arch/x86/kernel/traps.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/delay.h>
  24 #include <linux/errno.h>
  25 #include <linux/kexec.h>
  26 #include <linux/sched.h>
  27 #include <linux/timer.h>
  28 #include <linux/init.h>
  29 #include <linux/bug.h>
  30 #include <linux/nmi.h>
  31 #include <linux/mm.h>
  32 #include <linux/smp.h>
  33 #include <linux/io.h>
  34
  35 #ifdef CONFIG_EISA
  36 #include <linux/ioport.h>
  37 #include <linux/eisa.h>
  38 #endif
  39
  40 #ifdef CONFIG_MCA
  41 #include <linux/mca.h>
  42 #endif
  43
  44 #if defined(CONFIG_EDAC)
  45 #include <linux/edac.h>
  46 #endif
  47
  48 #include <asm/stacktrace.h>
  49 #include <asm/processor.h>
  50 #include <asm/debugreg.h>
  51 #include <asm/atomic.h>
  52 #include <asm/system.h>
  53 #include <asm/traps.h>
  54 #include <asm/desc.h>
  55 #include <asm/i387.h>
  56
  57 #include <mach_traps.h>
  58
  59 #ifdef CONFIG_X86_64
  60 #include <asm/pgalloc.h>
  61 #include <asm/proto.h>
  62 #include <asm/pda.h>
  63 #else
  64 #include <asm/processor-flags.h>
  65 #include <asm/arch_hooks.h>
  66 #include <asm/nmi.h>
  67 #include <asm/smp.h>
  68 #include <asm/io.h>
  69 #include <asm/traps.h>
  70
  71 #include "cpu/mcheck/mce.h"
  72
  73 asmlinkage int system_call(void);
  74
  75 /* Do we ignore FPU interrupts ? */
  76 char ignore_fpu_irq;
  77
  78 /*
  79  * The IDT has to be page-aligned to simplify the Pentium
  80  * F0 0F bug workaround.. We have a special link segment
  81  * for this.
  82  */
  83 gate_desc idt_table[256]
  84         __attribute__((__section__(".data.idt"))) = { { { { 0, 0 } } }, };
  85 #endif
  86
  87 DECLARE_BITMAP(used_vectors, NR_VECTORS);
  88 EXPORT_SYMBOL_GPL(used_vectors);
  89
  90 static int ignore_nmis;
  91
  92 static inline void conditional_sti(struct pt_regs *regs)
  93 {
  94         if (regs->flags & X86_EFLAGS_IF)
  95                 local_irq_enable();
  96 }
  97
  98 static inline void preempt_conditional_sti(struct pt_regs *regs)
  99 {
 100         inc_preempt_count();
 101         if (regs->flags & X86_EFLAGS_IF)
 102                 local_irq_enable();
 103 }
 104
 105 static inline void preempt_conditional_cli(struct pt_regs *regs)
 106 {
 107         if (regs->flags & X86_EFLAGS_IF)
 108                 local_irq_disable();
 109         dec_preempt_count();
 110 }
 111
 112 #ifdef CONFIG_X86_32
 113 static inline void
 114 die_if_kernel(const char *str, struct pt_regs *regs, long err)
 115 {
 116         if (!user_mode_vm(regs))
 117                 die(str, regs, err);
 118 }
 119
 120 /*
 121  * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
 122  * invalid offset set (the LAZY one) and the faulting thread has
 123  * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS,
 124  * we set the offset field correctly and return 1.
 125  */
 126 static int lazy_iobitmap_copy(void)
 127 {
 128         struct thread_struct *thread;
 129         struct tss_struct *tss;
 130         int cpu;
 131
 132         cpu = get_cpu();
 133         tss = &per_cpu(init_tss, cpu);
 134         thread = &current->thread;
 135
 136         if (tss->x86_tss.io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
 137             thread->io_bitmap_ptr) {
 138                 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
 139                        thread->io_bitmap_max);
 140                 /*
 141                  * If the previously set map was extending to higher ports
 142                  * than the current one, pad extra space with 0xff (no access).
