xref: /freebsd/sys/kern/kern_clock.c (revision 7d99ab9fd0cc2c1ce2ecef0ed6d0672c2a50b0cb)
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_kdtrace.h"
44 #include "opt_ntp.h"
45 #include "opt_watchdog.h"
46 
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/callout.h>
50 #include <sys/kdb.h>
51 #include <sys/kernel.h>
52 #include <sys/kthread.h>
53 #include <sys/ktr.h>
54 #include <sys/lock.h>
55 #include <sys/mutex.h>
56 #include <sys/proc.h>
57 #include <sys/resource.h>
58 #include <sys/resourcevar.h>
59 #include <sys/sched.h>
60 #include <sys/sdt.h>
61 #include <sys/signalvar.h>
62 #include <sys/sleepqueue.h>
63 #include <sys/smp.h>
64 #include <vm/vm.h>
65 #include <vm/pmap.h>
66 #include <vm/vm_map.h>
67 #include <sys/sysctl.h>
68 #include <sys/bus.h>
69 #include <sys/interrupt.h>
70 #include <sys/limits.h>
71 #include <sys/timetc.h>
72 
73 #ifdef GPROF
74 #include <sys/gmon.h>
75 #endif
76 
77 #ifdef HWPMC_HOOKS
78 #include <sys/pmckern.h>
79 PMC_SOFT_DEFINE( , , clock, hard);
80 PMC_SOFT_DEFINE( , , clock, stat);
81 PMC_SOFT_DEFINE_EX( , , clock, prof, \
82     cpu_startprofclock, cpu_stopprofclock);
83 #endif
84 
85 #ifdef DEVICE_POLLING
86 extern void hardclock_device_poll(void);
87 #endif /* DEVICE_POLLING */
88 
89 static void initclocks(void *dummy);
90 SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL);
91 
92 /* Spin-lock protecting profiling statistics. */
93 static struct mtx time_lock;
94 
95 SDT_PROVIDER_DECLARE(sched);
96 SDT_PROBE_DEFINE2(sched, , , tick, tick, "struct thread *", "struct proc *");
97 
98 static int
99 sysctl_kern_cp_time(SYSCTL_HANDLER_ARGS)
100 {
101 	int error;
102 	long cp_time[CPUSTATES];
103 #ifdef SCTL_MASK32
104 	int i;
105 	unsigned int cp_time32[CPUSTATES];
106 #endif
107 
108 	read_cpu_time(cp_time);
109 #ifdef SCTL_MASK32
110 	if (req->flags & SCTL_MASK32) {
111 		if (!req->oldptr)
112 			return SYSCTL_OUT(req, 0, sizeof(cp_time32));
113 		for (i = 0; i < CPUSTATES; i++)
114 			cp_time32[i] = (unsigned int)cp_time[i];
115 		error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32));
116 	} else
117 #endif
118 	{
119 		if (!req->oldptr)
120 			return SYSCTL_OUT(req, 0, sizeof(cp_time));
121 		error = SYSCTL_OUT(req, cp_time, sizeof(cp_time));
122 	}
123 	return error;
124 }
125 
126 SYSCTL_PROC(_kern, OID_AUTO, cp_time, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE,
127     0,0, sysctl_kern_cp_time, "LU", "CPU time statistics");
128 
129 static long empty[CPUSTATES];
130 
131 static int
132 sysctl_kern_cp_times(SYSCTL_HANDLER_ARGS)
133 {
134 	struct pcpu *pcpu;
135 	int error;
136 	int c;
137 	long *cp_time;
138 #ifdef SCTL_MASK32
139 	unsigned int cp_time32[CPUSTATES];
140 	int i;
141 #endif
142 
143 	if (!req->oldptr) {
144 #ifdef SCTL_MASK32
145 		if (req->flags & SCTL_MASK32)
146 			return SYSCTL_OUT(req, 0, sizeof(cp_time32) * (mp_maxid + 1));
147 		else
148 #endif
149 			return SYSCTL_OUT(req, 0, sizeof(long) * CPUSTATES * (mp_maxid + 1));
150 	}
151 	for (error = 0, c = 0; error == 0 && c <= mp_maxid; c++) {
152 		if (!CPU_ABSENT(c)) {
153 			pcpu = pcpu_find(c);
154 			cp_time = pcpu->pc_cp_time;
155 		} else {
156 			cp_time = empty;
157 		}
158 #ifdef SCTL_MASK32
159 		if (req->flags & SCTL_MASK32) {
160 			for (i = 0; i < CPUSTATES; i++)
161 				cp_time32[i] = (unsigned int)cp_time[i];
