arch/x86/kernel/traps.c

   1 /*
   2  *  Copyright (C) 1991, 1992  Linus Torvalds
   3  *  Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs
   4  *
   5  *  Pentium III FXSR, SSE support
   6  *      Gareth Hughes <gareth@valinux.com>, May 2000
   7  */
   8
   9 /*
  10  * Handle hardware traps and faults.
  11  */
  12 #include <linux/interrupt.h>
  13 #include <linux/kallsyms.h>
  14 #include <linux/spinlock.h>
  15 #include <linux/kprobes.h>
  16 #include <linux/uaccess.h>
  17 #include <linux/utsname.h>
  18 #include <linux/kdebug.h>
  19 #include <linux/kernel.h>
  20 #include <linux/module.h>
  21 #include <linux/ptrace.h>
  22 #include <linux/string.h>
  23 #include <linux/delay.h>
  24 #include <linux/errno.h>
  25 #include <linux/kexec.h>
  26 #include <linux/sched.h>
  27 #include <linux/timer.h>
  28 #include <linux/init.h>
  29 #include <linux/bug.h>
  30 #include <linux/nmi.h>
  31 #include <linux/mm.h>
  32 #include <linux/smp.h>
  33 #include <linux/io.h>
  34
  35 #ifdef CONFIG_EISA
  36 #include <linux/ioport.h>
  37 #include <linux/eisa.h>
  38 #endif
  39
  40 #ifdef CONFIG_MCA
  41 #include <linux/mca.h>
  42 #endif
  43
  44 #if defined(CONFIG_EDAC)
  45 #include <linux/edac.h>
  46 #endif
  47
  48 #include <asm/stacktrace.h>
  49 #include <asm/processor.h>
  50 #include <asm/debugreg.h>
  51 #include <asm/atomic.h>
  52 #include <asm/system.h>
  53 #include <asm/traps.h>
  54 #include <asm/desc.h>
  55 #include <asm/i387.h>
  56
  57 #include <mach_traps.h>
  58
  59 #ifdef CONFIG_X86_64
  60 #include <asm/pgalloc.h>
  61 #include <asm/proto.h>
  62 #else
  63 #include <asm/processor-flags.h>
  64 #include <asm/arch_hooks.h>
  65 #include <asm/traps.h>
  66
  67 #include "cpu/mcheck/mce.h"
  68
  69 asmlinkage int system_call(void);
  70
  71 /* Do we ignore FPU interrupts ? */
  72 char ignore_fpu_irq;
  73
  74 /*
  75  * The IDT has to be page-aligned to simplify the Pentium
  76  * F0 0F bug workaround.. We have a special link segment
  77  * for this.
  78  */
  79 gate_desc idt_table[256]
  80         __attribute__((__section__(".data.idt"))) = { { { { 0, 0 } } }, };
  81 #endif
  82
  83 DECLARE_BITMAP(used_vectors, NR_VECTORS);
  84 EXPORT_SYMBOL_GPL(used_vectors);
  85
  86 static int ignore_nmis;
  87
  88 static inline void conditional_sti(struct pt_regs *regs)
  89 {
  90         if (regs->flags & X86_EFLAGS_IF)
  91                 local_irq_enable();
  92 }
  93
  94 static inline void preempt_conditional_sti(struct pt_regs *regs)
  95 {
  96         inc_preempt_count();
  97         if (regs->flags & X86_EFLAGS_IF)
  98                 local_irq_enable();
  99 }
 100
 101 static inline void preempt_conditional_cli(struct pt_regs *regs)
 102 {
 103         if (regs->flags & X86_EFLAGS_IF)
 104                 local_irq_disable();
 105         dec_preempt_count();
 106 }
 107
 108 #ifdef CONFIG_X86_32
 109 static inline void
 110 die_if_kernel(const char *str, struct pt_regs *regs, long err)
 111 {
 112         if (!user_mode_vm(regs))
 113                 die(str, regs, err);
 114 }
 115
 116 /*
 117  * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
 118  * invalid offset set (the LAZY one) and the faulting thread has
 119  * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS,
 120  * we set the offset field correctly and return 1.
 121  */
 122 static int lazy_iobitmap_copy(void)
 123 {
 124         struct thread_struct *thread;
 125         struct tss_struct *tss;
 126         int cpu;
 127
 128         cpu = get_cpu();
 129         tss = &per_cpu(init_tss, cpu);
 130         thread = &current->thread;
 131
 132         if (tss->x86_tss.io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
 133             thread->io_bitmap_ptr) {
 134                 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
 135                        thread->io_bitmap_max);
 136                 /*
 137                  * If the previously set map was extending to higher ports
 138                  * than the current one, pad extra space with 0xff (no access).
