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/utsname.h>
  18 #include <linux/kdebug.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/stacktrace.h>
  49 #include <asm/processor.h>
  50 #include <asm/debugreg.h>
  51 #include <asm/atomic.h>
  52 #include <asm/system.h>
  53 #include <asm/traps.h>
  54 #include <asm/desc.h>
  55 #include <asm/i387.h>
  56
  57 #include <asm/mach_traps.h>
  58
  59 #ifdef CONFIG_X86_64
  60 #include <asm/pgalloc.h>
  61 #include <asm/proto.h>
  62 #else
  63 #include <asm/processor-flags.h>
  64 #include <asm/setup.h>
  65 #include <asm/traps.h>
  66
  67 #include "cpu/mcheck/mce.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.. We have a special link segment
  77  * for this.
  78  */
  79 gate_desc idt_table[256]
  80         __attribute__((__section__(".data.idt"))) = { { { { 0, 0 } } }, };
  81 #endif
  82
  83 DECLARE_BITMAP(used_vectors, NR_VECTORS);
  84 EXPORT_SYMBOL_GPL(used_vectors);
  85
  86 static int ignore_nmis;
  87
  88 static inline void conditional_sti(struct pt_regs *regs)
  89 {
  90         if (regs->flags & X86_EFLAGS_IF)
  91                 local_irq_enable();
  92 }
  93
  94 static inline void preempt_conditional_sti(struct pt_regs *regs)
  95 {
  96         inc_preempt_count();
  97         if (regs->flags & X86_EFLAGS_IF)
  98                 local_irq_enable();
  99 }
 100
 101 static inline void conditional_cli(struct pt_regs *regs)
 102 {
 103         if (regs->flags & X86_EFLAGS_IF)
 104                 local_irq_disable();
 105 }
 106
 107 static inline void preempt_conditional_cli(struct pt_regs *regs)
 108 {
 109         if (regs->flags & X86_EFLAGS_IF)
 110                 local_irq_disable();
 111         dec_preempt_count();
 112 }
 113
 114 #ifdef CONFIG_X86_32
 115 static inline void
 116 die_if_kernel(const char *str, struct pt_regs *regs, long err)
 117 {
 118         if (!user_mode_vm(regs))
 119                 die(str, regs, err);
 120 }
 121
 122 /*
 123  * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
 124  * invalid offset set (the LAZY one) and the faulting thread has
 125  * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS,
 126  * we set the offset field correctly and return 1.
 127  */
 128 static int lazy_iobitmap_copy(void)
 129 {
 130         struct thread_struct *thread;
 131         struct tss_struct *tss;
 132         int cpu;
 133
 134         cpu = get_cpu();
 135         tss = &per_cpu(init_tss, cpu);
 136         thread = &current->thread;
 137
 138         if (tss->x86_tss.io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
 139             thread->io_bitmap_ptr) {
 140                 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
 141                        thread->io_bitmap_max);
 142                 /*
 143                  * If the previously set map was extending to higher ports
 144                  * than the current one, pad extra space with 0xff (no access).
 145                  */
 146                 if (thread->io_bitmap_max < tss->io_bitmap_max) {
 147                         memset((char *) tss->io_bitmap +
 148                                 thread->io_bitmap_max, 0xff,
 149                                 tss->io_bitmap_max - thread->io_bitmap_max);
 150                 }
 151                 tss->io_bitmap_max = thread->io_bitmap_max;
 152                 tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
 153                 tss->io_bitmap_owner = thread;
 154                 put_cpu();
 155
 156                 return 1;
 157         }
 158         put_cpu();
 159
 160         return 0;
 161 }
 162 #endif
 163
 164 static void __kprobes
 165 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
 166         long error_code, siginfo_t *info)
 167 {
 168         struct task_struct *tsk = current;
 169
 170 #ifdef CONFIG_X86_32
 171         if (regs->flags & X86_VM_MASK) {
 172                 /*
 173                  * traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
 174                  * On nmi (interrupt 2), do_trap should not be called.
