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: stable/12/sys/kern/kern_clock.c 359794 2020-04-11 05:12:38Z jah $");
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/gtaskqueue.h>
53 #include <sys/kdb.h>
54 #include <sys/kernel.h>
55 #include <sys/kthread.h>
56 #include <sys/ktr.h>
57 #include <sys/lock.h>
58 #include <sys/mutex.h>
59 #include <sys/proc.h>
60 #include <sys/resource.h>
61 #include <sys/resourcevar.h>
62 #include <sys/sched.h>
63 #include <sys/sdt.h>
64 #include <sys/signalvar.h>
65 #include <sys/sleepqueue.h>
66 #include <sys/smp.h>
67 #include <vm/vm.h>
68 #include <vm/pmap.h>
69 #include <vm/vm_map.h>
70 #include <sys/sysctl.h>
71 #include <sys/bus.h>
72 #include <sys/interrupt.h>
73 #include <sys/limits.h>
74 #include <sys/timetc.h>
75
76 #ifdef GPROF
77 #include <sys/gmon.h>
78 #endif
79
80 #ifdef HWPMC_HOOKS
81 #include <sys/pmckern.h>
82 PMC_SOFT_DEFINE( , , clock, hard);
83 PMC_SOFT_DEFINE( , , clock, stat);
84 PMC_SOFT_DEFINE_EX( , , clock, prof, \
85 cpu_startprofclock, cpu_stopprofclock);
86 #endif
87
88 #ifdef DEVICE_POLLING
89 extern void hardclock_device_poll(void);
90 #endif /* DEVICE_POLLING */
91
92 static void initclocks(void *dummy);
93 SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL);
94
95 /* Spin-lock protecting profiling statistics. */
96 static struct mtx time_lock;
97
98 SDT_PROVIDER_DECLARE(sched);
99 SDT_PROBE_DEFINE2(sched, , , tick, "struct thread *", "struct proc *");
100
101 static int
sysctl_kern_cp_time(SYSCTL_HANDLER_ARGS)102 sysctl_kern_cp_time(SYSCTL_HANDLER_ARGS)
103 {
104 int error;
105 long cp_time[CPUSTATES];
106 #ifdef SCTL_MASK32
107 int i;
108 unsigned int cp_time32[CPUSTATES];
109 #endif
110
111 read_cpu_time(cp_time);
112 #ifdef SCTL_MASK32
113 if (req->flags & SCTL_MASK32) {
114 if (!req->oldptr)
115 return SYSCTL_OUT(req, 0, sizeof(cp_time32));
116 for (i = 0; i < CPUSTATES; i++)
117 cp_time32[i] = (unsigned int)cp_time[i];
118 error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32));
119 } else
120 #endif
121 {
122 if (!req->oldptr)
123 return SYSCTL_OUT(req, 0, sizeof(cp_time));
124 error = SYSCTL_OUT(req, cp_time, sizeof(cp_time));
125 }
126 return error;
127 }
128
129 SYSCTL_PROC(_kern, OID_AUTO, cp_time, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE,
130 0,0, sysctl_kern_cp_time, "LU", "CPU time statistics");
131
132 static long empty[CPUSTATES];
133
134 static int
sysctl_kern_cp_times(SYSCTL_HANDLER_ARGS)135 sysctl_kern_cp_times(SYSCTL_HANDLER_ARGS)
136 {
137 struct pcpu *pcpu;
138 int error;
139 int c;
140 long *cp_time;
141 #ifdef SCTL_MASK32
142 unsigned int cp_time32[CPUSTATES];
143 int i;
144 #endif
145
146 if (!req->oldptr) {
147 #ifdef SCTL_MASK32
148 if (req->flags & SCTL_MASK32)
149 return SYSCTL_OUT(req, 0, sizeof(cp_time32) * (mp_maxid + 1));
150 else
151 #endif
152 return SYSCTL_OUT(req, 0, sizeof(long) * CPUSTATES * (mp_maxid + 1));
153 }
154 for (error = 0, c = 0; error == 0 && c <= mp_maxid; c++) {
155 if (!CPU_ABSENT(c)) {
156 pcpu = pcpu_find(c);
157 cp_time = pcpu->pc_cp_time;
158 } else {
