arch/x86/kernel/traps_64.c

   1 /*
   2  *  Copyright (C) 1991, 1992  Linus Torvalds
   3  *  Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs
   4  *
   5  *  Pentium III FXSR, SSE support
   6  *      Gareth Hughes <gareth@valinux.com>, May 2000
   7  */
   8
   9 /*
  10  * Handle hardware traps and faults.
  11  */
  12 #include <linux/interrupt.h>
  13 #include <linux/kallsyms.h>
  14 #include <linux/spinlock.h>
  15 #include <linux/kprobes.h>
  16 #include <linux/uaccess.h>
  17 #include <linux/utsname.h>
  18 #include <linux/kdebug.h>
  19 #include <linux/kernel.h>
  20 #include <linux/module.h>
  21 #include <linux/ptrace.h>
  22 #include <linux/string.h>
  23 #include <linux/unwind.h>
  24 #include <linux/delay.h>
  25 #include <linux/errno.h>
  26 #include <linux/kexec.h>
  27 #include <linux/sched.h>
  28 #include <linux/timer.h>
  29 #include <linux/init.h>
  30 #include <linux/bug.h>
  31 #include <linux/nmi.h>
  32 #include <linux/mm.h>
  33 #include <linux/smp.h>
  34 #include <linux/io.h>
  35
  36 #ifdef CONFIG_EISA
  37 #include <linux/ioport.h>
  38 #include <linux/eisa.h>
  39 #endif
  40
  41 #ifdef CONFIG_MCA
  42 #include <linux/mca.h>
  43 #endif
  44
  45 #if defined(CONFIG_EDAC)
  46 #include <linux/edac.h>
  47 #endif
  48
  49 #include <asm/stacktrace.h>
  50 #include <asm/processor.h>
  51 #include <asm/debugreg.h>
  52 #include <asm/atomic.h>
  53 #include <asm/system.h>
  54 #include <asm/unwind.h>
  55 #include <asm/traps.h>
  56 #include <asm/desc.h>
  57 #include <asm/i387.h>
  58
  59 #include <mach_traps.h>
  60
  61 #ifdef CONFIG_X86_64
  62 #include <asm/pgalloc.h>
  63 #include <asm/proto.h>
  64 #include <asm/pda.h>
  65 #else
  66 #include <asm/processor-flags.h>
  67 #include <asm/arch_hooks.h>
  68 #include <asm/nmi.h>
  69 #include <asm/smp.h>
  70 #include <asm/io.h>
  71
  72 #include "cpu/mcheck/mce.h"
  73
  74 DECLARE_BITMAP(used_vectors, NR_VECTORS);
  75 EXPORT_SYMBOL_GPL(used_vectors);
  76
  77 asmlinkage int system_call(void);
  78
  79 /* Do we ignore FPU interrupts ? */
  80 char ignore_fpu_irq;
  81
  82 /*
  83  * The IDT has to be page-aligned to simplify the Pentium
  84  * F0 0F bug workaround.. We have a special link segment
  85  * for this.
  86  */
  87 gate_desc idt_table[256]
  88         __attribute__((__section__(".data.idt"))) = { { { { 0, 0 } } }, };
  89 #endif
  90
  91 static int ignore_nmis;
  92
  93 static inline void conditional_sti(struct pt_regs *regs)
  94 {
  95         if (regs->flags & X86_EFLAGS_IF)
  96                 local_irq_enable();
  97 }
  98
  99 static inline void preempt_conditional_sti(struct pt_regs *regs)
 100 {
 101         inc_preempt_count();
 102         if (regs->flags & X86_EFLAGS_IF)
 103                 local_irq_enable();
 104 }
 105
 106 static inline void preempt_conditional_cli(struct pt_regs *regs)
 107 {
 108         if (regs->flags & X86_EFLAGS_IF)
 109                 local_irq_disable();
 110         dec_preempt_count();
 111 }
 112
 113 #ifdef CONFIG_X86_32
 114 static inline void
 115 die_if_kernel(const char *str, struct pt_regs *regs, long err)
 116 {
 117         if (!user_mode_vm(regs))
 118                 die(str, regs, err);
 119 }
 120
 121 /*
 122  * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
 123  * invalid offset set (the LAZY one) and the faulting thread has
 124  * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS,
 125  * we set the offset field correctly and return 1.
 126  */
 127 static int lazy_iobitmap_copy(void)
 128 {
 129         struct thread_struct *thread;
 130         struct tss_struct *tss;
 131         int cpu;
 132
 133         cpu = get_cpu();
 134         tss = &per_cpu(init_tss, cpu);
 135         thread = &current->thread;
 136
 137         if (tss->x86_tss.io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
 138             thread->io_bitmap_ptr) {
 139                 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
 140                        thread->io_bitmap_max);
 141                 /*
 142                  * If the previously set map was extending to higher ports
 143                  * than the current one, pad extra space with 0xff (no access).
