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
  33 #include <asm/system.h>
  34 #include <asm/uaccess.h>
  35 #include <asm/io.h>
  36 #include <asm/atomic.h>
  37 #include <asm/debugreg.h>
  38 #include <asm/desc.h>
  39 #include <asm/i387.h>
  40 #include <asm/kdebug.h>
  41 #include <asm/processor.h>
  42
  43 #include <asm/smp.h>
  44 #include <asm/pgalloc.h>
  45 #include <asm/pda.h>
  46 #include <asm/proto.h>
  47 #include <asm/nmi.h>
  48
  49 #include <linux/irq.h>
  50
  51
  52 extern struct gate_struct idt_table[256];
  53
  54 asmlinkage void divide_error(void);
  55 asmlinkage void debug(void);
  56 asmlinkage void nmi(void);
  57 asmlinkage void int3(void);
  58 asmlinkage void overflow(void);
  59 asmlinkage void bounds(void);
  60 asmlinkage void invalid_op(void);
  61 asmlinkage void device_not_available(void);
  62 asmlinkage void double_fault(void);
  63 asmlinkage void coprocessor_segment_overrun(void);
  64 asmlinkage void invalid_TSS(void);
  65 asmlinkage void segment_not_present(void);
  66 asmlinkage void stack_segment(void);
  67 asmlinkage void general_protection(void);
  68 asmlinkage void page_fault(void);
  69 asmlinkage void coprocessor_error(void);
  70 asmlinkage void simd_coprocessor_error(void);
  71 asmlinkage void reserved(void);
  72 asmlinkage void alignment_check(void);
  73 asmlinkage void machine_check(void);
  74 asmlinkage void spurious_interrupt_bug(void);
  75 asmlinkage void call_debug(void);
  76
  77 struct notifier_block *die_chain;
  78 static DEFINE_SPINLOCK(die_notifier_lock);
  79
  80 int register_die_notifier(struct notifier_block *nb)
  81 {
  82         int err = 0;
  83         unsigned long flags;
  84         spin_lock_irqsave(&die_notifier_lock, flags);
  85         err = notifier_chain_register(&die_chain, nb);
  86         spin_unlock_irqrestore(&die_notifier_lock, flags);
  87         return err;
  88 }
  89
  90 static inline void conditional_sti(struct pt_regs *regs)
  91 {
  92         if (regs->eflags & X86_EFLAGS_IF)
  93                 local_irq_enable();
  94 }
  95
  96 static int kstack_depth_to_print = 10;
  97
  98 #ifdef CONFIG_KALLSYMS
  99 #include <linux/kallsyms.h>
 100 int printk_address(unsigned long address)
 101 {
 102         unsigned long offset = 0, symsize;
 103         const char *symname;
 104         char *modname;
 105         char *delim = ":";
 106         char namebuf[128];
 107
 108         symname = kallsyms_lookup(address, &symsize, &offset, &modname, namebuf);
 109         if (!symname)
 110                 return printk("[<%016lx>]", address);
 111         if (!modname)
 112                 modname = delim = "";
 113         return printk("<%016lx>{%s%s%s%s%+ld}",
 114                       address,delim,modname,delim,symname,offset);
 115 }
 116 #else
 117 int printk_address(unsigned long address)
 118 {
 119         return printk("[<%016lx>]", address);
 120 }
 121 #endif
 122
 123 unsigned long *in_exception_stack(int cpu, unsigned long stack)
 124 {
 125         int k;
 126         for (k = 0; k < N_EXCEPTION_STACKS; k++) {
 127                 struct tss_struct *tss = &per_cpu(init_tss, cpu);
 128                 unsigned long start = tss->ist[k] - EXCEPTION_STKSZ;
 129
 130                 if (stack >= start && stack < tss->ist[k])
 131                         return (unsigned long *)tss->ist[k];
 132         }
 133         return NULL;
 134 }
 135
 136 /*
 137  * x86-64 can have upto three kernel stacks:
 138  * process stack
 139  * interrupt stack
 140  * severe exception (double fault, nmi, stack fault) hardware stack
 141  * Check and process them in order.
