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
  35 #ifdef CONFIG_EISA
  36 #include <linux/ioport.h>
  37 #include <linux/eisa.h>
  38 #endif
  39
  40 #ifdef CONFIG_MCA
  41 #include <linux/mca.h>
  42 #endif
  43
  44 #if defined(CONFIG_EDAC)
  45 #include <linux/edac.h>
  46 #endif
  47
  48 #include <asm/kmemcheck.h>
  49 #include <asm/stacktrace.h>
  50 #include <asm/processor.h>
  51 #include <asm/debugreg.h>
  52 #include <asm/atomic.h>
  53 #include <asm/system.h>
  54 #include <asm/traps.h>
  55 #include <asm/desc.h>
  56 #include <asm/i387.h>
  57 #include <asm/mce.h>
  58
  59 #include <asm/mach_traps.h>
  60
  61 #ifdef CONFIG_X86_64
  62 #include <asm/x86_init.h>
  63 #include <asm/pgalloc.h>
  64 #include <asm/proto.h>
  65 #else
  66 #include <asm/processor-flags.h>
  67 #include <asm/setup.h>
  68
  69 asmlinkage int system_call(void);
  70
  71 /* Do we ignore FPU interrupts ? */
  72 char ignore_fpu_irq;
  73
  74 /*
  75  * The IDT has to be page-aligned to simplify the Pentium
  76  * F0 0F bug workaround.
  77  */
  78 gate_desc idt_table[NR_VECTORS] __page_aligned_data = { { { { 0, 0 } } }, };
  79 #endif
  80
  81 DECLARE_BITMAP(used_vectors, NR_VECTORS);
  82 EXPORT_SYMBOL_GPL(used_vectors);
  83
  84 static int ignore_nmis;
  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         inc_preempt_count();
  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         dec_preempt_count();
 110 }
 111
 112 static void __kprobes
 113 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
 114         long error_code, siginfo_t *info)
 115 {
 116         struct task_struct *tsk = current;
 117
 118 #ifdef CONFIG_X86_32
 119         if (regs->flags & X86_VM_MASK) {
 120                 /*
 121                  * traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
 122                  * On nmi (interrupt 2), do_trap should not be called.
 123                  */
 124                 if (trapnr < 6)
 125                         goto vm86_trap;
 126                 goto trap_signal;
 127         }
 128 #endif
 129
 130         if (!user_mode(regs))
 131                 goto kernel_trap;
 132
 133 #ifdef CONFIG_X86_32
 134 trap_signal:
 135 #endif
 136         /*
 137          * We want error_code and trap_no set for userspace faults and
 138          * kernelspace faults which result in die(), but not
 139          * kernelspace faults which are fixed up.  die() gives the
 140          * process no chance to handle the signal and notice the
 141          * kernel fault information, so that won't result in polluting
 142          * the information about previously queued, but not yet
 143          * delivered, faults.  See also do_general_protection below.
 144          */
 145         tsk->thread.error_code = error_code;
 146         tsk->thread.trap_no = trapnr;
 147
 148 #ifdef CONFIG_X86_64
 149         if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
 150             printk_ratelimit()) {
 151                 printk(KERN_INFO
 152                        "%s[%d] trap %s ip:%lx sp:%lx error:%lx",
 153                        tsk->comm, tsk->pid, str,
 154                        regs->ip, regs->sp, error_code);
 155                 print_vma_addr(" in ", regs->ip);
 156                 printk("\n");
 157         }
 158 #endif
 159
 160         if (info)
 161                 force_sig_info(signr, info, tsk);
 162         else
 163                 force_sig(signr, tsk);
 164         return;
 165
 166 kernel_trap:
 167         if (!fixup_exception(regs)) {
 168                 tsk->thread.error_code = error_code;
 169                 tsk->thread.trap_no = trapnr;
 170                 die(str, regs, error_code);
 171         }
 172         return;
 173
 174 #ifdef CONFIG_X86_32
 175 vm86_trap:
 176         if (handle_vm86_trap((struct kernel_vm86_regs *) regs,
 177                                                 error_code, trapnr))
 178                 goto trap_signal;
 179         return;
 180 #endif
 181 }
 182
 183 #define DO_ERROR(trapnr, signr, str, name)                              \
 184 dotraplinkage void do_##name(struct pt_regs *regs, long error_code)     \
 185 {                                                                       \
 186         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr)  \
 187                                                         == NOTIFY_STOP) \
 188                 return;                                                 \
