xref: /freebsd/sys/kern/kern_clock.c (revision f7c32ed617858bcd22f8d1b03199099d50125721)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)kern_clock.c	8.5 (Berkeley) 1/21/94
37  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include "opt_kdb.h"
43 #include "opt_device_polling.h"
44 #include "opt_hwpmc_hooks.h"
45 #include "opt_ntp.h"
46 #include "opt_watchdog.h"
47 
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/callout.h>
51 #include <sys/epoch.h>
52 #include <sys/eventhandler.h>
53 #include <sys/gtaskqueue.h>
54 #include <sys/kdb.h>
55 #include <sys/kernel.h>
56 #include <sys/kthread.h>
57 #include <sys/ktr.h>
58 #include <sys/lock.h>
59 #include <sys/mutex.h>
60 #include <sys/proc.h>
61 #include <sys/resource.h>
62 #include <sys/resourcevar.h>
63 #include <sys/sched.h>
64 #include <sys/sdt.h>
65 #include <sys/signalvar.h>
66 #include <sys/sleepqueue.h>
67 #include <sys/smp.h>
68 #include <vm/vm.h>
69 #include <vm/pmap.h>
70 #include <vm/vm_map.h>
71 #include <sys/sysctl.h>
72 #include <sys/bus.h>
73 #include <sys/interrupt.h>
74 #include <sys/limits.h>
75 #include <sys/timetc.h>
76 
77 #ifdef GPROF
78 #include <sys/gmon.h>
79 #endif
80 
81 #ifdef HWPMC_HOOKS
82 #include <sys/pmckern.h>
83 PMC_SOFT_DEFINE( , , clock, hard);
84 PMC_SOFT_DEFINE( , , clock, stat);
85 PMC_SOFT_DEFINE_EX( , , clock, prof, \
86     cpu_startprofclock, cpu_stopprofclock);
87 #endif
88 
89 #ifdef DEVICE_POLLING
90 extern void hardclock_device_poll(void);
91 #endif /* DEVICE_POLLING */
92 
93 /* Spin-lock protecting profiling statistics. */
94 static struct mtx time_lock;
95 
96 SDT_PROVIDER_DECLARE(sched);
97 SDT_PROBE_DEFINE2(sched, , , tick, "struct thread *", "struct proc *");
98 
99 static int
100 sysctl_kern_cp_time(SYSCTL_HANDLER_ARGS)
101 {
102 	int error;
103 	long cp_time[CPUSTATES];
104 #ifdef SCTL_MASK32
105 	int i;
106 	unsigned int cp_time32[CPUSTATES];
107 #endif
108 
109 	read_cpu_time(cp_time);
110 #ifdef SCTL_MASK32
111 	if (req->flags & SCTL_MASK32) {
112 		if (!req->oldptr)
113 			return SYSCTL_OUT(req, 0, sizeof(cp_time32));
114 		for (i = 0; i < CPUSTATES; i++)
115 			cp_time32[i] = (unsigned int)cp_time[i];
116 		error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32));
117 	} else
118 #endif
119 	{
120 		if (!req->oldptr)
121 			return SYSCTL_OUT(req, 0, sizeof(cp_time));
122 		error = SYSCTL_OUT(req, cp_time, sizeof(cp_time));
123 	}
124 	return error;
125 }
126 
127 SYSCTL_PROC(_kern, OID_AUTO, cp_time, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE,
128     0,0, sysctl_kern_cp_time, "LU", "CPU time statistics");
129 
130 static long empty[CPUSTATES];
131 
132 static int
133 sysctl_kern_cp_times(SYSCTL_HANDLER_ARGS)
134 {
135 	struct pcpu *pcpu;
136 	int error;
137 	int c;
138 	long *cp_time;
139 #ifdef SCTL_MASK32
140 	unsigned int cp_time32[CPUSTATES];
141 	int i;
142 #endif
143 
144 	if (!req->oldptr) {
145 #ifdef SCTL_MASK32
146 		if (req->flags & SCTL_MASK32)
147 			return SYSCTL_OUT(req, 0, sizeof(cp_time32) * (mp_maxid + 1));
148 		else
149 #endif
150 			return SYSCTL_OUT(req, 0, sizeof(long) * CPUSTATES * (mp_maxid + 1));
151 	}
152 	for (error = 0, c = 0; error == 0 && c <= mp_maxid; c++) {
153 		if (!CPU_ABSENT(c)) {
