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 #ifdef CONFIG_X86_32
 237 DO_ERROR(X86_TRAP_SS,     SIGBUS,  "stack segment",             stack_segment)
 238 #endif
 239 DO_ERROR(X86_TRAP_AC,     SIGBUS,  "alignment check",           alignment_check)
 240
 241 #ifdef CONFIG_X86_64
 242 /* Runs on IST stack */
 243 dotraplinkage void do_stack_segment(struct pt_regs *regs, long error_code)
 244 {
 245         enum ctx_state prev_state;
 246
 247         prev_state = exception_enter();
 248         if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
 249                        X86_TRAP_SS, SIGBUS) != NOTIFY_STOP) {
 250                 preempt_conditional_sti(regs);
 251                 do_trap(X86_TRAP_SS, SIGBUS, "stack segment", regs, error_code, NULL);
 252                 preempt_conditional_cli(regs);
 253         }
 254         exception_exit(prev_state);
 255 }
 256
 257 dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
 258 {
 259         static const char str[] = "double fault";
 260         struct task_struct *tsk = current;
 261
 262 #ifdef CONFIG_X86_ESPFIX64
 263         extern unsigned char native_irq_return_iret[];
 264
 265         /*
 266          * If IRET takes a non-IST fault on the espfix64 stack, then we
 267          * end up promoting it to a doublefault.  In that case, modify
 268          * the stack to make it look like we just entered the #GP
 269          * handler from user space, similar to bad_iret.
 270          */
 271         if (((long)regs->sp >> PGDIR_SHIFT) == ESPFIX_PGD_ENTRY &&
 272                 regs->cs == __KERNEL_CS &&
 273                 regs->ip == (unsigned long)native_irq_return_iret)
 274         {
 275                 struct pt_regs *normal_regs = task_pt_regs(current);
 276
 277                 /* Fake a #GP(0) from userspace. */
 278                 memmove(&normal_regs->ip, (void *)regs->sp, 5*8);
 279                 normal_regs->orig_ax = 0;  /* Missing (lost) #GP error code */
 280                 regs->ip = (unsigned long)general_protection;
 281                 regs->sp = (unsigned long)&normal_regs->orig_ax;
 282                 return;
 283         }
 284 #endif
 285
 286         exception_enter();
 287         /* Return not checked because double check cannot be ignored */
 288         notify_die(DIE_TRAP, str, regs, error_code, X86_TRAP_DF, SIGSEGV);
 289
 290         tsk->thread.error_code = error_code;
 291         tsk->thread.trap_nr = X86_TRAP_DF;
 292
 293 #ifdef CONFIG_DOUBLEFAULT
 294         df_debug(regs, error_code);
 295 #endif
 296         /*
 297          * This is always a kernel trap and never fixable (and thus must
 298          * never return).
 299          */
 300         for (;;)
 301                 die(str, regs, error_code);
 302 }
 303 #endif
 304
 305 dotraplinkage void
 306 do_general_protection(struct pt_regs *regs, long error_code)
 307 {
 308         struct task_struct *tsk;
 309         enum ctx_state prev_state;
 310
 311         prev_state = exception_enter();
 312         conditional_sti(regs);
 313
 314 #ifdef CONFIG_X86_32
 315         if (regs->flags & X86_VM_MASK) {
 316                 local_irq_enable();
 317                 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
 318                 goto exit;
 319         }
 320 #endif
 321
 322         tsk = current;
 323         if (!user_mode(regs)) {
 324                 if (fixup_exception(regs))
 325                         goto exit;
 326
 327                 tsk->thread.error_code = error_code;
 328                 tsk->thread.trap_nr = X86_TRAP_GP;
 329                 if (notify_die(DIE_GPF, "general protection fault", regs, error_code,
 330                                X86_TRAP_GP, SIGSEGV) != NOTIFY_STOP)
 331                         die("general protection fault", regs, error_code);
 332                 goto exit;
 333         }
 334
 335         tsk->thread.error_code = error_code;
 336         tsk->thread.trap_nr = X86_TRAP_GP;
 337
 338         if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
 339                         printk_ratelimit()) {
 340                 pr_info("%s[%d] general protection ip:%lx sp:%lx error:%lx",
 341                         tsk->comm, task_pid_nr(tsk),
 342                         regs->ip, regs->sp, error_code);
 343                 print_vma_addr(" in ", regs->ip);
 344                 pr_cont("\n");
 345         }
 346
 347         force_sig_info(SIGSEGV, SEND_SIG_PRIV, tsk);
 348 exit:
 349         exception_exit(prev_state);
 350 }
 351 NOKPROBE_SYMBOL(do_general_protection);
 352
 353 /* May run on IST stack. */
 354 dotraplinkage void notrace do_int3(struct pt_regs *regs, long error_code)
 355 {
 356         enum ctx_state prev_state;
 357
 358 #ifdef CONFIG_DYNAMIC_FTRACE
 359         /*
 360          * ftrace must be first, everything else may cause a recursive crash.
