arch/x86/kernel/traps.c

   1 /*
   2  *  Copyright (C) 1991, 1992  Linus Torvalds
   3  *  Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs
   4  *
   5  *  Pentium III FXSR, SSE support
   6  *      Gareth Hughes <gareth@valinux.com>, May 2000
   7  */
   8
   9 /*
  10  * Handle hardware traps and faults.
  11  */
  12 #include <linux/interrupt.h>
  13 #include <linux/kallsyms.h>
  14 #include <linux/spinlock.h>
  15 #include <linux/kprobes.h>
  16 #include <linux/uaccess.h>
  17 #include <linux/kdebug.h>
  18 #include <linux/kgdb.h>
  19 #include <linux/kernel.h>
  20 #include <linux/module.h>
  21 #include <linux/ptrace.h>
  22 #include <linux/string.h>
  23 #include <linux/delay.h>
  24 #include <linux/errno.h>
  25 #include <linux/kexec.h>
  26 #include <linux/sched.h>
  27 #include <linux/timer.h>
  28 #include <linux/init.h>
  29 #include <linux/bug.h>
  30 #include <linux/nmi.h>
  31 #include <linux/mm.h>
  32 #include <linux/smp.h>
  33 #include <linux/io.h>
  34 #include <trace/events/irq_vectors.h>
  35
  36 #ifdef CONFIG_EISA
  37 #include <linux/ioport.h>
  38 #include <linux/eisa.h>
  39 #endif
  40
  41 #ifdef CONFIG_MCA
  42 #include <linux/mca.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/system.h>
  55 #include <asm/traps.h>
  56 #include <asm/desc.h>
  57 #include <asm/i387.h>
  58 #include <asm/mce.h>
  59
  60 #include <asm/mach_traps.h>
  61
  62 #ifdef CONFIG_X86_64
  63 #include <asm/x86_init.h>
  64 #include <asm/pgalloc.h>
  65 #include <asm/proto.h>
  66 #else
  67 #include <asm/processor-flags.h>
  68 #include <asm/setup.h>
  69
  70 asmlinkage int system_call(void);
  71
  72 /* Do we ignore FPU interrupts ? */
  73 char ignore_fpu_irq;
  74
  75 /*
  76  * The IDT has to be page-aligned to simplify the Pentium
  77  * F0 0F bug workaround.
  78  */
  79 gate_desc idt_table[NR_VECTORS] __page_aligned_data = { { { { 0, 0 } } }, };
  80 #endif
  81
  82 DECLARE_BITMAP(used_vectors, NR_VECTORS);
  83 EXPORT_SYMBOL_GPL(used_vectors);
  84
  85 static int ignore_nmis;
  86
  87 int unknown_nmi_panic;
  88 /*
  89  * Prevent NMI reason port (0x61) being accessed simultaneously, can
  90  * only be used in NMI handler.
  91  */
  92 static DEFINE_RAW_SPINLOCK(nmi_reason_lock);
  93
  94 static inline void conditional_sti(struct pt_regs *regs)
  95 {
  96         if (regs->flags & X86_EFLAGS_IF)
  97                 local_irq_enable();
  98 }
  99
 100 static inline void preempt_conditional_sti(struct pt_regs *regs)
 101 {
 102         inc_preempt_count();
 103         if (regs->flags & X86_EFLAGS_IF)
 104                 local_irq_enable();
 105 }
 106
 107 static inline void conditional_cli(struct pt_regs *regs)
 108 {
 109         if (regs->flags & X86_EFLAGS_IF)
 110                 local_irq_disable();
 111 }
 112
 113 static inline void preempt_conditional_cli(struct pt_regs *regs)
 114 {
 115         if (regs->flags & X86_EFLAGS_IF)
 116                 local_irq_disable();
 117         dec_preempt_count();
 118 }
 119
 120 static void __kprobes
 121 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
 122         long error_code, siginfo_t *info)
 123 {
 124         struct task_struct *tsk = current;
 125
 126 #ifdef CONFIG_X86_32
 127         if (regs->flags & X86_VM_MASK) {
 128                 /*
 129                  * traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
 130                  * On nmi (interrupt 2), do_trap should not be called.
