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/reboot.h>
  19 #include <linux/delay.h>
  20 #include <linux/kexec.h>
  21 #include <linux/sched.h>
  22 #include <linux/sysrq.h>
  23 #include <linux/init.h>
  24 #include <linux/nmi.h>
  25 #include <linux/dmi.h>
  26
  27 #define PANIC_TIMER_STEP 100
  28 #define PANIC_BLINK_SPD 18
  29
  30 int panic_on_oops = CONFIG_PANIC_ON_OOPS_VALUE;
  31 static unsigned long tainted_mask;
  32 static int pause_on_oops;
  33 static int pause_on_oops_flag;
  34 static DEFINE_SPINLOCK(pause_on_oops_lock);
  35
  36 int panic_timeout;
  37 EXPORT_SYMBOL_GPL(panic_timeout);
  38
  39 ATOMIC_NOTIFIER_HEAD(panic_notifier_list);
  40
  41 EXPORT_SYMBOL(panic_notifier_list);
  42
  43 static long no_blink(int state)
  44 {
  45         return 0;
  46 }
  47
  48 /* Returns how long it waited in ms */
  49 long (*panic_blink)(int state);
  50 EXPORT_SYMBOL(panic_blink);
  51
  52 /*
  53  * Stop ourself in panic -- architecture code may override this
  54  */
  55 void __weak panic_smp_self_stop(void)
  56 {
  57         while (1)
  58                 cpu_relax();
  59 }
  60
  61 /**
  62  *      panic - halt the system
  63  *      @fmt: The text string to print
  64  *
  65  *      Display a message, then perform cleanups.
  66  *
  67  *      This function never returns.
  68  */
  69 void panic(const char *fmt, ...)
  70 {
  71         static DEFINE_SPINLOCK(panic_lock);
  72         static char buf[1024];
  73         va_list args;
  74         long i, i_next = 0;
  75         int state = 0;
  76
  77         /*
  78          * Disable local interrupts. This will prevent panic_smp_self_stop
  79          * from deadlocking the first cpu that invokes the panic, since
  80          * there is nothing to prevent an interrupt handler (that runs
  81          * after the panic_lock is acquired) from invoking panic again.
  82          */
  83         local_irq_disable();
  84
  85         /*
  86          * It's possible to come here directly from a panic-assertion and
  87          * not have preempt disabled. Some functions called from here want
  88          * preempt to be disabled. No point enabling it later though...
  89          *
  90          * Only one CPU is allowed to execute the panic code from here. For
  91          * multiple parallel invocations of panic, all other CPUs either
  92          * stop themself or will wait until they are stopped by the 1st CPU
  93          * with smp_send_stop().
  94          */
  95         if (!spin_trylock(&panic_lock))
  96                 panic_smp_self_stop();
  97
  98         console_verbose();
  99         bust_spinlocks(1);
 100         va_start(args, fmt);
 101         vsnprintf(buf, sizeof(buf), fmt, args);
 102         va_end(args);
 103         printk(KERN_EMERG "Kernel panic - not syncing: %s\n",buf);
 104 #ifdef CONFIG_DEBUG_BUGVERBOSE
 105         /*
 106          * Avoid nested stack-dumping if a panic occurs during oops processing
 107          */
 108         if (!test_taint(TAINT_DIE) && oops_in_progress <= 1)
 109                 dump_stack();
 110 #endif
 111
 112         /*
 113          * If we have crashed and we have a crash kernel loaded let it handle
 114          * everything else.
 115          * Do we want to call this before we try to display a message?
 116          */
 117         crash_kexec(NULL);
 118
 119         /*
 120          * Note smp_send_stop is the usual smp shutdown function, which
 121          * unfortunately means it may not be hardened to work in a panic
 122          * situation.
 123          */
 124         smp_send_stop();
 125
 126         kmsg_dump(KMSG_DUMP_PANIC);
 127
 128         atomic_notifier_call_chain(&panic_notifier_list, 0, buf);
 129
 130         bust_spinlocks(0);
 131
 132         if (!panic_blink)
 133                 panic_blink = no_blink;
 134
 135         if (panic_timeout > 0) {
 136                 /*
 137                  * Delay timeout seconds before rebooting the machine.