162 			error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32));
163 		} else
164 #endif
165 			error = SYSCTL_OUT(req, cp_time, sizeof(long) * CPUSTATES);
166 	}
167 	return error;
168 }
169 
170 SYSCTL_PROC(_kern, OID_AUTO, cp_times, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE,
171     0,0, sysctl_kern_cp_times, "LU", "per-CPU time statistics");
172 
173 #ifdef DEADLKRES
174 static const char *blessed[] = {
175 	"getblk",
176 	"so_snd_sx",
177 	"so_rcv_sx",
178 	NULL
179 };
180 static int slptime_threshold = 1800;
181 static int blktime_threshold = 900;
182 static int sleepfreq = 3;
183 
184 static void
185 deadlkres(void)
186 {
187 	struct proc *p;
188 	struct thread *td;
189 	void *wchan;
190 	int blkticks, i, slpticks, slptype, tryl, tticks;
191 
192 	tryl = 0;
193 	for (;;) {
194 		blkticks = blktime_threshold * hz;
195 		slpticks = slptime_threshold * hz;
196 
197 		/*
198 		 * Avoid to sleep on the sx_lock in order to avoid a possible
199 		 * priority inversion problem leading to starvation.
200 		 * If the lock can't be held after 100 tries, panic.
201 		 */
202 		if (!sx_try_slock(&allproc_lock)) {
203 			if (tryl > 100)
204 		panic("%s: possible deadlock detected on allproc_lock\n",
205 				    __func__);
206 			tryl++;
207 			pause("allproc", sleepfreq * hz);
208 			continue;
209 		}
210 		tryl = 0;
211 		FOREACH_PROC_IN_SYSTEM(p) {
212 			PROC_LOCK(p);
213 			if (p->p_state == PRS_NEW) {
214 				PROC_UNLOCK(p);
215 				continue;
216 			}
217 			FOREACH_THREAD_IN_PROC(p, td) {
218 
219 				/*
220 				 * Once a thread is found in "interesting"
221 				 * state a possible ticks wrap-up needs to be
222 				 * checked.
223 				 */
224 				thread_lock(td);
225 				if (TD_ON_LOCK(td) && ticks < td->td_blktick) {
226 
227 					/*
228 					 * The thread should be blocked on a
229 					 * turnstile, simply check if the
230 					 * turnstile channel is in good state.
231 					 */
232 					MPASS(td->td_blocked != NULL);
233 
234 					tticks = ticks - td->td_blktick;
235 					thread_unlock(td);
236 					if (tticks > blkticks) {
237 
238 						/*
239 						 * Accordingly with provided
240 						 * thresholds, this thread is
241 						 * stuck for too long on a
242 						 * turnstile.
243 						 */
244 						PROC_UNLOCK(p);
245 						sx_sunlock(&allproc_lock);
246 	panic("%s: possible deadlock detected for %p, blocked for %d ticks\n",
247 						    __func__, td, tticks);
248 					}
249 				} else if (TD_IS_SLEEPING(td) &&
250 				    TD_ON_SLEEPQ(td) &&
251 				    ticks < td->td_blktick) {
252 
253 					/*
254 					 * Check if the thread is sleeping on a
255 					 * lock, otherwise skip the check.
256 					 * Drop the thread lock in order to
257 					 * avoid a LOR with the sleepqueue
258 					 * spinlock.
259 					 */
260 					wchan = td->td_wchan;
261 					tticks = ticks - td->td_slptick;
262 					thread_unlock(td);
263 					slptype = sleepq_type(wchan);
264 					if ((slptype == SLEEPQ_SX ||
265 					    slptype == SLEEPQ_LK) &&
266 					    tticks > slpticks) {
267 
268 						/*
269 						 * Accordingly with provided
270 						 * thresholds, this thread is
271 						 * stuck for too long on a
272 						 * sleepqueue.
273 						 * However, being on a
274 						 * sleepqueue, we might still
275 						 * check for the blessed
276 						 * list.
