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
  64 #ifdef CONFIG_X86_64
  65 #include <asm/x86_init.h>
  66 #include <asm/pgalloc.h>
  67 #include <asm/proto.h>
  68
  69 /* No need to be aligned, but done to keep all IDTs defined the same way. */
  70 gate_desc debug_idt_table[NR_VECTORS] __page_aligned_bss;
  71 #else
  72 #include <asm/processor-flags.h>
  73 #include <asm/setup.h>
  74
  75 asmlinkage int system_call(void);
  76 #endif
  77
  78 /* Must be page-aligned because the real IDT is used in a fixmap. */
  79 gate_desc idt_table[NR_VECTORS] __page_aligned_bss;
  80
  81 DECLARE_BITMAP(used_vectors, NR_VECTORS);
  82 EXPORT_SYMBOL_GPL(used_vectors);
  83
  84 static inline void conditional_sti(struct pt_regs *regs)
  85 {
  86         if (regs->flags & X86_EFLAGS_IF)
  87                 local_irq_enable();
  88 }
  89
  90 static inline void preempt_conditional_sti(struct pt_regs *regs)
  91 {
  92         preempt_count_inc();
  93         if (regs->flags & X86_EFLAGS_IF)
  94                 local_irq_enable();
  95 }
  96
  97 static inline void conditional_cli(struct pt_regs *regs)
  98 {
  99         if (regs->flags & X86_EFLAGS_IF)
 100                 local_irq_disable();
 101 }
 102
 103 static inline void preempt_conditional_cli(struct pt_regs *regs)
 104 {
 105         if (regs->flags & X86_EFLAGS_IF)
 106                 local_irq_disable();
 107         preempt_count_dec();
 108 }
 109
 110 static nokprobe_inline int
 111 do_trap_no_signal(struct task_struct *tsk, int trapnr, char *str,
 112                   struct pt_regs *regs, long error_code)
 113 {
 114 #ifdef CONFIG_X86_32
 115         if (regs->flags & X86_VM_MASK) {
 116                 /*
 117                  * Traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
 118                  * On nmi (interrupt 2), do_trap should not be called.
 119                  */
 120                 if (trapnr < X86_TRAP_UD) {
 121                         if (!handle_vm86_trap((struct kernel_vm86_regs *) regs,
 122                                                 error_code, trapnr))
 123                                 return 0;
 124                 }
 125                 return -1;
 126         }
 127 #endif
 128         if (!user_mode(regs)) {
 129                 if (!fixup_exception(regs)) {
 130                         tsk->thread.error_code = error_code;
 131                         tsk->thread.trap_nr = trapnr;
 132                         die(str, regs, error_code);
 133                 }
 134                 return 0;
 135         }
 136
 137         return -1;
 138 }
 139
 140 static siginfo_t *fill_trap_info(struct pt_regs *regs, int signr, int trapnr,
 141                                 siginfo_t *info)
 142 {
 143         unsigned long siaddr;
 144         int sicode;
 145
 146         switch (trapnr) {
 147         default:
 148                 return SEND_SIG_PRIV;
 149
 150         case X86_TRAP_DE:
 151                 sicode = FPE_INTDIV;
 152                 siaddr = uprobe_get_trap_addr(regs);
 153                 break;
 154         case X86_TRAP_UD:
 155                 sicode = ILL_ILLOPN;
 156                 siaddr = uprobe_get_trap_addr(regs);
 157                 break;
 158         case X86_TRAP_AC:
 159                 sicode = BUS_ADRALN;
 160                 siaddr = 0;
 161                 break;
 162         }
 163
 164         info->si_signo = signr;
 165         info->si_errno = 0;
 166         info->si_code = sicode;
 167         info->si_addr = (void __user *)siaddr;
 168         return info;
 169 }
 170
 171 static void
 172 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
 173         long error_code, siginfo_t *info)
 174 {
 175         struct task_struct *tsk = current;
 176
 177
 178         if (!do_trap_no_signal(tsk, trapnr, str, regs, error_code))
 179                 return;
 180         /*
 181          * We want error_code and trap_nr set for userspace faults and
 182          * kernelspace faults which result in die(), but not
 183          * kernelspace faults which are fixed up.  die() gives the
 184          * process no chance to handle the signal and notice the
 185          * kernel fault information, so that won't result in polluting
 186          * the information about previously queued, but not yet
 187          * delivered, faults.  See also do_general_protection below.