 138                  * We can't use the "normal" timers since we just panicked.
 139                  */
 140                 printk(KERN_EMERG "Rebooting in %d seconds..", panic_timeout);
 141
 142                 for (i = 0; i < panic_timeout * 1000; i += PANIC_TIMER_STEP) {
 143                         touch_nmi_watchdog();
 144                         if (i >= i_next) {
 145                                 i += panic_blink(state ^= 1);
 146                                 i_next = i + 3600 / PANIC_BLINK_SPD;
 147                         }
 148                         mdelay(PANIC_TIMER_STEP);
 149                 }
 150         }
 151         if (panic_timeout != 0) {
 152                 /*
 153                  * This will not be a clean reboot, with everything
 154                  * shutting down.  But if there is a chance of
 155                  * rebooting the system it will be rebooted.
 156                  */
 157                 emergency_restart();
 158         }
 159 #ifdef __sparc__
 160         {
 161                 extern int stop_a_enabled;
 162                 /* Make sure the user can actually press Stop-A (L1-A) */
 163                 stop_a_enabled = 1;
 164                 printk(KERN_EMERG "Press Stop-A (L1-A) to return to the boot prom\n");
 165         }
 166 #endif
 167 #if defined(CONFIG_S390)
 168         {
 169                 unsigned long caller;
 170
 171                 caller = (unsigned long)__builtin_return_address(0);
 172                 disabled_wait(caller);
 173         }
 174 #endif
 175         local_irq_enable();
 176         for (i = 0; ; i += PANIC_TIMER_STEP) {
 177                 touch_softlockup_watchdog();
 178                 if (i >= i_next) {
 179                         i += panic_blink(state ^= 1);
 180                         i_next = i + 3600 / PANIC_BLINK_SPD;
 181                 }
 182                 mdelay(PANIC_TIMER_STEP);
 183         }
 184 }
 185
 186 EXPORT_SYMBOL(panic);
 187
 188
 189 struct tnt {
 190         u8      bit;
 191         char    true;
 192         char    false;
 193 };
 194
 195 static const struct tnt tnts[] = {
 196         { TAINT_PROPRIETARY_MODULE,     'P', 'G' },
 197         { TAINT_FORCED_MODULE,          'F', ' ' },
 198         { TAINT_UNSAFE_SMP,             'S', ' ' },
 199         { TAINT_FORCED_RMMOD,           'R', ' ' },
 200         { TAINT_MACHINE_CHECK,          'M', ' ' },
 201         { TAINT_BAD_PAGE,               'B', ' ' },
 202         { TAINT_USER,                   'U', ' ' },
 203         { TAINT_DIE,                    'D', ' ' },
 204         { TAINT_OVERRIDDEN_ACPI_TABLE,  'A', ' ' },
 205         { TAINT_WARN,                   'W', ' ' },
 206         { TAINT_CRAP,                   'C', ' ' },
 207         { TAINT_FIRMWARE_WORKAROUND,    'I', ' ' },
 208         { TAINT_OOT_MODULE,             'O', ' ' },
 209 };
 210
 211 /**
 212  *      print_tainted - return a string to represent the kernel taint state.
 213  *
 214  *  'P' - Proprietary module has been loaded.
 215  *  'F' - Module has been forcibly loaded.
 216  *  'S' - SMP with CPUs not designed for SMP.
 217  *  'R' - User forced a module unload.
 218  *  'M' - System experienced a machine check exception.
 219  *  'B' - System has hit bad_page.
 220  *  'U' - Userspace-defined naughtiness.
 221  *  'D' - Kernel has oopsed before
 222  *  'A' - ACPI table overridden.
 223  *  'W' - Taint on warning.
 224  *  'C' - modules from drivers/staging are loaded.
 225  *  'I' - Working around severe firmware bug.
 226  *  'O' - Out-of-tree module has been loaded.
 227  *
 228  *      The string is overwritten by the next call to print_tainted().
 229  */
 230 const char *print_tainted(void)
 231 {
 232         static char buf[ARRAY_SIZE(tnts) + sizeof("Tainted: ") + 1];
 233
 234         if (tainted_mask) {
 235                 char *s;
 236                 int i;
 237
 238                 s = buf + sprintf(buf, "Tainted: ");
 239                 for (i = 0; i < ARRAY_SIZE(tnts); i++) {
 240                         const struct tnt *t = &tnts[i];
 241                         *s++ = test_bit(t->bit, &tainted_mask) ?