 144                  */
 145                 if (thread->io_bitmap_max < tss->io_bitmap_max) {
 146                         memset((char *) tss->io_bitmap +
 147                                 thread->io_bitmap_max, 0xff,
 148                                 tss->io_bitmap_max - thread->io_bitmap_max);
 149                 }
 150                 tss->io_bitmap_max = thread->io_bitmap_max;
 151                 tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
 152                 tss->io_bitmap_owner = thread;
 153                 put_cpu();
 154
 155                 return 1;
 156         }
 157         put_cpu();
 158
 159         return 0;
 160 }
 161 #endif
 162
 163 static void __kprobes
 164 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
 165         long error_code, siginfo_t *info)
 166 {
 167         struct task_struct *tsk = current;
 168
 169 #ifdef CONFIG_X86_32
 170         if (regs->flags & X86_VM_MASK) {
 171                 /*
 172                  * traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
 173                  * On nmi (interrupt 2), do_trap should not be called.
 174                  */
 175                 if (trapnr < 6)
 176                         goto vm86_trap;
 177                 goto trap_signal;
 178         }
 179 #endif
 180
 181         if (!user_mode(regs))
 182                 goto kernel_trap;
 183
 184 #ifdef CONFIG_X86_32
 185 trap_signal:
 186 #endif
 187         /*
 188          * We want error_code and trap_no set for userspace faults and
 189          * kernelspace faults which result in die(), but not
 190          * kernelspace faults which are fixed up.  die() gives the
 191          * process no chance to handle the signal and notice the
 192          * kernel fault information, so that won't result in polluting
 193          * the information about previously queued, but not yet
 194          * delivered, faults.  See also do_general_protection below.
 195          */
 196         tsk->thread.error_code = error_code;
 197         tsk->thread.trap_no = trapnr;
 198
 199 #ifdef CONFIG_X86_64
 200         if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
 201             printk_ratelimit()) {
 202                 printk(KERN_INFO
 203                        "%s[%d] trap %s ip:%lx sp:%lx error:%lx",
 204                        tsk->comm, tsk->pid, str,
 205                        regs->ip, regs->sp, error_code);
 206                 print_vma_addr(" in ", regs->ip);
 207                 printk("\n");
 208         }
 209 #endif
 210
 211         if (info)
 212                 force_sig_info(signr, info, tsk);
 213         else
 214                 force_sig(signr, tsk);
 215         return;
 216
 217 kernel_trap:
 218         if (!fixup_exception(regs)) {
 219                 tsk->thread.error_code = error_code;
 220                 tsk->thread.trap_no = trapnr;
 221                 die(str, regs, error_code);
 222         }
 223         return;
 224
 225 #ifdef CONFIG_X86_32
 226 vm86_trap:
 227         if (handle_vm86_trap((struct kernel_vm86_regs *) regs,
 228                                                 error_code, trapnr))
 229                 goto trap_signal;
 230         return;
 231 #endif
 232 }
 233
 234 #define DO_ERROR(trapnr, signr, str, name)                              \
 235 dotraplinkage void do_##name(struct pt_regs *regs, long error_code)     \
 236 {                                                                       \
 237         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr)  \
 238                                                         == NOTIFY_STOP) \
 239                 return;                                                 \
 240         conditional_sti(regs);                                          \
 241         do_trap(trapnr, signr, str, regs, error_code, NULL);            \
 242 }
 243
 244 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr)         \
 245 dotraplinkage void do_##name(struct pt_regs *regs, long error_code)     \
 246 {                                                                       \
 247         siginfo_t info;                                                 \
 248         info.si_signo = signr;                                          \
 249         info.si_errno = 0;                                              \
 250         info.si_code = sicode;                                          \
 251         info.si_addr = (void __user *)siaddr;                           \
 252         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr)  \
 253                                                         == NOTIFY_STOP) \
 254                 return;                                                 \
 255         conditional_sti(regs);                                          \
 256         do_trap(trapnr, signr, str, regs, error_code, &info);           \
 257 }
 258
 259 DO_ERROR_INFO(0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip)
 260 DO_ERROR(4, SIGSEGV, "overflow", overflow)
 261 DO_ERROR(5, SIGSEGV, "bounds", bounds)
 262 DO_ERROR_INFO(6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip)
 263 DO_ERROR(9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
 264 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
 265 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
