xref: /freebsd/sys/kern/kern_clock.c (revision b197d4b893974c9eb4d7b38704c6d5c486235d6f)
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/eventhandler.h>
53 #include <sys/gtaskqueue.h>
54 #include <sys/kdb.h>
55 #include <sys/kernel.h>
56 #include <sys/kthread.h>
57 #include <sys/ktr.h>
58 #include <sys/lock.h>
59 #include <sys/mutex.h>
60 #include <sys/proc.h>
61 #include <sys/resource.h>
62 #include <sys/resourcevar.h>
63 #include <sys/sched.h>
64 #include <sys/sdt.h>
65 #include <sys/signalvar.h>
66 #include <sys/sleepqueue.h>
67 #include <sys/smp.h>
68 #include <vm/vm.h>
69 #include <vm/pmap.h>
70 #include <vm/vm_map.h>
71 #include <sys/sysctl.h>
72 #include <sys/bus.h>
73 #include <sys/interrupt.h>
74 #include <sys/limits.h>
75 #include <sys/timetc.h>
76 
77 #ifdef GPROF
78 #include <sys/gmon.h>
79 #endif
80 
81 #ifdef HWPMC_HOOKS
82 #include <sys/pmckern.h>
83 PMC_SOFT_DEFINE( , , clock, hard);
84 PMC_SOFT_DEFINE( , , clock, stat);
85 PMC_SOFT_DEFINE_EX( , , clock, prof, \
86     cpu_startprofclock, cpu_stopprofclock);
87 #endif
88 
89 #ifdef DEVICE_POLLING
90 extern void hardclock_device_poll(void);
91 #endif /* DEVICE_POLLING */
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 deadlres_td_on_lock(struct proc *p, struct thread *td, int blkticks)
187 {
188 	int tticks;
189 
190 	sx_assert(&allproc_lock, SX_LOCKED);
191 	PROC_LOCK_ASSERT(p, MA_OWNED);
192 	THREAD_LOCK_ASSERT(td, MA_OWNED);
193 	/*
194 	 * The thread should be blocked on a turnstile, simply check
195 	 * if the turnstile channel is in good state.
196 	 */
197 	MPASS(td->td_blocked != NULL);
198 
199 	tticks = ticks - td->td_blktick;
200 	if (tticks > blkticks)
201 		/*
202 		 * Accordingly with provided thresholds, this thread is stuck
203 		 * for too long on a turnstile.
204 		 */
205 		panic("%s: possible deadlock detected for %p (%s), "
206 		    "blocked for %d ticks\n", __func__,
207 		    td, sched_tdname(td), tticks);
208 }
209 
210 static void
211 deadlres_td_sleep_q(struct proc *p, struct thread *td, int slpticks)
212 {
213 	const void *wchan;
214 	int i, slptype, tticks;
215 
216 	sx_assert(&allproc_lock, SX_LOCKED);
217 	PROC_LOCK_ASSERT(p, MA_OWNED);
218 	THREAD_LOCK_ASSERT(td, MA_OWNED);
219 	/*
220 	 * Check if the thread is sleeping on a lock, otherwise skip the check.
221 	 * Drop the thread lock in order to avoid a LOR with the sleepqueue
222 	 * spinlock.
223 	 */
224 	wchan = td->td_wchan;
225 	tticks = ticks - td->td_slptick;
226 	slptype = sleepq_type(wchan);
227 	if ((slptype == SLEEPQ_SX || slptype == SLEEPQ_LK) &&
228 	    tticks > slpticks) {
229 		/*
230 		 * Accordingly with provided thresholds, this thread is stuck
231 		 * for too long on a sleepqueue.
232 		 * However, being on a sleepqueue, we might still check for the
233 		 * blessed list.