 188          */
 189         tsk->thread.error_code = error_code;
 190         tsk->thread.trap_nr = trapnr;
 191
 192 #ifdef CONFIG_X86_64
 193         if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
 194             printk_ratelimit()) {
 195                 pr_info("%s[%d] trap %s ip:%lx sp:%lx error:%lx",
 196                         tsk->comm, tsk->pid, str,
 197                         regs->ip, regs->sp, error_code);
 198                 print_vma_addr(" in ", regs->ip);
 199                 pr_cont("\n");
 200         }
 201 #endif
 202
 203         force_sig_info(signr, info ?: SEND_SIG_PRIV, tsk);
 204 }
 205 NOKPROBE_SYMBOL(do_trap);
 206
 207 static void do_error_trap(struct pt_regs *regs, long error_code, char *str,
 208                           unsigned long trapnr, int signr)
 209 {
 210         enum ctx_state prev_state = exception_enter();
 211         siginfo_t info;
 212
 213         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) !=
 214                         NOTIFY_STOP) {
 215                 conditional_sti(regs);
 216                 do_trap(trapnr, signr, str, regs, error_code,
 217                         fill_trap_info(regs, signr, trapnr, &info));
 218         }
 219
 220         exception_exit(prev_state);
 221 }
 222
 223 #define DO_ERROR(trapnr, signr, str, name)                              \
 224 dotraplinkage void do_##name(struct pt_regs *regs, long error_code)     \
 225 {                                                                       \
 226         do_error_trap(regs, error_code, str, trapnr, signr);            \
 227 }
 228
 229 DO_ERROR(X86_TRAP_DE,     SIGFPE,  "divide error",              divide_error)
 230 DO_ERROR(X86_TRAP_OF,     SIGSEGV, "overflow",                  overflow)
 231 DO_ERROR(X86_TRAP_BR,     SIGSEGV, "bounds",                    bounds)
 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
 290 do_general_protection(struct pt_regs *regs, long error_code)
 291 {
 292         struct task_struct *tsk;
 293         enum ctx_state prev_state;
 294
 295         prev_state = exception_enter();
 296         conditional_sti(regs);
 297
 298 #ifdef CONFIG_X86_32
 299         if (regs->flags & X86_VM_MASK) {
 300                 local_irq_enable();
 301                 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
 302                 goto exit;
 303         }
 304 #endif
 305
 306         tsk = current;
 307         if (!user_mode(regs)) {
 308                 if (fixup_exception(regs))
 309                         goto exit;
 310
 311                 tsk->thread.error_code = error_code;
 312                 tsk->thread.trap_nr = X86_TRAP_GP;
 313                 if (notify_die(DIE_GPF, "general protection fault", regs, error_code,
 314                                X86_TRAP_GP, SIGSEGV) != NOTIFY_STOP)
 315                         die("general protection fault", regs, error_code);
 316                 goto exit;
 317         }
 318
 319         tsk->thread.error_code = error_code;
 320         tsk->thread.trap_nr = X86_TRAP_GP;
 321
 322         if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
 323                         printk_ratelimit()) {
 324                 pr_info("%s[%d] general protection ip:%lx sp:%lx error:%lx",
 325                         tsk->comm, task_pid_nr(tsk),
 326                         regs->ip, regs->sp, error_code);
 327                 print_vma_addr(" in ", regs->ip);
 328                 pr_cont("\n");
 329         }
 330
 331         force_sig_info(SIGSEGV, SEND_SIG_PRIV, tsk);
 332 exit:
 333         exception_exit(prev_state);
 334 }
 335 NOKPROBE_SYMBOL(do_general_protection);
 336
 337 /* May run on IST stack. */
 338 dotraplinkage void notrace do_int3(struct pt_regs *regs, long error_code)
 339 {
 340         enum ctx_state prev_state;
 341
 342 #ifdef CONFIG_DYNAMIC_FTRACE
 343         /*
 344          * ftrace must be first, everything else may cause a recursive crash.
