xref: /freebsd/sys/kern/kern_clock.c (revision dd41de95a84d979615a2ef11df6850622bf6184e)
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 static void initclocks(void *dummy);
94 SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL);
95 
96 /* Spin-lock protecting profiling statistics. */
97 static struct mtx time_lock;
98 
99 SDT_PROVIDER_DECLARE(sched);
100 SDT_PROBE_DEFINE2(sched, , , tick, "struct thread *", "struct proc *");
101 
102 static int
103 sysctl_kern_cp_time(SYSCTL_HANDLER_ARGS)
104 {
105 	int error;
106 	long cp_time[CPUSTATES];
107 #ifdef SCTL_MASK32
108 	int i;
109 	unsigned int cp_time32[CPUSTATES];
110 #endif
111 
112 	read_cpu_time(cp_time);
113 #ifdef SCTL_MASK32
114 	if (req->flags & SCTL_MASK32) {
115 		if (!req->oldptr)
116 			return SYSCTL_OUT(req, 0, sizeof(cp_time32));
117 		for (i = 0; i < CPUSTATES; i++)
118 			cp_time32[i] = (unsigned int)cp_time[i];
119 		error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32));
120 	} else
121 #endif
122 	{
123 		if (!req->oldptr)
124 			return SYSCTL_OUT(req, 0, sizeof(cp_time));
125 		error = SYSCTL_OUT(req, cp_time, sizeof(cp_time));
126 	}
127 	return error;
128 }
129 
130 SYSCTL_PROC(_kern, OID_AUTO, cp_time, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE,
131     0,0, sysctl_kern_cp_time, "LU", "CPU time statistics");
132 
133 static long empty[CPUSTATES];
134 
135 static int
136 sysctl_kern_cp_times(SYSCTL_HANDLER_ARGS)
137 {
138 	struct pcpu *pcpu;
139 	int error;
140 	int c;
141 	long *cp_time;
142 #ifdef SCTL_MASK32
143 	unsigned int cp_time32[CPUSTATES];
144 	int i;
145 #endif
146 
147 	if (!req->oldptr) {
148 #ifdef SCTL_MASK32
149 		if (req->flags & SCTL_MASK32)
150 			return SYSCTL_OUT(req, 0, sizeof(cp_time32) * (mp_maxid + 1));
151 		else
152 #endif
153 			return SYSCTL_OUT(req, 0, sizeof(long) * CPUSTATES * (mp_maxid + 1));
154 	}
155 	for (error = 0, c = 0; error == 0 && c <= mp_maxid; c++) {
156 		if (!CPU_ABSENT(c)) {
157 			pcpu = pcpu_find(c);
158 			cp_time = pcpu->pc_cp_time;
159 		} else {
160 			cp_time = empty;
161 		}
162 #ifdef SCTL_MASK32
163 		if (req->flags & SCTL_MASK32) {
164 			for (i = 0; i < CPUSTATES; i++)
165 				cp_time32[i] = (unsigned int)cp_time[i];
166 			error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32));
167 		} else
168 #endif
169 			error = SYSCTL_OUT(req, cp_time, sizeof(long) * CPUSTATES);
170 	}
171 	return error;
172 }
173 
174 SYSCTL_PROC(_kern, OID_AUTO, cp_times, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE,
175     0,0, sysctl_kern_cp_times, "LU", "per-CPU time statistics");
176 
177 #ifdef DEADLKRES
178 static const char *blessed[] = {
179 	"getblk",
180 	"so_snd_sx",
181 	"so_rcv_sx",
182 	NULL
183 };
184 static int slptime_threshold = 1800;
185 static int blktime_threshold = 900;
186 static int sleepfreq = 3;
187 
188 static void
189 deadlres_td_on_lock(struct proc *p, struct thread *td, int blkticks)
190 {
191 	int tticks;
192 
193 	sx_assert(&allproc_lock, SX_LOCKED);
194 	PROC_LOCK_ASSERT(p, MA_OWNED);
195 	THREAD_LOCK_ASSERT(td, MA_OWNED);
196 	/*
197 	 * The thread should be blocked on a turnstile, simply check
198 	 * if the turnstile channel is in good state.
