arch/x86/kernel/traps_64.c

   1 /*
   2  *  Copyright (C) 1991, 1992  Linus Torvalds
   3  *  Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs
   4  *
   5  *  Pentium III FXSR, SSE support
   6  *      Gareth Hughes <gareth@valinux.com>, May 2000
   7  */
   8
   9 /*
  10  * 'Traps.c' handles hardware traps and faults after we have saved some
  11  * state in 'entry.S'.
  12  */
  13 #include <linux/sched.h>
  14 #include <linux/kernel.h>
  15 #include <linux/string.h>
  16 #include <linux/errno.h>
  17 #include <linux/ptrace.h>
  18 #include <linux/timer.h>
  19 #include <linux/mm.h>
  20 #include <linux/init.h>
  21 #include <linux/delay.h>
  22 #include <linux/spinlock.h>
  23 #include <linux/interrupt.h>
  24 #include <linux/kallsyms.h>
  25 #include <linux/module.h>
  26 #include <linux/moduleparam.h>
  27 #include <linux/nmi.h>
  28 #include <linux/kprobes.h>
  29 #include <linux/kexec.h>
  30 #include <linux/unwind.h>
  31 #include <linux/uaccess.h>
  32 #include <linux/bug.h>
  33 #include <linux/kdebug.h>
  34 #include <linux/utsname.h>
  35
  36 #if defined(CONFIG_EDAC)
  37 #include <linux/edac.h>
  38 #endif
  39
  40 #include <asm/system.h>
  41 #include <asm/io.h>
  42 #include <asm/atomic.h>
  43 #include <asm/debugreg.h>
  44 #include <asm/desc.h>
  45 #include <asm/i387.h>
  46 #include <asm/processor.h>
  47 #include <asm/unwind.h>
  48 #include <asm/smp.h>
  49 #include <asm/pgalloc.h>
  50 #include <asm/pda.h>
  51 #include <asm/proto.h>
  52 #include <asm/nmi.h>
  53 #include <asm/stacktrace.h>
  54
  55 asmlinkage void divide_error(void);
  56 asmlinkage void debug(void);
  57 asmlinkage void nmi(void);
  58 asmlinkage void int3(void);
  59 asmlinkage void overflow(void);
  60 asmlinkage void bounds(void);
  61 asmlinkage void invalid_op(void);
  62 asmlinkage void device_not_available(void);
  63 asmlinkage void double_fault(void);
  64 asmlinkage void coprocessor_segment_overrun(void);
  65 asmlinkage void invalid_TSS(void);
  66 asmlinkage void segment_not_present(void);
  67 asmlinkage void stack_segment(void);
  68 asmlinkage void general_protection(void);
  69 asmlinkage void page_fault(void);
  70 asmlinkage void coprocessor_error(void);
  71 asmlinkage void simd_coprocessor_error(void);
  72 asmlinkage void reserved(void);
  73 asmlinkage void alignment_check(void);
  74 asmlinkage void machine_check(void);
  75 asmlinkage void spurious_interrupt_bug(void);
  76
  77 static inline void conditional_sti(struct pt_regs *regs)
  78 {
  79         if (regs->eflags & X86_EFLAGS_IF)
  80                 local_irq_enable();
  81 }
  82
  83 static inline void preempt_conditional_sti(struct pt_regs *regs)
  84 {
  85         preempt_disable();
  86         if (regs->eflags & X86_EFLAGS_IF)
  87                 local_irq_enable();
  88 }
  89
  90 static inline void preempt_conditional_cli(struct pt_regs *regs)
  91 {
  92         if (regs->eflags & X86_EFLAGS_IF)
  93                 local_irq_disable();
  94         /* Make sure to not schedule here because we could be running
  95            on an exception stack. */
  96         preempt_enable_no_resched();
  97 }
  98
  99 int kstack_depth_to_print = 12;
 100
 101 #ifdef CONFIG_KALLSYMS
 102 void printk_address(unsigned long address)
 103 {
 104         unsigned long offset = 0, symsize;
 105         const char *symname;
 106         char *modname;
 107         char *delim = ":";
 108         char namebuf[128];
 109
 110         symname = kallsyms_lookup(address, &symsize, &offset,
 111                                         &modname, namebuf);
 112         if (!symname) {
 113                 printk(" [<%016lx>]\n", address);
 114                 return;
 115         }
 116         if (!modname)
 117                 modname = delim = "";
 118         printk(" [<%016lx>] %s%s%s%s+0x%lx/0x%lx\n",
 119                 address, delim, modname, delim, symname, offset, symsize);
 120 }
 121 #else
 122 void printk_address(unsigned long address)
 123 {
 124         printk(" [<%016lx>]\n", address);
 125 }
 126 #endif
 127
 128 static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
 129                                         unsigned *usedp, char **idp)
 130 {
 131         static char ids[][8] = {
 132                 [DEBUG_STACK - 1] = "#DB",
 133                 [NMI_STACK - 1] = "NMI",
 134                 [DOUBLEFAULT_STACK - 1] = "#DF",
 135                 [STACKFAULT_STACK - 1] = "#SS",
 136                 [MCE_STACK - 1] = "#MC",
 137 #if DEBUG_STKSZ > EXCEPTION_STKSZ
 138                 [N_EXCEPTION_STACKS ... N_EXCEPTION_STACKS + DEBUG_STKSZ / EXCEPTION_STKSZ - 2] = "#DB[?]"
