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         preempt_enable_no_resched();
 106 }
 107
 108 static int kstack_depth_to_print = 10;
 109
 110 #ifdef CONFIG_KALLSYMS
 111 #include <linux/kallsyms.h>
 112 int printk_address(unsigned long address)
 113 {
 114         unsigned long offset = 0, symsize;
 115         const char *symname;
 116         char *modname;
 117         char *delim = ":";
 118         char namebuf[128];
 119
 120         symname = kallsyms_lookup(address, &symsize, &offset, &modname, namebuf);
 121         if (!symname)
 122                 return printk("[<%016lx>]", address);
 123         if (!modname)
 124                 modname = delim = "";
 125         return printk("<%016lx>{%s%s%s%s%+ld}",
 126                       address, delim, modname, delim, symname, offset);
 127 }
 128 #else
 129 int printk_address(unsigned long address)
 130 {
 131         return printk("[<%016lx>]", address);
 132 }
 133 #endif
 134
 135 static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
 136                                         unsigned *usedp, const char **idp)
 137 {
 138         static char ids[][8] = {
 139                 [DEBUG_STACK - 1] = "#DB",
 140                 [NMI_STACK - 1] = "NMI",
 141                 [DOUBLEFAULT_STACK - 1] = "#DF",
 142                 [STACKFAULT_STACK - 1] = "#SS",
 143                 [MCE_STACK - 1] = "#MC",
 144 #if DEBUG_STKSZ > EXCEPTION_STKSZ
 145                 [N_EXCEPTION_STACKS ... N_EXCEPTION_STACKS + DEBUG_STKSZ / EXCEPTION_STKSZ - 2] = "#DB[?]"
 146 #endif
 147         };
 148         unsigned k;
 149
 150         for (k = 0; k < N_EXCEPTION_STACKS; k++) {
 151                 unsigned long end;
 152
 153                 switch (k + 1) {
 154 #if DEBUG_STKSZ > EXCEPTION_STKSZ
 155                 case DEBUG_STACK:
 156                         end = cpu_pda(cpu)->debugstack + DEBUG_STKSZ;
 157                         break;
 158 #endif
 159                 default:
 160                         end = per_cpu(init_tss, cpu).ist[k];
 161                         break;
 162                 }
 163                 if (stack >= end)
 164                         continue;
 165                 if (stack >= end - EXCEPTION_STKSZ) {
 166                         if (*usedp & (1U << k))
 167                                 break;
 168                         *usedp |= 1U << k;
 169                         *idp = ids[k];
 170                         return (unsigned long *)end;
 171                 }
 172 #if DEBUG_STKSZ > EXCEPTION_STKSZ
 173                 if (k == DEBUG_STACK - 1 && stack >= end - DEBUG_STKSZ) {
 174                         unsigned j = N_EXCEPTION_STACKS - 1;
 175
 176                         do {
 177                                 ++j;
 178                                 end -= EXCEPTION_STKSZ;
 179                                 ids[j][4] = '1' + (j - N_EXCEPTION_STACKS);
 180                         } while (stack < end - EXCEPTION_STKSZ);
 181                         if (*usedp & (1U << j))
 182                                 break;
 183                         *usedp |= 1U << j;
 184                         *idp = ids[j];
 185                         return (unsigned long *)end;
 186                 }
 187 #endif
 188         }
 189         return NULL;
 190 }
 191
 192 /*
 193  * x86-64 can have upto three kernel stacks:
 194  * process stack
 195  * interrupt stack
 196  * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
 197  */
 198
 199 void show_trace(unsigned long *stack)
 200 {
 201         const unsigned cpu = safe_smp_processor_id();
 202         unsigned long *irqstack_end = (unsigned long *)cpu_pda(cpu)->irqstackptr;
 203         int i;
 204         unsigned used = 0;
 205
 206         printk("\nCall Trace:");
 207
 208 #define HANDLE_STACK(cond) \
 209         do while (cond) { \
 210                 unsigned long addr = *stack++; \
 211                 if (kernel_text_address(addr)) { \
 212                         if (i > 50) { \
 213                                 printk("\n       "); \
 214                                 i = 0; \
 215                         } \
 216                         else \
 217                                 i += printk(" "); \
 218                         /* \
 219                          * If the address is either in the text segment of the \
 220                          * kernel, or in the region which contains vmalloc'ed \