 345          * See note by declaration of modifying_ftrace_code in ftrace.c
 346          */
 347         if (unlikely(atomic_read(&modifying_ftrace_code)) &&
 348             ftrace_int3_handler(regs))
 349                 return;
 350 #endif
 351         if (poke_int3_handler(regs))
 352                 return;
 353
 354         prev_state = exception_enter();
 355 #ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
 356         if (kgdb_ll_trap(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
 357                                 SIGTRAP) == NOTIFY_STOP)
 358                 goto exit;
 359 #endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
 360
 361 #ifdef CONFIG_KPROBES
 362         if (kprobe_int3_handler(regs))
 363                 goto exit;
 364 #endif
 365
 366         if (notify_die(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
 367                         SIGTRAP) == NOTIFY_STOP)
 368                 goto exit;
 369
 370         /*
 371          * Let others (NMI) know that the debug stack is in use
 372          * as we may switch to the interrupt stack.
 373          */
 374         debug_stack_usage_inc();
 375         preempt_conditional_sti(regs);
 376         do_trap(X86_TRAP_BP, SIGTRAP, "int3", regs, error_code, NULL);
 377         preempt_conditional_cli(regs);
 378         debug_stack_usage_dec();
 379 exit:
 380         exception_exit(prev_state);
 381 }
 382 NOKPROBE_SYMBOL(do_int3);
 383
 384 #ifdef CONFIG_X86_64
 385 /*
 386  * Help handler running on IST stack to switch back to user stack
 387  * for scheduling or signal handling. The actual stack switch is done in
 388  * entry.S
 389  */
 390 asmlinkage __visible struct pt_regs *sync_regs(struct pt_regs *eregs)
 391 {
 392         struct pt_regs *regs = eregs;
 393         /* Did already sync */
 394         if (eregs == (struct pt_regs *)eregs->sp)
 395                 ;
 396         /* Exception from user space */
 397         else if (user_mode(eregs))
 398                 regs = task_pt_regs(current);
 399         /*
 400          * Exception from kernel and interrupts are enabled. Move to
 401          * kernel process stack.
 402          */
 403         else if (eregs->flags & X86_EFLAGS_IF)
 404                 regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
 405         if (eregs != regs)
 406                 *regs = *eregs;
 407         return regs;
 408 }
 409 NOKPROBE_SYMBOL(sync_regs);
 410
 411 struct bad_iret_stack {
 412         void *error_entry_ret;
 413         struct pt_regs regs;
 414 };
 415
 416 asmlinkage __visible
 417 struct bad_iret_stack *fixup_bad_iret(struct bad_iret_stack *s)
 418 {
 419         /*
 420          * This is called from entry_64.S early in handling a fault
 421          * caused by a bad iret to user mode.  To handle the fault
 422          * correctly, we want move our stack frame to task_pt_regs
 423          * and we want to pretend that the exception came from the
 424          * iret target.
 425          */
 426         struct bad_iret_stack *new_stack =
 427                 container_of(task_pt_regs(current),
 428                              struct bad_iret_stack, regs);
 429
 430         /* Copy the IRET target to the new stack. */
 431         memmove(&new_stack->regs.ip, (void *)s->regs.sp, 5*8);
 432
 433         /* Copy the remainder of the stack from the current stack. */
 434         memmove(new_stack, s, offsetof(struct bad_iret_stack, regs.ip));
 435
 436         BUG_ON(!user_mode_vm(&new_stack->regs));
 437         return new_stack;
 438 }
 439 #endif
 440
 441 /*
 442  * Our handling of the processor debug registers is non-trivial.
 443  * We do not clear them on entry and exit from the kernel. Therefore
 444  * it is possible to get a watchpoint trap here from inside the kernel.