 139 #endif
 140         };
 141         unsigned k;
 142
 143         /*
 144          * Iterate over all exception stacks, and figure out whether
 145          * 'stack' is in one of them:
 146          */
 147         for (k = 0; k < N_EXCEPTION_STACKS; k++) {
 148                 unsigned long end = per_cpu(orig_ist, cpu).ist[k];
 149                 /*
 150                  * Is 'stack' above this exception frame's end?
 151                  * If yes then skip to the next frame.
 152                  */
 153                 if (stack >= end)
 154                         continue;
 155                 /*
 156                  * Is 'stack' above this exception frame's start address?
 157                  * If yes then we found the right frame.
 158                  */
 159                 if (stack >= end - EXCEPTION_STKSZ) {
 160                         /*
 161                          * Make sure we only iterate through an exception
 162                          * stack once. If it comes up for the second time
 163                          * then there's something wrong going on - just
 164                          * break out and return NULL:
 165                          */
 166                         if (*usedp & (1U << k))
 167                                 break;
 168                         *usedp |= 1U << k;
 169                         *idp = ids[k];
 170                         return (unsigned long *)end;
 171                 }
 172                 /*
 173                  * If this is a debug stack, and if it has a larger size than
 174                  * the usual exception stacks, then 'stack' might still
 175                  * be within the lower portion of the debug stack:
 176                  */
 177 #if DEBUG_STKSZ > EXCEPTION_STKSZ
 178                 if (k == DEBUG_STACK - 1 && stack >= end - DEBUG_STKSZ) {
 179                         unsigned j = N_EXCEPTION_STACKS - 1;
 180
 181                         /*
 182                          * Black magic. A large debug stack is composed of
 183                          * multiple exception stack entries, which we
 184                          * iterate through now. Dont look:
 185                          */
 186                         do {
 187                                 ++j;
 188                                 end -= EXCEPTION_STKSZ;
 189                                 ids[j][4] = '1' + (j - N_EXCEPTION_STACKS);
 190                         } while (stack < end - EXCEPTION_STKSZ);
 191                         if (*usedp & (1U << j))
 192                                 break;
 193                         *usedp |= 1U << j;
 194                         *idp = ids[j];
 195                         return (unsigned long *)end;
 196                 }
 197 #endif
 198         }
 199         return NULL;
 200 }
 201
 202 #define MSG(txt) ops->warning(data, txt)
 203
 204 /*
 205  * x86-64 can have up to three kernel stacks:
 206  * process stack
 207  * interrupt stack
 208  * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
 209  */
 210
 211 static inline int valid_stack_ptr(struct thread_info *tinfo, void *p)
 212 {
 213         void *t = (void *)tinfo;
 214         return p > t && p < t + THREAD_SIZE - 3;
 215 }
 216
 217 void dump_trace(struct task_struct *tsk, struct pt_regs *regs,
 218                 unsigned long *stack,
 219                 const struct stacktrace_ops *ops, void *data)
 220 {
 221         const unsigned cpu = get_cpu();
 222         unsigned long *irqstack_end = (unsigned long*)cpu_pda(cpu)->irqstackptr;
 223         unsigned used = 0;
 224         struct thread_info *tinfo;
 225
 226         if (!tsk)
 227                 tsk = current;
 228
 229         if (!stack) {
 230                 unsigned long dummy;
 231                 stack = &dummy;
 232                 if (tsk && tsk != current)
 233                         stack = (unsigned long *)tsk->thread.rsp;
 234         }
 235
 236         /*
 237          * Print function call entries within a stack. 'cond' is the
 238          * "end of stackframe" condition, that the 'stack++'
 239          * iteration will eventually trigger.
