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/cpu.h> 85 #include <machine/clock.h> 86 #include <machine/md_var.h> 87 #include <machine/smp.h> /* smp_active_mask, cpuid */ 88 #include <machine/vmparam.h> 89 #include <machine/thread.h> 90 91 #include <sys/signalvar.h> 92 93 #include <sys/wdog.h> 94 #include <dev/misc/gpio/gpio.h> 95 96 #ifndef PANIC_REBOOT_WAIT_TIME 97 #define PANIC_REBOOT_WAIT_TIME 15 /* default to 15 seconds */ 98 #endif 99 100 /* 101 * Note that stdarg.h and the ANSI style va_start macro is used for both 102 * ANSI and traditional C compilers. We use the machine version to stay 103 * within the confines of the kernel header files. 104 */ 105 #include <machine/stdarg.h> 106 107 #ifdef DDB 108 #include <ddb/ddb.h> 109 #ifdef DDB_UNATTENDED 110 int debugger_on_panic = 0; 111 #else 112 int debugger_on_panic = 1; 113 #endif 114 SYSCTL_INT(_debug, OID_AUTO, debugger_on_panic, CTLFLAG_RW, 115 &debugger_on_panic, 0, "Run debugger on kernel panic"); 116 117 #ifdef DDB_TRACE 118 int trace_on_panic = 1; 119 #else 120 int trace_on_panic = 0; 121 #endif 122 SYSCTL_INT(_debug, OID_AUTO, trace_on_panic, CTLFLAG_RW, 123 &trace_on_panic, 0, "Print stack trace on kernel panic"); 124 #endif 125 126 static int sync_on_panic = 0; 127 SYSCTL_INT(_kern, OID_AUTO, sync_on_panic, CTLFLAG_RW, 128 &sync_on_panic, 0, "Do a sync before rebooting from a panic"); 129 130 SYSCTL_NODE(_kern, OID_AUTO, shutdown, CTLFLAG_RW, 0, "Shutdown environment"); 131 132 #ifdef HW_WDOG 133 /* 134 * If there is a hardware watchdog, point this at the function needed to 135 * hold it off. 136 * It's needed when the kernel needs to do some lengthy operations. 137 * e.g. in wd.c when dumping core.. It's most annoying to have 138 * your precious core-dump only half written because the wdog kicked in. 139 */ 140 watchdog_tickle_fn wdog_tickler = NULL; 141 #endif /* HW_WDOG */ 142 143 /* 144 * Variable panicstr contains argument to first call to panic; used as flag 145 * to indicate that the kernel has already called panic. 146 */ 147 const char *panicstr; 148 149 int dumping; /* system is dumping */ 150 static struct dumperinfo dumper; /* selected dumper */ 151 152 globaldata_t panic_cpu_gd; /* which cpu took the panic */ 153 struct lwkt_tokref panic_tokens[LWKT_MAXTOKENS]; 154 int panic_tokens_count; 155 156 int bootverbose = 0; /* note: assignment to force non-bss */ 157 SYSCTL_INT(_debug, OID_AUTO, bootverbose, CTLFLAG_RW, 158 &bootverbose, 0, "Verbose kernel messages"); 159 160 int cold = 1; /* note: assignment to force non-bss */ 161 int dumplo; /* OBSOLETE - savecore compat */ 162 u_int64_t dumplo64; 163 164 static void boot (int) __dead2; 165 static int setdumpdev (cdev_t dev); 166 static void poweroff_wait (void *, int); 167 static void print_uptime (void); 168 static void shutdown_halt (void *junk, int howto); 169 static void shutdown_panic (void *junk, int howto); 170 static void shutdown_reset (void *junk, int howto); 171 static int shutdown_busycount1(struct buf *bp, void *info); 172 static int shutdown_busycount2(struct buf *bp, void *info); 173 static void shutdown_cleanup_proc(struct proc *p); 174 175 /* register various