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 "use_gpio.h" 42 43 #include <sys/param.h> 44 #include <sys/systm.h> 45 #include <sys/eventhandler.h> 46 #include <sys/buf.h> 47 #include <sys/disk.h> 48 #include <sys/diskslice.h> 49 #include <sys/reboot.h> 50 #include <sys/proc.h> 51 #include <sys/priv.h> 52 #include <sys/fcntl.h> /* FREAD */ 53 #include <sys/stat.h> /* S_IFCHR */ 54 #include <sys/vnode.h> 55 #include <sys/kernel.h> 56 #include <sys/kerneldump.h> 57 #include <sys/kthread.h> 58 #include <sys/malloc.h> 59 #include <sys/mount.h> 60 #include <sys/queue.h> 61 #include <sys/sysctl.h> 62 #include <sys/vkernel.h> 63 #include <sys/conf.h> 64 #include <sys/sysmsg.h> 65 #include <sys/device.h> 66 #include <sys/cons.h> 67 #include <sys/kbio.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 #include <sys/mplock2.h> 76 77 #include <machine/cpu.h> 78 #include <machine/clock.h> 79 #include <machine/md_var.h> 80 #include <machine/smp.h> /* smp_active_mask, cpuid */ 81 #include <machine/vmparam.h> 82 #include <machine/thread.h> 83 84 #include <sys/signalvar.h> 85 86 #if defined(WDOG_DISABLE_ON_PANIC) 87 #include <sys/wdog.h> 88 #endif 89 #include <dev/acpica/acpi_pvpanic/panic_notifier.h> 90 #if (NGPIO > 0) && defined(ERROR_LED_ON_PANIC) 91 #include <dev/misc/gpio/gpio.h> 92 #endif 93 94 #ifndef PANIC_REBOOT_WAIT_TIME 95 #define PANIC_REBOOT_WAIT_TIME 15 /* default to 15 seconds */ 96 #endif 97 98 /* 99 * Note that stdarg.h and the ANSI style va_start macro is used for both 100 * ANSI and traditional C compilers. We use the machine version to stay 101 * within the confines of the kernel header files. 102 */ 103 #include <machine/stdarg.h> 104 105 #ifdef DDB 106 #include <ddb/ddb.h> 107 #ifdef DDB_UNATTENDED 108 int debugger_on_panic = 0; 109 #else 110 int debugger_on_panic = 1; 111 #endif 112 SYSCTL_INT(_debug, OID_AUTO, debugger_on_panic, CTLFLAG_RW, 113 &debugger_on_panic, 0, "Run debugger on kernel panic"); 114 115 #ifdef DDB_TRACE 116 int trace_on_panic = 1; 117 #else 118 int trace_on_panic = 0; 119 #endif 120 SYSCTL_INT(_debug, OID_AUTO, trace_on_panic, CTLFLAG_RW, 121 &trace_on_panic, 0, "Print stack trace on kernel panic"); 122 #endif 123 124 static int sync_on_panic = 0; 125 SYSCTL_INT(_kern, OID_AUTO, sync_on_panic, CTLFLAG_RW, 126 &sync_on_panic, 0, "Do a sync before rebooting from a panic"); 127 128 SYSCTL_NODE(_kern, OID_AUTO, shutdown, CTLFLAG_RW, 0, "Shutdown environment"); 129 130 /* 131 * Variable panicstr contains argument to first call to panic; used as flag 132 * to indicate that the kernel has already called panic. 133 */ 134 const char *panicstr; 135 136 __read_mostly int dumping; /* system is dumping */ 137 static struct dumperinfo dumper; /* selected dumper */ 138 139 __read_frequently globaldata_t panic_cpu_gd; /* used in lock assertion */ 140 struct lwkt_tokref panic_tokens[LWKT_MAXTOKENS]; 141 int panic_tokens_count; 142 143 __read_mostly int bootverbose = 0; /* note: assignment to force non-bss */ 144 SYSCTL_INT(_debug, OID_AUTO, bootverbose, CTLFLAG_RW, 145 &bootverbose, 0, "Verbose kernel messages"); 146 147 int cold = 1; /* note: assignment to force non-bss */ 148 int dumplo; /* OBSOLETE - savecore compat */ 149 u_int64_t dumplo64; 150 151 static void boot (int) __dead2; 152 static int setdumpdev (cdev_t dev); 153 static void poweroff_wait (void *, int); 154 static void print_uptime (void); 155 static void shutdown_halt (void *junk, int howto); 156 static void shutdown_panic (void *junk, int howto); 157 static void shutdown_reset (void *junk, int howto); 158 static int shutdown_busycount1(struct buf *bp, void *info); 159 static int shutdown_busycount2(struct buf *bp, void *info); 160 static void shutdown_cleanup_proc(struct proc *p); 161 162 /* register various local shutdown events */ 163 static void 164 shutdown_conf(void *unused) 165 { 166 EVENTHANDLER_REGISTER(shutdown_final, poweroff_wait, NULL, SHUTDOWN_PRI_FIRST); 167 EVENTHANDLER_REGISTER(shutdown_final, shutdown_halt, NULL, SHUTDOWN_PRI_LAST + 100); 168 EVENTHANDLER_REGISTER(shutdown_final, shutdown_panic, NULL, SHUTDOWN_PRI_LAST + 100); 169 EVENTHANDLER_REGISTER(shutdown_final, shutdown_reset, NULL, SHUTDOWN_PRI_LAST + 200); 170 } 171 172 SYSINIT(shutdown_conf, SI_BOOT2_MACHDEP, SI_ORDER_ANY, shutdown_conf, NULL); 173 174 /* ARGSUSED */ 175 176 /* 177 * The system call that results in a reboot 178 * 179 * MPALMOSTSAFE 180 */ 181 int 182 sys_reboot(struct sysmsg *sysmsg, const struct reboot_args *uap) 183 { 184 struct thread *td = curthread; 185 int error; 186 187 if ((error = priv_check(td, PRIV_REBOOT))) 188 return (error); 189 190 get_mplock(); 191 boot(uap->opt); 192 rel_mplock(); 193 return (0); 194 } 195 196 /* 197 * Called by events that want to shut down.. e.g <CTL><ALT><DEL> on a PC 198 */ 199 static int shutdown_howto = 0; 200 201 void 202 shutdown_nice(int howto) 203 { 204 shutdown_howto = howto; 205 206 /* Send a signal to init(8) and have it shutdown the world */ 207 if (initproc != NULL) { 208 ksignal(initproc, SIGINT); 209 } else { 210 /* No init(8) running, so simply reboot */ 211 boot(RB_NOSYNC); 212 } 213 return; 214 } 215 static int waittime = -1; 216 struct pcb dumppcb; 217 struct thread *dumpthread; 218 219 static void 220 print_uptime(void) 221 { 222 int f; 223 struct timespec ts; 224 225 getnanouptime(&ts); 226 kprintf("Uptime: "); 227 f = 0; 228 if (ts.tv_sec >= 86400) { 229 kprintf("%ldd", ts.tv_sec / 86400); 230 ts.tv_sec %= 86400; 231 f = 1; 232 } 233 if (f || ts.tv_sec >= 3600) { 234 kprintf("%ldh", ts.tv_sec / 3600); 235 ts.tv_sec %= 3600; 236 f = 1; 237 } 238 if (f || ts.tv_sec >= 60) { 239 kprintf("%ldm", ts.tv_sec / 60); 240 ts.tv_sec %= 60; 241 f = 1; 242 } 243 kprintf("%lds\n", ts.tv_sec); 244 } 245 246 /* 247 * Go through the rigmarole of shutting down.. 248 * this used to be in machdep.c but I'll be dammned if I could see 249 * anything machine dependant in it. 250 */ 251 static void 252 boot(int howto) 253 { 254 /* 255 * Get rid of any user scheduler baggage and then give 256 * us a high priority. 257 */ 258 if (curthread->td_release) 259 curthread->td_release(curthread); 260 lwkt_setpri_self(TDPRI_MAX); 261 262 /* collect extra flags that shutdown_nice might have set */ 263 howto |= shutdown_howto; 264 265 /* 266 * We really want to shutdown on the BSP. Subsystems such as ACPI 267 * can't power-down the box otherwise. 