xref: /dflybsd-src/sys/kern/kern_shutdown.c (revision f8c7a42d831dbc6ff46a5448591fd28de34b649b)
1 /*-
2  * Copyright (c) 1986, 1988, 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  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)kern_shutdown.c	8.3 (Berkeley) 1/21/94
39  * $FreeBSD: src/sys/kern/kern_shutdown.c,v 1.72.2.12 2002/02/21 19:15:10 dillon Exp $
40  * $DragonFly: src/sys/kern/kern_shutdown.c,v 1.43 2006/12/18 20:41:01 dillon Exp $
41  */
42 
43 #include "opt_ddb.h"
44 #include "opt_ddb_trace.h"
45 #include "opt_hw_wdog.h"
46 #include "opt_panic.h"
47 #include "opt_show_busybufs.h"
48 
49 #include <sys/param.h>
50 #include <sys/systm.h>
51 #include <sys/eventhandler.h>
52 #include <sys/buf.h>
53 #include <sys/disklabel.h>
54 #include <sys/reboot.h>
55 #include <sys/proc.h>
56 #include <sys/vnode.h>
57 #include <sys/kernel.h>
58 #include <sys/kthread.h>
59 #include <sys/malloc.h>
60 #include <sys/mount.h>
61 #include <sys/queue.h>
62 #include <sys/sysctl.h>
63 #include <sys/vkernel.h>
64 #include <sys/conf.h>
65 #include <sys/sysproto.h>
66 #include <sys/device.h>
67 #include <sys/cons.h>
68 #include <sys/shm.h>
69 #include <sys/kern_syscall.h>
70 #include <vm/vm_map.h>
71 #include <vm/pmap.h>
72 
73 #include <sys/thread2.h>
74 #include <sys/buf2.h>
75 
76 #include <machine/pcb.h>
77 #include <machine/clock.h>
78 #include <machine/md_var.h>
79 #include <machine/smp.h>		/* smp_active_mask, cpuid */
80 
81 #include <sys/signalvar.h>
82 
83 #ifndef PANIC_REBOOT_WAIT_TIME
84 #define PANIC_REBOOT_WAIT_TIME 15 /* default to 15 seconds */
85 #endif
86 
87 /*
88  * Note that stdarg.h and the ANSI style va_start macro is used for both
89  * ANSI and traditional C compilers.  We use the machine version to stay
90  * within the confines of the kernel header files.
91  */
92 #include <machine/stdarg.h>
93 
94 #ifdef DDB
95 #ifdef DDB_UNATTENDED
96 int debugger_on_panic = 0;
97 #else
98 int debugger_on_panic = 1;
99 #endif
100 SYSCTL_INT(_debug, OID_AUTO, debugger_on_panic, CTLFLAG_RW,
101 	&debugger_on_panic, 0, "Run debugger on kernel panic");
102 
103 extern void db_print_backtrace(void);
104 
105 #ifdef DDB_TRACE
106 int trace_on_panic = 1;
107 #else
108 int trace_on_panic = 0;
109 #endif
110 SYSCTL_INT(_debug, OID_AUTO, trace_on_panic, CTLFLAG_RW,
111 	&trace_on_panic, 0, "Print stack trace on kernel panic");
112 #endif
113 
114 static int sync_on_panic = 1;
115 SYSCTL_INT(_kern, OID_AUTO, sync_on_panic, CTLFLAG_RW,
116 	&sync_on_panic, 0, "Do a sync before rebooting from a panic");
117 
118 SYSCTL_NODE(_kern, OID_AUTO, shutdown, CTLFLAG_RW, 0, "Shutdown environment");
119 
120 #ifdef	HW_WDOG
121 /*
122  * If there is a hardware watchdog, point this at the function needed to
123  * hold it off.
124  * It's needed when the kernel needs to do some lengthy operations.
125  * e.g. in wd.c when dumping core.. It's most annoying to have
126  * your precious core-dump only half written because the wdog kicked in.