 240          */
 241 #define HANDLE_STACK(cond) \
 242         do while (cond) { \
 243                 unsigned long addr = *stack++; \
 244                 /* Use unlocked access here because except for NMIs     \
 245                    we should be already protected against module unloads */ \
 246                 if (__kernel_text_address(addr)) { \
 247                         /* \
 248                          * If the address is either in the text segment of the \
 249                          * kernel, or in the region which contains vmalloc'ed \
 250                          * memory, it *may* be the address of a calling \
 251                          * routine; if so, print it so that someone tracing \
 252                          * down the cause of the crash will be able to figure \
 253                          * out the call path that was taken. \
 254                          */ \
 255                         ops->address(data, addr);   \
 256                 } \
 257         } while (0)
 258
 259         /*
 260          * Print function call entries in all stacks, starting at the
 261          * current stack address. If the stacks consist of nested
 262          * exceptions
 263          */
 264         for (;;) {
 265                 char *id;
 266                 unsigned long *estack_end;
 267                 estack_end = in_exception_stack(cpu, (unsigned long)stack,
 268                                                 &used, &id);
 269
 270                 if (estack_end) {
 271                         if (ops->stack(data, id) < 0)
 272                                 break;
 273                         HANDLE_STACK (stack < estack_end);
 274                         ops->stack(data, "<EOE>");
 275                         /*
 276                          * We link to the next stack via the
 277                          * second-to-last pointer (index -2 to end) in the
 278                          * exception stack:
 279                          */
 280                         stack = (unsigned long *) estack_end[-2];
 281                         continue;
 282                 }
 283                 if (irqstack_end) {
 284                         unsigned long *irqstack;
 285                         irqstack = irqstack_end -
 286                                 (IRQSTACKSIZE - 64) / sizeof(*irqstack);
 287
 288                         if (stack >= irqstack && stack < irqstack_end) {
 289                                 if (ops->stack(data, "IRQ") < 0)
 290                                         break;
 291                                 HANDLE_STACK (stack < irqstack_end);
 292                                 /*
 293                                  * We link to the next stack (which would be
 294                                  * the process stack normally) the last
 295                                  * pointer (index -1 to end) in the IRQ stack:
 296                                  */
 297                                 stack = (unsigned long *) (irqstack_end[-1]);
 298                                 irqstack_end = NULL;
 299                                 ops->stack(data, "EOI");
 300                                 continue;
 301                         }
 302                 }
 303                 break;
 304         }
 305
 306         /*
 307          * This handles the process stack:
 308          */
 309         tinfo = task_thread_info(tsk);
 310         HANDLE_STACK (valid_stack_ptr(tinfo, stack));
 311 #undef HANDLE_STACK
 312         put_cpu();
 313 }
 314 EXPORT_SYMBOL(dump_trace);
 315
 316 static void
 317 print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
 318 {
 319         print_symbol(msg, symbol);
 320         printk("\n");
 321 }
 322
 323 static void print_trace_warning(void *data, char *msg)
 324 {
 325         printk("%s\n", msg);
 326 }
 327
 328 static int print_trace_stack(void *data, char *name)
 329 {
 330         printk(" <%s> ", name);
 331         return 0;
 332 }
 333
 334 static void print_trace_address(void *data, unsigned long addr)
 335 {
 336         touch_nmi_watchdog();
 337         printk_address(addr);
 338 }
 339
 340 static const struct stacktrace_ops print_trace_ops = {
 341         .warning = print_trace_warning,
 342         .warning_symbol = print_trace_warning_symbol,
 343         .stack = print_trace_stack,
 344         .address = print_trace_address,
 345 };
 346
 347 void
 348 show_trace(struct task_struct *tsk, struct pt_regs *regs, unsigned long *stack)
 349 {
 350         printk("\nCall Trace:\n");
 351         dump_trace(tsk, regs, stack, &print_trace_ops, NULL);
 352         printk("\n");
 353 }
 354
 355 static void
 356 _show_stack(struct task_struct *tsk, struct pt_regs *regs, unsigned long *rsp)
 357 {
 358         unsigned long *stack;
 359         int i;
 360         const int cpu = smp_processor_id();
 361         unsigned long *irqstack_end = (unsigned long *) (cpu_pda(cpu)->irqstackptr);
 362         unsigned long *irqstack = (unsigned long *) (cpu_pda(cpu)->irqstackptr - IRQSTACKSIZE);
 363
 364         // debugging aid: "show_stack(NULL, NULL);" prints the
 365         // back trace for this cpu.
 366
 367         if (rsp == NULL) {
 368                 if (tsk)
 369                         rsp = (unsigned long *)tsk->thread.rsp;
 370                 else
 371                         rsp = (unsigned long *)&rsp;
 372         }
 373
 374         stack = rsp;
 375         for(i=0; i < kstack_depth_to_print; i++) {
 376                 if (stack >= irqstack && stack <= irqstack_end) {
 377                         if (stack == irqstack_end) {
 378                                 stack = (unsigned long *) (irqstack_end[-1]);
 379                                 printk(" <EOI> ");
 380                         }
 381                 } else {
 382                 if (((long) stack & (THREAD_SIZE-1)) == 0)
 383                         break;
 384                 }
 385                 if (i && ((i % 4) == 0))
 386                         printk("\n");
 387                 printk(" %016lx", *stack++);
 388                 touch_nmi_watchdog();
 389         }
 390         show_trace(tsk, regs, rsp);
 391 }
 392
 393 void show_stack(struct task_struct *tsk, unsigned long * rsp)
 394 {
 395         _show_stack(tsk, NULL, rsp);
 396 }
 397
 398 /*
 399  * The architecture-independent dump_stack generator
 400  */
 401 void dump_stack(void)
 402 {
 403         unsigned long dummy;
 404
 405         printk("Pid: %d, comm: %.20s %s %s %.*s\n",
 406                 current->pid, current->comm, print_tainted(),
 407                 init_utsname()->release,
 408                 (int)strcspn(init_utsname()->version, " "),
 409                 init_utsname()->version);
 410         show_trace(NULL, NULL, &dummy);
 411 }
 412
 413 EXPORT_SYMBOL(dump_stack);
 414
 415 void show_registers(struct pt_regs *regs)
 416 {
 417         int i;
 418         int in_kernel = !user_mode(regs);
 419         unsigned long rsp;
 420         const int cpu = smp_processor_id();
 421         struct task_struct *cur = cpu_pda(cpu)->pcurrent;
 422
 423         rsp = regs->rsp;
 424         printk("CPU %d ", cpu);
 425         __show_regs(regs);
 426         printk("Process %s (pid: %d, threadinfo %p, task %p)\n",
 427                 cur->comm, cur->pid, task_thread_info(cur), cur);
 428
 429         /*
 430          * When in-kernel, we also print out the stack and code at the
 431          * time of the fault..