277 						 */
278 						tryl = 0;
279 						for (i = 0; blessed[i] != NULL;
280 						    i++) {
281 							if (!strcmp(blessed[i],
282 							    td->td_wmesg)) {
283 								tryl = 1;
284 								break;
285 							}
286 						}
287 						if (tryl != 0) {
288 							tryl = 0;
289 							continue;
290 						}
291 						PROC_UNLOCK(p);
292 						sx_sunlock(&allproc_lock);
293 	panic("%s: possible deadlock detected for %p, blocked for %d ticks\n",
294 						    __func__, td, tticks);
295 					}
296 				} else
297 					thread_unlock(td);
298 			}
299 			PROC_UNLOCK(p);
300 		}
301 		sx_sunlock(&allproc_lock);
302 
303 		/* Sleep for sleepfreq seconds. */
304 		pause("-", sleepfreq * hz);
305 	}
306 }
307 
308 static struct kthread_desc deadlkres_kd = {
309 	"deadlkres",
310 	deadlkres,
311 	(struct thread **)NULL
312 };
313 
314 SYSINIT(deadlkres, SI_SUB_CLOCKS, SI_ORDER_ANY, kthread_start, &deadlkres_kd);
315 
316 static SYSCTL_NODE(_debug, OID_AUTO, deadlkres, CTLFLAG_RW, 0,
317     "Deadlock resolver");
318 SYSCTL_INT(_debug_deadlkres, OID_AUTO, slptime_threshold, CTLFLAG_RW,
319     &slptime_threshold, 0,
320     "Number of seconds within is valid to sleep on a sleepqueue");
321 SYSCTL_INT(_debug_deadlkres, OID_AUTO, blktime_threshold, CTLFLAG_RW,
322     &blktime_threshold, 0,
323     "Number of seconds within is valid to block on a turnstile");
324 SYSCTL_INT(_debug_deadlkres, OID_AUTO, sleepfreq, CTLFLAG_RW, &sleepfreq, 0,
325     "Number of seconds between any deadlock resolver thread run");
326 #endif	/* DEADLKRES */
327 
328 void
329 read_cpu_time(long *cp_time)
330 {
331 	struct pcpu *pc;
332 	int i, j;
333 
334 	/* Sum up global cp_time[]. */
335 	bzero(cp_time, sizeof(long) * CPUSTATES);
336 	CPU_FOREACH(i) {
337 		pc = pcpu_find(i);
338 		for (j = 0; j < CPUSTATES; j++)
339 			cp_time[j] += pc->pc_cp_time[j];
340 	}
341 }
342 
343 #ifdef SW_WATCHDOG
344 #include <sys/watchdog.h>
345 
346 static int watchdog_ticks;
347 static int watchdog_enabled;
348 static void watchdog_fire(void);
349 static void watchdog_config(void *, u_int, int *);
350 #endif /* SW_WATCHDOG */
351 
352 /*
353  * Clock handling routines.
354  *
355  * This code is written to operate with two timers that run independently of
356  * each other.
357  *
358  * The main timer, running hz times per second, is used to trigger interval
359  * timers, timeouts and rescheduling as needed.
360  *
361  * The second timer handles kernel and user profiling,
362  * and does resource use estimation.  If the second timer is programmable,
363  * it is randomized to avoid aliasing between the two clocks.  For example,
364  * the randomization prevents an adversary from always giving up the cpu
365  * just before its quantum expires.  Otherwise, it would never accumulate
366  * cpu ticks.  The mean frequency of the second timer is stathz.
367  *
368  * If no second timer exists, stathz will be zero; in this case we drive
369  * profiling and statistics off the main clock.  This WILL NOT be accurate;
370  * do not do it unless absolutely necessary.
371  *
372  * The statistics clock may (or may not) be run at a higher rate while
373  * profiling.  This profile clock runs at profhz.  We require that profhz
374  * be an integral multiple of stathz.
375  *
376  * If the statistics clock is running fast, it must be divided by the ratio
377  * profhz/stathz for statistics.  (For profiling, every tick counts.)
378  *
379  * Time-of-day is maintained using a "timecounter", which may or may
380  * not be related to the hardware generating the above mentioned
381  * interrupts.
382  */
383 
384 int	stathz;
385 int	profhz;
386 int	profprocs;
387 volatile int	ticks;
388 int	psratio;
389 
390 static DPCPU_DEFINE(int, pcputicks);	/* Per-CPU version of ticks. */
391 static int global_hardclock_run = 0;
392 
393 /*
394  * Initialize clock frequencies and start both clocks running.