 139                  */
 140                 if (thread->io_bitmap_max < tss->io_bitmap_max) {
 141                         memset((char *) tss->io_bitmap +
 142                                 thread->io_bitmap_max, 0xff,
 143                                 tss->io_bitmap_max - thread->io_bitmap_max);
 144                 }
 145                 tss->io_bitmap_max = thread->io_bitmap_max;
 146                 tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
 147                 tss->io_bitmap_owner = thread;
 148                 put_cpu();
 149
 150                 return 1;
 151         }
 152         put_cpu();
 153
 154         return 0;
 155 }
 156 #endif
 157
 158 static void __kprobes
 159 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
 160         long error_code, siginfo_t *info)
 161 {
 162         struct task_struct *tsk = current;
 163
 164 #ifdef CONFIG_X86_32
 165         if (regs->flags & X86_VM_MASK) {
 166                 /*
 167                  * traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
 168                  * On nmi (interrupt 2), do_trap should not be called.
 169                  */
 170                 if (trapnr < 6)
 171                         goto vm86_trap;
 172                 goto trap_signal;
 173         }
 174 #endif
 175
 176         if (!user_mode(regs))
 177                 goto kernel_trap;
 178
 179 #ifdef CONFIG_X86_32
 180 trap_signal:
 181 #endif
 182         /*
 183          * We want error_code and trap_no set for userspace faults and
 184          * kernelspace faults which result in die(), but not
 185          * kernelspace faults which are fixed up.  die() gives the
 186          * process no chance to handle the signal and notice the
 187          * kernel fault information, so that won't result in polluting
 188          * the information about previously queued, but not yet
 189          * delivered, faults.  See also do_general_protection below.
 190          */
 191         tsk->thread.error_code = error_code;
 192         tsk->thread.trap_no = trapnr;
 193
 194 #ifdef CONFIG_X86_64
 195         if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
 196             printk_ratelimit()) {
 197                 printk(KERN_INFO
 198                        "%s[%d] trap %s ip:%lx sp:%lx error:%lx",
 199                        tsk->comm, tsk->pid, str,
 200                        regs->ip, regs->sp, error_code);
 201                 print_vma_addr(" in ", regs->ip);
 202                 printk("\n");
 203         }
 204 #endif
 205
 206         if (info)
 207                 force_sig_info(signr, info, tsk);
 208         else
 209                 force_sig(signr, tsk);
 210         return;
 211
 212 kernel_trap:
 213         if (!fixup_exception(regs)) {
 214                 tsk->thread.error_code = error_code;
 215                 tsk->thread.trap_no = trapnr;
 216                 die(str, regs, error_code);
 217         }
 218         return;
 219
 220 #ifdef CONFIG_X86_32
 221 vm86_trap:
 222         if (handle_vm86_trap((struct kernel_vm86_regs *) regs,
 223                                                 error_code, trapnr))
 224                 goto trap_signal;
 225         return;
 226 #endif
 227 }
 228
 229 #define DO_ERROR(trapnr, signr, str, name)                              \
 230 dotraplinkage void do_##name(struct pt_regs *regs, long error_code)     \
 231 {                                                                       \
 232         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr)  \
 233                                                         == NOTIFY_STOP) \
 234                 return;                                                 \
 235         conditional_sti(regs);                                          \
 236         do_trap(trapnr, signr, str, regs, error_code, NULL);            \
 237 }
 238
 239 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr)         \
 240 dotraplinkage void do_##name(struct pt_regs *regs, long error_code)     \
 241 {                                                                       \
 242         siginfo_t info;                                                 \
 243         info.si_signo = signr;                                          \
 244         info.si_errno = 0;                                              \
 245         info.si_code = sicode;                                          \
 246         info.si_addr = (void __user *)siaddr;                           \
 247         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr)  \
 248                                                         == NOTIFY_STOP) \
 249                 return;                                                 \
 250         conditional_sti(regs);                                          \
 251         do_trap(trapnr, signr, str, regs, error_code, &info);           \
 252 }
 253
 254 DO_ERROR_INFO(0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip)
 255 DO_ERROR(4, SIGSEGV, "overflow", overflow)
 256 DO_ERROR(5, SIGSEGV, "bounds", bounds)
 257 DO_ERROR_INFO(6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip)
 258 DO_ERROR(9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
 259 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
 260 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
 261 #ifdef CONFIG_X86_32
 262 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
 263 #endif
 264 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
 265
 266 #ifdef CONFIG_X86_64
 267 /* Runs on IST stack */
 268 dotraplinkage void do_stack_segment(struct pt_regs *regs, long error_code)
 269 {
 270         if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
 271                         12, SIGBUS) == NOTIFY_STOP)
 272                 return;
 273         preempt_conditional_sti(regs);
 274         do_trap(12, SIGBUS, "stack segment", regs, error_code, NULL);
 275         preempt_conditional_cli(regs);
 276 }
 277
 278 dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
 279 {
 280         static const char str[] = "double fault";
 281         struct task_struct *tsk = current;
 282
 283         /* Return not checked because double check cannot be ignored */
 284         notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV);
 285
 286         tsk->thread.error_code = error_code;
 287         tsk->thread.trap_no = 8;
 288
 289         /*
 290          * This is always a kernel trap and never fixable (and thus must
 291          * never return).
