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 static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
 124                                         unsigned *usedp, const char **idp)
 125 {
 126         static const char ids[N_EXCEPTION_STACKS][8] = {
 127                 [DEBUG_STACK - 1] = "#DB",
 128                 [NMI_STACK - 1] = "NMI",
 129                 [DOUBLEFAULT_STACK - 1] = "#DF",
 130                 [STACKFAULT_STACK - 1] = "#SS",
 131                 [MCE_STACK - 1] = "#MC",
 132         };
 133         unsigned k;
 134
 135         for (k = 0; k < N_EXCEPTION_STACKS; k++) {
 136                 unsigned long end;
 137
 138                 end = per_cpu(init_tss, cpu).ist[k];
 139                 if (stack >= end)
 140                         continue;
 141                 if (stack >= end - EXCEPTION_STKSZ) {
 142                         if (*usedp & (1U << k))
 143                                 break;
 144                         *usedp |= 1U << k;
 145                         *idp = ids[k];
 146                         return (unsigned long *)end;
 147                 }
 148         }
 149         return NULL;
 150 }
 151
 152 /*
 153  * x86-64 can have upto three kernel stacks:
 154  * process stack
 155  * interrupt stack
 156  * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
 157  */
 158
 159 void show_trace(unsigned long *stack)
 160 {
 161         unsigned long addr;
 162         const unsigned cpu = safe_smp_processor_id();
 163         unsigned long *irqstack_end = (unsigned long *)cpu_pda[cpu].irqstackptr;
 164         int i;
 165         unsigned used = 0;
 166
 167         printk("\nCall Trace:");
 168
 169 #define HANDLE_STACK(cond) \
 170         do while (cond) { \
 171                 addr = *stack++; \
 172                 if (kernel_text_address(addr)) { \
 173                         /* \
 174                          * If the address is either in the text segment of the \
 175                          * kernel, or in the region which contains vmalloc'ed \
 176                          * memory, it *may* be the address of a calling \
 177                          * routine; if so, print it so that someone tracing \
 178                          * down the cause of the crash will be able to figure \
 179                          * out the call path that was taken. \
 180                          */ \
 181                         i += printk_address(addr); \
 182                         if (i > 50) { \
 183                                 printk("\n       "); \
 184                                 i = 0; \
 185                         } \
 186                         else \
 187                                 i += printk(" "); \
 188                 } \
 189         } while (0)
 190
 191         for(i = 0; ; ) {
 192                 const char *id;
 193                 unsigned long *estack_end;
 194                 estack_end = in_exception_stack(cpu, (unsigned long)stack,
 195                                                 &used, &id);
 196
 197                 if (estack_end) {
 198                         i += printk(" <%s> ", id);
 199                         HANDLE_STACK (stack < estack_end);
 200                         i += printk(" <EOE> ");
 201                         stack = (unsigned long *) estack_end[-2];
 202                         continue;
 203                 }
 204                 if (irqstack_end) {
 205                         unsigned long *irqstack;
 206                         irqstack = irqstack_end -
 207                                 (IRQSTACKSIZE - 64) / sizeof(*irqstack);
 208
 209                         if (stack >= irqstack && stack < irqstack_end) {
 210                                 i += printk(" <IRQ> ");
 211                                 HANDLE_STACK (stack < irqstack_end);
 212                                 stack = (unsigned long *) (irqstack_end[-1]);
 213                                 irqstack_end = NULL;
 214                                 i += printk(" <EOI> ");
 215                                 continue;
 216                         }
 217                 }
 218                 break;
 219         }
 220
 221         HANDLE_STACK (((long) stack & (THREAD_SIZE-1)) != 0);
 222 #undef HANDLE_STACK
 223         printk("\n");
 224 }
 225
 226 void show_stack(struct task_struct *tsk, unsigned long * rsp)
 227 {
 228         unsigned long *stack;
 229         int i;
 230         const int cpu = safe_smp_processor_id();
 231         unsigned long *irqstack_end = (unsigned long *) (cpu_pda[cpu].irqstackptr);
 232         unsigned long *irqstack = (unsigned long *) (cpu_pda[cpu].irqstackptr - IRQSTACKSIZE);
 233
 234         // debugging aid: "show_stack(NULL, NULL);" prints the
 235         // back trace for this cpu.
