xref: /freebsd/sys/kern/kern_clock.c (revision 6829dae12bb055451fa467da4589c43bd03b1e64)
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/epoch.h>
52 #include <sys/gtaskqueue.h>
53 #include <sys/kdb.h>
54 #include <sys/kernel.h>
55 #include <sys/kthread.h>
56 #include <sys/ktr.h>
57 #include <sys/lock.h>
58 #include <sys/mutex.h>
59 #include <sys/proc.h>
60 #include <sys/resource.h>
61 #include <sys/resourcevar.h>
62 #include <sys/sched.h>
63 #include <sys/sdt.h>
64 #include <sys/signalvar.h>
65 #include <sys/sleepqueue.h>
66 #include <sys/smp.h>
67 #include <vm/vm.h>
68 #include <vm/pmap.h>
69 #include <vm/vm_map.h>
70 #include <sys/sysctl.h>
71 #include <sys/bus.h>
72 #include <sys/interrupt.h>
73 #include <sys/limits.h>
74 #include <sys/timetc.h>
75 
76 #ifdef GPROF
77 #include <sys/gmon.h>
78 #endif
79 
80 #ifdef HWPMC_HOOKS
81 #include <sys/pmckern.h>
82 PMC_SOFT_DEFINE( , , clock, hard);
83 PMC_SOFT_DEFINE( , , clock, stat);
84 PMC_SOFT_DEFINE_EX( , , clock, prof, \
85     cpu_startprofclock, cpu_stopprofclock);
86 #endif
87 
88 #ifdef DEVICE_POLLING
89 extern void hardclock_device_poll(void);
90 #endif /* DEVICE_POLLING */
91 
92 static void initclocks(void *dummy);
93 SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL);
94 
95 /* Spin-lock protecting profiling statistics. */
96 static struct mtx time_lock;
97 
98 SDT_PROVIDER_DECLARE(sched);
99 SDT_PROBE_DEFINE2(sched, , , tick, "struct thread *", "struct proc *");
100 
101 static int
102 sysctl_kern_cp_time(SYSCTL_HANDLER_ARGS)
103 {
104 	int error;
105 	long cp_time[CPUSTATES];
106 #ifdef SCTL_MASK32
107 	int i;
108 	unsigned int cp_time32[CPUSTATES];
109 #endif
110 
111 	read_cpu_time(cp_time);
112 #ifdef SCTL_MASK32
113 	if (req->flags & SCTL_MASK32) {
114 		if (!req->oldptr)
115 			return SYSCTL_OUT(req, 0, sizeof(cp_time32));
116 		for (i = 0; i < CPUSTATES; i++)
117 			cp_time32[i] = (unsigned int)cp_time[i];
118 		error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32));
119 	} else
120 #endif
121 	{
122 		if (!req->oldptr)
123 			return SYSCTL_OUT(req, 0, sizeof(cp_time));
124 		error = SYSCTL_OUT(req, cp_time, sizeof(cp_time));
125 	}
126 	return error;
127 }
128 
129 SYSCTL_PROC(_kern, OID_AUTO, cp_time, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE,
130     0,0, sysctl_kern_cp_time, "LU", "CPU time statistics");
131 
132 static long empty[CPUSTATES];
133 
134 static int
135 sysctl_kern_cp_times(SYSCTL_HANDLER_ARGS)
136 {
137 	struct pcpu *pcpu;
138 	int error;
139 	int c;
140 	long *cp_time;
141 #ifdef SCTL_MASK32
142 	unsigned int cp_time32[CPUSTATES];
143 	int i;
144 #endif
145 
146 	if (!req->oldptr) {
147 #ifdef SCTL_MASK32
148 		if (req->flags & SCTL_MASK32)
149 			return SYSCTL_OUT(req, 0, sizeof(cp_time32) * (mp_maxid + 1));
150 		else
151 #endif
152 			return SYSCTL_OUT(req, 0, sizeof(long) * CPUSTATES * (mp_maxid + 1));
153 	}
154 	for (error = 0, c = 0; error == 0 && c <= mp_maxid; c++) {
155 		if (!CPU_ABSENT(c)) {
156 			pcpu = pcpu_find(c);
157 			cp_time = pcpu->pc_cp_time;
158 		} else {
