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/acpica/acpi_pvpanic/panic_notifier.h> 89 #include <dev/misc/gpio/gpio.h> 90 91 #ifndef PANIC_REBOOT_WAIT_TIME 92 #define PANIC_REBOOT_WAIT_TIME 15 /* default to 15 seconds */ 93 #endif 94 95 /* 96 * Note that stdarg.h and the ANSI style va_start macro is used for both 97 * ANSI and traditional C compilers. We use the machine version to stay 98 * within the confines of the kernel header files. 99 */ 100 #include <machine/stdarg.h> 101 102 #ifdef DDB 103 #include <ddb/ddb.h> 104 #ifdef DDB_UNATTENDED 105 int debugger_on_panic = 0; 106 #else 107 int debugger_on_panic = 1; 108 #endif 109 SYSCTL_INT(_debug, OID_AUTO, debugger_on_panic, CTLFLAG_RW, 110 &debugger_on_panic, 0, "Run debugger on kernel panic"); 111 112 #ifdef DDB_TRACE 113 int trace_on_panic = 1; 114 #else 115 int trace_on_panic = 0; 116 #endif 117 SYSCTL_INT(_debug, OID_AUTO, trace_on_panic, CTLFLAG_RW, 118 &trace_on_panic, 0, "Print stack trace on kernel panic"); 119 #endif 120 121 static int sync_on_panic = 0; 122 SYSCTL_INT(_kern, OID_AUTO, sync_on_panic, CTLFLAG_RW, 123 &sync_on_panic, 0, "Do a sync before rebooting from a panic"); 124 125 SYSCTL_NODE(_kern, OID_AUTO, shutdown, CTLFLAG_RW, 0, "Shutdown environment"); 126 127 /* 128 * Variable panicstr contains argument to first call to panic; used as flag 129 * to indicate that the kernel has already called panic. 130 */ 131 const char *panicstr; 132 133 int dumping; /* system is dumping */ 134 static struct dumperinfo dumper; /* selected dumper */ 135 136 globaldata_t panic_cpu_gd; /* which cpu took the panic */ 137 struct lwkt_tokref panic_tokens[LWKT_MAXTOKENS]; 138 int panic_tokens_count; 139 140 int bootverbose = 0; /* note: assignment to force non-bss */ 141 SYSCTL_INT(_debug, OID_AUTO, bootverbose, CTLFLAG_RW, 142 &bootverbose, 0, "Verbose kernel messages"); 143 144 int cold = 1; /* note: assignment to force non-bss */ 145 int dumplo; /* OBSOLETE - savecore compat */ 146 u_int64_t dumplo64; 147 148 static void boot (int) __dead2; 149 static int setdumpdev (cdev_t dev); 150 static void poweroff_wait (void *, int); 151 static void print_uptime (void); 152 static void shutdown_halt (void *junk, int howto); 153 static void shutdown_panic (void *junk, int howto); 154 static void shutdown_reset (void *junk, int howto); 155 static int shutdown_busycount1(struct buf *bp, void *info); 156 static int shutdown_busycount2(struct buf *bp, void *info); 157 static void shutdown_cleanup_proc(struct proc *p); 158 159 /* register various local shutdown events */ 160 static void 161 shutdown_conf(void *unused) 162 { 163 EVENTHANDLER_REGISTER(shutdown_final, poweroff_wait, NULL, SHUTDOWN_PRI_FIRST); 164 EVENTHANDLER_REGISTER(shutdown_final, shutdown_halt, NULL, SHUTDOWN_PRI_LAST + 100); 165 EVENTHANDLER_REGISTER(shutdown_final, shutdown_panic, NULL, SHUTDOWN_PRI_LAST + 100); 166 EVENTHANDLER_REGISTER(shutdown_final, shutdown_reset, NULL, SHUTDOWN_PRI_LAST + 200); 167 } 168 169 