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
37 #include <sys/cdefs.h>
38 #include "opt_kdb.h"
39 #include "opt_device_polling.h"
40 #include "opt_hwpmc_hooks.h"
41 #include "opt_ntp.h"
42 #include "opt_watchdog.h"
43
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/callout.h>
47 #include <sys/epoch.h>
48 #include <sys/eventhandler.h>
49 #include <sys/gtaskqueue.h>
50 #include <sys/kdb.h>
51 #include <sys/kernel.h>
52 #include <sys/kthread.h>
53 #include <sys/ktr.h>
54 #include <sys/lock.h>
55 #include <sys/mutex.h>
56 #include <sys/proc.h>
57 #include <sys/resource.h>
58 #include <sys/resourcevar.h>
59 #include <sys/sched.h>
60 #include <sys/sdt.h>
61 #include <sys/signalvar.h>
62 #include <sys/sleepqueue.h>
63 #include <sys/smp.h>
64 #include <vm/vm.h>
65 #include <vm/pmap.h>
66 #include <vm/vm_map.h>
67 #include <sys/sysctl.h>
68 #include <sys/bus.h>
69 #include <sys/interrupt.h>
70 #include <sys/limits.h>
71 #include <sys/timetc.h>
72
73 #ifdef HWPMC_HOOKS
74 #include <sys/pmckern.h>
75 PMC_SOFT_DEFINE( , , clock, hard);
76 PMC_SOFT_DEFINE( , , clock, stat);
77 PMC_SOFT_DEFINE_EX( , , clock, prof, \
78 cpu_startprofclock, cpu_stopprofclock);
79 #endif
80
81 #ifdef DEVICE_POLLING
82 extern void hardclock_device_poll(void);
83 #endif /* DEVICE_POLLING */
84
85 /* Spin-lock protecting profiling statistics. */
86 static struct mtx time_lock;
87
88 SDT_PROVIDER_DECLARE(sched);
89 SDT_PROBE_DEFINE2(sched, , , tick, "struct thread *", "struct proc *");
90
91 static int
sysctl_kern_cp_time(SYSCTL_HANDLER_ARGS)92 sysctl_kern_cp_time(SYSCTL_HANDLER_ARGS)
93 {
94 int error;
95 long cp_time[CPUSTATES];
96 #ifdef SCTL_MASK32
97 int i;
98 unsigned int cp_time32[CPUSTATES];
99 #endif
100
101 read_cpu_time(cp_time);
102 #ifdef SCTL_MASK32
103 if (req->flags & SCTL_MASK32) {
104 if (!req->oldptr)
105 return SYSCTL_OUT(req, 0, sizeof(cp_time32));
106 for (i = 0; i < CPUSTATES; i++)
107 cp_time32[i] = (unsigned int)cp_time[i];
108 error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32));
109 } else
110 #endif
111 {
112 if (!req->oldptr)
113 return SYSCTL_OUT(req, 0, sizeof(cp_time));
114 error = SYSCTL_OUT(req, cp_time, sizeof(cp_time));
115 }
116 return error;
117 }
118
119 SYSCTL_PROC(_kern, OID_AUTO, cp_time, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE,
120 0,0, sysctl_kern_cp_time, "LU", "CPU time statistics");
121
122 static long empty[CPUSTATES];
123
124 static int
sysctl_kern_cp_times(SYSCTL_HANDLER_ARGS)125 sysctl_kern_cp_times(SYSCTL_HANDLER_ARGS)
126 {
127 struct pcpu *pcpu;
128 int error;
129 int c;
130 long *cp_time;
131 #ifdef SCTL_MASK32
132 unsigned int cp_time32[CPUSTATES];
133 int i;
134 #endif
135
136 if (!req->oldptr) {
137 #ifdef SCTL_MASK32
138 if (req->flags & SCTL_MASK32)
139 return SYSCTL_OUT(req, 0, sizeof(cp_time32) * (mp_maxid + 1));
140 else
141 #endif
142 return SYSCTL_OUT(req, 0, sizeof(long) * CPUSTATES * (mp_maxid + 1));
143 }
144 for (error = 0, c = 0; error == 0 && c <= mp_maxid; c++) {
145 if (!CPU_ABSENT(c)) {
146 pcpu = pcpu_find(c);
147 cp_time = pcpu->pc_cp_time;
148 } else {
149 cp_time = empty;
150 }
151 #ifdef SCTL_MASK32
152 if (req->flags & SCTL_MASK32) {
153 for (i = 0; i < CPUSTATES; i++)
154 cp_time32[i] = (unsigned int)cp_time[i];
155 error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32));
156 } else
157 #endif
158 error = SYSCTL_OUT(req, cp_time, sizeof(long) * CPUSTATES);
159 }
160 return error;
161 }
162
163 SYSCTL_PROC(_kern, OID_AUTO, cp_times, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE,
164 0,0, sysctl_kern_cp_times, "LU", "per-CPU time statistics");
165
166 #ifdef DEADLKRES
167 static const char *blessed[] = {
168 "getblk",
169 "so_snd_sx",
170 "so_rcv_sx",
171 NULL
172 };
173 static int slptime_threshold = 1800;
174 static int blktime_threshold = 900;
175 static int sleepfreq = 3;
176
177 static void
deadlres_td_on_lock(struct proc * p,struct thread * td,int blkticks)178 deadlres_td_on_lock(struct proc *p, struct thread *td, int blkticks)
179 {
180 int tticks;
181
182 sx_assert(&allproc_lock, SX_LOCKED);
183 PROC_LOCK_ASSERT(p, MA_OWNED);
184 THREAD_LOCK_ASSERT(td, MA_OWNED);
185 /*
186 * The thread should be blocked on a turnstile, simply check
187 * if the turnstile channel is in good state.
