xref: /dflybsd-src/sys/kern/kern_shutdown.c (revision ae788f37fe53d5d1ca1e12a184a662192caad3c5)
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 1.62 2008/01/05 13:23:48 corecode 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/diskslice.h>
54 #include <sys/reboot.h>
55 #include <sys/proc.h>
56 #include <sys/priv.h>
57 #include <sys/fcntl.h>		/* FREAD	*/
58 #include <sys/stat.h>		/* S_IFCHR	*/
59 #include <sys/vnode.h>
60 #include <sys/kernel.h>
61 #include <sys/kthread.h>
62 #include <sys/malloc.h>
63 #include <sys/mount.h>
64 #include <sys/queue.h>
65 #include <sys/sysctl.h>
66 #include <sys/vkernel.h>
67 #include <sys/conf.h>
68 #include <sys/sysproto.h>
69 #include <sys/device.h>
70 #include <sys/cons.h>
71 #include <sys/shm.h>
72 #include <sys/kern_syscall.h>
73 #include <vm/vm_map.h>
74 #include <vm/pmap.h>
75 
76 #include <sys/thread2.h>
77 #include <sys/buf2.h>
78 
79 #include <machine/pcb.h>
80 #include <machine/clock.h>
81 #include <machine/md_var.h>
82 #include <machine/smp.h>		/* smp_active_mask, cpuid */
83 #include <machine/vmparam.h>
84 
85 #include <sys/signalvar.h>
86 
87 #include <sys/wdog.h>
88 #include <dev/misc/gpio/gpio.h>
89 
90 #ifndef PANIC_REBOOT_WAIT_TIME
91 #define PANIC_REBOOT_WAIT_TIME 15 /* default to 15 seconds */
92 #endif
93 
94 /*
95  * Note that stdarg.h and the ANSI style va_start macro is used for both
96  * ANSI and traditional C compilers.  We use the machine version to stay
97  * within the confines of the kernel header files.
98  */
99 #include <machine/stdarg.h>
100 
101 #ifdef DDB
102 #include <ddb/ddb.h>
103 #ifdef DDB_UNATTENDED
104 int debugger_on_panic = 0;
105 #else
106 int debugger_on_panic = 1;
107 #endif
108 SYSCTL_INT(_debug, OID_AUTO, debugger_on_panic, CTLFLAG_RW,
109 	&debugger_on_panic, 0, "Run debugger on kernel panic");
110 
111 #ifdef DDB_TRACE
112 int trace_on_panic = 1;
113 #else
114 int trace_on_panic = 0;
115 #endif
116 SYSCTL_INT(_debug, OID_AUTO, trace_on_panic, CTLFLAG_RW,
117 	&trace_on_panic, 0, "Print stack trace on kernel panic");
118 #endif
119 
120 static int sync_on_panic = 0;
121 SYSCTL_INT(_kern, OID_AUTO, sync_on_panic, CTLFLAG_RW,
122 	&sync_on_panic, 0, "Do a sync before rebooting from a panic");
123 
124 SYSCTL_NODE(_kern, OID_AUTO, shutdown, CTLFLAG_RW, 0, "Shutdown environment");
125 
126 #ifdef	HW_WDOG
127 /*
128  * If there is a hardware watchdog, point this at the function needed to
129  * hold it off.
130  * It's needed when the kernel needs to do some lengthy operations.
131  * e.g. in wd.c when dumping core.. It's most annoying to have
132  * your precious core-dump only half written because the wdog kicked in.
133  */
134 watchdog_tickle_fn wdog_tickler = NULL;
135 #endif	/* HW_WDOG */
136 
137 /*
138  * Variable panicstr contains argument to first call to panic; used as flag
139  * to indicate that the kernel has already called panic.
