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/kdebug.h>
  18 #include <linux/kernel.h>
  19 #include <linux/module.h>
  20 #include <linux/ptrace.h>
  21 #include <linux/string.h>
  22 #include <linux/delay.h>
  23 #include <linux/errno.h>
  24 #include <linux/kexec.h>
  25 #include <linux/sched.h>
  26 #include <linux/timer.h>
  27 #include <linux/init.h>
  28 #include <linux/bug.h>
  29 #include <linux/nmi.h>
  30 #include <linux/mm.h>
  31 #include <linux/smp.h>
  32 #include <linux/io.h>
  33
  34 #ifdef CONFIG_EISA
  35 #include <linux/ioport.h>
  36 #include <linux/eisa.h>
  37 #endif
  38
  39 #ifdef CONFIG_MCA
  40 #include <linux/mca.h>
  41 #endif
  42
  43 #if defined(CONFIG_EDAC)
  44 #include <linux/edac.h>
  45 #endif
  46
  47 #include <asm/kmemcheck.h>
  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 #include <asm/mce.h>
  57
  58 #include <asm/mach_traps.h>
  59
  60 #ifdef CONFIG_X86_64
  61 #include <asm/x86_init.h>
  62 #include <asm/pgalloc.h>
  63 #include <asm/proto.h>
  64 #else
  65 #include <asm/processor-flags.h>
  66 #include <asm/setup.h>
  67
  68 asmlinkage int system_call(void);
  69
  70 /* Do we ignore FPU interrupts ? */
  71 char ignore_fpu_irq;
  72
  73 /*
  74  * The IDT has to be page-aligned to simplify the Pentium
  75  * F0 0F bug workaround.
  76  */
  77 gate_desc idt_table[NR_VECTORS] __page_aligned_data = { { { { 0, 0 } } }, };
  78 #endif
  79
  80 DECLARE_BITMAP(used_vectors, NR_VECTORS);
  81 EXPORT_SYMBOL_GPL(used_vectors);
  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 conditional_cli(struct pt_regs *regs)
  99 {
 100         if (regs->flags & X86_EFLAGS_IF)
 101                 local_irq_disable();
 102 }
 103
 104 static inline void preempt_conditional_cli(struct pt_regs *regs)
 105 {
 106         if (regs->flags & X86_EFLAGS_IF)
 107                 local_irq_disable();
 108         dec_preempt_count();
 109 }
 110
 111 static void __kprobes
 112 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
 113         long error_code, siginfo_t *info)
 114 {
 115         struct task_struct *tsk = current;
 116
 117 #ifdef CONFIG_X86_32
 118         if (regs->flags & X86_VM_MASK) {
 119                 /*
 120                  * traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
 121                  * On nmi (interrupt 2), do_trap should not be called.
 122                  */
 123                 if (trapnr < 6)
 124                         goto vm86_trap;
 125                 goto trap_signal;
 126         }
 127 #endif
 128
 129         if (!user_mode(regs))
 130                 goto kernel_trap;
 131
 132 #ifdef CONFIG_X86_32
 133 trap_signal:
 134 #endif
 135         /*
 136          * We want error_code and trap_no set for userspace faults and
 137          * kernelspace faults which result in die(), but not
 138          * kernelspace faults which are fixed up.  die() gives the
 139          * process no chance to handle the signal and notice the
 140          * kernel fault information, so that won't result in polluting
 141          * the information about previously queued, but not yet
 142          * delivered, faults.  See also do_general_protection below.