395  */
396 /* ARGSUSED*/
397 static void
398 initclocks(dummy)
399 	void *dummy;
400 {
401 	register int i;
402 
403 	/*
404 	 * Set divisors to 1 (normal case) and let the machine-specific
405 	 * code do its bit.
406 	 */
407 	mtx_init(&time_lock, "time lock", NULL, MTX_DEF);
408 	cpu_initclocks();
409 
410 	/*
411 	 * Compute profhz/stathz, and fix profhz if needed.
412 	 */
413 	i = stathz ? stathz : hz;
414 	if (profhz == 0)
415 		profhz = i;
416 	psratio = profhz / i;
417 #ifdef SW_WATCHDOG
418 	EVENTHANDLER_REGISTER(watchdog_list, watchdog_config, NULL, 0);
419 #endif
420 }
421 
422 /*
423  * Each time the real-time timer fires, this function is called on all CPUs.
424  * Note that hardclock() calls hardclock_cpu() for the boot CPU, so only
425  * the other CPUs in the system need to call this function.
426  */
427 void
428 hardclock_cpu(int usermode)
429 {
430 	struct pstats *pstats;
431 	struct thread *td = curthread;
432 	struct proc *p = td->td_proc;
433 	int flags;
434 
435 	/*
436 	 * Run current process's virtual and profile time, as needed.
437 	 */
438 	pstats = p->p_stats;
439 	flags = 0;
440 	if (usermode &&
441 	    timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) {
442 		PROC_SLOCK(p);
443 		if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0)
444 			flags |= TDF_ALRMPEND | TDF_ASTPENDING;
445 		PROC_SUNLOCK(p);
446 	}
447 	if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) {
448 		PROC_SLOCK(p);
449 		if (itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0)
450 			flags |= TDF_PROFPEND | TDF_ASTPENDING;
451 		PROC_SUNLOCK(p);
452 	}
453 	thread_lock(td);
454 	sched_tick(1);
455 	td->td_flags |= flags;
456 	thread_unlock(td);
457 
458 #ifdef HWPMC_HOOKS
459 	if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid)))
460 		PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL);
461 	if (td->td_intr_frame != NULL)
462 		PMC_SOFT_CALL_TF( , , clock, hard, td->td_intr_frame);
463 #endif
464 	callout_process(sbinuptime());
465 }
466 
467 /*
468  * The real-time timer, interrupting hz times per second.
469  */
470 void
471 hardclock(int usermode, uintfptr_t pc)
472 {
473 
474 	atomic_add_int(&ticks, 1);
475 	hardclock_cpu(usermode);
476 	tc_ticktock(1);
477 	cpu_tick_calibration();
478 	/*
479 	 * If no separate statistics clock is available, run it from here.
480 	 *
481 	 * XXX: this only works for UP
482 	 */
483 	if (stathz == 0) {
484 		profclock(usermode, pc);
485 		statclock(usermode);
486 	}
487 #ifdef DEVICE_POLLING
488 	hardclock_device_poll();	/* this is very short and quick */
489 #endif /* DEVICE_POLLING */
490 #ifdef SW_WATCHDOG
491 	if (watchdog_enabled > 0 && --watchdog_ticks <= 0)
492 		watchdog_fire();
493 #endif /* SW_WATCHDOG */
494 }
495 
496 void
497 hardclock_cnt(int cnt, int usermode)
498 {
499 	struct pstats *pstats;
500 	struct thread *td = curthread;
501 	struct proc *p = td->td_proc;
502 	int *t = DPCPU_PTR(pcputicks);
503 	int flags, global, newticks;
504 #ifdef SW_WATCHDOG
505 	int i;
506 #endif /* SW_WATCHDOG */
507 
508 	/*
509 	 * Update per-CPU and possibly global ticks values.
510 	 */
511 	*t += cnt;
512 	do {
513 		global = ticks;
514 		newticks = *t - global;
515 		if (newticks <= 0) {
516 			if (newticks < -1)
517 				*t = global - 1;
518 			newticks = 0;
519 			break;
520 		}
521 	} while (!atomic_cmpset_int(&ticks, global, *t));
522 
523 	/*
524 	 * Run current process's virtual and profile time, as needed.