 143                  */
 144                 if (thread->io_bitmap_max < tss->io_bitmap_max) {
 145                         memset((char *) tss->io_bitmap +
 146                                 thread->io_bitmap_max, 0xff,
 147                                 tss->io_bitmap_max - thread->io_bitmap_max);
 148                 }
 149                 tss->io_bitmap_max = thread->io_bitmap_max;
 150                 tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
 151                 tss->io_bitmap_owner = thread;
 152                 put_cpu();
 153
 154                 return 1;
 155         }
 156         put_cpu();
 157
 158         return 0;
 159 }
 160 #endif
 161
 162 static void __kprobes
 163 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
 164         long error_code, siginfo_t *info)
 165 {
 166         struct task_struct *tsk = current;
 167
 168 #ifdef CONFIG_X86_32
 169         if (regs->flags & X86_VM_MASK) {
 170                 /*
 171                  * traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
 172                  * On nmi (interrupt 2), do_trap should not be called.
 173                  */
 174                 if (trapnr < 6)
 175                         goto vm86_trap;
 176                 goto trap_signal;
 177         }
 178 #endif
 179
 180         if (!user_mode(regs))
 181                 goto kernel_trap;
 182
 183 #ifdef CONFIG_X86_32
 184 trap_signal:
 185 #endif
 186         /*
 187          * We want error_code and trap_no set for userspace faults and
 188          * kernelspace faults which result in die(), but not
 189          * kernelspace faults which are fixed up.  die() gives the
 190          * process no chance to handle the signal and notice the
 191          * kernel fault information, so that won't result in polluting
 192          * the information about previously queued, but not yet
 193          * delivered, faults.  See also do_general_protection below.
 194          */
 195         tsk->thread.error_code = error_code;
 196         tsk->thread.trap_no = trapnr;
 197
 198 #ifdef CONFIG_X86_64
 199         if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
 200             printk_ratelimit()) {
 201                 printk(KERN_INFO
 202                        "%s[%d] trap %s ip:%lx sp:%lx error:%lx",
 203                        tsk->comm, tsk->pid, str,
 204                        regs->ip, regs->sp, error_code);
 205                 print_vma_addr(" in ", regs->ip);
 206                 printk("\n");
 207         }
 208 #endif
 209
 210         if (info)
 211                 force_sig_info(signr, info, tsk);
 212         else
 213                 force_sig(signr, tsk);
 214         return;
 215
 216 kernel_trap:
 217         if (!fixup_exception(regs)) {
 218                 tsk->thread.error_code = error_code;
 219                 tsk->thread.trap_no = trapnr;
 220                 die(str, regs, error_code);
 221         }
 222         return;
 223
 224 #ifdef CONFIG_X86_32
 225 vm86_trap:
 226         if (handle_vm86_trap((struct kernel_vm86_regs *) regs,
 227                                                 error_code, trapnr))
 228                 goto trap_signal;
 229         return;
 230 #endif
 231 }
 232
 233 #define DO_ERROR(trapnr, signr, str, name)                              \
 234 dotraplinkage void do_##name(struct pt_regs *regs, long error_code)     \
 235 {                                                                       \
 236         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr)  \
 237                                                         == NOTIFY_STOP) \
 238                 return;                                                 \
 239         conditional_sti(regs);                                          \
 240         do_trap(trapnr, signr, str, regs, error_code, NULL);            \
 241 }
 242
 243 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr)         \
 244 dotraplinkage void do_##name(struct pt_regs *regs, long error_code)     \
 245 {                                                                       \
 246         siginfo_t info;                                                 \
 247         info.si_signo = signr;                                          \
 248         info.si_errno = 0;                                              \
 249         info.si_code = sicode;                                          \
 250         info.si_addr = (void __user *)siaddr;                           \
 251         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr)  \
 252                                                         == NOTIFY_STOP) \
 253                 return;                                                 \
 254         conditional_sti(regs);                                          \
 255         do_trap(trapnr, signr, str, regs, error_code, &info);           \
 256 }
 257
 258 DO_ERROR_INFO(0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip)
 259 DO_ERROR(4, SIGSEGV, "overflow", overflow)
 260 DO_ERROR(5, SIGSEGV, "bounds", bounds)
 261 DO_ERROR_INFO(6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip)
 262 DO_ERROR(9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
 263 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
 264 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
 265 #ifdef CONFIG_X86_32
 266 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
 267 #endif
 268 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
 269
 270 #ifdef CONFIG_X86_64
 271 /* Runs on IST stack */
 272 dotraplinkage void do_stack_segment(struct pt_regs *regs, long error_code)
 273 {
 274         if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
 275                         12, SIGBUS) == NOTIFY_STOP)
 276                 return;
 277         preempt_conditional_sti(regs);
 278         do_trap(12, SIGBUS, "stack segment", regs, error_code, NULL);
 279         preempt_conditional_cli(regs);
 280 }
 281
 282 dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
 283 {
 284         static const char str[] = "double fault";
 285         struct task_struct *tsk = current;
 286
 287         /* Return not checked because double check cannot be ignored */
 288         notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV);
 289
 290         tsk->thread.error_code = error_code;
 291         tsk->thread.trap_no = 8;
 292
 293         /*
 294          * This is always a kernel trap and never fixable (and thus must
 295          * never return).
