
190 	void *wchan;
191 	int blkticks, i, slpticks, slptype, tryl, tticks;
192 
193 	tryl = 0;
194 	for (;;) {
195 		blkticks = blktime_threshold * hz;
196 		slpticks = slptime_threshold * hz;
197 
198 		/*
199 		 * Avoid to sleep on the sx_lock in order to avoid a possible
200 		 * priority inversion problem leading to starvation.
201 		 * If the lock can't be held after 100 tries, panic.
202 		 */
203 		if (!sx_try_slock(&allproc_lock)) {
204 			if (tryl > 100)
205 		panic("%s: possible deadlock detected on allproc_lock\n",
206 				    __func__);
207 			tryl++;
208 			pause("allproc", sleepfreq * hz);
209 			continue;
210 		}
211 		tryl = 0;
212 		FOREACH_PROC_IN_SYSTEM(p) {
213 			PROC_LOCK(p);
214 			if (p->p_state == PRS_NEW) {
215 				PROC_UNLOCK(p);
216 				continue;
217 			}
218 			FOREACH_THREAD_IN_PROC(p, td) {
219 
220 				thread_lock(td);
221 				if (TD_ON_LOCK(td)) {
222 
223 					/*
224 					 * The thread should be blocked on a
225 					 * turnstile, simply check if the
226 					 * turnstile channel is in good state.
227 					 */
228 					MPASS(td->td_blocked != NULL);
229 
230 					tticks = ticks - td->td_blktick;
231 					thread_unlock(td);
232 					if (tticks > blkticks) {
233 
234 						/*
235 						 * Accordingly with provided
236 						 * thresholds, this thread is
237 						 * stuck for too long on a
238 						 * turnstile.
239 						 */
240 						PROC_UNLOCK(p);
241 						sx_sunlock(&allproc_lock);
242 	panic("%s: possible deadlock detected for %p, blocked for %d ticks\n",
243 						    __func__, td, tticks);
244 					}
245 				} else if (TD_IS_SLEEPING(td) &&
246 				    TD_ON_SLEEPQ(td)) {
247 
248 					/*
249 					 * Check if the thread is sleeping on a
250 					 * lock, otherwise skip the check.
251 					 * Drop the thread lock in order to
252 					 * avoid a LOR with the sleepqueue
253 					 * spinlock.
254 					 */
255 					wchan = td->td_wchan;
256 					tticks = ticks - td->td_slptick;
257 					thread_unlock(td);
258 					slptype = sleepq_type(wchan);
259 					if ((slptype == SLEEPQ_SX ||
260 					    slptype == SLEEPQ_LK) &&
261 					    tticks > slpticks) {
262 
263 						/*
264 						 * Accordingly with provided
265 						 * thresholds, this thread is
266 						 * stuck for too long on a
267 						 * sleepqueue.
268 						 * However, being on a
269 						 * sleepqueue, we might still
270 						 * check for the blessed
271 						 * list.