199 	 */
200 	MPASS(td->td_blocked != NULL);
201 
202 	tticks = ticks - td->td_blktick;
203 	if (tticks > blkticks)
204 		/*
205 		 * Accordingly with provided thresholds, this thread is stuck
206 		 * for too long on a turnstile.
207 		 */
208 		panic("%s: possible deadlock detected for %p (%s), "
209 		    "blocked for %d ticks\n", __func__,
210 		    td, sched_tdname(td), tticks);
211 }
212 
213 static void
214 deadlres_td_sleep_q(struct proc *p, struct thread *td, int slpticks)
215 {
216 	const void *wchan;
217 	int i, slptype, tticks;
218 
219 	sx_assert(&allproc_lock, SX_LOCKED);
220 	PROC_LOCK_ASSERT(p, MA_OWNED);
221 	THREAD_LOCK_ASSERT(td, MA_OWNED);
222 	/*
223 	 * Check if the thread is sleeping on a lock, otherwise skip the check.
224 	 * Drop the thread lock in order to avoid a LOR with the sleepqueue
225 	 * spinlock.
226 	 */
227 	wchan = td->td_wchan;
228 	tticks = ticks - td->td_slptick;
229 	slptype = sleepq_type(wchan);
230 	if ((slptype == SLEEPQ_SX || slptype == SLEEPQ_LK) &&
231 	    tticks > slpticks) {
232 		/*
233 		 * Accordingly with provided thresholds, this thread is stuck
234 		 * for too long on a sleepqueue.
235 		 * However, being on a sleepqueue, we might still check for the
236 		 * blessed list.
237 		 */
238 		for (i = 0; blessed[i] != NULL; i++)
239 			if (!strcmp(blessed[i], td->td_wmesg))
240 				return;
241 
242 		panic("%s: possible deadlock detected for %p (%s), "
243 		    "blocked for %d ticks\n", __func__,
244 		    td, sched_tdname(td), tticks);
245 	}
246 }
247 
248 static void
249 deadlkres(void)
250 {
251 	struct proc *p;
252 	struct thread *td;
253 	int blkticks, slpticks, tryl;
254 
255 	tryl = 0;
256 	for (;;) {
257 		blkticks = blktime_threshold * hz;
258 		slpticks = slptime_threshold * hz;
259 
260 		/*
261 		 * Avoid to sleep on the sx_lock in order to avoid a
262 		 * possible priority inversion problem leading to
263 		 * starvation.
264 		 * If the lock can't be held after 100 tries, panic.
