xref: /freebsd/sys/kern/kern_clock.c (revision c243e4902be8df1e643c76b5f18b68bb77cc5268)
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 #endif
82 
83 #ifdef DEVICE_POLLING
84 extern void hardclock_device_poll(void);
85 #endif /* DEVICE_POLLING */
86 
87 static void initclocks(void *dummy);
88 SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL);
89 
90 /* Spin-lock protecting profiling statistics. */
91 static struct mtx time_lock;
92 
93 SDT_PROVIDER_DECLARE(sched);
94 SDT_PROBE_DEFINE2(sched, , , tick, tick, "struct thread *", "struct proc *");
95 
96 static int
97 sysctl_kern_cp_time(SYSCTL_HANDLER_ARGS)
98 {
99 	int error;
100 	long cp_time[CPUSTATES];
101 #ifdef SCTL_MASK32
102 	int i;
103 	unsigned int cp_time32[CPUSTATES];
104 #endif
105 
106 	read_cpu_time(cp_time);
107 #ifdef SCTL_MASK32
108 	if (req->flags & SCTL_MASK32) {
109 		if (!req->oldptr)
110 			return SYSCTL_OUT(req, 0, sizeof(cp_time32));
111 		for (i = 0; i < CPUSTATES; i++)
112 			cp_time32[i] = (unsigned int)cp_time[i];
113 		error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32));
114 	} else
115 #endif
116 	{
117 		if (!req->oldptr)
118 			return SYSCTL_OUT(req, 0, sizeof(cp_time));
119 		error = SYSCTL_OUT(req, cp_time, sizeof(cp_time));
120 	}
121 	return error;
122 }
123 
124 SYSCTL_PROC(_kern, OID_AUTO, cp_time, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE,
125     0,0, sysctl_kern_cp_time, "LU", "CPU time statistics");
126 
127 static long empty[CPUSTATES];
128 
129 static int
130 sysctl_kern_cp_times(SYSCTL_HANDLER_ARGS)
131 {
132 	struct pcpu *pcpu;
133 	int error;
134 	int c;
135 	long *cp_time;
136 #ifdef SCTL_MASK32
137 	unsigned int cp_time32[CPUSTATES];
138 	int i;
139 #endif
140 
141 	if (!req->oldptr) {
142 #ifdef SCTL_MASK32
143 		if (req->flags & SCTL_MASK32)
144 			return SYSCTL_OUT(req, 0, sizeof(cp_time32) * (mp_maxid + 1));
145 		else
146 #endif
147 			return SYSCTL_OUT(req, 0, sizeof(long) * CPUSTATES * (mp_maxid + 1));
148 	}
149 	for (error = 0, c = 0; error == 0 && c <= mp_maxid; c++) {
150 		if (!CPU_ABSENT(c)) {
151 			pcpu = pcpu_find(c);
152 			cp_time = pcpu->pc_cp_time;
153 		} else {
154 			cp_time = empty;
155 		}
156 #ifdef SCTL_MASK32
157 		if (req->flags & SCTL_MASK32) {
158 			for (i = 0; i < CPUSTATES; i++)
159 				cp_time32[i] = (unsigned int)cp_time[i];
160 			error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32));
161 		} else
162 #endif
163 			error = SYSCTL_OUT(req, cp_time, sizeof(long) * CPUSTATES);
164 	}
165 	return error;
166 }
167 
168 SYSCTL_PROC(_kern, OID_AUTO, cp_times, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE,
169     0,0, sysctl_kern_cp_times, "LU", "per-CPU time statistics");
170 
171 #ifdef DEADLKRES
172 static const char *blessed[] = {
173 	"getblk",
174 	"so_snd_sx",
175 	"so_rcv_sx",
176 	NULL
177 };
178 static int slptime_threshold = 1800;
179 static int blktime_threshold = 900;
180 static int sleepfreq = 3;
181 
182 static void
183 deadlkres(void)
184 {
185 	struct proc *p;
186 	struct thread *td;
187 	void *wchan;
188 	int blkticks, i, slpticks, slptype, tryl, tticks;
189 
190 	tryl = 0;
191 	for (;;) {
192 		blkticks = blktime_threshold * hz;
193 		slpticks = slptime_threshold * hz;
194 
195 		/*
196 		 * Avoid to sleep on the sx_lock in order to avoid a possible
197 		 * priority inversion problem leading to starvation.