 445  * However, the code in ./ptrace.c has ensured that the user can
 446  * only set watchpoints on userspace addresses. Therefore the in-kernel
 447  * watchpoint trap can only occur in code which is reading/writing
 448  * from user space. Such code must not hold kernel locks (since it
 449  * can equally take a page fault), therefore it is safe to call
 450  * force_sig_info even though that claims and releases locks.
 451  *
 452  * Code in ./signal.c ensures that the debug control register
 453  * is restored before we deliver any signal, and therefore that
 454  * user code runs with the correct debug control register even though
 455  * we clear it here.
 456  *
 457  * Being careful here means that we don't have to be as careful in a
 458  * lot of more complicated places (task switching can be a bit lazy
 459  * about restoring all the debug state, and ptrace doesn't have to
 460  * find every occurrence of the TF bit that could be saved away even
 461  * by user code)
 462  *
 463  * May run on IST stack.
 464  */
 465 dotraplinkage void do_debug(struct pt_regs *regs, long error_code)
 466 {
 467         struct task_struct *tsk = current;
 468         enum ctx_state prev_state;
 469         int user_icebp = 0;
 470         unsigned long dr6;
 471         int si_code;
 472
 473         prev_state = exception_enter();
 474
 475         get_debugreg(dr6, 6);
 476
 477         /* Filter out all the reserved bits which are preset to 1 */
 478         dr6 &= ~DR6_RESERVED;
 479
 480         /*
 481          * If dr6 has no reason to give us about the origin of this trap,
 482          * then it's very likely the result of an icebp/int01 trap.
 483          * User wants a sigtrap for that.
 484          */
 485         if (!dr6 && user_mode(regs))
 486                 user_icebp = 1;
 487
 488         /* Catch kmemcheck conditions first of all! */
 489         if ((dr6 & DR_STEP) && kmemcheck_trap(regs))
 490                 goto exit;
 491
 492         /* DR6 may or may not be cleared by the CPU */
 493         set_debugreg(0, 6);
 494
 495         /*
 496          * The processor cleared BTF, so don't mark that we need it set.
 497          */
 498         clear_tsk_thread_flag(tsk, TIF_BLOCKSTEP);
 499
 500         /* Store the virtualized DR6 value */
 501         tsk->thread.debugreg6 = dr6;
 502
 503 #ifdef CONFIG_KPROBES
 504         if (kprobe_debug_handler(regs))
 505                 goto exit;
 506 #endif
 507
 508         if (notify_die(DIE_DEBUG, "debug", regs, (long)&dr6, error_code,
 509                                                         SIGTRAP) == NOTIFY_STOP)
 510                 goto exit;
 511
 512         /*
 513          * Let others (NMI) know that the debug stack is in use
 514          * as we may switch to the interrupt stack.
 515          */
 516         debug_stack_usage_inc();
 517
 518         /* It's safe to allow irq's after DR6 has been saved */
 519         preempt_conditional_sti(regs);
 520
 521         if (regs->flags & X86_VM_MASK) {
 522                 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code,
 523                                         X86_TRAP_DB);
 524                 preempt_conditional_cli(regs);
 525                 debug_stack_usage_dec();
 526                 goto exit;
 527         }
 528
 529         /*
 530          * Single-stepping through system calls: ignore any exceptions in
 531          * kernel space, but re-enable TF when returning to user mode.
 532          *
 533          * We already checked v86 mode above, so we can check for kernel mode
 534          * by just checking the CPL of CS.