 361          * See note by declaration of modifying_ftrace_code in ftrace.c
 362          */
 363         if (unlikely(atomic_read(&modifying_ftrace_code)) &&
 364             ftrace_int3_handler(regs))
 365                 return;
 366 #endif
 367         if (poke_int3_handler(regs))
 368                 return;
 369
 370         prev_state = exception_enter();
 371 #ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
 372         if (kgdb_ll_trap(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
 373                                 SIGTRAP) == NOTIFY_STOP)
 374                 goto exit;
 375 #endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
 376
 377 #ifdef CONFIG_KPROBES
 378         if (kprobe_int3_handler(regs))
 379                 goto exit;
 380 #endif
 381
 382         if (notify_die(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
 383                         SIGTRAP) == NOTIFY_STOP)
 384                 goto exit;
 385
 386         /*
 387          * Let others (NMI) know that the debug stack is in use
 388          * as we may switch to the interrupt stack.
 389          */
 390         debug_stack_usage_inc();
 391         preempt_conditional_sti(regs);
 392         do_trap(X86_TRAP_BP, SIGTRAP, "int3", regs, error_code, NULL);
 393         preempt_conditional_cli(regs);
 394         debug_stack_usage_dec();
 395 exit:
 396         exception_exit(prev_state);
 397 }
 398 NOKPROBE_SYMBOL(do_int3);
 399
 400 #ifdef CONFIG_X86_64
 401 /*
 402  * Help handler running on IST stack to switch back to user stack
 403  * for scheduling or signal handling. The actual stack switch is done in
 404  * entry.S
 405  */
 406 asmlinkage __visible struct pt_regs *sync_regs(struct pt_regs *eregs)
 407 {
 408         struct pt_regs *regs = eregs;
 409         /* Did already sync */
 410         if (eregs == (struct pt_regs *)eregs->sp)
 411                 ;
 412         /* Exception from user space */
 413         else if (user_mode(eregs))
 414                 regs = task_pt_regs(current);
 415         /*
 416          * Exception from kernel and interrupts are enabled. Move to
 417          * kernel process stack.
 418          */
 419         else if (eregs->flags & X86_EFLAGS_IF)
 420                 regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
 421         if (eregs != regs)
 422                 *regs = *eregs;
 423         return regs;
 424 }
 425 NOKPROBE_SYMBOL(sync_regs);
 426 #endif
 427
 428 /*
 429  * Our handling of the processor debug registers is non-trivial.
 430  * We do not clear them on entry and exit from the kernel. Therefore
 431  * it is possible to get a watchpoint trap here from inside the kernel.
 432  * However, the code in ./ptrace.c has ensured that the user can
 433  * only set watchpoints on userspace addresses. Therefore the in-kernel
 434  * watchpoint trap can only occur in code which is reading/writing
 435  * from user space. Such code must not hold kernel locks (since it
 436  * can equally take a page fault), therefore it is safe to call
 437  * force_sig_info even though that claims and releases locks.