140  */
141 const char *panicstr;
142 
143 int dumping;				/* system is dumping */
144 #ifdef SMP
145 u_int panic_cpu_interlock;		/* panic interlock */
146 globaldata_t panic_cpu_gd;		/* which cpu took the panic */
147 #endif
148 
149 int bootverbose = 0;			/* note: assignment to force non-bss */
150 SYSCTL_INT(_debug, OID_AUTO, bootverbose, CTLFLAG_RW,
151 	   &bootverbose, 0, "Verbose kernel messages");
152 
153 int cold = 1;				/* note: assignment to force non-bss */
154 int dumplo;				/* OBSOLETE - savecore compat */
155 u_int64_t dumplo64;
156 
157 static void boot (int) __dead2;
158 static void dumpsys (void);
159 static int setdumpdev (cdev_t dev);
160 static void poweroff_wait (void *, int);
161 static void print_uptime (void);
162 static void shutdown_halt (void *junk, int howto);
163 static void shutdown_panic (void *junk, int howto);
164 static void shutdown_reset (void *junk, int howto);
165 static int shutdown_busycount1(struct buf *bp, void *info);
166 static int shutdown_busycount2(struct buf *bp, void *info);
167 static void shutdown_cleanup_proc(struct proc *p);
168 
169 /* register various local shutdown events */
170 static void
171 shutdown_conf(void *unused)
172 {
173 	EVENTHANDLER_REGISTER(shutdown_final, poweroff_wait, NULL, SHUTDOWN_PRI_FIRST);
174 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_halt, NULL, SHUTDOWN_PRI_LAST + 100);
175 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_panic, NULL, SHUTDOWN_PRI_LAST + 100);
176 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_reset, NULL, SHUTDOWN_PRI_LAST + 200);
177 }
178 
179 SYSINIT(shutdown_conf, SI_BOOT2_MACHDEP, SI_ORDER_ANY, shutdown_conf, NULL)
180 
181 /* ARGSUSED */
182 
183 /*
184  * The system call that results in a reboot
185  */
186 int
187 sys_reboot(struct reboot_args *uap)
188 {
189 	struct thread *td = curthread;
190 	int error;
191 
192 	if ((error = priv_check(td, PRIV_REBOOT)))
193 		return (error);
194 
195 	boot(uap->opt);
196 	return (0);
197 }
198 
199 /*
200  * Called by events that want to shut down.. e.g  <CTL><ALT><DEL> on a PC
201  */
202 static int shutdown_howto = 0;
203 
204 void
205 shutdown_nice(int howto)
206 {
207 	shutdown_howto = howto;
208 
209 	/* Send a signal to init(8) and have it shutdown the world */
210 	if (initproc != NULL) {
211 		ksignal(initproc, SIGINT);
212 	} else {
213 		/* No init(8) running, so simply reboot */
214 		boot(RB_NOSYNC);
215 	}
216 	return;
217 }
218 static int	waittime = -1;
219 static struct thread *dumpthread;
220 static struct pcb dumppcb;
221 
222 static void
223 print_uptime(void)
224 {
225 	int f;
226 	struct timespec ts;
227 
228 	getnanouptime(&ts);
229 	kprintf("Uptime: ");
230 	f = 0;
231 	if (ts.tv_sec >= 86400) {
232 		kprintf("%ldd", ts.tv_sec / 86400);
233 		ts.tv_sec %= 86400;
234 		f = 1;
235 	}
236 	if (f || ts.tv_sec >= 3600) {
237 		kprintf("%ldh", ts.tv_sec / 3600);
238 		ts.tv_sec %= 3600;
239 		f = 1;
240 	}
241 	if (f || ts.tv_sec >= 60) {
242 		kprintf("%ldm", ts.tv_sec / 60);
243 		ts.tv_sec %= 60;
244 		f = 1;
245 	}
246 	kprintf("%lds\n", ts.tv_sec);
247 }
248 
249 /*
250  *  Go through the rigmarole of shutting down..
251  * this used to be in machdep.c but I'll be dammned if I could see
252  * anything machine dependant in it.
253  */
254 static void
255 boot(int howto)
256 {
257 	/*
258 	 * Get rid of any user scheduler baggage and then give
259 	 * us a high priority.
260 	 */
261 	if (curthread->td_release)
262 		curthread->td_release(curthread);
263 	lwkt_setpri_self(TDPRI_MAX);
264 
265 	/* collect extra flags that shutdown_nice might have set */
266 	howto |= shutdown_howto;
267 
268 #ifdef SMP
269 	/*
270 	 * We really want to shutdown on the BSP.  Subsystems such as ACPI
271 	 * can't power-down the box otherwise.