 296          */
 297         for (;;)
 298                 die(str, regs, error_code);
 299 }
 300 #endif
 301
 302 dotraplinkage void __kprobes
 303 do_general_protection(struct pt_regs *regs, long error_code)
 304 {
 305         struct task_struct *tsk;
 306
 307         conditional_sti(regs);
 308
 309 #ifdef CONFIG_X86_32
 310         if (lazy_iobitmap_copy()) {
 311                 /* restart the faulting instruction */
 312                 return;
 313         }
 314
 315         if (regs->flags & X86_VM_MASK)
 316                 goto gp_in_vm86;
 317 #endif
 318
 319         tsk = current;
 320         if (!user_mode(regs))
 321                 goto gp_in_kernel;
 322
 323         tsk->thread.error_code = error_code;
 324         tsk->thread.trap_no = 13;
 325
 326         if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
 327                         printk_ratelimit()) {
 328                 printk(KERN_INFO
 329                         "%s[%d] general protection ip:%lx sp:%lx error:%lx",
 330                         tsk->comm, task_pid_nr(tsk),
 331                         regs->ip, regs->sp, error_code);
 332                 print_vma_addr(" in ", regs->ip);
 333                 printk("\n");
 334         }
 335
 336         force_sig(SIGSEGV, tsk);
 337         return;
 338
 339 #ifdef CONFIG_X86_32
 340 gp_in_vm86:
 341         local_irq_enable();
 342         handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
 343         return;
 344 #endif
 345
 346 gp_in_kernel:
 347         if (fixup_exception(regs))
 348                 return;
 349
 350         tsk->thread.error_code = error_code;
 351         tsk->thread.trap_no = 13;
 352         if (notify_die(DIE_GPF, "general protection fault", regs,
 353                                 error_code, 13, SIGSEGV) == NOTIFY_STOP)
 354                 return;
 355         die("general protection fault", regs, error_code);
 356 }
 357
 358 static notrace __kprobes void
 359 mem_parity_error(unsigned char reason, struct pt_regs *regs)
 360 {
 361         printk(KERN_EMERG
 362                 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
 363                         reason, smp_processor_id());
 364
 365         printk(KERN_EMERG
 366                 "You have some hardware problem, likely on the PCI bus.\n");
 367
 368 #if defined(CONFIG_EDAC)
 369         if (edac_handler_set()) {
 370                 edac_atomic_assert_error();
 371                 return;
 372         }
 373 #endif
 374
 375         if (panic_on_unrecovered_nmi)
 376                 panic("NMI: Not continuing");
 377
 378         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 379
 380         /* Clear and disable the memory parity error line. */
 381         reason = (reason & 0xf) | 4;
 382         outb(reason, 0x61);
 383 }
 384
 385 static notrace __kprobes void
 386 io_check_error(unsigned char reason, struct pt_regs *regs)
 387 {
 388         unsigned long i;
 389
 390         printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
 391         show_registers(regs);
 392
 393         /* Re-enable the IOCK line, wait for a few seconds */
 394         reason = (reason & 0xf) | 8;
 395         outb(reason, 0x61);
 396
 397         i = 2000;
 398         while (--i)
 399                 udelay(1000);
 400
 401         reason &= ~8;
 402         outb(reason, 0x61);
 403 }
 404
 405 static notrace __kprobes void
 406 unknown_nmi_error(unsigned char reason, struct pt_regs *regs)
 407 {
 408         if (notify_die(DIE_NMIUNKNOWN, "nmi", regs, reason, 2, SIGINT) ==
 409                         NOTIFY_STOP)
 410                 return;
 411 #ifdef CONFIG_MCA
 412         /*
 413          * Might actually be able to figure out what the guilty party
 414          * is:
 415          */
 416         if (MCA_bus) {
 417                 mca_handle_nmi();
 418                 return;
 419         }
 420 #endif
 421         printk(KERN_EMERG
 422                 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
 423                         reason, smp_processor_id());
 424
 425         printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
 426         if (panic_on_unrecovered_nmi)
 427                 panic("NMI: Not continuing");
 428
 429         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 430 }
 431
 432 static notrace __kprobes void default_do_nmi(struct pt_regs *regs)
 433 {
 434         unsigned char reason = 0;
 435         int cpu;
 436
 437         cpu = smp_processor_id();
 438
 439         /* Only the BSP gets external NMIs from the system. */
 440         if (!cpu)
 441                 reason = get_nmi_reason();
 442
 443         if (!(reason & 0xc0)) {
 444                 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
 445                                                                 == NOTIFY_STOP)
 446                         return;
 447 #ifdef CONFIG_X86_LOCAL_APIC
 448                 /*
 449                  * Ok, so this is none of the documented NMI sources,
 450                  * so it must be the NMI watchdog.
