xref: /freebsd/sys/kern/kern_clock.c (revision 5dae51da3da0cc94d17bd67b308fad304ebec7e0)
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  * 3. 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 #ifdef DEVICE_POLLING
385 static int devpoll_run = 0;
386 #endif
387 
388 /*
389  * Initialize clock frequencies and start both clocks running.
390  */
391 /* ARGSUSED*/
392 static void
393 initclocks(dummy)
394 	void *dummy;
395 {
396 #ifdef EARLY_AP_STARTUP
397 	struct proc *p;
398 	struct thread *td;
399 #endif
400 	register int i;
401 
402 	/*
403 	 * Set divisors to 1 (normal case) and let the machine-specific
404 	 * code do its bit.
405 	 */
406 	mtx_init(&time_lock, "time lock", NULL, MTX_DEF);
407 	cpu_initclocks();
408 
409 	/*
410 	 * Compute profhz/stathz, and fix profhz if needed.
411 	 */
412 	i = stathz ? stathz : hz;
413 	if (profhz == 0)
414 		profhz = i;
415 	psratio = profhz / i;
416 #ifdef SW_WATCHDOG
417 	EVENTHANDLER_REGISTER(watchdog_list, watchdog_config, NULL, 0);
418 #endif
419 	/*
420 	 * Arrange for ticks to wrap 10 minutes after boot to help catch
421 	 * sign problems sooner.
422 	 */
423 	ticks = INT_MAX - (hz * 10 * 60);
424 
425 #ifdef EARLY_AP_STARTUP
426 	/*
427 	 * Fixup the tick counts in any blocked or sleeping threads to
428 	 * account for the jump above.
429 	 */
430 	sx_slock(&allproc_lock);
431 	FOREACH_PROC_IN_SYSTEM(p) {
432 		PROC_LOCK(p);
433 		if (p->p_state == PRS_NEW) {
434 			PROC_UNLOCK(p);
435 			continue;
436 		}
437 		FOREACH_THREAD_IN_PROC(p, td) {
438 			thread_lock(td);
439 			if (TD_ON_LOCK(td)) {
440 				MPASS(td->td_blktick == 0);
441 				td->td_blktick = ticks;
442 			}
443 			if (TD_ON_SLEEPQ(td)) {
444 				MPASS(td->td_slptick == 0);
445 				td->td_slptick = ticks;
446 			}
447 			thread_unlock(td);
448 		}
449 		PROC_UNLOCK(p);
450 	}
451 	sx_sunlock(&allproc_lock);
452 #endif
453 }
454 
455 /*
456  * Each time the real-time timer fires, this function is called on all CPUs.
457  * Note that hardclock() calls hardclock_cpu() for the boot CPU, so only
458  * the other CPUs in the system need to call this function.
459  */
460 void
461 hardclock_cpu(int usermode)
462 {
463 	struct pstats *pstats;
464 	struct thread *td = curthread;
465 	struct proc *p = td->td_proc;
466 	int flags;
467 
468 	/*
469 	 * Run current process's virtual and profile time, as needed.
470 	 */
471 	pstats = p->p_stats;
472 	flags = 0;
473 	if (usermode &&
474 	    timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) {
475 		PROC_ITIMLOCK(p);
476 		if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0)
477 			flags |= TDF_ALRMPEND | TDF_ASTPENDING;
478 		PROC_ITIMUNLOCK(p);
479 	}
480 	if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) {
481 		PROC_ITIMLOCK(p);
482 		if (itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0)
483 			flags |= TDF_PROFPEND | TDF_ASTPENDING;
484 		PROC_ITIMUNLOCK(p);
485 	}
486 	thread_lock(td);
487 	td->td_flags |= flags;
488 	thread_unlock(td);
489 
490 #ifdef HWPMC_HOOKS
491 	if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid)))
492 		PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL);
493 	if (td->td_intr_frame != NULL)
494 		PMC_SOFT_CALL_TF( , , clock, hard, td->td_intr_frame);
495 #endif
496 	callout_process(sbinuptime());
497 }
498 
499 /*
500  * The real-time timer, interrupting hz times per second.
501  */
502 void
503 hardclock(int usermode, uintfptr_t pc)
504 {
505 
506 	atomic_add_int(&ticks, 1);
507 	hardclock_cpu(usermode);
508 	tc_ticktock(1);
509 	cpu_tick_calibration();
510 	/*
511 	 * If no separate statistics clock is available, run it from here.
