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. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)kern_shutdown.c 8.3 (Berkeley) 1/21/94 35 * $FreeBSD: src/sys/kern/kern_shutdown.c,v 1.72.2.12 2002/02/21 19:15:10 dillon Exp $ 36 */ 37 38 #include "opt_ddb.h" 39 #include "opt_ddb_trace.h" 40 #include "opt_panic.h" 41 #include "opt_show_busybufs.h" 42 #include "use_gpio.h" 43 44 #include <sys/param.h> 45 #include <sys/systm.h> 46 #include <sys/eventhandler.h> 47 #include <sys/buf.h> 48 #include <sys/disk.h> 49 #include <sys/diskslice.h> 50 #include <sys/reboot.h> 51 #include <sys/proc.h> 52 #include <sys/priv.h> 53 #include <sys/fcntl.h> /* FREAD */ 54 #include <sys/stat.h> /* S_IFCHR */ 55 #include <sys/vnode.h> 56 #include <sys/kernel.h> 57 #include <sys/kerneldump.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/kbio.h> 69 #include <sys/shm.h> 70 #include <sys/kern_syscall.h> 71 #include <vm/vm_map.h> 72 #include <vm/pmap.h> 73 74 #include <sys/thread2.h> 75 #include <sys/buf2.h> 76 #include <sys/mplock2.h> 77 78 #include <machine/cpu.h> 79 #include <machine/clock.h> 80 #include <machine/md_var.h> 81 #include <machine/smp.h> /* smp_active_mask, cpuid */ 82 #include <machine/vmparam.h> 83 #include <machine/thread.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 /* 127 * Variable panicstr contains argument to first call to panic; used as flag 128 * to indicate that the kernel has already called panic. 129 */ 130 const char *panicstr; 131 132 int dumping; /* system is dumping */ 133 static struct dumperinfo dumper; /* selected dumper */ 134 135 globaldata_t panic_cpu_gd; /* which cpu took the panic */ 136 struct lwkt_tokref panic_tokens[LWKT_MAXTOKENS]; 137 int panic_tokens_count; 138 139 int bootverbose = 0; /* note: assignment to force non-bss */ 140 SYSCTL_INT(_debug, OID_AUTO, bootverbose, CTLFLAG_RW, 141 &bootverbose, 0, "Verbose kernel messages"); 142 143 int cold = 1; /* note: assignment to force non-bss */ 144 int dumplo; /* OBSOLETE - savecore compat */ 145 u_int64_t dumplo64; 146 147 static void boot (int) __dead2; 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_BOOT2_MACHDEP, SI_ORDER_ANY, shutdown_conf, NULL); 169 170 /* ARGSUSED */ 171 172 /* 173 * The system call that results in a reboot 174 * 175 * MPALMOSTSAFE 176 */ 177 int 178 sys_reboot(struct reboot_args *uap) 179 { 180 struct thread *td = curthread; 181 int error; 182 183 if ((error = priv_check(td, PRIV_REBOOT))) 184 return (error); 185 186 get_mplock(); 187 boot(uap->opt); 188 rel_mplock(); 189 return (0); 190 } 191 192 /* 193 * Called by events that want to shut down.. e.g <CTL><ALT><DEL> on a PC 194 */ 195 static int shutdown_howto = 0; 196 197 void 198 shutdown_nice(int howto) 199 { 200 shutdown_howto = howto; 201 202 /* Send a signal to init(8) and have it shutdown the world */ 203 if (initproc != NULL) { 204 ksignal(initproc, SIGINT); 205 } else { 206 /* No init(8) running, so simply reboot */ 207 boot(RB_NOSYNC); 208 } 209 return; 210 } 211 static int waittime = -1; 212 struct pcb dumppcb; 213 struct thread *dumpthread; 214 215 static void 216 print_uptime(void) 217 { 218 int f; 219 struct timespec ts; 220 221 getnanouptime(&ts); 222 kprintf("Uptime: "); 223 f = 0; 224 if (ts.tv_sec >= 86400) { 225 kprintf("%ldd", ts.tv_sec / 86400); 226 ts.tv_sec %= 86400; 227 f = 1; 228 } 229 if (f || ts.tv_sec >= 3600) { 230 kprintf("%ldh", ts.tv_sec / 3600); 231 ts.tv_sec %= 3600; 232 f = 1; 233 } 234 if (f || ts.tv_sec >= 60) { 235 kprintf("%ldm", ts.tv_sec / 60); 236 ts.tv_sec %= 60; 237 f = 1; 238 } 239 kprintf("%lds\n", ts.tv_sec); 240 } 241 242 /* 243 * Go through the rigmarole of shutting down.. 244 * this used to be in machdep.c but I'll be dammned if I could see 245 * anything machine dependant in it. 246 */ 247 static void 248 boot(int howto) 249 { 250 /* 251 * Get rid of any user scheduler baggage and then give 252 * us a high priority. 253 */ 254 if (curthread->td_release) 255 curthread->td_release(curthread); 256 lwkt_setpri_self(TDPRI_MAX); 257 258 /* collect extra flags that shutdown_nice might have set */ 259 howto |= shutdown_howto; 260 261 /* 262 * We really want to shutdown on the BSP. Subsystems such as ACPI 263 * can't power-down the box otherwise. 264 */ 265 if (!CPUMASK_ISUP(smp_active_mask)) { 266 kprintf("boot() called on cpu#%d\n", mycpu->gd_cpuid); 267 } 268 if (panicstr == NULL && mycpu->gd_cpuid != 0) { 269 kprintf("Switching to cpu #0 for shutdown\n"); 270 lwkt_setcpu_self(globaldata_find(0)); 271 } 272 /* 273 * Do any callouts that should be done BEFORE syncing the filesystems. 274 */ 275 EVENTHANDLER_INVOKE(shutdown_pre_sync, howto); 276 277 /* 278 * Try to get rid of any remaining FS references. The calling 279 * process, proc0, and init may still hold references. The 280 * VFS cache subsystem may still hold a root reference to root. 281 * 282 * XXX this needs work. We really need to SIGSTOP all remaining 283 * processes in order to avoid blowups due to proc0's filesystem 284 * references going away. For now just make sure that the init 285 * process is stopped. 286 */ 287 if (panicstr == NULL) { 288 shutdown_cleanup_proc(curproc); 289 shutdown_cleanup_proc(&proc0); 290 if (initproc) { 291 if (initproc != curproc) { 292 ksignal(initproc, SIGSTOP); 293 tsleep(boot, 0, "shutdn", hz / 20); 294 } 295 shutdown_cleanup_proc(initproc); 296 } 297 vfs_cache_setroot(NULL, NULL); 298 } 299 300 /* 301 * Now sync filesystems 302 */ 303 if (!cold && (howto & RB_NOSYNC) == 0 && waittime < 0) { 304 int iter, nbusy, pbusy; 305 306 waittime = 0; 307 kprintf("\nsyncing disks... "); 308 309 sys_sync(NULL); /* YYY was sync(&proc0, NULL). why proc0 ? */ 310 311 /* 312 * With soft updates, some buffers that are 313 * written will be remarked as dirty until other 314 * buffers are written. 315 */ 316 for (iter = pbusy = 0; iter < 20; iter++) { 317 nbusy = scan_all_buffers(shutdown_busycount1, NULL); 318 if (nbusy == 0) 319 break; 320 kprintf("%d ", nbusy); 321 if (nbusy < pbusy) 322 iter = 0; 323 pbusy = nbusy; 324 /* 325 * XXX: 326 * Process soft update work queue if buffers don't sync 327 * after 6 iterations by permitting the syncer to run. 328 */ 329 if (iter > 5) 330 bio_ops_sync(NULL); 331 332 sys_sync(NULL); /* YYY was sync(&proc0, NULL). why proc0 ? */ 333 tsleep(boot, 0, "shutdn", hz * iter / 20 + 1); 334 } 335 kprintf("\n"); 336 /* 337 * Count only busy local buffers to prevent forcing 338 * a fsck if we're just a client of a wedged NFS server 339 */ 340 nbusy = scan_all_buffers(shutdown_busycount2, NULL); 341 if (nbusy) { 342 /* 343 * Failed to sync all blocks. Indicate this and don't 344 * unmount filesystems (thus forcing an fsck on reboot). 