arch/x86/kernel/traps_32.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  * 'Traps.c' handles hardware traps and faults after we have saved some
  11  * state in 'asm.s'.
  12  */
  13 #include <linux/interrupt.h>
  14 #include <linux/kallsyms.h>
  15 #include <linux/spinlock.h>
  16 #include <linux/highmem.h>
  17 #include <linux/kprobes.h>
  18 #include <linux/uaccess.h>
  19 #include <linux/utsname.h>
  20 #include <linux/kdebug.h>
  21 #include <linux/kernel.h>
  22 #include <linux/module.h>
  23 #include <linux/ptrace.h>
  24 #include <linux/string.h>
  25 #include <linux/unwind.h>
  26 #include <linux/delay.h>
  27 #include <linux/errno.h>
  28 #include <linux/kexec.h>
  29 #include <linux/sched.h>
  30 #include <linux/timer.h>
  31 #include <linux/init.h>
  32 #include <linux/bug.h>
  33 #include <linux/nmi.h>
  34 #include <linux/mm.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/processor-flags.h>
  50 #include <asm/arch_hooks.h>
  51 #include <asm/stacktrace.h>
  52 #include <asm/processor.h>
  53 #include <asm/debugreg.h>
  54 #include <asm/atomic.h>
  55 #include <asm/system.h>
  56 #include <asm/unwind.h>
  57 #include <asm/desc.h>
  58 #include <asm/i387.h>
  59 #include <asm/nmi.h>
  60 #include <asm/smp.h>
  61 #include <asm/io.h>
  62 #include <asm/traps.h>
  63
  64 #include "mach_traps.h"
  65 #include "cpu/mcheck/mce.h"
  66
  67 DECLARE_BITMAP(used_vectors, NR_VECTORS);
  68 EXPORT_SYMBOL_GPL(used_vectors);
  69
  70 asmlinkage int system_call(void);
  71
  72 /* Do we ignore FPU interrupts ? */
  73 char ignore_fpu_irq;
  74
  75 /*
  76  * The IDT has to be page-aligned to simplify the Pentium
  77  * F0 0F bug workaround.. We have a special link segment
  78  * for this.
  79  */
  80 gate_desc idt_table[256]
  81         __attribute__((__section__(".data.idt"))) = { { { { 0, 0 } } }, };
  82
  83 static int ignore_nmis;
  84
  85 static inline void conditional_sti(struct pt_regs *regs)
  86 {
  87         if (regs->flags & X86_EFLAGS_IF)
  88                 local_irq_enable();
  89 }
  90
  91 static inline void preempt_conditional_sti(struct pt_regs *regs)
  92 {
  93         inc_preempt_count();
  94         if (regs->flags & X86_EFLAGS_IF)
  95                 local_irq_enable();
  96 }
  97
  98 static inline void preempt_conditional_cli(struct pt_regs *regs)
  99 {
 100         if (regs->flags & X86_EFLAGS_IF)
 101                 local_irq_disable();
 102         dec_preempt_count();
 103 }
 104
 105 static inline void
 106 die_if_kernel(const char *str, struct pt_regs *regs, long err)
 107 {
 108         if (!user_mode_vm(regs))
 109                 die(str, regs, err);
 110 }
 111
 112 /*
 113  * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
 114  * invalid offset set (the LAZY one) and the faulting thread has
 115  * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS,
 116  * we set the offset field correctly and return 1.
 117  */
 118 static int lazy_iobitmap_copy(void)
 119 {
 120         struct thread_struct *thread;
 121         struct tss_struct *tss;
 122         int cpu;
 123
 124         cpu = get_cpu();
 125         tss = &per_cpu(init_tss, cpu);
 126         thread = &current->thread;
 127
 128         if (tss->x86_tss.io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
 129             thread->io_bitmap_ptr) {
 130                 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
 131                        thread->io_bitmap_max);
 132                 /*
 133                  * If the previously set map was extending to higher ports
 134                  * than the current one, pad extra space with 0xff (no access).