272 						 */
273 						tryl = 0;
274 						for (i = 0; blessed[i] != NULL;
275 						    i++) {
276 							if (!strcmp(blessed[i],
277 							    td->td_wmesg)) {
278 								tryl = 1;
279 								break;
280 							}
281 						}
282 						if (tryl != 0) {
283 							tryl = 0;
284 							continue;
285 						}
286 						PROC_UNLOCK(p);
287 						sx_sunlock(&allproc_lock);
288 	panic("%s: possible deadlock detected for %p, blocked for %d ticks\n",
289 						    __func__, td, tticks);
290 					}
291 				} else
292 					thread_unlock(td);
293 			}
294 			PROC_UNLOCK(p);
295 		}
296 		sx_sunlock(&allproc_lock);
297 
298 		/* Sleep for sleepfreq seconds. */
299 		pause("-", sleepfreq * hz);
300 	}
301 }
302 
303 static struct kthread_desc deadlkres_kd = {
304 	"deadlkres",
305 	deadlkres,
306 	(struct thread **)NULL
307 };
308 
309 SYSINIT(deadlkres, SI_SUB_CLOCKS, SI_ORDER_ANY, kthread_start, &deadlkres_kd);
310 
311 static SYSCTL_NODE(_debug, OID_AUTO, deadlkres, CTLFLAG_RW, 0,
312     "Deadlock resolver");
313 SYSCTL_INT(_debug_deadlkres, OID_AUTO, slptime_threshold, CTLFLAG_RW,
314     &slptime_threshold, 0,
315     "Number of seconds within is valid to sleep on a sleepqueue");
316 SYSCTL_INT(_debug_deadlkres, OID_AUTO, blktime_threshold, CTLFLAG_RW,
317     &blktime_threshold, 0,
318     "Number of seconds within is valid to block on a turnstile");
319 SYSCTL_INT(_debug_deadlkres, OID_AUTO, sleepfreq, CTLFLAG_RW, &sleepfreq, 0,
320     "Number of seconds between any deadlock resolver thread run");
321 #endif	/* DEADLKRES */
322 
323 void
324 read_cpu_time(long *cp_time)
325 {
326 	struct pcpu *pc;
327 	int i, j;
328 
329 	/* Sum up global cp_time[]. */
330 	bzero(cp_time, sizeof(long) * CPUSTATES);
331 	CPU_FOREACH(i) {
332 		pc = pcpu_find(i);
333 		for (j = 0; j < CPUSTATES; j++)
334 			cp_time[j] += pc->pc_cp_time[j];
335 	}
336 }
337 
338 #include <sys/watchdog.h>
339 
340 static int watchdog_ticks;
341 static int watchdog_enabled;
342 static void watchdog_fire(void);
343 static void watchdog_config(void *, u_int, int *);
344 
345 static void
346 watchdog_attach(void)
347 {
348 	EVENTHANDLER_REGISTER(watchdog_list, watchdog_config, NULL, 0);
349 }
350 
351 /*
352  * Clock handling routines.
353  *
354  * This code is written to operate with two timers that run independently of
355  * each other.
356  *
357  * The main timer, running hz times per second, is used to trigger interval
358  * timers, timeouts and rescheduling as needed.
359  *
360  * The second timer handles kernel and user profiling,
361  * and does resource use estimation.  If the second timer is programmable,
362  * it is randomized to avoid aliasing between the two clocks.  For example,
363  * the randomization prevents an adversary from always giving up the cpu
364  * just before its quantum expires.  Otherwise, it would never accumulate
365  * cpu ticks.  The mean frequency of the second timer is stathz.
366  *
367  * If no second timer exists, stathz will be zero; in this case we drive
368  * profiling and statistics off the main clock.  This WILL NOT be accurate;
369  * do not do it unless absolutely necessary.
370  *
371  * The statistics clock may (or may not) be run at a higher rate while
372  * profiling.  This profile clock runs at profhz.  We require that profhz
373  * be an integral multiple of stathz.
374  *
375  * If the statistics clock is running fast, it must be divided by the ratio
376  * profhz/stathz for statistics.  (For profiling, every tick counts.)
377  *
378  * Time-of-day is maintained using a "timecounter", which may or may
379  * not be related to the hardware generating the above mentioned
380  * interrupts.
381  */
382 
383 int	stathz;
384 int	profhz;
385 int	profprocs;
386 volatile int	ticks;
387 int	psratio;
388 
389 static DPCPU_DEFINE(int, pcputicks);	/* Per-CPU version of ticks. */
390 #ifdef DEVICE_POLLING
391 static int devpoll_run = 0;
392 #endif
393 
394 /*
395  * Initialize clock frequencies and start both clocks running.
396  */
397 /* ARGSUSED*/
398 static void
399 initclocks(void *dummy)
400 {
401 	int i;
402 
403 	/*
404 	 * Set divisors to 1 (normal case) and let the machine-specific
405 	 * code do its bit.
406 	 */
407 	mtx_init(&time_lock, "time lock", NULL, MTX_DEF);
408 	cpu_initclocks();
409 
410 	/*
411 	 * Compute profhz/stathz, and fix profhz if needed.
