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
  34 #include <asm/system.h>
  35 #include <asm/uaccess.h>
  36 #include <asm/io.h>
  37 #include <asm/atomic.h>
  38 #include <asm/debugreg.h>
  39 #include <asm/desc.h>
  40 #include <asm/i387.h>
  41 #include <asm/kdebug.h>
  42 #include <asm/processor.h>
  43
  44 #include <asm/smp.h>
  45 #include <asm/pgalloc.h>
  46 #include <asm/pda.h>
  47 #include <asm/proto.h>
  48 #include <asm/nmi.h>
  49
  50 #include <linux/irq.h>
  51
  52
  53 extern struct gate_struct idt_table[256];
  54
  55 asmlinkage void divide_error(void);
  56 asmlinkage void debug(void);
  57 asmlinkage void nmi(void);
  58 asmlinkage void int3(void);
  59 asmlinkage void overflow(void);
  60 asmlinkage void bounds(void);
  61 asmlinkage void invalid_op(void);
  62 asmlinkage void device_not_available(void);
  63 asmlinkage void double_fault(void);
  64 asmlinkage void coprocessor_segment_overrun(void);
  65 asmlinkage void invalid_TSS(void);
  66 asmlinkage void segment_not_present(void);
  67 asmlinkage void stack_segment(void);
  68 asmlinkage void general_protection(void);
  69 asmlinkage void page_fault(void);
  70 asmlinkage void coprocessor_error(void);
  71 asmlinkage void simd_coprocessor_error(void);
  72 asmlinkage void reserved(void);
  73 asmlinkage void alignment_check(void);
  74 asmlinkage void machine_check(void);
  75 asmlinkage void spurious_interrupt_bug(void);
  76 asmlinkage void call_debug(void);
  77
  78 struct notifier_block *die_chain;
  79 static DEFINE_SPINLOCK(die_notifier_lock);
  80
  81 int register_die_notifier(struct notifier_block *nb)
  82 {
  83         int err = 0;
  84         unsigned long flags;
  85         spin_lock_irqsave(&die_notifier_lock, flags);
  86         err = notifier_chain_register(&die_chain, nb);
  87         spin_unlock_irqrestore(&die_notifier_lock, flags);
  88         return err;
  89 }
  90
  91 static inline void conditional_sti(struct pt_regs *regs)
  92 {
  93         if (regs->eflags & X86_EFLAGS_IF)
  94                 local_irq_enable();
  95 }
  96
  97 static int kstack_depth_to_print = 10;
  98
  99 #ifdef CONFIG_KALLSYMS
 100 #include <linux/kallsyms.h>
 101 int printk_address(unsigned long address)
 102 {
 103         unsigned long offset = 0, symsize;
 104         const char *symname;
 105         char *modname;
 106         char *delim = ":";
 107         char namebuf[128];
 108
 109         symname = kallsyms_lookup(address, &symsize, &offset, &modname, namebuf);
 110         if (!symname)
 111                 return printk("[<%016lx>]", address);
 112         if (!modname)
 113                 modname = delim = "";
 114         return printk("<%016lx>{%s%s%s%s%+ld}",
 115                       address,delim,modname,delim,symname,offset);
 116 }
 117 #else
 118 int printk_address(unsigned long address)
 119 {
 120         return printk("[<%016lx>]", address);
 121 }
 122 #endif
 123
 124 static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
 125                                         unsigned *usedp, const char **idp)
 126 {
 127         static const char ids[N_EXCEPTION_STACKS][8] = {
 128                 [DEBUG_STACK - 1] = "#DB",
 129                 [NMI_STACK - 1] = "NMI",
 130                 [DOUBLEFAULT_STACK - 1] = "#DF",
 131                 [STACKFAULT_STACK - 1] = "#SS",
 132                 [MCE_STACK - 1] = "#MC",
 133         };
 134         unsigned k;
 135
 136         for (k = 0; k < N_EXCEPTION_STACKS; k++) {
 137                 unsigned long end;
 138
 139                 end = per_cpu(init_tss, cpu).ist[k];
 140                 if (stack >= end)
 141                         continue;
 142                 if (stack >= end - EXCEPTION_STKSZ) {
 143                         if (*usedp & (1U << k))
 144                                 break;