159 cp_time = empty;
160 }
161 #ifdef SCTL_MASK32
162 if (req->flags & SCTL_MASK32) {
163 for (i = 0; i < CPUSTATES; i++)
164 cp_time32[i] = (unsigned int)cp_time[i];
165 error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32));
166 } else
167 #endif
168 error = SYSCTL_OUT(req, cp_time, sizeof(long) * CPUSTATES);
169 }
170 return error;
171 }
172
173 SYSCTL_PROC(_kern, OID_AUTO, cp_times, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE,
174 0,0, sysctl_kern_cp_times, "LU", "per-CPU time statistics");
175
176 #ifdef DEADLKRES
177 static const char *blessed[] = {
178 "getblk",
179 "so_snd_sx",
180 "so_rcv_sx",
181 NULL
182 };
183 static int slptime_threshold = 1800;
184 static int blktime_threshold = 900;
185 static int sleepfreq = 3;
186
187 static void
deadlres_td_on_lock(struct proc * p,struct thread * td,int blkticks)188 deadlres_td_on_lock(struct proc *p, struct thread *td, int blkticks)
189 {
190 int tticks;
191
192 sx_assert(&allproc_lock, SX_LOCKED);
193 PROC_LOCK_ASSERT(p, MA_OWNED);
194 THREAD_LOCK_ASSERT(td, MA_OWNED);
195 /*
196 * The thread should be blocked on a turnstile, simply check
197 * if the turnstile channel is in good state.
198 */
199 MPASS(td->td_blocked != NULL);
200
201 tticks = ticks - td->td_blktick;
202 if (tticks > blkticks)
203 /*
204 * Accordingly with provided thresholds, this thread is stuck
205 * for too long on a turnstile.
206 */
207 panic("%s: possible deadlock detected for %p (%s), "
208 "blocked for %d ticks\n", __func__,
209 td, sched_tdname(td), tticks);
210 }
211
212 static void
deadlres_td_sleep_q(struct proc * p,struct thread * td,int slpticks)213 deadlres_td_sleep_q(struct proc *p, struct thread *td, int slpticks)
214 {
215 void *wchan;
216 int i, slptype, tticks;
217
218 sx_assert(&allproc_lock, SX_LOCKED);
219 PROC_LOCK_ASSERT(p, MA_OWNED);
220 THREAD_LOCK_ASSERT(td, MA_OWNED);
221 /*
222 * Check if the thread is sleeping on a lock, otherwise skip the check.
223 * Drop the thread lock in order to avoid a LOR with the sleepqueue
224 * spinlock.
225 */
226 wchan = td->td_wchan;
227 tticks = ticks - td->td_slptick;
228 slptype = sleepq_type(wchan);
229 if ((slptype == SLEEPQ_SX || slptype == SLEEPQ_LK) &&
230 tticks > slpticks) {
231
232 /*
233 * Accordingly with provided thresholds, this thread is stuck
234 * for too long on a sleepqueue.
235 * However, being on a sleepqueue, we might still check for the
236 * blessed list.
237 */
238 for (i = 0; blessed[i] != NULL; i++)
239 if (!strcmp(blessed[i], td->td_wmesg))
240 return;
241
242 panic("%s: possible deadlock detected for %p (%s), "
243 "blocked for %d ticks\n", __func__,
244 td, sched_tdname(td), tticks);
245 }
246 }
247
248 static void
deadlkres(void)249 deadlkres(void)
250 {
251 struct proc *p;
252 struct thread *td;
253 int blkticks, slpticks, tryl;
254
255 tryl = 0;
256 for (;;) {
257 blkticks = blktime_threshold * hz;
258 slpticks = slptime_threshold * hz;
259
260 /*
261 * Avoid to sleep on the sx_lock in order to avoid a
262 * possible priority inversion problem leading to
263 * starvation.