 142  */
 143
 144 void show_trace(unsigned long *stack)
 145 {
 146         unsigned long addr;
 147         unsigned long *irqstack, *irqstack_end, *estack_end;
 148         const int cpu = safe_smp_processor_id();
 149         int i;
 150
 151         printk("\nCall Trace:");
 152         i = 0;
 153
 154         estack_end = in_exception_stack(cpu, (unsigned long)stack);
 155         if (estack_end) {
 156                 while (stack < estack_end) {
 157                         addr = *stack++;
 158                         if (__kernel_text_address(addr)) {
 159                                 i += printk_address(addr);
 160                                 i += printk(" ");
 161                                 if (i > 50) {
 162                                         printk("\n");
 163                                         i = 0;
 164                                 }
 165                         }
 166                 }
 167                 i += printk(" <EOE> ");
 168                 i += 7;
 169                 stack = (unsigned long *) estack_end[-2];
 170         }
 171
 172         irqstack_end = (unsigned long *) (cpu_pda[cpu].irqstackptr);
 173         irqstack = (unsigned long *) (cpu_pda[cpu].irqstackptr - IRQSTACKSIZE + 64);
 174
 175         if (stack >= irqstack && stack < irqstack_end) {
 176                 printk("<IRQ> ");
 177                 while (stack < irqstack_end) {
 178                         addr = *stack++;
 179                         /*
 180                          * If the address is either in the text segment of the
 181                          * kernel, or in the region which contains vmalloc'ed
 182                          * memory, it *may* be the address of a calling
 183                          * routine; if so, print it so that someone tracing
 184                          * down the cause of the crash will be able to figure
 185                          * out the call path that was taken.
 186                          */
 187                          if (__kernel_text_address(addr)) {
 188                                  i += printk_address(addr);
 189                                  i += printk(" ");
 190                                  if (i > 50) {
 191                                         printk("\n       ");
 192                                          i = 0;
 193                                  }
 194                         }
 195                 }
 196                 stack = (unsigned long *) (irqstack_end[-1]);
 197                 printk(" <EOI> ");
 198                 i += 7;
 199         }
 200
 201         while (((long) stack & (THREAD_SIZE-1)) != 0) {
 202                 addr = *stack++;
 203                 if (__kernel_text_address(addr)) {
 204                         i += printk_address(addr);
 205                         i += printk(" ");
 206                         if (i > 50) {
 207                                 printk("\n       ");
 208                                          i = 0;
 209                         }
 210                 }
 211         }
 212         printk("\n");
 213 }
 214
 215 void show_stack(struct task_struct *tsk, unsigned long * rsp)
 216 {
 217         unsigned long *stack;
 218         int i;
 219         const int cpu = safe_smp_processor_id();
 220         unsigned long *irqstack_end = (unsigned long *) (cpu_pda[cpu].irqstackptr);
 221         unsigned long *irqstack = (unsigned long *) (cpu_pda[cpu].irqstackptr - IRQSTACKSIZE);
 222
 223         // debugging aid: "show_stack(NULL, NULL);" prints the
 224         // back trace for this cpu.
 225
 226         if (rsp == NULL) {
 227                 if (tsk)
 228                         rsp = (unsigned long *)tsk->thread.rsp;
 229                 else
 230                         rsp = (unsigned long *)&rsp;
 231         }
 232
 233         stack = rsp;
 234         for(i=0; i < kstack_depth_to_print; i++) {
 235                 if (stack >= irqstack && stack <= irqstack_end) {
 236                         if (stack == irqstack_end) {
 237                                 stack = (unsigned long *) (irqstack_end[-1]);
 238                                 printk(" <EOI> ");
 239                         }
 240                 } else {
 241                 if (((long) stack & (THREAD_SIZE-1)) == 0)
 242                         break;
 243                 }
 244                 if (i && ((i % 4) == 0))
 245                         printk("\n       ");
 246                 printk("%016lx ", *stack++);
 247                 touch_nmi_watchdog();
 248         }
 249         show_trace((unsigned long *)rsp);
 250 }
 251
 252 /*
 253  * The architecture-independent dump_stack generator
 254  */
 255 void dump_stack(void)
 256 {
 257         unsigned long dummy;
 258         show_trace(&dummy);
 259 }
 260
 261 EXPORT_SYMBOL(dump_stack);
 262
 263 void show_registers(struct pt_regs *regs)
 264 {
 265         int i;
 266         int in_kernel = (regs->cs & 3) == 0;
 267         unsigned long rsp;
 268         const int cpu = safe_smp_processor_id();
 269         struct task_struct *cur = cpu_pda[cpu].pcurrent;
 270
 271                 rsp = regs->rsp;
 272
 273         printk("CPU %d ", cpu);
 274         __show_regs(regs);
 275         printk("Process %s (pid: %d, threadinfo %p, task %p)\n",
 276                 cur->comm, cur->pid, cur->thread_info, cur);
 277
 278         /*
 279          * When in-kernel, we also print out the stack and code at the
 280          * time of the fault..