154 			pcpu = pcpu_find(c);
155 			cp_time = pcpu->pc_cp_time;
156 		} else {
157 			cp_time = empty;
158 		}
159 #ifdef SCTL_MASK32
160 		if (req->flags & SCTL_MASK32) {
161 			for (i = 0; i < CPUSTATES; i++)
162 				cp_time32[i] = (unsigned int)cp_time[i];
163 			error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32));
164 		} else
165 #endif
166 			error = SYSCTL_OUT(req, cp_time, sizeof(long) * CPUSTATES);
167 	}
168 	return error;
169 }
170 
171 SYSCTL_PROC(_kern, OID_AUTO, cp_times, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE,
172     0,0, sysctl_kern_cp_times, "LU", "per-CPU time statistics");
173 
174 #ifdef DEADLKRES
175 static const char *blessed[] = {
176 	"getblk",
177 	"so_snd_sx",
178 	"so_rcv_sx",
179 	NULL
180 };
181 static int slptime_threshold = 1800;
182 static int blktime_threshold = 900;
183 static int sleepfreq = 3;
184 
185 static void
186 deadlres_td_on_lock(struct proc *p, struct thread *td, int blkticks)
187 {
188 	int tticks;
189 
190 	sx_assert(&allproc_lock, SX_LOCKED);
191 	PROC_LOCK_ASSERT(p, MA_OWNED);
192 	THREAD_LOCK_ASSERT(td, MA_OWNED);
193 	/*
194 	 * The thread should be blocked on a turnstile, simply check
195 	 * if the turnstile channel is in good state.
196 	 */
197 	MPASS(td->td_blocked != NULL);
198 
199 	tticks = ticks - td->td_blktick;
200 	if (tticks > blkticks)
201 		/*
202 		 * Accordingly with provided thresholds, this thread is stuck
203 		 * for too long on a turnstile.
204 		 */
205 		panic("%s: possible deadlock detected for %p (%s), "
206 		    "blocked for %d ticks\n", __func__,
207 		    td, sched_tdname(td), tticks);
208 }
209 
210 static void
211 deadlres_td_sleep_q(struct proc *p, struct thread *td, int slpticks)
212 {
213 	const void *wchan;
214 	int i, slptype, tticks;
215 
216 	sx_assert(&allproc_lock, SX_LOCKED);
217 	PROC_LOCK_ASSERT(p, MA_OWNED);
218 	THREAD_LOCK_ASSERT(td, MA_OWNED);
219 	/*
220 	 * Check if the thread is sleeping on a lock, otherwise skip the check.
221 	 * Drop the thread lock in order to avoid a LOR with the sleepqueue
222 	 * spinlock.
223 	 */
224 	wchan = td->td_wchan;
225 	tticks = ticks - td->td_slptick;
226 	slptype = sleepq_type(wchan);
227 	if ((slptype == SLEEPQ_SX || slptype == SLEEPQ_LK) &&
228 	    tticks > slpticks) {
229 		/*
230 		 * Accordingly with provided thresholds, this thread is stuck
231 		 * for too long on a sleepqueue.
232 		 * However, being on a sleepqueue, we might still check for the
233 		 * blessed list.
234 		 */
235 		for (i = 0; blessed[i] != NULL; i++)
236 			if (!strcmp(blessed[i], td->td_wmesg))
237 				return;
238 
239 		panic("%s: possible deadlock detected for %p (%s), "
240 		    "blocked for %d ticks\n", __func__,
241 		    td, sched_tdname(td), tticks);
242 	}
243 }
244 
245 static void
246 deadlkres(void)
247 {
248 	struct proc *p;
249 	struct thread *td;
250 	int blkticks, slpticks, tryl;
251 
252 	tryl = 0;
253 	for (;;) {
254 		blkticks = blktime_threshold * hz;
255 		slpticks = slptime_threshold * hz;
256 
257 		/*
258 		 * Avoid to sleep on the sx_lock in order to avoid a
259 		 * possible priority inversion problem leading to
260 		 * starvation.
261 		 * If the lock can't be held after 100 tries, panic.