268 */ 269 if (!CPUMASK_ISUP(smp_active_mask)) { 270 kprintf("boot() called on cpu#%d\n", mycpu->gd_cpuid); 271 } 272 if (panicstr == NULL && mycpu->gd_cpuid != 0) { 273 kprintf("Switching to cpu #0 for shutdown\n"); 274 lwkt_setcpu_self(globaldata_find(0)); 275 } 276 /* 277 * Do any callouts that should be done BEFORE syncing the filesystems. 278 */ 279 EVENTHANDLER_INVOKE(shutdown_pre_sync, howto); 280 281 /* 282 * Try to get rid of any remaining FS references. The calling 283 * process, proc0, and init may still hold references. The 284 * VFS cache subsystem may still hold a root reference to root. 285 * 286 * XXX this needs work. We really need to SIGSTOP all remaining 287 * processes in order to avoid blowups due to proc0's filesystem 288 * references going away. For now just make sure that the init 289 * process is stopped. 290 */ 291 if (panicstr == NULL) { 292 shutdown_cleanup_proc(curproc); 293 shutdown_cleanup_proc(&proc0); 294 if (initproc) { 295 if (initproc != curproc) { 296 ksignal(initproc, SIGSTOP); 297 tsleep(boot, 0, "shutdn", hz / 20); 298 } 299 shutdown_cleanup_proc(initproc); 300 } 301 vfs_cache_setroot(NULL, NULL); 302 } 303 304 /* 305 * Now sync filesystems 306 */ 307 if (!cold && (howto & RB_NOSYNC) == 0 && waittime < 0) { 308 int iter, nbusy, pbusy; 309 int zcount; 310 311 waittime = 0; 312 zcount = 0; 313 kprintf("\nsyncing disks... "); 314 315 sys_sync(NULL, NULL); 316 317 /* 318 * With soft updates, some buffers that are written will be 319 * remarked as dirty until other buffers are written. 320 * 321 * sys_sync() usually runs asynchronously, to give us a 322 * better chance of syncing the rest of the filesystems when 323 * one or more of them are stuck. 324 */ 325 for (iter = pbusy = 0; iter < 20 + zcount; iter++) { 326 if (iter <= 10) 327 nbusy = scan_all_buffers(shutdown_busycount1, 328 &iter); 329 else 330 nbusy = scan_all_buffers(shutdown_busycount2, 331 &iter); 332 kprintf("%d ", nbusy); 333 if (nbusy == 0) { 334 if (++zcount == 3) 335 break; 336 } else { 337 zcount = 0; 338 } 339 340 /* 341 * There could be a lot to sync, only allow iter to 342 * proceed while there is progress. 343 */ 344 if (nbusy < pbusy) { 345 if (iter > 10) 346 iter = 10; 347 else 348 iter = 0; 349 } 350 pbusy = nbusy; 351 352 /* 353 * XXX: 354 * Process soft update work queue if buffers don't sync 355 * after 6 iterations by permitting the syncer to run. 356 */ 357 if (iter > 5) 358 bio_ops_sync(NULL); 359 360 sys_sync(NULL, NULL); 361 tsleep(boot, 0, "shutdn", hz * iter / 20 + 1); 362 } 363 kprintf("\n"); 364 365 if (zcount < 3) { 366 /* 367 * Failed to sync all blocks. Indicate this and don't 368 * unmount filesystems (thus forcing an fsck on reboot). 369 */ 370 kprintf("giving up on %d buffers\n", nbusy); 371 #ifdef DDB 372 if (debugger_on_panic) 373 Debugger("busy buffer problem"); 374 #endif /* DDB */ 375 tsleep(boot, 0, "shutdn", hz * 5 + 1); 376 } else { 377 kprintf("done\n"); 378 379 /* 380 * Unmount filesystems 381 */ 382 if (panicstr == NULL) 383 vfs_unmountall(1); 384 } 385 tsleep(boot, 0, "shutdn", hz / 10 + 1); 386 } 387 388 print_uptime(); 389 390 /* 391 * Dump before doing post_sync shutdown ops 392 */ 393 crit_enter(); 394 if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP && !cold) { 395 dumpsys(); 396 } 397 398 /* 399 * Ok, now do things that assume all filesystem activity has 400 * been completed. This will also call the device shutdown 401 * methods. 