 236
 237         if (rsp == NULL) {
 238                 if (tsk)
 239                         rsp = (unsigned long *)tsk->thread.rsp;
 240                 else
 241                         rsp = (unsigned long *)&rsp;
 242         }
 243
 244         stack = rsp;
 245         for(i=0; i < kstack_depth_to_print; i++) {
 246                 if (stack >= irqstack && stack <= irqstack_end) {
 247                         if (stack == irqstack_end) {
 248                                 stack = (unsigned long *) (irqstack_end[-1]);
 249                                 printk(" <EOI> ");
 250                         }
 251                 } else {
 252                 if (((long) stack & (THREAD_SIZE-1)) == 0)
 253                         break;
 254                 }
 255                 if (i && ((i % 4) == 0))
 256                         printk("\n       ");
 257                 printk("%016lx ", *stack++);
 258                 touch_nmi_watchdog();
 259         }
 260         show_trace((unsigned long *)rsp);
 261 }
 262
 263 /*
 264  * The architecture-independent dump_stack generator
 265  */
 266 void dump_stack(void)
 267 {
 268         unsigned long dummy;
 269         show_trace(&dummy);
 270 }
 271
 272 EXPORT_SYMBOL(dump_stack);
 273
 274 void show_registers(struct pt_regs *regs)
 275 {
 276         int i;
 277         int in_kernel = !user_mode(regs);
 278         unsigned long rsp;
 279         const int cpu = safe_smp_processor_id();
 280         struct task_struct *cur = cpu_pda[cpu].pcurrent;
 281
 282                 rsp = regs->rsp;
 283
 284         printk("CPU %d ", cpu);
 285         __show_regs(regs);
 286         printk("Process %s (pid: %d, threadinfo %p, task %p)\n",
 287                 cur->comm, cur->pid, cur->thread_info, cur);
 288
 289         /*
 290          * When in-kernel, we also print out the stack and code at the
 291          * time of the fault..
 292          */
 293         if (in_kernel) {
 294
 295                 printk("Stack: ");
 296                 show_stack(NULL, (unsigned long*)rsp);
 297
 298                 printk("\nCode: ");
 299                 if(regs->rip < PAGE_OFFSET)
 300                         goto bad;
 301
 302                 for(i=0;i<20;i++)
 303                 {
 304                         unsigned char c;
 305                         if(__get_user(c, &((unsigned char*)regs->rip)[i])) {
 306 bad:
 307                                 printk(" Bad RIP value.");
 308                                 break;
 309                         }
 310                         printk("%02x ", c);
 311                 }
 312         }
 313         printk("\n");
 314 }
 315
 316 void handle_BUG(struct pt_regs *regs)
 317 {
 318         struct bug_frame f;
 319         char tmp;
 320
 321         if (user_mode(regs))
 322                 return;
 323         if (__copy_from_user(&f, (struct bug_frame *) regs->rip,
 324                              sizeof(struct bug_frame)))
 325                 return;
 326         if ((unsigned long)f.filename < __PAGE_OFFSET ||
 327             f.ud2[0] != 0x0f || f.ud2[1] != 0x0b)
 328                 return;
 329         if (__get_user(tmp, f.filename))
 330                 f.filename = "unmapped filename";
 331         printk("----------- [cut here ] --------- [please bite here ] ---------\n");
 332         printk(KERN_ALERT "Kernel BUG at %.50s:%d\n", f.filename, f.line);
 333 }
 334
 335 #ifdef CONFIG_BUG
 336 void out_of_line_bug(void)
 337 {
 338         BUG();
 339 }
 340 #endif
 341
 342 static DEFINE_SPINLOCK(die_lock);
 343 static int die_owner = -1;
 344
 345 void oops_begin(void)
 346 {
 347         int cpu = safe_smp_processor_id();
 348         /* racy, but better than risking deadlock. */
 349         local_irq_disable();
 350         if (!spin_trylock(&die_lock)) {
 351                 if (cpu == die_owner)
 352                         /* nested oops. should stop eventually */;
 353                 else
 354                         spin_lock(&die_lock);
