xref: /freebsd/sys/kern/kern_clock.c (revision 5ac01ce026aa871e24065f931b5b5c36024c96f9)
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 		/*
234 		 * Accordingly with provided thresholds, this thread is stuck
235 		 * for too long on a sleepqueue.
236 		 * However, being on a sleepqueue, we might still check for the
237 		 * blessed list.
238 		 */
239 		for (i = 0; blessed[i] != NULL; i++)
240 			if (!strcmp(blessed[i], td->td_wmesg))
241 				return;
242 
243 		panic("%s: possible deadlock detected for %p (%s), "
244 		    "blocked for %d ticks\n", __func__,
245 		    td, sched_tdname(td), tticks);
246 	}
247 }
248 
249 static void
250 deadlkres(void)
251 {
252 	struct proc *p;
253 	struct thread *td;
254 	int blkticks, slpticks, tryl;
255 
256 	tryl = 0;
257 	for (;;) {
258 		blkticks = blktime_threshold * hz;
259 		slpticks = slptime_threshold * hz;
260 
261 		/*
262 		 * Avoid to sleep on the sx_lock in order to avoid a
263 		 * possible priority inversion problem leading to
264 		 * starvation.
265 		 * If the lock can't be held after 100 tries, panic.
266 		 */
267 		if (!sx_try_slock(&allproc_lock)) {
268 			if (tryl > 100)
269 				panic("%s: possible deadlock detected "
270 				    "on allproc_lock\n", __func__);
271 			tryl++;
272 			pause("allproc", sleepfreq * hz);
273 			continue;
274 		}
275 		tryl = 0;
276 		FOREACH_PROC_IN_SYSTEM(p) {
277 			PROC_LOCK(p);
278 			if (p->p_state == PRS_NEW) {
279 				PROC_UNLOCK(p);
280 				continue;
281 			}
282 			FOREACH_THREAD_IN_PROC(p, td) {
283 				thread_lock(td);
284 				if (TD_ON_LOCK(td))
285 					deadlres_td_on_lock(p, td,
286 					    blkticks);
287 				else if (TD_IS_SLEEPING(td))
288 					deadlres_td_sleep_q(p, td,
289 					    slpticks);
290 				thread_unlock(td);
291 			}
292 			PROC_UNLOCK(p);
293 		}
294 		sx_sunlock(&allproc_lock);
295 
296 		/* Sleep for sleepfreq seconds. */
297 		pause("-", sleepfreq * hz);
298 	}
299 }
300 
301 static struct kthread_desc deadlkres_kd = {
302 	"deadlkres",
303 	deadlkres,
304 	(struct thread **)NULL
305 };
306 
307 SYSINIT(deadlkres, SI_SUB_CLOCKS, SI_ORDER_ANY, kthread_start, &deadlkres_kd);
308 
309 static SYSCTL_NODE(_debug, OID_AUTO, deadlkres, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
310     "Deadlock resolver");
311 SYSCTL_INT(_debug_deadlkres, OID_AUTO, slptime_threshold, CTLFLAG_RW,
312     &slptime_threshold, 0,
313     "Number of seconds within is valid to sleep on a sleepqueue");
314 SYSCTL_INT(_debug_deadlkres, OID_AUTO, blktime_threshold, CTLFLAG_RW,
315     &blktime_threshold, 0,
316     "Number of seconds within is valid to block on a turnstile");
317 SYSCTL_INT(_debug_deadlkres, OID_AUTO, sleepfreq, CTLFLAG_RW, &sleepfreq, 0,
318     "Number of seconds between any deadlock resolver thread run");
319 #endif	/* DEADLKRES */
320 
321 void
322 read_cpu_time(long *cp_time)
323 {
324 	struct pcpu *pc;
325 	int i, j;
326 
327 	/* Sum up global cp_time[]. */
328 	bzero(cp_time, sizeof(long) * CPUSTATES);
329 	CPU_FOREACH(i) {
330 		pc = pcpu_find(i);
331 		for (j = 0; j < CPUSTATES; j++)
332 			cp_time[j] += pc->pc_cp_time[j];
333 	}
334 }
335 
336 #include <sys/watchdog.h>
337 
338 static int watchdog_ticks;
339 static int watchdog_enabled;
340 static void watchdog_fire(void);
341 static void watchdog_config(void *, u_int, int *);
342 
343 static void
344 watchdog_attach(void)
345 {
346 	EVENTHANDLER_REGISTER(watchdog_list, watchdog_config, NULL, 0);
347 }
348 
349 /*
350  * Clock handling routines.