402 */ 403 EVENTHANDLER_INVOKE(shutdown_post_sync, howto); 404 405 /* Now that we're going to really halt the system... */ 406 EVENTHANDLER_INVOKE(shutdown_final, howto); 407 408 for(;;) ; /* safety against shutdown_reset not working */ 409 /* NOTREACHED */ 410 } 411 412 /* 413 * Pass 1 - Figure out if there are any busy or dirty buffers still present. 414 * 415 * We ignore TMPFS mounts in this pass. 416 */ 417 static int 418 shutdown_busycount1(struct buf *bp, void *info __unused) 419 { 420 struct vnode *vp; 421 422 if ((vp = bp->b_vp) != NULL && vp->v_tag == VT_TMPFS) 423 return (0); 424 if ((bp->b_flags & B_INVAL) == 0 && BUF_LOCKINUSE(bp)) 425 return(1); 426 if ((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) 427 return (1); 428 return (0); 429 } 430 431 /* 432 * Pass 2 - only run after pass 1 has completed or has given up 433 * 434 * We ignore TMPFS, NFS, MFS, and SMBFS mounts in this pass. 435 */ 436 static int 437 shutdown_busycount2(struct buf *bp, void *info) 438 { 439 struct vnode *vp; 440 int *iterp = info; 441 const char *mpath; 442 443 /* 444 * Ignore tmpfs and nfs mounts 445 */ 446 if ((vp = bp->b_vp) != NULL) { 447 if (vp->v_tag == VT_TMPFS) 448 return (0); 449 if (vp->v_tag == VT_NFS) 450 return (0); 451 if (vp->v_tag == VT_MFS) 452 return (0); 453 if (vp->v_tag == VT_SMBFS) 454 return (0); 455 } 456 457 /* 458 * Only count buffers stuck on I/O, ignore everything else 459 */ 460 if (((bp->b_flags & B_INVAL) == 0 && BUF_LOCKINUSE(bp)) || 461 ((bp->b_flags & (B_DELWRI|B_INVAL)) == B_DELWRI)) { 462 /* 463 * Only count buffers undergoing write I/O 464 * on the related vnode. 465 */ 466 if (bp->b_vp == NULL || 467 bio_track_active(&bp->b_vp->v_track_write) == 0) { 468 return (0); 469 } 470 if (*iterp > 15) { 471 mpath = "?"; 472 if (bp->b_vp->v_mount) 473 mpath = bp->b_vp->v_mount->mnt_stat.f_mntonname; 474 475 kprintf("%p on %s, flags:%08x, loffset:%jd, " 476 "doffset:%jd\n", 477 bp, 478 mpath, 479 bp->b_flags, 480 (intmax_t)bp->b_loffset, 481 (intmax_t)bp->b_bio2.bio_offset); 482 } 483 return(1); 484 } 485 return(0); 486 } 487 488 /* 489 * If the shutdown was a clean halt, behave accordingly. 490 */ 491 static void 492 shutdown_halt(void *junk, int howto) 493 { 494 if (howto & RB_HALT) { 495 kprintf("\n"); 496 kprintf("The operating system has halted.\n"); 497 #ifdef _KERNEL_VIRTUAL 498 cpu_halt(); 499 #else 500 kprintf("Please press any key to reboot.\n\n"); 501 cnpoll(TRUE); 502 switch (cngetc()) { 503 case -1: /* No console, just die */ 504 cpu_halt(); 505 /* NOTREACHED */ 506 default: 507 howto &= ~RB_HALT; 508 break; 509 } 510 #endif 511 } 512 } 513 514 /* 515 * Check to see if the system paniced, pause and then reboot 516 * according to the specified delay. 