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
  13 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  14
  15 #include <linux/context_tracking.h>
  16 #include <linux/interrupt.h>
  17 #include <linux/kallsyms.h>
  18 #include <linux/spinlock.h>
  19 #include <linux/kprobes.h>
  20 #include <linux/uaccess.h>
  21 #include <linux/kdebug.h>
  22 #include <linux/kgdb.h>
  23 #include <linux/kernel.h>
  24 #include <linux/module.h>
  25 #include <linux/ptrace.h>
  26 #include <linux/uprobes.h>
  27 #include <linux/string.h>
  28 #include <linux/delay.h>
  29 #include <linux/errno.h>
  30 #include <linux/kexec.h>
  31 #include <linux/sched.h>
  32 #include <linux/timer.h>
  33 #include <linux/init.h>
  34 #include <linux/bug.h>
  35 #include <linux/nmi.h>
  36 #include <linux/mm.h>
  37 #include <linux/smp.h>
  38 #include <linux/io.h>
  39
  40 #ifdef CONFIG_EISA
  41 #include <linux/ioport.h>
  42 #include <linux/eisa.h>
  43 #endif
  44
  45 #if defined(CONFIG_EDAC)
  46 #include <linux/edac.h>
  47 #endif
  48
  49 #include <asm/kmemcheck.h>
  50 #include <asm/stacktrace.h>
  51 #include <asm/processor.h>
  52 #include <asm/debugreg.h>
  53 #include <linux/atomic.h>
  54 #include <asm/ftrace.h>
  55 #include <asm/traps.h>
  56 #include <asm/desc.h>
  57 #include <asm/i387.h>
  58 #include <asm/fpu-internal.h>
  59 #include <asm/mce.h>
  60 #include <asm/fixmap.h>
  61 #include <asm/mach_traps.h>
  62 #include <asm/alternative.h>
  63 #include <asm/mpx.h>
  64
  65 #ifdef CONFIG_X86_64
  66 #include <asm/x86_init.h>
  67 #include <asm/pgalloc.h>
  68 #include <asm/proto.h>
  69
  70 /* No need to be aligned, but done to keep all IDTs defined the same way. */
  71 gate_desc debug_idt_table[NR_VECTORS] __page_aligned_bss;
  72 #else
  73 #include <asm/processor-flags.h>
  74 #include <asm/setup.h>
  75
  76 asmlinkage int system_call(void);
  77 #endif
  78
  79 /* Must be page-aligned because the real IDT is used in a fixmap. */
  80 gate_desc idt_table[NR_VECTORS] __page_aligned_bss;
  81
  82 DECLARE_BITMAP(used_vectors, NR_VECTORS);
  83 EXPORT_SYMBOL_GPL(used_vectors);
  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         preempt_count_inc();
  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         preempt_count_dec();
 109 }
 110
 111 static nokprobe_inline int
 112 do_trap_no_signal(struct task_struct *tsk, int trapnr, char *str,
 113                   struct pt_regs *regs, long error_code)
 114 {
 115 #ifdef CONFIG_X86_32
 116         if (regs->flags & X86_VM_MASK) {
 117                 /*
 118                  * Traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
 119                  * On nmi (interrupt 2), do_trap should not be called.
 120                  */
 121                 if (trapnr < X86_TRAP_UD) {
 122                         if (!handle_vm86_trap((struct kernel_vm86_regs *) regs,
 123                                                 error_code, trapnr))
 124                                 return 0;
 125                 }
 126                 return -1;
 127         }
 128 #endif
 129         if (!user_mode(regs)) {
 130                 if (!fixup_exception(regs)) {
 131                         tsk->thread.error_code = error_code;
 132                         tsk->thread.trap_nr = trapnr;
 133                         die(str, regs, error_code);
 134                 }
 135                 return 0;
 136         }
 137
 138         return -1;
 139 }
 140
 141 static siginfo_t *fill_trap_info(struct pt_regs *regs, int signr, int trapnr,
 142                                 siginfo_t *info)
 143 {
 144         unsigned long siaddr;
 145         int sicode;
 146
 147         switch (trapnr) {
 148         default:
 149                 return SEND_SIG_PRIV;
 150
 151         case X86_TRAP_DE:
 152                 sicode = FPE_INTDIV;
 153                 siaddr = uprobe_get_trap_addr(regs);
 154                 break;
 155         case X86_TRAP_UD:
 156                 sicode = ILL_ILLOPN;
 157                 siaddr = uprobe_get_trap_addr(regs);
 158                 break;
 159         case X86_TRAP_AC:
 160                 sicode = BUS_ADRALN;
 161                 siaddr = 0;
 162                 break;
 163         }
 164
 165         info->si_signo = signr;
 166         info->si_errno = 0;
 167         info->si_code = sicode;
 168         info->si_addr = (void __user *)siaddr;
 169         return info;
 170 }
 171
 172 static void
 173 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
 174         long error_code, siginfo_t *info)
 175 {
 176         struct task_struct *tsk = current;
 177
 178
 179         if (!do_trap_no_signal(tsk, trapnr, str, regs, error_code))
 180                 return;
 181         /*
 182          * We want error_code and trap_nr set for userspace faults and
 183          * kernelspace faults which result in die(), but not
 184          * kernelspace faults which are fixed up.  die() gives the
 185          * process no chance to handle the signal and notice the
 186          * kernel fault information, so that won't result in polluting
 187          * the information about previously queued, but not yet
 188          * delivered, faults.  See also do_general_protection below.