 175                  */
 176                 if (trapnr < 6)
 177                         goto vm86_trap;
 178                 goto trap_signal;
 179         }
 180 #endif
 181
 182         if (!user_mode(regs))
 183                 goto kernel_trap;
 184
 185 #ifdef CONFIG_X86_32
 186 trap_signal:
 187 #endif
 188         /*
 189          * We want error_code and trap_no set for userspace faults and
 190          * kernelspace faults which result in die(), but not
 191          * kernelspace faults which are fixed up.  die() gives the
 192          * process no chance to handle the signal and notice the
 193          * kernel fault information, so that won't result in polluting
 194          * the information about previously queued, but not yet
 195          * delivered, faults.  See also do_general_protection below.
 196          */
 197         tsk->thread.error_code = error_code;
 198         tsk->thread.trap_no = trapnr;
 199
 200 #ifdef CONFIG_X86_64
 201         if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
 202             printk_ratelimit()) {
 203                 printk(KERN_INFO
 204                        "%s[%d] trap %s ip:%lx sp:%lx error:%lx",
 205                        tsk->comm, tsk->pid, str,
 206                        regs->ip, regs->sp, error_code);
 207                 print_vma_addr(" in ", regs->ip);
 208                 printk("\n");
 209         }
 210 #endif
 211
 212         if (info)
 213                 force_sig_info(signr, info, tsk);
 214         else
 215                 force_sig(signr, tsk);
 216         return;
 217
 218 kernel_trap:
 219         if (!fixup_exception(regs)) {
 220                 tsk->thread.error_code = error_code;
 221                 tsk->thread.trap_no = trapnr;
 222                 die(str, regs, error_code);
 223         }
 224         return;
 225
 226 #ifdef CONFIG_X86_32
 227 vm86_trap:
 228         if (handle_vm86_trap((struct kernel_vm86_regs *) regs,
 229                                                 error_code, trapnr))
 230                 goto trap_signal;
 231         return;
 232 #endif
 233 }
 234
 235 #define DO_ERROR(trapnr, signr, str, name)                              \
 236 dotraplinkage void do_##name(struct pt_regs *regs, long error_code)     \
 237 {                                                                       \
 238         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr)  \
 239                                                         == NOTIFY_STOP) \
 240                 return;                                                 \
 241         conditional_sti(regs);                                          \
 242         do_trap(trapnr, signr, str, regs, error_code, NULL);            \
 243 }
 244
 245 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr)         \
 246 dotraplinkage void do_##name(struct pt_regs *regs, long error_code)     \
 247 {                                                                       \
 248         siginfo_t info;                                                 \
 249         info.si_signo = signr;                                          \
 250         info.si_errno = 0;                                              \
 251         info.si_code = sicode;                                          \
 252         info.si_addr = (void __user *)siaddr;                           \
 253         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr)  \
 254                                                         == NOTIFY_STOP) \
 255                 return;                                                 \
 256         conditional_sti(regs);                                          \
 257         do_trap(trapnr, signr, str, regs, error_code, &info);           \
 258 }
 259
 260 DO_ERROR_INFO(0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip)
 261 DO_ERROR(4, SIGSEGV, "overflow", overflow)
 262 DO_ERROR(5, SIGSEGV, "bounds", bounds)
 263 DO_ERROR_INFO(6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip)
 264 DO_ERROR(9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
 265 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
 266 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
 267 #ifdef CONFIG_X86_32
 268 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
 269 #endif
 270 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
 271
 272 #ifdef CONFIG_X86_64
 273 /* Runs on IST stack */
 274 dotraplinkage void do_stack_segment(struct pt_regs *regs, long error_code)
 275 {
 276         if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
 277                         12, SIGBUS) == NOTIFY_STOP)
 278                 return;
 279         preempt_conditional_sti(regs);
 280         do_trap(12, SIGBUS, "stack segment", regs, error_code, NULL);
 281         preempt_conditional_cli(regs);
 282 }
 283
 284 dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
 285 {
 286         static const char str[] = "double fault";
 287         struct task_struct *tsk = current;
 288
 289         /* Return not checked because double check cannot be ignored */
 290         notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV);
 291
 292         tsk->thread.error_code = error_code;
 293         tsk->thread.trap_no = 8;
 294
 295         /*
 296          * This is always a kernel trap and never fixable (and thus must
 297          * never return).
