1 /* 2 * linux/kernel/panic.c 3 * 4 * Copyright (C) 1991, 1992 Linus Torvalds 5 */ 6 7 /* 8 * This function is used through-out the kernel (including mm and fs) 9 * to indicate a major problem. 10 */ 11 #include <linux/debug_locks.h> 12 #include <linux/interrupt.h> 13 #include <linux/kmsg_dump.h> 14 #include <linux/kallsyms.h> 15 #include <linux/notifier.h> 16 #include <linux/module.h> 17 #include <linux/random.h> 18 #include <linux/ftrace.h> 19 #include <linux/reboot.h> 20 #include <linux/delay.h> 21 #include <linux/kexec.h> 22 #include <linux/sched.h> 23 #include <linux/sysrq.h> 24 #include <linux/init.h> 25 #include <linux/nmi.h> 26 #include <linux/console.h> 27 #include <linux/bug.h> 28 29 #define PANIC_TIMER_STEP 100 30 #define PANIC_BLINK_SPD 18 31 32 int panic_on_oops = CONFIG_PANIC_ON_OOPS_VALUE; 33 static unsigned long tainted_mask; 34 static int pause_on_oops; 35 static int pause_on_oops_flag; 36 static DEFINE_SPINLOCK(pause_on_oops_lock); 37 bool crash_kexec_post_notifiers; 38 int panic_on_warn __read_mostly; 39 40 int panic_timeout = CONFIG_PANIC_TIMEOUT; 41 EXPORT_SYMBOL_GPL(panic_timeout); 42 43 ATOMIC_NOTIFIER_HEAD(panic_notifier_list); 44 45 EXPORT_SYMBOL(panic_notifier_list); 46 47 static long no_blink(int state) 48 { 49 return 0; 50 } 51 52 /* Returns how long it waited in ms */ 53 long (*panic_blink)(int state); 54 EXPORT_SYMBOL(panic_blink); 55 56 /* 57 * Stop ourself in panic -- architecture code may override this 58 */ 59 void __weak panic_smp_self_stop(void) 60 { 61 while (1) 62 cpu_relax(); 63 } 64 65 /* 66 * Stop ourselves in NMI context if another CPU has already panicked. Arch code 67 * may override this to prepare for crash dumping, e.g. save regs info. 68 */ 69 void __weak nmi_panic_self_stop(struct pt_regs *regs) 70 { 71 panic_smp_self_stop(); 72 } 73 74 /* 75 * Stop other CPUs in panic. Architecture dependent code may override this 76 * with more suitable version. For example, if the architecture supports 77 * crash dump, it should save registers of each stopped CPU and disable 78 * per-CPU features such as virtualization extensions. 79 */ 80 void __weak crash_smp_send_stop(void) 81 { 82 static int cpus_stopped; 83 84 /* 85 * This function can be called twice in panic path, but obviously 86 * we execute this only once. 87 */ 88 if (cpus_stopped) 89 return; 90 91 /* 92 * Note smp_send_stop is the usual smp shutdown function, which 93 * unfortunately means it may not be hardened to work in a panic 94 * situation. 95 */ 96 smp_send_stop(); 97 cpus_stopped = 1; 98 } 99 100 atomic_t panic_cpu = ATOMIC_INIT(PANIC_CPU_INVALID); 101 102 /* 103 * A variant of panic() called from NMI context. We return if we've already 104 * panicked on this CPU. If another CPU already panicked, loop in 105 * nmi_panic_self_stop() which can provide architecture dependent code such 106 * as saving register state for crash dump. 107 */ 108 void nmi_panic(struct pt_regs *regs, const char *msg) 109 { 110 int old_cpu, cpu; 111 112 cpu = raw_smp_processor_id(); 113 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, cpu); 114 115 if (old_cpu == PANIC_CPU_INVALID) 116 panic("%s", msg); 117 else if (old_cpu != cpu) 118 nmi_panic_self_stop(regs); 119 } 120 EXPORT_SYMBOL(nmi_panic); 121 122 /** 123 * panic - halt the system 124 * @fmt: The text string to print 125 * 126 * Display a message, then perform cleanups. 127 * 128 * This function never returns. 129 */ 130 void panic(const char *fmt, ...) 