345 */ 346 kprintf("giving up on %d buffers\n", nbusy); 347 #ifdef DDB 348 if (debugger_on_panic) 349 Debugger("busy buffer problem"); 350 #endif /* DDB */ 351 tsleep(boot, 0, "shutdn", hz * 5 + 1); 352 } else { 353 kprintf("done\n"); 354 /* 355 * Unmount filesystems 356 */ 357 if (panicstr == NULL) 358 vfs_unmountall(); 359 } 360 tsleep(boot, 0, "shutdn", hz / 10 + 1); 361 } 362 363 print_uptime(); 364 365 /* 366 * Dump before doing post_sync shutdown ops 367 */ 368 crit_enter(); 369 if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP && !cold) { 370 dumpsys(); 371 } 372 373 /* 374 * Ok, now do things that assume all filesystem activity has 375 * been completed. This will also call the device shutdown 376 * methods. 377 */ 378 EVENTHANDLER_INVOKE(shutdown_post_sync, howto); 379 380 /* Now that we're going to really halt the system... */ 381 EVENTHANDLER_INVOKE(shutdown_final, howto); 382 383 for(;;) ; /* safety against shutdown_reset not working */ 384 /* NOTREACHED */ 385 } 386 387 /* 388 * Pass 1 - Figure out if there are any busy or dirty buffers still present. 389 * 390 * We ignore TMPFS mounts in this pass. 391 */ 392 static int 393 shutdown_busycount1(struct buf *bp, void *info) 394 { 395 struct vnode *vp; 396 397 if ((vp = bp->b_vp) != NULL && vp->v_tag == VT_TMPFS) 398 return (0); 399 if ((bp->b_flags & B_INVAL) == 0 && BUF_REFCNT(bp) > 0) 400 return(1); 401 if ((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) 402 return (1); 403 return (0); 404 } 405 406 /* 407 * Pass 2 - only run after pass 1 has completed or has given up 408 * 409 * We ignore TMPFS, NFS, MFS, and SMBFS mounts in this pass. 410 */ 411 static int 412 shutdown_busycount2(struct buf *bp, void *info) 413 { 414 struct vnode *vp; 415 416 /* 417 * Ignore tmpfs and nfs mounts 418 */ 419 if ((vp = bp->b_vp) != NULL) { 420 if (vp->v_tag == VT_TMPFS) 421 return (0); 422 if (vp->v_tag == VT_NFS) 423 return (0); 424 if (vp->v_tag == VT_MFS) 425 return (0); 426 if (vp->v_tag == VT_SMBFS) 427 return (0); 428 } 429 430 /* 431 * Only count buffers stuck on I/O, ignore everything else 432 */ 433 if (((bp->b_flags & B_INVAL) == 0 && BUF_REFCNT(bp)) || 434 ((bp->b_flags & (B_DELWRI|B_INVAL)) == B_DELWRI)) { 435 /* 436 * Only count buffers undergoing write I/O 437 * on the related vnode. 438 */ 439 if (bp->b_vp == NULL || 440 bio_track_active(&bp->b_vp->v_track_write) == 0) { 441 return (0); 442 } 443 #if defined(SHOW_BUSYBUFS) || defined(DIAGNOSTIC) 444 kprintf( 445 "%p dev:?, flags:%08x, loffset:%jd, doffset:%jd\n", 446 bp, 447 bp->b_flags, (intmax_t)bp->b_loffset, 448 (intmax_t)bp->b_bio2.bio_offset); 449 #endif 450 return(1); 451 } 452 return(0); 453 } 454 455 /* 456 * If the shutdown was a clean halt, behave accordingly. 