arch/x86_64/kernel/traps.c

   1 /*
   2  *  linux/arch/x86-64/traps.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  *  Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs
   6  *
   7  *  Pentium III FXSR, SSE support
   8  *      Gareth Hughes <gareth@valinux.com>, May 2000
   9  */
  10
  11 /*
  12  * 'Traps.c' handles hardware traps and faults after we have saved some
  13  * state in 'entry.S'.
  14  */
  15 #include <linux/sched.h>
  16 #include <linux/kernel.h>
  17 #include <linux/string.h>
  18 #include <linux/errno.h>
  19 #include <linux/ptrace.h>
  20 #include <linux/timer.h>
  21 #include <linux/mm.h>
  22 #include <linux/init.h>
  23 #include <linux/delay.h>
  24 #include <linux/spinlock.h>
  25 #include <linux/interrupt.h>
  26 #include <linux/module.h>
  27 #include <linux/moduleparam.h>
  28 #include <linux/nmi.h>
  29 #include <linux/kprobes.h>
  30 #include <linux/kexec.h>
  31 #include <linux/unwind.h>
  32
  33 #include <asm/system.h>
  34 #include <asm/uaccess.h>
  35 #include <asm/io.h>
  36 #include <asm/atomic.h>
  37 #include <asm/debugreg.h>
  38 #include <asm/desc.h>
  39 #include <asm/i387.h>
  40 #include <asm/kdebug.h>
  41 #include <asm/processor.h>
  42 #include <asm/unwind.h>
  43 #include <asm/smp.h>
  44 #include <asm/pgalloc.h>
  45 #include <asm/pda.h>
  46 #include <asm/proto.h>
  47 #include <asm/nmi.h>
  48 #include <asm/stacktrace.h>
  49
  50 asmlinkage void divide_error(void);
  51 asmlinkage void debug(void);
  52 asmlinkage void nmi(void);
  53 asmlinkage void int3(void);
  54 asmlinkage void overflow(void);
  55 asmlinkage void bounds(void);
  56 asmlinkage void invalid_op(void);
  57 asmlinkage void device_not_available(void);
  58 asmlinkage void double_fault(void);
  59 asmlinkage void coprocessor_segment_overrun(void);
  60 asmlinkage void invalid_TSS(void);
  61 asmlinkage void segment_not_present(void);
  62 asmlinkage void stack_segment(void);
  63 asmlinkage void general_protection(void);
  64 asmlinkage void page_fault(void);
  65 asmlinkage void coprocessor_error(void);
  66 asmlinkage void simd_coprocessor_error(void);
  67 asmlinkage void reserved(void);
  68 asmlinkage void alignment_check(void);
  69 asmlinkage void machine_check(void);
  70 asmlinkage void spurious_interrupt_bug(void);
  71
  72 ATOMIC_NOTIFIER_HEAD(die_chain);
  73 EXPORT_SYMBOL(die_chain);
  74
  75 int register_die_notifier(struct notifier_block *nb)
  76 {
  77         vmalloc_sync_all();
  78         return atomic_notifier_chain_register(&die_chain, nb);
  79 }
  80 EXPORT_SYMBOL(register_die_notifier); /* used modular by kdb */
  81
  82 int unregister_die_notifier(struct notifier_block *nb)
  83 {
  84         return atomic_notifier_chain_unregister(&die_chain, nb);
  85 }
  86 EXPORT_SYMBOL(unregister_die_notifier); /* used modular by kdb */
  87
  88 static inline void conditional_sti(struct pt_regs *regs)
  89 {
  90         if (regs->eflags & X86_EFLAGS_IF)
  91                 local_irq_enable();
  92 }
  93
  94 static inline void preempt_conditional_sti(struct pt_regs *regs)
  95 {
  96         preempt_disable();
  97         if (regs->eflags & X86_EFLAGS_IF)
  98                 local_irq_enable();
  99 }
 100
 101 static inline void preempt_conditional_cli(struct pt_regs *regs)
 102 {
 103         if (regs->eflags & X86_EFLAGS_IF)
 104                 local_irq_disable();
 105         /* Make sure to not schedule here because we could be running
 106            on an exception stack. */
 107         preempt_enable_no_resched();
 108 }
 109
 110 static int kstack_depth_to_print = 12;
 111 #ifdef CONFIG_STACK_UNWIND
 112 static int call_trace = 1;
 113 #else
 114 #define call_trace (-1)
 115 #endif
 116
 117 #ifdef CONFIG_KALLSYMS
 118 # include <linux/kallsyms.h>
 119 void printk_address(unsigned long address)
 120 {
 121         unsigned long offset = 0, symsize;
 122         const char *symname;
 123         char *modname;
 124         char *delim = ":";
 125         char namebuf[128];
 126
 127         symname = kallsyms_lookup(address, &symsize, &offset,
 128                                         &modname, namebuf);
 129         if (!symname) {
 130                 printk(" [<%016lx>]\n", address);
 131                 return;
 132         }
 133         if (!modname)
 134                 modname = delim = "";
 135         printk(" [<%016lx>] %s%s%s%s+0x%lx/0x%lx\n",
 136                 address, delim, modname, delim, symname, offset, symsize);
 137 }
 138 #else
 139 void printk_address(unsigned long address)
 140 {
 141         printk(" [<%016lx>]\n", address);
 142 }
 143 #endif
 144
 145 static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
 146                                         unsigned *usedp, char **idp)
 147 {
 148         static char ids[][8] = {
 149                 [DEBUG_STACK - 1] = "#DB",
 150                 [NMI_STACK - 1] = "NMI",
 151                 [DOUBLEFAULT_STACK - 1] = "#DF",
 152                 [STACKFAULT_STACK - 1] = "#SS",
 153                 [MCE_STACK - 1] = "#MC",
 154 #if DEBUG_STKSZ > EXCEPTION_STKSZ
 155                 [N_EXCEPTION_STACKS ... N_EXCEPTION_STACKS + DEBUG_STKSZ / EXCEPTION_STKSZ - 2] = "#DB[?]"