198 		 * If the lock can't be held after 100 tries, panic.
199 		 */
200 		if (!sx_try_slock(&allproc_lock)) {
201 			if (tryl > 100)
202 		panic("%s: possible deadlock detected on allproc_lock\n",
203 				    __func__);
204 			tryl++;
205 			pause("allproc", sleepfreq * hz);
206 			continue;
207 		}
208 		tryl = 0;
209 		FOREACH_PROC_IN_SYSTEM(p) {
210 			PROC_LOCK(p);
211 			if (p->p_state == PRS_NEW) {
212 				PROC_UNLOCK(p);
213 				continue;
214 			}
215 			FOREACH_THREAD_IN_PROC(p, td) {
216 
217 				/*
218 				 * Once a thread is found in "interesting"
219 				 * state a possible ticks wrap-up needs to be
220 				 * checked.
221 				 */
222 				thread_lock(td);
223 				if (TD_ON_LOCK(td) && ticks < td->td_blktick) {
224 
225 					/*
226 					 * The thread should be blocked on a
227 					 * turnstile, simply check if the
228 					 * turnstile channel is in good state.
229 					 */
230 					MPASS(td->td_blocked != NULL);
231 
232 					tticks = ticks - td->td_blktick;
233 					thread_unlock(td);
234 					if (tticks > blkticks) {
235 
236 						/*
237 						 * Accordingly with provided
238 						 * thresholds, this thread is
239 						 * stuck for too long on a
240 						 * turnstile.
241 						 */
242 						PROC_UNLOCK(p);
243 						sx_sunlock(&allproc_lock);
244 	panic("%s: possible deadlock detected for %p, blocked for %d ticks\n",
245 						    __func__, td, tticks);
246 					}
247 				} else if (TD_IS_SLEEPING(td) &&
248 				    TD_ON_SLEEPQ(td) &&
249 				    ticks < td->td_blktick) {
250 
251 					/*
252 					 * Check if the thread is sleeping on a
253 					 * lock, otherwise skip the check.
254 					 * Drop the thread lock in order to
255 					 * avoid a LOR with the sleepqueue
256 					 * spinlock.
257 					 */
258 					wchan = td->td_wchan;
259 					tticks = ticks - td->td_slptick;
260 					thread_unlock(td);
261 					slptype = sleepq_type(wchan);
262 					if ((slptype == SLEEPQ_SX ||
263 					    slptype == SLEEPQ_LK) &&
264 					    tticks > slpticks) {
265 
266 						/*
267 						 * Accordingly with provided
268 						 * thresholds, this thread is
269 						 * stuck for too long on a
270 						 * sleepqueue.
271 						 * However, being on a
272 						 * sleepqueue, we might still
273 						 * check for the blessed
274 						 * list.