 292          */
 293         for (;;)
 294                 die(str, regs, error_code);
 295 }
 296 #endif
 297
 298 dotraplinkage void __kprobes
 299 do_general_protection(struct pt_regs *regs, long error_code)
 300 {
 301         struct task_struct *tsk;
 302
 303         conditional_sti(regs);
 304
 305 #ifdef CONFIG_X86_32
 306         if (lazy_iobitmap_copy()) {
 307                 /* restart the faulting instruction */
 308                 return;
 309         }
 310
 311         if (regs->flags & X86_VM_MASK)
 312                 goto gp_in_vm86;
 313 #endif
 314
 315         tsk = current;
 316         if (!user_mode(regs))
 317                 goto gp_in_kernel;
 318
 319         tsk->thread.error_code = error_code;
 320         tsk->thread.trap_no = 13;
 321
 322         if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
 323                         printk_ratelimit()) {
 324                 printk(KERN_INFO
 325                         "%s[%d] general protection ip:%lx sp:%lx error:%lx",
 326                         tsk->comm, task_pid_nr(tsk),
 327                         regs->ip, regs->sp, error_code);
 328                 print_vma_addr(" in ", regs->ip);
 329                 printk("\n");
 330         }
 331
 332         force_sig(SIGSEGV, tsk);
 333         return;
 334
 335 #ifdef CONFIG_X86_32
 336 gp_in_vm86:
 337         local_irq_enable();
 338         handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
 339         return;
 340 #endif
 341
 342 gp_in_kernel:
 343         if (fixup_exception(regs))
 344                 return;
 345
 346         tsk->thread.error_code = error_code;
 347         tsk->thread.trap_no = 13;
 348         if (notify_die(DIE_GPF, "general protection fault", regs,
 349                                 error_code, 13, SIGSEGV) == NOTIFY_STOP)
 350                 return;
 351         die("general protection fault", regs, error_code);
 352 }
 353
 354 static notrace __kprobes void
 355 mem_parity_error(unsigned char reason, struct pt_regs *regs)
 356 {
 357         printk(KERN_EMERG
 358                 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
 359                         reason, smp_processor_id());
 360
 361         printk(KERN_EMERG
 362                 "You have some hardware problem, likely on the PCI bus.\n");
 363
 364 #if defined(CONFIG_EDAC)
 365         if (edac_handler_set()) {
 366                 edac_atomic_assert_error();
 367                 return;
 368         }
 369 #endif
 370
 371         if (panic_on_unrecovered_nmi)
 372                 panic("NMI: Not continuing");
 373
 374         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 375
 376         /* Clear and disable the memory parity error line. */
 377         reason = (reason & 0xf) | 4;
 378         outb(reason, 0x61);
 379 }
 380
 381 static notrace __kprobes void
 382 io_check_error(unsigned char reason, struct pt_regs *regs)
 383 {
 384         unsigned long i;
 385
 386         printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
 387         show_registers(regs);
 388
 389         /* Re-enable the IOCK line, wait for a few seconds */
 390         reason = (reason & 0xf) | 8;
 391         outb(reason, 0x61);
 392
 393         i = 2000;
 394         while (--i)
 395                 udelay(1000);
 396
 397         reason &= ~8;
 398         outb(reason, 0x61);
 399 }
 400
 401 static notrace __kprobes void
 402 unknown_nmi_error(unsigned char reason, struct pt_regs *regs)
 403 {
 404         if (notify_die(DIE_NMIUNKNOWN, "nmi", regs, reason, 2, SIGINT) ==
 405                         NOTIFY_STOP)
 406                 return;
 407 #ifdef CONFIG_MCA
 408         /*
 409          * Might actually be able to figure out what the guilty party
 410          * is:
 411          */
 412         if (MCA_bus) {
 413                 mca_handle_nmi();
 414                 return;
 415         }
 416 #endif
 417         printk(KERN_EMERG
 418                 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
 419                         reason, smp_processor_id());
 420