 143          */
 144         tsk->thread.error_code = error_code;
 145         tsk->thread.trap_no = trapnr;
 146
 147 #ifdef CONFIG_X86_64
 148         if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
 149             printk_ratelimit()) {
 150                 printk(KERN_INFO
 151                        "%s[%d] trap %s ip:%lx sp:%lx error:%lx",
 152                        tsk->comm, tsk->pid, str,
 153                        regs->ip, regs->sp, error_code);
 154                 print_vma_addr(" in ", regs->ip);
 155                 printk("\n");
 156         }
 157 #endif
 158
 159         if (info)
 160                 force_sig_info(signr, info, tsk);
 161         else
 162                 force_sig(signr, tsk);
 163         return;
 164
 165 kernel_trap:
 166         if (!fixup_exception(regs)) {
 167                 tsk->thread.error_code = error_code;
 168                 tsk->thread.trap_no = trapnr;
 169                 die(str, regs, error_code);
 170         }
 171         return;
 172
 173 #ifdef CONFIG_X86_32
 174 vm86_trap:
 175         if (handle_vm86_trap((struct kernel_vm86_regs *) regs,
 176                                                 error_code, trapnr))
 177                 goto trap_signal;
 178         return;
 179 #endif
 180 }
 181
 182 #define DO_ERROR(trapnr, signr, str, name)                              \
 183 dotraplinkage void do_##name(struct pt_regs *regs, long error_code)     \
 184 {                                                                       \
 185         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr)  \
 186                                                         == NOTIFY_STOP) \
 187                 return;                                                 \
 188         conditional_sti(regs);                                          \
 189         do_trap(trapnr, signr, str, regs, error_code, NULL);            \
 190 }
 191
 192 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr)         \
 193 dotraplinkage void do_##name(struct pt_regs *regs, long error_code)     \
 194 {                                                                       \
 195         siginfo_t info;                                                 \
 196         info.si_signo = signr;                                          \
 197         info.si_errno = 0;                                              \
 198         info.si_code = sicode;                                          \
 199         info.si_addr = (void __user *)siaddr;                           \
 200         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr)  \
 201                                                         == NOTIFY_STOP) \
 202                 return;                                                 \
 203         conditional_sti(regs);                                          \
 204         do_trap(trapnr, signr, str, regs, error_code, &info);           \
 205 }
 206
 207 DO_ERROR_INFO(0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip)
 208 DO_ERROR(4, SIGSEGV, "overflow", overflow)
 209 DO_ERROR(5, SIGSEGV, "bounds", bounds)
 210 DO_ERROR_INFO(6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip)
 211 DO_ERROR(9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
 212 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
 213 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
 214 #ifdef CONFIG_X86_32
 215 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
 216 #endif
 217 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
 218
 219 #ifdef CONFIG_X86_64
 220 /* Runs on IST stack */
 221 dotraplinkage void do_stack_segment(struct pt_regs *regs, long error_code)
 222 {
 223         if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
 224                         12, SIGBUS) == NOTIFY_STOP)
 225                 return;
 226         preempt_conditional_sti(regs);
 227         do_trap(12, SIGBUS, "stack segment", regs, error_code, NULL);
 228         preempt_conditional_cli(regs);
 229 }
 230
 231 dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
 232 {
 233         static const char str[] = "double fault";
 234         struct task_struct *tsk = current;
 235
 236         /* Return not checked because double check cannot be ignored */
 237         notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV);
 238
 239         tsk->thread.error_code = error_code;
 240         tsk->thread.trap_no = 8;
 241
 242         /*
 243          * This is always a kernel trap and never fixable (and thus must
 244          * never return).
 245          */
 246         for (;;)
 247                 die(str, regs, error_code);
 248 }
 249 #endif
 250
 251 dotraplinkage void __kprobes
 252 do_general_protection(struct pt_regs *regs, long error_code)
 253 {
 254         struct task_struct *tsk;