275 						 */
276 						tryl = 0;
277 						for (i = 0; blessed[i] != NULL;
278 						    i++) {
279 							if (!strcmp(blessed[i],
280 							    td->td_wmesg)) {
281 								tryl = 1;
282 								break;
283 							}
284 						}
285 						if (tryl != 0) {
286 							tryl = 0;
287 							continue;
288 						}
289 						PROC_UNLOCK(p);
290 						sx_sunlock(&allproc_lock);
291 	panic("%s: possible deadlock detected for %p, blocked for %d ticks\n",
292 						    __func__, td, tticks);
293 					}
294 				} else
295 					thread_unlock(td);
296 			}
297 			PROC_UNLOCK(p);
298 		}
299 		sx_sunlock(&allproc_lock);
300 
301 		/* Sleep for sleepfreq seconds. */
302 		pause("-", sleepfreq * hz);
303 	}
304 }
305 
306 static struct kthread_desc deadlkres_kd = {
307 	"deadlkres",
308 	deadlkres,
309 	(struct thread **)NULL
310 };
311 
312 SYSINIT(deadlkres, SI_SUB_CLOCKS, SI_ORDER_ANY, kthread_start, &deadlkres_kd);
313 
314 static SYSCTL_NODE(_debug, OID_AUTO, deadlkres, CTLFLAG_RW, 0,
315     "Deadlock resolver");
316 SYSCTL_INT(_debug_deadlkres, OID_AUTO, slptime_threshold, CTLFLAG_RW,
317     &slptime_threshold, 0,
318     "Number of seconds within is valid to sleep on a sleepqueue");
319 SYSCTL_INT(_debug_deadlkres, OID_AUTO, blktime_threshold, CTLFLAG_RW,
320     &blktime_threshold, 0,
321     "Number of seconds within is valid to block on a turnstile");
322 SYSCTL_INT(_debug_deadlkres, OID_AUTO, sleepfreq, CTLFLAG_RW, &sleepfreq, 0,
323     "Number of seconds between any deadlock resolver thread run");
324 #endif	/* DEADLKRES */
325 
326 void
327 read_cpu_time(long *cp_time)
328 {
329 	struct pcpu *pc;
330 	int i, j;
331 
332 	/* Sum up global cp_time[]. */
333 	bzero(cp_time, sizeof(long) * CPUSTATES);
334 	CPU_FOREACH(i) {
335 		pc = pcpu_find(i);
336 		for (j = 0; j < CPUSTATES; j++)
337 			cp_time[j] += pc->pc_cp_time[j];
338 	}
339 }
340 
341 #ifdef SW_WATCHDOG
342 #include <sys/watchdog.h>
343 
344 static int watchdog_ticks;
345 static int watchdog_enabled;
346 static void watchdog_fire(void);
347 static void watchdog_config(void *, u_int, int *);
348 #endif /* SW_WATCHDOG */
349 
350 /*
351  * Clock handling routines.
352  *
353  * This code is written to operate with two timers that run independently of
354  * each other.
355  *
356  * The main timer, running hz times per second, is used to trigger interval
357  * timers, timeouts and rescheduling as needed.
358  *
359  * The second timer handles kernel and user profiling,
360  * and does resource use estimation.  If the second timer is programmable,
361  * it is randomized to avoid aliasing between the two clocks.  For example,
362  * the randomization prevents an adversary from always giving up the cpu
363  * just before its quantum expires.  Otherwise, it would never accumulate
364  * cpu ticks.  The mean frequency of the second timer is stathz.
365  *
366  * If no second timer exists, stathz will be zero; in this case we drive
367  * profiling and statistics off the main clock.  This WILL NOT be accurate;
368  * do not do it unless absolutely necessary.
369  *
370  * The statistics clock may (or may not) be run at a higher rate while
371  * profiling.  This profile clock runs at profhz.  We require that profhz
372  * be an integral multiple of stathz.
373  *
374  * If the statistics clock is running fast, it must be divided by the ratio
375  * profhz/stathz for statistics.  (For profiling, every tick counts.)
376  *
377  * Time-of-day is maintained using a "timecounter", which may or may
378  * not be related to the hardware generating the above mentioned
379  * interrupts.
380  */
381 
382 int	stathz;
383 int	profhz;
384 int	profprocs;
385 int	ticks;
386 int	psratio;
387 
388 static DPCPU_DEFINE(int, pcputicks);	/* Per-CPU version of ticks. */
389 static int global_hardclock_run = 0;
390 
391 /*
392  * Initialize clock frequencies and start both clocks running.
393  */
394 /* ARGSUSED*/
395 static void
396 initclocks(dummy)
397 	void *dummy;
398 {
399 	register 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 #ifdef SW_WATCHDOG
416 	EVENTHANDLER_REGISTER(watchdog_list, watchdog_config, NULL, 0);
417 #endif
418 }
419 
420 /*
421  * Each time the real-time timer fires, this function is called on all CPUs.
422  * Note that hardclock() calls hardclock_cpu() for the boot CPU, so only
423  * the other CPUs in the system need to call this function.
