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  *  $Id: traps.c,v 1.36 2002/03/24 11:09:10 ak Exp $
  11  */
  12
  13 /*
  14  * 'Traps.c' handles hardware traps and faults after we have saved some
  15  * state in 'entry.S'.
  16  */
  17 #include <linux/config.h>
  18 #include <linux/sched.h>
  19 #include <linux/kernel.h>
  20 #include <linux/string.h>
  21 #include <linux/errno.h>
  22 #include <linux/ptrace.h>
  23 #include <linux/timer.h>
  24 #include <linux/mm.h>
  25 #include <linux/init.h>
  26 #include <linux/delay.h>
  27 #include <linux/spinlock.h>
  28 #include <linux/interrupt.h>
  29 #include <linux/module.h>
  30 #include <linux/moduleparam.h>
  31 #include <linux/nmi.h>
  32 #include <linux/kprobes.h>
  33 #include <linux/kexec.h>
  34
  35 #include <asm/system.h>
  36 #include <asm/uaccess.h>
  37 #include <asm/io.h>
  38 #include <asm/atomic.h>
  39 #include <asm/debugreg.h>
  40 #include <asm/desc.h>
  41 #include <asm/i387.h>
  42 #include <asm/kdebug.h>
  43 #include <asm/processor.h>
  44
  45 #include <asm/smp.h>
  46 #include <asm/pgalloc.h>
  47 #include <asm/pda.h>
  48 #include <asm/proto.h>
  49 #include <asm/nmi.h>
  50
  51 asmlinkage void divide_error(void);
  52 asmlinkage void debug(void);
  53 asmlinkage void nmi(void);
  54 asmlinkage void int3(void);
  55 asmlinkage void overflow(void);
  56 asmlinkage void bounds(void);
  57 asmlinkage void invalid_op(void);
  58 asmlinkage void device_not_available(void);
  59 asmlinkage void double_fault(void);
  60 asmlinkage void coprocessor_segment_overrun(void);
  61 asmlinkage void invalid_TSS(void);
  62 asmlinkage void segment_not_present(void);
  63 asmlinkage void stack_segment(void);
  64 asmlinkage void general_protection(void);
  65 asmlinkage void page_fault(void);
  66 asmlinkage void coprocessor_error(void);
  67 asmlinkage void simd_coprocessor_error(void);
  68 asmlinkage void reserved(void);
  69 asmlinkage void alignment_check(void);
  70 asmlinkage void machine_check(void);
  71 asmlinkage void spurious_interrupt_bug(void);
  72
  73 ATOMIC_NOTIFIER_HEAD(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);
  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);
  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 = 10;
 111
 112 #ifdef CONFIG_KALLSYMS
 113 #include <linux/kallsyms.h>
 114 int printk_address(unsigned long address)
 115 {
 116         unsigned long offset = 0, symsize;
 117         const char *symname;
 118         char *modname;
 119         char *delim = ":";
 120         char namebuf[128];
 121
 122         symname = kallsyms_lookup(address, &symsize, &offset, &modname, namebuf);
 123         if (!symname)
 124                 return printk("[<%016lx>]", address);
 125         if (!modname)
 126                 modname = delim = "";
 127         return printk("<%016lx>{%s%s%s%s%+ld}",
 128                       address, delim, modname, delim, symname, offset);
 129 }
 130 #else
 131 int printk_address(unsigned long address)
 132 {
 133         return printk("[<%016lx>]", address);
 134 }
 135 #endif
 136
 137 static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
 138                                         unsigned *usedp, const char **idp)
 139 {
 140         static char ids[][8] = {
 141                 [DEBUG_STACK - 1] = "#DB",
 142                 [NMI_STACK - 1] = "NMI",
 143                 [DOUBLEFAULT_STACK - 1] = "#DF",
 144                 [STACKFAULT_STACK - 1] = "#SS",
 145                 [MCE_STACK - 1] = "#MC",
 146 #if DEBUG_STKSZ > EXCEPTION_STKSZ
 147                 [N_EXCEPTION_STACKS ... N_EXCEPTION_STACKS + DEBUG_STKSZ / EXCEPTION_STKSZ - 2] = "#DB[?]"