262 		 */
263 		if (!sx_try_slock(&allproc_lock)) {
264 			if (tryl > 100)
265 				panic("%s: possible deadlock detected "
266 				    "on allproc_lock\n", __func__);
267 			tryl++;
268 			pause("allproc", sleepfreq * hz);
269 			continue;
270 		}
271 		tryl = 0;
272 		FOREACH_PROC_IN_SYSTEM(p) {
273 			PROC_LOCK(p);
274 			if (p->p_state == PRS_NEW) {
275 				PROC_UNLOCK(p);
276 				continue;
277 			}
278 			FOREACH_THREAD_IN_PROC(p, td) {
279 				thread_lock(td);
280 				if (TD_ON_LOCK(td))
281 					deadlres_td_on_lock(p, td,
282 					    blkticks);
283 				else if (TD_IS_SLEEPING(td))
284 					deadlres_td_sleep_q(p, td,
285 					    slpticks);
286 				thread_unlock(td);
287 			}
288 			PROC_UNLOCK(p);
289 		}
290 		sx_sunlock(&allproc_lock);
291 
292 		/* Sleep for sleepfreq seconds. */
293 		pause("-", sleepfreq * hz);
294 	}
295 }
296 
297 static struct kthread_desc deadlkres_kd = {
298 	"deadlkres",
299 	deadlkres,
300 	(struct thread **)NULL
301 };
302 
303 SYSINIT(deadlkres, SI_SUB_CLOCKS, SI_ORDER_ANY, kthread_start, &deadlkres_kd);
304 
305 static SYSCTL_NODE(_debug, OID_AUTO, deadlkres, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
306     "Deadlock resolver");
307 SYSCTL_INT(_debug_deadlkres, OID_AUTO, slptime_threshold, CTLFLAG_RW,
308     &slptime_threshold, 0,
309     "Number of seconds within is valid to sleep on a sleepqueue");
310 SYSCTL_INT(_debug_deadlkres, OID_AUTO, blktime_threshold, CTLFLAG_RW,
311     &blktime_threshold, 0,
312     "Number of seconds within is valid to block on a turnstile");
313 SYSCTL_INT(_debug_deadlkres, OID_AUTO, sleepfreq, CTLFLAG_RW, &sleepfreq, 0,
314     "Number of seconds between any deadlock resolver thread run");
315 #endif	/* DEADLKRES */
316 
317 void
318 read_cpu_time(long *cp_time)
319 {
320 	struct pcpu *pc;
321 	int i, j;
322 
323 	/* Sum up global cp_time[]. */
324 	bzero(cp_time, sizeof(long) * CPUSTATES);
325 	CPU_FOREACH(i) {
326 		pc = pcpu_find(i);
327 		for (j = 0; j < CPUSTATES; j++)
328 			cp_time[j] += pc->pc_cp_time[j];
329 	}
330 }
331 
332 #include <sys/watchdog.h>
333 
334 static int watchdog_ticks;
335 static int watchdog_enabled;
336 static void watchdog_fire(void);
337 static void watchdog_config(void *, u_int, int *);
338 
339 static void
340 watchdog_attach(void)
341 {
342 	EVENTHANDLER_REGISTER(watchdog_list, watchdog_config, NULL, 0);
343 }
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 DPCPU_DEFINE_STATIC(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 static void
392 initclocks(void *dummy __unused)
393 {
394 	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 
411 #ifdef SW_WATCHDOG
412 	/* Enable hardclock watchdog now, even if a hardware watchdog exists. */
413 	watchdog_attach();
414 #else
415 	/* Volunteer to run a software watchdog. */
416 	if (wdog_software_attach == NULL)
417 		wdog_software_attach = watchdog_attach;
418 #endif
419 }
420 SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL);
421 
422 static __noinline void
423 hardclock_itimer(struct thread *td, struct pstats *pstats, int cnt, int usermode)
424 {
425 	struct proc *p;
426 	int flags;
427 
428 	flags = 0;
429 	p = td->td_proc;
430 	if (usermode &&
431 	    timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) {
432 		PROC_ITIMLOCK(p);
433 		if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL],
434 		    tick * cnt) == 0)
435 			flags |= TDF_ALRMPEND | TDF_ASTPENDING;
436 		PROC_ITIMUNLOCK(p);
437 	}
438 	if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) {
439 		PROC_ITIMLOCK(p);
440 		if (itimerdecr(&pstats->p_timer[ITIMER_PROF],
441 		    tick * cnt) == 0)
442 			flags |= TDF_PROFPEND | TDF_ASTPENDING;
443 		PROC_ITIMUNLOCK(p);
444 	}
445 	if (flags != 0) {
446 		thread_lock(td);
447 		td->td_flags |= flags;
448 		thread_unlock(td);
449 	}
450 }
451 
452 void
453 hardclock(int cnt, int usermode)
454 {
455 	struct pstats *pstats;
456 	struct thread *td = curthread;
457 	struct proc *p = td->td_proc;
458 	int *t = DPCPU_PTR(pcputicks);
459 	int global, i, newticks;
460 
461 	/*
462 	 * Update per-CPU and possibly global ticks values.