159 			cp_time = empty;
160 		}
161 #ifdef SCTL_MASK32
162 		if (req->flags & SCTL_MASK32) {
163 			for (i = 0; i < CPUSTATES; i++)
164 				cp_time32[i] = (unsigned int)cp_time[i];
165 			error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32));
166 		} else
167 #endif
168 			error = SYSCTL_OUT(req, cp_time, sizeof(long) * CPUSTATES);
169 	}
170 	return error;
171 }
172 
173 SYSCTL_PROC(_kern, OID_AUTO, cp_times, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE,
174     0,0, sysctl_kern_cp_times, "LU", "per-CPU time statistics");
175 
176 #ifdef DEADLKRES
177 static const char *blessed[] = {
178 	"getblk",
179 	"so_snd_sx",
180 	"so_rcv_sx",
181 	NULL
182 };
183 static int slptime_threshold = 1800;
184 static int blktime_threshold = 900;
185 static int sleepfreq = 3;
186 
187 static void
188 deadlres_td_on_lock(struct proc *p, struct thread *td, int blkticks)
189 {
190 	int tticks;
191 
192 	sx_assert(&allproc_lock, SX_LOCKED);
193 	PROC_LOCK_ASSERT(p, MA_OWNED);
194 	THREAD_LOCK_ASSERT(td, MA_OWNED);
195 	/*
196 	 * The thread should be blocked on a turnstile, simply check
197 	 * if the turnstile channel is in good state.
198 	 */
199 	MPASS(td->td_blocked != NULL);
200 
201 	tticks = ticks - td->td_blktick;
202 	if (tticks > blkticks)
203 		/*
204 		 * Accordingly with provided thresholds, this thread is stuck
205 		 * for too long on a turnstile.
206 		 */
207 		panic("%s: possible deadlock detected for %p, "
208 		    "blocked for %d ticks\n", __func__, td, tticks);
209 }
210 
211 static void
212 deadlres_td_sleep_q(struct proc *p, struct thread *td, int slpticks)
213 {
214 	void *wchan;
215 	int i, slptype, tticks;
216 
217 	sx_assert(&allproc_lock, SX_LOCKED);
218 	PROC_LOCK_ASSERT(p, MA_OWNED);
219 	THREAD_LOCK_ASSERT(td, MA_OWNED);
220 	/*
221 	 * Check if the thread is sleeping on a lock, otherwise skip the check.
222 	 * Drop the thread lock in order to avoid a LOR with the sleepqueue
223 	 * spinlock.
224 	 */
225 	wchan = td->td_wchan;
226 	tticks = ticks - td->td_slptick;
227 	slptype = sleepq_type(wchan);
228 	if ((slptype == SLEEPQ_SX || slptype == SLEEPQ_LK) &&
229 	    tticks > slpticks) {
230 
231 		/*
232 		 * Accordingly with provided thresholds, this thread is stuck
233 		 * for too long on a sleepqueue.
234 		 * However, being on a sleepqueue, we might still check for the
235 		 * blessed list.
236 		 */
237 		for (i = 0; blessed[i] != NULL; i++)
238 			if (!strcmp(blessed[i], td->td_wmesg))
239 				return;
240 
241 		panic("%s: possible deadlock detected for %p, "
242 		    "blocked for %d ticks\n", __func__, td, tticks);
243 	}
244 }
245 
246 static void
247 deadlkres(void)
248 {
249 	struct proc *p;
250 	struct thread *td;
251 	int blkticks, slpticks, tryl;
252 
253 	tryl = 0;
254 	for (;;) {
255 		blkticks = blktime_threshold * hz;
256 		slpticks = slptime_threshold * hz;
257 
258 		/*
259 		 * Avoid to sleep on the sx_lock in order to avoid a
260 		 * possible priority inversion problem leading to
261 		 * starvation.
262 		 * If the lock can't be held after 100 tries, panic.
