xref: /freebsd/sys/kern/kern_clock.c (revision 4abd7edcbde21ba7a089c7d1a0bba8f87ebece06)
1 /*-
2  * Copyright (c) 1982, 1986, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)kern_clock.c	8.5 (Berkeley) 1/21/94
35  */
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 #include "opt_kdb.h"
41 #include "opt_device_polling.h"
42 #include "opt_hwpmc_hooks.h"
43 #include "opt_ntp.h"
44 #include "opt_watchdog.h"
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/callout.h>
49 #include <sys/kdb.h>
50 #include <sys/kernel.h>
51 #include <sys/kthread.h>
52 #include <sys/ktr.h>
53 #include <sys/lock.h>
54 #include <sys/mutex.h>
55 #include <sys/proc.h>
56 #include <sys/resource.h>
57 #include <sys/resourcevar.h>
58 #include <sys/sched.h>
59 #include <sys/sdt.h>
60 #include <sys/signalvar.h>
61 #include <sys/sleepqueue.h>
62 #include <sys/smp.h>
63 #include <vm/vm.h>
64 #include <vm/pmap.h>
65 #include <vm/vm_map.h>
66 #include <sys/sysctl.h>
67 #include <sys/bus.h>
68 #include <sys/interrupt.h>
69 #include <sys/limits.h>
70 #include <sys/timetc.h>
71 
72 #ifdef GPROF
73 #include <sys/gmon.h>
74 #endif
75 
76 #ifdef HWPMC_HOOKS
77 #include <sys/pmckern.h>
78 PMC_SOFT_DEFINE( , , clock, hard);
79 PMC_SOFT_DEFINE( , , clock, stat);
80 PMC_SOFT_DEFINE_EX( , , clock, prof, \
81     cpu_startprofclock, cpu_stopprofclock);
82 #endif
83 
84 #ifdef DEVICE_POLLING
85 extern void hardclock_device_poll(void);
86 #endif /* DEVICE_POLLING */
87 
88 static void initclocks(void *dummy);
89 SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL);
90 
91 /* Spin-lock protecting profiling statistics. */
92 static struct mtx time_lock;
93 
94 SDT_PROVIDER_DECLARE(sched);
95 SDT_PROBE_DEFINE2(sched, , , tick, "struct thread *", "struct proc *");
96 
97 static int
98 sysctl_kern_cp_time(SYSCTL_HANDLER_ARGS)
99 {
100 	int error;
101 	long cp_time[CPUSTATES];
102 #ifdef SCTL_MASK32
103 	int i;
104 	unsigned int cp_time32[CPUSTATES];
105 #endif
106 
107 	read_cpu_time(cp_time);
108 #ifdef SCTL_MASK32
109 	if (req->flags & SCTL_MASK32) {
110 		if (!req->oldptr)
111 			return SYSCTL_OUT(req, 0, sizeof(cp_time32));
112 		for (i = 0; i < CPUSTATES; i++)
113 			cp_time32[i] = (unsigned int)cp_time[i];
114 		error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32));
115 	} else
116 #endif
117 	{
118 		if (!req->oldptr)
119 			return SYSCTL_OUT(req, 0, sizeof(cp_time));
120 		error = SYSCTL_OUT(req, cp_time, sizeof(cp_time));
121 	}
122 	return error;
123 }
124 
125 SYSCTL_PROC(_kern, OID_AUTO, cp_time, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE,
126     0,0, sysctl_kern_cp_time, "LU", "CPU time statistics");
127 
128 static long empty[CPUSTATES];
129 
130 static int
131 sysctl_kern_cp_times(SYSCTL_HANDLER_ARGS)
132 {
133 	struct pcpu *pcpu;
134 	int error;
135 	int c;
136 	long *cp_time;
137 #ifdef SCTL_MASK32
138 	unsigned int cp_time32[CPUSTATES];
139 	int i;
140 #endif
141 
142 	if (!req->oldptr) {
143 #ifdef SCTL_MASK32
144 		if (req->flags & SCTL_MASK32)