 131                  */
 132                 if (trapnr < 6)
 133                         goto vm86_trap;
 134                 goto trap_signal;
 135         }
 136 #endif
 137
 138         if (!user_mode(regs))
 139                 goto kernel_trap;
 140
 141 #ifdef CONFIG_X86_32
 142 trap_signal:
 143 #endif
 144         /*
 145          * We want error_code and trap_no set for userspace faults and
 146          * kernelspace faults which result in die(), but not
 147          * kernelspace faults which are fixed up.  die() gives the
 148          * process no chance to handle the signal and notice the
 149          * kernel fault information, so that won't result in polluting
 150          * the information about previously queued, but not yet
 151          * delivered, faults.  See also do_general_protection below.
 152          */
 153         tsk->thread.error_code = error_code;
 154         tsk->thread.trap_no = trapnr;
 155
 156 #ifdef CONFIG_X86_64
 157         if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
 158             printk_ratelimit()) {
 159                 printk(KERN_INFO
 160                        "%s[%d] trap %s ip:%lx sp:%lx error:%lx",
 161                        tsk->comm, tsk->pid, str,
 162                        regs->ip, regs->sp, error_code);
 163                 print_vma_addr(" in ", regs->ip);
 164                 printk("\n");
 165         }
 166 #endif
 167
 168         if (info)
 169                 force_sig_info(signr, info, tsk);
 170         else
 171                 force_sig(signr, tsk);
 172         return;
 173
 174 kernel_trap:
 175         if (!fixup_exception(regs)) {
 176                 tsk->thread.error_code = error_code;
 177                 tsk->thread.trap_no = trapnr;
 178                 die(str, regs, error_code);
 179         }
 180         return;
 181
 182 #ifdef CONFIG_X86_32
 183 vm86_trap:
 184         if (handle_vm86_trap((struct kernel_vm86_regs *) regs,
 185                                                 error_code, trapnr))
 186                 goto trap_signal;
 187         return;
 188 #endif
 189 }
 190
 191 #define DO_ERROR(trapnr, signr, str, name)                              \
 192 dotraplinkage void do_##name(struct pt_regs *regs, long error_code)     \
 193 {                                                                       \
 194         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr)  \
 195                                                         == NOTIFY_STOP) \
 196                 return;                                                 \
 197         conditional_sti(regs);                                          \
 198         do_trap(trapnr, signr, str, regs, error_code, NULL);            \
 199 }
 200
 201 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr)         \
 202 dotraplinkage void do_##name(struct pt_regs *regs, long error_code)     \
 203 {                                                                       \
 204         siginfo_t info;                                                 \
 205         info.si_signo = signr;                                          \
 206         info.si_errno = 0;                                              \
 207         info.si_code = sicode;                                          \
 208         info.si_addr = (void __user *)siaddr;                           \
 209         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr)  \
 210                                                         == NOTIFY_STOP) \
 211                 return;                                                 \
 212         conditional_sti(regs);                                          \
 213         do_trap(trapnr, signr, str, regs, error_code, &info);           \
 214 }
 215
 216 DO_ERROR_INFO(0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip)
 217 DO_ERROR(4, SIGSEGV, "overflow", overflow)
 218 DO_ERROR(5, SIGSEGV, "bounds", bounds)
 219 DO_ERROR_INFO(6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip)
 220 DO_ERROR(9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
 221 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
 222 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
 223 #ifdef CONFIG_X86_32
 224 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
 225 #endif
 226 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
 227
 228 #ifdef CONFIG_X86_64
 229 /* Runs on IST stack */
 230 dotraplinkage void do_stack_segment(struct pt_regs *regs, long error_code)
 231 {
 232         if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
 233                         12, SIGBUS) == NOTIFY_STOP)
 234                 return;
 235         preempt_conditional_sti(regs);
 236         do_trap(12, SIGBUS, "stack segment", regs, error_code, NULL);
 237         preempt_conditional_cli(regs);
 238 }
 239
 240 dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
 241 {
 242         static const char str[] = "double fault";
 243         struct task_struct *tsk = current;
 244
 245         /* Return not checked because double check cannot be ignored */
 246         notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV);
 247
 248         tsk->thread.error_code = error_code;
 249         tsk->thread.trap_no = 8;
 250
 251         /*
 252          * This is always a kernel trap and never fixable (and thus must
 253          * never return).
