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/fpu/internal.h>
  58 #include <asm/mce.h>
  59 #include <asm/fixmap.h>
  60 #include <asm/mach_traps.h>
  61 #include <asm/alternative.h>
  62 #include <asm/fpu/xstate.h>
  63 #include <asm/trace/mpx.h>
  64 #include <asm/mpx.h>
  65 #include <asm/vm86.h>
  66
  67 #ifdef CONFIG_X86_64
  68 #include <asm/x86_init.h>
  69 #include <asm/pgalloc.h>
  70 #include <asm/proto.h>
  71
  72 /* No need to be aligned, but done to keep all IDTs defined the same way. */
  73 gate_desc debug_idt_table[NR_VECTORS] __page_aligned_bss;
  74 #else
  75 #include <asm/processor-flags.h>
  76 #include <asm/setup.h>
  77 #include <asm/proto.h>
  78 #endif
  79
  80 /* Must be page-aligned because the real IDT is used in a fixmap. */
  81 gate_desc idt_table[NR_VECTORS] __page_aligned_bss;
  82
  83 DECLARE_BITMAP(used_vectors, NR_VECTORS);
  84 EXPORT_SYMBOL_GPL(used_vectors);
  85
  86 static inline void conditional_sti(struct pt_regs *regs)
  87 {
  88         if (regs->flags & X86_EFLAGS_IF)
  89                 local_irq_enable();
  90 }
  91
  92 static inline void preempt_conditional_sti(struct pt_regs *regs)
  93 {
  94         preempt_count_inc();
  95         if (regs->flags & X86_EFLAGS_IF)
  96                 local_irq_enable();
  97 }
  98
  99 static inline void conditional_cli(struct pt_regs *regs)
 100 {
 101         if (regs->flags & X86_EFLAGS_IF)
 102                 local_irq_disable();
 103 }
 104
 105 static inline void preempt_conditional_cli(struct pt_regs *regs)
 106 {
 107         if (regs->flags & X86_EFLAGS_IF)
 108                 local_irq_disable();
 109         preempt_count_dec();
 110 }
 111
 112 void ist_enter(struct pt_regs *regs)
 113 {
 114         if (user_mode(regs)) {
 115                 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
 116         } else {
 117                 /*
 118                  * We might have interrupted pretty much anything.  In
 119                  * fact, if we're a machine check, we can even interrupt
 120                  * NMI processing.  We don't want in_nmi() to return true,
 121                  * but we need to notify RCU.
 122                  */
 123                 rcu_nmi_enter();
 124         }
 125
 126         /*
 127          * We are atomic because we're on the IST stack; or we're on
 128          * x86_32, in which case we still shouldn't schedule; or we're
 129          * on x86_64 and entered from user mode, in which case we're
 130          * still atomic unless ist_begin_non_atomic is called.
 131          */
 132         preempt_count_add(HARDIRQ_OFFSET);
 133
 134         /* This code is a bit fragile.  Test it. */
 135         RCU_LOCKDEP_WARN(!rcu_is_watching(), "ist_enter didn't work");
 136 }
 137
 138 void ist_exit(struct pt_regs *regs)
 139 {
 140         preempt_count_sub(HARDIRQ_OFFSET);
 141
 142         if (!user_mode(regs))
 143                 rcu_nmi_exit();
 144 }
 145
 146 /**
 147  * ist_begin_non_atomic() - begin a non-atomic section in an IST exception
 148  * @regs:       regs passed to the IST exception handler
 149  *
 150  * IST exception handlers normally cannot schedule.  As a special
 151  * exception, if the exception interrupted userspace code (i.e.
 152  * user_mode(regs) would return true) and the exception was not
 153  * a double fault, it can be safe to schedule.  ist_begin_non_atomic()
 154  * begins a non-atomic section within an ist_enter()/ist_exit() region.
 155  * Callers are responsible for enabling interrupts themselves inside
 156  * the non-atomic section, and callers must call ist_end_non_atomic()
 157  * before ist_exit().
 158  */
 159 void ist_begin_non_atomic(struct pt_regs *regs)
 160 {
 161         BUG_ON(!user_mode(regs));
 162
 163         /*
 164          * Sanity check: we need to be on the normal thread stack.  This
 165          * will catch asm bugs and any attempt to use ist_preempt_enable
 166          * from double_fault.