 298          */
 299         for (;;)
 300                 die(str, regs, error_code);
 301 }
 302 #endif
 303
 304 dotraplinkage void __kprobes
 305 do_general_protection(struct pt_regs *regs, long error_code)
 306 {
 307         struct task_struct *tsk;
 308
 309         conditional_sti(regs);
 310
 311 #ifdef CONFIG_X86_32
 312         if (lazy_iobitmap_copy()) {
 313                 /* restart the faulting instruction */
 314                 return;
 315         }
 316
 317         if (regs->flags & X86_VM_MASK)
 318                 goto gp_in_vm86;
 319 #endif
 320
 321         tsk = current;
 322         if (!user_mode(regs))
 323                 goto gp_in_kernel;
 324
 325         tsk->thread.error_code = error_code;
 326         tsk->thread.trap_no = 13;
 327
 328         if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
 329                         printk_ratelimit()) {
 330                 printk(KERN_INFO
 331                         "%s[%d] general protection ip:%lx sp:%lx error:%lx",
 332                         tsk->comm, task_pid_nr(tsk),
 333                         regs->ip, regs->sp, error_code);
 334                 print_vma_addr(" in ", regs->ip);
 335                 printk("\n");
 336         }
 337
 338         force_sig(SIGSEGV, tsk);
 339         return;
 340
 341 #ifdef CONFIG_X86_32
 342 gp_in_vm86:
 343         local_irq_enable();
 344         handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
 345         return;
 346 #endif
 347
 348 gp_in_kernel:
 349         if (fixup_exception(regs))
 350                 return;
 351
 352         tsk->thread.error_code = error_code;
 353         tsk->thread.trap_no = 13;
 354         if (notify_die(DIE_GPF, "general protection fault", regs,
 355                                 error_code, 13, SIGSEGV) == NOTIFY_STOP)
 356                 return;
 357         die("general protection fault", regs, error_code);
 358 }
 359
 360 static notrace __kprobes void
 361 mem_parity_error(unsigned char reason, struct pt_regs *regs)
 362 {
 363         printk(KERN_EMERG
 364                 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
 365                         reason, smp_processor_id());
 366
 367         printk(KERN_EMERG
 368                 "You have some hardware problem, likely on the PCI bus.\n");
 369
 370 #if defined(CONFIG_EDAC)
 371         if (edac_handler_set()) {
 372                 edac_atomic_assert_error();
 373                 return;
 374         }
 375 #endif
 376
 377         if (panic_on_unrecovered_nmi)
 378                 panic("NMI: Not continuing");
 379
 380         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 381
 382         /* Clear and disable the memory parity error line. */
 383         reason = (reason & 0xf) | 4;
 384         outb(reason, 0x61);
 385 }
 386
 387 static notrace __kprobes void
 388 io_check_error(unsigned char reason, struct pt_regs *regs)
 389 {
 390         unsigned long i;
 391
 392         printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
 393         show_registers(regs);
 394
 395         /* Re-enable the IOCK line, wait for a few seconds */
 396         reason = (reason & 0xf) | 8;
 397         outb(reason, 0x61);
 398
 399         i = 2000;
 400         while (--i)
 401                 udelay(1000);
 402
 403         reason &= ~8;
 404         outb(reason, 0x61);
 405 }
 406
 407 static notrace __kprobes void
 408 unknown_nmi_error(unsigned char reason, struct pt_regs *regs)
 409 {
 410         if (notify_die(DIE_NMIUNKNOWN, "nmi", regs, reason, 2, SIGINT) ==
 411                         NOTIFY_STOP)
 412                 return;
 413 #ifdef CONFIG_MCA
 414         /*
 415          * Might actually be able to figure out what the guilty party
 416          * is:
 417          */
 418         if (MCA_bus) {
 419                 mca_handle_nmi();
 420                 return;
 421         }
 422 #endif
 423         printk(KERN_EMERG
 424                 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
 425                         reason, smp_processor_id());
 426
 427         printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
 428         if (panic_on_unrecovered_nmi)
 429                 panic("NMI: Not continuing");
 430
 431         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 432 }
 433
 434 static notrace __kprobes void default_do_nmi(struct pt_regs *regs)
 435 {
 436         unsigned char reason = 0;
 437         int cpu;
 438
 439         cpu = smp_processor_id();
 440
 441         /* Only the BSP gets external NMIs from the system. */
 442         if (!cpu)
 443                 reason = get_nmi_reason();
 444
 445         if (!(reason & 0xc0)) {
 446                 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
 447                                                                 == NOTIFY_STOP)
 448                         return;
 449 #ifdef CONFIG_X86_LOCAL_APIC
 450                 /*
 451                  * Ok, so this is none of the documented NMI sources,
 452                  * so it must be the NMI watchdog.