 255
 256         conditional_sti(regs);
 257
 258 #ifdef CONFIG_X86_32
 259         if (regs->flags & X86_VM_MASK)
 260                 goto gp_in_vm86;
 261 #endif
 262
 263         tsk = current;
 264         if (!user_mode(regs))
 265                 goto gp_in_kernel;
 266
 267         tsk->thread.error_code = error_code;
 268         tsk->thread.trap_no = 13;
 269
 270         if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
 271                         printk_ratelimit()) {
 272                 printk(KERN_INFO
 273                         "%s[%d] general protection ip:%lx sp:%lx error:%lx",
 274                         tsk->comm, task_pid_nr(tsk),
 275                         regs->ip, regs->sp, error_code);
 276                 print_vma_addr(" in ", regs->ip);
 277                 printk("\n");
 278         }
 279
 280         force_sig(SIGSEGV, tsk);
 281         return;
 282
 283 #ifdef CONFIG_X86_32
 284 gp_in_vm86:
 285         local_irq_enable();
 286         handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
 287         return;
 288 #endif
 289
 290 gp_in_kernel:
 291         if (fixup_exception(regs))
 292                 return;
 293
 294         tsk->thread.error_code = error_code;
 295         tsk->thread.trap_no = 13;
 296         if (notify_die(DIE_GPF, "general protection fault", regs,
 297                                 error_code, 13, SIGSEGV) == NOTIFY_STOP)
 298                 return;
 299         die("general protection fault", regs, error_code);
 300 }
 301
 302 static notrace __kprobes void
 303 mem_parity_error(unsigned char reason, struct pt_regs *regs)
 304 {
 305         printk(KERN_EMERG
 306                 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
 307                         reason, smp_processor_id());
 308
 309         printk(KERN_EMERG
 310                 "You have some hardware problem, likely on the PCI bus.\n");
 311
 312 #if defined(CONFIG_EDAC)
 313         if (edac_handler_set()) {
 314                 edac_atomic_assert_error();
 315                 return;
 316         }
 317 #endif
 318
 319         if (panic_on_unrecovered_nmi)
 320                 panic("NMI: Not continuing");
 321
 322         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 323
 324         /* Clear and disable the memory parity error line. */
 325         reason = (reason & 0xf) | 4;
 326         outb(reason, 0x61);
 327 }
 328
 329 static notrace __kprobes void
 330 io_check_error(unsigned char reason, struct pt_regs *regs)
 331 {
 332         unsigned long i;
 333
 334         printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
 335         show_registers(regs);
 336
 337         if (panic_on_io_nmi)
 338                 panic("NMI IOCK error: Not continuing");
 339
 340         /* Re-enable the IOCK line, wait for a few seconds */
 341         reason = (reason & 0xf) | 8;
 342         outb(reason, 0x61);
 343
 344         i = 2000;
 345         while (--i)
 346                 udelay(1000);
 347
 348         reason &= ~8;
 349         outb(reason, 0x61);
 350 }
 351
 352 static notrace __kprobes void
 353 unknown_nmi_error(unsigned char reason, struct pt_regs *regs)
 354 {
 355         if (notify_die(DIE_NMIUNKNOWN, "nmi", regs, reason, 2, SIGINT) ==
 356                         NOTIFY_STOP)
 357                 return;
 358 #ifdef CONFIG_MCA
 359         /*
 360          * Might actually be able to figure out what the guilty party
 361          * is:
 362          */
 363         if (MCA_bus) {
 364                 mca_handle_nmi();
 365                 return;
 366         }
 367 #endif
 368         printk(KERN_EMERG
 369                 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
 370                         reason, smp_processor_id());
 371
 372         printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
 373         if (panic_on_unrecovered_nmi)
 374                 panic("NMI: Not continuing");
 375
 376         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 377 }
 378
 379 static notrace __kprobes void default_do_nmi(struct pt_regs *regs)
 380 {
 381         unsigned char reason = 0;
 382         int cpu;
 383
 384         cpu = smp_processor_id();
 385
 386         /* Only the BSP gets external NMIs from the system. */
 387         if (!cpu)
 388                 reason = get_nmi_reason();
 389
 390         if (!(reason & 0xc0)) {
 391                 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
 392                                                                 == NOTIFY_STOP)
 393                         return;
 394 #ifdef CONFIG_X86_LOCAL_APIC
 395                 /*
 396                  * Ok, so this is none of the documented NMI sources,
 397                  * so it must be the NMI watchdog.