 266 #ifdef CONFIG_X86_32
 267 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
 268 #endif
 269 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
 270
 271 #ifdef CONFIG_X86_64
 272 /* Runs on IST stack */
 273 dotraplinkage void do_stack_segment(struct pt_regs *regs, long error_code)
 274 {
 275         if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
 276                         12, SIGBUS) == NOTIFY_STOP)
 277                 return;
 278         preempt_conditional_sti(regs);
 279         do_trap(12, SIGBUS, "stack segment", regs, error_code, NULL);
 280         preempt_conditional_cli(regs);
 281 }
 282
 283 dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
 284 {
 285         static const char str[] = "double fault";
 286         struct task_struct *tsk = current;
 287
 288         /* Return not checked because double check cannot be ignored */
 289         notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV);
 290
 291         tsk->thread.error_code = error_code;
 292         tsk->thread.trap_no = 8;
 293
 294         /* This is always a kernel trap and never fixable (and thus must
 295            never return). */
 296         for (;;)
 297                 die(str, regs, error_code);
 298 }
 299 #endif
 300
 301 dotraplinkage void __kprobes
 302 do_general_protection(struct pt_regs *regs, long error_code)
 303 {
 304         struct task_struct *tsk;
 305
 306         conditional_sti(regs);
 307
 308 #ifdef CONFIG_X86_32
 309         if (lazy_iobitmap_copy()) {
 310                 /* restart the faulting instruction */
 311                 return;
 312         }
 313
 314         if (regs->flags & X86_VM_MASK)
 315                 goto gp_in_vm86;
 316 #endif
 317
 318         tsk = current;
 319         if (!user_mode(regs))
 320                 goto gp_in_kernel;
 321
 322         tsk->thread.error_code = error_code;
 323         tsk->thread.trap_no = 13;
 324
 325         if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
 326                         printk_ratelimit()) {
 327                 printk(KERN_INFO
 328                         "%s[%d] general protection ip:%lx sp:%lx error:%lx",
 329                         tsk->comm, task_pid_nr(tsk),
 330                         regs->ip, regs->sp, error_code);
 331                 print_vma_addr(" in ", regs->ip);
 332                 printk("\n");
 333         }
 334
 335         force_sig(SIGSEGV, tsk);
 336         return;
 337
 338 #ifdef CONFIG_X86_32
 339 gp_in_vm86:
 340         local_irq_enable();
 341         handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
 342         return;
 343 #endif
 344
 345 gp_in_kernel:
 346         if (fixup_exception(regs))
 347                 return;
 348
 349         tsk->thread.error_code = error_code;
 350         tsk->thread.trap_no = 13;
 351         if (notify_die(DIE_GPF, "general protection fault", regs,
 352                                 error_code, 13, SIGSEGV) == NOTIFY_STOP)
 353                 return;
 354         die("general protection fault", regs, error_code);
 355 }
 356
 357 static notrace __kprobes void
 358 mem_parity_error(unsigned char reason, struct pt_regs *regs)
 359 {
 360         printk(KERN_EMERG
 361                 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
 362                         reason, smp_processor_id());
 363
 364         printk(KERN_EMERG
 365                 "You have some hardware problem, likely on the PCI bus.\n");
 366
 367 #if defined(CONFIG_EDAC)
 368         if (edac_handler_set()) {
 369                 edac_atomic_assert_error();
 370                 return;
 371         }
 372 #endif
 373
 374         if (panic_on_unrecovered_nmi)
 375                 panic("NMI: Not continuing");
 376
 377         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 378
 379         /* Clear and disable the memory parity error line. */
 380         reason = (reason & 0xf) | 4;
 381         outb(reason, 0x61);
 382 }
 383
 384 static notrace __kprobes void
 385 io_check_error(unsigned char reason, struct pt_regs *regs)
 386 {
 387         unsigned long i;
 388
 389         printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
 390         show_registers(regs);
 391
 392         /* Re-enable the IOCK line, wait for a few seconds */
 393         reason = (reason & 0xf) | 8;
 394         outb(reason, 0x61);
 395
 396         i = 2000;
 397         while (--i)
 398                 udelay(1000);
 399
 400         reason &= ~8;
 401         outb(reason, 0x61);
 402 }
 403
 404 static notrace __kprobes void
 405 unknown_nmi_error(unsigned char reason, struct pt_regs *regs)
 406 {
 407         if (notify_die(DIE_NMIUNKNOWN, "nmi", regs, reason, 2, SIGINT) ==
 408                         NOTIFY_STOP)
 409                 return;
 410 #ifdef CONFIG_MCA
 411         /*
 412          * Might actually be able to figure out what the guilty party
 413          * is:
 414          */
 415         if (MCA_bus) {
 416                 mca_handle_nmi();
 417                 return;
 418         }
 419 #endif
 420         printk(KERN_EMERG
 421                 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
 422                         reason, smp_processor_id());
 423
 424         printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
 425         if (panic_on_unrecovered_nmi)