 135                  */
 136                 if (thread->io_bitmap_max < tss->io_bitmap_max) {
 137                         memset((char *) tss->io_bitmap +
 138                                 thread->io_bitmap_max, 0xff,
 139                                 tss->io_bitmap_max - thread->io_bitmap_max);
 140                 }
 141                 tss->io_bitmap_max = thread->io_bitmap_max;
 142                 tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
 143                 tss->io_bitmap_owner = thread;
 144                 put_cpu();
 145
 146                 return 1;
 147         }
 148         put_cpu();
 149
 150         return 0;
 151 }
 152
 153 static void __kprobes
 154 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
 155         long error_code, siginfo_t *info)
 156 {
 157         struct task_struct *tsk = current;
 158
 159         if (regs->flags & X86_VM_MASK) {
 160                 /*
 161                  * traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
 162                  * On nmi (interrupt 2), do_trap should not be called.
 163                  */
 164                 if (trapnr < 6)
 165                         goto vm86_trap;
 166                 goto trap_signal;
 167         }
 168
 169         if (!user_mode(regs))
 170                 goto kernel_trap;
 171
 172 trap_signal:
 173         /*
 174          * We want error_code and trap_no set for userspace faults and
 175          * kernelspace faults which result in die(), but not
 176          * kernelspace faults which are fixed up.  die() gives the
 177          * process no chance to handle the signal and notice the
 178          * kernel fault information, so that won't result in polluting
 179          * the information about previously queued, but not yet
 180          * delivered, faults.  See also do_general_protection below.
 181          */
 182         tsk->thread.error_code = error_code;
 183         tsk->thread.trap_no = trapnr;
 184
 185         if (info)
 186                 force_sig_info(signr, info, tsk);
 187         else
 188                 force_sig(signr, tsk);
 189         return;
 190
 191 kernel_trap:
 192         if (!fixup_exception(regs)) {
 193                 tsk->thread.error_code = error_code;
 194                 tsk->thread.trap_no = trapnr;
 195                 die(str, regs, error_code);
 196         }
 197         return;
 198
 199 vm86_trap:
 200         if (handle_vm86_trap((struct kernel_vm86_regs *) regs,
 201                                                 error_code, trapnr))
 202                 goto trap_signal;
 203         return;
 204 }
 205
 206 #define DO_ERROR(trapnr, signr, str, name)                              \
 207 dotraplinkage void do_##name(struct pt_regs *regs, long error_code)     \
 208 {                                                                       \
 209         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr)  \
 210                                                         == NOTIFY_STOP) \
 211                 return;                                                 \
 212         conditional_sti(regs);                                          \
 213         do_trap(trapnr, signr, str, regs, error_code, NULL);            \
 214 }
 215
 216 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr)         \
 217 dotraplinkage void do_##name(struct pt_regs *regs, long error_code)     \
 218 {                                                                       \
 219         siginfo_t info;                                                 \
 220         info.si_signo = signr;                                          \
 221         info.si_errno = 0;                                              \
 222         info.si_code = sicode;                                          \
 223         info.si_addr = (void __user *)siaddr;                           \
 224         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr)  \
 225                                                         == NOTIFY_STOP) \
 226                 return;                                                 \
 227         conditional_sti(regs);                                          \
 228         do_trap(trapnr, signr, str, regs, error_code, &info);           \
 229 }
 230
 231 DO_ERROR_INFO(0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip)
 232 DO_ERROR(4, SIGSEGV, "overflow", overflow)
 233 DO_ERROR(5, SIGSEGV, "bounds", bounds)
 234 DO_ERROR_INFO(6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip)
 235 DO_ERROR(9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
 236 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
 237 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
 238 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
 239 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
 240
 241 dotraplinkage void __kprobes
 242 do_general_protection(struct pt_regs *regs, long error_code)
 243 {
 244         struct task_struct *tsk;
 245
 246         conditional_sti(regs);
 247
 248         if (lazy_iobitmap_copy()) {
 249                 /* restart the faulting instruction */
 250                 return;
 251         }
 252
 253         if (regs->flags & X86_VM_MASK)
 254                 goto gp_in_vm86;
 255
 256         tsk = current;
 257         if (!user_mode(regs))
 258                 goto gp_in_kernel;
 259
 260         tsk->thread.error_code = error_code;