265 		 */
266 		if (!sx_try_slock(&allproc_lock)) {
267 			if (tryl > 100)
268 				panic("%s: possible deadlock detected "
269 				    "on allproc_lock\n", __func__);
270 			tryl++;
271 			pause("allproc", sleepfreq * hz);
272 			continue;
273 		}
274 		tryl = 0;
275 		FOREACH_PROC_IN_SYSTEM(p) {
276 			PROC_LOCK(p);
277 			if (p->p_state == PRS_NEW) {
278 				PROC_UNLOCK(p);
279 				continue;
280 			}
281 			FOREACH_THREAD_IN_PROC(p, td) {
282 				thread_lock(td);
283 				if (TD_ON_LOCK(td))
284 					deadlres_td_on_lock(p, td,
285 					    blkticks);
286 				else if (TD_IS_SLEEPING(td))
287 					deadlres_td_sleep_q(p, td,
288 					    slpticks);
289 				thread_unlock(td);
290 			}
291 			PROC_UNLOCK(p);
292 		}
293 		sx_sunlock(&allproc_lock);
294 
295 		/* Sleep for sleepfreq seconds. */
296 		pause("-", sleepfreq * hz);
297 	}
298 }
299 
300 static struct kthread_desc deadlkres_kd = {
301 	"deadlkres",
302 	deadlkres,
303 	(struct thread **)NULL
304 };
305 
306 SYSINIT(deadlkres, SI_SUB_CLOCKS, SI_ORDER_ANY, kthread_start, &deadlkres_kd);
307 
308 static SYSCTL_NODE(_debug, OID_AUTO, deadlkres, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
309     "Deadlock resolver");
310 SYSCTL_INT(_debug_deadlkres, OID_AUTO, slptime_threshold, CTLFLAG_RW,
311     &slptime_threshold, 0,
312     "Number of seconds within is valid to sleep on a sleepqueue");
313 SYSCTL_INT(_debug_deadlkres, OID_AUTO, blktime_threshold, CTLFLAG_RW,
314     &blktime_threshold, 0,
315     "Number of seconds within is valid to block on a turnstile");
316 SYSCTL_INT(_debug_deadlkres, OID_AUTO, sleepfreq, CTLFLAG_RW, &sleepfreq, 0,
317     "Number of seconds between any deadlock resolver thread run");
318 #endif	/* DEADLKRES */
319 
320 void
321 read_cpu_time(long *cp_time)
322 {
323 	struct pcpu *pc;
324 	int i, j;
325 
326 	/* Sum up global cp_time[]. */
327 	bzero(cp_time, sizeof(long) * CPUSTATES);
328 	CPU_FOREACH(i) {
329 		pc = pcpu_find(i);
330 		for (j = 0; j < CPUSTATES; j++)
331 			cp_time[j] += pc->pc_cp_time[j];
332 	}
333 }
334 
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 
342 static void
343 watchdog_attach(void)
344 {
345 	EVENTHANDLER_REGISTER(watchdog_list, watchdog_config, NULL, 0);
346 }
347 
348 /*
349  * Clock handling routines.
350  *
351  * This code is written to operate with two timers that run independently of
352  * each other.
353  *
354  * The main timer, running hz times per second, is used to trigger interval
355  * timers, timeouts and rescheduling as needed.
356  *
357  * The second timer handles kernel and user profiling,
358  * and does resource use estimation.  If the second timer is programmable,
359  * it is randomized to avoid aliasing between the two clocks.  For example,
360  * the randomization prevents an adversary from always giving up the cpu
361  * just before its quantum expires.  Otherwise, it would never accumulate
362  * cpu ticks.  The mean frequency of the second timer is stathz.
363  *
364  * If no second timer exists, stathz will be zero; in this case we drive
365  * profiling and statistics off the main clock.  This WILL NOT be accurate;
366  * do not do it unless absolutely necessary.
367  *
368  * The statistics clock may (or may not) be run at a higher rate while
369  * profiling.  This profile clock runs at profhz.  We require that profhz
370  * be an integral multiple of stathz.
371  *
372  * If the statistics clock is running fast, it must be divided by the ratio
373  * profhz/stathz for statistics.  (For profiling, every tick counts.)
374  *
375  * Time-of-day is maintained using a "timecounter", which may or may
376  * not be related to the hardware generating the above mentioned
377  * interrupts.
378  */
379 
380 int	stathz;
381 int	profhz;
382 int	profprocs;
383 volatile int	ticks;
384 int	psratio;
385 
386 DPCPU_DEFINE_STATIC(int, pcputicks);	/* Per-CPU version of ticks. */
387 #ifdef DEVICE_POLLING
388 static int devpoll_run = 0;
389 #endif
390 
391 /*
392  * Initialize clock frequencies and start both clocks running.
393  */
394 /* ARGSUSED*/
395 static void
396 initclocks(void *dummy)
397 {
398 	int i;
399 
400 	/*
401 	 * Set divisors to 1 (normal case) and let the machine-specific
402 	 * code do its bit.