264 * If the lock can't be held after 100 tries, panic.
265 */
266 if (!sx_try_slock(&allproc_lock)) {
267 if (tryl > 100)
268 panic("%s: possible deadlock detected "
269 "on allproc_lock\n", __func__);
270 tryl++;
271 pause("allproc", sleepfreq * hz);
272 continue;
273 }
274 tryl = 0;
275 FOREACH_PROC_IN_SYSTEM(p) {
276 PROC_LOCK(p);
277 if (p->p_state == PRS_NEW) {
278 PROC_UNLOCK(p);
279 continue;
280 }
281 FOREACH_THREAD_IN_PROC(p, td) {
282 thread_lock(td);
283 if (TD_ON_LOCK(td))
284 deadlres_td_on_lock(p, td,
285 blkticks);
286 else if (TD_IS_SLEEPING(td) &&
287 TD_ON_SLEEPQ(td))
288 deadlres_td_sleep_q(p, td,
289 slpticks);
290 thread_unlock(td);
291 }
292 PROC_UNLOCK(p);
293 }
294 sx_sunlock(&allproc_lock);
295
296 /* Sleep for sleepfreq seconds. */
297 pause("-", sleepfreq * hz);
298 }
299 }
300
301 static struct kthread_desc deadlkres_kd = {
302 "deadlkres",
303 deadlkres,
304 (struct thread **)NULL
305 };
306
307 SYSINIT(deadlkres, SI_SUB_CLOCKS, SI_ORDER_ANY, kthread_start, &deadlkres_kd);
308
309 static SYSCTL_NODE(_debug, OID_AUTO, deadlkres, CTLFLAG_RW, 0,
310 "Deadlock resolver");
311 SYSCTL_INT(_debug_deadlkres, OID_AUTO, slptime_threshold, CTLFLAG_RW,
312 &slptime_threshold, 0,
313 "Number of seconds within is valid to sleep on a sleepqueue");
314 SYSCTL_INT(_debug_deadlkres, OID_AUTO, blktime_threshold, CTLFLAG_RW,
315 &blktime_threshold, 0,
316 "Number of seconds within is valid to block on a turnstile");
317 SYSCTL_INT(_debug_deadlkres, OID_AUTO, sleepfreq, CTLFLAG_RW, &sleepfreq, 0,
318 "Number of seconds between any deadlock resolver thread run");
319 #endif /* DEADLKRES */
320
321 void
read_cpu_time(long * cp_time)322 read_cpu_time(long *cp_time)
323 {
324 struct pcpu *pc;
325 int i, j;
326
327 /* Sum up global cp_time[]. */
328 bzero(cp_time, sizeof(long) * CPUSTATES);
329 CPU_FOREACH(i) {
330 pc = pcpu_find(i);
331 for (j = 0; j < CPUSTATES; j++)
332 cp_time[j] += pc->pc_cp_time[j];
333 }
334 }
335
336 #include <sys/watchdog.h>
337
338 static int watchdog_ticks;
339 static int watchdog_enabled;
340 static void watchdog_fire(void);
341 static void watchdog_config(void *, u_int, int *);
342
343 static void
watchdog_attach(void)344 watchdog_attach(void)
345 {
346 EVENTHANDLER_REGISTER(watchdog_list, watchdog_config, NULL, 0);
347 }
348
349 /*
350 * Clock handling routines.
351 *
352 * This code is written to operate with two timers that run independently of
353 * each other.
354 *
355 * The main timer, running hz times per second, is used to trigger interval
356 * timers, timeouts and rescheduling as needed.