127  */
128 watchdog_tickle_fn wdog_tickler = NULL;
129 #endif	/* HW_WDOG */
130 
131 /*
132  * Variable panicstr contains argument to first call to panic; used as flag
133  * to indicate that the kernel has already called panic.
134  */
135 const char *panicstr;
136 
137 int dumping;				/* system is dumping */
138 #ifdef SMP
139 u_int panic_cpu_interlock;		/* panic interlock */
140 globaldata_t panic_cpu_gd;		/* which cpu took the panic */
141 #endif
142 
143 int bootverbose = 0;			/* note: assignment to force non-bss */
144 int cold = 1;				/* note: assignment to force non-bss */
145 
146 static void boot (int) __dead2;
147 static void dumpsys (void);
148 static int setdumpdev (cdev_t dev);
149 static void poweroff_wait (void *, int);
150 static void print_uptime (void);
151 static void shutdown_halt (void *junk, int howto);
152 static void shutdown_panic (void *junk, int howto);
153 static void shutdown_reset (void *junk, int howto);
154 static int shutdown_busycount1(struct buf *bp, void *info);
155 static int shutdown_busycount2(struct buf *bp, void *info);
156 static void shutdown_cleanup_proc(struct proc *p);
157 
158 /* register various local shutdown events */
159 static void
160 shutdown_conf(void *unused)
161 {
162 	EVENTHANDLER_REGISTER(shutdown_final, poweroff_wait, NULL, SHUTDOWN_PRI_FIRST);
163 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_halt, NULL, SHUTDOWN_PRI_LAST + 100);
164 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_panic, NULL, SHUTDOWN_PRI_LAST + 100);
165 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_reset, NULL, SHUTDOWN_PRI_LAST + 200);
166 }
167 
168 SYSINIT(shutdown_conf, SI_SUB_INTRINSIC, SI_ORDER_ANY, shutdown_conf, NULL)
169 
170 /* ARGSUSED */
171 
172 /*
173  * The system call that results in a reboot
174  */
175 int
176 sys_reboot(struct reboot_args *uap)
177 {
178 	struct thread *td = curthread;
179 	int error;
180 
181 	if ((error = suser(td)))
182 		return (error);
183 
184 	boot(uap->opt);
185 	return (0);
186 }
187 
188 /*
189  * Called by events that want to shut down.. e.g  <CTL><ALT><DEL> on a PC
190  */
191 static int shutdown_howto = 0;
192 
193 void
194 shutdown_nice(int howto)
195 {
196 	shutdown_howto = howto;
197 
198 	/* Send a signal to init(8) and have it shutdown the world */
199 	if (initproc != NULL) {
200 		ksignal(initproc, SIGINT);
201 	} else {
202 		/* No init(8) running, so simply reboot */
203 		boot(RB_NOSYNC);
204 	}
205 	return;
206 }
207 static int	waittime = -1;
208 static struct thread *dumpthread;
209 static struct pcb dumppcb;
210 
211 static void
212 print_uptime()
213 {
214 	int f;
215 	struct timespec ts;
216 
217 	getnanouptime(&ts);
218 	printf("Uptime: ");
219 	f = 0;
220 	if (ts.tv_sec >= 86400) {
221 		printf("%ldd", ts.tv_sec / 86400);
222 		ts.tv_sec %= 86400;
223 		f = 1;
224 	}
225 	if (f || ts.tv_sec >= 3600) {
226 		printf("%ldh", ts.tv_sec / 3600);
227 		ts.tv_sec %= 3600;
228 		f = 1;
229 	}
230 	if (f || ts.tv_sec >= 60) {
231 		printf("%ldm", ts.tv_sec / 60);
232 		ts.tv_sec %= 60;
233 		f = 1;
234 	}
235 	printf("%lds\n", ts.tv_sec);
236 }
237 
238 /*
239  *  Go through the rigmarole of shutting down..
240  * this used to be in machdep.c but I'll be dammned if I could see
241  * anything machine dependant in it.