188 */
189 MPASS(td->td_blocked != NULL);
190
191 tticks = ticks - td->td_blktick;
192 if (tticks > blkticks)
193 /*
194 * Accordingly with provided thresholds, this thread is stuck
195 * for too long on a turnstile.
196 */
197 panic("%s: possible deadlock detected for %p (%s), "
198 "blocked for %d ticks\n", __func__,
199 td, sched_tdname(td), tticks);
200 }
201
202 static void
deadlres_td_sleep_q(struct proc * p,struct thread * td,int slpticks)203 deadlres_td_sleep_q(struct proc *p, struct thread *td, int slpticks)
204 {
205 const void *wchan;
206 int i, slptype, tticks;
207
208 sx_assert(&allproc_lock, SX_LOCKED);
209 PROC_LOCK_ASSERT(p, MA_OWNED);
210 THREAD_LOCK_ASSERT(td, MA_OWNED);
211 /*
212 * Check if the thread is sleeping on a lock, otherwise skip the check.
213 * Drop the thread lock in order to avoid a LOR with the sleepqueue
214 * spinlock.
215 */
216 wchan = td->td_wchan;
217 tticks = ticks - td->td_slptick;
218 slptype = sleepq_type(wchan);
219 if ((slptype == SLEEPQ_SX || slptype == SLEEPQ_LK) &&
220 tticks > slpticks) {
221 /*
222 * Accordingly with provided thresholds, this thread is stuck
223 * for too long on a sleepqueue.
224 * However, being on a sleepqueue, we might still check for the
225 * blessed list.
226 */
227 for (i = 0; blessed[i] != NULL; i++)
228 if (!strcmp(blessed[i], td->td_wmesg))
229 return;
230
231 panic("%s: possible deadlock detected for %p (%s), "
232 "blocked for %d ticks\n", __func__,
233 td, sched_tdname(td), tticks);
234 }
235 }
236
237 static void
deadlkres(void)238 deadlkres(void)
239 {
240 struct proc *p;
241 struct thread *td;
242 int blkticks, slpticks, tryl;
243
244 tryl = 0;
245 for (;;) {
246 blkticks = blktime_threshold * hz;
247 slpticks = slptime_threshold * hz;
248
249 /*
250 * Avoid to sleep on the sx_lock in order to avoid a
251 * possible priority inversion problem leading to
252 * starvation.
253 * If the lock can't be held after 100 tries, panic.
254 */
255 if (!sx_try_slock(&allproc_lock)) {
256 if (tryl > 100)
257 panic("%s: possible deadlock detected "
258 "on allproc_lock\n", __func__);
259 tryl++;
260 pause("allproc", sleepfreq * hz);
261 continue;
262 }
263 tryl = 0;
264 FOREACH_PROC_IN_SYSTEM(p) {
265 PROC_LOCK(p);
266 if (p->p_state == PRS_NEW) {
267 PROC_UNLOCK(p);
268 continue;
269 }
270 FOREACH_THREAD_IN_PROC(p, td) {
271 thread_lock(td);
272 if (TD_ON_LOCK(td))
273 deadlres_td_on_lock(p, td,
274 blkticks);
275 else if (TD_IS_SLEEPING(td))
276 deadlres_td_sleep_q(p, td,
277 slpticks);
278 thread_unlock(td);
279 }
280 PROC_UNLOCK(p);
281 }
282 sx_sunlock(&allproc_lock);
283
284 /* Sleep for sleepfreq seconds. */
285 pause("-", sleepfreq * hz);
286 }
287 }
288
289 static struct kthread_desc deadlkres_kd = {
290 "deadlkres",
291 deadlkres,
292 (struct thread **)NULL
293 };
294
295 SYSINIT(deadlkres, SI_SUB_CLOCKS, SI_ORDER_ANY, kthread_start, &deadlkres_kd);
296
297 static SYSCTL_NODE(_debug, OID_AUTO, deadlkres, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
298 "Deadlock resolver");
299 SYSCTL_INT(_debug_deadlkres, OID_AUTO, slptime_threshold, CTLFLAG_RWTUN,
300 &slptime_threshold, 0,
301 "Number of seconds within is valid to sleep on a sleepqueue");
302 SYSCTL_INT(_debug_deadlkres, OID_AUTO, blktime_threshold, CTLFLAG_RWTUN,
303 &blktime_threshold, 0,
304 "Number of seconds within is valid to block on a turnstile");
305 SYSCTL_INT(_debug_deadlkres, OID_AUTO, sleepfreq, CTLFLAG_RWTUN, &sleepfreq, 0,
306 "Number of seconds between any deadlock resolver thread run");
307 #endif /* DEADLKRES */
308
309 void