357 *
358 * The second timer handles kernel and user profiling,
359 * and does resource use estimation. If the second timer is programmable,
360 * it is randomized to avoid aliasing between the two clocks. For example,
361 * the randomization prevents an adversary from always giving up the cpu
362 * just before its quantum expires. Otherwise, it would never accumulate
363 * cpu ticks. The mean frequency of the second timer is stathz.
364 *
365 * If no second timer exists, stathz will be zero; in this case we drive
366 * profiling and statistics off the main clock. This WILL NOT be accurate;
367 * do not do it unless absolutely necessary.
368 *
369 * The statistics clock may (or may not) be run at a higher rate while
370 * profiling. This profile clock runs at profhz. We require that profhz
371 * be an integral multiple of stathz.
372 *
373 * If the statistics clock is running fast, it must be divided by the ratio
374 * profhz/stathz for statistics. (For profiling, every tick counts.)
375 *
376 * Time-of-day is maintained using a "timecounter", which may or may
377 * not be related to the hardware generating the above mentioned
378 * interrupts.
379 */
380
381 int stathz;
382 int profhz;
383 int profprocs;
384 volatile int ticks;
385 int psratio;
386
387 DPCPU_DEFINE_STATIC(int, pcputicks); /* Per-CPU version of ticks. */
388 #ifdef DEVICE_POLLING
389 static int devpoll_run = 0;
390 #endif
391
392 /*
393 * Initialize clock frequencies and start both clocks running.
394 */
395 /* ARGSUSED*/
396 static void
initclocks(void * dummy)397 initclocks(void *dummy)
398 {
399 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
416 #ifdef SW_WATCHDOG
417 /* Enable hardclock watchdog now, even if a hardware watchdog exists. */
418 watchdog_attach();
419 #else
420 /* Volunteer to run a software watchdog. */
421 if (wdog_software_attach == NULL)
422 wdog_software_attach = watchdog_attach;
423 #endif
424 }
425
426 void
hardclock(int cnt,int usermode)427 hardclock(int cnt, int usermode)
428 {
429 struct pstats *pstats;
430 struct thread *td = curthread;
431 struct proc *p = td->td_proc;
432 int *t = DPCPU_PTR(pcputicks);
433 int flags, global, newticks;
434 int i;
435
436 /*
437 * Update per-CPU and possibly global ticks values.
438 */
439 *t += cnt;
440 do {
441 global = ticks;
442 newticks = *t - global;
443 if (newticks <= 0) {
444 if (newticks < -1)
445 *t = global - 1;
446 newticks = 0;
447 break;
448 }
449 } while (!atomic_cmpset_int(&ticks, global, *t));
450
451 /*
452 * Run current process's virtual and profile time, as needed.
453 */
454 pstats = p->p_stats;
455 flags = 0;
456 if (usermode &&
457 timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) {
458 PROC_ITIMLOCK(p);
459 if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL],
460 tick * cnt) == 0)
461 flags |= TDF_ALRMPEND | TDF_ASTPENDING;
462 PROC_ITIMUNLOCK(p);
463 }
464 if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) {
465 PROC_ITIMLOCK(p);
466 if (itimerdecr(&pstats->p_timer[ITIMER_PROF],
467 tick * cnt) == 0)
468 flags |= TDF_PROFPEND | TDF_ASTPENDING;
469 PROC_ITIMUNLOCK(p);
470 }
471 if (flags != 0) {
472 thread_lock(td);
473 td->td_flags |= flags;
474 thread_unlock(td);
475 }
476
477 #ifdef HWPMC_HOOKS
478 if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid)))
479 PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL);
480 if (td->td_intr_frame != NULL)
481 PMC_SOFT_CALL_TF( , , clock, hard, td->td_intr_frame);
482 #endif
483 /* We are in charge to handle this tick duty. */
484 if (newticks > 0) {
485 tc_ticktock(newticks);
486 #ifdef DEVICE_POLLING
487 /* Dangerous and no need to call these things concurrently. */
488 if (atomic_cmpset_acq_int(&devpoll_run, 0, 1)) {
489 /* This is very short and quick. */
490 hardclock_device_poll();
491 atomic_store_rel_int(&devpoll_run, 0);
492 }
493 #endif /* DEVICE_POLLING */
494 if (watchdog_enabled > 0) {
495 i = atomic_fetchadd_int(&watchdog_ticks, -newticks);
496 if (i > 0 && i <= newticks)
497 watchdog_fire();
498 }
499 }
500 if (curcpu == CPU_FIRST())
501 cpu_tick_calibration();
502 if (__predict_false(DPCPU_GET(epoch_cb_count)))
503 GROUPTASK_ENQUEUE(DPCPU_PTR(epoch_cb_task));
504 }
505
506 void
hardclock_sync(int cpu)507 hardclock_sync(int cpu)
508 {
509 int *t;
510 KASSERT(!CPU_ABSENT(cpu), ("Absent CPU %d", cpu));
511 t = DPCPU_ID_PTR(cpu, pcputicks);
512
513 *t = ticks;
514 }
515
516 /*
517 * Compute number of ticks in the specified amount of time.