 535          */
 536         if ((dr6 & DR_STEP) && !user_mode(regs)) {
 537                 tsk->thread.debugreg6 &= ~DR_STEP;
 538                 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
 539                 regs->flags &= ~X86_EFLAGS_TF;
 540         }
 541         si_code = get_si_code(tsk->thread.debugreg6);
 542         if (tsk->thread.debugreg6 & (DR_STEP | DR_TRAP_BITS) || user_icebp)
 543                 send_sigtrap(tsk, regs, error_code, si_code);
 544         preempt_conditional_cli(regs);
 545         debug_stack_usage_dec();
 546
 547 exit:
 548         exception_exit(prev_state);
 549 }
 550 NOKPROBE_SYMBOL(do_debug);
 551
 552 /*
 553  * Note that we play around with the 'TS' bit in an attempt to get
 554  * the correct behaviour even in the presence of the asynchronous
 555  * IRQ13 behaviour
 556  */
 557 static void math_error(struct pt_regs *regs, int error_code, int trapnr)
 558 {
 559         struct task_struct *task = current;
 560         siginfo_t info;
 561         unsigned short err;
 562         char *str = (trapnr == X86_TRAP_MF) ? "fpu exception" :
 563                                                 "simd exception";
 564
 565         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, SIGFPE) == NOTIFY_STOP)
 566                 return;
 567         conditional_sti(regs);
 568
 569         if (!user_mode_vm(regs))
 570         {
 571                 if (!fixup_exception(regs)) {
 572                         task->thread.error_code = error_code;
 573                         task->thread.trap_nr = trapnr;
 574                         die(str, regs, error_code);
 575                 }
 576                 return;
 577         }
 578
 579         /*
 580          * Save the info for the exception handler and clear the error.
 581          */
 582         save_init_fpu(task);
 583         task->thread.trap_nr = trapnr;
 584         task->thread.error_code = error_code;
 585         info.si_signo = SIGFPE;
 586         info.si_errno = 0;
 587         info.si_addr = (void __user *)uprobe_get_trap_addr(regs);
 588         if (trapnr == X86_TRAP_MF) {
 589                 unsigned short cwd, swd;
 590                 /*
 591                  * (~cwd & swd) will mask out exceptions that are not set to unmasked
 592                  * status.  0x3f is the exception bits in these regs, 0x200 is the
 593                  * C1 reg you need in case of a stack fault, 0x040 is the stack
 594                  * fault bit.  We should only be taking one exception at a time,
 595                  * so if this combination doesn't produce any single exception,
 596                  * then we have a bad program that isn't synchronizing its FPU usage
 597                  * and it will suffer the consequences since we won't be able to
 598                  * fully reproduce the context of the exception
 599                  */
 600                 cwd = get_fpu_cwd(task);
 601                 swd = get_fpu_swd(task);
 602
 603                 err = swd & ~cwd;
 604         } else {
 605                 /*
 606                  * The SIMD FPU exceptions are handled a little differently, as there
 607                  * is only a single status/control register.  Thus, to determine which
 608                  * unmasked exception was caught we must mask the exception mask bits
 609                  * at 0x1f80, and then use these to mask the exception bits at 0x3f.
 610                  */
 611                 unsigned short mxcsr = get_fpu_mxcsr(task);
 612                 err = ~(mxcsr >> 7) & mxcsr;
 613         }
 614
 615         if (err & 0x001) {      /* Invalid op */
 616                 /*
 617                  * swd & 0x240 == 0x040: Stack Underflow
 618                  * swd & 0x240 == 0x240: Stack Overflow
 619                  * User must clear the SF bit (0x40) if set
 620                  */
 621                 info.si_code = FPE_FLTINV;
 622         } else if (err & 0x004) { /* Divide by Zero */
 623                 info.si_code = FPE_FLTDIV;
 624         } else if (err & 0x008) { /* Overflow */
 625                 info.si_code = FPE_FLTOVF;
 626         } else if (err & 0x012) { /* Denormal, Underflow */
 627                 info.si_code = FPE_FLTUND;
 628         } else if (err & 0x020) { /* Precision */
 629                 info.si_code = FPE_FLTRES;
 630         } else {
 631                 /*
 632                  * If we're using IRQ 13, or supposedly even some trap
 633                  * X86_TRAP_MF implementations, it's possible
 634                  * we get a spurious trap, which is not an error.