242  */
243 static void
244 boot(int howto)
245 {
246 	/*
247 	 * Get rid of any user scheduler baggage and then give
248 	 * us a high priority.
249 	 */
250 	if (curthread->td_release)
251 		curthread->td_release(curthread);
252 	lwkt_setpri_self(TDPRI_MAX);
253 
254 	/* collect extra flags that shutdown_nice might have set */
255 	howto |= shutdown_howto;
256 
257 #ifdef SMP
258 	/*
259 	 * We really want to shutdown on the BSP.  Subsystems such as ACPI
260 	 * can't power-down the box otherwise.
261 	 */
262 	if (smp_active_mask > 1) {
263 		printf("boot() called on cpu#%d\n", mycpu->gd_cpuid);
264 	}
265 	if (panicstr == NULL && mycpu->gd_cpuid != 0) {
266 		printf("Switching to cpu #0 for shutdown\n");
267 		lwkt_setcpu_self(globaldata_find(0));
268 	}
269 #endif
270 	/*
271 	 * Do any callouts that should be done BEFORE syncing the filesystems.
272 	 */
273 	EVENTHANDLER_INVOKE(shutdown_pre_sync, howto);
274 
275 	/*
276 	 * Try to get rid of any remaining FS references.  The calling
277 	 * process, proc0, and init may still hold references.  The
278 	 * VFS cache subsystem may still hold a root reference to root.
279 	 */
280 	if (panicstr == NULL) {
281 		shutdown_cleanup_proc(curproc);
282 		shutdown_cleanup_proc(&proc0);
283 		if (initproc)
284 			shutdown_cleanup_proc(initproc);
285 		vfs_cache_setroot(NULL, NULL);
286 	}
287 
288 	/*
289 	 * Now sync filesystems
290 	 */
291 	if (!cold && (howto & RB_NOSYNC) == 0 && waittime < 0) {
292 		int iter, nbusy, pbusy;
293 
294 		waittime = 0;
295 		printf("\nsyncing disks... ");
296 
297 		sys_sync(NULL);	/* YYY was sync(&proc0, NULL). why proc0 ? */
298 
299 		/*
300 		 * With soft updates, some buffers that are
301 		 * written will be remarked as dirty until other
302 		 * buffers are written.
303 		 */
304 		for (iter = pbusy = 0; iter < 20; iter++) {
305 			nbusy = scan_all_buffers(shutdown_busycount1, NULL);
306 			if (nbusy == 0)
307 				break;
308 			printf("%d ", nbusy);
309 			if (nbusy < pbusy)
310 				iter = 0;
311 			pbusy = nbusy;
312 			/*
313 			 * XXX:
314 			 * Process soft update work queue if buffers don't sync
315 			 * after 6 iterations by permitting the syncer to run.
316 			 */
317 			if (iter > 5 && bioops.io_sync)
318 				(*bioops.io_sync)(NULL);
319 			sys_sync(NULL); /* YYY was sync(&proc0, NULL). why proc0 ? */
320 			tsleep(boot, 0, "shutdn", hz * iter / 20 + 1);
321 		}
322 		printf("\n");
323 		/*
324 		 * Count only busy local buffers to prevent forcing
325 		 * a fsck if we're just a client of a wedged NFS server
326 		 */
327 		nbusy = scan_all_buffers(shutdown_busycount2, NULL);
328 		if (nbusy) {
329 			/*
330 			 * Failed to sync all blocks. Indicate this and don't
331 			 * unmount filesystems (thus forcing an fsck on reboot).
332 			 */
333 			printf("giving up on %d buffers\n", nbusy);
334 #ifdef DDB
335 			Debugger("busy buffer problem");
336 #endif /* DDB */
337 			tsleep(boot, 0, "shutdn", hz * 5 + 1);
338 		} else {
339 			printf("done\n");
340 			/*
341 			 * Unmount filesystems
342 			 */
343 			if (panicstr == NULL)
344 				vfs_unmountall();
345 		}
346 		tsleep(boot, 0, "shutdn", hz / 10 + 1);
347 	}
348 
349 	print_uptime();
350 
351 	/*
352 	 * Ok, now do things that assume all filesystem activity has
353 	 * been completed.