234 		 */
235 		for (i = 0; blessed[i] != NULL; i++)
236 			if (!strcmp(blessed[i], td->td_wmesg))
237 				return;
238 
239 		panic("%s: possible deadlock detected for %p (%s), "
240 		    "blocked for %d ticks\n", __func__,
241 		    td, sched_tdname(td), tticks);
242 	}
243 }
244 
245 static void
246 deadlkres(void)
247 {
248 	struct proc *p;
249 	struct thread *td;
250 	int blkticks, slpticks, tryl;
251 
252 	tryl = 0;
253 	for (;;) {
254 		blkticks = blktime_threshold * hz;
255 		slpticks = slptime_threshold * hz;
256 
257 		/*
258 		 * Avoid to sleep on the sx_lock in order to avoid a
259 		 * possible priority inversion problem leading to
260 		 * starvation.
261 		 * If the lock can't be held after 100 tries, panic.
262 		 */
263 		if (!sx_try_slock(&allproc_lock)) {
264 			if (tryl > 100)
265 				panic("%s: possible deadlock detected "
266 				    "on allproc_lock\n", __func__);
267 			tryl++;
268 			pause("allproc", sleepfreq * hz);
269 			continue;
270 		}
271 		tryl = 0;
272 		FOREACH_PROC_IN_SYSTEM(p) {
273 			PROC_LOCK(p);
274 			if (p->p_state == PRS_NEW) {
275 				PROC_UNLOCK(p);
276 				continue;
277 			}
278 			FOREACH_THREAD_IN_PROC(p, td) {
279 				thread_lock(td);
280 				if (TD_ON_LOCK(td))
281 					deadlres_td_on_lock(p, td,
282 					    blkticks);
283 				else if (TD_IS_SLEEPING(td))
284 					deadlres_td_sleep_q(p, td,
285 					    slpticks);
286 				thread_unlock(td);
287 			}
288 			PROC_UNLOCK(p);
289 		}
290 		sx_sunlock(&allproc_lock);
291 
292 		/* Sleep for sleepfreq seconds. */
293 		pause("-", sleepfreq * hz);
294 	}
295 }
296 
297 static struct kthread_desc deadlkres_kd = {
298 	"deadlkres",
299 	deadlkres,
300 	(struct thread **)NULL
301 };
302 
303 SYSINIT(deadlkres, SI_SUB_CLOCKS, SI_ORDER_ANY, kthread_start, &deadlkres_kd);
304 
305 static SYSCTL_NODE(_debug, OID_AUTO, deadlkres, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
306     "Deadlock resolver");
307 SYSCTL_INT(_debug_deadlkres, OID_AUTO, slptime_threshold, CTLFLAG_RW,
308     &slptime_threshold, 0,
309     "Number of seconds within is valid to sleep on a sleepqueue");
310 SYSCTL_INT(_debug_deadlkres, OID_AUTO, blktime_threshold, CTLFLAG_RW,
311     &blktime_threshold, 0,
312     "Number of seconds within is valid to block on a turnstile");
313 SYSCTL_INT(_debug_deadlkres, OID_AUTO, sleepfreq, CTLFLAG_RW, &sleepfreq, 0,
314     "Number of seconds between any deadlock resolver thread run");
315 #endif	/* DEADLKRES */
316 
317 void
318 read_cpu_time(long *cp_time)
319 {
320 	struct pcpu *pc;
321 	int i, j;
322 
323 	/* Sum up global cp_time[]. */
324 	bzero(cp_time, sizeof(long) * CPUSTATES);
325 	CPU_FOREACH(i) {
326 		pc = pcpu_find(i);
327 		for (j = 0; j < CPUSTATES; j++)
328 			cp_time[j] += pc->pc_cp_time[j];
329 	}
330 }
331 
332 #include <sys/watchdog.h>
333 
334 static int watchdog_ticks;
335 static int watchdog_enabled;
336 static void watchdog_fire(void);
337 static void watchdog_config(void *, u_int, int *);
338 
339 static void
340 watchdog_attach(void)
341 {
342 	EVENTHANDLER_REGISTER(watchdog_list, watchdog_config, NULL, 0);
343 }
344 
345 /*
346  * Clock handling routines.
347  *
348  * This code is written to operate with two timers that run independently of
349  * each other.
350  *
351  * The main timer, running hz times per second, is used to trigger interval
352  * timers, timeouts and rescheduling as needed.