 221                          * memory, it *may* be the address of a calling \
 222                          * routine; if so, print it so that someone tracing \
 223                          * down the cause of the crash will be able to figure \
 224                          * out the call path that was taken. \
 225                          */ \
 226                         i += printk_address(addr); \
 227                 } \
 228         } while (0)
 229
 230         for(i = 11; ; ) {
 231                 const char *id;
 232                 unsigned long *estack_end;
 233                 estack_end = in_exception_stack(cpu, (unsigned long)stack,
 234                                                 &used, &id);
 235
 236                 if (estack_end) {
 237                         i += printk(" <%s>", id);
 238                         HANDLE_STACK (stack < estack_end);
 239                         i += printk(" <EOE>");
 240                         stack = (unsigned long *) estack_end[-2];
 241                         continue;
 242                 }
 243                 if (irqstack_end) {
 244                         unsigned long *irqstack;
 245                         irqstack = irqstack_end -
 246                                 (IRQSTACKSIZE - 64) / sizeof(*irqstack);
 247
 248                         if (stack >= irqstack && stack < irqstack_end) {
 249                                 i += printk(" <IRQ>");
 250                                 HANDLE_STACK (stack < irqstack_end);
 251                                 stack = (unsigned long *) (irqstack_end[-1]);
 252                                 irqstack_end = NULL;
 253                                 i += printk(" <EOI>");
 254                                 continue;
 255                         }
 256                 }
 257                 break;
 258         }
 259
 260         HANDLE_STACK (((long) stack & (THREAD_SIZE-1)) != 0);
 261 #undef HANDLE_STACK
 262         printk("\n");
 263 }
 264
 265 void show_stack(struct task_struct *tsk, unsigned long * rsp)
 266 {
 267         unsigned long *stack;
 268         int i;
 269         const int cpu = safe_smp_processor_id();
 270         unsigned long *irqstack_end = (unsigned long *) (cpu_pda(cpu)->irqstackptr);
 271         unsigned long *irqstack = (unsigned long *) (cpu_pda(cpu)->irqstackptr - IRQSTACKSIZE);
 272
 273         // debugging aid: "show_stack(NULL, NULL);" prints the
 274         // back trace for this cpu.
 275
 276         if (rsp == NULL) {
 277                 if (tsk)
 278                         rsp = (unsigned long *)tsk->thread.rsp;
 279                 else
 280                         rsp = (unsigned long *)&rsp;
 281         }
 282
 283         stack = rsp;
 284         for(i=0; i < kstack_depth_to_print; i++) {
 285                 if (stack >= irqstack && stack <= irqstack_end) {
 286                         if (stack == irqstack_end) {
 287                                 stack = (unsigned long *) (irqstack_end[-1]);
 288                                 printk(" <EOI> ");
 289                         }
 290                 } else {
 291                 if (((long) stack & (THREAD_SIZE-1)) == 0)
 292                         break;
 293                 }
 294                 if (i && ((i % 4) == 0))
 295                         printk("\n       ");
 296                 printk("%016lx ", *stack++);
 297                 touch_nmi_watchdog();
 298         }
 299         show_trace((unsigned long *)rsp);
 300 }
 301
 302 /*
 303  * The architecture-independent dump_stack generator
 304  */
 305 void dump_stack(void)
 306 {
 307         unsigned long dummy;
 308         show_trace(&dummy);
 309 }
 310
 311 EXPORT_SYMBOL(dump_stack);
 312
 313 void show_registers(struct pt_regs *regs)
 314 {
 315         int i;
 316         int in_kernel = !user_mode(regs);
 317         unsigned long rsp;
 318         const int cpu = safe_smp_processor_id();
 319         struct task_struct *cur = cpu_pda(cpu)->pcurrent;
 320
 321                 rsp = regs->rsp;
 322
 323         printk("CPU %d ", cpu);
 324         __show_regs(regs);
 325         printk("Process %s (pid: %d, threadinfo %p, task %p)\n",
 326                 cur->comm, cur->pid, task_thread_info(cur), cur);
 327
 328         /*
 329          * When in-kernel, we also print out the stack and code at the
 330          * time of the fault..