SYSINIT(shutdown_conf, SI_BOOT2_MACHDEP, SI_ORDER_ANY, shutdown_conf, NULL); 170 171 /* ARGSUSED */ 172 173 /* 174 * The system call that results in a reboot 175 * 176 * MPALMOSTSAFE 177 */ 178 int 179 sys_reboot(struct reboot_args *uap) 180 { 181 struct thread *td = curthread; 182 int error; 183 184 if ((error = priv_check(td, PRIV_REBOOT))) 185 return (error); 186 187 get_mplock(); 188 boot(uap->opt); 189 rel_mplock(); 190 return (0); 191 } 192 193 /* 194 * Called by events that want to shut down.. e.g <CTL><ALT><DEL> on a PC 195 */ 196 static int shutdown_howto = 0; 197 198 void 199 shutdown_nice(int howto) 200 { 201 shutdown_howto = howto; 202 203 /* Send a signal to init(8) and have it shutdown the world */ 204 if (initproc != NULL) { 205 ksignal(initproc, SIGINT); 206 } else { 207 /* No init(8) running, so simply reboot */ 208 boot(RB_NOSYNC); 209 } 210 return; 211 } 212 static int waittime = -1; 213 struct pcb dumppcb; 214 struct thread *dumpthread; 215 216 static void 217 print_uptime(void) 218 { 219 int f; 220 struct timespec ts; 221 222 getnanouptime(&ts); 223 kprintf("Uptime: "); 224 f = 0; 225 if (ts.tv_sec >= 86400) { 226 kprintf("%ldd", ts.tv_sec / 86400); 227 ts.tv_sec %= 86400; 228 f = 1; 229 } 230 if (f || ts.tv_sec >= 3600) { 231 kprintf("%ldh", ts.tv_sec / 3600); 232 ts.tv_sec %= 3600; 233 f = 1; 234 } 235 if (f || ts.tv_sec >= 60) { 236 kprintf("%ldm", ts.tv_sec / 60); 237 ts.tv_sec %= 60; 238 f = 1; 239 } 240 kprintf("%lds\n", ts.tv_sec); 241 } 242 243 /* 244 * Go through the rigmarole of shutting down.. 245 * this used to be in machdep.c but I'll be dammned if I could see 246 * anything machine dependant in it. 247 */ 248 static void 249 boot(int howto) 250 { 251 /* 252 * Get rid of any user scheduler baggage and then give 253 * us a high priority. 254 */ 255 if (curthread->td_release) 256 curthread->td_release(curthread); 257 lwkt_setpri_self(TDPRI_MAX); 258 259 /* collect extra flags that shutdown_nice might have set */ 260 howto |= shutdown_howto; 261 262 /* 263 * We really want to shutdown on the BSP. Subsystems such as ACPI 264 * can't power-down the box otherwise. 265 */ 266 if (!CPUMASK_ISUP(smp_active_mask)) { 267 kprintf("boot() called on cpu#%d\n", mycpu->gd_cpuid); 268 } 269 if (panicstr == NULL && mycpu->gd_cpuid != 0) { 270 kprintf("Switching to cpu #0 for shutdown\n"); 271 lwkt_setcpu_self(globaldata_find(0)); 272 } 273 /* 274 * Do any callouts that should be done BEFORE syncing the filesystems. 275 */ 276 EVENTHANDLER_INVOKE(shutdown_pre_sync, howto); 277 278 /* 279 * Try to get rid of any remaining FS references. The calling 280 * process, proc0, and init may still hold references. The 281 * VFS cache subsystem may still hold a root reference to root. 282 * 283 * XXX this needs work. We really need to SIGSTOP all remaining 284 * processes in order to avoid blowups due to proc0's filesystem 285 * references going away. For now just make sure that the init 286 * process is stopped. 