272 	 */
273 	if (smp_active_mask > 1) {
274 		kprintf("boot() called on cpu#%d\n", mycpu->gd_cpuid);
275 	}
276 	if (panicstr == NULL && mycpu->gd_cpuid != 0) {
277 		kprintf("Switching to cpu #0 for shutdown\n");
278 		lwkt_setcpu_self(globaldata_find(0));
279 	}
280 #endif
281 	/*
282 	 * Do any callouts that should be done BEFORE syncing the filesystems.
283 	 */
284 	EVENTHANDLER_INVOKE(shutdown_pre_sync, howto);
285 
286 	/*
287 	 * Try to get rid of any remaining FS references.  The calling
288 	 * process, proc0, and init may still hold references.  The
289 	 * VFS cache subsystem may still hold a root reference to root.
290 	 *
291 	 * XXX this needs work.  We really need to SIGSTOP all remaining
292 	 * processes in order to avoid blowups due to proc0's filesystem
293 	 * references going away.  For now just make sure that the init
294 	 * process is stopped.
295 	 */
296 	if (panicstr == NULL) {
297 		shutdown_cleanup_proc(curproc);
298 		shutdown_cleanup_proc(&proc0);
299 		if (initproc) {
300 			if (initproc != curproc) {
301 				ksignal(initproc, SIGSTOP);
302 				tsleep(boot, 0, "shutdn", hz / 20);
303 			}
304 			shutdown_cleanup_proc(initproc);
305 		}
306 		vfs_cache_setroot(NULL, NULL);
307 	}
308 
309 	/*
310 	 * Now sync filesystems
311 	 */
312 	if (!cold && (howto & RB_NOSYNC) == 0 && waittime < 0) {
313 		int iter, nbusy, pbusy;
314 
315 		waittime = 0;
316 		kprintf("\nsyncing disks... ");
317 
318 		sys_sync(NULL);	/* YYY was sync(&proc0, NULL). why proc0 ? */
319 
320 		/*
321 		 * With soft updates, some buffers that are
322 		 * written will be remarked as dirty until other
323 		 * buffers are written.
324 		 */
325 		for (iter = pbusy = 0; iter < 20; iter++) {
326 			nbusy = scan_all_buffers(shutdown_busycount1, NULL);
327 			if (nbusy == 0)
328 				break;
329 			kprintf("%d ", nbusy);
330 			if (nbusy < pbusy)
331 				iter = 0;
332 			pbusy = nbusy;
333 			/*
334 			 * XXX:
335 			 * Process soft update work queue if buffers don't sync
336 			 * after 6 iterations by permitting the syncer to run.
337 			 */
338 			if (iter > 5)
339 				bio_ops_sync(NULL);
340 
341 			sys_sync(NULL); /* YYY was sync(&proc0, NULL). why proc0 ? */
342 			tsleep(boot, 0, "shutdn", hz * iter / 20 + 1);
343 		}
344 		kprintf("\n");
345 		/*
346 		 * Count only busy local buffers to prevent forcing
347 		 * a fsck if we're just a client of a wedged NFS server
348 		 */
349 		nbusy = scan_all_buffers(shutdown_busycount2, NULL);
350 		if (nbusy) {
351 			/*
352 			 * Failed to sync all blocks. Indicate this and don't
353 			 * unmount filesystems (thus forcing an fsck on reboot).
354 			 */
355 			kprintf("giving up on %d buffers\n", nbusy);
356 #ifdef DDB
357 			Debugger("busy buffer problem");
358 #endif /* DDB */
359 			tsleep(boot, 0, "shutdn", hz * 5 + 1);
360 		} else {
361 			kprintf("done\n");
362 			/*
363 			 * Unmount filesystems
364 			 */
365 			if (panicstr == NULL)
366 				vfs_unmountall();
367 		}
368 		tsleep(boot, 0, "shutdn", hz / 10 + 1);
369 	}
370 
371 	print_uptime();
372 
373 	/*
374 	 * Dump before doing post_sync shutdown ops
375 	 */
376 	crit_enter();
377 	if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP && !cold)
378 		dumpsys();
379 
380 	/*
381 	 * Ok, now do things that assume all filesystem activity has
382 	 * been completed.  This will also call the device shutdown
383 	 * methods.