 426                 panic("NMI: Not continuing");
 427
 428         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 429 }
 430
 431 #ifdef CONFIG_X86_32
 432 static DEFINE_SPINLOCK(nmi_print_lock);
 433
 434 void notrace __kprobes die_nmi(char *str, struct pt_regs *regs, int do_panic)
 435 {
 436         if (notify_die(DIE_NMIWATCHDOG, str, regs, 0, 2, SIGINT) == NOTIFY_STOP)
 437                 return;
 438
 439         spin_lock(&nmi_print_lock);
 440         /*
 441         * We are in trouble anyway, lets at least try
 442         * to get a message out:
 443         */
 444         bust_spinlocks(1);
 445         printk(KERN_EMERG "%s", str);
 446         printk(" on CPU%d, ip %08lx, registers:\n",
 447                 smp_processor_id(), regs->ip);
 448         show_registers(regs);
 449         if (do_panic)
 450                 panic("Non maskable interrupt");
 451         console_silent();
 452         spin_unlock(&nmi_print_lock);
 453         bust_spinlocks(0);
 454
 455         /*
 456          * If we are in kernel we are probably nested up pretty bad
 457          * and might aswell get out now while we still can:
 458          */
 459         if (!user_mode_vm(regs)) {
 460                 current->thread.trap_no = 2;
 461                 crash_kexec(regs);
 462         }
 463
 464         do_exit(SIGSEGV);
 465 }
 466 #endif
 467
 468 static notrace __kprobes void default_do_nmi(struct pt_regs *regs)
 469 {
 470         unsigned char reason = 0;
 471         int cpu;
 472
 473         cpu = smp_processor_id();
 474
 475         /* Only the BSP gets external NMIs from the system. */
 476         if (!cpu)
 477                 reason = get_nmi_reason();
 478
 479         if (!(reason & 0xc0)) {
 480                 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
 481                                                                 == NOTIFY_STOP)
 482                         return;
 483 #ifdef CONFIG_X86_LOCAL_APIC
 484                 /*
 485                  * Ok, so this is none of the documented NMI sources,
 486                  * so it must be the NMI watchdog.
 487                  */
 488                 if (nmi_watchdog_tick(regs, reason))
 489                         return;
 490                 if (!do_nmi_callback(regs, cpu))
 491                         unknown_nmi_error(reason, regs);
 492 #else
 493                 unknown_nmi_error(reason, regs);
 494 #endif
 495
 496                 return;
 497         }
 498         if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
 499                 return;
 500
 501         /* AK: following checks seem to be broken on modern chipsets. FIXME */
 502         if (reason & 0x80)
 503                 mem_parity_error(reason, regs);
 504         if (reason & 0x40)
 505                 io_check_error(reason, regs);
 506 #ifdef CONFIG_X86_32
 507         /*
 508          * Reassert NMI in case it became active meanwhile
 509          * as it's edge-triggered:
 510          */
 511         reassert_nmi();
 512 #endif
 513 }
 514
 515 dotraplinkage notrace __kprobes void
 516 do_nmi(struct pt_regs *regs, long error_code)
 517 {
 518         nmi_enter();
 519
 520 #ifdef CONFIG_X86_32
 521         { int cpu; cpu = smp_processor_id(); ++nmi_count(cpu); }
 522 #else
 523         add_pda(__nmi_count, 1);
 524 #endif
 525
 526         if (!ignore_nmis)
 527                 default_do_nmi(regs);
 528
 529         nmi_exit();
 530 }
 531
 532 void stop_nmi(void)
 533 {
 534         acpi_nmi_disable();
 535         ignore_nmis++;
 536 }
 537
 538 void restart_nmi(void)
 539 {
 540         ignore_nmis--;
 541         acpi_nmi_enable();
 542 }
 543
 544 /* May run on IST stack. */
 545 dotraplinkage void __kprobes do_int3(struct pt_regs *regs, long error_code)
 546 {
 547 #ifdef CONFIG_KPROBES
 548         if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
 549                         == NOTIFY_STOP)
 550                 return;
 551 #else
 552         if (notify_die(DIE_TRAP, "int3", regs, error_code, 3, SIGTRAP)
 553                         == NOTIFY_STOP)
 554                 return;
 555 #endif
 556
 557         preempt_conditional_sti(regs);
 558         do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
 559         preempt_conditional_cli(regs);
 560 }
 561
 562 #ifdef CONFIG_X86_64
 563 /* Help handler running on IST stack to switch back to user stack
 564    for scheduling or signal handling. The actual stack switch is done in
 565    entry.S */
 566 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
 567 {
 568         struct pt_regs *regs = eregs;
 569         /* Did already sync */
 570         if (eregs == (struct pt_regs *)eregs->sp)
 571                 ;
 572         /* Exception from user space */
 573         else if (user_mode(eregs))
 574                 regs = task_pt_regs(current);
 575         /* Exception from kernel and interrupts are enabled. Move to
 576            kernel process stack. */
 577         else if (eregs->flags & X86_EFLAGS_IF)
 578                 regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
 579         if (eregs != regs)
 580                 *regs = *eregs;
 581         return regs;
 582 }
 583 #endif
 584
 585 /*
 586  * Our handling of the processor debug registers is non-trivial.