 635                  */
 636                 return;
 637         }
 638         force_sig_info(SIGFPE, &info, task);
 639 }
 640
 641 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
 642 {
 643         enum ctx_state prev_state;
 644
 645         prev_state = exception_enter();
 646         math_error(regs, error_code, X86_TRAP_MF);
 647         exception_exit(prev_state);
 648 }
 649
 650 dotraplinkage void
 651 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
 652 {
 653         enum ctx_state prev_state;
 654
 655         prev_state = exception_enter();
 656         math_error(regs, error_code, X86_TRAP_XF);
 657         exception_exit(prev_state);
 658 }
 659
 660 dotraplinkage void
 661 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
 662 {
 663         conditional_sti(regs);
 664 #if 0
 665         /* No need to warn about this any longer. */
 666         pr_info("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
 667 #endif
 668 }
 669
 670 asmlinkage __visible void __attribute__((weak)) smp_thermal_interrupt(void)
 671 {
 672 }
 673
 674 asmlinkage __visible void __attribute__((weak)) smp_threshold_interrupt(void)
 675 {
 676 }
 677
 678 /*
 679  * 'math_state_restore()' saves the current math information in the
 680  * old math state array, and gets the new ones from the current task
 681  *
 682  * Careful.. There are problems with IBM-designed IRQ13 behaviour.
 683  * Don't touch unless you *really* know how it works.
 684  *
 685  * Must be called with kernel preemption disabled (eg with local
 686  * local interrupts as in the case of do_device_not_available).
 687  */
 688 void math_state_restore(void)
 689 {
 690         struct task_struct *tsk = current;
 691
 692         if (!tsk_used_math(tsk)) {
 693                 local_irq_enable();
 694                 /*
 695                  * does a slab alloc which can sleep
 696                  */
 697                 if (init_fpu(tsk)) {
 698                         /*
 699                          * ran out of memory!
 700                          */
 701                         do_group_exit(SIGKILL);
 702                         return;
 703                 }
 704                 local_irq_disable();
 705         }
 706
 707         __thread_fpu_begin(tsk);
 708
 709         /*
 710          * Paranoid restore. send a SIGSEGV if we fail to restore the state.
 711          */
 712         if (unlikely(restore_fpu_checking(tsk))) {
 713                 drop_init_fpu(tsk);
 714                 force_sig_info(SIGSEGV, SEND_SIG_PRIV, tsk);
 715                 return;
 716         }
 717
 718         tsk->thread.fpu_counter++;
 719 }
 720 EXPORT_SYMBOL_GPL(math_state_restore);
 721
 722 dotraplinkage void
 723 do_device_not_available(struct pt_regs *regs, long error_code)
 724 {
 725         enum ctx_state prev_state;
 726
 727         prev_state = exception_enter();
 728         BUG_ON(use_eager_fpu());
 729
 730 #ifdef CONFIG_MATH_EMULATION
 731         if (read_cr0() & X86_CR0_EM) {
 732                 struct math_emu_info info = { };
 733
 734                 conditional_sti(regs);
 735
 736                 info.regs = regs;
 737                 math_emulate(&info);
 738                 exception_exit(prev_state);
 739                 return;
 740         }
 741 #endif
 742         math_state_restore(); /* interrupts still off */
 743 #ifdef CONFIG_X86_32
 744         conditional_sti(regs);
 745 #endif
 746         exception_exit(prev_state);
 747 }
 748 NOKPROBE_SYMBOL(do_device_not_available);
 749
 750 #ifdef CONFIG_X86_32
 751 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
 752 {
 753         siginfo_t info;
 754         enum ctx_state prev_state;
 755
 756         prev_state = exception_enter();
 757         local_irq_enable();
 758
 759         info.si_signo = SIGILL;
 760         info.si_errno = 0;
 761         info.si_code = ILL_BADSTK;