518 */
519 int
tvtohz(struct timeval * tv)520 tvtohz(struct timeval *tv)
521 {
522 unsigned long ticks;
523 long sec, usec;
524
525 /*
526 * If the number of usecs in the whole seconds part of the time
527 * difference fits in a long, then the total number of usecs will
528 * fit in an unsigned long. Compute the total and convert it to
529 * ticks, rounding up and adding 1 to allow for the current tick
530 * to expire. Rounding also depends on unsigned long arithmetic
531 * to avoid overflow.
532 *
533 * Otherwise, if the number of ticks in the whole seconds part of
534 * the time difference fits in a long, then convert the parts to
535 * ticks separately and add, using similar rounding methods and
536 * overflow avoidance. This method would work in the previous
537 * case but it is slightly slower and assumes that hz is integral.
538 *
539 * Otherwise, round the time difference down to the maximum
540 * representable value.
541 *
542 * If ints have 32 bits, then the maximum value for any timeout in
543 * 10ms ticks is 248 days.
544 */
545 sec = tv->tv_sec;
546 usec = tv->tv_usec;
547 if (usec < 0) {
548 sec--;
549 usec += 1000000;
550 }
551 if (sec < 0) {
552 #ifdef DIAGNOSTIC
553 if (usec > 0) {
554 sec++;
555 usec -= 1000000;
556 }
557 printf("tvotohz: negative time difference %ld sec %ld usec\n",
558 sec, usec);
559 #endif
560 ticks = 1;
561 } else if (sec <= LONG_MAX / 1000000)
562 ticks = howmany(sec * 1000000 + (unsigned long)usec, tick) + 1;
563 else if (sec <= LONG_MAX / hz)
564 ticks = sec * hz
565 + howmany((unsigned long)usec, tick) + 1;
566 else
567 ticks = LONG_MAX;
568 if (ticks > INT_MAX)
569 ticks = INT_MAX;
570 return ((int)ticks);
571 }
572
573 /*
574 * Start profiling on a process.
575 *
576 * Kernel profiling passes proc0 which never exits and hence
577 * keeps the profile clock running constantly.
578 */
579 void
startprofclock(struct proc * p)580 startprofclock(struct proc *p)
581 {
582
583 PROC_LOCK_ASSERT(p, MA_OWNED);
584 if (p->p_flag & P_STOPPROF)
585 return;
586 if ((p->p_flag & P_PROFIL) == 0) {
587 p->p_flag |= P_PROFIL;
588 mtx_lock(&time_lock);
589 if (++profprocs == 1)
590 cpu_startprofclock();
591 mtx_unlock(&time_lock);
592 }
593 }
594
595 /*
596 * Stop profiling on a process.