 331          */
 332         if (in_kernel) {
 333
 334                 printk("Stack: ");
 335                 show_stack(NULL, (unsigned long*)rsp);
 336
 337                 printk("\nCode: ");
 338                 if (regs->rip < PAGE_OFFSET)
 339                         goto bad;
 340
 341                 for (i=0; i<20; i++) {
 342                         unsigned char c;
 343                         if (__get_user(c, &((unsigned char*)regs->rip)[i])) {
 344 bad:
 345                                 printk(" Bad RIP value.");
 346                                 break;
 347                         }
 348                         printk("%02x ", c);
 349                 }
 350         }
 351         printk("\n");
 352 }
 353
 354 void handle_BUG(struct pt_regs *regs)
 355 {
 356         struct bug_frame f;
 357         long len;
 358         const char *prefix = "";
 359
 360         if (user_mode(regs))
 361                 return;
 362         if (__copy_from_user(&f, (const void __user *) regs->rip,
 363                              sizeof(struct bug_frame)))
 364                 return;
 365         if (f.filename >= 0 ||
 366             f.ud2[0] != 0x0f || f.ud2[1] != 0x0b)
 367                 return;
 368         len = __strnlen_user((char *)(long)f.filename, PATH_MAX) - 1;
 369         if (len < 0 || len >= PATH_MAX)
 370                 f.filename = (int)(long)"unmapped filename";
 371         else if (len > 50) {
 372                 f.filename += len - 50;
 373                 prefix = "...";
 374         }
 375         printk("----------- [cut here ] --------- [please bite here ] ---------\n");
 376         printk(KERN_ALERT "Kernel BUG at %s%.50s:%d\n", prefix, (char *)(long)f.filename, f.line);
 377 }
 378
 379 #ifdef CONFIG_BUG
 380 void out_of_line_bug(void)
 381 {
 382         BUG();
 383 }
 384 #endif
 385
 386 static DEFINE_SPINLOCK(die_lock);
 387 static int die_owner = -1;
 388 static unsigned int die_nest_count;
 389
 390 unsigned __kprobes long oops_begin(void)
 391 {
 392         int cpu = safe_smp_processor_id();
 393         unsigned long flags;
 394
 395         /* racy, but better than risking deadlock. */
 396         local_irq_save(flags);
 397         if (!spin_trylock(&die_lock)) {
 398                 if (cpu == die_owner)
 399                         /* nested oops. should stop eventually */;
 400                 else
 401                         spin_lock(&die_lock);
 402         }
 403         die_nest_count++;
 404         die_owner = cpu;
 405         console_verbose();
 406         bust_spinlocks(1);
 407         return flags;
 408 }
 409
 410 void __kprobes oops_end(unsigned long flags)
 411 {
 412         die_owner = -1;
 413         bust_spinlocks(0);
 414         die_nest_count--;
 415         if (die_nest_count)
 416                 /* We still own the lock */
 417                 local_irq_restore(flags);
 418         else
 419                 /* Nest count reaches zero, release the lock. */
 420                 spin_unlock_irqrestore(&die_lock, flags);
 421         if (panic_on_oops)
 422                 panic("Oops");
 423 }
 424
 425 void __kprobes __die(const char * str, struct pt_regs * regs, long err)
 426 {
 427         static int die_counter;
 428         printk(KERN_EMERG "%s: %04lx [%u] ", str, err & 0xffff,++die_counter);
 429 #ifdef CONFIG_PREEMPT
 430         printk("PREEMPT ");
 431 #endif
 432 #ifdef CONFIG_SMP
 433         printk("SMP ");
 434 #endif
 435 #ifdef CONFIG_DEBUG_PAGEALLOC
 436         printk("DEBUG_PAGEALLOC");
 437 #endif
 438         printk("\n");
