1 /*-
2 * Copyright (c) 1982, 1986, 1991, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * @(#)kern_clock.c 8.5 (Berkeley) 1/21/94
35 */
36
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39
40 #include "opt_kdb.h"
41 #include "opt_device_polling.h"
42 #include "opt_hwpmc_hooks.h"
43 #include "opt_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/kdb.h>
50 #include <sys/kernel.h>
51 #include <sys/kthread.h>
52 #include <sys/ktr.h>
53 #include <sys/lock.h>
54 #include <sys/mutex.h>
55 #include <sys/proc.h>
56 #include <sys/resource.h>
57 #include <sys/resourcevar.h>
58 #include <sys/sched.h>
59 #include <sys/sdt.h>
60 #include <sys/signalvar.h>
61 #include <sys/sleepqueue.h>
62 #include <sys/smp.h>
63 #include <vm/vm.h>
64 #include <vm/pmap.h>
65 #include <vm/vm_map.h>
66 #include <sys/sysctl.h>
67 #include <sys/bus.h>
68 #include <sys/interrupt.h>
69 #include <sys/limits.h>
70 #include <sys/timetc.h>
71
72 #ifdef GPROF
73 #include <sys/gmon.h>
74 #endif
75
76 #ifdef HWPMC_HOOKS
77 #include <sys/pmckern.h>
78 PMC_SOFT_DEFINE( , , clock, hard);
79 PMC_SOFT_DEFINE( , , clock, stat);
80 PMC_SOFT_DEFINE_EX( , , clock, prof, \
81 cpu_startprofclock, cpu_stopprofclock);
82 #endif
83
84 #ifdef DEVICE_POLLING
85 extern void hardclock_device_poll(void);
86 #endif /* DEVICE_POLLING */
87
88 static void initclocks(void *dummy);
89 SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL);
90
91 /* Spin-lock protecting profiling statistics. */
92 static struct mtx time_lock;
93
94 SDT_PROVIDER_DECLARE(sched);
95 SDT_PROBE_DEFINE2(sched, , , tick, "struct thread *", "struct proc *");
96
97 static int
sysctl_kern_cp_time(SYSCTL_HANDLER_ARGS)98 sysctl_kern_cp_time(SYSCTL_HANDLER_ARGS)
99 {
100 int error;
101 long cp_time[CPUSTATES];
102 #ifdef SCTL_MASK32
103 int i;
104 unsigned int cp_time32[CPUSTATES];
105 #endif
106
107 read_cpu_time(cp_time);
108 #ifdef SCTL_MASK32
109 if (req->flags & SCTL_MASK32) {
110 if (!req->oldptr)
111 return SYSCTL_OUT(req, 0, sizeof(cp_time32));
112 for (i = 0; i < CPUSTATES; i++)
113 cp_time32[i] = (unsigned int)cp_time[i];
114 error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32));
115 } else
116 #endif
117 {
118 if (!req->oldptr)
119 return SYSCTL_OUT(req, 0, sizeof(cp_time));
120 error = SYSCTL_OUT(req, cp_time, sizeof(cp_time));
121 }
122 return error;
123 }
124
125 SYSCTL_PROC(_kern, OID_AUTO, cp_time, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE,
126 0,0, sysctl_kern_cp_time, "LU", "CPU time statistics");
127
128 static long empty[CPUSTATES];
129
130 static int
sysctl_kern_cp_times(SYSCTL_HANDLER_ARGS)131 sysctl_kern_cp_times(SYSCTL_HANDLER_ARGS)
132 {
133 struct pcpu *pcpu;
134 int error;
135 int c;
136 long *cp_time;
137 #ifdef SCTL_MASK32
138 unsigned int cp_time32[CPUSTATES];
139 int i;
140 #endif
141
142 if (!req->oldptr) {
143 #ifdef SCTL_MASK32
144 if (req->flags & SCTL_MASK32)
145 return SYSCTL_OUT(req, 0, sizeof(cp_time32) * (mp_maxid + 1));
146 else
147 #endif
148 return SYSCTL_OUT(req, 0, sizeof(long) * CPUSTATES * (mp_maxid + 1));
149 }