local shutdown events */ 176 static void 177 shutdown_conf(void *unused) 178 { 179 EVENTHANDLER_REGISTER(shutdown_final, poweroff_wait, NULL, SHUTDOWN_PRI_FIRST); 180 EVENTHANDLER_REGISTER(shutdown_final, shutdown_halt, NULL, SHUTDOWN_PRI_LAST + 100); 181 EVENTHANDLER_REGISTER(shutdown_final, shutdown_panic, NULL, SHUTDOWN_PRI_LAST + 100); 182 EVENTHANDLER_REGISTER(shutdown_final, shutdown_reset, NULL, SHUTDOWN_PRI_LAST + 200); 183 } 184 185 SYSINIT(shutdown_conf, SI_BOOT2_MACHDEP, SI_ORDER_ANY, shutdown_conf, NULL) 186 187 /* ARGSUSED */ 188 189 /* 190 * The system call that results in a reboot 191 * 192 * MPALMOSTSAFE 193 */ 194 int 195 sys_reboot(struct reboot_args *uap) 196 { 197 struct thread *td = curthread; 198 int error; 199 200 if ((error = priv_check(td, PRIV_REBOOT))) 201 return (error); 202 203 get_mplock(); 204 boot(uap->opt); 205 rel_mplock(); 206 return (0); 207 } 208 209 /* 210 * Called by events that want to shut down.. e.g <CTL><ALT><DEL> on a PC 211 */ 212 static int shutdown_howto = 0; 213 214 void 215 shutdown_nice(int howto) 216 { 217 shutdown_howto = howto; 218 219 /* Send a signal to init(8) and have it shutdown the world */ 220 if (initproc != NULL) { 221 ksignal(initproc, SIGINT); 222 } else { 223 /* No init(8) running, so simply reboot */ 224 boot(RB_NOSYNC); 225 } 226 return; 227 } 228 static int waittime = -1; 229 struct pcb dumppcb; 230 struct thread *dumpthread; 231 232 static void 233 print_uptime(void) 234 { 235 int f; 236 struct timespec ts; 237 238 getnanouptime(&ts); 239 kprintf("Uptime: "); 240 f = 0; 241 if (ts.tv_sec >= 86400) { 242 kprintf("%ldd", ts.tv_sec / 86400); 243 ts.tv_sec %= 86400; 244 f = 1; 245 } 246 if (f || ts.tv_sec >= 3600) { 247 kprintf("%ldh", ts.tv_sec / 3600); 248 ts.tv_sec %= 3600; 249 f = 1; 250 } 251 if (f || ts.tv_sec >= 60) { 252 kprintf("%ldm", ts.tv_sec / 60); 253 ts.tv_sec %= 60; 254 f = 1; 255 } 256 kprintf("%lds\n", ts.tv_sec); 257 } 258 259 /* 260 * Go through the rigmarole of shutting down.. 261 * this used to be in machdep.c but I'll be dammned if I could see 262 * anything machine dependant in it. 263 */ 264 static void 265 boot(int howto) 266 { 267 /* 268 * Get rid of any user scheduler baggage and then give 269 * us a high priority. 270 */ 271 if (curthread->td_release) 272 curthread->td_release(curthread); 273 lwkt_setpri_self(TDPRI_MAX); 274 275 /* collect extra flags that shutdown_nice might have set */ 276 howto |= shutdown_howto; 277 278 #ifdef SMP 279 /* 280 * We really want to shutdown on the BSP. Subsystems such as ACPI 281 * can't power-down the box otherwise. 282 */ 283 if (smp_active_mask > 1) { 284 kprintf("boot() called on cpu#%d\n", mycpu->gd_cpuid); 285 } 286 if (panicstr == NULL && mycpu->gd_cpuid != 0) { 287 kprintf("Switching to cpu #0 for shutdown\n"); 288 lwkt_setcpu_self(globaldata_find(0)); 289 } 290 #endif 291 /* 292 * Do any callouts that should be done BEFORE syncing the filesystems. 