 261         tsk->thread.trap_no = 13;
 262
 263         if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
 264                         printk_ratelimit()) {
 265                 printk(KERN_INFO
 266                         "%s[%d] general protection ip:%lx sp:%lx error:%lx",
 267                         tsk->comm, task_pid_nr(tsk),
 268                         regs->ip, regs->sp, error_code);
 269                 print_vma_addr(" in ", regs->ip);
 270                 printk("\n");
 271         }
 272
 273         force_sig(SIGSEGV, tsk);
 274         return;
 275
 276 gp_in_vm86:
 277         local_irq_enable();
 278         handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
 279         return;
 280
 281 gp_in_kernel:
 282         if (fixup_exception(regs))
 283                 return;
 284
 285         tsk->thread.error_code = error_code;
 286         tsk->thread.trap_no = 13;
 287         if (notify_die(DIE_GPF, "general protection fault", regs,
 288                                 error_code, 13, SIGSEGV) == NOTIFY_STOP)
 289                 return;
 290         die("general protection fault", regs, error_code);
 291 }
 292
 293 static notrace __kprobes void
 294 mem_parity_error(unsigned char reason, struct pt_regs *regs)
 295 {
 296         printk(KERN_EMERG
 297                 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
 298                         reason, smp_processor_id());
 299
 300         printk(KERN_EMERG
 301                 "You have some hardware problem, likely on the PCI bus.\n");
 302
 303 #if defined(CONFIG_EDAC)
 304         if (edac_handler_set()) {
 305                 edac_atomic_assert_error();
 306                 return;
 307         }
 308 #endif
 309
 310         if (panic_on_unrecovered_nmi)
 311                 panic("NMI: Not continuing");
 312
 313         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 314
 315         /* Clear and disable the memory parity error line. */
 316         clear_mem_error(reason);
 317 }
 318
 319 static notrace __kprobes void
 320 io_check_error(unsigned char reason, struct pt_regs *regs)
 321 {
 322         unsigned long i;
 323
 324         printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
 325         show_registers(regs);
 326
 327         /* Re-enable the IOCK line, wait for a few seconds */
 328         reason = (reason & 0xf) | 8;
 329         outb(reason, 0x61);
 330
 331         i = 2000;
 332         while (--i)
 333                 udelay(1000);
 334
 335         reason &= ~8;
 336         outb(reason, 0x61);
 337 }
 338
 339 static notrace __kprobes void
 340 unknown_nmi_error(unsigned char reason, struct pt_regs *regs)
 341 {
 342         if (notify_die(DIE_NMIUNKNOWN, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
 343                 return;
 344 #ifdef CONFIG_MCA
 345         /*
 346          * Might actually be able to figure out what the guilty party
 347          * is:
 348          */
 349         if (MCA_bus) {
 350                 mca_handle_nmi();
 351                 return;
 352         }
 353 #endif
 354         printk(KERN_EMERG
 355                 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
 356                         reason, smp_processor_id());
 357
 358         printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
 359         if (panic_on_unrecovered_nmi)
 360                 panic("NMI: Not continuing");
 361
 362         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 363 }
 364
 365 static DEFINE_SPINLOCK(nmi_print_lock);
 366
 367 void notrace __kprobes die_nmi(char *str, struct pt_regs *regs, int do_panic)
 368 {
 369         if (notify_die(DIE_NMIWATCHDOG, str, regs, 0, 2, SIGINT) == NOTIFY_STOP)
 370                 return;
 371
 372         spin_lock(&nmi_print_lock);
 373         /*
 374         * We are in trouble anyway, lets at least try
 375         * to get a message out:
 376         */
 377         bust_spinlocks(1);
 378         printk(KERN_EMERG "%s", str);
 379         printk(" on CPU%d, ip %08lx, registers:\n",
 380                 smp_processor_id(), regs->ip);
 381         show_registers(regs);
 382         if (do_panic)
 383                 panic("Non maskable interrupt");
 384         console_silent();
 385         spin_unlock(&nmi_print_lock);
 386         bust_spinlocks(0);
 387
 388         /*
 389          * If we are in kernel we are probably nested up pretty bad
 390          * and might aswell get out now while we still can:
 391          */
 392         if (!user_mode_vm(regs)) {
 393                 current->thread.trap_no = 2;
 394                 crash_kexec(regs);
 395         }
 396
 397         do_exit(SIGSEGV);
 398 }
 399
 400 static notrace __kprobes void default_do_nmi(struct pt_regs *regs)
 401 {
 402         unsigned char reason = 0;
 403         int cpu;
 404
 405         cpu = smp_processor_id();
 406
 407         /* Only the BSP gets external NMIs from the system. */
 408         if (!cpu)
 409                 reason = get_nmi_reason();
 410
 411         if (!(reason & 0xc0)) {
 412                 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
 413                                                                 == NOTIFY_STOP)
 414                         return;
 415 #ifdef CONFIG_X86_LOCAL_APIC
 416                 /*
 417                  * Ok, so this is none of the documented NMI sources,
 418                  * so it must be the NMI watchdog.