457 */ 458 static void 459 shutdown_halt(void *junk, int howto) 460 { 461 if (howto & RB_HALT) { 462 kprintf("\n"); 463 kprintf("The operating system has halted.\n"); 464 #ifdef _KERNEL_VIRTUAL 465 cpu_halt(); 466 #else 467 kprintf("Please press any key to reboot.\n\n"); 468 switch (cngetc()) { 469 case -1: /* No console, just die */ 470 cpu_halt(); 471 /* NOTREACHED */ 472 default: 473 howto &= ~RB_HALT; 474 break; 475 } 476 #endif 477 } 478 } 479 480 /* 481 * Check to see if the system paniced, pause and then reboot 482 * according to the specified delay. 483 */ 484 static void 485 shutdown_panic(void *junk, int howto) 486 { 487 int loop; 488 int c; 489 490 if (howto & RB_DUMP) { 491 if (PANIC_REBOOT_WAIT_TIME != 0) { 492 if (PANIC_REBOOT_WAIT_TIME != -1) { 493 kprintf("Automatic reboot in %d seconds - " 494 "press a key on the console to abort\n", 495 PANIC_REBOOT_WAIT_TIME); 496 for (loop = PANIC_REBOOT_WAIT_TIME * 10; 497 loop > 0; --loop) { 498 DELAY(1000 * 100); /* 1/10th second */ 499 /* Did user type a key? */ 500 c = cncheckc(); 501 if (c != -1 && c != NOKEY) 502 break; 503 } 504 if (!loop) 505 return; 506 } 507 } else { /* zero time specified - reboot NOW */ 508 return; 509 } 510 kprintf("--> Press a key on the console to reboot,\n"); 511 kprintf("--> or switch off the system now.\n"); 512 cngetc(); 513 } 514 } 515 516 /* 517 * Everything done, now reset 518 */ 519 static void 520 shutdown_reset(void *junk, int howto) 521 { 522 kprintf("Rebooting...\n"); 523 DELAY(1000000); /* wait 1 sec for kprintf's to complete and be read */ 524 /* cpu_boot(howto); */ /* doesn't do anything at the moment */ 525 cpu_reset(); 526 /* NOTREACHED */ /* assuming reset worked */ 527 } 528 529 /* 530 * Try to remove FS references in the specified process. This function 531 * is used during shutdown 532 */ 533 static 534 void 535 shutdown_cleanup_proc(struct proc *p) 536 { 537 struct filedesc *fdp; 538 struct vmspace *vm; 539 540 if (p == NULL) 541 return; 542 if ((fdp = p->p_fd) != NULL) { 543 kern_closefrom(0); 544 if (fdp->fd_cdir) { 545 cache_drop(&fdp->fd_ncdir); 546 vrele(fdp->fd_cdir); 547 fdp->fd_cdir = NULL; 548 } 549 if (fdp->fd_rdir) { 550 cache_drop(&fdp->fd_nrdir); 551 vrele(fdp->fd_rdir); 552 fdp->fd_rdir = NULL; 553 } 554 if (fdp->fd_jdir) { 555 cache_drop(&fdp->fd_njdir); 556 vrele(fdp->fd_jdir); 557 fdp->fd_jdir = NULL; 558 } 559 } 560 if (p->p_vkernel) 561 vkernel_exit(p); 562 if (p->p_textvp) { 563 vrele(p->p_textvp); 564 p->p_textvp = NULL; 565 } 566 vm = p->p_vmspace; 567 if (vm != NULL) { 568 pmap_remove_pages(vmspace_pmap(vm), 569 VM_MIN_USER_ADDRESS, 570 VM_MAX_USER_ADDRESS); 571 vm_map_remove(&vm->vm_map, 572 VM_MIN_USER_ADDRESS, 573 VM_MAX_USER_ADDRESS); 574 } 575 } 576 577 /* 578 * Magic number for savecore 579 * 580 * exported (symorder) and used at least by savecore(8) 581 * 582 * Mark it as used so that gcc doesn't optimize it away. 583 */ 584 __attribute__((__used__)) 585 static u_long const dumpmag = 0x8fca0101UL; 586 587 __attribute__((__used__)) 588 static int dumpsize = 0; /* also for savecore */ 589 