 453                  */
 454                 if (nmi_watchdog_tick(regs, reason))
 455                         return;
 456                 if (!do_nmi_callback(regs, cpu))
 457                         unknown_nmi_error(reason, regs);
 458 #else
 459                 unknown_nmi_error(reason, regs);
 460 #endif
 461
 462                 return;
 463         }
 464         if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
 465                 return;
 466
 467         /* AK: following checks seem to be broken on modern chipsets. FIXME */
 468         if (reason & 0x80)
 469                 mem_parity_error(reason, regs);
 470         if (reason & 0x40)
 471                 io_check_error(reason, regs);
 472 #ifdef CONFIG_X86_32
 473         /*
 474          * Reassert NMI in case it became active meanwhile
 475          * as it's edge-triggered:
 476          */
 477         reassert_nmi();
 478 #endif
 479 }
 480
 481 dotraplinkage notrace __kprobes void
 482 do_nmi(struct pt_regs *regs, long error_code)
 483 {
 484         nmi_enter();
 485
 486         inc_irq_stat(__nmi_count);
 487
 488         if (!ignore_nmis)
 489                 default_do_nmi(regs);
 490
 491         nmi_exit();
 492 }
 493
 494 void stop_nmi(void)
 495 {
 496         acpi_nmi_disable();
 497         ignore_nmis++;
 498 }
 499
 500 void restart_nmi(void)
 501 {
 502         ignore_nmis--;
 503         acpi_nmi_enable();
 504 }
 505
 506 /* May run on IST stack. */
 507 dotraplinkage void __kprobes do_int3(struct pt_regs *regs, long error_code)
 508 {
 509 #ifdef CONFIG_KPROBES
 510         if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
 511                         == NOTIFY_STOP)
 512                 return;
 513 #else
 514         if (notify_die(DIE_TRAP, "int3", regs, error_code, 3, SIGTRAP)
 515                         == NOTIFY_STOP)
 516                 return;
 517 #endif
 518
 519         preempt_conditional_sti(regs);
 520         do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
 521         preempt_conditional_cli(regs);
 522 }
 523
 524 #ifdef CONFIG_X86_64
 525 /*
 526  * Help handler running on IST stack to switch back to user stack
 527  * for scheduling or signal handling. The actual stack switch is done in
 528  * entry.S
 529  */
 530 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
 531 {
 532         struct pt_regs *regs = eregs;
 533         /* Did already sync */
 534         if (eregs == (struct pt_regs *)eregs->sp)
 535                 ;
 536         /* Exception from user space */
 537         else if (user_mode(eregs))
 538                 regs = task_pt_regs(current);
 539         /*
 540          * Exception from kernel and interrupts are enabled. Move to
 541          * kernel process stack.
 542          */
 543         else if (eregs->flags & X86_EFLAGS_IF)
 544                 regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
 545         if (eregs != regs)
 546                 *regs = *eregs;
 547         return regs;
 548 }
 549 #endif
 550
 551 /*
 552  * Our handling of the processor debug registers is non-trivial.
 553  * We do not clear them on entry and exit from the kernel. Therefore
 554  * it is possible to get a watchpoint trap here from inside the kernel.
 555  * However, the code in ./ptrace.c has ensured that the user can
 556  * only set watchpoints on userspace addresses. Therefore the in-kernel
 557  * watchpoint trap can only occur in code which is reading/writing
 558  * from user space. Such code must not hold kernel locks (since it
 559  * can equally take a page fault), therefore it is safe to call
 560  * force_sig_info even though that claims and releases locks.
 561  *
 562  * Code in ./signal.c ensures that the debug control register
 563  * is restored before we deliver any signal, and therefore that
 564  * user code runs with the correct debug control register even though
 565  * we clear it here.