 167          */
 168         BUG_ON((unsigned long)(current_top_of_stack() -
 169                                current_stack_pointer()) >= THREAD_SIZE);
 170
 171         preempt_count_sub(HARDIRQ_OFFSET);
 172 }
 173
 174 /**
 175  * ist_end_non_atomic() - begin a non-atomic section in an IST exception
 176  *
 177  * Ends a non-atomic section started with ist_begin_non_atomic().
 178  */
 179 void ist_end_non_atomic(void)
 180 {
 181         preempt_count_add(HARDIRQ_OFFSET);
 182 }
 183
 184 static nokprobe_inline int
 185 do_trap_no_signal(struct task_struct *tsk, int trapnr, char *str,
 186                   struct pt_regs *regs, long error_code)
 187 {
 188         if (v8086_mode(regs)) {
 189                 /*
 190                  * Traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
 191                  * On nmi (interrupt 2), do_trap should not be called.
 192                  */
 193                 if (trapnr < X86_TRAP_UD) {
 194                         if (!handle_vm86_trap((struct kernel_vm86_regs *) regs,
 195                                                 error_code, trapnr))
 196                                 return 0;
 197                 }
 198                 return -1;
 199         }
 200
 201         if (!user_mode(regs)) {
 202                 if (!fixup_exception(regs)) {
 203                         tsk->thread.error_code = error_code;
 204                         tsk->thread.trap_nr = trapnr;
 205                         die(str, regs, error_code);
 206                 }
 207                 return 0;
 208         }
 209
 210         return -1;
 211 }
 212
 213 static siginfo_t *fill_trap_info(struct pt_regs *regs, int signr, int trapnr,
 214                                 siginfo_t *info)
 215 {
 216         unsigned long siaddr;
 217         int sicode;
 218
 219         switch (trapnr) {
 220         default:
 221                 return SEND_SIG_PRIV;
 222
 223         case X86_TRAP_DE:
 224                 sicode = FPE_INTDIV;
 225                 siaddr = uprobe_get_trap_addr(regs);
 226                 break;
 227         case X86_TRAP_UD:
 228                 sicode = ILL_ILLOPN;
 229                 siaddr = uprobe_get_trap_addr(regs);
 230                 break;
 231         case X86_TRAP_AC:
 232                 sicode = BUS_ADRALN;
 233                 siaddr = 0;
 234                 break;
 235         }
 236
 237         info->si_signo = signr;
 238         info->si_errno = 0;
 239         info->si_code = sicode;
 240         info->si_addr = (void __user *)siaddr;
 241         return info;
 242 }
 243
 244 static void
 245 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
 246         long error_code, siginfo_t *info)
 247 {
 248         struct task_struct *tsk = current;
 249
 250
 251         if (!do_trap_no_signal(tsk, trapnr, str, regs, error_code))
 252                 return;
 253         /*
 254          * We want error_code and trap_nr set for userspace faults and
 255          * kernelspace faults which result in die(), but not
 256          * kernelspace faults which are fixed up.  die() gives the
 257          * process no chance to handle the signal and notice the
 258          * kernel fault information, so that won't result in polluting
 259          * the information about previously queued, but not yet
 260          * delivered, faults.  See also do_general_protection below.