517 */ 518 static void 519 shutdown_panic(void *junk, int howto) 520 { 521 int loop; 522 int c; 523 524 if (howto & RB_DUMP) { 525 if (PANIC_REBOOT_WAIT_TIME != 0) { 526 if (PANIC_REBOOT_WAIT_TIME != -1) { 527 kprintf("Automatic reboot in %d seconds - " 528 "press a key on the console to abort\n", 529 PANIC_REBOOT_WAIT_TIME); 530 for (loop = PANIC_REBOOT_WAIT_TIME * 10; 531 loop > 0; --loop) { 532 DELAY(1000 * 100); /* 1/10th second */ 533 /* Did user type a key? */ 534 c = cncheckc(); 535 if (c != -1 && c != NOKEY) 536 break; 537 } 538 if (!loop) 539 return; 540 } 541 } else { /* zero time specified - reboot NOW */ 542 return; 543 } 544 kprintf("--> Press a key on the console to reboot,\n"); 545 kprintf("--> or switch off the system now.\n"); 546 cngetc(); 547 } 548 } 549 550 /* 551 * Everything done, now reset 552 */ 553 static void 554 shutdown_reset(void *junk, int howto) 555 { 556 kprintf("Rebooting...\n"); 557 DELAY(1000000); /* wait 1 sec for kprintf's to complete and be read */ 558 /* cpu_boot(howto); */ /* doesn't do anything at the moment */ 559 cpu_reset(); 560 /* NOTREACHED */ /* assuming reset worked */ 561 } 562 563 /* 564 * Try to remove FS references in the specified process. This function 565 * is used during shutdown 566 */ 567 static 568 void 569 shutdown_cleanup_proc(struct proc *p) 570 { 571 struct filedesc *fdp; 572 struct vmspace *vm; 573 574 if (p == NULL) 575 return; 576 if ((fdp = p->p_fd) != NULL) { 577 kern_closefrom(0); 578 if (fdp->fd_cdir) { 579 cache_drop(&fdp->fd_ncdir); 580 vrele(fdp->fd_cdir); 581 fdp->fd_cdir = NULL; 582 } 583 if (fdp->fd_rdir) { 584 cache_drop(&fdp->fd_nrdir); 585 vrele(fdp->fd_rdir); 586 fdp->fd_rdir = NULL; 587 } 588 if (fdp->fd_jdir) { 589 cache_drop(&fdp->fd_njdir); 590 vrele(fdp->fd_jdir); 591 fdp->fd_jdir = NULL; 592 } 593 } 594 if (p->p_vkernel) 595 vkernel_exit(p); 596 if (p->p_textvp) { 597 vrele(p->p_textvp); 598 p->p_textvp = NULL; 599 } 600 if (p->p_textnch.ncp) 601 cache_drop(&p->p_textnch); 602 vm = p->p_vmspace; 603 if (vm != NULL) { 604 pmap_remove_pages(vmspace_pmap(vm), 605 VM_MIN_USER_ADDRESS, 606 VM_MAX_USER_ADDRESS); 607 vm_map_remove(&vm->vm_map, 608 VM_MIN_USER_ADDRESS, 609 VM_MAX_USER_ADDRESS); 610 } 611 } 612 613 /* 614 * Magic number for savecore 615 * 616 * exported (symorder) and used at least by savecore(8) 617 * 618 * Mark it as used so that gcc doesn't optimize it away. 619 */ 620 __attribute__((__used__)) 621 static u_long const dumpmag = 0x8fca0101UL; 622 623 __attribute__((__used__)) 624 static int dumpsize = 0; /* also for savecore */ 625 626 static int dodump = 1; 627 628 SYSCTL_INT(_machdep, OID_AUTO, do_dump, CTLFLAG_RW, &dodump, 0, 629 "Try to perform coredump on kernel panic"); 630 631 void 632 mkdumpheader(struct kerneldumpheader *kdh, char *magic, uint32_t archver, 633 uint64_t dumplen, uint32_t blksz) 634 { 635 bzero(kdh, sizeof(*kdh)); 636 strncpy(kdh->magic, magic, sizeof(kdh->magic)); 637 strncpy(kdh->architecture, MACHINE_ARCH, sizeof(kdh->architecture)); 638 kdh->version = htod32(KERNELDUMPVERSION); 639 kdh->architectureversion = htod32(archver); 640 kdh->dumplength = htod64(dumplen); 641 kdh->dumptime = htod64(time_second); 642 kdh->blocksize = htod32(blksz); 643 strncpy(kdh->hostname, hostname, sizeof(kdh->hostname)); 644 strncpy(kdh->versionstring, version, sizeof(kdh->versionstring)); 