 189         conditional_sti(regs);                                          \
 190         do_trap(trapnr, signr, str, regs, error_code, NULL);            \
 191 }
 192
 193 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr)         \
 194 dotraplinkage void do_##name(struct pt_regs *regs, long error_code)     \
 195 {                                                                       \
 196         siginfo_t info;                                                 \
 197         info.si_signo = signr;                                          \
 198         info.si_errno = 0;                                              \
 199         info.si_code = sicode;                                          \
 200         info.si_addr = (void __user *)siaddr;                           \
 201         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr)  \
 202                                                         == NOTIFY_STOP) \
 203                 return;                                                 \
 204         conditional_sti(regs);                                          \
 205         do_trap(trapnr, signr, str, regs, error_code, &info);           \
 206 }
 207
 208 DO_ERROR_INFO(0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip)
 209 DO_ERROR(4, SIGSEGV, "overflow", overflow)
 210 DO_ERROR(5, SIGSEGV, "bounds", bounds)
 211 DO_ERROR_INFO(6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip)
 212 DO_ERROR(9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
 213 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
 214 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
 215 #ifdef CONFIG_X86_32
 216 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
 217 #endif
 218 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
 219
 220 #ifdef CONFIG_X86_64
 221 /* Runs on IST stack */
 222 dotraplinkage void do_stack_segment(struct pt_regs *regs, long error_code)
 223 {
 224         if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
 225                         12, SIGBUS) == NOTIFY_STOP)
 226                 return;
 227         preempt_conditional_sti(regs);
 228         do_trap(12, SIGBUS, "stack segment", regs, error_code, NULL);
 229         preempt_conditional_cli(regs);
 230 }
 231
 232 dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
 233 {
 234         static const char str[] = "double fault";
 235         struct task_struct *tsk = current;
 236
 237         /* Return not checked because double check cannot be ignored */
 238         notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV);
 239
 240         tsk->thread.error_code = error_code;
 241         tsk->thread.trap_no = 8;
 242
 243         /*
 244          * This is always a kernel trap and never fixable (and thus must
 245          * never return).
 246          */
 247         for (;;)
 248                 die(str, regs, error_code);
 249 }
 250 #endif
 251
 252 dotraplinkage void __kprobes
 253 do_general_protection(struct pt_regs *regs, long error_code)
 254 {
 255         struct task_struct *tsk;
 256
 257         conditional_sti(regs);
 258
 259 #ifdef CONFIG_X86_32
 260         if (regs->flags & X86_VM_MASK)
 261                 goto gp_in_vm86;
 262 #endif
 263
 264         tsk = current;
 265         if (!user_mode(regs))
 266                 goto gp_in_kernel;
 267
 268         tsk->thread.error_code = error_code;
 269         tsk->thread.trap_no = 13;
 270
 271         if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
 272                         printk_ratelimit()) {
 273                 printk(KERN_INFO
 274                         "%s[%d] general protection ip:%lx sp:%lx error:%lx",
 275                         tsk->comm, task_pid_nr(tsk),
 276                         regs->ip, regs->sp, error_code);
 277                 print_vma_addr(" in ", regs->ip);
 278                 printk("\n");
 279         }
 280
 281         force_sig(SIGSEGV, tsk);
 282         return;
 283
 284 #ifdef CONFIG_X86_32
 285 gp_in_vm86:
 286         local_irq_enable();
 287         handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
 288         return;
 289 #endif
 290
 291 gp_in_kernel:
 292         if (fixup_exception(regs))
 293                 return;
 294
 295         tsk->thread.error_code = error_code;
 296         tsk->thread.trap_no = 13;
 297         if (notify_die(DIE_GPF, "general protection fault", regs,
 298                                 error_code, 13, SIGSEGV) == NOTIFY_STOP)