525 	 */
526 	pstats = p->p_stats;
527 	flags = 0;
528 	if (usermode &&
529 	    timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) {
530 		PROC_SLOCK(p);
531 		if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL],
532 		    tick * cnt) == 0)
533 			flags |= TDF_ALRMPEND | TDF_ASTPENDING;
534 		PROC_SUNLOCK(p);
535 	}
536 	if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) {
537 		PROC_SLOCK(p);
538 		if (itimerdecr(&pstats->p_timer[ITIMER_PROF],
539 		    tick * cnt) == 0)
540 			flags |= TDF_PROFPEND | TDF_ASTPENDING;
541 		PROC_SUNLOCK(p);
542 	}
543 	thread_lock(td);
544 	sched_tick(cnt);
545 	td->td_flags |= flags;
546 	thread_unlock(td);
547 
548 #ifdef	HWPMC_HOOKS
549 	if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid)))
550 		PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL);
551 	if (td->td_intr_frame != NULL)
552 		PMC_SOFT_CALL_TF( , , clock, hard, td->td_intr_frame);
553 #endif
554 	/* We are in charge to handle this tick duty. */
555 	if (newticks > 0) {
556 		/* Dangerous and no need to call these things concurrently. */
557 		if (atomic_cmpset_acq_int(&global_hardclock_run, 0, 1)) {
558 			tc_ticktock(newticks);
559 #ifdef DEVICE_POLLING
560 			/* This is very short and quick. */
561 			hardclock_device_poll();
562 #endif /* DEVICE_POLLING */
563 			atomic_store_rel_int(&global_hardclock_run, 0);
564 		}
565 #ifdef SW_WATCHDOG
566 		if (watchdog_enabled > 0) {
567 			i = atomic_fetchadd_int(&watchdog_ticks, -newticks);
568 			if (i > 0 && i <= newticks)
569 				watchdog_fire();
570 		}
571 #endif /* SW_WATCHDOG */
572 	}
573 	if (curcpu == CPU_FIRST())
574 		cpu_tick_calibration();
575 }
576 
577 void
578 hardclock_sync(int cpu)
579 {
580 	int	*t = DPCPU_ID_PTR(cpu, pcputicks);
581 
582 	*t = ticks;
583 }
584 
585 /*
586  * Compute number of ticks in the specified amount of time.
587  */
588 int
589 tvtohz(tv)
590 	struct timeval *tv;
591 {
592 	register unsigned long ticks;
593 	register long sec, usec;
594 
595 	/*
596 	 * If the number of usecs in the whole seconds part of the time
597 	 * difference fits in a long, then the total number of usecs will
598 	 * fit in an unsigned long.  Compute the total and convert it to
599 	 * ticks, rounding up and adding 1 to allow for the current tick
600 	 * to expire.  Rounding also depends on unsigned long arithmetic
601 	 * to avoid overflow.
602 	 *
603 	 * Otherwise, if the number of ticks in the whole seconds part of
604 	 * the time difference fits in a long, then convert the parts to
605 	 * ticks separately and add, using similar rounding methods and
606 	 * overflow avoidance.  This method would work in the previous
607 	 * case but it is slightly slower and assumes that hz is integral.
608 	 *
609 	 * Otherwise, round the time difference down to the maximum
610 	 * representable value.
611 	 *
612 	 * If ints have 32 bits, then the maximum value for any timeout in
613 	 * 10ms ticks is 248 days.
614 	 */
615 	sec = tv->tv_sec;
616 	usec = tv->tv_usec;
617 	if (usec < 0) {
618 		sec--;
619 		usec += 1000000;
620 	}
621 	if (sec < 0) {
622 #ifdef DIAGNOSTIC
623 		if (usec > 0) {
624 			sec++;
625 			usec -= 1000000;
626 		}
627 		printf("tvotohz: negative time difference %ld sec %ld usec\n",
628 		       sec, usec);
629 #endif
630 		ticks = 1;
631 	} else if (sec <= LONG_MAX / 1000000)
632 		ticks = (sec * 1000000 + (unsigned long)usec + (tick - 1))
633 			/ tick + 1;
634 	else if (sec <= LONG_MAX / hz)
635 		ticks = sec * hz
636 			+ ((unsigned long)usec + (tick - 1)) / tick + 1;
637 	else
638 		ticks = LONG_MAX;
639 	if (ticks > INT_MAX)
640 		ticks = INT_MAX;
641 	return ((int)ticks);
642 }
643 
644 /*
645  * Start profiling on a process.
646  *
647  * Kernel profiling passes proc0 which never exits and hence
648  * keeps the profile clock running constantly.