403 	 */
404 	mtx_init(&time_lock, "time lock", NULL, MTX_DEF);
405 	cpu_initclocks();
406 
407 	/*
408 	 * Compute profhz/stathz, and fix profhz if needed.
409 	 */
410 	i = stathz ? stathz : hz;
411 	if (profhz == 0)
412 		profhz = i;
413 	psratio = profhz / i;
414 
415 #ifdef SW_WATCHDOG
416 	/* Enable hardclock watchdog now, even if a hardware watchdog exists. */
417 	watchdog_attach();
418 #else
419 	/* Volunteer to run a software watchdog. */
420 	if (wdog_software_attach == NULL)
421 		wdog_software_attach = watchdog_attach;
422 #endif
423 }
424 
425 static __noinline void
426 hardclock_itimer(struct thread *td, struct pstats *pstats, int cnt, int usermode)
427 {
428 	struct proc *p;
429 	int flags;
430 
431 	flags = 0;
432 	p = td->td_proc;
433 	if (usermode &&
434 	    timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) {
435 		PROC_ITIMLOCK(p);
436 		if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL],
437 		    tick * cnt) == 0)
438 			flags |= TDF_ALRMPEND | TDF_ASTPENDING;
439 		PROC_ITIMUNLOCK(p);
440 	}
441 	if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) {
442 		PROC_ITIMLOCK(p);
443 		if (itimerdecr(&pstats->p_timer[ITIMER_PROF],
444 		    tick * cnt) == 0)
445 			flags |= TDF_PROFPEND | TDF_ASTPENDING;
446 		PROC_ITIMUNLOCK(p);
447 	}
448 	if (flags != 0) {
449 		thread_lock(td);
450 		td->td_flags |= flags;
451 		thread_unlock(td);
452 	}
453 }
454 
455 void
456 hardclock(int cnt, int usermode)
457 {
458 	struct pstats *pstats;
459 	struct thread *td = curthread;
460 	struct proc *p = td->td_proc;
461 	int *t = DPCPU_PTR(pcputicks);
462 	int global, i, newticks;
463 
464 	/*
465 	 * Update per-CPU and possibly global ticks values.
466 	 */
467 	*t += cnt;
468 	global = ticks;
469 	do {
470 		newticks = *t - global;
471 		if (newticks <= 0) {
472 			if (newticks < -1)
473 				*t = global - 1;
474 			newticks = 0;
475 			break;
476 		}
477 	} while (!atomic_fcmpset_int(&ticks, &global, *t));
478 
479 	/*
480 	 * Run current process's virtual and profile time, as needed.
481 	 */
482 	pstats = p->p_stats;
483 	if (__predict_false(
484 	    timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) ||
485 	    timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)))
486 		hardclock_itimer(td, pstats, cnt, usermode);
487 
488 #ifdef	HWPMC_HOOKS
489 	if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid)))
490 		PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL);
491 	if (td->td_intr_frame != NULL)
492 		PMC_SOFT_CALL_TF( , , clock, hard, td->td_intr_frame);
493 #endif
494 	/* We are in charge to handle this tick duty. */
495 	if (newticks > 0) {
496 		tc_ticktock(newticks);
497 #ifdef DEVICE_POLLING
498 		/* Dangerous and no need to call these things concurrently. */
499 		if (atomic_cmpset_acq_int(&devpoll_run, 0, 1)) {
500 			/* This is very short and quick. */
501 			hardclock_device_poll();
502 			atomic_store_rel_int(&devpoll_run, 0);
503 		}
504 #endif /* DEVICE_POLLING */
505 		if (watchdog_enabled > 0) {
506 			i = atomic_fetchadd_int(&watchdog_ticks, -newticks);
507 			if (i > 0 && i <= newticks)
508 				watchdog_fire();
509 		}
510 		intr_event_handle(clk_intr_event, NULL);
511 	}
512 	if (curcpu == CPU_FIRST())
513 		cpu_tick_calibration();
514 	if (__predict_false(DPCPU_GET(epoch_cb_count)))
515 		GROUPTASK_ENQUEUE(DPCPU_PTR(epoch_cb_task));
516 }
517 
518 void
519 hardclock_sync(int cpu)
520 {
521 	int *t;
522 	KASSERT(!CPU_ABSENT(cpu), ("Absent CPU %d", cpu));
523 	t = DPCPU_ID_PTR(cpu, pcputicks);
524 
525 	*t = ticks;
526 }
527 
528 /*
529  * Compute number of ticks in the specified amount of time.