351  *
352  * This code is written to operate with two timers that run independently of
353  * each other.
354  *
355  * The main timer, running hz times per second, is used to trigger interval
356  * timers, timeouts and rescheduling as needed.
357  *
358  * The second timer handles kernel and user profiling,
359  * and does resource use estimation.  If the second timer is programmable,
360  * it is randomized to avoid aliasing between the two clocks.  For example,
361  * the randomization prevents an adversary from always giving up the cpu
362  * just before its quantum expires.  Otherwise, it would never accumulate
363  * cpu ticks.  The mean frequency of the second timer is stathz.
364  *
365  * If no second timer exists, stathz will be zero; in this case we drive
366  * profiling and statistics off the main clock.  This WILL NOT be accurate;
367  * do not do it unless absolutely necessary.
368  *
369  * The statistics clock may (or may not) be run at a higher rate while
370  * profiling.  This profile clock runs at profhz.  We require that profhz
371  * be an integral multiple of stathz.
372  *
373  * If the statistics clock is running fast, it must be divided by the ratio
374  * profhz/stathz for statistics.  (For profiling, every tick counts.)
375  *
376  * Time-of-day is maintained using a "timecounter", which may or may
377  * not be related to the hardware generating the above mentioned
378  * interrupts.
379  */
380 
381 int	stathz;
382 int	profhz;
383 int	profprocs;
384 volatile int	ticks;
385 int	psratio;
386 
387 DPCPU_DEFINE_STATIC(int, pcputicks);	/* Per-CPU version of ticks. */
388 #ifdef DEVICE_POLLING
389 static int devpoll_run = 0;
390 #endif
391 
392 /*
393  * Initialize clock frequencies and start both clocks running.
394  */
395 /* ARGSUSED*/
396 static void
397 initclocks(void *dummy)
398 {
399 	int i;
400 
401 	/*
402 	 * Set divisors to 1 (normal case) and let the machine-specific
403 	 * code do its bit.
404 	 */
405 	mtx_init(&time_lock, "time lock", NULL, MTX_DEF);
406 	cpu_initclocks();
407 
408 	/*
409 	 * Compute profhz/stathz, and fix profhz if needed.
410 	 */
411 	i = stathz ? stathz : hz;
412 	if (profhz == 0)
413 		profhz = i;
414 	psratio = profhz / i;
415 
416 #ifdef SW_WATCHDOG
417 	/* Enable hardclock watchdog now, even if a hardware watchdog exists. */
418 	watchdog_attach();
419 #else
420 	/* Volunteer to run a software watchdog. */
421 	if (wdog_software_attach == NULL)
422 		wdog_software_attach = watchdog_attach;
423 #endif
424 }
425 
426 static __noinline void
427 hardclock_itimer(struct thread *td, struct pstats *pstats, int cnt, int usermode)
428 {
429 	struct proc *p;
430 	int flags;
431 
432 	flags = 0;
433 	p = td->td_proc;
434 	if (usermode &&
435 	    timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) {
436 		PROC_ITIMLOCK(p);
437 		if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL],
438 		    tick * cnt) == 0)
439 			flags |= TDF_ALRMPEND | TDF_ASTPENDING;
440 		PROC_ITIMUNLOCK(p);
441 	}
442 	if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) {
443 		PROC_ITIMLOCK(p);
444 		if (itimerdecr(&pstats->p_timer[ITIMER_PROF],
445 		    tick * cnt) == 0)
446 			flags |= TDF_PROFPEND | TDF_ASTPENDING;
447 		PROC_ITIMUNLOCK(p);
448 	}
449 	if (flags != 0) {
450 		thread_lock(td);
451 		td->td_flags |= flags;
452 		thread_unlock(td);
453 	}
454 }
455 
456 void
457 hardclock(int cnt, int usermode)
458 {
459 	struct pstats *pstats;
460 	struct thread *td = curthread;
461 	struct proc *p = td->td_proc;
462 	int *t = DPCPU_PTR(pcputicks);
463 	int global, i, newticks;
464 
465 	/*
466 	 * Update per-CPU and possibly global ticks values.