353  *
354  * The second timer handles kernel and user profiling,
355  * and does resource use estimation.  If the second timer is programmable,
356  * it is randomized to avoid aliasing between the two clocks.  For example,
357  * the randomization prevents an adversary from always giving up the cpu
358  * just before its quantum expires.  Otherwise, it would never accumulate
359  * cpu ticks.  The mean frequency of the second timer is stathz.
360  *
361  * If no second timer exists, stathz will be zero; in this case we drive
362  * profiling and statistics off the main clock.  This WILL NOT be accurate;
363  * do not do it unless absolutely necessary.
364  *
365  * The statistics clock may (or may not) be run at a higher rate while
366  * profiling.  This profile clock runs at profhz.  We require that profhz
367  * be an integral multiple of stathz.
368  *
369  * If the statistics clock is running fast, it must be divided by the ratio
370  * profhz/stathz for statistics.  (For profiling, every tick counts.)
371  *
372  * Time-of-day is maintained using a "timecounter", which may or may
373  * not be related to the hardware generating the above mentioned
374  * interrupts.
375  */
376 
377 int	stathz;
378 int	profhz;
379 int	profprocs;
380 volatile int	ticks;
381 int	psratio;
382 
383 DPCPU_DEFINE_STATIC(int, pcputicks);	/* Per-CPU version of ticks. */
384 #ifdef DEVICE_POLLING
385 static int devpoll_run = 0;
386 #endif
387 
388 static void
389 ast_oweupc(struct thread *td, int tda __unused)
390 {
391 	if ((td->td_proc->p_flag & P_PROFIL) == 0)
392 		return;
393 	addupc_task(td, td->td_profil_addr, td->td_profil_ticks);
394 	td->td_profil_ticks = 0;
395 	td->td_pflags &= ~TDP_OWEUPC;
396 }
397 
398 static void
399 ast_alrm(struct thread *td, int tda __unused)
400 {
401 	struct proc *p;
402 
403 	p = td->td_proc;
404 	PROC_LOCK(p);
405 	kern_psignal(p, SIGVTALRM);
406 	PROC_UNLOCK(p);
407 }
408 
409 static void
410 ast_prof(struct thread *td, int tda __unused)
411 {
412 	struct proc *p;
413 
414 	p = td->td_proc;
415 	PROC_LOCK(p);
416 	kern_psignal(p, SIGPROF);
417 	PROC_UNLOCK(p);
418 }
419 
420 /*
421  * Initialize clock frequencies and start both clocks running.
422  */
423 static void
424 initclocks(void *dummy __unused)
425 {
426 	int i;
427 
428 	/*
429 	 * Set divisors to 1 (normal case) and let the machine-specific
430 	 * code do its bit.
431 	 */
432 	mtx_init(&time_lock, "time lock", NULL, MTX_DEF);
433 	cpu_initclocks();
434 
435 	/*
436 	 * Compute profhz/stathz, and fix profhz if needed.
437 	 */
438 	i = stathz ? stathz : hz;
439 	if (profhz == 0)
440 		profhz = i;
441 	psratio = profhz / i;
442 
443 	ast_register(TDA_OWEUPC, ASTR_ASTF_REQUIRED, 0, ast_oweupc);
444 	ast_register(TDA_ALRM, ASTR_ASTF_REQUIRED, 0, ast_alrm);
445 	ast_register(TDA_PROF, ASTR_ASTF_REQUIRED, 0, ast_prof);
446 
447 #ifdef SW_WATCHDOG
448 	/* Enable hardclock watchdog now, even if a hardware watchdog exists. */
449 	watchdog_attach();
450 #else
451 	/* Volunteer to run a software watchdog. */
452 	if (wdog_software_attach == NULL)
453 		wdog_software_attach = watchdog_attach;
454 #endif
455 }
456 SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL);
457 
458 static __noinline void
459 hardclock_itimer(struct thread *td, struct pstats *pstats, int cnt, int usermode)
460 {
461 	struct proc *p;
462 	int ast;
463 
464 	ast = 0;
465 	p = td->td_proc;
466 	if (usermode &&
467 	    timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) {
468 		PROC_ITIMLOCK(p);
469 		if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL],
470 		    tick * cnt) == 0)
471 			ast |= TDAI(TDA_ALRM);
472 		PROC_ITIMUNLOCK(p);
473 	}
474 	if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) {
475 		PROC_ITIMLOCK(p);
476 		if (itimerdecr(&pstats->p_timer[ITIMER_PROF],
477 		    tick * cnt) == 0)
478 			ast |= TDAI(TDA_PROF);
479 		PROC_ITIMUNLOCK(p);
480 	}
481 	if (ast != 0)
482 		ast_sched_mask(td, ast);
483 }
484 
485 void
486 hardclock(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 global, i, newticks;
493 
494 	/*
495 	 * Update per-CPU and possibly global ticks values.