 261          */
 262         tsk->thread.error_code = error_code;
 263         tsk->thread.trap_nr = trapnr;
 264
 265 #ifdef CONFIG_X86_64
 266         if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
 267             printk_ratelimit()) {
 268                 pr_info("%s[%d] trap %s ip:%lx sp:%lx error:%lx",
 269                         tsk->comm, tsk->pid, str,
 270                         regs->ip, regs->sp, error_code);
 271                 print_vma_addr(" in ", regs->ip);
 272                 pr_cont("\n");
 273         }
 274 #endif
 275
 276         force_sig_info(signr, info ?: SEND_SIG_PRIV, tsk);
 277 }
 278 NOKPROBE_SYMBOL(do_trap);
 279
 280 static void do_error_trap(struct pt_regs *regs, long error_code, char *str,
 281                           unsigned long trapnr, int signr)
 282 {
 283         siginfo_t info;
 284
 285         RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
 286
 287         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) !=
 288                         NOTIFY_STOP) {
 289                 conditional_sti(regs);
 290                 do_trap(trapnr, signr, str, regs, error_code,
 291                         fill_trap_info(regs, signr, trapnr, &info));
 292         }
 293 }
 294
 295 #define DO_ERROR(trapnr, signr, str, name)                              \
 296 dotraplinkage void do_##name(struct pt_regs *regs, long error_code)     \
 297 {                                                                       \
 298         do_error_trap(regs, error_code, str, trapnr, signr);            \
 299 }
 300
 301 DO_ERROR(X86_TRAP_DE,     SIGFPE,  "divide error",              divide_error)
 302 DO_ERROR(X86_TRAP_OF,     SIGSEGV, "overflow",                  overflow)
 303 DO_ERROR(X86_TRAP_UD,     SIGILL,  "invalid opcode",            invalid_op)
 304 DO_ERROR(X86_TRAP_OLD_MF, SIGFPE,  "coprocessor segment overrun",coprocessor_segment_overrun)
 305 DO_ERROR(X86_TRAP_TS,     SIGSEGV, "invalid TSS",               invalid_TSS)
 306 DO_ERROR(X86_TRAP_NP,     SIGBUS,  "segment not present",       segment_not_present)
 307 DO_ERROR(X86_TRAP_SS,     SIGBUS,  "stack segment",             stack_segment)
 308 DO_ERROR(X86_TRAP_AC,     SIGBUS,  "alignment check",           alignment_check)
 309
 310 #ifdef CONFIG_X86_64
 311 /* Runs on IST stack */
 312 dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
 313 {
 314         static const char str[] = "double fault";
 315         struct task_struct *tsk = current;
 316
 317 #ifdef CONFIG_X86_ESPFIX64
 318         extern unsigned char native_irq_return_iret[];
 319
 320         /*
 321          * If IRET takes a non-IST fault on the espfix64 stack, then we
 322          * end up promoting it to a doublefault.  In that case, modify
 323          * the stack to make it look like we just entered the #GP
 324          * handler from user space, similar to bad_iret.
 325          *
 326          * No need for ist_enter here because we don't use RCU.
 327          */
 328         if (((long)regs->sp >> PGDIR_SHIFT) == ESPFIX_PGD_ENTRY &&
 329                 regs->cs == __KERNEL_CS &&
 330                 regs->ip == (unsigned long)native_irq_return_iret)
 331         {
 332                 struct pt_regs *normal_regs = task_pt_regs(current);
 333
 334                 /* Fake a #GP(0) from userspace. */
 335                 memmove(&normal_regs->ip, (void *)regs->sp, 5*8);
 336                 normal_regs->orig_ax = 0;  /* Missing (lost) #GP error code */
 337                 regs->ip = (unsigned long)general_protection;
 338                 regs->sp = (unsigned long)&normal_regs->orig_ax;
 339
 340                 return;
 341         }
 342 #endif
 343
 344         ist_enter(regs);
 345         notify_die(DIE_TRAP, str, regs, error_code, X86_TRAP_DF, SIGSEGV);
 346
 347         tsk->thread.error_code = error_code;
 348         tsk->thread.trap_nr = X86_TRAP_DF;
 349
 350 #ifdef CONFIG_DOUBLEFAULT
 351         df_debug(regs, error_code);
 352 #endif
 353         /*
 354          * This is always a kernel trap and never fixable (and thus must
 355          * never return).
 356          */
 357         for (;;)
 358                 die(str, regs, error_code);
 359 }
 360 #endif
 361
 362 dotraplinkage void do_bounds(struct pt_regs *regs, long error_code)
 363 {
 364         const struct mpx_bndcsr *bndcsr;
 365         siginfo_t *info;
 366
 367         RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
 368         if (notify_die(DIE_TRAP, "bounds", regs, error_code,
 369                         X86_TRAP_BR, SIGSEGV) == NOTIFY_STOP)
 370                 return;
 371         conditional_sti(regs);
 372
 373         if (!user_mode(regs))
 374                 die("bounds", regs, error_code);
 375
 376         if (!cpu_feature_enabled(X86_FEATURE_MPX)) {
 377                 /* The exception is not from Intel MPX */
 378                 goto exit_trap;
 379         }
 380
 381         /*
 382          * We need to look at BNDSTATUS to resolve this exception.
