xref: /freebsd/sys/kern/kern_clock.c (revision edf8578117e8844e02c0121147f45e4609b30680)
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 #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/epoch.h>
50 #include <sys/eventhandler.h>
51 #include <sys/gtaskqueue.h>
52 #include <sys/kdb.h>
53 #include <sys/kernel.h>
54 #include <sys/kthread.h>
55 #include <sys/ktr.h>
56 #include <sys/lock.h>
57 #include <sys/mutex.h>
58 #include <sys/proc.h>
59 #include <sys/resource.h>
60 #include <sys/resourcevar.h>
61 #include <sys/sched.h>
62 #include <sys/sdt.h>
63 #include <sys/signalvar.h>
64 #include <sys/sleepqueue.h>
65 #include <sys/smp.h>
66 #include <vm/vm.h>
67 #include <vm/pmap.h>
68 #include <vm/vm_map.h>
69 #include <sys/sysctl.h>
70 #include <sys/bus.h>
71 #include <sys/interrupt.h>
72 #include <sys/limits.h>
73 #include <sys/timetc.h>
74 
75 #ifdef HWPMC_HOOKS
76 #include <sys/pmckern.h>
77 PMC_SOFT_DEFINE( , , clock, hard);
78 PMC_SOFT_DEFINE( , , clock, stat);
79 PMC_SOFT_DEFINE_EX( , , clock, prof, \
80     cpu_startprofclock, cpu_stopprofclock);
81 #endif
82 
83 #ifdef DEVICE_POLLING
84 extern void hardclock_device_poll(void);
85 #endif /* DEVICE_POLLING */
86 
87 /* Spin-lock protecting profiling statistics. */
88 static struct mtx time_lock;
89 
90 SDT_PROVIDER_DECLARE(sched);
91 SDT_PROBE_DEFINE2(sched, , , tick, "struct thread *", "struct proc *");
92 
93 static int
94 sysctl_kern_cp_time(SYSCTL_HANDLER_ARGS)
95 {
96 	int error;
97 	long cp_time[CPUSTATES];
98 #ifdef SCTL_MASK32
99 	int i;
100 	unsigned int cp_time32[CPUSTATES];
101 #endif
102 
103 	read_cpu_time(cp_time);
104 #ifdef SCTL_MASK32
105 	if (req->flags & SCTL_MASK32) {
106 		if (!req->oldptr)
107 			return SYSCTL_OUT(req, 0, sizeof(cp_time32));
108 		for (i = 0; i < CPUSTATES; i++)
109 			cp_time32[i] = (unsigned int)cp_time[i];
110 		error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32));
111 	} else
112 #endif
113 	{
114 		if (!req->oldptr)
115 			return SYSCTL_OUT(req, 0, sizeof(cp_time));
116 		error = SYSCTL_OUT(req, cp_time, sizeof(cp_time));
117 	}
118 	return error;
119 }
120 
121 SYSCTL_PROC(_kern, OID_AUTO, cp_time, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE,
122     0,0, sysctl_kern_cp_time, "LU", "CPU time statistics");
123 
124 static long empty[CPUSTATES];
125 
126 static int
127 sysctl_kern_cp_times(SYSCTL_HANDLER_ARGS)
128 {
129 	struct pcpu *pcpu;
130 	int error;
131 	int c;
132 	long *cp_time;
133 #ifdef SCTL_MASK32
134 	unsigned int cp_time32[CPUSTATES];
135 	int i;
136 #endif
137 
138 	if (!req->oldptr) {
139 #ifdef SCTL_MASK32
140 		if (req->flags & SCTL_MASK32)
141 			return SYSCTL_OUT(req, 0, sizeof(cp_time32) * (mp_maxid + 1));
142 		else
143 #endif
144 			return SYSCTL_OUT(req, 0, sizeof(long) * CPUSTATES * (mp_maxid + 1));
145 	}
146 	for (error = 0, c = 0; error == 0 && c <= mp_maxid; c++) {
147 		if (!CPU_ABSENT(c)) {
148 			pcpu = pcpu_find(c);
149 			cp_time = pcpu->pc_cp_time;
150 		} else {
151 			cp_time = empty;
152 		}
153 #ifdef SCTL_MASK32
154 		if (req->flags & SCTL_MASK32) {