 438  *
 439  * Code in ./signal.c ensures that the debug control register
 440  * is restored before we deliver any signal, and therefore that
 441  * user code runs with the correct debug control register even though
 442  * we clear it here.
 443  *
 444  * Being careful here means that we don't have to be as careful in a
 445  * lot of more complicated places (task switching can be a bit lazy
 446  * about restoring all the debug state, and ptrace doesn't have to
 447  * find every occurrence of the TF bit that could be saved away even
 448  * by user code)
 449  *
 450  * May run on IST stack.
 451  */
 452 dotraplinkage void do_debug(struct pt_regs *regs, long error_code)
 453 {
 454         struct task_struct *tsk = current;
 455         enum ctx_state prev_state;
 456         int user_icebp = 0;
 457         unsigned long dr6;
 458         int si_code;
 459
 460         prev_state = exception_enter();
 461
 462         get_debugreg(dr6, 6);
 463
 464         /* Filter out all the reserved bits which are preset to 1 */
 465         dr6 &= ~DR6_RESERVED;
 466
 467         /*
 468          * If dr6 has no reason to give us about the origin of this trap,
 469          * then it's very likely the result of an icebp/int01 trap.
 470          * User wants a sigtrap for that.
 471          */
 472         if (!dr6 && user_mode(regs))
 473                 user_icebp = 1;
 474
 475         /* Catch kmemcheck conditions first of all! */
 476         if ((dr6 & DR_STEP) && kmemcheck_trap(regs))
 477                 goto exit;
 478
 479         /* DR6 may or may not be cleared by the CPU */
 480         set_debugreg(0, 6);
 481
 482         /*
 483          * The processor cleared BTF, so don't mark that we need it set.
 484          */
 485         clear_tsk_thread_flag(tsk, TIF_BLOCKSTEP);
 486
 487         /* Store the virtualized DR6 value */
 488         tsk->thread.debugreg6 = dr6;
 489
 490 #ifdef CONFIG_KPROBES
 491         if (kprobe_debug_handler(regs))
 492                 goto exit;
 493 #endif
 494
 495         if (notify_die(DIE_DEBUG, "debug", regs, (long)&dr6, error_code,
 496                                                         SIGTRAP) == NOTIFY_STOP)
 497                 goto exit;
 498
 499         /*
 500          * Let others (NMI) know that the debug stack is in use
 501          * as we may switch to the interrupt stack.
 502          */
 503         debug_stack_usage_inc();
 504
 505         /* It's safe to allow irq's after DR6 has been saved */
 506         preempt_conditional_sti(regs);
 507
 508         if (regs->flags & X86_VM_MASK) {
 509                 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code,
 510                                         X86_TRAP_DB);
 511                 preempt_conditional_cli(regs);
 512                 debug_stack_usage_dec();
 513                 goto exit;
 514         }
 515
 516         /*
 517          * Single-stepping through system calls: ignore any exceptions in
 518          * kernel space, but re-enable TF when returning to user mode.
 519          *
 520          * We already checked v86 mode above, so we can check for kernel mode
 521          * by just checking the CPL of CS.
 522          */
 523         if ((dr6 & DR_STEP) && !user_mode(regs)) {
 524                 tsk->thread.debugreg6 &= ~DR_STEP;
 525                 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
 526                 regs->flags &= ~X86_EFLAGS_TF;
 527         }
 528         si_code = get_si_code(tsk->thread.debugreg6);
 529         if (tsk->thread.debugreg6 & (DR_STEP | DR_TRAP_BITS) || user_icebp)
 530                 send_sigtrap(tsk, regs, error_code, si_code);
 531         preempt_conditional_cli(regs);
 532         debug_stack_usage_dec();
 533
 534 exit:
 535         exception_exit(prev_state);
 536 }
 537 NOKPROBE_SYMBOL(do_debug);
 538
 539 /*
 540  * Note that we play around with the 'TS' bit in an attempt to get
 541  * the correct behaviour even in the presence of the asynchronous
 542  * IRQ13 behaviour
 543  */
 544 static void math_error(struct pt_regs *regs, int error_code, int trapnr)
 545 {
 546         struct task_struct *task = current;
 547         siginfo_t info;
 548         unsigned short err;
 549         char *str = (trapnr == X86_TRAP_MF) ? "fpu exception" :
 550                                                 "simd exception";
 551
 552         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, SIGFPE) == NOTIFY_STOP)
 553                 return;
 554         conditional_sti(regs);
 555
 556         if (!user_mode_vm(regs))
 557         {
 558                 if (!fixup_exception(regs)) {
 559                         task->thread.error_code = error_code;
 560                         task->thread.trap_nr = trapnr;
 561                         die(str, regs, error_code);
 562                 }
 563                 return;
 564         }
 565
 566         /*
 567          * Save the info for the exception handler and clear the error.
