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.. We have a special link segment
  76  * for this.
  77  */
  78 gate_desc idt_table[NR_VECTORS]
  79         __attribute__((__section__(".data.idt"))) = { { { { 0, 0 } } }, };
  80 #endif
  81
  82 DECLARE_BITMAP(used_vectors, NR_VECTORS);
  83 EXPORT_SYMBOL_GPL(used_vectors);
  84
  85 static int ignore_nmis;
  86
  87 static inline void conditional_sti(struct pt_regs *regs)
  88 {
  89         if (regs->flags & X86_EFLAGS_IF)
  90                 local_irq_enable();
  91 }
  92
  93 static inline void preempt_conditional_sti(struct pt_regs *regs)
  94 {
  95         inc_preempt_count();
  96         if (regs->flags & X86_EFLAGS_IF)
  97                 local_irq_enable();
  98 }
  99
 100 static inline void conditional_cli(struct pt_regs *regs)
 101 {
 102         if (regs->flags & X86_EFLAGS_IF)
 103                 local_irq_disable();
 104 }
 105
 106 static inline void preempt_conditional_cli(struct pt_regs *regs)
 107 {
 108         if (regs->flags & X86_EFLAGS_IF)
 109                 local_irq_disable();
 110         dec_preempt_count();
 111 }
 112
 113 #ifdef CONFIG_X86_32
 114 static inline void
 115 die_if_kernel(const char *str, struct pt_regs *regs, long err)
 116 {
 117         if (!user_mode_vm(regs))
 118                 die(str, regs, err);
 119 }
 120 #endif
 121
 122 static void __kprobes
 123 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
 124         long error_code, siginfo_t *info)
 125 {
 126         struct task_struct *tsk = current;
 127
 128 #ifdef CONFIG_X86_32
 129         if (regs->flags & X86_VM_MASK) {
 130                 /*
 131                  * traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
 132                  * On nmi (interrupt 2), do_trap should not be called.
 133                  */
 134                 if (trapnr < 6)
 135                         goto vm86_trap;
 136                 goto trap_signal;
 137         }
 138 #endif
 139
 140         if (!user_mode(regs))
 141                 goto kernel_trap;
 142
 143 #ifdef CONFIG_X86_32
 144 trap_signal:
 145 #endif
 146         /*
 147          * We want error_code and trap_no set for userspace faults and
 148          * kernelspace faults which result in die(), but not
 149          * kernelspace faults which are fixed up.  die() gives the
 150          * process no chance to handle the signal and notice the
 151          * kernel fault information, so that won't result in polluting
 152          * the information about previously queued, but not yet
 153          * delivered, faults.  See also do_general_protection below.
 154          */
 155         tsk->thread.error_code = error_code;
 156         tsk->thread.trap_no = trapnr;
 157
 158 #ifdef CONFIG_X86_64
 159         if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
 160             printk_ratelimit()) {
 161                 printk(KERN_INFO
 162                        "%s[%d] trap %s ip:%lx sp:%lx error:%lx",
 163                        tsk->comm, tsk->pid, str,
 164                        regs->ip, regs->sp, error_code);
 165                 print_vma_addr(" in ", regs->ip);
 166                 printk("\n");
 167         }
 168 #endif
 169
 170         if (info)
 171                 force_sig_info(signr, info, tsk);
 172         else
 173                 force_sig(signr, tsk);
 174         return;
 175
 176 kernel_trap:
 177         if (!fixup_exception(regs)) {
 178                 tsk->thread.error_code = error_code;
 179                 tsk->thread.trap_no = trapnr;
 180                 die(str, regs, error_code);
 181         }
 182         return;
 183
 184 #ifdef CONFIG_X86_32
 185 vm86_trap:
 186         if (handle_vm86_trap((struct kernel_vm86_regs *) regs,
 187                                                 error_code, trapnr))
 188                 goto trap_signal;
 189         return;
 190 #endif
 191 }
 192
 193 #define DO_ERROR(trapnr, signr, str, name)                              \
 194 dotraplinkage void do_##name(struct pt_regs *regs, long error_code)     \