 148 #endif
 149         };
 150         unsigned k;
 151
 152         for (k = 0; k < N_EXCEPTION_STACKS; k++) {
 153                 unsigned long end;
 154
 155                 switch (k + 1) {
 156 #if DEBUG_STKSZ > EXCEPTION_STKSZ
 157                 case DEBUG_STACK:
 158                         end = cpu_pda(cpu)->debugstack + DEBUG_STKSZ;
 159                         break;
 160 #endif
 161                 default:
 162                         end = per_cpu(init_tss, cpu).ist[k];
 163                         break;
 164                 }
 165                 if (stack >= end)
 166                         continue;
 167                 if (stack >= end - EXCEPTION_STKSZ) {
 168                         if (*usedp & (1U << k))
 169                                 break;
 170                         *usedp |= 1U << k;
 171                         *idp = ids[k];
 172                         return (unsigned long *)end;
 173                 }
 174 #if DEBUG_STKSZ > EXCEPTION_STKSZ
 175                 if (k == DEBUG_STACK - 1 && stack >= end - DEBUG_STKSZ) {
 176                         unsigned j = N_EXCEPTION_STACKS - 1;
 177
 178                         do {
 179                                 ++j;
 180                                 end -= EXCEPTION_STKSZ;
 181                                 ids[j][4] = '1' + (j - N_EXCEPTION_STACKS);
 182                         } while (stack < end - EXCEPTION_STKSZ);
 183                         if (*usedp & (1U << j))
 184                                 break;
 185                         *usedp |= 1U << j;
 186                         *idp = ids[j];
 187                         return (unsigned long *)end;
 188                 }
 189 #endif
 190         }
 191         return NULL;
 192 }
 193
 194 /*
 195  * x86-64 can have upto three kernel stacks:
 196  * process stack
 197  * interrupt stack
 198  * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
 199  */
 200
 201 void show_trace(unsigned long *stack)
 202 {
 203         const unsigned cpu = safe_smp_processor_id();
 204         unsigned long *irqstack_end = (unsigned long *)cpu_pda(cpu)->irqstackptr;
 205         int i;
 206         unsigned used = 0;
 207
 208         printk("\nCall Trace:");
 209
 210 #define HANDLE_STACK(cond) \
 211         do while (cond) { \
 212                 unsigned long addr = *stack++; \
 213                 if (kernel_text_address(addr)) { \
 214                         if (i > 50) { \
 215                                 printk("\n       "); \
 216                                 i = 0; \
 217                         } \
 218                         else \
 219                                 i += printk(" "); \
 220                         /* \
 221                          * If the address is either in the text segment of the \
 222                          * kernel, or in the region which contains vmalloc'ed \
 223                          * memory, it *may* be the address of a calling \
 224                          * routine; if so, print it so that someone tracing \
 225                          * down the cause of the crash will be able to figure \
 226                          * out the call path that was taken. \
 227                          */ \
 228                         i += printk_address(addr); \
 229                 } \
 230         } while (0)
 231
 232         for(i = 11; ; ) {
 233                 const char *id;
 234                 unsigned long *estack_end;
 235                 estack_end = in_exception_stack(cpu, (unsigned long)stack,
 236                                                 &used, &id);
 237
 238                 if (estack_end) {
 239                         i += printk(" <%s>", id);
 240                         HANDLE_STACK (stack < estack_end);
 241                         i += printk(" <EOE>");
 242                         stack = (unsigned long *) estack_end[-2];
 243                         continue;
 244                 }
 245                 if (irqstack_end) {
 246                         unsigned long *irqstack;
 247                         irqstack = irqstack_end -
 248                                 (IRQSTACKSIZE - 64) / sizeof(*irqstack);
 249
 250                         if (stack >= irqstack && stack < irqstack_end) {
 251                                 i += printk(" <IRQ>");
 252                                 HANDLE_STACK (stack < irqstack_end);
 253                                 stack = (unsigned long *) (irqstack_end[-1]);
 254                                 irqstack_end = NULL;
 255                                 i += printk(" <EOI>");
 256                                 continue;
 257                         }
 258                 }
 259                 break;
 260         }
 261
 262         HANDLE_STACK (((long) stack & (THREAD_SIZE-1)) != 0);
 263 #undef HANDLE_STACK
 264         printk("\n");
 265 }
 266
 267 void show_stack(struct task_struct *tsk, unsigned long * rsp)
 268 {
 269         unsigned long *stack;
 270         int i;
 271         const int cpu = safe_smp_processor_id();
 272         unsigned long *irqstack_end = (unsigned long *) (cpu_pda(cpu)->irqstackptr);
 273         unsigned long *irqstack = (unsigned long *) (cpu_pda(cpu)->irqstackptr - IRQSTACKSIZE);
 274
 275         // debugging aid: "show_stack(NULL, NULL);" prints the
 276         // back trace for this cpu.
