xref: /dflybsd-src/sys/kern/kern_shutdown.c (revision 07caec20a93f40352fd28eec40bb9b40199edf40)
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 1.38 2006/09/24 19:43:55 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 static void boot (int) __dead2;
144 static void dumpsys (void);
145 static int setdumpdev (cdev_t dev);
146 static void poweroff_wait (void *, int);
147 static void print_uptime (void);
148 static void shutdown_halt (void *junk, int howto);
149 static void shutdown_panic (void *junk, int howto);
150 static void shutdown_reset (void *junk, int howto);
151 static int shutdown_busycount1(struct buf *bp, void *info);
152 static int shutdown_busycount2(struct buf *bp, void *info);
153 static void shutdown_cleanup_proc(struct proc *p);
154 
155 /* register various local shutdown events */
156 static void
157 shutdown_conf(void *unused)
158 {
159 	EVENTHANDLER_REGISTER(shutdown_final, poweroff_wait, NULL, SHUTDOWN_PRI_FIRST);
160 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_halt, NULL, SHUTDOWN_PRI_LAST + 100);
161 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_panic, NULL, SHUTDOWN_PRI_LAST + 100);
162 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_reset, NULL, SHUTDOWN_PRI_LAST + 200);
163 }
164 
165 SYSINIT(shutdown_conf, SI_SUB_INTRINSIC, SI_ORDER_ANY, shutdown_conf, NULL)
166 
167 /* ARGSUSED */
168 
169 /*
170  * The system call that results in a reboot
171  */
172 int
173 sys_reboot(struct reboot_args *uap)
174 {
175 	struct thread *td = curthread;
176 	int error;
177 
178 	if ((error = suser(td)))
179 		return (error);
180 
181 	boot(uap->opt);
182 	return (0);
183 }
184 
185 /*
186  * Called by events that want to shut down.. e.g  <CTL><ALT><DEL> on a PC
187  */
188 static int shutdown_howto = 0;
189 
190 void
191 shutdown_nice(int howto)
192 {
193 	shutdown_howto = howto;
194 
195 	/* Send a signal to init(8) and have it shutdown the world */
196 	if (initproc != NULL) {
197 		ksignal(initproc, SIGINT);
198 	} else {
199 		/* No init(8) running, so simply reboot */
200 		boot(RB_NOSYNC);
201 	}
202 	return;
203 }
204 static int	waittime = -1;
205 static struct thread *dumpthread;
206 static struct pcb dumppcb;
207 
208 static void
209 print_uptime()
210 {
211 	int f;
212 	struct timespec ts;
213 
214 	getnanouptime(&ts);
215 	printf("Uptime: ");
216 	f = 0;
217 	if (ts.tv_sec >= 86400) {
218 		printf("%ldd", ts.tv_sec / 86400);
219 		ts.tv_sec %= 86400;
220 		f = 1;
221 	}
222 	if (f || ts.tv_sec >= 3600) {
223 		printf("%ldh", ts.tv_sec / 3600);
224 		ts.tv_sec %= 3600;
225 		f = 1;
226 	}
227 	if (f || ts.tv_sec >= 60) {
228 		printf("%ldm", ts.tv_sec / 60);
229 		ts.tv_sec %= 60;
230 		f = 1;
231 	}
232 	printf("%lds\n", ts.tv_sec);
233 }
234 
235 /*
236  *  Go through the rigmarole of shutting down..
237  * this used to be in machdep.c but I'll be dammned if I could see
238  * anything machine dependant in it.
239  */
240 static void
241 boot(int howto)
242 {
243 	/*
244 	 * Get rid of any user scheduler baggage and then give
245 	 * us a high priority.
246 	 */
247 	if (curthread->td_release)
248 		curthread->td_release(curthread);
249 	lwkt_setpri_self(TDPRI_MAX);
250 
251 	/* collect extra flags that shutdown_nice might have set */
252 	howto |= shutdown_howto;
253 
254 #ifdef SMP
255 	/*
256 	 * We really want to shutdown on the BSP.  Subsystems such as ACPI
257 	 * can't power-down the box otherwise.