512 	 *
513 	 * XXX: this only works for UP
514 	 */
515 	if (stathz == 0) {
516 		profclock(usermode, pc);
517 		statclock(usermode);
518 	}
519 #ifdef DEVICE_POLLING
520 	hardclock_device_poll();	/* this is very short and quick */
521 #endif /* DEVICE_POLLING */
522 #ifdef SW_WATCHDOG
523 	if (watchdog_enabled > 0 && --watchdog_ticks <= 0)
524 		watchdog_fire();
525 #endif /* SW_WATCHDOG */
526 }
527 
528 void
529 hardclock_cnt(int cnt, int usermode)
530 {
531 	struct pstats *pstats;
532 	struct thread *td = curthread;
533 	struct proc *p = td->td_proc;
534 	int *t = DPCPU_PTR(pcputicks);
535 	int flags, global, newticks;
536 #ifdef SW_WATCHDOG
537 	int i;
538 #endif /* SW_WATCHDOG */
539 
540 	/*
541 	 * Update per-CPU and possibly global ticks values.
542 	 */
543 	*t += cnt;
544 	do {
545 		global = ticks;
546 		newticks = *t - global;
547 		if (newticks <= 0) {
548 			if (newticks < -1)
549 				*t = global - 1;
550 			newticks = 0;
551 			break;
552 		}
553 	} while (!atomic_cmpset_int(&ticks, global, *t));
554 
555 	/*
556 	 * Run current process's virtual and profile time, as needed.
557 	 */
558 	pstats = p->p_stats;
559 	flags = 0;
560 	if (usermode &&
561 	    timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) {
562 		PROC_ITIMLOCK(p);
563 		if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL],
564 		    tick * cnt) == 0)
565 			flags |= TDF_ALRMPEND | TDF_ASTPENDING;
566 		PROC_ITIMUNLOCK(p);
567 	}
568 	if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) {
569 		PROC_ITIMLOCK(p);
570 		if (itimerdecr(&pstats->p_timer[ITIMER_PROF],
571 		    tick * cnt) == 0)
572 			flags |= TDF_PROFPEND | TDF_ASTPENDING;
573 		PROC_ITIMUNLOCK(p);
574 	}
575 	if (flags != 0) {
576 		thread_lock(td);
577 		td->td_flags |= flags;
578 		thread_unlock(td);
579 	}
580 
581 #ifdef	HWPMC_HOOKS
582 	if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid)))
583 		PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL);
584 	if (td->td_intr_frame != NULL)
585 		PMC_SOFT_CALL_TF( , , clock, hard, td->td_intr_frame);
586 #endif
587 	/* We are in charge to handle this tick duty. */
588 	if (newticks > 0) {
589 		tc_ticktock(newticks);
590 #ifdef DEVICE_POLLING
591 		/* Dangerous and no need to call these things concurrently. */
592 		if (atomic_cmpset_acq_int(&devpoll_run, 0, 1)) {
593 			/* This is very short and quick. */
594 			hardclock_device_poll();
595 			atomic_store_rel_int(&devpoll_run, 0);
596 		}
597 #endif /* DEVICE_POLLING */
598 #ifdef SW_WATCHDOG
599 		if (watchdog_enabled > 0) {
600 			i = atomic_fetchadd_int(&watchdog_ticks, -newticks);
601 			if (i > 0 && i <= newticks)
602 				watchdog_fire();
603 		}
604 #endif /* SW_WATCHDOG */
605 	}
606 	if (curcpu == CPU_FIRST())
607 		cpu_tick_calibration();
608 }
609 
610 void
611 hardclock_sync(int cpu)
612 {
613 	int	*t = DPCPU_ID_PTR(cpu, pcputicks);
614 
615 	*t = ticks;
616 }
617 
618 /*
619  * Compute number of ticks in the specified amount of time.
620  */
621 int
622 tvtohz(tv)
623 	struct timeval *tv;
624 {
625 	register unsigned long ticks;
626 	register long sec, usec;
627 
628 	/*
629 	 * If the number of usecs in the whole seconds part of the time
630 	 * difference fits in a long, then the total number of usecs will
631 	 * fit in an unsigned long.  Compute the total and convert it to
632 	 * ticks, rounding up and adding 1 to allow for the current tick
633 	 * to expire.  Rounding also depends on unsigned long arithmetic
634 	 * to avoid overflow.
635 	 *
636 	 * Otherwise, if the number of ticks in the whole seconds part of
637 	 * the time difference fits in a long, then convert the parts to
638 	 * ticks separately and add, using similar rounding methods and
639 	 * overflow avoidance.  This method would work in the previous
640 	 * case but it is slightly slower and assumes that hz is integral.
641 	 *
642 	 * Otherwise, round the time difference down to the maximum
643 	 * representable value.
644 	 *
645 	 * If ints have 32 bits, then the maximum value for any timeout in
646 	 * 10ms ticks is 248 days.