384 	 */
385 	EVENTHANDLER_INVOKE(shutdown_post_sync, howto);
386 
387 	/* Now that we're going to really halt the system... */
388 	EVENTHANDLER_INVOKE(shutdown_final, howto);
389 
390 	for(;;) ;	/* safety against shutdown_reset not working */
391 	/* NOTREACHED */
392 }
393 
394 static int
395 shutdown_busycount1(struct buf *bp, void *info)
396 {
397 	if ((bp->b_flags & B_INVAL) == 0 && BUF_REFCNT(bp) > 0)
398 		return(1);
399 	if ((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI)
400 		return (1);
401 	return (0);
402 }
403 
404 static int
405 shutdown_busycount2(struct buf *bp, void *info)
406 {
407 	if (((bp->b_flags & B_INVAL) == 0 && BUF_REFCNT(bp)) ||
408 	    ((bp->b_flags & (B_DELWRI|B_INVAL)) == B_DELWRI)) {
409 		/*
410 		 * Only count buffers undergoing write I/O
411 		 * on the related vnode.
412 		 */
413 		if (bp->b_vp == NULL ||
414 		    bio_track_active(&bp->b_vp->v_track_write) == 0) {
415 			return (0);
416 		}
417 #if defined(SHOW_BUSYBUFS) || defined(DIAGNOSTIC)
418 		kprintf(
419 	    "%p dev:?, flags:%08x, loffset:%lld, doffset:%lld\n",
420 		    bp,
421 		    bp->b_flags, bp->b_loffset,
422 		    bp->b_bio2.bio_offset);
423 #endif
424 		return(1);
425 	}
426 	return(0);
427 }
428 
429 /*
430  * If the shutdown was a clean halt, behave accordingly.
431  */
432 static void
433 shutdown_halt(void *junk, int howto)
434 {
435 	if (howto & RB_HALT) {
436 		kprintf("\n");
437 		kprintf("The operating system has halted.\n");
438 #ifdef _KERNEL_VIRTUAL
439 		cpu_halt();
440 #else
441 		kprintf("Please press any key to reboot.\n\n");
442 		switch (cngetc()) {
443 		case -1:		/* No console, just die */
444 			cpu_halt();
445 			/* NOTREACHED */
446 		default:
447 			howto &= ~RB_HALT;
448 			break;
449 		}
450 #endif
451 	}
452 }
453 
454 /*
455  * Check to see if the system paniced, pause and then reboot
456  * according to the specified delay.
457  */
458 static void
459 shutdown_panic(void *junk, int howto)
460 {
461 	int loop;
462 
463 	if (howto & RB_DUMP) {
464 		if (PANIC_REBOOT_WAIT_TIME != 0) {
465 			if (PANIC_REBOOT_WAIT_TIME != -1) {
466 				kprintf("Automatic reboot in %d seconds - "
467 				       "press a key on the console to abort\n",
468 					PANIC_REBOOT_WAIT_TIME);
469 				for (loop = PANIC_REBOOT_WAIT_TIME * 10;
470 				     loop > 0; --loop) {
471 					DELAY(1000 * 100); /* 1/10th second */
472 					/* Did user type a key? */
473 					if (cncheckc() != -1)
474 						break;
475 				}
476 				if (!loop)
477 					return;
478 			}
479 		} else { /* zero time specified - reboot NOW */
480 			return;
481 		}
482 		kprintf("--> Press a key on the console to reboot,\n");
483 		kprintf("--> or switch off the system now.\n");
484 		cngetc();
485 	}
486 }
487 
488 /*
489  * Everything done, now reset
490  */
491 static void
492 shutdown_reset(void *junk, int howto)
493 {
494 	kprintf("Rebooting...\n");
495 	DELAY(1000000);	/* wait 1 sec for kprintf's to complete and be read */
496 	/* cpu_boot(howto); */ /* doesn't do anything at the moment */
497 	cpu_reset();
498 	/* NOTREACHED */ /* assuming reset worked */
499 }
500 
501 /*
502  * Try to remove FS references in the specified process.  This function
503  * is used during shutdown
504  */
505 static
506 void
507 shutdown_cleanup_proc(struct proc *p)
508 {
509 	struct filedesc *fdp;
510 	struct vmspace *vm;
511 
512 	if (p == NULL)
513 		return;
514 	if ((fdp = p->p_fd) != NULL) {
515 		kern_closefrom(0);
516 		if (fdp->fd_cdir) {
517 			cache_drop(&fdp->fd_ncdir);
518 			vrele(fdp->fd_cdir);