597 */
598 void
stopprofclock(struct proc * p)599 stopprofclock(struct proc *p)
600 {
601
602 PROC_LOCK_ASSERT(p, MA_OWNED);
603 if (p->p_flag & P_PROFIL) {
604 if (p->p_profthreads != 0) {
605 while (p->p_profthreads != 0) {
606 p->p_flag |= P_STOPPROF;
607 msleep(&p->p_profthreads, &p->p_mtx, PPAUSE,
608 "stopprof", 0);
609 }
610 }
611 if ((p->p_flag & P_PROFIL) == 0)
612 return;
613 p->p_flag &= ~P_PROFIL;
614 mtx_lock(&time_lock);
615 if (--profprocs == 0)
616 cpu_stopprofclock();
617 mtx_unlock(&time_lock);
618 }
619 }
620
621 /*
622 * Statistics clock. Updates rusage information and calls the scheduler
623 * to adjust priorities of the active thread.
624 *
625 * This should be called by all active processors.
626 */
627 void
statclock(int cnt,int usermode)628 statclock(int cnt, int usermode)
629 {
630 struct rusage *ru;
631 struct vmspace *vm;
632 struct thread *td;
633 struct proc *p;
634 long rss;
635 long *cp_time;
636 uint64_t runtime, new_switchtime;
637
638 td = curthread;
639 p = td->td_proc;
640
641 cp_time = (long *)PCPU_PTR(cp_time);
642 if (usermode) {
643 /*
644 * Charge the time as appropriate.
645 */
646 td->td_uticks += cnt;
647 if (p->p_nice > NZERO)
648 cp_time[CP_NICE] += cnt;
649 else
650 cp_time[CP_USER] += cnt;
651 } else {
652 /*
653 * Came from kernel mode, so we were:
654 * - handling an interrupt,
655 * - doing syscall or trap work on behalf of the current
656 * user process, or
657 * - spinning in the idle loop.
658 * Whichever it is, charge the time as appropriate.
659 * Note that we charge interrupts to the current process,
660 * regardless of whether they are ``for'' that process,
661 * so that we know how much of its real time was spent
662 * in ``non-process'' (i.e., interrupt) work.
663 */
664 if ((td->td_pflags & TDP_ITHREAD) ||
665 td->td_intr_nesting_level >= 2) {
666 td->td_iticks += cnt;
667 cp_time[CP_INTR] += cnt;
668 } else {
669 td->td_pticks += cnt;
670 td->td_sticks += cnt;
671 if (!TD_IS_IDLETHREAD(td))
672 cp_time[CP_SYS] += cnt;
673 else
674 cp_time[CP_IDLE] += cnt;
675 }
676 }
677
678 /* Update resource usage integrals and maximums. */
679 MPASS(p->p_vmspace != NULL);
680 vm = p->p_vmspace;
681 ru = &td->td_ru;
682 ru->ru_ixrss += pgtok(vm->vm_tsize) * cnt;
683 ru->ru_idrss += pgtok(vm->vm_dsize) * cnt;
684 ru->ru_isrss += pgtok(vm->vm_ssize) * cnt;
685 rss = pgtok(vmspace_resident_count(vm));
686 if (ru->ru_maxrss < rss)
687 ru->ru_maxrss = rss;
688 KTR_POINT2(KTR_SCHED, "thread", sched_tdname(td), "statclock",
689 "prio:%d", td->td_priority, "stathz:%d", (stathz)?stathz:hz);
690 SDT_PROBE2(sched, , , tick, td, td->td_proc);
691 thread_lock_flags(td, MTX_QUIET);
692
693 /*
694 * Compute the amount of time during which the current
695 * thread was running, and add that to its total so far.