424  */
425 void
426 hardclock_cpu(int usermode)
427 {
428 	struct pstats *pstats;
429 	struct thread *td = curthread;
430 	struct proc *p = td->td_proc;
431 	int flags;
432 
433 	/*
434 	 * Run current process's virtual and profile time, as needed.
435 	 */
436 	pstats = p->p_stats;
437 	flags = 0;
438 	if (usermode &&
439 	    timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) {
440 		PROC_SLOCK(p);
441 		if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0)
442 			flags |= TDF_ALRMPEND | TDF_ASTPENDING;
443 		PROC_SUNLOCK(p);
444 	}
445 	if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) {
446 		PROC_SLOCK(p);
447 		if (itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0)
448 			flags |= TDF_PROFPEND | TDF_ASTPENDING;
449 		PROC_SUNLOCK(p);
450 	}
451 	thread_lock(td);
452 	sched_tick(1);
453 	td->td_flags |= flags;
454 	thread_unlock(td);
455 
456 #ifdef HWPMC_HOOKS
457 	if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid)))
458 		PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL);
459 	if (td->td_intr_frame != NULL)
460 		PMC_SOFT_CALL_TF( , , clock, hard, td->td_intr_frame);
461 #endif
462 	callout_tick();
463 }
464 
465 /*
466  * The real-time timer, interrupting hz times per second.
467  */
468 void
469 hardclock(int usermode, uintfptr_t pc)
470 {
471 
472 	atomic_add_int((volatile int *)&ticks, 1);
473 	hardclock_cpu(usermode);
474 	tc_ticktock(1);
475 	cpu_tick_calibration();
476 	/*
477 	 * If no separate statistics clock is available, run it from here.
478 	 *
479 	 * XXX: this only works for UP
480 	 */
481 	if (stathz == 0) {
482 		profclock(usermode, pc);
483 		statclock(usermode);
484 	}
485 #ifdef DEVICE_POLLING
486 	hardclock_device_poll();	/* this is very short and quick */
487 #endif /* DEVICE_POLLING */
488 #ifdef SW_WATCHDOG
489 	if (watchdog_enabled > 0 && --watchdog_ticks <= 0)
490 		watchdog_fire();
491 #endif /* SW_WATCHDOG */
492 }
493 
494 void
495 hardclock_cnt(int cnt, int usermode)
496 {
497 	struct pstats *pstats;
498 	struct thread *td = curthread;
499 	struct proc *p = td->td_proc;
500 	int *t = DPCPU_PTR(pcputicks);
501 	int flags, global, newticks;
502 #ifdef SW_WATCHDOG
503 	int i;
504 #endif /* SW_WATCHDOG */
505 
506 	/*
507 	 * Update per-CPU and possibly global ticks values.
508 	 */
509 	*t += cnt;
510 	do {
511 		global = ticks;
512 		newticks = *t - global;
513 		if (newticks <= 0) {
514 			if (newticks < -1)
515 				*t = global - 1;
516 			newticks = 0;
517 			break;
518 		}
519 	} while (!atomic_cmpset_int(&ticks, global, *t));
520 
521 	/*
522 	 * Run current process's virtual and profile time, as needed.