 421         printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
 422         if (panic_on_unrecovered_nmi)
 423                 panic("NMI: Not continuing");
 424
 425         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 426 }
 427
 428 static notrace __kprobes void default_do_nmi(struct pt_regs *regs)
 429 {
 430         unsigned char reason = 0;
 431         int cpu;
 432
 433         cpu = smp_processor_id();
 434
 435         /* Only the BSP gets external NMIs from the system. */
 436         if (!cpu)
 437                 reason = get_nmi_reason();
 438
 439         if (!(reason & 0xc0)) {
 440                 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
 441                                                                 == NOTIFY_STOP)
 442                         return;
 443 #ifdef CONFIG_X86_LOCAL_APIC
 444                 /*
 445                  * Ok, so this is none of the documented NMI sources,
 446                  * so it must be the NMI watchdog.
 447                  */
 448                 if (nmi_watchdog_tick(regs, reason))
 449                         return;
 450                 if (!do_nmi_callback(regs, cpu))
 451                         unknown_nmi_error(reason, regs);
 452 #else
 453                 unknown_nmi_error(reason, regs);
 454 #endif
 455
 456                 return;
 457         }
 458         if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
 459                 return;
 460
 461         /* AK: following checks seem to be broken on modern chipsets. FIXME */
 462         if (reason & 0x80)
 463                 mem_parity_error(reason, regs);
 464         if (reason & 0x40)
 465                 io_check_error(reason, regs);
 466 #ifdef CONFIG_X86_32
 467         /*
 468          * Reassert NMI in case it became active meanwhile
 469          * as it's edge-triggered:
 470          */
 471         reassert_nmi();
 472 #endif
 473 }
 474
 475 dotraplinkage notrace __kprobes void
 476 do_nmi(struct pt_regs *regs, long error_code)
 477 {
 478         nmi_enter();
 479
 480         inc_irq_stat(__nmi_count);
 481
 482         if (!ignore_nmis)
 483                 default_do_nmi(regs);
 484
 485         nmi_exit();
 486 }
 487
 488 void stop_nmi(void)
 489 {
 490         acpi_nmi_disable();
 491         ignore_nmis++;
 492 }
 493
 494 void restart_nmi(void)
 495 {
 496         ignore_nmis--;
 497         acpi_nmi_enable();
 498 }
 499
 500 /* May run on IST stack. */
 501 dotraplinkage void __kprobes do_int3(struct pt_regs *regs, long error_code)
 502 {
 503 #ifdef CONFIG_KPROBES
 504         if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
 505                         == NOTIFY_STOP)
 506                 return;
 507 #else
 508         if (notify_die(DIE_TRAP, "int3", regs, error_code, 3, SIGTRAP)
 509                         == NOTIFY_STOP)
 510                 return;
 511 #endif
 512
 513         preempt_conditional_sti(regs);
 514         do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
 515         preempt_conditional_cli(regs);
 516 }
 517
 518 #ifdef CONFIG_X86_64
 519 /*
 520  * Help handler running on IST stack to switch back to user stack
 521  * for scheduling or signal handling. The actual stack switch is done in
 522  * entry.S
 523  */
 524 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
 525 {
 526         struct pt_regs *regs = eregs;
 527         /* Did already sync */
 528         if (eregs == (struct pt_regs *)eregs->sp)
 529                 ;
 530         /* Exception from user space */
 531         else if (user_mode(eregs))
 532                 regs = task_pt_regs(current);
 533         /*
 534          * Exception from kernel and interrupts are enabled. Move to
 535          * kernel process stack.
 536          */
 537         else if (eregs->flags & X86_EFLAGS_IF)
 538                 regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
 539         if (eregs != regs)
 540                 *regs = *eregs;
 541         return regs;
 542 }
 543 #endif
 544
 545 /*
 546  * Our handling of the processor debug registers is non-trivial.