 355         }
 356         die_owner = cpu;
 357         console_verbose();
 358         bust_spinlocks(1);
 359 }
 360
 361 void oops_end(void)
 362 {
 363         die_owner = -1;
 364         bust_spinlocks(0);
 365         spin_unlock(&die_lock);
 366         if (panic_on_oops)
 367                 panic("Oops");
 368 }
 369
 370 void __die(const char * str, struct pt_regs * regs, long err)
 371 {
 372         static int die_counter;
 373         printk(KERN_EMERG "%s: %04lx [%u] ", str, err & 0xffff,++die_counter);
 374 #ifdef CONFIG_PREEMPT
 375         printk("PREEMPT ");
 376 #endif
 377 #ifdef CONFIG_SMP
 378         printk("SMP ");
 379 #endif
 380 #ifdef CONFIG_DEBUG_PAGEALLOC
 381         printk("DEBUG_PAGEALLOC");
 382 #endif
 383         printk("\n");
 384         notify_die(DIE_OOPS, (char *)str, regs, err, 255, SIGSEGV);
 385         show_registers(regs);
 386         /* Executive summary in case the oops scrolled away */
 387         printk(KERN_ALERT "RIP ");
 388         printk_address(regs->rip);
 389         printk(" RSP <%016lx>\n", regs->rsp);
 390 }
 391
 392 void die(const char * str, struct pt_regs * regs, long err)
 393 {
 394         oops_begin();
 395         handle_BUG(regs);
 396         __die(str, regs, err);
 397         oops_end();
 398         do_exit(SIGSEGV);
 399 }
 400 static inline void die_if_kernel(const char * str, struct pt_regs * regs, long err)
 401 {
 402         if (!(regs->eflags & VM_MASK) && (regs->cs == __KERNEL_CS))
 403                 die(str, regs, err);
 404 }
 405
 406 void die_nmi(char *str, struct pt_regs *regs)
 407 {
 408         oops_begin();
 409         /*
 410          * We are in trouble anyway, lets at least try
 411          * to get a message out.
 412          */
 413         printk(str, safe_smp_processor_id());
 414         show_registers(regs);
 415         if (panic_on_timeout || panic_on_oops)
 416                 panic("nmi watchdog");
 417         printk("console shuts up ...\n");
 418         oops_end();
 419         do_exit(SIGSEGV);
 420 }
 421
 422 static void do_trap(int trapnr, int signr, char *str,
 423                            struct pt_regs * regs, long error_code, siginfo_t *info)
 424 {
 425         conditional_sti(regs);
 426
 427 #ifdef CONFIG_CHECKING
 428        {
 429                unsigned long gs;
 430                struct x8664_pda *pda = cpu_pda + safe_smp_processor_id();
 431                rdmsrl(MSR_GS_BASE, gs);
 432                if (gs != (unsigned long)pda) {
 433                        wrmsrl(MSR_GS_BASE, pda);
 434                        printk("%s: wrong gs %lx expected %p rip %lx\n", str, gs, pda,
 435                               regs->rip);
 436                }
 437        }
 438 #endif
 439
 440         if (user_mode(regs)) {
 441                 struct task_struct *tsk = current;
 442
 443                 if (exception_trace && unhandled_signal(tsk, signr))
 444                         printk(KERN_INFO
 445                                "%s[%d] trap %s rip:%lx rsp:%lx error:%lx\n",
 446                                tsk->comm, tsk->pid, str,
 447                                regs->rip,regs->rsp,error_code);
 448
 449                 tsk->thread.error_code = error_code;
 450                 tsk->thread.trap_no = trapnr;
 451                 if (info)
 452                         force_sig_info(signr, info, tsk);
 453                 else
 454                         force_sig(signr, tsk);
 455                 return;
 456         }
 457
 458
 459         /* kernel trap */
 460         {
 461                 const struct exception_table_entry *fixup;
 462                 fixup = search_exception_tables(regs->rip);
 463                 if (fixup) {
 464                         regs->rip = fixup->fixup;
 465                 } else
 466                         die(str, regs, error_code);
 467                 return;
 468         }
 469 }
 470
 471 #define DO_ERROR(trapnr, signr, str, name) \