263 		 */
264 		if (!sx_try_slock(&allproc_lock)) {
265 			if (tryl > 100)
266 				panic("%s: possible deadlock detected "
267 				    "on allproc_lock\n", __func__);
268 			tryl++;
269 			pause("allproc", sleepfreq * hz);
270 			continue;
271 		}
272 		tryl = 0;
273 		FOREACH_PROC_IN_SYSTEM(p) {
274 			PROC_LOCK(p);
275 			if (p->p_state == PRS_NEW) {
276 				PROC_UNLOCK(p);
277 				continue;
278 			}
279 			FOREACH_THREAD_IN_PROC(p, td) {
280 				thread_lock(td);
281 				if (TD_ON_LOCK(td))
282 					deadlres_td_on_lock(p, td,
283 					    blkticks);
284 				else if (TD_IS_SLEEPING(td) &&
285 				    TD_ON_SLEEPQ(td))
286 					deadlres_td_sleep_q(p, td,
287 					    slpticks);
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 #include <sys/watchdog.h>
335 
336 static int watchdog_ticks;
337 static int watchdog_enabled;
338 static void watchdog_fire(void);
339 static void watchdog_config(void *, u_int, int *);
340 
341 static void
342 watchdog_attach(void)
343 {
344 	EVENTHANDLER_REGISTER(watchdog_list, watchdog_config, NULL, 0);
345 }
346 
347 /*
348  * Clock handling routines.
349  *
350  * This code is written to operate with two timers that run independently of
351  * each other.
352  *
353  * The main timer, running hz times per second, is used to trigger interval
354  * timers, timeouts and rescheduling as needed.
355  *
356  * The second timer handles kernel and user profiling,
357  * and does resource use estimation.  If the second timer is programmable,
358  * it is randomized to avoid aliasing between the two clocks.  For example,
359  * the randomization prevents an adversary from always giving up the cpu
360  * just before its quantum expires.  Otherwise, it would never accumulate
361  * cpu ticks.  The mean frequency of the second timer is stathz.
362  *
363  * If no second timer exists, stathz will be zero; in this case we drive
364  * profiling and statistics off the main clock.  This WILL NOT be accurate;
365  * do not do it unless absolutely necessary.
366  *
367  * The statistics clock may (or may not) be run at a higher rate while
368  * profiling.  This profile clock runs at profhz.  We require that profhz
369  * be an integral multiple of stathz.
370  *
371  * If the statistics clock is running fast, it must be divided by the ratio
372  * profhz/stathz for statistics.  (For profiling, every tick counts.)
373  *
374  * Time-of-day is maintained using a "timecounter", which may or may
375  * not be related to the hardware generating the above mentioned
376  * interrupts.
377  */
378 
379 int	stathz;
380 int	profhz;
381 int	profprocs;
382 volatile int	ticks;
383 int	psratio;
384 
385 DPCPU_DEFINE_STATIC(int, pcputicks);	/* Per-CPU version of ticks. */
386 #ifdef DEVICE_POLLING
387 static int devpoll_run = 0;
388 #endif
389 
390 /*
391  * Initialize clock frequencies and start both clocks running.
392  */
393 /* ARGSUSED*/
394 static void
395 initclocks(void *dummy)
396 {
397 	int i;
398 
399 	/*
400 	 * Set divisors to 1 (normal case) and let the machine-specific
401 	 * code do its bit.
402 	 */
403 	mtx_init(&time_lock, "time lock", NULL, MTX_DEF);
404 	cpu_initclocks();
405 
406 	/*
407 	 * Compute profhz/stathz, and fix profhz if needed.
408 	 */
409 	i = stathz ? stathz : hz;
410 	if (profhz == 0)
411 		profhz = i;
412 	psratio = profhz / i;
413 
414 #ifdef SW_WATCHDOG
415 	/* Enable hardclock watchdog now, even if a hardware watchdog exists. */
416 	watchdog_attach();
417 #else
418 	/* Volunteer to run a software watchdog. */
419 	if (wdog_software_attach == NULL)
420 		wdog_software_attach = watchdog_attach;
421 #endif
422 }
423 
424 static __noinline void
425 hardclock_itimer(struct thread *td, struct pstats *pstats, int cnt, int usermode)
426 {
427 	struct proc *p;
428 	int flags;
429 
430 	flags = 0;
431 	p = td->td_proc;
432 	if (usermode &&
433 	    timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) {
434 		PROC_ITIMLOCK(p);
435 		if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL],
436 		    tick * cnt) == 0)
437 			flags |= TDF_ALRMPEND | TDF_ASTPENDING;
438 		PROC_ITIMUNLOCK(p);
439 	}
440 	if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) {
441 		PROC_ITIMLOCK(p);
442 		if (itimerdecr(&pstats->p_timer[ITIMER_PROF],
443 		    tick * cnt) == 0)
444 			flags |= TDF_PROFPEND | TDF_ASTPENDING;
445 		PROC_ITIMUNLOCK(p);
446 	}
447 	if (flags != 0) {
448 		thread_lock(td);
449 		td->td_flags |= flags;
450 		thread_unlock(td);
451 	}
452 }
453 
454 void
455 hardclock(int cnt, int usermode)
456 {
457 	struct pstats *pstats;
458 	struct thread *td = curthread;
459 	struct proc *p = td->td_proc;
460 	int *t = DPCPU_PTR(pcputicks);
461 	int global, i, newticks;
462 
463 	/*
464 	 * Update per-CPU and possibly global ticks values.
