1 /*- 2 * Copyright (c) 1986, 1988, 1991, 1993 3 * The Regents of the University of California. All rights reserved. 4 * (c) UNIX System Laboratories, Inc. 5 * All or some portions of this file are derived from material licensed 6 * to the University of California by American Telephone and Telegraph 7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8 * the permission of UNIX System Laboratories, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the University of 21 * California, Berkeley and its contributors. 22 * 4. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 * 38 * @(#)kern_shutdown.c 8.3 (Berkeley) 1/21/94 39 * $FreeBSD: src/sys/kern/kern_shutdown.c,v 1.72.2.12 2002/02/21 19:15:10 dillon Exp $ 40 * $DragonFly: src/sys/kern/kern_shutdown.c,v 1.62 2008/01/05 13:23:48 corecode Exp $ 41 */ 42 43 #include "opt_ddb.h" 44 #include "opt_ddb_trace.h" 45 #include "opt_hw_wdog.h" 46 #include "opt_panic.h" 47 #include "opt_show_busybufs.h" 48 #include "use_gpio.h" 49 50 #include <sys/param.h> 51 #include <sys/systm.h> 52 #include <sys/eventhandler.h> 53 #include <sys/buf.h> 54 #include <sys/disk.h> 55 #include <sys/diskslice.h> 56 #include <sys/reboot.h> 57 #include <sys/proc.h> 58 #include <sys/priv.h> 59 #include <sys/fcntl.h> /* FREAD */ 60 #include <sys/stat.h> /* S_IFCHR */ 61 #include <sys/vnode.h> 62 #include <sys/kernel.h> 63 #include <sys/kerneldump.h> 64 #include <sys/kthread.h> 65 #include <sys/malloc.h> 66 #include <sys/mount.h> 67 #include <sys/queue.h> 68 #include <sys/sysctl.h> 69 #include <sys/vkernel.h> 70 #include <sys/conf.h> 71 #include <sys/sysproto.h> 72 #include <sys/device.h> 73 #include <sys/cons.h> 74 #include <sys/shm.h> 75 #include <sys/kerneldump.h> 76 #include <sys/kern_syscall.h> 77 #include <vm/vm_map.h> 78 #include <vm/pmap.h> 79 80 #include <sys/thread2.h> 81 #include <sys/buf2.h> 82 #include <sys/mplock2.h> 83 84 #include <machine/clock.h> 85 #include <machine/md_var.h> 86 #include <machine/smp.h> /* smp_active_mask, cpuid */ 87 #include <machine/vmparam.h> 88 #include <machine/thread.h> 89 90 #include <sys/signalvar.h> 91 92 #include <sys/wdog.h> 93 #include <dev/misc/gpio/gpio.h> 94 95 #ifndef PANIC_REBOOT_WAIT_TIME 96 #define PANIC_REBOOT_WAIT_TIME 15 /* default to 15 seconds */ 97 #endif 98 99 /* 100 * Note that stdarg.h and the ANSI style va_start macro is used for both 101 * ANSI and traditional C compilers. We use the machine version to stay 102 * within the confines of the kernel header files. 103 */ 104 #include <machine/stdarg.h> 105 106 #ifdef DDB 107 #include <ddb/ddb.h> 108 #ifdef DDB_UNATTENDED 109 int debugger_on_panic = 0; 110 #else 111 int debugger_on_panic = 1; 112 #endif 113 SYSCTL_INT(_debug, OID_AUTO, debugger_on_panic, CTLFLAG_RW, 114 &debugger_on_panic, 0, "Run debugger on kernel panic"); 115 116 #ifdef DDB_TRACE 117 int trace_on_panic = 1; 118 #else 119 int trace_on_panic = 0; 120 #endif 121 SYSCTL_INT(_debug, OID_AUTO, trace_on_panic, CTLFLAG_RW, 122 &trace_on_panic, 0, "Print stack trace on kernel panic"); 123 #endif 124 125 static int sync_on_panic = 0; 126 SYSCTL_INT(_kern, OID_AUTO, sync_on_panic, CTLFLAG_RW, 127 &sync_on_panic, 0, "Do a sync before rebooting from a panic"); 128 129 SYSCTL_NODE(_kern, OID_AUTO, shutdown, CTLFLAG_RW, 0, "Shutdown environment"); 130 131 #ifdef HW_WDOG 132 /* 133 * If there is a hardware watchdog, point this at the function needed to 134 * hold it off. 135 * It's needed when the kernel needs to do some lengthy operations. 