 547  * We do not clear them on entry and exit from the kernel. Therefore
 548  * it is possible to get a watchpoint trap here from inside the kernel.
 549  * However, the code in ./ptrace.c has ensured that the user can
 550  * only set watchpoints on userspace addresses. Therefore the in-kernel
 551  * watchpoint trap can only occur in code which is reading/writing
 552  * from user space. Such code must not hold kernel locks (since it
 553  * can equally take a page fault), therefore it is safe to call
 554  * force_sig_info even though that claims and releases locks.
 555  *
 556  * Code in ./signal.c ensures that the debug control register
 557  * is restored before we deliver any signal, and therefore that
 558  * user code runs with the correct debug control register even though
 559  * we clear it here.
 560  *
 561  * Being careful here means that we don't have to be as careful in a
 562  * lot of more complicated places (task switching can be a bit lazy
 563  * about restoring all the debug state, and ptrace doesn't have to
 564  * find every occurrence of the TF bit that could be saved away even
 565  * by user code)
 566  *
 567  * May run on IST stack.
 568  */
 569 dotraplinkage void __kprobes do_debug(struct pt_regs *regs, long error_code)
 570 {
 571         struct task_struct *tsk = current;
 572         unsigned long condition;
 573         int si_code;
 574
 575         get_debugreg(condition, 6);
 576
 577         /*
 578          * The processor cleared BTF, so don't mark that we need it set.
 579          */
 580         clear_tsk_thread_flag(tsk, TIF_DEBUGCTLMSR);
 581         tsk->thread.debugctlmsr = 0;
 582
 583         if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
 584                                                 SIGTRAP) == NOTIFY_STOP)
 585                 return;
 586
 587         /* It's safe to allow irq's after DR6 has been saved */
 588         preempt_conditional_sti(regs);
 589
 590         /* Mask out spurious debug traps due to lazy DR7 setting */
 591         if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
 592                 if (!tsk->thread.debugreg7)
 593                         goto clear_dr7;
 594         }
 595
 596 #ifdef CONFIG_X86_32
 597         if (regs->flags & X86_VM_MASK)
 598                 goto debug_vm86;
 599 #endif
 600
 601         /* Save debug status register where ptrace can see it */
 602         tsk->thread.debugreg6 = condition;
 603
 604         /*
 605          * Single-stepping through TF: make sure we ignore any events in
 606          * kernel space (but re-enable TF when returning to user mode).
 607          */
 608         if (condition & DR_STEP) {
 609                 if (!user_mode(regs))
 610                         goto clear_TF_reenable;
 611         }
 612
 613         si_code = get_si_code(condition);
 614         /* Ok, finally something we can handle */
 615         send_sigtrap(tsk, regs, error_code, si_code);
 616
 617         /*
 618          * Disable additional traps. They'll be re-enabled when
 619          * the signal is delivered.
 620          */
 621 clear_dr7:
 622         set_debugreg(0, 7);
 623         preempt_conditional_cli(regs);
 624         return;
 625
 626 #ifdef CONFIG_X86_32
 627 debug_vm86:
 628         handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
 629         preempt_conditional_cli(regs);
 630         return;
 631 #endif
 632
 633 clear_TF_reenable:
 634         set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
 635         regs->flags &= ~X86_EFLAGS_TF;
 636         preempt_conditional_cli(regs);
 637         return;
 638 }
 639
 640 #ifdef CONFIG_X86_64
 641 static int kernel_math_error(struct pt_regs *regs, const char *str, int trapnr)
 642 {
 643         if (fixup_exception(regs))
 644                 return 1;
 645
 646         notify_die(DIE_GPF, str, regs, 0, trapnr, SIGFPE);
 647         /* Illegal floating point operation in the kernel */
 648         current->thread.trap_no = trapnr;
 649         die(str, regs, 0);
 650         return 0;
 651 }
 652 #endif
 653
 654 /*
 655  * Note that we play around with the 'TS' bit in an attempt to get
 656  * the correct behaviour even in the presence of the asynchronous
 657  * IRQ13 behaviour
 658  */
 659 void math_error(void __user *ip)
 660 {
 661         struct task_struct *task;
 662         siginfo_t info;
 663         unsigned short cwd, swd, err;
 664
 665         /*
 666          * Save the info for the exception handler and clear the error.