 587  * We do not clear them on entry and exit from the kernel. Therefore
 588  * it is possible to get a watchpoint trap here from inside the kernel.
 589  * However, the code in ./ptrace.c has ensured that the user can
 590  * only set watchpoints on userspace addresses. Therefore the in-kernel
 591  * watchpoint trap can only occur in code which is reading/writing
 592  * from user space. Such code must not hold kernel locks (since it
 593  * can equally take a page fault), therefore it is safe to call
 594  * force_sig_info even though that claims and releases locks.
 595  *
 596  * Code in ./signal.c ensures that the debug control register
 597  * is restored before we deliver any signal, and therefore that
 598  * user code runs with the correct debug control register even though
 599  * we clear it here.
 600  *
 601  * Being careful here means that we don't have to be as careful in a
 602  * lot of more complicated places (task switching can be a bit lazy
 603  * about restoring all the debug state, and ptrace doesn't have to
 604  * find every occurrence of the TF bit that could be saved away even
 605  * by user code)
 606  *
 607  * May run on IST stack.
 608  */
 609 dotraplinkage void __kprobes do_debug(struct pt_regs *regs, long error_code)
 610 {
 611         struct task_struct *tsk = current;
 612         unsigned long condition;
 613         int si_code;
 614
 615         get_debugreg(condition, 6);
 616
 617         /*
 618          * The processor cleared BTF, so don't mark that we need it set.
 619          */
 620         clear_tsk_thread_flag(tsk, TIF_DEBUGCTLMSR);
 621         tsk->thread.debugctlmsr = 0;
 622
 623         if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
 624                                                 SIGTRAP) == NOTIFY_STOP)
 625                 return;
 626
 627         /* It's safe to allow irq's after DR6 has been saved */
 628         preempt_conditional_sti(regs);
 629
 630         /* Mask out spurious debug traps due to lazy DR7 setting */
 631         if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
 632                 if (!tsk->thread.debugreg7)
 633                         goto clear_dr7;
 634         }
 635
 636 #ifdef CONFIG_X86_32
 637         if (regs->flags & X86_VM_MASK)
 638                 goto debug_vm86;
 639 #endif
 640
 641         /* Save debug status register where ptrace can see it */
 642         tsk->thread.debugreg6 = condition;
 643
 644         /*
 645          * Single-stepping through TF: make sure we ignore any events in
 646          * kernel space (but re-enable TF when returning to user mode).
 647          */
 648         if (condition & DR_STEP) {
 649                 if (!user_mode(regs))
 650                         goto clear_TF_reenable;
 651         }
 652
 653         si_code = get_si_code(condition);
 654         /* Ok, finally something we can handle */
 655         send_sigtrap(tsk, regs, error_code, si_code);
 656
 657         /*
 658          * Disable additional traps. They'll be re-enabled when
 659          * the signal is delivered.
 660          */
 661 clear_dr7:
 662         set_debugreg(0, 7);
 663         preempt_conditional_cli(regs);
 664         return;
 665
 666 #ifdef CONFIG_X86_32
 667 debug_vm86:
 668         handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
 669         preempt_conditional_cli(regs);
 670         return;
 671 #endif
 672
 673 clear_TF_reenable:
 674         set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
 675         regs->flags &= ~X86_EFLAGS_TF;
 676         preempt_conditional_cli(regs);
 677         return;
 678 }
 679
 680 #ifdef CONFIG_X86_64
 681 static int kernel_math_error(struct pt_regs *regs, const char *str, int trapnr)
 682 {
 683         if (fixup_exception(regs))
 684                 return 1;
 685
 686         notify_die(DIE_GPF, str, regs, 0, trapnr, SIGFPE);
 687         /* Illegal floating point operation in the kernel */
 688         current->thread.trap_no = trapnr;
 689         die(str, regs, 0);
 690         return 0;
 691 }
 692 #endif
 693
 694 /*
 695  * Note that we play around with the 'TS' bit in an attempt to get
 696  * the correct behaviour even in the presence of the asynchronous
 697  * IRQ13 behaviour
 698  */
 699 void math_error(void __user *ip)
 700 {
 701         struct task_struct *task;
 702         siginfo_t info;
 703         unsigned short cwd, swd;
 704
 705         /*
 706          * Save the info for the exception handler and clear the error.