 439         notify_die(DIE_OOPS, str, regs, err, current->thread.trap_no, SIGSEGV);
 440         show_registers(regs);
 441         /* Executive summary in case the oops scrolled away */
 442         printk(KERN_ALERT "RIP ");
 443         printk_address(regs->rip);
 444         printk(" RSP <%016lx>\n", regs->rsp);
 445         if (kexec_should_crash(current))
 446                 crash_kexec(regs);
 447 }
 448
 449 void die(const char * str, struct pt_regs * regs, long err)
 450 {
 451         unsigned long flags = oops_begin();
 452
 453         handle_BUG(regs);
 454         __die(str, regs, err);
 455         oops_end(flags);
 456         do_exit(SIGSEGV);
 457 }
 458
 459 void __kprobes die_nmi(char *str, struct pt_regs *regs)
 460 {
 461         unsigned long flags = oops_begin();
 462
 463         /*
 464          * We are in trouble anyway, lets at least try
 465          * to get a message out.
 466          */
 467         printk(str, safe_smp_processor_id());
 468         show_registers(regs);
 469         if (kexec_should_crash(current))
 470                 crash_kexec(regs);
 471         if (panic_on_timeout || panic_on_oops)
 472                 panic("nmi watchdog");
 473         printk("console shuts up ...\n");
 474         oops_end(flags);
 475         nmi_exit();
 476         local_irq_enable();
 477         do_exit(SIGSEGV);
 478 }
 479
 480 static void __kprobes do_trap(int trapnr, int signr, char *str,
 481                               struct pt_regs * regs, long error_code,
 482                               siginfo_t *info)
 483 {
 484         struct task_struct *tsk = current;
 485
 486         conditional_sti(regs);
 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         do_trap(trapnr, signr, str, regs, error_code, NULL); \
 525 }
 526
 527 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
 528 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
 529 { \
 530         siginfo_t info; \
 531         info.si_signo = signr; \
 532         info.si_errno = 0; \
 533         info.si_code = sicode; \
 534         info.si_addr = (void __user *)siaddr; \
 535         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 536                                                         == NOTIFY_STOP) \
 537                 return; \
 538         do_trap(trapnr, signr, str, regs, error_code, &info); \
 539 }
 540
 541 DO_ERROR_INFO( 0, SIGFPE,  "divide error", divide_error, FPE_INTDIV, regs->rip)
 542 DO_ERROR( 4, SIGSEGV, "overflow", overflow)
 543 DO_ERROR( 5, SIGSEGV, "bounds", bounds)
 544 DO_ERROR_INFO( 6, SIGILL,  "invalid opcode", invalid_op, ILL_ILLOPN, regs->rip)
 545 DO_ERROR( 7, SIGSEGV, "device not available", device_not_available)
 546 DO_ERROR( 9, SIGFPE,  "coprocessor segment overrun", coprocessor_segment_overrun)
 547 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
 548 DO_ERROR(11, SIGBUS,  "segment not present", segment_not_present)
 549 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
 550 DO_ERROR(18, SIGSEGV, "reserved", reserved)
 551 DO_ERROR(12, SIGBUS,  "stack segment", stack_segment)
 552
 553 asmlinkage void do_double_fault(struct pt_regs * regs, long error_code)
 554 {
 555         static const char str[] = "double fault";
 556         struct task_struct *tsk = current;
 557
 558         /* Return not checked because double check cannot be ignored */
 559         notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV);
 560
 561         tsk->thread.error_code = error_code;
 562         tsk->thread.trap_no = 8;
 563
 564         /* This is always a kernel trap and never fixable (and thus must