467 	 */
468 	*t += cnt;
469 	global = ticks;
470 	do {
471 		newticks = *t - global;
472 		if (newticks <= 0) {
473 			if (newticks < -1)
474 				*t = global - 1;
475 			newticks = 0;
476 			break;
477 		}
478 	} while (!atomic_fcmpset_int(&ticks, &global, *t));
479 
480 	/*
481 	 * Run current process's virtual and profile time, as needed.
482 	 */
483 	pstats = p->p_stats;
484 	if (__predict_false(
485 	    timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) ||
486 	    timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)))
487 		hardclock_itimer(td, pstats, cnt, usermode);
488 
489 #ifdef	HWPMC_HOOKS
490 	if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid)))
491 		PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL);
492 	if (td->td_intr_frame != NULL)
493 		PMC_SOFT_CALL_TF( , , clock, hard, td->td_intr_frame);
494 #endif
495 	/* We are in charge to handle this tick duty. */
496 	if (newticks > 0) {
497 		tc_ticktock(newticks);
498 #ifdef DEVICE_POLLING
499 		/* Dangerous and no need to call these things concurrently. */
500 		if (atomic_cmpset_acq_int(&devpoll_run, 0, 1)) {
501 			/* This is very short and quick. */
502 			hardclock_device_poll();
503 			atomic_store_rel_int(&devpoll_run, 0);
504 		}
505 #endif /* DEVICE_POLLING */
506 		if (watchdog_enabled > 0) {
507 			i = atomic_fetchadd_int(&watchdog_ticks, -newticks);
508 			if (i > 0 && i <= newticks)
509 				watchdog_fire();
510 		}
511 		intr_event_handle(clk_intr_event, NULL);
512 	}
513 	if (curcpu == CPU_FIRST())
514 		cpu_tick_calibration();
515 	if (__predict_false(DPCPU_GET(epoch_cb_count)))
516 		GROUPTASK_ENQUEUE(DPCPU_PTR(epoch_cb_task));
517 }
518 
519 void
520 hardclock_sync(int cpu)
521 {
522 	int *t;
523 	KASSERT(!CPU_ABSENT(cpu), ("Absent CPU %d", cpu));
524 	t = DPCPU_ID_PTR(cpu, pcputicks);
525 
526 	*t = ticks;
527 }
528 
529 /*
530  * Compute number of ticks in the specified amount of time.
531  */
532 int
533 tvtohz(struct timeval *tv)
534 {
535 	unsigned long ticks;
536 	long sec, usec;
537 
538 	/*
539 	 * If the number of usecs in the whole seconds part of the time
540 	 * difference fits in a long, then the total number of usecs will
541 	 * fit in an unsigned long.  Compute the total and convert it to
542 	 * ticks, rounding up and adding 1 to allow for the current tick
543 	 * to expire.  Rounding also depends on unsigned long arithmetic
544 	 * to avoid overflow.
545 	 *
546 	 * Otherwise, if the number of ticks in the whole seconds part of
547 	 * the time difference fits in a long, then convert the parts to
548 	 * ticks separately and add, using similar rounding methods and
549 	 * overflow avoidance.  This method would work in the previous
550 	 * case but it is slightly slower and assumes that hz is integral.
551 	 *
552 	 * Otherwise, round the time difference down to the maximum
553 	 * representable value.
554 	 *
555 	 * If ints have 32 bits, then the maximum value for any timeout in
556 	 * 10ms ticks is 248 days.
557 	 */
558 	sec = tv->tv_sec;
559 	usec = tv->tv_usec;
560 	if (usec < 0) {
561 		sec--;
562 		usec += 1000000;
563 	}
564 	if (sec < 0) {
565 #ifdef DIAGNOSTIC
566 		if (usec > 0) {
567 			sec++;
568 			usec -= 1000000;
569 		}
570 		printf("tvotohz: negative time difference %ld sec %ld usec\n",
571 		       sec, usec);
572 #endif
573 		ticks = 1;
574 	} else if (sec <= LONG_MAX / 1000000)
575 		ticks = howmany(sec * 1000000 + (unsigned long)usec, tick) + 1;
576 	else if (sec <= LONG_MAX / hz)
577 		ticks = sec * hz
578 			+ howmany((unsigned long)usec, tick) + 1;
579 	else
580 		ticks = LONG_MAX;
581 	if (ticks > INT_MAX)
582 		ticks = INT_MAX;
583 	return ((int)ticks);
584 }
585 
586 /*
587  * Start profiling on a process.
