xref: /freebsd/sys/kern/kern_clock.c (revision d5b0e70f7e04d971691517ce1304d86a1e367e2e)
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 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 /* Spin-lock protecting profiling statistics. */
90 static struct mtx time_lock;
91 
92 SDT_PROVIDER_DECLARE(sched);
93 SDT_PROBE_DEFINE2(sched, , , tick, "struct thread *", "struct proc *");
94 
95 static int
96 sysctl_kern_cp_time(SYSCTL_HANDLER_ARGS)
97 {
98 	int error;
99 	long cp_time[CPUSTATES];
100 #ifdef SCTL_MASK32
101 	int i;
102 	unsigned int cp_time32[CPUSTATES];
103 #endif
104 
105 	read_cpu_time(cp_time);
106 #ifdef SCTL_MASK32
107 	if (req->flags & SCTL_MASK32) {
108 		if (!req->oldptr)
109 			return SYSCTL_OUT(req, 0, sizeof(cp_time32));
110 		for (i = 0; i < CPUSTATES; i++)
111 			cp_time32[i] = (unsigned int)cp_time[i];
112 		error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32));
113 	} else
114 #endif
115 	{
116 		if (!req->oldptr)
117 			return SYSCTL_OUT(req, 0, sizeof(cp_time));
118 		error = SYSCTL_OUT(req, cp_time, sizeof(cp_time));
119 	}
120 	return error;
121 }
122 
123 SYSCTL_PROC(_kern, OID_AUTO, cp_time, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE,
124     0,0, sysctl_kern_cp_time, "LU", "CPU time statistics");
125 
126 static long empty[CPUSTATES];
127 
128 static int
129 sysctl_kern_cp_times(SYSCTL_HANDLER_ARGS)
130 {
131 	struct pcpu *pcpu;
132 	int error;
133 	int c;
134 	long *cp_time;
135 #ifdef SCTL_MASK32
136 	unsigned int cp_time32[CPUSTATES];
137 	int i;
138 #endif
139 
140 	if (!req->oldptr) {
141 #ifdef SCTL_MASK32
142 		if (req->flags & SCTL_MASK32)
143 			return SYSCTL_OUT(req, 0, sizeof(cp_time32) * (mp_maxid + 1));
144 		else
145 #endif
146 			return SYSCTL_OUT(req, 0, sizeof(long) * CPUSTATES * (mp_maxid + 1));
147 	}
148 	for (error = 0, c = 0; error == 0 && c <= mp_maxid; c++) {
149 		if (!CPU_ABSENT(c)) {
150 			pcpu = pcpu_find(c);
151 			cp_time = pcpu->pc_cp_time;
152 		} else {
153 			cp_time = empty;
154 		}
155 #ifdef SCTL_MASK32
156 		if (req->flags & SCTL_MASK32) {
157 			for (i = 0; i < CPUSTATES; i++)
158 				cp_time32[i] = (unsigned int)cp_time[i];
159 			error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32));
160 		} else
161 #endif
162 			error = SYSCTL_OUT(req, cp_time, sizeof(long) * CPUSTATES);
163 	}
164 	return error;
165 }
166 
167 SYSCTL_PROC(_kern, OID_AUTO, cp_times, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE,
168     0,0, sysctl_kern_cp_times, "LU", "per-CPU time statistics");
169 
170 #ifdef DEADLKRES
171 static const char *blessed[] = {
172 	"getblk",
173 	"so_snd_sx",
174 	"so_rcv_sx",
175 	NULL
176 };
177 static int slptime_threshold = 1800;
178 static int blktime_threshold = 900;
179 static int sleepfreq = 3;
180 
181 static void
182 deadlres_td_on_lock(struct proc *p, struct thread *td, int blkticks)
183 {
184 	int tticks;
185 
186 	sx_assert(&allproc_lock, SX_LOCKED);
187 	PROC_LOCK_ASSERT(p, MA_OWNED);
188 	THREAD_LOCK_ASSERT(td, MA_OWNED);
189 	/*
190 	 * The thread should be blocked on a turnstile, simply check
191 	 * if the turnstile channel is in good state.