 383          * A NULL here might mean that it is in its 'init state',
 384          * which is all zeros which indicates MPX was not
 385          * responsible for the exception.
 386          */
 387         bndcsr = get_xsave_field_ptr(XFEATURE_MASK_BNDCSR);
 388         if (!bndcsr)
 389                 goto exit_trap;
 390
 391         trace_bounds_exception_mpx(bndcsr);
 392         /*
 393          * The error code field of the BNDSTATUS register communicates status
 394          * information of a bound range exception #BR or operation involving
 395          * bound directory.
 396          */
 397         switch (bndcsr->bndstatus & MPX_BNDSTA_ERROR_CODE) {
 398         case 2: /* Bound directory has invalid entry. */
 399                 if (mpx_handle_bd_fault())
 400                         goto exit_trap;
 401                 break; /* Success, it was handled */
 402         case 1: /* Bound violation. */
 403                 info = mpx_generate_siginfo(regs);
 404                 if (IS_ERR(info)) {
 405                         /*
 406                          * We failed to decode the MPX instruction.  Act as if
 407                          * the exception was not caused by MPX.
 408                          */
 409                         goto exit_trap;
 410                 }
 411                 /*
 412                  * Success, we decoded the instruction and retrieved
 413                  * an 'info' containing the address being accessed
 414                  * which caused the exception.  This information
 415                  * allows and application to possibly handle the
 416                  * #BR exception itself.
 417                  */
 418                 do_trap(X86_TRAP_BR, SIGSEGV, "bounds", regs, error_code, info);
 419                 kfree(info);
 420                 break;
 421         case 0: /* No exception caused by Intel MPX operations. */
 422                 goto exit_trap;
 423         default:
 424                 die("bounds", regs, error_code);
 425         }
 426
 427         return;
 428
 429 exit_trap:
 430         /*
 431          * This path out is for all the cases where we could not
 432          * handle the exception in some way (like allocating a
 433          * table or telling userspace about it.  We will also end
 434          * up here if the kernel has MPX turned off at compile
 435          * time..
 436          */
 437         do_trap(X86_TRAP_BR, SIGSEGV, "bounds", regs, error_code, NULL);
 438 }
 439
 440 dotraplinkage void
 441 do_general_protection(struct pt_regs *regs, long error_code)
 442 {
 443         struct task_struct *tsk;
 444
 445         RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
 446         conditional_sti(regs);
 447
 448         if (v8086_mode(regs)) {
 449                 local_irq_enable();
 450                 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
 451                 return;
 452         }
 453
 454         tsk = current;
 455         if (!user_mode(regs)) {
 456                 if (fixup_exception(regs))
 457                         return;
 458
 459                 tsk->thread.error_code = error_code;
 460                 tsk->thread.trap_nr = X86_TRAP_GP;
 461                 if (notify_die(DIE_GPF, "general protection fault", regs, error_code,
 462                                X86_TRAP_GP, SIGSEGV) != NOTIFY_STOP)
 463                         die("general protection fault", regs, error_code);
 464                 return;
 465         }
 466
 467         tsk->thread.error_code = error_code;
 468         tsk->thread.trap_nr = X86_TRAP_GP;
 469
 470         if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
 471                         printk_ratelimit()) {
 472                 pr_info("%s[%d] general protection ip:%lx sp:%lx error:%lx",
 473                         tsk->comm, task_pid_nr(tsk),
 474                         regs->ip, regs->sp, error_code);
 475                 print_vma_addr(" in ", regs->ip);
 476                 pr_cont("\n");
 477         }
 478
 479         force_sig_info(SIGSEGV, SEND_SIG_PRIV, tsk);
 480 }
 481 NOKPROBE_SYMBOL(do_general_protection);
 482
 483 /* May run on IST stack. */
 484 dotraplinkage void notrace do_int3(struct pt_regs *regs, long error_code)
 485 {
 486 #ifdef CONFIG_DYNAMIC_FTRACE
 487         /*
 488          * ftrace must be first, everything else may cause a recursive crash.