 145                         *usedp |= 1U << k;
 146                         *idp = ids[k];
 147                         return (unsigned long *)end;
 148                 }
 149         }
 150         return NULL;
 151 }
 152
 153 /*
 154  * x86-64 can have upto three kernel stacks:
 155  * process stack
 156  * interrupt stack
 157  * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
 158  */
 159
 160 void show_trace(unsigned long *stack)
 161 {
 162         unsigned long addr;
 163         const unsigned cpu = safe_smp_processor_id();
 164         unsigned long *irqstack_end = (unsigned long *)cpu_pda[cpu].irqstackptr;
 165         int i;
 166         unsigned used = 0;
 167
 168         printk("\nCall Trace:");
 169
 170 #define HANDLE_STACK(cond) \
 171         do while (cond) { \
 172                 addr = *stack++; \
 173                 if (kernel_text_address(addr)) { \
 174                         /* \
 175                          * If the address is either in the text segment of the \
 176                          * kernel, or in the region which contains vmalloc'ed \
 177                          * memory, it *may* be the address of a calling \
 178                          * routine; if so, print it so that someone tracing \
 179                          * down the cause of the crash will be able to figure \
 180                          * out the call path that was taken. \
 181                          */ \
 182                         i += printk_address(addr); \
 183                         if (i > 50) { \
 184                                 printk("\n       "); \
 185                                 i = 0; \
 186                         } \
 187                         else \
 188                                 i += printk(" "); \
 189                 } \
 190         } while (0)
 191
 192         for(i = 0; ; ) {
 193                 const char *id;
 194                 unsigned long *estack_end;
 195                 estack_end = in_exception_stack(cpu, (unsigned long)stack,
 196                                                 &used, &id);
 197
 198                 if (estack_end) {
 199                         i += printk(" <%s> ", id);
 200                         HANDLE_STACK (stack < estack_end);
 201                         i += printk(" <EOE> ");
 202                         stack = (unsigned long *) estack_end[-2];
 203                         continue;
 204                 }
 205                 if (irqstack_end) {
 206                         unsigned long *irqstack;
 207                         irqstack = irqstack_end -
 208                                 (IRQSTACKSIZE - 64) / sizeof(*irqstack);
 209
 210                         if (stack >= irqstack && stack < irqstack_end) {
 211                                 i += printk(" <IRQ> ");
 212                                 HANDLE_STACK (stack < irqstack_end);
 213                                 stack = (unsigned long *) (irqstack_end[-1]);
 214                                 irqstack_end = NULL;
 215                                 i += printk(" <EOI> ");
 216                                 continue;
 217                         }
 218                 }
 219                 break;
 220         }
 221
 222         HANDLE_STACK (((long) stack & (THREAD_SIZE-1)) != 0);
 223 #undef HANDLE_STACK
 224         printk("\n");
 225 }
 226
 227 void show_stack(struct task_struct *tsk, unsigned long * rsp)
 228 {
 229         unsigned long *stack;
 230         int i;
 231         const int cpu = safe_smp_processor_id();
 232         unsigned long *irqstack_end = (unsigned long *) (cpu_pda[cpu].irqstackptr);
 233         unsigned long *irqstack = (unsigned long *) (cpu_pda[cpu].irqstackptr - IRQSTACKSIZE);
 234
 235         // debugging aid: "show_stack(NULL, NULL);" prints the
 236         // back trace for this cpu.