 277
 278         if (rsp == NULL) {
 279                 if (tsk)
 280                         rsp = (unsigned long *)tsk->thread.rsp;
 281                 else
 282                         rsp = (unsigned long *)&rsp;
 283         }
 284
 285         stack = rsp;
 286         for(i=0; i < kstack_depth_to_print; i++) {
 287                 if (stack >= irqstack && stack <= irqstack_end) {
 288                         if (stack == irqstack_end) {
 289                                 stack = (unsigned long *) (irqstack_end[-1]);
 290                                 printk(" <EOI> ");
 291                         }
 292                 } else {
 293                 if (((long) stack & (THREAD_SIZE-1)) == 0)
 294                         break;
 295                 }
 296                 if (i && ((i % 4) == 0))
 297                         printk("\n       ");
 298                 printk("%016lx ", *stack++);
 299                 touch_nmi_watchdog();
 300         }
 301         show_trace((unsigned long *)rsp);
 302 }
 303
 304 /*
 305  * The architecture-independent dump_stack generator
 306  */
 307 void dump_stack(void)
 308 {
 309         unsigned long dummy;
 310         show_trace(&dummy);
 311 }
 312
 313 EXPORT_SYMBOL(dump_stack);
 314
 315 void show_registers(struct pt_regs *regs)
 316 {
 317         int i;
 318         int in_kernel = !user_mode(regs);
 319         unsigned long rsp;
 320         const int cpu = safe_smp_processor_id();
 321         struct task_struct *cur = cpu_pda(cpu)->pcurrent;
 322
 323                 rsp = regs->rsp;
 324
 325         printk("CPU %d ", cpu);
 326         __show_regs(regs);
 327         printk("Process %s (pid: %d, threadinfo %p, task %p)\n",
 328                 cur->comm, cur->pid, task_thread_info(cur), cur);
 329
 330         /*
 331          * When in-kernel, we also print out the stack and code at the
 332          * time of the fault..
 333          */
 334         if (in_kernel) {
 335
 336                 printk("Stack: ");
 337                 show_stack(NULL, (unsigned long*)rsp);
 338
 339                 printk("\nCode: ");
 340                 if (regs->rip < PAGE_OFFSET)
 341                         goto bad;
 342
 343                 for (i=0; i<20; i++) {
 344                         unsigned char c;
 345                         if (__get_user(c, &((unsigned char*)regs->rip)[i])) {
 346 bad:
 347                                 printk(" Bad RIP value.");
 348                                 break;
 349                         }
 350                         printk("%02x ", c);
 351                 }
 352         }
 353         printk("\n");
 354 }
 355
 356 void handle_BUG(struct pt_regs *regs)
 357 {
 358         struct bug_frame f;
 359         long len;
 360         const char *prefix = "";
 361
 362         if (user_mode(regs))
 363                 return;
 364         if (__copy_from_user(&f, (const void __user *) regs->rip,
 365                              sizeof(struct bug_frame)))
 366                 return;
 367         if (f.filename >= 0 ||
 368             f.ud2[0] != 0x0f || f.ud2[1] != 0x0b)
 369                 return;
 370         len = __strnlen_user((char *)(long)f.filename, PATH_MAX) - 1;
 371         if (len < 0 || len >= PATH_MAX)
 372                 f.filename = (int)(long)"unmapped filename";
 373         else if (len > 50) {
 374                 f.filename += len - 50;
 375                 prefix = "...";
 376         }
 377         printk("----------- [cut here ] --------- [please bite here ] ---------\n");
 378         printk(KERN_ALERT "Kernel BUG at %s%.50s:%d\n", prefix, (char *)(long)f.filename, f.line);
 379 }
 380
 381 #ifdef CONFIG_BUG
 382 void out_of_line_bug(void)
 383 {
 384         BUG();
 385 }
 386 #endif
 387
 388 static DEFINE_SPINLOCK(die_lock);
 389 static int die_owner = -1;
 390 static unsigned int die_nest_count;
 391
 392 unsigned __kprobes long oops_begin(void)
 393 {
 394         int cpu = safe_smp_processor_id();
 395         unsigned long flags;
 396
 397         /* racy, but better than risking deadlock. */
 398         local_irq_save(flags);
 399         if (!spin_trylock(&die_lock)) {
 400                 if (cpu == die_owner)
 401                         /* nested oops. should stop eventually */;
 402                 else
 403                         spin_lock(&die_lock);
 404         }
 405         die_nest_count++;
 406         die_owner = cpu;
 407         console_verbose();
 408         bust_spinlocks(1);
 409         return flags;
 410 }
 411
 412 void __kprobes oops_end(unsigned long flags)