136 * e.g. in wd.c when dumping core.. It's most annoying to have 137 * your precious core-dump only half written because the wdog kicked in. 138 */ 139 watchdog_tickle_fn wdog_tickler = NULL; 140 #endif /* HW_WDOG */ 141 142 /* 143 * Variable panicstr contains argument to first call to panic; used as flag 144 * to indicate that the kernel has already called panic. 145 */ 146 const char *panicstr; 147 148 int dumping; /* system is dumping */ 149 static struct dumperinfo dumper; /* selected dumper */ 150 151 globaldata_t panic_cpu_gd; /* which cpu took the panic */ 152 struct lwkt_tokref panic_tokens[LWKT_MAXTOKENS]; 153 int panic_tokens_count; 154 155 int bootverbose = 0; /* note: assignment to force non-bss */ 156 SYSCTL_INT(_debug, OID_AUTO, bootverbose, CTLFLAG_RW, 157 &bootverbose, 0, "Verbose kernel messages"); 158 159 int cold = 1; /* note: assignment to force non-bss */ 160 int dumplo; /* OBSOLETE - savecore compat */ 161 u_int64_t dumplo64; 162 163 static void boot (int) __dead2; 164 static int setdumpdev (cdev_t dev); 165 static void poweroff_wait (void *, int); 166 static void print_uptime (void); 167 static void shutdown_halt (void *junk, int howto); 168 static void shutdown_panic (void *junk, int howto); 169 static void shutdown_reset (void *junk, int howto); 170 static int shutdown_busycount1(struct buf *bp, void *info); 171 static int shutdown_busycount2(struct buf *bp, void *info); 172 static void shutdown_cleanup_proc(struct proc *p); 173 174 /* register various local shutdown events */ 175 static void 176 shutdown_conf(void *unused) 177 { 178 EVENTHANDLER_REGISTER(shutdown_final, poweroff_wait, NULL, SHUTDOWN_PRI_FIRST); 179 EVENTHANDLER_REGISTER(shutdown_final, shutdown_halt, NULL, SHUTDOWN_PRI_LAST + 100); 180 EVENTHANDLER_REGISTER(shutdown_final, shutdown_panic, NULL, SHUTDOWN_PRI_LAST + 100); 181 EVENTHANDLER_REGISTER(shutdown_final, shutdown_reset, NULL, SHUTDOWN_PRI_LAST + 200); 182 } 183 184 SYSINIT(shutdown_conf, SI_BOOT2_MACHDEP, SI_ORDER_ANY, shutdown_conf, NULL) 185 186 /* ARGSUSED */ 187 188 /* 189 * The system call that results in a reboot 190 * 191 * MPALMOSTSAFE 192 */ 193 int 194 sys_reboot(struct reboot_args *uap) 195 { 196 struct thread *td = curthread; 197 int error; 198 199 if ((error = priv_check(td, PRIV_REBOOT))) 200 return (error); 201 202 get_mplock(); 203 boot(uap->opt); 204 rel_mplock(); 205 return (0); 206 } 207 208 /* 209 * Called by events that want to shut down.. e.g <CTL><ALT><DEL> on a PC 210 */ 211 static int shutdown_howto = 0; 212 213 void 214 shutdown_nice(int howto) 215 { 216 shutdown_howto = howto; 217 218 /* Send a signal to init(8) and have it shutdown the world */ 219 if (initproc != NULL) { 220 ksignal(initproc, SIGINT); 221 } else { 222 /* No init(8) running, so simply reboot */ 223 boot(RB_NOSYNC); 224 } 225 return; 226 } 227 static int waittime = -1; 228 struct pcb dumppcb; 229 struct thread *dumpthread; 230 231 static void 232 print_uptime(void) 233 { 234 int f; 235 struct timespec ts; 236 237 getnanouptime(&ts); 238 kprintf("Uptime: "); 239 f = 0; 240 if (ts.tv_sec >= 86400) { 241 kprintf("%ldd", ts.tv_sec / 86400); 242 ts.tv_sec %= 86400; 243 f = 1; 244 } 245 if (f || ts.tv_sec >= 