590 static int dodump = 1; 591 592 SYSCTL_INT(_machdep, OID_AUTO, do_dump, CTLFLAG_RW, &dodump, 0, 593 "Try to perform coredump on kernel panic"); 594 595 void 596 mkdumpheader(struct kerneldumpheader *kdh, char *magic, uint32_t archver, 597 uint64_t dumplen, uint32_t blksz) 598 { 599 bzero(kdh, sizeof(*kdh)); 600 strncpy(kdh->magic, magic, sizeof(kdh->magic)); 601 strncpy(kdh->architecture, MACHINE_ARCH, sizeof(kdh->architecture)); 602 kdh->version = htod32(KERNELDUMPVERSION); 603 kdh->architectureversion = htod32(archver); 604 kdh->dumplength = htod64(dumplen); 605 kdh->dumptime = htod64(time_second); 606 kdh->blocksize = htod32(blksz); 607 strncpy(kdh->hostname, hostname, sizeof(kdh->hostname)); 608 strncpy(kdh->versionstring, version, sizeof(kdh->versionstring)); 609 if (panicstr != NULL) 610 strncpy(kdh->panicstring, panicstr, sizeof(kdh->panicstring)); 611 kdh->parity = kerneldump_parity(kdh); 612 } 613 614 static int 615 setdumpdev(cdev_t dev) 616 { 617 int error; 618 int doopen; 619 620 if (dev == NULL) { 621 disk_dumpconf(NULL, 0/*off*/); 622 return (0); 623 } 624 625 /* 626 * We have to open the device before we can perform ioctls on it, 627 * or the slice/label data may not be present. Device opens are 628 * usually tracked by specfs, but the dump device can be set in 629 * early boot and may not be open so this is somewhat of a hack. 630 */ 631 doopen = (dev->si_sysref.refcnt == 1); 632 if (doopen) { 633 error = dev_dopen(dev, FREAD, S_IFCHR, proc0.p_ucred, NULL); 634 if (error) 635 return (error); 636 } 637 error = disk_dumpconf(dev, 1/*on*/); 638 639 return error; 640 } 641 642 /* ARGSUSED */ 643 static void dump_conf (void *dummy); 644 static void 645 dump_conf(void *dummy) 646 { 647 char *path; 648 cdev_t dev; 649 int _dummy; 650 651 path = kmalloc(MNAMELEN, M_TEMP, M_WAITOK); 652 if (TUNABLE_STR_FETCH("dumpdev", path, MNAMELEN) != 0) { 653 /* 654 * Make sure all disk devices created so far have also been 655 * probed, and also make sure that the newly created device 656 * nodes for probed disks are ready, too. 657 * 658 * XXX - Delay an additional 2 seconds to help drivers which 659 * pickup devices asynchronously and are not caught by 660 * CAM's initial probe. 661 */ 662 sync_devs(); 663 tsleep(&_dummy, 0, "syncer", hz*2); 664 665 dev = kgetdiskbyname(path); 666 if (dev != NULL) 667 dumpdev = dev; 668 } 669 kfree(path, M_TEMP); 670 if (setdumpdev(dumpdev) != 0) 671 dumpdev = NULL; 672 } 673 674 SYSINIT(dump_conf, SI_SUB_DUMP_CONF, SI_ORDER_FIRST, dump_conf, NULL); 675 676 static int 677 sysctl_kern_dumpdev(SYSCTL_HANDLER_ARGS) 678 { 679 int error; 680 udev_t ndumpdev; 681 682 ndumpdev = dev2udev(dumpdev); 683 error = sysctl_handle_opaque(oidp, &ndumpdev, sizeof ndumpdev, req); 684 if (error == 0 && req->newptr != NULL) 685 error = setdumpdev(udev2dev(ndumpdev, 0)); 686 return (error); 687 } 688 689 SYSCTL_PROC(_kern, KERN_DUMPDEV, dumpdev, CTLTYPE_OPAQUE|CTLFLAG_RW, 690 0, sizeof dumpdev, sysctl_kern_dumpdev, "T,udev_t", ""); 691 692 /* 693 * Panic is called on unresolvable fatal errors. It prints "panic: mesg", 694 * and then reboots. If we are called twice, then we avoid trying to sync 695 * the disks as this often leads to recursive panics. 