412 	 */
413 	i = stathz ? stathz : hz;
414 	if (profhz == 0)
415 		profhz = i;
416 	psratio = profhz / i;
417 
418 #ifdef SW_WATCHDOG
419 	/* Enable hardclock watchdog now, even if a hardware watchdog exists. */
420 	watchdog_attach();
421 #else
422 	/* Volunteer to run a software watchdog. */
423 	if (wdog_software_attach == NULL)
424 		wdog_software_attach = watchdog_attach;
425 #endif
426 }
427 
428 /*
429  * Each time the real-time timer fires, this function is called on all CPUs.
430  * Note that hardclock() calls hardclock_cpu() for the boot CPU, so only
431  * the other CPUs in the system need to call this function.
432  */
433 void
434 hardclock_cpu(int usermode)
435 {
436 	struct pstats *pstats;
437 	struct thread *td = curthread;
438 	struct proc *p = td->td_proc;
439 	int flags;
440 
441 	/*
442 	 * Run current process's virtual and profile time, as needed.
443 	 */
444 	pstats = p->p_stats;
445 	flags = 0;
446 	if (usermode &&
447 	    timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) {
448 		PROC_ITIMLOCK(p);
449 		if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0)
450 			flags |= TDF_ALRMPEND | TDF_ASTPENDING;
451 		PROC_ITIMUNLOCK(p);
452 	}
453 	if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) {
454 		PROC_ITIMLOCK(p);
455 		if (itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0)
456 			flags |= TDF_PROFPEND | TDF_ASTPENDING;
457 		PROC_ITIMUNLOCK(p);
458 	}
459 	thread_lock(td);
460 	td->td_flags |= flags;
461 	thread_unlock(td);
462 
463 #ifdef HWPMC_HOOKS
464 	if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid)))
465 		PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL);
466 	if (td->td_intr_frame != NULL)
467 		PMC_SOFT_CALL_TF( , , clock, hard, td->td_intr_frame);
468 #endif
469 	callout_process(sbinuptime());
470 }
471 
472 /*
473  * The real-time timer, interrupting hz times per second.
474  */
475 void
476 hardclock(int usermode, uintfptr_t pc)
477 {
478 
479 	atomic_add_int(&ticks, 1);
480 	hardclock_cpu(usermode);
481 	tc_ticktock(1);
482 	cpu_tick_calibration();
483 	/*
484 	 * If no separate statistics clock is available, run it from here.
485 	 *
486 	 * XXX: this only works for UP
487 	 */
488 	if (stathz == 0) {
489 		profclock(usermode, pc);
490 		statclock(usermode);
491 	}
492 #ifdef DEVICE_POLLING
493 	hardclock_device_poll();	/* this is very short and quick */
494 #endif /* DEVICE_POLLING */
495 	if (watchdog_enabled > 0 && --watchdog_ticks <= 0)
496 		watchdog_fire();
497 }
498 
499 void
500 hardclock_cnt(int cnt, int usermode)
501 {
502 	struct pstats *pstats;
503 	struct thread *td = curthread;
504 	struct proc *p = td->td_proc;
505 	int *t = DPCPU_PTR(pcputicks);
506 	int flags, global, newticks;
507 	int i;
508 
509 	/*
510 	 * Update per-CPU and possibly global ticks values.
511 	 */
512 	*t += cnt;
513 	do {
514 		global = ticks;
515 		newticks = *t - global;
516 		if (newticks <= 0) {
517 			if (newticks < -1)
518 				*t = global - 1;
519 			newticks = 0;
520 			break;
521 		}
522 	} while (!atomic_cmpset_int(&ticks, global, *t));
523 
524 	/*
525 	 * Run current process's virtual and profile time, as needed.