 281          */
 282         if (in_kernel) {
 283
 284                 printk("Stack: ");
 285                 show_stack(NULL, (unsigned long*)rsp);
 286
 287                 printk("\nCode: ");
 288                 if(regs->rip < PAGE_OFFSET)
 289                         goto bad;
 290
 291                 for(i=0;i<20;i++)
 292                 {
 293                         unsigned char c;
 294                         if(__get_user(c, &((unsigned char*)regs->rip)[i])) {
 295 bad:
 296                                 printk(" Bad RIP value.");
 297                                 break;
 298                         }
 299                         printk("%02x ", c);
 300                 }
 301         }
 302         printk("\n");
 303 }
 304
 305 void handle_BUG(struct pt_regs *regs)
 306 {
 307         struct bug_frame f;
 308         char tmp;
 309
 310         if (regs->cs & 3)
 311                 return;
 312         if (__copy_from_user(&f, (struct bug_frame *) regs->rip,
 313                              sizeof(struct bug_frame)))
 314                 return;
 315         if ((unsigned long)f.filename < __PAGE_OFFSET ||
 316             f.ud2[0] != 0x0f || f.ud2[1] != 0x0b)
 317                 return;
 318         if (__get_user(tmp, f.filename))
 319                 f.filename = "unmapped filename";
 320         printk("----------- [cut here ] --------- [please bite here ] ---------\n");
 321         printk(KERN_ALERT "Kernel BUG at %.50s:%d\n", f.filename, f.line);
 322 }
 323
 324 void out_of_line_bug(void)
 325 {
 326         BUG();
 327 }
 328
 329 static DEFINE_SPINLOCK(die_lock);
 330 static int die_owner = -1;
 331
 332 void oops_begin(void)
 333 {
 334         int cpu = safe_smp_processor_id();
 335         /* racy, but better than risking deadlock. */
 336         local_irq_disable();
 337         if (!spin_trylock(&die_lock)) {
 338                 if (cpu == die_owner)
 339                         /* nested oops. should stop eventually */;
 340                 else
 341                         spin_lock(&die_lock);
 342         }
 343         die_owner = cpu;
 344         console_verbose();
 345         bust_spinlocks(1);
 346 }
 347
 348 void oops_end(void)
 349 {
 350         die_owner = -1;
 351         bust_spinlocks(0);
 352         spin_unlock(&die_lock);
 353         if (panic_on_oops)
 354                 panic("Oops");
 355 }
 356
 357 void __die(const char * str, struct pt_regs * regs, long err)
 358 {
 359         static int die_counter;
 360         printk(KERN_EMERG "%s: %04lx [%u] ", str, err & 0xffff,++die_counter);
 361 #ifdef CONFIG_PREEMPT
 362         printk("PREEMPT ");
 363 #endif
 364 #ifdef CONFIG_SMP
 365         printk("SMP ");
 366 #endif
 367 #ifdef CONFIG_DEBUG_PAGEALLOC
 368         printk("DEBUG_PAGEALLOC");
 369 #endif
 370         printk("\n");
 371         notify_die(DIE_OOPS, (char *)str, regs, err, 255, SIGSEGV);
 372         show_registers(regs);
 373         /* Executive summary in case the oops scrolled away */
 374         printk(KERN_ALERT "RIP ");
 375         printk_address(regs->rip);
 376         printk(" RSP <%016lx>\n", regs->rsp);
 377 }
 378
 379 void die(const char * str, struct pt_regs * regs, long err)
 380 {
 381         oops_begin();
 382         handle_BUG(regs);
 383         __die(str, regs, err);
 384         oops_end();
 385         do_exit(SIGSEGV);
 386 }
 387 static inline void die_if_kernel(const char * str, struct pt_regs * regs, long err)
 388 {
 389         if (!(regs->eflags & VM_MASK) && (regs->cs == __KERNEL_CS))
 390                 die(str, regs, err);
 391 }
 392
 393 void die_nmi(char *str, struct pt_regs *regs)
 394 {
 395         oops_begin();
 396         /*
 397          * We are in trouble anyway, lets at least try
 398          * to get a message out.