287 */ 288 if (panicstr == NULL) { 289 shutdown_cleanup_proc(curproc); 290 shutdown_cleanup_proc(&proc0); 291 if (initproc) { 292 if (initproc != curproc) { 293 ksignal(initproc, SIGSTOP); 294 tsleep(boot, 0, "shutdn", hz / 20); 295 } 296 shutdown_cleanup_proc(initproc); 297 } 298 vfs_cache_setroot(NULL, NULL); 299 } 300 301 /* 302 * Now sync filesystems 303 */ 304 if (!cold && (howto & RB_NOSYNC) == 0 && waittime < 0) { 305 int iter, nbusy, pbusy; 306 int zcount; 307 308 waittime = 0; 309 zcount = 0; 310 kprintf("\nsyncing disks... "); 311 312 sys_sync(NULL); 313 314 /* 315 * With soft updates, some buffers that are written will be 316 * remarked as dirty until other buffers are written. 317 * 318 * sys_sync() usually runs asynchronously, to give us a 319 * better chance of syncing the rest of the filesystems when 320 * one or more of them are stuck. 321 */ 322 for (iter = pbusy = 0; iter < 20 + zcount; iter++) { 323 if (iter <= 10) 324 nbusy = scan_all_buffers(shutdown_busycount1, 325 &iter); 326 else 327 nbusy = scan_all_buffers(shutdown_busycount2, 328 &iter); 329 kprintf("%d ", nbusy); 330 if (nbusy == 0) { 331 if (++zcount == 3) 332 break; 333 } else { 334 zcount = 0; 335 } 336 337 /* 338 * There could be a lot to sync, only allow iter to 339 * proceed while there is progress. 340 */ 341 if (nbusy < pbusy) { 342 if (iter > 10) 343 iter = 10; 344 else 345 iter = 0; 346 } 347 pbusy = nbusy; 348 349 /* 350 * XXX: 351 * Process soft update work queue if buffers don't sync 352 * after 6 iterations by permitting the syncer to run. 353 */ 354 if (iter > 5) 355 bio_ops_sync(NULL); 356 357 sys_sync(NULL); 358 tsleep(boot, 0, "shutdn", hz * iter / 20 + 1); 359 } 360 kprintf("\n"); 361 362 if (zcount < 3) { 363 /* 364 * Failed to sync all blocks. Indicate this and don't 365 * unmount filesystems (thus forcing an fsck on reboot). 366 */ 367 kprintf("giving up on %d buffers\n", nbusy); 368 #ifdef DDB 369 if (debugger_on_panic) 370 Debugger("busy buffer problem"); 371 #endif /* DDB */ 372 tsleep(boot, 0, "shutdn", hz * 5 + 1); 373 } else { 374 kprintf("done\n"); 375 376 /* 377 * Unmount filesystems 378 */ 379 if (panicstr == NULL) 380 vfs_unmountall(); 381 } 382 tsleep(boot, 0, "shutdn", hz / 10 + 1); 383 } 384 385 print_uptime(); 386 387 /* 388 * Dump before doing post_sync shutdown ops 389 */ 390 crit_enter(); 391 if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP && !cold) { 392 dumpsys(); 393 } 394 395 /* 396 * Ok, now do things that assume all filesystem activity has 397 * been completed. This will also call the device shutdown 398 * methods. 399 */ 400 EVENTHANDLER_INVOKE(shutdown_post_sync, howto); 401 402 /* Now that we're going to really halt the system... */ 403 EVENTHANDLER_INVOKE(shutdown_final, howto); 404 405 for(;;) ; /* safety against shutdown_reset not working */ 406 /* NOTREACHED */ 407 } 408 409 /* 410 * Pass 1 - Figure out if there are any busy or dirty buffers still present. 