155 			for (i = 0; i < CPUSTATES; i++)
156 				cp_time32[i] = (unsigned int)cp_time[i];
157 			error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32));
158 		} else
159 #endif
160 			error = SYSCTL_OUT(req, cp_time, sizeof(long) * CPUSTATES);
161 	}
162 	return error;
163 }
164 
165 SYSCTL_PROC(_kern, OID_AUTO, cp_times, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE,
166     0,0, sysctl_kern_cp_times, "LU", "per-CPU time statistics");
167 
168 #ifdef DEADLKRES
169 static const char *blessed[] = {
170 	"getblk",
171 	"so_snd_sx",
172 	"so_rcv_sx",
173 	NULL
174 };
175 static int slptime_threshold = 1800;
176 static int blktime_threshold = 900;
177 static int sleepfreq = 3;
178 
179 static void
180 deadlres_td_on_lock(struct proc *p, struct thread *td, int blkticks)
181 {
182 	int tticks;
183 
184 	sx_assert(&allproc_lock, SX_LOCKED);
185 	PROC_LOCK_ASSERT(p, MA_OWNED);
186 	THREAD_LOCK_ASSERT(td, MA_OWNED);
187 	/*
188 	 * The thread should be blocked on a turnstile, simply check
189 	 * if the turnstile channel is in good state.
190 	 */
191 	MPASS(td->td_blocked != NULL);
192 
193 	tticks = ticks - td->td_blktick;
194 	if (tticks > blkticks)
195 		/*
196 		 * Accordingly with provided thresholds, this thread is stuck
197 		 * for too long on a turnstile.
198 		 */
199 		panic("%s: possible deadlock detected for %p (%s), "
200 		    "blocked for %d ticks\n", __func__,
201 		    td, sched_tdname(td), tticks);
202 }
203 
204 static void
205 deadlres_td_sleep_q(struct proc *p, struct thread *td, int slpticks)
206 {
207 	const void *wchan;
208 	int i, slptype, tticks;
209 
210 	sx_assert(&allproc_lock, SX_LOCKED);
211 	PROC_LOCK_ASSERT(p, MA_OWNED);
212 	THREAD_LOCK_ASSERT(td, MA_OWNED);
213 	/*
214 	 * Check if the thread is sleeping on a lock, otherwise skip the check.
215 	 * Drop the thread lock in order to avoid a LOR with the sleepqueue
216 	 * spinlock.
217 	 */
218 	wchan = td->td_wchan;
219 	tticks = ticks - td->td_slptick;
220 	slptype = sleepq_type(wchan);
221 	if ((slptype == SLEEPQ_SX || slptype == SLEEPQ_LK) &&
222 	    tticks > slpticks) {
223 		/*
224 		 * Accordingly with provided thresholds, this thread is stuck
225 		 * for too long on a sleepqueue.
226 		 * However, being on a sleepqueue, we might still check for the
227 		 * blessed list.
228 		 */
229 		for (i = 0; blessed[i] != NULL; i++)
230 			if (!strcmp(blessed[i], td->td_wmesg))
231 				return;
232 
233 		panic("%s: possible deadlock detected for %p (%s), "
234 		    "blocked for %d ticks\n", __func__,
235 		    td, sched_tdname(td), tticks);
236 	}
237 }
238 
239 static void
240 deadlkres(void)
241 {
242 	struct proc *p;
243 	struct thread *td;
244 	int blkticks, slpticks, tryl;
245 
246 	tryl = 0;
247 	for (;;) {
248 		blkticks = blktime_threshold * hz;
249 		slpticks = slptime_threshold * hz;
250 
251 		/*
252 		 * Avoid to sleep on the sx_lock in order to avoid a
253 		 * possible priority inversion problem leading to
254 		 * starvation.
255 		 * If the lock can't be held after 100 tries, panic.