463 	 */
464 	*t += cnt;
465 	global = ticks;
466 	do {
467 		newticks = *t - global;
468 		if (newticks <= 0) {
469 			if (newticks < -1)
470 				*t = global - 1;
471 			newticks = 0;
472 			break;
473 		}
474 	} while (!atomic_fcmpset_int(&ticks, &global, *t));
475 
476 	/*
477 	 * Run current process's virtual and profile time, as needed.
478 	 */
479 	pstats = p->p_stats;
480 	if (__predict_false(
481 	    timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) ||
482 	    timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)))
483 		hardclock_itimer(td, pstats, cnt, usermode);
484 
485 #ifdef	HWPMC_HOOKS
486 	if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid)))
487 		PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL);
488 	if (td->td_intr_frame != NULL)
489 		PMC_SOFT_CALL_TF( , , clock, hard, td->td_intr_frame);
490 #endif
491 	/* We are in charge to handle this tick duty. */
492 	if (newticks > 0) {
493 		tc_ticktock(newticks);
494 #ifdef DEVICE_POLLING
495 		/* Dangerous and no need to call these things concurrently. */
496 		if (atomic_cmpset_acq_int(&devpoll_run, 0, 1)) {
497 			/* This is very short and quick. */
498 			hardclock_device_poll();
499 			atomic_store_rel_int(&devpoll_run, 0);
500 		}
501 #endif /* DEVICE_POLLING */
502 		if (watchdog_enabled > 0) {
503 			i = atomic_fetchadd_int(&watchdog_ticks, -newticks);
504 			if (i > 0 && i <= newticks)
505 				watchdog_fire();
506 		}
507 		intr_event_handle(clk_intr_event, NULL);
508 	}
509 	if (curcpu == CPU_FIRST())
510 		cpu_tick_calibration();
511 	if (__predict_false(DPCPU_GET(epoch_cb_count)))
512 		GROUPTASK_ENQUEUE(DPCPU_PTR(epoch_cb_task));
513 }
514 
515 void
516 hardclock_sync(int cpu)
517 {
518 	int *t;
519 	KASSERT(!CPU_ABSENT(cpu), ("Absent CPU %d", cpu));
520 	t = DPCPU_ID_PTR(cpu, pcputicks);
521 
522 	*t = ticks;
523 }
524 
525 /*
526  * Compute number of ticks in the specified amount of time.
527  */
528 int
529 tvtohz(struct timeval *tv)
530 {
531 	unsigned long ticks;
532 	long sec, usec;
533 
534 	/*
535 	 * If the number of usecs in the whole seconds part of the time
536 	 * difference fits in a long, then the total number of usecs will
537 	 * fit in an unsigned long.  Compute the total and convert it to
538 	 * ticks, rounding up and adding 1 to allow for the current tick
539 	 * to expire.  Rounding also depends on unsigned long arithmetic
540 	 * to avoid overflow.
541 	 *
542 	 * Otherwise, if the number of ticks in the whole seconds part of
543 	 * the time difference fits in a long, then convert the parts to
544 	 * ticks separately and add, using similar rounding methods and
545 	 * overflow avoidance.  This method would work in the previous
546 	 * case but it is slightly slower and assumes that hz is integral.
547 	 *
548 	 * Otherwise, round the time difference down to the maximum
549 	 * representable value.
550 	 *
551 	 * If ints have 32 bits, then the maximum value for any timeout in
552 	 * 10ms ticks is 248 days.
553 	 */
554 	sec = tv->tv_sec;
555 	usec = tv->tv_usec;
556 	if (usec < 0) {
557 		sec--;
558 		usec += 1000000;
559 	}
560 	if (sec < 0) {
561 #ifdef DIAGNOSTIC
562 		if (usec > 0) {
563 			sec++;
564 			usec -= 1000000;
565 		}
566 		printf("tvotohz: negative time difference %ld sec %ld usec\n",
567 		       sec, usec);
568 #endif
569 		ticks = 1;
570 	} else if (sec <= LONG_MAX / 1000000)
571 		ticks = howmany(sec * 1000000 + (unsigned long)usec, tick) + 1;
572 	else if (sec <= LONG_MAX / hz)
573 		ticks = sec * hz
574 			+ howmany((unsigned long)usec, tick) + 1;
575 	else
576 		ticks = LONG_MAX;
577 	if (ticks > INT_MAX)
578 		ticks = INT_MAX;
579 	return ((int)ticks);
580 }
581 
582 /*
583  * Start profiling on a process.