 667          */
 668         task = current;
 669         save_init_fpu(task);
 670         task->thread.trap_no = 16;
 671         task->thread.error_code = 0;
 672         info.si_signo = SIGFPE;
 673         info.si_errno = 0;
 674         info.si_addr = ip;
 675         /*
 676          * (~cwd & swd) will mask out exceptions that are not set to unmasked
 677          * status.  0x3f is the exception bits in these regs, 0x200 is the
 678          * C1 reg you need in case of a stack fault, 0x040 is the stack
 679          * fault bit.  We should only be taking one exception at a time,
 680          * so if this combination doesn't produce any single exception,
 681          * then we have a bad program that isn't synchronizing its FPU usage
 682          * and it will suffer the consequences since we won't be able to
 683          * fully reproduce the context of the exception
 684          */
 685         cwd = get_fpu_cwd(task);
 686         swd = get_fpu_swd(task);
 687
 688         err = swd & ~cwd;
 689
 690         if (err & 0x001) {      /* Invalid op */
 691                 /*
 692                  * swd & 0x240 == 0x040: Stack Underflow
 693                  * swd & 0x240 == 0x240: Stack Overflow
 694                  * User must clear the SF bit (0x40) if set
 695                  */
 696                 info.si_code = FPE_FLTINV;
 697         } else if (err & 0x004) { /* Divide by Zero */
 698                 info.si_code = FPE_FLTDIV;
 699         } else if (err & 0x008) { /* Overflow */
 700                 info.si_code = FPE_FLTOVF;
 701         } else if (err & 0x012) { /* Denormal, Underflow */
 702                 info.si_code = FPE_FLTUND;
 703         } else if (err & 0x020) { /* Precision */
 704                 info.si_code = FPE_FLTRES;
 705         } else {
 706                 /*
 707                  * If we're using IRQ 13, or supposedly even some trap 16
 708                  * implementations, it's possible we get a spurious trap...
 709                  */
 710                 return;         /* Spurious trap, no error */
 711         }
 712         force_sig_info(SIGFPE, &info, task);
 713 }
 714
 715 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
 716 {
 717         conditional_sti(regs);
 718
 719 #ifdef CONFIG_X86_32
 720         ignore_fpu_irq = 1;
 721 #else
 722         if (!user_mode(regs) &&
 723             kernel_math_error(regs, "kernel x87 math error", 16))
 724                 return;
 725 #endif
 726
 727         math_error((void __user *)regs->ip);
 728 }
 729
 730 static void simd_math_error(void __user *ip)
 731 {
 732         struct task_struct *task;
 733         siginfo_t info;
 734         unsigned short mxcsr;
 735
 736         /*
 737          * Save the info for the exception handler and clear the error.
 738          */
 739         task = current;
 740         save_init_fpu(task);
 741         task->thread.trap_no = 19;
 742         task->thread.error_code = 0;
 743         info.si_signo = SIGFPE;
 744         info.si_errno = 0;
 745         info.si_code = __SI_FAULT;
 746         info.si_addr = ip;
 747         /*
 748          * The SIMD FPU exceptions are handled a little differently, as there
 749          * is only a single status/control register.  Thus, to determine which
 750          * unmasked exception was caught we must mask the exception mask bits
 751          * at 0x1f80, and then use these to mask the exception bits at 0x3f.
 752          */
 753         mxcsr = get_fpu_mxcsr(task);
 754         switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
 755         case 0x000:
 756         default:
 757                 break;
 758         case 0x001: /* Invalid Op */
 759                 info.si_code = FPE_FLTINV;
 760                 break;
 761         case 0x002: /* Denormalize */
 762         case 0x010: /* Underflow */
 763                 info.si_code = FPE_FLTUND;
 764                 break;
 765         case 0x004: /* Zero Divide */
 766                 info.si_code = FPE_FLTDIV;
 767                 break;
 768         case 0x008: /* Overflow */
 769                 info.si_code = FPE_FLTOVF;
 770                 break;
 771         case 0x020: /* Precision */
 772                 info.si_code = FPE_FLTRES;
 773                 break;
 774         }
 775         force_sig_info(SIGFPE, &info, task);
 776 }
 777
 778 dotraplinkage void
 779 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
 780 {
 781         conditional_sti(regs);
 782
 783 #ifdef CONFIG_X86_32
 784         if (cpu_has_xmm) {
 785                 /* Handle SIMD FPU exceptions on PIII+ processors. */
 786                 ignore_fpu_irq = 1;
 787                 simd_math_error((void __user *)regs->ip);
 788                 return;
 789         }
 790         /*
 791          * Handle strange cache flush from user space exception
 792          * in all other cases.  This is undocumented behaviour.