3600) { 246 kprintf("%ldh", ts.tv_sec / 3600); 247 ts.tv_sec %= 3600; 248 f = 1; 249 } 250 if (f || ts.tv_sec >= 60) { 251 kprintf("%ldm", ts.tv_sec / 60); 252 ts.tv_sec %= 60; 253 f = 1; 254 } 255 kprintf("%lds\n", ts.tv_sec); 256 } 257 258 /* 259 * Go through the rigmarole of shutting down.. 260 * this used to be in machdep.c but I'll be dammned if I could see 261 * anything machine dependant in it. 262 */ 263 static void 264 boot(int howto) 265 { 266 /* 267 * Get rid of any user scheduler baggage and then give 268 * us a high priority. 269 */ 270 if (curthread->td_release) 271 curthread->td_release(curthread); 272 lwkt_setpri_self(TDPRI_MAX); 273 274 /* collect extra flags that shutdown_nice might have set */ 275 howto |= shutdown_howto; 276 277 #ifdef SMP 278 /* 279 * We really want to shutdown on the BSP. Subsystems such as ACPI 280 * can't power-down the box otherwise. 281 */ 282 if (smp_active_mask > 1) { 283 kprintf("boot() called on cpu#%d\n", mycpu->gd_cpuid); 284 } 285 if (panicstr == NULL && mycpu->gd_cpuid != 0) { 286 kprintf("Switching to cpu #0 for shutdown\n"); 287 lwkt_setcpu_self(globaldata_find(0)); 288 } 289 #endif 290 /* 291 * Do any callouts that should be done BEFORE syncing the filesystems. 292 */ 293 EVENTHANDLER_INVOKE(shutdown_pre_sync, howto); 294 295 /* 296 * Try to get rid of any remaining FS references. The calling 297 * process, proc0, and init may still hold references. The 298 * VFS cache subsystem may still hold a root reference to root. 299 * 300 * XXX this needs work. We really need to SIGSTOP all remaining 301 * processes in order to avoid blowups due to proc0's filesystem 302 * references going away. For now just make sure that the init 303 * process is stopped. 304 */ 305 if (panicstr == NULL) { 306 shutdown_cleanup_proc(curproc); 307 shutdown_cleanup_proc(&proc0); 308 if (initproc) { 309 if (initproc != curproc) { 310 ksignal(initproc, SIGSTOP); 311 tsleep(boot, 0, "shutdn", hz / 20); 312 } 313 shutdown_cleanup_proc(initproc); 314 } 315 vfs_cache_setroot(NULL, NULL); 316 } 317 318 /* 319 * Now sync filesystems 320 */ 321 if (!cold && (howto & RB_NOSYNC) == 0 && waittime < 0) { 322 int iter, nbusy, pbusy; 323 324 waittime = 0; 325 kprintf("\nsyncing disks... "); 326 327 sys_sync(NULL); /* YYY was sync(&proc0, NULL). why proc0 ? */ 328 329 /* 330 * With soft updates, some buffers that are 331 * written will be remarked as dirty until other 332 * buffers are written. 333 */ 334 for (iter = pbusy = 0; iter < 20; iter++) { 335 nbusy = scan_all_buffers(shutdown_busycount1, NULL); 336 if (nbusy == 0) 337 break; 338 kprintf("%d ", nbusy); 339 if (nbusy < pbusy) 340 iter = 0; 341 pbusy = nbusy; 342 /* 343 * XXX: 344 * Process soft update work queue if buffers don't sync 345 * after 6 iterations by permitting the syncer to run. 346 */ 347 if (iter > 5) 348 bio_ops_sync(NULL); 349 350 sys_sync(NULL); /* YYY was sync(&proc0, NULL). why proc0 ? */ 351 tsleep(boot, 0, "shutdn", hz * iter / 20 + 1); 352 } 353 kprintf("\n"); 354 /* 355 * Count only busy local buffers to prevent forcing 356 * a fsck if we're just a client of a wedged NFS server 357 */ 358 nbusy = scan_all_buffers(shutdown_busycount2, NULL); 359 if (nbusy) { 360 /* 361 * Failed to sync all blocks. Indicate this and don't 362 * unmount filesystems (thus forcing an fsck on reboot). 