519 			fdp->fd_cdir = NULL;
520 		}
521 		if (fdp->fd_rdir) {
522 			cache_drop(&fdp->fd_nrdir);
523 			vrele(fdp->fd_rdir);
524 			fdp->fd_rdir = NULL;
525 		}
526 		if (fdp->fd_jdir) {
527 			cache_drop(&fdp->fd_njdir);
528 			vrele(fdp->fd_jdir);
529 			fdp->fd_jdir = NULL;
530 		}
531 	}
532 	if (p->p_vkernel)
533 		vkernel_exit(p);
534 	if (p->p_textvp) {
535 		vrele(p->p_textvp);
536 		p->p_textvp = NULL;
537 	}
538 	vm = p->p_vmspace;
539 	if (vm != NULL) {
540 		pmap_remove_pages(vmspace_pmap(vm),
541 				  VM_MIN_USER_ADDRESS,
542 				  VM_MAX_USER_ADDRESS);
543 		vm_map_remove(&vm->vm_map,
544 			      VM_MIN_USER_ADDRESS,
545 			      VM_MAX_USER_ADDRESS);
546 	}
547 }
548 
549 /*
550  * Magic number for savecore
551  *
552  * exported (symorder) and used at least by savecore(8)
553  *
554  * Mark it as used so that gcc doesn't optimize it away.
555  */
556 __attribute__((__used__))
557 	static u_long const dumpmag = 0x8fca0101UL;
558 
559 static int	dumpsize = 0;		/* also for savecore */
560 
561 static int	dodump = 1;
562 
563 SYSCTL_INT(_machdep, OID_AUTO, do_dump, CTLFLAG_RW, &dodump, 0,
564     "Try to perform coredump on kernel panic");
565 
566 static int
567 setdumpdev(cdev_t dev)
568 {
569 	struct partinfo pinfo;
570 	u_int64_t newdumplo;
571 	int error;
572 	int doopen;
573 
574 	if (dev == NULL) {
575 		dumpdev = dev;
576 		return (0);
577 	}
578 	bzero(&pinfo, sizeof(pinfo));
579 
580 	/*
581 	 * We have to open the device before we can perform ioctls on it,
582 	 * or the slice/label data may not be present.  Device opens are
583 	 * usually tracked by specfs, but the dump device can be set in
584 	 * early boot and may not be open so this is somewhat of a hack.
585 	 */
586 	doopen = (dev->si_sysref.refcnt == 1);
587 	if (doopen) {
588 		error = dev_dopen(dev, FREAD, S_IFCHR, proc0.p_ucred);
589 		if (error)
590 			return (error);
591 	}
592 	error = dev_dioctl(dev, DIOCGPART, (void *)&pinfo, 0,
593 			   proc0.p_ucred, NULL);
594 	if (doopen)
595 		dev_dclose(dev, FREAD, S_IFCHR);
596 	if (error || pinfo.media_blocks == 0 || pinfo.media_blksize == 0)
597 		return (ENXIO);
598 
599 	newdumplo = pinfo.media_blocks -
600 		    ((u_int64_t)Maxmem * PAGE_SIZE / DEV_BSIZE);
601 	if ((int64_t)newdumplo < (int64_t)pinfo.reserved_blocks)
602 		return (ENOSPC);
603 	dumpdev = dev;
604 	dumplo64 = newdumplo;
605 	return (0);
606 }
607 
608 
609 /* ARGSUSED */
610 static void dump_conf (void *dummy);
611 static void
612 dump_conf(void *dummy)
613 {
614 	char *path;
615 	cdev_t dev;
616 
617 	path = kmalloc(MNAMELEN, M_TEMP, M_WAITOK);
618 	if (TUNABLE_STR_FETCH("dumpdev", path, MNAMELEN) != 0) {
619 		dev = kgetdiskbyname(path);
620 		if (dev != NULL)
621 			dumpdev = dev;
622 	}
623 	kfree(path, M_TEMP);
624 	if (setdumpdev(dumpdev) != 0)
625 		dumpdev = NULL;
626 }
627 
628 SYSINIT(dump_conf, SI_SUB_DUMP_CONF, SI_ORDER_FIRST, dump_conf, NULL)
629 
630 static int
631 sysctl_kern_dumpdev(SYSCTL_HANDLER_ARGS)
632 {
633 	int error;
634 	udev_t ndumpdev;
635 
636 	ndumpdev = dev2udev(dumpdev);
637 	error = sysctl_handle_opaque(oidp, &ndumpdev, sizeof ndumpdev, req);
638 	if (error == 0 && req->newptr != NULL)
639 		error = setdumpdev(udev2dev(ndumpdev, 0));
640 	return (error);
641 }
642 
643 SYSCTL_PROC(_kern, KERN_DUMPDEV, dumpdev, CTLTYPE_OPAQUE|CTLFLAG_RW,
644 	0, sizeof dumpdev, sysctl_kern_dumpdev, "T,udev_t", "");
645 
646 /*
647  * Doadump comes here after turning off memory management and
648  * getting on the dump stack, either when called above, or by
649  * the auto-restart code.