 398                  */
 399                 if (nmi_watchdog_tick(regs, reason))
 400                         return;
 401                 if (!do_nmi_callback(regs, cpu))
 402                         unknown_nmi_error(reason, regs);
 403 #else
 404                 unknown_nmi_error(reason, regs);
 405 #endif
 406
 407                 return;
 408         }
 409         if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
 410                 return;
 411
 412         /* AK: following checks seem to be broken on modern chipsets. FIXME */
 413         if (reason & 0x80)
 414                 mem_parity_error(reason, regs);
 415         if (reason & 0x40)
 416                 io_check_error(reason, regs);
 417 #ifdef CONFIG_X86_32
 418         /*
 419          * Reassert NMI in case it became active meanwhile
 420          * as it's edge-triggered:
 421          */
 422         reassert_nmi();
 423 #endif
 424 }
 425
 426 dotraplinkage notrace __kprobes void
 427 do_nmi(struct pt_regs *regs, long error_code)
 428 {
 429         nmi_enter();
 430
 431         inc_irq_stat(__nmi_count);
 432
 433         if (!ignore_nmis)
 434                 default_do_nmi(regs);
 435
 436         nmi_exit();
 437 }
 438
 439 void stop_nmi(void)
 440 {
 441         acpi_nmi_disable();
 442         ignore_nmis++;
 443 }
 444
 445 void restart_nmi(void)
 446 {
 447         ignore_nmis--;
 448         acpi_nmi_enable();
 449 }
 450
 451 /* May run on IST stack. */
 452 dotraplinkage void __kprobes do_int3(struct pt_regs *regs, long error_code)
 453 {
 454 #ifdef CONFIG_KPROBES
 455         if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
 456                         == NOTIFY_STOP)
 457                 return;
 458 #else
 459         if (notify_die(DIE_TRAP, "int3", regs, error_code, 3, SIGTRAP)
 460                         == NOTIFY_STOP)
 461                 return;
 462 #endif
 463
 464         preempt_conditional_sti(regs);
 465         do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
 466         preempt_conditional_cli(regs);
 467 }
 468
 469 #ifdef CONFIG_X86_64
 470 /*
 471  * Help handler running on IST stack to switch back to user stack
 472  * for scheduling or signal handling. The actual stack switch is done in
 473  * entry.S
 474  */
 475 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
 476 {
 477         struct pt_regs *regs = eregs;
 478         /* Did already sync */
 479         if (eregs == (struct pt_regs *)eregs->sp)
 480                 ;
 481         /* Exception from user space */
 482         else if (user_mode(eregs))
 483                 regs = task_pt_regs(current);
 484         /*
 485          * Exception from kernel and interrupts are enabled. Move to
 486          * kernel process stack.
 487          */
 488         else if (eregs->flags & X86_EFLAGS_IF)
 489                 regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
 490         if (eregs != regs)
 491                 *regs = *eregs;
 492         return regs;
 493 }
 494 #endif
 495
 496 /*
 497  * Our handling of the processor debug registers is non-trivial.
 498  * We do not clear them on entry and exit from the kernel. Therefore
 499  * it is possible to get a watchpoint trap here from inside the kernel.
 500  * However, the code in ./ptrace.c has ensured that the user can
 501  * only set watchpoints on userspace addresses. Therefore the in-kernel
 502  * watchpoint trap can only occur in code which is reading/writing
 503  * from user space. Such code must not hold kernel locks (since it
 504  * can equally take a page fault), therefore it is safe to call
 505  * force_sig_info even though that claims and releases locks.
 506  *
 507  * Code in ./signal.c ensures that the debug control register
 508  * is restored before we deliver any signal, and therefore that
 509  * user code runs with the correct debug control register even though
 510  * we clear it here.
 511  *
 512  * Being careful here means that we don't have to be as careful in a
 513  * lot of more complicated places (task switching can be a bit lazy
 514  * about restoring all the debug state, and ptrace doesn't have to
 515  * find every occurrence of the TF bit that could be saved away even
 516  * by user code)
 517  *
 518  * May run on IST stack.
 519  */
 520 dotraplinkage void __kprobes do_debug(struct pt_regs *regs, long error_code)
 521 {
 522         struct task_struct *tsk = current;
 523         unsigned long dr6;
 524         int si_code;
 525
 526         get_debugreg(dr6, 6);
 527
 528         /* Filter out all the reserved bits which are preset to 1 */
 529         dr6 &= ~DR6_RESERVED;
 530
 531         /* Catch kmemcheck conditions first of all! */
 532         if ((dr6 & DR_STEP) && kmemcheck_trap(regs))
 533                 return;
 534
 535         /* DR6 may or may not be cleared by the CPU */
 536         set_debugreg(0, 6);
 537
 538         /*
 539          * The processor cleared BTF, so don't mark that we need it set.