 489          * See note by declaration of modifying_ftrace_code in ftrace.c
 490          */
 491         if (unlikely(atomic_read(&modifying_ftrace_code)) &&
 492             ftrace_int3_handler(regs))
 493                 return;
 494 #endif
 495         if (poke_int3_handler(regs))
 496                 return;
 497
 498         ist_enter(regs);
 499         RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
 500 #ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
 501         if (kgdb_ll_trap(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
 502                                 SIGTRAP) == NOTIFY_STOP)
 503                 goto exit;
 504 #endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
 505
 506 #ifdef CONFIG_KPROBES
 507         if (kprobe_int3_handler(regs))
 508                 goto exit;
 509 #endif
 510
 511         if (notify_die(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
 512                         SIGTRAP) == NOTIFY_STOP)
 513                 goto exit;
 514
 515         /*
 516          * Let others (NMI) know that the debug stack is in use
 517          * as we may switch to the interrupt stack.
 518          */
 519         debug_stack_usage_inc();
 520         preempt_conditional_sti(regs);
 521         do_trap(X86_TRAP_BP, SIGTRAP, "int3", regs, error_code, NULL);
 522         preempt_conditional_cli(regs);
 523         debug_stack_usage_dec();
 524 exit:
 525         ist_exit(regs);
 526 }
 527 NOKPROBE_SYMBOL(do_int3);
 528
 529 #ifdef CONFIG_X86_64
 530 /*
 531  * Help handler running on IST stack to switch off the IST stack if the
 532  * interrupted code was in user mode. The actual stack switch is done in
 533  * entry_64.S
 534  */
 535 asmlinkage __visible notrace struct pt_regs *sync_regs(struct pt_regs *eregs)
 536 {
 537         struct pt_regs *regs = task_pt_regs(current);
 538         *regs = *eregs;
 539         return regs;
 540 }
 541 NOKPROBE_SYMBOL(sync_regs);
 542
 543 struct bad_iret_stack {
 544         void *error_entry_ret;
 545         struct pt_regs regs;
 546 };
 547
 548 asmlinkage __visible notrace
 549 struct bad_iret_stack *fixup_bad_iret(struct bad_iret_stack *s)
 550 {
 551         /*
 552          * This is called from entry_64.S early in handling a fault
 553          * caused by a bad iret to user mode.  To handle the fault
 554          * correctly, we want move our stack frame to task_pt_regs
 555          * and we want to pretend that the exception came from the
 556          * iret target.
 557          */
 558         struct bad_iret_stack *new_stack =
 559                 container_of(task_pt_regs(current),
 560                              struct bad_iret_stack, regs);
 561
 562         /* Copy the IRET target to the new stack. */
 563         memmove(&new_stack->regs.ip, (void *)s->regs.sp, 5*8);
 564
 565         /* Copy the remainder of the stack from the current stack. */
 566         memmove(new_stack, s, offsetof(struct bad_iret_stack, regs.ip));
 567
 568         BUG_ON(!user_mode(&new_stack->regs));
 569         return new_stack;
 570 }
 571 NOKPROBE_SYMBOL(fixup_bad_iret);
 572 #endif
 573
 574 /*
 575  * Our handling of the processor debug registers is non-trivial.
 576  * We do not clear them on entry and exit from the kernel. Therefore
 577  * it is possible to get a watchpoint trap here from inside the kernel.
 578  * However, the code in ./ptrace.c has ensured that the user can
 579  * only set watchpoints on userspace addresses. Therefore the in-kernel
 580  * watchpoint trap can only occur in code which is reading/writing
 581  * from user space. Such code must not hold kernel locks (since it
 582  * can equally take a page fault), therefore it is safe to call
 583  * force_sig_info even though that claims and releases locks.
 584  *
 585  * Code in ./signal.c ensures that the debug control register
 586  * is restored before we deliver any signal, and therefore that
 587  * user code runs with the correct debug control register even though
 588  * we clear it here.