 566  *
 567  * Being careful here means that we don't have to be as careful in a
 568  * lot of more complicated places (task switching can be a bit lazy
 569  * about restoring all the debug state, and ptrace doesn't have to
 570  * find every occurrence of the TF bit that could be saved away even
 571  * by user code)
 572  *
 573  * May run on IST stack.
 574  */
 575 dotraplinkage void __kprobes do_debug(struct pt_regs *regs, long error_code)
 576 {
 577         struct task_struct *tsk = current;
 578         unsigned long condition;
 579         int si_code;
 580
 581         get_debugreg(condition, 6);
 582
 583         /*
 584          * The processor cleared BTF, so don't mark that we need it set.
 585          */
 586         clear_tsk_thread_flag(tsk, TIF_DEBUGCTLMSR);
 587         tsk->thread.debugctlmsr = 0;
 588
 589         if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
 590                                                 SIGTRAP) == NOTIFY_STOP)
 591                 return;
 592
 593         /* It's safe to allow irq's after DR6 has been saved */
 594         preempt_conditional_sti(regs);
 595
 596         /* Mask out spurious debug traps due to lazy DR7 setting */
 597         if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
 598                 if (!tsk->thread.debugreg7)
 599                         goto clear_dr7;
 600         }
 601
 602 #ifdef CONFIG_X86_32
 603         if (regs->flags & X86_VM_MASK)
 604                 goto debug_vm86;
 605 #endif
 606
 607         /* Save debug status register where ptrace can see it */
 608         tsk->thread.debugreg6 = condition;
 609
 610         /*
 611          * Single-stepping through TF: make sure we ignore any events in
 612          * kernel space (but re-enable TF when returning to user mode).
 613          */
 614         if (condition & DR_STEP) {
 615                 if (!user_mode(regs))
 616                         goto clear_TF_reenable;
 617         }
 618
 619         si_code = get_si_code(condition);
 620         /* Ok, finally something we can handle */
 621         send_sigtrap(tsk, regs, error_code, si_code);
 622
 623         /*
 624          * Disable additional traps. They'll be re-enabled when
 625          * the signal is delivered.
 626          */
 627 clear_dr7:
 628         set_debugreg(0, 7);
 629         preempt_conditional_cli(regs);
 630         return;
 631
 632 #ifdef CONFIG_X86_32
 633 debug_vm86:
 634         /* reenable preemption: handle_vm86_trap() might sleep */
 635         dec_preempt_count();
 636         handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
 637         conditional_cli(regs);
 638         return;
 639 #endif
 640
 641 clear_TF_reenable:
 642         set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
 643         regs->flags &= ~X86_EFLAGS_TF;
 644         preempt_conditional_cli(regs);
 645         return;
 646 }
 647
 648 #ifdef CONFIG_X86_64
 649 static int kernel_math_error(struct pt_regs *regs, const char *str, int trapnr)
 650 {
 651         if (fixup_exception(regs))
 652                 return 1;
 653
 654         notify_die(DIE_GPF, str, regs, 0, trapnr, SIGFPE);
 655         /* Illegal floating point operation in the kernel */
 656         current->thread.trap_no = trapnr;
 657         die(str, regs, 0);
 658         return 0;
 659 }
 660 #endif
 661
 662 /*
 663  * Note that we play around with the 'TS' bit in an attempt to get
 664  * the correct behaviour even in the presence of the asynchronous
 665  * IRQ13 behaviour
 666  */
 667 void math_error(void __user *ip)
 668 {
 669         struct task_struct *task;
 670         siginfo_t info;
 671         unsigned short cwd, swd, err;
 672
 673         /*
 674          * Save the info for the exception handler and clear the error.