 299                 return;
 300         die("general protection fault", regs, error_code);
 301 }
 302
 303 static notrace __kprobes void
 304 mem_parity_error(unsigned char reason, struct pt_regs *regs)
 305 {
 306         printk(KERN_EMERG
 307                 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
 308                         reason, smp_processor_id());
 309
 310         printk(KERN_EMERG
 311                 "You have some hardware problem, likely on the PCI bus.\n");
 312
 313 #if defined(CONFIG_EDAC)
 314         if (edac_handler_set()) {
 315                 edac_atomic_assert_error();
 316                 return;
 317         }
 318 #endif
 319
 320         if (panic_on_unrecovered_nmi)
 321                 panic("NMI: Not continuing");
 322
 323         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 324
 325         /* Clear and disable the memory parity error line. */
 326         reason = (reason & 0xf) | 4;
 327         outb(reason, 0x61);
 328 }
 329
 330 static notrace __kprobes void
 331 io_check_error(unsigned char reason, struct pt_regs *regs)
 332 {
 333         unsigned long i;
 334
 335         printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
 336         show_registers(regs);
 337
 338         if (panic_on_io_nmi)
 339                 panic("NMI IOCK error: Not continuing");
 340
 341         /* Re-enable the IOCK line, wait for a few seconds */
 342         reason = (reason & 0xf) | 8;
 343         outb(reason, 0x61);
 344
 345         i = 2000;
 346         while (--i)
 347                 udelay(1000);
 348
 349         reason &= ~8;
 350         outb(reason, 0x61);
 351 }
 352
 353 static notrace __kprobes void
 354 unknown_nmi_error(unsigned char reason, struct pt_regs *regs)
 355 {
 356         if (notify_die(DIE_NMIUNKNOWN, "nmi", regs, reason, 2, SIGINT) ==
 357                         NOTIFY_STOP)
 358                 return;
 359 #ifdef CONFIG_MCA
 360         /*
 361          * Might actually be able to figure out what the guilty party
 362          * is:
 363          */
 364         if (MCA_bus) {
 365                 mca_handle_nmi();
 366                 return;
 367         }
 368 #endif
 369         printk(KERN_EMERG
 370                 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
 371                         reason, smp_processor_id());
 372
 373         printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
 374         if (panic_on_unrecovered_nmi)
 375                 panic("NMI: Not continuing");
 376
 377         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 378 }
 379
 380 static notrace __kprobes void default_do_nmi(struct pt_regs *regs)
 381 {
 382         unsigned char reason = 0;
 383         int cpu;
 384
 385         cpu = smp_processor_id();
 386
 387         /* Only the BSP gets external NMIs from the system. */
 388         if (!cpu)
 389                 reason = get_nmi_reason();
 390
 391         if (!(reason & 0xc0)) {
 392                 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
 393                                                                 == NOTIFY_STOP)
 394                         return;
 395 #ifdef CONFIG_X86_LOCAL_APIC
 396                 /*
 397                  * Ok, so this is none of the documented NMI sources,
 398                  * so it must be the NMI watchdog.
 399                  */
 400                 if (nmi_watchdog_tick(regs, reason))
 401                         return;
 402                 if (!do_nmi_callback(regs, cpu))
 403                         unknown_nmi_error(reason, regs);
 404 #else
 405                 unknown_nmi_error(reason, regs);
 406 #endif
 407
 408                 return;
 409         }
 410         if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
 411                 return;
 412
 413         /* AK: following checks seem to be broken on modern chipsets. FIXME */
 414         if (reason & 0x80)
 415                 mem_parity_error(reason, regs);
 416         if (reason & 0x40)
 417                 io_check_error(reason, regs);
 418 #ifdef CONFIG_X86_32
 419         /*
 420          * Reassert NMI in case it became active meanwhile
 421          * as it's edge-triggered:
 422          */
 423         reassert_nmi();
 424 #endif
 425 }
 426
 427 dotraplinkage notrace __kprobes void
 428 do_nmi(struct pt_regs *regs, long error_code)
 429 {
 430         nmi_enter();
 431
 432         inc_irq_stat(__nmi_count);
 433
 434         if (!ignore_nmis)