256 		 */
257 		if (!sx_try_slock(&allproc_lock)) {
258 			if (tryl > 100)
259 				panic("%s: possible deadlock detected "
260 				    "on allproc_lock\n", __func__);
261 			tryl++;
262 			pause("allproc", sleepfreq * hz);
263 			continue;
264 		}
265 		tryl = 0;
266 		FOREACH_PROC_IN_SYSTEM(p) {
267 			PROC_LOCK(p);
268 			if (p->p_state == PRS_NEW) {
269 				PROC_UNLOCK(p);
270 				continue;
271 			}
272 			FOREACH_THREAD_IN_PROC(p, td) {
273 				thread_lock(td);
274 				if (TD_ON_LOCK(td))
275 					deadlres_td_on_lock(p, td,
276 					    blkticks);
277 				else if (TD_IS_SLEEPING(td))
278 					deadlres_td_sleep_q(p, td,
279 					    slpticks);
280 				thread_unlock(td);
281 			}
282 			PROC_UNLOCK(p);
283 		}
284 		sx_sunlock(&allproc_lock);
285 
286 		/* Sleep for sleepfreq seconds. */
287 		pause("-", sleepfreq * hz);
288 	}
289 }
290 
291 static struct kthread_desc deadlkres_kd = {
292 	"deadlkres",
293 	deadlkres,
294 	(struct thread **)NULL
295 };
296 
297 SYSINIT(deadlkres, SI_SUB_CLOCKS, SI_ORDER_ANY, kthread_start, &deadlkres_kd);
298 
299 static SYSCTL_NODE(_debug, OID_AUTO, deadlkres, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
300     "Deadlock resolver");
301 SYSCTL_INT(_debug_deadlkres, OID_AUTO, slptime_threshold, CTLFLAG_RWTUN,
302     &slptime_threshold, 0,
303     "Number of seconds within is valid to sleep on a sleepqueue");
304 SYSCTL_INT(_debug_deadlkres, OID_AUTO, blktime_threshold, CTLFLAG_RWTUN,
305     &blktime_threshold, 0,
306     "Number of seconds within is valid to block on a turnstile");
307 SYSCTL_INT(_debug_deadlkres, OID_AUTO, sleepfreq, CTLFLAG_RWTUN, &sleepfreq, 0,
308     "Number of seconds between any deadlock resolver thread run");
309 #endif	/* DEADLKRES */
310 
311 void
312 read_cpu_time(long *cp_time)
313 {
314 	struct pcpu *pc;
315 	int i, j;
316 
317 	/* Sum up global cp_time[]. */
318 	bzero(cp_time, sizeof(long) * CPUSTATES);
319 	CPU_FOREACH(i) {
320 		pc = pcpu_find(i);
321 		for (j = 0; j < CPUSTATES; j++)
322 			cp_time[j] += pc->pc_cp_time[j];
323 	}
324 }
325 
326 #include <sys/watchdog.h>
327 
328 static int watchdog_ticks;
329 static int watchdog_enabled;
330 static void watchdog_fire(void);
331 static void watchdog_config(void *, u_int, int *);
332 
333 static void
334 watchdog_attach(void)
335 {
336 	EVENTHANDLER_REGISTER(watchdog_list, watchdog_config, NULL, 0);
337 }
338 
339 /*
340  * Clock handling routines.
341  *
342  * This code is written to operate with two timers that run independently of
343  * each other.
344  *
345  * The main timer, running hz times per second, is used to trigger interval
346  * timers, timeouts and rescheduling as needed.
347  *
348  * The second timer handles kernel and user profiling,
349  * and does resource use estimation.  If the second timer is programmable,
350  * it is randomized to avoid aliasing between the two clocks.  For example,
351  * the randomization prevents an adversary from always giving up the cpu
352  * just before its quantum expires.  Otherwise, it would never accumulate
353  * cpu ticks.  The mean frequency of the second timer is stathz.
354  *
355  * If no second timer exists, stathz will be zero; in this case we drive
356  * profiling and statistics off the main clock.  This WILL NOT be accurate;
357  * do not do it unless absolutely necessary.
358  *
359  * The statistics clock may (or may not) be run at a higher rate while
360  * profiling.  This profile clock runs at profhz.  We require that profhz
361  * be an integral multiple of stathz.
362  *
363  * If the statistics clock is running fast, it must be divided by the ratio
364  * profhz/stathz for statistics.  (For profiling, every tick counts.)
365  *
366  * Time-of-day is maintained using a "timecounter", which may or may
367  * not be related to the hardware generating the above mentioned
368  * interrupts.