258 	 */
259 	if (smp_active_mask > 1) {
260 		printf("boot() called on cpu#%d\n", mycpu->gd_cpuid);
261 	}
262 	if (panicstr == NULL && mycpu->gd_cpuid != 0) {
263 		printf("Switching to cpu #0 for shutdown\n");
264 		lwkt_setcpu_self(globaldata_find(0));
265 	}
266 #endif
267 	/*
268 	 * Do any callouts that should be done BEFORE syncing the filesystems.
269 	 */
270 	EVENTHANDLER_INVOKE(shutdown_pre_sync, howto);
271 
272 	/*
273 	 * Try to get rid of any remaining FS references.  The calling
274 	 * process, proc0, and init may still hold references.  The
275 	 * VFS cache subsystem may still hold a root reference to root.
276 	 */
277 	if (panicstr == NULL) {
278 		shutdown_cleanup_proc(curproc);
279 		shutdown_cleanup_proc(&proc0);
280 		if (initproc)
281 			shutdown_cleanup_proc(initproc);
282 		vfs_cache_setroot(NULL, NULL);
283 	}
284 
285 	/*
286 	 * Now sync filesystems
287 	 */
288 	if (!cold && (howto & RB_NOSYNC) == 0 && waittime < 0) {
289 		int iter, nbusy, pbusy;
290 
291 		waittime = 0;
292 		printf("\nsyncing disks... ");
293 
294 		sys_sync(NULL);	/* YYY was sync(&proc0, NULL). why proc0 ? */
295 
296 		/*
297 		 * With soft updates, some buffers that are
298 		 * written will be remarked as dirty until other
299 		 * buffers are written.
300 		 */
301 		for (iter = pbusy = 0; iter < 20; iter++) {
302 			nbusy = scan_all_buffers(shutdown_busycount1, NULL);
303 			if (nbusy == 0)
304 				break;
305 			printf("%d ", nbusy);
306 			if (nbusy < pbusy)
307 				iter = 0;
308 			pbusy = nbusy;
309 			/*
310 			 * XXX:
311 			 * Process soft update work queue if buffers don't sync
312 			 * after 6 iterations by permitting the syncer to run.
313 			 */
314 			if (iter > 5 && bioops.io_sync)
315 				(*bioops.io_sync)(NULL);
316 			sys_sync(NULL); /* YYY was sync(&proc0, NULL). why proc0 ? */
317 			tsleep(boot, 0, "shutdn", hz * iter / 20 + 1);
318 		}
319 		printf("\n");
320 		/*
321 		 * Count only busy local buffers to prevent forcing
322 		 * a fsck if we're just a client of a wedged NFS server
323 		 */
324 		nbusy = scan_all_buffers(shutdown_busycount2, NULL);
325 		if (nbusy) {
326 			/*
327 			 * Failed to sync all blocks. Indicate this and don't
328 			 * unmount filesystems (thus forcing an fsck on reboot).
329 			 */
330 			printf("giving up on %d buffers\n", nbusy);
331 #ifdef DDB
332 			Debugger("busy buffer problem");
333 #endif /* DDB */
334 			tsleep(boot, 0, "shutdn", hz * 5 + 1);
335 		} else {
336 			printf("done\n");
337 			/*
338 			 * Unmount filesystems
339 			 */
340 			if (panicstr == NULL)
341 				vfs_unmountall();
342 		}
343 		tsleep(boot, 0, "shutdn", hz / 10 + 1);
344 	}
345 
346 	print_uptime();
347 
348 	/*
349 	 * Ok, now do things that assume all filesystem activity has
350 	 * been completed.
