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/kernel.h>
  19 #include <linux/module.h>
  20 #include <linux/ptrace.h>
  21 #include <linux/string.h>
  22 #include <linux/delay.h>
  23 #include <linux/errno.h>
  24 #include <linux/kexec.h>
  25 #include <linux/sched.h>
  26 #include <linux/timer.h>
  27 #include <linux/init.h>
  28 #include <linux/bug.h>
  29 #include <linux/nmi.h>
  30 #include <linux/mm.h>
  31 #include <linux/smp.h>
  32 #include <linux/io.h>
  33
  34 #ifdef CONFIG_EISA
  35 #include <linux/ioport.h>
  36 #include <linux/eisa.h>
  37 #endif
  38
  39 #ifdef CONFIG_MCA
  40 #include <linux/mca.h>
  41 #endif
  42
  43 #if defined(CONFIG_EDAC)
  44 #include <linux/edac.h>
  45 #endif
  46
  47 #include <asm/kmemcheck.h>
  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 #include <asm/mce.h>
  57
  58 #include <asm/mach_traps.h>
  59
  60 #ifdef CONFIG_X86_64
  61 #include <asm/x86_init.h>
  62 #include <asm/pgalloc.h>
  63 #include <asm/proto.h>
  64 #else
  65 #include <asm/processor-flags.h>
  66 #include <asm/setup.h>
  67
  68 asmlinkage int system_call(void);
  69
  70 /* Do we ignore FPU interrupts ? */
  71 char ignore_fpu_irq;
  72
  73 /*
  74  * The IDT has to be page-aligned to simplify the Pentium
  75  * F0 0F bug workaround.
  76  */
  77 gate_desc idt_table[NR_VECTORS] __page_aligned_data = { { { { 0, 0 } } }, };
  78 #endif
  79
  80 DECLARE_BITMAP(used_vectors, NR_VECTORS);
  81 EXPORT_SYMBOL_GPL(used_vectors);
  82
  83 static int ignore_nmis;
  84
  85 static inline void conditional_sti(struct pt_regs *regs)
  86 {
  87         if (regs->flags & X86_EFLAGS_IF)
  88                 local_irq_enable();
  89 }
  90
  91 static inline void preempt_conditional_sti(struct pt_regs *regs)
  92 {
  93         inc_preempt_count();
  94         if (regs->flags & X86_EFLAGS_IF)
  95                 local_irq_enable();
  96 }
  97
  98 static inline void conditional_cli(struct pt_regs *regs)
  99 {
 100         if (regs->flags & X86_EFLAGS_IF)
 101                 local_irq_disable();
 102 }
 103
 104 static inline void preempt_conditional_cli(struct pt_regs *regs)
 105 {
 106         if (regs->flags & X86_EFLAGS_IF)
 107                 local_irq_disable();
 108         dec_preempt_count();
 109 }
 110
 111 #ifdef CONFIG_X86_32
 112 static inline void
 113 die_if_kernel(const char *str, struct pt_regs *regs, long err)
 114 {
 115         if (!user_mode_vm(regs))
 116                 die(str, regs, err);
 117 }
 118 #endif
 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 notrace __kprobes void
 312 mem_parity_error(unsigned char reason, struct pt_regs *regs)
 313 {
 314         printk(KERN_EMERG
 315                 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
 316                         reason, smp_processor_id());
 317
 318         printk(KERN_EMERG
 319                 "You have some hardware problem, likely on the PCI bus.\n");
 320
 321 #if defined(CONFIG_EDAC)
 322         if (edac_handler_set()) {
 323                 edac_atomic_assert_error();
 324                 return;
 325         }
 326 #endif
 327
 328         if (panic_on_unrecovered_nmi)
 329                 panic("NMI: Not continuing");
 330
 331         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 332
 333         /* Clear and disable the memory parity error line. */
 334         reason = (reason & 0xf) | 4;
 335         outb(reason, 0x61);
 336 }
 337
 338 static notrace __kprobes void
 339 io_check_error(unsigned char reason, struct pt_regs *regs)
 340 {
 341         unsigned long i;
 342
 343         printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
 344         show_registers(regs);
 345
 346         if (panic_on_io_nmi)
 347                 panic("NMI IOCK error: Not continuing");
 348
 349         /* Re-enable the IOCK line, wait for a few seconds */
 350         reason = (reason & 0xf) | 8;
 351         outb(reason, 0x61);
 352
 353         i = 2000;
 354         while (--i)
 355                 udelay(1000);
 356
 357         reason &= ~8;
 358         outb(reason, 0x61);
 359 }
 360
 361 static notrace __kprobes void
 362 unknown_nmi_error(unsigned char reason, struct pt_regs *regs)
 363 {
 364         if (notify_die(DIE_NMIUNKNOWN, "nmi", regs, reason, 2, SIGINT) ==
 365                         NOTIFY_STOP)
 366                 return;
 367 #ifdef CONFIG_MCA
 368         /*