 472 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
 473 { \
 474         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 475                                                         == NOTIFY_STOP) \
 476                 return; \
 477         do_trap(trapnr, signr, str, regs, error_code, NULL); \
 478 }
 479
 480 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
 481 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
 482 { \
 483         siginfo_t info; \
 484         info.si_signo = signr; \
 485         info.si_errno = 0; \
 486         info.si_code = sicode; \
 487         info.si_addr = (void __user *)siaddr; \
 488         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 489                                                         == NOTIFY_STOP) \
 490                 return; \
 491         do_trap(trapnr, signr, str, regs, error_code, &info); \
 492 }
 493
 494 DO_ERROR_INFO( 0, SIGFPE,  "divide error", divide_error, FPE_INTDIV, regs->rip)
 495 DO_ERROR( 4, SIGSEGV, "overflow", overflow)
 496 DO_ERROR( 5, SIGSEGV, "bounds", bounds)
 497 DO_ERROR_INFO( 6, SIGILL,  "invalid operand", invalid_op, ILL_ILLOPN, regs->rip)
 498 DO_ERROR( 7, SIGSEGV, "device not available", device_not_available)
 499 DO_ERROR( 9, SIGFPE,  "coprocessor segment overrun", coprocessor_segment_overrun)
 500 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
 501 DO_ERROR(11, SIGBUS,  "segment not present", segment_not_present)
 502 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
 503 DO_ERROR(18, SIGSEGV, "reserved", reserved)
 504 DO_ERROR(12, SIGBUS,  "stack segment", stack_segment)
 505 DO_ERROR( 8, SIGSEGV, "double fault", double_fault)
 506
 507 asmlinkage void do_general_protection(struct pt_regs * regs, long error_code)
 508 {
 509         conditional_sti(regs);
 510
 511 #ifdef CONFIG_CHECKING
 512        {
 513                unsigned long gs;
 514                struct x8664_pda *pda = cpu_pda + safe_smp_processor_id();
 515                rdmsrl(MSR_GS_BASE, gs);
 516                if (gs != (unsigned long)pda) {
 517                        wrmsrl(MSR_GS_BASE, pda);
 518                        oops_in_progress++;
 519                        printk("general protection handler: wrong gs %lx expected %p\n", gs, pda);
 520                        oops_in_progress--;
 521                }
 522        }
 523 #endif
 524
 525         if (user_mode(regs)) {
 526                 struct task_struct *tsk = current;
 527
 528                 if (exception_trace && unhandled_signal(tsk, SIGSEGV))
 529                         printk(KERN_INFO
 530                        "%s[%d] general protection rip:%lx rsp:%lx error:%lx\n",
 531                                tsk->comm, tsk->pid,
 532                                regs->rip,regs->rsp,error_code);
 533
 534                 tsk->thread.error_code = error_code;
 535                 tsk->thread.trap_no = 13;
 536                 force_sig(SIGSEGV, tsk);
 537                 return;
 538         }
 539
 540         /* kernel gp */
 541         {
 542                 const struct exception_table_entry *fixup;
 543                 fixup = search_exception_tables(regs->rip);
 544                 if (fixup) {
 545                         regs->rip = fixup->fixup;
 546                         return;
 547                 }
 548                 if (notify_die(DIE_GPF, "general protection fault", regs,
 549                                         error_code, 13, SIGSEGV) == NOTIFY_STOP)
 550                         return;
 551                 die("general protection fault", regs, error_code);
 552         }
 553 }
 554
 555 static void mem_parity_error(unsigned char reason, struct pt_regs * regs)
 556 {
 557         printk("Uhhuh. NMI received. Dazed and confused, but trying to continue\n");
 558         printk("You probably have a hardware problem with your RAM chips\n");
 559
 560         /* Clear and disable the memory parity error line. */
 561         reason = (reason & 0xf) | 4;