650  */
651 static void
652 dumpsys(void)
653 {
654 	int	error;
655 
656 	savectx(&dumppcb);
657 	dumpthread = curthread;
658 	if (dumping++) {
659 		kprintf("Dump already in progress, bailing...\n");
660 		return;
661 	}
662 	if (!dodump)
663 		return;
664 	if (dumpdev == NULL)
665 		return;
666 	dumpsize = Maxmem;
667 	kprintf("\ndumping to dev %s, blockno %lld\n",
668 		devtoname(dumpdev),
669 		(long long)dumplo64);
670 	kprintf("dump ");
671 	error = dev_ddump(dumpdev);
672 	if (error == 0) {
673 		kprintf("succeeded\n");
674 		return;
675 	}
676 	kprintf("failed, reason: ");
677 	switch (error) {
678 	case ENOSYS:
679 	case ENODEV:
680 		kprintf("device doesn't support a dump routine\n");
681 		break;
682 
683 	case ENXIO:
684 		kprintf("device bad\n");
685 		break;
686 
687 	case EFAULT:
688 		kprintf("device not ready\n");
689 		break;
690 
691 	case EINVAL:
692 		kprintf("area improper\n");
693 		break;
694 
695 	case EIO:
696 		kprintf("i/o error\n");
697 		break;
698 
699 	case EINTR:
700 		kprintf("aborted from console\n");
701 		break;
702 
703 	default:
704 		kprintf("unknown, error = %d\n", error);
705 		break;
706 	}
707 }
708 
709 int
710 dumpstatus(vm_offset_t addr, off_t count)
711 {
712 	int c;
713 
714 	if (addr % (1024 * 1024) == 0) {
715 #ifdef HW_WDOG
716 		if (wdog_tickler)
717 			(*wdog_tickler)();
718 #endif
719 		kprintf("%ld ", (long)(count / (1024 * 1024)));
720 	}
721 
722 	if ((c = cncheckc()) == 0x03)
723 		return -1;
724 	else if (c != -1)
725 		kprintf("[CTRL-C to abort] ");
726 
727 	return 0;
728 }
729 
730 /*
731  * Panic is called on unresolvable fatal errors.  It prints "panic: mesg",
732  * and then reboots.  If we are called twice, then we avoid trying to sync
733  * the disks as this often leads to recursive panics.
734  */
735 void
736 panic(const char *fmt, ...)
737 {
738 	int bootopt, newpanic;
739 	__va_list ap;
740 	static char buf[256];
741 
742 #ifdef SMP
743 	/*
744 	 * If a panic occurs on multiple cpus before the first is able to
745 	 * halt the other cpus, only one cpu is allowed to take the panic.
746 	 * Attempt to be verbose about this situation but if the kprintf()
747 	 * itself panics don't let us overrun the kernel stack.
748 	 *
749 	 * Be very nasty about descheduling our thread at the lowest
750 	 * level possible in an attempt to freeze the thread without
751 	 * inducing further panics.
752 	 *
753 	 * Bumping gd_trap_nesting_level will also bypass assertions in
754 	 * lwkt_switch() and allow us to switch away even if we are a
755 	 * FAST interrupt or IPI.