 189          */
 190         tsk->thread.error_code = error_code;
 191         tsk->thread.trap_nr = trapnr;
 192
 193 #ifdef CONFIG_X86_64
 194         if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
 195             printk_ratelimit()) {
 196                 pr_info("%s[%d] trap %s ip:%lx sp:%lx error:%lx",
 197                         tsk->comm, tsk->pid, str,
 198                         regs->ip, regs->sp, error_code);
 199                 print_vma_addr(" in ", regs->ip);
 200                 pr_cont("\n");
 201         }
 202 #endif
 203
 204         force_sig_info(signr, info ?: SEND_SIG_PRIV, tsk);
 205 }
 206 NOKPROBE_SYMBOL(do_trap);
 207
 208 static void do_error_trap(struct pt_regs *regs, long error_code, char *str,
 209                           unsigned long trapnr, int signr)
 210 {
 211         enum ctx_state prev_state = exception_enter();
 212         siginfo_t info;
 213
 214         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) !=
 215                         NOTIFY_STOP) {
 216                 conditional_sti(regs);
 217                 do_trap(trapnr, signr, str, regs, error_code,
 218                         fill_trap_info(regs, signr, trapnr, &info));
 219         }
 220
 221         exception_exit(prev_state);
 222 }
 223
 224 #define DO_ERROR(trapnr, signr, str, name)                              \
 225 dotraplinkage void do_##name(struct pt_regs *regs, long error_code)     \
 226 {                                                                       \
 227         do_error_trap(regs, error_code, str, trapnr, signr);            \
 228 }
 229
 230 DO_ERROR(X86_TRAP_DE,     SIGFPE,  "divide error",              divide_error)
 231 DO_ERROR(X86_TRAP_OF,     SIGSEGV, "overflow",                  overflow)
 232 DO_ERROR(X86_TRAP_UD,     SIGILL,  "invalid opcode",            invalid_op)
 233 DO_ERROR(X86_TRAP_OLD_MF, SIGFPE,  "coprocessor segment overrun",coprocessor_segment_overrun)
 234 DO_ERROR(X86_TRAP_TS,     SIGSEGV, "invalid TSS",               invalid_TSS)
 235 DO_ERROR(X86_TRAP_NP,     SIGBUS,  "segment not present",       segment_not_present)
 236 DO_ERROR(X86_TRAP_SS,     SIGBUS,  "stack segment",             stack_segment)
 237 DO_ERROR(X86_TRAP_AC,     SIGBUS,  "alignment check",           alignment_check)
 238
 239 #ifdef CONFIG_X86_64
 240 /* Runs on IST stack */
 241 dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
 242 {
 243         static const char str[] = "double fault";
 244         struct task_struct *tsk = current;
 245
 246 #ifdef CONFIG_X86_ESPFIX64
 247         extern unsigned char native_irq_return_iret[];
 248
 249         /*
 250          * If IRET takes a non-IST fault on the espfix64 stack, then we
 251          * end up promoting it to a doublefault.  In that case, modify
 252          * the stack to make it look like we just entered the #GP
 253          * handler from user space, similar to bad_iret.