 254          */
 255         for (;;)
 256                 die(str, regs, error_code);
 257 }
 258 #endif
 259
 260 dotraplinkage void __kprobes
 261 do_general_protection(struct pt_regs *regs, long error_code)
 262 {
 263         struct task_struct *tsk;
 264
 265         conditional_sti(regs);
 266
 267 #ifdef CONFIG_X86_32
 268         if (regs->flags & X86_VM_MASK)
 269                 goto gp_in_vm86;
 270 #endif
 271
 272         tsk = current;
 273         if (!user_mode(regs))
 274                 goto gp_in_kernel;
 275
 276         tsk->thread.error_code = error_code;
 277         tsk->thread.trap_no = 13;
 278
 279         if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
 280                         printk_ratelimit()) {
 281                 printk(KERN_INFO
 282                         "%s[%d] general protection ip:%lx sp:%lx error:%lx",
 283                         tsk->comm, task_pid_nr(tsk),
 284                         regs->ip, regs->sp, error_code);
 285                 print_vma_addr(" in ", regs->ip);
 286                 printk("\n");
 287         }
 288
 289         force_sig(SIGSEGV, tsk);
 290         return;
 291
 292 #ifdef CONFIG_X86_32
 293 gp_in_vm86:
 294         local_irq_enable();
 295         handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
 296         return;
 297 #endif
 298
 299 gp_in_kernel:
 300         if (fixup_exception(regs))
 301                 return;
 302
 303         tsk->thread.error_code = error_code;
 304         tsk->thread.trap_no = 13;
 305         if (notify_die(DIE_GPF, "general protection fault", regs,
 306                                 error_code, 13, SIGSEGV) == NOTIFY_STOP)
 307                 return;
 308         die("general protection fault", regs, error_code);
 309 }
 310
 311 static int __init setup_unknown_nmi_panic(char *str)
 312 {
 313         unknown_nmi_panic = 1;
 314         return 1;
 315 }
 316 __setup("unknown_nmi_panic", setup_unknown_nmi_panic);
 317
 318 static notrace __kprobes void
 319 pci_serr_error(unsigned char reason, struct pt_regs *regs)
 320 {
 321         pr_emerg("NMI: PCI system error (SERR) for reason %02x on CPU %d.\n",
 322                  reason, smp_processor_id());
 323
 324         /*
 325          * On some machines, PCI SERR line is used to report memory
 326          * errors. EDAC makes use of it.