 451                  */
 452                 if (nmi_watchdog_tick(regs, reason))
 453                         return;
 454                 if (!do_nmi_callback(regs, cpu))
 455                         unknown_nmi_error(reason, regs);
 456 #else
 457                 unknown_nmi_error(reason, regs);
 458 #endif
 459
 460                 return;
 461         }
 462         if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
 463                 return;
 464
 465         /* AK: following checks seem to be broken on modern chipsets. FIXME */
 466         if (reason & 0x80)
 467                 mem_parity_error(reason, regs);
 468         if (reason & 0x40)
 469                 io_check_error(reason, regs);
 470 #ifdef CONFIG_X86_32
 471         /*
 472          * Reassert NMI in case it became active meanwhile
 473          * as it's edge-triggered:
 474          */
 475         reassert_nmi();
 476 #endif
 477 }
 478
 479 dotraplinkage notrace __kprobes void
 480 do_nmi(struct pt_regs *regs, long error_code)
 481 {
 482         nmi_enter();
 483
 484         inc_irq_stat(__nmi_count);
 485
 486         if (!ignore_nmis)
 487                 default_do_nmi(regs);
 488
 489         nmi_exit();
 490 }
 491
 492 void stop_nmi(void)
 493 {
 494         acpi_nmi_disable();
 495         ignore_nmis++;
 496 }
 497
 498 void restart_nmi(void)
 499 {
 500         ignore_nmis--;
 501         acpi_nmi_enable();
 502 }
 503
 504 /* May run on IST stack. */
 505 dotraplinkage void __kprobes do_int3(struct pt_regs *regs, long error_code)
 506 {
 507 #ifdef CONFIG_KPROBES
 508         if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
 509                         == NOTIFY_STOP)
 510                 return;
 511 #else
 512         if (notify_die(DIE_TRAP, "int3", regs, error_code, 3, SIGTRAP)
 513                         == NOTIFY_STOP)
 514                 return;
 515 #endif
 516
 517         preempt_conditional_sti(regs);
 518         do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
 519         preempt_conditional_cli(regs);
 520 }
 521
 522 #ifdef CONFIG_X86_64
 523 /*
 524  * Help handler running on IST stack to switch back to user stack
 525  * for scheduling or signal handling. The actual stack switch is done in
 526  * entry.S
 527  */
 528 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
 529 {
 530         struct pt_regs *regs = eregs;
 531         /* Did already sync */
 532         if (eregs == (struct pt_regs *)eregs->sp)
 533                 ;
 534         /* Exception from user space */
 535         else if (user_mode(eregs))
 536                 regs = task_pt_regs(current);
 537         /*
 538          * Exception from kernel and interrupts are enabled. Move to
 539          * kernel process stack.
 540          */
 541         else if (eregs->flags & X86_EFLAGS_IF)
 542                 regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
 543         if (eregs != regs)
 544                 *regs = *eregs;
 545         return regs;
 546 }
 547 #endif
 548
 549 /*
 550  * Our handling of the processor debug registers is non-trivial.
 551  * We do not clear them on entry and exit from the kernel. Therefore
 552  * it is possible to get a watchpoint trap here from inside the kernel.