 565            never return). */
 566         for (;;)
 567                 die(str, regs, error_code);
 568 }
 569
 570 asmlinkage void __kprobes do_general_protection(struct pt_regs * regs,
 571                                                 long error_code)
 572 {
 573         struct task_struct *tsk = current;
 574
 575         conditional_sti(regs);
 576
 577         tsk->thread.error_code = error_code;
 578         tsk->thread.trap_no = 13;
 579
 580         if (user_mode(regs)) {
 581                 if (exception_trace && unhandled_signal(tsk, SIGSEGV))
 582                         printk(KERN_INFO
 583                        "%s[%d] general protection rip:%lx rsp:%lx error:%lx\n",
 584                                tsk->comm, tsk->pid,
 585                                regs->rip, regs->rsp, error_code);
 586
 587                 force_sig(SIGSEGV, tsk);
 588                 return;
 589         }
 590
 591         /* kernel gp */
 592         {
 593                 const struct exception_table_entry *fixup;
 594                 fixup = search_exception_tables(regs->rip);
 595                 if (fixup) {
 596                         regs->rip = fixup->fixup;
 597                         return;
 598                 }
 599                 if (notify_die(DIE_GPF, "general protection fault", regs,
 600                                         error_code, 13, SIGSEGV) == NOTIFY_STOP)
 601                         return;
 602                 die("general protection fault", regs, error_code);
 603         }
 604 }
 605
 606 static __kprobes void
 607 mem_parity_error(unsigned char reason, struct pt_regs * regs)
 608 {
 609         printk("Uhhuh. NMI received. Dazed and confused, but trying to continue\n");
 610         printk("You probably have a hardware problem with your RAM chips\n");
 611
 612         /* Clear and disable the memory parity error line. */
 613         reason = (reason & 0xf) | 4;
 614         outb(reason, 0x61);
 615 }
 616
 617 static __kprobes void
 618 io_check_error(unsigned char reason, struct pt_regs * regs)
 619 {
 620         printk("NMI: IOCK error (debug interrupt?)\n");
 621         show_registers(regs);
 622
 623         /* Re-enable the IOCK line, wait for a few seconds */
 624         reason = (reason & 0xf) | 8;
 625         outb(reason, 0x61);
 626         mdelay(2000);
 627         reason &= ~8;
 628         outb(reason, 0x61);
 629 }
 630
 631 static __kprobes void
 632 unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
 633 {       printk("Uhhuh. NMI received for unknown reason %02x.\n", reason);
 634         printk("Dazed and confused, but trying to continue\n");
 635         printk("Do you have a strange power saving mode enabled?\n");
 636 }
 637
 638 /* Runs on IST stack. This code must keep interrupts off all the time.
 639    Nested NMIs are prevented by the CPU. */
 640 asmlinkage __kprobes void default_do_nmi(struct pt_regs *regs)
 641 {
 642         unsigned char reason = 0;
 643         int cpu;
 644
 645         cpu = smp_processor_id();
 646
 647         /* Only the BSP gets external NMIs from the system.  */
 648         if (!cpu)
 649                 reason = get_nmi_reason();
 650
 651         if (!(reason & 0xc0)) {
 652                 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
 653                                                                 == NOTIFY_STOP)
 654                         return;
 655 #ifdef CONFIG_X86_LOCAL_APIC
 656                 /*
 657                  * Ok, so this is none of the documented NMI sources,
 658                  * so it must be the NMI watchdog.