 327          */
 328 #if defined(CONFIG_EDAC)
 329         if (edac_handler_set()) {
 330                 edac_atomic_assert_error();
 331                 return;
 332         }
 333 #endif
 334
 335         if (panic_on_unrecovered_nmi)
 336                 panic("NMI: Not continuing");
 337
 338         pr_emerg("Dazed and confused, but trying to continue\n");
 339
 340         /* Clear and disable the PCI SERR error line. */
 341         reason = (reason & NMI_REASON_CLEAR_MASK) | NMI_REASON_CLEAR_SERR;
 342         outb(reason, NMI_REASON_PORT);
 343 }
 344
 345 static notrace __kprobes void
 346 io_check_error(unsigned char reason, struct pt_regs *regs)
 347 {
 348         unsigned long i;
 349
 350         pr_emerg(
 351         "NMI: IOCK error (debug interrupt?) for reason %02x on CPU %d.\n",
 352                  reason, smp_processor_id());
 353         show_registers(regs);
 354
 355         if (panic_on_io_nmi)
 356                 panic("NMI IOCK error: Not continuing");
 357
 358         /* Re-enable the IOCK line, wait for a few seconds */
 359         reason = (reason & NMI_REASON_CLEAR_MASK) | NMI_REASON_CLEAR_IOCHK;
 360         outb(reason, NMI_REASON_PORT);
 361
 362         i = 20000;
 363         while (--i) {
 364                 touch_nmi_watchdog();
 365                 udelay(100);
 366         }
 367
 368         reason &= ~NMI_REASON_CLEAR_IOCHK;
 369         outb(reason, NMI_REASON_PORT);
 370 }
 371
 372 static notrace __kprobes void
 373 unknown_nmi_error(unsigned char reason, struct pt_regs *regs)
 374 {
 375         if (notify_die(DIE_NMIUNKNOWN, "nmi", regs, reason, 2, SIGINT) ==
 376                         NOTIFY_STOP)
 377                 return;
 378 #ifdef CONFIG_MCA
 379         /*
 380          * Might actually be able to figure out what the guilty party
 381          * is:
 382          */
 383         if (MCA_bus) {
 384                 mca_handle_nmi();
 385                 return;
 386         }
 387 #endif
 388         pr_emerg("Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
 389                  reason, smp_processor_id());
 390
 391         pr_emerg("Do you have a strange power saving mode enabled?\n");
 392         if (unknown_nmi_panic || panic_on_unrecovered_nmi)
 393                 panic("NMI: Not continuing");
 394
 395         pr_emerg("Dazed and confused, but trying to continue\n");
 396 }
 397
 398 static notrace __kprobes void default_do_nmi(struct pt_regs *regs)
 399 {
 400         unsigned char reason = 0;
 401
 402         /*
 403          * CPU-specific NMI must be processed before non-CPU-specific
 404          * NMI, otherwise we may lose it, because the CPU-specific
 405          * NMI can not be detected/processed on other CPUs.
 406          */
 407         if (notify_die(DIE_NMI, "nmi", regs, 0, 2, SIGINT) == NOTIFY_STOP)
 408                 return;
 409
 410         /* Non-CPU-specific NMI: NMI sources can be processed on any CPU */
 411         raw_spin_lock(&nmi_reason_lock);
 412         reason = get_nmi_reason();
 413
 414         if (reason & NMI_REASON_MASK) {
 415                 if (reason & NMI_REASON_SERR)
 416                         pci_serr_error(reason, regs);
 417                 else if (reason & NMI_REASON_IOCHK)
 418                         io_check_error(reason, regs);
 419 #ifdef CONFIG_X86_32
 420                 /*
 421                  * Reassert NMI in case it became active
 422                  * meanwhile as it's edge-triggered:
 423                  */
 424                 reassert_nmi();
 425 #endif
 426                 raw_spin_unlock(&nmi_reason_lock);
 427                 return;
 428         }
 429         raw_spin_unlock(&nmi_reason_lock);
 430
 431         unknown_nmi_error(reason, regs);
 432 }
 433
 434 dotraplinkage notrace __kprobes void
 435 do_nmi(struct pt_regs *regs, long error_code)
 436 {
 437         nmi_enter();
 438         trace_irq_vector_entry(NMI_VECTOR);
 439
 440         inc_irq_stat(__nmi_count);
 441
 442         if (!ignore_nmis)
 443                 default_do_nmi(regs);
 444
 445         trace_irq_vector_exit(NMI_VECTOR);
 446         nmi_exit();
 447 }
 448
 449 void stop_nmi(void)
 450 {
 451         ignore_nmis++;
 452 }
 453
 454 void restart_nmi(void)