647 	 */
648 	sec = tv->tv_sec;
649 	usec = tv->tv_usec;
650 	if (usec < 0) {
651 		sec--;
652 		usec += 1000000;
653 	}
654 	if (sec < 0) {
655 #ifdef DIAGNOSTIC
656 		if (usec > 0) {
657 			sec++;
658 			usec -= 1000000;
659 		}
660 		printf("tvotohz: negative time difference %ld sec %ld usec\n",
661 		       sec, usec);
662 #endif
663 		ticks = 1;
664 	} else if (sec <= LONG_MAX / 1000000)
665 		ticks = howmany(sec * 1000000 + (unsigned long)usec, tick) + 1;
666 	else if (sec <= LONG_MAX / hz)
667 		ticks = sec * hz
668 			+ howmany((unsigned long)usec, tick) + 1;
669 	else
670 		ticks = LONG_MAX;
671 	if (ticks > INT_MAX)
672 		ticks = INT_MAX;
673 	return ((int)ticks);
674 }
675 
676 /*
677  * Start profiling on a process.
678  *
679  * Kernel profiling passes proc0 which never exits and hence
680  * keeps the profile clock running constantly.
681  */
682 void
683 startprofclock(p)
684 	register struct proc *p;
685 {
686 
687 	PROC_LOCK_ASSERT(p, MA_OWNED);
688 	if (p->p_flag & P_STOPPROF)
689 		return;
690 	if ((p->p_flag & P_PROFIL) == 0) {
691 		p->p_flag |= P_PROFIL;
692 		mtx_lock(&time_lock);
693 		if (++profprocs == 1)
694 			cpu_startprofclock();
695 		mtx_unlock(&time_lock);
696 	}
697 }
698 
699 /*
700  * Stop profiling on a process.
701  */
702 void
703 stopprofclock(p)
704 	register struct proc *p;
705 {
706 
707 	PROC_LOCK_ASSERT(p, MA_OWNED);
708 	if (p->p_flag & P_PROFIL) {
709 		if (p->p_profthreads != 0) {
710 			while (p->p_profthreads != 0) {
711 				p->p_flag |= P_STOPPROF;
712 				msleep(&p->p_profthreads, &p->p_mtx, PPAUSE,
713 				    "stopprof", 0);
714 			}
715 		}
716 		if ((p->p_flag & P_PROFIL) == 0)
717 			return;
718 		p->p_flag &= ~P_PROFIL;
719 		mtx_lock(&time_lock);
720 		if (--profprocs == 0)
721 			cpu_stopprofclock();
722 		mtx_unlock(&time_lock);
723 	}
724 }
725 
726 /*
727  * Statistics clock.  Updates rusage information and calls the scheduler
728  * to adjust priorities of the active thread.
729  *
730  * This should be called by all active processors.
731  */
732 void
733 statclock(int usermode)
734 {
735 
736 	statclock_cnt(1, usermode);
737 }
738 
739 void
740 statclock_cnt(int cnt, int usermode)
741 {
742 	struct rusage *ru;
743 	struct vmspace *vm;
744 	struct thread *td;
745 	struct proc *p;
746 	long rss;
747 	long *cp_time;
748 
749 	td = curthread;
750 	p = td->td_proc;
751 
752 	cp_time = (long *)PCPU_PTR(cp_time);
753 	if (usermode) {
754 		/*
755 		 * Charge the time as appropriate.
756 		 */
757 		td->td_uticks += cnt;
758 		if (p->p_nice > NZERO)
759 			cp_time[CP_NICE] += cnt;
760 		else
761 			cp_time[CP_USER] += cnt;
762 	} else {
763 		/*
764 		 * Came from kernel mode, so we were:
765 		 * - handling an interrupt,
766 		 * - doing syscall or trap work on behalf of the current
767 		 *   user process, or
768 		 * - spinning in the idle loop.
769 		 * Whichever it is, charge the time as appropriate.
770 		 * Note that we charge interrupts to the current process,
771 		 * regardless of whether they are ``for'' that process,
772 		 * so that we know how much of its real time was spent
773 		 * in ``non-process'' (i.e., interrupt) work.