584  *
585  * Kernel profiling passes proc0 which never exits and hence
586  * keeps the profile clock running constantly.
587  */
588 void
589 startprofclock(struct proc *p)
590 {
591 
592 	PROC_LOCK_ASSERT(p, MA_OWNED);
593 	if (p->p_flag & P_STOPPROF)
594 		return;
595 	if ((p->p_flag & P_PROFIL) == 0) {
596 		p->p_flag |= P_PROFIL;
597 		mtx_lock(&time_lock);
598 		if (++profprocs == 1)
599 			cpu_startprofclock();
600 		mtx_unlock(&time_lock);
601 	}
602 }
603 
604 /*
605  * Stop profiling on a process.
606  */
607 void
608 stopprofclock(struct proc *p)
609 {
610 
611 	PROC_LOCK_ASSERT(p, MA_OWNED);
612 	if (p->p_flag & P_PROFIL) {
613 		if (p->p_profthreads != 0) {
614 			while (p->p_profthreads != 0) {
615 				p->p_flag |= P_STOPPROF;
616 				msleep(&p->p_profthreads, &p->p_mtx, PPAUSE,
617 				    "stopprof", 0);
618 			}
619 		}
620 		if ((p->p_flag & P_PROFIL) == 0)
621 			return;
622 		p->p_flag &= ~P_PROFIL;
623 		mtx_lock(&time_lock);
624 		if (--profprocs == 0)
625 			cpu_stopprofclock();
626 		mtx_unlock(&time_lock);
627 	}
628 }
629 
630 /*
631  * Statistics clock.  Updates rusage information and calls the scheduler
632  * to adjust priorities of the active thread.
633  *
634  * This should be called by all active processors.
635  */
636 void
637 statclock(int cnt, int usermode)
638 {
639 	struct rusage *ru;
640 	struct vmspace *vm;
641 	struct thread *td;
642 	struct proc *p;
643 	long rss;
644 	long *cp_time;
645 	uint64_t runtime, new_switchtime;
646 
647 	td = curthread;
648 	p = td->td_proc;
649 
650 	cp_time = (long *)PCPU_PTR(cp_time);
651 	if (usermode) {
652 		/*
653 		 * Charge the time as appropriate.
654 		 */
655 		td->td_uticks += cnt;
656 		if (p->p_nice > NZERO)
657 			cp_time[CP_NICE] += cnt;
658 		else
659 			cp_time[CP_USER] += cnt;
660 	} else {
661 		/*
662 		 * Came from kernel mode, so we were:
663 		 * - handling an interrupt,
664 		 * - doing syscall or trap work on behalf of the current
665 		 *   user process, or
666 		 * - spinning in the idle loop.
667 		 * Whichever it is, charge the time as appropriate.
668 		 * Note that we charge interrupts to the current process,
669 		 * regardless of whether they are ``for'' that process,
670 		 * so that we know how much of its real time was spent
671 		 * in ``non-process'' (i.e., interrupt) work.
672 		 */
673 		if ((td->td_pflags & TDP_ITHREAD) ||
674 		    td->td_intr_nesting_level >= 2) {
675 			td->td_iticks += cnt;
676 			cp_time[CP_INTR] += cnt;
677 		} else {
678 			td->td_pticks += cnt;
679 			td->td_sticks += cnt;
680 			if (!TD_IS_IDLETHREAD(td))
681 				cp_time[CP_SYS] += cnt;
682 			else
683 				cp_time[CP_IDLE] += cnt;
684 		}
685 	}
686 
687 	/* Update resource usage integrals and maximums. */
688 	MPASS(p->p_vmspace != NULL);
689 	vm = p->p_vmspace;
690 	ru = &td->td_ru;
691 	ru->ru_ixrss += pgtok(vm->vm_tsize) * cnt;
692 	ru->ru_idrss += pgtok(vm->vm_dsize) * cnt;
693 	ru->ru_isrss += pgtok(vm->vm_ssize) * cnt;
694 	rss = pgtok(vmspace_resident_count(vm));
695 	if (ru->ru_maxrss < rss)
696 		ru->ru_maxrss = rss;
697 	KTR_POINT2(KTR_SCHED, "thread", sched_tdname(td), "statclock",
698 	    "prio:%d", td->td_priority, "stathz:%d", (stathz)?stathz:hz);
699 	SDT_PROBE2(sched, , , tick, td, td->td_proc);
700 	thread_lock_flags(td, MTX_QUIET);
701 
702 	/*
703 	 * Compute the amount of time during which the current
704 	 * thread was running, and add that to its total so far.
