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