 195 {                                                                       \
 196         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr)  \
 197                                                         == NOTIFY_STOP) \
 198                 return;                                                 \
 199         conditional_sti(regs);                                          \
 200         do_trap(trapnr, signr, str, regs, error_code, NULL);            \
 201 }
 202
 203 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr)         \
 204 dotraplinkage void do_##name(struct pt_regs *regs, long error_code)     \
 205 {                                                                       \
 206         siginfo_t info;                                                 \
 207         info.si_signo = signr;                                          \
 208         info.si_errno = 0;                                              \
 209         info.si_code = sicode;                                          \
 210         info.si_addr = (void __user *)siaddr;                           \
 211         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr)  \
 212                                                         == NOTIFY_STOP) \
 213                 return;                                                 \
 214         conditional_sti(regs);                                          \
 215         do_trap(trapnr, signr, str, regs, error_code, &info);           \
 216 }
 217
 218 DO_ERROR_INFO(0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip)
 219 DO_ERROR(4, SIGSEGV, "overflow", overflow)
 220 DO_ERROR(5, SIGSEGV, "bounds", bounds)
 221 DO_ERROR_INFO(6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip)
 222 DO_ERROR(9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
 223 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
 224 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
 225 #ifdef CONFIG_X86_32
 226 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
 227 #endif
 228 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
 229
 230 #ifdef CONFIG_X86_64
 231 /* Runs on IST stack */
 232 dotraplinkage void do_stack_segment(struct pt_regs *regs, long error_code)
 233 {
 234         if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
 235                         12, SIGBUS) == NOTIFY_STOP)
 236                 return;
 237         preempt_conditional_sti(regs);
 238         do_trap(12, SIGBUS, "stack segment", regs, error_code, NULL);
 239         preempt_conditional_cli(regs);
 240 }
 241
 242 dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
 243 {
 244         static const char str[] = "double fault";
 245         struct task_struct *tsk = current;
 246
 247         /* Return not checked because double check cannot be ignored */
 248         notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV);
 249
 250         tsk->thread.error_code = error_code;
 251         tsk->thread.trap_no = 8;
 252
 253         /*
 254          * This is always a kernel trap and never fixable (and thus must
 255          * never return).
 256          */
 257         for (;;)
 258                 die(str, regs, error_code);
 259 }
 260 #endif
 261
 262 dotraplinkage void __kprobes
 263 do_general_protection(struct pt_regs *regs, long error_code)
 264 {
 265         struct task_struct *tsk;
 266
 267         conditional_sti(regs);
 268
 269 #ifdef CONFIG_X86_32
 270         if (regs->flags & X86_VM_MASK)
 271                 goto gp_in_vm86;
 272 #endif
 273
 274         tsk = current;
 275         if (!user_mode(regs))
 276                 goto gp_in_kernel;
 277
 278         tsk->thread.error_code = error_code;
 279         tsk->thread.trap_no = 13;
 280
 281         if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
 282                         printk_ratelimit()) {
 283                 printk(KERN_INFO
 284                         "%s[%d] general protection ip:%lx sp:%lx error:%lx",
 285                         tsk->comm, task_pid_nr(tsk),
 286                         regs->ip, regs->sp, error_code);