354 	 */
355 	EVENTHANDLER_INVOKE(shutdown_post_sync, howto);
356 	crit_enter();
357 	if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP && !cold)
358 		dumpsys();
359 
360 	/* Now that we're going to really halt the system... */
361 	EVENTHANDLER_INVOKE(shutdown_final, howto);
362 
363 	for(;;) ;	/* safety against shutdown_reset not working */
364 	/* NOTREACHED */
365 }
366 
367 static int
368 shutdown_busycount1(struct buf *bp, void *info)
369 {
370 	if ((bp->b_flags & B_INVAL) == 0 && BUF_REFCNT(bp) > 0)
371 		return(1);
372 	if ((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI)
373 		return (1);
374 	return (0);
375 }
376 
377 static int
378 shutdown_busycount2(struct buf *bp, void *info)
379 {
380 	if (((bp->b_flags & B_INVAL) == 0 && BUF_REFCNT(bp)) ||
381 	    ((bp->b_flags & (B_DELWRI|B_INVAL)) == B_DELWRI)) {
382 		/*
383 		 * Only count buffers undergoing write I/O
384 		 * on the related vnode.
385 		 */
386 		if (bp->b_vp == NULL ||
387 		    bp->b_vp->v_track_write.bk_active == 0) {
388 			return (0);
389 		}
390 #if defined(SHOW_BUSYBUFS) || defined(DIAGNOSTIC)
391 		printf(
392 	    "%p dev:?, flags:%08x, loffset:%lld, doffset:%lld\n",
393 		    bp,
394 		    bp->b_flags, bp->b_loffset,
395 		    bp->b_bio2.bio_offset);
396 #endif
397 		return(1);
398 	}
399 	return(0);
400 }
401 
402 /*
403  * If the shutdown was a clean halt, behave accordingly.
404  */
405 static void
406 shutdown_halt(void *junk, int howto)
407 {
408 	if (howto & RB_HALT) {
409 		printf("\n");
410 		printf("The operating system has halted.\n");
411 		printf("Please press any key to reboot.\n\n");
412 		switch (cngetc()) {
413 		case -1:		/* No console, just die */
414 			cpu_halt();
415 			/* NOTREACHED */
416 		default:
417 			howto &= ~RB_HALT;
418 			break;
419 		}
420 	}
421 }
422 
423 /*
424  * Check to see if the system paniced, pause and then reboot
425  * according to the specified delay.
426  */
427 static void
428 shutdown_panic(void *junk, int howto)
429 {
430 	int loop;
431 
432 	if (howto & RB_DUMP) {
433 		if (PANIC_REBOOT_WAIT_TIME != 0) {
434 			if (PANIC_REBOOT_WAIT_TIME != -1) {
435 				printf("Automatic reboot in %d seconds - "
436 				       "press a key on the console to abort\n",
437 					PANIC_REBOOT_WAIT_TIME);
438 				for (loop = PANIC_REBOOT_WAIT_TIME * 10;
439 				     loop > 0; --loop) {
440 					DELAY(1000 * 100); /* 1/10th second */
441 					/* Did user type a key? */
442 					if (cncheckc() != -1)
443 						break;
444 				}
445 				if (!loop)
446 					return;
447 			}
448 		} else { /* zero time specified - reboot NOW */
449 			return;
450 		}
451 		printf("--> Press a key on the console to reboot,\n");
452 		printf("--> or switch off the system now.\n");
453 		cngetc();
454 	}
455 }
456 
457 /*
458  * Everything done, now reset
459  */
460 static void
461 shutdown_reset(void *junk, int howto)
462 {
463 	printf("Rebooting...\n");
464 	DELAY(1000000);	/* wait 1 sec for printf's to complete and be read */
465 	/* cpu_boot(howto); */ /* doesn't do anything at the moment */
466 	cpu_reset();
467 	/* NOTREACHED */ /* assuming reset worked */
468 }
469 
470 /*
471  * Try to remove FS references in the specified process.  This function