read_cpu_time(long * cp_time)310 read_cpu_time(long *cp_time)
311 {
312 struct pcpu *pc;
313 int i, j;
314
315 /* Sum up global cp_time[]. */
316 bzero(cp_time, sizeof(long) * CPUSTATES);
317 CPU_FOREACH(i) {
318 pc = pcpu_find(i);
319 for (j = 0; j < CPUSTATES; j++)
320 cp_time[j] += pc->pc_cp_time[j];
321 }
322 }
323
324 #include <sys/watchdog.h>
325
326 static long watchdog_ticks;
327 static int watchdog_enabled;
328 static void watchdog_fire(void);
329 static void watchdog_config(void *, u_int, int *);
330
331 static void
watchdog_attach(void)332 watchdog_attach(void)
333 {
334 EVENTHANDLER_REGISTER(watchdog_list, watchdog_config, NULL, 0);
335 }
336
337 /*
338 * Clock handling routines.
339 *
340 * This code is written to operate with two timers that run independently of
341 * each other.
342 *
343 * The main timer, running hz times per second, is used to trigger interval
344 * timers, timeouts and rescheduling as needed.
345 *
346 * The second timer handles kernel and user profiling,
347 * and does resource use estimation. If the second timer is programmable,
348 * it is randomized to avoid aliasing between the two clocks. For example,
349 * the randomization prevents an adversary from always giving up the cpu
350 * just before its quantum expires. Otherwise, it would never accumulate
351 * cpu ticks. The mean frequency of the second timer is stathz.
352 *
353 * If no second timer exists, stathz will be zero; in this case we drive
354 * profiling and statistics off the main clock. This WILL NOT be accurate;
355 * do not do it unless absolutely necessary.
356 *
357 * The statistics clock may (or may not) be run at a higher rate while
358 * profiling. This profile clock runs at profhz. We require that profhz
359 * be an integral multiple of stathz.
360 *
361 * If the statistics clock is running fast, it must be divided by the ratio
362 * profhz/stathz for statistics. (For profiling, every tick counts.)
363 *
364 * Time-of-day is maintained using a "timecounter", which may or may
365 * not be related to the hardware generating the above mentioned
366 * interrupts.
367 */
368
369 int stathz;
370 int profhz;
371 int profprocs;
372 int psratio;
373
374 DPCPU_DEFINE_STATIC(long, pcputicks); /* Per-CPU version of ticks. */
375 #ifdef DEVICE_POLLING
376 static int devpoll_run = 0;
377 #endif
378
379 static void
ast_oweupc(struct thread * td,int tda __unused)380 ast_oweupc(struct thread *td, int tda __unused)
381 {
382 if ((td->td_proc->p_flag & P_PROFIL) == 0)
383 return;
384 addupc_task(td, td->td_profil_addr, td->td_profil_ticks);
385 td->td_profil_ticks = 0;
386 td->td_pflags &= ~TDP_OWEUPC;
387 }
388
389 static void
ast_alrm(struct thread * td,int tda __unused)390 ast_alrm(struct thread *td, int tda __unused)
391 {
392 struct proc *p;
393
394 p = td->td_proc;
395 PROC_LOCK(p);
396 kern_psignal(p, SIGVTALRM);
397 PROC_UNLOCK(p);
398 }
399
400 static void
ast_prof(struct thread * td,int tda __unused)401 ast_prof(struct thread *td, int tda __unused)
402 {
403 struct proc *p;
404
405 p = td->td_proc;
406 PROC_LOCK(p);
407 kern_psignal(p, SIGPROF);
408 PROC_UNLOCK(p);
409 }
410
411 /*
412 * Initialize clock frequencies and start both clocks running.
413 */
414 static void
initclocks(void * dummy __unused)415 initclocks(void *dummy __unused)
416 {
417 int i;
418
419 /*
420 * Set divisors to 1 (normal case) and let the machine-specific
421 * code do its bit.
422 */
423 mtx_init(&time_lock, "time lock", NULL, MTX_DEF);
424 cpu_initclocks();
425
426 /*
427 * Compute profhz/stathz, and fix profhz if needed.