523 	 */
524 	pstats = p->p_stats;
525 	flags = 0;
526 	if (usermode &&
527 	    timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) {
528 		PROC_SLOCK(p);
529 		if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL],
530 		    tick * cnt) == 0)
531 			flags |= TDF_ALRMPEND | TDF_ASTPENDING;
532 		PROC_SUNLOCK(p);
533 	}
534 	if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) {
535 		PROC_SLOCK(p);
536 		if (itimerdecr(&pstats->p_timer[ITIMER_PROF],
537 		    tick * cnt) == 0)
538 			flags |= TDF_PROFPEND | TDF_ASTPENDING;
539 		PROC_SUNLOCK(p);
540 	}
541 	thread_lock(td);
542 	sched_tick(cnt);
543 	td->td_flags |= flags;
544 	thread_unlock(td);
545 
546 #ifdef	HWPMC_HOOKS
547 	if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid)))
548 		PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL);
549 	if (td->td_intr_frame != NULL)
550 		PMC_SOFT_CALL_TF( , , clock, hard, td->td_intr_frame);
551 #endif
552 	callout_tick();
553 	/* We are in charge to handle this tick duty. */
554 	if (newticks > 0) {
555 		/* Dangerous and no need to call these things concurrently. */
556 		if (atomic_cmpset_acq_int(&global_hardclock_run, 0, 1)) {
557 			tc_ticktock(newticks);
558 #ifdef DEVICE_POLLING
559 			/* This is very short and quick. */
560 			hardclock_device_poll();
561 #endif /* DEVICE_POLLING */
562 			atomic_store_rel_int(&global_hardclock_run, 0);
563 		}
564 #ifdef SW_WATCHDOG
565 		if (watchdog_enabled > 0) {
566 			i = atomic_fetchadd_int(&watchdog_ticks, -newticks);
567 			if (i > 0 && i <= newticks)
568 				watchdog_fire();
569 		}
570 #endif /* SW_WATCHDOG */
571 	}
572 	if (curcpu == CPU_FIRST())
573 		cpu_tick_calibration();
574 }
575 
576 void
577 hardclock_sync(int cpu)
578 {
579 	int	*t = DPCPU_ID_PTR(cpu, pcputicks);
580 
581 	*t = ticks;
582 }
583 
584 /*
585  * Compute number of ticks in the specified amount of time.
586  */
587 int
588 tvtohz(tv)
589 	struct timeval *tv;
590 {
591 	register unsigned long ticks;
592 	register long sec, usec;
593 
594 	/*
595 	 * If the number of usecs in the whole seconds part of the time
596 	 * difference fits in a long, then the total number of usecs will
597 	 * fit in an unsigned long.  Compute the total and convert it to
598 	 * ticks, rounding up and adding 1 to allow for the current tick
599 	 * to expire.  Rounding also depends on unsigned long arithmetic
600 	 * to avoid overflow.
601 	 *
602 	 * Otherwise, if the number of ticks in the whole seconds part of
603 	 * the time difference fits in a long, then convert the parts to
604 	 * ticks separately and add, using similar rounding methods and
605 	 * overflow avoidance.  This method would work in the previous
606 	 * case but it is slightly slower and assumes that hz is integral.
607 	 *
608 	 * Otherwise, round the time difference down to the maximum
609 	 * representable value.
610 	 *
611 	 * If ints have 32 bits, then the maximum value for any timeout in
612 	 * 10ms ticks is 248 days.
613 	 */
614 	sec = tv->tv_sec;
615 	usec = tv->tv_usec;
616 	if (usec < 0) {
617 		sec--;
618 		usec += 1000000;
619 	}
620 	if (sec < 0) {
621 #ifdef DIAGNOSTIC
622 		if (usec > 0) {
623 			sec++;
624 			usec -= 1000000;
625 		}
626 		printf("tvotohz: negative time difference %ld sec %ld usec\n",
627 		       sec, usec);
628 #endif
629 		ticks = 1;
630 	} else if (sec <= LONG_MAX / 1000000)
631 		ticks = (sec * 1000000 + (unsigned long)usec + (tick - 1))
632 			/ tick + 1;
633 	else if (sec <= LONG_MAX / hz)
634 		ticks = sec * hz
635 			+ ((unsigned long)usec + (tick - 1)) / tick + 1;
636 	else
637 		ticks = LONG_MAX;
638 	if (ticks > INT_MAX)
639 		ticks = INT_MAX;
640 	return ((int)ticks);
641 }
642 
643 /*
644  * Start profiling on a process.
645  *
646  * Kernel profiling passes proc0 which never exits and hence
647  * keeps the profile clock running constantly.
648  */
649 void
650 startprofclock(p)
651 	register struct proc *p;
652 {
653 
654 	PROC_LOCK_ASSERT(p, MA_OWNED);
655 	if (p->p_flag & P_STOPPROF)
656 		return;
657 	if ((p->p_flag & P_PROFIL) == 0) {
658 		p->p_flag |= P_PROFIL;
659 		mtx_lock(&time_lock);
660 		if (++profprocs == 1)
661 			cpu_startprofclock();
662 		mtx_unlock(&time_lock);
663 	}
664 }
665 
666 /*
667  * Stop profiling on a process.