 568          */
 569         save_init_fpu(task);
 570         task->thread.trap_nr = trapnr;
 571         task->thread.error_code = error_code;
 572         info.si_signo = SIGFPE;
 573         info.si_errno = 0;
 574         info.si_addr = (void __user *)uprobe_get_trap_addr(regs);
 575         if (trapnr == X86_TRAP_MF) {
 576                 unsigned short cwd, swd;
 577                 /*
 578                  * (~cwd & swd) will mask out exceptions that are not set to unmasked
 579                  * status.  0x3f is the exception bits in these regs, 0x200 is the
 580                  * C1 reg you need in case of a stack fault, 0x040 is the stack
 581                  * fault bit.  We should only be taking one exception at a time,
 582                  * so if this combination doesn't produce any single exception,
 583                  * then we have a bad program that isn't synchronizing its FPU usage
 584                  * and it will suffer the consequences since we won't be able to
 585                  * fully reproduce the context of the exception
 586                  */
 587                 cwd = get_fpu_cwd(task);
 588                 swd = get_fpu_swd(task);
 589
 590                 err = swd & ~cwd;
 591         } else {
 592                 /*
 593                  * The SIMD FPU exceptions are handled a little differently, as there
 594                  * is only a single status/control register.  Thus, to determine which
 595                  * unmasked exception was caught we must mask the exception mask bits
 596                  * at 0x1f80, and then use these to mask the exception bits at 0x3f.
 597                  */
 598                 unsigned short mxcsr = get_fpu_mxcsr(task);
 599                 err = ~(mxcsr >> 7) & mxcsr;
 600         }
 601
 602         if (err & 0x001) {      /* Invalid op */
 603                 /*
 604                  * swd & 0x240 == 0x040: Stack Underflow
 605                  * swd & 0x240 == 0x240: Stack Overflow
 606                  * User must clear the SF bit (0x40) if set
 607                  */
 608                 info.si_code = FPE_FLTINV;
 609         } else if (err & 0x004) { /* Divide by Zero */
 610                 info.si_code = FPE_FLTDIV;
 611         } else if (err & 0x008) { /* Overflow */
 612                 info.si_code = FPE_FLTOVF;
 613         } else if (err & 0x012) { /* Denormal, Underflow */
 614                 info.si_code = FPE_FLTUND;
 615         } else if (err & 0x020) { /* Precision */
 616                 info.si_code = FPE_FLTRES;
 617         } else {
 618                 /*
 619                  * If we're using IRQ 13, or supposedly even some trap
 620                  * X86_TRAP_MF implementations, it's possible
 621                  * we get a spurious trap, which is not an error.