530  */
531 int
532 tvtohz(struct timeval *tv)
533 {
534 	unsigned long ticks;
535 	long sec, usec;
536 
537 	/*
538 	 * If the number of usecs in the whole seconds part of the time
539 	 * difference fits in a long, then the total number of usecs will
540 	 * fit in an unsigned long.  Compute the total and convert it to
541 	 * ticks, rounding up and adding 1 to allow for the current tick
542 	 * to expire.  Rounding also depends on unsigned long arithmetic
543 	 * to avoid overflow.
544 	 *
545 	 * Otherwise, if the number of ticks in the whole seconds part of
546 	 * the time difference fits in a long, then convert the parts to
547 	 * ticks separately and add, using similar rounding methods and
548 	 * overflow avoidance.  This method would work in the previous
549 	 * case but it is slightly slower and assumes that hz is integral.
550 	 *
551 	 * Otherwise, round the time difference down to the maximum
552 	 * representable value.
553 	 *
554 	 * If ints have 32 bits, then the maximum value for any timeout in
555 	 * 10ms ticks is 248 days.
556 	 */
557 	sec = tv->tv_sec;
558 	usec = tv->tv_usec;
559 	if (usec < 0) {
560 		sec--;
561 		usec += 1000000;
562 	}
563 	if (sec < 0) {
564 #ifdef DIAGNOSTIC
565 		if (usec > 0) {
566 			sec++;
567 			usec -= 1000000;
568 		}
569 		printf("tvotohz: negative time difference %ld sec %ld usec\n",
570 		       sec, usec);
571 #endif
572 		ticks = 1;
573 	} else if (sec <= LONG_MAX / 1000000)
574 		ticks = howmany(sec * 1000000 + (unsigned long)usec, tick) + 1;
575 	else if (sec <= LONG_MAX / hz)
576 		ticks = sec * hz
577 			+ howmany((unsigned long)usec, tick) + 1;
578 	else
579 		ticks = LONG_MAX;
580 	if (ticks > INT_MAX)
581 		ticks = INT_MAX;
582 	return ((int)ticks);
583 }
584 
585 /*
586  * Start profiling on a process.
587  *
588  * Kernel profiling passes proc0 which never exits and hence
589  * keeps the profile clock running constantly.
590  */
591 void
592 startprofclock(struct proc *p)
593 {
594 
595 	PROC_LOCK_ASSERT(p, MA_OWNED);
596 	if (p->p_flag & P_STOPPROF)
597 		return;
598 	if ((p->p_flag & P_PROFIL) == 0) {
599 		p->p_flag |= P_PROFIL;
600 		mtx_lock(&time_lock);
601 		if (++profprocs == 1)
602 			cpu_startprofclock();
603 		mtx_unlock(&time_lock);
604 	}
605 }
606 
607 /*
608  * Stop profiling on a process.