145 			return SYSCTL_OUT(req, 0, sizeof(cp_time32) * (mp_maxid + 1));
146 		else
147 #endif
148 			return SYSCTL_OUT(req, 0, sizeof(long) * CPUSTATES * (mp_maxid + 1));
149 	}
150 	for (error = 0, c = 0; error == 0 && c <= mp_maxid; c++) {
151 		if (!CPU_ABSENT(c)) {
152 			pcpu = pcpu_find(c);
153 			cp_time = pcpu->pc_cp_time;
154 		} else {
155 			cp_time = empty;
156 		}
157 #ifdef SCTL_MASK32
158 		if (req->flags & SCTL_MASK32) {
159 			for (i = 0; i < CPUSTATES; i++)
160 				cp_time32[i] = (unsigned int)cp_time[i];
161 			error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32));
162 		} else
163 #endif
164 			error = SYSCTL_OUT(req, cp_time, sizeof(long) * CPUSTATES);
165 	}
166 	return error;
167 }
168 
169 SYSCTL_PROC(_kern, OID_AUTO, cp_times, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE,
170     0,0, sysctl_kern_cp_times, "LU", "per-CPU time statistics");
171 
172 #ifdef DEADLKRES
173 static const char *blessed[] = {
174 	"getblk",
175 	"so_snd_sx",
176 	"so_rcv_sx",
177 	NULL
178 };
179 static int slptime_threshold = 1800;
180 static int blktime_threshold = 900;
181 static int sleepfreq = 3;
182 
183 static void
184 deadlkres(void)
185 {
186 	struct proc *p;
187 	struct thread *td;
188 	void *wchan;
189 	int blkticks, i, slpticks, slptype, tryl, tticks;
190 
191 	tryl = 0;
192 	for (;;) {
193 		blkticks = blktime_threshold * hz;
194 		slpticks = slptime_threshold * hz;
195 
196 		/*
197 		 * Avoid to sleep on the sx_lock in order to avoid a possible
198 		 * priority inversion problem leading to starvation.
199 		 * If the lock can't be held after 100 tries, panic.
200 		 */
201 		if (!sx_try_slock(&allproc_lock)) {
202 			if (tryl > 100)
203 		panic("%s: possible deadlock detected on allproc_lock\n",
204 				    __func__);
205 			tryl++;
206 			pause("allproc", sleepfreq * hz);
207 			continue;
208 		}
209 		tryl = 0;
210 		FOREACH_PROC_IN_SYSTEM(p) {
211 			PROC_LOCK(p);
212 			if (p->p_state == PRS_NEW) {
213 				PROC_UNLOCK(p);
214 				continue;
215 			}
216 			FOREACH_THREAD_IN_PROC(p, td) {
217 
218 				thread_lock(td);
219 				if (TD_ON_LOCK(td)) {
220 
221 					/*
222 					 * The thread should be blocked on a
223 					 * turnstile, simply check if the
224 					 * turnstile channel is in good state.
225 					 */
226 					MPASS(td->td_blocked != NULL);
227 
228 					tticks = ticks - td->td_blktick;
229 					thread_unlock(td);
230 					if (tticks > blkticks) {
231 
232 						/*
233 						 * Accordingly with provided
234 						 * thresholds, this thread is
235 						 * stuck for too long on a
236 						 * turnstile.
237 						 */
238 						PROC_UNLOCK(p);
239 						sx_sunlock(&allproc_lock);
240 	panic("%s: possible deadlock detected for %p, blocked for %d ticks\n",
241 						    __func__, td, tticks);
242 					}
243 				} else if (TD_IS_SLEEPING(td) &&
244 				    TD_ON_SLEEPQ(td)) {
245 
246 					/*
247 					 * Check if the thread is sleeping on a
248 					 * lock, otherwise skip the check.