 237
 238         if (rsp == NULL) {
 239                 if (tsk)
 240                         rsp = (unsigned long *)tsk->thread.rsp;
 241                 else
 242                         rsp = (unsigned long *)&rsp;
 243         }
 244
 245         stack = rsp;
 246         for(i=0; i < kstack_depth_to_print; i++) {
 247                 if (stack >= irqstack && stack <= irqstack_end) {
 248                         if (stack == irqstack_end) {
 249                                 stack = (unsigned long *) (irqstack_end[-1]);
 250                                 printk(" <EOI> ");
 251                         }
 252                 } else {
 253                 if (((long) stack & (THREAD_SIZE-1)) == 0)
 254                         break;
 255                 }
 256                 if (i && ((i % 4) == 0))
 257                         printk("\n       ");
 258                 printk("%016lx ", *stack++);
 259                 touch_nmi_watchdog();
 260         }
 261         show_trace((unsigned long *)rsp);
 262 }
 263
 264 /*
 265  * The architecture-independent dump_stack generator
 266  */
 267 void dump_stack(void)
 268 {
 269         unsigned long dummy;
 270         show_trace(&dummy);
 271 }
 272
 273 EXPORT_SYMBOL(dump_stack);
 274
 275 void show_registers(struct pt_regs *regs)
 276 {
 277         int i;
 278         int in_kernel = !user_mode(regs);
 279         unsigned long rsp;
 280         const int cpu = safe_smp_processor_id();
 281         struct task_struct *cur = cpu_pda[cpu].pcurrent;
 282
 283                 rsp = regs->rsp;
 284
 285         printk("CPU %d ", cpu);
 286         __show_regs(regs);
 287         printk("Process %s (pid: %d, threadinfo %p, task %p)\n",
 288                 cur->comm, cur->pid, cur->thread_info, cur);
 289
 290         /*
 291          * When in-kernel, we also print out the stack and code at the
 292          * time of the fault..
 293          */
 294         if (in_kernel) {
 295
 296                 printk("Stack: ");
 297                 show_stack(NULL, (unsigned long*)rsp);
 298
 299                 printk("\nCode: ");
 300                 if(regs->rip < PAGE_OFFSET)
 301                         goto bad;
 302
 303                 for(i=0;i<20;i++)
 304                 {
 305                         unsigned char c;
 306                         if(__get_user(c, &((unsigned char*)regs->rip)[i])) {
 307 bad:
 308                                 printk(" Bad RIP value.");
 309                                 break;
 310                         }
 311                         printk("%02x ", c);
 312                 }
 313         }
 314         printk("\n");
 315 }
 316
 317 void handle_BUG(struct pt_regs *regs)
 318 {
 319         struct bug_frame f;
 320         char tmp;
 321
 322         if (user_mode(regs))
 323                 return;
 324         if (__copy_from_user(&f, (struct bug_frame *) regs->rip,
 325                              sizeof(struct bug_frame)))
 326                 return;
 327         if ((unsigned long)f.filename < __PAGE_OFFSET ||
 328             f.ud2[0] != 0x0f || f.ud2[1] != 0x0b)
 329                 return;
 330         if (__get_user(tmp, f.filename))
 331                 f.filename = "unmapped filename";
 332         printk("----------- [cut here ] --------- [please bite here ] ---------\n");
 333         printk(KERN_ALERT "Kernel BUG at %.50s:%d\n", f.filename, f.line);
 334 }
 335
 336 #ifdef CONFIG_BUG
 337 void out_of_line_bug(void)
 338 {
 339         BUG();
 340 }
 341 #endif
 342
 343 static DEFINE_SPINLOCK(die_lock);
 344 static int die_owner = -1;
 345
 346 void oops_begin(void)
 347 {
 348         int cpu = safe_smp_processor_id();
 349         /* racy, but better than risking deadlock. */
 350         local_irq_disable();
 351         if (!spin_trylock(&die_lock)) {
 352                 if (cpu == die_owner)
 353                         /* nested oops. should stop eventually */;
 354                 else
 355                         spin_lock(&die_lock);
 356         }
 357         die_owner = cpu;
 358         console_verbose();
 359         bust_spinlocks(1);
 360 }
 361
 362 void oops_end(void)
 363 {
 364         die_owner = -1;
 365         bust_spinlocks(0);
 366         spin_unlock(&die_lock);
 367         if (panic_on_oops)
 368                 panic("Oops");
 369 }
 370
 371 void __die(const char * str, struct pt_regs * regs, long err)
 372 {
 373         static int die_counter;
 374         printk(KERN_EMERG "%s: %04lx [%u] ", str, err & 0xffff,++die_counter);
 375 #ifdef CONFIG_PREEMPT
 376         printk("PREEMPT ");
 377 #endif
 378 #ifdef CONFIG_SMP
 379         printk("SMP ");
 380 #endif
 381 #ifdef CONFIG_DEBUG_PAGEALLOC
 382         printk("DEBUG_PAGEALLOC");
 383 #endif
 384         printk("\n");
 385         notify_die(DIE_OOPS, (char *)str, regs, err, 255, SIGSEGV);
 386         show_registers(regs);
 387         /* Executive summary in case the oops scrolled away */
 388         printk(KERN_ALERT "RIP ");
 389         printk_address(regs->rip);
 390         printk(" RSP <%016lx>\n", regs->rsp);
 391 }
 392
 393 void die(const char * str, struct pt_regs * regs, long err)
 394 {
 395         oops_begin();
 396         handle_BUG(regs);
 397         __die(str, regs, err);
 398         oops_end();
 399         do_exit(SIGSEGV);
 400 }
 401 static inline void die_if_kernel(const char * str, struct pt_regs * regs, long err)
 402 {
 403         if (!(regs->eflags & VM_MASK) && (regs->cs == __KERNEL_CS))
 404                 die(str, regs, err);
 405 }
 406
 407 void die_nmi(char *str, struct pt_regs *regs)
 408 {
 409         oops_begin();
 410         /*
 411          * We are in trouble anyway, lets at least try
 412          * to get a message out.