192 	 */
193 	MPASS(td->td_blocked != NULL);
194 
195 	tticks = ticks - td->td_blktick;
196 	if (tticks > blkticks)
197 		/*
198 		 * Accordingly with provided thresholds, this thread is stuck
199 		 * for too long on a turnstile.
200 		 */
201 		panic("%s: possible deadlock detected for %p (%s), "
202 		    "blocked for %d ticks\n", __func__,
203 		    td, sched_tdname(td), tticks);
204 }
205 
206 static void
207 deadlres_td_sleep_q(struct proc *p, struct thread *td, int slpticks)
208 {
209 	const void *wchan;
210 	int i, slptype, tticks;
211 
212 	sx_assert(&allproc_lock, SX_LOCKED);
213 	PROC_LOCK_ASSERT(p, MA_OWNED);
214 	THREAD_LOCK_ASSERT(td, MA_OWNED);
215 	/*
216 	 * Check if the thread is sleeping on a lock, otherwise skip the check.
217 	 * Drop the thread lock in order to avoid a LOR with the sleepqueue
218 	 * spinlock.
219 	 */
220 	wchan = td->td_wchan;
221 	tticks = ticks - td->td_slptick;
222 	slptype = sleepq_type(wchan);
223 	if ((slptype == SLEEPQ_SX || slptype == SLEEPQ_LK) &&
224 	    tticks > slpticks) {
225 		/*
226 		 * Accordingly with provided thresholds, this thread is stuck
227 		 * for too long on a sleepqueue.
228 		 * However, being on a sleepqueue, we might still check for the
229 		 * blessed list.
230 		 */
231 		for (i = 0; blessed[i] != NULL; i++)
232 			if (!strcmp(blessed[i], td->td_wmesg))
233 				return;
234 
235 		panic("%s: possible deadlock detected for %p (%s), "
236 		    "blocked for %d ticks\n", __func__,
237 		    td, sched_tdname(td), tticks);
238 	}
239 }
240 
241 static void
242 deadlkres(void)
243 {
244 	struct proc *p;
245 	struct thread *td;
246 	int blkticks, slpticks, tryl;
247 
248 	tryl = 0;
249 	for (;;) {
250 		blkticks = blktime_threshold * hz;
251 		slpticks = slptime_threshold * hz;
252 
253 		/*
254 		 * Avoid to sleep on the sx_lock in order to avoid a
255 		 * possible priority inversion problem leading to
256 		 * starvation.
257 		 * If the lock can't be held after 100 tries, panic.
258 		 */
259 		if (!sx_try_slock(&allproc_lock)) {
260 			if (tryl > 100)
261 				panic("%s: possible deadlock detected "
262 				    "on allproc_lock\n", __func__);
263 			tryl++;
264 			pause("allproc", sleepfreq * hz);
265 			continue;
266 		}
267 		tryl = 0;
268 		FOREACH_PROC_IN_SYSTEM(p) {
269 			PROC_LOCK(p);
270 			if (p->p_state == PRS_NEW) {
271 				PROC_UNLOCK(p);
272 				continue;
273 			}
274 			FOREACH_THREAD_IN_PROC(p, td) {
275 				thread_lock(td);
276 				if (TD_ON_LOCK(td))
277 					deadlres_td_on_lock(p, td,
278 					    blkticks);
279 				else if (TD_IS_SLEEPING(td))
280 					deadlres_td_sleep_q(p, td,
281 					    slpticks);
282 				thread_unlock(td);
283 			}
284 			PROC_UNLOCK(p);
285 		}
286 		sx_sunlock(&allproc_lock);
287 
288 		/* Sleep for sleepfreq seconds. */
289 		pause("-", sleepfreq * hz);
290 	}
291 }
292 
293 static struct kthread_desc deadlkres_kd = {
294 	"deadlkres",
295 	deadlkres,
296 	(struct thread **)NULL
297 };
298 
299 SYSINIT(deadlkres, SI_SUB_CLOCKS, SI_ORDER_ANY, kthread_start, &deadlkres_kd);
300 
301 static SYSCTL_NODE(_debug, OID_AUTO, deadlkres, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
302     "Deadlock resolver");
303 SYSCTL_INT(_debug_deadlkres, OID_AUTO, slptime_threshold, CTLFLAG_RW,
304     &slptime_threshold, 0,
305     "Number of seconds within is valid to sleep on a sleepqueue");
306 SYSCTL_INT(_debug_deadlkres, OID_AUTO, blktime_threshold, CTLFLAG_RW,