 156 #endif
 157         };
 158         unsigned k;
 159
 160         /*
 161          * Iterate over all exception stacks, and figure out whether
 162          * 'stack' is in one of them:
 163          */
 164         for (k = 0; k < N_EXCEPTION_STACKS; k++) {
 165                 unsigned long end = per_cpu(orig_ist, cpu).ist[k];
 166                 /*
 167                  * Is 'stack' above this exception frame's end?
 168                  * If yes then skip to the next frame.
 169                  */
 170                 if (stack >= end)
 171                         continue;
 172                 /*
 173                  * Is 'stack' above this exception frame's start address?
 174                  * If yes then we found the right frame.
 175                  */
 176                 if (stack >= end - EXCEPTION_STKSZ) {
 177                         /*
 178                          * Make sure we only iterate through an exception
 179                          * stack once. If it comes up for the second time
 180                          * then there's something wrong going on - just
 181                          * break out and return NULL:
 182                          */
 183                         if (*usedp & (1U << k))
 184                                 break;
 185                         *usedp |= 1U << k;
 186                         *idp = ids[k];
 187                         return (unsigned long *)end;
 188                 }
 189                 /*
 190                  * If this is a debug stack, and if it has a larger size than
 191                  * the usual exception stacks, then 'stack' might still
 192                  * be within the lower portion of the debug stack:
 193                  */
 194 #if DEBUG_STKSZ > EXCEPTION_STKSZ
 195                 if (k == DEBUG_STACK - 1 && stack >= end - DEBUG_STKSZ) {
 196                         unsigned j = N_EXCEPTION_STACKS - 1;
 197
 198                         /*
 199                          * Black magic. A large debug stack is composed of
 200                          * multiple exception stack entries, which we
 201                          * iterate through now. Dont look:
 202                          */
 203                         do {
 204                                 ++j;
 205                                 end -= EXCEPTION_STKSZ;
 206                                 ids[j][4] = '1' + (j - N_EXCEPTION_STACKS);
 207                         } while (stack < end - EXCEPTION_STKSZ);
 208                         if (*usedp & (1U << j))
 209                                 break;
 210                         *usedp |= 1U << j;
 211                         *idp = ids[j];
 212                         return (unsigned long *)end;
 213                 }
 214 #endif
 215         }
 216         return NULL;
 217 }
 218
 219 struct ops_and_data {
 220         struct stacktrace_ops *ops;
 221         void *data;
 222 };
 223
 224 static int dump_trace_unwind(struct unwind_frame_info *info, void *context)
 225 {
 226         struct ops_and_data *oad = (struct ops_and_data *)context;
 227         int n = 0;
 228
 229         while (unwind(info) == 0 && UNW_PC(info)) {
 230                 n++;
 231                 oad->ops->address(oad->data, UNW_PC(info));
 232                 if (arch_unw_user_mode(info))
 233                         break;
 234         }
 235         return n;
 236 }
 237
 238 /*
 239  * x86-64 can have upto three kernel stacks:
 240  * process stack
 241  * interrupt stack
 242  * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
 243  */
 244
 245 void dump_trace(struct task_struct *tsk, struct pt_regs *regs, unsigned long * stack,
 246                 struct stacktrace_ops *ops, void *data)
 247 {
 248         const unsigned cpu = smp_processor_id();
 249         unsigned long *irqstack_end = (unsigned long *)cpu_pda(cpu)->irqstackptr;
 250         unsigned used = 0;
 251
 252         if (!tsk)
 253                 tsk = current;
 254
 255         if (call_trace >= 0) {
 256                 int unw_ret = 0;
 257                 struct unwind_frame_info info;
 258                 struct ops_and_data oad = { .ops = ops, .data = data };
 259
 260                 if (regs) {
 261                         if (unwind_init_frame_info(&info, tsk, regs) == 0)
 262                                 unw_ret = dump_trace_unwind(&info, &oad);
 263                 } else if (tsk == current)
 264                         unw_ret = unwind_init_running(&info, dump_trace_unwind, &oad);
 265                 else {
 266                         if (unwind_init_blocked(&info, tsk) == 0)
 267                                 unw_ret = dump_trace_unwind(&info, &oad);
 268                 }
 269                 if (unw_ret > 0) {
 270                         if (call_trace == 1 && !arch_unw_user_mode(&info)) {
 271                                 ops->warning_symbol(data, "DWARF2 unwinder stuck at %s\n",
 272                                              UNW_PC(&info));
 273                                 if ((long)UNW_SP(&info) < 0) {
 274                                         ops->warning(data, "Leftover inexact backtrace:\n");
 275                                         stack = (unsigned long *)UNW_SP(&info);
 276                                         if (!stack)
 277                                                 return;
 278                                 } else
 279                                         ops->warning(data, "Full inexact backtrace again:\n");
 280                         } else if (call_trace >= 1)
 281                                 return;
 282                         else
 283                                 ops->warning(data, "Full inexact backtrace again:\n");
 284                 } else
 285                         ops->warning(data, "Inexact backtrace:\n");
 286         }
 287         if (!stack) {
 288                 unsigned long dummy;
 289                 stack = &dummy;
 290                 if (tsk && tsk != current)
 291                         stack = (unsigned long *)tsk->thread.rsp;
 292         }
 293
 294         /*
 295          * Print function call entries within a stack. 'cond' is the
 296          * "end of stackframe" condition, that the 'stack++'
 297          * iteration will eventually trigger.