363 */ 364 kprintf("giving up on %d buffers\n", nbusy); 365 #ifdef DDB 366 Debugger("busy buffer problem"); 367 #endif /* DDB */ 368 tsleep(boot, 0, "shutdn", hz * 5 + 1); 369 } else { 370 kprintf("done\n"); 371 /* 372 * Unmount filesystems 373 */ 374 if (panicstr == NULL) 375 vfs_unmountall(); 376 } 377 tsleep(boot, 0, "shutdn", hz / 10 + 1); 378 } 379 380 print_uptime(); 381 382 /* 383 * Dump before doing post_sync shutdown ops 384 */ 385 crit_enter(); 386 if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP && !cold) { 387 dumpsys(); 388 } 389 390 /* 391 * Ok, now do things that assume all filesystem activity has 392 * been completed. This will also call the device shutdown 393 * methods. 394 */ 395 EVENTHANDLER_INVOKE(shutdown_post_sync, howto); 396 397 /* Now that we're going to really halt the system... */ 398 EVENTHANDLER_INVOKE(shutdown_final, howto); 399 400 for(;;) ; /* safety against shutdown_reset not working */ 401 /* NOTREACHED */ 402 } 403 404 static int 405 shutdown_busycount1(struct buf *bp, void *info) 406 { 407 if ((bp->b_flags & B_INVAL) == 0 && BUF_REFCNT(bp) > 0) 408 return(1); 409 if ((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) 410 return (1); 411 return (0); 412 } 413 414 static int 415 shutdown_busycount2(struct buf *bp, void *info) 416 { 417 if (((bp->b_flags & B_INVAL) == 0 && BUF_REFCNT(bp)) || 418 ((bp->b_flags & (B_DELWRI|B_INVAL)) == B_DELWRI)) { 419 /* 420 * Only count buffers undergoing write I/O 421 * on the related vnode. 422 */ 423 if (bp->b_vp == NULL || 424 bio_track_active(&bp->b_vp->v_track_write) == 0) { 425 return (0); 426 } 427 #if defined(SHOW_BUSYBUFS) || defined(DIAGNOSTIC) 428 kprintf( 429 "%p dev:?, flags:%08x, loffset:%jd, doffset:%jd\n", 430 bp, 431 bp->b_flags, (intmax_t)bp->b_loffset, 432 (intmax_t)bp->b_bio2.bio_offset); 433 #endif 434 return(1); 435 } 436 return(0); 437 } 438 439 /* 440 * If the shutdown was a clean halt, behave accordingly. 441 */ 442 static void 443 shutdown_halt(void *junk, int howto) 444 { 445 if (howto & RB_HALT) { 446 kprintf("\n"); 447 kprintf("The operating system has halted.\n"); 448 #ifdef _KERNEL_VIRTUAL 449 cpu_halt(); 450 #else 451 kprintf("Please press any key to reboot.\n\n"); 452 switch (cngetc()) { 453 case -1: /* No console, just die */ 454 cpu_halt(); 455 /* NOTREACHED */ 456 default: 457 howto &= ~RB_HALT; 458 break; 459 } 460 #endif 461 } 462 } 463 464 /* 465 * Check to see if the system paniced, pause and then reboot 466 * according to the specified delay. 467 */ 468 static void 469 shutdown_panic(void *junk, int howto) 470 { 471 int loop; 472 473 if (howto & RB_DUMP) { 474 if (PANIC_REBOOT_WAIT_TIME != 0) { 475 if (PANIC_REBOOT_WAIT_TIME != -1) { 476 kprintf("Automatic reboot in %d seconds - " 477 "press a key on the console to abort\n", 478 PANIC_REBOOT_WAIT_TIME); 479 for (loop = PANIC_REBOOT_WAIT_TIME * 10; 480 loop > 0; --loop) { 481 DELAY(1000 * 100); /* 1/10th second */ 482 /* Did user type a key? */ 483 if (cncheckc() != -1) 484 break; 485 } 486 if (!loop) 487 return; 488 } 489 } else { /* zero time specified - reboot NOW */ 490 return; 491 } 492 kprintf("--> Press a key on the console to reboot,\n"); 493 kprintf("--> or switch off the system now.\n"); 494 cngetc(); 495 } 496 } 497 498 /* 499 * Everything done, now reset 500 */ 501 static void 502 shutdown_reset(void *junk, int howto) 503 { 504 kprintf("Rebooting...