 432          */
 433         if (in_kernel) {
 434                 printk("Stack: ");
 435                 _show_stack(NULL, regs, (unsigned long*)rsp);
 436
 437                 printk("\nCode: ");
 438                 if (regs->rip < PAGE_OFFSET)
 439                         goto bad;
 440
 441                 for (i=0; i<20; i++) {
 442                         unsigned char c;
 443                         if (__get_user(c, &((unsigned char*)regs->rip)[i])) {
 444 bad:
 445                                 printk(" Bad RIP value.");
 446                                 break;
 447                         }
 448                         printk("%02x ", c);
 449                 }
 450         }
 451         printk("\n");
 452 }
 453
 454 int is_valid_bugaddr(unsigned long rip)
 455 {
 456         unsigned short ud2;
 457
 458         if (__copy_from_user(&ud2, (const void __user *) rip, sizeof(ud2)))
 459                 return 0;
 460
 461         return ud2 == 0x0b0f;
 462 }
 463
 464 static raw_spinlock_t die_lock = __RAW_SPIN_LOCK_UNLOCKED;
 465 static int die_owner = -1;
 466 static unsigned int die_nest_count;
 467
 468 unsigned __kprobes long oops_begin(void)
 469 {
 470         int cpu;
 471         unsigned long flags;
 472
 473         oops_enter();
 474
 475         /* racy, but better than risking deadlock. */
 476         raw_local_irq_save(flags);
 477         cpu = smp_processor_id();
 478         if (!__raw_spin_trylock(&die_lock)) {
 479                 if (cpu == die_owner)
 480                         /* nested oops. should stop eventually */;
 481                 else
 482                         __raw_spin_lock(&die_lock);
 483         }
 484         die_nest_count++;
 485         die_owner = cpu;
 486         console_verbose();
 487         bust_spinlocks(1);
 488         return flags;
 489 }
 490
 491 void __kprobes oops_end(unsigned long flags)
 492 {
 493         die_owner = -1;
 494         bust_spinlocks(0);
 495         die_nest_count--;
 496         if (!die_nest_count)
 497                 /* Nest count reaches zero, release the lock. */
 498                 __raw_spin_unlock(&die_lock);
 499         raw_local_irq_restore(flags);
 500         if (panic_on_oops)
 501                 panic("Fatal exception");
 502         oops_exit();
 503 }
 504
 505 void __kprobes __die(const char * str, struct pt_regs * regs, long err)
 506 {
 507         static int die_counter;
 508         printk(KERN_EMERG "%s: %04lx [%u] ", str, err & 0xffff,++die_counter);
 509 #ifdef CONFIG_PREEMPT
 510         printk("PREEMPT ");
 511 #endif
 512 #ifdef CONFIG_SMP
 513         printk("SMP ");
 514 #endif
 515 #ifdef CONFIG_DEBUG_PAGEALLOC
 516         printk("DEBUG_PAGEALLOC");
 517 #endif
 518         printk("\n");
 519         notify_die(DIE_OOPS, str, regs, err, current->thread.trap_no, SIGSEGV);
 520         show_registers(regs);
 521         add_taint(TAINT_DIE);
 522         /* Executive summary in case the oops scrolled away */
 523         printk(KERN_ALERT "RIP ");
 524         printk_address(regs->rip);
 525         printk(" RSP <%016lx>\n", regs->rsp);
 526         if (kexec_should_crash(current))
 527                 crash_kexec(regs);
 528 }
 529
 530 void die(const char * str, struct pt_regs * regs, long err)
 531 {
 532         unsigned long flags = oops_begin();
 533
 534         if (!user_mode(regs))
 535                 report_bug(regs->rip, regs);
 536
 537         __die(str, regs, err);
 538         oops_end(flags);
 539         do_exit(SIGSEGV);
 540 }
 541
 542 void __kprobes die_nmi(char *str, struct pt_regs *regs, int do_panic)
 543 {
 544         unsigned long flags = oops_begin();
 545
 546         /*
 547          * We are in trouble anyway, lets at least try
 548          * to get a message out.