 455 {
 456         ignore_nmis--;
 457 }
 458
 459 /* May run on IST stack. */
 460 dotraplinkage void __kprobes do_int3(struct pt_regs *regs, long error_code)
 461 {
 462 #ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
 463         if (kgdb_ll_trap(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
 464                         == NOTIFY_STOP)
 465                 return;
 466 #endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
 467 #ifdef CONFIG_KPROBES
 468         if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
 469                         == NOTIFY_STOP)
 470                 return;
 471 #else
 472         if (notify_die(DIE_TRAP, "int3", regs, error_code, 3, SIGTRAP)
 473                         == NOTIFY_STOP)
 474                 return;
 475 #endif
 476
 477         preempt_conditional_sti(regs);
 478         do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
 479         preempt_conditional_cli(regs);
 480 }
 481
 482 #ifdef CONFIG_X86_64
 483 /*
 484  * Help handler running on IST stack to switch back to user stack
 485  * for scheduling or signal handling. The actual stack switch is done in
 486  * entry.S
 487  */
 488 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
 489 {
 490         struct pt_regs *regs = eregs;
 491         /* Did already sync */
 492         if (eregs == (struct pt_regs *)eregs->sp)
 493                 ;
 494         /* Exception from user space */
 495         else if (user_mode(eregs))
 496                 regs = task_pt_regs(current);
 497         /*
 498          * Exception from kernel and interrupts are enabled. Move to
 499          * kernel process stack.
 500          */
 501         else if (eregs->flags & X86_EFLAGS_IF)
 502                 regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
 503         if (eregs != regs)
 504                 *regs = *eregs;
 505         return regs;
 506 }
 507 #endif
 508
 509 /*
 510  * Our handling of the processor debug registers is non-trivial.
 511  * We do not clear them on entry and exit from the kernel. Therefore
 512  * it is possible to get a watchpoint trap here from inside the kernel.
 513  * However, the code in ./ptrace.c has ensured that the user can
 514  * only set watchpoints on userspace addresses. Therefore the in-kernel
 515  * watchpoint trap can only occur in code which is reading/writing
 516  * from user space. Such code must not hold kernel locks (since it
 517  * can equally take a page fault), therefore it is safe to call
 518  * force_sig_info even though that claims and releases locks.
 519  *
 520  * Code in ./signal.c ensures that the debug control register
 521  * is restored before we deliver any signal, and therefore that
 522  * user code runs with the correct debug control register even though
 523  * we clear it here.
 524  *
 525  * Being careful here means that we don't have to be as careful in a
 526  * lot of more complicated places (task switching can be a bit lazy
 527  * about restoring all the debug state, and ptrace doesn't have to
 528  * find every occurrence of the TF bit that could be saved away even
 529  * by user code)
 530  *
 531  * May run on IST stack.
 532  */
 533 dotraplinkage void __kprobes do_debug(struct pt_regs *regs, long error_code)
 534 {
 535         struct task_struct *tsk = current;
 536         int user_icebp = 0;
 537         unsigned long dr6;
 538         int si_code;
 539
 540         get_debugreg(dr6, 6);
 541
 542         /* Filter out all the reserved bits which are preset to 1 */
 543         dr6 &= ~DR6_RESERVED;
 544
 545         /*
 546          * If dr6 has no reason to give us about the origin of this trap,
 547          * then it's very likely the result of an icebp/int01 trap.
 548          * User wants a sigtrap for that.
 549          */
 550         if (!dr6 && user_mode(regs))
 551                 user_icebp = 1;
 552
 553         /* Catch kmemcheck conditions first of all! */
 554         if ((dr6 & DR_STEP) && kmemcheck_trap(regs))
 555                 return;
 556
 557         /* DR6 may or may not be cleared by the CPU */
 558         set_debugreg(0, 6);
 559
 560         /*
 561          * The processor cleared BTF, so don't mark that we need it set.