 399          */
 400         printk(str, safe_smp_processor_id());
 401         show_registers(regs);
 402         if (panic_on_timeout || panic_on_oops)
 403                 panic("nmi watchdog");
 404         printk("console shuts up ...\n");
 405         oops_end();
 406         do_exit(SIGSEGV);
 407 }
 408
 409 static void do_trap(int trapnr, int signr, char *str,
 410                            struct pt_regs * regs, long error_code, siginfo_t *info)
 411 {
 412         conditional_sti(regs);
 413
 414 #ifdef CONFIG_CHECKING
 415        {
 416                unsigned long gs;
 417                struct x8664_pda *pda = cpu_pda + safe_smp_processor_id();
 418                rdmsrl(MSR_GS_BASE, gs);
 419                if (gs != (unsigned long)pda) {
 420                        wrmsrl(MSR_GS_BASE, pda);
 421                        printk("%s: wrong gs %lx expected %p rip %lx\n", str, gs, pda,
 422                               regs->rip);
 423                }
 424        }
 425 #endif
 426
 427         if ((regs->cs & 3)  != 0) {
 428                 struct task_struct *tsk = current;
 429
 430                 if (exception_trace && unhandled_signal(tsk, signr))
 431                         printk(KERN_INFO
 432                                "%s[%d] trap %s rip:%lx rsp:%lx error:%lx\n",
 433                                tsk->comm, tsk->pid, str,
 434                                regs->rip,regs->rsp,error_code);
 435
 436                 tsk->thread.error_code = error_code;
 437                 tsk->thread.trap_no = trapnr;
 438                 if (info)
 439                         force_sig_info(signr, info, tsk);
 440                 else
 441                         force_sig(signr, tsk);
 442                 return;
 443         }
 444
 445
 446         /* kernel trap */
 447         {
 448                 const struct exception_table_entry *fixup;
 449                 fixup = search_exception_tables(regs->rip);
 450                 if (fixup) {
 451                         regs->rip = fixup->fixup;
 452                 } else
 453                         die(str, regs, error_code);
 454                 return;
 455         }
 456 }
 457
 458 #define DO_ERROR(trapnr, signr, str, name) \
 459 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
 460 { \
 461         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 462                                                         == NOTIFY_STOP) \
 463                 return; \
 464         do_trap(trapnr, signr, str, regs, error_code, NULL); \
 465 }
 466
 467 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
 468 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
 469 { \
 470         siginfo_t info; \
 471         info.si_signo = signr; \
 472         info.si_errno = 0; \
 473         info.si_code = sicode; \
 474         info.si_addr = (void __user *)siaddr; \
 475         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 476                                                         == NOTIFY_STOP) \
 477                 return; \
 478         do_trap(trapnr, signr, str, regs, error_code, &info); \
 479 }
 480
 481 DO_ERROR_INFO( 0, SIGFPE,  "divide error", divide_error, FPE_INTDIV, regs->rip)
 482 DO_ERROR( 4, SIGSEGV, "overflow", overflow)
 483 DO_ERROR( 5, SIGSEGV, "bounds", bounds)
 484 DO_ERROR_INFO( 6, SIGILL,  "invalid operand", invalid_op, ILL_ILLOPN, regs->rip)
 485 DO_ERROR( 7, SIGSEGV, "device not available", device_not_available)
 486 DO_ERROR( 9, SIGFPE,  "coprocessor segment overrun", coprocessor_segment_overrun)
 487 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
 488 DO_ERROR(11, SIGBUS,  "segment not present", segment_not_present)
 489 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
 490 DO_ERROR(18, SIGSEGV, "reserved", reserved)
 491
 492 #define DO_ERROR_STACK(trapnr, signr, str, name) \
 493 asmlinkage void *do_##name(struct pt_regs * regs, long error_code) \
 494 { \
 495         struct pt_regs *pr = ((struct pt_regs *)(current->thread.rsp0))-1; \
 496         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 497                                                         == NOTIFY_STOP) \
 498                 return regs; \