 553  * However, the code in ./ptrace.c has ensured that the user can
 554  * only set watchpoints on userspace addresses. Therefore the in-kernel
 555  * watchpoint trap can only occur in code which is reading/writing
 556  * from user space. Such code must not hold kernel locks (since it
 557  * can equally take a page fault), therefore it is safe to call
 558  * force_sig_info even though that claims and releases locks.
 559  *
 560  * Code in ./signal.c ensures that the debug control register
 561  * is restored before we deliver any signal, and therefore that
 562  * user code runs with the correct debug control register even though
 563  * we clear it here.
 564  *
 565  * Being careful here means that we don't have to be as careful in a
 566  * lot of more complicated places (task switching can be a bit lazy
 567  * about restoring all the debug state, and ptrace doesn't have to
 568  * find every occurrence of the TF bit that could be saved away even
 569  * by user code)
 570  *
 571  * May run on IST stack.
 572  */
 573 dotraplinkage void __kprobes do_debug(struct pt_regs *regs, long error_code)
 574 {
 575         struct task_struct *tsk = current;
 576         unsigned long condition;
 577         int si_code;
 578
 579         get_debugreg(condition, 6);
 580
 581         /*
 582          * The processor cleared BTF, so don't mark that we need it set.
 583          */
 584         clear_tsk_thread_flag(tsk, TIF_DEBUGCTLMSR);
 585         tsk->thread.debugctlmsr = 0;
 586
 587         if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
 588                                                 SIGTRAP) == NOTIFY_STOP)
 589                 return;
 590
 591         /* It's safe to allow irq's after DR6 has been saved */
 592         preempt_conditional_sti(regs);
 593
 594         /* Mask out spurious debug traps due to lazy DR7 setting */
 595         if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
 596                 if (!tsk->thread.debugreg7)
 597                         goto clear_dr7;
 598         }
 599
 600 #ifdef CONFIG_X86_32
 601         if (regs->flags & X86_VM_MASK)
 602                 goto debug_vm86;
 603 #endif
 604
 605         /* Save debug status register where ptrace can see it */
 606         tsk->thread.debugreg6 = condition;
 607
 608         /*
 609          * Single-stepping through TF: make sure we ignore any events in
 610          * kernel space (but re-enable TF when returning to user mode).
 611          */
 612         if (condition & DR_STEP) {
 613                 if (!user_mode(regs))
 614                         goto clear_TF_reenable;
 615         }
 616
 617         si_code = get_si_code(condition);
 618         /* Ok, finally something we can handle */
 619         send_sigtrap(tsk, regs, error_code, si_code);
 620
 621         /*
 622          * Disable additional traps. They'll be re-enabled when
 623          * the signal is delivered.
 624          */
 625 clear_dr7:
 626         set_debugreg(0, 7);
 627         preempt_conditional_cli(regs);
 628         return;
 629
 630 #ifdef CONFIG_X86_32
 631 debug_vm86:
 632         handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
 633         preempt_conditional_cli(regs);
 634         return;
 635 #endif
 636
 637 clear_TF_reenable:
 638         set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
 639         regs->flags &= ~X86_EFLAGS_TF;
 640         preempt_conditional_cli(regs);
 641         return;
 642 }
 643
 644 #ifdef CONFIG_X86_64
 645 static int kernel_math_error(struct pt_regs *regs, const char *str, int trapnr)
 646 {
 647         if (fixup_exception(regs))
 648                 return 1;
 649
 650         notify_die(DIE_GPF, str, regs, 0, trapnr, SIGFPE);
 651         /* Illegal floating point operation in the kernel */
 652         current->thread.trap_no = trapnr;
 653         die(str, regs, 0);
 654         return 0;
 655 }
 656 #endif
 657
 658 /*
 659  * Note that we play around with the 'TS' bit in an attempt to get
 660  * the correct behaviour even in the presence of the asynchronous
 661  * IRQ13 behaviour
 662  */
 663 void math_error(void __user *ip)
 664 {
 665         struct task_struct *task;
 666         siginfo_t info;
 667         unsigned short cwd, swd, err;
 668
 669         /*
 670          * Save the info for the exception handler and clear the error.