 793          */
 794         if (regs->flags & X86_VM_MASK) {
 795                 handle_vm86_fault((struct kernel_vm86_regs *)regs, error_code);
 796                 return;
 797         }
 798         current->thread.trap_no = 19;
 799         current->thread.error_code = error_code;
 800         die_if_kernel("cache flush denied", regs, error_code);
 801         force_sig(SIGSEGV, current);
 802 #else
 803         if (!user_mode(regs) &&
 804                         kernel_math_error(regs, "kernel simd math error", 19))
 805                 return;
 806         simd_math_error((void __user *)regs->ip);
 807 #endif
 808 }
 809
 810 dotraplinkage void
 811 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
 812 {
 813         conditional_sti(regs);
 814 #if 0
 815         /* No need to warn about this any longer. */
 816         printk(KERN_INFO "Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
 817 #endif
 818 }
 819
 820 #ifdef CONFIG_X86_32
 821 unsigned long patch_espfix_desc(unsigned long uesp, unsigned long kesp)
 822 {
 823         struct desc_struct *gdt = get_cpu_gdt_table(smp_processor_id());
 824         unsigned long base = (kesp - uesp) & -THREAD_SIZE;
 825         unsigned long new_kesp = kesp - base;
 826         unsigned long lim_pages = (new_kesp | (THREAD_SIZE - 1)) >> PAGE_SHIFT;
 827         __u64 desc = *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS];
 828
 829         /* Set up base for espfix segment */
 830         desc &= 0x00f0ff0000000000ULL;
 831         desc |= ((((__u64)base) << 16) & 0x000000ffffff0000ULL) |
 832                 ((((__u64)base) << 32) & 0xff00000000000000ULL) |
 833                 ((((__u64)lim_pages) << 32) & 0x000f000000000000ULL) |
 834                 (lim_pages & 0xffff);
 835         *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS] = desc;
 836
 837         return new_kesp;
 838 }
 839 #else
 840 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
 841 {
 842 }
 843
 844 asmlinkage void __attribute__((weak)) mce_threshold_interrupt(void)
 845 {
 846 }
 847 #endif
 848
 849 /*
 850  * 'math_state_restore()' saves the current math information in the
 851  * old math state array, and gets the new ones from the current task
 852  *
 853  * Careful.. There are problems with IBM-designed IRQ13 behaviour.
 854  * Don't touch unless you *really* know how it works.
 855  *
 856  * Must be called with kernel preemption disabled (in this case,
 857  * local interrupts are disabled at the call-site in entry.S).
 858  */
 859 asmlinkage void math_state_restore(void)
 860 {
 861         struct thread_info *thread = current_thread_info();
 862         struct task_struct *tsk = thread->task;
 863
 864         if (!tsk_used_math(tsk)) {
 865                 local_irq_enable();
 866                 /*
 867                  * does a slab alloc which can sleep
 868                  */
 869                 if (init_fpu(tsk)) {
 870                         /*
 871                          * ran out of memory!
 872                          */
 873                         do_group_exit(SIGKILL);
 874                         return;
 875                 }
 876                 local_irq_disable();
 877         }
 878
 879         clts();                         /* Allow maths ops (or we recurse) */
 880 #ifdef CONFIG_X86_32
 881         restore_fpu(tsk);
 882 #else
 883         /*
 884          * Paranoid restore. send a SIGSEGV if we fail to restore the state.