249 					 * Drop the thread lock in order to
250 					 * avoid a LOR with the sleepqueue
251 					 * spinlock.
252 					 */
253 					wchan = td->td_wchan;
254 					tticks = ticks - td->td_slptick;
255 					thread_unlock(td);
256 					slptype = sleepq_type(wchan);
257 					if ((slptype == SLEEPQ_SX ||
258 					    slptype == SLEEPQ_LK) &&
259 					    tticks > slpticks) {
260 
261 						/*
262 						 * Accordingly with provided
263 						 * thresholds, this thread is
264 						 * stuck for too long on a
265 						 * sleepqueue.
266 						 * However, being on a
267 						 * sleepqueue, we might still
268 						 * check for the blessed
269 						 * list.
270 						 */
271 						tryl = 0;
272 						for (i = 0; blessed[i] != NULL;
273 						    i++) {
274 							if (!strcmp(blessed[i],
275 							    td->td_wmesg)) {
276 								tryl = 1;
277 								break;
278 							}
279 						}
280 						if (tryl != 0) {
281 							tryl = 0;
282 							continue;
283 						}
284 						PROC_UNLOCK(p);
285 						sx_sunlock(&allproc_lock);
286 	panic("%s: possible deadlock detected for %p, blocked for %d ticks\n",
287 						    __func__, td, tticks);
288 					}
289 				} else
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, 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 #ifdef SW_WATCHDOG
337 #include <sys/watchdog.h>
338 
339 static int watchdog_ticks;
340 static int watchdog_enabled;
341 static void watchdog_fire(void);
342 static void watchdog_config(void *, u_int, int *);
343 #endif /* SW_WATCHDOG */
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 static DPCPU_DEFINE(int, pcputicks);	/* Per-CPU version of ticks. */
384 static int global_hardclock_run = 0;
385 
386 /*
387  * Initialize clock frequencies and start both clocks running.
388  */
389 /* ARGSUSED*/
390 static void
391 initclocks(dummy)
392 	void *dummy;
393 {
394 	register int i;
395 
396 	/*
397 	 * Set divisors to 1 (normal case) and let the machine-specific
398 	 * code do its bit.
399 	 */
400 	mtx_init(&time_lock, "time lock", NULL, MTX_DEF);
401 	cpu_initclocks();
402 
403 	/*
404 	 * Compute profhz/stathz, and fix profhz if needed.
405 	 */
406 	i = stathz ? stathz : hz;
407 	if (profhz == 0)
408 		profhz = i;
409 	psratio = profhz / i;
410 #ifdef SW_WATCHDOG
411 	EVENTHANDLER_REGISTER(watchdog_list, watchdog_config, NULL, 0);
412 #endif
413 }
414 
415 /*
416  * Each time the real-time timer fires, this function is called on all CPUs.
417  * Note that hardclock() calls hardclock_cpu() for the boot CPU, so only
418  * the other CPUs in the system need to call this function.
419  */
420 void
421 hardclock_cpu(int usermode)
422 {
423 	struct pstats *pstats;
424 	struct thread *td = curthread;
425 	struct proc *p = td->td_proc;
426 	int flags;
427 
428 	/*
429 	 * Run current process's virtual and profile time, as needed.
430 	 */
431 	pstats = p->p_stats;
432 	flags = 0;
433 	if (usermode &&
434 	    timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) {
435 		PROC_SLOCK(p);
436 		if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0)
437 			flags |= TDF_ALRMPEND | TDF_ASTPENDING;
438 		PROC_SUNLOCK(p);
439 	}
440 	if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) {
441 		PROC_SLOCK(p);
442 		if (itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0)
443 			flags |= TDF_PROFPEND | TDF_ASTPENDING;
444 		PROC_SUNLOCK(p);
445 	}
446 	thread_lock(td);
447 	sched_tick(1);
448 	td->td_flags |= flags;
449 	thread_unlock(td);
450 
451 #ifdef HWPMC_HOOKS
452 	if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid)))
453 		PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL);
454 	if (td->td_intr_frame != NULL)
455 		PMC_SOFT_CALL_TF( , , clock, hard, td->td_intr_frame);
456 #endif
457 	callout_process(sbinuptime());
458 }
459 
460 /*
461  * The real-time timer, interrupting hz times per second.