 671          */
 672         task = current;
 673         save_init_fpu(task);
 674         task->thread.trap_no = 16;
 675         task->thread.error_code = 0;
 676         info.si_signo = SIGFPE;
 677         info.si_errno = 0;
 678         info.si_addr = ip;
 679         /*
 680          * (~cwd & swd) will mask out exceptions that are not set to unmasked
 681          * status.  0x3f is the exception bits in these regs, 0x200 is the
 682          * C1 reg you need in case of a stack fault, 0x040 is the stack
 683          * fault bit.  We should only be taking one exception at a time,
 684          * so if this combination doesn't produce any single exception,
 685          * then we have a bad program that isn't synchronizing its FPU usage
 686          * and it will suffer the consequences since we won't be able to
 687          * fully reproduce the context of the exception
 688          */
 689         cwd = get_fpu_cwd(task);
 690         swd = get_fpu_swd(task);
 691
 692         err = swd & ~cwd;
 693
 694         if (err & 0x001) {      /* Invalid op */
 695                 /*
 696                  * swd & 0x240 == 0x040: Stack Underflow
 697                  * swd & 0x240 == 0x240: Stack Overflow
 698                  * User must clear the SF bit (0x40) if set
 699                  */
 700                 info.si_code = FPE_FLTINV;
 701         } else if (err & 0x004) { /* Divide by Zero */
 702                 info.si_code = FPE_FLTDIV;
 703         } else if (err & 0x008) { /* Overflow */
 704                 info.si_code = FPE_FLTOVF;
 705         } else if (err & 0x012) { /* Denormal, Underflow */
 706                 info.si_code = FPE_FLTUND;
 707         } else if (err & 0x020) { /* Precision */
 708                 info.si_code = FPE_FLTRES;
 709         } else {
 710                 /*
 711                  * If we're using IRQ 13, or supposedly even some trap 16
 712                  * implementations, it's possible we get a spurious trap...
 713                  */
 714                 return;         /* Spurious trap, no error */
 715         }
 716         force_sig_info(SIGFPE, &info, task);
 717 }
 718
 719 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
 720 {
 721         conditional_sti(regs);
 722
 723 #ifdef CONFIG_X86_32
 724         ignore_fpu_irq = 1;
 725 #else
 726         if (!user_mode(regs) &&
 727             kernel_math_error(regs, "kernel x87 math error", 16))
 728                 return;
 729 #endif
 730
 731         math_error((void __user *)regs->ip);
 732 }
 733
 734 static void simd_math_error(void __user *ip)
 735 {
 736         struct task_struct *task;
 737         siginfo_t info;
 738         unsigned short mxcsr;
 739
 740         /*
 741          * Save the info for the exception handler and clear the error.
 742          */
 743         task = current;
 744         save_init_fpu(task);
 745         task->thread.trap_no = 19;
 746         task->thread.error_code = 0;
 747         info.si_signo = SIGFPE;
 748         info.si_errno = 0;
 749         info.si_code = __SI_FAULT;
 750         info.si_addr = ip;
 751         /*
 752          * The SIMD FPU exceptions are handled a little differently, as there
 753          * is only a single status/control register.  Thus, to determine which
 754          * unmasked exception was caught we must mask the exception mask bits
 755          * at 0x1f80, and then use these to mask the exception bits at 0x3f.
 756          */
 757         mxcsr = get_fpu_mxcsr(task);
 758         switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
 759         case 0x000:
 760         default:
 761                 break;
 762         case 0x001: /* Invalid Op */
 763                 info.si_code = FPE_FLTINV;
 764                 break;
 765         case 0x002: /* Denormalize */
 766         case 0x010: /* Underflow */
 767                 info.si_code = FPE_FLTUND;
 768                 break;
 769         case 0x004: /* Zero Divide */
 770                 info.si_code = FPE_FLTDIV;
 771                 break;
 772         case 0x008: /* Overflow */
 773                 info.si_code = FPE_FLTOVF;
 774                 break;
 775         case 0x020: /* Precision */
 776                 info.si_code = FPE_FLTRES;
 777                 break;
 778         }
 779         force_sig_info(SIGFPE, &info, task);
 780 }
 781
 782 dotraplinkage void
 783 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
 784 {
 785         conditional_sti(regs);
 786
 787 #ifdef CONFIG_X86_32
 788         if (cpu_has_xmm) {
 789                 /* Handle SIMD FPU exceptions on PIII+ processors. */
 790                 ignore_fpu_irq = 1;
 791                 simd_math_error((void __user *)regs->ip);
 792                 return;
 793         }
 794         /*
 795          * Handle strange cache flush from user space exception
 796          * in all other cases.  This is undocumented behaviour.