 499         if (regs->cs & 3) { \
 500                 memcpy(pr, regs, sizeof(struct pt_regs)); \
 501                 regs = pr; \
 502         } \
 503         do_trap(trapnr, signr, str, regs, error_code, NULL); \
 504         return regs;            \
 505 }
 506
 507 DO_ERROR_STACK(12, SIGBUS,  "stack segment", stack_segment)
 508 DO_ERROR_STACK( 8, SIGSEGV, "double fault", double_fault)
 509
 510 asmlinkage void do_general_protection(struct pt_regs * regs, long error_code)
 511 {
 512         conditional_sti(regs);
 513
 514 #ifdef CONFIG_CHECKING
 515        {
 516                unsigned long gs;
 517                struct x8664_pda *pda = cpu_pda + safe_smp_processor_id();
 518                rdmsrl(MSR_GS_BASE, gs);
 519                if (gs != (unsigned long)pda) {
 520                        wrmsrl(MSR_GS_BASE, pda);
 521                        oops_in_progress++;
 522                        printk("general protection handler: wrong gs %lx expected %p\n", gs, pda);
 523                        oops_in_progress--;
 524                }
 525        }
 526 #endif
 527
 528         if ((regs->cs & 3)!=0) {
 529                 struct task_struct *tsk = current;
 530
 531                 if (exception_trace && unhandled_signal(tsk, SIGSEGV))
 532                         printk(KERN_INFO
 533                        "%s[%d] general protection rip:%lx rsp:%lx error:%lx\n",
 534                                tsk->comm, tsk->pid,
 535                                regs->rip,regs->rsp,error_code);
 536
 537                 tsk->thread.error_code = error_code;
 538                 tsk->thread.trap_no = 13;
 539                 force_sig(SIGSEGV, tsk);
 540                 return;
 541         }
 542
 543         /* kernel gp */
 544         {
 545                 const struct exception_table_entry *fixup;
 546                 fixup = search_exception_tables(regs->rip);
 547                 if (fixup) {
 548                         regs->rip = fixup->fixup;
 549                         return;
 550                 }
 551                 if (notify_die(DIE_GPF, "general protection fault", regs,
 552                                         error_code, 13, SIGSEGV) == NOTIFY_STOP)
 553                         return;
 554                 die("general protection fault", regs, error_code);
 555         }
 556 }
 557
 558 static void mem_parity_error(unsigned char reason, struct pt_regs * regs)
 559 {
 560         printk("Uhhuh. NMI received. Dazed and confused, but trying to continue\n");
 561         printk("You probably have a hardware problem with your RAM chips\n");
 562
 563         /* Clear and disable the memory parity error line. */
 564         reason = (reason & 0xf) | 4;
 565         outb(reason, 0x61);
 566 }
 567
 568 static void io_check_error(unsigned char reason, struct pt_regs * regs)
 569 {
 570         printk("NMI: IOCK error (debug interrupt?)\n");
 571         show_registers(regs);
 572
 573         /* Re-enable the IOCK line, wait for a few seconds */
 574         reason = (reason & 0xf) | 8;
 575         outb(reason, 0x61);
 576         mdelay(2000);
 577         reason &= ~8;
 578         outb(reason, 0x61);
 579 }
 580
 581 static void unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
 582 {       printk("Uhhuh. NMI received for unknown reason %02x.\n", reason);
 583         printk("Dazed and confused, but trying to continue\n");
 584         printk("Do you have a strange power saving mode enabled?\n");
 585 }
 586
 587 asmlinkage void default_do_nmi(struct pt_regs *regs)
 588 {
 589         unsigned char reason = 0;
 590
 591         /* Only the BSP gets external NMIs from the system.  */
 592         if (!smp_processor_id())
 593                 reason = get_nmi_reason();
 594
 595         if (!(reason & 0xc0)) {
 596                 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 0, SIGINT)
 597                                                                 == NOTIFY_STOP)
 598                         return;
 599 #ifdef CONFIG_X86_LOCAL_APIC
 600                 /*
 601                  * Ok, so this is none of the documented NMI sources,
 602                  * so it must be the NMI watchdog.