\n"); 505 DELAY(1000000); /* wait 1 sec for kprintf's to complete and be read */ 506 /* cpu_boot(howto); */ /* doesn't do anything at the moment */ 507 cpu_reset(); 508 /* NOTREACHED */ /* assuming reset worked */ 509 } 510 511 /* 512 * Try to remove FS references in the specified process. This function 513 * is used during shutdown 514 */ 515 static 516 void 517 shutdown_cleanup_proc(struct proc *p) 518 { 519 struct filedesc *fdp; 520 struct vmspace *vm; 521 522 if (p == NULL) 523 return; 524 if ((fdp = p->p_fd) != NULL) { 525 kern_closefrom(0); 526 if (fdp->fd_cdir) { 527 cache_drop(&fdp->fd_ncdir); 528 vrele(fdp->fd_cdir); 529 fdp->fd_cdir = NULL; 530 } 531 if (fdp->fd_rdir) { 532 cache_drop(&fdp->fd_nrdir); 533 vrele(fdp->fd_rdir); 534 fdp->fd_rdir = NULL; 535 } 536 if (fdp->fd_jdir) { 537 cache_drop(&fdp->fd_njdir); 538 vrele(fdp->fd_jdir); 539 fdp->fd_jdir = NULL; 540 } 541 } 542 if (p->p_vkernel) 543 vkernel_exit(p); 544 if (p->p_textvp) { 545 vrele(p->p_textvp); 546 p->p_textvp = NULL; 547 } 548 vm = p->p_vmspace; 549 if (vm != NULL) { 550 pmap_remove_pages(vmspace_pmap(vm), 551 VM_MIN_USER_ADDRESS, 552 VM_MAX_USER_ADDRESS); 553 vm_map_remove(&vm->vm_map, 554 VM_MIN_USER_ADDRESS, 555 VM_MAX_USER_ADDRESS); 556 } 557 } 558 559 /* 560 * Magic number for savecore 561 * 562 * exported (symorder) and used at least by savecore(8) 563 * 564 * Mark it as used so that gcc doesn't optimize it away. 565 */ 566 __attribute__((__used__)) 567 static u_long const dumpmag = 0x8fca0101UL; 568 569 __attribute__((__used__)) 570 static int dumpsize = 0; /* also for savecore */ 571 572 static int dodump = 1; 573 574 SYSCTL_INT(_machdep, OID_AUTO, do_dump, CTLFLAG_RW, &dodump, 0, 575 "Try to perform coredump on kernel panic"); 576 577 void 578 mkdumpheader(struct kerneldumpheader *kdh, char *magic, uint32_t archver, 579 uint64_t dumplen, uint32_t blksz) 580 { 581 bzero(kdh, sizeof(*kdh)); 582 strncpy(kdh->magic, magic, sizeof(kdh->magic)); 583 strncpy(kdh->architecture, MACHINE_ARCH, sizeof(kdh->architecture)); 584 kdh->version = htod32(KERNELDUMPVERSION); 585 kdh->architectureversion = htod32(archver); 586 kdh->dumplength = htod64(dumplen); 587 kdh->dumptime = htod64(time_second); 588 kdh->blocksize = htod32(blksz); 589 strncpy(kdh->hostname, hostname, sizeof(kdh->hostname)); 590 strncpy(kdh->versionstring, version, sizeof(kdh->versionstring)); 591 if (panicstr != NULL) 592 strncpy(kdh->panicstring, panicstr, sizeof(kdh->panicstring)); 593 kdh->parity = kerneldump_parity(kdh); 594 } 595 596 static int 597 setdumpdev(cdev_t dev) 598 { 599 int error; 600 int doopen; 601 602 if (dev == NULL) { 603 disk_dumpconf(NULL, 0/*off*/); 604 return (0); 605 } 606 607 /* 608 * We have to open the device before we can perform ioctls on it, 609 * or the slice/label data may not be present. Device opens are 610 * usually tracked by specfs, but the dump device can be set in 611 * early boot and may not be open so this is somewhat of a hack. 612 */ 613 doopen = (dev->si_sysref.refcnt == 1); 614 if (doopen) { 615 error = dev_dopen(dev, FREAD, S_IFCHR, proc0.p_ucred); 616 if (error) 617 return (error); 618 } 619 error = disk_dumpconf(dev, 1/*on*/); 620 621 return error; 622 } 623 624 /* ARGSUSED */ 625 static void dump_conf (void *dummy); 626 static void 627 dump_conf(void *dummy) 628 { 