307     &blktime_threshold, 0,
308     "Number of seconds within is valid to block on a turnstile");
309 SYSCTL_INT(_debug_deadlkres, OID_AUTO, sleepfreq, CTLFLAG_RW, &sleepfreq, 0,
310     "Number of seconds between any deadlock resolver thread run");
311 #endif	/* DEADLKRES */
312 
313 void
314 read_cpu_time(long *cp_time)
315 {
316 	struct pcpu *pc;
317 	int i, j;
318 
319 	/* Sum up global cp_time[]. */
320 	bzero(cp_time, sizeof(long) * CPUSTATES);
321 	CPU_FOREACH(i) {
322 		pc = pcpu_find(i);
323 		for (j = 0; j < CPUSTATES; j++)
324 			cp_time[j] += pc->pc_cp_time[j];
325 	}
326 }
327 
328 #include <sys/watchdog.h>
329 
330 static int watchdog_ticks;
331 static int watchdog_enabled;
332 static void watchdog_fire(void);
333 static void watchdog_config(void *, u_int, int *);
334 
335 static void
336 watchdog_attach(void)
337 {
338 	EVENTHANDLER_REGISTER(watchdog_list, watchdog_config, NULL, 0);
339 }
340 
341 /*
342  * Clock handling routines.
343  *
344  * This code is written to operate with two timers that run independently of
345  * each other.
346  *
347  * The main timer, running hz times per second, is used to trigger interval
348  * timers, timeouts and rescheduling as needed.
349  *
350  * The second timer handles kernel and user profiling,
351  * and does resource use estimation.  If the second timer is programmable,
352  * it is randomized to avoid aliasing between the two clocks.  For example,
353  * the randomization prevents an adversary from always giving up the cpu
354  * just before its quantum expires.  Otherwise, it would never accumulate
355  * cpu ticks.  The mean frequency of the second timer is stathz.
356  *
357  * If no second timer exists, stathz will be zero; in this case we drive
358  * profiling and statistics off the main clock.  This WILL NOT be accurate;
359  * do not do it unless absolutely necessary.
360  *
361  * The statistics clock may (or may not) be run at a higher rate while
362  * profiling.  This profile clock runs at profhz.  We require that profhz
363  * be an integral multiple of stathz.
364  *
365  * If the statistics clock is running fast, it must be divided by the ratio
366  * profhz/stathz for statistics.  (For profiling, every tick counts.)
367  *
368  * Time-of-day is maintained using a "timecounter", which may or may
369  * not be related to the hardware generating the above mentioned
370  * interrupts.
371  */
372 
373 int	stathz;
374 int	profhz;
375 int	profprocs;
376 volatile int	ticks;
377 int	psratio;
378 
379 DPCPU_DEFINE_STATIC(int, pcputicks);	/* Per-CPU version of ticks. */
380 #ifdef DEVICE_POLLING
381 static int devpoll_run = 0;
382 #endif
383 
384 static void
385 ast_oweupc(struct thread *td, int tda __unused)
386 {
387 	if ((td->td_proc->p_flag & P_PROFIL) == 0)
388 		return;
389 	addupc_task(td, td->td_profil_addr, td->td_profil_ticks);
390 	td->td_profil_ticks = 0;
391 	td->td_pflags &= ~TDP_OWEUPC;
392 }
393 
394 static void
395 ast_alrm(struct thread *td, int tda __unused)
396 {
397 	struct proc *p;
398 
399 	p = td->td_proc;
400 	PROC_LOCK(p);
401 	kern_psignal(p, SIGVTALRM);
402 	PROC_UNLOCK(p);
403 }
404 
405 static void
406 ast_prof(struct thread *td, int tda __unused)
407 {
408 	struct proc *p;
409 
410 	p = td->td_proc;
411 	PROC_LOCK(p);
412 	kern_psignal(p, SIGPROF);
413 	PROC_UNLOCK(p);
414 }
415 
416 /*
417  * Initialize clock frequencies and start both clocks running.