668  */
669 void
670 stopprofclock(p)
671 	register struct proc *p;
672 {
673 
674 	PROC_LOCK_ASSERT(p, MA_OWNED);
675 	if (p->p_flag & P_PROFIL) {
676 		if (p->p_profthreads != 0) {
677 			p->p_flag |= P_STOPPROF;
678 			while (p->p_profthreads != 0)
679 				msleep(&p->p_profthreads, &p->p_mtx, PPAUSE,
680 				    "stopprof", 0);
681 			p->p_flag &= ~P_STOPPROF;
682 		}
683 		if ((p->p_flag & P_PROFIL) == 0)
684 			return;
685 		p->p_flag &= ~P_PROFIL;
686 		mtx_lock(&time_lock);
687 		if (--profprocs == 0)
688 			cpu_stopprofclock();
689 		mtx_unlock(&time_lock);
690 	}
691 }
692 
693 /*
694  * Statistics clock.  Updates rusage information and calls the scheduler
695  * to adjust priorities of the active thread.
696  *
697  * This should be called by all active processors.
698  */
699 void
700 statclock(int usermode)
701 {
702 
703 	statclock_cnt(1, usermode);
704 }
705 
706 void
707 statclock_cnt(int cnt, int usermode)
708 {
709 	struct rusage *ru;
710 	struct vmspace *vm;
711 	struct thread *td;
712 	struct proc *p;
713 	long rss;
714 	long *cp_time;
715 
716 	td = curthread;
717 	p = td->td_proc;
718 
719 	cp_time = (long *)PCPU_PTR(cp_time);
720 	if (usermode) {
721 		/*
722 		 * Charge the time as appropriate.
723 		 */
724 		td->td_uticks += cnt;
725 		if (p->p_nice > NZERO)
726 			cp_time[CP_NICE] += cnt;
727 		else
728 			cp_time[CP_USER] += cnt;
729 	} else {
730 		/*
731 		 * Came from kernel mode, so we were:
732 		 * - handling an interrupt,
733 		 * - doing syscall or trap work on behalf of the current
734 		 *   user process, or
735 		 * - spinning in the idle loop.
736 		 * Whichever it is, charge the time as appropriate.
737 		 * Note that we charge interrupts to the current process,
738 		 * regardless of whether they are ``for'' that process,
739 		 * so that we know how much of its real time was spent
740 		 * in ``non-process'' (i.e., interrupt) work.
741 		 */
742 		if ((td->td_pflags & TDP_ITHREAD) ||
743 		    td->td_intr_nesting_level >= 2) {
744 			td->td_iticks += cnt;
745 			cp_time[CP_INTR] += cnt;
746 		} else {
747 			td->td_pticks += cnt;
748 			td->td_sticks += cnt;
749 			if (!TD_IS_IDLETHREAD(td))
750 				cp_time[CP_SYS] += cnt;
751 			else
752 				cp_time[CP_IDLE] += cnt;
753 		}
754 	}
755 
756 	/* Update resource usage integrals and maximums. */
757 	MPASS(p->p_vmspace != NULL);
758 	vm = p->p_vmspace;
759 	ru = &td->td_ru;
760 	ru->ru_ixrss += pgtok(vm->vm_tsize) * cnt;
761 	ru->ru_idrss += pgtok(vm->vm_dsize) * cnt;
762 	ru->ru_isrss += pgtok(vm->vm_ssize) * cnt;
763 	rss = pgtok(vmspace_resident_count(vm));
764 	if (ru->ru_maxrss < rss)
765 		ru->ru_maxrss = rss;
766 	KTR_POINT2(KTR_SCHED, "thread", sched_tdname(td), "statclock",
767 	    "prio:%d", td->td_priority, "stathz:%d", (stathz)?stathz:hz);
768 	SDT_PROBE2(sched, , , tick, td, td->td_proc);
769 	thread_lock_flags(td, MTX_QUIET);
770 	for ( ; cnt > 0; cnt--)
771 		sched_clock(td);
772 	thread_unlock(td);
773 #ifdef HWPMC_HOOKS
774 	if (td->td_intr_frame != NULL)
775 		PMC_SOFT_CALL_TF( , , clock, stat, td->td_intr_frame);
776 #endif
777 }
778 
779 void
780 profclock(int usermode, uintfptr_t pc)
781 {
782 
783 	profclock_cnt(1, usermode, pc);
784 }
785 
786 void
787 profclock_cnt(int cnt, int usermode, uintfptr_t pc)
788 {
789 	struct thread *td;
790 #ifdef GPROF
791 	struct gmonparam *g;
792 	uintfptr_t i;
793 #endif
794 
795 	td = curthread;
796 	if (usermode) {
797 		/*
798 		 * Came from user mode; CPU was in user state.