 707          */
 708         task = current;
 709         save_init_fpu(task);
 710         task->thread.trap_no = 16;
 711         task->thread.error_code = 0;
 712         info.si_signo = SIGFPE;
 713         info.si_errno = 0;
 714         info.si_code = __SI_FAULT;
 715         info.si_addr = ip;
 716         /*
 717          * (~cwd & swd) will mask out exceptions that are not set to unmasked
 718          * status.  0x3f is the exception bits in these regs, 0x200 is the
 719          * C1 reg you need in case of a stack fault, 0x040 is the stack
 720          * fault bit.  We should only be taking one exception at a time,
 721          * so if this combination doesn't produce any single exception,
 722          * then we have a bad program that isn't synchronizing its FPU usage
 723          * and it will suffer the consequences since we won't be able to
 724          * fully reproduce the context of the exception
 725          */
 726         cwd = get_fpu_cwd(task);
 727         swd = get_fpu_swd(task);
 728         switch (swd & ~cwd & 0x3f) {
 729         case 0x000: /* No unmasked exception */
 730 #ifdef CONFIG_X86_32
 731                 return;
 732 #endif
 733         default: /* Multiple exceptions */
 734                 break;
 735         case 0x001: /* Invalid Op */
 736                 /*
 737                  * swd & 0x240 == 0x040: Stack Underflow
 738                  * swd & 0x240 == 0x240: Stack Overflow
 739                  * User must clear the SF bit (0x40) if set
 740                  */
 741                 info.si_code = FPE_FLTINV;
 742                 break;
 743         case 0x002: /* Denormalize */
 744         case 0x010: /* Underflow */
 745                 info.si_code = FPE_FLTUND;
 746                 break;
 747         case 0x004: /* Zero Divide */
 748                 info.si_code = FPE_FLTDIV;
 749                 break;
 750         case 0x008: /* Overflow */
 751                 info.si_code = FPE_FLTOVF;
 752                 break;
 753         case 0x020: /* Precision */
 754                 info.si_code = FPE_FLTRES;
 755                 break;
 756         }
 757         force_sig_info(SIGFPE, &info, task);
 758 }
 759
 760 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
 761 {
 762         conditional_sti(regs);
 763
 764 #ifdef CONFIG_X86_32
 765         ignore_fpu_irq = 1;
 766 #else
 767         if (!user_mode(regs) &&
 768             kernel_math_error(regs, "kernel x87 math error", 16))
 769                 return;
 770 #endif
 771
 772         math_error((void __user *)regs->ip);
 773 }
 774
 775 static void simd_math_error(void __user *ip)
 776 {
 777         struct task_struct *task;
 778         siginfo_t info;
 779         unsigned short mxcsr;
 780
 781         /*
 782          * Save the info for the exception handler and clear the error.
 783          */
 784         task = current;
 785         save_init_fpu(task);
 786         task->thread.trap_no = 19;
 787         task->thread.error_code = 0;
 788         info.si_signo = SIGFPE;
 789         info.si_errno = 0;
 790         info.si_code = __SI_FAULT;
 791         info.si_addr = ip;
 792         /*
 793          * The SIMD FPU exceptions are handled a little differently, as there
 794          * is only a single status/control register.  Thus, to determine which
 795          * unmasked exception was caught we must mask the exception mask bits
 796          * at 0x1f80, and then use these to mask the exception bits at 0x3f.
 797          */
 798         mxcsr = get_fpu_mxcsr(task);
 799         switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
 800         case 0x000:
 801         default:
 802                 break;
 803         case 0x001: /* Invalid Op */
 804                 info.si_code = FPE_FLTINV;
 805                 break;
 806         case 0x002: /* Denormalize */
 807         case 0x010: /* Underflow */
 808                 info.si_code = FPE_FLTUND;
 809                 break;
 810         case 0x004: /* Zero Divide */
 811                 info.si_code = FPE_FLTDIV;
 812                 break;
 813         case 0x008: /* Overflow */
 814                 info.si_code = FPE_FLTOVF;
 815                 break;
 816         case 0x020: /* Precision */
 817                 info.si_code = FPE_FLTRES;
 818                 break;
 819         }
 820         force_sig_info(SIGFPE, &info, task);
 821 }
 822
 823 dotraplinkage void
 824 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
 825 {
 826         conditional_sti(regs);
 827
 828 #ifdef CONFIG_X86_32
 829         if (cpu_has_xmm) {
 830                 /* Handle SIMD FPU exceptions on PIII+ processors. */
 831                 ignore_fpu_irq = 1;
 832                 simd_math_error((void __user *)regs->ip);
 833                 return;
 834         }
 835         /*
 836          * Handle strange cache flush from user space exception
 837          * in all other cases.  This is undocumented behaviour.