150 for (error = 0, c = 0; error == 0 && c <= mp_maxid; c++) {
151 if (!CPU_ABSENT(c)) {
152 pcpu = pcpu_find(c);
153 cp_time = pcpu->pc_cp_time;
154 } else {
155 cp_time = empty;
156 }
157 #ifdef SCTL_MASK32
158 if (req->flags & SCTL_MASK32) {
159 for (i = 0; i < CPUSTATES; i++)
160 cp_time32[i] = (unsigned int)cp_time[i];
161 error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32));
162 } else
163 #endif
164 error = SYSCTL_OUT(req, cp_time, sizeof(long) * CPUSTATES);
165 }
166 return error;
167 }
168
169 SYSCTL_PROC(_kern, OID_AUTO, cp_times, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE,
170 0,0, sysctl_kern_cp_times, "LU", "per-CPU time statistics");
171
172 #ifdef DEADLKRES
173 static const char *blessed[] = {
174 "getblk",
175 "so_snd_sx",
176 "so_rcv_sx",
177 NULL
178 };
179 static int slptime_threshold = 1800;
180 static int blktime_threshold = 900;
181 static int sleepfreq = 3;
182
183 static void
deadlkres(void)184 deadlkres(void)
185 {
186 struct proc *p;
187 struct thread *td;
188 void *wchan;
189 int blkticks, i, slpticks, slptype, tryl, tticks;
190
191 tryl = 0;
192 for (;;) {
193 blkticks = blktime_threshold * hz;
194 slpticks = slptime_threshold * hz;
195
196 /*
197 * Avoid to sleep on the sx_lock in order to avoid a possible
198 * priority inversion problem leading to starvation.
199 * If the lock can't be held after 100 tries, panic.
200 */
201 if (!sx_try_slock(&allproc_lock)) {
202 if (tryl > 100)
203 panic("%s: possible deadlock detected on allproc_lock\n",
204 __func__);
205 tryl++;
206 pause("allproc", sleepfreq * hz);
207 continue;
208 }
209 tryl = 0;
210 FOREACH_PROC_IN_SYSTEM(p) {
211 PROC_LOCK(p);
212 if (p->p_state == PRS_NEW) {
213 PROC_UNLOCK(p);
214 continue;
215 }
216 FOREACH_THREAD_IN_PROC(p, td) {
217
218 thread_lock(td);
219 if (TD_ON_LOCK(td)) {
220
221 /*
222 * The thread should be blocked on a
223 * turnstile, simply check if the
224 * turnstile channel is in good state.
225 */
226 MPASS(td->td_blocked != NULL);
227
228 tticks = ticks - td->td_blktick;
229 thread_unlock(td);
230 if (tticks > blkticks) {
231
232 /*
233 * Accordingly with provided
234 * thresholds, this thread is
235 * stuck for too long on a
236 * turnstile.
237 */
238 PROC_UNLOCK(p);
239 sx_sunlock(&allproc_lock);
240 panic("%s: possible deadlock detected for %p, blocked for %d ticks\n",
241 __func__, td, tticks);
242 }
243 } else if (TD_IS_SLEEPING(td) &&
244 TD_ON_SLEEPQ(td)) {
245
246 /*
247 * Check if the thread is sleeping on a
248 * lock, otherwise skip the check.
249 * Drop the thread lock in order to
250 * avoid a LOR with the sleepqueue
251 * spinlock.
252 */
253 wchan = td->td_wchan;
254 tticks = ticks - td->td_slptick;
255 thread_unlock(td);
256 slptype = sleepq_type(wchan);
257 if ((slptype == SLEEPQ_SX ||
258 slptype == SLEEPQ_LK) &&
259 tticks > slpticks) {
260
261 /*
262 * Accordingly with provided
263 * thresholds, this thread is
264 * stuck for too long on a
265 * sleepqueue.
266 * However, being on a
267 * sleepqueue, we might still
268 * check for the blessed
269 * list.