 419                  */
 420                 if (nmi_watchdog_tick(regs, reason))
 421                         return;
 422                 if (!do_nmi_callback(regs, cpu))
 423                         unknown_nmi_error(reason, regs);
 424 #else
 425                 unknown_nmi_error(reason, regs);
 426 #endif
 427
 428                 return;
 429         }
 430         if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
 431                 return;
 432
 433         /* AK: following checks seem to be broken on modern chipsets. FIXME */
 434         if (reason & 0x80)
 435                 mem_parity_error(reason, regs);
 436         if (reason & 0x40)
 437                 io_check_error(reason, regs);
 438         /*
 439          * Reassert NMI in case it became active meanwhile
 440          * as it's edge-triggered:
 441          */
 442         reassert_nmi();
 443 }
 444
 445 dotraplinkage notrace __kprobes void
 446 do_nmi(struct pt_regs *regs, long error_code)
 447 {
 448         int cpu;
 449
 450         nmi_enter();
 451
 452         cpu = smp_processor_id();
 453
 454         ++nmi_count(cpu);
 455
 456         if (!ignore_nmis)
 457                 default_do_nmi(regs);
 458
 459         nmi_exit();
 460 }
 461
 462 void stop_nmi(void)
 463 {
 464         acpi_nmi_disable();
 465         ignore_nmis++;
 466 }
 467
 468 void restart_nmi(void)
 469 {
 470         ignore_nmis--;
 471         acpi_nmi_enable();
 472 }
 473
 474 dotraplinkage void __kprobes do_int3(struct pt_regs *regs, long error_code)
 475 {
 476 #ifdef CONFIG_KPROBES
 477         if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
 478                         == NOTIFY_STOP)
 479                 return;
 480 #else
 481         if (notify_die(DIE_TRAP, "int3", regs, error_code, 3, SIGTRAP)
 482                         == NOTIFY_STOP)
 483                 return;
 484 #endif
 485
 486         preempt_conditional_sti(regs);
 487         do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
 488         preempt_conditional_cli(regs);
 489 }
 490
 491 /*
 492  * Our handling of the processor debug registers is non-trivial.
 493  * We do not clear them on entry and exit from the kernel. Therefore
 494  * it is possible to get a watchpoint trap here from inside the kernel.
 495  * However, the code in ./ptrace.c has ensured that the user can
 496  * only set watchpoints on userspace addresses. Therefore the in-kernel
 497  * watchpoint trap can only occur in code which is reading/writing
 498  * from user space. Such code must not hold kernel locks (since it
 499  * can equally take a page fault), therefore it is safe to call
 500  * force_sig_info even though that claims and releases locks.
 501  *
 502  * Code in ./signal.c ensures that the debug control register
 503  * is restored before we deliver any signal, and therefore that
 504  * user code runs with the correct debug control register even though
 505  * we clear it here.
 506  *
 507  * Being careful here means that we don't have to be as careful in a
 508  * lot of more complicated places (task switching can be a bit lazy
 509  * about restoring all the debug state, and ptrace doesn't have to
 510  * find every occurrence of the TF bit that could be saved away even
 511  * by user code)
 512  */
 513 dotraplinkage void __kprobes do_debug(struct pt_regs *regs, long error_code)
 514 {
 515         struct task_struct *tsk = current;
 516         unsigned long condition;
 517         int si_code;
 518
 519         get_debugreg(condition, 6);
 520
 521         /*
 522          * The processor cleared BTF, so don't mark that we need it set.
 523          */
 524         clear_tsk_thread_flag(tsk, TIF_DEBUGCTLMSR);
 525         tsk->thread.debugctlmsr = 0;
 526
 527         if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
 528                                                 SIGTRAP) == NOTIFY_STOP)
 529                 return;
 530
 531         /* It's safe to allow irq's after DR6 has been saved */
 532         preempt_conditional_sti(regs);
 533
 534         /* Mask out spurious debug traps due to lazy DR7 setting */
 535         if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
 536                 if (!tsk->thread.debugreg7)
 537                         goto clear_dr7;
 538         }
 539
 540         if (regs->flags & X86_VM_MASK)
 541                 goto debug_vm86;
 542
 543         /* Save debug status register where ptrace can see it */
 544         tsk->thread.debugreg6 = condition;
 545
 546         /*
 547          * Single-stepping through TF: make sure we ignore any events in
 548          * kernel space (but re-enable TF when returning to user mode).