696 */
697 new_switchtime = cpu_ticks();
698 runtime = new_switchtime - PCPU_GET(switchtime);
699 td->td_runtime += runtime;
700 td->td_incruntime += runtime;
701 PCPU_SET(switchtime, new_switchtime);
702
703 for ( ; cnt > 0; cnt--)
704 sched_clock(td);
705 thread_unlock(td);
706 #ifdef HWPMC_HOOKS
707 if (td->td_intr_frame != NULL)
708 PMC_SOFT_CALL_TF( , , clock, stat, td->td_intr_frame);
709 #endif
710 }
711
712 void
profclock(int cnt,int usermode,uintfptr_t pc)713 profclock(int cnt, int usermode, uintfptr_t pc)
714 {
715 struct thread *td;
716 #ifdef GPROF
717 struct gmonparam *g;
718 uintfptr_t i;
719 #endif
720
721 td = curthread;
722 if (usermode) {
723 /*
724 * Came from user mode; CPU was in user state.
725 * If this process is being profiled, record the tick.
726 * if there is no related user location yet, don't
727 * bother trying to count it.
728 */
729 if (td->td_proc->p_flag & P_PROFIL)
730 addupc_intr(td, pc, cnt);
731 }
732 #ifdef GPROF
733 else {
734 /*
735 * Kernel statistics are just like addupc_intr, only easier.
736 */
737 g = &_gmonparam;
738 if (g->state == GMON_PROF_ON && pc >= g->lowpc) {
739 i = PC_TO_I(g, pc);
740 if (i < g->textsize) {
741 KCOUNT(g, i) += cnt;
742 }
743 }
744 }
745 #endif
746 #ifdef HWPMC_HOOKS
747 if (td->td_intr_frame != NULL)
748 PMC_SOFT_CALL_TF( , , clock, prof, td->td_intr_frame);
749 #endif
750 }
751
752 /*
753 * Return information about system clocks.
754 */
755 static int
sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS)756 sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS)
757 {
758 struct clockinfo clkinfo;
759 /*
760 * Construct clockinfo structure.
761 */
762 bzero(&clkinfo, sizeof(clkinfo));
763 clkinfo.hz = hz;
764 clkinfo.tick = tick;
765 clkinfo.profhz = profhz;
766 clkinfo.stathz = stathz ? stathz : hz;
767 return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req));
768 }
769
770 SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate,
771 CTLTYPE_STRUCT|CTLFLAG_RD|CTLFLAG_MPSAFE,
772 0, 0, sysctl_kern_clockrate, "S,clockinfo",
773 "Rate and period of various kernel clocks");
774
775 static void
watchdog_config(void * unused __unused,u_int cmd,int * error)776 watchdog_config(void *unused __unused, u_int cmd, int *error)
777 {
778 u_int u;
779
780 u = cmd & WD_INTERVAL;
781 if (u >= WD_TO_1SEC) {
782 watchdog_ticks = (1 << (u - WD_TO_1SEC)) * hz;
783 watchdog_enabled = 1;
784 *error = 0;
785 } else {
786 watchdog_enabled = 0;
787 }
788 }
789
790 /*
791 * Handle a watchdog timeout by dumping interrupt information and
792 * then either dropping to DDB or panicking.
793 */
794 static void
watchdog_fire(void)795 watchdog_fire(void)
796 {
797 int nintr;
798 uint64_t inttotal;
799 u_long *curintr;
800 char *curname;
801
802 curintr = intrcnt;
803 curname = intrnames;
804 inttotal = 0;
805 nintr = sintrcnt / sizeof(u_long);
806
807 printf("interrupt total\n");
808 while (--nintr >= 0) {
809 if (*curintr)
810 printf("%-12s %20lu\n", curname, *curintr);
811 curname += strlen(curname) + 1;
812 inttotal += *curintr++;
813 }
814 printf("Total %20ju\n", (uintmax_t)inttotal);
815
816 #if defined(KDB) && !defined(KDB_UNATTENDED)
817 kdb_backtrace();
818 kdb_enter(KDB_WHY_WATCHDOG, "watchdog timeout");
819 #else
820 panic("watchdog timeout");
821 #endif
822 }
823