293 */ 294 EVENTHANDLER_INVOKE(shutdown_pre_sync, howto); 295 296 /* 297 * Try to get rid of any remaining FS references. The calling 298 * process, proc0, and init may still hold references. The 299 * VFS cache subsystem may still hold a root reference to root. 300 * 301 * XXX this needs work. We really need to SIGSTOP all remaining 302 * processes in order to avoid blowups due to proc0's filesystem 303 * references going away. For now just make sure that the init 304 * process is stopped. 305 */ 306 if (panicstr == NULL) { 307 shutdown_cleanup_proc(curproc); 308 shutdown_cleanup_proc(&proc0); 309 if (initproc) { 310 if (initproc != curproc) { 311 ksignal(initproc, SIGSTOP); 312 tsleep(boot, 0, "shutdn", hz / 20); 313 } 314 shutdown_cleanup_proc(initproc); 315 } 316 vfs_cache_setroot(NULL, NULL); 317 } 318 319 /* 320 * Now sync filesystems 321 */ 322 if (!cold && (howto & RB_NOSYNC) == 0 && waittime < 0) { 323 int iter, nbusy, pbusy; 324 325 waittime = 0; 326 kprintf("\nsyncing disks... "); 327 328 sys_sync(NULL); /* YYY was sync(&proc0, NULL). why proc0 ? */ 329 330 /* 331 * With soft updates, some buffers that are 332 * written will be remarked as dirty until other 333 * buffers are written. 334 */ 335 for (iter = pbusy = 0; iter < 20; iter++) { 336 nbusy = scan_all_buffers(shutdown_busycount1, NULL); 337 if (nbusy == 0) 338 break; 339 kprintf("%d ", nbusy); 340 if (nbusy < pbusy) 341 iter = 0; 342 pbusy = nbusy; 343 /* 344 * XXX: 345 * Process soft update work queue if buffers don't sync 346 * after 6 iterations by permitting the syncer to run. 347 */ 348 if (iter > 5) 349 bio_ops_sync(NULL); 350 351 sys_sync(NULL); /* YYY was sync(&proc0, NULL). why proc0 ? */ 352 tsleep(boot, 0, "shutdn", hz * iter / 20 + 1); 353 } 354 kprintf("\n"); 355 /* 356 * Count only busy local buffers to prevent forcing 357 * a fsck if we're just a client of a wedged NFS server 358 */ 359 nbusy = scan_all_buffers(shutdown_busycount2, NULL); 360 if (nbusy) { 361 /* 362 * Failed to sync all blocks. Indicate this and don't 363 * unmount filesystems (thus forcing an fsck on reboot). 364 */ 365 kprintf("giving up on %d buffers\n", nbusy); 366 #ifdef DDB 367 Debugger("busy buffer problem"); 368 #endif /* DDB */ 369 tsleep(boot, 0, "shutdn", hz * 5 + 1); 370 } else { 371 kprintf("done\n"); 372 /* 373 * Unmount filesystems 374 */ 375 if (panicstr == NULL) 376 vfs_unmountall(); 377 } 378 tsleep(boot, 0, "shutdn", hz / 10 + 1); 379 } 380 381 print_uptime(); 382 383 /* 384 * Dump before doing post_sync shutdown ops 385 */ 386 crit_enter(); 387 if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP && !cold) { 388 dumpsys(); 389 } 390 391 /* 392 * Ok, now do things that assume all filesystem activity has 393 * been completed. This will also call the device shutdown 394 * methods. 395 */ 396 EVENTHANDLER_INVOKE(shutdown_post_sync, howto); 397 398 /* Now that we're going to really halt the system... */ 399 EVENTHANDLER_INVOKE(shutdown_final, howto); 400 401 for(;;) ; /* safety against shutdown_reset not working */ 402 /* NOTREACHED */ 403 } 404 405 static int 406 shutdown_busycount1(struct buf *bp, void *info) 407 { 408 if ((bp->b_flags & B_INVAL) == 0 && BUF_REFCNT(bp) > 0) 409 return(1); 410 if ((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) 411 return (1); 412 return (0); 413 } 414 415 static int 416 shutdown_busycount2(struct buf *bp, void *info) 417 { 418 if (((bp->b_flags & B_INVAL) == 0 && BUF_REFCNT(bp)) || 419 ((bp->b_flags & (B_DELWRI|B_INVAL)) == B_DELWRI)) { 420 /* 421 * Only count buffers undergoing write I/O 422 * on the related vnode. 