 413 {
 414         die_owner = -1;
 415         bust_spinlocks(0);
 416         die_nest_count--;
 417         if (die_nest_count)
 418                 /* We still own the lock */
 419                 local_irq_restore(flags);
 420         else
 421                 /* Nest count reaches zero, release the lock. */
 422                 spin_unlock_irqrestore(&die_lock, flags);
 423         if (panic_on_oops)
 424                 panic("Oops");
 425 }
 426
 427 void __kprobes __die(const char * str, struct pt_regs * regs, long err)
 428 {
 429         static int die_counter;
 430         printk(KERN_EMERG "%s: %04lx [%u] ", str, err & 0xffff,++die_counter);
 431 #ifdef CONFIG_PREEMPT
 432         printk("PREEMPT ");
 433 #endif
 434 #ifdef CONFIG_SMP
 435         printk("SMP ");
 436 #endif
 437 #ifdef CONFIG_DEBUG_PAGEALLOC
 438         printk("DEBUG_PAGEALLOC");
 439 #endif
 440         printk("\n");
 441         notify_die(DIE_OOPS, str, regs, err, current->thread.trap_no, SIGSEGV);
 442         show_registers(regs);
 443         /* Executive summary in case the oops scrolled away */
 444         printk(KERN_ALERT "RIP ");
 445         printk_address(regs->rip);
 446         printk(" RSP <%016lx>\n", regs->rsp);
 447         if (kexec_should_crash(current))
 448                 crash_kexec(regs);
 449 }
 450
 451 void die(const char * str, struct pt_regs * regs, long err)
 452 {
 453         unsigned long flags = oops_begin();
 454
 455         handle_BUG(regs);
 456         __die(str, regs, err);
 457         oops_end(flags);
 458         do_exit(SIGSEGV);
 459 }
 460
 461 void __kprobes die_nmi(char *str, struct pt_regs *regs)
 462 {
 463         unsigned long flags = oops_begin();
 464
 465         /*
 466          * We are in trouble anyway, lets at least try
 467          * to get a message out.
 468          */
 469         printk(str, safe_smp_processor_id());
 470         show_registers(regs);
 471         if (kexec_should_crash(current))
 472                 crash_kexec(regs);
 473         if (panic_on_timeout || panic_on_oops)
 474                 panic("nmi watchdog");
 475         printk("console shuts up ...\n");
 476         oops_end(flags);
 477         nmi_exit();
 478         local_irq_enable();
 479         do_exit(SIGSEGV);
 480 }
 481
 482 static void __kprobes do_trap(int trapnr, int signr, char *str,
 483                               struct pt_regs * regs, long error_code,
 484                               siginfo_t *info)
 485 {
 486         struct task_struct *tsk = current;
 487
 488         tsk->thread.error_code = error_code;
 489         tsk->thread.trap_no = trapnr;
 490
 491         if (user_mode(regs)) {
 492                 if (exception_trace && unhandled_signal(tsk, signr))
 493                         printk(KERN_INFO
 494                                "%s[%d] trap %s rip:%lx rsp:%lx error:%lx\n",
 495                                tsk->comm, tsk->pid, str,
 496                                regs->rip, regs->rsp, error_code);
 497
 498                 if (info)
 499                         force_sig_info(signr, info, tsk);
 500                 else
 501                         force_sig(signr, tsk);
 502                 return;
 503         }
 504
 505
 506         /* kernel trap */
 507         {
 508                 const struct exception_table_entry *fixup;
 509                 fixup = search_exception_tables(regs->rip);
 510                 if (fixup)
 511                         regs->rip = fixup->fixup;
 512                 else
 513                         die(str, regs, error_code);
 514                 return;
 515         }
 516 }
 517
 518 #define DO_ERROR(trapnr, signr, str, name) \
 519 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
 520 { \
 521         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 522                                                         == NOTIFY_STOP) \
 523                 return; \
 524         conditional_sti(regs);                                          \
 525         do_trap(trapnr, signr, str, regs, error_code, NULL); \
 526 }
 527
 528 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
 529 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
 530 { \
 531         siginfo_t info; \
 532         info.si_signo = signr; \
 533         info.si_errno = 0; \
 534         info.si_code = sicode; \
 535         info.si_addr = (void __user *)siaddr; \
 536         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 537                                                         == NOTIFY_STOP) \
 538                 return; \