629 char *path; 630 cdev_t dev; 631 632 path = kmalloc(MNAMELEN, M_TEMP, M_WAITOK); 633 if (TUNABLE_STR_FETCH("dumpdev", path, MNAMELEN) != 0) { 634 dev = kgetdiskbyname(path); 635 if (dev != NULL) 636 dumpdev = dev; 637 } 638 kfree(path, M_TEMP); 639 if (setdumpdev(dumpdev) != 0) 640 dumpdev = NULL; 641 } 642 643 SYSINIT(dump_conf, SI_SUB_DUMP_CONF, SI_ORDER_FIRST, dump_conf, NULL) 644 645 static int 646 sysctl_kern_dumpdev(SYSCTL_HANDLER_ARGS) 647 { 648 int error; 649 udev_t ndumpdev; 650 651 ndumpdev = dev2udev(dumpdev); 652 error = sysctl_handle_opaque(oidp, &ndumpdev, sizeof ndumpdev, req); 653 if (error == 0 && req->newptr != NULL) 654 error = setdumpdev(udev2dev(ndumpdev, 0)); 655 return (error); 656 } 657 658 SYSCTL_PROC(_kern, KERN_DUMPDEV, dumpdev, CTLTYPE_OPAQUE|CTLFLAG_RW, 659 0, sizeof dumpdev, sysctl_kern_dumpdev, "T,udev_t", ""); 660 661 /* 662 * Panic is called on unresolvable fatal errors. It prints "panic: mesg", 663 * and then reboots. If we are called twice, then we avoid trying to sync 664 * the disks as this often leads to recursive panics. 665 */ 666 void 667 panic(const char *fmt, ...) 668 { 669 int bootopt, newpanic; 670 globaldata_t gd = mycpu; 671 thread_t td = gd->gd_curthread; 672 __va_list ap; 673 static char buf[256]; 674 675 #ifdef SMP 676 /* 677 * If a panic occurs on multiple cpus before the first is able to 678 * halt the other cpus, only one cpu is allowed to take the panic. 679 * Attempt to be verbose about this situation but if the kprintf() 680 * itself panics don't let us overrun the kernel stack. 681 * 682 * Be very nasty about descheduling our thread at the lowest 683 * level possible in an attempt to freeze the thread without 684 * inducing further panics. 685 * 686 * Bumping gd_trap_nesting_level will also bypass assertions in 687 * lwkt_switch() and allow us to switch away even if we are a 688 * FAST interrupt or IPI. 689 * 690 * The setting of panic_cpu_gd also determines how kprintf() 691 * spin-locks itself. DDB can set panic_cpu_gd as well. 692 */ 693 for (;;) { 694 globaldata_t xgd = panic_cpu_gd; 695 696 /* 697 * Someone else got the panic cpu 698 */ 699 if (xgd && xgd != gd) { 700 crit_enter(); 701 ++mycpu->gd_trap_nesting_level; 702 if (mycpu->gd_trap_nesting_level < 25) { 703 kprintf("SECONDARY PANIC ON CPU %d THREAD %p\n", 704 mycpu->gd_cpuid, td); 705 } 706 td->td_release = NULL; /* be a grinch */ 707 for (;;) { 708 lwkt_deschedule_self(td); 709 lwkt_switch(); 710 } 711 /* NOT REACHED */ 712 /* --mycpu->gd_trap_nesting_level */ 713 /* crit_exit() */ 714 } 715 716 /* 717 * Reentrant panic 718 */ 719 if (xgd && xgd == gd) 720 break; 721 722 /* 723 * We got it 724 */ 725 if (atomic_cmpset_ptr(&panic_cpu_gd, NULL, gd)) 726 break; 727 } 728 #else 729 panic_cpu_gd = gd; 730 #endif 731 /* 732 * Try to get the system into a working state. Save information 733 * we are about to destroy. 734 */ 735 kvcreinitspin(); 736 if (panicstr == NULL) { 737 bcopy(td->td_toks_array, panic_tokens, sizeof(panic_tokens)); 738 panic_tokens_count = td->td_toks_stop - &td->td_toks_base; 739 } 740 lwkt_relalltokens(td); 741 td->td_toks_stop = &td->td_toks_base; 742 743 /* 744 * Setup 745 */ 746 bootopt = RB_AUTOBOOT | RB_DUMP; 747 if (sync_on_panic == 0) 748 bootopt |= RB_NOSYNC; 749 newpanic = 0; 750 if (panicstr) { 751 bootopt |= RB_NOSYNC; 752 } else { 753 panicstr = fmt; 754 newpanic = 1; 755 } 756 757 /* 758 * Format the panic string. 