 254          */
 255         if (((long)regs->sp >> PGDIR_SHIFT) == ESPFIX_PGD_ENTRY &&
 256                 regs->cs == __KERNEL_CS &&
 257                 regs->ip == (unsigned long)native_irq_return_iret)
 258         {
 259                 struct pt_regs *normal_regs = task_pt_regs(current);
 260
 261                 /* Fake a #GP(0) from userspace. */
 262                 memmove(&normal_regs->ip, (void *)regs->sp, 5*8);
 263                 normal_regs->orig_ax = 0;  /* Missing (lost) #GP error code */
 264                 regs->ip = (unsigned long)general_protection;
 265                 regs->sp = (unsigned long)&normal_regs->orig_ax;
 266                 return;
 267         }
 268 #endif
 269
 270         exception_enter();
 271         /* Return not checked because double check cannot be ignored */
 272         notify_die(DIE_TRAP, str, regs, error_code, X86_TRAP_DF, SIGSEGV);
 273
 274         tsk->thread.error_code = error_code;
 275         tsk->thread.trap_nr = X86_TRAP_DF;
 276
 277 #ifdef CONFIG_DOUBLEFAULT
 278         df_debug(regs, error_code);
 279 #endif
 280         /*
 281          * This is always a kernel trap and never fixable (and thus must
 282          * never return).
 283          */
 284         for (;;)
 285                 die(str, regs, error_code);
 286 }
 287 #endif
 288
 289 dotraplinkage void do_bounds(struct pt_regs *regs, long error_code)
 290 {
 291         struct task_struct *tsk = current;
 292         struct xsave_struct *xsave_buf;
 293         enum ctx_state prev_state;
 294         struct bndcsr *bndcsr;
 295         siginfo_t *info;
 296
 297         prev_state = exception_enter();
 298         if (notify_die(DIE_TRAP, "bounds", regs, error_code,
 299                         X86_TRAP_BR, SIGSEGV) == NOTIFY_STOP)
 300                 goto exit;
 301         conditional_sti(regs);
 302
 303         if (!user_mode(regs))
 304                 die("bounds", regs, error_code);
 305
 306         if (!cpu_feature_enabled(X86_FEATURE_MPX)) {
 307                 /* The exception is not from Intel MPX */
 308                 goto exit_trap;
 309         }
 310
 311         /*
 312          * We need to look at BNDSTATUS to resolve this exception.
 313          * It is not directly accessible, though, so we need to
 314          * do an xsave and then pull it out of the xsave buffer.
 315          */
 316         fpu_save_init(&tsk->thread.fpu);
 317         xsave_buf = &(tsk->thread.fpu.state->xsave);
 318         bndcsr = get_xsave_addr(xsave_buf, XSTATE_BNDCSR);
 319         if (!bndcsr)
 320                 goto exit_trap;
 321
 322         /*
 323          * The error code field of the BNDSTATUS register communicates status
 324          * information of a bound range exception #BR or operation involving
 325          * bound directory.
 326          */
 327         switch (bndcsr->bndstatus & MPX_BNDSTA_ERROR_CODE) {
 328         case 2: /* Bound directory has invalid entry. */
 329                 if (mpx_handle_bd_fault(xsave_buf))
 330                         goto exit_trap;
 331                 break; /* Success, it was handled */
 332         case 1: /* Bound violation. */
 333                 info = mpx_generate_siginfo(regs, xsave_buf);
 334                 if (IS_ERR(info)) {
 335                         /*
 336                          * We failed to decode the MPX instruction.  Act as if
 337                          * the exception was not caused by MPX.
 338                          */
 339                         goto exit_trap;
 340                 }
 341                 /*
 342                  * Success, we decoded the instruction and retrieved
 343                  * an 'info' containing the address being accessed
 344                  * which caused the exception.  This information
 345                  * allows and application to possibly handle the
 346                  * #BR exception itself.
 347                  */
 348                 do_trap(X86_TRAP_BR, SIGSEGV, "bounds", regs, error_code, info);
 349                 kfree(info);
 350                 break;
 351         case 0: /* No exception caused by Intel MPX operations. */
 352                 goto exit_trap;
 353         default:
 354                 die("bounds", regs, error_code);
 355         }
 356
 357 exit:
 358         exception_exit(prev_state);
 359         return;
 360 exit_trap:
 361         /*
 362          * This path out is for all the cases where we could not
 363          * handle the exception in some way (like allocating a
 364          * table or telling userspace about it.  We will also end
 365          * up here if the kernel has MPX turned off at compile
 366          * time..