 242                                         t->true : t->false;
 243                 }
 244                 *s = 0;
 245         } else
 246                 snprintf(buf, sizeof(buf), "Not tainted");
 247
 248         return buf;
 249 }
 250
 251 int test_taint(unsigned flag)
 252 {
 253         return test_bit(flag, &tainted_mask);
 254 }
 255 EXPORT_SYMBOL(test_taint);
 256
 257 unsigned long get_taint(void)
 258 {
 259         return tainted_mask;
 260 }
 261
 262 void add_taint(unsigned flag)
 263 {
 264         /*
 265          * Can't trust the integrity of the kernel anymore.
 266          * We don't call directly debug_locks_off() because the issue
 267          * is not necessarily serious enough to set oops_in_progress to 1
 268          * Also we want to keep up lockdep for staging/out-of-tree
 269          * development and post-warning case.
 270          */
 271         switch (flag) {
 272         case TAINT_CRAP:
 273         case TAINT_OOT_MODULE:
 274         case TAINT_WARN:
 275         case TAINT_FIRMWARE_WORKAROUND:
 276                 break;
 277
 278         default:
 279                 if (__debug_locks_off())
 280                         printk(KERN_WARNING "Disabling lock debugging due to kernel taint\n");
 281         }
 282
 283         set_bit(flag, &tainted_mask);
 284 }
 285 EXPORT_SYMBOL(add_taint);
 286
 287 static void spin_msec(int msecs)
 288 {
 289         int i;
 290
 291         for (i = 0; i < msecs; i++) {
 292                 touch_nmi_watchdog();
 293                 mdelay(1);
 294         }
 295 }
 296
 297 /*
 298  * It just happens that oops_enter() and oops_exit() are identically
 299  * implemented...
 300  */
 301 static void do_oops_enter_exit(void)
 302 {
 303         unsigned long flags;
 304         static int spin_counter;
 305
 306         if (!pause_on_oops)
 307                 return;
 308
 309         spin_lock_irqsave(&pause_on_oops_lock, flags);
 310         if (pause_on_oops_flag == 0) {
 311                 /* This CPU may now print the oops message */
 312                 pause_on_oops_flag = 1;
 313         } else {
 314                 /* We need to stall this CPU */
 315                 if (!spin_counter) {
 316                         /* This CPU gets to do the counting */
 317                         spin_counter = pause_on_oops;
 318                         do {
 319                                 spin_unlock(&pause_on_oops_lock);
 320                                 spin_msec(MSEC_PER_SEC);
 321                                 spin_lock(&pause_on_oops_lock);
 322                         } while (--spin_counter);
 323                         pause_on_oops_flag = 0;
 324                 } else {
 325                         /* This CPU waits for a different one */
 326                         while (spin_counter) {
 327                                 spin_unlock(&pause_on_oops_lock);
 328                                 spin_msec(1);
 329                                 spin_lock(&pause_on_oops_lock);
 330                         }
 331                 }
 332         }
 333         spin_unlock_irqrestore(&pause_on_oops_lock, flags);
 334 }
 335
 336 /*
 337  * Return true if the calling CPU is allowed to print oops-related info.
 338  * This is a bit racy..
 339  */
 340 int oops_may_print(void)
 341 {
 342         return pause_on_oops_flag == 0;
 343 }
 344
 345 /*
 346  * Called when the architecture enters its oops handler, before it prints
 347  * anything.  If this is the first CPU to oops, and it's oopsing the first
 348  * time then let it proceed.
 349  *
 350  * This is all enabled by the pause_on_oops kernel boot option.  We do all
 351  * this to ensure that oopses don't scroll off the screen.  It has the
 352  * side-effect of preventing later-oopsing CPUs from mucking up the display,
 353  * too.
 354  *
 355  * It turns out that the CPU which is allowed to print ends up pausing for
 356  * the right duration, whereas all the other CPUs pause for twice as long:
 357  * once in oops_enter(), once in oops_exit().