759 */ 760 __va_start(ap, fmt); 761 kvsnprintf(buf, sizeof(buf), fmt, ap); 762 if (panicstr == fmt) 763 panicstr = buf; 764 __va_end(ap); 765 kprintf("panic: %s\n", buf); 766 #ifdef SMP 767 /* two separate prints in case of an unmapped page and trap */ 768 kprintf("mp_lock = %08x; ", mp_lock); 769 kprintf("cpuid = %d\n", mycpu->gd_cpuid); 770 #endif 771 772 #if (NGPIO > 0) && defined(ERROR_LED_ON_PANIC) 773 led_switch("error", 1); 774 #endif 775 776 #if defined(WDOG_DISABLE_ON_PANIC) && defined(WATCHDOG_ENABLE) 777 wdog_disable(); 778 #endif 779 780 #if defined(DDB) 781 if (newpanic && trace_on_panic) 782 print_backtrace(-1); 783 if (debugger_on_panic) 784 Debugger("panic"); 785 #endif 786 boot(bootopt); 787 } 788 789 /* 790 * Support for poweroff delay. 791 */ 792 #ifndef POWEROFF_DELAY 793 # define POWEROFF_DELAY 5000 794 #endif 795 static int poweroff_delay = POWEROFF_DELAY; 796 797 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW, 798 &poweroff_delay, 0, ""); 799 800 static void 801 poweroff_wait(void *junk, int howto) 802 { 803 if(!(howto & RB_POWEROFF) || poweroff_delay <= 0) 804 return; 805 DELAY(poweroff_delay * 1000); 806 } 807 808 /* 809 * Some system processes (e.g. syncer) need to be stopped at appropriate 810 * points in their main loops prior to a system shutdown, so that they 811 * won't interfere with the shutdown process (e.g. by holding a disk buf 812 * to cause sync to fail). For each of these system processes, register 813 * shutdown_kproc() as a handler for one of shutdown events. 814 */ 815 static int kproc_shutdown_wait = 60; 816 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW, 817 &kproc_shutdown_wait, 0, ""); 818 819 void 820 shutdown_kproc(void *arg, int howto) 821 { 822 struct thread *td; 823 struct proc *p; 824 int error; 825 826 if (panicstr) 827 return; 828 829 td = (struct thread *)arg; 830 if ((p = td->td_proc) != NULL) { 831 kprintf("Waiting (max %d seconds) for system process `%s' to stop...", 832 kproc_shutdown_wait, p->p_comm); 833 } else { 834 kprintf("Waiting (max %d seconds) for system thread %s to stop...", 835 kproc_shutdown_wait, td->td_comm); 836 } 837 error = suspend_kproc(td, kproc_shutdown_wait * hz); 838 839 if (error == EWOULDBLOCK) 840 kprintf("timed out\n"); 841 else 842 kprintf("stopped\n"); 843 } 844 845 /* Registration of dumpers */ 846 int 847 set_dumper(struct dumperinfo *di) 848 { 849 if (di == NULL) { 850 bzero(&dumper, sizeof(dumper)); 851 return 0; 852 } 853 854 if (dumper.dumper != NULL) 855 return (EBUSY); 856 857 dumper = *di; 858 return 0; 859 } 860 861 void 862 dumpsys(void) 863 { 864 #if defined (_KERNEL_VIRTUAL) 865 /* VKERNELs don't support dumps */ 866 kprintf("VKERNEL doesn't support dumps\n"); 867 return; 868 #endif 869 /* 870 * If there is a dumper registered and we aren't dumping already, call 871 * the machine dependent dumpsys (md_dumpsys) to do the hard work. 872 * 873 * XXX: while right now the md_dumpsys() of x86 and x86_64 could be 874 * factored out completely into here, I rather keep them machine 875 * dependent in case we ever add a platform which does not share 876 * the same dumpsys() code, such as arm. 877 */ 878 if (dumper.dumper != NULL && !dumping) { 879 dumping++; 880 md_dumpsys(&dumper); 881 } 882 } 883