 622                  */
 623                 return;
 624         }
 625         force_sig_info(SIGFPE, &info, task);
 626 }
 627
 628 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
 629 {
 630         enum ctx_state prev_state;
 631
 632         prev_state = exception_enter();
 633         math_error(regs, error_code, X86_TRAP_MF);
 634         exception_exit(prev_state);
 635 }
 636
 637 dotraplinkage void
 638 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
 639 {
 640         enum ctx_state prev_state;
 641
 642         prev_state = exception_enter();
 643         math_error(regs, error_code, X86_TRAP_XF);
 644         exception_exit(prev_state);
 645 }
 646
 647 dotraplinkage void
 648 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
 649 {
 650         conditional_sti(regs);
 651 #if 0
 652         /* No need to warn about this any longer. */
 653         pr_info("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
 654 #endif
 655 }
 656
 657 asmlinkage __visible void __attribute__((weak)) smp_thermal_interrupt(void)
 658 {
 659 }
 660
 661 asmlinkage __visible void __attribute__((weak)) smp_threshold_interrupt(void)
 662 {
 663 }
 664
 665 /*
 666  * 'math_state_restore()' saves the current math information in the
 667  * old math state array, and gets the new ones from the current task
 668  *
 669  * Careful.. There are problems with IBM-designed IRQ13 behaviour.
 670  * Don't touch unless you *really* know how it works.
 671  *
 672  * Must be called with kernel preemption disabled (eg with local
 673  * local interrupts as in the case of do_device_not_available).
 674  */
 675 void math_state_restore(void)
 676 {
 677         struct task_struct *tsk = current;
 678
 679         if (!tsk_used_math(tsk)) {
 680                 local_irq_enable();
 681                 /*
 682                  * does a slab alloc which can sleep
 683                  */
 684                 if (init_fpu(tsk)) {
 685                         /*
 686                          * ran out of memory!
 687                          */
 688                         do_group_exit(SIGKILL);
 689                         return;
 690                 }
 691                 local_irq_disable();
 692         }
 693
 694         __thread_fpu_begin(tsk);
 695
 696         /*
 697          * Paranoid restore. send a SIGSEGV if we fail to restore the state.
 698          */
 699         if (unlikely(restore_fpu_checking(tsk))) {
 700                 drop_init_fpu(tsk);
 701                 force_sig_info(SIGSEGV, SEND_SIG_PRIV, tsk);
 702                 return;
 703         }
 704
 705         tsk->thread.fpu_counter++;
 706 }
 707 EXPORT_SYMBOL_GPL(math_state_restore);
 708
 709 dotraplinkage void
 710 do_device_not_available(struct pt_regs *regs, long error_code)
 711 {
 712         enum ctx_state prev_state;
 713
 714         prev_state = exception_enter();
 715         BUG_ON(use_eager_fpu());
 716
 717 #ifdef CONFIG_MATH_EMULATION
 718         if (read_cr0() & X86_CR0_EM) {
 719                 struct math_emu_info info = { };
 720
 721                 conditional_sti(regs);
 722
 723                 info.regs = regs;
 724                 math_emulate(&info);
 725                 exception_exit(prev_state);
 726                 return;
 727         }
 728 #endif
 729         math_state_restore(); /* interrupts still off */
 730 #ifdef CONFIG_X86_32
 731         conditional_sti(regs);
 732 #endif
 733         exception_exit(prev_state);
 734 }
 735 NOKPROBE_SYMBOL(do_device_not_available);
 736
 737 #ifdef CONFIG_X86_32
 738 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
 739 {
 740         siginfo_t info;
 741         enum ctx_state prev_state;
 742
 743         prev_state = exception_enter();
 744         local_irq_enable();
 745
 746         info.si_signo = SIGILL;
 747         info.si_errno = 0;
 748         info.si_code = ILL_BADSTK;