 298          */
 299 #define HANDLE_STACK(cond) \
 300         do while (cond) { \
 301                 unsigned long addr = *stack++; \
 302                 if (oops_in_progress ?          \
 303                         __kernel_text_address(addr) : \
 304                         kernel_text_address(addr)) { \
 305                         /* \
 306                          * If the address is either in the text segment of the \
 307                          * kernel, or in the region which contains vmalloc'ed \
 308                          * memory, it *may* be the address of a calling \
 309                          * routine; if so, print it so that someone tracing \
 310                          * down the cause of the crash will be able to figure \
 311                          * out the call path that was taken. \
 312                          */ \
 313                         ops->address(data, addr);   \
 314                 } \
 315         } while (0)
 316
 317         /*
 318          * Print function call entries in all stacks, starting at the
 319          * current stack address. If the stacks consist of nested
 320          * exceptions
 321          */
 322         for (;;) {
 323                 char *id;
 324                 unsigned long *estack_end;
 325                 estack_end = in_exception_stack(cpu, (unsigned long)stack,
 326                                                 &used, &id);
 327
 328                 if (estack_end) {
 329                         if (ops->stack(data, id) < 0)
 330                                 break;
 331                         HANDLE_STACK (stack < estack_end);
 332                         ops->stack(data, "<EOE>");
 333                         /*
 334                          * We link to the next stack via the
 335                          * second-to-last pointer (index -2 to end) in the
 336                          * exception stack:
 337                          */
 338                         stack = (unsigned long *) estack_end[-2];
 339                         continue;
 340                 }
 341                 if (irqstack_end) {
 342                         unsigned long *irqstack;
 343                         irqstack = irqstack_end -
 344                                 (IRQSTACKSIZE - 64) / sizeof(*irqstack);
 345
 346                         if (stack >= irqstack && stack < irqstack_end) {
 347                                 if (ops->stack(data, "IRQ") < 0)
 348                                         break;
 349                                 HANDLE_STACK (stack < irqstack_end);
 350                                 /*
 351                                  * We link to the next stack (which would be
 352                                  * the process stack normally) the last
 353                                  * pointer (index -1 to end) in the IRQ stack:
 354                                  */
 355                                 stack = (unsigned long *) (irqstack_end[-1]);
 356                                 irqstack_end = NULL;
 357                                 ops->stack(data, "EOI");
 358                                 continue;
 359                         }
 360                 }
 361                 break;
 362         }
 363
 364         /*
 365          * This handles the process stack:
 366          */
 367         HANDLE_STACK (((long) stack & (THREAD_SIZE-1)) != 0);
 368 #undef HANDLE_STACK
 369 }
 370 EXPORT_SYMBOL(dump_trace);
 371
 372 static void
 373 print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
 374 {
 375         print_symbol(msg, symbol);
 376         printk("\n");
 377 }
 378
 379 static void print_trace_warning(void *data, char *msg)
 380 {
 381         printk("%s\n", msg);
 382 }
 383
 384 static int print_trace_stack(void *data, char *name)
 385 {
 386         printk(" <%s> ", name);
 387         return 0;
 388 }
 389
 390 static void print_trace_address(void *data, unsigned long addr)
 391 {
 392         printk_address(addr);
 393 }
 394
 395 static struct stacktrace_ops print_trace_ops = {
 396         .warning = print_trace_warning,
 397         .warning_symbol = print_trace_warning_symbol,
 398         .stack = print_trace_stack,
 399         .address = print_trace_address,
 400 };
 401
 402 void
 403 show_trace(struct task_struct *tsk, struct pt_regs *regs, unsigned long *stack)
 404 {
 405         printk("\nCall Trace:\n");
 406         dump_trace(tsk, regs, stack, &print_trace_ops, NULL);
 407         printk("\n");
 408 }
 409
 410 static void
 411 _show_stack(struct task_struct *tsk, struct pt_regs *regs, unsigned long *rsp)
 412 {
 413         unsigned long *stack;
 414         int i;
 415         const int cpu = smp_processor_id();
 416         unsigned long *irqstack_end = (unsigned long *) (cpu_pda(cpu)->irqstackptr);
 417         unsigned long *irqstack = (unsigned long *) (cpu_pda(cpu)->irqstackptr - IRQSTACKSIZE);
 418
 419         // debugging aid: "show_stack(NULL, NULL);" prints the
 420         // back trace for this cpu.