588  *
589  * Kernel profiling passes proc0 which never exits and hence
590  * keeps the profile clock running constantly.
591  */
592 void
593 startprofclock(struct proc *p)
594 {
595 
596 	PROC_LOCK_ASSERT(p, MA_OWNED);
597 	if (p->p_flag & P_STOPPROF)
598 		return;
599 	if ((p->p_flag & P_PROFIL) == 0) {
600 		p->p_flag |= P_PROFIL;
601 		mtx_lock(&time_lock);
602 		if (++profprocs == 1)
603 			cpu_startprofclock();
604 		mtx_unlock(&time_lock);
605 	}
606 }
607 
608 /*
609  * Stop profiling on a process.
610  */
611 void
612 stopprofclock(struct proc *p)
613 {
614 
615 	PROC_LOCK_ASSERT(p, MA_OWNED);
616 	if (p->p_flag & P_PROFIL) {
617 		if (p->p_profthreads != 0) {
618 			while (p->p_profthreads != 0) {
619 				p->p_flag |= P_STOPPROF;
620 				msleep(&p->p_profthreads, &p->p_mtx, PPAUSE,
621 				    "stopprof", 0);
622 			}
623 		}
624 		if ((p->p_flag & P_PROFIL) == 0)
625 			return;
626 		p->p_flag &= ~P_PROFIL;
627 		mtx_lock(&time_lock);
628 		if (--profprocs == 0)
629 			cpu_stopprofclock();
630 		mtx_unlock(&time_lock);
631 	}
632 }
633 
634 /*
635  * Statistics clock.  Updates rusage information and calls the scheduler
636  * to adjust priorities of the active thread.
637  *
638  * This should be called by all active processors.
639  */
640 void
641 statclock(int cnt, int usermode)
642 {
643 	struct rusage *ru;
644 	struct vmspace *vm;
645 	struct thread *td;
646 	struct proc *p;
647 	long rss;
648 	long *cp_time;
649 	uint64_t runtime, new_switchtime;
650 
651 	td = curthread;
652 	p = td->td_proc;
653 
654 	cp_time = (long *)PCPU_PTR(cp_time);
655 	if (usermode) {
656 		/*
657 		 * Charge the time as appropriate.
658 		 */
659 		td->td_uticks += cnt;
660 		if (p->p_nice > NZERO)
661 			cp_time[CP_NICE] += cnt;
662 		else
663 			cp_time[CP_USER] += cnt;
664 	} else {
665 		/*
666 		 * Came from kernel mode, so we were:
667 		 * - handling an interrupt,
668 		 * - doing syscall or trap work on behalf of the current
669 		 *   user process, or
670 		 * - spinning in the idle loop.
671 		 * Whichever it is, charge the time as appropriate.
672 		 * Note that we charge interrupts to the current process,
673 		 * regardless of whether they are ``for'' that process,
674 		 * so that we know how much of its real time was spent
675 		 * in ``non-process'' (i.e., interrupt) work.
676 		 */
677 		if ((td->td_pflags & TDP_ITHREAD) ||
678 		    td->td_intr_nesting_level >= 2) {
679 			td->td_iticks += cnt;
680 			cp_time[CP_INTR] += cnt;
681 		} else {
682 			td->td_pticks += cnt;
683 			td->td_sticks += cnt;
684 			if (!TD_IS_IDLETHREAD(td))
685 				cp_time[CP_SYS] += cnt;
686 			else
687 				cp_time[CP_IDLE] += cnt;
688 		}
689 	}
690 
691 	/* Update resource usage integrals and maximums. */
692 	MPASS(p->p_vmspace != NULL);
693 	vm = p->p_vmspace;
694 	ru = &td->td_ru;
695 	ru->ru_ixrss += pgtok(vm->vm_tsize) * cnt;
696 	ru->ru_idrss += pgtok(vm->vm_dsize) * cnt;
697 	ru->ru_isrss += pgtok(vm->vm_ssize) * cnt;
698 	rss = pgtok(vmspace_resident_count(vm));
699 	if (ru->ru_maxrss < rss)
700 		ru->ru_maxrss = rss;
701 	KTR_POINT2(KTR_SCHED, "thread", sched_tdname(td), "statclock",
702 	    "prio:%d", td->td_priority, "stathz:%d", (stathz)?stathz:hz);
703 	SDT_PROBE2(sched, , , tick, td, td->td_proc);
704 	thread_lock_flags(td, MTX_QUIET);
705 
706 	/*
707 	 * Compute the amount of time during which the current
708 	 * thread was running, and add that to its total so far.