 675          */
 676         task = current;
 677         save_init_fpu(task);
 678         task->thread.trap_no = 16;
 679         task->thread.error_code = 0;
 680         info.si_signo = SIGFPE;
 681         info.si_errno = 0;
 682         info.si_addr = ip;
 683         /*
 684          * (~cwd & swd) will mask out exceptions that are not set to unmasked
 685          * status.  0x3f is the exception bits in these regs, 0x200 is the
 686          * C1 reg you need in case of a stack fault, 0x040 is the stack
 687          * fault bit.  We should only be taking one exception at a time,
 688          * so if this combination doesn't produce any single exception,
 689          * then we have a bad program that isn't synchronizing its FPU usage
 690          * and it will suffer the consequences since we won't be able to
 691          * fully reproduce the context of the exception
 692          */
 693         cwd = get_fpu_cwd(task);
 694         swd = get_fpu_swd(task);
 695
 696         err = swd & ~cwd;
 697
 698         if (err & 0x001) {      /* Invalid op */
 699                 /*
 700                  * swd & 0x240 == 0x040: Stack Underflow
 701                  * swd & 0x240 == 0x240: Stack Overflow
 702                  * User must clear the SF bit (0x40) if set
 703                  */
 704                 info.si_code = FPE_FLTINV;
 705         } else if (err & 0x004) { /* Divide by Zero */
 706                 info.si_code = FPE_FLTDIV;
 707         } else if (err & 0x008) { /* Overflow */
 708                 info.si_code = FPE_FLTOVF;
 709         } else if (err & 0x012) { /* Denormal, Underflow */
 710                 info.si_code = FPE_FLTUND;
 711         } else if (err & 0x020) { /* Precision */
 712                 info.si_code = FPE_FLTRES;
 713         } else {
 714                 /*
 715                  * If we're using IRQ 13, or supposedly even some trap 16
 716                  * implementations, it's possible we get a spurious trap...
 717                  */
 718                 return;         /* Spurious trap, no error */
 719         }
 720         force_sig_info(SIGFPE, &info, task);
 721 }
 722
 723 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
 724 {
 725         conditional_sti(regs);
 726
 727 #ifdef CONFIG_X86_32
 728         ignore_fpu_irq = 1;
 729 #else
 730         if (!user_mode(regs) &&
 731             kernel_math_error(regs, "kernel x87 math error", 16))
 732                 return;
 733 #endif
 734
 735         math_error((void __user *)regs->ip);
 736 }
 737
 738 static void simd_math_error(void __user *ip)
 739 {
 740         struct task_struct *task;
 741         siginfo_t info;
 742         unsigned short mxcsr;
 743
 744         /*
 745          * Save the info for the exception handler and clear the error.
 746          */
 747         task = current;
 748         save_init_fpu(task);
 749         task->thread.trap_no = 19;
 750         task->thread.error_code = 0;
 751         info.si_signo = SIGFPE;
 752         info.si_errno = 0;
 753         info.si_code = __SI_FAULT;
 754         info.si_addr = ip;
 755         /*
 756          * The SIMD FPU exceptions are handled a little differently, as there
 757          * is only a single status/control register.  Thus, to determine which
 758          * unmasked exception was caught we must mask the exception mask bits
 759          * at 0x1f80, and then use these to mask the exception bits at 0x3f.
 760          */
 761         mxcsr = get_fpu_mxcsr(task);
 762         switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
 763         case 0x000:
 764         default:
 765                 break;
 766         case 0x001: /* Invalid Op */
 767                 info.si_code = FPE_FLTINV;
 768                 break;
 769         case 0x002: /* Denormalize */
 770         case 0x010: /* Underflow */
 771                 info.si_code = FPE_FLTUND;
 772                 break;
 773         case 0x004: /* Zero Divide */
 774                 info.si_code = FPE_FLTDIV;
 775                 break;
 776         case 0x008: /* Overflow */
 777                 info.si_code = FPE_FLTOVF;
 778                 break;
 779         case 0x020: /* Precision */
 780                 info.si_code = FPE_FLTRES;
 781                 break;
 782         }
 783         force_sig_info(SIGFPE, &info, task);
 784 }
 785
 786 dotraplinkage void
 787 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
 788 {
 789         conditional_sti(regs);
 790
 791 #ifdef CONFIG_X86_32
 792         if (cpu_has_xmm) {
 793                 /* Handle SIMD FPU exceptions on PIII+ processors. */
 794                 ignore_fpu_irq = 1;
 795                 simd_math_error((void __user *)regs->ip);
 796                 return;
 797         }
 798         /*
 799          * Handle strange cache flush from user space exception
 800          * in all other cases.  This is undocumented behaviour.