270 */
271 tryl = 0;
272 for (i = 0; blessed[i] != NULL;
273 i++) {
274 if (!strcmp(blessed[i],
275 td->td_wmesg)) {
276 tryl = 1;
277 break;
278 }
279 }
280 if (tryl != 0) {
281 tryl = 0;
282 continue;
283 }
284 PROC_UNLOCK(p);
285 sx_sunlock(&allproc_lock);
286 panic("%s: possible deadlock detected for %p, blocked for %d ticks\n",
287 __func__, td, tticks);
288 }
289 } else
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 static DPCPU_DEFINE(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 /*
427 * Each time the real-time timer fires, this function is called on all CPUs.
428 * Note that hardclock() calls hardclock_cpu() for the boot CPU, so only
429 * the other CPUs in the system need to call this function.
430 */
431 void
hardclock_cpu(int usermode)432 hardclock_cpu(int usermode)
433 {
434 struct pstats *pstats;
435 struct thread *td = curthread;
436 struct proc *p = td->td_proc;
437 int flags;
438
439 /*
440 * Run current process's virtual and profile time, as needed.
441 */
442 pstats = p->p_stats;
443 flags = 0;
444 if (usermode &&
445 timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) {
446 PROC_ITIMLOCK(p);
447 if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0)
448 flags |= TDF_ALRMPEND | TDF_ASTPENDING;
449 PROC_ITIMUNLOCK(p);
450 }
451 if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) {
452 PROC_ITIMLOCK(p);
453 if (itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0)
454 flags |= TDF_PROFPEND | TDF_ASTPENDING;
455 PROC_ITIMUNLOCK(p);
456 }
457 thread_lock(td);
458 td->td_flags |= flags;
459 thread_unlock(td);
460
461 #ifdef HWPMC_HOOKS
462 if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid)))
463 PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL);
464 if (td->td_intr_frame != NULL)
465 PMC_SOFT_CALL_TF( , , clock, hard, td->td_intr_frame);
466 #endif
467 callout_process(sbinuptime());
468 }
469
470 /*
471 * The real-time timer, interrupting hz times per second.
472 */
473 void
hardclock(int usermode,uintfptr_t pc)474 hardclock(int usermode, uintfptr_t pc)
475 {
476
477 atomic_add_int(&ticks, 1);
478 hardclock_cpu(usermode);
479 tc_ticktock(1);
480 cpu_tick_calibration();
481 /*
482 * If no separate statistics clock is available, run it from here.
483 *
484 * XXX: this only works for UP
485 */
486 if (stathz == 0) {
487 profclock(usermode, pc);
488 statclock(usermode);
489 }
490 #ifdef DEVICE_POLLING
491 hardclock_device_poll(); /* this is very short and quick */
492 #endif /* DEVICE_POLLING */
493 if (watchdog_enabled > 0 && --watchdog_ticks <= 0)
494 watchdog_fire();
495 }
496
497 void
hardclock_cnt(int cnt,int usermode)498 hardclock_cnt(int cnt, int usermode)
499 {
500 struct pstats *pstats;
501 struct thread *td = curthread;
502 struct proc *p = td->td_proc;
503 int *t = DPCPU_PTR(pcputicks);
504 int flags, global, newticks;
505 int i;
506
507 /*
508 * Update per-CPU and possibly global ticks values.
509 */
510 *t += cnt;
511 do {
512 global = ticks;
513 newticks = *t - global;
514 if (newticks <= 0) {
515 if (newticks < -1)
516 *t = global - 1;
517 newticks = 0;
518 break;
519 }
520 } while (!atomic_cmpset_int(&ticks, global, *t));
521
522 /*
523 * Run current process's virtual and profile time, as needed.