 549          */
 550         printk(str, smp_processor_id());
 551         show_registers(regs);
 552         if (kexec_should_crash(current))
 553                 crash_kexec(regs);
 554         if (do_panic || panic_on_oops)
 555                 panic("Non maskable interrupt");
 556         oops_end(flags);
 557         nmi_exit();
 558         local_irq_enable();
 559         do_exit(SIGSEGV);
 560 }
 561
 562 static void __kprobes do_trap(int trapnr, int signr, char *str,
 563                               struct pt_regs * regs, long error_code,
 564                               siginfo_t *info)
 565 {
 566         struct task_struct *tsk = current;
 567
 568         if (user_mode(regs)) {
 569                 /*
 570                  * We want error_code and trap_no set for userspace
 571                  * faults and kernelspace faults which result in
 572                  * die(), but not kernelspace faults which are fixed
 573                  * up.  die() gives the process no chance to handle
 574                  * the signal and notice the kernel fault information,
 575                  * so that won't result in polluting the information
 576                  * about previously queued, but not yet delivered,
 577                  * faults.  See also do_general_protection below.
 578                  */
 579                 tsk->thread.error_code = error_code;
 580                 tsk->thread.trap_no = trapnr;
 581
 582                 if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
 583                     printk_ratelimit())
 584                         printk(KERN_INFO
 585                                "%s[%d] trap %s rip:%lx rsp:%lx error:%lx\n",
 586                                tsk->comm, tsk->pid, str,
 587                                regs->rip, regs->rsp, error_code);
 588
 589                 if (info)
 590                         force_sig_info(signr, info, tsk);
 591                 else
 592                         force_sig(signr, tsk);
 593                 return;
 594         }
 595
 596
 597         /* kernel trap */
 598         {
 599                 const struct exception_table_entry *fixup;
 600                 fixup = search_exception_tables(regs->rip);
 601                 if (fixup)
 602                         regs->rip = fixup->fixup;
 603                 else {
 604                         tsk->thread.error_code = error_code;
 605                         tsk->thread.trap_no = trapnr;
 606                         die(str, regs, error_code);
 607                 }
 608                 return;
 609         }
 610 }
 611
 612 #define DO_ERROR(trapnr, signr, str, name) \
 613 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
 614 { \
 615         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 616                                                         == NOTIFY_STOP) \
 617                 return; \
 618         conditional_sti(regs);                                          \
 619         do_trap(trapnr, signr, str, regs, error_code, NULL); \
 620 }
 621
 622 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
 623 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
 624 { \
 625         siginfo_t info; \
 626         info.si_signo = signr; \
 627         info.si_errno = 0; \
 628         info.si_code = sicode; \
 629         info.si_addr = (void __user *)siaddr; \
 630         trace_hardirqs_fixup(); \
 631         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 632                                                         == NOTIFY_STOP) \
 633                 return; \
 634         conditional_sti(regs);                                          \
 635         do_trap(trapnr, signr, str, regs, error_code, &info); \
 636 }
 637
 638 DO_ERROR_INFO( 0, SIGFPE,  "divide error", divide_error, FPE_INTDIV, regs->rip)
 639 DO_ERROR( 4, SIGSEGV, "overflow", overflow)
 640 DO_ERROR( 5, SIGSEGV, "bounds", bounds)
 641 DO_ERROR_INFO( 6, SIGILL,  "invalid opcode", invalid_op, ILL_ILLOPN, regs->rip)
 642 DO_ERROR( 7, SIGSEGV, "device not available", device_not_available)
 643 DO_ERROR( 9, SIGFPE,  "coprocessor segment overrun", coprocessor_segment_overrun)
 644 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
 645 DO_ERROR(11, SIGBUS,  "segment not present", segment_not_present)
 646 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
 647 DO_ERROR(18, SIGSEGV, "reserved", reserved)
 648
 649 /* Runs on IST stack */
 650 asmlinkage void do_stack_segment(struct pt_regs *regs, long error_code)
 651 {
 652         if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
 653                         12, SIGBUS) == NOTIFY_STOP)
 654                 return;
 655         preempt_conditional_sti(regs);
 656         do_trap(12, SIGBUS, "stack segment", regs, error_code, NULL);
 657         preempt_conditional_cli(regs);
 658 }
 659
 660 asmlinkage void do_double_fault(struct pt_regs * regs, long error_code)
 661 {
 662         static const char str[] = "double fault";
 663         struct task_struct *tsk = current;
 664
 665         /* Return not checked because double check cannot be ignored */
 666         notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV);
 667
 668         tsk->thread.error_code = error_code;
 669         tsk->thread.trap_no = 8;
 670
 671         /* This is always a kernel trap and never fixable (and thus must
 672            never return). */
 673         for (;;)
 674                 die(str, regs, error_code);
 675 }
 676
 677 asmlinkage void __kprobes do_general_protection(struct pt_regs * regs,
 678                                                 long error_code)
 679 {
 680         struct task_struct *tsk = current;
 681
 682         conditional_sti(regs);
 683
 684         if (user_mode(regs)) {
 685                 tsk->thread.error_code = error_code;
 686                 tsk->thread.trap_no = 13;
 687
 688                 if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
 689                     printk_ratelimit())
 690                         printk(KERN_INFO