 367          */
 368         do_trap(X86_TRAP_BR, SIGSEGV, "bounds", regs, error_code, NULL);
 369         exception_exit(prev_state);
 370 }
 371
 372 dotraplinkage void
 373 do_general_protection(struct pt_regs *regs, long error_code)
 374 {
 375         struct task_struct *tsk;
 376         enum ctx_state prev_state;
 377
 378         prev_state = exception_enter();
 379         conditional_sti(regs);
 380
 381 #ifdef CONFIG_X86_32
 382         if (regs->flags & X86_VM_MASK) {
 383                 local_irq_enable();
 384                 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
 385                 goto exit;
 386         }
 387 #endif
 388
 389         tsk = current;
 390         if (!user_mode(regs)) {
 391                 if (fixup_exception(regs))
 392                         goto exit;
 393
 394                 tsk->thread.error_code = error_code;
 395                 tsk->thread.trap_nr = X86_TRAP_GP;
 396                 if (notify_die(DIE_GPF, "general protection fault", regs, error_code,
 397                                X86_TRAP_GP, SIGSEGV) != NOTIFY_STOP)
 398                         die("general protection fault", regs, error_code);
 399                 goto exit;
 400         }
 401
 402         tsk->thread.error_code = error_code;
 403         tsk->thread.trap_nr = X86_TRAP_GP;
 404
 405         if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
 406                         printk_ratelimit()) {
 407                 pr_info("%s[%d] general protection ip:%lx sp:%lx error:%lx",
 408                         tsk->comm, task_pid_nr(tsk),
 409                         regs->ip, regs->sp, error_code);
 410                 print_vma_addr(" in ", regs->ip);
 411                 pr_cont("\n");
 412         }
 413
 414         force_sig_info(SIGSEGV, SEND_SIG_PRIV, tsk);
 415 exit:
 416         exception_exit(prev_state);
 417 }
 418 NOKPROBE_SYMBOL(do_general_protection);
 419
 420 /* May run on IST stack. */
 421 dotraplinkage void notrace do_int3(struct pt_regs *regs, long error_code)
 422 {
 423         enum ctx_state prev_state;
 424
 425 #ifdef CONFIG_DYNAMIC_FTRACE
 426         /*
 427          * ftrace must be first, everything else may cause a recursive crash.
 428          * See note by declaration of modifying_ftrace_code in ftrace.c
 429          */
 430         if (unlikely(atomic_read(&modifying_ftrace_code)) &&
 431             ftrace_int3_handler(regs))
 432                 return;
 433 #endif
 434         if (poke_int3_handler(regs))
 435                 return;
 436
 437         prev_state = exception_enter();
 438 #ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
 439         if (kgdb_ll_trap(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
 440                                 SIGTRAP) == NOTIFY_STOP)
 441                 goto exit;
 442 #endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
 443
 444 #ifdef CONFIG_KPROBES
 445         if (kprobe_int3_handler(regs))
 446                 goto exit;
 447 #endif
 448
 449         if (notify_die(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
 450                         SIGTRAP) == NOTIFY_STOP)
 451                 goto exit;
 452
 453         /*
 454          * Let others (NMI) know that the debug stack is in use
 455          * as we may switch to the interrupt stack.
 456          */
 457         debug_stack_usage_inc();
 458         preempt_conditional_sti(regs);
 459         do_trap(X86_TRAP_BP, SIGTRAP, "int3", regs, error_code, NULL);
 460         preempt_conditional_cli(regs);
 461         debug_stack_usage_dec();
 462 exit:
 463         exception_exit(prev_state);
 464 }
 465 NOKPROBE_SYMBOL(do_int3);
 466
 467 #ifdef CONFIG_X86_64
 468 /*
 469  * Help handler running on IST stack to switch off the IST stack if the
 470  * interrupted code was in user mode. The actual stack switch is done in
 471  * entry_64.S
 472  */
 473 asmlinkage __visible notrace struct pt_regs *sync_regs(struct pt_regs *eregs)
 474 {
 475         struct pt_regs *regs = task_pt_regs(current);
 476         *regs = *eregs;
 477         return regs;
 478 }
 479 NOKPROBE_SYMBOL(sync_regs);
 480
 481 struct bad_iret_stack {
 482         void *error_entry_ret;
 483         struct pt_regs regs;
 484 };
 485
 486 asmlinkage __visible notrace
 487 struct bad_iret_stack *fixup_bad_iret(struct bad_iret_stack *s)
 488 {
 489         /*
 490          * This is called from entry_64.S early in handling a fault
 491          * caused by a bad iret to user mode.  To handle the fault
 492          * correctly, we want move our stack frame to task_pt_regs
 493          * and we want to pretend that the exception came from the
 494          * iret target.