 589  *
 590  * Being careful here means that we don't have to be as careful in a
 591  * lot of more complicated places (task switching can be a bit lazy
 592  * about restoring all the debug state, and ptrace doesn't have to
 593  * find every occurrence of the TF bit that could be saved away even
 594  * by user code)
 595  *
 596  * May run on IST stack.
 597  */
 598 dotraplinkage void do_debug(struct pt_regs *regs, long error_code)
 599 {
 600         struct task_struct *tsk = current;
 601         int user_icebp = 0;
 602         unsigned long dr6;
 603         int si_code;
 604
 605         ist_enter(regs);
 606
 607         get_debugreg(dr6, 6);
 608
 609         /* Filter out all the reserved bits which are preset to 1 */
 610         dr6 &= ~DR6_RESERVED;
 611
 612         /*
 613          * If dr6 has no reason to give us about the origin of this trap,
 614          * then it's very likely the result of an icebp/int01 trap.
 615          * User wants a sigtrap for that.
 616          */
 617         if (!dr6 && user_mode(regs))
 618                 user_icebp = 1;
 619
 620         /* Catch kmemcheck conditions first of all! */
 621         if ((dr6 & DR_STEP) && kmemcheck_trap(regs))
 622                 goto exit;
 623
 624         /* DR6 may or may not be cleared by the CPU */
 625         set_debugreg(0, 6);
 626
 627         /*
 628          * The processor cleared BTF, so don't mark that we need it set.
 629          */
 630         clear_tsk_thread_flag(tsk, TIF_BLOCKSTEP);
 631
 632         /* Store the virtualized DR6 value */
 633         tsk->thread.debugreg6 = dr6;
 634
 635 #ifdef CONFIG_KPROBES
 636         if (kprobe_debug_handler(regs))
 637                 goto exit;
 638 #endif
 639
 640         if (notify_die(DIE_DEBUG, "debug", regs, (long)&dr6, error_code,
 641                                                         SIGTRAP) == NOTIFY_STOP)
 642                 goto exit;
 643
 644         /*
 645          * Let others (NMI) know that the debug stack is in use
 646          * as we may switch to the interrupt stack.
 647          */
 648         debug_stack_usage_inc();
 649
 650         /* It's safe to allow irq's after DR6 has been saved */
 651         preempt_conditional_sti(regs);
 652
 653         if (v8086_mode(regs)) {
 654                 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code,
 655                                         X86_TRAP_DB);
 656                 preempt_conditional_cli(regs);
 657                 debug_stack_usage_dec();
 658                 goto exit;
 659         }
 660
 661         /*
 662          * Single-stepping through system calls: ignore any exceptions in
 663          * kernel space, but re-enable TF when returning to user mode.
 664          *
 665          * We already checked v86 mode above, so we can check for kernel mode
 666          * by just checking the CPL of CS.
 667          */
 668         if ((dr6 & DR_STEP) && !user_mode(regs)) {
 669                 tsk->thread.debugreg6 &= ~DR_STEP;
 670                 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
 671                 regs->flags &= ~X86_EFLAGS_TF;
 672         }
 673         si_code = get_si_code(tsk->thread.debugreg6);
 674         if (tsk->thread.debugreg6 & (DR_STEP | DR_TRAP_BITS) || user_icebp)
 675                 send_sigtrap(tsk, regs, error_code, si_code);
 676         preempt_conditional_cli(regs);
 677         debug_stack_usage_dec();
 678
 679 exit:
 680         ist_exit(regs);
 681 }
 682 NOKPROBE_SYMBOL(do_debug);
 683
 684 /*
 685  * Note that we play around with the 'TS' bit in an attempt to get
 686  * the correct behaviour even in the presence of the asynchronous
 687  * IRQ13 behaviour
 688  */
 689 static void math_error(struct pt_regs *regs, int error_code, int trapnr)
 690 {
 691         struct task_struct *task = current;
 692         struct fpu *fpu = &task->thread.fpu;
 693         siginfo_t info;
 694         char *str = (trapnr == X86_TRAP_MF) ? "fpu exception" :
 695                                                 "simd exception";
 696
 697         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, SIGFPE) == NOTIFY_STOP)
 698                 return;
 699         conditional_sti(regs);
 700
 701         if (!user_mode(regs)) {
 702                 if (!fixup_exception(regs)) {
 703                         task->thread.error_code = error_code;
 704                         task->thread.trap_nr = trapnr;
 705                         die(str, regs, error_code);
 706                 }
 707                 return;
 708         }
 709
 710         /*
 711          * Save the info for the exception handler and clear the error.