609  */
610 void
611 stopprofclock(struct proc *p)
612 {
613 
614 	PROC_LOCK_ASSERT(p, MA_OWNED);
615 	if (p->p_flag & P_PROFIL) {
616 		if (p->p_profthreads != 0) {
617 			while (p->p_profthreads != 0) {
618 				p->p_flag |= P_STOPPROF;
619 				msleep(&p->p_profthreads, &p->p_mtx, PPAUSE,
620 				    "stopprof", 0);
621 			}
622 		}
623 		if ((p->p_flag & P_PROFIL) == 0)
624 			return;
625 		p->p_flag &= ~P_PROFIL;
626 		mtx_lock(&time_lock);
627 		if (--profprocs == 0)
628 			cpu_stopprofclock();
629 		mtx_unlock(&time_lock);
630 	}
631 }
632 
633 /*
634  * Statistics clock.  Updates rusage information and calls the scheduler
635  * to adjust priorities of the active thread.
636  *
637  * This should be called by all active processors.
638  */
639 void
640 statclock(int cnt, int usermode)
641 {
642 	struct rusage *ru;
643 	struct vmspace *vm;
644 	struct thread *td;
645 	struct proc *p;
646 	long rss;
647 	long *cp_time;
648 	uint64_t runtime, new_switchtime;
649 
650 	td = curthread;
651 	p = td->td_proc;
652 
653 	cp_time = (long *)PCPU_PTR(cp_time);
654 	if (usermode) {
655 		/*
656 		 * Charge the time as appropriate.
657 		 */
658 		td->td_uticks += cnt;
659 		if (p->p_nice > NZERO)
660 			cp_time[CP_NICE] += cnt;
661 		else
662 			cp_time[CP_USER] += cnt;
663 	} else {
664 		/*
665 		 * Came from kernel mode, so we were:
666 		 * - handling an interrupt,
667 		 * - doing syscall or trap work on behalf of the current
668 		 *   user process, or
669 		 * - spinning in the idle loop.
670 		 * Whichever it is, charge the time as appropriate.
671 		 * Note that we charge interrupts to the current process,
672 		 * regardless of whether they are ``for'' that process,
673 		 * so that we know how much of its real time was spent
674 		 * in ``non-process'' (i.e., interrupt) work.
675 		 */
676 		if ((td->td_pflags & TDP_ITHREAD) ||
677 		    td->td_intr_nesting_level >= 2) {
678 			td->td_iticks += cnt;
679 			cp_time[CP_INTR] += cnt;
680 		} else {
681 			td->td_pticks += cnt;
682 			td->td_sticks += cnt;
683 			if (!TD_IS_IDLETHREAD(td))
684 				cp_time[CP_SYS] += cnt;
685 			else
686 				cp_time[CP_IDLE] += cnt;
687 		}
688 	}
689 
690 	/* Update resource usage integrals and maximums. */
691 	MPASS(p->p_vmspace != NULL);
692 	vm = p->p_vmspace;
693 	ru = &td->td_ru;
694 	ru->ru_ixrss += pgtok(vm->vm_tsize) * cnt;
695 	ru->ru_idrss += pgtok(vm->vm_dsize) * cnt;
696 	ru->ru_isrss += pgtok(vm->vm_ssize) * cnt;
697 	rss = pgtok(vmspace_resident_count(vm));
698 	if (ru->ru_maxrss < rss)
699 		ru->ru_maxrss = rss;
700 	KTR_POINT2(KTR_SCHED, "thread", sched_tdname(td), "statclock",
701 	    "prio:%d", td->td_priority, "stathz:%d", (stathz)?stathz:hz);
702 	SDT_PROBE2(sched, , , tick, td, td->td_proc);
703 	thread_lock_flags(td, MTX_QUIET);
704 
705 	/*
706 	 * Compute the amount of time during which the current
707 	 * thread was running, and add that to its total so far.