465 	 */
466 	*t += cnt;
467 	global = ticks;
468 	do {
469 		newticks = *t - global;
470 		if (newticks <= 0) {
471 			if (newticks < -1)
472 				*t = global - 1;
473 			newticks = 0;
474 			break;
475 		}
476 	} while (!atomic_fcmpset_int(&ticks, &global, *t));
477 
478 	/*
479 	 * Run current process's virtual and profile time, as needed.
480 	 */
481 	pstats = p->p_stats;
482 	if (__predict_false(
483 	    timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) ||
484 	    timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)))
485 		hardclock_itimer(td, pstats, cnt, usermode);
486 
487 #ifdef	HWPMC_HOOKS
488 	if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid)))
489 		PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL);
490 	if (td->td_intr_frame != NULL)
491 		PMC_SOFT_CALL_TF( , , clock, hard, td->td_intr_frame);
492 #endif
493 	/* We are in charge to handle this tick duty. */
494 	if (newticks > 0) {
495 		tc_ticktock(newticks);
496 #ifdef DEVICE_POLLING
497 		/* Dangerous and no need to call these things concurrently. */
498 		if (atomic_cmpset_acq_int(&devpoll_run, 0, 1)) {
499 			/* This is very short and quick. */
500 			hardclock_device_poll();
501 			atomic_store_rel_int(&devpoll_run, 0);
502 		}
503 #endif /* DEVICE_POLLING */
504 		if (watchdog_enabled > 0) {
505 			i = atomic_fetchadd_int(&watchdog_ticks, -newticks);
506 			if (i > 0 && i <= newticks)
507 				watchdog_fire();
508 		}
509 	}
510 	if (curcpu == CPU_FIRST())
511 		cpu_tick_calibration();
512 	if (__predict_false(DPCPU_GET(epoch_cb_count)))
513 		GROUPTASK_ENQUEUE(DPCPU_PTR(epoch_cb_task));
514 }
515 
516 void
517 hardclock_sync(int cpu)
518 {
519 	int *t;
520 	KASSERT(!CPU_ABSENT(cpu), ("Absent CPU %d", cpu));
521 	t = DPCPU_ID_PTR(cpu, pcputicks);
522 
523 	*t = ticks;
524 }
525 
526 /*
527  * Compute number of ticks in the specified amount of time.
528  */
529 int
530 tvtohz(struct timeval *tv)
531 {
532 	unsigned long ticks;
533 	long sec, usec;
534 
535 	/*
536 	 * If the number of usecs in the whole seconds part of the time
537 	 * difference fits in a long, then the total number of usecs will
538 	 * fit in an unsigned long.  Compute the total and convert it to
539 	 * ticks, rounding up and adding 1 to allow for the current tick
540 	 * to expire.  Rounding also depends on unsigned long arithmetic
541 	 * to avoid overflow.
542 	 *
543 	 * Otherwise, if the number of ticks in the whole seconds part of
544 	 * the time difference fits in a long, then convert the parts to
545 	 * ticks separately and add, using similar rounding methods and
546 	 * overflow avoidance.  This method would work in the previous
547 	 * case but it is slightly slower and assumes that hz is integral.
548 	 *
549 	 * Otherwise, round the time difference down to the maximum
550 	 * representable value.
551 	 *
552 	 * If ints have 32 bits, then the maximum value for any timeout in
553 	 * 10ms ticks is 248 days.