 540          */
 541         clear_tsk_thread_flag(tsk, TIF_BLOCKSTEP);
 542
 543         /* Store the virtualized DR6 value */
 544         tsk->thread.debugreg6 = dr6;
 545
 546         if (notify_die(DIE_DEBUG, "debug", regs, PTR_ERR(&dr6), error_code,
 547                                                         SIGTRAP) == NOTIFY_STOP)
 548                 return;
 549
 550         /* It's safe to allow irq's after DR6 has been saved */
 551         preempt_conditional_sti(regs);
 552
 553         if (regs->flags & X86_VM_MASK) {
 554                 handle_vm86_trap((struct kernel_vm86_regs *) regs,
 555                                 error_code, 1);
 556                 return;
 557         }
 558
 559         /*
 560          * Single-stepping through system calls: ignore any exceptions in
 561          * kernel space, but re-enable TF when returning to user mode.
 562          *
 563          * We already checked v86 mode above, so we can check for kernel mode
 564          * by just checking the CPL of CS.
 565          */
 566         if ((dr6 & DR_STEP) && !user_mode(regs)) {
 567                 tsk->thread.debugreg6 &= ~DR_STEP;
 568                 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
 569                 regs->flags &= ~X86_EFLAGS_TF;
 570         }
 571         si_code = get_si_code(tsk->thread.debugreg6);
 572         if (tsk->thread.debugreg6 & (DR_STEP | DR_TRAP_BITS))
 573                 send_sigtrap(tsk, regs, error_code, si_code);
 574         preempt_conditional_cli(regs);
 575
 576         return;
 577 }
 578
 579 /*
 580  * Note that we play around with the 'TS' bit in an attempt to get
 581  * the correct behaviour even in the presence of the asynchronous
 582  * IRQ13 behaviour
 583  */
 584 void math_error(struct pt_regs *regs, int error_code, int trapnr)
 585 {
 586         struct task_struct *task = current;
 587         siginfo_t info;
 588         unsigned short err;
 589         char *str = (trapnr == 16) ? "fpu exception" : "simd exception";
 590
 591         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, SIGFPE) == NOTIFY_STOP)
 592                 return;
 593         conditional_sti(regs);
 594
 595         if (!user_mode_vm(regs))
 596         {
 597                 if (!fixup_exception(regs)) {
 598                         task->thread.error_code = error_code;
 599                         task->thread.trap_no = trapnr;
 600                         die(str, regs, error_code);
 601                 }
 602                 return;
 603         }
 604
 605         /*
 606          * Save the info for the exception handler and clear the error.
 607          */
 608         save_init_fpu(task);
 609         task->thread.trap_no = trapnr;
 610         task->thread.error_code = error_code;
 611         info.si_signo = SIGFPE;
 612         info.si_errno = 0;
 613         info.si_addr = (void __user *)regs->ip;
 614         if (trapnr == 16) {
 615                 unsigned short cwd, swd;
 616                 /*
 617                  * (~cwd & swd) will mask out exceptions that are not set to unmasked
 618                  * status.  0x3f is the exception bits in these regs, 0x200 is the
 619                  * C1 reg you need in case of a stack fault, 0x040 is the stack
 620                  * fault bit.  We should only be taking one exception at a time,
 621                  * so if this combination doesn't produce any single exception,
 622                  * then we have a bad program that isn't synchronizing its FPU usage
 623                  * and it will suffer the consequences since we won't be able to
 624                  * fully reproduce the context of the exception
 625                  */
 626                 cwd = get_fpu_cwd(task);
 627                 swd = get_fpu_swd(task);
 628
 629                 err = swd & ~cwd;
 630         } else {
 631                 /*
 632                  * The SIMD FPU exceptions are handled a little differently, as there
 633                  * is only a single status/control register.  Thus, to determine which
 634                  * unmasked exception was caught we must mask the exception mask bits
 635                  * at 0x1f80, and then use these to mask the exception bits at 0x3f.