 435                 default_do_nmi(regs);
 436
 437         nmi_exit();
 438 }
 439
 440 void stop_nmi(void)
 441 {
 442         acpi_nmi_disable();
 443         ignore_nmis++;
 444 }
 445
 446 void restart_nmi(void)
 447 {
 448         ignore_nmis--;
 449         acpi_nmi_enable();
 450 }
 451
 452 /* May run on IST stack. */
 453 dotraplinkage void __kprobes do_int3(struct pt_regs *regs, long error_code)
 454 {
 455 #ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
 456         if (kgdb_ll_trap(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
 457                         == NOTIFY_STOP)
 458                 return;
 459 #endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
 460 #ifdef CONFIG_KPROBES
 461         if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
 462                         == NOTIFY_STOP)
 463                 return;
 464 #else
 465         if (notify_die(DIE_TRAP, "int3", regs, error_code, 3, SIGTRAP)
 466                         == NOTIFY_STOP)
 467                 return;
 468 #endif
 469
 470         preempt_conditional_sti(regs);
 471         do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
 472         preempt_conditional_cli(regs);
 473 }
 474
 475 #ifdef CONFIG_X86_64
 476 /*
 477  * Help handler running on IST stack to switch back to user stack
 478  * for scheduling or signal handling. The actual stack switch is done in
 479  * entry.S
 480  */
 481 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
 482 {
 483         struct pt_regs *regs = eregs;
 484         /* Did already sync */
 485         if (eregs == (struct pt_regs *)eregs->sp)
 486                 ;
 487         /* Exception from user space */
 488         else if (user_mode(eregs))
 489                 regs = task_pt_regs(current);
 490         /*
 491          * Exception from kernel and interrupts are enabled. Move to
 492          * kernel process stack.
 493          */
 494         else if (eregs->flags & X86_EFLAGS_IF)
 495                 regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
 496         if (eregs != regs)
 497                 *regs = *eregs;
 498         return regs;
 499 }
 500 #endif
 501
 502 /*
 503  * Our handling of the processor debug registers is non-trivial.
 504  * We do not clear them on entry and exit from the kernel. Therefore
 505  * it is possible to get a watchpoint trap here from inside the kernel.
 506  * However, the code in ./ptrace.c has ensured that the user can
 507  * only set watchpoints on userspace addresses. Therefore the in-kernel
 508  * watchpoint trap can only occur in code which is reading/writing
 509  * from user space. Such code must not hold kernel locks (since it
 510  * can equally take a page fault), therefore it is safe to call
 511  * force_sig_info even though that claims and releases locks.
 512  *
 513  * Code in ./signal.c ensures that the debug control register
 514  * is restored before we deliver any signal, and therefore that
 515  * user code runs with the correct debug control register even though
 516  * we clear it here.
 517  *
 518  * Being careful here means that we don't have to be as careful in a
 519  * lot of more complicated places (task switching can be a bit lazy
 520  * about restoring all the debug state, and ptrace doesn't have to
 521  * find every occurrence of the TF bit that could be saved away even
 522  * by user code)
 523  *
 524  * May run on IST stack.
 525  */
 526 dotraplinkage void __kprobes do_debug(struct pt_regs *regs, long error_code)
 527 {
 528         struct task_struct *tsk = current;
 529         int user_icebp = 0;
 530         unsigned long dr6;
 531         int si_code;
 532
 533         get_debugreg(dr6, 6);
 534
 535         /* Filter out all the reserved bits which are preset to 1 */
 536         dr6 &= ~DR6_RESERVED;
 537
 538         /*
 539          * If dr6 has no reason to give us about the origin of this trap,
 540          * then it's very likely the result of an icebp/int01 trap.
 541          * User wants a sigtrap for that.
 542          */
 543         if (!dr6 && user_mode(regs))
 544                 user_icebp = 1;
 545
 546         /* Catch kmemcheck conditions first of all! */
 547         if ((dr6 & DR_STEP) && kmemcheck_trap(regs))
 548                 return;
 549
 550         /* DR6 may or may not be cleared by the CPU */
 551         set_debugreg(0, 6);
 552
 553         /*
 554          * The processor cleared BTF, so don't mark that we need it set.