131 { 132 static char buf[1024]; 133 va_list args; 134 long i, i_next = 0; 135 int state = 0; 136 int old_cpu, this_cpu; 137 bool _crash_kexec_post_notifiers = crash_kexec_post_notifiers; 138 139 /* 140 * Disable local interrupts. This will prevent panic_smp_self_stop 141 * from deadlocking the first cpu that invokes the panic, since 142 * there is nothing to prevent an interrupt handler (that runs 143 * after setting panic_cpu) from invoking panic() again. 144 */ 145 local_irq_disable(); 146 147 /* 148 * It's possible to come here directly from a panic-assertion and 149 * not have preempt disabled. Some functions called from here want 150 * preempt to be disabled. No point enabling it later though... 151 * 152 * Only one CPU is allowed to execute the panic code from here. For 153 * multiple parallel invocations of panic, all other CPUs either 154 * stop themself or will wait until they are stopped by the 1st CPU 155 * with smp_send_stop(). 156 * 157 * `old_cpu == PANIC_CPU_INVALID' means this is the 1st CPU which 158 * comes here, so go ahead. 159 * `old_cpu == this_cpu' means we came from nmi_panic() which sets 160 * panic_cpu to this CPU. In this case, this is also the 1st CPU. 161 */ 162 this_cpu = raw_smp_processor_id(); 163 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu); 164 165 if (old_cpu != PANIC_CPU_INVALID && old_cpu != this_cpu) 166 panic_smp_self_stop(); 167 168 console_verbose(); 169 bust_spinlocks(1); 170 va_start(args, fmt); 171 vsnprintf(buf, sizeof(buf), fmt, args); 172 va_end(args); 173 pr_emerg("Kernel panic - not syncing: %s\n", buf); 174 #ifdef CONFIG_DEBUG_BUGVERBOSE 175 /* 176 * Avoid nested stack-dumping if a panic occurs during oops processing 177 */ 178 if (!test_taint(TAINT_DIE) && oops_in_progress <= 1) 179 dump_stack(); 180 #endif 181 182 /* 183 * If we have crashed and we have a crash kernel loaded let it handle 184 * everything else. 185 * If we want to run this after calling panic_notifiers, pass 186 * the "crash_kexec_post_notifiers" option to the kernel. 187 * 188 * Bypass the panic_cpu check and call __crash_kexec directly. 189 */ 190 if (!_crash_kexec_post_notifiers) { 191 printk_nmi_flush_on_panic(); 192 __crash_kexec(NULL); 193 194 /* 195 * Note smp_send_stop is the usual smp shutdown function, which 196 * unfortunately means it may not be hardened to work in a 197 * panic situation. 198 */ 199 smp_send_stop(); 200 } else { 201 /* 202 * If we want to do crash dump after notifier calls and 203 * kmsg_dump, we will need architecture dependent extra 204 * works in addition to stopping other CPUs. 205 */ 206 crash_smp_send_stop(); 207 } 208 209 /* 210 * Run any panic handlers, including those that might need to 211 * add information to the kmsg dump output. 212 */ 213 atomic_notifier_call_chain(&panic_notifier_list, 0, buf); 214 215 /* Call flush even twice. It tries harder with a single online CPU */ 216 printk_nmi_flush_on_panic(); 217 kmsg_dump(KMSG_DUMP_PANIC); 218 219 /* 220 * If you doubt kdump always works fine in any situation, 221 * "crash_kexec_post_notifiers" offers you a chance to run 222 * panic_notifiers and dumping kmsg before kdump. 223 * Note: since some panic_notifiers can make crashed kernel 224 * more unstable, it can increase risks of the kdump failure too. 225 * 226 * Bypass the panic_cpu check and call __crash_kexec directly. 