418  */
419 static void
420 initclocks(void *dummy __unused)
421 {
422 	int i;
423 
424 	/*
425 	 * Set divisors to 1 (normal case) and let the machine-specific
426 	 * code do its bit.
427 	 */
428 	mtx_init(&time_lock, "time lock", NULL, MTX_DEF);
429 	cpu_initclocks();
430 
431 	/*
432 	 * Compute profhz/stathz, and fix profhz if needed.
433 	 */
434 	i = stathz ? stathz : hz;
435 	if (profhz == 0)
436 		profhz = i;
437 	psratio = profhz / i;
438 
439 	ast_register(TDA_OWEUPC, ASTR_ASTF_REQUIRED, 0, ast_oweupc);
440 	ast_register(TDA_ALRM, ASTR_ASTF_REQUIRED, 0, ast_alrm);
441 	ast_register(TDA_PROF, ASTR_ASTF_REQUIRED, 0, ast_prof);
442 
443 #ifdef SW_WATCHDOG
444 	/* Enable hardclock watchdog now, even if a hardware watchdog exists. */
445 	watchdog_attach();
446 #else
447 	/* Volunteer to run a software watchdog. */
448 	if (wdog_software_attach == NULL)
449 		wdog_software_attach = watchdog_attach;
450 #endif
451 }
452 SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL);
453 
454 static __noinline void
455 hardclock_itimer(struct thread *td, struct pstats *pstats, int cnt, int usermode)
456 {
457 	struct proc *p;
458 	int ast;
459 
460 	ast = 0;
461 	p = td->td_proc;
462 	if (usermode &&
463 	    timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) {
464 		PROC_ITIMLOCK(p);
465 		if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL],
466 		    tick * cnt) == 0)
467 			ast |= TDAI(TDA_ALRM);
468 		PROC_ITIMUNLOCK(p);
469 	}
470 	if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) {
471 		PROC_ITIMLOCK(p);
472 		if (itimerdecr(&pstats->p_timer[ITIMER_PROF],
473 		    tick * cnt) == 0)
474 			ast |= TDAI(TDA_PROF);
475 		PROC_ITIMUNLOCK(p);
476 	}
477 	if (ast != 0)
478 		ast_sched_mask(td, ast);
479 }
480 
481 void
482 hardclock(int cnt, int usermode)
483 {
484 	struct pstats *pstats;
485 	struct thread *td = curthread;
486 	struct proc *p = td->td_proc;
487 	int *t = DPCPU_PTR(pcputicks);
488 	int global, i, newticks;
489 
490 	/*
491 	 * Update per-CPU and possibly global ticks values.
492 	 */
493 	*t += cnt;
494 	global = ticks;
495 	do {
496 		newticks = *t - global;
497 		if (newticks <= 0) {
498 			if (newticks < -1)
499 				*t = global - 1;
500 			newticks = 0;
501 			break;
502 		}
503 	} while (!atomic_fcmpset_int(&ticks, &global, *t));
504 
505 	/*
506 	 * Run current process's virtual and profile time, as needed.