 421
 422         if (rsp == NULL) {
 423                 if (tsk)
 424                         rsp = (unsigned long *)tsk->thread.rsp;
 425                 else
 426                         rsp = (unsigned long *)&rsp;
 427         }
 428
 429         stack = rsp;
 430         for(i=0; i < kstack_depth_to_print; i++) {
 431                 if (stack >= irqstack && stack <= irqstack_end) {
 432                         if (stack == irqstack_end) {
 433                                 stack = (unsigned long *) (irqstack_end[-1]);
 434                                 printk(" <EOI> ");
 435                         }
 436                 } else {
 437                 if (((long) stack & (THREAD_SIZE-1)) == 0)
 438                         break;
 439                 }
 440                 if (i && ((i % 4) == 0))
 441                         printk("\n");
 442                 printk(" %016lx", *stack++);
 443                 touch_nmi_watchdog();
 444         }
 445         show_trace(tsk, regs, rsp);
 446 }
 447
 448 void show_stack(struct task_struct *tsk, unsigned long * rsp)
 449 {
 450         _show_stack(tsk, NULL, rsp);
 451 }
 452
 453 /*
 454  * The architecture-independent dump_stack generator
 455  */
 456 void dump_stack(void)
 457 {
 458         unsigned long dummy;
 459         show_trace(NULL, NULL, &dummy);
 460 }
 461
 462 EXPORT_SYMBOL(dump_stack);
 463
 464 void show_registers(struct pt_regs *regs)
 465 {
 466         int i;
 467         int in_kernel = !user_mode(regs);
 468         unsigned long rsp;
 469         const int cpu = smp_processor_id();
 470         struct task_struct *cur = cpu_pda(cpu)->pcurrent;
 471
 472                 rsp = regs->rsp;
 473
 474         printk("CPU %d ", cpu);
 475         __show_regs(regs);
 476         printk("Process %s (pid: %d, threadinfo %p, task %p)\n",
 477                 cur->comm, cur->pid, task_thread_info(cur), cur);
 478
 479         /*
 480          * When in-kernel, we also print out the stack and code at the
 481          * time of the fault..
 482          */
 483         if (in_kernel) {
 484
 485                 printk("Stack: ");
 486                 _show_stack(NULL, regs, (unsigned long*)rsp);
 487
 488                 printk("\nCode: ");
 489                 if (regs->rip < PAGE_OFFSET)
 490                         goto bad;
 491
 492                 for (i=0; i<20; i++) {
 493                         unsigned char c;
 494                         if (__get_user(c, &((unsigned char*)regs->rip)[i])) {
 495 bad:
 496                                 printk(" Bad RIP value.");
 497                                 break;
 498                         }
 499                         printk("%02x ", c);
 500                 }
 501         }
 502         printk("\n");
 503 }
 504
 505 void handle_BUG(struct pt_regs *regs)
 506 {
 507         struct bug_frame f;
 508         long len;
 509         const char *prefix = "";
 510
 511         if (user_mode(regs))
 512                 return;
 513         if (__copy_from_user(&f, (const void __user *) regs->rip,
 514                              sizeof(struct bug_frame)))
 515                 return;
 516         if (f.filename >= 0 ||
 517             f.ud2[0] != 0x0f || f.ud2[1] != 0x0b)
 518                 return;
 519         len = __strnlen_user((char *)(long)f.filename, PATH_MAX) - 1;
 520         if (len < 0 || len >= PATH_MAX)
 521                 f.filename = (int)(long)"unmapped filename";
 522         else if (len > 50) {
 523                 f.filename += len - 50;
 524                 prefix = "...";
 525         }
 526         printk("----------- [cut here ] --------- [please bite here ] ---------\n");
 527         printk(KERN_ALERT "Kernel BUG at %s%.50s:%d\n", prefix, (char *)(long)f.filename, f.line);
 528 }
 529
 530 #ifdef CONFIG_BUG
 531 void out_of_line_bug(void)
 532 {
 533         BUG();
 534 }
 535 EXPORT_SYMBOL(out_of_line_bug);
 536 #endif
 537
 538 static DEFINE_SPINLOCK(die_lock);
 539 static int die_owner = -1;
 540 static unsigned int die_nest_count;
 541
 542 unsigned __kprobes long oops_begin(void)
 543 {
 544         int cpu = smp_processor_id();
 545         unsigned long flags;
 546
 547         oops_enter();
 548
 549         /* racy, but better than risking deadlock. */
 550         local_irq_save(flags);
 551         if (!spin_trylock(&die_lock)) {
 552                 if (cpu == die_owner)
 553                         /* nested oops. should stop eventually */;
 554                 else
 555                         spin_lock(&die_lock);
 556         }
 557         die_nest_count++;
 558         die_owner = cpu;
 559         console_verbose();
 560         bust_spinlocks(1);
 561         return flags;
 562 }
 563
 564 void __kprobes oops_end(unsigned long flags)
 565 {
 566         die_owner = -1;
 567         bust_spinlocks(0);
 568         die_nest_count--;
 569         if (die_nest_count)
 570                 /* We still own the lock */
 571                 local_irq_restore(flags);
 572         else
 573                 /* Nest count reaches zero, release the lock. */
 574                 spin_unlock_irqrestore(&die_lock, flags);
 575         if (panic_on_oops)
 576                 panic("Fatal exception");
 577         oops_exit();
 578 }
 579
 580 void __kprobes __die(const char * str, struct pt_regs * regs, long err)
 581 {
 582         static int die_counter;
 583         printk(KERN_EMERG "%s: %04lx [%u] ", str, err & 0xffff,++die_counter);
 584 #ifdef CONFIG_PREEMPT
 585         printk("PREEMPT ");
 586 #endif
 587 #ifdef CONFIG_SMP
 588         printk("SMP ");
 589 #endif
 590 #ifdef CONFIG_DEBUG_PAGEALLOC
 591         printk("DEBUG_PAGEALLOC");
 592 #endif
 593         printk("\n");
 594         notify_die(DIE_OOPS, str, regs, err, current->thread.trap_no, SIGSEGV);
 595         show_registers(regs);
 596         /* Executive summary in case the oops scrolled away */
 597         printk(KERN_ALERT "RIP ");
 598         printk_address(regs->rip);
 599         printk(" RSP <%016lx>\n", regs->rsp);
 600         if (kexec_should_crash(current))
 601                 crash_kexec(regs);
 602 }
 603
 604 void die(const char * str, struct pt_regs * regs, long err)
 605 {
 606         unsigned long flags = oops_begin();
 607
 608         handle_BUG(regs);
 609         __die(str, regs, err);
 610         oops_end(flags);
 611         do_exit(SIGSEGV);
 612 }
 613
 614 void __kprobes die_nmi(char *str, struct pt_regs *regs, int do_panic)
 615 {
 616         unsigned long flags = oops_begin();
 617
 618         /*
 619          * We are in trouble anyway, lets at least try
 620          * to get a message out.