708 	 */
709 	new_switchtime = cpu_ticks();
710 	runtime = new_switchtime - PCPU_GET(switchtime);
711 	td->td_runtime += runtime;
712 	td->td_incruntime += runtime;
713 	PCPU_SET(switchtime, new_switchtime);
714 
715 	sched_clock(td, cnt);
716 	thread_unlock(td);
717 #ifdef HWPMC_HOOKS
718 	if (td->td_intr_frame != NULL)
719 		PMC_SOFT_CALL_TF( , , clock, stat, td->td_intr_frame);
720 #endif
721 }
722 
723 void
724 profclock(int cnt, int usermode, uintfptr_t pc)
725 {
726 	struct thread *td;
727 #ifdef GPROF
728 	struct gmonparam *g;
729 	uintfptr_t i;
730 #endif
731 
732 	td = curthread;
733 	if (usermode) {
734 		/*
735 		 * Came from user mode; CPU was in user state.
736 		 * If this process is being profiled, record the tick.
737 		 * if there is no related user location yet, don't
738 		 * bother trying to count it.
739 		 */
740 		if (td->td_proc->p_flag & P_PROFIL)
741 			addupc_intr(td, pc, cnt);
742 	}
743 #ifdef GPROF
744 	else {
745 		/*
746 		 * Kernel statistics are just like addupc_intr, only easier.
747 		 */
748 		g = &_gmonparam;
749 		if (g->state == GMON_PROF_ON && pc >= g->lowpc) {
750 			i = PC_TO_I(g, pc);
751 			if (i < g->textsize) {
752 				KCOUNT(g, i) += cnt;
753 			}
754 		}
755 	}
756 #endif
757 #ifdef HWPMC_HOOKS
758 	if (td->td_intr_frame != NULL)
759 		PMC_SOFT_CALL_TF( , , clock, prof, td->td_intr_frame);
760 #endif
761 }
762 
763 /*
764  * Return information about system clocks.
765  */
766 static int
767 sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS)
768 {
769 	struct clockinfo clkinfo;
770 	/*
771 	 * Construct clockinfo structure.
772 	 */
773 	bzero(&clkinfo, sizeof(clkinfo));
774 	clkinfo.hz = hz;
775 	clkinfo.tick = tick;
776 	clkinfo.profhz = profhz;
777 	clkinfo.stathz = stathz ? stathz : hz;
778 	return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req));
779 }
780 
781 SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate,
782 	CTLTYPE_STRUCT|CTLFLAG_RD|CTLFLAG_MPSAFE,
783 	0, 0, sysctl_kern_clockrate, "S,clockinfo",
784 	"Rate and period of various kernel clocks");
785 
786 static void
787 watchdog_config(void *unused __unused, u_int cmd, int *error)
788 {
789 	u_int u;
790 
791 	u = cmd & WD_INTERVAL;
792 	if (u >= WD_TO_1SEC) {
793 		watchdog_ticks = (1 << (u - WD_TO_1SEC)) * hz;
794 		watchdog_enabled = 1;
795 		*error = 0;
796 	} else {
797 		watchdog_enabled = 0;
798 	}
799 }
800 
801 /*
802  * Handle a watchdog timeout by dumping interrupt information and
803  * then either dropping to DDB or panicking.
804  */
805 static void
806 watchdog_fire(void)
807 {
808 	int nintr;
809 	uint64_t inttotal;
810 	u_long *curintr;
811 	char *curname;
812 
813 	curintr = intrcnt;
814 	curname = intrnames;
815 	inttotal = 0;
816 	nintr = sintrcnt / sizeof(u_long);
817 
818 	printf("interrupt                   total\n");
819 	while (--nintr >= 0) {
820 		if (*curintr)
821 			printf("%-12s %20lu\n", curname, *curintr);
822 		curname += strlen(curname) + 1;
823 		inttotal += *curintr++;
824 	}
825 	printf("Total        %20ju\n", (uintmax_t)inttotal);
826 
827 #if defined(KDB) && !defined(KDB_UNATTENDED)
828 	kdb_backtrace();
829 	kdb_enter(KDB_WHY_WATCHDOG, "watchdog timeout");
830 #else
831 	panic("watchdog timeout");
832 #endif
833 }
834