411 * 412 * We ignore TMPFS mounts in this pass. 413 */ 414 static int 415 shutdown_busycount1(struct buf *bp, void *info __unused) 416 { 417 struct vnode *vp; 418 419 if ((vp = bp->b_vp) != NULL && vp->v_tag == VT_TMPFS) 420 return (0); 421 if ((bp->b_flags & B_INVAL) == 0 && BUF_LOCKINUSE(bp)) 422 return(1); 423 if ((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) 424 return (1); 425 return (0); 426 } 427 428 /* 429 * Pass 2 - only run after pass 1 has completed or has given up 430 * 431 * We ignore TMPFS, NFS, MFS, and SMBFS mounts in this pass. 432 */ 433 static int 434 shutdown_busycount2(struct buf *bp, void *info) 435 { 436 struct vnode *vp; 437 int *iterp = info; 438 const char *mpath; 439 440 /* 441 * Ignore tmpfs and nfs mounts 442 */ 443 if ((vp = bp->b_vp) != NULL) { 444 if (vp->v_tag == VT_TMPFS) 445 return (0); 446 if (vp->v_tag == VT_NFS) 447 return (0); 448 if (vp->v_tag == VT_MFS) 449 return (0); 450 if (vp->v_tag == VT_SMBFS) 451 return (0); 452 } 453 454 /* 455 * Only count buffers stuck on I/O, ignore everything else 456 */ 457 if (((bp->b_flags & B_INVAL) == 0 && BUF_LOCKINUSE(bp)) || 458 ((bp->b_flags & (B_DELWRI|B_INVAL)) == B_DELWRI)) { 459 /* 460 * Only count buffers undergoing write I/O 461 * on the related vnode. 462 */ 463 if (bp->b_vp == NULL || 464 bio_track_active(&bp->b_vp->v_track_write) == 0) { 465 return (0); 466 } 467 if (*iterp > 15) { 468 mpath = "?"; 469 if (bp->b_vp->v_mount) 470 mpath = bp->b_vp->v_mount->mnt_stat.f_mntonname; 471 472 kprintf("%p on %s, flags:%08x, loffset:%jd, " 473 "doffset:%jd\n", 474 bp, 475 mpath, 476 bp->b_flags, 477 (intmax_t)bp->b_loffset, 478 (intmax_t)bp->b_bio2.bio_offset); 479 } 480 return(1); 481 } 482 return(0); 483 } 484 485 /* 486 * If the shutdown was a clean halt, behave accordingly. 487 */ 488 static void 489 shutdown_halt(void *junk, int howto) 490 { 491 if (howto & RB_HALT) { 492 kprintf("\n"); 493 kprintf("The operating system has halted.\n"); 494 #ifdef _KERNEL_VIRTUAL 495 cpu_halt(); 496 #else 497 kprintf("Please press any key to reboot.\n\n"); 498 switch (cngetc()) { 499 case -1: /* No console, just die */ 500 cpu_halt(); 501 /* NOTREACHED */ 502 default: 503 howto &= ~RB_HALT; 504 break; 505 } 506 #endif 507 } 508 } 509 510 /* 511 * Check to see if the system paniced, pause and then reboot 512 * according to the specified delay. 513 */ 514 static void 515 shutdown_panic(void *junk, int howto) 516 { 517 int loop; 518 int c; 519 520 if (howto & RB_DUMP) { 521 if (PANIC_REBOOT_WAIT_TIME != 0) { 522 if (PANIC_REBOOT_WAIT_TIME != -1) { 523 kprintf("Automatic reboot in %d seconds - " 524 "press a key on the console to abort\n", 525 PANIC_REBOOT_WAIT_TIME); 526 for (loop = PANIC_REBOOT_WAIT_TIME * 10; 527 loop > 0; --loop) { 528 DELAY(1000 * 100); /* 1/10th second */ 529 /* Did user type a key? */ 530 c = cncheckc(); 531 if (c != -1 && c != NOKEY) 532 break; 533 } 534 if (!loop) 535 return; 536 } 537 } else { /* zero time specified - reboot NOW */ 538 return; 539 } 540 kprintf("--> Press a key on the console to reboot,\n"); 541 kprintf("--> or switch off the system now.