756 	 */
757 	if (atomic_poll_acquire_int(&panic_cpu_interlock)) {
758 		panic_cpu_gd = mycpu;
759 	} else if (panic_cpu_gd != mycpu) {
760 		crit_enter();
761 		++mycpu->gd_trap_nesting_level;
762 		if (mycpu->gd_trap_nesting_level < 25) {
763 			kprintf("SECONDARY PANIC ON CPU %d THREAD %p\n",
764 				mycpu->gd_cpuid, curthread);
765 		}
766 		curthread->td_release = NULL;	/* be a grinch */
767 		for (;;) {
768 			lwkt_deschedule_self(curthread);
769 			lwkt_switch();
770 		}
771 		/* NOT REACHED */
772 		/* --mycpu->gd_trap_nesting_level */
773 		/* crit_exit() */
774 	}
775 #endif
776 	bootopt = RB_AUTOBOOT | RB_DUMP;
777 	if (sync_on_panic == 0)
778 		bootopt |= RB_NOSYNC;
779 	newpanic = 0;
780 	if (panicstr)
781 		bootopt |= RB_NOSYNC;
782 	else {
783 		panicstr = fmt;
784 		newpanic = 1;
785 	}
786 
787 	__va_start(ap, fmt);
788 	kvsnprintf(buf, sizeof(buf), fmt, ap);
789 	if (panicstr == fmt)
790 		panicstr = buf;
791 	__va_end(ap);
792 	kprintf("panic: %s\n", buf);
793 #ifdef SMP
794 	/* two separate prints in case of an unmapped page and trap */
795 	kprintf("mp_lock = %08x; ", mp_lock);
796 	kprintf("cpuid = %d\n", mycpu->gd_cpuid);
797 #endif
798 
799 #if (NGPIO > 0) && defined(ERROR_LED_ON_PANIC)
800 	led_switch("error", 1);
801 #endif
802 
803 #if defined(WDOG_DISABLE_ON_PANIC) && defined(WATCHDOG_ENABLE)
804 	wdog_disable();
805 #endif
806 
807 #if defined(DDB)
808 	if (newpanic && trace_on_panic)
809 		print_backtrace();
810 	if (debugger_on_panic)
811 		Debugger("panic");
812 #endif
813 	boot(bootopt);
814 }
815 
816 /*
817  * Support for poweroff delay.
818  */
819 #ifndef POWEROFF_DELAY
820 # define POWEROFF_DELAY 5000
821 #endif
822 static int poweroff_delay = POWEROFF_DELAY;
823 
824 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW,
825 	&poweroff_delay, 0, "");
826 
827 static void
828 poweroff_wait(void *junk, int howto)
829 {
830 	if(!(howto & RB_POWEROFF) || poweroff_delay <= 0)
831 		return;
832 	DELAY(poweroff_delay * 1000);
833 }
834 
835 /*
836  * Some system processes (e.g. syncer) need to be stopped at appropriate
837  * points in their main loops prior to a system shutdown, so that they
838  * won't interfere with the shutdown process (e.g. by holding a disk buf
839  * to cause sync to fail).  For each of these system processes, register
840  * shutdown_kproc() as a handler for one of shutdown events.
841  */
842 static int kproc_shutdown_wait = 60;
843 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW,
844     &kproc_shutdown_wait, 0, "");
845 
846 void
847 shutdown_kproc(void *arg, int howto)
848 {
849 	struct thread *td;
850 	struct proc *p;
851 	int error;
852 
853 	if (panicstr)
854 		return;
855 
856 	td = (struct thread *)arg;
857 	if ((p = td->td_proc) != NULL) {
858 	    kprintf("Waiting (max %d seconds) for system process `%s' to stop...",
859 		kproc_shutdown_wait, p->p_comm);
860 	} else {
861 	    kprintf("Waiting (max %d seconds) for system thread %s to stop...",
862 		kproc_shutdown_wait, td->td_comm);
863 	}
864 	error = suspend_kproc(td, kproc_shutdown_wait * hz);
865 
866 	if (error == EWOULDBLOCK)
867 		kprintf("timed out\n");
868 	else
869 		kprintf("stopped\n");
870 }
871