 413          */
 414         printk(str, safe_smp_processor_id());
 415         show_registers(regs);
 416         if (panic_on_timeout || panic_on_oops)
 417                 panic("nmi watchdog");
 418         printk("console shuts up ...\n");
 419         oops_end();
 420         do_exit(SIGSEGV);
 421 }
 422
 423 static void __kprobes do_trap(int trapnr, int signr, char *str,
 424                               struct pt_regs * regs, long error_code,
 425                               siginfo_t *info)
 426 {
 427         conditional_sti(regs);
 428
 429 #ifdef CONFIG_CHECKING
 430        {
 431                unsigned long gs;
 432                struct x8664_pda *pda = cpu_pda + safe_smp_processor_id();
 433                rdmsrl(MSR_GS_BASE, gs);
 434                if (gs != (unsigned long)pda) {
 435                        wrmsrl(MSR_GS_BASE, pda);
 436                        printk("%s: wrong gs %lx expected %p rip %lx\n", str, gs, pda,
 437                               regs->rip);
 438                }
 439        }
 440 #endif
 441
 442         if (user_mode(regs)) {
 443                 struct task_struct *tsk = current;
 444
 445                 if (exception_trace && unhandled_signal(tsk, signr))
 446                         printk(KERN_INFO
 447                                "%s[%d] trap %s rip:%lx rsp:%lx error:%lx\n",
 448                                tsk->comm, tsk->pid, str,
 449                                regs->rip,regs->rsp,error_code);
 450
 451                 tsk->thread.error_code = error_code;
 452                 tsk->thread.trap_no = trapnr;
 453                 if (info)
 454                         force_sig_info(signr, info, tsk);
 455                 else
 456                         force_sig(signr, tsk);
 457                 return;
 458         }
 459
 460
 461         /* kernel trap */
 462         {
 463                 const struct exception_table_entry *fixup;
 464                 fixup = search_exception_tables(regs->rip);
 465                 if (fixup) {
 466                         regs->rip = fixup->fixup;
 467                 } else
 468                         die(str, regs, error_code);
 469                 return;
 470         }
 471 }
 472
 473 #define DO_ERROR(trapnr, signr, str, name) \
 474 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
 475 { \
 476         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 477                                                         == NOTIFY_STOP) \
 478                 return; \
 479         do_trap(trapnr, signr, str, regs, error_code, NULL); \
 480 }
 481
 482 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
 483 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
 484 { \
 485         siginfo_t info; \
 486         info.si_signo = signr; \
 487         info.si_errno = 0; \
 488         info.si_code = sicode; \
 489         info.si_addr = (void __user *)siaddr; \
 490         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 491                                                         == NOTIFY_STOP) \
 492                 return; \
 493         do_trap(trapnr, signr, str, regs, error_code, &info); \
 494 }
 495
 496 DO_ERROR_INFO( 0, SIGFPE,  "divide error", divide_error, FPE_INTDIV, regs->rip)
 497 DO_ERROR( 4, SIGSEGV, "overflow", overflow)
 498 DO_ERROR( 5, SIGSEGV, "bounds", bounds)
 499 DO_ERROR_INFO( 6, SIGILL,  "invalid operand", invalid_op, ILL_ILLOPN, regs->rip)
 500 DO_ERROR( 7, SIGSEGV, "device not available", device_not_available)
 501 DO_ERROR( 9, SIGFPE,  "coprocessor segment overrun", coprocessor_segment_overrun)
 502 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
 503 DO_ERROR(11, SIGBUS,  "segment not present", segment_not_present)
 504 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
 505 DO_ERROR(18, SIGSEGV, "reserved", reserved)
 506 DO_ERROR(12, SIGBUS,  "stack segment", stack_segment)
 507 DO_ERROR( 8, SIGSEGV, "double fault", double_fault)
 508
 509 asmlinkage void __kprobes do_general_protection(struct pt_regs * regs,
 510                                                 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 (user_mode(regs)) {
 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 /* Runs on IST stack. This code must keep interrupts off all the time.