 562         outb(reason, 0x61);
 563 }
 564
 565 static void io_check_error(unsigned char reason, struct pt_regs * regs)
 566 {
 567         printk("NMI: IOCK error (debug interrupt?)\n");
 568         show_registers(regs);
 569
 570         /* Re-enable the IOCK line, wait for a few seconds */
 571         reason = (reason & 0xf) | 8;
 572         outb(reason, 0x61);
 573         mdelay(2000);
 574         reason &= ~8;
 575         outb(reason, 0x61);
 576 }
 577
 578 static void unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
 579 {       printk("Uhhuh. NMI received for unknown reason %02x.\n", reason);
 580         printk("Dazed and confused, but trying to continue\n");
 581         printk("Do you have a strange power saving mode enabled?\n");
 582 }
 583
 584 /* Runs on IST stack. This code must keep interrupts off all the time.
 585    Nested NMIs are prevented by the CPU. */
 586 asmlinkage void default_do_nmi(struct pt_regs *regs)
 587 {
 588         unsigned char reason = 0;
 589         int cpu;
 590
 591         cpu = smp_processor_id();
 592
 593         /* Only the BSP gets external NMIs from the system.  */
 594         if (!cpu)
 595                 reason = get_nmi_reason();
 596
 597         if (!cpu_online(cpu))
 598                 return;
 599
 600         if (!(reason & 0xc0)) {
 601                 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 0, SIGINT)
 602                                                                 == NOTIFY_STOP)
 603                         return;
 604 #ifdef CONFIG_X86_LOCAL_APIC
 605                 /*
 606                  * Ok, so this is none of the documented NMI sources,
 607                  * so it must be the NMI watchdog.
 608                  */
 609                 if (nmi_watchdog > 0) {
 610                         nmi_watchdog_tick(regs,reason);
 611                         return;
 612                 }
 613 #endif
 614                 unknown_nmi_error(reason, regs);
 615                 return;
 616         }
 617         if (notify_die(DIE_NMI, "nmi", regs, reason, 0, SIGINT) == NOTIFY_STOP)
 618                 return;
 619
 620         /* AK: following checks seem to be broken on modern chipsets. FIXME */
 621
 622         if (reason & 0x80)
 623                 mem_parity_error(reason, regs);
 624         if (reason & 0x40)
 625                 io_check_error(reason, regs);
 626 }
 627
 628 asmlinkage void do_int3(struct pt_regs * regs, long error_code)
 629 {
 630         if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP) == NOTIFY_STOP) {
 631                 return;
 632         }
 633         do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
 634         return;
 635 }
 636
 637 /* Help handler running on IST stack to switch back to user stack
 638    for scheduling or signal handling. The actual stack switch is done in
 639    entry.S */
 640 asmlinkage struct pt_regs *sync_regs(struct pt_regs *eregs)
 641 {
 642         struct pt_regs *regs = eregs;
 643         /* Did already sync */
 644         if (eregs == (struct pt_regs *)eregs->rsp)
 645                 ;
 646         /* Exception from user space */
 647         else if (user_mode(eregs))
 648                 regs = ((struct pt_regs *)current->thread.rsp0) - 1;
 649         /* Exception from kernel and interrupts are enabled. Move to
 650            kernel process stack. */
 651         else if (eregs->eflags & X86_EFLAGS_IF)
 652                 regs = (struct pt_regs *)(eregs->rsp -= sizeof(struct pt_regs));
 653         if (eregs != regs)
 654                 *regs = *eregs;
 655         return regs;
 656 }
 657
 658 /* runs on IST stack. */
 659 asmlinkage void do_debug(struct pt_regs * regs, unsigned long error_code)
 660 {
 661         unsigned long condition;
 662         struct task_struct *tsk = current;
 663         siginfo_t info;
 664
 665 #ifdef CONFIG_CHECKING
 666        {
 667                /* RED-PEN interaction with debugger - could destroy gs */