 885          */
 886         if (unlikely(restore_fpu_checking(tsk))) {
 887                 stts();
 888                 force_sig(SIGSEGV, tsk);
 889                 return;
 890         }
 891 #endif
 892         thread->status |= TS_USEDFPU;   /* So we fnsave on switch_to() */
 893         tsk->fpu_counter++;
 894 }
 895 EXPORT_SYMBOL_GPL(math_state_restore);
 896
 897 #ifndef CONFIG_MATH_EMULATION
 898 void math_emulate(struct math_emu_info *info)
 899 {
 900         printk(KERN_EMERG
 901                 "math-emulation not enabled and no coprocessor found.\n");
 902         printk(KERN_EMERG "killing %s.\n", current->comm);
 903         force_sig(SIGFPE, current);
 904         schedule();
 905 }
 906 #endif /* CONFIG_MATH_EMULATION */
 907
 908 dotraplinkage void __kprobes do_device_not_available(struct pt_regs regs)
 909 {
 910 #ifdef CONFIG_X86_32
 911         if (read_cr0() & X86_CR0_EM) {
 912                 struct math_emu_info info = { };
 913
 914                 conditional_sti(&regs);
 915
 916                 info.regs = &regs;
 917                 math_emulate(&info);
 918         } else {
 919                 math_state_restore(); /* interrupts still off */
 920                 conditional_sti(&regs);
 921         }
 922 #else
 923         math_state_restore();
 924 #endif
 925 }
 926
 927 #ifdef CONFIG_X86_32
 928 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
 929 {
 930         siginfo_t info;
 931         local_irq_enable();
 932
 933         info.si_signo = SIGILL;
 934         info.si_errno = 0;
 935         info.si_code = ILL_BADSTK;
 936         info.si_addr = 0;
 937         if (notify_die(DIE_TRAP, "iret exception",
 938                         regs, error_code, 32, SIGILL) == NOTIFY_STOP)
 939                 return;
 940         do_trap(32, SIGILL, "iret exception", regs, error_code, &info);
 941 }
 942 #endif
 943
 944 void __init trap_init(void)
 945 {
 946         int i;
 947
 948 #ifdef CONFIG_EISA
 949         void __iomem *p = early_ioremap(0x0FFFD9, 4);
 950
 951         if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
 952                 EISA_bus = 1;
 953         early_iounmap(p, 4);
 954 #endif
 955
 956         set_intr_gate(0, &divide_error);
 957         set_intr_gate_ist(1, &debug, DEBUG_STACK);
 958         set_intr_gate_ist(2, &nmi, NMI_STACK);
 959         /* int3 can be called from all */
 960         set_system_intr_gate_ist(3, &int3, DEBUG_STACK);
 961         /* int4 can be called from all */
 962         set_system_intr_gate(4, &overflow);
 963         set_intr_gate(5, &bounds);
 964         set_intr_gate(6, &invalid_op);
 965         set_intr_gate(7, &device_not_available);
 966 #ifdef CONFIG_X86_32
 967         set_task_gate(8, GDT_ENTRY_DOUBLEFAULT_TSS);
 968 #else
 969         set_intr_gate_ist(8, &double_fault, DOUBLEFAULT_STACK);
 970 #endif
 971         set_intr_gate(9, &coprocessor_segment_overrun);
 972         set_intr_gate(10, &invalid_TSS);
 973         set_intr_gate(11, &segment_not_present);
 974         set_intr_gate_ist(12, &stack_segment, STACKFAULT_STACK);
 975         set_intr_gate(13, &general_protection);
 976         set_intr_gate(14, &page_fault);
 977         set_intr_gate(15, &spurious_interrupt_bug);
 978         set_intr_gate(16, &coprocessor_error);
 979         set_intr_gate(17, &alignment_check);
 980 #ifdef CONFIG_X86_MCE
 981         set_intr_gate_ist(18, &machine_check, MCE_STACK);
 982 #endif
 983         set_intr_gate(19, &simd_coprocessor_error);
 984
 985 #ifdef CONFIG_IA32_EMULATION
 986         set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
 987 #endif
 988
 989 #ifdef CONFIG_X86_32
 990         if (cpu_has_fxsr) {
 991                 printk(KERN_INFO "Enabling fast FPU save and restore... ");
 992                 set_in_cr4(X86_CR4_OSFXSR);
 993                 printk("done.\n");
 994         }
 995         if (cpu_has_xmm) {
 996                 printk(KERN_INFO
 997                         "Enabling unmasked SIMD FPU exception support... ");
 998                 set_in_cr4(X86_CR4_OSXMMEXCPT);
 999                 printk("done.\n");
1000         }
1001
1002         set_system_trap_gate(SYSCALL_VECTOR, &system_call);
1003 #endif
1004
1005         /* Reserve all the builtin and the syscall vector: */
1006         for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
1007                 set_bit(i, used_vectors);
1008
1009 #ifdef CONFIG_X86_64
1010         set_bit(IA32_SYSCALL_VECTOR, used_vectors);
1011 #else
1012         set_bit(SYSCALL_VECTOR, used_vectors);
1013 #endif
1014         /*
1015          * Should be a barrier for any external CPU state:
1016          */
1017         cpu_init();
1018
1019 #ifdef CONFIG_X86_32
1020         trap_init_hook();
1021 #endif
1022 }