645 if (panicstr != NULL) 646 strncpy(kdh->panicstring, panicstr, sizeof(kdh->panicstring)); 647 kdh->parity = kerneldump_parity(kdh); 648 } 649 650 static int 651 setdumpdev(cdev_t dev) 652 { 653 int error; 654 int doopen; 655 656 if (dev == NULL) { 657 disk_dumpconf(NULL, 0/*off*/); 658 dumpdev = NULL; 659 return (0); 660 } 661 662 /* 663 * We have to open the device before we can perform ioctls on it, 664 * or the slice/label data may not be present. Device opens are 665 * usually tracked by specfs, but the dump device can be set in 666 * early boot and may not be open so this is somewhat of a hack. 667 */ 668 doopen = (dev->si_sysref.refcnt == 1); 669 if (doopen) { 670 error = dev_dopen(dev, FREAD, S_IFCHR, 671 proc0.p_ucred, NULL, NULL); 672 if (error) 673 return (error); 674 } 675 error = disk_dumpconf(dev, 1/*on*/); 676 if (error == 0) 677 dumpdev = dev; 678 679 return error; 680 } 681 682 /* ARGSUSED */ 683 static void dump_conf (void *dummy); 684 static void 685 dump_conf(void *dummy) 686 { 687 char *path; 688 cdev_t dev; 689 int _dummy; 690 691 path = kmalloc(MNAMELEN, M_TEMP, M_WAITOK); 692 if (TUNABLE_STR_FETCH("dumpdev", path, MNAMELEN) != 0) { 693 /* 694 * Make sure all disk devices created so far have also been 695 * probed, and also make sure that the newly created device 696 * nodes for probed disks are ready, too. 697 * 698 * XXX - Delay an additional 2 seconds to help drivers which 699 * pickup devices asynchronously and are not caught by 700 * CAM's initial probe. 701 */ 702 sync_devs(); 703 tsleep(&_dummy, 0, "syncer", hz*2); 704 705 dev = kgetdiskbyname(path); 706 if (dev != NULL) 707 dumpdev = dev; 708 } 709 kfree(path, M_TEMP); 710 if (setdumpdev(dumpdev) != 0) 711 dumpdev = NULL; 712 } 713 714 SYSINIT(dump_conf, SI_SUB_DUMP_CONF, SI_ORDER_FIRST, dump_conf, NULL); 715 716 static int 717 sysctl_kern_dumpdev(SYSCTL_HANDLER_ARGS) 718 { 719 int error; 720 dev_t ndumpdev; 721 722 ndumpdev = devid_from_dev(dumpdev); 723 error = sysctl_handle_opaque(oidp, &ndumpdev, sizeof ndumpdev, req); 724 if (error == 0 && req->newptr != NULL) 725 error = setdumpdev(dev_from_devid(ndumpdev, 0)); 726 return (error); 727 } 728 729 SYSCTL_PROC(_kern, KERN_DUMPDEV, dumpdev, CTLTYPE_OPAQUE|CTLFLAG_RW, 730 0, sizeof dumpdev, sysctl_kern_dumpdev, "T,udev_t", ""); 731 732 static struct panicerinfo *panic_notifier; 733 734 int 735 set_panic_notifier(struct panicerinfo *info) 736 { 737 if (info == NULL) 738 panic_notifier = NULL; 739 else if (panic_notifier != NULL) 740 return 1; 741 else 742 panic_notifier = info; 743 744 return 0; 745 } 746 747 /* 748 * Panic is called on unresolvable fatal errors. It prints "panic: mesg", 749 * and then reboots. If we are called twice, then we avoid trying to sync 750 * the disks as this often leads to recursive panics. 751 */ 752 void 753 panic(const char *fmt, ...) 754 { 755 int bootopt, newpanic; 756 globaldata_t gd = mycpu; 757 thread_t td = gd->gd_curthread; 758 __va_list ap; 759 static char buf[256]; 760 761 /* 762 * If a panic occurs on multiple cpus before the first is able to 763 * halt the other cpus, only one cpu is allowed to take the panic. 