 621          */
 622         printk(str, smp_processor_id());
 623         show_registers(regs);
 624         if (kexec_should_crash(current))
 625                 crash_kexec(regs);
 626         if (do_panic || panic_on_oops)
 627                 panic("Non maskable interrupt");
 628         oops_end(flags);
 629         nmi_exit();
 630         local_irq_enable();
 631         do_exit(SIGSEGV);
 632 }
 633
 634 static void __kprobes do_trap(int trapnr, int signr, char *str,
 635                               struct pt_regs * regs, long error_code,
 636                               siginfo_t *info)
 637 {
 638         struct task_struct *tsk = current;
 639
 640         tsk->thread.error_code = error_code;
 641         tsk->thread.trap_no = trapnr;
 642
 643         if (user_mode(regs)) {
 644                 if (exception_trace && unhandled_signal(tsk, signr))
 645                         printk(KERN_INFO
 646                                "%s[%d] trap %s rip:%lx rsp:%lx error:%lx\n",
 647                                tsk->comm, tsk->pid, str,
 648                                regs->rip, regs->rsp, error_code);
 649
 650                 if (info)
 651                         force_sig_info(signr, info, tsk);
 652                 else
 653                         force_sig(signr, tsk);
 654                 return;
 655         }
 656
 657
 658         /* kernel trap */
 659         {
 660                 const struct exception_table_entry *fixup;
 661                 fixup = search_exception_tables(regs->rip);
 662                 if (fixup)
 663                         regs->rip = fixup->fixup;
 664                 else
 665                         die(str, regs, error_code);
 666                 return;
 667         }
 668 }
 669
 670 #define DO_ERROR(trapnr, signr, str, name) \
 671 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
 672 { \
 673         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 674                                                         == NOTIFY_STOP) \
 675                 return; \
 676         conditional_sti(regs);                                          \
 677         do_trap(trapnr, signr, str, regs, error_code, NULL); \
 678 }
 679
 680 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
 681 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
 682 { \
 683         siginfo_t info; \
 684         info.si_signo = signr; \
 685         info.si_errno = 0; \
 686         info.si_code = sicode; \
 687         info.si_addr = (void __user *)siaddr; \
 688         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 689                                                         == NOTIFY_STOP) \
 690                 return; \
 691         conditional_sti(regs);                                          \
 692         do_trap(trapnr, signr, str, regs, error_code, &info); \
 693 }
 694
 695 DO_ERROR_INFO( 0, SIGFPE,  "divide error", divide_error, FPE_INTDIV, regs->rip)
 696 DO_ERROR( 4, SIGSEGV, "overflow", overflow)
 697 DO_ERROR( 5, SIGSEGV, "bounds", bounds)
 698 DO_ERROR_INFO( 6, SIGILL,  "invalid opcode", invalid_op, ILL_ILLOPN, regs->rip)
 699 DO_ERROR( 7, SIGSEGV, "device not available", device_not_available)
 700 DO_ERROR( 9, SIGFPE,  "coprocessor segment overrun", coprocessor_segment_overrun)
 701 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
 702 DO_ERROR(11, SIGBUS,  "segment not present", segment_not_present)
 703 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
 704 DO_ERROR(18, SIGSEGV, "reserved", reserved)
 705
 706 /* Runs on IST stack */
 707 asmlinkage void do_stack_segment(struct pt_regs *regs, long error_code)
 708 {
 709         if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
 710                         12, SIGBUS) == NOTIFY_STOP)
 711                 return;
 712         preempt_conditional_sti(regs);
 713         do_trap(12, SIGBUS, "stack segment", regs, error_code, NULL);
 714         preempt_conditional_cli(regs);
 715 }
 716
 717 asmlinkage void do_double_fault(struct pt_regs * regs, long error_code)
 718 {
 719         static const char str[] = "double fault";
 720         struct task_struct *tsk = current;
 721
 722         /* Return not checked because double check cannot be ignored */
 723         notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV);
 724
 725         tsk->thread.error_code = error_code;
 726         tsk->thread.trap_no = 8;
 727
 728         /* This is always a kernel trap and never fixable (and thus must
 729            never return). */
 730         for (;;)
 731                 die(str, regs, error_code);
 732 }
 733
 734 asmlinkage void __kprobes do_general_protection(struct pt_regs * regs,
 735                                                 long error_code)
 736 {
 737         struct task_struct *tsk = current;
 738
 739         conditional_sti(regs);
 740
 741         tsk->thread.error_code = error_code;
 742         tsk->thread.trap_no = 13;
 743
 744         if (user_mode(regs)) {
 745                 if (exception_trace && unhandled_signal(tsk, SIGSEGV))
 746                         printk(KERN_INFO
 747                        "%s[%d] general protection rip:%lx rsp:%lx error:%lx\n",
 748                                tsk->comm, tsk->pid,
 749                                regs->rip, regs->rsp, error_code);
 750
 751                 force_sig(SIGSEGV, tsk);
 752                 return;
 753         }
 754
 755         /* kernel gp */
 756         {
 757                 const struct exception_table_entry *fixup;
 758                 fixup = search_exception_tables(regs->rip);
 759                 if (fixup) {
 760                         regs->rip = fixup->fixup;
 761                         return;
 762                 }
 763                 if (notify_die(DIE_GPF, "general protection fault", regs,
 764                                         error_code, 13, SIGSEGV) == NOTIFY_STOP)
 765                         return;