\n"); 542 cngetc(); 543 } 544 } 545 546 /* 547 * Everything done, now reset 548 */ 549 static void 550 shutdown_reset(void *junk, int howto) 551 { 552 kprintf("Rebooting...\n"); 553 DELAY(1000000); /* wait 1 sec for kprintf's to complete and be read */ 554 /* cpu_boot(howto); */ /* doesn't do anything at the moment */ 555 cpu_reset(); 556 /* NOTREACHED */ /* assuming reset worked */ 557 } 558 559 /* 560 * Try to remove FS references in the specified process. This function 561 * is used during shutdown 562 */ 563 static 564 void 565 shutdown_cleanup_proc(struct proc *p) 566 { 567 struct filedesc *fdp; 568 struct vmspace *vm; 569 570 if (p == NULL) 571 return; 572 if ((fdp = p->p_fd) != NULL) { 573 kern_closefrom(0); 574 if (fdp->fd_cdir) { 575 cache_drop(&fdp->fd_ncdir); 576 vrele(fdp->fd_cdir); 577 fdp->fd_cdir = NULL; 578 } 579 if (fdp->fd_rdir) { 580 cache_drop(&fdp->fd_nrdir); 581 vrele(fdp->fd_rdir); 582 fdp->fd_rdir = NULL; 583 } 584 if (fdp->fd_jdir) { 585 cache_drop(&fdp->fd_njdir); 586 vrele(fdp->fd_jdir); 587 fdp->fd_jdir = NULL; 588 } 589 } 590 if (p->p_vkernel) 591 vkernel_exit(p); 592 if (p->p_textvp) { 593 vrele(p->p_textvp); 594 p->p_textvp = NULL; 595 } 596 vm = p->p_vmspace; 597 if (vm != NULL) { 598 pmap_remove_pages(vmspace_pmap(vm), 599 VM_MIN_USER_ADDRESS, 600 VM_MAX_USER_ADDRESS); 601 vm_map_remove(&vm->vm_map, 602 VM_MIN_USER_ADDRESS, 603 VM_MAX_USER_ADDRESS); 604 } 605 } 606 607 /* 608 * Magic number for savecore 609 * 610 * exported (symorder) and used at least by savecore(8) 611 * 612 * Mark it as used so that gcc doesn't optimize it away. 613 */ 614 __attribute__((__used__)) 615 static u_long const dumpmag = 0x8fca0101UL; 616 617 __attribute__((__used__)) 618 static int dumpsize = 0; /* also for savecore */ 619 620 static int dodump = 1; 621 622 SYSCTL_INT(_machdep, OID_AUTO, do_dump, CTLFLAG_RW, &dodump, 0, 623 "Try to perform coredump on kernel panic"); 624 625 void 626 mkdumpheader(struct kerneldumpheader *kdh, char *magic, uint32_t archver, 627 uint64_t dumplen, uint32_t blksz) 628 { 629 bzero(kdh, sizeof(*kdh)); 630 strncpy(kdh->magic, magic, sizeof(kdh->magic)); 631 strncpy(kdh->architecture, MACHINE_ARCH, sizeof(kdh->architecture)); 632 kdh->version = htod32(KERNELDUMPVERSION); 633 kdh->architectureversion = htod32(archver); 634 kdh->dumplength = htod64(dumplen); 635 kdh->dumptime = htod64(time_second); 636 kdh->blocksize = htod32(blksz); 637 strncpy(kdh->hostname, hostname, sizeof(kdh->hostname)); 638 strncpy(kdh->versionstring, version, sizeof(kdh->versionstring)); 639 if (panicstr != NULL) 640 strncpy(kdh->panicstring, panicstr, sizeof(kdh->panicstring)); 641 kdh->parity = kerneldump_parity(kdh); 642 } 643 644 static int 645 setdumpdev(cdev_t dev) 646 { 647 int error; 648 int doopen; 649 650 if (dev == NULL) { 651 disk_dumpconf(NULL, 0/*off*/); 652 dumpdev = NULL; 653 return (0); 654 } 655 656 /* 657 * We have to open the device before we can perform ioctls on it, 658 * or the slice/label data may not be