 539         conditional_sti(regs);                                          \
 540         do_trap(trapnr, signr, str, regs, error_code, &info); \
 541 }
 542
 543 DO_ERROR_INFO( 0, SIGFPE,  "divide error", divide_error, FPE_INTDIV, regs->rip)
 544 DO_ERROR( 4, SIGSEGV, "overflow", overflow)
 545 DO_ERROR( 5, SIGSEGV, "bounds", bounds)
 546 DO_ERROR_INFO( 6, SIGILL,  "invalid opcode", invalid_op, ILL_ILLOPN, regs->rip)
 547 DO_ERROR( 7, SIGSEGV, "device not available", device_not_available)
 548 DO_ERROR( 9, SIGFPE,  "coprocessor segment overrun", coprocessor_segment_overrun)
 549 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
 550 DO_ERROR(11, SIGBUS,  "segment not present", segment_not_present)
 551 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
 552 DO_ERROR(18, SIGSEGV, "reserved", reserved)
 553
 554 /* Runs on IST stack */
 555 asmlinkage void do_stack_segment(struct pt_regs *regs, long error_code)
 556 {
 557         if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
 558                         12, SIGBUS) == NOTIFY_STOP)
 559                 return;
 560         preempt_conditional_sti(regs);
 561         do_trap(12, SIGBUS, "stack segment", regs, error_code, NULL);
 562         preempt_conditional_cli(regs);
 563 }
 564
 565 asmlinkage void do_double_fault(struct pt_regs * regs, long error_code)
 566 {
 567         static const char str[] = "double fault";
 568         struct task_struct *tsk = current;
 569
 570         /* Return not checked because double check cannot be ignored */
 571         notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV);
 572
 573         tsk->thread.error_code = error_code;
 574         tsk->thread.trap_no = 8;
 575
 576         /* This is always a kernel trap and never fixable (and thus must
 577            never return). */
 578         for (;;)
 579                 die(str, regs, error_code);
 580 }
 581
 582 asmlinkage void __kprobes do_general_protection(struct pt_regs * regs,
 583                                                 long error_code)
 584 {
 585         struct task_struct *tsk = current;
 586
 587         conditional_sti(regs);
 588
 589         tsk->thread.error_code = error_code;
 590         tsk->thread.trap_no = 13;
 591
 592         if (user_mode(regs)) {
 593                 if (exception_trace && unhandled_signal(tsk, SIGSEGV))
 594                         printk(KERN_INFO
 595                        "%s[%d] general protection rip:%lx rsp:%lx error:%lx\n",
 596                                tsk->comm, tsk->pid,
 597                                regs->rip, regs->rsp, error_code);
 598
 599                 force_sig(SIGSEGV, tsk);
 600                 return;
 601         }
 602
 603         /* kernel gp */
 604         {
 605                 const struct exception_table_entry *fixup;
 606                 fixup = search_exception_tables(regs->rip);
 607                 if (fixup) {
 608                         regs->rip = fixup->fixup;
 609                         return;
 610                 }
 611                 if (notify_die(DIE_GPF, "general protection fault", regs,
 612                                         error_code, 13, SIGSEGV) == NOTIFY_STOP)
 613                         return;
 614                 die("general protection fault", regs, error_code);
 615         }
 616 }
 617
 618 static __kprobes void
 619 mem_parity_error(unsigned char reason, struct pt_regs * regs)
 620 {
 621         printk("Uhhuh. NMI received. Dazed and confused, but trying to continue\n");
 622         printk("You probably have a hardware problem with your RAM chips\n");
 623
 624         /* Clear and disable the memory parity error line. */
 625         reason = (reason & 0xf) | 4;
 626         outb(reason, 0x61);
 627 }
 628
 629 static __kprobes void
 630 io_check_error(unsigned char reason, struct pt_regs * regs)
 631 {
 632         printk("NMI: IOCK error (debug interrupt?)\n");
 633         show_registers(regs);
 634
 635         /* Re-enable the IOCK line, wait for a few seconds */
 636         reason = (reason & 0xf) | 8;
 637         outb(reason, 0x61);
 638         mdelay(2000);
 639         reason &= ~8;
 640         outb(reason, 0x61);
 641 }
 642
 643 static __kprobes void
 644 unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
 645 {       printk("Uhhuh. NMI received for unknown reason %02x.\n", reason);
 646         printk("Dazed and confused, but trying to continue\n");
 647         printk("Do you have a strange power saving mode enabled?\n");
 648 }
 649
 650 /* Runs on IST stack. This code must keep interrupts off all the time.