 636                  */
 637                 unsigned short mxcsr = get_fpu_mxcsr(task);
 638                 err = ~(mxcsr >> 7) & mxcsr;
 639         }
 640
 641         if (err & 0x001) {      /* Invalid op */
 642                 /*
 643                  * swd & 0x240 == 0x040: Stack Underflow
 644                  * swd & 0x240 == 0x240: Stack Overflow
 645                  * User must clear the SF bit (0x40) if set
 646                  */
 647                 info.si_code = FPE_FLTINV;
 648         } else if (err & 0x004) { /* Divide by Zero */
 649                 info.si_code = FPE_FLTDIV;
 650         } else if (err & 0x008) { /* Overflow */
 651                 info.si_code = FPE_FLTOVF;
 652         } else if (err & 0x012) { /* Denormal, Underflow */
 653                 info.si_code = FPE_FLTUND;
 654         } else if (err & 0x020) { /* Precision */
 655                 info.si_code = FPE_FLTRES;
 656         } else {
 657                 /*
 658                  * If we're using IRQ 13, or supposedly even some trap 16
 659                  * implementations, it's possible we get a spurious trap...
 660                  */
 661                 return;         /* Spurious trap, no error */
 662         }
 663         force_sig_info(SIGFPE, &info, task);
 664 }
 665
 666 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
 667 {
 668 #ifdef CONFIG_X86_32
 669         ignore_fpu_irq = 1;
 670 #endif
 671
 672         math_error(regs, error_code, 16);
 673 }
 674
 675 dotraplinkage void
 676 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
 677 {
 678         math_error(regs, error_code, 19);
 679 }
 680
 681 dotraplinkage void
 682 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
 683 {
 684         conditional_sti(regs);
 685 #if 0
 686         /* No need to warn about this any longer. */
 687         printk(KERN_INFO "Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
 688 #endif
 689 }
 690
 691 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
 692 {
 693 }
 694
 695 asmlinkage void __attribute__((weak)) smp_threshold_interrupt(void)
 696 {
 697 }
 698
 699 /*
 700  * __math_state_restore assumes that cr0.TS is already clear and the
 701  * fpu state is all ready for use.  Used during context switch.
 702  */
 703 void __math_state_restore(void)
 704 {
 705         struct thread_info *thread = current_thread_info();
 706         struct task_struct *tsk = thread->task;
 707
 708         /*
 709          * Paranoid restore. send a SIGSEGV if we fail to restore the state.
 710          */
 711         if (unlikely(restore_fpu_checking(tsk))) {
 712                 stts();
 713                 force_sig(SIGSEGV, tsk);
 714                 return;
 715         }
 716
 717         thread->status |= TS_USEDFPU;   /* So we fnsave on switch_to() */
 718         tsk->fpu_counter++;
 719 }
 720
 721 /*
 722  * 'math_state_restore()' saves the current math information in the
 723  * old math state array, and gets the new ones from the current task
 724  *
 725  * Careful.. There are problems with IBM-designed IRQ13 behaviour.
 726  * Don't touch unless you *really* know how it works.
 727  *
 728  * Must be called with kernel preemption disabled (in this case,
 729  * local interrupts are disabled at the call-site in entry.S).
 730  */
 731 asmlinkage void math_state_restore(void)
 732 {
 733         struct thread_info *thread = current_thread_info();
 734         struct task_struct *tsk = thread->task;
 735
 736         if (!tsk_used_math(tsk)) {
 737                 local_irq_enable();
 738                 /*
 739                  * does a slab alloc which can sleep
 740                  */
 741                 if (init_fpu(tsk)) {
 742                         /*
 743                          * ran out of memory!