369  */
370 
371 int	stathz;
372 int	profhz;
373 int	profprocs;
374 volatile int	ticks;
375 int	psratio;
376 
377 DPCPU_DEFINE_STATIC(int, pcputicks);	/* Per-CPU version of ticks. */
378 #ifdef DEVICE_POLLING
379 static int devpoll_run = 0;
380 #endif
381 
382 static void
383 ast_oweupc(struct thread *td, int tda __unused)
384 {
385 	if ((td->td_proc->p_flag & P_PROFIL) == 0)
386 		return;
387 	addupc_task(td, td->td_profil_addr, td->td_profil_ticks);
388 	td->td_profil_ticks = 0;
389 	td->td_pflags &= ~TDP_OWEUPC;
390 }
391 
392 static void
393 ast_alrm(struct thread *td, int tda __unused)
394 {
395 	struct proc *p;
396 
397 	p = td->td_proc;
398 	PROC_LOCK(p);
399 	kern_psignal(p, SIGVTALRM);
400 	PROC_UNLOCK(p);
401 }
402 
403 static void
404 ast_prof(struct thread *td, int tda __unused)
405 {
406 	struct proc *p;
407 
408 	p = td->td_proc;
409 	PROC_LOCK(p);
410 	kern_psignal(p, SIGPROF);
411 	PROC_UNLOCK(p);
412 }
413 
414 /*
415  * Initialize clock frequencies and start both clocks running.
416  */
417 static void
418 initclocks(void *dummy __unused)
419 {
420 	int i;
421 
422 	/*
423 	 * Set divisors to 1 (normal case) and let the machine-specific
424 	 * code do its bit.
425 	 */
426 	mtx_init(&time_lock, "time lock", NULL, MTX_DEF);
427 	cpu_initclocks();
428 
429 	/*
430 	 * Compute profhz/stathz, and fix profhz if needed.
431 	 */
432 	i = stathz ? stathz : hz;
433 	if (profhz == 0)
434 		profhz = i;
435 	psratio = profhz / i;
436 
437 	ast_register(TDA_OWEUPC, ASTR_ASTF_REQUIRED, 0, ast_oweupc);
438 	ast_register(TDA_ALRM, ASTR_ASTF_REQUIRED, 0, ast_alrm);
439 	ast_register(TDA_PROF, ASTR_ASTF_REQUIRED, 0, ast_prof);
440 
441 #ifdef SW_WATCHDOG
442 	/* Enable hardclock watchdog now, even if a hardware watchdog exists. */
443 	watchdog_attach();
444 #else
445 	/* Volunteer to run a software watchdog. */
446 	if (wdog_software_attach == NULL)
447 		wdog_software_attach = watchdog_attach;
448 #endif
449 }
450 SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL);
451 
452 static __noinline void
453 hardclock_itimer(struct thread *td, struct pstats *pstats, int cnt, int usermode)
454 {
455 	struct proc *p;
456 	int ast;
457 
458 	ast = 0;
459 	p = td->td_proc;
460 	if (usermode &&
461 	    timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) {
462 		PROC_ITIMLOCK(p);
463 		if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL],
464 		    tick * cnt) == 0)
465 			ast |= TDAI(TDA_ALRM);
466 		PROC_ITIMUNLOCK(p);
467 	}
468 	if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) {
469 		PROC_ITIMLOCK(p);
470 		if (itimerdecr(&pstats->p_timer[ITIMER_PROF],
471 		    tick * cnt) == 0)
472 			ast |= TDAI(TDA_PROF);
473 		PROC_ITIMUNLOCK(p);
474 	}
475 	if (ast != 0)
476 		ast_sched_mask(td, ast);
477 }
478 
479 void
480 hardclock(int cnt, int usermode)
481 {
482 	struct pstats *pstats;
483 	struct thread *td = curthread;
484 	struct proc *p = td->td_proc;
485 	int *t = DPCPU_PTR(pcputicks);
486 	int global, i, newticks;
487 
488 	/*
489 	 * Update per-CPU and possibly global ticks values.
490 	 */
491 	*t += cnt;
492 	global = ticks;
493 	do {
494 		newticks = *t - global;
495 		if (newticks <= 0) {
496 			if (newticks < -1)
497 				*t = global - 1;
498 			newticks = 0;
499 			break;
500 		}
501 	} while (!atomic_fcmpset_int(&ticks, &global, *t));
502 
503 	/*
504 	 * Run current process's virtual and profile time, as needed.