 691                        "%s[%d] general protection rip:%lx rsp:%lx error:%lx\n",
 692                                tsk->comm, tsk->pid,
 693                                regs->rip, regs->rsp, error_code);
 694
 695                 force_sig(SIGSEGV, tsk);
 696                 return;
 697         }
 698
 699         /* kernel gp */
 700         {
 701                 const struct exception_table_entry *fixup;
 702                 fixup = search_exception_tables(regs->rip);
 703                 if (fixup) {
 704                         regs->rip = fixup->fixup;
 705                         return;
 706                 }
 707
 708                 tsk->thread.error_code = error_code;
 709                 tsk->thread.trap_no = 13;
 710                 if (notify_die(DIE_GPF, "general protection fault", regs,
 711                                         error_code, 13, SIGSEGV) == NOTIFY_STOP)
 712                         return;
 713                 die("general protection fault", regs, error_code);
 714         }
 715 }
 716
 717 static __kprobes void
 718 mem_parity_error(unsigned char reason, struct pt_regs * regs)
 719 {
 720         printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x.\n",
 721                 reason);
 722         printk(KERN_EMERG "You have some hardware problem, likely on the PCI bus.\n");
 723
 724 #if defined(CONFIG_EDAC)
 725         if(edac_handler_set()) {
 726                 edac_atomic_assert_error();
 727                 return;
 728         }
 729 #endif
 730
 731         if (panic_on_unrecovered_nmi)
 732                 panic("NMI: Not continuing");
 733
 734         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 735
 736         /* Clear and disable the memory parity error line. */
 737         reason = (reason & 0xf) | 4;
 738         outb(reason, 0x61);
 739 }
 740
 741 static __kprobes void
 742 io_check_error(unsigned char reason, struct pt_regs * regs)
 743 {
 744         printk("NMI: IOCK error (debug interrupt?)\n");
 745         show_registers(regs);
 746
 747         /* Re-enable the IOCK line, wait for a few seconds */
 748         reason = (reason & 0xf) | 8;
 749         outb(reason, 0x61);
 750         mdelay(2000);
 751         reason &= ~8;
 752         outb(reason, 0x61);
 753 }
 754
 755 static __kprobes void
 756 unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
 757 {
 758         printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x.\n",
 759                 reason);
 760         printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
 761
 762         if (panic_on_unrecovered_nmi)
 763                 panic("NMI: Not continuing");
 764
 765         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 766 }
 767
 768 /* Runs on IST stack. This code must keep interrupts off all the time.
 769    Nested NMIs are prevented by the CPU. */
 770 asmlinkage __kprobes void default_do_nmi(struct pt_regs *regs)
 771 {
 772         unsigned char reason = 0;
 773         int cpu;
 774
 775         cpu = smp_processor_id();
 776
 777         /* Only the BSP gets external NMIs from the system.  */
 778         if (!cpu)
 779                 reason = get_nmi_reason();
 780
 781         if (!(reason & 0xc0)) {
 782                 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
 783                                                                 == NOTIFY_STOP)
 784                         return;
 785                 /*
 786                  * Ok, so this is none of the documented NMI sources,
 787                  * so it must be the NMI watchdog.
 788                  */
 789                 if (nmi_watchdog_tick(regs,reason))
 790                         return;
 791                 if (!do_nmi_callback(regs,cpu))
 792                         unknown_nmi_error(reason, regs);
 793
 794                 return;
 795         }
 796         if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
 797                 return;
 798
 799         /* AK: following checks seem to be broken on modern chipsets. FIXME */
 800
 801         if (reason & 0x80)
 802                 mem_parity_error(reason, regs);
 803         if (reason & 0x40)
 804                 io_check_error(reason, regs);
 805 }
 806
 807 /* runs on IST stack. */
 808 asmlinkage void __kprobes do_int3(struct pt_regs * regs, long error_code)
 809 {
 810         trace_hardirqs_fixup();
 811
 812         if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP) == NOTIFY_STOP) {
 813                 return;
 814         }
 815         preempt_conditional_sti(regs);
 816         do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
 817         preempt_conditional_cli(regs);
 818 }
 819
 820 /* Help handler running on IST stack to switch back to user stack
 821    for scheduling or signal handling. The actual stack switch is done in
 822    entry.S */
 823 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
 824 {
 825         struct pt_regs *regs = eregs;
 826         /* Did already sync */
 827         if (eregs == (struct pt_regs *)eregs->rsp)
 828                 ;
 829         /* Exception from user space */
 830         else if (user_mode(eregs))
 831                 regs = task_pt_regs(current);
 832         /* Exception from kernel and interrupts are enabled. Move to
 833            kernel process stack. */
 834         else if (eregs->eflags & X86_EFLAGS_IF)
 835                 regs = (struct pt_regs *)(eregs->rsp -= sizeof(struct pt_regs));
 836         if (eregs != regs)
 837                 *regs = *eregs;
 838         return regs;
 839 }
 840
 841 /* runs on IST stack. */
 842 asmlinkage void __kprobes do_debug(struct pt_regs * regs,
 843                                    unsigned long error_code)
 844 {
 845         unsigned long condition;
 846         struct task_struct *tsk = current;
 847         siginfo_t info;
 848
 849         trace_hardirqs_fixup();
 850
 851         get_debugreg(condition, 6);
 852
 853         /*
 854          * The processor cleared BTF, so don't mark that we need it set.