present. Device opens are 659 * usually tracked by specfs, but the dump device can be set in 660 * early boot and may not be open so this is somewhat of a hack. 661 */ 662 doopen = (dev->si_sysref.refcnt == 1); 663 if (doopen) { 664 error = dev_dopen(dev, FREAD, S_IFCHR, proc0.p_ucred, NULL); 665 if (error) 666 return (error); 667 } 668 error = disk_dumpconf(dev, 1/*on*/); 669 if (error == 0) 670 dumpdev = dev; 671 672 return error; 673 } 674 675 /* ARGSUSED */ 676 static void dump_conf (void *dummy); 677 static void 678 dump_conf(void *dummy) 679 { 680 char *path; 681 cdev_t dev; 682 int _dummy; 683 684 path = kmalloc(MNAMELEN, M_TEMP, M_WAITOK); 685 if (TUNABLE_STR_FETCH("dumpdev", path, MNAMELEN) != 0) { 686 /* 687 * Make sure all disk devices created so far have also been 688 * probed, and also make sure that the newly created device 689 * nodes for probed disks are ready, too. 690 * 691 * XXX - Delay an additional 2 seconds to help drivers which 692 * pickup devices asynchronously and are not caught by 693 * CAM's initial probe. 694 */ 695 sync_devs(); 696 tsleep(&_dummy, 0, "syncer", hz*2); 697 698 dev = kgetdiskbyname(path); 699 if (dev != NULL) 700 dumpdev = dev; 701 } 702 kfree(path, M_TEMP); 703 if (setdumpdev(dumpdev) != 0) 704 dumpdev = NULL; 705 } 706 707 SYSINIT(dump_conf, SI_SUB_DUMP_CONF, SI_ORDER_FIRST, dump_conf, NULL); 708 709 static int 710 sysctl_kern_dumpdev(SYSCTL_HANDLER_ARGS) 711 { 712 int error; 713 udev_t ndumpdev; 714 715 ndumpdev = dev2udev(dumpdev); 716 error = sysctl_handle_opaque(oidp, &ndumpdev, sizeof ndumpdev, req); 717 if (error == 0 && req->newptr != NULL) 718 error = setdumpdev(udev2dev(ndumpdev, 0)); 719 return (error); 720 } 721 722 SYSCTL_PROC(_kern, KERN_DUMPDEV, dumpdev, CTLTYPE_OPAQUE|CTLFLAG_RW, 723 0, sizeof dumpdev, sysctl_kern_dumpdev, "T,udev_t", ""); 724 725 static struct panicerinfo *panic_notifier; 726 727 int 728 set_panic_notifier(struct panicerinfo *info) 729 { 730 if (info == NULL) 731 panic_notifier = NULL; 732 else if (panic_notifier != NULL) 733 return 1; 734 else 735 panic_notifier = info; 736 737 return 0; 738 } 739 740 /* 741 * Panic is called on unresolvable fatal errors. It prints "panic: mesg", 742 * and then reboots. If we are called twice, then we avoid trying to sync 743 * the disks as this often leads to recursive panics. 744 */ 745 void 746 panic(const char *fmt, ...) 747 { 748 int bootopt, newpanic; 749 globaldata_t gd = mycpu; 750 thread_t td = gd->gd_curthread; 751 __va_list ap; 752 static char buf[256]; 753 754 /* 755 * If a panic occurs on multiple cpus before the first is able to 756 * halt the other cpus, only one cpu is allowed to take the panic. 757 * Attempt to be verbose about this situation but if the kprintf() 758 * itself panics don't let us overrun the kernel stack. 759 * 760 * Be very nasty about descheduling our thread at the lowest 761 * level possible in an attempt to freeze the thread without 762 * inducing further panics. 