 549          */
 550         if (condition & DR_STEP) {
 551                 if (!user_mode(regs))
 552                         goto clear_TF_reenable;
 553         }
 554
 555         si_code = get_si_code(condition);
 556         /* Ok, finally something we can handle */
 557         send_sigtrap(tsk, regs, error_code, si_code);
 558
 559         /*
 560          * Disable additional traps. They'll be re-enabled when
 561          * the signal is delivered.
 562          */
 563 clear_dr7:
 564         set_debugreg(0, 7);
 565         preempt_conditional_cli(regs);
 566         return;
 567
 568 debug_vm86:
 569         handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
 570         preempt_conditional_cli(regs);
 571         return;
 572
 573 clear_TF_reenable:
 574         set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
 575         regs->flags &= ~X86_EFLAGS_TF;
 576         preempt_conditional_cli(regs);
 577         return;
 578 }
 579
 580 /*
 581  * Note that we play around with the 'TS' bit in an attempt to get
 582  * the correct behaviour even in the presence of the asynchronous
 583  * IRQ13 behaviour
 584  */
 585 void math_error(void __user *ip)
 586 {
 587         struct task_struct *task;
 588         siginfo_t info;
 589         unsigned short cwd, swd;
 590
 591         /*
 592          * Save the info for the exception handler and clear the error.
 593          */
 594         task = current;
 595         save_init_fpu(task);
 596         task->thread.trap_no = 16;
 597         task->thread.error_code = 0;
 598         info.si_signo = SIGFPE;
 599         info.si_errno = 0;
 600         info.si_code = __SI_FAULT;
 601         info.si_addr = ip;
 602         /*
 603          * (~cwd & swd) will mask out exceptions that are not set to unmasked
 604          * status.  0x3f is the exception bits in these regs, 0x200 is the
 605          * C1 reg you need in case of a stack fault, 0x040 is the stack
 606          * fault bit.  We should only be taking one exception at a time,
 607          * so if this combination doesn't produce any single exception,
 608          * then we have a bad program that isn't synchronizing its FPU usage
 609          * and it will suffer the consequences since we won't be able to
 610          * fully reproduce the context of the exception
 611          */
 612         cwd = get_fpu_cwd(task);
 613         swd = get_fpu_swd(task);
 614         switch (swd & ~cwd & 0x3f) {
 615         case 0x000: /* No unmasked exception */
 616                 return;
 617         default: /* Multiple exceptions */
 618                 break;
 619         case 0x001: /* Invalid Op */
 620                 /*
 621                  * swd & 0x240 == 0x040: Stack Underflow
 622                  * swd & 0x240 == 0x240: Stack Overflow
 623                  * User must clear the SF bit (0x40) if set
 624                  */
 625                 info.si_code = FPE_FLTINV;
 626                 break;
 627         case 0x002: /* Denormalize */
 628         case 0x010: /* Underflow */
 629                 info.si_code = FPE_FLTUND;
 630                 break;
 631         case 0x004: /* Zero Divide */
 632                 info.si_code = FPE_FLTDIV;
 633                 break;
 634         case 0x008: /* Overflow */
 635                 info.si_code = FPE_FLTOVF;
 636                 break;
 637         case 0x020: /* Precision */
 638                 info.si_code = FPE_FLTRES;
 639                 break;
 640         }
 641         force_sig_info(SIGFPE, &info, task);
 642 }
 643
 644 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
 645 {
 646         conditional_sti(regs);
 647         ignore_fpu_irq = 1;
 648         math_error((void __user *)regs->ip);
 649 }
 650
 651 static void simd_math_error(void __user *ip)
 652 {
 653         struct task_struct *task;
 654         siginfo_t info;
 655         unsigned short mxcsr;
 656
 657         /*
 658          * Save the info for the exception handler and clear the error.
 659          */
 660         task = current;
 661         save_init_fpu(task);
 662         task->thread.trap_no = 19;
 663         task->thread.error_code = 0;
 664         info.si_signo = SIGFPE;
 665         info.si_errno = 0;
 666         info.si_code = __SI_FAULT;
 667         info.si_addr = ip;
 668         /*
 669          * The SIMD FPU exceptions are handled a little differently, as there
 670          * is only a single status/control register.  Thus, to determine which
 671          * unmasked exception was caught we must mask the exception mask bits
 672          * at 0x1f80, and then use these to mask the exception bits at 0x3f.