 369          * Might actually be able to figure out what the guilty party
 370          * is:
 371          */
 372         if (MCA_bus) {
 373                 mca_handle_nmi();
 374                 return;
 375         }
 376 #endif
 377         printk(KERN_EMERG
 378                 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
 379                         reason, smp_processor_id());
 380
 381         printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
 382         if (panic_on_unrecovered_nmi)
 383                 panic("NMI: Not continuing");
 384
 385         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 386 }
 387
 388 static notrace __kprobes void default_do_nmi(struct pt_regs *regs)
 389 {
 390         unsigned char reason = 0;
 391         int cpu;
 392
 393         cpu = smp_processor_id();
 394
 395         /* Only the BSP gets external NMIs from the system. */
 396         if (!cpu)
 397                 reason = get_nmi_reason();
 398
 399         if (!(reason & 0xc0)) {
 400                 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
 401                                                                 == NOTIFY_STOP)
 402                         return;
 403
 404 #ifdef CONFIG_X86_LOCAL_APIC
 405                 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT)
 406                                                         == NOTIFY_STOP)
 407                         return;
 408
 409 #ifndef CONFIG_LOCKUP_DETECTOR
 410                 /*
 411                  * Ok, so this is none of the documented NMI sources,
 412                  * so it must be the NMI watchdog.
 413                  */
 414                 if (nmi_watchdog_tick(regs, reason))
 415                         return;
 416                 if (!do_nmi_callback(regs, cpu))
 417 #endif /* !CONFIG_LOCKUP_DETECTOR */
 418                         unknown_nmi_error(reason, regs);
 419 #else
 420                 unknown_nmi_error(reason, regs);
 421 #endif
 422
 423                 return;
 424         }
 425         if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
 426                 return;
 427
 428         /* AK: following checks seem to be broken on modern chipsets. FIXME */
 429         if (reason & 0x80)
 430                 mem_parity_error(reason, regs);
 431         if (reason & 0x40)
 432                 io_check_error(reason, regs);
 433 #ifdef CONFIG_X86_32
 434         /*
 435          * Reassert NMI in case it became active meanwhile
 436          * as it's edge-triggered:
 437          */
 438         reassert_nmi();
 439 #endif
 440 }
 441
 442 dotraplinkage notrace __kprobes void
 443 do_nmi(struct pt_regs *regs, long error_code)
 444 {
 445         nmi_enter();
 446
 447         inc_irq_stat(__nmi_count);
 448
 449         if (!ignore_nmis)
 450                 default_do_nmi(regs);
 451
 452         nmi_exit();
 453 }
 454
 455 void stop_nmi(void)
 456 {
 457         acpi_nmi_disable();
 458         ignore_nmis++;
 459 }
 460
 461 void restart_nmi(void)
 462 {
 463         ignore_nmis--;
 464         acpi_nmi_enable();
 465 }
 466
 467 /* May run on IST stack. */
 468 dotraplinkage void __kprobes do_int3(struct pt_regs *regs, long error_code)
 469 {
 470 #ifdef CONFIG_KPROBES
 471         if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
 472                         == NOTIFY_STOP)
 473                 return;
 474 #else
 475         if (notify_die(DIE_TRAP, "int3", regs, error_code, 3, SIGTRAP)
 476                         == NOTIFY_STOP)
 477                 return;
 478 #endif
 479
 480         preempt_conditional_sti(regs);
 481         do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
 482         preempt_conditional_cli(regs);
 483 }
 484
 485 #ifdef CONFIG_X86_64
 486 /*
 487  * Help handler running on IST stack to switch back to user stack
 488  * for scheduling or signal handling. The actual stack switch is done in
 489  * entry.S
 490  */
 491 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
 492 {
 493         struct pt_regs *regs = eregs;
 494         /* Did already sync */
 495         if (eregs == (struct pt_regs *)eregs->sp)
 496                 ;
 497         /* Exception from user space */
 498         else if (user_mode(eregs))
 499                 regs = task_pt_regs(current);
 500         /*
 501          * Exception from kernel and interrupts are enabled. Move to
 502          * kernel process stack.