524 */
525 pstats = p->p_stats;
526 flags = 0;
527 if (usermode &&
528 timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) {
529 PROC_ITIMLOCK(p);
530 if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL],
531 tick * cnt) == 0)
532 flags |= TDF_ALRMPEND | TDF_ASTPENDING;
533 PROC_ITIMUNLOCK(p);
534 }
535 if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) {
536 PROC_ITIMLOCK(p);
537 if (itimerdecr(&pstats->p_timer[ITIMER_PROF],
538 tick * cnt) == 0)
539 flags |= TDF_PROFPEND | TDF_ASTPENDING;
540 PROC_ITIMUNLOCK(p);
541 }
542 if (flags != 0) {
543 thread_lock(td);
544 td->td_flags |= flags;
545 thread_unlock(td);
546 }
547
548 #ifdef HWPMC_HOOKS
549 if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid)))
550 PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL);
551 if (td->td_intr_frame != NULL)
552 PMC_SOFT_CALL_TF( , , clock, hard, td->td_intr_frame);
553 #endif
554 /* We are in charge to handle this tick duty. */
555 if (newticks > 0) {
556 tc_ticktock(newticks);
557 #ifdef DEVICE_POLLING
558 /* Dangerous and no need to call these things concurrently. */
559 if (atomic_cmpset_acq_int(&devpoll_run, 0, 1)) {
560 /* This is very short and quick. */
561 hardclock_device_poll();
562 atomic_store_rel_int(&devpoll_run, 0);
563 }
564 #endif /* DEVICE_POLLING */
565 if (watchdog_enabled > 0) {
566 i = atomic_fetchadd_int(&watchdog_ticks, -newticks);
567 if (i > 0 && i <= newticks)
568 watchdog_fire();
569 }
570 }
571 if (curcpu == CPU_FIRST())
572 cpu_tick_calibration();
573 }
574
575 void
hardclock_sync(int cpu)576 hardclock_sync(int cpu)
577 {
578 int *t = DPCPU_ID_PTR(cpu, pcputicks);
579
580 *t = ticks;
581 }
582
583 /*
584 * Compute number of ticks in the specified amount of time.
585 */
586 int
tvtohz(struct timeval * tv)587 tvtohz(struct timeval *tv)
588 {
589 unsigned long ticks;
590 long sec, usec;
591
592 /*
593 * If the number of usecs in the whole seconds part of the time
594 * difference fits in a long, then the total number of usecs will
595 * fit in an unsigned long. Compute the total and convert it to
596 * ticks, rounding up and adding 1 to allow for the current tick
597 * to expire. Rounding also depends on unsigned long arithmetic
598 * to avoid overflow.
599 *
600 * Otherwise, if the number of ticks in the whole seconds part of
601 * the time difference fits in a long, then convert the parts to
602 * ticks separately and add, using similar rounding methods and
603 * overflow avoidance. This method would work in the previous
604 * case but it is slightly slower and assumes that hz is integral.
605 *
606 * Otherwise, round the time difference down to the maximum
607 * representable value.
608 *
609 * If ints have 32 bits, then the maximum value for any timeout in
610 * 10ms ticks is 248 days.
611 */
612 sec = tv->tv_sec;
613 usec = tv->tv_usec;
614 if (usec < 0) {
615 sec--;
616 usec += 1000000;
617 }
618 if (sec < 0) {
619 #ifdef DIAGNOSTIC
620 if (usec > 0) {
621 sec++;
622 usec -= 1000000;
623 }
624 printf("tvotohz: negative time difference %ld sec %ld usec\n",
625 sec, usec);
626 #endif
627 ticks = 1;
628 } else if (sec <= LONG_MAX / 1000000)
629 ticks = howmany(sec * 1000000 + (unsigned long)usec, tick) + 1;
630 else if (sec <= LONG_MAX / hz)
631 ticks = sec * hz
632 + howmany((unsigned long)usec, tick) + 1;
633 else
634 ticks = LONG_MAX;
635 if (ticks > INT_MAX)
636 ticks = INT_MAX;
637 return ((int)ticks);
638 }
639
640 /*
641 * Start profiling on a process.
642 *
643 * Kernel profiling passes proc0 which never exits and hence
644 * keeps the profile clock running constantly.