649  */
650 void
651 startprofclock(p)
652 	register struct proc *p;
653 {
654 
655 	PROC_LOCK_ASSERT(p, MA_OWNED);
656 	if (p->p_flag & P_STOPPROF)
657 		return;
658 	if ((p->p_flag & P_PROFIL) == 0) {
659 		p->p_flag |= P_PROFIL;
660 		mtx_lock(&time_lock);
661 		if (++profprocs == 1)
662 			cpu_startprofclock();
663 		mtx_unlock(&time_lock);
664 	}
665 }
666 
667 /*
668  * Stop profiling on a process.
669  */
670 void
671 stopprofclock(p)
672 	register struct proc *p;
673 {
674 
675 	PROC_LOCK_ASSERT(p, MA_OWNED);
676 	if (p->p_flag & P_PROFIL) {
677 		if (p->p_profthreads != 0) {
678 			p->p_flag |= P_STOPPROF;
679 			while (p->p_profthreads != 0)
680 				msleep(&p->p_profthreads, &p->p_mtx, PPAUSE,
681 				    "stopprof", 0);
682 			p->p_flag &= ~P_STOPPROF;
683 		}
684 		if ((p->p_flag & P_PROFIL) == 0)
685 			return;
686 		p->p_flag &= ~P_PROFIL;
687 		mtx_lock(&time_lock);
688 		if (--profprocs == 0)
689 			cpu_stopprofclock();
690 		mtx_unlock(&time_lock);
691 	}
692 }
693 
694 /*
695  * Statistics clock.  Updates rusage information and calls the scheduler
696  * to adjust priorities of the active thread.
697  *
698  * This should be called by all active processors.
699  */
700 void
701 statclock(int usermode)
702 {
703 
704 	statclock_cnt(1, usermode);
705 }
706 
707 void
708 statclock_cnt(int cnt, int usermode)
709 {
710 	struct rusage *ru;
711 	struct vmspace *vm;
712 	struct thread *td;
713 	struct proc *p;
714 	long rss;
715 	long *cp_time;
716 
717 	td = curthread;
718 	p = td->td_proc;
719 
720 	cp_time = (long *)PCPU_PTR(cp_time);
721 	if (usermode) {
722 		/*
723 		 * Charge the time as appropriate.
724 		 */
725 		td->td_uticks += cnt;
726 		if (p->p_nice > NZERO)
727 			cp_time[CP_NICE] += cnt;
728 		else
729 			cp_time[CP_USER] += cnt;
730 	} else {
731 		/*
732 		 * Came from kernel mode, so we were:
733 		 * - handling an interrupt,
734 		 * - doing syscall or trap work on behalf of the current
735 		 *   user process, or
736 		 * - spinning in the idle loop.
737 		 * Whichever it is, charge the time as appropriate.
738 		 * Note that we charge interrupts to the current process,
739 		 * regardless of whether they are ``for'' that process,
740 		 * so that we know how much of its real time was spent
741 		 * in ``non-process'' (i.e., interrupt) work.
742 		 */
743 		if ((td->td_pflags & TDP_ITHREAD) ||
744 		    td->td_intr_nesting_level >= 2) {
745 			td->td_iticks += cnt;
746 			cp_time[CP_INTR] += cnt;
747 		} else {
748 			td->td_pticks += cnt;
749 			td->td_sticks += cnt;
750 			if (!TD_IS_IDLETHREAD(td))
751 				cp_time[CP_SYS] += cnt;
752 			else
753 				cp_time[CP_IDLE] += cnt;
754 		}
755 	}
756 
757 	/* Update resource usage integrals and maximums. */
758 	MPASS(p->p_vmspace != NULL);
759 	vm = p->p_vmspace;
760 	ru = &td->td_ru;
761 	ru->ru_ixrss += pgtok(vm->vm_tsize) * cnt;
762 	ru->ru_idrss += pgtok(vm->vm_dsize) * cnt;
763 	ru->ru_isrss += pgtok(vm->vm_ssize) * cnt;
764 	rss = pgtok(vmspace_resident_count(vm));
765 	if (ru->ru_maxrss < rss)
766 		ru->ru_maxrss = rss;
767 	KTR_POINT2(KTR_SCHED, "thread", sched_tdname(td), "statclock",
768 	    "prio:%d", td->td_priority, "stathz:%d", (stathz)?stathz:hz);
769 	SDT_PROBE2(sched, , , tick, td, td->td_proc);
770 	thread_lock_flags(td, MTX_QUIET);
771 	for ( ; cnt > 0; cnt--)
772 		sched_clock(td);
773 	thread_unlock(td);
774 #ifdef HWPMC_HOOKS
775 	if (td->td_intr_frame != NULL)
776 		PMC_SOFT_CALL_TF( , , clock, stat, td->td_intr_frame);
777 #endif
778 }
779 
780 void
781 profclock(int usermode, uintfptr_t pc)
782 {
783 
784 	profclock_cnt(1, usermode, pc);
785 }
786 
787 void
788 profclock_cnt(int cnt, int usermode, uintfptr_t pc)
789 {
790 	struct thread *td;
791 #ifdef GPROF
792 	struct gmonparam *g;
793 	uintfptr_t i;
794 #endif
795 
796 	td = curthread;
797 	if (usermode) {
798 		/*
799 		 * Came from user mode; CPU was in user state.