 766                 die("general protection fault", regs, error_code);
 767         }
 768 }
 769
 770 static __kprobes void
 771 mem_parity_error(unsigned char reason, struct pt_regs * regs)
 772 {
 773         printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x.\n",
 774                 reason);
 775         printk(KERN_EMERG "You probably have a hardware problem with your "
 776                 "RAM chips\n");
 777
 778         if (panic_on_unrecovered_nmi)
 779                 panic("NMI: Not continuing");
 780
 781         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 782
 783         /* Clear and disable the memory parity error line. */
 784         reason = (reason & 0xf) | 4;
 785         outb(reason, 0x61);
 786 }
 787
 788 static __kprobes void
 789 io_check_error(unsigned char reason, struct pt_regs * regs)
 790 {
 791         printk("NMI: IOCK error (debug interrupt?)\n");
 792         show_registers(regs);
 793
 794         /* Re-enable the IOCK line, wait for a few seconds */
 795         reason = (reason & 0xf) | 8;
 796         outb(reason, 0x61);
 797         mdelay(2000);
 798         reason &= ~8;
 799         outb(reason, 0x61);
 800 }
 801
 802 static __kprobes void
 803 unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
 804 {
 805         printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x.\n",
 806                 reason);
 807         printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
 808
 809         if (panic_on_unrecovered_nmi)
 810                 panic("NMI: Not continuing");
 811
 812         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 813 }
 814
 815 /* Runs on IST stack. This code must keep interrupts off all the time.
 816    Nested NMIs are prevented by the CPU. */
 817 asmlinkage __kprobes void default_do_nmi(struct pt_regs *regs)
 818 {
 819         unsigned char reason = 0;
 820         int cpu;
 821
 822         cpu = smp_processor_id();
 823
 824         /* Only the BSP gets external NMIs from the system.  */
 825         if (!cpu)
 826                 reason = get_nmi_reason();
 827
 828         if (!(reason & 0xc0)) {
 829                 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
 830                                                                 == NOTIFY_STOP)
 831                         return;
 832                 /*
 833                  * Ok, so this is none of the documented NMI sources,
 834                  * so it must be the NMI watchdog.
 835                  */
 836                 if (nmi_watchdog_tick(regs,reason))
 837                         return;
 838                 if (!do_nmi_callback(regs,cpu))
 839                         unknown_nmi_error(reason, regs);
 840
 841                 return;
 842         }
 843         if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
 844                 return;
 845
 846         /* AK: following checks seem to be broken on modern chipsets. FIXME */
 847
 848         if (reason & 0x80)
 849                 mem_parity_error(reason, regs);
 850         if (reason & 0x40)
 851                 io_check_error(reason, regs);
 852 }
 853
 854 /* runs on IST stack. */
 855 asmlinkage void __kprobes do_int3(struct pt_regs * regs, long error_code)
 856 {
 857         if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP) == NOTIFY_STOP) {
 858                 return;
 859         }
 860         preempt_conditional_sti(regs);
 861         do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
 862         preempt_conditional_cli(regs);
 863 }
 864
 865 /* Help handler running on IST stack to switch back to user stack
 866    for scheduling or signal handling. The actual stack switch is done in
 867    entry.S */
 868 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
 869 {
 870         struct pt_regs *regs = eregs;
 871         /* Did already sync */
 872         if (eregs == (struct pt_regs *)eregs->rsp)
 873                 ;
 874         /* Exception from user space */
 875         else if (user_mode(eregs))
 876                 regs = task_pt_regs(current);
 877         /* Exception from kernel and interrupts are enabled. Move to
 878            kernel process stack. */
 879         else if (eregs->eflags & X86_EFLAGS_IF)
 880                 regs = (struct pt_regs *)(eregs->rsp -= sizeof(struct pt_regs));
 881         if (eregs != regs)
 882                 *regs = *eregs;
 883         return regs;
 884 }
 885
 886 /* runs on IST stack. */
 887 asmlinkage void __kprobes do_debug(struct pt_regs * regs,
 888                                    unsigned long error_code)
 889 {
 890         unsigned long condition;
 891         struct task_struct *tsk = current;
 892         siginfo_t info;
 893
 894         get_debugreg(condition, 6);
 895
 896         if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
 897                                                 SIGTRAP) == NOTIFY_STOP)
 898                 return;
 899
 900         preempt_conditional_sti(regs);
 901
 902         /* Mask out spurious debug traps due to lazy DR7 setting */
 903         if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
 904                 if (!tsk->thread.debugreg7) {
 905                         goto clear_dr7;
 906                 }
 907         }
 908
 909         tsk->thread.debugreg6 = condition;
 910
 911         /* Mask out spurious TF errors due to lazy TF clearing */
 912         if (condition & DR_STEP) {
 913                 /*
 914                  * The TF error should be masked out only if the current
 915                  * process is not traced and if the TRAP flag has been set
 916                  * previously by a tracing process (condition detected by
 917                  * the PT_DTRACE flag); remember that the i386 TRAP flag
 918                  * can be modified by the process itself in user mode,
 919                  * allowing programs to debug themselves without the ptrace()
 920                  * interface.