764 * Attempt to be verbose about this situation but if the kprintf() 765 * itself panics don't let us overrun the kernel stack. 766 * 767 * Be very nasty about descheduling our thread at the lowest 768 * level possible in an attempt to freeze the thread without 769 * inducing further panics. 770 * 771 * Bumping gd_trap_nesting_level will also bypass assertions in 772 * lwkt_switch() and allow us to switch away even if we are a 773 * FAST interrupt or IPI. 774 * 775 * The setting of panic_cpu_gd also determines how kprintf() 776 * spin-locks itself. DDB can set panic_cpu_gd as well. 777 */ 778 for (;;) { 779 globaldata_t xgd = panic_cpu_gd; 780 781 /* 782 * Someone else got the panic cpu 783 */ 784 if (xgd && xgd != gd) { 785 crit_enter(); 786 ++mycpu->gd_trap_nesting_level; 787 if (mycpu->gd_trap_nesting_level < 25) { 788 kprintf("SECONDARY PANIC ON CPU %d THREAD %p\n", 789 mycpu->gd_cpuid, td); 790 } 791 td->td_release = NULL; /* be a grinch */ 792 for (;;) { 793 lwkt_deschedule_self(td); 794 lwkt_switch(); 795 } 796 /* NOT REACHED */ 797 /* --mycpu->gd_trap_nesting_level */ 798 /* crit_exit() */ 799 } 800 801 /* 802 * Reentrant panic 803 */ 804 if (xgd && xgd == gd) 805 break; 806 807 /* 808 * We got it 809 */ 810 if (atomic_cmpset_ptr(&panic_cpu_gd, NULL, gd)) 811 break; 812 } 813 /* 814 * Try to get the system into a working state. Save information 815 * we are about to destroy. 816 */ 817 kvcreinitspin(); 818 if (panicstr == NULL) { 819 bcopy(td->td_toks_array, panic_tokens, sizeof(panic_tokens)); 820 panic_tokens_count = td->td_toks_stop - &td->td_toks_base; 821 } 822 lwkt_relalltokens(td); 823 td->td_toks_stop = &td->td_toks_base; 824 if (gd->gd_spinlocks) 825 kprintf("panic with %d spinlocks held\n", gd->gd_spinlocks); 826 gd->gd_spinlocks = 0; 827 828 /* 829 * Setup 830 */ 831 bootopt = RB_AUTOBOOT | RB_DUMP; 832 if (sync_on_panic == 0) 833 bootopt |= RB_NOSYNC; 834 newpanic = 0; 835 if (panicstr) { 836 bootopt |= RB_NOSYNC; 837 } else { 838 panicstr = fmt; 839 newpanic = 1; 840 } 841 842 /* 843 * Format the panic string. 844 */ 845 __va_start(ap, fmt); 846 kvsnprintf(buf, sizeof(buf), fmt, ap); 847 if (panicstr == fmt) 848 panicstr = buf; 849 __va_end(ap); 850 if (panic_notifier != NULL) 851 panic_notifier->notifier(panic_notifier->arg); 852 kprintf("panic: %s\n", buf); 853 /* two separate prints in case of an unmapped page and trap */ 854 kprintf("cpuid = %d\n", mycpu->gd_cpuid); 855 856 #if (NGPIO > 0) && defined(ERROR_LED_ON_PANIC) 857 led_switch("error", 1); 858 #endif 859 860 #if defined(WDOG_DISABLE_ON_PANIC) 861 wdog_disable(); 862 #endif 863 864 /* 865 * Make sure kgdb knows who we are, there won't be a stoppcbs[] 866 * entry since our cpu wasn't stopped. 867 */ 868 savectx(&dumppcb); 869 dumpthread = curthread; 870 871 /* 872 * Enter the debugger or fall through & dump. Entering the 873 * debugger will stop cpus. If not entering the debugger stop 874 * cpus here. 875 * 876 * Limit the trace history to leave more panic data on a 877 * potentially row-limited console. 