472  * is used during shutdown
473  */
474 static
475 void
476 shutdown_cleanup_proc(struct proc *p)
477 {
478 	struct filedesc *fdp;
479 	struct vmspace *vm;
480 
481 	if (p == NULL)
482 		return;
483 	if ((fdp = p->p_fd) != NULL) {
484 		kern_closefrom(0);
485 		if (fdp->fd_cdir) {
486 			cache_drop(&fdp->fd_ncdir);
487 			vrele(fdp->fd_cdir);
488 			fdp->fd_cdir = NULL;
489 		}
490 		if (fdp->fd_rdir) {
491 			cache_drop(&fdp->fd_nrdir);
492 			vrele(fdp->fd_rdir);
493 			fdp->fd_rdir = NULL;
494 		}
495 		if (fdp->fd_jdir) {
496 			cache_drop(&fdp->fd_njdir);
497 			vrele(fdp->fd_jdir);
498 			fdp->fd_jdir = NULL;
499 		}
500 	}
501 	if (p->p_vkernel)
502 		vkernel_exit(p);
503 	if (p->p_textvp) {
504 		vrele(p->p_textvp);
505 		p->p_textvp = NULL;
506 	}
507 	vm = p->p_vmspace;
508 	if (vm != NULL) {
509 		pmap_remove_pages(vmspace_pmap(vm),
510 				  VM_MIN_USER_ADDRESS,
511 				  VM_MAX_USER_ADDRESS);
512 		vm_map_remove(&vm->vm_map,
513 			      VM_MIN_USER_ADDRESS,
514 			      VM_MAX_USER_ADDRESS);
515 	}
516 }
517 
518 /*
519  * Magic number for savecore
520  *
521  * exported (symorder) and used at least by savecore(8)
522  *
523  */
524 static u_long const	dumpmag = 0x8fca0101UL;
525 
526 static int	dumpsize = 0;		/* also for savecore */
527 
528 static int	dodump = 1;
529 
530 SYSCTL_INT(_machdep, OID_AUTO, do_dump, CTLFLAG_RW, &dodump, 0,
531     "Try to perform coredump on kernel panic");
532 
533 static int
534 setdumpdev(dev)
535 	cdev_t dev;
536 {
537 	int psize;
538 	long newdumplo;
539 
540 	if (dev == NOCDEV) {
541 		dumpdev = dev;
542 		return (0);
543 	}
544 	psize = dev_dpsize(dev);
545 	if (psize == -1)
546 		return (ENXIO);
547 	/*
548 	 * XXX should clean up checking in dumpsys() to be more like this.
549 	 */
550 	newdumplo = psize - Maxmem * (PAGE_SIZE / DEV_BSIZE);
551 	if (newdumplo <= LABELSECTOR)
552 		return (ENOSPC);
553 	dumpdev = dev;
554 	dumplo = newdumplo;
555 	return (0);
556 }
557 
558 
559 /* ARGSUSED */
560 static void dump_conf (void *dummy);
561 static void
562 dump_conf(dummy)
563 	void *dummy;
564 {
565 	char *path;
566 	cdev_t dev;
567 
568 	path = kmalloc(MNAMELEN, M_TEMP, M_WAITOK);
569 	if (TUNABLE_STR_FETCH("dumpdev", path, MNAMELEN) != 0) {
570 		dev = kgetdiskbyname(path);
571 		if (dev != NOCDEV)
572 			dumpdev = dev;
573 	}
574 	kfree(path, M_TEMP);
575 	if (setdumpdev(dumpdev) != 0)
576 		dumpdev = NOCDEV;
577 }
578 
579 SYSINIT(dump_conf, SI_SUB_DUMP_CONF, SI_ORDER_FIRST, dump_conf, NULL)
580 
581 static int
582 sysctl_kern_dumpdev(SYSCTL_HANDLER_ARGS)
583 {
584 	int error;
585 	udev_t ndumpdev;
586 
587 	ndumpdev = dev2udev(dumpdev);
588 	error = sysctl_handle_opaque(oidp, &ndumpdev, sizeof ndumpdev, req);
589 	if (error == 0 && req->newptr != NULL)
590 		error = setdumpdev(udev2dev(ndumpdev, 0));
591 	return (error);
592 }
593 
594 SYSCTL_PROC(_kern, KERN_DUMPDEV, dumpdev, CTLTYPE_OPAQUE|CTLFLAG_RW,
595 	0, sizeof dumpdev, sysctl_kern_dumpdev, "T,udev_t", "");
596 
597 /*
598  * Doadump comes here after turning off memory management and
599  * getting on the dump stack, either when called above, or by
600  * the auto-restart code.