 503          */
 504         else if (eregs->flags & X86_EFLAGS_IF)
 505                 regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
 506         if (eregs != regs)
 507                 *regs = *eregs;
 508         return regs;
 509 }
 510 #endif
 511
 512 /*
 513  * Our handling of the processor debug registers is non-trivial.
 514  * We do not clear them on entry and exit from the kernel. Therefore
 515  * it is possible to get a watchpoint trap here from inside the kernel.
 516  * However, the code in ./ptrace.c has ensured that the user can
 517  * only set watchpoints on userspace addresses. Therefore the in-kernel
 518  * watchpoint trap can only occur in code which is reading/writing
 519  * from user space. Such code must not hold kernel locks (since it
 520  * can equally take a page fault), therefore it is safe to call
 521  * force_sig_info even though that claims and releases locks.
 522  *
 523  * Code in ./signal.c ensures that the debug control register
 524  * is restored before we deliver any signal, and therefore that
 525  * user code runs with the correct debug control register even though
 526  * we clear it here.
 527  *
 528  * Being careful here means that we don't have to be as careful in a
 529  * lot of more complicated places (task switching can be a bit lazy
 530  * about restoring all the debug state, and ptrace doesn't have to
 531  * find every occurrence of the TF bit that could be saved away even
 532  * by user code)
 533  *
 534  * May run on IST stack.
 535  */
 536 dotraplinkage void __kprobes do_debug(struct pt_regs *regs, long error_code)
 537 {
 538         struct task_struct *tsk = current;
 539         unsigned long dr6;
 540         int si_code;
 541
 542         get_debugreg(dr6, 6);
 543
 544         /* Filter out all the reserved bits which are preset to 1 */
 545         dr6 &= ~DR6_RESERVED;
 546
 547         /* Catch kmemcheck conditions first of all! */
 548         if ((dr6 & DR_STEP) && kmemcheck_trap(regs))
 549                 return;
 550
 551         /* DR6 may or may not be cleared by the CPU */
 552         set_debugreg(0, 6);
 553         /*
 554          * The processor cleared BTF, so don't mark that we need it set.
 555          */
 556         clear_tsk_thread_flag(tsk, TIF_DEBUGCTLMSR);
 557         tsk->thread.debugctlmsr = 0;
 558
 559         /* Store the virtualized DR6 value */
 560         tsk->thread.debugreg6 = dr6;
 561
 562         if (notify_die(DIE_DEBUG, "debug", regs, PTR_ERR(&dr6), error_code,
 563                                                         SIGTRAP) == NOTIFY_STOP)
 564                 return;
 565
 566         /* It's safe to allow irq's after DR6 has been saved */
 567         preempt_conditional_sti(regs);
 568
 569         if (regs->flags & X86_VM_MASK) {
 570                 handle_vm86_trap((struct kernel_vm86_regs *) regs,
 571                                 error_code, 1);
 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))
 589                 send_sigtrap(tsk, regs, error_code, si_code);
 590         preempt_conditional_cli(regs);
 591
 592         return;
 593 }
 594
 595 #ifdef CONFIG_X86_64
 596 static int kernel_math_error(struct pt_regs *regs, const char *str, int trapnr)
 597 {
 598         if (fixup_exception(regs))
 599                 return 1;
 600
 601         notify_die(DIE_GPF, str, regs, 0, trapnr, SIGFPE);
 602         /* Illegal floating point operation in the kernel */
 603         current->thread.trap_no = trapnr;
 604         die(str, regs, 0);
 605         return 0;
 606 }
 607 #endif
 608
 609 /*
 610  * Note that we play around with the 'TS' bit in an attempt to get
 611  * the correct behaviour even in the presence of the asynchronous
 612  * IRQ13 behaviour
 613  */
 614 void math_error(void __user *ip)
 615 {
 616         struct task_struct *task;
 617         siginfo_t info;
 618         unsigned short cwd, swd, err;
 619
 620         /*
 621          * Save the info for the exception handler and clear the error.