 838          */
 839         if (regs->flags & X86_VM_MASK) {
 840                 handle_vm86_fault((struct kernel_vm86_regs *)regs, error_code);
 841                 return;
 842         }
 843         current->thread.trap_no = 19;
 844         current->thread.error_code = error_code;
 845         die_if_kernel("cache flush denied", regs, error_code);
 846         force_sig(SIGSEGV, current);
 847 #else
 848         if (!user_mode(regs) &&
 849                         kernel_math_error(regs, "kernel simd math error", 19))
 850                 return;
 851         simd_math_error((void __user *)regs->ip);
 852 #endif
 853 }
 854
 855 dotraplinkage void
 856 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
 857 {
 858         conditional_sti(regs);
 859 #if 0
 860         /* No need to warn about this any longer. */
 861         printk(KERN_INFO "Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
 862 #endif
 863 }
 864
 865 #ifdef CONFIG_X86_32
 866 unsigned long patch_espfix_desc(unsigned long uesp, unsigned long kesp)
 867 {
 868         struct desc_struct *gdt = get_cpu_gdt_table(smp_processor_id());
 869         unsigned long base = (kesp - uesp) & -THREAD_SIZE;
 870         unsigned long new_kesp = kesp - base;
 871         unsigned long lim_pages = (new_kesp | (THREAD_SIZE - 1)) >> PAGE_SHIFT;
 872         __u64 desc = *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS];
 873
 874         /* Set up base for espfix segment */
 875         desc &= 0x00f0ff0000000000ULL;
 876         desc |= ((((__u64)base) << 16) & 0x000000ffffff0000ULL) |
 877                 ((((__u64)base) << 32) & 0xff00000000000000ULL) |
 878                 ((((__u64)lim_pages) << 32) & 0x000f000000000000ULL) |
 879                 (lim_pages & 0xffff);
 880         *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS] = desc;
 881
 882         return new_kesp;
 883 }
 884 #else
 885 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
 886 {
 887 }
 888
 889 asmlinkage void __attribute__((weak)) mce_threshold_interrupt(void)
 890 {
 891 }
 892 #endif
 893
 894 /*
 895  * 'math_state_restore()' saves the current math information in the
 896  * old math state array, and gets the new ones from the current task
 897  *
 898  * Careful.. There are problems with IBM-designed IRQ13 behaviour.
 899  * Don't touch unless you *really* know how it works.
 900  *
 901  * Must be called with kernel preemption disabled (in this case,
 902  * local interrupts are disabled at the call-site in entry.S).
 903  */
 904 asmlinkage void math_state_restore(void)
 905 {
 906         struct thread_info *thread = current_thread_info();
 907         struct task_struct *tsk = thread->task;
 908
 909         if (!tsk_used_math(tsk)) {
 910                 local_irq_enable();
 911                 /*
 912                  * does a slab alloc which can sleep
 913                  */
 914                 if (init_fpu(tsk)) {
 915                         /*
 916                          * ran out of memory!
 917                          */
 918                         do_group_exit(SIGKILL);
 919                         return;
 920                 }
 921                 local_irq_disable();
 922         }
 923
 924         clts();                         /* Allow maths ops (or we recurse) */
 925 #ifdef CONFIG_X86_32
 926         restore_fpu(tsk);
 927 #else
 928         /*
 929          * Paranoid restore. send a SIGSEGV if we fail to restore the state.