 555          */
 556         clear_tsk_thread_flag(tsk, TIF_BLOCKSTEP);
 557
 558         /* Store the virtualized DR6 value */
 559         tsk->thread.debugreg6 = dr6;
 560
 561         if (notify_die(DIE_DEBUG, "debug", regs, PTR_ERR(&dr6), error_code,
 562                                                         SIGTRAP) == NOTIFY_STOP)
 563                 return;
 564
 565         /* It's safe to allow irq's after DR6 has been saved */
 566         preempt_conditional_sti(regs);
 567
 568         if (regs->flags & X86_VM_MASK) {
 569                 handle_vm86_trap((struct kernel_vm86_regs *) regs,
 570                                 error_code, 1);
 571                 preempt_conditional_cli(regs);
 572                 return;
 573         }
 574
 575         /*
 576          * Single-stepping through system calls: ignore any exceptions in
 577          * kernel space, but re-enable TF when returning to user mode.
 578          *
 579          * We already checked v86 mode above, so we can check for kernel mode
 580          * by just checking the CPL of CS.
 581          */
 582         if ((dr6 & DR_STEP) && !user_mode(regs)) {
 583                 tsk->thread.debugreg6 &= ~DR_STEP;
 584                 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
 585                 regs->flags &= ~X86_EFLAGS_TF;
 586         }
 587         si_code = get_si_code(tsk->thread.debugreg6);
 588         if (tsk->thread.debugreg6 & (DR_STEP | DR_TRAP_BITS) || user_icebp)
 589                 send_sigtrap(tsk, regs, error_code, si_code);
 590         preempt_conditional_cli(regs);
 591
 592         return;
 593 }
 594
 595 /*
 596  * Note that we play around with the 'TS' bit in an attempt to get
 597  * the correct behaviour even in the presence of the asynchronous
 598  * IRQ13 behaviour
 599  */
 600 void math_error(struct pt_regs *regs, int error_code, int trapnr)
 601 {
 602         struct task_struct *task = current;
 603         siginfo_t info;
 604         unsigned short err;
 605         char *str = (trapnr == 16) ? "fpu exception" : "simd exception";
 606
 607         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, SIGFPE) == NOTIFY_STOP)
 608                 return;
 609         conditional_sti(regs);
 610
 611         if (!user_mode_vm(regs))
 612         {
 613                 if (!fixup_exception(regs)) {
 614                         task->thread.error_code = error_code;
 615                         task->thread.trap_no = trapnr;
 616                         die(str, regs, error_code);
 617                 }
 618                 return;
 619         }
 620
 621         /*
 622          * Save the info for the exception handler and clear the error.
 623          */
 624         save_init_fpu(task);
 625         task->thread.trap_no = trapnr;
 626         task->thread.error_code = error_code;
 627         info.si_signo = SIGFPE;
 628         info.si_errno = 0;
 629         info.si_addr = (void __user *)regs->ip;
 630         if (trapnr == 16) {
 631                 unsigned short cwd, swd;
 632                 /*
 633                  * (~cwd & swd) will mask out exceptions that are not set to unmasked
 634                  * status.  0x3f is the exception bits in these regs, 0x200 is the
 635                  * C1 reg you need in case of a stack fault, 0x040 is the stack
 636                  * fault bit.  We should only be taking one exception at a time,
 637                  * so if this combination doesn't produce any single exception,
 638                  * then we have a bad program that isn't synchronizing its FPU usage
 639                  * and it will suffer the consequences since we won't be able to
 640                  * fully reproduce the context of the exception
 641                  */
 642                 cwd = get_fpu_cwd(task);
 643                 swd = get_fpu_swd(task);
 644
 645                 err = swd & ~cwd;
 646         } else {
 647                 /*
 648                  * The SIMD FPU exceptions are handled a little differently, as there
 649                  * is only a single status/control register.  Thus, to determine which
 650                  * unmasked exception was caught we must mask the exception mask bits
 651                  * at 0x1f80, and then use these to mask the exception bits at 0x3f.