 651    Nested NMIs are prevented by the CPU. */
 652 asmlinkage __kprobes void default_do_nmi(struct pt_regs *regs)
 653 {
 654         unsigned char reason = 0;
 655         int cpu;
 656
 657         cpu = smp_processor_id();
 658
 659         /* Only the BSP gets external NMIs from the system.  */
 660         if (!cpu)
 661                 reason = get_nmi_reason();
 662
 663         if (!(reason & 0xc0)) {
 664                 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
 665                                                                 == NOTIFY_STOP)
 666                         return;
 667 #ifdef CONFIG_X86_LOCAL_APIC
 668                 /*
 669                  * Ok, so this is none of the documented NMI sources,
 670                  * so it must be the NMI watchdog.
 671                  */
 672                 if (nmi_watchdog > 0) {
 673                         nmi_watchdog_tick(regs,reason);
 674                         return;
 675                 }
 676 #endif
 677                 unknown_nmi_error(reason, regs);
 678                 return;
 679         }
 680         if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
 681                 return;
 682
 683         /* AK: following checks seem to be broken on modern chipsets. FIXME */
 684
 685         if (reason & 0x80)
 686                 mem_parity_error(reason, regs);
 687         if (reason & 0x40)
 688                 io_check_error(reason, regs);
 689 }
 690
 691 /* runs on IST stack. */
 692 asmlinkage void __kprobes do_int3(struct pt_regs * regs, long error_code)
 693 {
 694         if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP) == NOTIFY_STOP) {
 695                 return;
 696         }
 697         preempt_conditional_sti(regs);
 698         do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
 699         preempt_conditional_cli(regs);
 700 }
 701
 702 /* Help handler running on IST stack to switch back to user stack
 703    for scheduling or signal handling. The actual stack switch is done in
 704    entry.S */
 705 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
 706 {
 707         struct pt_regs *regs = eregs;
 708         /* Did already sync */
 709         if (eregs == (struct pt_regs *)eregs->rsp)
 710                 ;
 711         /* Exception from user space */
 712         else if (user_mode(eregs))
 713                 regs = task_pt_regs(current);
 714         /* Exception from kernel and interrupts are enabled. Move to
 715            kernel process stack. */
 716         else if (eregs->eflags & X86_EFLAGS_IF)
 717                 regs = (struct pt_regs *)(eregs->rsp -= sizeof(struct pt_regs));
 718         if (eregs != regs)
 719                 *regs = *eregs;
 720         return regs;
 721 }
 722
 723 /* runs on IST stack. */
 724 asmlinkage void __kprobes do_debug(struct pt_regs * regs,
 725                                    unsigned long error_code)
 726 {
 727         unsigned long condition;
 728         struct task_struct *tsk = current;
 729         siginfo_t info;
 730
 731         get_debugreg(condition, 6);
 732
 733         if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
 734                                                 SIGTRAP) == NOTIFY_STOP)
 735                 return;
 736
 737         preempt_conditional_sti(regs);
 738
 739         /* Mask out spurious debug traps due to lazy DR7 setting */
 740         if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
 741                 if (!tsk->thread.debugreg7) {
 742                         goto clear_dr7;
 743                 }
 744         }
 745
 746         tsk->thread.debugreg6 = condition;
 747
 748         /* Mask out spurious TF errors due to lazy TF clearing */
 749         if (condition & DR_STEP) {
 750                 /*
 751                  * The TF error should be masked out only if the current
 752                  * process is not traced and if the TRAP flag has been set
 753                  * previously by a tracing process (condition detected by
 754                  * the PT_DTRACE flag); remember that the i386 TRAP flag
 755                  * can be modified by the process itself in user mode,
 756                  * allowing programs to debug themselves without the ptrace()
 757                  * interface.
 758                  */
 759                 if (!user_mode(regs))
 760                        goto clear_TF_reenable;
 761                 /*
 762                  * Was the TF flag set by a debugger? If so, clear it now,
 763                  * so that register information is correct.