 562          */
 563         clear_tsk_thread_flag(tsk, TIF_BLOCKSTEP);
 564
 565         /* Store the virtualized DR6 value */
 566         tsk->thread.debugreg6 = dr6;
 567
 568         if (notify_die(DIE_DEBUG, "debug", regs, PTR_ERR(&dr6), error_code,
 569                                                         SIGTRAP) == NOTIFY_STOP)
 570                 return;
 571
 572         /* It's safe to allow irq's after DR6 has been saved */
 573         preempt_conditional_sti(regs);
 574
 575         if (regs->flags & X86_VM_MASK) {
 576                 handle_vm86_trap((struct kernel_vm86_regs *) regs,
 577                                 error_code, 1);
 578                 preempt_conditional_cli(regs);
 579                 return;
 580         }
 581
 582         /*
 583          * Single-stepping through system calls: ignore any exceptions in
 584          * kernel space, but re-enable TF when returning to user mode.
 585          *
 586          * We already checked v86 mode above, so we can check for kernel mode
 587          * by just checking the CPL of CS.
 588          */
 589         if ((dr6 & DR_STEP) && !user_mode(regs)) {
 590                 tsk->thread.debugreg6 &= ~DR_STEP;
 591                 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
 592                 regs->flags &= ~X86_EFLAGS_TF;
 593         }
 594         si_code = get_si_code(tsk->thread.debugreg6);
 595         if (tsk->thread.debugreg6 & (DR_STEP | DR_TRAP_BITS) || user_icebp)
 596                 send_sigtrap(tsk, regs, error_code, si_code);
 597         preempt_conditional_cli(regs);
 598
 599         return;
 600 }
 601
 602 /*
 603  * Note that we play around with the 'TS' bit in an attempt to get
 604  * the correct behaviour even in the presence of the asynchronous
 605  * IRQ13 behaviour
 606  */
 607 void math_error(struct pt_regs *regs, int error_code, int trapnr)
 608 {
 609         struct task_struct *task = current;
 610         siginfo_t info;
 611         unsigned short err;
 612         char *str = (trapnr == 16) ? "fpu exception" : "simd exception";
 613
 614         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, SIGFPE) == NOTIFY_STOP)
 615                 return;
 616         conditional_sti(regs);
 617
 618         if (!user_mode_vm(regs))
 619         {
 620                 if (!fixup_exception(regs)) {
 621                         task->thread.error_code = error_code;
 622                         task->thread.trap_no = trapnr;
 623                         die(str, regs, error_code);
 624                 }
 625                 return;
 626         }
 627
 628         /*
 629          * Save the info for the exception handler and clear the error.
 630          */
 631         save_init_fpu(task);
 632         task->thread.trap_no = trapnr;
 633         task->thread.error_code = error_code;
 634         info.si_signo = SIGFPE;
 635         info.si_errno = 0;
 636         info.si_addr = (void __user *)regs->ip;
 637         if (trapnr == 16) {
 638                 unsigned short cwd, swd;
 639                 /*
 640                  * (~cwd & swd) will mask out exceptions that are not set to unmasked
 641                  * status.  0x3f is the exception bits in these regs, 0x200 is the
 642                  * C1 reg you need in case of a stack fault, 0x040 is the stack
 643                  * fault bit.  We should only be taking one exception at a time,
 644                  * so if this combination doesn't produce any single exception,
 645                  * then we have a bad program that isn't synchronizing its FPU usage
 646                  * and it will suffer the consequences since we won't be able to
 647                  * fully reproduce the context of the exception
 648                  */
 649                 cwd = get_fpu_cwd(task);
 650                 swd = get_fpu_swd(task);
 651
 652                 err = swd & ~cwd;
 653         } else {
 654                 /*
 655                  * The SIMD FPU exceptions are handled a little differently, as there
 656                  * is only a single status/control register.  Thus, to determine which
 657                  * unmasked exception was caught we must mask the exception mask bits
 658                  * at 0x1f80, and then use these to mask the exception bits at 0x3f.