351 	 */
352 	EVENTHANDLER_INVOKE(shutdown_post_sync, howto);
353 	crit_enter();
354 	if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP && !cold)
355 		dumpsys();
356 
357 	/* Now that we're going to really halt the system... */
358 	EVENTHANDLER_INVOKE(shutdown_final, howto);
359 
360 	for(;;) ;	/* safety against shutdown_reset not working */
361 	/* NOTREACHED */
362 }
363 
364 static int
365 shutdown_busycount1(struct buf *bp, void *info)
366 {
367 	if ((bp->b_flags & B_INVAL) == 0 && BUF_REFCNT(bp) > 0)
368 		return(1);
369 	if ((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI)
370 		return (1);
371 	return (0);
372 }
373 
374 static int
375 shutdown_busycount2(struct buf *bp, void *info)
376 {
377 	if (((bp->b_flags & B_INVAL) == 0 && BUF_REFCNT(bp)) ||
378 	    ((bp->b_flags & (B_DELWRI|B_INVAL)) == B_DELWRI)) {
379 		/*
380 		 * Only count buffers undergoing write I/O
381 		 * on the related vnode.
382 		 */
383 		if (bp->b_vp == NULL ||
384 		    bp->b_vp->v_track_write.bk_active == 0) {
385 			return (0);
386 		}
387 #if defined(SHOW_BUSYBUFS) || defined(DIAGNOSTIC)
388 		printf(
389 	    "%p dev:?, flags:%08x, loffset:%lld, doffset:%lld\n",
390 		    bp,
391 		    bp->b_flags, bp->b_loffset,
392 		    bp->b_bio2.bio_offset);
393 #endif
394 		return(1);
395 	}
396 	return(0);
397 }
398 
399 /*
400  * If the shutdown was a clean halt, behave accordingly.
401  */
402 static void
403 shutdown_halt(void *junk, int howto)
404 {
405 	if (howto & RB_HALT) {
406 		printf("\n");
407 		printf("The operating system has halted.\n");
408 		printf("Please press any key to reboot.\n\n");
409 		switch (cngetc()) {
410 		case -1:		/* No console, just die */
411 			cpu_halt();
412 			/* NOTREACHED */
413 		default:
414 			howto &= ~RB_HALT;
415 			break;
416 		}
417 	}
418 }
419 
420 /*
421  * Check to see if the system paniced, pause and then reboot
422  * according to the specified delay.
423  */
424 static void
425 shutdown_panic(void *junk, int howto)
426 {
427 	int loop;
428 
429 	if (howto & RB_DUMP) {
430 		if (PANIC_REBOOT_WAIT_TIME != 0) {
431 			if (PANIC_REBOOT_WAIT_TIME != -1) {
432 				printf("Automatic reboot in %d seconds - "
433 				       "press a key on the console to abort\n",
434 					PANIC_REBOOT_WAIT_TIME);
435 				for (loop = PANIC_REBOOT_WAIT_TIME * 10;
436 				     loop > 0; --loop) {
437 					DELAY(1000 * 100); /* 1/10th second */
438 					/* Did user type a key? */
439 					if (cncheckc() != -1)
440 						break;
441 				}
442 				if (!loop)
443 					return;
444 			}
445 		} else { /* zero time specified - reboot NOW */
446 			return;
447 		}
448 		printf("--> Press a key on the console to reboot,\n");
449 		printf("--> or switch off the system now.\n");
450 		cngetc();
451 	}
452 }
453 
454 /*
455  * Everything done, now reset
456  */
457 static void
458 shutdown_reset(void *junk, int howto)
459 {
460 	printf("Rebooting...\n");
461 	DELAY(1000000);	/* wait 1 sec for printf's to complete and be read */
462 	/* cpu_boot(howto); */ /* doesn't do anything at the moment */
463 	cpu_reset();
464 	/* NOTREACHED */ /* assuming reset worked */
465 }
466 
467 /*
468  * Try to remove FS references in the specified process.  This function
469  * is used during shutdown