 588    Nested NMIs are prevented by the CPU. */
 589 asmlinkage void default_do_nmi(struct pt_regs *regs)
 590 {
 591         unsigned char reason = 0;
 592         int cpu;
 593
 594         cpu = smp_processor_id();
 595
 596         /* Only the BSP gets external NMIs from the system.  */
 597         if (!cpu)
 598                 reason = get_nmi_reason();
 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 __kprobes 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 __kprobes do_debug(struct pt_regs * regs,
 660                                    unsigned long error_code)
 661 {
 662         unsigned long condition;
 663         struct task_struct *tsk = current;
 664         siginfo_t info;
 665
 666 #ifdef CONFIG_CHECKING
 667        {
 668                /* RED-PEN interaction with debugger - could destroy gs */
 669                unsigned long gs;
 670                struct x8664_pda *pda = cpu_pda + safe_smp_processor_id();
 671                rdmsrl(MSR_GS_BASE, gs);
 672                if (gs != (unsigned long)pda) {
 673                        wrmsrl(MSR_GS_BASE, pda);
 674                        printk("debug handler: wrong gs %lx expected %p\n", gs, pda);
 675                }
 676        }
 677 #endif
 678
 679         get_debugreg(condition, 6);
 680
 681         if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
 682                                                 SIGTRAP) == NOTIFY_STOP)
 683                 return;
 684
 685         conditional_sti(regs);
 686
 687         /* Mask out spurious debug traps due to lazy DR7 setting */
 688         if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
 689                 if (!tsk->thread.debugreg7) {
 690                         goto clear_dr7;
 691                 }
 692         }
 693
 694         tsk->thread.debugreg6 = condition;
 695
 696         /* Mask out spurious TF errors due to lazy TF clearing */
 697         if (condition & DR_STEP) {
 698                 /*
 699                  * The TF error should be masked out only if the current
 700                  * process is not traced and if the TRAP flag has been set
 701                  * previously by a tracing process (condition detected by
 702                  * the PT_DTRACE flag); remember that the i386 TRAP flag
 703                  * can be modified by the process itself in user mode,
 704                  * allowing programs to debug themselves without the ptrace()
 705                  * interface.
 706                  */
 707                 if (!user_mode(regs))
 708                        goto clear_TF_reenable;
 709                 /*
 710                  * Was the TF flag set by a debugger? If so, clear it now,
 711                  * so that register information is correct.