554 	 */
555 	sec = tv->tv_sec;
556 	usec = tv->tv_usec;
557 	if (usec < 0) {
558 		sec--;
559 		usec += 1000000;
560 	}
561 	if (sec < 0) {
562 #ifdef DIAGNOSTIC
563 		if (usec > 0) {
564 			sec++;
565 			usec -= 1000000;
566 		}
567 		printf("tvotohz: negative time difference %ld sec %ld usec\n",
568 		       sec, usec);
569 #endif
570 		ticks = 1;
571 	} else if (sec <= LONG_MAX / 1000000)
572 		ticks = howmany(sec * 1000000 + (unsigned long)usec, tick) + 1;
573 	else if (sec <= LONG_MAX / hz)
574 		ticks = sec * hz
575 			+ howmany((unsigned long)usec, tick) + 1;
576 	else
577 		ticks = LONG_MAX;
578 	if (ticks > INT_MAX)
579 		ticks = INT_MAX;
580 	return ((int)ticks);
581 }
582 
583 /*
584  * Start profiling on a process.
585  *
586  * Kernel profiling passes proc0 which never exits and hence
587  * keeps the profile clock running constantly.
588  */
589 void
590 startprofclock(struct proc *p)
591 {
592 
593 	PROC_LOCK_ASSERT(p, MA_OWNED);
594 	if (p->p_flag & P_STOPPROF)
595 		return;
596 	if ((p->p_flag & P_PROFIL) == 0) {
597 		p->p_flag |= P_PROFIL;
598 		mtx_lock(&time_lock);
599 		if (++profprocs == 1)
600 			cpu_startprofclock();
601 		mtx_unlock(&time_lock);
602 	}
603 }
604 
605 /*
606  * Stop profiling on a process.
607  */
608 void
609 stopprofclock(struct proc *p)
610 {
611 
612 	PROC_LOCK_ASSERT(p, MA_OWNED);
613 	if (p->p_flag & P_PROFIL) {
614 		if (p->p_profthreads != 0) {
615 			while (p->p_profthreads != 0) {
616 				p->p_flag |= P_STOPPROF;
617 				msleep(&p->p_profthreads, &p->p_mtx, PPAUSE,
618 				    "stopprof", 0);
619 			}
620 		}
621 		if ((p->p_flag & P_PROFIL) == 0)
622 			return;
623 		p->p_flag &= ~P_PROFIL;
624 		mtx_lock(&time_lock);
625 		if (--profprocs == 0)
626 			cpu_stopprofclock();
627 		mtx_unlock(&time_lock);
628 	}
629 }
630 
631 /*
632  * Statistics clock.  Updates rusage information and calls the scheduler
633  * to adjust priorities of the active thread.
634  *
635  * This should be called by all active processors.
636  */
637 void
638 statclock(int cnt, int usermode)
639 {
640 	struct rusage *ru;
641 	struct vmspace *vm;
642 	struct thread *td;
643 	struct proc *p;
644 	long rss;
645 	long *cp_time;
646 
647 	td = curthread;
648 	p = td->td_proc;
649 
650 	cp_time = (long *)PCPU_PTR(cp_time);
651 	if (usermode) {
652 		/*
653 		 * Charge the time as appropriate.
654 		 */
655 		td->td_uticks += cnt;
656 		if (p->p_nice > NZERO)
657 			cp_time[CP_NICE] += cnt;
658 		else
659 			cp_time[CP_USER] += cnt;
660 	} else {
661 		/*
662 		 * Came from kernel mode, so we were:
663 		 * - handling an interrupt,
664 		 * - doing syscall or trap work on behalf of the current
665 		 *   user process, or
666 		 * - spinning in the idle loop.
667 		 * Whichever it is, charge the time as appropriate.
668 		 * Note that we charge interrupts to the current process,
669 		 * regardless of whether they are ``for'' that process,
670 		 * so that we know how much of its real time was spent
671 		 * in ``non-process'' (i.e., interrupt) work.