 358  */
 359 void oops_enter(void)
 360 {
 361         tracing_off();
 362         /* can't trust the integrity of the kernel anymore: */
 363         debug_locks_off();
 364         do_oops_enter_exit();
 365 }
 366
 367 /*
 368  * 64-bit random ID for oopses:
 369  */
 370 static u64 oops_id;
 371
 372 static int init_oops_id(void)
 373 {
 374         if (!oops_id)
 375                 get_random_bytes(&oops_id, sizeof(oops_id));
 376         else
 377                 oops_id++;
 378
 379         return 0;
 380 }
 381 late_initcall(init_oops_id);
 382
 383 void print_oops_end_marker(void)
 384 {
 385         init_oops_id();
 386         printk(KERN_WARNING "---[ end trace %016llx ]---\n",
 387                 (unsigned long long)oops_id);
 388 }
 389
 390 /*
 391  * Called when the architecture exits its oops handler, after printing
 392  * everything.
 393  */
 394 void oops_exit(void)
 395 {
 396         do_oops_enter_exit();
 397         print_oops_end_marker();
 398         kmsg_dump(KMSG_DUMP_OOPS);
 399 }
 400
 401 #ifdef WANT_WARN_ON_SLOWPATH
 402 struct slowpath_args {
 403         const char *fmt;
 404         va_list args;
 405 };
 406
 407 static void warn_slowpath_common(const char *file, int line, void *caller,
 408                                  unsigned taint, struct slowpath_args *args)
 409 {
 410         const char *board;
 411
 412         printk(KERN_WARNING "------------[ cut here ]------------\n");
 413         printk(KERN_WARNING "WARNING: at %s:%d %pS()\n", file, line, caller);
 414         board = dmi_get_system_info(DMI_PRODUCT_NAME);
 415         if (board)
 416                 printk(KERN_WARNING "Hardware name: %s\n", board);
 417
 418         if (args)
 419                 vprintk(args->fmt, args->args);
 420
 421         print_modules();
 422         dump_stack();
 423         print_oops_end_marker();
 424         add_taint(taint);
 425 }
 426
 427 void warn_slowpath_fmt(const char *file, int line, const char *fmt, ...)
 428 {
 429         struct slowpath_args args;
 430
 431         args.fmt = fmt;
 432         va_start(args.args, fmt);
 433         warn_slowpath_common(file, line, __builtin_return_address(0),
 434                              TAINT_WARN, &args);
 435         va_end(args.args);
 436 }
 437 EXPORT_SYMBOL(warn_slowpath_fmt);
 438
 439 void warn_slowpath_fmt_taint(const char *file, int line,
 440                              unsigned taint, const char *fmt, ...)
 441 {
 442         struct slowpath_args args;
 443
 444         args.fmt = fmt;
 445         va_start(args.args, fmt);
 446         warn_slowpath_common(file, line, __builtin_return_address(0),
 447                              taint, &args);
 448         va_end(args.args);
 449 }
 450 EXPORT_SYMBOL(warn_slowpath_fmt_taint);
 451
 452 void warn_slowpath_null(const char *file, int line)
 453 {
 454         warn_slowpath_common(file, line, __builtin_return_address(0),
 455                              TAINT_WARN, NULL);
 456 }
 457 EXPORT_SYMBOL(warn_slowpath_null);
 458 #endif
 459
 460 #ifdef CONFIG_CC_STACKPROTECTOR
 461
 462 /*
 463  * Called when gcc's -fstack-protector feature is used, and
 464  * gcc detects corruption of the on-stack canary value
 465  */
 466 void __stack_chk_fail(void)
 467 {
 468         panic("stack-protector: Kernel stack is corrupted in: %p\n",
 469                 __builtin_return_address(0));
 470 }
 471 EXPORT_SYMBOL(__stack_chk_fail);
 472
 473 #endif
 474
 475 core_param(panic, panic_timeout, int, 0644);
 476 core_param(pause_on_oops, pause_on_oops, int, 0644);
 477
 478 static int __init oops_setup(char *s)
 479 {
 480         if (!s)
 481                 return -EINVAL;
 482         if (!strcmp(s, "panic"))
 483                 panic_on_oops = 1;
 484         return 0;
 485 }
 486 early_param("oops", oops_setup);