 603                  */
 604                 if (nmi_watchdog > 0) {
 605                         nmi_watchdog_tick(regs,reason);
 606                         return;
 607                 }
 608 #endif
 609                 unknown_nmi_error(reason, regs);
 610                 return;
 611         }
 612         if (notify_die(DIE_NMI, "nmi", regs, reason, 0, SIGINT) == NOTIFY_STOP)
 613                 return;
 614
 615         /* AK: following checks seem to be broken on modern chipsets. FIXME */
 616
 617         if (reason & 0x80)
 618                 mem_parity_error(reason, regs);
 619         if (reason & 0x40)
 620                 io_check_error(reason, regs);
 621 }
 622
 623 asmlinkage void do_int3(struct pt_regs * regs, long error_code)
 624 {
 625         if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP) == NOTIFY_STOP) {
 626                 return;
 627         }
 628         do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
 629         return;
 630 }
 631
 632 /* runs on IST stack. */
 633 asmlinkage void *do_debug(struct pt_regs * regs, unsigned long error_code)
 634 {
 635         struct pt_regs *pr;
 636         unsigned long condition;
 637         struct task_struct *tsk = current;
 638         siginfo_t info;
 639
 640         pr = (struct pt_regs *)(current->thread.rsp0)-1;
 641         if (regs->cs & 3) {
 642                 memcpy(pr, regs, sizeof(struct pt_regs));
 643                 regs = pr;
 644         }
 645
 646 #ifdef CONFIG_CHECKING
 647        {
 648                /* RED-PEN interaction with debugger - could destroy gs */
 649                unsigned long gs;
 650                struct x8664_pda *pda = cpu_pda + safe_smp_processor_id();
 651                rdmsrl(MSR_GS_BASE, gs);
 652                if (gs != (unsigned long)pda) {
 653                        wrmsrl(MSR_GS_BASE, pda);
 654                        printk("debug handler: wrong gs %lx expected %p\n", gs, pda);
 655                }
 656        }
 657 #endif
 658
 659         asm("movq %%db6,%0" : "=r" (condition));
 660
 661         if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
 662                                                 SIGTRAP) == NOTIFY_STOP) {
 663                 return regs;
 664         }
 665         conditional_sti(regs);
 666
 667         /* Mask out spurious debug traps due to lazy DR7 setting */
 668         if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
 669                 if (!tsk->thread.debugreg7) {
 670                         goto clear_dr7;
 671                 }
 672         }
 673
 674         tsk->thread.debugreg6 = condition;
 675
 676         /* Mask out spurious TF errors due to lazy TF clearing */
 677         if ((condition & DR_STEP) &&
 678             (notify_die(DIE_DEBUGSTEP, "debugstep", regs, condition,
 679                         1, SIGTRAP) != NOTIFY_STOP)) {
 680                 /*
 681                  * The TF error should be masked out only if the current
 682                  * process is not traced and if the TRAP flag has been set
 683                  * previously by a tracing process (condition detected by
 684                  * the PT_DTRACE flag); remember that the i386 TRAP flag
 685                  * can be modified by the process itself in user mode,
 686                  * allowing programs to debug themselves without the ptrace()
 687                  * interface.
 688                  */
 689                 if ((regs->cs & 3) == 0)
 690                        goto clear_TF_reenable;
 691                 if ((tsk->ptrace & (PT_DTRACE|PT_PTRACED)) == PT_DTRACE)
 692                         goto clear_TF;
 693         }
 694
 695         /* Ok, finally something we can handle */
 696         tsk->thread.trap_no = 1;
 697         tsk->thread.error_code = error_code;
 698         info.si_signo = SIGTRAP;
 699         info.si_errno = 0;
 700         info.si_code = TRAP_BRKPT;
 701         if ((regs->cs & 3) == 0)
 702                 goto clear_dr7;
 703
 704         info.si_addr = (void __user *)regs->rip;
 705         force_sig_info(SIGTRAP, &info, tsk);
 706 clear_dr7:
 707         asm volatile("movq %0,%%db7"::"r"(0UL));
 708         notify_die(DIE_DEBUG, "debug", regs, condition, 1, SIGTRAP);
 709         return regs;
 710
 711 clear_TF_reenable:
 712         set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
 713
 714 clear_TF:
 715         /* RED-PEN could cause spurious errors */
 716         if (notify_die(DIE_DEBUG, "debug2", regs, condition, 1, SIGTRAP)
 717                                                                 != NOTIFY_STOP)
 718         regs->eflags &= ~TF_MASK;
 719         return regs;
 720 }
 721