 712                  */
 713                 if (tsk->ptrace & PT_DTRACE) {
 714                         regs->eflags &= ~TF_MASK;
 715                         tsk->ptrace &= ~PT_DTRACE;
 716                 }
 717         }
 718
 719         /* Ok, finally something we can handle */
 720         tsk->thread.trap_no = 1;
 721         tsk->thread.error_code = error_code;
 722         info.si_signo = SIGTRAP;
 723         info.si_errno = 0;
 724         info.si_code = TRAP_BRKPT;
 725         if (!user_mode(regs))
 726                 goto clear_dr7;
 727
 728         info.si_addr = (void __user *)regs->rip;
 729         force_sig_info(SIGTRAP, &info, tsk);
 730 clear_dr7:
 731         set_debugreg(0UL, 7);
 732         return;
 733
 734 clear_TF_reenable:
 735         set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
 736         regs->eflags &= ~TF_MASK;
 737 }
 738
 739 static int kernel_math_error(struct pt_regs *regs, char *str)
 740 {
 741         const struct exception_table_entry *fixup;
 742         fixup = search_exception_tables(regs->rip);
 743         if (fixup) {
 744                 regs->rip = fixup->fixup;
 745                 return 1;
 746         }
 747         notify_die(DIE_GPF, str, regs, 0, 16, SIGFPE);
 748         /* Illegal floating point operation in the kernel */
 749         die(str, regs, 0);
 750         return 0;
 751 }
 752
 753 /*
 754  * Note that we play around with the 'TS' bit in an attempt to get
 755  * the correct behaviour even in the presence of the asynchronous
 756  * IRQ13 behaviour
 757  */
 758 asmlinkage void do_coprocessor_error(struct pt_regs *regs)
 759 {
 760         void __user *rip = (void __user *)(regs->rip);
 761         struct task_struct * task;
 762         siginfo_t info;
 763         unsigned short cwd, swd;
 764
 765         conditional_sti(regs);
 766         if (!user_mode(regs) &&
 767             kernel_math_error(regs, "kernel x87 math error"))
 768                 return;
 769
 770         /*
 771          * Save the info for the exception handler and clear the error.
 772          */
 773         task = current;
 774         save_init_fpu(task);
 775         task->thread.trap_no = 16;
 776         task->thread.error_code = 0;
 777         info.si_signo = SIGFPE;
 778         info.si_errno = 0;
 779         info.si_code = __SI_FAULT;
 780         info.si_addr = rip;
 781         /*
 782          * (~cwd & swd) will mask out exceptions that are not set to unmasked
 783          * status.  0x3f is the exception bits in these regs, 0x200 is the
 784          * C1 reg you need in case of a stack fault, 0x040 is the stack
 785          * fault bit.  We should only be taking one exception at a time,
 786          * so if this combination doesn't produce any single exception,
 787          * then we have a bad program that isn't synchronizing its FPU usage
 788          * and it will suffer the consequences since we won't be able to
 789          * fully reproduce the context of the exception
 790          */
 791         cwd = get_fpu_cwd(task);
 792         swd = get_fpu_swd(task);
 793         switch (((~cwd) & swd & 0x3f) | (swd & 0x240)) {
 794                 case 0x000:
 795                 default:
 796                         break;
 797                 case 0x001: /* Invalid Op */
 798                 case 0x041: /* Stack Fault */
 799                 case 0x241: /* Stack Fault | Direction */
 800                         info.si_code = FPE_FLTINV;
 801                         break;
 802                 case 0x002: /* Denormalize */
 803                 case 0x010: /* Underflow */
 804                         info.si_code = FPE_FLTUND;
 805                         break;
 806                 case 0x004: /* Zero Divide */
 807                         info.si_code = FPE_FLTDIV;
 808                         break;
 809                 case 0x008: /* Overflow */
 810                         info.si_code = FPE_FLTOVF;
 811                         break;
 812                 case 0x020: /* Precision */
 813                         info.si_code = FPE_FLTRES;
 814                         break;
 815         }
 816         force_sig_info(SIGFPE, &info, task);
 817 }
 818
 819 asmlinkage void bad_intr(void)
 820 {
 821         printk("bad interrupt");
 822 }
 823
 824 asmlinkage void do_simd_coprocessor_error(struct pt_regs *regs)
 825 {
 826         void __user *rip = (void __user *)(regs->rip);
 827         struct task_struct * task;
 828         siginfo_t info;
 829         unsigned short mxcsr;
 830
 831         conditional_sti(regs);
 832         if (!user_mode(regs) &&
 833                 kernel_math_error(regs, "kernel simd math error"))
 834                 return;
 835
 836         /*
 837          * Save the info for the exception handler and clear the error.
 838          */
 839         task = current;
 840         save_init_fpu(task);
 841         task->thread.trap_no = 19;
 842         task->thread.error_code = 0;
 843         info.si_signo = SIGFPE;
 844         info.si_errno = 0;
 845         info.si_code = __SI_FAULT;
 846         info.si_addr = rip;
 847         /*
 848          * The SIMD FPU exceptions are handled a little differently, as there
 849          * is only a single status/control register.  Thus, to determine which
 850          * unmasked exception was caught we must mask the exception mask bits
 851          * at 0x1f80, and then use these to mask the exception bits at 0x3f.