 762         info.si_addr = NULL;
 763         if (notify_die(DIE_TRAP, "iret exception", regs, error_code,
 764                         X86_TRAP_IRET, SIGILL) != NOTIFY_STOP) {
 765                 do_trap(X86_TRAP_IRET, SIGILL, "iret exception", regs, error_code,
 766                         &info);
 767         }
 768         exception_exit(prev_state);
 769 }
 770 #endif
 771
 772 /* Set of traps needed for early debugging. */
 773 void __init early_trap_init(void)
 774 {
 775         set_intr_gate_ist(X86_TRAP_DB, &debug, DEBUG_STACK);
 776         /* int3 can be called from all */
 777         set_system_intr_gate_ist(X86_TRAP_BP, &int3, DEBUG_STACK);
 778 #ifdef CONFIG_X86_32
 779         set_intr_gate(X86_TRAP_PF, page_fault);
 780 #endif
 781         load_idt(&idt_descr);
 782 }
 783
 784 void __init early_trap_pf_init(void)
 785 {
 786 #ifdef CONFIG_X86_64
 787         set_intr_gate(X86_TRAP_PF, page_fault);
 788 #endif
 789 }
 790
 791 void __init trap_init(void)
 792 {
 793         int i;
 794
 795 #ifdef CONFIG_EISA
 796         void __iomem *p = early_ioremap(0x0FFFD9, 4);
 797
 798         if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
 799                 EISA_bus = 1;
 800         early_iounmap(p, 4);
 801 #endif
 802
 803         set_intr_gate(X86_TRAP_DE, divide_error);
 804         set_intr_gate_ist(X86_TRAP_NMI, &nmi, NMI_STACK);
 805         /* int4 can be called from all */
 806         set_system_intr_gate(X86_TRAP_OF, &overflow);
 807         set_intr_gate(X86_TRAP_BR, bounds);
 808         set_intr_gate(X86_TRAP_UD, invalid_op);
 809         set_intr_gate(X86_TRAP_NM, device_not_available);
 810 #ifdef CONFIG_X86_32
 811         set_task_gate(X86_TRAP_DF, GDT_ENTRY_DOUBLEFAULT_TSS);
 812 #else
 813         set_intr_gate_ist(X86_TRAP_DF, &double_fault, DOUBLEFAULT_STACK);
 814 #endif
 815         set_intr_gate(X86_TRAP_OLD_MF, coprocessor_segment_overrun);
 816         set_intr_gate(X86_TRAP_TS, invalid_TSS);
 817         set_intr_gate(X86_TRAP_NP, segment_not_present);
 818         set_intr_gate(X86_TRAP_SS, stack_segment);
 819         set_intr_gate(X86_TRAP_GP, general_protection);
 820         set_intr_gate(X86_TRAP_SPURIOUS, spurious_interrupt_bug);
 821         set_intr_gate(X86_TRAP_MF, coprocessor_error);
 822         set_intr_gate(X86_TRAP_AC, alignment_check);
 823 #ifdef CONFIG_X86_MCE
 824         set_intr_gate_ist(X86_TRAP_MC, &machine_check, MCE_STACK);
 825 #endif
 826         set_intr_gate(X86_TRAP_XF, simd_coprocessor_error);
 827
 828         /* Reserve all the builtin and the syscall vector: */
 829         for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
 830                 set_bit(i, used_vectors);
 831
 832 #ifdef CONFIG_IA32_EMULATION
 833         set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
 834         set_bit(IA32_SYSCALL_VECTOR, used_vectors);
 835 #endif
 836
 837 #ifdef CONFIG_X86_32
 838         set_system_trap_gate(SYSCALL_VECTOR, &system_call);
 839         set_bit(SYSCALL_VECTOR, used_vectors);
 840 #endif
 841
 842         /*
 843          * Set the IDT descriptor to a fixed read-only location, so that the
 844          * "sidt" instruction will not leak the location of the kernel, and
 845          * to defend the IDT against arbitrary memory write vulnerabilities.
 846          * It will be reloaded in cpu_init() */
 847         __set_fixmap(FIX_RO_IDT, __pa_symbol(idt_table), PAGE_KERNEL_RO);
 848         idt_descr.address = fix_to_virt(FIX_RO_IDT);
 849
 850         /*
 851          * Should be a barrier for any external CPU state:
 852          */
 853         cpu_init();
 854
 855         x86_init.irqs.trap_init();
 856
 857 #ifdef CONFIG_X86_64
 858         memcpy(&debug_idt_table, &idt_table, IDT_ENTRIES * 16);
 859         set_nmi_gate(X86_TRAP_DB, &debug);
 860         set_nmi_gate(X86_TRAP_BP, &int3);
 861 #endif
 862 }