800 		 * If this process is being profiled, record the tick.
801 		 * if there is no related user location yet, don't
802 		 * bother trying to count it.
803 		 */
804 		if (td->td_proc->p_flag & P_PROFIL)
805 			addupc_intr(td, pc, cnt);
806 	}
807 #ifdef GPROF
808 	else {
809 		/*
810 		 * Kernel statistics are just like addupc_intr, only easier.
811 		 */
812 		g = &_gmonparam;
813 		if (g->state == GMON_PROF_ON && pc >= g->lowpc) {
814 			i = PC_TO_I(g, pc);
815 			if (i < g->textsize) {
816 				KCOUNT(g, i) += cnt;
817 			}
818 		}
819 	}
820 #endif
821 #ifdef HWPMC_HOOKS
822 	if (td->td_intr_frame != NULL)
823 		PMC_SOFT_CALL_TF( , , clock, prof, td->td_intr_frame);
824 #endif
825 }
826 
827 /*
828  * Return information about system clocks.
829  */
830 static int
831 sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS)
832 {
833 	struct clockinfo clkinfo;
834 	/*
835 	 * Construct clockinfo structure.
836 	 */
837 	bzero(&clkinfo, sizeof(clkinfo));
838 	clkinfo.hz = hz;
839 	clkinfo.tick = tick;
840 	clkinfo.profhz = profhz;
841 	clkinfo.stathz = stathz ? stathz : hz;
842 	return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req));
843 }
844 
845 SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate,
846 	CTLTYPE_STRUCT|CTLFLAG_RD|CTLFLAG_MPSAFE,
847 	0, 0, sysctl_kern_clockrate, "S,clockinfo",
848 	"Rate and period of various kernel clocks");
849 
850 #ifdef SW_WATCHDOG
851 
852 static void
853 watchdog_config(void *unused __unused, u_int cmd, int *error)
854 {
855 	u_int u;
856 
857 	u = cmd & WD_INTERVAL;
858 	if (u >= WD_TO_1SEC) {
859 		watchdog_ticks = (1 << (u - WD_TO_1SEC)) * hz;
860 		watchdog_enabled = 1;
861 		*error = 0;
862 	} else {
863 		watchdog_enabled = 0;
864 	}
865 }
866 
867 /*
868  * Handle a watchdog timeout by dumping interrupt information and
869  * then either dropping to DDB or panicking.
870  */
871 static void
872 watchdog_fire(void)
873 {
874 	int nintr;
875 	uint64_t inttotal;
876 	u_long *curintr;
877 	char *curname;
878 
879 	curintr = intrcnt;
880 	curname = intrnames;
881 	inttotal = 0;
882 	nintr = sintrcnt / sizeof(u_long);
883 
884 	printf("interrupt                   total\n");
885 	while (--nintr >= 0) {
886 		if (*curintr)
887 			printf("%-12s %20lu\n", curname, *curintr);
888 		curname += strlen(curname) + 1;
889 		inttotal += *curintr++;
890 	}
891 	printf("Total        %20ju\n", (uintmax_t)inttotal);
892 
893 #if defined(KDB) && !defined(KDB_UNATTENDED)
894 	kdb_backtrace();
895 	kdb_enter(KDB_WHY_WATCHDOG, "watchdog timeout");
896 #else
897 	panic("watchdog timeout");
898 #endif
899 }
900 
901 #endif /* SW_WATCHDOG */
902