526 	 */
527 	pstats = p->p_stats;
528 	flags = 0;
529 	if (usermode &&
530 	    timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) {
531 		PROC_ITIMLOCK(p);
532 		if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL],
533 		    tick * cnt) == 0)
534 			flags |= TDF_ALRMPEND | TDF_ASTPENDING;
535 		PROC_ITIMUNLOCK(p);
536 	}
537 	if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) {
538 		PROC_ITIMLOCK(p);
539 		if (itimerdecr(&pstats->p_timer[ITIMER_PROF],
540 		    tick * cnt) == 0)
541 			flags |= TDF_PROFPEND | TDF_ASTPENDING;
542 		PROC_ITIMUNLOCK(p);
543 	}
544 	if (flags != 0) {
545 		thread_lock(td);
546 		td->td_flags |= flags;
547 		thread_unlock(td);
548 	}
549 
550 #ifdef	HWPMC_HOOKS
551 	if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid)))
552 		PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL);
553 	if (td->td_intr_frame != NULL)
554 		PMC_SOFT_CALL_TF( , , clock, hard, td->td_intr_frame);
555 #endif
556 	/* We are in charge to handle this tick duty. */
557 	if (newticks > 0) {
558 		tc_ticktock(newticks);
559 #ifdef DEVICE_POLLING
560 		/* Dangerous and no need to call these things concurrently. */
561 		if (atomic_cmpset_acq_int(&devpoll_run, 0, 1)) {
562 			/* This is very short and quick. */
563 			hardclock_device_poll();
564 			atomic_store_rel_int(&devpoll_run, 0);
565 		}
566 #endif /* DEVICE_POLLING */
567 		if (watchdog_enabled > 0) {
568 			i = atomic_fetchadd_int(&watchdog_ticks, -newticks);
569 			if (i > 0 && i <= newticks)
570 				watchdog_fire();
571 		}
572 	}
573 	if (curcpu == CPU_FIRST())
574 		cpu_tick_calibration();
575 }
576 
577 void
578 hardclock_sync(int cpu)
579 {
580 	int *t;
581 	KASSERT(!CPU_ABSENT(cpu), ("Absent CPU %d", cpu));
582 	t = DPCPU_ID_PTR(cpu, pcputicks);
583 
584 	*t = ticks;
585 }
586 
587 /*
588  * Compute number of ticks in the specified amount of time.
589  */
590 int
591 tvtohz(struct timeval *tv)
592 {
593 	unsigned long ticks;
594 	long sec, usec;
595 
596 	/*
597 	 * If the number of usecs in the whole seconds part of the time
598 	 * difference fits in a long, then the total number of usecs will
599 	 * fit in an unsigned long.  Compute the total and convert it to
600 	 * ticks, rounding up and adding 1 to allow for the current tick
601 	 * to expire.  Rounding also depends on unsigned long arithmetic
602 	 * to avoid overflow.
603 	 *
604 	 * Otherwise, if the number of ticks in the whole seconds part of
605 	 * the time difference fits in a long, then convert the parts to
606 	 * ticks separately and add, using similar rounding methods and
607 	 * overflow avoidance.  This method would work in the previous
608 	 * case but it is slightly slower and assumes that hz is integral.
609 	 *
610 	 * Otherwise, round the time difference down to the maximum
611 	 * representable value.
612 	 *
613 	 * If ints have 32 bits, then the maximum value for any timeout in
614 	 * 10ms ticks is 248 days.
615 	 */
616 	sec = tv->tv_sec;
617 	usec = tv->tv_usec;
618 	if (usec < 0) {
619 		sec--;
620 		usec += 1000000;
621 	}
622 	if (sec < 0) {
623 #ifdef DIAGNOSTIC
624 		if (usec > 0) {
625 			sec++;
626 			usec -= 1000000;
627 		}
628 		printf("tvotohz: negative time difference %ld sec %ld usec\n",
629 		       sec, usec);
630 #endif
631 		ticks = 1;
632 	} else if (sec <= LONG_MAX / 1000000)
633 		ticks = howmany(sec * 1000000 + (unsigned long)usec, tick) + 1;
634 	else if (sec <= LONG_MAX / hz)
635 		ticks = sec * hz
636 			+ howmany((unsigned long)usec, tick) + 1;
637 	else
638 		ticks = LONG_MAX;
639 	if (ticks > INT_MAX)
640 		ticks = INT_MAX;
641 	return ((int)ticks);
642 }
643 
644 /*
645  * Start profiling on a process.
646  *
647  * Kernel profiling passes proc0 which never exits and hence
648  * keeps the profile clock running constantly.