763 * 764 * Bumping gd_trap_nesting_level will also bypass assertions in 765 * lwkt_switch() and allow us to switch away even if we are a 766 * FAST interrupt or IPI. 767 * 768 * The setting of panic_cpu_gd also determines how kprintf() 769 * spin-locks itself. DDB can set panic_cpu_gd as well. 770 */ 771 for (;;) { 772 globaldata_t xgd = panic_cpu_gd; 773 774 /* 775 * Someone else got the panic cpu 776 */ 777 if (xgd && xgd != gd) { 778 crit_enter(); 779 ++mycpu->gd_trap_nesting_level; 780 if (mycpu->gd_trap_nesting_level < 25) { 781 kprintf("SECONDARY PANIC ON CPU %d THREAD %p\n", 782 mycpu->gd_cpuid, td); 783 } 784 td->td_release = NULL; /* be a grinch */ 785 for (;;) { 786 lwkt_deschedule_self(td); 787 lwkt_switch(); 788 } 789 /* NOT REACHED */ 790 /* --mycpu->gd_trap_nesting_level */ 791 /* crit_exit() */ 792 } 793 794 /* 795 * Reentrant panic 796 */ 797 if (xgd && xgd == gd) 798 break; 799 800 /* 801 * We got it 802 */ 803 if (atomic_cmpset_ptr(&panic_cpu_gd, NULL, gd)) 804 break; 805 } 806 /* 807 * Try to get the system into a working state. Save information 808 * we are about to destroy. 809 */ 810 kvcreinitspin(); 811 if (panicstr == NULL) { 812 bcopy(td->td_toks_array, panic_tokens, sizeof(panic_tokens)); 813 panic_tokens_count = td->td_toks_stop - &td->td_toks_base; 814 } 815 lwkt_relalltokens(td); 816 td->td_toks_stop = &td->td_toks_base; 817 if (gd->gd_spinlocks) 818 kprintf("panic with %d spinlocks held\n", gd->gd_spinlocks); 819 gd->gd_spinlocks = 0; 820 821 /* 822 * Setup 823 */ 824 bootopt = RB_AUTOBOOT | RB_DUMP; 825 if (sync_on_panic == 0) 826 bootopt |= RB_NOSYNC; 827 newpanic = 0; 828 if (panicstr) { 829 bootopt |= RB_NOSYNC; 830 } else { 831 panicstr = fmt; 832 newpanic = 1; 833 } 834 835 /* 836 * Format the panic string. 837 */ 838 __va_start(ap, fmt); 839 kvsnprintf(buf, sizeof(buf), fmt, ap); 840 if (panicstr == fmt) 841 panicstr = buf; 842 __va_end(ap); 843 if (panic_notifier != NULL) 844 panic_notifier->notifier(panic_notifier->arg); 845 kprintf("panic: %s\n", buf); 846 /* two separate prints in case of an unmapped page and trap */ 847 kprintf("cpuid = %d\n", mycpu->gd_cpuid); 848 849 #if (NGPIO > 0) && defined(ERROR_LED_ON_PANIC) 850 led_switch("error", 1); 851 #endif 852 853 #if defined(WDOG_DISABLE_ON_PANIC) 854 wdog_disable(); 855 #endif 856 857 /* 858 * Make sure kgdb knows who we are, there won't be a stoppcbs[] 859 * entry since our cpu wasn't stopped. 860 */ 861 savectx(&dumppcb); 862 dumpthread = curthread; 863 864 /* 865 * Enter the debugger or fall through & dump. Entering the 866 * debugger will stop cpus. If not entering the debugger stop 867 * cpus here. 868 * 869 * Limit the trace history to leave more panic data on a 870 * potentially row-limited console. 871 */ 872 873 #if defined(DDB) 874 if (newpanic && trace_on_panic) 875 print_backtrace(6); 876 if (debugger_on_panic) 877 Debugger("panic"); 878 else 879 #endif 880 if (newpanic) 881 stop_cpus(mycpu->gd_other_cpus); 882 boot(bootopt); 883 } 884 885 /* 886 * Support for poweroff delay. 