462  */
463 void
464 hardclock(int usermode, uintfptr_t pc)
465 {
466 
467 	atomic_add_int(&ticks, 1);
468 	hardclock_cpu(usermode);
469 	tc_ticktock(1);
470 	cpu_tick_calibration();
471 	/*
472 	 * If no separate statistics clock is available, run it from here.
473 	 *
474 	 * XXX: this only works for UP
475 	 */
476 	if (stathz == 0) {
477 		profclock(usermode, pc);
478 		statclock(usermode);
479 	}
480 #ifdef DEVICE_POLLING
481 	hardclock_device_poll();	/* this is very short and quick */
482 #endif /* DEVICE_POLLING */
483 #ifdef SW_WATCHDOG
484 	if (watchdog_enabled > 0 && --watchdog_ticks <= 0)
485 		watchdog_fire();
486 #endif /* SW_WATCHDOG */
487 }
488 
489 void
490 hardclock_cnt(int cnt, int usermode)
491 {
492 	struct pstats *pstats;
493 	struct thread *td = curthread;
494 	struct proc *p = td->td_proc;
495 	int *t = DPCPU_PTR(pcputicks);
496 	int flags, global, newticks;
497 #ifdef SW_WATCHDOG
498 	int i;
499 #endif /* SW_WATCHDOG */
500 
501 	/*
502 	 * Update per-CPU and possibly global ticks values.
503 	 */
504 	*t += cnt;
505 	do {
506 		global = ticks;
507 		newticks = *t - global;
508 		if (newticks <= 0) {
509 			if (newticks < -1)
510 				*t = global - 1;
511 			newticks = 0;
512 			break;
513 		}
514 	} while (!atomic_cmpset_int(&ticks, global, *t));
515 
516 	/*
517 	 * Run current process's virtual and profile time, as needed.
518 	 */
519 	pstats = p->p_stats;
520 	flags = 0;
521 	if (usermode &&
522 	    timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) {
523 		PROC_SLOCK(p);
524 		if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL],
525 		    tick * cnt) == 0)
526 			flags |= TDF_ALRMPEND | TDF_ASTPENDING;
527 		PROC_SUNLOCK(p);
528 	}
529 	if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) {
530 		PROC_SLOCK(p);
531 		if (itimerdecr(&pstats->p_timer[ITIMER_PROF],
532 		    tick * cnt) == 0)
533 			flags |= TDF_PROFPEND | TDF_ASTPENDING;
534 		PROC_SUNLOCK(p);
535 	}
536 	thread_lock(td);
537 	sched_tick(cnt);
538 	td->td_flags |= flags;
539 	thread_unlock(td);
540 
541 #ifdef	HWPMC_HOOKS
542 	if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid)))
543 		PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL);
544 	if (td->td_intr_frame != NULL)
545 		PMC_SOFT_CALL_TF( , , clock, hard, td->td_intr_frame);
546 #endif
547 	/* We are in charge to handle this tick duty. */
548 	if (newticks > 0) {
549 		/* Dangerous and no need to call these things concurrently. */
550 		if (atomic_cmpset_acq_int(&global_hardclock_run, 0, 1)) {
551 			tc_ticktock(newticks);
552 #ifdef DEVICE_POLLING
553 			/* This is very short and quick. */
554 			hardclock_device_poll();
555 #endif /* DEVICE_POLLING */
556 			atomic_store_rel_int(&global_hardclock_run, 0);
557 		}
558 #ifdef SW_WATCHDOG
559 		if (watchdog_enabled > 0) {
560 			i = atomic_fetchadd_int(&watchdog_ticks, -newticks);
561 			if (i > 0 && i <= newticks)
562 				watchdog_fire();
563 		}
564 #endif /* SW_WATCHDOG */
565 	}
566 	if (curcpu == CPU_FIRST())
567 		cpu_tick_calibration();
568 }
569 
570 void
571 hardclock_sync(int cpu)
572 {
573 	int	*t = DPCPU_ID_PTR(cpu, pcputicks);
574 
575 	*t = ticks;
576 }
577 
578 /*
579  * Compute number of ticks in the specified amount of time.