 622          */
 623         task = current;
 624         save_init_fpu(task);
 625         task->thread.trap_no = 16;
 626         task->thread.error_code = 0;
 627         info.si_signo = SIGFPE;
 628         info.si_errno = 0;
 629         info.si_addr = ip;
 630         /*
 631          * (~cwd & swd) will mask out exceptions that are not set to unmasked
 632          * status.  0x3f is the exception bits in these regs, 0x200 is the
 633          * C1 reg you need in case of a stack fault, 0x040 is the stack
 634          * fault bit.  We should only be taking one exception at a time,
 635          * so if this combination doesn't produce any single exception,
 636          * then we have a bad program that isn't synchronizing its FPU usage
 637          * and it will suffer the consequences since we won't be able to
 638          * fully reproduce the context of the exception
 639          */
 640         cwd = get_fpu_cwd(task);
 641         swd = get_fpu_swd(task);
 642
 643         err = swd & ~cwd;
 644
 645         if (err & 0x001) {      /* Invalid op */
 646                 /*
 647                  * swd & 0x240 == 0x040: Stack Underflow
 648                  * swd & 0x240 == 0x240: Stack Overflow
 649                  * User must clear the SF bit (0x40) if set
 650                  */
 651                 info.si_code = FPE_FLTINV;
 652         } else if (err & 0x004) { /* Divide by Zero */
 653                 info.si_code = FPE_FLTDIV;
 654         } else if (err & 0x008) { /* Overflow */
 655                 info.si_code = FPE_FLTOVF;
 656         } else if (err & 0x012) { /* Denormal, Underflow */
 657                 info.si_code = FPE_FLTUND;
 658         } else if (err & 0x020) { /* Precision */
 659                 info.si_code = FPE_FLTRES;
 660         } else {
 661                 /*
 662                  * If we're using IRQ 13, or supposedly even some trap 16
 663                  * implementations, it's possible we get a spurious trap...
 664                  */
 665                 return;         /* Spurious trap, no error */
 666         }
 667         force_sig_info(SIGFPE, &info, task);
 668 }
 669
 670 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
 671 {
 672         conditional_sti(regs);
 673
 674 #ifdef CONFIG_X86_32
 675         ignore_fpu_irq = 1;
 676 #else
 677         if (!user_mode(regs) &&
 678             kernel_math_error(regs, "kernel x87 math error", 16))
 679                 return;
 680 #endif
 681
 682         math_error((void __user *)regs->ip);
 683 }
 684
 685 static void simd_math_error(void __user *ip)
 686 {
 687         struct task_struct *task;
 688         siginfo_t info;
 689         unsigned short mxcsr;
 690
 691         /*
 692          * Save the info for the exception handler and clear the error.
 693          */
 694         task = current;
 695         save_init_fpu(task);
 696         task->thread.trap_no = 19;
 697         task->thread.error_code = 0;
 698         info.si_signo = SIGFPE;
 699         info.si_errno = 0;
 700         info.si_code = __SI_FAULT;
 701         info.si_addr = ip;
 702         /*
 703          * The SIMD FPU exceptions are handled a little differently, as there
 704          * is only a single status/control register.  Thus, to determine which
 705          * unmasked exception was caught we must mask the exception mask bits
 706          * at 0x1f80, and then use these to mask the exception bits at 0x3f.