 659                  */
 660                 unsigned short mxcsr = get_fpu_mxcsr(task);
 661                 err = ~(mxcsr >> 7) & mxcsr;
 662         }
 663
 664         if (err & 0x001) {      /* Invalid op */
 665                 /*
 666                  * swd & 0x240 == 0x040: Stack Underflow
 667                  * swd & 0x240 == 0x240: Stack Overflow
 668                  * User must clear the SF bit (0x40) if set
 669                  */
 670                 info.si_code = FPE_FLTINV;
 671         } else if (err & 0x004) { /* Divide by Zero */
 672                 info.si_code = FPE_FLTDIV;
 673         } else if (err & 0x008) { /* Overflow */
 674                 info.si_code = FPE_FLTOVF;
 675         } else if (err & 0x012) { /* Denormal, Underflow */
 676                 info.si_code = FPE_FLTUND;
 677         } else if (err & 0x020) { /* Precision */
 678                 info.si_code = FPE_FLTRES;
 679         } else {
 680                 /*
 681                  * If we're using IRQ 13, or supposedly even some trap 16
 682                  * implementations, it's possible we get a spurious trap...
 683                  */
 684                 return;         /* Spurious trap, no error */
 685         }
 686         force_sig_info(SIGFPE, &info, task);
 687 }
 688
 689 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
 690 {
 691 #ifdef CONFIG_X86_32
 692         ignore_fpu_irq = 1;
 693 #endif
 694
 695         math_error(regs, error_code, 16);
 696 }
 697
 698 dotraplinkage void
 699 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
 700 {
 701         math_error(regs, error_code, 19);
 702 }
 703
 704 dotraplinkage void
 705 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
 706 {
 707         conditional_sti(regs);
 708 #if 0
 709         /* No need to warn about this any longer. */
 710         printk(KERN_INFO "Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
 711 #endif
 712 }
 713
 714 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
 715 {
 716 }
 717
 718 asmlinkage void __attribute__((weak)) smp_threshold_interrupt(void)
 719 {
 720 }
 721
 722 /*
 723  * __math_state_restore assumes that cr0.TS is already clear and the
 724  * fpu state is all ready for use.  Used during context switch.
 725  */
 726 void __math_state_restore(void)
 727 {
 728         struct thread_info *thread = current_thread_info();
 729         struct task_struct *tsk = thread->task;
 730
 731         /*
 732          * Paranoid restore. send a SIGSEGV if we fail to restore the state.
 733          */
 734         if (unlikely(restore_fpu_checking(tsk))) {
 735                 stts();
 736                 force_sig(SIGSEGV, tsk);
 737                 return;
 738         }
 739
 740         thread->status |= TS_USEDFPU;   /* So we fnsave on switch_to() */
 741         tsk->fpu_counter++;
 742 }
 743
 744 /*
 745  * 'math_state_restore()' saves the current math information in the
 746  * old math state array, and gets the new ones from the current task
 747  *
 748  * Careful.. There are problems with IBM-designed IRQ13 behaviour.
 749  * Don't touch unless you *really* know how it works.
 750  *
 751  * Must be called with kernel preemption disabled (in this case,
 752  * local interrupts are disabled at the call-site in entry.S).
 753  */
 754 asmlinkage void math_state_restore(void)
 755 {
 756         struct thread_info *thread = current_thread_info();
 757         struct task_struct *tsk = thread->task;
 758
 759         if (!tsk_used_math(tsk)) {
 760                 local_irq_enable();
 761                 /*
 762                  * does a slab alloc which can sleep
 763                  */
 764                 if (init_fpu(tsk)) {
 765                         /*
 766                          * ran out of memory!