507 	 */
508 	pstats = p->p_stats;
509 	if (__predict_false(
510 	    timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) ||
511 	    timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)))
512 		hardclock_itimer(td, pstats, cnt, usermode);
513 
514 #ifdef	HWPMC_HOOKS
515 	if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid)))
516 		PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL);
517 	if (td->td_intr_frame != NULL)
518 		PMC_SOFT_CALL_TF( , , clock, hard, td->td_intr_frame);
519 #endif
520 	/* We are in charge to handle this tick duty. */
521 	if (newticks > 0) {
522 		tc_ticktock(newticks);
523 #ifdef DEVICE_POLLING
524 		/* Dangerous and no need to call these things concurrently. */
525 		if (atomic_cmpset_acq_int(&devpoll_run, 0, 1)) {
526 			/* This is very short and quick. */
527 			hardclock_device_poll();
528 			atomic_store_rel_int(&devpoll_run, 0);
529 		}
530 #endif /* DEVICE_POLLING */
531 		if (watchdog_enabled > 0) {
532 			i = atomic_fetchadd_int(&watchdog_ticks, -newticks);
533 			if (i > 0 && i <= newticks)
534 				watchdog_fire();
535 		}
536 		intr_event_handle(clk_intr_event, NULL);
537 	}
538 	if (curcpu == CPU_FIRST())
539 		cpu_tick_calibration();
540 	if (__predict_false(DPCPU_GET(epoch_cb_count)))
541 		GROUPTASK_ENQUEUE(DPCPU_PTR(epoch_cb_task));
542 }
543 
544 void
545 hardclock_sync(int cpu)
546 {
547 	int *t;
548 	KASSERT(!CPU_ABSENT(cpu), ("Absent CPU %d", cpu));
549 	t = DPCPU_ID_PTR(cpu, pcputicks);
550 
551 	*t = ticks;
552 }
553 
554 /*
555  * Regular integer scaling formula without losing precision:
556  */
557 #define	TIME_INT_SCALE(value, mul, div) \
558 	(((value) / (div)) * (mul) + (((value) % (div)) * (mul)) / (div))
559 
560 /*
561  * Macro for converting seconds and microseconds into actual ticks,
562  * based on the given hz value:
563  */
564 #define	TIME_TO_TICKS(sec, usec, hz) \
565 	((sec) * (hz) + TIME_INT_SCALE(usec, hz, 1 << 6) / (1000000 >> 6))
566 
567 #define	TIME_ASSERT_VALID_HZ(hz)	\
568 	_Static_assert(TIME_TO_TICKS(INT_MAX / (hz) - 1, 999999, hz) >= 0 && \
569 		       TIME_TO_TICKS(INT_MAX / (hz) - 1, 999999, hz) < INT_MAX,	\
570 		       "tvtohz() can overflow the regular integer type")
571 
572 /*
573  * Compile time assert the maximum and minimum values to fit into a
574  * regular integer when computing TIME_TO_TICKS():
575  */
576 TIME_ASSERT_VALID_HZ(HZ_MAXIMUM);
577 TIME_ASSERT_VALID_HZ(HZ_MINIMUM);
578 
579 /*
580  * The formula is mostly linear, but test some more common values just
581  * in case:
582  */
583 TIME_ASSERT_VALID_HZ(1024);
584 TIME_ASSERT_VALID_HZ(1000);
585 TIME_ASSERT_VALID_HZ(128);
586 TIME_ASSERT_VALID_HZ(100);
587 
588 /*
589  * Compute number of ticks representing the specified amount of time.
590  * If the specified time is negative, a value of 1 is returned. This
591  * function returns a value from 1 up to and including INT_MAX.
592  */
593 int
594 tvtohz(struct timeval *tv)
595 {
596 	int retval;
597 
598 	/*
599 	 * The values passed here may come from user-space and these
600 	 * checks ensure "tv_usec" is within its allowed range:
601 	 */
602 
603 	/* check for tv_usec underflow */
604 	if (__predict_false(tv->tv_usec < 0)) {
605 		tv->tv_sec += tv->tv_usec / 1000000;
606 		tv->tv_usec = tv->tv_usec % 1000000;
607 		/* convert tv_usec to a positive value */
608 		if (__predict_true(tv->tv_usec < 0)) {
609 			tv->tv_usec += 1000000;
610 			tv->tv_sec -= 1;
611 		}
612 	/* check for tv_usec overflow */
613 	} else if (__predict_false(tv->tv_usec >= 1000000)) {
614 		tv->tv_sec += tv->tv_usec / 1000000;
615 		tv->tv_usec = tv->tv_usec % 1000000;
616 	}
617 
618 	/* check for tv_sec underflow */
619 	if (__predict_false(tv->tv_sec < 0))
620 		return (1);
621 	/* check for tv_sec overflow (including room for the tv_usec part) */
622 	else if (__predict_false(tv->tv_sec >= tick_seconds_max))
623 		return (INT_MAX);
624 
625 	/* cast to "int" to avoid platform differences */
626 	retval = TIME_TO_TICKS((int)tv->tv_sec, (int)tv->tv_usec, hz);
627 
628 	/* add one additional tick */
629 	return (retval + 1);
630 }
631 
632 /*
633  * Start profiling on a process.