 707          */
 708         mxcsr = get_fpu_mxcsr(task);
 709         switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
 710         case 0x000:
 711         default:
 712                 break;
 713         case 0x001: /* Invalid Op */
 714                 info.si_code = FPE_FLTINV;
 715                 break;
 716         case 0x002: /* Denormalize */
 717         case 0x010: /* Underflow */
 718                 info.si_code = FPE_FLTUND;
 719                 break;
 720         case 0x004: /* Zero Divide */
 721                 info.si_code = FPE_FLTDIV;
 722                 break;
 723         case 0x008: /* Overflow */
 724                 info.si_code = FPE_FLTOVF;
 725                 break;
 726         case 0x020: /* Precision */
 727                 info.si_code = FPE_FLTRES;
 728                 break;
 729         }
 730         force_sig_info(SIGFPE, &info, task);
 731 }
 732
 733 dotraplinkage void
 734 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
 735 {
 736         conditional_sti(regs);
 737
 738 #ifdef CONFIG_X86_32
 739         if (cpu_has_xmm) {
 740                 /* Handle SIMD FPU exceptions on PIII+ processors. */
 741                 ignore_fpu_irq = 1;
 742                 simd_math_error((void __user *)regs->ip);
 743                 return;
 744         }
 745         /*
 746          * Handle strange cache flush from user space exception
 747          * in all other cases.  This is undocumented behaviour.
 748          */
 749         if (regs->flags & X86_VM_MASK) {
 750                 handle_vm86_fault((struct kernel_vm86_regs *)regs, error_code);
 751                 return;
 752         }
 753         current->thread.trap_no = 19;
 754         current->thread.error_code = error_code;
 755         die_if_kernel("cache flush denied", regs, error_code);
 756         force_sig(SIGSEGV, current);
 757 #else
 758         if (!user_mode(regs) &&
 759                         kernel_math_error(regs, "kernel simd math error", 19))
 760                 return;
 761         simd_math_error((void __user *)regs->ip);
 762 #endif
 763 }
 764
 765 dotraplinkage void
 766 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
 767 {
 768         conditional_sti(regs);
 769 #if 0
 770         /* No need to warn about this any longer. */
 771         printk(KERN_INFO "Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
 772 #endif
 773 }
 774
 775 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
 776 {
 777 }
 778
 779 asmlinkage void __attribute__((weak)) smp_threshold_interrupt(void)
 780 {
 781 }
 782
 783 /*
 784  * __math_state_restore assumes that cr0.TS is already clear and the
 785  * fpu state is all ready for use.  Used during context switch.
 786  */
 787 void __math_state_restore(void)
 788 {
 789         struct thread_info *thread = current_thread_info();
 790         struct task_struct *tsk = thread->task;
 791
 792         /*
 793          * Paranoid restore. send a SIGSEGV if we fail to restore the state.
 794          */
 795         if (unlikely(restore_fpu_checking(tsk))) {
 796                 stts();
 797                 force_sig(SIGSEGV, tsk);
 798                 return;
 799         }
 800
 801         thread->status |= TS_USEDFPU;   /* So we fnsave on switch_to() */
 802         tsk->fpu_counter++;
 803 }
 804
 805 /*
 806  * 'math_state_restore()' saves the current math information in the
 807  * old math state array, and gets the new ones from the current task
 808  *
 809  * Careful.. There are problems with IBM-designed IRQ13 behaviour.
 810  * Don't touch unless you *really* know how it works.
 811  *
 812  * Must be called with kernel preemption disabled (in this case,
 813  * local interrupts are disabled at the call-site in entry.S).
 814  */
 815 asmlinkage void math_state_restore(void)
 816 {
 817         struct thread_info *thread = current_thread_info();
 818         struct task_struct *tsk = thread->task;
 819
 820         if (!tsk_used_math(tsk)) {
 821                 local_irq_enable();
 822                 /*
 823                  * does a slab alloc which can sleep
 824                  */
 825                 if (init_fpu(tsk)) {
 826                         /*
 827                          * ran out of memory!