696 */ 697 void 698 panic(const char *fmt, ...) 699 { 700 int bootopt, newpanic; 701 globaldata_t gd = mycpu; 702 thread_t td = gd->gd_curthread; 703 __va_list ap; 704 static char buf[256]; 705 706 /* 707 * If a panic occurs on multiple cpus before the first is able to 708 * halt the other cpus, only one cpu is allowed to take the panic. 709 * Attempt to be verbose about this situation but if the kprintf() 710 * itself panics don't let us overrun the kernel stack. 711 * 712 * Be very nasty about descheduling our thread at the lowest 713 * level possible in an attempt to freeze the thread without 714 * inducing further panics. 715 * 716 * Bumping gd_trap_nesting_level will also bypass assertions in 717 * lwkt_switch() and allow us to switch away even if we are a 718 * FAST interrupt or IPI. 719 * 720 * The setting of panic_cpu_gd also determines how kprintf() 721 * spin-locks itself. DDB can set panic_cpu_gd as well. 722 */ 723 for (;;) { 724 globaldata_t xgd = panic_cpu_gd; 725 726 /* 727 * Someone else got the panic cpu 728 */ 729 if (xgd && xgd != gd) { 730 crit_enter(); 731 ++mycpu->gd_trap_nesting_level; 732 if (mycpu->gd_trap_nesting_level < 25) { 733 kprintf("SECONDARY PANIC ON CPU %d THREAD %p\n", 734 mycpu->gd_cpuid, td); 735 } 736 td->td_release = NULL; /* be a grinch */ 737 for (;;) { 738 lwkt_deschedule_self(td); 739 lwkt_switch(); 740 } 741 /* NOT REACHED */ 742 /* --mycpu->gd_trap_nesting_level */ 743 /* crit_exit() */ 744 } 745 746 /* 747 * Reentrant panic 748 */ 749 if (xgd && xgd == gd) 750 break; 751 752 /* 753 * We got it 754 */ 755 if (atomic_cmpset_ptr(&panic_cpu_gd, NULL, gd)) 756 break; 757 } 758 /* 759 * Try to get the system into a working state. Save information 760 * we are about to destroy. 761 */ 762 kvcreinitspin(); 763 if (panicstr == NULL) { 764 bcopy(td->td_toks_array, panic_tokens, sizeof(panic_tokens)); 765 panic_tokens_count = td->td_toks_stop - &td->td_toks_base; 766 } 767 lwkt_relalltokens(td); 768 td->td_toks_stop = &td->td_toks_base; 769 if (gd->gd_spinlocks) 770 kprintf("panic with %d spinlocks held\n", gd->gd_spinlocks); 771 gd->gd_spinlocks = 0; 772 773 /* 774 * Setup 775 */ 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 /* 788 * Format the panic string. 789 */ 790 __va_start(ap, fmt); 791 kvsnprintf(buf, sizeof(buf), fmt, ap); 792 if (panicstr == fmt) 793 panicstr = buf; 794 __va_end(ap); 795 kprintf("panic: %s\n", buf); 796 /* two separate prints in case of an unmapped page and trap */ 797 kprintf("cpuid = %d\n", mycpu->gd_cpuid); 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) 804 wdog_disable(); 805 #endif 806 807 /* 808 * Enter the debugger or fall through & dump. Entering the 809 * debugger will stop cpus. If not entering the debugger stop 810 * cpus here. 811 * 812 * Limit the trace history to leave more panic data on a 813 * potentially row-limited console. 814 */ 815 #if defined(DDB) 816 if (newpanic && trace_on_panic) 817 print_backtrace(6); 818 if (debugger_on_panic) 819 Debugger("panic"); 820 else 821 #endif 822 if (newpanic) 823 stop_cpus(mycpu->gd_other_cpus); 824 boot(bootopt); 825 } 826 827 /* 828 * Support for poweroff delay. 