496 	 */
497 	*t += cnt;
498 	global = ticks;
499 	do {
500 		newticks = *t - global;
501 		if (newticks <= 0) {
502 			if (newticks < -1)
503 				*t = global - 1;
504 			newticks = 0;
505 			break;
506 		}
507 	} while (!atomic_fcmpset_int(&ticks, &global, *t));
508 
509 	/*
510 	 * Run current process's virtual and profile time, as needed.
511 	 */
512 	pstats = p->p_stats;
513 	if (__predict_false(
514 	    timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) ||
515 	    timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)))
516 		hardclock_itimer(td, pstats, cnt, usermode);
517 
518 #ifdef	HWPMC_HOOKS
519 	if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid)))
520 		PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL);
521 	if (td->td_intr_frame != NULL)
522 		PMC_SOFT_CALL_TF( , , clock, hard, td->td_intr_frame);
523 #endif
524 	/* We are in charge to handle this tick duty. */
525 	if (newticks > 0) {
526 		tc_ticktock(newticks);
527 #ifdef DEVICE_POLLING
528 		/* Dangerous and no need to call these things concurrently. */
529 		if (atomic_cmpset_acq_int(&devpoll_run, 0, 1)) {
530 			/* This is very short and quick. */
531 			hardclock_device_poll();
532 			atomic_store_rel_int(&devpoll_run, 0);
533 		}
534 #endif /* DEVICE_POLLING */
535 		if (watchdog_enabled > 0) {
536 			i = atomic_fetchadd_int(&watchdog_ticks, -newticks);
537 			if (i > 0 && i <= newticks)
538 				watchdog_fire();
539 		}
540 		intr_event_handle(clk_intr_event, NULL);
541 	}
542 	if (curcpu == CPU_FIRST())
543 		cpu_tick_calibration();
544 	if (__predict_false(DPCPU_GET(epoch_cb_count)))
545 		GROUPTASK_ENQUEUE(DPCPU_PTR(epoch_cb_task));
546 }
547 
548 void
549 hardclock_sync(int cpu)
550 {
551 	int *t;
552 	KASSERT(!CPU_ABSENT(cpu), ("Absent CPU %d", cpu));
553 	t = DPCPU_ID_PTR(cpu, pcputicks);
554 
555 	*t = ticks;
556 }
557 
558 /*
559  * Compute number of ticks in the specified amount of time.
560  */
561 int
562 tvtohz(struct timeval *tv)
563 {
564 	unsigned long ticks;
565 	long sec, usec;
566 
567 	/*
568 	 * If the number of usecs in the whole seconds part of the time
569 	 * difference fits in a long, then the total number of usecs will
570 	 * fit in an unsigned long.  Compute the total and convert it to
571 	 * ticks, rounding up and adding 1 to allow for the current tick
572 	 * to expire.  Rounding also depends on unsigned long arithmetic
573 	 * to avoid overflow.
574 	 *
575 	 * Otherwise, if the number of ticks in the whole seconds part of
576 	 * the time difference fits in a long, then convert the parts to
577 	 * ticks separately and add, using similar rounding methods and
578 	 * overflow avoidance.  This method would work in the previous
579 	 * case but it is slightly slower and assumes that hz is integral.
580 	 *
581 	 * Otherwise, round the time difference down to the maximum
582 	 * representable value.
583 	 *
584 	 * If ints have 32 bits, then the maximum value for any timeout in
585 	 * 10ms ticks is 248 days.