 828                          */
 829                         do_group_exit(SIGKILL);
 830                         return;
 831                 }
 832                 local_irq_disable();
 833         }
 834
 835         clts();                         /* Allow maths ops (or we recurse) */
 836
 837         __math_state_restore();
 838 }
 839 EXPORT_SYMBOL_GPL(math_state_restore);
 840
 841 #ifndef CONFIG_MATH_EMULATION
 842 void math_emulate(struct math_emu_info *info)
 843 {
 844         printk(KERN_EMERG
 845                 "math-emulation not enabled and no coprocessor found.\n");
 846         printk(KERN_EMERG "killing %s.\n", current->comm);
 847         force_sig(SIGFPE, current);
 848         schedule();
 849 }
 850 #endif /* CONFIG_MATH_EMULATION */
 851
 852 dotraplinkage void __kprobes
 853 do_device_not_available(struct pt_regs *regs, long error_code)
 854 {
 855 #ifdef CONFIG_X86_32
 856         if (read_cr0() & X86_CR0_EM) {
 857                 struct math_emu_info info = { };
 858
 859                 conditional_sti(regs);
 860
 861                 info.regs = regs;
 862                 math_emulate(&info);
 863         } else {
 864                 math_state_restore(); /* interrupts still off */
 865                 conditional_sti(regs);
 866         }
 867 #else
 868         math_state_restore();
 869 #endif
 870 }
 871
 872 #ifdef CONFIG_X86_32
 873 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
 874 {
 875         siginfo_t info;
 876         local_irq_enable();
 877
 878         info.si_signo = SIGILL;
 879         info.si_errno = 0;
 880         info.si_code = ILL_BADSTK;
 881         info.si_addr = NULL;
 882         if (notify_die(DIE_TRAP, "iret exception",
 883                         regs, error_code, 32, SIGILL) == NOTIFY_STOP)
 884                 return;
 885         do_trap(32, SIGILL, "iret exception", regs, error_code, &info);
 886 }
 887 #endif
 888
 889 void __init trap_init(void)
 890 {
 891         int i;
 892
 893 #ifdef CONFIG_EISA
 894         void __iomem *p = early_ioremap(0x0FFFD9, 4);
 895
 896         if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
 897                 EISA_bus = 1;
 898         early_iounmap(p, 4);
 899 #endif
 900
 901         set_intr_gate(0, &divide_error);
 902         set_intr_gate_ist(1, &debug, DEBUG_STACK);
 903         set_intr_gate_ist(2, &nmi, NMI_STACK);
 904         /* int3 can be called from all */
 905         set_system_intr_gate_ist(3, &int3, DEBUG_STACK);
 906         /* int4 can be called from all */
 907         set_system_intr_gate(4, &overflow);
 908         set_intr_gate(5, &bounds);
 909         set_intr_gate(6, &invalid_op);
 910         set_intr_gate(7, &device_not_available);
 911 #ifdef CONFIG_X86_32
 912         set_task_gate(8, GDT_ENTRY_DOUBLEFAULT_TSS);
 913 #else
 914         set_intr_gate_ist(8, &double_fault, DOUBLEFAULT_STACK);
 915 #endif
 916         set_intr_gate(9, &coprocessor_segment_overrun);
 917         set_intr_gate(10, &invalid_TSS);
 918         set_intr_gate(11, &segment_not_present);
 919         set_intr_gate_ist(12, &stack_segment, STACKFAULT_STACK);
 920         set_intr_gate(13, &general_protection);
 921         set_intr_gate(14, &page_fault);
 922         set_intr_gate(15, &spurious_interrupt_bug);
 923         set_intr_gate(16, &coprocessor_error);
 924         set_intr_gate(17, &alignment_check);
 925 #ifdef CONFIG_X86_MCE
 926         set_intr_gate_ist(18, &machine_check, MCE_STACK);
 927 #endif
 928         set_intr_gate(19, &simd_coprocessor_error);
 929
 930         /* Reserve all the builtin and the syscall vector: */
 931         for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
 932                 set_bit(i, used_vectors);
 933
 934 #ifdef CONFIG_IA32_EMULATION
 935         set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
 936         set_bit(IA32_SYSCALL_VECTOR, used_vectors);
 937 #endif
 938
 939 #ifdef CONFIG_X86_32
 940         if (cpu_has_fxsr) {
 941                 printk(KERN_INFO "Enabling fast FPU save and restore... ");
 942                 set_in_cr4(X86_CR4_OSFXSR);
 943                 printk("done.\n");
 944         }
 945         if (cpu_has_xmm) {
 946                 printk(KERN_INFO
 947                         "Enabling unmasked SIMD FPU exception support... ");
 948                 set_in_cr4(X86_CR4_OSXMMEXCPT);
 949                 printk("done.\n");
 950         }
 951
 952         set_system_trap_gate(SYSCALL_VECTOR, &system_call);
 953         set_bit(SYSCALL_VECTOR, used_vectors);
 954 #endif
 955
 956         /*
 957          * Should be a barrier for any external CPU state:
 958          */
 959         cpu_init();
 960
 961         x86_init.irqs.trap_init();
 962 }