 673          */
 674         mxcsr = get_fpu_mxcsr(task);
 675         switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
 676         case 0x000:
 677         default:
 678                 break;
 679         case 0x001: /* Invalid Op */
 680                 info.si_code = FPE_FLTINV;
 681                 break;
 682         case 0x002: /* Denormalize */
 683         case 0x010: /* Underflow */
 684                 info.si_code = FPE_FLTUND;
 685                 break;
 686         case 0x004: /* Zero Divide */
 687                 info.si_code = FPE_FLTDIV;
 688                 break;
 689         case 0x008: /* Overflow */
 690                 info.si_code = FPE_FLTOVF;
 691                 break;
 692         case 0x020: /* Precision */
 693                 info.si_code = FPE_FLTRES;
 694                 break;
 695         }
 696         force_sig_info(SIGFPE, &info, task);
 697 }
 698
 699 dotraplinkage void
 700 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
 701 {
 702         conditional_sti(regs);
 703
 704         if (cpu_has_xmm) {
 705                 /* Handle SIMD FPU exceptions on PIII+ processors. */
 706                 ignore_fpu_irq = 1;
 707                 simd_math_error((void __user *)regs->ip);
 708                 return;
 709         }
 710         /*
 711          * Handle strange cache flush from user space exception
 712          * in all other cases.  This is undocumented behaviour.
 713          */
 714         if (regs->flags & X86_VM_MASK) {
 715                 handle_vm86_fault((struct kernel_vm86_regs *)regs, error_code);
 716                 return;
 717         }
 718         current->thread.trap_no = 19;
 719         current->thread.error_code = error_code;
 720         die_if_kernel("cache flush denied", regs, error_code);
 721         force_sig(SIGSEGV, current);
 722 }
 723
 724 dotraplinkage void
 725 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
 726 {
 727         conditional_sti(regs);
 728 #if 0
 729         /* No need to warn about this any longer. */
 730         printk(KERN_INFO "Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
 731 #endif
 732 }
 733
 734 unsigned long patch_espfix_desc(unsigned long uesp, unsigned long kesp)
 735 {
 736         struct desc_struct *gdt = get_cpu_gdt_table(smp_processor_id());
 737         unsigned long base = (kesp - uesp) & -THREAD_SIZE;
 738         unsigned long new_kesp = kesp - base;
 739         unsigned long lim_pages = (new_kesp | (THREAD_SIZE - 1)) >> PAGE_SHIFT;
 740         __u64 desc = *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS];
 741
 742         /* Set up base for espfix segment */
 743         desc &= 0x00f0ff0000000000ULL;
 744         desc |= ((((__u64)base) << 16) & 0x000000ffffff0000ULL) |
 745                 ((((__u64)base) << 32) & 0xff00000000000000ULL) |
 746                 ((((__u64)lim_pages) << 32) & 0x000f000000000000ULL) |
 747                 (lim_pages & 0xffff);
 748         *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS] = desc;
 749
 750         return new_kesp;
 751 }
 752
 753 /*
 754  * 'math_state_restore()' saves the current math information in the
 755  * old math state array, and gets the new ones from the current task
 756  *
 757  * Careful.. There are problems with IBM-designed IRQ13 behaviour.
 758  * Don't touch unless you *really* know how it works.
 759  *
 760  * Must be called with kernel preemption disabled (in this case,
 761  * local interrupts are disabled at the call-site in entry.S).
 762  */
 763 asmlinkage void math_state_restore(void)
 764 {
 765         struct thread_info *thread = current_thread_info();
 766         struct task_struct *tsk = thread->task;
 767
 768         if (!tsk_used_math(tsk)) {
 769                 local_irq_enable();
 770                 /*
 771                  * does a slab alloc which can sleep
 772                  */
 773                 if (init_fpu(tsk)) {
 774                         /*
 775                          * ran out of memory!