 921                  */
 922                 if (!user_mode(regs))
 923                        goto clear_TF_reenable;
 924                 /*
 925                  * Was the TF flag set by a debugger? If so, clear it now,
 926                  * so that register information is correct.
 927                  */
 928                 if (tsk->ptrace & PT_DTRACE) {
 929                         regs->eflags &= ~TF_MASK;
 930                         tsk->ptrace &= ~PT_DTRACE;
 931                 }
 932         }
 933
 934         /* Ok, finally something we can handle */
 935         tsk->thread.trap_no = 1;
 936         tsk->thread.error_code = error_code;
 937         info.si_signo = SIGTRAP;
 938         info.si_errno = 0;
 939         info.si_code = TRAP_BRKPT;
 940         info.si_addr = user_mode(regs) ? (void __user *)regs->rip : NULL;
 941         force_sig_info(SIGTRAP, &info, tsk);
 942
 943 clear_dr7:
 944         set_debugreg(0UL, 7);
 945         preempt_conditional_cli(regs);
 946         return;
 947
 948 clear_TF_reenable:
 949         set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
 950         regs->eflags &= ~TF_MASK;
 951         preempt_conditional_cli(regs);
 952 }
 953
 954 static int kernel_math_error(struct pt_regs *regs, const char *str, int trapnr)
 955 {
 956         const struct exception_table_entry *fixup;
 957         fixup = search_exception_tables(regs->rip);
 958         if (fixup) {
 959                 regs->rip = fixup->fixup;
 960                 return 1;
 961         }
 962         notify_die(DIE_GPF, str, regs, 0, trapnr, SIGFPE);
 963         /* Illegal floating point operation in the kernel */
 964         current->thread.trap_no = trapnr;
 965         die(str, regs, 0);
 966         return 0;
 967 }
 968
 969 /*
 970  * Note that we play around with the 'TS' bit in an attempt to get
 971  * the correct behaviour even in the presence of the asynchronous
 972  * IRQ13 behaviour
 973  */
 974 asmlinkage void do_coprocessor_error(struct pt_regs *regs)
 975 {
 976         void __user *rip = (void __user *)(regs->rip);
 977         struct task_struct * task;
 978         siginfo_t info;
 979         unsigned short cwd, swd;
 980
 981         conditional_sti(regs);
 982         if (!user_mode(regs) &&
 983             kernel_math_error(regs, "kernel x87 math error", 16))
 984                 return;
 985
 986         /*
 987          * Save the info for the exception handler and clear the error.
 988          */
 989         task = current;
 990         save_init_fpu(task);
 991         task->thread.trap_no = 16;
 992         task->thread.error_code = 0;
 993         info.si_signo = SIGFPE;
 994         info.si_errno = 0;
 995         info.si_code = __SI_FAULT;
 996         info.si_addr = rip;
 997         /*
 998          * (~cwd & swd) will mask out exceptions that are not set to unmasked
 999          * status.  0x3f is the exception bits in these regs, 0x200 is the
1000          * C1 reg you need in case of a stack fault, 0x040 is the stack
1001          * fault bit.  We should only be taking one exception at a time,
1002          * so if this combination doesn't produce any single exception,
1003          * then we have a bad program that isn't synchronizing its FPU usage
1004          * and it will suffer the consequences since we won't be able to
1005          * fully reproduce the context of the exception
1006          */
1007         cwd = get_fpu_cwd(task);
1008         swd = get_fpu_swd(task);
1009         switch (swd & ~cwd & 0x3f) {
1010                 case 0x000:
1011                 default:
1012                         break;
1013                 case 0x001: /* Invalid Op */
1014                         /*
1015                          * swd & 0x240 == 0x040: Stack Underflow
1016                          * swd & 0x240 == 0x240: Stack Overflow
1017                          * User must clear the SF bit (0x40) if set
1018                          */
1019                         info.si_code = FPE_FLTINV;
1020                         break;
1021                 case 0x002: /* Denormalize */
1022                 case 0x010: /* Underflow */
1023                         info.si_code = FPE_FLTUND;
1024                         break;
1025                 case 0x004: /* Zero Divide */
1026                         info.si_code = FPE_FLTDIV;
1027                         break;
1028                 case 0x008: /* Overflow */
1029                         info.si_code = FPE_FLTOVF;
1030                         break;
1031                 case 0x020: /* Precision */
1032                         info.si_code = FPE_FLTRES;
1033                         break;
1034         }
1035         force_sig_info(SIGFPE, &info, task);
1036 }
1037
1038 asmlinkage void bad_intr(void)
1039 {
1040         printk("bad interrupt");
1041 }
1042
1043 asmlinkage void do_simd_coprocessor_error(struct pt_regs *regs)
1044 {
1045         void __user *rip = (void __user *)(regs->rip);
1046         struct task_struct * task;
1047         siginfo_t info;
1048         unsigned short mxcsr;
1049
1050         conditional_sti(regs);
1051         if (!user_mode(regs) &&
1052                 kernel_math_error(regs, "kernel simd math error", 19))
1053                 return;
1054
1055         /*
1056          * Save the info for the exception handler and clear the error.