 668                unsigned long gs;
 669                struct x8664_pda *pda = cpu_pda + safe_smp_processor_id();
 670                rdmsrl(MSR_GS_BASE, gs);
 671                if (gs != (unsigned long)pda) {
 672                        wrmsrl(MSR_GS_BASE, pda);
 673                        printk("debug handler: wrong gs %lx expected %p\n", gs, pda);
 674                }
 675        }
 676 #endif
 677
 678         get_debugreg(condition, 6);
 679
 680         if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
 681                                                 SIGTRAP) == NOTIFY_STOP)
 682                 return;
 683
 684         conditional_sti(regs);
 685
 686         /* Mask out spurious debug traps due to lazy DR7 setting */
 687         if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
 688                 if (!tsk->thread.debugreg7) {
 689                         goto clear_dr7;
 690                 }
 691         }
 692
 693         tsk->thread.debugreg6 = condition;
 694
 695         /* Mask out spurious TF errors due to lazy TF clearing */
 696         if (condition & DR_STEP) {
 697                 /*
 698                  * The TF error should be masked out only if the current
 699                  * process is not traced and if the TRAP flag has been set
 700                  * previously by a tracing process (condition detected by
 701                  * the PT_DTRACE flag); remember that the i386 TRAP flag
 702                  * can be modified by the process itself in user mode,
 703                  * allowing programs to debug themselves without the ptrace()
 704                  * interface.
 705                  */
 706                 if (!user_mode(regs))
 707                        goto clear_TF_reenable;
 708                 /*
 709                  * Was the TF flag set by a debugger? If so, clear it now,
 710                  * so that register information is correct.
 711                  */
 712                 if (tsk->ptrace & PT_DTRACE) {
 713                         regs->eflags &= ~TF_MASK;
 714                         tsk->ptrace &= ~PT_DTRACE;
 715                 }
 716         }
 717
 718         /* Ok, finally something we can handle */
 719         tsk->thread.trap_no = 1;
 720         tsk->thread.error_code = error_code;
 721         info.si_signo = SIGTRAP;
 722         info.si_errno = 0;
 723         info.si_code = TRAP_BRKPT;
 724         if (!user_mode(regs))
 725                 goto clear_dr7;
 726
 727         info.si_addr = (void __user *)regs->rip;
 728         force_sig_info(SIGTRAP, &info, tsk);
 729 clear_dr7:
 730         set_debugreg(0UL, 7);
 731         return;
 732
 733 clear_TF_reenable:
 734         set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
 735         regs->eflags &= ~TF_MASK;
 736 }
 737
 738 static int kernel_math_error(struct pt_regs *regs, char *str)
 739 {
 740         const struct exception_table_entry *fixup;
 741         fixup = search_exception_tables(regs->rip);
 742         if (fixup) {
 743                 regs->rip = fixup->fixup;
 744                 return 1;
 745         }
 746         notify_die(DIE_GPF, str, regs, 0, 16, SIGFPE);
 747         /* Illegal floating point operation in the kernel */
 748         die(str, regs, 0);
 749         return 0;
 750 }
 751
 752 /*
 753  * Note that we play around with the 'TS' bit in an attempt to get
 754  * the correct behaviour even in the presence of the asynchronous
 755  * IRQ13 behaviour
 756  */
 757 asmlinkage void do_coprocessor_error(struct pt_regs *regs)
 758 {
 759         void __user *rip = (void __user *)(regs->rip);
 760         struct task_struct * task;
 761         siginfo_t info;
 762         unsigned short cwd, swd;
 763
 764         conditional_sti(regs);
 765         if (!user_mode(regs) &&
 766             kernel_math_error(regs, "kernel x87 math error"))
 767                 return;
 768
 769         /*
 770          * Save the info for the exception handler and clear the error.