580  */
581 int
582 tvtohz(tv)
583 	struct timeval *tv;
584 {
585 	register unsigned long ticks;
586 	register long sec, usec;
587 
588 	/*
589 	 * If the number of usecs in the whole seconds part of the time
590 	 * difference fits in a long, then the total number of usecs will
591 	 * fit in an unsigned long.  Compute the total and convert it to
592 	 * ticks, rounding up and adding 1 to allow for the current tick
593 	 * to expire.  Rounding also depends on unsigned long arithmetic
594 	 * to avoid overflow.
595 	 *
596 	 * Otherwise, if the number of ticks in the whole seconds part of
597 	 * the time difference fits in a long, then convert the parts to
598 	 * ticks separately and add, using similar rounding methods and
599 	 * overflow avoidance.  This method would work in the previous
600 	 * case but it is slightly slower and assumes that hz is integral.
601 	 *
602 	 * Otherwise, round the time difference down to the maximum
603 	 * representable value.
604 	 *
605 	 * If ints have 32 bits, then the maximum value for any timeout in
606 	 * 10ms ticks is 248 days.
607 	 */
608 	sec = tv->tv_sec;
609 	usec = tv->tv_usec;
610 	if (usec < 0) {
611 		sec--;
612 		usec += 1000000;
613 	}
614 	if (sec < 0) {
615 #ifdef DIAGNOSTIC
616 		if (usec > 0) {
617 			sec++;
618 			usec -= 1000000;
619 		}
620 		printf("tvotohz: negative time difference %ld sec %ld usec\n",
621 		       sec, usec);
622 #endif
623 		ticks = 1;
624 	} else if (sec <= LONG_MAX / 1000000)
625 		ticks = (sec * 1000000 + (unsigned long)usec + (tick - 1))
626 			/ tick + 1;
627 	else if (sec <= LONG_MAX / hz)
628 		ticks = sec * hz
629 			+ ((unsigned long)usec + (tick - 1)) / tick + 1;
630 	else
631 		ticks = LONG_MAX;
632 	if (ticks > INT_MAX)
633 		ticks = INT_MAX;
634 	return ((int)ticks);
635 }
636 
637 /*
638  * Start profiling on a process.
639  *
640  * Kernel profiling passes proc0 which never exits and hence
641  * keeps the profile clock running constantly.
642  */
643 void
644 startprofclock(p)
645 	register struct proc *p;
646 {
647 
648 	PROC_LOCK_ASSERT(p, MA_OWNED);
649 	if (p->p_flag & P_STOPPROF)
650 		return;
651 	if ((p->p_flag & P_PROFIL) == 0) {
652 		p->p_flag |= P_PROFIL;
653 		mtx_lock(&time_lock);
654 		if (++profprocs == 1)
655 			cpu_startprofclock();
656 		mtx_unlock(&time_lock);
657 	}
658 }
659 
660 /*
661  * Stop profiling on a process.