645 */
646 void
startprofclock(struct proc * p)647 startprofclock(struct proc *p)
648 {
649
650 PROC_LOCK_ASSERT(p, MA_OWNED);
651 if (p->p_flag & P_STOPPROF)
652 return;
653 if ((p->p_flag & P_PROFIL) == 0) {
654 p->p_flag |= P_PROFIL;
655 mtx_lock(&time_lock);
656 if (++profprocs == 1)
657 cpu_startprofclock();
658 mtx_unlock(&time_lock);
659 }
660 }
661
662 /*
663 * Stop profiling on a process.
664 */
665 void
stopprofclock(struct proc * p)666 stopprofclock(struct proc *p)
667 {
668
669 PROC_LOCK_ASSERT(p, MA_OWNED);
670 if (p->p_flag & P_PROFIL) {
671 if (p->p_profthreads != 0) {
672 while (p->p_profthreads != 0) {
673 p->p_flag |= P_STOPPROF;
674 msleep(&p->p_profthreads, &p->p_mtx, PPAUSE,
675 "stopprof", 0);
676 }
677 }
678 if ((p->p_flag & P_PROFIL) == 0)
679 return;
680 p->p_flag &= ~P_PROFIL;
681 mtx_lock(&time_lock);
682 if (--profprocs == 0)
683 cpu_stopprofclock();
684 mtx_unlock(&time_lock);
685 }
686 }
687
688 /*
689 * Statistics clock. Updates rusage information and calls the scheduler
690 * to adjust priorities of the active thread.
691 *
692 * This should be called by all active processors.
693 */
694 void
statclock(int usermode)695 statclock(int usermode)
696 {
697
698 statclock_cnt(1, usermode);
699 }
700
701 void
statclock_cnt(int cnt,int usermode)702 statclock_cnt(int cnt, int usermode)
703 {
704 struct rusage *ru;
705 struct vmspace *vm;
706 struct thread *td;
707 struct proc *p;
708 long rss;
709 long *cp_time;
710
711 td = curthread;
712 p = td->td_proc;
713
714 cp_time = (long *)PCPU_PTR(cp_time);
715 if (usermode) {
716 /*
717 * Charge the time as appropriate.
718 */
719 td->td_uticks += cnt;
720 if (p->p_nice > NZERO)
721 cp_time[CP_NICE] += cnt;
722 else
723 cp_time[CP_USER] += cnt;
724 } else {
725 /*
726 * Came from kernel mode, so we were:
727 * - handling an interrupt,
728 * - doing syscall or trap work on behalf of the current
729 * user process, or
730 * - spinning in the idle loop.
731 * Whichever it is, charge the time as appropriate.
732 * Note that we charge interrupts to the current process,
733 * regardless of whether they are ``for'' that process,
734 * so that we know how much of its real time was spent
735 * in ``non-process'' (i.e., interrupt) work.
736 */
737 if ((td->td_pflags & TDP_ITHREAD) ||
738 td->td_intr_nesting_level >= 2) {
739 td->td_iticks += cnt;
740 cp_time[CP_INTR] += cnt;
741 } else {
742 td->td_pticks += cnt;
743 td->td_sticks += cnt;
744 if (!TD_IS_IDLETHREAD(td))
745 cp_time[CP_SYS] += cnt;
746 else
747 cp_time[CP_IDLE] += cnt;
748 }
749 }
750
751 /* Update resource usage integrals and maximums. */
752 MPASS(p->p_vmspace != NULL);
753 vm = p->p_vmspace;
754 ru = &td->td_ru;
755 ru->ru_ixrss += pgtok(vm->vm_tsize) * cnt;
756 ru->ru_idrss += pgtok(vm->vm_dsize) * cnt;
757 ru->ru_isrss += pgtok(vm->vm_ssize) * cnt;
758 rss = pgtok(vmspace_resident_count(vm));
759 if (ru->ru_maxrss < rss)
760 ru->ru_maxrss = rss;
761 KTR_POINT2(KTR_SCHED, "thread", sched_tdname(td), "statclock",
762 "prio:%d", td->td_priority, "stathz:%d", (stathz)?stathz:hz);
763 SDT_PROBE2(sched, , , tick, td, td->td_proc);
764 thread_lock_flags(td, MTX_QUIET);
765 for ( ; cnt > 0; cnt--)
766 sched_clock(td);
767 thread_unlock(td);
768 #ifdef HWPMC_HOOKS
769 if (td->td_intr_frame != NULL)
770 PMC_SOFT_CALL_TF( , , clock, stat, td->td_intr_frame);
771 #endif
772 }
773
774 void
profclock(int usermode,uintfptr_t pc)775 profclock(int usermode, uintfptr_t pc)
776 {
777
778 profclock_cnt(1, usermode, pc);
779 }
780
781 void
profclock_cnt(int cnt,int usermode,uintfptr_t pc)782 profclock_cnt(int cnt, int usermode, uintfptr_t pc)
783 {
784 struct thread *td;
785 #ifdef GPROF
786 struct gmonparam *g;
787 uintfptr_t i;
788 #endif
789
790 td = curthread;
791 if (usermode) {
792 /*
793 * Came from user mode; CPU was in user state.