 797          */
 798         if (regs->flags & X86_VM_MASK) {
 799                 handle_vm86_fault((struct kernel_vm86_regs *)regs, error_code);
 800                 return;
 801         }
 802         current->thread.trap_no = 19;
 803         current->thread.error_code = error_code;
 804         die_if_kernel("cache flush denied", regs, error_code);
 805         force_sig(SIGSEGV, current);
 806 #else
 807         if (!user_mode(regs) &&
 808                         kernel_math_error(regs, "kernel simd math error", 19))
 809                 return;
 810         simd_math_error((void __user *)regs->ip);
 811 #endif
 812 }
 813
 814 dotraplinkage void
 815 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
 816 {
 817         conditional_sti(regs);
 818 #if 0
 819         /* No need to warn about this any longer. */
 820         printk(KERN_INFO "Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
 821 #endif
 822 }
 823
 824 #ifdef CONFIG_X86_32
 825 unsigned long patch_espfix_desc(unsigned long uesp, unsigned long kesp)
 826 {
 827         struct desc_struct *gdt = get_cpu_gdt_table(smp_processor_id());
 828         unsigned long base = (kesp - uesp) & -THREAD_SIZE;
 829         unsigned long new_kesp = kesp - base;
 830         unsigned long lim_pages = (new_kesp | (THREAD_SIZE - 1)) >> PAGE_SHIFT;
 831         __u64 desc = *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS];
 832
 833         /* Set up base for espfix segment */
 834         desc &= 0x00f0ff0000000000ULL;
 835         desc |= ((((__u64)base) << 16) & 0x000000ffffff0000ULL) |
 836                 ((((__u64)base) << 32) & 0xff00000000000000ULL) |
 837                 ((((__u64)lim_pages) << 32) & 0x000f000000000000ULL) |
 838                 (lim_pages & 0xffff);
 839         *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS] = desc;
 840
 841         return new_kesp;
 842 }
 843 #else
 844 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
 845 {
 846 }
 847
 848 asmlinkage void __attribute__((weak)) mce_threshold_interrupt(void)
 849 {
 850 }
 851 #endif
 852
 853 /*
 854  * 'math_state_restore()' saves the current math information in the
 855  * old math state array, and gets the new ones from the current task
 856  *
 857  * Careful.. There are problems with IBM-designed IRQ13 behaviour.
 858  * Don't touch unless you *really* know how it works.
 859  *
 860  * Must be called with kernel preemption disabled (in this case,
 861  * local interrupts are disabled at the call-site in entry.S).
 862  */
 863 asmlinkage void math_state_restore(void)
 864 {
 865         struct thread_info *thread = current_thread_info();
 866         struct task_struct *tsk = thread->task;
 867
 868         if (!tsk_used_math(tsk)) {
 869                 local_irq_enable();
 870                 /*
 871                  * does a slab alloc which can sleep
 872                  */
 873                 if (init_fpu(tsk)) {
 874                         /*
 875                          * ran out of memory!
 876                          */
 877                         do_group_exit(SIGKILL);
 878                         return;
 879                 }
 880                 local_irq_disable();
 881         }
 882
 883         clts();                         /* Allow maths ops (or we recurse) */
 884 #ifdef CONFIG_X86_32
 885         restore_fpu(tsk);
 886 #else
 887         /*
 888          * Paranoid restore. send a SIGSEGV if we fail to restore the state.