662  */
663 void
664 stopprofclock(p)
665 	register struct proc *p;
666 {
667 
668 	PROC_LOCK_ASSERT(p, MA_OWNED);
669 	if (p->p_flag & P_PROFIL) {
670 		if (p->p_profthreads != 0) {
671 			p->p_flag |= P_STOPPROF;
672 			while (p->p_profthreads != 0)
673 				msleep(&p->p_profthreads, &p->p_mtx, PPAUSE,
674 				    "stopprof", 0);
675 			p->p_flag &= ~P_STOPPROF;
676 		}
677 		if ((p->p_flag & P_PROFIL) == 0)
678 			return;
679 		p->p_flag &= ~P_PROFIL;
680 		mtx_lock(&time_lock);
681 		if (--profprocs == 0)
682 			cpu_stopprofclock();
683 		mtx_unlock(&time_lock);
684 	}
685 }
686 
687 /*
688  * Statistics clock.  Updates rusage information and calls the scheduler
689  * to adjust priorities of the active thread.
690  *
691  * This should be called by all active processors.
692  */
693 void
694 statclock(int usermode)
695 {
696 
697 	statclock_cnt(1, usermode);
698 }
699 
700 void
701 statclock_cnt(int cnt, int usermode)
702 {
703 	struct rusage *ru;
704 	struct vmspace *vm;
705 	struct thread *td;
706 	struct proc *p;
707 	long rss;
708 	long *cp_time;
709 
710 	td = curthread;
711 	p = td->td_proc;
712 
713 	cp_time = (long *)PCPU_PTR(cp_time);
714 	if (usermode) {
715 		/*
716 		 * Charge the time as appropriate.
717 		 */
718 		td->td_uticks += cnt;
719 		if (p->p_nice > NZERO)
720 			cp_time[CP_NICE] += cnt;
721 		else
722 			cp_time[CP_USER] += cnt;
723 	} else {
724 		/*
725 		 * Came from kernel mode, so we were:
726 		 * - handling an interrupt,
727 		 * - doing syscall or trap work on behalf of the current
728 		 *   user process, or
729 		 * - spinning in the idle loop.
730 		 * Whichever it is, charge the time as appropriate.
731 		 * Note that we charge interrupts to the current process,
732 		 * regardless of whether they are ``for'' that process,
733 		 * so that we know how much of its real time was spent
734 		 * in ``non-process'' (i.e., interrupt) work.
735 		 */
736 		if ((td->td_pflags & TDP_ITHREAD) ||
737 		    td->td_intr_nesting_level >= 2) {
738 			td->td_iticks += cnt;
739 			cp_time[CP_INTR] += cnt;
740 		} else {
741 			td->td_pticks += cnt;
742 			td->td_sticks += cnt;
743 			if (!TD_IS_IDLETHREAD(td))
744 				cp_time[CP_SYS] += cnt;
745 			else
746 				cp_time[CP_IDLE] += cnt;
747 		}
748 	}
749 
750 	/* Update resource usage integrals and maximums. */
751 	MPASS(p->p_vmspace != NULL);
752 	vm = p->p_vmspace;
753 	ru = &td->td_ru;
754 	ru->ru_ixrss += pgtok(vm->vm_tsize) * cnt;
755 	ru->ru_idrss += pgtok(vm->vm_dsize) * cnt;
756 	ru->ru_isrss += pgtok(vm->vm_ssize) * cnt;
757 	rss = pgtok(vmspace_resident_count(vm));
758 	if (ru->ru_maxrss < rss)
759 		ru->ru_maxrss = rss;
760 	KTR_POINT2(KTR_SCHED, "thread", sched_tdname(td), "statclock",
761 	    "prio:%d", td->td_priority, "stathz:%d", (stathz)?stathz:hz);
762 	SDT_PROBE2(sched, , , tick, td, td->td_proc);
763 	thread_lock_flags(td, MTX_QUIET);
764 	for ( ; cnt > 0; cnt--)
765 		sched_clock(td);
766 	thread_unlock(td);
767 #ifdef HWPMC_HOOKS
768 	if (td->td_intr_frame != NULL)
769 		PMC_SOFT_CALL_TF( , , clock, stat, td->td_intr_frame);
770 #endif
771 }
772 
773 void
774 profclock(int usermode, uintfptr_t pc)
775 {
776 
777 	profclock_cnt(1, usermode, pc);
778 }
779 
780 void
781 profclock_cnt(int cnt, int usermode, uintfptr_t pc)
782 {
783 	struct thread *td;
784 #ifdef GPROF
785 	struct gmonparam *g;
786 	uintfptr_t i;
787 #endif
788 
789 	td = curthread;
790 	if (usermode) {
791 		/*
792 		 * Came from user mode; CPU was in user state.