 659                  */
 660                 if (nmi_watchdog > 0) {
 661                         nmi_watchdog_tick(regs,reason);
 662                         return;
 663                 }
 664 #endif
 665                 unknown_nmi_error(reason, regs);
 666                 return;
 667         }
 668         if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
 669                 return;
 670
 671         /* AK: following checks seem to be broken on modern chipsets. FIXME */
 672
 673         if (reason & 0x80)
 674                 mem_parity_error(reason, regs);
 675         if (reason & 0x40)
 676                 io_check_error(reason, regs);
 677 }
 678
 679 /* runs on IST stack. */
 680 asmlinkage void __kprobes do_int3(struct pt_regs * regs, long error_code)
 681 {
 682         if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP) == NOTIFY_STOP) {
 683                 return;
 684         }
 685         do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
 686         return;
 687 }
 688
 689 /* Help handler running on IST stack to switch back to user stack
 690    for scheduling or signal handling. The actual stack switch is done in
 691    entry.S */
 692 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
 693 {
 694         struct pt_regs *regs = eregs;
 695         /* Did already sync */
 696         if (eregs == (struct pt_regs *)eregs->rsp)
 697                 ;
 698         /* Exception from user space */
 699         else if (user_mode(eregs))
 700                 regs = task_pt_regs(current);
 701         /* Exception from kernel and interrupts are enabled. Move to
 702            kernel process stack. */
 703         else if (eregs->eflags & X86_EFLAGS_IF)
 704                 regs = (struct pt_regs *)(eregs->rsp -= sizeof(struct pt_regs));
 705         if (eregs != regs)
 706                 *regs = *eregs;
 707         return regs;
 708 }
 709
 710 /* runs on IST stack. */
 711 asmlinkage void __kprobes do_debug(struct pt_regs * regs,
 712                                    unsigned long error_code)
 713 {
 714         unsigned long condition;
 715         struct task_struct *tsk = current;
 716         siginfo_t info;
 717
 718         get_debugreg(condition, 6);
 719
 720         if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
 721                                                 SIGTRAP) == NOTIFY_STOP)
 722                 return;
 723
 724         preempt_conditional_sti(regs);
 725
 726         /* Mask out spurious debug traps due to lazy DR7 setting */
 727         if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
 728                 if (!tsk->thread.debugreg7) {
 729                         goto clear_dr7;
 730                 }
 731         }
 732
 733         tsk->thread.debugreg6 = condition;
 734
 735         /* Mask out spurious TF errors due to lazy TF clearing */
 736         if (condition & DR_STEP) {
 737                 /*
 738                  * The TF error should be masked out only if the current
 739                  * process is not traced and if the TRAP flag has been set
 740                  * previously by a tracing process (condition detected by
 741                  * the PT_DTRACE flag); remember that the i386 TRAP flag
 742                  * can be modified by the process itself in user mode,
 743                  * allowing programs to debug themselves without the ptrace()
 744                  * interface.
 745                  */
 746                 if (!user_mode(regs))
 747                        goto clear_TF_reenable;
 748                 /*
 749                  * Was the TF flag set by a debugger? If so, clear it now,
 750                  * so that register information is correct.
 751                  */
 752                 if (tsk->ptrace & PT_DTRACE) {
 753                         regs->eflags &= ~TF_MASK;
 754                         tsk->ptrace &= ~PT_DTRACE;
 755                 }
 756         }
 757
 758         /* Ok, finally something we can handle */
 759         tsk->thread.trap_no = 1;
 760         tsk->thread.error_code = error_code;
 761         info.si_signo = SIGTRAP;
 762         info.si_errno = 0;
 763         info.si_code = TRAP_BRKPT;
 764         info.si_addr = user_mode(regs) ? (void __user *)regs->rip : NULL;
 765         force_sig_info(SIGTRAP, &info, tsk);
 766
 767 clear_dr7:
 768         set_debugreg(0UL, 7);
 769         preempt_conditional_cli(regs);
 770         return;
 771
 772 clear_TF_reenable:
 773         set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
 774         regs->eflags &= ~TF_MASK;
 775         preempt_conditional_cli(regs);
 776 }
 777
 778 static int kernel_math_error(struct pt_regs *regs, const char *str, int trapnr)
 779 {
 780         const struct exception_table_entry *fixup;
 781         fixup = search_exception_tables(regs->rip);
 782         if (fixup) {
 783                 regs->rip = fixup->fixup;
 784                 return 1;
 785         }
 786         notify_die(DIE_GPF, str, regs, 0, trapnr, SIGFPE);
 787         /* Illegal floating point operation in the kernel */
 788         current->thread.trap_no = trapnr;
 789         die(str, regs, 0);
 790         return 0;
 791 }
 792
 793 /*
 794  * Note that we play around with the 'TS' bit in an attempt to get
 795  * the correct behaviour even in the presence of the asynchronous
 796  * IRQ13 behaviour
 797  */
 798 asmlinkage void do_coprocessor_error(struct pt_regs *regs)
 799 {
 800         void __user *rip = (void __user *)(regs->rip);
 801         struct task_struct * task;
 802         siginfo_t info;
 803         unsigned short cwd, swd;
 804
 805         conditional_sti(regs);
 806         if (!user_mode(regs) &&
 807             kernel_math_error(regs, "kernel x87 math error", 16))
 808                 return;
 809
 810         /*
 811          * Save the info for the exception handler and clear the error.