 801          */
 802         if (regs->flags & X86_VM_MASK) {
 803                 handle_vm86_fault((struct kernel_vm86_regs *)regs, error_code);
 804                 return;
 805         }
 806         current->thread.trap_no = 19;
 807         current->thread.error_code = error_code;
 808         die_if_kernel("cache flush denied", regs, error_code);
 809         force_sig(SIGSEGV, current);
 810 #else
 811         if (!user_mode(regs) &&
 812                         kernel_math_error(regs, "kernel simd math error", 19))
 813                 return;
 814         simd_math_error((void __user *)regs->ip);
 815 #endif
 816 }
 817
 818 dotraplinkage void
 819 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
 820 {
 821         conditional_sti(regs);
 822 #if 0
 823         /* No need to warn about this any longer. */
 824         printk(KERN_INFO "Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
 825 #endif
 826 }
 827
 828 #ifdef CONFIG_X86_32
 829 unsigned long patch_espfix_desc(unsigned long uesp, unsigned long kesp)
 830 {
 831         struct desc_struct *gdt = get_cpu_gdt_table(smp_processor_id());
 832         unsigned long base = (kesp - uesp) & -THREAD_SIZE;
 833         unsigned long new_kesp = kesp - base;
 834         unsigned long lim_pages = (new_kesp | (THREAD_SIZE - 1)) >> PAGE_SHIFT;
 835         __u64 desc = *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS];
 836
 837         /* Set up base for espfix segment */
 838         desc &= 0x00f0ff0000000000ULL;
 839         desc |= ((((__u64)base) << 16) & 0x000000ffffff0000ULL) |
 840                 ((((__u64)base) << 32) & 0xff00000000000000ULL) |
 841                 ((((__u64)lim_pages) << 32) & 0x000f000000000000ULL) |
 842                 (lim_pages & 0xffff);
 843         *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS] = desc;
 844
 845         return new_kesp;
 846 }
 847 #else
 848 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
 849 {
 850 }
 851
 852 asmlinkage void __attribute__((weak)) mce_threshold_interrupt(void)
 853 {
 854 }
 855 #endif
 856
 857 /*
 858  * 'math_state_restore()' saves the current math information in the
 859  * old math state array, and gets the new ones from the current task
 860  *
 861  * Careful.. There are problems with IBM-designed IRQ13 behaviour.
 862  * Don't touch unless you *really* know how it works.
 863  *
 864  * Must be called with kernel preemption disabled (in this case,
 865  * local interrupts are disabled at the call-site in entry.S).
 866  */
 867 asmlinkage void math_state_restore(void)
 868 {
 869         struct thread_info *thread = current_thread_info();
 870         struct task_struct *tsk = thread->task;
 871
 872         if (!tsk_used_math(tsk)) {
 873                 local_irq_enable();
 874                 /*
 875                  * does a slab alloc which can sleep
 876                  */
 877                 if (init_fpu(tsk)) {
 878                         /*
 879                          * ran out of memory!
 880                          */
 881                         do_group_exit(SIGKILL);
 882                         return;
 883                 }
 884                 local_irq_disable();
 885         }
 886
 887         clts();                         /* Allow maths ops (or we recurse) */
 888 #ifdef CONFIG_X86_32
 889         restore_fpu(tsk);
 890 #else
 891         /*
 892          * Paranoid restore. send a SIGSEGV if we fail to restore the state.