470  */
471 static
472 void
473 shutdown_cleanup_proc(struct proc *p)
474 {
475 	struct filedesc *fdp;
476 	struct vmspace *vm;
477 
478 	if (p == NULL)
479 		return;
480 	if ((fdp = p->p_fd) != NULL) {
481 		kern_closefrom(0);
482 		if (fdp->fd_cdir) {
483 			cache_drop(fdp->fd_ncdir);
484 			vrele(fdp->fd_cdir);
485 			fdp->fd_cdir = NULL;
486 			fdp->fd_ncdir = NULL;
487 		}
488 		if (fdp->fd_rdir) {
489 			cache_drop(fdp->fd_nrdir);
490 			vrele(fdp->fd_rdir);
491 			fdp->fd_rdir = NULL;
492 			fdp->fd_nrdir = NULL;
493 		}
494 		if (fdp->fd_jdir) {
495 			cache_drop(fdp->fd_njdir);
496 			vrele(fdp->fd_jdir);
497 			fdp->fd_jdir = NULL;
498 			fdp->fd_njdir = NULL;
499 		}
500 	}
501 	if (p->p_vkernel) {
502 		vkernel_drop(p->p_vkernel);
503 		p->p_vkernel = NULL;
504 	}
505 	if (p->p_textvp) {
506 		vrele(p->p_textvp);
507 		p->p_textvp = NULL;
508 	}
509 	vm = p->p_vmspace;
510 	if (vm != NULL) {
511 		pmap_remove_pages(vmspace_pmap(vm),
512 				  VM_MIN_ADDRESS,
513 				  VM_MAXUSER_ADDRESS);
514 		vm_map_remove(&vm->vm_map,
515 			      VM_MIN_ADDRESS,
516 			      VM_MAXUSER_ADDRESS);
517 	}
518 }
519 
520 /*
521  * Magic number for savecore
522  *
523  * exported (symorder) and used at least by savecore(8)
524  *
525  */
526 static u_long const	dumpmag = 0x8fca0101UL;
527 
528 static int	dumpsize = 0;		/* also for savecore */
529 
530 static int	dodump = 1;
531 
532 SYSCTL_INT(_machdep, OID_AUTO, do_dump, CTLFLAG_RW, &dodump, 0,
533     "Try to perform coredump on kernel panic");
534 
535 static int
536 setdumpdev(dev)
537 	cdev_t dev;
538 {
539 	int psize;
540 	long newdumplo;
541 
542 	if (dev == NOCDEV) {
543 		dumpdev = dev;
544 		return (0);
545 	}
546 	psize = dev_dpsize(dev);
547 	if (psize == -1)
548 		return (ENXIO);
549 	/*
550 	 * XXX should clean up checking in dumpsys() to be more like this.
551 	 */
552 	newdumplo = psize - Maxmem * (PAGE_SIZE / DEV_BSIZE);
553 	if (newdumplo <= LABELSECTOR)
554 		return (ENOSPC);
555 	dumpdev = dev;
556 	dumplo = newdumplo;
557 	return (0);
558 }
559 
560 
561 /* ARGSUSED */
562 static void dump_conf (void *dummy);
563 static void
564 dump_conf(dummy)
565 	void *dummy;
566 {
567 	char *path;
568 	cdev_t dev;
569 
570 	path = kmalloc(MNAMELEN, M_TEMP, M_WAITOK);
571 	if (TUNABLE_STR_FETCH("dumpdev", path, MNAMELEN) != 0) {
572 		dev = kgetdiskbyname(path);
573 		if (dev != NOCDEV)
574 			dumpdev = dev;
575 	}
576 	kfree(path, M_TEMP);
577 	if (setdumpdev(dumpdev) != 0)
578 		dumpdev = NOCDEV;
579 }
580 
581 SYSINIT(dump_conf, SI_SUB_DUMP_CONF, SI_ORDER_FIRST, dump_conf, NULL)
582 
583 static int
584 sysctl_kern_dumpdev(SYSCTL_HANDLER_ARGS)
585 {
586 	int error;
587 	udev_t ndumpdev;
588 
589 	ndumpdev = dev2udev(dumpdev);
590 	error = sysctl_handle_opaque(oidp, &ndumpdev, sizeof ndumpdev, req);
591 	if (error == 0 && req->newptr != NULL)
592 		error = setdumpdev(udev2dev(ndumpdev, 0));
593 	return (error);
594 }
595 
596 SYSCTL_PROC(_kern, KERN_DUMPDEV, dumpdev, CTLTYPE_OPAQUE|CTLFLAG_RW,
597 	0, sizeof dumpdev, sysctl_kern_dumpdev, "T,udev_t", "");
598 
599 /*
600  * Doadump comes here after turning off memory management and
601  * getting on the dump stack, either when called above, or by
602  * the auto-restart code.