227 */ 228 if (_crash_kexec_post_notifiers) 229 __crash_kexec(NULL); 230 231 bust_spinlocks(0); 232 233 /* 234 * We may have ended up stopping the CPU holding the lock (in 235 * smp_send_stop()) while still having some valuable data in the console 236 * buffer. Try to acquire the lock then release it regardless of the 237 * result. The release will also print the buffers out. Locks debug 238 * should be disabled to avoid reporting bad unlock balance when 239 * panic() is not being callled from OOPS. 240 */ 241 debug_locks_off(); 242 console_flush_on_panic(); 243 244 if (!panic_blink) 245 panic_blink = no_blink; 246 247 if (panic_timeout > 0) { 248 /* 249 * Delay timeout seconds before rebooting the machine. 250 * We can't use the "normal" timers since we just panicked. 251 */ 252 pr_emerg("Rebooting in %d seconds..", panic_timeout); 253 254 for (i = 0; i < panic_timeout * 1000; i += PANIC_TIMER_STEP) { 255 touch_nmi_watchdog(); 256 if (i >= i_next) { 257 i += panic_blink(state ^= 1); 258 i_next = i + 3600 / PANIC_BLINK_SPD; 259 } 260 mdelay(PANIC_TIMER_STEP); 261 } 262 } 263 if (panic_timeout != 0) { 264 /* 265 * This will not be a clean reboot, with everything 266 * shutting down. But if there is a chance of 267 * rebooting the system it will be rebooted. 268 */ 269 emergency_restart(); 270 } 271 #ifdef __sparc__ 272 { 273 extern int stop_a_enabled; 274 /* Make sure the user can actually press Stop-A (L1-A) */ 275 stop_a_enabled = 1; 276 pr_emerg("Press Stop-A (L1-A) to return to the boot prom\n"); 277 } 278 #endif 279 #if defined(CONFIG_S390) 280 { 281 unsigned long caller; 282 283 caller = (unsigned long)__builtin_return_address(0); 284 disabled_wait(caller); 285 } 286 #endif 287 pr_emerg("---[ end Kernel panic - not syncing: %s\n", buf); 288 local_irq_enable(); 289 for (i = 0; ; i += PANIC_TIMER_STEP) { 290 touch_softlockup_watchdog(); 291 if (i >= i_next) { 292 i += panic_blink(state ^= 1); 293 i_next = i + 3600 / PANIC_BLINK_SPD; 294 } 295 mdelay(PANIC_TIMER_STEP); 296 } 297 } 298 299 EXPORT_SYMBOL(panic); 300 301 302 struct tnt { 303 u8 bit; 304 char true; 305 char false; 306 }; 307 308 static const struct tnt tnts[] = { 309 { TAINT_PROPRIETARY_MODULE, 'P', 'G' }, 310 { TAINT_FORCED_MODULE, 'F', ' ' }, 311 { TAINT_CPU_OUT_OF_SPEC, 'S', ' ' }, 312 { TAINT_FORCED_RMMOD, 'R', ' ' }, 313 { TAINT_MACHINE_CHECK, 'M', ' ' }, 314 { TAINT_BAD_PAGE, 'B', ' ' }, 315 { TAINT_USER, 'U', ' ' }, 316 { TAINT_DIE, 'D', ' ' }, 317 { TAINT_OVERRIDDEN_ACPI_TABLE, 'A', ' ' }, 318 { TAINT_WARN, 'W', ' ' }, 319 { TAINT_CRAP, 'C', ' ' }, 320 { TAINT_FIRMWARE_WORKAROUND, 'I', ' ' }, 321 { TAINT_OOT_MODULE, 'O', ' ' }, 322 { TAINT_UNSIGNED_MODULE, 'E', ' ' }, 323 { TAINT_SOFTLOCKUP, 'L', ' ' }, 324 { TAINT_LIVEPATCH, 'K', ' ' }, 325 }; 326 327 /** 328 * print_tainted - return a string to represent the kernel taint state. 329 * 330 * 'P' - Proprietary module has been loaded. 331 * 'F' - Module has been forcibly loaded. 332 * 'S' - SMP with CPUs not designed for SMP. 333 * 'R' - User forced a module unload. 334 * 'M' - System experienced a machine check exception. 335 * 'B' - System has hit bad_page. 336 * 'U' - Userspace-defined naughtiness. 337 * 'D' - Kernel has oopsed before 338 * 'A' - ACPI table overridden. 339 * 'W' - Taint on warning. 340 * 'C' - modules from drivers/staging are loaded. 341 * 'I' - Working around severe firmware bug. 342 * 'O' - Out-of-tree module has been loaded. 343 * 'E' - Unsigned module has been loaded. 344 * 'L' - A soft lockup has previously occurred. 345 * 'K' - Kernel has been live patched. 346 * 347 * The string is overwritten by the next call to print_tainted(). 