 764                  */
 765                 if (tsk->ptrace & PT_DTRACE) {
 766                         regs->eflags &= ~TF_MASK;
 767                         tsk->ptrace &= ~PT_DTRACE;
 768                 }
 769         }
 770
 771         /* Ok, finally something we can handle */
 772         tsk->thread.trap_no = 1;
 773         tsk->thread.error_code = error_code;
 774         info.si_signo = SIGTRAP;
 775         info.si_errno = 0;
 776         info.si_code = TRAP_BRKPT;
 777         info.si_addr = user_mode(regs) ? (void __user *)regs->rip : NULL;
 778         force_sig_info(SIGTRAP, &info, tsk);
 779
 780 clear_dr7:
 781         set_debugreg(0UL, 7);
 782         preempt_conditional_cli(regs);
 783         return;
 784
 785 clear_TF_reenable:
 786         set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
 787         regs->eflags &= ~TF_MASK;
 788         preempt_conditional_cli(regs);
 789 }
 790
 791 static int kernel_math_error(struct pt_regs *regs, const char *str, int trapnr)
 792 {
 793         const struct exception_table_entry *fixup;
 794         fixup = search_exception_tables(regs->rip);
 795         if (fixup) {
 796                 regs->rip = fixup->fixup;
 797                 return 1;
 798         }
 799         notify_die(DIE_GPF, str, regs, 0, trapnr, SIGFPE);
 800         /* Illegal floating point operation in the kernel */
 801         current->thread.trap_no = trapnr;
 802         die(str, regs, 0);
 803         return 0;
 804 }
 805
 806 /*
 807  * Note that we play around with the 'TS' bit in an attempt to get
 808  * the correct behaviour even in the presence of the asynchronous
 809  * IRQ13 behaviour
 810  */
 811 asmlinkage void do_coprocessor_error(struct pt_regs *regs)
 812 {
 813         void __user *rip = (void __user *)(regs->rip);
 814         struct task_struct * task;
 815         siginfo_t info;
 816         unsigned short cwd, swd;
 817
 818         conditional_sti(regs);
 819         if (!user_mode(regs) &&
 820             kernel_math_error(regs, "kernel x87 math error", 16))
 821                 return;
 822
 823         /*
 824          * Save the info for the exception handler and clear the error.
 825          */
 826         task = current;
 827         save_init_fpu(task);
 828         task->thread.trap_no = 16;
 829         task->thread.error_code = 0;
 830         info.si_signo = SIGFPE;
 831         info.si_errno = 0;
 832         info.si_code = __SI_FAULT;
 833         info.si_addr = rip;
 834         /*
 835          * (~cwd & swd) will mask out exceptions that are not set to unmasked
 836          * status.  0x3f is the exception bits in these regs, 0x200 is the
 837          * C1 reg you need in case of a stack fault, 0x040 is the stack
 838          * fault bit.  We should only be taking one exception at a time,
 839          * so if this combination doesn't produce any single exception,
 840          * then we have a bad program that isn't synchronizing its FPU usage
 841          * and it will suffer the consequences since we won't be able to
 842          * fully reproduce the context of the exception
 843          */
 844         cwd = get_fpu_cwd(task);
 845         swd = get_fpu_swd(task);
 846         switch (swd & ~cwd & 0x3f) {
 847                 case 0x000:
 848                 default:
 849                         break;
 850                 case 0x001: /* Invalid Op */
 851                         /*
 852                          * swd & 0x240 == 0x040: Stack Underflow
 853                          * swd & 0x240 == 0x240: Stack Overflow
 854                          * User must clear the SF bit (0x40) if set
 855                          */
 856                         info.si_code = FPE_FLTINV;
 857                         break;
 858                 case 0x002: /* Denormalize */
 859                 case 0x010: /* Underflow */
 860                         info.si_code = FPE_FLTUND;
 861                         break;
 862                 case 0x004: /* Zero Divide */
 863                         info.si_code = FPE_FLTDIV;
 864                         break;
 865                 case 0x008: /* Overflow */
 866                         info.si_code = FPE_FLTOVF;
 867                         break;
 868                 case 0x020: /* Precision */
 869                         info.si_code = FPE_FLTRES;
 870                         break;
 871         }
 872         force_sig_info(SIGFPE, &info, task);
 873 }
 874
 875 asmlinkage void bad_intr(void)
 876 {
 877         printk("bad interrupt");
 878 }
 879
 880 asmlinkage void do_simd_coprocessor_error(struct pt_regs *regs)
 881 {
 882         void __user *rip = (void __user *)(regs->rip);
 883         struct task_struct * task;
 884         siginfo_t info;
 885         unsigned short mxcsr;
 886
 887         conditional_sti(regs);
 888         if (!user_mode(regs) &&
 889                 kernel_math_error(regs, "kernel simd math error", 19))
 890                 return;
 891
 892         /*
 893          * Save the info for the exception handler and clear the error.