603  */
604 static void
605 dumpsys(void)
606 {
607 	int	error;
608 
609 	savectx(&dumppcb);
610 	dumpthread = curthread;
611 	if (dumping++) {
612 		printf("Dump already in progress, bailing...\n");
613 		return;
614 	}
615 	if (!dodump)
616 		return;
617 	if (dumpdev == NOCDEV)
618 		return;
619 	dumpsize = Maxmem;
620 	printf("\ndumping to dev %s, offset %ld\n", devtoname(dumpdev), dumplo);
621 	printf("dump ");
622 	error = dev_ddump(dumpdev);
623 	if (error == 0) {
624 		printf("succeeded\n");
625 		return;
626 	}
627 	printf("failed, reason: ");
628 	switch (error) {
629 	case ENOSYS:
630 	case ENODEV:
631 		printf("device doesn't support a dump routine\n");
632 		break;
633 
634 	case ENXIO:
635 		printf("device bad\n");
636 		break;
637 
638 	case EFAULT:
639 		printf("device not ready\n");
640 		break;
641 
642 	case EINVAL:
643 		printf("area improper\n");
644 		break;
645 
646 	case EIO:
647 		printf("i/o error\n");
648 		break;
649 
650 	case EINTR:
651 		printf("aborted from console\n");
652 		break;
653 
654 	default:
655 		printf("unknown, error = %d\n", error);
656 		break;
657 	}
658 }
659 
660 int
661 dumpstatus(vm_offset_t addr, off_t count)
662 {
663 	int c;
664 
665 	if (addr % (1024 * 1024) == 0) {
666 #ifdef HW_WDOG
667 		if (wdog_tickler)
668 			(*wdog_tickler)();
669 #endif
670 		printf("%ld ", (long)(count / (1024 * 1024)));
671 	}
672 
673 	if ((c = cncheckc()) == 0x03)
674 		return -1;
675 	else if (c != -1)
676 		printf("[CTRL-C to abort] ");
677 
678 	return 0;
679 }
680 
681 /*
682  * Panic is called on unresolvable fatal errors.  It prints "panic: mesg",
683  * and then reboots.  If we are called twice, then we avoid trying to sync
684  * the disks as this often leads to recursive panics.
685  */
686 void
687 panic(const char *fmt, ...)
688 {
689 	int bootopt, newpanic;
690 	__va_list ap;
691 	static char buf[256];
692 
693 #ifdef SMP
694 	/*
695 	 * If a panic occurs on multiple cpus before the first is able to
696 	 * halt the other cpus, only one cpu is allowed to take the panic.
697 	 * Attempt to be verbose about this situation but if the printf()
698 	 * itself panics don't let us overrun the kernel stack.
699 	 *
700 	 * Be very nasty about descheduling our thread at the lowest
701 	 * level possible in an attempt to freeze the thread without
702 	 * inducing further panics.
703 	 *
704 	 * Bumping gd_trap_nesting_level will also bypass assertions in
705 	 * lwkt_switch() and allow us to switch away even if we are a
706 	 * FAST interrupt or IPI.