586 	 */
587 	sec = tv->tv_sec;
588 	usec = tv->tv_usec;
589 	if (usec < 0) {
590 		sec--;
591 		usec += 1000000;
592 	}
593 	if (sec < 0) {
594 #ifdef DIAGNOSTIC
595 		if (usec > 0) {
596 			sec++;
597 			usec -= 1000000;
598 		}
599 		printf("tvotohz: negative time difference %ld sec %ld usec\n",
600 		       sec, usec);
601 #endif
602 		ticks = 1;
603 	} else if (sec <= LONG_MAX / 1000000)
604 		ticks = howmany(sec * 1000000 + (unsigned long)usec, tick) + 1;
605 	else if (sec <= LONG_MAX / hz)
606 		ticks = sec * hz
607 			+ howmany((unsigned long)usec, tick) + 1;
608 	else
609 		ticks = LONG_MAX;
610 	if (ticks > INT_MAX)
611 		ticks = INT_MAX;
612 	return ((int)ticks);
613 }
614 
615 /*
616  * Start profiling on a process.
617  *
618  * Kernel profiling passes proc0 which never exits and hence
619  * keeps the profile clock running constantly.
620  */
621 void
622 startprofclock(struct proc *p)
623 {
624 
625 	PROC_LOCK_ASSERT(p, MA_OWNED);
626 	if (p->p_flag & P_STOPPROF)
627 		return;
628 	if ((p->p_flag & P_PROFIL) == 0) {
629 		p->p_flag |= P_PROFIL;
630 		mtx_lock(&time_lock);
631 		if (++profprocs == 1)
632 			cpu_startprofclock();
633 		mtx_unlock(&time_lock);
634 	}
635 }
636 
637 /*
638  * Stop profiling on a process.
639  */
640 void
641 stopprofclock(struct proc *p)
642 {
643 
644 	PROC_LOCK_ASSERT(p, MA_OWNED);
645 	if (p->p_flag & P_PROFIL) {
646 		if (p->p_profthreads != 0) {
647 			while (p->p_profthreads != 0) {
648 				p->p_flag |= P_STOPPROF;
649 				msleep(&p->p_profthreads, &p->p_mtx, PPAUSE,
650 				    "stopprof", 0);
651 			}
652 		}
653 		if ((p->p_flag & P_PROFIL) == 0)
654 			return;
655 		p->p_flag &= ~P_PROFIL;
656 		mtx_lock(&time_lock);
657 		if (--profprocs == 0)
658 			cpu_stopprofclock();
659 		mtx_unlock(&time_lock);
660 	}
661 }
662 
663 /*
664  * Statistics clock.  Updates rusage information and calls the scheduler
665  * to adjust priorities of the active thread.
666  *
667  * This should be called by all active processors.
668  */
669 void
670 statclock(int cnt, int usermode)
671 {
672 	struct rusage *ru;
673 	struct vmspace *vm;
674 	struct thread *td;
675 	struct proc *p;
676 	long rss;
677 	long *cp_time;
678 	uint64_t runtime, new_switchtime;
679 
680 	td = curthread;
681 	p = td->td_proc;
682 
683 	cp_time = (long *)PCPU_PTR(cp_time);
684 	if (usermode) {
685 		/*
686 		 * Charge the time as appropriate.
687 		 */
688 		td->td_uticks += cnt;
689 		if (p->p_nice > NZERO)
690 			cp_time[CP_NICE] += cnt;
691 		else
692 			cp_time[CP_USER] += cnt;
693 	} else {
694 		/*
695 		 * Came from kernel mode, so we were:
696 		 * - handling an interrupt,
697 		 * - doing syscall or trap work on behalf of the current
698 		 *   user process, or
699 		 * - spinning in the idle loop.
700 		 * Whichever it is, charge the time as appropriate.
701 		 * Note that we charge interrupts to the current process,
702 		 * regardless of whether they are ``for'' that process,
703 		 * so that we know how much of its real time was spent
704 		 * in ``non-process'' (i.e., interrupt) work.