 771          */
 772         task = current;
 773         save_init_fpu(task);
 774         task->thread.trap_no = 16;
 775         task->thread.error_code = 0;
 776         info.si_signo = SIGFPE;
 777         info.si_errno = 0;
 778         info.si_code = __SI_FAULT;
 779         info.si_addr = rip;
 780         /*
 781          * (~cwd & swd) will mask out exceptions that are not set to unmasked
 782          * status.  0x3f is the exception bits in these regs, 0x200 is the
 783          * C1 reg you need in case of a stack fault, 0x040 is the stack
 784          * fault bit.  We should only be taking one exception at a time,
 785          * so if this combination doesn't produce any single exception,
 786          * then we have a bad program that isn't synchronizing its FPU usage
 787          * and it will suffer the consequences since we won't be able to
 788          * fully reproduce the context of the exception
 789          */
 790         cwd = get_fpu_cwd(task);
 791         swd = get_fpu_swd(task);
 792         switch (((~cwd) & swd & 0x3f) | (swd & 0x240)) {
 793                 case 0x000:
 794                 default:
 795                         break;
 796                 case 0x001: /* Invalid Op */
 797                 case 0x041: /* Stack Fault */
 798                 case 0x241: /* Stack Fault | Direction */
 799                         info.si_code = FPE_FLTINV;
 800                         break;
 801                 case 0x002: /* Denormalize */
 802                 case 0x010: /* Underflow */
 803                         info.si_code = FPE_FLTUND;
 804                         break;
 805                 case 0x004: /* Zero Divide */
 806                         info.si_code = FPE_FLTDIV;
 807                         break;
 808                 case 0x008: /* Overflow */
 809                         info.si_code = FPE_FLTOVF;
 810                         break;
 811                 case 0x020: /* Precision */
 812                         info.si_code = FPE_FLTRES;
 813                         break;
 814         }
 815         force_sig_info(SIGFPE, &info, task);
 816 }
 817
 818 asmlinkage void bad_intr(void)
 819 {
 820         printk("bad interrupt");
 821 }
 822
 823 asmlinkage void do_simd_coprocessor_error(struct pt_regs *regs)
 824 {
 825         void __user *rip = (void __user *)(regs->rip);
 826         struct task_struct * task;
 827         siginfo_t info;
 828         unsigned short mxcsr;
 829
 830         conditional_sti(regs);
 831         if (!user_mode(regs) &&
 832                 kernel_math_error(regs, "kernel simd math error"))
 833                 return;
 834
 835         /*
 836          * Save the info for the exception handler and clear the error.
 837          */
 838         task = current;
 839         save_init_fpu(task);
 840         task->thread.trap_no = 19;
 841         task->thread.error_code = 0;
 842         info.si_signo = SIGFPE;
 843         info.si_errno = 0;
 844         info.si_code = __SI_FAULT;
 845         info.si_addr = rip;
 846         /*
 847          * The SIMD FPU exceptions are handled a little differently, as there
 848          * is only a single status/control register.  Thus, to determine which
 849          * unmasked exception was caught we must mask the exception mask bits
 850          * at 0x1f80, and then use these to mask the exception bits at 0x3f.