 712          */
 713         fpu__save(fpu);
 714
 715         task->thread.trap_nr    = trapnr;
 716         task->thread.error_code = error_code;
 717         info.si_signo           = SIGFPE;
 718         info.si_errno           = 0;
 719         info.si_addr            = (void __user *)uprobe_get_trap_addr(regs);
 720
 721         info.si_code = fpu__exception_code(fpu, trapnr);
 722
 723         /* Retry when we get spurious exceptions: */
 724         if (!info.si_code)
 725                 return;
 726
 727         force_sig_info(SIGFPE, &info, task);
 728 }
 729
 730 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
 731 {
 732         RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
 733         math_error(regs, error_code, X86_TRAP_MF);
 734 }
 735
 736 dotraplinkage void
 737 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
 738 {
 739         RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
 740         math_error(regs, error_code, X86_TRAP_XF);
 741 }
 742
 743 dotraplinkage void
 744 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
 745 {
 746         conditional_sti(regs);
 747 }
 748
 749 dotraplinkage void
 750 do_device_not_available(struct pt_regs *regs, long error_code)
 751 {
 752         RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
 753         BUG_ON(use_eager_fpu());
 754
 755 #ifdef CONFIG_MATH_EMULATION
 756         if (read_cr0() & X86_CR0_EM) {
 757                 struct math_emu_info info = { };
 758
 759                 conditional_sti(regs);
 760
 761                 info.regs = regs;
 762                 math_emulate(&info);
 763                 return;
 764         }
 765 #endif
 766         fpu__restore(&current->thread.fpu); /* interrupts still off */
 767 #ifdef CONFIG_X86_32
 768         conditional_sti(regs);
 769 #endif
 770 }
 771 NOKPROBE_SYMBOL(do_device_not_available);
 772
 773 #ifdef CONFIG_X86_32
 774 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
 775 {
 776         siginfo_t info;
 777
 778         RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
 779         local_irq_enable();
 780
 781         info.si_signo = SIGILL;
 782         info.si_errno = 0;
 783         info.si_code = ILL_BADSTK;
 784         info.si_addr = NULL;
 785         if (notify_die(DIE_TRAP, "iret exception", regs, error_code,
 786                         X86_TRAP_IRET, SIGILL) != NOTIFY_STOP) {
 787                 do_trap(X86_TRAP_IRET, SIGILL, "iret exception", regs, error_code,
 788                         &info);
 789         }
 790 }
 791 #endif
 792
 793 /* Set of traps needed for early debugging. */
 794 void __init early_trap_init(void)
 795 {
 796         /*
 797          * Don't use IST to set DEBUG_STACK as it doesn't work until TSS
 798          * is ready in cpu_init() <-- trap_init(). Before trap_init(),
 799          * CPU runs at ring 0 so it is impossible to hit an invalid
 800          * stack.  Using the original stack works well enough at this
 801          * early stage. DEBUG_STACK will be equipped after cpu_init() in
 802          * trap_init().
 803          *
 804          * We don't need to set trace_idt_table like set_intr_gate(),
 805          * since we don't have trace_debug and it will be reset to
 806          * 'debug' in trap_init() by set_intr_gate_ist().