774 		 */
775 		if ((td->td_pflags & TDP_ITHREAD) ||
776 		    td->td_intr_nesting_level >= 2) {
777 			td->td_iticks += cnt;
778 			cp_time[CP_INTR] += cnt;
779 		} else {
780 			td->td_pticks += cnt;
781 			td->td_sticks += cnt;
782 			if (!TD_IS_IDLETHREAD(td))
783 				cp_time[CP_SYS] += cnt;
784 			else
785 				cp_time[CP_IDLE] += cnt;
786 		}
787 	}
788 
789 	/* Update resource usage integrals and maximums. */
790 	MPASS(p->p_vmspace != NULL);
791 	vm = p->p_vmspace;
792 	ru = &td->td_ru;
793 	ru->ru_ixrss += pgtok(vm->vm_tsize) * cnt;
794 	ru->ru_idrss += pgtok(vm->vm_dsize) * cnt;
795 	ru->ru_isrss += pgtok(vm->vm_ssize) * cnt;
796 	rss = pgtok(vmspace_resident_count(vm));
797 	if (ru->ru_maxrss < rss)
798 		ru->ru_maxrss = rss;
799 	KTR_POINT2(KTR_SCHED, "thread", sched_tdname(td), "statclock",
800 	    "prio:%d", td->td_priority, "stathz:%d", (stathz)?stathz:hz);
801 	SDT_PROBE2(sched, , , tick, td, td->td_proc);
802 	thread_lock_flags(td, MTX_QUIET);
803 	for ( ; cnt > 0; cnt--)
804 		sched_clock(td);
805 	thread_unlock(td);
806 #ifdef HWPMC_HOOKS
807 	if (td->td_intr_frame != NULL)
808 		PMC_SOFT_CALL_TF( , , clock, stat, td->td_intr_frame);
809 #endif
810 }
811 
812 void
813 profclock(int usermode, uintfptr_t pc)
814 {
815 
816 	profclock_cnt(1, usermode, pc);
817 }
818 
819 void
820 profclock_cnt(int cnt, int usermode, uintfptr_t pc)
821 {
822 	struct thread *td;
823 #ifdef GPROF
824 	struct gmonparam *g;
825 	uintfptr_t i;
826 #endif
827 
828 	td = curthread;
829 	if (usermode) {
830 		/*
831 		 * Came from user mode; CPU was in user state.
832 		 * If this process is being profiled, record the tick.
833 		 * if there is no related user location yet, don't
834 		 * bother trying to count it.
835 		 */
836 		if (td->td_proc->p_flag & P_PROFIL)
837 			addupc_intr(td, pc, cnt);
838 	}
839 #ifdef GPROF
840 	else {
841 		/*
842 		 * Kernel statistics are just like addupc_intr, only easier.
843 		 */
844 		g = &_gmonparam;
845 		if (g->state == GMON_PROF_ON && pc >= g->lowpc) {
846 			i = PC_TO_I(g, pc);
847 			if (i < g->textsize) {
848 				KCOUNT(g, i) += cnt;
849 			}
850 		}
851 	}
852 #endif
853 #ifdef HWPMC_HOOKS
854 	if (td->td_intr_frame != NULL)
855 		PMC_SOFT_CALL_TF( , , clock, prof, td->td_intr_frame);
856 #endif
857 }
858 
859 /*
860  * Return information about system clocks.
861  */
862 static int
863 sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS)
864 {
865 	struct clockinfo clkinfo;
866 	/*
867 	 * Construct clockinfo structure.
868 	 */
869 	bzero(&clkinfo, sizeof(clkinfo));
870 	clkinfo.hz = hz;
871 	clkinfo.tick = tick;
872 	clkinfo.profhz = profhz;
873 	clkinfo.stathz = stathz ? stathz : hz;
874 	return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req));
875 }
876 
877 SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate,
878 	CTLTYPE_STRUCT|CTLFLAG_RD|CTLFLAG_MPSAFE,
879 	0, 0, sysctl_kern_clockrate, "S,clockinfo",
880 	"Rate and period of various kernel clocks");
881 
882 #ifdef SW_WATCHDOG
883 
884 static void
885 watchdog_config(void *unused __unused, u_int cmd, int *error)
886 {
887 	u_int u;
888 
889 	u = cmd & WD_INTERVAL;
890 	if (u >= WD_TO_1SEC) {
891 		watchdog_ticks = (1 << (u - WD_TO_1SEC)) * hz;
892 		watchdog_enabled = 1;
893 		*error = 0;
894 	} else {
895 		watchdog_enabled = 0;
896 	}
897 }
898 
899 /*
900  * Handle a watchdog timeout by dumping interrupt information and
901  * then either dropping to DDB or panicking.
902  */
903 static void
904 watchdog_fire(void)
905 {
906 	int nintr;
907 	uint64_t inttotal;
908 	u_long *curintr;
909 	char *curname;
910 
911 	curintr = intrcnt;
912 	curname = intrnames;
913 	inttotal = 0;
914 	nintr = sintrcnt / sizeof(u_long);
915 
916 	printf("interrupt                   total\n");
917 	while (--nintr >= 0) {
918 		if (*curintr)
919 			printf("%-12s %20lu\n", curname, *curintr);
920 		curname += strlen(curname) + 1;
921 		inttotal += *curintr++;
922 	}
923 	printf("Total        %20ju\n", (uintmax_t)inttotal);
924 
925 #if defined(KDB) && !defined(KDB_UNATTENDED)
926 	kdb_backtrace();
927 	kdb_enter(KDB_WHY_WATCHDOG, "watchdog timeout");
928 #else
929 	panic("watchdog timeout");
930 #endif
931 }
932 
933 #endif /* SW_WATCHDOG */
934