 851          */
 852         mxcsr = get_fpu_mxcsr(task);
 853         switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
 854                 case 0x000:
 855                 default:
 856                         break;
 857                 case 0x001: /* Invalid Op */
 858                         info.si_code = FPE_FLTINV;
 859                         break;
 860                 case 0x002: /* Denormalize */
 861                 case 0x010: /* Underflow */
 862                         info.si_code = FPE_FLTUND;
 863                         break;
 864                 case 0x004: /* Zero Divide */
 865                         info.si_code = FPE_FLTDIV;
 866                         break;
 867                 case 0x008: /* Overflow */
 868                         info.si_code = FPE_FLTOVF;
 869                         break;
 870                 case 0x020: /* Precision */
 871                         info.si_code = FPE_FLTRES;
 872                         break;
 873         }
 874         force_sig_info(SIGFPE, &info, task);
 875 }
 876
 877 asmlinkage void do_spurious_interrupt_bug(struct pt_regs * regs)
 878 {
 879 }
 880
 881 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
 882 {
 883 }
 884
 885 /*
 886  *  'math_state_restore()' saves the current math information in the
 887  * old math state array, and gets the new ones from the current task
 888  *
 889  * Careful.. There are problems with IBM-designed IRQ13 behaviour.
 890  * Don't touch unless you *really* know how it works.
 891  */
 892 asmlinkage void math_state_restore(void)
 893 {
 894         struct task_struct *me = current;
 895         clts();                 /* Allow maths ops (or we recurse) */
 896
 897         if (!used_math())
 898                 init_fpu(me);
 899         restore_fpu_checking(&me->thread.i387.fxsave);
 900         me->thread_info->status |= TS_USEDFPU;
 901 }
 902
 903 void do_call_debug(struct pt_regs *regs)
 904 {
 905         notify_die(DIE_CALL, "debug call", regs, 0, 255, SIGINT);
 906 }
 907
 908 void __init trap_init(void)
 909 {
 910         set_intr_gate(0,&divide_error);
 911         set_intr_gate_ist(1,&debug,DEBUG_STACK);
 912         set_intr_gate_ist(2,&nmi,NMI_STACK);
 913         set_system_gate(3,&int3);
 914         set_system_gate(4,&overflow);   /* int4-5 can be called from all */
 915         set_system_gate(5,&bounds);
 916         set_intr_gate(6,&invalid_op);
 917         set_intr_gate(7,&device_not_available);
 918         set_intr_gate_ist(8,&double_fault, DOUBLEFAULT_STACK);
 919         set_intr_gate(9,&coprocessor_segment_overrun);
 920         set_intr_gate(10,&invalid_TSS);
 921         set_intr_gate(11,&segment_not_present);
 922         set_intr_gate_ist(12,&stack_segment,STACKFAULT_STACK);
 923         set_intr_gate(13,&general_protection);
 924         set_intr_gate(14,&page_fault);
 925         set_intr_gate(15,&spurious_interrupt_bug);
 926         set_intr_gate(16,&coprocessor_error);
 927         set_intr_gate(17,&alignment_check);
 928 #ifdef CONFIG_X86_MCE
 929         set_intr_gate_ist(18,&machine_check, MCE_STACK);
 930 #endif
 931         set_intr_gate(19,&simd_coprocessor_error);
 932
 933 #ifdef CONFIG_IA32_EMULATION
 934         set_system_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
 935 #endif
 936
 937         set_intr_gate(KDB_VECTOR, call_debug);
 938
 939         /*
 940          * Should be a barrier for any external CPU state.
 941          */
 942         cpu_init();
 943 }
 944
 945
 946 /* Actual parsing is done early in setup.c. */
 947 static int __init oops_dummy(char *s)
 948 {
 949         panic_on_oops = 1;
 950         return -1;
 951 }
 952 __setup("oops=", oops_dummy);
 953
 954 static int __init kstack_setup(char *s)
 955 {
 956         kstack_depth_to_print = simple_strtoul(s,NULL,0);
 957         return 0;
 958 }
 959 __setup("kstack=", kstack_setup);
 960