634  *
635  * Kernel profiling passes proc0 which never exits and hence
636  * keeps the profile clock running constantly.
637  */
638 void
639 startprofclock(struct proc *p)
640 {
641 
642 	PROC_LOCK_ASSERT(p, MA_OWNED);
643 	if (p->p_flag & P_STOPPROF)
644 		return;
645 	if ((p->p_flag & P_PROFIL) == 0) {
646 		p->p_flag |= P_PROFIL;
647 		mtx_lock(&time_lock);
648 		if (++profprocs == 1)
649 			cpu_startprofclock();
650 		mtx_unlock(&time_lock);
651 	}
652 }
653 
654 /*
655  * Stop profiling on a process.
656  */
657 void
658 stopprofclock(struct proc *p)
659 {
660 
661 	PROC_LOCK_ASSERT(p, MA_OWNED);
662 	if (p->p_flag & P_PROFIL) {
663 		if (p->p_profthreads != 0) {
664 			while (p->p_profthreads != 0) {
665 				p->p_flag |= P_STOPPROF;
666 				msleep(&p->p_profthreads, &p->p_mtx, PPAUSE,
667 				    "stopprof", 0);
668 			}
669 		}
670 		if ((p->p_flag & P_PROFIL) == 0)
671 			return;
672 		p->p_flag &= ~P_PROFIL;
673 		mtx_lock(&time_lock);
674 		if (--profprocs == 0)
675 			cpu_stopprofclock();
676 		mtx_unlock(&time_lock);
677 	}
678 }
679 
680 /*
681  * Statistics clock.  Updates rusage information and calls the scheduler
682  * to adjust priorities of the active thread.
683  *
684  * This should be called by all active processors.
685  */
686 void
687 statclock(int cnt, int usermode)
688 {
689 	struct rusage *ru;
690 	struct vmspace *vm;
691 	struct thread *td;
692 	struct proc *p;
693 	long rss;
694 	long *cp_time;
695 	uint64_t runtime, new_switchtime;
696 
697 	td = curthread;
698 	p = td->td_proc;
699 
700 	cp_time = (long *)PCPU_PTR(cp_time);
701 	if (usermode) {
702 		/*
703 		 * Charge the time as appropriate.
704 		 */
705 		td->td_uticks += cnt;
706 		if (p->p_nice > NZERO)
707 			cp_time[CP_NICE] += cnt;
708 		else
709 			cp_time[CP_USER] += cnt;
710 	} else {
711 		/*
712 		 * Came from kernel mode, so we were:
713 		 * - handling an interrupt,
714 		 * - doing syscall or trap work on behalf of the current
715 		 *   user process, or
716 		 * - spinning in the idle loop.
717 		 * Whichever it is, charge the time as appropriate.
718 		 * Note that we charge interrupts to the current process,
719 		 * regardless of whether they are ``for'' that process,
720 		 * so that we know how much of its real time was spent
721 		 * in ``non-process'' (i.e., interrupt) work.
722 		 */
723 		if ((td->td_pflags & TDP_ITHREAD) ||
724 		    td->td_intr_nesting_level >= 2) {
725 			td->td_iticks += cnt;
726 			cp_time[CP_INTR] += cnt;
727 		} else {
728 			td->td_pticks += cnt;
729 			td->td_sticks += cnt;
730 			if (!TD_IS_IDLETHREAD(td))
731 				cp_time[CP_SYS] += cnt;
732 			else
733 				cp_time[CP_IDLE] += cnt;
734 		}
735 	}
736 
737 	/* Update resource usage integrals and maximums. */
738 	MPASS(p->p_vmspace != NULL);
739 	vm = p->p_vmspace;
740 	ru = &td->td_ru;
741 	ru->ru_ixrss += pgtok(vm->vm_tsize) * cnt;
742 	ru->ru_idrss += pgtok(vm->vm_dsize) * cnt;
743 	ru->ru_isrss += pgtok(vm->vm_ssize) * cnt;
744 	rss = pgtok(vmspace_resident_count(vm));
745 	if (ru->ru_maxrss < rss)
746 		ru->ru_maxrss = rss;
747 	KTR_POINT2(KTR_SCHED, "thread", sched_tdname(td), "statclock",
748 	    "prio:%d", td->td_priority, "stathz:%d", (stathz)?stathz:hz);
749 	SDT_PROBE2(sched, , , tick, td, td->td_proc);
750 	thread_lock_flags(td, MTX_QUIET);
751 
752 	/*
753 	 * Compute the amount of time during which the current
754 	 * thread was running, and add that to its total so far.