423 */ 424 if (bp->b_vp == NULL || 425 bio_track_active(&bp->b_vp->v_track_write) == 0) { 426 return (0); 427 } 428 #if defined(SHOW_BUSYBUFS) || defined(DIAGNOSTIC) 429 kprintf( 430 "%p dev:?, flags:%08x, loffset:%jd, doffset:%jd\n", 431 bp, 432 bp->b_flags, (intmax_t)bp->b_loffset, 433 (intmax_t)bp->b_bio2.bio_offset); 434 #endif 435 return(1); 436 } 437 return(0); 438 } 439 440 /* 441 * If the shutdown was a clean halt, behave accordingly. 442 */ 443 static void 444 shutdown_halt(void *junk, int howto) 445 { 446 if (howto & RB_HALT) { 447 kprintf("\n"); 448 kprintf("The operating system has halted.\n"); 449 #ifdef _KERNEL_VIRTUAL 450 cpu_halt(); 451 #else 452 kprintf("Please press any key to reboot.\n\n"); 453 switch (cngetc()) { 454 case -1: /* No console, just die */ 455 cpu_halt(); 456 /* NOTREACHED */ 457 default: 458 howto &= ~RB_HALT; 459 break; 460 } 461 #endif 462 } 463 } 464 465 /* 466 * Check to see if the system paniced, pause and then reboot 467 * according to the specified delay. 468 */ 469 static void 470 shutdown_panic(void *junk, int howto) 471 { 472 int loop; 473 474 if (howto & RB_DUMP) { 475 if (PANIC_REBOOT_WAIT_TIME != 0) { 476 if (PANIC_REBOOT_WAIT_TIME != -1) { 477 kprintf("Automatic reboot in %d seconds - " 478 "press a key on the console to abort\n", 479 PANIC_REBOOT_WAIT_TIME); 480 for (loop = PANIC_REBOOT_WAIT_TIME * 10; 481 loop > 0; --loop) { 482 DELAY(1000 * 100); /* 1/10th second */ 483 /* Did user type a key? */ 484 if (cncheckc() != -1) 485 break; 486 } 487 if (!loop) 488 return; 489 } 490 } else { /* zero time specified - reboot NOW */ 491 return; 492 } 493 kprintf("--> Press a key on the console to reboot,\n"); 494 kprintf("--> or switch off the system now.\n"); 495 cngetc(); 496 } 497 } 498 499 /* 500 * Everything done, now reset 501 */ 502 static void 503 shutdown_reset(void *junk, int howto) 504 { 505 kprintf("Rebooting...\n"); 506 DELAY(1000000); /* wait 1 sec for kprintf's to complete and be read */ 507 /* cpu_boot(howto); */ /* doesn't do anything at the moment */ 508 cpu_reset(); 509 /* NOTREACHED */ /* assuming reset worked */ 510 } 511 512 /* 513 * Try to remove FS references in the specified process. This function 514 * is used during shutdown 515 */ 516 static 517 void 518 shutdown_cleanup_proc(struct proc *p) 519 { 520 struct filedesc *fdp; 521 struct vmspace *vm; 522 523 if (p == NULL) 524 return; 525 if ((fdp = p->p_fd) != NULL) { 526 kern_closefrom(0); 527 if (fdp->fd_cdir) { 528 cache_drop(&fdp->fd_ncdir); 529 vrele(fdp->fd_cdir); 530 fdp->fd_cdir = NULL; 531 } 532 if (fdp->fd_rdir) { 533 cache_drop(&fdp->fd_nrdir); 534 vrele(fdp->fd_rdir); 535 fdp->fd_rdir = NULL; 536 } 537 if (fdp->fd_jdir) { 538 cache_drop(&fdp->fd_njdir); 539 vrele(fdp->fd_jdir); 540 fdp->fd_jdir = NULL; 541 } 542 } 543 if (p->p_vkernel) 544 vkernel_exit(p); 545 if (p->p_textvp) { 546 vrele(p->p_textvp); 547 p->p_textvp = NULL; 548 } 549 vm = p->p_vmspace; 550 if (vm != NULL) { 551 pmap_remove_pages(vmspace_pmap(vm), 