672 		 */
673 		if ((td->td_pflags & TDP_ITHREAD) ||
674 		    td->td_intr_nesting_level >= 2) {
675 			td->td_iticks += cnt;
676 			cp_time[CP_INTR] += cnt;
677 		} else {
678 			td->td_pticks += cnt;
679 			td->td_sticks += cnt;
680 			if (!TD_IS_IDLETHREAD(td))
681 				cp_time[CP_SYS] += cnt;
682 			else
683 				cp_time[CP_IDLE] += cnt;
684 		}
685 	}
686 
687 	/* Update resource usage integrals and maximums. */
688 	MPASS(p->p_vmspace != NULL);
689 	vm = p->p_vmspace;
690 	ru = &td->td_ru;
691 	ru->ru_ixrss += pgtok(vm->vm_tsize) * cnt;
692 	ru->ru_idrss += pgtok(vm->vm_dsize) * cnt;
693 	ru->ru_isrss += pgtok(vm->vm_ssize) * cnt;
694 	rss = pgtok(vmspace_resident_count(vm));
695 	if (ru->ru_maxrss < rss)
696 		ru->ru_maxrss = rss;
697 	KTR_POINT2(KTR_SCHED, "thread", sched_tdname(td), "statclock",
698 	    "prio:%d", td->td_priority, "stathz:%d", (stathz)?stathz:hz);
699 	SDT_PROBE2(sched, , , tick, td, td->td_proc);
700 	thread_lock_flags(td, MTX_QUIET);
701 	for ( ; cnt > 0; cnt--)
702 		sched_clock(td);
703 	thread_unlock(td);
704 #ifdef HWPMC_HOOKS
705 	if (td->td_intr_frame != NULL)
706 		PMC_SOFT_CALL_TF( , , clock, stat, td->td_intr_frame);
707 #endif
708 }
709 
710 void
711 profclock(int cnt, 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, cnt);
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) += cnt;
740 			}
741 		}
742 	}
743 #endif
744 #ifdef HWPMC_HOOKS
745 	if (td->td_intr_frame != NULL)
746 		PMC_SOFT_CALL_TF( , , clock, prof, td->td_intr_frame);
747 #endif
748 }
749 
750 /*
751  * Return information about system clocks.
752  */
753 static int
754 sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS)
755 {
756 	struct clockinfo clkinfo;
757 	/*
758 	 * Construct clockinfo structure.
759 	 */
760 	bzero(&clkinfo, sizeof(clkinfo));
761 	clkinfo.hz = hz;
762 	clkinfo.tick = tick;
763 	clkinfo.profhz = profhz;
764 	clkinfo.stathz = stathz ? stathz : hz;
765 	return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req));
766 }
767 
768 SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate,
769 	CTLTYPE_STRUCT|CTLFLAG_RD|CTLFLAG_MPSAFE,
770 	0, 0, sysctl_kern_clockrate, "S,clockinfo",
771 	"Rate and period of various kernel clocks");
772 
773 static void
774 watchdog_config(void *unused __unused, u_int cmd, int *error)
775 {
776 	u_int u;
777 
778 	u = cmd & WD_INTERVAL;
779 	if (u >= WD_TO_1SEC) {
780 		watchdog_ticks = (1 << (u - WD_TO_1SEC)) * hz;
781 		watchdog_enabled = 1;
782 		*error = 0;
783 	} else {
784 		watchdog_enabled = 0;
785 	}
786 }
787 
788 /*
789  * Handle a watchdog timeout by dumping interrupt information and
790  * then either dropping to DDB or panicking.
791  */
792 static void
793 watchdog_fire(void)
794 {
795 	int nintr;
796 	uint64_t inttotal;
797 	u_long *curintr;
798 	char *curname;
799 
800 	curintr = intrcnt;
801 	curname = intrnames;
802 	inttotal = 0;
803 	nintr = sintrcnt / sizeof(u_long);
804 
805 	printf("interrupt                   total\n");
806 	while (--nintr >= 0) {
807 		if (*curintr)
808 			printf("%-12s %20lu\n", curname, *curintr);
809 		curname += strlen(curname) + 1;
810 		inttotal += *curintr++;
811 	}
812 	printf("Total        %20ju\n", (uintmax_t)inttotal);
813 
814 #if defined(KDB) && !defined(KDB_UNATTENDED)
815 	kdb_backtrace();
816 	kdb_enter(KDB_WHY_WATCHDOG, "watchdog timeout");
817 #else
818 	panic("watchdog timeout");
819 #endif
820 }
821