829 */ 830 #ifndef POWEROFF_DELAY 831 # define POWEROFF_DELAY 5000 832 #endif 833 static int poweroff_delay = POWEROFF_DELAY; 834 835 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW, 836 &poweroff_delay, 0, ""); 837 838 static void 839 poweroff_wait(void *junk, int howto) 840 { 841 if(!(howto & RB_POWEROFF) || poweroff_delay <= 0) 842 return; 843 DELAY(poweroff_delay * 1000); 844 } 845 846 /* 847 * Some system processes (e.g. syncer) need to be stopped at appropriate 848 * points in their main loops prior to a system shutdown, so that they 849 * won't interfere with the shutdown process (e.g. by holding a disk buf 850 * to cause sync to fail). For each of these system processes, register 851 * shutdown_kproc() as a handler for one of shutdown events. 852 */ 853 static int kproc_shutdown_wait = 60; 854 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW, 855 &kproc_shutdown_wait, 0, ""); 856 857 void 858 shutdown_kproc(void *arg, int howto) 859 { 860 struct thread *td; 861 struct proc *p; 862 int error; 863 864 if (panicstr) 865 return; 866 867 td = (struct thread *)arg; 868 if ((p = td->td_proc) != NULL) { 869 kprintf("Waiting (max %d seconds) for system process `%s' to stop...", 870 kproc_shutdown_wait, p->p_comm); 871 } else { 872 kprintf("Waiting (max %d seconds) for system thread %s to stop...", 873 kproc_shutdown_wait, td->td_comm); 874 } 875 error = suspend_kproc(td, kproc_shutdown_wait * hz); 876 877 if (error == EWOULDBLOCK) 878 kprintf("timed out\n"); 879 else 880 kprintf("stopped\n"); 881 } 882 883 /* Registration of dumpers */ 884 int 885 set_dumper(struct dumperinfo *di) 886 { 887 if (di == NULL) { 888 bzero(&dumper, sizeof(dumper)); 889 return 0; 890 } 891 892 if (dumper.dumper != NULL) 893 return (EBUSY); 894 895 dumper = *di; 896 return 0; 897 } 898 899 void 900 dumpsys(void) 901 { 902 #if defined (_KERNEL_VIRTUAL) 903 /* vkernels don't support dumps */ 904 kprintf("vkernels don't support dumps\n"); 905 return; 906 #endif 907 /* 908 * If there is a dumper registered and we aren't dumping already, call 909 * the machine dependent dumpsys (md_dumpsys) to do the hard work. 910 * 911 * XXX: while right now the md_dumpsys() of x86 and x86_64 could be 912 * factored out completely into here, I rather keep them machine 913 * dependent in case we ever add a platform which does not share 914 * the same dumpsys() code, such as arm. 915 */ 916 if (dumper.dumper != NULL && !dumping) { 917 dumping++; 918 md_dumpsys(&dumper); 919 } 920 } 921 922 int dump_stop_usertds = 0; 923 924 static 925 void 926 need_user_resched_remote(void *dummy) 927 { 928 need_user_resched(); 929 } 930 931 void 932 dump_reactivate_cpus(void) 933 { 934 globaldata_t gd; 935 int cpu, seq; 936 937 dump_stop_usertds = 1; 938 939 need_user_resched(); 940 941 for (cpu = 0; cpu < ncpus; cpu++) { 942 gd = globaldata_find(cpu); 943 seq = lwkt_send_ipiq(gd, need_user_resched_remote, NULL); 944 lwkt_wait_ipiq(gd, seq); 945 } 946 947 restart_cpus(stopped_cpus); 948 } 949