649  */
650 void
651 startprofclock(struct proc *p)
652 {
653 
654 	PROC_LOCK_ASSERT(p, MA_OWNED);
655 	if (p->p_flag & P_STOPPROF)
656 		return;
657 	if ((p->p_flag & P_PROFIL) == 0) {
658 		p->p_flag |= P_PROFIL;
659 		mtx_lock(&time_lock);
660 		if (++profprocs == 1)
661 			cpu_startprofclock();
662 		mtx_unlock(&time_lock);
663 	}
664 }
665 
666 /*
667  * Stop profiling on a process.
668  */
669 void
670 stopprofclock(struct proc *p)
671 {
672 
673 	PROC_LOCK_ASSERT(p, MA_OWNED);
674 	if (p->p_flag & P_PROFIL) {
675 		if (p->p_profthreads != 0) {
676 			while (p->p_profthreads != 0) {
677 				p->p_flag |= P_STOPPROF;
678 				msleep(&p->p_profthreads, &p->p_mtx, PPAUSE,
679 				    "stopprof", 0);
680 			}
681 		}
682 		if ((p->p_flag & P_PROFIL) == 0)
683 			return;
684 		p->p_flag &= ~P_PROFIL;
685 		mtx_lock(&time_lock);
686 		if (--profprocs == 0)
687 			cpu_stopprofclock();
688 		mtx_unlock(&time_lock);
689 	}
690 }
691 
692 /*
693  * Statistics clock.  Updates rusage information and calls the scheduler
694  * to adjust priorities of the active thread.
695  *
696  * This should be called by all active processors.
697  */
698 void
699 statclock(int usermode)
700 {
701 
702 	statclock_cnt(1, usermode);
703 }
704 
705 void
706 statclock_cnt(int cnt, int usermode)
707 {
708 	struct rusage *ru;
709 	struct vmspace *vm;
710 	struct thread *td;
711 	struct proc *p;
712 	long rss;
713 	long *cp_time;
714 
715 	td = curthread;
716 	p = td->td_proc;
717 
718 	cp_time = (long *)PCPU_PTR(cp_time);
719 	if (usermode) {
720 		/*
721 		 * Charge the time as appropriate.
722 		 */
723 		td->td_uticks += cnt;
724 		if (p->p_nice > NZERO)
725 			cp_time[CP_NICE] += cnt;
726 		else
727 			cp_time[CP_USER] += cnt;
728 	} else {
729 		/*
730 		 * Came from kernel mode, so we were:
731 		 * - handling an interrupt,
732 		 * - doing syscall or trap work on behalf of the current
733 		 *   user process, or
734 		 * - spinning in the idle loop.
735 		 * Whichever it is, charge the time as appropriate.
736 		 * Note that we charge interrupts to the current process,
737 		 * regardless of whether they are ``for'' that process,
738 		 * so that we know how much of its real time was spent
739 		 * in ``non-process'' (i.e., interrupt) work.
740 		 */
741 		if ((td->td_pflags & TDP_ITHREAD) ||
742 		    td->td_intr_nesting_level >= 2) {
743 			td->td_iticks += cnt;
744 			cp_time[CP_INTR] += cnt;
745 		} else {
746 			td->td_pticks += cnt;
747 			td->td_sticks += cnt;
748 			if (!TD_IS_IDLETHREAD(td))
749 				cp_time[CP_SYS] += cnt;
750 			else
751 				cp_time[CP_IDLE] += cnt;
752 		}
753 	}
754 
755 	/* Update resource usage integrals and maximums. */
756 	MPASS(p->p_vmspace != NULL);
757 	vm = p->p_vmspace;
758 	ru = &td->td_ru;
759 	ru->ru_ixrss += pgtok(vm->vm_tsize) * cnt;
760 	ru->ru_idrss += pgtok(vm->vm_dsize) * cnt;
761 	ru->ru_isrss += pgtok(vm->vm_ssize) * cnt;
762 	rss = pgtok(vmspace_resident_count(vm));
763 	if (ru->ru_maxrss < rss)
764 		ru->ru_maxrss = rss;
765 	KTR_POINT2(KTR_SCHED, "thread", sched_tdname(td), "statclock",
766 	    "prio:%d", td->td_priority, "stathz:%d", (stathz)?stathz:hz);
767 	SDT_PROBE2(sched, , , tick, td, td->td_proc);
768 	thread_lock_flags(td, MTX_QUIET);
769 	for ( ; cnt > 0; cnt--)
770 		sched_clock(td);
771 	thread_unlock(td);
772 #ifdef HWPMC_HOOKS
773 	if (td->td_intr_frame != NULL)
774 		PMC_SOFT_CALL_TF( , , clock, stat, td->td_intr_frame);
775 #endif
776 }
777 
778 void
779 profclock(int usermode, uintfptr_t pc)
780 {
781 
782 	profclock_cnt(1, usermode, pc);
783 }
784 
785 void
786 profclock_cnt(int cnt, int usermode, uintfptr_t pc)
787 {
788 	struct thread *td;
789 #ifdef GPROF
790 	struct gmonparam *g;
791 	uintfptr_t i;
792 #endif
793 
794 	td = curthread;
795 	if (usermode) {
796 		/*
797 		 * Came from user mode; CPU was in user state.