 495          */
 496         struct bad_iret_stack *new_stack =
 497                 container_of(task_pt_regs(current),
 498                              struct bad_iret_stack, regs);
 499
 500         /* Copy the IRET target to the new stack. */
 501         memmove(&new_stack->regs.ip, (void *)s->regs.sp, 5*8);
 502
 503         /* Copy the remainder of the stack from the current stack. */
 504         memmove(new_stack, s, offsetof(struct bad_iret_stack, regs.ip));
 505
 506         BUG_ON(!user_mode_vm(&new_stack->regs));
 507         return new_stack;
 508 }
 509 NOKPROBE_SYMBOL(fixup_bad_iret);
 510 #endif
 511
 512 /*
 513  * Our handling of the processor debug registers is non-trivial.
 514  * We do not clear them on entry and exit from the kernel. Therefore
 515  * it is possible to get a watchpoint trap here from inside the kernel.
 516  * However, the code in ./ptrace.c has ensured that the user can
 517  * only set watchpoints on userspace addresses. Therefore the in-kernel
 518  * watchpoint trap can only occur in code which is reading/writing
 519  * from user space. Such code must not hold kernel locks (since it
 520  * can equally take a page fault), therefore it is safe to call
 521  * force_sig_info even though that claims and releases locks.
 522  *
 523  * Code in ./signal.c ensures that the debug control register
 524  * is restored before we deliver any signal, and therefore that
 525  * user code runs with the correct debug control register even though
 526  * we clear it here.
 527  *
 528  * Being careful here means that we don't have to be as careful in a
 529  * lot of more complicated places (task switching can be a bit lazy
 530  * about restoring all the debug state, and ptrace doesn't have to
 531  * find every occurrence of the TF bit that could be saved away even
 532  * by user code)
 533  *
 534  * May run on IST stack.
 535  */
 536 dotraplinkage void do_debug(struct pt_regs *regs, long error_code)
 537 {
 538         struct task_struct *tsk = current;
 539         enum ctx_state prev_state;
 540         int user_icebp = 0;
 541         unsigned long dr6;
 542         int si_code;
 543
 544         prev_state = exception_enter();
 545
 546         get_debugreg(dr6, 6);
 547
 548         /* Filter out all the reserved bits which are preset to 1 */
 549         dr6 &= ~DR6_RESERVED;
 550
 551         /*
 552          * If dr6 has no reason to give us about the origin of this trap,
 553          * then it's very likely the result of an icebp/int01 trap.
 554          * User wants a sigtrap for that.
 555          */
 556         if (!dr6 && user_mode(regs))
 557                 user_icebp = 1;
 558
 559         /* Catch kmemcheck conditions first of all! */
 560         if ((dr6 & DR_STEP) && kmemcheck_trap(regs))
 561                 goto exit;
 562
 563         /* DR6 may or may not be cleared by the CPU */
 564         set_debugreg(0, 6);
 565
 566         /*
 567          * The processor cleared BTF, so don't mark that we need it set.
 568          */
 569         clear_tsk_thread_flag(tsk, TIF_BLOCKSTEP);
 570
 571         /* Store the virtualized DR6 value */
 572         tsk->thread.debugreg6 = dr6;
 573
 574 #ifdef CONFIG_KPROBES
 575         if (kprobe_debug_handler(regs))
 576                 goto exit;
 577 #endif
 578
 579         if (notify_die(DIE_DEBUG, "debug", regs, (long)&dr6, error_code,
 580                                                         SIGTRAP) == NOTIFY_STOP)
 581                 goto exit;
 582
 583         /*
 584          * Let others (NMI) know that the debug stack is in use
 585          * as we may switch to the interrupt stack.
 586          */
 587         debug_stack_usage_inc();
 588
 589         /* It's safe to allow irq's after DR6 has been saved */
 590         preempt_conditional_sti(regs);
 591
 592         if (regs->flags & X86_VM_MASK) {
 593                 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code,
 594                                         X86_TRAP_DB);
 595                 preempt_conditional_cli(regs);
 596                 debug_stack_usage_dec();
 597                 goto exit;
 598         }
 599
 600         /*
 601          * Single-stepping through system calls: ignore any exceptions in
 602          * kernel space, but re-enable TF when returning to user mode.