 767                          */
 768                         do_group_exit(SIGKILL);
 769                         return;
 770                 }
 771                 local_irq_disable();
 772         }
 773
 774         clts();                         /* Allow maths ops (or we recurse) */
 775
 776         __math_state_restore();
 777 }
 778 EXPORT_SYMBOL_GPL(math_state_restore);
 779
 780 dotraplinkage void __kprobes
 781 do_device_not_available(struct pt_regs *regs, long error_code)
 782 {
 783 #ifdef CONFIG_MATH_EMULATION
 784         if (read_cr0() & X86_CR0_EM) {
 785                 struct math_emu_info info = { };
 786
 787                 conditional_sti(regs);
 788
 789                 info.regs = regs;
 790                 math_emulate(&info);
 791                 return;
 792         }
 793 #endif
 794         math_state_restore(); /* interrupts still off */
 795 #ifdef CONFIG_X86_32
 796         conditional_sti(regs);
 797 #endif
 798 }
 799
 800 #ifdef CONFIG_X86_32
 801 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
 802 {
 803         siginfo_t info;
 804         local_irq_enable();
 805
 806         info.si_signo = SIGILL;
 807         info.si_errno = 0;
 808         info.si_code = ILL_BADSTK;
 809         info.si_addr = NULL;
 810         if (notify_die(DIE_TRAP, "iret exception",
 811                         regs, error_code, 32, SIGILL) == NOTIFY_STOP)
 812                 return;
 813         do_trap(32, SIGILL, "iret exception", regs, error_code, &info);
 814 }
 815 #endif
 816
 817 /* Set of traps needed for early debugging. */
 818 void __init early_trap_init(void)
 819 {
 820         set_intr_gate_ist(1, &debug, DEBUG_STACK);
 821         /* int3 can be called from all */
 822         set_system_intr_gate_ist(3, &int3, DEBUG_STACK);
 823         set_intr_gate(14, &page_fault);
 824         load_idt(&idt_descr);
 825 }
 826
 827 void __init trap_init(void)
 828 {
 829         int i;
 830
 831 #ifdef CONFIG_EISA
 832         void __iomem *p = early_ioremap(0x0FFFD9, 4);
 833
 834         if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
 835                 EISA_bus = 1;
 836         early_iounmap(p, 4);
 837 #endif
 838
 839         set_intr_gate(0, &divide_error);
 840         set_intr_gate_ist(2, &nmi, NMI_STACK);
 841         /* int4 can be called from all */
 842         set_system_intr_gate(4, &overflow);
 843         set_intr_gate(5, &bounds);
 844         set_intr_gate(6, &invalid_op);
 845         set_intr_gate(7, &device_not_available);
 846 #ifdef CONFIG_X86_32
 847         set_task_gate(8, GDT_ENTRY_DOUBLEFAULT_TSS);
 848 #else
 849         set_intr_gate_ist(8, &double_fault, DOUBLEFAULT_STACK);
 850 #endif
 851         set_intr_gate(9, &coprocessor_segment_overrun);
 852         set_intr_gate(10, &invalid_TSS);
 853         set_intr_gate(11, &segment_not_present);
 854         set_intr_gate_ist(12, &stack_segment, STACKFAULT_STACK);
 855         set_intr_gate(13, &general_protection);
 856         set_intr_gate(15, &spurious_interrupt_bug);
 857         set_intr_gate(16, &coprocessor_error);
 858         set_intr_gate(17, &alignment_check);
 859 #ifdef CONFIG_X86_MCE
 860         set_intr_gate_ist(18, &machine_check, MCE_STACK);
 861 #endif
 862         set_intr_gate(19, &simd_coprocessor_error);
 863
 864         /* Reserve all the builtin and the syscall vector: */
 865         for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
 866                 set_bit(i, used_vectors);
 867
 868 #ifdef CONFIG_IA32_EMULATION
 869         set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
 870         set_bit(IA32_SYSCALL_VECTOR, used_vectors);
 871 #endif
 872
 873 #ifdef CONFIG_X86_32
 874         set_system_trap_gate(SYSCALL_VECTOR, &system_call);
 875         set_bit(SYSCALL_VECTOR, used_vectors);
 876 #endif
 877
 878         /*
 879          * Should be a barrier for any external CPU state:
 880          */
 881         cpu_init();
 882
 883         x86_init.irqs.trap_init();
 884 }