 722 static int kernel_math_error(struct pt_regs *regs, char *str)
 723 {
 724         const struct exception_table_entry *fixup;
 725         fixup = search_exception_tables(regs->rip);
 726         if (fixup) {
 727                 regs->rip = fixup->fixup;
 728                 return 1;
 729         }
 730         notify_die(DIE_GPF, str, regs, 0, 16, SIGFPE);
 731 #if 0
 732         /* This should be a die, but warn only for now */
 733         die(str, regs, 0);
 734 #else
 735         printk(KERN_DEBUG "%s: %s at ", current->comm, str);
 736         printk_address(regs->rip);
 737         printk("\n");
 738 #endif
 739         return 0;
 740 }
 741
 742 /*
 743  * Note that we play around with the 'TS' bit in an attempt to get
 744  * the correct behaviour even in the presence of the asynchronous
 745  * IRQ13 behaviour
 746  */
 747 asmlinkage void do_coprocessor_error(struct pt_regs *regs)
 748 {
 749         void __user *rip = (void __user *)(regs->rip);
 750         struct task_struct * task;
 751         siginfo_t info;
 752         unsigned short cwd, swd;
 753
 754         conditional_sti(regs);
 755         if ((regs->cs & 3) == 0 &&
 756             kernel_math_error(regs, "kernel x87 math error"))
 757                 return;
 758
 759         /*
 760          * Save the info for the exception handler and clear the error.
 761          */
 762         task = current;
 763         save_init_fpu(task);
 764         task->thread.trap_no = 16;
 765         task->thread.error_code = 0;
 766         info.si_signo = SIGFPE;
 767         info.si_errno = 0;
 768         info.si_code = __SI_FAULT;
 769         info.si_addr = rip;
 770         /*
 771          * (~cwd & swd) will mask out exceptions that are not set to unmasked
 772          * status.  0x3f is the exception bits in these regs, 0x200 is the
 773          * C1 reg you need in case of a stack fault, 0x040 is the stack
 774          * fault bit.  We should only be taking one exception at a time,
 775          * so if this combination doesn't produce any single exception,
 776          * then we have a bad program that isn't synchronizing its FPU usage
 777          * and it will suffer the consequences since we won't be able to
 778          * fully reproduce the context of the exception
 779          */
 780         cwd = get_fpu_cwd(task);
 781         swd = get_fpu_swd(task);
 782         switch (((~cwd) & swd & 0x3f) | (swd & 0x240)) {
 783                 case 0x000:
 784                 default:
 785                         break;
 786                 case 0x001: /* Invalid Op */
 787                 case 0x041: /* Stack Fault */
 788                 case 0x241: /* Stack Fault | Direction */
 789                         info.si_code = FPE_FLTINV;
 790                         break;
 791                 case 0x002: /* Denormalize */
 792                 case 0x010: /* Underflow */
 793                         info.si_code = FPE_FLTUND;
 794                         break;
 795                 case 0x004: /* Zero Divide */
 796                         info.si_code = FPE_FLTDIV;
 797                         break;
 798                 case 0x008: /* Overflow */
 799                         info.si_code = FPE_FLTOVF;
 800                         break;
 801                 case 0x020: /* Precision */
 802                         info.si_code = FPE_FLTRES;
 803                         break;
 804         }
 805         force_sig_info(SIGFPE, &info, task);
 806 }
 807
 808 asmlinkage void bad_intr(void)
 809 {
 810         printk("bad interrupt");
 811 }
 812
 813 asmlinkage void do_simd_coprocessor_error(struct pt_regs *regs)
 814 {
 815         void __user *rip = (void __user *)(regs->rip);
 816         struct task_struct * task;
 817         siginfo_t info;
 818         unsigned short mxcsr;
 819
 820         conditional_sti(regs);
 821         if ((regs->cs & 3) == 0 &&
 822                 kernel_math_error(regs, "simd math error"))
 823                 return;
 824
 825         /*
 826          * Save the info for the exception handler and clear the error.
 827          */
 828         task = current;
 829         save_init_fpu(task);
 830         task->thread.trap_no = 19;
 831         task->thread.error_code = 0;
 832         info.si_signo = SIGFPE;
 833         info.si_errno = 0;
 834         info.si_code = __SI_FAULT;
 835         info.si_addr = rip;
 836         /*
 837          * The SIMD FPU exceptions are handled a little differently, as there
 838          * is only a single status/control register.  Thus, to determine which
 839          * unmasked exception was caught we must mask the exception mask bits
 840          * at 0x1f80, and then use these to mask the exception bits at 0x3f.