 852          */
 853         mxcsr = get_fpu_mxcsr(task);
 854         switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
 855                 case 0x000:
 856                 default:
 857                         break;
 858                 case 0x001: /* Invalid Op */
 859                         info.si_code = FPE_FLTINV;
 860                         break;
 861                 case 0x002: /* Denormalize */
 862                 case 0x010: /* Underflow */
 863                         info.si_code = FPE_FLTUND;
 864                         break;
 865                 case 0x004: /* Zero Divide */
 866                         info.si_code = FPE_FLTDIV;
 867                         break;
 868                 case 0x008: /* Overflow */
 869                         info.si_code = FPE_FLTOVF;
 870                         break;
 871                 case 0x020: /* Precision */
 872                         info.si_code = FPE_FLTRES;
 873                         break;
 874         }
 875         force_sig_info(SIGFPE, &info, task);
 876 }
 877
 878 asmlinkage void do_spurious_interrupt_bug(struct pt_regs * regs)
 879 {
 880 }
 881
 882 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
 883 {
 884 }
 885
 886 /*
 887  *  'math_state_restore()' saves the current math information in the
 888  * old math state array, and gets the new ones from the current task
 889  *
 890  * Careful.. There are problems with IBM-designed IRQ13 behaviour.
 891  * Don't touch unless you *really* know how it works.
 892  */
 893 asmlinkage void math_state_restore(void)
 894 {
 895         struct task_struct *me = current;
 896         clts();                 /* Allow maths ops (or we recurse) */
 897
 898         if (!used_math())
 899                 init_fpu(me);
 900         restore_fpu_checking(&me->thread.i387.fxsave);
 901         me->thread_info->status |= TS_USEDFPU;
 902 }
 903
 904 void do_call_debug(struct pt_regs *regs)
 905 {
 906         notify_die(DIE_CALL, "debug call", regs, 0, 255, SIGINT);
 907 }
 908
 909 void __init trap_init(void)
 910 {
 911         set_intr_gate(0,&divide_error);
 912         set_intr_gate_ist(1,&debug,DEBUG_STACK);
 913         set_intr_gate_ist(2,&nmi,NMI_STACK);
 914         set_system_gate(3,&int3);
 915         set_system_gate(4,&overflow);   /* int4-5 can be called from all */
 916         set_system_gate(5,&bounds);
 917         set_intr_gate(6,&invalid_op);
 918         set_intr_gate(7,&device_not_available);
 919         set_intr_gate_ist(8,&double_fault, DOUBLEFAULT_STACK);
 920         set_intr_gate(9,&coprocessor_segment_overrun);
 921         set_intr_gate(10,&invalid_TSS);
 922         set_intr_gate(11,&segment_not_present);
 923         set_intr_gate_ist(12,&stack_segment,STACKFAULT_STACK);
 924         set_intr_gate(13,&general_protection);
 925         set_intr_gate(14,&page_fault);
 926         set_intr_gate(15,&spurious_interrupt_bug);
 927         set_intr_gate(16,&coprocessor_error);
 928         set_intr_gate(17,&alignment_check);
 929 #ifdef CONFIG_X86_MCE
 930         set_intr_gate_ist(18,&machine_check, MCE_STACK);
 931 #endif
 932         set_intr_gate(19,&simd_coprocessor_error);
 933
 934 #ifdef CONFIG_IA32_EMULATION
 935         set_system_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
 936 #endif
 937
 938         set_intr_gate(KDB_VECTOR, call_debug);
 939
 940         /*
 941          * Should be a barrier for any external CPU state.
 942          */
 943         cpu_init();
 944 }
 945
 946
 947 /* Actual parsing is done early in setup.c. */
 948 static int __init oops_dummy(char *s)
 949 {
 950         panic_on_oops = 1;
 951         return -1;
 952 }
 953 __setup("oops=", oops_dummy);
 954
 955 static int __init kstack_setup(char *s)
 956 {
 957         kstack_depth_to_print = simple_strtoul(s,NULL,0);
 958         return 0;
 959 }
 960 __setup("kstack=", kstack_setup);
 961