 893          */
 894         if (unlikely(restore_fpu_checking(tsk))) {
 895                 stts();
 896                 force_sig(SIGSEGV, tsk);
 897                 return;
 898         }
 899 #endif
 900         thread->status |= TS_USEDFPU;   /* So we fnsave on switch_to() */
 901         tsk->fpu_counter++;
 902 }
 903 EXPORT_SYMBOL_GPL(math_state_restore);
 904
 905 #ifndef CONFIG_MATH_EMULATION
 906 void math_emulate(struct math_emu_info *info)
 907 {
 908         printk(KERN_EMERG
 909                 "math-emulation not enabled and no coprocessor found.\n");
 910         printk(KERN_EMERG "killing %s.\n", current->comm);
 911         force_sig(SIGFPE, current);
 912         schedule();
 913 }
 914 #endif /* CONFIG_MATH_EMULATION */
 915
 916 dotraplinkage void __kprobes
 917 do_device_not_available(struct pt_regs *regs, long error_code)
 918 {
 919 #ifdef CONFIG_X86_32
 920         if (read_cr0() & X86_CR0_EM) {
 921                 struct math_emu_info info = { };
 922
 923                 conditional_sti(regs);
 924
 925                 info.regs = regs;
 926                 math_emulate(&info);
 927         } else {
 928                 math_state_restore(); /* interrupts still off */
 929                 conditional_sti(regs);
 930         }
 931 #else
 932         math_state_restore();
 933 #endif
 934 }
 935
 936 #ifdef CONFIG_X86_32
 937 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
 938 {
 939         siginfo_t info;
 940         local_irq_enable();
 941
 942         info.si_signo = SIGILL;
 943         info.si_errno = 0;
 944         info.si_code = ILL_BADSTK;
 945         info.si_addr = 0;
 946         if (notify_die(DIE_TRAP, "iret exception",
 947                         regs, error_code, 32, SIGILL) == NOTIFY_STOP)
 948                 return;
 949         do_trap(32, SIGILL, "iret exception", regs, error_code, &info);
 950 }
 951 #endif
 952
 953 void __init trap_init(void)
 954 {
 955         int i;
 956
 957 #ifdef CONFIG_EISA
 958         void __iomem *p = early_ioremap(0x0FFFD9, 4);
 959
 960         if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
 961                 EISA_bus = 1;
 962         early_iounmap(p, 4);
 963 #endif
 964
 965         set_intr_gate(0, &divide_error);
 966         set_intr_gate_ist(1, &debug, DEBUG_STACK);
 967         set_intr_gate_ist(2, &nmi, NMI_STACK);
 968         /* int3 can be called from all */
 969         set_system_intr_gate_ist(3, &int3, DEBUG_STACK);
 970         /* int4 can be called from all */
 971         set_system_intr_gate(4, &overflow);
 972         set_intr_gate(5, &bounds);
 973         set_intr_gate(6, &invalid_op);
 974         set_intr_gate(7, &device_not_available);
 975 #ifdef CONFIG_X86_32
 976         set_task_gate(8, GDT_ENTRY_DOUBLEFAULT_TSS);
 977 #else
 978         set_intr_gate_ist(8, &double_fault, DOUBLEFAULT_STACK);
 979 #endif
 980         set_intr_gate(9, &coprocessor_segment_overrun);
 981         set_intr_gate(10, &invalid_TSS);
 982         set_intr_gate(11, &segment_not_present);
 983         set_intr_gate_ist(12, &stack_segment, STACKFAULT_STACK);
 984         set_intr_gate(13, &general_protection);
 985         set_intr_gate(14, &page_fault);
 986         set_intr_gate(15, &spurious_interrupt_bug);
 987         set_intr_gate(16, &coprocessor_error);
 988         set_intr_gate(17, &alignment_check);
 989 #ifdef CONFIG_X86_MCE
 990         set_intr_gate_ist(18, &machine_check, MCE_STACK);
 991 #endif
 992         set_intr_gate(19, &simd_coprocessor_error);
 993
 994 #ifdef CONFIG_IA32_EMULATION
 995         set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
 996 #endif
 997
 998 #ifdef CONFIG_X86_32
 999         if (cpu_has_fxsr) {
1000                 printk(KERN_INFO "Enabling fast FPU save and restore... ");
1001                 set_in_cr4(X86_CR4_OSFXSR);
1002                 printk("done.\n");
1003         }
1004         if (cpu_has_xmm) {
1005                 printk(KERN_INFO
1006                         "Enabling unmasked SIMD FPU exception support... ");
1007                 set_in_cr4(X86_CR4_OSXMMEXCPT);
1008                 printk("done.\n");
1009         }
1010
1011         set_system_trap_gate(SYSCALL_VECTOR, &system_call);
1012 #endif
1013
1014         /* Reserve all the builtin and the syscall vector: */
1015         for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
1016                 set_bit(i, used_vectors);
1017
1018 #ifdef CONFIG_X86_64
1019         set_bit(IA32_SYSCALL_VECTOR, used_vectors);
1020 #else
1021         set_bit(SYSCALL_VECTOR, used_vectors);
1022 #endif
1023         /*
1024          * Should be a barrier for any external CPU state:
1025          */
1026         cpu_init();
1027
1028 #ifdef CONFIG_X86_32
1029         x86_quirk_trap_init();
1030 #endif
1031 }