 287                 print_vma_addr(" in ", regs->ip);
 288                 printk("\n");
 289         }
 290
 291         force_sig(SIGSEGV, tsk);
 292         return;
 293
 294 #ifdef CONFIG_X86_32
 295 gp_in_vm86:
 296         local_irq_enable();
 297         handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
 298         return;
 299 #endif
 300
 301 gp_in_kernel:
 302         if (fixup_exception(regs))
 303                 return;
 304
 305         tsk->thread.error_code = error_code;
 306         tsk->thread.trap_no = 13;
 307         if (notify_die(DIE_GPF, "general protection fault", regs,
 308                                 error_code, 13, SIGSEGV) == NOTIFY_STOP)
 309                 return;
 310         die("general protection fault", regs, error_code);
 311 }
 312
 313 static notrace __kprobes void
 314 mem_parity_error(unsigned char reason, struct pt_regs *regs)
 315 {
 316         printk(KERN_EMERG
 317                 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
 318                         reason, smp_processor_id());
 319
 320         printk(KERN_EMERG
 321                 "You have some hardware problem, likely on the PCI bus.\n");
 322
 323 #if defined(CONFIG_EDAC)
 324         if (edac_handler_set()) {
 325                 edac_atomic_assert_error();
 326                 return;
 327         }
 328 #endif
 329
 330         if (panic_on_unrecovered_nmi)
 331                 panic("NMI: Not continuing");
 332
 333         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 334
 335         /* Clear and disable the memory parity error line. */
 336         reason = (reason & 0xf) | 4;
 337         outb(reason, 0x61);
 338 }
 339
 340 static notrace __kprobes void
 341 io_check_error(unsigned char reason, struct pt_regs *regs)
 342 {
 343         unsigned long i;
 344
 345         printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
 346         show_registers(regs);
 347
 348         if (panic_on_io_nmi)
 349                 panic("NMI IOCK error: Not continuing");
 350
 351         /* Re-enable the IOCK line, wait for a few seconds */
 352         reason = (reason & 0xf) | 8;
 353         outb(reason, 0x61);
 354
 355         i = 2000;
 356         while (--i)
 357                 udelay(1000);
 358
 359         reason &= ~8;
 360         outb(reason, 0x61);
 361 }
 362
 363 static notrace __kprobes void
 364 unknown_nmi_error(unsigned char reason, struct pt_regs *regs)
 365 {
 366         if (notify_die(DIE_NMIUNKNOWN, "nmi", regs, reason, 2, SIGINT) ==
 367                         NOTIFY_STOP)
 368                 return;
 369 #ifdef CONFIG_MCA
 370         /*
 371          * Might actually be able to figure out what the guilty party
 372          * is:
 373          */
 374         if (MCA_bus) {
 375                 mca_handle_nmi();
 376                 return;
 377         }
 378 #endif
 379         printk(KERN_EMERG
 380                 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
 381                         reason, smp_processor_id());
 382
 383         printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
 384         if (panic_on_unrecovered_nmi)
 385                 panic("NMI: Not continuing");
 386
 387         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 388 }
 389
 390 static notrace __kprobes void default_do_nmi(struct pt_regs *regs)
 391 {
 392         unsigned char reason = 0;
 393         int cpu;
 394
 395         cpu = smp_processor_id();
 396
 397         /* Only the BSP gets external NMIs from the system. */
 398         if (!cpu)
 399                 reason = get_nmi_reason();
 400
 401         if (!(reason & 0xc0)) {
 402                 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
 403                                                                 == NOTIFY_STOP)
 404                         return;
 405 #ifdef CONFIG_X86_LOCAL_APIC
 406                 /*
 407                  * Ok, so this is none of the documented NMI sources,
 408                  * so it must be the NMI watchdog.