707 	 */
708 	if (atomic_poll_acquire_int(&panic_cpu_interlock)) {
709 		panic_cpu_gd = mycpu;
710 	} else if (panic_cpu_gd != mycpu) {
711 		crit_enter();
712 		++mycpu->gd_trap_nesting_level;
713 		if (mycpu->gd_trap_nesting_level < 25) {
714 			printf("SECONDARY PANIC ON CPU %d THREAD %p\n",
715 				mycpu->gd_cpuid, curthread);
716 		}
717 		curthread->td_release = NULL;	/* be a grinch */
718 		for (;;) {
719 			lwkt_deschedule_self(curthread);
720 			lwkt_switch();
721 		}
722 		/* NOT REACHED */
723 		/* --mycpu->gd_trap_nesting_level */
724 		/* crit_exit() */
725 	}
726 #endif
727 	bootopt = RB_AUTOBOOT | RB_DUMP;
728 	if (sync_on_panic == 0)
729 		bootopt |= RB_NOSYNC;
730 	newpanic = 0;
731 	if (panicstr)
732 		bootopt |= RB_NOSYNC;
733 	else {
734 		panicstr = fmt;
735 		newpanic = 1;
736 	}
737 
738 	__va_start(ap, fmt);
739 	(void)vsnprintf(buf, sizeof(buf), fmt, ap);
740 	if (panicstr == fmt)
741 		panicstr = buf;
742 	__va_end(ap);
743 	printf("panic: %s\n", buf);
744 #ifdef SMP
745 	/* three separate prints in case of an unmapped page and trap */
746 	printf("mp_lock = %08x; ", mp_lock);
747 	printf("cpuid = %d; ", mycpu->gd_cpuid);
748 	printf("lapic.id = %08x\n", lapic.id);
749 #endif
750 
751 #if defined(DDB)
752 	if (newpanic && trace_on_panic)
753 		db_print_backtrace();
754 	if (debugger_on_panic)
755 		Debugger ("panic");
756 #endif
757 	boot(bootopt);
758 }
759 
760 /*
761  * Support for poweroff delay.
762  */
763 #ifndef POWEROFF_DELAY
764 # define POWEROFF_DELAY 5000
765 #endif
766 static int poweroff_delay = POWEROFF_DELAY;
767 
768 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW,
769 	&poweroff_delay, 0, "");
770 
771 static void
772 poweroff_wait(void *junk, int howto)
773 {
774 	if(!(howto & RB_POWEROFF) || poweroff_delay <= 0)
775 		return;
776 	DELAY(poweroff_delay * 1000);
777 }
778 
779 /*
780  * Some system processes (e.g. syncer) need to be stopped at appropriate
781  * points in their main loops prior to a system shutdown, so that they
782  * won't interfere with the shutdown process (e.g. by holding a disk buf
783  * to cause sync to fail).  For each of these system processes, register
784  * shutdown_kproc() as a handler for one of shutdown events.
785  */
786 static int kproc_shutdown_wait = 60;
787 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW,
788     &kproc_shutdown_wait, 0, "");
789 
790 void
791 shutdown_kproc(void *arg, int howto)
792 {
793 	struct thread *td;
794 	struct proc *p;
795 	int error;
796 
797 	if (panicstr)
798 		return;
799 
800 	td = (struct thread *)arg;
801 	if ((p = td->td_proc) != NULL) {
802 	    printf("Waiting (max %d seconds) for system process `%s' to stop...",
803 		kproc_shutdown_wait, p->p_comm);
804 	} else {
805 	    printf("Waiting (max %d seconds) for system thread %s to stop...",
806 		kproc_shutdown_wait, td->td_comm);
807 	}
808 	error = suspend_kproc(td, kproc_shutdown_wait * hz);
809 
810 	if (error == EWOULDBLOCK)
811 		printf("timed out\n");
812 	else
813 		printf("stopped\n");
814 }
815