799 		 * If this process is being profiled, record the tick.
800 		 * if there is no related user location yet, don't
801 		 * bother trying to count it.
802 		 */
803 		if (td->td_proc->p_flag & P_PROFIL)
804 			addupc_intr(td, pc, cnt);
805 	}
806 #ifdef GPROF
807 	else {
808 		/*
809 		 * Kernel statistics are just like addupc_intr, only easier.
810 		 */
811 		g = &_gmonparam;
812 		if (g->state == GMON_PROF_ON && pc >= g->lowpc) {
813 			i = PC_TO_I(g, pc);
814 			if (i < g->textsize) {
815 				KCOUNT(g, i) += cnt;
816 			}
817 		}
818 	}
819 #endif
820 }
821 
822 /*
823  * Return information about system clocks.
824  */
825 static int
826 sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS)
827 {
828 	struct clockinfo clkinfo;
829 	/*
830 	 * Construct clockinfo structure.
831 	 */
832 	bzero(&clkinfo, sizeof(clkinfo));
833 	clkinfo.hz = hz;
834 	clkinfo.tick = tick;
835 	clkinfo.profhz = profhz;
836 	clkinfo.stathz = stathz ? stathz : hz;
837 	return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req));
838 }
839 
840 SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate,
841 	CTLTYPE_STRUCT|CTLFLAG_RD|CTLFLAG_MPSAFE,
842 	0, 0, sysctl_kern_clockrate, "S,clockinfo",
843 	"Rate and period of various kernel clocks");
844 
845 #ifdef SW_WATCHDOG
846 
847 static void
848 watchdog_config(void *unused __unused, u_int cmd, int *error)
849 {
850 	u_int u;
851 
852 	u = cmd & WD_INTERVAL;
853 	if (u >= WD_TO_1SEC) {
854 		watchdog_ticks = (1 << (u - WD_TO_1SEC)) * hz;
855 		watchdog_enabled = 1;
856 		*error = 0;
857 	} else {
858 		watchdog_enabled = 0;
859 	}
860 }
861 
862 /*
863  * Handle a watchdog timeout by dumping interrupt information and
864  * then either dropping to DDB or panicking.
865  */
866 static void
867 watchdog_fire(void)
868 {
869 	int nintr;
870 	uint64_t inttotal;
871 	u_long *curintr;
872 	char *curname;
873 
874 	curintr = intrcnt;
875 	curname = intrnames;
876 	inttotal = 0;
877 	nintr = sintrcnt / sizeof(u_long);
878 
879 	printf("interrupt                   total\n");
880 	while (--nintr >= 0) {
881 		if (*curintr)
882 			printf("%-12s %20lu\n", curname, *curintr);
883 		curname += strlen(curname) + 1;
884 		inttotal += *curintr++;
885 	}
886 	printf("Total        %20ju\n", (uintmax_t)inttotal);
887 
888 #if defined(KDB) && !defined(KDB_UNATTENDED)
889 	kdb_backtrace();
890 	kdb_enter(KDB_WHY_WATCHDOG, "watchdog timeout");
891 #else
892 	panic("watchdog timeout");
893 #endif
894 }
895 
896 #endif /* SW_WATCHDOG */
897