601  */
602 static void
603 dumpsys(void)
604 {
605 	int	error;
606 
607 	savectx(&dumppcb);
608 	dumpthread = curthread;
609 	if (dumping++) {
610 		printf("Dump already in progress, bailing...\n");
611 		return;
612 	}
613 	if (!dodump)
614 		return;
615 	if (dumpdev == NOCDEV)
616 		return;
617 	dumpsize = Maxmem;
618 	printf("\ndumping to dev %s, offset %ld\n", devtoname(dumpdev), dumplo);
619 	printf("dump ");
620 	error = dev_ddump(dumpdev);
621 	if (error == 0) {
622 		printf("succeeded\n");
623 		return;
624 	}
625 	printf("failed, reason: ");
626 	switch (error) {
627 	case ENOSYS:
628 	case ENODEV:
629 		printf("device doesn't support a dump routine\n");
630 		break;
631 
632 	case ENXIO:
633 		printf("device bad\n");
634 		break;
635 
636 	case EFAULT:
637 		printf("device not ready\n");
638 		break;
639 
640 	case EINVAL:
641 		printf("area improper\n");
642 		break;
643 
644 	case EIO:
645 		printf("i/o error\n");
646 		break;
647 
648 	case EINTR:
649 		printf("aborted from console\n");
650 		break;
651 
652 	default:
653 		printf("unknown, error = %d\n", error);
654 		break;
655 	}
656 }
657 
658 int
659 dumpstatus(vm_offset_t addr, off_t count)
660 {
661 	int c;
662 
663 	if (addr % (1024 * 1024) == 0) {
664 #ifdef HW_WDOG
665 		if (wdog_tickler)
666 			(*wdog_tickler)();
667 #endif
668 		printf("%ld ", (long)(count / (1024 * 1024)));
669 	}
670 
671 	if ((c = cncheckc()) == 0x03)
672 		return -1;
673 	else if (c != -1)
674 		printf("[CTRL-C to abort] ");
675 
676 	return 0;
677 }
678 
679 /*
680  * Panic is called on unresolvable fatal errors.  It prints "panic: mesg",
681  * and then reboots.  If we are called twice, then we avoid trying to sync
682  * the disks as this often leads to recursive panics.
683  */
684 void
685 panic(const char *fmt, ...)
686 {
687 	int bootopt, newpanic;
688 	__va_list ap;
689 	static char buf[256];
690 
691 #ifdef SMP
692 	/*
693 	 * If a panic occurs on multiple cpus before the first is able to
694 	 * halt the other cpus, only one cpu is allowed to take the panic.
695 	 * Attempt to be verbose about this situation but if the printf()
696 	 * itself panics don't let us overrun the kernel stack.
697 	 *
698 	 * Be very nasty about descheduling our thread at the lowest
699 	 * level possible in an attempt to freeze the thread without
700 	 * inducing further panics.
701 	 *
702 	 * Bumping gd_trap_nesting_level will also bypass assertions in
703 	 * lwkt_switch() and allow us to switch away even if we are a
704 	 * FAST interrupt or IPI.