 855          */
 856         clear_tsk_thread_flag(tsk, TIF_DEBUGCTLMSR);
 857         tsk->thread.debugctlmsr = 0;
 858
 859         if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
 860                                                 SIGTRAP) == NOTIFY_STOP)
 861                 return;
 862
 863         preempt_conditional_sti(regs);
 864
 865         /* Mask out spurious debug traps due to lazy DR7 setting */
 866         if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
 867                 if (!tsk->thread.debugreg7) {
 868                         goto clear_dr7;
 869                 }
 870         }
 871
 872         tsk->thread.debugreg6 = condition;
 873
 874
 875         /*
 876          * Single-stepping through TF: make sure we ignore any events in
 877          * kernel space (but re-enable TF when returning to user mode).
 878          */
 879         if (condition & DR_STEP) {
 880                 if (!user_mode(regs))
 881                        goto clear_TF_reenable;
 882         }
 883
 884         /* Ok, finally something we can handle */
 885         tsk->thread.trap_no = 1;
 886         tsk->thread.error_code = error_code;
 887         info.si_signo = SIGTRAP;
 888         info.si_errno = 0;
 889         info.si_code = TRAP_BRKPT;
 890         info.si_addr = user_mode(regs) ? (void __user *)regs->rip : NULL;
 891         force_sig_info(SIGTRAP, &info, tsk);
 892
 893 clear_dr7:
 894         set_debugreg(0UL, 7);
 895         preempt_conditional_cli(regs);
 896         return;
 897
 898 clear_TF_reenable:
 899         set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
 900         regs->eflags &= ~TF_MASK;
 901         preempt_conditional_cli(regs);
 902 }
 903
 904 static int kernel_math_error(struct pt_regs *regs, const char *str, int trapnr)
 905 {
 906         const struct exception_table_entry *fixup;
 907         fixup = search_exception_tables(regs->rip);
 908         if (fixup) {
 909                 regs->rip = fixup->fixup;
 910                 return 1;
 911         }
 912         notify_die(DIE_GPF, str, regs, 0, trapnr, SIGFPE);
 913         /* Illegal floating point operation in the kernel */
 914         current->thread.trap_no = trapnr;
 915         die(str, regs, 0);
 916         return 0;
 917 }
 918
 919 /*
 920  * Note that we play around with the 'TS' bit in an attempt to get
 921  * the correct behaviour even in the presence of the asynchronous
 922  * IRQ13 behaviour
 923  */
 924 asmlinkage void do_coprocessor_error(struct pt_regs *regs)
 925 {
 926         void __user *rip = (void __user *)(regs->rip);
 927         struct task_struct * task;
 928         siginfo_t info;
 929         unsigned short cwd, swd;
 930
 931         conditional_sti(regs);
 932         if (!user_mode(regs) &&
 933             kernel_math_error(regs, "kernel x87 math error", 16))
 934                 return;
 935
 936         /*
 937          * Save the info for the exception handler and clear the error.
 938          */
 939         task = current;
 940         save_init_fpu(task);
 941         task->thread.trap_no = 16;
 942         task->thread.error_code = 0;
 943         info.si_signo = SIGFPE;
 944         info.si_errno = 0;
 945         info.si_code = __SI_FAULT;
 946         info.si_addr = rip;
 947         /*
 948          * (~cwd & swd) will mask out exceptions that are not set to unmasked
 949          * status.  0x3f is the exception bits in these regs, 0x200 is the
 950          * C1 reg you need in case of a stack fault, 0x040 is the stack
 951          * fault bit.  We should only be taking one exception at a time,
 952          * so if this combination doesn't produce any single exception,
 953          * then we have a bad program that isn't synchronizing its FPU usage
 954          * and it will suffer the consequences since we won't be able to
 955          * fully reproduce the context of the exception
 956          */
 957         cwd = get_fpu_cwd(task);
 958         swd = get_fpu_swd(task);
 959         switch (swd & ~cwd & 0x3f) {
 960                 case 0x000:
 961                 default:
 962                         break;
 963                 case 0x001: /* Invalid Op */
 964                         /*
 965                          * swd & 0x240 == 0x040: Stack Underflow
 966                          * swd & 0x240 == 0x240: Stack Overflow
 967                          * User must clear the SF bit (0x40) if set
 968                          */
 969                         info.si_code = FPE_FLTINV;
 970                         break;
 971                 case 0x002: /* Denormalize */
 972                 case 0x010: /* Underflow */
 973                         info.si_code = FPE_FLTUND;
 974                         break;
 975                 case 0x004: /* Zero Divide */
 976                         info.si_code = FPE_FLTDIV;
 977                         break;
 978                 case 0x008: /* Overflow */
 979                         info.si_code = FPE_FLTOVF;
 980                         break;
 981                 case 0x020: /* Precision */
 982                         info.si_code = FPE_FLTRES;
 983                         break;
 984         }
 985         force_sig_info(SIGFPE, &info, task);
 986 }
 987
 988 asmlinkage void bad_intr(void)
 989 {
 990         printk("bad interrupt");
 991 }
 992
 993 asmlinkage void do_simd_coprocessor_error(struct pt_regs *regs)
 994 {
 995         void __user *rip = (void __user *)(regs->rip);
 996         struct task_struct * task;
 997         siginfo_t info;
 998         unsigned short mxcsr;
 999
1000         conditional_sti(regs);
1001         if (!user_mode(regs) &&
1002                 kernel_math_error(regs, "kernel simd math error", 19))
1003                 return;
1004
1005         /*
1006          * Save the info for the exception handler and clear the error.