348 */ 349 const char *print_tainted(void) 350 { 351 static char buf[ARRAY_SIZE(tnts) + sizeof("Tainted: ")]; 352 353 if (tainted_mask) { 354 char *s; 355 int i; 356 357 s = buf + sprintf(buf, "Tainted: "); 358 for (i = 0; i < ARRAY_SIZE(tnts); i++) { 359 const struct tnt *t = &tnts[i]; 360 *s++ = test_bit(t->bit, &tainted_mask) ? 361 t->true : t->false; 362 } 363 *s = 0; 364 } else 365 snprintf(buf, sizeof(buf), "Not tainted"); 366 367 return buf; 368 } 369 370 int test_taint(unsigned flag) 371 { 372 return test_bit(flag, &tainted_mask); 373 } 374 EXPORT_SYMBOL(test_taint); 375 376 unsigned long get_taint(void) 377 { 378 return tainted_mask; 379 } 380 381 /** 382 * add_taint: add a taint flag if not already set. 383 * @flag: one of the TAINT_* constants. 384 * @lockdep_ok: whether lock debugging is still OK. 385 * 386 * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for 387 * some notewortht-but-not-corrupting cases, it can be set to true. 388 */ 389 void add_taint(unsigned flag, enum lockdep_ok lockdep_ok) 390 { 391 if (lockdep_ok == LOCKDEP_NOW_UNRELIABLE && __debug_locks_off()) 392 pr_warn("Disabling lock debugging due to kernel taint\n"); 393 394 set_bit(flag, &tainted_mask); 395 } 396 EXPORT_SYMBOL(add_taint); 397 398 static void spin_msec(int msecs) 399 { 400 int i; 401 402 for (i = 0; i < msecs; i++) { 403 touch_nmi_watchdog(); 404 mdelay(1); 405 } 406 } 407 408 /* 409 * It just happens that oops_enter() and oops_exit() are identically 410 * implemented... 411 */ 412 static void do_oops_enter_exit(void) 413 { 414 unsigned long flags; 415 static int spin_counter; 416 417 if (!pause_on_oops) 418 return; 419 420 spin_lock_irqsave(&pause_on_oops_lock, flags); 421 if (pause_on_oops_flag == 0) { 422 /* This CPU may now print the oops message */ 423 pause_on_oops_flag = 1; 424 } else { 425 /* We need to stall this CPU */ 426 if (!spin_counter) { 427 /* This CPU gets to do the counting */ 428 spin_counter = pause_on_oops; 429 do { 430 spin_unlock(&pause_on_oops_lock); 431 spin_msec(MSEC_PER_SEC); 432 spin_lock(&pause_on_oops_lock); 433 } while (--spin_counter); 434 pause_on_oops_flag = 0; 435 } else { 436 /* This CPU waits for a different one */ 437 while (spin_counter) { 438 spin_unlock(&pause_on_oops_lock); 439 spin_msec(1); 440 spin_lock(&pause_on_oops_lock); 441 } 442 } 443 } 444 spin_unlock_irqrestore(&pause_on_oops_lock, flags); 445 } 446 447 /* 448 * Return true if the calling CPU is allowed to print oops-related info. 449 * This is a bit racy.. 450 */ 451 int oops_may_print(void) 452 { 453 return pause_on_oops_flag == 0; 454 } 455 456 /* 457 * Called when the architecture enters its oops handler, before it prints 458 * anything. If this is the first CPU to oops, and it's oopsing the first 459 * time then let it proceed. 460 * 461 * This is all enabled by the pause_on_oops kernel boot option. We do all 462 * this to ensure that oopses don't scroll off the screen. It has the 463 * side-effect of preventing later-oopsing CPUs from mucking up the display, 464 * too. 465 * 466 * It turns out that the CPU which is allowed to print ends up pausing for 467 * the right duration, whereas all the other CPUs pause for twice as long: 468 * once in oops_enter(), once in oops_exit(). 