 409                  */
 410                 if (nmi_watchdog_tick(regs, reason))
 411                         return;
 412                 if (!do_nmi_callback(regs, cpu))
 413                         unknown_nmi_error(reason, regs);
 414 #else
 415                 unknown_nmi_error(reason, regs);
 416 #endif
 417
 418                 return;
 419         }
 420         if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
 421                 return;
 422
 423         /* AK: following checks seem to be broken on modern chipsets. FIXME */
 424         if (reason & 0x80)
 425                 mem_parity_error(reason, regs);
 426         if (reason & 0x40)
 427                 io_check_error(reason, regs);
 428 #ifdef CONFIG_X86_32
 429         /*
 430          * Reassert NMI in case it became active meanwhile
 431          * as it's edge-triggered:
 432          */
 433         reassert_nmi();
 434 #endif
 435 }
 436
 437 dotraplinkage notrace __kprobes void
 438 do_nmi(struct pt_regs *regs, long error_code)
 439 {
 440         nmi_enter();
 441
 442         inc_irq_stat(__nmi_count);
 443
 444         if (!ignore_nmis)
 445                 default_do_nmi(regs);
 446
 447         nmi_exit();
 448 }
 449
 450 void stop_nmi(void)
 451 {
 452         acpi_nmi_disable();
 453         ignore_nmis++;
 454 }
 455
 456 void restart_nmi(void)
 457 {
 458         ignore_nmis--;
 459         acpi_nmi_enable();
 460 }
 461
 462 /* May run on IST stack. */
 463 dotraplinkage void __kprobes do_int3(struct pt_regs *regs, long error_code)
 464 {
 465 #ifdef CONFIG_KPROBES
 466         if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
 467                         == NOTIFY_STOP)
 468                 return;
 469 #else
 470         if (notify_die(DIE_TRAP, "int3", regs, error_code, 3, SIGTRAP)
 471                         == NOTIFY_STOP)
 472                 return;
 473 #endif
 474
 475         preempt_conditional_sti(regs);
 476         do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
 477         preempt_conditional_cli(regs);
 478 }
 479
 480 #ifdef CONFIG_X86_64
 481 /*
 482  * Help handler running on IST stack to switch back to user stack
 483  * for scheduling or signal handling. The actual stack switch is done in
 484  * entry.S
 485  */
 486 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
 487 {
 488         struct pt_regs *regs = eregs;
 489         /* Did already sync */
 490         if (eregs == (struct pt_regs *)eregs->sp)
 491                 ;
 492         /* Exception from user space */
 493         else if (user_mode(eregs))
 494                 regs = task_pt_regs(current);
 495         /*
 496          * Exception from kernel and interrupts are enabled. Move to
 497          * kernel process stack.
 498          */
 499         else if (eregs->flags & X86_EFLAGS_IF)
 500                 regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
 501         if (eregs != regs)
 502                 *regs = *eregs;
 503         return regs;
 504 }
 505 #endif
 506
 507 /*
 508  * Our handling of the processor debug registers is non-trivial.
 509  * We do not clear them on entry and exit from the kernel. Therefore
 510  * it is possible to get a watchpoint trap here from inside the kernel.
 511  * However, the code in ./ptrace.c has ensured that the user can
 512  * only set watchpoints on userspace addresses. Therefore the in-kernel
 513  * watchpoint trap can only occur in code which is reading/writing
 514  * from user space. Such code must not hold kernel locks (since it
 515  * can equally take a page fault), therefore it is safe to call
 516  * force_sig_info even though that claims and releases locks.
 517  *
 518  * Code in ./signal.c ensures that the debug control register
 519  * is restored before we deliver any signal, and therefore that
 520  * user code runs with the correct debug control register even though
 521  * we clear it here.
 522  *
 523  * Being careful here means that we don't have to be as careful in a
 524  * lot of more complicated places (task switching can be a bit lazy
 525  * about restoring all the debug state, and ptrace doesn't have to
 526  * find every occurrence of the TF bit that could be saved away even
 527  * by user code)
 528  *
 529  * May run on IST stack.
 530  */
 531 dotraplinkage void __kprobes do_debug(struct pt_regs *regs, long error_code)
 532 {
 533         struct task_struct *tsk = current;
 534         unsigned long condition;
 535         int si_code;
 536
 537         get_debugreg(condition, 6);
 538
 539         /* Catch kmemcheck conditions first of all! */
 540         if (condition & DR_STEP && kmemcheck_trap(regs))
 541                 return;
 542
 543         /*
 544          * The processor cleared BTF, so don't mark that we need it set.