552 VM_MIN_USER_ADDRESS, 553 VM_MAX_USER_ADDRESS); 554 vm_map_remove(&vm->vm_map, 555 VM_MIN_USER_ADDRESS, 556 VM_MAX_USER_ADDRESS); 557 } 558 } 559 560 /* 561 * Magic number for savecore 562 * 563 * exported (symorder) and used at least by savecore(8) 564 * 565 * Mark it as used so that gcc doesn't optimize it away. 566 */ 567 __attribute__((__used__)) 568 static u_long const dumpmag = 0x8fca0101UL; 569 570 __attribute__((__used__)) 571 static int dumpsize = 0; /* also for savecore */ 572 573 static int dodump = 1; 574 575 SYSCTL_INT(_machdep, OID_AUTO, do_dump, CTLFLAG_RW, &dodump, 0, 576 "Try to perform coredump on kernel panic"); 577 578 void 579 mkdumpheader(struct kerneldumpheader *kdh, char *magic, uint32_t archver, 580 uint64_t dumplen, uint32_t blksz) 581 { 582 bzero(kdh, sizeof(*kdh)); 583 strncpy(kdh->magic, magic, sizeof(kdh->magic)); 584 strncpy(kdh->architecture, MACHINE_ARCH, sizeof(kdh->architecture)); 585 kdh->version = htod32(KERNELDUMPVERSION); 586 kdh->architectureversion = htod32(archver); 587 kdh->dumplength = htod64(dumplen); 588 kdh->dumptime = htod64(time_second); 589 kdh->blocksize = htod32(blksz); 590 strncpy(kdh->hostname, hostname, sizeof(kdh->hostname)); 591 strncpy(kdh->versionstring, version, sizeof(kdh->versionstring)); 592 if (panicstr != NULL) 593 strncpy(kdh->panicstring, panicstr, sizeof(kdh->panicstring)); 594 kdh->parity = kerneldump_parity(kdh); 595 } 596 597 static int 598 setdumpdev(cdev_t dev) 599 { 600 int error; 601 int doopen; 602 603 if (dev == NULL) { 604 disk_dumpconf(NULL, 0/*off*/); 605 return (0); 606 } 607 608 /* 609 * We have to open the device before we can perform ioctls on it, 610 * or the slice/label data may not be present. Device opens are 611 * usually tracked by specfs, but the dump device can be set in 612 * early boot and may not be open so this is somewhat of a hack. 613 */ 614 doopen = (dev->si_sysref.refcnt == 1); 615 if (doopen) { 616 error = dev_dopen(dev, FREAD, S_IFCHR, proc0.p_ucred); 617 if (error) 618 return (error); 619 } 620 error = disk_dumpconf(dev, 1/*on*/); 621 622 return error; 623 } 624 625 /* ARGSUSED */ 626 static void dump_conf (void *dummy); 627 static void 628 dump_conf(void *dummy) 629 { 630 char *path; 631 cdev_t dev; 632 633 path = kmalloc(MNAMELEN, M_TEMP, M_WAITOK); 634 if (TUNABLE_STR_FETCH("dumpdev", path, MNAMELEN) != 0) { 635 sync_devs(); 636 dev = kgetdiskbyname(path); 637 if (dev != NULL) 638 dumpdev = dev; 639 } 640 kfree(path, M_TEMP); 641 if (setdumpdev(dumpdev) != 0) 642 dumpdev = NULL; 643 } 644 645 SYSINIT(dump_conf, SI_SUB_DUMP_CONF, SI_ORDER_FIRST, dump_conf, NULL) 646 647 static int 648 sysctl_kern_dumpdev(SYSCTL_HANDLER_ARGS) 649 { 650 int error; 651 udev_t ndumpdev; 652 653 ndumpdev = dev2udev(dumpdev); 654 error = sysctl_handle_opaque(oidp, &ndumpdev, sizeof ndumpdev, req); 655 if (error == 0 && req->newptr != NULL) 656 error = setdumpdev(udev2dev(ndumpdev, 0)); 657 return (error); 658 } 659 660 SYSCTL_PROC(_kern, KERN_DUMPDEV, dumpdev, CTLTYPE_OPAQUE|CTLFLAG_RW, 661 0, sizeof dumpdev, sysctl_kern_dumpdev, "T,udev_t", ""); 662 663 /* 664 * Panic is called on unresolvable fatal errors. It prints "panic: mesg", 665 * and then reboots. If we are called twice, then we avoid trying to sync 666 * the disks as this often leads to recursive panics. 