505 	 */
506 	pstats = p->p_stats;
507 	if (__predict_false(
508 	    timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) ||
509 	    timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)))
510 		hardclock_itimer(td, pstats, cnt, usermode);
511 
512 #ifdef	HWPMC_HOOKS
513 	if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid)))
514 		PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL);
515 	if (td->td_intr_frame != NULL)
516 		PMC_SOFT_CALL_TF( , , clock, hard, td->td_intr_frame);
517 #endif
518 	/* We are in charge to handle this tick duty. */
519 	if (newticks > 0) {
520 		tc_ticktock(newticks);
521 #ifdef DEVICE_POLLING
522 		/* Dangerous and no need to call these things concurrently. */
523 		if (atomic_cmpset_acq_int(&devpoll_run, 0, 1)) {
524 			/* This is very short and quick. */
525 			hardclock_device_poll();
526 			atomic_store_rel_int(&devpoll_run, 0);
527 		}
528 #endif /* DEVICE_POLLING */
529 		if (watchdog_enabled > 0) {
530 			i = atomic_fetchadd_int(&watchdog_ticks, -newticks);
531 			if (i > 0 && i <= newticks)
532 				watchdog_fire();
533 		}
534 		intr_event_handle(clk_intr_event, NULL);
535 	}
536 	if (curcpu == CPU_FIRST())
537 		cpu_tick_calibration();
538 	if (__predict_false(DPCPU_GET(epoch_cb_count)))
539 		GROUPTASK_ENQUEUE(DPCPU_PTR(epoch_cb_task));
540 }
541 
542 void
543 hardclock_sync(int cpu)
544 {
545 	int *t;
546 	KASSERT(!CPU_ABSENT(cpu), ("Absent CPU %d", cpu));
547 	t = DPCPU_ID_PTR(cpu, pcputicks);
548 
549 	*t = ticks;
550 }
551 
552 /*
553  * Regular integer scaling formula without losing precision:
554  */
555 #define	TIME_INT_SCALE(value, mul, div) \
556 	(((value) / (div)) * (mul) + (((value) % (div)) * (mul)) / (div))
557 
558 /*
559  * Macro for converting seconds and microseconds into actual ticks,
560  * based on the given hz value:
561  */
562 #define	TIME_TO_TICKS(sec, usec, hz) \
563 	((sec) * (hz) + TIME_INT_SCALE(usec, hz, 1 << 6) / (1000000 >> 6))
564 
565 #define	TIME_ASSERT_VALID_HZ(hz)	\
566 	_Static_assert(TIME_TO_TICKS(INT_MAX / (hz) - 1, 999999, hz) >= 0 && \
567 		       TIME_TO_TICKS(INT_MAX / (hz) - 1, 999999, hz) < INT_MAX,	\
568 		       "tvtohz() can overflow the regular integer type")
569 
570 /*
571  * Compile time assert the maximum and minimum values to fit into a
572  * regular integer when computing TIME_TO_TICKS():
573  */
574 TIME_ASSERT_VALID_HZ(HZ_MAXIMUM);
575 TIME_ASSERT_VALID_HZ(HZ_MINIMUM);
576 
577 /*
578  * The formula is mostly linear, but test some more common values just
579  * in case:
580  */
581 TIME_ASSERT_VALID_HZ(1024);
582 TIME_ASSERT_VALID_HZ(1000);
583 TIME_ASSERT_VALID_HZ(128);
584 TIME_ASSERT_VALID_HZ(100);
585 
586 /*
587  * Compute number of ticks representing the specified amount of time.
588  * If the specified time is negative, a value of 1 is returned. This
589  * function returns a value from 1 up to and including INT_MAX.