1057          */
1058         task = current;
1059         save_init_fpu(task);
1060         task->thread.trap_no = 19;
1061         task->thread.error_code = 0;
1062         info.si_signo = SIGFPE;
1063         info.si_errno = 0;
1064         info.si_code = __SI_FAULT;
1065         info.si_addr = rip;
1066         /*
1067          * The SIMD FPU exceptions are handled a little differently, as there
1068          * is only a single status/control register.  Thus, to determine which
1069          * unmasked exception was caught we must mask the exception mask bits
1070          * at 0x1f80, and then use these to mask the exception bits at 0x3f.
1071          */
1072         mxcsr = get_fpu_mxcsr(task);
1073         switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
1074                 case 0x000:
1075                 default:
1076                         break;
1077                 case 0x001: /* Invalid Op */
1078                         info.si_code = FPE_FLTINV;
1079                         break;
1080                 case 0x002: /* Denormalize */
1081                 case 0x010: /* Underflow */
1082                         info.si_code = FPE_FLTUND;
1083                         break;
1084                 case 0x004: /* Zero Divide */
1085                         info.si_code = FPE_FLTDIV;
1086                         break;
1087                 case 0x008: /* Overflow */
1088                         info.si_code = FPE_FLTOVF;
1089                         break;
1090                 case 0x020: /* Precision */
1091                         info.si_code = FPE_FLTRES;
1092                         break;
1093         }
1094         force_sig_info(SIGFPE, &info, task);
1095 }
1096
1097 asmlinkage void do_spurious_interrupt_bug(struct pt_regs * regs)
1098 {
1099 }
1100
1101 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
1102 {
1103 }
1104
1105 asmlinkage void __attribute__((weak)) mce_threshold_interrupt(void)
1106 {
1107 }
1108
1109 /*
1110  *  'math_state_restore()' saves the current math information in the
1111  * old math state array, and gets the new ones from the current task
1112  *
1113  * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1114  * Don't touch unless you *really* know how it works.
1115  */
1116 asmlinkage void math_state_restore(void)
1117 {
1118         struct task_struct *me = current;
1119         clts();                 /* Allow maths ops (or we recurse) */
1120
1121         if (!used_math())
1122                 init_fpu(me);
1123         restore_fpu_checking(&me->thread.i387.fxsave);
1124         task_thread_info(me)->status |= TS_USEDFPU;
1125         me->fpu_counter++;
1126 }
1127
1128 void __init trap_init(void)
1129 {
1130         set_intr_gate(0,&divide_error);
1131         set_intr_gate_ist(1,&debug,DEBUG_STACK);
1132         set_intr_gate_ist(2,&nmi,NMI_STACK);
1133         set_system_gate_ist(3,&int3,DEBUG_STACK); /* int3 can be called from all */
1134         set_system_gate(4,&overflow);   /* int4 can be called from all */
1135         set_intr_gate(5,&bounds);
1136         set_intr_gate(6,&invalid_op);
1137         set_intr_gate(7,&device_not_available);
1138         set_intr_gate_ist(8,&double_fault, DOUBLEFAULT_STACK);
1139         set_intr_gate(9,&coprocessor_segment_overrun);
1140         set_intr_gate(10,&invalid_TSS);
1141         set_intr_gate(11,&segment_not_present);
1142         set_intr_gate_ist(12,&stack_segment,STACKFAULT_STACK);
1143         set_intr_gate(13,&general_protection);
1144         set_intr_gate(14,&page_fault);
1145         set_intr_gate(15,&spurious_interrupt_bug);
1146         set_intr_gate(16,&coprocessor_error);
1147         set_intr_gate(17,&alignment_check);
1148 #ifdef CONFIG_X86_MCE
1149         set_intr_gate_ist(18,&machine_check, MCE_STACK);
1150 #endif
1151         set_intr_gate(19,&simd_coprocessor_error);
1152
1153 #ifdef CONFIG_IA32_EMULATION
1154         set_system_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
1155 #endif
1156
1157         /*
1158          * Should be a barrier for any external CPU state.
1159          */
1160         cpu_init();
1161 }
1162
1163
1164 static int __init oops_setup(char *s)
1165 {
1166         if (!s)
1167                 return -EINVAL;
1168         if (!strcmp(s, "panic"))
1169                 panic_on_oops = 1;
1170         return 0;
1171 }
1172 early_param("oops", oops_setup);
1173
1174 static int __init kstack_setup(char *s)
1175 {
1176         if (!s)
1177                 return -EINVAL;
1178         kstack_depth_to_print = simple_strtoul(s,NULL,0);
1179         return 0;
1180 }
1181 early_param("kstack", kstack_setup);
1182
1183 #ifdef CONFIG_STACK_UNWIND
1184 static int __init call_trace_setup(char *s)
1185 {
1186         if (!s)
1187                 return -EINVAL;
1188         if (strcmp(s, "old") == 0)
1189                 call_trace = -1;
1190         else if (strcmp(s, "both") == 0)
1191                 call_trace = 0;
1192         else if (strcmp(s, "newfallback") == 0)
1193                 call_trace = 1;
1194         else if (strcmp(s, "new") == 0)
1195                 call_trace = 2;
1196         return 0;
1197 }
1198 early_param("call_trace", call_trace_setup);
1199 #endif