469 */ 470 void oops_enter(void) 471 { 472 tracing_off(); 473 /* can't trust the integrity of the kernel anymore: */ 474 debug_locks_off(); 475 do_oops_enter_exit(); 476 } 477 478 /* 479 * 64-bit random ID for oopses: 480 */ 481 static u64 oops_id; 482 483 static int init_oops_id(void) 484 { 485 if (!oops_id) 486 get_random_bytes(&oops_id, sizeof(oops_id)); 487 else 488 oops_id++; 489 490 return 0; 491 } 492 late_initcall(init_oops_id); 493 494 void print_oops_end_marker(void) 495 { 496 init_oops_id(); 497 pr_warn("---[ end trace %016llx ]---\n", (unsigned long long)oops_id); 498 } 499 500 /* 501 * Called when the architecture exits its oops handler, after printing 502 * everything. 503 */ 504 void oops_exit(void) 505 { 506 do_oops_enter_exit(); 507 print_oops_end_marker(); 508 kmsg_dump(KMSG_DUMP_OOPS); 509 } 510 511 struct warn_args { 512 const char *fmt; 513 va_list args; 514 }; 515 516 void __warn(const char *file, int line, void *caller, unsigned taint, 517 struct pt_regs *regs, struct warn_args *args) 518 { 519 disable_trace_on_warning(); 520 521 pr_warn("------------[ cut here ]------------\n"); 522 523 if (file) 524 pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS\n", 525 raw_smp_processor_id(), current->pid, file, line, 526 caller); 527 else 528 pr_warn("WARNING: CPU: %d PID: %d at %pS\n", 529 raw_smp_processor_id(), current->pid, caller); 530 531 if (args) 532 vprintk(args->fmt, args->args); 533 534 if (panic_on_warn) { 535 /* 536 * This thread may hit another WARN() in the panic path. 537 * Resetting this prevents additional WARN() from panicking the 538 * system on this thread. Other threads are blocked by the 539 * panic_mutex in panic(). 540 */ 541 panic_on_warn = 0; 542 panic("panic_on_warn set ...\n"); 543 } 544 545 print_modules(); 546 547 if (regs) 548 show_regs(regs); 549 else 550 dump_stack(); 551 552 print_oops_end_marker(); 553 554 /* Just a warning, don't kill lockdep. */ 555 add_taint(taint, LOCKDEP_STILL_OK); 556 } 557 558 #ifdef WANT_WARN_ON_SLOWPATH 559 void warn_slowpath_fmt(const char *file, int line, const char *fmt, ...) 560 { 561 struct warn_args args; 562 563 args.fmt = fmt; 564 va_start(args.args, fmt); 565 __warn(file, line, __builtin_return_address(0), TAINT_WARN, NULL, 566 &args); 567 va_end(args.args); 568 } 569 EXPORT_SYMBOL(warn_slowpath_fmt); 570 571 void warn_slowpath_fmt_taint(const char *file, int line, 572 unsigned taint, const char *fmt, ...) 573 { 574 struct warn_args args; 575 576 args.fmt = fmt; 577 va_start(args.args, fmt); 578 __warn(file, line, __builtin_return_address(0), taint, NULL, &args); 579 va_end(args.args); 580 } 581 EXPORT_SYMBOL(warn_slowpath_fmt_taint); 582 583 void warn_slowpath_null(const char *file, int line) 584 { 585 __warn(file, line, __builtin_return_address(0), TAINT_WARN, NULL, NULL); 586 } 587 EXPORT_SYMBOL(warn_slowpath_null); 588 #endif 589 590 #ifdef CONFIG_CC_STACKPROTECTOR 591 592 /* 593 * Called when gcc's -fstack-protector feature is used, and 594 * gcc detects corruption of the on-stack canary value 595 */ 596 __visible void __stack_chk_fail(void) 597 { 598 panic("stack-protector: Kernel stack is corrupted in: %p\n", 599 __builtin_return_address(0)); 600 } 601 EXPORT_SYMBOL(__stack_chk_fail); 602 603 #endif 604 605 core_param(panic, panic_timeout, int, 0644); 606 core_param(pause_on_oops, pause_on_oops, int, 0644); 607 core_param(panic_on_warn, panic_on_warn, int, 0644); 608 core_param(crash_kexec_post_notifiers, crash_kexec_post_notifiers, bool, 0644); 609 610 static int __init oops_setup(char *s) 611 { 612 if (!s) 613 return -EINVAL; 614 if (!strcmp(s, "panic")) 615 panic_on_oops = 1; 616 return 0; 617 } 618 early_param("oops", oops_setup); 619