 652                  */
 653                 unsigned short mxcsr = get_fpu_mxcsr(task);
 654                 err = ~(mxcsr >> 7) & mxcsr;
 655         }
 656
 657         if (err & 0x001) {      /* Invalid op */
 658                 /*
 659                  * swd & 0x240 == 0x040: Stack Underflow
 660                  * swd & 0x240 == 0x240: Stack Overflow
 661                  * User must clear the SF bit (0x40) if set
 662                  */
 663                 info.si_code = FPE_FLTINV;
 664         } else if (err & 0x004) { /* Divide by Zero */
 665                 info.si_code = FPE_FLTDIV;
 666         } else if (err & 0x008) { /* Overflow */
 667                 info.si_code = FPE_FLTOVF;
 668         } else if (err & 0x012) { /* Denormal, Underflow */
 669                 info.si_code = FPE_FLTUND;
 670         } else if (err & 0x020) { /* Precision */
 671                 info.si_code = FPE_FLTRES;
 672         } else {
 673                 /*
 674                  * If we're using IRQ 13, or supposedly even some trap 16
 675                  * implementations, it's possible we get a spurious trap...
 676                  */
 677                 return;         /* Spurious trap, no error */
 678         }
 679         force_sig_info(SIGFPE, &info, task);
 680 }
 681
 682 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
 683 {
 684 #ifdef CONFIG_X86_32
 685         ignore_fpu_irq = 1;
 686 #endif
 687
 688         math_error(regs, error_code, 16);
 689 }
 690
 691 dotraplinkage void
 692 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
 693 {
 694         math_error(regs, error_code, 19);
 695 }
 696
 697 dotraplinkage void
 698 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
 699 {
 700         conditional_sti(regs);
 701 #if 0
 702         /* No need to warn about this any longer. */
 703         printk(KERN_INFO "Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
 704 #endif
 705 }
 706
 707 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
 708 {
 709 }
 710
 711 asmlinkage void __attribute__((weak)) smp_threshold_interrupt(void)
 712 {
 713 }
 714
 715 /*
 716  * __math_state_restore assumes that cr0.TS is already clear and the
 717  * fpu state is all ready for use.  Used during context switch.
 718  */
 719 void __math_state_restore(void)
 720 {
 721         struct thread_info *thread = current_thread_info();
 722         struct task_struct *tsk = thread->task;
 723
 724         /*
 725          * Paranoid restore. send a SIGSEGV if we fail to restore the state.
 726          */
 727         if (unlikely(restore_fpu_checking(tsk))) {
 728                 stts();
 729                 force_sig(SIGSEGV, tsk);
 730                 return;
 731         }
 732
 733         thread->status |= TS_USEDFPU;   /* So we fnsave on switch_to() */
 734         tsk->fpu_counter++;
 735 }
 736
 737 /*
 738  * 'math_state_restore()' saves the current math information in the
 739  * old math state array, and gets the new ones from the current task
 740  *
 741  * Careful.. There are problems with IBM-designed IRQ13 behaviour.
 742  * Don't touch unless you *really* know how it works.
 743  *
 744  * Must be called with kernel preemption disabled (in this case,
 745  * local interrupts are disabled at the call-site in entry.S).
 746  */
 747 asmlinkage void math_state_restore(void)
 748 {
 749         struct thread_info *thread = current_thread_info();
 750         struct task_struct *tsk = thread->task;
 751
 752         if (!tsk_used_math(tsk)) {
 753                 local_irq_enable();
 754                 /*
 755                  * does a slab alloc which can sleep
 756                  */
 757                 if (init_fpu(tsk)) {
 758                         /*
 759                          * ran out of memory!