590  */
591 int
592 tvtohz(struct timeval *tv)
593 {
594 	int retval;
595 
596 	/*
597 	 * The values passed here may come from user-space and these
598 	 * checks ensure "tv_usec" is within its allowed range:
599 	 */
600 
601 	/* check for tv_usec underflow */
602 	if (__predict_false(tv->tv_usec < 0)) {
603 		tv->tv_sec += tv->tv_usec / 1000000;
604 		tv->tv_usec = tv->tv_usec % 1000000;
605 		/* convert tv_usec to a positive value */
606 		if (__predict_true(tv->tv_usec < 0)) {
607 			tv->tv_usec += 1000000;
608 			tv->tv_sec -= 1;
609 		}
610 	/* check for tv_usec overflow */
611 	} else if (__predict_false(tv->tv_usec >= 1000000)) {
612 		tv->tv_sec += tv->tv_usec / 1000000;
613 		tv->tv_usec = tv->tv_usec % 1000000;
614 	}
615 
616 	/* check for tv_sec underflow */
617 	if (__predict_false(tv->tv_sec < 0))
618 		return (1);
619 	/* check for tv_sec overflow (including room for the tv_usec part) */
620 	else if (__predict_false(tv->tv_sec >= tick_seconds_max))
621 		return (INT_MAX);
622 
623 	/* cast to "int" to avoid platform differences */
624 	retval = TIME_TO_TICKS((int)tv->tv_sec, (int)tv->tv_usec, hz);
625 
626 	/* add one additional tick */
627 	return (retval + 1);
628 }
629 
630 /*
631  * Start profiling on a process.
632  *
633  * Kernel profiling passes proc0 which never exits and hence
634  * keeps the profile clock running constantly.
635  */
636 void
637 startprofclock(struct proc *p)
638 {
639 
640 	PROC_LOCK_ASSERT(p, MA_OWNED);
641 	if (p->p_flag & P_STOPPROF)
642 		return;
643 	if ((p->p_flag & P_PROFIL) == 0) {
644 		p->p_flag |= P_PROFIL;
645 		mtx_lock(&time_lock);
646 		if (++profprocs == 1)
647 			cpu_startprofclock();
648 		mtx_unlock(&time_lock);
649 	}
650 }
651 
652 /*
653  * Stop profiling on a process.
654  */
655 void
656 stopprofclock(struct proc *p)
657 {
658 
659 	PROC_LOCK_ASSERT(p, MA_OWNED);
660 	if (p->p_flag & P_PROFIL) {
661 		if (p->p_profthreads != 0) {
662 			while (p->p_profthreads != 0) {
663 				p->p_flag |= P_STOPPROF;
664 				msleep(&p->p_profthreads, &p->p_mtx, PPAUSE,
665 				    "stopprof", 0);
666 			}
667 		}
668 		if ((p->p_flag & P_PROFIL) == 0)
669 			return;
670 		p->p_flag &= ~P_PROFIL;
671 		mtx_lock(&time_lock);
672 		if (--profprocs == 0)
673 			cpu_stopprofclock();
674 		mtx_unlock(&time_lock);
675 	}
676 }
677 
678 /*
679  * Statistics clock.  Updates rusage information and calls the scheduler
680  * to adjust priorities of the active thread.
681  *
682  * This should be called by all active processors.
683  */
684 void
685 statclock(int cnt, int usermode)
686 {
687 	struct rusage *ru;
688 	struct vmspace *vm;
689 	struct thread *td;
690 	struct proc *p;
691 	long rss;
692 	long *cp_time;
693 	uint64_t runtime, new_switchtime;
694 
695 	td = curthread;
696 	p = td->td_proc;
697 
698 	cp_time = (long *)PCPU_PTR(cp_time);
699 	if (usermode) {
700 		/*
701 		 * Charge the time as appropriate.
702 		 */
703 		td->td_uticks += cnt;
704 		if (p->p_nice > NZERO)
705 			cp_time[CP_NICE] += cnt;
706 		else
707 			cp_time[CP_USER] += cnt;
708 	} else {
709 		/*
710 		 * Came from kernel mode, so we were:
711 		 * - handling an interrupt,
712 		 * - doing syscall or trap work on behalf of the current
713 		 *   user process, or
714 		 * - spinning in the idle loop.
715 		 * Whichever it is, charge the time as appropriate.
716 		 * Note that we charge interrupts to the current process,
717 		 * regardless of whether they are ``for'' that process,
718 		 * so that we know how much of its real time was spent
719 		 * in ``non-process'' (i.e., interrupt) work.