709 	 */
710 	new_switchtime = cpu_ticks();
711 	runtime = new_switchtime - PCPU_GET(switchtime);
712 	td->td_runtime += runtime;
713 	td->td_incruntime += runtime;
714 	PCPU_SET(switchtime, new_switchtime);
715 
716 	sched_clock(td, cnt);
717 	thread_unlock(td);
718 #ifdef HWPMC_HOOKS
719 	if (td->td_intr_frame != NULL)
720 		PMC_SOFT_CALL_TF( , , clock, stat, td->td_intr_frame);
721 #endif
722 }
723 
724 void
725 profclock(int cnt, int usermode, uintfptr_t pc)
726 {
727 	struct thread *td;
728 #ifdef GPROF
729 	struct gmonparam *g;
730 	uintfptr_t i;
731 #endif
732 
733 	td = curthread;
734 	if (usermode) {
735 		/*
736 		 * Came from user mode; CPU was in user state.
737 		 * If this process is being profiled, record the tick.
738 		 * if there is no related user location yet, don't
739 		 * bother trying to count it.
740 		 */
741 		if (td->td_proc->p_flag & P_PROFIL)
742 			addupc_intr(td, pc, cnt);
743 	}
744 #ifdef GPROF
745 	else {
746 		/*
747 		 * Kernel statistics are just like addupc_intr, only easier.
748 		 */
749 		g = &_gmonparam;
750 		if (g->state == GMON_PROF_ON && pc >= g->lowpc) {
751 			i = PC_TO_I(g, pc);
752 			if (i < g->textsize) {
753 				KCOUNT(g, i) += cnt;
754 			}
755 		}
756 	}
757 #endif
758 #ifdef HWPMC_HOOKS
759 	if (td->td_intr_frame != NULL)
760 		PMC_SOFT_CALL_TF( , , clock, prof, td->td_intr_frame);
761 #endif
762 }
763 
764 /*
765  * Return information about system clocks.
766  */
767 static int
768 sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS)
769 {
770 	struct clockinfo clkinfo;
771 	/*
772 	 * Construct clockinfo structure.
773 	 */
774 	bzero(&clkinfo, sizeof(clkinfo));
775 	clkinfo.hz = hz;
776 	clkinfo.tick = tick;
777 	clkinfo.profhz = profhz;
778 	clkinfo.stathz = stathz ? stathz : hz;
779 	return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req));
780 }
781 
782 SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate,
783 	CTLTYPE_STRUCT|CTLFLAG_RD|CTLFLAG_MPSAFE,
784 	0, 0, sysctl_kern_clockrate, "S,clockinfo",
785 	"Rate and period of various kernel clocks");
786 
787 static void
788 watchdog_config(void *unused __unused, u_int cmd, int *error)
789 {
790 	u_int u;
791 
792 	u = cmd & WD_INTERVAL;
793 	if (u >= WD_TO_1SEC) {
794 		watchdog_ticks = (1 << (u - WD_TO_1SEC)) * hz;
795 		watchdog_enabled = 1;
796 		*error = 0;
797 	} else {
798 		watchdog_enabled = 0;
799 	}
800 }
801 
802 /*
803  * Handle a watchdog timeout by dumping interrupt information and
804  * then either dropping to DDB or panicking.
805  */
806 static void
807 watchdog_fire(void)
808 {
809 	int nintr;
810 	uint64_t inttotal;
811 	u_long *curintr;
812 	char *curname;
813 
814 	curintr = intrcnt;
815 	curname = intrnames;
816 	inttotal = 0;
817 	nintr = sintrcnt / sizeof(u_long);
818 
819 	printf("interrupt                   total\n");
820 	while (--nintr >= 0) {
821 		if (*curintr)
822 			printf("%-12s %20lu\n", curname, *curintr);
823 		curname += strlen(curname) + 1;
824 		inttotal += *curintr++;
825 	}
826 	printf("Total        %20ju\n", (uintmax_t)inttotal);
827 
828 #if defined(KDB) && !defined(KDB_UNATTENDED)
829 	kdb_backtrace();
830 	kdb_enter(KDB_WHY_WATCHDOG, "watchdog timeout");
831 #else
832 	panic("watchdog timeout");
833 #endif
834 }
835