 744                          */
 745                         do_group_exit(SIGKILL);
 746                         return;
 747                 }
 748                 local_irq_disable();
 749         }
 750
 751         clts();                         /* Allow maths ops (or we recurse) */
 752
 753         __math_state_restore();
 754 }
 755 EXPORT_SYMBOL_GPL(math_state_restore);
 756
 757 #ifndef CONFIG_MATH_EMULATION
 758 void math_emulate(struct math_emu_info *info)
 759 {
 760         printk(KERN_EMERG
 761                 "math-emulation not enabled and no coprocessor found.\n");
 762         printk(KERN_EMERG "killing %s.\n", current->comm);
 763         force_sig(SIGFPE, current);
 764         schedule();
 765 }
 766 #endif /* CONFIG_MATH_EMULATION */
 767
 768 dotraplinkage void __kprobes
 769 do_device_not_available(struct pt_regs *regs, long error_code)
 770 {
 771 #ifdef CONFIG_X86_32
 772         if (read_cr0() & X86_CR0_EM) {
 773                 struct math_emu_info info = { };
 774
 775                 conditional_sti(regs);
 776
 777                 info.regs = regs;
 778                 math_emulate(&info);
 779         } else {
 780                 math_state_restore(); /* interrupts still off */
 781                 conditional_sti(regs);
 782         }
 783 #else
 784         math_state_restore();
 785 #endif
 786 }
 787
 788 #ifdef CONFIG_X86_32
 789 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
 790 {
 791         siginfo_t info;
 792         local_irq_enable();
 793
 794         info.si_signo = SIGILL;
 795         info.si_errno = 0;
 796         info.si_code = ILL_BADSTK;
 797         info.si_addr = NULL;
 798         if (notify_die(DIE_TRAP, "iret exception",
 799                         regs, error_code, 32, SIGILL) == NOTIFY_STOP)
 800                 return;
 801         do_trap(32, SIGILL, "iret exception", regs, error_code, &info);
 802 }
 803 #endif
 804
 805 void __init trap_init(void)
 806 {
 807         int i;
 808
 809 #ifdef CONFIG_EISA
 810         void __iomem *p = early_ioremap(0x0FFFD9, 4);
 811
 812         if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
 813                 EISA_bus = 1;
 814         early_iounmap(p, 4);
 815 #endif
 816
 817         set_intr_gate(0, &divide_error);
 818         set_intr_gate_ist(1, &debug, DEBUG_STACK);
 819         set_intr_gate_ist(2, &nmi, NMI_STACK);
 820         /* int3 can be called from all */
 821         set_system_intr_gate_ist(3, &int3, DEBUG_STACK);
 822         /* int4 can be called from all */
 823         set_system_intr_gate(4, &overflow);
 824         set_intr_gate(5, &bounds);
 825         set_intr_gate(6, &invalid_op);
 826         set_intr_gate(7, &device_not_available);
 827 #ifdef CONFIG_X86_32
 828         set_task_gate(8, GDT_ENTRY_DOUBLEFAULT_TSS);
 829 #else
 830         set_intr_gate_ist(8, &double_fault, DOUBLEFAULT_STACK);
 831 #endif
 832         set_intr_gate(9, &coprocessor_segment_overrun);
 833         set_intr_gate(10, &invalid_TSS);
 834         set_intr_gate(11, &segment_not_present);
 835         set_intr_gate_ist(12, &stack_segment, STACKFAULT_STACK);
 836         set_intr_gate(13, &general_protection);
 837         set_intr_gate(14, &page_fault);
 838         set_intr_gate(15, &spurious_interrupt_bug);
 839         set_intr_gate(16, &coprocessor_error);
 840         set_intr_gate(17, &alignment_check);
 841 #ifdef CONFIG_X86_MCE
 842         set_intr_gate_ist(18, &machine_check, MCE_STACK);
 843 #endif
 844         set_intr_gate(19, &simd_coprocessor_error);
 845
 846         /* Reserve all the builtin and the syscall vector: */
 847         for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
 848                 set_bit(i, used_vectors);
 849
 850 #ifdef CONFIG_IA32_EMULATION
 851         set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
 852         set_bit(IA32_SYSCALL_VECTOR, used_vectors);
 853 #endif
 854
 855 #ifdef CONFIG_X86_32
 856         if (cpu_has_fxsr) {
 857                 printk(KERN_INFO "Enabling fast FPU save and restore... ");
 858                 set_in_cr4(X86_CR4_OSFXSR);
 859                 printk("done.\n");
 860         }
 861         if (cpu_has_xmm) {
 862                 printk(KERN_INFO
 863                         "Enabling unmasked SIMD FPU exception support... ");
 864                 set_in_cr4(X86_CR4_OSXMMEXCPT);
 865                 printk("done.\n");
 866         }
 867
 868         set_system_trap_gate(SYSCALL_VECTOR, &system_call);
 869         set_bit(SYSCALL_VECTOR, used_vectors);
 870 #endif
 871
 872         /*
 873          * Should be a barrier for any external CPU state:
 874          */
 875         cpu_init();
 876
 877         x86_init.irqs.trap_init();
 878 }