 930          */
 931         if (unlikely(restore_fpu_checking(tsk))) {
 932                 stts();
 933                 force_sig(SIGSEGV, tsk);
 934                 return;
 935         }
 936 #endif
 937         thread->status |= TS_USEDFPU;   /* So we fnsave on switch_to() */
 938         tsk->fpu_counter++;
 939 }
 940 EXPORT_SYMBOL_GPL(math_state_restore);
 941
 942 #ifndef CONFIG_MATH_EMULATION
 943 asmlinkage void math_emulate(long arg)
 944 {
 945         printk(KERN_EMERG
 946                 "math-emulation not enabled and no coprocessor found.\n");
 947         printk(KERN_EMERG "killing %s.\n", current->comm);
 948         force_sig(SIGFPE, current);
 949         schedule();
 950 }
 951 #endif /* CONFIG_MATH_EMULATION */
 952
 953 dotraplinkage void __kprobes
 954 do_device_not_available(struct pt_regs *regs, long error)
 955 {
 956 #ifdef CONFIG_X86_32
 957         if (read_cr0() & X86_CR0_EM) {
 958                 conditional_sti(regs);
 959                 math_emulate(0);
 960         } else {
 961                 math_state_restore(); /* interrupts still off */
 962                 conditional_sti(regs);
 963         }
 964 #else
 965         math_state_restore();
 966 #endif
 967 }
 968
 969 #ifdef CONFIG_X86_32
 970 #ifdef CONFIG_X86_MCE
 971 dotraplinkage void __kprobes do_machine_check(struct pt_regs *regs, long error)
 972 {
 973         conditional_sti(regs);
 974         machine_check_vector(regs, error);
 975 }
 976 #endif
 977
 978 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
 979 {
 980         siginfo_t info;
 981         local_irq_enable();
 982
 983         info.si_signo = SIGILL;
 984         info.si_errno = 0;
 985         info.si_code = ILL_BADSTK;
 986         info.si_addr = 0;
 987         if (notify_die(DIE_TRAP, "iret exception",
 988                         regs, error_code, 32, SIGILL) == NOTIFY_STOP)
 989                 return;
 990         do_trap(32, SIGILL, "iret exception", regs, error_code, &info);
 991 }
 992 #endif
 993
 994 void __init trap_init(void)
 995 {
 996 #ifdef CONFIG_X86_32
 997         int i;
 998 #endif
 999
1000 #ifdef CONFIG_EISA
1001         void __iomem *p = early_ioremap(0x0FFFD9, 4);
1002
1003         if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
1004                 EISA_bus = 1;
1005         early_iounmap(p, 4);
1006 #endif
1007
1008         set_intr_gate(0, &divide_error);
1009         set_intr_gate_ist(1, &debug, DEBUG_STACK);
1010         set_intr_gate_ist(2, &nmi, NMI_STACK);
1011         /* int3 can be called from all */
1012         set_system_intr_gate_ist(3, &int3, DEBUG_STACK);
1013         /* int4 can be called from all */
1014         set_system_intr_gate(4, &overflow);
1015         set_intr_gate(5, &bounds);
1016         set_intr_gate(6, &invalid_op);
1017         set_intr_gate(7, &device_not_available);
1018 #ifdef CONFIG_X86_32
1019         set_task_gate(8, GDT_ENTRY_DOUBLEFAULT_TSS);
1020 #else
1021         set_intr_gate_ist(8, &double_fault, DOUBLEFAULT_STACK);
1022 #endif
1023         set_intr_gate(9, &coprocessor_segment_overrun);
1024         set_intr_gate(10, &invalid_TSS);
1025         set_intr_gate(11, &segment_not_present);
1026         set_intr_gate_ist(12, &stack_segment, STACKFAULT_STACK);
1027         set_intr_gate(13, &general_protection);
1028         set_intr_gate(14, &page_fault);
1029         set_intr_gate(15, &spurious_interrupt_bug);
1030         set_intr_gate(16, &coprocessor_error);
1031         set_intr_gate(17, &alignment_check);
1032 #ifdef CONFIG_X86_MCE
1033         set_intr_gate_ist(18, &machine_check, MCE_STACK);
1034 #endif
1035         set_intr_gate(19, &simd_coprocessor_error);
1036
1037 #ifdef CONFIG_IA32_EMULATION
1038         set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
1039 #endif
1040
1041 #ifdef CONFIG_X86_32
1042         if (cpu_has_fxsr) {
1043                 printk(KERN_INFO "Enabling fast FPU save and restore... ");
1044                 set_in_cr4(X86_CR4_OSFXSR);
1045                 printk("done.\n");
1046         }
1047         if (cpu_has_xmm) {
1048                 printk(KERN_INFO
1049                         "Enabling unmasked SIMD FPU exception support... ");
1050                 set_in_cr4(X86_CR4_OSXMMEXCPT);
1051                 printk("done.\n");
1052         }
1053
1054         set_system_trap_gate(SYSCALL_VECTOR, &system_call);
1055
1056         /* Reserve all the builtin and the syscall vector: */
1057         for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
1058                 set_bit(i, used_vectors);
1059
1060         set_bit(SYSCALL_VECTOR, used_vectors);
1061 #endif
1062         /*
1063          * Should be a barrier for any external CPU state:
1064          */
1065         cpu_init();
1066
1067 #ifdef CONFIG_X86_32
1068         trap_init_hook();
1069 #endif
1070 }