720 		 */
721 		if ((td->td_pflags & TDP_ITHREAD) ||
722 		    td->td_intr_nesting_level >= 2) {
723 			td->td_iticks += cnt;
724 			cp_time[CP_INTR] += cnt;
725 		} else {
726 			td->td_pticks += cnt;
727 			td->td_sticks += cnt;
728 			if (!TD_IS_IDLETHREAD(td))
729 				cp_time[CP_SYS] += cnt;
730 			else
731 				cp_time[CP_IDLE] += cnt;
732 		}
733 	}
734 
735 	/* Update resource usage integrals and maximums. */
736 	MPASS(p->p_vmspace != NULL);
737 	vm = p->p_vmspace;
738 	ru = &td->td_ru;
739 	ru->ru_ixrss += pgtok(vm->vm_tsize) * cnt;
740 	ru->ru_idrss += pgtok(vm->vm_dsize) * cnt;
741 	ru->ru_isrss += pgtok(vm->vm_ssize) * cnt;
742 	rss = pgtok(vmspace_resident_count(vm));
743 	if (ru->ru_maxrss < rss)
744 		ru->ru_maxrss = rss;
745 	KTR_POINT2(KTR_SCHED, "thread", sched_tdname(td), "statclock",
746 	    "prio:%d", td->td_priority, "stathz:%d", (stathz)?stathz:hz);
747 	SDT_PROBE2(sched, , , tick, td, td->td_proc);
748 	thread_lock_flags(td, MTX_QUIET);
749 
750 	/*
751 	 * Compute the amount of time during which the current
752 	 * thread was running, and add that to its total so far.
753 	 */
754 	new_switchtime = cpu_ticks();
755 	runtime = new_switchtime - PCPU_GET(switchtime);
756 	td->td_runtime += runtime;
757 	td->td_incruntime += runtime;
758 	PCPU_SET(switchtime, new_switchtime);
759 
760 	sched_clock(td, cnt);
761 	thread_unlock(td);
762 #ifdef HWPMC_HOOKS
763 	if (td->td_intr_frame != NULL)
764 		PMC_SOFT_CALL_TF( , , clock, stat, td->td_intr_frame);
765 #endif
766 }
767 
768 void
769 profclock(int cnt, int usermode, uintfptr_t pc)
770 {
771 	struct thread *td;
772 
773 	td = curthread;
774 	if (usermode) {
775 		/*
776 		 * Came from user mode; CPU was in user state.
777 		 * If this process is being profiled, record the tick.
778 		 * if there is no related user location yet, don't
779 		 * bother trying to count it.
780 		 */
781 		if (td->td_proc->p_flag & P_PROFIL)
782 			addupc_intr(td, pc, cnt);
783 	}
784 #ifdef HWPMC_HOOKS
785 	if (td->td_intr_frame != NULL)
786 		PMC_SOFT_CALL_TF( , , clock, prof, td->td_intr_frame);
787 #endif
788 }
789 
790 /*
791  * Return information about system clocks.
792  */
793 static int
794 sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS)
795 {
796 	struct clockinfo clkinfo;
797 	/*
798 	 * Construct clockinfo structure.
799 	 */
800 	bzero(&clkinfo, sizeof(clkinfo));
801 	clkinfo.hz = hz;
802 	clkinfo.tick = tick;
803 	clkinfo.profhz = profhz;
804 	clkinfo.stathz = stathz ? stathz : hz;
805 	return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req));
806 }
807 
808 SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate,
809 	CTLTYPE_STRUCT|CTLFLAG_RD|CTLFLAG_MPSAFE,
810 	0, 0, sysctl_kern_clockrate, "S,clockinfo",
811 	"Rate and period of various kernel clocks");
812 
813 static void
814 watchdog_config(void *unused __unused, u_int cmd, int *error)
815 {
816 	u_int u;
817 
818 	u = cmd & WD_INTERVAL;
819 	if (u >= WD_TO_1SEC) {
820 		watchdog_ticks = (1 << (u - WD_TO_1SEC)) * hz;
821 		watchdog_enabled = 1;
822 		*error = 0;
823 	} else {
824 		watchdog_enabled = 0;
825 	}
826 }
827 
828 /*
829  * Handle a watchdog timeout by dropping to DDB or panicking.
830  */
831 static void
832 watchdog_fire(void)
833 {
834 
835 #if defined(KDB) && !defined(KDB_UNATTENDED)
836 	kdb_backtrace();
837 	kdb_enter(KDB_WHY_WATCHDOG, "watchdog timeout");
838 #else
839 	panic("watchdog timeout");
840 #endif
841 }
842