 812          */
 813         task = current;
 814         save_init_fpu(task);
 815         task->thread.trap_no = 16;
 816         task->thread.error_code = 0;
 817         info.si_signo = SIGFPE;
 818         info.si_errno = 0;
 819         info.si_code = __SI_FAULT;
 820         info.si_addr = rip;
 821         /*
 822          * (~cwd & swd) will mask out exceptions that are not set to unmasked
 823          * status.  0x3f is the exception bits in these regs, 0x200 is the
 824          * C1 reg you need in case of a stack fault, 0x040 is the stack
 825          * fault bit.  We should only be taking one exception at a time,
 826          * so if this combination doesn't produce any single exception,
 827          * then we have a bad program that isn't synchronizing its FPU usage
 828          * and it will suffer the consequences since we won't be able to
 829          * fully reproduce the context of the exception
 830          */
 831         cwd = get_fpu_cwd(task);
 832         swd = get_fpu_swd(task);
 833         switch (swd & ~cwd & 0x3f) {
 834                 case 0x000:
 835                 default:
 836                         break;
 837                 case 0x001: /* Invalid Op */
 838                         /*
 839                          * swd & 0x240 == 0x040: Stack Underflow
 840                          * swd & 0x240 == 0x240: Stack Overflow
 841                          * User must clear the SF bit (0x40) if set
 842                          */
 843                         info.si_code = FPE_FLTINV;
 844                         break;
 845                 case 0x002: /* Denormalize */
 846                 case 0x010: /* Underflow */
 847                         info.si_code = FPE_FLTUND;
 848                         break;
 849                 case 0x004: /* Zero Divide */
 850                         info.si_code = FPE_FLTDIV;
 851                         break;
 852                 case 0x008: /* Overflow */
 853                         info.si_code = FPE_FLTOVF;
 854                         break;
 855                 case 0x020: /* Precision */
 856                         info.si_code = FPE_FLTRES;
 857                         break;
 858         }
 859         force_sig_info(SIGFPE, &info, task);
 860 }
 861
 862 asmlinkage void bad_intr(void)
 863 {
 864         printk("bad interrupt");
 865 }
 866
 867 asmlinkage void do_simd_coprocessor_error(struct pt_regs *regs)
 868 {
 869         void __user *rip = (void __user *)(regs->rip);
 870         struct task_struct * task;
 871         siginfo_t info;
 872         unsigned short mxcsr;
 873
 874         conditional_sti(regs);
 875         if (!user_mode(regs) &&
 876                 kernel_math_error(regs, "kernel simd math error", 19))
 877                 return;
 878
 879         /*
 880          * Save the info for the exception handler and clear the error.
 881          */
 882         task = current;
 883         save_init_fpu(task);
 884         task->thread.trap_no = 19;
 885         task->thread.error_code = 0;
 886         info.si_signo = SIGFPE;
 887         info.si_errno = 0;
 888         info.si_code = __SI_FAULT;
 889         info.si_addr = rip;
 890         /*
 891          * The SIMD FPU exceptions are handled a little differently, as there
 892          * is only a single status/control register.  Thus, to determine which
 893          * unmasked exception was caught we must mask the exception mask bits
 894          * at 0x1f80, and then use these to mask the exception bits at 0x3f.