755 	 */
756 	new_switchtime = cpu_ticks();
757 	runtime = new_switchtime - PCPU_GET(switchtime);
758 	td->td_runtime += runtime;
759 	td->td_incruntime += runtime;
760 	PCPU_SET(switchtime, new_switchtime);
761 
762 	sched_clock(td, cnt);
763 	thread_unlock(td);
764 #ifdef HWPMC_HOOKS
765 	if (td->td_intr_frame != NULL)
766 		PMC_SOFT_CALL_TF( , , clock, stat, td->td_intr_frame);
767 #endif
768 }
769 
770 void
771 profclock(int cnt, int usermode, uintfptr_t pc)
772 {
773 	struct thread *td;
774 
775 	td = curthread;
776 	if (usermode) {
777 		/*
778 		 * Came from user mode; CPU was in user state.
779 		 * If this process is being profiled, record the tick.
780 		 * if there is no related user location yet, don't
781 		 * bother trying to count it.
782 		 */
783 		if (td->td_proc->p_flag & P_PROFIL)
784 			addupc_intr(td, pc, cnt);
785 	}
786 #ifdef HWPMC_HOOKS
787 	if (td->td_intr_frame != NULL)
788 		PMC_SOFT_CALL_TF( , , clock, prof, td->td_intr_frame);
789 #endif
790 }
791 
792 /*
793  * Return information about system clocks.
794  */
795 static int
796 sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS)
797 {
798 	struct clockinfo clkinfo;
799 	/*
800 	 * Construct clockinfo structure.
801 	 */
802 	bzero(&clkinfo, sizeof(clkinfo));
803 	clkinfo.hz = hz;
804 	clkinfo.tick = tick;
805 	clkinfo.profhz = profhz;
806 	clkinfo.stathz = stathz ? stathz : hz;
807 	return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req));
808 }
809 
810 SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate,
811 	CTLTYPE_STRUCT|CTLFLAG_RD|CTLFLAG_MPSAFE,
812 	0, 0, sysctl_kern_clockrate, "S,clockinfo",
813 	"Rate and period of various kernel clocks");
814 
815 static void
816 watchdog_config(void *unused __unused, u_int cmd, int *error)
817 {
818 	u_int u;
819 
820 	u = cmd & WD_INTERVAL;
821 	if (u >= WD_TO_1SEC) {
822 		watchdog_ticks = (1 << (u - WD_TO_1SEC)) * hz;
823 		watchdog_enabled = 1;
824 		*error = 0;
825 	} else {
826 		watchdog_enabled = 0;
827 	}
828 }
829 
830 /*
831  * Handle a watchdog timeout by dumping interrupt information and
832  * then either dropping to DDB or panicking.
833  */
834 static void
835 watchdog_fire(void)
836 {
837 	int nintr;
838 	uint64_t inttotal;
839 	u_long *curintr;
840 	char *curname;
841 
842 	curintr = intrcnt;
843 	curname = intrnames;
844 	inttotal = 0;
845 	nintr = sintrcnt / sizeof(u_long);
846 
847 	printf("interrupt                   total\n");
848 	while (--nintr >= 0) {
849 		if (*curintr)
850 			printf("%-12s %20lu\n", curname, *curintr);
851 		curname += strlen(curname) + 1;
852 		inttotal += *curintr++;
853 	}
854 	printf("Total        %20ju\n", (uintmax_t)inttotal);
855 
856 #if defined(KDB) && !defined(KDB_UNATTENDED)
857 	kdb_backtrace();
858 	kdb_enter(KDB_WHY_WATCHDOG, "watchdog timeout");
859 #else
860 	panic("watchdog timeout");
861 #endif
862 }
863