 807          */
 808         set_intr_gate_notrace(X86_TRAP_DB, debug);
 809         /* int3 can be called from all */
 810         set_system_intr_gate(X86_TRAP_BP, &int3);
 811 #ifdef CONFIG_X86_32
 812         set_intr_gate(X86_TRAP_PF, page_fault);
 813 #endif
 814         load_idt(&idt_descr);
 815 }
 816
 817 void __init early_trap_pf_init(void)
 818 {
 819 #ifdef CONFIG_X86_64
 820         set_intr_gate(X86_TRAP_PF, page_fault);
 821 #endif
 822 }
 823
 824 void __init trap_init(void)
 825 {
 826         int i;
 827
 828 #ifdef CONFIG_EISA
 829         void __iomem *p = early_ioremap(0x0FFFD9, 4);
 830
 831         if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
 832                 EISA_bus = 1;
 833         early_iounmap(p, 4);
 834 #endif
 835
 836         set_intr_gate(X86_TRAP_DE, divide_error);
 837         set_intr_gate_ist(X86_TRAP_NMI, &nmi, NMI_STACK);
 838         /* int4 can be called from all */
 839         set_system_intr_gate(X86_TRAP_OF, &overflow);
 840         set_intr_gate(X86_TRAP_BR, bounds);
 841         set_intr_gate(X86_TRAP_UD, invalid_op);
 842         set_intr_gate(X86_TRAP_NM, device_not_available);
 843 #ifdef CONFIG_X86_32
 844         set_task_gate(X86_TRAP_DF, GDT_ENTRY_DOUBLEFAULT_TSS);
 845 #else
 846         set_intr_gate_ist(X86_TRAP_DF, &double_fault, DOUBLEFAULT_STACK);
 847 #endif
 848         set_intr_gate(X86_TRAP_OLD_MF, coprocessor_segment_overrun);
 849         set_intr_gate(X86_TRAP_TS, invalid_TSS);
 850         set_intr_gate(X86_TRAP_NP, segment_not_present);
 851         set_intr_gate(X86_TRAP_SS, stack_segment);
 852         set_intr_gate(X86_TRAP_GP, general_protection);
 853         set_intr_gate(X86_TRAP_SPURIOUS, spurious_interrupt_bug);
 854         set_intr_gate(X86_TRAP_MF, coprocessor_error);
 855         set_intr_gate(X86_TRAP_AC, alignment_check);
 856 #ifdef CONFIG_X86_MCE
 857         set_intr_gate_ist(X86_TRAP_MC, &machine_check, MCE_STACK);
 858 #endif
 859         set_intr_gate(X86_TRAP_XF, simd_coprocessor_error);
 860
 861         /* Reserve all the builtin and the syscall vector: */
 862         for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
 863                 set_bit(i, used_vectors);
 864
 865 #ifdef CONFIG_IA32_EMULATION
 866         set_system_intr_gate(IA32_SYSCALL_VECTOR, entry_INT80_compat);
 867         set_bit(IA32_SYSCALL_VECTOR, used_vectors);
 868 #endif
 869
 870 #ifdef CONFIG_X86_32
 871         set_system_trap_gate(IA32_SYSCALL_VECTOR, entry_INT80_32);
 872         set_bit(IA32_SYSCALL_VECTOR, used_vectors);
 873 #endif
 874
 875         /*
 876          * Set the IDT descriptor to a fixed read-only location, so that the
 877          * "sidt" instruction will not leak the location of the kernel, and
 878          * to defend the IDT against arbitrary memory write vulnerabilities.
 879          * It will be reloaded in cpu_init() */
 880         __set_fixmap(FIX_RO_IDT, __pa_symbol(idt_table), PAGE_KERNEL_RO);
 881         idt_descr.address = fix_to_virt(FIX_RO_IDT);
 882
 883         /*
 884          * Should be a barrier for any external CPU state:
 885          */
 886         cpu_init();
 887
 888         /*
 889          * X86_TRAP_DB and X86_TRAP_BP have been set
 890          * in early_trap_init(). However, ITS works only after
 891          * cpu_init() loads TSS. See comments in early_trap_init().
 892          */
 893         set_intr_gate_ist(X86_TRAP_DB, &debug, DEBUG_STACK);
 894         /* int3 can be called from all */
 895         set_system_intr_gate_ist(X86_TRAP_BP, &int3, DEBUG_STACK);
 896
 897         x86_init.irqs.trap_init();
 898
 899 #ifdef CONFIG_X86_64
 900         memcpy(&debug_idt_table, &idt_table, IDT_ENTRIES * 16);
 901         set_nmi_gate(X86_TRAP_DB, &debug);
 902         set_nmi_gate(X86_TRAP_BP, &int3);
 903 #endif
 904 }