705 		 */
706 		if ((td->td_pflags & TDP_ITHREAD) ||
707 		    td->td_intr_nesting_level >= 2) {
708 			td->td_iticks += cnt;
709 			cp_time[CP_INTR] += cnt;
710 		} else {
711 			td->td_pticks += cnt;
712 			td->td_sticks += cnt;
713 			if (!TD_IS_IDLETHREAD(td))
714 				cp_time[CP_SYS] += cnt;
715 			else
716 				cp_time[CP_IDLE] += cnt;
717 		}
718 	}
719 
720 	/* Update resource usage integrals and maximums. */
721 	MPASS(p->p_vmspace != NULL);
722 	vm = p->p_vmspace;
723 	ru = &td->td_ru;
724 	ru->ru_ixrss += pgtok(vm->vm_tsize) * cnt;
725 	ru->ru_idrss += pgtok(vm->vm_dsize) * cnt;
726 	ru->ru_isrss += pgtok(vm->vm_ssize) * cnt;
727 	rss = pgtok(vmspace_resident_count(vm));
728 	if (ru->ru_maxrss < rss)
729 		ru->ru_maxrss = rss;
730 	KTR_POINT2(KTR_SCHED, "thread", sched_tdname(td), "statclock",
731 	    "prio:%d", td->td_priority, "stathz:%d", (stathz)?stathz:hz);
732 	SDT_PROBE2(sched, , , tick, td, td->td_proc);
733 	thread_lock_flags(td, MTX_QUIET);
734 
735 	/*
736 	 * Compute the amount of time during which the current
737 	 * thread was running, and add that to its total so far.
738 	 */
739 	new_switchtime = cpu_ticks();
740 	runtime = new_switchtime - PCPU_GET(switchtime);
741 	td->td_runtime += runtime;
742 	td->td_incruntime += runtime;
743 	PCPU_SET(switchtime, new_switchtime);
744 
745 	sched_clock(td, cnt);
746 	thread_unlock(td);
747 #ifdef HWPMC_HOOKS
748 	if (td->td_intr_frame != NULL)
749 		PMC_SOFT_CALL_TF( , , clock, stat, td->td_intr_frame);
750 #endif
751 }
752 
753 void
754 profclock(int cnt, int usermode, uintfptr_t pc)
755 {
756 	struct thread *td;
757 #ifdef GPROF
758 	struct gmonparam *g;
759 	uintfptr_t i;
760 #endif
761 
762 	td = curthread;
763 	if (usermode) {
764 		/*
765 		 * Came from user mode; CPU was in user state.
766 		 * If this process is being profiled, record the tick.
767 		 * if there is no related user location yet, don't
768 		 * bother trying to count it.
769 		 */
770 		if (td->td_proc->p_flag & P_PROFIL)
771 			addupc_intr(td, pc, cnt);
772 	}
773 #ifdef GPROF
774 	else {
775 		/*
776 		 * Kernel statistics are just like addupc_intr, only easier.
777 		 */
778 		g = &_gmonparam;
779 		if (g->state == GMON_PROF_ON && pc >= g->lowpc) {
780 			i = PC_TO_I(g, pc);
781 			if (i < g->textsize) {
782 				KCOUNT(g, i) += cnt;
783 			}
784 		}
785 	}
786 #endif
787 #ifdef HWPMC_HOOKS
788 	if (td->td_intr_frame != NULL)
789 		PMC_SOFT_CALL_TF( , , clock, prof, td->td_intr_frame);
790 #endif
791 }
792 
793 /*
794  * Return information about system clocks.
795  */
796 static int
797 sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS)
798 {
799 	struct clockinfo clkinfo;
800 	/*
801 	 * Construct clockinfo structure.
802 	 */
803 	bzero(&clkinfo, sizeof(clkinfo));
804 	clkinfo.hz = hz;
805 	clkinfo.tick = tick;
806 	clkinfo.profhz = profhz;
807 	clkinfo.stathz = stathz ? stathz : hz;
808 	return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req));
809 }
810 
811 SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate,
812 	CTLTYPE_STRUCT|CTLFLAG_RD|CTLFLAG_MPSAFE,
813 	0, 0, sysctl_kern_clockrate, "S,clockinfo",
814 	"Rate and period of various kernel clocks");
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