667 */ 668 void 669 panic(const char *fmt, ...) 670 { 671 int bootopt, newpanic; 672 globaldata_t gd = mycpu; 673 thread_t td = gd->gd_curthread; 674 __va_list ap; 675 static char buf[256]; 676 677 #ifdef SMP 678 /* 679 * If a panic occurs on multiple cpus before the first is able to 680 * halt the other cpus, only one cpu is allowed to take the panic. 681 * Attempt to be verbose about this situation but if the kprintf() 682 * itself panics don't let us overrun the kernel stack. 683 * 684 * Be very nasty about descheduling our thread at the lowest 685 * level possible in an attempt to freeze the thread without 686 * inducing further panics. 687 * 688 * Bumping gd_trap_nesting_level will also bypass assertions in 689 * lwkt_switch() and allow us to switch away even if we are a 690 * FAST interrupt or IPI. 691 * 692 * The setting of panic_cpu_gd also determines how kprintf() 693 * spin-locks itself. DDB can set panic_cpu_gd as well. 694 */ 695 for (;;) { 696 globaldata_t xgd = panic_cpu_gd; 697 698 /* 699 * Someone else got the panic cpu 700 */ 701 if (xgd && xgd != gd) { 702 crit_enter(); 703 ++mycpu->gd_trap_nesting_level; 704 if (mycpu->gd_trap_nesting_level < 25) { 705 kprintf("SECONDARY PANIC ON CPU %d THREAD %p\n", 706 mycpu->gd_cpuid, td); 707 } 708 td->td_release = NULL; /* be a grinch */ 709 for (;;) { 710 lwkt_deschedule_self(td); 711 lwkt_switch(); 712 } 713 /* NOT REACHED */ 714 /* --mycpu->gd_trap_nesting_level */ 715 /* crit_exit() */ 716 } 717 718 /* 719 * Reentrant panic 720 */ 721 if (xgd && xgd == gd) 722 break; 723 724 /* 725 * We got it 726 */ 727 if (atomic_cmpset_ptr(&panic_cpu_gd, NULL, gd)) 728 break; 729 } 730 #else 731 panic_cpu_gd = gd; 732 #endif 733 /* 734 * Try to get the system into a working state. Save information 735 * we are about to destroy. 736 */ 737 kvcreinitspin(); 738 if (panicstr == NULL) { 739 bcopy(td->td_toks_array, panic_tokens, sizeof(panic_tokens)); 740 panic_tokens_count = td->td_toks_stop - &td->td_toks_base; 741 } 742 lwkt_relalltokens(td); 743 td->td_toks_stop = &td->td_toks_base; 744 745 /* 746 * Setup 747 */ 748 bootopt = RB_AUTOBOOT | RB_DUMP; 749 if (sync_on_panic == 0) 750 bootopt |= RB_NOSYNC; 751 newpanic = 0; 752 if (panicstr) { 753 bootopt |= RB_NOSYNC; 754 } else { 755 panicstr = fmt; 756 newpanic = 1; 757 } 758 759 /* 760 * Format the panic string. 761 */ 762 __va_start(ap, fmt); 763 kvsnprintf(buf, sizeof(buf), fmt, ap); 764 if (panicstr == fmt) 765 panicstr = buf; 766 __va_end(ap); 767 kprintf("panic: %s\n", buf); 768 #ifdef SMP 769 /* two separate prints in case of an unmapped page and trap */ 770 kprintf("mp_lock = %08x; ", mp_lock); 771 kprintf("cpuid = %d\n", mycpu->gd_cpuid); 772 #endif 773 774 #if (NGPIO > 0) && defined(ERROR_LED_ON_PANIC) 775 led_switch("error", 1); 776 #endif 777 778 #if defined(WDOG_DISABLE_ON_PANIC) && defined(WATCHDOG_ENABLE) 779 wdog_disable(); 780 #endif 781 782 #if defined(DDB) 783 if (newpanic && trace_on_panic) 784 print_backtrace(-1); 785 if (debugger_on_panic) 786 Debugger("panic"); 787 #endif 788 boot(bootopt); 789 } 790 791 /* 792 * Support for poweroff delay. 