 894          */
 895         task = current;
 896         save_init_fpu(task);
 897         task->thread.trap_no = 19;
 898         task->thread.error_code = 0;
 899         info.si_signo = SIGFPE;
 900         info.si_errno = 0;
 901         info.si_code = __SI_FAULT;
 902         info.si_addr = rip;
 903         /*
 904          * The SIMD FPU exceptions are handled a little differently, as there
 905          * is only a single status/control register.  Thus, to determine which
 906          * unmasked exception was caught we must mask the exception mask bits
 907          * at 0x1f80, and then use these to mask the exception bits at 0x3f.
 908          */
 909         mxcsr = get_fpu_mxcsr(task);
 910         switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
 911                 case 0x000:
 912                 default:
 913                         break;
 914                 case 0x001: /* Invalid Op */
 915                         info.si_code = FPE_FLTINV;
 916                         break;
 917                 case 0x002: /* Denormalize */
 918                 case 0x010: /* Underflow */
 919                         info.si_code = FPE_FLTUND;
 920                         break;
 921                 case 0x004: /* Zero Divide */
 922                         info.si_code = FPE_FLTDIV;
 923                         break;
 924                 case 0x008: /* Overflow */
 925                         info.si_code = FPE_FLTOVF;
 926                         break;
 927                 case 0x020: /* Precision */
 928                         info.si_code = FPE_FLTRES;
 929                         break;
 930         }
 931         force_sig_info(SIGFPE, &info, task);
 932 }
 933
 934 asmlinkage void do_spurious_interrupt_bug(struct pt_regs * regs)
 935 {
 936 }
 937
 938 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
 939 {
 940 }
 941
 942 asmlinkage void __attribute__((weak)) mce_threshold_interrupt(void)
 943 {
 944 }
 945
 946 /*
 947  *  'math_state_restore()' saves the current math information in the
 948  * old math state array, and gets the new ones from the current task
 949  *
 950  * Careful.. There are problems with IBM-designed IRQ13 behaviour.
 951  * Don't touch unless you *really* know how it works.
 952  */
 953 asmlinkage void math_state_restore(void)
 954 {
 955         struct task_struct *me = current;
 956         clts();                 /* Allow maths ops (or we recurse) */
 957
 958         if (!used_math())
 959                 init_fpu(me);
 960         restore_fpu_checking(&me->thread.i387.fxsave);
 961         task_thread_info(me)->status |= TS_USEDFPU;
 962 }
 963
 964 void __init trap_init(void)
 965 {
 966         set_intr_gate(0,&divide_error);
 967         set_intr_gate_ist(1,&debug,DEBUG_STACK);
 968         set_intr_gate_ist(2,&nmi,NMI_STACK);
 969         set_system_gate_ist(3,&int3,DEBUG_STACK); /* int3 can be called from all */
 970         set_system_gate(4,&overflow);   /* int4 can be called from all */
 971         set_intr_gate(5,&bounds);
 972         set_intr_gate(6,&invalid_op);
 973         set_intr_gate(7,&device_not_available);
 974         set_intr_gate_ist(8,&double_fault, DOUBLEFAULT_STACK);
 975         set_intr_gate(9,&coprocessor_segment_overrun);
 976         set_intr_gate(10,&invalid_TSS);
 977         set_intr_gate(11,&segment_not_present);
 978         set_intr_gate_ist(12,&stack_segment,STACKFAULT_STACK);
 979         set_intr_gate(13,&general_protection);
 980         set_intr_gate(14,&page_fault);
 981         set_intr_gate(15,&spurious_interrupt_bug);
 982         set_intr_gate(16,&coprocessor_error);
 983         set_intr_gate(17,&alignment_check);
 984 #ifdef CONFIG_X86_MCE
 985         set_intr_gate_ist(18,&machine_check, MCE_STACK);
 986 #endif
 987         set_intr_gate(19,&simd_coprocessor_error);
 988
 989 #ifdef CONFIG_IA32_EMULATION
 990         set_system_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
 991 #endif
 992
 993         /*
 994          * Should be a barrier for any external CPU state.
 995          */
 996         cpu_init();
 997 }
 998
 999
1000 /* Actual parsing is done early in setup.c. */
1001 static int __init oops_dummy(char *s)
1002 {
1003         panic_on_oops = 1;
1004         return 1;
1005 }
1006 __setup("oops=", oops_dummy);
1007
1008 static int __init kstack_setup(char *s)
1009 {
1010         kstack_depth_to_print = simple_strtoul(s,NULL,0);
1011         return 1;
1012 }
1013 __setup("kstack=", kstack_setup);
1014