 603          *
 604          * We already checked v86 mode above, so we can check for kernel mode
 605          * by just checking the CPL of CS.
 606          */
 607         if ((dr6 & DR_STEP) && !user_mode(regs)) {
 608                 tsk->thread.debugreg6 &= ~DR_STEP;
 609                 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
 610                 regs->flags &= ~X86_EFLAGS_TF;
 611         }
 612         si_code = get_si_code(tsk->thread.debugreg6);
 613         if (tsk->thread.debugreg6 & (DR_STEP | DR_TRAP_BITS) || user_icebp)
 614                 send_sigtrap(tsk, regs, error_code, si_code);
 615         preempt_conditional_cli(regs);
 616         debug_stack_usage_dec();
 617
 618 exit:
 619         exception_exit(prev_state);
 620 }
 621 NOKPROBE_SYMBOL(do_debug);
 622
 623 /*
 624  * Note that we play around with the 'TS' bit in an attempt to get
 625  * the correct behaviour even in the presence of the asynchronous
 626  * IRQ13 behaviour
 627  */
 628 static void math_error(struct pt_regs *regs, int error_code, int trapnr)
 629 {
 630         struct task_struct *task = current;
 631         siginfo_t info;
 632         unsigned short err;
 633         char *str = (trapnr == X86_TRAP_MF) ? "fpu exception" :
 634                                                 "simd exception";
 635
 636         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, SIGFPE) == NOTIFY_STOP)
 637                 return;
 638         conditional_sti(regs);
 639
 640         if (!user_mode_vm(regs))
 641         {
 642                 if (!fixup_exception(regs)) {
 643                         task->thread.error_code = error_code;
 644                         task->thread.trap_nr = trapnr;
 645                         die(str, regs, error_code);
 646                 }
 647                 return;
 648         }
 649
 650         /*
 651          * Save the info for the exception handler and clear the error.
 652          */
 653         save_init_fpu(task);
 654         task->thread.trap_nr = trapnr;
 655         task->thread.error_code = error_code;
 656         info.si_signo = SIGFPE;
 657         info.si_errno = 0;
 658         info.si_addr = (void __user *)uprobe_get_trap_addr(regs);
 659         if (trapnr == X86_TRAP_MF) {
 660                 unsigned short cwd, swd;
 661                 /*
 662                  * (~cwd & swd) will mask out exceptions that are not set to unmasked
 663                  * status.  0x3f is the exception bits in these regs, 0x200 is the
 664                  * C1 reg you need in case of a stack fault, 0x040 is the stack
 665                  * fault bit.  We should only be taking one exception at a time,
 666                  * so if this combination doesn't produce any single exception,
 667                  * then we have a bad program that isn't synchronizing its FPU usage
 668                  * and it will suffer the consequences since we won't be able to
 669                  * fully reproduce the context of the exception
 670                  */
 671                 cwd = get_fpu_cwd(task);
 672                 swd = get_fpu_swd(task);
 673
 674                 err = swd & ~cwd;
 675         } else {
 676                 /*
 677                  * The SIMD FPU exceptions are handled a little differently, as there
 678                  * is only a single status/control register.  Thus, to determine which
 679                  * unmasked exception was caught we must mask the exception mask bits
 680                  * at 0x1f80, and then use these to mask the exception bits at 0x3f.
 681                  */
 682                 unsigned short mxcsr = get_fpu_mxcsr(task);
 683                 err = ~(mxcsr >> 7) & mxcsr;
 684         }
 685
 686         if (err & 0x001) {      /* Invalid op */
 687                 /*
 688                  * swd & 0x240 == 0x040: Stack Underflow
 689                  * swd & 0x240 == 0x240: Stack Overflow
 690                  * User must clear the SF bit (0x40) if set
 691                  */
 692                 info.si_code = FPE_FLTINV;
 693         } else if (err & 0x004) { /* Divide by Zero */
 694                 info.si_code = FPE_FLTDIV;
 695         } else if (err & 0x008) { /* Overflow */
 696                 info.si_code = FPE_FLTOVF;
 697         } else if (err & 0x012) { /* Denormal, Underflow */
 698                 info.si_code = FPE_FLTUND;
 699         } else if (err & 0x020) { /* Precision */
 700                 info.si_code = FPE_FLTRES;
 701         } else {
 702                 /*
 703                  * If we're using IRQ 13, or supposedly even some trap
 704                  * X86_TRAP_MF implementations, it's possible
 705                  * we get a spurious trap, which is not an error.