 749         info.si_addr = NULL;
 750         if (notify_die(DIE_TRAP, "iret exception", regs, error_code,
 751                         X86_TRAP_IRET, SIGILL) != NOTIFY_STOP) {
 752                 do_trap(X86_TRAP_IRET, SIGILL, "iret exception", regs, error_code,
 753                         &info);
 754         }
 755         exception_exit(prev_state);
 756 }
 757 #endif
 758
 759 /* Set of traps needed for early debugging. */
 760 void __init early_trap_init(void)
 761 {
 762         set_intr_gate_ist(X86_TRAP_DB, &debug, DEBUG_STACK);
 763         /* int3 can be called from all */
 764         set_system_intr_gate_ist(X86_TRAP_BP, &int3, DEBUG_STACK);
 765 #ifdef CONFIG_X86_32
 766         set_intr_gate(X86_TRAP_PF, page_fault);
 767 #endif
 768         load_idt(&idt_descr);
 769 }
 770
 771 void __init early_trap_pf_init(void)
 772 {
 773 #ifdef CONFIG_X86_64
 774         set_intr_gate(X86_TRAP_PF, page_fault);
 775 #endif
 776 }
 777
 778 void __init trap_init(void)
 779 {
 780         int i;
 781
 782 #ifdef CONFIG_EISA
 783         void __iomem *p = early_ioremap(0x0FFFD9, 4);
 784
 785         if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
 786                 EISA_bus = 1;
 787         early_iounmap(p, 4);
 788 #endif
 789
 790         set_intr_gate(X86_TRAP_DE, divide_error);
 791         set_intr_gate_ist(X86_TRAP_NMI, &nmi, NMI_STACK);
 792         /* int4 can be called from all */
 793         set_system_intr_gate(X86_TRAP_OF, &overflow);
 794         set_intr_gate(X86_TRAP_BR, bounds);
 795         set_intr_gate(X86_TRAP_UD, invalid_op);
 796         set_intr_gate(X86_TRAP_NM, device_not_available);
 797 #ifdef CONFIG_X86_32
 798         set_task_gate(X86_TRAP_DF, GDT_ENTRY_DOUBLEFAULT_TSS);
 799 #else
 800         set_intr_gate_ist(X86_TRAP_DF, &double_fault, DOUBLEFAULT_STACK);
 801 #endif
 802         set_intr_gate(X86_TRAP_OLD_MF, coprocessor_segment_overrun);
 803         set_intr_gate(X86_TRAP_TS, invalid_TSS);
 804         set_intr_gate(X86_TRAP_NP, segment_not_present);
 805         set_intr_gate_ist(X86_TRAP_SS, &stack_segment, STACKFAULT_STACK);
 806         set_intr_gate(X86_TRAP_GP, general_protection);
 807         set_intr_gate(X86_TRAP_SPURIOUS, spurious_interrupt_bug);
 808         set_intr_gate(X86_TRAP_MF, coprocessor_error);
 809         set_intr_gate(X86_TRAP_AC, alignment_check);
 810 #ifdef CONFIG_X86_MCE
 811         set_intr_gate_ist(X86_TRAP_MC, &machine_check, MCE_STACK);
 812 #endif
 813         set_intr_gate(X86_TRAP_XF, simd_coprocessor_error);
 814
 815         /* Reserve all the builtin and the syscall vector: */
 816         for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
 817                 set_bit(i, used_vectors);
 818
 819 #ifdef CONFIG_IA32_EMULATION
 820         set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
 821         set_bit(IA32_SYSCALL_VECTOR, used_vectors);
 822 #endif
 823
 824 #ifdef CONFIG_X86_32
 825         set_system_trap_gate(SYSCALL_VECTOR, &system_call);
 826         set_bit(SYSCALL_VECTOR, used_vectors);
 827 #endif
 828
 829         /*
 830          * Set the IDT descriptor to a fixed read-only location, so that the
 831          * "sidt" instruction will not leak the location of the kernel, and
 832          * to defend the IDT against arbitrary memory write vulnerabilities.
 833          * It will be reloaded in cpu_init() */
 834         __set_fixmap(FIX_RO_IDT, __pa_symbol(idt_table), PAGE_KERNEL_RO);
 835         idt_descr.address = fix_to_virt(FIX_RO_IDT);
 836
 837         /*
 838          * Should be a barrier for any external CPU state:
 839          */
 840         cpu_init();
 841
 842         x86_init.irqs.trap_init();
 843
 844 #ifdef CONFIG_X86_64
 845         memcpy(&debug_idt_table, &idt_table, IDT_ENTRIES * 16);
 846         set_nmi_gate(X86_TRAP_DB, &debug);
 847         set_nmi_gate(X86_TRAP_BP, &int3);
 848 #endif
 849 }