 545          */
 546         clear_tsk_thread_flag(tsk, TIF_DEBUGCTLMSR);
 547         tsk->thread.debugctlmsr = 0;
 548
 549         if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
 550                                                 SIGTRAP) == NOTIFY_STOP)
 551                 return;
 552
 553         /* It's safe to allow irq's after DR6 has been saved */
 554         preempt_conditional_sti(regs);
 555
 556         /* Mask out spurious debug traps due to lazy DR7 setting */
 557         if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
 558                 if (!tsk->thread.debugreg7)
 559                         goto clear_dr7;
 560         }
 561
 562 #ifdef CONFIG_X86_32
 563         if (regs->flags & X86_VM_MASK)
 564                 goto debug_vm86;
 565 #endif
 566
 567         /* Save debug status register where ptrace can see it */
 568         tsk->thread.debugreg6 = condition;
 569
 570         /*
 571          * Single-stepping through TF: make sure we ignore any events in
 572          * kernel space (but re-enable TF when returning to user mode).
 573          */
 574         if (condition & DR_STEP) {
 575                 if (!user_mode(regs))
 576                         goto clear_TF_reenable;
 577         }
 578
 579         si_code = get_si_code(condition);
 580         /* Ok, finally something we can handle */
 581         send_sigtrap(tsk, regs, error_code, si_code);
 582
 583         /*
 584          * Disable additional traps. They'll be re-enabled when
 585          * the signal is delivered.
 586          */
 587 clear_dr7:
 588         set_debugreg(0, 7);
 589         preempt_conditional_cli(regs);
 590         return;
 591
 592 #ifdef CONFIG_X86_32
 593 debug_vm86:
 594         /* reenable preemption: handle_vm86_trap() might sleep */
 595         dec_preempt_count();
 596         handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
 597         conditional_cli(regs);
 598         return;
 599 #endif
 600
 601 clear_TF_reenable:
 602         set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
 603         regs->flags &= ~X86_EFLAGS_TF;
 604         preempt_conditional_cli(regs);
 605         return;
 606 }
 607
 608 #ifdef CONFIG_X86_64
 609 static int kernel_math_error(struct pt_regs *regs, const char *str, int trapnr)
 610 {
 611         if (fixup_exception(regs))
 612                 return 1;
 613
 614         notify_die(DIE_GPF, str, regs, 0, trapnr, SIGFPE);
 615         /* Illegal floating point operation in the kernel */
 616         current->thread.trap_no = trapnr;
 617         die(str, regs, 0);
 618         return 0;
 619 }
 620 #endif
 621
 622 /*
 623  * Note that we play around with the 'TS' bit in an attempt to get
 624  * the correct behaviour even in the presence of the asynchronous
 625  * IRQ13 behaviour
 626  */
 627 void math_error(void __user *ip)
 628 {
 629         struct task_struct *task;
 630         siginfo_t info;
 631         unsigned short cwd, swd, err;
 632
 633         /*
 634          * Save the info for the exception handler and clear the error.