 895          */
 896         mxcsr = get_fpu_mxcsr(task);
 897         switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
 898                 case 0x000:
 899                 default:
 900                         break;
 901                 case 0x001: /* Invalid Op */
 902                         info.si_code = FPE_FLTINV;
 903                         break;
 904                 case 0x002: /* Denormalize */
 905                 case 0x010: /* Underflow */
 906                         info.si_code = FPE_FLTUND;
 907                         break;
 908                 case 0x004: /* Zero Divide */
 909                         info.si_code = FPE_FLTDIV;
 910                         break;
 911                 case 0x008: /* Overflow */
 912                         info.si_code = FPE_FLTOVF;
 913                         break;
 914                 case 0x020: /* Precision */
 915                         info.si_code = FPE_FLTRES;
 916                         break;
 917         }
 918         force_sig_info(SIGFPE, &info, task);
 919 }
 920
 921 asmlinkage void do_spurious_interrupt_bug(struct pt_regs * regs)
 922 {
 923 }
 924
 925 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
 926 {
 927 }
 928
 929 asmlinkage void __attribute__((weak)) mce_threshold_interrupt(void)
 930 {
 931 }
 932
 933 /*
 934  *  'math_state_restore()' saves the current math information in the
 935  * old math state array, and gets the new ones from the current task
 936  *
 937  * Careful.. There are problems with IBM-designed IRQ13 behaviour.
 938  * Don't touch unless you *really* know how it works.
 939  */
 940 asmlinkage void math_state_restore(void)
 941 {
 942         struct task_struct *me = current;
 943         clts();                 /* Allow maths ops (or we recurse) */
 944
 945         if (!used_math())
 946                 init_fpu(me);
 947         restore_fpu_checking(&me->thread.i387.fxsave);
 948         task_thread_info(me)->status |= TS_USEDFPU;
 949 }
 950
 951 void __init trap_init(void)
 952 {
 953         set_intr_gate(0,&divide_error);
 954         set_intr_gate_ist(1,&debug,DEBUG_STACK);
 955         set_intr_gate_ist(2,&nmi,NMI_STACK);
 956         set_system_gate_ist(3,&int3,DEBUG_STACK); /* int3 can be called from all */
 957         set_system_gate(4,&overflow);   /* int4 can be called from all */
 958         set_intr_gate(5,&bounds);
 959         set_intr_gate(6,&invalid_op);
 960         set_intr_gate(7,&device_not_available);
 961         set_intr_gate_ist(8,&double_fault, DOUBLEFAULT_STACK);
 962         set_intr_gate(9,&coprocessor_segment_overrun);
 963         set_intr_gate(10,&invalid_TSS);
 964         set_intr_gate(11,&segment_not_present);
 965         set_intr_gate_ist(12,&stack_segment,STACKFAULT_STACK);
 966         set_intr_gate(13,&general_protection);
 967         set_intr_gate(14,&page_fault);
 968         set_intr_gate(15,&spurious_interrupt_bug);
 969         set_intr_gate(16,&coprocessor_error);
 970         set_intr_gate(17,&alignment_check);
 971 #ifdef CONFIG_X86_MCE
 972         set_intr_gate_ist(18,&machine_check, MCE_STACK);
 973 #endif
 974         set_intr_gate(19,&simd_coprocessor_error);
 975
 976 #ifdef CONFIG_IA32_EMULATION
 977         set_system_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
 978 #endif
 979
 980         /*
 981          * Should be a barrier for any external CPU state.
 982          */
 983         cpu_init();
 984 }
 985
 986
 987 /* Actual parsing is done early in setup.c. */
 988 static int __init oops_dummy(char *s)
 989 {
 990         panic_on_oops = 1;
 991         return 1;
 992 }
 993 __setup("oops=", oops_dummy);
 994
 995 static int __init kstack_setup(char *s)
 996 {
 997         kstack_depth_to_print = simple_strtoul(s,NULL,0);
 998         return 1;
 999 }
1000 __setup("kstack=", kstack_setup);
1001