793 		 * If this process is being profiled, record the tick.
794 		 * if there is no related user location yet, don't
795 		 * bother trying to count it.
796 		 */
797 		if (td->td_proc->p_flag & P_PROFIL)
798 			addupc_intr(td, pc, cnt);
799 	}
800 #ifdef GPROF
801 	else {
802 		/*
803 		 * Kernel statistics are just like addupc_intr, only easier.
804 		 */
805 		g = &_gmonparam;
806 		if (g->state == GMON_PROF_ON && pc >= g->lowpc) {
807 			i = PC_TO_I(g, pc);
808 			if (i < g->textsize) {
809 				KCOUNT(g, i) += cnt;
810 			}
811 		}
812 	}
813 #endif
814 #ifdef HWPMC_HOOKS
815 	if (td->td_intr_frame != NULL)
816 		PMC_SOFT_CALL_TF( , , clock, prof, td->td_intr_frame);
817 #endif
818 }
819 
820 /*
821  * Return information about system clocks.
822  */
823 static int
824 sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS)
825 {
826 	struct clockinfo clkinfo;
827 	/*
828 	 * Construct clockinfo structure.
829 	 */
830 	bzero(&clkinfo, sizeof(clkinfo));
831 	clkinfo.hz = hz;
832 	clkinfo.tick = tick;
833 	clkinfo.profhz = profhz;
834 	clkinfo.stathz = stathz ? stathz : hz;
835 	return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req));
836 }
837 
838 SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate,
839 	CTLTYPE_STRUCT|CTLFLAG_RD|CTLFLAG_MPSAFE,
840 	0, 0, sysctl_kern_clockrate, "S,clockinfo",
841 	"Rate and period of various kernel clocks");
842 
843 #ifdef SW_WATCHDOG
844 
845 static void
846 watchdog_config(void *unused __unused, u_int cmd, int *error)
847 {
848 	u_int u;
849 
850 	u = cmd & WD_INTERVAL;
851 	if (u >= WD_TO_1SEC) {
852 		watchdog_ticks = (1 << (u - WD_TO_1SEC)) * hz;
853 		watchdog_enabled = 1;
854 		*error = 0;
855 	} else {
856 		watchdog_enabled = 0;
857 	}
858 }
859 
860 /*
861  * Handle a watchdog timeout by dumping interrupt information and
862  * then either dropping to DDB or panicking.
863  */
864 static void
865 watchdog_fire(void)
866 {
867 	int nintr;
868 	uint64_t inttotal;
869 	u_long *curintr;
870 	char *curname;
871 
872 	curintr = intrcnt;
873 	curname = intrnames;
874 	inttotal = 0;
875 	nintr = sintrcnt / sizeof(u_long);
876 
877 	printf("interrupt                   total\n");
878 	while (--nintr >= 0) {
879 		if (*curintr)
880 			printf("%-12s %20lu\n", curname, *curintr);
881 		curname += strlen(curname) + 1;
882 		inttotal += *curintr++;
883 	}
884 	printf("Total        %20ju\n", (uintmax_t)inttotal);
885 
886 #if defined(KDB) && !defined(KDB_UNATTENDED)
887 	kdb_backtrace();
888 	kdb_enter(KDB_WHY_WATCHDOG, "watchdog timeout");
889 #else
890 	panic("watchdog timeout");
891 #endif
892 }
893 
894 #endif /* SW_WATCHDOG */
895