 635          */
 636         task = current;
 637         save_init_fpu(task);
 638         task->thread.trap_no = 16;
 639         task->thread.error_code = 0;
 640         info.si_signo = SIGFPE;
 641         info.si_errno = 0;
 642         info.si_addr = ip;
 643         /*
 644          * (~cwd & swd) will mask out exceptions that are not set to unmasked
 645          * status.  0x3f is the exception bits in these regs, 0x200 is the
 646          * C1 reg you need in case of a stack fault, 0x040 is the stack
 647          * fault bit.  We should only be taking one exception at a time,
 648          * so if this combination doesn't produce any single exception,
 649          * then we have a bad program that isn't synchronizing its FPU usage
 650          * and it will suffer the consequences since we won't be able to
 651          * fully reproduce the context of the exception
 652          */
 653         cwd = get_fpu_cwd(task);
 654         swd = get_fpu_swd(task);
 655
 656         err = swd & ~cwd;
 657
 658         if (err & 0x001) {      /* Invalid op */
 659                 /*
 660                  * swd & 0x240 == 0x040: Stack Underflow
 661                  * swd & 0x240 == 0x240: Stack Overflow
 662                  * User must clear the SF bit (0x40) if set
 663                  */
 664                 info.si_code = FPE_FLTINV;
 665         } else if (err & 0x004) { /* Divide by Zero */
 666                 info.si_code = FPE_FLTDIV;
 667         } else if (err & 0x008) { /* Overflow */
 668                 info.si_code = FPE_FLTOVF;
 669         } else if (err & 0x012) { /* Denormal, Underflow */
 670                 info.si_code = FPE_FLTUND;
 671         } else if (err & 0x020) { /* Precision */
 672                 info.si_code = FPE_FLTRES;
 673         } else {
 674                 /*
 675                  * If we're using IRQ 13, or supposedly even some trap 16
 676                  * implementations, it's possible we get a spurious trap...
 677                  */
 678                 return;         /* Spurious trap, no error */
 679         }
 680         force_sig_info(SIGFPE, &info, task);
 681 }
 682
 683 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
 684 {
 685         conditional_sti(regs);
 686
 687 #ifdef CONFIG_X86_32
 688         ignore_fpu_irq = 1;
 689 #else
 690         if (!user_mode(regs) &&
 691             kernel_math_error(regs, "kernel x87 math error", 16))
 692                 return;
 693 #endif
 694
 695         math_error((void __user *)regs->ip);
 696 }
 697
 698 static void simd_math_error(void __user *ip)
 699 {
 700         struct task_struct *task;
 701         siginfo_t info;
 702         unsigned short mxcsr;
 703
 704         /*
 705          * Save the info for the exception handler and clear the error.
 706          */
 707         task = current;
 708         save_init_fpu(task);
 709         task->thread.trap_no = 19;
 710         task->thread.error_code = 0;
 711         info.si_signo = SIGFPE;
 712         info.si_errno = 0;
 713         info.si_code = __SI_FAULT;
 714         info.si_addr = ip;
 715         /*
 716          * The SIMD FPU exceptions are handled a little differently, as there
 717          * is only a single status/control register.  Thus, to determine which
 718          * unmasked exception was caught we must mask the exception mask bits
 719          * at 0x1f80, and then use these to mask the exception bits at 0x3f.
 720          */
 721         mxcsr = get_fpu_mxcsr(task);
 722         switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
 723         case 0x000:
 724         default:
 725                 break;
 726         case 0x001: /* Invalid Op */
 727                 info.si_code = FPE_FLTINV;
 728                 break;
 729         case 0x002: /* Denormalize */
 730         case 0x010: /* Underflow */
 731                 info.si_code = FPE_FLTUND;
 732                 break;
 733         case 0x004: /* Zero Divide */
 734                 info.si_code = FPE_FLTDIV;
 735                 break;
 736         case 0x008: /* Overflow */
 737                 info.si_code = FPE_FLTOVF;
 738                 break;
 739         case 0x020: /* Precision */
 740                 info.si_code = FPE_FLTRES;
 741                 break;
 742         }
 743         force_sig_info(SIGFPE, &info, task);
 744 }
 745
 746 dotraplinkage void
 747 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
 748 {
 749         conditional_sti(regs);
 750
 751 #ifdef CONFIG_X86_32
 752         if (cpu_has_xmm) {
 753                 /* Handle SIMD FPU exceptions on PIII+ processors. */
 754                 ignore_fpu_irq = 1;
 755                 simd_math_error((void __user *)regs->ip);
 756                 return;
 757         }
 758         /*
 759          * Handle strange cache flush from user space exception
 760          * in all other cases.  This is undocumented behaviour.