798 		 * If this process is being profiled, record the tick.
799 		 * if there is no related user location yet, don't
800 		 * bother trying to count it.
801 		 */
802 		if (td->td_proc->p_flag & P_PROFIL)
803 			addupc_intr(td, pc, cnt);
804 	}
805 #ifdef GPROF
806 	else {
807 		/*
808 		 * Kernel statistics are just like addupc_intr, only easier.
809 		 */
810 		g = &_gmonparam;
811 		if (g->state == GMON_PROF_ON && pc >= g->lowpc) {
812 			i = PC_TO_I(g, pc);
813 			if (i < g->textsize) {
814 				KCOUNT(g, i) += cnt;
815 			}
816 		}
817 	}
818 #endif
819 #ifdef HWPMC_HOOKS
820 	if (td->td_intr_frame != NULL)
821 		PMC_SOFT_CALL_TF( , , clock, prof, td->td_intr_frame);
822 #endif
823 }
824 
825 /*
826  * Return information about system clocks.
827  */
828 static int
829 sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS)
830 {
831 	struct clockinfo clkinfo;
832 	/*
833 	 * Construct clockinfo structure.
834 	 */
835 	bzero(&clkinfo, sizeof(clkinfo));
836 	clkinfo.hz = hz;
837 	clkinfo.tick = tick;
838 	clkinfo.profhz = profhz;
839 	clkinfo.stathz = stathz ? stathz : hz;
840 	return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req));
841 }
842 
843 SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate,
844 	CTLTYPE_STRUCT|CTLFLAG_RD|CTLFLAG_MPSAFE,
845 	0, 0, sysctl_kern_clockrate, "S,clockinfo",
846 	"Rate and period of various kernel clocks");
847 
848 static void
849 watchdog_config(void *unused __unused, u_int cmd, int *error)
850 {
851 	u_int u;
852 
853 	u = cmd & WD_INTERVAL;
854 	if (u >= WD_TO_1SEC) {
855 		watchdog_ticks = (1 << (u - WD_TO_1SEC)) * hz;
856 		watchdog_enabled = 1;
857 		*error = 0;
858 	} else {
859 		watchdog_enabled = 0;
860 	}
861 }
862 
863 /*
864  * Handle a watchdog timeout by dumping interrupt information and
865  * then either dropping to DDB or panicking.
866  */
867 static void
868 watchdog_fire(void)
869 {
870 	int nintr;
871 	uint64_t inttotal;
872 	u_long *curintr;
873 	char *curname;
874 
875 	curintr = intrcnt;
876 	curname = intrnames;
877 	inttotal = 0;
878 	nintr = sintrcnt / sizeof(u_long);
879 
880 	printf("interrupt                   total\n");
881 	while (--nintr >= 0) {
882 		if (*curintr)
883 			printf("%-12s %20lu\n", curname, *curintr);
884 		curname += strlen(curname) + 1;
885 		inttotal += *curintr++;
886 	}
887 	printf("Total        %20ju\n", (uintmax_t)inttotal);
888 
889 #if defined(KDB) && !defined(KDB_UNATTENDED)
890 	kdb_backtrace();
891 	kdb_enter(KDB_WHY_WATCHDOG, "watchdog timeout");
892 #else
893 	panic("watchdog timeout");
894 #endif
895 }
896