 889          */
 890         if (unlikely(restore_fpu_checking(tsk))) {
 891                 stts();
 892                 force_sig(SIGSEGV, tsk);
 893                 return;
 894         }
 895 #endif
 896         thread->status |= TS_USEDFPU;   /* So we fnsave on switch_to() */
 897         tsk->fpu_counter++;
 898 }
 899 EXPORT_SYMBOL_GPL(math_state_restore);
 900
 901 #ifndef CONFIG_MATH_EMULATION
 902 asmlinkage void math_emulate(long arg)
 903 {
 904         printk(KERN_EMERG
 905                 "math-emulation not enabled and no coprocessor found.\n");
 906         printk(KERN_EMERG "killing %s.\n", current->comm);
 907         force_sig(SIGFPE, current);
 908         schedule();
 909 }
 910 #endif /* CONFIG_MATH_EMULATION */
 911
 912 dotraplinkage void __kprobes
 913 do_device_not_available(struct pt_regs *regs, long error)
 914 {
 915 #ifdef CONFIG_X86_32
 916         if (read_cr0() & X86_CR0_EM) {
 917                 conditional_sti(regs);
 918                 math_emulate(0);
 919         } else {
 920                 math_state_restore(); /* interrupts still off */
 921                 conditional_sti(regs);
 922         }
 923 #else
 924         math_state_restore();
 925 #endif
 926 }
 927
 928 #ifdef CONFIG_X86_32
 929 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
 930 {
 931         siginfo_t info;
 932         local_irq_enable();
 933
 934         info.si_signo = SIGILL;
 935         info.si_errno = 0;
 936         info.si_code = ILL_BADSTK;
 937         info.si_addr = 0;
 938         if (notify_die(DIE_TRAP, "iret exception",
 939                         regs, error_code, 32, SIGILL) == NOTIFY_STOP)
 940                 return;
 941         do_trap(32, SIGILL, "iret exception", regs, error_code, &info);
 942 }
 943 #endif
 944
 945 void __init trap_init(void)
 946 {
 947         int i;
 948
 949 #ifdef CONFIG_EISA
 950         void __iomem *p = early_ioremap(0x0FFFD9, 4);
 951
 952         if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
 953                 EISA_bus = 1;
 954         early_iounmap(p, 4);
 955 #endif
 956
 957         set_intr_gate(0, &divide_error);
 958         set_intr_gate_ist(1, &debug, DEBUG_STACK);
 959         set_intr_gate_ist(2, &nmi, NMI_STACK);
 960         /* int3 can be called from all */
 961         set_system_intr_gate_ist(3, &int3, DEBUG_STACK);
 962         /* int4 can be called from all */
 963         set_system_intr_gate(4, &overflow);
 964         set_intr_gate(5, &bounds);
 965         set_intr_gate(6, &invalid_op);
 966         set_intr_gate(7, &device_not_available);
 967 #ifdef CONFIG_X86_32
 968         set_task_gate(8, GDT_ENTRY_DOUBLEFAULT_TSS);
 969 #else
 970         set_intr_gate_ist(8, &double_fault, DOUBLEFAULT_STACK);
 971 #endif
 972         set_intr_gate(9, &coprocessor_segment_overrun);
 973         set_intr_gate(10, &invalid_TSS);
 974         set_intr_gate(11, &segment_not_present);
 975         set_intr_gate_ist(12, &stack_segment, STACKFAULT_STACK);
 976         set_intr_gate(13, &general_protection);
 977         set_intr_gate(14, &page_fault);
 978         set_intr_gate(15, &spurious_interrupt_bug);
 979         set_intr_gate(16, &coprocessor_error);
 980         set_intr_gate(17, &alignment_check);
 981 #ifdef CONFIG_X86_MCE
 982         set_intr_gate_ist(18, &machine_check, MCE_STACK);
 983 #endif
 984         set_intr_gate(19, &simd_coprocessor_error);
 985
 986 #ifdef CONFIG_IA32_EMULATION
 987         set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
 988 #endif
 989
 990 #ifdef CONFIG_X86_32
 991         if (cpu_has_fxsr) {
 992                 printk(KERN_INFO "Enabling fast FPU save and restore... ");
 993                 set_in_cr4(X86_CR4_OSFXSR);
 994                 printk("done.\n");
 995         }
 996         if (cpu_has_xmm) {
 997                 printk(KERN_INFO
 998                         "Enabling unmasked SIMD FPU exception support... ");
 999                 set_in_cr4(X86_CR4_OSXMMEXCPT);
1000                 printk("done.\n");
1001         }
1002
1003         set_system_trap_gate(SYSCALL_VECTOR, &system_call);
1004 #endif
1005
1006         /* Reserve all the builtin and the syscall vector: */
1007         for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
1008                 set_bit(i, used_vectors);
1009
1010 #ifdef CONFIG_X86_64
1011         set_bit(IA32_SYSCALL_VECTOR, used_vectors);
1012 #else
1013         set_bit(SYSCALL_VECTOR, used_vectors);
1014 #endif
1015         /*
1016          * Should be a barrier for any external CPU state:
1017          */
1018         cpu_init();
1019
1020 #ifdef CONFIG_X86_32
1021         trap_init_hook();
1022 #endif
1023 }