 761          */
 762         if (regs->flags & X86_VM_MASK) {
 763                 handle_vm86_fault((struct kernel_vm86_regs *)regs, error_code);
 764                 return;
 765         }
 766         current->thread.trap_no = 19;
 767         current->thread.error_code = error_code;
 768         die_if_kernel("cache flush denied", regs, error_code);
 769         force_sig(SIGSEGV, current);
 770 #else
 771         if (!user_mode(regs) &&
 772                         kernel_math_error(regs, "kernel simd math error", 19))
 773                 return;
 774         simd_math_error((void __user *)regs->ip);
 775 #endif
 776 }
 777
 778 dotraplinkage void
 779 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
 780 {
 781         conditional_sti(regs);
 782 #if 0
 783         /* No need to warn about this any longer. */
 784         printk(KERN_INFO "Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
 785 #endif
 786 }
 787
 788 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
 789 {
 790 }
 791
 792 asmlinkage void __attribute__((weak)) smp_threshold_interrupt(void)
 793 {
 794 }
 795
 796 /*
 797  * __math_state_restore assumes that cr0.TS is already clear and the
 798  * fpu state is all ready for use.  Used during context switch.
 799  */
 800 void __math_state_restore(void)
 801 {
 802         struct thread_info *thread = current_thread_info();
 803         struct task_struct *tsk = thread->task;
 804
 805         /*
 806          * Paranoid restore. send a SIGSEGV if we fail to restore the state.
 807          */
 808         if (unlikely(restore_fpu_checking(tsk))) {
 809                 stts();
 810                 force_sig(SIGSEGV, tsk);
 811                 return;
 812         }
 813
 814         thread->status |= TS_USEDFPU;   /* So we fnsave on switch_to() */
 815         tsk->fpu_counter++;
 816 }
 817
 818 /*
 819  * 'math_state_restore()' saves the current math information in the
 820  * old math state array, and gets the new ones from the current task
 821  *
 822  * Careful.. There are problems with IBM-designed IRQ13 behaviour.
 823  * Don't touch unless you *really* know how it works.
 824  *
 825  * Must be called with kernel preemption disabled (in this case,
 826  * local interrupts are disabled at the call-site in entry.S).
 827  */
 828 asmlinkage void math_state_restore(void)
 829 {
 830         struct thread_info *thread = current_thread_info();
 831         struct task_struct *tsk = thread->task;
 832
 833         if (!tsk_used_math(tsk)) {
 834                 local_irq_enable();
 835                 /*
 836                  * does a slab alloc which can sleep
 837                  */
 838                 if (init_fpu(tsk)) {
 839                         /*
 840                          * ran out of memory!
 841                          */
 842                         do_group_exit(SIGKILL);
 843                         return;
 844                 }
 845                 local_irq_disable();
 846         }
 847
 848         clts();                         /* Allow maths ops (or we recurse) */
 849
 850         __math_state_restore();
 851 }
 852 EXPORT_SYMBOL_GPL(math_state_restore);
 853
 854 #ifndef CONFIG_MATH_EMULATION
 855 void math_emulate(struct math_emu_info *info)
 856 {
 857         printk(KERN_EMERG
 858                 "math-emulation not enabled and no coprocessor found.\n");
 859         printk(KERN_EMERG "killing %s.\n", current->comm);
 860         force_sig(SIGFPE, current);
 861         schedule();
 862 }
 863 #endif /* CONFIG_MATH_EMULATION */
 864
 865 dotraplinkage void __kprobes
 866 do_device_not_available(struct pt_regs *regs, long error_code)
 867 {
 868 #ifdef CONFIG_X86_32
 869         if (read_cr0() & X86_CR0_EM) {
