kernel/panic.c

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