428 */
429 i = stathz ? stathz : hz;
430 if (profhz == 0)
431 profhz = i;
432 psratio = profhz / i;
433
434 ast_register(TDA_OWEUPC, ASTR_ASTF_REQUIRED, 0, ast_oweupc);
435 ast_register(TDA_ALRM, ASTR_ASTF_REQUIRED, 0, ast_alrm);
436 ast_register(TDA_PROF, ASTR_ASTF_REQUIRED, 0, ast_prof);
437
438 #ifdef SW_WATCHDOG
439 /* Enable hardclock watchdog now, even if a hardware watchdog exists. */
440 watchdog_attach();
441 #else
442 /* Volunteer to run a software watchdog. */
443 if (wdog_software_attach == NULL)
444 wdog_software_attach = watchdog_attach;
445 #endif
446 }
447 SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL);
448
449 static __noinline void
hardclock_itimer(struct thread * td,struct pstats * pstats,int cnt,int usermode)450 hardclock_itimer(struct thread *td, struct pstats *pstats, int cnt, int usermode)
451 {
452 struct proc *p;
453 int ast;
454
455 ast = 0;
456 p = td->td_proc;
457 if (usermode &&
458 timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) {
459 PROC_ITIMLOCK(p);
460 if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL],
461 tick * cnt) == 0)
462 ast |= TDAI(TDA_ALRM);
463 PROC_ITIMUNLOCK(p);
464 }
465 if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) {
466 PROC_ITIMLOCK(p);
467 if (itimerdecr(&pstats->p_timer[ITIMER_PROF],
468 tick * cnt) == 0)
469 ast |= TDAI(TDA_PROF);
470 PROC_ITIMUNLOCK(p);
471 }
472 if (ast != 0)
473 ast_sched_mask(td, ast);
474 }
475
476 void
hardclock(int cnt,int usermode)477 hardclock(int cnt, int usermode)
478 {
479 struct pstats *pstats;
480 struct thread *td = curthread;
481 struct proc *p = td->td_proc;
482 long global, newticks, *t;
483
484 /*
485 * Update per-CPU and possibly global ticks values.
486 */
487 t = DPCPU_PTR(pcputicks);
488 *t += cnt;
489 global = atomic_load_long(&ticksl);
490 do {
491 newticks = *t - global;
492 if (newticks <= 0) {
493 if (newticks < -1)
494 *t = global - 1;
495 newticks = 0;
496 break;
497 }
498 } while (!atomic_fcmpset_long(&ticksl, &global, *t));
499
500 /*
501 * Run current process's virtual and profile time, as needed.
502 */
503 pstats = p->p_stats;
504 if (__predict_false(
505 timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) ||
506 timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)))
507 hardclock_itimer(td, pstats, cnt, usermode);
508
509 #ifdef HWPMC_HOOKS
510 if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid)))
511 PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL);
512 if (td->td_intr_frame != NULL)
513 PMC_SOFT_CALL_TF( , , clock, hard, td->td_intr_frame);
514 #endif
515 /* We are in charge to handle this tick duty. */
516 if (newticks > 0) {
517 tc_ticktock(newticks);
518 #ifdef DEVICE_POLLING
519 /* Dangerous and no need to call these things concurrently. */
520 if (atomic_cmpset_acq_int(&devpoll_run, 0, 1)) {
521 /* This is very short and quick. */
522 hardclock_device_poll();
523 atomic_store_rel_int(&devpoll_run, 0);
524 }
525 #endif /* DEVICE_POLLING */
526 if (watchdog_enabled > 0) {
527 long left;
528
529 left = atomic_fetchadd_long(&watchdog_ticks, -newticks);
530 if (left > 0 && left <= newticks)
531 watchdog_fire();
532 }
533 intr_event_handle(clk_intr_event, NULL);
534 }
535 if (curcpu == CPU_FIRST())
536 cpu_tick_calibration();
537 if (__predict_false(DPCPU_GET(epoch_cb_count)))
538 GROUPTASK_ENQUEUE(DPCPU_PTR(epoch_cb_task));
539 }
540
541 void
hardclock_sync(int cpu)542 hardclock_sync(int cpu)
543 {
544 long *t;
545
546 KASSERT(!CPU_ABSENT(cpu), ("Absent CPU %d", cpu));
547
548 t = DPCPU_ID_PTR(cpu, pcputicks);
549 *t = ticksl;
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
tvtohz(struct timeval * tv)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
startprofclock(struct proc * p)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
stopprofclock(struct proc * p)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
statclock(int cnt,int usermode)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
profclock(int cnt,int usermode,uintfptr_t pc)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
sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS)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
watchdog_config(void * unused __unused,u_int cmd,int * error)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
watchdog_fire(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