705 	 */
706 	new_switchtime = cpu_ticks();
707 	runtime = new_switchtime - PCPU_GET(switchtime);
708 	td->td_runtime += runtime;
709 	td->td_incruntime += runtime;
710 	PCPU_SET(switchtime, new_switchtime);
711 
712 	sched_clock(td, cnt);
713 	thread_unlock(td);
714 #ifdef HWPMC_HOOKS
715 	if (td->td_intr_frame != NULL)
716 		PMC_SOFT_CALL_TF( , , clock, stat, td->td_intr_frame);
717 #endif
718 }
719 
720 void
721 profclock(int cnt, int usermode, uintfptr_t pc)
722 {
723 	struct thread *td;
724 #ifdef GPROF
725 	struct gmonparam *g;
726 	uintfptr_t i;
727 #endif
728 
729 	td = curthread;
730 	if (usermode) {
731 		/*
732 		 * Came from user mode; CPU was in user state.
733 		 * If this process is being profiled, record the tick.
734 		 * if there is no related user location yet, don't
735 		 * bother trying to count it.
736 		 */
737 		if (td->td_proc->p_flag & P_PROFIL)
738 			addupc_intr(td, pc, cnt);
739 	}
740 #ifdef GPROF
741 	else {
742 		/*
743 		 * Kernel statistics are just like addupc_intr, only easier.
744 		 */
745 		g = &_gmonparam;
746 		if (g->state == GMON_PROF_ON && pc >= g->lowpc) {
747 			i = PC_TO_I(g, pc);
748 			if (i < g->textsize) {
749 				KCOUNT(g, i) += cnt;
750 			}
751 		}
752 	}
753 #endif
754 #ifdef HWPMC_HOOKS
755 	if (td->td_intr_frame != NULL)
756 		PMC_SOFT_CALL_TF( , , clock, prof, td->td_intr_frame);
757 #endif
758 }
759 
760 /*
761  * Return information about system clocks.
762  */
763 static int
764 sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS)
765 {
766 	struct clockinfo clkinfo;
767 	/*
768 	 * Construct clockinfo structure.
769 	 */
770 	bzero(&clkinfo, sizeof(clkinfo));
771 	clkinfo.hz = hz;
772 	clkinfo.tick = tick;
773 	clkinfo.profhz = profhz;
774 	clkinfo.stathz = stathz ? stathz : hz;
775 	return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req));
776 }
777 
778 SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate,
779 	CTLTYPE_STRUCT|CTLFLAG_RD|CTLFLAG_MPSAFE,
780 	0, 0, sysctl_kern_clockrate, "S,clockinfo",
781 	"Rate and period of various kernel clocks");
782 
783 static void
784 watchdog_config(void *unused __unused, u_int cmd, int *error)
785 {
786 	u_int u;
787 
788 	u = cmd & WD_INTERVAL;
789 	if (u >= WD_TO_1SEC) {
790 		watchdog_ticks = (1 << (u - WD_TO_1SEC)) * hz;
791 		watchdog_enabled = 1;
792 		*error = 0;
793 	} else {
794 		watchdog_enabled = 0;
795 	}
796 }
797 
798 /*
799  * Handle a watchdog timeout by dumping interrupt information and
800  * then either dropping to DDB or panicking.
801  */
802 static void
803 watchdog_fire(void)
804 {
805 	int nintr;
806 	uint64_t inttotal;
807 	u_long *curintr;
808 	char *curname;
809 
810 	curintr = intrcnt;
811 	curname = intrnames;
812 	inttotal = 0;
813 	nintr = sintrcnt / sizeof(u_long);
814 
815 	printf("interrupt                   total\n");
816 	while (--nintr >= 0) {
817 		if (*curintr)
818 			printf("%-12s %20lu\n", curname, *curintr);
819 		curname += strlen(curname) + 1;
820 		inttotal += *curintr++;
821 	}
822 	printf("Total        %20ju\n", (uintmax_t)inttotal);
823 
824 #if defined(KDB) && !defined(KDB_UNATTENDED)
825 	kdb_backtrace();
826 	kdb_enter(KDB_WHY_WATCHDOG, "watchdog timeout");
827 #else
828 	panic("watchdog timeout");
829 #endif
830 }
831