 706                  */
 707                 return;
 708         }
 709         force_sig_info(SIGFPE, &info, task);
 710 }
 711
 712 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
 713 {
 714         enum ctx_state prev_state;
 715
 716         prev_state = exception_enter();
 717         math_error(regs, error_code, X86_TRAP_MF);
 718         exception_exit(prev_state);
 719 }
 720
 721 dotraplinkage void
 722 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
 723 {
 724         enum ctx_state prev_state;
 725
 726         prev_state = exception_enter();
 727         math_error(regs, error_code, X86_TRAP_XF);
 728         exception_exit(prev_state);
 729 }
 730
 731 dotraplinkage void
 732 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
 733 {
 734         conditional_sti(regs);
 735 #if 0
 736         /* No need to warn about this any longer. */
 737         pr_info("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
 738 #endif
 739 }
 740
 741 asmlinkage __visible void __attribute__((weak)) smp_thermal_interrupt(void)
 742 {
 743 }
 744
 745 asmlinkage __visible void __attribute__((weak)) smp_threshold_interrupt(void)
 746 {
 747 }
 748
 749 /*
 750  * 'math_state_restore()' saves the current math information in the
 751  * old math state array, and gets the new ones from the current task
 752  *
 753  * Careful.. There are problems with IBM-designed IRQ13 behaviour.
 754  * Don't touch unless you *really* know how it works.
 755  *
 756  * Must be called with kernel preemption disabled (eg with local
 757  * local interrupts as in the case of do_device_not_available).
 758  */
 759 void math_state_restore(void)
 760 {
 761         struct task_struct *tsk = current;
 762
 763         if (!tsk_used_math(tsk)) {
 764                 local_irq_enable();
 765                 /*
 766                  * does a slab alloc which can sleep
 767                  */
 768                 if (init_fpu(tsk)) {
 769                         /*
 770                          * ran out of memory!
 771                          */
 772                         do_group_exit(SIGKILL);
 773                         return;
 774                 }
 775                 local_irq_disable();
 776         }
 777
 778         __thread_fpu_begin(tsk);
 779
 780         /*
 781          * Paranoid restore. send a SIGSEGV if we fail to restore the state.
 782          */
 783         if (unlikely(restore_fpu_checking(tsk))) {
 784                 drop_init_fpu(tsk);
 785                 force_sig_info(SIGSEGV, SEND_SIG_PRIV, tsk);
 786                 return;
 787         }
 788
 789         tsk->thread.fpu_counter++;
 790 }
 791 EXPORT_SYMBOL_GPL(math_state_restore);
 792
 793 dotraplinkage void
 794 do_device_not_available(struct pt_regs *regs, long error_code)
 795 {
 796         enum ctx_state prev_state;
 797
 798         prev_state = exception_enter();
 799         BUG_ON(use_eager_fpu());
 800
 801 #ifdef CONFIG_MATH_EMULATION
 802         if (read_cr0() & X86_CR0_EM) {
 803                 struct math_emu_info info = { };
 804
 805                 conditional_sti(regs);
 806
 807                 info.regs = regs;
 808                 math_emulate(&info);
 809                 exception_exit(prev_state);
 810                 return;
 811         }
 812 #endif
 813         math_state_restore(); /* interrupts still off */
 814 #ifdef CONFIG_X86_32
 815         conditional_sti(regs);
 816 #endif
 817         exception_exit(prev_state);
 818 }
 819 NOKPROBE_SYMBOL(do_device_not_available);
 820
 821 #ifdef CONFIG_X86_32
 822 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
 823 {
 824         siginfo_t info;
 825         enum ctx_state prev_state;
 826
 827         prev_state = exception_enter();
 828         local_irq_enable();
 829
 830         info.si_signo = SIGILL;
 831         info.si_errno = 0;
 832         info.si_code = ILL_BADSTK;