793 */ 794 #ifndef POWEROFF_DELAY 795 # define POWEROFF_DELAY 5000 796 #endif 797 static int poweroff_delay = POWEROFF_DELAY; 798 799 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW, 800 &poweroff_delay, 0, ""); 801 802 static void 803 poweroff_wait(void *junk, int howto) 804 { 805 if(!(howto & RB_POWEROFF) || poweroff_delay <= 0) 806 return; 807 DELAY(poweroff_delay * 1000); 808 } 809 810 /* 811 * Some system processes (e.g. syncer) need to be stopped at appropriate 812 * points in their main loops prior to a system shutdown, so that they 813 * won't interfere with the shutdown process (e.g. by holding a disk buf 814 * to cause sync to fail). For each of these system processes, register 815 * shutdown_kproc() as a handler for one of shutdown events. 816 */ 817 static int kproc_shutdown_wait = 60; 818 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW, 819 &kproc_shutdown_wait, 0, ""); 820 821 void 822 shutdown_kproc(void *arg, int howto) 823 { 824 struct thread *td; 825 struct proc *p; 826 int error; 827 828 if (panicstr) 829 return; 830 831 td = (struct thread *)arg; 832 if ((p = td->td_proc) != NULL) { 833 kprintf("Waiting (max %d seconds) for system process `%s' to stop...", 834 kproc_shutdown_wait, p->p_comm); 835 } else { 836 kprintf("Waiting (max %d seconds) for system thread %s to stop...", 837 kproc_shutdown_wait, td->td_comm); 838 } 839 error = suspend_kproc(td, kproc_shutdown_wait * hz); 840 841 if (error == EWOULDBLOCK) 842 kprintf("timed out\n"); 843 else 844 kprintf("stopped\n"); 845 } 846 847 /* Registration of dumpers */ 848 int 849 set_dumper(struct dumperinfo *di) 850 { 851 if (di == NULL) { 852 bzero(&dumper, sizeof(dumper)); 853 return 0; 854 } 855 856 if (dumper.dumper != NULL) 857 return (EBUSY); 858 859 dumper = *di; 860 return 0; 861 } 862 863 void 864 dumpsys(void) 865 { 866 #if defined (_KERNEL_VIRTUAL) 867 /* VKERNELs don't support dumps */ 868 kprintf("VKERNEL doesn't support dumps\n"); 869 return; 870 #endif 871 /* 872 * If there is a dumper registered and we aren't dumping already, call 873 * the machine dependent dumpsys (md_dumpsys) to do the hard work. 874 * 875 * XXX: while right now the md_dumpsys() of x86 and x86_64 could be 876 * factored out completely into here, I rather keep them machine 877 * dependent in case we ever add a platform which does not share 878 * the same dumpsys() code, such as arm. 879 */ 880 if (dumper.dumper != NULL && !dumping) { 881 dumping++; 882 md_dumpsys(&dumper); 883 } 884 } 885 886 int dump_stop_usertds = 0; 887 888 #ifdef SMP 889 static 890 void 891 need_user_resched_remote(void *dummy) 892 { 893 need_user_resched(); 894 } 895 #endif 896 897 void 898 dump_reactivate_cpus(void) 899 { 900 #ifdef SMP 901 globaldata_t gd; 902 int cpu, seq; 903 #endif 904 905 dump_stop_usertds = 1; 906 907 need_user_resched(); 908 909 #ifdef SMP 910 for (cpu = 0; cpu < ncpus; cpu++) { 911 gd = globaldata_find(cpu); 912 seq = lwkt_send_ipiq(gd, need_user_resched_remote, NULL); 913 lwkt_wait_ipiq(gd, seq); 914 } 915 916 restart_cpus(stopped_cpus); 917 #endif 918 } 919