 760                          */
 761                         do_group_exit(SIGKILL);
 762                         return;
 763                 }
 764                 local_irq_disable();
 765         }
 766
 767         clts();                         /* Allow maths ops (or we recurse) */
 768
 769         __math_state_restore();
 770 }
 771 EXPORT_SYMBOL_GPL(math_state_restore);
 772
 773 #ifndef CONFIG_MATH_EMULATION
 774 void math_emulate(struct math_emu_info *info)
 775 {
 776         printk(KERN_EMERG
 777                 "math-emulation not enabled and no coprocessor found.\n");
 778         printk(KERN_EMERG "killing %s.\n", current->comm);
 779         force_sig(SIGFPE, current);
 780         schedule();
 781 }
 782 #endif /* CONFIG_MATH_EMULATION */
 783
 784 dotraplinkage void __kprobes
 785 do_device_not_available(struct pt_regs *regs, long error_code)
 786 {
 787 #ifdef CONFIG_X86_32
 788         if (read_cr0() & X86_CR0_EM) {
 789                 struct math_emu_info info = { };
 790
 791                 conditional_sti(regs);
 792
 793                 info.regs = regs;
 794                 math_emulate(&info);
 795         } else {
 796                 math_state_restore(); /* interrupts still off */
 797                 conditional_sti(regs);
 798         }
 799 #else
 800         math_state_restore();
 801 #endif
 802 }
 803
 804 #ifdef CONFIG_X86_32
 805 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
 806 {
 807         siginfo_t info;
 808         local_irq_enable();
 809
 810         info.si_signo = SIGILL;
 811         info.si_errno = 0;
 812         info.si_code = ILL_BADSTK;
 813         info.si_addr = NULL;
 814         if (notify_die(DIE_TRAP, "iret exception",
 815                         regs, error_code, 32, SIGILL) == NOTIFY_STOP)
 816                 return;
 817         do_trap(32, SIGILL, "iret exception", regs, error_code, &info);
 818 }
 819 #endif
 820
 821 /* Set of traps needed for early debugging. */
 822 void __init early_trap_init(void)
 823 {
 824         set_intr_gate_ist(1, &debug, DEBUG_STACK);
 825         /* int3 can be called from all */
 826         set_system_intr_gate_ist(3, &int3, DEBUG_STACK);
 827         set_intr_gate(14, &page_fault);
 828         load_idt(&idt_descr);
 829 }
 830
 831 void __init trap_init(void)
 832 {
 833         int i;
 834
 835 #ifdef CONFIG_EISA
 836         void __iomem *p = early_ioremap(0x0FFFD9, 4);
 837
 838         if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
 839                 EISA_bus = 1;
 840         early_iounmap(p, 4);
 841 #endif
 842
 843         set_intr_gate(0, &divide_error);
 844         set_intr_gate_ist(2, &nmi, NMI_STACK);
 845         /* int4 can be called from all */
 846         set_system_intr_gate(4, &overflow);
 847         set_intr_gate(5, &bounds);
 848         set_intr_gate(6, &invalid_op);
 849         set_intr_gate(7, &device_not_available);
 850 #ifdef CONFIG_X86_32
 851         set_task_gate(8, GDT_ENTRY_DOUBLEFAULT_TSS);
 852 #else
 853         set_intr_gate_ist(8, &double_fault, DOUBLEFAULT_STACK);
 854 #endif
 855         set_intr_gate(9, &coprocessor_segment_overrun);
 856         set_intr_gate(10, &invalid_TSS);
 857         set_intr_gate(11, &segment_not_present);
 858         set_intr_gate_ist(12, &stack_segment, STACKFAULT_STACK);
 859         set_intr_gate(13, &general_protection);
 860         set_intr_gate(15, &spurious_interrupt_bug);
 861         set_intr_gate(16, &coprocessor_error);
 862         set_intr_gate(17, &alignment_check);
 863 #ifdef CONFIG_X86_MCE
 864         set_intr_gate_ist(18, &machine_check, MCE_STACK);
 865 #endif
 866         set_intr_gate(19, &simd_coprocessor_error);
 867
 868         /* Reserve all the builtin and the syscall vector: */
 869         for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
 870                 set_bit(i, used_vectors);
 871
 872 #ifdef CONFIG_IA32_EMULATION
 873         set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
 874         set_bit(IA32_SYSCALL_VECTOR, used_vectors);
 875 #endif
 876
 877 #ifdef CONFIG_X86_32
 878         if (cpu_has_fxsr) {
 879                 printk(KERN_INFO "Enabling fast FPU save and restore... ");
 880                 set_in_cr4(X86_CR4_OSFXSR);
 881                 printk("done.\n");
 882         }
 883         if (cpu_has_xmm) {
 884                 printk(KERN_INFO
 885                         "Enabling unmasked SIMD FPU exception support... ");
 886                 set_in_cr4(X86_CR4_OSXMMEXCPT);
 887                 printk("done.\n");
 888         }
 889
 890         set_system_trap_gate(SYSCALL_VECTOR, &system_call);
 891         set_bit(SYSCALL_VECTOR, used_vectors);
 892 #endif
 893
 894         /*
 895          * Should be a barrier for any external CPU state:
 896          */
 897         cpu_init();
 898
 899         x86_init.irqs.trap_init();
 900 }