 776                          */
 777                         do_group_exit(SIGKILL);
 778                         return;
 779                 }
 780                 local_irq_disable();
 781         }
 782
 783         clts();                         /* Allow maths ops (or we recurse) */
 784         restore_fpu(tsk);
 785         thread->status |= TS_USEDFPU;   /* So we fnsave on switch_to() */
 786         tsk->fpu_counter++;
 787 }
 788 EXPORT_SYMBOL_GPL(math_state_restore);
 789
 790 #ifndef CONFIG_MATH_EMULATION
 791
 792 asmlinkage void math_emulate(long arg)
 793 {
 794         printk(KERN_EMERG
 795                 "math-emulation not enabled and no coprocessor found.\n");
 796         printk(KERN_EMERG "killing %s.\n", current->comm);
 797         force_sig(SIGFPE, current);
 798         schedule();
 799 }
 800
 801 #endif /* CONFIG_MATH_EMULATION */
 802
 803 dotraplinkage void __kprobes
 804 do_device_not_available(struct pt_regs *regs, long error)
 805 {
 806         if (read_cr0() & X86_CR0_EM) {
 807                 conditional_sti(regs);
 808                 math_emulate(0);
 809         } else {
 810                 math_state_restore(); /* interrupts still off */
 811                 conditional_sti(regs);
 812         }
 813 }
 814
 815 #ifdef CONFIG_X86_MCE
 816 dotraplinkage void __kprobes do_machine_check(struct pt_regs *regs, long error)
 817 {
 818         conditional_sti(regs);
 819         machine_check_vector(regs, error);
 820 }
 821 #endif
 822
 823 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
 824 {
 825         siginfo_t info;
 826         local_irq_enable();
 827
 828         info.si_signo = SIGILL;
 829         info.si_errno = 0;
 830         info.si_code = ILL_BADSTK;
 831         info.si_addr = 0;
 832         if (notify_die(DIE_TRAP, "iret exception",
 833                         regs, error_code, 32, SIGILL) == NOTIFY_STOP)
 834                 return;
 835         do_trap(32, SIGILL, "iret exception", regs, error_code, &info);
 836 }
 837
 838 void __init trap_init(void)
 839 {
 840         int i;
 841
 842 #ifdef CONFIG_EISA
 843         void __iomem *p = early_ioremap(0x0FFFD9, 4);
 844
 845         if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
 846                 EISA_bus = 1;
 847         early_iounmap(p, 4);
 848 #endif
 849
 850         set_intr_gate(0, &divide_error);
 851         set_intr_gate_ist(1, &debug, DEBUG_STACK);
 852         set_intr_gate_ist(2, &nmi, NMI_STACK);
 853         /* int3 can be called from all */
 854         set_system_intr_gate_ist(3, &int3, DEBUG_STACK);
 855         /* int4 can be called from all */
 856         set_system_intr_gate(4, &overflow);
 857         set_intr_gate(5, &bounds);
 858         set_intr_gate(6, &invalid_op);
 859         set_intr_gate(7, &device_not_available);
 860         set_task_gate(8, GDT_ENTRY_DOUBLEFAULT_TSS);
 861         set_intr_gate(9, &coprocessor_segment_overrun);
 862         set_intr_gate(10, &invalid_TSS);
 863         set_intr_gate(11, &segment_not_present);
 864         set_intr_gate_ist(12, &stack_segment, STACKFAULT_STACK);
 865         set_intr_gate(13, &general_protection);
 866         set_intr_gate(14, &page_fault);
 867         set_intr_gate(15, &spurious_interrupt_bug);
 868         set_intr_gate(16, &coprocessor_error);
 869         set_intr_gate(17, &alignment_check);
 870 #ifdef CONFIG_X86_MCE
 871         set_intr_gate_ist(18, &machine_check, MCE_STACK);
 872 #endif
 873         set_intr_gate(19, &simd_coprocessor_error);
 874
 875         if (cpu_has_fxsr) {
 876                 printk(KERN_INFO "Enabling fast FPU save and restore... ");
 877                 set_in_cr4(X86_CR4_OSFXSR);
 878                 printk("done.\n");
 879         }
 880         if (cpu_has_xmm) {
 881                 printk(KERN_INFO
 882                         "Enabling unmasked SIMD FPU exception support... ");
 883                 set_in_cr4(X86_CR4_OSXMMEXCPT);
 884                 printk("done.\n");
 885         }
 886
 887         set_system_trap_gate(SYSCALL_VECTOR, &system_call);
 888
 889         /* Reserve all the builtin and the syscall vector: */
 890         for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
 891                 set_bit(i, used_vectors);
 892
 893         set_bit(SYSCALL_VECTOR, used_vectors);
 894
 895         /*
 896          * Should be a barrier for any external CPU state:
 897          */
 898         cpu_init();
 899
 900         trap_init_hook();
 901 }