705 	 */
706 	if (atomic_poll_acquire_int(&panic_cpu_interlock)) {
707 		panic_cpu_gd = mycpu;
708 	} else if (panic_cpu_gd != mycpu) {
709 		crit_enter();
710 		++mycpu->gd_trap_nesting_level;
711 		if (mycpu->gd_trap_nesting_level < 25) {
712 			printf("SECONDARY PANIC ON CPU %d THREAD %p\n",
713 				mycpu->gd_cpuid, curthread);
714 		}
715 		curthread->td_release = NULL;	/* be a grinch */
716 		for (;;) {
717 			lwkt_deschedule_self(curthread);
718 			lwkt_switch();
719 		}
720 		/* NOT REACHED */
721 		/* --mycpu->gd_trap_nesting_level */
722 		/* crit_exit() */
723 	}
724 #endif
725 	bootopt = RB_AUTOBOOT | RB_DUMP;
726 	if (sync_on_panic == 0)
727 		bootopt |= RB_NOSYNC;
728 	newpanic = 0;
729 	if (panicstr)
730 		bootopt |= RB_NOSYNC;
731 	else {
732 		panicstr = fmt;
733 		newpanic = 1;
734 	}
735 
736 	__va_start(ap, fmt);
737 	kvsnprintf(buf, sizeof(buf), fmt, ap);
738 	if (panicstr == fmt)
739 		panicstr = buf;
740 	__va_end(ap);
741 	printf("panic: %s\n", buf);
742 #ifdef SMP
743 	/* three separate prints in case of an unmapped page and trap */
744 	printf("mp_lock = %08x; ", mp_lock);
745 	printf("cpuid = %d; ", mycpu->gd_cpuid);
746 	printf("lapic.id = %08x\n", lapic.id);
747 #endif
748 
749 #if defined(DDB)
750 	if (newpanic && trace_on_panic)
751 		db_print_backtrace();
752 	if (debugger_on_panic)
753 		Debugger ("panic");
754 #endif
755 	boot(bootopt);
756 }
757 
758 /*
759  * Support for poweroff delay.
760  */
761 #ifndef POWEROFF_DELAY
762 # define POWEROFF_DELAY 5000
763 #endif
764 static int poweroff_delay = POWEROFF_DELAY;
765 
766 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW,
767 	&poweroff_delay, 0, "");
768 
769 static void
770 poweroff_wait(void *junk, int howto)
771 {
772 	if(!(howto & RB_POWEROFF) || poweroff_delay <= 0)
773 		return;
774 	DELAY(poweroff_delay * 1000);
775 }
776 
777 /*
778  * Some system processes (e.g. syncer) need to be stopped at appropriate
779  * points in their main loops prior to a system shutdown, so that they
780  * won't interfere with the shutdown process (e.g. by holding a disk buf
781  * to cause sync to fail).  For each of these system processes, register
782  * shutdown_kproc() as a handler for one of shutdown events.
783  */
784 static int kproc_shutdown_wait = 60;
785 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW,
786     &kproc_shutdown_wait, 0, "");
787 
788 void
789 shutdown_kproc(void *arg, int howto)
790 {
791 	struct thread *td;
792 	struct proc *p;
793 	int error;
794 
795 	if (panicstr)
796 		return;
797 
798 	td = (struct thread *)arg;
799 	if ((p = td->td_proc) != NULL) {
800 	    printf("Waiting (max %d seconds) for system process `%s' to stop...",
801 		kproc_shutdown_wait, p->p_comm);
802 	} else {
803 	    printf("Waiting (max %d seconds) for system thread %s to stop...",
804 		kproc_shutdown_wait, td->td_comm);
805 	}
806 	error = suspend_kproc(td, kproc_shutdown_wait * hz);
807 
808 	if (error == EWOULDBLOCK)
809 		printf("timed out\n");
810 	else
811 		printf("stopped\n");
812 }
813