 870                 struct math_emu_info info = { };
 871
 872                 conditional_sti(regs);
 873
 874                 info.regs = regs;
 875                 math_emulate(&info);
 876         } else {
 877                 math_state_restore(); /* interrupts still off */
 878                 conditional_sti(regs);
 879         }
 880 #else
 881         math_state_restore();
 882 #endif
 883 }
 884
 885 #ifdef CONFIG_X86_32
 886 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
 887 {
 888         siginfo_t info;
 889         local_irq_enable();
 890
 891         info.si_signo = SIGILL;
 892         info.si_errno = 0;
 893         info.si_code = ILL_BADSTK;
 894         info.si_addr = NULL;
 895         if (notify_die(DIE_TRAP, "iret exception",
 896                         regs, error_code, 32, SIGILL) == NOTIFY_STOP)
 897                 return;
 898         do_trap(32, SIGILL, "iret exception", regs, error_code, &info);
 899 }
 900 #endif
 901
 902 void __init trap_init(void)
 903 {
 904         int i;
 905
 906 #ifdef CONFIG_EISA
 907         void __iomem *p = early_ioremap(0x0FFFD9, 4);
 908
 909         if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
 910                 EISA_bus = 1;
 911         early_iounmap(p, 4);
 912 #endif
 913
 914         set_intr_gate(0, &divide_error);
 915         set_intr_gate_ist(1, &debug, DEBUG_STACK);
 916         set_intr_gate_ist(2, &nmi, NMI_STACK);
 917         /* int3 can be called from all */
 918         set_system_intr_gate_ist(3, &int3, DEBUG_STACK);
 919         /* int4 can be called from all */
 920         set_system_intr_gate(4, &overflow);
 921         set_intr_gate(5, &bounds);
 922         set_intr_gate(6, &invalid_op);
 923         set_intr_gate(7, &device_not_available);
 924 #ifdef CONFIG_X86_32
 925         set_task_gate(8, GDT_ENTRY_DOUBLEFAULT_TSS);
 926 #else
 927         set_intr_gate_ist(8, &double_fault, DOUBLEFAULT_STACK);
 928 #endif
 929         set_intr_gate(9, &coprocessor_segment_overrun);
 930         set_intr_gate(10, &invalid_TSS);
 931         set_intr_gate(11, &segment_not_present);
 932         set_intr_gate_ist(12, &stack_segment, STACKFAULT_STACK);
 933         set_intr_gate(13, &general_protection);
 934         set_intr_gate(14, &page_fault);
 935         set_intr_gate(15, &spurious_interrupt_bug);
 936         set_intr_gate(16, &coprocessor_error);
 937         set_intr_gate(17, &alignment_check);
 938 #ifdef CONFIG_X86_MCE
 939         set_intr_gate_ist(18, &machine_check, MCE_STACK);
 940 #endif
 941         set_intr_gate(19, &simd_coprocessor_error);
 942
 943         /* Reserve all the builtin and the syscall vector: */
 944         for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
 945                 set_bit(i, used_vectors);
 946
 947 #ifdef CONFIG_IA32_EMULATION
 948         set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
 949         set_bit(IA32_SYSCALL_VECTOR, used_vectors);
 950 #endif
 951
 952 #ifdef CONFIG_X86_32
 953         if (cpu_has_fxsr) {
 954                 printk(KERN_INFO "Enabling fast FPU save and restore... ");
 955                 set_in_cr4(X86_CR4_OSFXSR);
 956                 printk("done.\n");
 957         }
 958         if (cpu_has_xmm) {
 959                 printk(KERN_INFO
 960                         "Enabling unmasked SIMD FPU exception support... ");
 961                 set_in_cr4(X86_CR4_OSXMMEXCPT);
 962                 printk("done.\n");
 963         }
 964
 965         set_system_trap_gate(SYSCALL_VECTOR, &system_call);
 966         set_bit(SYSCALL_VECTOR, used_vectors);
 967 #endif
 968
 969         /*
 970          * Should be a barrier for any external CPU state:
 971          */
 972         cpu_init();
 973
 974         x86_init.irqs.trap_init();
 975 }