887 */ 888 #ifndef POWEROFF_DELAY 889 # define POWEROFF_DELAY 5000 890 #endif 891 static int poweroff_delay = POWEROFF_DELAY; 892 893 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW, 894 &poweroff_delay, 0, ""); 895 896 static void 897 poweroff_wait(void *junk, int howto) 898 { 899 if(!(howto & RB_POWEROFF) || poweroff_delay <= 0) 900 return; 901 DELAY(poweroff_delay * 1000); 902 } 903 904 /* 905 * Some system processes (e.g. syncer) need to be stopped at appropriate 906 * points in their main loops prior to a system shutdown, so that they 907 * won't interfere with the shutdown process (e.g. by holding a disk buf 908 * to cause sync to fail). For each of these system processes, register 909 * shutdown_kproc() as a handler for one of shutdown events. 910 */ 911 static int kproc_shutdown_wait = 60; 912 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW, 913 &kproc_shutdown_wait, 0, ""); 914 915 void 916 shutdown_kproc(void *arg, int howto) 917 { 918 struct thread *td; 919 struct proc *p; 920 int error; 921 922 if (panicstr) 923 return; 924 925 td = (struct thread *)arg; 926 if ((p = td->td_proc) != NULL) { 927 kprintf("Waiting (max %d seconds) for system process `%s' to stop...", 928 kproc_shutdown_wait, p->p_comm); 929 } else { 930 kprintf("Waiting (max %d seconds) for system thread %s to stop...", 931 kproc_shutdown_wait, td->td_comm); 932 } 933 error = suspend_kproc(td, kproc_shutdown_wait * hz); 934 935 if (error == EWOULDBLOCK) 936 kprintf("timed out\n"); 937 else 938 kprintf("stopped\n"); 939 } 940 941 /* Registration of dumpers */ 942 int 943 set_dumper(struct dumperinfo *di) 944 { 945 if (di == NULL) { 946 bzero(&dumper, sizeof(dumper)); 947 return 0; 948 } 949 950 if (dumper.dumper != NULL) 951 return (EBUSY); 952 953 dumper = *di; 954 return 0; 955 } 956 957 void 958 dumpsys(void) 959 { 960 #if defined (_KERNEL_VIRTUAL) 961 /* vkernels don't support dumps */ 962 kprintf("vkernels don't support dumps\n"); 963 return; 964 #endif 965 /* 966 * If there is a dumper registered and we aren't dumping already, call 967 * the machine dependent dumpsys (md_dumpsys) to do the hard work. 968 * 969 * XXX: while right now the md_dumpsys() of x86 and x86_64 could be 970 * factored out completely into here, I rather keep them machine 971 * dependent in case we ever add a platform which does not share 972 * the same dumpsys() code, such as arm. 973 */ 974 if (dumper.dumper != NULL && !dumping) { 975 dumping++; 976 md_dumpsys(&dumper); 977 } 978 } 979 980 int dump_stop_usertds = 0; 981 982 static 983 void 984 need_user_resched_remote(void *dummy) 985 { 986 need_user_resched(); 987 } 988 989 void 990 dump_reactivate_cpus(void) 991 { 992 globaldata_t gd; 993 int cpu, seq; 994 995 dump_stop_usertds = 1; 996 997 need_user_resched(); 998 999 for (cpu = 0; cpu < ncpus; cpu++) { 1000 gd = globaldata_find(cpu); 1001 seq = lwkt_send_ipiq(gd, need_user_resched_remote, NULL); 1002 lwkt_wait_ipiq(gd, seq); 1003 } 1004 1005 restart_cpus(stopped_cpus); 1006 } 1007