794 * If this process is being profiled, record the tick.
795 * if there is no related user location yet, don't
796 * bother trying to count it.
797 */
798 if (td->td_proc->p_flag & P_PROFIL)
799 addupc_intr(td, pc, cnt);
800 }
801 #ifdef GPROF
802 else {
803 /*
804 * Kernel statistics are just like addupc_intr, only easier.
805 */
806 g = &_gmonparam;
807 if (g->state == GMON_PROF_ON && pc >= g->lowpc) {
808 i = PC_TO_I(g, pc);
809 if (i < g->textsize) {
810 KCOUNT(g, i) += cnt;
811 }
812 }
813 }
814 #endif
815 #ifdef HWPMC_HOOKS
816 if (td->td_intr_frame != NULL)
817 PMC_SOFT_CALL_TF( , , clock, prof, td->td_intr_frame);
818 #endif
819 }
820
821 /*
822 * Return information about system clocks.
823 */
824 static int
sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS)825 sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS)
826 {
827 struct clockinfo clkinfo;
828 /*
829 * Construct clockinfo structure.
830 */
831 bzero(&clkinfo, sizeof(clkinfo));
832 clkinfo.hz = hz;
833 clkinfo.tick = tick;
834 clkinfo.profhz = profhz;
835 clkinfo.stathz = stathz ? stathz : hz;
836 return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req));
837 }
838
839 SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate,
840 CTLTYPE_STRUCT|CTLFLAG_RD|CTLFLAG_MPSAFE,
841 0, 0, sysctl_kern_clockrate, "S,clockinfo",
842 "Rate and period of various kernel clocks");
843
844 static void
watchdog_config(void * unused __unused,u_int cmd,int * error)845 watchdog_config(void *unused __unused, u_int cmd, int *error)
846 {
847 u_int u;
848
849 u = cmd & WD_INTERVAL;
850 if (u >= WD_TO_1SEC) {
851 watchdog_ticks = (1 << (u - WD_TO_1SEC)) * hz;
852 watchdog_enabled = 1;
853 *error = 0;
854 } else {
855 watchdog_enabled = 0;
856 }
857 }
858
859 /*
860 * Handle a watchdog timeout by dumping interrupt information and
861 * then either dropping to DDB or panicking.
862 */
863 static void
watchdog_fire(void)864 watchdog_fire(void)
865 {
866 int nintr;
867 uint64_t inttotal;
868 u_long *curintr;
869 char *curname;
870
871 curintr = intrcnt;
872 curname = intrnames;
873 inttotal = 0;
874 nintr = sintrcnt / sizeof(u_long);
875
876 printf("interrupt total\n");
877 while (--nintr >= 0) {
878 if (*curintr)
879 printf("%-12s %20lu\n", curname, *curintr);
880 curname += strlen(curname) + 1;
881 inttotal += *curintr++;
882 }
883 printf("Total %20ju\n", (uintmax_t)inttotal);
884
885 #if defined(KDB) && !defined(KDB_UNATTENDED)
886 kdb_backtrace();
887 kdb_enter(KDB_WHY_WATCHDOG, "watchdog timeout");
888 #else
889 panic("watchdog timeout");
890 #endif
891 }
892