 841          */
 842         mxcsr = get_fpu_mxcsr(task);
 843         switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
 844                 case 0x000:
 845                 default:
 846                         break;
 847                 case 0x001: /* Invalid Op */
 848                         info.si_code = FPE_FLTINV;
 849                         break;
 850                 case 0x002: /* Denormalize */
 851                 case 0x010: /* Underflow */
 852                         info.si_code = FPE_FLTUND;
 853                         break;
 854                 case 0x004: /* Zero Divide */
 855                         info.si_code = FPE_FLTDIV;
 856                         break;
 857                 case 0x008: /* Overflow */
 858                         info.si_code = FPE_FLTOVF;
 859                         break;
 860                 case 0x020: /* Precision */
 861                         info.si_code = FPE_FLTRES;
 862                         break;
 863         }
 864         force_sig_info(SIGFPE, &info, task);
 865 }
 866
 867 asmlinkage void do_spurious_interrupt_bug(struct pt_regs * regs)
 868 {
 869 }
 870
 871 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
 872 {
 873 }
 874
 875 /*
 876  *  'math_state_restore()' saves the current math information in the
 877  * old math state array, and gets the new ones from the current task
 878  *
 879  * Careful.. There are problems with IBM-designed IRQ13 behaviour.
 880  * Don't touch unless you *really* know how it works.
 881  */
 882 asmlinkage void math_state_restore(void)
 883 {
 884         struct task_struct *me = current;
 885         clts();                 /* Allow maths ops (or we recurse) */
 886
 887         if (!used_math())
 888                 init_fpu(me);
 889         restore_fpu_checking(&me->thread.i387.fxsave);
 890         me->thread_info->status |= TS_USEDFPU;
 891 }
 892
 893 void do_call_debug(struct pt_regs *regs)
 894 {
 895         notify_die(DIE_CALL, "debug call", regs, 0, 255, SIGINT);
 896 }
 897
 898 void __init trap_init(void)
 899 {
 900         set_intr_gate(0,&divide_error);
 901         set_intr_gate_ist(1,&debug,DEBUG_STACK);
 902         set_intr_gate_ist(2,&nmi,NMI_STACK);
 903         set_system_gate(3,&int3);
 904         set_system_gate(4,&overflow);   /* int4-5 can be called from all */
 905         set_system_gate(5,&bounds);
 906         set_intr_gate(6,&invalid_op);
 907         set_intr_gate(7,&device_not_available);
 908         set_intr_gate_ist(8,&double_fault, DOUBLEFAULT_STACK);
 909         set_intr_gate(9,&coprocessor_segment_overrun);
 910         set_intr_gate(10,&invalid_TSS);
 911         set_intr_gate(11,&segment_not_present);
 912         set_intr_gate_ist(12,&stack_segment,STACKFAULT_STACK);
 913         set_intr_gate(13,&general_protection);
 914         set_intr_gate(14,&page_fault);
 915         set_intr_gate(15,&spurious_interrupt_bug);
 916         set_intr_gate(16,&coprocessor_error);
 917         set_intr_gate(17,&alignment_check);
 918 #ifdef CONFIG_X86_MCE
 919         set_intr_gate_ist(18,&machine_check, MCE_STACK);
 920 #endif
 921         set_intr_gate(19,&simd_coprocessor_error);
 922
 923 #ifdef CONFIG_IA32_EMULATION
 924         set_system_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
 925 #endif
 926
 927         set_intr_gate(KDB_VECTOR, call_debug);
 928
 929         /*
 930          * Should be a barrier for any external CPU state.
 931          */
 932         cpu_init();
 933 }
 934
 935
 936 /* Actual parsing is done early in setup.c. */
 937 static int __init oops_dummy(char *s)
 938 {
 939         panic_on_oops = 1;
 940         return -1;
 941 }
 942 __setup("oops=", oops_dummy);
 943
 944 static int __init kstack_setup(char *s)
 945 {
 946         kstack_depth_to_print = simple_strtoul(s,NULL,0);
 947         return 0;
 948 }
 949 __setup("kstack=", kstack_setup);
 950