878 */ 879 880 #if defined(DDB) 881 if (newpanic && trace_on_panic) 882 print_backtrace(6); 883 if (debugger_on_panic) 884 Debugger("panic"); 885 else 886 #endif 887 if (newpanic) 888 stop_cpus(mycpu->gd_other_cpus); 889 boot(bootopt); 890 } 891 892 /* 893 * Support for poweroff delay. 894 */ 895 #ifndef POWEROFF_DELAY 896 # define POWEROFF_DELAY 5000 897 #endif 898 static int poweroff_delay = POWEROFF_DELAY; 899 900 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW, 901 &poweroff_delay, 0, ""); 902 903 static void 904 poweroff_wait(void *junk, int howto) 905 { 906 if(!(howto & RB_POWEROFF) || poweroff_delay <= 0) 907 return; 908 DELAY(poweroff_delay * 1000); 909 } 910 911 /* 912 * Some system processes (e.g. syncer) need to be stopped at appropriate 913 * points in their main loops prior to a system shutdown, so that they 914 * won't interfere with the shutdown process (e.g. by holding a disk buf 915 * to cause sync to fail). For each of these system processes, register 916 * shutdown_kproc() as a handler for one of shutdown events. 917 */ 918 static int kproc_shutdown_wait = 60; 919 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW, 920 &kproc_shutdown_wait, 0, ""); 921 922 void 923 shutdown_kproc(void *arg, int howto) 924 { 925 struct thread *td; 926 struct proc *p; 927 int error; 928 929 if (panicstr) 930 return; 931 932 td = (struct thread *)arg; 933 if ((p = td->td_proc) != NULL) { 934 kprintf("Waiting (max %d seconds) for system process `%s' to stop...", 935 kproc_shutdown_wait, p->p_comm); 936 } else { 937 kprintf("Waiting (max %d seconds) for system thread %s to stop...", 938 kproc_shutdown_wait, td->td_comm); 939 } 940 error = suspend_kproc(td, kproc_shutdown_wait * hz); 941 942 if (error == EWOULDBLOCK) 943 kprintf("timed out\n"); 944 else 945 kprintf("stopped\n"); 946 } 947 948 /* Registration of dumpers */ 949 int 950 set_dumper(struct dumperinfo *di) 951 { 952 if (di == NULL) { 953 bzero(&dumper, sizeof(dumper)); 954 return 0; 955 } 956 957 if (dumper.dumper != NULL) 958 return (EBUSY); 959 960 dumper = *di; 961 return 0; 962 } 963 964 void 965 dumpsys(void) 966 { 967 #if defined (_KERNEL_VIRTUAL) 968 /* vkernels don't support dumps */ 969 kprintf("vkernels don't support dumps\n"); 970 return; 971 #endif 972 /* 973 * If there is a dumper registered and we aren't dumping already, call 974 * the machine dependent dumpsys (md_dumpsys) to do the hard work. 975 * 976 * XXX: while right now the md_dumpsys() of x86 and x86_64 could be 977 * factored out completely into here, I rather keep them machine 978 * dependent in case we ever add a platform which does not share 979 * the same dumpsys() code, such as arm. 980 */ 981 if (dumper.dumper != NULL && !dumping) { 982 dumping++; 983 md_dumpsys(&dumper); 984 } 985 } 986 987 __read_frequently int dump_stop_usertds = 0; 988 989 static 990 void 991 need_user_resched_remote(void *dummy) 992 { 993 need_user_resched(); 994 } 995 996 void 997 dump_reactivate_cpus(void) 998 { 999 globaldata_t gd; 1000 int cpu, seq; 1001 1002 dump_stop_usertds = 1; 1003 1004 need_user_resched(); 1005 1006 for (cpu = 0; cpu < ncpus; cpu++) { 1007 gd = globaldata_find(cpu); 1008 seq = lwkt_send_ipiq(gd, need_user_resched_remote, NULL); 1009 lwkt_wait_ipiq(gd, seq); 1010 } 1011 1012 restart_cpus(stopped_cpus); 1013 } 1014