 833         info.si_addr = NULL;
 834         if (notify_die(DIE_TRAP, "iret exception", regs, error_code,
 835                         X86_TRAP_IRET, SIGILL) != NOTIFY_STOP) {
 836                 do_trap(X86_TRAP_IRET, SIGILL, "iret exception", regs, error_code,
 837                         &info);
 838         }
 839         exception_exit(prev_state);
 840 }
 841 #endif
 842
 843 /* Set of traps needed for early debugging. */
 844 void __init early_trap_init(void)
 845 {
 846         set_intr_gate_ist(X86_TRAP_DB, &debug, DEBUG_STACK);
 847         /* int3 can be called from all */
 848         set_system_intr_gate_ist(X86_TRAP_BP, &int3, DEBUG_STACK);
 849 #ifdef CONFIG_X86_32
 850         set_intr_gate(X86_TRAP_PF, page_fault);
 851 #endif
 852         load_idt(&idt_descr);
 853 }
 854
 855 void __init early_trap_pf_init(void)
 856 {
 857 #ifdef CONFIG_X86_64
 858         set_intr_gate(X86_TRAP_PF, page_fault);
 859 #endif
 860 }
 861
 862 void __init trap_init(void)
 863 {
 864         int i;
 865
 866 #ifdef CONFIG_EISA
 867         void __iomem *p = early_ioremap(0x0FFFD9, 4);
 868
 869         if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
 870                 EISA_bus = 1;
 871         early_iounmap(p, 4);
 872 #endif
 873
 874         set_intr_gate(X86_TRAP_DE, divide_error);
 875         set_intr_gate_ist(X86_TRAP_NMI, &nmi, NMI_STACK);
 876         /* int4 can be called from all */
 877         set_system_intr_gate(X86_TRAP_OF, &overflow);
 878         set_intr_gate(X86_TRAP_BR, bounds);
 879         set_intr_gate(X86_TRAP_UD, invalid_op);
 880         set_intr_gate(X86_TRAP_NM, device_not_available);
 881 #ifdef CONFIG_X86_32
 882         set_task_gate(X86_TRAP_DF, GDT_ENTRY_DOUBLEFAULT_TSS);
 883 #else
 884         set_intr_gate_ist(X86_TRAP_DF, &double_fault, DOUBLEFAULT_STACK);
 885 #endif
 886         set_intr_gate(X86_TRAP_OLD_MF, coprocessor_segment_overrun);
 887         set_intr_gate(X86_TRAP_TS, invalid_TSS);
 888         set_intr_gate(X86_TRAP_NP, segment_not_present);
 889         set_intr_gate(X86_TRAP_SS, stack_segment);
 890         set_intr_gate(X86_TRAP_GP, general_protection);
 891         set_intr_gate(X86_TRAP_SPURIOUS, spurious_interrupt_bug);
 892         set_intr_gate(X86_TRAP_MF, coprocessor_error);
 893         set_intr_gate(X86_TRAP_AC, alignment_check);
 894 #ifdef CONFIG_X86_MCE
 895         set_intr_gate_ist(X86_TRAP_MC, &machine_check, MCE_STACK);
 896 #endif
 897         set_intr_gate(X86_TRAP_XF, simd_coprocessor_error);
 898
 899         /* Reserve all the builtin and the syscall vector: */
 900         for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
 901                 set_bit(i, used_vectors);
 902
 903 #ifdef CONFIG_IA32_EMULATION
 904         set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
 905         set_bit(IA32_SYSCALL_VECTOR, used_vectors);
 906 #endif
 907
 908 #ifdef CONFIG_X86_32
 909         set_system_trap_gate(SYSCALL_VECTOR, &system_call);
 910         set_bit(SYSCALL_VECTOR, used_vectors);
 911 #endif
 912
 913         /*
 914          * Set the IDT descriptor to a fixed read-only location, so that the
 915          * "sidt" instruction will not leak the location of the kernel, and
 916          * to defend the IDT against arbitrary memory write vulnerabilities.
 917          * It will be reloaded in cpu_init() */
 918         __set_fixmap(FIX_RO_IDT, __pa_symbol(idt_table), PAGE_KERNEL_RO);
 919         idt_descr.address = fix_to_virt(FIX_RO_IDT);
 920
 921         /*
 922          * Should be a barrier for any external CPU state:
 923          */
 924         cpu_init();
 925
 926         x86_init.irqs.trap_init();
 927
 928 #ifdef CONFIG_X86_64
 929         memcpy(&debug_idt_table, &idt_table, IDT_ENTRIES * 16);
 930         set_nmi_gate(X86_TRAP_DB, &debug);
 931         set_nmi_gate(X86_TRAP_BP, &int3);
 932 #endif
 933 }