1007          */
1008         task = current;
1009         save_init_fpu(task);
1010         task->thread.trap_no = 19;
1011         task->thread.error_code = 0;
1012         info.si_signo = SIGFPE;
1013         info.si_errno = 0;
1014         info.si_code = __SI_FAULT;
1015         info.si_addr = rip;
1016         /*
1017          * The SIMD FPU exceptions are handled a little differently, as there
1018          * is only a single status/control register.  Thus, to determine which
1019          * unmasked exception was caught we must mask the exception mask bits
1020          * at 0x1f80, and then use these to mask the exception bits at 0x3f.
1021          */
1022         mxcsr = get_fpu_mxcsr(task);
1023         switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
1024                 case 0x000:
1025                 default:
1026                         break;
1027                 case 0x001: /* Invalid Op */
1028                         info.si_code = FPE_FLTINV;
1029                         break;
1030                 case 0x002: /* Denormalize */
1031                 case 0x010: /* Underflow */
1032                         info.si_code = FPE_FLTUND;
1033                         break;
1034                 case 0x004: /* Zero Divide */
1035                         info.si_code = FPE_FLTDIV;
1036                         break;
1037                 case 0x008: /* Overflow */
1038                         info.si_code = FPE_FLTOVF;
1039                         break;
1040                 case 0x020: /* Precision */
1041                         info.si_code = FPE_FLTRES;
1042                         break;
1043         }
1044         force_sig_info(SIGFPE, &info, task);
1045 }
1046
1047 asmlinkage void do_spurious_interrupt_bug(struct pt_regs * regs)
1048 {
1049 }
1050
1051 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
1052 {
1053 }
1054
1055 asmlinkage void __attribute__((weak)) mce_threshold_interrupt(void)
1056 {
1057 }
1058
1059 /*
1060  *  'math_state_restore()' saves the current math information in the
1061  * old math state array, and gets the new ones from the current task
1062  *
1063  * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1064  * Don't touch unless you *really* know how it works.
1065  */
1066 asmlinkage void math_state_restore(void)
1067 {
1068         struct task_struct *me = current;
1069         clts();                 /* Allow maths ops (or we recurse) */
1070
1071         if (!used_math())
1072                 init_fpu(me);
1073         restore_fpu_checking(&me->thread.i387.fxsave);
1074         task_thread_info(me)->status |= TS_USEDFPU;
1075         me->fpu_counter++;
1076 }
1077
1078 void __init trap_init(void)
1079 {
1080         set_intr_gate(0,&divide_error);
1081         set_intr_gate_ist(1,&debug,DEBUG_STACK);
1082         set_intr_gate_ist(2,&nmi,NMI_STACK);
1083         set_system_gate_ist(3,&int3,DEBUG_STACK); /* int3 can be called from all */
1084         set_system_gate(4,&overflow);   /* int4 can be called from all */
1085         set_intr_gate(5,&bounds);
1086         set_intr_gate(6,&invalid_op);
1087         set_intr_gate(7,&device_not_available);
1088         set_intr_gate_ist(8,&double_fault, DOUBLEFAULT_STACK);
1089         set_intr_gate(9,&coprocessor_segment_overrun);
1090         set_intr_gate(10,&invalid_TSS);
1091         set_intr_gate(11,&segment_not_present);
1092         set_intr_gate_ist(12,&stack_segment,STACKFAULT_STACK);
1093         set_intr_gate(13,&general_protection);
1094         set_intr_gate(14,&page_fault);
1095         set_intr_gate(15,&spurious_interrupt_bug);
1096         set_intr_gate(16,&coprocessor_error);
1097         set_intr_gate(17,&alignment_check);
1098 #ifdef CONFIG_X86_MCE
1099         set_intr_gate_ist(18,&machine_check, MCE_STACK);
1100 #endif
1101         set_intr_gate(19,&simd_coprocessor_error);
1102
1103 #ifdef CONFIG_IA32_EMULATION
1104         set_system_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
1105 #endif
1106
1107         /*
1108          * Should be a barrier for any external CPU state.
1109          */
1110         cpu_init();
1111 }
1112
1113
1114 static int __init oops_setup(char *s)
1115 {
1116         if (!s)
1117                 return -EINVAL;
1118         if (!strcmp(s, "panic"))
1119                 panic_on_oops = 1;
1120         return 0;
1121 }
1122 early_param("oops", oops_setup);
1123
1124 static int __init kstack_setup(char *s)
1125 {
1126         if (!s)
1127                 return -EINVAL;
1128         kstack_depth_to_print = simple_strtoul(s,NULL,0);
1129         return 0;
1130 }
1131 early_param("kstack", kstack_setup);